]>
Commit | Line | Data |
---|---|---|
c1d7c514 | 1 | // SPDX-License-Identifier: GPL-2.0 |
6cbd5570 CM |
2 | /* |
3 | * Copyright (C) 2007 Oracle. All rights reserved. | |
6cbd5570 CM |
4 | */ |
5 | ||
79154b1b | 6 | #include <linux/fs.h> |
5a0e3ad6 | 7 | #include <linux/slab.h> |
34088780 | 8 | #include <linux/sched.h> |
d3c2fdcf | 9 | #include <linux/writeback.h> |
5f39d397 | 10 | #include <linux/pagemap.h> |
5f2cc086 | 11 | #include <linux/blkdev.h> |
8ea05e3a | 12 | #include <linux/uuid.h> |
79154b1b CM |
13 | #include "ctree.h" |
14 | #include "disk-io.h" | |
15 | #include "transaction.h" | |
925baedd | 16 | #include "locking.h" |
e02119d5 | 17 | #include "tree-log.h" |
581bb050 | 18 | #include "inode-map.h" |
733f4fbb | 19 | #include "volumes.h" |
8dabb742 | 20 | #include "dev-replace.h" |
fcebe456 | 21 | #include "qgroup.h" |
79154b1b | 22 | |
0f7d52f4 CM |
23 | #define BTRFS_ROOT_TRANS_TAG 0 |
24 | ||
e8c9f186 | 25 | static const unsigned int btrfs_blocked_trans_types[TRANS_STATE_MAX] = { |
4a9d8bde | 26 | [TRANS_STATE_RUNNING] = 0U, |
bcf3a3e7 NB |
27 | [TRANS_STATE_BLOCKED] = __TRANS_START, |
28 | [TRANS_STATE_COMMIT_START] = (__TRANS_START | __TRANS_ATTACH), | |
29 | [TRANS_STATE_COMMIT_DOING] = (__TRANS_START | | |
4a9d8bde MX |
30 | __TRANS_ATTACH | |
31 | __TRANS_JOIN), | |
bcf3a3e7 | 32 | [TRANS_STATE_UNBLOCKED] = (__TRANS_START | |
4a9d8bde MX |
33 | __TRANS_ATTACH | |
34 | __TRANS_JOIN | | |
35 | __TRANS_JOIN_NOLOCK), | |
bcf3a3e7 | 36 | [TRANS_STATE_COMPLETED] = (__TRANS_START | |
4a9d8bde MX |
37 | __TRANS_ATTACH | |
38 | __TRANS_JOIN | | |
39 | __TRANS_JOIN_NOLOCK), | |
40 | }; | |
41 | ||
724e2315 | 42 | void btrfs_put_transaction(struct btrfs_transaction *transaction) |
79154b1b | 43 | { |
9b64f57d ER |
44 | WARN_ON(refcount_read(&transaction->use_count) == 0); |
45 | if (refcount_dec_and_test(&transaction->use_count)) { | |
a4abeea4 | 46 | BUG_ON(!list_empty(&transaction->list)); |
c46effa6 | 47 | WARN_ON(!RB_EMPTY_ROOT(&transaction->delayed_refs.href_root)); |
1262133b | 48 | if (transaction->delayed_refs.pending_csums) |
ab8d0fc4 JM |
49 | btrfs_err(transaction->fs_info, |
50 | "pending csums is %llu", | |
51 | transaction->delayed_refs.pending_csums); | |
6df9a95e JB |
52 | while (!list_empty(&transaction->pending_chunks)) { |
53 | struct extent_map *em; | |
54 | ||
55 | em = list_first_entry(&transaction->pending_chunks, | |
56 | struct extent_map, list); | |
57 | list_del_init(&em->list); | |
58 | free_extent_map(em); | |
59 | } | |
7785a663 FM |
60 | /* |
61 | * If any block groups are found in ->deleted_bgs then it's | |
62 | * because the transaction was aborted and a commit did not | |
63 | * happen (things failed before writing the new superblock | |
64 | * and calling btrfs_finish_extent_commit()), so we can not | |
65 | * discard the physical locations of the block groups. | |
66 | */ | |
67 | while (!list_empty(&transaction->deleted_bgs)) { | |
68 | struct btrfs_block_group_cache *cache; | |
69 | ||
70 | cache = list_first_entry(&transaction->deleted_bgs, | |
71 | struct btrfs_block_group_cache, | |
72 | bg_list); | |
73 | list_del_init(&cache->bg_list); | |
74 | btrfs_put_block_group_trimming(cache); | |
75 | btrfs_put_block_group(cache); | |
76 | } | |
4b5faeac | 77 | kfree(transaction); |
78fae27e | 78 | } |
79154b1b CM |
79 | } |
80 | ||
663dfbb0 FM |
81 | static void clear_btree_io_tree(struct extent_io_tree *tree) |
82 | { | |
83 | spin_lock(&tree->lock); | |
b666a9cd DS |
84 | /* |
85 | * Do a single barrier for the waitqueue_active check here, the state | |
86 | * of the waitqueue should not change once clear_btree_io_tree is | |
87 | * called. | |
88 | */ | |
89 | smp_mb(); | |
663dfbb0 FM |
90 | while (!RB_EMPTY_ROOT(&tree->state)) { |
91 | struct rb_node *node; | |
92 | struct extent_state *state; | |
93 | ||
94 | node = rb_first(&tree->state); | |
95 | state = rb_entry(node, struct extent_state, rb_node); | |
96 | rb_erase(&state->rb_node, &tree->state); | |
97 | RB_CLEAR_NODE(&state->rb_node); | |
98 | /* | |
99 | * btree io trees aren't supposed to have tasks waiting for | |
100 | * changes in the flags of extent states ever. | |
101 | */ | |
102 | ASSERT(!waitqueue_active(&state->wq)); | |
103 | free_extent_state(state); | |
351810c1 DS |
104 | |
105 | cond_resched_lock(&tree->lock); | |
663dfbb0 FM |
106 | } |
107 | spin_unlock(&tree->lock); | |
108 | } | |
109 | ||
16916a88 | 110 | static noinline void switch_commit_roots(struct btrfs_transaction *trans) |
817d52f8 | 111 | { |
16916a88 | 112 | struct btrfs_fs_info *fs_info = trans->fs_info; |
9e351cc8 JB |
113 | struct btrfs_root *root, *tmp; |
114 | ||
115 | down_write(&fs_info->commit_root_sem); | |
116 | list_for_each_entry_safe(root, tmp, &trans->switch_commits, | |
117 | dirty_list) { | |
118 | list_del_init(&root->dirty_list); | |
119 | free_extent_buffer(root->commit_root); | |
120 | root->commit_root = btrfs_root_node(root); | |
121 | if (is_fstree(root->objectid)) | |
122 | btrfs_unpin_free_ino(root); | |
663dfbb0 | 123 | clear_btree_io_tree(&root->dirty_log_pages); |
9e351cc8 | 124 | } |
2b9dbef2 JB |
125 | |
126 | /* We can free old roots now. */ | |
127 | spin_lock(&trans->dropped_roots_lock); | |
128 | while (!list_empty(&trans->dropped_roots)) { | |
129 | root = list_first_entry(&trans->dropped_roots, | |
130 | struct btrfs_root, root_list); | |
131 | list_del_init(&root->root_list); | |
132 | spin_unlock(&trans->dropped_roots_lock); | |
133 | btrfs_drop_and_free_fs_root(fs_info, root); | |
134 | spin_lock(&trans->dropped_roots_lock); | |
135 | } | |
136 | spin_unlock(&trans->dropped_roots_lock); | |
9e351cc8 | 137 | up_write(&fs_info->commit_root_sem); |
817d52f8 JB |
138 | } |
139 | ||
0860adfd MX |
140 | static inline void extwriter_counter_inc(struct btrfs_transaction *trans, |
141 | unsigned int type) | |
142 | { | |
143 | if (type & TRANS_EXTWRITERS) | |
144 | atomic_inc(&trans->num_extwriters); | |
145 | } | |
146 | ||
147 | static inline void extwriter_counter_dec(struct btrfs_transaction *trans, | |
148 | unsigned int type) | |
149 | { | |
150 | if (type & TRANS_EXTWRITERS) | |
151 | atomic_dec(&trans->num_extwriters); | |
152 | } | |
153 | ||
154 | static inline void extwriter_counter_init(struct btrfs_transaction *trans, | |
155 | unsigned int type) | |
156 | { | |
157 | atomic_set(&trans->num_extwriters, ((type & TRANS_EXTWRITERS) ? 1 : 0)); | |
158 | } | |
159 | ||
160 | static inline int extwriter_counter_read(struct btrfs_transaction *trans) | |
161 | { | |
162 | return atomic_read(&trans->num_extwriters); | |
178260b2 MX |
163 | } |
164 | ||
d352ac68 CM |
165 | /* |
166 | * either allocate a new transaction or hop into the existing one | |
167 | */ | |
2ff7e61e JM |
168 | static noinline int join_transaction(struct btrfs_fs_info *fs_info, |
169 | unsigned int type) | |
79154b1b CM |
170 | { |
171 | struct btrfs_transaction *cur_trans; | |
a4abeea4 | 172 | |
19ae4e81 | 173 | spin_lock(&fs_info->trans_lock); |
d43317dc | 174 | loop: |
49b25e05 | 175 | /* The file system has been taken offline. No new transactions. */ |
87533c47 | 176 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { |
19ae4e81 | 177 | spin_unlock(&fs_info->trans_lock); |
49b25e05 JM |
178 | return -EROFS; |
179 | } | |
180 | ||
19ae4e81 | 181 | cur_trans = fs_info->running_transaction; |
a4abeea4 | 182 | if (cur_trans) { |
871383be | 183 | if (cur_trans->aborted) { |
19ae4e81 | 184 | spin_unlock(&fs_info->trans_lock); |
49b25e05 | 185 | return cur_trans->aborted; |
871383be | 186 | } |
4a9d8bde | 187 | if (btrfs_blocked_trans_types[cur_trans->state] & type) { |
178260b2 MX |
188 | spin_unlock(&fs_info->trans_lock); |
189 | return -EBUSY; | |
190 | } | |
9b64f57d | 191 | refcount_inc(&cur_trans->use_count); |
13c5a93e | 192 | atomic_inc(&cur_trans->num_writers); |
0860adfd | 193 | extwriter_counter_inc(cur_trans, type); |
19ae4e81 | 194 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 | 195 | return 0; |
79154b1b | 196 | } |
19ae4e81 | 197 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 | 198 | |
354aa0fb MX |
199 | /* |
200 | * If we are ATTACH, we just want to catch the current transaction, | |
201 | * and commit it. If there is no transaction, just return ENOENT. | |
202 | */ | |
203 | if (type == TRANS_ATTACH) | |
204 | return -ENOENT; | |
205 | ||
4a9d8bde MX |
206 | /* |
207 | * JOIN_NOLOCK only happens during the transaction commit, so | |
208 | * it is impossible that ->running_transaction is NULL | |
209 | */ | |
210 | BUG_ON(type == TRANS_JOIN_NOLOCK); | |
211 | ||
4b5faeac | 212 | cur_trans = kmalloc(sizeof(*cur_trans), GFP_NOFS); |
a4abeea4 JB |
213 | if (!cur_trans) |
214 | return -ENOMEM; | |
d43317dc | 215 | |
19ae4e81 JS |
216 | spin_lock(&fs_info->trans_lock); |
217 | if (fs_info->running_transaction) { | |
d43317dc CM |
218 | /* |
219 | * someone started a transaction after we unlocked. Make sure | |
4a9d8bde | 220 | * to redo the checks above |
d43317dc | 221 | */ |
4b5faeac | 222 | kfree(cur_trans); |
d43317dc | 223 | goto loop; |
87533c47 | 224 | } else if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { |
e4b50e14 | 225 | spin_unlock(&fs_info->trans_lock); |
4b5faeac | 226 | kfree(cur_trans); |
7b8b92af | 227 | return -EROFS; |
79154b1b | 228 | } |
d43317dc | 229 | |
ab8d0fc4 | 230 | cur_trans->fs_info = fs_info; |
a4abeea4 | 231 | atomic_set(&cur_trans->num_writers, 1); |
0860adfd | 232 | extwriter_counter_init(cur_trans, type); |
a4abeea4 JB |
233 | init_waitqueue_head(&cur_trans->writer_wait); |
234 | init_waitqueue_head(&cur_trans->commit_wait); | |
161c3549 | 235 | init_waitqueue_head(&cur_trans->pending_wait); |
4a9d8bde | 236 | cur_trans->state = TRANS_STATE_RUNNING; |
a4abeea4 JB |
237 | /* |
238 | * One for this trans handle, one so it will live on until we | |
239 | * commit the transaction. | |
240 | */ | |
9b64f57d | 241 | refcount_set(&cur_trans->use_count, 2); |
161c3549 | 242 | atomic_set(&cur_trans->pending_ordered, 0); |
3204d33c | 243 | cur_trans->flags = 0; |
afd48513 | 244 | cur_trans->start_time = ktime_get_seconds(); |
a4abeea4 | 245 | |
a099d0fd AM |
246 | memset(&cur_trans->delayed_refs, 0, sizeof(cur_trans->delayed_refs)); |
247 | ||
c46effa6 | 248 | cur_trans->delayed_refs.href_root = RB_ROOT; |
3368d001 | 249 | cur_trans->delayed_refs.dirty_extent_root = RB_ROOT; |
d7df2c79 | 250 | atomic_set(&cur_trans->delayed_refs.num_entries, 0); |
20b297d6 JS |
251 | |
252 | /* | |
253 | * although the tree mod log is per file system and not per transaction, | |
254 | * the log must never go across transaction boundaries. | |
255 | */ | |
256 | smp_mb(); | |
31b1a2bd | 257 | if (!list_empty(&fs_info->tree_mod_seq_list)) |
5d163e0e | 258 | WARN(1, KERN_ERR "BTRFS: tree_mod_seq_list not empty when creating a fresh transaction\n"); |
31b1a2bd | 259 | if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log)) |
5d163e0e | 260 | WARN(1, KERN_ERR "BTRFS: tree_mod_log rb tree not empty when creating a fresh transaction\n"); |
fc36ed7e | 261 | atomic64_set(&fs_info->tree_mod_seq, 0); |
20b297d6 | 262 | |
a4abeea4 JB |
263 | spin_lock_init(&cur_trans->delayed_refs.lock); |
264 | ||
265 | INIT_LIST_HEAD(&cur_trans->pending_snapshots); | |
6df9a95e | 266 | INIT_LIST_HEAD(&cur_trans->pending_chunks); |
9e351cc8 | 267 | INIT_LIST_HEAD(&cur_trans->switch_commits); |
ce93ec54 | 268 | INIT_LIST_HEAD(&cur_trans->dirty_bgs); |
1bbc621e | 269 | INIT_LIST_HEAD(&cur_trans->io_bgs); |
2b9dbef2 | 270 | INIT_LIST_HEAD(&cur_trans->dropped_roots); |
1bbc621e | 271 | mutex_init(&cur_trans->cache_write_mutex); |
cb723e49 | 272 | cur_trans->num_dirty_bgs = 0; |
ce93ec54 | 273 | spin_lock_init(&cur_trans->dirty_bgs_lock); |
e33e17ee | 274 | INIT_LIST_HEAD(&cur_trans->deleted_bgs); |
2b9dbef2 | 275 | spin_lock_init(&cur_trans->dropped_roots_lock); |
19ae4e81 | 276 | list_add_tail(&cur_trans->list, &fs_info->trans_list); |
a4abeea4 | 277 | extent_io_tree_init(&cur_trans->dirty_pages, |
c6100a4b | 278 | fs_info->btree_inode); |
19ae4e81 JS |
279 | fs_info->generation++; |
280 | cur_trans->transid = fs_info->generation; | |
281 | fs_info->running_transaction = cur_trans; | |
49b25e05 | 282 | cur_trans->aborted = 0; |
19ae4e81 | 283 | spin_unlock(&fs_info->trans_lock); |
15ee9bc7 | 284 | |
79154b1b CM |
285 | return 0; |
286 | } | |
287 | ||
d352ac68 | 288 | /* |
d397712b CM |
289 | * this does all the record keeping required to make sure that a reference |
290 | * counted root is properly recorded in a given transaction. This is required | |
291 | * to make sure the old root from before we joined the transaction is deleted | |
292 | * when the transaction commits | |
