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 | } | |
4b5faeac | 96 | kfree(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 | ||
4b5faeac | 231 | cur_trans = kmalloc(sizeof(*cur_trans), GFP_NOFS); |
a4abeea4 JB |
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 | */ |
4b5faeac | 241 | kfree(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); |
4b5faeac | 245 | kfree(cur_trans); |
7b8b92af | 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, |
c6100a4b | 297 | fs_info->btree_inode); |
19ae4e81 JS |
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 | 497 | h = current->journal_info; |
b50fff81 DS |
498 | refcount_inc(&h->use_count); |
499 | WARN_ON(refcount_read(&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; |
b50fff81 | 570 | refcount_set(&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 | |
7a7eaa40 | 661 | struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root) |
f9295749 | 662 | { |
003d7c59 JM |
663 | return start_transaction(root, 0, TRANS_JOIN, BTRFS_RESERVE_NO_FLUSH, |
664 | true); | |
f9295749 CM |
665 | } |
666 | ||
7a7eaa40 | 667 | struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root) |
0af3d00b | 668 | { |
575a75d6 | 669 | return start_transaction(root, 0, TRANS_JOIN_NOLOCK, |
003d7c59 | 670 | BTRFS_RESERVE_NO_FLUSH, true); |
0af3d00b JB |
671 | } |
672 | ||
7a7eaa40 | 673 | struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root) |
9ca9ee09 | 674 | { |
575a75d6 | 675 | return start_transaction(root, 0, TRANS_USERSPACE, |
003d7c59 | 676 | BTRFS_RESERVE_NO_FLUSH, true); |
9ca9ee09 SW |
677 | } |
678 | ||
d4edf39b MX |
679 | /* |
680 | * btrfs_attach_transaction() - catch the running transaction | |
681 | * | |
682 | * It is used when we want to commit the current the transaction, but | |
683 | * don't want to start a new one. | |
684 | * | |
685 | * Note: If this function return -ENOENT, it just means there is no | |
686 | * running transaction. But it is possible that the inactive transaction | |
687 | * is still in the memory, not fully on disk. If you hope there is no | |
688 | * inactive transaction in the fs when -ENOENT is returned, you should | |
689 | * invoke | |
690 | * btrfs_attach_transaction_barrier() | |
691 | */ | |
354aa0fb | 692 | struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root) |
60376ce4 | 693 | { |
575a75d6 | 694 | return start_transaction(root, 0, TRANS_ATTACH, |
003d7c59 | 695 | BTRFS_RESERVE_NO_FLUSH, true); |
60376ce4 JB |
696 | } |
697 | ||
d4edf39b | 698 | /* |
90b6d283 | 699 | * btrfs_attach_transaction_barrier() - catch the running transaction |
d4edf39b MX |
700 | * |
701 | * It is similar to the above function, the differentia is this one | |
702 | * will wait for all the inactive transactions until they fully | |
703 | * complete. | |
704 | */ | |
705 | struct btrfs_trans_handle * | |
706 | btrfs_attach_transaction_barrier(struct btrfs_root *root) | |
707 | { | |
708 | struct btrfs_trans_handle *trans; | |
709 | ||
575a75d6 | 710 | trans = start_transaction(root, 0, TRANS_ATTACH, |
003d7c59 | 711 | BTRFS_RESERVE_NO_FLUSH, true); |
d4edf39b | 712 | if (IS_ERR(trans) && PTR_ERR(trans) == -ENOENT) |
2ff7e61e | 713 | btrfs_wait_for_commit(root->fs_info, 0); |
d4edf39b MX |
714 | |
715 | return trans; | |
716 | } | |
717 | ||
d352ac68 | 718 | /* wait for a transaction commit to be fully complete */ |
2ff7e61e | 719 | static noinline void wait_for_commit(struct btrfs_transaction *commit) |
89ce8a63 | 720 | { |
4a9d8bde | 721 | wait_event(commit->commit_wait, commit->state == TRANS_STATE_COMPLETED); |
89ce8a63 CM |
722 | } |
723 | ||
2ff7e61e | 724 | int btrfs_wait_for_commit(struct btrfs_fs_info *fs_info, u64 transid) |
46204592 SW |
725 | { |
726 | struct btrfs_transaction *cur_trans = NULL, *t; | |
8cd2807f | 727 | int ret = 0; |
46204592 | 728 | |
46204592 | 729 | if (transid) { |
0b246afa | 730 | if (transid <= fs_info->last_trans_committed) |
a4abeea4 | 731 | goto out; |
46204592 SW |
732 | |
733 | /* find specified transaction */ | |
0b246afa JM |
734 | spin_lock(&fs_info->trans_lock); |
735 | list_for_each_entry(t, &fs_info->trans_list, list) { | |
46204592 SW |
736 | if (t->transid == transid) { |
737 | cur_trans = t; | |
9b64f57d | 738 | refcount_inc(&cur_trans->use_count); |
8cd2807f | 739 | ret = 0; |
46204592 SW |
740 | break; |
741 | } | |
8cd2807f MX |
742 | if (t->transid > transid) { |
743 | ret = 0; | |
46204592 | 744 | break; |
8cd2807f | 745 | } |
46204592 | 746 | } |
0b246afa | 747 | spin_unlock(&fs_info->trans_lock); |
42383020 SW |
748 | |
749 | /* | |
750 | * The specified transaction doesn't exist, or we | |
751 | * raced with btrfs_commit_transaction | |
752 | */ | |
753 | if (!cur_trans) { | |
0b246afa | 754 | if (transid > fs_info->last_trans_committed) |
42383020 | 755 | ret = -EINVAL; |
8cd2807f | 756 | goto out; |
42383020 | 757 | } |
46204592 SW |
758 | } else { |
759 | /* find newest transaction that is committing | committed */ | |
0b246afa JM |
760 | spin_lock(&fs_info->trans_lock); |
761 | list_for_each_entry_reverse(t, &fs_info->trans_list, | |
46204592 | 762 | list) { |
4a9d8bde MX |
763 | if (t->state >= TRANS_STATE_COMMIT_START) { |
764 | if (t->state == TRANS_STATE_COMPLETED) | |
3473f3c0 | 765 | break; |
46204592 | 766 | cur_trans = t; |
9b64f57d | 767 | refcount_inc(&cur_trans->use_count); |
46204592 SW |
768 | break; |
769 | } | |
770 | } | |
0b246afa | 771 | spin_unlock(&fs_info->trans_lock); |
46204592 | 772 | if (!cur_trans) |
a4abeea4 | 773 | goto out; /* nothing committing|committed */ |
46204592 SW |
774 | } |
775 | ||
2ff7e61e | 776 | wait_for_commit(cur_trans); |
724e2315 | 777 | btrfs_put_transaction(cur_trans); |
a4abeea4 | 778 | out: |
46204592 SW |
779 | return ret; |
780 | } | |
781 | ||
2ff7e61e | 782 | void btrfs_throttle(struct btrfs_fs_info *fs_info) |
37d1aeee | 783 | { |
0b246afa | 784 | if (!atomic_read(&fs_info->open_ioctl_trans)) |
2ff7e61e | 785 | wait_current_trans(fs_info); |
37d1aeee CM |
786 | } |
787 | ||
2ff7e61e | 788 | static int should_end_transaction(struct btrfs_trans_handle *trans) |
8929ecfa | 789 | { |
2ff7e61e | 790 | struct btrfs_fs_info *fs_info = trans->fs_info; |
0b246afa | 791 | |
7c777430 | 792 | if (btrfs_check_space_for_delayed_refs(trans, fs_info)) |
1be41b78 | 793 | return 1; |
36ba022a | 794 | |
2ff7e61e | 795 | return !!btrfs_block_rsv_check(&fs_info->global_block_rsv, 5); |
8929ecfa YZ |
796 | } |
797 | ||
3a45bb20 | 798 | int btrfs_should_end_transaction(struct btrfs_trans_handle *trans) |
8929ecfa YZ |
799 | { |
800 | struct btrfs_transaction *cur_trans = trans->transaction; | |
3a45bb20 | 801 | struct btrfs_fs_info *fs_info = trans->fs_info; |
8929ecfa | 802 | int updates; |
49b25e05 | 803 | int err; |
8929ecfa | 804 | |
a4abeea4 | 805 | smp_mb(); |
4a9d8bde MX |
806 | if (cur_trans->state >= TRANS_STATE_BLOCKED || |
807 | cur_trans->delayed_refs.flushing) | |
8929ecfa YZ |
808 | return 1; |
809 | ||
810 | updates = trans->delayed_ref_updates; | |
811 | trans->delayed_ref_updates = 0; | |
49b25e05 | 812 | if (updates) { |
2ff7e61e | 813 | err = btrfs_run_delayed_refs(trans, fs_info, updates * 2); |
49b25e05 JM |
814 | if (err) /* Error code will also eval true */ |
815 | return err; | |
816 | } | |
8929ecfa | 817 | |
2ff7e61e | 818 | return should_end_transaction(trans); |
8929ecfa YZ |
819 | } |
820 | ||
dc60c525 NB |
821 | static void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans) |
822 | ||
0e34693f | 823 | { |
dc60c525 NB |
824 | struct btrfs_fs_info *fs_info = trans->fs_info; |
825 | ||
0e34693f NB |
826 | if (!trans->block_rsv) { |
827 | ASSERT(!trans->bytes_reserved); | |
828 | return; | |
829 | } | |
830 | ||
831 | if (!trans->bytes_reserved) | |
832 | return; | |
833 | ||
834 | ASSERT(trans->block_rsv == &fs_info->trans_block_rsv); | |
835 | trace_btrfs_space_reservation(fs_info, "transaction", | |
836 | trans->transid, trans->bytes_reserved, 0); | |
837 | btrfs_block_rsv_release(fs_info, trans->block_rsv, | |
838 | trans->bytes_reserved); | |
839 | trans->bytes_reserved = 0; | |
840 | } | |
841 | ||
89ce8a63 | 842 | static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, |
3a45bb20 | 843 | int throttle) |
79154b1b | 844 | { |
3a45bb20 | 845 | struct btrfs_fs_info *info = trans->fs_info; |
8929ecfa | 846 | struct btrfs_transaction *cur_trans = trans->transaction; |
31b9655f | 847 | u64 transid = trans->transid; |
1be41b78 | 848 | unsigned long cur = trans->delayed_ref_updates; |
a698d075 | 849 | int lock = (trans->type != TRANS_JOIN_NOLOCK); |
4edc2ca3 | 850 | int err = 0; |
a79b7d4b | 851 | int must_run_delayed_refs = 0; |
c3e69d58 | 852 | |
b50fff81 DS |
853 | if (refcount_read(&trans->use_count) > 1) { |
854 | refcount_dec(&trans->use_count); | |
2a1eb461 JB |
855 | trans->block_rsv = trans->orig_rsv; |
856 | return 0; | |
857 | } | |
858 | ||
dc60c525 | 859 | btrfs_trans_release_metadata(trans); |
4c13d758 | 860 | trans->block_rsv = NULL; |
c5567237 | 861 | |
ea658bad | 862 | if (!list_empty(&trans->new_bgs)) |
6c686b35 | 863 | btrfs_create_pending_block_groups(trans); |
ea658bad | 864 | |
1be41b78 | 865 | trans->delayed_ref_updates = 0; |
a79b7d4b CM |
866 | if (!trans->sync) { |
867 | must_run_delayed_refs = | |
2ff7e61e | 868 | btrfs_should_throttle_delayed_refs(trans, info); |
0a2b2a84 | 869 | cur = max_t(unsigned long, cur, 32); |
a79b7d4b CM |
870 | |
871 | /* | |
