Commit | Line | Data |
---|---|---|
6cbd5570 CM |
1 | /* |
2 | * Copyright (C) 2007 Oracle. All rights reserved. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public | |
6 | * License v2 as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, | |
9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public | |
14 | * License along with this program; if not, write to the | |
15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
16 | * Boston, MA 021110-1307, USA. | |
17 | */ | |
18 | ||
79154b1b | 19 | #include <linux/fs.h> |
5a0e3ad6 | 20 | #include <linux/slab.h> |
34088780 | 21 | #include <linux/sched.h> |
d3c2fdcf | 22 | #include <linux/writeback.h> |
5f39d397 | 23 | #include <linux/pagemap.h> |
5f2cc086 | 24 | #include <linux/blkdev.h> |
8ea05e3a | 25 | #include <linux/uuid.h> |
79154b1b CM |
26 | #include "ctree.h" |
27 | #include "disk-io.h" | |
28 | #include "transaction.h" | |
925baedd | 29 | #include "locking.h" |
e02119d5 | 30 | #include "tree-log.h" |
581bb050 | 31 | #include "inode-map.h" |
733f4fbb | 32 | #include "volumes.h" |
8dabb742 | 33 | #include "dev-replace.h" |
fcebe456 | 34 | #include "qgroup.h" |
79154b1b | 35 | |
0f7d52f4 CM |
36 | #define BTRFS_ROOT_TRANS_TAG 0 |
37 | ||
4a9d8bde MX |
38 | static unsigned int btrfs_blocked_trans_types[TRANS_STATE_MAX] = { |
39 | [TRANS_STATE_RUNNING] = 0U, | |
40 | [TRANS_STATE_BLOCKED] = (__TRANS_USERSPACE | | |
41 | __TRANS_START), | |
42 | [TRANS_STATE_COMMIT_START] = (__TRANS_USERSPACE | | |
43 | __TRANS_START | | |
44 | __TRANS_ATTACH), | |
45 | [TRANS_STATE_COMMIT_DOING] = (__TRANS_USERSPACE | | |
46 | __TRANS_START | | |
47 | __TRANS_ATTACH | | |
48 | __TRANS_JOIN), | |
49 | [TRANS_STATE_UNBLOCKED] = (__TRANS_USERSPACE | | |
50 | __TRANS_START | | |
51 | __TRANS_ATTACH | | |
52 | __TRANS_JOIN | | |
53 | __TRANS_JOIN_NOLOCK), | |
54 | [TRANS_STATE_COMPLETED] = (__TRANS_USERSPACE | | |
55 | __TRANS_START | | |
56 | __TRANS_ATTACH | | |
57 | __TRANS_JOIN | | |
58 | __TRANS_JOIN_NOLOCK), | |
59 | }; | |
60 | ||
724e2315 | 61 | void btrfs_put_transaction(struct btrfs_transaction *transaction) |
79154b1b | 62 | { |
13c5a93e JB |
63 | WARN_ON(atomic_read(&transaction->use_count) == 0); |
64 | if (atomic_dec_and_test(&transaction->use_count)) { | |
a4abeea4 | 65 | BUG_ON(!list_empty(&transaction->list)); |
c46effa6 | 66 | WARN_ON(!RB_EMPTY_ROOT(&transaction->delayed_refs.href_root)); |
6df9a95e JB |
67 | while (!list_empty(&transaction->pending_chunks)) { |
68 | struct extent_map *em; | |
69 | ||
70 | em = list_first_entry(&transaction->pending_chunks, | |
71 | struct extent_map, list); | |
72 | list_del_init(&em->list); | |
73 | free_extent_map(em); | |
74 | } | |
2c90e5d6 | 75 | kmem_cache_free(btrfs_transaction_cachep, transaction); |
78fae27e | 76 | } |
79154b1b CM |
77 | } |
78 | ||
9e351cc8 JB |
79 | static noinline void switch_commit_roots(struct btrfs_transaction *trans, |
80 | struct btrfs_fs_info *fs_info) | |
817d52f8 | 81 | { |
9e351cc8 JB |
82 | struct btrfs_root *root, *tmp; |
83 | ||
84 | down_write(&fs_info->commit_root_sem); | |
85 | list_for_each_entry_safe(root, tmp, &trans->switch_commits, | |
86 | dirty_list) { | |
87 | list_del_init(&root->dirty_list); | |
88 | free_extent_buffer(root->commit_root); | |
89 | root->commit_root = btrfs_root_node(root); | |
90 | if (is_fstree(root->objectid)) | |
91 | btrfs_unpin_free_ino(root); | |
92 | } | |
93 | up_write(&fs_info->commit_root_sem); | |
817d52f8 JB |
94 | } |
95 | ||
0860adfd MX |
96 | static inline void extwriter_counter_inc(struct btrfs_transaction *trans, |
97 | unsigned int type) | |
98 | { | |
99 | if (type & TRANS_EXTWRITERS) | |
100 | atomic_inc(&trans->num_extwriters); | |
101 | } | |
102 | ||
103 | static inline void extwriter_counter_dec(struct btrfs_transaction *trans, | |
104 | unsigned int type) | |
105 | { | |
106 | if (type & TRANS_EXTWRITERS) | |
107 | atomic_dec(&trans->num_extwriters); | |
108 | } | |
109 | ||
110 | static inline void extwriter_counter_init(struct btrfs_transaction *trans, | |
111 | unsigned int type) | |
112 | { | |
113 | atomic_set(&trans->num_extwriters, ((type & TRANS_EXTWRITERS) ? 1 : 0)); | |
114 | } | |
115 | ||
116 | static inline int extwriter_counter_read(struct btrfs_transaction *trans) | |
117 | { | |
118 | return atomic_read(&trans->num_extwriters); | |
178260b2 MX |
119 | } |
120 | ||
d352ac68 CM |
121 | /* |
122 | * either allocate a new transaction or hop into the existing one | |
123 | */ | |
0860adfd | 124 | static noinline int join_transaction(struct btrfs_root *root, unsigned int type) |
79154b1b CM |
125 | { |
126 | struct btrfs_transaction *cur_trans; | |
19ae4e81 | 127 | struct btrfs_fs_info *fs_info = root->fs_info; |
a4abeea4 | 128 | |
19ae4e81 | 129 | spin_lock(&fs_info->trans_lock); |
d43317dc | 130 | loop: |
49b25e05 | 131 | /* The file system has been taken offline. No new transactions. */ |
87533c47 | 132 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { |
19ae4e81 | 133 | spin_unlock(&fs_info->trans_lock); |
49b25e05 JM |
134 | return -EROFS; |
135 | } | |
136 | ||
19ae4e81 | 137 | cur_trans = fs_info->running_transaction; |
a4abeea4 | 138 | if (cur_trans) { |
871383be | 139 | if (cur_trans->aborted) { |
19ae4e81 | 140 | spin_unlock(&fs_info->trans_lock); |
49b25e05 | 141 | return cur_trans->aborted; |
871383be | 142 | } |
4a9d8bde | 143 | if (btrfs_blocked_trans_types[cur_trans->state] & type) { |
178260b2 MX |
144 | spin_unlock(&fs_info->trans_lock); |
145 | return -EBUSY; | |
146 | } | |
a4abeea4 | 147 | atomic_inc(&cur_trans->use_count); |
13c5a93e | 148 | atomic_inc(&cur_trans->num_writers); |
0860adfd | 149 | extwriter_counter_inc(cur_trans, type); |
19ae4e81 | 150 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 | 151 | return 0; |
79154b1b | 152 | } |
19ae4e81 | 153 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 | 154 | |
354aa0fb MX |
155 | /* |
156 | * If we are ATTACH, we just want to catch the current transaction, | |
157 | * and commit it. If there is no transaction, just return ENOENT. | |
158 | */ | |
159 | if (type == TRANS_ATTACH) | |
160 | return -ENOENT; | |
161 | ||
4a9d8bde MX |
162 | /* |
163 | * JOIN_NOLOCK only happens during the transaction commit, so | |
164 | * it is impossible that ->running_transaction is NULL | |
165 | */ | |
166 | BUG_ON(type == TRANS_JOIN_NOLOCK); | |
167 | ||
a4abeea4 JB |
168 | cur_trans = kmem_cache_alloc(btrfs_transaction_cachep, GFP_NOFS); |
169 | if (!cur_trans) | |
170 | return -ENOMEM; | |
d43317dc | 171 | |
19ae4e81 JS |
172 | spin_lock(&fs_info->trans_lock); |
173 | if (fs_info->running_transaction) { | |
d43317dc CM |
174 | /* |
175 | * someone started a transaction after we unlocked. Make sure | |
4a9d8bde | 176 | * to redo the checks above |
d43317dc | 177 | */ |
a4abeea4 | 178 | kmem_cache_free(btrfs_transaction_cachep, cur_trans); |
d43317dc | 179 | goto loop; |
87533c47 | 180 | } else if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { |
e4b50e14 | 181 | spin_unlock(&fs_info->trans_lock); |
7b8b92af JB |
182 | kmem_cache_free(btrfs_transaction_cachep, cur_trans); |
183 | return -EROFS; | |
79154b1b | 184 | } |
d43317dc | 185 | |
a4abeea4 | 186 | atomic_set(&cur_trans->num_writers, 1); |
0860adfd | 187 | extwriter_counter_init(cur_trans, type); |
a4abeea4 JB |
188 | init_waitqueue_head(&cur_trans->writer_wait); |
189 | init_waitqueue_head(&cur_trans->commit_wait); | |
4a9d8bde | 190 | cur_trans->state = TRANS_STATE_RUNNING; |
a4abeea4 JB |
191 | /* |
192 | * One for this trans handle, one so it will live on until we | |
193 | * commit the transaction. | |
194 | */ | |
195 | atomic_set(&cur_trans->use_count, 2); | |
a4abeea4 JB |
196 | cur_trans->start_time = get_seconds(); |
197 | ||
c46effa6 | 198 | cur_trans->delayed_refs.href_root = RB_ROOT; |
d7df2c79 | 199 | atomic_set(&cur_trans->delayed_refs.num_entries, 0); |
a4abeea4 JB |
200 | cur_trans->delayed_refs.num_heads_ready = 0; |
201 | cur_trans->delayed_refs.num_heads = 0; | |
202 | cur_trans->delayed_refs.flushing = 0; | |
203 | cur_trans->delayed_refs.run_delayed_start = 0; | |
20b297d6 JS |
204 | |
205 | /* | |
206 | * although the tree mod log is per file system and not per transaction, | |
207 | * the log must never go across transaction boundaries. | |
208 | */ | |
209 | smp_mb(); | |
31b1a2bd | 210 | if (!list_empty(&fs_info->tree_mod_seq_list)) |
efe120a0 | 211 | WARN(1, KERN_ERR "BTRFS: tree_mod_seq_list not empty when " |
20b297d6 | 212 | "creating a fresh transaction\n"); |
31b1a2bd | 213 | if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log)) |
efe120a0 | 214 | WARN(1, KERN_ERR "BTRFS: tree_mod_log rb tree not empty when " |
20b297d6 | 215 | "creating a fresh transaction\n"); |
fc36ed7e | 216 | atomic64_set(&fs_info->tree_mod_seq, 0); |
20b297d6 | 217 | |
a4abeea4 JB |
218 | spin_lock_init(&cur_trans->delayed_refs.lock); |
219 | ||
220 | INIT_LIST_HEAD(&cur_trans->pending_snapshots); | |
569e0f35 | 221 | INIT_LIST_HEAD(&cur_trans->ordered_operations); |
6df9a95e | 222 | INIT_LIST_HEAD(&cur_trans->pending_chunks); |
9e351cc8 | 223 | INIT_LIST_HEAD(&cur_trans->switch_commits); |
19ae4e81 | 224 | list_add_tail(&cur_trans->list, &fs_info->trans_list); |
a4abeea4 | 225 | extent_io_tree_init(&cur_trans->dirty_pages, |
19ae4e81 JS |
226 | fs_info->btree_inode->i_mapping); |
227 | fs_info->generation++; | |
228 | cur_trans->transid = fs_info->generation; | |
229 | fs_info->running_transaction = cur_trans; | |
49b25e05 | 230 | cur_trans->aborted = 0; |
19ae4e81 | 231 | spin_unlock(&fs_info->trans_lock); |
15ee9bc7 | 232 | |
79154b1b CM |
233 | return 0; |
234 | } | |
235 | ||
d352ac68 | 236 | /* |
d397712b CM |
237 | * this does all the record keeping required to make sure that a reference |
238 | * counted root is properly recorded in a given transaction. This is required | |
239 | * to make sure the old root from before we joined the transaction is deleted | |
240 | * when the transaction commits | |
d352ac68 | 241 | */ |
7585717f | 242 | static int record_root_in_trans(struct btrfs_trans_handle *trans, |
a4abeea4 | 243 | struct btrfs_root *root) |
6702ed49 | 244 | { |
27cdeb70 MX |
245 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
246 | root->last_trans < trans->transid) { | |
6702ed49 | 247 | WARN_ON(root == root->fs_info->extent_root); |
5d4f98a2 YZ |
248 | WARN_ON(root->commit_root != root->node); |
249 | ||
7585717f | 250 | /* |
27cdeb70 | 251 | * see below for IN_TRANS_SETUP usage rules |
7585717f CM |
