]>
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 | 30 | __TRANS_ATTACH | |
a6d155d2 FM |
31 | __TRANS_JOIN | |
32 | __TRANS_JOIN_NOSTART), | |
bcf3a3e7 | 33 | [TRANS_STATE_UNBLOCKED] = (__TRANS_START | |
4a9d8bde MX |
34 | __TRANS_ATTACH | |
35 | __TRANS_JOIN | | |
a6d155d2 FM |
36 | __TRANS_JOIN_NOLOCK | |
37 | __TRANS_JOIN_NOSTART), | |
bcf3a3e7 | 38 | [TRANS_STATE_COMPLETED] = (__TRANS_START | |
4a9d8bde MX |
39 | __TRANS_ATTACH | |
40 | __TRANS_JOIN | | |
a6d155d2 FM |
41 | __TRANS_JOIN_NOLOCK | |
42 | __TRANS_JOIN_NOSTART), | |
4a9d8bde MX |
43 | }; |
44 | ||
724e2315 | 45 | void btrfs_put_transaction(struct btrfs_transaction *transaction) |
79154b1b | 46 | { |
9b64f57d ER |
47 | WARN_ON(refcount_read(&transaction->use_count) == 0); |
48 | if (refcount_dec_and_test(&transaction->use_count)) { | |
a4abeea4 | 49 | BUG_ON(!list_empty(&transaction->list)); |
5c9d028b LB |
50 | WARN_ON(!RB_EMPTY_ROOT( |
51 | &transaction->delayed_refs.href_root.rb_root)); | |
1262133b | 52 | if (transaction->delayed_refs.pending_csums) |
ab8d0fc4 JM |
53 | btrfs_err(transaction->fs_info, |
54 | "pending csums is %llu", | |
55 | transaction->delayed_refs.pending_csums); | |
7785a663 FM |
56 | /* |
57 | * If any block groups are found in ->deleted_bgs then it's | |
58 | * because the transaction was aborted and a commit did not | |
59 | * happen (things failed before writing the new superblock | |
60 | * and calling btrfs_finish_extent_commit()), so we can not | |
61 | * discard the physical locations of the block groups. | |
62 | */ | |
63 | while (!list_empty(&transaction->deleted_bgs)) { | |
64 | struct btrfs_block_group_cache *cache; | |
65 | ||
66 | cache = list_first_entry(&transaction->deleted_bgs, | |
67 | struct btrfs_block_group_cache, | |
68 | bg_list); | |
69 | list_del_init(&cache->bg_list); | |
70 | btrfs_put_block_group_trimming(cache); | |
71 | btrfs_put_block_group(cache); | |
72 | } | |
bbbf7243 | 73 | WARN_ON(!list_empty(&transaction->dev_update_list)); |
4b5faeac | 74 | kfree(transaction); |
78fae27e | 75 | } |
79154b1b CM |
76 | } |
77 | ||
16916a88 | 78 | static noinline void switch_commit_roots(struct btrfs_transaction *trans) |
817d52f8 | 79 | { |
16916a88 | 80 | struct btrfs_fs_info *fs_info = trans->fs_info; |
9e351cc8 JB |
81 | struct btrfs_root *root, *tmp; |
82 | ||
83 | down_write(&fs_info->commit_root_sem); | |
84 | list_for_each_entry_safe(root, tmp, &trans->switch_commits, | |
85 | dirty_list) { | |
86 | list_del_init(&root->dirty_list); | |
87 | free_extent_buffer(root->commit_root); | |
88 | root->commit_root = btrfs_root_node(root); | |
4fd786e6 | 89 | if (is_fstree(root->root_key.objectid)) |
9e351cc8 | 90 | btrfs_unpin_free_ino(root); |
41e7acd3 | 91 | extent_io_tree_release(&root->dirty_log_pages); |
370a11b8 | 92 | btrfs_qgroup_clean_swapped_blocks(root); |
9e351cc8 | 93 | } |
2b9dbef2 JB |
94 | |
95 | /* We can free old roots now. */ | |
96 | spin_lock(&trans->dropped_roots_lock); | |
97 | while (!list_empty(&trans->dropped_roots)) { | |
98 | root = list_first_entry(&trans->dropped_roots, | |
99 | struct btrfs_root, root_list); | |
100 | list_del_init(&root->root_list); | |
101 | spin_unlock(&trans->dropped_roots_lock); | |
102 | btrfs_drop_and_free_fs_root(fs_info, root); | |
103 | spin_lock(&trans->dropped_roots_lock); | |
104 | } | |
105 | spin_unlock(&trans->dropped_roots_lock); | |
9e351cc8 | 106 | up_write(&fs_info->commit_root_sem); |
817d52f8 JB |
107 | } |
108 | ||
0860adfd MX |
109 | static inline void extwriter_counter_inc(struct btrfs_transaction *trans, |
110 | unsigned int type) | |
111 | { | |
112 | if (type & TRANS_EXTWRITERS) | |
113 | atomic_inc(&trans->num_extwriters); | |
114 | } | |
115 | ||
116 | static inline void extwriter_counter_dec(struct btrfs_transaction *trans, | |
117 | unsigned int type) | |
118 | { | |
119 | if (type & TRANS_EXTWRITERS) | |
120 | atomic_dec(&trans->num_extwriters); | |
121 | } | |
122 | ||
123 | static inline void extwriter_counter_init(struct btrfs_transaction *trans, | |
124 | unsigned int type) | |
125 | { | |
126 | atomic_set(&trans->num_extwriters, ((type & TRANS_EXTWRITERS) ? 1 : 0)); | |
127 | } | |
128 | ||
129 | static inline int extwriter_counter_read(struct btrfs_transaction *trans) | |
130 | { | |
131 | return atomic_read(&trans->num_extwriters); | |
178260b2 MX |
132 | } |
133 | ||
fb6dea26 JB |
134 | /* |
135 | * To be called after all the new block groups attached to the transaction | |
136 | * handle have been created (btrfs_create_pending_block_groups()). | |
137 | */ | |
138 | void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans) | |
139 | { | |
140 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
141 | ||
142 | if (!trans->chunk_bytes_reserved) | |
143 | return; | |
144 | ||
145 | WARN_ON_ONCE(!list_empty(&trans->new_bgs)); | |
146 | ||
147 | btrfs_block_rsv_release(fs_info, &fs_info->chunk_block_rsv, | |
148 | trans->chunk_bytes_reserved); | |
149 | trans->chunk_bytes_reserved = 0; | |
150 | } | |
151 | ||
d352ac68 CM |
152 | /* |
153 | * either allocate a new transaction or hop into the existing one | |
154 | */ | |
2ff7e61e JM |
155 | static noinline int join_transaction(struct btrfs_fs_info *fs_info, |
156 | unsigned int type) | |
79154b1b CM |
157 | { |
158 | struct btrfs_transaction *cur_trans; | |
a4abeea4 | 159 | |
19ae4e81 | 160 | spin_lock(&fs_info->trans_lock); |
d43317dc | 161 | loop: |
49b25e05 | 162 | /* The file system has been taken offline. No new transactions. */ |
87533c47 | 163 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { |
19ae4e81 | 164 | spin_unlock(&fs_info->trans_lock); |
49b25e05 JM |
165 | return -EROFS; |
166 | } | |
167 | ||
19ae4e81 | 168 | cur_trans = fs_info->running_transaction; |
a4abeea4 | 169 | if (cur_trans) { |
871383be | 170 | if (cur_trans->aborted) { |
19ae4e81 | 171 | spin_unlock(&fs_info->trans_lock); |
49b25e05 | 172 | return cur_trans->aborted; |
871383be | 173 | } |
4a9d8bde | 174 | if (btrfs_blocked_trans_types[cur_trans->state] & type) { |
178260b2 MX |
175 | spin_unlock(&fs_info->trans_lock); |
176 | return -EBUSY; | |
177 | } | |
9b64f57d | 178 | refcount_inc(&cur_trans->use_count); |
13c5a93e | 179 | atomic_inc(&cur_trans->num_writers); |
0860adfd | 180 | extwriter_counter_inc(cur_trans, type); |
19ae4e81 | 181 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 | 182 | return 0; |
79154b1b | 183 | } |
19ae4e81 | 184 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 | 185 | |
354aa0fb MX |
186 | /* |
187 | * If we are ATTACH, we just want to catch the current transaction, | |
188 | * and commit it. If there is no transaction, just return ENOENT. | |
189 | */ | |
190 | if (type == TRANS_ATTACH) | |
191 | return -ENOENT; | |
192 | ||
4a9d8bde MX |
193 | /* |
194 | * JOIN_NOLOCK only happens during the transaction commit, so | |
195 | * it is impossible that ->running_transaction is NULL | |
196 | */ | |
197 | BUG_ON(type == TRANS_JOIN_NOLOCK); | |
198 | ||
4b5faeac | 199 | cur_trans = kmalloc(sizeof(*cur_trans), GFP_NOFS); |
a4abeea4 JB |
200 | if (!cur_trans) |
201 | return -ENOMEM; | |
d43317dc | 202 | |
19ae4e81 JS |
203 | spin_lock(&fs_info->trans_lock); |
204 | if (fs_info->running_transaction) { | |
d43317dc CM |
205 | /* |
206 | * someone started a transaction after we unlocked. Make sure | |
4a9d8bde | 207 | * to redo the checks above |
d43317dc | 208 | */ |
4b5faeac | 209 | kfree(cur_trans); |
d43317dc | 210 | goto loop; |
87533c47 | 211 | } else if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { |
e4b50e14 | 212 | spin_unlock(&fs_info->trans_lock); |
4b5faeac | 213 | kfree(cur_trans); |
7b8b92af | 214 | return -EROFS; |
79154b1b | 215 | } |
d43317dc | 216 | |
ab8d0fc4 | 217 | cur_trans->fs_info = fs_info; |
a4abeea4 | 218 | atomic_set(&cur_trans->num_writers, 1); |
0860adfd | 219 | extwriter_counter_init(cur_trans, type); |
a4abeea4 JB |
220 | init_waitqueue_head(&cur_trans->writer_wait); |
221 | init_waitqueue_head(&cur_trans->commit_wait); | |
4a9d8bde | 222 | cur_trans->state = TRANS_STATE_RUNNING; |
a4abeea4 JB |
223 | /* |
224 | * One for this trans handle, one so it will live on until we | |
225 | * commit the transaction. | |
226 | */ | |
9b64f57d | 227 | refcount_set(&cur_trans->use_count, 2); |
3204d33c | 228 | cur_trans->flags = 0; |
afd48513 | 229 | cur_trans->start_time = ktime_get_seconds(); |
a4abeea4 | 230 | |
a099d0fd AM |
231 | memset(&cur_trans->delayed_refs, 0, sizeof(cur_trans->delayed_refs)); |
232 | ||
5c9d028b | 233 | cur_trans->delayed_refs.href_root = RB_ROOT_CACHED; |
3368d001 | 234 | cur_trans->delayed_refs.dirty_extent_root = RB_ROOT; |
d7df2c79 | 235 | atomic_set(&cur_trans->delayed_refs.num_entries, 0); |
20b297d6 JS |
236 | |
237 | /* | |
238 | * although the tree mod log is per file system and not per transaction, | |
239 | * the log must never go across transaction boundaries. | |
240 | */ | |
241 | smp_mb(); | |
31b1a2bd | 242 | if (!list_empty(&fs_info->tree_mod_seq_list)) |
5d163e0e | 243 | WARN(1, KERN_ERR "BTRFS: tree_mod_seq_list not empty when creating a fresh transaction\n"); |
31b1a2bd | 244 | if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log)) |
5d163e0e | 245 | WARN(1, KERN_ERR "BTRFS: tree_mod_log rb tree not empty when creating a fresh transaction\n"); |
fc36ed7e | 246 | atomic64_set(&fs_info->tree_mod_seq, 0); |
20b297d6 | 247 | |
a4abeea4 JB |
248 | spin_lock_init(&cur_trans->delayed_refs.lock); |
249 | ||
250 | INIT_LIST_HEAD(&cur_trans->pending_snapshots); | |
bbbf7243 | 251 | INIT_LIST_HEAD(&cur_trans->dev_update_list); |
9e351cc8 | 252 | INIT_LIST_HEAD(&cur_trans->switch_commits); |
ce93ec54 | 253 | INIT_LIST_HEAD(&cur_trans->dirty_bgs); |
1bbc621e | 254 | INIT_LIST_HEAD(&cur_trans->io_bgs); |
2b9dbef2 | 255 | INIT_LIST_HEAD(&cur_trans->dropped_roots); |
1bbc621e | 256 | mutex_init(&cur_trans->cache_write_mutex); |
ce93ec54 | 257 | spin_lock_init(&cur_trans->dirty_bgs_lock); |
e33e17ee | 258 | INIT_LIST_HEAD(&cur_trans->deleted_bgs); |
2b9dbef2 | 259 | spin_lock_init(&cur_trans->dropped_roots_lock); |
19ae4e81 | 260 | list_add_tail(&cur_trans->list, &fs_info->trans_list); |
c258d6e3 | 261 | extent_io_tree_init(fs_info, &cur_trans->dirty_pages, |
43eb5f29 | 262 | IO_TREE_TRANS_DIRTY_PAGES, fs_info->btree_inode); |
19ae4e81 JS |
263 | fs_info->generation++; |
264 | cur_trans->transid = fs_info->generation; | |
265 | fs_info->running_transaction = cur_trans; | |
49b25e05 | 266 | cur_trans->aborted = 0; |
19ae4e81 | 267 | spin_unlock(&fs_info->trans_lock); |
15ee9bc7 | 268 | |
79154b1b CM |
269 | return 0; |
270 | } | |
271 | ||
d352ac68 | 272 | /* |
d397712b CM |
273 | * this does all the record keeping required to make sure that a reference |
274 | * counted root is properly recorded in a given transaction. This is required | |
275 | * to make sure the old root from before we joined the transaction is deleted | |
276 | * when the transaction commits | |
d352ac68 | 277 | */ |
7585717f | 278 | static int record_root_in_trans(struct btrfs_trans_handle *trans, |
6426c7ad QW |
279 | struct btrfs_root *root, |
280 | int force) | |
6702ed49 | 281 | { |
0b246afa JM |
282 | struct btrfs_fs_info *fs_info = root->fs_info; |
283 | ||
6426c7ad QW |
284 | if ((test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
285 | root->last_trans < trans->transid) || force) { | |
0b246afa | 286 | WARN_ON(root == fs_info->extent_root); |
4d31778a | 287 | WARN_ON(!force && root->commit_root != root->node); |
5d4f98a2 | 288 | |
7585717f | 289 | /* |
27cdeb70 | 290 | * see below for IN_TRANS_SETUP usage rules |
7585717f CM |
291 | * we have the reloc mutex held now, so there |
292 | * is only one writer in this function | |
293 | */ | |
27cdeb70 | 294 | set_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); |
7585717f | 295 | |
27cdeb70 | 296 | /* make sure readers find IN_TRANS_SETUP before |
7585717f CM |
297 | * they find our root->last_trans update |
298 | */ | |
299 | smp_wmb(); | |
300 | ||
0b246afa | 301 | spin_lock(&fs_info->fs_roots_radix_lock); |
6426c7ad | 302 | if (root->last_trans == trans->transid && !force) { |
0b246afa | 303 | spin_unlock(&fs_info->fs_roots_radix_lock); |
a4abeea4 JB |
304 | return 0; |
305 | } | |
0b246afa JM |
306 | radix_tree_tag_set(&fs_info->fs_roots_radix, |
307 | (unsigned long)root->root_key.objectid, | |
308 | BTRFS_ROOT_TRANS_TAG); | |
309 | spin_unlock(&fs_info->fs_roots_radix_lock); | |
7585717f CM |
310 | root->last_trans = trans->transid; |
311 | ||
312 | /* this is pretty tricky. We don't want to | |
313 | * take the relocation lock in btrfs_record_root_in_trans | |
314 | * unless we're really doing the first setup for this root in | |
