]>
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; |
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 CM |
197 | ret = join_transaction(root); |
198 | BUG_ON(ret); | |
0f7d52f4 | 199 | |
a22285a6 YZ |
200 | cur_trans = root->fs_info->running_transaction; |
201 | cur_trans->use_count++; | |
0af3d00b JB |
202 | if (type != TRANS_JOIN_NOLOCK) |
203 | mutex_unlock(&root->fs_info->trans_mutex); | |
a22285a6 YZ |
204 | |
205 | h->transid = cur_trans->transid; | |
206 | h->transaction = cur_trans; | |
79154b1b | 207 | h->blocks_used = 0; |
d2fb3437 | 208 | h->block_group = 0; |
a22285a6 | 209 | h->bytes_reserved = 0; |
56bec294 | 210 | h->delayed_ref_updates = 0; |
f0486c68 | 211 | h->block_rsv = NULL; |
b7ec40d7 | 212 | |
a22285a6 YZ |
213 | smp_mb(); |
214 | if (cur_trans->blocked && may_wait_transaction(root, type)) { | |
215 | btrfs_commit_transaction(h, root); | |
216 | goto again; | |
217 | } | |
218 | ||
219 | if (num_items > 0) { | |
8bb8ab2e | 220 | ret = btrfs_trans_reserve_metadata(h, root, num_items); |
a22285a6 YZ |
221 | if (ret == -EAGAIN) { |
222 | btrfs_commit_transaction(h, root); | |
223 | goto again; | |
224 | } | |
225 | if (ret < 0) { | |
226 | btrfs_end_transaction(h, root); | |
227 | return ERR_PTR(ret); | |
228 | } | |
229 | } | |
9ed74f2d | 230 | |
0af3d00b JB |
231 | if (type != TRANS_JOIN_NOLOCK) |
232 | mutex_lock(&root->fs_info->trans_mutex); | |
5d4f98a2 | 233 | record_root_in_trans(h, root); |
0af3d00b JB |
234 | if (type != TRANS_JOIN_NOLOCK) |
235 | mutex_unlock(&root->fs_info->trans_mutex); | |
a22285a6 YZ |
236 | |
237 | if (!current->journal_info && type != TRANS_USERSPACE) | |
238 | current->journal_info = h; | |
79154b1b CM |
239 | return h; |
240 | } | |
241 | ||
f9295749 | 242 | struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, |
a22285a6 | 243 | int num_items) |
f9295749 | 244 | { |
a22285a6 | 245 | return start_transaction(root, num_items, TRANS_START); |
f9295749 CM |
246 | } |
247 | struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root, | |
248 | int num_blocks) | |
249 | { | |
a22285a6 | 250 | return start_transaction(root, 0, TRANS_JOIN); |
f9295749 CM |
251 | } |
252 | ||
0af3d00b JB |
253 | struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root, |
254 | int num_blocks) | |
255 | { | |
256 | return start_transaction(root, 0, TRANS_JOIN_NOLOCK); | |
257 | } | |
258 | ||
9ca9ee09 SW |
259 | struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *r, |
260 | int num_blocks) | |
261 | { | |
a22285a6 | 262 | return start_transaction(r, 0, TRANS_USERSPACE); |
9ca9ee09 SW |
263 | } |
264 | ||
d352ac68 | 265 | /* wait for a transaction commit to be fully complete */ |
89ce8a63 CM |
266 | static noinline int wait_for_commit(struct btrfs_root *root, |
267 | struct btrfs_transaction *commit) | |
268 | { | |
269 | DEFINE_WAIT(wait); | |
270 | mutex_lock(&root->fs_info->trans_mutex); | |
d397712b | 271 | while (!commit->commit_done) { |
89ce8a63 CM |
272 | prepare_to_wait(&commit->commit_wait, &wait, |
273 | TASK_UNINTERRUPTIBLE); | |
274 | if (commit->commit_done) | |
275 | break; | |
276 | mutex_unlock(&root->fs_info->trans_mutex); | |
277 | schedule(); | |
278 | mutex_lock(&root->fs_info->trans_mutex); | |
279 | } | |
280 | mutex_unlock(&root->fs_info->trans_mutex); | |
281 | finish_wait(&commit->commit_wait, &wait); | |
282 | return 0; | |
283 | } | |
284 | ||
46204592 SW |
285 | int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid) |
286 | { | |
287 | struct btrfs_transaction *cur_trans = NULL, *t; | |
288 | int ret; | |
289 | ||
290 | mutex_lock(&root->fs_info->trans_mutex); | |
291 | ||
292 | ret = 0; | |
293 | if (transid) { | |
294 | if (transid <= root->fs_info->last_trans_committed) | |
295 | goto out_unlock; | |
296 | ||
297 | /* find specified transaction */ | |
298 | list_for_each_entry(t, &root->fs_info->trans_list, list) { | |
299 | if (t->transid == transid) { | |
300 | cur_trans = t; | |
301 | break; | |
302 | } | |
303 | if (t->transid > transid) | |
304 | break; | |
305 | } | |
306 | ret = -EINVAL; | |
307 | if (!cur_trans) | |
308 | goto out_unlock; /* bad transid */ | |
309 | } else { | |
310 | /* find newest transaction that is committing | committed */ | |
311 | list_for_each_entry_reverse(t, &root->fs_info->trans_list, | |
312 | list) { | |
313 | if (t->in_commit) { | |
314 | if (t->commit_done) | |
315 | goto out_unlock; | |
316 | cur_trans = t; | |
317 | break; | |
318 | } | |
319 | } | |
320 | if (!cur_trans) | |
321 | goto out_unlock; /* nothing committing|committed */ | |
322 | } | |
323 | ||
324 | cur_trans->use_count++; | |
325 | mutex_unlock(&root->fs_info->trans_mutex); | |
326 | ||
327 | wait_for_commit(root, cur_trans); | |
328 | ||
329 | mutex_lock(&root->fs_info->trans_mutex); | |
330 | put_transaction(cur_trans); | |
331 | ret = 0; | |
332 | out_unlock: | |
333 | mutex_unlock(&root->fs_info->trans_mutex); | |
334 | return ret; | |
335 | } | |
336 | ||
5d4f98a2 | 337 | #if 0 |
d352ac68 | 338 | /* |
d397712b CM |
339 | * rate limit against the drop_snapshot code. This helps to slow down new |
340 | * operations if the drop_snapshot code isn't able to keep up. | |
d352ac68 | 341 | */ |
37d1aeee | 342 | static void throttle_on_drops(struct btrfs_root *root) |
ab78c84d CM |
343 | { |
344 | struct btrfs_fs_info *info = root->fs_info; | |
2dd3e67b | 345 | int harder_count = 0; |
ab78c84d | 346 | |
2dd3e67b | 347 | harder: |
ab78c84d CM |
348 | if (atomic_read(&info->throttles)) { |
349 | DEFINE_WAIT(wait); | |
350 | int thr; | |
ab78c84d CM |
351 | thr = atomic_read(&info->throttle_gen); |
352 | ||
353 | do { | |
354 | prepare_to_wait(&info->transaction_throttle, | |
355 | &wait, TASK_UNINTERRUPTIBLE); | |
356 | if (!atomic_read(&info->throttles)) { | |
357 | finish_wait(&info->transaction_throttle, &wait); | |
358 | break; | |
359 | } | |
360 | schedule(); | |