872 | * don't make the caller wait if they are from a NOLOCK | |
873 | * or ATTACH transaction, it will deadlock with commit | |
874 | */ | |
875 | if (must_run_delayed_refs == 1 && | |
876 | (trans->type & (__TRANS_JOIN_NOLOCK | __TRANS_ATTACH))) | |
877 | must_run_delayed_refs = 2; | |
56bec294 | 878 | } |
bb721703 | 879 | |
dc60c525 | 880 | btrfs_trans_release_metadata(trans); |
0e721106 | 881 | trans->block_rsv = NULL; |
56bec294 | 882 | |
ea658bad | 883 | if (!list_empty(&trans->new_bgs)) |
6c686b35 | 884 | btrfs_create_pending_block_groups(trans); |
ea658bad | 885 | |
4fbcdf66 FM |
886 | btrfs_trans_release_chunk_metadata(trans); |
887 | ||
0b246afa | 888 | if (lock && !atomic_read(&info->open_ioctl_trans) && |
2ff7e61e | 889 | should_end_transaction(trans) && |
20c7bcec | 890 | READ_ONCE(cur_trans->state) == TRANS_STATE_RUNNING) { |
4a9d8bde MX |
891 | spin_lock(&info->trans_lock); |
892 | if (cur_trans->state == TRANS_STATE_RUNNING) | |
893 | cur_trans->state = TRANS_STATE_BLOCKED; | |
894 | spin_unlock(&info->trans_lock); | |
a4abeea4 | 895 | } |
8929ecfa | 896 | |
20c7bcec | 897 | if (lock && READ_ONCE(cur_trans->state) == TRANS_STATE_BLOCKED) { |
3bbb24b2 | 898 | if (throttle) |
3a45bb20 | 899 | return btrfs_commit_transaction(trans); |
3bbb24b2 | 900 | else |
8929ecfa YZ |
901 | wake_up_process(info->transaction_kthread); |
902 | } | |
903 | ||
0860adfd | 904 | if (trans->type & __TRANS_FREEZABLE) |
0b246afa | 905 | sb_end_intwrite(info->sb); |
6df7881a | 906 | |
8929ecfa | 907 | WARN_ON(cur_trans != info->running_transaction); |
13c5a93e JB |
908 | WARN_ON(atomic_read(&cur_trans->num_writers) < 1); |
909 | atomic_dec(&cur_trans->num_writers); | |
0860adfd | 910 | extwriter_counter_dec(cur_trans, trans->type); |
89ce8a63 | 911 | |
a83342aa DS |
912 | /* |
913 | * Make sure counter is updated before we wake up waiters. | |
914 | */ | |
99d16cbc | 915 | smp_mb(); |
79154b1b CM |
916 | if (waitqueue_active(&cur_trans->writer_wait)) |
917 | wake_up(&cur_trans->writer_wait); | |
724e2315 | 918 | btrfs_put_transaction(cur_trans); |
9ed74f2d JB |
919 | |
920 | if (current->journal_info == trans) | |
921 | current->journal_info = NULL; | |
ab78c84d | 922 | |
24bbcf04 | 923 | if (throttle) |
2ff7e61e | 924 | btrfs_run_delayed_iputs(info); |
24bbcf04 | 925 | |
49b25e05 | 926 | if (trans->aborted || |
0b246afa | 927 | test_bit(BTRFS_FS_STATE_ERROR, &info->fs_state)) { |
4e121c06 | 928 | wake_up_process(info->transaction_kthread); |
4edc2ca3 | 929 | err = -EIO; |
4e121c06 | 930 | } |
49b25e05 | 931 | |
4edc2ca3 | 932 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
a79b7d4b | 933 | if (must_run_delayed_refs) { |
2ff7e61e | 934 | btrfs_async_run_delayed_refs(info, cur, transid, |
a79b7d4b CM |
935 | must_run_delayed_refs == 1); |
936 | } | |
4edc2ca3 | 937 | return err; |
79154b1b CM |
938 | } |
939 | ||
3a45bb20 | 940 | int btrfs_end_transaction(struct btrfs_trans_handle *trans) |
89ce8a63 | 941 | { |
3a45bb20 | 942 | return __btrfs_end_transaction(trans, 0); |
89ce8a63 CM |
943 | } |
944 | ||
3a45bb20 | 945 | int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans) |
89ce8a63 | 946 | { |
3a45bb20 | 947 | return __btrfs_end_transaction(trans, 1); |
16cdcec7 MX |
948 | } |
949 | ||
d352ac68 CM |
950 | /* |
951 | * when btree blocks are allocated, they have some corresponding bits set for | |
952 | * them in one of two extent_io trees. This is used to make sure all of | |
690587d1 | 953 | * those extents are sent to disk but does not wait on them |
d352ac68 | 954 | */ |
2ff7e61e | 955 | int btrfs_write_marked_extents(struct btrfs_fs_info *fs_info, |
8cef4e16 | 956 | struct extent_io_tree *dirty_pages, int mark) |
79154b1b | 957 | { |
777e6bd7 | 958 | int err = 0; |
7c4452b9 | 959 | int werr = 0; |
0b246afa | 960 | struct address_space *mapping = fs_info->btree_inode->i_mapping; |
e6138876 | 961 | struct extent_state *cached_state = NULL; |
777e6bd7 | 962 | u64 start = 0; |
5f39d397 | 963 | u64 end; |
7c4452b9 | 964 | |
6300463b | 965 | atomic_inc(&BTRFS_I(fs_info->btree_inode)->sync_writers); |
1728366e | 966 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
e6138876 | 967 | mark, &cached_state)) { |
663dfbb0 FM |
968 | bool wait_writeback = false; |
969 | ||
970 | err = convert_extent_bit(dirty_pages, start, end, | |
971 | EXTENT_NEED_WAIT, | |
210aa277 | 972 | mark, &cached_state); |
663dfbb0 FM |
973 | /* |
974 | * convert_extent_bit can return -ENOMEM, which is most of the | |
975 | * time a temporary error. So when it happens, ignore the error | |
976 | * and wait for writeback of this range to finish - because we | |
977 | * failed to set the bit EXTENT_NEED_WAIT for the range, a call | |
bf89d38f JM |
978 | * to __btrfs_wait_marked_extents() would not know that |
979 | * writeback for this range started and therefore wouldn't | |
980 | * wait for it to finish - we don't want to commit a | |
981 | * superblock that points to btree nodes/leafs for which | |
982 | * writeback hasn't finished yet (and without errors). | |
663dfbb0 FM |
983 | * We cleanup any entries left in the io tree when committing |
984 | * the transaction (through clear_btree_io_tree()). | |
985 | */ | |
986 | if (err == -ENOMEM) { | |
987 | err = 0; | |
988 | wait_writeback = true; | |
989 | } | |
990 | if (!err) | |
991 | err = filemap_fdatawrite_range(mapping, start, end); | |
1728366e JB |
992 | if (err) |
993 | werr = err; | |
663dfbb0 FM |
994 | else if (wait_writeback) |
995 | werr = filemap_fdatawait_range(mapping, start, end); | |
e38e2ed7 | 996 | free_extent_state(cached_state); |
663dfbb0 | 997 | cached_state = NULL; |
1728366e JB |
998 | cond_resched(); |
999 | start = end + 1; | |
7c4452b9 | 1000 | } |
6300463b | 1001 | atomic_dec(&BTRFS_I(fs_info->btree_inode)->sync_writers); |
690587d1 CM |
1002 | return werr; |
1003 | } | |
1004 | ||
1005 | /* | |
1006 | * when btree blocks are allocated, they have some corresponding bits set for | |
1007 | * them in one of two extent_io trees. This is used to make sure all of | |
1008 | * those extents are on disk for transaction or log commit. We wait | |
1009 | * on all the pages and clear them from the dirty pages state tree | |
1010 | */ | |
bf89d38f JM |
1011 | static int __btrfs_wait_marked_extents(struct btrfs_fs_info *fs_info, |
1012 | struct extent_io_tree *dirty_pages) | |
690587d1 | 1013 | { |
690587d1 CM |
1014 | int err = 0; |
1015 | int werr = 0; | |
0b246afa | 1016 | struct address_space *mapping = fs_info->btree_inode->i_mapping; |
e6138876 | 1017 | struct extent_state *cached_state = NULL; |
690587d1 CM |
1018 | u64 start = 0; |
1019 | u64 end; | |
777e6bd7 | 1020 | |
1728366e | 1021 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
e6138876 | 1022 | EXTENT_NEED_WAIT, &cached_state)) { |
663dfbb0 FM |
1023 | /* |
1024 | * Ignore -ENOMEM errors returned by clear_extent_bit(). | |
1025 | * When committing the transaction, we'll remove any entries | |
1026 | * left in the io tree. For a log commit, we don't remove them | |
1027 | * after committing the log because the tree can be accessed | |
1028 | * concurrently - we do it only at transaction commit time when | |
1029 | * it's safe to do it (through clear_btree_io_tree()). | |
1030 | */ | |
1031 | err = clear_extent_bit(dirty_pages, start, end, | |
ae0f1625 | 1032 | EXTENT_NEED_WAIT, 0, 0, &cached_state); |
663dfbb0 FM |
1033 | if (err == -ENOMEM) |
1034 | err = 0; | |
1035 | if (!err) | |
1036 | err = filemap_fdatawait_range(mapping, start, end); | |
1728366e JB |
1037 | if (err) |
1038 | werr = err; | |
e38e2ed7 FM |
1039 | free_extent_state(cached_state); |
1040 | cached_state = NULL; | |
1728366e JB |
1041 | cond_resched(); |
1042 | start = end + 1; | |
777e6bd7 | 1043 | } |
7c4452b9 CM |
1044 | if (err) |
1045 | werr = err; | |
bf89d38f JM |
1046 | return werr; |
1047 | } | |
656f30db | 1048 | |
bf89d38f JM |
1049 | int btrfs_wait_extents(struct btrfs_fs_info *fs_info, |
1050 | struct extent_io_tree *dirty_pages) | |
1051 | { | |
1052 | bool errors = false; | |
1053 | int err; | |
656f30db | 1054 | |
bf89d38f JM |
1055 | err = __btrfs_wait_marked_extents(fs_info, dirty_pages); |
1056 | if (test_and_clear_bit(BTRFS_FS_BTREE_ERR, &fs_info->flags)) | |
1057 | errors = true; | |
1058 | ||
1059 | if (errors && !err) | |
1060 | err = -EIO; | |
1061 | return err; | |
1062 | } | |
656f30db | 1063 | |
bf89d38f JM |
1064 | int btrfs_wait_tree_log_extents(struct btrfs_root *log_root, int mark) |
1065 | { | |
1066 | struct btrfs_fs_info *fs_info = log_root->fs_info; | |
1067 | struct extent_io_tree *dirty_pages = &log_root->dirty_log_pages; | |
1068 | bool errors = false; | |
1069 | int err; | |
656f30db | 1070 | |
bf89d38f JM |
1071 | ASSERT(log_root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID); |
1072 | ||
1073 | err = __btrfs_wait_marked_extents(fs_info, dirty_pages); | |
1074 | if ((mark & EXTENT_DIRTY) && | |
1075 | test_and_clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags)) | |
1076 | errors = true; | |
1077 | ||
1078 | if ((mark & EXTENT_NEW) && | |
1079 | test_and_clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags)) | |
1080 | errors = true; | |
1081 | ||
1082 | if (errors && !err) | |
1083 | err = -EIO; | |
1084 | return err; | |
79154b1b CM |
1085 | } |
1086 | ||
690587d1 | 1087 | /* |
c9b577c0 NB |
1088 | * When btree blocks are allocated the corresponding extents are marked dirty. |
