]>
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); | |
db5b493a TI |
60 | if (!cur_trans) |
61 | return -ENOMEM; | |
0f7d52f4 | 62 | root->fs_info->generation++; |
15ee9bc7 JB |
63 | cur_trans->num_writers = 1; |
64 | cur_trans->num_joined = 0; | |
0f7d52f4 | 65 | cur_trans->transid = root->fs_info->generation; |
79154b1b CM |
66 | init_waitqueue_head(&cur_trans->writer_wait); |
67 | init_waitqueue_head(&cur_trans->commit_wait); | |
68 | cur_trans->in_commit = 0; | |
f9295749 | 69 | cur_trans->blocked = 0; |
d5719762 | 70 | cur_trans->use_count = 1; |
79154b1b | 71 | cur_trans->commit_done = 0; |
08607c1b | 72 | cur_trans->start_time = get_seconds(); |
56bec294 | 73 | |
6bef4d31 | 74 | cur_trans->delayed_refs.root = RB_ROOT; |
56bec294 | 75 | cur_trans->delayed_refs.num_entries = 0; |
c3e69d58 CM |
76 | cur_trans->delayed_refs.num_heads_ready = 0; |
77 | cur_trans->delayed_refs.num_heads = 0; | |
56bec294 | 78 | cur_trans->delayed_refs.flushing = 0; |
c3e69d58 | 79 | cur_trans->delayed_refs.run_delayed_start = 0; |
56bec294 CM |
80 | spin_lock_init(&cur_trans->delayed_refs.lock); |
81 | ||
3063d29f | 82 | INIT_LIST_HEAD(&cur_trans->pending_snapshots); |
8fd17795 | 83 | list_add_tail(&cur_trans->list, &root->fs_info->trans_list); |
d1310b2e | 84 | extent_io_tree_init(&cur_trans->dirty_pages, |
5f39d397 CM |
85 | root->fs_info->btree_inode->i_mapping, |
86 | GFP_NOFS); | |
48ec2cf8 CM |
87 | spin_lock(&root->fs_info->new_trans_lock); |
88 | root->fs_info->running_transaction = cur_trans; | |
89 | spin_unlock(&root->fs_info->new_trans_lock); | |
15ee9bc7 JB |
90 | } else { |
91 | cur_trans->num_writers++; | |
92 | cur_trans->num_joined++; | |
79154b1b | 93 | } |
15ee9bc7 | 94 | |
79154b1b CM |
95 | return 0; |
96 | } | |
97 | ||
d352ac68 | 98 | /* |
d397712b CM |
99 | * this does all the record keeping required to make sure that a reference |
100 | * counted root is properly recorded in a given transaction. This is required | |
101 | * to make sure the old root from before we joined the transaction is deleted | |
102 | * when the transaction commits | |
d352ac68 | 103 | */ |
5d4f98a2 YZ |
104 | static noinline int record_root_in_trans(struct btrfs_trans_handle *trans, |
105 | struct btrfs_root *root) | |
6702ed49 | 106 | { |
5d4f98a2 | 107 | if (root->ref_cows && root->last_trans < trans->transid) { |
6702ed49 | 108 | WARN_ON(root == root->fs_info->extent_root); |
5d4f98a2 YZ |
109 | WARN_ON(root->commit_root != root->node); |
110 | ||
111 | radix_tree_tag_set(&root->fs_info->fs_roots_radix, | |
112 | (unsigned long)root->root_key.objectid, | |
113 | BTRFS_ROOT_TRANS_TAG); | |
114 | root->last_trans = trans->transid; | |
115 | btrfs_init_reloc_root(trans, root); | |
116 | } | |
117 | return 0; | |
118 | } | |
bcc63abb | 119 | |
5d4f98a2 YZ |
120 | int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, |
121 | struct btrfs_root *root) | |
122 | { | |
123 | if (!root->ref_cows) | |
124 | return 0; | |
bcc63abb | 125 | |
5d4f98a2 YZ |
126 | mutex_lock(&root->fs_info->trans_mutex); |
127 | if (root->last_trans == trans->transid) { | |
128 | mutex_unlock(&root->fs_info->trans_mutex); | |
129 | return 0; | |
6702ed49 | 130 | } |
5d4f98a2 YZ |
131 | |
132 | record_root_in_trans(trans, root); | |
133 | mutex_unlock(&root->fs_info->trans_mutex); | |
6702ed49 CM |
134 | return 0; |
135 | } | |
136 | ||
d352ac68 CM |
137 | /* wait for commit against the current transaction to become unblocked |
138 | * when this is done, it is safe to start a new transaction, but the current | |
139 | * transaction might not be fully on disk. | |
140 | */ | |
37d1aeee | 141 | static void wait_current_trans(struct btrfs_root *root) |
79154b1b | 142 | { |
f9295749 | 143 | struct btrfs_transaction *cur_trans; |
79154b1b | 144 | |
f9295749 | 145 | cur_trans = root->fs_info->running_transaction; |
37d1aeee | 146 | if (cur_trans && cur_trans->blocked) { |
f9295749 CM |
147 | DEFINE_WAIT(wait); |
148 | cur_trans->use_count++; | |
d397712b | 149 | while (1) { |
f9295749 CM |
150 | prepare_to_wait(&root->fs_info->transaction_wait, &wait, |
151 | TASK_UNINTERRUPTIBLE); | |
471fa17d | 152 | if (!cur_trans->blocked) |
f9295749 | 153 | break; |
471fa17d ZL |
154 | mutex_unlock(&root->fs_info->trans_mutex); |
155 | schedule(); | |
156 | mutex_lock(&root->fs_info->trans_mutex); | |
f9295749 | 157 | } |
471fa17d | 158 | finish_wait(&root->fs_info->transaction_wait, &wait); |
f9295749 CM |
159 | put_transaction(cur_trans); |
160 | } | |
37d1aeee CM |
161 | } |
162 | ||
249ac1e5 JB |
163 | enum btrfs_trans_type { |
164 | TRANS_START, | |
165 | TRANS_JOIN, | |
166 | TRANS_USERSPACE, | |
0af3d00b | 167 | TRANS_JOIN_NOLOCK, |
249ac1e5 JB |
168 | }; |
169 | ||
a22285a6 YZ |
170 | static int may_wait_transaction(struct btrfs_root *root, int type) |
171 | { | |
172 | if (!root->fs_info->log_root_recovering && | |
173 | ((type == TRANS_START && !root->fs_info->open_ioctl_trans) || | |
174 | type == TRANS_USERSPACE)) | |
175 | return 1; | |
176 | return 0; | |
177 | } | |
178 | ||
e02119d5 | 179 | static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root, |
a22285a6 | 180 | u64 num_items, int type) |
37d1aeee | 181 | { |
a22285a6 YZ |
182 | struct btrfs_trans_handle *h; |
183 | struct btrfs_transaction *cur_trans; | |
37d1aeee | 184 | int ret; |
acce952b | 185 | |
186 | if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) | |
187 | return ERR_PTR(-EROFS); | |
a22285a6 YZ |
188 | again: |
189 | h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS); | |
190 | if (!h) | |
191 | return ERR_PTR(-ENOMEM); | |
37d1aeee | 192 | |
0af3d00b JB |
193 | if (type != TRANS_JOIN_NOLOCK) |
194 | mutex_lock(&root->fs_info->trans_mutex); | |
a22285a6 | 195 | if (may_wait_transaction(root, type)) |
37d1aeee | 196 | wait_current_trans(root); |
a22285a6 | 197 | |
79154b1b | 198 | ret = join_transaction(root); |
db5b493a TI |
199 | if (ret < 0) { |
200 | if (type != TRANS_JOIN_NOLOCK) | |
201 | mutex_unlock(&root->fs_info->trans_mutex); | |
202 | return ERR_PTR(ret); | |
203 | } | |
0f7d52f4 | 204 | |
a22285a6 YZ |
205 | cur_trans = root->fs_info->running_transaction; |
206 | cur_trans->use_count++; | |
0af3d00b JB |
207 | if (type != TRANS_JOIN_NOLOCK) |
208 | mutex_unlock(&root->fs_info->trans_mutex); | |
a22285a6 YZ |
209 | |
210 | h->transid = cur_trans->transid; | |
211 | h->transaction = cur_trans; | |
79154b1b | 212 | h->blocks_used = 0; |
d2fb3437 | 213 | h->block_group = 0; |
a22285a6 | 214 | h->bytes_reserved = 0; |
56bec294 | 215 | h->delayed_ref_updates = 0; |
f0486c68 | 216 | h->block_rsv = NULL; |
b7ec40d7 | 217 | |
a22285a6 YZ |
218 | smp_mb(); |
