Commit | Line | Data |
---|---|---|
6cbd5570 CM |
1 | /* |
2 | * Copyright (C) 2007 Oracle. All rights reserved. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public | |
6 | * License v2 as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, | |
9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public | |
14 | * License along with this program; if not, write to the | |
15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
16 | * Boston, MA 021110-1307, USA. | |
17 | */ | |
18 | ||
79154b1b | 19 | #include <linux/fs.h> |
5a0e3ad6 | 20 | #include <linux/slab.h> |
34088780 | 21 | #include <linux/sched.h> |
d3c2fdcf | 22 | #include <linux/writeback.h> |
5f39d397 | 23 | #include <linux/pagemap.h> |
5f2cc086 | 24 | #include <linux/blkdev.h> |
8ea05e3a | 25 | #include <linux/uuid.h> |
79154b1b CM |
26 | #include "ctree.h" |
27 | #include "disk-io.h" | |
28 | #include "transaction.h" | |
925baedd | 29 | #include "locking.h" |
e02119d5 | 30 | #include "tree-log.h" |
581bb050 | 31 | #include "inode-map.h" |
733f4fbb | 32 | #include "volumes.h" |
8dabb742 | 33 | #include "dev-replace.h" |
fcebe456 | 34 | #include "qgroup.h" |
79154b1b | 35 | |
0f7d52f4 CM |
36 | #define BTRFS_ROOT_TRANS_TAG 0 |
37 | ||
e8c9f186 | 38 | static const unsigned int btrfs_blocked_trans_types[TRANS_STATE_MAX] = { |
4a9d8bde MX |
39 | [TRANS_STATE_RUNNING] = 0U, |
40 | [TRANS_STATE_BLOCKED] = (__TRANS_USERSPACE | | |
41 | __TRANS_START), | |
42 | [TRANS_STATE_COMMIT_START] = (__TRANS_USERSPACE | | |
43 | __TRANS_START | | |
44 | __TRANS_ATTACH), | |
45 | [TRANS_STATE_COMMIT_DOING] = (__TRANS_USERSPACE | | |
46 | __TRANS_START | | |
47 | __TRANS_ATTACH | | |
48 | __TRANS_JOIN), | |
49 | [TRANS_STATE_UNBLOCKED] = (__TRANS_USERSPACE | | |
50 | __TRANS_START | | |
51 | __TRANS_ATTACH | | |
52 | __TRANS_JOIN | | |
53 | __TRANS_JOIN_NOLOCK), | |
54 | [TRANS_STATE_COMPLETED] = (__TRANS_USERSPACE | | |
55 | __TRANS_START | | |
56 | __TRANS_ATTACH | | |
57 | __TRANS_JOIN | | |
58 | __TRANS_JOIN_NOLOCK), | |
59 | }; | |
60 | ||
724e2315 | 61 | void btrfs_put_transaction(struct btrfs_transaction *transaction) |
79154b1b | 62 | { |
13c5a93e JB |
63 | WARN_ON(atomic_read(&transaction->use_count) == 0); |
64 | if (atomic_dec_and_test(&transaction->use_count)) { | |
a4abeea4 | 65 | BUG_ON(!list_empty(&transaction->list)); |
c46effa6 | 66 | WARN_ON(!RB_EMPTY_ROOT(&transaction->delayed_refs.href_root)); |
1262133b JB |
67 | if (transaction->delayed_refs.pending_csums) |
68 | printk(KERN_ERR "pending csums is %llu\n", | |
69 | transaction->delayed_refs.pending_csums); | |
6df9a95e JB |
70 | while (!list_empty(&transaction->pending_chunks)) { |
71 | struct extent_map *em; | |
72 | ||
73 | em = list_first_entry(&transaction->pending_chunks, | |
74 | struct extent_map, list); | |
75 | list_del_init(&em->list); | |
76 | free_extent_map(em); | |
77 | } | |
2c90e5d6 | 78 | kmem_cache_free(btrfs_transaction_cachep, transaction); |
78fae27e | 79 | } |
79154b1b CM |
80 | } |
81 | ||
663dfbb0 FM |
82 | static void clear_btree_io_tree(struct extent_io_tree *tree) |
83 | { | |
84 | spin_lock(&tree->lock); | |
85 | while (!RB_EMPTY_ROOT(&tree->state)) { | |
86 | struct rb_node *node; | |
87 | struct extent_state *state; | |
88 | ||
89 | node = rb_first(&tree->state); | |
90 | state = rb_entry(node, struct extent_state, rb_node); | |
91 | rb_erase(&state->rb_node, &tree->state); | |
92 | RB_CLEAR_NODE(&state->rb_node); | |
93 | /* | |
94 | * btree io trees aren't supposed to have tasks waiting for | |
95 | * changes in the flags of extent states ever. | |
96 | */ | |
97 | ASSERT(!waitqueue_active(&state->wq)); | |
98 | free_extent_state(state); | |
351810c1 DS |
99 | |
100 | cond_resched_lock(&tree->lock); | |
663dfbb0 FM |
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 | 119 | } |
2b9dbef2 JB |
120 | |
121 | /* We can free old roots now. */ | |
122 | spin_lock(&trans->dropped_roots_lock); | |
123 | while (!list_empty(&trans->dropped_roots)) { | |
124 | root = list_first_entry(&trans->dropped_roots, | |
125 | struct btrfs_root, root_list); | |
126 | list_del_init(&root->root_list); | |
127 | spin_unlock(&trans->dropped_roots_lock); | |
128 | btrfs_drop_and_free_fs_root(fs_info, root); | |
129 | spin_lock(&trans->dropped_roots_lock); | |
130 | } | |
131 | spin_unlock(&trans->dropped_roots_lock); | |
9e351cc8 | 132 | up_write(&fs_info->commit_root_sem); |
817d52f8 JB |
133 | } |
134 | ||
0860adfd MX |
135 | static inline void extwriter_counter_inc(struct btrfs_transaction *trans, |
136 | unsigned int type) | |
137 | { | |
138 | if (type & TRANS_EXTWRITERS) | |
139 | atomic_inc(&trans->num_extwriters); | |
140 | } | |
141 | ||
142 | static inline void extwriter_counter_dec(struct btrfs_transaction *trans, | |
143 | unsigned int type) | |
144 | { | |
145 | if (type & TRANS_EXTWRITERS) | |
146 | atomic_dec(&trans->num_extwriters); | |
147 | } | |
148 | ||
149 | static inline void extwriter_counter_init(struct btrfs_transaction *trans, | |
150 | unsigned int type) | |
151 | { | |
152 | atomic_set(&trans->num_extwriters, ((type & TRANS_EXTWRITERS) ? 1 : 0)); | |
153 | } | |
154 | ||
155 | static inline int extwriter_counter_read(struct btrfs_transaction *trans) | |
156 | { | |
157 | return atomic_read(&trans->num_extwriters); | |
178260b2 MX |
158 | } |
159 | ||
d352ac68 CM |
160 | /* |
161 | * either allocate a new transaction or hop into the existing one | |
162 | */ | |
0860adfd | 163 | static noinline int join_transaction(struct btrfs_root *root, unsigned int type) |
79154b1b CM |
164 | { |
165 | struct btrfs_transaction *cur_trans; | |
19ae4e81 | 166 | struct btrfs_fs_info *fs_info = root->fs_info; |
a4abeea4 | 167 | |
19ae4e81 | 168 | spin_lock(&fs_info->trans_lock); |
d43317dc | 169 | loop: |
49b25e05 | 170 | /* The file system has been taken offline. No new transactions. */ |
87533c47 | 171 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { |
19ae4e81 | 172 | spin_unlock(&fs_info->trans_lock); |
49b25e05 JM |
173 | return -EROFS; |
174 | } | |
175 | ||
19ae4e81 | 176 | cur_trans = fs_info->running_transaction; |
a4abeea4 | 177 | if (cur_trans) { |
871383be | 178 | if (cur_trans->aborted) { |
19ae4e81 | 179 | spin_unlock(&fs_info->trans_lock); |
49b25e05 | 180 | return cur_trans->aborted; |
871383be | 181 | } |
4a9d8bde | 182 | if (btrfs_blocked_trans_types[cur_trans->state] & type) { |
178260b2 MX |
183 | spin_unlock(&fs_info->trans_lock); |
184 | return -EBUSY; | |
185 | } | |
a4abeea4 | 186 | atomic_inc(&cur_trans->use_count); |
13c5a93e | 187 | atomic_inc(&cur_trans->num_writers); |
0860adfd | 188 | extwriter_counter_inc(cur_trans, type); |
19ae4e81 | 189 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 | 190 | return 0; |
79154b1b | 191 | } |
19ae4e81 | 192 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 | 193 | |
354aa0fb MX |
194 | /* |
195 | * If we are ATTACH, we just want to catch the current transaction, | |
196 | * and commit it. If there is no transaction, just return ENOENT. | |
197 | */ | |
198 | if (type == TRANS_ATTACH) | |
199 | return -ENOENT; | |
200 | ||
4a9d8bde MX |
201 | /* |
202 | * JOIN_NOLOCK only happens during the transaction commit, so | |
203 | * it is impossible that ->running_transaction is NULL | |
204 | */ | |
205 | BUG_ON(type == TRANS_JOIN_NOLOCK); | |
206 | ||
a4abeea4 JB |
207 | cur_trans = kmem_cache_alloc(btrfs_transaction_cachep, GFP_NOFS); |
208 | if (!cur_trans) | |
209 | return -ENOMEM; | |
d43317dc | 210 | |
19ae4e81 JS |
211 | spin_lock(&fs_info->trans_lock); |
212 | if (fs_info->running_transaction) { | |
d43317dc CM |
213 | /* |
214 | * someone started a transaction after we unlocked. Make sure | |
4a9d8bde | 215 | * to redo the checks above |
d43317dc | 216 | */ |
a4abeea4 | 217 | kmem_cache_free(btrfs_transaction_cachep, cur_trans); |
d43317dc | 218 | goto loop; |
87533c47 | 219 | } else if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { |
e4b50e14 | 220 | spin_unlock(&fs_info->trans_lock); |
7b8b92af JB |
221 | kmem_cache_free(btrfs_transaction_cachep, cur_trans); |
222 | return -EROFS; | |
79154b1b | 223 | } |
d43317dc | 224 | |
a4abeea4 | 225 | atomic_set(&cur_trans->num_writers, 1); |
0860adfd | 226 | extwriter_counter_init(cur_trans, type); |
a4abeea4 JB |
227 | init_waitqueue_head(&cur_trans->writer_wait); |
228 | init_waitqueue_head(&cur_trans->commit_wait); | |
4a9d8bde | 229 | cur_trans->state = TRANS_STATE_RUNNING; |
a4abeea4 JB |
230 | /* |
231 | * One for this trans handle, one so it will live on until we | |
232 | * commit the transaction. | |
233 | */ | |
234 | atomic_set(&cur_trans->use_count, 2); | |
13212b54 | 235 | cur_trans->have_free_bgs = 0; |
a4abeea4 | 236 | cur_trans->start_time = get_seconds(); |
1bbc621e | 237 | cur_trans->dirty_bg_run = 0; |
a4abeea4 | 238 | |
a099d0fd AM |
239 | memset(&cur_trans->delayed_refs, 0, sizeof(cur_trans->delayed_refs)); |
240 | ||
c46effa6 | 241 | cur_trans->delayed_refs.href_root = RB_ROOT; |
3368d001 | 242 | cur_trans->delayed_refs.dirty_extent_root = RB_ROOT; |
d7df2c79 | 243 | atomic_set(&cur_trans->delayed_refs.num_entries, 0); |
20b297d6 JS |
244 | |
245 | /* | |
246 | * although the tree mod log is per file system and not per transaction, | |
247 | * the log must never go across transaction boundaries. | |
248 | */ | |
249 | smp_mb(); | |
31b1a2bd | 250 | if (!list_empty(&fs_info->tree_mod_seq_list)) |
efe120a0 | 251 | WARN(1, KERN_ERR "BTRFS: tree_mod_seq_list not empty when " |
20b297d6 | 252 | "creating a fresh transaction\n"); |
31b1a2bd | 253 | if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log)) |
efe120a0 | 254 | WARN(1, KERN_ERR "BTRFS: tree_mod_log rb tree not empty when " |
20b297d6 | 255 | "creating a fresh transaction\n"); |
fc36ed7e | 256 | atomic64_set(&fs_info->tree_mod_seq, 0); |
20b297d6 | 257 | |
a4abeea4 JB |
258 | spin_lock_init(&cur_trans->delayed_refs.lock); |
259 | ||
260 | INIT_LIST_HEAD(&cur_trans->pending_snapshots); | |
6df9a95e | 261 | INIT_LIST_HEAD(&cur_trans->pending_chunks); |
9e351cc8 | 262 | INIT_LIST_HEAD(&cur_trans->switch_commits); |
50d9aa99 | 263 | INIT_LIST_HEAD(&cur_trans->pending_ordered); |
ce93ec54 | 264 | INIT_LIST_HEAD(&cur_trans->dirty_bgs); |
1bbc621e | 265 | INIT_LIST_HEAD(&cur_trans->io_bgs); |
2b9dbef2 | 266 | INIT_LIST_HEAD(&cur_trans->dropped_roots); |
1bbc621e | 267 | mutex_init(&cur_trans->cache_write_mutex); |
cb723e49 | 268 | cur_trans->num_dirty_bgs = 0; |
ce93ec54 | 269 | spin_lock_init(&cur_trans->dirty_bgs_lock); |
e33e17ee JM |
270 | INIT_LIST_HEAD(&cur_trans->deleted_bgs); |
271 | spin_lock_init(&cur_trans->deleted_bgs_lock); | |
2b9dbef2 | 272 | spin_lock_init(&cur_trans->dropped_roots_lock); |
19ae4e81 | 273 | list_add_tail(&cur_trans->list, &fs_info->trans_list); |
a4abeea4 | 274 | extent_io_tree_init(&cur_trans->dirty_pages, |
19ae4e81 JS |
275 | fs_info->btree_inode->i_mapping); |
276 | fs_info->generation++; | |
277 | cur_trans->transid = fs_info->generation; | |
278 | fs_info->running_transaction = cur_trans; | |
49b25e05 | 279 | cur_trans->aborted = 0; |
19ae4e81 | 280 | spin_unlock(&fs_info->trans_lock); |
15ee9bc7 | 281 | |
79154b1b CM |
282 | return 0; |
283 | } | |
284 | ||
d352ac68 | 285 | /* |
d397712b CM |
286 | * this does all the record keeping required to make sure that a reference |
287 | * counted root is properly recorded in a given transaction. This is required | |
288 | * to make sure the old root from before we joined the transaction is deleted | |
289 | * when the transaction commits | |
d352ac68 | 290 | */ |
7585717f | 291 | static int record_root_in_trans(struct btrfs_trans_handle *trans, |
a4abeea4 | 292 | struct btrfs_root *root) |
6702ed49 | 293 | { |
27cdeb70 MX |
