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 && | |
501407aa JB |
305 | trans->state < TRANS_STATE_UNBLOCKED && |
306 | !trans->aborted); | |
4a9d8bde MX |
307 | } |
308 | ||
d352ac68 CM |
309 | /* wait for commit against the current transaction to become unblocked |
310 | * when this is done, it is safe to start a new transaction, but the current | |
311 | * transaction might not be fully on disk. | |
312 | */ | |
37d1aeee | 313 | static void wait_current_trans(struct btrfs_root *root) |
79154b1b | 314 | { |
f9295749 | 315 | struct btrfs_transaction *cur_trans; |
79154b1b | 316 | |
a4abeea4 | 317 | spin_lock(&root->fs_info->trans_lock); |
f9295749 | 318 | cur_trans = root->fs_info->running_transaction; |
4a9d8bde | 319 | if (cur_trans && is_transaction_blocked(cur_trans)) { |
13c5a93e | 320 | atomic_inc(&cur_trans->use_count); |
a4abeea4 | 321 | spin_unlock(&root->fs_info->trans_lock); |
72d63ed6 LZ |
322 | |
323 | wait_event(root->fs_info->transaction_wait, | |
501407aa JB |
324 | cur_trans->state >= TRANS_STATE_UNBLOCKED || |
325 | cur_trans->aborted); | |
f9295749 | 326 | put_transaction(cur_trans); |
a4abeea4 JB |
327 | } else { |
328 | spin_unlock(&root->fs_info->trans_lock); | |
f9295749 | 329 | } |
37d1aeee CM |
330 | } |
331 | ||
a22285a6 YZ |
332 | static int may_wait_transaction(struct btrfs_root *root, int type) |
333 | { | |
a4abeea4 JB |
334 | if (root->fs_info->log_root_recovering) |
335 | return 0; | |
336 | ||
337 | if (type == TRANS_USERSPACE) | |
338 | return 1; | |
339 | ||
340 | if (type == TRANS_START && | |
341 | !atomic_read(&root->fs_info->open_ioctl_trans)) | |
a22285a6 | 342 | return 1; |
a4abeea4 | 343 | |
a22285a6 YZ |
344 | return 0; |
345 | } | |
346 | ||
08e007d2 | 347 | static struct btrfs_trans_handle * |
0860adfd | 348 | start_transaction(struct btrfs_root *root, u64 num_items, unsigned int type, |
08e007d2 | 349 | enum btrfs_reserve_flush_enum flush) |
37d1aeee | 350 | { |
a22285a6 YZ |
351 | struct btrfs_trans_handle *h; |
352 | struct btrfs_transaction *cur_trans; | |
b5009945 | 353 | u64 num_bytes = 0; |
37d1aeee | 354 | int ret; |
c5567237 | 355 | u64 qgroup_reserved = 0; |
acce952b | 356 | |
87533c47 | 357 | if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) |
acce952b | 358 | return ERR_PTR(-EROFS); |
2a1eb461 JB |
359 | |
360 | if (current->journal_info) { | |
0860adfd | 361 | WARN_ON(type & TRANS_EXTWRITERS); |
2a1eb461 JB |
362 | h = current->journal_info; |
363 | h->use_count++; | |
b7d5b0a8 | 364 | WARN_ON(h->use_count > 2); |
2a1eb461 JB |
365 | h->orig_rsv = h->block_rsv; |
366 | h->block_rsv = NULL; | |
367 | goto got_it; | |
368 | } | |
b5009945 JB |
369 | |
370 | /* | |
371 | * Do the reservation before we join the transaction so we can do all | |
372 | * the appropriate flushing if need be. | |
373 | */ | |
374 | if (num_items > 0 && root != root->fs_info->chunk_root) { | |
c5567237 AJ |
375 | if (root->fs_info->quota_enabled && |
376 | is_fstree(root->root_key.objectid)) { | |
377 | qgroup_reserved = num_items * root->leafsize; | |
378 | ret = btrfs_qgroup_reserve(root, qgroup_reserved); | |
379 | if (ret) | |
380 | return ERR_PTR(ret); | |
381 | } | |
382 | ||
b5009945 | 383 | num_bytes = btrfs_calc_trans_metadata_size(root, num_items); |
08e007d2 MX |
384 | ret = btrfs_block_rsv_add(root, |
385 | &root->fs_info->trans_block_rsv, | |
386 | num_bytes, flush); | |
b5009945 | 387 | if (ret) |
843fcf35 | 388 | goto reserve_fail; |
b5009945 | 389 | } |
a22285a6 YZ |
390 | again: |
391 | h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS); | |
843fcf35 MX |
392 | if (!h) { |
393 | ret = -ENOMEM; | |
394 | goto alloc_fail; | |
395 | } | |
37d1aeee | 396 | |
98114659 JB |
397 | /* |
398 | * If we are JOIN_NOLOCK we're already committing a transaction and | |
399 | * waiting on this guy, so we don't need to do the sb_start_intwrite | |
400 | * because we're already holding a ref. We need this because we could | |
401 | * have raced in and did an fsync() on a file which can kick a commit | |
402 | * and then we deadlock with somebody doing a freeze. | |
354aa0fb MX |
403 | * |
404 | * If we are ATTACH, it means we just want to catch the current | |
405 | * transaction and commit it, so we needn't do sb_start_intwrite(). | |
98114659 | 406 | */ |
0860adfd | 407 | if (type & __TRANS_FREEZABLE) |
60376ce4 | 408 | sb_start_intwrite(root->fs_info->sb); |
b2b5ef5c | 409 | |
a22285a6 | 410 | if (may_wait_transaction(root, type)) |
37d1aeee | 411 | wait_current_trans(root); |
a22285a6 | 412 | |
a4abeea4 | 413 | do { |
354aa0fb | 414 | ret = join_transaction(root, type); |
178260b2 | 415 | if (ret == -EBUSY) { |
a4abeea4 | 416 | wait_current_trans(root); |
178260b2 MX |
417 | if (unlikely(type == TRANS_ATTACH)) |
418 | ret = -ENOENT; | |
419 | } | |
a4abeea4 JB |
420 | } while (ret == -EBUSY); |
421 | ||
db5b493a | 422 | if (ret < 0) { |
354aa0fb MX |
423 | /* We must get the transaction if we are JOIN_NOLOCK. */ |
424 | BUG_ON(type == TRANS_JOIN_NOLOCK); | |
843fcf35 | 425 | goto join_fail; |
db5b493a | 426 | } |
0f7d52f4 | 427 | |
a22285a6 | 428 | cur_trans = root->fs_info->running_transaction; |
a22285a6 YZ |
429 | |
430 | h->transid = cur_trans->transid; | |
431 | h->transaction = cur_trans; | |
79154b1b | 432 | h->blocks_used = 0; |
a22285a6 | 433 | h->bytes_reserved = 0; |
d13603ef | 434 | h->root = root; |
56bec294 | 435 | h->delayed_ref_updates = 0; |
2a1eb461 | 436 | h->use_count = 1; |
0e721106 | 437 | h->adding_csums = 0; |
f0486c68 | 438 | h->block_rsv = NULL; |
2a1eb461 | 439 | h->orig_rsv = NULL; |
49b25e05 | 440 | h->aborted = 0; |
4b824906 | 441 | h->qgroup_reserved = 0; |
bed92eae | 442 | h->delayed_ref_elem.seq = 0; |
a698d075 | 443 | h->type = type; |
c6b305a8 | 444 | h->allocating_chunk = false; |
bed92eae | 445 | INIT_LIST_HEAD(&h->qgroup_ref_list); |
ea658bad | 446 | INIT_LIST_HEAD(&h->new_bgs); |
b7ec40d7 | 447 | |
a22285a6 | 448 | smp_mb(); |
4a9d8bde MX |
449 | if (cur_trans->state >= TRANS_STATE_BLOCKED && |
450 | may_wait_transaction(root, type)) { | |
a22285a6 YZ |
451 | btrfs_commit_transaction(h, root); |
452 | goto again; | |
453 | } | |
454 | ||
b5009945 | 455 | if (num_bytes) { |
8c2a3ca2 | 456 | trace_btrfs_space_reservation(root->fs_info, "transaction", |
2bcc0328 | 457 | h->transid, num_bytes, 1); |
b5009945 JB |
458 | h->block_rsv = &root->fs_info->trans_block_rsv; |
459 | h->bytes_reserved = num_bytes; | |
a22285a6 | 460 | } |
4b824906 | 461 | h->qgroup_reserved = qgroup_reserved; |
9ed74f2d | 462 | |
2a1eb461 | 463 | got_it: |
a4abeea4 | 464 | btrfs_record_root_in_trans(h, root); |
a22285a6 YZ |
465 | |
466 | if (!current->journal_info && type != TRANS_USERSPACE) | |
