]>
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); |
6df9a95e JB |
66 | while (!list_empty(&transaction->pending_chunks)) { |
67 | struct extent_map *em; | |
68 | ||
69 | em = list_first_entry(&transaction->pending_chunks, | |
70 | struct extent_map, list); | |
71 | list_del_init(&em->list); | |
72 | free_extent_map(em); | |
73 | } | |
2c90e5d6 | 74 | kmem_cache_free(btrfs_transaction_cachep, transaction); |
78fae27e | 75 | } |
79154b1b CM |
76 | } |
77 | ||
817d52f8 JB |
78 | static noinline void switch_commit_root(struct btrfs_root *root) |
79 | { | |
817d52f8 JB |
80 | free_extent_buffer(root->commit_root); |
81 | root->commit_root = btrfs_root_node(root); | |
817d52f8 JB |
82 | } |
83 | ||
0860adfd MX |
84 | static inline void extwriter_counter_inc(struct btrfs_transaction *trans, |
85 | unsigned int type) | |
86 | { | |
87 | if (type & TRANS_EXTWRITERS) | |
88 | atomic_inc(&trans->num_extwriters); | |
89 | } | |
90 | ||
91 | static inline void extwriter_counter_dec(struct btrfs_transaction *trans, | |
92 | unsigned int type) | |
93 | { | |
94 | if (type & TRANS_EXTWRITERS) | |
95 | atomic_dec(&trans->num_extwriters); | |
96 | } | |
97 | ||
98 | static inline void extwriter_counter_init(struct btrfs_transaction *trans, | |
99 | unsigned int type) | |
100 | { | |
101 | atomic_set(&trans->num_extwriters, ((type & TRANS_EXTWRITERS) ? 1 : 0)); | |
102 | } | |
103 | ||
104 | static inline int extwriter_counter_read(struct btrfs_transaction *trans) | |
105 | { | |
106 | return atomic_read(&trans->num_extwriters); | |
178260b2 MX |
107 | } |
108 | ||
d352ac68 CM |
109 | /* |
110 | * either allocate a new transaction or hop into the existing one | |
111 | */ | |
0860adfd | 112 | static noinline int join_transaction(struct btrfs_root *root, unsigned int type) |
79154b1b CM |
113 | { |
114 | struct btrfs_transaction *cur_trans; | |
19ae4e81 | 115 | struct btrfs_fs_info *fs_info = root->fs_info; |
a4abeea4 | 116 | |
19ae4e81 | 117 | spin_lock(&fs_info->trans_lock); |
d43317dc | 118 | loop: |
49b25e05 | 119 | /* The file system has been taken offline. No new transactions. */ |
87533c47 | 120 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { |
19ae4e81 | 121 | spin_unlock(&fs_info->trans_lock); |
49b25e05 JM |
122 | return -EROFS; |
123 | } | |
124 | ||
19ae4e81 | 125 | cur_trans = fs_info->running_transaction; |
a4abeea4 | 126 | if (cur_trans) { |
871383be | 127 | if (cur_trans->aborted) { |
19ae4e81 | 128 | spin_unlock(&fs_info->trans_lock); |
49b25e05 | 129 | return cur_trans->aborted; |
871383be | 130 | } |
4a9d8bde | 131 | if (btrfs_blocked_trans_types[cur_trans->state] & type) { |
178260b2 MX |
132 | spin_unlock(&fs_info->trans_lock); |
133 | return -EBUSY; | |
134 | } | |
a4abeea4 | 135 | atomic_inc(&cur_trans->use_count); |
13c5a93e | 136 | atomic_inc(&cur_trans->num_writers); |
0860adfd | 137 | extwriter_counter_inc(cur_trans, type); |
19ae4e81 | 138 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 | 139 | return 0; |
79154b1b | 140 | } |
19ae4e81 | 141 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 | 142 | |
354aa0fb MX |
143 | /* |
144 | * If we are ATTACH, we just want to catch the current transaction, | |
145 | * and commit it. If there is no transaction, just return ENOENT. | |
146 | */ | |
147 | if (type == TRANS_ATTACH) | |
148 | return -ENOENT; | |
149 | ||
4a9d8bde MX |
150 | /* |
151 | * JOIN_NOLOCK only happens during the transaction commit, so | |
152 | * it is impossible that ->running_transaction is NULL | |
153 | */ | |
154 | BUG_ON(type == TRANS_JOIN_NOLOCK); | |
155 | ||
a4abeea4 JB |
156 | cur_trans = kmem_cache_alloc(btrfs_transaction_cachep, GFP_NOFS); |
157 | if (!cur_trans) | |
158 | return -ENOMEM; | |
d43317dc | 159 | |
19ae4e81 JS |
160 | spin_lock(&fs_info->trans_lock); |
161 | if (fs_info->running_transaction) { | |
d43317dc CM |
162 | /* |
163 | * someone started a transaction after we unlocked. Make sure | |
4a9d8bde | 164 | * to redo the checks above |
d43317dc | 165 | */ |
a4abeea4 | 166 | kmem_cache_free(btrfs_transaction_cachep, cur_trans); |
d43317dc | 167 | goto loop; |
87533c47 | 168 | } else if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { |
e4b50e14 | 169 | spin_unlock(&fs_info->trans_lock); |
7b8b92af JB |
170 | kmem_cache_free(btrfs_transaction_cachep, cur_trans); |
171 | return -EROFS; | |
79154b1b | 172 | } |
d43317dc | 173 | |
a4abeea4 | 174 | atomic_set(&cur_trans->num_writers, 1); |
0860adfd | 175 | extwriter_counter_init(cur_trans, type); |
a4abeea4 JB |
176 | init_waitqueue_head(&cur_trans->writer_wait); |
177 | init_waitqueue_head(&cur_trans->commit_wait); | |
4a9d8bde | 178 | cur_trans->state = TRANS_STATE_RUNNING; |
a4abeea4 JB |
179 | /* |
180 | * One for this trans handle, one so it will live on until we | |
181 | * commit the transaction. | |
182 | */ | |
183 | atomic_set(&cur_trans->use_count, 2); | |
a4abeea4 JB |
184 | cur_trans->start_time = get_seconds(); |
185 | ||
186 | cur_trans->delayed_refs.root = RB_ROOT; | |
187 | cur_trans->delayed_refs.num_entries = 0; | |
188 | cur_trans->delayed_refs.num_heads_ready = 0; | |
189 | cur_trans->delayed_refs.num_heads = 0; | |
190 | cur_trans->delayed_refs.flushing = 0; | |
191 | cur_trans->delayed_refs.run_delayed_start = 0; | |
20b297d6 JS |
192 | |
193 | /* | |
194 | * although the tree mod log is per file system and not per transaction, | |
195 | * the log must never go across transaction boundaries. | |
196 | */ | |
197 | smp_mb(); | |
31b1a2bd JL |
198 | if (!list_empty(&fs_info->tree_mod_seq_list)) |
199 | WARN(1, KERN_ERR "btrfs: tree_mod_seq_list not empty when " | |
20b297d6 | 200 | "creating a fresh transaction\n"); |
31b1a2bd JL |
201 | if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log)) |
202 | WARN(1, KERN_ERR "btrfs: tree_mod_log rb tree not empty when " | |
20b297d6 | 203 | "creating a fresh transaction\n"); |
fc36ed7e | 204 | atomic64_set(&fs_info->tree_mod_seq, 0); |
20b297d6 | 205 | |
a4abeea4 | 206 | spin_lock_init(&cur_trans->delayed_refs.lock); |
bb721703 CM |
207 | atomic_set(&cur_trans->delayed_refs.procs_running_refs, 0); |
208 | atomic_set(&cur_trans->delayed_refs.ref_seq, 0); | |
209 | init_waitqueue_head(&cur_trans->delayed_refs.wait); | |
a4abeea4 JB |
210 | |
211 | INIT_LIST_HEAD(&cur_trans->pending_snapshots); | |
569e0f35 | 212 | INIT_LIST_HEAD(&cur_trans->ordered_operations); |
6df9a95e | 213 | INIT_LIST_HEAD(&cur_trans->pending_chunks); |
19ae4e81 | 214 | list_add_tail(&cur_trans->list, &fs_info->trans_list); |
a4abeea4 | 215 | extent_io_tree_init(&cur_trans->dirty_pages, |
19ae4e81 JS |
216 | fs_info->btree_inode->i_mapping); |
217 | fs_info->generation++; | |
218 | cur_trans->transid = fs_info->generation; | |
219 | fs_info->running_transaction = cur_trans; | |
49b25e05 | 220 | cur_trans->aborted = 0; |
19ae4e81 | 221 | spin_unlock(&fs_info->trans_lock); |
15ee9bc7 | 222 | |
79154b1b CM |
223 | return 0; |
224 | } | |
225 | ||
d352ac68 | 226 | /* |
d397712b CM |
227 | * this does all the record keeping required to make sure that a reference |
228 | * counted root is properly recorded in a given transaction. This is required | |
229 | * to make sure the old root from before we joined the transaction is deleted | |
230 | * when the transaction commits | |
d352ac68 | 231 | */ |
7585717f | 232 | static int record_root_in_trans(struct btrfs_trans_handle *trans, |
a4abeea4 | 233 | struct btrfs_root *root) |
6702ed49 | 234 | { |
5d4f98a2 | 235 | if (root->ref_cows && root->last_trans < trans->transid) { |
6702ed49 | 236 | WARN_ON(root == root->fs_info->extent_root); |
5d4f98a2 YZ |
237 | WARN_ON(root->commit_root != root->node); |
238 | ||
7585717f CM |
239 | /* |
240 | * see below for in_trans_setup usage rules | |
241 | * we have the reloc mutex held now, so there | |
242 | * is only one writer in this function | |
243 | */ | |
244 | root->in_trans_setup = 1; | |
245 | ||
246 | /* make sure readers find in_trans_setup before | |
247 | * they find our root->last_trans update | |
248 | */ | |
249 | smp_wmb(); | |
250 | ||
a4abeea4 JB |
251 | spin_lock(&root->fs_info->fs_roots_radix_lock); |
252 | if (root->last_trans == trans->transid) { | |
253 | spin_unlock(&root->fs_info->fs_roots_radix_lock); | |
254 | return 0; | |
255 | } | |
5d4f98a2 YZ |
256 | radix_tree_tag_set(&root->fs_info->fs_roots_radix, |
257 | (unsigned long)root->root_key.objectid, | |
258 | BTRFS_ROOT_TRANS_TAG); | |
a4abeea4 | 259 | spin_unlock(&root->fs_info->fs_roots_radix_lock); |
7585717f CM |
260 | root->last_trans = trans->transid; |
261 | ||
262 | /* this is pretty tricky. We don't want to | |
263 | * take the relocation lock in btrfs_record_root_in_trans | |
264 | * unless we're really doing the first setup for this root in | |
265 | * this transaction. | |
266 | * | |
267 | * Normally we'd use root->last_trans as a flag to decide | |
268 | * if we want to take the expensive mutex. | |
269 | * | |
270 | * But, we have to set root->last_trans before we | |
271 | * init the relocation root, otherwise, we trip over warnings | |
272 | * in ctree.c. The solution used here is to flag ourselves | |
273 | * with root->in_trans_setup. When this is 1, we're still | |
274 | * fixing up the reloc trees and everyone must wait. | |
275 | * | |
276 | * When this is zero, they can trust root->last_trans and fly | |
277 | * through btrfs_record_root_in_trans without having to take the | |
278 | * lock. smp_wmb() makes sure that all the writes above are | |
279 | * done before we pop in the zero below | |
280 | */ | |
5d4f98a2 | 281 | btrfs_init_reloc_root(trans, root); |
7585717f CM |
282 | smp_wmb(); |
283 | root->in_trans_setup = 0; | |
5d4f98a2 YZ |
284 | } |
285 | return 0; | |
286 | } | |
bcc63abb | 287 | |
7585717f CM |
288 | |
289 | int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, | |
