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 | 199 | if (ret < 0) { |
6e8df2ae | 200 | kmem_cache_free(btrfs_trans_handle_cachep, h); |
db5b493a TI |
201 | if (type != TRANS_JOIN_NOLOCK) |
202 | mutex_unlock(&root->fs_info->trans_mutex); | |
203 | return ERR_PTR(ret); | |
204 | } | |
0f7d52f4 | 205 | |
a22285a6 YZ |
206 | cur_trans = root->fs_info->running_transaction; |
207 | cur_trans->use_count++; | |
0af3d00b JB |
208 | if (type != TRANS_JOIN_NOLOCK) |
209 | mutex_unlock(&root->fs_info->trans_mutex); | |
a22285a6 YZ |
210 | |
211 | h->transid = cur_trans->transid; | |
212 | h->transaction = cur_trans; | |
79154b1b | 213 | h->blocks_used = 0; |
d2fb3437 | 214 | h->block_group = 0; |
a22285a6 | 215 | h->bytes_reserved = 0; |
56bec294 | 216 | h->delayed_ref_updates = 0; |
f0486c68 | 217 | h->block_rsv = NULL; |
b7ec40d7 | 218 | |
a22285a6 YZ |
219 | smp_mb(); |
220 | if (cur_trans->blocked && may_wait_transaction(root, type)) { | |
221 | btrfs_commit_transaction(h, root); | |
222 | goto again; | |
223 | } | |
224 | ||
225 | if (num_items > 0) { | |
8bb8ab2e | 226 | ret = btrfs_trans_reserve_metadata(h, root, num_items); |
a22285a6 YZ |
227 | if (ret == -EAGAIN) { |
228 | btrfs_commit_transaction(h, root); | |
229 | goto again; | |
230 | } | |
231 | if (ret < 0) { | |
232 | btrfs_end_transaction(h, root); | |
233 | return ERR_PTR(ret); | |
234 | } | |
235 | } | |
9ed74f2d | 236 | |
0af3d00b JB |
237 | if (type != TRANS_JOIN_NOLOCK) |
238 | mutex_lock(&root->fs_info->trans_mutex); | |
5d4f98a2 | 239 | record_root_in_trans(h, root); |
0af3d00b JB |
240 | if (type != TRANS_JOIN_NOLOCK) |
241 | mutex_unlock(&root->fs_info->trans_mutex); | |
a22285a6 YZ |
242 | |
243 | if (!current->journal_info && type != TRANS_USERSPACE) | |
244 | current->journal_info = h; | |
79154b1b CM |
245 | return h; |
246 | } | |
247 | ||
f9295749 | 248 | struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, |
a22285a6 | 249 | int num_items) |
f9295749 | 250 | { |
a22285a6 | 251 | return start_transaction(root, num_items, TRANS_START); |
f9295749 CM |
252 | } |
253 | struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root, | |
254 | int num_blocks) | |
255 | { | |
a22285a6 | 256 | return start_transaction(root, 0, TRANS_JOIN); |
f9295749 CM |
257 | } |
258 | ||
0af3d00b JB |
259 | struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root, |
260 | int num_blocks) | |
261 | { | |
262 | return start_transaction(root, 0, TRANS_JOIN_NOLOCK); | |
263 | } | |
264 | ||
9ca9ee09 SW |
265 | struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *r, |
266 | int num_blocks) | |
267 | { | |
a22285a6 | 268 | return start_transaction(r, 0, TRANS_USERSPACE); |
9ca9ee09 SW |
269 | } |
270 | ||
d352ac68 | 271 | /* wait for a transaction commit to be fully complete */ |
89ce8a63 CM |
272 | static noinline int wait_for_commit(struct btrfs_root *root, |
273 | struct btrfs_transaction *commit) | |
274 | { | |
275 | DEFINE_WAIT(wait); | |
276 | mutex_lock(&root->fs_info->trans_mutex); | |
d397712b | 277 | while (!commit->commit_done) { |
89ce8a63 CM |
278 | prepare_to_wait(&commit->commit_wait, &wait, |
279 | TASK_UNINTERRUPTIBLE); | |
280 | if (commit->commit_done) | |
281 | break; | |
282 | mutex_unlock(&root->fs_info->trans_mutex); | |
283 | schedule(); | |
284 | mutex_lock(&root->fs_info->trans_mutex); | |
285 | } | |
286 | mutex_unlock(&root->fs_info->trans_mutex); | |
