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 LB |
65 | WARN_ON(!RB_EMPTY_ROOT(&transaction->delayed_refs.root)); |
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 | ||
817d52f8 JB |
79 | static noinline void switch_commit_root(struct btrfs_root *root) |
80 | { | |
817d52f8 JB |
81 | free_extent_buffer(root->commit_root); |
82 | root->commit_root = btrfs_root_node(root); | |
817d52f8 JB |
83 | } |
84 | ||
0860adfd MX |
85 | static inline void extwriter_counter_inc(struct btrfs_transaction *trans, |
86 | unsigned int type) | |
87 | { | |
88 | if (type & TRANS_EXTWRITERS) | |
89 | atomic_inc(&trans->num_extwriters); | |
90 | } | |
91 | ||
92 | static inline void extwriter_counter_dec(struct btrfs_transaction *trans, | |
93 | unsigned int type) | |
94 | { | |
95 | if (type & TRANS_EXTWRITERS) | |
96 | atomic_dec(&trans->num_extwriters); | |
97 | } | |
98 | ||
99 | static inline void extwriter_counter_init(struct btrfs_transaction *trans, | |
100 | unsigned int type) | |
101 | { | |
102 | atomic_set(&trans->num_extwriters, ((type & TRANS_EXTWRITERS) ? 1 : 0)); | |
103 | } | |
104 | ||
105 | static inline int extwriter_counter_read(struct btrfs_transaction *trans) | |
106 | { | |
107 | return atomic_read(&trans->num_extwriters); | |
178260b2 MX |
108 | } |
109 | ||
d352ac68 CM |
110 | /* |
111 | * either allocate a new transaction or hop into the existing one | |
112 | */ | |
0860adfd | 113 | static noinline int join_transaction(struct btrfs_root *root, unsigned int type) |
79154b1b CM |
114 | { |
115 | struct btrfs_transaction *cur_trans; | |
19ae4e81 | 116 | struct btrfs_fs_info *fs_info = root->fs_info; |
a4abeea4 | 117 | |
19ae4e81 | 118 | spin_lock(&fs_info->trans_lock); |
d43317dc | 119 | loop: |
49b25e05 | 120 | /* The file system has been taken offline. No new transactions. */ |
87533c47 | 121 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { |
19ae4e81 | 122 | spin_unlock(&fs_info->trans_lock); |
49b25e05 JM |
123 | return -EROFS; |
124 | } | |
125 | ||
19ae4e81 | 126 | cur_trans = fs_info->running_transaction; |
a4abeea4 | 127 | if (cur_trans) { |
871383be | 128 | if (cur_trans->aborted) { |
19ae4e81 | 129 | spin_unlock(&fs_info->trans_lock); |
49b25e05 | 130 | return cur_trans->aborted; |
871383be | 131 | } |
4a9d8bde | 132 | if (btrfs_blocked_trans_types[cur_trans->state] & type) { |
178260b2 MX |
133 | spin_unlock(&fs_info->trans_lock); |
134 | return -EBUSY; | |
135 | } | |
a4abeea4 | 136 | atomic_inc(&cur_trans->use_count); |
13c5a93e | 137 | atomic_inc(&cur_trans->num_writers); |
0860adfd | 138 | extwriter_counter_inc(cur_trans, type); |
19ae4e81 | 139 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 | 140 | return 0; |
79154b1b | 141 | } |
19ae4e81 | 142 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 | 143 | |
354aa0fb MX |
144 | /* |
145 | * If we are ATTACH, we just want to catch the current transaction, | |
146 | * and commit it. If there is no transaction, just return ENOENT. | |
147 | */ | |
148 | if (type == TRANS_ATTACH) | |
149 | return -ENOENT; | |
150 | ||
4a9d8bde MX |
151 | /* |
152 | * JOIN_NOLOCK only happens during the transaction commit, so | |
153 | * it is impossible that ->running_transaction is NULL | |
154 | */ | |
155 | BUG_ON(type == TRANS_JOIN_NOLOCK); | |
156 | ||
a4abeea4 JB |
157 | cur_trans = kmem_cache_alloc(btrfs_transaction_cachep, GFP_NOFS); |
158 | if (!cur_trans) | |
159 | return -ENOMEM; | |
d43317dc | 160 | |
19ae4e81 JS |
161 | spin_lock(&fs_info->trans_lock); |
162 | if (fs_info->running_transaction) { | |
d43317dc CM |
163 | /* |
164 | * someone started a transaction after we unlocked. Make sure | |
4a9d8bde | 165 | * to redo the checks above |
d43317dc | 166 | */ |
a4abeea4 | 167 | kmem_cache_free(btrfs_transaction_cachep, cur_trans); |
d43317dc | 168 | goto loop; |
87533c47 | 169 | } else if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { |
e4b50e14 | 170 | spin_unlock(&fs_info->trans_lock); |
7b8b92af JB |
171 | kmem_cache_free(btrfs_transaction_cachep, cur_trans); |
172 | return -EROFS; | |
79154b1b | 173 | } |
d43317dc | 174 | |
a4abeea4 | 175 | atomic_set(&cur_trans->num_writers, 1); |
0860adfd | 176 | extwriter_counter_init(cur_trans, type); |
a4abeea4 JB |
177 | init_waitqueue_head(&cur_trans->writer_wait); |
178 | init_waitqueue_head(&cur_trans->commit_wait); | |
4a9d8bde | 179 | cur_trans->state = TRANS_STATE_RUNNING; |
a4abeea4 JB |
180 | /* |
181 | * One for this trans handle, one so it will live on until we | |
182 | * commit the transaction. | |
183 | */ | |
184 | atomic_set(&cur_trans->use_count, 2); | |
a4abeea4 JB |
185 | cur_trans->start_time = get_seconds(); |
186 | ||
187 | cur_trans->delayed_refs.root = RB_ROOT; | |
c46effa6 | 188 | cur_trans->delayed_refs.href_root = RB_ROOT; |
a4abeea4 JB |
189 | cur_trans->delayed_refs.num_entries = 0; |
190 | cur_trans->delayed_refs.num_heads_ready = 0; | |
191 | cur_trans->delayed_refs.num_heads = 0; | |
192 | cur_trans->delayed_refs.flushing = 0; | |
193 | cur_trans->delayed_refs.run_delayed_start = 0; | |
20b297d6 JS |
194 | |
195 | /* | |
196 | * although the tree mod log is per file system and not per transaction, | |
197 | * the log must never go across transaction boundaries. | |
198 | */ | |
199 | smp_mb(); | |
31b1a2bd | 200 | if (!list_empty(&fs_info->tree_mod_seq_list)) |
efe120a0 | 201 | WARN(1, KERN_ERR "BTRFS: tree_mod_seq_list not empty when " |
20b297d6 | 202 | "creating a fresh transaction\n"); |
31b1a2bd | 203 | if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log)) |
efe120a0 | 204 | WARN(1, KERN_ERR "BTRFS: tree_mod_log rb tree not empty when " |
20b297d6 | 205 | "creating a fresh transaction\n"); |
fc36ed7e | 206 | atomic64_set(&fs_info->tree_mod_seq, 0); |
20b297d6 | 207 | |
a4abeea4 | 208 | spin_lock_init(&cur_trans->delayed_refs.lock); |
bb721703 CM |
209 | atomic_set(&cur_trans->delayed_refs.procs_running_refs, 0); |
210 | atomic_set(&cur_trans->delayed_refs.ref_seq, 0); | |
211 | init_waitqueue_head(&cur_trans->delayed_refs.wait); | |
a4abeea4 JB |
212 | |
213 | INIT_LIST_HEAD(&cur_trans->pending_snapshots); | |
569e0f35 | 214 | INIT_LIST_HEAD(&cur_trans->ordered_operations); |
6df9a95e | 215 | INIT_LIST_HEAD(&cur_trans->pending_chunks); |
19ae4e81 | 216 | list_add_tail(&cur_trans->list, &fs_info->trans_list); |
a4abeea4 | 217 | extent_io_tree_init(&cur_trans->dirty_pages, |
19ae4e81 JS |
218 | fs_info->btree_inode->i_mapping); |
219 | fs_info->generation++; | |
220 | cur_trans->transid = fs_info->generation; | |
221 | fs_info->running_transaction = cur_trans; | |
49b25e05 | 222 | cur_trans->aborted = 0; |
19ae4e81 | 223 | spin_unlock(&fs_info->trans_lock); |
15ee9bc7 | 224 | |
79154b1b CM |
225 | return 0; |
226 | } | |
227 | ||
d352ac68 | 228 | /* |
d397712b CM |
229 | * this does all the record keeping required to make sure that a reference |
230 | * counted root is properly recorded in a given transaction. This is required | |
231 | * to make sure the old root from before we joined the transaction is deleted | |
232 | * when the transaction commits | |
d352ac68 | 233 | */ |
7585717f | 234 | static int record_root_in_trans(struct btrfs_trans_handle *trans, |
a4abeea4 | 235 | struct btrfs_root *root) |
6702ed49 | 236 | { |
5d4f98a2 | 237 | if (root->ref_cows && root->last_trans < trans->transid) { |
6702ed49 | 238 | WARN_ON(root == root->fs_info->extent_root); |
5d4f98a2 YZ |
239 | WARN_ON(root->commit_root != root->node); |
240 | ||
7585717f CM |
241 | /* |
242 | * see below for in_trans_setup usage rules | |
243 | * we have the reloc mutex held now, so there | |
244 | * is only one writer in this function | |
245 | */ | |
246 | root->in_trans_setup = 1; | |
247 | ||
248 | /* make sure readers find in_trans_setup before | |
249 | * they find our root->last_trans update | |
250 | */ | |
251 | smp_wmb(); | |
252 | ||
a4abeea4 JB |
253 | spin_lock(&root->fs_info->fs_roots_radix_lock); |
254 | if (root->last_trans == trans->transid) { | |
255 | spin_unlock(&root->fs_info->fs_roots_radix_lock); | |
256 | return 0; | |
257 | } | |
5d4f98a2 YZ |
258 | radix_tree_tag_set(&root->fs_info->fs_roots_radix, |
259 | (unsigned long)root->root_key.objectid, | |
260 | BTRFS_ROOT_TRANS_TAG); | |
a4abeea4 | 261 | spin_unlock(&root->fs_info->fs_roots_radix_lock); |
7585717f CM |
262 | root->last_trans = trans->transid; |
263 | ||
264 | /* this is pretty tricky. We don't want to | |
265 | * take the relocation lock in btrfs_record_root_in_trans | |
266 | * unless we're really doing the first setup for this root in | |
267 | * this transaction. | |
268 | * | |
269 | * Normally we'd use root->last_trans as a flag to decide | |
270 | * if we want to take the expensive mutex. | |
271 | * | |
272 | * But, we have to set root->last_trans before we | |
273 | * init the relocation root, otherwise, we trip over warnings | |
274 | * in ctree.c. The solution used here is to flag ourselves | |
275 | * with root->in_trans_setup. When this is 1, we're still | |
276 | * fixing up the reloc trees and everyone must wait. | |
277 | * | |
278 | * When this is zero, they can trust root->last_trans and fly | |
279 | * through btrfs_record_root_in_trans without having to take the | |
280 | * lock. smp_wmb() makes sure that all the writes above are | |
281 | * done before we pop in the zero below | |
282 | */ | |
5d4f98a2 | 283 | btrfs_init_reloc_root(trans, root); |
7585717f CM |
284 | smp_wmb(); |
285 | root->in_trans_setup = 0; | |
5d4f98a2 YZ |
286 | } |
287 | return 0; | |
288 | } | |
bcc63abb | 289 | |
7585717f CM |
290 | |
291 | int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, | |
292 | struct btrfs_root *root) | |
293 | { | |
294 | if (!root->ref_cows) | |
295 | return 0; | |