219 | if (cur_trans->blocked && may_wait_transaction(root, type)) { | |
220 | btrfs_commit_transaction(h, root); | |
221 | goto again; | |
222 | } | |
223 | ||
224 | if (num_items > 0) { | |
8bb8ab2e | 225 | ret = btrfs_trans_reserve_metadata(h, root, num_items); |
a22285a6 YZ |
226 | if (ret == -EAGAIN) { |
227 | btrfs_commit_transaction(h, root); | |
228 | goto again; | |
229 | } | |
230 | if (ret < 0) { | |
231 | btrfs_end_transaction(h, root); | |
232 | return ERR_PTR(ret); | |
233 | } | |
234 | } | |
9ed74f2d | 235 | |
0af3d00b JB |
236 | if (type != TRANS_JOIN_NOLOCK) |
237 | mutex_lock(&root->fs_info->trans_mutex); | |
5d4f98a2 | 238 | record_root_in_trans(h, root); |
0af3d00b JB |
239 | if (type != TRANS_JOIN_NOLOCK) |
240 | mutex_unlock(&root->fs_info->trans_mutex); | |
a22285a6 YZ |
241 | |
242 | if (!current->journal_info && type != TRANS_USERSPACE) | |
243 | current->journal_info = h; | |
79154b1b CM |
244 | return h; |
245 | } | |
246 | ||
f9295749 | 247 | struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, |
a22285a6 | 248 | int num_items) |
f9295749 | 249 | { |
a22285a6 | 250 | return start_transaction(root, num_items, TRANS_START); |
f9295749 CM |
251 | } |
252 | struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root, | |
253 | int num_blocks) | |
254 | { | |
a22285a6 | 255 | return start_transaction(root, 0, TRANS_JOIN); |
f9295749 CM |
256 | } |
257 | ||
0af3d00b JB |
258 | struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root, |
259 | int num_blocks) | |
260 | { | |
261 | return start_transaction(root, 0, TRANS_JOIN_NOLOCK); | |
262 | } | |
263 | ||
9ca9ee09 SW |
264 | struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *r, |
265 | int num_blocks) | |
266 | { | |
a22285a6 | 267 | return start_transaction(r, 0, TRANS_USERSPACE); |
9ca9ee09 SW |
268 | } |
269 | ||
d352ac68 | 270 | /* wait for a transaction commit to be fully complete */ |
89ce8a63 CM |
271 | static noinline int wait_for_commit(struct btrfs_root *root, |
272 | struct btrfs_transaction *commit) | |
273 | { | |
274 | DEFINE_WAIT(wait); | |
275 | mutex_lock(&root->fs_info->trans_mutex); | |
d397712b | 276 | while (!commit->commit_done) { |
89ce8a63 CM |
277 | prepare_to_wait(&commit->commit_wait, &wait, |
278 | TASK_UNINTERRUPTIBLE); | |
279 | if (commit->commit_done) | |
280 | break; | |
281 | mutex_unlock(&root->fs_info->trans_mutex); | |
282 | schedule(); | |
283 | mutex_lock(&root->fs_info->trans_mutex); | |
284 | } | |
285 | mutex_unlock(&root->fs_info->trans_mutex); | |
286 | finish_wait(&commit->commit_wait, &wait); | |
287 | return 0; | |
288 | } | |
289 | ||
46204592 SW |
290 | int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid) |
291 | { | |
292 | struct btrfs_transaction *cur_trans = NULL, *t; | |
293 | int ret; | |
294 | ||
295 | mutex_lock(&root->fs_info->trans_mutex); | |
296 | ||
297 | ret = 0; | |
298 | if (transid) { | |
299 | if (transid <= root->fs_info->last_trans_committed) | |
300 | goto out_unlock; | |
301 | ||
302 | /* find specified transaction */ | |
303 | list_for_each_entry(t, &root->fs_info->trans_list, list) { | |
304 | if (t->transid == transid) { | |
305 | cur_trans = t; | |
306 | break; | |
307 | } | |
308 | if (t->transid > transid) | |
309 | break; | |
310 | } | |
311 | ret = -EINVAL; | |
312 | if (!cur_trans) | |
313 | goto out_unlock; /* bad transid */ | |
314 | } else { | |
315 | /* find newest transaction that is committing | committed */ | |
316 | list_for_each_entry_reverse(t, &root->fs_info->trans_list, | |
317 | list) { | |
318 | if (t->in_commit) { | |
319 | if (t->commit_done) | |
320 | goto out_unlock; | |
321 | cur_trans = t; | |
322 | break; | |
323 | } | |
324 | } | |
325 | if (!cur_trans) | |
326 | goto out_unlock; /* nothing committing|committed */ | |
327 | } | |
328 | ||
329 | cur_trans->use_count++; | |
330 | mutex_unlock(&root->fs_info->trans_mutex); | |
331 | ||
332 | wait_for_commit(root, cur_trans); | |
333 | ||
334 | mutex_lock(&root->fs_info->trans_mutex); | |
335 | put_transaction(cur_trans); | |
336 | ret = 0; | |
337 | out_unlock: | |
338 | mutex_unlock(&root->fs_info->trans_mutex); | |
339 | return ret; | |
340 | } | |
341 | ||
5d4f98a2 | 342 | #if 0 |
d352ac68 | 343 | /* |
d397712b CM |
344 | * rate limit against the drop_snapshot code. This helps to slow down new |
345 | * operations if the drop_snapshot code isn't able to keep up. | |
d352ac68 | 346 | */ |
37d1aeee | 347 | static void throttle_on_drops(struct btrfs_root *root) |
ab78c84d CM |
348 | { |
349 | struct btrfs_fs_info *info = root->fs_info; | |
2dd3e67b | 350 | int harder_count = 0; |
ab78c84d | 351 | |
2dd3e67b | 352 | harder: |
ab78c84d CM |
353 | if (atomic_read(&info->throttles)) { |
354 | DEFINE_WAIT(wait); | |
355 | int thr; | |
ab78c84d CM |
356 | thr = atomic_read(&info->throttle_gen); |
357 | ||
358 | do { | |
359 | prepare_to_wait(&info->transaction_throttle, | |
360 | &wait, TASK_UNINTERRUPTIBLE); | |
361 | if (!atomic_read(&info->throttles)) { | |
362 | finish_wait(&info->transaction_throttle, &wait); | |
363 | break; | |
364 | } | |
365 | schedule(); | |
366 | finish_wait(&info->transaction_throttle, &wait); | |
367 | } while (thr == atomic_read(&info->throttle_gen)); | |
2dd3e67b CM |
368 | harder_count++; |
369 | ||
370 | if (root->fs_info->total_ref_cache_size > 1 * 1024 * 1024 && | |
371 | harder_count < 2) | |
372 | goto harder; | |
373 | ||
374 | if (root->fs_info->total_ref_cache_size > 5 * 1024 * 1024 && | |
375 | harder_count < 10) | |
376 | goto harder; | |
377 | ||
378 | if (root->fs_info->total_ref_cache_size > 10 * 1024 * 1024 && | |
379 | harder_count < 20) | |
380 | goto harder; | |
ab78c84d CM |
381 | } |
382 | } | |
5d4f98a2 | 383 | #endif |
ab78c84d | 384 | |
37d1aeee CM |
385 | void btrfs_throttle(struct btrfs_root *root) |
386 | { | |
387 | mutex_lock(&root->fs_info->trans_mutex); | |
9ca9ee09 SW |
388 | if (!root->fs_info->open_ioctl_trans) |
389 | wait_current_trans(root); | |
37d1aeee | 390 | mutex_unlock(&root->fs_info->trans_mutex); |
37d1aeee CM |
391 | } |
392 | ||
8929ecfa YZ |
393 | static int should_end_transaction(struct btrfs_trans_handle *trans, |
394 | struct btrfs_root *root) | |
395 | { | |
396 | int ret; | |
397 | ret = btrfs_block_rsv_check(trans, root, | |
398 | &root->fs_info->global_block_rsv, 0, 5); | |
399 | return ret ? 1 : 0; | |
400 | } | |
401 | ||
402 | int btrfs_should_end_transaction(struct btrfs_trans_handle *trans, | |
403 | struct btrfs_root *root) | |
404 | { | |
405 | struct btrfs_transaction *cur_trans = trans->transaction; | |
406 | int updates; | |
407 | ||
408 | if (cur_trans->blocked || cur_trans->delayed_refs.flushing) | |