294 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
295 | root->last_trans < trans->transid) { | |
6702ed49 | 296 | WARN_ON(root == root->fs_info->extent_root); |
5d4f98a2 YZ |
297 | WARN_ON(root->commit_root != root->node); |
298 | ||
7585717f | 299 | /* |
27cdeb70 | 300 | * see below for IN_TRANS_SETUP usage rules |
7585717f CM |
301 | * we have the reloc mutex held now, so there |
302 | * is only one writer in this function | |
303 | */ | |
27cdeb70 | 304 | set_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); |
7585717f | 305 | |
27cdeb70 | 306 | /* make sure readers find IN_TRANS_SETUP before |
7585717f CM |
307 | * they find our root->last_trans update |
308 | */ | |
309 | smp_wmb(); | |
310 | ||
a4abeea4 JB |
311 | spin_lock(&root->fs_info->fs_roots_radix_lock); |
312 | if (root->last_trans == trans->transid) { | |
313 | spin_unlock(&root->fs_info->fs_roots_radix_lock); | |
314 | return 0; | |
315 | } | |
5d4f98a2 YZ |
316 | radix_tree_tag_set(&root->fs_info->fs_roots_radix, |
317 | (unsigned long)root->root_key.objectid, | |
318 | BTRFS_ROOT_TRANS_TAG); | |
a4abeea4 | 319 | spin_unlock(&root->fs_info->fs_roots_radix_lock); |
7585717f CM |
320 | root->last_trans = trans->transid; |
321 | ||
322 | /* this is pretty tricky. We don't want to | |
323 | * take the relocation lock in btrfs_record_root_in_trans | |
324 | * unless we're really doing the first setup for this root in | |
325 | * this transaction. | |
326 | * | |
327 | * Normally we'd use root->last_trans as a flag to decide | |
328 | * if we want to take the expensive mutex. | |
329 | * | |
330 | * But, we have to set root->last_trans before we | |
331 | * init the relocation root, otherwise, we trip over warnings | |
332 | * in ctree.c. The solution used here is to flag ourselves | |
27cdeb70 | 333 | * with root IN_TRANS_SETUP. When this is 1, we're still |
7585717f CM |
334 | * fixing up the reloc trees and everyone must wait. |
335 | * | |
336 | * When this is zero, they can trust root->last_trans and fly | |
337 | * through btrfs_record_root_in_trans without having to take the | |
338 | * lock. smp_wmb() makes sure that all the writes above are | |
339 | * done before we pop in the zero below | |
340 | */ | |
5d4f98a2 | 341 | btrfs_init_reloc_root(trans, root); |
c7548af6 | 342 | smp_mb__before_atomic(); |
27cdeb70 | 343 | clear_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); |
5d4f98a2 YZ |
344 | } |
345 | return 0; | |
346 | } | |
bcc63abb | 347 | |
7585717f | 348 | |
2b9dbef2 JB |
349 | void btrfs_add_dropped_root(struct btrfs_trans_handle *trans, |
350 | struct btrfs_root *root) | |
351 | { | |
352 | struct btrfs_transaction *cur_trans = trans->transaction; | |
353 | ||
354 | /* Add ourselves to the transaction dropped list */ | |
355 | spin_lock(&cur_trans->dropped_roots_lock); | |
356 | list_add_tail(&root->root_list, &cur_trans->dropped_roots); | |
357 | spin_unlock(&cur_trans->dropped_roots_lock); | |
358 | ||
359 | /* Make sure we don't try to update the root at commit time */ | |
360 | spin_lock(&root->fs_info->fs_roots_radix_lock); | |
361 | radix_tree_tag_clear(&root->fs_info->fs_roots_radix, | |
362 | (unsigned long)root->root_key.objectid, | |
363 | BTRFS_ROOT_TRANS_TAG); | |
364 | spin_unlock(&root->fs_info->fs_roots_radix_lock); | |
365 | } | |
366 | ||
7585717f CM |
367 | int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, |
368 | struct btrfs_root *root) | |
369 | { | |
27cdeb70 | 370 | if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state)) |
7585717f CM |
371 | return 0; |
372 | ||
373 | /* | |
27cdeb70 | 374 | * see record_root_in_trans for comments about IN_TRANS_SETUP usage |
7585717f CM |
375 | * and barriers |
376 | */ | |
377 | smp_rmb(); | |
378 | if (root->last_trans == trans->transid && | |
27cdeb70 | 379 | !test_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state)) |
7585717f CM |
380 | return 0; |
381 | ||
382 | mutex_lock(&root->fs_info->reloc_mutex); | |
383 | record_root_in_trans(trans, root); | |
384 | mutex_unlock(&root->fs_info->reloc_mutex); | |
385 | ||
386 | return 0; | |
387 | } | |
388 | ||
4a9d8bde MX |
389 | static inline int is_transaction_blocked(struct btrfs_transaction *trans) |
390 | { | |
391 | return (trans->state >= TRANS_STATE_BLOCKED && | |
501407aa JB |
392 | trans->state < TRANS_STATE_UNBLOCKED && |
393 | !trans->aborted); | |
4a9d8bde MX |
394 | } |
395 | ||
d352ac68 CM |
396 | /* wait for commit against the current transaction to become unblocked |
397 | * when this is done, it is safe to start a new transaction, but the current | |
398 | * transaction might not be fully on disk. | |
399 | */ | |
37d1aeee | 400 | static void wait_current_trans(struct btrfs_root *root) |
79154b1b | 401 | { |
f9295749 | 402 | struct btrfs_transaction *cur_trans; |
79154b1b | 403 | |
a4abeea4 | 404 | spin_lock(&root->fs_info->trans_lock); |
f9295749 | 405 | cur_trans = root->fs_info->running_transaction; |
4a9d8bde | 406 | if (cur_trans && is_transaction_blocked(cur_trans)) { |
13c5a93e | 407 | atomic_inc(&cur_trans->use_count); |
a4abeea4 | 408 | spin_unlock(&root->fs_info->trans_lock); |
72d63ed6 LZ |
409 | |
410 | wait_event(root->fs_info->transaction_wait, | |
501407aa JB |
411 | cur_trans->state >= TRANS_STATE_UNBLOCKED || |
412 | cur_trans->aborted); | |
724e2315 | 413 | btrfs_put_transaction(cur_trans); |
a4abeea4 JB |
414 | } else { |
415 | spin_unlock(&root->fs_info->trans_lock); | |
f9295749 | 416 | } |
37d1aeee CM |
417 | } |
418 | ||
a22285a6 YZ |
419 | static int may_wait_transaction(struct btrfs_root *root, int type) |
420 | { | |
a4abeea4 JB |
421 | if (root->fs_info->log_root_recovering) |
422 | return 0; | |
423 | ||
424 | if (type == TRANS_USERSPACE) | |
425 | return 1; | |
426 | ||
427 | if (type == TRANS_START && | |
428 | !atomic_read(&root->fs_info->open_ioctl_trans)) | |
a22285a6 | 429 | return 1; |
a4abeea4 | 430 | |
a22285a6 YZ |
431 | return 0; |
432 | } | |
433 | ||
20dd2cbf MX |
434 | static inline bool need_reserve_reloc_root(struct btrfs_root *root) |
435 | { | |
436 | if (!root->fs_info->reloc_ctl || | |
27cdeb70 | 437 | !test_bit(BTRFS_ROOT_REF_COWS, &root->state) || |
20dd2cbf MX |
438 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || |
439 | root->reloc_root) | |
440 | return false; | |
441 | ||
442 | return true; | |
443 | } | |
444 | ||
08e007d2 | 445 | static struct btrfs_trans_handle * |
0860adfd | 446 | start_transaction(struct btrfs_root *root, u64 num_items, unsigned int type, |
08e007d2 | 447 | enum btrfs_reserve_flush_enum flush) |
37d1aeee | 448 | { |
a22285a6 YZ |
449 | struct btrfs_trans_handle *h; |
450 | struct btrfs_transaction *cur_trans; | |
b5009945 | 451 | u64 num_bytes = 0; |
c5567237 | 452 | u64 qgroup_reserved = 0; |
20dd2cbf MX |
453 | bool reloc_reserved = false; |
454 | int ret; | |
acce952b | 455 | |
46c4e71e | 456 | /* Send isn't supposed to start transactions. */ |
2755a0de | 457 | ASSERT(current->journal_info != BTRFS_SEND_TRANS_STUB); |
46c4e71e | 458 | |
87533c47 | 459 | if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) |
acce952b | 460 | return ERR_PTR(-EROFS); |
2a1eb461 | 461 | |
46c4e71e | 462 | if (current->journal_info) { |
0860adfd | 463 | WARN_ON(type & TRANS_EXTWRITERS); |
2a1eb461 JB |
464 | h = current->journal_info; |
465 | h->use_count++; | |
b7d5b0a8 | 466 | WARN_ON(h->use_count > 2); |
2a1eb461 JB |
467 | h->orig_rsv = h->block_rsv; |
468 | h->block_rsv = NULL; | |
469 | goto got_it; | |
470 | } | |
b5009945 JB |
471 | |
472 | /* | |
473 | * Do the reservation before we join the transaction so we can do all | |
474 | * the appropriate flushing if need be. | |
475 | */ | |
476 | if (num_items > 0 && root != root->fs_info->chunk_root) { | |
c5567237 AJ |
477 | if (root->fs_info->quota_enabled && |
478 | is_fstree(root->root_key.objectid)) { | |
707e8a07 | 479 | qgroup_reserved = num_items * root->nodesize; |
c5567237 AJ |
480 | ret = btrfs_qgroup_reserve(root, qgroup_reserved); |
481 | if (ret) | |
482 | return ERR_PTR(ret); | |
483 | } | |
484 | ||
b5009945 | 485 | num_bytes = btrfs_calc_trans_metadata_size(root, num_items); |
20dd2cbf MX |
486 | /* |
487 | * Do the reservation for the relocation root creation | |
488 | */ | |
ee39b432 | 489 | if (need_reserve_reloc_root(root)) { |
20dd2cbf MX |
490 | num_bytes += root->nodesize; |
491 | reloc_reserved = true; | |
492 | } | |
493 | ||
08e007d2 MX |
494 | ret = btrfs_block_rsv_add(root, |
495 | &root->fs_info->trans_block_rsv, | |
496 | num_bytes, flush); | |
b5009945 | 497 | if (ret) |
843fcf35 | 498 | goto reserve_fail; |
b5009945 | 499 | } |
a22285a6 | 500 | again: |
f2f767e7 | 501 | h = kmem_cache_zalloc(btrfs_trans_handle_cachep, GFP_NOFS); |
843fcf35 MX |
502 | if (!h) { |
503 | ret = -ENOMEM; | |
504 | goto alloc_fail; | |
505 | } | |
37d1aeee | 506 | |
98114659 JB |
507 | /* |
508 | * If we are JOIN_NOLOCK we're already committing a transaction and | |
509 | * waiting on this guy, so we don't need to do the sb_start_intwrite | |
510 | * because we're already holding a ref. We need this because we could | |
511 | * have raced in and did an fsync() on a file which can kick a commit | |
512 | * and then we deadlock with somebody doing a freeze. | |
354aa0fb MX |
513 | * |
514 | * If we are ATTACH, it means we just want to catch the current | |
515 | * transaction and commit it, so we needn't do sb_start_intwrite(). | |
98114659 | 516 | */ |
0860adfd | 517 | if (type & __TRANS_FREEZABLE) |
60376ce4 | 518 | sb_start_intwrite(root->fs_info->sb); |
b2b5ef5c | 519 | |
a22285a6 | 520 | if (may_wait_transaction(root, type)) |
37d1aeee | 521 | wait_current_trans(root); |
a22285a6 | 522 | |
a4abeea4 | 523 | do { |
354aa0fb | 524 | ret = join_transaction(root, type); |
178260b2 | 525 | if (ret == -EBUSY) { |
a4abeea4 | 526 | wait_current_trans(root); |
178260b2 MX |
527 | if (unlikely(type == TRANS_ATTACH)) |
528 | ret = -ENOENT; | |
529 | } | |
a4abeea4 JB |
530 | } while (ret == -EBUSY); |
531 | ||
db5b493a | 532 | if (ret < 0) { |
354aa0fb MX |
533 | /* We must get the transaction if we are JOIN_NOLOCK. */ |
534 | BUG_ON(type == TRANS_JOIN_NOLOCK); | |
843fcf35 | 535 | goto join_fail; |
db5b493a | 536 | } |
0f7d52f4 | 537 | |
a22285a6 | 538 | cur_trans = root->fs_info->running_transaction; |
a22285a6 YZ |
539 | |
540 | h->transid = cur_trans->transid; | |
541 | h->transaction = cur_trans; | |
d13603ef | 542 | h->root = root; |
2a1eb461 | 543 | h->use_count = 1; |
a698d075 | 544 | h->type = type; |
d9a0540a | 545 | h->can_flush_pending_bgs = true; |
bed92eae | 546 | INIT_LIST_HEAD(&h->qgroup_ref_list); |
ea658bad | 547 | INIT_LIST_HEAD(&h->new_bgs); |
50d9aa99 | 548 | INIT_LIST_HEAD(&h->ordered); |
b7ec40d7 | 549 | |
a22285a6 | 550 | smp_mb(); |
4a9d8bde MX |
551 | if (cur_trans->state >= TRANS_STATE_BLOCKED && |
552 | may_wait_transaction(root, type)) { | |
abdd2e80 | 553 | current->journal_info = h; |
a22285a6 YZ |
554 | btrfs_commit_transaction(h, root); |
555 | goto again; | |
556 | } | |
557 | ||
b5009945 | 558 | if (num_bytes) { |
8c2a3ca2 | 559 | trace_btrfs_space_reservation(root->fs_info, "transaction", |
2bcc0328 | 560 | h->transid, num_bytes, 1); |
b5009945 JB |
561 | h->block_rsv = &root->fs_info->trans_block_rsv; |
562 | h->bytes_reserved = num_bytes; | |
20dd2cbf | 563 | h->reloc_reserved = reloc_reserved; |
a22285a6 | 564 | } |
4b824906 | 565 | h->qgroup_reserved = qgroup_reserved; |
9ed74f2d | 566 | |
2a1eb461 | 567 | got_it: |
a4abeea4 | 568 | btrfs_record_root_in_trans(h, root); |
a22285a6 YZ |
569 | |
570 | if (!current->journal_info && type != TRANS_USERSPACE) | |
571 | current->journal_info = h; | |
79154b1b | 572 | return h; |
843fcf35 MX |