361 | finish_wait(&info->transaction_throttle, &wait); | |
362 | } while (thr == atomic_read(&info->throttle_gen)); | |
2dd3e67b CM |
363 | harder_count++; |
364 | ||
365 | if (root->fs_info->total_ref_cache_size > 1 * 1024 * 1024 && | |
366 | harder_count < 2) | |
367 | goto harder; | |
368 | ||
369 | if (root->fs_info->total_ref_cache_size > 5 * 1024 * 1024 && | |
370 | harder_count < 10) | |
371 | goto harder; | |
372 | ||
373 | if (root->fs_info->total_ref_cache_size > 10 * 1024 * 1024 && | |
374 | harder_count < 20) | |
375 | goto harder; | |
ab78c84d CM |
376 | } |
377 | } | |
5d4f98a2 | 378 | #endif |
ab78c84d | 379 | |
37d1aeee CM |
380 | void btrfs_throttle(struct btrfs_root *root) |
381 | { | |
382 | mutex_lock(&root->fs_info->trans_mutex); | |
9ca9ee09 SW |
383 | if (!root->fs_info->open_ioctl_trans) |
384 | wait_current_trans(root); | |
37d1aeee | 385 | mutex_unlock(&root->fs_info->trans_mutex); |
37d1aeee CM |
386 | } |
387 | ||
8929ecfa YZ |
388 | static int should_end_transaction(struct btrfs_trans_handle *trans, |
389 | struct btrfs_root *root) | |
390 | { | |
391 | int ret; | |
392 | ret = btrfs_block_rsv_check(trans, root, | |
393 | &root->fs_info->global_block_rsv, 0, 5); | |
394 | return ret ? 1 : 0; | |
395 | } | |
396 | ||
397 | int btrfs_should_end_transaction(struct btrfs_trans_handle *trans, | |
398 | struct btrfs_root *root) | |
399 | { | |
400 | struct btrfs_transaction *cur_trans = trans->transaction; | |
401 | int updates; | |
402 | ||
403 | if (cur_trans->blocked || cur_trans->delayed_refs.flushing) | |
404 | return 1; | |
405 | ||
406 | updates = trans->delayed_ref_updates; | |
407 | trans->delayed_ref_updates = 0; | |
408 | if (updates) | |
409 | btrfs_run_delayed_refs(trans, root, updates); | |
410 | ||
411 | return should_end_transaction(trans, root); | |
412 | } | |
413 | ||
89ce8a63 | 414 | static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, |
0af3d00b | 415 | struct btrfs_root *root, int throttle, int lock) |
79154b1b | 416 | { |
8929ecfa | 417 | struct btrfs_transaction *cur_trans = trans->transaction; |
ab78c84d | 418 | struct btrfs_fs_info *info = root->fs_info; |
c3e69d58 CM |
419 | int count = 0; |
420 | ||
421 | while (count < 4) { | |
422 | unsigned long cur = trans->delayed_ref_updates; | |
423 | trans->delayed_ref_updates = 0; | |
424 | if (cur && | |
425 | trans->transaction->delayed_refs.num_heads_ready > 64) { | |
426 | trans->delayed_ref_updates = 0; | |
b7ec40d7 CM |
427 | |
428 | /* | |
429 | * do a full flush if the transaction is trying | |
430 | * to close | |
431 | */ | |
432 | if (trans->transaction->delayed_refs.flushing) | |
433 | cur = 0; | |
c3e69d58 CM |
434 | btrfs_run_delayed_refs(trans, root, cur); |
435 | } else { | |
436 | break; | |
437 | } | |
438 | count++; | |
56bec294 CM |
439 | } |
440 | ||
a22285a6 YZ |
441 | btrfs_trans_release_metadata(trans, root); |
442 | ||
0af3d00b | 443 | if (lock && !root->fs_info->open_ioctl_trans && |
8929ecfa YZ |
444 | should_end_transaction(trans, root)) |
445 | trans->transaction->blocked = 1; | |
446 | ||
0af3d00b | 447 | if (lock && cur_trans->blocked && !cur_trans->in_commit) { |
8929ecfa YZ |
448 | if (throttle) |
449 | return btrfs_commit_transaction(trans, root); | |
450 | else | |
451 | wake_up_process(info->transaction_kthread); | |
452 | } | |
453 | ||
0af3d00b JB |
454 | if (lock) |
455 | mutex_lock(&info->trans_mutex); | |
8929ecfa | 456 | WARN_ON(cur_trans != info->running_transaction); |
d5719762 | 457 | WARN_ON(cur_trans->num_writers < 1); |
ccd467d6 | 458 | cur_trans->num_writers--; |
89ce8a63 | 459 | |
99d16cbc | 460 | smp_mb(); |
79154b1b CM |
461 | if (waitqueue_active(&cur_trans->writer_wait)) |
462 | wake_up(&cur_trans->writer_wait); | |
79154b1b | 463 | put_transaction(cur_trans); |
0af3d00b JB |
464 | if (lock) |
465 | mutex_unlock(&info->trans_mutex); | |
9ed74f2d JB |
466 | |
467 | if (current->journal_info == trans) | |
468 | current->journal_info = NULL; | |
d6025579 | 469 | memset(trans, 0, sizeof(*trans)); |
2c90e5d6 | 470 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
ab78c84d | 471 | |
24bbcf04 YZ |
472 | if (throttle) |
473 | btrfs_run_delayed_iputs(root); | |
474 | ||
79154b1b CM |
475 | return 0; |
476 | } | |
477 | ||
89ce8a63 CM |
478 | int btrfs_end_transaction(struct btrfs_trans_handle *trans, |
479 | struct btrfs_root *root) | |
480 | { | |
0af3d00b | 481 | return __btrfs_end_transaction(trans, root, 0, 1); |
89ce8a63 CM |
482 | } |
483 | ||
484 | int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans, | |
485 | struct btrfs_root *root) | |
486 | { | |
0af3d00b JB |
487 | return __btrfs_end_transaction(trans, root, 1, 1); |
488 | } | |
489 | ||
490 | int btrfs_end_transaction_nolock(struct btrfs_trans_handle *trans, | |
491 | struct btrfs_root *root) | |
492 | { | |
493 | return __btrfs_end_transaction(trans, root, 0, 0); | |
89ce8a63 CM |
494 | } |
495 | ||
d352ac68 CM |
496 | /* |
497 | * when btree blocks are allocated, they have some corresponding bits set for | |
498 | * them in one of two extent_io trees. This is used to make sure all of | |
690587d1 | 499 | * those extents are sent to disk but does not wait on them |
d352ac68 | 500 | */ |
690587d1 | 501 | int btrfs_write_marked_extents(struct btrfs_root *root, |
8cef4e16 | 502 | struct extent_io_tree *dirty_pages, int mark) |
79154b1b | 503 | { |
7c4452b9 | 504 | int ret; |
777e6bd7 | 505 | int err = 0; |
7c4452b9 CM |
506 | int werr = 0; |
507 | struct page *page; | |
7c4452b9 | 508 | struct inode *btree_inode = root->fs_info->btree_inode; |
777e6bd7 | 509 | u64 start = 0; |
5f39d397 CM |
510 | u64 end; |
511 | unsigned long index; | |
7c4452b9 | 512 | |
d397712b | 513 | while (1) { |
777e6bd7 | 514 | ret = find_first_extent_bit(dirty_pages, start, &start, &end, |
8cef4e16 | 515 | mark); |
5f39d397 | 516 | if (ret) |
7c4452b9 | 517 | break; |
d397712b | 518 | while (start <= end) { |