409 | return 1; | |
410 | ||
411 | updates = trans->delayed_ref_updates; | |
412 | trans->delayed_ref_updates = 0; | |
413 | if (updates) | |
414 | btrfs_run_delayed_refs(trans, root, updates); | |
415 | ||
416 | return should_end_transaction(trans, root); | |
417 | } | |
418 | ||
89ce8a63 | 419 | static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, |
0af3d00b | 420 | struct btrfs_root *root, int throttle, int lock) |
79154b1b | 421 | { |
8929ecfa | 422 | struct btrfs_transaction *cur_trans = trans->transaction; |
ab78c84d | 423 | struct btrfs_fs_info *info = root->fs_info; |
c3e69d58 CM |
424 | int count = 0; |
425 | ||
426 | while (count < 4) { | |
427 | unsigned long cur = trans->delayed_ref_updates; | |
428 | trans->delayed_ref_updates = 0; | |
429 | if (cur && | |
430 | trans->transaction->delayed_refs.num_heads_ready > 64) { | |
431 | trans->delayed_ref_updates = 0; | |
b7ec40d7 CM |
432 | |
433 | /* | |
434 | * do a full flush if the transaction is trying | |
435 | * to close | |
436 | */ | |
437 | if (trans->transaction->delayed_refs.flushing) | |
438 | cur = 0; | |
c3e69d58 CM |
439 | btrfs_run_delayed_refs(trans, root, cur); |
440 | } else { | |
441 | break; | |
442 | } | |
443 | count++; | |
56bec294 CM |
444 | } |
445 | ||
a22285a6 YZ |
446 | btrfs_trans_release_metadata(trans, root); |
447 | ||
0af3d00b | 448 | if (lock && !root->fs_info->open_ioctl_trans && |
8929ecfa YZ |
449 | should_end_transaction(trans, root)) |
450 | trans->transaction->blocked = 1; | |
451 | ||
0af3d00b | 452 | if (lock && cur_trans->blocked && !cur_trans->in_commit) { |
8929ecfa YZ |
453 | if (throttle) |
454 | return btrfs_commit_transaction(trans, root); | |
455 | else | |
456 | wake_up_process(info->transaction_kthread); | |
457 | } | |
458 | ||
0af3d00b JB |
459 | if (lock) |
460 | mutex_lock(&info->trans_mutex); | |
8929ecfa | 461 | WARN_ON(cur_trans != info->running_transaction); |
d5719762 | 462 | WARN_ON(cur_trans->num_writers < 1); |
ccd467d6 | 463 | cur_trans->num_writers--; |
89ce8a63 | 464 | |
99d16cbc | 465 | smp_mb(); |
79154b1b CM |
466 | if (waitqueue_active(&cur_trans->writer_wait)) |
467 | wake_up(&cur_trans->writer_wait); | |
79154b1b | 468 | put_transaction(cur_trans); |
0af3d00b JB |
469 | if (lock) |
470 | mutex_unlock(&info->trans_mutex); | |
9ed74f2d JB |
471 | |
472 | if (current->journal_info == trans) | |
473 | current->journal_info = NULL; | |
d6025579 | 474 | memset(trans, 0, sizeof(*trans)); |
2c90e5d6 | 475 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
ab78c84d | 476 | |
24bbcf04 YZ |
477 | if (throttle) |
478 | btrfs_run_delayed_iputs(root); | |
479 | ||
79154b1b CM |
480 | return 0; |
481 | } | |
482 | ||
89ce8a63 CM |
483 | int btrfs_end_transaction(struct btrfs_trans_handle *trans, |
484 | struct btrfs_root *root) | |
485 | { | |
0af3d00b | 486 | return __btrfs_end_transaction(trans, root, 0, 1); |
89ce8a63 CM |
487 | } |
488 | ||
489 | int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans, | |
490 | struct btrfs_root *root) | |
491 | { | |
0af3d00b JB |
492 | return __btrfs_end_transaction(trans, root, 1, 1); |
493 | } | |
494 | ||
495 | int btrfs_end_transaction_nolock(struct btrfs_trans_handle *trans, | |
496 | struct btrfs_root *root) | |
497 | { | |
498 | return __btrfs_end_transaction(trans, root, 0, 0); | |
89ce8a63 CM |
499 | } |
500 | ||
d352ac68 CM |
501 | /* |
502 | * when btree blocks are allocated, they have some corresponding bits set for | |
503 | * them in one of two extent_io trees. This is used to make sure all of | |
690587d1 | 504 | * those extents are sent to disk but does not wait on them |
d352ac68 | 505 | */ |
690587d1 | 506 | int btrfs_write_marked_extents(struct btrfs_root *root, |
8cef4e16 | 507 | struct extent_io_tree *dirty_pages, int mark) |
79154b1b | 508 | { |
7c4452b9 | 509 | int ret; |
777e6bd7 | 510 | int err = 0; |
7c4452b9 CM |
511 | int werr = 0; |
512 | struct page *page; | |
7c4452b9 | 513 | struct inode *btree_inode = root->fs_info->btree_inode; |
777e6bd7 | 514 | u64 start = 0; |
5f39d397 CM |
515 | u64 end; |
516 | unsigned long index; | |
7c4452b9 | 517 | |
d397712b | 518 | while (1) { |
777e6bd7 | 519 | ret = find_first_extent_bit(dirty_pages, start, &start, &end, |
8cef4e16 | 520 | mark); |
5f39d397 | 521 | if (ret) |
7c4452b9 | 522 | break; |
d397712b | 523 | while (start <= end) { |
777e6bd7 CM |
524 | cond_resched(); |
525 | ||
5f39d397 | 526 | index = start >> PAGE_CACHE_SHIFT; |
35ebb934 | 527 | start = (u64)(index + 1) << PAGE_CACHE_SHIFT; |
4bef0848 | 528 | page = find_get_page(btree_inode->i_mapping, index); |
7c4452b9 CM |
529 | if (!page) |
530 | continue; | |
4bef0848 CM |
531 | |
532 | btree_lock_page_hook(page); | |
533 | if (!page->mapping) { | |
534 | unlock_page(page); | |
535 | page_cache_release(page); | |
536 | continue; | |
537 | } | |
538 | ||
6702ed49 CM |
539 | if (PageWriteback(page)) { |
540 | if (PageDirty(page)) | |
541 | wait_on_page_writeback(page); | |
542 | else { | |
543 | unlock_page(page); | |
544 | page_cache_release(page); | |
545 | continue; | |
546 | } | |
547 | } | |
7c4452b9 CM |
548 | err = write_one_page(page, 0); |
549 | if (err) | |
550 | werr = err; | |
551 | page_cache_release(page); | |
552 | } | |
553 | } | |
690587d1 CM |
554 | if (err) |
555 | werr = err; | |
556 | return werr; | |
557 | } | |
558 | ||
559 | /* | |
560 | * when btree blocks are allocated, they have some corresponding bits set for | |
561 | * them in one of two extent_io trees. This is used to make sure all of | |
562 | * those extents are on disk for transaction or log commit. We wait | |
563 | * on all the pages and clear them from the dirty pages state tree | |
564 | */ | |
565 | int btrfs_wait_marked_extents(struct btrfs_root *root, | |
8cef4e16 | 566 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 CM |
567 | { |
568 | int ret; | |
569 | int err = 0; | |
570 | int werr = 0; | |
571 | struct page *page; | |
572 | struct inode *btree_inode = root->fs_info->btree_inode; | |
573 | u64 start = 0; | |
574 | u64 end; | |
575 | unsigned long index; | |
576 | ||
d397712b | 577 | while (1) { |
8cef4e16 YZ |
578 | ret = find_first_extent_bit(dirty_pages, start, &start, &end, |
579 | mark); | |
777e6bd7 CM |
580 | if (ret) |
581 | break; | |
582 | ||
8cef4e16 | 583 | clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS); |
d397712b | 584 | while (start <= end) { |
777e6bd7 CM |
585 | index = start >> PAGE_CACHE_SHIFT; |
586 | start = (u64)(index + 1) << PAGE_CACHE_SHIFT; | |
587 | page = find_get_page(btree_inode->i_mapping, index); | |
588 | if (!page) | |
589 | continue; | |
590 | if (PageDirty(page)) { | |