1089 | * This function ensures such extents are persisted on disk for transaction or | |
1090 | * log commit. | |
1091 | * | |
1092 | * @trans: transaction whose dirty pages we'd like to write | |
690587d1 | 1093 | */ |
c9b577c0 NB |
1094 | static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans, |
1095 | struct btrfs_fs_info *fs_info) | |
690587d1 CM |
1096 | { |
1097 | int ret; | |
1098 | int ret2; | |
c9b577c0 | 1099 | struct extent_io_tree *dirty_pages = &trans->transaction->dirty_pages; |
c6adc9cc | 1100 | struct blk_plug plug; |
690587d1 | 1101 | |
c6adc9cc | 1102 | blk_start_plug(&plug); |
c9b577c0 | 1103 | ret = btrfs_write_marked_extents(fs_info, dirty_pages, EXTENT_DIRTY); |
c6adc9cc | 1104 | blk_finish_plug(&plug); |
bf89d38f | 1105 | ret2 = btrfs_wait_extents(fs_info, dirty_pages); |
bf0da8c1 | 1106 | |
c9b577c0 NB |
1107 | clear_btree_io_tree(&trans->transaction->dirty_pages); |
1108 | ||
bf0da8c1 CM |
1109 | if (ret) |
1110 | return ret; | |
c9b577c0 | 1111 | else if (ret2) |
bf0da8c1 | 1112 | return ret2; |
c9b577c0 NB |
1113 | else |
1114 | return 0; | |
d0c803c4 CM |
1115 | } |
1116 | ||
d352ac68 CM |
1117 | /* |
1118 | * this is used to update the root pointer in the tree of tree roots. | |
1119 | * | |
1120 | * But, in the case of the extent allocation tree, updating the root | |
1121 | * pointer may allocate blocks which may change the root of the extent | |
1122 | * allocation tree. | |
1123 | * | |
1124 | * So, this loops and repeats and makes sure the cowonly root didn't | |
1125 | * change while the root pointer was being updated in the metadata. | |
1126 | */ | |
0b86a832 CM |
1127 | static int update_cowonly_root(struct btrfs_trans_handle *trans, |
1128 | struct btrfs_root *root) | |
79154b1b CM |
1129 | { |
1130 | int ret; | |
0b86a832 | 1131 | u64 old_root_bytenr; |
86b9f2ec | 1132 | u64 old_root_used; |
0b246afa JM |
1133 | struct btrfs_fs_info *fs_info = root->fs_info; |
1134 | struct btrfs_root *tree_root = fs_info->tree_root; | |
79154b1b | 1135 | |
86b9f2ec | 1136 | old_root_used = btrfs_root_used(&root->root_item); |
56bec294 | 1137 | |
d397712b | 1138 | while (1) { |
0b86a832 | 1139 | old_root_bytenr = btrfs_root_bytenr(&root->root_item); |
86b9f2ec | 1140 | if (old_root_bytenr == root->node->start && |
ea526d18 | 1141 | old_root_used == btrfs_root_used(&root->root_item)) |
79154b1b | 1142 | break; |
87ef2bb4 | 1143 | |
5d4f98a2 | 1144 | btrfs_set_root_node(&root->root_item, root->node); |
79154b1b | 1145 | ret = btrfs_update_root(trans, tree_root, |
0b86a832 CM |
1146 | &root->root_key, |
1147 | &root->root_item); | |
49b25e05 JM |
1148 | if (ret) |
1149 | return ret; | |
56bec294 | 1150 | |
86b9f2ec | 1151 | old_root_used = btrfs_root_used(&root->root_item); |
0b86a832 | 1152 | } |
276e680d | 1153 | |
0b86a832 CM |
1154 | return 0; |
1155 | } | |
1156 | ||
d352ac68 CM |
1157 | /* |
1158 | * update all the cowonly tree roots on disk | |
49b25e05 JM |
1159 | * |
1160 | * The error handling in this function may not be obvious. Any of the | |
1161 | * failures will cause the file system to go offline. We still need | |
1162 | * to clean up the delayed refs. | |
d352ac68 | 1163 | */ |
5d4f98a2 | 1164 | static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans, |
2ff7e61e | 1165 | struct btrfs_fs_info *fs_info) |
0b86a832 | 1166 | { |
ea526d18 | 1167 | struct list_head *dirty_bgs = &trans->transaction->dirty_bgs; |
1bbc621e | 1168 | struct list_head *io_bgs = &trans->transaction->io_bgs; |
0b86a832 | 1169 | struct list_head *next; |
84234f3a | 1170 | struct extent_buffer *eb; |
56bec294 | 1171 | int ret; |
84234f3a YZ |
1172 | |
1173 | eb = btrfs_lock_root_node(fs_info->tree_root); | |
49b25e05 JM |
1174 | ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, |
1175 | 0, &eb); | |
84234f3a YZ |
1176 | btrfs_tree_unlock(eb); |
1177 | free_extent_buffer(eb); | |
0b86a832 | 1178 | |
49b25e05 JM |
1179 | if (ret) |
1180 | return ret; | |
1181 | ||
2ff7e61e | 1182 | ret = btrfs_run_delayed_refs(trans, fs_info, (unsigned long)-1); |
49b25e05 JM |
1183 | if (ret) |
1184 | return ret; | |
87ef2bb4 | 1185 | |
0b246afa | 1186 | ret = btrfs_run_dev_stats(trans, fs_info); |
c16ce190 JB |
1187 | if (ret) |
1188 | return ret; | |
0b246afa | 1189 | ret = btrfs_run_dev_replace(trans, fs_info); |
c16ce190 JB |
1190 | if (ret) |
1191 | return ret; | |
0b246afa | 1192 | ret = btrfs_run_qgroups(trans, fs_info); |
c16ce190 JB |
1193 | if (ret) |
1194 | return ret; | |
546adb0d | 1195 | |
2ff7e61e | 1196 | ret = btrfs_setup_space_cache(trans, fs_info); |
dcdf7f6d JB |
1197 | if (ret) |
1198 | return ret; | |
1199 | ||
546adb0d | 1200 | /* run_qgroups might have added some more refs */ |
2ff7e61e | 1201 | ret = btrfs_run_delayed_refs(trans, fs_info, (unsigned long)-1); |
c16ce190 JB |
1202 | if (ret) |
1203 | return ret; | |
ea526d18 | 1204 | again: |
d397712b | 1205 | while (!list_empty(&fs_info->dirty_cowonly_roots)) { |
2ff7e61e | 1206 | struct btrfs_root *root; |
0b86a832 CM |
1207 | next = fs_info->dirty_cowonly_roots.next; |
1208 | list_del_init(next); | |
1209 | root = list_entry(next, struct btrfs_root, dirty_list); | |
e7070be1 | 1210 | clear_bit(BTRFS_ROOT_DIRTY, &root->state); |
87ef2bb4 | 1211 | |
9e351cc8 JB |
1212 | if (root != fs_info->extent_root) |
1213 | list_add_tail(&root->dirty_list, | |
1214 | &trans->transaction->switch_commits); | |
49b25e05 JM |
1215 | ret = update_cowonly_root(trans, root); |
1216 | if (ret) | |
1217 | return ret; | |
2ff7e61e | 1218 | ret = btrfs_run_delayed_refs(trans, fs_info, (unsigned long)-1); |
ea526d18 JB |
1219 | if (ret) |
1220 | return ret; | |
79154b1b | 1221 | } |
276e680d | 1222 | |
1bbc621e | 1223 | while (!list_empty(dirty_bgs) || !list_empty(io_bgs)) { |
2ff7e61e | 1224 | ret = btrfs_write_dirty_block_groups(trans, fs_info); |
ea526d18 JB |
1225 | if (ret) |
1226 | return ret; | |
2ff7e61e | 1227 | ret = btrfs_run_delayed_refs(trans, fs_info, (unsigned long)-1); |
ea526d18 JB |
1228 | if (ret) |
1229 | return ret; | |
1230 | } | |
1231 | ||
1232 | if (!list_empty(&fs_info->dirty_cowonly_roots)) | |
1233 | goto again; | |
1234 | ||
9e351cc8 JB |
1235 | list_add_tail(&fs_info->extent_root->dirty_list, |
1236 | &trans->transaction->switch_commits); | |
8dabb742 SB |
1237 | btrfs_after_dev_replace_commit(fs_info); |
1238 | ||
79154b1b CM |
1239 | return 0; |
1240 | } | |
1241 | ||
d352ac68 CM |
1242 | /* |
1243 | * dead roots are old snapshots that need to be deleted. This allocates | |
1244 | * a dirty root struct and adds it into the list of dead roots that need to | |
1245 | * be deleted | |
1246 | */ | |
cfad392b | 1247 | void btrfs_add_dead_root(struct btrfs_root *root) |
5eda7b5e | 1248 | { |
0b246afa JM |
1249 | struct btrfs_fs_info *fs_info = root->fs_info; |
1250 | ||
1251 | spin_lock(&fs_info->trans_lock); | |
cfad392b | 1252 | if (list_empty(&root->root_list)) |
0b246afa JM |
1253 | list_add_tail(&root->root_list, &fs_info->dead_roots); |
1254 | spin_unlock(&fs_info->trans_lock); | |
5eda7b5e CM |
1255 | } |
1256 | ||
d352ac68 | 1257 | /* |
5d4f98a2 | 1258 | * update all the cowonly tree roots on disk |
d352ac68 | 1259 | */ |
5d4f98a2 | 1260 | static noinline int commit_fs_roots(struct btrfs_trans_handle *trans, |
5b4aacef | 1261 | struct btrfs_fs_info *fs_info) |
0f7d52f4 | 1262 | { |
0f7d52f4 | 1263 | struct btrfs_root *gang[8]; |
0f7d52f4 CM |
1264 | int i; |
1265 | int ret; | |
54aa1f4d CM |
1266 | int err = 0; |
1267 | ||
a4abeea4 | 1268 | spin_lock(&fs_info->fs_roots_radix_lock); |
d397712b | 1269 | while (1) { |
5d4f98a2 YZ |
1270 | ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, |
1271 | (void **)gang, 0, | |
0f7d52f4 CM |
1272 | ARRAY_SIZE(gang), |
1273 | BTRFS_ROOT_TRANS_TAG); | |
1274 | if (ret == 0) | |
1275 | break; | |
1276 | for (i = 0; i < ret; i++) { | |
5b4aacef | 1277 | struct btrfs_root *root = gang[i]; |
5d4f98a2 YZ |
1278 | radix_tree_tag_clear(&fs_info->fs_roots_radix, |
1279 | (unsigned long)root->root_key.objectid, | |
1280 | BTRFS_ROOT_TRANS_TAG); | |
a4abeea4 | 1281 | spin_unlock(&fs_info->fs_roots_radix_lock); |
31153d81 | 1282 | |
e02119d5 | 1283 | btrfs_free_log(trans, root); |
5d4f98a2 | 1284 | btrfs_update_reloc_root(trans, root); |
d68fc57b | 1285 | btrfs_orphan_commit_root(trans, root); |
bcc63abb | 1286 | |
82d5902d LZ |
1287 | btrfs_save_ino_cache(root, trans); |
1288 | ||
f1ebcc74 | 1289 | /* see comments in should_cow_block() */ |
27cdeb70 | 1290 | clear_bit(BTRFS_ROOT_FORCE_COW, &root->state); |
c7548af6 | 1291 | smp_mb__after_atomic(); |
f1ebcc74 | 1292 | |
978d910d | 1293 | if (root->commit_root != root->node) { |
9e351cc8 JB |
1294 | list_add_tail(&root->dirty_list, |
1295 | &trans->transaction->switch_commits); | |
978d910d YZ |
1296 | btrfs_set_root_node(&root->root_item, |
1297 | root->node); | |
1298 | } | |
5d4f98a2 | 1299 | |
5d4f98a2 | 1300 | err = btrfs_update_root(trans, fs_info->tree_root, |
0f7d52f4 CM |
1301 | &root->root_key, |
1302 | &root->root_item); | |
a4abeea4 | 1303 | spin_lock(&fs_info->fs_roots_radix_lock); |
54aa1f4d CM |
1304 | if (err) |
1305 | break; | |
7174109c | 1306 | btrfs_qgroup_free_meta_all(root); |
0f7d52f4 CM |
1307 | } |
1308 | } | |
a4abeea4 | 1309 | spin_unlock(&fs_info->fs_roots_radix_lock); |
54aa1f4d | 1310 | return err; |
0f7d52f4 CM |
1311 | } |
1312 | ||
d352ac68 | 1313 | /* |
de78b51a ES |
1314 | * defrag a given btree. |