290 | struct btrfs_root *root) | |
291 | { | |
292 | if (!root->ref_cows) | |
293 | return 0; | |
294 | ||
295 | /* | |
296 | * see record_root_in_trans for comments about in_trans_setup usage | |
297 | * and barriers | |
298 | */ | |
299 | smp_rmb(); | |
300 | if (root->last_trans == trans->transid && | |
301 | !root->in_trans_setup) | |
302 | return 0; | |
303 | ||
304 | mutex_lock(&root->fs_info->reloc_mutex); | |
305 | record_root_in_trans(trans, root); | |
306 | mutex_unlock(&root->fs_info->reloc_mutex); | |
307 | ||
308 | return 0; | |
309 | } | |
310 | ||
4a9d8bde MX |
311 | static inline int is_transaction_blocked(struct btrfs_transaction *trans) |
312 | { | |
313 | return (trans->state >= TRANS_STATE_BLOCKED && | |
501407aa JB |
314 | trans->state < TRANS_STATE_UNBLOCKED && |
315 | !trans->aborted); | |
4a9d8bde MX |
316 | } |
317 | ||
d352ac68 CM |
318 | /* wait for commit against the current transaction to become unblocked |
319 | * when this is done, it is safe to start a new transaction, but the current | |
320 | * transaction might not be fully on disk. | |
321 | */ | |
37d1aeee | 322 | static void wait_current_trans(struct btrfs_root *root) |
79154b1b | 323 | { |
f9295749 | 324 | struct btrfs_transaction *cur_trans; |
79154b1b | 325 | |
a4abeea4 | 326 | spin_lock(&root->fs_info->trans_lock); |
f9295749 | 327 | cur_trans = root->fs_info->running_transaction; |
4a9d8bde | 328 | if (cur_trans && is_transaction_blocked(cur_trans)) { |
13c5a93e | 329 | atomic_inc(&cur_trans->use_count); |
a4abeea4 | 330 | spin_unlock(&root->fs_info->trans_lock); |
72d63ed6 LZ |
331 | |
332 | wait_event(root->fs_info->transaction_wait, | |
501407aa JB |
333 | cur_trans->state >= TRANS_STATE_UNBLOCKED || |
334 | cur_trans->aborted); | |
f9295749 | 335 | put_transaction(cur_trans); |
a4abeea4 JB |
336 | } else { |
337 | spin_unlock(&root->fs_info->trans_lock); | |
f9295749 | 338 | } |
37d1aeee CM |
339 | } |
340 | ||
a22285a6 YZ |
341 | static int may_wait_transaction(struct btrfs_root *root, int type) |
342 | { | |
a4abeea4 JB |
343 | if (root->fs_info->log_root_recovering) |
344 | return 0; | |
345 | ||
346 | if (type == TRANS_USERSPACE) | |
347 | return 1; | |
348 | ||
349 | if (type == TRANS_START && | |
350 | !atomic_read(&root->fs_info->open_ioctl_trans)) | |
a22285a6 | 351 | return 1; |
a4abeea4 | 352 | |
a22285a6 YZ |
353 | return 0; |
354 | } | |
355 | ||
08e007d2 | 356 | static struct btrfs_trans_handle * |
0860adfd | 357 | start_transaction(struct btrfs_root *root, u64 num_items, unsigned int type, |
08e007d2 | 358 | enum btrfs_reserve_flush_enum flush) |
37d1aeee | 359 | { |
a22285a6 YZ |
360 | struct btrfs_trans_handle *h; |
361 | struct btrfs_transaction *cur_trans; | |
b5009945 | 362 | u64 num_bytes = 0; |
37d1aeee | 363 | int ret; |
c5567237 | 364 | u64 qgroup_reserved = 0; |
acce952b | 365 | |
87533c47 | 366 | if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) |
acce952b | 367 | return ERR_PTR(-EROFS); |
2a1eb461 JB |
368 | |
369 | if (current->journal_info) { | |
0860adfd | 370 | WARN_ON(type & TRANS_EXTWRITERS); |
2a1eb461 JB |
371 | h = current->journal_info; |
372 | h->use_count++; | |
b7d5b0a8 | 373 | WARN_ON(h->use_count > 2); |
2a1eb461 JB |
374 | h->orig_rsv = h->block_rsv; |
375 | h->block_rsv = NULL; | |
376 | goto got_it; | |
377 | } | |
b5009945 JB |
378 | |
379 | /* | |
380 | * Do the reservation before we join the transaction so we can do all | |
381 | * the appropriate flushing if need be. | |
382 | */ | |
383 | if (num_items > 0 && root != root->fs_info->chunk_root) { | |
c5567237 AJ |
384 | if (root->fs_info->quota_enabled && |
385 | is_fstree(root->root_key.objectid)) { | |
386 | qgroup_reserved = num_items * root->leafsize; | |
387 | ret = btrfs_qgroup_reserve(root, qgroup_reserved); | |
388 | if (ret) | |
389 | return ERR_PTR(ret); | |
390 | } | |
391 | ||
b5009945 | 392 | num_bytes = btrfs_calc_trans_metadata_size(root, num_items); |
08e007d2 MX |
393 | ret = btrfs_block_rsv_add(root, |
394 | &root->fs_info->trans_block_rsv, | |
395 | num_bytes, flush); | |
b5009945 | 396 | if (ret) |
843fcf35 | 397 | goto reserve_fail; |
b5009945 | 398 | } |
a22285a6 YZ |
399 | again: |
400 | h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS); | |
843fcf35 MX |
401 | if (!h) { |
402 | ret = -ENOMEM; | |
403 | goto alloc_fail; | |
404 | } | |
37d1aeee | 405 | |
98114659 JB |
406 | /* |
407 | * If we are JOIN_NOLOCK we're already committing a transaction and | |
408 | * waiting on this guy, so we don't need to do the sb_start_intwrite | |
409 | * because we're already holding a ref. We need this because we could | |
410 | * have raced in and did an fsync() on a file which can kick a commit | |
411 | * and then we deadlock with somebody doing a freeze. | |
354aa0fb MX |
412 | * |
413 | * If we are ATTACH, it means we just want to catch the current | |
414 | * transaction and commit it, so we needn't do sb_start_intwrite(). | |
98114659 | 415 | */ |
0860adfd | 416 | if (type & __TRANS_FREEZABLE) |
60376ce4 | 417 | sb_start_intwrite(root->fs_info->sb); |
b2b5ef5c | 418 | |
a22285a6 | 419 | if (may_wait_transaction(root, type)) |
37d1aeee | 420 | wait_current_trans(root); |
a22285a6 | 421 | |
a4abeea4 | 422 | do { |
354aa0fb | 423 | ret = join_transaction(root, type); |
178260b2 | 424 | if (ret == -EBUSY) { |
a4abeea4 | 425 | wait_current_trans(root); |
178260b2 MX |
426 | if (unlikely(type == TRANS_ATTACH)) |
427 | ret = -ENOENT; | |
428 | } | |
a4abeea4 JB |
429 | } while (ret == -EBUSY); |
430 | ||
db5b493a | 431 | if (ret < 0) { |
354aa0fb MX |
432 | /* We must get the transaction if we are JOIN_NOLOCK. */ |
433 | BUG_ON(type == TRANS_JOIN_NOLOCK); | |
843fcf35 | 434 | goto join_fail; |
db5b493a | 435 | } |
0f7d52f4 | 436 | |
a22285a6 | 437 | cur_trans = root->fs_info->running_transaction; |
a22285a6 YZ |
438 | |
439 | h->transid = cur_trans->transid; | |
440 | h->transaction = cur_trans; | |
79154b1b | 441 | h->blocks_used = 0; |
a22285a6 | 442 | h->bytes_reserved = 0; |
d13603ef | 443 | h->root = root; |
56bec294 | 444 | h->delayed_ref_updates = 0; |
2a1eb461 | 445 | h->use_count = 1; |
0e721106 | 446 | h->adding_csums = 0; |
f0486c68 | 447 | h->block_rsv = NULL; |
2a1eb461 | 448 | h->orig_rsv = NULL; |
49b25e05 | 449 | h->aborted = 0; |
4b824906 | 450 | h->qgroup_reserved = 0; |
bed92eae | 451 | h->delayed_ref_elem.seq = 0; |
a698d075 | 452 | h->type = type; |
c6b305a8 | 453 | h->allocating_chunk = false; |
bed92eae | 454 | INIT_LIST_HEAD(&h->qgroup_ref_list); |
ea658bad | 455 | INIT_LIST_HEAD(&h->new_bgs); |
b7ec40d7 | 456 | |
a22285a6 | 457 | smp_mb(); |
4a9d8bde MX |
458 | if (cur_trans->state >= TRANS_STATE_BLOCKED && |
459 | may_wait_transaction(root, type)) { | |
a22285a6 YZ |
460 | btrfs_commit_transaction(h, root); |
461 | goto again; | |
462 | } | |
463 | ||
b5009945 | 464 | if (num_bytes) { |
8c2a3ca2 | 465 | trace_btrfs_space_reservation(root->fs_info, "transaction", |
2bcc0328 | 466 | h->transid, num_bytes, 1); |
b5009945 JB |
467 | h->block_rsv = &root->fs_info->trans_block_rsv; |
468 | h->bytes_reserved = num_bytes; | |
a22285a6 | 469 | } |
4b824906 | 470 | h->qgroup_reserved = qgroup_reserved; |
9ed74f2d | 471 | |
2a1eb461 | 472 | got_it: |
a4abeea4 | 473 | btrfs_record_root_in_trans(h, root); |
a22285a6 YZ |
474 | |
475 | if (!current->journal_info && type != TRANS_USERSPACE) | |
476 | current->journal_info = h; | |
79154b1b | 477 | return h; |
843fcf35 MX |
478 | |
479 | join_fail: | |
0860adfd | 480 | if (type & __TRANS_FREEZABLE) |
843fcf35 MX |
481 | sb_end_intwrite(root->fs_info->sb); |
482 | kmem_cache_free(btrfs_trans_handle_cachep, h); | |
483 | alloc_fail: | |
484 | if (num_bytes) | |
485 | btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv, | |
486 | num_bytes); | |
487 | reserve_fail: | |
488 | if (qgroup_reserved) | |
489 | btrfs_qgroup_free(root, qgroup_reserved); | |
490 | return ERR_PTR(ret); | |
79154b1b CM |
491 | } |
492 | ||
f9295749 | 493 | struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, |
a22285a6 | 494 | int num_items) |
f9295749 | 495 | { |
08e007d2 MX |
496 | return start_transaction(root, num_items, TRANS_START, |
497 | BTRFS_RESERVE_FLUSH_ALL); | |
f9295749 | 498 | } |
8407aa46 | 499 | |
08e007d2 | 500 | struct btrfs_trans_handle *btrfs_start_transaction_lflush( |
8407aa46 MX |
501 | struct btrfs_root *root, int num_items) |
502 | { | |
08e007d2 MX |
503 | return start_transaction(root, num_items, TRANS_START, |
504 | BTRFS_RESERVE_FLUSH_LIMIT); | |
8407aa46 MX |
505 | } |
506 | ||
7a7eaa40 | 507 | struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root) |
f9295749 | 508 | { |
8407aa46 | 509 | return start_transaction(root, 0, TRANS_JOIN, 0); |
f9295749 CM |
510 | } |
511 | ||
7a7eaa40 | 512 | struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root) |
0af3d00b | 513 | { |
8407aa46 | 514 | return start_transaction(root, 0, TRANS_JOIN_NOLOCK, 0); |
0af3d00b JB |
515 | } |
516 | ||
7a7eaa40 | 517 | struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root) |
9ca9ee09 | 518 | { |
8407aa46 | 519 | return start_transaction(root, 0, TRANS_USERSPACE, 0); |
9ca9ee09 SW |
520 | } |
521 | ||
d4edf39b MX |
522 | /* |
523 | * btrfs_attach_transaction() - catch the running transaction | |
524 | * | |
525 | * It is used when we want to commit the current the transaction, but | |
526 | * don't want to start a new one. | |
527 | * | |
528 | * Note: If this function return -ENOENT, it just means there is no | |
529 | * running transaction. But it is possible that the inactive transaction | |
530 | * is still in the memory, not fully on disk. If you hope there is no | |
531 | * inactive transaction in the fs when -ENOENT is returned, you should | |
532 | * invoke | |
533 | * btrfs_attach_transaction_barrier() | |
534 | */ | |