573 | |
574 | join_fail: | |
0860adfd | 575 | if (type & __TRANS_FREEZABLE) |
843fcf35 MX |
576 | sb_end_intwrite(root->fs_info->sb); |
577 | kmem_cache_free(btrfs_trans_handle_cachep, h); | |
578 | alloc_fail: | |
579 | if (num_bytes) | |
580 | btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv, | |
581 | num_bytes); | |
582 | reserve_fail: | |
583 | if (qgroup_reserved) | |
584 | btrfs_qgroup_free(root, qgroup_reserved); | |
585 | return ERR_PTR(ret); | |
79154b1b CM |
586 | } |
587 | ||
f9295749 | 588 | struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, |
a22285a6 | 589 | int num_items) |
f9295749 | 590 | { |
08e007d2 MX |
591 | return start_transaction(root, num_items, TRANS_START, |
592 | BTRFS_RESERVE_FLUSH_ALL); | |
f9295749 | 593 | } |
8407aa46 | 594 | |
08e007d2 | 595 | struct btrfs_trans_handle *btrfs_start_transaction_lflush( |
8407aa46 MX |
596 | struct btrfs_root *root, int num_items) |
597 | { | |
08e007d2 MX |
598 | return start_transaction(root, num_items, TRANS_START, |
599 | BTRFS_RESERVE_FLUSH_LIMIT); | |
8407aa46 MX |
600 | } |
601 | ||
7a7eaa40 | 602 | struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root) |
f9295749 | 603 | { |
8407aa46 | 604 | return start_transaction(root, 0, TRANS_JOIN, 0); |
f9295749 CM |
605 | } |
606 | ||
7a7eaa40 | 607 | struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root) |
0af3d00b | 608 | { |
8407aa46 | 609 | return start_transaction(root, 0, TRANS_JOIN_NOLOCK, 0); |
0af3d00b JB |
610 | } |
611 | ||
7a7eaa40 | 612 | struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root) |
9ca9ee09 | 613 | { |
8407aa46 | 614 | return start_transaction(root, 0, TRANS_USERSPACE, 0); |
9ca9ee09 SW |
615 | } |
616 | ||
d4edf39b MX |
617 | /* |
618 | * btrfs_attach_transaction() - catch the running transaction | |
619 | * | |
620 | * It is used when we want to commit the current the transaction, but | |
621 | * don't want to start a new one. | |
622 | * | |
623 | * Note: If this function return -ENOENT, it just means there is no | |
624 | * running transaction. But it is possible that the inactive transaction | |
625 | * is still in the memory, not fully on disk. If you hope there is no | |
626 | * inactive transaction in the fs when -ENOENT is returned, you should | |
627 | * invoke | |
628 | * btrfs_attach_transaction_barrier() | |
629 | */ | |
354aa0fb | 630 | struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root) |
60376ce4 | 631 | { |
354aa0fb | 632 | return start_transaction(root, 0, TRANS_ATTACH, 0); |
60376ce4 JB |
633 | } |
634 | ||
d4edf39b | 635 | /* |
90b6d283 | 636 | * btrfs_attach_transaction_barrier() - catch the running transaction |
d4edf39b MX |
637 | * |
638 | * It is similar to the above function, the differentia is this one | |
639 | * will wait for all the inactive transactions until they fully | |
640 | * complete. | |
641 | */ | |
642 | struct btrfs_trans_handle * | |
643 | btrfs_attach_transaction_barrier(struct btrfs_root *root) | |
644 | { | |
645 | struct btrfs_trans_handle *trans; | |
646 | ||
647 | trans = start_transaction(root, 0, TRANS_ATTACH, 0); | |
648 | if (IS_ERR(trans) && PTR_ERR(trans) == -ENOENT) | |
649 | btrfs_wait_for_commit(root, 0); | |
650 | ||
651 | return trans; | |
652 | } | |
653 | ||
d352ac68 | 654 | /* wait for a transaction commit to be fully complete */ |
b9c8300c | 655 | static noinline void wait_for_commit(struct btrfs_root *root, |
89ce8a63 CM |
656 | struct btrfs_transaction *commit) |
657 | { | |
4a9d8bde | 658 | wait_event(commit->commit_wait, commit->state == TRANS_STATE_COMPLETED); |
89ce8a63 CM |
659 | } |
660 | ||
46204592 SW |
661 | int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid) |
662 | { | |
663 | struct btrfs_transaction *cur_trans = NULL, *t; | |
8cd2807f | 664 | int ret = 0; |
46204592 | 665 | |
46204592 SW |
666 | if (transid) { |
667 | if (transid <= root->fs_info->last_trans_committed) | |
a4abeea4 | 668 | goto out; |
46204592 SW |
669 | |
670 | /* find specified transaction */ | |
a4abeea4 | 671 | spin_lock(&root->fs_info->trans_lock); |
46204592 SW |
672 | list_for_each_entry(t, &root->fs_info->trans_list, list) { |
673 | if (t->transid == transid) { | |
674 | cur_trans = t; | |
a4abeea4 | 675 | atomic_inc(&cur_trans->use_count); |
8cd2807f | 676 | ret = 0; |
46204592 SW |
677 | break; |
678 | } | |
8cd2807f MX |
679 | if (t->transid > transid) { |
680 | ret = 0; | |
46204592 | 681 | break; |
8cd2807f | 682 | } |
46204592 | 683 | } |
a4abeea4 | 684 | spin_unlock(&root->fs_info->trans_lock); |
42383020 SW |
685 | |
686 | /* | |
687 | * The specified transaction doesn't exist, or we | |
688 | * raced with btrfs_commit_transaction | |
689 | */ | |
690 | if (!cur_trans) { | |
691 | if (transid > root->fs_info->last_trans_committed) | |
692 | ret = -EINVAL; | |
8cd2807f | 693 | goto out; |
42383020 | 694 | } |
46204592 SW |
695 | } else { |
696 | /* find newest transaction that is committing | committed */ | |
a4abeea4 | 697 | spin_lock(&root->fs_info->trans_lock); |
46204592 SW |
698 | list_for_each_entry_reverse(t, &root->fs_info->trans_list, |
699 | list) { | |
4a9d8bde MX |
700 | if (t->state >= TRANS_STATE_COMMIT_START) { |
701 | if (t->state == TRANS_STATE_COMPLETED) | |
3473f3c0 | 702 | break; |
46204592 | 703 | cur_trans = t; |
a4abeea4 | 704 | atomic_inc(&cur_trans->use_count); |
46204592 SW |
705 | break; |
706 | } | |
707 | } | |
a4abeea4 | 708 | spin_unlock(&root->fs_info->trans_lock); |
46204592 | 709 | if (!cur_trans) |
a4abeea4 | 710 | goto out; /* nothing committing|committed */ |
46204592 SW |
711 | } |
712 | ||
46204592 | 713 | wait_for_commit(root, cur_trans); |
724e2315 | 714 | btrfs_put_transaction(cur_trans); |
a4abeea4 | 715 | out: |
46204592 SW |
716 | return ret; |
717 | } | |
718 | ||
37d1aeee CM |
719 | void btrfs_throttle(struct btrfs_root *root) |
720 | { | |
a4abeea4 | 721 | if (!atomic_read(&root->fs_info->open_ioctl_trans)) |
9ca9ee09 | 722 | wait_current_trans(root); |
37d1aeee CM |
723 | } |
724 | ||
8929ecfa YZ |
725 | static int should_end_transaction(struct btrfs_trans_handle *trans, |
726 | struct btrfs_root *root) | |
727 | { | |
1be41b78 | 728 | if (root->fs_info->global_block_rsv.space_info->full && |
0a2b2a84 | 729 | btrfs_check_space_for_delayed_refs(trans, root)) |
1be41b78 | 730 | return 1; |
36ba022a | 731 | |
1be41b78 | 732 | return !!btrfs_block_rsv_check(root, &root->fs_info->global_block_rsv, 5); |
8929ecfa YZ |
733 | } |
734 | ||
735 | int btrfs_should_end_transaction(struct btrfs_trans_handle *trans, | |
736 | struct btrfs_root *root) | |
737 | { | |
738 | struct btrfs_transaction *cur_trans = trans->transaction; | |
739 | int updates; | |
49b25e05 | 740 | int err; |
8929ecfa | 741 | |
a4abeea4 | 742 | smp_mb(); |
4a9d8bde MX |
743 | if (cur_trans->state >= TRANS_STATE_BLOCKED || |
744 | cur_trans->delayed_refs.flushing) | |
8929ecfa YZ |
745 | return 1; |
746 | ||
747 | updates = trans->delayed_ref_updates; | |
748 | trans->delayed_ref_updates = 0; | |
49b25e05 | 749 | if (updates) { |
28ed1345 | 750 | err = btrfs_run_delayed_refs(trans, root, updates * 2); |
49b25e05 JM |
751 | if (err) /* Error code will also eval true */ |
752 | return err; | |
753 | } | |
8929ecfa YZ |
754 | |
755 | return should_end_transaction(trans, root); | |
756 | } | |
757 | ||
89ce8a63 | 758 | static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, |
a698d075 | 759 | struct btrfs_root *root, int throttle) |
79154b1b | 760 | { |
8929ecfa | 761 | struct btrfs_transaction *cur_trans = trans->transaction; |
ab78c84d | 762 | struct btrfs_fs_info *info = root->fs_info; |
1be41b78 | 763 | unsigned long cur = trans->delayed_ref_updates; |
a698d075 | 764 | int lock = (trans->type != TRANS_JOIN_NOLOCK); |
4edc2ca3 | 765 | int err = 0; |
a79b7d4b | 766 | int must_run_delayed_refs = 0; |
c3e69d58 | 767 | |
3bbb24b2 JB |
768 | if (trans->use_count > 1) { |
769 | trans->use_count--; | |
2a1eb461 JB |
770 | trans->block_rsv = trans->orig_rsv; |
771 | return 0; | |
772 | } | |
773 | ||
b24e03db | 774 | btrfs_trans_release_metadata(trans, root); |
4c13d758 | 775 | trans->block_rsv = NULL; |
c5567237 | 776 | |
ea658bad JB |
777 | if (!list_empty(&trans->new_bgs)) |
778 | btrfs_create_pending_block_groups(trans, root); | |
779 | ||
50d9aa99 JB |
780 | if (!list_empty(&trans->ordered)) { |
781 | spin_lock(&info->trans_lock); | |
d3efe084 | 782 | list_splice_init(&trans->ordered, &cur_trans->pending_ordered); |
50d9aa99 JB |
783 | spin_unlock(&info->trans_lock); |
784 | } | |
785 | ||
1be41b78 | 786 | trans->delayed_ref_updates = 0; |
a79b7d4b CM |
787 | if (!trans->sync) { |
788 | must_run_delayed_refs = | |
789 | btrfs_should_throttle_delayed_refs(trans, root); | |
0a2b2a84 | 790 | cur = max_t(unsigned long, cur, 32); |
a79b7d4b CM |
791 | |
792 | /* | |
793 | * don't make the caller wait if they are from a NOLOCK | |
794 | * or ATTACH transaction, it will deadlock with commit | |
795 | */ | |
796 | if (must_run_delayed_refs == 1 && | |
797 | (trans->type & (__TRANS_JOIN_NOLOCK | __TRANS_ATTACH))) | |
798 | must_run_delayed_refs = 2; | |
56bec294 | 799 | } |
bb721703 | 800 | |
fcebe456 JB |
801 | if (trans->qgroup_reserved) { |
802 | /* | |
803 | * the same root has to be passed here between start_transaction | |
804 | * and end_transaction. Subvolume quota depends on this. | |
805 | */ | |
806 | btrfs_qgroup_free(trans->root, trans->qgroup_reserved); | |
807 | trans->qgroup_reserved = 0; | |
808 | } | |
809 | ||
0e721106 JB |
810 | btrfs_trans_release_metadata(trans, root); |
811 | trans->block_rsv = NULL; | |
56bec294 | 812 | |
ea658bad JB |
813 | if (!list_empty(&trans->new_bgs)) |
814 | btrfs_create_pending_block_groups(trans, root); | |
815 | ||
4fbcdf66 FM |
816 | btrfs_trans_release_chunk_metadata(trans); |
817 | ||
a4abeea4 | 818 | if (lock && !atomic_read(&root->fs_info->open_ioctl_trans) && |
4a9d8bde MX |
819 | should_end_transaction(trans, root) && |
820 | ACCESS_ONCE(cur_trans->state) == TRANS_STATE_RUNNING) { | |
821 | spin_lock(&info->trans_lock); | |
822 | if (cur_trans->state == TRANS_STATE_RUNNING) | |
823 | cur_trans->state = TRANS_STATE_BLOCKED; | |
824 | spin_unlock(&info->trans_lock); | |
a4abeea4 | 825 | } |
8929ecfa | 826 | |
4a9d8bde | 827 | if (lock && ACCESS_ONCE(cur_trans->state) == TRANS_STATE_BLOCKED) { |
3bbb24b2 | 828 | if (throttle) |
8929ecfa | 829 | return btrfs_commit_transaction(trans, root); |
3bbb24b2 | 830 | else |
8929ecfa YZ |
831 | wake_up_process(info->transaction_kthread); |
832 | } | |
833 | ||
0860adfd | 834 | if (trans->type & __TRANS_FREEZABLE) |
98114659 | 835 | sb_end_intwrite(root->fs_info->sb); |
6df7881a | 836 | |
8929ecfa | 837 | WARN_ON(cur_trans != info->running_transaction); |
13c5a93e JB |
838 | WARN_ON(atomic_read(&cur_trans->num_writers) < 1); |
839 | atomic_dec(&cur_trans->num_writers); | |
0860adfd | 840 | extwriter_counter_dec(cur_trans, trans->type); |
89ce8a63 | 841 | |
99d16cbc | 842 | smp_mb(); |
79154b1b CM |
843 | if (waitqueue_active(&cur_trans->writer_wait)) |
844 | wake_up(&cur_trans->writer_wait); | |
724e2315 | 845 | btrfs_put_transaction(cur_trans); |
9ed74f2d JB |
846 | |
847 | if (current->journal_info == trans) | |
848 | current->journal_info = NULL; | |
ab78c84d | 849 | |
24bbcf04 YZ |
850 | if (throttle) |
851 | btrfs_run_delayed_iputs(root); | |
852 | ||
49b25e05 | 853 | if (trans->aborted || |
4e121c06 JB |
854 | test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) { |
855 | wake_up_process(info->transaction_kthread); | |
4edc2ca3 | 856 | err = -EIO; |
4e121c06 | 857 | } |
edf39272 | 858 | assert_qgroups_uptodate(trans); |
49b25e05 | 859 | |
4edc2ca3 | 860 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