1315 | * Every leaf in the btree is read and defragged. | |
d352ac68 | 1316 | */ |
de78b51a | 1317 | int btrfs_defrag_root(struct btrfs_root *root) |
e9d0b13b CM |
1318 | { |
1319 | struct btrfs_fs_info *info = root->fs_info; | |
e9d0b13b | 1320 | struct btrfs_trans_handle *trans; |
8929ecfa | 1321 | int ret; |
e9d0b13b | 1322 | |
27cdeb70 | 1323 | if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state)) |
e9d0b13b | 1324 | return 0; |
8929ecfa | 1325 | |
6b80053d | 1326 | while (1) { |
8929ecfa YZ |
1327 | trans = btrfs_start_transaction(root, 0); |
1328 | if (IS_ERR(trans)) | |
1329 | return PTR_ERR(trans); | |
1330 | ||
de78b51a | 1331 | ret = btrfs_defrag_leaves(trans, root); |
8929ecfa | 1332 | |
3a45bb20 | 1333 | btrfs_end_transaction(trans); |
2ff7e61e | 1334 | btrfs_btree_balance_dirty(info); |
e9d0b13b CM |
1335 | cond_resched(); |
1336 | ||
ab8d0fc4 | 1337 | if (btrfs_fs_closing(info) || ret != -EAGAIN) |
e9d0b13b | 1338 | break; |
210549eb | 1339 | |
ab8d0fc4 JM |
1340 | if (btrfs_defrag_cancelled(info)) { |
1341 | btrfs_debug(info, "defrag_root cancelled"); | |
210549eb DS |
1342 | ret = -EAGAIN; |
1343 | break; | |
1344 | } | |
e9d0b13b | 1345 | } |
27cdeb70 | 1346 | clear_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state); |
8929ecfa | 1347 | return ret; |
e9d0b13b CM |
1348 | } |
1349 | ||
6426c7ad QW |
1350 | /* |
1351 | * Do all special snapshot related qgroup dirty hack. | |
1352 | * | |
1353 | * Will do all needed qgroup inherit and dirty hack like switch commit | |
1354 | * roots inside one transaction and write all btree into disk, to make | |
1355 | * qgroup works. | |
1356 | */ | |
1357 | static int qgroup_account_snapshot(struct btrfs_trans_handle *trans, | |
1358 | struct btrfs_root *src, | |
1359 | struct btrfs_root *parent, | |
1360 | struct btrfs_qgroup_inherit *inherit, | |
1361 | u64 dst_objectid) | |
1362 | { | |
1363 | struct btrfs_fs_info *fs_info = src->fs_info; | |
1364 | int ret; | |
1365 | ||
1366 | /* | |
1367 | * Save some performance in the case that qgroups are not | |
1368 | * enabled. If this check races with the ioctl, rescan will | |
1369 | * kick in anyway. | |
1370 | */ | |
9ea6e2b5 | 1371 | if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) |
6426c7ad | 1372 | return 0; |
6426c7ad QW |
1373 | |
1374 | /* | |
1375 | * We are going to commit transaction, see btrfs_commit_transaction() | |
1376 | * comment for reason locking tree_log_mutex | |
1377 | */ | |
1378 | mutex_lock(&fs_info->tree_log_mutex); | |
1379 | ||
5b4aacef | 1380 | ret = commit_fs_roots(trans, fs_info); |
6426c7ad QW |
1381 | if (ret) |
1382 | goto out; | |
6426c7ad QW |
1383 | ret = btrfs_qgroup_account_extents(trans, fs_info); |
1384 | if (ret < 0) | |
1385 | goto out; | |
1386 | ||
1387 | /* Now qgroup are all updated, we can inherit it to new qgroups */ | |
1388 | ret = btrfs_qgroup_inherit(trans, fs_info, | |
1389 | src->root_key.objectid, dst_objectid, | |
1390 | inherit); | |
1391 | if (ret < 0) | |
1392 | goto out; | |
1393 | ||
1394 | /* | |
1395 | * Now we do a simplified commit transaction, which will: | |
1396 | * 1) commit all subvolume and extent tree | |
1397 | * To ensure all subvolume and extent tree have a valid | |
1398 | * commit_root to accounting later insert_dir_item() | |
1399 | * 2) write all btree blocks onto disk | |
1400 | * This is to make sure later btree modification will be cowed | |
1401 | * Or commit_root can be populated and cause wrong qgroup numbers | |
1402 | * In this simplified commit, we don't really care about other trees | |
1403 | * like chunk and root tree, as they won't affect qgroup. | |
1404 | * And we don't write super to avoid half committed status. | |
1405 | */ | |
2ff7e61e | 1406 | ret = commit_cowonly_roots(trans, fs_info); |
6426c7ad QW |
1407 | if (ret) |
1408 | goto out; | |
1409 | switch_commit_roots(trans->transaction, fs_info); | |
bf89d38f | 1410 | ret = btrfs_write_and_wait_transaction(trans, fs_info); |
6426c7ad | 1411 | if (ret) |
f7af3934 | 1412 | btrfs_handle_fs_error(fs_info, ret, |
6426c7ad QW |
1413 | "Error while writing out transaction for qgroup"); |
1414 | ||
1415 | out: | |
1416 | mutex_unlock(&fs_info->tree_log_mutex); | |
1417 | ||
1418 | /* | |
1419 | * Force parent root to be updated, as we recorded it before so its | |
1420 | * last_trans == cur_transid. | |
1421 | * Or it won't be committed again onto disk after later | |
1422 | * insert_dir_item() | |
1423 | */ | |
1424 | if (!ret) | |
1425 | record_root_in_trans(trans, parent, 1); | |
1426 | return ret; | |
1427 | } | |
1428 | ||
d352ac68 CM |
1429 | /* |
1430 | * new snapshots need to be created at a very specific time in the | |
aec8030a MX |
1431 | * transaction commit. This does the actual creation. |
1432 | * | |
1433 | * Note: | |
1434 | * If the error which may affect the commitment of the current transaction | |
1435 | * happens, we should return the error number. If the error which just affect | |
1436 | * the creation of the pending snapshots, just return 0. | |
d352ac68 | 1437 | */ |
80b6794d | 1438 | static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, |
3063d29f CM |
1439 | struct btrfs_fs_info *fs_info, |
1440 | struct btrfs_pending_snapshot *pending) | |
1441 | { | |
1442 | struct btrfs_key key; | |
80b6794d | 1443 | struct btrfs_root_item *new_root_item; |
3063d29f CM |
1444 | struct btrfs_root *tree_root = fs_info->tree_root; |
1445 | struct btrfs_root *root = pending->root; | |
6bdb72de | 1446 | struct btrfs_root *parent_root; |
98c9942a | 1447 | struct btrfs_block_rsv *rsv; |
6bdb72de | 1448 | struct inode *parent_inode; |
42874b3d MX |
1449 | struct btrfs_path *path; |
1450 | struct btrfs_dir_item *dir_item; | |
a22285a6 | 1451 | struct dentry *dentry; |
3063d29f | 1452 | struct extent_buffer *tmp; |
925baedd | 1453 | struct extent_buffer *old; |
04b285f3 | 1454 | struct timespec cur_time; |
aec8030a | 1455 | int ret = 0; |
d68fc57b | 1456 | u64 to_reserve = 0; |
6bdb72de | 1457 | u64 index = 0; |
a22285a6 | 1458 | u64 objectid; |
b83cc969 | 1459 | u64 root_flags; |
8ea05e3a | 1460 | uuid_le new_uuid; |
3063d29f | 1461 | |
8546b570 DS |
1462 | ASSERT(pending->path); |
1463 | path = pending->path; | |
42874b3d | 1464 | |
b0c0ea63 DS |
1465 | ASSERT(pending->root_item); |
1466 | new_root_item = pending->root_item; | |
a22285a6 | 1467 | |
aec8030a MX |
1468 | pending->error = btrfs_find_free_objectid(tree_root, &objectid); |
1469 | if (pending->error) | |
6fa9700e | 1470 | goto no_free_objectid; |
3063d29f | 1471 | |
d6726335 QW |
1472 | /* |
1473 | * Make qgroup to skip current new snapshot's qgroupid, as it is | |
1474 | * accounted by later btrfs_qgroup_inherit(). | |
1475 | */ | |
1476 | btrfs_set_skip_qgroup(trans, objectid); | |
1477 | ||
147d256e | 1478 | btrfs_reloc_pre_snapshot(pending, &to_reserve); |
d68fc57b YZ |
1479 | |
1480 | if (to_reserve > 0) { | |
aec8030a MX |
1481 | pending->error = btrfs_block_rsv_add(root, |
1482 | &pending->block_rsv, | |
1483 | to_reserve, | |
1484 | BTRFS_RESERVE_NO_FLUSH); | |
1485 | if (pending->error) | |
d6726335 | 1486 | goto clear_skip_qgroup; |
d68fc57b YZ |
1487 | } |
1488 | ||
3063d29f | 1489 | key.objectid = objectid; |
a22285a6 YZ |
1490 | key.offset = (u64)-1; |
1491 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3063d29f | 1492 | |
6fa9700e | 1493 | rsv = trans->block_rsv; |
a22285a6 | 1494 | trans->block_rsv = &pending->block_rsv; |
2382c5cc | 1495 | trans->bytes_reserved = trans->block_rsv->reserved; |
0b246afa | 1496 | trace_btrfs_space_reservation(fs_info, "transaction", |
88d3a5aa JB |
1497 | trans->transid, |
1498 | trans->bytes_reserved, 1); | |
a22285a6 | 1499 | dentry = pending->dentry; |
e9662f70 | 1500 | parent_inode = pending->dir; |
a22285a6 | 1501 | parent_root = BTRFS_I(parent_inode)->root; |
6426c7ad | 1502 | record_root_in_trans(trans, parent_root, 0); |
a22285a6 | 1503 | |
c2050a45 | 1504 | cur_time = current_time(parent_inode); |
04b285f3 | 1505 | |
3063d29f CM |
1506 | /* |
1507 | * insert the directory item | |
1508 | */ | |
877574e2 | 1509 | ret = btrfs_set_inode_index(BTRFS_I(parent_inode), &index); |
49b25e05 | 1510 | BUG_ON(ret); /* -ENOMEM */ |
42874b3d MX |
1511 | |
1512 | /* check if there is a file/dir which has the same name. */ | |
1513 | dir_item = btrfs_lookup_dir_item(NULL, parent_root, path, | |
4a0cc7ca | 1514 | btrfs_ino(BTRFS_I(parent_inode)), |
42874b3d MX |
1515 | dentry->d_name.name, |
1516 | dentry->d_name.len, 0); | |
1517 | if (dir_item != NULL && !IS_ERR(dir_item)) { | |
fe66a05a | 1518 | pending->error = -EEXIST; |
aec8030a | 1519 | goto dir_item_existed; |
42874b3d MX |
1520 | } else if (IS_ERR(dir_item)) { |
1521 | ret = PTR_ERR(dir_item); | |
66642832 | 1522 | btrfs_abort_transaction(trans, ret); |
8732d44f | 1523 | goto fail; |
79787eaa | 1524 | } |
42874b3d | 1525 | btrfs_release_path(path); |
52c26179 | 1526 | |
e999376f CM |
1527 | /* |
1528 | * pull in the delayed directory update | |
1529 | * and the delayed inode item | |
1530 | * otherwise we corrupt the FS during | |
1531 | * snapshot | |
1532 | */ | |
e5c304e6 | 1533 | ret = btrfs_run_delayed_items(trans); |
8732d44f | 1534 | if (ret) { /* Transaction aborted */ |
66642832 | 1535 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1536 | goto fail; |
1537 | } | |
e999376f | 1538 | |
6426c7ad | 1539 | record_root_in_trans(trans, root, 0); |
6bdb72de SW |
