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> |
79154b1b CM |
25 | #include "ctree.h" |
26 | #include "disk-io.h" | |
27 | #include "transaction.h" | |
925baedd | 28 | #include "locking.h" |
e02119d5 | 29 | #include "tree-log.h" |
581bb050 | 30 | #include "inode-map.h" |
79154b1b | 31 | |
0f7d52f4 CM |
32 | #define BTRFS_ROOT_TRANS_TAG 0 |
33 | ||
80b6794d | 34 | static noinline void put_transaction(struct btrfs_transaction *transaction) |
79154b1b | 35 | { |
13c5a93e JB |
36 | WARN_ON(atomic_read(&transaction->use_count) == 0); |
37 | if (atomic_dec_and_test(&transaction->use_count)) { | |
2c90e5d6 CM |
38 | memset(transaction, 0, sizeof(*transaction)); |
39 | kmem_cache_free(btrfs_transaction_cachep, transaction); | |
78fae27e | 40 | } |
79154b1b CM |
41 | } |
42 | ||
817d52f8 JB |
43 | static noinline void switch_commit_root(struct btrfs_root *root) |
44 | { | |
817d52f8 JB |
45 | free_extent_buffer(root->commit_root); |
46 | root->commit_root = btrfs_root_node(root); | |
817d52f8 JB |
47 | } |
48 | ||
d352ac68 CM |
49 | /* |
50 | * either allocate a new transaction or hop into the existing one | |
51 | */ | |
80b6794d | 52 | static noinline int join_transaction(struct btrfs_root *root) |
79154b1b CM |
53 | { |
54 | struct btrfs_transaction *cur_trans; | |
55 | cur_trans = root->fs_info->running_transaction; | |
56 | if (!cur_trans) { | |
2c90e5d6 CM |
57 | cur_trans = kmem_cache_alloc(btrfs_transaction_cachep, |
58 | GFP_NOFS); | |
db5b493a TI |
59 | if (!cur_trans) |
60 | return -ENOMEM; | |
0f7d52f4 | 61 | root->fs_info->generation++; |
13c5a93e | 62 | atomic_set(&cur_trans->num_writers, 1); |
15ee9bc7 | 63 | cur_trans->num_joined = 0; |
0f7d52f4 | 64 | cur_trans->transid = root->fs_info->generation; |
79154b1b CM |
65 | init_waitqueue_head(&cur_trans->writer_wait); |
66 | init_waitqueue_head(&cur_trans->commit_wait); | |
67 | cur_trans->in_commit = 0; | |
f9295749 | 68 | cur_trans->blocked = 0; |
13c5a93e | 69 | atomic_set(&cur_trans->use_count, 1); |
79154b1b | 70 | cur_trans->commit_done = 0; |
08607c1b | 71 | cur_trans->start_time = get_seconds(); |
56bec294 | 72 | |
6bef4d31 | 73 | cur_trans->delayed_refs.root = RB_ROOT; |
56bec294 | 74 | cur_trans->delayed_refs.num_entries = 0; |
c3e69d58 CM |
75 | cur_trans->delayed_refs.num_heads_ready = 0; |
76 | cur_trans->delayed_refs.num_heads = 0; | |
56bec294 | 77 | cur_trans->delayed_refs.flushing = 0; |
c3e69d58 | 78 | cur_trans->delayed_refs.run_delayed_start = 0; |
56bec294 CM |
79 | spin_lock_init(&cur_trans->delayed_refs.lock); |
80 | ||
3063d29f | 81 | INIT_LIST_HEAD(&cur_trans->pending_snapshots); |
8fd17795 | 82 | list_add_tail(&cur_trans->list, &root->fs_info->trans_list); |
d1310b2e | 83 | extent_io_tree_init(&cur_trans->dirty_pages, |
f993c883 | 84 | root->fs_info->btree_inode->i_mapping); |
48ec2cf8 CM |
85 | spin_lock(&root->fs_info->new_trans_lock); |
86 | root->fs_info->running_transaction = cur_trans; | |
87 | spin_unlock(&root->fs_info->new_trans_lock); | |
15ee9bc7 | 88 | } else { |
13c5a93e | 89 | atomic_inc(&cur_trans->num_writers); |
15ee9bc7 | 90 | cur_trans->num_joined++; |
79154b1b | 91 | } |
15ee9bc7 | 92 | |
79154b1b CM |
93 | return 0; |
94 | } | |
95 | ||
d352ac68 | 96 | /* |
d397712b CM |
97 | * this does all the record keeping required to make sure that a reference |
98 | * counted root is properly recorded in a given transaction. This is required | |
99 | * to make sure the old root from before we joined the transaction is deleted | |
100 | * when the transaction commits | |
d352ac68 | 101 | */ |
5d4f98a2 YZ |
102 | static noinline int record_root_in_trans(struct btrfs_trans_handle *trans, |
103 | struct btrfs_root *root) | |
6702ed49 | 104 | { |
5d4f98a2 | 105 | if (root->ref_cows && root->last_trans < trans->transid) { |
6702ed49 | 106 | WARN_ON(root == root->fs_info->extent_root); |
5d4f98a2 YZ |
107 | WARN_ON(root->commit_root != root->node); |
108 | ||
109 | radix_tree_tag_set(&root->fs_info->fs_roots_radix, | |
110 | (unsigned long)root->root_key.objectid, | |
111 | BTRFS_ROOT_TRANS_TAG); | |
112 | root->last_trans = trans->transid; | |
113 | btrfs_init_reloc_root(trans, root); | |
114 | } | |
115 | return 0; | |
116 | } | |
bcc63abb | 117 | |
5d4f98a2 YZ |
118 | int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, |
119 | struct btrfs_root *root) | |
120 | { | |
121 | if (!root->ref_cows) | |
122 | return 0; | |
bcc63abb | 123 | |
5d4f98a2 YZ |
124 | mutex_lock(&root->fs_info->trans_mutex); |
125 | if (root->last_trans == trans->transid) { | |
126 | mutex_unlock(&root->fs_info->trans_mutex); | |
127 | return 0; | |
6702ed49 | 128 | } |
5d4f98a2 YZ |
129 | |
130 | record_root_in_trans(trans, root); | |
131 | mutex_unlock(&root->fs_info->trans_mutex); | |
6702ed49 CM |
132 | return 0; |
133 | } | |
134 | ||
d352ac68 CM |
135 | /* wait for commit against the current transaction to become unblocked |
136 | * when this is done, it is safe to start a new transaction, but the current | |
137 | * transaction might not be fully on disk. | |
138 | */ | |
37d1aeee | 139 | static void wait_current_trans(struct btrfs_root *root) |
79154b1b | 140 | { |
f9295749 | 141 | struct btrfs_transaction *cur_trans; |
79154b1b | 142 | |
f9295749 | 143 | cur_trans = root->fs_info->running_transaction; |
37d1aeee | 144 | if (cur_trans && cur_trans->blocked) { |
f9295749 | 145 | DEFINE_WAIT(wait); |
13c5a93e | 146 | atomic_inc(&cur_trans->use_count); |
d397712b | 147 | while (1) { |
f9295749 CM |
148 | prepare_to_wait(&root->fs_info->transaction_wait, &wait, |
149 | TASK_UNINTERRUPTIBLE); | |
471fa17d | 150 | if (!cur_trans->blocked) |
f9295749 | 151 | break; |
471fa17d ZL |
152 | mutex_unlock(&root->fs_info->trans_mutex); |
153 | schedule(); | |
154 | mutex_lock(&root->fs_info->trans_mutex); | |
f9295749 | 155 | } |
471fa17d | 156 | finish_wait(&root->fs_info->transaction_wait, &wait); |
f9295749 CM |
157 | put_transaction(cur_trans); |
158 | } | |
37d1aeee CM |
159 | } |
160 | ||
249ac1e5 JB |
161 | enum btrfs_trans_type { |
162 | TRANS_START, | |
163 | TRANS_JOIN, | |
164 | TRANS_USERSPACE, | |
0af3d00b | 165 | TRANS_JOIN_NOLOCK, |
249ac1e5 JB |
166 | }; |
167 | ||
a22285a6 YZ |
168 | static int may_wait_transaction(struct btrfs_root *root, int type) |
169 | { | |
170 | if (!root->fs_info->log_root_recovering && | |
171 | ((type == TRANS_START && !root->fs_info->open_ioctl_trans) || | |
172 | type == TRANS_USERSPACE)) | |
173 | return 1; | |
174 | return 0; | |