315 | * this transaction. | |
316 | * | |
317 | * Normally we'd use root->last_trans as a flag to decide | |
318 | * if we want to take the expensive mutex. | |
319 | * | |
320 | * But, we have to set root->last_trans before we | |
321 | * init the relocation root, otherwise, we trip over warnings | |
322 | * in ctree.c. The solution used here is to flag ourselves | |
27cdeb70 | 323 | * with root IN_TRANS_SETUP. When this is 1, we're still |
7585717f CM |
324 | * fixing up the reloc trees and everyone must wait. |
325 | * | |
326 | * When this is zero, they can trust root->last_trans and fly | |
327 | * through btrfs_record_root_in_trans without having to take the | |
328 | * lock. smp_wmb() makes sure that all the writes above are | |
329 | * done before we pop in the zero below | |
330 | */ | |
5d4f98a2 | 331 | btrfs_init_reloc_root(trans, root); |
c7548af6 | 332 | smp_mb__before_atomic(); |
27cdeb70 | 333 | clear_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); |
5d4f98a2 YZ |
334 | } |
335 | return 0; | |
336 | } | |
bcc63abb | 337 | |
7585717f | 338 | |
2b9dbef2 JB |
339 | void btrfs_add_dropped_root(struct btrfs_trans_handle *trans, |
340 | struct btrfs_root *root) | |
341 | { | |
0b246afa | 342 | struct btrfs_fs_info *fs_info = root->fs_info; |
2b9dbef2 JB |
343 | struct btrfs_transaction *cur_trans = trans->transaction; |
344 | ||
345 | /* Add ourselves to the transaction dropped list */ | |
346 | spin_lock(&cur_trans->dropped_roots_lock); | |
347 | list_add_tail(&root->root_list, &cur_trans->dropped_roots); | |
348 | spin_unlock(&cur_trans->dropped_roots_lock); | |
349 | ||
350 | /* Make sure we don't try to update the root at commit time */ | |
0b246afa JM |
351 | spin_lock(&fs_info->fs_roots_radix_lock); |
352 | radix_tree_tag_clear(&fs_info->fs_roots_radix, | |
2b9dbef2 JB |
353 | (unsigned long)root->root_key.objectid, |
354 | BTRFS_ROOT_TRANS_TAG); | |
0b246afa | 355 | spin_unlock(&fs_info->fs_roots_radix_lock); |
2b9dbef2 JB |
356 | } |
357 | ||
7585717f CM |
358 | int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, |
359 | struct btrfs_root *root) | |
360 | { | |
0b246afa JM |
361 | struct btrfs_fs_info *fs_info = root->fs_info; |
362 | ||
27cdeb70 | 363 | if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state)) |
7585717f CM |
364 | return 0; |
365 | ||
366 | /* | |
27cdeb70 | 367 | * see record_root_in_trans for comments about IN_TRANS_SETUP usage |
7585717f CM |
368 | * and barriers |
369 | */ | |
370 | smp_rmb(); | |
371 | if (root->last_trans == trans->transid && | |
27cdeb70 | 372 | !test_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state)) |
7585717f CM |
373 | return 0; |
374 | ||
0b246afa | 375 | mutex_lock(&fs_info->reloc_mutex); |
6426c7ad | 376 | record_root_in_trans(trans, root, 0); |
0b246afa | 377 | mutex_unlock(&fs_info->reloc_mutex); |
7585717f CM |
378 | |
379 | return 0; | |
380 | } | |
381 | ||
4a9d8bde MX |
382 | static inline int is_transaction_blocked(struct btrfs_transaction *trans) |
383 | { | |
384 | return (trans->state >= TRANS_STATE_BLOCKED && | |
501407aa JB |
385 | trans->state < TRANS_STATE_UNBLOCKED && |
386 | !trans->aborted); | |
4a9d8bde MX |
387 | } |
388 | ||
d352ac68 CM |
389 | /* wait for commit against the current transaction to become unblocked |
390 | * when this is done, it is safe to start a new transaction, but the current | |
391 | * transaction might not be fully on disk. | |
392 | */ | |
2ff7e61e | 393 | static void wait_current_trans(struct btrfs_fs_info *fs_info) |
79154b1b | 394 | { |
f9295749 | 395 | struct btrfs_transaction *cur_trans; |
79154b1b | 396 | |
0b246afa JM |
397 | spin_lock(&fs_info->trans_lock); |
398 | cur_trans = fs_info->running_transaction; | |
4a9d8bde | 399 | if (cur_trans && is_transaction_blocked(cur_trans)) { |
9b64f57d | 400 | refcount_inc(&cur_trans->use_count); |
0b246afa | 401 | spin_unlock(&fs_info->trans_lock); |
72d63ed6 | 402 | |
0b246afa | 403 | wait_event(fs_info->transaction_wait, |
501407aa JB |
404 | cur_trans->state >= TRANS_STATE_UNBLOCKED || |
405 | cur_trans->aborted); | |
724e2315 | 406 | btrfs_put_transaction(cur_trans); |
a4abeea4 | 407 | } else { |
0b246afa | 408 | spin_unlock(&fs_info->trans_lock); |
f9295749 | 409 | } |
37d1aeee CM |
410 | } |
411 | ||
2ff7e61e | 412 | static int may_wait_transaction(struct btrfs_fs_info *fs_info, int type) |
a22285a6 | 413 | { |
0b246afa | 414 | if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) |
a4abeea4 JB |
415 | return 0; |
416 | ||
92e2f7e3 | 417 | if (type == TRANS_START) |
a22285a6 | 418 | return 1; |
a4abeea4 | 419 | |
a22285a6 YZ |
420 | return 0; |
421 | } | |
422 | ||
20dd2cbf MX |
423 | static inline bool need_reserve_reloc_root(struct btrfs_root *root) |
424 | { | |
0b246afa JM |
425 | struct btrfs_fs_info *fs_info = root->fs_info; |
426 | ||
427 | if (!fs_info->reloc_ctl || | |
27cdeb70 | 428 | !test_bit(BTRFS_ROOT_REF_COWS, &root->state) || |
20dd2cbf MX |
429 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || |
430 | root->reloc_root) | |
431 | return false; | |
432 | ||
433 | return true; | |
434 | } | |
435 | ||
08e007d2 | 436 | static struct btrfs_trans_handle * |
5aed1dd8 | 437 | start_transaction(struct btrfs_root *root, unsigned int num_items, |
003d7c59 JM |
438 | unsigned int type, enum btrfs_reserve_flush_enum flush, |
439 | bool enforce_qgroups) | |
37d1aeee | 440 | { |
0b246afa | 441 | struct btrfs_fs_info *fs_info = root->fs_info; |
ba2c4d4e | 442 | struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv; |
a22285a6 YZ |
443 | struct btrfs_trans_handle *h; |
444 | struct btrfs_transaction *cur_trans; | |
b5009945 | 445 | u64 num_bytes = 0; |
c5567237 | 446 | u64 qgroup_reserved = 0; |
20dd2cbf MX |
447 | bool reloc_reserved = false; |
448 | int ret; | |
acce952b | 449 | |
46c4e71e | 450 | /* Send isn't supposed to start transactions. */ |
2755a0de | 451 | ASSERT(current->journal_info != BTRFS_SEND_TRANS_STUB); |
46c4e71e | 452 | |
0b246afa | 453 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) |
acce952b | 454 | return ERR_PTR(-EROFS); |
2a1eb461 | 455 | |
46c4e71e | 456 | if (current->journal_info) { |
0860adfd | 457 | WARN_ON(type & TRANS_EXTWRITERS); |
2a1eb461 | 458 | h = current->journal_info; |
b50fff81 DS |
459 | refcount_inc(&h->use_count); |
460 | WARN_ON(refcount_read(&h->use_count) > 2); | |
2a1eb461 JB |
461 | h->orig_rsv = h->block_rsv; |
462 | h->block_rsv = NULL; | |
463 | goto got_it; | |
464 | } | |
b5009945 JB |
465 | |
466 | /* | |
467 | * Do the reservation before we join the transaction so we can do all | |
468 | * the appropriate flushing if need be. | |
469 | */ | |
003d7c59 | 470 | if (num_items && root != fs_info->chunk_root) { |
ba2c4d4e JB |
471 | struct btrfs_block_rsv *rsv = &fs_info->trans_block_rsv; |
472 | u64 delayed_refs_bytes = 0; | |
473 | ||
0b246afa | 474 | qgroup_reserved = num_items * fs_info->nodesize; |
733e03a0 QW |
475 | ret = btrfs_qgroup_reserve_meta_pertrans(root, qgroup_reserved, |
476 | enforce_qgroups); | |
7174109c QW |
477 | if (ret) |
478 | return ERR_PTR(ret); | |
c5567237 | 479 | |
ba2c4d4e JB |
480 | /* |
481 | * We want to reserve all the bytes we may need all at once, so | |
482 | * we only do 1 enospc flushing cycle per transaction start. We | |
483 | * accomplish this by simply assuming we'll do 2 x num_items | |
484 | * worth of delayed refs updates in this trans handle, and | |
485 | * refill that amount for whatever is missing in the reserve. | |
486 | */ | |
0b246afa | 487 | num_bytes = btrfs_calc_trans_metadata_size(fs_info, num_items); |
ba2c4d4e JB |
488 | if (delayed_refs_rsv->full == 0) { |
489 | delayed_refs_bytes = num_bytes; | |
490 | num_bytes <<= 1; | |
491 | } | |
492 | ||
20dd2cbf MX |
493 | /* |
494 | * Do the reservation for the relocation root creation | |
495 | */ | |
ee39b432 | 496 | if (need_reserve_reloc_root(root)) { |
0b246afa | 497 | num_bytes += fs_info->nodesize; |
20dd2cbf MX |
498 | reloc_reserved = true; |
499 | } | |
500 | ||
ba2c4d4e JB |
501 | ret = btrfs_block_rsv_add(root, rsv, num_bytes, flush); |
502 | if (ret) | |
503 | goto reserve_fail; | |
504 | if (delayed_refs_bytes) { | |
505 | btrfs_migrate_to_delayed_refs_rsv(fs_info, rsv, | |
506 | delayed_refs_bytes); | |
507 | num_bytes -= delayed_refs_bytes; | |
508 | } | |
509 | } else if (num_items == 0 && flush == BTRFS_RESERVE_FLUSH_ALL && | |
510 | !delayed_refs_rsv->full) { | |
511 | /* | |
512 | * Some people call with btrfs_start_transaction(root, 0) | |
513 | * because they can be throttled, but have some other mechanism | |
514 | * for reserving space. We still want these guys to refill the | |
515 | * delayed block_rsv so just add 1 items worth of reservation | |
516 | * here. | |
517 | */ | |
518 | ret = btrfs_delayed_refs_rsv_refill(fs_info, flush); | |
b5009945 | 519 | if (ret) |
843fcf35 | 520 | goto reserve_fail; |
b5009945 | 521 | } |
a22285a6 | 522 | again: |
f2f767e7 | 523 | h = kmem_cache_zalloc(btrfs_trans_handle_cachep, GFP_NOFS); |
843fcf35 MX |
524 | if (!h) { |
525 | ret = -ENOMEM; | |
526 | goto alloc_fail; | |
527 | } | |
37d1aeee | 528 | |
98114659 JB |
529 | /* |
530 | * If we are JOIN_NOLOCK we're already committing a transaction and | |
531 | * waiting on this guy, so we don't need to do the sb_start_intwrite | |
532 | * because we're already holding a ref. We need this because we could | |
533 | * have raced in and did an fsync() on a file which can kick a commit | |
534 | * and then we deadlock with somebody doing a freeze. | |
354aa0fb MX |
535 | * |
536 | * If we are ATTACH, it means we just want to catch the current | |
537 | * transaction and commit it, so we needn't do sb_start_intwrite(). | |
98114659 | 538 | */ |
0860adfd | 539 | if (type & __TRANS_FREEZABLE) |
0b246afa | 540 | sb_start_intwrite(fs_info->sb); |
b2b5ef5c | 541 | |
2ff7e61e JM |
542 | if (may_wait_transaction(fs_info, type)) |
543 | wait_current_trans(fs_info); | |
a22285a6 | 544 | |
a4abeea4 | 545 | do { |
2ff7e61e | 546 | ret = join_transaction(fs_info, type); |
178260b2 | 547 | if (ret == -EBUSY) { |
2ff7e61e | 548 | wait_current_trans(fs_info); |
a6d155d2 FM |
549 | if (unlikely(type == TRANS_ATTACH || |
550 | type == TRANS_JOIN_NOSTART)) | |
178260b2 MX |
551 | ret = -ENOENT; |
552 | } | |
a4abeea4 JB |
553 | } while (ret == -EBUSY); |
554 | ||
a43f7f82 | 555 | if (ret < 0) |
843fcf35 | 556 | goto join_fail; |
0f7d52f4 | 557 | |
0b246afa | 558 | cur_trans = fs_info->running_transaction; |
a22285a6 YZ |
559 | |
560 | h->transid = cur_trans->transid; | |
561 | h->transaction = cur_trans; | |
d13603ef | 562 | h->root = root; |
b50fff81 | 563 | refcount_set(&h->use_count, 1); |
64b63580 | 564 | h->fs_info = root->fs_info; |
7174109c | 565 | |
a698d075 | 566 | h->type = type; |
d9a0540a | 567 | h->can_flush_pending_bgs = true; |
ea658bad | 568 | INIT_LIST_HEAD(&h->new_bgs); |
b7ec40d7 | 569 | |
a22285a6 | 570 | smp_mb(); |
4a9d8bde | 571 | if (cur_trans->state >= TRANS_STATE_BLOCKED && |
2ff7e61e | 572 | may_wait_transaction(fs_info, type)) { |
abdd2e80 | 573 | current->journal_info = h; |
3a45bb20 | 574 | btrfs_commit_transaction(h); |
a22285a6 YZ |
575 | goto again; |
576 | } | |
577 | ||
b5009945 | 578 | if (num_bytes) { |
0b246afa | 579 | trace_btrfs_space_reservation(fs_info, "transaction", |
2bcc0328 | 580 | h->transid, num_bytes, 1); |
0b246afa | 581 | h->block_rsv = &fs_info->trans_block_rsv; |
b5009945 | 582 | h->bytes_reserved = num_bytes; |
20dd2cbf | 583 | h->reloc_reserved = reloc_reserved; |
a22285a6 | 584 | } |
9ed74f2d | 585 | |
2a1eb461 | 586 | got_it: |
a4abeea4 | 587 | btrfs_record_root_in_trans(h, root); |
a22285a6 | 588 | |
bcf3a3e7 | 589 | if (!current->journal_info) |
a22285a6 | 590 | current->journal_info = h; |
79154b1b | 591 | return h; |
843fcf35 MX |
592 | |
593 | join_fail: | |
0860adfd | 594 | if (type & __TRANS_FREEZABLE) |
0b246afa | 595 | sb_end_intwrite(fs_info->sb); |
843fcf35 MX |
596 | kmem_cache_free(btrfs_trans_handle_cachep, h); |
597 | alloc_fail: | |
598 | if (num_bytes) | |
2ff7e61e | 599 | btrfs_block_rsv_release(fs_info, &fs_info->trans_block_rsv, |
843fcf35 MX |
600 | num_bytes); |
601 | reserve_fail: | |
733e03a0 | 602 | btrfs_qgroup_free_meta_pertrans(root, qgroup_reserved); |
843fcf35 | 603 | return ERR_PTR(ret); |
79154b1b CM |
604 | } |
605 | ||
f9295749 | 606 | struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, |
5aed1dd8 | 607 | unsigned int num_items) |
f9295749 | 608 | { |
08e007d2 | 609 | return start_transaction(root, num_items, TRANS_START, |
003d7c59 | 610 | BTRFS_RESERVE_FLUSH_ALL, true); |
f9295749 | 611 | } |
003d7c59 | 612 | |