1540 | btrfs_set_root_last_snapshot(&root->root_item, trans->transid); |
1541 | memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); | |
08fe4db1 | 1542 | btrfs_check_and_init_root_item(new_root_item); |
6bdb72de | 1543 | |
b83cc969 LZ |
1544 | root_flags = btrfs_root_flags(new_root_item); |
1545 | if (pending->readonly) | |
1546 | root_flags |= BTRFS_ROOT_SUBVOL_RDONLY; | |
1547 | else | |
1548 | root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY; | |
1549 | btrfs_set_root_flags(new_root_item, root_flags); | |
1550 | ||
8ea05e3a AB |
1551 | btrfs_set_root_generation_v2(new_root_item, |
1552 | trans->transid); | |
1553 | uuid_le_gen(&new_uuid); | |
1554 | memcpy(new_root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE); | |
1555 | memcpy(new_root_item->parent_uuid, root->root_item.uuid, | |
1556 | BTRFS_UUID_SIZE); | |
70023da2 SB |
1557 | if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) { |
1558 | memset(new_root_item->received_uuid, 0, | |
1559 | sizeof(new_root_item->received_uuid)); | |
1560 | memset(&new_root_item->stime, 0, sizeof(new_root_item->stime)); | |
1561 | memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime)); | |
1562 | btrfs_set_root_stransid(new_root_item, 0); | |
1563 | btrfs_set_root_rtransid(new_root_item, 0); | |
1564 | } | |
3cae210f QW |
1565 | btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec); |
1566 | btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec); | |
8ea05e3a | 1567 | btrfs_set_root_otransid(new_root_item, trans->transid); |
8ea05e3a | 1568 | |
6bdb72de | 1569 | old = btrfs_lock_root_node(root); |
49b25e05 | 1570 | ret = btrfs_cow_block(trans, root, old, NULL, 0, &old); |
79787eaa JM |
1571 | if (ret) { |
1572 | btrfs_tree_unlock(old); | |
1573 | free_extent_buffer(old); | |
66642832 | 1574 | btrfs_abort_transaction(trans, ret); |
8732d44f | 1575 | goto fail; |
79787eaa | 1576 | } |
49b25e05 | 1577 | |
6bdb72de SW |
1578 | btrfs_set_lock_blocking(old); |
1579 | ||
49b25e05 | 1580 | ret = btrfs_copy_root(trans, root, old, &tmp, objectid); |
79787eaa | 1581 | /* clean up in any case */ |
6bdb72de SW |
1582 | btrfs_tree_unlock(old); |
1583 | free_extent_buffer(old); | |
8732d44f | 1584 | if (ret) { |
66642832 | 1585 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1586 | goto fail; |
1587 | } | |
f1ebcc74 | 1588 | /* see comments in should_cow_block() */ |
27cdeb70 | 1589 | set_bit(BTRFS_ROOT_FORCE_COW, &root->state); |
f1ebcc74 LB |
1590 | smp_wmb(); |
1591 | ||
6bdb72de | 1592 | btrfs_set_root_node(new_root_item, tmp); |
a22285a6 YZ |
1593 | /* record when the snapshot was created in key.offset */ |
1594 | key.offset = trans->transid; | |
1595 | ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); | |
6bdb72de SW |
1596 | btrfs_tree_unlock(tmp); |
1597 | free_extent_buffer(tmp); | |
8732d44f | 1598 | if (ret) { |
66642832 | 1599 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1600 | goto fail; |
1601 | } | |
6bdb72de | 1602 | |
a22285a6 YZ |
1603 | /* |
1604 | * insert root back/forward references | |
1605 | */ | |
6bccf3ab | 1606 | ret = btrfs_add_root_ref(trans, fs_info, objectid, |
0660b5af | 1607 | parent_root->root_key.objectid, |
4a0cc7ca | 1608 | btrfs_ino(BTRFS_I(parent_inode)), index, |
a22285a6 | 1609 | dentry->d_name.name, dentry->d_name.len); |
8732d44f | 1610 | if (ret) { |
66642832 | 1611 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1612 | goto fail; |
1613 | } | |
0660b5af | 1614 | |
a22285a6 | 1615 | key.offset = (u64)-1; |
0b246afa | 1616 | pending->snap = btrfs_read_fs_root_no_name(fs_info, &key); |
79787eaa JM |
1617 | if (IS_ERR(pending->snap)) { |
1618 | ret = PTR_ERR(pending->snap); | |
66642832 | 1619 | btrfs_abort_transaction(trans, ret); |
8732d44f | 1620 | goto fail; |
79787eaa | 1621 | } |
d68fc57b | 1622 | |
49b25e05 | 1623 | ret = btrfs_reloc_post_snapshot(trans, pending); |
8732d44f | 1624 | if (ret) { |
66642832 | 1625 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1626 | goto fail; |
1627 | } | |
361048f5 | 1628 | |
2ff7e61e | 1629 | ret = btrfs_run_delayed_refs(trans, fs_info, (unsigned long)-1); |
8732d44f | 1630 | if (ret) { |
66642832 | 1631 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1632 | goto fail; |
1633 | } | |
42874b3d | 1634 | |
6426c7ad QW |
1635 | /* |
1636 | * Do special qgroup accounting for snapshot, as we do some qgroup | |
1637 | * snapshot hack to do fast snapshot. | |
1638 | * To co-operate with that hack, we do hack again. | |
1639 | * Or snapshot will be greatly slowed down by a subtree qgroup rescan | |
1640 | */ | |
1641 | ret = qgroup_account_snapshot(trans, root, parent_root, | |
1642 | pending->inherit, objectid); | |
1643 | if (ret < 0) | |
1644 | goto fail; | |
1645 | ||
42874b3d MX |
1646 | ret = btrfs_insert_dir_item(trans, parent_root, |
1647 | dentry->d_name.name, dentry->d_name.len, | |
8e7611cf | 1648 | BTRFS_I(parent_inode), &key, |
42874b3d MX |
1649 | BTRFS_FT_DIR, index); |
1650 | /* We have check then name at the beginning, so it is impossible. */ | |
9c52057c | 1651 | BUG_ON(ret == -EEXIST || ret == -EOVERFLOW); |
8732d44f | 1652 | if (ret) { |
66642832 | 1653 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1654 | goto fail; |
1655 | } | |
42874b3d | 1656 | |
6ef06d27 | 1657 | btrfs_i_size_write(BTRFS_I(parent_inode), parent_inode->i_size + |
42874b3d | 1658 | dentry->d_name.len * 2); |
04b285f3 | 1659 | parent_inode->i_mtime = parent_inode->i_ctime = |
c2050a45 | 1660 | current_time(parent_inode); |
be6aef60 | 1661 | ret = btrfs_update_inode_fallback(trans, parent_root, parent_inode); |
dd5f9615 | 1662 | if (ret) { |
66642832 | 1663 | btrfs_abort_transaction(trans, ret); |
dd5f9615 SB |
1664 | goto fail; |
1665 | } | |
6bccf3ab | 1666 | ret = btrfs_uuid_tree_add(trans, fs_info, new_uuid.b, |
dd5f9615 SB |
1667 | BTRFS_UUID_KEY_SUBVOL, objectid); |
1668 | if (ret) { | |
66642832 | 1669 | btrfs_abort_transaction(trans, ret); |
dd5f9615 SB |
1670 | goto fail; |
1671 | } | |
1672 | if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) { | |
6bccf3ab | 1673 | ret = btrfs_uuid_tree_add(trans, fs_info, |
dd5f9615 SB |
1674 | new_root_item->received_uuid, |
1675 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, | |
1676 | objectid); | |
1677 | if (ret && ret != -EEXIST) { | |
66642832 | 1678 | btrfs_abort_transaction(trans, ret); |
dd5f9615 SB |
1679 | goto fail; |
1680 | } | |
1681 | } | |
d6726335 | 1682 | |
2ff7e61e | 1683 | ret = btrfs_run_delayed_refs(trans, fs_info, (unsigned long)-1); |
d6726335 | 1684 | if (ret) { |
66642832 | 1685 | btrfs_abort_transaction(trans, ret); |
d6726335 QW |
1686 | goto fail; |
1687 | } | |
1688 | ||
3063d29f | 1689 | fail: |
aec8030a MX |
1690 | pending->error = ret; |
1691 | dir_item_existed: | |
98c9942a | 1692 | trans->block_rsv = rsv; |
2382c5cc | 1693 | trans->bytes_reserved = 0; |
d6726335 QW |
1694 | clear_skip_qgroup: |
1695 | btrfs_clear_skip_qgroup(trans); | |
6fa9700e MX |
1696 | no_free_objectid: |
1697 | kfree(new_root_item); | |
b0c0ea63 | 1698 | pending->root_item = NULL; |
42874b3d | 1699 | btrfs_free_path(path); |
8546b570 DS |
1700 | pending->path = NULL; |
1701 | ||
49b25e05 | 1702 | return ret; |
3063d29f CM |
1703 | } |
1704 | ||
d352ac68 CM |
1705 | /* |
1706 | * create all the snapshots we've scheduled for creation | |
1707 | */ | |
80b6794d CM |
1708 | static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans, |
1709 | struct btrfs_fs_info *fs_info) | |
3de4586c | 1710 | { |
aec8030a | 1711 | struct btrfs_pending_snapshot *pending, *next; |
3de4586c | 1712 | struct list_head *head = &trans->transaction->pending_snapshots; |
aec8030a | 1713 | int ret = 0; |
3de4586c | 1714 | |
aec8030a MX |
1715 | list_for_each_entry_safe(pending, next, head, list) { |
1716 | list_del(&pending->list); | |
1717 | ret = create_pending_snapshot(trans, fs_info, pending); | |
1718 | if (ret) | |
1719 | break; | |
1720 | } | |
1721 | return ret; | |
3de4586c CM |
1722 | } |
1723 | ||
2ff7e61e | 1724 | static void update_super_roots(struct btrfs_fs_info *fs_info) |
5d4f98a2 YZ |
1725 | { |
1726 | struct btrfs_root_item *root_item; | |
1727 | struct btrfs_super_block *super; | |
1728 | ||
0b246afa | 1729 | super = fs_info->super_copy; |
5d4f98a2 | 1730 | |
0b246afa | 1731 | root_item = &fs_info->chunk_root->root_item; |
093e037c DS |
1732 | super->chunk_root = root_item->bytenr; |
1733 | super->chunk_root_generation = root_item->generation; | |
1734 | super->chunk_root_level = root_item->level; | |
5d4f98a2 | 1735 | |
0b246afa | 1736 | root_item = &fs_info->tree_root->root_item; |
093e037c DS |
1737 | super->root = root_item->bytenr; |
1738 | super->generation = root_item->generation; | |
1739 | super->root_level = root_item->level; | |
0b246afa | 1740 | if (btrfs_test_opt(fs_info, SPACE_CACHE)) |
093e037c | 1741 | super->cache_generation = root_item->generation; |
0b246afa | 1742 | if (test_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags)) |
093e037c | 1743 | super->uuid_tree_generation = root_item->generation; |
5d4f98a2 YZ |
1744 | } |
1745 | ||
f36f3042 CM |
1746 | int btrfs_transaction_in_commit(struct btrfs_fs_info *info) |
1747 | { | |
4a9d8bde | 1748 | struct btrfs_transaction *trans; |
f36f3042 | 1749 | int ret = 0; |
4a9d8bde | 1750 | |
a4abeea4 | 1751 | spin_lock(&info->trans_lock); |
4a9d8bde MX |
1752 | trans = info->running_transaction; |
