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 | { |
9b64f57d ER |
63 | WARN_ON(refcount_read(&transaction->use_count) == 0); |
64 | if (refcount_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) |
ab8d0fc4 JM |
68 | btrfs_err(transaction->fs_info, |
69 | "pending csums is %llu", | |
70 | transaction->delayed_refs.pending_csums); | |
6df9a95e JB |
71 | while (!list_empty(&transaction->pending_chunks)) { |
72 | struct extent_map *em; | |
73 | ||
74 | em = list_first_entry(&transaction->pending_chunks, | |
75 | struct extent_map, list); | |
76 | list_del_init(&em->list); | |
77 | free_extent_map(em); | |
78 | } | |
7785a663 FM |
79 | /* |
80 | * If any block groups are found in ->deleted_bgs then it's | |
81 | * because the transaction was aborted and a commit did not | |
82 | * happen (things failed before writing the new superblock | |
83 | * and calling btrfs_finish_extent_commit()), so we can not | |
84 | * discard the physical locations of the block groups. | |
85 | */ | |
86 | while (!list_empty(&transaction->deleted_bgs)) { | |
87 | struct btrfs_block_group_cache *cache; | |
88 | ||
89 | cache = list_first_entry(&transaction->deleted_bgs, | |
90 | struct btrfs_block_group_cache, | |
91 | bg_list); | |
92 | list_del_init(&cache->bg_list); | |
93 | btrfs_put_block_group_trimming(cache); | |
94 | btrfs_put_block_group(cache); | |
95 | } | |
2c90e5d6 | 96 | kmem_cache_free(btrfs_transaction_cachep, transaction); |
78fae27e | 97 | } |
79154b1b CM |
98 | } |
99 | ||
663dfbb0 FM |
100 | static void clear_btree_io_tree(struct extent_io_tree *tree) |
101 | { | |
102 | spin_lock(&tree->lock); | |
b666a9cd DS |
103 | /* |
104 | * Do a single barrier for the waitqueue_active check here, the state | |
105 | * of the waitqueue should not change once clear_btree_io_tree is | |
106 | * called. | |
107 | */ | |
108 | smp_mb(); | |
663dfbb0 FM |
109 | while (!RB_EMPTY_ROOT(&tree->state)) { |
110 | struct rb_node *node; | |
111 | struct extent_state *state; | |
112 | ||
113 | node = rb_first(&tree->state); | |
114 | state = rb_entry(node, struct extent_state, rb_node); | |
115 | rb_erase(&state->rb_node, &tree->state); | |
116 | RB_CLEAR_NODE(&state->rb_node); | |
117 | /* | |
118 | * btree io trees aren't supposed to have tasks waiting for | |
119 | * changes in the flags of extent states ever. | |
120 | */ | |
121 | ASSERT(!waitqueue_active(&state->wq)); | |
122 | free_extent_state(state); | |
351810c1 DS |
123 | |
124 | cond_resched_lock(&tree->lock); | |
663dfbb0 FM |
125 | } |
126 | spin_unlock(&tree->lock); | |
127 | } | |
128 | ||
9e351cc8 JB |
129 | static noinline void switch_commit_roots(struct btrfs_transaction *trans, |
130 | struct btrfs_fs_info *fs_info) | |
817d52f8 | 131 | { |
9e351cc8 JB |
132 | struct btrfs_root *root, *tmp; |
133 | ||
134 | down_write(&fs_info->commit_root_sem); | |
135 | list_for_each_entry_safe(root, tmp, &trans->switch_commits, | |
136 | dirty_list) { | |
137 | list_del_init(&root->dirty_list); | |
138 | free_extent_buffer(root->commit_root); | |
139 | root->commit_root = btrfs_root_node(root); | |
140 | if (is_fstree(root->objectid)) | |
141 | btrfs_unpin_free_ino(root); | |
663dfbb0 | 142 | clear_btree_io_tree(&root->dirty_log_pages); |
9e351cc8 | 143 | } |
2b9dbef2 JB |
144 | |
145 | /* We can free old roots now. */ | |
146 | spin_lock(&trans->dropped_roots_lock); | |
147 | while (!list_empty(&trans->dropped_roots)) { | |
148 | root = list_first_entry(&trans->dropped_roots, | |
149 | struct btrfs_root, root_list); | |
150 | list_del_init(&root->root_list); | |
151 | spin_unlock(&trans->dropped_roots_lock); | |
152 | btrfs_drop_and_free_fs_root(fs_info, root); | |
153 | spin_lock(&trans->dropped_roots_lock); | |
154 | } | |
155 | spin_unlock(&trans->dropped_roots_lock); | |
9e351cc8 | 156 | up_write(&fs_info->commit_root_sem); |
817d52f8 JB |
157 | } |
158 | ||
0860adfd MX |
159 | static inline void extwriter_counter_inc(struct btrfs_transaction *trans, |
160 | unsigned int type) | |
161 | { | |
162 | if (type & TRANS_EXTWRITERS) | |
163 | atomic_inc(&trans->num_extwriters); | |
164 | } | |
165 | ||
166 | static inline void extwriter_counter_dec(struct btrfs_transaction *trans, | |
167 | unsigned int type) | |
168 | { | |
169 | if (type & TRANS_EXTWRITERS) | |
170 | atomic_dec(&trans->num_extwriters); | |
171 | } | |
172 | ||
173 | static inline void extwriter_counter_init(struct btrfs_transaction *trans, | |
174 | unsigned int type) | |
175 | { | |
176 | atomic_set(&trans->num_extwriters, ((type & TRANS_EXTWRITERS) ? 1 : 0)); | |
177 | } | |
178 | ||
179 | static inline int extwriter_counter_read(struct btrfs_transaction *trans) | |
180 | { | |
181 | return atomic_read(&trans->num_extwriters); | |
178260b2 MX |
182 | } |
183 | ||
d352ac68 CM |
184 | /* |
185 | * either allocate a new transaction or hop into the existing one | |
186 | */ | |
2ff7e61e JM |
187 | static noinline int join_transaction(struct btrfs_fs_info *fs_info, |
188 | unsigned int type) | |
79154b1b CM |
189 | { |
190 | struct btrfs_transaction *cur_trans; | |
a4abeea4 | 191 | |
19ae4e81 | 192 | spin_lock(&fs_info->trans_lock); |
d43317dc | 193 | loop: |
49b25e05 | 194 | /* The file system has been taken offline. No new transactions. */ |
87533c47 | 195 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { |
19ae4e81 | 196 | spin_unlock(&fs_info->trans_lock); |
49b25e05 JM |
197 | return -EROFS; |
198 | } | |
199 | ||
19ae4e81 | 200 | cur_trans = fs_info->running_transaction; |
a4abeea4 | 201 | if (cur_trans) { |
871383be | 202 | if (cur_trans->aborted) { |
19ae4e81 | 203 | spin_unlock(&fs_info->trans_lock); |
49b25e05 | 204 | return cur_trans->aborted; |
871383be | 205 | } |
4a9d8bde | 206 | if (btrfs_blocked_trans_types[cur_trans->state] & type) { |
178260b2 MX |
207 | spin_unlock(&fs_info->trans_lock); |
208 | return -EBUSY; | |
209 | } | |
9b64f57d | 210 | refcount_inc(&cur_trans->use_count); |
13c5a93e | 211 | atomic_inc(&cur_trans->num_writers); |
0860adfd | 212 | extwriter_counter_inc(cur_trans, type); |
19ae4e81 | 213 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 | 214 | return 0; |
79154b1b | 215 | } |
19ae4e81 | 216 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 | 217 | |
354aa0fb MX |
218 | /* |
219 | * If we are ATTACH, we just want to catch the current transaction, | |
220 | * and commit it. If there is no transaction, just return ENOENT. | |
221 | */ | |
222 | if (type == TRANS_ATTACH) | |
223 | return -ENOENT; | |
224 | ||
4a9d8bde MX |
225 | /* |
226 | * JOIN_NOLOCK only happens during the transaction commit, so | |
227 | * it is impossible that ->running_transaction is NULL | |
228 | */ | |
229 | BUG_ON(type == TRANS_JOIN_NOLOCK); | |
230 | ||
a4abeea4 JB |
231 | cur_trans = kmem_cache_alloc(btrfs_transaction_cachep, GFP_NOFS); |
232 | if (!cur_trans) | |
233 | return -ENOMEM; | |
d43317dc | 234 | |
19ae4e81 JS |
235 | spin_lock(&fs_info->trans_lock); |
236 | if (fs_info->running_transaction) { | |
d43317dc CM |
237 | /* |
238 | * someone started a transaction after we unlocked. Make sure | |
4a9d8bde | 239 | * to redo the checks above |
d43317dc | 240 | */ |
a4abeea4 | 241 | kmem_cache_free(btrfs_transaction_cachep, cur_trans); |
d43317dc | 242 | goto loop; |
87533c47 | 243 | } else if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { |
e4b50e14 | 244 | spin_unlock(&fs_info->trans_lock); |
7b8b92af JB |
245 | kmem_cache_free(btrfs_transaction_cachep, cur_trans); |
246 | return -EROFS; | |
79154b1b | 247 | } |
d43317dc | 248 | |
ab8d0fc4 | 249 | cur_trans->fs_info = fs_info; |
a4abeea4 | 250 | atomic_set(&cur_trans->num_writers, 1); |
0860adfd | 251 | extwriter_counter_init(cur_trans, type); |
a4abeea4 JB |
252 | init_waitqueue_head(&cur_trans->writer_wait); |
253 | init_waitqueue_head(&cur_trans->commit_wait); | |
161c3549 | 254 | init_waitqueue_head(&cur_trans->pending_wait); |
4a9d8bde | 255 | cur_trans->state = TRANS_STATE_RUNNING; |
a4abeea4 JB |
256 | /* |
257 | * One for this trans handle, one so it will live on until we | |
258 | * commit the transaction. | |
259 | */ | |
9b64f57d | 260 | refcount_set(&cur_trans->use_count, 2); |
161c3549 | 261 | atomic_set(&cur_trans->pending_ordered, 0); |
3204d33c | 262 | cur_trans->flags = 0; |
a4abeea4 JB |
263 | cur_trans->start_time = get_seconds(); |
264 | ||
a099d0fd AM |
265 | memset(&cur_trans->delayed_refs, 0, sizeof(cur_trans->delayed_refs)); |
266 | ||
c46effa6 | 267 | cur_trans->delayed_refs.href_root = RB_ROOT; |
3368d001 | 268 | cur_trans->delayed_refs.dirty_extent_root = RB_ROOT; |
d7df2c79 | 269 | atomic_set(&cur_trans->delayed_refs.num_entries, 0); |
20b297d6 JS |
270 | |
271 | /* | |
272 | * although the tree mod log is per file system and not per transaction, | |
273 | * the log must never go across transaction boundaries. | |
274 | */ | |
275 | smp_mb(); | |
31b1a2bd | 276 | if (!list_empty(&fs_info->tree_mod_seq_list)) |
5d163e0e | 277 | WARN(1, KERN_ERR "BTRFS: tree_mod_seq_list not empty when creating a fresh transaction\n"); |
31b1a2bd | 278 | if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log)) |
5d163e0e | 279 | WARN(1, KERN_ERR "BTRFS: tree_mod_log rb tree not empty when creating a fresh transaction\n"); |
fc36ed7e | 280 | atomic64_set(&fs_info->tree_mod_seq, 0); |
20b297d6 | 281 | |
a4abeea4 JB |
282 | spin_lock_init(&cur_trans->delayed_refs.lock); |
283 | ||
284 | INIT_LIST_HEAD(&cur_trans->pending_snapshots); | |
6df9a95e | 285 | INIT_LIST_HEAD(&cur_trans->pending_chunks); |
9e351cc8 | 286 | INIT_LIST_HEAD(&cur_trans->switch_commits); |
ce93ec54 | 287 | INIT_LIST_HEAD(&cur_trans->dirty_bgs); |
1bbc621e | 288 | INIT_LIST_HEAD(&cur_trans->io_bgs); |
2b9dbef2 | 289 | INIT_LIST_HEAD(&cur_trans->dropped_roots); |
1bbc621e | 290 | mutex_init(&cur_trans->cache_write_mutex); |
cb723e49 | 291 | cur_trans->num_dirty_bgs = 0; |
ce93ec54 | 292 | spin_lock_init(&cur_trans->dirty_bgs_lock); |
e33e17ee | 293 | INIT_LIST_HEAD(&cur_trans->deleted_bgs); |
2b9dbef2 | 294 | spin_lock_init(&cur_trans->dropped_roots_lock); |
19ae4e81 | 295 | list_add_tail(&cur_trans->list, &fs_info->trans_list); |
a4abeea4 | 296 | extent_io_tree_init(&cur_trans->dirty_pages, |
19ae4e81 JS |
297 | fs_info->btree_inode->i_mapping); |
298 | fs_info->generation++; | |
299 | cur_trans->transid = fs_info->generation; | |
300 | fs_info->running_transaction = cur_trans; | |
49b25e05 | 301 | cur_trans->aborted = 0; |
19ae4e81 | 302 | spin_unlock(&fs_info->trans_lock); |
15ee9bc7 | 303 | |
79154b1b CM |
304 | return 0; |
305 | } | |
306 | ||
d352ac68 | 307 | /* |
d397712b CM |
308 | * this does all the record keeping required to make sure that a reference |
309 | * counted root is properly recorded in a given transaction. This is required | |
310 | * to make sure the old root from before we joined the transaction is deleted | |
311 | * when the transaction commits | |
d352ac68 | 312 | */ |
7585717f | 313 | static int record_root_in_trans(struct btrfs_trans_handle *trans, |
6426c7ad QW |
314 | struct btrfs_root *root, |
315 | int force) | |
6702ed49 | 316 | { |
0b246afa JM |
317 | struct btrfs_fs_info *fs_info = root->fs_info; |
318 | ||
6426c7ad QW |
319 | if ((test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
320 | root->last_trans < trans->transid) || force) { | |
0b246afa | 321 | WARN_ON(root == fs_info->extent_root); |
5d4f98a2 YZ |
322 | WARN_ON(root->commit_root != root->node); |
323 | ||
7585717f | 324 | /* |
27cdeb70 | 325 | * see below for IN_TRANS_SETUP usage rules |
7585717f CM |
326 | * we have the reloc mutex held now, so there |
327 | * is only one writer in this function | |
328 | */ | |
27cdeb70 | 329 | set_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); |
7585717f | 330 | |
27cdeb70 | 331 | /* make sure readers find IN_TRANS_SETUP before |
7585717f CM |
332 | * they find our root->last_trans update |
333 | */ | |
334 | smp_wmb(); | |
335 | ||
0b246afa | 336 | spin_lock(&fs_info->fs_roots_radix_lock); |
6426c7ad | 337 | if (root->last_trans == trans->transid && !force) { |
0b246afa | 338 | spin_unlock(&fs_info->fs_roots_radix_lock); |
a4abeea4 JB |
339 | return 0; |
340 | } | |
0b246afa JM |
341 | radix_tree_tag_set(&fs_info->fs_roots_radix, |
342 | (unsigned long)root->root_key.objectid, | |
343 | BTRFS_ROOT_TRANS_TAG); | |
344 | spin_unlock(&fs_info->fs_roots_radix_lock); | |
7585717f CM |
345 | root->last_trans = trans->transid; |
