Commit | Line | Data |
---|---|---|
6cbd5570 CM |
1 | /* |
2 | * Copyright (C) 2007 Oracle. All rights reserved. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public | |
6 | * License v2 as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, | |
9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public | |
14 | * License along with this program; if not, write to the | |
15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
16 | * Boston, MA 021110-1307, USA. | |
17 | */ | |
18 | ||
79154b1b | 19 | #include <linux/fs.h> |
5a0e3ad6 | 20 | #include <linux/slab.h> |
34088780 | 21 | #include <linux/sched.h> |
d3c2fdcf | 22 | #include <linux/writeback.h> |
5f39d397 | 23 | #include <linux/pagemap.h> |
5f2cc086 | 24 | #include <linux/blkdev.h> |
8ea05e3a | 25 | #include <linux/uuid.h> |
79154b1b CM |
26 | #include "ctree.h" |
27 | #include "disk-io.h" | |
28 | #include "transaction.h" | |
925baedd | 29 | #include "locking.h" |
e02119d5 | 30 | #include "tree-log.h" |
581bb050 | 31 | #include "inode-map.h" |
733f4fbb | 32 | #include "volumes.h" |
8dabb742 | 33 | #include "dev-replace.h" |
79154b1b | 34 | |
0f7d52f4 CM |
35 | #define BTRFS_ROOT_TRANS_TAG 0 |
36 | ||
4a9d8bde MX |
37 | static unsigned int btrfs_blocked_trans_types[TRANS_STATE_MAX] = { |
38 | [TRANS_STATE_RUNNING] = 0U, | |
39 | [TRANS_STATE_BLOCKED] = (__TRANS_USERSPACE | | |
40 | __TRANS_START), | |
41 | [TRANS_STATE_COMMIT_START] = (__TRANS_USERSPACE | | |
42 | __TRANS_START | | |
43 | __TRANS_ATTACH), | |
44 | [TRANS_STATE_COMMIT_DOING] = (__TRANS_USERSPACE | | |
45 | __TRANS_START | | |
46 | __TRANS_ATTACH | | |
47 | __TRANS_JOIN), | |
48 | [TRANS_STATE_UNBLOCKED] = (__TRANS_USERSPACE | | |
49 | __TRANS_START | | |
50 | __TRANS_ATTACH | | |
51 | __TRANS_JOIN | | |
52 | __TRANS_JOIN_NOLOCK), | |
53 | [TRANS_STATE_COMPLETED] = (__TRANS_USERSPACE | | |
54 | __TRANS_START | | |
55 | __TRANS_ATTACH | | |
56 | __TRANS_JOIN | | |
57 | __TRANS_JOIN_NOLOCK), | |
58 | }; | |
59 | ||
724e2315 | 60 | void btrfs_put_transaction(struct btrfs_transaction *transaction) |
79154b1b | 61 | { |
13c5a93e JB |
62 | WARN_ON(atomic_read(&transaction->use_count) == 0); |
63 | if (atomic_dec_and_test(&transaction->use_count)) { | |
a4abeea4 | 64 | BUG_ON(!list_empty(&transaction->list)); |
c46effa6 | 65 | WARN_ON(!RB_EMPTY_ROOT(&transaction->delayed_refs.href_root)); |
6df9a95e JB |
66 | while (!list_empty(&transaction->pending_chunks)) { |
67 | struct extent_map *em; | |
68 | ||
69 | em = list_first_entry(&transaction->pending_chunks, | |
70 | struct extent_map, list); | |
71 | list_del_init(&em->list); | |
72 | free_extent_map(em); | |
73 | } | |
2c90e5d6 | 74 | kmem_cache_free(btrfs_transaction_cachep, transaction); |
78fae27e | 75 | } |
79154b1b CM |
76 | } |
77 | ||
9e351cc8 JB |
78 | static noinline void switch_commit_roots(struct btrfs_transaction *trans, |
79 | struct btrfs_fs_info *fs_info) | |
817d52f8 | 80 | { |
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); | |
89 | if (is_fstree(root->objectid)) | |
90 | btrfs_unpin_free_ino(root); | |
91 | } | |
92 | up_write(&fs_info->commit_root_sem); | |
817d52f8 JB |
93 | } |
94 | ||
0860adfd MX |
95 | static inline void extwriter_counter_inc(struct btrfs_transaction *trans, |
96 | unsigned int type) | |
97 | { | |
98 | if (type & TRANS_EXTWRITERS) | |
99 | atomic_inc(&trans->num_extwriters); | |
100 | } | |
101 | ||
102 | static inline void extwriter_counter_dec(struct btrfs_transaction *trans, | |
103 | unsigned int type) | |
104 | { | |
105 | if (type & TRANS_EXTWRITERS) | |
106 | atomic_dec(&trans->num_extwriters); | |
107 | } | |
108 | ||
109 | static inline void extwriter_counter_init(struct btrfs_transaction *trans, | |
110 | unsigned int type) | |
111 | { | |
112 | atomic_set(&trans->num_extwriters, ((type & TRANS_EXTWRITERS) ? 1 : 0)); | |
113 | } | |
114 | ||
115 | static inline int extwriter_counter_read(struct btrfs_transaction *trans) | |
116 | { | |
117 | return atomic_read(&trans->num_extwriters); | |
178260b2 MX |
118 | } |
119 | ||
d352ac68 CM |
120 | /* |
121 | * either allocate a new transaction or hop into the existing one | |
122 | */ | |
0860adfd | 123 | static noinline int join_transaction(struct btrfs_root *root, unsigned int type) |
79154b1b CM |
124 | { |
125 | struct btrfs_transaction *cur_trans; | |
19ae4e81 | 126 | struct btrfs_fs_info *fs_info = root->fs_info; |
a4abeea4 | 127 | |
19ae4e81 | 128 | spin_lock(&fs_info->trans_lock); |
d43317dc | 129 | loop: |
49b25e05 | 130 | /* The file system has been taken offline. No new transactions. */ |
87533c47 | 131 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { |
19ae4e81 | 132 | spin_unlock(&fs_info->trans_lock); |
49b25e05 JM |
133 | return -EROFS; |
134 | } | |
135 | ||
19ae4e81 | 136 | cur_trans = fs_info->running_transaction; |
a4abeea4 | 137 | if (cur_trans) { |
871383be | 138 | if (cur_trans->aborted) { |
19ae4e81 | 139 | spin_unlock(&fs_info->trans_lock); |
49b25e05 | 140 | return cur_trans->aborted; |
871383be | 141 | } |
4a9d8bde | 142 | if (btrfs_blocked_trans_types[cur_trans->state] & type) { |
178260b2 MX |
143 | spin_unlock(&fs_info->trans_lock); |
144 | return -EBUSY; | |
145 | } | |
a4abeea4 | 146 | atomic_inc(&cur_trans->use_count); |
13c5a93e | 147 | atomic_inc(&cur_trans->num_writers); |
0860adfd | 148 | extwriter_counter_inc(cur_trans, type); |
19ae4e81 | 149 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 | 150 | return 0; |
79154b1b | 151 | } |
19ae4e81 | 152 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 | 153 | |
354aa0fb MX |
154 | /* |
155 | * If we are ATTACH, we just want to catch the current transaction, | |
156 | * and commit it. If there is no transaction, just return ENOENT. | |
157 | */ | |
158 | if (type == TRANS_ATTACH) | |
159 | return -ENOENT; | |
160 | ||
4a9d8bde MX |
161 | /* |
162 | * JOIN_NOLOCK only happens during the transaction commit, so | |
163 | * it is impossible that ->running_transaction is NULL | |
164 | */ | |
165 | BUG_ON(type == TRANS_JOIN_NOLOCK); | |
166 | ||
a4abeea4 JB |
167 | cur_trans = kmem_cache_alloc(btrfs_transaction_cachep, GFP_NOFS); |
168 | if (!cur_trans) | |
169 | return -ENOMEM; | |
d43317dc | 170 | |
19ae4e81 JS |
171 | spin_lock(&fs_info->trans_lock); |
172 | if (fs_info->running_transaction) { | |
d43317dc CM |
173 | /* |
174 | * someone started a transaction after we unlocked. Make sure | |
4a9d8bde | 175 | * to redo the checks above |
d43317dc | 176 | */ |
a4abeea4 | 177 | kmem_cache_free(btrfs_transaction_cachep, cur_trans); |
d43317dc | 178 | goto loop; |
87533c47 | 179 | } else if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { |
e4b50e14 | 180 | spin_unlock(&fs_info->trans_lock); |
7b8b92af JB |
181 | kmem_cache_free(btrfs_transaction_cachep, cur_trans); |
182 | return -EROFS; | |
79154b1b | 183 | } |
d43317dc | 184 | |
a4abeea4 | 185 | atomic_set(&cur_trans->num_writers, 1); |
0860adfd | 186 | extwriter_counter_init(cur_trans, type); |
a4abeea4 JB |
187 | init_waitqueue_head(&cur_trans->writer_wait); |
188 | init_waitqueue_head(&cur_trans->commit_wait); | |
4a9d8bde | 189 | cur_trans->state = TRANS_STATE_RUNNING; |
a4abeea4 JB |
190 | /* |
191 | * One for this trans handle, one so it will live on until we | |
192 | * commit the transaction. | |
193 | */ | |
194 | atomic_set(&cur_trans->use_count, 2); | |
a4abeea4 JB |
195 | cur_trans->start_time = get_seconds(); |
196 | ||
c46effa6 | 197 | cur_trans->delayed_refs.href_root = RB_ROOT; |
d7df2c79 | 198 | atomic_set(&cur_trans->delayed_refs.num_entries, 0); |
a4abeea4 JB |
199 | cur_trans->delayed_refs.num_heads_ready = 0; |
200 | cur_trans->delayed_refs.num_heads = 0; | |
201 | cur_trans->delayed_refs.flushing = 0; | |
202 | cur_trans->delayed_refs.run_delayed_start = 0; | |
20b297d6 JS |
203 | |
204 | /* | |
205 | * although the tree mod log is per file system and not per transaction, | |
206 | * the log must never go across transaction boundaries. | |
207 | */ | |
208 | smp_mb(); | |
31b1a2bd | 209 | if (!list_empty(&fs_info->tree_mod_seq_list)) |
efe120a0 | 210 | WARN(1, KERN_ERR "BTRFS: tree_mod_seq_list not empty when " |
20b297d6 | 211 | "creating a fresh transaction\n"); |
31b1a2bd | 212 | if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log)) |
efe120a0 | 213 | WARN(1, KERN_ERR "BTRFS: tree_mod_log rb tree not empty when " |
20b297d6 | 214 | "creating a fresh transaction\n"); |
fc36ed7e | 215 | atomic64_set(&fs_info->tree_mod_seq, 0); |
20b297d6 | 216 | |
a4abeea4 JB |
217 | spin_lock_init(&cur_trans->delayed_refs.lock); |
218 | ||
219 | INIT_LIST_HEAD(&cur_trans->pending_snapshots); | |
569e0f35 | 220 | INIT_LIST_HEAD(&cur_trans->ordered_operations); |
6df9a95e | 221 | INIT_LIST_HEAD(&cur_trans->pending_chunks); |
9e351cc8 | 222 | INIT_LIST_HEAD(&cur_trans->switch_commits); |
19ae4e81 | 223 | list_add_tail(&cur_trans->list, &fs_info->trans_list); |
a4abeea4 | 224 | extent_io_tree_init(&cur_trans->dirty_pages, |
19ae4e81 JS |
225 | fs_info->btree_inode->i_mapping); |
226 | fs_info->generation++; | |
227 | cur_trans->transid = fs_info->generation; | |
228 | fs_info->running_transaction = cur_trans; | |
49b25e05 | 229 | cur_trans->aborted = 0; |
19ae4e81 | 230 | spin_unlock(&fs_info->trans_lock); |
15ee9bc7 | 231 | |
79154b1b CM |
232 | return 0; |
233 | } | |
234 | ||
d352ac68 | 235 | /* |
d397712b CM |
236 | * this does all the record keeping required to make sure that a reference |
237 | * counted root is properly recorded in a given transaction. This is required | |
238 | * to make sure the old root from before we joined the transaction is deleted | |
239 | * when the transaction commits | |
d352ac68 | 240 | */ |
7585717f | 241 | static int record_root_in_trans(struct btrfs_trans_handle *trans, |
a4abeea4 | 242 | struct btrfs_root *root) |
6702ed49 | 243 | { |
27cdeb70 MX |
244 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
245 | root->last_trans < trans->transid) { | |
6702ed49 | 246 | WARN_ON(root == root->fs_info->extent_root); |
5d4f98a2 YZ |
247 | WARN_ON(root->commit_root != root->node); |
248 | ||
7585717f | 249 | /* |
27cdeb70 | 250 | * see below for IN_TRANS_SETUP usage rules |
7585717f CM |
251 | * we have the reloc mutex held now, so there |
252 | * is only one writer in this function | |
253 | */ | |
27cdeb70 | 254 | set_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); |
7585717f | 255 | |