8eab77ff FM |
613 | struct btrfs_trans_handle *btrfs_start_transaction_fallback_global_rsv( |
614 | struct btrfs_root *root, | |
615 | unsigned int num_items, | |
616 | int min_factor) | |
617 | { | |
0b246afa | 618 | struct btrfs_fs_info *fs_info = root->fs_info; |
8eab77ff FM |
619 | struct btrfs_trans_handle *trans; |
620 | u64 num_bytes; | |
621 | int ret; | |
622 | ||
003d7c59 JM |
623 | /* |
624 | * We have two callers: unlink and block group removal. The | |
625 | * former should succeed even if we will temporarily exceed | |
626 | * quota and the latter operates on the extent root so | |
627 | * qgroup enforcement is ignored anyway. | |
628 | */ | |
629 | trans = start_transaction(root, num_items, TRANS_START, | |
630 | BTRFS_RESERVE_FLUSH_ALL, false); | |
8eab77ff FM |
631 | if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC) |
632 | return trans; | |
633 | ||
634 | trans = btrfs_start_transaction(root, 0); | |
635 | if (IS_ERR(trans)) | |
636 | return trans; | |
637 | ||
0b246afa JM |
638 | num_bytes = btrfs_calc_trans_metadata_size(fs_info, num_items); |
639 | ret = btrfs_cond_migrate_bytes(fs_info, &fs_info->trans_block_rsv, | |
640 | num_bytes, min_factor); | |
8eab77ff | 641 | if (ret) { |
3a45bb20 | 642 | btrfs_end_transaction(trans); |
8eab77ff FM |
643 | return ERR_PTR(ret); |
644 | } | |
645 | ||
0b246afa | 646 | trans->block_rsv = &fs_info->trans_block_rsv; |
8eab77ff | 647 | trans->bytes_reserved = num_bytes; |
0b246afa | 648 | trace_btrfs_space_reservation(fs_info, "transaction", |
88d3a5aa | 649 | trans->transid, num_bytes, 1); |
8eab77ff FM |
650 | |
651 | return trans; | |
652 | } | |
8407aa46 | 653 | |
7a7eaa40 | 654 | struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root) |
f9295749 | 655 | { |
003d7c59 JM |
656 | return start_transaction(root, 0, TRANS_JOIN, BTRFS_RESERVE_NO_FLUSH, |
657 | true); | |
f9295749 CM |
658 | } |
659 | ||
7a7eaa40 | 660 | struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root) |
0af3d00b | 661 | { |
575a75d6 | 662 | return start_transaction(root, 0, TRANS_JOIN_NOLOCK, |
003d7c59 | 663 | BTRFS_RESERVE_NO_FLUSH, true); |
0af3d00b JB |
664 | } |
665 | ||
a6d155d2 FM |
666 | /* |
667 | * Similar to regular join but it never starts a transaction when none is | |
668 | * running or after waiting for the current one to finish. | |
669 | */ | |
670 | struct btrfs_trans_handle *btrfs_join_transaction_nostart(struct btrfs_root *root) | |
671 | { | |
672 | return start_transaction(root, 0, TRANS_JOIN_NOSTART, | |
673 | BTRFS_RESERVE_NO_FLUSH, true); | |
674 | } | |
675 | ||
d4edf39b MX |
676 | /* |
677 | * btrfs_attach_transaction() - catch the running transaction | |
678 | * | |
679 | * It is used when we want to commit the current the transaction, but | |
680 | * don't want to start a new one. | |
681 | * | |
682 | * Note: If this function return -ENOENT, it just means there is no | |
683 | * running transaction. But it is possible that the inactive transaction | |
684 | * is still in the memory, not fully on disk. If you hope there is no | |
685 | * inactive transaction in the fs when -ENOENT is returned, you should | |
686 | * invoke | |
687 | * btrfs_attach_transaction_barrier() | |
688 | */ | |
354aa0fb | 689 | struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root) |
60376ce4 | 690 | { |
575a75d6 | 691 | return start_transaction(root, 0, TRANS_ATTACH, |
003d7c59 | 692 | BTRFS_RESERVE_NO_FLUSH, true); |
60376ce4 JB |
693 | } |
694 | ||
d4edf39b | 695 | /* |
90b6d283 | 696 | * btrfs_attach_transaction_barrier() - catch the running transaction |
d4edf39b | 697 | * |
52042d8e | 698 | * It is similar to the above function, the difference is this one |
d4edf39b MX |
699 | * will wait for all the inactive transactions until they fully |
700 | * complete. | |
701 | */ | |
702 | struct btrfs_trans_handle * | |
703 | btrfs_attach_transaction_barrier(struct btrfs_root *root) | |
704 | { | |
705 | struct btrfs_trans_handle *trans; | |
706 | ||
575a75d6 | 707 | trans = start_transaction(root, 0, TRANS_ATTACH, |
003d7c59 | 708 | BTRFS_RESERVE_NO_FLUSH, true); |
8d9e220c | 709 | if (trans == ERR_PTR(-ENOENT)) |
2ff7e61e | 710 | btrfs_wait_for_commit(root->fs_info, 0); |
d4edf39b MX |
711 | |
712 | return trans; | |
713 | } | |
714 | ||
d352ac68 | 715 | /* wait for a transaction commit to be fully complete */ |
2ff7e61e | 716 | static noinline void wait_for_commit(struct btrfs_transaction *commit) |
89ce8a63 | 717 | { |
4a9d8bde | 718 | wait_event(commit->commit_wait, commit->state == TRANS_STATE_COMPLETED); |
89ce8a63 CM |
719 | } |
720 | ||
2ff7e61e | 721 | int btrfs_wait_for_commit(struct btrfs_fs_info *fs_info, u64 transid) |
46204592 SW |
722 | { |
723 | struct btrfs_transaction *cur_trans = NULL, *t; | |
8cd2807f | 724 | int ret = 0; |
46204592 | 725 | |
46204592 | 726 | if (transid) { |
0b246afa | 727 | if (transid <= fs_info->last_trans_committed) |
a4abeea4 | 728 | goto out; |
46204592 SW |
729 | |
730 | /* find specified transaction */ | |
0b246afa JM |
731 | spin_lock(&fs_info->trans_lock); |
732 | list_for_each_entry(t, &fs_info->trans_list, list) { | |
46204592 SW |
733 | if (t->transid == transid) { |
734 | cur_trans = t; | |
9b64f57d | 735 | refcount_inc(&cur_trans->use_count); |
8cd2807f | 736 | ret = 0; |
46204592 SW |
737 | break; |
738 | } | |
8cd2807f MX |
739 | if (t->transid > transid) { |
740 | ret = 0; | |
46204592 | 741 | break; |
8cd2807f | 742 | } |
46204592 | 743 | } |
0b246afa | 744 | spin_unlock(&fs_info->trans_lock); |
42383020 SW |
745 | |
746 | /* | |
747 | * The specified transaction doesn't exist, or we | |
748 | * raced with btrfs_commit_transaction | |
749 | */ | |
750 | if (!cur_trans) { | |
0b246afa | 751 | if (transid > fs_info->last_trans_committed) |
42383020 | 752 | ret = -EINVAL; |
8cd2807f | 753 | goto out; |
42383020 | 754 | } |
46204592 SW |
755 | } else { |
756 | /* find newest transaction that is committing | committed */ | |
0b246afa JM |
757 | spin_lock(&fs_info->trans_lock); |
758 | list_for_each_entry_reverse(t, &fs_info->trans_list, | |
46204592 | 759 | list) { |
4a9d8bde MX |
760 | if (t->state >= TRANS_STATE_COMMIT_START) { |
761 | if (t->state == TRANS_STATE_COMPLETED) | |
3473f3c0 | 762 | break; |
46204592 | 763 | cur_trans = t; |
9b64f57d | 764 | refcount_inc(&cur_trans->use_count); |
46204592 SW |
765 | break; |
766 | } | |
767 | } | |
0b246afa | 768 | spin_unlock(&fs_info->trans_lock); |
46204592 | 769 | if (!cur_trans) |
a4abeea4 | 770 | goto out; /* nothing committing|committed */ |
46204592 SW |
771 | } |
772 | ||
2ff7e61e | 773 | wait_for_commit(cur_trans); |
724e2315 | 774 | btrfs_put_transaction(cur_trans); |
a4abeea4 | 775 | out: |
46204592 SW |
776 | return ret; |
777 | } | |
778 | ||
2ff7e61e | 779 | void btrfs_throttle(struct btrfs_fs_info *fs_info) |
37d1aeee | 780 | { |
92e2f7e3 | 781 | wait_current_trans(fs_info); |
37d1aeee CM |
782 | } |
783 | ||
2ff7e61e | 784 | static int should_end_transaction(struct btrfs_trans_handle *trans) |
8929ecfa | 785 | { |
2ff7e61e | 786 | struct btrfs_fs_info *fs_info = trans->fs_info; |
0b246afa | 787 | |
64403612 | 788 | if (btrfs_check_space_for_delayed_refs(fs_info)) |
1be41b78 | 789 | return 1; |
36ba022a | 790 | |
2ff7e61e | 791 | return !!btrfs_block_rsv_check(&fs_info->global_block_rsv, 5); |
8929ecfa YZ |
792 | } |
793 | ||
3a45bb20 | 794 | int btrfs_should_end_transaction(struct btrfs_trans_handle *trans) |
8929ecfa YZ |
795 | { |
796 | struct btrfs_transaction *cur_trans = trans->transaction; | |
8929ecfa | 797 | |
a4abeea4 | 798 | smp_mb(); |
4a9d8bde MX |
799 | if (cur_trans->state >= TRANS_STATE_BLOCKED || |
800 | cur_trans->delayed_refs.flushing) | |
8929ecfa YZ |
801 | return 1; |
802 | ||
2ff7e61e | 803 | return should_end_transaction(trans); |
8929ecfa YZ |
804 | } |
805 | ||
dc60c525 NB |
806 | static void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans) |
807 | ||
0e34693f | 808 | { |
dc60c525 NB |
809 | struct btrfs_fs_info *fs_info = trans->fs_info; |
810 | ||
0e34693f NB |
811 | if (!trans->block_rsv) { |
812 | ASSERT(!trans->bytes_reserved); | |
813 | return; | |
814 | } | |
815 | ||
816 | if (!trans->bytes_reserved) | |
817 | return; | |
818 | ||
819 | ASSERT(trans->block_rsv == &fs_info->trans_block_rsv); | |
820 | trace_btrfs_space_reservation(fs_info, "transaction", | |
821 | trans->transid, trans->bytes_reserved, 0); | |
822 | btrfs_block_rsv_release(fs_info, trans->block_rsv, | |
823 | trans->bytes_reserved); | |
824 | trans->bytes_reserved = 0; | |
825 | } | |
826 | ||
89ce8a63 | 827 | static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, |
3a45bb20 | 828 | int throttle) |
79154b1b | 829 | { |
3a45bb20 | 830 | struct btrfs_fs_info *info = trans->fs_info; |
8929ecfa | 831 | struct btrfs_transaction *cur_trans = trans->transaction; |
a698d075 | 832 | int lock = (trans->type != TRANS_JOIN_NOLOCK); |
4edc2ca3 | 833 | int err = 0; |
c3e69d58 | 834 | |
b50fff81 DS |
835 | if (refcount_read(&trans->use_count) > 1) { |
836 | refcount_dec(&trans->use_count); | |
2a1eb461 JB |
837 | trans->block_rsv = trans->orig_rsv; |
838 | return 0; | |
839 | } | |
840 | ||
dc60c525 | 841 | btrfs_trans_release_metadata(trans); |
4c13d758 | 842 | trans->block_rsv = NULL; |
c5567237 | 843 | |
119e80df | 844 | btrfs_create_pending_block_groups(trans); |
ea658bad | 845 | |
4fbcdf66 FM |
846 | btrfs_trans_release_chunk_metadata(trans); |
847 | ||
20c7bcec | 848 | if (lock && READ_ONCE(cur_trans->state) == TRANS_STATE_BLOCKED) { |
3bbb24b2 | 849 | if (throttle) |
3a45bb20 | 850 | return btrfs_commit_transaction(trans); |
3bbb24b2 | 851 | else |
8929ecfa YZ |
852 | wake_up_process(info->transaction_kthread); |
853 | } | |
854 | ||
0860adfd | 855 | if (trans->type & __TRANS_FREEZABLE) |
0b246afa | 856 | sb_end_intwrite(info->sb); |
6df7881a | 857 | |
8929ecfa | 858 | WARN_ON(cur_trans != info->running_transaction); |
13c5a93e JB |
859 | WARN_ON(atomic_read(&cur_trans->num_writers) < 1); |
860 | atomic_dec(&cur_trans->num_writers); | |
0860adfd | 861 | extwriter_counter_dec(cur_trans, trans->type); |
89ce8a63 | 862 | |
093258e6 | 863 | cond_wake_up(&cur_trans->writer_wait); |
724e2315 | 864 | btrfs_put_transaction(cur_trans); |
9ed74f2d JB |
865 | |
866 | if (current->journal_info == trans) | |
867 | current->journal_info = NULL; | |
ab78c84d | 868 | |
24bbcf04 | 869 | if (throttle) |
2ff7e61e | 870 | btrfs_run_delayed_iputs(info); |
24bbcf04 | 871 | |
49b25e05 | 872 | if (trans->aborted || |
0b246afa | 873 | test_bit(BTRFS_FS_STATE_ERROR, &info->fs_state)) { |
4e121c06 | 874 | wake_up_process(info->transaction_kthread); |
4edc2ca3 | 875 | err = -EIO; |
4e121c06 | 876 | } |
49b25e05 | 877 | |
4edc2ca3 DJ |
878 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
879 | return err; | |
79154b1b CM |
880 | } |
881 | ||
3a45bb20 | 882 | int btrfs_end_transaction(struct btrfs_trans_handle *trans) |
89ce8a63 | 883 | { |
3a45bb20 | 884 | return __btrfs_end_transaction(trans, 0); |
89ce8a63 CM |
885 | } |
886 | ||
3a45bb20 | 887 | int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans) |
89ce8a63 | 888 | { |
3a45bb20 | 889 | return __btrfs_end_transaction(trans, 1); |
16cdcec7 MX |
890 | } |
891 | ||
d352ac68 CM |
892 | /* |
893 | * when btree blocks are allocated, they have some corresponding bits set for | |
894 | * them in one of two extent_io trees. This is used to make sure all of | |
690587d1 | 895 | * those extents are sent to disk but does not wait on them |
d352ac68 | 896 | */ |
2ff7e61e | 897 | int btrfs_write_marked_extents(struct btrfs_fs_info *fs_info, |
8cef4e16 | 898 | struct extent_io_tree *dirty_pages, int mark) |
79154b1b | 899 | { |
777e6bd7 | 900 | int err = 0; |
7c4452b9 | 901 | int werr = 0; |
0b246afa | 902 | struct address_space *mapping = fs_info->btree_inode->i_mapping; |
e6138876 | 903 | struct extent_state *cached_state = NULL; |
777e6bd7 | 904 | u64 start = 0; |
5f39d397 | 905 | u64 end; |
7c4452b9 | 906 | |
6300463b | 907 | atomic_inc(&BTRFS_I(fs_info->btree_inode)->sync_writers); |
1728366e | 908 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
e6138876 | 909 | mark, &cached_state)) { |
663dfbb0 FM |
910 | bool wait_writeback = false; |
911 | ||
912 | err = convert_extent_bit(dirty_pages, start, end, | |
913 | EXTENT_NEED_WAIT, | |
210aa277 | 914 | mark, &cached_state); |
663dfbb0 FM |
915 | /* |
916 | * convert_extent_bit can return -ENOMEM, which is most of the | |
917 | * time a temporary error. So when it happens, ignore the error | |
918 | * and wait for writeback of this range to finish - because we | |
919 | * failed to set the bit EXTENT_NEED_WAIT for the range, a call | |
