Commit | Line | Data |
---|---|---|
6cbd5570 CM |
1 | /* |
2 | * Copyright (C) 2007 Oracle. All rights reserved. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public | |
6 | * License v2 as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, | |
9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public | |
14 | * License along with this program; if not, write to the | |
15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
16 | * Boston, MA 021110-1307, USA. | |
17 | */ | |
18 | ||
79154b1b | 19 | #include <linux/fs.h> |
5a0e3ad6 | 20 | #include <linux/slab.h> |
34088780 | 21 | #include <linux/sched.h> |
d3c2fdcf | 22 | #include <linux/writeback.h> |
5f39d397 | 23 | #include <linux/pagemap.h> |
5f2cc086 | 24 | #include <linux/blkdev.h> |
8ea05e3a | 25 | #include <linux/uuid.h> |
79154b1b CM |
26 | #include "ctree.h" |
27 | #include "disk-io.h" | |
28 | #include "transaction.h" | |
925baedd | 29 | #include "locking.h" |
e02119d5 | 30 | #include "tree-log.h" |
581bb050 | 31 | #include "inode-map.h" |
733f4fbb | 32 | #include "volumes.h" |
8dabb742 | 33 | #include "dev-replace.h" |
79154b1b | 34 | |
0f7d52f4 CM |
35 | #define BTRFS_ROOT_TRANS_TAG 0 |
36 | ||
4a9d8bde MX |
37 | static unsigned int btrfs_blocked_trans_types[TRANS_STATE_MAX] = { |
38 | [TRANS_STATE_RUNNING] = 0U, | |
39 | [TRANS_STATE_BLOCKED] = (__TRANS_USERSPACE | | |
40 | __TRANS_START), | |
41 | [TRANS_STATE_COMMIT_START] = (__TRANS_USERSPACE | | |
42 | __TRANS_START | | |
43 | __TRANS_ATTACH), | |
44 | [TRANS_STATE_COMMIT_DOING] = (__TRANS_USERSPACE | | |
45 | __TRANS_START | | |
46 | __TRANS_ATTACH | | |
47 | __TRANS_JOIN), | |
48 | [TRANS_STATE_UNBLOCKED] = (__TRANS_USERSPACE | | |
49 | __TRANS_START | | |
50 | __TRANS_ATTACH | | |
51 | __TRANS_JOIN | | |
52 | __TRANS_JOIN_NOLOCK), | |
53 | [TRANS_STATE_COMPLETED] = (__TRANS_USERSPACE | | |
54 | __TRANS_START | | |
55 | __TRANS_ATTACH | | |
56 | __TRANS_JOIN | | |
57 | __TRANS_JOIN_NOLOCK), | |
58 | }; | |
59 | ||
724e2315 | 60 | void btrfs_put_transaction(struct btrfs_transaction *transaction) |
79154b1b | 61 | { |
13c5a93e JB |
62 | WARN_ON(atomic_read(&transaction->use_count) == 0); |
63 | if (atomic_dec_and_test(&transaction->use_count)) { | |
a4abeea4 | 64 | BUG_ON(!list_empty(&transaction->list)); |
c46effa6 | 65 | WARN_ON(!RB_EMPTY_ROOT(&transaction->delayed_refs.href_root)); |
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 | ||
c46effa6 | 186 | cur_trans->delayed_refs.href_root = RB_ROOT; |
d7df2c79 | 187 | atomic_set(&cur_trans->delayed_refs.num_entries, 0); |
a4abeea4 JB |
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 | 198 | if (!list_empty(&fs_info->tree_mod_seq_list)) |
efe120a0 | 199 | WARN(1, KERN_ERR "BTRFS: tree_mod_seq_list not empty when " |
20b297d6 | 200 | "creating a fresh transaction\n"); |
31b1a2bd | 201 | if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log)) |
efe120a0 | 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 JB |
206 | spin_lock_init(&cur_trans->delayed_refs.lock); |
207 | ||
208 | INIT_LIST_HEAD(&cur_trans->pending_snapshots); | |
569e0f35 | 209 | INIT_LIST_HEAD(&cur_trans->ordered_operations); |
6df9a95e | 210 | INIT_LIST_HEAD(&cur_trans->pending_chunks); |
19ae4e81 | 211 | list_add_tail(&cur_trans->list, &fs_info->trans_list); |
a4abeea4 | 212 | extent_io_tree_init(&cur_trans->dirty_pages, |
19ae4e81 JS |
213 | fs_info->btree_inode->i_mapping); |
214 | fs_info->generation++; | |
215 | cur_trans->transid = fs_info->generation; | |
216 | fs_info->running_transaction = cur_trans; | |
49b25e05 | 217 | cur_trans->aborted = 0; |
19ae4e81 | 218 | spin_unlock(&fs_info->trans_lock); |
15ee9bc7 | 219 | |
79154b1b CM |
220 | return 0; |
221 | } | |
222 | ||
d352ac68 | 223 | /* |
d397712b CM |
224 | * this does all the record keeping required to make sure that a reference |
225 | * counted root is properly recorded in a given transaction. This is required | |
226 | * to make sure the old root from before we joined the transaction is deleted | |
227 | * when the transaction commits | |
d352ac68 | 228 | */ |
7585717f | 229 | static int record_root_in_trans(struct btrfs_trans_handle *trans, |
a4abeea4 | 230 | struct btrfs_root *root) |
6702ed49 | 231 | { |
5d4f98a2 | 232 | if (root->ref_cows && root->last_trans < trans->transid) { |
6702ed49 | 233 | WARN_ON(root == root->fs_info->extent_root); |
5d4f98a2 YZ |
234 | WARN_ON(root->commit_root != root->node); |
235 | ||
7585717f CM |
236 | /* |
237 | * see below for in_trans_setup usage rules | |
238 | * we have the reloc mutex held now, so there | |
239 | * is only one writer in this function | |
240 | */ | |
241 | root->in_trans_setup = 1; | |
242 | ||
243 | /* make sure readers find in_trans_setup before | |
244 | * they find our root->last_trans update | |
245 | */ | |
246 | smp_wmb(); | |
247 | ||
a4abeea4 JB |
248 | spin_lock(&root->fs_info->fs_roots_radix_lock); |
249 | if (root->last_trans == trans->transid) { | |
250 | spin_unlock(&root->fs_info->fs_roots_radix_lock); | |
251 | return 0; | |
252 | } | |
5d4f98a2 YZ |
253 | radix_tree_tag_set(&root->fs_info->fs_roots_radix, |
254 | (unsigned long)root->root_key.objectid, | |
255 | BTRFS_ROOT_TRANS_TAG); | |
a4abeea4 | 256 | spin_unlock(&root->fs_info->fs_roots_radix_lock); |
7585717f CM |
257 | root->last_trans = trans->transid; |
258 | ||
259 | /* this is pretty tricky. We don't want to | |
260 | * take the relocation lock in btrfs_record_root_in_trans | |
261 | * unless we're really doing the first setup for this root in | |
262 | * this transaction. | |
263 | * | |
264 | * Normally we'd use root->last_trans as a flag to decide | |
265 | * if we want to take the expensive mutex. | |
266 | * | |
267 | * But, we have to set root->last_trans before we | |
268 | * init the relocation root, otherwise, we trip over warnings | |
269 | * in ctree.c. The solution used here is to flag ourselves | |
270 | * with root->in_trans_setup. When this is 1, we're still | |
271 | * fixing up the reloc trees and everyone must wait. | |
272 | * | |
273 | * When this is zero, they can trust root->last_trans and fly | |
274 | * through btrfs_record_root_in_trans without having to take the | |
275 | * lock. smp_wmb() makes sure that all the writes above are | |
276 | * done before we pop in the zero below | |
277 | */ | |
5d4f98a2 | 278 | btrfs_init_reloc_root(trans, root); |
7585717f CM |
279 | smp_wmb(); |
280 | root->in_trans_setup = 0; | |
5d4f98a2 YZ |
281 | } |
282 | return 0; | |
283 | } | |
bcc63abb | 284 | |
7585717f CM |
285 | |
286 | int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, | |
287 | struct btrfs_root *root) | |
288 | { | |
289 | if (!root->ref_cows) | |
290 | return 0; | |
291 | ||
292 | /* | |
293 | * see record_root_in_trans for comments about in_trans_setup usage | |
294 | * and barriers | |
295 | */ | |
296 | smp_rmb(); | |
297 | if (root->last_trans == trans->transid && | |
298 | !root->in_trans_setup) | |
299 | return 0; | |
300 | ||
301 | mutex_lock(&root->fs_info->reloc_mutex); | |
302 | record_root_in_trans(trans, root); | |
303 | mutex_unlock(&root->fs_info->reloc_mutex); | |
304 | ||
305 | return 0; | |
306 | } | |
307 | ||
4a9d8bde MX |
308 | static inline int is_transaction_blocked(struct btrfs_transaction *trans) |
309 | { | |
310 | return (trans->state >= TRANS_STATE_BLOCKED && | |
501407aa JB |
311 | trans->state < TRANS_STATE_UNBLOCKED && |
312 | !trans->aborted); | |
4a9d8bde MX |
313 | } |
314 | ||
d352ac68 CM |
315 | /* wait for commit against the current transaction to become unblocked |
316 | * when this is done, it is safe to start a new transaction, but the current | |
317 | * transaction might not be fully on disk. | |
318 | */ | |
37d1aeee | 319 | static void wait_current_trans(struct btrfs_root *root) |
79154b1b | 320 | { |
f9295749 | 321 | struct btrfs_transaction *cur_trans; |
79154b1b | 322 | |
a4abeea4 | 323 | spin_lock(&root->fs_info->trans_lock); |
f9295749 | 324 | cur_trans = root->fs_info->running_transaction; |
4a9d8bde | 325 | if (cur_trans && is_transaction_blocked(cur_trans)) { |
13c5a93e | 326 | atomic_inc(&cur_trans->use_count); |
a4abeea4 | 327 | spin_unlock(&root->fs_info->trans_lock); |
72d63ed6 LZ |
328 | |
329 | wait_event(root->fs_info->transaction_wait, | |
501407aa JB |
330 | cur_trans->state >= TRANS_STATE_UNBLOCKED || |
331 | cur_trans->aborted); | |
724e2315 | 332 | btrfs_put_transaction(cur_trans); |
a4abeea4 JB |
333 | } else { |
334 | spin_unlock(&root->fs_info->trans_lock); | |
f9295749 | 335 | } |
37d1aeee CM |
336 | } |
337 | ||
a22285a6 YZ |
338 | static int may_wait_transaction(struct btrfs_root *root, int type) |
339 | { | |
a4abeea4 JB |
340 | if (root->fs_info->log_root_recovering) |
341 | return 0; | |
342 | ||
343 | if (type == TRANS_USERSPACE) | |
344 | return 1; | |
345 | ||
346 | if (type == TRANS_START && | |
347 | !atomic_read(&root->fs_info->open_ioctl_trans)) | |
a22285a6 | 348 | return 1; |
a4abeea4 | 349 | |
a22285a6 YZ |
350 | return 0; |
351 | } | |
352 | ||
20dd2cbf MX |
353 | static inline bool need_reserve_reloc_root(struct btrfs_root *root) |
354 | { | |
355 | if (!root->fs_info->reloc_ctl || | |
356 | !root->ref_cows || | |
357 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || | |
358 | root->reloc_root) | |
359 | return false; | |
360 | ||
361 | return true; | |
362 | } | |
363 | ||
08e007d2 | 364 | static struct btrfs_trans_handle * |
0860adfd | 365 | start_transaction(struct btrfs_root *root, u64 num_items, unsigned int type, |
08e007d2 | 366 | enum btrfs_reserve_flush_enum flush) |
37d1aeee | 367 | { |
a22285a6 YZ |
368 | struct btrfs_trans_handle *h; |
369 | struct btrfs_transaction *cur_trans; | |