27cdeb70 | 256 | /* make sure readers find IN_TRANS_SETUP before |
7585717f CM |
257 | * they find our root->last_trans update |
258 | */ | |
259 | smp_wmb(); | |
260 | ||
a4abeea4 JB |
261 | spin_lock(&root->fs_info->fs_roots_radix_lock); |
262 | if (root->last_trans == trans->transid) { | |
263 | spin_unlock(&root->fs_info->fs_roots_radix_lock); | |
264 | return 0; | |
265 | } | |
5d4f98a2 YZ |
266 | radix_tree_tag_set(&root->fs_info->fs_roots_radix, |
267 | (unsigned long)root->root_key.objectid, | |
268 | BTRFS_ROOT_TRANS_TAG); | |
a4abeea4 | 269 | spin_unlock(&root->fs_info->fs_roots_radix_lock); |
7585717f CM |
270 | root->last_trans = trans->transid; |
271 | ||
272 | /* this is pretty tricky. We don't want to | |
273 | * take the relocation lock in btrfs_record_root_in_trans | |
274 | * unless we're really doing the first setup for this root in | |
275 | * this transaction. | |
276 | * | |
277 | * Normally we'd use root->last_trans as a flag to decide | |
278 | * if we want to take the expensive mutex. | |
279 | * | |
280 | * But, we have to set root->last_trans before we | |
281 | * init the relocation root, otherwise, we trip over warnings | |
282 | * in ctree.c. The solution used here is to flag ourselves | |
27cdeb70 | 283 | * with root IN_TRANS_SETUP. When this is 1, we're still |
7585717f CM |
284 | * fixing up the reloc trees and everyone must wait. |
285 | * | |
286 | * When this is zero, they can trust root->last_trans and fly | |
287 | * through btrfs_record_root_in_trans without having to take the | |
288 | * lock. smp_wmb() makes sure that all the writes above are | |
289 | * done before we pop in the zero below | |
290 | */ | |
5d4f98a2 | 291 | btrfs_init_reloc_root(trans, root); |
27cdeb70 MX |
292 | smp_mb__before_clear_bit(); |
293 | clear_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); | |
5d4f98a2 YZ |
294 | } |
295 | return 0; | |
296 | } | |
bcc63abb | 297 | |
7585717f CM |
298 | |
299 | int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, | |
300 | struct btrfs_root *root) | |
301 | { | |
27cdeb70 | 302 | if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state)) |
7585717f CM |
303 | return 0; |
304 | ||
305 | /* | |
27cdeb70 | 306 | * see record_root_in_trans for comments about IN_TRANS_SETUP usage |
7585717f CM |
307 | * and barriers |
308 | */ | |
309 | smp_rmb(); | |
310 | if (root->last_trans == trans->transid && | |
27cdeb70 | 311 | !test_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state)) |
7585717f CM |
312 | return 0; |
313 | ||
314 | mutex_lock(&root->fs_info->reloc_mutex); | |
315 | record_root_in_trans(trans, root); | |
316 | mutex_unlock(&root->fs_info->reloc_mutex); | |
317 | ||
318 | return 0; | |
319 | } | |
320 | ||
4a9d8bde MX |
321 | static inline int is_transaction_blocked(struct btrfs_transaction *trans) |
322 | { | |
323 | return (trans->state >= TRANS_STATE_BLOCKED && | |
501407aa JB |
324 | trans->state < TRANS_STATE_UNBLOCKED && |
325 | !trans->aborted); | |
4a9d8bde MX |
326 | } |
327 | ||
d352ac68 CM |
328 | /* wait for commit against the current transaction to become unblocked |
329 | * when this is done, it is safe to start a new transaction, but the current | |
330 | * transaction might not be fully on disk. | |
331 | */ | |
37d1aeee | 332 | static void wait_current_trans(struct btrfs_root *root) |
79154b1b | 333 | { |
f9295749 | 334 | struct btrfs_transaction *cur_trans; |
79154b1b | 335 | |
a4abeea4 | 336 | spin_lock(&root->fs_info->trans_lock); |
f9295749 | 337 | cur_trans = root->fs_info->running_transaction; |
4a9d8bde | 338 | if (cur_trans && is_transaction_blocked(cur_trans)) { |
13c5a93e | 339 | atomic_inc(&cur_trans->use_count); |
a4abeea4 | 340 | spin_unlock(&root->fs_info->trans_lock); |
72d63ed6 LZ |
341 | |
342 | wait_event(root->fs_info->transaction_wait, | |
501407aa JB |
343 | cur_trans->state >= TRANS_STATE_UNBLOCKED || |
344 | cur_trans->aborted); | |
724e2315 | 345 | btrfs_put_transaction(cur_trans); |
a4abeea4 JB |
346 | } else { |
347 | spin_unlock(&root->fs_info->trans_lock); | |
f9295749 | 348 | } |
37d1aeee CM |
349 | } |
350 | ||
a22285a6 YZ |
351 | static int may_wait_transaction(struct btrfs_root *root, int type) |
352 | { | |
a4abeea4 JB |
353 | if (root->fs_info->log_root_recovering) |
354 | return 0; | |
355 | ||
356 | if (type == TRANS_USERSPACE) | |
357 | return 1; | |
358 | ||
359 | if (type == TRANS_START && | |
360 | !atomic_read(&root->fs_info->open_ioctl_trans)) | |
a22285a6 | 361 | return 1; |
a4abeea4 | 362 | |
a22285a6 YZ |
363 | return 0; |
364 | } | |
365 | ||
20dd2cbf MX |
366 | static inline bool need_reserve_reloc_root(struct btrfs_root *root) |
367 | { | |
368 | if (!root->fs_info->reloc_ctl || | |
27cdeb70 | 369 | !test_bit(BTRFS_ROOT_REF_COWS, &root->state) || |
20dd2cbf MX |
370 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || |
371 | root->reloc_root) | |
372 | return false; | |
373 | ||
374 | return true; | |
375 | } | |
376 | ||
08e007d2 | 377 | static struct btrfs_trans_handle * |
0860adfd | 378 | start_transaction(struct btrfs_root *root, u64 num_items, unsigned int type, |
08e007d2 | 379 | enum btrfs_reserve_flush_enum flush) |
37d1aeee | 380 | { |
a22285a6 YZ |
381 | struct btrfs_trans_handle *h; |
382 | struct btrfs_transaction *cur_trans; | |
b5009945 | 383 | u64 num_bytes = 0; |
c5567237 | 384 | u64 qgroup_reserved = 0; |
20dd2cbf MX |
385 | bool reloc_reserved = false; |
386 | int ret; | |
acce952b | 387 | |
87533c47 | 388 | if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) |
acce952b | 389 | return ERR_PTR(-EROFS); |
2a1eb461 | 390 | |
a26e8c9f JB |
391 | if (current->journal_info && |
392 | current->journal_info != (void *)BTRFS_SEND_TRANS_STUB) { | |
0860adfd | 393 | WARN_ON(type & TRANS_EXTWRITERS); |
2a1eb461 JB |
394 | h = current->journal_info; |
395 | h->use_count++; | |
b7d5b0a8 | 396 | WARN_ON(h->use_count > 2); |
2a1eb461 JB |
397 | h->orig_rsv = h->block_rsv; |
398 | h->block_rsv = NULL; | |
399 | goto got_it; | |
400 | } | |
b5009945 JB |
401 | |
402 | /* | |
403 | * Do the reservation before we join the transaction so we can do all | |
404 | * the appropriate flushing if need be. | |
405 | */ | |
406 | if (num_items > 0 && root != root->fs_info->chunk_root) { | |
c5567237 AJ |
407 | if (root->fs_info->quota_enabled && |
408 | is_fstree(root->root_key.objectid)) { | |
409 | qgroup_reserved = num_items * root->leafsize; | |
410 | ret = btrfs_qgroup_reserve(root, qgroup_reserved); | |
411 | if (ret) | |
412 | return ERR_PTR(ret); | |
413 | } | |
414 | ||
b5009945 | 415 | num_bytes = btrfs_calc_trans_metadata_size(root, num_items); |
20dd2cbf MX |
416 | /* |
417 | * Do the reservation for the relocation root creation | |
418 | */ | |
419 | if (unlikely(need_reserve_reloc_root(root))) { | |
420 | num_bytes += root->nodesize; | |
421 | reloc_reserved = true; | |
422 | } | |
423 | ||
08e007d2 MX |
424 | ret = btrfs_block_rsv_add(root, |
425 | &root->fs_info->trans_block_rsv, | |
426 | num_bytes, flush); | |
b5009945 | 427 | if (ret) |
843fcf35 | 428 | goto reserve_fail; |
b5009945 | 429 | } |
a22285a6 YZ |
430 | again: |
431 | h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS); | |
843fcf35 MX |
432 | if (!h) { |
433 | ret = -ENOMEM; | |
434 | goto alloc_fail; | |
435 | } | |
37d1aeee | 436 | |
98114659 JB |
437 | /* |
438 | * If we are JOIN_NOLOCK we're already committing a transaction and | |
439 | * waiting on this guy, so we don't need to do the sb_start_intwrite | |
440 | * because we're already holding a ref. We need this because we could | |
441 | * have raced in and did an fsync() on a file which can kick a commit | |
442 | * and then we deadlock with somebody doing a freeze. | |
354aa0fb MX |
443 | * |
444 | * If we are ATTACH, it means we just want to catch the current | |
445 | * transaction and commit it, so we needn't do sb_start_intwrite(). | |
98114659 | 446 | */ |
0860adfd | 447 | if (type & __TRANS_FREEZABLE) |
60376ce4 | 448 | sb_start_intwrite(root->fs_info->sb); |
b2b5ef5c | 449 | |
a22285a6 | 450 | if (may_wait_transaction(root, type)) |
37d1aeee | 451 | wait_current_trans(root); |
a22285a6 | 452 | |
a4abeea4 | 453 | do { |
354aa0fb | 454 | ret = join_transaction(root, type); |
178260b2 | 455 | if (ret == -EBUSY) { |
a4abeea4 | 456 | wait_current_trans(root); |
178260b2 MX |
457 | if (unlikely(type == TRANS_ATTACH)) |
458 | ret = -ENOENT; | |
459 | } | |
a4abeea4 JB |
460 | } while (ret == -EBUSY); |
461 | ||
db5b493a | 462 | if (ret < 0) { |
354aa0fb MX |
463 | /* We must get the transaction if we are JOIN_NOLOCK. */ |
464 | BUG_ON(type == TRANS_JOIN_NOLOCK); | |
843fcf35 | 465 | goto join_fail; |
db5b493a | 466 | } |
0f7d52f4 | 467 | |
a22285a6 | 468 | cur_trans = root->fs_info->running_transaction; |
a22285a6 YZ |
469 | |
470 | h->transid = cur_trans->transid; | |
471 | h->transaction = cur_trans; | |
79154b1b | 472 | h->blocks_used = 0; |
a22285a6 | 473 | h->bytes_reserved = 0; |
d13603ef | 474 | h->root = root; |
56bec294 | 475 | h->delayed_ref_updates = 0; |
2a1eb461 | 476 | h->use_count = 1; |
0e721106 | 477 | h->adding_csums = 0; |
f0486c68 | 478 | h->block_rsv = NULL; |
2a1eb461 | 479 | h->orig_rsv = NULL; |
49b25e05 | 480 | h->aborted = 0; |
4b824906 | 481 | h->qgroup_reserved = 0; |
bed92eae | 482 | h->delayed_ref_elem.seq = 0; |
a698d075 | 483 | h->type = type; |
c6b305a8 | 484 | h->allocating_chunk = false; |
20dd2cbf | 485 | h->reloc_reserved = false; |
5039eddc | 486 | h->sync = false; |
bed92eae | 487 | INIT_LIST_HEAD(&h->qgroup_ref_list); |
ea658bad | 488 | INIT_LIST_HEAD(&h->new_bgs); |
b7ec40d7 | 489 | |
a22285a6 | 490 | smp_mb(); |
4a9d8bde MX |
491 | if (cur_trans->state >= TRANS_STATE_BLOCKED && |
492 | may_wait_transaction(root, type)) { | |
a22285a6 YZ |
493 | btrfs_commit_transaction(h, root); |
494 | goto again; | |
495 | } | |
496 | ||
b5009945 | 497 | if (num_bytes) { |
8c2a3ca2 | 498 | trace_btrfs_space_reservation(root->fs_info, "transaction", |
2bcc0328 | 499 | h->transid, num_bytes, 1); |
b5009945 JB |
500 | h->block_rsv = &root->fs_info->trans_block_rsv; |
501 | h->bytes_reserved = num_bytes; | |
20dd2cbf | 502 | h->reloc_reserved = reloc_reserved; |
a22285a6 | 503 | } |
4b824906 | 504 | h->qgroup_reserved = qgroup_reserved; |
9ed74f2d | 505 | |
2a1eb461 | 506 | got_it: |
a4abeea4 | 507 | btrfs_record_root_in_trans(h, root); |
a22285a6 YZ |
508 | |
