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" |
79154b1b | 30 | |
0f7d52f4 CM |
31 | #define BTRFS_ROOT_TRANS_TAG 0 |
32 | ||
80b6794d | 33 | static noinline void put_transaction(struct btrfs_transaction *transaction) |
79154b1b | 34 | { |
2c90e5d6 | 35 | WARN_ON(transaction->use_count == 0); |
79154b1b | 36 | transaction->use_count--; |
78fae27e | 37 | if (transaction->use_count == 0) { |
8fd17795 | 38 | list_del_init(&transaction->list); |
2c90e5d6 CM |
39 | memset(transaction, 0, sizeof(*transaction)); |
40 | kmem_cache_free(btrfs_transaction_cachep, transaction); | |
78fae27e | 41 | } |
79154b1b CM |
42 | } |
43 | ||
817d52f8 JB |
44 | static noinline void switch_commit_root(struct btrfs_root *root) |
45 | { | |
817d52f8 JB |
46 | free_extent_buffer(root->commit_root); |
47 | root->commit_root = btrfs_root_node(root); | |
817d52f8 JB |
48 | } |
49 | ||
d352ac68 CM |
50 | /* |
51 | * either allocate a new transaction or hop into the existing one | |
52 | */ | |
80b6794d | 53 | static noinline int join_transaction(struct btrfs_root *root) |
79154b1b CM |
54 | { |
55 | struct btrfs_transaction *cur_trans; | |
56 | cur_trans = root->fs_info->running_transaction; | |
57 | if (!cur_trans) { | |
2c90e5d6 CM |
58 | cur_trans = kmem_cache_alloc(btrfs_transaction_cachep, |
59 | GFP_NOFS); | |
79154b1b | 60 | BUG_ON(!cur_trans); |
0f7d52f4 | 61 | root->fs_info->generation++; |
15ee9bc7 JB |
62 | cur_trans->num_writers = 1; |
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; |
d5719762 | 69 | 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, |
5f39d397 CM |
84 | root->fs_info->btree_inode->i_mapping, |
85 | GFP_NOFS); | |
48ec2cf8 CM |
86 | spin_lock(&root->fs_info->new_trans_lock); |
87 | root->fs_info->running_transaction = cur_trans; | |
88 | spin_unlock(&root->fs_info->new_trans_lock); | |
15ee9bc7 JB |
89 | } else { |
90 | cur_trans->num_writers++; | |
91 | cur_trans->num_joined++; | |
79154b1b | 92 | } |
15ee9bc7 | 93 | |
79154b1b CM |
94 | return 0; |
95 | } | |
96 | ||
d352ac68 | 97 | /* |
d397712b CM |
98 | * this does all the record keeping required to make sure that a reference |
99 | * counted root is properly recorded in a given transaction. This is required | |
100 | * to make sure the old root from before we joined the transaction is deleted | |
101 | * when the transaction commits | |
d352ac68 | 102 | */ |
5d4f98a2 YZ |
103 | static noinline int record_root_in_trans(struct btrfs_trans_handle *trans, |
104 | struct btrfs_root *root) | |
6702ed49 | 105 | { |
5d4f98a2 | 106 | if (root->ref_cows && root->last_trans < trans->transid) { |
6702ed49 | 107 | WARN_ON(root == root->fs_info->extent_root); |
5d4f98a2 YZ |
108 | WARN_ON(root->commit_root != root->node); |
109 | ||
110 | radix_tree_tag_set(&root->fs_info->fs_roots_radix, | |
111 | (unsigned long)root->root_key.objectid, | |
112 | BTRFS_ROOT_TRANS_TAG); | |
113 | root->last_trans = trans->transid; | |
114 | btrfs_init_reloc_root(trans, root); | |
115 | } | |
116 | return 0; | |
117 | } | |
bcc63abb | 118 | |
5d4f98a2 YZ |
119 | int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, |
120 | struct btrfs_root *root) | |
121 | { | |
122 | if (!root->ref_cows) | |
123 | return 0; | |
bcc63abb | 124 | |
5d4f98a2 YZ |
125 | mutex_lock(&root->fs_info->trans_mutex); |
126 | if (root->last_trans == trans->transid) { | |
127 | mutex_unlock(&root->fs_info->trans_mutex); | |
128 | return 0; | |
6702ed49 | 129 | } |
5d4f98a2 YZ |
130 | |
131 | record_root_in_trans(trans, root); | |
132 | mutex_unlock(&root->fs_info->trans_mutex); | |
6702ed49 CM |
133 | return 0; |
134 | } | |
135 | ||
d352ac68 CM |
136 | /* wait for commit against the current transaction to become unblocked |
137 | * when this is done, it is safe to start a new transaction, but the current | |
138 | * transaction might not be fully on disk. | |
139 | */ | |
37d1aeee | 140 | static void wait_current_trans(struct btrfs_root *root) |
79154b1b | 141 | { |
f9295749 | 142 | struct btrfs_transaction *cur_trans; |
79154b1b | 143 | |
f9295749 | 144 | cur_trans = root->fs_info->running_transaction; |
37d1aeee | 145 | if (cur_trans && cur_trans->blocked) { |
f9295749 CM |
146 | DEFINE_WAIT(wait); |
147 | cur_trans->use_count++; | |
d397712b | 148 | while (1) { |
f9295749 CM |
149 | prepare_to_wait(&root->fs_info->transaction_wait, &wait, |
150 | TASK_UNINTERRUPTIBLE); | |
471fa17d | 151 | if (!cur_trans->blocked) |
f9295749 | 152 | break; |
471fa17d ZL |
153 | mutex_unlock(&root->fs_info->trans_mutex); |
154 | schedule(); | |
155 | mutex_lock(&root->fs_info->trans_mutex); | |
f9295749 | 156 | } |
471fa17d | 157 | finish_wait(&root->fs_info->transaction_wait, &wait); |
f9295749 CM |
158 | put_transaction(cur_trans); |
159 | } | |
37d1aeee CM |
160 | } |
161 | ||
249ac1e5 JB |
162 | enum btrfs_trans_type { |
163 | TRANS_START, | |
164 | TRANS_JOIN, | |
165 | TRANS_USERSPACE, | |
0af3d00b | 166 | TRANS_JOIN_NOLOCK, |
249ac1e5 JB |
167 | }; |
168 | ||
a22285a6 YZ |
169 | static int may_wait_transaction(struct btrfs_root *root, int type) |
170 | { | |
171 | if (!root->fs_info->log_root_recovering && | |
172 | ((type == TRANS_START && !root->fs_info->open_ioctl_trans) || | |
173 | type == TRANS_USERSPACE)) | |
174 | return 1; | |
175 | return 0; | |
176 | } | |
177 | ||
e02119d5 | 178 | static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root, |
a22285a6 | 179 | u64 num_items, int type) |
37d1aeee | 180 | { |
a22285a6 YZ |
181 | struct btrfs_trans_handle *h; |
182 | struct btrfs_transaction *cur_trans; | |
37d1aeee | 183 | int ret; |
a22285a6 YZ |
184 | again: |
185 | h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS); | |
186 | if (!h) | |
187 | return ERR_PTR(-ENOMEM); | |
37d1aeee | 188 | |
0af3d00b JB |
189 | if (type != TRANS_JOIN_NOLOCK) |
190 | mutex_lock(&root->fs_info->trans_mutex); | |
a22285a6 | 191 | if (may_wait_transaction(root, type)) |
37d1aeee | 192 | wait_current_trans(root); |
a22285a6 | 193 | |
79154b1b CM |
194 | ret = join_transaction(root); |
195 | BUG_ON(ret); | |
0f7d52f4 | 196 | |
a22285a6 YZ |
197 | cur_trans = root->fs_info->running_transaction; |
198 | cur_trans->use_count++; | |
0af3d00b JB |
199 | if (type != TRANS_JOIN_NOLOCK) |
200 | mutex_unlock(&root->fs_info->trans_mutex); | |
a22285a6 YZ |
201 | |
202 | h->transid = cur_trans->transid; | |
203 | h->transaction = cur_trans; | |
79154b1b | 204 | h->blocks_used = 0; |
d2fb3437 | 205 | h->block_group = 0; |
a22285a6 | 206 | h->bytes_reserved = 0; |
56bec294 | 207 | h->delayed_ref_updates = 0; |
f0486c68 | 208 | h->block_rsv = NULL; |