287 | finish_wait(&commit->commit_wait, &wait); | |
288 | return 0; | |
289 | } | |
290 | ||
46204592 SW |
291 | int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid) |
292 | { | |
293 | struct btrfs_transaction *cur_trans = NULL, *t; | |
294 | int ret; | |
295 | ||
296 | mutex_lock(&root->fs_info->trans_mutex); | |
297 | ||
298 | ret = 0; | |
299 | if (transid) { | |
300 | if (transid <= root->fs_info->last_trans_committed) | |
301 | goto out_unlock; | |
302 | ||
303 | /* find specified transaction */ | |
304 | list_for_each_entry(t, &root->fs_info->trans_list, list) { | |
305 | if (t->transid == transid) { | |
306 | cur_trans = t; | |
307 | break; | |
308 | } | |
309 | if (t->transid > transid) | |
310 | break; | |
311 | } | |
312 | ret = -EINVAL; | |
313 | if (!cur_trans) | |
314 | goto out_unlock; /* bad transid */ | |
315 | } else { | |
316 | /* find newest transaction that is committing | committed */ | |
317 | list_for_each_entry_reverse(t, &root->fs_info->trans_list, | |
318 | list) { | |
319 | if (t->in_commit) { | |
320 | if (t->commit_done) | |
321 | goto out_unlock; | |
322 | cur_trans = t; | |
323 | break; | |
324 | } | |
325 | } | |
326 | if (!cur_trans) | |
327 | goto out_unlock; /* nothing committing|committed */ | |
328 | } | |
329 | ||
330 | cur_trans->use_count++; | |
331 | mutex_unlock(&root->fs_info->trans_mutex); | |
332 | ||
333 | wait_for_commit(root, cur_trans); | |
334 | ||
335 | mutex_lock(&root->fs_info->trans_mutex); | |
336 | put_transaction(cur_trans); | |
337 | ret = 0; | |
338 | out_unlock: | |
339 | mutex_unlock(&root->fs_info->trans_mutex); | |
340 | return ret; | |
341 | } | |
342 | ||
5d4f98a2 | 343 | #if 0 |
d352ac68 | 344 | /* |
d397712b CM |
345 | * rate limit against the drop_snapshot code. This helps to slow down new |
346 | * operations if the drop_snapshot code isn't able to keep up. | |
d352ac68 | 347 | */ |
37d1aeee | 348 | static void throttle_on_drops(struct btrfs_root *root) |
ab78c84d CM |
349 | { |
350 | struct btrfs_fs_info *info = root->fs_info; | |
2dd3e67b | 351 | int harder_count = 0; |
ab78c84d | 352 | |
2dd3e67b | 353 | harder: |
ab78c84d CM |
354 | if (atomic_read(&info->throttles)) { |
355 | DEFINE_WAIT(wait); | |
356 | int thr; | |
ab78c84d CM |
357 | thr = atomic_read(&info->throttle_gen); |
358 | ||
359 | do { | |
360 | prepare_to_wait(&info->transaction_throttle, | |
361 | &wait, TASK_UNINTERRUPTIBLE); | |
362 | if (!atomic_read(&info->throttles)) { | |
363 | finish_wait(&info->transaction_throttle, &wait); | |
364 | break; | |
365 | } | |
366 | schedule(); | |
367 | finish_wait(&info->transaction_throttle, &wait); | |
368 | } while (thr == atomic_read(&info->throttle_gen)); | |
2dd3e67b CM |
369 | harder_count++; |
370 | ||
371 | if (root->fs_info->total_ref_cache_size > 1 * 1024 * 1024 && | |
372 | harder_count < 2) | |
373 | goto harder; | |
374 | ||
375 | if (root->fs_info->total_ref_cache_size > 5 * 1024 * 1024 && | |
376 | harder_count < 10) | |
377 | goto harder; | |
378 | ||
379 | if (root->fs_info->total_ref_cache_size > 10 * 1024 * 1024 && | |
380 | harder_count < 20) | |
381 | goto harder; | |
ab78c84d CM |
382 | } |
383 | } | |
5d4f98a2 | 384 | #endif |
ab78c84d | 385 | |
37d1aeee CM |
386 | void btrfs_throttle(struct btrfs_root *root) |
387 | { | |
388 | mutex_lock(&root->fs_info->trans_mutex); | |
9ca9ee09 SW |
389 | if (!root->fs_info->open_ioctl_trans) |
390 | wait_current_trans(root); | |