467 | current->journal_info = h; | |
79154b1b | 468 | return h; |
843fcf35 MX |
469 | |
470 | join_fail: | |
0860adfd | 471 | if (type & __TRANS_FREEZABLE) |
843fcf35 MX |
472 | sb_end_intwrite(root->fs_info->sb); |
473 | kmem_cache_free(btrfs_trans_handle_cachep, h); | |
474 | alloc_fail: | |
475 | if (num_bytes) | |
476 | btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv, | |
477 | num_bytes); | |
478 | reserve_fail: | |
479 | if (qgroup_reserved) | |
480 | btrfs_qgroup_free(root, qgroup_reserved); | |
481 | return ERR_PTR(ret); | |
79154b1b CM |
482 | } |
483 | ||
f9295749 | 484 | struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, |
a22285a6 | 485 | int num_items) |
f9295749 | 486 | { |
08e007d2 MX |
487 | return start_transaction(root, num_items, TRANS_START, |
488 | BTRFS_RESERVE_FLUSH_ALL); | |
f9295749 | 489 | } |
8407aa46 | 490 | |
08e007d2 | 491 | struct btrfs_trans_handle *btrfs_start_transaction_lflush( |
8407aa46 MX |
492 | struct btrfs_root *root, int num_items) |
493 | { | |
08e007d2 MX |
494 | return start_transaction(root, num_items, TRANS_START, |
495 | BTRFS_RESERVE_FLUSH_LIMIT); | |
8407aa46 MX |
496 | } |
497 | ||
7a7eaa40 | 498 | struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root) |
f9295749 | 499 | { |
8407aa46 | 500 | return start_transaction(root, 0, TRANS_JOIN, 0); |
f9295749 CM |
501 | } |
502 | ||
7a7eaa40 | 503 | struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root) |
0af3d00b | 504 | { |
8407aa46 | 505 | return start_transaction(root, 0, TRANS_JOIN_NOLOCK, 0); |
0af3d00b JB |
506 | } |
507 | ||
7a7eaa40 | 508 | struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root) |
9ca9ee09 | 509 | { |
8407aa46 | 510 | return start_transaction(root, 0, TRANS_USERSPACE, 0); |
9ca9ee09 SW |
511 | } |
512 | ||
d4edf39b MX |
513 | /* |
514 | * btrfs_attach_transaction() - catch the running transaction | |
515 | * | |
516 | * It is used when we want to commit the current the transaction, but | |
517 | * don't want to start a new one. | |
518 | * | |
519 | * Note: If this function return -ENOENT, it just means there is no | |
520 | * running transaction. But it is possible that the inactive transaction | |
521 | * is still in the memory, not fully on disk. If you hope there is no | |
522 | * inactive transaction in the fs when -ENOENT is returned, you should | |
523 | * invoke | |
524 | * btrfs_attach_transaction_barrier() | |
525 | */ | |
354aa0fb | 526 | struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root) |
60376ce4 | 527 | { |
354aa0fb | 528 | return start_transaction(root, 0, TRANS_ATTACH, 0); |
60376ce4 JB |
529 | } |
530 | ||
d4edf39b | 531 | /* |
90b6d283 | 532 | * btrfs_attach_transaction_barrier() - catch the running transaction |
d4edf39b MX |
533 | * |
534 | * It is similar to the above function, the differentia is this one | |
535 | * will wait for all the inactive transactions until they fully | |
536 | * complete. | |
537 | */ | |
538 | struct btrfs_trans_handle * | |
539 | btrfs_attach_transaction_barrier(struct btrfs_root *root) | |
540 | { | |
541 | struct btrfs_trans_handle *trans; | |
542 | ||
543 | trans = start_transaction(root, 0, TRANS_ATTACH, 0); | |
544 | if (IS_ERR(trans) && PTR_ERR(trans) == -ENOENT) | |
545 | btrfs_wait_for_commit(root, 0); | |
546 | ||
547 | return trans; | |
548 | } | |
549 | ||
d352ac68 | 550 | /* wait for a transaction commit to be fully complete */ |
b9c8300c | 551 | static noinline void wait_for_commit(struct btrfs_root *root, |
89ce8a63 CM |
552 | struct btrfs_transaction *commit) |
553 | { | |
4a9d8bde | 554 | wait_event(commit->commit_wait, commit->state == TRANS_STATE_COMPLETED); |
89ce8a63 CM |
555 | } |
556 | ||
46204592 SW |
557 | int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid) |
558 | { | |
559 | struct btrfs_transaction *cur_trans = NULL, *t; | |
8cd2807f | 560 | int ret = 0; |
46204592 | 561 | |
46204592 SW |
562 | if (transid) { |
563 | if (transid <= root->fs_info->last_trans_committed) | |
a4abeea4 | 564 | goto out; |
46204592 | 565 | |
8cd2807f | 566 | ret = -EINVAL; |
46204592 | 567 | /* find specified transaction */ |
a4abeea4 | 568 | spin_lock(&root->fs_info->trans_lock); |
46204592 SW |
569 | list_for_each_entry(t, &root->fs_info->trans_list, list) { |
570 | if (t->transid == transid) { | |
571 | cur_trans = t; | |
a4abeea4 | 572 | atomic_inc(&cur_trans->use_count); |
8cd2807f | 573 | ret = 0; |
46204592 SW |
574 | break; |
575 | } | |
8cd2807f MX |
576 | if (t->transid > transid) { |
577 | ret = 0; | |
46204592 | 578 | break; |
8cd2807f | 579 | } |
46204592 | 580 | } |
a4abeea4 | 581 | spin_unlock(&root->fs_info->trans_lock); |
8cd2807f | 582 | /* The specified transaction doesn't exist */ |
46204592 | 583 | if (!cur_trans) |
8cd2807f | 584 | goto out; |
46204592 SW |
585 | } else { |
586 | /* find newest transaction that is committing | committed */ | |
a4abeea4 | 587 | spin_lock(&root->fs_info->trans_lock); |
46204592 SW |
588 | list_for_each_entry_reverse(t, &root->fs_info->trans_list, |
589 | list) { | |
4a9d8bde MX |
590 | if (t->state >= TRANS_STATE_COMMIT_START) { |
591 | if (t->state == TRANS_STATE_COMPLETED) | |
3473f3c0 | 592 | break; |
46204592 | 593 | cur_trans = t; |
a4abeea4 | 594 | atomic_inc(&cur_trans->use_count); |
46204592 SW |
595 | break; |
596 | } | |
597 | } | |
a4abeea4 | 598 | spin_unlock(&root->fs_info->trans_lock); |
46204592 | 599 | if (!cur_trans) |
a4abeea4 | 600 | goto out; /* nothing committing|committed */ |
46204592 SW |
601 | } |
602 | ||
46204592 | 603 | wait_for_commit(root, cur_trans); |
46204592 | 604 | put_transaction(cur_trans); |
a4abeea4 | 605 | out: |
46204592 SW |
606 | return ret; |
607 | } | |
608 | ||
37d1aeee CM |
609 | void btrfs_throttle(struct btrfs_root *root) |
610 | { | |
a4abeea4 | 611 | if (!atomic_read(&root->fs_info->open_ioctl_trans)) |
9ca9ee09 | 612 | wait_current_trans(root); |
37d1aeee CM |
613 | } |
614 | ||
8929ecfa YZ |
615 | static int should_end_transaction(struct btrfs_trans_handle *trans, |
616 | struct btrfs_root *root) | |
617 | { | |
1be41b78 JB |
618 | if (root->fs_info->global_block_rsv.space_info->full && |
619 | btrfs_should_throttle_delayed_refs(trans, root)) | |
620 | return 1; | |
36ba022a | 621 | |
1be41b78 | 622 | return !!btrfs_block_rsv_check(root, &root->fs_info->global_block_rsv, 5); |
8929ecfa YZ |
623 | } |
624 | ||
625 | int btrfs_should_end_transaction(struct btrfs_trans_handle *trans, | |
626 | struct btrfs_root *root) | |
627 | { | |
628 | struct btrfs_transaction *cur_trans = trans->transaction; | |
629 | int updates; | |
49b25e05 | 630 | int err; |
8929ecfa | 631 | |
a4abeea4 | 632 | smp_mb(); |
4a9d8bde MX |
633 | if (cur_trans->state >= TRANS_STATE_BLOCKED || |