296 | ||
297 | /* | |
298 | * see record_root_in_trans for comments about in_trans_setup usage | |
299 | * and barriers | |
300 | */ | |
301 | smp_rmb(); | |
302 | if (root->last_trans == trans->transid && | |
303 | !root->in_trans_setup) | |
304 | return 0; | |
305 | ||
306 | mutex_lock(&root->fs_info->reloc_mutex); | |
307 | record_root_in_trans(trans, root); | |
308 | mutex_unlock(&root->fs_info->reloc_mutex); | |
309 | ||
310 | return 0; | |
311 | } | |
312 | ||
4a9d8bde MX |
313 | static inline int is_transaction_blocked(struct btrfs_transaction *trans) |
314 | { | |
315 | return (trans->state >= TRANS_STATE_BLOCKED && | |
501407aa JB |
316 | trans->state < TRANS_STATE_UNBLOCKED && |
317 | !trans->aborted); | |
4a9d8bde MX |
318 | } |
319 | ||
d352ac68 CM |
320 | /* wait for commit against the current transaction to become unblocked |
321 | * when this is done, it is safe to start a new transaction, but the current | |
322 | * transaction might not be fully on disk. | |
323 | */ | |
37d1aeee | 324 | static void wait_current_trans(struct btrfs_root *root) |
79154b1b | 325 | { |
f9295749 | 326 | struct btrfs_transaction *cur_trans; |
79154b1b | 327 | |
a4abeea4 | 328 | spin_lock(&root->fs_info->trans_lock); |
f9295749 | 329 | cur_trans = root->fs_info->running_transaction; |
4a9d8bde | 330 | if (cur_trans && is_transaction_blocked(cur_trans)) { |
13c5a93e | 331 | atomic_inc(&cur_trans->use_count); |
a4abeea4 | 332 | spin_unlock(&root->fs_info->trans_lock); |
72d63ed6 LZ |
333 | |
334 | wait_event(root->fs_info->transaction_wait, | |
501407aa JB |
335 | cur_trans->state >= TRANS_STATE_UNBLOCKED || |
336 | cur_trans->aborted); | |
724e2315 | 337 | btrfs_put_transaction(cur_trans); |
a4abeea4 JB |
338 | } else { |
339 | spin_unlock(&root->fs_info->trans_lock); | |
f9295749 | 340 | } |
37d1aeee CM |
341 | } |
342 | ||
a22285a6 YZ |
343 | static int may_wait_transaction(struct btrfs_root *root, int type) |
344 | { | |
a4abeea4 JB |
345 | if (root->fs_info->log_root_recovering) |
346 | return 0; | |
347 | ||
348 | if (type == TRANS_USERSPACE) | |
349 | return 1; | |
350 | ||
351 | if (type == TRANS_START && | |
352 | !atomic_read(&root->fs_info->open_ioctl_trans)) | |
a22285a6 | 353 | return 1; |
a4abeea4 | 354 | |
a22285a6 YZ |
355 | return 0; |
356 | } | |
357 | ||
20dd2cbf MX |
358 | static inline bool need_reserve_reloc_root(struct btrfs_root *root) |
359 | { | |
360 | if (!root->fs_info->reloc_ctl || | |
361 | !root->ref_cows || | |
362 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || | |
363 | root->reloc_root) | |
364 | return false; | |
365 | ||
366 | return true; | |
367 | } | |
368 | ||
08e007d2 | 369 | static struct btrfs_trans_handle * |
0860adfd | 370 | start_transaction(struct btrfs_root *root, u64 num_items, unsigned int type, |
08e007d2 | 371 | enum btrfs_reserve_flush_enum flush) |
37d1aeee | 372 | { |
a22285a6 YZ |
373 | struct btrfs_trans_handle *h; |
374 | struct btrfs_transaction *cur_trans; | |
b5009945 | 375 | u64 num_bytes = 0; |
c5567237 | 376 | u64 qgroup_reserved = 0; |
20dd2cbf MX |
377 | bool reloc_reserved = false; |
378 | int ret; | |
acce952b | 379 | |
87533c47 | 380 | if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) |
acce952b | 381 | return ERR_PTR(-EROFS); |
2a1eb461 JB |
382 | |
383 | if (current->journal_info) { | |
0860adfd | 384 | WARN_ON(type & TRANS_EXTWRITERS); |
2a1eb461 JB |
385 | h = current->journal_info; |
386 | h->use_count++; | |
b7d5b0a8 | 387 | WARN_ON(h->use_count > 2); |
2a1eb461 JB |
388 | h->orig_rsv = h->block_rsv; |
389 | h->block_rsv = NULL; | |
390 | goto got_it; | |
391 | } | |
b5009945 JB |
392 | |
393 | /* | |
394 | * Do the reservation before we join the transaction so we can do all | |
395 | * the appropriate flushing if need be. | |
396 | */ | |
397 | if (num_items > 0 && root != root->fs_info->chunk_root) { | |
c5567237 AJ |
398 | if (root->fs_info->quota_enabled && |
399 | is_fstree(root->root_key.objectid)) { | |
400 | qgroup_reserved = num_items * root->leafsize; | |
401 | ret = btrfs_qgroup_reserve(root, qgroup_reserved); | |
402 | if (ret) | |
403 | return ERR_PTR(ret); | |
404 | } | |
405 | ||
b5009945 | 406 | num_bytes = btrfs_calc_trans_metadata_size(root, num_items); |
20dd2cbf MX |
407 | /* |
408 | * Do the reservation for the relocation root creation | |
409 | */ | |
410 | if (unlikely(need_reserve_reloc_root(root))) { | |
411 | num_bytes += root->nodesize; | |
412 | reloc_reserved = true; | |
413 | } | |
414 | ||
08e007d2 MX |
415 | ret = btrfs_block_rsv_add(root, |
416 | &root->fs_info->trans_block_rsv, | |
417 | num_bytes, flush); | |
b5009945 | 418 | if (ret) |
843fcf35 | 419 | goto reserve_fail; |
b5009945 | 420 | } |
a22285a6 YZ |
421 | again: |
422 | h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS); | |
843fcf35 MX |
423 | if (!h) { |
424 | ret = -ENOMEM; | |
425 | goto alloc_fail; | |
426 | } | |
37d1aeee | 427 | |
98114659 JB |
428 | /* |
429 | * If we are JOIN_NOLOCK we're already committing a transaction and | |
430 | * waiting on this guy, so we don't need to do the sb_start_intwrite | |
431 | * because we're already holding a ref. We need this because we could | |
432 | * have raced in and did an fsync() on a file which can kick a commit | |
433 | * and then we deadlock with somebody doing a freeze. | |
354aa0fb MX |
434 | * |
435 | * If we are ATTACH, it means we just want to catch the current | |
436 | * transaction and commit it, so we needn't do sb_start_intwrite(). | |
98114659 | 437 | */ |
0860adfd | 438 | if (type & __TRANS_FREEZABLE) |
60376ce4 | 439 | sb_start_intwrite(root->fs_info->sb); |
b2b5ef5c | 440 | |
a22285a6 | 441 | if (may_wait_transaction(root, type)) |
37d1aeee | 442 | wait_current_trans(root); |
a22285a6 | 443 | |
a4abeea4 | 444 | do { |
354aa0fb | 445 | ret = join_transaction(root, type); |
178260b2 | 446 | if (ret == -EBUSY) { |
a4abeea4 | 447 | wait_current_trans(root); |
178260b2 MX |
448 | if (unlikely(type == TRANS_ATTACH)) |
449 | ret = -ENOENT; | |
450 | } | |
a4abeea4 JB |
451 | } while (ret == -EBUSY); |
452 | ||
db5b493a | 453 | if (ret < 0) { |
354aa0fb MX |
454 | /* We must get the transaction if we are JOIN_NOLOCK. */ |
455 | BUG_ON(type == TRANS_JOIN_NOLOCK); | |
843fcf35 | 456 | goto join_fail; |
db5b493a | 457 | } |
0f7d52f4 | 458 | |
a22285a6 | 459 | cur_trans = root->fs_info->running_transaction; |
a22285a6 YZ |
460 | |
461 | h->transid = cur_trans->transid; | |
462 | h->transaction = cur_trans; | |
79154b1b | 463 | h->blocks_used = 0; |
a22285a6 | 464 | h->bytes_reserved = 0; |
d13603ef | 465 | h->root = root; |
56bec294 | 466 | h->delayed_ref_updates = 0; |
2a1eb461 | 467 | h->use_count = 1; |
0e721106 | 468 | h->adding_csums = 0; |
f0486c68 | 469 | h->block_rsv = NULL; |
2a1eb461 | 470 | h->orig_rsv = NULL; |
49b25e05 | 471 | h->aborted = 0; |
4b824906 | 472 | h->qgroup_reserved = 0; |
bed92eae | 473 | h->delayed_ref_elem.seq = 0; |
a698d075 | 474 | h->type = type; |
c6b305a8 | 475 | h->allocating_chunk = false; |
20dd2cbf | 476 | h->reloc_reserved = false; |
5039eddc | 477 | h->sync = false; |
bed92eae | 478 | INIT_LIST_HEAD(&h->qgroup_ref_list); |
ea658bad | 479 | INIT_LIST_HEAD(&h->new_bgs); |
b7ec40d7 | 480 | |
a22285a6 | 481 | smp_mb(); |
4a9d8bde MX |
482 | if (cur_trans->state >= TRANS_STATE_BLOCKED && |
483 | may_wait_transaction(root, type)) { | |
a22285a6 YZ |
484 | btrfs_commit_transaction(h, root); |
485 | goto again; | |
486 | } | |
487 | ||
b5009945 | 488 | if (num_bytes) { |
8c2a3ca2 | 489 | trace_btrfs_space_reservation(root->fs_info, "transaction", |
2bcc0328 | 490 | h->transid, num_bytes, 1); |
b5009945 JB |
491 | h->block_rsv = &root->fs_info->trans_block_rsv; |
492 | h->bytes_reserved = num_bytes; | |
20dd2cbf | 493 | h->reloc_reserved = reloc_reserved; |
a22285a6 | 494 | } |
4b824906 | 495 | h->qgroup_reserved = qgroup_reserved; |
9ed74f2d | 496 | |
2a1eb461 | 497 | got_it: |
a4abeea4 | 498 | btrfs_record_root_in_trans(h, root); |
a22285a6 YZ |
499 | |
500 | if (!current->journal_info && type != TRANS_USERSPACE) | |
501 | current->journal_info = h; | |
79154b1b | 502 | return h; |
843fcf35 MX |
503 | |
504 | join_fail: | |
0860adfd | 505 | if (type & __TRANS_FREEZABLE) |
843fcf35 MX |
506 | sb_end_intwrite(root->fs_info->sb); |
507 | kmem_cache_free(btrfs_trans_handle_cachep, h); | |
508 | alloc_fail: | |
509 | if (num_bytes) | |
510 | btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv, | |
511 | num_bytes); | |
512 | reserve_fail: | |
513 | if (qgroup_reserved) | |
514 | btrfs_qgroup_free(root, qgroup_reserved); | |
515 | return ERR_PTR(ret); | |
79154b1b CM |
516 | } |
517 | ||
f9295749 | 518 | struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, |
a22285a6 | 519 | int num_items) |
f9295749 | 520 | { |
08e007d2 MX |
521 | return start_transaction(root, num_items, TRANS_START, |
522 | BTRFS_RESERVE_FLUSH_ALL); | |
f9295749 | 523 | } |
8407aa46 | 524 | |
08e007d2 | 525 | struct btrfs_trans_handle *btrfs_start_transaction_lflush( |
8407aa46 MX |
526 | struct btrfs_root *root, int num_items) |
527 | { | |
08e007d2 MX |
528 | return start_transaction(root, num_items, TRANS_START, |
529 | BTRFS_RESERVE_FLUSH_LIMIT); | |
8407aa46 MX |
530 | } |
531 | ||
7a7eaa40 | 532 | struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root) |
f9295749 | 533 | { |
8407aa46 | 534 | return start_transaction(root, 0, TRANS_JOIN, 0); |
f9295749 CM |
535 | } |
536 | ||
7a7eaa40 | 537 | struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root) |
0af3d00b | 538 | { |
8407aa46 | 539 | return start_transaction(root, 0, TRANS_JOIN_NOLOCK, 0); |
0af3d00b JB |
540 | } |
541 | ||
7a7eaa40 | 542 | struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root) |