a79b7d4b CM |
861 | if (must_run_delayed_refs) { |
862 | btrfs_async_run_delayed_refs(root, cur, | |
863 | must_run_delayed_refs == 1); | |
864 | } | |
4edc2ca3 | 865 | return err; |
79154b1b CM |
866 | } |
867 | ||
89ce8a63 CM |
868 | int btrfs_end_transaction(struct btrfs_trans_handle *trans, |
869 | struct btrfs_root *root) | |
870 | { | |
98ad43be | 871 | return __btrfs_end_transaction(trans, root, 0); |
89ce8a63 CM |
872 | } |
873 | ||
874 | int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans, | |
875 | struct btrfs_root *root) | |
876 | { | |
98ad43be | 877 | return __btrfs_end_transaction(trans, root, 1); |
16cdcec7 MX |
878 | } |
879 | ||
d352ac68 CM |
880 | /* |
881 | * when btree blocks are allocated, they have some corresponding bits set for | |
882 | * them in one of two extent_io trees. This is used to make sure all of | |
690587d1 | 883 | * those extents are sent to disk but does not wait on them |
d352ac68 | 884 | */ |
690587d1 | 885 | int btrfs_write_marked_extents(struct btrfs_root *root, |
8cef4e16 | 886 | struct extent_io_tree *dirty_pages, int mark) |
79154b1b | 887 | { |
777e6bd7 | 888 | int err = 0; |
7c4452b9 | 889 | int werr = 0; |
1728366e | 890 | struct address_space *mapping = root->fs_info->btree_inode->i_mapping; |
e6138876 | 891 | struct extent_state *cached_state = NULL; |
777e6bd7 | 892 | u64 start = 0; |
5f39d397 | 893 | u64 end; |
7c4452b9 | 894 | |
1728366e | 895 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
e6138876 | 896 | mark, &cached_state)) { |
663dfbb0 FM |
897 | bool wait_writeback = false; |
898 | ||
899 | err = convert_extent_bit(dirty_pages, start, end, | |
900 | EXTENT_NEED_WAIT, | |
901 | mark, &cached_state, GFP_NOFS); | |
902 | /* | |
903 | * convert_extent_bit can return -ENOMEM, which is most of the | |
904 | * time a temporary error. So when it happens, ignore the error | |
905 | * and wait for writeback of this range to finish - because we | |
906 | * failed to set the bit EXTENT_NEED_WAIT for the range, a call | |
907 | * to btrfs_wait_marked_extents() would not know that writeback | |
908 | * for this range started and therefore wouldn't wait for it to | |
909 | * finish - we don't want to commit a superblock that points to | |
910 | * btree nodes/leafs for which writeback hasn't finished yet | |
911 | * (and without errors). | |
912 | * We cleanup any entries left in the io tree when committing | |
913 | * the transaction (through clear_btree_io_tree()). | |
914 | */ | |
915 | if (err == -ENOMEM) { | |
916 | err = 0; | |
917 | wait_writeback = true; | |
918 | } | |
919 | if (!err) | |
920 | err = filemap_fdatawrite_range(mapping, start, end); | |
1728366e JB |
921 | if (err) |
922 | werr = err; | |
663dfbb0 FM |
923 | else if (wait_writeback) |
924 | werr = filemap_fdatawait_range(mapping, start, end); | |
e38e2ed7 | 925 | free_extent_state(cached_state); |
663dfbb0 | 926 | cached_state = NULL; |
1728366e JB |
927 | cond_resched(); |
928 | start = end + 1; | |
7c4452b9 | 929 | } |
690587d1 CM |
930 | return werr; |
931 | } | |
932 | ||
933 | /* | |
934 | * when btree blocks are allocated, they have some corresponding bits set for | |
935 | * them in one of two extent_io trees. This is used to make sure all of | |
936 | * those extents are on disk for transaction or log commit. We wait | |
937 | * on all the pages and clear them from the dirty pages state tree | |
938 | */ | |
939 | int btrfs_wait_marked_extents(struct btrfs_root *root, | |
8cef4e16 | 940 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 | 941 | { |
690587d1 CM |
942 | int err = 0; |
943 | int werr = 0; | |
1728366e | 944 | struct address_space *mapping = root->fs_info->btree_inode->i_mapping; |
e6138876 | 945 | struct extent_state *cached_state = NULL; |
690587d1 CM |
946 | u64 start = 0; |
947 | u64 end; | |
656f30db FM |
948 | struct btrfs_inode *btree_ino = BTRFS_I(root->fs_info->btree_inode); |
949 | bool errors = false; | |
777e6bd7 | 950 | |
1728366e | 951 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
e6138876 | 952 | EXTENT_NEED_WAIT, &cached_state)) { |
663dfbb0 FM |
953 | /* |
954 | * Ignore -ENOMEM errors returned by clear_extent_bit(). | |
955 | * When committing the transaction, we'll remove any entries | |
956 | * left in the io tree. For a log commit, we don't remove them | |
957 | * after committing the log because the tree can be accessed | |
958 | * concurrently - we do it only at transaction commit time when | |
959 | * it's safe to do it (through clear_btree_io_tree()). | |
960 | */ | |
961 | err = clear_extent_bit(dirty_pages, start, end, | |
962 | EXTENT_NEED_WAIT, | |
963 | 0, 0, &cached_state, GFP_NOFS); | |
964 | if (err == -ENOMEM) | |
965 | err = 0; | |
966 | if (!err) | |
967 | err = filemap_fdatawait_range(mapping, start, end); | |
1728366e JB |
968 | if (err) |
969 | werr = err; | |
e38e2ed7 FM |
970 | free_extent_state(cached_state); |
971 | cached_state = NULL; | |
1728366e JB |
972 | cond_resched(); |
973 | start = end + 1; | |
777e6bd7 | 974 | } |
7c4452b9 CM |
975 | if (err) |
976 | werr = err; | |
656f30db FM |
977 | |
978 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { | |
979 | if ((mark & EXTENT_DIRTY) && | |
980 | test_and_clear_bit(BTRFS_INODE_BTREE_LOG1_ERR, | |
981 | &btree_ino->runtime_flags)) | |
982 | errors = true; | |
983 | ||
984 | if ((mark & EXTENT_NEW) && | |
985 | test_and_clear_bit(BTRFS_INODE_BTREE_LOG2_ERR, | |
986 | &btree_ino->runtime_flags)) | |
987 | errors = true; | |
988 | } else { | |
989 | if (test_and_clear_bit(BTRFS_INODE_BTREE_ERR, | |
990 | &btree_ino->runtime_flags)) | |
991 | errors = true; | |
992 | } | |
993 | ||
994 | if (errors && !werr) | |
995 | werr = -EIO; | |
996 | ||
7c4452b9 | 997 | return werr; |
79154b1b CM |
998 | } |
999 | ||
690587d1 CM |
1000 | /* |
1001 | * when btree blocks are allocated, they have some corresponding bits set for | |
1002 | * them in one of two extent_io trees. This is used to make sure all of | |
1003 | * those extents are on disk for transaction or log commit | |
1004 | */ | |
171170c1 | 1005 | static int btrfs_write_and_wait_marked_extents(struct btrfs_root *root, |
8cef4e16 | 1006 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 CM |
1007 | { |
1008 | int ret; | |
1009 | int ret2; | |
c6adc9cc | 1010 | struct blk_plug plug; |
690587d1 | 1011 | |
c6adc9cc | 1012 | blk_start_plug(&plug); |
8cef4e16 | 1013 | ret = btrfs_write_marked_extents(root, dirty_pages, mark); |
c6adc9cc | 1014 | blk_finish_plug(&plug); |
8cef4e16 | 1015 | ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark); |
bf0da8c1 CM |
1016 | |
1017 | if (ret) | |
1018 | return ret; | |
1019 | if (ret2) | |
1020 | return ret2; | |
1021 | return 0; | |
690587d1 CM |
1022 | } |
1023 | ||
663dfbb0 | 1024 | static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans, |
d0c803c4 CM |
1025 | struct btrfs_root *root) |
1026 | { | |
663dfbb0 FM |
1027 | int ret; |
1028 | ||
1029 | ret = btrfs_write_and_wait_marked_extents(root, | |
8cef4e16 YZ |
1030 | &trans->transaction->dirty_pages, |
1031 | EXTENT_DIRTY); | |
663dfbb0 FM |
1032 | clear_btree_io_tree(&trans->transaction->dirty_pages); |
1033 | ||
1034 | return ret; | |
d0c803c4 CM |
1035 | } |
1036 | ||
d352ac68 CM |
1037 | /* |
1038 | * this is used to update the root pointer in the tree of tree roots. | |
1039 | * | |
1040 | * But, in the case of the extent allocation tree, updating the root | |
1041 | * pointer may allocate blocks which may change the root of the extent | |
1042 | * allocation tree. | |
1043 | * | |
1044 | * So, this loops and repeats and makes sure the cowonly root didn't | |
1045 | * change while the root pointer was being updated in the metadata. | |
1046 | */ | |
0b86a832 CM |
1047 | static int update_cowonly_root(struct btrfs_trans_handle *trans, |
1048 | struct btrfs_root *root) | |
79154b1b CM |
1049 | { |
1050 | int ret; | |
0b86a832 | 1051 | u64 old_root_bytenr; |
86b9f2ec | 1052 | u64 old_root_used; |
0b86a832 | 1053 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
79154b1b | 1054 | |
86b9f2ec | 1055 | old_root_used = btrfs_root_used(&root->root_item); |
56bec294 | 1056 | |
d397712b | 1057 | while (1) { |
0b86a832 | 1058 | old_root_bytenr = btrfs_root_bytenr(&root->root_item); |
86b9f2ec | 1059 | if (old_root_bytenr == root->node->start && |
ea526d18 | 1060 | old_root_used == btrfs_root_used(&root->root_item)) |
79154b1b | 1061 | break; |
87ef2bb4 | 1062 | |
5d4f98a2 | 1063 | btrfs_set_root_node(&root->root_item, root->node); |
79154b1b | 1064 | ret = btrfs_update_root(trans, tree_root, |
0b86a832 CM |
1065 | &root->root_key, |
1066 | &root->root_item); | |
49b25e05 JM |
1067 | if (ret) |
1068 | return ret; | |
56bec294 | 1069 | |
86b9f2ec | 1070 | old_root_used = btrfs_root_used(&root->root_item); |
0b86a832 | 1071 | } |
276e680d | 1072 | |
0b86a832 CM |
1073 | return 0; |
1074 | } | |
1075 | ||
d352ac68 CM |
1076 | /* |
1077 | * update all the cowonly tree roots on disk | |
49b25e05 JM |
1078 | * |
1079 | * The error handling in this function may not be obvious. Any of the | |
1080 | * failures will cause the file system to go offline. We still need | |
1081 | * to clean up the delayed refs. | |
d352ac68 | 1082 | */ |
5d4f98a2 YZ |
1083 | static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans, |
1084 | struct btrfs_root *root) | |
0b86a832 CM |
1085 | { |
1086 | struct btrfs_fs_info *fs_info = root->fs_info; | |
ea526d18 | 1087 | struct list_head *dirty_bgs = &trans->transaction->dirty_bgs; |
1bbc621e | 1088 | struct list_head *io_bgs = &trans->transaction->io_bgs; |
0b86a832 | 1089 | struct list_head *next; |
84234f3a | 1090 | struct extent_buffer *eb; |
56bec294 | 1091 | int ret; |
84234f3a YZ |
1092 | |
1093 | eb = btrfs_lock_root_node(fs_info->tree_root); | |
49b25e05 JM |
1094 | ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, |
1095 | 0, &eb); | |
84234f3a YZ |
1096 | btrfs_tree_unlock(eb); |
1097 | free_extent_buffer(eb); | |
0b86a832 | 1098 | |
49b25e05 JM |
1099 | if (ret) |
1100 | return ret; | |
1101 | ||
56bec294 | 1102 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
49b25e05 JM |
1103 | if (ret) |
1104 | return ret; | |
87ef2bb4 | 1105 | |
733f4fbb | 1106 | ret = btrfs_run_dev_stats(trans, root->fs_info); |
c16ce190 JB |
1107 | if (ret) |
1108 | return ret; | |
8dabb742 | 1109 | ret = btrfs_run_dev_replace(trans, root->fs_info); |
c16ce190 JB |
1110 | if (ret) |
1111 | return ret; | |
546adb0d | 1112 | ret = btrfs_run_qgroups(trans, root->fs_info); |
c16ce190 JB |
1113 | if (ret) |
1114 | return ret; | |
546adb0d | 1115 | |
dcdf7f6d JB |
1116 | ret = btrfs_setup_space_cache(trans, root); |
1117 | if (ret) | |
1118 | return ret; | |
1119 | ||
546adb0d JS |
1120 | /* run_qgroups might have added some more refs */ |
1121 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); | |
c16ce190 JB |
1122 | if (ret) |
1123 | return ret; | |
ea526d18 | 1124 | again: |
d397712b | 1125 | while (!list_empty(&fs_info->dirty_cowonly_roots)) { |
0b86a832 CM |
1126 | next = fs_info->dirty_cowonly_roots.next; |
1127 | list_del_init(next); | |
1128 | root = list_entry(next, struct btrfs_root, dirty_list); | |
e7070be1 | 1129 | clear_bit(BTRFS_ROOT_DIRTY, &root->state); |
87ef2bb4 | 1130 | |
9e351cc8 JB |
1131 | if (root != fs_info->extent_root) |
1132 | list_add_tail(&root->dirty_list, | |
1133 | &trans->transaction->switch_commits); | |
49b25e05 JM |
1134 | ret = update_cowonly_root(trans, root); |
1135 | if (ret) | |
1136 | return ret; | |
ea526d18 JB |
1137 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
1138 | if (ret) | |
1139 | return ret; | |
79154b1b | 1140 | } |
276e680d | 1141 | |