252 | * we have the reloc mutex held now, so there |
253 | * is only one writer in this function | |
254 | */ | |
27cdeb70 | 255 | set_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); |
7585717f | 256 | |
27cdeb70 | 257 | /* make sure readers find IN_TRANS_SETUP before |
7585717f CM |
258 | * they find our root->last_trans update |
259 | */ | |
260 | smp_wmb(); | |
261 | ||
a4abeea4 JB |
262 | spin_lock(&root->fs_info->fs_roots_radix_lock); |
263 | if (root->last_trans == trans->transid) { | |
264 | spin_unlock(&root->fs_info->fs_roots_radix_lock); | |
265 | return 0; | |
266 | } | |
5d4f98a2 YZ |
267 | radix_tree_tag_set(&root->fs_info->fs_roots_radix, |
268 | (unsigned long)root->root_key.objectid, | |
269 | BTRFS_ROOT_TRANS_TAG); | |
a4abeea4 | 270 | spin_unlock(&root->fs_info->fs_roots_radix_lock); |
7585717f CM |
271 | root->last_trans = trans->transid; |
272 | ||
273 | /* this is pretty tricky. We don't want to | |
274 | * take the relocation lock in btrfs_record_root_in_trans | |
275 | * unless we're really doing the first setup for this root in | |
276 | * this transaction. | |
277 | * | |
278 | * Normally we'd use root->last_trans as a flag to decide | |
279 | * if we want to take the expensive mutex. | |
280 | * | |
281 | * But, we have to set root->last_trans before we | |
282 | * init the relocation root, otherwise, we trip over warnings | |
283 | * in ctree.c. The solution used here is to flag ourselves | |
27cdeb70 | 284 | * with root IN_TRANS_SETUP. When this is 1, we're still |
7585717f CM |
285 | * fixing up the reloc trees and everyone must wait. |
286 | * | |
287 | * When this is zero, they can trust root->last_trans and fly | |
288 | * through btrfs_record_root_in_trans without having to take the | |
289 | * lock. smp_wmb() makes sure that all the writes above are | |
290 | * done before we pop in the zero below | |
291 | */ | |
5d4f98a2 | 292 | btrfs_init_reloc_root(trans, root); |
27cdeb70 MX |
293 | smp_mb__before_clear_bit(); |
294 | clear_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); | |
5d4f98a2 YZ |
295 | } |
296 | return 0; | |
297 | } | |
bcc63abb | 298 | |
7585717f CM |
299 | |
300 | int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, | |
301 | struct btrfs_root *root) | |
302 | { | |
27cdeb70 | 303 | if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state)) |
7585717f CM |
304 | return 0; |
305 | ||
306 | /* | |
27cdeb70 | 307 | * see record_root_in_trans for comments about IN_TRANS_SETUP usage |
7585717f CM |
308 | * and barriers |
309 | */ | |
310 | smp_rmb(); | |
311 | if (root->last_trans == trans->transid && | |
27cdeb70 | 312 | !test_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state)) |
7585717f CM |
313 | return 0; |
314 | ||
315 | mutex_lock(&root->fs_info->reloc_mutex); | |
316 | record_root_in_trans(trans, root); | |
317 | mutex_unlock(&root->fs_info->reloc_mutex); | |
318 | ||
319 | return 0; | |
320 | } | |
321 | ||
4a9d8bde MX |
322 | static inline int is_transaction_blocked(struct btrfs_transaction *trans) |
323 | { | |
324 | return (trans->state >= TRANS_STATE_BLOCKED && | |
501407aa JB |
325 | trans->state < TRANS_STATE_UNBLOCKED && |
326 | !trans->aborted); | |
4a9d8bde MX |
327 | } |
328 | ||
d352ac68 CM |
329 | /* wait for commit against the current transaction to become unblocked |
330 | * when this is done, it is safe to start a new transaction, but the current | |
331 | * transaction might not be fully on disk. | |
332 | */ | |
37d1aeee | 333 | static void wait_current_trans(struct btrfs_root *root) |
79154b1b | 334 | { |
f9295749 | 335 | struct btrfs_transaction *cur_trans; |
79154b1b | 336 | |
a4abeea4 | 337 | spin_lock(&root->fs_info->trans_lock); |
f9295749 | 338 | cur_trans = root->fs_info->running_transaction; |
4a9d8bde | 339 | if (cur_trans && is_transaction_blocked(cur_trans)) { |
13c5a93e | 340 | atomic_inc(&cur_trans->use_count); |
a4abeea4 | 341 | spin_unlock(&root->fs_info->trans_lock); |
72d63ed6 LZ |
342 | |
343 | wait_event(root->fs_info->transaction_wait, | |
501407aa JB |
344 | cur_trans->state >= TRANS_STATE_UNBLOCKED || |
345 | cur_trans->aborted); | |
724e2315 | 346 | btrfs_put_transaction(cur_trans); |
a4abeea4 JB |
347 | } else { |
348 | spin_unlock(&root->fs_info->trans_lock); | |
f9295749 | 349 | } |
37d1aeee CM |
350 | } |
351 | ||
a22285a6 YZ |
352 | static int may_wait_transaction(struct btrfs_root *root, int type) |
353 | { | |
a4abeea4 JB |
354 | if (root->fs_info->log_root_recovering) |
355 | return 0; | |
356 | ||
357 | if (type == TRANS_USERSPACE) | |
358 | return 1; | |
359 | ||
360 | if (type == TRANS_START && | |
361 | !atomic_read(&root->fs_info->open_ioctl_trans)) | |
a22285a6 | 362 | return 1; |
a4abeea4 | 363 | |
a22285a6 YZ |
364 | return 0; |
365 | } | |
366 | ||
20dd2cbf MX |
367 | static inline bool need_reserve_reloc_root(struct btrfs_root *root) |
368 | { | |
369 | if (!root->fs_info->reloc_ctl || | |
27cdeb70 | 370 | !test_bit(BTRFS_ROOT_REF_COWS, &root->state) || |
20dd2cbf MX |
371 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || |
372 | root->reloc_root) | |
373 | return false; | |
374 | ||
375 | return true; | |
376 | } | |
377 | ||
08e007d2 | 378 | static struct btrfs_trans_handle * |
0860adfd | 379 | start_transaction(struct btrfs_root *root, u64 num_items, unsigned int type, |
08e007d2 | 380 | enum btrfs_reserve_flush_enum flush) |
37d1aeee | 381 | { |
a22285a6 YZ |
382 | struct btrfs_trans_handle *h; |
383 | struct btrfs_transaction *cur_trans; | |
b5009945 | 384 | u64 num_bytes = 0; |
c5567237 | 385 | u64 qgroup_reserved = 0; |
20dd2cbf MX |
386 | bool reloc_reserved = false; |
387 | int ret; | |
acce952b | 388 | |
87533c47 | 389 | if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) |
acce952b | 390 | return ERR_PTR(-EROFS); |
2a1eb461 | 391 | |
a26e8c9f JB |
392 | if (current->journal_info && |
393 | current->journal_info != (void *)BTRFS_SEND_TRANS_STUB) { | |
0860adfd | 394 | WARN_ON(type & TRANS_EXTWRITERS); |
2a1eb461 JB |
395 | h = current->journal_info; |
396 | h->use_count++; | |
b7d5b0a8 | 397 | WARN_ON(h->use_count > 2); |
2a1eb461 JB |
398 | h->orig_rsv = h->block_rsv; |
399 | h->block_rsv = NULL; | |
400 | goto got_it; | |
401 | } | |
b5009945 JB |
402 | |
403 | /* | |
404 | * Do the reservation before we join the transaction so we can do all | |
405 | * the appropriate flushing if need be. | |
406 | */ | |
407 | if (num_items > 0 && root != root->fs_info->chunk_root) { | |
c5567237 AJ |
408 | if (root->fs_info->quota_enabled && |
409 | is_fstree(root->root_key.objectid)) { | |
410 | qgroup_reserved = num_items * root->leafsize; | |
411 | ret = btrfs_qgroup_reserve(root, qgroup_reserved); | |
412 | if (ret) | |
413 | return ERR_PTR(ret); | |
414 | } | |
415 | ||
b5009945 | 416 | num_bytes = btrfs_calc_trans_metadata_size(root, num_items); |
20dd2cbf MX |
417 | /* |
418 | * Do the reservation for the relocation root creation | |
419 | */ | |
420 | if (unlikely(need_reserve_reloc_root(root))) { | |
421 | num_bytes += root->nodesize; | |
422 | reloc_reserved = true; | |
423 | } | |
424 | ||
08e007d2 MX |
425 | ret = btrfs_block_rsv_add(root, |
426 | &root->fs_info->trans_block_rsv, | |
427 | num_bytes, flush); | |
b5009945 | 428 | if (ret) |
843fcf35 | 429 | goto reserve_fail; |
b5009945 | 430 | } |
a22285a6 YZ |
431 | again: |
432 | h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS); | |
843fcf35 MX |
433 | if (!h) { |
434 | ret = -ENOMEM; | |
435 | goto alloc_fail; | |
436 | } | |
37d1aeee | 437 | |
98114659 JB |
438 | /* |
439 | * If we are JOIN_NOLOCK we're already committing a transaction and | |
440 | * waiting on this guy, so we don't need to do the sb_start_intwrite | |
441 | * because we're already holding a ref. We need this because we could | |
442 | * have raced in and did an fsync() on a file which can kick a commit | |
443 | * and then we deadlock with somebody doing a freeze. | |
354aa0fb MX |
444 | * |
445 | * If we are ATTACH, it means we just want to catch the current | |
446 | * transaction and commit it, so we needn't do sb_start_intwrite(). | |
98114659 | 447 | */ |
0860adfd | 448 | if (type & __TRANS_FREEZABLE) |
60376ce4 | 449 | sb_start_intwrite(root->fs_info->sb); |
b2b5ef5c | 450 | |
a22285a6 | 451 | if (may_wait_transaction(root, type)) |
37d1aeee | 452 | wait_current_trans(root); |
a22285a6 | 453 | |
a4abeea4 | 454 | do { |
354aa0fb | 455 | ret = join_transaction(root, type); |
178260b2 | 456 | if (ret == -EBUSY) { |
a4abeea4 | 457 | wait_current_trans(root); |
178260b2 MX |
458 | if (unlikely(type == TRANS_ATTACH)) |
459 | ret = -ENOENT; | |
460 | } | |
a4abeea4 JB |
461 | } while (ret == -EBUSY); |
462 | ||
db5b493a | 463 | if (ret < 0) { |
354aa0fb MX |
464 | /* We must get the transaction if we are JOIN_NOLOCK. */ |
465 | BUG_ON(type == TRANS_JOIN_NOLOCK); | |
843fcf35 | 466 | goto join_fail; |
db5b493a | 467 | } |
0f7d52f4 | 468 | |
a22285a6 | 469 | cur_trans = root->fs_info->running_transaction; |
a22285a6 YZ |
470 | |
471 | h->transid = cur_trans->transid; | |
472 | h->transaction = cur_trans; | |
79154b1b | 473 | h->blocks_used = 0; |
a22285a6 | 474 | h->bytes_reserved = 0; |
d13603ef | 475 | h->root = root; |
56bec294 | 476 | h->delayed_ref_updates = 0; |
2a1eb461 | 477 | h->use_count = 1; |
0e721106 | 478 | h->adding_csums = 0; |
f0486c68 | 479 | h->block_rsv = NULL; |
2a1eb461 | 480 | h->orig_rsv = NULL; |
49b25e05 | 481 | h->aborted = 0; |
4b824906 | 482 | h->qgroup_reserved = 0; |
bed92eae | 483 | h->delayed_ref_elem.seq = 0; |
a698d075 | 484 | h->type = type; |
c6b305a8 | 485 | h->allocating_chunk = false; |
20dd2cbf | 486 | h->reloc_reserved = false; |
5039eddc | 487 | h->sync = false; |
bed92eae | 488 | INIT_LIST_HEAD(&h->qgroup_ref_list); |
ea658bad | 489 | INIT_LIST_HEAD(&h->new_bgs); |
b7ec40d7 | 490 | |
a22285a6 | 491 | smp_mb(); |
4a9d8bde MX |
492 | if (cur_trans->state >= TRANS_STATE_BLOCKED && |
493 | may_wait_transaction(root, type)) { | |
a22285a6 YZ |
494 | btrfs_commit_transaction(h, root); |
495 | goto again; | |
496 | } | |
497 | ||
b5009945 | 498 | if (num_bytes) { |
8c2a3ca2 | 499 | trace_btrfs_space_reservation(root->fs_info, "transaction", |
2bcc0328 | 500 | h->transid, num_bytes, 1); |
b5009945 JB |
501 | h->block_rsv = &root->fs_info->trans_block_rsv; |
502 | h->bytes_reserved = num_bytes; | |
20dd2cbf | 503 | h->reloc_reserved = reloc_reserved; |
a22285a6 | 504 | } |
4b824906 | 505 | h->qgroup_reserved = qgroup_reserved; |