777e6bd7 CM |
519 | cond_resched(); |
520 | ||
5f39d397 | 521 | index = start >> PAGE_CACHE_SHIFT; |
35ebb934 | 522 | start = (u64)(index + 1) << PAGE_CACHE_SHIFT; |
4bef0848 | 523 | page = find_get_page(btree_inode->i_mapping, index); |
7c4452b9 CM |
524 | if (!page) |
525 | continue; | |
4bef0848 CM |
526 | |
527 | btree_lock_page_hook(page); | |
528 | if (!page->mapping) { | |
529 | unlock_page(page); | |
530 | page_cache_release(page); | |
531 | continue; | |
532 | } | |
533 | ||
6702ed49 CM |
534 | if (PageWriteback(page)) { |
535 | if (PageDirty(page)) | |
536 | wait_on_page_writeback(page); | |
537 | else { | |
538 | unlock_page(page); | |
539 | page_cache_release(page); | |
540 | continue; | |
541 | } | |
542 | } | |
7c4452b9 CM |
543 | err = write_one_page(page, 0); |
544 | if (err) | |
545 | werr = err; | |
546 | page_cache_release(page); | |
547 | } | |
548 | } | |
690587d1 CM |
549 | if (err) |
550 | werr = err; | |
551 | return werr; | |
552 | } | |
553 | ||
554 | /* | |
555 | * when btree blocks are allocated, they have some corresponding bits set for | |
556 | * them in one of two extent_io trees. This is used to make sure all of | |
557 | * those extents are on disk for transaction or log commit. We wait | |
558 | * on all the pages and clear them from the dirty pages state tree | |
559 | */ | |
560 | int btrfs_wait_marked_extents(struct btrfs_root *root, | |
8cef4e16 | 561 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 CM |
562 | { |
563 | int ret; | |
564 | int err = 0; | |
565 | int werr = 0; | |
566 | struct page *page; | |
567 | struct inode *btree_inode = root->fs_info->btree_inode; | |
568 | u64 start = 0; | |
569 | u64 end; | |
570 | unsigned long index; | |
571 | ||
d397712b | 572 | while (1) { |
8cef4e16 YZ |
573 | ret = find_first_extent_bit(dirty_pages, start, &start, &end, |
574 | mark); | |
777e6bd7 CM |
575 | if (ret) |
576 | break; | |
577 | ||
8cef4e16 | 578 | clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS); |
d397712b | 579 | while (start <= end) { |
777e6bd7 CM |
580 | index = start >> PAGE_CACHE_SHIFT; |
581 | start = (u64)(index + 1) << PAGE_CACHE_SHIFT; | |
582 | page = find_get_page(btree_inode->i_mapping, index); | |
583 | if (!page) | |
584 | continue; | |
585 | if (PageDirty(page)) { | |
4bef0848 CM |
586 | btree_lock_page_hook(page); |
587 | wait_on_page_writeback(page); | |
777e6bd7 CM |
588 | err = write_one_page(page, 0); |
589 | if (err) | |
590 | werr = err; | |
591 | } | |
105d931d | 592 | wait_on_page_writeback(page); |
777e6bd7 CM |
593 | page_cache_release(page); |
594 | cond_resched(); | |
595 | } | |
596 | } | |
7c4452b9 CM |
597 | if (err) |
598 | werr = err; | |
599 | return werr; | |
79154b1b CM |
600 | } |
601 | ||
690587d1 CM |
602 | /* |
603 | * when btree blocks are allocated, they have some corresponding bits set for | |
604 | * them in one of two extent_io trees. This is used to make sure all of | |
605 | * those extents are on disk for transaction or log commit | |
606 | */ | |
607 | int btrfs_write_and_wait_marked_extents(struct btrfs_root *root, | |
8cef4e16 | 608 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 CM |
609 | { |
610 | int ret; | |
611 | int ret2; | |
612 | ||
8cef4e16 YZ |
613 | ret = btrfs_write_marked_extents(root, dirty_pages, mark); |
614 | ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark); | |
690587d1 CM |
615 | return ret || ret2; |
616 | } | |
617 | ||
d0c803c4 CM |
618 | int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans, |
619 | struct btrfs_root *root) | |
620 | { | |
621 | if (!trans || !trans->transaction) { | |
622 | struct inode *btree_inode; | |
623 | btree_inode = root->fs_info->btree_inode; | |
624 | return filemap_write_and_wait(btree_inode->i_mapping); | |
625 | } | |
626 | return btrfs_write_and_wait_marked_extents(root, | |
8cef4e16 YZ |
627 | &trans->transaction->dirty_pages, |
628 | EXTENT_DIRTY); | |
d0c803c4 CM |
629 | } |
630 | ||
d352ac68 CM |
631 | /* |
632 | * this is used to update the root pointer in the tree of tree roots. | |
633 | * | |
634 | * But, in the case of the extent allocation tree, updating the root | |
635 | * pointer may allocate blocks which may change the root of the extent | |
636 | * allocation tree. | |
637 | * | |
638 | * So, this loops and repeats and makes sure the cowonly root didn't | |
639 | * change while the root pointer was being updated in the metadata. | |
640 | */ | |
0b86a832 CM |
641 | static int update_cowonly_root(struct btrfs_trans_handle *trans, |
642 | struct btrfs_root *root) | |
79154b1b CM |
643 | { |
644 | int ret; | |
0b86a832 | 645 | u64 old_root_bytenr; |
86b9f2ec | 646 | u64 old_root_used; |
0b86a832 | 647 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
79154b1b | 648 | |
86b9f2ec | 649 | old_root_used = btrfs_root_used(&root->root_item); |
0b86a832 | 650 | btrfs_write_dirty_block_groups(trans, root); |
56bec294 | 651 | |
d397712b | 652 | while (1) { |
0b86a832 | 653 | old_root_bytenr = btrfs_root_bytenr(&root->root_item); |
86b9f2ec YZ |
654 | if (old_root_bytenr == root->node->start && |
655 | old_root_used == btrfs_root_used(&root->root_item)) | |
79154b1b | 656 | break; |
87ef2bb4 | 657 | |
5d4f98a2 | 658 | btrfs_set_root_node(&root->root_item, root->node); |
79154b1b | 659 | ret = btrfs_update_root(trans, tree_root, |
0b86a832 CM |
660 | &root->root_key, |
661 | &root->root_item); | |
79154b1b | 662 | BUG_ON(ret); |
56bec294 | 663 | |
86b9f2ec | 664 | old_root_used = btrfs_root_used(&root->root_item); |
4a8c9a62 | 665 | ret = btrfs_write_dirty_block_groups(trans, root); |
56bec294 | 666 | BUG_ON(ret); |
0b86a832 | 667 | } |
276e680d YZ |
668 | |
669 | if (root != root->fs_info->extent_root) | |
670 | switch_commit_root(root); | |
671 | ||
0b86a832 CM |
672 | return 0; |
673 | } | |
674 | ||
d352ac68 CM |
675 | /* |
676 | * update all the cowonly tree roots on disk | |
677 | */ | |