37d1aeee | 391 | mutex_unlock(&root->fs_info->trans_mutex); |
37d1aeee CM |
392 | } |
393 | ||
8929ecfa YZ |
394 | static int should_end_transaction(struct btrfs_trans_handle *trans, |
395 | struct btrfs_root *root) | |
396 | { | |
397 | int ret; | |
398 | ret = btrfs_block_rsv_check(trans, root, | |
399 | &root->fs_info->global_block_rsv, 0, 5); | |
400 | return ret ? 1 : 0; | |
401 | } | |
402 | ||
403 | int btrfs_should_end_transaction(struct btrfs_trans_handle *trans, | |
404 | struct btrfs_root *root) | |
405 | { | |
406 | struct btrfs_transaction *cur_trans = trans->transaction; | |
407 | int updates; | |
408 | ||
409 | if (cur_trans->blocked || cur_trans->delayed_refs.flushing) | |
410 | return 1; | |
411 | ||
412 | updates = trans->delayed_ref_updates; | |
413 | trans->delayed_ref_updates = 0; | |
414 | if (updates) | |
415 | btrfs_run_delayed_refs(trans, root, updates); | |
416 | ||
417 | return should_end_transaction(trans, root); | |
418 | } | |
419 | ||
89ce8a63 | 420 | static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, |
0af3d00b | 421 | struct btrfs_root *root, int throttle, int lock) |
79154b1b | 422 | { |
8929ecfa | 423 | struct btrfs_transaction *cur_trans = trans->transaction; |
ab78c84d | 424 | struct btrfs_fs_info *info = root->fs_info; |
c3e69d58 CM |
425 | int count = 0; |
426 | ||
427 | while (count < 4) { | |
428 | unsigned long cur = trans->delayed_ref_updates; | |
429 | trans->delayed_ref_updates = 0; | |
430 | if (cur && | |
431 | trans->transaction->delayed_refs.num_heads_ready > 64) { | |
432 | trans->delayed_ref_updates = 0; | |
b7ec40d7 CM |
433 | |
434 | /* | |
435 | * do a full flush if the transaction is trying | |
436 | * to close | |
437 | */ | |
438 | if (trans->transaction->delayed_refs.flushing) | |
439 | cur = 0; | |
c3e69d58 CM |
440 | btrfs_run_delayed_refs(trans, root, cur); |
441 | } else { | |
442 | break; | |
443 | } | |
444 | count++; | |
56bec294 CM |
445 | } |
446 | ||
a22285a6 YZ |
447 | btrfs_trans_release_metadata(trans, root); |
448 | ||
0af3d00b | 449 | if (lock && !root->fs_info->open_ioctl_trans && |
8929ecfa YZ |
450 | should_end_transaction(trans, root)) |
451 | trans->transaction->blocked = 1; | |
452 | ||
0af3d00b | 453 | if (lock && cur_trans->blocked && !cur_trans->in_commit) { |
8929ecfa YZ |
454 | if (throttle) |
455 | return btrfs_commit_transaction(trans, root); | |
456 | else | |
457 | wake_up_process(info->transaction_kthread); | |
458 | } | |
459 | ||
0af3d00b JB |
460 | if (lock) |
461 | mutex_lock(&info->trans_mutex); | |
8929ecfa | 462 | WARN_ON(cur_trans != info->running_transaction); |
d5719762 | 463 | WARN_ON(cur_trans->num_writers < 1); |
ccd467d6 | 464 | cur_trans->num_writers--; |
89ce8a63 | 465 | |
99d16cbc | 466 | smp_mb(); |
79154b1b CM |
467 | if (waitqueue_active(&cur_trans->writer_wait)) |
468 | wake_up(&cur_trans->writer_wait); | |
79154b1b | 469 | put_transaction(cur_trans); |
0af3d00b JB |
470 | if (lock) |
471 | mutex_unlock(&info->trans_mutex); | |
9ed74f2d JB |
472 | |
473 | if (current->journal_info == trans) | |
474 | current->journal_info = NULL; | |
d6025579 | 475 | memset(trans, 0, sizeof(*trans)); |
2c90e5d6 | 476 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
ab78c84d | 477 | |
24bbcf04 YZ |
478 | if (throttle) |
479 | btrfs_run_delayed_iputs(root); | |
480 | ||
79154b1b CM |
481 | return 0; |
482 | } | |
483 | ||
89ce8a63 CM |
484 | int btrfs_end_transaction(struct btrfs_trans_handle *trans, |