d352ac68 | 293 | */ |
7585717f | 294 | static int record_root_in_trans(struct btrfs_trans_handle *trans, |
6426c7ad QW |
295 | struct btrfs_root *root, |
296 | int force) | |
6702ed49 | 297 | { |
0b246afa JM |
298 | struct btrfs_fs_info *fs_info = root->fs_info; |
299 | ||
6426c7ad QW |
300 | if ((test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
301 | root->last_trans < trans->transid) || force) { | |
0b246afa | 302 | WARN_ON(root == fs_info->extent_root); |
4d31778a | 303 | WARN_ON(!force && root->commit_root != root->node); |
5d4f98a2 | 304 | |
7585717f | 305 | /* |
27cdeb70 | 306 | * see below for IN_TRANS_SETUP usage rules |
7585717f CM |
307 | * we have the reloc mutex held now, so there |
308 | * is only one writer in this function | |
309 | */ | |
27cdeb70 | 310 | set_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); |
7585717f | 311 | |
27cdeb70 | 312 | /* make sure readers find IN_TRANS_SETUP before |
7585717f CM |
313 | * they find our root->last_trans update |
314 | */ | |
315 | smp_wmb(); | |
316 | ||
0b246afa | 317 | spin_lock(&fs_info->fs_roots_radix_lock); |
6426c7ad | 318 | if (root->last_trans == trans->transid && !force) { |
0b246afa | 319 | spin_unlock(&fs_info->fs_roots_radix_lock); |
a4abeea4 JB |
320 | return 0; |
321 | } | |
0b246afa JM |
322 | radix_tree_tag_set(&fs_info->fs_roots_radix, |
323 | (unsigned long)root->root_key.objectid, | |
324 | BTRFS_ROOT_TRANS_TAG); | |
325 | spin_unlock(&fs_info->fs_roots_radix_lock); | |
7585717f CM |
326 | root->last_trans = trans->transid; |
327 | ||
328 | /* this is pretty tricky. We don't want to | |
329 | * take the relocation lock in btrfs_record_root_in_trans | |
330 | * unless we're really doing the first setup for this root in | |
331 | * this transaction. | |
332 | * | |
333 | * Normally we'd use root->last_trans as a flag to decide | |
334 | * if we want to take the expensive mutex. | |
335 | * | |
336 | * But, we have to set root->last_trans before we | |
337 | * init the relocation root, otherwise, we trip over warnings | |
338 | * in ctree.c. The solution used here is to flag ourselves | |
27cdeb70 | 339 | * with root IN_TRANS_SETUP. When this is 1, we're still |
7585717f CM |
340 | * fixing up the reloc trees and everyone must wait. |
341 | * | |
342 | * When this is zero, they can trust root->last_trans and fly | |
343 | * through btrfs_record_root_in_trans without having to take the | |
344 | * lock. smp_wmb() makes sure that all the writes above are | |
345 | * done before we pop in the zero below | |
346 | */ | |
5d4f98a2 | 347 | btrfs_init_reloc_root(trans, root); |
c7548af6 | 348 | smp_mb__before_atomic(); |
27cdeb70 | 349 | clear_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); |
5d4f98a2 YZ |
350 | } |
351 | return 0; | |
352 | } | |
bcc63abb | 353 | |
7585717f | 354 | |
2b9dbef2 JB |
355 | void btrfs_add_dropped_root(struct btrfs_trans_handle *trans, |
356 | struct btrfs_root *root) | |
357 | { | |
0b246afa | 358 | struct btrfs_fs_info *fs_info = root->fs_info; |
2b9dbef2 JB |
359 | struct btrfs_transaction *cur_trans = trans->transaction; |
360 | ||
361 | /* Add ourselves to the transaction dropped list */ | |
362 | spin_lock(&cur_trans->dropped_roots_lock); | |
363 | list_add_tail(&root->root_list, &cur_trans->dropped_roots); | |
364 | spin_unlock(&cur_trans->dropped_roots_lock); | |
365 | ||
366 | /* Make sure we don't try to update the root at commit time */ | |
0b246afa JM |
367 | spin_lock(&fs_info->fs_roots_radix_lock); |
368 | radix_tree_tag_clear(&fs_info->fs_roots_radix, | |
2b9dbef2 JB |
369 | (unsigned long)root->root_key.objectid, |
370 | BTRFS_ROOT_TRANS_TAG); | |
0b246afa | 371 | spin_unlock(&fs_info->fs_roots_radix_lock); |
2b9dbef2 JB |
372 | } |
373 | ||
7585717f CM |
374 | int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, |
375 | struct btrfs_root *root) | |
376 | { | |
0b246afa JM |
377 | struct btrfs_fs_info *fs_info = root->fs_info; |
378 | ||
27cdeb70 | 379 | if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state)) |
7585717f CM |
380 | return 0; |
381 | ||
382 | /* | |
27cdeb70 | 383 | * see record_root_in_trans for comments about IN_TRANS_SETUP usage |
7585717f CM |
384 | * and barriers |
385 | */ | |
386 | smp_rmb(); | |
387 | if (root->last_trans == trans->transid && | |
27cdeb70 | 388 | !test_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state)) |
7585717f CM |
389 | return 0; |
390 | ||
0b246afa | 391 | mutex_lock(&fs_info->reloc_mutex); |
6426c7ad | 392 | record_root_in_trans(trans, root, 0); |
0b246afa | 393 | mutex_unlock(&fs_info->reloc_mutex); |
7585717f CM |
394 | |
395 | return 0; | |
396 | } | |
397 | ||
4a9d8bde MX |
398 | static inline int is_transaction_blocked(struct btrfs_transaction *trans) |
399 | { | |
400 | return (trans->state >= TRANS_STATE_BLOCKED && | |
501407aa JB |
401 | trans->state < TRANS_STATE_UNBLOCKED && |
402 | !trans->aborted); | |
4a9d8bde MX |
403 | } |
404 | ||
d352ac68 CM |
405 | /* wait for commit against the current transaction to become unblocked |
406 | * when this is done, it is safe to start a new transaction, but the current | |
407 | * transaction might not be fully on disk. | |
408 | */ | |
2ff7e61e | 409 | static void wait_current_trans(struct btrfs_fs_info *fs_info) |
79154b1b | 410 | { |
f9295749 | 411 | struct btrfs_transaction *cur_trans; |
79154b1b | 412 | |
0b246afa JM |
413 | spin_lock(&fs_info->trans_lock); |
414 | cur_trans = fs_info->running_transaction; | |
4a9d8bde | 415 | if (cur_trans && is_transaction_blocked(cur_trans)) { |
9b64f57d | 416 | refcount_inc(&cur_trans->use_count); |
0b246afa | 417 | spin_unlock(&fs_info->trans_lock); |
72d63ed6 | 418 | |
0b246afa | 419 | wait_event(fs_info->transaction_wait, |
501407aa JB |
420 | cur_trans->state >= TRANS_STATE_UNBLOCKED || |
421 | cur_trans->aborted); | |
724e2315 | 422 | btrfs_put_transaction(cur_trans); |
a4abeea4 | 423 | } else { |
0b246afa | 424 | spin_unlock(&fs_info->trans_lock); |
f9295749 | 425 | } |
37d1aeee CM |
426 | } |
427 | ||
2ff7e61e | 428 | static int may_wait_transaction(struct btrfs_fs_info *fs_info, int type) |
a22285a6 | 429 | { |
0b246afa | 430 | if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) |
a4abeea4 JB |
431 | return 0; |
432 | ||
92e2f7e3 | 433 | if (type == TRANS_START) |
a22285a6 | 434 | return 1; |
a4abeea4 | 435 | |
a22285a6 YZ |
436 | return 0; |
437 | } | |
438 | ||
20dd2cbf MX |
439 | static inline bool need_reserve_reloc_root(struct btrfs_root *root) |
440 | { | |
0b246afa JM |
441 | struct btrfs_fs_info *fs_info = root->fs_info; |
442 | ||
443 | if (!fs_info->reloc_ctl || | |
27cdeb70 | 444 | !test_bit(BTRFS_ROOT_REF_COWS, &root->state) || |
20dd2cbf MX |
445 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || |
446 | root->reloc_root) | |
447 | return false; | |
448 | ||
449 | return true; | |
450 | } | |
451 | ||
08e007d2 | 452 | static struct btrfs_trans_handle * |
5aed1dd8 | 453 | start_transaction(struct btrfs_root *root, unsigned int num_items, |
003d7c59 JM |
454 | unsigned int type, enum btrfs_reserve_flush_enum flush, |
455 | bool enforce_qgroups) | |
37d1aeee | 456 | { |
0b246afa JM |
457 | struct btrfs_fs_info *fs_info = root->fs_info; |
458 | ||
a22285a6 YZ |
459 | struct btrfs_trans_handle *h; |
460 | struct btrfs_transaction *cur_trans; | |
b5009945 | 461 | u64 num_bytes = 0; |
c5567237 | 462 | u64 qgroup_reserved = 0; |
20dd2cbf MX |
463 | bool reloc_reserved = false; |
464 | int ret; | |
acce952b | 465 | |
46c4e71e | 466 | /* Send isn't supposed to start transactions. */ |
2755a0de | 467 | ASSERT(current->journal_info != BTRFS_SEND_TRANS_STUB); |
46c4e71e | 468 | |
0b246afa | 469 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) |
acce952b | 470 | return ERR_PTR(-EROFS); |
2a1eb461 | 471 | |
46c4e71e | 472 | if (current->journal_info) { |
0860adfd | 473 | WARN_ON(type & TRANS_EXTWRITERS); |
2a1eb461 | 474 | h = current->journal_info; |
b50fff81 DS |
475 | refcount_inc(&h->use_count); |
476 | WARN_ON(refcount_read(&h->use_count) > 2); | |
2a1eb461 JB |
477 | h->orig_rsv = h->block_rsv; |
478 | h->block_rsv = NULL; | |
479 | goto got_it; | |
480 | } | |
b5009945 JB |
481 | |
482 | /* | |
483 | * Do the reservation before we join the transaction so we can do all | |
484 | * the appropriate flushing if need be. | |
485 | */ | |
003d7c59 | 486 | if (num_items && root != fs_info->chunk_root) { |
0b246afa | 487 | qgroup_reserved = num_items * fs_info->nodesize; |
733e03a0 QW |
488 | ret = btrfs_qgroup_reserve_meta_pertrans(root, qgroup_reserved, |
489 | enforce_qgroups); | |
7174109c QW |
490 | if (ret) |
491 | return ERR_PTR(ret); | |
c5567237 | 492 | |
0b246afa | 493 | num_bytes = btrfs_calc_trans_metadata_size(fs_info, num_items); |
20dd2cbf MX |
494 | /* |
495 | * Do the reservation for the relocation root creation | |
496 | */ | |
ee39b432 | 497 | if (need_reserve_reloc_root(root)) { |
0b246afa | 498 | num_bytes += fs_info->nodesize; |
20dd2cbf MX |
499 | reloc_reserved = true; |
500 | } | |
501 | ||
0b246afa | 502 | ret = btrfs_block_rsv_add(root, &fs_info->trans_block_rsv, |
08e007d2 | 503 | num_bytes, flush); |
b5009945 | 504 | if (ret) |
843fcf35 | 505 | goto reserve_fail; |
b5009945 | 506 | } |
a22285a6 | 507 | again: |
f2f767e7 | 508 | h = kmem_cache_zalloc(btrfs_trans_handle_cachep, GFP_NOFS); |
843fcf35 MX |
509 | if (!h) { |
510 | ret = -ENOMEM; | |
511 | goto alloc_fail; | |
512 | } | |
37d1aeee | 513 | |
98114659 JB |
514 | /* |
515 | * If we are JOIN_NOLOCK we're already committing a transaction and | |
516 | * waiting on this guy, so we don't need to do the sb_start_intwrite | |
517 | * because we're already holding a ref. We need this because we could | |
518 | * have raced in and did an fsync() on a file which can kick a commit | |
519 | * and then we deadlock with somebody doing a freeze. | |
354aa0fb MX |
520 | * |
521 | * If we are ATTACH, it means we just want to catch the current | |
522 | * transaction and commit it, so we needn't do sb_start_intwrite(). | |
98114659 | 523 | */ |
0860adfd | 524 | if (type & __TRANS_FREEZABLE) |
0b246afa | 525 | sb_start_intwrite(fs_info->sb); |
b2b5ef5c | 526 | |
2ff7e61e JM |
527 | if (may_wait_transaction(fs_info, type)) |
528 | wait_current_trans(fs_info); | |
a22285a6 | 529 | |
a4abeea4 | 530 | do { |
2ff7e61e | 531 | ret = join_transaction(fs_info, type); |
178260b2 | 532 | if (ret == -EBUSY) { |
2ff7e61e | 533 | wait_current_trans(fs_info); |
178260b2 MX |
534 | if (unlikely(type == TRANS_ATTACH)) |
535 | ret = -ENOENT; | |
536 | } | |
a4abeea4 JB |
537 | } while (ret == -EBUSY); |
538 | ||
a43f7f82 | 539 | if (ret < 0) |
843fcf35 | 540 | goto join_fail; |
0f7d52f4 | 541 | |
0b246afa | 542 | cur_trans = fs_info->running_transaction; |
a22285a6 YZ |
543 | |
544 | h->transid = cur_trans->transid; | |
545 | h->transaction = cur_trans; | |
d13603ef | 546 | h->root = root; |
b50fff81 | 547 | refcount_set(&h->use_count, 1); |
64b63580 | 548 | h->fs_info = root->fs_info; |
7174109c | 549 | |
a698d075 | 550 | h->type = type; |
d9a0540a | 551 | h->can_flush_pending_bgs = true; |
ea658bad | 552 | INIT_LIST_HEAD(&h->new_bgs); |
b7ec40d7 | 553 | |
a22285a6 | 554 | smp_mb(); |
4a9d8bde | 555 | if (cur_trans->state >= TRANS_STATE_BLOCKED && |
2ff7e61e | 556 | may_wait_transaction(fs_info, type)) { |
abdd2e80 | 557 | current->journal_info = h; |
3a45bb20 | 558 | btrfs_commit_transaction(h); |
a22285a6 YZ |
559 | goto again; |
560 | } | |
561 | ||
b5009945 | 562 | if (num_bytes) { |
0b246afa | 563 | trace_btrfs_space_reservation(fs_info, "transaction", |
2bcc0328 | 564 | h->transid, num_bytes, 1); |
0b246afa | 565 | h->block_rsv = &fs_info->trans_block_rsv; |
b5009945 | 566 | h->bytes_reserved = num_bytes; |
20dd2cbf | 567 | h->reloc_reserved = reloc_reserved; |
a22285a6 | 568 | } |
9ed74f2d | 569 | |
2a1eb461 | 570 | got_it: |
a4abeea4 | 571 | btrfs_record_root_in_trans(h, root); |
a22285a6 | 572 | |
bcf3a3e7 | 573 | if (!current->journal_info) |
a22285a6 | 574 | current->journal_info = h; |
79154b1b | 575 | return h; |
843fcf35 MX |
576 | |
577 | join_fail: | |
0860adfd | 578 | if (type & __TRANS_FREEZABLE) |
0b246afa | 579 | sb_end_intwrite(fs_info->sb); |
843fcf35 MX |
580 | kmem_cache_free(btrfs_trans_handle_cachep, h); |
581 | alloc_fail: | |
582 | if (num_bytes) | |
2ff7e61e | 583 | btrfs_block_rsv_release(fs_info, &fs_info->trans_block_rsv, |
843fcf35 MX |
584 | num_bytes); |
585 | reserve_fail: | |
733e03a0 | 586 | btrfs_qgroup_free_meta_pertrans(root, qgroup_reserved); |
843fcf35 | 587 | return ERR_PTR(ret); |
79154b1b CM |
588 | } |
589 | ||
f9295749 | 590 | struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, |