9ca9ee09 | 543 | { |
8407aa46 | 544 | return start_transaction(root, 0, TRANS_USERSPACE, 0); |
9ca9ee09 SW |
545 | } |
546 | ||
d4edf39b MX |
547 | /* |
548 | * btrfs_attach_transaction() - catch the running transaction | |
549 | * | |
550 | * It is used when we want to commit the current the transaction, but | |
551 | * don't want to start a new one. | |
552 | * | |
553 | * Note: If this function return -ENOENT, it just means there is no | |
554 | * running transaction. But it is possible that the inactive transaction | |
555 | * is still in the memory, not fully on disk. If you hope there is no | |
556 | * inactive transaction in the fs when -ENOENT is returned, you should | |
557 | * invoke | |
558 | * btrfs_attach_transaction_barrier() | |
559 | */ | |
354aa0fb | 560 | struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root) |
60376ce4 | 561 | { |
354aa0fb | 562 | return start_transaction(root, 0, TRANS_ATTACH, 0); |
60376ce4 JB |
563 | } |
564 | ||
d4edf39b | 565 | /* |
90b6d283 | 566 | * btrfs_attach_transaction_barrier() - catch the running transaction |
d4edf39b MX |
567 | * |
568 | * It is similar to the above function, the differentia is this one | |
569 | * will wait for all the inactive transactions until they fully | |
570 | * complete. | |
571 | */ | |
572 | struct btrfs_trans_handle * | |
573 | btrfs_attach_transaction_barrier(struct btrfs_root *root) | |
574 | { | |
575 | struct btrfs_trans_handle *trans; | |
576 | ||
577 | trans = start_transaction(root, 0, TRANS_ATTACH, 0); | |
578 | if (IS_ERR(trans) && PTR_ERR(trans) == -ENOENT) | |
579 | btrfs_wait_for_commit(root, 0); | |
580 | ||
581 | return trans; | |
582 | } | |
583 | ||
d352ac68 | 584 | /* wait for a transaction commit to be fully complete */ |
b9c8300c | 585 | static noinline void wait_for_commit(struct btrfs_root *root, |
89ce8a63 CM |
586 | struct btrfs_transaction *commit) |
587 | { | |
4a9d8bde | 588 | wait_event(commit->commit_wait, commit->state == TRANS_STATE_COMPLETED); |
89ce8a63 CM |
589 | } |
590 | ||
46204592 SW |
591 | int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid) |
592 | { | |
593 | struct btrfs_transaction *cur_trans = NULL, *t; | |
8cd2807f | 594 | int ret = 0; |
46204592 | 595 | |
46204592 SW |
596 | if (transid) { |
597 | if (transid <= root->fs_info->last_trans_committed) | |
a4abeea4 | 598 | goto out; |
46204592 | 599 | |
8cd2807f | 600 | ret = -EINVAL; |
46204592 | 601 | /* find specified transaction */ |
a4abeea4 | 602 | spin_lock(&root->fs_info->trans_lock); |
46204592 SW |
603 | list_for_each_entry(t, &root->fs_info->trans_list, list) { |
604 | if (t->transid == transid) { | |
605 | cur_trans = t; | |
a4abeea4 | 606 | atomic_inc(&cur_trans->use_count); |
8cd2807f | 607 | ret = 0; |
46204592 SW |
608 | break; |
609 | } | |
8cd2807f MX |
610 | if (t->transid > transid) { |
611 | ret = 0; | |
46204592 | 612 | break; |
8cd2807f | 613 | } |
46204592 | 614 | } |
a4abeea4 | 615 | spin_unlock(&root->fs_info->trans_lock); |
8cd2807f | 616 | /* The specified transaction doesn't exist */ |
46204592 | 617 | if (!cur_trans) |
8cd2807f | 618 | goto out; |
46204592 SW |
619 | } else { |
620 | /* find newest transaction that is committing | committed */ | |
a4abeea4 | 621 | spin_lock(&root->fs_info->trans_lock); |
46204592 SW |
622 | list_for_each_entry_reverse(t, &root->fs_info->trans_list, |
623 | list) { | |
4a9d8bde MX |
624 | if (t->state >= TRANS_STATE_COMMIT_START) { |
625 | if (t->state == TRANS_STATE_COMPLETED) | |
3473f3c0 | 626 | break; |
46204592 | 627 | cur_trans = t; |
a4abeea4 | 628 | atomic_inc(&cur_trans->use_count); |
46204592 SW |
629 | break; |
630 | } | |
631 | } | |
a4abeea4 | 632 | spin_unlock(&root->fs_info->trans_lock); |
46204592 | 633 | if (!cur_trans) |
a4abeea4 | 634 | goto out; /* nothing committing|committed */ |
46204592 SW |
635 | } |
636 | ||
46204592 | 637 | wait_for_commit(root, cur_trans); |
724e2315 | 638 | btrfs_put_transaction(cur_trans); |
a4abeea4 | 639 | out: |
46204592 SW |
640 | return ret; |
641 | } | |
642 | ||
37d1aeee CM |
643 | void btrfs_throttle(struct btrfs_root *root) |
644 | { | |
a4abeea4 | 645 | if (!atomic_read(&root->fs_info->open_ioctl_trans)) |
9ca9ee09 | 646 | wait_current_trans(root); |
37d1aeee CM |
647 | } |
648 | ||
8929ecfa YZ |
649 | static int should_end_transaction(struct btrfs_trans_handle *trans, |
650 | struct btrfs_root *root) | |
651 | { | |
1be41b78 JB |
652 | if (root->fs_info->global_block_rsv.space_info->full && |
653 | btrfs_should_throttle_delayed_refs(trans, root)) | |
654 | return 1; | |
36ba022a | 655 | |
1be41b78 | 656 | return !!btrfs_block_rsv_check(root, &root->fs_info->global_block_rsv, 5); |
8929ecfa YZ |
657 | } |
658 | ||
659 | int btrfs_should_end_transaction(struct btrfs_trans_handle *trans, | |
660 | struct btrfs_root *root) | |
661 | { | |
662 | struct btrfs_transaction *cur_trans = trans->transaction; | |
663 | int updates; | |
49b25e05 | 664 | int err; |
8929ecfa | 665 | |
a4abeea4 | 666 | smp_mb(); |
4a9d8bde MX |
667 | if (cur_trans->state >= TRANS_STATE_BLOCKED || |
668 | cur_trans->delayed_refs.flushing) | |
8929ecfa YZ |
669 | return 1; |
670 | ||
671 | updates = trans->delayed_ref_updates; | |
672 | trans->delayed_ref_updates = 0; | |
49b25e05 JM |
673 | if (updates) { |
674 | err = btrfs_run_delayed_refs(trans, root, updates); | |
675 | if (err) /* Error code will also eval true */ | |
676 | return err; | |
677 | } | |
8929ecfa YZ |
678 | |
679 | return should_end_transaction(trans, root); | |
680 | } | |
681 | ||
89ce8a63 | 682 | static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, |
a698d075 | 683 | struct btrfs_root *root, int throttle) |
79154b1b | 684 | { |
8929ecfa | 685 | struct btrfs_transaction *cur_trans = trans->transaction; |
ab78c84d | 686 | struct btrfs_fs_info *info = root->fs_info; |
1be41b78 | 687 | unsigned long cur = trans->delayed_ref_updates; |
a698d075 | 688 | int lock = (trans->type != TRANS_JOIN_NOLOCK); |
4edc2ca3 | 689 | int err = 0; |
c3e69d58 | 690 | |
2a1eb461 JB |
691 | if (--trans->use_count) { |
692 | trans->block_rsv = trans->orig_rsv; | |
693 | return 0; | |
694 | } | |
695 | ||
edf39272 JS |
696 | /* |
697 | * do the qgroup accounting as early as possible | |
698 | */ | |
699 | err = btrfs_delayed_refs_qgroup_accounting(trans, info); | |
700 | ||
b24e03db | 701 | btrfs_trans_release_metadata(trans, root); |
4c13d758 | 702 | trans->block_rsv = NULL; |
c5567237 AJ |
703 | |
704 | if (trans->qgroup_reserved) { | |
7c2ec3f0 LB |
705 | /* |
706 | * the same root has to be passed here between start_transaction | |
707 | * and end_transaction. Subvolume quota depends on this. | |
708 | */ | |
709 | btrfs_qgroup_free(trans->root, trans->qgroup_reserved); | |
c5567237 AJ |
710 | trans->qgroup_reserved = 0; |
711 | } | |
712 | ||
ea658bad JB |
713 | if (!list_empty(&trans->new_bgs)) |
714 | btrfs_create_pending_block_groups(trans, root); | |
715 | ||
1be41b78 | 716 | trans->delayed_ref_updates = 0; |
5039eddc | 717 | if (!trans->sync && btrfs_should_throttle_delayed_refs(trans, root)) { |
1be41b78 | 718 | cur = max_t(unsigned long, cur, 1); |
c3e69d58 | 719 | trans->delayed_ref_updates = 0; |
1be41b78 | 720 | btrfs_run_delayed_refs(trans, root, cur); |
56bec294 | 721 | } |
bb721703 | 722 | |
0e721106 JB |
723 | btrfs_trans_release_metadata(trans, root); |
724 | trans->block_rsv = NULL; | |
56bec294 | 725 | |
ea658bad JB |
726 | if (!list_empty(&trans->new_bgs)) |
727 | btrfs_create_pending_block_groups(trans, root); | |
728 | ||
a4abeea4 | 729 | if (lock && !atomic_read(&root->fs_info->open_ioctl_trans) && |
4a9d8bde MX |
730 | should_end_transaction(trans, root) && |
731 | ACCESS_ONCE(cur_trans->state) == TRANS_STATE_RUNNING) { | |
732 | spin_lock(&info->trans_lock); | |
733 | if (cur_trans->state == TRANS_STATE_RUNNING) | |
734 | cur_trans->state = TRANS_STATE_BLOCKED; | |
735 | spin_unlock(&info->trans_lock); | |
a4abeea4 | 736 | } |
8929ecfa | 737 | |
4a9d8bde | 738 | if (lock && ACCESS_ONCE(cur_trans->state) == TRANS_STATE_BLOCKED) { |
81317fde JB |
739 | if (throttle) { |
740 | /* | |
741 | * We may race with somebody else here so end up having | |
742 | * to call end_transaction on ourselves again, so inc | |
743 | * our use_count. | |
744 | */ | |
745 | trans->use_count++; | |
8929ecfa | 746 | return btrfs_commit_transaction(trans, root); |
81317fde | 747 | } else { |
8929ecfa | 748 | wake_up_process(info->transaction_kthread); |
81317fde | 749 | } |
8929ecfa YZ |
750 | } |
751 | ||
0860adfd | 752 | if (trans->type & __TRANS_FREEZABLE) |
98114659 | 753 | sb_end_intwrite(root->fs_info->sb); |
6df7881a | 754 | |
8929ecfa | 755 | WARN_ON(cur_trans != info->running_transaction); |
13c5a93e JB |
756 | WARN_ON(atomic_read(&cur_trans->num_writers) < 1); |
757 | atomic_dec(&cur_trans->num_writers); | |
0860adfd | 758 | extwriter_counter_dec(cur_trans, trans->type); |
89ce8a63 | 759 | |
99d16cbc | 760 | smp_mb(); |
79154b1b CM |
761 | if (waitqueue_active(&cur_trans->writer_wait)) |
762 | wake_up(&cur_trans->writer_wait); | |
724e2315 | 763 | btrfs_put_transaction(cur_trans); |
9ed74f2d JB |
764 | |
765 | if (current->journal_info == trans) | |
766 | current->journal_info = NULL; | |
ab78c84d | 767 | |
24bbcf04 YZ |
768 | if (throttle) |
769 | btrfs_run_delayed_iputs(root); | |
770 | ||
49b25e05 | 771 | if (trans->aborted || |
4e121c06 JB |
772 | test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) { |
773 | wake_up_process(info->transaction_kthread); | |
4edc2ca3 | 774 | err = -EIO; |
4e121c06 | 775 | } |
edf39272 | 776 | assert_qgroups_uptodate(trans); |
49b25e05 | 777 | |
4edc2ca3 DJ |
778 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
779 | return err; | |