346 | ||
347 | /* this is pretty tricky. We don't want to | |
348 | * take the relocation lock in btrfs_record_root_in_trans | |
349 | * unless we're really doing the first setup for this root in | |
350 | * this transaction. | |
351 | * | |
352 | * Normally we'd use root->last_trans as a flag to decide | |
353 | * if we want to take the expensive mutex. | |
354 | * | |
355 | * But, we have to set root->last_trans before we | |
356 | * init the relocation root, otherwise, we trip over warnings | |
357 | * in ctree.c. The solution used here is to flag ourselves | |
27cdeb70 | 358 | * with root IN_TRANS_SETUP. When this is 1, we're still |
7585717f CM |
359 | * fixing up the reloc trees and everyone must wait. |
360 | * | |
361 | * When this is zero, they can trust root->last_trans and fly | |
362 | * through btrfs_record_root_in_trans without having to take the | |
363 | * lock. smp_wmb() makes sure that all the writes above are | |
364 | * done before we pop in the zero below | |
365 | */ | |
5d4f98a2 | 366 | btrfs_init_reloc_root(trans, root); |
c7548af6 | 367 | smp_mb__before_atomic(); |
27cdeb70 | 368 | clear_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); |
5d4f98a2 YZ |
369 | } |
370 | return 0; | |
371 | } | |
bcc63abb | 372 | |
7585717f | 373 | |
2b9dbef2 JB |
374 | void btrfs_add_dropped_root(struct btrfs_trans_handle *trans, |
375 | struct btrfs_root *root) | |
376 | { | |
0b246afa | 377 | struct btrfs_fs_info *fs_info = root->fs_info; |
2b9dbef2 JB |
378 | struct btrfs_transaction *cur_trans = trans->transaction; |
379 | ||
380 | /* Add ourselves to the transaction dropped list */ | |
381 | spin_lock(&cur_trans->dropped_roots_lock); | |
382 | list_add_tail(&root->root_list, &cur_trans->dropped_roots); | |
383 | spin_unlock(&cur_trans->dropped_roots_lock); | |
384 | ||
385 | /* Make sure we don't try to update the root at commit time */ | |
0b246afa JM |
386 | spin_lock(&fs_info->fs_roots_radix_lock); |
387 | radix_tree_tag_clear(&fs_info->fs_roots_radix, | |
2b9dbef2 JB |
388 | (unsigned long)root->root_key.objectid, |
389 | BTRFS_ROOT_TRANS_TAG); | |
0b246afa | 390 | spin_unlock(&fs_info->fs_roots_radix_lock); |
2b9dbef2 JB |
391 | } |
392 | ||
7585717f CM |
393 | int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, |
394 | struct btrfs_root *root) | |
395 | { | |
0b246afa JM |
396 | struct btrfs_fs_info *fs_info = root->fs_info; |
397 | ||
27cdeb70 | 398 | if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state)) |
7585717f CM |
399 | return 0; |
400 | ||
401 | /* | |
27cdeb70 | 402 | * see record_root_in_trans for comments about IN_TRANS_SETUP usage |
7585717f CM |
403 | * and barriers |
404 | */ | |
405 | smp_rmb(); | |
406 | if (root->last_trans == trans->transid && | |
27cdeb70 | 407 | !test_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state)) |
7585717f CM |
408 | return 0; |
409 | ||
0b246afa | 410 | mutex_lock(&fs_info->reloc_mutex); |
6426c7ad | 411 | record_root_in_trans(trans, root, 0); |
0b246afa | 412 | mutex_unlock(&fs_info->reloc_mutex); |
7585717f CM |
413 | |
414 | return 0; | |
415 | } | |
416 | ||
4a9d8bde MX |
417 | static inline int is_transaction_blocked(struct btrfs_transaction *trans) |
418 | { | |
419 | return (trans->state >= TRANS_STATE_BLOCKED && | |
501407aa JB |
420 | trans->state < TRANS_STATE_UNBLOCKED && |
421 | !trans->aborted); | |
4a9d8bde MX |
422 | } |
423 | ||
d352ac68 CM |
424 | /* wait for commit against the current transaction to become unblocked |
425 | * when this is done, it is safe to start a new transaction, but the current | |
426 | * transaction might not be fully on disk. | |
427 | */ | |
2ff7e61e | 428 | static void wait_current_trans(struct btrfs_fs_info *fs_info) |
79154b1b | 429 | { |
f9295749 | 430 | struct btrfs_transaction *cur_trans; |
79154b1b | 431 | |
0b246afa JM |
432 | spin_lock(&fs_info->trans_lock); |
433 | cur_trans = fs_info->running_transaction; | |
4a9d8bde | 434 | if (cur_trans && is_transaction_blocked(cur_trans)) { |
9b64f57d | 435 | refcount_inc(&cur_trans->use_count); |
0b246afa | 436 | spin_unlock(&fs_info->trans_lock); |
72d63ed6 | 437 | |
0b246afa | 438 | wait_event(fs_info->transaction_wait, |
501407aa JB |
439 | cur_trans->state >= TRANS_STATE_UNBLOCKED || |
440 | cur_trans->aborted); | |
724e2315 | 441 | btrfs_put_transaction(cur_trans); |
a4abeea4 | 442 | } else { |
0b246afa | 443 | spin_unlock(&fs_info->trans_lock); |
f9295749 | 444 | } |
37d1aeee CM |
445 | } |
446 | ||
2ff7e61e | 447 | static int may_wait_transaction(struct btrfs_fs_info *fs_info, int type) |
a22285a6 | 448 | { |
0b246afa | 449 | if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) |
a4abeea4 JB |
450 | return 0; |
451 | ||
452 | if (type == TRANS_USERSPACE) | |
453 | return 1; | |
454 | ||
455 | if (type == TRANS_START && | |
0b246afa | 456 | !atomic_read(&fs_info->open_ioctl_trans)) |
a22285a6 | 457 | return 1; |
a4abeea4 | 458 | |
a22285a6 YZ |
459 | return 0; |
460 | } | |
461 | ||
20dd2cbf MX |
462 | static inline bool need_reserve_reloc_root(struct btrfs_root *root) |
463 | { | |
0b246afa JM |
464 | struct btrfs_fs_info *fs_info = root->fs_info; |
465 | ||
466 | if (!fs_info->reloc_ctl || | |
27cdeb70 | 467 | !test_bit(BTRFS_ROOT_REF_COWS, &root->state) || |
20dd2cbf MX |
468 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || |
469 | root->reloc_root) | |
470 | return false; | |
471 | ||
472 | return true; | |
473 | } | |
474 | ||
08e007d2 | 475 | static struct btrfs_trans_handle * |
5aed1dd8 | 476 | start_transaction(struct btrfs_root *root, unsigned int num_items, |
003d7c59 JM |
477 | unsigned int type, enum btrfs_reserve_flush_enum flush, |
478 | bool enforce_qgroups) | |
37d1aeee | 479 | { |
0b246afa JM |
480 | struct btrfs_fs_info *fs_info = root->fs_info; |
481 | ||
a22285a6 YZ |
482 | struct btrfs_trans_handle *h; |
483 | struct btrfs_transaction *cur_trans; | |
b5009945 | 484 | u64 num_bytes = 0; |
c5567237 | 485 | u64 qgroup_reserved = 0; |
20dd2cbf MX |
486 | bool reloc_reserved = false; |
487 | int ret; | |
acce952b | 488 | |
46c4e71e | 489 | /* Send isn't supposed to start transactions. */ |
2755a0de | 490 | ASSERT(current->journal_info != BTRFS_SEND_TRANS_STUB); |
46c4e71e | 491 | |
0b246afa | 492 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) |
acce952b | 493 | return ERR_PTR(-EROFS); |
2a1eb461 | 494 | |
46c4e71e | 495 | if (current->journal_info) { |
0860adfd | 496 | WARN_ON(type & TRANS_EXTWRITERS); |
2a1eb461 JB |
497 | h = current->journal_info; |
498 | h->use_count++; | |
b7d5b0a8 | 499 | WARN_ON(h->use_count > 2); |
2a1eb461 JB |
500 | h->orig_rsv = h->block_rsv; |
501 | h->block_rsv = NULL; | |
502 | goto got_it; | |
503 | } | |
b5009945 JB |
504 | |
505 | /* | |
506 | * Do the reservation before we join the transaction so we can do all | |
507 | * the appropriate flushing if need be. | |
508 | */ | |
003d7c59 | 509 | if (num_items && root != fs_info->chunk_root) { |
0b246afa | 510 | qgroup_reserved = num_items * fs_info->nodesize; |
003d7c59 JM |
511 | ret = btrfs_qgroup_reserve_meta(root, qgroup_reserved, |
512 | enforce_qgroups); | |
7174109c QW |
513 | if (ret) |
514 | return ERR_PTR(ret); | |
c5567237 | 515 | |
0b246afa | 516 | num_bytes = btrfs_calc_trans_metadata_size(fs_info, num_items); |
20dd2cbf MX |
517 | /* |
518 | * Do the reservation for the relocation root creation | |
519 | */ | |
ee39b432 | 520 | if (need_reserve_reloc_root(root)) { |
0b246afa | 521 | num_bytes += fs_info->nodesize; |
20dd2cbf MX |
522 | reloc_reserved = true; |
523 | } | |
524 | ||
0b246afa | 525 | ret = btrfs_block_rsv_add(root, &fs_info->trans_block_rsv, |
08e007d2 | 526 | num_bytes, flush); |
b5009945 | 527 | if (ret) |
843fcf35 | 528 | goto reserve_fail; |
b5009945 | 529 | } |
a22285a6 | 530 | again: |
f2f767e7 | 531 | h = kmem_cache_zalloc(btrfs_trans_handle_cachep, GFP_NOFS); |
843fcf35 MX |
532 | if (!h) { |
533 | ret = -ENOMEM; | |
534 | goto alloc_fail; | |
535 | } | |
37d1aeee | 536 | |
98114659 JB |
537 | /* |
538 | * If we are JOIN_NOLOCK we're already committing a transaction and | |
539 | * waiting on this guy, so we don't need to do the sb_start_intwrite | |
540 | * because we're already holding a ref. We need this because we could | |
541 | * have raced in and did an fsync() on a file which can kick a commit | |
542 | * and then we deadlock with somebody doing a freeze. | |
354aa0fb MX |
543 | * |
544 | * If we are ATTACH, it means we just want to catch the current | |
545 | * transaction and commit it, so we needn't do sb_start_intwrite(). | |
98114659 | 546 | */ |
0860adfd | 547 | if (type & __TRANS_FREEZABLE) |
0b246afa | 548 | sb_start_intwrite(fs_info->sb); |
b2b5ef5c | 549 | |
2ff7e61e JM |
550 | if (may_wait_transaction(fs_info, type)) |
551 | wait_current_trans(fs_info); | |
a22285a6 | 552 | |
a4abeea4 | 553 | do { |
2ff7e61e | 554 | ret = join_transaction(fs_info, type); |
178260b2 | 555 | if (ret == -EBUSY) { |
2ff7e61e | 556 | wait_current_trans(fs_info); |
178260b2 MX |
557 | if (unlikely(type == TRANS_ATTACH)) |
558 | ret = -ENOENT; | |
559 | } | |
a4abeea4 JB |
560 | } while (ret == -EBUSY); |
561 | ||
a43f7f82 | 562 | if (ret < 0) |
843fcf35 | 563 | goto join_fail; |
0f7d52f4 | 564 | |
0b246afa | 565 | cur_trans = fs_info->running_transaction; |
a22285a6 YZ |
566 | |
567 | h->transid = cur_trans->transid; | |
568 | h->transaction = cur_trans; | |
d13603ef | 569 | h->root = root; |
2a1eb461 | 570 | h->use_count = 1; |
64b63580 | 571 | h->fs_info = root->fs_info; |
7174109c | 572 | |
a698d075 | 573 | h->type = type; |
d9a0540a | 574 | h->can_flush_pending_bgs = true; |
ea658bad | 575 | INIT_LIST_HEAD(&h->new_bgs); |
b7ec40d7 | 576 | |
a22285a6 | 577 | smp_mb(); |
4a9d8bde | 578 | if (cur_trans->state >= TRANS_STATE_BLOCKED && |
2ff7e61e | 579 | may_wait_transaction(fs_info, type)) { |
abdd2e80 | 580 | current->journal_info = h; |
3a45bb20 | 581 | btrfs_commit_transaction(h); |
a22285a6 YZ |
582 | goto again; |
583 | } | |
584 | ||
b5009945 | 585 | if (num_bytes) { |
0b246afa | 586 | trace_btrfs_space_reservation(fs_info, "transaction", |
2bcc0328 | 587 | h->transid, num_bytes, 1); |
0b246afa | 588 | h->block_rsv = &fs_info->trans_block_rsv; |
b5009945 | 589 | h->bytes_reserved = num_bytes; |
20dd2cbf | 590 | h->reloc_reserved = reloc_reserved; |
a22285a6 | 591 | } |
9ed74f2d | 592 | |
2a1eb461 | 593 | got_it: |
a4abeea4 | 594 | btrfs_record_root_in_trans(h, root); |
a22285a6 YZ |
595 | |
596 | if (!current->journal_info && type != TRANS_USERSPACE) | |
597 | current->journal_info = h; | |
79154b1b | 598 | return h; |
843fcf35 MX |
599 | |
600 | join_fail: | |
0860adfd | 601 | if (type & __TRANS_FREEZABLE) |
0b246afa | 602 | sb_end_intwrite(fs_info->sb); |
843fcf35 MX |
603 | kmem_cache_free(btrfs_trans_handle_cachep, h); |
604 | alloc_fail: | |
605 | if (num_bytes) | |
2ff7e61e | 606 | btrfs_block_rsv_release(fs_info, &fs_info->trans_block_rsv, |
843fcf35 MX |
607 | num_bytes); |
608 | reserve_fail: | |
7174109c | 609 | btrfs_qgroup_free_meta(root, qgroup_reserved); |
843fcf35 | 610 | return ERR_PTR(ret); |
79154b1b CM |
611 | } |
612 | ||
f9295749 | 613 | struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, |
5aed1dd8 | 614 | unsigned int num_items) |
f9295749 | 615 | { |
08e007d2 | 616 | return start_transaction(root, num_items, TRANS_START, |
003d7c59 | 617 | BTRFS_RESERVE_FLUSH_ALL, true); |
f9295749 | 618 | } |
003d7c59 | 619 | |
8eab77ff FM |
620 | struct btrfs_trans_handle *btrfs_start_transaction_fallback_global_rsv( |
621 | struct btrfs_root *root, | |
622 | unsigned int num_items, | |
623 | int min_factor) | |
624 | { | |
0b246afa | 625 | struct btrfs_fs_info *fs_info = root->fs_info; |
8eab77ff FM |
626 | struct btrfs_trans_handle *trans; |
627 | u64 num_bytes; | |
628 | int ret; | |
629 | ||
003d7c59 JM |
630 | /* |
631 | * We have two callers: unlink and block group removal. The | |
632 | * former should succeed even if we will temporarily exceed | |
633 | * quota and the latter operates on the extent root so | |
634 | * qgroup enforcement is ignored anyway. | |
635 | */ | |
636 | trans = start_transaction(root, num_items, TRANS_START, | |
637 | BTRFS_RESERVE_FLUSH_ALL, false); | |
8eab77ff FM |
638 | if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC) |
639 | return trans; | |
640 | ||
641 | trans = btrfs_start_transaction(root, 0); | |