9ed74f2d | 506 | |
2a1eb461 | 507 | got_it: |
a4abeea4 | 508 | btrfs_record_root_in_trans(h, root); |
a22285a6 YZ |
509 | |
510 | if (!current->journal_info && type != TRANS_USERSPACE) | |
511 | current->journal_info = h; | |
79154b1b | 512 | return h; |
843fcf35 MX |
513 | |
514 | join_fail: | |
0860adfd | 515 | if (type & __TRANS_FREEZABLE) |
843fcf35 MX |
516 | sb_end_intwrite(root->fs_info->sb); |
517 | kmem_cache_free(btrfs_trans_handle_cachep, h); | |
518 | alloc_fail: | |
519 | if (num_bytes) | |
520 | btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv, | |
521 | num_bytes); | |
522 | reserve_fail: | |
523 | if (qgroup_reserved) | |
524 | btrfs_qgroup_free(root, qgroup_reserved); | |
525 | return ERR_PTR(ret); | |
79154b1b CM |
526 | } |
527 | ||
f9295749 | 528 | struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, |
a22285a6 | 529 | int num_items) |
f9295749 | 530 | { |
08e007d2 MX |
531 | return start_transaction(root, num_items, TRANS_START, |
532 | BTRFS_RESERVE_FLUSH_ALL); | |
f9295749 | 533 | } |
8407aa46 | 534 | |
08e007d2 | 535 | struct btrfs_trans_handle *btrfs_start_transaction_lflush( |
8407aa46 MX |
536 | struct btrfs_root *root, int num_items) |
537 | { | |
08e007d2 MX |
538 | return start_transaction(root, num_items, TRANS_START, |
539 | BTRFS_RESERVE_FLUSH_LIMIT); | |
8407aa46 MX |
540 | } |
541 | ||
7a7eaa40 | 542 | struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root) |
f9295749 | 543 | { |
8407aa46 | 544 | return start_transaction(root, 0, TRANS_JOIN, 0); |
f9295749 CM |
545 | } |
546 | ||
7a7eaa40 | 547 | struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root) |
0af3d00b | 548 | { |
8407aa46 | 549 | return start_transaction(root, 0, TRANS_JOIN_NOLOCK, 0); |
0af3d00b JB |
550 | } |
551 | ||
7a7eaa40 | 552 | struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root) |
9ca9ee09 | 553 | { |
8407aa46 | 554 | return start_transaction(root, 0, TRANS_USERSPACE, 0); |
9ca9ee09 SW |
555 | } |
556 | ||
d4edf39b MX |
557 | /* |
558 | * btrfs_attach_transaction() - catch the running transaction | |
559 | * | |
560 | * It is used when we want to commit the current the transaction, but | |
561 | * don't want to start a new one. | |
562 | * | |
563 | * Note: If this function return -ENOENT, it just means there is no | |
564 | * running transaction. But it is possible that the inactive transaction | |
565 | * is still in the memory, not fully on disk. If you hope there is no | |
566 | * inactive transaction in the fs when -ENOENT is returned, you should | |
567 | * invoke | |
568 | * btrfs_attach_transaction_barrier() | |
569 | */ | |
354aa0fb | 570 | struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root) |
60376ce4 | 571 | { |
354aa0fb | 572 | return start_transaction(root, 0, TRANS_ATTACH, 0); |
60376ce4 JB |
573 | } |
574 | ||
d4edf39b | 575 | /* |
90b6d283 | 576 | * btrfs_attach_transaction_barrier() - catch the running transaction |
d4edf39b MX |
577 | * |
578 | * It is similar to the above function, the differentia is this one | |
579 | * will wait for all the inactive transactions until they fully | |
580 | * complete. | |
581 | */ | |
582 | struct btrfs_trans_handle * | |
583 | btrfs_attach_transaction_barrier(struct btrfs_root *root) | |
584 | { | |
585 | struct btrfs_trans_handle *trans; | |
586 | ||
587 | trans = start_transaction(root, 0, TRANS_ATTACH, 0); | |
588 | if (IS_ERR(trans) && PTR_ERR(trans) == -ENOENT) | |
589 | btrfs_wait_for_commit(root, 0); | |
590 | ||
591 | return trans; | |
592 | } | |
593 | ||
d352ac68 | 594 | /* wait for a transaction commit to be fully complete */ |
b9c8300c | 595 | static noinline void wait_for_commit(struct btrfs_root *root, |
89ce8a63 CM |
596 | struct btrfs_transaction *commit) |
597 | { | |
4a9d8bde | 598 | wait_event(commit->commit_wait, commit->state == TRANS_STATE_COMPLETED); |
89ce8a63 CM |
599 | } |
600 | ||
46204592 SW |
601 | int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid) |
602 | { | |
603 | struct btrfs_transaction *cur_trans = NULL, *t; | |
8cd2807f | 604 | int ret = 0; |
46204592 | 605 | |
46204592 SW |
606 | if (transid) { |
607 | if (transid <= root->fs_info->last_trans_committed) | |
a4abeea4 | 608 | goto out; |
46204592 | 609 | |
8cd2807f | 610 | ret = -EINVAL; |
46204592 | 611 | /* find specified transaction */ |
a4abeea4 | 612 | spin_lock(&root->fs_info->trans_lock); |
46204592 SW |
613 | list_for_each_entry(t, &root->fs_info->trans_list, list) { |
614 | if (t->transid == transid) { | |
615 | cur_trans = t; | |
a4abeea4 | 616 | atomic_inc(&cur_trans->use_count); |
8cd2807f | 617 | ret = 0; |
46204592 SW |
618 | break; |
619 | } | |
8cd2807f MX |
620 | if (t->transid > transid) { |
621 | ret = 0; | |
46204592 | 622 | break; |
8cd2807f | 623 | } |
46204592 | 624 | } |
a4abeea4 | 625 | spin_unlock(&root->fs_info->trans_lock); |
8cd2807f | 626 | /* The specified transaction doesn't exist */ |
46204592 | 627 | if (!cur_trans) |
8cd2807f | 628 | goto out; |
46204592 SW |
629 | } else { |
630 | /* find newest transaction that is committing | committed */ | |
a4abeea4 | 631 | spin_lock(&root->fs_info->trans_lock); |
46204592 SW |
632 | list_for_each_entry_reverse(t, &root->fs_info->trans_list, |
633 | list) { | |
4a9d8bde MX |
634 | if (t->state >= TRANS_STATE_COMMIT_START) { |
635 | if (t->state == TRANS_STATE_COMPLETED) | |
3473f3c0 | 636 | break; |
46204592 | 637 | cur_trans = t; |
a4abeea4 | 638 | atomic_inc(&cur_trans->use_count); |
46204592 SW |
639 | break; |
640 | } | |
641 | } | |
a4abeea4 | 642 | spin_unlock(&root->fs_info->trans_lock); |
46204592 | 643 | if (!cur_trans) |
a4abeea4 | 644 | goto out; /* nothing committing|committed */ |
46204592 SW |
645 | } |
646 | ||
46204592 | 647 | wait_for_commit(root, cur_trans); |
724e2315 | 648 | btrfs_put_transaction(cur_trans); |
a4abeea4 | 649 | out: |
46204592 SW |
650 | return ret; |
651 | } | |
652 | ||
37d1aeee CM |
653 | void btrfs_throttle(struct btrfs_root *root) |
654 | { | |
a4abeea4 | 655 | if (!atomic_read(&root->fs_info->open_ioctl_trans)) |
9ca9ee09 | 656 | wait_current_trans(root); |
37d1aeee CM |
657 | } |
658 | ||
8929ecfa YZ |
659 | static int should_end_transaction(struct btrfs_trans_handle *trans, |
660 | struct btrfs_root *root) | |
661 | { | |
1be41b78 | 662 | if (root->fs_info->global_block_rsv.space_info->full && |
0a2b2a84 | 663 | btrfs_check_space_for_delayed_refs(trans, root)) |
1be41b78 | 664 | return 1; |
36ba022a | 665 | |
1be41b78 | 666 | return !!btrfs_block_rsv_check(root, &root->fs_info->global_block_rsv, 5); |
8929ecfa YZ |
667 | } |
668 | ||
669 | int btrfs_should_end_transaction(struct btrfs_trans_handle *trans, | |
670 | struct btrfs_root *root) | |
671 | { | |
672 | struct btrfs_transaction *cur_trans = trans->transaction; | |
673 | int updates; | |
49b25e05 | 674 | int err; |
8929ecfa | 675 | |
a4abeea4 | 676 | smp_mb(); |
4a9d8bde MX |
677 | if (cur_trans->state >= TRANS_STATE_BLOCKED || |
678 | cur_trans->delayed_refs.flushing) | |
8929ecfa YZ |
679 | return 1; |
680 | ||
681 | updates = trans->delayed_ref_updates; | |
682 | trans->delayed_ref_updates = 0; | |
49b25e05 JM |
683 | if (updates) { |
684 | err = btrfs_run_delayed_refs(trans, root, updates); | |
685 | if (err) /* Error code will also eval true */ | |
686 | return err; | |
687 | } | |
8929ecfa YZ |
688 | |
689 | return should_end_transaction(trans, root); | |
690 | } | |
691 | ||
89ce8a63 | 692 | static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, |
a698d075 | 693 | struct btrfs_root *root, int throttle) |
79154b1b | 694 | { |
8929ecfa | 695 | struct btrfs_transaction *cur_trans = trans->transaction; |
ab78c84d | 696 | struct btrfs_fs_info *info = root->fs_info; |
1be41b78 | 697 | unsigned long cur = trans->delayed_ref_updates; |
a698d075 | 698 | int lock = (trans->type != TRANS_JOIN_NOLOCK); |
4edc2ca3 | 699 | int err = 0; |
a79b7d4b | 700 | int must_run_delayed_refs = 0; |
c3e69d58 | 701 | |
3bbb24b2 JB |
702 | if (trans->use_count > 1) { |
703 | trans->use_count--; | |
2a1eb461 JB |
704 | trans->block_rsv = trans->orig_rsv; |
705 | return 0; | |
706 | } | |
707 | ||
b24e03db | 708 | btrfs_trans_release_metadata(trans, root); |
4c13d758 | 709 | trans->block_rsv = NULL; |
c5567237 | 710 | |
ea658bad JB |
711 | if (!list_empty(&trans->new_bgs)) |
712 | btrfs_create_pending_block_groups(trans, root); | |
713 | ||
1be41b78 | 714 | trans->delayed_ref_updates = 0; |
a79b7d4b CM |
715 | if (!trans->sync) { |
716 | must_run_delayed_refs = | |
717 | btrfs_should_throttle_delayed_refs(trans, root); | |
0a2b2a84 | 718 | cur = max_t(unsigned long, cur, 32); |
a79b7d4b CM |
719 | |
720 | /* | |
721 | * don't make the caller wait if they are from a NOLOCK | |
722 | * or ATTACH transaction, it will deadlock with commit | |
723 | */ | |
724 | if (must_run_delayed_refs == 1 && | |
725 | (trans->type & (__TRANS_JOIN_NOLOCK | __TRANS_ATTACH))) | |
726 | must_run_delayed_refs = 2; | |
56bec294 | 727 | } |
bb721703 | 728 | |
fcebe456 JB |
729 | if (trans->qgroup_reserved) { |
730 | /* | |
731 | * the same root has to be passed here between start_transaction | |
732 | * and end_transaction. Subvolume quota depends on this. | |
733 | */ | |
734 | btrfs_qgroup_free(trans->root, trans->qgroup_reserved); | |
735 | trans->qgroup_reserved = 0; | |
736 | } | |
737 | ||
0e721106 JB |
738 | btrfs_trans_release_metadata(trans, root); |
739 | trans->block_rsv = NULL; | |
56bec294 | 740 | |
ea658bad JB |
741 | if (!list_empty(&trans->new_bgs)) |
742 | btrfs_create_pending_block_groups(trans, root); | |
743 | ||
a4abeea4 | 744 | if (lock && !atomic_read(&root->fs_info->open_ioctl_trans) && |
4a9d8bde MX |
745 | should_end_transaction(trans, root) && |
746 | ACCESS_ONCE(cur_trans->state) == TRANS_STATE_RUNNING) { | |
747 | spin_lock(&info->trans_lock); | |
748 | if (cur_trans->state == TRANS_STATE_RUNNING) | |
749 | cur_trans->state = TRANS_STATE_BLOCKED; | |
750 | spin_unlock(&info->trans_lock); | |
a4abeea4 | 751 | } |
8929ecfa | 752 | |
4a9d8bde | 753 | if (lock && ACCESS_ONCE(cur_trans->state) == TRANS_STATE_BLOCKED) { |
3bbb24b2 | 754 | if (throttle) |
8929ecfa | 755 | return btrfs_commit_transaction(trans, root); |
3bbb24b2 | 756 | else |
8929ecfa YZ |
757 | wake_up_process(info->transaction_kthread); |
758 | } | |
759 | ||
0860adfd | 760 | if (trans->type & __TRANS_FREEZABLE) |
98114659 | 761 | sb_end_intwrite(root->fs_info->sb); |
6df7881a | 762 | |
8929ecfa | 763 | WARN_ON(cur_trans != info->running_transaction); |