354aa0fb | 535 | struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root) |
60376ce4 | 536 | { |
354aa0fb | 537 | return start_transaction(root, 0, TRANS_ATTACH, 0); |
60376ce4 JB |
538 | } |
539 | ||
d4edf39b | 540 | /* |
90b6d283 | 541 | * btrfs_attach_transaction_barrier() - catch the running transaction |
d4edf39b MX |
542 | * |
543 | * It is similar to the above function, the differentia is this one | |
544 | * will wait for all the inactive transactions until they fully | |
545 | * complete. | |
546 | */ | |
547 | struct btrfs_trans_handle * | |
548 | btrfs_attach_transaction_barrier(struct btrfs_root *root) | |
549 | { | |
550 | struct btrfs_trans_handle *trans; | |
551 | ||
552 | trans = start_transaction(root, 0, TRANS_ATTACH, 0); | |
553 | if (IS_ERR(trans) && PTR_ERR(trans) == -ENOENT) | |
554 | btrfs_wait_for_commit(root, 0); | |
555 | ||
556 | return trans; | |
557 | } | |
558 | ||
d352ac68 | 559 | /* wait for a transaction commit to be fully complete */ |
b9c8300c | 560 | static noinline void wait_for_commit(struct btrfs_root *root, |
89ce8a63 CM |
561 | struct btrfs_transaction *commit) |
562 | { | |
4a9d8bde | 563 | wait_event(commit->commit_wait, commit->state == TRANS_STATE_COMPLETED); |
89ce8a63 CM |
564 | } |
565 | ||
46204592 SW |
566 | int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid) |
567 | { | |
568 | struct btrfs_transaction *cur_trans = NULL, *t; | |
8cd2807f | 569 | int ret = 0; |
46204592 | 570 | |
46204592 SW |
571 | if (transid) { |
572 | if (transid <= root->fs_info->last_trans_committed) | |
a4abeea4 | 573 | goto out; |
46204592 | 574 | |
8cd2807f | 575 | ret = -EINVAL; |
46204592 | 576 | /* find specified transaction */ |
a4abeea4 | 577 | spin_lock(&root->fs_info->trans_lock); |
46204592 SW |
578 | list_for_each_entry(t, &root->fs_info->trans_list, list) { |
579 | if (t->transid == transid) { | |
580 | cur_trans = t; | |
a4abeea4 | 581 | atomic_inc(&cur_trans->use_count); |
8cd2807f | 582 | ret = 0; |
46204592 SW |
583 | break; |
584 | } | |
8cd2807f MX |
585 | if (t->transid > transid) { |
586 | ret = 0; | |
46204592 | 587 | break; |
8cd2807f | 588 | } |
46204592 | 589 | } |
a4abeea4 | 590 | spin_unlock(&root->fs_info->trans_lock); |
8cd2807f | 591 | /* The specified transaction doesn't exist */ |
46204592 | 592 | if (!cur_trans) |
8cd2807f | 593 | goto out; |
46204592 SW |
594 | } else { |
595 | /* find newest transaction that is committing | committed */ | |
a4abeea4 | 596 | spin_lock(&root->fs_info->trans_lock); |
46204592 SW |
597 | list_for_each_entry_reverse(t, &root->fs_info->trans_list, |
598 | list) { | |
4a9d8bde MX |
599 | if (t->state >= TRANS_STATE_COMMIT_START) { |
600 | if (t->state == TRANS_STATE_COMPLETED) | |
3473f3c0 | 601 | break; |
46204592 | 602 | cur_trans = t; |
a4abeea4 | 603 | atomic_inc(&cur_trans->use_count); |
46204592 SW |
604 | break; |
605 | } | |
606 | } | |
a4abeea4 | 607 | spin_unlock(&root->fs_info->trans_lock); |
46204592 | 608 | if (!cur_trans) |
a4abeea4 | 609 | goto out; /* nothing committing|committed */ |
46204592 SW |
610 | } |
611 | ||
46204592 | 612 | wait_for_commit(root, cur_trans); |
46204592 | 613 | put_transaction(cur_trans); |
a4abeea4 | 614 | out: |
46204592 SW |
615 | return ret; |
616 | } | |
617 | ||
37d1aeee CM |
618 | void btrfs_throttle(struct btrfs_root *root) |
619 | { | |
a4abeea4 | 620 | if (!atomic_read(&root->fs_info->open_ioctl_trans)) |
9ca9ee09 | 621 | wait_current_trans(root); |
37d1aeee CM |
622 | } |
623 | ||
8929ecfa YZ |
624 | static int should_end_transaction(struct btrfs_trans_handle *trans, |
625 | struct btrfs_root *root) | |
626 | { | |
1be41b78 JB |
627 | if (root->fs_info->global_block_rsv.space_info->full && |
628 | btrfs_should_throttle_delayed_refs(trans, root)) | |
629 | return 1; | |
36ba022a | 630 | |
1be41b78 | 631 | return !!btrfs_block_rsv_check(root, &root->fs_info->global_block_rsv, 5); |
8929ecfa YZ |
632 | } |
633 | ||
634 | int btrfs_should_end_transaction(struct btrfs_trans_handle *trans, | |
635 | struct btrfs_root *root) | |
636 | { | |
637 | struct btrfs_transaction *cur_trans = trans->transaction; | |
638 | int updates; | |
49b25e05 | 639 | int err; |
8929ecfa | 640 | |
a4abeea4 | 641 | smp_mb(); |
4a9d8bde MX |
642 | if (cur_trans->state >= TRANS_STATE_BLOCKED || |
643 | cur_trans->delayed_refs.flushing) | |
8929ecfa YZ |
644 | return 1; |
645 | ||
646 | updates = trans->delayed_ref_updates; | |
647 | trans->delayed_ref_updates = 0; | |
49b25e05 JM |
648 | if (updates) { |
649 | err = btrfs_run_delayed_refs(trans, root, updates); | |
650 | if (err) /* Error code will also eval true */ | |
651 | return err; | |
652 | } | |
8929ecfa YZ |
653 | |
654 | return should_end_transaction(trans, root); | |
655 | } | |
656 | ||
89ce8a63 | 657 | static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, |
a698d075 | 658 | struct btrfs_root *root, int throttle) |
79154b1b | 659 | { |
8929ecfa | 660 | struct btrfs_transaction *cur_trans = trans->transaction; |
ab78c84d | 661 | struct btrfs_fs_info *info = root->fs_info; |
1be41b78 | 662 | unsigned long cur = trans->delayed_ref_updates; |
a698d075 | 663 | int lock = (trans->type != TRANS_JOIN_NOLOCK); |
4edc2ca3 | 664 | int err = 0; |
c3e69d58 | 665 | |
2a1eb461 JB |
666 | if (--trans->use_count) { |
667 | trans->block_rsv = trans->orig_rsv; | |
668 | return 0; | |
669 | } | |
670 | ||
edf39272 JS |
671 | /* |
672 | * do the qgroup accounting as early as possible | |
673 | */ | |
674 | err = btrfs_delayed_refs_qgroup_accounting(trans, info); | |
675 | ||
b24e03db | 676 | btrfs_trans_release_metadata(trans, root); |
4c13d758 | 677 | trans->block_rsv = NULL; |
c5567237 AJ |
678 | |
679 | if (trans->qgroup_reserved) { | |
7c2ec3f0 LB |
680 | /* |
681 | * the same root has to be passed here between start_transaction | |
682 | * and end_transaction. Subvolume quota depends on this. | |
683 | */ | |
684 | btrfs_qgroup_free(trans->root, trans->qgroup_reserved); | |
c5567237 AJ |
685 | trans->qgroup_reserved = 0; |
686 | } | |
687 | ||
ea658bad JB |
688 | if (!list_empty(&trans->new_bgs)) |
689 | btrfs_create_pending_block_groups(trans, root); | |
690 | ||
1be41b78 JB |
691 | trans->delayed_ref_updates = 0; |
692 | if (btrfs_should_throttle_delayed_refs(trans, root)) { | |
693 | cur = max_t(unsigned long, cur, 1); | |
c3e69d58 | 694 | trans->delayed_ref_updates = 0; |
1be41b78 | 695 | btrfs_run_delayed_refs(trans, root, cur); |
56bec294 | 696 | } |
bb721703 | 697 | |
0e721106 JB |
698 | btrfs_trans_release_metadata(trans, root); |
699 | trans->block_rsv = NULL; | |
56bec294 | 700 | |
ea658bad JB |
701 | if (!list_empty(&trans->new_bgs)) |
702 | btrfs_create_pending_block_groups(trans, root); | |
703 | ||
a4abeea4 | 704 | if (lock && !atomic_read(&root->fs_info->open_ioctl_trans) && |
4a9d8bde MX |
705 | should_end_transaction(trans, root) && |
706 | ACCESS_ONCE(cur_trans->state) == TRANS_STATE_RUNNING) { | |
707 | spin_lock(&info->trans_lock); | |
708 | if (cur_trans->state == TRANS_STATE_RUNNING) | |
709 | cur_trans->state = TRANS_STATE_BLOCKED; | |
710 | spin_unlock(&info->trans_lock); | |
a4abeea4 | 711 | } |
8929ecfa | 712 | |
4a9d8bde | 713 | if (lock && ACCESS_ONCE(cur_trans->state) == TRANS_STATE_BLOCKED) { |
81317fde JB |
714 | if (throttle) { |
715 | /* | |
716 | * We may race with somebody else here so end up having | |
717 | * to call end_transaction on ourselves again, so inc | |
718 | * our use_count. | |
719 | */ | |
720 | trans->use_count++; | |
8929ecfa | 721 | return btrfs_commit_transaction(trans, root); |
81317fde | 722 | } else { |
8929ecfa | 723 | wake_up_process(info->transaction_kthread); |
81317fde | 724 | } |
8929ecfa YZ |
725 | } |
726 | ||
0860adfd | 727 | if (trans->type & __TRANS_FREEZABLE) |
98114659 | 728 | sb_end_intwrite(root->fs_info->sb); |
6df7881a | 729 | |
8929ecfa | 730 | WARN_ON(cur_trans != info->running_transaction); |
13c5a93e JB |
731 | WARN_ON(atomic_read(&cur_trans->num_writers) < 1); |
732 | atomic_dec(&cur_trans->num_writers); | |
0860adfd | 733 | extwriter_counter_dec(cur_trans, trans->type); |
89ce8a63 | 734 | |
99d16cbc | 735 | smp_mb(); |
79154b1b CM |
736 | if (waitqueue_active(&cur_trans->writer_wait)) |
737 | wake_up(&cur_trans->writer_wait); | |
79154b1b | 738 | put_transaction(cur_trans); |
9ed74f2d JB |
739 | |
740 | if (current->journal_info == trans) | |
741 | current->journal_info = NULL; | |
ab78c84d | 742 | |
24bbcf04 YZ |
743 | if (throttle) |
744 | btrfs_run_delayed_iputs(root); | |
745 | ||
49b25e05 | 746 | if (trans->aborted || |
4e121c06 JB |
747 | test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) { |
748 | wake_up_process(info->transaction_kthread); | |
4edc2ca3 | 749 | err = -EIO; |
4e121c06 | 750 | } |
edf39272 | 751 | assert_qgroups_uptodate(trans); |
49b25e05 | 752 | |
4edc2ca3 DJ |
753 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
754 | return err; | |
79154b1b CM |
755 | } |
756 | ||
89ce8a63 CM |
757 | int btrfs_end_transaction(struct btrfs_trans_handle *trans, |
758 | struct btrfs_root *root) | |
759 | { | |
98ad43be | 760 | return __btrfs_end_transaction(trans, root, 0); |
89ce8a63 CM |
761 | } |
762 | ||
763 | int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans, | |
764 | struct btrfs_root *root) | |
765 | { | |
98ad43be | 766 | return __btrfs_end_transaction(trans, root, 1); |
16cdcec7 MX |
767 | } |
768 | ||
769 | int btrfs_end_transaction_dmeta(struct btrfs_trans_handle *trans, | |
770 | struct btrfs_root *root) | |
771 | { | |
a698d075 | 772 | return __btrfs_end_transaction(trans, root, 1); |
89ce8a63 CM |
773 | } |
774 | ||
d352ac68 CM |
775 | /* |
776 | * when btree blocks are allocated, they have some corresponding bits set for | |
777 | * them in one of two extent_io trees. This is used to make sure all of | |
690587d1 | 778 | * those extents are sent to disk but does not wait on them |