b5009945 | 370 | u64 num_bytes = 0; |
c5567237 | 371 | u64 qgroup_reserved = 0; |
20dd2cbf MX |
372 | bool reloc_reserved = false; |
373 | int ret; | |
acce952b | 374 | |
87533c47 | 375 | if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) |
acce952b | 376 | return ERR_PTR(-EROFS); |
2a1eb461 JB |
377 | |
378 | if (current->journal_info) { | |
0860adfd | 379 | WARN_ON(type & TRANS_EXTWRITERS); |
2a1eb461 JB |
380 | h = current->journal_info; |
381 | h->use_count++; | |
b7d5b0a8 | 382 | WARN_ON(h->use_count > 2); |
2a1eb461 JB |
383 | h->orig_rsv = h->block_rsv; |
384 | h->block_rsv = NULL; | |
385 | goto got_it; | |
386 | } | |
b5009945 JB |
387 | |
388 | /* | |
389 | * Do the reservation before we join the transaction so we can do all | |
390 | * the appropriate flushing if need be. | |
391 | */ | |
392 | if (num_items > 0 && root != root->fs_info->chunk_root) { | |
c5567237 AJ |
393 | if (root->fs_info->quota_enabled && |
394 | is_fstree(root->root_key.objectid)) { | |
395 | qgroup_reserved = num_items * root->leafsize; | |
396 | ret = btrfs_qgroup_reserve(root, qgroup_reserved); | |
397 | if (ret) | |
398 | return ERR_PTR(ret); | |
399 | } | |
400 | ||
b5009945 | 401 | num_bytes = btrfs_calc_trans_metadata_size(root, num_items); |
20dd2cbf MX |
402 | /* |
403 | * Do the reservation for the relocation root creation | |
404 | */ | |
405 | if (unlikely(need_reserve_reloc_root(root))) { | |
406 | num_bytes += root->nodesize; | |
407 | reloc_reserved = true; | |
408 | } | |
409 | ||
08e007d2 MX |
410 | ret = btrfs_block_rsv_add(root, |
411 | &root->fs_info->trans_block_rsv, | |
412 | num_bytes, flush); | |
b5009945 | 413 | if (ret) |
843fcf35 | 414 | goto reserve_fail; |
b5009945 | 415 | } |
a22285a6 YZ |
416 | again: |
417 | h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS); | |
843fcf35 MX |
418 | if (!h) { |
419 | ret = -ENOMEM; | |
420 | goto alloc_fail; | |
421 | } | |
37d1aeee | 422 | |
98114659 JB |
423 | /* |
424 | * If we are JOIN_NOLOCK we're already committing a transaction and | |
425 | * waiting on this guy, so we don't need to do the sb_start_intwrite | |
426 | * because we're already holding a ref. We need this because we could | |
427 | * have raced in and did an fsync() on a file which can kick a commit | |
428 | * and then we deadlock with somebody doing a freeze. | |
354aa0fb MX |
429 | * |
430 | * If we are ATTACH, it means we just want to catch the current | |
431 | * transaction and commit it, so we needn't do sb_start_intwrite(). | |
98114659 | 432 | */ |
0860adfd | 433 | if (type & __TRANS_FREEZABLE) |
60376ce4 | 434 | sb_start_intwrite(root->fs_info->sb); |
b2b5ef5c | 435 | |
a22285a6 | 436 | if (may_wait_transaction(root, type)) |
37d1aeee | 437 | wait_current_trans(root); |
a22285a6 | 438 | |
a4abeea4 | 439 | do { |
354aa0fb | 440 | ret = join_transaction(root, type); |
178260b2 | 441 | if (ret == -EBUSY) { |
a4abeea4 | 442 | wait_current_trans(root); |
178260b2 MX |
443 | if (unlikely(type == TRANS_ATTACH)) |
444 | ret = -ENOENT; | |
445 | } | |
a4abeea4 JB |
446 | } while (ret == -EBUSY); |
447 | ||
db5b493a | 448 | if (ret < 0) { |
354aa0fb MX |
449 | /* We must get the transaction if we are JOIN_NOLOCK. */ |
450 | BUG_ON(type == TRANS_JOIN_NOLOCK); | |
843fcf35 | 451 | goto join_fail; |
db5b493a | 452 | } |
0f7d52f4 | 453 | |
a22285a6 | 454 | cur_trans = root->fs_info->running_transaction; |
a22285a6 YZ |
455 | |
456 | h->transid = cur_trans->transid; | |
457 | h->transaction = cur_trans; | |
79154b1b | 458 | h->blocks_used = 0; |
a22285a6 | 459 | h->bytes_reserved = 0; |
d13603ef | 460 | h->root = root; |
56bec294 | 461 | h->delayed_ref_updates = 0; |
2a1eb461 | 462 | h->use_count = 1; |
0e721106 | 463 | h->adding_csums = 0; |
f0486c68 | 464 | h->block_rsv = NULL; |
2a1eb461 | 465 | h->orig_rsv = NULL; |
49b25e05 | 466 | h->aborted = 0; |
4b824906 | 467 | h->qgroup_reserved = 0; |
bed92eae | 468 | h->delayed_ref_elem.seq = 0; |
a698d075 | 469 | h->type = type; |
c6b305a8 | 470 | h->allocating_chunk = false; |
20dd2cbf | 471 | h->reloc_reserved = false; |
5039eddc | 472 | h->sync = false; |
bed92eae | 473 | INIT_LIST_HEAD(&h->qgroup_ref_list); |
ea658bad | 474 | INIT_LIST_HEAD(&h->new_bgs); |
b7ec40d7 | 475 | |
a22285a6 | 476 | smp_mb(); |
4a9d8bde MX |
477 | if (cur_trans->state >= TRANS_STATE_BLOCKED && |
478 | may_wait_transaction(root, type)) { | |
a22285a6 YZ |
479 | btrfs_commit_transaction(h, root); |
480 | goto again; | |
481 | } | |
482 | ||
b5009945 | 483 | if (num_bytes) { |
8c2a3ca2 | 484 | trace_btrfs_space_reservation(root->fs_info, "transaction", |
2bcc0328 | 485 | h->transid, num_bytes, 1); |
b5009945 JB |
486 | h->block_rsv = &root->fs_info->trans_block_rsv; |
487 | h->bytes_reserved = num_bytes; | |
20dd2cbf | 488 | h->reloc_reserved = reloc_reserved; |
a22285a6 | 489 | } |
4b824906 | 490 | h->qgroup_reserved = qgroup_reserved; |
9ed74f2d | 491 | |
2a1eb461 | 492 | got_it: |
a4abeea4 | 493 | btrfs_record_root_in_trans(h, root); |
a22285a6 YZ |
494 | |
495 | if (!current->journal_info && type != TRANS_USERSPACE) | |
496 | current->journal_info = h; | |
79154b1b | 497 | return h; |
843fcf35 MX |
498 | |
499 | join_fail: | |
0860adfd | 500 | if (type & __TRANS_FREEZABLE) |
843fcf35 MX |
501 | sb_end_intwrite(root->fs_info->sb); |
502 | kmem_cache_free(btrfs_trans_handle_cachep, h); | |
503 | alloc_fail: | |
504 | if (num_bytes) | |
505 | btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv, | |
506 | num_bytes); | |
507 | reserve_fail: | |
508 | if (qgroup_reserved) | |
509 | btrfs_qgroup_free(root, qgroup_reserved); | |
510 | return ERR_PTR(ret); | |
79154b1b CM |
511 | } |
512 | ||
f9295749 | 513 | struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, |
a22285a6 | 514 | int num_items) |
f9295749 | 515 | { |
08e007d2 MX |
516 | return start_transaction(root, num_items, TRANS_START, |
517 | BTRFS_RESERVE_FLUSH_ALL); | |
f9295749 | 518 | } |
8407aa46 | 519 | |
08e007d2 | 520 | struct btrfs_trans_handle *btrfs_start_transaction_lflush( |
8407aa46 MX |
521 | struct btrfs_root *root, int num_items) |
522 | { | |
08e007d2 MX |
523 | return start_transaction(root, num_items, TRANS_START, |
524 | BTRFS_RESERVE_FLUSH_LIMIT); | |
8407aa46 MX |
525 | } |
526 | ||
7a7eaa40 | 527 | struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root) |
f9295749 | 528 | { |
8407aa46 | 529 | return start_transaction(root, 0, TRANS_JOIN, 0); |
f9295749 CM |
530 | } |
531 | ||
7a7eaa40 | 532 | struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root) |
0af3d00b | 533 | { |
8407aa46 | 534 | return start_transaction(root, 0, TRANS_JOIN_NOLOCK, 0); |
0af3d00b JB |
535 | } |
536 | ||
7a7eaa40 | 537 | struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root) |
9ca9ee09 | 538 | { |
8407aa46 | 539 | return start_transaction(root, 0, TRANS_USERSPACE, 0); |
9ca9ee09 SW |
540 | } |
541 | ||
d4edf39b MX |
542 | /* |
543 | * btrfs_attach_transaction() - catch the running transaction | |
544 | * | |
545 | * It is used when we want to commit the current the transaction, but | |
546 | * don't want to start a new one. | |
547 | * | |
548 | * Note: If this function return -ENOENT, it just means there is no | |
549 | * running transaction. But it is possible that the inactive transaction | |
550 | * is still in the memory, not fully on disk. If you hope there is no | |
551 | * inactive transaction in the fs when -ENOENT is returned, you should | |
552 | * invoke | |
553 | * btrfs_attach_transaction_barrier() | |
554 | */ | |
354aa0fb | 555 | struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root) |
60376ce4 | 556 | { |
354aa0fb | 557 | return start_transaction(root, 0, TRANS_ATTACH, 0); |
60376ce4 JB |
558 | } |
559 | ||
d4edf39b | 560 | /* |
90b6d283 | 561 | * btrfs_attach_transaction_barrier() - catch the running transaction |
d4edf39b MX |
562 | * |
563 | * It is similar to the above function, the differentia is this one | |
564 | * will wait for all the inactive transactions until they fully | |
565 | * complete. | |
566 | */ | |
567 | struct btrfs_trans_handle * | |
568 | btrfs_attach_transaction_barrier(struct btrfs_root *root) | |
569 | { | |
570 | struct btrfs_trans_handle *trans; | |
571 | ||
572 | trans = start_transaction(root, 0, TRANS_ATTACH, 0); | |
573 | if (IS_ERR(trans) && PTR_ERR(trans) == -ENOENT) | |
574 | btrfs_wait_for_commit(root, 0); | |
575 | ||
576 | return trans; | |
577 | } | |
578 | ||
d352ac68 | 579 | /* wait for a transaction commit to be fully complete */ |
b9c8300c | 580 | static noinline void wait_for_commit(struct btrfs_root *root, |
89ce8a63 CM |
581 | struct btrfs_transaction *commit) |
582 | { | |
4a9d8bde | 583 | wait_event(commit->commit_wait, commit->state == TRANS_STATE_COMPLETED); |
89ce8a63 CM |
584 | } |
585 | ||
46204592 SW |
586 | int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid) |
587 | { | |
588 | struct btrfs_transaction *cur_trans = NULL, *t; | |
8cd2807f | 589 | int ret = 0; |
46204592 | 590 | |
46204592 SW |
591 | if (transid) { |
592 | if (transid <= root->fs_info->last_trans_committed) | |
a4abeea4 | 593 | goto out; |
46204592 | 594 | |
8cd2807f | 595 | ret = -EINVAL; |
46204592 | 596 | /* find specified transaction */ |
a4abeea4 | 597 | spin_lock(&root->fs_info->trans_lock); |
46204592 SW |
598 | list_for_each_entry(t, &root->fs_info->trans_list, list) { |
599 | if (t->transid == transid) { | |
600 | cur_trans = t; | |
a4abeea4 | 601 | atomic_inc(&cur_trans->use_count); |
8cd2807f | 602 | ret = 0; |
46204592 SW |
603 | break; |
604 | } | |
8cd2807f MX |
605 | if (t->transid > transid) { |
606 | ret = 0; | |
46204592 | 607 | break; |