175 | } | |
176 | ||
e02119d5 | 177 | static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root, |
a22285a6 | 178 | u64 num_items, int type) |
37d1aeee | 179 | { |
a22285a6 YZ |
180 | struct btrfs_trans_handle *h; |
181 | struct btrfs_transaction *cur_trans; | |
06d5a589 | 182 | int retries = 0; |
37d1aeee | 183 | int ret; |
acce952b | 184 | |
185 | if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) | |
186 | return ERR_PTR(-EROFS); | |
a22285a6 YZ |
187 | again: |
188 | h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS); | |
189 | if (!h) | |
190 | return ERR_PTR(-ENOMEM); | |
37d1aeee | 191 | |
0af3d00b JB |
192 | if (type != TRANS_JOIN_NOLOCK) |
193 | mutex_lock(&root->fs_info->trans_mutex); | |
a22285a6 | 194 | if (may_wait_transaction(root, type)) |
37d1aeee | 195 | wait_current_trans(root); |
a22285a6 | 196 | |
79154b1b | 197 | ret = join_transaction(root); |
db5b493a | 198 | if (ret < 0) { |
6e8df2ae | 199 | kmem_cache_free(btrfs_trans_handle_cachep, h); |
db5b493a TI |
200 | if (type != TRANS_JOIN_NOLOCK) |
201 | mutex_unlock(&root->fs_info->trans_mutex); | |
202 | return ERR_PTR(ret); | |
203 | } | |
0f7d52f4 | 204 | |
a22285a6 | 205 | cur_trans = root->fs_info->running_transaction; |
13c5a93e | 206 | atomic_inc(&cur_trans->use_count); |
0af3d00b JB |
207 | if (type != TRANS_JOIN_NOLOCK) |
208 | mutex_unlock(&root->fs_info->trans_mutex); | |
a22285a6 YZ |
209 | |
210 | h->transid = cur_trans->transid; | |
211 | h->transaction = cur_trans; | |
79154b1b | 212 | h->blocks_used = 0; |
d2fb3437 | 213 | h->block_group = 0; |
a22285a6 | 214 | h->bytes_reserved = 0; |
56bec294 | 215 | h->delayed_ref_updates = 0; |
f0486c68 | 216 | h->block_rsv = NULL; |
b7ec40d7 | 217 | |
a22285a6 YZ |
218 | smp_mb(); |
219 | if (cur_trans->blocked && may_wait_transaction(root, type)) { | |
220 | btrfs_commit_transaction(h, root); | |
221 | goto again; | |
222 | } | |
223 | ||
224 | if (num_items > 0) { | |
8bb8ab2e | 225 | ret = btrfs_trans_reserve_metadata(h, root, num_items); |
06d5a589 JB |
226 | if (ret == -EAGAIN && !retries) { |
227 | retries++; | |
a22285a6 YZ |
228 | btrfs_commit_transaction(h, root); |
229 | goto again; | |
06d5a589 JB |
230 | } else if (ret == -EAGAIN) { |
231 | /* | |
232 | * We have already retried and got EAGAIN, so really we | |
233 | * don't have space, so set ret to -ENOSPC. | |
234 | */ | |
235 | ret = -ENOSPC; | |
a22285a6 | 236 | } |
06d5a589 | 237 | |
a22285a6 YZ |
238 | if (ret < 0) { |
239 | btrfs_end_transaction(h, root); | |
240 | return ERR_PTR(ret); | |
241 | } | |
242 | } | |
9ed74f2d | 243 | |
0af3d00b JB |
244 | if (type != TRANS_JOIN_NOLOCK) |
245 | mutex_lock(&root->fs_info->trans_mutex); | |
5d4f98a2 | 246 | record_root_in_trans(h, root); |
0af3d00b JB |
247 | if (type != TRANS_JOIN_NOLOCK) |
248 | mutex_unlock(&root->fs_info->trans_mutex); | |
a22285a6 YZ |
249 | |
250 | if (!current->journal_info && type != TRANS_USERSPACE) | |
251 | current->journal_info = h; | |
79154b1b CM |
252 | return h; |
253 | } | |
254 | ||
f9295749 | 255 | struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, |
a22285a6 | 256 | int num_items) |
f9295749 | 257 | { |
a22285a6 | 258 | return start_transaction(root, num_items, TRANS_START); |
f9295749 CM |
259 | } |
260 | struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root, | |
261 | int num_blocks) | |
262 | { | |
a22285a6 | 263 | return start_transaction(root, 0, TRANS_JOIN); |
f9295749 CM |
264 | } |
265 | ||
0af3d00b JB |
266 | struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root, |
267 | int num_blocks) | |
268 | { | |
269 | return start_transaction(root, 0, TRANS_JOIN_NOLOCK); | |
270 | } | |
271 | ||
9ca9ee09 SW |
272 | struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *r, |
273 | int num_blocks) | |
274 | { | |
a22285a6 | 275 | return start_transaction(r, 0, TRANS_USERSPACE); |
9ca9ee09 SW |
276 | } |
277 | ||
d352ac68 | 278 | /* wait for a transaction commit to be fully complete */ |
89ce8a63 CM |
279 | static noinline int wait_for_commit(struct btrfs_root *root, |
280 | struct btrfs_transaction *commit) | |
281 | { | |
282 | DEFINE_WAIT(wait); | |
283 | mutex_lock(&root->fs_info->trans_mutex); | |
d397712b | 284 | while (!commit->commit_done) { |
89ce8a63 CM |
285 | prepare_to_wait(&commit->commit_wait, &wait, |
286 | TASK_UNINTERRUPTIBLE); | |
287 | if (commit->commit_done) | |
288 | break; | |
289 | mutex_unlock(&root->fs_info->trans_mutex); | |
290 | schedule(); | |
291 | mutex_lock(&root->fs_info->trans_mutex); | |
292 | } | |
293 | mutex_unlock(&root->fs_info->trans_mutex); | |
294 | finish_wait(&commit->commit_wait, &wait); | |
295 | return 0; | |
296 | } | |
297 | ||
46204592 SW |
298 | int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid) |
299 | { | |
300 | struct btrfs_transaction *cur_trans = NULL, *t; | |
301 | int ret; | |
302 | ||
303 | mutex_lock(&root->fs_info->trans_mutex); | |
304 | ||
305 | ret = 0; | |
306 | if (transid) { | |
307 | if (transid <= root->fs_info->last_trans_committed) | |
308 | goto out_unlock; | |
309 | ||
310 | /* find specified transaction */ | |
311 | list_for_each_entry(t, &root->fs_info->trans_list, list) { | |
312 | if (t->transid == transid) { | |
313 | cur_trans = t; | |
314 | break; | |
315 | } | |
316 | if (t->transid > transid) | |
317 | break; | |
318 | } | |
319 | ret = -EINVAL; | |
320 | if (!cur_trans) | |
321 | goto out_unlock; /* bad transid */ | |
322 | } else { | |
323 | /* find newest transaction that is committing | committed */ | |
324 | list_for_each_entry_reverse(t, &root->fs_info->trans_list, | |
325 | list) { | |
326 | if (t->in_commit) { | |
327 | if (t->commit_done) | |
328 | goto out_unlock; | |
329 | cur_trans = t; | |
330 | break; | |
331 | } | |
332 | } | |
333 | if (!cur_trans) | |
334 | goto out_unlock; /* nothing committing|committed */ | |
335 | } | |
336 | ||
13c5a93e | 337 | atomic_inc(&cur_trans->use_count); |
46204592 SW |
338 | mutex_unlock(&root->fs_info->trans_mutex); |
339 | ||
340 | wait_for_commit(root, cur_trans); | |
341 | ||
342 | mutex_lock(&root->fs_info->trans_mutex); | |
343 | put_transaction(cur_trans); | |
344 | ret = 0; | |
345 | out_unlock: | |
346 | mutex_unlock(&root->fs_info->trans_mutex); | |
347 | return ret; | |
348 | } | |
349 | ||
37d1aeee CM |
350 | void btrfs_throttle(struct btrfs_root *root) |
351 | { | |
352 | mutex_lock(&root->fs_info->trans_mutex); | |
9ca9ee09 SW |
353 | if (!root->fs_info->open_ioctl_trans) |