d352ac68 | 779 | */ |
690587d1 | 780 | int btrfs_write_marked_extents(struct btrfs_root *root, |
8cef4e16 | 781 | struct extent_io_tree *dirty_pages, int mark) |
79154b1b | 782 | { |
777e6bd7 | 783 | int err = 0; |
7c4452b9 | 784 | int werr = 0; |
1728366e | 785 | struct address_space *mapping = root->fs_info->btree_inode->i_mapping; |
e6138876 | 786 | struct extent_state *cached_state = NULL; |
777e6bd7 | 787 | u64 start = 0; |
5f39d397 | 788 | u64 end; |
7c4452b9 | 789 | |
1728366e | 790 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
e6138876 JB |
791 | mark, &cached_state)) { |
792 | convert_extent_bit(dirty_pages, start, end, EXTENT_NEED_WAIT, | |
793 | mark, &cached_state, GFP_NOFS); | |
794 | cached_state = NULL; | |
1728366e JB |
795 | err = filemap_fdatawrite_range(mapping, start, end); |
796 | if (err) | |
797 | werr = err; | |
798 | cond_resched(); | |
799 | start = end + 1; | |
7c4452b9 | 800 | } |
690587d1 CM |
801 | if (err) |
802 | werr = err; | |
803 | return werr; | |
804 | } | |
805 | ||
806 | /* | |
807 | * when btree blocks are allocated, they have some corresponding bits set for | |
808 | * them in one of two extent_io trees. This is used to make sure all of | |
809 | * those extents are on disk for transaction or log commit. We wait | |
810 | * on all the pages and clear them from the dirty pages state tree | |
811 | */ | |
812 | int btrfs_wait_marked_extents(struct btrfs_root *root, | |
8cef4e16 | 813 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 | 814 | { |
690587d1 CM |
815 | int err = 0; |
816 | int werr = 0; | |
1728366e | 817 | struct address_space *mapping = root->fs_info->btree_inode->i_mapping; |
e6138876 | 818 | struct extent_state *cached_state = NULL; |
690587d1 CM |
819 | u64 start = 0; |
820 | u64 end; | |
777e6bd7 | 821 | |
1728366e | 822 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
e6138876 JB |
823 | EXTENT_NEED_WAIT, &cached_state)) { |
824 | clear_extent_bit(dirty_pages, start, end, EXTENT_NEED_WAIT, | |
825 | 0, 0, &cached_state, GFP_NOFS); | |
1728366e JB |
826 | err = filemap_fdatawait_range(mapping, start, end); |
827 | if (err) | |
828 | werr = err; | |
829 | cond_resched(); | |
830 | start = end + 1; | |
777e6bd7 | 831 | } |
7c4452b9 CM |
832 | if (err) |
833 | werr = err; | |
834 | return werr; | |
79154b1b CM |
835 | } |
836 | ||
690587d1 CM |
837 | /* |
838 | * when btree blocks are allocated, they have some corresponding bits set for | |
839 | * them in one of two extent_io trees. This is used to make sure all of | |
840 | * those extents are on disk for transaction or log commit | |
841 | */ | |
171170c1 | 842 | static int btrfs_write_and_wait_marked_extents(struct btrfs_root *root, |
8cef4e16 | 843 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 CM |
844 | { |
845 | int ret; | |
846 | int ret2; | |
c6adc9cc | 847 | struct blk_plug plug; |
690587d1 | 848 | |
c6adc9cc | 849 | blk_start_plug(&plug); |
8cef4e16 | 850 | ret = btrfs_write_marked_extents(root, dirty_pages, mark); |
c6adc9cc | 851 | blk_finish_plug(&plug); |
8cef4e16 | 852 | ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark); |
bf0da8c1 CM |
853 | |
854 | if (ret) | |
855 | return ret; | |
856 | if (ret2) | |
857 | return ret2; | |
858 | return 0; | |
690587d1 CM |
859 | } |
860 | ||
d0c803c4 CM |
861 | int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans, |
862 | struct btrfs_root *root) | |
863 | { | |
864 | if (!trans || !trans->transaction) { | |
865 | struct inode *btree_inode; | |
866 | btree_inode = root->fs_info->btree_inode; | |
867 | return filemap_write_and_wait(btree_inode->i_mapping); | |
868 | } | |
869 | return btrfs_write_and_wait_marked_extents(root, | |
8cef4e16 YZ |
870 | &trans->transaction->dirty_pages, |
871 | EXTENT_DIRTY); | |
d0c803c4 CM |
872 | } |
873 | ||
d352ac68 CM |
874 | /* |
875 | * this is used to update the root pointer in the tree of tree roots. | |
876 | * | |
877 | * But, in the case of the extent allocation tree, updating the root | |
878 | * pointer may allocate blocks which may change the root of the extent | |
879 | * allocation tree. | |
880 | * | |
881 | * So, this loops and repeats and makes sure the cowonly root didn't | |
882 | * change while the root pointer was being updated in the metadata. | |
883 | */ | |
0b86a832 CM |
884 | static int update_cowonly_root(struct btrfs_trans_handle *trans, |
885 | struct btrfs_root *root) | |
79154b1b CM |
886 | { |
887 | int ret; | |
0b86a832 | 888 | u64 old_root_bytenr; |
86b9f2ec | 889 | u64 old_root_used; |
0b86a832 | 890 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
79154b1b | 891 | |
86b9f2ec | 892 | old_root_used = btrfs_root_used(&root->root_item); |
0b86a832 | 893 | btrfs_write_dirty_block_groups(trans, root); |
56bec294 | 894 | |
d397712b | 895 | while (1) { |
0b86a832 | 896 | old_root_bytenr = btrfs_root_bytenr(&root->root_item); |
86b9f2ec YZ |
897 | if (old_root_bytenr == root->node->start && |
898 | old_root_used == btrfs_root_used(&root->root_item)) | |
79154b1b | 899 | break; |
87ef2bb4 | 900 | |
5d4f98a2 | 901 | btrfs_set_root_node(&root->root_item, root->node); |
79154b1b | 902 | ret = btrfs_update_root(trans, tree_root, |
0b86a832 CM |
903 | &root->root_key, |
904 | &root->root_item); | |
49b25e05 JM |
905 | if (ret) |
906 | return ret; | |
56bec294 | 907 | |
86b9f2ec | 908 | old_root_used = btrfs_root_used(&root->root_item); |
4a8c9a62 | 909 | ret = btrfs_write_dirty_block_groups(trans, root); |
49b25e05 JM |
910 | if (ret) |
911 | return ret; | |
0b86a832 | 912 | } |
276e680d YZ |
913 | |
914 | if (root != root->fs_info->extent_root) | |
915 | switch_commit_root(root); | |
916 | ||
0b86a832 CM |
917 | return 0; |
918 | } | |
919 | ||
d352ac68 CM |
920 | /* |
921 | * update all the cowonly tree roots on disk | |
49b25e05 JM |
922 | * |
923 | * The error handling in this function may not be obvious. Any of the | |
924 | * failures will cause the file system to go offline. We still need | |
925 | * to clean up the delayed refs. | |
d352ac68 | 926 | */ |
5d4f98a2 YZ |
927 | static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans, |
928 | struct btrfs_root *root) | |
0b86a832 CM |
929 | { |
930 | struct btrfs_fs_info *fs_info = root->fs_info; | |
931 | struct list_head *next; | |
84234f3a | 932 | struct extent_buffer *eb; |
56bec294 | 933 | int ret; |
84234f3a | 934 | |
56bec294 | 935 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
49b25e05 JM |
936 | if (ret) |
937 | return ret; | |
87ef2bb4 | 938 | |
84234f3a | 939 | eb = btrfs_lock_root_node(fs_info->tree_root); |
49b25e05 JM |
940 | ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, |
941 | 0, &eb); | |
84234f3a YZ |
942 | btrfs_tree_unlock(eb); |
943 | free_extent_buffer(eb); | |
0b86a832 | 944 | |
49b25e05 JM |
945 | if (ret) |
946 | return ret; | |
947 | ||
56bec294 | 948 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
49b25e05 JM |
949 | if (ret) |
950 | return ret; | |
87ef2bb4 | 951 | |
733f4fbb | 952 | ret = btrfs_run_dev_stats(trans, root->fs_info); |
c16ce190 JB |
953 | if (ret) |
954 | return ret; | |
8dabb742 | 955 | ret = btrfs_run_dev_replace(trans, root->fs_info); |
c16ce190 JB |
956 | if (ret) |
957 | return ret; | |
546adb0d | 958 | ret = btrfs_run_qgroups(trans, root->fs_info); |
c16ce190 JB |
959 | if (ret) |
960 | return ret; | |
546adb0d JS |
961 | |
962 | /* run_qgroups might have added some more refs */ | |
963 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); | |
c16ce190 JB |
964 | if (ret) |
965 | return ret; | |
546adb0d | 966 | |
d397712b | 967 | while (!list_empty(&fs_info->dirty_cowonly_roots)) { |
0b86a832 CM |
968 | next = fs_info->dirty_cowonly_roots.next; |
969 | list_del_init(next); | |
970 | root = list_entry(next, struct btrfs_root, dirty_list); | |
87ef2bb4 | 971 | |
49b25e05 JM |
972 | ret = update_cowonly_root(trans, root); |
973 | if (ret) | |
974 | return ret; | |
79154b1b | 975 | } |
276e680d YZ |
976 | |
977 | down_write(&fs_info->extent_commit_sem); | |
978 | switch_commit_root(fs_info->extent_root); | |
979 | up_write(&fs_info->extent_commit_sem); | |
980 | ||
8dabb742 SB |
981 | btrfs_after_dev_replace_commit(fs_info); |
982 | ||
79154b1b CM |
983 | return 0; |
984 | } | |
985 | ||
d352ac68 CM |
986 | /* |
987 | * dead roots are old snapshots that need to be deleted. This allocates | |
988 | * a dirty root struct and adds it into the list of dead roots that need to | |
989 | * be deleted | |
990 | */ | |
cfad392b | 991 | void btrfs_add_dead_root(struct btrfs_root *root) |
5eda7b5e | 992 | { |
a4abeea4 | 993 | spin_lock(&root->fs_info->trans_lock); |
cfad392b JB |
994 | if (list_empty(&root->root_list)) |
995 | list_add_tail(&root->root_list, &root->fs_info->dead_roots); | |
a4abeea4 | 996 | spin_unlock(&root->fs_info->trans_lock); |
5eda7b5e CM |
997 | } |
998 | ||
d352ac68 | 999 | /* |
5d4f98a2 | 1000 | * update all the cowonly tree roots on disk |
d352ac68 | 1001 | */ |
5d4f98a2 YZ |
1002 | static noinline int commit_fs_roots(struct btrfs_trans_handle *trans, |
1003 | struct btrfs_root *root) | |
0f7d52f4 | 1004 | { |
0f7d52f4 | 1005 | struct btrfs_root *gang[8]; |
5d4f98a2 | 1006 | struct btrfs_fs_info *fs_info = root->fs_info; |
0f7d52f4 CM |
1007 | int i; |
1008 | int ret; | |
54aa1f4d CM |
1009 | int err = 0; |
1010 | ||
a4abeea4 | 1011 | spin_lock(&fs_info->fs_roots_radix_lock); |
d397712b | 1012 | while (1) { |
5d4f98a2 YZ |
1013 | ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, |
1014 | (void **)gang, 0, | |
0f7d52f4 CM |
1015 | ARRAY_SIZE(gang), |
1016 | BTRFS_ROOT_TRANS_TAG); | |
1017 | if (ret == 0) | |
1018 | break; | |
1019 | for (i = 0; i < ret; i++) { | |
1020 | root = gang[i]; | |
5d4f98a2 YZ |
1021 | radix_tree_tag_clear(&fs_info->fs_roots_radix, |
1022 | (unsigned long)root->root_key.objectid, | |
1023 | BTRFS_ROOT_TRANS_TAG); | |
