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 | ||
4a9d8bde MX |
38 | static unsigned int btrfs_blocked_trans_types[TRANS_STATE_MAX] = { |
39 | [TRANS_STATE_RUNNING] = 0U, | |
40 | [TRANS_STATE_BLOCKED] = (__TRANS_USERSPACE | | |
41 | __TRANS_START), | |
42 | [TRANS_STATE_COMMIT_START] = (__TRANS_USERSPACE | | |
43 | __TRANS_START | | |
44 | __TRANS_ATTACH), | |
45 | [TRANS_STATE_COMMIT_DOING] = (__TRANS_USERSPACE | | |
46 | __TRANS_START | | |
47 | __TRANS_ATTACH | | |
48 | __TRANS_JOIN), | |
49 | [TRANS_STATE_UNBLOCKED] = (__TRANS_USERSPACE | | |
50 | __TRANS_START | | |
51 | __TRANS_ATTACH | | |
52 | __TRANS_JOIN | | |
53 | __TRANS_JOIN_NOLOCK), | |
54 | [TRANS_STATE_COMPLETED] = (__TRANS_USERSPACE | | |
55 | __TRANS_START | | |
56 | __TRANS_ATTACH | | |
57 | __TRANS_JOIN | | |
58 | __TRANS_JOIN_NOLOCK), | |
59 | }; | |
60 | ||
724e2315 | 61 | void btrfs_put_transaction(struct btrfs_transaction *transaction) |
79154b1b | 62 | { |
13c5a93e JB |
63 | WARN_ON(atomic_read(&transaction->use_count) == 0); |
64 | if (atomic_dec_and_test(&transaction->use_count)) { | |
a4abeea4 | 65 | BUG_ON(!list_empty(&transaction->list)); |
c46effa6 | 66 | WARN_ON(!RB_EMPTY_ROOT(&transaction->delayed_refs.href_root)); |
6df9a95e JB |
67 | while (!list_empty(&transaction->pending_chunks)) { |
68 | struct extent_map *em; | |
69 | ||
70 | em = list_first_entry(&transaction->pending_chunks, | |
71 | struct extent_map, list); | |
72 | list_del_init(&em->list); | |
73 | free_extent_map(em); | |
74 | } | |
2c90e5d6 | 75 | kmem_cache_free(btrfs_transaction_cachep, transaction); |
78fae27e | 76 | } |
79154b1b CM |
77 | } |
78 | ||
663dfbb0 FM |
79 | static void clear_btree_io_tree(struct extent_io_tree *tree) |
80 | { | |
81 | spin_lock(&tree->lock); | |
82 | while (!RB_EMPTY_ROOT(&tree->state)) { | |
83 | struct rb_node *node; | |
84 | struct extent_state *state; | |
85 | ||
86 | node = rb_first(&tree->state); | |
87 | state = rb_entry(node, struct extent_state, rb_node); | |
88 | rb_erase(&state->rb_node, &tree->state); | |
89 | RB_CLEAR_NODE(&state->rb_node); | |
90 | /* | |
91 | * btree io trees aren't supposed to have tasks waiting for | |
92 | * changes in the flags of extent states ever. | |
93 | */ | |
94 | ASSERT(!waitqueue_active(&state->wq)); | |
95 | free_extent_state(state); | |
96 | if (need_resched()) { | |
97 | spin_unlock(&tree->lock); | |
98 | cond_resched(); | |
99 | spin_lock(&tree->lock); | |
100 | } | |
101 | } | |
102 | spin_unlock(&tree->lock); | |
103 | } | |
104 | ||
9e351cc8 JB |
105 | static noinline void switch_commit_roots(struct btrfs_transaction *trans, |
106 | struct btrfs_fs_info *fs_info) | |
817d52f8 | 107 | { |
9e351cc8 JB |
108 | struct btrfs_root *root, *tmp; |
109 | ||
110 | down_write(&fs_info->commit_root_sem); | |
111 | list_for_each_entry_safe(root, tmp, &trans->switch_commits, | |
112 | dirty_list) { | |
113 | list_del_init(&root->dirty_list); | |
114 | free_extent_buffer(root->commit_root); | |
115 | root->commit_root = btrfs_root_node(root); | |
116 | if (is_fstree(root->objectid)) | |
117 | btrfs_unpin_free_ino(root); | |
663dfbb0 | 118 | clear_btree_io_tree(&root->dirty_log_pages); |
9e351cc8 JB |
119 | } |
120 | up_write(&fs_info->commit_root_sem); | |
817d52f8 JB |
121 | } |
122 | ||
0860adfd MX |
123 | static inline void extwriter_counter_inc(struct btrfs_transaction *trans, |
124 | unsigned int type) | |
125 | { | |
126 | if (type & TRANS_EXTWRITERS) | |
127 | atomic_inc(&trans->num_extwriters); | |
128 | } | |
129 | ||
130 | static inline void extwriter_counter_dec(struct btrfs_transaction *trans, | |
131 | unsigned int type) | |
132 | { | |
133 | if (type & TRANS_EXTWRITERS) | |
134 | atomic_dec(&trans->num_extwriters); | |
135 | } | |
136 | ||
137 | static inline void extwriter_counter_init(struct btrfs_transaction *trans, | |
138 | unsigned int type) | |
139 | { | |
140 | atomic_set(&trans->num_extwriters, ((type & TRANS_EXTWRITERS) ? 1 : 0)); | |
141 | } | |
142 | ||
143 | static inline int extwriter_counter_read(struct btrfs_transaction *trans) | |
144 | { | |
145 | return atomic_read(&trans->num_extwriters); | |
178260b2 MX |
146 | } |
147 | ||
d352ac68 CM |
148 | /* |
149 | * either allocate a new transaction or hop into the existing one | |
150 | */ | |
0860adfd | 151 | static noinline int join_transaction(struct btrfs_root *root, unsigned int type) |
79154b1b CM |
152 | { |
153 | struct btrfs_transaction *cur_trans; | |
19ae4e81 | 154 | struct btrfs_fs_info *fs_info = root->fs_info; |
a4abeea4 | 155 | |
19ae4e81 | 156 | spin_lock(&fs_info->trans_lock); |
d43317dc | 157 | loop: |
49b25e05 | 158 | /* The file system has been taken offline. No new transactions. */ |
87533c47 | 159 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { |
19ae4e81 | 160 | spin_unlock(&fs_info->trans_lock); |
49b25e05 JM |
161 | return -EROFS; |
162 | } | |
163 | ||
19ae4e81 | 164 | cur_trans = fs_info->running_transaction; |
a4abeea4 | 165 | if (cur_trans) { |
871383be | 166 | if (cur_trans->aborted) { |
19ae4e81 | 167 | spin_unlock(&fs_info->trans_lock); |
49b25e05 | 168 | return cur_trans->aborted; |
871383be | 169 | } |
4a9d8bde | 170 | if (btrfs_blocked_trans_types[cur_trans->state] & type) { |
178260b2 MX |
171 | spin_unlock(&fs_info->trans_lock); |
172 | return -EBUSY; | |
173 | } | |
a4abeea4 | 174 | atomic_inc(&cur_trans->use_count); |
13c5a93e | 175 | atomic_inc(&cur_trans->num_writers); |
0860adfd | 176 | extwriter_counter_inc(cur_trans, type); |
19ae4e81 | 177 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 | 178 | return 0; |
79154b1b | 179 | } |
19ae4e81 | 180 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 | 181 | |
354aa0fb MX |
182 | /* |
183 | * If we are ATTACH, we just want to catch the current transaction, | |
184 | * and commit it. If there is no transaction, just return ENOENT. | |
185 | */ | |
186 | if (type == TRANS_ATTACH) | |
187 | return -ENOENT; | |
188 | ||
4a9d8bde MX |
189 | /* |
190 | * JOIN_NOLOCK only happens during the transaction commit, so | |
191 | * it is impossible that ->running_transaction is NULL | |
192 | */ | |
193 | BUG_ON(type == TRANS_JOIN_NOLOCK); | |
194 | ||
a4abeea4 JB |
195 | cur_trans = kmem_cache_alloc(btrfs_transaction_cachep, GFP_NOFS); |
196 | if (!cur_trans) | |
197 | return -ENOMEM; | |
d43317dc | 198 | |
19ae4e81 JS |
199 | spin_lock(&fs_info->trans_lock); |
200 | if (fs_info->running_transaction) { | |
d43317dc CM |
201 | /* |
202 | * someone started a transaction after we unlocked. Make sure | |
4a9d8bde | 203 | * to redo the checks above |
d43317dc | 204 | */ |
a4abeea4 | 205 | kmem_cache_free(btrfs_transaction_cachep, cur_trans); |
d43317dc | 206 | goto loop; |
87533c47 | 207 | } else if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { |
e4b50e14 | 208 | spin_unlock(&fs_info->trans_lock); |
7b8b92af JB |
209 | kmem_cache_free(btrfs_transaction_cachep, cur_trans); |
210 | return -EROFS; | |
79154b1b | 211 | } |
d43317dc | 212 | |
a4abeea4 | 213 | atomic_set(&cur_trans->num_writers, 1); |
0860adfd | 214 | extwriter_counter_init(cur_trans, type); |
a4abeea4 JB |
215 | init_waitqueue_head(&cur_trans->writer_wait); |
216 | init_waitqueue_head(&cur_trans->commit_wait); | |
4a9d8bde | 217 | cur_trans->state = TRANS_STATE_RUNNING; |
a4abeea4 JB |
218 | /* |
219 | * One for this trans handle, one so it will live on until we | |
220 | * commit the transaction. | |
221 | */ | |
222 | atomic_set(&cur_trans->use_count, 2); | |
a4abeea4 JB |
223 | cur_trans->start_time = get_seconds(); |
224 | ||
c46effa6 | 225 | cur_trans->delayed_refs.href_root = RB_ROOT; |
d7df2c79 | 226 | atomic_set(&cur_trans->delayed_refs.num_entries, 0); |
a4abeea4 JB |
227 | cur_trans->delayed_refs.num_heads_ready = 0; |
228 | cur_trans->delayed_refs.num_heads = 0; | |
229 | cur_trans->delayed_refs.flushing = 0; | |
230 | cur_trans->delayed_refs.run_delayed_start = 0; | |
20b297d6 JS |
231 | |
232 | /* | |
233 | * although the tree mod log is per file system and not per transaction, | |
234 | * the log must never go across transaction boundaries. | |
235 | */ | |
236 | smp_mb(); | |
31b1a2bd | 237 | if (!list_empty(&fs_info->tree_mod_seq_list)) |
efe120a0 | 238 | WARN(1, KERN_ERR "BTRFS: tree_mod_seq_list not empty when " |
20b297d6 | 239 | "creating a fresh transaction\n"); |
31b1a2bd | 240 | if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log)) |
efe120a0 | 241 | WARN(1, KERN_ERR "BTRFS: tree_mod_log rb tree not empty when " |
20b297d6 | 242 | "creating a fresh transaction\n"); |
fc36ed7e | 243 | atomic64_set(&fs_info->tree_mod_seq, 0); |
20b297d6 | 244 | |
a4abeea4 JB |
245 | spin_lock_init(&cur_trans->delayed_refs.lock); |
246 | ||
247 | INIT_LIST_HEAD(&cur_trans->pending_snapshots); | |
6df9a95e | 248 | INIT_LIST_HEAD(&cur_trans->pending_chunks); |
9e351cc8 | 249 | INIT_LIST_HEAD(&cur_trans->switch_commits); |
50d9aa99 | 250 | INIT_LIST_HEAD(&cur_trans->pending_ordered); |
ce93ec54 JB |
251 | INIT_LIST_HEAD(&cur_trans->dirty_bgs); |
252 | spin_lock_init(&cur_trans->dirty_bgs_lock); | |
19ae4e81 | 253 | list_add_tail(&cur_trans->list, &fs_info->trans_list); |
a4abeea4 | 254 | extent_io_tree_init(&cur_trans->dirty_pages, |
19ae4e81 JS |
255 | fs_info->btree_inode->i_mapping); |
256 | fs_info->generation++; | |
257 | cur_trans->transid = fs_info->generation; | |
258 | fs_info->running_transaction = cur_trans; | |
49b25e05 | 259 | cur_trans->aborted = 0; |
19ae4e81 | 260 | spin_unlock(&fs_info->trans_lock); |
15ee9bc7 | 261 | |
79154b1b CM |
262 | return 0; |
263 | } | |
264 | ||
d352ac68 | 265 | /* |
d397712b CM |
266 | * this does all the record keeping required to make sure that a reference |
267 | * counted root is properly recorded in a given transaction. This is required | |
268 | * to make sure the old root from before we joined the transaction is deleted | |
269 | * when the transaction commits | |
d352ac68 | 270 | */ |
7585717f | 271 | static int record_root_in_trans(struct btrfs_trans_handle *trans, |
a4abeea4 | 272 | struct btrfs_root *root) |
6702ed49 | 273 | { |
27cdeb70 MX |
274 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
275 | root->last_trans < trans->transid) { | |
6702ed49 | 276 | WARN_ON(root == root->fs_info->extent_root); |
5d4f98a2 YZ |
277 | WARN_ON(root->commit_root != root->node); |
278 | ||
7585717f | 279 | /* |
27cdeb70 | 280 | * see below for IN_TRANS_SETUP usage rules |
7585717f CM |
281 | * we have the reloc mutex held now, so there |
282 | * is only one writer in this function | |
283 | */ | |
27cdeb70 | 284 | set_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); |
7585717f | 285 | |
27cdeb70 | 286 | /* make sure readers find IN_TRANS_SETUP before |
7585717f CM |
287 | * they find our root->last_trans update |
288 | */ | |
289 | smp_wmb(); | |
290 | ||
a4abeea4 JB |
291 | spin_lock(&root->fs_info->fs_roots_radix_lock); |
292 | if (root->last_trans == trans->transid) { | |
293 | spin_unlock(&root->fs_info->fs_roots_radix_lock); | |
294 | return 0; | |
295 | } | |
5d4f98a2 YZ |