5d4f98a2 YZ |
678 | static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans, |
679 | struct btrfs_root *root) | |
0b86a832 CM |
680 | { |
681 | struct btrfs_fs_info *fs_info = root->fs_info; | |
682 | struct list_head *next; | |
84234f3a | 683 | struct extent_buffer *eb; |
56bec294 | 684 | int ret; |
84234f3a | 685 | |
56bec294 CM |
686 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
687 | BUG_ON(ret); | |
87ef2bb4 | 688 | |
84234f3a | 689 | eb = btrfs_lock_root_node(fs_info->tree_root); |
9fa8cfe7 | 690 | btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 0, &eb); |
84234f3a YZ |
691 | btrfs_tree_unlock(eb); |
692 | free_extent_buffer(eb); | |
0b86a832 | 693 | |
56bec294 CM |
694 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
695 | BUG_ON(ret); | |
87ef2bb4 | 696 | |
d397712b | 697 | while (!list_empty(&fs_info->dirty_cowonly_roots)) { |
0b86a832 CM |
698 | next = fs_info->dirty_cowonly_roots.next; |
699 | list_del_init(next); | |
700 | root = list_entry(next, struct btrfs_root, dirty_list); | |
87ef2bb4 | 701 | |
0b86a832 | 702 | update_cowonly_root(trans, root); |
79154b1b | 703 | } |
276e680d YZ |
704 | |
705 | down_write(&fs_info->extent_commit_sem); | |
706 | switch_commit_root(fs_info->extent_root); | |
707 | up_write(&fs_info->extent_commit_sem); | |
708 | ||
79154b1b CM |
709 | return 0; |
710 | } | |
711 | ||
d352ac68 CM |
712 | /* |
713 | * dead roots are old snapshots that need to be deleted. This allocates | |
714 | * a dirty root struct and adds it into the list of dead roots that need to | |
715 | * be deleted | |
716 | */ | |
5d4f98a2 | 717 | int btrfs_add_dead_root(struct btrfs_root *root) |
5eda7b5e | 718 | { |
b48652c1 | 719 | mutex_lock(&root->fs_info->trans_mutex); |
5d4f98a2 | 720 | list_add(&root->root_list, &root->fs_info->dead_roots); |
b48652c1 | 721 | mutex_unlock(&root->fs_info->trans_mutex); |
5eda7b5e CM |
722 | return 0; |
723 | } | |
724 | ||
d352ac68 | 725 | /* |
5d4f98a2 | 726 | * update all the cowonly tree roots on disk |
d352ac68 | 727 | */ |
5d4f98a2 YZ |
728 | static noinline int commit_fs_roots(struct btrfs_trans_handle *trans, |
729 | struct btrfs_root *root) | |
0f7d52f4 | 730 | { |
0f7d52f4 | 731 | struct btrfs_root *gang[8]; |
5d4f98a2 | 732 | struct btrfs_fs_info *fs_info = root->fs_info; |
0f7d52f4 CM |
733 | int i; |
734 | int ret; | |
54aa1f4d CM |
735 | int err = 0; |
736 | ||
d397712b | 737 | while (1) { |
5d4f98a2 YZ |
738 | ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, |
739 | (void **)gang, 0, | |
0f7d52f4 CM |
740 | ARRAY_SIZE(gang), |
741 | BTRFS_ROOT_TRANS_TAG); | |
742 | if (ret == 0) | |
743 | break; | |
744 | for (i = 0; i < ret; i++) { | |
745 | root = gang[i]; | |
5d4f98a2 YZ |
746 | radix_tree_tag_clear(&fs_info->fs_roots_radix, |
747 | (unsigned long)root->root_key.objectid, | |
748 | BTRFS_ROOT_TRANS_TAG); | |
31153d81 | 749 | |
e02119d5 | 750 | btrfs_free_log(trans, root); |
5d4f98a2 | 751 | btrfs_update_reloc_root(trans, root); |
d68fc57b | 752 | btrfs_orphan_commit_root(trans, root); |
bcc63abb | 753 | |
978d910d | 754 | if (root->commit_root != root->node) { |
817d52f8 | 755 | switch_commit_root(root); |
978d910d YZ |
756 | btrfs_set_root_node(&root->root_item, |
757 | root->node); | |
758 | } | |
5d4f98a2 | 759 | |
5d4f98a2 | 760 | err = btrfs_update_root(trans, fs_info->tree_root, |
0f7d52f4 CM |
761 | &root->root_key, |
762 | &root->root_item); | |
54aa1f4d CM |
763 | if (err) |
764 | break; | |
0f7d52f4 CM |
765 | } |
766 | } | |
54aa1f4d | 767 | return err; |
0f7d52f4 CM |
768 | } |
769 | ||
d352ac68 CM |
770 | /* |
771 | * defrag a given btree. If cacheonly == 1, this won't read from the disk, | |
772 | * otherwise every leaf in the btree is read and defragged. | |
773 | */ | |
e9d0b13b CM |
774 | int btrfs_defrag_root(struct btrfs_root *root, int cacheonly) |
775 | { | |
776 | struct btrfs_fs_info *info = root->fs_info; | |
e9d0b13b | 777 | struct btrfs_trans_handle *trans; |
8929ecfa | 778 | int ret; |
d3c2fdcf | 779 | unsigned long nr; |
e9d0b13b | 780 | |
8929ecfa | 781 | if (xchg(&root->defrag_running, 1)) |
e9d0b13b | 782 | return 0; |
8929ecfa | 783 | |
6b80053d | 784 | while (1) { |
8929ecfa YZ |
785 | trans = btrfs_start_transaction(root, 0); |
786 | if (IS_ERR(trans)) | |
787 | return PTR_ERR(trans); | |
788 | ||
e9d0b13b | 789 | ret = btrfs_defrag_leaves(trans, root, cacheonly); |
8929ecfa | 790 | |
d3c2fdcf | 791 | nr = trans->blocks_used; |
e9d0b13b | 792 | btrfs_end_transaction(trans, root); |
d3c2fdcf | 793 | btrfs_btree_balance_dirty(info->tree_root, nr); |
e9d0b13b CM |
794 | cond_resched(); |
795 | ||
3f157a2f | 796 | if (root->fs_info->closing || ret != -EAGAIN) |
e9d0b13b CM |
797 | break; |
798 | } | |
799 | root->defrag_running = 0; | |
8929ecfa | 800 | return ret; |
e9d0b13b CM |
801 | } |
802 | ||
2c47e605 | 803 | #if 0 |
b7ec40d7 CM |
804 | /* |
805 | * when dropping snapshots, we generate a ton of delayed refs, and it makes | |
806 | * sense not to join the transaction while it is trying to flush the current | |
807 | * queue of delayed refs out. | |
808 | * | |
809 | * This is used by the drop snapshot code only | |
810 | */ | |
811 | static noinline int wait_transaction_pre_flush(struct btrfs_fs_info *info) | |
812 | { | |
813 | DEFINE_WAIT(wait); | |
814 | ||
815 | mutex_lock(&info->trans_mutex); | |
816 | while (info->running_transaction && | |
817 | info->running_transaction->delayed_refs.flushing) { | |
818 | prepare_to_wait(&info->transaction_wait, &wait, | |
819 | TASK_UNINTERRUPTIBLE); | |
820 | mutex_unlock(&info->trans_mutex); | |
59bc5c75 | 821 | |
b7ec40d7 | 822 | schedule(); |
59bc5c75 | 823 | |
b7ec40d7 CM |
824 | mutex_lock(&info->trans_mutex); |
825 | finish_wait(&info->transaction_wait, &wait); | |
826 | } | |
827 | mutex_unlock(&info->trans_mutex); | |
828 | return 0; | |
829 | } | |
830 | ||
d352ac68 CM |
831 | /* |