354 | wait_current_trans(root); | |
37d1aeee | 355 | mutex_unlock(&root->fs_info->trans_mutex); |
37d1aeee CM |
356 | } |
357 | ||
8929ecfa YZ |
358 | static int should_end_transaction(struct btrfs_trans_handle *trans, |
359 | struct btrfs_root *root) | |
360 | { | |
361 | int ret; | |
362 | ret = btrfs_block_rsv_check(trans, root, | |
363 | &root->fs_info->global_block_rsv, 0, 5); | |
364 | return ret ? 1 : 0; | |
365 | } | |
366 | ||
367 | int btrfs_should_end_transaction(struct btrfs_trans_handle *trans, | |
368 | struct btrfs_root *root) | |
369 | { | |
370 | struct btrfs_transaction *cur_trans = trans->transaction; | |
371 | int updates; | |
372 | ||
373 | if (cur_trans->blocked || cur_trans->delayed_refs.flushing) | |
374 | return 1; | |
375 | ||
376 | updates = trans->delayed_ref_updates; | |
377 | trans->delayed_ref_updates = 0; | |
378 | if (updates) | |
379 | btrfs_run_delayed_refs(trans, root, updates); | |
380 | ||
381 | return should_end_transaction(trans, root); | |
382 | } | |
383 | ||
89ce8a63 | 384 | static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, |
0af3d00b | 385 | struct btrfs_root *root, int throttle, int lock) |
79154b1b | 386 | { |
8929ecfa | 387 | struct btrfs_transaction *cur_trans = trans->transaction; |
ab78c84d | 388 | struct btrfs_fs_info *info = root->fs_info; |
c3e69d58 CM |
389 | int count = 0; |
390 | ||
391 | while (count < 4) { | |
392 | unsigned long cur = trans->delayed_ref_updates; | |
393 | trans->delayed_ref_updates = 0; | |
394 | if (cur && | |
395 | trans->transaction->delayed_refs.num_heads_ready > 64) { | |
396 | trans->delayed_ref_updates = 0; | |
b7ec40d7 CM |
397 | |
398 | /* | |
399 | * do a full flush if the transaction is trying | |
400 | * to close | |
401 | */ | |
402 | if (trans->transaction->delayed_refs.flushing) | |
403 | cur = 0; | |
c3e69d58 CM |
404 | btrfs_run_delayed_refs(trans, root, cur); |
405 | } else { | |
406 | break; | |
407 | } | |
408 | count++; | |
56bec294 CM |
409 | } |
410 | ||
a22285a6 YZ |
411 | btrfs_trans_release_metadata(trans, root); |
412 | ||
0af3d00b | 413 | if (lock && !root->fs_info->open_ioctl_trans && |
8929ecfa YZ |
414 | should_end_transaction(trans, root)) |
415 | trans->transaction->blocked = 1; | |
416 | ||
0af3d00b | 417 | if (lock && cur_trans->blocked && !cur_trans->in_commit) { |
8929ecfa YZ |
418 | if (throttle) |
419 | return btrfs_commit_transaction(trans, root); | |
420 | else | |
421 | wake_up_process(info->transaction_kthread); | |
422 | } | |
423 | ||
8929ecfa | 424 | WARN_ON(cur_trans != info->running_transaction); |
13c5a93e JB |
425 | WARN_ON(atomic_read(&cur_trans->num_writers) < 1); |
426 | atomic_dec(&cur_trans->num_writers); | |
89ce8a63 | 427 | |
99d16cbc | 428 | smp_mb(); |
79154b1b CM |
429 | if (waitqueue_active(&cur_trans->writer_wait)) |
430 | wake_up(&cur_trans->writer_wait); | |
79154b1b | 431 | put_transaction(cur_trans); |
9ed74f2d JB |
432 | |
433 | if (current->journal_info == trans) | |
434 | current->journal_info = NULL; | |
d6025579 | 435 | memset(trans, 0, sizeof(*trans)); |
2c90e5d6 | 436 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
ab78c84d | 437 | |
24bbcf04 YZ |
438 | if (throttle) |
439 | btrfs_run_delayed_iputs(root); | |
440 | ||
79154b1b CM |
441 | return 0; |
442 | } | |
443 | ||
89ce8a63 CM |
444 | int btrfs_end_transaction(struct btrfs_trans_handle *trans, |
445 | struct btrfs_root *root) | |
446 | { | |
16cdcec7 MX |
447 | int ret; |
448 | ||
449 | ret = __btrfs_end_transaction(trans, root, 0, 1); | |
450 | if (ret) | |
451 | return ret; | |
452 | return 0; | |
89ce8a63 CM |
453 | } |
454 | ||
455 | int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans, | |
456 | struct btrfs_root *root) | |
457 | { | |
16cdcec7 MX |
458 | int ret; |
459 | ||
460 | ret = __btrfs_end_transaction(trans, root, 1, 1); | |
461 | if (ret) | |
462 | return ret; | |
463 | return 0; | |
0af3d00b JB |
464 | } |
465 | ||
466 | int btrfs_end_transaction_nolock(struct btrfs_trans_handle *trans, | |
467 | struct btrfs_root *root) | |
468 | { | |
16cdcec7 MX |
469 | int ret; |
470 | ||
471 | ret = __btrfs_end_transaction(trans, root, 0, 0); | |
472 | if (ret) | |
473 | return ret; | |
474 | return 0; | |
475 | } | |
476 | ||
477 | int btrfs_end_transaction_dmeta(struct btrfs_trans_handle *trans, | |
478 | struct btrfs_root *root) | |
479 | { | |
480 | return __btrfs_end_transaction(trans, root, 1, 1); | |
89ce8a63 CM |
481 | } |
482 | ||
d352ac68 CM |
483 | /* |
484 | * when btree blocks are allocated, they have some corresponding bits set for | |
485 | * them in one of two extent_io trees. This is used to make sure all of | |
690587d1 | 486 | * those extents are sent to disk but does not wait on them |
d352ac68 | 487 | */ |
690587d1 | 488 | int btrfs_write_marked_extents(struct btrfs_root *root, |
8cef4e16 | 489 | struct extent_io_tree *dirty_pages, int mark) |
79154b1b | 490 | { |
7c4452b9 | 491 | int ret; |
777e6bd7 | 492 | int err = 0; |
7c4452b9 CM |
493 | int werr = 0; |
494 | struct page *page; | |
7c4452b9 | 495 | struct inode *btree_inode = root->fs_info->btree_inode; |
777e6bd7 | 496 | u64 start = 0; |
5f39d397 CM |
497 | u64 end; |
498 | unsigned long index; | |
7c4452b9 | 499 | |
d397712b | 500 | while (1) { |
777e6bd7 | 501 | ret = find_first_extent_bit(dirty_pages, start, &start, &end, |
8cef4e16 | 502 | mark); |
5f39d397 | 503 | if (ret) |
7c4452b9 | 504 | break; |
d397712b | 505 | while (start <= end) { |
777e6bd7 CM |
506 | cond_resched(); |
507 | ||
5f39d397 | 508 | index = start >> PAGE_CACHE_SHIFT; |
35ebb934 | 509 | start = (u64)(index + 1) << PAGE_CACHE_SHIFT; |
4bef0848 | 510 | page = find_get_page(btree_inode->i_mapping, index); |
7c4452b9 CM |
511 | if (!page) |
512 | continue; | |
4bef0848 CM |
513 | |
514 | btree_lock_page_hook(page); | |
515 | if (!page->mapping) { | |
516 | unlock_page(page); | |
517 | page_cache_release(page); | |
518 | continue; | |
519 | } | |
520 | ||
6702ed49 CM |
521 | if (PageWriteback(page)) { |
522 | if (PageDirty(page)) | |
523 | wait_on_page_writeback(page); | |
524 | else { | |
525 | unlock_page(page); | |
526 | page_cache_release(page); | |
527 | continue; | |
528 | } | |
529 | } | |
7c4452b9 CM |
530 | err = write_one_page(page, 0); |
531 | if (err) | |
532 | werr = err; | |
533 | page_cache_release(page); | |
534 | } | |
535 | } | |
690587d1 CM |
536 | if (err) |
537 | werr = err; | |
538 | return werr; | |
539 | } | |
540 | ||
541 | /* | |