485 | struct btrfs_root *root) | |
486 | { | |
0af3d00b | 487 | return __btrfs_end_transaction(trans, root, 0, 1); |
89ce8a63 CM |
488 | } |
489 | ||
490 | int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans, | |
491 | struct btrfs_root *root) | |
492 | { | |
0af3d00b JB |
493 | return __btrfs_end_transaction(trans, root, 1, 1); |
494 | } | |
495 | ||
496 | int btrfs_end_transaction_nolock(struct btrfs_trans_handle *trans, | |
497 | struct btrfs_root *root) | |
498 | { | |
499 | return __btrfs_end_transaction(trans, root, 0, 0); | |
89ce8a63 CM |
500 | } |
501 | ||
d352ac68 CM |
502 | /* |
503 | * when btree blocks are allocated, they have some corresponding bits set for | |
504 | * them in one of two extent_io trees. This is used to make sure all of | |
690587d1 | 505 | * those extents are sent to disk but does not wait on them |
d352ac68 | 506 | */ |
690587d1 | 507 | int btrfs_write_marked_extents(struct btrfs_root *root, |
8cef4e16 | 508 | struct extent_io_tree *dirty_pages, int mark) |
79154b1b | 509 | { |
7c4452b9 | 510 | int ret; |
777e6bd7 | 511 | int err = 0; |
7c4452b9 CM |
512 | int werr = 0; |
513 | struct page *page; | |
7c4452b9 | 514 | struct inode *btree_inode = root->fs_info->btree_inode; |
777e6bd7 | 515 | u64 start = 0; |
5f39d397 CM |
516 | u64 end; |
517 | unsigned long index; | |
7c4452b9 | 518 | |
d397712b | 519 | while (1) { |
777e6bd7 | 520 | ret = find_first_extent_bit(dirty_pages, start, &start, &end, |
8cef4e16 | 521 | mark); |
5f39d397 | 522 | if (ret) |
7c4452b9 | 523 | break; |
d397712b | 524 | while (start <= end) { |
777e6bd7 CM |
525 | cond_resched(); |
526 | ||
5f39d397 | 527 | index = start >> PAGE_CACHE_SHIFT; |
35ebb934 | 528 | start = (u64)(index + 1) << PAGE_CACHE_SHIFT; |
4bef0848 | 529 | page = find_get_page(btree_inode->i_mapping, index); |
7c4452b9 CM |
530 | if (!page) |
531 | continue; | |
4bef0848 CM |
532 | |
533 | btree_lock_page_hook(page); | |
534 | if (!page->mapping) { | |
535 | unlock_page(page); | |
536 | page_cache_release(page); | |
537 | continue; | |
538 | } | |
539 | ||
6702ed49 CM |
540 | if (PageWriteback(page)) { |
541 | if (PageDirty(page)) | |
542 | wait_on_page_writeback(page); | |
543 | else { | |
544 | unlock_page(page); | |
545 | page_cache_release(page); | |
546 | continue; | |
547 | } | |
548 | } | |
7c4452b9 CM |
549 | err = write_one_page(page, 0); |
550 | if (err) | |
551 | werr = err; | |
552 | page_cache_release(page); | |
553 | } | |
554 | } | |
690587d1 CM |
555 | if (err) |
556 | werr = err; | |
557 | return werr; | |
558 | } | |
559 | ||
560 | /* | |
561 | * when btree blocks are allocated, they have some corresponding bits set for | |
562 | * them in one of two extent_io trees. This is used to make sure all of | |
563 | * those extents are on disk for transaction or log commit. We wait | |
564 | * on all the pages and clear them from the dirty pages state tree | |
565 | */ | |
566 | int btrfs_wait_marked_extents(struct btrfs_root *root, | |
8cef4e16 | 567 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 CM |
568 | { |
569 | int ret; | |
570 | int err = 0; | |
571 | int werr = 0; | |
572 | struct page *page; | |
573 | struct inode *btree_inode = root->fs_info->btree_inode; | |
574 | u64 start = 0; | |
575 | u64 end; | |
576 | unsigned long index; | |
577 | ||
d397712b | 578 | while (1) { |
8cef4e16 YZ |
579 | ret = find_first_extent_bit(dirty_pages, start, &start, &end, |