296 | radix_tree_tag_set(&root->fs_info->fs_roots_radix, |
297 | (unsigned long)root->root_key.objectid, | |
298 | BTRFS_ROOT_TRANS_TAG); | |
a4abeea4 | 299 | spin_unlock(&root->fs_info->fs_roots_radix_lock); |
7585717f CM |
300 | root->last_trans = trans->transid; |
301 | ||
302 | /* this is pretty tricky. We don't want to | |
303 | * take the relocation lock in btrfs_record_root_in_trans | |
304 | * unless we're really doing the first setup for this root in | |
305 | * this transaction. | |
306 | * | |
307 | * Normally we'd use root->last_trans as a flag to decide | |
308 | * if we want to take the expensive mutex. | |
309 | * | |
310 | * But, we have to set root->last_trans before we | |
311 | * init the relocation root, otherwise, we trip over warnings | |
312 | * in ctree.c. The solution used here is to flag ourselves | |
27cdeb70 | 313 | * with root IN_TRANS_SETUP. When this is 1, we're still |
7585717f CM |
314 | * fixing up the reloc trees and everyone must wait. |
315 | * | |
316 | * When this is zero, they can trust root->last_trans and fly | |
317 | * through btrfs_record_root_in_trans without having to take the | |
318 | * lock. smp_wmb() makes sure that all the writes above are | |
319 | * done before we pop in the zero below | |
320 | */ | |
5d4f98a2 | 321 | btrfs_init_reloc_root(trans, root); |
c7548af6 | 322 | smp_mb__before_atomic(); |
27cdeb70 | 323 | clear_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); |
5d4f98a2 YZ |
324 | } |
325 | return 0; | |
326 | } | |
bcc63abb | 327 | |
7585717f CM |
328 | |
329 | int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, | |
330 | struct btrfs_root *root) | |
331 | { | |
27cdeb70 | 332 | if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state)) |
7585717f CM |
333 | return 0; |
334 | ||
335 | /* | |
27cdeb70 | 336 | * see record_root_in_trans for comments about IN_TRANS_SETUP usage |
7585717f CM |
337 | * and barriers |
338 | */ | |
339 | smp_rmb(); | |
340 | if (root->last_trans == trans->transid && | |
27cdeb70 | 341 | !test_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state)) |
7585717f CM |
342 | return 0; |
343 | ||
344 | mutex_lock(&root->fs_info->reloc_mutex); | |
345 | record_root_in_trans(trans, root); | |
346 | mutex_unlock(&root->fs_info->reloc_mutex); | |
347 | ||
348 | return 0; | |
349 | } | |
350 | ||
4a9d8bde MX |
351 | static inline int is_transaction_blocked(struct btrfs_transaction *trans) |
352 | { | |
353 | return (trans->state >= TRANS_STATE_BLOCKED && | |
501407aa JB |
354 | trans->state < TRANS_STATE_UNBLOCKED && |
355 | !trans->aborted); | |
4a9d8bde MX |
356 | } |
357 | ||
d352ac68 CM |
358 | /* wait for commit against the current transaction to become unblocked |
359 | * when this is done, it is safe to start a new transaction, but the current | |
360 | * transaction might not be fully on disk. | |
361 | */ | |
37d1aeee | 362 | static void wait_current_trans(struct btrfs_root *root) |
79154b1b | 363 | { |
f9295749 | 364 | struct btrfs_transaction *cur_trans; |
79154b1b | 365 | |
a4abeea4 | 366 | spin_lock(&root->fs_info->trans_lock); |
f9295749 | 367 | cur_trans = root->fs_info->running_transaction; |
4a9d8bde | 368 | if (cur_trans && is_transaction_blocked(cur_trans)) { |
13c5a93e | 369 | atomic_inc(&cur_trans->use_count); |
a4abeea4 | 370 | spin_unlock(&root->fs_info->trans_lock); |
72d63ed6 LZ |
371 | |
372 | wait_event(root->fs_info->transaction_wait, | |
501407aa JB |
373 | cur_trans->state >= TRANS_STATE_UNBLOCKED || |
374 | cur_trans->aborted); | |
724e2315 | 375 | btrfs_put_transaction(cur_trans); |
a4abeea4 JB |
376 | } else { |
377 | spin_unlock(&root->fs_info->trans_lock); | |
f9295749 | 378 | } |
37d1aeee CM |
379 | } |
380 | ||
a22285a6 YZ |
381 | static int may_wait_transaction(struct btrfs_root *root, int type) |
382 | { | |
a4abeea4 JB |
383 | if (root->fs_info->log_root_recovering) |
384 | return 0; | |
385 | ||
386 | if (type == TRANS_USERSPACE) | |
387 | return 1; | |
388 | ||
389 | if (type == TRANS_START && | |
390 | !atomic_read(&root->fs_info->open_ioctl_trans)) | |
a22285a6 | 391 | return 1; |
a4abeea4 | 392 | |
a22285a6 YZ |
393 | return 0; |
394 | } | |
395 | ||
20dd2cbf MX |
396 | static inline bool need_reserve_reloc_root(struct btrfs_root *root) |
397 | { | |
398 | if (!root->fs_info->reloc_ctl || | |
27cdeb70 | 399 | !test_bit(BTRFS_ROOT_REF_COWS, &root->state) || |
20dd2cbf MX |
400 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || |
401 | root->reloc_root) | |
402 | return false; | |
403 | ||
404 | return true; | |
405 | } | |
406 | ||
08e007d2 | 407 | static struct btrfs_trans_handle * |
0860adfd | 408 | start_transaction(struct btrfs_root *root, u64 num_items, unsigned int type, |
08e007d2 | 409 | enum btrfs_reserve_flush_enum flush) |
37d1aeee | 410 | { |
a22285a6 YZ |
411 | struct btrfs_trans_handle *h; |
412 | struct btrfs_transaction *cur_trans; | |
b5009945 | 413 | u64 num_bytes = 0; |
c5567237 | 414 | u64 qgroup_reserved = 0; |
20dd2cbf MX |
415 | bool reloc_reserved = false; |
416 | int ret; | |
acce952b | 417 | |
46c4e71e | 418 | /* Send isn't supposed to start transactions. */ |
2755a0de | 419 | ASSERT(current->journal_info != BTRFS_SEND_TRANS_STUB); |
46c4e71e | 420 | |
87533c47 | 421 | if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) |
acce952b | 422 | return ERR_PTR(-EROFS); |
2a1eb461 | 423 | |
46c4e71e | 424 | if (current->journal_info) { |
0860adfd | 425 | WARN_ON(type & TRANS_EXTWRITERS); |
2a1eb461 JB |
426 | h = current->journal_info; |
427 | h->use_count++; | |
b7d5b0a8 | 428 | WARN_ON(h->use_count > 2); |
2a1eb461 JB |
429 | h->orig_rsv = h->block_rsv; |
430 | h->block_rsv = NULL; | |
431 | goto got_it; | |
432 | } | |
b5009945 JB |
433 | |
434 | /* | |
435 | * Do the reservation before we join the transaction so we can do all | |
436 | * the appropriate flushing if need be. | |
437 | */ | |
438 | if (num_items > 0 && root != root->fs_info->chunk_root) { | |
c5567237 AJ |
439 | if (root->fs_info->quota_enabled && |
440 | is_fstree(root->root_key.objectid)) { | |
707e8a07 | 441 | qgroup_reserved = num_items * root->nodesize; |
c5567237 AJ |
442 | ret = btrfs_qgroup_reserve(root, qgroup_reserved); |
443 | if (ret) | |
444 | return ERR_PTR(ret); | |
445 | } | |
446 | ||
b5009945 | 447 | num_bytes = btrfs_calc_trans_metadata_size(root, num_items); |
20dd2cbf MX |
448 | /* |
449 | * Do the reservation for the relocation root creation | |
450 | */ | |
ee39b432 | 451 | if (need_reserve_reloc_root(root)) { |
20dd2cbf MX |
452 | num_bytes += root->nodesize; |
453 | reloc_reserved = true; | |
454 | } | |
455 | ||
08e007d2 MX |
456 | ret = btrfs_block_rsv_add(root, |
457 | &root->fs_info->trans_block_rsv, | |
458 | num_bytes, flush); | |
b5009945 | 459 | if (ret) |
843fcf35 | 460 | goto reserve_fail; |
b5009945 | 461 | } |
a22285a6 YZ |
462 | again: |
463 | h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS); | |
843fcf35 MX |
464 | if (!h) { |
465 | ret = -ENOMEM; | |
466 | goto alloc_fail; | |
467 | } | |
37d1aeee | 468 | |
98114659 JB |
469 | /* |
470 | * If we are JOIN_NOLOCK we're already committing a transaction and | |
471 | * waiting on this guy, so we don't need to do the sb_start_intwrite | |
472 | * because we're already holding a ref. We need this because we could | |
473 | * have raced in and did an fsync() on a file which can kick a commit | |
474 | * and then we deadlock with somebody doing a freeze. | |
354aa0fb MX |
475 | * |
476 | * If we are ATTACH, it means we just want to catch the current | |
477 | * transaction and commit it, so we needn't do sb_start_intwrite(). | |
98114659 | 478 | */ |
0860adfd | 479 | if (type & __TRANS_FREEZABLE) |
60376ce4 | 480 | sb_start_intwrite(root->fs_info->sb); |
b2b5ef5c | 481 | |
a22285a6 | 482 | if (may_wait_transaction(root, type)) |
37d1aeee | 483 | wait_current_trans(root); |
a22285a6 | 484 | |
a4abeea4 | 485 | do { |
354aa0fb | 486 | ret = join_transaction(root, type); |
178260b2 | 487 | if (ret == -EBUSY) { |
a4abeea4 | 488 | wait_current_trans(root); |
178260b2 MX |
489 | if (unlikely(type == TRANS_ATTACH)) |
490 | ret = -ENOENT; | |
491 | } | |
a4abeea4 JB |
492 | } while (ret == -EBUSY); |
493 | ||
db5b493a | 494 | if (ret < 0) { |
354aa0fb MX |
495 | /* We must get the transaction if we are JOIN_NOLOCK. */ |
496 | BUG_ON(type == TRANS_JOIN_NOLOCK); | |
843fcf35 | 497 | goto join_fail; |
db5b493a | 498 | } |
0f7d52f4 | 499 | |
a22285a6 | 500 | cur_trans = root->fs_info->running_transaction; |
a22285a6 YZ |
501 | |
502 | h->transid = cur_trans->transid; | |
503 | h->transaction = cur_trans; | |
79154b1b | 504 | h->blocks_used = 0; |
a22285a6 | 505 | h->bytes_reserved = 0; |
d13603ef | 506 | h->root = root; |
56bec294 | 507 | h->delayed_ref_updates = 0; |
2a1eb461 | 508 | h->use_count = 1; |
0e721106 | 509 | h->adding_csums = 0; |
f0486c68 | 510 | h->block_rsv = NULL; |
2a1eb461 | 511 | h->orig_rsv = NULL; |
49b25e05 | 512 | h->aborted = 0; |
4b824906 | 513 | h->qgroup_reserved = 0; |
bed92eae | 514 | h->delayed_ref_elem.seq = 0; |
a698d075 | 515 | h->type = type; |
c6b305a8 | 516 | h->allocating_chunk = false; |
20dd2cbf | 517 | h->reloc_reserved = false; |
5039eddc | 518 | h->sync = false; |
bed92eae | 519 | INIT_LIST_HEAD(&h->qgroup_ref_list); |
ea658bad | 520 | INIT_LIST_HEAD(&h->new_bgs); |
50d9aa99 | 521 | INIT_LIST_HEAD(&h->ordered); |
b7ec40d7 | 522 | |
a22285a6 | 523 | smp_mb(); |
4a9d8bde MX |
524 | if (cur_trans->state >= TRANS_STATE_BLOCKED && |
525 | may_wait_transaction(root, type)) { | |
abdd2e80 | 526 | current->journal_info = h; |
a22285a6 YZ |
527 | btrfs_commit_transaction(h, root); |
528 | goto again; | |
529 | } | |
530 | ||
b5009945 | 531 | if (num_bytes) { |
8c2a3ca2 | 532 | trace_btrfs_space_reservation(root->fs_info, "transaction", |
2bcc0328 | 533 | h->transid, num_bytes, 1); |
b5009945 JB |
534 | h->block_rsv = &root->fs_info->trans_block_rsv; |
535 | h->bytes_reserved = num_bytes; | |
20dd2cbf | 536 | h->reloc_reserved = reloc_reserved; |
a22285a6 | 537 | } |
4b824906 | 538 | h->qgroup_reserved = qgroup_reserved; |
9ed74f2d | 539 | |
2a1eb461 | 540 | got_it: |
a4abeea4 | 541 | btrfs_record_root_in_trans(h, root); |
a22285a6 YZ |
542 | |
543 | if (!current->journal_info && type != TRANS_USERSPACE) | |
544 | current->journal_info = h; | |
79154b1b | 545 | return h; |
843fcf35 MX |
546 | |
547 | join_fail: | |
0860adfd | 548 | if (type & __TRANS_FREEZABLE) |
843fcf35 MX |
549 | sb_end_intwrite(root->fs_info->sb); |
550 | kmem_cache_free(btrfs_trans_handle_cachep, h); | |
551 | alloc_fail: | |
552 | if (num_bytes) | |
553 | btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv, | |
554 | num_bytes); | |
555 | reserve_fail: | |
556 | if (qgroup_reserved) | |
557 | btrfs_qgroup_free(root, qgroup_reserved); | |
558 | return ERR_PTR(ret); | |
79154b1b CM |
559 | } |
560 | ||
f9295749 | 561 | struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, |
a22285a6 | 562 | int num_items) |
f9295749 | 563 | { |
08e007d2 MX |
564 | return start_transaction(root, num_items, TRANS_START, |
