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)); |
5c9d028b LB |
47 | WARN_ON(!RB_EMPTY_ROOT( |
48 | &transaction->delayed_refs.href_root.rb_root)); | |
1262133b | 49 | if (transaction->delayed_refs.pending_csums) |
ab8d0fc4 JM |
50 | btrfs_err(transaction->fs_info, |
51 | "pending csums is %llu", | |
52 | transaction->delayed_refs.pending_csums); | |
6df9a95e JB |
53 | while (!list_empty(&transaction->pending_chunks)) { |
54 | struct extent_map *em; | |
55 | ||
56 | em = list_first_entry(&transaction->pending_chunks, | |
57 | struct extent_map, list); | |
58 | list_del_init(&em->list); | |
59 | free_extent_map(em); | |
60 | } | |
7785a663 FM |
61 | /* |
62 | * If any block groups are found in ->deleted_bgs then it's | |
63 | * because the transaction was aborted and a commit did not | |
64 | * happen (things failed before writing the new superblock | |
65 | * and calling btrfs_finish_extent_commit()), so we can not | |
66 | * discard the physical locations of the block groups. | |
67 | */ | |
68 | while (!list_empty(&transaction->deleted_bgs)) { | |
69 | struct btrfs_block_group_cache *cache; | |
70 | ||
71 | cache = list_first_entry(&transaction->deleted_bgs, | |
72 | struct btrfs_block_group_cache, | |
73 | bg_list); | |
74 | list_del_init(&cache->bg_list); | |
75 | btrfs_put_block_group_trimming(cache); | |
76 | btrfs_put_block_group(cache); | |
77 | } | |
4b5faeac | 78 | kfree(transaction); |
78fae27e | 79 | } |
79154b1b CM |
80 | } |
81 | ||
663dfbb0 FM |
82 | static void clear_btree_io_tree(struct extent_io_tree *tree) |
83 | { | |
84 | spin_lock(&tree->lock); | |
b666a9cd DS |
85 | /* |
86 | * Do a single barrier for the waitqueue_active check here, the state | |
87 | * of the waitqueue should not change once clear_btree_io_tree is | |
88 | * called. | |
89 | */ | |
90 | smp_mb(); | |
663dfbb0 FM |
91 | while (!RB_EMPTY_ROOT(&tree->state)) { |
92 | struct rb_node *node; | |
93 | struct extent_state *state; | |
94 | ||
95 | node = rb_first(&tree->state); | |
96 | state = rb_entry(node, struct extent_state, rb_node); | |
97 | rb_erase(&state->rb_node, &tree->state); | |
98 | RB_CLEAR_NODE(&state->rb_node); | |
99 | /* | |
100 | * btree io trees aren't supposed to have tasks waiting for | |
101 | * changes in the flags of extent states ever. | |
102 | */ | |
103 | ASSERT(!waitqueue_active(&state->wq)); | |
104 | free_extent_state(state); | |
351810c1 DS |
105 | |
106 | cond_resched_lock(&tree->lock); | |
663dfbb0 FM |
107 | } |
108 | spin_unlock(&tree->lock); | |
109 | } | |
110 | ||
16916a88 | 111 | static noinline void switch_commit_roots(struct btrfs_transaction *trans) |
817d52f8 | 112 | { |
16916a88 | 113 | struct btrfs_fs_info *fs_info = trans->fs_info; |
9e351cc8 JB |
114 | struct btrfs_root *root, *tmp; |
115 | ||
116 | down_write(&fs_info->commit_root_sem); | |
117 | list_for_each_entry_safe(root, tmp, &trans->switch_commits, | |
118 | dirty_list) { | |
119 | list_del_init(&root->dirty_list); | |
120 | free_extent_buffer(root->commit_root); | |
121 | root->commit_root = btrfs_root_node(root); | |
4fd786e6 | 122 | if (is_fstree(root->root_key.objectid)) |
9e351cc8 | 123 | btrfs_unpin_free_ino(root); |
663dfbb0 | 124 | clear_btree_io_tree(&root->dirty_log_pages); |
9e351cc8 | 125 | } |
2b9dbef2 JB |
126 | |
127 | /* We can free old roots now. */ | |
128 | spin_lock(&trans->dropped_roots_lock); | |
129 | while (!list_empty(&trans->dropped_roots)) { | |
130 | root = list_first_entry(&trans->dropped_roots, | |
131 | struct btrfs_root, root_list); | |
132 | list_del_init(&root->root_list); | |
133 | spin_unlock(&trans->dropped_roots_lock); | |
134 | btrfs_drop_and_free_fs_root(fs_info, root); | |
135 | spin_lock(&trans->dropped_roots_lock); | |
136 | } | |
137 | spin_unlock(&trans->dropped_roots_lock); | |
9e351cc8 | 138 | up_write(&fs_info->commit_root_sem); |
817d52f8 JB |
139 | } |
140 | ||
0860adfd MX |
141 | static inline void extwriter_counter_inc(struct btrfs_transaction *trans, |
142 | unsigned int type) | |
143 | { | |
144 | if (type & TRANS_EXTWRITERS) | |
145 | atomic_inc(&trans->num_extwriters); | |
146 | } | |
147 | ||
148 | static inline void extwriter_counter_dec(struct btrfs_transaction *trans, | |
149 | unsigned int type) | |
150 | { | |
151 | if (type & TRANS_EXTWRITERS) | |
152 | atomic_dec(&trans->num_extwriters); | |
153 | } | |
154 | ||
155 | static inline void extwriter_counter_init(struct btrfs_transaction *trans, | |
156 | unsigned int type) | |
157 | { | |
158 | atomic_set(&trans->num_extwriters, ((type & TRANS_EXTWRITERS) ? 1 : 0)); | |
159 | } | |
160 | ||
161 | static inline int extwriter_counter_read(struct btrfs_transaction *trans) | |
162 | { | |
163 | return atomic_read(&trans->num_extwriters); | |
178260b2 MX |
164 | } |
165 | ||
d352ac68 CM |
166 | /* |
167 | * either allocate a new transaction or hop into the existing one | |
168 | */ | |
2ff7e61e JM |
169 | static noinline int join_transaction(struct btrfs_fs_info *fs_info, |
170 | unsigned int type) | |
79154b1b CM |
171 | { |
172 | struct btrfs_transaction *cur_trans; | |
a4abeea4 | 173 | |
19ae4e81 | 174 | spin_lock(&fs_info->trans_lock); |
d43317dc | 175 | loop: |
49b25e05 | 176 | /* The file system has been taken offline. No new transactions. */ |
87533c47 | 177 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { |
19ae4e81 | 178 | spin_unlock(&fs_info->trans_lock); |
49b25e05 JM |
179 | return -EROFS; |
180 | } | |
181 | ||
19ae4e81 | 182 | cur_trans = fs_info->running_transaction; |
a4abeea4 | 183 | if (cur_trans) { |
871383be | 184 | if (cur_trans->aborted) { |
19ae4e81 | 185 | spin_unlock(&fs_info->trans_lock); |
49b25e05 | 186 | return cur_trans->aborted; |
871383be | 187 | } |
4a9d8bde | 188 | if (btrfs_blocked_trans_types[cur_trans->state] & type) { |
178260b2 MX |
189 | spin_unlock(&fs_info->trans_lock); |
190 | return -EBUSY; | |
191 | } | |
9b64f57d | 192 | refcount_inc(&cur_trans->use_count); |
13c5a93e | 193 | atomic_inc(&cur_trans->num_writers); |
0860adfd | 194 | extwriter_counter_inc(cur_trans, type); |
19ae4e81 | 195 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 | 196 | return 0; |
79154b1b | 197 | } |
19ae4e81 | 198 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 | 199 | |
354aa0fb MX |
200 | /* |
201 | * If we are ATTACH, we just want to catch the current transaction, | |
202 | * and commit it. If there is no transaction, just return ENOENT. | |
203 | */ | |
204 | if (type == TRANS_ATTACH) | |
205 | return -ENOENT; | |
206 | ||
4a9d8bde MX |
207 | /* |
208 | * JOIN_NOLOCK only happens during the transaction commit, so | |
209 | * it is impossible that ->running_transaction is NULL | |
210 | */ | |
211 | BUG_ON(type == TRANS_JOIN_NOLOCK); | |
212 | ||
4b5faeac | 213 | cur_trans = kmalloc(sizeof(*cur_trans), GFP_NOFS); |
a4abeea4 JB |
214 | if (!cur_trans) |
215 | return -ENOMEM; | |
d43317dc | 216 | |
19ae4e81 JS |
217 | spin_lock(&fs_info->trans_lock); |
218 | if (fs_info->running_transaction) { | |
d43317dc CM |
219 | /* |
220 | * someone started a transaction after we unlocked. Make sure | |
4a9d8bde | 221 | * to redo the checks above |
d43317dc | 222 | */ |
4b5faeac | 223 | kfree(cur_trans); |
d43317dc | 224 | goto loop; |
87533c47 | 225 | } else if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { |
e4b50e14 | 226 | spin_unlock(&fs_info->trans_lock); |
4b5faeac | 227 | kfree(cur_trans); |
7b8b92af | 228 | return -EROFS; |
79154b1b | 229 | } |
d43317dc | 230 | |
ab8d0fc4 | 231 | cur_trans->fs_info = fs_info; |
a4abeea4 | 232 | atomic_set(&cur_trans->num_writers, 1); |
0860adfd | 233 | extwriter_counter_init(cur_trans, type); |
a4abeea4 JB |
234 | init_waitqueue_head(&cur_trans->writer_wait); |
235 | init_waitqueue_head(&cur_trans->commit_wait); | |
161c3549 | 236 | init_waitqueue_head(&cur_trans->pending_wait); |
4a9d8bde | 237 | cur_trans->state = TRANS_STATE_RUNNING; |
a4abeea4 JB |
238 | /* |
239 | * One for this trans handle, one so it will live on until we | |
240 | * commit the transaction. | |
241 | */ | |
9b64f57d | 242 | refcount_set(&cur_trans->use_count, 2); |
161c3549 | 243 | atomic_set(&cur_trans->pending_ordered, 0); |
3204d33c | 244 | cur_trans->flags = 0; |
afd48513 | 245 | cur_trans->start_time = ktime_get_seconds(); |
a4abeea4 | 246 | |
a099d0fd AM |
247 | memset(&cur_trans->delayed_refs, 0, sizeof(cur_trans->delayed_refs)); |
248 | ||
5c9d028b | 249 | cur_trans->delayed_refs.href_root = RB_ROOT_CACHED; |
3368d001 | 250 | cur_trans->delayed_refs.dirty_extent_root = RB_ROOT; |
d7df2c79 | 251 | atomic_set(&cur_trans->delayed_refs.num_entries, 0); |
20b297d6 JS |
252 | |
253 | /* | |
254 | * although the tree mod log is per file system and not per transaction, | |
255 | * the log must never go across transaction boundaries. | |
256 | */ | |
257 | smp_mb(); | |
31b1a2bd | 258 | if (!list_empty(&fs_info->tree_mod_seq_list)) |
5d163e0e | 259 | WARN(1, KERN_ERR "BTRFS: tree_mod_seq_list not empty when creating a fresh transaction\n"); |
31b1a2bd | 260 | if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log)) |
5d163e0e | 261 | WARN(1, KERN_ERR "BTRFS: tree_mod_log rb tree not empty when creating a fresh transaction\n"); |
fc36ed7e | 262 | atomic64_set(&fs_info->tree_mod_seq, 0); |
20b297d6 | 263 | |
a4abeea4 JB |
264 | spin_lock_init(&cur_trans->delayed_refs.lock); |
265 | ||
266 | INIT_LIST_HEAD(&cur_trans->pending_snapshots); | |
6df9a95e | 267 | INIT_LIST_HEAD(&cur_trans->pending_chunks); |
9e351cc8 | 268 | INIT_LIST_HEAD(&cur_trans->switch_commits); |
ce93ec54 | 269 | INIT_LIST_HEAD(&cur_trans->dirty_bgs); |
1bbc621e | 270 | INIT_LIST_HEAD(&cur_trans->io_bgs); |
2b9dbef2 | 271 | INIT_LIST_HEAD(&cur_trans->dropped_roots); |
1bbc621e | 272 | mutex_init(&cur_trans->cache_write_mutex); |
cb723e49 | 273 | cur_trans->num_dirty_bgs = 0; |
ce93ec54 | 274 | spin_lock_init(&cur_trans->dirty_bgs_lock); |
e33e17ee | 275 | INIT_LIST_HEAD(&cur_trans->deleted_bgs); |
2b9dbef2 | 276 | spin_lock_init(&cur_trans->dropped_roots_lock); |
19ae4e81 | 277 | list_add_tail(&cur_trans->list, &fs_info->trans_list); |
a4abeea4 | 278 | extent_io_tree_init(&cur_trans->dirty_pages, |
c6100a4b | 279 | fs_info->btree_inode); |
19ae4e81 JS |
280 | fs_info->generation++; |
281 | cur_trans->transid = fs_info->generation; | |
282 | fs_info->running_transaction = cur_trans; | |
49b25e05 | 283 | cur_trans->aborted = 0; |
19ae4e81 | 284 | spin_unlock(&fs_info->trans_lock); |
15ee9bc7 | 285 | |
79154b1b CM |
286 | return 0; |
287 | } | |
288 | ||
d352ac68 | 289 | /* |
d397712b CM |
290 | * this does all the record keeping required to make sure that a reference |
291 | * counted root is properly recorded in a given transaction. This is required | |
292 | * to make sure the old root from before we joined the transaction is deleted | |
293 | * when the transaction commits | |
d352ac68 | 294 | */ |
7585717f | 295 | static int record_root_in_trans(struct btrfs_trans_handle *trans, |
6426c7ad QW |
296 | struct btrfs_root *root, |
297 | int force) | |
6702ed49 | 298 | { |
0b246afa JM |
299 | struct btrfs_fs_info *fs_info = root->fs_info; |
300 | ||
6426c7ad QW |
301 | if ((test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
302 | root->last_trans < trans->transid) || force) { | |
0b246afa | 303 | WARN_ON(root == fs_info->extent_root); |
4d31778a | 304 | WARN_ON(!force && root->commit_root != root->node); |
5d4f98a2 | 305 | |
7585717f | 306 | /* |
27cdeb70 | 307 | * see below for IN_TRANS_SETUP usage rules |
7585717f CM |
308 | * we have the reloc mutex held now, so there |
309 | * is only one writer in this function | |
310 | */ | |
27cdeb70 | 311 | set_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); |
7585717f | 312 | |
27cdeb70 | 313 | /* make sure readers find IN_TRANS_SETUP before |
7585717f CM |
314 | * they find our root->last_trans update |
315 | */ | |
316 | smp_wmb(); | |
317 | ||
0b246afa | 318 | spin_lock(&fs_info->fs_roots_radix_lock); |
6426c7ad | 319 | if (root->last_trans == trans->transid && !force) { |
0b246afa | 320 | spin_unlock(&fs_info->fs_roots_radix_lock); |
a4abeea4 JB |
321 | return 0; |
322 | } | |
0b246afa JM |
323 | radix_tree_tag_set(&fs_info->fs_roots_radix, |
324 | (unsigned long)root->root_key.objectid, | |
325 | BTRFS_ROOT_TRANS_TAG); | |
326 | spin_unlock(&fs_info->fs_roots_radix_lock); | |
7585717f CM |
327 | root->last_trans = trans->transid; |
328 | ||
329 | /* this is pretty tricky. We don't want to | |