79154b1b CM |
780 | } |
781 | ||
89ce8a63 CM |
782 | int btrfs_end_transaction(struct btrfs_trans_handle *trans, |
783 | struct btrfs_root *root) | |
784 | { | |
98ad43be | 785 | return __btrfs_end_transaction(trans, root, 0); |
89ce8a63 CM |
786 | } |
787 | ||
788 | int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans, | |
789 | struct btrfs_root *root) | |
790 | { | |
98ad43be | 791 | return __btrfs_end_transaction(trans, root, 1); |
16cdcec7 MX |
792 | } |
793 | ||
d352ac68 CM |
794 | /* |
795 | * when btree blocks are allocated, they have some corresponding bits set for | |
796 | * them in one of two extent_io trees. This is used to make sure all of | |
690587d1 | 797 | * those extents are sent to disk but does not wait on them |
d352ac68 | 798 | */ |
690587d1 | 799 | int btrfs_write_marked_extents(struct btrfs_root *root, |
8cef4e16 | 800 | struct extent_io_tree *dirty_pages, int mark) |
79154b1b | 801 | { |
777e6bd7 | 802 | int err = 0; |
7c4452b9 | 803 | int werr = 0; |
1728366e | 804 | struct address_space *mapping = root->fs_info->btree_inode->i_mapping; |
e6138876 | 805 | struct extent_state *cached_state = NULL; |
777e6bd7 | 806 | u64 start = 0; |
5f39d397 | 807 | u64 end; |
7c4452b9 | 808 | |
1728366e | 809 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
e6138876 JB |
810 | mark, &cached_state)) { |
811 | convert_extent_bit(dirty_pages, start, end, EXTENT_NEED_WAIT, | |
812 | mark, &cached_state, GFP_NOFS); | |
813 | cached_state = NULL; | |
1728366e JB |
814 | err = filemap_fdatawrite_range(mapping, start, end); |
815 | if (err) | |
816 | werr = err; | |
817 | cond_resched(); | |
818 | start = end + 1; | |
7c4452b9 | 819 | } |
690587d1 CM |
820 | if (err) |
821 | werr = err; | |
822 | return werr; | |
823 | } | |
824 | ||
825 | /* | |
826 | * when btree blocks are allocated, they have some corresponding bits set for | |
827 | * them in one of two extent_io trees. This is used to make sure all of | |
828 | * those extents are on disk for transaction or log commit. We wait | |
829 | * on all the pages and clear them from the dirty pages state tree | |
830 | */ | |
831 | int btrfs_wait_marked_extents(struct btrfs_root *root, | |
8cef4e16 | 832 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 | 833 | { |
690587d1 CM |
834 | int err = 0; |
835 | int werr = 0; | |
1728366e | 836 | struct address_space *mapping = root->fs_info->btree_inode->i_mapping; |
e6138876 | 837 | struct extent_state *cached_state = NULL; |
690587d1 CM |
838 | u64 start = 0; |
839 | u64 end; | |
777e6bd7 | 840 | |
1728366e | 841 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
e6138876 JB |
842 | EXTENT_NEED_WAIT, &cached_state)) { |
843 | clear_extent_bit(dirty_pages, start, end, EXTENT_NEED_WAIT, | |
844 | 0, 0, &cached_state, GFP_NOFS); | |
1728366e JB |
845 | err = filemap_fdatawait_range(mapping, start, end); |
846 | if (err) | |
847 | werr = err; | |
848 | cond_resched(); | |
849 | start = end + 1; | |
777e6bd7 | 850 | } |
7c4452b9 CM |
851 | if (err) |
852 | werr = err; | |
853 | return werr; | |
79154b1b CM |
854 | } |
855 | ||
690587d1 CM |
856 | /* |
857 | * when btree blocks are allocated, they have some corresponding bits set for | |
858 | * them in one of two extent_io trees. This is used to make sure all of | |
859 | * those extents are on disk for transaction or log commit | |
860 | */ | |
171170c1 | 861 | static int btrfs_write_and_wait_marked_extents(struct btrfs_root *root, |
8cef4e16 | 862 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 CM |
863 | { |
864 | int ret; | |
865 | int ret2; | |
c6adc9cc | 866 | struct blk_plug plug; |
690587d1 | 867 | |
c6adc9cc | 868 | blk_start_plug(&plug); |
8cef4e16 | 869 | ret = btrfs_write_marked_extents(root, dirty_pages, mark); |
c6adc9cc | 870 | blk_finish_plug(&plug); |
8cef4e16 | 871 | ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark); |
bf0da8c1 CM |
872 | |
873 | if (ret) | |
874 | return ret; | |
875 | if (ret2) | |
876 | return ret2; | |
877 | return 0; | |
690587d1 CM |
878 | } |
879 | ||
d0c803c4 CM |
880 | int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans, |
881 | struct btrfs_root *root) | |
882 | { | |
883 | if (!trans || !trans->transaction) { | |
884 | struct inode *btree_inode; | |
885 | btree_inode = root->fs_info->btree_inode; | |
886 | return filemap_write_and_wait(btree_inode->i_mapping); | |
887 | } | |
888 | return btrfs_write_and_wait_marked_extents(root, | |
8cef4e16 YZ |
889 | &trans->transaction->dirty_pages, |
890 | EXTENT_DIRTY); | |
d0c803c4 CM |
891 | } |
892 | ||
d352ac68 CM |
893 | /* |
894 | * this is used to update the root pointer in the tree of tree roots. | |
895 | * | |
896 | * But, in the case of the extent allocation tree, updating the root | |
897 | * pointer may allocate blocks which may change the root of the extent | |
898 | * allocation tree. | |
899 | * | |
900 | * So, this loops and repeats and makes sure the cowonly root didn't | |
901 | * change while the root pointer was being updated in the metadata. | |
902 | */ | |
0b86a832 CM |
903 | static int update_cowonly_root(struct btrfs_trans_handle *trans, |
904 | struct btrfs_root *root) | |
79154b1b CM |
905 | { |
906 | int ret; | |
0b86a832 | 907 | u64 old_root_bytenr; |
86b9f2ec | 908 | u64 old_root_used; |
0b86a832 | 909 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
79154b1b | 910 | |
86b9f2ec | 911 | old_root_used = btrfs_root_used(&root->root_item); |
0b86a832 | 912 | btrfs_write_dirty_block_groups(trans, root); |
56bec294 | 913 | |
d397712b | 914 | while (1) { |
0b86a832 | 915 | old_root_bytenr = btrfs_root_bytenr(&root->root_item); |
86b9f2ec YZ |
916 | if (old_root_bytenr == root->node->start && |
917 | old_root_used == btrfs_root_used(&root->root_item)) | |
79154b1b | 918 | break; |
87ef2bb4 | 919 | |
5d4f98a2 | 920 | btrfs_set_root_node(&root->root_item, root->node); |
79154b1b | 921 | ret = btrfs_update_root(trans, tree_root, |
0b86a832 CM |
922 | &root->root_key, |
923 | &root->root_item); | |
49b25e05 JM |
924 | if (ret) |
925 | return ret; | |
56bec294 | 926 | |
86b9f2ec | 927 | old_root_used = btrfs_root_used(&root->root_item); |
4a8c9a62 | 928 | ret = btrfs_write_dirty_block_groups(trans, root); |
49b25e05 JM |
929 | if (ret) |
930 | return ret; | |
0b86a832 | 931 | } |
276e680d YZ |
932 | |
933 | if (root != root->fs_info->extent_root) | |
934 | switch_commit_root(root); | |
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 | |
49b25e05 JM |
991 | ret = update_cowonly_root(trans, root); |
992 | if (ret) | |
993 | return ret; | |
79154b1b | 994 | } |
276e680d YZ |
995 | |
996 | down_write(&fs_info->extent_commit_sem); | |
997 | switch_commit_root(fs_info->extent_root); | |
998 | up_write(&fs_info->extent_commit_sem); | |
999 | ||
8dabb742 SB |
1000 | btrfs_after_dev_replace_commit(fs_info); |
1001 | ||
79154b1b CM |
1002 | return 0; |
1003 | } | |
1004 | ||
d352ac68 CM |
1005 | /* |
1006 | * dead roots are old snapshots that need to be deleted. This allocates | |
1007 | * a dirty root struct and adds it into the list of dead roots that need to | |
1008 | * be deleted | |
1009 | */ | |
cfad392b | 1010 | void btrfs_add_dead_root(struct btrfs_root *root) |
5eda7b5e | 1011 | { |
a4abeea4 | 1012 | spin_lock(&root->fs_info->trans_lock); |
cfad392b JB |
1013 | if (list_empty(&root->root_list)) |
1014 | list_add_tail(&root->root_list, &root->fs_info->dead_roots); | |
a4abeea4 | 1015 | spin_unlock(&root->fs_info->trans_lock); |
5eda7b5e CM |
1016 | } |
1017 | ||
d352ac68 | 1018 | /* |
5d4f98a2 | 1019 | * update all the cowonly tree roots on disk |
d352ac68 | 1020 | */ |
5d4f98a2 YZ |
1021 | static noinline int commit_fs_roots(struct btrfs_trans_handle *trans, |
1022 | struct btrfs_root *root) | |
0f7d52f4 | 1023 | { |
0f7d52f4 | 1024 | struct btrfs_root *gang[8]; |
5d4f98a2 | 1025 | struct btrfs_fs_info *fs_info = root->fs_info; |
0f7d52f4 CM |
1026 | int i; |
1027 | int ret; | |
54aa1f4d CM |
1028 | int err = 0; |
1029 | ||
a4abeea4 | 1030 | spin_lock(&fs_info->fs_roots_radix_lock); |
d397712b | 1031 | while (1) { |
5d4f98a2 YZ |
1032 | ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, |
1033 | (void **)gang, 0, | |
0f7d52f4 CM |
1034 | ARRAY_SIZE(gang), |
1035 | BTRFS_ROOT_TRANS_TAG); | |
1036 | if (ret == 0) | |
1037 | break; | |
1038 | for (i = 0; i < ret; i++) { | |
1039 | root = gang[i]; | |
5d4f98a2 YZ |
1040 | radix_tree_tag_clear(&fs_info->fs_roots_radix, |
1041 | (unsigned long)root->root_key.objectid, | |
1042 | BTRFS_ROOT_TRANS_TAG); | |
a4abeea4 | 1043 | spin_unlock(&fs_info->fs_roots_radix_lock); |
31153d81 | 1044 | |
e02119d5 | 1045 | btrfs_free_log(trans, root); |
5d4f98a2 | 1046 | btrfs_update_reloc_root(trans, root); |
d68fc57b | 1047 | btrfs_orphan_commit_root(trans, root); |
bcc63abb | 1048 | |
82d5902d LZ |
1049 | btrfs_save_ino_cache(root, trans); |
1050 | ||
f1ebcc74 LB |
1051 | /* see comments in should_cow_block() */ |
1052 | root->force_cow = 0; | |
1053 | smp_wmb(); | |
1054 | ||
978d910d | 1055 | if (root->commit_root != root->node) { |
581bb050 | 1056 | mutex_lock(&root->fs_commit_mutex); |
817d52f8 | 1057 | switch_commit_root(root); |
581bb050 LZ |
1058 | btrfs_unpin_free_ino(root); |
1059 | mutex_unlock(&root->fs_commit_mutex); | |
1060 | ||
978d910d YZ |
1061 | btrfs_set_root_node(&root->root_item, |
1062 | root->node); | |
1063 | } | |
5d4f98a2 | 1064 | |
5d4f98a2 | 1065 | err = btrfs_update_root(trans, fs_info->tree_root, |
0f7d52f4 CM |
1066 | &root->root_key, |
1067 | &root->root_item); | |
a4abeea4 | 1068 | spin_lock(&fs_info->fs_roots_radix_lock); |
54aa1f4d CM |
1069 | if (err) |
1070 | break; | |
0f7d52f4 CM |
1071 | } |
1072 | } | |
a4abeea4 | 1073 | spin_unlock(&fs_info->fs_roots_radix_lock); |