642 | if (IS_ERR(trans)) | |
643 | return trans; | |
644 | ||
0b246afa JM |
645 | num_bytes = btrfs_calc_trans_metadata_size(fs_info, num_items); |
646 | ret = btrfs_cond_migrate_bytes(fs_info, &fs_info->trans_block_rsv, | |
647 | num_bytes, min_factor); | |
8eab77ff | 648 | if (ret) { |
3a45bb20 | 649 | btrfs_end_transaction(trans); |
8eab77ff FM |
650 | return ERR_PTR(ret); |
651 | } | |
652 | ||
0b246afa | 653 | trans->block_rsv = &fs_info->trans_block_rsv; |
8eab77ff | 654 | trans->bytes_reserved = num_bytes; |
0b246afa | 655 | trace_btrfs_space_reservation(fs_info, "transaction", |
88d3a5aa | 656 | trans->transid, num_bytes, 1); |
8eab77ff FM |
657 | |
658 | return trans; | |
659 | } | |
8407aa46 | 660 | |
08e007d2 | 661 | struct btrfs_trans_handle *btrfs_start_transaction_lflush( |
5aed1dd8 AM |
662 | struct btrfs_root *root, |
663 | unsigned int num_items) | |
8407aa46 | 664 | { |
08e007d2 | 665 | return start_transaction(root, num_items, TRANS_START, |
003d7c59 | 666 | BTRFS_RESERVE_FLUSH_LIMIT, true); |
8407aa46 MX |
667 | } |
668 | ||
7a7eaa40 | 669 | struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root) |
f9295749 | 670 | { |
003d7c59 JM |
671 | return start_transaction(root, 0, TRANS_JOIN, BTRFS_RESERVE_NO_FLUSH, |
672 | true); | |
f9295749 CM |
673 | } |
674 | ||
7a7eaa40 | 675 | struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root) |
0af3d00b | 676 | { |
575a75d6 | 677 | return start_transaction(root, 0, TRANS_JOIN_NOLOCK, |
003d7c59 | 678 | BTRFS_RESERVE_NO_FLUSH, true); |
0af3d00b JB |
679 | } |
680 | ||
7a7eaa40 | 681 | struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root) |
9ca9ee09 | 682 | { |
575a75d6 | 683 | return start_transaction(root, 0, TRANS_USERSPACE, |
003d7c59 | 684 | BTRFS_RESERVE_NO_FLUSH, true); |
9ca9ee09 SW |
685 | } |
686 | ||
d4edf39b MX |
687 | /* |
688 | * btrfs_attach_transaction() - catch the running transaction | |
689 | * | |
690 | * It is used when we want to commit the current the transaction, but | |
691 | * don't want to start a new one. | |
692 | * | |
693 | * Note: If this function return -ENOENT, it just means there is no | |
694 | * running transaction. But it is possible that the inactive transaction | |
695 | * is still in the memory, not fully on disk. If you hope there is no | |
696 | * inactive transaction in the fs when -ENOENT is returned, you should | |
697 | * invoke | |
698 | * btrfs_attach_transaction_barrier() | |
699 | */ | |
354aa0fb | 700 | struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root) |
60376ce4 | 701 | { |
575a75d6 | 702 | return start_transaction(root, 0, TRANS_ATTACH, |
003d7c59 | 703 | BTRFS_RESERVE_NO_FLUSH, true); |
60376ce4 JB |
704 | } |
705 | ||
d4edf39b | 706 | /* |
90b6d283 | 707 | * btrfs_attach_transaction_barrier() - catch the running transaction |
d4edf39b MX |
708 | * |
709 | * It is similar to the above function, the differentia is this one | |
710 | * will wait for all the inactive transactions until they fully | |
711 | * complete. | |
712 | */ | |
713 | struct btrfs_trans_handle * | |
714 | btrfs_attach_transaction_barrier(struct btrfs_root *root) | |
715 | { | |
716 | struct btrfs_trans_handle *trans; | |
717 | ||
575a75d6 | 718 | trans = start_transaction(root, 0, TRANS_ATTACH, |
003d7c59 | 719 | BTRFS_RESERVE_NO_FLUSH, true); |
d4edf39b | 720 | if (IS_ERR(trans) && PTR_ERR(trans) == -ENOENT) |
2ff7e61e | 721 | btrfs_wait_for_commit(root->fs_info, 0); |
d4edf39b MX |
722 | |
723 | return trans; | |
724 | } | |
725 | ||
d352ac68 | 726 | /* wait for a transaction commit to be fully complete */ |
2ff7e61e | 727 | static noinline void wait_for_commit(struct btrfs_transaction *commit) |
89ce8a63 | 728 | { |
4a9d8bde | 729 | wait_event(commit->commit_wait, commit->state == TRANS_STATE_COMPLETED); |
89ce8a63 CM |
730 | } |
731 | ||
2ff7e61e | 732 | int btrfs_wait_for_commit(struct btrfs_fs_info *fs_info, u64 transid) |
46204592 SW |
733 | { |
734 | struct btrfs_transaction *cur_trans = NULL, *t; | |
8cd2807f | 735 | int ret = 0; |
46204592 | 736 | |
46204592 | 737 | if (transid) { |
0b246afa | 738 | if (transid <= fs_info->last_trans_committed) |
a4abeea4 | 739 | goto out; |
46204592 SW |
740 | |
741 | /* find specified transaction */ | |
0b246afa JM |
742 | spin_lock(&fs_info->trans_lock); |
743 | list_for_each_entry(t, &fs_info->trans_list, list) { | |
46204592 SW |
744 | if (t->transid == transid) { |
745 | cur_trans = t; | |
9b64f57d | 746 | refcount_inc(&cur_trans->use_count); |
8cd2807f | 747 | ret = 0; |
46204592 SW |
748 | break; |
749 | } | |
8cd2807f MX |
750 | if (t->transid > transid) { |
751 | ret = 0; | |
46204592 | 752 | break; |
8cd2807f | 753 | } |
46204592 | 754 | } |
0b246afa | 755 | spin_unlock(&fs_info->trans_lock); |
42383020 SW |
756 | |
757 | /* | |
758 | * The specified transaction doesn't exist, or we | |
759 | * raced with btrfs_commit_transaction | |
760 | */ | |
761 | if (!cur_trans) { | |
0b246afa | 762 | if (transid > fs_info->last_trans_committed) |
42383020 | 763 | ret = -EINVAL; |
8cd2807f | 764 | goto out; |
42383020 | 765 | } |
46204592 SW |
766 | } else { |
767 | /* find newest transaction that is committing | committed */ | |
0b246afa JM |
768 | spin_lock(&fs_info->trans_lock); |
769 | list_for_each_entry_reverse(t, &fs_info->trans_list, | |
46204592 | 770 | list) { |
4a9d8bde MX |
771 | if (t->state >= TRANS_STATE_COMMIT_START) { |
772 | if (t->state == TRANS_STATE_COMPLETED) | |
3473f3c0 | 773 | break; |
46204592 | 774 | cur_trans = t; |
9b64f57d | 775 | refcount_inc(&cur_trans->use_count); |
46204592 SW |
776 | break; |
777 | } | |
778 | } | |
0b246afa | 779 | spin_unlock(&fs_info->trans_lock); |
46204592 | 780 | if (!cur_trans) |
a4abeea4 | 781 | goto out; /* nothing committing|committed */ |
46204592 SW |
782 | } |
783 | ||
2ff7e61e | 784 | wait_for_commit(cur_trans); |
724e2315 | 785 | btrfs_put_transaction(cur_trans); |
a4abeea4 | 786 | out: |
46204592 SW |
787 | return ret; |
788 | } | |
789 | ||
2ff7e61e | 790 | void btrfs_throttle(struct btrfs_fs_info *fs_info) |
37d1aeee | 791 | { |
0b246afa | 792 | if (!atomic_read(&fs_info->open_ioctl_trans)) |
2ff7e61e | 793 | wait_current_trans(fs_info); |
37d1aeee CM |
794 | } |
795 | ||
2ff7e61e | 796 | static int should_end_transaction(struct btrfs_trans_handle *trans) |
8929ecfa | 797 | { |
2ff7e61e | 798 | struct btrfs_fs_info *fs_info = trans->fs_info; |
0b246afa JM |
799 | |
800 | if (fs_info->global_block_rsv.space_info->full && | |
2ff7e61e | 801 | btrfs_check_space_for_delayed_refs(trans, fs_info)) |
1be41b78 | 802 | return 1; |
36ba022a | 803 | |
2ff7e61e | 804 | return !!btrfs_block_rsv_check(&fs_info->global_block_rsv, 5); |
8929ecfa YZ |
805 | } |
806 | ||
3a45bb20 | 807 | int btrfs_should_end_transaction(struct btrfs_trans_handle *trans) |
8929ecfa YZ |
808 | { |
809 | struct btrfs_transaction *cur_trans = trans->transaction; | |
3a45bb20 | 810 | struct btrfs_fs_info *fs_info = trans->fs_info; |
8929ecfa | 811 | int updates; |
49b25e05 | 812 | int err; |
8929ecfa | 813 | |
a4abeea4 | 814 | smp_mb(); |
4a9d8bde MX |
815 | if (cur_trans->state >= TRANS_STATE_BLOCKED || |
816 | cur_trans->delayed_refs.flushing) | |
8929ecfa YZ |
817 | return 1; |
818 | ||
819 | updates = trans->delayed_ref_updates; | |
820 | trans->delayed_ref_updates = 0; | |
49b25e05 | 821 | if (updates) { |
2ff7e61e | 822 | err = btrfs_run_delayed_refs(trans, fs_info, updates * 2); |
49b25e05 JM |
823 | if (err) /* Error code will also eval true */ |
824 | return err; | |
825 | } | |
8929ecfa | 826 | |
2ff7e61e | 827 | return should_end_transaction(trans); |
8929ecfa YZ |
828 | } |
829 | ||
89ce8a63 | 830 | static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, |
3a45bb20 | 831 | int throttle) |
79154b1b | 832 | { |
3a45bb20 | 833 | struct btrfs_fs_info *info = trans->fs_info; |
8929ecfa | 834 | struct btrfs_transaction *cur_trans = trans->transaction; |
31b9655f | 835 | u64 transid = trans->transid; |
1be41b78 | 836 | unsigned long cur = trans->delayed_ref_updates; |
a698d075 | 837 | int lock = (trans->type != TRANS_JOIN_NOLOCK); |
4edc2ca3 | 838 | int err = 0; |
a79b7d4b | 839 | int must_run_delayed_refs = 0; |
c3e69d58 | 840 | |
3bbb24b2 JB |
841 | if (trans->use_count > 1) { |
842 | trans->use_count--; | |
2a1eb461 JB |
843 | trans->block_rsv = trans->orig_rsv; |
844 | return 0; | |
845 | } | |
846 | ||
2ff7e61e | 847 | btrfs_trans_release_metadata(trans, info); |
4c13d758 | 848 | trans->block_rsv = NULL; |
c5567237 | 849 | |
ea658bad | 850 | if (!list_empty(&trans->new_bgs)) |
2ff7e61e | 851 | btrfs_create_pending_block_groups(trans, info); |
ea658bad | 852 | |
1be41b78 | 853 | trans->delayed_ref_updates = 0; |
a79b7d4b CM |
854 | if (!trans->sync) { |
855 | must_run_delayed_refs = | |
2ff7e61e | 856 | btrfs_should_throttle_delayed_refs(trans, info); |
0a2b2a84 | 857 | cur = max_t(unsigned long, cur, 32); |
a79b7d4b CM |
858 | |
859 | /* | |
860 | * don't make the caller wait if they are from a NOLOCK | |
861 | * or ATTACH transaction, it will deadlock with commit | |
862 | */ | |
863 | if (must_run_delayed_refs == 1 && | |
864 | (trans->type & (__TRANS_JOIN_NOLOCK | __TRANS_ATTACH))) | |
865 | must_run_delayed_refs = 2; | |
56bec294 | 866 | } |
bb721703 | 867 | |
2ff7e61e | 868 | btrfs_trans_release_metadata(trans, info); |
0e721106 | 869 | trans->block_rsv = NULL; |
56bec294 | 870 | |
ea658bad | 871 | if (!list_empty(&trans->new_bgs)) |
2ff7e61e | 872 | btrfs_create_pending_block_groups(trans, info); |
ea658bad | 873 | |
4fbcdf66 FM |
874 | btrfs_trans_release_chunk_metadata(trans); |
875 | ||
0b246afa | 876 | if (lock && !atomic_read(&info->open_ioctl_trans) && |
2ff7e61e | 877 | should_end_transaction(trans) && |
20c7bcec | 878 | READ_ONCE(cur_trans->state) == TRANS_STATE_RUNNING) { |
4a9d8bde MX |
879 | spin_lock(&info->trans_lock); |
880 | if (cur_trans->state == TRANS_STATE_RUNNING) | |
881 | cur_trans->state = TRANS_STATE_BLOCKED; | |
882 | spin_unlock(&info->trans_lock); | |
a4abeea4 | 883 | } |
8929ecfa | 884 | |
20c7bcec | 885 | if (lock && READ_ONCE(cur_trans->state) == TRANS_STATE_BLOCKED) { |
3bbb24b2 | 886 | if (throttle) |
3a45bb20 | 887 | return btrfs_commit_transaction(trans); |
3bbb24b2 | 888 | else |
8929ecfa YZ |
889 | wake_up_process(info->transaction_kthread); |
890 | } | |
891 | ||
0860adfd | 892 | if (trans->type & __TRANS_FREEZABLE) |
0b246afa | 893 | sb_end_intwrite(info->sb); |
6df7881a | 894 | |
8929ecfa | 895 | WARN_ON(cur_trans != info->running_transaction); |
13c5a93e JB |
896 | WARN_ON(atomic_read(&cur_trans->num_writers) < 1); |
897 | atomic_dec(&cur_trans->num_writers); | |
0860adfd | 898 | extwriter_counter_dec(cur_trans, trans->type); |
89ce8a63 | 899 | |
a83342aa DS |
900 | /* |
901 | * Make sure counter is updated before we wake up waiters. | |
902 | */ | |
99d16cbc | 903 | smp_mb(); |
79154b1b CM |
904 | if (waitqueue_active(&cur_trans->writer_wait)) |
905 | wake_up(&cur_trans->writer_wait); | |
724e2315 | 906 | btrfs_put_transaction(cur_trans); |
9ed74f2d JB |
907 | |
908 | if (current->journal_info == trans) | |
909 | current->journal_info = NULL; | |
ab78c84d | 910 | |
24bbcf04 | 911 | if (throttle) |
2ff7e61e | 912 | btrfs_run_delayed_iputs(info); |
24bbcf04 | 913 | |
49b25e05 | 914 | if (trans->aborted || |
0b246afa | 915 | test_bit(BTRFS_FS_STATE_ERROR, &info->fs_state)) { |
4e121c06 | 916 | wake_up_process(info->transaction_kthread); |
4edc2ca3 | 917 | err = -EIO; |
4e121c06 | 918 | } |
49b25e05 | 919 | |
4edc2ca3 | 920 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
a79b7d4b | 921 | if (must_run_delayed_refs) { |
2ff7e61e | 922 | btrfs_async_run_delayed_refs(info, cur, transid, |
a79b7d4b CM |
923 | must_run_delayed_refs == 1); |
924 | } | |
4edc2ca3 | 925 | return err; |
79154b1b CM |
926 | } |
927 | ||
3a45bb20 | 928 | int btrfs_end_transaction(struct btrfs_trans_handle *trans) |
89ce8a63 | 929 | { |
3a45bb20 | 930 | return __btrfs_end_transaction(trans, 0); |
89ce8a63 CM |
931 | } |
932 | ||
3a45bb20 | 933 | int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans) |
89ce8a63 | 934 | { |
3a45bb20 | 935 | return __btrfs_end_transaction(trans, 1); |
16cdcec7 MX |
936 | } |
937 | ||
d352ac68 CM |
938 | /* |
939 | * when btree blocks are allocated, they have some corresponding bits set for | |
940 | * them in one of two extent_io trees. This is used to make sure all of | |