634 | cur_trans->delayed_refs.flushing) | |
8929ecfa YZ |
635 | return 1; |
636 | ||
637 | updates = trans->delayed_ref_updates; | |
638 | trans->delayed_ref_updates = 0; | |
49b25e05 JM |
639 | if (updates) { |
640 | err = btrfs_run_delayed_refs(trans, root, updates); | |
641 | if (err) /* Error code will also eval true */ | |
642 | return err; | |
643 | } | |
8929ecfa YZ |
644 | |
645 | return should_end_transaction(trans, root); | |
646 | } | |
647 | ||
89ce8a63 | 648 | static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, |
a698d075 | 649 | struct btrfs_root *root, int throttle) |
79154b1b | 650 | { |
8929ecfa | 651 | struct btrfs_transaction *cur_trans = trans->transaction; |
ab78c84d | 652 | struct btrfs_fs_info *info = root->fs_info; |
1be41b78 | 653 | unsigned long cur = trans->delayed_ref_updates; |
a698d075 | 654 | int lock = (trans->type != TRANS_JOIN_NOLOCK); |
4edc2ca3 | 655 | int err = 0; |
c3e69d58 | 656 | |
2a1eb461 JB |
657 | if (--trans->use_count) { |
658 | trans->block_rsv = trans->orig_rsv; | |
659 | return 0; | |
660 | } | |
661 | ||
edf39272 JS |
662 | /* |
663 | * do the qgroup accounting as early as possible | |
664 | */ | |
665 | err = btrfs_delayed_refs_qgroup_accounting(trans, info); | |
666 | ||
b24e03db | 667 | btrfs_trans_release_metadata(trans, root); |
4c13d758 | 668 | trans->block_rsv = NULL; |
c5567237 AJ |
669 | |
670 | if (trans->qgroup_reserved) { | |
7c2ec3f0 LB |
671 | /* |
672 | * the same root has to be passed here between start_transaction | |
673 | * and end_transaction. Subvolume quota depends on this. | |
674 | */ | |
675 | btrfs_qgroup_free(trans->root, trans->qgroup_reserved); | |
c5567237 AJ |
676 | trans->qgroup_reserved = 0; |
677 | } | |
678 | ||
ea658bad JB |
679 | if (!list_empty(&trans->new_bgs)) |
680 | btrfs_create_pending_block_groups(trans, root); | |
681 | ||
1be41b78 JB |
682 | trans->delayed_ref_updates = 0; |
683 | if (btrfs_should_throttle_delayed_refs(trans, root)) { | |
684 | cur = max_t(unsigned long, cur, 1); | |
c3e69d58 | 685 | trans->delayed_ref_updates = 0; |
1be41b78 | 686 | btrfs_run_delayed_refs(trans, root, cur); |
56bec294 | 687 | } |
bb721703 | 688 | |
0e721106 JB |
689 | btrfs_trans_release_metadata(trans, root); |
690 | trans->block_rsv = NULL; | |
56bec294 | 691 | |
ea658bad JB |
692 | if (!list_empty(&trans->new_bgs)) |
693 | btrfs_create_pending_block_groups(trans, root); | |
694 | ||
a4abeea4 | 695 | if (lock && !atomic_read(&root->fs_info->open_ioctl_trans) && |
4a9d8bde MX |
696 | should_end_transaction(trans, root) && |
697 | ACCESS_ONCE(cur_trans->state) == TRANS_STATE_RUNNING) { | |
698 | spin_lock(&info->trans_lock); | |
699 | if (cur_trans->state == TRANS_STATE_RUNNING) | |
700 | cur_trans->state = TRANS_STATE_BLOCKED; | |
701 | spin_unlock(&info->trans_lock); | |
a4abeea4 | 702 | } |
8929ecfa | 703 | |
4a9d8bde | 704 | if (lock && ACCESS_ONCE(cur_trans->state) == TRANS_STATE_BLOCKED) { |
81317fde JB |
705 | if (throttle) { |
706 | /* | |
707 | * We may race with somebody else here so end up having | |
708 | * to call end_transaction on ourselves again, so inc | |
709 | * our use_count. | |
710 | */ | |
711 | trans->use_count++; | |
8929ecfa | 712 | return btrfs_commit_transaction(trans, root); |
81317fde | 713 | } else { |
8929ecfa | 714 | wake_up_process(info->transaction_kthread); |
81317fde | 715 | } |
8929ecfa YZ |
716 | } |
717 | ||
0860adfd | 718 | if (trans->type & __TRANS_FREEZABLE) |
98114659 | 719 | sb_end_intwrite(root->fs_info->sb); |
6df7881a | 720 | |
8929ecfa | 721 | WARN_ON(cur_trans != info->running_transaction); |
13c5a93e JB |
722 | WARN_ON(atomic_read(&cur_trans->num_writers) < 1); |
723 | atomic_dec(&cur_trans->num_writers); | |
0860adfd | 724 | extwriter_counter_dec(cur_trans, trans->type); |
89ce8a63 | 725 | |
99d16cbc | 726 | smp_mb(); |
79154b1b CM |
727 | if (waitqueue_active(&cur_trans->writer_wait)) |
728 | wake_up(&cur_trans->writer_wait); | |
79154b1b | 729 | put_transaction(cur_trans); |
9ed74f2d JB |
730 | |
731 | if (current->journal_info == trans) | |
732 | current->journal_info = NULL; | |
ab78c84d | 733 | |
24bbcf04 YZ |
734 | if (throttle) |
735 | btrfs_run_delayed_iputs(root); | |
736 | ||
49b25e05 | 737 | if (trans->aborted || |
87533c47 | 738 | test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) |
4edc2ca3 | 739 | err = -EIO; |
edf39272 | 740 | assert_qgroups_uptodate(trans); |
49b25e05 | 741 | |
4edc2ca3 DJ |
742 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
743 | return err; | |
79154b1b CM |
744 | } |
745 | ||
89ce8a63 CM |
746 | int btrfs_end_transaction(struct btrfs_trans_handle *trans, |
747 | struct btrfs_root *root) | |
748 | { | |
98ad43be | 749 | return __btrfs_end_transaction(trans, root, 0); |
89ce8a63 CM |
750 | } |
751 | ||
752 | int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans, | |
753 | struct btrfs_root *root) | |
754 | { | |
98ad43be | 755 | return __btrfs_end_transaction(trans, root, 1); |
16cdcec7 MX |
756 | } |
757 | ||
758 | int btrfs_end_transaction_dmeta(struct btrfs_trans_handle *trans, | |
759 | struct btrfs_root *root) | |
760 | { | |
a698d075 | 761 | return __btrfs_end_transaction(trans, root, 1); |
89ce8a63 CM |
762 | } |
763 | ||
d352ac68 CM |
764 | /* |
765 | * when btree blocks are allocated, they have some corresponding bits set for | |
766 | * them in one of two extent_io trees. This is used to make sure all of | |
690587d1 | 767 | * those extents are sent to disk but does not wait on them |
d352ac68 | 768 | */ |
690587d1 | 769 | int btrfs_write_marked_extents(struct btrfs_root *root, |
8cef4e16 | 770 | struct extent_io_tree *dirty_pages, int mark) |
79154b1b | 771 | { |
777e6bd7 | 772 | int err = 0; |
7c4452b9 | 773 | int werr = 0; |
1728366e | 774 | struct address_space *mapping = root->fs_info->btree_inode->i_mapping; |
e6138876 | 775 | struct extent_state *cached_state = NULL; |
777e6bd7 | 776 | u64 start = 0; |
5f39d397 | 777 | u64 end; |
7c4452b9 | 778 | |
1728366e | 779 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
e6138876 JB |
780 | mark, &cached_state)) { |
781 | convert_extent_bit(dirty_pages, start, end, EXTENT_NEED_WAIT, | |
782 | mark, &cached_state, GFP_NOFS); | |
783 | cached_state = NULL; | |
1728366e JB |
784 | err = filemap_fdatawrite_range(mapping, start, end); |
785 | if (err) | |
786 | werr = err; | |
787 | cond_resched(); | |
788 | start = end + 1; | |
7c4452b9 | 789 | } |
690587d1 CM |
790 | if (err) |
791 | werr = err; | |
792 | return werr; | |
793 | } | |
794 | ||
795 | /* | |
796 | * when btree blocks are allocated, they have some corresponding bits set for | |
797 | * them in one of two extent_io trees. This is used to make sure all of | |