54aa1f4d | 1074 | return err; |
0f7d52f4 CM |
1075 | } |
1076 | ||
d352ac68 | 1077 | /* |
de78b51a ES |
1078 | * defrag a given btree. |
1079 | * Every leaf in the btree is read and defragged. | |
d352ac68 | 1080 | */ |
de78b51a | 1081 | int btrfs_defrag_root(struct btrfs_root *root) |
e9d0b13b CM |
1082 | { |
1083 | struct btrfs_fs_info *info = root->fs_info; | |
e9d0b13b | 1084 | struct btrfs_trans_handle *trans; |
8929ecfa | 1085 | int ret; |
e9d0b13b | 1086 | |
8929ecfa | 1087 | if (xchg(&root->defrag_running, 1)) |
e9d0b13b | 1088 | return 0; |
8929ecfa | 1089 | |
6b80053d | 1090 | while (1) { |
8929ecfa YZ |
1091 | trans = btrfs_start_transaction(root, 0); |
1092 | if (IS_ERR(trans)) | |
1093 | return PTR_ERR(trans); | |
1094 | ||
de78b51a | 1095 | ret = btrfs_defrag_leaves(trans, root); |
8929ecfa | 1096 | |
e9d0b13b | 1097 | btrfs_end_transaction(trans, root); |
b53d3f5d | 1098 | btrfs_btree_balance_dirty(info->tree_root); |
e9d0b13b CM |
1099 | cond_resched(); |
1100 | ||
7841cb28 | 1101 | if (btrfs_fs_closing(root->fs_info) || ret != -EAGAIN) |
e9d0b13b | 1102 | break; |
210549eb DS |
1103 | |
1104 | if (btrfs_defrag_cancelled(root->fs_info)) { | |
efe120a0 | 1105 | pr_debug("BTRFS: defrag_root cancelled\n"); |
210549eb DS |
1106 | ret = -EAGAIN; |
1107 | break; | |
1108 | } | |
e9d0b13b CM |
1109 | } |
1110 | root->defrag_running = 0; | |
8929ecfa | 1111 | return ret; |
e9d0b13b CM |
1112 | } |
1113 | ||
d352ac68 CM |
1114 | /* |
1115 | * new snapshots need to be created at a very specific time in the | |
aec8030a MX |
1116 | * transaction commit. This does the actual creation. |
1117 | * | |
1118 | * Note: | |
1119 | * If the error which may affect the commitment of the current transaction | |
1120 | * happens, we should return the error number. If the error which just affect | |
1121 | * the creation of the pending snapshots, just return 0. | |
d352ac68 | 1122 | */ |
80b6794d | 1123 | static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, |
3063d29f CM |
1124 | struct btrfs_fs_info *fs_info, |
1125 | struct btrfs_pending_snapshot *pending) | |
1126 | { | |
1127 | struct btrfs_key key; | |
80b6794d | 1128 | struct btrfs_root_item *new_root_item; |
3063d29f CM |
1129 | struct btrfs_root *tree_root = fs_info->tree_root; |
1130 | struct btrfs_root *root = pending->root; | |
6bdb72de | 1131 | struct btrfs_root *parent_root; |
98c9942a | 1132 | struct btrfs_block_rsv *rsv; |
6bdb72de | 1133 | struct inode *parent_inode; |
42874b3d MX |
1134 | struct btrfs_path *path; |
1135 | struct btrfs_dir_item *dir_item; | |
a22285a6 | 1136 | struct dentry *dentry; |
3063d29f | 1137 | struct extent_buffer *tmp; |
925baedd | 1138 | struct extent_buffer *old; |
8ea05e3a | 1139 | struct timespec cur_time = CURRENT_TIME; |
aec8030a | 1140 | int ret = 0; |
d68fc57b | 1141 | u64 to_reserve = 0; |
6bdb72de | 1142 | u64 index = 0; |
a22285a6 | 1143 | u64 objectid; |
b83cc969 | 1144 | u64 root_flags; |
8ea05e3a | 1145 | uuid_le new_uuid; |
3063d29f | 1146 | |
42874b3d MX |
1147 | path = btrfs_alloc_path(); |
1148 | if (!path) { | |
aec8030a MX |
1149 | pending->error = -ENOMEM; |
1150 | return 0; | |
42874b3d MX |
1151 | } |
1152 | ||
80b6794d CM |
1153 | new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS); |
1154 | if (!new_root_item) { | |
aec8030a | 1155 | pending->error = -ENOMEM; |
6fa9700e | 1156 | goto root_item_alloc_fail; |
80b6794d | 1157 | } |
a22285a6 | 1158 | |
aec8030a MX |
1159 | pending->error = btrfs_find_free_objectid(tree_root, &objectid); |
1160 | if (pending->error) | |
6fa9700e | 1161 | goto no_free_objectid; |
3063d29f | 1162 | |
3fd0a558 | 1163 | btrfs_reloc_pre_snapshot(trans, pending, &to_reserve); |
d68fc57b YZ |
1164 | |
1165 | if (to_reserve > 0) { | |
aec8030a MX |
1166 | pending->error = btrfs_block_rsv_add(root, |
1167 | &pending->block_rsv, | |
1168 | to_reserve, | |
1169 | BTRFS_RESERVE_NO_FLUSH); | |
1170 | if (pending->error) | |
6fa9700e | 1171 | goto no_free_objectid; |
d68fc57b YZ |
1172 | } |
1173 | ||
aec8030a MX |
1174 | pending->error = btrfs_qgroup_inherit(trans, fs_info, |
1175 | root->root_key.objectid, | |
1176 | objectid, pending->inherit); | |
1177 | if (pending->error) | |
6fa9700e | 1178 | goto no_free_objectid; |
6f72c7e2 | 1179 | |
3063d29f | 1180 | key.objectid = objectid; |
a22285a6 YZ |
1181 | key.offset = (u64)-1; |
1182 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3063d29f | 1183 | |
6fa9700e | 1184 | rsv = trans->block_rsv; |
a22285a6 | 1185 | trans->block_rsv = &pending->block_rsv; |
2382c5cc | 1186 | trans->bytes_reserved = trans->block_rsv->reserved; |
3de4586c | 1187 | |
a22285a6 | 1188 | dentry = pending->dentry; |
e9662f70 | 1189 | parent_inode = pending->dir; |
a22285a6 | 1190 | parent_root = BTRFS_I(parent_inode)->root; |
7585717f | 1191 | record_root_in_trans(trans, parent_root); |
a22285a6 | 1192 | |
3063d29f CM |
1193 | /* |
1194 | * insert the directory item | |
1195 | */ | |
3de4586c | 1196 | ret = btrfs_set_inode_index(parent_inode, &index); |
49b25e05 | 1197 | BUG_ON(ret); /* -ENOMEM */ |
42874b3d MX |
1198 | |
1199 | /* check if there is a file/dir which has the same name. */ | |
1200 | dir_item = btrfs_lookup_dir_item(NULL, parent_root, path, | |
1201 | btrfs_ino(parent_inode), | |
1202 | dentry->d_name.name, | |
1203 | dentry->d_name.len, 0); | |
1204 | if (dir_item != NULL && !IS_ERR(dir_item)) { | |
fe66a05a | 1205 | pending->error = -EEXIST; |
aec8030a | 1206 | goto dir_item_existed; |
42874b3d MX |
1207 | } else if (IS_ERR(dir_item)) { |
1208 | ret = PTR_ERR(dir_item); | |
8732d44f MX |
1209 | btrfs_abort_transaction(trans, root, ret); |
1210 | goto fail; | |
79787eaa | 1211 | } |
42874b3d | 1212 | btrfs_release_path(path); |
52c26179 | 1213 | |
e999376f CM |
1214 | /* |
1215 | * pull in the delayed directory update | |
1216 | * and the delayed inode item | |
1217 | * otherwise we corrupt the FS during | |
1218 | * snapshot | |
1219 | */ | |
1220 | ret = btrfs_run_delayed_items(trans, root); | |
8732d44f MX |
1221 | if (ret) { /* Transaction aborted */ |
1222 | btrfs_abort_transaction(trans, root, ret); | |
1223 | goto fail; | |
1224 | } | |
e999376f | 1225 | |
7585717f | 1226 | record_root_in_trans(trans, root); |
6bdb72de SW |
1227 | btrfs_set_root_last_snapshot(&root->root_item, trans->transid); |
1228 | memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); | |
08fe4db1 | 1229 | btrfs_check_and_init_root_item(new_root_item); |
6bdb72de | 1230 | |
b83cc969 LZ |
1231 | root_flags = btrfs_root_flags(new_root_item); |
1232 | if (pending->readonly) | |
1233 | root_flags |= BTRFS_ROOT_SUBVOL_RDONLY; | |
1234 | else | |
1235 | root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY; | |
1236 | btrfs_set_root_flags(new_root_item, root_flags); | |
1237 | ||
8ea05e3a AB |
1238 | btrfs_set_root_generation_v2(new_root_item, |
1239 | trans->transid); | |
1240 | uuid_le_gen(&new_uuid); | |
1241 | memcpy(new_root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE); | |
1242 | memcpy(new_root_item->parent_uuid, root->root_item.uuid, | |
1243 | BTRFS_UUID_SIZE); | |
70023da2 SB |
1244 | if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) { |
1245 | memset(new_root_item->received_uuid, 0, | |
1246 | sizeof(new_root_item->received_uuid)); | |
1247 | memset(&new_root_item->stime, 0, sizeof(new_root_item->stime)); | |
1248 | memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime)); | |
1249 | btrfs_set_root_stransid(new_root_item, 0); | |
1250 | btrfs_set_root_rtransid(new_root_item, 0); | |
1251 | } | |
3cae210f QW |
1252 | btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec); |
1253 | btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec); | |
8ea05e3a | 1254 | btrfs_set_root_otransid(new_root_item, trans->transid); |
8ea05e3a | 1255 | |
6bdb72de | 1256 | old = btrfs_lock_root_node(root); |
49b25e05 | 1257 | ret = btrfs_cow_block(trans, root, old, NULL, 0, &old); |
79787eaa JM |
1258 | if (ret) { |
1259 | btrfs_tree_unlock(old); | |
1260 | free_extent_buffer(old); | |
8732d44f MX |
1261 | btrfs_abort_transaction(trans, root, ret); |
1262 | goto fail; | |
79787eaa | 1263 | } |
49b25e05 | 1264 | |
6bdb72de SW |
1265 | btrfs_set_lock_blocking(old); |
1266 | ||
49b25e05 | 1267 | ret = btrfs_copy_root(trans, root, old, &tmp, objectid); |
79787eaa | 1268 | /* clean up in any case */ |
6bdb72de SW |
1269 | btrfs_tree_unlock(old); |
1270 | free_extent_buffer(old); | |
8732d44f MX |
1271 | if (ret) { |
1272 | btrfs_abort_transaction(trans, root, ret); | |
1273 | goto fail; | |
1274 | } | |
6bdb72de | 1275 | |
f1ebcc74 LB |
1276 | /* see comments in should_cow_block() */ |
1277 | root->force_cow = 1; | |
1278 | smp_wmb(); | |
1279 | ||
6bdb72de | 1280 | btrfs_set_root_node(new_root_item, tmp); |
a22285a6 YZ |
1281 | /* record when the snapshot was created in key.offset */ |
1282 | key.offset = trans->transid; | |
1283 | ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); | |
6bdb72de SW |
1284 | btrfs_tree_unlock(tmp); |
1285 | free_extent_buffer(tmp); | |
8732d44f MX |
1286 | if (ret) { |
1287 | btrfs_abort_transaction(trans, root, ret); | |
1288 | goto fail; | |
1289 | } | |
6bdb72de | 1290 | |
a22285a6 YZ |
1291 | /* |
1292 | * insert root back/forward references | |
1293 | */ | |
1294 | ret = btrfs_add_root_ref(trans, tree_root, objectid, | |
0660b5af | 1295 | parent_root->root_key.objectid, |
33345d01 | 1296 | btrfs_ino(parent_inode), index, |
a22285a6 | 1297 | dentry->d_name.name, dentry->d_name.len); |
8732d44f MX |
1298 | if (ret) { |
1299 | btrfs_abort_transaction(trans, root, ret); | |
1300 | goto fail; | |
1301 | } | |
0660b5af | 1302 | |
a22285a6 YZ |
1303 | key.offset = (u64)-1; |