542 | * when btree blocks are allocated, they have some corresponding bits set for | |
543 | * them in one of two extent_io trees. This is used to make sure all of | |
544 | * those extents are on disk for transaction or log commit. We wait | |
545 | * on all the pages and clear them from the dirty pages state tree | |
546 | */ | |
547 | int btrfs_wait_marked_extents(struct btrfs_root *root, | |
8cef4e16 | 548 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 CM |
549 | { |
550 | int ret; | |
551 | int err = 0; | |
552 | int werr = 0; | |
553 | struct page *page; | |
554 | struct inode *btree_inode = root->fs_info->btree_inode; | |
555 | u64 start = 0; | |
556 | u64 end; | |
557 | unsigned long index; | |
558 | ||
d397712b | 559 | while (1) { |
8cef4e16 YZ |
560 | ret = find_first_extent_bit(dirty_pages, start, &start, &end, |
561 | mark); | |
777e6bd7 CM |
562 | if (ret) |
563 | break; | |
564 | ||
8cef4e16 | 565 | clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS); |
d397712b | 566 | while (start <= end) { |
777e6bd7 CM |
567 | index = start >> PAGE_CACHE_SHIFT; |
568 | start = (u64)(index + 1) << PAGE_CACHE_SHIFT; | |
569 | page = find_get_page(btree_inode->i_mapping, index); | |
570 | if (!page) | |
571 | continue; | |
572 | if (PageDirty(page)) { | |
4bef0848 CM |
573 | btree_lock_page_hook(page); |
574 | wait_on_page_writeback(page); | |
777e6bd7 CM |
575 | err = write_one_page(page, 0); |
576 | if (err) | |
577 | werr = err; | |
578 | } | |
105d931d | 579 | wait_on_page_writeback(page); |
777e6bd7 CM |
580 | page_cache_release(page); |
581 | cond_resched(); | |
582 | } | |
583 | } | |
7c4452b9 CM |
584 | if (err) |
585 | werr = err; | |
586 | return werr; | |
79154b1b CM |
587 | } |
588 | ||
690587d1 CM |
589 | /* |
590 | * when btree blocks are allocated, they have some corresponding bits set for | |
591 | * them in one of two extent_io trees. This is used to make sure all of | |
592 | * those extents are on disk for transaction or log commit | |
593 | */ | |
594 | int btrfs_write_and_wait_marked_extents(struct btrfs_root *root, | |
8cef4e16 | 595 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 CM |
596 | { |
597 | int ret; | |
598 | int ret2; | |
599 | ||
8cef4e16 YZ |
600 | ret = btrfs_write_marked_extents(root, dirty_pages, mark); |
601 | ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark); | |
690587d1 CM |
602 | return ret || ret2; |
603 | } | |
604 | ||
d0c803c4 CM |
605 | int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans, |
606 | struct btrfs_root *root) | |
607 | { | |
608 | if (!trans || !trans->transaction) { | |
609 | struct inode *btree_inode; | |
610 | btree_inode = root->fs_info->btree_inode; | |
611 | return filemap_write_and_wait(btree_inode->i_mapping); | |
612 | } | |
613 | return btrfs_write_and_wait_marked_extents(root, | |
8cef4e16 YZ |
614 | &trans->transaction->dirty_pages, |
615 | EXTENT_DIRTY); | |
d0c803c4 CM |
616 | } |
617 | ||
d352ac68 CM |
618 | /* |
619 | * this is used to update the root pointer in the tree of tree roots. | |
620 | * | |
621 | * But, in the case of the extent allocation tree, updating the root | |
622 | * pointer may allocate blocks which may change the root of the extent | |
623 | * allocation tree. | |
624 | * | |
625 | * So, this loops and repeats and makes sure the cowonly root didn't | |
626 | * change while the root pointer was being updated in the metadata. | |
627 | */ | |
0b86a832 CM |
628 | static int update_cowonly_root(struct btrfs_trans_handle *trans, |
629 | struct btrfs_root *root) | |
79154b1b CM |
630 | { |
631 | int ret; | |
0b86a832 | 632 | u64 old_root_bytenr; |
86b9f2ec | 633 | u64 old_root_used; |
0b86a832 | 634 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
79154b1b | 635 | |
86b9f2ec | 636 | old_root_used = btrfs_root_used(&root->root_item); |
0b86a832 | 637 | btrfs_write_dirty_block_groups(trans, root); |
56bec294 | 638 | |
d397712b | 639 | while (1) { |
0b86a832 | 640 | old_root_bytenr = btrfs_root_bytenr(&root->root_item); |
86b9f2ec YZ |
641 | if (old_root_bytenr == root->node->start && |
642 | old_root_used == btrfs_root_used(&root->root_item)) | |
79154b1b | 643 | break; |
87ef2bb4 | 644 | |
5d4f98a2 | 645 | btrfs_set_root_node(&root->root_item, root->node); |
79154b1b | 646 | ret = btrfs_update_root(trans, tree_root, |
0b86a832 CM |
647 | &root->root_key, |
648 | &root->root_item); | |
79154b1b | 649 | BUG_ON(ret); |
56bec294 | 650 | |
86b9f2ec | 651 | old_root_used = btrfs_root_used(&root->root_item); |
4a8c9a62 | 652 | ret = btrfs_write_dirty_block_groups(trans, root); |
56bec294 | 653 | BUG_ON(ret); |
0b86a832 | 654 | } |
276e680d YZ |
655 | |
656 | if (root != root->fs_info->extent_root) | |
657 | switch_commit_root(root); | |
658 | ||
0b86a832 CM |
659 | return 0; |
660 | } | |
661 | ||
d352ac68 CM |
662 | /* |
663 | * update all the cowonly tree roots on disk | |
664 | */ | |
5d4f98a2 YZ |
665 | static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans, |
666 | struct btrfs_root *root) | |
0b86a832 CM |
667 | { |
668 | struct btrfs_fs_info *fs_info = root->fs_info; | |
669 | struct list_head *next; | |
84234f3a | 670 | struct extent_buffer *eb; |
56bec294 | 671 | int ret; |
84234f3a | 672 | |
56bec294 CM |
673 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
674 | BUG_ON(ret); | |
87ef2bb4 | 675 | |
84234f3a | 676 | eb = btrfs_lock_root_node(fs_info->tree_root); |
9fa8cfe7 | 677 | btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 0, &eb); |
84234f3a YZ |
678 | btrfs_tree_unlock(eb); |
679 | free_extent_buffer(eb); | |
0b86a832 | 680 | |
56bec294 CM |
681 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
682 | BUG_ON(ret); | |
87ef2bb4 | 683 | |
d397712b | 684 | while (!list_empty(&fs_info->dirty_cowonly_roots)) { |
0b86a832 CM |
685 | next = fs_info->dirty_cowonly_roots.next; |
686 | list_del_init(next); | |
687 | root = list_entry(next, struct btrfs_root, dirty_list); | |
87ef2bb4 | 688 | |
0b86a832 | 689 | update_cowonly_root(trans, root); |
79154b1b | 690 | } |
276e680d YZ |
691 | |
692 | down_write(&fs_info->extent_commit_sem); | |
693 | switch_commit_root(fs_info->extent_root); | |
694 | up_write(&fs_info->extent_commit_sem); | |
695 | ||
79154b1b CM |
696 | return 0; |
697 | } | |
698 | ||
d352ac68 CM |
699 | /* |
700 | * dead roots are old snapshots that need to be deleted. This allocates | |
701 | * a dirty root struct and adds it into the list of dead roots that need to | |
702 | * be deleted | |
703 | */ | |