565 | BTRFS_RESERVE_FLUSH_ALL); | |
f9295749 | 566 | } |
8407aa46 | 567 | |
08e007d2 | 568 | struct btrfs_trans_handle *btrfs_start_transaction_lflush( |
8407aa46 MX |
569 | struct btrfs_root *root, int num_items) |
570 | { | |
08e007d2 MX |
571 | return start_transaction(root, num_items, TRANS_START, |
572 | BTRFS_RESERVE_FLUSH_LIMIT); | |
8407aa46 MX |
573 | } |
574 | ||
7a7eaa40 | 575 | struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root) |
f9295749 | 576 | { |
8407aa46 | 577 | return start_transaction(root, 0, TRANS_JOIN, 0); |
f9295749 CM |
578 | } |
579 | ||
7a7eaa40 | 580 | struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root) |
0af3d00b | 581 | { |
8407aa46 | 582 | return start_transaction(root, 0, TRANS_JOIN_NOLOCK, 0); |
0af3d00b JB |
583 | } |
584 | ||
7a7eaa40 | 585 | struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root) |
9ca9ee09 | 586 | { |
8407aa46 | 587 | return start_transaction(root, 0, TRANS_USERSPACE, 0); |
9ca9ee09 SW |
588 | } |
589 | ||
d4edf39b MX |
590 | /* |
591 | * btrfs_attach_transaction() - catch the running transaction | |
592 | * | |
593 | * It is used when we want to commit the current the transaction, but | |
594 | * don't want to start a new one. | |
595 | * | |
596 | * Note: If this function return -ENOENT, it just means there is no | |
597 | * running transaction. But it is possible that the inactive transaction | |
598 | * is still in the memory, not fully on disk. If you hope there is no | |
599 | * inactive transaction in the fs when -ENOENT is returned, you should | |
600 | * invoke | |
601 | * btrfs_attach_transaction_barrier() | |
602 | */ | |
354aa0fb | 603 | struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root) |
60376ce4 | 604 | { |
354aa0fb | 605 | return start_transaction(root, 0, TRANS_ATTACH, 0); |
60376ce4 JB |
606 | } |
607 | ||
d4edf39b | 608 | /* |
90b6d283 | 609 | * btrfs_attach_transaction_barrier() - catch the running transaction |
d4edf39b MX |
610 | * |
611 | * It is similar to the above function, the differentia is this one | |
612 | * will wait for all the inactive transactions until they fully | |
613 | * complete. | |
614 | */ | |
615 | struct btrfs_trans_handle * | |
616 | btrfs_attach_transaction_barrier(struct btrfs_root *root) | |
617 | { | |
618 | struct btrfs_trans_handle *trans; | |
619 | ||
620 | trans = start_transaction(root, 0, TRANS_ATTACH, 0); | |
621 | if (IS_ERR(trans) && PTR_ERR(trans) == -ENOENT) | |
622 | btrfs_wait_for_commit(root, 0); | |
623 | ||
624 | return trans; | |
625 | } | |
626 | ||
d352ac68 | 627 | /* wait for a transaction commit to be fully complete */ |
b9c8300c | 628 | static noinline void wait_for_commit(struct btrfs_root *root, |
89ce8a63 CM |
629 | struct btrfs_transaction *commit) |
630 | { | |
4a9d8bde | 631 | wait_event(commit->commit_wait, commit->state == TRANS_STATE_COMPLETED); |
89ce8a63 CM |
632 | } |
633 | ||
46204592 SW |
634 | int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid) |
635 | { | |
636 | struct btrfs_transaction *cur_trans = NULL, *t; | |
8cd2807f | 637 | int ret = 0; |
46204592 | 638 | |
46204592 SW |
639 | if (transid) { |
640 | if (transid <= root->fs_info->last_trans_committed) | |
a4abeea4 | 641 | goto out; |
46204592 SW |
642 | |
643 | /* find specified transaction */ | |
a4abeea4 | 644 | spin_lock(&root->fs_info->trans_lock); |
46204592 SW |
645 | list_for_each_entry(t, &root->fs_info->trans_list, list) { |
646 | if (t->transid == transid) { | |
647 | cur_trans = t; | |
a4abeea4 | 648 | atomic_inc(&cur_trans->use_count); |
8cd2807f | 649 | ret = 0; |
46204592 SW |
650 | break; |
651 | } | |
8cd2807f MX |
652 | if (t->transid > transid) { |
653 | ret = 0; | |
46204592 | 654 | break; |
8cd2807f | 655 | } |
46204592 | 656 | } |
a4abeea4 | 657 | spin_unlock(&root->fs_info->trans_lock); |
42383020 SW |
658 | |
659 | /* | |
660 | * The specified transaction doesn't exist, or we | |
661 | * raced with btrfs_commit_transaction | |
662 | */ | |
663 | if (!cur_trans) { | |
664 | if (transid > root->fs_info->last_trans_committed) | |
665 | ret = -EINVAL; | |
8cd2807f | 666 | goto out; |
42383020 | 667 | } |
46204592 SW |
668 | } else { |
669 | /* find newest transaction that is committing | committed */ | |
a4abeea4 | 670 | spin_lock(&root->fs_info->trans_lock); |
46204592 SW |
671 | list_for_each_entry_reverse(t, &root->fs_info->trans_list, |
672 | list) { | |
4a9d8bde MX |
673 | if (t->state >= TRANS_STATE_COMMIT_START) { |
674 | if (t->state == TRANS_STATE_COMPLETED) | |
3473f3c0 | 675 | break; |
46204592 | 676 | cur_trans = t; |
a4abeea4 | 677 | atomic_inc(&cur_trans->use_count); |
46204592 SW |
678 | break; |
679 | } | |
680 | } | |
a4abeea4 | 681 | spin_unlock(&root->fs_info->trans_lock); |
46204592 | 682 | if (!cur_trans) |
a4abeea4 | 683 | goto out; /* nothing committing|committed */ |
46204592 SW |
684 | } |
685 | ||
46204592 | 686 | wait_for_commit(root, cur_trans); |
724e2315 | 687 | btrfs_put_transaction(cur_trans); |
a4abeea4 | 688 | out: |
46204592 SW |
689 | return ret; |
690 | } | |
691 | ||
37d1aeee CM |
692 | void btrfs_throttle(struct btrfs_root *root) |
693 | { | |
a4abeea4 | 694 | if (!atomic_read(&root->fs_info->open_ioctl_trans)) |
9ca9ee09 | 695 | wait_current_trans(root); |
37d1aeee CM |
696 | } |
697 | ||
8929ecfa YZ |
698 | static int should_end_transaction(struct btrfs_trans_handle *trans, |
699 | struct btrfs_root *root) | |
700 | { | |
1be41b78 | 701 | if (root->fs_info->global_block_rsv.space_info->full && |
0a2b2a84 | 702 | btrfs_check_space_for_delayed_refs(trans, root)) |
1be41b78 | 703 | return 1; |
36ba022a | 704 | |
1be41b78 | 705 | return !!btrfs_block_rsv_check(root, &root->fs_info->global_block_rsv, 5); |
8929ecfa YZ |
706 | } |
707 | ||
708 | int btrfs_should_end_transaction(struct btrfs_trans_handle *trans, | |
709 | struct btrfs_root *root) | |
710 | { | |
711 | struct btrfs_transaction *cur_trans = trans->transaction; | |
712 | int updates; | |
49b25e05 | 713 | int err; |
8929ecfa | 714 | |
a4abeea4 | 715 | smp_mb(); |
4a9d8bde MX |
716 | if (cur_trans->state >= TRANS_STATE_BLOCKED || |
717 | cur_trans->delayed_refs.flushing) | |
8929ecfa YZ |
718 | return 1; |
719 | ||
720 | updates = trans->delayed_ref_updates; | |
721 | trans->delayed_ref_updates = 0; | |
49b25e05 JM |
722 | if (updates) { |
723 | err = btrfs_run_delayed_refs(trans, root, updates); | |
724 | if (err) /* Error code will also eval true */ | |
725 | return err; | |
726 | } | |
8929ecfa YZ |
727 | |
728 | return should_end_transaction(trans, root); | |
729 | } | |
730 | ||
89ce8a63 | 731 | static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, |
a698d075 | 732 | struct btrfs_root *root, int throttle) |
79154b1b | 733 | { |
8929ecfa | 734 | struct btrfs_transaction *cur_trans = trans->transaction; |
ab78c84d | 735 | struct btrfs_fs_info *info = root->fs_info; |
1be41b78 | 736 | unsigned long cur = trans->delayed_ref_updates; |
a698d075 | 737 | int lock = (trans->type != TRANS_JOIN_NOLOCK); |
4edc2ca3 | 738 | int err = 0; |
a79b7d4b | 739 | int must_run_delayed_refs = 0; |
c3e69d58 | 740 | |
3bbb24b2 JB |
741 | if (trans->use_count > 1) { |
742 | trans->use_count--; | |
2a1eb461 JB |
743 | trans->block_rsv = trans->orig_rsv; |
744 | return 0; | |
745 | } | |
746 | ||
b24e03db | 747 | btrfs_trans_release_metadata(trans, root); |
4c13d758 | 748 | trans->block_rsv = NULL; |
c5567237 | 749 | |
ea658bad JB |
750 | if (!list_empty(&trans->new_bgs)) |
751 | btrfs_create_pending_block_groups(trans, root); | |
752 | ||
50d9aa99 JB |
753 | if (!list_empty(&trans->ordered)) { |
754 | spin_lock(&info->trans_lock); | |
755 | list_splice(&trans->ordered, &cur_trans->pending_ordered); | |
756 | spin_unlock(&info->trans_lock); | |
757 | } | |
758 | ||
1be41b78 | 759 | trans->delayed_ref_updates = 0; |
a79b7d4b CM |
760 | if (!trans->sync) { |
761 | must_run_delayed_refs = | |
762 | btrfs_should_throttle_delayed_refs(trans, root); | |
0a2b2a84 | 763 | cur = max_t(unsigned long, cur, 32); |
a79b7d4b CM |
764 | |
765 | /* | |
766 | * don't make the caller wait if they are from a NOLOCK | |
767 | * or ATTACH transaction, it will deadlock with commit | |
768 | */ | |
769 | if (must_run_delayed_refs == 1 && | |
770 | (trans->type & (__TRANS_JOIN_NOLOCK | __TRANS_ATTACH))) | |
771 | must_run_delayed_refs = 2; | |
56bec294 | 772 | } |
bb721703 | 773 | |
fcebe456 JB |
774 | if (trans->qgroup_reserved) { |
775 | /* | |
776 | * the same root has to be passed here between start_transaction | |
777 | * and end_transaction. Subvolume quota depends on this. | |
778 | */ | |
779 | btrfs_qgroup_free(trans->root, trans->qgroup_reserved); | |
780 | trans->qgroup_reserved = 0; | |
781 | } | |
782 | ||
0e721106 JB |
783 | btrfs_trans_release_metadata(trans, root); |
784 | trans->block_rsv = NULL; | |
56bec294 | 785 | |
ea658bad JB |
786 | if (!list_empty(&trans->new_bgs)) |
787 | btrfs_create_pending_block_groups(trans, root); | |
788 | ||
a4abeea4 | 789 | if (lock && !atomic_read(&root->fs_info->open_ioctl_trans) && |
4a9d8bde MX |
790 | should_end_transaction(trans, root) && |
791 | ACCESS_ONCE(cur_trans->state) == TRANS_STATE_RUNNING) { | |
792 | spin_lock(&info->trans_lock); | |
793 | if (cur_trans->state == TRANS_STATE_RUNNING) | |
794 | cur_trans->state = TRANS_STATE_BLOCKED; | |
795 | spin_unlock(&info->trans_lock); | |
a4abeea4 | 796 | } |
8929ecfa | 797 | |
4a9d8bde | 798 | if (lock && ACCESS_ONCE(cur_trans->state) == TRANS_STATE_BLOCKED) { |
3bbb24b2 | 799 | if (throttle) |
8929ecfa | 800 | return btrfs_commit_transaction(trans, root); |
3bbb24b2 | 801 | else |
8929ecfa YZ |
802 | wake_up_process(info->transaction_kthread); |
803 | } | |
804 | ||
0860adfd | 805 | if (trans->type & __TRANS_FREEZABLE) |
98114659 | 806 | sb_end_intwrite(root->fs_info->sb); |
6df7881a | 807 | |
8929ecfa | 808 | WARN_ON(cur_trans != info->running_transaction); |
13c5a93e JB |
809 | WARN_ON(atomic_read(&cur_trans->num_writers) < 1); |
810 | atomic_dec(&cur_trans->num_writers); | |
0860adfd | 811 | extwriter_counter_dec(cur_trans, trans->type); |
89ce8a63 | 812 | |
99d16cbc | 813 | smp_mb(); |
79154b1b CM |
814 | if (waitqueue_active(&cur_trans->writer_wait)) |
815 | wake_up(&cur_trans->writer_wait); | |
724e2315 | 816 | btrfs_put_transaction(cur_trans); |
9ed74f2d JB |
817 | |
818 | if (current->journal_info == trans) | |
819 | current->journal_info = NULL; | |
ab78c84d | 820 | |
24bbcf04 YZ |
821 | if (throttle) |
822 | btrfs_run_delayed_iputs(root); | |
823 | ||
49b25e05 | 824 | if (trans->aborted || |
4e121c06 JB |
825 | test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) { |
826 | wake_up_process(info->transaction_kthread); | |
4edc2ca3 | 827 | err = -EIO; |
4e121c06 | 828 | } |
edf39272 | 829 | assert_qgroups_uptodate(trans); |
49b25e05 | 830 | |
4edc2ca3 | 831 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
a79b7d4b CM |
832 | if (must_run_delayed_refs) { |
833 | btrfs_async_run_delayed_refs(root, cur, | |
834 | must_run_delayed_refs == 1); | |
835 | } | |