330 | * take the relocation lock in btrfs_record_root_in_trans | |
331 | * unless we're really doing the first setup for this root in | |
332 | * this transaction. | |
333 | * | |
334 | * Normally we'd use root->last_trans as a flag to decide | |
335 | * if we want to take the expensive mutex. | |
336 | * | |
337 | * But, we have to set root->last_trans before we | |
338 | * init the relocation root, otherwise, we trip over warnings | |
339 | * in ctree.c. The solution used here is to flag ourselves | |
27cdeb70 | 340 | * with root IN_TRANS_SETUP. When this is 1, we're still |
7585717f CM |
341 | * fixing up the reloc trees and everyone must wait. |
342 | * | |
343 | * When this is zero, they can trust root->last_trans and fly | |
344 | * through btrfs_record_root_in_trans without having to take the | |
345 | * lock. smp_wmb() makes sure that all the writes above are | |
346 | * done before we pop in the zero below | |
347 | */ | |
5d4f98a2 | 348 | btrfs_init_reloc_root(trans, root); |
c7548af6 | 349 | smp_mb__before_atomic(); |
27cdeb70 | 350 | clear_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); |
5d4f98a2 YZ |
351 | } |
352 | return 0; | |
353 | } | |
bcc63abb | 354 | |
7585717f | 355 | |
2b9dbef2 JB |
356 | void btrfs_add_dropped_root(struct btrfs_trans_handle *trans, |
357 | struct btrfs_root *root) | |
358 | { | |
0b246afa | 359 | struct btrfs_fs_info *fs_info = root->fs_info; |
2b9dbef2 JB |
360 | struct btrfs_transaction *cur_trans = trans->transaction; |
361 | ||
362 | /* Add ourselves to the transaction dropped list */ | |
363 | spin_lock(&cur_trans->dropped_roots_lock); | |
364 | list_add_tail(&root->root_list, &cur_trans->dropped_roots); | |
365 | spin_unlock(&cur_trans->dropped_roots_lock); | |
366 | ||
367 | /* Make sure we don't try to update the root at commit time */ | |
0b246afa JM |
368 | spin_lock(&fs_info->fs_roots_radix_lock); |
369 | radix_tree_tag_clear(&fs_info->fs_roots_radix, | |
2b9dbef2 JB |
370 | (unsigned long)root->root_key.objectid, |
371 | BTRFS_ROOT_TRANS_TAG); | |
0b246afa | 372 | spin_unlock(&fs_info->fs_roots_radix_lock); |
2b9dbef2 JB |
373 | } |
374 | ||
7585717f CM |
375 | int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, |
376 | struct btrfs_root *root) | |
377 | { | |
0b246afa JM |
378 | struct btrfs_fs_info *fs_info = root->fs_info; |
379 | ||
27cdeb70 | 380 | if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state)) |
7585717f CM |
381 | return 0; |
382 | ||
383 | /* | |
27cdeb70 | 384 | * see record_root_in_trans for comments about IN_TRANS_SETUP usage |
7585717f CM |
385 | * and barriers |
386 | */ | |
387 | smp_rmb(); | |
388 | if (root->last_trans == trans->transid && | |
27cdeb70 | 389 | !test_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state)) |
7585717f CM |
390 | return 0; |
391 | ||
0b246afa | 392 | mutex_lock(&fs_info->reloc_mutex); |
6426c7ad | 393 | record_root_in_trans(trans, root, 0); |
0b246afa | 394 | mutex_unlock(&fs_info->reloc_mutex); |
7585717f CM |
395 | |
396 | return 0; | |
397 | } | |
398 | ||
4a9d8bde MX |
399 | static inline int is_transaction_blocked(struct btrfs_transaction *trans) |
400 | { | |
401 | return (trans->state >= TRANS_STATE_BLOCKED && | |
501407aa JB |
402 | trans->state < TRANS_STATE_UNBLOCKED && |
403 | !trans->aborted); | |
4a9d8bde MX |
404 | } |
405 | ||
d352ac68 CM |
406 | /* wait for commit against the current transaction to become unblocked |
407 | * when this is done, it is safe to start a new transaction, but the current | |
408 | * transaction might not be fully on disk. | |
409 | */ | |
2ff7e61e | 410 | static void wait_current_trans(struct btrfs_fs_info *fs_info) |
79154b1b | 411 | { |
f9295749 | 412 | struct btrfs_transaction *cur_trans; |
79154b1b | 413 | |
0b246afa JM |
414 | spin_lock(&fs_info->trans_lock); |
415 | cur_trans = fs_info->running_transaction; | |
4a9d8bde | 416 | if (cur_trans && is_transaction_blocked(cur_trans)) { |
9b64f57d | 417 | refcount_inc(&cur_trans->use_count); |
0b246afa | 418 | spin_unlock(&fs_info->trans_lock); |
72d63ed6 | 419 | |
0b246afa | 420 | wait_event(fs_info->transaction_wait, |
501407aa JB |
421 | cur_trans->state >= TRANS_STATE_UNBLOCKED || |
422 | cur_trans->aborted); | |
724e2315 | 423 | btrfs_put_transaction(cur_trans); |
a4abeea4 | 424 | } else { |
0b246afa | 425 | spin_unlock(&fs_info->trans_lock); |
f9295749 | 426 | } |
37d1aeee CM |
427 | } |
428 | ||
2ff7e61e | 429 | static int may_wait_transaction(struct btrfs_fs_info *fs_info, int type) |
a22285a6 | 430 | { |
0b246afa | 431 | if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) |
a4abeea4 JB |
432 | return 0; |
433 | ||
92e2f7e3 | 434 | if (type == TRANS_START) |
a22285a6 | 435 | return 1; |
a4abeea4 | 436 | |
a22285a6 YZ |
437 | return 0; |
438 | } | |
439 | ||
20dd2cbf MX |
440 | static inline bool need_reserve_reloc_root(struct btrfs_root *root) |
441 | { | |
0b246afa JM |
442 | struct btrfs_fs_info *fs_info = root->fs_info; |
443 | ||
444 | if (!fs_info->reloc_ctl || | |
27cdeb70 | 445 | !test_bit(BTRFS_ROOT_REF_COWS, &root->state) || |
20dd2cbf MX |
446 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || |
447 | root->reloc_root) | |
448 | return false; | |
449 | ||
450 | return true; | |
451 | } | |
452 | ||
08e007d2 | 453 | static struct btrfs_trans_handle * |
5aed1dd8 | 454 | start_transaction(struct btrfs_root *root, unsigned int num_items, |
003d7c59 JM |
455 | unsigned int type, enum btrfs_reserve_flush_enum flush, |
456 | bool enforce_qgroups) | |
37d1aeee | 457 | { |
0b246afa JM |
458 | struct btrfs_fs_info *fs_info = root->fs_info; |
459 | ||
a22285a6 YZ |
460 | struct btrfs_trans_handle *h; |
461 | struct btrfs_transaction *cur_trans; | |
b5009945 | 462 | u64 num_bytes = 0; |
c5567237 | 463 | u64 qgroup_reserved = 0; |
20dd2cbf MX |
464 | bool reloc_reserved = false; |
465 | int ret; | |
acce952b | 466 | |
46c4e71e | 467 | /* Send isn't supposed to start transactions. */ |
2755a0de | 468 | ASSERT(current->journal_info != BTRFS_SEND_TRANS_STUB); |
46c4e71e | 469 | |
0b246afa | 470 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) |
acce952b | 471 | return ERR_PTR(-EROFS); |
2a1eb461 | 472 | |
46c4e71e | 473 | if (current->journal_info) { |
0860adfd | 474 | WARN_ON(type & TRANS_EXTWRITERS); |
2a1eb461 | 475 | h = current->journal_info; |
b50fff81 DS |
476 | refcount_inc(&h->use_count); |
477 | WARN_ON(refcount_read(&h->use_count) > 2); | |
2a1eb461 JB |
478 | h->orig_rsv = h->block_rsv; |
479 | h->block_rsv = NULL; | |
480 | goto got_it; | |
481 | } | |
b5009945 JB |
482 | |
483 | /* | |
484 | * Do the reservation before we join the transaction so we can do all | |
485 | * the appropriate flushing if need be. | |
486 | */ | |
003d7c59 | 487 | if (num_items && root != fs_info->chunk_root) { |
0b246afa | 488 | qgroup_reserved = num_items * fs_info->nodesize; |
733e03a0 QW |
489 | ret = btrfs_qgroup_reserve_meta_pertrans(root, qgroup_reserved, |
490 | enforce_qgroups); | |
7174109c QW |
491 | if (ret) |
492 | return ERR_PTR(ret); | |
c5567237 | 493 | |
0b246afa | 494 | num_bytes = btrfs_calc_trans_metadata_size(fs_info, num_items); |
20dd2cbf MX |
495 | /* |
496 | * Do the reservation for the relocation root creation | |
497 | */ | |
ee39b432 | 498 | if (need_reserve_reloc_root(root)) { |
0b246afa | 499 | num_bytes += fs_info->nodesize; |
20dd2cbf MX |
500 | reloc_reserved = true; |
501 | } | |
502 | ||
0b246afa | 503 | ret = btrfs_block_rsv_add(root, &fs_info->trans_block_rsv, |
08e007d2 | 504 | num_bytes, flush); |
b5009945 | 505 | if (ret) |
843fcf35 | 506 | goto reserve_fail; |
b5009945 | 507 | } |
a22285a6 | 508 | again: |
f2f767e7 | 509 | h = kmem_cache_zalloc(btrfs_trans_handle_cachep, GFP_NOFS); |
843fcf35 MX |
510 | if (!h) { |
511 | ret = -ENOMEM; | |
512 | goto alloc_fail; | |
513 | } | |
37d1aeee | 514 | |
98114659 JB |
515 | /* |
516 | * If we are JOIN_NOLOCK we're already committing a transaction and | |
517 | * waiting on this guy, so we don't need to do the sb_start_intwrite | |
518 | * because we're already holding a ref. We need this because we could | |
519 | * have raced in and did an fsync() on a file which can kick a commit | |
520 | * and then we deadlock with somebody doing a freeze. | |
354aa0fb MX |
521 | * |
522 | * If we are ATTACH, it means we just want to catch the current | |
523 | * transaction and commit it, so we needn't do sb_start_intwrite(). | |
98114659 | 524 | */ |
0860adfd | 525 | if (type & __TRANS_FREEZABLE) |
0b246afa | 526 | sb_start_intwrite(fs_info->sb); |
b2b5ef5c | 527 | |
2ff7e61e JM |
528 | if (may_wait_transaction(fs_info, type)) |
529 | wait_current_trans(fs_info); | |
a22285a6 | 530 | |
a4abeea4 | 531 | do { |
2ff7e61e | 532 | ret = join_transaction(fs_info, type); |
178260b2 | 533 | if (ret == -EBUSY) { |
2ff7e61e | 534 | wait_current_trans(fs_info); |
178260b2 MX |
535 | if (unlikely(type == TRANS_ATTACH)) |
536 | ret = -ENOENT; | |
537 | } | |
a4abeea4 JB |
538 | } while (ret == -EBUSY); |
539 | ||
a43f7f82 | 540 | if (ret < 0) |
843fcf35 | 541 | goto join_fail; |
0f7d52f4 | 542 | |
0b246afa | 543 | cur_trans = fs_info->running_transaction; |
a22285a6 YZ |
544 | |
545 | h->transid = cur_trans->transid; | |
546 | h->transaction = cur_trans; | |
d13603ef | 547 | h->root = root; |
b50fff81 | 548 | refcount_set(&h->use_count, 1); |
64b63580 | 549 | h->fs_info = root->fs_info; |
7174109c | 550 | |
a698d075 | 551 | h->type = type; |
d9a0540a | 552 | h->can_flush_pending_bgs = true; |
ea658bad | 553 | INIT_LIST_HEAD(&h->new_bgs); |
b7ec40d7 | 554 | |
a22285a6 | 555 | smp_mb(); |
4a9d8bde | 556 | if (cur_trans->state >= TRANS_STATE_BLOCKED && |
2ff7e61e | 557 | may_wait_transaction(fs_info, type)) { |
abdd2e80 | 558 | current->journal_info = h; |
3a45bb20 | 559 | btrfs_commit_transaction(h); |
a22285a6 YZ |
560 | goto again; |
561 | } | |
562 | ||
b5009945 | 563 | if (num_bytes) { |
0b246afa | 564 | trace_btrfs_space_reservation(fs_info, "transaction", |
2bcc0328 | 565 | h->transid, num_bytes, 1); |
0b246afa | 566 | h->block_rsv = &fs_info->trans_block_rsv; |
b5009945 | 567 | h->bytes_reserved = num_bytes; |
20dd2cbf | 568 | h->reloc_reserved = reloc_reserved; |
a22285a6 | 569 | } |
9ed74f2d | 570 | |
2a1eb461 | 571 | got_it: |
a4abeea4 | 572 | btrfs_record_root_in_trans(h, root); |
a22285a6 | 573 | |
bcf3a3e7 | 574 | if (!current->journal_info) |
a22285a6 | 575 | current->journal_info = h; |
79154b1b | 576 | return h; |
843fcf35 MX |
577 | |
578 | join_fail: | |
0860adfd | 579 | if (type & __TRANS_FREEZABLE) |
0b246afa | 580 | sb_end_intwrite(fs_info->sb); |
843fcf35 MX |
581 | kmem_cache_free(btrfs_trans_handle_cachep, h); |
582 | alloc_fail: | |
583 | if (num_bytes) | |
2ff7e61e | 584 | btrfs_block_rsv_release(fs_info, &fs_info->trans_block_rsv, |
843fcf35 MX |
585 | num_bytes); |
586 | reserve_fail: | |
733e03a0 | 587 | btrfs_qgroup_free_meta_pertrans(root, qgroup_reserved); |
843fcf35 | 588 | return ERR_PTR(ret); |
79154b1b CM |
589 | } |
590 | ||
f9295749 | 591 | struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, |
5aed1dd8 | 592 | unsigned int num_items) |
f9295749 | 593 | { |
08e007d2 | 594 | return start_transaction(root, num_items, TRANS_START, |
003d7c59 | 595 | BTRFS_RESERVE_FLUSH_ALL, true); |
f9295749 | 596 | } |
003d7c59 | 597 | |
8eab77ff FM |
598 | struct btrfs_trans_handle *btrfs_start_transaction_fallback_global_rsv( |
599 | struct btrfs_root *root, | |
600 | unsigned int num_items, | |
601 | int min_factor) | |
602 | { | |
0b246afa | 603 | struct btrfs_fs_info *fs_info = root->fs_info; |
8eab77ff FM |
604 | struct btrfs_trans_handle *trans; |
605 | u64 num_bytes; | |
606 | int ret; | |
607 | ||
003d7c59 JM |
608 | /* |
609 | * We have two callers: unlink and block group removal. The | |
610 | * former should succeed even if we will temporarily exceed | |
611 | * quota and the latter operates on the extent root so | |
612 | * qgroup enforcement is ignored anyway. | |
613 | */ | |
614 | trans = start_transaction(root, num_items, TRANS_START, | |
615 | BTRFS_RESERVE_FLUSH_ALL, false); | |
8eab77ff FM |
616 | if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC) |
617 | return trans; | |
618 | ||
619 | trans = btrfs_start_transaction(root, 0); | |
620 | if (IS_ERR(trans)) | |
621 | return trans; | |
622 | ||
0b246afa JM |
623 | num_bytes = btrfs_calc_trans_metadata_size(fs_info, num_items); |
624 | ret = btrfs_cond_migrate_bytes(fs_info, &fs_info->trans_block_rsv, | |