5d4f98a2 | 704 | int btrfs_add_dead_root(struct btrfs_root *root) |
5eda7b5e | 705 | { |
b48652c1 | 706 | mutex_lock(&root->fs_info->trans_mutex); |
5d4f98a2 | 707 | list_add(&root->root_list, &root->fs_info->dead_roots); |
b48652c1 | 708 | mutex_unlock(&root->fs_info->trans_mutex); |
5eda7b5e CM |
709 | return 0; |
710 | } | |
711 | ||
d352ac68 | 712 | /* |
5d4f98a2 | 713 | * update all the cowonly tree roots on disk |
d352ac68 | 714 | */ |
5d4f98a2 YZ |
715 | static noinline int commit_fs_roots(struct btrfs_trans_handle *trans, |
716 | struct btrfs_root *root) | |
0f7d52f4 | 717 | { |
0f7d52f4 | 718 | struct btrfs_root *gang[8]; |
5d4f98a2 | 719 | struct btrfs_fs_info *fs_info = root->fs_info; |
0f7d52f4 CM |
720 | int i; |
721 | int ret; | |
54aa1f4d CM |
722 | int err = 0; |
723 | ||
d397712b | 724 | while (1) { |
5d4f98a2 YZ |
725 | ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, |
726 | (void **)gang, 0, | |
0f7d52f4 CM |
727 | ARRAY_SIZE(gang), |
728 | BTRFS_ROOT_TRANS_TAG); | |
729 | if (ret == 0) | |
730 | break; | |
731 | for (i = 0; i < ret; i++) { | |
732 | root = gang[i]; | |
5d4f98a2 YZ |
733 | radix_tree_tag_clear(&fs_info->fs_roots_radix, |
734 | (unsigned long)root->root_key.objectid, | |
735 | BTRFS_ROOT_TRANS_TAG); | |
31153d81 | 736 | |
e02119d5 | 737 | btrfs_free_log(trans, root); |
5d4f98a2 | 738 | btrfs_update_reloc_root(trans, root); |
d68fc57b | 739 | btrfs_orphan_commit_root(trans, root); |
bcc63abb | 740 | |
82d5902d LZ |
741 | btrfs_save_ino_cache(root, trans); |
742 | ||
978d910d | 743 | if (root->commit_root != root->node) { |
581bb050 | 744 | mutex_lock(&root->fs_commit_mutex); |
817d52f8 | 745 | switch_commit_root(root); |
581bb050 LZ |
746 | btrfs_unpin_free_ino(root); |
747 | mutex_unlock(&root->fs_commit_mutex); | |
748 | ||
978d910d YZ |
749 | btrfs_set_root_node(&root->root_item, |
750 | root->node); | |
751 | } | |
5d4f98a2 | 752 | |
5d4f98a2 | 753 | err = btrfs_update_root(trans, fs_info->tree_root, |
0f7d52f4 CM |
754 | &root->root_key, |
755 | &root->root_item); | |
54aa1f4d CM |
756 | if (err) |
757 | break; | |
0f7d52f4 CM |
758 | } |
759 | } | |
54aa1f4d | 760 | return err; |
0f7d52f4 CM |
761 | } |
762 | ||
d352ac68 CM |
763 | /* |
764 | * defrag a given btree. If cacheonly == 1, this won't read from the disk, | |
765 | * otherwise every leaf in the btree is read and defragged. | |
766 | */ | |
e9d0b13b CM |
767 | int btrfs_defrag_root(struct btrfs_root *root, int cacheonly) |
768 | { | |
769 | struct btrfs_fs_info *info = root->fs_info; | |
e9d0b13b | 770 | struct btrfs_trans_handle *trans; |
8929ecfa | 771 | int ret; |
d3c2fdcf | 772 | unsigned long nr; |
e9d0b13b | 773 | |
8929ecfa | 774 | if (xchg(&root->defrag_running, 1)) |
e9d0b13b | 775 | return 0; |
8929ecfa | 776 | |
6b80053d | 777 | while (1) { |
8929ecfa YZ |
778 | trans = btrfs_start_transaction(root, 0); |
779 | if (IS_ERR(trans)) | |
780 | return PTR_ERR(trans); | |
781 | ||
e9d0b13b | 782 | ret = btrfs_defrag_leaves(trans, root, cacheonly); |
8929ecfa | 783 | |
d3c2fdcf | 784 | nr = trans->blocks_used; |
e9d0b13b | 785 | btrfs_end_transaction(trans, root); |
d3c2fdcf | 786 | btrfs_btree_balance_dirty(info->tree_root, nr); |
e9d0b13b CM |
787 | cond_resched(); |
788 | ||
3f157a2f | 789 | if (root->fs_info->closing || ret != -EAGAIN) |
e9d0b13b CM |
790 | break; |
791 | } | |
792 | root->defrag_running = 0; | |
8929ecfa | 793 | return ret; |
e9d0b13b CM |
794 | } |
795 | ||
d352ac68 CM |
796 | /* |
797 | * new snapshots need to be created at a very specific time in the | |
798 | * transaction commit. This does the actual creation | |
799 | */ | |
80b6794d | 800 | static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, |
3063d29f CM |
801 | struct btrfs_fs_info *fs_info, |
802 | struct btrfs_pending_snapshot *pending) | |
803 | { | |
804 | struct btrfs_key key; | |
80b6794d | 805 | struct btrfs_root_item *new_root_item; |
3063d29f CM |
806 | struct btrfs_root *tree_root = fs_info->tree_root; |
807 | struct btrfs_root *root = pending->root; | |
6bdb72de SW |
808 | struct btrfs_root *parent_root; |
809 | struct inode *parent_inode; | |
6a912213 | 810 | struct dentry *parent; |
a22285a6 | 811 | struct dentry *dentry; |
3063d29f | 812 | struct extent_buffer *tmp; |
925baedd | 813 | struct extent_buffer *old; |
3063d29f | 814 | int ret; |
d68fc57b | 815 | u64 to_reserve = 0; |
6bdb72de | 816 | u64 index = 0; |
a22285a6 | 817 | u64 objectid; |
b83cc969 | 818 | u64 root_flags; |
3063d29f | 819 | |
80b6794d CM |
820 | new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS); |
821 | if (!new_root_item) { | |
a22285a6 | 822 | pending->error = -ENOMEM; |
80b6794d CM |
823 | goto fail; |
824 | } | |
a22285a6 | 825 | |
581bb050 | 826 | ret = btrfs_find_free_objectid(tree_root, &objectid); |
a22285a6 YZ |
827 | if (ret) { |
828 | pending->error = ret; | |
3063d29f | 829 | goto fail; |
a22285a6 | 830 | } |
3063d29f | 831 | |
3fd0a558 | 832 | btrfs_reloc_pre_snapshot(trans, pending, &to_reserve); |
d68fc57b YZ |
833 | btrfs_orphan_pre_snapshot(trans, pending, &to_reserve); |
834 | ||
835 | if (to_reserve > 0) { | |
836 | ret = btrfs_block_rsv_add(trans, root, &pending->block_rsv, | |
8bb8ab2e | 837 | to_reserve); |
d68fc57b YZ |
838 | if (ret) { |
839 | pending->error = ret; | |
840 | goto fail; | |
841 | } | |
842 | } | |
843 | ||
3063d29f | 844 | key.objectid = objectid; |
a22285a6 YZ |
845 | key.offset = (u64)-1; |
846 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3063d29f | 847 | |
a22285a6 | 848 | trans->block_rsv = &pending->block_rsv; |
3de4586c | 849 | |
a22285a6 | 850 | dentry = pending->dentry; |
6a912213 JB |
851 | parent = dget_parent(dentry); |
852 | parent_inode = parent->d_inode; | |
a22285a6 | 853 | parent_root = BTRFS_I(parent_inode)->root; |
6bdb72de | 854 | record_root_in_trans(trans, parent_root); |
a22285a6 | 855 | |
3063d29f CM |
856 | /* |
857 | * insert the directory item | |
858 | */ | |
3de4586c | 859 | ret = btrfs_set_inode_index(parent_inode, &index); |
6bdb72de | 860 | BUG_ON(ret); |
0660b5af | 861 | ret = btrfs_insert_dir_item(trans, parent_root, |
a22285a6 | 862 | dentry->d_name.name, dentry->d_name.len, |
16cdcec7 | 863 | parent_inode, &key, |
a22285a6 | 864 | BTRFS_FT_DIR, index); |
6bdb72de | 865 | BUG_ON(ret); |
0660b5af | 866 | |
a22285a6 YZ |
867 | btrfs_i_size_write(parent_inode, parent_inode->i_size + |
868 | dentry->d_name.len * 2); | |
52c26179 YZ |
869 | ret = btrfs_update_inode(trans, parent_root, parent_inode); |