1bbc621e | 1142 | while (!list_empty(dirty_bgs) || !list_empty(io_bgs)) { |
ea526d18 JB |
1143 | ret = btrfs_write_dirty_block_groups(trans, root); |
1144 | if (ret) | |
1145 | return ret; | |
1146 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); | |
1147 | if (ret) | |
1148 | return ret; | |
1149 | } | |
1150 | ||
1151 | if (!list_empty(&fs_info->dirty_cowonly_roots)) | |
1152 | goto again; | |
1153 | ||
9e351cc8 JB |
1154 | list_add_tail(&fs_info->extent_root->dirty_list, |
1155 | &trans->transaction->switch_commits); | |
8dabb742 SB |
1156 | btrfs_after_dev_replace_commit(fs_info); |
1157 | ||
79154b1b CM |
1158 | return 0; |
1159 | } | |
1160 | ||
d352ac68 CM |
1161 | /* |
1162 | * dead roots are old snapshots that need to be deleted. This allocates | |
1163 | * a dirty root struct and adds it into the list of dead roots that need to | |
1164 | * be deleted | |
1165 | */ | |
cfad392b | 1166 | void btrfs_add_dead_root(struct btrfs_root *root) |
5eda7b5e | 1167 | { |
a4abeea4 | 1168 | spin_lock(&root->fs_info->trans_lock); |
cfad392b JB |
1169 | if (list_empty(&root->root_list)) |
1170 | list_add_tail(&root->root_list, &root->fs_info->dead_roots); | |
a4abeea4 | 1171 | spin_unlock(&root->fs_info->trans_lock); |
5eda7b5e CM |
1172 | } |
1173 | ||
d352ac68 | 1174 | /* |
5d4f98a2 | 1175 | * update all the cowonly tree roots on disk |
d352ac68 | 1176 | */ |
5d4f98a2 YZ |
1177 | static noinline int commit_fs_roots(struct btrfs_trans_handle *trans, |
1178 | struct btrfs_root *root) | |
0f7d52f4 | 1179 | { |
0f7d52f4 | 1180 | struct btrfs_root *gang[8]; |
5d4f98a2 | 1181 | struct btrfs_fs_info *fs_info = root->fs_info; |
0f7d52f4 CM |
1182 | int i; |
1183 | int ret; | |
54aa1f4d CM |
1184 | int err = 0; |
1185 | ||
a4abeea4 | 1186 | spin_lock(&fs_info->fs_roots_radix_lock); |
d397712b | 1187 | while (1) { |
5d4f98a2 YZ |
1188 | ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, |
1189 | (void **)gang, 0, | |
0f7d52f4 CM |
1190 | ARRAY_SIZE(gang), |
1191 | BTRFS_ROOT_TRANS_TAG); | |
1192 | if (ret == 0) | |
1193 | break; | |
1194 | for (i = 0; i < ret; i++) { | |
1195 | root = gang[i]; | |
5d4f98a2 YZ |
1196 | radix_tree_tag_clear(&fs_info->fs_roots_radix, |
1197 | (unsigned long)root->root_key.objectid, | |
1198 | BTRFS_ROOT_TRANS_TAG); | |
a4abeea4 | 1199 | spin_unlock(&fs_info->fs_roots_radix_lock); |
31153d81 | 1200 | |
e02119d5 | 1201 | btrfs_free_log(trans, root); |
5d4f98a2 | 1202 | btrfs_update_reloc_root(trans, root); |
d68fc57b | 1203 | btrfs_orphan_commit_root(trans, root); |
bcc63abb | 1204 | |
82d5902d LZ |
1205 | btrfs_save_ino_cache(root, trans); |
1206 | ||
f1ebcc74 | 1207 | /* see comments in should_cow_block() */ |
27cdeb70 | 1208 | clear_bit(BTRFS_ROOT_FORCE_COW, &root->state); |
c7548af6 | 1209 | smp_mb__after_atomic(); |
f1ebcc74 | 1210 | |
978d910d | 1211 | if (root->commit_root != root->node) { |
9e351cc8 JB |
1212 | list_add_tail(&root->dirty_list, |
1213 | &trans->transaction->switch_commits); | |
978d910d YZ |
1214 | btrfs_set_root_node(&root->root_item, |
1215 | root->node); | |
1216 | } | |
5d4f98a2 | 1217 | |
5d4f98a2 | 1218 | err = btrfs_update_root(trans, fs_info->tree_root, |
0f7d52f4 CM |
1219 | &root->root_key, |
1220 | &root->root_item); | |
a4abeea4 | 1221 | spin_lock(&fs_info->fs_roots_radix_lock); |
54aa1f4d CM |
1222 | if (err) |
1223 | break; | |
0f7d52f4 CM |
1224 | } |
1225 | } | |
a4abeea4 | 1226 | spin_unlock(&fs_info->fs_roots_radix_lock); |
54aa1f4d | 1227 | return err; |
0f7d52f4 CM |
1228 | } |
1229 | ||
d352ac68 | 1230 | /* |
de78b51a ES |
1231 | * defrag a given btree. |
1232 | * Every leaf in the btree is read and defragged. | |
d352ac68 | 1233 | */ |
de78b51a | 1234 | int btrfs_defrag_root(struct btrfs_root *root) |
e9d0b13b CM |
1235 | { |
1236 | struct btrfs_fs_info *info = root->fs_info; | |
e9d0b13b | 1237 | struct btrfs_trans_handle *trans; |
8929ecfa | 1238 | int ret; |
e9d0b13b | 1239 | |
27cdeb70 | 1240 | if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state)) |
e9d0b13b | 1241 | return 0; |
8929ecfa | 1242 | |
6b80053d | 1243 | while (1) { |
8929ecfa YZ |
1244 | trans = btrfs_start_transaction(root, 0); |
1245 | if (IS_ERR(trans)) | |
1246 | return PTR_ERR(trans); | |
1247 | ||
de78b51a | 1248 | ret = btrfs_defrag_leaves(trans, root); |
8929ecfa | 1249 | |
e9d0b13b | 1250 | btrfs_end_transaction(trans, root); |
b53d3f5d | 1251 | btrfs_btree_balance_dirty(info->tree_root); |
e9d0b13b CM |
1252 | cond_resched(); |
1253 | ||
7841cb28 | 1254 | if (btrfs_fs_closing(root->fs_info) || ret != -EAGAIN) |
e9d0b13b | 1255 | break; |
210549eb DS |
1256 | |
1257 | if (btrfs_defrag_cancelled(root->fs_info)) { | |
efe120a0 | 1258 | pr_debug("BTRFS: defrag_root cancelled\n"); |
210549eb DS |
1259 | ret = -EAGAIN; |
1260 | break; | |
1261 | } | |
e9d0b13b | 1262 | } |
27cdeb70 | 1263 | clear_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state); |
8929ecfa | 1264 | return ret; |
e9d0b13b CM |
1265 | } |
1266 | ||
d352ac68 CM |
1267 | /* |
1268 | * new snapshots need to be created at a very specific time in the | |
aec8030a MX |
1269 | * transaction commit. This does the actual creation. |
1270 | * | |
1271 | * Note: | |
1272 | * If the error which may affect the commitment of the current transaction | |
1273 | * happens, we should return the error number. If the error which just affect | |
1274 | * the creation of the pending snapshots, just return 0. | |
d352ac68 | 1275 | */ |
80b6794d | 1276 | static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, |
3063d29f CM |
1277 | struct btrfs_fs_info *fs_info, |
1278 | struct btrfs_pending_snapshot *pending) | |
1279 | { | |
1280 | struct btrfs_key key; | |
80b6794d | 1281 | struct btrfs_root_item *new_root_item; |
3063d29f CM |
1282 | struct btrfs_root *tree_root = fs_info->tree_root; |
1283 | struct btrfs_root *root = pending->root; | |
6bdb72de | 1284 | struct btrfs_root *parent_root; |
98c9942a | 1285 | struct btrfs_block_rsv *rsv; |
6bdb72de | 1286 | struct inode *parent_inode; |
42874b3d MX |
1287 | struct btrfs_path *path; |
1288 | struct btrfs_dir_item *dir_item; | |
a22285a6 | 1289 | struct dentry *dentry; |
3063d29f | 1290 | struct extent_buffer *tmp; |
925baedd | 1291 | struct extent_buffer *old; |
8ea05e3a | 1292 | struct timespec cur_time = CURRENT_TIME; |
aec8030a | 1293 | int ret = 0; |
d68fc57b | 1294 | u64 to_reserve = 0; |
6bdb72de | 1295 | u64 index = 0; |
a22285a6 | 1296 | u64 objectid; |
b83cc969 | 1297 | u64 root_flags; |
8ea05e3a | 1298 | uuid_le new_uuid; |
3063d29f | 1299 | |
42874b3d MX |
1300 | path = btrfs_alloc_path(); |
1301 | if (!path) { | |
aec8030a MX |
1302 | pending->error = -ENOMEM; |
1303 | return 0; | |
42874b3d MX |
1304 | } |
1305 | ||
80b6794d CM |
1306 | new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS); |
1307 | if (!new_root_item) { | |
aec8030a | 1308 | pending->error = -ENOMEM; |
6fa9700e | 1309 | goto root_item_alloc_fail; |
80b6794d | 1310 | } |
a22285a6 | 1311 | |
aec8030a MX |
1312 | pending->error = btrfs_find_free_objectid(tree_root, &objectid); |
1313 | if (pending->error) | |
6fa9700e | 1314 | goto no_free_objectid; |
3063d29f | 1315 | |
d6726335 QW |
1316 | /* |
1317 | * Make qgroup to skip current new snapshot's qgroupid, as it is | |
1318 | * accounted by later btrfs_qgroup_inherit(). | |
1319 | */ | |
1320 | btrfs_set_skip_qgroup(trans, objectid); | |
1321 | ||
147d256e | 1322 | btrfs_reloc_pre_snapshot(pending, &to_reserve); |
d68fc57b YZ |
1323 | |
1324 | if (to_reserve > 0) { | |
aec8030a MX |
1325 | pending->error = btrfs_block_rsv_add(root, |
1326 | &pending->block_rsv, | |
1327 | to_reserve, | |
1328 | BTRFS_RESERVE_NO_FLUSH); | |
1329 | if (pending->error) | |
d6726335 | 1330 | goto clear_skip_qgroup; |
d68fc57b YZ |
1331 | } |
1332 | ||
3063d29f | 1333 | key.objectid = objectid; |
a22285a6 YZ |
1334 | key.offset = (u64)-1; |
1335 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3063d29f | 1336 | |
6fa9700e | 1337 | rsv = trans->block_rsv; |
a22285a6 | 1338 | trans->block_rsv = &pending->block_rsv; |
2382c5cc | 1339 | trans->bytes_reserved = trans->block_rsv->reserved; |
3de4586c | 1340 | |
a22285a6 | 1341 | dentry = pending->dentry; |
e9662f70 | 1342 | parent_inode = pending->dir; |
a22285a6 | 1343 | parent_root = BTRFS_I(parent_inode)->root; |
7585717f | 1344 | record_root_in_trans(trans, parent_root); |
a22285a6 | 1345 | |
3063d29f CM |
1346 | /* |
1347 | * insert the directory item | |
1348 | */ | |
3de4586c | 1349 | ret = btrfs_set_inode_index(parent_inode, &index); |
49b25e05 | 1350 | BUG_ON(ret); /* -ENOMEM */ |
42874b3d MX |
1351 | |
1352 | /* check if there is a file/dir which has the same name. */ | |
1353 | dir_item = btrfs_lookup_dir_item(NULL, parent_root, path, | |
1354 | btrfs_ino(parent_inode), | |
1355 | dentry->d_name.name, | |
1356 | dentry->d_name.len, 0); | |
1357 | if (dir_item != NULL && !IS_ERR(dir_item)) { | |
fe66a05a | 1358 | pending->error = -EEXIST; |
aec8030a | 1359 | goto dir_item_existed; |
42874b3d MX |
1360 | } else if (IS_ERR(dir_item)) { |
1361 | ret = PTR_ERR(dir_item); | |
8732d44f MX |
1362 | btrfs_abort_transaction(trans, root, ret); |
1363 | goto fail; | |
79787eaa | 1364 | } |
42874b3d | 1365 | btrfs_release_path(path); |
52c26179 | 1366 | |
e999376f CM |
1367 | /* |
1368 | * pull in the delayed directory update | |
1369 | * and the delayed inode item | |
1370 | * otherwise we corrupt the FS during | |
1371 | * snapshot | |
1372 | */ | |
1373 | ret = btrfs_run_delayed_items(trans, root); | |
8732d44f MX |
1374 | if (ret) { /* Transaction aborted */ |
1375 | btrfs_abort_transaction(trans, root, ret); | |
1376 | goto fail; | |
1377 | } | |
e999376f | 1378 | |
7585717f | 1379 | record_root_in_trans(trans, root); |
6bdb72de SW |
1380 | btrfs_set_root_last_snapshot(&root->root_item, trans->transid); |
1381 | memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); | |
08fe4db1 | 1382 | btrfs_check_and_init_root_item(new_root_item); |
6bdb72de | 1383 | |
b83cc969 LZ |
1384 | root_flags = btrfs_root_flags(new_root_item); |
1385 | if (pending->readonly) | |
1386 | root_flags |= BTRFS_ROOT_SUBVOL_RDONLY; | |
1387 | else | |
1388 | root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY; | |
1389 | btrfs_set_root_flags(new_root_item, root_flags); | |
1390 | ||
8ea05e3a AB |
1391 | btrfs_set_root_generation_v2(new_root_item, |
1392 | trans->transid); | |
1393 | uuid_le_gen(&new_uuid); | |
1394 | memcpy(new_root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE); | |
1395 | memcpy(new_root_item->parent_uuid, root->root_item.uuid, | |
1396 | BTRFS_UUID_SIZE); | |
70023da2 SB |
1397 | if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) { |
1398 | memset(new_root_item->received_uuid, 0, | |
1399 | sizeof(new_root_item->received_uuid)); | |
1400 | memset(&new_root_item->stime, 0, sizeof(new_root_item->stime)); | |
1401 | memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime)); | |
1402 | btrfs_set_root_stransid(new_root_item, 0); | |
1403 | btrfs_set_root_rtransid(new_root_item, 0); | |
1404 | } | |
3cae210f QW |
1405 | btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec); |
1406 | btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec); | |
8ea05e3a | 1407 | btrfs_set_root_otransid(new_root_item, trans->transid); |
8ea05e3a | 1408 | |
6bdb72de | 1409 | old = btrfs_lock_root_node(root); |
49b25e05 | 1410 | ret = btrfs_cow_block(trans, root, old, NULL, 0, &old); |
79787eaa JM |
1411 | if (ret) { |
1412 | btrfs_tree_unlock(old); | |
1413 | free_extent_buffer(old); | |
8732d44f MX |
1414 | btrfs_abort_transaction(trans, root, ret); |
1415 | goto fail; | |
79787eaa | 1416 | } |
49b25e05 | 1417 | |
6bdb72de SW |