5aed1dd8 | 591 | unsigned int num_items) |
f9295749 | 592 | { |
08e007d2 | 593 | return start_transaction(root, num_items, TRANS_START, |
003d7c59 | 594 | BTRFS_RESERVE_FLUSH_ALL, true); |
f9295749 | 595 | } |
003d7c59 | 596 | |
8eab77ff FM |
597 | struct btrfs_trans_handle *btrfs_start_transaction_fallback_global_rsv( |
598 | struct btrfs_root *root, | |
599 | unsigned int num_items, | |
600 | int min_factor) | |
601 | { | |
0b246afa | 602 | struct btrfs_fs_info *fs_info = root->fs_info; |
8eab77ff FM |
603 | struct btrfs_trans_handle *trans; |
604 | u64 num_bytes; | |
605 | int ret; | |
606 | ||
003d7c59 JM |
607 | /* |
608 | * We have two callers: unlink and block group removal. The | |
609 | * former should succeed even if we will temporarily exceed | |
610 | * quota and the latter operates on the extent root so | |
611 | * qgroup enforcement is ignored anyway. | |
612 | */ | |
613 | trans = start_transaction(root, num_items, TRANS_START, | |
614 | BTRFS_RESERVE_FLUSH_ALL, false); | |
8eab77ff FM |
615 | if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC) |
616 | return trans; | |
617 | ||
618 | trans = btrfs_start_transaction(root, 0); | |
619 | if (IS_ERR(trans)) | |
620 | return trans; | |
621 | ||
0b246afa JM |
622 | num_bytes = btrfs_calc_trans_metadata_size(fs_info, num_items); |
623 | ret = btrfs_cond_migrate_bytes(fs_info, &fs_info->trans_block_rsv, | |
624 | num_bytes, min_factor); | |
8eab77ff | 625 | if (ret) { |
3a45bb20 | 626 | btrfs_end_transaction(trans); |
8eab77ff FM |
627 | return ERR_PTR(ret); |
628 | } | |
629 | ||
0b246afa | 630 | trans->block_rsv = &fs_info->trans_block_rsv; |
8eab77ff | 631 | trans->bytes_reserved = num_bytes; |
0b246afa | 632 | trace_btrfs_space_reservation(fs_info, "transaction", |
88d3a5aa | 633 | trans->transid, num_bytes, 1); |
8eab77ff FM |
634 | |
635 | return trans; | |
636 | } | |
8407aa46 | 637 | |
7a7eaa40 | 638 | struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root) |
f9295749 | 639 | { |
003d7c59 JM |
640 | return start_transaction(root, 0, TRANS_JOIN, BTRFS_RESERVE_NO_FLUSH, |
641 | true); | |
f9295749 CM |
642 | } |
643 | ||
7a7eaa40 | 644 | struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root) |
0af3d00b | 645 | { |
575a75d6 | 646 | return start_transaction(root, 0, TRANS_JOIN_NOLOCK, |
003d7c59 | 647 | BTRFS_RESERVE_NO_FLUSH, true); |
0af3d00b JB |
648 | } |
649 | ||
d4edf39b MX |
650 | /* |
651 | * btrfs_attach_transaction() - catch the running transaction | |
652 | * | |
653 | * It is used when we want to commit the current the transaction, but | |
654 | * don't want to start a new one. | |
655 | * | |
656 | * Note: If this function return -ENOENT, it just means there is no | |
657 | * running transaction. But it is possible that the inactive transaction | |
658 | * is still in the memory, not fully on disk. If you hope there is no | |
659 | * inactive transaction in the fs when -ENOENT is returned, you should | |
660 | * invoke | |
661 | * btrfs_attach_transaction_barrier() | |
662 | */ | |
354aa0fb | 663 | struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root) |
60376ce4 | 664 | { |
575a75d6 | 665 | return start_transaction(root, 0, TRANS_ATTACH, |
003d7c59 | 666 | BTRFS_RESERVE_NO_FLUSH, true); |
60376ce4 JB |
667 | } |
668 | ||
d4edf39b | 669 | /* |
90b6d283 | 670 | * btrfs_attach_transaction_barrier() - catch the running transaction |
d4edf39b MX |
671 | * |
672 | * It is similar to the above function, the differentia is this one | |
673 | * will wait for all the inactive transactions until they fully | |
674 | * complete. | |
675 | */ | |
676 | struct btrfs_trans_handle * | |
677 | btrfs_attach_transaction_barrier(struct btrfs_root *root) | |
678 | { | |
679 | struct btrfs_trans_handle *trans; | |
680 | ||
575a75d6 | 681 | trans = start_transaction(root, 0, TRANS_ATTACH, |
003d7c59 | 682 | BTRFS_RESERVE_NO_FLUSH, true); |
d4edf39b | 683 | if (IS_ERR(trans) && PTR_ERR(trans) == -ENOENT) |
2ff7e61e | 684 | btrfs_wait_for_commit(root->fs_info, 0); |
d4edf39b MX |
685 | |
686 | return trans; | |
687 | } | |
688 | ||
d352ac68 | 689 | /* wait for a transaction commit to be fully complete */ |
2ff7e61e | 690 | static noinline void wait_for_commit(struct btrfs_transaction *commit) |
89ce8a63 | 691 | { |
4a9d8bde | 692 | wait_event(commit->commit_wait, commit->state == TRANS_STATE_COMPLETED); |
89ce8a63 CM |
693 | } |
694 | ||
2ff7e61e | 695 | int btrfs_wait_for_commit(struct btrfs_fs_info *fs_info, u64 transid) |
46204592 SW |
696 | { |
697 | struct btrfs_transaction *cur_trans = NULL, *t; | |
8cd2807f | 698 | int ret = 0; |
46204592 | 699 | |
46204592 | 700 | if (transid) { |
0b246afa | 701 | if (transid <= fs_info->last_trans_committed) |
a4abeea4 | 702 | goto out; |
46204592 SW |
703 | |
704 | /* find specified transaction */ | |
0b246afa JM |
705 | spin_lock(&fs_info->trans_lock); |
706 | list_for_each_entry(t, &fs_info->trans_list, list) { | |
46204592 SW |
707 | if (t->transid == transid) { |
708 | cur_trans = t; | |
9b64f57d | 709 | refcount_inc(&cur_trans->use_count); |
8cd2807f | 710 | ret = 0; |
46204592 SW |
711 | break; |
712 | } | |
8cd2807f MX |
713 | if (t->transid > transid) { |
714 | ret = 0; | |
46204592 | 715 | break; |
8cd2807f | 716 | } |
46204592 | 717 | } |
0b246afa | 718 | spin_unlock(&fs_info->trans_lock); |
42383020 SW |
719 | |
720 | /* | |
721 | * The specified transaction doesn't exist, or we | |
722 | * raced with btrfs_commit_transaction | |
723 | */ | |
724 | if (!cur_trans) { | |
0b246afa | 725 | if (transid > fs_info->last_trans_committed) |
42383020 | 726 | ret = -EINVAL; |
8cd2807f | 727 | goto out; |
42383020 | 728 | } |
46204592 SW |
729 | } else { |
730 | /* find newest transaction that is committing | committed */ | |
0b246afa JM |
731 | spin_lock(&fs_info->trans_lock); |
732 | list_for_each_entry_reverse(t, &fs_info->trans_list, | |
46204592 | 733 | list) { |
4a9d8bde MX |
734 | if (t->state >= TRANS_STATE_COMMIT_START) { |
735 | if (t->state == TRANS_STATE_COMPLETED) | |
3473f3c0 | 736 | break; |
46204592 | 737 | cur_trans = t; |
9b64f57d | 738 | refcount_inc(&cur_trans->use_count); |
46204592 SW |
739 | break; |
740 | } | |
741 | } | |
0b246afa | 742 | spin_unlock(&fs_info->trans_lock); |
46204592 | 743 | if (!cur_trans) |
a4abeea4 | 744 | goto out; /* nothing committing|committed */ |
46204592 SW |
745 | } |
746 | ||
2ff7e61e | 747 | wait_for_commit(cur_trans); |
724e2315 | 748 | btrfs_put_transaction(cur_trans); |
a4abeea4 | 749 | out: |
46204592 SW |
750 | return ret; |
751 | } | |
752 | ||
2ff7e61e | 753 | void btrfs_throttle(struct btrfs_fs_info *fs_info) |
37d1aeee | 754 | { |
92e2f7e3 | 755 | wait_current_trans(fs_info); |
37d1aeee CM |
756 | } |
757 | ||
2ff7e61e | 758 | static int should_end_transaction(struct btrfs_trans_handle *trans) |
8929ecfa | 759 | { |
2ff7e61e | 760 | struct btrfs_fs_info *fs_info = trans->fs_info; |
0b246afa | 761 | |
7c777430 | 762 | if (btrfs_check_space_for_delayed_refs(trans, fs_info)) |
1be41b78 | 763 | return 1; |
36ba022a | 764 | |
2ff7e61e | 765 | return !!btrfs_block_rsv_check(&fs_info->global_block_rsv, 5); |
8929ecfa YZ |
766 | } |
767 | ||
3a45bb20 | 768 | int btrfs_should_end_transaction(struct btrfs_trans_handle *trans) |
8929ecfa YZ |
769 | { |
770 | struct btrfs_transaction *cur_trans = trans->transaction; | |
771 | int updates; | |
49b25e05 | 772 | int err; |
8929ecfa | 773 | |
a4abeea4 | 774 | smp_mb(); |
4a9d8bde MX |
775 | if (cur_trans->state >= TRANS_STATE_BLOCKED || |
776 | cur_trans->delayed_refs.flushing) | |
8929ecfa YZ |
777 | return 1; |
778 | ||
779 | updates = trans->delayed_ref_updates; | |
780 | trans->delayed_ref_updates = 0; | |
49b25e05 | 781 | if (updates) { |
c79a70b1 | 782 | err = btrfs_run_delayed_refs(trans, updates * 2); |
49b25e05 JM |
783 | if (err) /* Error code will also eval true */ |
784 | return err; | |
785 | } | |
8929ecfa | 786 | |
2ff7e61e | 787 | return should_end_transaction(trans); |
8929ecfa YZ |
788 | } |
789 | ||
dc60c525 NB |
790 | static void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans) |
791 | ||
0e34693f | 792 | { |
dc60c525 NB |
793 | struct btrfs_fs_info *fs_info = trans->fs_info; |
794 | ||
0e34693f NB |
795 | if (!trans->block_rsv) { |
796 | ASSERT(!trans->bytes_reserved); | |
797 | return; | |
798 | } | |
799 | ||
800 | if (!trans->bytes_reserved) | |
801 | return; | |
802 | ||
803 | ASSERT(trans->block_rsv == &fs_info->trans_block_rsv); | |
804 | trace_btrfs_space_reservation(fs_info, "transaction", | |
805 | trans->transid, trans->bytes_reserved, 0); | |
806 | btrfs_block_rsv_release(fs_info, trans->block_rsv, | |
807 | trans->bytes_reserved); | |
808 | trans->bytes_reserved = 0; | |
809 | } | |
810 | ||
89ce8a63 | 811 | static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, |
3a45bb20 | 812 | int throttle) |
79154b1b | 813 | { |
3a45bb20 | 814 | struct btrfs_fs_info *info = trans->fs_info; |
8929ecfa | 815 | struct btrfs_transaction *cur_trans = trans->transaction; |
31b9655f | 816 | u64 transid = trans->transid; |
1be41b78 | 817 | unsigned long cur = trans->delayed_ref_updates; |
a698d075 | 818 | int lock = (trans->type != TRANS_JOIN_NOLOCK); |
4edc2ca3 | 819 | int err = 0; |
a79b7d4b | 820 | int must_run_delayed_refs = 0; |
c3e69d58 | 821 | |
b50fff81 DS |
822 | if (refcount_read(&trans->use_count) > 1) { |
823 | refcount_dec(&trans->use_count); | |
2a1eb461 JB |
824 | trans->block_rsv = trans->orig_rsv; |
825 | return 0; | |
826 | } | |
827 | ||
dc60c525 | 828 | btrfs_trans_release_metadata(trans); |
4c13d758 | 829 | trans->block_rsv = NULL; |
c5567237 | 830 | |
ea658bad | 831 | if (!list_empty(&trans->new_bgs)) |
6c686b35 | 832 | btrfs_create_pending_block_groups(trans); |
ea658bad | 833 | |
1be41b78 | 834 | trans->delayed_ref_updates = 0; |
a79b7d4b CM |
835 | if (!trans->sync) { |
836 | must_run_delayed_refs = | |
2ff7e61e | 837 | btrfs_should_throttle_delayed_refs(trans, info); |
0a2b2a84 | 838 | cur = max_t(unsigned long, cur, 32); |
a79b7d4b CM |
839 | |
840 | /* | |
841 | * don't make the caller wait if they are from a NOLOCK | |
842 | * or ATTACH transaction, it will deadlock with commit | |
843 | */ | |
844 | if (must_run_delayed_refs == 1 && | |
845 | (trans->type & (__TRANS_JOIN_NOLOCK | __TRANS_ATTACH))) | |
846 | must_run_delayed_refs = 2; | |
56bec294 | 847 | } |
bb721703 | 848 | |
dc60c525 | 849 | btrfs_trans_release_metadata(trans); |
0e721106 | 850 | trans->block_rsv = NULL; |
56bec294 | 851 | |
ea658bad | 852 | if (!list_empty(&trans->new_bgs)) |
6c686b35 | 853 | btrfs_create_pending_block_groups(trans); |
ea658bad | 854 | |
4fbcdf66 FM |
855 | btrfs_trans_release_chunk_metadata(trans); |
856 | ||
92e2f7e3 | 857 | if (lock && should_end_transaction(trans) && |
20c7bcec | 858 | READ_ONCE(cur_trans->state) == TRANS_STATE_RUNNING) { |
4a9d8bde MX |
859 | spin_lock(&info->trans_lock); |
860 | if (cur_trans->state == TRANS_STATE_RUNNING) | |
861 | cur_trans->state = TRANS_STATE_BLOCKED; | |
862 | spin_unlock(&info->trans_lock); | |
a4abeea4 | 863 | } |
8929ecfa | 864 | |
20c7bcec | 865 | if (lock && READ_ONCE(cur_trans->state) == TRANS_STATE_BLOCKED) { |
3bbb24b2 | 866 | if (throttle) |
3a45bb20 | 867 | return btrfs_commit_transaction(trans); |
3bbb24b2 | 868 | else |
8929ecfa YZ |
869 | wake_up_process(info->transaction_kthread); |
870 | } | |
871 | ||
0860adfd | 872 | if (trans->type & __TRANS_FREEZABLE) |
0b246afa | 873 | sb_end_intwrite(info->sb); |
6df7881a | 874 | |
8929ecfa | 875 | WARN_ON(cur_trans != info->running_transaction); |
13c5a93e JB |
876 | WARN_ON(atomic_read(&cur_trans->num_writers) < 1); |
877 | atomic_dec(&cur_trans->num_writers); | |
0860adfd | 878 | extwriter_counter_dec(cur_trans, trans->type); |
89ce8a63 | 879 | |
093258e6 | 880 | cond_wake_up(&cur_trans->writer_wait); |
724e2315 | 881 | btrfs_put_transaction(cur_trans); |
9ed74f2d JB |
882 | |
883 | if (current->journal_info == trans) | |
884 | current->journal_info = NULL; | |
ab78c84d | 885 | |
24bbcf04 | 886 | if (throttle) |
2ff7e61e | 887 | btrfs_run_delayed_iputs(info); |
24bbcf04 | 888 | |
49b25e05 | 889 | if (trans->aborted || |
0b246afa | 890 | test_bit(BTRFS_FS_STATE_ERROR, &info->fs_state)) { |
4e121c06 | 891 | wake_up_process(info->transaction_kthread); |
4edc2ca3 | 892 | err = -EIO; |
4e121c06 | 893 | } |
49b25e05 | 894 | |
4edc2ca3 | 895 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
a79b7d4b | 896 | if (must_run_delayed_refs) { |
2ff7e61e | 897 | btrfs_async_run_delayed_refs(info, cur, transid, |
a79b7d4b CM |
898 | must_run_delayed_refs == 1); |
899 | } | |
