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 | ||
48a3b636 | 60 | static void 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)); |
00f04b88 | 65 | WARN_ON(transaction->delayed_refs.root.rb_node); |
2c90e5d6 | 66 | kmem_cache_free(btrfs_transaction_cachep, transaction); |
78fae27e | 67 | } |
79154b1b CM |
68 | } |
69 | ||
817d52f8 JB |
70 | static noinline void switch_commit_root(struct btrfs_root *root) |
71 | { | |
817d52f8 JB |
72 | free_extent_buffer(root->commit_root); |
73 | root->commit_root = btrfs_root_node(root); | |
817d52f8 JB |
74 | } |
75 | ||
0860adfd MX |
76 | static inline void extwriter_counter_inc(struct btrfs_transaction *trans, |
77 | unsigned int type) | |
78 | { | |
79 | if (type & TRANS_EXTWRITERS) | |
80 | atomic_inc(&trans->num_extwriters); | |
81 | } | |
82 | ||
83 | static inline void extwriter_counter_dec(struct btrfs_transaction *trans, | |
84 | unsigned int type) | |
85 | { | |
86 | if (type & TRANS_EXTWRITERS) | |
87 | atomic_dec(&trans->num_extwriters); | |
88 | } | |
89 | ||
90 | static inline void extwriter_counter_init(struct btrfs_transaction *trans, | |
91 | unsigned int type) | |
92 | { | |
93 | atomic_set(&trans->num_extwriters, ((type & TRANS_EXTWRITERS) ? 1 : 0)); | |
94 | } | |
95 | ||
96 | static inline int extwriter_counter_read(struct btrfs_transaction *trans) | |
97 | { | |
98 | return atomic_read(&trans->num_extwriters); | |
178260b2 MX |
99 | } |
100 | ||
d352ac68 CM |
101 | /* |
102 | * either allocate a new transaction or hop into the existing one | |
103 | */ | |
0860adfd | 104 | static noinline int join_transaction(struct btrfs_root *root, unsigned int type) |
79154b1b CM |
105 | { |
106 | struct btrfs_transaction *cur_trans; | |
19ae4e81 | 107 | struct btrfs_fs_info *fs_info = root->fs_info; |
a4abeea4 | 108 | |
19ae4e81 | 109 | spin_lock(&fs_info->trans_lock); |
d43317dc | 110 | loop: |
49b25e05 | 111 | /* The file system has been taken offline. No new transactions. */ |
87533c47 | 112 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { |
19ae4e81 | 113 | spin_unlock(&fs_info->trans_lock); |
49b25e05 JM |
114 | return -EROFS; |
115 | } | |
116 | ||
19ae4e81 | 117 | cur_trans = fs_info->running_transaction; |
a4abeea4 | 118 | if (cur_trans) { |
871383be | 119 | if (cur_trans->aborted) { |
19ae4e81 | 120 | spin_unlock(&fs_info->trans_lock); |
49b25e05 | 121 | return cur_trans->aborted; |
871383be | 122 | } |
4a9d8bde | 123 | if (btrfs_blocked_trans_types[cur_trans->state] & type) { |
178260b2 MX |
124 | spin_unlock(&fs_info->trans_lock); |
125 | return -EBUSY; | |
126 | } | |
a4abeea4 | 127 | atomic_inc(&cur_trans->use_count); |
13c5a93e | 128 | atomic_inc(&cur_trans->num_writers); |
0860adfd | 129 | extwriter_counter_inc(cur_trans, type); |
19ae4e81 | 130 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 | 131 | return 0; |
79154b1b | 132 | } |
19ae4e81 | 133 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 | 134 | |
354aa0fb MX |
135 | /* |
136 | * If we are ATTACH, we just want to catch the current transaction, | |
137 | * and commit it. If there is no transaction, just return ENOENT. | |
138 | */ | |
139 | if (type == TRANS_ATTACH) | |
140 | return -ENOENT; | |
141 | ||
4a9d8bde MX |
142 | /* |
143 | * JOIN_NOLOCK only happens during the transaction commit, so | |
144 | * it is impossible that ->running_transaction is NULL | |
145 | */ | |
146 | BUG_ON(type == TRANS_JOIN_NOLOCK); | |
147 | ||
a4abeea4 JB |
148 | cur_trans = kmem_cache_alloc(btrfs_transaction_cachep, GFP_NOFS); |
149 | if (!cur_trans) | |
150 | return -ENOMEM; | |
d43317dc | 151 | |
19ae4e81 JS |
152 | spin_lock(&fs_info->trans_lock); |
153 | if (fs_info->running_transaction) { | |
d43317dc CM |
154 | /* |
155 | * someone started a transaction after we unlocked. Make sure | |
4a9d8bde | 156 | * to redo the checks above |
d43317dc | 157 | */ |
a4abeea4 | 158 | kmem_cache_free(btrfs_transaction_cachep, cur_trans); |
d43317dc | 159 | goto loop; |
87533c47 | 160 | } else if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { |
e4b50e14 | 161 | spin_unlock(&fs_info->trans_lock); |
7b8b92af JB |
162 | kmem_cache_free(btrfs_transaction_cachep, cur_trans); |
163 | return -EROFS; | |
79154b1b | 164 | } |
d43317dc | 165 | |
a4abeea4 | 166 | atomic_set(&cur_trans->num_writers, 1); |
0860adfd | 167 | extwriter_counter_init(cur_trans, type); |
a4abeea4 JB |
168 | init_waitqueue_head(&cur_trans->writer_wait); |
169 | init_waitqueue_head(&cur_trans->commit_wait); | |
4a9d8bde | 170 | cur_trans->state = TRANS_STATE_RUNNING; |
a4abeea4 JB |
171 | /* |
172 | * One for this trans handle, one so it will live on until we | |
173 | * commit the transaction. | |
174 | */ | |
175 | atomic_set(&cur_trans->use_count, 2); | |
a4abeea4 JB |
176 | cur_trans->start_time = get_seconds(); |
177 | ||
178 | cur_trans->delayed_refs.root = RB_ROOT; | |
179 | cur_trans->delayed_refs.num_entries = 0; | |
180 | cur_trans->delayed_refs.num_heads_ready = 0; | |
181 | cur_trans->delayed_refs.num_heads = 0; | |
182 | cur_trans->delayed_refs.flushing = 0; | |
183 | cur_trans->delayed_refs.run_delayed_start = 0; | |
20b297d6 JS |
184 | |
185 | /* | |
186 | * although the tree mod log is per file system and not per transaction, | |
187 | * the log must never go across transaction boundaries. | |
188 | */ | |
189 | smp_mb(); | |
31b1a2bd JL |
190 | if (!list_empty(&fs_info->tree_mod_seq_list)) |
191 | WARN(1, KERN_ERR "btrfs: tree_mod_seq_list not empty when " | |
20b297d6 | 192 | "creating a fresh transaction\n"); |
31b1a2bd JL |
193 | if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log)) |
194 | WARN(1, KERN_ERR "btrfs: tree_mod_log rb tree not empty when " | |
20b297d6 | 195 | "creating a fresh transaction\n"); |
fc36ed7e | 196 | atomic64_set(&fs_info->tree_mod_seq, 0); |
20b297d6 | 197 | |
a4abeea4 | 198 | spin_lock_init(&cur_trans->delayed_refs.lock); |
bb721703 CM |
199 | atomic_set(&cur_trans->delayed_refs.procs_running_refs, 0); |
200 | atomic_set(&cur_trans->delayed_refs.ref_seq, 0); | |
201 | init_waitqueue_head(&cur_trans->delayed_refs.wait); | |
a4abeea4 JB |
202 | |
203 | INIT_LIST_HEAD(&cur_trans->pending_snapshots); | |
569e0f35 | 204 | INIT_LIST_HEAD(&cur_trans->ordered_operations); |
19ae4e81 | 205 | list_add_tail(&cur_trans->list, &fs_info->trans_list); |
a4abeea4 | 206 | extent_io_tree_init(&cur_trans->dirty_pages, |
19ae4e81 JS |
207 | fs_info->btree_inode->i_mapping); |
208 | fs_info->generation++; | |
209 | cur_trans->transid = fs_info->generation; | |
210 | fs_info->running_transaction = cur_trans; | |
49b25e05 | 211 | cur_trans->aborted = 0; |
19ae4e81 | 212 | spin_unlock(&fs_info->trans_lock); |
15ee9bc7 | 213 | |
79154b1b CM |
214 | return 0; |
215 | } | |
216 | ||
d352ac68 | 217 | /* |
d397712b CM |
218 | * this does all the record keeping required to make sure that a reference |
219 | * counted root is properly recorded in a given transaction. This is required | |
220 | * to make sure the old root from before we joined the transaction is deleted | |
221 | * when the transaction commits | |
d352ac68 | 222 | */ |
7585717f | 223 | static int record_root_in_trans(struct btrfs_trans_handle *trans, |
a4abeea4 | 224 | struct btrfs_root *root) |
6702ed49 | 225 | { |
5d4f98a2 | 226 | if (root->ref_cows && root->last_trans < trans->transid) { |
6702ed49 | 227 | WARN_ON(root == root->fs_info->extent_root); |
5d4f98a2 YZ |
228 | WARN_ON(root->commit_root != root->node); |
229 | ||
7585717f CM |
230 | /* |
231 | * see below for in_trans_setup usage rules | |
232 | * we have the reloc mutex held now, so there | |
233 | * is only one writer in this function | |
234 | */ | |
235 | root->in_trans_setup = 1; | |
236 | ||
237 | /* make sure readers find in_trans_setup before | |
238 | * they find our root->last_trans update | |
239 | */ | |
240 | smp_wmb(); | |
241 | ||
a4abeea4 JB |
242 | spin_lock(&root->fs_info->fs_roots_radix_lock); |
243 | if (root->last_trans == trans->transid) { | |
244 | spin_unlock(&root->fs_info->fs_roots_radix_lock); | |
245 | return 0; | |
246 | } | |
5d4f98a2 YZ |
247 | radix_tree_tag_set(&root->fs_info->fs_roots_radix, |
248 | (unsigned long)root->root_key.objectid, | |
249 | BTRFS_ROOT_TRANS_TAG); | |
a4abeea4 | 250 | spin_unlock(&root->fs_info->fs_roots_radix_lock); |
7585717f CM |
251 | root->last_trans = trans->transid; |
252 | ||
253 | /* this is pretty tricky. We don't want to | |
254 | * take the relocation lock in btrfs_record_root_in_trans | |
255 | * unless we're really doing the first setup for this root in | |
256 | * this transaction. | |
257 | * | |
258 | * Normally we'd use root->last_trans as a flag to decide | |
259 | * if we want to take the expensive mutex. | |
260 | * | |
261 | * But, we have to set root->last_trans before we | |
262 | * init the relocation root, otherwise, we trip over warnings | |
263 | * in ctree.c. The solution used here is to flag ourselves | |
264 | * with root->in_trans_setup. When this is 1, we're still | |
265 | * fixing up the reloc trees and everyone must wait. | |
266 | * | |
267 | * When this is zero, they can trust root->last_trans and fly | |
268 | * through btrfs_record_root_in_trans without having to take the | |
269 | * lock. smp_wmb() makes sure that all the writes above are | |
270 | * done before we pop in the zero below | |
271 | */ | |
5d4f98a2 | 272 | btrfs_init_reloc_root(trans, root); |
7585717f CM |
273 | smp_wmb(); |
274 | root->in_trans_setup = 0; | |
5d4f98a2 YZ |
275 | } |
276 | return 0; | |
277 | } | |
bcc63abb | 278 | |
7585717f CM |
279 | |
280 | int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, | |
281 | struct btrfs_root *root) | |