1418 | btrfs_set_lock_blocking(old); |
1419 | ||
49b25e05 | 1420 | ret = btrfs_copy_root(trans, root, old, &tmp, objectid); |
79787eaa | 1421 | /* clean up in any case */ |
6bdb72de SW |
1422 | btrfs_tree_unlock(old); |
1423 | free_extent_buffer(old); | |
8732d44f MX |
1424 | if (ret) { |
1425 | btrfs_abort_transaction(trans, root, ret); | |
1426 | goto fail; | |
1427 | } | |
f1ebcc74 | 1428 | /* see comments in should_cow_block() */ |
27cdeb70 | 1429 | set_bit(BTRFS_ROOT_FORCE_COW, &root->state); |
f1ebcc74 LB |
1430 | smp_wmb(); |
1431 | ||
6bdb72de | 1432 | btrfs_set_root_node(new_root_item, tmp); |
a22285a6 YZ |
1433 | /* record when the snapshot was created in key.offset */ |
1434 | key.offset = trans->transid; | |
1435 | ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); | |
6bdb72de SW |
1436 | btrfs_tree_unlock(tmp); |
1437 | free_extent_buffer(tmp); | |
8732d44f MX |
1438 | if (ret) { |
1439 | btrfs_abort_transaction(trans, root, ret); | |
1440 | goto fail; | |
1441 | } | |
6bdb72de | 1442 | |
a22285a6 YZ |
1443 | /* |
1444 | * insert root back/forward references | |
1445 | */ | |
1446 | ret = btrfs_add_root_ref(trans, tree_root, objectid, | |
0660b5af | 1447 | parent_root->root_key.objectid, |
33345d01 | 1448 | btrfs_ino(parent_inode), index, |
a22285a6 | 1449 | dentry->d_name.name, dentry->d_name.len); |
8732d44f MX |
1450 | if (ret) { |
1451 | btrfs_abort_transaction(trans, root, ret); | |
1452 | goto fail; | |
1453 | } | |
0660b5af | 1454 | |
a22285a6 YZ |
1455 | key.offset = (u64)-1; |
1456 | pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key); | |
79787eaa JM |
1457 | if (IS_ERR(pending->snap)) { |
1458 | ret = PTR_ERR(pending->snap); | |
8732d44f MX |
1459 | btrfs_abort_transaction(trans, root, ret); |
1460 | goto fail; | |
79787eaa | 1461 | } |
d68fc57b | 1462 | |
49b25e05 | 1463 | ret = btrfs_reloc_post_snapshot(trans, pending); |
8732d44f MX |
1464 | if (ret) { |
1465 | btrfs_abort_transaction(trans, root, ret); | |
1466 | goto fail; | |
1467 | } | |
361048f5 MX |
1468 | |
1469 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); | |
8732d44f MX |
1470 | if (ret) { |
1471 | btrfs_abort_transaction(trans, root, ret); | |
1472 | goto fail; | |
1473 | } | |
42874b3d MX |
1474 | |
1475 | ret = btrfs_insert_dir_item(trans, parent_root, | |
1476 | dentry->d_name.name, dentry->d_name.len, | |
1477 | parent_inode, &key, | |
1478 | BTRFS_FT_DIR, index); | |
1479 | /* We have check then name at the beginning, so it is impossible. */ | |
9c52057c | 1480 | BUG_ON(ret == -EEXIST || ret == -EOVERFLOW); |
8732d44f MX |
1481 | if (ret) { |
1482 | btrfs_abort_transaction(trans, root, ret); | |
1483 | goto fail; | |
1484 | } | |
42874b3d MX |
1485 | |
1486 | btrfs_i_size_write(parent_inode, parent_inode->i_size + | |
1487 | dentry->d_name.len * 2); | |
1488 | parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME; | |
be6aef60 | 1489 | ret = btrfs_update_inode_fallback(trans, parent_root, parent_inode); |
dd5f9615 SB |
1490 | if (ret) { |
1491 | btrfs_abort_transaction(trans, root, ret); | |
1492 | goto fail; | |
1493 | } | |
1494 | ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root, new_uuid.b, | |
1495 | BTRFS_UUID_KEY_SUBVOL, objectid); | |
1496 | if (ret) { | |
8732d44f | 1497 | btrfs_abort_transaction(trans, root, ret); |
dd5f9615 SB |
1498 | goto fail; |
1499 | } | |
1500 | if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) { | |
1501 | ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root, | |
1502 | new_root_item->received_uuid, | |
1503 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, | |
1504 | objectid); | |
1505 | if (ret && ret != -EEXIST) { | |
1506 | btrfs_abort_transaction(trans, root, ret); | |
1507 | goto fail; | |
1508 | } | |
1509 | } | |
d6726335 QW |
1510 | |
1511 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); | |
1512 | if (ret) { | |
1513 | btrfs_abort_transaction(trans, root, ret); | |
1514 | goto fail; | |
1515 | } | |
1516 | ||
1517 | /* | |
1518 | * account qgroup counters before qgroup_inherit() | |
1519 | */ | |
1520 | ret = btrfs_qgroup_prepare_account_extents(trans, fs_info); | |
1521 | if (ret) | |
1522 | goto fail; | |
1523 | ret = btrfs_qgroup_account_extents(trans, fs_info); | |
1524 | if (ret) | |
1525 | goto fail; | |
1526 | ret = btrfs_qgroup_inherit(trans, fs_info, | |
1527 | root->root_key.objectid, | |
1528 | objectid, pending->inherit); | |
1529 | if (ret) { | |
1530 | btrfs_abort_transaction(trans, root, ret); | |
1531 | goto fail; | |
1532 | } | |
1533 | ||
3063d29f | 1534 | fail: |
aec8030a MX |
1535 | pending->error = ret; |
1536 | dir_item_existed: | |
98c9942a | 1537 | trans->block_rsv = rsv; |
2382c5cc | 1538 | trans->bytes_reserved = 0; |
d6726335 QW |
1539 | clear_skip_qgroup: |
1540 | btrfs_clear_skip_qgroup(trans); | |
6fa9700e MX |
1541 | no_free_objectid: |
1542 | kfree(new_root_item); | |
1543 | root_item_alloc_fail: | |
42874b3d | 1544 | btrfs_free_path(path); |
49b25e05 | 1545 | return ret; |
3063d29f CM |
1546 | } |
1547 | ||
d352ac68 CM |
1548 | /* |
1549 | * create all the snapshots we've scheduled for creation | |
1550 | */ | |
80b6794d CM |
1551 | static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans, |
1552 | struct btrfs_fs_info *fs_info) | |
3de4586c | 1553 | { |
aec8030a | 1554 | struct btrfs_pending_snapshot *pending, *next; |
3de4586c | 1555 | struct list_head *head = &trans->transaction->pending_snapshots; |
aec8030a | 1556 | int ret = 0; |
3de4586c | 1557 | |
aec8030a MX |
1558 | list_for_each_entry_safe(pending, next, head, list) { |
1559 | list_del(&pending->list); | |
1560 | ret = create_pending_snapshot(trans, fs_info, pending); | |
1561 | if (ret) | |
1562 | break; | |
1563 | } | |
1564 | return ret; | |
3de4586c CM |
1565 | } |
1566 | ||
5d4f98a2 YZ |
1567 | static void update_super_roots(struct btrfs_root *root) |
1568 | { | |
1569 | struct btrfs_root_item *root_item; | |
1570 | struct btrfs_super_block *super; | |
1571 | ||
6c41761f | 1572 | super = root->fs_info->super_copy; |
5d4f98a2 YZ |
1573 | |
1574 | root_item = &root->fs_info->chunk_root->root_item; | |
1575 | super->chunk_root = root_item->bytenr; | |
1576 | super->chunk_root_generation = root_item->generation; | |
1577 | super->chunk_root_level = root_item->level; | |
1578 | ||
1579 | root_item = &root->fs_info->tree_root->root_item; | |
1580 | super->root = root_item->bytenr; | |
1581 | super->generation = root_item->generation; | |
1582 | super->root_level = root_item->level; | |
73bc1876 | 1583 | if (btrfs_test_opt(root, SPACE_CACHE)) |
0af3d00b | 1584 | super->cache_generation = root_item->generation; |
70f80175 SB |
1585 | if (root->fs_info->update_uuid_tree_gen) |
1586 | super->uuid_tree_generation = root_item->generation; | |
5d4f98a2 YZ |
1587 | } |
1588 | ||
f36f3042 CM |
1589 | int btrfs_transaction_in_commit(struct btrfs_fs_info *info) |
1590 | { | |
4a9d8bde | 1591 | struct btrfs_transaction *trans; |
f36f3042 | 1592 | int ret = 0; |
4a9d8bde | 1593 | |
a4abeea4 | 1594 | spin_lock(&info->trans_lock); |
4a9d8bde MX |
1595 | trans = info->running_transaction; |
1596 | if (trans) | |
1597 | ret = (trans->state >= TRANS_STATE_COMMIT_START); | |
a4abeea4 | 1598 | spin_unlock(&info->trans_lock); |
f36f3042 CM |
1599 | return ret; |
1600 | } | |
1601 | ||
8929ecfa YZ |
1602 | int btrfs_transaction_blocked(struct btrfs_fs_info *info) |
1603 | { | |
4a9d8bde | 1604 | struct btrfs_transaction *trans; |
8929ecfa | 1605 | int ret = 0; |
4a9d8bde | 1606 | |
a4abeea4 | 1607 | spin_lock(&info->trans_lock); |
4a9d8bde MX |
1608 | trans = info->running_transaction; |
1609 | if (trans) | |
1610 | ret = is_transaction_blocked(trans); | |
a4abeea4 | 1611 | spin_unlock(&info->trans_lock); |
8929ecfa YZ |
1612 | return ret; |
1613 | } | |
1614 | ||
bb9c12c9 SW |
1615 | /* |
1616 | * wait for the current transaction commit to start and block subsequent | |
1617 | * transaction joins | |
1618 | */ | |
1619 | static void wait_current_trans_commit_start(struct btrfs_root *root, | |
1620 | struct btrfs_transaction *trans) | |
1621 | { | |
4a9d8bde | 1622 | wait_event(root->fs_info->transaction_blocked_wait, |
501407aa JB |
1623 | trans->state >= TRANS_STATE_COMMIT_START || |
1624 | trans->aborted); | |
bb9c12c9 SW |
1625 | } |
1626 | ||
1627 | /* | |
1628 | * wait for the current transaction to start and then become unblocked. | |
1629 | * caller holds ref. | |
1630 | */ | |
1631 | static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root, | |
1632 | struct btrfs_transaction *trans) | |
1633 | { | |
72d63ed6 | 1634 | wait_event(root->fs_info->transaction_wait, |
501407aa JB |
1635 | trans->state >= TRANS_STATE_UNBLOCKED || |
1636 | trans->aborted); | |
bb9c12c9 SW |
1637 | } |
1638 | ||
1639 | /* | |
1640 | * commit transactions asynchronously. once btrfs_commit_transaction_async | |
1641 | * returns, any subsequent transaction will not be allowed to join. | |
1642 | */ | |
1643 | struct btrfs_async_commit { | |
1644 | struct btrfs_trans_handle *newtrans; | |
1645 | struct btrfs_root *root; | |
7892b5af | 1646 | struct work_struct work; |
bb9c12c9 SW |
1647 | }; |
1648 | ||
1649 | static void do_async_commit(struct work_struct *work) | |
1650 | { | |
1651 | struct btrfs_async_commit *ac = | |
7892b5af | 1652 | container_of(work, struct btrfs_async_commit, work); |
bb9c12c9 | 1653 | |
6fc4e354 SW |
1654 | /* |
1655 | * We've got freeze protection passed with the transaction. | |
1656 | * Tell lockdep about it. | |
1657 | */ | |
b1a06a4b | 1658 | if (ac->newtrans->type & __TRANS_FREEZABLE) |
bee9182d | 1659 | __sb_writers_acquired(ac->root->fs_info->sb, SB_FREEZE_FS); |
6fc4e354 | 1660 | |
e209db7a SW |
1661 | current->journal_info = ac->newtrans; |
1662 | ||
bb9c12c9 SW |
1663 | btrfs_commit_transaction(ac->newtrans, ac->root); |
1664 | kfree(ac); | |
1665 | } | |
1666 | ||
1667 | int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans, | |
1668 | struct btrfs_root *root, | |
1669 | int wait_for_unblock) | |
1670 | { | |
1671 | struct btrfs_async_commit *ac; | |
1672 | struct btrfs_transaction *cur_trans; | |
1673 | ||
1674 | ac = kmalloc(sizeof(*ac), GFP_NOFS); | |
db5b493a TI |
1675 | if (!ac) |
1676 | return -ENOMEM; | |
bb9c12c9 | 1677 | |
7892b5af | 1678 | INIT_WORK(&ac->work, do_async_commit); |
bb9c12c9 | 1679 | ac->root = root; |
7a7eaa40 | 1680 | ac->newtrans = btrfs_join_transaction(root); |
3612b495 TI |
1681 | if (IS_ERR(ac->newtrans)) { |
1682 | int err = PTR_ERR(ac->newtrans); | |
1683 | kfree(ac); | |
1684 | return err; | |
1685 | } | |
bb9c12c9 SW |
1686 | |
1687 | /* take transaction reference */ | |
bb9c12c9 | 1688 | cur_trans = trans->transaction; |
13c5a93e | 1689 | atomic_inc(&cur_trans->use_count); |
bb9c12c9 SW |
1690 | |
1691 | btrfs_end_transaction(trans, root); | |
6fc4e354 SW |
1692 | |
1693 | /* | |
1694 | * Tell lockdep we've released the freeze rwsem, since the | |
1695 | * async commit thread will be the one to unlock it. | |
1696 | */ | |
b1a06a4b | 1697 | if (ac->newtrans->type & __TRANS_FREEZABLE) |
bee9182d | 1698 | __sb_writers_release(root->fs_info->sb, SB_FREEZE_FS); |
6fc4e354 | 1699 | |
7892b5af | 1700 | schedule_work(&ac->work); |
bb9c12c9 SW |
1701 | |
1702 | /* wait for transaction to start and unblock */ | |
bb9c12c9 SW |
1703 | if (wait_for_unblock) |
1704 | wait_current_trans_commit_start_and_unblock(root, cur_trans); | |
1705 | else | |
1706 | wait_current_trans_commit_start(root, cur_trans); | |
bb9c12c9 | 1707 | |
38e88054 SW |
1708 | if (current->journal_info == trans) |
1709 | current->journal_info = NULL; | |
1710 | ||
724e2315 | 1711 | btrfs_put_transaction(cur_trans); |
bb9c12c9 SW |
1712 | return 0; |