798 | * those extents are on disk for transaction or log commit. We wait | |
799 | * on all the pages and clear them from the dirty pages state tree | |
800 | */ | |
801 | int btrfs_wait_marked_extents(struct btrfs_root *root, | |
8cef4e16 | 802 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 | 803 | { |
690587d1 CM |
804 | int err = 0; |
805 | int werr = 0; | |
1728366e | 806 | struct address_space *mapping = root->fs_info->btree_inode->i_mapping; |
e6138876 | 807 | struct extent_state *cached_state = NULL; |
690587d1 CM |
808 | u64 start = 0; |
809 | u64 end; | |
777e6bd7 | 810 | |
1728366e | 811 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
e6138876 JB |
812 | EXTENT_NEED_WAIT, &cached_state)) { |
813 | clear_extent_bit(dirty_pages, start, end, EXTENT_NEED_WAIT, | |
814 | 0, 0, &cached_state, GFP_NOFS); | |
1728366e JB |
815 | err = filemap_fdatawait_range(mapping, start, end); |
816 | if (err) | |
817 | werr = err; | |
818 | cond_resched(); | |
819 | start = end + 1; | |
777e6bd7 | 820 | } |
7c4452b9 CM |
821 | if (err) |
822 | werr = err; | |
823 | return werr; | |
79154b1b CM |
824 | } |
825 | ||
690587d1 CM |
826 | /* |
827 | * when btree blocks are allocated, they have some corresponding bits set for | |
828 | * them in one of two extent_io trees. This is used to make sure all of | |
829 | * those extents are on disk for transaction or log commit | |
830 | */ | |
831 | int btrfs_write_and_wait_marked_extents(struct btrfs_root *root, | |
8cef4e16 | 832 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 CM |
833 | { |
834 | int ret; | |
835 | int ret2; | |
c6adc9cc | 836 | struct blk_plug plug; |
690587d1 | 837 | |
c6adc9cc | 838 | blk_start_plug(&plug); |
8cef4e16 | 839 | ret = btrfs_write_marked_extents(root, dirty_pages, mark); |
c6adc9cc | 840 | blk_finish_plug(&plug); |
8cef4e16 | 841 | ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark); |
bf0da8c1 CM |
842 | |
843 | if (ret) | |
844 | return ret; | |
845 | if (ret2) | |
846 | return ret2; | |
847 | return 0; | |
690587d1 CM |
848 | } |
849 | ||
d0c803c4 CM |
850 | int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans, |
851 | struct btrfs_root *root) | |
852 | { | |
853 | if (!trans || !trans->transaction) { | |
854 | struct inode *btree_inode; | |
855 | btree_inode = root->fs_info->btree_inode; | |
856 | return filemap_write_and_wait(btree_inode->i_mapping); | |
857 | } | |
858 | return btrfs_write_and_wait_marked_extents(root, | |
8cef4e16 YZ |
859 | &trans->transaction->dirty_pages, |
860 | EXTENT_DIRTY); | |
d0c803c4 CM |
861 | } |
862 | ||
d352ac68 CM |
863 | /* |
864 | * this is used to update the root pointer in the tree of tree roots. | |
865 | * | |
866 | * But, in the case of the extent allocation tree, updating the root | |
867 | * pointer may allocate blocks which may change the root of the extent | |
868 | * allocation tree. | |
869 | * | |
870 | * So, this loops and repeats and makes sure the cowonly root didn't | |
871 | * change while the root pointer was being updated in the metadata. | |
872 | */ | |
0b86a832 CM |
873 | static int update_cowonly_root(struct btrfs_trans_handle *trans, |
874 | struct btrfs_root *root) | |
79154b1b CM |
875 | { |
876 | int ret; | |
0b86a832 | 877 | u64 old_root_bytenr; |
86b9f2ec | 878 | u64 old_root_used; |
0b86a832 | 879 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
79154b1b | 880 | |
86b9f2ec | 881 | old_root_used = btrfs_root_used(&root->root_item); |
0b86a832 | 882 | btrfs_write_dirty_block_groups(trans, root); |
56bec294 | 883 | |
d397712b | 884 | while (1) { |
0b86a832 | 885 | old_root_bytenr = btrfs_root_bytenr(&root->root_item); |
86b9f2ec YZ |
886 | if (old_root_bytenr == root->node->start && |
887 | old_root_used == btrfs_root_used(&root->root_item)) | |
79154b1b | 888 | break; |
87ef2bb4 | 889 | |
5d4f98a2 | 890 | btrfs_set_root_node(&root->root_item, root->node); |
79154b1b | 891 | ret = btrfs_update_root(trans, tree_root, |
0b86a832 CM |
892 | &root->root_key, |
893 | &root->root_item); | |
49b25e05 JM |
894 | if (ret) |
895 | return ret; | |
56bec294 | 896 | |
86b9f2ec | 897 | old_root_used = btrfs_root_used(&root->root_item); |
4a8c9a62 | 898 | ret = btrfs_write_dirty_block_groups(trans, root); |
49b25e05 JM |
899 | if (ret) |
900 | return ret; | |
0b86a832 | 901 | } |
276e680d YZ |
902 | |
903 | if (root != root->fs_info->extent_root) | |
904 | switch_commit_root(root); | |
905 | ||
0b86a832 CM |
906 | return 0; |
907 | } | |
908 | ||
d352ac68 CM |
909 | /* |
910 | * update all the cowonly tree roots on disk | |
49b25e05 JM |
911 | * |
912 | * The error handling in this function may not be obvious. Any of the | |
913 | * failures will cause the file system to go offline. We still need | |
914 | * to clean up the delayed refs. | |
d352ac68 | 915 | */ |
5d4f98a2 YZ |
916 | static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans, |
917 | struct btrfs_root *root) | |
0b86a832 CM |
918 | { |
919 | struct btrfs_fs_info *fs_info = root->fs_info; | |
920 | struct list_head *next; | |
84234f3a | 921 | struct extent_buffer *eb; |
56bec294 | 922 | int ret; |
84234f3a | 923 | |
56bec294 | 924 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
49b25e05 JM |
925 | if (ret) |
926 | return ret; | |
87ef2bb4 | 927 | |
84234f3a | 928 | eb = btrfs_lock_root_node(fs_info->tree_root); |
49b25e05 JM |
929 | ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, |
930 | 0, &eb); | |
84234f3a YZ |
931 | btrfs_tree_unlock(eb); |
932 | free_extent_buffer(eb); | |
0b86a832 | 933 | |
49b25e05 JM |
934 | if (ret) |
935 | return ret; | |
936 | ||
56bec294 | 937 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
49b25e05 JM |
938 | if (ret) |
939 | return ret; | |
87ef2bb4 | 940 | |
733f4fbb | 941 | ret = btrfs_run_dev_stats(trans, root->fs_info); |
8dabb742 SB |
942 | WARN_ON(ret); |
943 | ret = btrfs_run_dev_replace(trans, root->fs_info); | |
944 | WARN_ON(ret); | |
733f4fbb | 945 | |
546adb0d JS |
946 | ret = btrfs_run_qgroups(trans, root->fs_info); |
947 | BUG_ON(ret); | |
948 | ||
949 | /* run_qgroups might have added some more refs */ | |
950 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); | |
951 | BUG_ON(ret); | |
952 | ||
d397712b | 953 | while (!list_empty(&fs_info->dirty_cowonly_roots)) { |
0b86a832 CM |
954 | next = fs_info->dirty_cowonly_roots.next; |
955 | list_del_init(next); | |
956 | root = list_entry(next, struct btrfs_root, dirty_list); | |
87ef2bb4 | 957 | |
49b25e05 JM |
958 | ret = update_cowonly_root(trans, root); |
959 | if (ret) | |
960 | return ret; | |
79154b1b | 961 | } |
276e680d YZ |
962 | |
963 | down_write(&fs_info->extent_commit_sem); | |