282 | { | |
283 | if (!root->ref_cows) | |
284 | return 0; | |
285 | ||
286 | /* | |
287 | * see record_root_in_trans for comments about in_trans_setup usage | |
288 | * and barriers | |
289 | */ | |
290 | smp_rmb(); | |
291 | if (root->last_trans == trans->transid && | |
292 | !root->in_trans_setup) | |
293 | return 0; | |
294 | ||
295 | mutex_lock(&root->fs_info->reloc_mutex); | |
296 | record_root_in_trans(trans, root); | |
297 | mutex_unlock(&root->fs_info->reloc_mutex); | |
298 | ||
299 | return 0; | |
300 | } | |
301 | ||
4a9d8bde MX |
302 | static inline int is_transaction_blocked(struct btrfs_transaction *trans) |
303 | { | |
304 | return (trans->state >= TRANS_STATE_BLOCKED && | |
305 | trans->state < TRANS_STATE_UNBLOCKED); | |
306 | } | |
307 | ||
d352ac68 CM |
308 | /* wait for commit against the current transaction to become unblocked |
309 | * when this is done, it is safe to start a new transaction, but the current | |
310 | * transaction might not be fully on disk. | |
311 | */ | |
37d1aeee | 312 | static void wait_current_trans(struct btrfs_root *root) |
79154b1b | 313 | { |
f9295749 | 314 | struct btrfs_transaction *cur_trans; |
79154b1b | 315 | |
a4abeea4 | 316 | spin_lock(&root->fs_info->trans_lock); |
f9295749 | 317 | cur_trans = root->fs_info->running_transaction; |
4a9d8bde | 318 | if (cur_trans && is_transaction_blocked(cur_trans)) { |
13c5a93e | 319 | atomic_inc(&cur_trans->use_count); |
a4abeea4 | 320 | spin_unlock(&root->fs_info->trans_lock); |
72d63ed6 LZ |
321 | |
322 | wait_event(root->fs_info->transaction_wait, | |
4a9d8bde | 323 | cur_trans->state >= TRANS_STATE_UNBLOCKED); |
f9295749 | 324 | put_transaction(cur_trans); |
a4abeea4 JB |
325 | } else { |
326 | spin_unlock(&root->fs_info->trans_lock); | |
f9295749 | 327 | } |
37d1aeee CM |
328 | } |
329 | ||
a22285a6 YZ |
330 | static int may_wait_transaction(struct btrfs_root *root, int type) |
331 | { | |
a4abeea4 JB |
332 | if (root->fs_info->log_root_recovering) |
333 | return 0; | |
334 | ||
335 | if (type == TRANS_USERSPACE) | |
336 | return 1; | |
337 | ||
338 | if (type == TRANS_START && | |
339 | !atomic_read(&root->fs_info->open_ioctl_trans)) | |
a22285a6 | 340 | return 1; |
a4abeea4 | 341 | |
a22285a6 YZ |
342 | return 0; |
343 | } | |
344 | ||
08e007d2 | 345 | static struct btrfs_trans_handle * |
0860adfd | 346 | start_transaction(struct btrfs_root *root, u64 num_items, unsigned int type, |
08e007d2 | 347 | enum btrfs_reserve_flush_enum flush) |
37d1aeee | 348 | { |
a22285a6 YZ |
349 | struct btrfs_trans_handle *h; |
350 | struct btrfs_transaction *cur_trans; | |
b5009945 | 351 | u64 num_bytes = 0; |
37d1aeee | 352 | int ret; |
c5567237 | 353 | u64 qgroup_reserved = 0; |
acce952b | 354 | |
87533c47 | 355 | if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) |
acce952b | 356 | return ERR_PTR(-EROFS); |
2a1eb461 JB |
357 | |
358 | if (current->journal_info) { | |
0860adfd | 359 | WARN_ON(type & TRANS_EXTWRITERS); |
2a1eb461 JB |
360 | h = current->journal_info; |
361 | h->use_count++; | |
b7d5b0a8 | 362 | WARN_ON(h->use_count > 2); |
2a1eb461 JB |
363 | h->orig_rsv = h->block_rsv; |
364 | h->block_rsv = NULL; | |
365 | goto got_it; | |
366 | } | |
b5009945 JB |
367 | |
368 | /* | |
369 | * Do the reservation before we join the transaction so we can do all | |
370 | * the appropriate flushing if need be. | |
371 | */ | |
372 | if (num_items > 0 && root != root->fs_info->chunk_root) { | |
c5567237 AJ |
373 | if (root->fs_info->quota_enabled && |
374 | is_fstree(root->root_key.objectid)) { | |
375 | qgroup_reserved = num_items * root->leafsize; | |
376 | ret = btrfs_qgroup_reserve(root, qgroup_reserved); | |
377 | if (ret) | |
378 | return ERR_PTR(ret); | |
379 | } | |
380 | ||
b5009945 | 381 | num_bytes = btrfs_calc_trans_metadata_size(root, num_items); |
08e007d2 MX |
382 | ret = btrfs_block_rsv_add(root, |
383 | &root->fs_info->trans_block_rsv, | |
384 | num_bytes, flush); | |
b5009945 | 385 | if (ret) |
843fcf35 | 386 | goto reserve_fail; |
b5009945 | 387 | } |
a22285a6 YZ |
388 | again: |
389 | h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS); | |
843fcf35 MX |
390 | if (!h) { |
391 | ret = -ENOMEM; | |
392 | goto alloc_fail; | |
393 | } | |
37d1aeee | 394 | |
98114659 JB |
395 | /* |
396 | * If we are JOIN_NOLOCK we're already committing a transaction and | |
397 | * waiting on this guy, so we don't need to do the sb_start_intwrite | |
398 | * because we're already holding a ref. We need this because we could | |
399 | * have raced in and did an fsync() on a file which can kick a commit | |
400 | * and then we deadlock with somebody doing a freeze. | |
354aa0fb MX |
401 | * |
402 | * If we are ATTACH, it means we just want to catch the current | |
403 | * transaction and commit it, so we needn't do sb_start_intwrite(). | |
98114659 | 404 | */ |
0860adfd | 405 | if (type & __TRANS_FREEZABLE) |
60376ce4 | 406 | sb_start_intwrite(root->fs_info->sb); |
b2b5ef5c | 407 | |
a22285a6 | 408 | if (may_wait_transaction(root, type)) |
37d1aeee | 409 | wait_current_trans(root); |
a22285a6 | 410 | |
a4abeea4 | 411 | do { |
354aa0fb | 412 | ret = join_transaction(root, type); |
178260b2 | 413 | if (ret == -EBUSY) { |
a4abeea4 | 414 | wait_current_trans(root); |
178260b2 MX |
415 | if (unlikely(type == TRANS_ATTACH)) |
416 | ret = -ENOENT; | |
417 | } | |
a4abeea4 JB |
418 | } while (ret == -EBUSY); |
419 | ||
db5b493a | 420 | if (ret < 0) { |
354aa0fb MX |
421 | /* We must get the transaction if we are JOIN_NOLOCK. */ |
422 | BUG_ON(type == TRANS_JOIN_NOLOCK); | |
843fcf35 | 423 | goto join_fail; |
db5b493a | 424 | } |
0f7d52f4 | 425 | |
a22285a6 | 426 | cur_trans = root->fs_info->running_transaction; |
a22285a6 YZ |
427 | |
428 | h->transid = cur_trans->transid; | |
429 | h->transaction = cur_trans; | |
79154b1b | 430 | h->blocks_used = 0; |
a22285a6 | 431 | h->bytes_reserved = 0; |
d13603ef | 432 | h->root = root; |
56bec294 | 433 | h->delayed_ref_updates = 0; |
2a1eb461 | 434 | h->use_count = 1; |
0e721106 | 435 | h->adding_csums = 0; |
f0486c68 | 436 | h->block_rsv = NULL; |
2a1eb461 | 437 | h->orig_rsv = NULL; |
49b25e05 | 438 | h->aborted = 0; |
4b824906 | 439 | h->qgroup_reserved = 0; |
bed92eae | 440 | h->delayed_ref_elem.seq = 0; |
a698d075 | 441 | h->type = type; |
c6b305a8 | 442 | h->allocating_chunk = false; |
bed92eae | 443 | INIT_LIST_HEAD(&h->qgroup_ref_list); |
ea658bad | 444 | INIT_LIST_HEAD(&h->new_bgs); |
b7ec40d7 | 445 | |
a22285a6 | 446 | smp_mb(); |
4a9d8bde MX |
447 | if (cur_trans->state >= TRANS_STATE_BLOCKED && |
448 | may_wait_transaction(root, type)) { | |
a22285a6 YZ |
449 | btrfs_commit_transaction(h, root); |
450 | goto again; | |
451 | } | |
452 | ||
b5009945 | 453 | if (num_bytes) { |
8c2a3ca2 | 454 | trace_btrfs_space_reservation(root->fs_info, "transaction", |
2bcc0328 | 455 | h->transid, num_bytes, 1); |
b5009945 JB |
456 | h->block_rsv = &root->fs_info->trans_block_rsv; |
457 | h->bytes_reserved = num_bytes; | |
a22285a6 | 458 | } |
4b824906 | 459 | h->qgroup_reserved = qgroup_reserved; |
9ed74f2d | 460 | |
2a1eb461 | 461 | got_it: |
a4abeea4 | 462 | btrfs_record_root_in_trans(h, root); |
a22285a6 YZ |
463 | |
464 | if (!current->journal_info && type != TRANS_USERSPACE) | |
465 | current->journal_info = h; | |
79154b1b | 466 | return h; |
843fcf35 MX |
467 | |
468 | join_fail: | |
0860adfd | 469 | if (type & __TRANS_FREEZABLE) |
843fcf35 MX |
470 | sb_end_intwrite(root->fs_info->sb); |
471 | kmem_cache_free(btrfs_trans_handle_cachep, h); | |
472 | alloc_fail: | |
473 | if (num_bytes) | |
474 | btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv, | |
475 | num_bytes); | |
476 | reserve_fail: | |
477 | if (qgroup_reserved) | |
478 | btrfs_qgroup_free(root, qgroup_reserved); | |
479 | return ERR_PTR(ret); | |
79154b1b CM |
480 | } |
481 | ||
f9295749 | 482 | struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, |
a22285a6 | 483 | int num_items) |
f9295749 | 484 | { |
08e007d2 MX |
485 | return start_transaction(root, num_items, TRANS_START, |
486 | BTRFS_RESERVE_FLUSH_ALL); | |
f9295749 | 487 | } |
8407aa46 | 488 | |
08e007d2 | 489 | struct btrfs_trans_handle *btrfs_start_transaction_lflush( |
8407aa46 MX |
490 | struct btrfs_root *root, int num_items) |
491 | { | |
08e007d2 MX |
492 | return start_transaction(root, num_items, TRANS_START, |
493 | BTRFS_RESERVE_FLUSH_LIMIT); | |
8407aa46 MX |
494 | } |
495 | ||
7a7eaa40 | 496 | struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root) |
f9295749 | 497 | { |
8407aa46 | 498 | return start_transaction(root, 0, TRANS_JOIN, 0); |
f9295749 CM |
499 | } |
500 | ||
7a7eaa40 | 501 | struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root) |
0af3d00b | 502 | { |
8407aa46 | 503 | return start_transaction(root, 0, TRANS_JOIN_NOLOCK, 0); |
0af3d00b JB |
504 | } |
505 | ||
7a7eaa40 | 506 | struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root) |
9ca9ee09 | 507 | { |
8407aa46 | 508 | return start_transaction(root, 0, TRANS_USERSPACE, 0); |
9ca9ee09 SW |
509 | } |
510 | ||
d4edf39b MX |
511 | /* |
512 | * btrfs_attach_transaction() - catch the running transaction | |
513 | * | |
514 | * It is used when we want to commit the current the transaction, but | |
515 | * don't want to start a new one. | |
516 | * | |
517 | * Note: If this function return -ENOENT, it just means there is no | |
518 | * running transaction. But it is possible that the inactive transaction | |
519 | * is still in the memory, not fully on disk. If you hope there is no | |
520 | * inactive transaction in the fs when -ENOENT is returned, you should | |
521 | * invoke | |