13c5a93e JB |
764 | WARN_ON(atomic_read(&cur_trans->num_writers) < 1); |
765 | atomic_dec(&cur_trans->num_writers); | |
0860adfd | 766 | extwriter_counter_dec(cur_trans, trans->type); |
89ce8a63 | 767 | |
99d16cbc | 768 | smp_mb(); |
79154b1b CM |
769 | if (waitqueue_active(&cur_trans->writer_wait)) |
770 | wake_up(&cur_trans->writer_wait); | |
724e2315 | 771 | btrfs_put_transaction(cur_trans); |
9ed74f2d JB |
772 | |
773 | if (current->journal_info == trans) | |
774 | current->journal_info = NULL; | |
ab78c84d | 775 | |
24bbcf04 YZ |
776 | if (throttle) |
777 | btrfs_run_delayed_iputs(root); | |
778 | ||
49b25e05 | 779 | if (trans->aborted || |
4e121c06 JB |
780 | test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) { |
781 | wake_up_process(info->transaction_kthread); | |
4edc2ca3 | 782 | err = -EIO; |
4e121c06 | 783 | } |
edf39272 | 784 | assert_qgroups_uptodate(trans); |
49b25e05 | 785 | |
4edc2ca3 | 786 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
a79b7d4b CM |
787 | if (must_run_delayed_refs) { |
788 | btrfs_async_run_delayed_refs(root, cur, | |
789 | must_run_delayed_refs == 1); | |
790 | } | |
4edc2ca3 | 791 | return err; |
79154b1b CM |
792 | } |
793 | ||
89ce8a63 CM |
794 | int btrfs_end_transaction(struct btrfs_trans_handle *trans, |
795 | struct btrfs_root *root) | |
796 | { | |
98ad43be | 797 | return __btrfs_end_transaction(trans, root, 0); |
89ce8a63 CM |
798 | } |
799 | ||
800 | int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans, | |
801 | struct btrfs_root *root) | |
802 | { | |
98ad43be | 803 | return __btrfs_end_transaction(trans, root, 1); |
16cdcec7 MX |
804 | } |
805 | ||
d352ac68 CM |
806 | /* |
807 | * when btree blocks are allocated, they have some corresponding bits set for | |
808 | * them in one of two extent_io trees. This is used to make sure all of | |
690587d1 | 809 | * those extents are sent to disk but does not wait on them |
d352ac68 | 810 | */ |
690587d1 | 811 | int btrfs_write_marked_extents(struct btrfs_root *root, |
8cef4e16 | 812 | struct extent_io_tree *dirty_pages, int mark) |
79154b1b | 813 | { |
777e6bd7 | 814 | int err = 0; |
7c4452b9 | 815 | int werr = 0; |
1728366e | 816 | struct address_space *mapping = root->fs_info->btree_inode->i_mapping; |
e6138876 | 817 | struct extent_state *cached_state = NULL; |
777e6bd7 | 818 | u64 start = 0; |
5f39d397 | 819 | u64 end; |
7c4452b9 | 820 | |
1728366e | 821 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
e6138876 JB |
822 | mark, &cached_state)) { |
823 | convert_extent_bit(dirty_pages, start, end, EXTENT_NEED_WAIT, | |
824 | mark, &cached_state, GFP_NOFS); | |
825 | cached_state = NULL; | |
1728366e JB |
826 | err = filemap_fdatawrite_range(mapping, start, end); |
827 | if (err) | |
828 | werr = err; | |
829 | cond_resched(); | |
830 | start = end + 1; | |
7c4452b9 | 831 | } |
690587d1 CM |
832 | if (err) |
833 | werr = err; | |
834 | return werr; | |
835 | } | |
836 | ||
837 | /* | |
838 | * when btree blocks are allocated, they have some corresponding bits set for | |
839 | * them in one of two extent_io trees. This is used to make sure all of | |
840 | * those extents are on disk for transaction or log commit. We wait | |
841 | * on all the pages and clear them from the dirty pages state tree | |
842 | */ | |
843 | int btrfs_wait_marked_extents(struct btrfs_root *root, | |
8cef4e16 | 844 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 | 845 | { |
690587d1 CM |
846 | int err = 0; |
847 | int werr = 0; | |
1728366e | 848 | struct address_space *mapping = root->fs_info->btree_inode->i_mapping; |
e6138876 | 849 | struct extent_state *cached_state = NULL; |
690587d1 CM |
850 | u64 start = 0; |
851 | u64 end; | |
777e6bd7 | 852 | |
1728366e | 853 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
e6138876 JB |
854 | EXTENT_NEED_WAIT, &cached_state)) { |
855 | clear_extent_bit(dirty_pages, start, end, EXTENT_NEED_WAIT, | |
856 | 0, 0, &cached_state, GFP_NOFS); | |
1728366e JB |
857 | err = filemap_fdatawait_range(mapping, start, end); |
858 | if (err) | |
859 | werr = err; | |
860 | cond_resched(); | |
861 | start = end + 1; | |
777e6bd7 | 862 | } |
7c4452b9 CM |
863 | if (err) |
864 | werr = err; | |
865 | return werr; | |
79154b1b CM |
866 | } |
867 | ||
690587d1 CM |
868 | /* |
869 | * when btree blocks are allocated, they have some corresponding bits set for | |
870 | * them in one of two extent_io trees. This is used to make sure all of | |
871 | * those extents are on disk for transaction or log commit | |
872 | */ | |
171170c1 | 873 | static int btrfs_write_and_wait_marked_extents(struct btrfs_root *root, |
8cef4e16 | 874 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 CM |
875 | { |
876 | int ret; | |
877 | int ret2; | |
c6adc9cc | 878 | struct blk_plug plug; |
690587d1 | 879 | |
c6adc9cc | 880 | blk_start_plug(&plug); |
8cef4e16 | 881 | ret = btrfs_write_marked_extents(root, dirty_pages, mark); |
c6adc9cc | 882 | blk_finish_plug(&plug); |
8cef4e16 | 883 | ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark); |
bf0da8c1 CM |
884 | |
885 | if (ret) | |
886 | return ret; | |
887 | if (ret2) | |
888 | return ret2; | |
889 | return 0; | |
690587d1 CM |
890 | } |
891 | ||
d0c803c4 CM |
892 | int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans, |
893 | struct btrfs_root *root) | |
894 | { | |
895 | if (!trans || !trans->transaction) { | |
896 | struct inode *btree_inode; | |
897 | btree_inode = root->fs_info->btree_inode; | |
898 | return filemap_write_and_wait(btree_inode->i_mapping); | |
899 | } | |
900 | return btrfs_write_and_wait_marked_extents(root, | |
8cef4e16 YZ |
901 | &trans->transaction->dirty_pages, |
902 | EXTENT_DIRTY); | |
d0c803c4 CM |
903 | } |
904 | ||
d352ac68 CM |
905 | /* |
906 | * this is used to update the root pointer in the tree of tree roots. | |
907 | * | |
908 | * But, in the case of the extent allocation tree, updating the root | |
909 | * pointer may allocate blocks which may change the root of the extent | |
910 | * allocation tree. | |
911 | * | |
912 | * So, this loops and repeats and makes sure the cowonly root didn't | |
913 | * change while the root pointer was being updated in the metadata. | |
914 | */ | |
0b86a832 CM |
915 | static int update_cowonly_root(struct btrfs_trans_handle *trans, |
916 | struct btrfs_root *root) | |
79154b1b CM |
917 | { |
918 | int ret; | |
0b86a832 | 919 | u64 old_root_bytenr; |
86b9f2ec | 920 | u64 old_root_used; |
0b86a832 | 921 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
79154b1b | 922 | |
86b9f2ec | 923 | old_root_used = btrfs_root_used(&root->root_item); |
0b86a832 | 924 | btrfs_write_dirty_block_groups(trans, root); |
56bec294 | 925 | |
d397712b | 926 | while (1) { |
0b86a832 | 927 | old_root_bytenr = btrfs_root_bytenr(&root->root_item); |
86b9f2ec YZ |
928 | if (old_root_bytenr == root->node->start && |
929 | old_root_used == btrfs_root_used(&root->root_item)) | |
79154b1b | 930 | break; |
87ef2bb4 | 931 | |
5d4f98a2 | 932 | btrfs_set_root_node(&root->root_item, root->node); |
79154b1b | 933 | ret = btrfs_update_root(trans, tree_root, |
0b86a832 CM |
934 | &root->root_key, |
935 | &root->root_item); | |
49b25e05 JM |
936 | if (ret) |
937 | return ret; | |
56bec294 | 938 | |
86b9f2ec | 939 | old_root_used = btrfs_root_used(&root->root_item); |
4a8c9a62 | 940 | ret = btrfs_write_dirty_block_groups(trans, root); |
49b25e05 JM |
941 | if (ret) |
942 | return ret; | |
0b86a832 | 943 | } |
276e680d | 944 | |
0b86a832 CM |
945 | return 0; |
946 | } | |
947 | ||
d352ac68 CM |
948 | /* |
949 | * update all the cowonly tree roots on disk | |
49b25e05 JM |
950 | * |
951 | * The error handling in this function may not be obvious. Any of the | |
952 | * failures will cause the file system to go offline. We still need | |
953 | * to clean up the delayed refs. | |
d352ac68 | 954 | */ |
5d4f98a2 YZ |
955 | static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans, |
956 | struct btrfs_root *root) | |
0b86a832 CM |
957 | { |
958 | struct btrfs_fs_info *fs_info = root->fs_info; | |
959 | struct list_head *next; | |
84234f3a | 960 | struct extent_buffer *eb; |
56bec294 | 961 | int ret; |
84234f3a | 962 | |
56bec294 | 963 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
49b25e05 JM |
964 | if (ret) |
965 | return ret; | |
87ef2bb4 | 966 | |
84234f3a | 967 | eb = btrfs_lock_root_node(fs_info->tree_root); |
49b25e05 JM |
968 | ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, |
969 | 0, &eb); | |
84234f3a YZ |
970 | btrfs_tree_unlock(eb); |
971 | free_extent_buffer(eb); | |
0b86a832 | 972 | |
49b25e05 JM |
973 | if (ret) |
974 | return ret; | |
975 | ||
56bec294 | 976 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
49b25e05 JM |
977 | if (ret) |
978 | return ret; | |
87ef2bb4 | 979 | |
733f4fbb | 980 | ret = btrfs_run_dev_stats(trans, root->fs_info); |
c16ce190 JB |
981 | if (ret) |
982 | return ret; | |
8dabb742 | 983 | ret = btrfs_run_dev_replace(trans, root->fs_info); |
c16ce190 JB |
984 | if (ret) |
985 | return ret; | |
546adb0d | 986 | ret = btrfs_run_qgroups(trans, root->fs_info); |
c16ce190 JB |
987 | if (ret) |
988 | return ret; | |
546adb0d JS |
989 | |
990 | /* run_qgroups might have added some more refs */ | |
991 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); | |
c16ce190 JB |
992 | if (ret) |
993 | return ret; | |
546adb0d | 994 | |
d397712b | 995 | while (!list_empty(&fs_info->dirty_cowonly_roots)) { |
0b86a832 CM |
996 | next = fs_info->dirty_cowonly_roots.next; |
997 | list_del_init(next); | |
998 | root = list_entry(next, struct btrfs_root, dirty_list); | |
87ef2bb4 | 999 | |
9e351cc8 JB |
1000 | if (root != fs_info->extent_root) |
1001 | list_add_tail(&root->dirty_list, | |
1002 | &trans->transaction->switch_commits); | |
49b25e05 JM |
1003 | ret = update_cowonly_root(trans, root); |
1004 | if (ret) | |
1005 | return ret; | |
79154b1b | 1006 | } |
276e680d | 1007 | |
9e351cc8 JB |
1008 | list_add_tail(&fs_info->extent_root->dirty_list, |
1009 | &trans->transaction->switch_commits); | |
8dabb742 SB |
1010 | btrfs_after_dev_replace_commit(fs_info); |
1011 | ||
79154b1b CM |
1012 | return 0; |
1013 | } | |
1014 | ||
d352ac68 CM |
1015 | /* |
1016 | * dead roots are old snapshots that need to be deleted. This allocates | |
1017 | * a dirty root struct and adds it into the list of dead roots that need to | |
1018 | * be deleted | |
1019 | */ | |
cfad392b | 1020 | void btrfs_add_dead_root(struct btrfs_root *root) |