964 | switch_commit_root(fs_info->extent_root); | |
965 | up_write(&fs_info->extent_commit_sem); | |
966 | ||
8dabb742 SB |
967 | btrfs_after_dev_replace_commit(fs_info); |
968 | ||
79154b1b CM |
969 | return 0; |
970 | } | |
971 | ||
d352ac68 CM |
972 | /* |
973 | * dead roots are old snapshots that need to be deleted. This allocates | |
974 | * a dirty root struct and adds it into the list of dead roots that need to | |
975 | * be deleted | |
976 | */ | |
5d4f98a2 | 977 | int btrfs_add_dead_root(struct btrfs_root *root) |
5eda7b5e | 978 | { |
a4abeea4 | 979 | spin_lock(&root->fs_info->trans_lock); |
9d1a2a3a | 980 | list_add_tail(&root->root_list, &root->fs_info->dead_roots); |
a4abeea4 | 981 | spin_unlock(&root->fs_info->trans_lock); |
5eda7b5e CM |
982 | return 0; |
983 | } | |
984 | ||
d352ac68 | 985 | /* |
5d4f98a2 | 986 | * update all the cowonly tree roots on disk |
d352ac68 | 987 | */ |
5d4f98a2 YZ |
988 | static noinline int commit_fs_roots(struct btrfs_trans_handle *trans, |
989 | struct btrfs_root *root) | |
0f7d52f4 | 990 | { |
0f7d52f4 | 991 | struct btrfs_root *gang[8]; |
5d4f98a2 | 992 | struct btrfs_fs_info *fs_info = root->fs_info; |
0f7d52f4 CM |
993 | int i; |
994 | int ret; | |
54aa1f4d CM |
995 | int err = 0; |
996 | ||
a4abeea4 | 997 | spin_lock(&fs_info->fs_roots_radix_lock); |
d397712b | 998 | while (1) { |
5d4f98a2 YZ |
999 | ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, |
1000 | (void **)gang, 0, | |
0f7d52f4 CM |
1001 | ARRAY_SIZE(gang), |
1002 | BTRFS_ROOT_TRANS_TAG); | |
1003 | if (ret == 0) | |
1004 | break; | |
1005 | for (i = 0; i < ret; i++) { | |
1006 | root = gang[i]; | |
5d4f98a2 YZ |
1007 | radix_tree_tag_clear(&fs_info->fs_roots_radix, |
1008 | (unsigned long)root->root_key.objectid, | |
1009 | BTRFS_ROOT_TRANS_TAG); | |
a4abeea4 | 1010 | spin_unlock(&fs_info->fs_roots_radix_lock); |
31153d81 | 1011 | |
e02119d5 | 1012 | btrfs_free_log(trans, root); |
5d4f98a2 | 1013 | btrfs_update_reloc_root(trans, root); |
d68fc57b | 1014 | btrfs_orphan_commit_root(trans, root); |
bcc63abb | 1015 | |
82d5902d LZ |
1016 | btrfs_save_ino_cache(root, trans); |
1017 | ||
f1ebcc74 LB |
1018 | /* see comments in should_cow_block() */ |
1019 | root->force_cow = 0; | |
1020 | smp_wmb(); | |
1021 | ||
978d910d | 1022 | if (root->commit_root != root->node) { |
581bb050 | 1023 | mutex_lock(&root->fs_commit_mutex); |
817d52f8 | 1024 | switch_commit_root(root); |
581bb050 LZ |
1025 | btrfs_unpin_free_ino(root); |
1026 | mutex_unlock(&root->fs_commit_mutex); | |
1027 | ||
978d910d YZ |
1028 | btrfs_set_root_node(&root->root_item, |
1029 | root->node); | |
1030 | } | |
5d4f98a2 | 1031 | |
5d4f98a2 | 1032 | err = btrfs_update_root(trans, fs_info->tree_root, |
0f7d52f4 CM |
1033 | &root->root_key, |
1034 | &root->root_item); | |
a4abeea4 | 1035 | spin_lock(&fs_info->fs_roots_radix_lock); |
54aa1f4d CM |
1036 | if (err) |
1037 | break; | |
0f7d52f4 CM |
1038 | } |
1039 | } | |
a4abeea4 | 1040 | spin_unlock(&fs_info->fs_roots_radix_lock); |
54aa1f4d | 1041 | return err; |
0f7d52f4 CM |
1042 | } |
1043 | ||
d352ac68 | 1044 | /* |
de78b51a ES |
1045 | * defrag a given btree. |
1046 | * Every leaf in the btree is read and defragged. | |
d352ac68 | 1047 | */ |
de78b51a | 1048 | int btrfs_defrag_root(struct btrfs_root *root) |
e9d0b13b CM |
1049 | { |
1050 | struct btrfs_fs_info *info = root->fs_info; | |
e9d0b13b | 1051 | struct btrfs_trans_handle *trans; |
8929ecfa | 1052 | int ret; |
e9d0b13b | 1053 | |
8929ecfa | 1054 | if (xchg(&root->defrag_running, 1)) |
e9d0b13b | 1055 | return 0; |
8929ecfa | 1056 | |
6b80053d | 1057 | while (1) { |
8929ecfa YZ |
1058 | trans = btrfs_start_transaction(root, 0); |
1059 | if (IS_ERR(trans)) | |
1060 | return PTR_ERR(trans); | |
1061 | ||
de78b51a | 1062 | ret = btrfs_defrag_leaves(trans, root); |
8929ecfa | 1063 | |
e9d0b13b | 1064 | btrfs_end_transaction(trans, root); |
b53d3f5d | 1065 | btrfs_btree_balance_dirty(info->tree_root); |
e9d0b13b CM |
1066 | cond_resched(); |
1067 | ||
7841cb28 | 1068 | if (btrfs_fs_closing(root->fs_info) || ret != -EAGAIN) |
e9d0b13b | 1069 | break; |
210549eb DS |
1070 | |
1071 | if (btrfs_defrag_cancelled(root->fs_info)) { | |
1072 | printk(KERN_DEBUG "btrfs: defrag_root cancelled\n"); | |
1073 | ret = -EAGAIN; | |
1074 | break; | |
1075 | } | |
e9d0b13b CM |
1076 | } |
1077 | root->defrag_running = 0; | |
8929ecfa | 1078 | return ret; |
e9d0b13b CM |
1079 | } |
1080 | ||
d352ac68 CM |
1081 | /* |
1082 | * new snapshots need to be created at a very specific time in the | |
aec8030a MX |
1083 | * transaction commit. This does the actual creation. |
1084 | * | |
1085 | * Note: | |
1086 | * If the error which may affect the commitment of the current transaction | |
1087 | * happens, we should return the error number. If the error which just affect | |
1088 | * the creation of the pending snapshots, just return 0. | |
d352ac68 | 1089 | */ |
80b6794d | 1090 | static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, |
3063d29f CM |
1091 | struct btrfs_fs_info *fs_info, |
1092 | struct btrfs_pending_snapshot *pending) | |
1093 | { | |
1094 | struct btrfs_key key; | |
80b6794d | 1095 | struct btrfs_root_item *new_root_item; |
3063d29f CM |
1096 | struct btrfs_root *tree_root = fs_info->tree_root; |
1097 | struct btrfs_root *root = pending->root; | |
6bdb72de | 1098 | struct btrfs_root *parent_root; |
98c9942a | 1099 | struct btrfs_block_rsv *rsv; |
6bdb72de | 1100 | struct inode *parent_inode; |
42874b3d MX |
1101 | struct btrfs_path *path; |
1102 | struct btrfs_dir_item *dir_item; | |
a22285a6 | 1103 | struct dentry *dentry; |
3063d29f | 1104 | struct extent_buffer *tmp; |
925baedd | 1105 | struct extent_buffer *old; |
8ea05e3a | 1106 | struct timespec cur_time = CURRENT_TIME; |
aec8030a | 1107 | int ret = 0; |
d68fc57b | 1108 | u64 to_reserve = 0; |
6bdb72de | 1109 | u64 index = 0; |
a22285a6 | 1110 | u64 objectid; |
b83cc969 | 1111 | u64 root_flags; |
8ea05e3a | 1112 | uuid_le new_uuid; |
3063d29f | 1113 | |
42874b3d MX |
1114 | path = btrfs_alloc_path(); |
1115 | if (!path) { | |
aec8030a MX |
1116 | pending->error = -ENOMEM; |
1117 | return 0; | |
42874b3d MX |
1118 | } |
1119 | ||
80b6794d CM |
1120 | new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS); |
1121 | if (!new_root_item) { | |
aec8030a | 1122 | pending->error = -ENOMEM; |
6fa9700e | 1123 | goto root_item_alloc_fail; |
80b6794d | 1124 | } |
a22285a6 | 1125 | |
aec8030a MX |
1126 | pending->error = btrfs_find_free_objectid(tree_root, &objectid); |
1127 | if (pending->error) | |
6fa9700e | 1128 | goto no_free_objectid; |
3063d29f | 1129 | |
3fd0a558 | 1130 | btrfs_reloc_pre_snapshot(trans, pending, &to_reserve); |
d68fc57b YZ |
1131 | |