8cd2807f | 608 | } |
46204592 | 609 | } |
a4abeea4 | 610 | spin_unlock(&root->fs_info->trans_lock); |
8cd2807f | 611 | /* The specified transaction doesn't exist */ |
46204592 | 612 | if (!cur_trans) |
8cd2807f | 613 | goto out; |
46204592 SW |
614 | } else { |
615 | /* find newest transaction that is committing | committed */ | |
a4abeea4 | 616 | spin_lock(&root->fs_info->trans_lock); |
46204592 SW |
617 | list_for_each_entry_reverse(t, &root->fs_info->trans_list, |
618 | list) { | |
4a9d8bde MX |
619 | if (t->state >= TRANS_STATE_COMMIT_START) { |
620 | if (t->state == TRANS_STATE_COMPLETED) | |
3473f3c0 | 621 | break; |
46204592 | 622 | cur_trans = t; |
a4abeea4 | 623 | atomic_inc(&cur_trans->use_count); |
46204592 SW |
624 | break; |
625 | } | |
626 | } | |
a4abeea4 | 627 | spin_unlock(&root->fs_info->trans_lock); |
46204592 | 628 | if (!cur_trans) |
a4abeea4 | 629 | goto out; /* nothing committing|committed */ |
46204592 SW |
630 | } |
631 | ||
46204592 | 632 | wait_for_commit(root, cur_trans); |
724e2315 | 633 | btrfs_put_transaction(cur_trans); |
a4abeea4 | 634 | out: |
46204592 SW |
635 | return ret; |
636 | } | |
637 | ||
37d1aeee CM |
638 | void btrfs_throttle(struct btrfs_root *root) |
639 | { | |
a4abeea4 | 640 | if (!atomic_read(&root->fs_info->open_ioctl_trans)) |
9ca9ee09 | 641 | wait_current_trans(root); |
37d1aeee CM |
642 | } |
643 | ||
8929ecfa YZ |
644 | static int should_end_transaction(struct btrfs_trans_handle *trans, |
645 | struct btrfs_root *root) | |
646 | { | |
1be41b78 | 647 | if (root->fs_info->global_block_rsv.space_info->full && |
0a2b2a84 | 648 | btrfs_check_space_for_delayed_refs(trans, root)) |
1be41b78 | 649 | return 1; |
36ba022a | 650 | |
1be41b78 | 651 | return !!btrfs_block_rsv_check(root, &root->fs_info->global_block_rsv, 5); |
8929ecfa YZ |
652 | } |
653 | ||
654 | int btrfs_should_end_transaction(struct btrfs_trans_handle *trans, | |
655 | struct btrfs_root *root) | |
656 | { | |
657 | struct btrfs_transaction *cur_trans = trans->transaction; | |
658 | int updates; | |
49b25e05 | 659 | int err; |
8929ecfa | 660 | |
a4abeea4 | 661 | smp_mb(); |
4a9d8bde MX |
662 | if (cur_trans->state >= TRANS_STATE_BLOCKED || |
663 | cur_trans->delayed_refs.flushing) | |
8929ecfa YZ |
664 | return 1; |
665 | ||
666 | updates = trans->delayed_ref_updates; | |
667 | trans->delayed_ref_updates = 0; | |
49b25e05 JM |
668 | if (updates) { |
669 | err = btrfs_run_delayed_refs(trans, root, updates); | |
670 | if (err) /* Error code will also eval true */ | |
671 | return err; | |
672 | } | |
8929ecfa YZ |
673 | |
674 | return should_end_transaction(trans, root); | |
675 | } | |
676 | ||
89ce8a63 | 677 | static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, |
a698d075 | 678 | struct btrfs_root *root, int throttle) |
79154b1b | 679 | { |
8929ecfa | 680 | struct btrfs_transaction *cur_trans = trans->transaction; |
ab78c84d | 681 | struct btrfs_fs_info *info = root->fs_info; |
1be41b78 | 682 | unsigned long cur = trans->delayed_ref_updates; |
a698d075 | 683 | int lock = (trans->type != TRANS_JOIN_NOLOCK); |
4edc2ca3 | 684 | int err = 0; |
c3e69d58 | 685 | |
2a1eb461 JB |
686 | if (--trans->use_count) { |
687 | trans->block_rsv = trans->orig_rsv; | |
688 | return 0; | |
689 | } | |
690 | ||
edf39272 JS |
691 | /* |
692 | * do the qgroup accounting as early as possible | |
693 | */ | |
694 | err = btrfs_delayed_refs_qgroup_accounting(trans, info); | |
695 | ||
b24e03db | 696 | btrfs_trans_release_metadata(trans, root); |
4c13d758 | 697 | trans->block_rsv = NULL; |
c5567237 AJ |
698 | |
699 | if (trans->qgroup_reserved) { | |
7c2ec3f0 LB |
700 | /* |
701 | * the same root has to be passed here between start_transaction | |
702 | * and end_transaction. Subvolume quota depends on this. | |
703 | */ | |
704 | btrfs_qgroup_free(trans->root, trans->qgroup_reserved); | |
c5567237 AJ |
705 | trans->qgroup_reserved = 0; |
706 | } | |
707 | ||
ea658bad JB |
708 | if (!list_empty(&trans->new_bgs)) |
709 | btrfs_create_pending_block_groups(trans, root); | |
710 | ||
1be41b78 | 711 | trans->delayed_ref_updates = 0; |
5039eddc | 712 | if (!trans->sync && btrfs_should_throttle_delayed_refs(trans, root)) { |
0a2b2a84 | 713 | cur = max_t(unsigned long, cur, 32); |
c3e69d58 | 714 | trans->delayed_ref_updates = 0; |
1be41b78 | 715 | btrfs_run_delayed_refs(trans, root, cur); |
56bec294 | 716 | } |
bb721703 | 717 | |
0e721106 JB |
718 | btrfs_trans_release_metadata(trans, root); |
719 | trans->block_rsv = NULL; | |
56bec294 | 720 | |
ea658bad JB |
721 | if (!list_empty(&trans->new_bgs)) |
722 | btrfs_create_pending_block_groups(trans, root); | |
723 | ||
a4abeea4 | 724 | if (lock && !atomic_read(&root->fs_info->open_ioctl_trans) && |
4a9d8bde MX |
725 | should_end_transaction(trans, root) && |
726 | ACCESS_ONCE(cur_trans->state) == TRANS_STATE_RUNNING) { | |
727 | spin_lock(&info->trans_lock); | |
728 | if (cur_trans->state == TRANS_STATE_RUNNING) | |
729 | cur_trans->state = TRANS_STATE_BLOCKED; | |
730 | spin_unlock(&info->trans_lock); | |
a4abeea4 | 731 | } |
8929ecfa | 732 | |
4a9d8bde | 733 | if (lock && ACCESS_ONCE(cur_trans->state) == TRANS_STATE_BLOCKED) { |
81317fde JB |
734 | if (throttle) { |
735 | /* | |
736 | * We may race with somebody else here so end up having | |
737 | * to call end_transaction on ourselves again, so inc | |
738 | * our use_count. | |
739 | */ | |
740 | trans->use_count++; | |
8929ecfa | 741 | return btrfs_commit_transaction(trans, root); |
81317fde | 742 | } else { |
8929ecfa | 743 | wake_up_process(info->transaction_kthread); |
81317fde | 744 | } |
8929ecfa YZ |
745 | } |
746 | ||
0860adfd | 747 | if (trans->type & __TRANS_FREEZABLE) |
98114659 | 748 | sb_end_intwrite(root->fs_info->sb); |
6df7881a | 749 | |
8929ecfa | 750 | WARN_ON(cur_trans != info->running_transaction); |
13c5a93e JB |
751 | WARN_ON(atomic_read(&cur_trans->num_writers) < 1); |
752 | atomic_dec(&cur_trans->num_writers); | |
0860adfd | 753 | extwriter_counter_dec(cur_trans, trans->type); |
89ce8a63 | 754 | |
99d16cbc | 755 | smp_mb(); |
79154b1b CM |
756 | if (waitqueue_active(&cur_trans->writer_wait)) |
757 | wake_up(&cur_trans->writer_wait); | |
724e2315 | 758 | btrfs_put_transaction(cur_trans); |
9ed74f2d JB |
759 | |
760 | if (current->journal_info == trans) | |
761 | current->journal_info = NULL; | |
ab78c84d | 762 | |
24bbcf04 YZ |
763 | if (throttle) |
764 | btrfs_run_delayed_iputs(root); | |
765 | ||
49b25e05 | 766 | if (trans->aborted || |
4e121c06 JB |
767 | test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) { |
768 | wake_up_process(info->transaction_kthread); | |
4edc2ca3 | 769 | err = -EIO; |
4e121c06 | 770 | } |
edf39272 | 771 | assert_qgroups_uptodate(trans); |
49b25e05 | 772 | |
4edc2ca3 DJ |
773 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
774 | return err; | |
79154b1b CM |
775 | } |
776 | ||
89ce8a63 CM |
777 | int btrfs_end_transaction(struct btrfs_trans_handle *trans, |
778 | struct btrfs_root *root) | |
779 | { | |
98ad43be | 780 | return __btrfs_end_transaction(trans, root, 0); |
89ce8a63 CM |
781 | } |
782 | ||
783 | int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans, | |
784 | struct btrfs_root *root) | |
785 | { | |
98ad43be | 786 | return __btrfs_end_transaction(trans, root, 1); |
16cdcec7 MX |
787 | } |
788 | ||
d352ac68 CM |
789 | /* |
790 | * when btree blocks are allocated, they have some corresponding bits set for | |
791 | * them in one of two extent_io trees. This is used to make sure all of | |
690587d1 | 792 | * those extents are sent to disk but does not wait on them |
d352ac68 | 793 | */ |
690587d1 | 794 | int btrfs_write_marked_extents(struct btrfs_root *root, |
8cef4e16 | 795 | struct extent_io_tree *dirty_pages, int mark) |
79154b1b | 796 | { |
777e6bd7 | 797 | int err = 0; |
7c4452b9 | 798 | int werr = 0; |
1728366e | 799 | struct address_space *mapping = root->fs_info->btree_inode->i_mapping; |
e6138876 | 800 | struct extent_state *cached_state = NULL; |
777e6bd7 | 801 | u64 start = 0; |
5f39d397 | 802 | u64 end; |
7c4452b9 | 803 | |
1728366e | 804 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
e6138876 JB |
805 | mark, &cached_state)) { |
806 | convert_extent_bit(dirty_pages, start, end, EXTENT_NEED_WAIT, | |
807 | mark, &cached_state, GFP_NOFS); | |
808 | cached_state = NULL; | |
1728366e JB |
809 | err = filemap_fdatawrite_range(mapping, start, end); |
810 | if (err) | |
811 | werr = err; | |
812 | cond_resched(); | |
813 | start = end + 1; | |
7c4452b9 | 814 | } |
690587d1 CM |
815 | if (err) |
816 | werr = err; | |
817 | return werr; | |
818 | } | |
819 | ||
820 | /* | |
821 | * when btree blocks are allocated, they have some corresponding bits set for | |
822 | * them in one of two extent_io trees. This is used to make sure all of | |
823 | * those extents are on disk for transaction or log commit. We wait | |
824 | * on all the pages and clear them from the dirty pages state tree | |
825 | */ | |
826 | int btrfs_wait_marked_extents(struct btrfs_root *root, | |
8cef4e16 | 827 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 | 828 | { |
690587d1 CM |
829 | int err = 0; |
830 | int werr = 0; | |
1728366e | 831 | struct address_space *mapping = root->fs_info->btree_inode->i_mapping; |
e6138876 | 832 | struct extent_state *cached_state = NULL; |
690587d1 CM |
833 | u64 start = 0; |
834 | u64 end; | |
777e6bd7 | 835 | |
1728366e | 836 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
e6138876 JB |
837 | EXTENT_NEED_WAIT, &cached_state)) { |
838 | clear_extent_bit(dirty_pages, start, end, EXTENT_NEED_WAIT, | |
839 | 0, 0, &cached_state, GFP_NOFS); | |
1728366e JB |