a4abeea4 | 1024 | spin_unlock(&fs_info->fs_roots_radix_lock); |
31153d81 | 1025 | |
e02119d5 | 1026 | btrfs_free_log(trans, root); |
5d4f98a2 | 1027 | btrfs_update_reloc_root(trans, root); |
d68fc57b | 1028 | btrfs_orphan_commit_root(trans, root); |
bcc63abb | 1029 | |
82d5902d LZ |
1030 | btrfs_save_ino_cache(root, trans); |
1031 | ||
f1ebcc74 LB |
1032 | /* see comments in should_cow_block() */ |
1033 | root->force_cow = 0; | |
1034 | smp_wmb(); | |
1035 | ||
978d910d | 1036 | if (root->commit_root != root->node) { |
581bb050 | 1037 | mutex_lock(&root->fs_commit_mutex); |
817d52f8 | 1038 | switch_commit_root(root); |
581bb050 LZ |
1039 | btrfs_unpin_free_ino(root); |
1040 | mutex_unlock(&root->fs_commit_mutex); | |
1041 | ||
978d910d YZ |
1042 | btrfs_set_root_node(&root->root_item, |
1043 | root->node); | |
1044 | } | |
5d4f98a2 | 1045 | |
5d4f98a2 | 1046 | err = btrfs_update_root(trans, fs_info->tree_root, |
0f7d52f4 CM |
1047 | &root->root_key, |
1048 | &root->root_item); | |
a4abeea4 | 1049 | spin_lock(&fs_info->fs_roots_radix_lock); |
54aa1f4d CM |
1050 | if (err) |
1051 | break; | |
0f7d52f4 CM |
1052 | } |
1053 | } | |
a4abeea4 | 1054 | spin_unlock(&fs_info->fs_roots_radix_lock); |
54aa1f4d | 1055 | return err; |
0f7d52f4 CM |
1056 | } |
1057 | ||
d352ac68 | 1058 | /* |
de78b51a ES |
1059 | * defrag a given btree. |
1060 | * Every leaf in the btree is read and defragged. | |
d352ac68 | 1061 | */ |
de78b51a | 1062 | int btrfs_defrag_root(struct btrfs_root *root) |
e9d0b13b CM |
1063 | { |
1064 | struct btrfs_fs_info *info = root->fs_info; | |
e9d0b13b | 1065 | struct btrfs_trans_handle *trans; |
8929ecfa | 1066 | int ret; |
e9d0b13b | 1067 | |
8929ecfa | 1068 | if (xchg(&root->defrag_running, 1)) |
e9d0b13b | 1069 | return 0; |
8929ecfa | 1070 | |
6b80053d | 1071 | while (1) { |
8929ecfa YZ |
1072 | trans = btrfs_start_transaction(root, 0); |
1073 | if (IS_ERR(trans)) | |
1074 | return PTR_ERR(trans); | |
1075 | ||
de78b51a | 1076 | ret = btrfs_defrag_leaves(trans, root); |
8929ecfa | 1077 | |
e9d0b13b | 1078 | btrfs_end_transaction(trans, root); |
b53d3f5d | 1079 | btrfs_btree_balance_dirty(info->tree_root); |
e9d0b13b CM |
1080 | cond_resched(); |
1081 | ||
7841cb28 | 1082 | if (btrfs_fs_closing(root->fs_info) || ret != -EAGAIN) |
e9d0b13b | 1083 | break; |
210549eb DS |
1084 | |
1085 | if (btrfs_defrag_cancelled(root->fs_info)) { | |
1086 | printk(KERN_DEBUG "btrfs: defrag_root cancelled\n"); | |
1087 | ret = -EAGAIN; | |
1088 | break; | |
1089 | } | |
e9d0b13b CM |
1090 | } |
1091 | root->defrag_running = 0; | |
8929ecfa | 1092 | return ret; |
e9d0b13b CM |
1093 | } |
1094 | ||
d352ac68 CM |
1095 | /* |
1096 | * new snapshots need to be created at a very specific time in the | |
aec8030a MX |
1097 | * transaction commit. This does the actual creation. |
1098 | * | |
1099 | * Note: | |
1100 | * If the error which may affect the commitment of the current transaction | |
1101 | * happens, we should return the error number. If the error which just affect | |
1102 | * the creation of the pending snapshots, just return 0. | |
d352ac68 | 1103 | */ |
80b6794d | 1104 | static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, |
3063d29f CM |
1105 | struct btrfs_fs_info *fs_info, |
1106 | struct btrfs_pending_snapshot *pending) | |
1107 | { | |
1108 | struct btrfs_key key; | |
80b6794d | 1109 | struct btrfs_root_item *new_root_item; |
3063d29f CM |
1110 | struct btrfs_root *tree_root = fs_info->tree_root; |
1111 | struct btrfs_root *root = pending->root; | |
6bdb72de | 1112 | struct btrfs_root *parent_root; |
98c9942a | 1113 | struct btrfs_block_rsv *rsv; |
6bdb72de | 1114 | struct inode *parent_inode; |
42874b3d MX |
1115 | struct btrfs_path *path; |
1116 | struct btrfs_dir_item *dir_item; | |
a22285a6 | 1117 | struct dentry *dentry; |
3063d29f | 1118 | struct extent_buffer *tmp; |
925baedd | 1119 | struct extent_buffer *old; |
8ea05e3a | 1120 | struct timespec cur_time = CURRENT_TIME; |
aec8030a | 1121 | int ret = 0; |
d68fc57b | 1122 | u64 to_reserve = 0; |
6bdb72de | 1123 | u64 index = 0; |
a22285a6 | 1124 | u64 objectid; |
b83cc969 | 1125 | u64 root_flags; |
8ea05e3a | 1126 | uuid_le new_uuid; |
3063d29f | 1127 | |
42874b3d MX |
1128 | path = btrfs_alloc_path(); |
1129 | if (!path) { | |
aec8030a MX |
1130 | pending->error = -ENOMEM; |
1131 | return 0; | |
42874b3d MX |
1132 | } |
1133 | ||
80b6794d CM |
1134 | new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS); |
1135 | if (!new_root_item) { | |
aec8030a | 1136 | pending->error = -ENOMEM; |
6fa9700e | 1137 | goto root_item_alloc_fail; |
80b6794d | 1138 | } |
a22285a6 | 1139 | |
aec8030a MX |
1140 | pending->error = btrfs_find_free_objectid(tree_root, &objectid); |
1141 | if (pending->error) | |
6fa9700e | 1142 | goto no_free_objectid; |
3063d29f | 1143 | |
3fd0a558 | 1144 | btrfs_reloc_pre_snapshot(trans, pending, &to_reserve); |
d68fc57b YZ |
1145 | |
1146 | if (to_reserve > 0) { | |
aec8030a MX |
1147 | pending->error = btrfs_block_rsv_add(root, |
1148 | &pending->block_rsv, | |
1149 | to_reserve, | |
1150 | BTRFS_RESERVE_NO_FLUSH); | |
1151 | if (pending->error) | |
6fa9700e | 1152 | goto no_free_objectid; |
d68fc57b YZ |
1153 | } |
1154 | ||
aec8030a MX |
1155 | pending->error = btrfs_qgroup_inherit(trans, fs_info, |
1156 | root->root_key.objectid, | |
1157 | objectid, pending->inherit); | |
1158 | if (pending->error) | |
6fa9700e | 1159 | goto no_free_objectid; |
6f72c7e2 | 1160 | |
3063d29f | 1161 | key.objectid = objectid; |
a22285a6 YZ |
1162 | key.offset = (u64)-1; |
1163 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3063d29f | 1164 | |
6fa9700e | 1165 | rsv = trans->block_rsv; |
a22285a6 | 1166 | trans->block_rsv = &pending->block_rsv; |
2382c5cc | 1167 | trans->bytes_reserved = trans->block_rsv->reserved; |
3de4586c | 1168 | |
a22285a6 | 1169 | dentry = pending->dentry; |
e9662f70 | 1170 | parent_inode = pending->dir; |
a22285a6 | 1171 | parent_root = BTRFS_I(parent_inode)->root; |
7585717f | 1172 | record_root_in_trans(trans, parent_root); |
a22285a6 | 1173 | |
3063d29f CM |
1174 | /* |
1175 | * insert the directory item | |
1176 | */ | |
3de4586c | 1177 | ret = btrfs_set_inode_index(parent_inode, &index); |
49b25e05 | 1178 | BUG_ON(ret); /* -ENOMEM */ |
42874b3d MX |
1179 | |
1180 | /* check if there is a file/dir which has the same name. */ | |
1181 | dir_item = btrfs_lookup_dir_item(NULL, parent_root, path, | |
1182 | btrfs_ino(parent_inode), | |
1183 | dentry->d_name.name, | |
1184 | dentry->d_name.len, 0); | |
1185 | if (dir_item != NULL && !IS_ERR(dir_item)) { | |
fe66a05a | 1186 | pending->error = -EEXIST; |
aec8030a | 1187 | goto dir_item_existed; |
42874b3d MX |
1188 | } else if (IS_ERR(dir_item)) { |
1189 | ret = PTR_ERR(dir_item); | |
8732d44f MX |
1190 | btrfs_abort_transaction(trans, root, ret); |
1191 | goto fail; | |
79787eaa | 1192 | } |
42874b3d | 1193 | btrfs_release_path(path); |
52c26179 | 1194 | |
e999376f CM |
1195 | /* |
1196 | * pull in the delayed directory update | |
1197 | * and the delayed inode item | |
1198 | * otherwise we corrupt the FS during | |
1199 | * snapshot | |
1200 | */ | |
1201 | ret = btrfs_run_delayed_items(trans, root); | |
8732d44f MX |
1202 | if (ret) { /* Transaction aborted */ |
1203 | btrfs_abort_transaction(trans, root, ret); | |
1204 | goto fail; | |
1205 | } | |
e999376f | 1206 | |
7585717f | 1207 | record_root_in_trans(trans, root); |
6bdb72de SW |
1208 | btrfs_set_root_last_snapshot(&root->root_item, trans->transid); |
1209 | memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); | |
08fe4db1 | 1210 | btrfs_check_and_init_root_item(new_root_item); |
6bdb72de | 1211 | |
b83cc969 LZ |
1212 | root_flags = btrfs_root_flags(new_root_item); |
1213 | if (pending->readonly) | |
1214 | root_flags |= BTRFS_ROOT_SUBVOL_RDONLY; | |
1215 | else | |
1216 | root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY; | |
1217 | btrfs_set_root_flags(new_root_item, root_flags); | |
1218 | ||
8ea05e3a AB |
1219 | btrfs_set_root_generation_v2(new_root_item, |
1220 | trans->transid); | |
1221 | uuid_le_gen(&new_uuid); | |
1222 | memcpy(new_root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE); | |
1223 | memcpy(new_root_item->parent_uuid, root->root_item.uuid, | |
1224 | BTRFS_UUID_SIZE); | |
70023da2 SB |
1225 | if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) { |
1226 | memset(new_root_item->received_uuid, 0, | |
1227 | sizeof(new_root_item->received_uuid)); | |
1228 | memset(&new_root_item->stime, 0, sizeof(new_root_item->stime)); | |
1229 | memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime)); | |
1230 | btrfs_set_root_stransid(new_root_item, 0); | |
1231 | btrfs_set_root_rtransid(new_root_item, 0); | |
1232 | } | |
3cae210f QW |
1233 | btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec); |
1234 | btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec); | |
8ea05e3a | 1235 | btrfs_set_root_otransid(new_root_item, trans->transid); |
8ea05e3a | 1236 | |
6bdb72de | 1237 | old = btrfs_lock_root_node(root); |
49b25e05 | 1238 | ret = btrfs_cow_block(trans, root, old, NULL, 0, &old); |
79787eaa JM |
1239 | if (ret) { |
1240 | btrfs_tree_unlock(old); | |
1241 | free_extent_buffer(old); | |
8732d44f MX |
1242 | btrfs_abort_transaction(trans, root, ret); |
1243 | goto fail; | |
79787eaa | 1244 | } |
49b25e05 | 1245 | |
6bdb72de SW |
1246 | btrfs_set_lock_blocking(old); |
1247 | ||
49b25e05 | 1248 | ret = btrfs_copy_root(trans, root, old, &tmp, objectid); |
79787eaa | 1249 | /* clean up in any case */ |
6bdb72de SW |
1250 | btrfs_tree_unlock(old); |
1251 | free_extent_buffer(old); | |
8732d44f MX |
1252 | if (ret) { |
1253 | btrfs_abort_transaction(trans, root, ret); | |
1254 | goto fail; | |
1255 | } | |
6bdb72de | 1256 | |
f1ebcc74 LB |
1257 | /* see comments in should_cow_block() */ |