5eda7b5e | 1021 | { |
a4abeea4 | 1022 | spin_lock(&root->fs_info->trans_lock); |
cfad392b JB |
1023 | if (list_empty(&root->root_list)) |
1024 | list_add_tail(&root->root_list, &root->fs_info->dead_roots); | |
a4abeea4 | 1025 | spin_unlock(&root->fs_info->trans_lock); |
5eda7b5e CM |
1026 | } |
1027 | ||
d352ac68 | 1028 | /* |
5d4f98a2 | 1029 | * update all the cowonly tree roots on disk |
d352ac68 | 1030 | */ |
5d4f98a2 YZ |
1031 | static noinline int commit_fs_roots(struct btrfs_trans_handle *trans, |
1032 | struct btrfs_root *root) | |
0f7d52f4 | 1033 | { |
0f7d52f4 | 1034 | struct btrfs_root *gang[8]; |
5d4f98a2 | 1035 | struct btrfs_fs_info *fs_info = root->fs_info; |
0f7d52f4 CM |
1036 | int i; |
1037 | int ret; | |
54aa1f4d CM |
1038 | int err = 0; |
1039 | ||
a4abeea4 | 1040 | spin_lock(&fs_info->fs_roots_radix_lock); |
d397712b | 1041 | while (1) { |
5d4f98a2 YZ |
1042 | ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, |
1043 | (void **)gang, 0, | |
0f7d52f4 CM |
1044 | ARRAY_SIZE(gang), |
1045 | BTRFS_ROOT_TRANS_TAG); | |
1046 | if (ret == 0) | |
1047 | break; | |
1048 | for (i = 0; i < ret; i++) { | |
1049 | root = gang[i]; | |
5d4f98a2 YZ |
1050 | radix_tree_tag_clear(&fs_info->fs_roots_radix, |
1051 | (unsigned long)root->root_key.objectid, | |
1052 | BTRFS_ROOT_TRANS_TAG); | |
a4abeea4 | 1053 | spin_unlock(&fs_info->fs_roots_radix_lock); |
31153d81 | 1054 | |
e02119d5 | 1055 | btrfs_free_log(trans, root); |
5d4f98a2 | 1056 | btrfs_update_reloc_root(trans, root); |
d68fc57b | 1057 | btrfs_orphan_commit_root(trans, root); |
bcc63abb | 1058 | |
82d5902d LZ |
1059 | btrfs_save_ino_cache(root, trans); |
1060 | ||
f1ebcc74 | 1061 | /* see comments in should_cow_block() */ |
27cdeb70 MX |
1062 | clear_bit(BTRFS_ROOT_FORCE_COW, &root->state); |
1063 | smp_mb__after_clear_bit(); | |
f1ebcc74 | 1064 | |
978d910d | 1065 | if (root->commit_root != root->node) { |
9e351cc8 JB |
1066 | list_add_tail(&root->dirty_list, |
1067 | &trans->transaction->switch_commits); | |
978d910d YZ |
1068 | btrfs_set_root_node(&root->root_item, |
1069 | root->node); | |
1070 | } | |
5d4f98a2 | 1071 | |
5d4f98a2 | 1072 | err = btrfs_update_root(trans, fs_info->tree_root, |
0f7d52f4 CM |
1073 | &root->root_key, |
1074 | &root->root_item); | |
a4abeea4 | 1075 | spin_lock(&fs_info->fs_roots_radix_lock); |
54aa1f4d CM |
1076 | if (err) |
1077 | break; | |
0f7d52f4 CM |
1078 | } |
1079 | } | |
a4abeea4 | 1080 | spin_unlock(&fs_info->fs_roots_radix_lock); |
54aa1f4d | 1081 | return err; |
0f7d52f4 CM |
1082 | } |
1083 | ||
d352ac68 | 1084 | /* |
de78b51a ES |
1085 | * defrag a given btree. |
1086 | * Every leaf in the btree is read and defragged. | |
d352ac68 | 1087 | */ |
de78b51a | 1088 | int btrfs_defrag_root(struct btrfs_root *root) |
e9d0b13b CM |
1089 | { |
1090 | struct btrfs_fs_info *info = root->fs_info; | |
e9d0b13b | 1091 | struct btrfs_trans_handle *trans; |
8929ecfa | 1092 | int ret; |
e9d0b13b | 1093 | |
27cdeb70 | 1094 | if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state)) |
e9d0b13b | 1095 | return 0; |
8929ecfa | 1096 | |
6b80053d | 1097 | while (1) { |
8929ecfa YZ |
1098 | trans = btrfs_start_transaction(root, 0); |
1099 | if (IS_ERR(trans)) | |
1100 | return PTR_ERR(trans); | |
1101 | ||
de78b51a | 1102 | ret = btrfs_defrag_leaves(trans, root); |
8929ecfa | 1103 | |
e9d0b13b | 1104 | btrfs_end_transaction(trans, root); |
b53d3f5d | 1105 | btrfs_btree_balance_dirty(info->tree_root); |
e9d0b13b CM |
1106 | cond_resched(); |
1107 | ||
7841cb28 | 1108 | if (btrfs_fs_closing(root->fs_info) || ret != -EAGAIN) |
e9d0b13b | 1109 | break; |
210549eb DS |
1110 | |
1111 | if (btrfs_defrag_cancelled(root->fs_info)) { | |
efe120a0 | 1112 | pr_debug("BTRFS: defrag_root cancelled\n"); |
210549eb DS |
1113 | ret = -EAGAIN; |
1114 | break; | |
1115 | } | |
e9d0b13b | 1116 | } |
27cdeb70 | 1117 | clear_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state); |
8929ecfa | 1118 | return ret; |
e9d0b13b CM |
1119 | } |
1120 | ||
d352ac68 CM |
1121 | /* |
1122 | * new snapshots need to be created at a very specific time in the | |
aec8030a MX |
1123 | * transaction commit. This does the actual creation. |
1124 | * | |
1125 | * Note: | |
1126 | * If the error which may affect the commitment of the current transaction | |
1127 | * happens, we should return the error number. If the error which just affect | |
1128 | * the creation of the pending snapshots, just return 0. | |
d352ac68 | 1129 | */ |
80b6794d | 1130 | static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, |
3063d29f CM |
1131 | struct btrfs_fs_info *fs_info, |
1132 | struct btrfs_pending_snapshot *pending) | |
1133 | { | |
1134 | struct btrfs_key key; | |
80b6794d | 1135 | struct btrfs_root_item *new_root_item; |
3063d29f CM |
1136 | struct btrfs_root *tree_root = fs_info->tree_root; |
1137 | struct btrfs_root *root = pending->root; | |
6bdb72de | 1138 | struct btrfs_root *parent_root; |
98c9942a | 1139 | struct btrfs_block_rsv *rsv; |
6bdb72de | 1140 | struct inode *parent_inode; |
42874b3d MX |
1141 | struct btrfs_path *path; |
1142 | struct btrfs_dir_item *dir_item; | |
a22285a6 | 1143 | struct dentry *dentry; |
3063d29f | 1144 | struct extent_buffer *tmp; |
925baedd | 1145 | struct extent_buffer *old; |
8ea05e3a | 1146 | struct timespec cur_time = CURRENT_TIME; |
aec8030a | 1147 | int ret = 0; |
d68fc57b | 1148 | u64 to_reserve = 0; |
6bdb72de | 1149 | u64 index = 0; |
a22285a6 | 1150 | u64 objectid; |
b83cc969 | 1151 | u64 root_flags; |
8ea05e3a | 1152 | uuid_le new_uuid; |
3063d29f | 1153 | |
42874b3d MX |
1154 | path = btrfs_alloc_path(); |
1155 | if (!path) { | |
aec8030a MX |
1156 | pending->error = -ENOMEM; |
1157 | return 0; | |
42874b3d MX |
1158 | } |
1159 | ||
80b6794d CM |
1160 | new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS); |
1161 | if (!new_root_item) { | |
aec8030a | 1162 | pending->error = -ENOMEM; |
6fa9700e | 1163 | goto root_item_alloc_fail; |
80b6794d | 1164 | } |
a22285a6 | 1165 | |
aec8030a MX |
1166 | pending->error = btrfs_find_free_objectid(tree_root, &objectid); |
1167 | if (pending->error) | |
6fa9700e | 1168 | goto no_free_objectid; |
3063d29f | 1169 | |
3fd0a558 | 1170 | btrfs_reloc_pre_snapshot(trans, pending, &to_reserve); |
d68fc57b YZ |
1171 | |
1172 | if (to_reserve > 0) { | |
aec8030a MX |
1173 | pending->error = btrfs_block_rsv_add(root, |
1174 | &pending->block_rsv, | |
1175 | to_reserve, | |
1176 | BTRFS_RESERVE_NO_FLUSH); | |
1177 | if (pending->error) | |
6fa9700e | 1178 | goto no_free_objectid; |
d68fc57b YZ |
1179 | } |
1180 | ||
3063d29f | 1181 | key.objectid = objectid; |
a22285a6 YZ |
1182 | key.offset = (u64)-1; |
1183 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3063d29f | 1184 | |
6fa9700e | 1185 | rsv = trans->block_rsv; |
a22285a6 | 1186 | trans->block_rsv = &pending->block_rsv; |
2382c5cc | 1187 | trans->bytes_reserved = trans->block_rsv->reserved; |
3de4586c | 1188 | |
a22285a6 | 1189 | dentry = pending->dentry; |
e9662f70 | 1190 | parent_inode = pending->dir; |
a22285a6 | 1191 | parent_root = BTRFS_I(parent_inode)->root; |
7585717f | 1192 | record_root_in_trans(trans, parent_root); |
a22285a6 | 1193 | |
3063d29f CM |
1194 | /* |
1195 | * insert the directory item | |
1196 | */ | |
3de4586c | 1197 | ret = btrfs_set_inode_index(parent_inode, &index); |
49b25e05 | 1198 | BUG_ON(ret); /* -ENOMEM */ |
42874b3d MX |
1199 | |
1200 | /* check if there is a file/dir which has the same name. */ | |
1201 | dir_item = btrfs_lookup_dir_item(NULL, parent_root, path, | |
1202 | btrfs_ino(parent_inode), | |
1203 | dentry->d_name.name, | |
1204 | dentry->d_name.len, 0); | |
1205 | if (dir_item != NULL && !IS_ERR(dir_item)) { | |
fe66a05a | 1206 | pending->error = -EEXIST; |
aec8030a | 1207 | goto dir_item_existed; |
42874b3d MX |
1208 | } else if (IS_ERR(dir_item)) { |
1209 | ret = PTR_ERR(dir_item); | |
8732d44f MX |
1210 | btrfs_abort_transaction(trans, root, ret); |
1211 | goto fail; | |
79787eaa | 1212 | } |
42874b3d | 1213 | btrfs_release_path(path); |
52c26179 | 1214 | |
e999376f CM |
1215 | /* |
1216 | * pull in the delayed directory update | |
1217 | * and the delayed inode item | |
1218 | * otherwise we corrupt the FS during | |
1219 | * snapshot | |
1220 | */ | |
1221 | ret = btrfs_run_delayed_items(trans, root); | |
8732d44f MX |
1222 | if (ret) { /* Transaction aborted */ |
1223 | btrfs_abort_transaction(trans, root, ret); | |
1224 | goto fail; | |
1225 | } | |
e999376f | 1226 | |
7585717f | 1227 | record_root_in_trans(trans, root); |
6bdb72de SW |
1228 | btrfs_set_root_last_snapshot(&root->root_item, trans->transid); |
1229 | memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); | |
08fe4db1 | 1230 | btrfs_check_and_init_root_item(new_root_item); |
6bdb72de | 1231 | |
b83cc969 LZ |
1232 | root_flags = btrfs_root_flags(new_root_item); |
1233 | if (pending->readonly) | |
1234 | root_flags |= BTRFS_ROOT_SUBVOL_RDONLY; | |
1235 | else | |
1236 | root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY; | |
1237 | btrfs_set_root_flags(new_root_item, root_flags); | |
1238 | ||
8ea05e3a AB |
1239 | btrfs_set_root_generation_v2(new_root_item, |
1240 | trans->transid); | |
1241 | uuid_le_gen(&new_uuid); | |
1242 | memcpy(new_root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE); | |
1243 | memcpy(new_root_item->parent_uuid, root->root_item.uuid, | |
1244 | BTRFS_UUID_SIZE); | |
70023da2 SB |
1245 | if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) { |
1246 | memset(new_root_item->received_uuid, 0, | |
1247 | sizeof(new_root_item->received_uuid)); | |
1248 | memset(&new_root_item->stime, 0, sizeof(new_root_item->stime)); | |
1249 | memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime)); | |
1250 | btrfs_set_root_stransid(new_root_item, 0); | |
1251 | btrfs_set_root_rtransid(new_root_item, 0); | |
1252 | } | |
3cae210f QW |
1253 | btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec); |
1254 | btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec); | |
8ea05e3a | 1255 | btrfs_set_root_otransid(new_root_item, trans->transid); |
8ea05e3a | 1256 | |
6bdb72de | 1257 | old = btrfs_lock_root_node(root); |
49b25e05 | 1258 | ret = btrfs_cow_block(trans, root, old, NULL, 0, &old); |
79787eaa JM |
1259 | if (ret) { |
1260 | btrfs_tree_unlock(old); | |
1261 | free_extent_buffer(old); | |
8732d44f MX |
1262 | btrfs_abort_transaction(trans, root, ret); |
1263 | goto fail; | |
79787eaa | 1264 | } |
49b25e05 | 1265 | |
6bdb72de SW |