bf89d38f JM |
920 | * to __btrfs_wait_marked_extents() would not know that |
921 | * writeback for this range started and therefore wouldn't | |
922 | * wait for it to finish - we don't want to commit a | |
923 | * superblock that points to btree nodes/leafs for which | |
924 | * writeback hasn't finished yet (and without errors). | |
663dfbb0 | 925 | * We cleanup any entries left in the io tree when committing |
41e7acd3 | 926 | * the transaction (through extent_io_tree_release()). |
663dfbb0 FM |
927 | */ |
928 | if (err == -ENOMEM) { | |
929 | err = 0; | |
930 | wait_writeback = true; | |
931 | } | |
932 | if (!err) | |
933 | err = filemap_fdatawrite_range(mapping, start, end); | |
1728366e JB |
934 | if (err) |
935 | werr = err; | |
663dfbb0 FM |
936 | else if (wait_writeback) |
937 | werr = filemap_fdatawait_range(mapping, start, end); | |
e38e2ed7 | 938 | free_extent_state(cached_state); |
663dfbb0 | 939 | cached_state = NULL; |
1728366e JB |
940 | cond_resched(); |
941 | start = end + 1; | |
7c4452b9 | 942 | } |
6300463b | 943 | atomic_dec(&BTRFS_I(fs_info->btree_inode)->sync_writers); |
690587d1 CM |
944 | return werr; |
945 | } | |
946 | ||
947 | /* | |
948 | * when btree blocks are allocated, they have some corresponding bits set for | |
949 | * them in one of two extent_io trees. This is used to make sure all of | |
950 | * those extents are on disk for transaction or log commit. We wait | |
951 | * on all the pages and clear them from the dirty pages state tree | |
952 | */ | |
bf89d38f JM |
953 | static int __btrfs_wait_marked_extents(struct btrfs_fs_info *fs_info, |
954 | struct extent_io_tree *dirty_pages) | |
690587d1 | 955 | { |
690587d1 CM |
956 | int err = 0; |
957 | int werr = 0; | |
0b246afa | 958 | struct address_space *mapping = fs_info->btree_inode->i_mapping; |
e6138876 | 959 | struct extent_state *cached_state = NULL; |
690587d1 CM |
960 | u64 start = 0; |
961 | u64 end; | |
777e6bd7 | 962 | |
1728366e | 963 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
e6138876 | 964 | EXTENT_NEED_WAIT, &cached_state)) { |
663dfbb0 FM |
965 | /* |
966 | * Ignore -ENOMEM errors returned by clear_extent_bit(). | |
967 | * When committing the transaction, we'll remove any entries | |
968 | * left in the io tree. For a log commit, we don't remove them | |
969 | * after committing the log because the tree can be accessed | |
970 | * concurrently - we do it only at transaction commit time when | |
41e7acd3 | 971 | * it's safe to do it (through extent_io_tree_release()). |
663dfbb0 FM |
972 | */ |
973 | err = clear_extent_bit(dirty_pages, start, end, | |
ae0f1625 | 974 | EXTENT_NEED_WAIT, 0, 0, &cached_state); |
663dfbb0 FM |
975 | if (err == -ENOMEM) |
976 | err = 0; | |
977 | if (!err) | |
978 | err = filemap_fdatawait_range(mapping, start, end); | |
1728366e JB |
979 | if (err) |
980 | werr = err; | |
e38e2ed7 FM |
981 | free_extent_state(cached_state); |
982 | cached_state = NULL; | |
1728366e JB |
983 | cond_resched(); |
984 | start = end + 1; | |
777e6bd7 | 985 | } |
7c4452b9 CM |
986 | if (err) |
987 | werr = err; | |
bf89d38f JM |
988 | return werr; |
989 | } | |
656f30db | 990 | |
bf89d38f JM |
991 | int btrfs_wait_extents(struct btrfs_fs_info *fs_info, |
992 | struct extent_io_tree *dirty_pages) | |
993 | { | |
994 | bool errors = false; | |
995 | int err; | |
656f30db | 996 | |
bf89d38f JM |
997 | err = __btrfs_wait_marked_extents(fs_info, dirty_pages); |
998 | if (test_and_clear_bit(BTRFS_FS_BTREE_ERR, &fs_info->flags)) | |
999 | errors = true; | |
1000 | ||
1001 | if (errors && !err) | |
1002 | err = -EIO; | |
1003 | return err; | |
1004 | } | |
656f30db | 1005 | |
bf89d38f JM |
1006 | int btrfs_wait_tree_log_extents(struct btrfs_root *log_root, int mark) |
1007 | { | |
1008 | struct btrfs_fs_info *fs_info = log_root->fs_info; | |
1009 | struct extent_io_tree *dirty_pages = &log_root->dirty_log_pages; | |
1010 | bool errors = false; | |
1011 | int err; | |
656f30db | 1012 | |
bf89d38f JM |
1013 | ASSERT(log_root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID); |
1014 | ||
1015 | err = __btrfs_wait_marked_extents(fs_info, dirty_pages); | |
1016 | if ((mark & EXTENT_DIRTY) && | |
1017 | test_and_clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags)) | |
1018 | errors = true; | |
1019 | ||
1020 | if ((mark & EXTENT_NEW) && | |
1021 | test_and_clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags)) | |
1022 | errors = true; | |
1023 | ||
1024 | if (errors && !err) | |
1025 | err = -EIO; | |
1026 | return err; | |
79154b1b CM |
1027 | } |
1028 | ||
690587d1 | 1029 | /* |
c9b577c0 NB |
1030 | * When btree blocks are allocated the corresponding extents are marked dirty. |
1031 | * This function ensures such extents are persisted on disk for transaction or | |
1032 | * log commit. | |
1033 | * | |
1034 | * @trans: transaction whose dirty pages we'd like to write | |
690587d1 | 1035 | */ |
70458a58 | 1036 | static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans) |
690587d1 CM |
1037 | { |
1038 | int ret; | |
1039 | int ret2; | |
c9b577c0 | 1040 | struct extent_io_tree *dirty_pages = &trans->transaction->dirty_pages; |
70458a58 | 1041 | struct btrfs_fs_info *fs_info = trans->fs_info; |
c6adc9cc | 1042 | struct blk_plug plug; |
690587d1 | 1043 | |
c6adc9cc | 1044 | blk_start_plug(&plug); |
c9b577c0 | 1045 | ret = btrfs_write_marked_extents(fs_info, dirty_pages, EXTENT_DIRTY); |
c6adc9cc | 1046 | blk_finish_plug(&plug); |
bf89d38f | 1047 | ret2 = btrfs_wait_extents(fs_info, dirty_pages); |
bf0da8c1 | 1048 | |
41e7acd3 | 1049 | extent_io_tree_release(&trans->transaction->dirty_pages); |
c9b577c0 | 1050 | |
bf0da8c1 CM |
1051 | if (ret) |
1052 | return ret; | |
c9b577c0 | 1053 | else if (ret2) |
bf0da8c1 | 1054 | return ret2; |
c9b577c0 NB |
1055 | else |
1056 | return 0; | |
d0c803c4 CM |
1057 | } |
1058 | ||
d352ac68 CM |
1059 | /* |
1060 | * this is used to update the root pointer in the tree of tree roots. | |
1061 | * | |
1062 | * But, in the case of the extent allocation tree, updating the root | |
1063 | * pointer may allocate blocks which may change the root of the extent | |
1064 | * allocation tree. | |
1065 | * | |
1066 | * So, this loops and repeats and makes sure the cowonly root didn't | |
1067 | * change while the root pointer was being updated in the metadata. | |
1068 | */ | |
0b86a832 CM |
1069 | static int update_cowonly_root(struct btrfs_trans_handle *trans, |
1070 | struct btrfs_root *root) | |
79154b1b CM |
1071 | { |
1072 | int ret; | |
0b86a832 | 1073 | u64 old_root_bytenr; |
86b9f2ec | 1074 | u64 old_root_used; |
0b246afa JM |
1075 | struct btrfs_fs_info *fs_info = root->fs_info; |
1076 | struct btrfs_root *tree_root = fs_info->tree_root; | |
79154b1b | 1077 | |
86b9f2ec | 1078 | old_root_used = btrfs_root_used(&root->root_item); |
56bec294 | 1079 | |
d397712b | 1080 | while (1) { |
0b86a832 | 1081 | old_root_bytenr = btrfs_root_bytenr(&root->root_item); |
86b9f2ec | 1082 | if (old_root_bytenr == root->node->start && |
ea526d18 | 1083 | old_root_used == btrfs_root_used(&root->root_item)) |
79154b1b | 1084 | break; |
87ef2bb4 | 1085 | |
5d4f98a2 | 1086 | btrfs_set_root_node(&root->root_item, root->node); |
79154b1b | 1087 | ret = btrfs_update_root(trans, tree_root, |
0b86a832 CM |
1088 | &root->root_key, |
1089 | &root->root_item); | |
49b25e05 JM |
1090 | if (ret) |
1091 | return ret; | |
56bec294 | 1092 | |
86b9f2ec | 1093 | old_root_used = btrfs_root_used(&root->root_item); |
0b86a832 | 1094 | } |
276e680d | 1095 | |
0b86a832 CM |
1096 | return 0; |
1097 | } | |
1098 | ||
d352ac68 CM |
1099 | /* |
1100 | * update all the cowonly tree roots on disk | |
49b25e05 JM |
1101 | * |
1102 | * The error handling in this function may not be obvious. Any of the | |
1103 | * failures will cause the file system to go offline. We still need | |
1104 | * to clean up the delayed refs. | |
d352ac68 | 1105 | */ |
9386d8bc | 1106 | static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans) |
0b86a832 | 1107 | { |
9386d8bc | 1108 | struct btrfs_fs_info *fs_info = trans->fs_info; |
ea526d18 | 1109 | struct list_head *dirty_bgs = &trans->transaction->dirty_bgs; |
1bbc621e | 1110 | struct list_head *io_bgs = &trans->transaction->io_bgs; |
0b86a832 | 1111 | struct list_head *next; |
84234f3a | 1112 | struct extent_buffer *eb; |
56bec294 | 1113 | int ret; |
84234f3a YZ |
1114 | |
1115 | eb = btrfs_lock_root_node(fs_info->tree_root); | |
49b25e05 JM |
1116 | ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, |
1117 | 0, &eb); | |
84234f3a YZ |
1118 | btrfs_tree_unlock(eb); |
1119 | free_extent_buffer(eb); | |
0b86a832 | 1120 | |
49b25e05 JM |
1121 | if (ret) |
1122 | return ret; | |
1123 | ||
c79a70b1 | 1124 | ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); |
49b25e05 JM |
1125 | if (ret) |
1126 | return ret; | |
87ef2bb4 | 1127 | |
196c9d8d | 1128 | ret = btrfs_run_dev_stats(trans); |
c16ce190 JB |
1129 | if (ret) |
1130 | return ret; | |
2b584c68 | 1131 | ret = btrfs_run_dev_replace(trans); |
c16ce190 JB |
1132 | if (ret) |
1133 | return ret; | |
280f8bd2 | 1134 | ret = btrfs_run_qgroups(trans); |
c16ce190 JB |
1135 | if (ret) |
1136 | return ret; | |
546adb0d | 1137 | |
bbebb3e0 | 1138 | ret = btrfs_setup_space_cache(trans); |
dcdf7f6d JB |
1139 | if (ret) |
1140 | return ret; | |
1141 | ||
546adb0d | 1142 | /* run_qgroups might have added some more refs */ |
c79a70b1 | 1143 | ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); |
c16ce190 JB |
1144 | if (ret) |
1145 | return ret; | |
ea526d18 | 1146 | again: |
d397712b | 1147 | while (!list_empty(&fs_info->dirty_cowonly_roots)) { |
2ff7e61e | 1148 | struct btrfs_root *root; |
0b86a832 CM |
1149 | next = fs_info->dirty_cowonly_roots.next; |
1150 | list_del_init(next); | |
1151 | root = list_entry(next, struct btrfs_root, dirty_list); | |
e7070be1 | 1152 | clear_bit(BTRFS_ROOT_DIRTY, &root->state); |
87ef2bb4 | 1153 | |
9e351cc8 JB |
1154 | if (root != fs_info->extent_root) |
1155 | list_add_tail(&root->dirty_list, | |
1156 | &trans->transaction->switch_commits); | |
49b25e05 JM |
1157 | ret = update_cowonly_root(trans, root); |
1158 | if (ret) | |
1159 | return ret; | |
c79a70b1 | 1160 | ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); |
ea526d18 JB |
1161 | if (ret) |
1162 | return ret; | |
79154b1b | 1163 | } |
276e680d | 1164 | |
1bbc621e | 1165 | while (!list_empty(dirty_bgs) || !list_empty(io_bgs)) { |
5742d15f | 1166 | ret = btrfs_write_dirty_block_groups(trans); |
ea526d18 JB |
1167 | if (ret) |
1168 | return ret; | |
c79a70b1 | 1169 | ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); |
ea526d18 JB |
1170 | if (ret) |
1171 | return ret; | |
1172 | } | |
1173 | ||
1174 | if (!list_empty(&fs_info->dirty_cowonly_roots)) | |
1175 | goto again; | |
1176 | ||
9e351cc8 JB |
1177 | list_add_tail(&fs_info->extent_root->dirty_list, |
1178 | &trans->transaction->switch_commits); | |
9f6cbcbb DS |
1179 | |
1180 | /* Update dev-replace pointer once everything is committed */ | |
1181 | fs_info->dev_replace.committed_cursor_left = | |
1182 | fs_info->dev_replace.cursor_left_last_write_of_item; | |
8dabb742 | 1183 | |
79154b1b CM |
1184 | return 0; |
1185 | } | |
1186 | ||
d352ac68 CM |
1187 | /* |
1188 | * dead roots are old snapshots that need to be deleted. This allocates | |
1189 | * a dirty root struct and adds it into the list of dead roots that need to | |
1190 | * be deleted | |
1191 | */ | |
cfad392b | 1192 | void btrfs_add_dead_root(struct btrfs_root *root) |
5eda7b5e | 1193 | { |
0b246afa JM |
1194 | struct btrfs_fs_info *fs_info = root->fs_info; |
1195 | ||
1196 | spin_lock(&fs_info->trans_lock); | |
cfad392b | 1197 | if (list_empty(&root->root_list)) |
0b246afa JM |
1198 | list_add_tail(&root->root_list, &fs_info->dead_roots); |
1199 | spin_unlock(&fs_info->trans_lock); | |
5eda7b5e CM |
1200 | } |
1201 | ||
d352ac68 | 1202 | /* |
5d4f98a2 | 1203 | * update all the cowonly tree roots on disk |
d352ac68 | 1204 | */ |
7e4443d9 | 1205 | static noinline int commit_fs_roots(struct btrfs_trans_handle *trans) |
0f7d52f4 | 1206 | { |
7e4443d9 | 1207 | struct btrfs_fs_info *fs_info = trans->fs_info; |
0f7d52f4 | 1208 | struct btrfs_root *gang[8]; |
0f7d52f4 CM |
1209 | int i; |
1210 | int ret; | |
54aa1f4d CM |
1211 | int err = 0; |
1212 | ||
a4abeea4 | 1213 | spin_lock(&fs_info->fs_roots_radix_lock); |
d397712b | 1214 | while (1) { |
5d4f98a2 YZ |
1215 | ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, |
1216 | (void **)gang, 0, | |
0f7d52f4 CM |
1217 | ARRAY_SIZE(gang), |
1218 | BTRFS_ROOT_TRANS_TAG); | |
1219 | if (ret == 0) | |
1220 | break; | |