509 | if (!current->journal_info && type != TRANS_USERSPACE) | |
510 | current->journal_info = h; | |
79154b1b | 511 | return h; |
843fcf35 MX |
512 | |
513 | join_fail: | |
0860adfd | 514 | if (type & __TRANS_FREEZABLE) |
843fcf35 MX |
515 | sb_end_intwrite(root->fs_info->sb); |
516 | kmem_cache_free(btrfs_trans_handle_cachep, h); | |
517 | alloc_fail: | |
518 | if (num_bytes) | |
519 | btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv, | |
520 | num_bytes); | |
521 | reserve_fail: | |
522 | if (qgroup_reserved) | |
523 | btrfs_qgroup_free(root, qgroup_reserved); | |
524 | return ERR_PTR(ret); | |
79154b1b CM |
525 | } |
526 | ||
f9295749 | 527 | struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, |
a22285a6 | 528 | int num_items) |
f9295749 | 529 | { |
08e007d2 MX |
530 | return start_transaction(root, num_items, TRANS_START, |
531 | BTRFS_RESERVE_FLUSH_ALL); | |
f9295749 | 532 | } |
8407aa46 | 533 | |
08e007d2 | 534 | struct btrfs_trans_handle *btrfs_start_transaction_lflush( |
8407aa46 MX |
535 | struct btrfs_root *root, int num_items) |
536 | { | |
08e007d2 MX |
537 | return start_transaction(root, num_items, TRANS_START, |
538 | BTRFS_RESERVE_FLUSH_LIMIT); | |
8407aa46 MX |
539 | } |
540 | ||
7a7eaa40 | 541 | struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root) |
f9295749 | 542 | { |
8407aa46 | 543 | return start_transaction(root, 0, TRANS_JOIN, 0); |
f9295749 CM |
544 | } |
545 | ||
7a7eaa40 | 546 | struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root) |
0af3d00b | 547 | { |
8407aa46 | 548 | return start_transaction(root, 0, TRANS_JOIN_NOLOCK, 0); |
0af3d00b JB |
549 | } |
550 | ||
7a7eaa40 | 551 | struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root) |
9ca9ee09 | 552 | { |
8407aa46 | 553 | return start_transaction(root, 0, TRANS_USERSPACE, 0); |
9ca9ee09 SW |
554 | } |
555 | ||
d4edf39b MX |
556 | /* |
557 | * btrfs_attach_transaction() - catch the running transaction | |
558 | * | |
559 | * It is used when we want to commit the current the transaction, but | |
560 | * don't want to start a new one. | |
561 | * | |
562 | * Note: If this function return -ENOENT, it just means there is no | |
563 | * running transaction. But it is possible that the inactive transaction | |
564 | * is still in the memory, not fully on disk. If you hope there is no | |
565 | * inactive transaction in the fs when -ENOENT is returned, you should | |
566 | * invoke | |
567 | * btrfs_attach_transaction_barrier() | |
568 | */ | |
354aa0fb | 569 | struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root) |
60376ce4 | 570 | { |
354aa0fb | 571 | return start_transaction(root, 0, TRANS_ATTACH, 0); |
60376ce4 JB |
572 | } |
573 | ||
d4edf39b | 574 | /* |
90b6d283 | 575 | * btrfs_attach_transaction_barrier() - catch the running transaction |
d4edf39b MX |
576 | * |
577 | * It is similar to the above function, the differentia is this one | |
578 | * will wait for all the inactive transactions until they fully | |
579 | * complete. | |
580 | */ | |
581 | struct btrfs_trans_handle * | |
582 | btrfs_attach_transaction_barrier(struct btrfs_root *root) | |
583 | { | |
584 | struct btrfs_trans_handle *trans; | |
585 | ||
586 | trans = start_transaction(root, 0, TRANS_ATTACH, 0); | |
587 | if (IS_ERR(trans) && PTR_ERR(trans) == -ENOENT) | |
588 | btrfs_wait_for_commit(root, 0); | |
589 | ||
590 | return trans; | |
591 | } | |
592 | ||
d352ac68 | 593 | /* wait for a transaction commit to be fully complete */ |
b9c8300c | 594 | static noinline void wait_for_commit(struct btrfs_root *root, |
89ce8a63 CM |
595 | struct btrfs_transaction *commit) |
596 | { | |
4a9d8bde | 597 | wait_event(commit->commit_wait, commit->state == TRANS_STATE_COMPLETED); |
89ce8a63 CM |
598 | } |
599 | ||
46204592 SW |
600 | int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid) |
601 | { | |
602 | struct btrfs_transaction *cur_trans = NULL, *t; | |
8cd2807f | 603 | int ret = 0; |
46204592 | 604 | |
46204592 SW |
605 | if (transid) { |
606 | if (transid <= root->fs_info->last_trans_committed) | |
a4abeea4 | 607 | goto out; |
46204592 | 608 | |
8cd2807f | 609 | ret = -EINVAL; |
46204592 | 610 | /* find specified transaction */ |
a4abeea4 | 611 | spin_lock(&root->fs_info->trans_lock); |
46204592 SW |
612 | list_for_each_entry(t, &root->fs_info->trans_list, list) { |
613 | if (t->transid == transid) { | |
614 | cur_trans = t; | |
a4abeea4 | 615 | atomic_inc(&cur_trans->use_count); |
8cd2807f | 616 | ret = 0; |
46204592 SW |
617 | break; |
618 | } | |
8cd2807f MX |
619 | if (t->transid > transid) { |
620 | ret = 0; | |
46204592 | 621 | break; |
8cd2807f | 622 | } |
46204592 | 623 | } |
a4abeea4 | 624 | spin_unlock(&root->fs_info->trans_lock); |
8cd2807f | 625 | /* The specified transaction doesn't exist */ |
46204592 | 626 | if (!cur_trans) |
8cd2807f | 627 | goto out; |
46204592 SW |
628 | } else { |
629 | /* find newest transaction that is committing | committed */ | |
a4abeea4 | 630 | spin_lock(&root->fs_info->trans_lock); |
46204592 SW |
631 | list_for_each_entry_reverse(t, &root->fs_info->trans_list, |
632 | list) { | |
4a9d8bde MX |
633 | if (t->state >= TRANS_STATE_COMMIT_START) { |
634 | if (t->state == TRANS_STATE_COMPLETED) | |
3473f3c0 | 635 | break; |
46204592 | 636 | cur_trans = t; |
a4abeea4 | 637 | atomic_inc(&cur_trans->use_count); |
46204592 SW |
638 | break; |
639 | } | |
640 | } | |
a4abeea4 | 641 | spin_unlock(&root->fs_info->trans_lock); |
46204592 | 642 | if (!cur_trans) |
a4abeea4 | 643 | goto out; /* nothing committing|committed */ |
46204592 SW |
644 | } |
645 | ||
46204592 | 646 | wait_for_commit(root, cur_trans); |
724e2315 | 647 | btrfs_put_transaction(cur_trans); |
a4abeea4 | 648 | out: |
46204592 SW |
649 | return ret; |
650 | } | |
651 | ||
37d1aeee CM |
652 | void btrfs_throttle(struct btrfs_root *root) |
653 | { | |
a4abeea4 | 654 | if (!atomic_read(&root->fs_info->open_ioctl_trans)) |
9ca9ee09 | 655 | wait_current_trans(root); |
37d1aeee CM |
656 | } |
657 | ||
8929ecfa YZ |
658 | static int should_end_transaction(struct btrfs_trans_handle *trans, |
659 | struct btrfs_root *root) | |
660 | { | |
1be41b78 | 661 | if (root->fs_info->global_block_rsv.space_info->full && |
0a2b2a84 | 662 | btrfs_check_space_for_delayed_refs(trans, root)) |
1be41b78 | 663 | return 1; |
36ba022a | 664 | |
1be41b78 | 665 | return !!btrfs_block_rsv_check(root, &root->fs_info->global_block_rsv, 5); |
8929ecfa YZ |
666 | } |
667 | ||
668 | int btrfs_should_end_transaction(struct btrfs_trans_handle *trans, | |
669 | struct btrfs_root *root) | |
670 | { | |
671 | struct btrfs_transaction *cur_trans = trans->transaction; | |
672 | int updates; | |
49b25e05 | 673 | int err; |
8929ecfa | 674 | |
a4abeea4 | 675 | smp_mb(); |
4a9d8bde MX |
676 | if (cur_trans->state >= TRANS_STATE_BLOCKED || |
677 | cur_trans->delayed_refs.flushing) | |
8929ecfa YZ |
678 | return 1; |
679 | ||
680 | updates = trans->delayed_ref_updates; | |
681 | trans->delayed_ref_updates = 0; | |
49b25e05 JM |
682 | if (updates) { |
683 | err = btrfs_run_delayed_refs(trans, root, updates); | |
684 | if (err) /* Error code will also eval true */ | |
685 | return err; | |
686 | } | |
8929ecfa YZ |
687 | |
688 | return should_end_transaction(trans, root); | |
689 | } | |
690 | ||
89ce8a63 | 691 | static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, |
a698d075 | 692 | struct btrfs_root *root, int throttle) |
79154b1b | 693 | { |
8929ecfa | 694 | struct btrfs_transaction *cur_trans = trans->transaction; |
ab78c84d | 695 | struct btrfs_fs_info *info = root->fs_info; |
1be41b78 | 696 | unsigned long cur = trans->delayed_ref_updates; |
a698d075 | 697 | int lock = (trans->type != TRANS_JOIN_NOLOCK); |
4edc2ca3 | 698 | int err = 0; |
c3e69d58 | 699 | |
3bbb24b2 JB |
700 | if (trans->use_count > 1) { |
701 | trans->use_count--; | |
2a1eb461 JB |
702 | trans->block_rsv = trans->orig_rsv; |
703 | return 0; | |
704 | } | |
705 | ||
edf39272 JS |
706 | /* |
707 | * do the qgroup accounting as early as possible | |
708 | */ | |
709 | err = btrfs_delayed_refs_qgroup_accounting(trans, info); | |
710 | ||
b24e03db | 711 | btrfs_trans_release_metadata(trans, root); |
4c13d758 | 712 | trans->block_rsv = NULL; |
c5567237 AJ |
713 | |
714 | if (trans->qgroup_reserved) { | |
7c2ec3f0 LB |
715 | /* |
716 | * the same root has to be passed here between start_transaction | |
717 | * and end_transaction. Subvolume quota depends on this. | |
718 | */ | |
719 | btrfs_qgroup_free(trans->root, trans->qgroup_reserved); | |
c5567237 AJ |
720 | trans->qgroup_reserved = 0; |
721 | } | |
722 | ||
ea658bad JB |
723 | if (!list_empty(&trans->new_bgs)) |
724 | btrfs_create_pending_block_groups(trans, root); | |
725 | ||
1be41b78 | 726 | trans->delayed_ref_updates = 0; |
5039eddc | 727 | if (!trans->sync && btrfs_should_throttle_delayed_refs(trans, root)) { |
0a2b2a84 | 728 | cur = max_t(unsigned long, cur, 32); |
c3e69d58 | 729 | trans->delayed_ref_updates = 0; |
1be41b78 | 730 | btrfs_run_delayed_refs(trans, root, cur); |
56bec294 | 731 | } |
bb721703 | 732 | |
0e721106 JB |
733 | btrfs_trans_release_metadata(trans, root); |
734 | trans->block_rsv = NULL; | |
56bec294 | 735 | |
ea658bad JB |
736 | if (!list_empty(&trans->new_bgs)) |
737 | btrfs_create_pending_block_groups(trans, root); | |
738 | ||
a4abeea4 | 739 | if (lock && !atomic_read(&root->fs_info->open_ioctl_trans) && |
4a9d8bde MX |
740 | should_end_transaction(trans, root) && |
741 | ACCESS_ONCE(cur_trans->state) == TRANS_STATE_RUNNING) { | |
742 | spin_lock(&info->trans_lock); | |
743 | if (cur_trans->state == TRANS_STATE_RUNNING) | |
744 | cur_trans->state = TRANS_STATE_BLOCKED; | |
745 | spin_unlock(&info->trans_lock); | |
a4abeea4 | 746 | } |
8929ecfa | 747 | |
4a9d8bde | 748 | if (lock && ACCESS_ONCE(cur_trans->state) == TRANS_STATE_BLOCKED) { |
3bbb24b2 | 749 | if (throttle) |
8929ecfa | 750 | return btrfs_commit_transaction(trans, root); |
3bbb24b2 | 751 | else |
8929ecfa YZ |
752 | wake_up_process(info->transaction_kthread); |
753 | } | |
754 | ||
0860adfd | 755 | if (trans->type & __TRANS_FREEZABLE) |
98114659 | 756 | sb_end_intwrite(root->fs_info->sb); |
6df7881a | 757 | |
8929ecfa | 758 | WARN_ON(cur_trans != info->running_transaction); |
13c5a93e JB |
759 | WARN_ON(atomic_read(&cur_trans->num_writers) < 1); |
760 | atomic_dec(&cur_trans->num_writers); | |
0860adfd | 761 | extwriter_counter_dec(cur_trans, trans->type); |