b7ec40d7 | 209 | |
a22285a6 YZ |
210 | smp_mb(); |
211 | if (cur_trans->blocked && may_wait_transaction(root, type)) { | |
212 | btrfs_commit_transaction(h, root); | |
213 | goto again; | |
214 | } | |
215 | ||
216 | if (num_items > 0) { | |
8bb8ab2e | 217 | ret = btrfs_trans_reserve_metadata(h, root, num_items); |
a22285a6 YZ |
218 | if (ret == -EAGAIN) { |
219 | btrfs_commit_transaction(h, root); | |
220 | goto again; | |
221 | } | |
222 | if (ret < 0) { | |
223 | btrfs_end_transaction(h, root); | |
224 | return ERR_PTR(ret); | |
225 | } | |
226 | } | |
9ed74f2d | 227 | |
0af3d00b JB |
228 | if (type != TRANS_JOIN_NOLOCK) |
229 | mutex_lock(&root->fs_info->trans_mutex); | |
5d4f98a2 | 230 | record_root_in_trans(h, root); |
0af3d00b JB |
231 | if (type != TRANS_JOIN_NOLOCK) |
232 | mutex_unlock(&root->fs_info->trans_mutex); | |
a22285a6 YZ |
233 | |
234 | if (!current->journal_info && type != TRANS_USERSPACE) | |
235 | current->journal_info = h; | |
79154b1b CM |
236 | return h; |
237 | } | |
238 | ||
f9295749 | 239 | struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, |
a22285a6 | 240 | int num_items) |
f9295749 | 241 | { |
a22285a6 | 242 | return start_transaction(root, num_items, TRANS_START); |
f9295749 CM |
243 | } |
244 | struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root, | |
245 | int num_blocks) | |
246 | { | |
a22285a6 | 247 | return start_transaction(root, 0, TRANS_JOIN); |
f9295749 CM |
248 | } |
249 | ||
0af3d00b JB |
250 | struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root, |
251 | int num_blocks) | |
252 | { | |
253 | return start_transaction(root, 0, TRANS_JOIN_NOLOCK); | |
254 | } | |
255 | ||
9ca9ee09 SW |
256 | struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *r, |
257 | int num_blocks) | |
258 | { | |
a22285a6 | 259 | return start_transaction(r, 0, TRANS_USERSPACE); |
9ca9ee09 SW |
260 | } |
261 | ||
d352ac68 | 262 | /* wait for a transaction commit to be fully complete */ |
89ce8a63 CM |
263 | static noinline int wait_for_commit(struct btrfs_root *root, |
264 | struct btrfs_transaction *commit) | |
265 | { | |
266 | DEFINE_WAIT(wait); | |
267 | mutex_lock(&root->fs_info->trans_mutex); | |
d397712b | 268 | while (!commit->commit_done) { |
89ce8a63 CM |
269 | prepare_to_wait(&commit->commit_wait, &wait, |
270 | TASK_UNINTERRUPTIBLE); | |
271 | if (commit->commit_done) | |
272 | break; | |
273 | mutex_unlock(&root->fs_info->trans_mutex); | |
274 | schedule(); | |
275 | mutex_lock(&root->fs_info->trans_mutex); | |
276 | } | |
277 | mutex_unlock(&root->fs_info->trans_mutex); | |
278 | finish_wait(&commit->commit_wait, &wait); | |
279 | return 0; | |
280 | } | |
281 | ||
46204592 SW |
282 | int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid) |
283 | { | |
284 | struct btrfs_transaction *cur_trans = NULL, *t; | |
285 | int ret; | |
286 | ||
287 | mutex_lock(&root->fs_info->trans_mutex); | |
288 | ||
289 | ret = 0; | |
290 | if (transid) { | |
291 | if (transid <= root->fs_info->last_trans_committed) | |
292 | goto out_unlock; | |
293 | ||
294 | /* find specified transaction */ | |
295 | list_for_each_entry(t, &root->fs_info->trans_list, list) { | |
296 | if (t->transid == transid) { | |
297 | cur_trans = t; | |
298 | break; | |
299 | } | |
300 | if (t->transid > transid) | |
301 | break; | |
302 | } | |
303 | ret = -EINVAL; | |
304 | if (!cur_trans) | |
305 | goto out_unlock; /* bad transid */ | |
306 | } else { | |
307 | /* find newest transaction that is committing | committed */ | |
308 | list_for_each_entry_reverse(t, &root->fs_info->trans_list, | |
309 | list) { | |
310 | if (t->in_commit) { | |
311 | if (t->commit_done) | |
312 | goto out_unlock; | |
313 | cur_trans = t; | |
314 | break; | |
315 | } | |
316 | } | |
317 | if (!cur_trans) | |
318 | goto out_unlock; /* nothing committing|committed */ | |
319 | } | |
320 | ||
321 | cur_trans->use_count++; | |
322 | mutex_unlock(&root->fs_info->trans_mutex); | |
323 | ||
324 | wait_for_commit(root, cur_trans); | |
325 | ||
326 | mutex_lock(&root->fs_info->trans_mutex); | |
327 | put_transaction(cur_trans); | |
328 | ret = 0; | |
329 | out_unlock: | |
330 | mutex_unlock(&root->fs_info->trans_mutex); | |
331 | return ret; | |
332 | } | |
333 | ||
5d4f98a2 | 334 | #if 0 |
d352ac68 | 335 | /* |
d397712b CM |
336 | * rate limit against the drop_snapshot code. This helps to slow down new |
337 | * operations if the drop_snapshot code isn't able to keep up. | |
d352ac68 | 338 | */ |
37d1aeee | 339 | static void throttle_on_drops(struct btrfs_root *root) |
ab78c84d CM |
340 | { |
341 | struct btrfs_fs_info *info = root->fs_info; | |
2dd3e67b | 342 | int harder_count = 0; |
ab78c84d | 343 | |
2dd3e67b | 344 | harder: |
ab78c84d CM |
345 | if (atomic_read(&info->throttles)) { |
346 | DEFINE_WAIT(wait); | |
347 | int thr; | |
ab78c84d CM |
348 | thr = atomic_read(&info->throttle_gen); |
349 | ||
350 | do { | |
351 | prepare_to_wait(&info->transaction_throttle, | |
352 | &wait, TASK_UNINTERRUPTIBLE); | |
353 | if (!atomic_read(&info->throttles)) { | |
354 | finish_wait(&info->transaction_throttle, &wait); | |
355 | break; | |
356 | } | |
357 | schedule(); | |
358 | finish_wait(&info->transaction_throttle, &wait); | |
359 | } while (thr == atomic_read(&info->throttle_gen)); | |
2dd3e67b CM |
360 | harder_count++; |
361 | ||
362 | if (root->fs_info->total_ref_cache_size > 1 * 1024 * 1024 && | |
363 | harder_count < 2) | |
364 | goto harder; | |
365 | ||
366 | if (root->fs_info->total_ref_cache_size > 5 * 1024 * 1024 && | |
367 | harder_count < 10) | |
368 | goto harder; | |
369 | ||
370 | if (root->fs_info->total_ref_cache_size > 10 * 1024 * 1024 && | |
371 | harder_count < 20) | |
372 | goto harder; | |
ab78c84d CM |
373 | } |
374 | } | |
5d4f98a2 | 375 | #endif |
ab78c84d | 376 | |
37d1aeee CM |
377 | void btrfs_throttle(struct btrfs_root *root) |
378 | { | |
379 | mutex_lock(&root->fs_info->trans_mutex); | |
9ca9ee09 SW |
380 | if (!root->fs_info->open_ioctl_trans) |
381 | wait_current_trans(root); | |
37d1aeee | 382 | mutex_unlock(&root->fs_info->trans_mutex); |
37d1aeee CM |
383 | } |
384 | ||
8929ecfa YZ |
385 | static int should_end_transaction(struct btrfs_trans_handle *trans, |
386 | struct btrfs_root *root) | |
387 | { | |
388 | int ret; | |
389 | ret = btrfs_block_rsv_check(trans, root, | |
390 | &root->fs_info->global_block_rsv, 0, 5); | |
391 | return ret ? 1 : 0; | |
392 | } | |
393 | ||
394 | int btrfs_should_end_transaction(struct btrfs_trans_handle *trans, | |
395 | struct btrfs_root *root) | |
396 | { | |