4bef0848 CM |
591 | btree_lock_page_hook(page); |
592 | wait_on_page_writeback(page); | |
777e6bd7 CM |
593 | err = write_one_page(page, 0); |
594 | if (err) | |
595 | werr = err; | |
596 | } | |
105d931d | 597 | wait_on_page_writeback(page); |
777e6bd7 CM |
598 | page_cache_release(page); |
599 | cond_resched(); | |
600 | } | |
601 | } | |
7c4452b9 CM |
602 | if (err) |
603 | werr = err; | |
604 | return werr; | |
79154b1b CM |
605 | } |
606 | ||
690587d1 CM |
607 | /* |
608 | * when btree blocks are allocated, they have some corresponding bits set for | |
609 | * them in one of two extent_io trees. This is used to make sure all of | |
610 | * those extents are on disk for transaction or log commit | |
611 | */ | |
612 | int btrfs_write_and_wait_marked_extents(struct btrfs_root *root, | |
8cef4e16 | 613 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 CM |
614 | { |
615 | int ret; | |
616 | int ret2; | |
617 | ||
8cef4e16 YZ |
618 | ret = btrfs_write_marked_extents(root, dirty_pages, mark); |
619 | ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark); | |
690587d1 CM |
620 | return ret || ret2; |
621 | } | |
622 | ||
d0c803c4 CM |
623 | int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans, |
624 | struct btrfs_root *root) | |
625 | { | |
626 | if (!trans || !trans->transaction) { | |
627 | struct inode *btree_inode; | |
628 | btree_inode = root->fs_info->btree_inode; | |
629 | return filemap_write_and_wait(btree_inode->i_mapping); | |
630 | } | |
631 | return btrfs_write_and_wait_marked_extents(root, | |
8cef4e16 YZ |
632 | &trans->transaction->dirty_pages, |
633 | EXTENT_DIRTY); | |
d0c803c4 CM |
634 | } |
635 | ||
d352ac68 CM |
636 | /* |
637 | * this is used to update the root pointer in the tree of tree roots. | |
638 | * | |
639 | * But, in the case of the extent allocation tree, updating the root | |
640 | * pointer may allocate blocks which may change the root of the extent | |
641 | * allocation tree. | |
642 | * | |
643 | * So, this loops and repeats and makes sure the cowonly root didn't | |
644 | * change while the root pointer was being updated in the metadata. | |
645 | */ | |
0b86a832 CM |
646 | static int update_cowonly_root(struct btrfs_trans_handle *trans, |
647 | struct btrfs_root *root) | |
79154b1b CM |
648 | { |
649 | int ret; | |
0b86a832 | 650 | u64 old_root_bytenr; |
86b9f2ec | 651 | u64 old_root_used; |
0b86a832 | 652 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
79154b1b | 653 | |
86b9f2ec | 654 | old_root_used = btrfs_root_used(&root->root_item); |
0b86a832 | 655 | btrfs_write_dirty_block_groups(trans, root); |
56bec294 | 656 | |
d397712b | 657 | while (1) { |
0b86a832 | 658 | old_root_bytenr = btrfs_root_bytenr(&root->root_item); |
86b9f2ec YZ |
659 | if (old_root_bytenr == root->node->start && |
660 | old_root_used == btrfs_root_used(&root->root_item)) | |
79154b1b | 661 | break; |
87ef2bb4 | 662 | |
5d4f98a2 | 663 | btrfs_set_root_node(&root->root_item, root->node); |
79154b1b | 664 | ret = btrfs_update_root(trans, tree_root, |
0b86a832 CM |
665 | &root->root_key, |
666 | &root->root_item); | |
79154b1b | 667 | BUG_ON(ret); |
56bec294 | 668 | |
86b9f2ec | 669 | old_root_used = btrfs_root_used(&root->root_item); |
4a8c9a62 | 670 | ret = btrfs_write_dirty_block_groups(trans, root); |
56bec294 | 671 | BUG_ON(ret); |
0b86a832 | 672 | } |
276e680d YZ |
673 | |
674 | if (root != root->fs_info->extent_root) | |
675 | switch_commit_root(root); | |
676 | ||
0b86a832 CM |
677 | return 0; |
678 | } | |
679 | ||
d352ac68 CM |
680 | /* |
681 | * update all the cowonly tree roots on disk | |
682 | */ | |
5d4f98a2 YZ |
683 | static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans, |
684 | struct btrfs_root *root) | |
0b86a832 CM |
685 | { |
686 | struct btrfs_fs_info *fs_info = root->fs_info; | |
687 | struct list_head *next; | |
84234f3a | 688 | struct extent_buffer *eb; |
56bec294 | 689 | int ret; |
84234f3a | 690 | |
56bec294 CM |
691 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
692 | BUG_ON(ret); | |
87ef2bb4 | 693 | |
84234f3a | 694 | eb = btrfs_lock_root_node(fs_info->tree_root); |
9fa8cfe7 | 695 | btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 0, &eb); |
84234f3a YZ |
696 | btrfs_tree_unlock(eb); |
697 | free_extent_buffer(eb); | |
0b86a832 | 698 | |
56bec294 CM |
699 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
700 | BUG_ON(ret); | |
87ef2bb4 | 701 | |
d397712b | 702 | while (!list_empty(&fs_info->dirty_cowonly_roots)) { |
0b86a832 CM |
703 | next = fs_info->dirty_cowonly_roots.next; |
704 | list_del_init(next); | |
705 | root = list_entry(next, struct btrfs_root, dirty_list); | |
87ef2bb4 | 706 | |
0b86a832 | 707 | update_cowonly_root(trans, root); |
79154b1b | 708 | } |
276e680d YZ |
709 | |
710 | down_write(&fs_info->extent_commit_sem); | |
711 | switch_commit_root(fs_info->extent_root); | |
712 | up_write(&fs_info->extent_commit_sem); | |
713 | ||
79154b1b CM |
714 | return 0; |
715 | } | |
716 | ||
d352ac68 CM |
717 | /* |
718 | * dead roots are old snapshots that need to be deleted. This allocates | |
719 | * a dirty root struct and adds it into the list of dead roots that need to | |
720 | * be deleted | |
721 | */ | |
5d4f98a2 | 722 | int btrfs_add_dead_root(struct btrfs_root *root) |
5eda7b5e | 723 | { |
b48652c1 | 724 | mutex_lock(&root->fs_info->trans_mutex); |
5d4f98a2 | 725 | list_add(&root->root_list, &root->fs_info->dead_roots); |
b48652c1 | 726 | mutex_unlock(&root->fs_info->trans_mutex); |
5eda7b5e CM |
727 | return 0; |
728 | } | |
729 | ||
d352ac68 | 730 | /* |
5d4f98a2 | 731 | * update all the cowonly tree roots on disk |
d352ac68 | 732 | */ |
5d4f98a2 YZ |
733 | static noinline int commit_fs_roots(struct btrfs_trans_handle *trans, |
734 | struct btrfs_root *root) | |
0f7d52f4 | 735 | { |
0f7d52f4 | 736 | struct btrfs_root *gang[8]; |
5d4f98a2 | 737 | struct btrfs_fs_info *fs_info = root->fs_info; |
0f7d52f4 CM |
738 | int i; |
739 | int ret; | |
54aa1f4d CM |
740 | int err = 0; |
741 | ||
d397712b | 742 | while (1) { |
5d4f98a2 YZ |
743 | ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, |
744 | (void **)gang, 0, | |
0f7d52f4 CM |
745 | ARRAY_SIZE(gang), |
746 | BTRFS_ROOT_TRANS_TAG); | |
747 | if (ret == 0) | |
748 | break; | |
749 | for (i = 0; i < ret; i++) { | |
750 | root = gang[i]; | |
5d4f98a2 YZ |
751 | radix_tree_tag_clear(&fs_info->fs_roots_radix, |
752 | (unsigned long)root->root_key.objectid, | |
753 | BTRFS_ROOT_TRANS_TAG); | |
31153d81 | 754 | |
e02119d5 | 755 | btrfs_free_log(trans, root); |
5d4f98a2 | 756 | btrfs_update_reloc_root(trans, root); |
d68fc57b | 757 | btrfs_orphan_commit_root(trans, root); |