89ce8a63 | 762 | |
99d16cbc | 763 | smp_mb(); |
79154b1b CM |
764 | if (waitqueue_active(&cur_trans->writer_wait)) |
765 | wake_up(&cur_trans->writer_wait); | |
724e2315 | 766 | btrfs_put_transaction(cur_trans); |
9ed74f2d JB |
767 | |
768 | if (current->journal_info == trans) | |
769 | current->journal_info = NULL; | |
ab78c84d | 770 | |
24bbcf04 YZ |
771 | if (throttle) |
772 | btrfs_run_delayed_iputs(root); | |
773 | ||
49b25e05 | 774 | if (trans->aborted || |
4e121c06 JB |
775 | test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) { |
776 | wake_up_process(info->transaction_kthread); | |
4edc2ca3 | 777 | err = -EIO; |
4e121c06 | 778 | } |
edf39272 | 779 | assert_qgroups_uptodate(trans); |
49b25e05 | 780 | |
4edc2ca3 DJ |
781 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
782 | return err; | |
79154b1b CM |
783 | } |
784 | ||
89ce8a63 CM |
785 | int btrfs_end_transaction(struct btrfs_trans_handle *trans, |
786 | struct btrfs_root *root) | |
787 | { | |
98ad43be | 788 | return __btrfs_end_transaction(trans, root, 0); |
89ce8a63 CM |
789 | } |
790 | ||
791 | int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans, | |
792 | struct btrfs_root *root) | |
793 | { | |
98ad43be | 794 | return __btrfs_end_transaction(trans, root, 1); |
16cdcec7 MX |
795 | } |
796 | ||
d352ac68 CM |
797 | /* |
798 | * when btree blocks are allocated, they have some corresponding bits set for | |
799 | * them in one of two extent_io trees. This is used to make sure all of | |
690587d1 | 800 | * those extents are sent to disk but does not wait on them |
d352ac68 | 801 | */ |
690587d1 | 802 | int btrfs_write_marked_extents(struct btrfs_root *root, |
8cef4e16 | 803 | struct extent_io_tree *dirty_pages, int mark) |
79154b1b | 804 | { |
777e6bd7 | 805 | int err = 0; |
7c4452b9 | 806 | int werr = 0; |
1728366e | 807 | struct address_space *mapping = root->fs_info->btree_inode->i_mapping; |
e6138876 | 808 | struct extent_state *cached_state = NULL; |
777e6bd7 | 809 | u64 start = 0; |
5f39d397 | 810 | u64 end; |
7c4452b9 | 811 | |
1728366e | 812 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
e6138876 JB |
813 | mark, &cached_state)) { |
814 | convert_extent_bit(dirty_pages, start, end, EXTENT_NEED_WAIT, | |
815 | mark, &cached_state, GFP_NOFS); | |
816 | cached_state = NULL; | |
1728366e JB |
817 | err = filemap_fdatawrite_range(mapping, start, end); |
818 | if (err) | |
819 | werr = err; | |
820 | cond_resched(); | |
821 | start = end + 1; | |
7c4452b9 | 822 | } |
690587d1 CM |
823 | if (err) |
824 | werr = err; | |
825 | return werr; | |
826 | } | |
827 | ||
828 | /* | |
829 | * when btree blocks are allocated, they have some corresponding bits set for | |
830 | * them in one of two extent_io trees. This is used to make sure all of | |
831 | * those extents are on disk for transaction or log commit. We wait | |
832 | * on all the pages and clear them from the dirty pages state tree | |
833 | */ | |
834 | int btrfs_wait_marked_extents(struct btrfs_root *root, | |
8cef4e16 | 835 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 | 836 | { |
690587d1 CM |
837 | int err = 0; |
838 | int werr = 0; | |
1728366e | 839 | struct address_space *mapping = root->fs_info->btree_inode->i_mapping; |
e6138876 | 840 | struct extent_state *cached_state = NULL; |
690587d1 CM |
841 | u64 start = 0; |
842 | u64 end; | |
777e6bd7 | 843 | |
1728366e | 844 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
e6138876 JB |
845 | EXTENT_NEED_WAIT, &cached_state)) { |
846 | clear_extent_bit(dirty_pages, start, end, EXTENT_NEED_WAIT, | |
847 | 0, 0, &cached_state, GFP_NOFS); | |
1728366e JB |
848 | err = filemap_fdatawait_range(mapping, start, end); |
849 | if (err) | |
850 | werr = err; | |
851 | cond_resched(); | |
852 | start = end + 1; | |
777e6bd7 | 853 | } |
7c4452b9 CM |
854 | if (err) |
855 | werr = err; | |
856 | return werr; | |
79154b1b CM |
857 | } |
858 | ||
690587d1 CM |
859 | /* |
860 | * when btree blocks are allocated, they have some corresponding bits set for | |
861 | * them in one of two extent_io trees. This is used to make sure all of | |
862 | * those extents are on disk for transaction or log commit | |
863 | */ | |
171170c1 | 864 | static int btrfs_write_and_wait_marked_extents(struct btrfs_root *root, |
8cef4e16 | 865 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 CM |
866 | { |
867 | int ret; | |
868 | int ret2; | |
c6adc9cc | 869 | struct blk_plug plug; |
690587d1 | 870 | |
c6adc9cc | 871 | blk_start_plug(&plug); |
8cef4e16 | 872 | ret = btrfs_write_marked_extents(root, dirty_pages, mark); |
c6adc9cc | 873 | blk_finish_plug(&plug); |
8cef4e16 | 874 | ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark); |
bf0da8c1 CM |
875 | |
876 | if (ret) | |
877 | return ret; | |
878 | if (ret2) | |
879 | return ret2; | |
880 | return 0; | |
690587d1 CM |
881 | } |
882 | ||
d0c803c4 CM |
883 | int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans, |
884 | struct btrfs_root *root) | |
885 | { | |
886 | if (!trans || !trans->transaction) { | |
887 | struct inode *btree_inode; | |
888 | btree_inode = root->fs_info->btree_inode; | |
889 | return filemap_write_and_wait(btree_inode->i_mapping); | |
890 | } | |
891 | return btrfs_write_and_wait_marked_extents(root, | |
8cef4e16 YZ |
892 | &trans->transaction->dirty_pages, |
893 | EXTENT_DIRTY); | |
d0c803c4 CM |
894 | } |
895 | ||
d352ac68 CM |
896 | /* |
897 | * this is used to update the root pointer in the tree of tree roots. | |
898 | * | |
899 | * But, in the case of the extent allocation tree, updating the root | |
900 | * pointer may allocate blocks which may change the root of the extent | |
901 | * allocation tree. | |
902 | * | |
903 | * So, this loops and repeats and makes sure the cowonly root didn't | |
904 | * change while the root pointer was being updated in the metadata. | |
905 | */ | |
0b86a832 CM |
906 | static int update_cowonly_root(struct btrfs_trans_handle *trans, |
907 | struct btrfs_root *root) | |
79154b1b CM |
908 | { |
909 | int ret; | |
0b86a832 | 910 | u64 old_root_bytenr; |
86b9f2ec | 911 | u64 old_root_used; |
0b86a832 | 912 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
79154b1b | 913 | |
86b9f2ec | 914 | old_root_used = btrfs_root_used(&root->root_item); |
0b86a832 | 915 | btrfs_write_dirty_block_groups(trans, root); |
56bec294 | 916 | |
d397712b | 917 | while (1) { |
0b86a832 | 918 | old_root_bytenr = btrfs_root_bytenr(&root->root_item); |
86b9f2ec YZ |
919 | if (old_root_bytenr == root->node->start && |
920 | old_root_used == btrfs_root_used(&root->root_item)) | |
79154b1b | 921 | break; |
87ef2bb4 | 922 | |
5d4f98a2 | 923 | btrfs_set_root_node(&root->root_item, root->node); |
79154b1b | 924 | ret = btrfs_update_root(trans, tree_root, |
0b86a832 CM |
925 | &root->root_key, |
926 | &root->root_item); | |
49b25e05 JM |
927 | if (ret) |
928 | return ret; | |
56bec294 | 929 | |
86b9f2ec | 930 | old_root_used = btrfs_root_used(&root->root_item); |
4a8c9a62 | 931 | ret = btrfs_write_dirty_block_groups(trans, root); |
49b25e05 JM |
932 | if (ret) |
933 | return ret; | |
0b86a832 | 934 | } |
276e680d | 935 | |
0b86a832 CM |
936 | return 0; |
937 | } | |
938 | ||
d352ac68 CM |
939 | /* |
940 | * update all the cowonly tree roots on disk | |
49b25e05 JM |
941 | * |
942 | * The error handling in this function may not be obvious. Any of the | |
943 | * failures will cause the file system to go offline. We still need | |
944 | * to clean up the delayed refs. | |
d352ac68 | 945 | */ |
5d4f98a2 YZ |
946 | static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans, |
947 | struct btrfs_root *root) | |
0b86a832 CM |
948 | { |
949 | struct btrfs_fs_info *fs_info = root->fs_info; | |
950 | struct list_head *next; | |
84234f3a | 951 | struct extent_buffer *eb; |
56bec294 | 952 | int ret; |
84234f3a | 953 | |
56bec294 | 954 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
49b25e05 JM |
955 | if (ret) |
956 | return ret; | |
87ef2bb4 | 957 | |
84234f3a | 958 | eb = btrfs_lock_root_node(fs_info->tree_root); |
49b25e05 JM |
959 | ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, |
960 | 0, &eb); | |
84234f3a YZ |
961 | btrfs_tree_unlock(eb); |
962 | free_extent_buffer(eb); | |
0b86a832 | 963 | |
49b25e05 JM |
964 | if (ret) |
965 | return ret; | |
966 | ||
56bec294 | 967 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
49b25e05 JM |
968 | if (ret) |
969 | return ret; | |
87ef2bb4 | 970 | |
733f4fbb | 971 | ret = btrfs_run_dev_stats(trans, root->fs_info); |
c16ce190 JB |
972 | if (ret) |
973 | return ret; | |
8dabb742 | 974 | ret = btrfs_run_dev_replace(trans, root->fs_info); |
c16ce190 JB |
975 | if (ret) |
976 | return ret; | |
546adb0d | 977 | ret = btrfs_run_qgroups(trans, root->fs_info); |
c16ce190 JB |
978 | if (ret) |
979 | return ret; | |
546adb0d JS |
980 | |
981 | /* run_qgroups might have added some more refs */ | |
982 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); | |
c16ce190 JB |
983 | if (ret) |
984 | return ret; | |
546adb0d | 985 | |
d397712b | 986 | while (!list_empty(&fs_info->dirty_cowonly_roots)) { |
0b86a832 CM |
987 | next = fs_info->dirty_cowonly_roots.next; |
988 | list_del_init(next); | |
989 | root = list_entry(next, struct btrfs_root, dirty_list); | |
87ef2bb4 | 990 | |
9e351cc8 JB |
991 | if (root != fs_info->extent_root) |
992 | list_add_tail(&root->dirty_list, | |
993 | &trans->transaction->switch_commits); | |
49b25e05 JM |
994 | ret = update_cowonly_root(trans, root); |
995 | if (ret) | |
996 | return ret; | |
79154b1b | 997 | } |
276e680d | 998 | |
9e351cc8 JB |
999 | list_add_tail(&fs_info->extent_root->dirty_list, |
1000 | &trans->transaction->switch_commits); | |
8dabb742 SB |
1001 | btrfs_after_dev_replace_commit(fs_info); |
1002 | ||
79154b1b CM |
1003 | return 0; |
1004 | } | |
1005 | ||
d352ac68 CM |
1006 | /* |
1007 | * dead roots are old snapshots that need to be deleted. This allocates | |
1008 | * a dirty root struct and adds it into the list of dead roots that need to | |
1009 | * be deleted | |
1010 | */ | |
cfad392b | 1011 | void btrfs_add_dead_root(struct btrfs_root *root) |
5eda7b5e | 1012 | { |
a4abeea4 | 1013 | spin_lock(&root->fs_info->trans_lock); |
cfad392b JB |
1014 | if (list_empty(&root->root_list)) |
1015 | list_add_tail(&root->root_list, &root->fs_info->dead_roots); | |
a4abeea4 | 1016 | spin_unlock(&root->fs_info->trans_lock); |