397 | struct btrfs_transaction *cur_trans = trans->transaction; | |
398 | int updates; | |
399 | ||
400 | if (cur_trans->blocked || cur_trans->delayed_refs.flushing) | |
401 | return 1; | |
402 | ||
403 | updates = trans->delayed_ref_updates; | |
404 | trans->delayed_ref_updates = 0; | |
405 | if (updates) | |
406 | btrfs_run_delayed_refs(trans, root, updates); | |
407 | ||
408 | return should_end_transaction(trans, root); | |
409 | } | |
410 | ||
89ce8a63 | 411 | static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, |
0af3d00b | 412 | struct btrfs_root *root, int throttle, int lock) |
79154b1b | 413 | { |
8929ecfa | 414 | struct btrfs_transaction *cur_trans = trans->transaction; |
ab78c84d | 415 | struct btrfs_fs_info *info = root->fs_info; |
c3e69d58 CM |
416 | int count = 0; |
417 | ||
418 | while (count < 4) { | |
419 | unsigned long cur = trans->delayed_ref_updates; | |
420 | trans->delayed_ref_updates = 0; | |
421 | if (cur && | |
422 | trans->transaction->delayed_refs.num_heads_ready > 64) { | |
423 | trans->delayed_ref_updates = 0; | |
b7ec40d7 CM |
424 | |
425 | /* | |
426 | * do a full flush if the transaction is trying | |
427 | * to close | |
428 | */ | |
429 | if (trans->transaction->delayed_refs.flushing) | |
430 | cur = 0; | |
c3e69d58 CM |
431 | btrfs_run_delayed_refs(trans, root, cur); |
432 | } else { | |
433 | break; | |
434 | } | |
435 | count++; | |
56bec294 CM |
436 | } |
437 | ||
a22285a6 YZ |
438 | btrfs_trans_release_metadata(trans, root); |
439 | ||
0af3d00b | 440 | if (lock && !root->fs_info->open_ioctl_trans && |
8929ecfa YZ |
441 | should_end_transaction(trans, root)) |
442 | trans->transaction->blocked = 1; | |
443 | ||
0af3d00b | 444 | if (lock && cur_trans->blocked && !cur_trans->in_commit) { |
8929ecfa YZ |
445 | if (throttle) |
446 | return btrfs_commit_transaction(trans, root); | |
447 | else | |
448 | wake_up_process(info->transaction_kthread); | |
449 | } | |
450 | ||
0af3d00b JB |
451 | if (lock) |
452 | mutex_lock(&info->trans_mutex); | |
8929ecfa | 453 | WARN_ON(cur_trans != info->running_transaction); |
d5719762 | 454 | WARN_ON(cur_trans->num_writers < 1); |
ccd467d6 | 455 | cur_trans->num_writers--; |
89ce8a63 | 456 | |
99d16cbc | 457 | smp_mb(); |
79154b1b CM |
458 | if (waitqueue_active(&cur_trans->writer_wait)) |
459 | wake_up(&cur_trans->writer_wait); | |
79154b1b | 460 | put_transaction(cur_trans); |
0af3d00b JB |
461 | if (lock) |
462 | mutex_unlock(&info->trans_mutex); | |
9ed74f2d JB |
463 | |
464 | if (current->journal_info == trans) | |
465 | current->journal_info = NULL; | |
d6025579 | 466 | memset(trans, 0, sizeof(*trans)); |
2c90e5d6 | 467 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
ab78c84d | 468 | |
24bbcf04 YZ |
469 | if (throttle) |
470 | btrfs_run_delayed_iputs(root); | |
471 | ||
79154b1b CM |
472 | return 0; |
473 | } | |
474 | ||
89ce8a63 CM |
475 | int btrfs_end_transaction(struct btrfs_trans_handle *trans, |
476 | struct btrfs_root *root) | |
477 | { | |
0af3d00b | 478 | return __btrfs_end_transaction(trans, root, 0, 1); |
89ce8a63 CM |
479 | } |
480 | ||
481 | int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans, | |
482 | struct btrfs_root *root) | |
483 | { | |
0af3d00b JB |
484 | return __btrfs_end_transaction(trans, root, 1, 1); |
485 | } | |
486 | ||
487 | int btrfs_end_transaction_nolock(struct btrfs_trans_handle *trans, | |
488 | struct btrfs_root *root) | |
489 | { | |
490 | return __btrfs_end_transaction(trans, root, 0, 0); | |
89ce8a63 CM |
491 | } |
492 | ||
d352ac68 CM |
493 | /* |
494 | * when btree blocks are allocated, they have some corresponding bits set for | |
495 | * them in one of two extent_io trees. This is used to make sure all of | |
690587d1 | 496 | * those extents are sent to disk but does not wait on them |
d352ac68 | 497 | */ |
690587d1 | 498 | int btrfs_write_marked_extents(struct btrfs_root *root, |
8cef4e16 | 499 | struct extent_io_tree *dirty_pages, int mark) |
79154b1b | 500 | { |
7c4452b9 | 501 | int ret; |
777e6bd7 | 502 | int err = 0; |
7c4452b9 CM |
503 | int werr = 0; |
504 | struct page *page; | |
7c4452b9 | 505 | struct inode *btree_inode = root->fs_info->btree_inode; |
777e6bd7 | 506 | u64 start = 0; |
5f39d397 CM |
507 | u64 end; |
508 | unsigned long index; | |
7c4452b9 | 509 | |
d397712b | 510 | while (1) { |
777e6bd7 | 511 | ret = find_first_extent_bit(dirty_pages, start, &start, &end, |
8cef4e16 | 512 | mark); |
5f39d397 | 513 | if (ret) |
7c4452b9 | 514 | break; |
d397712b | 515 | while (start <= end) { |
777e6bd7 CM |
516 | cond_resched(); |
517 | ||
5f39d397 | 518 | index = start >> PAGE_CACHE_SHIFT; |
35ebb934 | 519 | start = (u64)(index + 1) << PAGE_CACHE_SHIFT; |
4bef0848 | 520 | page = find_get_page(btree_inode->i_mapping, index); |
7c4452b9 CM |
521 | if (!page) |
522 | continue; | |
4bef0848 CM |
523 | |
524 | btree_lock_page_hook(page); | |
525 | if (!page->mapping) { | |
526 | unlock_page(page); | |
527 | page_cache_release(page); | |
528 | continue; | |
529 | } | |
530 | ||
6702ed49 CM |
531 | if (PageWriteback(page)) { |
532 | if (PageDirty(page)) | |
533 | wait_on_page_writeback(page); | |
534 | else { | |
535 | unlock_page(page); | |
536 | page_cache_release(page); | |
537 | continue; | |
538 | } | |
539 | } | |
7c4452b9 CM |
540 | err = write_one_page(page, 0); |
541 | if (err) | |
542 | werr = err; | |
543 | page_cache_release(page); | |
544 | } | |
545 | } | |
690587d1 CM |
546 | if (err) |
547 | werr = err; | |
548 | return werr; | |
549 | } | |
550 | ||
551 | /* | |
552 | * when btree blocks are allocated, they have some corresponding bits set for | |
553 | * them in one of two extent_io trees. This is used to make sure all of | |
554 | * those extents are on disk for transaction or log commit. We wait | |
555 | * on all the pages and clear them from the dirty pages state tree | |
556 | */ | |
557 | int btrfs_wait_marked_extents(struct btrfs_root *root, | |
8cef4e16 | 558 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 CM |
559 | { |
560 | int ret; | |
561 | int err = 0; | |
562 | int werr = 0; | |
563 | struct page *page; | |
564 | struct inode *btree_inode = root->fs_info->btree_inode; | |
565 | u64 start = 0; | |
566 | u64 end; | |
567 | unsigned long index; | |
568 | ||
d397712b | 569 | while (1) { |
8cef4e16 YZ |
570 | ret = find_first_extent_bit(dirty_pages, start, &start, &end, |
571 | mark); | |
777e6bd7 CM |
572 | if (ret) |
573 | break; | |
574 | ||
8cef4e16 | 575 | clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS); |
d397712b | 576 | while (start <= end) { |
777e6bd7 CM |