4edc2ca3 | 836 | return err; |
79154b1b CM |
837 | } |
838 | ||
89ce8a63 CM |
839 | int btrfs_end_transaction(struct btrfs_trans_handle *trans, |
840 | struct btrfs_root *root) | |
841 | { | |
98ad43be | 842 | return __btrfs_end_transaction(trans, root, 0); |
89ce8a63 CM |
843 | } |
844 | ||
845 | int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans, | |
846 | struct btrfs_root *root) | |
847 | { | |
98ad43be | 848 | return __btrfs_end_transaction(trans, root, 1); |
16cdcec7 MX |
849 | } |
850 | ||
d352ac68 CM |
851 | /* |
852 | * when btree blocks are allocated, they have some corresponding bits set for | |
853 | * them in one of two extent_io trees. This is used to make sure all of | |
690587d1 | 854 | * those extents are sent to disk but does not wait on them |
d352ac68 | 855 | */ |
690587d1 | 856 | int btrfs_write_marked_extents(struct btrfs_root *root, |
8cef4e16 | 857 | struct extent_io_tree *dirty_pages, int mark) |
79154b1b | 858 | { |
777e6bd7 | 859 | int err = 0; |
7c4452b9 | 860 | int werr = 0; |
1728366e | 861 | struct address_space *mapping = root->fs_info->btree_inode->i_mapping; |
e6138876 | 862 | struct extent_state *cached_state = NULL; |
777e6bd7 | 863 | u64 start = 0; |
5f39d397 | 864 | u64 end; |
7c4452b9 | 865 | |
1728366e | 866 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
e6138876 | 867 | mark, &cached_state)) { |
663dfbb0 FM |
868 | bool wait_writeback = false; |
869 | ||
870 | err = convert_extent_bit(dirty_pages, start, end, | |
871 | EXTENT_NEED_WAIT, | |
872 | mark, &cached_state, GFP_NOFS); | |
873 | /* | |
874 | * convert_extent_bit can return -ENOMEM, which is most of the | |
875 | * time a temporary error. So when it happens, ignore the error | |
876 | * and wait for writeback of this range to finish - because we | |
877 | * failed to set the bit EXTENT_NEED_WAIT for the range, a call | |
878 | * to btrfs_wait_marked_extents() would not know that writeback | |
879 | * for this range started and therefore wouldn't wait for it to | |
880 | * finish - we don't want to commit a superblock that points to | |
881 | * btree nodes/leafs for which writeback hasn't finished yet | |
882 | * (and without errors). | |
883 | * We cleanup any entries left in the io tree when committing | |
884 | * the transaction (through clear_btree_io_tree()). | |
885 | */ | |
886 | if (err == -ENOMEM) { | |
887 | err = 0; | |
888 | wait_writeback = true; | |
889 | } | |
890 | if (!err) | |
891 | err = filemap_fdatawrite_range(mapping, start, end); | |
1728366e JB |
892 | if (err) |
893 | werr = err; | |
663dfbb0 FM |
894 | else if (wait_writeback) |
895 | werr = filemap_fdatawait_range(mapping, start, end); | |
e38e2ed7 | 896 | free_extent_state(cached_state); |
663dfbb0 | 897 | cached_state = NULL; |
1728366e JB |
898 | cond_resched(); |
899 | start = end + 1; | |
7c4452b9 | 900 | } |
690587d1 CM |
901 | return werr; |
902 | } | |
903 | ||
904 | /* | |
905 | * when btree blocks are allocated, they have some corresponding bits set for | |
906 | * them in one of two extent_io trees. This is used to make sure all of | |
907 | * those extents are on disk for transaction or log commit. We wait | |
908 | * on all the pages and clear them from the dirty pages state tree | |
909 | */ | |
910 | int btrfs_wait_marked_extents(struct btrfs_root *root, | |
8cef4e16 | 911 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 | 912 | { |
690587d1 CM |
913 | int err = 0; |
914 | int werr = 0; | |
1728366e | 915 | struct address_space *mapping = root->fs_info->btree_inode->i_mapping; |
e6138876 | 916 | struct extent_state *cached_state = NULL; |
690587d1 CM |
917 | u64 start = 0; |
918 | u64 end; | |
656f30db FM |
919 | struct btrfs_inode *btree_ino = BTRFS_I(root->fs_info->btree_inode); |
920 | bool errors = false; | |
777e6bd7 | 921 | |
1728366e | 922 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
e6138876 | 923 | EXTENT_NEED_WAIT, &cached_state)) { |
663dfbb0 FM |
924 | /* |
925 | * Ignore -ENOMEM errors returned by clear_extent_bit(). | |
926 | * When committing the transaction, we'll remove any entries | |
927 | * left in the io tree. For a log commit, we don't remove them | |
928 | * after committing the log because the tree can be accessed | |
929 | * concurrently - we do it only at transaction commit time when | |
930 | * it's safe to do it (through clear_btree_io_tree()). | |
931 | */ | |
932 | err = clear_extent_bit(dirty_pages, start, end, | |
933 | EXTENT_NEED_WAIT, | |
934 | 0, 0, &cached_state, GFP_NOFS); | |
935 | if (err == -ENOMEM) | |
936 | err = 0; | |
937 | if (!err) | |
938 | err = filemap_fdatawait_range(mapping, start, end); | |
1728366e JB |
939 | if (err) |
940 | werr = err; | |
e38e2ed7 FM |
941 | free_extent_state(cached_state); |
942 | cached_state = NULL; | |
1728366e JB |
943 | cond_resched(); |
944 | start = end + 1; | |
777e6bd7 | 945 | } |
7c4452b9 CM |
946 | if (err) |
947 | werr = err; | |
656f30db FM |
948 | |
949 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { | |
950 | if ((mark & EXTENT_DIRTY) && | |
951 | test_and_clear_bit(BTRFS_INODE_BTREE_LOG1_ERR, | |
952 | &btree_ino->runtime_flags)) | |
953 | errors = true; | |
954 | ||
955 | if ((mark & EXTENT_NEW) && | |
956 | test_and_clear_bit(BTRFS_INODE_BTREE_LOG2_ERR, | |
957 | &btree_ino->runtime_flags)) | |
958 | errors = true; | |
959 | } else { | |
960 | if (test_and_clear_bit(BTRFS_INODE_BTREE_ERR, | |
961 | &btree_ino->runtime_flags)) | |
962 | errors = true; | |
963 | } | |
964 | ||
965 | if (errors && !werr) | |
966 | werr = -EIO; | |
967 | ||
7c4452b9 | 968 | return werr; |
79154b1b CM |
969 | } |
970 | ||
690587d1 CM |
971 | /* |
972 | * when btree blocks are allocated, they have some corresponding bits set for | |
973 | * them in one of two extent_io trees. This is used to make sure all of | |
974 | * those extents are on disk for transaction or log commit | |
975 | */ | |
171170c1 | 976 | static int btrfs_write_and_wait_marked_extents(struct btrfs_root *root, |
8cef4e16 | 977 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 CM |
978 | { |
979 | int ret; | |
980 | int ret2; | |
c6adc9cc | 981 | struct blk_plug plug; |
690587d1 | 982 | |
c6adc9cc | 983 | blk_start_plug(&plug); |
8cef4e16 | 984 | ret = btrfs_write_marked_extents(root, dirty_pages, mark); |
c6adc9cc | 985 | blk_finish_plug(&plug); |
8cef4e16 | 986 | ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark); |
bf0da8c1 CM |
987 | |
988 | if (ret) | |
989 | return ret; | |
990 | if (ret2) | |
991 | return ret2; | |
992 | return 0; | |
690587d1 CM |
993 | } |
994 | ||
663dfbb0 | 995 | static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans, |
d0c803c4 CM |
996 | struct btrfs_root *root) |
997 | { | |
663dfbb0 FM |
998 | int ret; |
999 | ||
1000 | ret = btrfs_write_and_wait_marked_extents(root, | |
8cef4e16 YZ |
1001 | &trans->transaction->dirty_pages, |
1002 | EXTENT_DIRTY); | |
663dfbb0 FM |
1003 | clear_btree_io_tree(&trans->transaction->dirty_pages); |
1004 | ||
1005 | return ret; | |
d0c803c4 CM |
1006 | } |
1007 | ||
d352ac68 CM |
1008 | /* |
1009 | * this is used to update the root pointer in the tree of tree roots. | |
1010 | * | |
1011 | * But, in the case of the extent allocation tree, updating the root | |
1012 | * pointer may allocate blocks which may change the root of the extent | |
1013 | * allocation tree. | |
1014 | * | |
1015 | * So, this loops and repeats and makes sure the cowonly root didn't | |
1016 | * change while the root pointer was being updated in the metadata. | |
1017 | */ | |
0b86a832 CM |
1018 | static int update_cowonly_root(struct btrfs_trans_handle *trans, |
1019 | struct btrfs_root *root) | |
79154b1b CM |
1020 | { |
1021 | int ret; | |
0b86a832 | 1022 | u64 old_root_bytenr; |
86b9f2ec | 1023 | u64 old_root_used; |
0b86a832 | 1024 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
e7070be1 | 1025 | bool extent_root = (root->objectid == BTRFS_EXTENT_TREE_OBJECTID); |
79154b1b | 1026 | |
86b9f2ec | 1027 | old_root_used = btrfs_root_used(&root->root_item); |
0b86a832 | 1028 | btrfs_write_dirty_block_groups(trans, root); |
56bec294 | 1029 | |
d397712b | 1030 | while (1) { |
0b86a832 | 1031 | old_root_bytenr = btrfs_root_bytenr(&root->root_item); |
86b9f2ec | 1032 | if (old_root_bytenr == root->node->start && |
ce93ec54 JB |
1033 | old_root_used == btrfs_root_used(&root->root_item) && |
1034 | (!extent_root || | |
1035 | list_empty(&trans->transaction->dirty_bgs))) | |
79154b1b | 1036 | break; |
87ef2bb4 | 1037 | |
5d4f98a2 | 1038 | btrfs_set_root_node(&root->root_item, root->node); |
79154b1b | 1039 | ret = btrfs_update_root(trans, tree_root, |
0b86a832 CM |
1040 | &root->root_key, |
1041 | &root->root_item); | |
49b25e05 JM |
1042 | if (ret) |
1043 | return ret; | |
56bec294 | 1044 | |
86b9f2ec | 1045 | old_root_used = btrfs_root_used(&root->root_item); |
e7070be1 JB |
1046 | if (extent_root) { |
1047 | ret = btrfs_write_dirty_block_groups(trans, root); | |
1048 | if (ret) | |
1049 | return ret; | |
1050 | } | |
1051 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); | |
49b25e05 JM |
1052 | if (ret) |
1053 | return ret; | |
ce93ec54 JB |
1054 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
1055 | if (ret) | |
1056 | return ret; | |
0b86a832 | 1057 | } |
276e680d | 1058 | |
0b86a832 CM |
1059 | return 0; |
1060 | } | |
1061 | ||
d352ac68 CM |
1062 | /* |
1063 | * update all the cowonly tree roots on disk | |
49b25e05 JM |
1064 | * |
1065 | * The error handling in this function may not be obvious. Any of the | |
1066 | * failures will cause the file system to go offline. We still need | |
1067 | * to clean up the delayed refs. | |
d352ac68 | 1068 | */ |
5d4f98a2 YZ |
1069 | static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans, |
1070 | struct btrfs_root *root) | |
0b86a832 CM |
1071 | { |
1072 | struct btrfs_fs_info *fs_info = root->fs_info; | |
1073 | struct list_head *next; | |
84234f3a | 1074 | struct extent_buffer *eb; |
56bec294 | 1075 | int ret; |
84234f3a YZ |
1076 | |
1077 | eb = btrfs_lock_root_node(fs_info->tree_root); | |
49b25e05 JM |
1078 | ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, |
1079 | 0, &eb); | |
84234f3a YZ |
1080 | btrfs_tree_unlock(eb); |
1081 | free_extent_buffer(eb); | |
0b86a832 | 1082 | |
49b25e05 JM |
1083 | if (ret) |
1084 | return ret; | |
1085 | ||
56bec294 | 1086 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
49b25e05 JM |
1087 | if (ret) |
1088 | return ret; | |
87ef2bb4 | 1089 | |
733f4fbb | 1090 | ret = btrfs_run_dev_stats(trans, root->fs_info); |
c16ce190 JB |
1091 | if (ret) |
1092 | return ret; | |
8dabb742 | 1093 | ret = btrfs_run_dev_replace(trans, root->fs_info); |
c16ce190 JB |
1094 | if (ret) |
1095 | return ret; | |
546adb0d | 1096 | ret = btrfs_run_qgroups(trans, root->fs_info); |
c16ce190 JB |
1097 | if (ret) |
1098 | return ret; | |
546adb0d JS |
1099 | |
1100 | /* run_qgroups might have added some more refs */ | |
1101 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); | |
c16ce190 JB |
1102 | if (ret) |