bcc63abb | 758 | |
978d910d | 759 | if (root->commit_root != root->node) { |
817d52f8 | 760 | switch_commit_root(root); |
978d910d YZ |
761 | btrfs_set_root_node(&root->root_item, |
762 | root->node); | |
763 | } | |
5d4f98a2 | 764 | |
5d4f98a2 | 765 | err = btrfs_update_root(trans, fs_info->tree_root, |
0f7d52f4 CM |
766 | &root->root_key, |
767 | &root->root_item); | |
54aa1f4d CM |
768 | if (err) |
769 | break; | |
0f7d52f4 CM |
770 | } |
771 | } | |
54aa1f4d | 772 | return err; |
0f7d52f4 CM |
773 | } |
774 | ||
d352ac68 CM |
775 | /* |
776 | * defrag a given btree. If cacheonly == 1, this won't read from the disk, | |
777 | * otherwise every leaf in the btree is read and defragged. | |
778 | */ | |
e9d0b13b CM |
779 | int btrfs_defrag_root(struct btrfs_root *root, int cacheonly) |
780 | { | |
781 | struct btrfs_fs_info *info = root->fs_info; | |
e9d0b13b | 782 | struct btrfs_trans_handle *trans; |
8929ecfa | 783 | int ret; |
d3c2fdcf | 784 | unsigned long nr; |
e9d0b13b | 785 | |
8929ecfa | 786 | if (xchg(&root->defrag_running, 1)) |
e9d0b13b | 787 | return 0; |
8929ecfa | 788 | |
6b80053d | 789 | while (1) { |
8929ecfa YZ |
790 | trans = btrfs_start_transaction(root, 0); |
791 | if (IS_ERR(trans)) | |
792 | return PTR_ERR(trans); | |
793 | ||
e9d0b13b | 794 | ret = btrfs_defrag_leaves(trans, root, cacheonly); |
8929ecfa | 795 | |
d3c2fdcf | 796 | nr = trans->blocks_used; |
e9d0b13b | 797 | btrfs_end_transaction(trans, root); |
d3c2fdcf | 798 | btrfs_btree_balance_dirty(info->tree_root, nr); |
e9d0b13b CM |
799 | cond_resched(); |
800 | ||
3f157a2f | 801 | if (root->fs_info->closing || ret != -EAGAIN) |
e9d0b13b CM |
802 | break; |
803 | } | |
804 | root->defrag_running = 0; | |
8929ecfa | 805 | return ret; |
e9d0b13b CM |
806 | } |
807 | ||
2c47e605 | 808 | #if 0 |
b7ec40d7 CM |
809 | /* |
810 | * when dropping snapshots, we generate a ton of delayed refs, and it makes | |
811 | * sense not to join the transaction while it is trying to flush the current | |
812 | * queue of delayed refs out. | |
813 | * | |
814 | * This is used by the drop snapshot code only | |
815 | */ | |
816 | static noinline int wait_transaction_pre_flush(struct btrfs_fs_info *info) | |
817 | { | |
818 | DEFINE_WAIT(wait); | |
819 | ||
820 | mutex_lock(&info->trans_mutex); | |
821 | while (info->running_transaction && | |
822 | info->running_transaction->delayed_refs.flushing) { | |
823 | prepare_to_wait(&info->transaction_wait, &wait, | |
824 | TASK_UNINTERRUPTIBLE); | |
825 | mutex_unlock(&info->trans_mutex); | |
59bc5c75 | 826 | |
b7ec40d7 | 827 | schedule(); |
59bc5c75 | 828 | |
b7ec40d7 CM |
829 | mutex_lock(&info->trans_mutex); |
830 | finish_wait(&info->transaction_wait, &wait); | |
831 | } | |
832 | mutex_unlock(&info->trans_mutex); | |
833 | return 0; | |
834 | } | |
835 | ||
d352ac68 CM |
836 | /* |
837 | * Given a list of roots that need to be deleted, call btrfs_drop_snapshot on | |
838 | * all of them | |
839 | */ | |
5d4f98a2 | 840 | int btrfs_drop_dead_root(struct btrfs_root *root) |
0f7d52f4 | 841 | { |
0f7d52f4 | 842 | struct btrfs_trans_handle *trans; |
5d4f98a2 | 843 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
d3c2fdcf | 844 | unsigned long nr; |
5d4f98a2 | 845 | int ret; |
58176a96 | 846 | |
5d4f98a2 YZ |
847 | while (1) { |
848 | /* | |
849 | * we don't want to jump in and create a bunch of | |
850 | * delayed refs if the transaction is starting to close | |
851 | */ | |
852 | wait_transaction_pre_flush(tree_root->fs_info); | |
853 | trans = btrfs_start_transaction(tree_root, 1); | |
a2135011 | 854 | |
5d4f98a2 YZ |
855 | /* |
856 | * we've joined a transaction, make sure it isn't | |
857 | * closing right now | |
858 | */ | |
859 | if (trans->transaction->delayed_refs.flushing) { | |
860 | btrfs_end_transaction(trans, tree_root); | |
861 | continue; | |
9f3a7427 | 862 | } |
58176a96 | 863 | |
5d4f98a2 YZ |
864 | ret = btrfs_drop_snapshot(trans, root); |
865 | if (ret != -EAGAIN) | |
866 | break; | |
a2135011 | 867 | |
5d4f98a2 YZ |
868 | ret = btrfs_update_root(trans, tree_root, |
869 | &root->root_key, | |
870 | &root->root_item); | |
871 | if (ret) | |
54aa1f4d | 872 | break; |
bcc63abb | 873 | |
d3c2fdcf | 874 | nr = trans->blocks_used; |
0f7d52f4 CM |
875 | ret = btrfs_end_transaction(trans, tree_root); |
876 | BUG_ON(ret); | |
5eda7b5e | 877 | |
d3c2fdcf | 878 | btrfs_btree_balance_dirty(tree_root, nr); |
4dc11904 | 879 | cond_resched(); |
0f7d52f4 | 880 | } |
5d4f98a2 YZ |
881 | BUG_ON(ret); |
882 | ||
883 | ret = btrfs_del_root(trans, tree_root, &root->root_key); | |
884 | BUG_ON(ret); | |
885 | ||
886 | nr = trans->blocks_used; | |
887 | ret = btrfs_end_transaction(trans, tree_root); | |
888 | BUG_ON(ret); | |
889 | ||
890 | free_extent_buffer(root->node); | |
891 | free_extent_buffer(root->commit_root); | |
892 | kfree(root); | |
893 | ||
894 | btrfs_btree_balance_dirty(tree_root, nr); | |
54aa1f4d | 895 | return ret; |
0f7d52f4 | 896 | } |
2c47e605 | 897 | #endif |
0f7d52f4 | 898 | |
d352ac68 CM |
899 | /* |
900 | * new snapshots need to be created at a very specific time in the | |
901 | * transaction commit. This does the actual creation | |
902 | */ | |
80b6794d | 903 | static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, |
3063d29f CM |
904 | struct btrfs_fs_info *fs_info, |
905 | struct btrfs_pending_snapshot *pending) | |
906 | { | |
907 | struct btrfs_key key; | |
80b6794d | 908 | struct btrfs_root_item *new_root_item; |
3063d29f CM |
909 | struct btrfs_root *tree_root = fs_info->tree_root; |
910 | struct btrfs_root *root = pending->root; | |
6bdb72de SW |
911 | struct btrfs_root *parent_root; |
912 | struct inode *parent_inode; | |
6a912213 | 913 | struct dentry *parent; |
a22285a6 | 914 | struct dentry *dentry; |
3063d29f | 915 | struct extent_buffer *tmp; |
925baedd | 916 | struct extent_buffer *old; |
3063d29f | 917 | int ret; |
d68fc57b | 918 | u64 to_reserve = 0; |
6bdb72de | 919 | u64 index = 0; |
a22285a6 | 920 | u64 objectid; |
b83cc969 | 921 | u64 root_flags; |
3063d29f | 922 | |
80b6794d CM |
923 | new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS); |
924 | if (!new_root_item) { | |
a22285a6 | 925 | pending->error = -ENOMEM; |
80b6794d CM |
926 | goto fail; |
927 | } | |
a22285a6 | 928 | |
3063d29f | 929 | ret = btrfs_find_free_objectid(trans, tree_root, 0, &objectid); |
a22285a6 YZ |
930 | if (ret) { |
931 | pending->error = ret; | |
3063d29f | 932 | goto fail; |
a22285a6 | 933 | } |
3063d29f | 934 | |
3fd0a558 | 935 | btrfs_reloc_pre_snapshot(trans, pending, &to_reserve); |
d68fc57b YZ |
936 | btrfs_orphan_pre_snapshot(trans, pending, &to_reserve); |
937 | ||