1132 | if (to_reserve > 0) { | |
aec8030a MX |
1133 | pending->error = btrfs_block_rsv_add(root, |
1134 | &pending->block_rsv, | |
1135 | to_reserve, | |
1136 | BTRFS_RESERVE_NO_FLUSH); | |
1137 | if (pending->error) | |
6fa9700e | 1138 | goto no_free_objectid; |
d68fc57b YZ |
1139 | } |
1140 | ||
aec8030a MX |
1141 | pending->error = btrfs_qgroup_inherit(trans, fs_info, |
1142 | root->root_key.objectid, | |
1143 | objectid, pending->inherit); | |
1144 | if (pending->error) | |
6fa9700e | 1145 | goto no_free_objectid; |
6f72c7e2 | 1146 | |
3063d29f | 1147 | key.objectid = objectid; |
a22285a6 YZ |
1148 | key.offset = (u64)-1; |
1149 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3063d29f | 1150 | |
6fa9700e | 1151 | rsv = trans->block_rsv; |
a22285a6 | 1152 | trans->block_rsv = &pending->block_rsv; |
2382c5cc | 1153 | trans->bytes_reserved = trans->block_rsv->reserved; |
3de4586c | 1154 | |
a22285a6 | 1155 | dentry = pending->dentry; |
e9662f70 | 1156 | parent_inode = pending->dir; |
a22285a6 | 1157 | parent_root = BTRFS_I(parent_inode)->root; |
7585717f | 1158 | record_root_in_trans(trans, parent_root); |
a22285a6 | 1159 | |
3063d29f CM |
1160 | /* |
1161 | * insert the directory item | |
1162 | */ | |
3de4586c | 1163 | ret = btrfs_set_inode_index(parent_inode, &index); |
49b25e05 | 1164 | BUG_ON(ret); /* -ENOMEM */ |
42874b3d MX |
1165 | |
1166 | /* check if there is a file/dir which has the same name. */ | |
1167 | dir_item = btrfs_lookup_dir_item(NULL, parent_root, path, | |
1168 | btrfs_ino(parent_inode), | |
1169 | dentry->d_name.name, | |
1170 | dentry->d_name.len, 0); | |
1171 | if (dir_item != NULL && !IS_ERR(dir_item)) { | |
fe66a05a | 1172 | pending->error = -EEXIST; |
aec8030a | 1173 | goto dir_item_existed; |
42874b3d MX |
1174 | } else if (IS_ERR(dir_item)) { |
1175 | ret = PTR_ERR(dir_item); | |
8732d44f MX |
1176 | btrfs_abort_transaction(trans, root, ret); |
1177 | goto fail; | |
79787eaa | 1178 | } |
42874b3d | 1179 | btrfs_release_path(path); |
52c26179 | 1180 | |
e999376f CM |
1181 | /* |
1182 | * pull in the delayed directory update | |
1183 | * and the delayed inode item | |
1184 | * otherwise we corrupt the FS during | |
1185 | * snapshot | |
1186 | */ | |
1187 | ret = btrfs_run_delayed_items(trans, root); | |
8732d44f MX |
1188 | if (ret) { /* Transaction aborted */ |
1189 | btrfs_abort_transaction(trans, root, ret); | |
1190 | goto fail; | |
1191 | } | |
e999376f | 1192 | |
7585717f | 1193 | record_root_in_trans(trans, root); |
6bdb72de SW |
1194 | btrfs_set_root_last_snapshot(&root->root_item, trans->transid); |
1195 | memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); | |
08fe4db1 | 1196 | btrfs_check_and_init_root_item(new_root_item); |
6bdb72de | 1197 | |
b83cc969 LZ |
1198 | root_flags = btrfs_root_flags(new_root_item); |
1199 | if (pending->readonly) | |
1200 | root_flags |= BTRFS_ROOT_SUBVOL_RDONLY; | |
1201 | else | |
1202 | root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY; | |
1203 | btrfs_set_root_flags(new_root_item, root_flags); | |
1204 | ||
8ea05e3a AB |
1205 | btrfs_set_root_generation_v2(new_root_item, |
1206 | trans->transid); | |
1207 | uuid_le_gen(&new_uuid); | |
1208 | memcpy(new_root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE); | |
1209 | memcpy(new_root_item->parent_uuid, root->root_item.uuid, | |
1210 | BTRFS_UUID_SIZE); | |
70023da2 SB |
1211 | if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) { |
1212 | memset(new_root_item->received_uuid, 0, | |
1213 | sizeof(new_root_item->received_uuid)); | |
1214 | memset(&new_root_item->stime, 0, sizeof(new_root_item->stime)); | |
1215 | memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime)); | |
1216 | btrfs_set_root_stransid(new_root_item, 0); | |
1217 | btrfs_set_root_rtransid(new_root_item, 0); | |
1218 | } | |
8ea05e3a | 1219 | new_root_item->otime.sec = cpu_to_le64(cur_time.tv_sec); |
dadd1105 | 1220 | new_root_item->otime.nsec = cpu_to_le32(cur_time.tv_nsec); |
8ea05e3a | 1221 | btrfs_set_root_otransid(new_root_item, trans->transid); |
8ea05e3a | 1222 | |
6bdb72de | 1223 | old = btrfs_lock_root_node(root); |
49b25e05 | 1224 | ret = btrfs_cow_block(trans, root, old, NULL, 0, &old); |
79787eaa JM |
1225 | if (ret) { |
1226 | btrfs_tree_unlock(old); | |
1227 | free_extent_buffer(old); | |
8732d44f MX |
1228 | btrfs_abort_transaction(trans, root, ret); |
1229 | goto fail; | |
79787eaa | 1230 | } |
49b25e05 | 1231 | |
6bdb72de SW |
1232 | btrfs_set_lock_blocking(old); |
1233 | ||
49b25e05 | 1234 | ret = btrfs_copy_root(trans, root, old, &tmp, objectid); |
79787eaa | 1235 | /* clean up in any case */ |
6bdb72de SW |
1236 | btrfs_tree_unlock(old); |
1237 | free_extent_buffer(old); | |
8732d44f MX |
1238 | if (ret) { |
1239 | btrfs_abort_transaction(trans, root, ret); | |
1240 | goto fail; | |
1241 | } | |
6bdb72de | 1242 | |
f1ebcc74 LB |
1243 | /* see comments in should_cow_block() */ |
1244 | root->force_cow = 1; | |
1245 | smp_wmb(); | |
1246 | ||
6bdb72de | 1247 | btrfs_set_root_node(new_root_item, tmp); |
a22285a6 YZ |
1248 | /* record when the snapshot was created in key.offset */ |
1249 | key.offset = trans->transid; | |
1250 | ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); | |
6bdb72de SW |
1251 | btrfs_tree_unlock(tmp); |
1252 | free_extent_buffer(tmp); | |
8732d44f MX |
1253 | if (ret) { |
1254 | btrfs_abort_transaction(trans, root, ret); | |
1255 | goto fail; | |
1256 | } | |
6bdb72de | 1257 | |
a22285a6 YZ |
1258 | /* |
1259 | * insert root back/forward references | |
1260 | */ | |
1261 | ret = btrfs_add_root_ref(trans, tree_root, objectid, | |
0660b5af | 1262 | parent_root->root_key.objectid, |
33345d01 | 1263 | btrfs_ino(parent_inode), index, |
a22285a6 | 1264 | dentry->d_name.name, dentry->d_name.len); |
8732d44f MX |
1265 | if (ret) { |
1266 | btrfs_abort_transaction(trans, root, ret); | |
1267 | goto fail; | |
1268 | } | |
0660b5af | 1269 | |
a22285a6 YZ |
1270 | key.offset = (u64)-1; |
1271 | pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key); | |
79787eaa JM |
1272 | if (IS_ERR(pending->snap)) { |
1273 | ret = PTR_ERR(pending->snap); | |
8732d44f MX |
1274 | btrfs_abort_transaction(trans, root, ret); |
1275 | goto fail; | |
79787eaa | 1276 | } |
d68fc57b | 1277 | |
49b25e05 | 1278 | ret = btrfs_reloc_post_snapshot(trans, pending); |
8732d44f MX |
1279 | if (ret) { |
1280 | btrfs_abort_transaction(trans, root, ret); | |
1281 | goto fail; | |
1282 | } | |
361048f5 MX |
1283 | |
1284 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); | |
8732d44f MX |
1285 | if (ret) { |
1286 | btrfs_abort_transaction(trans, root, ret); | |
1287 | goto fail; | |
1288 | } | |
42874b3d MX |
1289 | |