1221 | for (i = 0; i < ret; i++) { | |
5b4aacef | 1222 | struct btrfs_root *root = gang[i]; |
5d4f98a2 YZ |
1223 | radix_tree_tag_clear(&fs_info->fs_roots_radix, |
1224 | (unsigned long)root->root_key.objectid, | |
1225 | BTRFS_ROOT_TRANS_TAG); | |
a4abeea4 | 1226 | spin_unlock(&fs_info->fs_roots_radix_lock); |
31153d81 | 1227 | |
e02119d5 | 1228 | btrfs_free_log(trans, root); |
5d4f98a2 | 1229 | btrfs_update_reloc_root(trans, root); |
bcc63abb | 1230 | |
82d5902d LZ |
1231 | btrfs_save_ino_cache(root, trans); |
1232 | ||
f1ebcc74 | 1233 | /* see comments in should_cow_block() */ |
27cdeb70 | 1234 | clear_bit(BTRFS_ROOT_FORCE_COW, &root->state); |
c7548af6 | 1235 | smp_mb__after_atomic(); |
f1ebcc74 | 1236 | |
978d910d | 1237 | if (root->commit_root != root->node) { |
9e351cc8 JB |
1238 | list_add_tail(&root->dirty_list, |
1239 | &trans->transaction->switch_commits); | |
978d910d YZ |
1240 | btrfs_set_root_node(&root->root_item, |
1241 | root->node); | |
1242 | } | |
5d4f98a2 | 1243 | |
5d4f98a2 | 1244 | err = btrfs_update_root(trans, fs_info->tree_root, |
0f7d52f4 CM |
1245 | &root->root_key, |
1246 | &root->root_item); | |
a4abeea4 | 1247 | spin_lock(&fs_info->fs_roots_radix_lock); |
54aa1f4d CM |
1248 | if (err) |
1249 | break; | |
733e03a0 | 1250 | btrfs_qgroup_free_meta_all_pertrans(root); |
0f7d52f4 CM |
1251 | } |
1252 | } | |
a4abeea4 | 1253 | spin_unlock(&fs_info->fs_roots_radix_lock); |
54aa1f4d | 1254 | return err; |
0f7d52f4 CM |
1255 | } |
1256 | ||
d352ac68 | 1257 | /* |
de78b51a ES |
1258 | * defrag a given btree. |
1259 | * Every leaf in the btree is read and defragged. | |
d352ac68 | 1260 | */ |
de78b51a | 1261 | int btrfs_defrag_root(struct btrfs_root *root) |
e9d0b13b CM |
1262 | { |
1263 | struct btrfs_fs_info *info = root->fs_info; | |
e9d0b13b | 1264 | struct btrfs_trans_handle *trans; |
8929ecfa | 1265 | int ret; |
e9d0b13b | 1266 | |
27cdeb70 | 1267 | if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state)) |
e9d0b13b | 1268 | return 0; |
8929ecfa | 1269 | |
6b80053d | 1270 | while (1) { |
8929ecfa YZ |
1271 | trans = btrfs_start_transaction(root, 0); |
1272 | if (IS_ERR(trans)) | |
1273 | return PTR_ERR(trans); | |
1274 | ||
de78b51a | 1275 | ret = btrfs_defrag_leaves(trans, root); |
8929ecfa | 1276 | |
3a45bb20 | 1277 | btrfs_end_transaction(trans); |
2ff7e61e | 1278 | btrfs_btree_balance_dirty(info); |
e9d0b13b CM |
1279 | cond_resched(); |
1280 | ||
ab8d0fc4 | 1281 | if (btrfs_fs_closing(info) || ret != -EAGAIN) |
e9d0b13b | 1282 | break; |
210549eb | 1283 | |
ab8d0fc4 JM |
1284 | if (btrfs_defrag_cancelled(info)) { |
1285 | btrfs_debug(info, "defrag_root cancelled"); | |
210549eb DS |
1286 | ret = -EAGAIN; |
1287 | break; | |
1288 | } | |
e9d0b13b | 1289 | } |
27cdeb70 | 1290 | clear_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state); |
8929ecfa | 1291 | return ret; |
e9d0b13b CM |
1292 | } |
1293 | ||
6426c7ad QW |
1294 | /* |
1295 | * Do all special snapshot related qgroup dirty hack. | |
1296 | * | |
1297 | * Will do all needed qgroup inherit and dirty hack like switch commit | |
1298 | * roots inside one transaction and write all btree into disk, to make | |
1299 | * qgroup works. | |
1300 | */ | |
1301 | static int qgroup_account_snapshot(struct btrfs_trans_handle *trans, | |
1302 | struct btrfs_root *src, | |
1303 | struct btrfs_root *parent, | |
1304 | struct btrfs_qgroup_inherit *inherit, | |
1305 | u64 dst_objectid) | |
1306 | { | |
1307 | struct btrfs_fs_info *fs_info = src->fs_info; | |
1308 | int ret; | |
1309 | ||
1310 | /* | |
1311 | * Save some performance in the case that qgroups are not | |
1312 | * enabled. If this check races with the ioctl, rescan will | |
1313 | * kick in anyway. | |
1314 | */ | |
9ea6e2b5 | 1315 | if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) |
6426c7ad | 1316 | return 0; |
6426c7ad | 1317 | |
4d31778a | 1318 | /* |
52042d8e | 1319 | * Ensure dirty @src will be committed. Or, after coming |
4d31778a QW |
1320 | * commit_fs_roots() and switch_commit_roots(), any dirty but not |
1321 | * recorded root will never be updated again, causing an outdated root | |
1322 | * item. | |
1323 | */ | |
1324 | record_root_in_trans(trans, src, 1); | |
1325 | ||
6426c7ad QW |
1326 | /* |
1327 | * We are going to commit transaction, see btrfs_commit_transaction() | |
1328 | * comment for reason locking tree_log_mutex | |
1329 | */ | |
1330 | mutex_lock(&fs_info->tree_log_mutex); | |
1331 | ||
7e4443d9 | 1332 | ret = commit_fs_roots(trans); |
6426c7ad QW |
1333 | if (ret) |
1334 | goto out; | |
460fb20a | 1335 | ret = btrfs_qgroup_account_extents(trans); |
6426c7ad QW |
1336 | if (ret < 0) |
1337 | goto out; | |
1338 | ||
1339 | /* Now qgroup are all updated, we can inherit it to new qgroups */ | |
a9377422 | 1340 | ret = btrfs_qgroup_inherit(trans, src->root_key.objectid, dst_objectid, |
6426c7ad QW |
1341 | inherit); |
1342 | if (ret < 0) | |
1343 | goto out; | |
1344 | ||
1345 | /* | |
1346 | * Now we do a simplified commit transaction, which will: | |
1347 | * 1) commit all subvolume and extent tree | |
1348 | * To ensure all subvolume and extent tree have a valid | |
1349 | * commit_root to accounting later insert_dir_item() | |
1350 | * 2) write all btree blocks onto disk | |
1351 | * This is to make sure later btree modification will be cowed | |
1352 | * Or commit_root can be populated and cause wrong qgroup numbers | |
1353 | * In this simplified commit, we don't really care about other trees | |
1354 | * like chunk and root tree, as they won't affect qgroup. | |
1355 | * And we don't write super to avoid half committed status. | |
1356 | */ | |
9386d8bc | 1357 | ret = commit_cowonly_roots(trans); |
6426c7ad QW |
1358 | if (ret) |
1359 | goto out; | |
16916a88 | 1360 | switch_commit_roots(trans->transaction); |
70458a58 | 1361 | ret = btrfs_write_and_wait_transaction(trans); |
6426c7ad | 1362 | if (ret) |
f7af3934 | 1363 | btrfs_handle_fs_error(fs_info, ret, |
6426c7ad QW |
1364 | "Error while writing out transaction for qgroup"); |
1365 | ||
1366 | out: | |
1367 | mutex_unlock(&fs_info->tree_log_mutex); | |
1368 | ||
1369 | /* | |
1370 | * Force parent root to be updated, as we recorded it before so its | |
1371 | * last_trans == cur_transid. | |
1372 | * Or it won't be committed again onto disk after later | |
1373 | * insert_dir_item() | |
1374 | */ | |
1375 | if (!ret) | |
1376 | record_root_in_trans(trans, parent, 1); | |
1377 | return ret; | |
1378 | } | |
1379 | ||
d352ac68 CM |
1380 | /* |
1381 | * new snapshots need to be created at a very specific time in the | |
aec8030a MX |
1382 | * transaction commit. This does the actual creation. |
1383 | * | |
1384 | * Note: | |
1385 | * If the error which may affect the commitment of the current transaction | |
1386 | * happens, we should return the error number. If the error which just affect | |
1387 | * the creation of the pending snapshots, just return 0. | |
d352ac68 | 1388 | */ |
80b6794d | 1389 | static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, |
3063d29f CM |
1390 | struct btrfs_pending_snapshot *pending) |
1391 | { | |
08d50ca3 NB |
1392 | |
1393 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
3063d29f | 1394 | struct btrfs_key key; |
80b6794d | 1395 | struct btrfs_root_item *new_root_item; |
3063d29f CM |
1396 | struct btrfs_root *tree_root = fs_info->tree_root; |
1397 | struct btrfs_root *root = pending->root; | |
6bdb72de | 1398 | struct btrfs_root *parent_root; |
98c9942a | 1399 | struct btrfs_block_rsv *rsv; |
6bdb72de | 1400 | struct inode *parent_inode; |
42874b3d MX |
1401 | struct btrfs_path *path; |
1402 | struct btrfs_dir_item *dir_item; | |
a22285a6 | 1403 | struct dentry *dentry; |
3063d29f | 1404 | struct extent_buffer *tmp; |
925baedd | 1405 | struct extent_buffer *old; |
95582b00 | 1406 | struct timespec64 cur_time; |
aec8030a | 1407 | int ret = 0; |
d68fc57b | 1408 | u64 to_reserve = 0; |
6bdb72de | 1409 | u64 index = 0; |
a22285a6 | 1410 | u64 objectid; |
b83cc969 | 1411 | u64 root_flags; |
8ea05e3a | 1412 | uuid_le new_uuid; |
3063d29f | 1413 | |
8546b570 DS |
1414 | ASSERT(pending->path); |
1415 | path = pending->path; | |
42874b3d | 1416 | |
b0c0ea63 DS |
1417 | ASSERT(pending->root_item); |
1418 | new_root_item = pending->root_item; | |
a22285a6 | 1419 | |
aec8030a MX |
1420 | pending->error = btrfs_find_free_objectid(tree_root, &objectid); |
1421 | if (pending->error) | |
6fa9700e | 1422 | goto no_free_objectid; |
3063d29f | 1423 | |
d6726335 QW |
1424 | /* |
1425 | * Make qgroup to skip current new snapshot's qgroupid, as it is | |
1426 | * accounted by later btrfs_qgroup_inherit(). | |
1427 | */ | |
1428 | btrfs_set_skip_qgroup(trans, objectid); | |
1429 | ||
147d256e | 1430 | btrfs_reloc_pre_snapshot(pending, &to_reserve); |
d68fc57b YZ |
1431 | |
1432 | if (to_reserve > 0) { | |
aec8030a MX |
1433 | pending->error = btrfs_block_rsv_add(root, |
1434 | &pending->block_rsv, | |
1435 | to_reserve, | |
1436 | BTRFS_RESERVE_NO_FLUSH); | |
1437 | if (pending->error) | |
d6726335 | 1438 | goto clear_skip_qgroup; |
d68fc57b YZ |
1439 | } |
1440 | ||
3063d29f | 1441 | key.objectid = objectid; |
a22285a6 YZ |
1442 | key.offset = (u64)-1; |
1443 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3063d29f | 1444 | |
6fa9700e | 1445 | rsv = trans->block_rsv; |
a22285a6 | 1446 | trans->block_rsv = &pending->block_rsv; |
2382c5cc | 1447 | trans->bytes_reserved = trans->block_rsv->reserved; |
0b246afa | 1448 | trace_btrfs_space_reservation(fs_info, "transaction", |
88d3a5aa JB |
1449 | trans->transid, |
1450 | trans->bytes_reserved, 1); | |
a22285a6 | 1451 | dentry = pending->dentry; |
e9662f70 | 1452 | parent_inode = pending->dir; |
a22285a6 | 1453 | parent_root = BTRFS_I(parent_inode)->root; |
6426c7ad | 1454 | record_root_in_trans(trans, parent_root, 0); |
a22285a6 | 1455 | |
c2050a45 | 1456 | cur_time = current_time(parent_inode); |
04b285f3 | 1457 | |
3063d29f CM |
1458 | /* |
1459 | * insert the directory item | |
1460 | */ | |
877574e2 | 1461 | ret = btrfs_set_inode_index(BTRFS_I(parent_inode), &index); |
49b25e05 | 1462 | BUG_ON(ret); /* -ENOMEM */ |
42874b3d MX |
1463 | |
1464 | /* check if there is a file/dir which has the same name. */ | |
1465 | dir_item = btrfs_lookup_dir_item(NULL, parent_root, path, | |
4a0cc7ca | 1466 | btrfs_ino(BTRFS_I(parent_inode)), |
42874b3d MX |
1467 | dentry->d_name.name, |
1468 | dentry->d_name.len, 0); | |
1469 | if (dir_item != NULL && !IS_ERR(dir_item)) { | |
fe66a05a | 1470 | pending->error = -EEXIST; |
aec8030a | 1471 | goto dir_item_existed; |
42874b3d MX |
1472 | } else if (IS_ERR(dir_item)) { |
1473 | ret = PTR_ERR(dir_item); | |
66642832 | 1474 | btrfs_abort_transaction(trans, ret); |
8732d44f | 1475 | goto fail; |
79787eaa | 1476 | } |
42874b3d | 1477 | btrfs_release_path(path); |
52c26179 | 1478 | |
e999376f CM |
1479 | /* |
1480 | * pull in the delayed directory update | |
1481 | * and the delayed inode item | |
1482 | * otherwise we corrupt the FS during | |
1483 | * snapshot | |
1484 | */ | |
e5c304e6 | 1485 | ret = btrfs_run_delayed_items(trans); |
8732d44f | 1486 | if (ret) { /* Transaction aborted */ |
66642832 | 1487 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1488 | goto fail; |
1489 | } | |
e999376f | 1490 | |
6426c7ad | 1491 | record_root_in_trans(trans, root, 0); |
6bdb72de SW |
1492 | btrfs_set_root_last_snapshot(&root->root_item, trans->transid); |
1493 | memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); | |
08fe4db1 | 1494 | btrfs_check_and_init_root_item(new_root_item); |
6bdb72de | 1495 | |
b83cc969 LZ |
1496 | root_flags = btrfs_root_flags(new_root_item); |
1497 | if (pending->readonly) | |
1498 | root_flags |= BTRFS_ROOT_SUBVOL_RDONLY; | |
1499 | else | |
1500 | root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY; | |
1501 | btrfs_set_root_flags(new_root_item, root_flags); | |
1502 | ||
8ea05e3a AB |
1503 | btrfs_set_root_generation_v2(new_root_item, |
1504 | trans->transid); | |
1505 | uuid_le_gen(&new_uuid); | |
1506 | memcpy(new_root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE); | |
1507 | memcpy(new_root_item->parent_uuid, root->root_item.uuid, | |
1508 | BTRFS_UUID_SIZE); | |
70023da2 SB |
1509 | if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) { |
1510 | memset(new_root_item->received_uuid, 0, | |
1511 | sizeof(new_root_item->received_uuid)); | |
1512 | memset(&new_root_item->stime, 0, sizeof(new_root_item->stime)); | |
1513 | memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime)); | |
1514 | btrfs_set_root_stransid(new_root_item, 0); | |
1515 | btrfs_set_root_rtransid(new_root_item, 0); | |
1516 | } | |