522 | * btrfs_attach_transaction_barrier() | |
523 | */ | |
354aa0fb | 524 | struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root) |
60376ce4 | 525 | { |
354aa0fb | 526 | return start_transaction(root, 0, TRANS_ATTACH, 0); |
60376ce4 JB |
527 | } |
528 | ||
d4edf39b MX |
529 | /* |
530 | * btrfs_attach_transaction() - catch the running transaction | |
531 | * | |
532 | * It is similar to the above function, the differentia is this one | |
533 | * will wait for all the inactive transactions until they fully | |
534 | * complete. | |
535 | */ | |
536 | struct btrfs_trans_handle * | |
537 | btrfs_attach_transaction_barrier(struct btrfs_root *root) | |
538 | { | |
539 | struct btrfs_trans_handle *trans; | |
540 | ||
541 | trans = start_transaction(root, 0, TRANS_ATTACH, 0); | |
542 | if (IS_ERR(trans) && PTR_ERR(trans) == -ENOENT) | |
543 | btrfs_wait_for_commit(root, 0); | |
544 | ||
545 | return trans; | |
546 | } | |
547 | ||
d352ac68 | 548 | /* wait for a transaction commit to be fully complete */ |
b9c8300c | 549 | static noinline void wait_for_commit(struct btrfs_root *root, |
89ce8a63 CM |
550 | struct btrfs_transaction *commit) |
551 | { | |
4a9d8bde | 552 | wait_event(commit->commit_wait, commit->state == TRANS_STATE_COMPLETED); |
89ce8a63 CM |
553 | } |
554 | ||
46204592 SW |
555 | int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid) |
556 | { | |
557 | struct btrfs_transaction *cur_trans = NULL, *t; | |
8cd2807f | 558 | int ret = 0; |
46204592 | 559 | |
46204592 SW |
560 | if (transid) { |
561 | if (transid <= root->fs_info->last_trans_committed) | |
a4abeea4 | 562 | goto out; |
46204592 | 563 | |
8cd2807f | 564 | ret = -EINVAL; |
46204592 | 565 | /* find specified transaction */ |
a4abeea4 | 566 | spin_lock(&root->fs_info->trans_lock); |
46204592 SW |
567 | list_for_each_entry(t, &root->fs_info->trans_list, list) { |
568 | if (t->transid == transid) { | |
569 | cur_trans = t; | |
a4abeea4 | 570 | atomic_inc(&cur_trans->use_count); |
8cd2807f | 571 | ret = 0; |
46204592 SW |
572 | break; |
573 | } | |
8cd2807f MX |
574 | if (t->transid > transid) { |
575 | ret = 0; | |
46204592 | 576 | break; |
8cd2807f | 577 | } |
46204592 | 578 | } |
a4abeea4 | 579 | spin_unlock(&root->fs_info->trans_lock); |
8cd2807f | 580 | /* The specified transaction doesn't exist */ |
46204592 | 581 | if (!cur_trans) |
8cd2807f | 582 | goto out; |
46204592 SW |
583 | } else { |
584 | /* find newest transaction that is committing | committed */ | |
a4abeea4 | 585 | spin_lock(&root->fs_info->trans_lock); |
46204592 SW |
586 | list_for_each_entry_reverse(t, &root->fs_info->trans_list, |
587 | list) { | |
4a9d8bde MX |
588 | if (t->state >= TRANS_STATE_COMMIT_START) { |
589 | if (t->state == TRANS_STATE_COMPLETED) | |
3473f3c0 | 590 | break; |
46204592 | 591 | cur_trans = t; |
a4abeea4 | 592 | atomic_inc(&cur_trans->use_count); |
46204592 SW |
593 | break; |
594 | } | |
595 | } | |
a4abeea4 | 596 | spin_unlock(&root->fs_info->trans_lock); |
46204592 | 597 | if (!cur_trans) |
a4abeea4 | 598 | goto out; /* nothing committing|committed */ |
46204592 SW |
599 | } |
600 | ||
46204592 | 601 | wait_for_commit(root, cur_trans); |
46204592 | 602 | put_transaction(cur_trans); |
a4abeea4 | 603 | out: |
46204592 SW |
604 | return ret; |
605 | } | |
606 | ||
37d1aeee CM |
607 | void btrfs_throttle(struct btrfs_root *root) |
608 | { | |
a4abeea4 | 609 | if (!atomic_read(&root->fs_info->open_ioctl_trans)) |
9ca9ee09 | 610 | wait_current_trans(root); |
37d1aeee CM |
611 | } |
612 | ||
8929ecfa YZ |
613 | static int should_end_transaction(struct btrfs_trans_handle *trans, |
614 | struct btrfs_root *root) | |
615 | { | |
616 | int ret; | |
36ba022a JB |
617 | |
618 | ret = btrfs_block_rsv_check(root, &root->fs_info->global_block_rsv, 5); | |
8929ecfa YZ |
619 | return ret ? 1 : 0; |
620 | } | |
621 | ||
622 | int btrfs_should_end_transaction(struct btrfs_trans_handle *trans, | |
623 | struct btrfs_root *root) | |
624 | { | |
625 | struct btrfs_transaction *cur_trans = trans->transaction; | |
626 | int updates; | |
49b25e05 | 627 | int err; |
8929ecfa | 628 | |
a4abeea4 | 629 | smp_mb(); |
4a9d8bde MX |
630 | if (cur_trans->state >= TRANS_STATE_BLOCKED || |
631 | cur_trans->delayed_refs.flushing) | |
8929ecfa YZ |
632 | return 1; |
633 | ||
634 | updates = trans->delayed_ref_updates; | |
635 | trans->delayed_ref_updates = 0; | |
49b25e05 JM |
636 | if (updates) { |
637 | err = btrfs_run_delayed_refs(trans, root, updates); | |
638 | if (err) /* Error code will also eval true */ | |
639 | return err; | |
640 | } | |
8929ecfa YZ |
641 | |
642 | return should_end_transaction(trans, root); | |
643 | } | |
644 | ||
89ce8a63 | 645 | static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, |
a698d075 | 646 | struct btrfs_root *root, int throttle) |
79154b1b | 647 | { |
8929ecfa | 648 | struct btrfs_transaction *cur_trans = trans->transaction; |
ab78c84d | 649 | struct btrfs_fs_info *info = root->fs_info; |
c3e69d58 | 650 | int count = 0; |
a698d075 | 651 | int lock = (trans->type != TRANS_JOIN_NOLOCK); |
4edc2ca3 | 652 | int err = 0; |
c3e69d58 | 653 | |
2a1eb461 JB |
654 | if (--trans->use_count) { |
655 | trans->block_rsv = trans->orig_rsv; | |
656 | return 0; | |
657 | } | |
658 | ||
edf39272 JS |
659 | /* |
660 | * do the qgroup accounting as early as possible | |
661 | */ | |
662 | err = btrfs_delayed_refs_qgroup_accounting(trans, info); | |
663 | ||
b24e03db | 664 | btrfs_trans_release_metadata(trans, root); |
4c13d758 | 665 | trans->block_rsv = NULL; |
c5567237 AJ |
666 | |
667 | if (trans->qgroup_reserved) { | |
7c2ec3f0 LB |
668 | /* |
669 | * the same root has to be passed here between start_transaction | |
670 | * and end_transaction. Subvolume quota depends on this. | |
671 | */ | |
672 | btrfs_qgroup_free(trans->root, trans->qgroup_reserved); | |
c5567237 AJ |
673 | trans->qgroup_reserved = 0; |
674 | } | |
675 | ||
ea658bad JB |
676 | if (!list_empty(&trans->new_bgs)) |
677 | btrfs_create_pending_block_groups(trans, root); | |
678 | ||
bb721703 | 679 | while (count < 1) { |
c3e69d58 CM |
680 | unsigned long cur = trans->delayed_ref_updates; |
681 | trans->delayed_ref_updates = 0; | |
682 | if (cur && | |
683 | trans->transaction->delayed_refs.num_heads_ready > 64) { | |
684 | trans->delayed_ref_updates = 0; | |
685 | btrfs_run_delayed_refs(trans, root, cur); | |
686 | } else { | |
687 | break; | |
688 | } | |
689 | count++; | |
56bec294 | 690 | } |
bb721703 | 691 | |
0e721106 JB |
692 | btrfs_trans_release_metadata(trans, root); |
693 | trans->block_rsv = NULL; | |
56bec294 | 694 | |
ea658bad JB |
695 | if (!list_empty(&trans->new_bgs)) |
696 | btrfs_create_pending_block_groups(trans, root); | |
697 | ||
a4abeea4 | 698 | if (lock && !atomic_read(&root->fs_info->open_ioctl_trans) && |
4a9d8bde MX |
699 | should_end_transaction(trans, root) && |
700 | ACCESS_ONCE(cur_trans->state) == TRANS_STATE_RUNNING) { | |
701 | spin_lock(&info->trans_lock); | |
702 | if (cur_trans->state == TRANS_STATE_RUNNING) | |
703 | cur_trans->state = TRANS_STATE_BLOCKED; | |
704 | spin_unlock(&info->trans_lock); | |
a4abeea4 | 705 | } |
8929ecfa | 706 | |
4a9d8bde | 707 | if (lock && ACCESS_ONCE(cur_trans->state) == TRANS_STATE_BLOCKED) { |
81317fde JB |
708 | if (throttle) { |
709 | /* | |
710 | * We may race with somebody else here so end up having | |
711 | * to call end_transaction on ourselves again, so inc | |
712 | * our use_count. | |
713 | */ | |
714 | trans->use_count++; | |
8929ecfa | 715 | return btrfs_commit_transaction(trans, root); |
81317fde | 716 | } else { |
8929ecfa | 717 | wake_up_process(info->transaction_kthread); |
81317fde | 718 | } |
8929ecfa YZ |
719 | } |
720 | ||
0860adfd | 721 | if (trans->type & __TRANS_FREEZABLE) |
98114659 | 722 | sb_end_intwrite(root->fs_info->sb); |
6df7881a | 723 | |
8929ecfa | 724 | WARN_ON(cur_trans != info->running_transaction); |
13c5a93e JB |
725 | WARN_ON(atomic_read(&cur_trans->num_writers) < 1); |
726 | atomic_dec(&cur_trans->num_writers); | |
0860adfd | 727 | extwriter_counter_dec(cur_trans, trans->type); |
89ce8a63 | 728 | |
99d16cbc | 729 | smp_mb(); |
79154b1b CM |
730 | if (waitqueue_active(&cur_trans->writer_wait)) |
731 | wake_up(&cur_trans->writer_wait); | |
79154b1b | 732 | put_transaction(cur_trans); |
9ed74f2d JB |
733 | |
734 | if (current->journal_info == trans) | |
735 | current->journal_info = NULL; | |
ab78c84d | 736 | |
24bbcf04 YZ |
737 | if (throttle) |
738 | btrfs_run_delayed_iputs(root); | |
739 | ||
49b25e05 | 740 | if (trans->aborted || |
87533c47 | 741 | test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) |
4edc2ca3 | 742 | err = -EIO; |
edf39272 | 743 | assert_qgroups_uptodate(trans); |
49b25e05 | 744 | |
4edc2ca3 DJ |
745 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
746 | return err; | |
79154b1b CM |
747 | } |
748 | ||
89ce8a63 CM |
749 | int btrfs_end_transaction(struct btrfs_trans_handle *trans, |
750 | struct btrfs_root *root) | |
751 | { | |
98ad43be | 752 | return __btrfs_end_transaction(trans, root, 0); |
89ce8a63 CM |
753 | } |
754 | ||
755 | int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans, | |
756 | struct btrfs_root *root) | |
757 | { | |
98ad43be | 758 | return __btrfs_end_transaction(trans, root, 1); |
16cdcec7 MX |
759 | } |
760 | ||
761 | int btrfs_end_transaction_dmeta(struct btrfs_trans_handle *trans, | |
762 | struct btrfs_root *root) | |
763 | { | |
a698d075 | 764 | return __btrfs_end_transaction(trans, root, 1); |
89ce8a63 CM |
765 | } |
766 | ||
d352ac68 CM |
767 | /* |
768 | * when btree blocks are allocated, they have some corresponding bits set for | |