690587d1 | 941 | * those extents are sent to disk but does not wait on them |
d352ac68 | 942 | */ |
2ff7e61e | 943 | int btrfs_write_marked_extents(struct btrfs_fs_info *fs_info, |
8cef4e16 | 944 | struct extent_io_tree *dirty_pages, int mark) |
79154b1b | 945 | { |
777e6bd7 | 946 | int err = 0; |
7c4452b9 | 947 | int werr = 0; |
0b246afa | 948 | struct address_space *mapping = fs_info->btree_inode->i_mapping; |
e6138876 | 949 | struct extent_state *cached_state = NULL; |
777e6bd7 | 950 | u64 start = 0; |
5f39d397 | 951 | u64 end; |
7c4452b9 | 952 | |
1728366e | 953 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
e6138876 | 954 | mark, &cached_state)) { |
663dfbb0 FM |
955 | bool wait_writeback = false; |
956 | ||
957 | err = convert_extent_bit(dirty_pages, start, end, | |
958 | EXTENT_NEED_WAIT, | |
210aa277 | 959 | mark, &cached_state); |
663dfbb0 FM |
960 | /* |
961 | * convert_extent_bit can return -ENOMEM, which is most of the | |
962 | * time a temporary error. So when it happens, ignore the error | |
963 | * and wait for writeback of this range to finish - because we | |
964 | * failed to set the bit EXTENT_NEED_WAIT for the range, a call | |
bf89d38f JM |
965 | * to __btrfs_wait_marked_extents() would not know that |
966 | * writeback for this range started and therefore wouldn't | |
967 | * wait for it to finish - we don't want to commit a | |
968 | * superblock that points to btree nodes/leafs for which | |
969 | * writeback hasn't finished yet (and without errors). | |
663dfbb0 FM |
970 | * We cleanup any entries left in the io tree when committing |
971 | * the transaction (through clear_btree_io_tree()). | |
972 | */ | |
973 | if (err == -ENOMEM) { | |
974 | err = 0; | |
975 | wait_writeback = true; | |
976 | } | |
977 | if (!err) | |
978 | err = filemap_fdatawrite_range(mapping, start, end); | |
1728366e JB |
979 | if (err) |
980 | werr = err; | |
663dfbb0 FM |
981 | else if (wait_writeback) |
982 | werr = filemap_fdatawait_range(mapping, start, end); | |
e38e2ed7 | 983 | free_extent_state(cached_state); |
663dfbb0 | 984 | cached_state = NULL; |
1728366e JB |
985 | cond_resched(); |
986 | start = end + 1; | |
7c4452b9 | 987 | } |
690587d1 CM |
988 | return werr; |
989 | } | |
990 | ||
991 | /* | |
992 | * when btree blocks are allocated, they have some corresponding bits set for | |
993 | * them in one of two extent_io trees. This is used to make sure all of | |
994 | * those extents are on disk for transaction or log commit. We wait | |
995 | * on all the pages and clear them from the dirty pages state tree | |
996 | */ | |
bf89d38f JM |
997 | static int __btrfs_wait_marked_extents(struct btrfs_fs_info *fs_info, |
998 | struct extent_io_tree *dirty_pages) | |
690587d1 | 999 | { |
690587d1 CM |
1000 | int err = 0; |
1001 | int werr = 0; | |
0b246afa | 1002 | struct address_space *mapping = fs_info->btree_inode->i_mapping; |
e6138876 | 1003 | struct extent_state *cached_state = NULL; |
690587d1 CM |
1004 | u64 start = 0; |
1005 | u64 end; | |
777e6bd7 | 1006 | |
1728366e | 1007 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
e6138876 | 1008 | EXTENT_NEED_WAIT, &cached_state)) { |
663dfbb0 FM |
1009 | /* |
1010 | * Ignore -ENOMEM errors returned by clear_extent_bit(). | |
1011 | * When committing the transaction, we'll remove any entries | |
1012 | * left in the io tree. For a log commit, we don't remove them | |
1013 | * after committing the log because the tree can be accessed | |
1014 | * concurrently - we do it only at transaction commit time when | |
1015 | * it's safe to do it (through clear_btree_io_tree()). | |
1016 | */ | |
1017 | err = clear_extent_bit(dirty_pages, start, end, | |
1018 | EXTENT_NEED_WAIT, | |
1019 | 0, 0, &cached_state, GFP_NOFS); | |
1020 | if (err == -ENOMEM) | |
1021 | err = 0; | |
1022 | if (!err) | |
1023 | err = filemap_fdatawait_range(mapping, start, end); | |
1728366e JB |
1024 | if (err) |
1025 | werr = err; | |
e38e2ed7 FM |
1026 | free_extent_state(cached_state); |
1027 | cached_state = NULL; | |
1728366e JB |
1028 | cond_resched(); |
1029 | start = end + 1; | |
777e6bd7 | 1030 | } |
7c4452b9 CM |
1031 | if (err) |
1032 | werr = err; | |
bf89d38f JM |
1033 | return werr; |
1034 | } | |
656f30db | 1035 | |
bf89d38f JM |
1036 | int btrfs_wait_extents(struct btrfs_fs_info *fs_info, |
1037 | struct extent_io_tree *dirty_pages) | |
1038 | { | |
1039 | bool errors = false; | |
1040 | int err; | |
656f30db | 1041 | |
bf89d38f JM |
1042 | err = __btrfs_wait_marked_extents(fs_info, dirty_pages); |
1043 | if (test_and_clear_bit(BTRFS_FS_BTREE_ERR, &fs_info->flags)) | |
1044 | errors = true; | |
1045 | ||
1046 | if (errors && !err) | |
1047 | err = -EIO; | |
1048 | return err; | |
1049 | } | |
656f30db | 1050 | |
bf89d38f JM |
1051 | int btrfs_wait_tree_log_extents(struct btrfs_root *log_root, int mark) |
1052 | { | |
1053 | struct btrfs_fs_info *fs_info = log_root->fs_info; | |
1054 | struct extent_io_tree *dirty_pages = &log_root->dirty_log_pages; | |
1055 | bool errors = false; | |
1056 | int err; | |
656f30db | 1057 | |
bf89d38f JM |
1058 | ASSERT(log_root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID); |
1059 | ||
1060 | err = __btrfs_wait_marked_extents(fs_info, dirty_pages); | |
1061 | if ((mark & EXTENT_DIRTY) && | |
1062 | test_and_clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags)) | |
1063 | errors = true; | |
1064 | ||
1065 | if ((mark & EXTENT_NEW) && | |
1066 | test_and_clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags)) | |
1067 | errors = true; | |
1068 | ||
1069 | if (errors && !err) | |
1070 | err = -EIO; | |
1071 | return err; | |
79154b1b CM |
1072 | } |
1073 | ||
690587d1 CM |
1074 | /* |
1075 | * when btree blocks are allocated, they have some corresponding bits set for | |
1076 | * them in one of two extent_io trees. This is used to make sure all of | |
1077 | * those extents are on disk for transaction or log commit | |
1078 | */ | |
bf89d38f | 1079 | static int btrfs_write_and_wait_marked_extents(struct btrfs_fs_info *fs_info, |
8cef4e16 | 1080 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 CM |
1081 | { |
1082 | int ret; | |
1083 | int ret2; | |
c6adc9cc | 1084 | struct blk_plug plug; |
690587d1 | 1085 | |
c6adc9cc | 1086 | blk_start_plug(&plug); |
bf89d38f | 1087 | ret = btrfs_write_marked_extents(fs_info, dirty_pages, mark); |
c6adc9cc | 1088 | blk_finish_plug(&plug); |
bf89d38f | 1089 | ret2 = btrfs_wait_extents(fs_info, dirty_pages); |
bf0da8c1 CM |
1090 | |
1091 | if (ret) | |
1092 | return ret; | |
1093 | if (ret2) | |
1094 | return ret2; | |
1095 | return 0; | |
690587d1 CM |
1096 | } |
1097 | ||
663dfbb0 | 1098 | static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans, |
bf89d38f | 1099 | struct btrfs_fs_info *fs_info) |
d0c803c4 | 1100 | { |
663dfbb0 FM |
1101 | int ret; |
1102 | ||
bf89d38f | 1103 | ret = btrfs_write_and_wait_marked_extents(fs_info, |
8cef4e16 YZ |
1104 | &trans->transaction->dirty_pages, |
1105 | EXTENT_DIRTY); | |
663dfbb0 FM |
1106 | clear_btree_io_tree(&trans->transaction->dirty_pages); |
1107 | ||
1108 | return ret; | |
d0c803c4 CM |
1109 | } |
1110 | ||
d352ac68 CM |
1111 | /* |
1112 | * this is used to update the root pointer in the tree of tree roots. | |
1113 | * | |
1114 | * But, in the case of the extent allocation tree, updating the root | |
1115 | * pointer may allocate blocks which may change the root of the extent | |
1116 | * allocation tree. | |
1117 | * | |
1118 | * So, this loops and repeats and makes sure the cowonly root didn't | |
1119 | * change while the root pointer was being updated in the metadata. | |
1120 | */ | |
0b86a832 CM |
1121 | static int update_cowonly_root(struct btrfs_trans_handle *trans, |
1122 | struct btrfs_root *root) | |
79154b1b CM |
1123 | { |
1124 | int ret; | |
0b86a832 | 1125 | u64 old_root_bytenr; |
86b9f2ec | 1126 | u64 old_root_used; |
0b246afa JM |
1127 | struct btrfs_fs_info *fs_info = root->fs_info; |
1128 | struct btrfs_root *tree_root = fs_info->tree_root; | |
79154b1b | 1129 | |
86b9f2ec | 1130 | old_root_used = btrfs_root_used(&root->root_item); |
56bec294 | 1131 | |
d397712b | 1132 | while (1) { |
0b86a832 | 1133 | old_root_bytenr = btrfs_root_bytenr(&root->root_item); |
86b9f2ec | 1134 | if (old_root_bytenr == root->node->start && |
ea526d18 | 1135 | old_root_used == btrfs_root_used(&root->root_item)) |
79154b1b | 1136 | break; |
87ef2bb4 | 1137 | |
5d4f98a2 | 1138 | btrfs_set_root_node(&root->root_item, root->node); |
79154b1b | 1139 | ret = btrfs_update_root(trans, tree_root, |
0b86a832 CM |
1140 | &root->root_key, |
1141 | &root->root_item); | |
49b25e05 JM |
1142 | if (ret) |
1143 | return ret; | |
56bec294 | 1144 | |
86b9f2ec | 1145 | old_root_used = btrfs_root_used(&root->root_item); |
0b86a832 | 1146 | } |
276e680d | 1147 | |
0b86a832 CM |
1148 | return 0; |
1149 | } | |
1150 | ||
d352ac68 CM |
1151 | /* |
1152 | * update all the cowonly tree roots on disk | |
49b25e05 JM |
1153 | * |
1154 | * The error handling in this function may not be obvious. Any of the | |
1155 | * failures will cause the file system to go offline. We still need | |
1156 | * to clean up the delayed refs. | |
d352ac68 | 1157 | */ |
5d4f98a2 | 1158 | static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans, |
2ff7e61e | 1159 | struct btrfs_fs_info *fs_info) |
0b86a832 | 1160 | { |
ea526d18 | 1161 | struct list_head *dirty_bgs = &trans->transaction->dirty_bgs; |
1bbc621e | 1162 | struct list_head *io_bgs = &trans->transaction->io_bgs; |
0b86a832 | 1163 | struct list_head *next; |
84234f3a | 1164 | struct extent_buffer *eb; |
56bec294 | 1165 | int ret; |
84234f3a YZ |
1166 | |
1167 | eb = btrfs_lock_root_node(fs_info->tree_root); | |
49b25e05 JM |
1168 | ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, |
1169 | 0, &eb); | |
84234f3a YZ |
1170 | btrfs_tree_unlock(eb); |
1171 | free_extent_buffer(eb); | |
0b86a832 | 1172 | |
49b25e05 JM |
1173 | if (ret) |
1174 | return ret; | |
1175 | ||
2ff7e61e | 1176 | ret = btrfs_run_delayed_refs(trans, fs_info, (unsigned long)-1); |
49b25e05 JM |
1177 | if (ret) |
1178 | return ret; | |
87ef2bb4 | 1179 | |
0b246afa | 1180 | ret = btrfs_run_dev_stats(trans, fs_info); |
c16ce190 JB |
1181 | if (ret) |
1182 | return ret; | |
0b246afa | 1183 | ret = btrfs_run_dev_replace(trans, fs_info); |
c16ce190 JB |
1184 | if (ret) |
1185 | return ret; | |
0b246afa | 1186 | ret = btrfs_run_qgroups(trans, fs_info); |
c16ce190 JB |
1187 | if (ret) |
1188 | return ret; | |
546adb0d | 1189 | |
2ff7e61e | 1190 | ret = btrfs_setup_space_cache(trans, fs_info); |
dcdf7f6d JB |
1191 | if (ret) |
1192 | return ret; | |
1193 | ||
546adb0d | 1194 | /* run_qgroups might have added some more refs */ |
2ff7e61e | 1195 | ret = btrfs_run_delayed_refs(trans, fs_info, (unsigned long)-1); |
c16ce190 JB |
1196 | if (ret) |
1197 | return ret; | |
ea526d18 | 1198 | again: |
d397712b | 1199 | while (!list_empty(&fs_info->dirty_cowonly_roots)) { |
2ff7e61e | 1200 | struct btrfs_root *root; |
0b86a832 CM |
1201 | next = fs_info->dirty_cowonly_roots.next; |
1202 | list_del_init(next); | |
1203 | root = list_entry(next, struct btrfs_root, dirty_list); | |
e7070be1 | 1204 | clear_bit(BTRFS_ROOT_DIRTY, &root->state); |
87ef2bb4 | 1205 | |
9e351cc8 JB |
1206 | if (root != fs_info->extent_root) |
1207 | list_add_tail(&root->dirty_list, | |
1208 | &trans->transaction->switch_commits); | |
49b25e05 JM |
1209 | ret = update_cowonly_root(trans, root); |
1210 | if (ret) | |
1211 | return ret; | |
2ff7e61e | 1212 | ret = btrfs_run_delayed_refs(trans, fs_info, (unsigned long)-1); |
ea526d18 JB |
1213 | if (ret) |
1214 | return ret; | |
79154b1b | 1215 | } |
276e680d | 1216 | |
1bbc621e | 1217 | while (!list_empty(dirty_bgs) || !list_empty(io_bgs)) { |
2ff7e61e | 1218 | ret = btrfs_write_dirty_block_groups(trans, fs_info); |
ea526d18 JB |
1219 | if (ret) |
1220 | return ret; | |
2ff7e61e | 1221 | ret = btrfs_run_delayed_refs(trans, fs_info, (unsigned long)-1); |
ea526d18 JB |
1222 | if (ret) |
1223 | return ret; | |
1224 | } | |
1225 | ||
1226 | if (!list_empty(&fs_info->dirty_cowonly_roots)) | |
1227 | goto again; | |
1228 | ||
9e351cc8 JB |
1229 | list_add_tail(&fs_info->extent_root->dirty_list, |
1230 | &trans->transaction->switch_commits); | |
8dabb742 SB |
1231 | btrfs_after_dev_replace_commit(fs_info); |
1232 | ||
79154b1b CM |
1233 | return 0; |
1234 | } | |