4edc2ca3 | 900 | return err; |
79154b1b CM |
901 | } |
902 | ||
3a45bb20 | 903 | int btrfs_end_transaction(struct btrfs_trans_handle *trans) |
89ce8a63 | 904 | { |
3a45bb20 | 905 | return __btrfs_end_transaction(trans, 0); |
89ce8a63 CM |
906 | } |
907 | ||
3a45bb20 | 908 | int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans) |
89ce8a63 | 909 | { |
3a45bb20 | 910 | return __btrfs_end_transaction(trans, 1); |
16cdcec7 MX |
911 | } |
912 | ||
d352ac68 CM |
913 | /* |
914 | * when btree blocks are allocated, they have some corresponding bits set for | |
915 | * them in one of two extent_io trees. This is used to make sure all of | |
690587d1 | 916 | * those extents are sent to disk but does not wait on them |
d352ac68 | 917 | */ |
2ff7e61e | 918 | int btrfs_write_marked_extents(struct btrfs_fs_info *fs_info, |
8cef4e16 | 919 | struct extent_io_tree *dirty_pages, int mark) |
79154b1b | 920 | { |
777e6bd7 | 921 | int err = 0; |
7c4452b9 | 922 | int werr = 0; |
0b246afa | 923 | struct address_space *mapping = fs_info->btree_inode->i_mapping; |
e6138876 | 924 | struct extent_state *cached_state = NULL; |
777e6bd7 | 925 | u64 start = 0; |
5f39d397 | 926 | u64 end; |
7c4452b9 | 927 | |
6300463b | 928 | atomic_inc(&BTRFS_I(fs_info->btree_inode)->sync_writers); |
1728366e | 929 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
e6138876 | 930 | mark, &cached_state)) { |
663dfbb0 FM |
931 | bool wait_writeback = false; |
932 | ||
933 | err = convert_extent_bit(dirty_pages, start, end, | |
934 | EXTENT_NEED_WAIT, | |
210aa277 | 935 | mark, &cached_state); |
663dfbb0 FM |
936 | /* |
937 | * convert_extent_bit can return -ENOMEM, which is most of the | |
938 | * time a temporary error. So when it happens, ignore the error | |
939 | * and wait for writeback of this range to finish - because we | |
940 | * failed to set the bit EXTENT_NEED_WAIT for the range, a call | |
bf89d38f JM |
941 | * to __btrfs_wait_marked_extents() would not know that |
942 | * writeback for this range started and therefore wouldn't | |
943 | * wait for it to finish - we don't want to commit a | |
944 | * superblock that points to btree nodes/leafs for which | |
945 | * writeback hasn't finished yet (and without errors). | |
663dfbb0 FM |
946 | * We cleanup any entries left in the io tree when committing |
947 | * the transaction (through clear_btree_io_tree()). | |
948 | */ | |
949 | if (err == -ENOMEM) { | |
950 | err = 0; | |
951 | wait_writeback = true; | |
952 | } | |
953 | if (!err) | |
954 | err = filemap_fdatawrite_range(mapping, start, end); | |
1728366e JB |
955 | if (err) |
956 | werr = err; | |
663dfbb0 FM |
957 | else if (wait_writeback) |
958 | werr = filemap_fdatawait_range(mapping, start, end); | |
e38e2ed7 | 959 | free_extent_state(cached_state); |
663dfbb0 | 960 | cached_state = NULL; |
1728366e JB |
961 | cond_resched(); |
962 | start = end + 1; | |
7c4452b9 | 963 | } |
6300463b | 964 | atomic_dec(&BTRFS_I(fs_info->btree_inode)->sync_writers); |
690587d1 CM |
965 | return werr; |
966 | } | |
967 | ||
968 | /* | |
969 | * when btree blocks are allocated, they have some corresponding bits set for | |
970 | * them in one of two extent_io trees. This is used to make sure all of | |
971 | * those extents are on disk for transaction or log commit. We wait | |
972 | * on all the pages and clear them from the dirty pages state tree | |
973 | */ | |
bf89d38f JM |
974 | static int __btrfs_wait_marked_extents(struct btrfs_fs_info *fs_info, |
975 | struct extent_io_tree *dirty_pages) | |
690587d1 | 976 | { |
690587d1 CM |
977 | int err = 0; |
978 | int werr = 0; | |
0b246afa | 979 | struct address_space *mapping = fs_info->btree_inode->i_mapping; |
e6138876 | 980 | struct extent_state *cached_state = NULL; |
690587d1 CM |
981 | u64 start = 0; |
982 | u64 end; | |
777e6bd7 | 983 | |
1728366e | 984 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
e6138876 | 985 | EXTENT_NEED_WAIT, &cached_state)) { |
663dfbb0 FM |
986 | /* |
987 | * Ignore -ENOMEM errors returned by clear_extent_bit(). | |
988 | * When committing the transaction, we'll remove any entries | |
989 | * left in the io tree. For a log commit, we don't remove them | |
990 | * after committing the log because the tree can be accessed | |
991 | * concurrently - we do it only at transaction commit time when | |
992 | * it's safe to do it (through clear_btree_io_tree()). | |
993 | */ | |
994 | err = clear_extent_bit(dirty_pages, start, end, | |
ae0f1625 | 995 | EXTENT_NEED_WAIT, 0, 0, &cached_state); |
663dfbb0 FM |
996 | if (err == -ENOMEM) |
997 | err = 0; | |
998 | if (!err) | |
999 | err = filemap_fdatawait_range(mapping, start, end); | |
1728366e JB |
1000 | if (err) |
1001 | werr = err; | |
e38e2ed7 FM |
1002 | free_extent_state(cached_state); |
1003 | cached_state = NULL; | |
1728366e JB |
1004 | cond_resched(); |
1005 | start = end + 1; | |
777e6bd7 | 1006 | } |
7c4452b9 CM |
1007 | if (err) |
1008 | werr = err; | |
bf89d38f JM |
1009 | return werr; |
1010 | } | |
656f30db | 1011 | |
bf89d38f JM |
1012 | int btrfs_wait_extents(struct btrfs_fs_info *fs_info, |
1013 | struct extent_io_tree *dirty_pages) | |
1014 | { | |
1015 | bool errors = false; | |
1016 | int err; | |
656f30db | 1017 | |
bf89d38f JM |
1018 | err = __btrfs_wait_marked_extents(fs_info, dirty_pages); |
1019 | if (test_and_clear_bit(BTRFS_FS_BTREE_ERR, &fs_info->flags)) | |
1020 | errors = true; | |
1021 | ||
1022 | if (errors && !err) | |
1023 | err = -EIO; | |
1024 | return err; | |
1025 | } | |
656f30db | 1026 | |
bf89d38f JM |
1027 | int btrfs_wait_tree_log_extents(struct btrfs_root *log_root, int mark) |
1028 | { | |
1029 | struct btrfs_fs_info *fs_info = log_root->fs_info; | |
1030 | struct extent_io_tree *dirty_pages = &log_root->dirty_log_pages; | |
1031 | bool errors = false; | |
1032 | int err; | |
656f30db | 1033 | |
bf89d38f JM |
1034 | ASSERT(log_root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID); |
1035 | ||
1036 | err = __btrfs_wait_marked_extents(fs_info, dirty_pages); | |
1037 | if ((mark & EXTENT_DIRTY) && | |
1038 | test_and_clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags)) | |
1039 | errors = true; | |
1040 | ||
1041 | if ((mark & EXTENT_NEW) && | |
1042 | test_and_clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags)) | |
1043 | errors = true; | |
1044 | ||
1045 | if (errors && !err) | |
1046 | err = -EIO; | |
1047 | return err; | |
79154b1b CM |
1048 | } |
1049 | ||
690587d1 | 1050 | /* |
c9b577c0 NB |
1051 | * When btree blocks are allocated the corresponding extents are marked dirty. |
1052 | * This function ensures such extents are persisted on disk for transaction or | |
1053 | * log commit. | |
1054 | * | |
1055 | * @trans: transaction whose dirty pages we'd like to write | |
690587d1 | 1056 | */ |
70458a58 | 1057 | static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans) |
690587d1 CM |
1058 | { |
1059 | int ret; | |
1060 | int ret2; | |
c9b577c0 | 1061 | struct extent_io_tree *dirty_pages = &trans->transaction->dirty_pages; |
70458a58 | 1062 | struct btrfs_fs_info *fs_info = trans->fs_info; |
c6adc9cc | 1063 | struct blk_plug plug; |
690587d1 | 1064 | |
c6adc9cc | 1065 | blk_start_plug(&plug); |
c9b577c0 | 1066 | ret = btrfs_write_marked_extents(fs_info, dirty_pages, EXTENT_DIRTY); |
c6adc9cc | 1067 | blk_finish_plug(&plug); |
bf89d38f | 1068 | ret2 = btrfs_wait_extents(fs_info, dirty_pages); |
bf0da8c1 | 1069 | |
c9b577c0 NB |
1070 | clear_btree_io_tree(&trans->transaction->dirty_pages); |
1071 | ||
bf0da8c1 CM |
1072 | if (ret) |
1073 | return ret; | |
c9b577c0 | 1074 | else if (ret2) |
bf0da8c1 | 1075 | return ret2; |
c9b577c0 NB |
1076 | else |
1077 | return 0; | |
d0c803c4 CM |
1078 | } |
1079 | ||
d352ac68 CM |
1080 | /* |
1081 | * this is used to update the root pointer in the tree of tree roots. | |
1082 | * | |
1083 | * But, in the case of the extent allocation tree, updating the root | |
1084 | * pointer may allocate blocks which may change the root of the extent | |
1085 | * allocation tree. | |
1086 | * | |
1087 | * So, this loops and repeats and makes sure the cowonly root didn't | |
1088 | * change while the root pointer was being updated in the metadata. | |
1089 | */ | |
0b86a832 CM |
1090 | static int update_cowonly_root(struct btrfs_trans_handle *trans, |
1091 | struct btrfs_root *root) | |
79154b1b CM |
1092 | { |
1093 | int ret; | |
0b86a832 | 1094 | u64 old_root_bytenr; |
86b9f2ec | 1095 | u64 old_root_used; |
0b246afa JM |
1096 | struct btrfs_fs_info *fs_info = root->fs_info; |
1097 | struct btrfs_root *tree_root = fs_info->tree_root; | |
79154b1b | 1098 | |
86b9f2ec | 1099 | old_root_used = btrfs_root_used(&root->root_item); |
56bec294 | 1100 | |
d397712b | 1101 | while (1) { |
0b86a832 | 1102 | old_root_bytenr = btrfs_root_bytenr(&root->root_item); |
86b9f2ec | 1103 | if (old_root_bytenr == root->node->start && |
ea526d18 | 1104 | old_root_used == btrfs_root_used(&root->root_item)) |
79154b1b | 1105 | break; |
87ef2bb4 | 1106 | |
5d4f98a2 | 1107 | btrfs_set_root_node(&root->root_item, root->node); |
79154b1b | 1108 | ret = btrfs_update_root(trans, tree_root, |
0b86a832 CM |
1109 | &root->root_key, |
1110 | &root->root_item); | |
49b25e05 JM |
1111 | if (ret) |
1112 | return ret; | |
56bec294 | 1113 | |
86b9f2ec | 1114 | old_root_used = btrfs_root_used(&root->root_item); |
0b86a832 | 1115 | } |
276e680d | 1116 | |
0b86a832 CM |
1117 | return 0; |
1118 | } | |
1119 | ||
d352ac68 CM |
1120 | /* |
1121 | * update all the cowonly tree roots on disk | |
49b25e05 JM |
1122 | * |
1123 | * The error handling in this function may not be obvious. Any of the | |
1124 | * failures will cause the file system to go offline. We still need | |
1125 | * to clean up the delayed refs. | |
d352ac68 | 1126 | */ |
9386d8bc | 1127 | static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans) |
0b86a832 | 1128 | { |
9386d8bc | 1129 | struct btrfs_fs_info *fs_info = trans->fs_info; |
ea526d18 | 1130 | struct list_head *dirty_bgs = &trans->transaction->dirty_bgs; |
1bbc621e | 1131 | struct list_head *io_bgs = &trans->transaction->io_bgs; |
0b86a832 | 1132 | struct list_head *next; |
84234f3a | 1133 | struct extent_buffer *eb; |
56bec294 | 1134 | int ret; |
84234f3a YZ |
1135 | |
1136 | eb = btrfs_lock_root_node(fs_info->tree_root); | |
49b25e05 JM |
1137 | ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, |
1138 | 0, &eb); | |
84234f3a YZ |
1139 | btrfs_tree_unlock(eb); |
1140 | free_extent_buffer(eb); | |
0b86a832 | 1141 | |
49b25e05 JM |
1142 | if (ret) |
1143 | return ret; | |
1144 | ||
c79a70b1 | 1145 | ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); |
49b25e05 JM |
1146 | if (ret) |
1147 | return ret; | |
87ef2bb4 | 1148 | |
0b246afa | 1149 | ret = btrfs_run_dev_stats(trans, fs_info); |
c16ce190 JB |
1150 | if (ret) |
1151 | return ret; | |
0b246afa | 1152 | ret = btrfs_run_dev_replace(trans, fs_info); |
c16ce190 JB |
1153 | if (ret) |
1154 | return ret; | |
280f8bd2 | 1155 | ret = btrfs_run_qgroups(trans); |
c16ce190 JB |
1156 | if (ret) |
1157 | return ret; | |
546adb0d | 1158 | |
2ff7e61e | 1159 | ret = btrfs_setup_space_cache(trans, fs_info); |
dcdf7f6d JB |
1160 | if (ret) |
1161 | return ret; | |
1162 | ||
546adb0d | 1163 | /* run_qgroups might have added some more refs */ |
c79a70b1 | 1164 | ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); |
c16ce190 JB |
1165 | if (ret) |
1166 | return ret; | |
ea526d18 | 1167 | again: |
d397712b | 1168 | while (!list_empty(&fs_info->dirty_cowonly_roots)) { |
2ff7e61e | 1169 | struct btrfs_root *root; |
0b86a832 CM |
1170 | next = fs_info->dirty_cowonly_roots.next; |
1171 | list_del_init(next); | |
1172 | root = list_entry(next, struct btrfs_root, dirty_list); | |
e7070be1 | 1173 | clear_bit(BTRFS_ROOT_DIRTY, &root->state); |
87ef2bb4 | 1174 | |
9e351cc8 JB |
1175 | if (root != fs_info->extent_root) |
1176 | list_add_tail(&root->dirty_list, | |
1177 | &trans->transaction->switch_commits); | |
49b25e05 JM |
1178 | ret = update_cowonly_root(trans, root); |
1179 | if (ret) | |
1180 | return ret; | |
c79a70b1 | 1181 | ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); |
ea526d18 JB |
1182 | if (ret) |
1183 | return ret; | |
79154b1b | 1184 | } |
276e680d | 1185 | |
1bbc621e | 1186 | while (!list_empty(dirty_bgs) || !list_empty(io_bgs)) { |
2ff7e61e | 1187 | ret = btrfs_write_dirty_block_groups(trans, fs_info); |
ea526d18 JB |
1188 | if (ret) |
1189 | return ret; | |
c79a70b1 | 1190 | ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); |
ea526d18 JB |
1191 | if (ret) |
1192 | return ret; | |
1193 | } | |
1194 | ||
1195 | if (!list_empty(&fs_info->dirty_cowonly_roots)) | |