1304 | pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key); | |
79787eaa JM |
1305 | if (IS_ERR(pending->snap)) { |
1306 | ret = PTR_ERR(pending->snap); | |
8732d44f MX |
1307 | btrfs_abort_transaction(trans, root, ret); |
1308 | goto fail; | |
79787eaa | 1309 | } |
d68fc57b | 1310 | |
49b25e05 | 1311 | ret = btrfs_reloc_post_snapshot(trans, pending); |
8732d44f MX |
1312 | if (ret) { |
1313 | btrfs_abort_transaction(trans, root, ret); | |
1314 | goto fail; | |
1315 | } | |
361048f5 MX |
1316 | |
1317 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); | |
8732d44f MX |
1318 | if (ret) { |
1319 | btrfs_abort_transaction(trans, root, ret); | |
1320 | goto fail; | |
1321 | } | |
42874b3d MX |
1322 | |
1323 | ret = btrfs_insert_dir_item(trans, parent_root, | |
1324 | dentry->d_name.name, dentry->d_name.len, | |
1325 | parent_inode, &key, | |
1326 | BTRFS_FT_DIR, index); | |
1327 | /* We have check then name at the beginning, so it is impossible. */ | |
9c52057c | 1328 | BUG_ON(ret == -EEXIST || ret == -EOVERFLOW); |
8732d44f MX |
1329 | if (ret) { |
1330 | btrfs_abort_transaction(trans, root, ret); | |
1331 | goto fail; | |
1332 | } | |
42874b3d MX |
1333 | |
1334 | btrfs_i_size_write(parent_inode, parent_inode->i_size + | |
1335 | dentry->d_name.len * 2); | |
1336 | parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME; | |
be6aef60 | 1337 | ret = btrfs_update_inode_fallback(trans, parent_root, parent_inode); |
dd5f9615 SB |
1338 | if (ret) { |
1339 | btrfs_abort_transaction(trans, root, ret); | |
1340 | goto fail; | |
1341 | } | |
1342 | ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root, new_uuid.b, | |
1343 | BTRFS_UUID_KEY_SUBVOL, objectid); | |
1344 | if (ret) { | |
8732d44f | 1345 | btrfs_abort_transaction(trans, root, ret); |
dd5f9615 SB |
1346 | goto fail; |
1347 | } | |
1348 | if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) { | |
1349 | ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root, | |
1350 | new_root_item->received_uuid, | |
1351 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, | |
1352 | objectid); | |
1353 | if (ret && ret != -EEXIST) { | |
1354 | btrfs_abort_transaction(trans, root, ret); | |
1355 | goto fail; | |
1356 | } | |
1357 | } | |
3063d29f | 1358 | fail: |
aec8030a MX |
1359 | pending->error = ret; |
1360 | dir_item_existed: | |
98c9942a | 1361 | trans->block_rsv = rsv; |
2382c5cc | 1362 | trans->bytes_reserved = 0; |
6fa9700e MX |
1363 | no_free_objectid: |
1364 | kfree(new_root_item); | |
1365 | root_item_alloc_fail: | |
42874b3d | 1366 | btrfs_free_path(path); |
49b25e05 | 1367 | return ret; |
3063d29f CM |
1368 | } |
1369 | ||
d352ac68 CM |
1370 | /* |
1371 | * create all the snapshots we've scheduled for creation | |
1372 | */ | |
80b6794d CM |
1373 | static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans, |
1374 | struct btrfs_fs_info *fs_info) | |
3de4586c | 1375 | { |
aec8030a | 1376 | struct btrfs_pending_snapshot *pending, *next; |
3de4586c | 1377 | struct list_head *head = &trans->transaction->pending_snapshots; |
aec8030a | 1378 | int ret = 0; |
3de4586c | 1379 | |
aec8030a MX |
1380 | list_for_each_entry_safe(pending, next, head, list) { |
1381 | list_del(&pending->list); | |
1382 | ret = create_pending_snapshot(trans, fs_info, pending); | |
1383 | if (ret) | |
1384 | break; | |
1385 | } | |
1386 | return ret; | |
3de4586c CM |
1387 | } |
1388 | ||
5d4f98a2 YZ |
1389 | static void update_super_roots(struct btrfs_root *root) |
1390 | { | |
1391 | struct btrfs_root_item *root_item; | |
1392 | struct btrfs_super_block *super; | |
1393 | ||
6c41761f | 1394 | super = root->fs_info->super_copy; |
5d4f98a2 YZ |
1395 | |
1396 | root_item = &root->fs_info->chunk_root->root_item; | |
1397 | super->chunk_root = root_item->bytenr; | |
1398 | super->chunk_root_generation = root_item->generation; | |
1399 | super->chunk_root_level = root_item->level; | |
1400 | ||
1401 | root_item = &root->fs_info->tree_root->root_item; | |
1402 | super->root = root_item->bytenr; | |
1403 | super->generation = root_item->generation; | |
1404 | super->root_level = root_item->level; | |
73bc1876 | 1405 | if (btrfs_test_opt(root, SPACE_CACHE)) |
0af3d00b | 1406 | super->cache_generation = root_item->generation; |
70f80175 SB |
1407 | if (root->fs_info->update_uuid_tree_gen) |
1408 | super->uuid_tree_generation = root_item->generation; | |
5d4f98a2 YZ |
1409 | } |
1410 | ||
f36f3042 CM |
1411 | int btrfs_transaction_in_commit(struct btrfs_fs_info *info) |
1412 | { | |
4a9d8bde | 1413 | struct btrfs_transaction *trans; |
f36f3042 | 1414 | int ret = 0; |
4a9d8bde | 1415 | |
a4abeea4 | 1416 | spin_lock(&info->trans_lock); |
4a9d8bde MX |
1417 | trans = info->running_transaction; |
1418 | if (trans) | |
1419 | ret = (trans->state >= TRANS_STATE_COMMIT_START); | |
a4abeea4 | 1420 | spin_unlock(&info->trans_lock); |
f36f3042 CM |
1421 | return ret; |
1422 | } | |
1423 | ||
8929ecfa YZ |
1424 | int btrfs_transaction_blocked(struct btrfs_fs_info *info) |
1425 | { | |
4a9d8bde | 1426 | struct btrfs_transaction *trans; |
8929ecfa | 1427 | int ret = 0; |
4a9d8bde | 1428 | |
a4abeea4 | 1429 | spin_lock(&info->trans_lock); |
4a9d8bde MX |
1430 | trans = info->running_transaction; |
1431 | if (trans) | |
1432 | ret = is_transaction_blocked(trans); | |
a4abeea4 | 1433 | spin_unlock(&info->trans_lock); |
8929ecfa YZ |
1434 | return ret; |
1435 | } | |
1436 | ||
bb9c12c9 SW |
1437 | /* |
1438 | * wait for the current transaction commit to start and block subsequent | |
1439 | * transaction joins | |
1440 | */ | |
1441 | static void wait_current_trans_commit_start(struct btrfs_root *root, | |
1442 | struct btrfs_transaction *trans) | |
1443 | { | |
4a9d8bde | 1444 | wait_event(root->fs_info->transaction_blocked_wait, |
501407aa JB |
1445 | trans->state >= TRANS_STATE_COMMIT_START || |
1446 | trans->aborted); | |
bb9c12c9 SW |
1447 | } |
1448 | ||
1449 | /* | |
1450 | * wait for the current transaction to start and then become unblocked. | |
1451 | * caller holds ref. | |
1452 | */ | |
1453 | static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root, | |
1454 | struct btrfs_transaction *trans) | |
1455 | { | |
72d63ed6 | 1456 | wait_event(root->fs_info->transaction_wait, |
501407aa JB |
1457 | trans->state >= TRANS_STATE_UNBLOCKED || |
1458 | trans->aborted); | |
bb9c12c9 SW |
1459 | } |
1460 | ||
1461 | /* | |
1462 | * commit transactions asynchronously. once btrfs_commit_transaction_async | |
1463 | * returns, any subsequent transaction will not be allowed to join. | |
1464 | */ | |
1465 | struct btrfs_async_commit { | |
1466 | struct btrfs_trans_handle *newtrans; | |
1467 | struct btrfs_root *root; | |
7892b5af | 1468 | struct work_struct work; |
bb9c12c9 SW |
1469 | }; |
1470 | ||
1471 | static void do_async_commit(struct work_struct *work) | |
1472 | { | |
1473 | struct btrfs_async_commit *ac = | |
7892b5af | 1474 | container_of(work, struct btrfs_async_commit, work); |
bb9c12c9 | 1475 | |
6fc4e354 SW |
1476 | /* |
1477 | * We've got freeze protection passed with the transaction. | |
1478 | * Tell lockdep about it. | |
1479 | */ | |
b1a06a4b | 1480 | if (ac->newtrans->type & __TRANS_FREEZABLE) |
ff7c1d33 MX |
1481 | rwsem_acquire_read( |
1482 | &ac->root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1], | |
1483 | 0, 1, _THIS_IP_); | |
6fc4e354 | 1484 | |
e209db7a SW |
1485 | current->journal_info = ac->newtrans; |
1486 | ||
bb9c12c9 SW |
1487 | btrfs_commit_transaction(ac->newtrans, ac->root); |
1488 | kfree(ac); | |
1489 | } | |
1490 | ||
1491 | int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans, | |
1492 | struct btrfs_root *root, | |
1493 | int wait_for_unblock) | |
1494 | { | |
1495 | struct btrfs_async_commit *ac; | |
1496 | struct btrfs_transaction *cur_trans; | |
1497 | ||
1498 | ac = kmalloc(sizeof(*ac), GFP_NOFS); | |
db5b493a TI |
1499 | if (!ac) |
1500 | return -ENOMEM; | |
bb9c12c9 | 1501 | |
7892b5af | 1502 | INIT_WORK(&ac->work, do_async_commit); |
bb9c12c9 | 1503 | ac->root = root; |
7a7eaa40 | 1504 | ac->newtrans = btrfs_join_transaction(root); |
3612b495 TI |
1505 | if (IS_ERR(ac->newtrans)) { |
1506 | int err = PTR_ERR(ac->newtrans); | |
1507 | kfree(ac); | |
1508 | return err; | |
1509 | } | |
bb9c12c9 SW |
1510 | |
1511 | /* take transaction reference */ | |
bb9c12c9 | 1512 | cur_trans = trans->transaction; |
13c5a93e | 1513 | atomic_inc(&cur_trans->use_count); |
bb9c12c9 SW |
1514 | |
1515 | btrfs_end_transaction(trans, root); | |
6fc4e354 SW |
1516 | |
1517 | /* | |
1518 | * Tell lockdep we've released the freeze rwsem, since the | |
1519 | * async commit thread will be the one to unlock it. | |
1520 | */ | |
b1a06a4b | 1521 | if (ac->newtrans->type & __TRANS_FREEZABLE) |
ff7c1d33 MX |
1522 | rwsem_release( |
1523 | &root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1], | |
1524 | 1, _THIS_IP_); | |
6fc4e354 | 1525 | |
7892b5af | 1526 | schedule_work(&ac->work); |
bb9c12c9 SW |
1527 | |
1528 | /* wait for transaction to start and unblock */ | |
bb9c12c9 SW |
1529 | if (wait_for_unblock) |
1530 | wait_current_trans_commit_start_and_unblock(root, cur_trans); | |
1531 | else | |
1532 | wait_current_trans_commit_start(root, cur_trans); | |
bb9c12c9 | 1533 | |
38e88054 SW |
1534 | if (current->journal_info == trans) |
1535 | current->journal_info = NULL; | |
1536 | ||
724e2315 | 1537 | btrfs_put_transaction(cur_trans); |
bb9c12c9 SW |
1538 | return 0; |
1539 | } | |
1540 | ||
49b25e05 JM |
1541 | |
1542 | static void cleanup_transaction(struct btrfs_trans_handle *trans, | |
7b8b92af | 1543 | struct btrfs_root *root, int err) |
49b25e05 JM |
1544 | { |
1545 | struct btrfs_transaction *cur_trans = trans->transaction; | |
f094ac32 | 1546 | DEFINE_WAIT(wait); |