625 | num_bytes, min_factor); | |
8eab77ff | 626 | if (ret) { |
3a45bb20 | 627 | btrfs_end_transaction(trans); |
8eab77ff FM |
628 | return ERR_PTR(ret); |
629 | } | |
630 | ||
0b246afa | 631 | trans->block_rsv = &fs_info->trans_block_rsv; |
8eab77ff | 632 | trans->bytes_reserved = num_bytes; |
0b246afa | 633 | trace_btrfs_space_reservation(fs_info, "transaction", |
88d3a5aa | 634 | trans->transid, num_bytes, 1); |
8eab77ff FM |
635 | |
636 | return trans; | |
637 | } | |
8407aa46 | 638 | |
7a7eaa40 | 639 | struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root) |
f9295749 | 640 | { |
003d7c59 JM |
641 | return start_transaction(root, 0, TRANS_JOIN, BTRFS_RESERVE_NO_FLUSH, |
642 | true); | |
f9295749 CM |
643 | } |
644 | ||
7a7eaa40 | 645 | struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root) |
0af3d00b | 646 | { |
575a75d6 | 647 | return start_transaction(root, 0, TRANS_JOIN_NOLOCK, |
003d7c59 | 648 | BTRFS_RESERVE_NO_FLUSH, true); |
0af3d00b JB |
649 | } |
650 | ||
d4edf39b MX |
651 | /* |
652 | * btrfs_attach_transaction() - catch the running transaction | |
653 | * | |
654 | * It is used when we want to commit the current the transaction, but | |
655 | * don't want to start a new one. | |
656 | * | |
657 | * Note: If this function return -ENOENT, it just means there is no | |
658 | * running transaction. But it is possible that the inactive transaction | |
659 | * is still in the memory, not fully on disk. If you hope there is no | |
660 | * inactive transaction in the fs when -ENOENT is returned, you should | |
661 | * invoke | |
662 | * btrfs_attach_transaction_barrier() | |
663 | */ | |
354aa0fb | 664 | struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root) |
60376ce4 | 665 | { |
575a75d6 | 666 | return start_transaction(root, 0, TRANS_ATTACH, |
003d7c59 | 667 | BTRFS_RESERVE_NO_FLUSH, true); |
60376ce4 JB |
668 | } |
669 | ||
d4edf39b | 670 | /* |
90b6d283 | 671 | * btrfs_attach_transaction_barrier() - catch the running transaction |
d4edf39b MX |
672 | * |
673 | * It is similar to the above function, the differentia is this one | |
674 | * will wait for all the inactive transactions until they fully | |
675 | * complete. | |
676 | */ | |
677 | struct btrfs_trans_handle * | |
678 | btrfs_attach_transaction_barrier(struct btrfs_root *root) | |
679 | { | |
680 | struct btrfs_trans_handle *trans; | |
681 | ||
575a75d6 | 682 | trans = start_transaction(root, 0, TRANS_ATTACH, |
003d7c59 | 683 | BTRFS_RESERVE_NO_FLUSH, true); |
8d9e220c | 684 | if (trans == ERR_PTR(-ENOENT)) |
2ff7e61e | 685 | btrfs_wait_for_commit(root->fs_info, 0); |
d4edf39b MX |
686 | |
687 | return trans; | |
688 | } | |
689 | ||
d352ac68 | 690 | /* wait for a transaction commit to be fully complete */ |
2ff7e61e | 691 | static noinline void wait_for_commit(struct btrfs_transaction *commit) |
89ce8a63 | 692 | { |
4a9d8bde | 693 | wait_event(commit->commit_wait, commit->state == TRANS_STATE_COMPLETED); |
89ce8a63 CM |
694 | } |
695 | ||
2ff7e61e | 696 | int btrfs_wait_for_commit(struct btrfs_fs_info *fs_info, u64 transid) |
46204592 SW |
697 | { |
698 | struct btrfs_transaction *cur_trans = NULL, *t; | |
8cd2807f | 699 | int ret = 0; |
46204592 | 700 | |
46204592 | 701 | if (transid) { |
0b246afa | 702 | if (transid <= fs_info->last_trans_committed) |
a4abeea4 | 703 | goto out; |
46204592 SW |
704 | |
705 | /* find specified transaction */ | |
0b246afa JM |
706 | spin_lock(&fs_info->trans_lock); |
707 | list_for_each_entry(t, &fs_info->trans_list, list) { | |
46204592 SW |
708 | if (t->transid == transid) { |
709 | cur_trans = t; | |
9b64f57d | 710 | refcount_inc(&cur_trans->use_count); |
8cd2807f | 711 | ret = 0; |
46204592 SW |
712 | break; |
713 | } | |
8cd2807f MX |
714 | if (t->transid > transid) { |
715 | ret = 0; | |
46204592 | 716 | break; |
8cd2807f | 717 | } |
46204592 | 718 | } |
0b246afa | 719 | spin_unlock(&fs_info->trans_lock); |
42383020 SW |
720 | |
721 | /* | |
722 | * The specified transaction doesn't exist, or we | |
723 | * raced with btrfs_commit_transaction | |
724 | */ | |
725 | if (!cur_trans) { | |
0b246afa | 726 | if (transid > fs_info->last_trans_committed) |
42383020 | 727 | ret = -EINVAL; |
8cd2807f | 728 | goto out; |
42383020 | 729 | } |
46204592 SW |
730 | } else { |
731 | /* find newest transaction that is committing | committed */ | |
0b246afa JM |
732 | spin_lock(&fs_info->trans_lock); |
733 | list_for_each_entry_reverse(t, &fs_info->trans_list, | |
46204592 | 734 | list) { |
4a9d8bde MX |
735 | if (t->state >= TRANS_STATE_COMMIT_START) { |
736 | if (t->state == TRANS_STATE_COMPLETED) | |
3473f3c0 | 737 | break; |
46204592 | 738 | cur_trans = t; |
9b64f57d | 739 | refcount_inc(&cur_trans->use_count); |
46204592 SW |
740 | break; |
741 | } | |
742 | } | |
0b246afa | 743 | spin_unlock(&fs_info->trans_lock); |
46204592 | 744 | if (!cur_trans) |
a4abeea4 | 745 | goto out; /* nothing committing|committed */ |
46204592 SW |
746 | } |
747 | ||
2ff7e61e | 748 | wait_for_commit(cur_trans); |
724e2315 | 749 | btrfs_put_transaction(cur_trans); |
a4abeea4 | 750 | out: |
46204592 SW |
751 | return ret; |
752 | } | |
753 | ||
2ff7e61e | 754 | void btrfs_throttle(struct btrfs_fs_info *fs_info) |
37d1aeee | 755 | { |
92e2f7e3 | 756 | wait_current_trans(fs_info); |
37d1aeee CM |
757 | } |
758 | ||
2ff7e61e | 759 | static int should_end_transaction(struct btrfs_trans_handle *trans) |
8929ecfa | 760 | { |
2ff7e61e | 761 | struct btrfs_fs_info *fs_info = trans->fs_info; |
0b246afa | 762 | |
7c777430 | 763 | if (btrfs_check_space_for_delayed_refs(trans, fs_info)) |
1be41b78 | 764 | return 1; |
36ba022a | 765 | |
2ff7e61e | 766 | return !!btrfs_block_rsv_check(&fs_info->global_block_rsv, 5); |
8929ecfa YZ |
767 | } |
768 | ||
3a45bb20 | 769 | int btrfs_should_end_transaction(struct btrfs_trans_handle *trans) |
8929ecfa YZ |
770 | { |
771 | struct btrfs_transaction *cur_trans = trans->transaction; | |
772 | int updates; | |
49b25e05 | 773 | int err; |
8929ecfa | 774 | |
a4abeea4 | 775 | smp_mb(); |
4a9d8bde MX |
776 | if (cur_trans->state >= TRANS_STATE_BLOCKED || |
777 | cur_trans->delayed_refs.flushing) | |
8929ecfa YZ |
778 | return 1; |
779 | ||
780 | updates = trans->delayed_ref_updates; | |
781 | trans->delayed_ref_updates = 0; | |
49b25e05 | 782 | if (updates) { |
c79a70b1 | 783 | err = btrfs_run_delayed_refs(trans, updates * 2); |
49b25e05 JM |
784 | if (err) /* Error code will also eval true */ |
785 | return err; | |
786 | } | |
8929ecfa | 787 | |
2ff7e61e | 788 | return should_end_transaction(trans); |
8929ecfa YZ |
789 | } |
790 | ||
dc60c525 NB |
791 | static void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans) |
792 | ||
0e34693f | 793 | { |
dc60c525 NB |
794 | struct btrfs_fs_info *fs_info = trans->fs_info; |
795 | ||
0e34693f NB |
796 | if (!trans->block_rsv) { |
797 | ASSERT(!trans->bytes_reserved); | |
798 | return; | |
799 | } | |
800 | ||
801 | if (!trans->bytes_reserved) | |
802 | return; | |
803 | ||
804 | ASSERT(trans->block_rsv == &fs_info->trans_block_rsv); | |
805 | trace_btrfs_space_reservation(fs_info, "transaction", | |
806 | trans->transid, trans->bytes_reserved, 0); | |
807 | btrfs_block_rsv_release(fs_info, trans->block_rsv, | |
808 | trans->bytes_reserved); | |
809 | trans->bytes_reserved = 0; | |
810 | } | |
811 | ||
89ce8a63 | 812 | static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, |
3a45bb20 | 813 | int throttle) |
79154b1b | 814 | { |
3a45bb20 | 815 | struct btrfs_fs_info *info = trans->fs_info; |
8929ecfa | 816 | struct btrfs_transaction *cur_trans = trans->transaction; |
31b9655f | 817 | u64 transid = trans->transid; |
1be41b78 | 818 | unsigned long cur = trans->delayed_ref_updates; |
a698d075 | 819 | int lock = (trans->type != TRANS_JOIN_NOLOCK); |
4edc2ca3 | 820 | int err = 0; |
a79b7d4b | 821 | int must_run_delayed_refs = 0; |
c3e69d58 | 822 | |
b50fff81 DS |
823 | if (refcount_read(&trans->use_count) > 1) { |
824 | refcount_dec(&trans->use_count); | |
2a1eb461 JB |
825 | trans->block_rsv = trans->orig_rsv; |
826 | return 0; | |
827 | } | |
828 | ||
dc60c525 | 829 | btrfs_trans_release_metadata(trans); |
4c13d758 | 830 | trans->block_rsv = NULL; |
c5567237 | 831 | |
ea658bad | 832 | if (!list_empty(&trans->new_bgs)) |
6c686b35 | 833 | btrfs_create_pending_block_groups(trans); |
ea658bad | 834 | |
1be41b78 | 835 | trans->delayed_ref_updates = 0; |
a79b7d4b CM |
836 | if (!trans->sync) { |
837 | must_run_delayed_refs = | |
2ff7e61e | 838 | btrfs_should_throttle_delayed_refs(trans, info); |
0a2b2a84 | 839 | cur = max_t(unsigned long, cur, 32); |
a79b7d4b CM |
840 | |
841 | /* | |
842 | * don't make the caller wait if they are from a NOLOCK | |
843 | * or ATTACH transaction, it will deadlock with commit | |
844 | */ | |
845 | if (must_run_delayed_refs == 1 && | |
846 | (trans->type & (__TRANS_JOIN_NOLOCK | __TRANS_ATTACH))) | |
847 | must_run_delayed_refs = 2; | |
56bec294 | 848 | } |
bb721703 | 849 | |
dc60c525 | 850 | btrfs_trans_release_metadata(trans); |
0e721106 | 851 | trans->block_rsv = NULL; |
56bec294 | 852 | |
ea658bad | 853 | if (!list_empty(&trans->new_bgs)) |
6c686b35 | 854 | btrfs_create_pending_block_groups(trans); |
ea658bad | 855 | |
4fbcdf66 FM |
856 | btrfs_trans_release_chunk_metadata(trans); |
857 | ||
92e2f7e3 | 858 | if (lock && should_end_transaction(trans) && |
20c7bcec | 859 | READ_ONCE(cur_trans->state) == TRANS_STATE_RUNNING) { |
4a9d8bde MX |
860 | spin_lock(&info->trans_lock); |
861 | if (cur_trans->state == TRANS_STATE_RUNNING) | |
862 | cur_trans->state = TRANS_STATE_BLOCKED; | |
863 | spin_unlock(&info->trans_lock); | |
a4abeea4 | 864 | } |
8929ecfa | 865 | |
20c7bcec | 866 | if (lock && READ_ONCE(cur_trans->state) == TRANS_STATE_BLOCKED) { |
3bbb24b2 | 867 | if (throttle) |
3a45bb20 | 868 | return btrfs_commit_transaction(trans); |
3bbb24b2 | 869 | else |
8929ecfa YZ |
870 | wake_up_process(info->transaction_kthread); |
871 | } | |
872 | ||
0860adfd | 873 | if (trans->type & __TRANS_FREEZABLE) |
0b246afa | 874 | sb_end_intwrite(info->sb); |
6df7881a | 875 | |
8929ecfa | 876 | WARN_ON(cur_trans != info->running_transaction); |
13c5a93e JB |
877 | WARN_ON(atomic_read(&cur_trans->num_writers) < 1); |
878 | atomic_dec(&cur_trans->num_writers); | |
0860adfd | 879 | extwriter_counter_dec(cur_trans, trans->type); |
89ce8a63 | 880 | |
093258e6 | 881 | cond_wake_up(&cur_trans->writer_wait); |
724e2315 | 882 | btrfs_put_transaction(cur_trans); |
9ed74f2d JB |
883 | |
884 | if (current->journal_info == trans) | |
885 | current->journal_info = NULL; | |
ab78c84d | 886 | |
24bbcf04 | 887 | if (throttle) |
2ff7e61e | 888 | btrfs_run_delayed_iputs(info); |
24bbcf04 | 889 | |
49b25e05 | 890 | if (trans->aborted || |
0b246afa | 891 | test_bit(BTRFS_FS_STATE_ERROR, &info->fs_state)) { |
4e121c06 | 892 | wake_up_process(info->transaction_kthread); |
4edc2ca3 | 893 | err = -EIO; |
4e121c06 | 894 | } |
49b25e05 | 895 | |
4edc2ca3 | 896 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
a79b7d4b | 897 | if (must_run_delayed_refs) { |
2ff7e61e | 898 | btrfs_async_run_delayed_refs(info, cur, transid, |
a79b7d4b CM |
899 | must_run_delayed_refs == 1); |
900 | } | |
4edc2ca3 | 901 | return err; |
79154b1b CM |
902 | } |
903 | ||
3a45bb20 | 904 | int btrfs_end_transaction(struct btrfs_trans_handle *trans) |
89ce8a63 | 905 | { |
3a45bb20 | 906 | return __btrfs_end_transaction(trans, 0); |
89ce8a63 CM |
907 | } |
908 | ||
3a45bb20 | 909 | int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans) |
89ce8a63 | 910 | { |
3a45bb20 | 911 | return __btrfs_end_transaction(trans, 1); |
16cdcec7 MX |
912 | } |
913 | ||
d352ac68 CM |
914 | /* |
915 | * when btree blocks are allocated, they have some corresponding bits set for | |
916 | * them in one of two extent_io trees. This is used to make sure all of | |
690587d1 | 917 | * those extents are sent to disk but does not wait on them |
d352ac68 | 918 | */ |
2ff7e61e | 919 | int btrfs_write_marked_extents(struct btrfs_fs_info *fs_info, |
8cef4e16 | 920 | struct extent_io_tree *dirty_pages, int mark) |
79154b1b | 921 | { |
777e6bd7 | 922 | int err = 0; |
7c4452b9 | 923 | int werr = 0; |
0b246afa | 924 | struct address_space *mapping = fs_info->btree_inode->i_mapping; |
e6138876 | 925 | struct extent_state *cached_state = NULL; |
777e6bd7 | 926 | u64 start = 0; |
5f39d397 | 927 | u64 end; |
7c4452b9 | 928 | |