1196 | goto again; | |
1197 | ||
9e351cc8 JB |
1198 | list_add_tail(&fs_info->extent_root->dirty_list, |
1199 | &trans->transaction->switch_commits); | |
8dabb742 SB |
1200 | btrfs_after_dev_replace_commit(fs_info); |
1201 | ||
79154b1b CM |
1202 | return 0; |
1203 | } | |
1204 | ||
d352ac68 CM |
1205 | /* |
1206 | * dead roots are old snapshots that need to be deleted. This allocates | |
1207 | * a dirty root struct and adds it into the list of dead roots that need to | |
1208 | * be deleted | |
1209 | */ | |
cfad392b | 1210 | void btrfs_add_dead_root(struct btrfs_root *root) |
5eda7b5e | 1211 | { |
0b246afa JM |
1212 | struct btrfs_fs_info *fs_info = root->fs_info; |
1213 | ||
1214 | spin_lock(&fs_info->trans_lock); | |
cfad392b | 1215 | if (list_empty(&root->root_list)) |
0b246afa JM |
1216 | list_add_tail(&root->root_list, &fs_info->dead_roots); |
1217 | spin_unlock(&fs_info->trans_lock); | |
5eda7b5e CM |
1218 | } |
1219 | ||
d352ac68 | 1220 | /* |
5d4f98a2 | 1221 | * update all the cowonly tree roots on disk |
d352ac68 | 1222 | */ |
7e4443d9 | 1223 | static noinline int commit_fs_roots(struct btrfs_trans_handle *trans) |
0f7d52f4 | 1224 | { |
7e4443d9 | 1225 | struct btrfs_fs_info *fs_info = trans->fs_info; |
0f7d52f4 | 1226 | struct btrfs_root *gang[8]; |
0f7d52f4 CM |
1227 | int i; |
1228 | int ret; | |
54aa1f4d CM |
1229 | int err = 0; |
1230 | ||
a4abeea4 | 1231 | spin_lock(&fs_info->fs_roots_radix_lock); |
d397712b | 1232 | while (1) { |
5d4f98a2 YZ |
1233 | ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, |
1234 | (void **)gang, 0, | |
0f7d52f4 CM |
1235 | ARRAY_SIZE(gang), |
1236 | BTRFS_ROOT_TRANS_TAG); | |
1237 | if (ret == 0) | |
1238 | break; | |
1239 | for (i = 0; i < ret; i++) { | |
5b4aacef | 1240 | struct btrfs_root *root = gang[i]; |
5d4f98a2 YZ |
1241 | radix_tree_tag_clear(&fs_info->fs_roots_radix, |
1242 | (unsigned long)root->root_key.objectid, | |
1243 | BTRFS_ROOT_TRANS_TAG); | |
a4abeea4 | 1244 | spin_unlock(&fs_info->fs_roots_radix_lock); |
31153d81 | 1245 | |
e02119d5 | 1246 | btrfs_free_log(trans, root); |
5d4f98a2 | 1247 | btrfs_update_reloc_root(trans, root); |
bcc63abb | 1248 | |
82d5902d LZ |
1249 | btrfs_save_ino_cache(root, trans); |
1250 | ||
f1ebcc74 | 1251 | /* see comments in should_cow_block() */ |
27cdeb70 | 1252 | clear_bit(BTRFS_ROOT_FORCE_COW, &root->state); |
c7548af6 | 1253 | smp_mb__after_atomic(); |
f1ebcc74 | 1254 | |
978d910d | 1255 | if (root->commit_root != root->node) { |
9e351cc8 JB |
1256 | list_add_tail(&root->dirty_list, |
1257 | &trans->transaction->switch_commits); | |
978d910d YZ |
1258 | btrfs_set_root_node(&root->root_item, |
1259 | root->node); | |
1260 | } | |
5d4f98a2 | 1261 | |
5d4f98a2 | 1262 | err = btrfs_update_root(trans, fs_info->tree_root, |
0f7d52f4 CM |
1263 | &root->root_key, |
1264 | &root->root_item); | |
a4abeea4 | 1265 | spin_lock(&fs_info->fs_roots_radix_lock); |
54aa1f4d CM |
1266 | if (err) |
1267 | break; | |
733e03a0 | 1268 | btrfs_qgroup_free_meta_all_pertrans(root); |
0f7d52f4 CM |
1269 | } |
1270 | } | |
a4abeea4 | 1271 | spin_unlock(&fs_info->fs_roots_radix_lock); |
54aa1f4d | 1272 | return err; |
0f7d52f4 CM |
1273 | } |
1274 | ||
d352ac68 | 1275 | /* |
de78b51a ES |
1276 | * defrag a given btree. |
1277 | * Every leaf in the btree is read and defragged. | |
d352ac68 | 1278 | */ |
de78b51a | 1279 | int btrfs_defrag_root(struct btrfs_root *root) |
e9d0b13b CM |
1280 | { |
1281 | struct btrfs_fs_info *info = root->fs_info; | |
e9d0b13b | 1282 | struct btrfs_trans_handle *trans; |
8929ecfa | 1283 | int ret; |
e9d0b13b | 1284 | |
27cdeb70 | 1285 | if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state)) |
e9d0b13b | 1286 | return 0; |
8929ecfa | 1287 | |
6b80053d | 1288 | while (1) { |
8929ecfa YZ |
1289 | trans = btrfs_start_transaction(root, 0); |
1290 | if (IS_ERR(trans)) | |
1291 | return PTR_ERR(trans); | |
1292 | ||
de78b51a | 1293 | ret = btrfs_defrag_leaves(trans, root); |
8929ecfa | 1294 | |
3a45bb20 | 1295 | btrfs_end_transaction(trans); |
2ff7e61e | 1296 | btrfs_btree_balance_dirty(info); |
e9d0b13b CM |
1297 | cond_resched(); |
1298 | ||
ab8d0fc4 | 1299 | if (btrfs_fs_closing(info) || ret != -EAGAIN) |
e9d0b13b | 1300 | break; |
210549eb | 1301 | |
ab8d0fc4 JM |
1302 | if (btrfs_defrag_cancelled(info)) { |
1303 | btrfs_debug(info, "defrag_root cancelled"); | |
210549eb DS |
1304 | ret = -EAGAIN; |
1305 | break; | |
1306 | } | |
e9d0b13b | 1307 | } |
27cdeb70 | 1308 | clear_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state); |
8929ecfa | 1309 | return ret; |
e9d0b13b CM |
1310 | } |
1311 | ||
6426c7ad QW |
1312 | /* |
1313 | * Do all special snapshot related qgroup dirty hack. | |
1314 | * | |
1315 | * Will do all needed qgroup inherit and dirty hack like switch commit | |
1316 | * roots inside one transaction and write all btree into disk, to make | |
1317 | * qgroup works. | |
1318 | */ | |
1319 | static int qgroup_account_snapshot(struct btrfs_trans_handle *trans, | |
1320 | struct btrfs_root *src, | |
1321 | struct btrfs_root *parent, | |
1322 | struct btrfs_qgroup_inherit *inherit, | |
1323 | u64 dst_objectid) | |
1324 | { | |
1325 | struct btrfs_fs_info *fs_info = src->fs_info; | |
1326 | int ret; | |
1327 | ||
1328 | /* | |
1329 | * Save some performance in the case that qgroups are not | |
1330 | * enabled. If this check races with the ioctl, rescan will | |
1331 | * kick in anyway. | |
1332 | */ | |
9ea6e2b5 | 1333 | if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) |
6426c7ad | 1334 | return 0; |
6426c7ad | 1335 | |
4d31778a QW |
1336 | /* |
1337 | * Ensure dirty @src will be commited. Or, after comming | |
1338 | * commit_fs_roots() and switch_commit_roots(), any dirty but not | |
1339 | * recorded root will never be updated again, causing an outdated root | |
1340 | * item. | |
1341 | */ | |
1342 | record_root_in_trans(trans, src, 1); | |
1343 | ||
6426c7ad QW |
1344 | /* |
1345 | * We are going to commit transaction, see btrfs_commit_transaction() | |
1346 | * comment for reason locking tree_log_mutex | |
1347 | */ | |
1348 | mutex_lock(&fs_info->tree_log_mutex); | |
1349 | ||
7e4443d9 | 1350 | ret = commit_fs_roots(trans); |
6426c7ad QW |
1351 | if (ret) |
1352 | goto out; | |
460fb20a | 1353 | ret = btrfs_qgroup_account_extents(trans); |
6426c7ad QW |
1354 | if (ret < 0) |
1355 | goto out; | |
1356 | ||
1357 | /* Now qgroup are all updated, we can inherit it to new qgroups */ | |
1358 | ret = btrfs_qgroup_inherit(trans, fs_info, | |
1359 | src->root_key.objectid, dst_objectid, | |
1360 | inherit); | |
1361 | if (ret < 0) | |
1362 | goto out; | |
1363 | ||
1364 | /* | |
1365 | * Now we do a simplified commit transaction, which will: | |
1366 | * 1) commit all subvolume and extent tree | |
1367 | * To ensure all subvolume and extent tree have a valid | |
1368 | * commit_root to accounting later insert_dir_item() | |
1369 | * 2) write all btree blocks onto disk | |
1370 | * This is to make sure later btree modification will be cowed | |
1371 | * Or commit_root can be populated and cause wrong qgroup numbers | |
1372 | * In this simplified commit, we don't really care about other trees | |
1373 | * like chunk and root tree, as they won't affect qgroup. | |
1374 | * And we don't write super to avoid half committed status. | |
1375 | */ | |
9386d8bc | 1376 | ret = commit_cowonly_roots(trans); |
6426c7ad QW |
1377 | if (ret) |
1378 | goto out; | |
16916a88 | 1379 | switch_commit_roots(trans->transaction); |
70458a58 | 1380 | ret = btrfs_write_and_wait_transaction(trans); |
6426c7ad | 1381 | if (ret) |
f7af3934 | 1382 | btrfs_handle_fs_error(fs_info, ret, |
6426c7ad QW |
1383 | "Error while writing out transaction for qgroup"); |
1384 | ||
1385 | out: | |
1386 | mutex_unlock(&fs_info->tree_log_mutex); | |
1387 | ||
1388 | /* | |
1389 | * Force parent root to be updated, as we recorded it before so its | |
1390 | * last_trans == cur_transid. | |
1391 | * Or it won't be committed again onto disk after later | |
1392 | * insert_dir_item() | |
1393 | */ | |
1394 | if (!ret) | |
1395 | record_root_in_trans(trans, parent, 1); | |
1396 | return ret; | |
1397 | } | |
1398 | ||
d352ac68 CM |
1399 | /* |
1400 | * new snapshots need to be created at a very specific time in the | |
aec8030a MX |
1401 | * transaction commit. This does the actual creation. |
1402 | * | |
1403 | * Note: | |
1404 | * If the error which may affect the commitment of the current transaction | |
1405 | * happens, we should return the error number. If the error which just affect | |
1406 | * the creation of the pending snapshots, just return 0. | |
d352ac68 | 1407 | */ |
80b6794d | 1408 | static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, |
3063d29f CM |
1409 | struct btrfs_pending_snapshot *pending) |
1410 | { | |
08d50ca3 NB |
1411 | |
1412 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
3063d29f | 1413 | struct btrfs_key key; |
80b6794d | 1414 | struct btrfs_root_item *new_root_item; |
3063d29f CM |
1415 | struct btrfs_root *tree_root = fs_info->tree_root; |
1416 | struct btrfs_root *root = pending->root; | |
6bdb72de | 1417 | struct btrfs_root *parent_root; |
98c9942a | 1418 | struct btrfs_block_rsv *rsv; |
6bdb72de | 1419 | struct inode *parent_inode; |
42874b3d MX |
1420 | struct btrfs_path *path; |
1421 | struct btrfs_dir_item *dir_item; | |
a22285a6 | 1422 | struct dentry *dentry; |
3063d29f | 1423 | struct extent_buffer *tmp; |
925baedd | 1424 | struct extent_buffer *old; |
95582b00 | 1425 | struct timespec64 cur_time; |
aec8030a | 1426 | int ret = 0; |
d68fc57b | 1427 | u64 to_reserve = 0; |
6bdb72de | 1428 | u64 index = 0; |
a22285a6 | 1429 | u64 objectid; |
b83cc969 | 1430 | u64 root_flags; |
8ea05e3a | 1431 | uuid_le new_uuid; |
3063d29f | 1432 | |
8546b570 DS |
1433 | ASSERT(pending->path); |
1434 | path = pending->path; | |
42874b3d | 1435 | |
b0c0ea63 DS |
1436 | ASSERT(pending->root_item); |
1437 | new_root_item = pending->root_item; | |
a22285a6 | 1438 | |
aec8030a MX |
1439 | pending->error = btrfs_find_free_objectid(tree_root, &objectid); |
1440 | if (pending->error) | |
6fa9700e | 1441 | goto no_free_objectid; |
3063d29f | 1442 | |
d6726335 QW |
1443 | /* |
1444 | * Make qgroup to skip current new snapshot's qgroupid, as it is | |
1445 | * accounted by later btrfs_qgroup_inherit(). | |
1446 | */ | |
1447 | btrfs_set_skip_qgroup(trans, objectid); | |
1448 | ||
147d256e | 1449 | btrfs_reloc_pre_snapshot(pending, &to_reserve); |
d68fc57b YZ |
1450 | |
1451 | if (to_reserve > 0) { | |
aec8030a MX |
1452 | pending->error = btrfs_block_rsv_add(root, |
1453 | &pending->block_rsv, | |
1454 | to_reserve, | |
1455 | BTRFS_RESERVE_NO_FLUSH); | |
1456 | if (pending->error) | |
d6726335 | 1457 | goto clear_skip_qgroup; |
d68fc57b YZ |
1458 | } |
1459 | ||
3063d29f | 1460 | key.objectid = objectid; |
a22285a6 YZ |
1461 | key.offset = (u64)-1; |
1462 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3063d29f | 1463 | |
6fa9700e | 1464 | rsv = trans->block_rsv; |
a22285a6 | 1465 | trans->block_rsv = &pending->block_rsv; |
2382c5cc | 1466 | trans->bytes_reserved = trans->block_rsv->reserved; |
0b246afa | 1467 | trace_btrfs_space_reservation(fs_info, "transaction", |
88d3a5aa JB |
1468 | trans->transid, |
1469 | trans->bytes_reserved, 1); | |
a22285a6 | 1470 | dentry = pending->dentry; |
e9662f70 | 1471 | parent_inode = pending->dir; |
a22285a6 | 1472 | parent_root = BTRFS_I(parent_inode)->root; |
6426c7ad | 1473 | record_root_in_trans(trans, parent_root, 0); |
a22285a6 | 1474 | |
c2050a45 | 1475 | cur_time = current_time(parent_inode); |
04b285f3 | 1476 | |
3063d29f CM |
1477 | /* |
1478 | * insert the directory item | |
1479 | */ | |
877574e2 | 1480 | ret = btrfs_set_inode_index(BTRFS_I(parent_inode), &index); |
49b25e05 | 1481 | BUG_ON(ret); /* -ENOMEM */ |
42874b3d MX |
1482 | |
1483 | /* check if there is a file/dir which has the same name. */ | |
1484 | dir_item = btrfs_lookup_dir_item(NULL, parent_root, path, | |
4a0cc7ca | 1485 | btrfs_ino(BTRFS_I(parent_inode)), |
42874b3d MX |
1486 | dentry->d_name.name, |
1487 | dentry->d_name.len, 0); | |
1488 | if (dir_item != NULL && !IS_ERR(dir_item)) { | |
fe66a05a | 1489 | pending->error = -EEXIST; |
aec8030a | 1490 | goto dir_item_existed; |
42874b3d MX |
1491 | } else if (IS_ERR(dir_item)) { |
1492 | ret = PTR_ERR(dir_item); | |
66642832 | 1493 | btrfs_abort_transaction(trans, ret); |
8732d44f | 1494 | goto fail; |
79787eaa | 1495 | } |
42874b3d | 1496 | btrfs_release_path(path); |
52c26179 | 1497 | |
e999376f CM |
1498 | /* |
1499 | * pull in the delayed directory update | |