832 | * Given a list of roots that need to be deleted, call btrfs_drop_snapshot on | |
833 | * all of them | |
834 | */ | |
5d4f98a2 | 835 | int btrfs_drop_dead_root(struct btrfs_root *root) |
0f7d52f4 | 836 | { |
0f7d52f4 | 837 | struct btrfs_trans_handle *trans; |
5d4f98a2 | 838 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
d3c2fdcf | 839 | unsigned long nr; |
5d4f98a2 | 840 | int ret; |
58176a96 | 841 | |
5d4f98a2 YZ |
842 | while (1) { |
843 | /* | |
844 | * we don't want to jump in and create a bunch of | |
845 | * delayed refs if the transaction is starting to close | |
846 | */ | |
847 | wait_transaction_pre_flush(tree_root->fs_info); | |
848 | trans = btrfs_start_transaction(tree_root, 1); | |
a2135011 | 849 | |
5d4f98a2 YZ |
850 | /* |
851 | * we've joined a transaction, make sure it isn't | |
852 | * closing right now | |
853 | */ | |
854 | if (trans->transaction->delayed_refs.flushing) { | |
855 | btrfs_end_transaction(trans, tree_root); | |
856 | continue; | |
9f3a7427 | 857 | } |
58176a96 | 858 | |
5d4f98a2 YZ |
859 | ret = btrfs_drop_snapshot(trans, root); |
860 | if (ret != -EAGAIN) | |
861 | break; | |
a2135011 | 862 | |
5d4f98a2 YZ |
863 | ret = btrfs_update_root(trans, tree_root, |
864 | &root->root_key, | |
865 | &root->root_item); | |
866 | if (ret) | |
54aa1f4d | 867 | break; |
bcc63abb | 868 | |
d3c2fdcf | 869 | nr = trans->blocks_used; |
0f7d52f4 CM |
870 | ret = btrfs_end_transaction(trans, tree_root); |
871 | BUG_ON(ret); | |
5eda7b5e | 872 | |
d3c2fdcf | 873 | btrfs_btree_balance_dirty(tree_root, nr); |
4dc11904 | 874 | cond_resched(); |
0f7d52f4 | 875 | } |
5d4f98a2 YZ |
876 | BUG_ON(ret); |
877 | ||
878 | ret = btrfs_del_root(trans, tree_root, &root->root_key); | |
879 | BUG_ON(ret); | |
880 | ||
881 | nr = trans->blocks_used; | |
882 | ret = btrfs_end_transaction(trans, tree_root); | |
883 | BUG_ON(ret); | |
884 | ||
885 | free_extent_buffer(root->node); | |
886 | free_extent_buffer(root->commit_root); | |
887 | kfree(root); | |
888 | ||
889 | btrfs_btree_balance_dirty(tree_root, nr); | |
54aa1f4d | 890 | return ret; |
0f7d52f4 | 891 | } |
2c47e605 | 892 | #endif |
0f7d52f4 | 893 | |
d352ac68 CM |
894 | /* |
895 | * new snapshots need to be created at a very specific time in the | |
896 | * transaction commit. This does the actual creation | |
897 | */ | |
80b6794d | 898 | static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, |
3063d29f CM |
899 | struct btrfs_fs_info *fs_info, |
900 | struct btrfs_pending_snapshot *pending) | |
901 | { | |
902 | struct btrfs_key key; | |
80b6794d | 903 | struct btrfs_root_item *new_root_item; |
3063d29f CM |
904 | struct btrfs_root *tree_root = fs_info->tree_root; |
905 | struct btrfs_root *root = pending->root; | |
6bdb72de SW |
906 | struct btrfs_root *parent_root; |
907 | struct inode *parent_inode; | |
6a912213 | 908 | struct dentry *parent; |
a22285a6 | 909 | struct dentry *dentry; |
3063d29f | 910 | struct extent_buffer *tmp; |
925baedd | 911 | struct extent_buffer *old; |
3063d29f | 912 | int ret; |
d68fc57b | 913 | u64 to_reserve = 0; |
6bdb72de | 914 | u64 index = 0; |
a22285a6 | 915 | u64 objectid; |
b83cc969 | 916 | u64 root_flags; |
3063d29f | 917 | |
80b6794d CM |
918 | new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS); |
919 | if (!new_root_item) { | |
a22285a6 | 920 | pending->error = -ENOMEM; |
80b6794d CM |
921 | goto fail; |
922 | } | |
a22285a6 | 923 | |
3063d29f | 924 | ret = btrfs_find_free_objectid(trans, tree_root, 0, &objectid); |
a22285a6 YZ |
925 | if (ret) { |
926 | pending->error = ret; | |
3063d29f | 927 | goto fail; |
a22285a6 | 928 | } |
3063d29f | 929 | |
3fd0a558 | 930 | btrfs_reloc_pre_snapshot(trans, pending, &to_reserve); |
d68fc57b YZ |
931 | btrfs_orphan_pre_snapshot(trans, pending, &to_reserve); |
932 | ||
933 | if (to_reserve > 0) { | |
934 | ret = btrfs_block_rsv_add(trans, root, &pending->block_rsv, | |
8bb8ab2e | 935 | to_reserve); |
d68fc57b YZ |
936 | if (ret) { |
937 | pending->error = ret; | |
938 | goto fail; | |
939 | } | |
940 | } | |
941 | ||
3063d29f | 942 | key.objectid = objectid; |
a22285a6 YZ |
943 | key.offset = (u64)-1; |
944 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3063d29f | 945 | |
a22285a6 | 946 | trans->block_rsv = &pending->block_rsv; |
3de4586c | 947 | |
a22285a6 | 948 | dentry = pending->dentry; |
6a912213 JB |
949 | parent = dget_parent(dentry); |
950 | parent_inode = parent->d_inode; | |
a22285a6 | 951 | parent_root = BTRFS_I(parent_inode)->root; |
6bdb72de | 952 | record_root_in_trans(trans, parent_root); |
a22285a6 | 953 | |
3063d29f CM |
954 | /* |
955 | * insert the directory item | |
956 | */ | |
3de4586c | 957 | ret = btrfs_set_inode_index(parent_inode, &index); |
6bdb72de | 958 | BUG_ON(ret); |
0660b5af | 959 | ret = btrfs_insert_dir_item(trans, parent_root, |
a22285a6 YZ |
960 | dentry->d_name.name, dentry->d_name.len, |
961 | parent_inode->i_ino, &key, | |
962 | BTRFS_FT_DIR, index); | |
6bdb72de | 963 | BUG_ON(ret); |
0660b5af | 964 | |
a22285a6 YZ |
965 | btrfs_i_size_write(parent_inode, parent_inode->i_size + |
966 | dentry->d_name.len * 2); | |
52c26179 YZ |
967 | ret = btrfs_update_inode(trans, parent_root, parent_inode); |
968 | BUG_ON(ret); | |
969 | ||
6bdb72de SW |
970 | record_root_in_trans(trans, root); |
971 | btrfs_set_root_last_snapshot(&root->root_item, trans->transid); | |
972 | memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); | |
973 | ||
b83cc969 LZ |
974 | root_flags = btrfs_root_flags(new_root_item); |
975 | if (pending->readonly) | |
976 | root_flags |= BTRFS_ROOT_SUBVOL_RDONLY; | |
977 | else | |
978 | root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY; | |
979 | btrfs_set_root_flags(new_root_item, root_flags); | |
980 | ||
6bdb72de SW |
981 | old = btrfs_lock_root_node(root); |
982 | btrfs_cow_block(trans, root, old, NULL, 0, &old); | |
983 | btrfs_set_lock_blocking(old); | |
984 | ||