1235 | ||
d352ac68 CM |
1236 | /* |
1237 | * dead roots are old snapshots that need to be deleted. This allocates | |
1238 | * a dirty root struct and adds it into the list of dead roots that need to | |
1239 | * be deleted | |
1240 | */ | |
cfad392b | 1241 | void btrfs_add_dead_root(struct btrfs_root *root) |
5eda7b5e | 1242 | { |
0b246afa JM |
1243 | struct btrfs_fs_info *fs_info = root->fs_info; |
1244 | ||
1245 | spin_lock(&fs_info->trans_lock); | |
cfad392b | 1246 | if (list_empty(&root->root_list)) |
0b246afa JM |
1247 | list_add_tail(&root->root_list, &fs_info->dead_roots); |
1248 | spin_unlock(&fs_info->trans_lock); | |
5eda7b5e CM |
1249 | } |
1250 | ||
d352ac68 | 1251 | /* |
5d4f98a2 | 1252 | * update all the cowonly tree roots on disk |
d352ac68 | 1253 | */ |
5d4f98a2 | 1254 | static noinline int commit_fs_roots(struct btrfs_trans_handle *trans, |
5b4aacef | 1255 | struct btrfs_fs_info *fs_info) |
0f7d52f4 | 1256 | { |
0f7d52f4 | 1257 | struct btrfs_root *gang[8]; |
0f7d52f4 CM |
1258 | int i; |
1259 | int ret; | |
54aa1f4d CM |
1260 | int err = 0; |
1261 | ||
a4abeea4 | 1262 | spin_lock(&fs_info->fs_roots_radix_lock); |
d397712b | 1263 | while (1) { |
5d4f98a2 YZ |
1264 | ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, |
1265 | (void **)gang, 0, | |
0f7d52f4 CM |
1266 | ARRAY_SIZE(gang), |
1267 | BTRFS_ROOT_TRANS_TAG); | |
1268 | if (ret == 0) | |
1269 | break; | |
1270 | for (i = 0; i < ret; i++) { | |
5b4aacef | 1271 | struct btrfs_root *root = gang[i]; |
5d4f98a2 YZ |
1272 | radix_tree_tag_clear(&fs_info->fs_roots_radix, |
1273 | (unsigned long)root->root_key.objectid, | |
1274 | BTRFS_ROOT_TRANS_TAG); | |
a4abeea4 | 1275 | spin_unlock(&fs_info->fs_roots_radix_lock); |
31153d81 | 1276 | |
e02119d5 | 1277 | btrfs_free_log(trans, root); |
5d4f98a2 | 1278 | btrfs_update_reloc_root(trans, root); |
d68fc57b | 1279 | btrfs_orphan_commit_root(trans, root); |
bcc63abb | 1280 | |
82d5902d LZ |
1281 | btrfs_save_ino_cache(root, trans); |
1282 | ||
f1ebcc74 | 1283 | /* see comments in should_cow_block() */ |
27cdeb70 | 1284 | clear_bit(BTRFS_ROOT_FORCE_COW, &root->state); |
c7548af6 | 1285 | smp_mb__after_atomic(); |
f1ebcc74 | 1286 | |
978d910d | 1287 | if (root->commit_root != root->node) { |
9e351cc8 JB |
1288 | list_add_tail(&root->dirty_list, |
1289 | &trans->transaction->switch_commits); | |
978d910d YZ |
1290 | btrfs_set_root_node(&root->root_item, |
1291 | root->node); | |
1292 | } | |
5d4f98a2 | 1293 | |
5d4f98a2 | 1294 | err = btrfs_update_root(trans, fs_info->tree_root, |
0f7d52f4 CM |
1295 | &root->root_key, |
1296 | &root->root_item); | |
a4abeea4 | 1297 | spin_lock(&fs_info->fs_roots_radix_lock); |
54aa1f4d CM |
1298 | if (err) |
1299 | break; | |
7174109c | 1300 | btrfs_qgroup_free_meta_all(root); |
0f7d52f4 CM |
1301 | } |
1302 | } | |
a4abeea4 | 1303 | spin_unlock(&fs_info->fs_roots_radix_lock); |
54aa1f4d | 1304 | return err; |
0f7d52f4 CM |
1305 | } |
1306 | ||
d352ac68 | 1307 | /* |
de78b51a ES |
1308 | * defrag a given btree. |
1309 | * Every leaf in the btree is read and defragged. | |
d352ac68 | 1310 | */ |
de78b51a | 1311 | int btrfs_defrag_root(struct btrfs_root *root) |
e9d0b13b CM |
1312 | { |
1313 | struct btrfs_fs_info *info = root->fs_info; | |
e9d0b13b | 1314 | struct btrfs_trans_handle *trans; |
8929ecfa | 1315 | int ret; |
e9d0b13b | 1316 | |
27cdeb70 | 1317 | if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state)) |
e9d0b13b | 1318 | return 0; |
8929ecfa | 1319 | |
6b80053d | 1320 | while (1) { |
8929ecfa YZ |
1321 | trans = btrfs_start_transaction(root, 0); |
1322 | if (IS_ERR(trans)) | |
1323 | return PTR_ERR(trans); | |
1324 | ||
de78b51a | 1325 | ret = btrfs_defrag_leaves(trans, root); |
8929ecfa | 1326 | |
3a45bb20 | 1327 | btrfs_end_transaction(trans); |
2ff7e61e | 1328 | btrfs_btree_balance_dirty(info); |
e9d0b13b CM |
1329 | cond_resched(); |
1330 | ||
ab8d0fc4 | 1331 | if (btrfs_fs_closing(info) || ret != -EAGAIN) |
e9d0b13b | 1332 | break; |
210549eb | 1333 | |
ab8d0fc4 JM |
1334 | if (btrfs_defrag_cancelled(info)) { |
1335 | btrfs_debug(info, "defrag_root cancelled"); | |
210549eb DS |
1336 | ret = -EAGAIN; |
1337 | break; | |
1338 | } | |
e9d0b13b | 1339 | } |
27cdeb70 | 1340 | clear_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state); |
8929ecfa | 1341 | return ret; |
e9d0b13b CM |
1342 | } |
1343 | ||
6426c7ad QW |
1344 | /* |
1345 | * Do all special snapshot related qgroup dirty hack. | |
1346 | * | |
1347 | * Will do all needed qgroup inherit and dirty hack like switch commit | |
1348 | * roots inside one transaction and write all btree into disk, to make | |
1349 | * qgroup works. | |
1350 | */ | |
1351 | static int qgroup_account_snapshot(struct btrfs_trans_handle *trans, | |
1352 | struct btrfs_root *src, | |
1353 | struct btrfs_root *parent, | |
1354 | struct btrfs_qgroup_inherit *inherit, | |
1355 | u64 dst_objectid) | |
1356 | { | |
1357 | struct btrfs_fs_info *fs_info = src->fs_info; | |
1358 | int ret; | |
1359 | ||
1360 | /* | |
1361 | * Save some performance in the case that qgroups are not | |
1362 | * enabled. If this check races with the ioctl, rescan will | |
1363 | * kick in anyway. | |
1364 | */ | |
9ea6e2b5 | 1365 | if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) |
6426c7ad | 1366 | return 0; |
6426c7ad QW |
1367 | |
1368 | /* | |
1369 | * We are going to commit transaction, see btrfs_commit_transaction() | |
1370 | * comment for reason locking tree_log_mutex | |
1371 | */ | |
1372 | mutex_lock(&fs_info->tree_log_mutex); | |
1373 | ||
5b4aacef | 1374 | ret = commit_fs_roots(trans, fs_info); |
6426c7ad QW |
1375 | if (ret) |
1376 | goto out; | |
1377 | ret = btrfs_qgroup_prepare_account_extents(trans, fs_info); | |
1378 | if (ret < 0) | |
1379 | goto out; | |
1380 | ret = btrfs_qgroup_account_extents(trans, fs_info); | |
1381 | if (ret < 0) | |
1382 | goto out; | |
1383 | ||
1384 | /* Now qgroup are all updated, we can inherit it to new qgroups */ | |
1385 | ret = btrfs_qgroup_inherit(trans, fs_info, | |
1386 | src->root_key.objectid, dst_objectid, | |
1387 | inherit); | |
1388 | if (ret < 0) | |
1389 | goto out; | |
1390 | ||
1391 | /* | |
1392 | * Now we do a simplified commit transaction, which will: | |
1393 | * 1) commit all subvolume and extent tree | |
1394 | * To ensure all subvolume and extent tree have a valid | |
1395 | * commit_root to accounting later insert_dir_item() | |
1396 | * 2) write all btree blocks onto disk | |
1397 | * This is to make sure later btree modification will be cowed | |
1398 | * Or commit_root can be populated and cause wrong qgroup numbers | |
1399 | * In this simplified commit, we don't really care about other trees | |
1400 | * like chunk and root tree, as they won't affect qgroup. | |
1401 | * And we don't write super to avoid half committed status. | |
1402 | */ | |
2ff7e61e | 1403 | ret = commit_cowonly_roots(trans, fs_info); |
6426c7ad QW |
1404 | if (ret) |
1405 | goto out; | |
1406 | switch_commit_roots(trans->transaction, fs_info); | |
bf89d38f | 1407 | ret = btrfs_write_and_wait_transaction(trans, fs_info); |
6426c7ad | 1408 | if (ret) |
f7af3934 | 1409 | btrfs_handle_fs_error(fs_info, ret, |
6426c7ad QW |
1410 | "Error while writing out transaction for qgroup"); |
1411 | ||
1412 | out: | |
1413 | mutex_unlock(&fs_info->tree_log_mutex); | |
1414 | ||
1415 | /* | |
1416 | * Force parent root to be updated, as we recorded it before so its | |
1417 | * last_trans == cur_transid. | |
1418 | * Or it won't be committed again onto disk after later | |
1419 | * insert_dir_item() | |
1420 | */ | |
1421 | if (!ret) | |
1422 | record_root_in_trans(trans, parent, 1); | |
1423 | return ret; | |
1424 | } | |
1425 | ||
d352ac68 CM |
1426 | /* |
1427 | * new snapshots need to be created at a very specific time in the | |
aec8030a MX |
1428 | * transaction commit. This does the actual creation. |
1429 | * | |
1430 | * Note: | |
1431 | * If the error which may affect the commitment of the current transaction | |
1432 | * happens, we should return the error number. If the error which just affect | |
1433 | * the creation of the pending snapshots, just return 0. | |
d352ac68 | 1434 | */ |
80b6794d | 1435 | static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, |
3063d29f CM |
1436 | struct btrfs_fs_info *fs_info, |
1437 | struct btrfs_pending_snapshot *pending) | |
1438 | { | |
1439 | struct btrfs_key key; | |
80b6794d | 1440 | struct btrfs_root_item *new_root_item; |
3063d29f CM |
1441 | struct btrfs_root *tree_root = fs_info->tree_root; |
1442 | struct btrfs_root *root = pending->root; | |
6bdb72de | 1443 | struct btrfs_root *parent_root; |
98c9942a | 1444 | struct btrfs_block_rsv *rsv; |
6bdb72de | 1445 | struct inode *parent_inode; |
42874b3d MX |
1446 | struct btrfs_path *path; |
1447 | struct btrfs_dir_item *dir_item; | |
a22285a6 | 1448 | struct dentry *dentry; |
3063d29f | 1449 | struct extent_buffer *tmp; |
925baedd | 1450 | struct extent_buffer *old; |
04b285f3 | 1451 | struct timespec cur_time; |
aec8030a | 1452 | int ret = 0; |
d68fc57b | 1453 | u64 to_reserve = 0; |
6bdb72de | 1454 | u64 index = 0; |
a22285a6 | 1455 | u64 objectid; |
b83cc969 | 1456 | u64 root_flags; |
8ea05e3a | 1457 | uuid_le new_uuid; |
3063d29f | 1458 | |
8546b570 DS |
1459 | ASSERT(pending->path); |
1460 | path = pending->path; | |
42874b3d | 1461 | |
b0c0ea63 DS |
1462 | ASSERT(pending->root_item); |
1463 | new_root_item = pending->root_item; | |
a22285a6 | 1464 | |
aec8030a MX |
1465 | pending->error = btrfs_find_free_objectid(tree_root, &objectid); |
1466 | if (pending->error) | |
6fa9700e | 1467 | goto no_free_objectid; |
3063d29f | 1468 | |
d6726335 QW |
1469 | /* |
1470 | * Make qgroup to skip current new snapshot's qgroupid, as it is | |
1471 | * accounted by later btrfs_qgroup_inherit(). | |
1472 | */ | |
1473 | btrfs_set_skip_qgroup(trans, objectid); | |
1474 | ||
147d256e | 1475 | btrfs_reloc_pre_snapshot(pending, &to_reserve); |
d68fc57b YZ |
1476 | |
1477 | if (to_reserve > 0) { | |
aec8030a MX |
1478 | pending->error = btrfs_block_rsv_add(root, |
1479 | &pending->block_rsv, | |
1480 | to_reserve, | |
1481 | BTRFS_RESERVE_NO_FLUSH); | |
1482 | if (pending->error) | |
d6726335 | 1483 | goto clear_skip_qgroup; |
d68fc57b YZ |
1484 | } |
1485 | ||
3063d29f | 1486 | key.objectid = objectid; |
a22285a6 YZ |
1487 | key.offset = (u64)-1; |
1488 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3063d29f | 1489 | |
6fa9700e | 1490 | rsv = trans->block_rsv; |
a22285a6 | 1491 | trans->block_rsv = &pending->block_rsv; |
2382c5cc | 1492 | trans->bytes_reserved = trans->block_rsv->reserved; |
0b246afa | 1493 | trace_btrfs_space_reservation(fs_info, "transaction", |
88d3a5aa JB |
1494 | trans->transid, |
1495 | trans->bytes_reserved, 1); | |
a22285a6 | 1496 | dentry = pending->dentry; |
e9662f70 | 1497 | parent_inode = pending->dir; |
a22285a6 | 1498 | parent_root = BTRFS_I(parent_inode)->root; |
6426c7ad | 1499 | record_root_in_trans(trans, parent_root, 0); |
a22285a6 | 1500 | |
c2050a45 | 1501 | cur_time = current_time(parent_inode); |
04b285f3 | 1502 | |
3063d29f CM |
1503 | /* |
1504 | * insert the directory item | |
1505 | */ | |
877574e2 | 1506 | ret = btrfs_set_inode_index(BTRFS_I(parent_inode), &index); |
49b25e05 | 1507 | BUG_ON(ret); /* -ENOMEM */ |
42874b3d MX |
1508 | |
1509 | /* check if there is a file/dir which has the same name. */ | |
1510 | dir_item = btrfs_lookup_dir_item(NULL, parent_root, path, | |
4a0cc7ca | 1511 | btrfs_ino(BTRFS_I(parent_inode)), |
42874b3d MX |
1512 | dentry->d_name.name, |
1513 | dentry->d_name.len, 0); | |
1514 | if (dir_item != NULL && !IS_ERR(dir_item)) { | |
fe66a05a | 1515 | pending->error = -EEXIST; |
aec8030a | 1516 | goto dir_item_existed; |
42874b3d MX |
1517 | } else if (IS_ERR(dir_item)) { |
1518 | ret = PTR_ERR(dir_item); | |
66642832 | 1519 | btrfs_abort_transaction(trans, ret); |
8732d44f | 1520 | goto fail; |
79787eaa | 1521 | } |
42874b3d | 1522 | btrfs_release_path(path); |
52c26179 | 1523 | |
e999376f CM |
1524 | /* |
1525 | * pull in the delayed directory update | |
1526 | * and the delayed inode item | |
1527 | * otherwise we corrupt the FS during | |
1528 | * snapshot | |
1529 | */ | |
2ff7e61e | 1530 | ret = btrfs_run_delayed_items(trans, fs_info); |
8732d44f | 1531 | if (ret) { /* Transaction aborted */ |