840 | err = filemap_fdatawait_range(mapping, start, end); |
841 | if (err) | |
842 | werr = err; | |
843 | cond_resched(); | |
844 | start = end + 1; | |
777e6bd7 | 845 | } |
7c4452b9 CM |
846 | if (err) |
847 | werr = err; | |
848 | return werr; | |
79154b1b CM |
849 | } |
850 | ||
690587d1 CM |
851 | /* |
852 | * when btree blocks are allocated, they have some corresponding bits set for | |
853 | * them in one of two extent_io trees. This is used to make sure all of | |
854 | * those extents are on disk for transaction or log commit | |
855 | */ | |
171170c1 | 856 | static int btrfs_write_and_wait_marked_extents(struct btrfs_root *root, |
8cef4e16 | 857 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 CM |
858 | { |
859 | int ret; | |
860 | int ret2; | |
c6adc9cc | 861 | struct blk_plug plug; |
690587d1 | 862 | |
c6adc9cc | 863 | blk_start_plug(&plug); |
8cef4e16 | 864 | ret = btrfs_write_marked_extents(root, dirty_pages, mark); |
c6adc9cc | 865 | blk_finish_plug(&plug); |
8cef4e16 | 866 | ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark); |
bf0da8c1 CM |
867 | |
868 | if (ret) | |
869 | return ret; | |
870 | if (ret2) | |
871 | return ret2; | |
872 | return 0; | |
690587d1 CM |
873 | } |
874 | ||
d0c803c4 CM |
875 | int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans, |
876 | struct btrfs_root *root) | |
877 | { | |
878 | if (!trans || !trans->transaction) { | |
879 | struct inode *btree_inode; | |
880 | btree_inode = root->fs_info->btree_inode; | |
881 | return filemap_write_and_wait(btree_inode->i_mapping); | |
882 | } | |
883 | return btrfs_write_and_wait_marked_extents(root, | |
8cef4e16 YZ |
884 | &trans->transaction->dirty_pages, |
885 | EXTENT_DIRTY); | |
d0c803c4 CM |
886 | } |
887 | ||
d352ac68 CM |
888 | /* |
889 | * this is used to update the root pointer in the tree of tree roots. | |
890 | * | |
891 | * But, in the case of the extent allocation tree, updating the root | |
892 | * pointer may allocate blocks which may change the root of the extent | |
893 | * allocation tree. | |
894 | * | |
895 | * So, this loops and repeats and makes sure the cowonly root didn't | |
896 | * change while the root pointer was being updated in the metadata. | |
897 | */ | |
0b86a832 CM |
898 | static int update_cowonly_root(struct btrfs_trans_handle *trans, |
899 | struct btrfs_root *root) | |
79154b1b CM |
900 | { |
901 | int ret; | |
0b86a832 | 902 | u64 old_root_bytenr; |
86b9f2ec | 903 | u64 old_root_used; |
0b86a832 | 904 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
79154b1b | 905 | |
86b9f2ec | 906 | old_root_used = btrfs_root_used(&root->root_item); |
0b86a832 | 907 | btrfs_write_dirty_block_groups(trans, root); |
56bec294 | 908 | |
d397712b | 909 | while (1) { |
0b86a832 | 910 | old_root_bytenr = btrfs_root_bytenr(&root->root_item); |
86b9f2ec YZ |
911 | if (old_root_bytenr == root->node->start && |
912 | old_root_used == btrfs_root_used(&root->root_item)) | |
79154b1b | 913 | break; |
87ef2bb4 | 914 | |
5d4f98a2 | 915 | btrfs_set_root_node(&root->root_item, root->node); |
79154b1b | 916 | ret = btrfs_update_root(trans, tree_root, |
0b86a832 CM |
917 | &root->root_key, |
918 | &root->root_item); | |
49b25e05 JM |
919 | if (ret) |
920 | return ret; | |
56bec294 | 921 | |
86b9f2ec | 922 | old_root_used = btrfs_root_used(&root->root_item); |
4a8c9a62 | 923 | ret = btrfs_write_dirty_block_groups(trans, root); |
49b25e05 JM |
924 | if (ret) |
925 | return ret; | |
0b86a832 | 926 | } |
276e680d YZ |
927 | |
928 | if (root != root->fs_info->extent_root) | |
929 | switch_commit_root(root); | |
930 | ||
0b86a832 CM |
931 | return 0; |
932 | } | |
933 | ||
d352ac68 CM |
934 | /* |
935 | * update all the cowonly tree roots on disk | |
49b25e05 JM |
936 | * |
937 | * The error handling in this function may not be obvious. Any of the | |
938 | * failures will cause the file system to go offline. We still need | |
939 | * to clean up the delayed refs. | |
d352ac68 | 940 | */ |
5d4f98a2 YZ |
941 | static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans, |
942 | struct btrfs_root *root) | |
0b86a832 CM |
943 | { |
944 | struct btrfs_fs_info *fs_info = root->fs_info; | |
945 | struct list_head *next; | |
84234f3a | 946 | struct extent_buffer *eb; |
56bec294 | 947 | int ret; |
84234f3a | 948 | |
56bec294 | 949 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
49b25e05 JM |
950 | if (ret) |
951 | return ret; | |
87ef2bb4 | 952 | |
84234f3a | 953 | eb = btrfs_lock_root_node(fs_info->tree_root); |
49b25e05 JM |
954 | ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, |
955 | 0, &eb); | |
84234f3a YZ |
956 | btrfs_tree_unlock(eb); |
957 | free_extent_buffer(eb); | |
0b86a832 | 958 | |
49b25e05 JM |
959 | if (ret) |
960 | return ret; | |
961 | ||
56bec294 | 962 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
49b25e05 JM |
963 | if (ret) |
964 | return ret; | |
87ef2bb4 | 965 | |
733f4fbb | 966 | ret = btrfs_run_dev_stats(trans, root->fs_info); |
c16ce190 JB |
967 | if (ret) |
968 | return ret; | |
8dabb742 | 969 | ret = btrfs_run_dev_replace(trans, root->fs_info); |
c16ce190 JB |
970 | if (ret) |
971 | return ret; | |
546adb0d | 972 | ret = btrfs_run_qgroups(trans, root->fs_info); |
c16ce190 JB |
973 | if (ret) |
974 | return ret; | |
546adb0d JS |
975 | |
976 | /* run_qgroups might have added some more refs */ | |
977 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); | |
c16ce190 JB |
978 | if (ret) |
979 | return ret; | |
546adb0d | 980 | |
d397712b | 981 | while (!list_empty(&fs_info->dirty_cowonly_roots)) { |
0b86a832 CM |
982 | next = fs_info->dirty_cowonly_roots.next; |
983 | list_del_init(next); | |
984 | root = list_entry(next, struct btrfs_root, dirty_list); | |
87ef2bb4 | 985 | |
49b25e05 JM |
986 | ret = update_cowonly_root(trans, root); |
987 | if (ret) | |
988 | return ret; | |
79154b1b | 989 | } |
276e680d YZ |
990 | |
991 | down_write(&fs_info->extent_commit_sem); | |
992 | switch_commit_root(fs_info->extent_root); | |
993 | up_write(&fs_info->extent_commit_sem); | |
994 | ||
8dabb742 SB |
995 | btrfs_after_dev_replace_commit(fs_info); |
996 | ||
79154b1b CM |
997 | return 0; |
998 | } | |
999 | ||
d352ac68 CM |
1000 | /* |
1001 | * dead roots are old snapshots that need to be deleted. This allocates | |
1002 | * a dirty root struct and adds it into the list of dead roots that need to | |
1003 | * be deleted | |
1004 | */ | |
cfad392b | 1005 | void btrfs_add_dead_root(struct btrfs_root *root) |
5eda7b5e | 1006 | { |
a4abeea4 | 1007 | spin_lock(&root->fs_info->trans_lock); |
cfad392b JB |
1008 | if (list_empty(&root->root_list)) |
1009 | list_add_tail(&root->root_list, &root->fs_info->dead_roots); | |
a4abeea4 | 1010 | spin_unlock(&root->fs_info->trans_lock); |
5eda7b5e CM |
1011 | } |
1012 | ||
d352ac68 | 1013 | /* |
5d4f98a2 | 1014 | * update all the cowonly tree roots on disk |
d352ac68 | 1015 | */ |
5d4f98a2 YZ |
1016 | static noinline int commit_fs_roots(struct btrfs_trans_handle *trans, |
1017 | struct btrfs_root *root) | |
0f7d52f4 | 1018 | { |
0f7d52f4 | 1019 | struct btrfs_root *gang[8]; |
5d4f98a2 | 1020 | struct btrfs_fs_info *fs_info = root->fs_info; |
0f7d52f4 CM |
1021 | int i; |
1022 | int ret; | |
54aa1f4d CM |
1023 | int err = 0; |
1024 | ||
a4abeea4 | 1025 | spin_lock(&fs_info->fs_roots_radix_lock); |
d397712b | 1026 | while (1) { |
5d4f98a2 YZ |
1027 | ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, |
1028 | (void **)gang, 0, | |
0f7d52f4 CM |
1029 | ARRAY_SIZE(gang), |
1030 | BTRFS_ROOT_TRANS_TAG); | |
1031 | if (ret == 0) | |
1032 | break; | |
1033 | for (i = 0; i < ret; i++) { | |
1034 | root = gang[i]; | |
5d4f98a2 YZ |
1035 | radix_tree_tag_clear(&fs_info->fs_roots_radix, |
1036 | (unsigned long)root->root_key.objectid, | |
1037 | BTRFS_ROOT_TRANS_TAG); | |
a4abeea4 | 1038 | spin_unlock(&fs_info->fs_roots_radix_lock); |
31153d81 | 1039 | |
e02119d5 | 1040 | btrfs_free_log(trans, root); |
5d4f98a2 | 1041 | btrfs_update_reloc_root(trans, root); |
d68fc57b | 1042 | btrfs_orphan_commit_root(trans, root); |
bcc63abb | 1043 | |
82d5902d LZ |
1044 | btrfs_save_ino_cache(root, trans); |
1045 | ||
f1ebcc74 LB |
1046 | /* see comments in should_cow_block() */ |
1047 | root->force_cow = 0; | |
1048 | smp_wmb(); | |
1049 | ||
978d910d | 1050 | if (root->commit_root != root->node) { |
581bb050 | 1051 | mutex_lock(&root->fs_commit_mutex); |
817d52f8 | 1052 | switch_commit_root(root); |
581bb050 LZ |
1053 | btrfs_unpin_free_ino(root); |
1054 | mutex_unlock(&root->fs_commit_mutex); | |
1055 | ||
978d910d YZ |
1056 | btrfs_set_root_node(&root->root_item, |
1057 | root->node); | |
1058 | } | |
5d4f98a2 | 1059 | |
5d4f98a2 | 1060 | err = btrfs_update_root(trans, fs_info->tree_root, |
0f7d52f4 CM |
1061 | &root->root_key, |
1062 | &root->root_item); | |
a4abeea4 | 1063 | spin_lock(&fs_info->fs_roots_radix_lock); |
54aa1f4d CM |
1064 | if (err) |
1065 | break; | |
0f7d52f4 CM |
1066 | } |
1067 | } | |
a4abeea4 | 1068 | spin_unlock(&fs_info->fs_roots_radix_lock); |
54aa1f4d | 1069 | return err; |
0f7d52f4 CM |
1070 | } |
1071 | ||
d352ac68 | 1072 | /* |
de78b51a ES |
1073 | * defrag a given btree. |
1074 | * Every leaf in the btree is read and defragged. | |
d352ac68 | 1075 | */ |
de78b51a | 1076 | int btrfs_defrag_root(struct btrfs_root *root) |
e9d0b13b CM |
1077 | { |
1078 | struct btrfs_fs_info *info = root->fs_info; | |
e9d0b13b | 1079 | struct btrfs_trans_handle *trans; |
8929ecfa | 1080 | int ret; |
e9d0b13b | 1081 | |
8929ecfa | 1082 | if (xchg(&root->defrag_running, 1)) |
e9d0b13b | 1083 | return 0; |
8929ecfa | 1084 | |
6b80053d | 1085 | while (1) { |
8929ecfa YZ |
1086 | trans = btrfs_start_transaction(root, 0); |