577 | index = start >> PAGE_CACHE_SHIFT; |
578 | start = (u64)(index + 1) << PAGE_CACHE_SHIFT; | |
579 | page = find_get_page(btree_inode->i_mapping, index); | |
580 | if (!page) | |
581 | continue; | |
582 | if (PageDirty(page)) { | |
4bef0848 CM |
583 | btree_lock_page_hook(page); |
584 | wait_on_page_writeback(page); | |
777e6bd7 CM |
585 | err = write_one_page(page, 0); |
586 | if (err) | |
587 | werr = err; | |
588 | } | |
105d931d | 589 | wait_on_page_writeback(page); |
777e6bd7 CM |
590 | page_cache_release(page); |
591 | cond_resched(); | |
592 | } | |
593 | } | |
7c4452b9 CM |
594 | if (err) |
595 | werr = err; | |
596 | return werr; | |
79154b1b CM |
597 | } |
598 | ||
690587d1 CM |
599 | /* |
600 | * when btree blocks are allocated, they have some corresponding bits set for | |
601 | * them in one of two extent_io trees. This is used to make sure all of | |
602 | * those extents are on disk for transaction or log commit | |
603 | */ | |
604 | int btrfs_write_and_wait_marked_extents(struct btrfs_root *root, | |
8cef4e16 | 605 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 CM |
606 | { |
607 | int ret; | |
608 | int ret2; | |
609 | ||
8cef4e16 YZ |
610 | ret = btrfs_write_marked_extents(root, dirty_pages, mark); |
611 | ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark); | |
690587d1 CM |
612 | return ret || ret2; |
613 | } | |
614 | ||
d0c803c4 CM |
615 | int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans, |
616 | struct btrfs_root *root) | |
617 | { | |
618 | if (!trans || !trans->transaction) { | |
619 | struct inode *btree_inode; | |
620 | btree_inode = root->fs_info->btree_inode; | |
621 | return filemap_write_and_wait(btree_inode->i_mapping); | |
622 | } | |
623 | return btrfs_write_and_wait_marked_extents(root, | |
8cef4e16 YZ |
624 | &trans->transaction->dirty_pages, |
625 | EXTENT_DIRTY); | |
d0c803c4 CM |
626 | } |
627 | ||
d352ac68 CM |
628 | /* |
629 | * this is used to update the root pointer in the tree of tree roots. | |
630 | * | |
631 | * But, in the case of the extent allocation tree, updating the root | |
632 | * pointer may allocate blocks which may change the root of the extent | |
633 | * allocation tree. | |
634 | * | |
635 | * So, this loops and repeats and makes sure the cowonly root didn't | |
636 | * change while the root pointer was being updated in the metadata. | |
637 | */ | |
0b86a832 CM |
638 | static int update_cowonly_root(struct btrfs_trans_handle *trans, |
639 | struct btrfs_root *root) | |
79154b1b CM |
640 | { |
641 | int ret; | |
0b86a832 | 642 | u64 old_root_bytenr; |
86b9f2ec | 643 | u64 old_root_used; |
0b86a832 | 644 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
79154b1b | 645 | |
86b9f2ec | 646 | old_root_used = btrfs_root_used(&root->root_item); |
0b86a832 | 647 | btrfs_write_dirty_block_groups(trans, root); |
56bec294 | 648 | |
d397712b | 649 | while (1) { |
0b86a832 | 650 | old_root_bytenr = btrfs_root_bytenr(&root->root_item); |
86b9f2ec YZ |
651 | if (old_root_bytenr == root->node->start && |
652 | old_root_used == btrfs_root_used(&root->root_item)) | |
79154b1b | 653 | break; |
87ef2bb4 | 654 | |
5d4f98a2 | 655 | btrfs_set_root_node(&root->root_item, root->node); |
79154b1b | 656 | ret = btrfs_update_root(trans, tree_root, |
0b86a832 CM |
657 | &root->root_key, |
658 | &root->root_item); | |
79154b1b | 659 | BUG_ON(ret); |
56bec294 | 660 | |
86b9f2ec | 661 | old_root_used = btrfs_root_used(&root->root_item); |
4a8c9a62 | 662 | ret = btrfs_write_dirty_block_groups(trans, root); |
56bec294 | 663 | BUG_ON(ret); |
0b86a832 | 664 | } |
276e680d YZ |
665 | |
666 | if (root != root->fs_info->extent_root) | |
667 | switch_commit_root(root); | |
668 | ||
0b86a832 CM |
669 | return 0; |
670 | } | |
671 | ||
d352ac68 CM |
672 | /* |
673 | * update all the cowonly tree roots on disk | |
674 | */ | |
5d4f98a2 YZ |
675 | static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans, |
676 | struct btrfs_root *root) | |
0b86a832 CM |
677 | { |
678 | struct btrfs_fs_info *fs_info = root->fs_info; | |
679 | struct list_head *next; | |
84234f3a | 680 | struct extent_buffer *eb; |
56bec294 | 681 | int ret; |
84234f3a | 682 | |
56bec294 CM |
683 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
684 | BUG_ON(ret); | |
87ef2bb4 | 685 | |
84234f3a | 686 | eb = btrfs_lock_root_node(fs_info->tree_root); |
9fa8cfe7 | 687 | btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 0, &eb); |
84234f3a YZ |
688 | btrfs_tree_unlock(eb); |
689 | free_extent_buffer(eb); | |
0b86a832 | 690 | |
56bec294 CM |
691 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
692 | BUG_ON(ret); | |
87ef2bb4 | 693 | |
d397712b | 694 | while (!list_empty(&fs_info->dirty_cowonly_roots)) { |
0b86a832 CM |
695 | next = fs_info->dirty_cowonly_roots.next; |
696 | list_del_init(next); | |
697 | root = list_entry(next, struct btrfs_root, dirty_list); | |
87ef2bb4 | 698 | |
0b86a832 | 699 | update_cowonly_root(trans, root); |
79154b1b | 700 | } |
276e680d YZ |
701 | |
702 | down_write(&fs_info->extent_commit_sem); | |
703 | switch_commit_root(fs_info->extent_root); | |
704 | up_write(&fs_info->extent_commit_sem); | |
705 | ||
79154b1b CM |
706 | return 0; |
707 | } | |
708 | ||
d352ac68 CM |
709 | /* |
710 | * dead roots are old snapshots that need to be deleted. This allocates | |
711 | * a dirty root struct and adds it into the list of dead roots that need to | |
712 | * be deleted | |
713 | */ | |
5d4f98a2 | 714 | int btrfs_add_dead_root(struct btrfs_root *root) |
5eda7b5e | 715 | { |
b48652c1 | 716 | mutex_lock(&root->fs_info->trans_mutex); |
5d4f98a2 | 717 | list_add(&root->root_list, &root->fs_info->dead_roots); |
b48652c1 | 718 | mutex_unlock(&root->fs_info->trans_mutex); |
5eda7b5e CM |
719 | return 0; |
720 | } | |
721 | ||
d352ac68 | 722 | /* |
5d4f98a2 | 723 | * update all the cowonly tree roots on disk |
d352ac68 | 724 | */ |
5d4f98a2 YZ |
725 | static noinline int commit_fs_roots(struct btrfs_trans_handle *trans, |
726 | struct btrfs_root *root) | |
0f7d52f4 | 727 | { |
0f7d52f4 | 728 | struct btrfs_root *gang[8]; |
5d4f98a2 | 729 | struct btrfs_fs_info *fs_info = root->fs_info; |
0f7d52f4 CM |
730 | int i; |
731 | int ret; | |
54aa1f4d CM |
732 | int err = 0; |
733 | ||
d397712b | 734 | while (1) { |
5d4f98a2 YZ |
735 | ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, |
736 | (void **)gang, 0, | |
0f7d52f4 CM |
737 | ARRAY_SIZE(gang), |
738 | BTRFS_ROOT_TRANS_TAG); | |
739 | if (ret == 0) | |
740 | break; | |
741 | for (i = 0; i < ret; i++) { | |
742 | root = gang[i]; | |