769 | * them in one of two extent_io trees. This is used to make sure all of | |
690587d1 | 770 | * those extents are sent to disk but does not wait on them |
d352ac68 | 771 | */ |
690587d1 | 772 | int btrfs_write_marked_extents(struct btrfs_root *root, |
8cef4e16 | 773 | struct extent_io_tree *dirty_pages, int mark) |
79154b1b | 774 | { |
777e6bd7 | 775 | int err = 0; |
7c4452b9 | 776 | int werr = 0; |
1728366e | 777 | struct address_space *mapping = root->fs_info->btree_inode->i_mapping; |
e6138876 | 778 | struct extent_state *cached_state = NULL; |
777e6bd7 | 779 | u64 start = 0; |
5f39d397 | 780 | u64 end; |
7c4452b9 | 781 | |
1728366e | 782 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
e6138876 JB |
783 | mark, &cached_state)) { |
784 | convert_extent_bit(dirty_pages, start, end, EXTENT_NEED_WAIT, | |
785 | mark, &cached_state, GFP_NOFS); | |
786 | cached_state = NULL; | |
1728366e JB |
787 | err = filemap_fdatawrite_range(mapping, start, end); |
788 | if (err) | |
789 | werr = err; | |
790 | cond_resched(); | |
791 | start = end + 1; | |
7c4452b9 | 792 | } |
690587d1 CM |
793 | if (err) |
794 | werr = err; | |
795 | return werr; | |
796 | } | |
797 | ||
798 | /* | |
799 | * when btree blocks are allocated, they have some corresponding bits set for | |
800 | * them in one of two extent_io trees. This is used to make sure all of | |
801 | * those extents are on disk for transaction or log commit. We wait | |
802 | * on all the pages and clear them from the dirty pages state tree | |
803 | */ | |
804 | int btrfs_wait_marked_extents(struct btrfs_root *root, | |
8cef4e16 | 805 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 | 806 | { |
690587d1 CM |
807 | int err = 0; |
808 | int werr = 0; | |
1728366e | 809 | struct address_space *mapping = root->fs_info->btree_inode->i_mapping; |
e6138876 | 810 | struct extent_state *cached_state = NULL; |
690587d1 CM |
811 | u64 start = 0; |
812 | u64 end; | |
777e6bd7 | 813 | |
1728366e | 814 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
e6138876 JB |
815 | EXTENT_NEED_WAIT, &cached_state)) { |
816 | clear_extent_bit(dirty_pages, start, end, EXTENT_NEED_WAIT, | |
817 | 0, 0, &cached_state, GFP_NOFS); | |
1728366e JB |
818 | err = filemap_fdatawait_range(mapping, start, end); |
819 | if (err) | |
820 | werr = err; | |
821 | cond_resched(); | |
822 | start = end + 1; | |
777e6bd7 | 823 | } |
7c4452b9 CM |
824 | if (err) |
825 | werr = err; | |
826 | return werr; | |
79154b1b CM |
827 | } |
828 | ||
690587d1 CM |
829 | /* |
830 | * when btree blocks are allocated, they have some corresponding bits set for | |
831 | * them in one of two extent_io trees. This is used to make sure all of | |
832 | * those extents are on disk for transaction or log commit | |
833 | */ | |
834 | int btrfs_write_and_wait_marked_extents(struct btrfs_root *root, | |
8cef4e16 | 835 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 CM |
836 | { |
837 | int ret; | |
838 | int ret2; | |
c6adc9cc | 839 | struct blk_plug plug; |
690587d1 | 840 | |
c6adc9cc | 841 | blk_start_plug(&plug); |
8cef4e16 | 842 | ret = btrfs_write_marked_extents(root, dirty_pages, mark); |
c6adc9cc | 843 | blk_finish_plug(&plug); |
8cef4e16 | 844 | ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark); |
bf0da8c1 CM |
845 | |
846 | if (ret) | |
847 | return ret; | |
848 | if (ret2) | |
849 | return ret2; | |
850 | return 0; | |
690587d1 CM |
851 | } |
852 | ||
d0c803c4 CM |
853 | int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans, |
854 | struct btrfs_root *root) | |
855 | { | |
856 | if (!trans || !trans->transaction) { | |
857 | struct inode *btree_inode; | |
858 | btree_inode = root->fs_info->btree_inode; | |
859 | return filemap_write_and_wait(btree_inode->i_mapping); | |
860 | } | |
861 | return btrfs_write_and_wait_marked_extents(root, | |
8cef4e16 YZ |
862 | &trans->transaction->dirty_pages, |
863 | EXTENT_DIRTY); | |
d0c803c4 CM |
864 | } |
865 | ||
d352ac68 CM |
866 | /* |
867 | * this is used to update the root pointer in the tree of tree roots. | |
868 | * | |
869 | * But, in the case of the extent allocation tree, updating the root | |
870 | * pointer may allocate blocks which may change the root of the extent | |
871 | * allocation tree. | |
872 | * | |
873 | * So, this loops and repeats and makes sure the cowonly root didn't | |
874 | * change while the root pointer was being updated in the metadata. | |
875 | */ | |
0b86a832 CM |
876 | static int update_cowonly_root(struct btrfs_trans_handle *trans, |
877 | struct btrfs_root *root) | |
79154b1b CM |
878 | { |
879 | int ret; | |
0b86a832 | 880 | u64 old_root_bytenr; |
86b9f2ec | 881 | u64 old_root_used; |
0b86a832 | 882 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
79154b1b | 883 | |
86b9f2ec | 884 | old_root_used = btrfs_root_used(&root->root_item); |
0b86a832 | 885 | btrfs_write_dirty_block_groups(trans, root); |
56bec294 | 886 | |
d397712b | 887 | while (1) { |
0b86a832 | 888 | old_root_bytenr = btrfs_root_bytenr(&root->root_item); |
86b9f2ec YZ |
889 | if (old_root_bytenr == root->node->start && |
890 | old_root_used == btrfs_root_used(&root->root_item)) | |
79154b1b | 891 | break; |
87ef2bb4 | 892 | |
5d4f98a2 | 893 | btrfs_set_root_node(&root->root_item, root->node); |
79154b1b | 894 | ret = btrfs_update_root(trans, tree_root, |
0b86a832 CM |
895 | &root->root_key, |
896 | &root->root_item); | |
49b25e05 JM |
897 | if (ret) |
898 | return ret; | |
56bec294 | 899 | |
86b9f2ec | 900 | old_root_used = btrfs_root_used(&root->root_item); |
4a8c9a62 | 901 | ret = btrfs_write_dirty_block_groups(trans, root); |
49b25e05 JM |
902 | if (ret) |
903 | return ret; | |
0b86a832 | 904 | } |
276e680d YZ |
905 | |
906 | if (root != root->fs_info->extent_root) | |
907 | switch_commit_root(root); | |
908 | ||
0b86a832 CM |
909 | return 0; |
910 | } | |
911 | ||
d352ac68 CM |
912 | /* |
913 | * update all the cowonly tree roots on disk | |
49b25e05 JM |
914 | * |
915 | * The error handling in this function may not be obvious. Any of the | |
916 | * failures will cause the file system to go offline. We still need | |
917 | * to clean up the delayed refs. | |
d352ac68 | 918 | */ |
5d4f98a2 YZ |
919 | static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans, |
920 | struct btrfs_root *root) | |
0b86a832 CM |
921 | { |
922 | struct btrfs_fs_info *fs_info = root->fs_info; | |
923 | struct list_head *next; | |
84234f3a | 924 | struct extent_buffer *eb; |
56bec294 | 925 | int ret; |
84234f3a | 926 | |
56bec294 | 927 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
49b25e05 JM |
928 | if (ret) |
929 | return ret; | |
87ef2bb4 | 930 | |
84234f3a | 931 | eb = btrfs_lock_root_node(fs_info->tree_root); |
49b25e05 JM |
932 | ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, |
933 | 0, &eb); | |
84234f3a YZ |
934 | btrfs_tree_unlock(eb); |
935 | free_extent_buffer(eb); | |
0b86a832 | 936 | |
49b25e05 JM |
937 | if (ret) |
938 | return ret; | |
939 | ||
56bec294 | 940 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
49b25e05 JM |
941 | if (ret) |
942 | return ret; | |
87ef2bb4 | 943 | |
733f4fbb | 944 | ret = btrfs_run_dev_stats(trans, root->fs_info); |
8dabb742 SB |
945 | WARN_ON(ret); |
946 | ret = btrfs_run_dev_replace(trans, root->fs_info); | |
947 | WARN_ON(ret); | |
733f4fbb | 948 | |
546adb0d JS |
949 | ret = btrfs_run_qgroups(trans, root->fs_info); |
950 | BUG_ON(ret); | |
951 | ||
952 | /* run_qgroups might have added some more refs */ | |
953 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); | |
954 | BUG_ON(ret); | |
955 | ||
d397712b | 956 | while (!list_empty(&fs_info->dirty_cowonly_roots)) { |
0b86a832 CM |
957 | next = fs_info->dirty_cowonly_roots.next; |
958 | list_del_init(next); | |
959 | root = list_entry(next, struct btrfs_root, dirty_list); | |
87ef2bb4 | 960 | |
49b25e05 JM |
961 | ret = update_cowonly_root(trans, root); |
962 | if (ret) | |
963 | return ret; | |
79154b1b | 964 | } |
276e680d YZ |
965 | |
966 | down_write(&fs_info->extent_commit_sem); | |
967 | switch_commit_root(fs_info->extent_root); | |
968 | up_write(&fs_info->extent_commit_sem); | |
969 | ||
8dabb742 SB |
970 | btrfs_after_dev_replace_commit(fs_info); |
971 | ||
79154b1b CM |
972 | return 0; |
973 | } | |
974 | ||
d352ac68 CM |
975 | /* |
976 | * dead roots are old snapshots that need to be deleted. This allocates | |
977 | * a dirty root struct and adds it into the list of dead roots that need to | |
978 | * be deleted | |
979 | */ | |
5d4f98a2 | 980 | int btrfs_add_dead_root(struct btrfs_root *root) |
5eda7b5e | 981 | { |
a4abeea4 | 982 | spin_lock(&root->fs_info->trans_lock); |
9d1a2a3a | 983 | list_add_tail(&root->root_list, &root->fs_info->dead_roots); |
a4abeea4 | 984 | spin_unlock(&root->fs_info->trans_lock); |
5eda7b5e CM |
985 | return 0; |
986 | } | |
987 | ||
d352ac68 | 988 | /* |
5d4f98a2 | 989 | * update all the cowonly tree roots on disk |
d352ac68 | 990 | */ |
5d4f98a2 YZ |
991 | static noinline int commit_fs_roots(struct btrfs_trans_handle *trans, |
992 | struct btrfs_root *root) | |
0f7d52f4 | 993 | { |
0f7d52f4 | 994 | struct btrfs_root *gang[8]; |
5d4f98a2 | 995 | struct btrfs_fs_info *fs_info = root->fs_info; |
0f7d52f4 CM |
996 | int i; |
997 | int ret; | |
54aa1f4d CM |
998 | int err = 0; |
999 | ||
a4abeea4 | 1000 | spin_lock(&fs_info->fs_roots_radix_lock); |
d397712b | 1001 | while (1) { |
5d4f98a2 YZ |
1002 | ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, |
1003 | (void **)gang, 0, | |
0f7d52f4 CM |
1004 | ARRAY_SIZE(gang), |
1005 | BTRFS_ROOT_TRANS_TAG); | |
1006 | if (ret == 0) | |
1007 | break; | |
1008 | for (i = 0; i < ret; i++) { | |
1009 | root = gang[i]; | |