1087 | if (IS_ERR(trans)) | |
1088 | return PTR_ERR(trans); | |
1089 | ||
de78b51a | 1090 | ret = btrfs_defrag_leaves(trans, root); |
8929ecfa | 1091 | |
e9d0b13b | 1092 | btrfs_end_transaction(trans, root); |
b53d3f5d | 1093 | btrfs_btree_balance_dirty(info->tree_root); |
e9d0b13b CM |
1094 | cond_resched(); |
1095 | ||
7841cb28 | 1096 | if (btrfs_fs_closing(root->fs_info) || ret != -EAGAIN) |
e9d0b13b | 1097 | break; |
210549eb DS |
1098 | |
1099 | if (btrfs_defrag_cancelled(root->fs_info)) { | |
efe120a0 | 1100 | pr_debug("BTRFS: defrag_root cancelled\n"); |
210549eb DS |
1101 | ret = -EAGAIN; |
1102 | break; | |
1103 | } | |
e9d0b13b CM |
1104 | } |
1105 | root->defrag_running = 0; | |
8929ecfa | 1106 | return ret; |
e9d0b13b CM |
1107 | } |
1108 | ||
d352ac68 CM |
1109 | /* |
1110 | * new snapshots need to be created at a very specific time in the | |
aec8030a MX |
1111 | * transaction commit. This does the actual creation. |
1112 | * | |
1113 | * Note: | |
1114 | * If the error which may affect the commitment of the current transaction | |
1115 | * happens, we should return the error number. If the error which just affect | |
1116 | * the creation of the pending snapshots, just return 0. | |
d352ac68 | 1117 | */ |
80b6794d | 1118 | static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, |
3063d29f CM |
1119 | struct btrfs_fs_info *fs_info, |
1120 | struct btrfs_pending_snapshot *pending) | |
1121 | { | |
1122 | struct btrfs_key key; | |
80b6794d | 1123 | struct btrfs_root_item *new_root_item; |
3063d29f CM |
1124 | struct btrfs_root *tree_root = fs_info->tree_root; |
1125 | struct btrfs_root *root = pending->root; | |
6bdb72de | 1126 | struct btrfs_root *parent_root; |
98c9942a | 1127 | struct btrfs_block_rsv *rsv; |
6bdb72de | 1128 | struct inode *parent_inode; |
42874b3d MX |
1129 | struct btrfs_path *path; |
1130 | struct btrfs_dir_item *dir_item; | |
a22285a6 | 1131 | struct dentry *dentry; |
3063d29f | 1132 | struct extent_buffer *tmp; |
925baedd | 1133 | struct extent_buffer *old; |
8ea05e3a | 1134 | struct timespec cur_time = CURRENT_TIME; |
aec8030a | 1135 | int ret = 0; |
d68fc57b | 1136 | u64 to_reserve = 0; |
6bdb72de | 1137 | u64 index = 0; |
a22285a6 | 1138 | u64 objectid; |
b83cc969 | 1139 | u64 root_flags; |
8ea05e3a | 1140 | uuid_le new_uuid; |
3063d29f | 1141 | |
42874b3d MX |
1142 | path = btrfs_alloc_path(); |
1143 | if (!path) { | |
aec8030a MX |
1144 | pending->error = -ENOMEM; |
1145 | return 0; | |
42874b3d MX |
1146 | } |
1147 | ||
80b6794d CM |
1148 | new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS); |
1149 | if (!new_root_item) { | |
aec8030a | 1150 | pending->error = -ENOMEM; |
6fa9700e | 1151 | goto root_item_alloc_fail; |
80b6794d | 1152 | } |
a22285a6 | 1153 | |
aec8030a MX |
1154 | pending->error = btrfs_find_free_objectid(tree_root, &objectid); |
1155 | if (pending->error) | |
6fa9700e | 1156 | goto no_free_objectid; |
3063d29f | 1157 | |
3fd0a558 | 1158 | btrfs_reloc_pre_snapshot(trans, pending, &to_reserve); |
d68fc57b YZ |
1159 | |
1160 | if (to_reserve > 0) { | |
aec8030a MX |
1161 | pending->error = btrfs_block_rsv_add(root, |
1162 | &pending->block_rsv, | |
1163 | to_reserve, | |
1164 | BTRFS_RESERVE_NO_FLUSH); | |
1165 | if (pending->error) | |
6fa9700e | 1166 | goto no_free_objectid; |
d68fc57b YZ |
1167 | } |
1168 | ||
aec8030a MX |
1169 | pending->error = btrfs_qgroup_inherit(trans, fs_info, |
1170 | root->root_key.objectid, | |
1171 | objectid, pending->inherit); | |
1172 | if (pending->error) | |
6fa9700e | 1173 | goto no_free_objectid; |
6f72c7e2 | 1174 | |
3063d29f | 1175 | key.objectid = objectid; |
a22285a6 YZ |
1176 | key.offset = (u64)-1; |
1177 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3063d29f | 1178 | |
6fa9700e | 1179 | rsv = trans->block_rsv; |
a22285a6 | 1180 | trans->block_rsv = &pending->block_rsv; |
2382c5cc | 1181 | trans->bytes_reserved = trans->block_rsv->reserved; |
3de4586c | 1182 | |
a22285a6 | 1183 | dentry = pending->dentry; |
e9662f70 | 1184 | parent_inode = pending->dir; |
a22285a6 | 1185 | parent_root = BTRFS_I(parent_inode)->root; |
7585717f | 1186 | record_root_in_trans(trans, parent_root); |
a22285a6 | 1187 | |
3063d29f CM |
1188 | /* |
1189 | * insert the directory item | |
1190 | */ | |
3de4586c | 1191 | ret = btrfs_set_inode_index(parent_inode, &index); |
49b25e05 | 1192 | BUG_ON(ret); /* -ENOMEM */ |
42874b3d MX |
1193 | |
1194 | /* check if there is a file/dir which has the same name. */ | |
1195 | dir_item = btrfs_lookup_dir_item(NULL, parent_root, path, | |
1196 | btrfs_ino(parent_inode), | |
1197 | dentry->d_name.name, | |
1198 | dentry->d_name.len, 0); | |
1199 | if (dir_item != NULL && !IS_ERR(dir_item)) { | |
fe66a05a | 1200 | pending->error = -EEXIST; |
aec8030a | 1201 | goto dir_item_existed; |
42874b3d MX |
1202 | } else if (IS_ERR(dir_item)) { |
1203 | ret = PTR_ERR(dir_item); | |
8732d44f MX |
1204 | btrfs_abort_transaction(trans, root, ret); |
1205 | goto fail; | |
79787eaa | 1206 | } |
42874b3d | 1207 | btrfs_release_path(path); |
52c26179 | 1208 | |
e999376f CM |
1209 | /* |
1210 | * pull in the delayed directory update | |
1211 | * and the delayed inode item | |
1212 | * otherwise we corrupt the FS during | |
1213 | * snapshot | |
1214 | */ | |
1215 | ret = btrfs_run_delayed_items(trans, root); | |
8732d44f MX |
1216 | if (ret) { /* Transaction aborted */ |
1217 | btrfs_abort_transaction(trans, root, ret); | |
1218 | goto fail; | |
1219 | } | |
e999376f | 1220 | |
7585717f | 1221 | record_root_in_trans(trans, root); |
6bdb72de SW |
1222 | btrfs_set_root_last_snapshot(&root->root_item, trans->transid); |
1223 | memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); | |
08fe4db1 | 1224 | btrfs_check_and_init_root_item(new_root_item); |
6bdb72de | 1225 | |
b83cc969 LZ |
1226 | root_flags = btrfs_root_flags(new_root_item); |
1227 | if (pending->readonly) | |
1228 | root_flags |= BTRFS_ROOT_SUBVOL_RDONLY; | |
1229 | else | |
1230 | root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY; | |
1231 | btrfs_set_root_flags(new_root_item, root_flags); | |
1232 | ||
8ea05e3a AB |
1233 | btrfs_set_root_generation_v2(new_root_item, |
1234 | trans->transid); | |
1235 | uuid_le_gen(&new_uuid); | |
1236 | memcpy(new_root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE); | |
1237 | memcpy(new_root_item->parent_uuid, root->root_item.uuid, | |
1238 | BTRFS_UUID_SIZE); | |
70023da2 SB |
1239 | if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) { |
1240 | memset(new_root_item->received_uuid, 0, | |
1241 | sizeof(new_root_item->received_uuid)); | |
1242 | memset(&new_root_item->stime, 0, sizeof(new_root_item->stime)); | |
1243 | memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime)); | |
1244 | btrfs_set_root_stransid(new_root_item, 0); | |
1245 | btrfs_set_root_rtransid(new_root_item, 0); | |
1246 | } | |
3cae210f QW |
1247 | btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec); |
1248 | btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec); | |
8ea05e3a | 1249 | btrfs_set_root_otransid(new_root_item, trans->transid); |
8ea05e3a | 1250 | |
6bdb72de | 1251 | old = btrfs_lock_root_node(root); |
49b25e05 | 1252 | ret = btrfs_cow_block(trans, root, old, NULL, 0, &old); |
79787eaa JM |
1253 | if (ret) { |
1254 | btrfs_tree_unlock(old); | |
1255 | free_extent_buffer(old); | |
8732d44f MX |
1256 | btrfs_abort_transaction(trans, root, ret); |
1257 | goto fail; | |
79787eaa | 1258 | } |
49b25e05 | 1259 | |
6bdb72de SW |
1260 | btrfs_set_lock_blocking(old); |
1261 | ||
49b25e05 | 1262 | ret = btrfs_copy_root(trans, root, old, &tmp, objectid); |
79787eaa | 1263 | /* clean up in any case */ |
6bdb72de SW |
1264 | btrfs_tree_unlock(old); |
1265 | free_extent_buffer(old); | |
8732d44f MX |
1266 | if (ret) { |
1267 | btrfs_abort_transaction(trans, root, ret); | |
1268 | goto fail; | |
1269 | } | |
6bdb72de | 1270 | |
f1ebcc74 LB |
1271 | /* see comments in should_cow_block() */ |
1272 | root->force_cow = 1; | |
1273 | smp_wmb(); | |
1274 | ||
6bdb72de | 1275 | btrfs_set_root_node(new_root_item, tmp); |
a22285a6 YZ |
1276 | /* record when the snapshot was created in key.offset */ |
1277 | key.offset = trans->transid; | |
1278 | ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); | |
6bdb72de SW |
1279 | btrfs_tree_unlock(tmp); |
1280 | free_extent_buffer(tmp); | |
8732d44f MX |
1281 | if (ret) { |
1282 | btrfs_abort_transaction(trans, root, ret); | |
1283 | goto fail; | |
1284 | } | |
6bdb72de | 1285 | |
a22285a6 YZ |
1286 | /* |
1287 | * insert root back/forward references | |
1288 | */ | |
1289 | ret = btrfs_add_root_ref(trans, tree_root, objectid, | |
0660b5af | 1290 | parent_root->root_key.objectid, |
33345d01 | 1291 | btrfs_ino(parent_inode), index, |
a22285a6 | 1292 | dentry->d_name.name, dentry->d_name.len); |
8732d44f MX |
1293 | if (ret) { |
1294 | btrfs_abort_transaction(trans, root, ret); | |
1295 | goto fail; | |
1296 | } | |
0660b5af | 1297 | |
a22285a6 YZ |
1298 | key.offset = (u64)-1; |
1299 | pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key); | |
79787eaa JM |
1300 | if (IS_ERR(pending->snap)) { |
1301 | ret = PTR_ERR(pending->snap); | |
8732d44f MX |
1302 | btrfs_abort_transaction(trans, root, ret); |
1303 | goto fail; | |
79787eaa | 1304 | } |
d68fc57b | 1305 | |
49b25e05 | 1306 | ret = btrfs_reloc_post_snapshot(trans, pending); |
8732d44f MX |
1307 | if (ret) { |
1308 | btrfs_abort_transaction(trans, root, ret); | |
1309 | goto fail; | |
1310 | } | |
361048f5 MX |
1311 | |
1312 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); | |
8732d44f MX |
1313 | if (ret) { |