5d4f98a2 YZ |
743 | radix_tree_tag_clear(&fs_info->fs_roots_radix, |
744 | (unsigned long)root->root_key.objectid, | |
745 | BTRFS_ROOT_TRANS_TAG); | |
31153d81 | 746 | |
e02119d5 | 747 | btrfs_free_log(trans, root); |
5d4f98a2 | 748 | btrfs_update_reloc_root(trans, root); |
d68fc57b | 749 | btrfs_orphan_commit_root(trans, root); |
bcc63abb | 750 | |
978d910d | 751 | if (root->commit_root != root->node) { |
817d52f8 | 752 | switch_commit_root(root); |
978d910d YZ |
753 | btrfs_set_root_node(&root->root_item, |
754 | root->node); | |
755 | } | |
5d4f98a2 | 756 | |
5d4f98a2 | 757 | err = btrfs_update_root(trans, fs_info->tree_root, |
0f7d52f4 CM |
758 | &root->root_key, |
759 | &root->root_item); | |
54aa1f4d CM |
760 | if (err) |
761 | break; | |
0f7d52f4 CM |
762 | } |
763 | } | |
54aa1f4d | 764 | return err; |
0f7d52f4 CM |
765 | } |
766 | ||
d352ac68 CM |
767 | /* |
768 | * defrag a given btree. If cacheonly == 1, this won't read from the disk, | |
769 | * otherwise every leaf in the btree is read and defragged. | |
770 | */ | |
e9d0b13b CM |
771 | int btrfs_defrag_root(struct btrfs_root *root, int cacheonly) |
772 | { | |
773 | struct btrfs_fs_info *info = root->fs_info; | |
e9d0b13b | 774 | struct btrfs_trans_handle *trans; |
8929ecfa | 775 | int ret; |
d3c2fdcf | 776 | unsigned long nr; |
e9d0b13b | 777 | |
8929ecfa | 778 | if (xchg(&root->defrag_running, 1)) |
e9d0b13b | 779 | return 0; |
8929ecfa | 780 | |
6b80053d | 781 | while (1) { |
8929ecfa YZ |
782 | trans = btrfs_start_transaction(root, 0); |
783 | if (IS_ERR(trans)) | |
784 | return PTR_ERR(trans); | |
785 | ||
e9d0b13b | 786 | ret = btrfs_defrag_leaves(trans, root, cacheonly); |
8929ecfa | 787 | |
d3c2fdcf | 788 | nr = trans->blocks_used; |
e9d0b13b | 789 | btrfs_end_transaction(trans, root); |
d3c2fdcf | 790 | btrfs_btree_balance_dirty(info->tree_root, nr); |
e9d0b13b CM |
791 | cond_resched(); |
792 | ||
3f157a2f | 793 | if (root->fs_info->closing || ret != -EAGAIN) |
e9d0b13b CM |
794 | break; |
795 | } | |
796 | root->defrag_running = 0; | |
8929ecfa | 797 | return ret; |
e9d0b13b CM |
798 | } |
799 | ||
2c47e605 | 800 | #if 0 |
b7ec40d7 CM |
801 | /* |
802 | * when dropping snapshots, we generate a ton of delayed refs, and it makes | |
803 | * sense not to join the transaction while it is trying to flush the current | |
804 | * queue of delayed refs out. | |
805 | * | |
806 | * This is used by the drop snapshot code only | |
807 | */ | |
808 | static noinline int wait_transaction_pre_flush(struct btrfs_fs_info *info) | |
809 | { | |
810 | DEFINE_WAIT(wait); | |
811 | ||
812 | mutex_lock(&info->trans_mutex); | |
813 | while (info->running_transaction && | |
814 | info->running_transaction->delayed_refs.flushing) { | |
815 | prepare_to_wait(&info->transaction_wait, &wait, | |
816 | TASK_UNINTERRUPTIBLE); | |
817 | mutex_unlock(&info->trans_mutex); | |
59bc5c75 | 818 | |
b7ec40d7 | 819 | schedule(); |
59bc5c75 | 820 | |
b7ec40d7 CM |
821 | mutex_lock(&info->trans_mutex); |
822 | finish_wait(&info->transaction_wait, &wait); | |
823 | } | |
824 | mutex_unlock(&info->trans_mutex); | |
825 | return 0; | |
826 | } | |
827 | ||
d352ac68 CM |
828 | /* |
829 | * Given a list of roots that need to be deleted, call btrfs_drop_snapshot on | |
830 | * all of them | |
831 | */ | |
5d4f98a2 | 832 | int btrfs_drop_dead_root(struct btrfs_root *root) |
0f7d52f4 | 833 | { |
0f7d52f4 | 834 | struct btrfs_trans_handle *trans; |
5d4f98a2 | 835 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
d3c2fdcf | 836 | unsigned long nr; |
5d4f98a2 | 837 | int ret; |
58176a96 | 838 | |
5d4f98a2 YZ |
839 | while (1) { |
840 | /* | |
841 | * we don't want to jump in and create a bunch of | |
842 | * delayed refs if the transaction is starting to close | |
843 | */ | |
844 | wait_transaction_pre_flush(tree_root->fs_info); | |
845 | trans = btrfs_start_transaction(tree_root, 1); | |
a2135011 | 846 | |
5d4f98a2 YZ |
847 | /* |
848 | * we've joined a transaction, make sure it isn't | |
849 | * closing right now | |
850 | */ | |
851 | if (trans->transaction->delayed_refs.flushing) { | |
852 | btrfs_end_transaction(trans, tree_root); | |
853 | continue; | |
9f3a7427 | 854 | } |
58176a96 | 855 | |
5d4f98a2 YZ |
856 | ret = btrfs_drop_snapshot(trans, root); |
857 | if (ret != -EAGAIN) | |
858 | break; | |
a2135011 | 859 | |
5d4f98a2 YZ |
860 | ret = btrfs_update_root(trans, tree_root, |
861 | &root->root_key, | |
862 | &root->root_item); | |
863 | if (ret) | |
54aa1f4d | 864 | break; |
bcc63abb | 865 | |
d3c2fdcf | 866 | nr = trans->blocks_used; |
0f7d52f4 CM |
867 | ret = btrfs_end_transaction(trans, tree_root); |
868 | BUG_ON(ret); | |
5eda7b5e | 869 | |
d3c2fdcf | 870 | btrfs_btree_balance_dirty(tree_root, nr); |
4dc11904 | 871 | cond_resched(); |
0f7d52f4 | 872 | } |
5d4f98a2 YZ |
873 | BUG_ON(ret); |
874 | ||
875 | ret = btrfs_del_root(trans, tree_root, &root->root_key); | |
876 | BUG_ON(ret); | |
877 | ||
878 | nr = trans->blocks_used; | |
879 | ret = btrfs_end_transaction(trans, tree_root); | |
880 | BUG_ON(ret); | |
881 | ||
882 | free_extent_buffer(root->node); | |
883 | free_extent_buffer(root->commit_root); | |
884 | kfree(root); | |
885 | ||
886 | btrfs_btree_balance_dirty(tree_root, nr); | |
54aa1f4d | 887 | return ret; |
0f7d52f4 | 888 | } |
2c47e605 | 889 | #endif |
0f7d52f4 | 890 | |
d352ac68 CM |
891 | /* |
892 | * new snapshots need to be created at a very specific time in the | |
893 | * transaction commit. This does the actual creation | |
894 | */ | |
80b6794d | 895 | static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, |
3063d29f CM |
896 | struct btrfs_fs_info *fs_info, |
897 | struct btrfs_pending_snapshot *pending) | |
898 | { | |
899 | struct btrfs_key key; | |
80b6794d | 900 | struct btrfs_root_item *new_root_item; |
3063d29f CM |
901 | struct btrfs_root *tree_root = fs_info->tree_root; |
902 | struct btrfs_root *root = pending->root; | |
6bdb72de SW |
903 | struct btrfs_root *parent_root; |
904 | struct inode *parent_inode; | |
6a912213 | 905 | struct dentry *parent; |
a22285a6 | 906 | struct dentry *dentry; |
3063d29f | 907 | struct extent_buffer *tmp; |
925baedd | 908 | struct extent_buffer *old; |
3063d29f | 909 | int ret; |
d68fc57b | 910 | u64 to_reserve = 0; |
6bdb72de | 911 | u64 index = 0; |
a22285a6 | 912 | u64 objectid; |
3063d29f | 913 | |
80b6794d CM |
914 | new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS); |
915 | if (!new_root_item) { | |
a22285a6 | 916 | pending->error = -ENOMEM; |
80b6794d CM |
917 | goto fail; |
918 | } | |
a22285a6 | 919 | |