1266 | btrfs_set_lock_blocking(old); |
1267 | ||
49b25e05 | 1268 | ret = btrfs_copy_root(trans, root, old, &tmp, objectid); |
79787eaa | 1269 | /* clean up in any case */ |
6bdb72de SW |
1270 | btrfs_tree_unlock(old); |
1271 | free_extent_buffer(old); | |
8732d44f MX |
1272 | if (ret) { |
1273 | btrfs_abort_transaction(trans, root, ret); | |
1274 | goto fail; | |
1275 | } | |
6bdb72de | 1276 | |
fcebe456 JB |
1277 | /* |
1278 | * We need to flush delayed refs in order to make sure all of our quota | |
1279 | * operations have been done before we call btrfs_qgroup_inherit. | |
1280 | */ | |
1281 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); | |
1282 | if (ret) { | |
1283 | btrfs_abort_transaction(trans, root, ret); | |
1284 | goto fail; | |
1285 | } | |
1286 | ||
1287 | pending->error = btrfs_qgroup_inherit(trans, fs_info, | |
1288 | root->root_key.objectid, | |
1289 | objectid, pending->inherit); | |
1290 | if (pending->error) | |
1291 | goto no_free_objectid; | |
1292 | ||
f1ebcc74 | 1293 | /* see comments in should_cow_block() */ |
27cdeb70 | 1294 | set_bit(BTRFS_ROOT_FORCE_COW, &root->state); |
f1ebcc74 LB |
1295 | smp_wmb(); |
1296 | ||
6bdb72de | 1297 | btrfs_set_root_node(new_root_item, tmp); |
a22285a6 YZ |
1298 | /* record when the snapshot was created in key.offset */ |
1299 | key.offset = trans->transid; | |
1300 | ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); | |
6bdb72de SW |
1301 | btrfs_tree_unlock(tmp); |
1302 | free_extent_buffer(tmp); | |
8732d44f MX |
1303 | if (ret) { |
1304 | btrfs_abort_transaction(trans, root, ret); | |
1305 | goto fail; | |
1306 | } | |
6bdb72de | 1307 | |
a22285a6 YZ |
1308 | /* |
1309 | * insert root back/forward references | |
1310 | */ | |
1311 | ret = btrfs_add_root_ref(trans, tree_root, objectid, | |
0660b5af | 1312 | parent_root->root_key.objectid, |
33345d01 | 1313 | btrfs_ino(parent_inode), index, |
a22285a6 | 1314 | dentry->d_name.name, dentry->d_name.len); |
8732d44f MX |
1315 | if (ret) { |
1316 | btrfs_abort_transaction(trans, root, ret); | |
1317 | goto fail; | |
1318 | } | |
0660b5af | 1319 | |
a22285a6 YZ |
1320 | key.offset = (u64)-1; |
1321 | pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key); | |
79787eaa JM |
1322 | if (IS_ERR(pending->snap)) { |
1323 | ret = PTR_ERR(pending->snap); | |
8732d44f MX |
1324 | btrfs_abort_transaction(trans, root, ret); |
1325 | goto fail; | |
79787eaa | 1326 | } |
d68fc57b | 1327 | |
49b25e05 | 1328 | ret = btrfs_reloc_post_snapshot(trans, pending); |
8732d44f MX |
1329 | if (ret) { |
1330 | btrfs_abort_transaction(trans, root, ret); | |
1331 | goto fail; | |
1332 | } | |
361048f5 MX |
1333 | |
1334 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); | |
8732d44f MX |
1335 | if (ret) { |
1336 | btrfs_abort_transaction(trans, root, ret); | |
1337 | goto fail; | |
1338 | } | |
42874b3d MX |
1339 | |
1340 | ret = btrfs_insert_dir_item(trans, parent_root, | |
1341 | dentry->d_name.name, dentry->d_name.len, | |
1342 | parent_inode, &key, | |
1343 | BTRFS_FT_DIR, index); | |
1344 | /* We have check then name at the beginning, so it is impossible. */ | |
9c52057c | 1345 | BUG_ON(ret == -EEXIST || ret == -EOVERFLOW); |
8732d44f MX |
1346 | if (ret) { |
1347 | btrfs_abort_transaction(trans, root, ret); | |
1348 | goto fail; | |
1349 | } | |
42874b3d MX |
1350 | |
1351 | btrfs_i_size_write(parent_inode, parent_inode->i_size + | |
1352 | dentry->d_name.len * 2); | |
1353 | parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME; | |
be6aef60 | 1354 | ret = btrfs_update_inode_fallback(trans, parent_root, parent_inode); |
dd5f9615 SB |
1355 | if (ret) { |
1356 | btrfs_abort_transaction(trans, root, ret); | |
1357 | goto fail; | |
1358 | } | |
1359 | ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root, new_uuid.b, | |
1360 | BTRFS_UUID_KEY_SUBVOL, objectid); | |
1361 | if (ret) { | |
8732d44f | 1362 | btrfs_abort_transaction(trans, root, ret); |
dd5f9615 SB |
1363 | goto fail; |
1364 | } | |
1365 | if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) { | |
1366 | ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root, | |
1367 | new_root_item->received_uuid, | |
1368 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, | |
1369 | objectid); | |
1370 | if (ret && ret != -EEXIST) { | |
1371 | btrfs_abort_transaction(trans, root, ret); | |
1372 | goto fail; | |
1373 | } | |
1374 | } | |
3063d29f | 1375 | fail: |
aec8030a MX |
1376 | pending->error = ret; |
1377 | dir_item_existed: | |
98c9942a | 1378 | trans->block_rsv = rsv; |
2382c5cc | 1379 | trans->bytes_reserved = 0; |
6fa9700e MX |
1380 | no_free_objectid: |
1381 | kfree(new_root_item); | |
1382 | root_item_alloc_fail: | |
42874b3d | 1383 | btrfs_free_path(path); |
49b25e05 | 1384 | return ret; |
3063d29f CM |
1385 | } |
1386 | ||
d352ac68 CM |
1387 | /* |
1388 | * create all the snapshots we've scheduled for creation | |
1389 | */ | |
80b6794d CM |
1390 | static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans, |
1391 | struct btrfs_fs_info *fs_info) | |
3de4586c | 1392 | { |
aec8030a | 1393 | struct btrfs_pending_snapshot *pending, *next; |
3de4586c | 1394 | struct list_head *head = &trans->transaction->pending_snapshots; |
aec8030a | 1395 | int ret = 0; |
3de4586c | 1396 | |
aec8030a MX |
1397 | list_for_each_entry_safe(pending, next, head, list) { |
1398 | list_del(&pending->list); | |
1399 | ret = create_pending_snapshot(trans, fs_info, pending); | |
1400 | if (ret) | |
1401 | break; | |
1402 | } | |
1403 | return ret; | |
3de4586c CM |
1404 | } |
1405 | ||
5d4f98a2 YZ |
1406 | static void update_super_roots(struct btrfs_root *root) |
1407 | { | |
1408 | struct btrfs_root_item *root_item; | |
1409 | struct btrfs_super_block *super; | |
1410 | ||
6c41761f | 1411 | super = root->fs_info->super_copy; |
5d4f98a2 YZ |
1412 | |
1413 | root_item = &root->fs_info->chunk_root->root_item; | |
1414 | super->chunk_root = root_item->bytenr; | |
1415 | super->chunk_root_generation = root_item->generation; | |
1416 | super->chunk_root_level = root_item->level; | |
1417 | ||
1418 | root_item = &root->fs_info->tree_root->root_item; | |
1419 | super->root = root_item->bytenr; | |
1420 | super->generation = root_item->generation; | |
1421 | super->root_level = root_item->level; | |
73bc1876 | 1422 | if (btrfs_test_opt(root, SPACE_CACHE)) |
0af3d00b | 1423 | super->cache_generation = root_item->generation; |
70f80175 SB |
1424 | if (root->fs_info->update_uuid_tree_gen) |
1425 | super->uuid_tree_generation = root_item->generation; | |
5d4f98a2 YZ |
1426 | } |
1427 | ||
f36f3042 CM |
1428 | int btrfs_transaction_in_commit(struct btrfs_fs_info *info) |
1429 | { | |
4a9d8bde | 1430 | struct btrfs_transaction *trans; |
f36f3042 | 1431 | int ret = 0; |
4a9d8bde | 1432 | |
a4abeea4 | 1433 | spin_lock(&info->trans_lock); |
4a9d8bde MX |
1434 | trans = info->running_transaction; |
1435 | if (trans) | |
1436 | ret = (trans->state >= TRANS_STATE_COMMIT_START); | |
a4abeea4 | 1437 | spin_unlock(&info->trans_lock); |
f36f3042 CM |
1438 | return ret; |
1439 | } | |
1440 | ||
8929ecfa YZ |
1441 | int btrfs_transaction_blocked(struct btrfs_fs_info *info) |
1442 | { | |
4a9d8bde | 1443 | struct btrfs_transaction *trans; |
8929ecfa | 1444 | int ret = 0; |
4a9d8bde | 1445 | |
a4abeea4 | 1446 | spin_lock(&info->trans_lock); |
4a9d8bde MX |
1447 | trans = info->running_transaction; |
1448 | if (trans) | |
1449 | ret = is_transaction_blocked(trans); | |
a4abeea4 | 1450 | spin_unlock(&info->trans_lock); |
8929ecfa YZ |
1451 | return ret; |
1452 | } | |
1453 | ||
bb9c12c9 SW |
1454 | /* |
1455 | * wait for the current transaction commit to start and block subsequent | |
1456 | * transaction joins | |
1457 | */ | |
1458 | static void wait_current_trans_commit_start(struct btrfs_root *root, | |
1459 | struct btrfs_transaction *trans) | |
1460 | { | |
4a9d8bde | 1461 | wait_event(root->fs_info->transaction_blocked_wait, |
501407aa JB |
1462 | trans->state >= TRANS_STATE_COMMIT_START || |
1463 | trans->aborted); | |
bb9c12c9 SW |
1464 | } |
1465 | ||
1466 | /* | |
1467 | * wait for the current transaction to start and then become unblocked. | |
1468 | * caller holds ref. | |
1469 | */ | |
1470 | static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root, | |
1471 | struct btrfs_transaction *trans) | |
1472 | { | |
72d63ed6 | 1473 | wait_event(root->fs_info->transaction_wait, |
501407aa JB |
1474 | trans->state >= TRANS_STATE_UNBLOCKED || |
1475 | trans->aborted); | |
bb9c12c9 SW |
1476 | } |
1477 | ||
1478 | /* | |
1479 | * commit transactions asynchronously. once btrfs_commit_transaction_async | |
1480 | * returns, any subsequent transaction will not be allowed to join. | |
1481 | */ | |
1482 | struct btrfs_async_commit { | |
1483 | struct btrfs_trans_handle *newtrans; | |
1484 | struct btrfs_root *root; | |
7892b5af | 1485 | struct work_struct work; |
bb9c12c9 SW |
1486 | }; |
1487 | ||
1488 | static void do_async_commit(struct work_struct *work) | |
1489 | { | |
1490 | struct btrfs_async_commit *ac = | |
7892b5af | 1491 | container_of(work, struct btrfs_async_commit, work); |
bb9c12c9 | 1492 | |
6fc4e354 SW |
1493 | /* |
1494 | * We've got freeze protection passed with the transaction. | |
1495 | * Tell lockdep about it. | |
1496 | */ | |
b1a06a4b | 1497 | if (ac->newtrans->type & __TRANS_FREEZABLE) |
ff7c1d33 MX |
1498 | rwsem_acquire_read( |
1499 | &ac->root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1], | |
1500 | 0, 1, _THIS_IP_); | |
6fc4e354 | 1501 | |
e209db7a SW |
1502 | current->journal_info = ac->newtrans; |
1503 | ||
bb9c12c9 SW |
1504 | btrfs_commit_transaction(ac->newtrans, ac->root); |
1505 | kfree(ac); | |
1506 | } | |
1507 | ||
1508 | int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans, | |
1509 | struct btrfs_root *root, | |
1510 | int wait_for_unblock) | |
1511 | { | |
1512 | struct btrfs_async_commit *ac; | |
1513 | struct btrfs_transaction *cur_trans; | |
1514 | ||
1515 | ac = kmalloc(sizeof(*ac), GFP_NOFS); | |
db5b493a TI |
1516 | if (!ac) |
1517 | return -ENOMEM; | |
bb9c12c9 | 1518 | |
7892b5af | 1519 | INIT_WORK(&ac->work, do_async_commit); |
bb9c12c9 | 1520 | ac->root = root; |
7a7eaa40 | 1521 | ac->newtrans = btrfs_join_transaction(root); |
3612b495 TI |
1522 | if (IS_ERR(ac->newtrans)) { |
1523 | int err = PTR_ERR(ac->newtrans); | |
1524 | kfree(ac); | |
1525 | return err; | |
1526 | } | |
bb9c12c9 SW |
1527 | |
1528 | /* take transaction reference */ | |