66642832 | 1532 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1533 | goto fail; |
1534 | } | |
e999376f | 1535 | |
6426c7ad | 1536 | record_root_in_trans(trans, root, 0); |
6bdb72de SW |
1537 | btrfs_set_root_last_snapshot(&root->root_item, trans->transid); |
1538 | memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); | |
08fe4db1 | 1539 | btrfs_check_and_init_root_item(new_root_item); |
6bdb72de | 1540 | |
b83cc969 LZ |
1541 | root_flags = btrfs_root_flags(new_root_item); |
1542 | if (pending->readonly) | |
1543 | root_flags |= BTRFS_ROOT_SUBVOL_RDONLY; | |
1544 | else | |
1545 | root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY; | |
1546 | btrfs_set_root_flags(new_root_item, root_flags); | |
1547 | ||
8ea05e3a AB |
1548 | btrfs_set_root_generation_v2(new_root_item, |
1549 | trans->transid); | |
1550 | uuid_le_gen(&new_uuid); | |
1551 | memcpy(new_root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE); | |
1552 | memcpy(new_root_item->parent_uuid, root->root_item.uuid, | |
1553 | BTRFS_UUID_SIZE); | |
70023da2 SB |
1554 | if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) { |
1555 | memset(new_root_item->received_uuid, 0, | |
1556 | sizeof(new_root_item->received_uuid)); | |
1557 | memset(&new_root_item->stime, 0, sizeof(new_root_item->stime)); | |
1558 | memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime)); | |
1559 | btrfs_set_root_stransid(new_root_item, 0); | |
1560 | btrfs_set_root_rtransid(new_root_item, 0); | |
1561 | } | |
3cae210f QW |
1562 | btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec); |
1563 | btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec); | |
8ea05e3a | 1564 | btrfs_set_root_otransid(new_root_item, trans->transid); |
8ea05e3a | 1565 | |
6bdb72de | 1566 | old = btrfs_lock_root_node(root); |
49b25e05 | 1567 | ret = btrfs_cow_block(trans, root, old, NULL, 0, &old); |
79787eaa JM |
1568 | if (ret) { |
1569 | btrfs_tree_unlock(old); | |
1570 | free_extent_buffer(old); | |
66642832 | 1571 | btrfs_abort_transaction(trans, ret); |
8732d44f | 1572 | goto fail; |
79787eaa | 1573 | } |
49b25e05 | 1574 | |
6bdb72de SW |
1575 | btrfs_set_lock_blocking(old); |
1576 | ||
49b25e05 | 1577 | ret = btrfs_copy_root(trans, root, old, &tmp, objectid); |
79787eaa | 1578 | /* clean up in any case */ |
6bdb72de SW |
1579 | btrfs_tree_unlock(old); |
1580 | free_extent_buffer(old); | |
8732d44f | 1581 | if (ret) { |
66642832 | 1582 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1583 | goto fail; |
1584 | } | |
f1ebcc74 | 1585 | /* see comments in should_cow_block() */ |
27cdeb70 | 1586 | set_bit(BTRFS_ROOT_FORCE_COW, &root->state); |
f1ebcc74 LB |
1587 | smp_wmb(); |
1588 | ||
6bdb72de | 1589 | btrfs_set_root_node(new_root_item, tmp); |
a22285a6 YZ |
1590 | /* record when the snapshot was created in key.offset */ |
1591 | key.offset = trans->transid; | |
1592 | ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); | |
6bdb72de SW |
1593 | btrfs_tree_unlock(tmp); |
1594 | free_extent_buffer(tmp); | |
8732d44f | 1595 | if (ret) { |
66642832 | 1596 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1597 | goto fail; |
1598 | } | |
6bdb72de | 1599 | |
a22285a6 YZ |
1600 | /* |
1601 | * insert root back/forward references | |
1602 | */ | |
6bccf3ab | 1603 | ret = btrfs_add_root_ref(trans, fs_info, objectid, |
0660b5af | 1604 | parent_root->root_key.objectid, |
4a0cc7ca | 1605 | btrfs_ino(BTRFS_I(parent_inode)), index, |
a22285a6 | 1606 | dentry->d_name.name, dentry->d_name.len); |
8732d44f | 1607 | if (ret) { |
66642832 | 1608 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1609 | goto fail; |
1610 | } | |
0660b5af | 1611 | |
a22285a6 | 1612 | key.offset = (u64)-1; |
0b246afa | 1613 | pending->snap = btrfs_read_fs_root_no_name(fs_info, &key); |
79787eaa JM |
1614 | if (IS_ERR(pending->snap)) { |
1615 | ret = PTR_ERR(pending->snap); | |
66642832 | 1616 | btrfs_abort_transaction(trans, ret); |
8732d44f | 1617 | goto fail; |
79787eaa | 1618 | } |
d68fc57b | 1619 | |
49b25e05 | 1620 | ret = btrfs_reloc_post_snapshot(trans, pending); |
8732d44f | 1621 | if (ret) { |
66642832 | 1622 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1623 | goto fail; |
1624 | } | |
361048f5 | 1625 | |
2ff7e61e | 1626 | ret = btrfs_run_delayed_refs(trans, fs_info, (unsigned long)-1); |
8732d44f | 1627 | if (ret) { |
66642832 | 1628 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1629 | goto fail; |
1630 | } | |
42874b3d | 1631 | |
6426c7ad QW |
1632 | /* |
1633 | * Do special qgroup accounting for snapshot, as we do some qgroup | |
1634 | * snapshot hack to do fast snapshot. | |
1635 | * To co-operate with that hack, we do hack again. | |
1636 | * Or snapshot will be greatly slowed down by a subtree qgroup rescan | |
1637 | */ | |
1638 | ret = qgroup_account_snapshot(trans, root, parent_root, | |
1639 | pending->inherit, objectid); | |
1640 | if (ret < 0) | |
1641 | goto fail; | |
1642 | ||
42874b3d MX |
1643 | ret = btrfs_insert_dir_item(trans, parent_root, |
1644 | dentry->d_name.name, dentry->d_name.len, | |
8e7611cf | 1645 | BTRFS_I(parent_inode), &key, |
42874b3d MX |
1646 | BTRFS_FT_DIR, index); |
1647 | /* We have check then name at the beginning, so it is impossible. */ | |
9c52057c | 1648 | BUG_ON(ret == -EEXIST || ret == -EOVERFLOW); |
8732d44f | 1649 | if (ret) { |
66642832 | 1650 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1651 | goto fail; |
1652 | } | |
42874b3d | 1653 | |
6ef06d27 | 1654 | btrfs_i_size_write(BTRFS_I(parent_inode), parent_inode->i_size + |
42874b3d | 1655 | dentry->d_name.len * 2); |
04b285f3 | 1656 | parent_inode->i_mtime = parent_inode->i_ctime = |
c2050a45 | 1657 | current_time(parent_inode); |
be6aef60 | 1658 | ret = btrfs_update_inode_fallback(trans, parent_root, parent_inode); |
dd5f9615 | 1659 | if (ret) { |
66642832 | 1660 | btrfs_abort_transaction(trans, ret); |
dd5f9615 SB |
1661 | goto fail; |
1662 | } | |
6bccf3ab | 1663 | ret = btrfs_uuid_tree_add(trans, fs_info, new_uuid.b, |
dd5f9615 SB |
1664 | BTRFS_UUID_KEY_SUBVOL, objectid); |
1665 | if (ret) { | |
66642832 | 1666 | btrfs_abort_transaction(trans, ret); |
dd5f9615 SB |
1667 | goto fail; |
1668 | } | |
1669 | if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) { | |
6bccf3ab | 1670 | ret = btrfs_uuid_tree_add(trans, fs_info, |
dd5f9615 SB |
1671 | new_root_item->received_uuid, |
1672 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, | |
1673 | objectid); | |
1674 | if (ret && ret != -EEXIST) { | |
66642832 | 1675 | btrfs_abort_transaction(trans, ret); |
dd5f9615 SB |
1676 | goto fail; |
1677 | } | |
1678 | } | |
d6726335 | 1679 | |
2ff7e61e | 1680 | ret = btrfs_run_delayed_refs(trans, fs_info, (unsigned long)-1); |
d6726335 | 1681 | if (ret) { |
66642832 | 1682 | btrfs_abort_transaction(trans, ret); |
d6726335 QW |
1683 | goto fail; |
1684 | } | |
1685 | ||
3063d29f | 1686 | fail: |
aec8030a MX |
1687 | pending->error = ret; |
1688 | dir_item_existed: | |
98c9942a | 1689 | trans->block_rsv = rsv; |
2382c5cc | 1690 | trans->bytes_reserved = 0; |
d6726335 QW |
1691 | clear_skip_qgroup: |
1692 | btrfs_clear_skip_qgroup(trans); | |
6fa9700e MX |
1693 | no_free_objectid: |
1694 | kfree(new_root_item); | |
b0c0ea63 | 1695 | pending->root_item = NULL; |
42874b3d | 1696 | btrfs_free_path(path); |
8546b570 DS |
1697 | pending->path = NULL; |
1698 | ||
49b25e05 | 1699 | return ret; |
3063d29f CM |
1700 | } |
1701 | ||
d352ac68 CM |
1702 | /* |
1703 | * create all the snapshots we've scheduled for creation | |
1704 | */ | |
80b6794d CM |
1705 | static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans, |
1706 | struct btrfs_fs_info *fs_info) | |
3de4586c | 1707 | { |
aec8030a | 1708 | struct btrfs_pending_snapshot *pending, *next; |
3de4586c | 1709 | struct list_head *head = &trans->transaction->pending_snapshots; |
aec8030a | 1710 | int ret = 0; |
3de4586c | 1711 | |
aec8030a MX |
1712 | list_for_each_entry_safe(pending, next, head, list) { |
1713 | list_del(&pending->list); | |
1714 | ret = create_pending_snapshot(trans, fs_info, pending); | |
1715 | if (ret) | |
1716 | break; | |
1717 | } | |
1718 | return ret; | |
3de4586c CM |
1719 | } |
1720 | ||
2ff7e61e | 1721 | static void update_super_roots(struct btrfs_fs_info *fs_info) |
5d4f98a2 YZ |
1722 | { |
1723 | struct btrfs_root_item *root_item; | |
1724 | struct btrfs_super_block *super; | |
1725 | ||
0b246afa | 1726 | super = fs_info->super_copy; |
5d4f98a2 | 1727 | |
0b246afa | 1728 | root_item = &fs_info->chunk_root->root_item; |
5d4f98a2 YZ |
1729 | super->chunk_root = root_item->bytenr; |
1730 | super->chunk_root_generation = root_item->generation; | |
1731 | super->chunk_root_level = root_item->level; | |
1732 | ||
0b246afa | 1733 | root_item = &fs_info->tree_root->root_item; |
5d4f98a2 YZ |
1734 | super->root = root_item->bytenr; |
1735 | super->generation = root_item->generation; | |
1736 | super->root_level = root_item->level; | |
0b246afa | 1737 | if (btrfs_test_opt(fs_info, SPACE_CACHE)) |
0af3d00b | 1738 | super->cache_generation = root_item->generation; |
0b246afa | 1739 | if (test_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags)) |
70f80175 | 1740 | super->uuid_tree_generation = root_item->generation; |
5d4f98a2 YZ |
1741 | } |
1742 | ||
f36f3042 CM |
1743 | int btrfs_transaction_in_commit(struct btrfs_fs_info *info) |
1744 | { | |
4a9d8bde | 1745 | struct btrfs_transaction *trans; |
f36f3042 | 1746 | int ret = 0; |
4a9d8bde | 1747 | |
a4abeea4 | 1748 | spin_lock(&info->trans_lock); |
4a9d8bde MX |
1749 | trans = info->running_transaction; |
1750 | if (trans) | |
1751 | ret = (trans->state >= TRANS_STATE_COMMIT_START); | |
a4abeea4 | 1752 | spin_unlock(&info->trans_lock); |
f36f3042 CM |
1753 | return ret; |
1754 | } | |
1755 | ||
8929ecfa YZ |
1756 | int btrfs_transaction_blocked(struct btrfs_fs_info *info) |
1757 | { | |
4a9d8bde | 1758 | struct btrfs_transaction *trans; |
8929ecfa | 1759 | int ret = 0; |
4a9d8bde | 1760 | |
a4abeea4 | 1761 | spin_lock(&info->trans_lock); |
4a9d8bde MX |
1762 | trans = info->running_transaction; |
1763 | if (trans) | |
1764 | ret = is_transaction_blocked(trans); | |
a4abeea4 | 1765 | spin_unlock(&info->trans_lock); |
8929ecfa YZ |
1766 | return ret; |
1767 | } | |
1768 | ||
bb9c12c9 SW |
1769 | /* |
1770 | * wait for the current transaction commit to start and block subsequent | |
1771 | * transaction joins | |
1772 | */ | |
2ff7e61e | 1773 | static void wait_current_trans_commit_start(struct btrfs_fs_info *fs_info, |
bb9c12c9 SW |
1774 | struct btrfs_transaction *trans) |
1775 | { | |
2ff7e61e JM |
1776 | wait_event(fs_info->transaction_blocked_wait, |
1777 | trans->state >= TRANS_STATE_COMMIT_START || trans->aborted); | |
bb9c12c9 SW |
1778 | } |
1779 | ||
1780 | /* | |
1781 | * wait for the current transaction to start and then become unblocked. | |
1782 | * caller holds ref. | |
1783 | */ | |
2ff7e61e JM |
1784 | static void wait_current_trans_commit_start_and_unblock( |
1785 | struct btrfs_fs_info *fs_info, | |
1786 | struct btrfs_transaction *trans) | |
bb9c12c9 | 1787 | { |
2ff7e61e JM |
1788 | wait_event(fs_info->transaction_wait, |
1789 | trans->state >= TRANS_STATE_UNBLOCKED || trans->aborted); | |
bb9c12c9 SW |
1790 | } |
1791 | ||
1792 | /* | |
1793 | * commit transactions asynchronously. once btrfs_commit_transaction_async | |
1794 | * returns, any subsequent transaction will not be allowed to join. | |
1795 | */ | |
1796 | struct btrfs_async_commit { | |
1797 | struct btrfs_trans_handle *newtrans; | |
7892b5af | 1798 | struct work_struct work; |
bb9c12c9 SW |
1799 | }; |
1800 | ||
1801 | static void do_async_commit(struct work_struct *work) | |
1802 | { | |
1803 | struct btrfs_async_commit *ac = | |
7892b5af | 1804 | container_of(work, struct btrfs_async_commit, work); |
bb9c12c9 | 1805 | |
6fc4e354 SW |
1806 | /* |
1807 | * We've got freeze protection passed with the transaction. | |
1808 | * Tell lockdep about it. | |
1809 | */ | |
b1a06a4b | 1810 | if (ac->newtrans->type & __TRANS_FREEZABLE) |
3a45bb20 | 1811 | __sb_writers_acquired(ac->newtrans->fs_info->sb, SB_FREEZE_FS); |
6fc4e354 | 1812 | |
e209db7a SW |
1813 | current->journal_info = ac->newtrans; |
1814 | ||
3a45bb20 | 1815 | btrfs_commit_transaction(ac->newtrans); |
bb9c12c9 SW |
1816 | kfree(ac); |
1817 | } | |
1818 | ||
1819 | int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans, | |
bb9c12c9 SW |
1820 | int wait_for_unblock) |