870 | BUG_ON(ret); | |
871 | ||
6bdb72de SW |
872 | record_root_in_trans(trans, root); |
873 | btrfs_set_root_last_snapshot(&root->root_item, trans->transid); | |
874 | memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); | |
08fe4db1 | 875 | btrfs_check_and_init_root_item(new_root_item); |
6bdb72de | 876 | |
b83cc969 LZ |
877 | root_flags = btrfs_root_flags(new_root_item); |
878 | if (pending->readonly) | |
879 | root_flags |= BTRFS_ROOT_SUBVOL_RDONLY; | |
880 | else | |
881 | root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY; | |
882 | btrfs_set_root_flags(new_root_item, root_flags); | |
883 | ||
6bdb72de SW |
884 | old = btrfs_lock_root_node(root); |
885 | btrfs_cow_block(trans, root, old, NULL, 0, &old); | |
886 | btrfs_set_lock_blocking(old); | |
887 | ||
888 | btrfs_copy_root(trans, root, old, &tmp, objectid); | |
889 | btrfs_tree_unlock(old); | |
890 | free_extent_buffer(old); | |
891 | ||
892 | btrfs_set_root_node(new_root_item, tmp); | |
a22285a6 YZ |
893 | /* record when the snapshot was created in key.offset */ |
894 | key.offset = trans->transid; | |
895 | ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); | |
6bdb72de SW |
896 | btrfs_tree_unlock(tmp); |
897 | free_extent_buffer(tmp); | |
a22285a6 | 898 | BUG_ON(ret); |
6bdb72de | 899 | |
a22285a6 YZ |
900 | /* |
901 | * insert root back/forward references | |
902 | */ | |
903 | ret = btrfs_add_root_ref(trans, tree_root, objectid, | |
0660b5af | 904 | parent_root->root_key.objectid, |
33345d01 | 905 | btrfs_ino(parent_inode), index, |
a22285a6 | 906 | dentry->d_name.name, dentry->d_name.len); |
0660b5af | 907 | BUG_ON(ret); |
6a912213 | 908 | dput(parent); |
0660b5af | 909 | |
a22285a6 YZ |
910 | key.offset = (u64)-1; |
911 | pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key); | |
912 | BUG_ON(IS_ERR(pending->snap)); | |
d68fc57b | 913 | |
3fd0a558 | 914 | btrfs_reloc_post_snapshot(trans, pending); |
d68fc57b | 915 | btrfs_orphan_post_snapshot(trans, pending); |
3063d29f | 916 | fail: |
6bdb72de | 917 | kfree(new_root_item); |
a22285a6 YZ |
918 | btrfs_block_rsv_release(root, &pending->block_rsv, (u64)-1); |
919 | return 0; | |
3063d29f CM |
920 | } |
921 | ||
d352ac68 CM |
922 | /* |
923 | * create all the snapshots we've scheduled for creation | |
924 | */ | |
80b6794d CM |
925 | static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans, |
926 | struct btrfs_fs_info *fs_info) | |
3de4586c CM |
927 | { |
928 | struct btrfs_pending_snapshot *pending; | |
929 | struct list_head *head = &trans->transaction->pending_snapshots; | |
3de4586c CM |
930 | int ret; |
931 | ||
c6e30871 | 932 | list_for_each_entry(pending, head, list) { |
16cdcec7 MX |
933 | /* |
934 | * We must deal with the delayed items before creating | |
935 | * snapshots, or we will create a snapthot with inconsistent | |
936 | * information. | |
937 | */ | |
938 | ret = btrfs_run_delayed_items(trans, fs_info->fs_root); | |
939 | BUG_ON(ret); | |
940 | ||
3de4586c CM |
941 | ret = create_pending_snapshot(trans, fs_info, pending); |
942 | BUG_ON(ret); | |
943 | } | |
944 | return 0; | |
945 | } | |
946 | ||
5d4f98a2 YZ |
947 | static void update_super_roots(struct btrfs_root *root) |
948 | { | |
949 | struct btrfs_root_item *root_item; | |
950 | struct btrfs_super_block *super; | |
951 | ||
952 | super = &root->fs_info->super_copy; | |
953 | ||
954 | root_item = &root->fs_info->chunk_root->root_item; | |
955 | super->chunk_root = root_item->bytenr; | |
956 | super->chunk_root_generation = root_item->generation; | |
957 | super->chunk_root_level = root_item->level; | |
958 | ||
959 | root_item = &root->fs_info->tree_root->root_item; | |
960 | super->root = root_item->bytenr; | |
961 | super->generation = root_item->generation; | |
962 | super->root_level = root_item->level; | |
0af3d00b JB |
963 | if (super->cache_generation != 0 || btrfs_test_opt(root, SPACE_CACHE)) |
964 | super->cache_generation = root_item->generation; | |
5d4f98a2 YZ |
965 | } |
966 | ||
f36f3042 CM |
967 | int btrfs_transaction_in_commit(struct btrfs_fs_info *info) |
968 | { | |
969 | int ret = 0; | |
970 | spin_lock(&info->new_trans_lock); | |
971 | if (info->running_transaction) | |
972 | ret = info->running_transaction->in_commit; | |
973 | spin_unlock(&info->new_trans_lock); | |
974 | return ret; | |
975 | } | |
976 | ||
8929ecfa YZ |
977 | int btrfs_transaction_blocked(struct btrfs_fs_info *info) |
978 | { | |
979 | int ret = 0; | |
980 | spin_lock(&info->new_trans_lock); | |
981 | if (info->running_transaction) | |
982 | ret = info->running_transaction->blocked; | |
983 | spin_unlock(&info->new_trans_lock); | |
984 | return ret; | |
985 | } | |
986 | ||
bb9c12c9 SW |
987 | /* |
988 | * wait for the current transaction commit to start and block subsequent | |
989 | * transaction joins | |
990 | */ | |
991 | static void wait_current_trans_commit_start(struct btrfs_root *root, | |
992 | struct btrfs_transaction *trans) | |
993 | { | |
994 | DEFINE_WAIT(wait); | |
995 | ||
996 | if (trans->in_commit) | |
997 | return; | |
998 | ||
999 | while (1) { | |
1000 | prepare_to_wait(&root->fs_info->transaction_blocked_wait, &wait, | |
1001 | TASK_UNINTERRUPTIBLE); | |
1002 | if (trans->in_commit) { | |
1003 | finish_wait(&root->fs_info->transaction_blocked_wait, | |
1004 | &wait); | |
1005 | break; | |
1006 | } | |
1007 | mutex_unlock(&root->fs_info->trans_mutex); | |
1008 | schedule(); | |
1009 | mutex_lock(&root->fs_info->trans_mutex); | |
1010 | finish_wait(&root->fs_info->transaction_blocked_wait, &wait); | |
1011 | } | |
1012 | } | |
1013 | ||
1014 | /* | |
1015 | * wait for the current transaction to start and then become unblocked. | |
1016 | * caller holds ref. | |
1017 | */ | |
1018 | static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root, | |
1019 | struct btrfs_transaction *trans) | |
1020 | { | |
1021 | DEFINE_WAIT(wait); | |
1022 | ||
1023 | if (trans->commit_done || (trans->in_commit && !trans->blocked)) | |
1024 | return; | |
1025 | ||
1026 | while (1) { | |
1027 | prepare_to_wait(&root->fs_info->transaction_wait, &wait, | |
1028 | TASK_UNINTERRUPTIBLE); | |
1029 | if (trans->commit_done || | |
1030 | (trans->in_commit && !trans->blocked)) { | |
1031 | finish_wait(&root->fs_info->transaction_wait, | |
1032 | &wait); | |
1033 | break; | |
1034 | } | |
1035 | mutex_unlock(&root->fs_info->trans_mutex); | |
1036 | schedule(); | |
1037 | mutex_lock(&root->fs_info->trans_mutex); | |