1314 | btrfs_abort_transaction(trans, root, ret); | |
1315 | goto fail; | |
1316 | } | |
42874b3d MX |
1317 | |
1318 | ret = btrfs_insert_dir_item(trans, parent_root, | |
1319 | dentry->d_name.name, dentry->d_name.len, | |
1320 | parent_inode, &key, | |
1321 | BTRFS_FT_DIR, index); | |
1322 | /* We have check then name at the beginning, so it is impossible. */ | |
9c52057c | 1323 | BUG_ON(ret == -EEXIST || ret == -EOVERFLOW); |
8732d44f MX |
1324 | if (ret) { |
1325 | btrfs_abort_transaction(trans, root, ret); | |
1326 | goto fail; | |
1327 | } | |
42874b3d MX |
1328 | |
1329 | btrfs_i_size_write(parent_inode, parent_inode->i_size + | |
1330 | dentry->d_name.len * 2); | |
1331 | parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME; | |
be6aef60 | 1332 | ret = btrfs_update_inode_fallback(trans, parent_root, parent_inode); |
dd5f9615 SB |
1333 | if (ret) { |
1334 | btrfs_abort_transaction(trans, root, ret); | |
1335 | goto fail; | |
1336 | } | |
1337 | ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root, new_uuid.b, | |
1338 | BTRFS_UUID_KEY_SUBVOL, objectid); | |
1339 | if (ret) { | |
8732d44f | 1340 | btrfs_abort_transaction(trans, root, ret); |
dd5f9615 SB |
1341 | goto fail; |
1342 | } | |
1343 | if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) { | |
1344 | ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root, | |
1345 | new_root_item->received_uuid, | |
1346 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, | |
1347 | objectid); | |
1348 | if (ret && ret != -EEXIST) { | |
1349 | btrfs_abort_transaction(trans, root, ret); | |
1350 | goto fail; | |
1351 | } | |
1352 | } | |
3063d29f | 1353 | fail: |
aec8030a MX |
1354 | pending->error = ret; |
1355 | dir_item_existed: | |
98c9942a | 1356 | trans->block_rsv = rsv; |
2382c5cc | 1357 | trans->bytes_reserved = 0; |
6fa9700e MX |
1358 | no_free_objectid: |
1359 | kfree(new_root_item); | |
1360 | root_item_alloc_fail: | |
42874b3d | 1361 | btrfs_free_path(path); |
49b25e05 | 1362 | return ret; |
3063d29f CM |
1363 | } |
1364 | ||
d352ac68 CM |
1365 | /* |
1366 | * create all the snapshots we've scheduled for creation | |
1367 | */ | |
80b6794d CM |
1368 | static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans, |
1369 | struct btrfs_fs_info *fs_info) | |
3de4586c | 1370 | { |
aec8030a | 1371 | struct btrfs_pending_snapshot *pending, *next; |
3de4586c | 1372 | struct list_head *head = &trans->transaction->pending_snapshots; |
aec8030a | 1373 | int ret = 0; |
3de4586c | 1374 | |
aec8030a MX |
1375 | list_for_each_entry_safe(pending, next, head, list) { |
1376 | list_del(&pending->list); | |
1377 | ret = create_pending_snapshot(trans, fs_info, pending); | |
1378 | if (ret) | |
1379 | break; | |
1380 | } | |
1381 | return ret; | |
3de4586c CM |
1382 | } |
1383 | ||
5d4f98a2 YZ |
1384 | static void update_super_roots(struct btrfs_root *root) |
1385 | { | |
1386 | struct btrfs_root_item *root_item; | |
1387 | struct btrfs_super_block *super; | |
1388 | ||
6c41761f | 1389 | super = root->fs_info->super_copy; |
5d4f98a2 YZ |
1390 | |
1391 | root_item = &root->fs_info->chunk_root->root_item; | |
1392 | super->chunk_root = root_item->bytenr; | |
1393 | super->chunk_root_generation = root_item->generation; | |
1394 | super->chunk_root_level = root_item->level; | |
1395 | ||
1396 | root_item = &root->fs_info->tree_root->root_item; | |
1397 | super->root = root_item->bytenr; | |
1398 | super->generation = root_item->generation; | |
1399 | super->root_level = root_item->level; | |
73bc1876 | 1400 | if (btrfs_test_opt(root, SPACE_CACHE)) |
0af3d00b | 1401 | super->cache_generation = root_item->generation; |
70f80175 SB |
1402 | if (root->fs_info->update_uuid_tree_gen) |
1403 | super->uuid_tree_generation = root_item->generation; | |
5d4f98a2 YZ |
1404 | } |
1405 | ||
f36f3042 CM |
1406 | int btrfs_transaction_in_commit(struct btrfs_fs_info *info) |
1407 | { | |
4a9d8bde | 1408 | struct btrfs_transaction *trans; |
f36f3042 | 1409 | int ret = 0; |
4a9d8bde | 1410 | |
a4abeea4 | 1411 | spin_lock(&info->trans_lock); |
4a9d8bde MX |
1412 | trans = info->running_transaction; |
1413 | if (trans) | |
1414 | ret = (trans->state >= TRANS_STATE_COMMIT_START); | |
a4abeea4 | 1415 | spin_unlock(&info->trans_lock); |
f36f3042 CM |
1416 | return ret; |
1417 | } | |
1418 | ||
8929ecfa YZ |
1419 | int btrfs_transaction_blocked(struct btrfs_fs_info *info) |
1420 | { | |
4a9d8bde | 1421 | struct btrfs_transaction *trans; |
8929ecfa | 1422 | int ret = 0; |
4a9d8bde | 1423 | |
a4abeea4 | 1424 | spin_lock(&info->trans_lock); |
4a9d8bde MX |
1425 | trans = info->running_transaction; |
1426 | if (trans) | |
1427 | ret = is_transaction_blocked(trans); | |
a4abeea4 | 1428 | spin_unlock(&info->trans_lock); |
8929ecfa YZ |
1429 | return ret; |
1430 | } | |
1431 | ||
bb9c12c9 SW |
1432 | /* |
1433 | * wait for the current transaction commit to start and block subsequent | |
1434 | * transaction joins | |
1435 | */ | |
1436 | static void wait_current_trans_commit_start(struct btrfs_root *root, | |
1437 | struct btrfs_transaction *trans) | |
1438 | { | |
4a9d8bde | 1439 | wait_event(root->fs_info->transaction_blocked_wait, |
501407aa JB |
1440 | trans->state >= TRANS_STATE_COMMIT_START || |
1441 | trans->aborted); | |
bb9c12c9 SW |
1442 | } |
1443 | ||
1444 | /* | |
1445 | * wait for the current transaction to start and then become unblocked. | |
1446 | * caller holds ref. | |
1447 | */ | |
1448 | static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root, | |
1449 | struct btrfs_transaction *trans) | |
1450 | { | |
72d63ed6 | 1451 | wait_event(root->fs_info->transaction_wait, |
501407aa JB |
1452 | trans->state >= TRANS_STATE_UNBLOCKED || |
1453 | trans->aborted); | |
bb9c12c9 SW |
1454 | } |
1455 | ||
1456 | /* | |
1457 | * commit transactions asynchronously. once btrfs_commit_transaction_async | |
1458 | * returns, any subsequent transaction will not be allowed to join. | |
1459 | */ | |
1460 | struct btrfs_async_commit { | |
1461 | struct btrfs_trans_handle *newtrans; | |
1462 | struct btrfs_root *root; | |
7892b5af | 1463 | struct work_struct work; |
bb9c12c9 SW |
1464 | }; |
1465 | ||
1466 | static void do_async_commit(struct work_struct *work) | |
1467 | { | |
1468 | struct btrfs_async_commit *ac = | |
7892b5af | 1469 | container_of(work, struct btrfs_async_commit, work); |
bb9c12c9 | 1470 | |
6fc4e354 SW |
1471 | /* |
1472 | * We've got freeze protection passed with the transaction. | |
1473 | * Tell lockdep about it. | |
1474 | */ | |
b1a06a4b | 1475 | if (ac->newtrans->type & __TRANS_FREEZABLE) |
ff7c1d33 MX |
1476 | rwsem_acquire_read( |
1477 | &ac->root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1], | |
1478 | 0, 1, _THIS_IP_); | |
6fc4e354 | 1479 | |
e209db7a SW |
1480 | current->journal_info = ac->newtrans; |
1481 | ||
bb9c12c9 SW |
1482 | btrfs_commit_transaction(ac->newtrans, ac->root); |
1483 | kfree(ac); | |
1484 | } | |
1485 | ||
1486 | int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans, | |
1487 | struct btrfs_root *root, | |
1488 | int wait_for_unblock) | |
1489 | { | |
1490 | struct btrfs_async_commit *ac; | |
1491 | struct btrfs_transaction *cur_trans; | |
1492 | ||
1493 | ac = kmalloc(sizeof(*ac), GFP_NOFS); | |
db5b493a TI |
1494 | if (!ac) |
1495 | return -ENOMEM; | |
bb9c12c9 | 1496 | |
7892b5af | 1497 | INIT_WORK(&ac->work, do_async_commit); |
bb9c12c9 | 1498 | ac->root = root; |
7a7eaa40 | 1499 | ac->newtrans = btrfs_join_transaction(root); |
3612b495 TI |
1500 | if (IS_ERR(ac->newtrans)) { |
1501 | int err = PTR_ERR(ac->newtrans); | |
1502 | kfree(ac); | |
1503 | return err; | |
1504 | } | |
bb9c12c9 SW |
1505 | |
1506 | /* take transaction reference */ | |
bb9c12c9 | 1507 | cur_trans = trans->transaction; |
13c5a93e | 1508 | atomic_inc(&cur_trans->use_count); |
bb9c12c9 SW |
1509 | |
1510 | btrfs_end_transaction(trans, root); | |
6fc4e354 SW |
1511 | |
1512 | /* | |
1513 | * Tell lockdep we've released the freeze rwsem, since the | |
1514 | * async commit thread will be the one to unlock it. | |
1515 | */ | |
b1a06a4b | 1516 | if (ac->newtrans->type & __TRANS_FREEZABLE) |
ff7c1d33 MX |
1517 | rwsem_release( |
1518 | &root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1], | |
1519 | 1, _THIS_IP_); | |
6fc4e354 | 1520 | |
7892b5af | 1521 | schedule_work(&ac->work); |
bb9c12c9 SW |
1522 | |
1523 | /* wait for transaction to start and unblock */ | |
bb9c12c9 SW |
1524 | if (wait_for_unblock) |
1525 | wait_current_trans_commit_start_and_unblock(root, cur_trans); | |
1526 | else | |
1527 | wait_current_trans_commit_start(root, cur_trans); | |
bb9c12c9 | 1528 | |
38e88054 SW |
1529 | if (current->journal_info == trans) |
1530 | current->journal_info = NULL; | |
1531 | ||
724e2315 | 1532 | btrfs_put_transaction(cur_trans); |
bb9c12c9 SW |
1533 | return 0; |
1534 | } | |
1535 | ||
49b25e05 JM |
1536 | |
1537 | static void cleanup_transaction(struct btrfs_trans_handle *trans, | |
7b8b92af | 1538 | struct btrfs_root *root, int err) |
49b25e05 JM |
1539 | { |
1540 | struct btrfs_transaction *cur_trans = trans->transaction; | |
f094ac32 | 1541 | DEFINE_WAIT(wait); |
49b25e05 JM |
1542 | |
1543 | WARN_ON(trans->use_count > 1); | |
1544 | ||
7b8b92af JB |
1545 | btrfs_abort_transaction(trans, root, err); |
1546 | ||
49b25e05 | 1547 | spin_lock(&root->fs_info->trans_lock); |
66b6135b | 1548 | |
25d8c284 MX |
1549 | /* |
1550 | * If the transaction is removed from the list, it means this | |
1551 | * transaction has been committed successfully, so it is impossible | |
1552 | * to call the cleanup function. | |
1553 | */ | |
1554 | BUG_ON(list_empty(&cur_trans->list)); | |
66b6135b | 1555 | |