5d4f98a2 YZ |
1010 | radix_tree_tag_clear(&fs_info->fs_roots_radix, |
1011 | (unsigned long)root->root_key.objectid, | |
1012 | BTRFS_ROOT_TRANS_TAG); | |
a4abeea4 | 1013 | spin_unlock(&fs_info->fs_roots_radix_lock); |
31153d81 | 1014 | |
e02119d5 | 1015 | btrfs_free_log(trans, root); |
5d4f98a2 | 1016 | btrfs_update_reloc_root(trans, root); |
d68fc57b | 1017 | btrfs_orphan_commit_root(trans, root); |
bcc63abb | 1018 | |
82d5902d LZ |
1019 | btrfs_save_ino_cache(root, trans); |
1020 | ||
f1ebcc74 LB |
1021 | /* see comments in should_cow_block() */ |
1022 | root->force_cow = 0; | |
1023 | smp_wmb(); | |
1024 | ||
978d910d | 1025 | if (root->commit_root != root->node) { |
581bb050 | 1026 | mutex_lock(&root->fs_commit_mutex); |
817d52f8 | 1027 | switch_commit_root(root); |
581bb050 LZ |
1028 | btrfs_unpin_free_ino(root); |
1029 | mutex_unlock(&root->fs_commit_mutex); | |
1030 | ||
978d910d YZ |
1031 | btrfs_set_root_node(&root->root_item, |
1032 | root->node); | |
1033 | } | |
5d4f98a2 | 1034 | |
5d4f98a2 | 1035 | err = btrfs_update_root(trans, fs_info->tree_root, |
0f7d52f4 CM |
1036 | &root->root_key, |
1037 | &root->root_item); | |
a4abeea4 | 1038 | spin_lock(&fs_info->fs_roots_radix_lock); |
54aa1f4d CM |
1039 | if (err) |
1040 | break; | |
0f7d52f4 CM |
1041 | } |
1042 | } | |
a4abeea4 | 1043 | spin_unlock(&fs_info->fs_roots_radix_lock); |
54aa1f4d | 1044 | return err; |
0f7d52f4 CM |
1045 | } |
1046 | ||
d352ac68 | 1047 | /* |
de78b51a ES |
1048 | * defrag a given btree. |
1049 | * Every leaf in the btree is read and defragged. | |
d352ac68 | 1050 | */ |
de78b51a | 1051 | int btrfs_defrag_root(struct btrfs_root *root) |
e9d0b13b CM |
1052 | { |
1053 | struct btrfs_fs_info *info = root->fs_info; | |
e9d0b13b | 1054 | struct btrfs_trans_handle *trans; |
8929ecfa | 1055 | int ret; |
e9d0b13b | 1056 | |
8929ecfa | 1057 | if (xchg(&root->defrag_running, 1)) |
e9d0b13b | 1058 | return 0; |
8929ecfa | 1059 | |
6b80053d | 1060 | while (1) { |
8929ecfa YZ |
1061 | trans = btrfs_start_transaction(root, 0); |
1062 | if (IS_ERR(trans)) | |
1063 | return PTR_ERR(trans); | |
1064 | ||
de78b51a | 1065 | ret = btrfs_defrag_leaves(trans, root); |
8929ecfa | 1066 | |
e9d0b13b | 1067 | btrfs_end_transaction(trans, root); |
b53d3f5d | 1068 | btrfs_btree_balance_dirty(info->tree_root); |
e9d0b13b CM |
1069 | cond_resched(); |
1070 | ||
7841cb28 | 1071 | if (btrfs_fs_closing(root->fs_info) || ret != -EAGAIN) |
e9d0b13b | 1072 | break; |
210549eb DS |
1073 | |
1074 | if (btrfs_defrag_cancelled(root->fs_info)) { | |
1075 | printk(KERN_DEBUG "btrfs: defrag_root cancelled\n"); | |
1076 | ret = -EAGAIN; | |
1077 | break; | |
1078 | } | |
e9d0b13b CM |
1079 | } |
1080 | root->defrag_running = 0; | |
8929ecfa | 1081 | return ret; |
e9d0b13b CM |
1082 | } |
1083 | ||
d352ac68 CM |
1084 | /* |
1085 | * new snapshots need to be created at a very specific time in the | |
aec8030a MX |
1086 | * transaction commit. This does the actual creation. |
1087 | * | |
1088 | * Note: | |
1089 | * If the error which may affect the commitment of the current transaction | |
1090 | * happens, we should return the error number. If the error which just affect | |
1091 | * the creation of the pending snapshots, just return 0. | |
d352ac68 | 1092 | */ |
80b6794d | 1093 | static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, |
3063d29f CM |
1094 | struct btrfs_fs_info *fs_info, |
1095 | struct btrfs_pending_snapshot *pending) | |
1096 | { | |
1097 | struct btrfs_key key; | |
80b6794d | 1098 | struct btrfs_root_item *new_root_item; |
3063d29f CM |
1099 | struct btrfs_root *tree_root = fs_info->tree_root; |
1100 | struct btrfs_root *root = pending->root; | |
6bdb72de | 1101 | struct btrfs_root *parent_root; |
98c9942a | 1102 | struct btrfs_block_rsv *rsv; |
6bdb72de | 1103 | struct inode *parent_inode; |
42874b3d MX |
1104 | struct btrfs_path *path; |
1105 | struct btrfs_dir_item *dir_item; | |
a22285a6 | 1106 | struct dentry *dentry; |
3063d29f | 1107 | struct extent_buffer *tmp; |
925baedd | 1108 | struct extent_buffer *old; |
8ea05e3a | 1109 | struct timespec cur_time = CURRENT_TIME; |
aec8030a | 1110 | int ret = 0; |
d68fc57b | 1111 | u64 to_reserve = 0; |
6bdb72de | 1112 | u64 index = 0; |
a22285a6 | 1113 | u64 objectid; |
b83cc969 | 1114 | u64 root_flags; |
8ea05e3a | 1115 | uuid_le new_uuid; |
3063d29f | 1116 | |
42874b3d MX |
1117 | path = btrfs_alloc_path(); |
1118 | if (!path) { | |
aec8030a MX |
1119 | pending->error = -ENOMEM; |
1120 | return 0; | |
42874b3d MX |
1121 | } |
1122 | ||
80b6794d CM |
1123 | new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS); |
1124 | if (!new_root_item) { | |
aec8030a | 1125 | pending->error = -ENOMEM; |
6fa9700e | 1126 | goto root_item_alloc_fail; |
80b6794d | 1127 | } |
a22285a6 | 1128 | |
aec8030a MX |
1129 | pending->error = btrfs_find_free_objectid(tree_root, &objectid); |
1130 | if (pending->error) | |
6fa9700e | 1131 | goto no_free_objectid; |
3063d29f | 1132 | |
3fd0a558 | 1133 | btrfs_reloc_pre_snapshot(trans, pending, &to_reserve); |
d68fc57b YZ |
1134 | |
1135 | if (to_reserve > 0) { | |
aec8030a MX |
1136 | pending->error = btrfs_block_rsv_add(root, |
1137 | &pending->block_rsv, | |
1138 | to_reserve, | |
1139 | BTRFS_RESERVE_NO_FLUSH); | |
1140 | if (pending->error) | |
6fa9700e | 1141 | goto no_free_objectid; |
d68fc57b YZ |
1142 | } |
1143 | ||
aec8030a MX |
1144 | pending->error = btrfs_qgroup_inherit(trans, fs_info, |
1145 | root->root_key.objectid, | |
1146 | objectid, pending->inherit); | |
1147 | if (pending->error) | |
6fa9700e | 1148 | goto no_free_objectid; |
6f72c7e2 | 1149 | |
3063d29f | 1150 | key.objectid = objectid; |
a22285a6 YZ |
1151 | key.offset = (u64)-1; |
1152 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3063d29f | 1153 | |
6fa9700e | 1154 | rsv = trans->block_rsv; |
a22285a6 | 1155 | trans->block_rsv = &pending->block_rsv; |
2382c5cc | 1156 | trans->bytes_reserved = trans->block_rsv->reserved; |
3de4586c | 1157 | |
a22285a6 | 1158 | dentry = pending->dentry; |
e9662f70 | 1159 | parent_inode = pending->dir; |
a22285a6 | 1160 | parent_root = BTRFS_I(parent_inode)->root; |
7585717f | 1161 | record_root_in_trans(trans, parent_root); |
a22285a6 | 1162 | |
3063d29f CM |
1163 | /* |
1164 | * insert the directory item | |
1165 | */ | |
3de4586c | 1166 | ret = btrfs_set_inode_index(parent_inode, &index); |
49b25e05 | 1167 | BUG_ON(ret); /* -ENOMEM */ |
42874b3d MX |
1168 | |
1169 | /* check if there is a file/dir which has the same name. */ | |
1170 | dir_item = btrfs_lookup_dir_item(NULL, parent_root, path, | |
1171 | btrfs_ino(parent_inode), | |
1172 | dentry->d_name.name, | |
1173 | dentry->d_name.len, 0); | |
1174 | if (dir_item != NULL && !IS_ERR(dir_item)) { | |
fe66a05a | 1175 | pending->error = -EEXIST; |
aec8030a | 1176 | goto dir_item_existed; |
42874b3d MX |
1177 | } else if (IS_ERR(dir_item)) { |
1178 | ret = PTR_ERR(dir_item); | |
8732d44f MX |
1179 | btrfs_abort_transaction(trans, root, ret); |
1180 | goto fail; | |
79787eaa | 1181 | } |
42874b3d | 1182 | btrfs_release_path(path); |
52c26179 | 1183 | |
e999376f CM |
1184 | /* |
1185 | * pull in the delayed directory update | |
1186 | * and the delayed inode item | |
1187 | * otherwise we corrupt the FS during | |
1188 | * snapshot | |
1189 | */ | |
1190 | ret = btrfs_run_delayed_items(trans, root); | |
8732d44f MX |
1191 | if (ret) { /* Transaction aborted */ |
1192 | btrfs_abort_transaction(trans, root, ret); | |
1193 | goto fail; | |
1194 | } | |
e999376f | 1195 | |
7585717f | 1196 | record_root_in_trans(trans, root); |
6bdb72de SW |
1197 | btrfs_set_root_last_snapshot(&root->root_item, trans->transid); |
1198 | memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); | |
08fe4db1 | 1199 | btrfs_check_and_init_root_item(new_root_item); |
6bdb72de | 1200 | |
b83cc969 LZ |
1201 | root_flags = btrfs_root_flags(new_root_item); |
1202 | if (pending->readonly) | |
1203 | root_flags |= BTRFS_ROOT_SUBVOL_RDONLY; | |
1204 | else | |
1205 | root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY; | |
1206 | btrfs_set_root_flags(new_root_item, root_flags); | |
1207 | ||
8ea05e3a AB |
1208 | btrfs_set_root_generation_v2(new_root_item, |
1209 | trans->transid); | |
1210 | uuid_le_gen(&new_uuid); | |
1211 | memcpy(new_root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE); | |
1212 | memcpy(new_root_item->parent_uuid, root->root_item.uuid, | |
1213 | BTRFS_UUID_SIZE); | |
70023da2 SB |
1214 | if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) { |
1215 | memset(new_root_item->received_uuid, 0, | |
1216 | sizeof(new_root_item->received_uuid)); | |
1217 | memset(&new_root_item->stime, 0, sizeof(new_root_item->stime)); | |
1218 | memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime)); | |
1219 | btrfs_set_root_stransid(new_root_item, 0); | |
1220 | btrfs_set_root_rtransid(new_root_item, 0); | |
1221 | } | |
8ea05e3a | 1222 | new_root_item->otime.sec = cpu_to_le64(cur_time.tv_sec); |
dadd1105 | 1223 | new_root_item->otime.nsec = cpu_to_le32(cur_time.tv_nsec); |
8ea05e3a | 1224 | btrfs_set_root_otransid(new_root_item, trans->transid); |
8ea05e3a | 1225 | |
6bdb72de | 1226 | old = btrfs_lock_root_node(root); |
49b25e05 | 1227 | ret = btrfs_cow_block(trans, root, old, NULL, 0, &old); |
79787eaa JM |
1228 | if (ret) { |
1229 | btrfs_tree_unlock(old); | |
1230 | free_extent_buffer(old); | |
8732d44f MX |
1231 | btrfs_abort_transaction(trans, root, ret); |
1232 | goto fail; | |
79787eaa | 1233 | } |
49b25e05 | 1234 | |
6bdb72de SW |
1235 | btrfs_set_lock_blocking(old); |
1236 | ||
49b25e05 | 1237 | ret = btrfs_copy_root(trans, root, old, &tmp, objectid); |