49b25e05 JM |
1547 | |
1548 | WARN_ON(trans->use_count > 1); | |
1549 | ||
7b8b92af JB |
1550 | btrfs_abort_transaction(trans, root, err); |
1551 | ||
49b25e05 | 1552 | spin_lock(&root->fs_info->trans_lock); |
66b6135b | 1553 | |
25d8c284 MX |
1554 | /* |
1555 | * If the transaction is removed from the list, it means this | |
1556 | * transaction has been committed successfully, so it is impossible | |
1557 | * to call the cleanup function. | |
1558 | */ | |
1559 | BUG_ON(list_empty(&cur_trans->list)); | |
66b6135b | 1560 | |
49b25e05 | 1561 | list_del_init(&cur_trans->list); |
d7096fc3 | 1562 | if (cur_trans == root->fs_info->running_transaction) { |
4a9d8bde | 1563 | cur_trans->state = TRANS_STATE_COMMIT_DOING; |
f094ac32 LB |
1564 | spin_unlock(&root->fs_info->trans_lock); |
1565 | wait_event(cur_trans->writer_wait, | |
1566 | atomic_read(&cur_trans->num_writers) == 1); | |
1567 | ||
1568 | spin_lock(&root->fs_info->trans_lock); | |
d7096fc3 | 1569 | } |
49b25e05 JM |
1570 | spin_unlock(&root->fs_info->trans_lock); |
1571 | ||
1572 | btrfs_cleanup_one_transaction(trans->transaction, root); | |
1573 | ||
4a9d8bde MX |
1574 | spin_lock(&root->fs_info->trans_lock); |
1575 | if (cur_trans == root->fs_info->running_transaction) | |
1576 | root->fs_info->running_transaction = NULL; | |
1577 | spin_unlock(&root->fs_info->trans_lock); | |
1578 | ||
e0228285 JB |
1579 | if (trans->type & __TRANS_FREEZABLE) |
1580 | sb_end_intwrite(root->fs_info->sb); | |
724e2315 JB |
1581 | btrfs_put_transaction(cur_trans); |
1582 | btrfs_put_transaction(cur_trans); | |
49b25e05 JM |
1583 | |
1584 | trace_btrfs_transaction_commit(root); | |
1585 | ||
1586 | btrfs_scrub_continue(root); | |
1587 | ||
1588 | if (current->journal_info == trans) | |
1589 | current->journal_info = NULL; | |
1590 | ||
1591 | kmem_cache_free(btrfs_trans_handle_cachep, trans); | |
1592 | } | |
1593 | ||
ca469637 MX |
1594 | static int btrfs_flush_all_pending_stuffs(struct btrfs_trans_handle *trans, |
1595 | struct btrfs_root *root) | |
1596 | { | |
ca469637 MX |
1597 | int ret; |
1598 | ||
ca469637 | 1599 | ret = btrfs_run_delayed_items(trans, root); |
ca469637 MX |
1600 | /* |
1601 | * running the delayed items may have added new refs. account | |
1602 | * them now so that they hinder processing of more delayed refs | |
1603 | * as little as possible. | |
1604 | */ | |
80d94fb3 FDBM |
1605 | if (ret) { |
1606 | btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info); | |
1607 | return ret; | |
1608 | } | |
1609 | ||
1610 | ret = btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info); | |
1611 | if (ret) | |
1612 | return ret; | |
ca469637 MX |
1613 | |
1614 | /* | |
1615 | * rename don't use btrfs_join_transaction, so, once we | |
1616 | * set the transaction to blocked above, we aren't going | |
1617 | * to get any new ordered operations. We can safely run | |
1618 | * it here and no for sure that nothing new will be added | |
1619 | * to the list | |
1620 | */ | |
569e0f35 | 1621 | ret = btrfs_run_ordered_operations(trans, root, 1); |
ca469637 | 1622 | |
eebc6084 | 1623 | return ret; |
ca469637 MX |
1624 | } |
1625 | ||
82436617 MX |
1626 | static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info) |
1627 | { | |
1628 | if (btrfs_test_opt(fs_info->tree_root, FLUSHONCOMMIT)) | |
91aef86f | 1629 | return btrfs_start_delalloc_roots(fs_info, 1); |
82436617 MX |
1630 | return 0; |
1631 | } | |
1632 | ||
1633 | static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info) | |
1634 | { | |
1635 | if (btrfs_test_opt(fs_info->tree_root, FLUSHONCOMMIT)) | |
b0244199 | 1636 | btrfs_wait_ordered_roots(fs_info, -1); |
82436617 MX |
1637 | } |
1638 | ||
79154b1b CM |
1639 | int btrfs_commit_transaction(struct btrfs_trans_handle *trans, |
1640 | struct btrfs_root *root) | |
1641 | { | |
49b25e05 | 1642 | struct btrfs_transaction *cur_trans = trans->transaction; |
8fd17795 | 1643 | struct btrfs_transaction *prev_trans = NULL; |
25287e0a | 1644 | int ret; |
79154b1b | 1645 | |
569e0f35 | 1646 | ret = btrfs_run_ordered_operations(trans, root, 0); |
25287e0a MX |
1647 | if (ret) { |
1648 | btrfs_abort_transaction(trans, root, ret); | |
e4a2bcac JB |
1649 | btrfs_end_transaction(trans, root); |
1650 | return ret; | |
25287e0a | 1651 | } |
5a3f23d5 | 1652 | |
8d25a086 MX |
1653 | /* Stop the commit early if ->aborted is set */ |
1654 | if (unlikely(ACCESS_ONCE(cur_trans->aborted))) { | |
25287e0a | 1655 | ret = cur_trans->aborted; |
e4a2bcac JB |
1656 | btrfs_end_transaction(trans, root); |
1657 | return ret; | |
25287e0a | 1658 | } |
49b25e05 | 1659 | |
56bec294 CM |
1660 | /* make a pass through all the delayed refs we have so far |
1661 | * any runnings procs may add more while we are here | |
1662 | */ | |
1663 | ret = btrfs_run_delayed_refs(trans, root, 0); | |
e4a2bcac JB |
1664 | if (ret) { |
1665 | btrfs_end_transaction(trans, root); | |
1666 | return ret; | |
1667 | } | |
56bec294 | 1668 | |
0e721106 JB |
1669 | btrfs_trans_release_metadata(trans, root); |
1670 | trans->block_rsv = NULL; | |
272d26d0 MX |
1671 | if (trans->qgroup_reserved) { |
1672 | btrfs_qgroup_free(root, trans->qgroup_reserved); | |
1673 | trans->qgroup_reserved = 0; | |
1674 | } | |
0e721106 | 1675 | |
b7ec40d7 | 1676 | cur_trans = trans->transaction; |
49b25e05 | 1677 | |
56bec294 CM |
1678 | /* |
1679 | * set the flushing flag so procs in this transaction have to | |
1680 | * start sending their work down. | |
1681 | */ | |
b7ec40d7 | 1682 | cur_trans->delayed_refs.flushing = 1; |
1be41b78 | 1683 | smp_wmb(); |
56bec294 | 1684 | |
ea658bad JB |
1685 | if (!list_empty(&trans->new_bgs)) |
1686 | btrfs_create_pending_block_groups(trans, root); | |
1687 | ||
c3e69d58 | 1688 | ret = btrfs_run_delayed_refs(trans, root, 0); |
e4a2bcac JB |
1689 | if (ret) { |
1690 | btrfs_end_transaction(trans, root); | |
1691 | return ret; | |
1692 | } | |
56bec294 | 1693 | |
4a9d8bde MX |
1694 | spin_lock(&root->fs_info->trans_lock); |
1695 | if (cur_trans->state >= TRANS_STATE_COMMIT_START) { | |
1696 | spin_unlock(&root->fs_info->trans_lock); | |
13c5a93e | 1697 | atomic_inc(&cur_trans->use_count); |
49b25e05 | 1698 | ret = btrfs_end_transaction(trans, root); |
ccd467d6 | 1699 | |
b9c8300c | 1700 | wait_for_commit(root, cur_trans); |
15ee9bc7 | 1701 | |
724e2315 | 1702 | btrfs_put_transaction(cur_trans); |
15ee9bc7 | 1703 | |
49b25e05 | 1704 | return ret; |
79154b1b | 1705 | } |
4313b399 | 1706 | |
4a9d8bde | 1707 | cur_trans->state = TRANS_STATE_COMMIT_START; |
bb9c12c9 SW |
1708 | wake_up(&root->fs_info->transaction_blocked_wait); |
1709 | ||
ccd467d6 CM |
1710 | if (cur_trans->list.prev != &root->fs_info->trans_list) { |
1711 | prev_trans = list_entry(cur_trans->list.prev, | |
1712 | struct btrfs_transaction, list); | |
4a9d8bde | 1713 | if (prev_trans->state != TRANS_STATE_COMPLETED) { |
13c5a93e | 1714 | atomic_inc(&prev_trans->use_count); |
a4abeea4 | 1715 | spin_unlock(&root->fs_info->trans_lock); |
ccd467d6 CM |
1716 | |
1717 | wait_for_commit(root, prev_trans); | |
ccd467d6 | 1718 | |
724e2315 | 1719 | btrfs_put_transaction(prev_trans); |
a4abeea4 JB |
1720 | } else { |
1721 | spin_unlock(&root->fs_info->trans_lock); | |
ccd467d6 | 1722 | } |
a4abeea4 JB |
1723 | } else { |
1724 | spin_unlock(&root->fs_info->trans_lock); | |
ccd467d6 | 1725 | } |
15ee9bc7 | 1726 | |
0860adfd MX |
1727 | extwriter_counter_dec(cur_trans, trans->type); |
1728 | ||
82436617 MX |
1729 | ret = btrfs_start_delalloc_flush(root->fs_info); |
1730 | if (ret) | |
1731 | goto cleanup_transaction; | |
1732 | ||
581227d0 MX |
1733 | ret = btrfs_flush_all_pending_stuffs(trans, root); |
1734 | if (ret) | |
1735 | goto cleanup_transaction; | |
15ee9bc7 | 1736 | |
581227d0 MX |
1737 | wait_event(cur_trans->writer_wait, |
1738 | extwriter_counter_read(cur_trans) == 0); | |
15ee9bc7 | 1739 | |
581227d0 | 1740 | /* some pending stuffs might be added after the previous flush. */ |
ca469637 MX |
1741 | ret = btrfs_flush_all_pending_stuffs(trans, root); |
1742 | if (ret) | |
1743 | goto cleanup_transaction; | |
1744 | ||
82436617 | 1745 | btrfs_wait_delalloc_flush(root->fs_info); |
cb7ab021 WS |
1746 | |
1747 | btrfs_scrub_pause(root); | |
ed0ca140 JB |
1748 | /* |
1749 | * Ok now we need to make sure to block out any other joins while we | |
1750 | * commit the transaction. We could have started a join before setting | |
4a9d8bde | 1751 | * COMMIT_DOING so make sure to wait for num_writers to == 1 again. |
ed0ca140 JB |
1752 | */ |
1753 | spin_lock(&root->fs_info->trans_lock); | |
4a9d8bde | 1754 | cur_trans->state = TRANS_STATE_COMMIT_DOING; |
ed0ca140 JB |
1755 | spin_unlock(&root->fs_info->trans_lock); |
1756 | wait_event(cur_trans->writer_wait, | |
1757 | atomic_read(&cur_trans->num_writers) == 1); | |
1758 | ||
2cba30f1 MX |
1759 | /* ->aborted might be set after the previous check, so check it */ |
1760 | if (unlikely(ACCESS_ONCE(cur_trans->aborted))) { | |
1761 | ret = cur_trans->aborted; | |
1762 | goto cleanup_transaction; | |
1763 | } | |
7585717f CM |
1764 | /* |
1765 | * the reloc mutex makes sure that we stop | |
1766 | * the balancing code from coming in and moving | |
1767 | * extents around in the middle of the commit | |
1768 | */ | |
1769 | mutex_lock(&root->fs_info->reloc_mutex); | |
1770 | ||
42874b3d MX |
1771 | /* |
1772 | * We needn't worry about the delayed items because we will | |
1773 | * deal with them in create_pending_snapshot(), which is the | |
1774 | * core function of the snapshot creation. | |
1775 | */ | |
1776 | ret = create_pending_snapshots(trans, root->fs_info); | |
49b25e05 JM |
1777 | if (ret) { |
1778 | mutex_unlock(&root->fs_info->reloc_mutex); | |
1779 | goto cleanup_transaction; | |
1780 | } | |