985 | btrfs_copy_root(trans, root, old, &tmp, objectid); | |
986 | btrfs_tree_unlock(old); | |
987 | free_extent_buffer(old); | |
988 | ||
989 | btrfs_set_root_node(new_root_item, tmp); | |
a22285a6 YZ |
990 | /* record when the snapshot was created in key.offset */ |
991 | key.offset = trans->transid; | |
992 | ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); | |
6bdb72de SW |
993 | btrfs_tree_unlock(tmp); |
994 | free_extent_buffer(tmp); | |
a22285a6 | 995 | BUG_ON(ret); |
6bdb72de | 996 | |
a22285a6 YZ |
997 | /* |
998 | * insert root back/forward references | |
999 | */ | |
1000 | ret = btrfs_add_root_ref(trans, tree_root, objectid, | |
0660b5af | 1001 | parent_root->root_key.objectid, |
a22285a6 YZ |
1002 | parent_inode->i_ino, index, |
1003 | dentry->d_name.name, dentry->d_name.len); | |
0660b5af | 1004 | BUG_ON(ret); |
6a912213 | 1005 | dput(parent); |
0660b5af | 1006 | |
a22285a6 YZ |
1007 | key.offset = (u64)-1; |
1008 | pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key); | |
1009 | BUG_ON(IS_ERR(pending->snap)); | |
d68fc57b | 1010 | |
3fd0a558 | 1011 | btrfs_reloc_post_snapshot(trans, pending); |
d68fc57b | 1012 | btrfs_orphan_post_snapshot(trans, pending); |
3063d29f | 1013 | fail: |
6bdb72de | 1014 | kfree(new_root_item); |
a22285a6 YZ |
1015 | btrfs_block_rsv_release(root, &pending->block_rsv, (u64)-1); |
1016 | return 0; | |
3063d29f CM |
1017 | } |
1018 | ||
d352ac68 CM |
1019 | /* |
1020 | * create all the snapshots we've scheduled for creation | |
1021 | */ | |
80b6794d CM |
1022 | static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans, |
1023 | struct btrfs_fs_info *fs_info) | |
3de4586c CM |
1024 | { |
1025 | struct btrfs_pending_snapshot *pending; | |
1026 | struct list_head *head = &trans->transaction->pending_snapshots; | |
3de4586c CM |
1027 | int ret; |
1028 | ||
c6e30871 | 1029 | list_for_each_entry(pending, head, list) { |
3de4586c CM |
1030 | ret = create_pending_snapshot(trans, fs_info, pending); |
1031 | BUG_ON(ret); | |
1032 | } | |
1033 | return 0; | |
1034 | } | |
1035 | ||
5d4f98a2 YZ |
1036 | static void update_super_roots(struct btrfs_root *root) |
1037 | { | |
1038 | struct btrfs_root_item *root_item; | |
1039 | struct btrfs_super_block *super; | |
1040 | ||
1041 | super = &root->fs_info->super_copy; | |
1042 | ||
1043 | root_item = &root->fs_info->chunk_root->root_item; | |
1044 | super->chunk_root = root_item->bytenr; | |
1045 | super->chunk_root_generation = root_item->generation; | |
1046 | super->chunk_root_level = root_item->level; | |
1047 | ||
1048 | root_item = &root->fs_info->tree_root->root_item; | |
1049 | super->root = root_item->bytenr; | |
1050 | super->generation = root_item->generation; | |
1051 | super->root_level = root_item->level; | |
0af3d00b JB |
1052 | if (super->cache_generation != 0 || btrfs_test_opt(root, SPACE_CACHE)) |
1053 | super->cache_generation = root_item->generation; | |
5d4f98a2 YZ |
1054 | } |
1055 | ||
f36f3042 CM |
1056 | int btrfs_transaction_in_commit(struct btrfs_fs_info *info) |
1057 | { | |
1058 | int ret = 0; | |
1059 | spin_lock(&info->new_trans_lock); | |
1060 | if (info->running_transaction) | |
1061 | ret = info->running_transaction->in_commit; | |
1062 | spin_unlock(&info->new_trans_lock); | |
1063 | return ret; | |
1064 | } | |
1065 | ||
8929ecfa YZ |
1066 | int btrfs_transaction_blocked(struct btrfs_fs_info *info) |
1067 | { | |
1068 | int ret = 0; | |
1069 | spin_lock(&info->new_trans_lock); | |
1070 | if (info->running_transaction) | |
1071 | ret = info->running_transaction->blocked; | |
1072 | spin_unlock(&info->new_trans_lock); | |
1073 | return ret; | |
1074 | } | |
1075 | ||
bb9c12c9 SW |
1076 | /* |
1077 | * wait for the current transaction commit to start and block subsequent | |
1078 | * transaction joins | |
1079 | */ | |
1080 | static void wait_current_trans_commit_start(struct btrfs_root *root, | |
1081 | struct btrfs_transaction *trans) | |
1082 | { | |
1083 | DEFINE_WAIT(wait); | |
1084 | ||
1085 | if (trans->in_commit) | |
1086 | return; | |
1087 | ||
1088 | while (1) { | |
1089 | prepare_to_wait(&root->fs_info->transaction_blocked_wait, &wait, | |
1090 | TASK_UNINTERRUPTIBLE); | |
1091 | if (trans->in_commit) { | |
1092 | finish_wait(&root->fs_info->transaction_blocked_wait, | |
1093 | &wait); | |
1094 | break; | |
1095 | } | |
1096 | mutex_unlock(&root->fs_info->trans_mutex); | |
1097 | schedule(); | |
1098 | mutex_lock(&root->fs_info->trans_mutex); | |
1099 | finish_wait(&root->fs_info->transaction_blocked_wait, &wait); | |
1100 | } | |
1101 | } | |
1102 | ||
1103 | /* | |
1104 | * wait for the current transaction to start and then become unblocked. | |
1105 | * caller holds ref. | |
1106 | */ | |
1107 | static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root, | |
1108 | struct btrfs_transaction *trans) | |
1109 | { | |
1110 | DEFINE_WAIT(wait); | |
1111 | ||
1112 | if (trans->commit_done || (trans->in_commit && !trans->blocked)) | |
1113 | return; | |
1114 | ||
1115 | while (1) { | |
1116 | prepare_to_wait(&root->fs_info->transaction_wait, &wait, | |
1117 | TASK_UNINTERRUPTIBLE); | |
1118 | if (trans->commit_done || | |
1119 | (trans->in_commit && !trans->blocked)) { | |
1120 | finish_wait(&root->fs_info->transaction_wait, | |
1121 | &wait); | |
1122 | break; | |
1123 | } | |
1124 | mutex_unlock(&root->fs_info->trans_mutex); | |
1125 | schedule(); | |
1126 | mutex_lock(&root->fs_info->trans_mutex); | |
1127 | finish_wait(&root->fs_info->transaction_wait, | |
1128 | &wait); | |
1129 | } | |
1130 | } | |
1131 | ||
1132 | /* | |
1133 | * commit transactions asynchronously. once btrfs_commit_transaction_async | |
1134 | * returns, any subsequent transaction will not be allowed to join. | |
1135 | */ | |
1136 | struct btrfs_async_commit { | |
1137 | struct btrfs_trans_handle *newtrans; | |
1138 | struct btrfs_root *root; | |
1139 | struct delayed_work work; | |