1103 | return ret; | |
546adb0d | 1104 | |
d397712b | 1105 | while (!list_empty(&fs_info->dirty_cowonly_roots)) { |
0b86a832 CM |
1106 | next = fs_info->dirty_cowonly_roots.next; |
1107 | list_del_init(next); | |
1108 | root = list_entry(next, struct btrfs_root, dirty_list); | |
e7070be1 | 1109 | clear_bit(BTRFS_ROOT_DIRTY, &root->state); |
87ef2bb4 | 1110 | |
9e351cc8 JB |
1111 | if (root != fs_info->extent_root) |
1112 | list_add_tail(&root->dirty_list, | |
1113 | &trans->transaction->switch_commits); | |
49b25e05 JM |
1114 | ret = update_cowonly_root(trans, root); |
1115 | if (ret) | |
1116 | return ret; | |
79154b1b | 1117 | } |
276e680d | 1118 | |
9e351cc8 JB |
1119 | list_add_tail(&fs_info->extent_root->dirty_list, |
1120 | &trans->transaction->switch_commits); | |
8dabb742 SB |
1121 | btrfs_after_dev_replace_commit(fs_info); |
1122 | ||
79154b1b CM |
1123 | return 0; |
1124 | } | |
1125 | ||
d352ac68 CM |
1126 | /* |
1127 | * dead roots are old snapshots that need to be deleted. This allocates | |
1128 | * a dirty root struct and adds it into the list of dead roots that need to | |
1129 | * be deleted | |
1130 | */ | |
cfad392b | 1131 | void btrfs_add_dead_root(struct btrfs_root *root) |
5eda7b5e | 1132 | { |
a4abeea4 | 1133 | spin_lock(&root->fs_info->trans_lock); |
cfad392b JB |
1134 | if (list_empty(&root->root_list)) |
1135 | list_add_tail(&root->root_list, &root->fs_info->dead_roots); | |
a4abeea4 | 1136 | spin_unlock(&root->fs_info->trans_lock); |
5eda7b5e CM |
1137 | } |
1138 | ||
d352ac68 | 1139 | /* |
5d4f98a2 | 1140 | * update all the cowonly tree roots on disk |
d352ac68 | 1141 | */ |
5d4f98a2 YZ |
1142 | static noinline int commit_fs_roots(struct btrfs_trans_handle *trans, |
1143 | struct btrfs_root *root) | |
0f7d52f4 | 1144 | { |
0f7d52f4 | 1145 | struct btrfs_root *gang[8]; |
5d4f98a2 | 1146 | struct btrfs_fs_info *fs_info = root->fs_info; |
0f7d52f4 CM |
1147 | int i; |
1148 | int ret; | |
54aa1f4d CM |
1149 | int err = 0; |
1150 | ||
a4abeea4 | 1151 | spin_lock(&fs_info->fs_roots_radix_lock); |
d397712b | 1152 | while (1) { |
5d4f98a2 YZ |
1153 | ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, |
1154 | (void **)gang, 0, | |
0f7d52f4 CM |
1155 | ARRAY_SIZE(gang), |
1156 | BTRFS_ROOT_TRANS_TAG); | |
1157 | if (ret == 0) | |
1158 | break; | |
1159 | for (i = 0; i < ret; i++) { | |
1160 | root = gang[i]; | |
5d4f98a2 YZ |
1161 | radix_tree_tag_clear(&fs_info->fs_roots_radix, |
1162 | (unsigned long)root->root_key.objectid, | |
1163 | BTRFS_ROOT_TRANS_TAG); | |
a4abeea4 | 1164 | spin_unlock(&fs_info->fs_roots_radix_lock); |
31153d81 | 1165 | |
e02119d5 | 1166 | btrfs_free_log(trans, root); |
5d4f98a2 | 1167 | btrfs_update_reloc_root(trans, root); |
d68fc57b | 1168 | btrfs_orphan_commit_root(trans, root); |
bcc63abb | 1169 | |
82d5902d LZ |
1170 | btrfs_save_ino_cache(root, trans); |
1171 | ||
f1ebcc74 | 1172 | /* see comments in should_cow_block() */ |
27cdeb70 | 1173 | clear_bit(BTRFS_ROOT_FORCE_COW, &root->state); |
c7548af6 | 1174 | smp_mb__after_atomic(); |
f1ebcc74 | 1175 | |
978d910d | 1176 | if (root->commit_root != root->node) { |
9e351cc8 JB |
1177 | list_add_tail(&root->dirty_list, |
1178 | &trans->transaction->switch_commits); | |
978d910d YZ |
1179 | btrfs_set_root_node(&root->root_item, |
1180 | root->node); | |
1181 | } | |
5d4f98a2 | 1182 | |
5d4f98a2 | 1183 | err = btrfs_update_root(trans, fs_info->tree_root, |
0f7d52f4 CM |
1184 | &root->root_key, |
1185 | &root->root_item); | |
a4abeea4 | 1186 | spin_lock(&fs_info->fs_roots_radix_lock); |
54aa1f4d CM |
1187 | if (err) |
1188 | break; | |
0f7d52f4 CM |
1189 | } |
1190 | } | |
a4abeea4 | 1191 | spin_unlock(&fs_info->fs_roots_radix_lock); |
54aa1f4d | 1192 | return err; |
0f7d52f4 CM |
1193 | } |
1194 | ||
d352ac68 | 1195 | /* |
de78b51a ES |
1196 | * defrag a given btree. |
1197 | * Every leaf in the btree is read and defragged. | |
d352ac68 | 1198 | */ |
de78b51a | 1199 | int btrfs_defrag_root(struct btrfs_root *root) |
e9d0b13b CM |
1200 | { |
1201 | struct btrfs_fs_info *info = root->fs_info; | |
e9d0b13b | 1202 | struct btrfs_trans_handle *trans; |
8929ecfa | 1203 | int ret; |
e9d0b13b | 1204 | |
27cdeb70 | 1205 | if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state)) |
e9d0b13b | 1206 | return 0; |
8929ecfa | 1207 | |
6b80053d | 1208 | while (1) { |
8929ecfa YZ |
1209 | trans = btrfs_start_transaction(root, 0); |
1210 | if (IS_ERR(trans)) | |
1211 | return PTR_ERR(trans); | |
1212 | ||
de78b51a | 1213 | ret = btrfs_defrag_leaves(trans, root); |
8929ecfa | 1214 | |
e9d0b13b | 1215 | btrfs_end_transaction(trans, root); |
b53d3f5d | 1216 | btrfs_btree_balance_dirty(info->tree_root); |
e9d0b13b CM |
1217 | cond_resched(); |
1218 | ||
7841cb28 | 1219 | if (btrfs_fs_closing(root->fs_info) || ret != -EAGAIN) |
e9d0b13b | 1220 | break; |
210549eb DS |
1221 | |
1222 | if (btrfs_defrag_cancelled(root->fs_info)) { | |
efe120a0 | 1223 | pr_debug("BTRFS: defrag_root cancelled\n"); |
210549eb DS |
1224 | ret = -EAGAIN; |
1225 | break; | |
1226 | } | |
e9d0b13b | 1227 | } |
27cdeb70 | 1228 | clear_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state); |
8929ecfa | 1229 | return ret; |
e9d0b13b CM |
1230 | } |
1231 | ||
d352ac68 CM |
1232 | /* |
1233 | * new snapshots need to be created at a very specific time in the | |
aec8030a MX |
1234 | * transaction commit. This does the actual creation. |
1235 | * | |
1236 | * Note: | |
1237 | * If the error which may affect the commitment of the current transaction | |
1238 | * happens, we should return the error number. If the error which just affect | |
1239 | * the creation of the pending snapshots, just return 0. | |
d352ac68 | 1240 | */ |
80b6794d | 1241 | static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, |
3063d29f CM |
1242 | struct btrfs_fs_info *fs_info, |
1243 | struct btrfs_pending_snapshot *pending) | |
1244 | { | |
1245 | struct btrfs_key key; | |
80b6794d | 1246 | struct btrfs_root_item *new_root_item; |
3063d29f CM |
1247 | struct btrfs_root *tree_root = fs_info->tree_root; |
1248 | struct btrfs_root *root = pending->root; | |
6bdb72de | 1249 | struct btrfs_root *parent_root; |
98c9942a | 1250 | struct btrfs_block_rsv *rsv; |
6bdb72de | 1251 | struct inode *parent_inode; |
42874b3d MX |
1252 | struct btrfs_path *path; |
1253 | struct btrfs_dir_item *dir_item; | |
a22285a6 | 1254 | struct dentry *dentry; |
3063d29f | 1255 | struct extent_buffer *tmp; |
925baedd | 1256 | struct extent_buffer *old; |
8ea05e3a | 1257 | struct timespec cur_time = CURRENT_TIME; |
aec8030a | 1258 | int ret = 0; |
d68fc57b | 1259 | u64 to_reserve = 0; |
6bdb72de | 1260 | u64 index = 0; |
a22285a6 | 1261 | u64 objectid; |
b83cc969 | 1262 | u64 root_flags; |
8ea05e3a | 1263 | uuid_le new_uuid; |
3063d29f | 1264 | |
42874b3d MX |
1265 | path = btrfs_alloc_path(); |
1266 | if (!path) { | |
aec8030a MX |
1267 | pending->error = -ENOMEM; |
1268 | return 0; | |
42874b3d MX |
1269 | } |
1270 | ||
80b6794d CM |
1271 | new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS); |
1272 | if (!new_root_item) { | |
aec8030a | 1273 | pending->error = -ENOMEM; |
6fa9700e | 1274 | goto root_item_alloc_fail; |
80b6794d | 1275 | } |
a22285a6 | 1276 | |
aec8030a MX |
1277 | pending->error = btrfs_find_free_objectid(tree_root, &objectid); |
1278 | if (pending->error) | |
6fa9700e | 1279 | goto no_free_objectid; |
3063d29f | 1280 | |
3fd0a558 | 1281 | btrfs_reloc_pre_snapshot(trans, pending, &to_reserve); |
d68fc57b YZ |
1282 | |
1283 | if (to_reserve > 0) { | |
aec8030a MX |
1284 | pending->error = btrfs_block_rsv_add(root, |
1285 | &pending->block_rsv, | |
1286 | to_reserve, | |
1287 | BTRFS_RESERVE_NO_FLUSH); | |
1288 | if (pending->error) | |
6fa9700e | 1289 | goto no_free_objectid; |
d68fc57b YZ |
1290 | } |
1291 | ||
3063d29f | 1292 | key.objectid = objectid; |
a22285a6 YZ |
1293 | key.offset = (u64)-1; |
1294 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3063d29f | 1295 | |
6fa9700e | 1296 | rsv = trans->block_rsv; |
a22285a6 | 1297 | trans->block_rsv = &pending->block_rsv; |
2382c5cc | 1298 | trans->bytes_reserved = trans->block_rsv->reserved; |
3de4586c | 1299 | |
a22285a6 | 1300 | dentry = pending->dentry; |
e9662f70 | 1301 | parent_inode = pending->dir; |
a22285a6 | 1302 | parent_root = BTRFS_I(parent_inode)->root; |
7585717f | 1303 | record_root_in_trans(trans, parent_root); |
a22285a6 | 1304 | |
3063d29f CM |
1305 | /* |
1306 | * insert the directory item | |
1307 | */ | |
3de4586c | 1308 | ret = btrfs_set_inode_index(parent_inode, &index); |
49b25e05 | 1309 | BUG_ON(ret); /* -ENOMEM */ |
42874b3d MX |
1310 | |
1311 | /* check if there is a file/dir which has the same name. */ | |
1312 | dir_item = btrfs_lookup_dir_item(NULL, parent_root, path, | |
1313 | btrfs_ino(parent_inode), | |
1314 | dentry->d_name.name, | |
1315 | dentry->d_name.len, 0); | |
1316 | if (dir_item != NULL && !IS_ERR(dir_item)) { | |
fe66a05a | 1317 | pending->error = -EEXIST; |
aec8030a | 1318 | goto dir_item_existed; |
42874b3d MX |
1319 | } else if (IS_ERR(dir_item)) { |
1320 | ret = PTR_ERR(dir_item); | |
8732d44f MX |
1321 | btrfs_abort_transaction(trans, root, ret); |
1322 | goto fail; | |
79787eaa | 1323 | } |
42874b3d | 1324 | btrfs_release_path(path); |
52c26179 | 1325 | |
e999376f CM |
1326 | /* |
1327 | * pull in the delayed directory update | |
1328 | * and the delayed inode item | |
1329 | * otherwise we corrupt the FS during | |
1330 | * snapshot | |
1331 | */ | |
1332 | ret = btrfs_run_delayed_items(trans, root); | |
8732d44f MX |
1333 | if (ret) { /* Transaction aborted */ |
1334 | btrfs_abort_transaction(trans, root, ret); | |
1335 | goto fail; | |
1336 | } | |
e999376f | 1337 | |
7585717f | 1338 | record_root_in_trans(trans, root); |
6bdb72de SW |
1339 | btrfs_set_root_last_snapshot(&root->root_item, trans->transid); |
1340 | memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); | |
08fe4db1 | 1341 | btrfs_check_and_init_root_item(new_root_item); |
6bdb72de | 1342 | |
b83cc969 LZ |
1343 | root_flags = btrfs_root_flags(new_root_item); |
1344 | if (pending->readonly) | |
1345 | root_flags |= BTRFS_ROOT_SUBVOL_RDONLY; | |
1346 | else | |
1347 | root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY; | |
1348 | btrfs_set_root_flags(new_root_item, root_flags); | |
1349 | ||
8ea05e3a AB |
1350 | btrfs_set_root_generation_v2(new_root_item, |
1351 | trans->transid); | |
1352 | uuid_le_gen(&new_uuid); | |
1353 | memcpy(new_root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE); | |
1354 | memcpy(new_root_item->parent_uuid, root->root_item.uuid, | |
1355 | BTRFS_UUID_SIZE); | |
70023da2 SB |
1356 | if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) { |
1357 | memset(new_root_item->received_uuid, 0, | |
1358 | sizeof(new_root_item->received_uuid)); | |
1359 | memset(&new_root_item->stime, 0, sizeof(new_root_item->stime)); | |
1360 | memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime)); | |
1361 | btrfs_set_root_stransid(new_root_item, 0); | |