79787eaa | 1238 | /* clean up in any case */ |
6bdb72de SW |
1239 | btrfs_tree_unlock(old); |
1240 | free_extent_buffer(old); | |
8732d44f MX |
1241 | if (ret) { |
1242 | btrfs_abort_transaction(trans, root, ret); | |
1243 | goto fail; | |
1244 | } | |
6bdb72de | 1245 | |
f1ebcc74 LB |
1246 | /* see comments in should_cow_block() */ |
1247 | root->force_cow = 1; | |
1248 | smp_wmb(); | |
1249 | ||
6bdb72de | 1250 | btrfs_set_root_node(new_root_item, tmp); |
a22285a6 YZ |
1251 | /* record when the snapshot was created in key.offset */ |
1252 | key.offset = trans->transid; | |
1253 | ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); | |
6bdb72de SW |
1254 | btrfs_tree_unlock(tmp); |
1255 | free_extent_buffer(tmp); | |
8732d44f MX |
1256 | if (ret) { |
1257 | btrfs_abort_transaction(trans, root, ret); | |
1258 | goto fail; | |
1259 | } | |
6bdb72de | 1260 | |
a22285a6 YZ |
1261 | /* |
1262 | * insert root back/forward references | |
1263 | */ | |
1264 | ret = btrfs_add_root_ref(trans, tree_root, objectid, | |
0660b5af | 1265 | parent_root->root_key.objectid, |
33345d01 | 1266 | btrfs_ino(parent_inode), index, |
a22285a6 | 1267 | dentry->d_name.name, dentry->d_name.len); |
8732d44f MX |
1268 | if (ret) { |
1269 | btrfs_abort_transaction(trans, root, ret); | |
1270 | goto fail; | |
1271 | } | |
0660b5af | 1272 | |
a22285a6 YZ |
1273 | key.offset = (u64)-1; |
1274 | pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key); | |
79787eaa JM |
1275 | if (IS_ERR(pending->snap)) { |
1276 | ret = PTR_ERR(pending->snap); | |
8732d44f MX |
1277 | btrfs_abort_transaction(trans, root, ret); |
1278 | goto fail; | |
79787eaa | 1279 | } |
d68fc57b | 1280 | |
49b25e05 | 1281 | ret = btrfs_reloc_post_snapshot(trans, pending); |
8732d44f MX |
1282 | if (ret) { |
1283 | btrfs_abort_transaction(trans, root, ret); | |
1284 | goto fail; | |
1285 | } | |
361048f5 MX |
1286 | |
1287 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); | |
8732d44f MX |
1288 | if (ret) { |
1289 | btrfs_abort_transaction(trans, root, ret); | |
1290 | goto fail; | |
1291 | } | |
42874b3d MX |
1292 | |
1293 | ret = btrfs_insert_dir_item(trans, parent_root, | |
1294 | dentry->d_name.name, dentry->d_name.len, | |
1295 | parent_inode, &key, | |
1296 | BTRFS_FT_DIR, index); | |
1297 | /* We have check then name at the beginning, so it is impossible. */ | |
9c52057c | 1298 | BUG_ON(ret == -EEXIST || ret == -EOVERFLOW); |
8732d44f MX |
1299 | if (ret) { |
1300 | btrfs_abort_transaction(trans, root, ret); | |
1301 | goto fail; | |
1302 | } | |
42874b3d MX |
1303 | |
1304 | btrfs_i_size_write(parent_inode, parent_inode->i_size + | |
1305 | dentry->d_name.len * 2); | |
1306 | parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME; | |
be6aef60 | 1307 | ret = btrfs_update_inode_fallback(trans, parent_root, parent_inode); |
42874b3d | 1308 | if (ret) |
8732d44f | 1309 | btrfs_abort_transaction(trans, root, ret); |
3063d29f | 1310 | fail: |
aec8030a MX |
1311 | pending->error = ret; |
1312 | dir_item_existed: | |
98c9942a | 1313 | trans->block_rsv = rsv; |
2382c5cc | 1314 | trans->bytes_reserved = 0; |
6fa9700e MX |
1315 | no_free_objectid: |
1316 | kfree(new_root_item); | |
1317 | root_item_alloc_fail: | |
42874b3d | 1318 | btrfs_free_path(path); |
49b25e05 | 1319 | return ret; |
3063d29f CM |
1320 | } |
1321 | ||
d352ac68 CM |
1322 | /* |
1323 | * create all the snapshots we've scheduled for creation | |
1324 | */ | |
80b6794d CM |
1325 | static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans, |
1326 | struct btrfs_fs_info *fs_info) | |
3de4586c | 1327 | { |
aec8030a | 1328 | struct btrfs_pending_snapshot *pending, *next; |
3de4586c | 1329 | struct list_head *head = &trans->transaction->pending_snapshots; |
aec8030a | 1330 | int ret = 0; |
3de4586c | 1331 | |
aec8030a MX |
1332 | list_for_each_entry_safe(pending, next, head, list) { |
1333 | list_del(&pending->list); | |
1334 | ret = create_pending_snapshot(trans, fs_info, pending); | |
1335 | if (ret) | |
1336 | break; | |
1337 | } | |
1338 | return ret; | |
3de4586c CM |
1339 | } |
1340 | ||
5d4f98a2 YZ |
1341 | static void update_super_roots(struct btrfs_root *root) |
1342 | { | |
1343 | struct btrfs_root_item *root_item; | |
1344 | struct btrfs_super_block *super; | |
1345 | ||
6c41761f | 1346 | super = root->fs_info->super_copy; |
5d4f98a2 YZ |
1347 | |
1348 | root_item = &root->fs_info->chunk_root->root_item; | |
1349 | super->chunk_root = root_item->bytenr; | |
1350 | super->chunk_root_generation = root_item->generation; | |
1351 | super->chunk_root_level = root_item->level; | |
1352 | ||
1353 | root_item = &root->fs_info->tree_root->root_item; | |
1354 | super->root = root_item->bytenr; | |
1355 | super->generation = root_item->generation; | |
1356 | super->root_level = root_item->level; | |
73bc1876 | 1357 | if (btrfs_test_opt(root, SPACE_CACHE)) |
0af3d00b | 1358 | super->cache_generation = root_item->generation; |
5d4f98a2 YZ |
1359 | } |
1360 | ||
f36f3042 CM |
1361 | int btrfs_transaction_in_commit(struct btrfs_fs_info *info) |
1362 | { | |
4a9d8bde | 1363 | struct btrfs_transaction *trans; |
f36f3042 | 1364 | int ret = 0; |
4a9d8bde | 1365 | |
a4abeea4 | 1366 | spin_lock(&info->trans_lock); |
4a9d8bde MX |
1367 | trans = info->running_transaction; |
1368 | if (trans) | |
1369 | ret = (trans->state >= TRANS_STATE_COMMIT_START); | |
a4abeea4 | 1370 | spin_unlock(&info->trans_lock); |
f36f3042 CM |
1371 | return ret; |
1372 | } | |
1373 | ||
8929ecfa YZ |
1374 | int btrfs_transaction_blocked(struct btrfs_fs_info *info) |
1375 | { | |
4a9d8bde | 1376 | struct btrfs_transaction *trans; |
8929ecfa | 1377 | int ret = 0; |
4a9d8bde | 1378 | |
a4abeea4 | 1379 | spin_lock(&info->trans_lock); |
4a9d8bde MX |
1380 | trans = info->running_transaction; |
1381 | if (trans) | |
1382 | ret = is_transaction_blocked(trans); | |
a4abeea4 | 1383 | spin_unlock(&info->trans_lock); |
8929ecfa YZ |
1384 | return ret; |
1385 | } | |
1386 | ||
bb9c12c9 SW |
1387 | /* |
1388 | * wait for the current transaction commit to start and block subsequent | |
1389 | * transaction joins | |
1390 | */ | |
1391 | static void wait_current_trans_commit_start(struct btrfs_root *root, | |
1392 | struct btrfs_transaction *trans) | |
1393 | { | |
4a9d8bde MX |
1394 | wait_event(root->fs_info->transaction_blocked_wait, |
1395 | trans->state >= TRANS_STATE_COMMIT_START); | |
bb9c12c9 SW |
1396 | } |
1397 | ||
1398 | /* | |
1399 | * wait for the current transaction to start and then become unblocked. | |
1400 | * caller holds ref. | |
1401 | */ | |
1402 | static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root, | |
1403 | struct btrfs_transaction *trans) | |
1404 | { | |
72d63ed6 | 1405 | wait_event(root->fs_info->transaction_wait, |
4a9d8bde | 1406 | trans->state >= TRANS_STATE_UNBLOCKED); |
bb9c12c9 SW |
1407 | } |
1408 | ||
1409 | /* | |
1410 | * commit transactions asynchronously. once btrfs_commit_transaction_async | |
1411 | * returns, any subsequent transaction will not be allowed to join. | |
1412 | */ | |
1413 | struct btrfs_async_commit { | |
1414 | struct btrfs_trans_handle *newtrans; | |
1415 | struct btrfs_root *root; | |
7892b5af | 1416 | struct work_struct work; |
bb9c12c9 SW |
1417 | }; |
1418 | ||
1419 | static void do_async_commit(struct work_struct *work) | |
1420 | { | |
1421 | struct btrfs_async_commit *ac = | |
7892b5af | 1422 | container_of(work, struct btrfs_async_commit, work); |
bb9c12c9 | 1423 | |
6fc4e354 SW |
1424 | /* |
1425 | * We've got freeze protection passed with the transaction. | |
1426 | * Tell lockdep about it. | |
1427 | */ | |
ff7c1d33 MX |
1428 | if (ac->newtrans->type < TRANS_JOIN_NOLOCK) |
1429 | rwsem_acquire_read( | |
1430 | &ac->root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1], | |
1431 | 0, 1, _THIS_IP_); | |
6fc4e354 | 1432 | |
e209db7a SW |
1433 | current->journal_info = ac->newtrans; |
1434 | ||
bb9c12c9 SW |
1435 | btrfs_commit_transaction(ac->newtrans, ac->root); |
1436 | kfree(ac); | |
1437 | } | |
1438 | ||
1439 | int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans, | |
1440 | struct btrfs_root *root, | |
1441 | int wait_for_unblock) | |
1442 | { | |
1443 | struct btrfs_async_commit *ac; | |
1444 | struct btrfs_transaction *cur_trans; | |
1445 | ||
1446 | ac = kmalloc(sizeof(*ac), GFP_NOFS); | |
db5b493a TI |
1447 | if (!ac) |
1448 | return -ENOMEM; | |
bb9c12c9 | 1449 | |
7892b5af | 1450 | INIT_WORK(&ac->work, do_async_commit); |
bb9c12c9 | 1451 | ac->root = root; |
7a7eaa40 | 1452 | ac->newtrans = btrfs_join_transaction(root); |
3612b495 TI |
1453 | if (IS_ERR(ac->newtrans)) { |
1454 | int err = PTR_ERR(ac->newtrans); | |
1455 | kfree(ac); | |
1456 | return err; | |
1457 | } | |
bb9c12c9 SW |
1458 | |
1459 | /* take transaction reference */ | |
bb9c12c9 | 1460 | cur_trans = trans->transaction; |
13c5a93e | 1461 | atomic_inc(&cur_trans->use_count); |
bb9c12c9 SW |
1462 | |
1463 | btrfs_end_transaction(trans, root); | |
6fc4e354 SW |
1464 | |
1465 | /* | |
1466 | * Tell lockdep we've released the freeze rwsem, since the | |
1467 | * async commit thread will be the one to unlock it. | |
1468 | */ | |
ff7c1d33 MX |
1469 | if (trans->type < TRANS_JOIN_NOLOCK) |
1470 | rwsem_release( | |
1471 | &root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1], | |
1472 | 1, _THIS_IP_); | |
6fc4e354 | 1473 | |
7892b5af | 1474 | schedule_work(&ac->work); |
bb9c12c9 SW |
1475 | |
1476 | /* wait for transaction to start and unblock */ | |
bb9c12c9 SW |
1477 | if (wait_for_unblock) |
1478 | wait_current_trans_commit_start_and_unblock(root, cur_trans); | |
1479 | else | |