1500 | * and the delayed inode item | |
1501 | * otherwise we corrupt the FS during | |
1502 | * snapshot | |
1503 | */ | |
e5c304e6 | 1504 | ret = btrfs_run_delayed_items(trans); |
8732d44f | 1505 | if (ret) { /* Transaction aborted */ |
66642832 | 1506 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1507 | goto fail; |
1508 | } | |
e999376f | 1509 | |
6426c7ad | 1510 | record_root_in_trans(trans, root, 0); |
6bdb72de SW |
1511 | btrfs_set_root_last_snapshot(&root->root_item, trans->transid); |
1512 | memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); | |
08fe4db1 | 1513 | btrfs_check_and_init_root_item(new_root_item); |
6bdb72de | 1514 | |
b83cc969 LZ |
1515 | root_flags = btrfs_root_flags(new_root_item); |
1516 | if (pending->readonly) | |
1517 | root_flags |= BTRFS_ROOT_SUBVOL_RDONLY; | |
1518 | else | |
1519 | root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY; | |
1520 | btrfs_set_root_flags(new_root_item, root_flags); | |
1521 | ||
8ea05e3a AB |
1522 | btrfs_set_root_generation_v2(new_root_item, |
1523 | trans->transid); | |
1524 | uuid_le_gen(&new_uuid); | |
1525 | memcpy(new_root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE); | |
1526 | memcpy(new_root_item->parent_uuid, root->root_item.uuid, | |
1527 | BTRFS_UUID_SIZE); | |
70023da2 SB |
1528 | if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) { |
1529 | memset(new_root_item->received_uuid, 0, | |
1530 | sizeof(new_root_item->received_uuid)); | |
1531 | memset(&new_root_item->stime, 0, sizeof(new_root_item->stime)); | |
1532 | memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime)); | |
1533 | btrfs_set_root_stransid(new_root_item, 0); | |
1534 | btrfs_set_root_rtransid(new_root_item, 0); | |
1535 | } | |
3cae210f QW |
1536 | btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec); |
1537 | btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec); | |
8ea05e3a | 1538 | btrfs_set_root_otransid(new_root_item, trans->transid); |
8ea05e3a | 1539 | |
6bdb72de | 1540 | old = btrfs_lock_root_node(root); |
49b25e05 | 1541 | ret = btrfs_cow_block(trans, root, old, NULL, 0, &old); |
79787eaa JM |
1542 | if (ret) { |
1543 | btrfs_tree_unlock(old); | |
1544 | free_extent_buffer(old); | |
66642832 | 1545 | btrfs_abort_transaction(trans, ret); |
8732d44f | 1546 | goto fail; |
79787eaa | 1547 | } |
49b25e05 | 1548 | |
6bdb72de SW |
1549 | btrfs_set_lock_blocking(old); |
1550 | ||
49b25e05 | 1551 | ret = btrfs_copy_root(trans, root, old, &tmp, objectid); |
79787eaa | 1552 | /* clean up in any case */ |
6bdb72de SW |
1553 | btrfs_tree_unlock(old); |
1554 | free_extent_buffer(old); | |
8732d44f | 1555 | if (ret) { |
66642832 | 1556 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1557 | goto fail; |
1558 | } | |
f1ebcc74 | 1559 | /* see comments in should_cow_block() */ |
27cdeb70 | 1560 | set_bit(BTRFS_ROOT_FORCE_COW, &root->state); |
f1ebcc74 LB |
1561 | smp_wmb(); |
1562 | ||
6bdb72de | 1563 | btrfs_set_root_node(new_root_item, tmp); |
a22285a6 YZ |
1564 | /* record when the snapshot was created in key.offset */ |
1565 | key.offset = trans->transid; | |
1566 | ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); | |
6bdb72de SW |
1567 | btrfs_tree_unlock(tmp); |
1568 | free_extent_buffer(tmp); | |
8732d44f | 1569 | if (ret) { |
66642832 | 1570 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1571 | goto fail; |
1572 | } | |
6bdb72de | 1573 | |
a22285a6 YZ |
1574 | /* |
1575 | * insert root back/forward references | |
1576 | */ | |
6bccf3ab | 1577 | ret = btrfs_add_root_ref(trans, fs_info, objectid, |
0660b5af | 1578 | parent_root->root_key.objectid, |
4a0cc7ca | 1579 | btrfs_ino(BTRFS_I(parent_inode)), index, |
a22285a6 | 1580 | dentry->d_name.name, dentry->d_name.len); |
8732d44f | 1581 | if (ret) { |
66642832 | 1582 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1583 | goto fail; |
1584 | } | |
0660b5af | 1585 | |
a22285a6 | 1586 | key.offset = (u64)-1; |
0b246afa | 1587 | pending->snap = btrfs_read_fs_root_no_name(fs_info, &key); |
79787eaa JM |
1588 | if (IS_ERR(pending->snap)) { |
1589 | ret = PTR_ERR(pending->snap); | |
66642832 | 1590 | btrfs_abort_transaction(trans, ret); |
8732d44f | 1591 | goto fail; |
79787eaa | 1592 | } |
d68fc57b | 1593 | |
49b25e05 | 1594 | ret = btrfs_reloc_post_snapshot(trans, pending); |
8732d44f | 1595 | if (ret) { |
66642832 | 1596 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1597 | goto fail; |
1598 | } | |
361048f5 | 1599 | |
c79a70b1 | 1600 | ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); |
8732d44f | 1601 | if (ret) { |
66642832 | 1602 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1603 | goto fail; |
1604 | } | |
42874b3d | 1605 | |
6426c7ad QW |
1606 | /* |
1607 | * Do special qgroup accounting for snapshot, as we do some qgroup | |
1608 | * snapshot hack to do fast snapshot. | |
1609 | * To co-operate with that hack, we do hack again. | |
1610 | * Or snapshot will be greatly slowed down by a subtree qgroup rescan | |
1611 | */ | |
1612 | ret = qgroup_account_snapshot(trans, root, parent_root, | |
1613 | pending->inherit, objectid); | |
1614 | if (ret < 0) | |
1615 | goto fail; | |
1616 | ||
42874b3d MX |
1617 | ret = btrfs_insert_dir_item(trans, parent_root, |
1618 | dentry->d_name.name, dentry->d_name.len, | |
8e7611cf | 1619 | BTRFS_I(parent_inode), &key, |
42874b3d MX |
1620 | BTRFS_FT_DIR, index); |
1621 | /* We have check then name at the beginning, so it is impossible. */ | |
9c52057c | 1622 | BUG_ON(ret == -EEXIST || ret == -EOVERFLOW); |
8732d44f | 1623 | if (ret) { |
66642832 | 1624 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1625 | goto fail; |
1626 | } | |
42874b3d | 1627 | |
6ef06d27 | 1628 | btrfs_i_size_write(BTRFS_I(parent_inode), parent_inode->i_size + |
42874b3d | 1629 | dentry->d_name.len * 2); |
04b285f3 | 1630 | parent_inode->i_mtime = parent_inode->i_ctime = |
c2050a45 | 1631 | current_time(parent_inode); |
be6aef60 | 1632 | ret = btrfs_update_inode_fallback(trans, parent_root, parent_inode); |
dd5f9615 | 1633 | if (ret) { |
66642832 | 1634 | btrfs_abort_transaction(trans, ret); |
dd5f9615 SB |
1635 | goto fail; |
1636 | } | |
cdb345a8 LF |
1637 | ret = btrfs_uuid_tree_add(trans, new_uuid.b, BTRFS_UUID_KEY_SUBVOL, |
1638 | objectid); | |
dd5f9615 | 1639 | if (ret) { |
66642832 | 1640 | btrfs_abort_transaction(trans, ret); |
dd5f9615 SB |
1641 | goto fail; |
1642 | } | |
1643 | if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) { | |
cdb345a8 | 1644 | ret = btrfs_uuid_tree_add(trans, new_root_item->received_uuid, |
dd5f9615 SB |
1645 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, |
1646 | objectid); | |
1647 | if (ret && ret != -EEXIST) { | |
66642832 | 1648 | btrfs_abort_transaction(trans, ret); |
dd5f9615 SB |
1649 | goto fail; |
1650 | } | |
1651 | } | |
d6726335 | 1652 | |
c79a70b1 | 1653 | ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); |
d6726335 | 1654 | if (ret) { |
66642832 | 1655 | btrfs_abort_transaction(trans, ret); |
d6726335 QW |
1656 | goto fail; |
1657 | } | |
1658 | ||
3063d29f | 1659 | fail: |
aec8030a MX |
1660 | pending->error = ret; |
1661 | dir_item_existed: | |
98c9942a | 1662 | trans->block_rsv = rsv; |
2382c5cc | 1663 | trans->bytes_reserved = 0; |
d6726335 QW |
1664 | clear_skip_qgroup: |
1665 | btrfs_clear_skip_qgroup(trans); | |
6fa9700e MX |
1666 | no_free_objectid: |
1667 | kfree(new_root_item); | |
b0c0ea63 | 1668 | pending->root_item = NULL; |
42874b3d | 1669 | btrfs_free_path(path); |
8546b570 DS |
1670 | pending->path = NULL; |
1671 | ||
49b25e05 | 1672 | return ret; |
3063d29f CM |
1673 | } |
1674 | ||
d352ac68 CM |
1675 | /* |
1676 | * create all the snapshots we've scheduled for creation | |
1677 | */ | |
08d50ca3 | 1678 | static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans) |
3de4586c | 1679 | { |
aec8030a | 1680 | struct btrfs_pending_snapshot *pending, *next; |
3de4586c | 1681 | struct list_head *head = &trans->transaction->pending_snapshots; |
aec8030a | 1682 | int ret = 0; |
3de4586c | 1683 | |
aec8030a MX |
1684 | list_for_each_entry_safe(pending, next, head, list) { |
1685 | list_del(&pending->list); | |
08d50ca3 | 1686 | ret = create_pending_snapshot(trans, pending); |
aec8030a MX |
1687 | if (ret) |
1688 | break; | |
1689 | } | |
1690 | return ret; | |
3de4586c CM |
1691 | } |
1692 | ||
2ff7e61e | 1693 | static void update_super_roots(struct btrfs_fs_info *fs_info) |
5d4f98a2 YZ |
1694 | { |
1695 | struct btrfs_root_item *root_item; | |
1696 | struct btrfs_super_block *super; | |
1697 | ||
0b246afa | 1698 | super = fs_info->super_copy; |
5d4f98a2 | 1699 | |
0b246afa | 1700 | root_item = &fs_info->chunk_root->root_item; |
093e037c DS |
1701 | super->chunk_root = root_item->bytenr; |
1702 | super->chunk_root_generation = root_item->generation; | |
1703 | super->chunk_root_level = root_item->level; | |
5d4f98a2 | 1704 | |
0b246afa | 1705 | root_item = &fs_info->tree_root->root_item; |
093e037c DS |
1706 | super->root = root_item->bytenr; |
1707 | super->generation = root_item->generation; | |
1708 | super->root_level = root_item->level; | |
0b246afa | 1709 | if (btrfs_test_opt(fs_info, SPACE_CACHE)) |
093e037c | 1710 | super->cache_generation = root_item->generation; |
0b246afa | 1711 | if (test_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags)) |
093e037c | 1712 | super->uuid_tree_generation = root_item->generation; |
5d4f98a2 YZ |
1713 | } |
1714 | ||
f36f3042 CM |
1715 | int btrfs_transaction_in_commit(struct btrfs_fs_info *info) |
1716 | { | |
4a9d8bde | 1717 | struct btrfs_transaction *trans; |
f36f3042 | 1718 | int ret = 0; |
4a9d8bde | 1719 | |
a4abeea4 | 1720 | spin_lock(&info->trans_lock); |
4a9d8bde MX |
1721 | trans = info->running_transaction; |
1722 | if (trans) | |
1723 | ret = (trans->state >= TRANS_STATE_COMMIT_START); | |
a4abeea4 | 1724 | spin_unlock(&info->trans_lock); |
f36f3042 CM |
1725 | return ret; |
1726 | } | |
1727 | ||
8929ecfa YZ |
1728 | int btrfs_transaction_blocked(struct btrfs_fs_info *info) |
1729 | { | |
4a9d8bde | 1730 | struct btrfs_transaction *trans; |
8929ecfa | 1731 | int ret = 0; |
4a9d8bde | 1732 | |
a4abeea4 | 1733 | spin_lock(&info->trans_lock); |
4a9d8bde MX |
1734 | trans = info->running_transaction; |
1735 | if (trans) | |
1736 | ret = is_transaction_blocked(trans); | |
a4abeea4 | 1737 | spin_unlock(&info->trans_lock); |
8929ecfa YZ |
1738 | return ret; |
1739 | } | |
1740 | ||
bb9c12c9 SW |
1741 | /* |
1742 | * wait for the current transaction commit to start and block subsequent | |
1743 | * transaction joins | |
1744 | */ | |
2ff7e61e | 1745 | static void wait_current_trans_commit_start(struct btrfs_fs_info *fs_info, |
bb9c12c9 SW |
1746 | struct btrfs_transaction *trans) |
1747 | { | |
2ff7e61e JM |
1748 | wait_event(fs_info->transaction_blocked_wait, |
1749 | trans->state >= TRANS_STATE_COMMIT_START || trans->aborted); | |
bb9c12c9 SW |
1750 | } |
1751 | ||
1752 | /* | |
1753 | * wait for the current transaction to start and then become unblocked. | |
1754 | * caller holds ref. | |
1755 | */ | |
2ff7e61e JM |
1756 | static void wait_current_trans_commit_start_and_unblock( |
1757 | struct btrfs_fs_info *fs_info, | |
1758 | struct btrfs_transaction *trans) | |
bb9c12c9 | 1759 | { |
2ff7e61e JM |
1760 | wait_event(fs_info->transaction_wait, |
1761 | trans->state >= TRANS_STATE_UNBLOCKED || trans->aborted); | |
bb9c12c9 SW |
1762 | } |
1763 | ||
1764 | /* | |
1765 | * commit transactions asynchronously. once btrfs_commit_transaction_async | |
1766 | * returns, any subsequent transaction will not be allowed to join. | |
1767 | */ | |
1768 | struct btrfs_async_commit { | |
1769 | struct btrfs_trans_handle *newtrans; | |
7892b5af | 1770 | struct work_struct work; |
bb9c12c9 SW |
1771 | }; |
1772 | ||
1773 | static void do_async_commit(struct work_struct *work) | |
1774 | { | |
1775 | struct btrfs_async_commit *ac = | |
7892b5af | 1776 | container_of(work, struct btrfs_async_commit, work); |
bb9c12c9 | 1777 | |
6fc4e354 SW |
1778 | /* |
1779 | * We've got freeze protection passed with the transaction. | |
1780 | * Tell lockdep about it. | |
1781 | */ | |
b1a06a4b | 1782 | if (ac->newtrans->type & __TRANS_FREEZABLE) |
3a45bb20 | 1783 | __sb_writers_acquired(ac->newtrans->fs_info->sb, SB_FREEZE_FS); |
6fc4e354 | 1784 | |
e209db7a SW |
1785 | current->journal_info = ac->newtrans; |
1786 | ||
3a45bb20 | 1787 | btrfs_commit_transaction(ac->newtrans); |
bb9c12c9 SW |
1788 | kfree(ac); |
1789 | } | |
1790 | ||
1791 | int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans, | |
bb9c12c9 SW |
1792 | int wait_for_unblock) |
1793 | { | |