6300463b | 929 | atomic_inc(&BTRFS_I(fs_info->btree_inode)->sync_writers); |
1728366e | 930 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
e6138876 | 931 | mark, &cached_state)) { |
663dfbb0 FM |
932 | bool wait_writeback = false; |
933 | ||
934 | err = convert_extent_bit(dirty_pages, start, end, | |
935 | EXTENT_NEED_WAIT, | |
210aa277 | 936 | mark, &cached_state); |
663dfbb0 FM |
937 | /* |
938 | * convert_extent_bit can return -ENOMEM, which is most of the | |
939 | * time a temporary error. So when it happens, ignore the error | |
940 | * and wait for writeback of this range to finish - because we | |
941 | * failed to set the bit EXTENT_NEED_WAIT for the range, a call | |
bf89d38f JM |
942 | * to __btrfs_wait_marked_extents() would not know that |
943 | * writeback for this range started and therefore wouldn't | |
944 | * wait for it to finish - we don't want to commit a | |
945 | * superblock that points to btree nodes/leafs for which | |
946 | * writeback hasn't finished yet (and without errors). | |
663dfbb0 FM |
947 | * We cleanup any entries left in the io tree when committing |
948 | * the transaction (through clear_btree_io_tree()). | |
949 | */ | |
950 | if (err == -ENOMEM) { | |
951 | err = 0; | |
952 | wait_writeback = true; | |
953 | } | |
954 | if (!err) | |
955 | err = filemap_fdatawrite_range(mapping, start, end); | |
1728366e JB |
956 | if (err) |
957 | werr = err; | |
663dfbb0 FM |
958 | else if (wait_writeback) |
959 | werr = filemap_fdatawait_range(mapping, start, end); | |
e38e2ed7 | 960 | free_extent_state(cached_state); |
663dfbb0 | 961 | cached_state = NULL; |
1728366e JB |
962 | cond_resched(); |
963 | start = end + 1; | |
7c4452b9 | 964 | } |
6300463b | 965 | atomic_dec(&BTRFS_I(fs_info->btree_inode)->sync_writers); |
690587d1 CM |
966 | return werr; |
967 | } | |
968 | ||
969 | /* | |
970 | * when btree blocks are allocated, they have some corresponding bits set for | |
971 | * them in one of two extent_io trees. This is used to make sure all of | |
972 | * those extents are on disk for transaction or log commit. We wait | |
973 | * on all the pages and clear them from the dirty pages state tree | |
974 | */ | |
bf89d38f JM |
975 | static int __btrfs_wait_marked_extents(struct btrfs_fs_info *fs_info, |
976 | struct extent_io_tree *dirty_pages) | |
690587d1 | 977 | { |
690587d1 CM |
978 | int err = 0; |
979 | int werr = 0; | |
0b246afa | 980 | struct address_space *mapping = fs_info->btree_inode->i_mapping; |
e6138876 | 981 | struct extent_state *cached_state = NULL; |
690587d1 CM |
982 | u64 start = 0; |
983 | u64 end; | |
777e6bd7 | 984 | |
1728366e | 985 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
e6138876 | 986 | EXTENT_NEED_WAIT, &cached_state)) { |
663dfbb0 FM |
987 | /* |
988 | * Ignore -ENOMEM errors returned by clear_extent_bit(). | |
989 | * When committing the transaction, we'll remove any entries | |
990 | * left in the io tree. For a log commit, we don't remove them | |
991 | * after committing the log because the tree can be accessed | |
992 | * concurrently - we do it only at transaction commit time when | |
993 | * it's safe to do it (through clear_btree_io_tree()). | |
994 | */ | |
995 | err = clear_extent_bit(dirty_pages, start, end, | |
ae0f1625 | 996 | EXTENT_NEED_WAIT, 0, 0, &cached_state); |
663dfbb0 FM |
997 | if (err == -ENOMEM) |
998 | err = 0; | |
999 | if (!err) | |
1000 | err = filemap_fdatawait_range(mapping, start, end); | |
1728366e JB |
1001 | if (err) |
1002 | werr = err; | |
e38e2ed7 FM |
1003 | free_extent_state(cached_state); |
1004 | cached_state = NULL; | |
1728366e JB |
1005 | cond_resched(); |
1006 | start = end + 1; | |
777e6bd7 | 1007 | } |
7c4452b9 CM |
1008 | if (err) |
1009 | werr = err; | |
bf89d38f JM |
1010 | return werr; |
1011 | } | |
656f30db | 1012 | |
bf89d38f JM |
1013 | int btrfs_wait_extents(struct btrfs_fs_info *fs_info, |
1014 | struct extent_io_tree *dirty_pages) | |
1015 | { | |
1016 | bool errors = false; | |
1017 | int err; | |
656f30db | 1018 | |
bf89d38f JM |
1019 | err = __btrfs_wait_marked_extents(fs_info, dirty_pages); |
1020 | if (test_and_clear_bit(BTRFS_FS_BTREE_ERR, &fs_info->flags)) | |
1021 | errors = true; | |
1022 | ||
1023 | if (errors && !err) | |
1024 | err = -EIO; | |
1025 | return err; | |
1026 | } | |
656f30db | 1027 | |
bf89d38f JM |
1028 | int btrfs_wait_tree_log_extents(struct btrfs_root *log_root, int mark) |
1029 | { | |
1030 | struct btrfs_fs_info *fs_info = log_root->fs_info; | |
1031 | struct extent_io_tree *dirty_pages = &log_root->dirty_log_pages; | |
1032 | bool errors = false; | |
1033 | int err; | |
656f30db | 1034 | |
bf89d38f JM |
1035 | ASSERT(log_root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID); |
1036 | ||
1037 | err = __btrfs_wait_marked_extents(fs_info, dirty_pages); | |
1038 | if ((mark & EXTENT_DIRTY) && | |
1039 | test_and_clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags)) | |
1040 | errors = true; | |
1041 | ||
1042 | if ((mark & EXTENT_NEW) && | |
1043 | test_and_clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags)) | |
1044 | errors = true; | |
1045 | ||
1046 | if (errors && !err) | |
1047 | err = -EIO; | |
1048 | return err; | |
79154b1b CM |
1049 | } |
1050 | ||
690587d1 | 1051 | /* |
c9b577c0 NB |
1052 | * When btree blocks are allocated the corresponding extents are marked dirty. |
1053 | * This function ensures such extents are persisted on disk for transaction or | |
1054 | * log commit. | |
1055 | * | |
1056 | * @trans: transaction whose dirty pages we'd like to write | |
690587d1 | 1057 | */ |
70458a58 | 1058 | static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans) |
690587d1 CM |
1059 | { |
1060 | int ret; | |
1061 | int ret2; | |
c9b577c0 | 1062 | struct extent_io_tree *dirty_pages = &trans->transaction->dirty_pages; |
70458a58 | 1063 | struct btrfs_fs_info *fs_info = trans->fs_info; |
c6adc9cc | 1064 | struct blk_plug plug; |
690587d1 | 1065 | |
c6adc9cc | 1066 | blk_start_plug(&plug); |
c9b577c0 | 1067 | ret = btrfs_write_marked_extents(fs_info, dirty_pages, EXTENT_DIRTY); |
c6adc9cc | 1068 | blk_finish_plug(&plug); |
bf89d38f | 1069 | ret2 = btrfs_wait_extents(fs_info, dirty_pages); |
bf0da8c1 | 1070 | |
c9b577c0 NB |
1071 | clear_btree_io_tree(&trans->transaction->dirty_pages); |
1072 | ||
bf0da8c1 CM |
1073 | if (ret) |
1074 | return ret; | |
c9b577c0 | 1075 | else if (ret2) |
bf0da8c1 | 1076 | return ret2; |
c9b577c0 NB |
1077 | else |
1078 | return 0; | |
d0c803c4 CM |
1079 | } |
1080 | ||
d352ac68 CM |
1081 | /* |
1082 | * this is used to update the root pointer in the tree of tree roots. | |
1083 | * | |
1084 | * But, in the case of the extent allocation tree, updating the root | |
1085 | * pointer may allocate blocks which may change the root of the extent | |
1086 | * allocation tree. | |
1087 | * | |
1088 | * So, this loops and repeats and makes sure the cowonly root didn't | |
1089 | * change while the root pointer was being updated in the metadata. | |
1090 | */ | |
0b86a832 CM |
1091 | static int update_cowonly_root(struct btrfs_trans_handle *trans, |
1092 | struct btrfs_root *root) | |
79154b1b CM |
1093 | { |
1094 | int ret; | |
0b86a832 | 1095 | u64 old_root_bytenr; |
86b9f2ec | 1096 | u64 old_root_used; |
0b246afa JM |
1097 | struct btrfs_fs_info *fs_info = root->fs_info; |
1098 | struct btrfs_root *tree_root = fs_info->tree_root; | |
79154b1b | 1099 | |
86b9f2ec | 1100 | old_root_used = btrfs_root_used(&root->root_item); |
56bec294 | 1101 | |
d397712b | 1102 | while (1) { |
0b86a832 | 1103 | old_root_bytenr = btrfs_root_bytenr(&root->root_item); |
86b9f2ec | 1104 | if (old_root_bytenr == root->node->start && |
ea526d18 | 1105 | old_root_used == btrfs_root_used(&root->root_item)) |
79154b1b | 1106 | break; |
87ef2bb4 | 1107 | |
5d4f98a2 | 1108 | btrfs_set_root_node(&root->root_item, root->node); |
79154b1b | 1109 | ret = btrfs_update_root(trans, tree_root, |
0b86a832 CM |
1110 | &root->root_key, |
1111 | &root->root_item); | |
49b25e05 JM |
1112 | if (ret) |
1113 | return ret; | |
56bec294 | 1114 | |
86b9f2ec | 1115 | old_root_used = btrfs_root_used(&root->root_item); |
0b86a832 | 1116 | } |
276e680d | 1117 | |
0b86a832 CM |
1118 | return 0; |
1119 | } | |
1120 | ||
d352ac68 CM |
1121 | /* |
1122 | * update all the cowonly tree roots on disk | |
49b25e05 JM |
1123 | * |
1124 | * The error handling in this function may not be obvious. Any of the | |
1125 | * failures will cause the file system to go offline. We still need | |
1126 | * to clean up the delayed refs. | |
d352ac68 | 1127 | */ |
9386d8bc | 1128 | static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans) |
0b86a832 | 1129 | { |
9386d8bc | 1130 | struct btrfs_fs_info *fs_info = trans->fs_info; |
ea526d18 | 1131 | struct list_head *dirty_bgs = &trans->transaction->dirty_bgs; |
1bbc621e | 1132 | struct list_head *io_bgs = &trans->transaction->io_bgs; |
0b86a832 | 1133 | struct list_head *next; |
84234f3a | 1134 | struct extent_buffer *eb; |
56bec294 | 1135 | int ret; |
84234f3a YZ |
1136 | |
1137 | eb = btrfs_lock_root_node(fs_info->tree_root); | |
49b25e05 JM |
1138 | ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, |
1139 | 0, &eb); | |
84234f3a YZ |
1140 | btrfs_tree_unlock(eb); |
1141 | free_extent_buffer(eb); | |
0b86a832 | 1142 | |
49b25e05 JM |
1143 | if (ret) |
1144 | return ret; | |
1145 | ||
c79a70b1 | 1146 | ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); |
49b25e05 JM |
1147 | if (ret) |
1148 | return ret; | |
87ef2bb4 | 1149 | |
0b246afa | 1150 | ret = btrfs_run_dev_stats(trans, fs_info); |
c16ce190 JB |
1151 | if (ret) |
1152 | return ret; | |
0b246afa | 1153 | ret = btrfs_run_dev_replace(trans, fs_info); |
c16ce190 JB |
1154 | if (ret) |
1155 | return ret; | |
280f8bd2 | 1156 | ret = btrfs_run_qgroups(trans); |
c16ce190 JB |
1157 | if (ret) |
1158 | return ret; | |
546adb0d | 1159 | |
2ff7e61e | 1160 | ret = btrfs_setup_space_cache(trans, fs_info); |
dcdf7f6d JB |
1161 | if (ret) |
1162 | return ret; | |
1163 | ||
546adb0d | 1164 | /* run_qgroups might have added some more refs */ |
c79a70b1 | 1165 | ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); |
c16ce190 JB |
1166 | if (ret) |
1167 | return ret; | |
ea526d18 | 1168 | again: |
d397712b | 1169 | while (!list_empty(&fs_info->dirty_cowonly_roots)) { |
2ff7e61e | 1170 | struct btrfs_root *root; |
0b86a832 CM |
1171 | next = fs_info->dirty_cowonly_roots.next; |
1172 | list_del_init(next); | |
1173 | root = list_entry(next, struct btrfs_root, dirty_list); | |
e7070be1 | 1174 | clear_bit(BTRFS_ROOT_DIRTY, &root->state); |
87ef2bb4 | 1175 | |
9e351cc8 JB |
1176 | if (root != fs_info->extent_root) |
1177 | list_add_tail(&root->dirty_list, | |
1178 | &trans->transaction->switch_commits); | |
49b25e05 JM |
1179 | ret = update_cowonly_root(trans, root); |
1180 | if (ret) | |
1181 | return ret; | |
c79a70b1 | 1182 | ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); |
ea526d18 JB |
1183 | if (ret) |
1184 | return ret; | |
79154b1b | 1185 | } |
276e680d | 1186 | |
1bbc621e | 1187 | while (!list_empty(dirty_bgs) || !list_empty(io_bgs)) { |
2ff7e61e | 1188 | ret = btrfs_write_dirty_block_groups(trans, fs_info); |
ea526d18 JB |
1189 | if (ret) |
1190 | return ret; | |
c79a70b1 | 1191 | ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); |
ea526d18 JB |
1192 | if (ret) |
1193 | return ret; | |
1194 | } | |
1195 | ||
1196 | if (!list_empty(&fs_info->dirty_cowonly_roots)) | |
1197 | goto again; | |
1198 | ||
9e351cc8 JB |
1199 | list_add_tail(&fs_info->extent_root->dirty_list, |
1200 | &trans->transaction->switch_commits); | |
9f6cbcbb DS |
1201 | |
1202 | /* Update dev-replace pointer once everything is committed */ | |
1203 | fs_info->dev_replace.committed_cursor_left = | |
1204 | fs_info->dev_replace.cursor_left_last_write_of_item; | |
8dabb742 | 1205 | |
79154b1b CM |
1206 | return 0; |
1207 | } | |
1208 | ||
d352ac68 CM |
1209 | /* |
1210 | * dead roots are old snapshots that need to be deleted. This allocates | |
1211 | * a dirty root struct and adds it into the list of dead roots that need to | |
1212 | * be deleted | |
1213 | */ | |
cfad392b | 1214 | void btrfs_add_dead_root(struct btrfs_root *root) |
5eda7b5e | 1215 | { |
0b246afa JM |
1216 | struct btrfs_fs_info *fs_info = root->fs_info; |
1217 | ||
1218 | spin_lock(&fs_info->trans_lock); | |
cfad392b | 1219 | if (list_empty(&root->root_list)) |
0b246afa JM |