1480 | wait_current_trans_commit_start(root, cur_trans); | |
bb9c12c9 | 1481 | |
38e88054 SW |
1482 | if (current->journal_info == trans) |
1483 | current->journal_info = NULL; | |
1484 | ||
1485 | put_transaction(cur_trans); | |
bb9c12c9 SW |
1486 | return 0; |
1487 | } | |
1488 | ||
49b25e05 JM |
1489 | |
1490 | static void cleanup_transaction(struct btrfs_trans_handle *trans, | |
7b8b92af | 1491 | struct btrfs_root *root, int err) |
49b25e05 JM |
1492 | { |
1493 | struct btrfs_transaction *cur_trans = trans->transaction; | |
f094ac32 | 1494 | DEFINE_WAIT(wait); |
49b25e05 JM |
1495 | |
1496 | WARN_ON(trans->use_count > 1); | |
1497 | ||
7b8b92af JB |
1498 | btrfs_abort_transaction(trans, root, err); |
1499 | ||
49b25e05 | 1500 | spin_lock(&root->fs_info->trans_lock); |
66b6135b | 1501 | |
25d8c284 MX |
1502 | /* |
1503 | * If the transaction is removed from the list, it means this | |
1504 | * transaction has been committed successfully, so it is impossible | |
1505 | * to call the cleanup function. | |
1506 | */ | |
1507 | BUG_ON(list_empty(&cur_trans->list)); | |
66b6135b | 1508 | |
49b25e05 | 1509 | list_del_init(&cur_trans->list); |
d7096fc3 | 1510 | if (cur_trans == root->fs_info->running_transaction) { |
4a9d8bde | 1511 | cur_trans->state = TRANS_STATE_COMMIT_DOING; |
f094ac32 LB |
1512 | spin_unlock(&root->fs_info->trans_lock); |
1513 | wait_event(cur_trans->writer_wait, | |
1514 | atomic_read(&cur_trans->num_writers) == 1); | |
1515 | ||
1516 | spin_lock(&root->fs_info->trans_lock); | |
d7096fc3 | 1517 | } |
49b25e05 JM |
1518 | spin_unlock(&root->fs_info->trans_lock); |
1519 | ||
1520 | btrfs_cleanup_one_transaction(trans->transaction, root); | |
1521 | ||
4a9d8bde MX |
1522 | spin_lock(&root->fs_info->trans_lock); |
1523 | if (cur_trans == root->fs_info->running_transaction) | |
1524 | root->fs_info->running_transaction = NULL; | |
1525 | spin_unlock(&root->fs_info->trans_lock); | |
1526 | ||
49b25e05 JM |
1527 | put_transaction(cur_trans); |
1528 | put_transaction(cur_trans); | |
1529 | ||
1530 | trace_btrfs_transaction_commit(root); | |
1531 | ||
1532 | btrfs_scrub_continue(root); | |
1533 | ||
1534 | if (current->journal_info == trans) | |
1535 | current->journal_info = NULL; | |
1536 | ||
1537 | kmem_cache_free(btrfs_trans_handle_cachep, trans); | |
1538 | } | |
1539 | ||
ca469637 MX |
1540 | static int btrfs_flush_all_pending_stuffs(struct btrfs_trans_handle *trans, |
1541 | struct btrfs_root *root) | |
1542 | { | |
ca469637 MX |
1543 | int ret; |
1544 | ||
ca469637 MX |
1545 | ret = btrfs_run_delayed_items(trans, root); |
1546 | if (ret) | |
1547 | return ret; | |
1548 | ||
1549 | /* | |
1550 | * running the delayed items may have added new refs. account | |
1551 | * them now so that they hinder processing of more delayed refs | |
1552 | * as little as possible. | |
1553 | */ | |
1554 | btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info); | |
1555 | ||
1556 | /* | |
1557 | * rename don't use btrfs_join_transaction, so, once we | |
1558 | * set the transaction to blocked above, we aren't going | |
1559 | * to get any new ordered operations. We can safely run | |
1560 | * it here and no for sure that nothing new will be added | |
1561 | * to the list | |
1562 | */ | |
569e0f35 | 1563 | ret = btrfs_run_ordered_operations(trans, root, 1); |
ca469637 | 1564 | |
eebc6084 | 1565 | return ret; |
ca469637 MX |
1566 | } |
1567 | ||
82436617 MX |
1568 | static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info) |
1569 | { | |
1570 | if (btrfs_test_opt(fs_info->tree_root, FLUSHONCOMMIT)) | |
1571 | return btrfs_start_all_delalloc_inodes(fs_info, 1); | |
1572 | return 0; | |
1573 | } | |
1574 | ||
1575 | static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info) | |
1576 | { | |
1577 | if (btrfs_test_opt(fs_info->tree_root, FLUSHONCOMMIT)) | |
1578 | btrfs_wait_all_ordered_extents(fs_info, 1); | |
1579 | } | |
1580 | ||
79154b1b CM |
1581 | int btrfs_commit_transaction(struct btrfs_trans_handle *trans, |
1582 | struct btrfs_root *root) | |
1583 | { | |
49b25e05 | 1584 | struct btrfs_transaction *cur_trans = trans->transaction; |
8fd17795 | 1585 | struct btrfs_transaction *prev_trans = NULL; |
25287e0a | 1586 | int ret; |
79154b1b | 1587 | |
569e0f35 | 1588 | ret = btrfs_run_ordered_operations(trans, root, 0); |
25287e0a MX |
1589 | if (ret) { |
1590 | btrfs_abort_transaction(trans, root, ret); | |
e4a2bcac JB |
1591 | btrfs_end_transaction(trans, root); |
1592 | return ret; | |
25287e0a | 1593 | } |
5a3f23d5 | 1594 | |
8d25a086 MX |
1595 | /* Stop the commit early if ->aborted is set */ |
1596 | if (unlikely(ACCESS_ONCE(cur_trans->aborted))) { | |
25287e0a | 1597 | ret = cur_trans->aborted; |
e4a2bcac JB |
1598 | btrfs_end_transaction(trans, root); |
1599 | return ret; | |
25287e0a | 1600 | } |
49b25e05 | 1601 | |
56bec294 CM |
1602 | /* make a pass through all the delayed refs we have so far |
1603 | * any runnings procs may add more while we are here | |
1604 | */ | |
1605 | ret = btrfs_run_delayed_refs(trans, root, 0); | |
e4a2bcac JB |
1606 | if (ret) { |
1607 | btrfs_end_transaction(trans, root); | |
1608 | return ret; | |
1609 | } | |
56bec294 | 1610 | |
0e721106 JB |
1611 | btrfs_trans_release_metadata(trans, root); |
1612 | trans->block_rsv = NULL; | |
272d26d0 MX |
1613 | if (trans->qgroup_reserved) { |
1614 | btrfs_qgroup_free(root, trans->qgroup_reserved); | |
1615 | trans->qgroup_reserved = 0; | |
1616 | } | |
0e721106 | 1617 | |
b7ec40d7 | 1618 | cur_trans = trans->transaction; |
49b25e05 | 1619 | |
56bec294 CM |
1620 | /* |
1621 | * set the flushing flag so procs in this transaction have to | |
1622 | * start sending their work down. | |
1623 | */ | |
b7ec40d7 | 1624 | cur_trans->delayed_refs.flushing = 1; |
56bec294 | 1625 | |
ea658bad JB |
1626 | if (!list_empty(&trans->new_bgs)) |
1627 | btrfs_create_pending_block_groups(trans, root); | |
1628 | ||
c3e69d58 | 1629 | ret = btrfs_run_delayed_refs(trans, root, 0); |
e4a2bcac JB |
1630 | if (ret) { |
1631 | btrfs_end_transaction(trans, root); | |
1632 | return ret; | |
1633 | } | |
56bec294 | 1634 | |
4a9d8bde MX |
1635 | spin_lock(&root->fs_info->trans_lock); |
1636 | if (cur_trans->state >= TRANS_STATE_COMMIT_START) { | |
1637 | spin_unlock(&root->fs_info->trans_lock); | |
13c5a93e | 1638 | atomic_inc(&cur_trans->use_count); |
49b25e05 | 1639 | ret = btrfs_end_transaction(trans, root); |
ccd467d6 | 1640 | |
b9c8300c | 1641 | wait_for_commit(root, cur_trans); |
15ee9bc7 | 1642 | |
79154b1b | 1643 | put_transaction(cur_trans); |
15ee9bc7 | 1644 | |
49b25e05 | 1645 | return ret; |
79154b1b | 1646 | } |
4313b399 | 1647 | |
4a9d8bde | 1648 | cur_trans->state = TRANS_STATE_COMMIT_START; |
bb9c12c9 SW |
1649 | wake_up(&root->fs_info->transaction_blocked_wait); |
1650 | ||
ccd467d6 CM |
1651 | if (cur_trans->list.prev != &root->fs_info->trans_list) { |
1652 | prev_trans = list_entry(cur_trans->list.prev, | |
1653 | struct btrfs_transaction, list); | |
4a9d8bde | 1654 | if (prev_trans->state != TRANS_STATE_COMPLETED) { |
13c5a93e | 1655 | atomic_inc(&prev_trans->use_count); |
a4abeea4 | 1656 | spin_unlock(&root->fs_info->trans_lock); |
ccd467d6 CM |
1657 | |
1658 | wait_for_commit(root, prev_trans); | |
ccd467d6 | 1659 | |
15ee9bc7 | 1660 | put_transaction(prev_trans); |
a4abeea4 JB |
1661 | } else { |
1662 | spin_unlock(&root->fs_info->trans_lock); | |
ccd467d6 | 1663 | } |
a4abeea4 JB |
1664 | } else { |
1665 | spin_unlock(&root->fs_info->trans_lock); | |
ccd467d6 | 1666 | } |
15ee9bc7 | 1667 | |
0860adfd MX |
1668 | extwriter_counter_dec(cur_trans, trans->type); |
1669 | ||
82436617 MX |
1670 | ret = btrfs_start_delalloc_flush(root->fs_info); |
1671 | if (ret) | |
1672 | goto cleanup_transaction; | |
1673 | ||
581227d0 MX |
1674 | ret = btrfs_flush_all_pending_stuffs(trans, root); |
1675 | if (ret) | |
1676 | goto cleanup_transaction; | |
15ee9bc7 | 1677 | |
581227d0 MX |
1678 | wait_event(cur_trans->writer_wait, |
1679 | extwriter_counter_read(cur_trans) == 0); | |
15ee9bc7 | 1680 | |
581227d0 | 1681 | /* some pending stuffs might be added after the previous flush. */ |
ca469637 MX |
1682 | ret = btrfs_flush_all_pending_stuffs(trans, root); |
1683 | if (ret) | |
1684 | goto cleanup_transaction; | |
1685 | ||
82436617 | 1686 | btrfs_wait_delalloc_flush(root->fs_info); |
ed0ca140 JB |
1687 | /* |
1688 | * Ok now we need to make sure to block out any other joins while we | |
1689 | * commit the transaction. We could have started a join before setting | |
4a9d8bde | 1690 | * COMMIT_DOING so make sure to wait for num_writers to == 1 again. |
ed0ca140 JB |
1691 | */ |
1692 | spin_lock(&root->fs_info->trans_lock); | |
4a9d8bde | 1693 | cur_trans->state = TRANS_STATE_COMMIT_DOING; |
ed0ca140 JB |
1694 | spin_unlock(&root->fs_info->trans_lock); |
1695 | wait_event(cur_trans->writer_wait, | |
1696 | atomic_read(&cur_trans->num_writers) == 1); | |
1697 | ||
2cba30f1 MX |
1698 | /* ->aborted might be set after the previous check, so check it */ |
1699 | if (unlikely(ACCESS_ONCE(cur_trans->aborted))) { | |
1700 | ret = cur_trans->aborted; | |
1701 | goto cleanup_transaction; | |
1702 | } | |
7585717f CM |
1703 | /* |
1704 | * the reloc mutex makes sure that we stop | |
1705 | * the balancing code from coming in and moving | |
1706 | * extents around in the middle of the commit | |
1707 | */ | |
1708 | mutex_lock(&root->fs_info->reloc_mutex); | |
1709 | ||
42874b3d MX |
1710 | /* |
1711 | * We needn't worry about the delayed items because we will | |
1712 | * deal with them in create_pending_snapshot(), which is the | |
1713 | * core function of the snapshot creation. | |
1714 | */ | |