bb9c12c9 | 1529 | cur_trans = trans->transaction; |
13c5a93e | 1530 | atomic_inc(&cur_trans->use_count); |
bb9c12c9 SW |
1531 | |
1532 | btrfs_end_transaction(trans, root); | |
6fc4e354 SW |
1533 | |
1534 | /* | |
1535 | * Tell lockdep we've released the freeze rwsem, since the | |
1536 | * async commit thread will be the one to unlock it. | |
1537 | */ | |
b1a06a4b | 1538 | if (ac->newtrans->type & __TRANS_FREEZABLE) |
ff7c1d33 MX |
1539 | rwsem_release( |
1540 | &root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1], | |
1541 | 1, _THIS_IP_); | |
6fc4e354 | 1542 | |
7892b5af | 1543 | schedule_work(&ac->work); |
bb9c12c9 SW |
1544 | |
1545 | /* wait for transaction to start and unblock */ | |
bb9c12c9 SW |
1546 | if (wait_for_unblock) |
1547 | wait_current_trans_commit_start_and_unblock(root, cur_trans); | |
1548 | else | |
1549 | wait_current_trans_commit_start(root, cur_trans); | |
bb9c12c9 | 1550 | |
38e88054 SW |
1551 | if (current->journal_info == trans) |
1552 | current->journal_info = NULL; | |
1553 | ||
724e2315 | 1554 | btrfs_put_transaction(cur_trans); |
bb9c12c9 SW |
1555 | return 0; |
1556 | } | |
1557 | ||
49b25e05 JM |
1558 | |
1559 | static void cleanup_transaction(struct btrfs_trans_handle *trans, | |
7b8b92af | 1560 | struct btrfs_root *root, int err) |
49b25e05 JM |
1561 | { |
1562 | struct btrfs_transaction *cur_trans = trans->transaction; | |
f094ac32 | 1563 | DEFINE_WAIT(wait); |
49b25e05 JM |
1564 | |
1565 | WARN_ON(trans->use_count > 1); | |
1566 | ||
7b8b92af JB |
1567 | btrfs_abort_transaction(trans, root, err); |
1568 | ||
49b25e05 | 1569 | spin_lock(&root->fs_info->trans_lock); |
66b6135b | 1570 | |
25d8c284 MX |
1571 | /* |
1572 | * If the transaction is removed from the list, it means this | |
1573 | * transaction has been committed successfully, so it is impossible | |
1574 | * to call the cleanup function. | |
1575 | */ | |
1576 | BUG_ON(list_empty(&cur_trans->list)); | |
66b6135b | 1577 | |
49b25e05 | 1578 | list_del_init(&cur_trans->list); |
d7096fc3 | 1579 | if (cur_trans == root->fs_info->running_transaction) { |
4a9d8bde | 1580 | cur_trans->state = TRANS_STATE_COMMIT_DOING; |
f094ac32 LB |
1581 | spin_unlock(&root->fs_info->trans_lock); |
1582 | wait_event(cur_trans->writer_wait, | |
1583 | atomic_read(&cur_trans->num_writers) == 1); | |
1584 | ||
1585 | spin_lock(&root->fs_info->trans_lock); | |
d7096fc3 | 1586 | } |
49b25e05 JM |
1587 | spin_unlock(&root->fs_info->trans_lock); |
1588 | ||
1589 | btrfs_cleanup_one_transaction(trans->transaction, root); | |
1590 | ||
4a9d8bde MX |
1591 | spin_lock(&root->fs_info->trans_lock); |
1592 | if (cur_trans == root->fs_info->running_transaction) | |
1593 | root->fs_info->running_transaction = NULL; | |
1594 | spin_unlock(&root->fs_info->trans_lock); | |
1595 | ||
e0228285 JB |
1596 | if (trans->type & __TRANS_FREEZABLE) |
1597 | sb_end_intwrite(root->fs_info->sb); | |
724e2315 JB |
1598 | btrfs_put_transaction(cur_trans); |
1599 | btrfs_put_transaction(cur_trans); | |
49b25e05 JM |
1600 | |
1601 | trace_btrfs_transaction_commit(root); | |
1602 | ||
49b25e05 JM |
1603 | if (current->journal_info == trans) |
1604 | current->journal_info = NULL; | |
c0af8f0b | 1605 | btrfs_scrub_cancel(root->fs_info); |
49b25e05 JM |
1606 | |
1607 | kmem_cache_free(btrfs_trans_handle_cachep, trans); | |
1608 | } | |
1609 | ||
ca469637 MX |
1610 | static int btrfs_flush_all_pending_stuffs(struct btrfs_trans_handle *trans, |
1611 | struct btrfs_root *root) | |
1612 | { | |
ca469637 MX |
1613 | int ret; |
1614 | ||
ca469637 | 1615 | ret = btrfs_run_delayed_items(trans, root); |
ca469637 MX |
1616 | /* |
1617 | * running the delayed items may have added new refs. account | |
1618 | * them now so that they hinder processing of more delayed refs | |
1619 | * as little as possible. | |
1620 | */ | |
80d94fb3 FDBM |
1621 | if (ret) |
1622 | return ret; | |
ca469637 MX |
1623 | |
1624 | /* | |
1625 | * rename don't use btrfs_join_transaction, so, once we | |
1626 | * set the transaction to blocked above, we aren't going | |
1627 | * to get any new ordered operations. We can safely run | |
1628 | * it here and no for sure that nothing new will be added | |
1629 | * to the list | |
1630 | */ | |
569e0f35 | 1631 | ret = btrfs_run_ordered_operations(trans, root, 1); |
ca469637 | 1632 | |
eebc6084 | 1633 | return ret; |
ca469637 MX |
1634 | } |
1635 | ||
82436617 MX |
1636 | static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info) |
1637 | { | |
1638 | if (btrfs_test_opt(fs_info->tree_root, FLUSHONCOMMIT)) | |
6c255e67 | 1639 | return btrfs_start_delalloc_roots(fs_info, 1, -1); |
82436617 MX |
1640 | return 0; |
1641 | } | |
1642 | ||
1643 | static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info) | |
1644 | { | |
1645 | if (btrfs_test_opt(fs_info->tree_root, FLUSHONCOMMIT)) | |
b0244199 | 1646 | btrfs_wait_ordered_roots(fs_info, -1); |
82436617 MX |
1647 | } |
1648 | ||
79154b1b CM |
1649 | int btrfs_commit_transaction(struct btrfs_trans_handle *trans, |
1650 | struct btrfs_root *root) | |
1651 | { | |
49b25e05 | 1652 | struct btrfs_transaction *cur_trans = trans->transaction; |
8fd17795 | 1653 | struct btrfs_transaction *prev_trans = NULL; |
25287e0a | 1654 | int ret; |
79154b1b | 1655 | |
569e0f35 | 1656 | ret = btrfs_run_ordered_operations(trans, root, 0); |
25287e0a MX |
1657 | if (ret) { |
1658 | btrfs_abort_transaction(trans, root, ret); | |
e4a2bcac JB |
1659 | btrfs_end_transaction(trans, root); |
1660 | return ret; | |
25287e0a | 1661 | } |
5a3f23d5 | 1662 | |
8d25a086 MX |
1663 | /* Stop the commit early if ->aborted is set */ |
1664 | if (unlikely(ACCESS_ONCE(cur_trans->aborted))) { | |
25287e0a | 1665 | ret = cur_trans->aborted; |
e4a2bcac JB |
1666 | btrfs_end_transaction(trans, root); |
1667 | return ret; | |
25287e0a | 1668 | } |
49b25e05 | 1669 | |
56bec294 CM |
1670 | /* make a pass through all the delayed refs we have so far |
1671 | * any runnings procs may add more while we are here | |
1672 | */ | |
1673 | ret = btrfs_run_delayed_refs(trans, root, 0); | |
e4a2bcac JB |
1674 | if (ret) { |
1675 | btrfs_end_transaction(trans, root); | |
1676 | return ret; | |
1677 | } | |
56bec294 | 1678 | |
0e721106 JB |
1679 | btrfs_trans_release_metadata(trans, root); |
1680 | trans->block_rsv = NULL; | |
272d26d0 MX |
1681 | if (trans->qgroup_reserved) { |
1682 | btrfs_qgroup_free(root, trans->qgroup_reserved); | |
1683 | trans->qgroup_reserved = 0; | |
1684 | } | |
0e721106 | 1685 | |
b7ec40d7 | 1686 | cur_trans = trans->transaction; |
49b25e05 | 1687 | |
56bec294 CM |
1688 | /* |
1689 | * set the flushing flag so procs in this transaction have to | |
1690 | * start sending their work down. | |
1691 | */ | |
b7ec40d7 | 1692 | cur_trans->delayed_refs.flushing = 1; |
1be41b78 | 1693 | smp_wmb(); |
56bec294 | 1694 | |
ea658bad JB |
1695 | if (!list_empty(&trans->new_bgs)) |
1696 | btrfs_create_pending_block_groups(trans, root); | |
1697 | ||
c3e69d58 | 1698 | ret = btrfs_run_delayed_refs(trans, root, 0); |
e4a2bcac JB |
1699 | if (ret) { |
1700 | btrfs_end_transaction(trans, root); | |
1701 | return ret; | |
1702 | } | |
56bec294 | 1703 | |
4a9d8bde MX |
1704 | spin_lock(&root->fs_info->trans_lock); |
1705 | if (cur_trans->state >= TRANS_STATE_COMMIT_START) { | |
1706 | spin_unlock(&root->fs_info->trans_lock); | |
13c5a93e | 1707 | atomic_inc(&cur_trans->use_count); |
49b25e05 | 1708 | ret = btrfs_end_transaction(trans, root); |
ccd467d6 | 1709 | |
b9c8300c | 1710 | wait_for_commit(root, cur_trans); |
15ee9bc7 | 1711 | |
724e2315 | 1712 | btrfs_put_transaction(cur_trans); |
15ee9bc7 | 1713 | |
49b25e05 | 1714 | return ret; |
79154b1b | 1715 | } |
4313b399 | 1716 | |
4a9d8bde | 1717 | cur_trans->state = TRANS_STATE_COMMIT_START; |
bb9c12c9 SW |
1718 | wake_up(&root->fs_info->transaction_blocked_wait); |
1719 | ||
ccd467d6 CM |
1720 | if (cur_trans->list.prev != &root->fs_info->trans_list) { |
1721 | prev_trans = list_entry(cur_trans->list.prev, | |
1722 | struct btrfs_transaction, list); | |
4a9d8bde | 1723 | if (prev_trans->state != TRANS_STATE_COMPLETED) { |
13c5a93e | 1724 | atomic_inc(&prev_trans->use_count); |
a4abeea4 | 1725 | spin_unlock(&root->fs_info->trans_lock); |
ccd467d6 CM |
1726 | |
1727 | wait_for_commit(root, prev_trans); | |
ccd467d6 | 1728 | |
724e2315 | 1729 | btrfs_put_transaction(prev_trans); |
a4abeea4 JB |
1730 | } else { |
1731 | spin_unlock(&root->fs_info->trans_lock); | |
ccd467d6 | 1732 | } |
a4abeea4 JB |
1733 | } else { |
1734 | spin_unlock(&root->fs_info->trans_lock); | |
ccd467d6 | 1735 | } |
15ee9bc7 | 1736 | |
0860adfd MX |
1737 | extwriter_counter_dec(cur_trans, trans->type); |
1738 | ||
82436617 MX |
1739 | ret = btrfs_start_delalloc_flush(root->fs_info); |
1740 | if (ret) | |
1741 | goto cleanup_transaction; | |
1742 | ||
581227d0 MX |
1743 | ret = btrfs_flush_all_pending_stuffs(trans, root); |
1744 | if (ret) | |
1745 | goto cleanup_transaction; | |
15ee9bc7 | 1746 | |
581227d0 MX |
1747 | wait_event(cur_trans->writer_wait, |
1748 | extwriter_counter_read(cur_trans) == 0); | |
15ee9bc7 | 1749 | |
581227d0 | 1750 | /* some pending stuffs might be added after the previous flush. */ |
ca469637 MX |
1751 | ret = btrfs_flush_all_pending_stuffs(trans, root); |
1752 | if (ret) | |
1753 | goto cleanup_transaction; | |
1754 | ||
82436617 | 1755 | btrfs_wait_delalloc_flush(root->fs_info); |
cb7ab021 WS |
1756 | |
1757 | btrfs_scrub_pause(root); | |
ed0ca140 JB |
1758 | /* |
1759 | * Ok now we need to make sure to block out any other joins while we | |
1760 | * commit the transaction. We could have started a join before setting | |
4a9d8bde | 1761 | * COMMIT_DOING so make sure to wait for num_writers to == 1 again. |
ed0ca140 JB |
1762 | */ |
1763 | spin_lock(&root->fs_info->trans_lock); | |
4a9d8bde | 1764 | cur_trans->state = TRANS_STATE_COMMIT_DOING; |
ed0ca140 JB |
1765 | spin_unlock(&root->fs_info->trans_lock); |
1766 | wait_event(cur_trans->writer_wait, | |
1767 | atomic_read(&cur_trans->num_writers) == 1); | |
1768 | ||
2cba30f1 MX |
1769 | /* ->aborted might be set after the previous check, so check it */ |
1770 | if (unlikely(ACCESS_ONCE(cur_trans->aborted))) { | |
1771 | ret = cur_trans->aborted; | |
6cf7f77e | 1772 | goto scrub_continue; |
2cba30f1 | 1773 | } |
7585717f CM |
1774 | /* |
1775 | * the reloc mutex makes sure that we stop | |
1776 | * the balancing code from coming in and moving | |
1777 | * extents around in the middle of the commit | |
1778 | */ | |