3cae210f QW |
1517 | btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec); |
1518 | btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec); | |
8ea05e3a | 1519 | btrfs_set_root_otransid(new_root_item, trans->transid); |
8ea05e3a | 1520 | |
6bdb72de | 1521 | old = btrfs_lock_root_node(root); |
49b25e05 | 1522 | ret = btrfs_cow_block(trans, root, old, NULL, 0, &old); |
79787eaa JM |
1523 | if (ret) { |
1524 | btrfs_tree_unlock(old); | |
1525 | free_extent_buffer(old); | |
66642832 | 1526 | btrfs_abort_transaction(trans, ret); |
8732d44f | 1527 | goto fail; |
79787eaa | 1528 | } |
49b25e05 | 1529 | |
8bead258 | 1530 | btrfs_set_lock_blocking_write(old); |
6bdb72de | 1531 | |
49b25e05 | 1532 | ret = btrfs_copy_root(trans, root, old, &tmp, objectid); |
79787eaa | 1533 | /* clean up in any case */ |
6bdb72de SW |
1534 | btrfs_tree_unlock(old); |
1535 | free_extent_buffer(old); | |
8732d44f | 1536 | if (ret) { |
66642832 | 1537 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1538 | goto fail; |
1539 | } | |
f1ebcc74 | 1540 | /* see comments in should_cow_block() */ |
27cdeb70 | 1541 | set_bit(BTRFS_ROOT_FORCE_COW, &root->state); |
f1ebcc74 LB |
1542 | smp_wmb(); |
1543 | ||
6bdb72de | 1544 | btrfs_set_root_node(new_root_item, tmp); |
a22285a6 YZ |
1545 | /* record when the snapshot was created in key.offset */ |
1546 | key.offset = trans->transid; | |
1547 | ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); | |
6bdb72de SW |
1548 | btrfs_tree_unlock(tmp); |
1549 | free_extent_buffer(tmp); | |
8732d44f | 1550 | if (ret) { |
66642832 | 1551 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1552 | goto fail; |
1553 | } | |
6bdb72de | 1554 | |
a22285a6 YZ |
1555 | /* |
1556 | * insert root back/forward references | |
1557 | */ | |
6025c19f | 1558 | ret = btrfs_add_root_ref(trans, objectid, |
0660b5af | 1559 | parent_root->root_key.objectid, |
4a0cc7ca | 1560 | btrfs_ino(BTRFS_I(parent_inode)), index, |
a22285a6 | 1561 | dentry->d_name.name, dentry->d_name.len); |
8732d44f | 1562 | if (ret) { |
66642832 | 1563 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1564 | goto fail; |
1565 | } | |
0660b5af | 1566 | |
a22285a6 | 1567 | key.offset = (u64)-1; |
0b246afa | 1568 | pending->snap = btrfs_read_fs_root_no_name(fs_info, &key); |
79787eaa JM |
1569 | if (IS_ERR(pending->snap)) { |
1570 | ret = PTR_ERR(pending->snap); | |
66642832 | 1571 | btrfs_abort_transaction(trans, ret); |
8732d44f | 1572 | goto fail; |
79787eaa | 1573 | } |
d68fc57b | 1574 | |
49b25e05 | 1575 | ret = btrfs_reloc_post_snapshot(trans, pending); |
8732d44f | 1576 | if (ret) { |
66642832 | 1577 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1578 | goto fail; |
1579 | } | |
361048f5 | 1580 | |
c79a70b1 | 1581 | ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); |
8732d44f | 1582 | if (ret) { |
66642832 | 1583 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1584 | goto fail; |
1585 | } | |
42874b3d | 1586 | |
6426c7ad QW |
1587 | /* |
1588 | * Do special qgroup accounting for snapshot, as we do some qgroup | |
1589 | * snapshot hack to do fast snapshot. | |
1590 | * To co-operate with that hack, we do hack again. | |
1591 | * Or snapshot will be greatly slowed down by a subtree qgroup rescan | |
1592 | */ | |
1593 | ret = qgroup_account_snapshot(trans, root, parent_root, | |
1594 | pending->inherit, objectid); | |
1595 | if (ret < 0) | |
1596 | goto fail; | |
1597 | ||
684572df LF |
1598 | ret = btrfs_insert_dir_item(trans, dentry->d_name.name, |
1599 | dentry->d_name.len, BTRFS_I(parent_inode), | |
1600 | &key, BTRFS_FT_DIR, index); | |
42874b3d | 1601 | /* We have check then name at the beginning, so it is impossible. */ |
9c52057c | 1602 | BUG_ON(ret == -EEXIST || ret == -EOVERFLOW); |
8732d44f | 1603 | if (ret) { |
66642832 | 1604 | btrfs_abort_transaction(trans, ret); |
8732d44f MX |
1605 | goto fail; |
1606 | } | |
42874b3d | 1607 | |
6ef06d27 | 1608 | btrfs_i_size_write(BTRFS_I(parent_inode), parent_inode->i_size + |
42874b3d | 1609 | dentry->d_name.len * 2); |
04b285f3 | 1610 | parent_inode->i_mtime = parent_inode->i_ctime = |
c2050a45 | 1611 | current_time(parent_inode); |
be6aef60 | 1612 | ret = btrfs_update_inode_fallback(trans, parent_root, parent_inode); |
dd5f9615 | 1613 | if (ret) { |
66642832 | 1614 | btrfs_abort_transaction(trans, ret); |
dd5f9615 SB |
1615 | goto fail; |
1616 | } | |
cdb345a8 LF |
1617 | ret = btrfs_uuid_tree_add(trans, new_uuid.b, BTRFS_UUID_KEY_SUBVOL, |
1618 | objectid); | |
dd5f9615 | 1619 | if (ret) { |
66642832 | 1620 | btrfs_abort_transaction(trans, ret); |
dd5f9615 SB |
1621 | goto fail; |
1622 | } | |
1623 | if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) { | |
cdb345a8 | 1624 | ret = btrfs_uuid_tree_add(trans, new_root_item->received_uuid, |
dd5f9615 SB |
1625 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, |
1626 | objectid); | |
1627 | if (ret && ret != -EEXIST) { | |
66642832 | 1628 | btrfs_abort_transaction(trans, ret); |
dd5f9615 SB |
1629 | goto fail; |
1630 | } | |
1631 | } | |
d6726335 | 1632 | |
c79a70b1 | 1633 | ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); |
d6726335 | 1634 | if (ret) { |
66642832 | 1635 | btrfs_abort_transaction(trans, ret); |
d6726335 QW |
1636 | goto fail; |
1637 | } | |
1638 | ||
3063d29f | 1639 | fail: |
aec8030a MX |
1640 | pending->error = ret; |
1641 | dir_item_existed: | |
98c9942a | 1642 | trans->block_rsv = rsv; |
2382c5cc | 1643 | trans->bytes_reserved = 0; |
d6726335 QW |
1644 | clear_skip_qgroup: |
1645 | btrfs_clear_skip_qgroup(trans); | |
6fa9700e MX |
1646 | no_free_objectid: |
1647 | kfree(new_root_item); | |
b0c0ea63 | 1648 | pending->root_item = NULL; |
42874b3d | 1649 | btrfs_free_path(path); |
8546b570 DS |
1650 | pending->path = NULL; |
1651 | ||
49b25e05 | 1652 | return ret; |
3063d29f CM |
1653 | } |
1654 | ||
d352ac68 CM |
1655 | /* |
1656 | * create all the snapshots we've scheduled for creation | |
1657 | */ | |
08d50ca3 | 1658 | static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans) |
3de4586c | 1659 | { |
aec8030a | 1660 | struct btrfs_pending_snapshot *pending, *next; |
3de4586c | 1661 | struct list_head *head = &trans->transaction->pending_snapshots; |
aec8030a | 1662 | int ret = 0; |
3de4586c | 1663 | |
aec8030a MX |
1664 | list_for_each_entry_safe(pending, next, head, list) { |
1665 | list_del(&pending->list); | |
08d50ca3 | 1666 | ret = create_pending_snapshot(trans, pending); |
aec8030a MX |
1667 | if (ret) |
1668 | break; | |
1669 | } | |
1670 | return ret; | |
3de4586c CM |
1671 | } |
1672 | ||
2ff7e61e | 1673 | static void update_super_roots(struct btrfs_fs_info *fs_info) |
5d4f98a2 YZ |
1674 | { |
1675 | struct btrfs_root_item *root_item; | |
1676 | struct btrfs_super_block *super; | |
1677 | ||
0b246afa | 1678 | super = fs_info->super_copy; |
5d4f98a2 | 1679 | |
0b246afa | 1680 | root_item = &fs_info->chunk_root->root_item; |
093e037c DS |
1681 | super->chunk_root = root_item->bytenr; |
1682 | super->chunk_root_generation = root_item->generation; | |
1683 | super->chunk_root_level = root_item->level; | |
5d4f98a2 | 1684 | |
0b246afa | 1685 | root_item = &fs_info->tree_root->root_item; |
093e037c DS |
1686 | super->root = root_item->bytenr; |
1687 | super->generation = root_item->generation; | |
1688 | super->root_level = root_item->level; | |
0b246afa | 1689 | if (btrfs_test_opt(fs_info, SPACE_CACHE)) |
093e037c | 1690 | super->cache_generation = root_item->generation; |
0b246afa | 1691 | if (test_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags)) |
093e037c | 1692 | super->uuid_tree_generation = root_item->generation; |
5d4f98a2 YZ |
1693 | } |
1694 | ||
f36f3042 CM |
1695 | int btrfs_transaction_in_commit(struct btrfs_fs_info *info) |
1696 | { | |
4a9d8bde | 1697 | struct btrfs_transaction *trans; |
f36f3042 | 1698 | int ret = 0; |
4a9d8bde | 1699 | |
a4abeea4 | 1700 | spin_lock(&info->trans_lock); |
4a9d8bde MX |
1701 | trans = info->running_transaction; |
1702 | if (trans) | |
1703 | ret = (trans->state >= TRANS_STATE_COMMIT_START); | |
a4abeea4 | 1704 | spin_unlock(&info->trans_lock); |
f36f3042 CM |
1705 | return ret; |
1706 | } | |
1707 | ||
8929ecfa YZ |
1708 | int btrfs_transaction_blocked(struct btrfs_fs_info *info) |
1709 | { | |
4a9d8bde | 1710 | struct btrfs_transaction *trans; |
8929ecfa | 1711 | int ret = 0; |
4a9d8bde | 1712 | |
a4abeea4 | 1713 | spin_lock(&info->trans_lock); |
4a9d8bde MX |
1714 | trans = info->running_transaction; |
1715 | if (trans) | |
1716 | ret = is_transaction_blocked(trans); | |
a4abeea4 | 1717 | spin_unlock(&info->trans_lock); |
8929ecfa YZ |
1718 | return ret; |
1719 | } | |
1720 | ||
bb9c12c9 SW |
1721 | /* |
1722 | * wait for the current transaction commit to start and block subsequent | |
1723 | * transaction joins | |
1724 | */ | |
2ff7e61e | 1725 | static void wait_current_trans_commit_start(struct btrfs_fs_info *fs_info, |
bb9c12c9 SW |
1726 | struct btrfs_transaction *trans) |
1727 | { | |
2ff7e61e JM |
1728 | wait_event(fs_info->transaction_blocked_wait, |
1729 | trans->state >= TRANS_STATE_COMMIT_START || trans->aborted); | |
bb9c12c9 SW |
1730 | } |
1731 | ||
1732 | /* | |
1733 | * wait for the current transaction to start and then become unblocked. | |
1734 | * caller holds ref. | |
1735 | */ | |
2ff7e61e JM |
1736 | static void wait_current_trans_commit_start_and_unblock( |
1737 | struct btrfs_fs_info *fs_info, | |
1738 | struct btrfs_transaction *trans) | |
bb9c12c9 | 1739 | { |
2ff7e61e JM |
1740 | wait_event(fs_info->transaction_wait, |
1741 | trans->state >= TRANS_STATE_UNBLOCKED || trans->aborted); | |
bb9c12c9 SW |
1742 | } |
1743 | ||
1744 | /* | |
1745 | * commit transactions asynchronously. once btrfs_commit_transaction_async | |
1746 | * returns, any subsequent transaction will not be allowed to join. | |
1747 | */ | |
1748 | struct btrfs_async_commit { | |
1749 | struct btrfs_trans_handle *newtrans; | |
7892b5af | 1750 | struct work_struct work; |
bb9c12c9 SW |
1751 | }; |
1752 | ||
1753 | static void do_async_commit(struct work_struct *work) | |
1754 | { | |
1755 | struct btrfs_async_commit *ac = | |
7892b5af | 1756 | container_of(work, struct btrfs_async_commit, work); |
bb9c12c9 | 1757 | |
6fc4e354 SW |
1758 | /* |
1759 | * We've got freeze protection passed with the transaction. | |
1760 | * Tell lockdep about it. | |
1761 | */ | |
b1a06a4b | 1762 | if (ac->newtrans->type & __TRANS_FREEZABLE) |
3a45bb20 | 1763 | __sb_writers_acquired(ac->newtrans->fs_info->sb, SB_FREEZE_FS); |
6fc4e354 | 1764 | |
e209db7a SW |
1765 | current->journal_info = ac->newtrans; |
1766 | ||
3a45bb20 | 1767 | btrfs_commit_transaction(ac->newtrans); |
bb9c12c9 SW |
1768 | kfree(ac); |
1769 | } | |
1770 | ||
1771 | int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans, | |
bb9c12c9 SW |
1772 | int wait_for_unblock) |
1773 | { | |
3a45bb20 | 1774 | struct btrfs_fs_info *fs_info = trans->fs_info; |
bb9c12c9 SW |
1775 | struct btrfs_async_commit *ac; |
1776 | struct btrfs_transaction *cur_trans; | |
1777 | ||
1778 | ac = kmalloc(sizeof(*ac), GFP_NOFS); | |
db5b493a TI |
1779 | if (!ac) |
1780 | return -ENOMEM; | |
bb9c12c9 | 1781 | |
7892b5af | 1782 | INIT_WORK(&ac->work, do_async_commit); |
3a45bb20 | 1783 | ac->newtrans = btrfs_join_transaction(trans->root); |
3612b495 TI |
1784 | if (IS_ERR(ac->newtrans)) { |
1785 | int err = PTR_ERR(ac->newtrans); | |
1786 | kfree(ac); | |
1787 | return err; | |
1788 | } | |
bb9c12c9 SW |
1789 | |
1790 | /* take transaction reference */ | |
bb9c12c9 | 1791 | cur_trans = trans->transaction; |
9b64f57d | 1792 | refcount_inc(&cur_trans->use_count); |
bb9c12c9 | 1793 | |
3a45bb20 | 1794 | btrfs_end_transaction(trans); |
6fc4e354 SW |
1795 | |
1796 | /* | |
1797 | * Tell lockdep we've released the freeze rwsem, since the | |
1798 | * async commit thread will be the one to unlock it. | |
1799 | */ | |
b1a06a4b | 1800 | if (ac->newtrans->type & __TRANS_FREEZABLE) |
0b246afa | 1801 | __sb_writers_release(fs_info->sb, SB_FREEZE_FS); |
6fc4e354 | 1802 | |
7892b5af | 1803 | schedule_work(&ac->work); |
bb9c12c9 SW |
1804 | |
1805 | /* wait for transaction to start and unblock */ | |
bb9c12c9 | 1806 | if (wait_for_unblock) |
2ff7e61e | 1807 | wait_current_trans_commit_start_and_unblock(fs_info, cur_trans); |
bb9c12c9 | 1808 | else |
2ff7e61e | 1809 | wait_current_trans_commit_start(fs_info, cur_trans); |
bb9c12c9 | 1810 | |
38e88054 SW |
1811 | if (current->journal_info == trans) |
1812 | current->journal_info = NULL; | |
1813 | ||
724e2315 | 1814 | btrfs_put_transaction(cur_trans); |
bb9c12c9 SW |
1815 | return 0; |
1816 | } | |
1817 | ||
49b25e05 | 1818 | |