1715 | ret = create_pending_snapshots(trans, root->fs_info); | |
49b25e05 JM |
1716 | if (ret) { |
1717 | mutex_unlock(&root->fs_info->reloc_mutex); | |
1718 | goto cleanup_transaction; | |
1719 | } | |
3063d29f | 1720 | |
42874b3d MX |
1721 | /* |
1722 | * We insert the dir indexes of the snapshots and update the inode | |
1723 | * of the snapshots' parents after the snapshot creation, so there | |
1724 | * are some delayed items which are not dealt with. Now deal with | |
1725 | * them. | |
1726 | * | |
1727 | * We needn't worry that this operation will corrupt the snapshots, | |
1728 | * because all the tree which are snapshoted will be forced to COW | |
1729 | * the nodes and leaves. | |
1730 | */ | |
1731 | ret = btrfs_run_delayed_items(trans, root); | |
49b25e05 JM |
1732 | if (ret) { |
1733 | mutex_unlock(&root->fs_info->reloc_mutex); | |
1734 | goto cleanup_transaction; | |
1735 | } | |
16cdcec7 | 1736 | |
56bec294 | 1737 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
49b25e05 JM |
1738 | if (ret) { |
1739 | mutex_unlock(&root->fs_info->reloc_mutex); | |
1740 | goto cleanup_transaction; | |
1741 | } | |
56bec294 | 1742 | |
e999376f CM |
1743 | /* |
1744 | * make sure none of the code above managed to slip in a | |
1745 | * delayed item | |
1746 | */ | |
1747 | btrfs_assert_delayed_root_empty(root); | |
1748 | ||
2c90e5d6 | 1749 | WARN_ON(cur_trans != trans->transaction); |
dc17ff8f | 1750 | |
a2de733c | 1751 | btrfs_scrub_pause(root); |
e02119d5 CM |
1752 | /* btrfs_commit_tree_roots is responsible for getting the |
1753 | * various roots consistent with each other. Every pointer | |
1754 | * in the tree of tree roots has to point to the most up to date | |
1755 | * root for every subvolume and other tree. So, we have to keep | |
1756 | * the tree logging code from jumping in and changing any | |
1757 | * of the trees. | |
1758 | * | |
1759 | * At this point in the commit, there can't be any tree-log | |
1760 | * writers, but a little lower down we drop the trans mutex | |
1761 | * and let new people in. By holding the tree_log_mutex | |
1762 | * from now until after the super is written, we avoid races | |
1763 | * with the tree-log code. | |
1764 | */ | |
1765 | mutex_lock(&root->fs_info->tree_log_mutex); | |
1766 | ||
5d4f98a2 | 1767 | ret = commit_fs_roots(trans, root); |
49b25e05 JM |
1768 | if (ret) { |
1769 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
871383be | 1770 | mutex_unlock(&root->fs_info->reloc_mutex); |
49b25e05 JM |
1771 | goto cleanup_transaction; |
1772 | } | |
54aa1f4d | 1773 | |
5d4f98a2 | 1774 | /* commit_fs_roots gets rid of all the tree log roots, it is now |
e02119d5 CM |
1775 | * safe to free the root of tree log roots |
1776 | */ | |
1777 | btrfs_free_log_root_tree(trans, root->fs_info); | |
1778 | ||
5d4f98a2 | 1779 | ret = commit_cowonly_roots(trans, root); |
49b25e05 JM |
1780 | if (ret) { |
1781 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
871383be | 1782 | mutex_unlock(&root->fs_info->reloc_mutex); |
49b25e05 JM |
1783 | goto cleanup_transaction; |
1784 | } | |
54aa1f4d | 1785 | |
2cba30f1 MX |
1786 | /* |
1787 | * The tasks which save the space cache and inode cache may also | |
1788 | * update ->aborted, check it. | |
1789 | */ | |
1790 | if (unlikely(ACCESS_ONCE(cur_trans->aborted))) { | |
1791 | ret = cur_trans->aborted; | |
1792 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
1793 | mutex_unlock(&root->fs_info->reloc_mutex); | |
1794 | goto cleanup_transaction; | |
1795 | } | |
1796 | ||
11833d66 YZ |
1797 | btrfs_prepare_extent_commit(trans, root); |
1798 | ||
78fae27e | 1799 | cur_trans = root->fs_info->running_transaction; |
5d4f98a2 YZ |
1800 | |
1801 | btrfs_set_root_node(&root->fs_info->tree_root->root_item, | |
1802 | root->fs_info->tree_root->node); | |
817d52f8 | 1803 | switch_commit_root(root->fs_info->tree_root); |
5d4f98a2 YZ |
1804 | |
1805 | btrfs_set_root_node(&root->fs_info->chunk_root->root_item, | |
1806 | root->fs_info->chunk_root->node); | |
817d52f8 | 1807 | switch_commit_root(root->fs_info->chunk_root); |
5d4f98a2 | 1808 | |
edf39272 | 1809 | assert_qgroups_uptodate(trans); |
5d4f98a2 | 1810 | update_super_roots(root); |
e02119d5 CM |
1811 | |
1812 | if (!root->fs_info->log_root_recovering) { | |
6c41761f DS |
1813 | btrfs_set_super_log_root(root->fs_info->super_copy, 0); |
1814 | btrfs_set_super_log_root_level(root->fs_info->super_copy, 0); | |
e02119d5 CM |
1815 | } |
1816 | ||
6c41761f DS |
1817 | memcpy(root->fs_info->super_for_commit, root->fs_info->super_copy, |
1818 | sizeof(*root->fs_info->super_copy)); | |
ccd467d6 | 1819 | |
a4abeea4 | 1820 | spin_lock(&root->fs_info->trans_lock); |
4a9d8bde | 1821 | cur_trans->state = TRANS_STATE_UNBLOCKED; |
a4abeea4 | 1822 | root->fs_info->running_transaction = NULL; |
a4abeea4 | 1823 | spin_unlock(&root->fs_info->trans_lock); |
7585717f | 1824 | mutex_unlock(&root->fs_info->reloc_mutex); |
b7ec40d7 | 1825 | |
f9295749 | 1826 | wake_up(&root->fs_info->transaction_wait); |
e6dcd2dc | 1827 | |
79154b1b | 1828 | ret = btrfs_write_and_wait_transaction(trans, root); |
49b25e05 JM |
1829 | if (ret) { |
1830 | btrfs_error(root->fs_info, ret, | |
08748810 | 1831 | "Error while writing out transaction"); |
49b25e05 JM |
1832 | mutex_unlock(&root->fs_info->tree_log_mutex); |
1833 | goto cleanup_transaction; | |
1834 | } | |
1835 | ||
1836 | ret = write_ctree_super(trans, root, 0); | |
1837 | if (ret) { | |
1838 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
1839 | goto cleanup_transaction; | |
1840 | } | |
4313b399 | 1841 | |
e02119d5 CM |
1842 | /* |
1843 | * the super is written, we can safely allow the tree-loggers | |
1844 | * to go about their business | |
1845 | */ | |
1846 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
1847 | ||
11833d66 | 1848 | btrfs_finish_extent_commit(trans, root); |
4313b399 | 1849 | |
15ee9bc7 | 1850 | root->fs_info->last_trans_committed = cur_trans->transid; |
4a9d8bde MX |
1851 | /* |
1852 | * We needn't acquire the lock here because there is no other task | |
1853 | * which can change it. | |
1854 | */ | |
1855 | cur_trans->state = TRANS_STATE_COMPLETED; | |
2c90e5d6 | 1856 | wake_up(&cur_trans->commit_wait); |
3de4586c | 1857 | |
a4abeea4 | 1858 | spin_lock(&root->fs_info->trans_lock); |
13c5a93e | 1859 | list_del_init(&cur_trans->list); |
a4abeea4 JB |
1860 | spin_unlock(&root->fs_info->trans_lock); |
1861 | ||
78fae27e | 1862 | put_transaction(cur_trans); |
79154b1b | 1863 | put_transaction(cur_trans); |
58176a96 | 1864 | |
0860adfd | 1865 | if (trans->type & __TRANS_FREEZABLE) |
354aa0fb | 1866 | sb_end_intwrite(root->fs_info->sb); |
b2b5ef5c | 1867 | |
1abe9b8a | 1868 | trace_btrfs_transaction_commit(root); |
1869 | ||
a2de733c AJ |
1870 | btrfs_scrub_continue(root); |
1871 | ||
9ed74f2d JB |
1872 | if (current->journal_info == trans) |
1873 | current->journal_info = NULL; | |
1874 | ||
2c90e5d6 | 1875 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
24bbcf04 YZ |
1876 | |
1877 | if (current != root->fs_info->transaction_kthread) | |
1878 | btrfs_run_delayed_iputs(root); | |
1879 | ||
79154b1b | 1880 | return ret; |
49b25e05 JM |
1881 | |
1882 | cleanup_transaction: | |
0e721106 JB |
1883 | btrfs_trans_release_metadata(trans, root); |
1884 | trans->block_rsv = NULL; | |
272d26d0 MX |
1885 | if (trans->qgroup_reserved) { |
1886 | btrfs_qgroup_free(root, trans->qgroup_reserved); | |
1887 | trans->qgroup_reserved = 0; | |
1888 | } | |
c2cf52eb | 1889 | btrfs_warn(root->fs_info, "Skipping commit of aborted transaction."); |
49b25e05 JM |
1890 | if (current->journal_info == trans) |
1891 | current->journal_info = NULL; | |
7b8b92af | 1892 | cleanup_transaction(trans, root, ret); |
49b25e05 JM |
1893 | |
1894 | return ret; | |
79154b1b CM |
1895 | } |
1896 | ||
d352ac68 | 1897 | /* |
9d1a2a3a DS |
1898 | * return < 0 if error |
1899 | * 0 if there are no more dead_roots at the time of call | |
1900 | * 1 there are more to be processed, call me again | |
1901 | * | |
1902 | * The return value indicates there are certainly more snapshots to delete, but | |
1903 | * if there comes a new one during processing, it may return 0. We don't mind, | |
1904 | * because btrfs_commit_super will poke cleaner thread and it will process it a | |
1905 | * few seconds later. | |
d352ac68 | 1906 | */ |
9d1a2a3a | 1907 | int btrfs_clean_one_deleted_snapshot(struct btrfs_root *root) |
e9d0b13b | 1908 | { |
9d1a2a3a | 1909 | int ret; |
5d4f98a2 YZ |
1910 | struct btrfs_fs_info *fs_info = root->fs_info; |
1911 | ||
a4abeea4 | 1912 | spin_lock(&fs_info->trans_lock); |
9d1a2a3a DS |
1913 | if (list_empty(&fs_info->dead_roots)) { |
1914 | spin_unlock(&fs_info->trans_lock); | |
1915 | return 0; | |
1916 | } | |
1917 | root = list_first_entry(&fs_info->dead_roots, | |
1918 | struct btrfs_root, root_list); | |
1919 | list_del(&root->root_list); | |
a4abeea4 | 1920 | spin_unlock(&fs_info->trans_lock); |
e9d0b13b | 1921 | |
9d1a2a3a DS |
1922 | pr_debug("btrfs: cleaner removing %llu\n", |
1923 | (unsigned long long)root->objectid); | |
76dda93c | 1924 | |
9d1a2a3a | 1925 | btrfs_kill_all_delayed_nodes(root); |
16cdcec7 | 1926 | |
9d1a2a3a DS |
1927 | if (btrfs_header_backref_rev(root->node) < |
1928 | BTRFS_MIXED_BACKREF_REV) | |
1929 | ret = btrfs_drop_snapshot(root, NULL, 0, 0); | |
1930 | else | |
1931 | ret = btrfs_drop_snapshot(root, NULL, 1, 0); | |
1932 | /* | |
1933 | * If we encounter a transaction abort during snapshot cleaning, we | |
1934 | * don't want to crash here | |
1935 | */ | |
1936 | BUG_ON(ret < 0 && ret != -EAGAIN && ret != -EROFS); | |
1937 | return 1; | |
e9d0b13b | 1938 | } |