1258 | root->force_cow = 1; | |
1259 | smp_wmb(); | |
1260 | ||
6bdb72de | 1261 | btrfs_set_root_node(new_root_item, tmp); |
a22285a6 YZ |
1262 | /* record when the snapshot was created in key.offset */ |
1263 | key.offset = trans->transid; | |
1264 | ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); | |
6bdb72de SW |
1265 | btrfs_tree_unlock(tmp); |
1266 | free_extent_buffer(tmp); | |
8732d44f MX |
1267 | if (ret) { |
1268 | btrfs_abort_transaction(trans, root, ret); | |
1269 | goto fail; | |
1270 | } | |
6bdb72de | 1271 | |
a22285a6 YZ |
1272 | /* |
1273 | * insert root back/forward references | |
1274 | */ | |
1275 | ret = btrfs_add_root_ref(trans, tree_root, objectid, | |
0660b5af | 1276 | parent_root->root_key.objectid, |
33345d01 | 1277 | btrfs_ino(parent_inode), index, |
a22285a6 | 1278 | dentry->d_name.name, dentry->d_name.len); |
8732d44f MX |
1279 | if (ret) { |
1280 | btrfs_abort_transaction(trans, root, ret); | |
1281 | goto fail; | |
1282 | } | |
0660b5af | 1283 | |
a22285a6 YZ |
1284 | key.offset = (u64)-1; |
1285 | pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key); | |
79787eaa JM |
1286 | if (IS_ERR(pending->snap)) { |
1287 | ret = PTR_ERR(pending->snap); | |
8732d44f MX |
1288 | btrfs_abort_transaction(trans, root, ret); |
1289 | goto fail; | |
79787eaa | 1290 | } |
d68fc57b | 1291 | |
49b25e05 | 1292 | ret = btrfs_reloc_post_snapshot(trans, pending); |
8732d44f MX |
1293 | if (ret) { |
1294 | btrfs_abort_transaction(trans, root, ret); | |
1295 | goto fail; | |
1296 | } | |
361048f5 MX |
1297 | |
1298 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); | |
8732d44f MX |
1299 | if (ret) { |
1300 | btrfs_abort_transaction(trans, root, ret); | |
1301 | goto fail; | |
1302 | } | |
42874b3d MX |
1303 | |
1304 | ret = btrfs_insert_dir_item(trans, parent_root, | |
1305 | dentry->d_name.name, dentry->d_name.len, | |
1306 | parent_inode, &key, | |
1307 | BTRFS_FT_DIR, index); | |
1308 | /* We have check then name at the beginning, so it is impossible. */ | |
9c52057c | 1309 | BUG_ON(ret == -EEXIST || ret == -EOVERFLOW); |
8732d44f MX |
1310 | if (ret) { |
1311 | btrfs_abort_transaction(trans, root, ret); | |
1312 | goto fail; | |
1313 | } | |
42874b3d MX |
1314 | |
1315 | btrfs_i_size_write(parent_inode, parent_inode->i_size + | |
1316 | dentry->d_name.len * 2); | |
1317 | parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME; | |
be6aef60 | 1318 | ret = btrfs_update_inode_fallback(trans, parent_root, parent_inode); |
dd5f9615 SB |
1319 | if (ret) { |
1320 | btrfs_abort_transaction(trans, root, ret); | |
1321 | goto fail; | |
1322 | } | |
1323 | ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root, new_uuid.b, | |
1324 | BTRFS_UUID_KEY_SUBVOL, objectid); | |
1325 | if (ret) { | |
8732d44f | 1326 | btrfs_abort_transaction(trans, root, ret); |
dd5f9615 SB |
1327 | goto fail; |
1328 | } | |
1329 | if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) { | |
1330 | ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root, | |
1331 | new_root_item->received_uuid, | |
1332 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, | |
1333 | objectid); | |
1334 | if (ret && ret != -EEXIST) { | |
1335 | btrfs_abort_transaction(trans, root, ret); | |
1336 | goto fail; | |
1337 | } | |
1338 | } | |
3063d29f | 1339 | fail: |
aec8030a MX |
1340 | pending->error = ret; |
1341 | dir_item_existed: | |
98c9942a | 1342 | trans->block_rsv = rsv; |
2382c5cc | 1343 | trans->bytes_reserved = 0; |
6fa9700e MX |
1344 | no_free_objectid: |
1345 | kfree(new_root_item); | |
1346 | root_item_alloc_fail: | |
42874b3d | 1347 | btrfs_free_path(path); |
49b25e05 | 1348 | return ret; |
3063d29f CM |
1349 | } |
1350 | ||
d352ac68 CM |
1351 | /* |
1352 | * create all the snapshots we've scheduled for creation | |
1353 | */ | |
80b6794d CM |
1354 | static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans, |
1355 | struct btrfs_fs_info *fs_info) | |
3de4586c | 1356 | { |
aec8030a | 1357 | struct btrfs_pending_snapshot *pending, *next; |
3de4586c | 1358 | struct list_head *head = &trans->transaction->pending_snapshots; |
aec8030a | 1359 | int ret = 0; |
3de4586c | 1360 | |
aec8030a MX |
1361 | list_for_each_entry_safe(pending, next, head, list) { |
1362 | list_del(&pending->list); | |
1363 | ret = create_pending_snapshot(trans, fs_info, pending); | |
1364 | if (ret) | |
1365 | break; | |
1366 | } | |
1367 | return ret; | |
3de4586c CM |
1368 | } |
1369 | ||
5d4f98a2 YZ |
1370 | static void update_super_roots(struct btrfs_root *root) |
1371 | { | |
1372 | struct btrfs_root_item *root_item; | |
1373 | struct btrfs_super_block *super; | |
1374 | ||
6c41761f | 1375 | super = root->fs_info->super_copy; |
5d4f98a2 YZ |
1376 | |
1377 | root_item = &root->fs_info->chunk_root->root_item; | |
1378 | super->chunk_root = root_item->bytenr; | |
1379 | super->chunk_root_generation = root_item->generation; | |
1380 | super->chunk_root_level = root_item->level; | |
1381 | ||
1382 | root_item = &root->fs_info->tree_root->root_item; | |
1383 | super->root = root_item->bytenr; | |
1384 | super->generation = root_item->generation; | |
1385 | super->root_level = root_item->level; | |
73bc1876 | 1386 | if (btrfs_test_opt(root, SPACE_CACHE)) |
0af3d00b | 1387 | super->cache_generation = root_item->generation; |
70f80175 SB |
1388 | if (root->fs_info->update_uuid_tree_gen) |
1389 | super->uuid_tree_generation = root_item->generation; | |
5d4f98a2 YZ |
1390 | } |
1391 | ||
f36f3042 CM |
1392 | int btrfs_transaction_in_commit(struct btrfs_fs_info *info) |
1393 | { | |
4a9d8bde | 1394 | struct btrfs_transaction *trans; |
f36f3042 | 1395 | int ret = 0; |
4a9d8bde | 1396 | |
a4abeea4 | 1397 | spin_lock(&info->trans_lock); |
4a9d8bde MX |
1398 | trans = info->running_transaction; |
1399 | if (trans) | |
1400 | ret = (trans->state >= TRANS_STATE_COMMIT_START); | |
a4abeea4 | 1401 | spin_unlock(&info->trans_lock); |
f36f3042 CM |
1402 | return ret; |
1403 | } | |
1404 | ||
8929ecfa YZ |
1405 | int btrfs_transaction_blocked(struct btrfs_fs_info *info) |
1406 | { | |
4a9d8bde | 1407 | struct btrfs_transaction *trans; |
8929ecfa | 1408 | int ret = 0; |
4a9d8bde | 1409 | |
a4abeea4 | 1410 | spin_lock(&info->trans_lock); |
4a9d8bde MX |
1411 | trans = info->running_transaction; |
1412 | if (trans) | |
1413 | ret = is_transaction_blocked(trans); | |
a4abeea4 | 1414 | spin_unlock(&info->trans_lock); |
8929ecfa YZ |
1415 | return ret; |
1416 | } | |
1417 | ||
bb9c12c9 SW |
1418 | /* |
1419 | * wait for the current transaction commit to start and block subsequent | |
1420 | * transaction joins | |
1421 | */ | |
1422 | static void wait_current_trans_commit_start(struct btrfs_root *root, | |
1423 | struct btrfs_transaction *trans) | |
1424 | { | |
4a9d8bde | 1425 | wait_event(root->fs_info->transaction_blocked_wait, |
501407aa JB |
1426 | trans->state >= TRANS_STATE_COMMIT_START || |
1427 | trans->aborted); | |
bb9c12c9 SW |
1428 | } |
1429 | ||
1430 | /* | |
1431 | * wait for the current transaction to start and then become unblocked. | |
1432 | * caller holds ref. | |
1433 | */ | |
1434 | static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root, | |
1435 | struct btrfs_transaction *trans) | |
1436 | { | |
72d63ed6 | 1437 | wait_event(root->fs_info->transaction_wait, |
501407aa JB |
1438 | trans->state >= TRANS_STATE_UNBLOCKED || |
1439 | trans->aborted); | |
bb9c12c9 SW |
1440 | } |
1441 | ||
1442 | /* | |
1443 | * commit transactions asynchronously. once btrfs_commit_transaction_async | |
1444 | * returns, any subsequent transaction will not be allowed to join. | |
1445 | */ | |
1446 | struct btrfs_async_commit { | |
1447 | struct btrfs_trans_handle *newtrans; | |
1448 | struct btrfs_root *root; | |
7892b5af | 1449 | struct work_struct work; |
bb9c12c9 SW |
1450 | }; |
1451 | ||
1452 | static void do_async_commit(struct work_struct *work) | |
1453 | { | |
1454 | struct btrfs_async_commit *ac = | |
7892b5af | 1455 | container_of(work, struct btrfs_async_commit, work); |
bb9c12c9 | 1456 | |
6fc4e354 SW |
1457 | /* |
1458 | * We've got freeze protection passed with the transaction. | |
1459 | * Tell lockdep about it. | |
1460 | */ | |
ff7c1d33 MX |
1461 | if (ac->newtrans->type < TRANS_JOIN_NOLOCK) |
1462 | rwsem_acquire_read( | |
1463 | &ac->root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1], | |
1464 | 0, 1, _THIS_IP_); | |
6fc4e354 | 1465 | |
e209db7a SW |
1466 | current->journal_info = ac->newtrans; |
1467 | ||
bb9c12c9 SW |
1468 | btrfs_commit_transaction(ac->newtrans, ac->root); |
1469 | kfree(ac); | |
1470 | } | |
1471 | ||
1472 | int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans, | |
1473 | struct btrfs_root *root, | |
1474 | int wait_for_unblock) | |
1475 | { | |
1476 | struct btrfs_async_commit *ac; | |
1477 | struct btrfs_transaction *cur_trans; | |
1478 | ||
1479 | ac = kmalloc(sizeof(*ac), GFP_NOFS); | |
db5b493a TI |
1480 | if (!ac) |
1481 | return -ENOMEM; | |
bb9c12c9 | 1482 | |
7892b5af | 1483 | INIT_WORK(&ac->work, do_async_commit); |
bb9c12c9 | 1484 | ac->root = root; |
7a7eaa40 | 1485 | ac->newtrans = btrfs_join_transaction(root); |
3612b495 TI |
1486 | if (IS_ERR(ac->newtrans)) { |
1487 | int err = PTR_ERR(ac->newtrans); | |
1488 | kfree(ac); | |
1489 | return err; | |
1490 | } | |
bb9c12c9 SW |
1491 | |
1492 | /* take transaction reference */ | |
bb9c12c9 | 1493 | cur_trans = trans->transaction; |
13c5a93e | 1494 | atomic_inc(&cur_trans->use_count); |
bb9c12c9 SW |
1495 | |
1496 | btrfs_end_transaction(trans, root); | |
6fc4e354 SW |
1497 | |
1498 | /* | |
1499 | * Tell lockdep we've released the freeze rwsem, since the | |
1500 | * async commit thread will be the one to unlock it. | |
1501 | */ | |
ff7c1d33 MX |
1502 | if (trans->type < TRANS_JOIN_NOLOCK) |
1503 | rwsem_release( | |