1220 | list_add_tail(&root->root_list, &fs_info->dead_roots); |
1221 | spin_unlock(&fs_info->trans_lock); | |
5eda7b5e CM |
1222 | } |
1223 | ||
d352ac68 | 1224 | /* |
5d4f98a2 | 1225 | * update all the cowonly tree roots on disk |
d352ac68 | 1226 | */ |
7e4443d9 | 1227 | static noinline int commit_fs_roots(struct btrfs_trans_handle *trans) |
0f7d52f4 | 1228 | { |
7e4443d9 | 1229 | struct btrfs_fs_info *fs_info = trans->fs_info; |
0f7d52f4 | 1230 | struct btrfs_root *gang[8]; |
0f7d52f4 CM |
1231 | int i; |
1232 | int ret; | |
54aa1f4d CM |
1233 | int err = 0; |
1234 | ||
a4abeea4 | 1235 | spin_lock(&fs_info->fs_roots_radix_lock); |
d397712b | 1236 | while (1) { |
5d4f98a2 YZ |
1237 | ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, |
1238 | (void **)gang, 0, | |
0f7d52f4 CM |
1239 | ARRAY_SIZE(gang), |
1240 | BTRFS_ROOT_TRANS_TAG); | |
1241 | if (ret == 0) | |
1242 | break; | |
1243 | for (i = 0; i < ret; i++) { | |
5b4aacef | 1244 | struct btrfs_root *root = gang[i]; |
5d4f98a2 YZ |
1245 | radix_tree_tag_clear(&fs_info->fs_roots_radix, |
1246 | (unsigned long)root->root_key.objectid, | |
1247 | BTRFS_ROOT_TRANS_TAG); | |
a4abeea4 | 1248 | spin_unlock(&fs_info->fs_roots_radix_lock); |
31153d81 | 1249 | |
e02119d5 | 1250 | btrfs_free_log(trans, root); |
5d4f98a2 | 1251 | btrfs_update_reloc_root(trans, root); |
bcc63abb | 1252 | |
82d5902d LZ |
1253 | btrfs_save_ino_cache(root, trans); |
1254 | ||
f1ebcc74 | 1255 | /* see comments in should_cow_block() */ |
27cdeb70 | 1256 | clear_bit(BTRFS_ROOT_FORCE_COW, &root->state); |
c7548af6 | 1257 | smp_mb__after_atomic(); |
f1ebcc74 | 1258 | |
978d910d | 1259 | if (root->commit_root != root->node) { |
9e351cc8 JB |
1260 | list_add_tail(&root->dirty_list, |
1261 | &trans->transaction->switch_commits); | |
978d910d YZ |
1262 | btrfs_set_root_node(&root->root_item, |
1263 | root->node); | |
1264 | } | |
5d4f98a2 | 1265 | |
5d4f98a2 | 1266 | err = btrfs_update_root(trans, fs_info->tree_root, |
0f7d52f4 CM |
1267 | &root->root_key, |
1268 | &root->root_item); | |
a4abeea4 | 1269 | spin_lock(&fs_info->fs_roots_radix_lock); |
54aa1f4d CM |
1270 | if (err) |
1271 | break; | |
733e03a0 | 1272 | btrfs_qgroup_free_meta_all_pertrans(root); |
0f7d52f4 CM |
1273 | } |
1274 | } | |
a4abeea4 | 1275 | spin_unlock(&fs_info->fs_roots_radix_lock); |
54aa1f4d | 1276 | return err; |
0f7d52f4 CM |
1277 | } |
1278 | ||
d352ac68 | 1279 | /* |
de78b51a ES |
1280 | * defrag a given btree. |
1281 | * Every leaf in the btree is read and defragged. | |
d352ac68 | 1282 | */ |
de78b51a | 1283 | int btrfs_defrag_root(struct btrfs_root *root) |
e9d0b13b CM |
1284 | { |
1285 | struct btrfs_fs_info *info = root->fs_info; | |
e9d0b13b | 1286 | struct btrfs_trans_handle *trans; |
8929ecfa | 1287 | int ret; |
e9d0b13b | 1288 | |
27cdeb70 | 1289 | if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state)) |
e9d0b13b | 1290 | return 0; |
8929ecfa | 1291 | |
6b80053d | 1292 | while (1) { |
8929ecfa YZ |
1293 | trans = btrfs_start_transaction(root, 0); |
1294 | if (IS_ERR(trans)) | |
1295 | return PTR_ERR(trans); | |
1296 | ||
de78b51a | 1297 | ret = btrfs_defrag_leaves(trans, root); |
8929ecfa | 1298 | |
3a45bb20 | 1299 | btrfs_end_transaction(trans); |
2ff7e61e | 1300 | btrfs_btree_balance_dirty(info); |
e9d0b13b CM |
1301 | cond_resched(); |
1302 | ||
ab8d0fc4 | 1303 | if (btrfs_fs_closing(info) || ret != -EAGAIN) |
e9d0b13b | 1304 | break; |
210549eb | 1305 | |
ab8d0fc4 JM |
1306 | if (btrfs_defrag_cancelled(info)) { |
1307 | btrfs_debug(info, "defrag_root cancelled"); | |
210549eb DS |
1308 | ret = -EAGAIN; |
1309 | break; | |
1310 | } | |
e9d0b13b | 1311 | } |
27cdeb70 | 1312 | clear_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state); |
8929ecfa | 1313 | return ret; |
e9d0b13b CM |
1314 | } |
1315 | ||
6426c7ad QW |
1316 | /* |
1317 | * Do all special snapshot related qgroup dirty hack. | |
1318 | * | |
1319 | * Will do all needed qgroup inherit and dirty hack like switch commit | |
1320 | * roots inside one transaction and write all btree into disk, to make | |
1321 | * qgroup works. | |
1322 | */ | |
1323 | static int qgroup_account_snapshot(struct btrfs_trans_handle *trans, | |
1324 | struct btrfs_root *src, | |
1325 | struct btrfs_root *parent, | |
1326 | struct btrfs_qgroup_inherit *inherit, | |
1327 | u64 dst_objectid) | |
1328 | { | |
1329 | struct btrfs_fs_info *fs_info = src->fs_info; | |
1330 | int ret; | |
1331 | ||
1332 | /* | |
1333 | * Save some performance in the case that qgroups are not | |
1334 | * enabled. If this check races with the ioctl, rescan will | |
1335 | * kick in anyway. | |
1336 | */ | |
9ea6e2b5 | 1337 | if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) |
6426c7ad | 1338 | return 0; |
6426c7ad | 1339 | |
4d31778a QW |
1340 | /* |
1341 | * Ensure dirty @src will be commited. Or, after comming | |
1342 | * commit_fs_roots() and switch_commit_roots(), any dirty but not | |
1343 | * recorded root will never be updated again, causing an outdated root | |
1344 | * item. | |
1345 | */ | |
1346 | record_root_in_trans(trans, src, 1); | |
1347 | ||
6426c7ad QW |
1348 | /* |
1349 | * We are going to commit transaction, see btrfs_commit_transaction() | |
1350 | * comment for reason locking tree_log_mutex | |
1351 | */ | |
1352 | mutex_lock(&fs_info->tree_log_mutex); | |
1353 | ||
7e4443d9 | 1354 | ret = commit_fs_roots(trans); |
6426c7ad QW |
1355 | if (ret) |
1356 | goto out; | |
460fb20a | 1357 | ret = btrfs_qgroup_account_extents(trans); |
6426c7ad QW |
1358 | if (ret < 0) |
1359 | goto out; | |
1360 | ||
1361 | /* Now qgroup are all updated, we can inherit it to new qgroups */ | |
a9377422 | 1362 | ret = btrfs_qgroup_inherit(trans, src->root_key.objectid, dst_objectid, |
6426c7ad QW |
1363 | inherit); |
1364 | if (ret < 0) | |
1365 | goto out; | |
1366 | ||
1367 | /* | |
1368 | * Now we do a simplified commit transaction, which will: | |
1369 | * 1) commit all subvolume and extent tree | |
1370 | * To ensure all subvolume and extent tree have a valid | |
1371 | * commit_root to accounting later insert_dir_item() | |
1372 | * 2) write all btree blocks onto disk | |
1373 | * This is to make sure later btree modification will be cowed | |
1374 | * Or commit_root can be populated and cause wrong qgroup numbers | |
1375 | * In this simplified commit, we don't really care about other trees | |
1376 | * like chunk and root tree, as they won't affect qgroup. | |
1377 | * And we don't write super to avoid half committed status. | |
1378 | */ | |
9386d8bc | 1379 | ret = commit_cowonly_roots(trans); |
6426c7ad QW |
1380 | if (ret) |
1381 | goto out; | |
16916a88 | 1382 | switch_commit_roots(trans->transaction); |
70458a58 | 1383 | ret = btrfs_write_and_wait_transaction(trans); |
6426c7ad | 1384 | if (ret) |
f7af3934 | 1385 | btrfs_handle_fs_error(fs_info, ret, |
6426c7ad QW |
1386 | "Error while writing out transaction for qgroup"); |
1387 | ||
1388 | out: | |
1389 | mutex_unlock(&fs_info->tree_log_mutex); | |
1390 | ||
1391 | /* | |
1392 | * Force parent root to be updated, as we recorded it before so its | |
1393 | * last_trans == cur_transid. | |
1394 | * Or it won't be committed again onto disk after later | |
1395 | * insert_dir_item() | |
1396 | */ | |
1397 | if (!ret) | |
1398 | record_root_in_trans(trans, parent, 1); | |
1399 | return ret; | |
1400 | } | |
1401 | ||
d352ac68 CM |
1402 | /* |
1403 | * new snapshots need to be created at a very specific time in the | |
aec8030a MX |
1404 | * transaction commit. This does the actual creation. |
1405 | * | |
1406 | * Note: | |
1407 | * If the error which may affect the commitment of the current transaction | |
1408 | * happens, we should return the error number. If the error which just affect | |
1409 | * the creation of the pending snapshots, just return 0. | |
d352ac68 | 1410 | */ |
80b6794d | 1411 | static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, |
3063d29f CM |
1412 | struct btrfs_pending_snapshot *pending) |
1413 | { | |
08d50ca3 NB |
1414 | |
1415 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
3063d29f | 1416 | struct btrfs_key key; |
80b6794d | 1417 | struct btrfs_root_item *new_root_item; |
3063d29f CM |
1418 | struct btrfs_root *tree_root = fs_info->tree_root; |
1419 | struct btrfs_root *root = pending->root; | |
6bdb72de | 1420 | struct btrfs_root *parent_root; |
98c9942a | 1421 | struct btrfs_block_rsv *rsv; |
6bdb72de | 1422 | struct inode *parent_inode; |
42874b3d MX |
1423 | struct btrfs_path *path; |
1424 | struct btrfs_dir_item *dir_item; | |
a22285a6 | 1425 | struct dentry *dentry; |
3063d29f | 1426 | struct extent_buffer *tmp; |
925baedd | 1427 | struct extent_buffer *old; |
95582b00 | 1428 | struct timespec64 cur_time; |
aec8030a | 1429 | int ret = 0; |
d68fc57b | 1430 | u64 to_reserve = 0; |
6bdb72de | 1431 | u64 index = 0; |
a22285a6 | 1432 | u64 objectid; |
b83cc969 | 1433 | u64 root_flags; |
8ea05e3a | 1434 | uuid_le new_uuid; |
3063d29f | 1435 | |
8546b570 DS |
1436 | ASSERT(pending->path); |
1437 | path = pending->path; | |
42874b3d | 1438 | |
b0c0ea63 DS |
1439 | ASSERT(pending->root_item); |
1440 | new_root_item = pending->root_item; | |
a22285a6 | 1441 | |
aec8030a MX |
1442 | pending->error = btrfs_find_free_objectid(tree_root, &objectid); |
1443 | if (pending->error) | |
6fa9700e | 1444 | goto no_free_objectid; |
3063d29f | 1445 | |
d6726335 QW |
1446 | /* |
1447 | * Make qgroup to skip current new snapshot's qgroupid, as it is | |
1448 | * accounted by later btrfs_qgroup_inherit(). | |
1449 | */ | |
1450 | btrfs_set_skip_qgroup(trans, objectid); | |
1451 | ||
147d256e | 1452 | btrfs_reloc_pre_snapshot(pending, &to_reserve); |
d68fc57b YZ |
1453 | |
1454 | if (to_reserve > 0) { | |
aec8030a MX |
1455 | pending->error = btrfs_block_rsv_add(root, |
1456 | &pending->block_rsv, | |
1457 | to_reserve, | |
1458 | BTRFS_RESERVE_NO_FLUSH); | |
1459 | if (pending->error) | |
d6726335 | 1460 | goto clear_skip_qgroup; |
d68fc57b YZ |
1461 | } |
1462 | ||
3063d29f | 1463 | key.objectid = objectid; |
a22285a6 YZ |
1464 | key.offset = (u64)-1; |
1465 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3063d29f | 1466 | |
6fa9700e | 1467 | rsv = trans->block_rsv; |
a22285a6 | 1468 | trans->block_rsv = &pending->block_rsv; |
2382c5cc | 1469 | trans->bytes_reserved = trans->block_rsv->reserved; |
0b246afa | 1470 | trace_btrfs_space_reservation(fs_info, "transaction", |
88d3a5aa JB |
1471 | trans->transid, |
1472 | trans->bytes_reserved, 1); | |
a22285a6 | 1473 | dentry = pending->dentry; |
e9662f70 | 1474 | parent_inode = pending->dir; |
a22285a6 | 1475 | parent_root = BTRFS_I(parent_inode)->root; |
6426c7ad | 1476 | record_root_in_trans(trans, parent_root, 0); |
a22285a6 | 1477 | |
c2050a45 | 1478 | cur_time = current_time(parent_inode); |
04b285f3 | 1479 | |
3063d29f CM |
1480 | /* |
1481 | * insert the directory item | |
1482 | */ | |
877574e2 | 1483 | ret = btrfs_set_inode_index(BTRFS_I(parent_inode), &index); |
49b25e05 | 1484 | BUG_ON(ret); /* -ENOMEM */ |
42874b3d MX |
1485 | |
1486 | /* check if there is a file/dir which has the same name. */ | |
1487 | dir_item = btrfs_lookup_dir_item(NULL, parent_root, path, | |
4a0cc7ca | 1488 | btrfs_ino(BTRFS_I(parent_inode)), |
42874b3d MX |
1489 | dentry->d_name.name, |
1490 | dentry->d_name.len, 0); | |
1491 | if (dir_item != NULL && !IS_ERR(dir_item)) { | |
fe66a05a | 1492 | pending->error = -EEXIST; |
aec8030a | 1493 | goto dir_item_existed; |
42874b3d MX |
1494 | } else if (IS_ERR(dir_item)) { |
1495 | ret = PTR_ERR(dir_item); | |
66642832 | 1496 | btrfs_abort_transaction(trans, ret); |
8732d44f | 1497 | goto fail; |
79787eaa | 1498 | } |
42874b3d | 1499 | btrfs_release_path(path); |
52c26179 | 1500 | |
e999376f CM |
1501 | /* |
1502 | * pull in the delayed directory update | |
1503 | * and the delayed inode item | |
1504 | * otherwise we corrupt the FS during | |
1505 | * snapshot | |
1506 | */ | |
e5c304e6 | 1507 | ret = btrfs_run_delayed_items(trans); |
8732d44f | 1508 | if (ret) { /* Transaction aborted */ |
66642832 | 1509 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1510 | goto fail; |
1511 | } | |
e999376f | 1512 | |
6426c7ad | 1513 | record_root_in_trans(trans, root, 0); |
6bdb72de SW |
1514 | btrfs_set_root_last_snapshot(&root->root_item, trans->transid); |
1515 | memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); | |
08fe4db1 | 1516 | btrfs_check_and_init_root_item(new_root_item); |