49b25e05 | 1556 | list_del_init(&cur_trans->list); |
d7096fc3 | 1557 | if (cur_trans == root->fs_info->running_transaction) { |
4a9d8bde | 1558 | cur_trans->state = TRANS_STATE_COMMIT_DOING; |
f094ac32 LB |
1559 | spin_unlock(&root->fs_info->trans_lock); |
1560 | wait_event(cur_trans->writer_wait, | |
1561 | atomic_read(&cur_trans->num_writers) == 1); | |
1562 | ||
1563 | spin_lock(&root->fs_info->trans_lock); | |
d7096fc3 | 1564 | } |
49b25e05 JM |
1565 | spin_unlock(&root->fs_info->trans_lock); |
1566 | ||
1567 | btrfs_cleanup_one_transaction(trans->transaction, root); | |
1568 | ||
4a9d8bde MX |
1569 | spin_lock(&root->fs_info->trans_lock); |
1570 | if (cur_trans == root->fs_info->running_transaction) | |
1571 | root->fs_info->running_transaction = NULL; | |
1572 | spin_unlock(&root->fs_info->trans_lock); | |
1573 | ||
e0228285 JB |
1574 | if (trans->type & __TRANS_FREEZABLE) |
1575 | sb_end_intwrite(root->fs_info->sb); | |
724e2315 JB |
1576 | btrfs_put_transaction(cur_trans); |
1577 | btrfs_put_transaction(cur_trans); | |
49b25e05 JM |
1578 | |
1579 | trace_btrfs_transaction_commit(root); | |
1580 | ||
1581 | btrfs_scrub_continue(root); | |
1582 | ||
1583 | if (current->journal_info == trans) | |
1584 | current->journal_info = NULL; | |
1585 | ||
1586 | kmem_cache_free(btrfs_trans_handle_cachep, trans); | |
1587 | } | |
1588 | ||
ca469637 MX |
1589 | static int btrfs_flush_all_pending_stuffs(struct btrfs_trans_handle *trans, |
1590 | struct btrfs_root *root) | |
1591 | { | |
ca469637 MX |
1592 | int ret; |
1593 | ||
ca469637 | 1594 | ret = btrfs_run_delayed_items(trans, root); |
ca469637 MX |
1595 | /* |
1596 | * running the delayed items may have added new refs. account | |
1597 | * them now so that they hinder processing of more delayed refs | |
1598 | * as little as possible. | |
1599 | */ | |
80d94fb3 FDBM |
1600 | if (ret) { |
1601 | btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info); | |
1602 | return ret; | |
1603 | } | |
1604 | ||
1605 | ret = btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info); | |
1606 | if (ret) | |
1607 | return ret; | |
ca469637 MX |
1608 | |
1609 | /* | |
1610 | * rename don't use btrfs_join_transaction, so, once we | |
1611 | * set the transaction to blocked above, we aren't going | |
1612 | * to get any new ordered operations. We can safely run | |
1613 | * it here and no for sure that nothing new will be added | |
1614 | * to the list | |
1615 | */ | |
569e0f35 | 1616 | ret = btrfs_run_ordered_operations(trans, root, 1); |
ca469637 | 1617 | |
eebc6084 | 1618 | return ret; |
ca469637 MX |
1619 | } |
1620 | ||
82436617 MX |
1621 | static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info) |
1622 | { | |
1623 | if (btrfs_test_opt(fs_info->tree_root, FLUSHONCOMMIT)) | |
91aef86f | 1624 | return btrfs_start_delalloc_roots(fs_info, 1); |
82436617 MX |
1625 | return 0; |
1626 | } | |
1627 | ||
1628 | static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info) | |
1629 | { | |
1630 | if (btrfs_test_opt(fs_info->tree_root, FLUSHONCOMMIT)) | |
b0244199 | 1631 | btrfs_wait_ordered_roots(fs_info, -1); |
82436617 MX |
1632 | } |
1633 | ||
79154b1b CM |
1634 | int btrfs_commit_transaction(struct btrfs_trans_handle *trans, |
1635 | struct btrfs_root *root) | |
1636 | { | |
49b25e05 | 1637 | struct btrfs_transaction *cur_trans = trans->transaction; |
8fd17795 | 1638 | struct btrfs_transaction *prev_trans = NULL; |
25287e0a | 1639 | int ret; |
79154b1b | 1640 | |
569e0f35 | 1641 | ret = btrfs_run_ordered_operations(trans, root, 0); |
25287e0a MX |
1642 | if (ret) { |
1643 | btrfs_abort_transaction(trans, root, ret); | |
e4a2bcac JB |
1644 | btrfs_end_transaction(trans, root); |
1645 | return ret; | |
25287e0a | 1646 | } |
5a3f23d5 | 1647 | |
8d25a086 MX |
1648 | /* Stop the commit early if ->aborted is set */ |
1649 | if (unlikely(ACCESS_ONCE(cur_trans->aborted))) { | |
25287e0a | 1650 | ret = cur_trans->aborted; |
e4a2bcac JB |
1651 | btrfs_end_transaction(trans, root); |
1652 | return ret; | |
25287e0a | 1653 | } |
49b25e05 | 1654 | |
56bec294 CM |
1655 | /* make a pass through all the delayed refs we have so far |
1656 | * any runnings procs may add more while we are here | |
1657 | */ | |
1658 | ret = btrfs_run_delayed_refs(trans, root, 0); | |
e4a2bcac JB |
1659 | if (ret) { |
1660 | btrfs_end_transaction(trans, root); | |
1661 | return ret; | |
1662 | } | |
56bec294 | 1663 | |
0e721106 JB |
1664 | btrfs_trans_release_metadata(trans, root); |
1665 | trans->block_rsv = NULL; | |
272d26d0 MX |
1666 | if (trans->qgroup_reserved) { |
1667 | btrfs_qgroup_free(root, trans->qgroup_reserved); | |
1668 | trans->qgroup_reserved = 0; | |
1669 | } | |
0e721106 | 1670 | |
b7ec40d7 | 1671 | cur_trans = trans->transaction; |
49b25e05 | 1672 | |
56bec294 CM |
1673 | /* |
1674 | * set the flushing flag so procs in this transaction have to | |
1675 | * start sending their work down. | |
1676 | */ | |
b7ec40d7 | 1677 | cur_trans->delayed_refs.flushing = 1; |
1be41b78 | 1678 | smp_wmb(); |
56bec294 | 1679 | |
ea658bad JB |
1680 | if (!list_empty(&trans->new_bgs)) |
1681 | btrfs_create_pending_block_groups(trans, root); | |
1682 | ||
c3e69d58 | 1683 | ret = btrfs_run_delayed_refs(trans, root, 0); |
e4a2bcac JB |
1684 | if (ret) { |
1685 | btrfs_end_transaction(trans, root); | |
1686 | return ret; | |
1687 | } | |
56bec294 | 1688 | |
4a9d8bde MX |
1689 | spin_lock(&root->fs_info->trans_lock); |
1690 | if (cur_trans->state >= TRANS_STATE_COMMIT_START) { | |
1691 | spin_unlock(&root->fs_info->trans_lock); | |
13c5a93e | 1692 | atomic_inc(&cur_trans->use_count); |
49b25e05 | 1693 | ret = btrfs_end_transaction(trans, root); |
ccd467d6 | 1694 | |
b9c8300c | 1695 | wait_for_commit(root, cur_trans); |
15ee9bc7 | 1696 | |
724e2315 | 1697 | btrfs_put_transaction(cur_trans); |
15ee9bc7 | 1698 | |
49b25e05 | 1699 | return ret; |
79154b1b | 1700 | } |
4313b399 | 1701 | |
4a9d8bde | 1702 | cur_trans->state = TRANS_STATE_COMMIT_START; |
bb9c12c9 SW |
1703 | wake_up(&root->fs_info->transaction_blocked_wait); |
1704 | ||
ccd467d6 CM |
1705 | if (cur_trans->list.prev != &root->fs_info->trans_list) { |
1706 | prev_trans = list_entry(cur_trans->list.prev, | |
1707 | struct btrfs_transaction, list); | |
4a9d8bde | 1708 | if (prev_trans->state != TRANS_STATE_COMPLETED) { |
13c5a93e | 1709 | atomic_inc(&prev_trans->use_count); |
a4abeea4 | 1710 | spin_unlock(&root->fs_info->trans_lock); |
ccd467d6 CM |
1711 | |
1712 | wait_for_commit(root, prev_trans); | |
ccd467d6 | 1713 | |
724e2315 | 1714 | btrfs_put_transaction(prev_trans); |
a4abeea4 JB |
1715 | } else { |
1716 | spin_unlock(&root->fs_info->trans_lock); | |
ccd467d6 | 1717 | } |
a4abeea4 JB |
1718 | } else { |
1719 | spin_unlock(&root->fs_info->trans_lock); | |
ccd467d6 | 1720 | } |
15ee9bc7 | 1721 | |
0860adfd MX |
1722 | extwriter_counter_dec(cur_trans, trans->type); |
1723 | ||
82436617 MX |
1724 | ret = btrfs_start_delalloc_flush(root->fs_info); |
1725 | if (ret) | |
1726 | goto cleanup_transaction; | |
1727 | ||
581227d0 MX |
1728 | ret = btrfs_flush_all_pending_stuffs(trans, root); |
1729 | if (ret) | |
1730 | goto cleanup_transaction; | |
15ee9bc7 | 1731 | |
581227d0 MX |
1732 | wait_event(cur_trans->writer_wait, |
1733 | extwriter_counter_read(cur_trans) == 0); | |
15ee9bc7 | 1734 | |
581227d0 | 1735 | /* some pending stuffs might be added after the previous flush. */ |
ca469637 MX |
1736 | ret = btrfs_flush_all_pending_stuffs(trans, root); |
1737 | if (ret) | |
1738 | goto cleanup_transaction; | |
1739 | ||
82436617 | 1740 | btrfs_wait_delalloc_flush(root->fs_info); |
cb7ab021 WS |
1741 | |
1742 | btrfs_scrub_pause(root); | |
ed0ca140 JB |
1743 | /* |
1744 | * Ok now we need to make sure to block out any other joins while we | |
1745 | * commit the transaction. We could have started a join before setting | |
4a9d8bde | 1746 | * COMMIT_DOING so make sure to wait for num_writers to == 1 again. |
ed0ca140 JB |
1747 | */ |
1748 | spin_lock(&root->fs_info->trans_lock); | |
4a9d8bde | 1749 | cur_trans->state = TRANS_STATE_COMMIT_DOING; |
ed0ca140 JB |
1750 | spin_unlock(&root->fs_info->trans_lock); |
1751 | wait_event(cur_trans->writer_wait, | |
1752 | atomic_read(&cur_trans->num_writers) == 1); | |
1753 | ||
2cba30f1 MX |
1754 | /* ->aborted might be set after the previous check, so check it */ |
1755 | if (unlikely(ACCESS_ONCE(cur_trans->aborted))) { | |
1756 | ret = cur_trans->aborted; | |
1757 | goto cleanup_transaction; | |
1758 | } | |
7585717f CM |
1759 | /* |
1760 | * the reloc mutex makes sure that we stop | |
1761 | * the balancing code from coming in and moving | |
1762 | * extents around in the middle of the commit | |
1763 | */ | |
1764 | mutex_lock(&root->fs_info->reloc_mutex); | |
1765 | ||
42874b3d MX |
1766 | /* |
1767 | * We needn't worry about the delayed items because we will | |
1768 | * deal with them in create_pending_snapshot(), which is the | |
1769 | * core function of the snapshot creation. | |
1770 | */ | |
1771 | ret = create_pending_snapshots(trans, root->fs_info); | |
49b25e05 JM |
1772 | if (ret) { |
1773 | mutex_unlock(&root->fs_info->reloc_mutex); | |
1774 | goto cleanup_transaction; | |
1775 | } | |
3063d29f | 1776 | |
42874b3d MX |
1777 | /* |
1778 | * We insert the dir indexes of the snapshots and update the inode | |
1779 | * of the snapshots' parents after the snapshot creation, so there | |
1780 | * are some delayed items which are not dealt with. Now deal with | |
1781 | * them. | |
1782 | * | |
1783 | * We needn't worry that this operation will corrupt the snapshots, | |
1784 | * because all the tree which are snapshoted will be forced to COW | |