3063d29f | 1781 | |
42874b3d MX |
1782 | /* |
1783 | * We insert the dir indexes of the snapshots and update the inode | |
1784 | * of the snapshots' parents after the snapshot creation, so there | |
1785 | * are some delayed items which are not dealt with. Now deal with | |
1786 | * them. | |
1787 | * | |
1788 | * We needn't worry that this operation will corrupt the snapshots, | |
1789 | * because all the tree which are snapshoted will be forced to COW | |
1790 | * the nodes and leaves. | |
1791 | */ | |
1792 | ret = btrfs_run_delayed_items(trans, root); | |
49b25e05 JM |
1793 | if (ret) { |
1794 | mutex_unlock(&root->fs_info->reloc_mutex); | |
1795 | goto cleanup_transaction; | |
1796 | } | |
16cdcec7 | 1797 | |
56bec294 | 1798 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
49b25e05 JM |
1799 | if (ret) { |
1800 | mutex_unlock(&root->fs_info->reloc_mutex); | |
1801 | goto cleanup_transaction; | |
1802 | } | |
56bec294 | 1803 | |
e999376f CM |
1804 | /* |
1805 | * make sure none of the code above managed to slip in a | |
1806 | * delayed item | |
1807 | */ | |
1808 | btrfs_assert_delayed_root_empty(root); | |
1809 | ||
2c90e5d6 | 1810 | WARN_ON(cur_trans != trans->transaction); |
dc17ff8f | 1811 | |
e02119d5 CM |
1812 | /* btrfs_commit_tree_roots is responsible for getting the |
1813 | * various roots consistent with each other. Every pointer | |
1814 | * in the tree of tree roots has to point to the most up to date | |
1815 | * root for every subvolume and other tree. So, we have to keep | |
1816 | * the tree logging code from jumping in and changing any | |
1817 | * of the trees. | |
1818 | * | |
1819 | * At this point in the commit, there can't be any tree-log | |
1820 | * writers, but a little lower down we drop the trans mutex | |
1821 | * and let new people in. By holding the tree_log_mutex | |
1822 | * from now until after the super is written, we avoid races | |
1823 | * with the tree-log code. | |
1824 | */ | |
1825 | mutex_lock(&root->fs_info->tree_log_mutex); | |
1826 | ||
5d4f98a2 | 1827 | ret = commit_fs_roots(trans, root); |
49b25e05 JM |
1828 | if (ret) { |
1829 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
871383be | 1830 | mutex_unlock(&root->fs_info->reloc_mutex); |
49b25e05 JM |
1831 | goto cleanup_transaction; |
1832 | } | |
54aa1f4d | 1833 | |
5d4f98a2 | 1834 | /* commit_fs_roots gets rid of all the tree log roots, it is now |
e02119d5 CM |
1835 | * safe to free the root of tree log roots |
1836 | */ | |
1837 | btrfs_free_log_root_tree(trans, root->fs_info); | |
1838 | ||
5d4f98a2 | 1839 | ret = commit_cowonly_roots(trans, root); |
49b25e05 JM |
1840 | if (ret) { |
1841 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
871383be | 1842 | mutex_unlock(&root->fs_info->reloc_mutex); |
49b25e05 JM |
1843 | goto cleanup_transaction; |
1844 | } | |
54aa1f4d | 1845 | |
2cba30f1 MX |
1846 | /* |
1847 | * The tasks which save the space cache and inode cache may also | |
1848 | * update ->aborted, check it. | |
1849 | */ | |
1850 | if (unlikely(ACCESS_ONCE(cur_trans->aborted))) { | |
1851 | ret = cur_trans->aborted; | |
1852 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
1853 | mutex_unlock(&root->fs_info->reloc_mutex); | |
1854 | goto cleanup_transaction; | |
1855 | } | |
1856 | ||
11833d66 YZ |
1857 | btrfs_prepare_extent_commit(trans, root); |
1858 | ||
78fae27e | 1859 | cur_trans = root->fs_info->running_transaction; |
5d4f98a2 YZ |
1860 | |
1861 | btrfs_set_root_node(&root->fs_info->tree_root->root_item, | |
1862 | root->fs_info->tree_root->node); | |
817d52f8 | 1863 | switch_commit_root(root->fs_info->tree_root); |
5d4f98a2 YZ |
1864 | |
1865 | btrfs_set_root_node(&root->fs_info->chunk_root->root_item, | |
1866 | root->fs_info->chunk_root->node); | |
817d52f8 | 1867 | switch_commit_root(root->fs_info->chunk_root); |
5d4f98a2 | 1868 | |
edf39272 | 1869 | assert_qgroups_uptodate(trans); |
5d4f98a2 | 1870 | update_super_roots(root); |
e02119d5 | 1871 | |
60e7cd3a JB |
1872 | btrfs_set_super_log_root(root->fs_info->super_copy, 0); |
1873 | btrfs_set_super_log_root_level(root->fs_info->super_copy, 0); | |
6c41761f DS |
1874 | memcpy(root->fs_info->super_for_commit, root->fs_info->super_copy, |
1875 | sizeof(*root->fs_info->super_copy)); | |
ccd467d6 | 1876 | |
a4abeea4 | 1877 | spin_lock(&root->fs_info->trans_lock); |
4a9d8bde | 1878 | cur_trans->state = TRANS_STATE_UNBLOCKED; |
a4abeea4 | 1879 | root->fs_info->running_transaction = NULL; |
a4abeea4 | 1880 | spin_unlock(&root->fs_info->trans_lock); |
7585717f | 1881 | mutex_unlock(&root->fs_info->reloc_mutex); |
b7ec40d7 | 1882 | |
f9295749 | 1883 | wake_up(&root->fs_info->transaction_wait); |
e6dcd2dc | 1884 | |
79154b1b | 1885 | ret = btrfs_write_and_wait_transaction(trans, root); |
49b25e05 JM |
1886 | if (ret) { |
1887 | btrfs_error(root->fs_info, ret, | |
08748810 | 1888 | "Error while writing out transaction"); |
49b25e05 JM |
1889 | mutex_unlock(&root->fs_info->tree_log_mutex); |
1890 | goto cleanup_transaction; | |
1891 | } | |
1892 | ||
1893 | ret = write_ctree_super(trans, root, 0); | |
1894 | if (ret) { | |
1895 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
1896 | goto cleanup_transaction; | |
1897 | } | |
4313b399 | 1898 | |
e02119d5 CM |
1899 | /* |
1900 | * the super is written, we can safely allow the tree-loggers | |
1901 | * to go about their business | |
1902 | */ | |
1903 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
1904 | ||
11833d66 | 1905 | btrfs_finish_extent_commit(trans, root); |
4313b399 | 1906 | |
15ee9bc7 | 1907 | root->fs_info->last_trans_committed = cur_trans->transid; |
4a9d8bde MX |
1908 | /* |
1909 | * We needn't acquire the lock here because there is no other task | |
1910 | * which can change it. | |
1911 | */ | |
1912 | cur_trans->state = TRANS_STATE_COMPLETED; | |
2c90e5d6 | 1913 | wake_up(&cur_trans->commit_wait); |
3de4586c | 1914 | |
a4abeea4 | 1915 | spin_lock(&root->fs_info->trans_lock); |
13c5a93e | 1916 | list_del_init(&cur_trans->list); |
a4abeea4 JB |
1917 | spin_unlock(&root->fs_info->trans_lock); |
1918 | ||
724e2315 JB |
1919 | btrfs_put_transaction(cur_trans); |
1920 | btrfs_put_transaction(cur_trans); | |
58176a96 | 1921 | |
0860adfd | 1922 | if (trans->type & __TRANS_FREEZABLE) |
354aa0fb | 1923 | sb_end_intwrite(root->fs_info->sb); |
b2b5ef5c | 1924 | |
1abe9b8a | 1925 | trace_btrfs_transaction_commit(root); |
1926 | ||
a2de733c AJ |
1927 | btrfs_scrub_continue(root); |
1928 | ||
9ed74f2d JB |
1929 | if (current->journal_info == trans) |
1930 | current->journal_info = NULL; | |
1931 | ||
2c90e5d6 | 1932 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
24bbcf04 YZ |
1933 | |
1934 | if (current != root->fs_info->transaction_kthread) | |
1935 | btrfs_run_delayed_iputs(root); | |
1936 | ||
79154b1b | 1937 | return ret; |
49b25e05 JM |
1938 | |
1939 | cleanup_transaction: | |
0e721106 JB |
1940 | btrfs_trans_release_metadata(trans, root); |
1941 | trans->block_rsv = NULL; | |
272d26d0 MX |
1942 | if (trans->qgroup_reserved) { |
1943 | btrfs_qgroup_free(root, trans->qgroup_reserved); | |
1944 | trans->qgroup_reserved = 0; | |
1945 | } | |
c2cf52eb | 1946 | btrfs_warn(root->fs_info, "Skipping commit of aborted transaction."); |
49b25e05 JM |
1947 | if (current->journal_info == trans) |
1948 | current->journal_info = NULL; | |
7b8b92af | 1949 | cleanup_transaction(trans, root, ret); |
49b25e05 JM |
1950 | |
1951 | return ret; | |
79154b1b CM |
1952 | } |
1953 | ||
d352ac68 | 1954 | /* |
9d1a2a3a DS |
1955 | * return < 0 if error |
1956 | * 0 if there are no more dead_roots at the time of call | |
1957 | * 1 there are more to be processed, call me again | |
1958 | * | |
1959 | * The return value indicates there are certainly more snapshots to delete, but | |
1960 | * if there comes a new one during processing, it may return 0. We don't mind, | |
1961 | * because btrfs_commit_super will poke cleaner thread and it will process it a | |
1962 | * few seconds later. | |
d352ac68 | 1963 | */ |
9d1a2a3a | 1964 | int btrfs_clean_one_deleted_snapshot(struct btrfs_root *root) |
e9d0b13b | 1965 | { |
9d1a2a3a | 1966 | int ret; |
5d4f98a2 YZ |
1967 | struct btrfs_fs_info *fs_info = root->fs_info; |
1968 | ||
a4abeea4 | 1969 | spin_lock(&fs_info->trans_lock); |
9d1a2a3a DS |
1970 | if (list_empty(&fs_info->dead_roots)) { |
1971 | spin_unlock(&fs_info->trans_lock); | |
1972 | return 0; | |
1973 | } | |
1974 | root = list_first_entry(&fs_info->dead_roots, | |
1975 | struct btrfs_root, root_list); | |
18f687d5 WS |
1976 | /* |
1977 | * Make sure root is not involved in send, | |
1978 | * if we fail with first root, we return | |
1979 | * directly rather than continue. | |
1980 | */ | |
1981 | spin_lock(&root->root_item_lock); | |
1982 | if (root->send_in_progress) { | |
1983 | spin_unlock(&fs_info->trans_lock); | |
1984 | spin_unlock(&root->root_item_lock); | |
1985 | return 0; | |
1986 | } | |
1987 | spin_unlock(&root->root_item_lock); | |
1988 | ||
cfad392b | 1989 | list_del_init(&root->root_list); |
a4abeea4 | 1990 | spin_unlock(&fs_info->trans_lock); |
e9d0b13b | 1991 | |
efe120a0 | 1992 | pr_debug("BTRFS: cleaner removing %llu\n", root->objectid); |
76dda93c | 1993 | |
9d1a2a3a | 1994 | btrfs_kill_all_delayed_nodes(root); |
16cdcec7 | 1995 | |
9d1a2a3a DS |
1996 | if (btrfs_header_backref_rev(root->node) < |
1997 | BTRFS_MIXED_BACKREF_REV) | |
1998 | ret = btrfs_drop_snapshot(root, NULL, 0, 0); | |
1999 | else | |
2000 | ret = btrfs_drop_snapshot(root, NULL, 1, 0); | |
2001 | /* | |
2002 | * If we encounter a transaction abort during snapshot cleaning, we | |
2003 | * don't want to crash here | |
2004 | */ | |
6596a928 | 2005 | return (ret < 0) ? 0 : 1; |
e9d0b13b | 2006 | } |