1362 | btrfs_set_root_rtransid(new_root_item, 0); | |
1363 | } | |
3cae210f QW |
1364 | btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec); |
1365 | btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec); | |
8ea05e3a | 1366 | btrfs_set_root_otransid(new_root_item, trans->transid); |
8ea05e3a | 1367 | |
6bdb72de | 1368 | old = btrfs_lock_root_node(root); |
49b25e05 | 1369 | ret = btrfs_cow_block(trans, root, old, NULL, 0, &old); |
79787eaa JM |
1370 | if (ret) { |
1371 | btrfs_tree_unlock(old); | |
1372 | free_extent_buffer(old); | |
8732d44f MX |
1373 | btrfs_abort_transaction(trans, root, ret); |
1374 | goto fail; | |
79787eaa | 1375 | } |
49b25e05 | 1376 | |
6bdb72de SW |
1377 | btrfs_set_lock_blocking(old); |
1378 | ||
49b25e05 | 1379 | ret = btrfs_copy_root(trans, root, old, &tmp, objectid); |
79787eaa | 1380 | /* clean up in any case */ |
6bdb72de SW |
1381 | btrfs_tree_unlock(old); |
1382 | free_extent_buffer(old); | |
8732d44f MX |
1383 | if (ret) { |
1384 | btrfs_abort_transaction(trans, root, ret); | |
1385 | goto fail; | |
1386 | } | |
6bdb72de | 1387 | |
fcebe456 JB |
1388 | /* |
1389 | * We need to flush delayed refs in order to make sure all of our quota | |
1390 | * operations have been done before we call btrfs_qgroup_inherit. | |
1391 | */ | |
1392 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); | |
1393 | if (ret) { | |
1394 | btrfs_abort_transaction(trans, root, ret); | |
1395 | goto fail; | |
1396 | } | |
1397 | ||
47a306a7 ES |
1398 | ret = btrfs_qgroup_inherit(trans, fs_info, |
1399 | root->root_key.objectid, | |
1400 | objectid, pending->inherit); | |
1401 | if (ret) { | |
1402 | btrfs_abort_transaction(trans, root, ret); | |
1403 | goto fail; | |
1404 | } | |
fcebe456 | 1405 | |
f1ebcc74 | 1406 | /* see comments in should_cow_block() */ |
27cdeb70 | 1407 | set_bit(BTRFS_ROOT_FORCE_COW, &root->state); |
f1ebcc74 LB |
1408 | smp_wmb(); |
1409 | ||
6bdb72de | 1410 | btrfs_set_root_node(new_root_item, tmp); |
a22285a6 YZ |
1411 | /* record when the snapshot was created in key.offset */ |
1412 | key.offset = trans->transid; | |
1413 | ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); | |
6bdb72de SW |
1414 | btrfs_tree_unlock(tmp); |
1415 | free_extent_buffer(tmp); | |
8732d44f MX |
1416 | if (ret) { |
1417 | btrfs_abort_transaction(trans, root, ret); | |
1418 | goto fail; | |
1419 | } | |
6bdb72de | 1420 | |
a22285a6 YZ |
1421 | /* |
1422 | * insert root back/forward references | |
1423 | */ | |
1424 | ret = btrfs_add_root_ref(trans, tree_root, objectid, | |
0660b5af | 1425 | parent_root->root_key.objectid, |
33345d01 | 1426 | btrfs_ino(parent_inode), index, |
a22285a6 | 1427 | dentry->d_name.name, dentry->d_name.len); |
8732d44f MX |
1428 | if (ret) { |
1429 | btrfs_abort_transaction(trans, root, ret); | |
1430 | goto fail; | |
1431 | } | |
0660b5af | 1432 | |
a22285a6 YZ |
1433 | key.offset = (u64)-1; |
1434 | pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key); | |
79787eaa JM |
1435 | if (IS_ERR(pending->snap)) { |
1436 | ret = PTR_ERR(pending->snap); | |
8732d44f MX |
1437 | btrfs_abort_transaction(trans, root, ret); |
1438 | goto fail; | |
79787eaa | 1439 | } |
d68fc57b | 1440 | |
49b25e05 | 1441 | ret = btrfs_reloc_post_snapshot(trans, pending); |
8732d44f MX |
1442 | if (ret) { |
1443 | btrfs_abort_transaction(trans, root, ret); | |
1444 | goto fail; | |
1445 | } | |
361048f5 MX |
1446 | |
1447 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); | |
8732d44f MX |
1448 | if (ret) { |
1449 | btrfs_abort_transaction(trans, root, ret); | |
1450 | goto fail; | |
1451 | } | |
42874b3d MX |
1452 | |
1453 | ret = btrfs_insert_dir_item(trans, parent_root, | |
1454 | dentry->d_name.name, dentry->d_name.len, | |
1455 | parent_inode, &key, | |
1456 | BTRFS_FT_DIR, index); | |
1457 | /* We have check then name at the beginning, so it is impossible. */ | |
9c52057c | 1458 | BUG_ON(ret == -EEXIST || ret == -EOVERFLOW); |
8732d44f MX |
1459 | if (ret) { |
1460 | btrfs_abort_transaction(trans, root, ret); | |
1461 | goto fail; | |
1462 | } | |
42874b3d MX |
1463 | |
1464 | btrfs_i_size_write(parent_inode, parent_inode->i_size + | |
1465 | dentry->d_name.len * 2); | |
1466 | parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME; | |
be6aef60 | 1467 | ret = btrfs_update_inode_fallback(trans, parent_root, parent_inode); |
dd5f9615 SB |
1468 | if (ret) { |
1469 | btrfs_abort_transaction(trans, root, ret); | |
1470 | goto fail; | |
1471 | } | |
1472 | ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root, new_uuid.b, | |
1473 | BTRFS_UUID_KEY_SUBVOL, objectid); | |
1474 | if (ret) { | |
8732d44f | 1475 | btrfs_abort_transaction(trans, root, ret); |
dd5f9615 SB |
1476 | goto fail; |
1477 | } | |
1478 | if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) { | |
1479 | ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root, | |
1480 | new_root_item->received_uuid, | |
1481 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, | |
1482 | objectid); | |
1483 | if (ret && ret != -EEXIST) { | |
1484 | btrfs_abort_transaction(trans, root, ret); | |
1485 | goto fail; | |
1486 | } | |
1487 | } | |
3063d29f | 1488 | fail: |
aec8030a MX |
1489 | pending->error = ret; |
1490 | dir_item_existed: | |
98c9942a | 1491 | trans->block_rsv = rsv; |
2382c5cc | 1492 | trans->bytes_reserved = 0; |
6fa9700e MX |
1493 | no_free_objectid: |
1494 | kfree(new_root_item); | |
1495 | root_item_alloc_fail: | |
42874b3d | 1496 | btrfs_free_path(path); |
49b25e05 | 1497 | return ret; |
3063d29f CM |
1498 | } |
1499 | ||
d352ac68 CM |
1500 | /* |
1501 | * create all the snapshots we've scheduled for creation | |
1502 | */ | |
80b6794d CM |
1503 | static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans, |
1504 | struct btrfs_fs_info *fs_info) | |
3de4586c | 1505 | { |
aec8030a | 1506 | struct btrfs_pending_snapshot *pending, *next; |
3de4586c | 1507 | struct list_head *head = &trans->transaction->pending_snapshots; |
aec8030a | 1508 | int ret = 0; |
3de4586c | 1509 | |
aec8030a MX |
1510 | list_for_each_entry_safe(pending, next, head, list) { |
1511 | list_del(&pending->list); | |
1512 | ret = create_pending_snapshot(trans, fs_info, pending); | |
1513 | if (ret) | |
1514 | break; | |
1515 | } | |
1516 | return ret; | |
3de4586c CM |
1517 | } |
1518 | ||
5d4f98a2 YZ |
1519 | static void update_super_roots(struct btrfs_root *root) |
1520 | { | |
1521 | struct btrfs_root_item *root_item; | |
1522 | struct btrfs_super_block *super; | |
1523 | ||
6c41761f | 1524 | super = root->fs_info->super_copy; |
5d4f98a2 YZ |
1525 | |
1526 | root_item = &root->fs_info->chunk_root->root_item; | |
1527 | super->chunk_root = root_item->bytenr; | |
1528 | super->chunk_root_generation = root_item->generation; | |
1529 | super->chunk_root_level = root_item->level; | |
1530 | ||
1531 | root_item = &root->fs_info->tree_root->root_item; | |
1532 | super->root = root_item->bytenr; | |
1533 | super->generation = root_item->generation; | |
1534 | super->root_level = root_item->level; | |
73bc1876 | 1535 | if (btrfs_test_opt(root, SPACE_CACHE)) |
0af3d00b | 1536 | super->cache_generation = root_item->generation; |
70f80175 SB |
1537 | if (root->fs_info->update_uuid_tree_gen) |
1538 | super->uuid_tree_generation = root_item->generation; | |
5d4f98a2 YZ |
1539 | } |
1540 | ||
f36f3042 CM |
1541 | int btrfs_transaction_in_commit(struct btrfs_fs_info *info) |
1542 | { | |
4a9d8bde | 1543 | struct btrfs_transaction *trans; |
f36f3042 | 1544 | int ret = 0; |
4a9d8bde | 1545 | |
a4abeea4 | 1546 | spin_lock(&info->trans_lock); |
4a9d8bde MX |
1547 | trans = info->running_transaction; |
1548 | if (trans) | |
1549 | ret = (trans->state >= TRANS_STATE_COMMIT_START); | |
a4abeea4 | 1550 | spin_unlock(&info->trans_lock); |
f36f3042 CM |
1551 | return ret; |
1552 | } | |
1553 | ||
8929ecfa YZ |
1554 | int btrfs_transaction_blocked(struct btrfs_fs_info *info) |
1555 | { | |
4a9d8bde | 1556 | struct btrfs_transaction *trans; |
8929ecfa | 1557 | int ret = 0; |
4a9d8bde | 1558 | |
a4abeea4 | 1559 | spin_lock(&info->trans_lock); |
4a9d8bde MX |
1560 | trans = info->running_transaction; |
1561 | if (trans) | |
1562 | ret = is_transaction_blocked(trans); | |
a4abeea4 | 1563 | spin_unlock(&info->trans_lock); |
8929ecfa YZ |
1564 | return ret; |
1565 | } | |
1566 | ||
bb9c12c9 SW |
1567 | /* |
1568 | * wait for the current transaction commit to start and block subsequent | |
1569 | * transaction joins | |
1570 | */ | |
1571 | static void wait_current_trans_commit_start(struct btrfs_root *root, | |
1572 | struct btrfs_transaction *trans) | |
1573 | { | |
4a9d8bde | 1574 | wait_event(root->fs_info->transaction_blocked_wait, |
501407aa JB |
1575 | trans->state >= TRANS_STATE_COMMIT_START || |
1576 | trans->aborted); | |
bb9c12c9 SW |
1577 | } |
1578 | ||
1579 | /* | |
1580 | * wait for the current transaction to start and then become unblocked. | |
1581 | * caller holds ref. | |
1582 | */ | |
1583 | static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root, | |
1584 | struct btrfs_transaction *trans) | |
1585 | { | |
72d63ed6 | 1586 | wait_event(root->fs_info->transaction_wait, |
501407aa JB |
1587 | trans->state >= TRANS_STATE_UNBLOCKED || |
1588 | trans->aborted); | |
bb9c12c9 SW |
1589 | } |
1590 | ||
1591 | /* | |
1592 | * commit transactions asynchronously. once btrfs_commit_transaction_async | |
1593 | * returns, any subsequent transaction will not be allowed to join. | |
1594 | */ | |
1595 | struct btrfs_async_commit { | |
1596 | struct btrfs_trans_handle *newtrans; | |
1597 | struct btrfs_root *root; | |
7892b5af | 1598 | struct work_struct work; |
bb9c12c9 SW |
1599 | }; |
1600 | ||
1601 | static void do_async_commit(struct work_struct *work) | |
1602 | { | |
1603 | struct btrfs_async_commit *ac = | |
7892b5af | 1604 | container_of(work, struct btrfs_async_commit, work); |
bb9c12c9 | 1605 | |
6fc4e354 SW |
1606 | /* |
1607 | * We've got freeze protection passed with the transaction. | |
1608 | * Tell lockdep about it. | |
1609 | */ | |
b1a06a4b | 1610 | if (ac->newtrans->type & __TRANS_FREEZABLE) |
ff7c1d33 MX |
1611 | rwsem_acquire_read( |
1612 | &ac->root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1], | |
1613 | 0, 1, _THIS_IP_); | |
6fc4e354 | 1614 | |
e209db7a SW |
1615 | current->journal_info = ac->newtrans; |
1616 | ||
bb9c12c9 SW |
1617 | btrfs_commit_transaction(ac->newtrans, ac->root); |
1618 | kfree(ac); | |
1619 | } | |
1620 | ||
1621 | int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans, | |
1622 | struct btrfs_root *root, | |
1623 | int wait_for_unblock) | |
1624 | { | |
1625 | struct btrfs_async_commit *ac; | |
1626 | struct btrfs_transaction *cur_trans; | |
1627 | ||
1628 | ac = kmalloc(sizeof(*ac), GFP_NOFS); | |
db5b493a TI |
1629 | if (!ac) |
1630 | return -ENOMEM; | |
bb9c12c9 | 1631 | |
7892b5af | 1632 | INIT_WORK(&ac->work, do_async_commit); |
bb9c12c9 | 1633 | ac->root = root; |
7a7eaa40 | 1634 | ac->newtrans = btrfs_join_transaction(root); |
3612b495 TI |
1635 | if (IS_ERR(ac->newtrans)) { |