1290 | ret = btrfs_insert_dir_item(trans, parent_root, | |
1291 | dentry->d_name.name, dentry->d_name.len, | |
1292 | parent_inode, &key, | |
1293 | BTRFS_FT_DIR, index); | |
1294 | /* We have check then name at the beginning, so it is impossible. */ | |
9c52057c | 1295 | BUG_ON(ret == -EEXIST || ret == -EOVERFLOW); |
8732d44f MX |
1296 | if (ret) { |
1297 | btrfs_abort_transaction(trans, root, ret); | |
1298 | goto fail; | |
1299 | } | |
42874b3d MX |
1300 | |
1301 | btrfs_i_size_write(parent_inode, parent_inode->i_size + | |
1302 | dentry->d_name.len * 2); | |
1303 | parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME; | |
be6aef60 | 1304 | ret = btrfs_update_inode_fallback(trans, parent_root, parent_inode); |
42874b3d | 1305 | if (ret) |
8732d44f | 1306 | btrfs_abort_transaction(trans, root, ret); |
3063d29f | 1307 | fail: |
aec8030a MX |
1308 | pending->error = ret; |
1309 | dir_item_existed: | |
98c9942a | 1310 | trans->block_rsv = rsv; |
2382c5cc | 1311 | trans->bytes_reserved = 0; |
6fa9700e MX |
1312 | no_free_objectid: |
1313 | kfree(new_root_item); | |
1314 | root_item_alloc_fail: | |
42874b3d | 1315 | btrfs_free_path(path); |
49b25e05 | 1316 | return ret; |
3063d29f CM |
1317 | } |
1318 | ||
d352ac68 CM |
1319 | /* |
1320 | * create all the snapshots we've scheduled for creation | |
1321 | */ | |
80b6794d CM |
1322 | static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans, |
1323 | struct btrfs_fs_info *fs_info) | |
3de4586c | 1324 | { |
aec8030a | 1325 | struct btrfs_pending_snapshot *pending, *next; |
3de4586c | 1326 | struct list_head *head = &trans->transaction->pending_snapshots; |
aec8030a | 1327 | int ret = 0; |
3de4586c | 1328 | |
aec8030a MX |
1329 | list_for_each_entry_safe(pending, next, head, list) { |
1330 | list_del(&pending->list); | |
1331 | ret = create_pending_snapshot(trans, fs_info, pending); | |
1332 | if (ret) | |
1333 | break; | |
1334 | } | |
1335 | return ret; | |
3de4586c CM |
1336 | } |
1337 | ||
5d4f98a2 YZ |
1338 | static void update_super_roots(struct btrfs_root *root) |
1339 | { | |
1340 | struct btrfs_root_item *root_item; | |
1341 | struct btrfs_super_block *super; | |
1342 | ||
6c41761f | 1343 | super = root->fs_info->super_copy; |
5d4f98a2 YZ |
1344 | |
1345 | root_item = &root->fs_info->chunk_root->root_item; | |
1346 | super->chunk_root = root_item->bytenr; | |
1347 | super->chunk_root_generation = root_item->generation; | |
1348 | super->chunk_root_level = root_item->level; | |
1349 | ||
1350 | root_item = &root->fs_info->tree_root->root_item; | |
1351 | super->root = root_item->bytenr; | |
1352 | super->generation = root_item->generation; | |
1353 | super->root_level = root_item->level; | |
73bc1876 | 1354 | if (btrfs_test_opt(root, SPACE_CACHE)) |
0af3d00b | 1355 | super->cache_generation = root_item->generation; |
5d4f98a2 YZ |
1356 | } |
1357 | ||
f36f3042 CM |
1358 | int btrfs_transaction_in_commit(struct btrfs_fs_info *info) |
1359 | { | |
4a9d8bde | 1360 | struct btrfs_transaction *trans; |
f36f3042 | 1361 | int ret = 0; |
4a9d8bde | 1362 | |
a4abeea4 | 1363 | spin_lock(&info->trans_lock); |
4a9d8bde MX |
1364 | trans = info->running_transaction; |
1365 | if (trans) | |
1366 | ret = (trans->state >= TRANS_STATE_COMMIT_START); | |
a4abeea4 | 1367 | spin_unlock(&info->trans_lock); |
f36f3042 CM |
1368 | return ret; |
1369 | } | |
1370 | ||
8929ecfa YZ |
1371 | int btrfs_transaction_blocked(struct btrfs_fs_info *info) |
1372 | { | |
4a9d8bde | 1373 | struct btrfs_transaction *trans; |
8929ecfa | 1374 | int ret = 0; |
4a9d8bde | 1375 | |
a4abeea4 | 1376 | spin_lock(&info->trans_lock); |
4a9d8bde MX |
1377 | trans = info->running_transaction; |
1378 | if (trans) | |
1379 | ret = is_transaction_blocked(trans); | |
a4abeea4 | 1380 | spin_unlock(&info->trans_lock); |
8929ecfa YZ |
1381 | return ret; |
1382 | } | |
1383 | ||
bb9c12c9 SW |
1384 | /* |
1385 | * wait for the current transaction commit to start and block subsequent | |
1386 | * transaction joins | |
1387 | */ | |
1388 | static void wait_current_trans_commit_start(struct btrfs_root *root, | |
1389 | struct btrfs_transaction *trans) | |
1390 | { | |
4a9d8bde | 1391 | wait_event(root->fs_info->transaction_blocked_wait, |
501407aa JB |
1392 | trans->state >= TRANS_STATE_COMMIT_START || |
1393 | trans->aborted); | |
bb9c12c9 SW |
1394 | } |
1395 | ||
1396 | /* | |
1397 | * wait for the current transaction to start and then become unblocked. | |
1398 | * caller holds ref. | |
1399 | */ | |
1400 | static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root, | |
1401 | struct btrfs_transaction *trans) | |
1402 | { | |
72d63ed6 | 1403 | wait_event(root->fs_info->transaction_wait, |
501407aa JB |
1404 | trans->state >= TRANS_STATE_UNBLOCKED || |
1405 | trans->aborted); | |
bb9c12c9 SW |
1406 | } |
1407 | ||
1408 | /* | |
1409 | * commit transactions asynchronously. once btrfs_commit_transaction_async | |
1410 | * returns, any subsequent transaction will not be allowed to join. | |
1411 | */ | |
1412 | struct btrfs_async_commit { | |
1413 | struct btrfs_trans_handle *newtrans; | |
1414 | struct btrfs_root *root; | |
7892b5af | 1415 | struct work_struct work; |
bb9c12c9 SW |
1416 | }; |
1417 | ||
1418 | static void do_async_commit(struct work_struct *work) | |
1419 | { | |
1420 | struct btrfs_async_commit *ac = | |
7892b5af | 1421 | container_of(work, struct btrfs_async_commit, work); |
bb9c12c9 | 1422 | |
6fc4e354 SW |
1423 | /* |
1424 | * We've got freeze protection passed with the transaction. | |
1425 | * Tell lockdep about it. | |
1426 | */ | |
ff7c1d33 MX |
1427 | if (ac->newtrans->type < TRANS_JOIN_NOLOCK) |
1428 | rwsem_acquire_read( | |
1429 | &ac->root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1], | |
1430 | 0, 1, _THIS_IP_); | |
6fc4e354 | 1431 | |
e209db7a SW |
1432 | current->journal_info = ac->newtrans; |
1433 | ||
bb9c12c9 SW |
1434 | btrfs_commit_transaction(ac->newtrans, ac->root); |
1435 | kfree(ac); | |
1436 | } | |
1437 | ||
1438 | int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans, | |
1439 | struct btrfs_root *root, | |
1440 | int wait_for_unblock) | |
1441 | { | |
1442 | struct btrfs_async_commit *ac; | |
1443 | struct btrfs_transaction *cur_trans; | |
1444 | ||
1445 | ac = kmalloc(sizeof(*ac), GFP_NOFS); | |
db5b493a TI |
1446 | if (!ac) |
1447 | return -ENOMEM; | |
bb9c12c9 | 1448 | |
7892b5af | 1449 | INIT_WORK(&ac->work, do_async_commit); |
bb9c12c9 | 1450 | ac->root = root; |
7a7eaa40 | 1451 | ac->newtrans = btrfs_join_transaction(root); |
3612b495 TI |
1452 | if (IS_ERR(ac->newtrans)) { |
1453 | int err = PTR_ERR(ac->newtrans); | |
1454 | kfree(ac); | |
1455 | return err; | |
1456 | } | |
bb9c12c9 SW |
1457 | |