3063d29f | 920 | ret = btrfs_find_free_objectid(trans, tree_root, 0, &objectid); |
a22285a6 YZ |
921 | if (ret) { |
922 | pending->error = ret; | |
3063d29f | 923 | goto fail; |
a22285a6 | 924 | } |
3063d29f | 925 | |
3fd0a558 | 926 | btrfs_reloc_pre_snapshot(trans, pending, &to_reserve); |
d68fc57b YZ |
927 | btrfs_orphan_pre_snapshot(trans, pending, &to_reserve); |
928 | ||
929 | if (to_reserve > 0) { | |
930 | ret = btrfs_block_rsv_add(trans, root, &pending->block_rsv, | |
8bb8ab2e | 931 | to_reserve); |
d68fc57b YZ |
932 | if (ret) { |
933 | pending->error = ret; | |
934 | goto fail; | |
935 | } | |
936 | } | |
937 | ||
3063d29f | 938 | key.objectid = objectid; |
a22285a6 YZ |
939 | key.offset = (u64)-1; |
940 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3063d29f | 941 | |
a22285a6 | 942 | trans->block_rsv = &pending->block_rsv; |
3de4586c | 943 | |
a22285a6 | 944 | dentry = pending->dentry; |
6a912213 JB |
945 | parent = dget_parent(dentry); |
946 | parent_inode = parent->d_inode; | |
a22285a6 | 947 | parent_root = BTRFS_I(parent_inode)->root; |
6bdb72de | 948 | record_root_in_trans(trans, parent_root); |
a22285a6 | 949 | |
3063d29f CM |
950 | /* |
951 | * insert the directory item | |
952 | */ | |
3de4586c | 953 | ret = btrfs_set_inode_index(parent_inode, &index); |
6bdb72de | 954 | BUG_ON(ret); |
0660b5af | 955 | ret = btrfs_insert_dir_item(trans, parent_root, |
a22285a6 YZ |
956 | dentry->d_name.name, dentry->d_name.len, |
957 | parent_inode->i_ino, &key, | |
958 | BTRFS_FT_DIR, index); | |
6bdb72de | 959 | BUG_ON(ret); |
0660b5af | 960 | |
a22285a6 YZ |
961 | btrfs_i_size_write(parent_inode, parent_inode->i_size + |
962 | dentry->d_name.len * 2); | |
52c26179 YZ |
963 | ret = btrfs_update_inode(trans, parent_root, parent_inode); |
964 | BUG_ON(ret); | |
965 | ||
6bdb72de SW |
966 | record_root_in_trans(trans, root); |
967 | btrfs_set_root_last_snapshot(&root->root_item, trans->transid); | |
968 | memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); | |
969 | ||
970 | old = btrfs_lock_root_node(root); | |
971 | btrfs_cow_block(trans, root, old, NULL, 0, &old); | |
972 | btrfs_set_lock_blocking(old); | |
973 | ||
974 | btrfs_copy_root(trans, root, old, &tmp, objectid); | |
975 | btrfs_tree_unlock(old); | |
976 | free_extent_buffer(old); | |
977 | ||
978 | btrfs_set_root_node(new_root_item, tmp); | |
a22285a6 YZ |
979 | /* record when the snapshot was created in key.offset */ |
980 | key.offset = trans->transid; | |
981 | ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); | |
6bdb72de SW |
982 | btrfs_tree_unlock(tmp); |
983 | free_extent_buffer(tmp); | |
a22285a6 | 984 | BUG_ON(ret); |
6bdb72de | 985 | |
a22285a6 YZ |
986 | /* |
987 | * insert root back/forward references | |
988 | */ | |
989 | ret = btrfs_add_root_ref(trans, tree_root, objectid, | |
0660b5af | 990 | parent_root->root_key.objectid, |
a22285a6 YZ |
991 | parent_inode->i_ino, index, |
992 | dentry->d_name.name, dentry->d_name.len); | |
0660b5af | 993 | BUG_ON(ret); |
6a912213 | 994 | dput(parent); |
0660b5af | 995 | |
a22285a6 YZ |
996 | key.offset = (u64)-1; |
997 | pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key); | |
998 | BUG_ON(IS_ERR(pending->snap)); | |
d68fc57b | 999 | |
3fd0a558 | 1000 | btrfs_reloc_post_snapshot(trans, pending); |
d68fc57b | 1001 | btrfs_orphan_post_snapshot(trans, pending); |
3063d29f | 1002 | fail: |
6bdb72de | 1003 | kfree(new_root_item); |
a22285a6 YZ |
1004 | btrfs_block_rsv_release(root, &pending->block_rsv, (u64)-1); |
1005 | return 0; | |
3063d29f CM |
1006 | } |
1007 | ||
d352ac68 CM |
1008 | /* |
1009 | * create all the snapshots we've scheduled for creation | |
1010 | */ | |
80b6794d CM |
1011 | static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans, |
1012 | struct btrfs_fs_info *fs_info) | |
3de4586c CM |
1013 | { |
1014 | struct btrfs_pending_snapshot *pending; | |
1015 | struct list_head *head = &trans->transaction->pending_snapshots; | |
3de4586c CM |
1016 | int ret; |
1017 | ||
c6e30871 | 1018 | list_for_each_entry(pending, head, list) { |
3de4586c CM |
1019 | ret = create_pending_snapshot(trans, fs_info, pending); |
1020 | BUG_ON(ret); | |
1021 | } | |
1022 | return 0; | |
1023 | } | |
1024 | ||
5d4f98a2 YZ |
1025 | static void update_super_roots(struct btrfs_root *root) |
1026 | { | |
1027 | struct btrfs_root_item *root_item; | |
1028 | struct btrfs_super_block *super; | |
1029 | ||
1030 | super = &root->fs_info->super_copy; | |
1031 | ||
1032 | root_item = &root->fs_info->chunk_root->root_item; | |
1033 | super->chunk_root = root_item->bytenr; | |
1034 | super->chunk_root_generation = root_item->generation; | |
1035 | super->chunk_root_level = root_item->level; | |
1036 | ||
1037 | root_item = &root->fs_info->tree_root->root_item; | |
1038 | super->root = root_item->bytenr; | |
1039 | super->generation = root_item->generation; | |
1040 | super->root_level = root_item->level; | |
0af3d00b JB |
1041 | if (super->cache_generation != 0 || btrfs_test_opt(root, SPACE_CACHE)) |
1042 | super->cache_generation = root_item->generation; | |
5d4f98a2 YZ |
1043 | } |
1044 | ||
f36f3042 CM |
1045 | int btrfs_transaction_in_commit(struct btrfs_fs_info *info) |
1046 | { | |
1047 | int ret = 0; | |
1048 | spin_lock(&info->new_trans_lock); | |
1049 | if (info->running_transaction) | |
1050 | ret = info->running_transaction->in_commit; | |
1051 | spin_unlock(&info->new_trans_lock); | |
1052 | return ret; | |
1053 | } | |
1054 | ||
8929ecfa YZ |
1055 | int btrfs_transaction_blocked(struct btrfs_fs_info *info) |
1056 | { | |
1057 | int ret = 0; | |
1058 | spin_lock(&info->new_trans_lock); | |
1059 | if (info->running_transaction) | |
1060 | ret = info->running_transaction->blocked; | |
1061 | spin_unlock(&info->new_trans_lock); | |
1062 | return ret; | |
1063 | } | |
1064 | ||
bb9c12c9 SW |
1065 | /* |
1066 | * wait for the current transaction commit to start and block subsequent | |
1067 | * transaction joins | |
1068 | */ | |
1069 | static void wait_current_trans_commit_start(struct btrfs_root *root, | |
1070 | struct btrfs_transaction *trans) | |
1071 | { | |
1072 | DEFINE_WAIT(wait); | |
1073 | ||
1074 | if (trans->in_commit) | |
1075 | return; | |
1076 | ||
1077 | while (1) { | |
1078 | prepare_to_wait(&root->fs_info->transaction_blocked_wait, &wait, | |
1079 | TASK_UNINTERRUPTIBLE); | |
1080 | if (trans->in_commit) { | |
1081 | finish_wait(&root->fs_info->transaction_blocked_wait, | |
1082 | &wait); | |
1083 | break; | |
1084 | } | |
1085 | mutex_unlock(&root->fs_info->trans_mutex); | |
1086 | schedule(); | |