1636 | int err = PTR_ERR(ac->newtrans); | |
1637 | kfree(ac); | |
1638 | return err; | |
1639 | } | |
bb9c12c9 SW |
1640 | |
1641 | /* take transaction reference */ | |
bb9c12c9 | 1642 | cur_trans = trans->transaction; |
13c5a93e | 1643 | atomic_inc(&cur_trans->use_count); |
bb9c12c9 SW |
1644 | |
1645 | btrfs_end_transaction(trans, root); | |
6fc4e354 SW |
1646 | |
1647 | /* | |
1648 | * Tell lockdep we've released the freeze rwsem, since the | |
1649 | * async commit thread will be the one to unlock it. | |
1650 | */ | |
b1a06a4b | 1651 | if (ac->newtrans->type & __TRANS_FREEZABLE) |
ff7c1d33 MX |
1652 | rwsem_release( |
1653 | &root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1], | |
1654 | 1, _THIS_IP_); | |
6fc4e354 | 1655 | |
7892b5af | 1656 | schedule_work(&ac->work); |
bb9c12c9 SW |
1657 | |
1658 | /* wait for transaction to start and unblock */ | |
bb9c12c9 SW |
1659 | if (wait_for_unblock) |
1660 | wait_current_trans_commit_start_and_unblock(root, cur_trans); | |
1661 | else | |
1662 | wait_current_trans_commit_start(root, cur_trans); | |
bb9c12c9 | 1663 | |
38e88054 SW |
1664 | if (current->journal_info == trans) |
1665 | current->journal_info = NULL; | |
1666 | ||
724e2315 | 1667 | btrfs_put_transaction(cur_trans); |
bb9c12c9 SW |
1668 | return 0; |
1669 | } | |
1670 | ||
49b25e05 JM |
1671 | |
1672 | static void cleanup_transaction(struct btrfs_trans_handle *trans, | |
7b8b92af | 1673 | struct btrfs_root *root, int err) |
49b25e05 JM |
1674 | { |
1675 | struct btrfs_transaction *cur_trans = trans->transaction; | |
f094ac32 | 1676 | DEFINE_WAIT(wait); |
49b25e05 JM |
1677 | |
1678 | WARN_ON(trans->use_count > 1); | |
1679 | ||
7b8b92af JB |
1680 | btrfs_abort_transaction(trans, root, err); |
1681 | ||
49b25e05 | 1682 | spin_lock(&root->fs_info->trans_lock); |
66b6135b | 1683 | |
25d8c284 MX |
1684 | /* |
1685 | * If the transaction is removed from the list, it means this | |
1686 | * transaction has been committed successfully, so it is impossible | |
1687 | * to call the cleanup function. | |
1688 | */ | |
1689 | BUG_ON(list_empty(&cur_trans->list)); | |
66b6135b | 1690 | |
49b25e05 | 1691 | list_del_init(&cur_trans->list); |
d7096fc3 | 1692 | if (cur_trans == root->fs_info->running_transaction) { |
4a9d8bde | 1693 | cur_trans->state = TRANS_STATE_COMMIT_DOING; |
f094ac32 LB |
1694 | spin_unlock(&root->fs_info->trans_lock); |
1695 | wait_event(cur_trans->writer_wait, | |
1696 | atomic_read(&cur_trans->num_writers) == 1); | |
1697 | ||
1698 | spin_lock(&root->fs_info->trans_lock); | |
d7096fc3 | 1699 | } |
49b25e05 JM |
1700 | spin_unlock(&root->fs_info->trans_lock); |
1701 | ||
1702 | btrfs_cleanup_one_transaction(trans->transaction, root); | |
1703 | ||
4a9d8bde MX |
1704 | spin_lock(&root->fs_info->trans_lock); |
1705 | if (cur_trans == root->fs_info->running_transaction) | |
1706 | root->fs_info->running_transaction = NULL; | |
1707 | spin_unlock(&root->fs_info->trans_lock); | |
1708 | ||
e0228285 JB |
1709 | if (trans->type & __TRANS_FREEZABLE) |
1710 | sb_end_intwrite(root->fs_info->sb); | |
724e2315 JB |
1711 | btrfs_put_transaction(cur_trans); |
1712 | btrfs_put_transaction(cur_trans); | |
49b25e05 JM |
1713 | |
1714 | trace_btrfs_transaction_commit(root); | |
1715 | ||
49b25e05 JM |
1716 | if (current->journal_info == trans) |
1717 | current->journal_info = NULL; | |
c0af8f0b | 1718 | btrfs_scrub_cancel(root->fs_info); |
49b25e05 JM |
1719 | |
1720 | kmem_cache_free(btrfs_trans_handle_cachep, trans); | |
1721 | } | |
1722 | ||
82436617 MX |
1723 | static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info) |
1724 | { | |
1725 | if (btrfs_test_opt(fs_info->tree_root, FLUSHONCOMMIT)) | |
6c255e67 | 1726 | return btrfs_start_delalloc_roots(fs_info, 1, -1); |
82436617 MX |
1727 | return 0; |
1728 | } | |
1729 | ||
1730 | static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info) | |
1731 | { | |
1732 | if (btrfs_test_opt(fs_info->tree_root, FLUSHONCOMMIT)) | |
b0244199 | 1733 | btrfs_wait_ordered_roots(fs_info, -1); |
82436617 MX |
1734 | } |
1735 | ||
50d9aa99 JB |
1736 | static inline void |
1737 | btrfs_wait_pending_ordered(struct btrfs_transaction *cur_trans, | |
1738 | struct btrfs_fs_info *fs_info) | |
1739 | { | |
1740 | struct btrfs_ordered_extent *ordered; | |
1741 | ||
1742 | spin_lock(&fs_info->trans_lock); | |
1743 | while (!list_empty(&cur_trans->pending_ordered)) { | |
1744 | ordered = list_first_entry(&cur_trans->pending_ordered, | |
1745 | struct btrfs_ordered_extent, | |
1746 | trans_list); | |
1747 | list_del_init(&ordered->trans_list); | |
1748 | spin_unlock(&fs_info->trans_lock); | |
1749 | ||
1750 | wait_event(ordered->wait, test_bit(BTRFS_ORDERED_COMPLETE, | |
1751 | &ordered->flags)); | |
1752 | btrfs_put_ordered_extent(ordered); | |
1753 | spin_lock(&fs_info->trans_lock); | |
1754 | } | |
1755 | spin_unlock(&fs_info->trans_lock); | |
1756 | } | |
1757 | ||
79154b1b CM |
1758 | int btrfs_commit_transaction(struct btrfs_trans_handle *trans, |
1759 | struct btrfs_root *root) | |
1760 | { | |
49b25e05 | 1761 | struct btrfs_transaction *cur_trans = trans->transaction; |
8fd17795 | 1762 | struct btrfs_transaction *prev_trans = NULL; |
656f30db | 1763 | struct btrfs_inode *btree_ino = BTRFS_I(root->fs_info->btree_inode); |
25287e0a | 1764 | int ret; |
79154b1b | 1765 | |
8d25a086 MX |
1766 | /* Stop the commit early if ->aborted is set */ |
1767 | if (unlikely(ACCESS_ONCE(cur_trans->aborted))) { | |
25287e0a | 1768 | ret = cur_trans->aborted; |
e4a2bcac JB |
1769 | btrfs_end_transaction(trans, root); |
1770 | return ret; | |
25287e0a | 1771 | } |
49b25e05 | 1772 | |
56bec294 CM |
1773 | /* make a pass through all the delayed refs we have so far |
1774 | * any runnings procs may add more while we are here | |
1775 | */ | |
1776 | ret = btrfs_run_delayed_refs(trans, root, 0); | |
e4a2bcac JB |
1777 | if (ret) { |
1778 | btrfs_end_transaction(trans, root); | |
1779 | return ret; | |
1780 | } | |
56bec294 | 1781 | |
0e721106 JB |
1782 | btrfs_trans_release_metadata(trans, root); |
1783 | trans->block_rsv = NULL; | |
272d26d0 MX |
1784 | if (trans->qgroup_reserved) { |
1785 | btrfs_qgroup_free(root, trans->qgroup_reserved); | |
1786 | trans->qgroup_reserved = 0; | |
1787 | } | |
0e721106 | 1788 | |
b7ec40d7 | 1789 | cur_trans = trans->transaction; |
49b25e05 | 1790 | |
56bec294 CM |
1791 | /* |
1792 | * set the flushing flag so procs in this transaction have to | |
1793 | * start sending their work down. | |
1794 | */ | |
b7ec40d7 | 1795 | cur_trans->delayed_refs.flushing = 1; |
1be41b78 | 1796 | smp_wmb(); |
56bec294 | 1797 | |
ea658bad JB |
1798 | if (!list_empty(&trans->new_bgs)) |
1799 | btrfs_create_pending_block_groups(trans, root); | |
1800 | ||
c3e69d58 | 1801 | ret = btrfs_run_delayed_refs(trans, root, 0); |
e4a2bcac JB |
1802 | if (ret) { |
1803 | btrfs_end_transaction(trans, root); | |
1804 | return ret; | |
1805 | } | |
56bec294 | 1806 | |
4a9d8bde | 1807 | spin_lock(&root->fs_info->trans_lock); |
50d9aa99 | 1808 | list_splice(&trans->ordered, &cur_trans->pending_ordered); |
4a9d8bde MX |
1809 | if (cur_trans->state >= TRANS_STATE_COMMIT_START) { |
1810 | spin_unlock(&root->fs_info->trans_lock); | |
13c5a93e | 1811 | atomic_inc(&cur_trans->use_count); |
49b25e05 | 1812 | ret = btrfs_end_transaction(trans, root); |
ccd467d6 | 1813 | |
b9c8300c | 1814 | wait_for_commit(root, cur_trans); |
15ee9bc7 | 1815 | |
724e2315 | 1816 | btrfs_put_transaction(cur_trans); |
15ee9bc7 | 1817 | |
49b25e05 | 1818 | return ret; |
79154b1b | 1819 | } |
4313b399 | 1820 | |
4a9d8bde | 1821 | cur_trans->state = TRANS_STATE_COMMIT_START; |
bb9c12c9 SW |
1822 | wake_up(&root->fs_info->transaction_blocked_wait); |
1823 | ||
ccd467d6 CM |
1824 | if (cur_trans->list.prev != &root->fs_info->trans_list) { |
1825 | prev_trans = list_entry(cur_trans->list.prev, | |
1826 | struct btrfs_transaction, list); | |
4a9d8bde | 1827 | if (prev_trans->state != TRANS_STATE_COMPLETED) { |
13c5a93e | 1828 | atomic_inc(&prev_trans->use_count); |
a4abeea4 | 1829 | spin_unlock(&root->fs_info->trans_lock); |
ccd467d6 CM |
1830 | |
1831 | wait_for_commit(root, prev_trans); | |
ccd467d6 | 1832 | |
724e2315 | 1833 | btrfs_put_transaction(prev_trans); |
a4abeea4 JB |
1834 | } else { |
1835 | spin_unlock(&root->fs_info->trans_lock); | |
ccd467d6 | 1836 | } |
a4abeea4 JB |
1837 | } else { |
1838 | spin_unlock(&root->fs_info->trans_lock); | |
ccd467d6 | 1839 | } |
15ee9bc7 | 1840 | |
0860adfd MX |
1841 | extwriter_counter_dec(cur_trans, trans->type); |
1842 | ||
82436617 MX |
1843 | ret = btrfs_start_delalloc_flush(root->fs_info); |
1844 | if (ret) | |
1845 | goto cleanup_transaction; | |
1846 | ||
8d875f95 | 1847 | ret = btrfs_run_delayed_items(trans, root); |
581227d0 MX |
1848 | if (ret) |
1849 | goto cleanup_transaction; | |
15ee9bc7 | 1850 | |
581227d0 MX |
1851 | wait_event(cur_trans->writer_wait, |
1852 | extwriter_counter_read(cur_trans) == 0); | |
15ee9bc7 | 1853 | |
581227d0 | 1854 | /* some pending stuffs might be added after the previous flush. */ |
8d875f95 | 1855 | ret = btrfs_run_delayed_items(trans, root); |
ca469637 MX |
1856 | if (ret) |
1857 | goto cleanup_transaction; | |
1858 | ||
82436617 | 1859 | btrfs_wait_delalloc_flush(root->fs_info); |
cb7ab021 | 1860 | |
50d9aa99 JB |
1861 | btrfs_wait_pending_ordered(cur_trans, root->fs_info); |
1862 | ||
cb7ab021 | 1863 | btrfs_scrub_pause(root); |
ed0ca140 JB |
1864 | /* |
1865 | * Ok now we need to make sure to block out any other joins while we | |
1866 | * commit the transaction. We could have started a join before setting | |
4a9d8bde | 1867 | * COMMIT_DOING so make sure to wait for num_writers to == 1 again. |
ed0ca140 JB |
1868 | */ |
1869 | spin_lock(&root->fs_info->trans_lock); | |
4a9d8bde | 1870 | cur_trans->state = TRANS_STATE_COMMIT_DOING; |
ed0ca140 JB |
1871 | spin_unlock(&root->fs_info->trans_lock); |
1872 | wait_event(cur_trans->writer_wait, | |
1873 | atomic_read(&cur_trans->num_writers) == 1); | |
1874 | ||
2cba30f1 MX |
1875 | /* ->aborted might be set after the previous check, so check it */ |
1876 | if (unlikely(ACCESS_ONCE(cur_trans->aborted))) { | |
1877 | ret = cur_trans->aborted; | |
6cf7f77e | 1878 | goto scrub_continue; |
2cba30f1 | 1879 | } |
7585717f CM |
1880 | /* |
1881 | * the reloc mutex makes sure that we stop | |
1882 | * the balancing code from coming in and moving | |
1883 | * extents around in the middle of the commit | |
1884 | */ | |
1885 | mutex_lock(&root->fs_info->reloc_mutex); | |
1886 | ||
42874b3d MX |
1887 | /* |
1888 | * We needn't worry about the delayed items because we will | |
1889 | * deal with them in create_pending_snapshot(), which is the | |
1890 | * core function of the snapshot creation. | |
1891 | */ | |
1892 | ret = create_pending_snapshots(trans, root->fs_info); | |
49b25e05 JM |
1893 | if (ret) { |
1894 | mutex_unlock(&root->fs_info->reloc_mutex); | |
6cf7f77e | 1895 | goto scrub_continue; |
49b25e05 | 1896 | } |
3063d29f | 1897 | |
42874b3d MX |
1898 | /* |