580 | mark); | |
777e6bd7 CM |
581 | if (ret) |
582 | break; | |
583 | ||
8cef4e16 | 584 | clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS); |
d397712b | 585 | while (start <= end) { |
777e6bd7 CM |
586 | index = start >> PAGE_CACHE_SHIFT; |
587 | start = (u64)(index + 1) << PAGE_CACHE_SHIFT; | |
588 | page = find_get_page(btree_inode->i_mapping, index); | |
589 | if (!page) | |
590 | continue; | |
591 | if (PageDirty(page)) { | |
4bef0848 CM |
592 | btree_lock_page_hook(page); |
593 | wait_on_page_writeback(page); | |
777e6bd7 CM |
594 | err = write_one_page(page, 0); |
595 | if (err) | |
596 | werr = err; | |
597 | } | |
105d931d | 598 | wait_on_page_writeback(page); |
777e6bd7 CM |
599 | page_cache_release(page); |
600 | cond_resched(); | |
601 | } | |
602 | } | |
7c4452b9 CM |
603 | if (err) |
604 | werr = err; | |
605 | return werr; | |
79154b1b CM |
606 | } |
607 | ||
690587d1 CM |
608 | /* |
609 | * when btree blocks are allocated, they have some corresponding bits set for | |
610 | * them in one of two extent_io trees. This is used to make sure all of | |
611 | * those extents are on disk for transaction or log commit | |
612 | */ | |
613 | int btrfs_write_and_wait_marked_extents(struct btrfs_root *root, | |
8cef4e16 | 614 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 CM |
615 | { |
616 | int ret; | |
617 | int ret2; | |
618 | ||
8cef4e16 YZ |
619 | ret = btrfs_write_marked_extents(root, dirty_pages, mark); |
620 | ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark); | |
690587d1 CM |
621 | return ret || ret2; |
622 | } | |
623 | ||
d0c803c4 CM |
624 | int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans, |
625 | struct btrfs_root *root) | |
626 | { | |
627 | if (!trans || !trans->transaction) { | |
628 | struct inode *btree_inode; | |
629 | btree_inode = root->fs_info->btree_inode; | |
630 | return filemap_write_and_wait(btree_inode->i_mapping); | |
631 | } | |
632 | return btrfs_write_and_wait_marked_extents(root, | |
8cef4e16 YZ |
633 | &trans->transaction->dirty_pages, |
634 | EXTENT_DIRTY); | |
d0c803c4 CM |
635 | } |
636 | ||
d352ac68 CM |
637 | /* |
638 | * this is used to update the root pointer in the tree of tree roots. | |
639 | * | |
640 | * But, in the case of the extent allocation tree, updating the root | |
641 | * pointer may allocate blocks which may change the root of the extent | |
642 | * allocation tree. | |
643 | * | |
644 | * So, this loops and repeats and makes sure the cowonly root didn't | |
645 | * change while the root pointer was being updated in the metadata. | |
646 | */ | |
0b86a832 CM |
647 | static int update_cowonly_root(struct btrfs_trans_handle *trans, |
648 | struct btrfs_root *root) | |
79154b1b CM |
649 | { |
650 | int ret; | |
0b86a832 | 651 | u64 old_root_bytenr; |
86b9f2ec | 652 | u64 old_root_used; |
0b86a832 | 653 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
79154b1b | 654 | |
86b9f2ec | 655 | old_root_used = btrfs_root_used(&root->root_item); |
0b86a832 | 656 | btrfs_write_dirty_block_groups(trans, root); |
56bec294 | 657 | |
d397712b | 658 | while (1) { |
0b86a832 | 659 | old_root_bytenr = btrfs_root_bytenr(&root->root_item); |
86b9f2ec YZ |
660 | if (old_root_bytenr == root->node->start && |
661 | old_root_used == btrfs_root_used(&root->root_item)) | |
79154b1b | 662 | break; |
87ef2bb4 | 663 | |
5d4f98a2 | 664 | btrfs_set_root_node(&root->root_item, root->node); |