6bdb72de | 1517 | |
b83cc969 LZ |
1518 | root_flags = btrfs_root_flags(new_root_item); |
1519 | if (pending->readonly) | |
1520 | root_flags |= BTRFS_ROOT_SUBVOL_RDONLY; | |
1521 | else | |
1522 | root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY; | |
1523 | btrfs_set_root_flags(new_root_item, root_flags); | |
1524 | ||
8ea05e3a AB |
1525 | btrfs_set_root_generation_v2(new_root_item, |
1526 | trans->transid); | |
1527 | uuid_le_gen(&new_uuid); | |
1528 | memcpy(new_root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE); | |
1529 | memcpy(new_root_item->parent_uuid, root->root_item.uuid, | |
1530 | BTRFS_UUID_SIZE); | |
70023da2 SB |
1531 | if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) { |
1532 | memset(new_root_item->received_uuid, 0, | |
1533 | sizeof(new_root_item->received_uuid)); | |
1534 | memset(&new_root_item->stime, 0, sizeof(new_root_item->stime)); | |
1535 | memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime)); | |
1536 | btrfs_set_root_stransid(new_root_item, 0); | |
1537 | btrfs_set_root_rtransid(new_root_item, 0); | |
1538 | } | |
3cae210f QW |
1539 | btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec); |
1540 | btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec); | |
8ea05e3a | 1541 | btrfs_set_root_otransid(new_root_item, trans->transid); |
8ea05e3a | 1542 | |
6bdb72de | 1543 | old = btrfs_lock_root_node(root); |
49b25e05 | 1544 | ret = btrfs_cow_block(trans, root, old, NULL, 0, &old); |
79787eaa JM |
1545 | if (ret) { |
1546 | btrfs_tree_unlock(old); | |
1547 | free_extent_buffer(old); | |
66642832 | 1548 | btrfs_abort_transaction(trans, ret); |
8732d44f | 1549 | goto fail; |
79787eaa | 1550 | } |
49b25e05 | 1551 | |
6bdb72de SW |
1552 | btrfs_set_lock_blocking(old); |
1553 | ||
49b25e05 | 1554 | ret = btrfs_copy_root(trans, root, old, &tmp, objectid); |
79787eaa | 1555 | /* clean up in any case */ |
6bdb72de SW |
1556 | btrfs_tree_unlock(old); |
1557 | free_extent_buffer(old); | |
8732d44f | 1558 | if (ret) { |
66642832 | 1559 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1560 | goto fail; |
1561 | } | |
f1ebcc74 | 1562 | /* see comments in should_cow_block() */ |
27cdeb70 | 1563 | set_bit(BTRFS_ROOT_FORCE_COW, &root->state); |
f1ebcc74 LB |
1564 | smp_wmb(); |
1565 | ||
6bdb72de | 1566 | btrfs_set_root_node(new_root_item, tmp); |
a22285a6 YZ |
1567 | /* record when the snapshot was created in key.offset */ |
1568 | key.offset = trans->transid; | |
1569 | ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); | |
6bdb72de SW |
1570 | btrfs_tree_unlock(tmp); |
1571 | free_extent_buffer(tmp); | |
8732d44f | 1572 | if (ret) { |
66642832 | 1573 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1574 | goto fail; |
1575 | } | |
6bdb72de | 1576 | |
a22285a6 YZ |
1577 | /* |
1578 | * insert root back/forward references | |
1579 | */ | |
6025c19f | 1580 | ret = btrfs_add_root_ref(trans, objectid, |
0660b5af | 1581 | parent_root->root_key.objectid, |
4a0cc7ca | 1582 | btrfs_ino(BTRFS_I(parent_inode)), index, |
a22285a6 | 1583 | dentry->d_name.name, dentry->d_name.len); |
8732d44f | 1584 | if (ret) { |
66642832 | 1585 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1586 | goto fail; |
1587 | } | |
0660b5af | 1588 | |
a22285a6 | 1589 | key.offset = (u64)-1; |
0b246afa | 1590 | pending->snap = btrfs_read_fs_root_no_name(fs_info, &key); |
79787eaa JM |
1591 | if (IS_ERR(pending->snap)) { |
1592 | ret = PTR_ERR(pending->snap); | |
66642832 | 1593 | btrfs_abort_transaction(trans, ret); |
8732d44f | 1594 | goto fail; |
79787eaa | 1595 | } |
d68fc57b | 1596 | |
49b25e05 | 1597 | ret = btrfs_reloc_post_snapshot(trans, pending); |
8732d44f | 1598 | if (ret) { |
66642832 | 1599 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1600 | goto fail; |
1601 | } | |
361048f5 | 1602 | |
c79a70b1 | 1603 | ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); |
8732d44f | 1604 | if (ret) { |
66642832 | 1605 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1606 | goto fail; |
1607 | } | |
42874b3d | 1608 | |
6426c7ad QW |
1609 | /* |
1610 | * Do special qgroup accounting for snapshot, as we do some qgroup | |
1611 | * snapshot hack to do fast snapshot. | |
1612 | * To co-operate with that hack, we do hack again. | |
1613 | * Or snapshot will be greatly slowed down by a subtree qgroup rescan | |
1614 | */ | |
1615 | ret = qgroup_account_snapshot(trans, root, parent_root, | |
1616 | pending->inherit, objectid); | |
1617 | if (ret < 0) | |
1618 | goto fail; | |
1619 | ||
684572df LF |
1620 | ret = btrfs_insert_dir_item(trans, dentry->d_name.name, |
1621 | dentry->d_name.len, BTRFS_I(parent_inode), | |
1622 | &key, BTRFS_FT_DIR, index); | |
42874b3d | 1623 | /* We have check then name at the beginning, so it is impossible. */ |
9c52057c | 1624 | BUG_ON(ret == -EEXIST || ret == -EOVERFLOW); |
8732d44f | 1625 | if (ret) { |
66642832 | 1626 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1627 | goto fail; |
1628 | } | |
42874b3d | 1629 | |
6ef06d27 | 1630 | btrfs_i_size_write(BTRFS_I(parent_inode), parent_inode->i_size + |
42874b3d | 1631 | dentry->d_name.len * 2); |
04b285f3 | 1632 | parent_inode->i_mtime = parent_inode->i_ctime = |
c2050a45 | 1633 | current_time(parent_inode); |
be6aef60 | 1634 | ret = btrfs_update_inode_fallback(trans, parent_root, parent_inode); |
dd5f9615 | 1635 | if (ret) { |
66642832 | 1636 | btrfs_abort_transaction(trans, ret); |
dd5f9615 SB |
1637 | goto fail; |
1638 | } | |
cdb345a8 LF |
1639 | ret = btrfs_uuid_tree_add(trans, new_uuid.b, BTRFS_UUID_KEY_SUBVOL, |
1640 | objectid); | |
dd5f9615 | 1641 | if (ret) { |
66642832 | 1642 | btrfs_abort_transaction(trans, ret); |
dd5f9615 SB |
1643 | goto fail; |
1644 | } | |
1645 | if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) { | |
cdb345a8 | 1646 | ret = btrfs_uuid_tree_add(trans, new_root_item->received_uuid, |
dd5f9615 SB |
1647 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, |
1648 | objectid); | |
1649 | if (ret && ret != -EEXIST) { | |
66642832 | 1650 | btrfs_abort_transaction(trans, ret); |
dd5f9615 SB |
1651 | goto fail; |
1652 | } | |
1653 | } | |
d6726335 | 1654 | |
c79a70b1 | 1655 | ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); |
d6726335 | 1656 | if (ret) { |
66642832 | 1657 | btrfs_abort_transaction(trans, ret); |
d6726335 QW |
1658 | goto fail; |
1659 | } | |
1660 | ||
3063d29f | 1661 | fail: |
aec8030a MX |
1662 | pending->error = ret; |
1663 | dir_item_existed: | |
98c9942a | 1664 | trans->block_rsv = rsv; |
2382c5cc | 1665 | trans->bytes_reserved = 0; |
d6726335 QW |
1666 | clear_skip_qgroup: |
1667 | btrfs_clear_skip_qgroup(trans); | |
6fa9700e MX |
1668 | no_free_objectid: |
1669 | kfree(new_root_item); | |
b0c0ea63 | 1670 | pending->root_item = NULL; |
42874b3d | 1671 | btrfs_free_path(path); |
8546b570 DS |
1672 | pending->path = NULL; |
1673 | ||
49b25e05 | 1674 | return ret; |
3063d29f CM |
1675 | } |
1676 | ||
d352ac68 CM |
1677 | /* |
1678 | * create all the snapshots we've scheduled for creation | |
1679 | */ | |
08d50ca3 | 1680 | static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans) |
3de4586c | 1681 | { |
aec8030a | 1682 | struct btrfs_pending_snapshot *pending, *next; |
3de4586c | 1683 | struct list_head *head = &trans->transaction->pending_snapshots; |
aec8030a | 1684 | int ret = 0; |
3de4586c | 1685 | |
aec8030a MX |
1686 | list_for_each_entry_safe(pending, next, head, list) { |
1687 | list_del(&pending->list); | |
08d50ca3 | 1688 | ret = create_pending_snapshot(trans, pending); |
aec8030a MX |
1689 | if (ret) |
1690 | break; | |
1691 | } | |
1692 | return ret; | |
3de4586c CM |
1693 | } |
1694 | ||
2ff7e61e | 1695 | static void update_super_roots(struct btrfs_fs_info *fs_info) |
5d4f98a2 YZ |
1696 | { |
1697 | struct btrfs_root_item *root_item; | |
1698 | struct btrfs_super_block *super; | |
1699 | ||
0b246afa | 1700 | super = fs_info->super_copy; |
5d4f98a2 | 1701 | |
0b246afa | 1702 | root_item = &fs_info->chunk_root->root_item; |
093e037c DS |
1703 | super->chunk_root = root_item->bytenr; |
1704 | super->chunk_root_generation = root_item->generation; | |
1705 | super->chunk_root_level = root_item->level; | |
5d4f98a2 | 1706 | |
0b246afa | 1707 | root_item = &fs_info->tree_root->root_item; |
093e037c DS |
1708 | super->root = root_item->bytenr; |
1709 | super->generation = root_item->generation; | |
1710 | super->root_level = root_item->level; | |
0b246afa | 1711 | if (btrfs_test_opt(fs_info, SPACE_CACHE)) |
093e037c | 1712 | super->cache_generation = root_item->generation; |
0b246afa | 1713 | if (test_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags)) |
093e037c | 1714 | super->uuid_tree_generation = root_item->generation; |
5d4f98a2 YZ |
1715 | } |
1716 | ||
f36f3042 CM |
1717 | int btrfs_transaction_in_commit(struct btrfs_fs_info *info) |
1718 | { | |
4a9d8bde | 1719 | struct btrfs_transaction *trans; |
f36f3042 | 1720 | int ret = 0; |
4a9d8bde | 1721 | |
a4abeea4 | 1722 | spin_lock(&info->trans_lock); |
4a9d8bde MX |
1723 | trans = info->running_transaction; |
1724 | if (trans) | |
1725 | ret = (trans->state >= TRANS_STATE_COMMIT_START); | |
a4abeea4 | 1726 | spin_unlock(&info->trans_lock); |
f36f3042 CM |
1727 | return ret; |
1728 | } | |
1729 | ||
8929ecfa YZ |
1730 | int btrfs_transaction_blocked(struct btrfs_fs_info *info) |
1731 | { | |
4a9d8bde | 1732 | struct btrfs_transaction *trans; |
8929ecfa | 1733 | int ret = 0; |
4a9d8bde | 1734 | |
a4abeea4 | 1735 | spin_lock(&info->trans_lock); |
4a9d8bde MX |
1736 | trans = info->running_transaction; |
1737 | if (trans) | |
1738 | ret = is_transaction_blocked(trans); | |
a4abeea4 | 1739 | spin_unlock(&info->trans_lock); |
8929ecfa YZ |
1740 | return ret; |
1741 | } | |
1742 | ||
bb9c12c9 SW |
1743 | /* |
1744 | * wait for the current transaction commit to start and block subsequent | |
1745 | * transaction joins | |
1746 | */ | |
2ff7e61e | 1747 | static void wait_current_trans_commit_start(struct btrfs_fs_info *fs_info, |
bb9c12c9 SW |
1748 | struct btrfs_transaction *trans) |
1749 | { | |
2ff7e61e JM |
1750 | wait_event(fs_info->transaction_blocked_wait, |
1751 | trans->state >= TRANS_STATE_COMMIT_START || trans->aborted); | |
bb9c12c9 SW |
1752 | } |
1753 | ||
1754 | /* | |
1755 | * wait for the current transaction to start and then become unblocked. | |
1756 | * caller holds ref. | |
1757 | */ | |
2ff7e61e JM |
1758 | static void wait_current_trans_commit_start_and_unblock( |
1759 | struct btrfs_fs_info *fs_info, | |
1760 | struct btrfs_transaction *trans) | |
bb9c12c9 | 1761 | { |
2ff7e61e JM |
1762 | wait_event(fs_info->transaction_wait, |
1763 | trans->state >= TRANS_STATE_UNBLOCKED || trans->aborted); | |
bb9c12c9 SW |
1764 | } |
1765 | ||
1766 | /* | |
1767 | * commit transactions asynchronously. once btrfs_commit_transaction_async | |
1768 | * returns, any subsequent transaction will not be allowed to join. | |
1769 | */ | |
1770 | struct btrfs_async_commit { | |
1771 | struct btrfs_trans_handle *newtrans; | |
7892b5af | 1772 | struct work_struct work; |
bb9c12c9 SW |
1773 | }; |
1774 | ||
1775 | static void do_async_commit(struct work_struct *work) | |
1776 | { | |
1777 | struct btrfs_async_commit *ac = | |
7892b5af | 1778 | container_of(work, struct btrfs_async_commit, work); |
bb9c12c9 | 1779 | |
6fc4e354 SW |
1780 | /* |
1781 | * We've got freeze protection passed with the transaction. | |
1782 | * Tell lockdep about it. | |
1783 | */ | |
b1a06a4b | 1784 | if (ac->newtrans->type & __TRANS_FREEZABLE) |
3a45bb20 | 1785 | __sb_writers_acquired(ac->newtrans->fs_info->sb, SB_FREEZE_FS); |
6fc4e354 | 1786 | |
e209db7a SW |
1787 | current->journal_info = ac->newtrans; |
1788 | ||
3a45bb20 | 1789 | btrfs_commit_transaction(ac->newtrans); |
bb9c12c9 SW |
1790 | kfree(ac); |
1791 | } | |
1792 | ||
1793 | int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans, | |
bb9c12c9 SW |
1794 | int wait_for_unblock) |
1795 | { | |
3a45bb20 | 1796 | struct btrfs_fs_info *fs_info = trans->fs_info; |
bb9c12c9 SW |
1797 | struct btrfs_async_commit *ac; |
1798 | struct btrfs_transaction *cur_trans; | |
1799 | ||
1800 | ac = kmalloc(sizeof(*ac), GFP_NOFS); | |
db5b493a TI |
1801 | if (!ac) |
1802 | return -ENOMEM; | |
bb9c12c9 | 1803 | |
7892b5af | 1804 | INIT_WORK(&ac->work, do_async_commit); |
3a45bb20 | 1805 | ac->newtrans = btrfs_join_transaction(trans->root); |
3612b495 TI |