1713 | } | |
1714 | ||
49b25e05 JM |
1715 | |
1716 | static void cleanup_transaction(struct btrfs_trans_handle *trans, | |
7b8b92af | 1717 | struct btrfs_root *root, int err) |
49b25e05 JM |
1718 | { |
1719 | struct btrfs_transaction *cur_trans = trans->transaction; | |
f094ac32 | 1720 | DEFINE_WAIT(wait); |
49b25e05 JM |
1721 | |
1722 | WARN_ON(trans->use_count > 1); | |
1723 | ||
7b8b92af JB |
1724 | btrfs_abort_transaction(trans, root, err); |
1725 | ||
49b25e05 | 1726 | spin_lock(&root->fs_info->trans_lock); |
66b6135b | 1727 | |
25d8c284 MX |
1728 | /* |
1729 | * If the transaction is removed from the list, it means this | |
1730 | * transaction has been committed successfully, so it is impossible | |
1731 | * to call the cleanup function. | |
1732 | */ | |
1733 | BUG_ON(list_empty(&cur_trans->list)); | |
66b6135b | 1734 | |
49b25e05 | 1735 | list_del_init(&cur_trans->list); |
d7096fc3 | 1736 | if (cur_trans == root->fs_info->running_transaction) { |
4a9d8bde | 1737 | cur_trans->state = TRANS_STATE_COMMIT_DOING; |
f094ac32 LB |
1738 | spin_unlock(&root->fs_info->trans_lock); |
1739 | wait_event(cur_trans->writer_wait, | |
1740 | atomic_read(&cur_trans->num_writers) == 1); | |
1741 | ||
1742 | spin_lock(&root->fs_info->trans_lock); | |
d7096fc3 | 1743 | } |
49b25e05 JM |
1744 | spin_unlock(&root->fs_info->trans_lock); |
1745 | ||
1746 | btrfs_cleanup_one_transaction(trans->transaction, root); | |
1747 | ||
4a9d8bde MX |
1748 | spin_lock(&root->fs_info->trans_lock); |
1749 | if (cur_trans == root->fs_info->running_transaction) | |
1750 | root->fs_info->running_transaction = NULL; | |
1751 | spin_unlock(&root->fs_info->trans_lock); | |
1752 | ||
e0228285 JB |
1753 | if (trans->type & __TRANS_FREEZABLE) |
1754 | sb_end_intwrite(root->fs_info->sb); | |
724e2315 JB |
1755 | btrfs_put_transaction(cur_trans); |
1756 | btrfs_put_transaction(cur_trans); | |
49b25e05 JM |
1757 | |
1758 | trace_btrfs_transaction_commit(root); | |
1759 | ||
49b25e05 JM |
1760 | if (current->journal_info == trans) |
1761 | current->journal_info = NULL; | |
c0af8f0b | 1762 | btrfs_scrub_cancel(root->fs_info); |
49b25e05 JM |
1763 | |
1764 | kmem_cache_free(btrfs_trans_handle_cachep, trans); | |
1765 | } | |
1766 | ||
82436617 MX |
1767 | static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info) |
1768 | { | |
1769 | if (btrfs_test_opt(fs_info->tree_root, FLUSHONCOMMIT)) | |
6c255e67 | 1770 | return btrfs_start_delalloc_roots(fs_info, 1, -1); |
82436617 MX |
1771 | return 0; |
1772 | } | |
1773 | ||
1774 | static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info) | |
1775 | { | |
1776 | if (btrfs_test_opt(fs_info->tree_root, FLUSHONCOMMIT)) | |
b0244199 | 1777 | btrfs_wait_ordered_roots(fs_info, -1); |
82436617 MX |
1778 | } |
1779 | ||
50d9aa99 JB |
1780 | static inline void |
1781 | btrfs_wait_pending_ordered(struct btrfs_transaction *cur_trans, | |
1782 | struct btrfs_fs_info *fs_info) | |
1783 | { | |
1784 | struct btrfs_ordered_extent *ordered; | |
1785 | ||
1786 | spin_lock(&fs_info->trans_lock); | |
1787 | while (!list_empty(&cur_trans->pending_ordered)) { | |
1788 | ordered = list_first_entry(&cur_trans->pending_ordered, | |
1789 | struct btrfs_ordered_extent, | |
1790 | trans_list); | |
1791 | list_del_init(&ordered->trans_list); | |
1792 | spin_unlock(&fs_info->trans_lock); | |
1793 | ||
1794 | wait_event(ordered->wait, test_bit(BTRFS_ORDERED_COMPLETE, | |
1795 | &ordered->flags)); | |
1796 | btrfs_put_ordered_extent(ordered); | |
1797 | spin_lock(&fs_info->trans_lock); | |
1798 | } | |
1799 | spin_unlock(&fs_info->trans_lock); | |
1800 | } | |
1801 | ||
79154b1b CM |
1802 | int btrfs_commit_transaction(struct btrfs_trans_handle *trans, |
1803 | struct btrfs_root *root) | |
1804 | { | |
49b25e05 | 1805 | struct btrfs_transaction *cur_trans = trans->transaction; |
8fd17795 | 1806 | struct btrfs_transaction *prev_trans = NULL; |
656f30db | 1807 | struct btrfs_inode *btree_ino = BTRFS_I(root->fs_info->btree_inode); |
25287e0a | 1808 | int ret; |
79154b1b | 1809 | |
8d25a086 MX |
1810 | /* Stop the commit early if ->aborted is set */ |
1811 | if (unlikely(ACCESS_ONCE(cur_trans->aborted))) { | |
25287e0a | 1812 | ret = cur_trans->aborted; |
e4a2bcac JB |
1813 | btrfs_end_transaction(trans, root); |
1814 | return ret; | |
25287e0a | 1815 | } |
49b25e05 | 1816 | |
56bec294 CM |
1817 | /* make a pass through all the delayed refs we have so far |
1818 | * any runnings procs may add more while we are here | |
1819 | */ | |
1820 | ret = btrfs_run_delayed_refs(trans, root, 0); | |
e4a2bcac JB |
1821 | if (ret) { |
1822 | btrfs_end_transaction(trans, root); | |
1823 | return ret; | |
1824 | } | |
56bec294 | 1825 | |
0e721106 JB |
1826 | btrfs_trans_release_metadata(trans, root); |
1827 | trans->block_rsv = NULL; | |
272d26d0 MX |
1828 | if (trans->qgroup_reserved) { |
1829 | btrfs_qgroup_free(root, trans->qgroup_reserved); | |
1830 | trans->qgroup_reserved = 0; | |
1831 | } | |
0e721106 | 1832 | |
b7ec40d7 | 1833 | cur_trans = trans->transaction; |
49b25e05 | 1834 | |
56bec294 CM |
1835 | /* |
1836 | * set the flushing flag so procs in this transaction have to | |
1837 | * start sending their work down. | |
1838 | */ | |
b7ec40d7 | 1839 | cur_trans->delayed_refs.flushing = 1; |
1be41b78 | 1840 | smp_wmb(); |
56bec294 | 1841 | |
ea658bad JB |
1842 | if (!list_empty(&trans->new_bgs)) |
1843 | btrfs_create_pending_block_groups(trans, root); | |
1844 | ||
c3e69d58 | 1845 | ret = btrfs_run_delayed_refs(trans, root, 0); |
e4a2bcac JB |
1846 | if (ret) { |
1847 | btrfs_end_transaction(trans, root); | |
1848 | return ret; | |
1849 | } | |
56bec294 | 1850 | |
1bbc621e CM |
1851 | if (!cur_trans->dirty_bg_run) { |
1852 | int run_it = 0; | |
1853 | ||
1854 | /* this mutex is also taken before trying to set | |
1855 | * block groups readonly. We need to make sure | |
1856 | * that nobody has set a block group readonly | |
1857 | * after a extents from that block group have been | |
1858 | * allocated for cache files. btrfs_set_block_group_ro | |
1859 | * will wait for the transaction to commit if it | |
1860 | * finds dirty_bg_run = 1 | |
1861 | * | |
1862 | * The dirty_bg_run flag is also used to make sure only | |
1863 | * one process starts all the block group IO. It wouldn't | |
1864 | * hurt to have more than one go through, but there's no | |
1865 | * real advantage to it either. | |
1866 | */ | |
1867 | mutex_lock(&root->fs_info->ro_block_group_mutex); | |
1868 | if (!cur_trans->dirty_bg_run) { | |
1869 | run_it = 1; | |
1870 | cur_trans->dirty_bg_run = 1; | |
1871 | } | |
1872 | mutex_unlock(&root->fs_info->ro_block_group_mutex); | |
1873 | ||
1874 | if (run_it) | |
1875 | ret = btrfs_start_dirty_block_groups(trans, root); | |
1876 | } | |
1877 | if (ret) { | |
1878 | btrfs_end_transaction(trans, root); | |
1879 | return ret; | |
1880 | } | |
1881 | ||
4a9d8bde | 1882 | spin_lock(&root->fs_info->trans_lock); |
d3efe084 | 1883 | list_splice_init(&trans->ordered, &cur_trans->pending_ordered); |
4a9d8bde MX |
1884 | if (cur_trans->state >= TRANS_STATE_COMMIT_START) { |
1885 | spin_unlock(&root->fs_info->trans_lock); | |
13c5a93e | 1886 | atomic_inc(&cur_trans->use_count); |
49b25e05 | 1887 | ret = btrfs_end_transaction(trans, root); |
ccd467d6 | 1888 | |
b9c8300c | 1889 | wait_for_commit(root, cur_trans); |
15ee9bc7 | 1890 | |
b4924a0f LB |
1891 | if (unlikely(cur_trans->aborted)) |
1892 | ret = cur_trans->aborted; | |
1893 | ||
724e2315 | 1894 | btrfs_put_transaction(cur_trans); |
15ee9bc7 | 1895 | |
49b25e05 | 1896 | return ret; |
79154b1b | 1897 | } |
4313b399 | 1898 | |
4a9d8bde | 1899 | cur_trans->state = TRANS_STATE_COMMIT_START; |
bb9c12c9 SW |
1900 | wake_up(&root->fs_info->transaction_blocked_wait); |
1901 | ||
ccd467d6 CM |
1902 | if (cur_trans->list.prev != &root->fs_info->trans_list) { |
1903 | prev_trans = list_entry(cur_trans->list.prev, | |
1904 | struct btrfs_transaction, list); | |
4a9d8bde | 1905 | if (prev_trans->state != TRANS_STATE_COMPLETED) { |
13c5a93e | 1906 | atomic_inc(&prev_trans->use_count); |
a4abeea4 | 1907 | spin_unlock(&root->fs_info->trans_lock); |
ccd467d6 CM |
1908 | |
1909 | wait_for_commit(root, prev_trans); | |
1f9b8c8f | 1910 | ret = prev_trans->aborted; |
ccd467d6 | 1911 | |
724e2315 | 1912 | btrfs_put_transaction(prev_trans); |
1f9b8c8f FM |
1913 | if (ret) |
1914 | goto cleanup_transaction; | |
a4abeea4 JB |
1915 | } else { |
1916 | spin_unlock(&root->fs_info->trans_lock); | |
ccd467d6 | 1917 | } |
a4abeea4 JB |
1918 | } else { |
1919 | spin_unlock(&root->fs_info->trans_lock); | |
ccd467d6 | 1920 | } |
15ee9bc7 | 1921 | |
0860adfd MX |
1922 | extwriter_counter_dec(cur_trans, trans->type); |
1923 | ||
82436617 MX |
1924 | ret = btrfs_start_delalloc_flush(root->fs_info); |
1925 | if (ret) | |
1926 | goto cleanup_transaction; | |
1927 | ||
8d875f95 | 1928 | ret = btrfs_run_delayed_items(trans, root); |
581227d0 MX |
1929 | if (ret) |
1930 | goto cleanup_transaction; | |
15ee9bc7 | 1931 | |
581227d0 MX |
1932 | wait_event(cur_trans->writer_wait, |
1933 | extwriter_counter_read(cur_trans) == 0); | |
15ee9bc7 | 1934 | |
581227d0 | 1935 | /* some pending stuffs might be added after the previous flush. */ |
8d875f95 | 1936 | ret = btrfs_run_delayed_items(trans, root); |
ca469637 MX |
1937 | if (ret) |
1938 | goto cleanup_transaction; | |
1939 | ||
82436617 | 1940 | btrfs_wait_delalloc_flush(root->fs_info); |
cb7ab021 | 1941 | |
50d9aa99 JB |
1942 | btrfs_wait_pending_ordered(cur_trans, root->fs_info); |
1943 | ||
cb7ab021 | 1944 | btrfs_scrub_pause(root); |
ed0ca140 JB |
1945 | /* |
1946 | * Ok now we need to make sure to block out any other joins while we | |
1947 | * commit the transaction. We could have started a join before setting | |
4a9d8bde | 1948 | * COMMIT_DOING so make sure to wait for num_writers to == 1 again. |
ed0ca140 JB |
1949 | */ |
1950 | spin_lock(&root->fs_info->trans_lock); | |
4a9d8bde | 1951 | cur_trans->state = TRANS_STATE_COMMIT_DOING; |
ed0ca140 JB |
1952 | spin_unlock(&root->fs_info->trans_lock); |
1953 | wait_event(cur_trans->writer_wait, | |
1954 | atomic_read(&cur_trans->num_writers) == 1); | |
1955 | ||
2cba30f1 MX |
1956 | /* ->aborted might be set after the previous check, so check it */ |
1957 | if (unlikely(ACCESS_ONCE(cur_trans->aborted))) { | |
1958 | ret = cur_trans->aborted; | |
6cf7f77e | 1959 | goto scrub_continue; |
2cba30f1 | 1960 | } |
7585717f CM |
1961 | /* |
1962 | * the reloc mutex makes sure that we stop | |
1963 | * the balancing code from coming in and moving | |
1964 | * extents around in the middle of the commit | |
1965 | */ | |
1966 | mutex_lock(&root->fs_info->reloc_mutex); | |
1967 | ||
42874b3d MX |
1968 | /* |
1969 | * We needn't worry about the delayed items because we will | |
1970 | * deal with them in create_pending_snapshot(), which is the | |
1971 | * core function of the snapshot creation. | |
1972 | */ | |
1973 | ret = create_pending_snapshots(trans, root->fs_info); | |
49b25e05 JM |
1974 | if (ret) { |
1975 | mutex_unlock(&root->fs_info->reloc_mutex); | |
6cf7f77e | 1976 | goto scrub_continue; |
49b25e05 | 1977 | } |
3063d29f | 1978 | |
42874b3d MX |
1979 | /* |
1980 | * We insert the dir indexes of the snapshots and update the inode | |
1981 | * of the snapshots' parents after the snapshot creation, so there | |
1982 | * are some delayed items which are not dealt with. Now deal with | |
1983 | * them. | |
1984 | * | |
1985 | * We needn't worry that this operation will corrupt the snapshots, | |
1986 | * because all the tree which are snapshoted will be forced to COW | |
1987 | * the nodes and leaves. | |
1988 | */ | |
1989 | ret = btrfs_run_delayed_items(trans, root); | |