1779 | mutex_lock(&root->fs_info->reloc_mutex); | |
1780 | ||
42874b3d MX |
1781 | /* |
1782 | * We needn't worry about the delayed items because we will | |
1783 | * deal with them in create_pending_snapshot(), which is the | |
1784 | * core function of the snapshot creation. | |
1785 | */ | |
1786 | ret = create_pending_snapshots(trans, root->fs_info); | |
49b25e05 JM |
1787 | if (ret) { |
1788 | mutex_unlock(&root->fs_info->reloc_mutex); | |
6cf7f77e | 1789 | goto scrub_continue; |
49b25e05 | 1790 | } |
3063d29f | 1791 | |
42874b3d MX |
1792 | /* |
1793 | * We insert the dir indexes of the snapshots and update the inode | |
1794 | * of the snapshots' parents after the snapshot creation, so there | |
1795 | * are some delayed items which are not dealt with. Now deal with | |
1796 | * them. | |
1797 | * | |
1798 | * We needn't worry that this operation will corrupt the snapshots, | |
1799 | * because all the tree which are snapshoted will be forced to COW | |
1800 | * the nodes and leaves. | |
1801 | */ | |
1802 | ret = btrfs_run_delayed_items(trans, root); | |
49b25e05 JM |
1803 | if (ret) { |
1804 | mutex_unlock(&root->fs_info->reloc_mutex); | |
6cf7f77e | 1805 | goto scrub_continue; |
49b25e05 | 1806 | } |
16cdcec7 | 1807 | |
56bec294 | 1808 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
49b25e05 JM |
1809 | if (ret) { |
1810 | mutex_unlock(&root->fs_info->reloc_mutex); | |
6cf7f77e | 1811 | goto scrub_continue; |
49b25e05 | 1812 | } |
56bec294 | 1813 | |
e999376f CM |
1814 | /* |
1815 | * make sure none of the code above managed to slip in a | |
1816 | * delayed item | |
1817 | */ | |
1818 | btrfs_assert_delayed_root_empty(root); | |
1819 | ||
2c90e5d6 | 1820 | WARN_ON(cur_trans != trans->transaction); |
dc17ff8f | 1821 | |
e02119d5 CM |
1822 | /* btrfs_commit_tree_roots is responsible for getting the |
1823 | * various roots consistent with each other. Every pointer | |
1824 | * in the tree of tree roots has to point to the most up to date | |
1825 | * root for every subvolume and other tree. So, we have to keep | |
1826 | * the tree logging code from jumping in and changing any | |
1827 | * of the trees. | |
1828 | * | |
1829 | * At this point in the commit, there can't be any tree-log | |
1830 | * writers, but a little lower down we drop the trans mutex | |
1831 | * and let new people in. By holding the tree_log_mutex | |
1832 | * from now until after the super is written, we avoid races | |
1833 | * with the tree-log code. | |
1834 | */ | |
1835 | mutex_lock(&root->fs_info->tree_log_mutex); | |
1836 | ||
5d4f98a2 | 1837 | ret = commit_fs_roots(trans, root); |
49b25e05 JM |
1838 | if (ret) { |
1839 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
871383be | 1840 | mutex_unlock(&root->fs_info->reloc_mutex); |
6cf7f77e | 1841 | goto scrub_continue; |
49b25e05 | 1842 | } |
54aa1f4d | 1843 | |
3818aea2 QW |
1844 | /* |
1845 | * Since the transaction is done, we should set the inode map cache flag | |
1846 | * before any other comming transaction. | |
1847 | */ | |
1848 | if (btrfs_test_opt(root, CHANGE_INODE_CACHE)) | |
1849 | btrfs_set_opt(root->fs_info->mount_opt, INODE_MAP_CACHE); | |
1850 | else | |
1851 | btrfs_clear_opt(root->fs_info->mount_opt, INODE_MAP_CACHE); | |
1852 | ||
5d4f98a2 | 1853 | /* commit_fs_roots gets rid of all the tree log roots, it is now |
e02119d5 CM |
1854 | * safe to free the root of tree log roots |
1855 | */ | |
1856 | btrfs_free_log_root_tree(trans, root->fs_info); | |
1857 | ||
5d4f98a2 | 1858 | ret = commit_cowonly_roots(trans, root); |
49b25e05 JM |
1859 | if (ret) { |
1860 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
871383be | 1861 | mutex_unlock(&root->fs_info->reloc_mutex); |
6cf7f77e | 1862 | goto scrub_continue; |
49b25e05 | 1863 | } |
54aa1f4d | 1864 | |
2cba30f1 MX |
1865 | /* |
1866 | * The tasks which save the space cache and inode cache may also | |
1867 | * update ->aborted, check it. | |
1868 | */ | |
1869 | if (unlikely(ACCESS_ONCE(cur_trans->aborted))) { | |
1870 | ret = cur_trans->aborted; | |
1871 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
1872 | mutex_unlock(&root->fs_info->reloc_mutex); | |
6cf7f77e | 1873 | goto scrub_continue; |
2cba30f1 MX |
1874 | } |
1875 | ||
11833d66 YZ |
1876 | btrfs_prepare_extent_commit(trans, root); |
1877 | ||
78fae27e | 1878 | cur_trans = root->fs_info->running_transaction; |
5d4f98a2 YZ |
1879 | |
1880 | btrfs_set_root_node(&root->fs_info->tree_root->root_item, | |
1881 | root->fs_info->tree_root->node); | |
9e351cc8 JB |
1882 | list_add_tail(&root->fs_info->tree_root->dirty_list, |
1883 | &cur_trans->switch_commits); | |
5d4f98a2 YZ |
1884 | |
1885 | btrfs_set_root_node(&root->fs_info->chunk_root->root_item, | |
1886 | root->fs_info->chunk_root->node); | |
9e351cc8 JB |
1887 | list_add_tail(&root->fs_info->chunk_root->dirty_list, |
1888 | &cur_trans->switch_commits); | |
1889 | ||
1890 | switch_commit_roots(cur_trans, root->fs_info); | |
5d4f98a2 | 1891 | |
edf39272 | 1892 | assert_qgroups_uptodate(trans); |
5d4f98a2 | 1893 | update_super_roots(root); |
e02119d5 | 1894 | |
60e7cd3a JB |
1895 | btrfs_set_super_log_root(root->fs_info->super_copy, 0); |
1896 | btrfs_set_super_log_root_level(root->fs_info->super_copy, 0); | |
6c41761f DS |
1897 | memcpy(root->fs_info->super_for_commit, root->fs_info->super_copy, |
1898 | sizeof(*root->fs_info->super_copy)); | |
ccd467d6 | 1899 | |
a4abeea4 | 1900 | spin_lock(&root->fs_info->trans_lock); |
4a9d8bde | 1901 | cur_trans->state = TRANS_STATE_UNBLOCKED; |
a4abeea4 | 1902 | root->fs_info->running_transaction = NULL; |
a4abeea4 | 1903 | spin_unlock(&root->fs_info->trans_lock); |
7585717f | 1904 | mutex_unlock(&root->fs_info->reloc_mutex); |
b7ec40d7 | 1905 | |
f9295749 | 1906 | wake_up(&root->fs_info->transaction_wait); |
e6dcd2dc | 1907 | |
79154b1b | 1908 | ret = btrfs_write_and_wait_transaction(trans, root); |
49b25e05 JM |
1909 | if (ret) { |
1910 | btrfs_error(root->fs_info, ret, | |
08748810 | 1911 | "Error while writing out transaction"); |
49b25e05 | 1912 | mutex_unlock(&root->fs_info->tree_log_mutex); |
6cf7f77e | 1913 | goto scrub_continue; |
49b25e05 JM |
1914 | } |
1915 | ||
1916 | ret = write_ctree_super(trans, root, 0); | |
1917 | if (ret) { | |
1918 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
6cf7f77e | 1919 | goto scrub_continue; |
49b25e05 | 1920 | } |
4313b399 | 1921 | |
e02119d5 CM |
1922 | /* |
1923 | * the super is written, we can safely allow the tree-loggers | |
1924 | * to go about their business | |
1925 | */ | |
1926 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
1927 | ||
11833d66 | 1928 | btrfs_finish_extent_commit(trans, root); |
4313b399 | 1929 | |
15ee9bc7 | 1930 | root->fs_info->last_trans_committed = cur_trans->transid; |
4a9d8bde MX |
1931 | /* |
1932 | * We needn't acquire the lock here because there is no other task | |
1933 | * which can change it. | |
1934 | */ | |
1935 | cur_trans->state = TRANS_STATE_COMPLETED; | |
2c90e5d6 | 1936 | wake_up(&cur_trans->commit_wait); |
3de4586c | 1937 | |
a4abeea4 | 1938 | spin_lock(&root->fs_info->trans_lock); |
13c5a93e | 1939 | list_del_init(&cur_trans->list); |
a4abeea4 JB |
1940 | spin_unlock(&root->fs_info->trans_lock); |
1941 | ||
724e2315 JB |
1942 | btrfs_put_transaction(cur_trans); |
1943 | btrfs_put_transaction(cur_trans); | |
58176a96 | 1944 | |
0860adfd | 1945 | if (trans->type & __TRANS_FREEZABLE) |
354aa0fb | 1946 | sb_end_intwrite(root->fs_info->sb); |
b2b5ef5c | 1947 | |
1abe9b8a | 1948 | trace_btrfs_transaction_commit(root); |
1949 | ||
a2de733c AJ |
1950 | btrfs_scrub_continue(root); |
1951 | ||
9ed74f2d JB |
1952 | if (current->journal_info == trans) |
1953 | current->journal_info = NULL; | |
1954 | ||
2c90e5d6 | 1955 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
24bbcf04 YZ |
1956 | |
1957 | if (current != root->fs_info->transaction_kthread) | |
1958 | btrfs_run_delayed_iputs(root); | |
1959 | ||
79154b1b | 1960 | return ret; |
49b25e05 | 1961 | |
6cf7f77e WS |
1962 | scrub_continue: |
1963 | btrfs_scrub_continue(root); | |
49b25e05 | 1964 | cleanup_transaction: |
0e721106 JB |
1965 | btrfs_trans_release_metadata(trans, root); |
1966 | trans->block_rsv = NULL; | |
272d26d0 MX |
1967 | if (trans->qgroup_reserved) { |
1968 | btrfs_qgroup_free(root, trans->qgroup_reserved); | |
1969 | trans->qgroup_reserved = 0; | |
1970 | } | |
c2cf52eb | 1971 | btrfs_warn(root->fs_info, "Skipping commit of aborted transaction."); |
49b25e05 JM |
1972 | if (current->journal_info == trans) |
1973 | current->journal_info = NULL; | |
7b8b92af | 1974 | cleanup_transaction(trans, root, ret); |
49b25e05 JM |
1975 | |
1976 | return ret; | |
79154b1b CM |
1977 | } |
1978 | ||
d352ac68 | 1979 | /* |
9d1a2a3a DS |
1980 | * return < 0 if error |
1981 | * 0 if there are no more dead_roots at the time of call | |
1982 | * 1 there are more to be processed, call me again | |
1983 | * | |
1984 | * The return value indicates there are certainly more snapshots to delete, but | |
1985 | * if there comes a new one during processing, it may return 0. We don't mind, | |
1986 | * because btrfs_commit_super will poke cleaner thread and it will process it a | |
1987 | * few seconds later. | |
d352ac68 | 1988 | */ |
9d1a2a3a | 1989 | int btrfs_clean_one_deleted_snapshot(struct btrfs_root *root) |
e9d0b13b | 1990 | { |
9d1a2a3a | 1991 | int ret; |
5d4f98a2 YZ |
1992 | struct btrfs_fs_info *fs_info = root->fs_info; |
1993 | ||
a4abeea4 | 1994 | spin_lock(&fs_info->trans_lock); |
9d1a2a3a DS |
1995 | if (list_empty(&fs_info->dead_roots)) { |
1996 | spin_unlock(&fs_info->trans_lock); | |
1997 | return 0; | |
1998 | } | |
1999 | root = list_first_entry(&fs_info->dead_roots, | |
2000 | struct btrfs_root, root_list); | |
cfad392b | 2001 | list_del_init(&root->root_list); |
a4abeea4 | 2002 | spin_unlock(&fs_info->trans_lock); |
e9d0b13b | 2003 | |
efe120a0 | 2004 | pr_debug("BTRFS: cleaner removing %llu\n", root->objectid); |
76dda93c | 2005 | |
9d1a2a3a | 2006 | btrfs_kill_all_delayed_nodes(root); |
16cdcec7 | 2007 | |
9d1a2a3a DS |
2008 | if (btrfs_header_backref_rev(root->node) < |
2009 | BTRFS_MIXED_BACKREF_REV) | |
2010 | ret = btrfs_drop_snapshot(root, NULL, 0, 0); | |
2011 | else | |
2012 | ret = btrfs_drop_snapshot(root, NULL, 1, 0); | |
2013 | /* | |
2014 | * If we encounter a transaction abort during snapshot cleaning, we | |
2015 | * don't want to crash here | |
2016 | */ | |
6596a928 | 2017 | return (ret < 0) ? 0 : 1; |
e9d0b13b | 2018 | } |