1821 | { | |
3a45bb20 | 1822 | struct btrfs_fs_info *fs_info = trans->fs_info; |
bb9c12c9 SW |
1823 | struct btrfs_async_commit *ac; |
1824 | struct btrfs_transaction *cur_trans; | |
1825 | ||
1826 | ac = kmalloc(sizeof(*ac), GFP_NOFS); | |
db5b493a TI |
1827 | if (!ac) |
1828 | return -ENOMEM; | |
bb9c12c9 | 1829 | |
7892b5af | 1830 | INIT_WORK(&ac->work, do_async_commit); |
3a45bb20 | 1831 | ac->newtrans = btrfs_join_transaction(trans->root); |
3612b495 TI |
1832 | if (IS_ERR(ac->newtrans)) { |
1833 | int err = PTR_ERR(ac->newtrans); | |
1834 | kfree(ac); | |
1835 | return err; | |
1836 | } | |
bb9c12c9 SW |
1837 | |
1838 | /* take transaction reference */ | |
bb9c12c9 | 1839 | cur_trans = trans->transaction; |
9b64f57d | 1840 | refcount_inc(&cur_trans->use_count); |
bb9c12c9 | 1841 | |
3a45bb20 | 1842 | btrfs_end_transaction(trans); |
6fc4e354 SW |
1843 | |
1844 | /* | |
1845 | * Tell lockdep we've released the freeze rwsem, since the | |
1846 | * async commit thread will be the one to unlock it. | |
1847 | */ | |
b1a06a4b | 1848 | if (ac->newtrans->type & __TRANS_FREEZABLE) |
0b246afa | 1849 | __sb_writers_release(fs_info->sb, SB_FREEZE_FS); |
6fc4e354 | 1850 | |
7892b5af | 1851 | schedule_work(&ac->work); |
bb9c12c9 SW |
1852 | |
1853 | /* wait for transaction to start and unblock */ | |
bb9c12c9 | 1854 | if (wait_for_unblock) |
2ff7e61e | 1855 | wait_current_trans_commit_start_and_unblock(fs_info, cur_trans); |
bb9c12c9 | 1856 | else |
2ff7e61e | 1857 | wait_current_trans_commit_start(fs_info, cur_trans); |
bb9c12c9 | 1858 | |
38e88054 SW |
1859 | if (current->journal_info == trans) |
1860 | current->journal_info = NULL; | |
1861 | ||
724e2315 | 1862 | btrfs_put_transaction(cur_trans); |
bb9c12c9 SW |
1863 | return 0; |
1864 | } | |
1865 | ||
49b25e05 JM |
1866 | |
1867 | static void cleanup_transaction(struct btrfs_trans_handle *trans, | |
7b8b92af | 1868 | struct btrfs_root *root, int err) |
49b25e05 | 1869 | { |
0b246afa | 1870 | struct btrfs_fs_info *fs_info = root->fs_info; |
49b25e05 | 1871 | struct btrfs_transaction *cur_trans = trans->transaction; |
f094ac32 | 1872 | DEFINE_WAIT(wait); |
49b25e05 JM |
1873 | |
1874 | WARN_ON(trans->use_count > 1); | |
1875 | ||
66642832 | 1876 | btrfs_abort_transaction(trans, err); |
7b8b92af | 1877 | |
0b246afa | 1878 | spin_lock(&fs_info->trans_lock); |
66b6135b | 1879 | |
25d8c284 MX |
1880 | /* |
1881 | * If the transaction is removed from the list, it means this | |
1882 | * transaction has been committed successfully, so it is impossible | |
1883 | * to call the cleanup function. | |
1884 | */ | |
1885 | BUG_ON(list_empty(&cur_trans->list)); | |
66b6135b | 1886 | |
49b25e05 | 1887 | list_del_init(&cur_trans->list); |
0b246afa | 1888 | if (cur_trans == fs_info->running_transaction) { |
4a9d8bde | 1889 | cur_trans->state = TRANS_STATE_COMMIT_DOING; |
0b246afa | 1890 | spin_unlock(&fs_info->trans_lock); |
f094ac32 LB |
1891 | wait_event(cur_trans->writer_wait, |
1892 | atomic_read(&cur_trans->num_writers) == 1); | |
1893 | ||
0b246afa | 1894 | spin_lock(&fs_info->trans_lock); |
d7096fc3 | 1895 | } |
0b246afa | 1896 | spin_unlock(&fs_info->trans_lock); |
49b25e05 | 1897 | |
2ff7e61e | 1898 | btrfs_cleanup_one_transaction(trans->transaction, fs_info); |
49b25e05 | 1899 | |
0b246afa JM |
1900 | spin_lock(&fs_info->trans_lock); |
1901 | if (cur_trans == fs_info->running_transaction) | |
1902 | fs_info->running_transaction = NULL; | |
1903 | spin_unlock(&fs_info->trans_lock); | |
4a9d8bde | 1904 | |
e0228285 | 1905 | if (trans->type & __TRANS_FREEZABLE) |
0b246afa | 1906 | sb_end_intwrite(fs_info->sb); |
724e2315 JB |
1907 | btrfs_put_transaction(cur_trans); |
1908 | btrfs_put_transaction(cur_trans); | |
49b25e05 JM |
1909 | |
1910 | trace_btrfs_transaction_commit(root); | |
1911 | ||
49b25e05 JM |
1912 | if (current->journal_info == trans) |
1913 | current->journal_info = NULL; | |
0b246afa | 1914 | btrfs_scrub_cancel(fs_info); |
49b25e05 JM |
1915 | |
1916 | kmem_cache_free(btrfs_trans_handle_cachep, trans); | |
1917 | } | |
1918 | ||
82436617 MX |
1919 | static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info) |
1920 | { | |
3cdde224 | 1921 | if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) |
6c255e67 | 1922 | return btrfs_start_delalloc_roots(fs_info, 1, -1); |
82436617 MX |
1923 | return 0; |
1924 | } | |
1925 | ||
1926 | static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info) | |
1927 | { | |
3cdde224 | 1928 | if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) |
578def7c | 1929 | btrfs_wait_ordered_roots(fs_info, -1, 0, (u64)-1); |
82436617 MX |
1930 | } |
1931 | ||
50d9aa99 | 1932 | static inline void |
161c3549 | 1933 | btrfs_wait_pending_ordered(struct btrfs_transaction *cur_trans) |
50d9aa99 | 1934 | { |
161c3549 JB |
1935 | wait_event(cur_trans->pending_wait, |
1936 | atomic_read(&cur_trans->pending_ordered) == 0); | |
50d9aa99 JB |
1937 | } |
1938 | ||
3a45bb20 | 1939 | int btrfs_commit_transaction(struct btrfs_trans_handle *trans) |
79154b1b | 1940 | { |
3a45bb20 | 1941 | struct btrfs_fs_info *fs_info = trans->fs_info; |
49b25e05 | 1942 | struct btrfs_transaction *cur_trans = trans->transaction; |
8fd17795 | 1943 | struct btrfs_transaction *prev_trans = NULL; |
25287e0a | 1944 | int ret; |
79154b1b | 1945 | |
8d25a086 | 1946 | /* Stop the commit early if ->aborted is set */ |
20c7bcec | 1947 | if (unlikely(READ_ONCE(cur_trans->aborted))) { |
25287e0a | 1948 | ret = cur_trans->aborted; |
3a45bb20 | 1949 | btrfs_end_transaction(trans); |
e4a2bcac | 1950 | return ret; |
25287e0a | 1951 | } |
49b25e05 | 1952 | |
56bec294 CM |
1953 | /* make a pass through all the delayed refs we have so far |
1954 | * any runnings procs may add more while we are here | |
1955 | */ | |
2ff7e61e | 1956 | ret = btrfs_run_delayed_refs(trans, fs_info, 0); |
e4a2bcac | 1957 | if (ret) { |
3a45bb20 | 1958 | btrfs_end_transaction(trans); |
e4a2bcac JB |
1959 | return ret; |
1960 | } | |
56bec294 | 1961 | |
2ff7e61e | 1962 | btrfs_trans_release_metadata(trans, fs_info); |
0e721106 JB |
1963 | trans->block_rsv = NULL; |
1964 | ||
b7ec40d7 | 1965 | cur_trans = trans->transaction; |
49b25e05 | 1966 | |
56bec294 CM |
1967 | /* |
1968 | * set the flushing flag so procs in this transaction have to | |
1969 | * start sending their work down. | |
1970 | */ | |
b7ec40d7 | 1971 | cur_trans->delayed_refs.flushing = 1; |
1be41b78 | 1972 | smp_wmb(); |
56bec294 | 1973 | |
ea658bad | 1974 | if (!list_empty(&trans->new_bgs)) |
2ff7e61e | 1975 | btrfs_create_pending_block_groups(trans, fs_info); |
ea658bad | 1976 | |
2ff7e61e | 1977 | ret = btrfs_run_delayed_refs(trans, fs_info, 0); |
e4a2bcac | 1978 | if (ret) { |
3a45bb20 | 1979 | btrfs_end_transaction(trans); |
e4a2bcac JB |
1980 | return ret; |
1981 | } | |
56bec294 | 1982 | |
3204d33c | 1983 | if (!test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &cur_trans->flags)) { |
1bbc621e CM |
1984 | int run_it = 0; |
1985 | ||
1986 | /* this mutex is also taken before trying to set | |
1987 | * block groups readonly. We need to make sure | |
1988 | * that nobody has set a block group readonly | |
1989 | * after a extents from that block group have been | |
1990 | * allocated for cache files. btrfs_set_block_group_ro | |
1991 | * will wait for the transaction to commit if it | |
3204d33c | 1992 | * finds BTRFS_TRANS_DIRTY_BG_RUN set. |
1bbc621e | 1993 | * |
3204d33c JB |
1994 | * The BTRFS_TRANS_DIRTY_BG_RUN flag is also used to make sure |
1995 | * only one process starts all the block group IO. It wouldn't | |
1bbc621e CM |
1996 | * hurt to have more than one go through, but there's no |
1997 | * real advantage to it either. | |
1998 | */ | |
0b246afa | 1999 | mutex_lock(&fs_info->ro_block_group_mutex); |
3204d33c JB |
2000 | if (!test_and_set_bit(BTRFS_TRANS_DIRTY_BG_RUN, |
2001 | &cur_trans->flags)) | |
1bbc621e | 2002 | run_it = 1; |
0b246afa | 2003 | mutex_unlock(&fs_info->ro_block_group_mutex); |
1bbc621e CM |
2004 | |
2005 | if (run_it) | |
2ff7e61e | 2006 | ret = btrfs_start_dirty_block_groups(trans, fs_info); |
1bbc621e CM |
2007 | } |
2008 | if (ret) { | |
3a45bb20 | 2009 | btrfs_end_transaction(trans); |
1bbc621e CM |
2010 | return ret; |
2011 | } | |
2012 | ||
0b246afa | 2013 | spin_lock(&fs_info->trans_lock); |
4a9d8bde | 2014 | if (cur_trans->state >= TRANS_STATE_COMMIT_START) { |
0b246afa | 2015 | spin_unlock(&fs_info->trans_lock); |
9b64f57d | 2016 | refcount_inc(&cur_trans->use_count); |
3a45bb20 | 2017 | ret = btrfs_end_transaction(trans); |
ccd467d6 | 2018 | |
2ff7e61e | 2019 | wait_for_commit(cur_trans); |
15ee9bc7 | 2020 | |
b4924a0f LB |
2021 | if (unlikely(cur_trans->aborted)) |
2022 | ret = cur_trans->aborted; | |
2023 | ||
724e2315 | 2024 | btrfs_put_transaction(cur_trans); |
15ee9bc7 | 2025 | |
49b25e05 | 2026 | return ret; |
79154b1b | 2027 | } |
4313b399 | 2028 | |
4a9d8bde | 2029 | cur_trans->state = TRANS_STATE_COMMIT_START; |
0b246afa | 2030 | wake_up(&fs_info->transaction_blocked_wait); |
bb9c12c9 | 2031 | |
0b246afa | 2032 | if (cur_trans->list.prev != &fs_info->trans_list) { |
ccd467d6 CM |
2033 | prev_trans = list_entry(cur_trans->list.prev, |
2034 | struct btrfs_transaction, list); | |
4a9d8bde | 2035 | if (prev_trans->state != TRANS_STATE_COMPLETED) { |
9b64f57d | 2036 | refcount_inc(&prev_trans->use_count); |
0b246afa | 2037 | spin_unlock(&fs_info->trans_lock); |
ccd467d6 | 2038 | |
2ff7e61e | 2039 | wait_for_commit(prev_trans); |
1f9b8c8f | 2040 | ret = prev_trans->aborted; |
ccd467d6 | 2041 | |
724e2315 | 2042 | btrfs_put_transaction(prev_trans); |
1f9b8c8f FM |
2043 | if (ret) |
2044 | goto cleanup_transaction; | |
a4abeea4 | 2045 | } else { |
0b246afa | 2046 | spin_unlock(&fs_info->trans_lock); |
ccd467d6 | 2047 | } |
a4abeea4 | 2048 | } else { |
0b246afa | 2049 | spin_unlock(&fs_info->trans_lock); |
ccd467d6 | 2050 | } |
15ee9bc7 | 2051 | |
0860adfd MX |
2052 | extwriter_counter_dec(cur_trans, trans->type); |
2053 | ||
0b246afa | 2054 | ret = btrfs_start_delalloc_flush(fs_info); |
82436617 MX |
2055 | if (ret) |
2056 | goto cleanup_transaction; | |
2057 | ||
2ff7e61e | 2058 | ret = btrfs_run_delayed_items(trans, fs_info); |
581227d0 MX |
2059 | if (ret) |
2060 | goto cleanup_transaction; | |
15ee9bc7 | 2061 | |
581227d0 MX |
2062 | wait_event(cur_trans->writer_wait, |
2063 | extwriter_counter_read(cur_trans) == 0); | |
15ee9bc7 | 2064 | |
581227d0 | 2065 | /* some pending stuffs might be added after the previous flush. */ |
2ff7e61e | 2066 | ret = btrfs_run_delayed_items(trans, fs_info); |
ca469637 MX |
2067 | if (ret) |
2068 | goto cleanup_transaction; | |
2069 | ||
0b246afa | 2070 | btrfs_wait_delalloc_flush(fs_info); |
cb7ab021 | 2071 | |
161c3549 | 2072 | btrfs_wait_pending_ordered(cur_trans); |
50d9aa99 | 2073 | |
2ff7e61e | 2074 | btrfs_scrub_pause(fs_info); |
ed0ca140 JB |
2075 | /* |
2076 | * Ok now we need to make sure to block out any other joins while we | |
2077 | * commit the transaction. We could have started a join before setting | |
4a9d8bde | 2078 | * COMMIT_DOING so make sure to wait for num_writers to == 1 again. |
ed0ca140 | 2079 | */ |
0b246afa | 2080 | spin_lock(&fs_info->trans_lock); |
4a9d8bde | 2081 | cur_trans->state = TRANS_STATE_COMMIT_DOING; |
0b246afa | 2082 | spin_unlock(&fs_info->trans_lock); |
ed0ca140 JB |
2083 | wait_event(cur_trans->writer_wait, |
2084 | atomic_read(&cur_trans->num_writers) == 1); | |
2085 | ||
2cba30f1 | 2086 | /* ->aborted might be set after the previous check, so check it */ |
20c7bcec | 2087 | if (unlikely(READ_ONCE(cur_trans->aborted))) { |
2cba30f1 | 2088 | ret = cur_trans->aborted; |
6cf7f77e | 2089 | goto scrub_continue; |
2cba30f1 | 2090 | } |
7585717f CM |
2091 | /* |
2092 | * the reloc mutex makes sure that we stop | |
2093 | * the balancing code from coming in and moving | |
2094 | * extents around in the middle of the commit | |
2095 | */ | |
0b246afa | 2096 | mutex_lock(&fs_info->reloc_mutex); |
7585717f | 2097 | |
42874b3d MX |
2098 | /* |
2099 | * We needn't worry about the delayed items because we will | |
2100 | * deal with them in create_pending_snapshot(), which is the | |
2101 | * core function of the snapshot creation. | |
2102 | */ | |
0b246afa | 2103 | ret = create_pending_snapshots(trans, fs_info); |
49b25e05 | 2104 | if (ret) { |
0b246afa | 2105 | mutex_unlock(&fs_info->reloc_mutex); |