49b25e05 JM |
1990 | if (ret) { |
1991 | mutex_unlock(&root->fs_info->reloc_mutex); | |
6cf7f77e | 1992 | goto scrub_continue; |
49b25e05 | 1993 | } |
16cdcec7 | 1994 | |
56bec294 | 1995 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
49b25e05 JM |
1996 | if (ret) { |
1997 | mutex_unlock(&root->fs_info->reloc_mutex); | |
6cf7f77e | 1998 | goto scrub_continue; |
49b25e05 | 1999 | } |
56bec294 | 2000 | |
0ed4792a QW |
2001 | /* Reocrd old roots for later qgroup accounting */ |
2002 | ret = btrfs_qgroup_prepare_account_extents(trans, root->fs_info); | |
2003 | if (ret) { | |
2004 | mutex_unlock(&root->fs_info->reloc_mutex); | |
2005 | goto scrub_continue; | |
2006 | } | |
2007 | ||
e999376f CM |
2008 | /* |
2009 | * make sure none of the code above managed to slip in a | |
2010 | * delayed item | |
2011 | */ | |
2012 | btrfs_assert_delayed_root_empty(root); | |
2013 | ||
2c90e5d6 | 2014 | WARN_ON(cur_trans != trans->transaction); |
dc17ff8f | 2015 | |
e02119d5 CM |
2016 | /* btrfs_commit_tree_roots is responsible for getting the |
2017 | * various roots consistent with each other. Every pointer | |
2018 | * in the tree of tree roots has to point to the most up to date | |
2019 | * root for every subvolume and other tree. So, we have to keep | |
2020 | * the tree logging code from jumping in and changing any | |
2021 | * of the trees. | |
2022 | * | |
2023 | * At this point in the commit, there can't be any tree-log | |
2024 | * writers, but a little lower down we drop the trans mutex | |
2025 | * and let new people in. By holding the tree_log_mutex | |
2026 | * from now until after the super is written, we avoid races | |
2027 | * with the tree-log code. | |
2028 | */ | |
2029 | mutex_lock(&root->fs_info->tree_log_mutex); | |
2030 | ||
5d4f98a2 | 2031 | ret = commit_fs_roots(trans, root); |
49b25e05 JM |
2032 | if (ret) { |
2033 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
871383be | 2034 | mutex_unlock(&root->fs_info->reloc_mutex); |
6cf7f77e | 2035 | goto scrub_continue; |
49b25e05 | 2036 | } |
54aa1f4d | 2037 | |
3818aea2 | 2038 | /* |
7e1876ac DS |
2039 | * Since the transaction is done, we can apply the pending changes |
2040 | * before the next transaction. | |
3818aea2 | 2041 | */ |
572d9ab7 | 2042 | btrfs_apply_pending_changes(root->fs_info); |
3818aea2 | 2043 | |
5d4f98a2 | 2044 | /* commit_fs_roots gets rid of all the tree log roots, it is now |
e02119d5 CM |
2045 | * safe to free the root of tree log roots |
2046 | */ | |
2047 | btrfs_free_log_root_tree(trans, root->fs_info); | |
2048 | ||
0ed4792a QW |
2049 | /* |
2050 | * Since fs roots are all committed, we can get a quite accurate | |
2051 | * new_roots. So let's do quota accounting. | |
2052 | */ | |
2053 | ret = btrfs_qgroup_account_extents(trans, root->fs_info); | |
2054 | if (ret < 0) { | |
2055 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
2056 | mutex_unlock(&root->fs_info->reloc_mutex); | |
2057 | goto scrub_continue; | |
2058 | } | |
2059 | ||
5d4f98a2 | 2060 | ret = commit_cowonly_roots(trans, root); |
49b25e05 JM |
2061 | if (ret) { |
2062 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
871383be | 2063 | mutex_unlock(&root->fs_info->reloc_mutex); |
6cf7f77e | 2064 | goto scrub_continue; |
49b25e05 | 2065 | } |
54aa1f4d | 2066 | |
2cba30f1 MX |
2067 | /* |
2068 | * The tasks which save the space cache and inode cache may also | |
2069 | * update ->aborted, check it. | |
2070 | */ | |
2071 | if (unlikely(ACCESS_ONCE(cur_trans->aborted))) { | |
2072 | ret = cur_trans->aborted; | |
2073 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
2074 | mutex_unlock(&root->fs_info->reloc_mutex); | |
6cf7f77e | 2075 | goto scrub_continue; |
2cba30f1 MX |
2076 | } |
2077 | ||
11833d66 YZ |
2078 | btrfs_prepare_extent_commit(trans, root); |
2079 | ||
78fae27e | 2080 | cur_trans = root->fs_info->running_transaction; |
5d4f98a2 YZ |
2081 | |
2082 | btrfs_set_root_node(&root->fs_info->tree_root->root_item, | |
2083 | root->fs_info->tree_root->node); | |
9e351cc8 JB |
2084 | list_add_tail(&root->fs_info->tree_root->dirty_list, |
2085 | &cur_trans->switch_commits); | |
5d4f98a2 YZ |
2086 | |
2087 | btrfs_set_root_node(&root->fs_info->chunk_root->root_item, | |
2088 | root->fs_info->chunk_root->node); | |
9e351cc8 JB |
2089 | list_add_tail(&root->fs_info->chunk_root->dirty_list, |
2090 | &cur_trans->switch_commits); | |
2091 | ||
2092 | switch_commit_roots(cur_trans, root->fs_info); | |
5d4f98a2 | 2093 | |
edf39272 | 2094 | assert_qgroups_uptodate(trans); |
ce93ec54 | 2095 | ASSERT(list_empty(&cur_trans->dirty_bgs)); |
1bbc621e | 2096 | ASSERT(list_empty(&cur_trans->io_bgs)); |
5d4f98a2 | 2097 | update_super_roots(root); |
e02119d5 | 2098 | |
60e7cd3a JB |
2099 | btrfs_set_super_log_root(root->fs_info->super_copy, 0); |
2100 | btrfs_set_super_log_root_level(root->fs_info->super_copy, 0); | |
6c41761f DS |
2101 | memcpy(root->fs_info->super_for_commit, root->fs_info->super_copy, |
2102 | sizeof(*root->fs_info->super_copy)); | |
ccd467d6 | 2103 | |
935e5cc9 | 2104 | btrfs_update_commit_device_size(root->fs_info); |
ce7213c7 | 2105 | btrfs_update_commit_device_bytes_used(root, cur_trans); |
935e5cc9 | 2106 | |
656f30db FM |
2107 | clear_bit(BTRFS_INODE_BTREE_LOG1_ERR, &btree_ino->runtime_flags); |
2108 | clear_bit(BTRFS_INODE_BTREE_LOG2_ERR, &btree_ino->runtime_flags); | |
2109 | ||
4fbcdf66 FM |
2110 | btrfs_trans_release_chunk_metadata(trans); |
2111 | ||
a4abeea4 | 2112 | spin_lock(&root->fs_info->trans_lock); |
4a9d8bde | 2113 | cur_trans->state = TRANS_STATE_UNBLOCKED; |
a4abeea4 | 2114 | root->fs_info->running_transaction = NULL; |
a4abeea4 | 2115 | spin_unlock(&root->fs_info->trans_lock); |
7585717f | 2116 | mutex_unlock(&root->fs_info->reloc_mutex); |
b7ec40d7 | 2117 | |
f9295749 | 2118 | wake_up(&root->fs_info->transaction_wait); |
e6dcd2dc | 2119 | |
79154b1b | 2120 | ret = btrfs_write_and_wait_transaction(trans, root); |
49b25e05 JM |
2121 | if (ret) { |
2122 | btrfs_error(root->fs_info, ret, | |
08748810 | 2123 | "Error while writing out transaction"); |
49b25e05 | 2124 | mutex_unlock(&root->fs_info->tree_log_mutex); |
6cf7f77e | 2125 | goto scrub_continue; |
49b25e05 JM |
2126 | } |
2127 | ||
2128 | ret = write_ctree_super(trans, root, 0); | |
2129 | if (ret) { | |
2130 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
6cf7f77e | 2131 | goto scrub_continue; |
49b25e05 | 2132 | } |
4313b399 | 2133 | |
e02119d5 CM |
2134 | /* |
2135 | * the super is written, we can safely allow the tree-loggers | |
2136 | * to go about their business | |
2137 | */ | |
2138 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
2139 | ||
11833d66 | 2140 | btrfs_finish_extent_commit(trans, root); |
4313b399 | 2141 | |
13212b54 ZL |
2142 | if (cur_trans->have_free_bgs) |
2143 | btrfs_clear_space_info_full(root->fs_info); | |
2144 | ||
15ee9bc7 | 2145 | root->fs_info->last_trans_committed = cur_trans->transid; |
4a9d8bde MX |
2146 | /* |
2147 | * We needn't acquire the lock here because there is no other task | |
2148 | * which can change it. | |
2149 | */ | |
2150 | cur_trans->state = TRANS_STATE_COMPLETED; | |
2c90e5d6 | 2151 | wake_up(&cur_trans->commit_wait); |
3de4586c | 2152 | |
a4abeea4 | 2153 | spin_lock(&root->fs_info->trans_lock); |
13c5a93e | 2154 | list_del_init(&cur_trans->list); |
a4abeea4 JB |
2155 | spin_unlock(&root->fs_info->trans_lock); |
2156 | ||
724e2315 JB |
2157 | btrfs_put_transaction(cur_trans); |
2158 | btrfs_put_transaction(cur_trans); | |
58176a96 | 2159 | |
0860adfd | 2160 | if (trans->type & __TRANS_FREEZABLE) |
354aa0fb | 2161 | sb_end_intwrite(root->fs_info->sb); |
b2b5ef5c | 2162 | |
1abe9b8a | 2163 | trace_btrfs_transaction_commit(root); |
2164 | ||
a2de733c AJ |
2165 | btrfs_scrub_continue(root); |
2166 | ||
9ed74f2d JB |
2167 | if (current->journal_info == trans) |
2168 | current->journal_info = NULL; | |
2169 | ||
2c90e5d6 | 2170 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
24bbcf04 | 2171 | |
8a733013 ZL |
2172 | if (current != root->fs_info->transaction_kthread && |
2173 | current != root->fs_info->cleaner_kthread) | |
24bbcf04 YZ |
2174 | btrfs_run_delayed_iputs(root); |
2175 | ||
79154b1b | 2176 | return ret; |
49b25e05 | 2177 | |
6cf7f77e WS |
2178 | scrub_continue: |
2179 | btrfs_scrub_continue(root); | |
49b25e05 | 2180 | cleanup_transaction: |
0e721106 | 2181 | btrfs_trans_release_metadata(trans, root); |
4fbcdf66 | 2182 | btrfs_trans_release_chunk_metadata(trans); |
0e721106 | 2183 | trans->block_rsv = NULL; |
272d26d0 MX |
2184 | if (trans->qgroup_reserved) { |
2185 | btrfs_qgroup_free(root, trans->qgroup_reserved); | |
2186 | trans->qgroup_reserved = 0; | |
2187 | } | |
c2cf52eb | 2188 | btrfs_warn(root->fs_info, "Skipping commit of aborted transaction."); |
49b25e05 JM |
2189 | if (current->journal_info == trans) |
2190 | current->journal_info = NULL; | |
7b8b92af | 2191 | cleanup_transaction(trans, root, ret); |
49b25e05 JM |
2192 | |
2193 | return ret; | |
79154b1b CM |
2194 | } |
2195 | ||
d352ac68 | 2196 | /* |
9d1a2a3a DS |
2197 | * return < 0 if error |
2198 | * 0 if there are no more dead_roots at the time of call | |
2199 | * 1 there are more to be processed, call me again | |
2200 | * | |
2201 | * The return value indicates there are certainly more snapshots to delete, but | |
2202 | * if there comes a new one during processing, it may return 0. We don't mind, | |
2203 | * because btrfs_commit_super will poke cleaner thread and it will process it a | |
2204 | * few seconds later. | |
d352ac68 | 2205 | */ |
9d1a2a3a | 2206 | int btrfs_clean_one_deleted_snapshot(struct btrfs_root *root) |
e9d0b13b | 2207 | { |
9d1a2a3a | 2208 | int ret; |
5d4f98a2 YZ |
2209 | struct btrfs_fs_info *fs_info = root->fs_info; |
2210 | ||
a4abeea4 | 2211 | spin_lock(&fs_info->trans_lock); |
9d1a2a3a DS |
2212 | if (list_empty(&fs_info->dead_roots)) { |
2213 | spin_unlock(&fs_info->trans_lock); | |
2214 | return 0; | |
2215 | } | |
2216 | root = list_first_entry(&fs_info->dead_roots, | |
2217 | struct btrfs_root, root_list); | |
cfad392b | 2218 | list_del_init(&root->root_list); |
a4abeea4 | 2219 | spin_unlock(&fs_info->trans_lock); |
e9d0b13b | 2220 | |
efe120a0 | 2221 | pr_debug("BTRFS: cleaner removing %llu\n", root->objectid); |
76dda93c | 2222 | |
9d1a2a3a | 2223 | btrfs_kill_all_delayed_nodes(root); |
16cdcec7 | 2224 | |
9d1a2a3a DS |
2225 | if (btrfs_header_backref_rev(root->node) < |
2226 | BTRFS_MIXED_BACKREF_REV) | |
2227 | ret = btrfs_drop_snapshot(root, NULL, 0, 0); | |
2228 | else | |
2229 | ret = btrfs_drop_snapshot(root, NULL, 1, 0); | |
32471dc2 | 2230 | |
6596a928 | 2231 | return (ret < 0) ? 0 : 1; |
e9d0b13b | 2232 | } |
572d9ab7 DS |
2233 | |
2234 | void btrfs_apply_pending_changes(struct btrfs_fs_info *fs_info) | |
2235 | { | |
2236 | unsigned long prev; | |
2237 | unsigned long bit; | |
2238 | ||
6c9fe14f | 2239 | prev = xchg(&fs_info->pending_changes, 0); |
572d9ab7 DS |
2240 | if (!prev) |
2241 | return; | |
2242 | ||
7e1876ac DS |
2243 | bit = 1 << BTRFS_PENDING_SET_INODE_MAP_CACHE; |
2244 | if (prev & bit) | |
2245 | btrfs_set_opt(fs_info->mount_opt, INODE_MAP_CACHE); | |
2246 | prev &= ~bit; | |
2247 | ||
2248 | bit = 1 << BTRFS_PENDING_CLEAR_INODE_MAP_CACHE; | |
2249 | if (prev & bit) | |
2250 | btrfs_clear_opt(fs_info->mount_opt, INODE_MAP_CACHE); | |
2251 | prev &= ~bit; | |
2252 | ||
d51033d0 DS |
2253 | bit = 1 << BTRFS_PENDING_COMMIT; |
2254 | if (prev & bit) | |
2255 | btrfs_debug(fs_info, "pending commit done"); | |
2256 | prev &= ~bit; | |
2257 | ||
572d9ab7 DS |
2258 | if (prev) |
2259 | btrfs_warn(fs_info, | |
2260 | "unknown pending changes left 0x%lx, ignoring", prev); | |
2261 | } |