1087 | mutex_lock(&root->fs_info->trans_mutex); | |
1088 | finish_wait(&root->fs_info->transaction_blocked_wait, &wait); | |
1089 | } | |
1090 | } | |
1091 | ||
1092 | /* | |
1093 | * wait for the current transaction to start and then become unblocked. | |
1094 | * caller holds ref. | |
1095 | */ | |
1096 | static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root, | |
1097 | struct btrfs_transaction *trans) | |
1098 | { | |
1099 | DEFINE_WAIT(wait); | |
1100 | ||
1101 | if (trans->commit_done || (trans->in_commit && !trans->blocked)) | |
1102 | return; | |
1103 | ||
1104 | while (1) { | |
1105 | prepare_to_wait(&root->fs_info->transaction_wait, &wait, | |
1106 | TASK_UNINTERRUPTIBLE); | |
1107 | if (trans->commit_done || | |
1108 | (trans->in_commit && !trans->blocked)) { | |
1109 | finish_wait(&root->fs_info->transaction_wait, | |
1110 | &wait); | |
1111 | break; | |
1112 | } | |
1113 | mutex_unlock(&root->fs_info->trans_mutex); | |
1114 | schedule(); | |
1115 | mutex_lock(&root->fs_info->trans_mutex); | |
1116 | finish_wait(&root->fs_info->transaction_wait, | |
1117 | &wait); | |
1118 | } | |
1119 | } | |
1120 | ||
1121 | /* | |
1122 | * commit transactions asynchronously. once btrfs_commit_transaction_async | |
1123 | * returns, any subsequent transaction will not be allowed to join. | |
1124 | */ | |
1125 | struct btrfs_async_commit { | |
1126 | struct btrfs_trans_handle *newtrans; | |
1127 | struct btrfs_root *root; | |
1128 | struct delayed_work work; | |
1129 | }; | |
1130 | ||
1131 | static void do_async_commit(struct work_struct *work) | |
1132 | { | |
1133 | struct btrfs_async_commit *ac = | |
1134 | container_of(work, struct btrfs_async_commit, work.work); | |
1135 | ||
1136 | btrfs_commit_transaction(ac->newtrans, ac->root); | |
1137 | kfree(ac); | |
1138 | } | |
1139 | ||
1140 | int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans, | |
1141 | struct btrfs_root *root, | |
1142 | int wait_for_unblock) | |
1143 | { | |
1144 | struct btrfs_async_commit *ac; | |
1145 | struct btrfs_transaction *cur_trans; | |
1146 | ||
1147 | ac = kmalloc(sizeof(*ac), GFP_NOFS); | |
1148 | BUG_ON(!ac); | |
1149 | ||
1150 | INIT_DELAYED_WORK(&ac->work, do_async_commit); | |
1151 | ac->root = root; | |
1152 | ac->newtrans = btrfs_join_transaction(root, 0); | |
1153 | ||
1154 | /* take transaction reference */ | |
1155 | mutex_lock(&root->fs_info->trans_mutex); | |
1156 | cur_trans = trans->transaction; | |
1157 | cur_trans->use_count++; | |
1158 | mutex_unlock(&root->fs_info->trans_mutex); | |
1159 | ||
1160 | btrfs_end_transaction(trans, root); | |
1161 | schedule_delayed_work(&ac->work, 0); | |
1162 | ||
1163 | /* wait for transaction to start and unblock */ | |
1164 | mutex_lock(&root->fs_info->trans_mutex); | |
1165 | if (wait_for_unblock) | |
1166 | wait_current_trans_commit_start_and_unblock(root, cur_trans); | |
1167 | else | |
1168 | wait_current_trans_commit_start(root, cur_trans); | |
1169 | put_transaction(cur_trans); | |
1170 | mutex_unlock(&root->fs_info->trans_mutex); | |
1171 | ||
1172 | return 0; | |
1173 | } | |
1174 | ||
1175 | /* | |
1176 | * btrfs_transaction state sequence: | |
1177 | * in_commit = 0, blocked = 0 (initial) | |
1178 | * in_commit = 1, blocked = 1 | |
1179 | * blocked = 0 | |
1180 | * commit_done = 1 | |
1181 | */ | |
79154b1b CM |
1182 | int btrfs_commit_transaction(struct btrfs_trans_handle *trans, |
1183 | struct btrfs_root *root) | |
1184 | { | |
15ee9bc7 | 1185 | unsigned long joined = 0; |
79154b1b | 1186 | struct btrfs_transaction *cur_trans; |
8fd17795 | 1187 | struct btrfs_transaction *prev_trans = NULL; |
79154b1b | 1188 | DEFINE_WAIT(wait); |
15ee9bc7 | 1189 | int ret; |
89573b9c CM |
1190 | int should_grow = 0; |
1191 | unsigned long now = get_seconds(); | |
dccae999 | 1192 | int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT); |
79154b1b | 1193 | |
5a3f23d5 CM |
1194 | btrfs_run_ordered_operations(root, 0); |
1195 | ||
56bec294 CM |
1196 | /* make a pass through all the delayed refs we have so far |
1197 | * any runnings procs may add more while we are here | |
1198 | */ | |
1199 | ret = btrfs_run_delayed_refs(trans, root, 0); | |
1200 | BUG_ON(ret); | |
1201 | ||
a22285a6 YZ |
1202 | btrfs_trans_release_metadata(trans, root); |
1203 | ||
b7ec40d7 | 1204 | cur_trans = trans->transaction; |
56bec294 CM |
1205 | /* |
1206 | * set the flushing flag so procs in this transaction have to | |
1207 | * start sending their work down. | |
1208 | */ | |
b7ec40d7 | 1209 | cur_trans->delayed_refs.flushing = 1; |
56bec294 | 1210 | |
c3e69d58 | 1211 | ret = btrfs_run_delayed_refs(trans, root, 0); |
56bec294 CM |
1212 | BUG_ON(ret); |
1213 | ||
79154b1b | 1214 | mutex_lock(&root->fs_info->trans_mutex); |
b7ec40d7 CM |
1215 | if (cur_trans->in_commit) { |
1216 | cur_trans->use_count++; | |
ccd467d6 | 1217 | mutex_unlock(&root->fs_info->trans_mutex); |
79154b1b | 1218 | btrfs_end_transaction(trans, root); |
ccd467d6 | 1219 | |
79154b1b CM |
1220 | ret = wait_for_commit(root, cur_trans); |
1221 | BUG_ON(ret); | |
15ee9bc7 JB |
1222 | |
1223 | mutex_lock(&root->fs_info->trans_mutex); | |
79154b1b | 1224 | put_transaction(cur_trans); |
15ee9bc7 JB |
1225 | mutex_unlock(&root->fs_info->trans_mutex); |
1226 | ||
79154b1b CM |
1227 | return 0; |
1228 | } | |
4313b399 | 1229 | |
2c90e5d6 | 1230 | trans->transaction->in_commit = 1; |
f9295749 | 1231 | trans->transaction->blocked = 1; |
bb9c12c9 SW |
1232 | wake_up(&root->fs_info->transaction_blocked_wait); |
1233 | ||
ccd467d6 CM |
1234 | if (cur_trans->list.prev != &root->fs_info->trans_list) { |
1235 | prev_trans = list_entry(cur_trans->list.prev, | |
1236 | struct btrfs_transaction, list); | |
1237 | if (!prev_trans->commit_done) { | |
1238 | prev_trans->use_count++; | |
ccd467d6 CM |
1239 | mutex_unlock(&root->fs_info->trans_mutex); |
1240 | ||
1241 | wait_for_commit(root, prev_trans); | |
ccd467d6 | 1242 | |
ccd467d6 | 1243 | mutex_lock(&root->fs_info->trans_mutex); |
15ee9bc7 | 1244 | put_transaction(prev_trans); |
ccd467d6 CM |
1245 | } |
1246 | } | |
15ee9bc7 | 1247 | |
89573b9c CM |
1248 | if (now < cur_trans->start_time || now - cur_trans->start_time < 1) |
1249 | should_grow = 1; | |
1250 | ||
15ee9bc7 | 1251 | do { |
7ea394f1 | 1252 | int snap_pending = 0; |
15ee9bc7 | 1253 | joined = cur_trans->num_joined; |
7ea394f1 YZ |
1254 | if (!list_empty(&trans->transaction->pending_snapshots)) |
1255 | snap_pending = 1; | |
1256 | ||
2c90e5d6 | 1257 | WARN_ON(cur_trans != trans->transaction); |
79154b1b | 1258 | mutex_unlock(&root->fs_info->trans_mutex); |