3a45bb20 | 1794 | struct btrfs_fs_info *fs_info = trans->fs_info; |
bb9c12c9 SW |
1795 | struct btrfs_async_commit *ac; |
1796 | struct btrfs_transaction *cur_trans; | |
1797 | ||
1798 | ac = kmalloc(sizeof(*ac), GFP_NOFS); | |
db5b493a TI |
1799 | if (!ac) |
1800 | return -ENOMEM; | |
bb9c12c9 | 1801 | |
7892b5af | 1802 | INIT_WORK(&ac->work, do_async_commit); |
3a45bb20 | 1803 | ac->newtrans = btrfs_join_transaction(trans->root); |
3612b495 TI |
1804 | if (IS_ERR(ac->newtrans)) { |
1805 | int err = PTR_ERR(ac->newtrans); | |
1806 | kfree(ac); | |
1807 | return err; | |
1808 | } | |
bb9c12c9 SW |
1809 | |
1810 | /* take transaction reference */ | |
bb9c12c9 | 1811 | cur_trans = trans->transaction; |
9b64f57d | 1812 | refcount_inc(&cur_trans->use_count); |
bb9c12c9 | 1813 | |
3a45bb20 | 1814 | btrfs_end_transaction(trans); |
6fc4e354 SW |
1815 | |
1816 | /* | |
1817 | * Tell lockdep we've released the freeze rwsem, since the | |
1818 | * async commit thread will be the one to unlock it. | |
1819 | */ | |
b1a06a4b | 1820 | if (ac->newtrans->type & __TRANS_FREEZABLE) |
0b246afa | 1821 | __sb_writers_release(fs_info->sb, SB_FREEZE_FS); |
6fc4e354 | 1822 | |
7892b5af | 1823 | schedule_work(&ac->work); |
bb9c12c9 SW |
1824 | |
1825 | /* wait for transaction to start and unblock */ | |
bb9c12c9 | 1826 | if (wait_for_unblock) |
2ff7e61e | 1827 | wait_current_trans_commit_start_and_unblock(fs_info, cur_trans); |
bb9c12c9 | 1828 | else |
2ff7e61e | 1829 | wait_current_trans_commit_start(fs_info, cur_trans); |
bb9c12c9 | 1830 | |
38e88054 SW |
1831 | if (current->journal_info == trans) |
1832 | current->journal_info = NULL; | |
1833 | ||
724e2315 | 1834 | btrfs_put_transaction(cur_trans); |
bb9c12c9 SW |
1835 | return 0; |
1836 | } | |
1837 | ||
49b25e05 | 1838 | |
97cb39bb | 1839 | static void cleanup_transaction(struct btrfs_trans_handle *trans, int err) |
49b25e05 | 1840 | { |
97cb39bb | 1841 | struct btrfs_fs_info *fs_info = trans->fs_info; |
49b25e05 | 1842 | struct btrfs_transaction *cur_trans = trans->transaction; |
f094ac32 | 1843 | DEFINE_WAIT(wait); |
49b25e05 | 1844 | |
b50fff81 | 1845 | WARN_ON(refcount_read(&trans->use_count) > 1); |
49b25e05 | 1846 | |
66642832 | 1847 | btrfs_abort_transaction(trans, err); |
7b8b92af | 1848 | |
0b246afa | 1849 | spin_lock(&fs_info->trans_lock); |
66b6135b | 1850 | |
25d8c284 MX |
1851 | /* |
1852 | * If the transaction is removed from the list, it means this | |
1853 | * transaction has been committed successfully, so it is impossible | |
1854 | * to call the cleanup function. | |
1855 | */ | |
1856 | BUG_ON(list_empty(&cur_trans->list)); | |
66b6135b | 1857 | |
49b25e05 | 1858 | list_del_init(&cur_trans->list); |
0b246afa | 1859 | if (cur_trans == fs_info->running_transaction) { |
4a9d8bde | 1860 | cur_trans->state = TRANS_STATE_COMMIT_DOING; |
0b246afa | 1861 | spin_unlock(&fs_info->trans_lock); |
f094ac32 LB |
1862 | wait_event(cur_trans->writer_wait, |
1863 | atomic_read(&cur_trans->num_writers) == 1); | |
1864 | ||
0b246afa | 1865 | spin_lock(&fs_info->trans_lock); |
d7096fc3 | 1866 | } |
0b246afa | 1867 | spin_unlock(&fs_info->trans_lock); |
49b25e05 | 1868 | |
2ff7e61e | 1869 | btrfs_cleanup_one_transaction(trans->transaction, fs_info); |
49b25e05 | 1870 | |
0b246afa JM |
1871 | spin_lock(&fs_info->trans_lock); |
1872 | if (cur_trans == fs_info->running_transaction) | |
1873 | fs_info->running_transaction = NULL; | |
1874 | spin_unlock(&fs_info->trans_lock); | |
4a9d8bde | 1875 | |
e0228285 | 1876 | if (trans->type & __TRANS_FREEZABLE) |
0b246afa | 1877 | sb_end_intwrite(fs_info->sb); |
724e2315 JB |
1878 | btrfs_put_transaction(cur_trans); |
1879 | btrfs_put_transaction(cur_trans); | |
49b25e05 | 1880 | |
97cb39bb | 1881 | trace_btrfs_transaction_commit(trans->root); |
49b25e05 | 1882 | |
49b25e05 JM |
1883 | if (current->journal_info == trans) |
1884 | current->journal_info = NULL; | |
0b246afa | 1885 | btrfs_scrub_cancel(fs_info); |
49b25e05 JM |
1886 | |
1887 | kmem_cache_free(btrfs_trans_handle_cachep, trans); | |
1888 | } | |
1889 | ||
82436617 MX |
1890 | static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info) |
1891 | { | |
ce8ea7cc JB |
1892 | /* |
1893 | * We use writeback_inodes_sb here because if we used | |
1894 | * btrfs_start_delalloc_roots we would deadlock with fs freeze. | |
1895 | * Currently are holding the fs freeze lock, if we do an async flush | |
1896 | * we'll do btrfs_join_transaction() and deadlock because we need to | |
1897 | * wait for the fs freeze lock. Using the direct flushing we benefit | |
1898 | * from already being in a transaction and our join_transaction doesn't | |
1899 | * have to re-take the fs freeze lock. | |
1900 | */ | |
3cdde224 | 1901 | if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) |
ce8ea7cc | 1902 | writeback_inodes_sb(fs_info->sb, WB_REASON_SYNC); |
82436617 MX |
1903 | return 0; |
1904 | } | |
1905 | ||
1906 | static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info) | |
1907 | { | |
3cdde224 | 1908 | if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) |
6374e57a | 1909 | btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1); |
82436617 MX |
1910 | } |
1911 | ||
50d9aa99 | 1912 | static inline void |
161c3549 | 1913 | btrfs_wait_pending_ordered(struct btrfs_transaction *cur_trans) |
50d9aa99 | 1914 | { |
161c3549 JB |
1915 | wait_event(cur_trans->pending_wait, |
1916 | atomic_read(&cur_trans->pending_ordered) == 0); | |
50d9aa99 JB |
1917 | } |
1918 | ||
3a45bb20 | 1919 | int btrfs_commit_transaction(struct btrfs_trans_handle *trans) |
79154b1b | 1920 | { |
3a45bb20 | 1921 | struct btrfs_fs_info *fs_info = trans->fs_info; |
49b25e05 | 1922 | struct btrfs_transaction *cur_trans = trans->transaction; |
8fd17795 | 1923 | struct btrfs_transaction *prev_trans = NULL; |
25287e0a | 1924 | int ret; |
79154b1b | 1925 | |
8d25a086 | 1926 | /* Stop the commit early if ->aborted is set */ |
20c7bcec | 1927 | if (unlikely(READ_ONCE(cur_trans->aborted))) { |
25287e0a | 1928 | ret = cur_trans->aborted; |
3a45bb20 | 1929 | btrfs_end_transaction(trans); |
e4a2bcac | 1930 | return ret; |
25287e0a | 1931 | } |
49b25e05 | 1932 | |
56bec294 CM |
1933 | /* make a pass through all the delayed refs we have so far |
1934 | * any runnings procs may add more while we are here | |
1935 | */ | |
c79a70b1 | 1936 | ret = btrfs_run_delayed_refs(trans, 0); |
e4a2bcac | 1937 | if (ret) { |
3a45bb20 | 1938 | btrfs_end_transaction(trans); |
e4a2bcac JB |
1939 | return ret; |
1940 | } | |
56bec294 | 1941 | |
dc60c525 | 1942 | btrfs_trans_release_metadata(trans); |
0e721106 JB |
1943 | trans->block_rsv = NULL; |
1944 | ||
b7ec40d7 | 1945 | cur_trans = trans->transaction; |
49b25e05 | 1946 | |
56bec294 CM |
1947 | /* |
1948 | * set the flushing flag so procs in this transaction have to | |
1949 | * start sending their work down. | |
1950 | */ | |
b7ec40d7 | 1951 | cur_trans->delayed_refs.flushing = 1; |
1be41b78 | 1952 | smp_wmb(); |
56bec294 | 1953 | |
ea658bad | 1954 | if (!list_empty(&trans->new_bgs)) |
6c686b35 | 1955 | btrfs_create_pending_block_groups(trans); |
ea658bad | 1956 | |
c79a70b1 | 1957 | ret = btrfs_run_delayed_refs(trans, 0); |
e4a2bcac | 1958 | if (ret) { |
3a45bb20 | 1959 | btrfs_end_transaction(trans); |
e4a2bcac JB |
1960 | return ret; |
1961 | } | |
56bec294 | 1962 | |
3204d33c | 1963 | if (!test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &cur_trans->flags)) { |
1bbc621e CM |
1964 | int run_it = 0; |
1965 | ||
1966 | /* this mutex is also taken before trying to set | |
1967 | * block groups readonly. We need to make sure | |
1968 | * that nobody has set a block group readonly | |
1969 | * after a extents from that block group have been | |
1970 | * allocated for cache files. btrfs_set_block_group_ro | |
1971 | * will wait for the transaction to commit if it | |
3204d33c | 1972 | * finds BTRFS_TRANS_DIRTY_BG_RUN set. |
1bbc621e | 1973 | * |
3204d33c JB |
1974 | * The BTRFS_TRANS_DIRTY_BG_RUN flag is also used to make sure |
1975 | * only one process starts all the block group IO. It wouldn't | |
1bbc621e CM |
1976 | * hurt to have more than one go through, but there's no |
1977 | * real advantage to it either. | |
1978 | */ | |
0b246afa | 1979 | mutex_lock(&fs_info->ro_block_group_mutex); |
3204d33c JB |
1980 | if (!test_and_set_bit(BTRFS_TRANS_DIRTY_BG_RUN, |
1981 | &cur_trans->flags)) | |
1bbc621e | 1982 | run_it = 1; |
0b246afa | 1983 | mutex_unlock(&fs_info->ro_block_group_mutex); |
1bbc621e | 1984 | |
f9cacae3 | 1985 | if (run_it) { |
21217054 | 1986 | ret = btrfs_start_dirty_block_groups(trans); |
f9cacae3 NB |
1987 | if (ret) { |
1988 | btrfs_end_transaction(trans); | |
1989 | return ret; | |
1990 | } | |
1991 | } | |
1bbc621e CM |
1992 | } |
1993 | ||
0b246afa | 1994 | spin_lock(&fs_info->trans_lock); |
4a9d8bde | 1995 | if (cur_trans->state >= TRANS_STATE_COMMIT_START) { |
0b246afa | 1996 | spin_unlock(&fs_info->trans_lock); |
9b64f57d | 1997 | refcount_inc(&cur_trans->use_count); |
3a45bb20 | 1998 | ret = btrfs_end_transaction(trans); |
ccd467d6 | 1999 | |
2ff7e61e | 2000 | wait_for_commit(cur_trans); |
15ee9bc7 | 2001 | |
b4924a0f LB |
2002 | if (unlikely(cur_trans->aborted)) |
2003 | ret = cur_trans->aborted; | |
2004 | ||
724e2315 | 2005 | btrfs_put_transaction(cur_trans); |
15ee9bc7 | 2006 | |
49b25e05 | 2007 | return ret; |
79154b1b | 2008 | } |
4313b399 | 2009 | |
4a9d8bde | 2010 | cur_trans->state = TRANS_STATE_COMMIT_START; |
0b246afa | 2011 | wake_up(&fs_info->transaction_blocked_wait); |
bb9c12c9 | 2012 | |
0b246afa | 2013 | if (cur_trans->list.prev != &fs_info->trans_list) { |
ccd467d6 CM |
2014 | prev_trans = list_entry(cur_trans->list.prev, |
2015 | struct btrfs_transaction, list); | |
4a9d8bde | 2016 | if (prev_trans->state != TRANS_STATE_COMPLETED) { |
9b64f57d | 2017 | refcount_inc(&prev_trans->use_count); |
0b246afa | 2018 | spin_unlock(&fs_info->trans_lock); |
ccd467d6 | 2019 | |
2ff7e61e | 2020 | wait_for_commit(prev_trans); |
1f9b8c8f | 2021 | ret = prev_trans->aborted; |
ccd467d6 | 2022 | |
724e2315 | 2023 | btrfs_put_transaction(prev_trans); |
1f9b8c8f FM |
2024 | if (ret) |
2025 | goto cleanup_transaction; | |
a4abeea4 | 2026 | } else { |
0b246afa | 2027 | spin_unlock(&fs_info->trans_lock); |
ccd467d6 | 2028 | } |
a4abeea4 | 2029 | } else { |
0b246afa | 2030 | spin_unlock(&fs_info->trans_lock); |
ccd467d6 | 2031 | } |
15ee9bc7 | 2032 | |
0860adfd MX |
2033 | extwriter_counter_dec(cur_trans, trans->type); |
2034 | ||
0b246afa | 2035 | ret = btrfs_start_delalloc_flush(fs_info); |
82436617 MX |
2036 | if (ret) |
2037 | goto cleanup_transaction; | |
2038 | ||
e5c304e6 | 2039 | ret = btrfs_run_delayed_items(trans); |
581227d0 MX |
2040 | if (ret) |
2041 | goto cleanup_transaction; | |
15ee9bc7 | 2042 | |
581227d0 MX |
2043 | wait_event(cur_trans->writer_wait, |
2044 | extwriter_counter_read(cur_trans) == 0); | |
15ee9bc7 | 2045 | |
581227d0 | 2046 | /* some pending stuffs might be added after the previous flush. */ |
e5c304e6 | 2047 | ret = btrfs_run_delayed_items(trans); |
ca469637 MX |
2048 | if (ret) |
2049 | goto cleanup_transaction; | |
2050 | ||
0b246afa | 2051 | btrfs_wait_delalloc_flush(fs_info); |
cb7ab021 | 2052 | |
161c3549 | 2053 | btrfs_wait_pending_ordered(cur_trans); |
50d9aa99 | 2054 | |
2ff7e61e | 2055 | btrfs_scrub_pause(fs_info); |
ed0ca140 JB |
2056 | /* |
2057 | * Ok now we need to make sure to block out any other joins while we | |
2058 | * commit the transaction. We could have started a join before setting | |
4a9d8bde | 2059 | * COMMIT_DOING so make sure to wait for num_writers to == 1 again. |
ed0ca140 | 2060 | */ |
0b246afa | 2061 | spin_lock(&fs_info->trans_lock); |
4a9d8bde | 2062 | cur_trans->state = TRANS_STATE_COMMIT_DOING; |
0b246afa | 2063 | spin_unlock(&fs_info->trans_lock); |
ed0ca140 JB |
2064 | wait_event(cur_trans->writer_wait, |
2065 | atomic_read(&cur_trans->num_writers) == 1); | |
2066 | ||
2cba30f1 | 2067 | /* ->aborted might be set after the previous check, so check it */ |
20c7bcec | 2068 | if (unlikely(READ_ONCE(cur_trans->aborted))) { |
2cba30f1 | 2069 | ret = cur_trans->aborted; |
6cf7f77e | 2070 | goto scrub_continue; |
2cba30f1 | 2071 | } |
7585717f CM |
2072 | /* |
2073 | * the reloc mutex makes sure that we stop | |
2074 | * the balancing code from coming in and moving | |
2075 | * extents around in the middle of the commit | |
2076 | */ | |
0b246afa | 2077 | mutex_lock(&fs_info->reloc_mutex); |
7585717f | 2078 | |
42874b3d MX |
2079 | /* |
2080 | * We needn't worry about the delayed items because we will | |
2081 | * deal with them in create_pending_snapshot(), which is the | |
2082 | * core function of the snapshot creation. | |