1753 | if (trans) | |
1754 | ret = (trans->state >= TRANS_STATE_COMMIT_START); | |
a4abeea4 | 1755 | spin_unlock(&info->trans_lock); |
f36f3042 CM |
1756 | return ret; |
1757 | } | |
1758 | ||
8929ecfa YZ |
1759 | int btrfs_transaction_blocked(struct btrfs_fs_info *info) |
1760 | { | |
4a9d8bde | 1761 | struct btrfs_transaction *trans; |
8929ecfa | 1762 | int ret = 0; |
4a9d8bde | 1763 | |
a4abeea4 | 1764 | spin_lock(&info->trans_lock); |
4a9d8bde MX |
1765 | trans = info->running_transaction; |
1766 | if (trans) | |
1767 | ret = is_transaction_blocked(trans); | |
a4abeea4 | 1768 | spin_unlock(&info->trans_lock); |
8929ecfa YZ |
1769 | return ret; |
1770 | } | |
1771 | ||
bb9c12c9 SW |
1772 | /* |
1773 | * wait for the current transaction commit to start and block subsequent | |
1774 | * transaction joins | |
1775 | */ | |
2ff7e61e | 1776 | static void wait_current_trans_commit_start(struct btrfs_fs_info *fs_info, |
bb9c12c9 SW |
1777 | struct btrfs_transaction *trans) |
1778 | { | |
2ff7e61e JM |
1779 | wait_event(fs_info->transaction_blocked_wait, |
1780 | trans->state >= TRANS_STATE_COMMIT_START || trans->aborted); | |
bb9c12c9 SW |
1781 | } |
1782 | ||
1783 | /* | |
1784 | * wait for the current transaction to start and then become unblocked. | |
1785 | * caller holds ref. | |
1786 | */ | |
2ff7e61e JM |
1787 | static void wait_current_trans_commit_start_and_unblock( |
1788 | struct btrfs_fs_info *fs_info, | |
1789 | struct btrfs_transaction *trans) | |
bb9c12c9 | 1790 | { |
2ff7e61e JM |
1791 | wait_event(fs_info->transaction_wait, |
1792 | trans->state >= TRANS_STATE_UNBLOCKED || trans->aborted); | |
bb9c12c9 SW |
1793 | } |
1794 | ||
1795 | /* | |
1796 | * commit transactions asynchronously. once btrfs_commit_transaction_async | |
1797 | * returns, any subsequent transaction will not be allowed to join. | |
1798 | */ | |
1799 | struct btrfs_async_commit { | |
1800 | struct btrfs_trans_handle *newtrans; | |
7892b5af | 1801 | struct work_struct work; |
bb9c12c9 SW |
1802 | }; |
1803 | ||
1804 | static void do_async_commit(struct work_struct *work) | |
1805 | { | |
1806 | struct btrfs_async_commit *ac = | |
7892b5af | 1807 | container_of(work, struct btrfs_async_commit, work); |
bb9c12c9 | 1808 | |
6fc4e354 SW |
1809 | /* |
1810 | * We've got freeze protection passed with the transaction. | |
1811 | * Tell lockdep about it. | |
1812 | */ | |
b1a06a4b | 1813 | if (ac->newtrans->type & __TRANS_FREEZABLE) |
3a45bb20 | 1814 | __sb_writers_acquired(ac->newtrans->fs_info->sb, SB_FREEZE_FS); |
6fc4e354 | 1815 | |
e209db7a SW |
1816 | current->journal_info = ac->newtrans; |
1817 | ||
3a45bb20 | 1818 | btrfs_commit_transaction(ac->newtrans); |
bb9c12c9 SW |
1819 | kfree(ac); |
1820 | } | |
1821 | ||
1822 | int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans, | |
bb9c12c9 SW |
1823 | int wait_for_unblock) |
1824 | { | |
3a45bb20 | 1825 | struct btrfs_fs_info *fs_info = trans->fs_info; |
bb9c12c9 SW |
1826 | struct btrfs_async_commit *ac; |
1827 | struct btrfs_transaction *cur_trans; | |
1828 | ||
1829 | ac = kmalloc(sizeof(*ac), GFP_NOFS); | |
db5b493a TI |
1830 | if (!ac) |
1831 | return -ENOMEM; | |
bb9c12c9 | 1832 | |
7892b5af | 1833 | INIT_WORK(&ac->work, do_async_commit); |
3a45bb20 | 1834 | ac->newtrans = btrfs_join_transaction(trans->root); |
3612b495 TI |
1835 | if (IS_ERR(ac->newtrans)) { |
1836 | int err = PTR_ERR(ac->newtrans); | |
1837 | kfree(ac); | |
1838 | return err; | |
1839 | } | |
bb9c12c9 SW |
1840 | |
1841 | /* take transaction reference */ | |
bb9c12c9 | 1842 | cur_trans = trans->transaction; |
9b64f57d | 1843 | refcount_inc(&cur_trans->use_count); |
bb9c12c9 | 1844 | |
3a45bb20 | 1845 | btrfs_end_transaction(trans); |
6fc4e354 SW |
1846 | |
1847 | /* | |
1848 | * Tell lockdep we've released the freeze rwsem, since the | |
1849 | * async commit thread will be the one to unlock it. | |
1850 | */ | |
b1a06a4b | 1851 | if (ac->newtrans->type & __TRANS_FREEZABLE) |
0b246afa | 1852 | __sb_writers_release(fs_info->sb, SB_FREEZE_FS); |
6fc4e354 | 1853 | |
7892b5af | 1854 | schedule_work(&ac->work); |
bb9c12c9 SW |
1855 | |
1856 | /* wait for transaction to start and unblock */ | |
bb9c12c9 | 1857 | if (wait_for_unblock) |
2ff7e61e | 1858 | wait_current_trans_commit_start_and_unblock(fs_info, cur_trans); |
bb9c12c9 | 1859 | else |
2ff7e61e | 1860 | wait_current_trans_commit_start(fs_info, cur_trans); |
bb9c12c9 | 1861 | |
38e88054 SW |
1862 | if (current->journal_info == trans) |
1863 | current->journal_info = NULL; | |
1864 | ||
724e2315 | 1865 | btrfs_put_transaction(cur_trans); |
bb9c12c9 SW |
1866 | return 0; |
1867 | } | |
1868 | ||
49b25e05 JM |
1869 | |
1870 | static void cleanup_transaction(struct btrfs_trans_handle *trans, | |
7b8b92af | 1871 | struct btrfs_root *root, int err) |
49b25e05 | 1872 | { |
0b246afa | 1873 | struct btrfs_fs_info *fs_info = root->fs_info; |
49b25e05 | 1874 | struct btrfs_transaction *cur_trans = trans->transaction; |
f094ac32 | 1875 | DEFINE_WAIT(wait); |
49b25e05 | 1876 | |
b50fff81 | 1877 | WARN_ON(refcount_read(&trans->use_count) > 1); |
49b25e05 | 1878 | |
66642832 | 1879 | btrfs_abort_transaction(trans, err); |
7b8b92af | 1880 | |
0b246afa | 1881 | spin_lock(&fs_info->trans_lock); |
66b6135b | 1882 | |
25d8c284 MX |
1883 | /* |
1884 | * If the transaction is removed from the list, it means this | |
1885 | * transaction has been committed successfully, so it is impossible | |
1886 | * to call the cleanup function. | |
1887 | */ | |
1888 | BUG_ON(list_empty(&cur_trans->list)); | |
66b6135b | 1889 | |
49b25e05 | 1890 | list_del_init(&cur_trans->list); |
0b246afa | 1891 | if (cur_trans == fs_info->running_transaction) { |
4a9d8bde | 1892 | cur_trans->state = TRANS_STATE_COMMIT_DOING; |
0b246afa | 1893 | spin_unlock(&fs_info->trans_lock); |
f094ac32 LB |
1894 | wait_event(cur_trans->writer_wait, |
1895 | atomic_read(&cur_trans->num_writers) == 1); | |
1896 | ||
0b246afa | 1897 | spin_lock(&fs_info->trans_lock); |
d7096fc3 | 1898 | } |
0b246afa | 1899 | spin_unlock(&fs_info->trans_lock); |
49b25e05 | 1900 | |
2ff7e61e | 1901 | btrfs_cleanup_one_transaction(trans->transaction, fs_info); |
49b25e05 | 1902 | |
0b246afa JM |
1903 | spin_lock(&fs_info->trans_lock); |
1904 | if (cur_trans == fs_info->running_transaction) | |
1905 | fs_info->running_transaction = NULL; | |
1906 | spin_unlock(&fs_info->trans_lock); | |
4a9d8bde | 1907 | |
e0228285 | 1908 | if (trans->type & __TRANS_FREEZABLE) |
0b246afa | 1909 | sb_end_intwrite(fs_info->sb); |
724e2315 JB |
1910 | btrfs_put_transaction(cur_trans); |
1911 | btrfs_put_transaction(cur_trans); | |
49b25e05 JM |
1912 | |
1913 | trace_btrfs_transaction_commit(root); | |
1914 | ||
49b25e05 JM |
1915 | if (current->journal_info == trans) |
1916 | current->journal_info = NULL; | |
0b246afa | 1917 | btrfs_scrub_cancel(fs_info); |
49b25e05 JM |
1918 | |
1919 | kmem_cache_free(btrfs_trans_handle_cachep, trans); | |
1920 | } | |
1921 | ||
82436617 MX |
1922 | static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info) |
1923 | { | |
ce8ea7cc JB |
1924 | /* |
1925 | * We use writeback_inodes_sb here because if we used | |
1926 | * btrfs_start_delalloc_roots we would deadlock with fs freeze. | |
1927 | * Currently are holding the fs freeze lock, if we do an async flush | |
1928 | * we'll do btrfs_join_transaction() and deadlock because we need to | |
1929 | * wait for the fs freeze lock. Using the direct flushing we benefit | |
1930 | * from already being in a transaction and our join_transaction doesn't | |
1931 | * have to re-take the fs freeze lock. | |
1932 | */ | |
3cdde224 | 1933 | if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) |
ce8ea7cc | 1934 | writeback_inodes_sb(fs_info->sb, WB_REASON_SYNC); |
82436617 MX |
1935 | return 0; |
1936 | } | |
1937 | ||
1938 | static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info) | |
1939 | { | |
3cdde224 | 1940 | if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) |
6374e57a | 1941 | btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1); |
82436617 MX |
1942 | } |
1943 | ||
50d9aa99 | 1944 | static inline void |
161c3549 | 1945 | btrfs_wait_pending_ordered(struct btrfs_transaction *cur_trans) |
50d9aa99 | 1946 | { |
161c3549 JB |
1947 | wait_event(cur_trans->pending_wait, |
1948 | atomic_read(&cur_trans->pending_ordered) == 0); | |
50d9aa99 JB |
1949 | } |
1950 | ||
3a45bb20 | 1951 | int btrfs_commit_transaction(struct btrfs_trans_handle *trans) |
79154b1b | 1952 | { |
3a45bb20 | 1953 | struct btrfs_fs_info *fs_info = trans->fs_info; |
49b25e05 | 1954 | struct btrfs_transaction *cur_trans = trans->transaction; |
8fd17795 | 1955 | struct btrfs_transaction *prev_trans = NULL; |
25287e0a | 1956 | int ret; |
79154b1b | 1957 | |
8d25a086 | 1958 | /* Stop the commit early if ->aborted is set */ |
20c7bcec | 1959 | if (unlikely(READ_ONCE(cur_trans->aborted))) { |
25287e0a | 1960 | ret = cur_trans->aborted; |
3a45bb20 | 1961 | btrfs_end_transaction(trans); |
e4a2bcac | 1962 | return ret; |
25287e0a | 1963 | } |
49b25e05 | 1964 | |
56bec294 CM |
1965 | /* make a pass through all the delayed refs we have so far |
1966 | * any runnings procs may add more while we are here | |
1967 | */ | |
2ff7e61e | 1968 | ret = btrfs_run_delayed_refs(trans, fs_info, 0); |
e4a2bcac | 1969 | if (ret) { |
3a45bb20 | 1970 | btrfs_end_transaction(trans); |
e4a2bcac JB |
1971 | return ret; |