1140 | }; | |
1141 | ||
1142 | static void do_async_commit(struct work_struct *work) | |
1143 | { | |
1144 | struct btrfs_async_commit *ac = | |
1145 | container_of(work, struct btrfs_async_commit, work.work); | |
1146 | ||
1147 | btrfs_commit_transaction(ac->newtrans, ac->root); | |
1148 | kfree(ac); | |
1149 | } | |
1150 | ||
1151 | int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans, | |
1152 | struct btrfs_root *root, | |
1153 | int wait_for_unblock) | |
1154 | { | |
1155 | struct btrfs_async_commit *ac; | |
1156 | struct btrfs_transaction *cur_trans; | |
1157 | ||
1158 | ac = kmalloc(sizeof(*ac), GFP_NOFS); | |
1159 | BUG_ON(!ac); | |
1160 | ||
1161 | INIT_DELAYED_WORK(&ac->work, do_async_commit); | |
1162 | ac->root = root; | |
1163 | ac->newtrans = btrfs_join_transaction(root, 0); | |
3612b495 TI |
1164 | if (IS_ERR(ac->newtrans)) { |
1165 | int err = PTR_ERR(ac->newtrans); | |
1166 | kfree(ac); | |
1167 | return err; | |
1168 | } | |
bb9c12c9 SW |
1169 | |
1170 | /* take transaction reference */ | |
1171 | mutex_lock(&root->fs_info->trans_mutex); | |
1172 | cur_trans = trans->transaction; | |
1173 | cur_trans->use_count++; | |
1174 | mutex_unlock(&root->fs_info->trans_mutex); | |
1175 | ||
1176 | btrfs_end_transaction(trans, root); | |
1177 | schedule_delayed_work(&ac->work, 0); | |
1178 | ||
1179 | /* wait for transaction to start and unblock */ | |
1180 | mutex_lock(&root->fs_info->trans_mutex); | |
1181 | if (wait_for_unblock) | |
1182 | wait_current_trans_commit_start_and_unblock(root, cur_trans); | |
1183 | else | |
1184 | wait_current_trans_commit_start(root, cur_trans); | |
1185 | put_transaction(cur_trans); | |
1186 | mutex_unlock(&root->fs_info->trans_mutex); | |
1187 | ||
1188 | return 0; | |
1189 | } | |
1190 | ||
1191 | /* | |
1192 | * btrfs_transaction state sequence: | |
1193 | * in_commit = 0, blocked = 0 (initial) | |
1194 | * in_commit = 1, blocked = 1 | |
1195 | * blocked = 0 | |
1196 | * commit_done = 1 | |
1197 | */ | |
79154b1b CM |
1198 | int btrfs_commit_transaction(struct btrfs_trans_handle *trans, |
1199 | struct btrfs_root *root) | |
1200 | { | |
15ee9bc7 | 1201 | unsigned long joined = 0; |
79154b1b | 1202 | struct btrfs_transaction *cur_trans; |
8fd17795 | 1203 | struct btrfs_transaction *prev_trans = NULL; |
79154b1b | 1204 | DEFINE_WAIT(wait); |
15ee9bc7 | 1205 | int ret; |
89573b9c CM |
1206 | int should_grow = 0; |
1207 | unsigned long now = get_seconds(); | |
dccae999 | 1208 | int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT); |
79154b1b | 1209 | |
5a3f23d5 CM |
1210 | btrfs_run_ordered_operations(root, 0); |
1211 | ||
56bec294 CM |
1212 | /* make a pass through all the delayed refs we have so far |
1213 | * any runnings procs may add more while we are here | |
1214 | */ | |
1215 | ret = btrfs_run_delayed_refs(trans, root, 0); | |
1216 | BUG_ON(ret); | |
1217 | ||
a22285a6 YZ |
1218 | btrfs_trans_release_metadata(trans, root); |
1219 | ||
b7ec40d7 | 1220 | cur_trans = trans->transaction; |
56bec294 CM |
1221 | /* |
1222 | * set the flushing flag so procs in this transaction have to | |
1223 | * start sending their work down. | |
1224 | */ | |
b7ec40d7 | 1225 | cur_trans->delayed_refs.flushing = 1; |
56bec294 | 1226 | |
c3e69d58 | 1227 | ret = btrfs_run_delayed_refs(trans, root, 0); |
56bec294 CM |
1228 | BUG_ON(ret); |
1229 | ||
79154b1b | 1230 | mutex_lock(&root->fs_info->trans_mutex); |
b7ec40d7 CM |
1231 | if (cur_trans->in_commit) { |
1232 | cur_trans->use_count++; | |
ccd467d6 | 1233 | mutex_unlock(&root->fs_info->trans_mutex); |
79154b1b | 1234 | btrfs_end_transaction(trans, root); |
ccd467d6 | 1235 | |
79154b1b CM |
1236 | ret = wait_for_commit(root, cur_trans); |
1237 | BUG_ON(ret); | |
15ee9bc7 JB |
1238 | |
1239 | mutex_lock(&root->fs_info->trans_mutex); | |
79154b1b | 1240 | put_transaction(cur_trans); |
15ee9bc7 JB |
1241 | mutex_unlock(&root->fs_info->trans_mutex); |
1242 | ||
79154b1b CM |
1243 | return 0; |
1244 | } | |
4313b399 | 1245 | |
2c90e5d6 | 1246 | trans->transaction->in_commit = 1; |
f9295749 | 1247 | trans->transaction->blocked = 1; |
bb9c12c9 SW |
1248 | wake_up(&root->fs_info->transaction_blocked_wait); |
1249 | ||
ccd467d6 CM |
1250 | if (cur_trans->list.prev != &root->fs_info->trans_list) { |
1251 | prev_trans = list_entry(cur_trans->list.prev, | |
1252 | struct btrfs_transaction, list); | |
1253 | if (!prev_trans->commit_done) { | |
1254 | prev_trans->use_count++; | |
ccd467d6 CM |
1255 | mutex_unlock(&root->fs_info->trans_mutex); |
1256 | ||
1257 | wait_for_commit(root, prev_trans); | |
ccd467d6 | 1258 | |
ccd467d6 | 1259 | mutex_lock(&root->fs_info->trans_mutex); |
15ee9bc7 | 1260 | put_transaction(prev_trans); |
ccd467d6 CM |
1261 | } |
1262 | } | |
15ee9bc7 | 1263 | |
89573b9c CM |
1264 | if (now < cur_trans->start_time || now - cur_trans->start_time < 1) |
1265 | should_grow = 1; | |
1266 | ||
15ee9bc7 | 1267 | do { |
7ea394f1 | 1268 | int snap_pending = 0; |
15ee9bc7 | 1269 | joined = cur_trans->num_joined; |
7ea394f1 YZ |
1270 | if (!list_empty(&trans->transaction->pending_snapshots)) |
1271 | snap_pending = 1; | |
1272 | ||
2c90e5d6 | 1273 | WARN_ON(cur_trans != trans->transaction); |
79154b1b | 1274 | mutex_unlock(&root->fs_info->trans_mutex); |
15ee9bc7 | 1275 | |
0bdb1db2 | 1276 | if (flush_on_commit || snap_pending) { |
24bbcf04 YZ |
1277 | btrfs_start_delalloc_inodes(root, 1); |
1278 | ret = btrfs_wait_ordered_extents(root, 0, 1); | |
ebecd3d9 | 1279 | BUG_ON(ret); |
7ea394f1 YZ |
1280 | } |
1281 | ||
5a3f23d5 CM |
1282 | /* |
1283 | * rename don't use btrfs_join_transaction, so, once we | |
1284 | * set the transaction to blocked above, we aren't going | |
1285 | * to get any new ordered operations. We can safely run | |
1286 | * it here and no for sure that nothing new will be added | |
1287 | * to the list | |
1288 | */ | |
1289 | btrfs_run_ordered_operations(root, 1); | |
1290 | ||