5eda7b5e CM |
1017 | } |
1018 | ||
d352ac68 | 1019 | /* |
5d4f98a2 | 1020 | * update all the cowonly tree roots on disk |
d352ac68 | 1021 | */ |
5d4f98a2 YZ |
1022 | static noinline int commit_fs_roots(struct btrfs_trans_handle *trans, |
1023 | struct btrfs_root *root) | |
0f7d52f4 | 1024 | { |
0f7d52f4 | 1025 | struct btrfs_root *gang[8]; |
5d4f98a2 | 1026 | struct btrfs_fs_info *fs_info = root->fs_info; |
0f7d52f4 CM |
1027 | int i; |
1028 | int ret; | |
54aa1f4d CM |
1029 | int err = 0; |
1030 | ||
a4abeea4 | 1031 | spin_lock(&fs_info->fs_roots_radix_lock); |
d397712b | 1032 | while (1) { |
5d4f98a2 YZ |
1033 | ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, |
1034 | (void **)gang, 0, | |
0f7d52f4 CM |
1035 | ARRAY_SIZE(gang), |
1036 | BTRFS_ROOT_TRANS_TAG); | |
1037 | if (ret == 0) | |
1038 | break; | |
1039 | for (i = 0; i < ret; i++) { | |
1040 | root = gang[i]; | |
5d4f98a2 YZ |
1041 | radix_tree_tag_clear(&fs_info->fs_roots_radix, |
1042 | (unsigned long)root->root_key.objectid, | |
1043 | BTRFS_ROOT_TRANS_TAG); | |
a4abeea4 | 1044 | spin_unlock(&fs_info->fs_roots_radix_lock); |
31153d81 | 1045 | |
e02119d5 | 1046 | btrfs_free_log(trans, root); |
5d4f98a2 | 1047 | btrfs_update_reloc_root(trans, root); |
d68fc57b | 1048 | btrfs_orphan_commit_root(trans, root); |
bcc63abb | 1049 | |
82d5902d LZ |
1050 | btrfs_save_ino_cache(root, trans); |
1051 | ||
f1ebcc74 | 1052 | /* see comments in should_cow_block() */ |
27cdeb70 MX |
1053 | clear_bit(BTRFS_ROOT_FORCE_COW, &root->state); |
1054 | smp_mb__after_clear_bit(); | |
f1ebcc74 | 1055 | |
978d910d | 1056 | if (root->commit_root != root->node) { |
9e351cc8 JB |
1057 | list_add_tail(&root->dirty_list, |
1058 | &trans->transaction->switch_commits); | |
978d910d YZ |
1059 | btrfs_set_root_node(&root->root_item, |
1060 | root->node); | |
1061 | } | |
5d4f98a2 | 1062 | |
5d4f98a2 | 1063 | err = btrfs_update_root(trans, fs_info->tree_root, |
0f7d52f4 CM |
1064 | &root->root_key, |
1065 | &root->root_item); | |
a4abeea4 | 1066 | spin_lock(&fs_info->fs_roots_radix_lock); |
54aa1f4d CM |
1067 | if (err) |
1068 | break; | |
0f7d52f4 CM |
1069 | } |
1070 | } | |
a4abeea4 | 1071 | spin_unlock(&fs_info->fs_roots_radix_lock); |
54aa1f4d | 1072 | return err; |
0f7d52f4 CM |
1073 | } |
1074 | ||
d352ac68 | 1075 | /* |
de78b51a ES |
1076 | * defrag a given btree. |
1077 | * Every leaf in the btree is read and defragged. | |
d352ac68 | 1078 | */ |
de78b51a | 1079 | int btrfs_defrag_root(struct btrfs_root *root) |
e9d0b13b CM |
1080 | { |
1081 | struct btrfs_fs_info *info = root->fs_info; | |
e9d0b13b | 1082 | struct btrfs_trans_handle *trans; |
8929ecfa | 1083 | int ret; |
e9d0b13b | 1084 | |
27cdeb70 | 1085 | if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state)) |
e9d0b13b | 1086 | return 0; |
8929ecfa | 1087 | |
6b80053d | 1088 | while (1) { |
8929ecfa YZ |
1089 | trans = btrfs_start_transaction(root, 0); |
1090 | if (IS_ERR(trans)) | |
1091 | return PTR_ERR(trans); | |
1092 | ||
de78b51a | 1093 | ret = btrfs_defrag_leaves(trans, root); |
8929ecfa | 1094 | |
e9d0b13b | 1095 | btrfs_end_transaction(trans, root); |
b53d3f5d | 1096 | btrfs_btree_balance_dirty(info->tree_root); |
e9d0b13b CM |
1097 | cond_resched(); |
1098 | ||
7841cb28 | 1099 | if (btrfs_fs_closing(root->fs_info) || ret != -EAGAIN) |
e9d0b13b | 1100 | break; |
210549eb DS |
1101 | |
1102 | if (btrfs_defrag_cancelled(root->fs_info)) { | |
efe120a0 | 1103 | pr_debug("BTRFS: defrag_root cancelled\n"); |
210549eb DS |
1104 | ret = -EAGAIN; |
1105 | break; | |
1106 | } | |
e9d0b13b | 1107 | } |
27cdeb70 | 1108 | clear_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state); |
8929ecfa | 1109 | return ret; |
e9d0b13b CM |
1110 | } |
1111 | ||
d352ac68 CM |
1112 | /* |
1113 | * new snapshots need to be created at a very specific time in the | |
aec8030a MX |
1114 | * transaction commit. This does the actual creation. |
1115 | * | |
1116 | * Note: | |
1117 | * If the error which may affect the commitment of the current transaction | |
1118 | * happens, we should return the error number. If the error which just affect | |
1119 | * the creation of the pending snapshots, just return 0. | |
d352ac68 | 1120 | */ |
80b6794d | 1121 | static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, |
3063d29f CM |
1122 | struct btrfs_fs_info *fs_info, |
1123 | struct btrfs_pending_snapshot *pending) | |
1124 | { | |
1125 | struct btrfs_key key; | |
80b6794d | 1126 | struct btrfs_root_item *new_root_item; |
3063d29f CM |
1127 | struct btrfs_root *tree_root = fs_info->tree_root; |
1128 | struct btrfs_root *root = pending->root; | |
6bdb72de | 1129 | struct btrfs_root *parent_root; |
98c9942a | 1130 | struct btrfs_block_rsv *rsv; |
6bdb72de | 1131 | struct inode *parent_inode; |
42874b3d MX |
1132 | struct btrfs_path *path; |
1133 | struct btrfs_dir_item *dir_item; | |
a22285a6 | 1134 | struct dentry *dentry; |
3063d29f | 1135 | struct extent_buffer *tmp; |
925baedd | 1136 | struct extent_buffer *old; |
8ea05e3a | 1137 | struct timespec cur_time = CURRENT_TIME; |
aec8030a | 1138 | int ret = 0; |
d68fc57b | 1139 | u64 to_reserve = 0; |
6bdb72de | 1140 | u64 index = 0; |
a22285a6 | 1141 | u64 objectid; |
b83cc969 | 1142 | u64 root_flags; |
8ea05e3a | 1143 | uuid_le new_uuid; |
3063d29f | 1144 | |
42874b3d MX |
1145 | path = btrfs_alloc_path(); |
1146 | if (!path) { | |
aec8030a MX |
1147 | pending->error = -ENOMEM; |
1148 | return 0; | |
42874b3d MX |
1149 | } |
1150 | ||
80b6794d CM |
1151 | new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS); |
1152 | if (!new_root_item) { | |
aec8030a | 1153 | pending->error = -ENOMEM; |
6fa9700e | 1154 | goto root_item_alloc_fail; |
80b6794d | 1155 | } |
a22285a6 | 1156 | |
aec8030a MX |
1157 | pending->error = btrfs_find_free_objectid(tree_root, &objectid); |
1158 | if (pending->error) | |
6fa9700e | 1159 | goto no_free_objectid; |
3063d29f | 1160 | |
3fd0a558 | 1161 | btrfs_reloc_pre_snapshot(trans, pending, &to_reserve); |
d68fc57b YZ |
1162 | |
1163 | if (to_reserve > 0) { | |
aec8030a MX |
1164 | pending->error = btrfs_block_rsv_add(root, |
1165 | &pending->block_rsv, | |
1166 | to_reserve, | |
1167 | BTRFS_RESERVE_NO_FLUSH); | |
1168 | if (pending->error) | |
6fa9700e | 1169 | goto no_free_objectid; |
d68fc57b YZ |
1170 | } |
1171 | ||
aec8030a MX |
1172 | pending->error = btrfs_qgroup_inherit(trans, fs_info, |
1173 | root->root_key.objectid, | |
1174 | objectid, pending->inherit); | |
1175 | if (pending->error) | |
6fa9700e | 1176 | goto no_free_objectid; |
6f72c7e2 | 1177 | |
3063d29f | 1178 | key.objectid = objectid; |
a22285a6 YZ |
1179 | key.offset = (u64)-1; |
1180 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3063d29f | 1181 | |
6fa9700e | 1182 | rsv = trans->block_rsv; |
a22285a6 | 1183 | trans->block_rsv = &pending->block_rsv; |
2382c5cc | 1184 | trans->bytes_reserved = trans->block_rsv->reserved; |
3de4586c | 1185 | |
a22285a6 | 1186 | dentry = pending->dentry; |
e9662f70 | 1187 | parent_inode = pending->dir; |
a22285a6 | 1188 | parent_root = BTRFS_I(parent_inode)->root; |
7585717f | 1189 | record_root_in_trans(trans, parent_root); |
a22285a6 | 1190 | |
3063d29f CM |
1191 | /* |
1192 | * insert the directory item | |
1193 | */ | |
3de4586c | 1194 | ret = btrfs_set_inode_index(parent_inode, &index); |
49b25e05 | 1195 | BUG_ON(ret); /* -ENOMEM */ |
42874b3d MX |
1196 | |
1197 | /* check if there is a file/dir which has the same name. */ | |
1198 | dir_item = btrfs_lookup_dir_item(NULL, parent_root, path, | |
1199 | btrfs_ino(parent_inode), | |
1200 | dentry->d_name.name, | |
1201 | dentry->d_name.len, 0); | |
1202 | if (dir_item != NULL && !IS_ERR(dir_item)) { | |
fe66a05a | 1203 | pending->error = -EEXIST; |
aec8030a | 1204 | goto dir_item_existed; |
42874b3d MX |
1205 | } else if (IS_ERR(dir_item)) { |
1206 | ret = PTR_ERR(dir_item); | |
8732d44f MX |
1207 | btrfs_abort_transaction(trans, root, ret); |
1208 | goto fail; | |
79787eaa | 1209 | } |
42874b3d | 1210 | btrfs_release_path(path); |
52c26179 | 1211 | |
e999376f CM |
1212 | /* |
1213 | * pull in the delayed directory update | |
1214 | * and the delayed inode item | |
1215 | * otherwise we corrupt the FS during | |
1216 | * snapshot | |
1217 | */ | |
1218 | ret = btrfs_run_delayed_items(trans, root); | |
8732d44f MX |
1219 | if (ret) { /* Transaction aborted */ |
1220 | btrfs_abort_transaction(trans, root, ret); | |
1221 | goto fail; | |
1222 | } | |
e999376f | 1223 | |
7585717f | 1224 | record_root_in_trans(trans, root); |
6bdb72de SW |
1225 | btrfs_set_root_last_snapshot(&root->root_item, trans->transid); |
1226 | memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); | |
08fe4db1 | 1227 | btrfs_check_and_init_root_item(new_root_item); |
6bdb72de | 1228 | |
b83cc969 LZ |
1229 | root_flags = btrfs_root_flags(new_root_item); |
1230 | if (pending->readonly) | |
1231 | root_flags |= BTRFS_ROOT_SUBVOL_RDONLY; | |
1232 | else | |
1233 | root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY; | |
1234 | btrfs_set_root_flags(new_root_item, root_flags); | |
1235 | ||
8ea05e3a AB |
1236 | btrfs_set_root_generation_v2(new_root_item, |
1237 | trans->transid); | |
1238 | uuid_le_gen(&new_uuid); | |
1239 | memcpy(new_root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE); | |
1240 | memcpy(new_root_item->parent_uuid, root->root_item.uuid, | |
1241 | BTRFS_UUID_SIZE); | |
70023da2 SB |
1242 | if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) { |
1243 | memset(new_root_item->received_uuid, 0, | |
1244 | sizeof(new_root_item->received_uuid)); | |
1245 | memset(&new_root_item->stime, 0, sizeof(new_root_item->stime)); | |
1246 | memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime)); | |
1247 | btrfs_set_root_stransid(new_root_item, 0); | |
1248 | btrfs_set_root_rtransid(new_root_item, 0); | |
1249 | } | |
3cae210f QW |
1250 | btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec); |
1251 | btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec); | |
8ea05e3a | 1252 | btrfs_set_root_otransid(new_root_item, trans->transid); |
8ea05e3a | 1253 | |
6bdb72de | 1254 | old = btrfs_lock_root_node(root); |
49b25e05 | 1255 | ret = btrfs_cow_block(trans, root, old, NULL, 0, &old); |
79787eaa JM |
1256 | if (ret) { |
1257 | btrfs_tree_unlock(old); | |
1258 | free_extent_buffer(old); | |
8732d44f MX |