1972 | } | |
56bec294 | 1973 | |
dc60c525 | 1974 | btrfs_trans_release_metadata(trans); |
0e721106 JB |
1975 | trans->block_rsv = NULL; |
1976 | ||
b7ec40d7 | 1977 | cur_trans = trans->transaction; |
49b25e05 | 1978 | |
56bec294 CM |
1979 | /* |
1980 | * set the flushing flag so procs in this transaction have to | |
1981 | * start sending their work down. | |
1982 | */ | |
b7ec40d7 | 1983 | cur_trans->delayed_refs.flushing = 1; |
1be41b78 | 1984 | smp_wmb(); |
56bec294 | 1985 | |
ea658bad | 1986 | if (!list_empty(&trans->new_bgs)) |
6c686b35 | 1987 | btrfs_create_pending_block_groups(trans); |
ea658bad | 1988 | |
2ff7e61e | 1989 | ret = btrfs_run_delayed_refs(trans, fs_info, 0); |
e4a2bcac | 1990 | if (ret) { |
3a45bb20 | 1991 | btrfs_end_transaction(trans); |
e4a2bcac JB |
1992 | return ret; |
1993 | } | |
56bec294 | 1994 | |
3204d33c | 1995 | if (!test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &cur_trans->flags)) { |
1bbc621e CM |
1996 | int run_it = 0; |
1997 | ||
1998 | /* this mutex is also taken before trying to set | |
1999 | * block groups readonly. We need to make sure | |
2000 | * that nobody has set a block group readonly | |
2001 | * after a extents from that block group have been | |
2002 | * allocated for cache files. btrfs_set_block_group_ro | |
2003 | * will wait for the transaction to commit if it | |
3204d33c | 2004 | * finds BTRFS_TRANS_DIRTY_BG_RUN set. |
1bbc621e | 2005 | * |
3204d33c JB |
2006 | * The BTRFS_TRANS_DIRTY_BG_RUN flag is also used to make sure |
2007 | * only one process starts all the block group IO. It wouldn't | |
1bbc621e CM |
2008 | * hurt to have more than one go through, but there's no |
2009 | * real advantage to it either. | |
2010 | */ | |
0b246afa | 2011 | mutex_lock(&fs_info->ro_block_group_mutex); |
3204d33c JB |
2012 | if (!test_and_set_bit(BTRFS_TRANS_DIRTY_BG_RUN, |
2013 | &cur_trans->flags)) | |
1bbc621e | 2014 | run_it = 1; |
0b246afa | 2015 | mutex_unlock(&fs_info->ro_block_group_mutex); |
1bbc621e CM |
2016 | |
2017 | if (run_it) | |
21217054 | 2018 | ret = btrfs_start_dirty_block_groups(trans); |
1bbc621e CM |
2019 | } |
2020 | if (ret) { | |
3a45bb20 | 2021 | btrfs_end_transaction(trans); |
1bbc621e CM |
2022 | return ret; |
2023 | } | |
2024 | ||
0b246afa | 2025 | spin_lock(&fs_info->trans_lock); |
4a9d8bde | 2026 | if (cur_trans->state >= TRANS_STATE_COMMIT_START) { |
0b246afa | 2027 | spin_unlock(&fs_info->trans_lock); |
9b64f57d | 2028 | refcount_inc(&cur_trans->use_count); |
3a45bb20 | 2029 | ret = btrfs_end_transaction(trans); |
ccd467d6 | 2030 | |
2ff7e61e | 2031 | wait_for_commit(cur_trans); |
15ee9bc7 | 2032 | |
b4924a0f LB |
2033 | if (unlikely(cur_trans->aborted)) |
2034 | ret = cur_trans->aborted; | |
2035 | ||
724e2315 | 2036 | btrfs_put_transaction(cur_trans); |
15ee9bc7 | 2037 | |
49b25e05 | 2038 | return ret; |
79154b1b | 2039 | } |
4313b399 | 2040 | |
4a9d8bde | 2041 | cur_trans->state = TRANS_STATE_COMMIT_START; |
0b246afa | 2042 | wake_up(&fs_info->transaction_blocked_wait); |
bb9c12c9 | 2043 | |
0b246afa | 2044 | if (cur_trans->list.prev != &fs_info->trans_list) { |
ccd467d6 CM |
2045 | prev_trans = list_entry(cur_trans->list.prev, |
2046 | struct btrfs_transaction, list); | |
4a9d8bde | 2047 | if (prev_trans->state != TRANS_STATE_COMPLETED) { |
9b64f57d | 2048 | refcount_inc(&prev_trans->use_count); |
0b246afa | 2049 | spin_unlock(&fs_info->trans_lock); |
ccd467d6 | 2050 | |
2ff7e61e | 2051 | wait_for_commit(prev_trans); |
1f9b8c8f | 2052 | ret = prev_trans->aborted; |
ccd467d6 | 2053 | |
724e2315 | 2054 | btrfs_put_transaction(prev_trans); |
1f9b8c8f FM |
2055 | if (ret) |
2056 | goto cleanup_transaction; | |
a4abeea4 | 2057 | } else { |
0b246afa | 2058 | spin_unlock(&fs_info->trans_lock); |
ccd467d6 | 2059 | } |
a4abeea4 | 2060 | } else { |
0b246afa | 2061 | spin_unlock(&fs_info->trans_lock); |
ccd467d6 | 2062 | } |
15ee9bc7 | 2063 | |
0860adfd MX |
2064 | extwriter_counter_dec(cur_trans, trans->type); |
2065 | ||
0b246afa | 2066 | ret = btrfs_start_delalloc_flush(fs_info); |
82436617 MX |
2067 | if (ret) |
2068 | goto cleanup_transaction; | |
2069 | ||
e5c304e6 | 2070 | ret = btrfs_run_delayed_items(trans); |
581227d0 MX |
2071 | if (ret) |
2072 | goto cleanup_transaction; | |
15ee9bc7 | 2073 | |
581227d0 MX |
2074 | wait_event(cur_trans->writer_wait, |
2075 | extwriter_counter_read(cur_trans) == 0); | |
15ee9bc7 | 2076 | |
581227d0 | 2077 | /* some pending stuffs might be added after the previous flush. */ |
e5c304e6 | 2078 | ret = btrfs_run_delayed_items(trans); |
ca469637 MX |
2079 | if (ret) |
2080 | goto cleanup_transaction; | |
2081 | ||
0b246afa | 2082 | btrfs_wait_delalloc_flush(fs_info); |
cb7ab021 | 2083 | |
161c3549 | 2084 | btrfs_wait_pending_ordered(cur_trans); |
50d9aa99 | 2085 | |
2ff7e61e | 2086 | btrfs_scrub_pause(fs_info); |
ed0ca140 JB |
2087 | /* |
2088 | * Ok now we need to make sure to block out any other joins while we | |
2089 | * commit the transaction. We could have started a join before setting | |
4a9d8bde | 2090 | * COMMIT_DOING so make sure to wait for num_writers to == 1 again. |
ed0ca140 | 2091 | */ |
0b246afa | 2092 | spin_lock(&fs_info->trans_lock); |
4a9d8bde | 2093 | cur_trans->state = TRANS_STATE_COMMIT_DOING; |
0b246afa | 2094 | spin_unlock(&fs_info->trans_lock); |
ed0ca140 JB |
2095 | wait_event(cur_trans->writer_wait, |
2096 | atomic_read(&cur_trans->num_writers) == 1); | |
2097 | ||
2cba30f1 | 2098 | /* ->aborted might be set after the previous check, so check it */ |
20c7bcec | 2099 | if (unlikely(READ_ONCE(cur_trans->aborted))) { |
2cba30f1 | 2100 | ret = cur_trans->aborted; |
6cf7f77e | 2101 | goto scrub_continue; |
2cba30f1 | 2102 | } |
7585717f CM |
2103 | /* |
2104 | * the reloc mutex makes sure that we stop | |
2105 | * the balancing code from coming in and moving | |
2106 | * extents around in the middle of the commit | |
2107 | */ | |
0b246afa | 2108 | mutex_lock(&fs_info->reloc_mutex); |
7585717f | 2109 | |
42874b3d MX |
2110 | /* |
2111 | * We needn't worry about the delayed items because we will | |
2112 | * deal with them in create_pending_snapshot(), which is the | |
2113 | * core function of the snapshot creation. | |
2114 | */ | |
0b246afa | 2115 | ret = create_pending_snapshots(trans, fs_info); |
49b25e05 | 2116 | if (ret) { |
0b246afa | 2117 | mutex_unlock(&fs_info->reloc_mutex); |
6cf7f77e | 2118 | goto scrub_continue; |
49b25e05 | 2119 | } |
3063d29f | 2120 | |
42874b3d MX |
2121 | /* |
2122 | * We insert the dir indexes of the snapshots and update the inode | |
2123 | * of the snapshots' parents after the snapshot creation, so there | |
2124 | * are some delayed items which are not dealt with. Now deal with | |
2125 | * them. | |
2126 | * | |
2127 | * We needn't worry that this operation will corrupt the snapshots, | |
2128 | * because all the tree which are snapshoted will be forced to COW | |
2129 | * the nodes and leaves. | |
2130 | */ | |
e5c304e6 | 2131 | ret = btrfs_run_delayed_items(trans); |
49b25e05 | 2132 | if (ret) { |
0b246afa | 2133 | mutex_unlock(&fs_info->reloc_mutex); |
6cf7f77e | 2134 | goto scrub_continue; |
49b25e05 | 2135 | } |
16cdcec7 | 2136 | |
2ff7e61e | 2137 | ret = btrfs_run_delayed_refs(trans, fs_info, (unsigned long)-1); |
49b25e05 | 2138 | if (ret) { |
0b246afa | 2139 | mutex_unlock(&fs_info->reloc_mutex); |
6cf7f77e | 2140 | goto scrub_continue; |
49b25e05 | 2141 | } |
56bec294 | 2142 | |
e999376f CM |
2143 | /* |
2144 | * make sure none of the code above managed to slip in a | |
2145 | * delayed item | |
2146 | */ | |
ccdf9b30 | 2147 | btrfs_assert_delayed_root_empty(fs_info); |
e999376f | 2148 | |
2c90e5d6 | 2149 | WARN_ON(cur_trans != trans->transaction); |
dc17ff8f | 2150 | |
e02119d5 CM |
2151 | /* btrfs_commit_tree_roots is responsible for getting the |
2152 | * various roots consistent with each other. Every pointer | |
2153 | * in the tree of tree roots has to point to the most up to date | |
2154 | * root for every subvolume and other tree. So, we have to keep | |
2155 | * the tree logging code from jumping in and changing any | |
2156 | * of the trees. | |
2157 | * | |
2158 | * At this point in the commit, there can't be any tree-log | |
2159 | * writers, but a little lower down we drop the trans mutex | |
2160 | * and let new people in. By holding the tree_log_mutex | |
2161 | * from now until after the super is written, we avoid races | |
2162 | * with the tree-log code. | |
2163 | */ | |
0b246afa | 2164 | mutex_lock(&fs_info->tree_log_mutex); |
e02119d5 | 2165 | |
0b246afa | 2166 | ret = commit_fs_roots(trans, fs_info); |
49b25e05 | 2167 | if (ret) { |
0b246afa JM |
2168 | mutex_unlock(&fs_info->tree_log_mutex); |
2169 | mutex_unlock(&fs_info->reloc_mutex); | |
6cf7f77e | 2170 | goto scrub_continue; |
49b25e05 | 2171 | } |
54aa1f4d | 2172 | |
3818aea2 | 2173 | /* |
7e1876ac DS |
2174 | * Since the transaction is done, we can apply the pending changes |
2175 | * before the next transaction. | |
3818aea2 | 2176 | */ |
0b246afa | 2177 | btrfs_apply_pending_changes(fs_info); |
3818aea2 | 2178 | |
5d4f98a2 | 2179 | /* commit_fs_roots gets rid of all the tree log roots, it is now |
e02119d5 CM |
2180 | * safe to free the root of tree log roots |
2181 | */ | |
0b246afa | 2182 | btrfs_free_log_root_tree(trans, fs_info); |
e02119d5 | 2183 | |
82bafb38 QW |
2184 | /* |
2185 | * commit_fs_roots() can call btrfs_save_ino_cache(), which generates | |