1504 | &root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1], | |
1505 | 1, _THIS_IP_); | |
6fc4e354 | 1506 | |
7892b5af | 1507 | schedule_work(&ac->work); |
bb9c12c9 SW |
1508 | |
1509 | /* wait for transaction to start and unblock */ | |
bb9c12c9 SW |
1510 | if (wait_for_unblock) |
1511 | wait_current_trans_commit_start_and_unblock(root, cur_trans); | |
1512 | else | |
1513 | wait_current_trans_commit_start(root, cur_trans); | |
bb9c12c9 | 1514 | |
38e88054 SW |
1515 | if (current->journal_info == trans) |
1516 | current->journal_info = NULL; | |
1517 | ||
1518 | put_transaction(cur_trans); | |
bb9c12c9 SW |
1519 | return 0; |
1520 | } | |
1521 | ||
49b25e05 JM |
1522 | |
1523 | static void cleanup_transaction(struct btrfs_trans_handle *trans, | |
7b8b92af | 1524 | struct btrfs_root *root, int err) |
49b25e05 JM |
1525 | { |
1526 | struct btrfs_transaction *cur_trans = trans->transaction; | |
f094ac32 | 1527 | DEFINE_WAIT(wait); |
49b25e05 JM |
1528 | |
1529 | WARN_ON(trans->use_count > 1); | |
1530 | ||
7b8b92af JB |
1531 | btrfs_abort_transaction(trans, root, err); |
1532 | ||
49b25e05 | 1533 | spin_lock(&root->fs_info->trans_lock); |
66b6135b | 1534 | |
25d8c284 MX |
1535 | /* |
1536 | * If the transaction is removed from the list, it means this | |
1537 | * transaction has been committed successfully, so it is impossible | |
1538 | * to call the cleanup function. | |
1539 | */ | |
1540 | BUG_ON(list_empty(&cur_trans->list)); | |
66b6135b | 1541 | |
49b25e05 | 1542 | list_del_init(&cur_trans->list); |
d7096fc3 | 1543 | if (cur_trans == root->fs_info->running_transaction) { |
4a9d8bde | 1544 | cur_trans->state = TRANS_STATE_COMMIT_DOING; |
f094ac32 LB |
1545 | spin_unlock(&root->fs_info->trans_lock); |
1546 | wait_event(cur_trans->writer_wait, | |
1547 | atomic_read(&cur_trans->num_writers) == 1); | |
1548 | ||
1549 | spin_lock(&root->fs_info->trans_lock); | |
d7096fc3 | 1550 | } |
49b25e05 JM |
1551 | spin_unlock(&root->fs_info->trans_lock); |
1552 | ||
1553 | btrfs_cleanup_one_transaction(trans->transaction, root); | |
1554 | ||
4a9d8bde MX |
1555 | spin_lock(&root->fs_info->trans_lock); |
1556 | if (cur_trans == root->fs_info->running_transaction) | |
1557 | root->fs_info->running_transaction = NULL; | |
1558 | spin_unlock(&root->fs_info->trans_lock); | |
1559 | ||
e0228285 JB |
1560 | if (trans->type & __TRANS_FREEZABLE) |
1561 | sb_end_intwrite(root->fs_info->sb); | |
49b25e05 JM |
1562 | put_transaction(cur_trans); |
1563 | put_transaction(cur_trans); | |
1564 | ||
1565 | trace_btrfs_transaction_commit(root); | |
1566 | ||
1567 | btrfs_scrub_continue(root); | |
1568 | ||
1569 | if (current->journal_info == trans) | |
1570 | current->journal_info = NULL; | |
1571 | ||
1572 | kmem_cache_free(btrfs_trans_handle_cachep, trans); | |
1573 | } | |
1574 | ||
ca469637 MX |
1575 | static int btrfs_flush_all_pending_stuffs(struct btrfs_trans_handle *trans, |
1576 | struct btrfs_root *root) | |
1577 | { | |
ca469637 MX |
1578 | int ret; |
1579 | ||
ca469637 MX |
1580 | ret = btrfs_run_delayed_items(trans, root); |
1581 | if (ret) | |
1582 | return ret; | |
1583 | ||
1584 | /* | |
1585 | * running the delayed items may have added new refs. account | |
1586 | * them now so that they hinder processing of more delayed refs | |
1587 | * as little as possible. | |
1588 | */ | |
1589 | btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info); | |
1590 | ||
1591 | /* | |
1592 | * rename don't use btrfs_join_transaction, so, once we | |
1593 | * set the transaction to blocked above, we aren't going | |
1594 | * to get any new ordered operations. We can safely run | |
1595 | * it here and no for sure that nothing new will be added | |
1596 | * to the list | |
1597 | */ | |
569e0f35 | 1598 | ret = btrfs_run_ordered_operations(trans, root, 1); |
ca469637 | 1599 | |
eebc6084 | 1600 | return ret; |
ca469637 MX |
1601 | } |
1602 | ||
82436617 MX |
1603 | static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info) |
1604 | { | |
1605 | if (btrfs_test_opt(fs_info->tree_root, FLUSHONCOMMIT)) | |
1606 | return btrfs_start_all_delalloc_inodes(fs_info, 1); | |
1607 | return 0; | |
1608 | } | |
1609 | ||
1610 | static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info) | |
1611 | { | |
1612 | if (btrfs_test_opt(fs_info->tree_root, FLUSHONCOMMIT)) | |
f0de181c | 1613 | btrfs_wait_all_ordered_extents(fs_info); |
82436617 MX |
1614 | } |
1615 | ||
79154b1b CM |
1616 | int btrfs_commit_transaction(struct btrfs_trans_handle *trans, |
1617 | struct btrfs_root *root) | |
1618 | { | |
49b25e05 | 1619 | struct btrfs_transaction *cur_trans = trans->transaction; |
8fd17795 | 1620 | struct btrfs_transaction *prev_trans = NULL; |
25287e0a | 1621 | int ret; |
79154b1b | 1622 | |
569e0f35 | 1623 | ret = btrfs_run_ordered_operations(trans, root, 0); |
25287e0a MX |
1624 | if (ret) { |
1625 | btrfs_abort_transaction(trans, root, ret); | |
e4a2bcac JB |
1626 | btrfs_end_transaction(trans, root); |
1627 | return ret; | |
25287e0a | 1628 | } |
5a3f23d5 | 1629 | |
8d25a086 MX |
1630 | /* Stop the commit early if ->aborted is set */ |
1631 | if (unlikely(ACCESS_ONCE(cur_trans->aborted))) { | |
25287e0a | 1632 | ret = cur_trans->aborted; |
e4a2bcac JB |
1633 | btrfs_end_transaction(trans, root); |
1634 | return ret; | |
25287e0a | 1635 | } |
49b25e05 | 1636 | |
56bec294 CM |
1637 | /* make a pass through all the delayed refs we have so far |
1638 | * any runnings procs may add more while we are here | |
1639 | */ | |
1640 | ret = btrfs_run_delayed_refs(trans, root, 0); | |
e4a2bcac JB |
1641 | if (ret) { |
1642 | btrfs_end_transaction(trans, root); | |
1643 | return ret; | |
1644 | } | |
56bec294 | 1645 | |
0e721106 JB |
1646 | btrfs_trans_release_metadata(trans, root); |
1647 | trans->block_rsv = NULL; | |
272d26d0 MX |
1648 | if (trans->qgroup_reserved) { |
1649 | btrfs_qgroup_free(root, trans->qgroup_reserved); | |
1650 | trans->qgroup_reserved = 0; | |
1651 | } | |
0e721106 | 1652 | |
b7ec40d7 | 1653 | cur_trans = trans->transaction; |
49b25e05 | 1654 | |
56bec294 CM |
1655 | /* |
1656 | * set the flushing flag so procs in this transaction have to | |
1657 | * start sending their work down. | |
1658 | */ | |
b7ec40d7 | 1659 | cur_trans->delayed_refs.flushing = 1; |
1be41b78 | 1660 | smp_wmb(); |
56bec294 | 1661 | |
ea658bad JB |
1662 | if (!list_empty(&trans->new_bgs)) |
1663 | btrfs_create_pending_block_groups(trans, root); | |
1664 | ||
c3e69d58 | 1665 | ret = btrfs_run_delayed_refs(trans, root, 0); |
e4a2bcac JB |
1666 | if (ret) { |
1667 | btrfs_end_transaction(trans, root); | |
1668 | return ret; | |
1669 | } | |
56bec294 | 1670 | |
4a9d8bde MX |
1671 | spin_lock(&root->fs_info->trans_lock); |
1672 | if (cur_trans->state >= TRANS_STATE_COMMIT_START) { | |
1673 | spin_unlock(&root->fs_info->trans_lock); | |
13c5a93e | 1674 | atomic_inc(&cur_trans->use_count); |
49b25e05 | 1675 | ret = btrfs_end_transaction(trans, root); |
ccd467d6 | 1676 | |
b9c8300c | 1677 | wait_for_commit(root, cur_trans); |
15ee9bc7 | 1678 | |
79154b1b | 1679 | put_transaction(cur_trans); |
15ee9bc7 | 1680 | |
49b25e05 | 1681 | return ret; |
79154b1b | 1682 | } |
4313b399 | 1683 | |
4a9d8bde | 1684 | cur_trans->state = TRANS_STATE_COMMIT_START; |
bb9c12c9 SW |
1685 | wake_up(&root->fs_info->transaction_blocked_wait); |
1686 | ||
ccd467d6 CM |
1687 | if (cur_trans->list.prev != &root->fs_info->trans_list) { |
1688 | prev_trans = list_entry(cur_trans->list.prev, | |
1689 | struct btrfs_transaction, list); | |
4a9d8bde | 1690 | if (prev_trans->state != TRANS_STATE_COMPLETED) { |
13c5a93e | 1691 | atomic_inc(&prev_trans->use_count); |
a4abeea4 | 1692 | spin_unlock(&root->fs_info->trans_lock); |
ccd467d6 CM |
1693 | |
1694 | wait_for_commit(root, prev_trans); | |
ccd467d6 | 1695 | |
15ee9bc7 | 1696 | put_transaction(prev_trans); |
a4abeea4 JB |
1697 | } else { |
1698 | spin_unlock(&root->fs_info->trans_lock); | |
ccd467d6 | 1699 | } |
a4abeea4 JB |
1700 | } else { |
1701 | spin_unlock(&root->fs_info->trans_lock); | |
ccd467d6 | 1702 | } |
15ee9bc7 | 1703 | |
0860adfd MX |
1704 | extwriter_counter_dec(cur_trans, trans->type); |
1705 | ||
82436617 MX |
1706 | ret = btrfs_start_delalloc_flush(root->fs_info); |
1707 | if (ret) | |
1708 | goto cleanup_transaction; | |
1709 | ||
581227d0 MX |
1710 | ret = btrfs_flush_all_pending_stuffs(trans, root); |
1711 | if (ret) | |
1712 | goto cleanup_transaction; | |
15ee9bc7 | 1713 | |
581227d0 MX |
1714 | wait_event(cur_trans->writer_wait, |
1715 | extwriter_counter_read(cur_trans) == 0); | |
15ee9bc7 | 1716 | |
581227d0 | 1717 | /* some pending stuffs might be added after the previous flush. */ |
ca469637 MX |
1718 | ret = btrfs_flush_all_pending_stuffs(trans, root); |
1719 | if (ret) | |
1720 | goto cleanup_transaction; | |
1721 | ||
82436617 | 1722 | btrfs_wait_delalloc_flush(root->fs_info); |
ed0ca140 JB |
1723 | /* |
1724 | * Ok now we need to make sure to block out any other joins while we | |
1725 | * commit the transaction. We could have started a join before setting | |
4a9d8bde | 1726 | * COMMIT_DOING so make sure to wait for num_writers to == 1 again. |
ed0ca140 JB |
1727 | */ |
1728 | spin_lock(&root->fs_info->trans_lock); | |
4a9d8bde | 1729 | cur_trans->state = TRANS_STATE_COMMIT_DOING; |
ed0ca140 JB |
1730 | spin_unlock(&root->fs_info->trans_lock); |
1731 | wait_event(cur_trans->writer_wait, | |
1732 | atomic_read(&cur_trans->num_writers) == 1); | |
1733 | ||
2cba30f1 MX |
1734 | /* ->aborted might be set after the previous check, so check it */ |
1735 | if (unlikely(ACCESS_ONCE(cur_trans->aborted))) { | |
1736 | ret = cur_trans->aborted; | |
1737 | goto cleanup_transaction; | |
1738 | } | |
7585717f CM |
1739 | /* |
1740 | * the reloc mutex makes sure that we stop | |
1741 | * the balancing code from coming in and moving | |
1742 | * extents around in the middle of the commit | |