1259 | btrfs_abort_transaction(trans, root, ret); |
1260 | goto fail; | |
79787eaa | 1261 | } |
49b25e05 | 1262 | |
6bdb72de SW |
1263 | btrfs_set_lock_blocking(old); |
1264 | ||
49b25e05 | 1265 | ret = btrfs_copy_root(trans, root, old, &tmp, objectid); |
79787eaa | 1266 | /* clean up in any case */ |
6bdb72de SW |
1267 | btrfs_tree_unlock(old); |
1268 | free_extent_buffer(old); | |
8732d44f MX |
1269 | if (ret) { |
1270 | btrfs_abort_transaction(trans, root, ret); | |
1271 | goto fail; | |
1272 | } | |
6bdb72de | 1273 | |
f1ebcc74 | 1274 | /* see comments in should_cow_block() */ |
27cdeb70 | 1275 | set_bit(BTRFS_ROOT_FORCE_COW, &root->state); |
f1ebcc74 LB |
1276 | smp_wmb(); |
1277 | ||
6bdb72de | 1278 | btrfs_set_root_node(new_root_item, tmp); |
a22285a6 YZ |
1279 | /* record when the snapshot was created in key.offset */ |
1280 | key.offset = trans->transid; | |
1281 | ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); | |
6bdb72de SW |
1282 | btrfs_tree_unlock(tmp); |
1283 | free_extent_buffer(tmp); | |
8732d44f MX |
1284 | if (ret) { |
1285 | btrfs_abort_transaction(trans, root, ret); | |
1286 | goto fail; | |
1287 | } | |
6bdb72de | 1288 | |
a22285a6 YZ |
1289 | /* |
1290 | * insert root back/forward references | |
1291 | */ | |
1292 | ret = btrfs_add_root_ref(trans, tree_root, objectid, | |
0660b5af | 1293 | parent_root->root_key.objectid, |
33345d01 | 1294 | btrfs_ino(parent_inode), index, |
a22285a6 | 1295 | dentry->d_name.name, dentry->d_name.len); |
8732d44f MX |
1296 | if (ret) { |
1297 | btrfs_abort_transaction(trans, root, ret); | |
1298 | goto fail; | |
1299 | } | |
0660b5af | 1300 | |
a22285a6 YZ |
1301 | key.offset = (u64)-1; |
1302 | pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key); | |
79787eaa JM |
1303 | if (IS_ERR(pending->snap)) { |
1304 | ret = PTR_ERR(pending->snap); | |
8732d44f MX |
1305 | btrfs_abort_transaction(trans, root, ret); |
1306 | goto fail; | |
79787eaa | 1307 | } |
d68fc57b | 1308 | |
49b25e05 | 1309 | ret = btrfs_reloc_post_snapshot(trans, pending); |
8732d44f MX |
1310 | if (ret) { |
1311 | btrfs_abort_transaction(trans, root, ret); | |
1312 | goto fail; | |
1313 | } | |
361048f5 MX |
1314 | |
1315 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); | |
8732d44f MX |
1316 | if (ret) { |
1317 | btrfs_abort_transaction(trans, root, ret); | |
1318 | goto fail; | |
1319 | } | |
42874b3d MX |
1320 | |
1321 | ret = btrfs_insert_dir_item(trans, parent_root, | |
1322 | dentry->d_name.name, dentry->d_name.len, | |
1323 | parent_inode, &key, | |
1324 | BTRFS_FT_DIR, index); | |
1325 | /* We have check then name at the beginning, so it is impossible. */ | |
9c52057c | 1326 | BUG_ON(ret == -EEXIST || ret == -EOVERFLOW); |
8732d44f MX |
1327 | if (ret) { |
1328 | btrfs_abort_transaction(trans, root, ret); | |
1329 | goto fail; | |
1330 | } | |
42874b3d MX |
1331 | |
1332 | btrfs_i_size_write(parent_inode, parent_inode->i_size + | |
1333 | dentry->d_name.len * 2); | |
1334 | parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME; | |
be6aef60 | 1335 | ret = btrfs_update_inode_fallback(trans, parent_root, parent_inode); |
dd5f9615 SB |
1336 | if (ret) { |
1337 | btrfs_abort_transaction(trans, root, ret); | |
1338 | goto fail; | |
1339 | } | |
1340 | ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root, new_uuid.b, | |
1341 | BTRFS_UUID_KEY_SUBVOL, objectid); | |
1342 | if (ret) { | |
8732d44f | 1343 | btrfs_abort_transaction(trans, root, ret); |
dd5f9615 SB |
1344 | goto fail; |
1345 | } | |
1346 | if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) { | |
1347 | ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root, | |
1348 | new_root_item->received_uuid, | |
1349 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, | |
1350 | objectid); | |
1351 | if (ret && ret != -EEXIST) { | |
1352 | btrfs_abort_transaction(trans, root, ret); | |
1353 | goto fail; | |
1354 | } | |
1355 | } | |
3063d29f | 1356 | fail: |
aec8030a MX |
1357 | pending->error = ret; |
1358 | dir_item_existed: | |
98c9942a | 1359 | trans->block_rsv = rsv; |
2382c5cc | 1360 | trans->bytes_reserved = 0; |
6fa9700e MX |
1361 | no_free_objectid: |
1362 | kfree(new_root_item); | |
1363 | root_item_alloc_fail: | |
42874b3d | 1364 | btrfs_free_path(path); |
49b25e05 | 1365 | return ret; |
3063d29f CM |
1366 | } |
1367 | ||
d352ac68 CM |
1368 | /* |
1369 | * create all the snapshots we've scheduled for creation | |
1370 | */ | |
80b6794d CM |
1371 | static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans, |
1372 | struct btrfs_fs_info *fs_info) | |
3de4586c | 1373 | { |
aec8030a | 1374 | struct btrfs_pending_snapshot *pending, *next; |
3de4586c | 1375 | struct list_head *head = &trans->transaction->pending_snapshots; |
aec8030a | 1376 | int ret = 0; |
3de4586c | 1377 | |
aec8030a MX |
1378 | list_for_each_entry_safe(pending, next, head, list) { |
1379 | list_del(&pending->list); | |
1380 | ret = create_pending_snapshot(trans, fs_info, pending); | |
1381 | if (ret) | |
1382 | break; | |
1383 | } | |
1384 | return ret; | |
3de4586c CM |
1385 | } |
1386 | ||
5d4f98a2 YZ |
1387 | static void update_super_roots(struct btrfs_root *root) |
1388 | { | |
1389 | struct btrfs_root_item *root_item; | |
1390 | struct btrfs_super_block *super; | |
1391 | ||
6c41761f | 1392 | super = root->fs_info->super_copy; |
5d4f98a2 YZ |
1393 | |
1394 | root_item = &root->fs_info->chunk_root->root_item; | |
1395 | super->chunk_root = root_item->bytenr; | |
1396 | super->chunk_root_generation = root_item->generation; | |
1397 | super->chunk_root_level = root_item->level; | |
1398 | ||
1399 | root_item = &root->fs_info->tree_root->root_item; | |
1400 | super->root = root_item->bytenr; | |
1401 | super->generation = root_item->generation; | |
1402 | super->root_level = root_item->level; | |
73bc1876 | 1403 | if (btrfs_test_opt(root, SPACE_CACHE)) |
0af3d00b | 1404 | super->cache_generation = root_item->generation; |
70f80175 SB |
1405 | if (root->fs_info->update_uuid_tree_gen) |
1406 | super->uuid_tree_generation = root_item->generation; | |
5d4f98a2 YZ |
1407 | } |
1408 | ||
f36f3042 CM |
1409 | int btrfs_transaction_in_commit(struct btrfs_fs_info *info) |
1410 | { | |
4a9d8bde | 1411 | struct btrfs_transaction *trans; |
f36f3042 | 1412 | int ret = 0; |
4a9d8bde | 1413 | |
a4abeea4 | 1414 | spin_lock(&info->trans_lock); |
4a9d8bde MX |
1415 | trans = info->running_transaction; |
1416 | if (trans) | |
1417 | ret = (trans->state >= TRANS_STATE_COMMIT_START); | |
a4abeea4 | 1418 | spin_unlock(&info->trans_lock); |
f36f3042 CM |
1419 | return ret; |
1420 | } | |
1421 | ||
8929ecfa YZ |
1422 | int btrfs_transaction_blocked(struct btrfs_fs_info *info) |
1423 | { | |
4a9d8bde | 1424 | struct btrfs_transaction *trans; |
8929ecfa | 1425 | int ret = 0; |
4a9d8bde | 1426 | |
a4abeea4 | 1427 | spin_lock(&info->trans_lock); |
4a9d8bde MX |
1428 | trans = info->running_transaction; |
1429 | if (trans) | |
1430 | ret = is_transaction_blocked(trans); | |
a4abeea4 | 1431 | spin_unlock(&info->trans_lock); |
8929ecfa YZ |
1432 | return ret; |
1433 | } | |
1434 | ||
bb9c12c9 SW |
1435 | /* |
1436 | * wait for the current transaction commit to start and block subsequent | |
1437 | * transaction joins | |
1438 | */ | |
1439 | static void wait_current_trans_commit_start(struct btrfs_root *root, | |
1440 | struct btrfs_transaction *trans) | |
1441 | { | |
4a9d8bde | 1442 | wait_event(root->fs_info->transaction_blocked_wait, |
501407aa JB |
1443 | trans->state >= TRANS_STATE_COMMIT_START || |
1444 | trans->aborted); | |
bb9c12c9 SW |
1445 | } |
1446 | ||
1447 | /* | |
1448 | * wait for the current transaction to start and then become unblocked. | |
1449 | * caller holds ref. | |
1450 | */ | |
1451 | static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root, | |
1452 | struct btrfs_transaction *trans) | |
1453 | { | |
72d63ed6 | 1454 | wait_event(root->fs_info->transaction_wait, |
501407aa JB |
1455 | trans->state >= TRANS_STATE_UNBLOCKED || |
1456 | trans->aborted); | |
bb9c12c9 SW |
1457 | } |
1458 | ||
1459 | /* | |
1460 | * commit transactions asynchronously. once btrfs_commit_transaction_async | |
1461 | * returns, any subsequent transaction will not be allowed to join. | |
1462 | */ | |
1463 | struct btrfs_async_commit { | |
1464 | struct btrfs_trans_handle *newtrans; | |
1465 | struct btrfs_root *root; | |
7892b5af | 1466 | struct work_struct work; |
bb9c12c9 SW |
1467 | }; |
1468 | ||
1469 | static void do_async_commit(struct work_struct *work) | |
1470 | { | |
1471 | struct btrfs_async_commit *ac = | |
7892b5af | 1472 | container_of(work, struct btrfs_async_commit, work); |
bb9c12c9 | 1473 | |
6fc4e354 SW |
1474 | /* |
1475 | * We've got freeze protection passed with the transaction. | |
1476 | * Tell lockdep about it. | |
1477 | */ | |
b1a06a4b | 1478 | if (ac->newtrans->type & __TRANS_FREEZABLE) |
ff7c1d33 MX |
1479 | rwsem_acquire_read( |
1480 | &ac->root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1], | |
1481 | 0, 1, _THIS_IP_); | |
6fc4e354 | 1482 | |
e209db7a SW |
1483 | current->journal_info = ac->newtrans; |
1484 | ||
bb9c12c9 SW |
1485 | btrfs_commit_transaction(ac->newtrans, ac->root); |
1486 | kfree(ac); | |
1487 | } | |
1488 | ||
1489 | int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans, | |
1490 | struct btrfs_root *root, | |
1491 | int wait_for_unblock) | |
1492 | { | |
1493 | struct btrfs_async_commit *ac; | |
1494 | struct btrfs_transaction *cur_trans; | |
1495 | ||
1496 | ac = kmalloc(sizeof(*ac), GFP_NOFS); | |
db5b493a TI |
1497 | if (!ac) |
1498 | return -ENOMEM; | |
bb9c12c9 | 1499 | |
7892b5af | 1500 | INIT_WORK(&ac->work, do_async_commit); |
bb9c12c9 | 1501 | ac->root = root; |
7a7eaa40 | 1502 | ac->newtrans = btrfs_join_transaction(root); |
3612b495 TI |
1503 | if (IS_ERR(ac->newtrans)) { |
1504 | int err = PTR_ERR(ac->newtrans); | |
1505 | kfree(ac); | |
1506 | return err; | |
1507 | } | |
bb9c12c9 SW |
1508 | |
1509 | /* take transaction reference */ | |
bb9c12c9 | 1510 | cur_trans = trans->transaction; |
13c5a93e | 1511 | atomic_inc(&cur_trans->use_count); |
bb9c12c9 SW |
1512 | |
1513 | btrfs_end_transaction(trans, root); | |
6fc4e354 SW |
1514 | |
1515 | /* | |
1516 | * Tell lockdep we've released the freeze rwsem, since the | |
1517 | * async commit thread will be the one to unlock it. | |
1518 | */ | |
b1a06a4b | 1519 | if (ac->newtrans->type & __TRANS_FREEZABLE) |