1038 | finish_wait(&root->fs_info->transaction_wait, | |
1039 | &wait); | |
1040 | } | |
1041 | } | |
1042 | ||
1043 | /* | |
1044 | * commit transactions asynchronously. once btrfs_commit_transaction_async | |
1045 | * returns, any subsequent transaction will not be allowed to join. | |
1046 | */ | |
1047 | struct btrfs_async_commit { | |
1048 | struct btrfs_trans_handle *newtrans; | |
1049 | struct btrfs_root *root; | |
1050 | struct delayed_work work; | |
1051 | }; | |
1052 | ||
1053 | static void do_async_commit(struct work_struct *work) | |
1054 | { | |
1055 | struct btrfs_async_commit *ac = | |
1056 | container_of(work, struct btrfs_async_commit, work.work); | |
1057 | ||
1058 | btrfs_commit_transaction(ac->newtrans, ac->root); | |
1059 | kfree(ac); | |
1060 | } | |
1061 | ||
1062 | int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans, | |
1063 | struct btrfs_root *root, | |
1064 | int wait_for_unblock) | |
1065 | { | |
1066 | struct btrfs_async_commit *ac; | |
1067 | struct btrfs_transaction *cur_trans; | |
1068 | ||
1069 | ac = kmalloc(sizeof(*ac), GFP_NOFS); | |
db5b493a TI |
1070 | if (!ac) |
1071 | return -ENOMEM; | |
bb9c12c9 SW |
1072 | |
1073 | INIT_DELAYED_WORK(&ac->work, do_async_commit); | |
1074 | ac->root = root; | |
1075 | ac->newtrans = btrfs_join_transaction(root, 0); | |
3612b495 TI |
1076 | if (IS_ERR(ac->newtrans)) { |
1077 | int err = PTR_ERR(ac->newtrans); | |
1078 | kfree(ac); | |
1079 | return err; | |
1080 | } | |
bb9c12c9 SW |
1081 | |
1082 | /* take transaction reference */ | |
1083 | mutex_lock(&root->fs_info->trans_mutex); | |
1084 | cur_trans = trans->transaction; | |
13c5a93e | 1085 | atomic_inc(&cur_trans->use_count); |
bb9c12c9 SW |
1086 | mutex_unlock(&root->fs_info->trans_mutex); |
1087 | ||
1088 | btrfs_end_transaction(trans, root); | |
1089 | schedule_delayed_work(&ac->work, 0); | |
1090 | ||
1091 | /* wait for transaction to start and unblock */ | |
1092 | mutex_lock(&root->fs_info->trans_mutex); | |
1093 | if (wait_for_unblock) | |
1094 | wait_current_trans_commit_start_and_unblock(root, cur_trans); | |
1095 | else | |
1096 | wait_current_trans_commit_start(root, cur_trans); | |
1097 | put_transaction(cur_trans); | |
1098 | mutex_unlock(&root->fs_info->trans_mutex); | |
1099 | ||
1100 | return 0; | |
1101 | } | |
1102 | ||
1103 | /* | |
1104 | * btrfs_transaction state sequence: | |
1105 | * in_commit = 0, blocked = 0 (initial) | |
1106 | * in_commit = 1, blocked = 1 | |
1107 | * blocked = 0 | |
1108 | * commit_done = 1 | |
1109 | */ | |
79154b1b CM |
1110 | int btrfs_commit_transaction(struct btrfs_trans_handle *trans, |
1111 | struct btrfs_root *root) | |
1112 | { | |
15ee9bc7 | 1113 | unsigned long joined = 0; |
79154b1b | 1114 | struct btrfs_transaction *cur_trans; |
8fd17795 | 1115 | struct btrfs_transaction *prev_trans = NULL; |
79154b1b | 1116 | DEFINE_WAIT(wait); |
15ee9bc7 | 1117 | int ret; |
89573b9c CM |
1118 | int should_grow = 0; |
1119 | unsigned long now = get_seconds(); | |
dccae999 | 1120 | int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT); |
79154b1b | 1121 | |
5a3f23d5 CM |
1122 | btrfs_run_ordered_operations(root, 0); |
1123 | ||
56bec294 CM |
1124 | /* make a pass through all the delayed refs we have so far |
1125 | * any runnings procs may add more while we are here | |
1126 | */ | |
1127 | ret = btrfs_run_delayed_refs(trans, root, 0); | |
1128 | BUG_ON(ret); | |
1129 | ||
a22285a6 YZ |
1130 | btrfs_trans_release_metadata(trans, root); |
1131 | ||
b7ec40d7 | 1132 | cur_trans = trans->transaction; |
56bec294 CM |
1133 | /* |
1134 | * set the flushing flag so procs in this transaction have to | |
1135 | * start sending their work down. | |
1136 | */ | |
b7ec40d7 | 1137 | cur_trans->delayed_refs.flushing = 1; |
56bec294 | 1138 | |
c3e69d58 | 1139 | ret = btrfs_run_delayed_refs(trans, root, 0); |
56bec294 CM |
1140 | BUG_ON(ret); |
1141 | ||
79154b1b | 1142 | mutex_lock(&root->fs_info->trans_mutex); |
b7ec40d7 | 1143 | if (cur_trans->in_commit) { |
13c5a93e | 1144 | atomic_inc(&cur_trans->use_count); |
ccd467d6 | 1145 | mutex_unlock(&root->fs_info->trans_mutex); |
79154b1b | 1146 | btrfs_end_transaction(trans, root); |
ccd467d6 | 1147 | |
79154b1b CM |
1148 | ret = wait_for_commit(root, cur_trans); |
1149 | BUG_ON(ret); | |
15ee9bc7 JB |
1150 | |
1151 | mutex_lock(&root->fs_info->trans_mutex); | |
79154b1b | 1152 | put_transaction(cur_trans); |
15ee9bc7 JB |
1153 | mutex_unlock(&root->fs_info->trans_mutex); |
1154 | ||
79154b1b CM |
1155 | return 0; |
1156 | } | |
4313b399 | 1157 | |
2c90e5d6 | 1158 | trans->transaction->in_commit = 1; |
f9295749 | 1159 | trans->transaction->blocked = 1; |
bb9c12c9 SW |
1160 | wake_up(&root->fs_info->transaction_blocked_wait); |
1161 | ||
ccd467d6 CM |
1162 | if (cur_trans->list.prev != &root->fs_info->trans_list) { |
1163 | prev_trans = list_entry(cur_trans->list.prev, | |
1164 | struct btrfs_transaction, list); | |
1165 | if (!prev_trans->commit_done) { | |
13c5a93e | 1166 | atomic_inc(&prev_trans->use_count); |
ccd467d6 CM |
1167 | mutex_unlock(&root->fs_info->trans_mutex); |
1168 | ||
1169 | wait_for_commit(root, prev_trans); | |
ccd467d6 | 1170 | |
ccd467d6 | 1171 | mutex_lock(&root->fs_info->trans_mutex); |
15ee9bc7 | 1172 | put_transaction(prev_trans); |
ccd467d6 CM |
1173 | } |
1174 | } | |
15ee9bc7 | 1175 | |
89573b9c CM |
1176 | if (now < cur_trans->start_time || now - cur_trans->start_time < 1) |
1177 | should_grow = 1; | |
1178 | ||
15ee9bc7 | 1179 | do { |
7ea394f1 | 1180 | int snap_pending = 0; |
15ee9bc7 | 1181 | joined = cur_trans->num_joined; |
7ea394f1 YZ |
1182 | if (!list_empty(&trans->transaction->pending_snapshots)) |
1183 | snap_pending = 1; | |
1184 | ||
2c90e5d6 | 1185 | WARN_ON(cur_trans != trans->transaction); |
79154b1b | 1186 | mutex_unlock(&root->fs_info->trans_mutex); |
15ee9bc7 | 1187 | |
0bdb1db2 | 1188 | if (flush_on_commit || snap_pending) { |
24bbcf04 YZ |
1189 | btrfs_start_delalloc_inodes(root, 1); |
1190 | ret = btrfs_wait_ordered_extents(root, 0, 1); | |
ebecd3d9 | 1191 | BUG_ON(ret); |
7ea394f1 YZ |
1192 | } |
1193 | ||
16cdcec7 MX |
1194 | ret = btrfs_run_delayed_items(trans, root); |
1195 | BUG_ON(ret); | |
1196 | ||
5a3f23d5 CM |
1197 | /* |
1198 | * rename don't use btrfs_join_transaction, so, once we | |
1199 | * set the transaction to blocked above, we aren't going | |
1200 | * to get any new ordered operations. We can safely run | |
1201 | * it here and no for sure that nothing new will be added | |