938 | if (to_reserve > 0) { | |
939 | ret = btrfs_block_rsv_add(trans, root, &pending->block_rsv, | |
8bb8ab2e | 940 | to_reserve); |
d68fc57b YZ |
941 | if (ret) { |
942 | pending->error = ret; | |
943 | goto fail; | |
944 | } | |
945 | } | |
946 | ||
3063d29f | 947 | key.objectid = objectid; |
a22285a6 YZ |
948 | key.offset = (u64)-1; |
949 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3063d29f | 950 | |
a22285a6 | 951 | trans->block_rsv = &pending->block_rsv; |
3de4586c | 952 | |
a22285a6 | 953 | dentry = pending->dentry; |
6a912213 JB |
954 | parent = dget_parent(dentry); |
955 | parent_inode = parent->d_inode; | |
a22285a6 | 956 | parent_root = BTRFS_I(parent_inode)->root; |
6bdb72de | 957 | record_root_in_trans(trans, parent_root); |
a22285a6 | 958 | |
3063d29f CM |
959 | /* |
960 | * insert the directory item | |
961 | */ | |
3de4586c | 962 | ret = btrfs_set_inode_index(parent_inode, &index); |
6bdb72de | 963 | BUG_ON(ret); |
0660b5af | 964 | ret = btrfs_insert_dir_item(trans, parent_root, |
a22285a6 YZ |
965 | dentry->d_name.name, dentry->d_name.len, |
966 | parent_inode->i_ino, &key, | |
967 | BTRFS_FT_DIR, index); | |
6bdb72de | 968 | BUG_ON(ret); |
0660b5af | 969 | |
a22285a6 YZ |
970 | btrfs_i_size_write(parent_inode, parent_inode->i_size + |
971 | dentry->d_name.len * 2); | |
52c26179 YZ |
972 | ret = btrfs_update_inode(trans, parent_root, parent_inode); |
973 | BUG_ON(ret); | |
974 | ||
6bdb72de SW |
975 | record_root_in_trans(trans, root); |
976 | btrfs_set_root_last_snapshot(&root->root_item, trans->transid); | |
977 | memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); | |
978 | ||
b83cc969 LZ |
979 | root_flags = btrfs_root_flags(new_root_item); |
980 | if (pending->readonly) | |
981 | root_flags |= BTRFS_ROOT_SUBVOL_RDONLY; | |
982 | else | |
983 | root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY; | |
984 | btrfs_set_root_flags(new_root_item, root_flags); | |
985 | ||
6bdb72de SW |
986 | old = btrfs_lock_root_node(root); |
987 | btrfs_cow_block(trans, root, old, NULL, 0, &old); | |
988 | btrfs_set_lock_blocking(old); | |
989 | ||
990 | btrfs_copy_root(trans, root, old, &tmp, objectid); | |
991 | btrfs_tree_unlock(old); | |
992 | free_extent_buffer(old); | |
993 | ||
994 | btrfs_set_root_node(new_root_item, tmp); | |
a22285a6 YZ |
995 | /* record when the snapshot was created in key.offset */ |
996 | key.offset = trans->transid; | |
997 | ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); | |
6bdb72de SW |
998 | btrfs_tree_unlock(tmp); |
999 | free_extent_buffer(tmp); | |
a22285a6 | 1000 | BUG_ON(ret); |
6bdb72de | 1001 | |
a22285a6 YZ |
1002 | /* |
1003 | * insert root back/forward references | |
1004 | */ | |
1005 | ret = btrfs_add_root_ref(trans, tree_root, objectid, | |
0660b5af | 1006 | parent_root->root_key.objectid, |
a22285a6 YZ |
1007 | parent_inode->i_ino, index, |
1008 | dentry->d_name.name, dentry->d_name.len); | |
0660b5af | 1009 | BUG_ON(ret); |
6a912213 | 1010 | dput(parent); |
0660b5af | 1011 | |
a22285a6 YZ |
1012 | key.offset = (u64)-1; |
1013 | pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key); | |
1014 | BUG_ON(IS_ERR(pending->snap)); | |
d68fc57b | 1015 | |
3fd0a558 | 1016 | btrfs_reloc_post_snapshot(trans, pending); |
d68fc57b | 1017 | btrfs_orphan_post_snapshot(trans, pending); |
3063d29f | 1018 | fail: |
6bdb72de | 1019 | kfree(new_root_item); |
a22285a6 YZ |
1020 | btrfs_block_rsv_release(root, &pending->block_rsv, (u64)-1); |
1021 | return 0; | |
3063d29f CM |
1022 | } |
1023 | ||
d352ac68 CM |
1024 | /* |
1025 | * create all the snapshots we've scheduled for creation | |
1026 | */ | |
80b6794d CM |
1027 | static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans, |
1028 | struct btrfs_fs_info *fs_info) | |
3de4586c CM |
1029 | { |
1030 | struct btrfs_pending_snapshot *pending; | |
1031 | struct list_head *head = &trans->transaction->pending_snapshots; | |
3de4586c CM |
1032 | int ret; |
1033 | ||
c6e30871 | 1034 | list_for_each_entry(pending, head, list) { |
3de4586c CM |
1035 | ret = create_pending_snapshot(trans, fs_info, pending); |
1036 | BUG_ON(ret); | |
1037 | } | |
1038 | return 0; | |
1039 | } | |
1040 | ||
5d4f98a2 YZ |
1041 | static void update_super_roots(struct btrfs_root *root) |
1042 | { | |
1043 | struct btrfs_root_item *root_item; | |
1044 | struct btrfs_super_block *super; | |
1045 | ||
1046 | super = &root->fs_info->super_copy; | |
1047 | ||
1048 | root_item = &root->fs_info->chunk_root->root_item; | |
1049 | super->chunk_root = root_item->bytenr; | |
1050 | super->chunk_root_generation = root_item->generation; | |
1051 | super->chunk_root_level = root_item->level; | |
1052 | ||
1053 | root_item = &root->fs_info->tree_root->root_item; | |
1054 | super->root = root_item->bytenr; | |
1055 | super->generation = root_item->generation; | |
1056 | super->root_level = root_item->level; | |
0af3d00b JB |
1057 | if (super->cache_generation != 0 || btrfs_test_opt(root, SPACE_CACHE)) |
1058 | super->cache_generation = root_item->generation; | |
5d4f98a2 YZ |
1059 | } |
1060 | ||
f36f3042 CM |
1061 | int btrfs_transaction_in_commit(struct btrfs_fs_info *info) |
1062 | { | |
1063 | int ret = 0; | |
1064 | spin_lock(&info->new_trans_lock); | |
1065 | if (info->running_transaction) | |
1066 | ret = info->running_transaction->in_commit; | |
1067 | spin_unlock(&info->new_trans_lock); | |
1068 | return ret; | |
1069 | } | |
1070 | ||
8929ecfa YZ |
1071 | int btrfs_transaction_blocked(struct btrfs_fs_info *info) |
1072 | { | |
1073 | int ret = 0; | |
1074 | spin_lock(&info->new_trans_lock); | |
1075 | if (info->running_transaction) | |
1076 | ret = info->running_transaction->blocked; | |
1077 | spin_unlock(&info->new_trans_lock); | |
1078 | return ret; | |
1079 | } | |
1080 | ||
bb9c12c9 SW |
1081 | /* |
1082 | * wait for the current transaction commit to start and block subsequent | |
1083 | * transaction joins | |
1084 | */ | |
1085 | static void wait_current_trans_commit_start(struct btrfs_root *root, | |
1086 | struct btrfs_transaction *trans) | |
1087 | { | |
1088 | DEFINE_WAIT(wait); | |
1089 | ||
1090 | if (trans->in_commit) | |
1091 | return; | |
1092 | ||
1093 | while (1) { | |
1094 | prepare_to_wait(&root->fs_info->transaction_blocked_wait, &wait, | |
1095 | TASK_UNINTERRUPTIBLE); | |
1096 | if (trans->in_commit) { | |
1097 | finish_wait(&root->fs_info->transaction_blocked_wait, | |
1098 | &wait); | |
1099 | break; | |
1100 | } | |
1101 | mutex_unlock(&root->fs_info->trans_mutex); | |
1102 | schedule(); | |
1103 | mutex_lock(&root->fs_info->trans_mutex); | |
1104 | finish_wait(&root->fs_info->transaction_blocked_wait, &wait); | |