ff7c1d33 MX |
1520 | rwsem_release( |
1521 | &root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1], | |
1522 | 1, _THIS_IP_); | |
6fc4e354 | 1523 | |
7892b5af | 1524 | schedule_work(&ac->work); |
bb9c12c9 SW |
1525 | |
1526 | /* wait for transaction to start and unblock */ | |
bb9c12c9 SW |
1527 | if (wait_for_unblock) |
1528 | wait_current_trans_commit_start_and_unblock(root, cur_trans); | |
1529 | else | |
1530 | wait_current_trans_commit_start(root, cur_trans); | |
bb9c12c9 | 1531 | |
38e88054 SW |
1532 | if (current->journal_info == trans) |
1533 | current->journal_info = NULL; | |
1534 | ||
724e2315 | 1535 | btrfs_put_transaction(cur_trans); |
bb9c12c9 SW |
1536 | return 0; |
1537 | } | |
1538 | ||
49b25e05 JM |
1539 | |
1540 | static void cleanup_transaction(struct btrfs_trans_handle *trans, | |
7b8b92af | 1541 | struct btrfs_root *root, int err) |
49b25e05 JM |
1542 | { |
1543 | struct btrfs_transaction *cur_trans = trans->transaction; | |
f094ac32 | 1544 | DEFINE_WAIT(wait); |
49b25e05 JM |
1545 | |
1546 | WARN_ON(trans->use_count > 1); | |
1547 | ||
7b8b92af JB |
1548 | btrfs_abort_transaction(trans, root, err); |
1549 | ||
49b25e05 | 1550 | spin_lock(&root->fs_info->trans_lock); |
66b6135b | 1551 | |
25d8c284 MX |
1552 | /* |
1553 | * If the transaction is removed from the list, it means this | |
1554 | * transaction has been committed successfully, so it is impossible | |
1555 | * to call the cleanup function. | |
1556 | */ | |
1557 | BUG_ON(list_empty(&cur_trans->list)); | |
66b6135b | 1558 | |
49b25e05 | 1559 | list_del_init(&cur_trans->list); |
d7096fc3 | 1560 | if (cur_trans == root->fs_info->running_transaction) { |
4a9d8bde | 1561 | cur_trans->state = TRANS_STATE_COMMIT_DOING; |
f094ac32 LB |
1562 | spin_unlock(&root->fs_info->trans_lock); |
1563 | wait_event(cur_trans->writer_wait, | |
1564 | atomic_read(&cur_trans->num_writers) == 1); | |
1565 | ||
1566 | spin_lock(&root->fs_info->trans_lock); | |
d7096fc3 | 1567 | } |
49b25e05 JM |
1568 | spin_unlock(&root->fs_info->trans_lock); |
1569 | ||
1570 | btrfs_cleanup_one_transaction(trans->transaction, root); | |
1571 | ||
4a9d8bde MX |
1572 | spin_lock(&root->fs_info->trans_lock); |
1573 | if (cur_trans == root->fs_info->running_transaction) | |
1574 | root->fs_info->running_transaction = NULL; | |
1575 | spin_unlock(&root->fs_info->trans_lock); | |
1576 | ||
e0228285 JB |
1577 | if (trans->type & __TRANS_FREEZABLE) |
1578 | sb_end_intwrite(root->fs_info->sb); | |
724e2315 JB |
1579 | btrfs_put_transaction(cur_trans); |
1580 | btrfs_put_transaction(cur_trans); | |
49b25e05 JM |
1581 | |
1582 | trace_btrfs_transaction_commit(root); | |
1583 | ||
49b25e05 JM |
1584 | if (current->journal_info == trans) |
1585 | current->journal_info = NULL; | |
c0af8f0b | 1586 | btrfs_scrub_cancel(root->fs_info); |
49b25e05 JM |
1587 | |
1588 | kmem_cache_free(btrfs_trans_handle_cachep, trans); | |
1589 | } | |
1590 | ||
ca469637 MX |
1591 | static int btrfs_flush_all_pending_stuffs(struct btrfs_trans_handle *trans, |
1592 | struct btrfs_root *root) | |
1593 | { | |
ca469637 MX |
1594 | int ret; |
1595 | ||
ca469637 | 1596 | ret = btrfs_run_delayed_items(trans, root); |
ca469637 MX |
1597 | /* |
1598 | * running the delayed items may have added new refs. account | |
1599 | * them now so that they hinder processing of more delayed refs | |
1600 | * as little as possible. | |
1601 | */ | |
80d94fb3 FDBM |
1602 | if (ret) { |
1603 | btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info); | |
1604 | return ret; | |
1605 | } | |
1606 | ||
1607 | ret = btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info); | |
1608 | if (ret) | |
1609 | return ret; | |
ca469637 MX |
1610 | |
1611 | /* | |
1612 | * rename don't use btrfs_join_transaction, so, once we | |
1613 | * set the transaction to blocked above, we aren't going | |
1614 | * to get any new ordered operations. We can safely run | |
1615 | * it here and no for sure that nothing new will be added | |
1616 | * to the list | |
1617 | */ | |
569e0f35 | 1618 | ret = btrfs_run_ordered_operations(trans, root, 1); |
ca469637 | 1619 | |
eebc6084 | 1620 | return ret; |
ca469637 MX |
1621 | } |
1622 | ||
82436617 MX |
1623 | static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info) |
1624 | { | |
1625 | if (btrfs_test_opt(fs_info->tree_root, FLUSHONCOMMIT)) | |
6c255e67 | 1626 | return btrfs_start_delalloc_roots(fs_info, 1, -1); |
82436617 MX |
1627 | return 0; |
1628 | } | |
1629 | ||
1630 | static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info) | |
1631 | { | |
1632 | if (btrfs_test_opt(fs_info->tree_root, FLUSHONCOMMIT)) | |
b0244199 | 1633 | btrfs_wait_ordered_roots(fs_info, -1); |
82436617 MX |
1634 | } |
1635 | ||
79154b1b CM |
1636 | int btrfs_commit_transaction(struct btrfs_trans_handle *trans, |
1637 | struct btrfs_root *root) | |
1638 | { | |
49b25e05 | 1639 | struct btrfs_transaction *cur_trans = trans->transaction; |
8fd17795 | 1640 | struct btrfs_transaction *prev_trans = NULL; |
25287e0a | 1641 | int ret; |
79154b1b | 1642 | |
569e0f35 | 1643 | ret = btrfs_run_ordered_operations(trans, root, 0); |
25287e0a MX |
1644 | if (ret) { |
1645 | btrfs_abort_transaction(trans, root, ret); | |
e4a2bcac JB |
1646 | btrfs_end_transaction(trans, root); |
1647 | return ret; | |
25287e0a | 1648 | } |
5a3f23d5 | 1649 | |
8d25a086 MX |
1650 | /* Stop the commit early if ->aborted is set */ |
1651 | if (unlikely(ACCESS_ONCE(cur_trans->aborted))) { | |
25287e0a | 1652 | ret = cur_trans->aborted; |
e4a2bcac JB |
1653 | btrfs_end_transaction(trans, root); |
1654 | return ret; | |
25287e0a | 1655 | } |
49b25e05 | 1656 | |
56bec294 CM |
1657 | /* make a pass through all the delayed refs we have so far |
1658 | * any runnings procs may add more while we are here | |
1659 | */ | |
1660 | ret = btrfs_run_delayed_refs(trans, root, 0); | |
e4a2bcac JB |
1661 | if (ret) { |
1662 | btrfs_end_transaction(trans, root); | |
1663 | return ret; | |
1664 | } | |
56bec294 | 1665 | |
0e721106 JB |
1666 | btrfs_trans_release_metadata(trans, root); |
1667 | trans->block_rsv = NULL; | |
272d26d0 MX |
1668 | if (trans->qgroup_reserved) { |
1669 | btrfs_qgroup_free(root, trans->qgroup_reserved); | |
1670 | trans->qgroup_reserved = 0; | |
1671 | } | |
0e721106 | 1672 | |
b7ec40d7 | 1673 | cur_trans = trans->transaction; |
49b25e05 | 1674 | |
56bec294 CM |
1675 | /* |
1676 | * set the flushing flag so procs in this transaction have to | |
1677 | * start sending their work down. | |
1678 | */ | |
b7ec40d7 | 1679 | cur_trans->delayed_refs.flushing = 1; |
1be41b78 | 1680 | smp_wmb(); |
56bec294 | 1681 | |
ea658bad JB |
1682 | if (!list_empty(&trans->new_bgs)) |
1683 | btrfs_create_pending_block_groups(trans, root); | |
1684 | ||
c3e69d58 | 1685 | ret = btrfs_run_delayed_refs(trans, root, 0); |
e4a2bcac JB |
1686 | if (ret) { |
1687 | btrfs_end_transaction(trans, root); | |
1688 | return ret; | |
1689 | } | |
56bec294 | 1690 | |
4a9d8bde MX |
1691 | spin_lock(&root->fs_info->trans_lock); |
1692 | if (cur_trans->state >= TRANS_STATE_COMMIT_START) { | |
1693 | spin_unlock(&root->fs_info->trans_lock); | |
13c5a93e | 1694 | atomic_inc(&cur_trans->use_count); |
49b25e05 | 1695 | ret = btrfs_end_transaction(trans, root); |
ccd467d6 | 1696 | |
b9c8300c | 1697 | wait_for_commit(root, cur_trans); |
15ee9bc7 | 1698 | |
724e2315 | 1699 | btrfs_put_transaction(cur_trans); |
15ee9bc7 | 1700 | |
49b25e05 | 1701 | return ret; |
79154b1b | 1702 | } |
4313b399 | 1703 | |
4a9d8bde | 1704 | cur_trans->state = TRANS_STATE_COMMIT_START; |
bb9c12c9 SW |
1705 | wake_up(&root->fs_info->transaction_blocked_wait); |
1706 | ||
ccd467d6 CM |
1707 | if (cur_trans->list.prev != &root->fs_info->trans_list) { |
1708 | prev_trans = list_entry(cur_trans->list.prev, | |
1709 | struct btrfs_transaction, list); | |
4a9d8bde | 1710 | if (prev_trans->state != TRANS_STATE_COMPLETED) { |
13c5a93e | 1711 | atomic_inc(&prev_trans->use_count); |
a4abeea4 | 1712 | spin_unlock(&root->fs_info->trans_lock); |
ccd467d6 CM |
1713 | |
1714 | wait_for_commit(root, prev_trans); | |
ccd467d6 | 1715 | |
724e2315 | 1716 | btrfs_put_transaction(prev_trans); |
a4abeea4 JB |
1717 | } else { |
1718 | spin_unlock(&root->fs_info->trans_lock); | |
ccd467d6 | 1719 | } |
a4abeea4 JB |
1720 | } else { |
1721 | spin_unlock(&root->fs_info->trans_lock); | |
ccd467d6 | 1722 | } |
15ee9bc7 | 1723 | |
0860adfd MX |
1724 | extwriter_counter_dec(cur_trans, trans->type); |
1725 | ||
82436617 MX |
1726 | ret = btrfs_start_delalloc_flush(root->fs_info); |
1727 | if (ret) | |
1728 | goto cleanup_transaction; | |
1729 | ||
581227d0 MX |
1730 | ret = btrfs_flush_all_pending_stuffs(trans, root); |
1731 | if (ret) | |
1732 | goto cleanup_transaction; | |
15ee9bc7 | 1733 | |
581227d0 MX |
1734 | wait_event(cur_trans->writer_wait, |
1735 | extwriter_counter_read(cur_trans) == 0); | |
15ee9bc7 | 1736 | |
581227d0 | 1737 | /* some pending stuffs might be added after the previous flush. */ |
ca469637 MX |
1738 | ret = btrfs_flush_all_pending_stuffs(trans, root); |
1739 | if (ret) | |
1740 | goto cleanup_transaction; | |
1741 | ||
82436617 | 1742 | btrfs_wait_delalloc_flush(root->fs_info); |
cb7ab021 WS |
1743 | |
1744 | btrfs_scrub_pause(root); | |
ed0ca140 JB |
1745 | /* |
1746 | * Ok now we need to make sure to block out any other joins while we | |
1747 | * commit the transaction. We could have started a join before setting | |
4a9d8bde | 1748 | * COMMIT_DOING so make sure to wait for num_writers to == 1 again. |
ed0ca140 JB |
1749 | */ |
1750 | spin_lock(&root->fs_info->trans_lock); | |
4a9d8bde | 1751 | cur_trans->state = TRANS_STATE_COMMIT_DOING; |
ed0ca140 JB |
1752 | spin_unlock(&root->fs_info->trans_lock); |
1753 | wait_event(cur_trans->writer_wait, | |
1754 | atomic_read(&cur_trans->num_writers) == 1); | |
1755 | ||
2cba30f1 MX |
1756 | /* ->aborted might be set after the previous check, so check it */ |
1757 | if (unlikely(ACCESS_ONCE(cur_trans->aborted))) { | |
1758 | ret = cur_trans->aborted; | |
6cf7f77e | 1759 | goto scrub_continue; |
2cba30f1 | 1760 | } |
7585717f CM |
1761 | /* |
1762 | * the reloc mutex makes sure that we stop | |
1763 | * the balancing code from coming in and moving | |
1764 | * extents around in the middle of the commit | |
1765 | */ | |
1766 | mutex_lock(&root->fs_info->reloc_mutex); | |
1767 | ||
42874b3d MX |
1768 | /* |