6cf7f77e | 2106 | goto scrub_continue; |
49b25e05 | 2107 | } |
3063d29f | 2108 | |
42874b3d MX |
2109 | /* |
2110 | * We insert the dir indexes of the snapshots and update the inode | |
2111 | * of the snapshots' parents after the snapshot creation, so there | |
2112 | * are some delayed items which are not dealt with. Now deal with | |
2113 | * them. | |
2114 | * | |
2115 | * We needn't worry that this operation will corrupt the snapshots, | |
2116 | * because all the tree which are snapshoted will be forced to COW | |
2117 | * the nodes and leaves. | |
2118 | */ | |
2ff7e61e | 2119 | ret = btrfs_run_delayed_items(trans, fs_info); |
49b25e05 | 2120 | if (ret) { |
0b246afa | 2121 | mutex_unlock(&fs_info->reloc_mutex); |
6cf7f77e | 2122 | goto scrub_continue; |
49b25e05 | 2123 | } |
16cdcec7 | 2124 | |
2ff7e61e | 2125 | ret = btrfs_run_delayed_refs(trans, fs_info, (unsigned long)-1); |
49b25e05 | 2126 | if (ret) { |
0b246afa | 2127 | mutex_unlock(&fs_info->reloc_mutex); |
6cf7f77e | 2128 | goto scrub_continue; |
49b25e05 | 2129 | } |
56bec294 | 2130 | |
0ed4792a | 2131 | /* Reocrd old roots for later qgroup accounting */ |
0b246afa | 2132 | ret = btrfs_qgroup_prepare_account_extents(trans, fs_info); |
0ed4792a | 2133 | if (ret) { |
0b246afa | 2134 | mutex_unlock(&fs_info->reloc_mutex); |
0ed4792a QW |
2135 | goto scrub_continue; |
2136 | } | |
2137 | ||
e999376f CM |
2138 | /* |
2139 | * make sure none of the code above managed to slip in a | |
2140 | * delayed item | |
2141 | */ | |
ccdf9b30 | 2142 | btrfs_assert_delayed_root_empty(fs_info); |
e999376f | 2143 | |
2c90e5d6 | 2144 | WARN_ON(cur_trans != trans->transaction); |
dc17ff8f | 2145 | |
e02119d5 CM |
2146 | /* btrfs_commit_tree_roots is responsible for getting the |
2147 | * various roots consistent with each other. Every pointer | |
2148 | * in the tree of tree roots has to point to the most up to date | |
2149 | * root for every subvolume and other tree. So, we have to keep | |
2150 | * the tree logging code from jumping in and changing any | |
2151 | * of the trees. | |
2152 | * | |
2153 | * At this point in the commit, there can't be any tree-log | |
2154 | * writers, but a little lower down we drop the trans mutex | |
2155 | * and let new people in. By holding the tree_log_mutex | |
2156 | * from now until after the super is written, we avoid races | |
2157 | * with the tree-log code. | |
2158 | */ | |
0b246afa | 2159 | mutex_lock(&fs_info->tree_log_mutex); |
e02119d5 | 2160 | |
0b246afa | 2161 | ret = commit_fs_roots(trans, fs_info); |
49b25e05 | 2162 | if (ret) { |
0b246afa JM |
2163 | mutex_unlock(&fs_info->tree_log_mutex); |
2164 | mutex_unlock(&fs_info->reloc_mutex); | |
6cf7f77e | 2165 | goto scrub_continue; |
49b25e05 | 2166 | } |
54aa1f4d | 2167 | |
3818aea2 | 2168 | /* |
7e1876ac DS |
2169 | * Since the transaction is done, we can apply the pending changes |
2170 | * before the next transaction. | |
3818aea2 | 2171 | */ |
0b246afa | 2172 | btrfs_apply_pending_changes(fs_info); |
3818aea2 | 2173 | |
5d4f98a2 | 2174 | /* commit_fs_roots gets rid of all the tree log roots, it is now |
e02119d5 CM |
2175 | * safe to free the root of tree log roots |
2176 | */ | |
0b246afa | 2177 | btrfs_free_log_root_tree(trans, fs_info); |
e02119d5 | 2178 | |
0ed4792a QW |
2179 | /* |
2180 | * Since fs roots are all committed, we can get a quite accurate | |
2181 | * new_roots. So let's do quota accounting. | |
2182 | */ | |
0b246afa | 2183 | ret = btrfs_qgroup_account_extents(trans, fs_info); |
0ed4792a | 2184 | if (ret < 0) { |
0b246afa JM |
2185 | mutex_unlock(&fs_info->tree_log_mutex); |
2186 | mutex_unlock(&fs_info->reloc_mutex); | |
0ed4792a QW |
2187 | goto scrub_continue; |
2188 | } | |
2189 | ||
2ff7e61e | 2190 | ret = commit_cowonly_roots(trans, fs_info); |
49b25e05 | 2191 | if (ret) { |
0b246afa JM |
2192 | mutex_unlock(&fs_info->tree_log_mutex); |
2193 | mutex_unlock(&fs_info->reloc_mutex); | |
6cf7f77e | 2194 | goto scrub_continue; |
49b25e05 | 2195 | } |
54aa1f4d | 2196 | |
2cba30f1 MX |
2197 | /* |
2198 | * The tasks which save the space cache and inode cache may also | |
2199 | * update ->aborted, check it. | |
2200 | */ | |
20c7bcec | 2201 | if (unlikely(READ_ONCE(cur_trans->aborted))) { |
2cba30f1 | 2202 | ret = cur_trans->aborted; |
0b246afa JM |
2203 | mutex_unlock(&fs_info->tree_log_mutex); |
2204 | mutex_unlock(&fs_info->reloc_mutex); | |
6cf7f77e | 2205 | goto scrub_continue; |
2cba30f1 MX |
2206 | } |
2207 | ||
8b74c03e | 2208 | btrfs_prepare_extent_commit(fs_info); |
11833d66 | 2209 | |
0b246afa | 2210 | cur_trans = fs_info->running_transaction; |
5d4f98a2 | 2211 | |
0b246afa JM |
2212 | btrfs_set_root_node(&fs_info->tree_root->root_item, |
2213 | fs_info->tree_root->node); | |
2214 | list_add_tail(&fs_info->tree_root->dirty_list, | |
9e351cc8 | 2215 | &cur_trans->switch_commits); |
5d4f98a2 | 2216 | |
0b246afa JM |
2217 | btrfs_set_root_node(&fs_info->chunk_root->root_item, |
2218 | fs_info->chunk_root->node); | |
2219 | list_add_tail(&fs_info->chunk_root->dirty_list, | |
9e351cc8 JB |
2220 | &cur_trans->switch_commits); |
2221 | ||
0b246afa | 2222 | switch_commit_roots(cur_trans, fs_info); |
5d4f98a2 | 2223 | |
ce93ec54 | 2224 | ASSERT(list_empty(&cur_trans->dirty_bgs)); |
1bbc621e | 2225 | ASSERT(list_empty(&cur_trans->io_bgs)); |
2ff7e61e | 2226 | update_super_roots(fs_info); |
e02119d5 | 2227 | |
0b246afa JM |
2228 | btrfs_set_super_log_root(fs_info->super_copy, 0); |
2229 | btrfs_set_super_log_root_level(fs_info->super_copy, 0); | |
2230 | memcpy(fs_info->super_for_commit, fs_info->super_copy, | |
2231 | sizeof(*fs_info->super_copy)); | |
ccd467d6 | 2232 | |
0b246afa | 2233 | btrfs_update_commit_device_size(fs_info); |
2ff7e61e | 2234 | btrfs_update_commit_device_bytes_used(fs_info, cur_trans); |
935e5cc9 | 2235 | |
0b246afa JM |
2236 | clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags); |
2237 | clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags); | |
656f30db | 2238 | |
4fbcdf66 FM |
2239 | btrfs_trans_release_chunk_metadata(trans); |
2240 | ||
0b246afa | 2241 | spin_lock(&fs_info->trans_lock); |
4a9d8bde | 2242 | cur_trans->state = TRANS_STATE_UNBLOCKED; |
0b246afa JM |
2243 | fs_info->running_transaction = NULL; |
2244 | spin_unlock(&fs_info->trans_lock); | |
2245 | mutex_unlock(&fs_info->reloc_mutex); | |
b7ec40d7 | 2246 | |
0b246afa | 2247 | wake_up(&fs_info->transaction_wait); |
e6dcd2dc | 2248 | |
bf89d38f | 2249 | ret = btrfs_write_and_wait_transaction(trans, fs_info); |
49b25e05 | 2250 | if (ret) { |
0b246afa JM |
2251 | btrfs_handle_fs_error(fs_info, ret, |
2252 | "Error while writing out transaction"); | |
2253 | mutex_unlock(&fs_info->tree_log_mutex); | |
6cf7f77e | 2254 | goto scrub_continue; |
49b25e05 JM |
2255 | } |
2256 | ||
eece6a9c | 2257 | ret = write_all_supers(fs_info, 0); |
49b25e05 | 2258 | if (ret) { |
0b246afa | 2259 | mutex_unlock(&fs_info->tree_log_mutex); |
6cf7f77e | 2260 | goto scrub_continue; |
49b25e05 | 2261 | } |
4313b399 | 2262 | |
e02119d5 CM |
2263 | /* |
2264 | * the super is written, we can safely allow the tree-loggers | |
2265 | * to go about their business | |
2266 | */ | |
0b246afa | 2267 | mutex_unlock(&fs_info->tree_log_mutex); |
e02119d5 | 2268 | |
2ff7e61e | 2269 | btrfs_finish_extent_commit(trans, fs_info); |
4313b399 | 2270 | |
3204d33c | 2271 | if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS, &cur_trans->flags)) |
0b246afa | 2272 | btrfs_clear_space_info_full(fs_info); |
13212b54 | 2273 | |
0b246afa | 2274 | fs_info->last_trans_committed = cur_trans->transid; |
4a9d8bde MX |
2275 | /* |
2276 | * We needn't acquire the lock here because there is no other task | |
2277 | * which can change it. | |
2278 | */ | |
2279 | cur_trans->state = TRANS_STATE_COMPLETED; | |
2c90e5d6 | 2280 | wake_up(&cur_trans->commit_wait); |
3de4586c | 2281 | |
0b246afa | 2282 | spin_lock(&fs_info->trans_lock); |
13c5a93e | 2283 | list_del_init(&cur_trans->list); |
0b246afa | 2284 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 | 2285 | |
724e2315 JB |
2286 | btrfs_put_transaction(cur_trans); |
2287 | btrfs_put_transaction(cur_trans); | |
58176a96 | 2288 | |
0860adfd | 2289 | if (trans->type & __TRANS_FREEZABLE) |
0b246afa | 2290 | sb_end_intwrite(fs_info->sb); |
b2b5ef5c | 2291 | |
3a45bb20 | 2292 | trace_btrfs_transaction_commit(trans->root); |
1abe9b8a | 2293 | |
2ff7e61e | 2294 | btrfs_scrub_continue(fs_info); |
a2de733c | 2295 | |
9ed74f2d JB |
2296 | if (current->journal_info == trans) |
2297 | current->journal_info = NULL; | |
2298 | ||
2c90e5d6 | 2299 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
24bbcf04 | 2300 | |
9e7cc91a WX |
2301 | /* |
2302 | * If fs has been frozen, we can not handle delayed iputs, otherwise | |
2303 | * it'll result in deadlock about SB_FREEZE_FS. | |
2304 | */ | |
0b246afa JM |
2305 | if (current != fs_info->transaction_kthread && |
2306 | current != fs_info->cleaner_kthread && !fs_info->fs_frozen) | |
2ff7e61e | 2307 | btrfs_run_delayed_iputs(fs_info); |
24bbcf04 | 2308 | |
79154b1b | 2309 | return ret; |
49b25e05 | 2310 | |
6cf7f77e | 2311 | scrub_continue: |
2ff7e61e | 2312 | btrfs_scrub_continue(fs_info); |
49b25e05 | 2313 | cleanup_transaction: |
2ff7e61e | 2314 | btrfs_trans_release_metadata(trans, fs_info); |
4fbcdf66 | 2315 | btrfs_trans_release_chunk_metadata(trans); |
0e721106 | 2316 | trans->block_rsv = NULL; |
0b246afa | 2317 | btrfs_warn(fs_info, "Skipping commit of aborted transaction."); |
49b25e05 JM |
2318 | if (current->journal_info == trans) |
2319 | current->journal_info = NULL; | |
3a45bb20 | 2320 | cleanup_transaction(trans, trans->root, ret); |
49b25e05 JM |
2321 | |
2322 | return ret; | |
79154b1b CM |
2323 | } |
2324 | ||
d352ac68 | 2325 | /* |
9d1a2a3a DS |
2326 | * return < 0 if error |
2327 | * 0 if there are no more dead_roots at the time of call | |
2328 | * 1 there are more to be processed, call me again | |
2329 | * | |
2330 | * The return value indicates there are certainly more snapshots to delete, but | |
2331 | * if there comes a new one during processing, it may return 0. We don't mind, | |
2332 | * because btrfs_commit_super will poke cleaner thread and it will process it a | |
2333 | * few seconds later. | |
d352ac68 | 2334 | */ |
9d1a2a3a | 2335 | int btrfs_clean_one_deleted_snapshot(struct btrfs_root *root) |
e9d0b13b | 2336 | { |
9d1a2a3a | 2337 | int ret; |
5d4f98a2 YZ |
2338 | struct btrfs_fs_info *fs_info = root->fs_info; |
2339 | ||
a4abeea4 | 2340 | spin_lock(&fs_info->trans_lock); |
9d1a2a3a DS |
2341 | if (list_empty(&fs_info->dead_roots)) { |
2342 | spin_unlock(&fs_info->trans_lock); | |
2343 | return 0; | |
2344 | } | |
2345 | root = list_first_entry(&fs_info->dead_roots, | |
2346 | struct btrfs_root, root_list); | |
cfad392b | 2347 | list_del_init(&root->root_list); |
a4abeea4 | 2348 | spin_unlock(&fs_info->trans_lock); |
e9d0b13b | 2349 | |
ab8d0fc4 | 2350 | btrfs_debug(fs_info, "cleaner removing %llu", root->objectid); |
76dda93c | 2351 | |
9d1a2a3a | 2352 | btrfs_kill_all_delayed_nodes(root); |
16cdcec7 | 2353 | |
9d1a2a3a DS |
2354 | if (btrfs_header_backref_rev(root->node) < |
2355 | BTRFS_MIXED_BACKREF_REV) | |
2356 | ret = btrfs_drop_snapshot(root, NULL, 0, 0); | |
2357 | else | |
2358 | ret = btrfs_drop_snapshot(root, NULL, 1, 0); | |
32471dc2 | 2359 | |
6596a928 | 2360 | return (ret < 0) ? 0 : 1; |
e9d0b13b | 2361 | } |
572d9ab7 DS |
2362 | |
2363 | void btrfs_apply_pending_changes(struct btrfs_fs_info *fs_info) | |
2364 | { | |
2365 | unsigned long prev; | |
2366 | unsigned long bit; | |
2367 | ||
6c9fe14f | 2368 | prev = xchg(&fs_info->pending_changes, 0); |
572d9ab7 DS |
2369 | if (!prev) |
2370 | return; | |
2371 | ||
7e1876ac DS |
2372 | bit = 1 << BTRFS_PENDING_SET_INODE_MAP_CACHE; |
2373 | if (prev & bit) | |
2374 | btrfs_set_opt(fs_info->mount_opt, INODE_MAP_CACHE); | |
2375 | prev &= ~bit; | |
2376 | ||
2377 | bit = 1 << BTRFS_PENDING_CLEAR_INODE_MAP_CACHE; | |
2378 | if (prev & bit) | |
2379 | btrfs_clear_opt(fs_info->mount_opt, INODE_MAP_CACHE); | |
2380 | prev &= ~bit; | |
2381 | ||
d51033d0 DS |
2382 | bit = 1 << BTRFS_PENDING_COMMIT; |
2383 | if (prev & bit) | |
2384 | btrfs_debug(fs_info, "pending commit done"); | |
2385 | prev &= ~bit; | |
2386 | ||
572d9ab7 DS |
2387 | if (prev) |
2388 | btrfs_warn(fs_info, | |
2389 | "unknown pending changes left 0x%lx, ignoring", prev); | |
2390 | } |