1105 | } | |
1106 | } | |
1107 | ||
1108 | /* | |
1109 | * wait for the current transaction to start and then become unblocked. | |
1110 | * caller holds ref. | |
1111 | */ | |
1112 | static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root, | |
1113 | struct btrfs_transaction *trans) | |
1114 | { | |
1115 | DEFINE_WAIT(wait); | |
1116 | ||
1117 | if (trans->commit_done || (trans->in_commit && !trans->blocked)) | |
1118 | return; | |
1119 | ||
1120 | while (1) { | |
1121 | prepare_to_wait(&root->fs_info->transaction_wait, &wait, | |
1122 | TASK_UNINTERRUPTIBLE); | |
1123 | if (trans->commit_done || | |
1124 | (trans->in_commit && !trans->blocked)) { | |
1125 | finish_wait(&root->fs_info->transaction_wait, | |
1126 | &wait); | |
1127 | break; | |
1128 | } | |
1129 | mutex_unlock(&root->fs_info->trans_mutex); | |
1130 | schedule(); | |
1131 | mutex_lock(&root->fs_info->trans_mutex); | |
1132 | finish_wait(&root->fs_info->transaction_wait, | |
1133 | &wait); | |
1134 | } | |
1135 | } | |
1136 | ||
1137 | /* | |
1138 | * commit transactions asynchronously. once btrfs_commit_transaction_async | |
1139 | * returns, any subsequent transaction will not be allowed to join. | |
1140 | */ | |
1141 | struct btrfs_async_commit { | |
1142 | struct btrfs_trans_handle *newtrans; | |
1143 | struct btrfs_root *root; | |
1144 | struct delayed_work work; | |
1145 | }; | |
1146 | ||
1147 | static void do_async_commit(struct work_struct *work) | |
1148 | { | |
1149 | struct btrfs_async_commit *ac = | |
1150 | container_of(work, struct btrfs_async_commit, work.work); | |
1151 | ||
1152 | btrfs_commit_transaction(ac->newtrans, ac->root); | |
1153 | kfree(ac); | |
1154 | } | |
1155 | ||
1156 | int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans, | |
1157 | struct btrfs_root *root, | |
1158 | int wait_for_unblock) | |
1159 | { | |
1160 | struct btrfs_async_commit *ac; | |
1161 | struct btrfs_transaction *cur_trans; | |
1162 | ||
1163 | ac = kmalloc(sizeof(*ac), GFP_NOFS); | |
db5b493a TI |
1164 | if (!ac) |
1165 | return -ENOMEM; | |
bb9c12c9 SW |
1166 | |
1167 | INIT_DELAYED_WORK(&ac->work, do_async_commit); | |
1168 | ac->root = root; | |
1169 | ac->newtrans = btrfs_join_transaction(root, 0); | |
3612b495 TI |
1170 | if (IS_ERR(ac->newtrans)) { |
1171 | int err = PTR_ERR(ac->newtrans); | |
1172 | kfree(ac); | |
1173 | return err; | |
1174 | } | |
bb9c12c9 SW |
1175 | |
1176 | /* take transaction reference */ | |
1177 | mutex_lock(&root->fs_info->trans_mutex); | |
1178 | cur_trans = trans->transaction; | |
1179 | cur_trans->use_count++; | |
1180 | mutex_unlock(&root->fs_info->trans_mutex); | |
1181 | ||
1182 | btrfs_end_transaction(trans, root); | |
1183 | schedule_delayed_work(&ac->work, 0); | |
1184 | ||
1185 | /* wait for transaction to start and unblock */ | |
1186 | mutex_lock(&root->fs_info->trans_mutex); | |
1187 | if (wait_for_unblock) | |
1188 | wait_current_trans_commit_start_and_unblock(root, cur_trans); | |
1189 | else | |
1190 | wait_current_trans_commit_start(root, cur_trans); | |
1191 | put_transaction(cur_trans); | |
1192 | mutex_unlock(&root->fs_info->trans_mutex); | |
1193 | ||
1194 | return 0; | |
1195 | } | |
1196 | ||
1197 | /* | |
1198 | * btrfs_transaction state sequence: | |
1199 | * in_commit = 0, blocked = 0 (initial) | |
1200 | * in_commit = 1, blocked = 1 | |
1201 | * blocked = 0 | |
1202 | * commit_done = 1 | |
1203 | */ | |
79154b1b CM |
1204 | int btrfs_commit_transaction(struct btrfs_trans_handle *trans, |
1205 | struct btrfs_root *root) | |
1206 | { | |
15ee9bc7 | 1207 | unsigned long joined = 0; |
79154b1b | 1208 | struct btrfs_transaction *cur_trans; |
8fd17795 | 1209 | struct btrfs_transaction *prev_trans = NULL; |
79154b1b | 1210 | DEFINE_WAIT(wait); |
15ee9bc7 | 1211 | int ret; |
89573b9c CM |
1212 | int should_grow = 0; |
1213 | unsigned long now = get_seconds(); | |
dccae999 | 1214 | int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT); |
79154b1b | 1215 | |
5a3f23d5 CM |
1216 | btrfs_run_ordered_operations(root, 0); |
1217 | ||
56bec294 CM |
1218 | /* make a pass through all the delayed refs we have so far |
1219 | * any runnings procs may add more while we are here | |
1220 | */ | |
1221 | ret = btrfs_run_delayed_refs(trans, root, 0); | |
1222 | BUG_ON(ret); | |
1223 | ||
a22285a6 YZ |
1224 | btrfs_trans_release_metadata(trans, root); |
1225 | ||
b7ec40d7 | 1226 | cur_trans = trans->transaction; |
56bec294 CM |
1227 | /* |
1228 | * set the flushing flag so procs in this transaction have to | |
1229 | * start sending their work down. | |
1230 | */ | |
b7ec40d7 | 1231 | cur_trans->delayed_refs.flushing = 1; |
56bec294 | 1232 | |
c3e69d58 | 1233 | ret = btrfs_run_delayed_refs(trans, root, 0); |
56bec294 CM |
1234 | BUG_ON(ret); |
1235 | ||
79154b1b | 1236 | mutex_lock(&root->fs_info->trans_mutex); |
b7ec40d7 CM |
1237 | if (cur_trans->in_commit) { |
1238 | cur_trans->use_count++; | |
ccd467d6 | 1239 | mutex_unlock(&root->fs_info->trans_mutex); |
79154b1b | 1240 | btrfs_end_transaction(trans, root); |
ccd467d6 | 1241 | |
79154b1b CM |
1242 | ret = wait_for_commit(root, cur_trans); |
1243 | BUG_ON(ret); | |
15ee9bc7 JB |
1244 | |
1245 | mutex_lock(&root->fs_info->trans_mutex); | |
79154b1b | 1246 | put_transaction(cur_trans); |
15ee9bc7 JB |
1247 | mutex_unlock(&root->fs_info->trans_mutex); |
1248 | ||
79154b1b CM |
1249 | return 0; |
1250 | } | |
4313b399 | 1251 | |
2c90e5d6 | 1252 | trans->transaction->in_commit = 1; |
f9295749 | 1253 | trans->transaction->blocked = 1; |
bb9c12c9 SW |
1254 | wake_up(&root->fs_info->transaction_blocked_wait); |
1255 | ||
ccd467d6 CM |
1256 | if (cur_trans->list.prev != &root->fs_info->trans_list) { |
1257 | prev_trans = list_entry(cur_trans->list.prev, | |
1258 | struct btrfs_transaction, list); | |
1259 | if (!prev_trans->commit_done) { | |
1260 | prev_trans->use_count++; | |
ccd467d6 CM |
1261 | mutex_unlock(&root->fs_info->trans_mutex); |
1262 | ||
1263 | wait_for_commit(root, prev_trans); | |
ccd467d6 | 1264 | |
ccd467d6 | 1265 | mutex_lock(&root->fs_info->trans_mutex); |
15ee9bc7 | 1266 | put_transaction(prev_trans); |
ccd467d6 CM |
1267 | } |
1268 | } | |
15ee9bc7 | 1269 | |
89573b9c CM |
1270 | if (now < cur_trans->start_time || now - cur_trans->start_time < 1) |
1271 | should_grow = 1; | |
1272 | ||
15ee9bc7 | 1273 | do { |
7ea394f1 | 1274 | int snap_pending = 0; |
15ee9bc7 | 1275 | joined = cur_trans->num_joined; |
7ea394f1 YZ |
1276 | if (!list_empty(&trans->transaction->pending_snapshots)) |
1277 | snap_pending = 1; | |
1278 | ||
2c90e5d6 | 1279 | WARN_ON(cur_trans != trans->transaction); |
79154b1b | 1280 | mutex_unlock(&root->fs_info->trans_mutex); |