1769 | * We needn't worry about the delayed items because we will | |
1770 | * deal with them in create_pending_snapshot(), which is the | |
1771 | * core function of the snapshot creation. | |
1772 | */ | |
1773 | ret = create_pending_snapshots(trans, root->fs_info); | |
49b25e05 JM |
1774 | if (ret) { |
1775 | mutex_unlock(&root->fs_info->reloc_mutex); | |
6cf7f77e | 1776 | goto scrub_continue; |
49b25e05 | 1777 | } |
3063d29f | 1778 | |
42874b3d MX |
1779 | /* |
1780 | * We insert the dir indexes of the snapshots and update the inode | |
1781 | * of the snapshots' parents after the snapshot creation, so there | |
1782 | * are some delayed items which are not dealt with. Now deal with | |
1783 | * them. | |
1784 | * | |
1785 | * We needn't worry that this operation will corrupt the snapshots, | |
1786 | * because all the tree which are snapshoted will be forced to COW | |
1787 | * the nodes and leaves. | |
1788 | */ | |
1789 | ret = btrfs_run_delayed_items(trans, root); | |
49b25e05 JM |
1790 | if (ret) { |
1791 | mutex_unlock(&root->fs_info->reloc_mutex); | |
6cf7f77e | 1792 | goto scrub_continue; |
49b25e05 | 1793 | } |
16cdcec7 | 1794 | |
56bec294 | 1795 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
49b25e05 JM |
1796 | if (ret) { |
1797 | mutex_unlock(&root->fs_info->reloc_mutex); | |
6cf7f77e | 1798 | goto scrub_continue; |
49b25e05 | 1799 | } |
56bec294 | 1800 | |
e999376f CM |
1801 | /* |
1802 | * make sure none of the code above managed to slip in a | |
1803 | * delayed item | |
1804 | */ | |
1805 | btrfs_assert_delayed_root_empty(root); | |
1806 | ||
2c90e5d6 | 1807 | WARN_ON(cur_trans != trans->transaction); |
dc17ff8f | 1808 | |
e02119d5 CM |
1809 | /* btrfs_commit_tree_roots is responsible for getting the |
1810 | * various roots consistent with each other. Every pointer | |
1811 | * in the tree of tree roots has to point to the most up to date | |
1812 | * root for every subvolume and other tree. So, we have to keep | |
1813 | * the tree logging code from jumping in and changing any | |
1814 | * of the trees. | |
1815 | * | |
1816 | * At this point in the commit, there can't be any tree-log | |
1817 | * writers, but a little lower down we drop the trans mutex | |
1818 | * and let new people in. By holding the tree_log_mutex | |
1819 | * from now until after the super is written, we avoid races | |
1820 | * with the tree-log code. | |
1821 | */ | |
1822 | mutex_lock(&root->fs_info->tree_log_mutex); | |
1823 | ||
5d4f98a2 | 1824 | ret = commit_fs_roots(trans, root); |
49b25e05 JM |
1825 | if (ret) { |
1826 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
871383be | 1827 | mutex_unlock(&root->fs_info->reloc_mutex); |
6cf7f77e | 1828 | goto scrub_continue; |
49b25e05 | 1829 | } |
54aa1f4d | 1830 | |
3818aea2 QW |
1831 | /* |
1832 | * Since the transaction is done, we should set the inode map cache flag | |
1833 | * before any other comming transaction. | |
1834 | */ | |
1835 | if (btrfs_test_opt(root, CHANGE_INODE_CACHE)) | |
1836 | btrfs_set_opt(root->fs_info->mount_opt, INODE_MAP_CACHE); | |
1837 | else | |
1838 | btrfs_clear_opt(root->fs_info->mount_opt, INODE_MAP_CACHE); | |
1839 | ||
5d4f98a2 | 1840 | /* commit_fs_roots gets rid of all the tree log roots, it is now |
e02119d5 CM |
1841 | * safe to free the root of tree log roots |
1842 | */ | |
1843 | btrfs_free_log_root_tree(trans, root->fs_info); | |
1844 | ||
5d4f98a2 | 1845 | ret = commit_cowonly_roots(trans, root); |
49b25e05 JM |
1846 | if (ret) { |
1847 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
871383be | 1848 | mutex_unlock(&root->fs_info->reloc_mutex); |
6cf7f77e | 1849 | goto scrub_continue; |
49b25e05 | 1850 | } |
54aa1f4d | 1851 | |
2cba30f1 MX |
1852 | /* |
1853 | * The tasks which save the space cache and inode cache may also | |
1854 | * update ->aborted, check it. | |
1855 | */ | |
1856 | if (unlikely(ACCESS_ONCE(cur_trans->aborted))) { | |
1857 | ret = cur_trans->aborted; | |
1858 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
1859 | mutex_unlock(&root->fs_info->reloc_mutex); | |
6cf7f77e | 1860 | goto scrub_continue; |
2cba30f1 MX |
1861 | } |
1862 | ||
11833d66 YZ |
1863 | btrfs_prepare_extent_commit(trans, root); |
1864 | ||
78fae27e | 1865 | cur_trans = root->fs_info->running_transaction; |
5d4f98a2 YZ |
1866 | |
1867 | btrfs_set_root_node(&root->fs_info->tree_root->root_item, | |
1868 | root->fs_info->tree_root->node); | |
9e351cc8 JB |
1869 | list_add_tail(&root->fs_info->tree_root->dirty_list, |
1870 | &cur_trans->switch_commits); | |
5d4f98a2 YZ |
1871 | |
1872 | btrfs_set_root_node(&root->fs_info->chunk_root->root_item, | |
1873 | root->fs_info->chunk_root->node); | |
9e351cc8 JB |
1874 | list_add_tail(&root->fs_info->chunk_root->dirty_list, |
1875 | &cur_trans->switch_commits); | |
1876 | ||
1877 | switch_commit_roots(cur_trans, root->fs_info); | |
5d4f98a2 | 1878 | |
edf39272 | 1879 | assert_qgroups_uptodate(trans); |
5d4f98a2 | 1880 | update_super_roots(root); |
e02119d5 | 1881 | |
60e7cd3a JB |
1882 | btrfs_set_super_log_root(root->fs_info->super_copy, 0); |
1883 | btrfs_set_super_log_root_level(root->fs_info->super_copy, 0); | |
6c41761f DS |
1884 | memcpy(root->fs_info->super_for_commit, root->fs_info->super_copy, |
1885 | sizeof(*root->fs_info->super_copy)); | |
ccd467d6 | 1886 | |
a4abeea4 | 1887 | spin_lock(&root->fs_info->trans_lock); |
4a9d8bde | 1888 | cur_trans->state = TRANS_STATE_UNBLOCKED; |
a4abeea4 | 1889 | root->fs_info->running_transaction = NULL; |
a4abeea4 | 1890 | spin_unlock(&root->fs_info->trans_lock); |
7585717f | 1891 | mutex_unlock(&root->fs_info->reloc_mutex); |
b7ec40d7 | 1892 | |
f9295749 | 1893 | wake_up(&root->fs_info->transaction_wait); |
e6dcd2dc | 1894 | |
79154b1b | 1895 | ret = btrfs_write_and_wait_transaction(trans, root); |
49b25e05 JM |
1896 | if (ret) { |
1897 | btrfs_error(root->fs_info, ret, | |
08748810 | 1898 | "Error while writing out transaction"); |
49b25e05 | 1899 | mutex_unlock(&root->fs_info->tree_log_mutex); |
6cf7f77e | 1900 | goto scrub_continue; |
49b25e05 JM |
1901 | } |
1902 | ||
1903 | ret = write_ctree_super(trans, root, 0); | |
1904 | if (ret) { | |
1905 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
6cf7f77e | 1906 | goto scrub_continue; |
49b25e05 | 1907 | } |
4313b399 | 1908 | |
e02119d5 CM |
1909 | /* |
1910 | * the super is written, we can safely allow the tree-loggers | |
1911 | * to go about their business | |
1912 | */ | |
1913 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
1914 | ||
11833d66 | 1915 | btrfs_finish_extent_commit(trans, root); |
4313b399 | 1916 | |
15ee9bc7 | 1917 | root->fs_info->last_trans_committed = cur_trans->transid; |
4a9d8bde MX |
1918 | /* |
1919 | * We needn't acquire the lock here because there is no other task | |
1920 | * which can change it. | |
1921 | */ | |
1922 | cur_trans->state = TRANS_STATE_COMPLETED; | |
2c90e5d6 | 1923 | wake_up(&cur_trans->commit_wait); |
3de4586c | 1924 | |
a4abeea4 | 1925 | spin_lock(&root->fs_info->trans_lock); |
13c5a93e | 1926 | list_del_init(&cur_trans->list); |
a4abeea4 JB |
1927 | spin_unlock(&root->fs_info->trans_lock); |
1928 | ||
724e2315 JB |
1929 | btrfs_put_transaction(cur_trans); |
1930 | btrfs_put_transaction(cur_trans); | |
58176a96 | 1931 | |
0860adfd | 1932 | if (trans->type & __TRANS_FREEZABLE) |
354aa0fb | 1933 | sb_end_intwrite(root->fs_info->sb); |
b2b5ef5c | 1934 | |
1abe9b8a | 1935 | trace_btrfs_transaction_commit(root); |
1936 | ||
a2de733c AJ |
1937 | btrfs_scrub_continue(root); |
1938 | ||
9ed74f2d JB |
1939 | if (current->journal_info == trans) |
1940 | current->journal_info = NULL; | |
1941 | ||
2c90e5d6 | 1942 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
24bbcf04 YZ |
1943 | |
1944 | if (current != root->fs_info->transaction_kthread) | |
1945 | btrfs_run_delayed_iputs(root); | |
1946 | ||
79154b1b | 1947 | return ret; |
49b25e05 | 1948 | |
6cf7f77e WS |
1949 | scrub_continue: |
1950 | btrfs_scrub_continue(root); | |
49b25e05 | 1951 | cleanup_transaction: |
0e721106 JB |
1952 | btrfs_trans_release_metadata(trans, root); |
1953 | trans->block_rsv = NULL; | |
272d26d0 MX |
1954 | if (trans->qgroup_reserved) { |
1955 | btrfs_qgroup_free(root, trans->qgroup_reserved); | |
1956 | trans->qgroup_reserved = 0; | |
1957 | } | |
c2cf52eb | 1958 | btrfs_warn(root->fs_info, "Skipping commit of aborted transaction."); |
49b25e05 JM |
1959 | if (current->journal_info == trans) |
1960 | current->journal_info = NULL; | |
7b8b92af | 1961 | cleanup_transaction(trans, root, ret); |
49b25e05 JM |
1962 | |
1963 | return ret; | |
79154b1b CM |
1964 | } |
1965 | ||
d352ac68 | 1966 | /* |
9d1a2a3a DS |
1967 | * return < 0 if error |
1968 | * 0 if there are no more dead_roots at the time of call | |
1969 | * 1 there are more to be processed, call me again | |
1970 | * | |
1971 | * The return value indicates there are certainly more snapshots to delete, but | |
1972 | * if there comes a new one during processing, it may return 0. We don't mind, | |
1973 | * because btrfs_commit_super will poke cleaner thread and it will process it a | |
1974 | * few seconds later. | |
d352ac68 | 1975 | */ |
9d1a2a3a | 1976 | int btrfs_clean_one_deleted_snapshot(struct btrfs_root *root) |
e9d0b13b | 1977 | { |
9d1a2a3a | 1978 | int ret; |
5d4f98a2 YZ |
1979 | struct btrfs_fs_info *fs_info = root->fs_info; |
1980 | ||
a4abeea4 | 1981 | spin_lock(&fs_info->trans_lock); |
9d1a2a3a DS |
1982 | if (list_empty(&fs_info->dead_roots)) { |
1983 | spin_unlock(&fs_info->trans_lock); | |
1984 | return 0; | |
1985 | } | |
1986 | root = list_first_entry(&fs_info->dead_roots, | |
1987 | struct btrfs_root, root_list); | |
cfad392b | 1988 | list_del_init(&root->root_list); |
a4abeea4 | 1989 | spin_unlock(&fs_info->trans_lock); |
e9d0b13b | 1990 | |
efe120a0 | 1991 | pr_debug("BTRFS: cleaner removing %llu\n", root->objectid); |
76dda93c | 1992 | |
9d1a2a3a | 1993 | btrfs_kill_all_delayed_nodes(root); |
16cdcec7 | 1994 | |
9d1a2a3a DS |
1995 | if (btrfs_header_backref_rev(root->node) < |
1996 | BTRFS_MIXED_BACKREF_REV) | |
1997 | ret = btrfs_drop_snapshot(root, NULL, 0, 0); | |
1998 | else | |
1999 | ret = btrfs_drop_snapshot(root, NULL, 1, 0); | |
2000 | /* | |
2001 | * If we encounter a transaction abort during snapshot cleaning, we | |
2002 | * don't want to crash here | |
2003 | */ | |
6596a928 | 2004 | return (ret < 0) ? 0 : 1; |
e9d0b13b | 2005 | } |