1806 | if (IS_ERR(ac->newtrans)) { |
1807 | int err = PTR_ERR(ac->newtrans); | |
1808 | kfree(ac); | |
1809 | return err; | |
1810 | } | |
bb9c12c9 SW |
1811 | |
1812 | /* take transaction reference */ | |
bb9c12c9 | 1813 | cur_trans = trans->transaction; |
9b64f57d | 1814 | refcount_inc(&cur_trans->use_count); |
bb9c12c9 | 1815 | |
3a45bb20 | 1816 | btrfs_end_transaction(trans); |
6fc4e354 SW |
1817 | |
1818 | /* | |
1819 | * Tell lockdep we've released the freeze rwsem, since the | |
1820 | * async commit thread will be the one to unlock it. | |
1821 | */ | |
b1a06a4b | 1822 | if (ac->newtrans->type & __TRANS_FREEZABLE) |
0b246afa | 1823 | __sb_writers_release(fs_info->sb, SB_FREEZE_FS); |
6fc4e354 | 1824 | |
7892b5af | 1825 | schedule_work(&ac->work); |
bb9c12c9 SW |
1826 | |
1827 | /* wait for transaction to start and unblock */ | |
bb9c12c9 | 1828 | if (wait_for_unblock) |
2ff7e61e | 1829 | wait_current_trans_commit_start_and_unblock(fs_info, cur_trans); |
bb9c12c9 | 1830 | else |
2ff7e61e | 1831 | wait_current_trans_commit_start(fs_info, cur_trans); |
bb9c12c9 | 1832 | |
38e88054 SW |
1833 | if (current->journal_info == trans) |
1834 | current->journal_info = NULL; | |
1835 | ||
724e2315 | 1836 | btrfs_put_transaction(cur_trans); |
bb9c12c9 SW |
1837 | return 0; |
1838 | } | |
1839 | ||
49b25e05 | 1840 | |
97cb39bb | 1841 | static void cleanup_transaction(struct btrfs_trans_handle *trans, int err) |
49b25e05 | 1842 | { |
97cb39bb | 1843 | struct btrfs_fs_info *fs_info = trans->fs_info; |
49b25e05 | 1844 | struct btrfs_transaction *cur_trans = trans->transaction; |
f094ac32 | 1845 | DEFINE_WAIT(wait); |
49b25e05 | 1846 | |
b50fff81 | 1847 | WARN_ON(refcount_read(&trans->use_count) > 1); |
49b25e05 | 1848 | |
66642832 | 1849 | btrfs_abort_transaction(trans, err); |
7b8b92af | 1850 | |
0b246afa | 1851 | spin_lock(&fs_info->trans_lock); |
66b6135b | 1852 | |
25d8c284 MX |
1853 | /* |
1854 | * If the transaction is removed from the list, it means this | |
1855 | * transaction has been committed successfully, so it is impossible | |
1856 | * to call the cleanup function. | |
1857 | */ | |
1858 | BUG_ON(list_empty(&cur_trans->list)); | |
66b6135b | 1859 | |
49b25e05 | 1860 | list_del_init(&cur_trans->list); |
0b246afa | 1861 | if (cur_trans == fs_info->running_transaction) { |
4a9d8bde | 1862 | cur_trans->state = TRANS_STATE_COMMIT_DOING; |
0b246afa | 1863 | spin_unlock(&fs_info->trans_lock); |
f094ac32 LB |
1864 | wait_event(cur_trans->writer_wait, |
1865 | atomic_read(&cur_trans->num_writers) == 1); | |
1866 | ||
0b246afa | 1867 | spin_lock(&fs_info->trans_lock); |
d7096fc3 | 1868 | } |
0b246afa | 1869 | spin_unlock(&fs_info->trans_lock); |
49b25e05 | 1870 | |
2ff7e61e | 1871 | btrfs_cleanup_one_transaction(trans->transaction, fs_info); |
49b25e05 | 1872 | |
0b246afa JM |
1873 | spin_lock(&fs_info->trans_lock); |
1874 | if (cur_trans == fs_info->running_transaction) | |
1875 | fs_info->running_transaction = NULL; | |
1876 | spin_unlock(&fs_info->trans_lock); | |
4a9d8bde | 1877 | |
e0228285 | 1878 | if (trans->type & __TRANS_FREEZABLE) |
0b246afa | 1879 | sb_end_intwrite(fs_info->sb); |
724e2315 JB |
1880 | btrfs_put_transaction(cur_trans); |
1881 | btrfs_put_transaction(cur_trans); | |
49b25e05 | 1882 | |
97cb39bb | 1883 | trace_btrfs_transaction_commit(trans->root); |
49b25e05 | 1884 | |
49b25e05 JM |
1885 | if (current->journal_info == trans) |
1886 | current->journal_info = NULL; | |
0b246afa | 1887 | btrfs_scrub_cancel(fs_info); |
49b25e05 JM |
1888 | |
1889 | kmem_cache_free(btrfs_trans_handle_cachep, trans); | |
1890 | } | |
1891 | ||
82436617 MX |
1892 | static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info) |
1893 | { | |
ce8ea7cc JB |
1894 | /* |
1895 | * We use writeback_inodes_sb here because if we used | |
1896 | * btrfs_start_delalloc_roots we would deadlock with fs freeze. | |
1897 | * Currently are holding the fs freeze lock, if we do an async flush | |
1898 | * we'll do btrfs_join_transaction() and deadlock because we need to | |
1899 | * wait for the fs freeze lock. Using the direct flushing we benefit | |
1900 | * from already being in a transaction and our join_transaction doesn't | |
1901 | * have to re-take the fs freeze lock. | |
1902 | */ | |
3cdde224 | 1903 | if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) |
ce8ea7cc | 1904 | writeback_inodes_sb(fs_info->sb, WB_REASON_SYNC); |
82436617 MX |
1905 | return 0; |
1906 | } | |
1907 | ||
1908 | static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info) | |
1909 | { | |
3cdde224 | 1910 | if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) |
6374e57a | 1911 | btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1); |
82436617 MX |
1912 | } |
1913 | ||
50d9aa99 | 1914 | static inline void |
161c3549 | 1915 | btrfs_wait_pending_ordered(struct btrfs_transaction *cur_trans) |
50d9aa99 | 1916 | { |
161c3549 JB |
1917 | wait_event(cur_trans->pending_wait, |
1918 | atomic_read(&cur_trans->pending_ordered) == 0); | |
50d9aa99 JB |
1919 | } |
1920 | ||
3a45bb20 | 1921 | int btrfs_commit_transaction(struct btrfs_trans_handle *trans) |
79154b1b | 1922 | { |
3a45bb20 | 1923 | struct btrfs_fs_info *fs_info = trans->fs_info; |
49b25e05 | 1924 | struct btrfs_transaction *cur_trans = trans->transaction; |
8fd17795 | 1925 | struct btrfs_transaction *prev_trans = NULL; |
25287e0a | 1926 | int ret; |
79154b1b | 1927 | |
8d25a086 | 1928 | /* Stop the commit early if ->aborted is set */ |
20c7bcec | 1929 | if (unlikely(READ_ONCE(cur_trans->aborted))) { |
25287e0a | 1930 | ret = cur_trans->aborted; |
3a45bb20 | 1931 | btrfs_end_transaction(trans); |
e4a2bcac | 1932 | return ret; |
25287e0a | 1933 | } |
49b25e05 | 1934 | |
56bec294 CM |
1935 | /* make a pass through all the delayed refs we have so far |
1936 | * any runnings procs may add more while we are here | |
1937 | */ | |
c79a70b1 | 1938 | ret = btrfs_run_delayed_refs(trans, 0); |
e4a2bcac | 1939 | if (ret) { |
3a45bb20 | 1940 | btrfs_end_transaction(trans); |
e4a2bcac JB |
1941 | return ret; |
1942 | } | |
56bec294 | 1943 | |
dc60c525 | 1944 | btrfs_trans_release_metadata(trans); |
0e721106 JB |
1945 | trans->block_rsv = NULL; |
1946 | ||
b7ec40d7 | 1947 | cur_trans = trans->transaction; |
49b25e05 | 1948 | |
56bec294 CM |
1949 | /* |
1950 | * set the flushing flag so procs in this transaction have to | |
1951 | * start sending their work down. | |
1952 | */ | |
b7ec40d7 | 1953 | cur_trans->delayed_refs.flushing = 1; |
1be41b78 | 1954 | smp_wmb(); |
56bec294 | 1955 | |
ea658bad | 1956 | if (!list_empty(&trans->new_bgs)) |
6c686b35 | 1957 | btrfs_create_pending_block_groups(trans); |
ea658bad | 1958 | |
c79a70b1 | 1959 | ret = btrfs_run_delayed_refs(trans, 0); |
e4a2bcac | 1960 | if (ret) { |
3a45bb20 | 1961 | btrfs_end_transaction(trans); |
e4a2bcac JB |
1962 | return ret; |
1963 | } | |
56bec294 | 1964 | |
3204d33c | 1965 | if (!test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &cur_trans->flags)) { |
1bbc621e CM |
1966 | int run_it = 0; |
1967 | ||
1968 | /* this mutex is also taken before trying to set | |
1969 | * block groups readonly. We need to make sure | |
1970 | * that nobody has set a block group readonly | |
1971 | * after a extents from that block group have been | |
1972 | * allocated for cache files. btrfs_set_block_group_ro | |
1973 | * will wait for the transaction to commit if it | |
3204d33c | 1974 | * finds BTRFS_TRANS_DIRTY_BG_RUN set. |
1bbc621e | 1975 | * |
3204d33c JB |
1976 | * The BTRFS_TRANS_DIRTY_BG_RUN flag is also used to make sure |
1977 | * only one process starts all the block group IO. It wouldn't | |
1bbc621e CM |
1978 | * hurt to have more than one go through, but there's no |
1979 | * real advantage to it either. | |
1980 | */ | |
0b246afa | 1981 | mutex_lock(&fs_info->ro_block_group_mutex); |
3204d33c JB |
1982 | if (!test_and_set_bit(BTRFS_TRANS_DIRTY_BG_RUN, |
1983 | &cur_trans->flags)) | |
1bbc621e | 1984 | run_it = 1; |
0b246afa | 1985 | mutex_unlock(&fs_info->ro_block_group_mutex); |
1bbc621e | 1986 | |
f9cacae3 | 1987 | if (run_it) { |
21217054 | 1988 | ret = btrfs_start_dirty_block_groups(trans); |
f9cacae3 NB |
1989 | if (ret) { |
1990 | btrfs_end_transaction(trans); | |
1991 | return ret; | |
1992 | } | |
1993 | } | |
1bbc621e CM |
1994 | } |
1995 | ||
0b246afa | 1996 | spin_lock(&fs_info->trans_lock); |
4a9d8bde | 1997 | if (cur_trans->state >= TRANS_STATE_COMMIT_START) { |
0b246afa | 1998 | spin_unlock(&fs_info->trans_lock); |
9b64f57d | 1999 | refcount_inc(&cur_trans->use_count); |
3a45bb20 | 2000 | ret = btrfs_end_transaction(trans); |
ccd467d6 | 2001 | |
2ff7e61e | 2002 | wait_for_commit(cur_trans); |
15ee9bc7 | 2003 | |
b4924a0f LB |
2004 | if (unlikely(cur_trans->aborted)) |
2005 | ret = cur_trans->aborted; | |
2006 | ||
724e2315 | 2007 | btrfs_put_transaction(cur_trans); |
15ee9bc7 | 2008 | |
49b25e05 | 2009 | return ret; |
79154b1b | 2010 | } |
4313b399 | 2011 | |
4a9d8bde | 2012 | cur_trans->state = TRANS_STATE_COMMIT_START; |
0b246afa | 2013 | wake_up(&fs_info->transaction_blocked_wait); |
bb9c12c9 | 2014 | |
0b246afa | 2015 | if (cur_trans->list.prev != &fs_info->trans_list) { |
ccd467d6 CM |
2016 | prev_trans = list_entry(cur_trans->list.prev, |
2017 | struct btrfs_transaction, list); | |
4a9d8bde | 2018 | if (prev_trans->state != TRANS_STATE_COMPLETED) { |
9b64f57d | 2019 | refcount_inc(&prev_trans->use_count); |
0b246afa | 2020 | spin_unlock(&fs_info->trans_lock); |
ccd467d6 | 2021 | |
2ff7e61e | 2022 | wait_for_commit(prev_trans); |
1f9b8c8f | 2023 | ret = prev_trans->aborted; |
ccd467d6 | 2024 | |
724e2315 | 2025 | btrfs_put_transaction(prev_trans); |
1f9b8c8f FM |
2026 | if (ret) |
2027 | goto cleanup_transaction; | |
a4abeea4 | 2028 | } else { |
0b246afa | 2029 | spin_unlock(&fs_info->trans_lock); |
ccd467d6 | 2030 | } |
a4abeea4 | 2031 | } else { |
0b246afa | 2032 | spin_unlock(&fs_info->trans_lock); |
ccd467d6 | 2033 | } |
15ee9bc7 | 2034 | |
0860adfd MX |
2035 | extwriter_counter_dec(cur_trans, trans->type); |
2036 | ||
0b246afa | 2037 | ret = btrfs_start_delalloc_flush(fs_info); |
82436617 MX |
2038 | if (ret) |
2039 | goto cleanup_transaction; | |
2040 | ||
e5c304e6 | 2041 | ret = btrfs_run_delayed_items(trans); |
581227d0 MX |
2042 | if (ret) |
2043 | goto cleanup_transaction; | |
15ee9bc7 | 2044 | |
581227d0 MX |
2045 | wait_event(cur_trans->writer_wait, |
2046 | extwriter_counter_read(cur_trans) == 0); | |
15ee9bc7 | 2047 | |
581227d0 | 2048 | /* some pending stuffs might be added after the previous flush. */ |
e5c304e6 | 2049 | ret = btrfs_run_delayed_items(trans); |
ca469637 MX |
2050 | if (ret) |
2051 | goto cleanup_transaction; | |
2052 | ||
0b246afa | 2053 | btrfs_wait_delalloc_flush(fs_info); |
cb7ab021 | 2054 | |
161c3549 | 2055 | btrfs_wait_pending_ordered(cur_trans); |
50d9aa99 | 2056 | |
2ff7e61e | 2057 | btrfs_scrub_pause(fs_info); |
ed0ca140 JB |
2058 | /* |
2059 | * Ok now we need to make sure to block out any other joins while we | |
2060 | * commit the transaction. We could have started a join before setting | |
4a9d8bde | 2061 | * COMMIT_DOING so make sure to wait for num_writers to == 1 again. |
ed0ca140 | 2062 | */ |
0b246afa | 2063 | spin_lock(&fs_info->trans_lock); |
4a9d8bde | 2064 | cur_trans->state = TRANS_STATE_COMMIT_DOING; |
0b246afa | 2065 | spin_unlock(&fs_info->trans_lock); |
ed0ca140 JB |
2066 | wait_event(cur_trans->writer_wait, |
2067 | atomic_read(&cur_trans->num_writers) == 1); | |
2068 | ||
2cba30f1 | 2069 | /* ->aborted might be set after the previous check, so check it */ |
20c7bcec | 2070 | if (unlikely(READ_ONCE(cur_trans->aborted))) { |
2cba30f1 | 2071 | ret = cur_trans->aborted; |
6cf7f77e | 2072 | goto scrub_continue; |
2cba30f1 | 2073 | } |
7585717f CM |
2074 | /* |
2075 | * the reloc mutex makes sure that we stop | |
2076 | * the balancing code from coming in and moving | |
2077 | * extents around in the middle of the commit | |
2078 | */ | |
0b246afa | 2079 | mutex_lock(&fs_info->reloc_mutex); |
7585717f | 2080 | |
42874b3d MX |
2081 | /* |
2082 | * We needn't worry about the delayed items because we will | |
2083 | * deal with them in create_pending_snapshot(), which is the | |
2084 | * core function of the snapshot creation. | |
2085 | */ | |
08d50ca3 | 2086 | ret = create_pending_snapshots(trans); |
49b25e05 | 2087 | if (ret) { |
0b246afa | 2088 | mutex_unlock(&fs_info->reloc_mutex); |
6cf7f77e | 2089 | goto scrub_continue; |