97cb39bb | 1819 | static void cleanup_transaction(struct btrfs_trans_handle *trans, int err) |
49b25e05 | 1820 | { |
97cb39bb | 1821 | struct btrfs_fs_info *fs_info = trans->fs_info; |
49b25e05 JM |
1822 | struct btrfs_transaction *cur_trans = trans->transaction; |
1823 | ||
b50fff81 | 1824 | WARN_ON(refcount_read(&trans->use_count) > 1); |
49b25e05 | 1825 | |
66642832 | 1826 | btrfs_abort_transaction(trans, err); |
7b8b92af | 1827 | |
0b246afa | 1828 | spin_lock(&fs_info->trans_lock); |
66b6135b | 1829 | |
25d8c284 MX |
1830 | /* |
1831 | * If the transaction is removed from the list, it means this | |
1832 | * transaction has been committed successfully, so it is impossible | |
1833 | * to call the cleanup function. | |
1834 | */ | |
1835 | BUG_ON(list_empty(&cur_trans->list)); | |
66b6135b | 1836 | |
49b25e05 | 1837 | list_del_init(&cur_trans->list); |
0b246afa | 1838 | if (cur_trans == fs_info->running_transaction) { |
4a9d8bde | 1839 | cur_trans->state = TRANS_STATE_COMMIT_DOING; |
0b246afa | 1840 | spin_unlock(&fs_info->trans_lock); |
f094ac32 LB |
1841 | wait_event(cur_trans->writer_wait, |
1842 | atomic_read(&cur_trans->num_writers) == 1); | |
1843 | ||
0b246afa | 1844 | spin_lock(&fs_info->trans_lock); |
d7096fc3 | 1845 | } |
0b246afa | 1846 | spin_unlock(&fs_info->trans_lock); |
49b25e05 | 1847 | |
2ff7e61e | 1848 | btrfs_cleanup_one_transaction(trans->transaction, fs_info); |
49b25e05 | 1849 | |
0b246afa JM |
1850 | spin_lock(&fs_info->trans_lock); |
1851 | if (cur_trans == fs_info->running_transaction) | |
1852 | fs_info->running_transaction = NULL; | |
1853 | spin_unlock(&fs_info->trans_lock); | |
4a9d8bde | 1854 | |
e0228285 | 1855 | if (trans->type & __TRANS_FREEZABLE) |
0b246afa | 1856 | sb_end_intwrite(fs_info->sb); |
724e2315 JB |
1857 | btrfs_put_transaction(cur_trans); |
1858 | btrfs_put_transaction(cur_trans); | |
49b25e05 | 1859 | |
97cb39bb | 1860 | trace_btrfs_transaction_commit(trans->root); |
49b25e05 | 1861 | |
49b25e05 JM |
1862 | if (current->journal_info == trans) |
1863 | current->journal_info = NULL; | |
0b246afa | 1864 | btrfs_scrub_cancel(fs_info); |
49b25e05 JM |
1865 | |
1866 | kmem_cache_free(btrfs_trans_handle_cachep, trans); | |
1867 | } | |
1868 | ||
c7cc64a9 DS |
1869 | /* |
1870 | * Release reserved delayed ref space of all pending block groups of the | |
1871 | * transaction and remove them from the list | |
1872 | */ | |
1873 | static void btrfs_cleanup_pending_block_groups(struct btrfs_trans_handle *trans) | |
1874 | { | |
1875 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
1876 | struct btrfs_block_group_cache *block_group, *tmp; | |
1877 | ||
1878 | list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) { | |
1879 | btrfs_delayed_refs_rsv_release(fs_info, 1); | |
1880 | list_del_init(&block_group->bg_list); | |
1881 | } | |
1882 | } | |
1883 | ||
609e804d | 1884 | static inline int btrfs_start_delalloc_flush(struct btrfs_trans_handle *trans) |
82436617 | 1885 | { |
609e804d FM |
1886 | struct btrfs_fs_info *fs_info = trans->fs_info; |
1887 | ||
ce8ea7cc JB |
1888 | /* |
1889 | * We use writeback_inodes_sb here because if we used | |
1890 | * btrfs_start_delalloc_roots we would deadlock with fs freeze. | |
1891 | * Currently are holding the fs freeze lock, if we do an async flush | |
1892 | * we'll do btrfs_join_transaction() and deadlock because we need to | |
1893 | * wait for the fs freeze lock. Using the direct flushing we benefit | |
1894 | * from already being in a transaction and our join_transaction doesn't | |
1895 | * have to re-take the fs freeze lock. | |
1896 | */ | |
609e804d | 1897 | if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) { |
ce8ea7cc | 1898 | writeback_inodes_sb(fs_info->sb, WB_REASON_SYNC); |
609e804d FM |
1899 | } else { |
1900 | struct btrfs_pending_snapshot *pending; | |
1901 | struct list_head *head = &trans->transaction->pending_snapshots; | |
1902 | ||
1903 | /* | |
1904 | * Flush dellaloc for any root that is going to be snapshotted. | |
1905 | * This is done to avoid a corrupted version of files, in the | |
1906 | * snapshots, that had both buffered and direct IO writes (even | |
1907 | * if they were done sequentially) due to an unordered update of | |
1908 | * the inode's size on disk. | |
1909 | */ | |
1910 | list_for_each_entry(pending, head, list) { | |
1911 | int ret; | |
1912 | ||
1913 | ret = btrfs_start_delalloc_snapshot(pending->root); | |
1914 | if (ret) | |
1915 | return ret; | |
1916 | } | |
1917 | } | |
82436617 MX |
1918 | return 0; |
1919 | } | |
1920 | ||
609e804d | 1921 | static inline void btrfs_wait_delalloc_flush(struct btrfs_trans_handle *trans) |
82436617 | 1922 | { |
609e804d FM |
1923 | struct btrfs_fs_info *fs_info = trans->fs_info; |
1924 | ||
1925 | if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) { | |
6374e57a | 1926 | btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1); |
609e804d FM |
1927 | } else { |
1928 | struct btrfs_pending_snapshot *pending; | |
1929 | struct list_head *head = &trans->transaction->pending_snapshots; | |
1930 | ||
1931 | /* | |
1932 | * Wait for any dellaloc that we started previously for the roots | |
1933 | * that are going to be snapshotted. This is to avoid a corrupted | |
1934 | * version of files in the snapshots that had both buffered and | |
1935 | * direct IO writes (even if they were done sequentially). | |
1936 | */ | |
1937 | list_for_each_entry(pending, head, list) | |
1938 | btrfs_wait_ordered_extents(pending->root, | |
1939 | U64_MAX, 0, U64_MAX); | |
1940 | } | |
82436617 MX |
1941 | } |
1942 | ||
3a45bb20 | 1943 | int btrfs_commit_transaction(struct btrfs_trans_handle *trans) |
79154b1b | 1944 | { |
3a45bb20 | 1945 | struct btrfs_fs_info *fs_info = trans->fs_info; |
49b25e05 | 1946 | struct btrfs_transaction *cur_trans = trans->transaction; |
8fd17795 | 1947 | struct btrfs_transaction *prev_trans = NULL; |
25287e0a | 1948 | int ret; |
79154b1b | 1949 | |
8d25a086 | 1950 | /* Stop the commit early if ->aborted is set */ |
20c7bcec | 1951 | if (unlikely(READ_ONCE(cur_trans->aborted))) { |
25287e0a | 1952 | ret = cur_trans->aborted; |
3a45bb20 | 1953 | btrfs_end_transaction(trans); |
e4a2bcac | 1954 | return ret; |
25287e0a | 1955 | } |
49b25e05 | 1956 | |
f45c752b JB |
1957 | btrfs_trans_release_metadata(trans); |
1958 | trans->block_rsv = NULL; | |
1959 | ||
56bec294 CM |
1960 | /* make a pass through all the delayed refs we have so far |
1961 | * any runnings procs may add more while we are here | |
1962 | */ | |
c79a70b1 | 1963 | ret = btrfs_run_delayed_refs(trans, 0); |
e4a2bcac | 1964 | if (ret) { |
3a45bb20 | 1965 | btrfs_end_transaction(trans); |
e4a2bcac JB |
1966 | return ret; |
1967 | } | |
56bec294 | 1968 | |
b7ec40d7 | 1969 | cur_trans = trans->transaction; |
49b25e05 | 1970 | |
56bec294 CM |
1971 | /* |
1972 | * set the flushing flag so procs in this transaction have to | |
1973 | * start sending their work down. | |
1974 | */ | |
b7ec40d7 | 1975 | cur_trans->delayed_refs.flushing = 1; |
1be41b78 | 1976 | smp_wmb(); |
56bec294 | 1977 | |
119e80df | 1978 | btrfs_create_pending_block_groups(trans); |
ea658bad | 1979 | |
c79a70b1 | 1980 | ret = btrfs_run_delayed_refs(trans, 0); |
e4a2bcac | 1981 | if (ret) { |
3a45bb20 | 1982 | btrfs_end_transaction(trans); |
e4a2bcac JB |
1983 | return ret; |
1984 | } | |
56bec294 | 1985 | |
3204d33c | 1986 | if (!test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &cur_trans->flags)) { |
1bbc621e CM |
1987 | int run_it = 0; |
1988 | ||
1989 | /* this mutex is also taken before trying to set | |
1990 | * block groups readonly. We need to make sure | |
1991 | * that nobody has set a block group readonly | |
1992 | * after a extents from that block group have been | |
1993 | * allocated for cache files. btrfs_set_block_group_ro | |
1994 | * will wait for the transaction to commit if it | |
3204d33c | 1995 | * finds BTRFS_TRANS_DIRTY_BG_RUN set. |
1bbc621e | 1996 | * |
3204d33c JB |
1997 | * The BTRFS_TRANS_DIRTY_BG_RUN flag is also used to make sure |
1998 | * only one process starts all the block group IO. It wouldn't | |
1bbc621e CM |
1999 | * hurt to have more than one go through, but there's no |
2000 | * real advantage to it either. | |
2001 | */ | |
0b246afa | 2002 | mutex_lock(&fs_info->ro_block_group_mutex); |
3204d33c JB |
2003 | if (!test_and_set_bit(BTRFS_TRANS_DIRTY_BG_RUN, |
2004 | &cur_trans->flags)) | |
1bbc621e | 2005 | run_it = 1; |
0b246afa | 2006 | mutex_unlock(&fs_info->ro_block_group_mutex); |
1bbc621e | 2007 | |
f9cacae3 | 2008 | if (run_it) { |
21217054 | 2009 | ret = btrfs_start_dirty_block_groups(trans); |
f9cacae3 NB |
2010 | if (ret) { |
2011 | btrfs_end_transaction(trans); | |
2012 | return ret; | |
2013 | } | |
2014 | } | |
1bbc621e CM |
2015 | } |
2016 | ||
0b246afa | 2017 | spin_lock(&fs_info->trans_lock); |
4a9d8bde | 2018 | if (cur_trans->state >= TRANS_STATE_COMMIT_START) { |
0b246afa | 2019 | spin_unlock(&fs_info->trans_lock); |
9b64f57d | 2020 | refcount_inc(&cur_trans->use_count); |
3a45bb20 | 2021 | ret = btrfs_end_transaction(trans); |
ccd467d6 | 2022 | |
2ff7e61e | 2023 | wait_for_commit(cur_trans); |
15ee9bc7 | 2024 | |
b4924a0f LB |
2025 | if (unlikely(cur_trans->aborted)) |
2026 | ret = cur_trans->aborted; | |
2027 | ||
724e2315 | 2028 | btrfs_put_transaction(cur_trans); |
15ee9bc7 | 2029 | |
49b25e05 | 2030 | return ret; |
79154b1b | 2031 | } |
4313b399 | 2032 | |
4a9d8bde | 2033 | cur_trans->state = TRANS_STATE_COMMIT_START; |
0b246afa | 2034 | wake_up(&fs_info->transaction_blocked_wait); |
bb9c12c9 | 2035 | |
0b246afa | 2036 | if (cur_trans->list.prev != &fs_info->trans_list) { |
ccd467d6 CM |
2037 | prev_trans = list_entry(cur_trans->list.prev, |
2038 | struct btrfs_transaction, list); | |
4a9d8bde | 2039 | if (prev_trans->state != TRANS_STATE_COMPLETED) { |
9b64f57d | 2040 | refcount_inc(&prev_trans->use_count); |
0b246afa | 2041 | spin_unlock(&fs_info->trans_lock); |
ccd467d6 | 2042 | |
2ff7e61e | 2043 | wait_for_commit(prev_trans); |
1f9b8c8f | 2044 | ret = prev_trans->aborted; |
ccd467d6 | 2045 | |
724e2315 | 2046 | btrfs_put_transaction(prev_trans); |
1f9b8c8f FM |
2047 | if (ret) |
2048 | goto cleanup_transaction; | |
a4abeea4 | 2049 | } else { |
0b246afa | 2050 | spin_unlock(&fs_info->trans_lock); |
ccd467d6 | 2051 | } |
a4abeea4 | 2052 | } else { |
0b246afa | 2053 | spin_unlock(&fs_info->trans_lock); |
cb2d3dad FM |
2054 | /* |
2055 | * The previous transaction was aborted and was already removed | |
2056 | * from the list of transactions at fs_info->trans_list. So we | |
2057 | * abort to prevent writing a new superblock that reflects a | |
2058 | * corrupt state (pointing to trees with unwritten nodes/leafs). | |
2059 | */ | |
2060 | if (test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state)) { | |
2061 | ret = -EROFS; | |
2062 | goto cleanup_transaction; | |
2063 | } | |
ccd467d6 | 2064 | } |
15ee9bc7 | 2065 | |
0860adfd MX |
2066 | extwriter_counter_dec(cur_trans, trans->type); |
2067 | ||
609e804d | 2068 | ret = btrfs_start_delalloc_flush(trans); |
82436617 MX |
2069 | if (ret) |
2070 | goto cleanup_transaction; | |
2071 | ||
e5c304e6 | 2072 | ret = btrfs_run_delayed_items(trans); |
581227d0 MX |
2073 | if (ret) |
2074 | goto cleanup_transaction; | |
15ee9bc7 | 2075 | |
581227d0 MX |
2076 | wait_event(cur_trans->writer_wait, |
2077 | extwriter_counter_read(cur_trans) == 0); | |
15ee9bc7 | 2078 | |
581227d0 | 2079 | /* some pending stuffs might be added after the previous flush. */ |
e5c304e6 | 2080 | ret = btrfs_run_delayed_items(trans); |
ca469637 MX |
2081 | if (ret) |
2082 | goto cleanup_transaction; | |
2083 | ||
609e804d | 2084 | btrfs_wait_delalloc_flush(trans); |
cb7ab021 | 2085 | |
2ff7e61e | 2086 | btrfs_scrub_pause(fs_info); |
ed0ca140 JB |
2087 | /* |
2088 | * Ok now we need to make sure to block out any other joins while we | |
2089 | * commit the transaction. We could have started a join before setting | |
4a9d8bde | 2090 | * COMMIT_DOING so make sure to wait for num_writers to == 1 again. |
ed0ca140 | 2091 | */ |
0b246afa | 2092 | spin_lock(&fs_info->trans_lock); |
4a9d8bde | 2093 | cur_trans->state = TRANS_STATE_COMMIT_DOING; |
0b246afa | 2094 | spin_unlock(&fs_info->trans_lock); |
ed0ca140 JB |
2095 | wait_event(cur_trans->writer_wait, |
2096 | atomic_read(&cur_trans->num_writers) == 1); | |
2097 | ||
2cba30f1 | 2098 | /* ->aborted might be set after the previous check, so check it */ |
20c7bcec | 2099 | if (unlikely(READ_ONCE(cur_trans->aborted))) { |
2cba30f1 | 2100 | ret = cur_trans->aborted; |
6cf7f77e | 2101 | goto scrub_continue; |
2cba30f1 | 2102 | } |
7585717f CM |
2103 | /* |