2083 | */ | |
08d50ca3 | 2084 | ret = create_pending_snapshots(trans); |
49b25e05 | 2085 | if (ret) { |
0b246afa | 2086 | mutex_unlock(&fs_info->reloc_mutex); |
6cf7f77e | 2087 | goto scrub_continue; |
49b25e05 | 2088 | } |
3063d29f | 2089 | |
42874b3d MX |
2090 | /* |
2091 | * We insert the dir indexes of the snapshots and update the inode | |
2092 | * of the snapshots' parents after the snapshot creation, so there | |
2093 | * are some delayed items which are not dealt with. Now deal with | |
2094 | * them. | |
2095 | * | |
2096 | * We needn't worry that this operation will corrupt the snapshots, | |
2097 | * because all the tree which are snapshoted will be forced to COW | |
2098 | * the nodes and leaves. | |
2099 | */ | |
e5c304e6 | 2100 | ret = btrfs_run_delayed_items(trans); |
49b25e05 | 2101 | if (ret) { |
0b246afa | 2102 | mutex_unlock(&fs_info->reloc_mutex); |
6cf7f77e | 2103 | goto scrub_continue; |
49b25e05 | 2104 | } |
16cdcec7 | 2105 | |
c79a70b1 | 2106 | ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); |
49b25e05 | 2107 | if (ret) { |
0b246afa | 2108 | mutex_unlock(&fs_info->reloc_mutex); |
6cf7f77e | 2109 | goto scrub_continue; |
49b25e05 | 2110 | } |
56bec294 | 2111 | |
e999376f CM |
2112 | /* |
2113 | * make sure none of the code above managed to slip in a | |
2114 | * delayed item | |
2115 | */ | |
ccdf9b30 | 2116 | btrfs_assert_delayed_root_empty(fs_info); |
e999376f | 2117 | |
2c90e5d6 | 2118 | WARN_ON(cur_trans != trans->transaction); |
dc17ff8f | 2119 | |
e02119d5 CM |
2120 | /* btrfs_commit_tree_roots is responsible for getting the |
2121 | * various roots consistent with each other. Every pointer | |
2122 | * in the tree of tree roots has to point to the most up to date | |
2123 | * root for every subvolume and other tree. So, we have to keep | |
2124 | * the tree logging code from jumping in and changing any | |
2125 | * of the trees. | |
2126 | * | |
2127 | * At this point in the commit, there can't be any tree-log | |
2128 | * writers, but a little lower down we drop the trans mutex | |
2129 | * and let new people in. By holding the tree_log_mutex | |
2130 | * from now until after the super is written, we avoid races | |
2131 | * with the tree-log code. | |
2132 | */ | |
0b246afa | 2133 | mutex_lock(&fs_info->tree_log_mutex); |
e02119d5 | 2134 | |
7e4443d9 | 2135 | ret = commit_fs_roots(trans); |
49b25e05 | 2136 | if (ret) { |
0b246afa JM |
2137 | mutex_unlock(&fs_info->tree_log_mutex); |
2138 | mutex_unlock(&fs_info->reloc_mutex); | |
6cf7f77e | 2139 | goto scrub_continue; |
49b25e05 | 2140 | } |
54aa1f4d | 2141 | |
3818aea2 | 2142 | /* |
7e1876ac DS |
2143 | * Since the transaction is done, we can apply the pending changes |
2144 | * before the next transaction. | |
3818aea2 | 2145 | */ |
0b246afa | 2146 | btrfs_apply_pending_changes(fs_info); |
3818aea2 | 2147 | |
5d4f98a2 | 2148 | /* commit_fs_roots gets rid of all the tree log roots, it is now |
e02119d5 CM |
2149 | * safe to free the root of tree log roots |
2150 | */ | |
0b246afa | 2151 | btrfs_free_log_root_tree(trans, fs_info); |
e02119d5 | 2152 | |
82bafb38 QW |
2153 | /* |
2154 | * commit_fs_roots() can call btrfs_save_ino_cache(), which generates | |
2155 | * new delayed refs. Must handle them or qgroup can be wrong. | |
2156 | */ | |
c79a70b1 | 2157 | ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); |
82bafb38 QW |
2158 | if (ret) { |
2159 | mutex_unlock(&fs_info->tree_log_mutex); | |
2160 | mutex_unlock(&fs_info->reloc_mutex); | |
2161 | goto scrub_continue; | |
2162 | } | |
2163 | ||
0ed4792a QW |
2164 | /* |
2165 | * Since fs roots are all committed, we can get a quite accurate | |
2166 | * new_roots. So let's do quota accounting. | |
2167 | */ | |
460fb20a | 2168 | ret = btrfs_qgroup_account_extents(trans); |
0ed4792a | 2169 | if (ret < 0) { |
0b246afa JM |
2170 | mutex_unlock(&fs_info->tree_log_mutex); |
2171 | mutex_unlock(&fs_info->reloc_mutex); | |
0ed4792a QW |
2172 | goto scrub_continue; |
2173 | } | |
2174 | ||
9386d8bc | 2175 | ret = commit_cowonly_roots(trans); |
49b25e05 | 2176 | if (ret) { |
0b246afa JM |
2177 | mutex_unlock(&fs_info->tree_log_mutex); |
2178 | mutex_unlock(&fs_info->reloc_mutex); | |
6cf7f77e | 2179 | goto scrub_continue; |
49b25e05 | 2180 | } |
54aa1f4d | 2181 | |
2cba30f1 MX |
2182 | /* |
2183 | * The tasks which save the space cache and inode cache may also | |
2184 | * update ->aborted, check it. | |
2185 | */ | |
20c7bcec | 2186 | if (unlikely(READ_ONCE(cur_trans->aborted))) { |
2cba30f1 | 2187 | ret = cur_trans->aborted; |
0b246afa JM |
2188 | mutex_unlock(&fs_info->tree_log_mutex); |
2189 | mutex_unlock(&fs_info->reloc_mutex); | |
6cf7f77e | 2190 | goto scrub_continue; |
2cba30f1 MX |
2191 | } |
2192 | ||
8b74c03e | 2193 | btrfs_prepare_extent_commit(fs_info); |
11833d66 | 2194 | |
0b246afa | 2195 | cur_trans = fs_info->running_transaction; |
5d4f98a2 | 2196 | |
0b246afa JM |
2197 | btrfs_set_root_node(&fs_info->tree_root->root_item, |
2198 | fs_info->tree_root->node); | |
2199 | list_add_tail(&fs_info->tree_root->dirty_list, | |
9e351cc8 | 2200 | &cur_trans->switch_commits); |
5d4f98a2 | 2201 | |
0b246afa JM |
2202 | btrfs_set_root_node(&fs_info->chunk_root->root_item, |
2203 | fs_info->chunk_root->node); | |
2204 | list_add_tail(&fs_info->chunk_root->dirty_list, | |
9e351cc8 JB |
2205 | &cur_trans->switch_commits); |
2206 | ||
16916a88 | 2207 | switch_commit_roots(cur_trans); |
5d4f98a2 | 2208 | |
ce93ec54 | 2209 | ASSERT(list_empty(&cur_trans->dirty_bgs)); |
1bbc621e | 2210 | ASSERT(list_empty(&cur_trans->io_bgs)); |
2ff7e61e | 2211 | update_super_roots(fs_info); |
e02119d5 | 2212 | |
0b246afa JM |
2213 | btrfs_set_super_log_root(fs_info->super_copy, 0); |
2214 | btrfs_set_super_log_root_level(fs_info->super_copy, 0); | |
2215 | memcpy(fs_info->super_for_commit, fs_info->super_copy, | |
2216 | sizeof(*fs_info->super_copy)); | |
ccd467d6 | 2217 | |
0b246afa | 2218 | btrfs_update_commit_device_size(fs_info); |
e9b919b1 | 2219 | btrfs_update_commit_device_bytes_used(cur_trans); |
935e5cc9 | 2220 | |
0b246afa JM |
2221 | clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags); |
2222 | clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags); | |
656f30db | 2223 | |
4fbcdf66 FM |
2224 | btrfs_trans_release_chunk_metadata(trans); |
2225 | ||
0b246afa | 2226 | spin_lock(&fs_info->trans_lock); |
4a9d8bde | 2227 | cur_trans->state = TRANS_STATE_UNBLOCKED; |
0b246afa JM |
2228 | fs_info->running_transaction = NULL; |
2229 | spin_unlock(&fs_info->trans_lock); | |
2230 | mutex_unlock(&fs_info->reloc_mutex); | |
b7ec40d7 | 2231 | |
0b246afa | 2232 | wake_up(&fs_info->transaction_wait); |
e6dcd2dc | 2233 | |
70458a58 | 2234 | ret = btrfs_write_and_wait_transaction(trans); |
49b25e05 | 2235 | if (ret) { |
0b246afa JM |
2236 | btrfs_handle_fs_error(fs_info, ret, |
2237 | "Error while writing out transaction"); | |
2238 | mutex_unlock(&fs_info->tree_log_mutex); | |
6cf7f77e | 2239 | goto scrub_continue; |
49b25e05 JM |
2240 | } |
2241 | ||
eece6a9c | 2242 | ret = write_all_supers(fs_info, 0); |
e02119d5 CM |
2243 | /* |
2244 | * the super is written, we can safely allow the tree-loggers | |
2245 | * to go about their business | |
2246 | */ | |
0b246afa | 2247 | mutex_unlock(&fs_info->tree_log_mutex); |
c1f32b7c AJ |
2248 | if (ret) |
2249 | goto scrub_continue; | |
e02119d5 | 2250 | |
5ead2dd0 | 2251 | btrfs_finish_extent_commit(trans); |
4313b399 | 2252 | |
3204d33c | 2253 | if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS, &cur_trans->flags)) |
0b246afa | 2254 | btrfs_clear_space_info_full(fs_info); |
13212b54 | 2255 | |
0b246afa | 2256 | fs_info->last_trans_committed = cur_trans->transid; |
4a9d8bde MX |
2257 | /* |
2258 | * We needn't acquire the lock here because there is no other task | |
2259 | * which can change it. | |
2260 | */ | |
2261 | cur_trans->state = TRANS_STATE_COMPLETED; | |
2c90e5d6 | 2262 | wake_up(&cur_trans->commit_wait); |
a514d638 | 2263 | clear_bit(BTRFS_FS_NEED_ASYNC_COMMIT, &fs_info->flags); |
3de4586c | 2264 | |
0b246afa | 2265 | spin_lock(&fs_info->trans_lock); |
13c5a93e | 2266 | list_del_init(&cur_trans->list); |
0b246afa | 2267 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 | 2268 | |
724e2315 JB |
2269 | btrfs_put_transaction(cur_trans); |
2270 | btrfs_put_transaction(cur_trans); | |
58176a96 | 2271 | |
0860adfd | 2272 | if (trans->type & __TRANS_FREEZABLE) |
0b246afa | 2273 | sb_end_intwrite(fs_info->sb); |
b2b5ef5c | 2274 | |
3a45bb20 | 2275 | trace_btrfs_transaction_commit(trans->root); |
1abe9b8a | 2276 | |
2ff7e61e | 2277 | btrfs_scrub_continue(fs_info); |
a2de733c | 2278 | |
9ed74f2d JB |
2279 | if (current->journal_info == trans) |
2280 | current->journal_info = NULL; | |
2281 | ||
2c90e5d6 | 2282 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
24bbcf04 | 2283 | |
9e7cc91a WX |
2284 | /* |
2285 | * If fs has been frozen, we can not handle delayed iputs, otherwise | |
2286 | * it'll result in deadlock about SB_FREEZE_FS. | |
2287 | */ | |
0b246afa | 2288 | if (current != fs_info->transaction_kthread && |
fac03c8d DS |
2289 | current != fs_info->cleaner_kthread && |
2290 | !test_bit(BTRFS_FS_FROZEN, &fs_info->flags)) | |
2ff7e61e | 2291 | btrfs_run_delayed_iputs(fs_info); |
24bbcf04 | 2292 | |
79154b1b | 2293 | return ret; |
49b25e05 | 2294 | |
6cf7f77e | 2295 | scrub_continue: |
2ff7e61e | 2296 | btrfs_scrub_continue(fs_info); |
49b25e05 | 2297 | cleanup_transaction: |
dc60c525 | 2298 | btrfs_trans_release_metadata(trans); |
4fbcdf66 | 2299 | btrfs_trans_release_chunk_metadata(trans); |
0e721106 | 2300 | trans->block_rsv = NULL; |
0b246afa | 2301 | btrfs_warn(fs_info, "Skipping commit of aborted transaction."); |
49b25e05 JM |
2302 | if (current->journal_info == trans) |
2303 | current->journal_info = NULL; | |
97cb39bb | 2304 | cleanup_transaction(trans, ret); |
49b25e05 JM |
2305 | |
2306 | return ret; | |
79154b1b CM |
2307 | } |
2308 | ||
d352ac68 | 2309 | /* |
9d1a2a3a DS |
2310 | * return < 0 if error |
2311 | * 0 if there are no more dead_roots at the time of call | |
2312 | * 1 there are more to be processed, call me again | |
2313 | * | |
2314 | * The return value indicates there are certainly more snapshots to delete, but | |
2315 | * if there comes a new one during processing, it may return 0. We don't mind, | |
2316 | * because btrfs_commit_super will poke cleaner thread and it will process it a | |
2317 | * few seconds later. | |
d352ac68 | 2318 | */ |
9d1a2a3a | 2319 | int btrfs_clean_one_deleted_snapshot(struct btrfs_root *root) |
e9d0b13b | 2320 | { |
9d1a2a3a | 2321 | int ret; |
5d4f98a2 YZ |
2322 | struct btrfs_fs_info *fs_info = root->fs_info; |
2323 | ||
a4abeea4 | 2324 | spin_lock(&fs_info->trans_lock); |
9d1a2a3a DS |
2325 | if (list_empty(&fs_info->dead_roots)) { |
2326 | spin_unlock(&fs_info->trans_lock); | |
2327 | return 0; | |
2328 | } | |
2329 | root = list_first_entry(&fs_info->dead_roots, | |
2330 | struct btrfs_root, root_list); | |
cfad392b | 2331 | list_del_init(&root->root_list); |
a4abeea4 | 2332 | spin_unlock(&fs_info->trans_lock); |
e9d0b13b | 2333 | |
ab8d0fc4 | 2334 | btrfs_debug(fs_info, "cleaner removing %llu", root->objectid); |
76dda93c | 2335 | |
9d1a2a3a | 2336 | btrfs_kill_all_delayed_nodes(root); |
16cdcec7 | 2337 | |
9d1a2a3a DS |
2338 | if (btrfs_header_backref_rev(root->node) < |
2339 | BTRFS_MIXED_BACKREF_REV) | |
2340 | ret = btrfs_drop_snapshot(root, NULL, 0, 0); | |
2341 | else | |
2342 | ret = btrfs_drop_snapshot(root, NULL, 1, 0); | |
32471dc2 | 2343 | |
6596a928 | 2344 | return (ret < 0) ? 0 : 1; |
e9d0b13b | 2345 | } |
572d9ab7 DS |
2346 | |
2347 | void btrfs_apply_pending_changes(struct btrfs_fs_info *fs_info) | |
2348 | { | |
2349 | unsigned long prev; | |
2350 | unsigned long bit; | |
2351 | ||
6c9fe14f | 2352 | prev = xchg(&fs_info->pending_changes, 0); |
572d9ab7 DS |
2353 | if (!prev) |
2354 | return; | |
2355 | ||
7e1876ac DS |
2356 | bit = 1 << BTRFS_PENDING_SET_INODE_MAP_CACHE; |
2357 | if (prev & bit) | |
2358 | btrfs_set_opt(fs_info->mount_opt, INODE_MAP_CACHE); | |
2359 | prev &= ~bit; | |
2360 | ||
2361 | bit = 1 << BTRFS_PENDING_CLEAR_INODE_MAP_CACHE; | |
2362 | if (prev & bit) | |
2363 | btrfs_clear_opt(fs_info->mount_opt, INODE_MAP_CACHE); | |
2364 | prev &= ~bit; | |
2365 | ||
d51033d0 DS |
2366 | bit = 1 << BTRFS_PENDING_COMMIT; |
2367 | if (prev & bit) | |
2368 | btrfs_debug(fs_info, "pending commit done"); | |
2369 | prev &= ~bit; | |
2370 | ||
572d9ab7 DS |
2371 | if (prev) |
2372 | btrfs_warn(fs_info, | |
2373 | "unknown pending changes left 0x%lx, ignoring", prev); | |
2374 | } |