1785 | * the nodes and leaves. | |
1786 | */ | |
1787 | ret = btrfs_run_delayed_items(trans, root); | |
49b25e05 JM |
1788 | if (ret) { |
1789 | mutex_unlock(&root->fs_info->reloc_mutex); | |
1790 | goto cleanup_transaction; | |
1791 | } | |
16cdcec7 | 1792 | |
56bec294 | 1793 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
49b25e05 JM |
1794 | if (ret) { |
1795 | mutex_unlock(&root->fs_info->reloc_mutex); | |
1796 | goto cleanup_transaction; | |
1797 | } | |
56bec294 | 1798 | |
e999376f CM |
1799 | /* |
1800 | * make sure none of the code above managed to slip in a | |
1801 | * delayed item | |
1802 | */ | |
1803 | btrfs_assert_delayed_root_empty(root); | |
1804 | ||
2c90e5d6 | 1805 | WARN_ON(cur_trans != trans->transaction); |
dc17ff8f | 1806 | |
e02119d5 CM |
1807 | /* btrfs_commit_tree_roots is responsible for getting the |
1808 | * various roots consistent with each other. Every pointer | |
1809 | * in the tree of tree roots has to point to the most up to date | |
1810 | * root for every subvolume and other tree. So, we have to keep | |
1811 | * the tree logging code from jumping in and changing any | |
1812 | * of the trees. | |
1813 | * | |
1814 | * At this point in the commit, there can't be any tree-log | |
1815 | * writers, but a little lower down we drop the trans mutex | |
1816 | * and let new people in. By holding the tree_log_mutex | |
1817 | * from now until after the super is written, we avoid races | |
1818 | * with the tree-log code. | |
1819 | */ | |
1820 | mutex_lock(&root->fs_info->tree_log_mutex); | |
1821 | ||
5d4f98a2 | 1822 | ret = commit_fs_roots(trans, root); |
49b25e05 JM |
1823 | if (ret) { |
1824 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
871383be | 1825 | mutex_unlock(&root->fs_info->reloc_mutex); |
49b25e05 JM |
1826 | goto cleanup_transaction; |
1827 | } | |
54aa1f4d | 1828 | |
3818aea2 QW |
1829 | /* |
1830 | * Since the transaction is done, we should set the inode map cache flag | |
1831 | * before any other comming transaction. | |
1832 | */ | |
1833 | if (btrfs_test_opt(root, CHANGE_INODE_CACHE)) | |
1834 | btrfs_set_opt(root->fs_info->mount_opt, INODE_MAP_CACHE); | |
1835 | else | |
1836 | btrfs_clear_opt(root->fs_info->mount_opt, INODE_MAP_CACHE); | |
1837 | ||
5d4f98a2 | 1838 | /* commit_fs_roots gets rid of all the tree log roots, it is now |
e02119d5 CM |
1839 | * safe to free the root of tree log roots |
1840 | */ | |
1841 | btrfs_free_log_root_tree(trans, root->fs_info); | |
1842 | ||
5d4f98a2 | 1843 | ret = commit_cowonly_roots(trans, root); |
49b25e05 JM |
1844 | if (ret) { |
1845 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
871383be | 1846 | mutex_unlock(&root->fs_info->reloc_mutex); |
49b25e05 JM |
1847 | goto cleanup_transaction; |
1848 | } | |
54aa1f4d | 1849 | |
2cba30f1 MX |
1850 | /* |
1851 | * The tasks which save the space cache and inode cache may also | |
1852 | * update ->aborted, check it. | |
1853 | */ | |
1854 | if (unlikely(ACCESS_ONCE(cur_trans->aborted))) { | |
1855 | ret = cur_trans->aborted; | |
1856 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
1857 | mutex_unlock(&root->fs_info->reloc_mutex); | |
1858 | goto cleanup_transaction; | |
1859 | } | |
1860 | ||
11833d66 YZ |
1861 | btrfs_prepare_extent_commit(trans, root); |
1862 | ||
78fae27e | 1863 | cur_trans = root->fs_info->running_transaction; |
5d4f98a2 YZ |
1864 | |
1865 | btrfs_set_root_node(&root->fs_info->tree_root->root_item, | |
1866 | root->fs_info->tree_root->node); | |
817d52f8 | 1867 | switch_commit_root(root->fs_info->tree_root); |
5d4f98a2 YZ |
1868 | |
1869 | btrfs_set_root_node(&root->fs_info->chunk_root->root_item, | |
1870 | root->fs_info->chunk_root->node); | |
817d52f8 | 1871 | switch_commit_root(root->fs_info->chunk_root); |
5d4f98a2 | 1872 | |
edf39272 | 1873 | assert_qgroups_uptodate(trans); |
5d4f98a2 | 1874 | update_super_roots(root); |
e02119d5 | 1875 | |
60e7cd3a JB |
1876 | btrfs_set_super_log_root(root->fs_info->super_copy, 0); |
1877 | btrfs_set_super_log_root_level(root->fs_info->super_copy, 0); | |
6c41761f DS |
1878 | memcpy(root->fs_info->super_for_commit, root->fs_info->super_copy, |
1879 | sizeof(*root->fs_info->super_copy)); | |
ccd467d6 | 1880 | |
a4abeea4 | 1881 | spin_lock(&root->fs_info->trans_lock); |
4a9d8bde | 1882 | cur_trans->state = TRANS_STATE_UNBLOCKED; |
a4abeea4 | 1883 | root->fs_info->running_transaction = NULL; |
a4abeea4 | 1884 | spin_unlock(&root->fs_info->trans_lock); |
7585717f | 1885 | mutex_unlock(&root->fs_info->reloc_mutex); |
b7ec40d7 | 1886 | |
f9295749 | 1887 | wake_up(&root->fs_info->transaction_wait); |
e6dcd2dc | 1888 | |
79154b1b | 1889 | ret = btrfs_write_and_wait_transaction(trans, root); |
49b25e05 JM |
1890 | if (ret) { |
1891 | btrfs_error(root->fs_info, ret, | |
08748810 | 1892 | "Error while writing out transaction"); |
49b25e05 JM |
1893 | mutex_unlock(&root->fs_info->tree_log_mutex); |
1894 | goto cleanup_transaction; | |
1895 | } | |
1896 | ||
1897 | ret = write_ctree_super(trans, root, 0); | |
1898 | if (ret) { | |
1899 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
1900 | goto cleanup_transaction; | |
1901 | } | |
4313b399 | 1902 | |
e02119d5 CM |
1903 | /* |
1904 | * the super is written, we can safely allow the tree-loggers | |
1905 | * to go about their business | |
1906 | */ | |
1907 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
1908 | ||
11833d66 | 1909 | btrfs_finish_extent_commit(trans, root); |
4313b399 | 1910 | |
15ee9bc7 | 1911 | root->fs_info->last_trans_committed = cur_trans->transid; |
4a9d8bde MX |
1912 | /* |
1913 | * We needn't acquire the lock here because there is no other task | |
1914 | * which can change it. | |
1915 | */ | |
1916 | cur_trans->state = TRANS_STATE_COMPLETED; | |
2c90e5d6 | 1917 | wake_up(&cur_trans->commit_wait); |
3de4586c | 1918 | |
a4abeea4 | 1919 | spin_lock(&root->fs_info->trans_lock); |
13c5a93e | 1920 | list_del_init(&cur_trans->list); |
a4abeea4 JB |
1921 | spin_unlock(&root->fs_info->trans_lock); |
1922 | ||
724e2315 JB |
1923 | btrfs_put_transaction(cur_trans); |
1924 | btrfs_put_transaction(cur_trans); | |
58176a96 | 1925 | |
0860adfd | 1926 | if (trans->type & __TRANS_FREEZABLE) |
354aa0fb | 1927 | sb_end_intwrite(root->fs_info->sb); |
b2b5ef5c | 1928 | |
1abe9b8a | 1929 | trace_btrfs_transaction_commit(root); |
1930 | ||
a2de733c AJ |
1931 | btrfs_scrub_continue(root); |
1932 | ||
9ed74f2d JB |
1933 | if (current->journal_info == trans) |
1934 | current->journal_info = NULL; | |
1935 | ||
2c90e5d6 | 1936 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
24bbcf04 YZ |
1937 | |
1938 | if (current != root->fs_info->transaction_kthread) | |
1939 | btrfs_run_delayed_iputs(root); | |
1940 | ||
79154b1b | 1941 | return ret; |
49b25e05 JM |
1942 | |
1943 | cleanup_transaction: | |
0e721106 JB |
1944 | btrfs_trans_release_metadata(trans, root); |
1945 | trans->block_rsv = NULL; | |
272d26d0 MX |
1946 | if (trans->qgroup_reserved) { |
1947 | btrfs_qgroup_free(root, trans->qgroup_reserved); | |
1948 | trans->qgroup_reserved = 0; | |
1949 | } | |
c2cf52eb | 1950 | btrfs_warn(root->fs_info, "Skipping commit of aborted transaction."); |
49b25e05 JM |
1951 | if (current->journal_info == trans) |
1952 | current->journal_info = NULL; | |
7b8b92af | 1953 | cleanup_transaction(trans, root, ret); |
49b25e05 JM |
1954 | |
1955 | return ret; | |
79154b1b CM |
1956 | } |
1957 | ||
d352ac68 | 1958 | /* |
9d1a2a3a DS |
1959 | * return < 0 if error |
1960 | * 0 if there are no more dead_roots at the time of call | |
1961 | * 1 there are more to be processed, call me again | |
1962 | * | |
1963 | * The return value indicates there are certainly more snapshots to delete, but | |
1964 | * if there comes a new one during processing, it may return 0. We don't mind, | |
1965 | * because btrfs_commit_super will poke cleaner thread and it will process it a | |
1966 | * few seconds later. | |
d352ac68 | 1967 | */ |
9d1a2a3a | 1968 | int btrfs_clean_one_deleted_snapshot(struct btrfs_root *root) |
e9d0b13b | 1969 | { |
9d1a2a3a | 1970 | int ret; |
5d4f98a2 YZ |
1971 | struct btrfs_fs_info *fs_info = root->fs_info; |
1972 | ||
a4abeea4 | 1973 | spin_lock(&fs_info->trans_lock); |
9d1a2a3a DS |
1974 | if (list_empty(&fs_info->dead_roots)) { |
1975 | spin_unlock(&fs_info->trans_lock); | |
1976 | return 0; | |
1977 | } | |
1978 | root = list_first_entry(&fs_info->dead_roots, | |
1979 | struct btrfs_root, root_list); | |
18f687d5 WS |
1980 | /* |
1981 | * Make sure root is not involved in send, | |
1982 | * if we fail with first root, we return | |
1983 | * directly rather than continue. | |
1984 | */ | |
1985 | spin_lock(&root->root_item_lock); | |
1986 | if (root->send_in_progress) { | |
1987 | spin_unlock(&fs_info->trans_lock); | |
1988 | spin_unlock(&root->root_item_lock); | |
1989 | return 0; | |
1990 | } | |
1991 | spin_unlock(&root->root_item_lock); | |
1992 | ||
cfad392b | 1993 | list_del_init(&root->root_list); |
a4abeea4 | 1994 | spin_unlock(&fs_info->trans_lock); |
e9d0b13b | 1995 | |
efe120a0 | 1996 | pr_debug("BTRFS: cleaner removing %llu\n", root->objectid); |
76dda93c | 1997 | |
9d1a2a3a | 1998 | btrfs_kill_all_delayed_nodes(root); |
16cdcec7 | 1999 | |
9d1a2a3a DS |
2000 | if (btrfs_header_backref_rev(root->node) < |
2001 | BTRFS_MIXED_BACKREF_REV) | |
2002 | ret = btrfs_drop_snapshot(root, NULL, 0, 0); | |
2003 | else | |
2004 | ret = btrfs_drop_snapshot(root, NULL, 1, 0); | |
2005 | /* | |
2006 | * If we encounter a transaction abort during snapshot cleaning, we | |
2007 | * don't want to crash here | |
2008 | */ | |
6596a928 | 2009 | return (ret < 0) ? 0 : 1; |
e9d0b13b | 2010 | } |