1458 | /* take transaction reference */ | |
bb9c12c9 | 1459 | cur_trans = trans->transaction; |
13c5a93e | 1460 | atomic_inc(&cur_trans->use_count); |
bb9c12c9 SW |
1461 | |
1462 | btrfs_end_transaction(trans, root); | |
6fc4e354 SW |
1463 | |
1464 | /* | |
1465 | * Tell lockdep we've released the freeze rwsem, since the | |
1466 | * async commit thread will be the one to unlock it. | |
1467 | */ | |
ff7c1d33 MX |
1468 | if (trans->type < TRANS_JOIN_NOLOCK) |
1469 | rwsem_release( | |
1470 | &root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1], | |
1471 | 1, _THIS_IP_); | |
6fc4e354 | 1472 | |
7892b5af | 1473 | schedule_work(&ac->work); |
bb9c12c9 SW |
1474 | |
1475 | /* wait for transaction to start and unblock */ | |
bb9c12c9 SW |
1476 | if (wait_for_unblock) |
1477 | wait_current_trans_commit_start_and_unblock(root, cur_trans); | |
1478 | else | |
1479 | wait_current_trans_commit_start(root, cur_trans); | |
bb9c12c9 | 1480 | |
38e88054 SW |
1481 | if (current->journal_info == trans) |
1482 | current->journal_info = NULL; | |
1483 | ||
1484 | put_transaction(cur_trans); | |
bb9c12c9 SW |
1485 | return 0; |
1486 | } | |
1487 | ||
49b25e05 JM |
1488 | |
1489 | static void cleanup_transaction(struct btrfs_trans_handle *trans, | |
7b8b92af | 1490 | struct btrfs_root *root, int err) |
49b25e05 JM |
1491 | { |
1492 | struct btrfs_transaction *cur_trans = trans->transaction; | |
f094ac32 | 1493 | DEFINE_WAIT(wait); |
49b25e05 JM |
1494 | |
1495 | WARN_ON(trans->use_count > 1); | |
1496 | ||
7b8b92af JB |
1497 | btrfs_abort_transaction(trans, root, err); |
1498 | ||
49b25e05 | 1499 | spin_lock(&root->fs_info->trans_lock); |
66b6135b | 1500 | |
25d8c284 MX |
1501 | /* |
1502 | * If the transaction is removed from the list, it means this | |
1503 | * transaction has been committed successfully, so it is impossible | |
1504 | * to call the cleanup function. | |
1505 | */ | |
1506 | BUG_ON(list_empty(&cur_trans->list)); | |
66b6135b | 1507 | |
49b25e05 | 1508 | list_del_init(&cur_trans->list); |
d7096fc3 | 1509 | if (cur_trans == root->fs_info->running_transaction) { |
4a9d8bde | 1510 | cur_trans->state = TRANS_STATE_COMMIT_DOING; |
f094ac32 LB |
1511 | spin_unlock(&root->fs_info->trans_lock); |
1512 | wait_event(cur_trans->writer_wait, | |
1513 | atomic_read(&cur_trans->num_writers) == 1); | |
1514 | ||
1515 | spin_lock(&root->fs_info->trans_lock); | |
d7096fc3 | 1516 | } |
49b25e05 JM |
1517 | spin_unlock(&root->fs_info->trans_lock); |
1518 | ||
1519 | btrfs_cleanup_one_transaction(trans->transaction, root); | |
1520 | ||
4a9d8bde MX |
1521 | spin_lock(&root->fs_info->trans_lock); |
1522 | if (cur_trans == root->fs_info->running_transaction) | |
1523 | root->fs_info->running_transaction = NULL; | |
1524 | spin_unlock(&root->fs_info->trans_lock); | |
1525 | ||
49b25e05 JM |
1526 | put_transaction(cur_trans); |
1527 | put_transaction(cur_trans); | |
1528 | ||
1529 | trace_btrfs_transaction_commit(root); | |
1530 | ||
1531 | btrfs_scrub_continue(root); | |
1532 | ||
1533 | if (current->journal_info == trans) | |
1534 | current->journal_info = NULL; | |
1535 | ||
1536 | kmem_cache_free(btrfs_trans_handle_cachep, trans); | |
1537 | } | |
1538 | ||
ca469637 MX |
1539 | static int btrfs_flush_all_pending_stuffs(struct btrfs_trans_handle *trans, |
1540 | struct btrfs_root *root) | |
1541 | { | |
ca469637 MX |
1542 | int ret; |
1543 | ||
ca469637 MX |
1544 | ret = btrfs_run_delayed_items(trans, root); |
1545 | if (ret) | |
1546 | return ret; | |
1547 | ||
1548 | /* | |
1549 | * running the delayed items may have added new refs. account | |
1550 | * them now so that they hinder processing of more delayed refs | |
1551 | * as little as possible. | |
1552 | */ | |
1553 | btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info); | |
1554 | ||
1555 | /* | |
1556 | * rename don't use btrfs_join_transaction, so, once we | |
1557 | * set the transaction to blocked above, we aren't going | |
1558 | * to get any new ordered operations. We can safely run | |
1559 | * it here and no for sure that nothing new will be added | |
1560 | * to the list | |
1561 | */ | |
569e0f35 | 1562 | ret = btrfs_run_ordered_operations(trans, root, 1); |
ca469637 | 1563 | |
eebc6084 | 1564 | return ret; |
ca469637 MX |
1565 | } |
1566 | ||
82436617 MX |
1567 | static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info) |
1568 | { | |
1569 | if (btrfs_test_opt(fs_info->tree_root, FLUSHONCOMMIT)) | |
1570 | return btrfs_start_all_delalloc_inodes(fs_info, 1); | |
1571 | return 0; | |
1572 | } | |
1573 | ||
1574 | static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info) | |
1575 | { | |
1576 | if (btrfs_test_opt(fs_info->tree_root, FLUSHONCOMMIT)) | |
1577 | btrfs_wait_all_ordered_extents(fs_info, 1); | |
1578 | } | |
1579 | ||
79154b1b CM |
1580 | int btrfs_commit_transaction(struct btrfs_trans_handle *trans, |
1581 | struct btrfs_root *root) | |
1582 | { | |
49b25e05 | 1583 | struct btrfs_transaction *cur_trans = trans->transaction; |
8fd17795 | 1584 | struct btrfs_transaction *prev_trans = NULL; |
25287e0a | 1585 | int ret; |
79154b1b | 1586 | |
569e0f35 | 1587 | ret = btrfs_run_ordered_operations(trans, root, 0); |
25287e0a MX |
1588 | if (ret) { |
1589 | btrfs_abort_transaction(trans, root, ret); | |
e4a2bcac JB |
1590 | btrfs_end_transaction(trans, root); |
1591 | return ret; | |
25287e0a | 1592 | } |
5a3f23d5 | 1593 | |
8d25a086 MX |
1594 | /* Stop the commit early if ->aborted is set */ |
1595 | if (unlikely(ACCESS_ONCE(cur_trans->aborted))) { | |
25287e0a | 1596 | ret = cur_trans->aborted; |
e4a2bcac JB |
1597 | btrfs_end_transaction(trans, root); |
1598 | return ret; | |
25287e0a | 1599 | } |
49b25e05 | 1600 | |
56bec294 CM |
1601 | /* make a pass through all the delayed refs we have so far |
1602 | * any runnings procs may add more while we are here | |
1603 | */ | |
1604 | ret = btrfs_run_delayed_refs(trans, root, 0); | |
e4a2bcac JB |
1605 | if (ret) { |
1606 | btrfs_end_transaction(trans, root); | |
1607 | return ret; | |
1608 | } | |
56bec294 | 1609 | |
0e721106 JB |
1610 | btrfs_trans_release_metadata(trans, root); |
1611 | trans->block_rsv = NULL; | |
272d26d0 MX |
1612 | if (trans->qgroup_reserved) { |
1613 | btrfs_qgroup_free(root, trans->qgroup_reserved); | |
1614 | trans->qgroup_reserved = 0; | |
1615 | } | |
0e721106 | 1616 | |
b7ec40d7 | 1617 | cur_trans = trans->transaction; |
49b25e05 | 1618 | |
56bec294 CM |
1619 | /* |
1620 | * set the flushing flag so procs in this transaction have to | |
1621 | * start sending their work down. | |
1622 | */ | |
b7ec40d7 | 1623 | cur_trans->delayed_refs.flushing = 1; |
1be41b78 | 1624 | smp_wmb(); |
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 | } |