2104 | * the reloc mutex makes sure that we stop | |
2105 | * the balancing code from coming in and moving | |
2106 | * extents around in the middle of the commit | |
2107 | */ | |
0b246afa | 2108 | mutex_lock(&fs_info->reloc_mutex); |
7585717f | 2109 | |
42874b3d MX |
2110 | /* |
2111 | * We needn't worry about the delayed items because we will | |
2112 | * deal with them in create_pending_snapshot(), which is the | |
2113 | * core function of the snapshot creation. | |
2114 | */ | |
08d50ca3 | 2115 | ret = create_pending_snapshots(trans); |
49b25e05 | 2116 | if (ret) { |
0b246afa | 2117 | mutex_unlock(&fs_info->reloc_mutex); |
6cf7f77e | 2118 | goto scrub_continue; |
49b25e05 | 2119 | } |
3063d29f | 2120 | |
42874b3d MX |
2121 | /* |
2122 | * We insert the dir indexes of the snapshots and update the inode | |
2123 | * of the snapshots' parents after the snapshot creation, so there | |
2124 | * are some delayed items which are not dealt with. Now deal with | |
2125 | * them. | |
2126 | * | |
2127 | * We needn't worry that this operation will corrupt the snapshots, | |
2128 | * because all the tree which are snapshoted will be forced to COW | |
2129 | * the nodes and leaves. | |
2130 | */ | |
e5c304e6 | 2131 | ret = btrfs_run_delayed_items(trans); |
49b25e05 | 2132 | if (ret) { |
0b246afa | 2133 | mutex_unlock(&fs_info->reloc_mutex); |
6cf7f77e | 2134 | goto scrub_continue; |
49b25e05 | 2135 | } |
16cdcec7 | 2136 | |
c79a70b1 | 2137 | ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); |
49b25e05 | 2138 | if (ret) { |
0b246afa | 2139 | mutex_unlock(&fs_info->reloc_mutex); |
6cf7f77e | 2140 | goto scrub_continue; |
49b25e05 | 2141 | } |
56bec294 | 2142 | |
e999376f CM |
2143 | /* |
2144 | * make sure none of the code above managed to slip in a | |
2145 | * delayed item | |
2146 | */ | |
ccdf9b30 | 2147 | btrfs_assert_delayed_root_empty(fs_info); |
e999376f | 2148 | |
2c90e5d6 | 2149 | WARN_ON(cur_trans != trans->transaction); |
dc17ff8f | 2150 | |
e02119d5 CM |
2151 | /* btrfs_commit_tree_roots is responsible for getting the |
2152 | * various roots consistent with each other. Every pointer | |
2153 | * in the tree of tree roots has to point to the most up to date | |
2154 | * root for every subvolume and other tree. So, we have to keep | |
2155 | * the tree logging code from jumping in and changing any | |
2156 | * of the trees. | |
2157 | * | |
2158 | * At this point in the commit, there can't be any tree-log | |
2159 | * writers, but a little lower down we drop the trans mutex | |
2160 | * and let new people in. By holding the tree_log_mutex | |
2161 | * from now until after the super is written, we avoid races | |
2162 | * with the tree-log code. | |
2163 | */ | |
0b246afa | 2164 | mutex_lock(&fs_info->tree_log_mutex); |
e02119d5 | 2165 | |
7e4443d9 | 2166 | ret = commit_fs_roots(trans); |
49b25e05 | 2167 | if (ret) { |
0b246afa JM |
2168 | mutex_unlock(&fs_info->tree_log_mutex); |
2169 | mutex_unlock(&fs_info->reloc_mutex); | |
6cf7f77e | 2170 | goto scrub_continue; |
49b25e05 | 2171 | } |
54aa1f4d | 2172 | |
3818aea2 | 2173 | /* |
7e1876ac DS |
2174 | * Since the transaction is done, we can apply the pending changes |
2175 | * before the next transaction. | |
3818aea2 | 2176 | */ |
0b246afa | 2177 | btrfs_apply_pending_changes(fs_info); |
3818aea2 | 2178 | |
5d4f98a2 | 2179 | /* commit_fs_roots gets rid of all the tree log roots, it is now |
e02119d5 CM |
2180 | * safe to free the root of tree log roots |
2181 | */ | |
0b246afa | 2182 | btrfs_free_log_root_tree(trans, fs_info); |
e02119d5 | 2183 | |
82bafb38 QW |
2184 | /* |
2185 | * commit_fs_roots() can call btrfs_save_ino_cache(), which generates | |
2186 | * new delayed refs. Must handle them or qgroup can be wrong. | |
2187 | */ | |
c79a70b1 | 2188 | ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); |
82bafb38 QW |
2189 | if (ret) { |
2190 | mutex_unlock(&fs_info->tree_log_mutex); | |
2191 | mutex_unlock(&fs_info->reloc_mutex); | |
2192 | goto scrub_continue; | |
2193 | } | |
2194 | ||
0ed4792a QW |
2195 | /* |
2196 | * Since fs roots are all committed, we can get a quite accurate | |
2197 | * new_roots. So let's do quota accounting. | |
2198 | */ | |
460fb20a | 2199 | ret = btrfs_qgroup_account_extents(trans); |
0ed4792a | 2200 | if (ret < 0) { |
0b246afa JM |
2201 | mutex_unlock(&fs_info->tree_log_mutex); |
2202 | mutex_unlock(&fs_info->reloc_mutex); | |
0ed4792a QW |
2203 | goto scrub_continue; |
2204 | } | |
2205 | ||
9386d8bc | 2206 | ret = commit_cowonly_roots(trans); |
49b25e05 | 2207 | if (ret) { |
0b246afa JM |
2208 | mutex_unlock(&fs_info->tree_log_mutex); |
2209 | mutex_unlock(&fs_info->reloc_mutex); | |
6cf7f77e | 2210 | goto scrub_continue; |
49b25e05 | 2211 | } |
54aa1f4d | 2212 | |
2cba30f1 MX |
2213 | /* |
2214 | * The tasks which save the space cache and inode cache may also | |
2215 | * update ->aborted, check it. | |
2216 | */ | |
20c7bcec | 2217 | if (unlikely(READ_ONCE(cur_trans->aborted))) { |
2cba30f1 | 2218 | ret = cur_trans->aborted; |
0b246afa JM |
2219 | mutex_unlock(&fs_info->tree_log_mutex); |
2220 | mutex_unlock(&fs_info->reloc_mutex); | |
6cf7f77e | 2221 | goto scrub_continue; |
2cba30f1 MX |
2222 | } |
2223 | ||
8b74c03e | 2224 | btrfs_prepare_extent_commit(fs_info); |
11833d66 | 2225 | |
0b246afa | 2226 | cur_trans = fs_info->running_transaction; |
5d4f98a2 | 2227 | |
0b246afa JM |
2228 | btrfs_set_root_node(&fs_info->tree_root->root_item, |
2229 | fs_info->tree_root->node); | |
2230 | list_add_tail(&fs_info->tree_root->dirty_list, | |
9e351cc8 | 2231 | &cur_trans->switch_commits); |
5d4f98a2 | 2232 | |
0b246afa JM |
2233 | btrfs_set_root_node(&fs_info->chunk_root->root_item, |
2234 | fs_info->chunk_root->node); | |
2235 | list_add_tail(&fs_info->chunk_root->dirty_list, | |
9e351cc8 JB |
2236 | &cur_trans->switch_commits); |
2237 | ||
16916a88 | 2238 | switch_commit_roots(cur_trans); |
5d4f98a2 | 2239 | |
ce93ec54 | 2240 | ASSERT(list_empty(&cur_trans->dirty_bgs)); |
1bbc621e | 2241 | ASSERT(list_empty(&cur_trans->io_bgs)); |
2ff7e61e | 2242 | update_super_roots(fs_info); |
e02119d5 | 2243 | |
0b246afa JM |
2244 | btrfs_set_super_log_root(fs_info->super_copy, 0); |
2245 | btrfs_set_super_log_root_level(fs_info->super_copy, 0); | |
2246 | memcpy(fs_info->super_for_commit, fs_info->super_copy, | |
2247 | sizeof(*fs_info->super_copy)); | |
ccd467d6 | 2248 | |
bbbf7243 | 2249 | btrfs_commit_device_sizes(cur_trans); |
935e5cc9 | 2250 | |
0b246afa JM |
2251 | clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags); |
2252 | clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags); | |
656f30db | 2253 | |
4fbcdf66 FM |
2254 | btrfs_trans_release_chunk_metadata(trans); |
2255 | ||
0b246afa | 2256 | spin_lock(&fs_info->trans_lock); |
4a9d8bde | 2257 | cur_trans->state = TRANS_STATE_UNBLOCKED; |
0b246afa JM |
2258 | fs_info->running_transaction = NULL; |
2259 | spin_unlock(&fs_info->trans_lock); | |
2260 | mutex_unlock(&fs_info->reloc_mutex); | |
b7ec40d7 | 2261 | |
0b246afa | 2262 | wake_up(&fs_info->transaction_wait); |
e6dcd2dc | 2263 | |
70458a58 | 2264 | ret = btrfs_write_and_wait_transaction(trans); |
49b25e05 | 2265 | if (ret) { |
0b246afa JM |
2266 | btrfs_handle_fs_error(fs_info, ret, |
2267 | "Error while writing out transaction"); | |
2268 | mutex_unlock(&fs_info->tree_log_mutex); | |
6cf7f77e | 2269 | goto scrub_continue; |
49b25e05 JM |
2270 | } |
2271 | ||
eece6a9c | 2272 | ret = write_all_supers(fs_info, 0); |
e02119d5 CM |
2273 | /* |
2274 | * the super is written, we can safely allow the tree-loggers | |
2275 | * to go about their business | |
2276 | */ | |
0b246afa | 2277 | mutex_unlock(&fs_info->tree_log_mutex); |
c1f32b7c AJ |
2278 | if (ret) |
2279 | goto scrub_continue; | |
e02119d5 | 2280 | |
5ead2dd0 | 2281 | btrfs_finish_extent_commit(trans); |
4313b399 | 2282 | |
3204d33c | 2283 | if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS, &cur_trans->flags)) |
0b246afa | 2284 | btrfs_clear_space_info_full(fs_info); |
13212b54 | 2285 | |
0b246afa | 2286 | fs_info->last_trans_committed = cur_trans->transid; |
4a9d8bde MX |
2287 | /* |
2288 | * We needn't acquire the lock here because there is no other task | |
2289 | * which can change it. | |
2290 | */ | |
2291 | cur_trans->state = TRANS_STATE_COMPLETED; | |
2c90e5d6 | 2292 | wake_up(&cur_trans->commit_wait); |
a514d638 | 2293 | clear_bit(BTRFS_FS_NEED_ASYNC_COMMIT, &fs_info->flags); |
3de4586c | 2294 | |
0b246afa | 2295 | spin_lock(&fs_info->trans_lock); |
13c5a93e | 2296 | list_del_init(&cur_trans->list); |
0b246afa | 2297 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 | 2298 | |
724e2315 JB |
2299 | btrfs_put_transaction(cur_trans); |
2300 | btrfs_put_transaction(cur_trans); | |
58176a96 | 2301 | |
0860adfd | 2302 | if (trans->type & __TRANS_FREEZABLE) |
0b246afa | 2303 | sb_end_intwrite(fs_info->sb); |
b2b5ef5c | 2304 | |
3a45bb20 | 2305 | trace_btrfs_transaction_commit(trans->root); |
1abe9b8a | 2306 | |
2ff7e61e | 2307 | btrfs_scrub_continue(fs_info); |
a2de733c | 2308 | |
9ed74f2d JB |
2309 | if (current->journal_info == trans) |
2310 | current->journal_info = NULL; | |
2311 | ||
2c90e5d6 | 2312 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
24bbcf04 | 2313 | |
79154b1b | 2314 | return ret; |
49b25e05 | 2315 | |
6cf7f77e | 2316 | scrub_continue: |
2ff7e61e | 2317 | btrfs_scrub_continue(fs_info); |
49b25e05 | 2318 | cleanup_transaction: |
dc60c525 | 2319 | btrfs_trans_release_metadata(trans); |
c7cc64a9 | 2320 | btrfs_cleanup_pending_block_groups(trans); |
4fbcdf66 | 2321 | btrfs_trans_release_chunk_metadata(trans); |
0e721106 | 2322 | trans->block_rsv = NULL; |
0b246afa | 2323 | btrfs_warn(fs_info, "Skipping commit of aborted transaction."); |
49b25e05 JM |
2324 | if (current->journal_info == trans) |
2325 | current->journal_info = NULL; | |
97cb39bb | 2326 | cleanup_transaction(trans, ret); |
49b25e05 JM |
2327 | |
2328 | return ret; | |
79154b1b CM |
2329 | } |
2330 | ||
d352ac68 | 2331 | /* |
9d1a2a3a DS |
2332 | * return < 0 if error |
2333 | * 0 if there are no more dead_roots at the time of call | |
2334 | * 1 there are more to be processed, call me again | |
2335 | * | |
2336 | * The return value indicates there are certainly more snapshots to delete, but | |
2337 | * if there comes a new one during processing, it may return 0. We don't mind, | |
2338 | * because btrfs_commit_super will poke cleaner thread and it will process it a | |
2339 | * few seconds later. | |
d352ac68 | 2340 | */ |
9d1a2a3a | 2341 | int btrfs_clean_one_deleted_snapshot(struct btrfs_root *root) |
e9d0b13b | 2342 | { |
9d1a2a3a | 2343 | int ret; |
5d4f98a2 YZ |
2344 | struct btrfs_fs_info *fs_info = root->fs_info; |
2345 | ||
a4abeea4 | 2346 | spin_lock(&fs_info->trans_lock); |
9d1a2a3a DS |
2347 | if (list_empty(&fs_info->dead_roots)) { |
2348 | spin_unlock(&fs_info->trans_lock); | |
2349 | return 0; | |
2350 | } | |
2351 | root = list_first_entry(&fs_info->dead_roots, | |
2352 | struct btrfs_root, root_list); | |
cfad392b | 2353 | list_del_init(&root->root_list); |
a4abeea4 | 2354 | spin_unlock(&fs_info->trans_lock); |
e9d0b13b | 2355 | |
4fd786e6 | 2356 | btrfs_debug(fs_info, "cleaner removing %llu", root->root_key.objectid); |
76dda93c | 2357 | |
9d1a2a3a | 2358 | btrfs_kill_all_delayed_nodes(root); |
16cdcec7 | 2359 | |
9d1a2a3a DS |
2360 | if (btrfs_header_backref_rev(root->node) < |
2361 | BTRFS_MIXED_BACKREF_REV) | |
2362 | ret = btrfs_drop_snapshot(root, NULL, 0, 0); | |
2363 | else | |
2364 | ret = btrfs_drop_snapshot(root, NULL, 1, 0); | |
32471dc2 | 2365 | |
6596a928 | 2366 | return (ret < 0) ? 0 : 1; |
e9d0b13b | 2367 | } |
572d9ab7 DS |
2368 | |
2369 | void btrfs_apply_pending_changes(struct btrfs_fs_info *fs_info) | |
2370 | { | |
2371 | unsigned long prev; | |
2372 | unsigned long bit; | |
2373 | ||
6c9fe14f | 2374 | prev = xchg(&fs_info->pending_changes, 0); |
572d9ab7 DS |
2375 | if (!prev) |
2376 | return; | |
2377 | ||
7e1876ac DS |
2378 | bit = 1 << BTRFS_PENDING_SET_INODE_MAP_CACHE; |
2379 | if (prev & bit) | |
2380 | btrfs_set_opt(fs_info->mount_opt, INODE_MAP_CACHE); | |
2381 | prev &= ~bit; | |
2382 | ||
2383 | bit = 1 << BTRFS_PENDING_CLEAR_INODE_MAP_CACHE; | |
2384 | if (prev & bit) | |
2385 | btrfs_clear_opt(fs_info->mount_opt, INODE_MAP_CACHE); | |
2386 | prev &= ~bit; | |
2387 | ||
d51033d0 DS |
2388 | bit = 1 << BTRFS_PENDING_COMMIT; |
2389 | if (prev & bit) | |
2390 | btrfs_debug(fs_info, "pending commit done"); | |
2391 | prev &= ~bit; | |
2392 | ||
572d9ab7 DS |
2393 | if (prev) |
2394 | btrfs_warn(fs_info, | |
2395 | "unknown pending changes left 0x%lx, ignoring", prev); | |
2396 | } |