ed3b3d31 CM |
1291 | prepare_to_wait(&cur_trans->writer_wait, &wait, |
1292 | TASK_UNINTERRUPTIBLE); | |
1293 | ||
89573b9c | 1294 | smp_mb(); |
99d16cbc SW |
1295 | if (cur_trans->num_writers > 1) |
1296 | schedule_timeout(MAX_SCHEDULE_TIMEOUT); | |
1297 | else if (should_grow) | |
1298 | schedule_timeout(1); | |
15ee9bc7 | 1299 | |
79154b1b | 1300 | mutex_lock(&root->fs_info->trans_mutex); |
15ee9bc7 JB |
1301 | finish_wait(&cur_trans->writer_wait, &wait); |
1302 | } while (cur_trans->num_writers > 1 || | |
89573b9c | 1303 | (should_grow && cur_trans->num_joined != joined)); |
15ee9bc7 | 1304 | |
3063d29f CM |
1305 | ret = create_pending_snapshots(trans, root->fs_info); |
1306 | BUG_ON(ret); | |
1307 | ||
56bec294 CM |
1308 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
1309 | BUG_ON(ret); | |
1310 | ||
2c90e5d6 | 1311 | WARN_ON(cur_trans != trans->transaction); |
dc17ff8f | 1312 | |
e02119d5 CM |
1313 | /* btrfs_commit_tree_roots is responsible for getting the |
1314 | * various roots consistent with each other. Every pointer | |
1315 | * in the tree of tree roots has to point to the most up to date | |
1316 | * root for every subvolume and other tree. So, we have to keep | |
1317 | * the tree logging code from jumping in and changing any | |
1318 | * of the trees. | |
1319 | * | |
1320 | * At this point in the commit, there can't be any tree-log | |
1321 | * writers, but a little lower down we drop the trans mutex | |
1322 | * and let new people in. By holding the tree_log_mutex | |
1323 | * from now until after the super is written, we avoid races | |
1324 | * with the tree-log code. | |
1325 | */ | |
1326 | mutex_lock(&root->fs_info->tree_log_mutex); | |
1327 | ||
5d4f98a2 | 1328 | ret = commit_fs_roots(trans, root); |
54aa1f4d CM |
1329 | BUG_ON(ret); |
1330 | ||
5d4f98a2 | 1331 | /* commit_fs_roots gets rid of all the tree log roots, it is now |
e02119d5 CM |
1332 | * safe to free the root of tree log roots |
1333 | */ | |
1334 | btrfs_free_log_root_tree(trans, root->fs_info); | |
1335 | ||
5d4f98a2 | 1336 | ret = commit_cowonly_roots(trans, root); |
79154b1b | 1337 | BUG_ON(ret); |
54aa1f4d | 1338 | |
11833d66 YZ |
1339 | btrfs_prepare_extent_commit(trans, root); |
1340 | ||
78fae27e | 1341 | cur_trans = root->fs_info->running_transaction; |
cee36a03 | 1342 | spin_lock(&root->fs_info->new_trans_lock); |
78fae27e | 1343 | root->fs_info->running_transaction = NULL; |
cee36a03 | 1344 | spin_unlock(&root->fs_info->new_trans_lock); |
5d4f98a2 YZ |
1345 | |
1346 | btrfs_set_root_node(&root->fs_info->tree_root->root_item, | |
1347 | root->fs_info->tree_root->node); | |
817d52f8 | 1348 | switch_commit_root(root->fs_info->tree_root); |
5d4f98a2 YZ |
1349 | |
1350 | btrfs_set_root_node(&root->fs_info->chunk_root->root_item, | |
1351 | root->fs_info->chunk_root->node); | |
817d52f8 | 1352 | switch_commit_root(root->fs_info->chunk_root); |
5d4f98a2 YZ |
1353 | |
1354 | update_super_roots(root); | |
e02119d5 CM |
1355 | |
1356 | if (!root->fs_info->log_root_recovering) { | |
1357 | btrfs_set_super_log_root(&root->fs_info->super_copy, 0); | |
1358 | btrfs_set_super_log_root_level(&root->fs_info->super_copy, 0); | |
1359 | } | |
1360 | ||
a061fc8d CM |
1361 | memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy, |
1362 | sizeof(root->fs_info->super_copy)); | |
ccd467d6 | 1363 | |
f9295749 | 1364 | trans->transaction->blocked = 0; |
b7ec40d7 | 1365 | |
f9295749 | 1366 | wake_up(&root->fs_info->transaction_wait); |
e6dcd2dc | 1367 | |
78fae27e | 1368 | mutex_unlock(&root->fs_info->trans_mutex); |
79154b1b CM |
1369 | ret = btrfs_write_and_wait_transaction(trans, root); |
1370 | BUG_ON(ret); | |
a512bbf8 | 1371 | write_ctree_super(trans, root, 0); |
4313b399 | 1372 | |
e02119d5 CM |
1373 | /* |
1374 | * the super is written, we can safely allow the tree-loggers | |
1375 | * to go about their business | |
1376 | */ | |
1377 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
1378 | ||
11833d66 | 1379 | btrfs_finish_extent_commit(trans, root); |
4313b399 | 1380 | |
1a40e23b ZY |
1381 | mutex_lock(&root->fs_info->trans_mutex); |
1382 | ||
2c90e5d6 | 1383 | cur_trans->commit_done = 1; |
b7ec40d7 | 1384 | |
15ee9bc7 | 1385 | root->fs_info->last_trans_committed = cur_trans->transid; |
817d52f8 | 1386 | |
2c90e5d6 | 1387 | wake_up(&cur_trans->commit_wait); |
3de4586c | 1388 | |
78fae27e | 1389 | put_transaction(cur_trans); |
79154b1b | 1390 | put_transaction(cur_trans); |
58176a96 | 1391 | |
78fae27e | 1392 | mutex_unlock(&root->fs_info->trans_mutex); |
3de4586c | 1393 | |
9ed74f2d JB |
1394 | if (current->journal_info == trans) |
1395 | current->journal_info = NULL; | |
1396 | ||
2c90e5d6 | 1397 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
24bbcf04 YZ |
1398 | |
1399 | if (current != root->fs_info->transaction_kthread) | |
1400 | btrfs_run_delayed_iputs(root); | |
1401 | ||
79154b1b CM |
1402 | return ret; |
1403 | } | |
1404 | ||
d352ac68 CM |
1405 | /* |
1406 | * interface function to delete all the snapshots we have scheduled for deletion | |
1407 | */ | |
e9d0b13b CM |
1408 | int btrfs_clean_old_snapshots(struct btrfs_root *root) |
1409 | { | |
5d4f98a2 YZ |
1410 | LIST_HEAD(list); |
1411 | struct btrfs_fs_info *fs_info = root->fs_info; | |
1412 | ||
1413 | mutex_lock(&fs_info->trans_mutex); | |
1414 | list_splice_init(&fs_info->dead_roots, &list); | |
1415 | mutex_unlock(&fs_info->trans_mutex); | |
e9d0b13b | 1416 | |
5d4f98a2 YZ |
1417 | while (!list_empty(&list)) { |
1418 | root = list_entry(list.next, struct btrfs_root, root_list); | |
76dda93c YZ |
1419 | list_del(&root->root_list); |
1420 | ||
1421 | if (btrfs_header_backref_rev(root->node) < | |
1422 | BTRFS_MIXED_BACKREF_REV) | |
3fd0a558 | 1423 | btrfs_drop_snapshot(root, NULL, 0); |
76dda93c | 1424 | else |
3fd0a558 | 1425 | btrfs_drop_snapshot(root, NULL, 1); |
e9d0b13b CM |
1426 | } |
1427 | return 0; | |
1428 | } |