49b25e05 | 2090 | } |
3063d29f | 2091 | |
42874b3d MX |
2092 | /* |
2093 | * We insert the dir indexes of the snapshots and update the inode | |
2094 | * of the snapshots' parents after the snapshot creation, so there | |
2095 | * are some delayed items which are not dealt with. Now deal with | |
2096 | * them. | |
2097 | * | |
2098 | * We needn't worry that this operation will corrupt the snapshots, | |
2099 | * because all the tree which are snapshoted will be forced to COW | |
2100 | * the nodes and leaves. | |
2101 | */ | |
e5c304e6 | 2102 | ret = btrfs_run_delayed_items(trans); |
49b25e05 | 2103 | if (ret) { |
0b246afa | 2104 | mutex_unlock(&fs_info->reloc_mutex); |
6cf7f77e | 2105 | goto scrub_continue; |
49b25e05 | 2106 | } |
16cdcec7 | 2107 | |
c79a70b1 | 2108 | ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); |
49b25e05 | 2109 | if (ret) { |
0b246afa | 2110 | mutex_unlock(&fs_info->reloc_mutex); |
6cf7f77e | 2111 | goto scrub_continue; |
49b25e05 | 2112 | } |
56bec294 | 2113 | |
e999376f CM |
2114 | /* |
2115 | * make sure none of the code above managed to slip in a | |
2116 | * delayed item | |
2117 | */ | |
ccdf9b30 | 2118 | btrfs_assert_delayed_root_empty(fs_info); |
e999376f | 2119 | |
2c90e5d6 | 2120 | WARN_ON(cur_trans != trans->transaction); |
dc17ff8f | 2121 | |
e02119d5 CM |
2122 | /* btrfs_commit_tree_roots is responsible for getting the |
2123 | * various roots consistent with each other. Every pointer | |
2124 | * in the tree of tree roots has to point to the most up to date | |
2125 | * root for every subvolume and other tree. So, we have to keep | |
2126 | * the tree logging code from jumping in and changing any | |
2127 | * of the trees. | |
2128 | * | |
2129 | * At this point in the commit, there can't be any tree-log | |
2130 | * writers, but a little lower down we drop the trans mutex | |
2131 | * and let new people in. By holding the tree_log_mutex | |
2132 | * from now until after the super is written, we avoid races | |
2133 | * with the tree-log code. | |
2134 | */ | |
0b246afa | 2135 | mutex_lock(&fs_info->tree_log_mutex); |
e02119d5 | 2136 | |
7e4443d9 | 2137 | ret = commit_fs_roots(trans); |
49b25e05 | 2138 | if (ret) { |
0b246afa JM |
2139 | mutex_unlock(&fs_info->tree_log_mutex); |
2140 | mutex_unlock(&fs_info->reloc_mutex); | |
6cf7f77e | 2141 | goto scrub_continue; |
49b25e05 | 2142 | } |
54aa1f4d | 2143 | |
3818aea2 | 2144 | /* |
7e1876ac DS |
2145 | * Since the transaction is done, we can apply the pending changes |
2146 | * before the next transaction. | |
3818aea2 | 2147 | */ |
0b246afa | 2148 | btrfs_apply_pending_changes(fs_info); |
3818aea2 | 2149 | |
5d4f98a2 | 2150 | /* commit_fs_roots gets rid of all the tree log roots, it is now |
e02119d5 CM |
2151 | * safe to free the root of tree log roots |
2152 | */ | |
0b246afa | 2153 | btrfs_free_log_root_tree(trans, fs_info); |
e02119d5 | 2154 | |
82bafb38 QW |
2155 | /* |
2156 | * commit_fs_roots() can call btrfs_save_ino_cache(), which generates | |
2157 | * new delayed refs. Must handle them or qgroup can be wrong. | |
2158 | */ | |
c79a70b1 | 2159 | ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); |
82bafb38 QW |
2160 | if (ret) { |
2161 | mutex_unlock(&fs_info->tree_log_mutex); | |
2162 | mutex_unlock(&fs_info->reloc_mutex); | |
2163 | goto scrub_continue; | |
2164 | } | |
2165 | ||
0ed4792a QW |
2166 | /* |
2167 | * Since fs roots are all committed, we can get a quite accurate | |
2168 | * new_roots. So let's do quota accounting. | |
2169 | */ | |
460fb20a | 2170 | ret = btrfs_qgroup_account_extents(trans); |
0ed4792a | 2171 | if (ret < 0) { |
0b246afa JM |
2172 | mutex_unlock(&fs_info->tree_log_mutex); |
2173 | mutex_unlock(&fs_info->reloc_mutex); | |
0ed4792a QW |
2174 | goto scrub_continue; |
2175 | } | |
2176 | ||
9386d8bc | 2177 | ret = commit_cowonly_roots(trans); |
49b25e05 | 2178 | if (ret) { |
0b246afa JM |
2179 | mutex_unlock(&fs_info->tree_log_mutex); |
2180 | mutex_unlock(&fs_info->reloc_mutex); | |
6cf7f77e | 2181 | goto scrub_continue; |
49b25e05 | 2182 | } |
54aa1f4d | 2183 | |
2cba30f1 MX |
2184 | /* |
2185 | * The tasks which save the space cache and inode cache may also | |
2186 | * update ->aborted, check it. | |
2187 | */ | |
20c7bcec | 2188 | if (unlikely(READ_ONCE(cur_trans->aborted))) { |
2cba30f1 | 2189 | ret = cur_trans->aborted; |
0b246afa JM |
2190 | mutex_unlock(&fs_info->tree_log_mutex); |
2191 | mutex_unlock(&fs_info->reloc_mutex); | |
6cf7f77e | 2192 | goto scrub_continue; |
2cba30f1 MX |
2193 | } |
2194 | ||
8b74c03e | 2195 | btrfs_prepare_extent_commit(fs_info); |
11833d66 | 2196 | |
0b246afa | 2197 | cur_trans = fs_info->running_transaction; |
5d4f98a2 | 2198 | |
0b246afa JM |
2199 | btrfs_set_root_node(&fs_info->tree_root->root_item, |
2200 | fs_info->tree_root->node); | |
2201 | list_add_tail(&fs_info->tree_root->dirty_list, | |
9e351cc8 | 2202 | &cur_trans->switch_commits); |
5d4f98a2 | 2203 | |
0b246afa JM |
2204 | btrfs_set_root_node(&fs_info->chunk_root->root_item, |
2205 | fs_info->chunk_root->node); | |
2206 | list_add_tail(&fs_info->chunk_root->dirty_list, | |
9e351cc8 JB |
2207 | &cur_trans->switch_commits); |
2208 | ||
16916a88 | 2209 | switch_commit_roots(cur_trans); |
5d4f98a2 | 2210 | |
ce93ec54 | 2211 | ASSERT(list_empty(&cur_trans->dirty_bgs)); |
1bbc621e | 2212 | ASSERT(list_empty(&cur_trans->io_bgs)); |
2ff7e61e | 2213 | update_super_roots(fs_info); |
e02119d5 | 2214 | |
0b246afa JM |
2215 | btrfs_set_super_log_root(fs_info->super_copy, 0); |
2216 | btrfs_set_super_log_root_level(fs_info->super_copy, 0); | |
2217 | memcpy(fs_info->super_for_commit, fs_info->super_copy, | |
2218 | sizeof(*fs_info->super_copy)); | |
ccd467d6 | 2219 | |
0b246afa | 2220 | btrfs_update_commit_device_size(fs_info); |
e9b919b1 | 2221 | btrfs_update_commit_device_bytes_used(cur_trans); |
935e5cc9 | 2222 | |
0b246afa JM |
2223 | clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags); |
2224 | clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags); | |
656f30db | 2225 | |
4fbcdf66 FM |
2226 | btrfs_trans_release_chunk_metadata(trans); |
2227 | ||
0b246afa | 2228 | spin_lock(&fs_info->trans_lock); |
4a9d8bde | 2229 | cur_trans->state = TRANS_STATE_UNBLOCKED; |
0b246afa JM |
2230 | fs_info->running_transaction = NULL; |
2231 | spin_unlock(&fs_info->trans_lock); | |
2232 | mutex_unlock(&fs_info->reloc_mutex); | |
b7ec40d7 | 2233 | |
0b246afa | 2234 | wake_up(&fs_info->transaction_wait); |
e6dcd2dc | 2235 | |
70458a58 | 2236 | ret = btrfs_write_and_wait_transaction(trans); |
49b25e05 | 2237 | if (ret) { |
0b246afa JM |
2238 | btrfs_handle_fs_error(fs_info, ret, |
2239 | "Error while writing out transaction"); | |
2240 | mutex_unlock(&fs_info->tree_log_mutex); | |
6cf7f77e | 2241 | goto scrub_continue; |
49b25e05 JM |
2242 | } |
2243 | ||
eece6a9c | 2244 | ret = write_all_supers(fs_info, 0); |
e02119d5 CM |
2245 | /* |
2246 | * the super is written, we can safely allow the tree-loggers | |
2247 | * to go about their business | |
2248 | */ | |
0b246afa | 2249 | mutex_unlock(&fs_info->tree_log_mutex); |
c1f32b7c AJ |
2250 | if (ret) |
2251 | goto scrub_continue; | |
e02119d5 | 2252 | |
5ead2dd0 | 2253 | btrfs_finish_extent_commit(trans); |
4313b399 | 2254 | |
3204d33c | 2255 | if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS, &cur_trans->flags)) |
0b246afa | 2256 | btrfs_clear_space_info_full(fs_info); |
13212b54 | 2257 | |
0b246afa | 2258 | fs_info->last_trans_committed = cur_trans->transid; |
4a9d8bde MX |
2259 | /* |
2260 | * We needn't acquire the lock here because there is no other task | |
2261 | * which can change it. | |
2262 | */ | |
2263 | cur_trans->state = TRANS_STATE_COMPLETED; | |
2c90e5d6 | 2264 | wake_up(&cur_trans->commit_wait); |
a514d638 | 2265 | clear_bit(BTRFS_FS_NEED_ASYNC_COMMIT, &fs_info->flags); |
3de4586c | 2266 | |
0b246afa | 2267 | spin_lock(&fs_info->trans_lock); |
13c5a93e | 2268 | list_del_init(&cur_trans->list); |
0b246afa | 2269 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 | 2270 | |
724e2315 JB |
2271 | btrfs_put_transaction(cur_trans); |
2272 | btrfs_put_transaction(cur_trans); | |
58176a96 | 2273 | |
0860adfd | 2274 | if (trans->type & __TRANS_FREEZABLE) |
0b246afa | 2275 | sb_end_intwrite(fs_info->sb); |
b2b5ef5c | 2276 | |
3a45bb20 | 2277 | trace_btrfs_transaction_commit(trans->root); |
1abe9b8a | 2278 | |
2ff7e61e | 2279 | btrfs_scrub_continue(fs_info); |
a2de733c | 2280 | |
9ed74f2d JB |
2281 | if (current->journal_info == trans) |
2282 | current->journal_info = NULL; | |
2283 | ||
2c90e5d6 | 2284 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
24bbcf04 | 2285 | |
9e7cc91a WX |
2286 | /* |
2287 | * If fs has been frozen, we can not handle delayed iputs, otherwise | |
2288 | * it'll result in deadlock about SB_FREEZE_FS. | |
2289 | */ | |
0b246afa | 2290 | if (current != fs_info->transaction_kthread && |
fac03c8d DS |
2291 | current != fs_info->cleaner_kthread && |
2292 | !test_bit(BTRFS_FS_FROZEN, &fs_info->flags)) | |
2ff7e61e | 2293 | btrfs_run_delayed_iputs(fs_info); |
24bbcf04 | 2294 | |
79154b1b | 2295 | return ret; |
49b25e05 | 2296 | |
6cf7f77e | 2297 | scrub_continue: |
2ff7e61e | 2298 | btrfs_scrub_continue(fs_info); |
49b25e05 | 2299 | cleanup_transaction: |
dc60c525 | 2300 | btrfs_trans_release_metadata(trans); |
4fbcdf66 | 2301 | btrfs_trans_release_chunk_metadata(trans); |
0e721106 | 2302 | trans->block_rsv = NULL; |
0b246afa | 2303 | btrfs_warn(fs_info, "Skipping commit of aborted transaction."); |
49b25e05 JM |
2304 | if (current->journal_info == trans) |
2305 | current->journal_info = NULL; | |
97cb39bb | 2306 | cleanup_transaction(trans, ret); |
49b25e05 JM |
2307 | |
2308 | return ret; | |
79154b1b CM |
2309 | } |
2310 | ||
d352ac68 | 2311 | /* |
9d1a2a3a DS |
2312 | * return < 0 if error |
2313 | * 0 if there are no more dead_roots at the time of call | |
2314 | * 1 there are more to be processed, call me again | |
2315 | * | |
2316 | * The return value indicates there are certainly more snapshots to delete, but | |
2317 | * if there comes a new one during processing, it may return 0. We don't mind, | |
2318 | * because btrfs_commit_super will poke cleaner thread and it will process it a | |
2319 | * few seconds later. | |
d352ac68 | 2320 | */ |
9d1a2a3a | 2321 | int btrfs_clean_one_deleted_snapshot(struct btrfs_root *root) |
e9d0b13b | 2322 | { |
9d1a2a3a | 2323 | int ret; |
5d4f98a2 YZ |
2324 | struct btrfs_fs_info *fs_info = root->fs_info; |
2325 | ||
a4abeea4 | 2326 | spin_lock(&fs_info->trans_lock); |
9d1a2a3a DS |
2327 | if (list_empty(&fs_info->dead_roots)) { |
2328 | spin_unlock(&fs_info->trans_lock); | |
2329 | return 0; | |
2330 | } | |
2331 | root = list_first_entry(&fs_info->dead_roots, | |
2332 | struct btrfs_root, root_list); | |
cfad392b | 2333 | list_del_init(&root->root_list); |
a4abeea4 | 2334 | spin_unlock(&fs_info->trans_lock); |
e9d0b13b | 2335 | |
4fd786e6 | 2336 | btrfs_debug(fs_info, "cleaner removing %llu", root->root_key.objectid); |
76dda93c | 2337 | |
9d1a2a3a | 2338 | btrfs_kill_all_delayed_nodes(root); |
16cdcec7 | 2339 | |
9d1a2a3a DS |
2340 | if (btrfs_header_backref_rev(root->node) < |
2341 | BTRFS_MIXED_BACKREF_REV) | |
2342 | ret = btrfs_drop_snapshot(root, NULL, 0, 0); | |
2343 | else | |
2344 | ret = btrfs_drop_snapshot(root, NULL, 1, 0); | |
32471dc2 | 2345 | |
6596a928 | 2346 | return (ret < 0) ? 0 : 1; |
e9d0b13b | 2347 | } |
572d9ab7 DS |
2348 | |
2349 | void btrfs_apply_pending_changes(struct btrfs_fs_info *fs_info) | |
2350 | { | |
2351 | unsigned long prev; | |
2352 | unsigned long bit; | |
2353 | ||
6c9fe14f | 2354 | prev = xchg(&fs_info->pending_changes, 0); |
572d9ab7 DS |
2355 | if (!prev) |
2356 | return; | |
2357 | ||
7e1876ac DS |
2358 | bit = 1 << BTRFS_PENDING_SET_INODE_MAP_CACHE; |
2359 | if (prev & bit) | |
2360 | btrfs_set_opt(fs_info->mount_opt, INODE_MAP_CACHE); | |
2361 | prev &= ~bit; | |
2362 | ||
2363 | bit = 1 << BTRFS_PENDING_CLEAR_INODE_MAP_CACHE; | |
2364 | if (prev & bit) | |
2365 | btrfs_clear_opt(fs_info->mount_opt, INODE_MAP_CACHE); | |
2366 | prev &= ~bit; | |
2367 | ||
d51033d0 DS |
2368 | bit = 1 << BTRFS_PENDING_COMMIT; |
2369 | if (prev & bit) | |
2370 | btrfs_debug(fs_info, "pending commit done"); | |
2371 | prev &= ~bit; | |
2372 | ||
572d9ab7 DS |
2373 | if (prev) |
2374 | btrfs_warn(fs_info, | |
2375 | "unknown pending changes left 0x%lx, ignoring", prev); | |
2376 | } |