15ee9bc7 | 1281 | |
0bdb1db2 | 1282 | if (flush_on_commit || snap_pending) { |
24bbcf04 YZ |
1283 | btrfs_start_delalloc_inodes(root, 1); |
1284 | ret = btrfs_wait_ordered_extents(root, 0, 1); | |
ebecd3d9 | 1285 | BUG_ON(ret); |
7ea394f1 YZ |
1286 | } |
1287 | ||
5a3f23d5 CM |
1288 | /* |
1289 | * rename don't use btrfs_join_transaction, so, once we | |
1290 | * set the transaction to blocked above, we aren't going | |
1291 | * to get any new ordered operations. We can safely run | |
1292 | * it here and no for sure that nothing new will be added | |
1293 | * to the list | |
1294 | */ | |
1295 | btrfs_run_ordered_operations(root, 1); | |
1296 | ||
ed3b3d31 CM |
1297 | prepare_to_wait(&cur_trans->writer_wait, &wait, |
1298 | TASK_UNINTERRUPTIBLE); | |
1299 | ||
89573b9c | 1300 | smp_mb(); |
99d16cbc SW |
1301 | if (cur_trans->num_writers > 1) |
1302 | schedule_timeout(MAX_SCHEDULE_TIMEOUT); | |
1303 | else if (should_grow) | |
1304 | schedule_timeout(1); | |
15ee9bc7 | 1305 | |
79154b1b | 1306 | mutex_lock(&root->fs_info->trans_mutex); |
15ee9bc7 JB |
1307 | finish_wait(&cur_trans->writer_wait, &wait); |
1308 | } while (cur_trans->num_writers > 1 || | |
89573b9c | 1309 | (should_grow && cur_trans->num_joined != joined)); |
15ee9bc7 | 1310 | |
3063d29f CM |
1311 | ret = create_pending_snapshots(trans, root->fs_info); |
1312 | BUG_ON(ret); | |
1313 | ||
56bec294 CM |
1314 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
1315 | BUG_ON(ret); | |
1316 | ||
2c90e5d6 | 1317 | WARN_ON(cur_trans != trans->transaction); |
dc17ff8f | 1318 | |
e02119d5 CM |
1319 | /* btrfs_commit_tree_roots is responsible for getting the |
1320 | * various roots consistent with each other. Every pointer | |
1321 | * in the tree of tree roots has to point to the most up to date | |
1322 | * root for every subvolume and other tree. So, we have to keep | |
1323 | * the tree logging code from jumping in and changing any | |
1324 | * of the trees. | |
1325 | * | |
1326 | * At this point in the commit, there can't be any tree-log | |
1327 | * writers, but a little lower down we drop the trans mutex | |
1328 | * and let new people in. By holding the tree_log_mutex | |
1329 | * from now until after the super is written, we avoid races | |
1330 | * with the tree-log code. | |
1331 | */ | |
1332 | mutex_lock(&root->fs_info->tree_log_mutex); | |
1333 | ||
5d4f98a2 | 1334 | ret = commit_fs_roots(trans, root); |
54aa1f4d CM |
1335 | BUG_ON(ret); |
1336 | ||
5d4f98a2 | 1337 | /* commit_fs_roots gets rid of all the tree log roots, it is now |
e02119d5 CM |
1338 | * safe to free the root of tree log roots |
1339 | */ | |
1340 | btrfs_free_log_root_tree(trans, root->fs_info); | |
1341 | ||
5d4f98a2 | 1342 | ret = commit_cowonly_roots(trans, root); |
79154b1b | 1343 | BUG_ON(ret); |
54aa1f4d | 1344 | |
11833d66 YZ |
1345 | btrfs_prepare_extent_commit(trans, root); |
1346 | ||
78fae27e | 1347 | cur_trans = root->fs_info->running_transaction; |
cee36a03 | 1348 | spin_lock(&root->fs_info->new_trans_lock); |
78fae27e | 1349 | root->fs_info->running_transaction = NULL; |
cee36a03 | 1350 | spin_unlock(&root->fs_info->new_trans_lock); |
5d4f98a2 YZ |
1351 | |
1352 | btrfs_set_root_node(&root->fs_info->tree_root->root_item, | |
1353 | root->fs_info->tree_root->node); | |
817d52f8 | 1354 | switch_commit_root(root->fs_info->tree_root); |
5d4f98a2 YZ |
1355 | |
1356 | btrfs_set_root_node(&root->fs_info->chunk_root->root_item, | |
1357 | root->fs_info->chunk_root->node); | |
817d52f8 | 1358 | switch_commit_root(root->fs_info->chunk_root); |
5d4f98a2 YZ |
1359 | |
1360 | update_super_roots(root); | |
e02119d5 CM |
1361 | |
1362 | if (!root->fs_info->log_root_recovering) { | |
1363 | btrfs_set_super_log_root(&root->fs_info->super_copy, 0); | |
1364 | btrfs_set_super_log_root_level(&root->fs_info->super_copy, 0); | |
1365 | } | |
1366 | ||
a061fc8d CM |
1367 | memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy, |
1368 | sizeof(root->fs_info->super_copy)); | |
ccd467d6 | 1369 | |
f9295749 | 1370 | trans->transaction->blocked = 0; |
b7ec40d7 | 1371 | |
f9295749 | 1372 | wake_up(&root->fs_info->transaction_wait); |
e6dcd2dc | 1373 | |
78fae27e | 1374 | mutex_unlock(&root->fs_info->trans_mutex); |
79154b1b CM |
1375 | ret = btrfs_write_and_wait_transaction(trans, root); |
1376 | BUG_ON(ret); | |
a512bbf8 | 1377 | write_ctree_super(trans, root, 0); |
4313b399 | 1378 | |
e02119d5 CM |
1379 | /* |
1380 | * the super is written, we can safely allow the tree-loggers | |
1381 | * to go about their business | |
1382 | */ | |
1383 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
1384 | ||
11833d66 | 1385 | btrfs_finish_extent_commit(trans, root); |
4313b399 | 1386 | |
1a40e23b ZY |
1387 | mutex_lock(&root->fs_info->trans_mutex); |
1388 | ||
2c90e5d6 | 1389 | cur_trans->commit_done = 1; |
b7ec40d7 | 1390 | |
15ee9bc7 | 1391 | root->fs_info->last_trans_committed = cur_trans->transid; |
817d52f8 | 1392 | |
2c90e5d6 | 1393 | wake_up(&cur_trans->commit_wait); |
3de4586c | 1394 | |
78fae27e | 1395 | put_transaction(cur_trans); |
79154b1b | 1396 | put_transaction(cur_trans); |
58176a96 | 1397 | |
1abe9b8a | 1398 | trace_btrfs_transaction_commit(root); |
1399 | ||
78fae27e | 1400 | mutex_unlock(&root->fs_info->trans_mutex); |
3de4586c | 1401 | |
9ed74f2d JB |
1402 | if (current->journal_info == trans) |
1403 | current->journal_info = NULL; | |
1404 | ||
2c90e5d6 | 1405 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
24bbcf04 YZ |
1406 | |
1407 | if (current != root->fs_info->transaction_kthread) | |
1408 | btrfs_run_delayed_iputs(root); | |
1409 | ||
79154b1b CM |
1410 | return ret; |
1411 | } | |
1412 | ||
d352ac68 CM |
1413 | /* |
1414 | * interface function to delete all the snapshots we have scheduled for deletion | |
1415 | */ | |
e9d0b13b CM |
1416 | int btrfs_clean_old_snapshots(struct btrfs_root *root) |
1417 | { | |
5d4f98a2 YZ |
1418 | LIST_HEAD(list); |
1419 | struct btrfs_fs_info *fs_info = root->fs_info; | |
1420 | ||
1421 | mutex_lock(&fs_info->trans_mutex); | |
1422 | list_splice_init(&fs_info->dead_roots, &list); | |
1423 | mutex_unlock(&fs_info->trans_mutex); | |
e9d0b13b | 1424 | |
5d4f98a2 YZ |
1425 | while (!list_empty(&list)) { |
1426 | root = list_entry(list.next, struct btrfs_root, root_list); | |
76dda93c YZ |
1427 | list_del(&root->root_list); |
1428 | ||
1429 | if (btrfs_header_backref_rev(root->node) < | |
1430 | BTRFS_MIXED_BACKREF_REV) | |
3fd0a558 | 1431 | btrfs_drop_snapshot(root, NULL, 0); |
76dda93c | 1432 | else |
3fd0a558 | 1433 | btrfs_drop_snapshot(root, NULL, 1); |
e9d0b13b CM |
1434 | } |
1435 | return 0; | |
1436 | } |