1743 | */ | |
1744 | mutex_lock(&root->fs_info->reloc_mutex); | |
1745 | ||
42874b3d MX |
1746 | /* |
1747 | * We needn't worry about the delayed items because we will | |
1748 | * deal with them in create_pending_snapshot(), which is the | |
1749 | * core function of the snapshot creation. | |
1750 | */ | |
1751 | ret = create_pending_snapshots(trans, root->fs_info); | |
49b25e05 JM |
1752 | if (ret) { |
1753 | mutex_unlock(&root->fs_info->reloc_mutex); | |
1754 | goto cleanup_transaction; | |
1755 | } | |
3063d29f | 1756 | |
42874b3d MX |
1757 | /* |
1758 | * We insert the dir indexes of the snapshots and update the inode | |
1759 | * of the snapshots' parents after the snapshot creation, so there | |
1760 | * are some delayed items which are not dealt with. Now deal with | |
1761 | * them. | |
1762 | * | |
1763 | * We needn't worry that this operation will corrupt the snapshots, | |
1764 | * because all the tree which are snapshoted will be forced to COW | |
1765 | * the nodes and leaves. | |
1766 | */ | |
1767 | ret = btrfs_run_delayed_items(trans, root); | |
49b25e05 JM |
1768 | if (ret) { |
1769 | mutex_unlock(&root->fs_info->reloc_mutex); | |
1770 | goto cleanup_transaction; | |
1771 | } | |
16cdcec7 | 1772 | |
56bec294 | 1773 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
49b25e05 JM |
1774 | if (ret) { |
1775 | mutex_unlock(&root->fs_info->reloc_mutex); | |
1776 | goto cleanup_transaction; | |
1777 | } | |
56bec294 | 1778 | |
e999376f CM |
1779 | /* |
1780 | * make sure none of the code above managed to slip in a | |
1781 | * delayed item | |
1782 | */ | |
1783 | btrfs_assert_delayed_root_empty(root); | |
1784 | ||
2c90e5d6 | 1785 | WARN_ON(cur_trans != trans->transaction); |
dc17ff8f | 1786 | |
a2de733c | 1787 | btrfs_scrub_pause(root); |
e02119d5 CM |
1788 | /* btrfs_commit_tree_roots is responsible for getting the |
1789 | * various roots consistent with each other. Every pointer | |
1790 | * in the tree of tree roots has to point to the most up to date | |
1791 | * root for every subvolume and other tree. So, we have to keep | |
1792 | * the tree logging code from jumping in and changing any | |
1793 | * of the trees. | |
1794 | * | |
1795 | * At this point in the commit, there can't be any tree-log | |
1796 | * writers, but a little lower down we drop the trans mutex | |
1797 | * and let new people in. By holding the tree_log_mutex | |
1798 | * from now until after the super is written, we avoid races | |
1799 | * with the tree-log code. | |
1800 | */ | |
1801 | mutex_lock(&root->fs_info->tree_log_mutex); | |
1802 | ||
5d4f98a2 | 1803 | ret = commit_fs_roots(trans, root); |
49b25e05 JM |
1804 | if (ret) { |
1805 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
871383be | 1806 | mutex_unlock(&root->fs_info->reloc_mutex); |
49b25e05 JM |
1807 | goto cleanup_transaction; |
1808 | } | |
54aa1f4d | 1809 | |
5d4f98a2 | 1810 | /* commit_fs_roots gets rid of all the tree log roots, it is now |
e02119d5 CM |
1811 | * safe to free the root of tree log roots |
1812 | */ | |
1813 | btrfs_free_log_root_tree(trans, root->fs_info); | |
1814 | ||
5d4f98a2 | 1815 | ret = commit_cowonly_roots(trans, root); |
49b25e05 JM |
1816 | if (ret) { |
1817 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
871383be | 1818 | mutex_unlock(&root->fs_info->reloc_mutex); |
49b25e05 JM |
1819 | goto cleanup_transaction; |
1820 | } | |
54aa1f4d | 1821 | |
2cba30f1 MX |
1822 | /* |
1823 | * The tasks which save the space cache and inode cache may also | |
1824 | * update ->aborted, check it. | |
1825 | */ | |
1826 | if (unlikely(ACCESS_ONCE(cur_trans->aborted))) { | |
1827 | ret = cur_trans->aborted; | |
1828 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
1829 | mutex_unlock(&root->fs_info->reloc_mutex); | |
1830 | goto cleanup_transaction; | |
1831 | } | |
1832 | ||
11833d66 YZ |
1833 | btrfs_prepare_extent_commit(trans, root); |
1834 | ||
78fae27e | 1835 | cur_trans = root->fs_info->running_transaction; |
5d4f98a2 YZ |
1836 | |
1837 | btrfs_set_root_node(&root->fs_info->tree_root->root_item, | |
1838 | root->fs_info->tree_root->node); | |
817d52f8 | 1839 | switch_commit_root(root->fs_info->tree_root); |
5d4f98a2 YZ |
1840 | |
1841 | btrfs_set_root_node(&root->fs_info->chunk_root->root_item, | |
1842 | root->fs_info->chunk_root->node); | |
817d52f8 | 1843 | switch_commit_root(root->fs_info->chunk_root); |
5d4f98a2 | 1844 | |
edf39272 | 1845 | assert_qgroups_uptodate(trans); |
5d4f98a2 | 1846 | update_super_roots(root); |
e02119d5 | 1847 | |
60e7cd3a JB |
1848 | btrfs_set_super_log_root(root->fs_info->super_copy, 0); |
1849 | btrfs_set_super_log_root_level(root->fs_info->super_copy, 0); | |
6c41761f DS |
1850 | memcpy(root->fs_info->super_for_commit, root->fs_info->super_copy, |
1851 | sizeof(*root->fs_info->super_copy)); | |
ccd467d6 | 1852 | |
a4abeea4 | 1853 | spin_lock(&root->fs_info->trans_lock); |
4a9d8bde | 1854 | cur_trans->state = TRANS_STATE_UNBLOCKED; |
a4abeea4 | 1855 | root->fs_info->running_transaction = NULL; |
a4abeea4 | 1856 | spin_unlock(&root->fs_info->trans_lock); |
7585717f | 1857 | mutex_unlock(&root->fs_info->reloc_mutex); |
b7ec40d7 | 1858 | |
f9295749 | 1859 | wake_up(&root->fs_info->transaction_wait); |
e6dcd2dc | 1860 | |
79154b1b | 1861 | ret = btrfs_write_and_wait_transaction(trans, root); |
49b25e05 JM |
1862 | if (ret) { |
1863 | btrfs_error(root->fs_info, ret, | |
08748810 | 1864 | "Error while writing out transaction"); |
49b25e05 JM |
1865 | mutex_unlock(&root->fs_info->tree_log_mutex); |
1866 | goto cleanup_transaction; | |
1867 | } | |
1868 | ||
1869 | ret = write_ctree_super(trans, root, 0); | |
1870 | if (ret) { | |
1871 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
1872 | goto cleanup_transaction; | |
1873 | } | |
4313b399 | 1874 | |
e02119d5 CM |
1875 | /* |
1876 | * the super is written, we can safely allow the tree-loggers | |
1877 | * to go about their business | |
1878 | */ | |
1879 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
1880 | ||
11833d66 | 1881 | btrfs_finish_extent_commit(trans, root); |
4313b399 | 1882 | |
15ee9bc7 | 1883 | root->fs_info->last_trans_committed = cur_trans->transid; |
4a9d8bde MX |
1884 | /* |
1885 | * We needn't acquire the lock here because there is no other task | |
1886 | * which can change it. | |
1887 | */ | |
1888 | cur_trans->state = TRANS_STATE_COMPLETED; | |
2c90e5d6 | 1889 | wake_up(&cur_trans->commit_wait); |
3de4586c | 1890 | |
a4abeea4 | 1891 | spin_lock(&root->fs_info->trans_lock); |
13c5a93e | 1892 | list_del_init(&cur_trans->list); |
a4abeea4 JB |
1893 | spin_unlock(&root->fs_info->trans_lock); |
1894 | ||
78fae27e | 1895 | put_transaction(cur_trans); |
79154b1b | 1896 | put_transaction(cur_trans); |
58176a96 | 1897 | |
0860adfd | 1898 | if (trans->type & __TRANS_FREEZABLE) |
354aa0fb | 1899 | sb_end_intwrite(root->fs_info->sb); |
b2b5ef5c | 1900 | |
1abe9b8a | 1901 | trace_btrfs_transaction_commit(root); |
1902 | ||
a2de733c AJ |
1903 | btrfs_scrub_continue(root); |
1904 | ||
9ed74f2d JB |
1905 | if (current->journal_info == trans) |
1906 | current->journal_info = NULL; | |
1907 | ||
2c90e5d6 | 1908 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
24bbcf04 YZ |
1909 | |
1910 | if (current != root->fs_info->transaction_kthread) | |
1911 | btrfs_run_delayed_iputs(root); | |
1912 | ||
79154b1b | 1913 | return ret; |
49b25e05 JM |
1914 | |
1915 | cleanup_transaction: | |
0e721106 JB |
1916 | btrfs_trans_release_metadata(trans, root); |
1917 | trans->block_rsv = NULL; | |
272d26d0 MX |
1918 | if (trans->qgroup_reserved) { |
1919 | btrfs_qgroup_free(root, trans->qgroup_reserved); | |
1920 | trans->qgroup_reserved = 0; | |
1921 | } | |
c2cf52eb | 1922 | btrfs_warn(root->fs_info, "Skipping commit of aborted transaction."); |
49b25e05 JM |
1923 | if (current->journal_info == trans) |
1924 | current->journal_info = NULL; | |
7b8b92af | 1925 | cleanup_transaction(trans, root, ret); |
49b25e05 JM |
1926 | |
1927 | return ret; | |
79154b1b CM |
1928 | } |
1929 | ||
d352ac68 | 1930 | /* |
9d1a2a3a DS |
1931 | * return < 0 if error |
1932 | * 0 if there are no more dead_roots at the time of call | |
1933 | * 1 there are more to be processed, call me again | |
1934 | * | |
1935 | * The return value indicates there are certainly more snapshots to delete, but | |
1936 | * if there comes a new one during processing, it may return 0. We don't mind, | |
1937 | * because btrfs_commit_super will poke cleaner thread and it will process it a | |
1938 | * few seconds later. | |
d352ac68 | 1939 | */ |
9d1a2a3a | 1940 | int btrfs_clean_one_deleted_snapshot(struct btrfs_root *root) |
e9d0b13b | 1941 | { |
9d1a2a3a | 1942 | int ret; |
5d4f98a2 YZ |
1943 | struct btrfs_fs_info *fs_info = root->fs_info; |
1944 | ||
a4abeea4 | 1945 | spin_lock(&fs_info->trans_lock); |
9d1a2a3a DS |
1946 | if (list_empty(&fs_info->dead_roots)) { |
1947 | spin_unlock(&fs_info->trans_lock); | |
1948 | return 0; | |
1949 | } | |
1950 | root = list_first_entry(&fs_info->dead_roots, | |
1951 | struct btrfs_root, root_list); | |
cfad392b | 1952 | list_del_init(&root->root_list); |
a4abeea4 | 1953 | spin_unlock(&fs_info->trans_lock); |
e9d0b13b | 1954 | |
c1c9ff7c | 1955 | pr_debug("btrfs: cleaner removing %llu\n", root->objectid); |
76dda93c | 1956 | |
9d1a2a3a | 1957 | btrfs_kill_all_delayed_nodes(root); |
16cdcec7 | 1958 | |
9d1a2a3a DS |
1959 | if (btrfs_header_backref_rev(root->node) < |
1960 | BTRFS_MIXED_BACKREF_REV) | |
1961 | ret = btrfs_drop_snapshot(root, NULL, 0, 0); | |
1962 | else | |
1963 | ret = btrfs_drop_snapshot(root, NULL, 1, 0); | |
1964 | /* | |
1965 | * If we encounter a transaction abort during snapshot cleaning, we | |
1966 | * don't want to crash here | |
1967 | */ | |
6596a928 | 1968 | return (ret < 0) ? 0 : 1; |
e9d0b13b | 1969 | } |