2186 | * new delayed refs. Must handle them or qgroup can be wrong. | |
2187 | */ | |
2188 | ret = btrfs_run_delayed_refs(trans, fs_info, (unsigned long)-1); | |
2189 | if (ret) { | |
2190 | mutex_unlock(&fs_info->tree_log_mutex); | |
2191 | mutex_unlock(&fs_info->reloc_mutex); | |
2192 | goto scrub_continue; | |
2193 | } | |
2194 | ||
0ed4792a QW |
2195 | /* |
2196 | * Since fs roots are all committed, we can get a quite accurate | |
2197 | * new_roots. So let's do quota accounting. | |
2198 | */ | |
0b246afa | 2199 | ret = btrfs_qgroup_account_extents(trans, fs_info); |
0ed4792a | 2200 | if (ret < 0) { |
0b246afa JM |
2201 | mutex_unlock(&fs_info->tree_log_mutex); |
2202 | mutex_unlock(&fs_info->reloc_mutex); | |
0ed4792a QW |
2203 | goto scrub_continue; |
2204 | } | |
2205 | ||
2ff7e61e | 2206 | ret = commit_cowonly_roots(trans, fs_info); |
49b25e05 | 2207 | if (ret) { |
0b246afa JM |
2208 | mutex_unlock(&fs_info->tree_log_mutex); |
2209 | mutex_unlock(&fs_info->reloc_mutex); | |
6cf7f77e | 2210 | goto scrub_continue; |
49b25e05 | 2211 | } |
54aa1f4d | 2212 | |
2cba30f1 MX |
2213 | /* |
2214 | * The tasks which save the space cache and inode cache may also | |
2215 | * update ->aborted, check it. | |
2216 | */ | |
20c7bcec | 2217 | if (unlikely(READ_ONCE(cur_trans->aborted))) { |
2cba30f1 | 2218 | ret = cur_trans->aborted; |
0b246afa JM |
2219 | mutex_unlock(&fs_info->tree_log_mutex); |
2220 | mutex_unlock(&fs_info->reloc_mutex); | |
6cf7f77e | 2221 | goto scrub_continue; |
2cba30f1 MX |
2222 | } |
2223 | ||
8b74c03e | 2224 | btrfs_prepare_extent_commit(fs_info); |
11833d66 | 2225 | |
0b246afa | 2226 | cur_trans = fs_info->running_transaction; |
5d4f98a2 | 2227 | |
0b246afa JM |
2228 | btrfs_set_root_node(&fs_info->tree_root->root_item, |
2229 | fs_info->tree_root->node); | |
2230 | list_add_tail(&fs_info->tree_root->dirty_list, | |
9e351cc8 | 2231 | &cur_trans->switch_commits); |
5d4f98a2 | 2232 | |
0b246afa JM |
2233 | btrfs_set_root_node(&fs_info->chunk_root->root_item, |
2234 | fs_info->chunk_root->node); | |
2235 | list_add_tail(&fs_info->chunk_root->dirty_list, | |
9e351cc8 JB |
2236 | &cur_trans->switch_commits); |
2237 | ||
0b246afa | 2238 | switch_commit_roots(cur_trans, fs_info); |
5d4f98a2 | 2239 | |
ce93ec54 | 2240 | ASSERT(list_empty(&cur_trans->dirty_bgs)); |
1bbc621e | 2241 | ASSERT(list_empty(&cur_trans->io_bgs)); |
2ff7e61e | 2242 | update_super_roots(fs_info); |
e02119d5 | 2243 | |
0b246afa JM |
2244 | btrfs_set_super_log_root(fs_info->super_copy, 0); |
2245 | btrfs_set_super_log_root_level(fs_info->super_copy, 0); | |
2246 | memcpy(fs_info->super_for_commit, fs_info->super_copy, | |
2247 | sizeof(*fs_info->super_copy)); | |
ccd467d6 | 2248 | |
0b246afa | 2249 | btrfs_update_commit_device_size(fs_info); |
2ff7e61e | 2250 | btrfs_update_commit_device_bytes_used(fs_info, cur_trans); |
935e5cc9 | 2251 | |
0b246afa JM |
2252 | clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags); |
2253 | clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags); | |
656f30db | 2254 | |
4fbcdf66 FM |
2255 | btrfs_trans_release_chunk_metadata(trans); |
2256 | ||
0b246afa | 2257 | spin_lock(&fs_info->trans_lock); |
4a9d8bde | 2258 | cur_trans->state = TRANS_STATE_UNBLOCKED; |
0b246afa JM |
2259 | fs_info->running_transaction = NULL; |
2260 | spin_unlock(&fs_info->trans_lock); | |
2261 | mutex_unlock(&fs_info->reloc_mutex); | |
b7ec40d7 | 2262 | |
0b246afa | 2263 | wake_up(&fs_info->transaction_wait); |
e6dcd2dc | 2264 | |
bf89d38f | 2265 | ret = btrfs_write_and_wait_transaction(trans, fs_info); |
49b25e05 | 2266 | if (ret) { |
0b246afa JM |
2267 | btrfs_handle_fs_error(fs_info, ret, |
2268 | "Error while writing out transaction"); | |
2269 | mutex_unlock(&fs_info->tree_log_mutex); | |
6cf7f77e | 2270 | goto scrub_continue; |
49b25e05 JM |
2271 | } |
2272 | ||
eece6a9c | 2273 | ret = write_all_supers(fs_info, 0); |
e02119d5 CM |
2274 | /* |
2275 | * the super is written, we can safely allow the tree-loggers | |
2276 | * to go about their business | |
2277 | */ | |
0b246afa | 2278 | mutex_unlock(&fs_info->tree_log_mutex); |
c1f32b7c AJ |
2279 | if (ret) |
2280 | goto scrub_continue; | |
e02119d5 | 2281 | |
2ff7e61e | 2282 | btrfs_finish_extent_commit(trans, fs_info); |
4313b399 | 2283 | |
3204d33c | 2284 | if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS, &cur_trans->flags)) |
0b246afa | 2285 | btrfs_clear_space_info_full(fs_info); |
13212b54 | 2286 | |
0b246afa | 2287 | fs_info->last_trans_committed = cur_trans->transid; |
4a9d8bde MX |
2288 | /* |
2289 | * We needn't acquire the lock here because there is no other task | |
2290 | * which can change it. | |
2291 | */ | |
2292 | cur_trans->state = TRANS_STATE_COMPLETED; | |
2c90e5d6 | 2293 | wake_up(&cur_trans->commit_wait); |
3de4586c | 2294 | |
0b246afa | 2295 | spin_lock(&fs_info->trans_lock); |
13c5a93e | 2296 | list_del_init(&cur_trans->list); |
0b246afa | 2297 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 | 2298 | |
724e2315 JB |
2299 | btrfs_put_transaction(cur_trans); |
2300 | btrfs_put_transaction(cur_trans); | |
58176a96 | 2301 | |
0860adfd | 2302 | if (trans->type & __TRANS_FREEZABLE) |
0b246afa | 2303 | sb_end_intwrite(fs_info->sb); |
b2b5ef5c | 2304 | |
3a45bb20 | 2305 | trace_btrfs_transaction_commit(trans->root); |
1abe9b8a | 2306 | |
2ff7e61e | 2307 | btrfs_scrub_continue(fs_info); |
a2de733c | 2308 | |
9ed74f2d JB |
2309 | if (current->journal_info == trans) |
2310 | current->journal_info = NULL; | |
2311 | ||
2c90e5d6 | 2312 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
24bbcf04 | 2313 | |
9e7cc91a WX |
2314 | /* |
2315 | * If fs has been frozen, we can not handle delayed iputs, otherwise | |
2316 | * it'll result in deadlock about SB_FREEZE_FS. | |
2317 | */ | |
0b246afa | 2318 | if (current != fs_info->transaction_kthread && |
fac03c8d DS |
2319 | current != fs_info->cleaner_kthread && |
2320 | !test_bit(BTRFS_FS_FROZEN, &fs_info->flags)) | |
2ff7e61e | 2321 | btrfs_run_delayed_iputs(fs_info); |
24bbcf04 | 2322 | |
79154b1b | 2323 | return ret; |
49b25e05 | 2324 | |
6cf7f77e | 2325 | scrub_continue: |
2ff7e61e | 2326 | btrfs_scrub_continue(fs_info); |
49b25e05 | 2327 | cleanup_transaction: |
dc60c525 | 2328 | btrfs_trans_release_metadata(trans); |
4fbcdf66 | 2329 | btrfs_trans_release_chunk_metadata(trans); |
0e721106 | 2330 | trans->block_rsv = NULL; |
0b246afa | 2331 | btrfs_warn(fs_info, "Skipping commit of aborted transaction."); |
49b25e05 JM |
2332 | if (current->journal_info == trans) |
2333 | current->journal_info = NULL; | |
3a45bb20 | 2334 | cleanup_transaction(trans, trans->root, ret); |
49b25e05 JM |
2335 | |
2336 | return ret; | |
79154b1b CM |
2337 | } |
2338 | ||
d352ac68 | 2339 | /* |
9d1a2a3a DS |
2340 | * return < 0 if error |
2341 | * 0 if there are no more dead_roots at the time of call | |
2342 | * 1 there are more to be processed, call me again | |
2343 | * | |
2344 | * The return value indicates there are certainly more snapshots to delete, but | |
2345 | * if there comes a new one during processing, it may return 0. We don't mind, | |
2346 | * because btrfs_commit_super will poke cleaner thread and it will process it a | |
2347 | * few seconds later. | |
d352ac68 | 2348 | */ |
9d1a2a3a | 2349 | int btrfs_clean_one_deleted_snapshot(struct btrfs_root *root) |
e9d0b13b | 2350 | { |
9d1a2a3a | 2351 | int ret; |
5d4f98a2 YZ |
2352 | struct btrfs_fs_info *fs_info = root->fs_info; |
2353 | ||
a4abeea4 | 2354 | spin_lock(&fs_info->trans_lock); |
9d1a2a3a DS |
2355 | if (list_empty(&fs_info->dead_roots)) { |
2356 | spin_unlock(&fs_info->trans_lock); | |
2357 | return 0; | |
2358 | } | |
2359 | root = list_first_entry(&fs_info->dead_roots, | |
2360 | struct btrfs_root, root_list); | |
cfad392b | 2361 | list_del_init(&root->root_list); |
a4abeea4 | 2362 | spin_unlock(&fs_info->trans_lock); |
e9d0b13b | 2363 | |
ab8d0fc4 | 2364 | btrfs_debug(fs_info, "cleaner removing %llu", root->objectid); |
76dda93c | 2365 | |
9d1a2a3a | 2366 | btrfs_kill_all_delayed_nodes(root); |
16cdcec7 | 2367 | |
9d1a2a3a DS |
2368 | if (btrfs_header_backref_rev(root->node) < |
2369 | BTRFS_MIXED_BACKREF_REV) | |
2370 | ret = btrfs_drop_snapshot(root, NULL, 0, 0); | |
2371 | else | |
2372 | ret = btrfs_drop_snapshot(root, NULL, 1, 0); | |
32471dc2 | 2373 | |
6596a928 | 2374 | return (ret < 0) ? 0 : 1; |
e9d0b13b | 2375 | } |
572d9ab7 DS |
2376 | |
2377 | void btrfs_apply_pending_changes(struct btrfs_fs_info *fs_info) | |
2378 | { | |
2379 | unsigned long prev; | |
2380 | unsigned long bit; | |
2381 | ||
6c9fe14f | 2382 | prev = xchg(&fs_info->pending_changes, 0); |
572d9ab7 DS |
2383 | if (!prev) |
2384 | return; | |
2385 | ||
7e1876ac DS |
2386 | bit = 1 << BTRFS_PENDING_SET_INODE_MAP_CACHE; |
2387 | if (prev & bit) | |
2388 | btrfs_set_opt(fs_info->mount_opt, INODE_MAP_CACHE); | |
2389 | prev &= ~bit; | |
2390 | ||
2391 | bit = 1 << BTRFS_PENDING_CLEAR_INODE_MAP_CACHE; | |
2392 | if (prev & bit) | |
2393 | btrfs_clear_opt(fs_info->mount_opt, INODE_MAP_CACHE); | |
2394 | prev &= ~bit; | |
2395 | ||
d51033d0 DS |
2396 | bit = 1 << BTRFS_PENDING_COMMIT; |
2397 | if (prev & bit) | |
2398 | btrfs_debug(fs_info, "pending commit done"); | |
2399 | prev &= ~bit; | |
2400 | ||
572d9ab7 DS |
2401 | if (prev) |
2402 | btrfs_warn(fs_info, | |
2403 | "unknown pending changes left 0x%lx, ignoring", prev); | |
2404 | } |