1899 | * We insert the dir indexes of the snapshots and update the inode | |
1900 | * of the snapshots' parents after the snapshot creation, so there | |
1901 | * are some delayed items which are not dealt with. Now deal with | |
1902 | * them. | |
1903 | * | |
1904 | * We needn't worry that this operation will corrupt the snapshots, | |
1905 | * because all the tree which are snapshoted will be forced to COW | |
1906 | * the nodes and leaves. | |
1907 | */ | |
1908 | ret = btrfs_run_delayed_items(trans, root); | |
49b25e05 JM |
1909 | if (ret) { |
1910 | mutex_unlock(&root->fs_info->reloc_mutex); | |
6cf7f77e | 1911 | goto scrub_continue; |
49b25e05 | 1912 | } |
16cdcec7 | 1913 | |
56bec294 | 1914 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
49b25e05 JM |
1915 | if (ret) { |
1916 | mutex_unlock(&root->fs_info->reloc_mutex); | |
6cf7f77e | 1917 | goto scrub_continue; |
49b25e05 | 1918 | } |
56bec294 | 1919 | |
e999376f CM |
1920 | /* |
1921 | * make sure none of the code above managed to slip in a | |
1922 | * delayed item | |
1923 | */ | |
1924 | btrfs_assert_delayed_root_empty(root); | |
1925 | ||
2c90e5d6 | 1926 | WARN_ON(cur_trans != trans->transaction); |
dc17ff8f | 1927 | |
e02119d5 CM |
1928 | /* btrfs_commit_tree_roots is responsible for getting the |
1929 | * various roots consistent with each other. Every pointer | |
1930 | * in the tree of tree roots has to point to the most up to date | |
1931 | * root for every subvolume and other tree. So, we have to keep | |
1932 | * the tree logging code from jumping in and changing any | |
1933 | * of the trees. | |
1934 | * | |
1935 | * At this point in the commit, there can't be any tree-log | |
1936 | * writers, but a little lower down we drop the trans mutex | |
1937 | * and let new people in. By holding the tree_log_mutex | |
1938 | * from now until after the super is written, we avoid races | |
1939 | * with the tree-log code. | |
1940 | */ | |
1941 | mutex_lock(&root->fs_info->tree_log_mutex); | |
1942 | ||
5d4f98a2 | 1943 | ret = commit_fs_roots(trans, root); |
49b25e05 JM |
1944 | if (ret) { |
1945 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
871383be | 1946 | mutex_unlock(&root->fs_info->reloc_mutex); |
6cf7f77e | 1947 | goto scrub_continue; |
49b25e05 | 1948 | } |
54aa1f4d | 1949 | |
3818aea2 | 1950 | /* |
7e1876ac DS |
1951 | * Since the transaction is done, we can apply the pending changes |
1952 | * before the next transaction. | |
3818aea2 | 1953 | */ |
572d9ab7 | 1954 | btrfs_apply_pending_changes(root->fs_info); |
3818aea2 | 1955 | |
5d4f98a2 | 1956 | /* commit_fs_roots gets rid of all the tree log roots, it is now |
e02119d5 CM |
1957 | * safe to free the root of tree log roots |
1958 | */ | |
1959 | btrfs_free_log_root_tree(trans, root->fs_info); | |
1960 | ||
5d4f98a2 | 1961 | ret = commit_cowonly_roots(trans, root); |
49b25e05 JM |
1962 | if (ret) { |
1963 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
871383be | 1964 | mutex_unlock(&root->fs_info->reloc_mutex); |
6cf7f77e | 1965 | goto scrub_continue; |
49b25e05 | 1966 | } |
54aa1f4d | 1967 | |
2cba30f1 MX |
1968 | /* |
1969 | * The tasks which save the space cache and inode cache may also | |
1970 | * update ->aborted, check it. | |
1971 | */ | |
1972 | if (unlikely(ACCESS_ONCE(cur_trans->aborted))) { | |
1973 | ret = cur_trans->aborted; | |
1974 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
1975 | mutex_unlock(&root->fs_info->reloc_mutex); | |
6cf7f77e | 1976 | goto scrub_continue; |
2cba30f1 MX |
1977 | } |
1978 | ||
11833d66 YZ |
1979 | btrfs_prepare_extent_commit(trans, root); |
1980 | ||
78fae27e | 1981 | cur_trans = root->fs_info->running_transaction; |
5d4f98a2 YZ |
1982 | |
1983 | btrfs_set_root_node(&root->fs_info->tree_root->root_item, | |
1984 | root->fs_info->tree_root->node); | |
9e351cc8 JB |
1985 | list_add_tail(&root->fs_info->tree_root->dirty_list, |
1986 | &cur_trans->switch_commits); | |
5d4f98a2 YZ |
1987 | |
1988 | btrfs_set_root_node(&root->fs_info->chunk_root->root_item, | |
1989 | root->fs_info->chunk_root->node); | |
9e351cc8 JB |
1990 | list_add_tail(&root->fs_info->chunk_root->dirty_list, |
1991 | &cur_trans->switch_commits); | |
1992 | ||
1993 | switch_commit_roots(cur_trans, root->fs_info); | |
5d4f98a2 | 1994 | |
edf39272 | 1995 | assert_qgroups_uptodate(trans); |
ce93ec54 | 1996 | ASSERT(list_empty(&cur_trans->dirty_bgs)); |
5d4f98a2 | 1997 | update_super_roots(root); |
e02119d5 | 1998 | |
60e7cd3a JB |
1999 | btrfs_set_super_log_root(root->fs_info->super_copy, 0); |
2000 | btrfs_set_super_log_root_level(root->fs_info->super_copy, 0); | |
6c41761f DS |
2001 | memcpy(root->fs_info->super_for_commit, root->fs_info->super_copy, |
2002 | sizeof(*root->fs_info->super_copy)); | |
ccd467d6 | 2003 | |
935e5cc9 | 2004 | btrfs_update_commit_device_size(root->fs_info); |
ce7213c7 | 2005 | btrfs_update_commit_device_bytes_used(root, cur_trans); |
935e5cc9 | 2006 | |
656f30db FM |
2007 | clear_bit(BTRFS_INODE_BTREE_LOG1_ERR, &btree_ino->runtime_flags); |
2008 | clear_bit(BTRFS_INODE_BTREE_LOG2_ERR, &btree_ino->runtime_flags); | |
2009 | ||
a4abeea4 | 2010 | spin_lock(&root->fs_info->trans_lock); |
4a9d8bde | 2011 | cur_trans->state = TRANS_STATE_UNBLOCKED; |
a4abeea4 | 2012 | root->fs_info->running_transaction = NULL; |
a4abeea4 | 2013 | spin_unlock(&root->fs_info->trans_lock); |
7585717f | 2014 | mutex_unlock(&root->fs_info->reloc_mutex); |
b7ec40d7 | 2015 | |
f9295749 | 2016 | wake_up(&root->fs_info->transaction_wait); |
e6dcd2dc | 2017 | |
79154b1b | 2018 | ret = btrfs_write_and_wait_transaction(trans, root); |
49b25e05 JM |
2019 | if (ret) { |
2020 | btrfs_error(root->fs_info, ret, | |
08748810 | 2021 | "Error while writing out transaction"); |
49b25e05 | 2022 | mutex_unlock(&root->fs_info->tree_log_mutex); |
6cf7f77e | 2023 | goto scrub_continue; |
49b25e05 JM |
2024 | } |
2025 | ||
2026 | ret = write_ctree_super(trans, root, 0); | |
2027 | if (ret) { | |
2028 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
6cf7f77e | 2029 | goto scrub_continue; |
49b25e05 | 2030 | } |
4313b399 | 2031 | |
e02119d5 CM |
2032 | /* |
2033 | * the super is written, we can safely allow the tree-loggers | |
2034 | * to go about their business | |
2035 | */ | |
2036 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
2037 | ||
11833d66 | 2038 | btrfs_finish_extent_commit(trans, root); |
4313b399 | 2039 | |
15ee9bc7 | 2040 | root->fs_info->last_trans_committed = cur_trans->transid; |
4a9d8bde MX |
2041 | /* |
2042 | * We needn't acquire the lock here because there is no other task | |
2043 | * which can change it. | |
2044 | */ | |
2045 | cur_trans->state = TRANS_STATE_COMPLETED; | |
2c90e5d6 | 2046 | wake_up(&cur_trans->commit_wait); |
3de4586c | 2047 | |
a4abeea4 | 2048 | spin_lock(&root->fs_info->trans_lock); |
13c5a93e | 2049 | list_del_init(&cur_trans->list); |
a4abeea4 JB |
2050 | spin_unlock(&root->fs_info->trans_lock); |
2051 | ||
724e2315 JB |
2052 | btrfs_put_transaction(cur_trans); |
2053 | btrfs_put_transaction(cur_trans); | |
58176a96 | 2054 | |
0860adfd | 2055 | if (trans->type & __TRANS_FREEZABLE) |
354aa0fb | 2056 | sb_end_intwrite(root->fs_info->sb); |
b2b5ef5c | 2057 | |
1abe9b8a | 2058 | trace_btrfs_transaction_commit(root); |
2059 | ||
a2de733c AJ |
2060 | btrfs_scrub_continue(root); |
2061 | ||
9ed74f2d JB |
2062 | if (current->journal_info == trans) |
2063 | current->journal_info = NULL; | |
2064 | ||
2c90e5d6 | 2065 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
24bbcf04 YZ |
2066 | |
2067 | if (current != root->fs_info->transaction_kthread) | |
2068 | btrfs_run_delayed_iputs(root); | |
2069 | ||
79154b1b | 2070 | return ret; |
49b25e05 | 2071 | |
6cf7f77e WS |
2072 | scrub_continue: |
2073 | btrfs_scrub_continue(root); | |
49b25e05 | 2074 | cleanup_transaction: |
0e721106 JB |
2075 | btrfs_trans_release_metadata(trans, root); |
2076 | trans->block_rsv = NULL; | |
272d26d0 MX |
2077 | if (trans->qgroup_reserved) { |
2078 | btrfs_qgroup_free(root, trans->qgroup_reserved); | |
2079 | trans->qgroup_reserved = 0; | |
2080 | } | |
c2cf52eb | 2081 | btrfs_warn(root->fs_info, "Skipping commit of aborted transaction."); |
49b25e05 JM |
2082 | if (current->journal_info == trans) |
2083 | current->journal_info = NULL; | |
7b8b92af | 2084 | cleanup_transaction(trans, root, ret); |
49b25e05 JM |
2085 | |
2086 | return ret; | |
79154b1b CM |
2087 | } |
2088 | ||
d352ac68 | 2089 | /* |
9d1a2a3a DS |
2090 | * return < 0 if error |
2091 | * 0 if there are no more dead_roots at the time of call | |
2092 | * 1 there are more to be processed, call me again | |
2093 | * | |
2094 | * The return value indicates there are certainly more snapshots to delete, but | |
2095 | * if there comes a new one during processing, it may return 0. We don't mind, | |
2096 | * because btrfs_commit_super will poke cleaner thread and it will process it a | |
2097 | * few seconds later. | |
d352ac68 | 2098 | */ |
9d1a2a3a | 2099 | int btrfs_clean_one_deleted_snapshot(struct btrfs_root *root) |
e9d0b13b | 2100 | { |
9d1a2a3a | 2101 | int ret; |
5d4f98a2 YZ |
2102 | struct btrfs_fs_info *fs_info = root->fs_info; |
2103 | ||
a4abeea4 | 2104 | spin_lock(&fs_info->trans_lock); |
9d1a2a3a DS |
2105 | if (list_empty(&fs_info->dead_roots)) { |
2106 | spin_unlock(&fs_info->trans_lock); | |
2107 | return 0; | |
2108 | } | |
2109 | root = list_first_entry(&fs_info->dead_roots, | |
2110 | struct btrfs_root, root_list); | |
cfad392b | 2111 | list_del_init(&root->root_list); |
a4abeea4 | 2112 | spin_unlock(&fs_info->trans_lock); |
e9d0b13b | 2113 | |
efe120a0 | 2114 | pr_debug("BTRFS: cleaner removing %llu\n", root->objectid); |
76dda93c | 2115 | |
9d1a2a3a | 2116 | btrfs_kill_all_delayed_nodes(root); |
16cdcec7 | 2117 | |
9d1a2a3a DS |
2118 | if (btrfs_header_backref_rev(root->node) < |
2119 | BTRFS_MIXED_BACKREF_REV) | |
2120 | ret = btrfs_drop_snapshot(root, NULL, 0, 0); | |
2121 | else | |
2122 | ret = btrfs_drop_snapshot(root, NULL, 1, 0); | |
32471dc2 | 2123 | |
6596a928 | 2124 | return (ret < 0) ? 0 : 1; |
e9d0b13b | 2125 | } |
572d9ab7 DS |
2126 | |
2127 | void btrfs_apply_pending_changes(struct btrfs_fs_info *fs_info) | |
2128 | { | |
2129 | unsigned long prev; | |
2130 | unsigned long bit; | |
2131 | ||
2132 | prev = cmpxchg(&fs_info->pending_changes, 0, 0); | |
2133 | if (!prev) | |
2134 | return; | |
2135 | ||
7e1876ac DS |
2136 | bit = 1 << BTRFS_PENDING_SET_INODE_MAP_CACHE; |
2137 | if (prev & bit) | |
2138 | btrfs_set_opt(fs_info->mount_opt, INODE_MAP_CACHE); | |
2139 | prev &= ~bit; | |
2140 | ||
2141 | bit = 1 << BTRFS_PENDING_CLEAR_INODE_MAP_CACHE; | |
2142 | if (prev & bit) | |
2143 | btrfs_clear_opt(fs_info->mount_opt, INODE_MAP_CACHE); | |
2144 | prev &= ~bit; | |
2145 | ||
d51033d0 DS |
2146 | bit = 1 << BTRFS_PENDING_COMMIT; |
2147 | if (prev & bit) | |
2148 | btrfs_debug(fs_info, "pending commit done"); | |
2149 | prev &= ~bit; | |
2150 | ||
572d9ab7 DS |
2151 | if (prev) |
2152 | btrfs_warn(fs_info, | |
2153 | "unknown pending changes left 0x%lx, ignoring", prev); | |
2154 | } |