15ee9bc7 | 1259 | |
0bdb1db2 | 1260 | if (flush_on_commit || snap_pending) { |
24bbcf04 YZ |
1261 | btrfs_start_delalloc_inodes(root, 1); |
1262 | ret = btrfs_wait_ordered_extents(root, 0, 1); | |
ebecd3d9 | 1263 | BUG_ON(ret); |
7ea394f1 YZ |
1264 | } |
1265 | ||
5a3f23d5 CM |
1266 | /* |
1267 | * rename don't use btrfs_join_transaction, so, once we | |
1268 | * set the transaction to blocked above, we aren't going | |
1269 | * to get any new ordered operations. We can safely run | |
1270 | * it here and no for sure that nothing new will be added | |
1271 | * to the list | |
1272 | */ | |
1273 | btrfs_run_ordered_operations(root, 1); | |
1274 | ||
ed3b3d31 CM |
1275 | prepare_to_wait(&cur_trans->writer_wait, &wait, |
1276 | TASK_UNINTERRUPTIBLE); | |
1277 | ||
89573b9c | 1278 | smp_mb(); |
99d16cbc SW |
1279 | if (cur_trans->num_writers > 1) |
1280 | schedule_timeout(MAX_SCHEDULE_TIMEOUT); | |
1281 | else if (should_grow) | |
1282 | schedule_timeout(1); | |
15ee9bc7 | 1283 | |
79154b1b | 1284 | mutex_lock(&root->fs_info->trans_mutex); |
15ee9bc7 JB |
1285 | finish_wait(&cur_trans->writer_wait, &wait); |
1286 | } while (cur_trans->num_writers > 1 || | |
89573b9c | 1287 | (should_grow && cur_trans->num_joined != joined)); |
15ee9bc7 | 1288 | |
3063d29f CM |
1289 | ret = create_pending_snapshots(trans, root->fs_info); |
1290 | BUG_ON(ret); | |
1291 | ||
56bec294 CM |
1292 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
1293 | BUG_ON(ret); | |
1294 | ||
2c90e5d6 | 1295 | WARN_ON(cur_trans != trans->transaction); |
dc17ff8f | 1296 | |
e02119d5 CM |
1297 | /* btrfs_commit_tree_roots is responsible for getting the |
1298 | * various roots consistent with each other. Every pointer | |
1299 | * in the tree of tree roots has to point to the most up to date | |
1300 | * root for every subvolume and other tree. So, we have to keep | |
1301 | * the tree logging code from jumping in and changing any | |
1302 | * of the trees. | |
1303 | * | |
1304 | * At this point in the commit, there can't be any tree-log | |
1305 | * writers, but a little lower down we drop the trans mutex | |
1306 | * and let new people in. By holding the tree_log_mutex | |
1307 | * from now until after the super is written, we avoid races | |
1308 | * with the tree-log code. | |
1309 | */ | |
1310 | mutex_lock(&root->fs_info->tree_log_mutex); | |
1311 | ||
5d4f98a2 | 1312 | ret = commit_fs_roots(trans, root); |
54aa1f4d CM |
1313 | BUG_ON(ret); |
1314 | ||
5d4f98a2 | 1315 | /* commit_fs_roots gets rid of all the tree log roots, it is now |
e02119d5 CM |
1316 | * safe to free the root of tree log roots |
1317 | */ | |
1318 | btrfs_free_log_root_tree(trans, root->fs_info); | |
1319 | ||
5d4f98a2 | 1320 | ret = commit_cowonly_roots(trans, root); |
79154b1b | 1321 | BUG_ON(ret); |
54aa1f4d | 1322 | |
11833d66 YZ |
1323 | btrfs_prepare_extent_commit(trans, root); |
1324 | ||
78fae27e | 1325 | cur_trans = root->fs_info->running_transaction; |
cee36a03 | 1326 | spin_lock(&root->fs_info->new_trans_lock); |
78fae27e | 1327 | root->fs_info->running_transaction = NULL; |
cee36a03 | 1328 | spin_unlock(&root->fs_info->new_trans_lock); |
5d4f98a2 YZ |
1329 | |
1330 | btrfs_set_root_node(&root->fs_info->tree_root->root_item, | |
1331 | root->fs_info->tree_root->node); | |
817d52f8 | 1332 | switch_commit_root(root->fs_info->tree_root); |
5d4f98a2 YZ |
1333 | |
1334 | btrfs_set_root_node(&root->fs_info->chunk_root->root_item, | |
1335 | root->fs_info->chunk_root->node); | |
817d52f8 | 1336 | switch_commit_root(root->fs_info->chunk_root); |
5d4f98a2 YZ |
1337 | |
1338 | update_super_roots(root); | |
e02119d5 CM |
1339 | |
1340 | if (!root->fs_info->log_root_recovering) { | |
1341 | btrfs_set_super_log_root(&root->fs_info->super_copy, 0); | |
1342 | btrfs_set_super_log_root_level(&root->fs_info->super_copy, 0); | |
1343 | } | |
1344 | ||
a061fc8d CM |
1345 | memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy, |
1346 | sizeof(root->fs_info->super_copy)); | |
ccd467d6 | 1347 | |
f9295749 | 1348 | trans->transaction->blocked = 0; |
b7ec40d7 | 1349 | |
f9295749 | 1350 | wake_up(&root->fs_info->transaction_wait); |
e6dcd2dc | 1351 | |
78fae27e | 1352 | mutex_unlock(&root->fs_info->trans_mutex); |
79154b1b CM |
1353 | ret = btrfs_write_and_wait_transaction(trans, root); |
1354 | BUG_ON(ret); | |
a512bbf8 | 1355 | write_ctree_super(trans, root, 0); |
4313b399 | 1356 | |
e02119d5 CM |
1357 | /* |
1358 | * the super is written, we can safely allow the tree-loggers | |
1359 | * to go about their business | |
1360 | */ | |
1361 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
1362 | ||
11833d66 | 1363 | btrfs_finish_extent_commit(trans, root); |
4313b399 | 1364 | |
1a40e23b ZY |
1365 | mutex_lock(&root->fs_info->trans_mutex); |
1366 | ||
2c90e5d6 | 1367 | cur_trans->commit_done = 1; |
b7ec40d7 | 1368 | |
15ee9bc7 | 1369 | root->fs_info->last_trans_committed = cur_trans->transid; |
817d52f8 | 1370 | |
2c90e5d6 | 1371 | wake_up(&cur_trans->commit_wait); |
3de4586c | 1372 | |
78fae27e | 1373 | put_transaction(cur_trans); |
79154b1b | 1374 | put_transaction(cur_trans); |
58176a96 | 1375 | |
78fae27e | 1376 | mutex_unlock(&root->fs_info->trans_mutex); |
3de4586c | 1377 | |
9ed74f2d JB |
1378 | if (current->journal_info == trans) |
1379 | current->journal_info = NULL; | |
1380 | ||
2c90e5d6 | 1381 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
24bbcf04 YZ |
1382 | |
1383 | if (current != root->fs_info->transaction_kthread) | |
1384 | btrfs_run_delayed_iputs(root); | |
1385 | ||
79154b1b CM |
1386 | return ret; |
1387 | } | |
1388 | ||
d352ac68 CM |
1389 | /* |
1390 | * interface function to delete all the snapshots we have scheduled for deletion | |
1391 | */ | |
e9d0b13b CM |
1392 | int btrfs_clean_old_snapshots(struct btrfs_root *root) |
1393 | { | |
5d4f98a2 YZ |
1394 | LIST_HEAD(list); |
1395 | struct btrfs_fs_info *fs_info = root->fs_info; | |
1396 | ||
1397 | mutex_lock(&fs_info->trans_mutex); | |
1398 | list_splice_init(&fs_info->dead_roots, &list); | |
1399 | mutex_unlock(&fs_info->trans_mutex); | |
e9d0b13b | 1400 | |
5d4f98a2 YZ |
1401 | while (!list_empty(&list)) { |
1402 | root = list_entry(list.next, struct btrfs_root, root_list); | |
76dda93c YZ |
1403 | list_del(&root->root_list); |
1404 | ||
1405 | if (btrfs_header_backref_rev(root->node) < | |
1406 | BTRFS_MIXED_BACKREF_REV) | |
3fd0a558 | 1407 | btrfs_drop_snapshot(root, NULL, 0); |
76dda93c | 1408 | else |
3fd0a558 | 1409 | btrfs_drop_snapshot(root, NULL, 1); |
e9d0b13b CM |
1410 | } |
1411 | return 0; | |
1412 | } |