1202 | * to the list | |
1203 | */ | |
1204 | btrfs_run_ordered_operations(root, 1); | |
1205 | ||
ed3b3d31 CM |
1206 | prepare_to_wait(&cur_trans->writer_wait, &wait, |
1207 | TASK_UNINTERRUPTIBLE); | |
1208 | ||
89573b9c | 1209 | smp_mb(); |
13c5a93e | 1210 | if (atomic_read(&cur_trans->num_writers) > 1) |
99d16cbc SW |
1211 | schedule_timeout(MAX_SCHEDULE_TIMEOUT); |
1212 | else if (should_grow) | |
1213 | schedule_timeout(1); | |
15ee9bc7 | 1214 | |
79154b1b | 1215 | mutex_lock(&root->fs_info->trans_mutex); |
15ee9bc7 | 1216 | finish_wait(&cur_trans->writer_wait, &wait); |
13c5a93e | 1217 | } while (atomic_read(&cur_trans->num_writers) > 1 || |
89573b9c | 1218 | (should_grow && cur_trans->num_joined != joined)); |
15ee9bc7 | 1219 | |
3063d29f CM |
1220 | ret = create_pending_snapshots(trans, root->fs_info); |
1221 | BUG_ON(ret); | |
1222 | ||
16cdcec7 MX |
1223 | ret = btrfs_run_delayed_items(trans, root); |
1224 | BUG_ON(ret); | |
1225 | ||
56bec294 CM |
1226 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
1227 | BUG_ON(ret); | |
1228 | ||
2c90e5d6 | 1229 | WARN_ON(cur_trans != trans->transaction); |
dc17ff8f | 1230 | |
a2de733c | 1231 | btrfs_scrub_pause(root); |
e02119d5 CM |
1232 | /* btrfs_commit_tree_roots is responsible for getting the |
1233 | * various roots consistent with each other. Every pointer | |
1234 | * in the tree of tree roots has to point to the most up to date | |
1235 | * root for every subvolume and other tree. So, we have to keep | |
1236 | * the tree logging code from jumping in and changing any | |
1237 | * of the trees. | |
1238 | * | |
1239 | * At this point in the commit, there can't be any tree-log | |
1240 | * writers, but a little lower down we drop the trans mutex | |
1241 | * and let new people in. By holding the tree_log_mutex | |
1242 | * from now until after the super is written, we avoid races | |
1243 | * with the tree-log code. | |
1244 | */ | |
1245 | mutex_lock(&root->fs_info->tree_log_mutex); | |
1246 | ||
5d4f98a2 | 1247 | ret = commit_fs_roots(trans, root); |
54aa1f4d CM |
1248 | BUG_ON(ret); |
1249 | ||
5d4f98a2 | 1250 | /* commit_fs_roots gets rid of all the tree log roots, it is now |
e02119d5 CM |
1251 | * safe to free the root of tree log roots |
1252 | */ | |
1253 | btrfs_free_log_root_tree(trans, root->fs_info); | |
1254 | ||
5d4f98a2 | 1255 | ret = commit_cowonly_roots(trans, root); |
79154b1b | 1256 | BUG_ON(ret); |
54aa1f4d | 1257 | |
11833d66 YZ |
1258 | btrfs_prepare_extent_commit(trans, root); |
1259 | ||
78fae27e | 1260 | cur_trans = root->fs_info->running_transaction; |
cee36a03 | 1261 | spin_lock(&root->fs_info->new_trans_lock); |
78fae27e | 1262 | root->fs_info->running_transaction = NULL; |
cee36a03 | 1263 | spin_unlock(&root->fs_info->new_trans_lock); |
5d4f98a2 YZ |
1264 | |
1265 | btrfs_set_root_node(&root->fs_info->tree_root->root_item, | |
1266 | root->fs_info->tree_root->node); | |
817d52f8 | 1267 | switch_commit_root(root->fs_info->tree_root); |
5d4f98a2 YZ |
1268 | |
1269 | btrfs_set_root_node(&root->fs_info->chunk_root->root_item, | |
1270 | root->fs_info->chunk_root->node); | |
817d52f8 | 1271 | switch_commit_root(root->fs_info->chunk_root); |
5d4f98a2 YZ |
1272 | |
1273 | update_super_roots(root); | |
e02119d5 CM |
1274 | |
1275 | if (!root->fs_info->log_root_recovering) { | |
1276 | btrfs_set_super_log_root(&root->fs_info->super_copy, 0); | |
1277 | btrfs_set_super_log_root_level(&root->fs_info->super_copy, 0); | |
1278 | } | |
1279 | ||
a061fc8d CM |
1280 | memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy, |
1281 | sizeof(root->fs_info->super_copy)); | |
ccd467d6 | 1282 | |
f9295749 | 1283 | trans->transaction->blocked = 0; |
b7ec40d7 | 1284 | |
f9295749 | 1285 | wake_up(&root->fs_info->transaction_wait); |
e6dcd2dc | 1286 | |
78fae27e | 1287 | mutex_unlock(&root->fs_info->trans_mutex); |
79154b1b CM |
1288 | ret = btrfs_write_and_wait_transaction(trans, root); |
1289 | BUG_ON(ret); | |
a512bbf8 | 1290 | write_ctree_super(trans, root, 0); |
4313b399 | 1291 | |
e02119d5 CM |
1292 | /* |
1293 | * the super is written, we can safely allow the tree-loggers | |
1294 | * to go about their business | |
1295 | */ | |
1296 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
1297 | ||
11833d66 | 1298 | btrfs_finish_extent_commit(trans, root); |
4313b399 | 1299 | |
1a40e23b ZY |
1300 | mutex_lock(&root->fs_info->trans_mutex); |
1301 | ||
2c90e5d6 | 1302 | cur_trans->commit_done = 1; |
b7ec40d7 | 1303 | |
15ee9bc7 | 1304 | root->fs_info->last_trans_committed = cur_trans->transid; |
817d52f8 | 1305 | |
2c90e5d6 | 1306 | wake_up(&cur_trans->commit_wait); |
3de4586c | 1307 | |
13c5a93e | 1308 | list_del_init(&cur_trans->list); |
78fae27e | 1309 | put_transaction(cur_trans); |
79154b1b | 1310 | put_transaction(cur_trans); |
58176a96 | 1311 | |
1abe9b8a | 1312 | trace_btrfs_transaction_commit(root); |
1313 | ||
78fae27e | 1314 | mutex_unlock(&root->fs_info->trans_mutex); |
3de4586c | 1315 | |
a2de733c AJ |
1316 | btrfs_scrub_continue(root); |
1317 | ||
9ed74f2d JB |
1318 | if (current->journal_info == trans) |
1319 | current->journal_info = NULL; | |
1320 | ||
2c90e5d6 | 1321 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
24bbcf04 YZ |
1322 | |
1323 | if (current != root->fs_info->transaction_kthread) | |
1324 | btrfs_run_delayed_iputs(root); | |
1325 | ||
79154b1b CM |
1326 | return ret; |
1327 | } | |
1328 | ||
d352ac68 CM |
1329 | /* |
1330 | * interface function to delete all the snapshots we have scheduled for deletion | |
1331 | */ | |
e9d0b13b CM |
1332 | int btrfs_clean_old_snapshots(struct btrfs_root *root) |
1333 | { | |
5d4f98a2 YZ |
1334 | LIST_HEAD(list); |
1335 | struct btrfs_fs_info *fs_info = root->fs_info; | |
1336 | ||
1337 | mutex_lock(&fs_info->trans_mutex); | |
1338 | list_splice_init(&fs_info->dead_roots, &list); | |
1339 | mutex_unlock(&fs_info->trans_mutex); | |
e9d0b13b | 1340 | |
5d4f98a2 YZ |
1341 | while (!list_empty(&list)) { |
1342 | root = list_entry(list.next, struct btrfs_root, root_list); | |
76dda93c YZ |
1343 | list_del(&root->root_list); |
1344 | ||
16cdcec7 MX |
1345 | btrfs_kill_all_delayed_nodes(root); |
1346 | ||
76dda93c YZ |
1347 | if (btrfs_header_backref_rev(root->node) < |
1348 | BTRFS_MIXED_BACKREF_REV) | |
3fd0a558 | 1349 | btrfs_drop_snapshot(root, NULL, 0); |
76dda93c | 1350 | else |
3fd0a558 | 1351 | btrfs_drop_snapshot(root, NULL, 1); |
e9d0b13b CM |
1352 | } |
1353 | return 0; | |
1354 | } |