79154b1b | 665 | ret = btrfs_update_root(trans, tree_root, |
0b86a832 CM |
666 | &root->root_key, |
667 | &root->root_item); | |
79154b1b | 668 | BUG_ON(ret); |
56bec294 | 669 | |
86b9f2ec | 670 | old_root_used = btrfs_root_used(&root->root_item); |
4a8c9a62 | 671 | ret = btrfs_write_dirty_block_groups(trans, root); |
56bec294 | 672 | BUG_ON(ret); |
0b86a832 | 673 | } |
276e680d YZ |
674 | |
675 | if (root != root->fs_info->extent_root) | |
676 | switch_commit_root(root); | |
677 | ||
0b86a832 CM |
678 | return 0; |
679 | } | |
680 | ||
d352ac68 CM |
681 | /* |
682 | * update all the cowonly tree roots on disk | |
683 | */ | |
5d4f98a2 YZ |
684 | static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans, |
685 | struct btrfs_root *root) | |
0b86a832 CM |
686 | { |
687 | struct btrfs_fs_info *fs_info = root->fs_info; | |
688 | struct list_head *next; | |
84234f3a | 689 | struct extent_buffer *eb; |
56bec294 | 690 | int ret; |
84234f3a | 691 | |
56bec294 CM |
692 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
693 | BUG_ON(ret); | |
87ef2bb4 | 694 | |
84234f3a | 695 | eb = btrfs_lock_root_node(fs_info->tree_root); |
9fa8cfe7 | 696 | btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 0, &eb); |
84234f3a YZ |
697 | btrfs_tree_unlock(eb); |
698 | free_extent_buffer(eb); | |
0b86a832 | 699 | |
56bec294 CM |
700 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
701 | BUG_ON(ret); | |
87ef2bb4 | 702 | |
d397712b | 703 | while (!list_empty(&fs_info->dirty_cowonly_roots)) { |
0b86a832 CM |
704 | next = fs_info->dirty_cowonly_roots.next; |
705 | list_del_init(next); | |
706 | root = list_entry(next, struct btrfs_root, dirty_list); | |
87ef2bb4 | 707 | |
0b86a832 | 708 | update_cowonly_root(trans, root); |
79154b1b | 709 | } |
276e680d YZ |
710 | |
711 | down_write(&fs_info->extent_commit_sem); | |
712 | switch_commit_root(fs_info->extent_root); | |
713 | up_write(&fs_info->extent_commit_sem); | |
714 | ||
79154b1b CM |
715 | return 0; |
716 | } | |
717 | ||
d352ac68 CM |
718 | /* |
719 | * dead roots are old snapshots that need to be deleted. This allocates | |
720 | * a dirty root struct and adds it into the list of dead roots that need to | |
721 | * be deleted | |
722 | */ | |
5d4f98a2 | 723 | int btrfs_add_dead_root(struct btrfs_root *root) |
5eda7b5e | 724 | { |
b48652c1 | 725 | mutex_lock(&root->fs_info->trans_mutex); |
5d4f98a2 | 726 | list_add(&root->root_list, &root->fs_info->dead_roots); |
b48652c1 | 727 | mutex_unlock(&root->fs_info->trans_mutex); |
5eda7b5e CM |
728 | return 0; |
729 | } | |
730 | ||
d352ac68 | 731 | /* |
5d4f98a2 | 732 | * update all the cowonly tree roots on disk |
d352ac68 | 733 | */ |
5d4f98a2 YZ |
734 | static noinline int commit_fs_roots(struct btrfs_trans_handle *trans, |
735 | struct btrfs_root *root) | |
0f7d52f4 | 736 | { |
0f7d52f4 | 737 | struct btrfs_root *gang[8]; |
5d4f98a2 | 738 | struct btrfs_fs_info *fs_info = root->fs_info; |
0f7d52f4 CM |
739 | int i; |
740 | int ret; | |
54aa1f4d CM |
741 | int err = 0; |
742 | ||
d397712b | 743 | while (1) { |
5d4f98a2 YZ |
744 | ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, |
745 | (void **)gang, 0, | |
0f7d52f4 CM |
746 | ARRAY_SIZE(gang), |
747 | BTRFS_ROOT_TRANS_TAG); | |
748 | if (ret == 0) | |
749 | break; | |
750 | for (i = 0; i < ret; i++) { | |
751 | root = gang[i]; | |
5d4f98a2 YZ |
752 | radix_tree_tag_clear(&fs_info->fs_roots_radix, |
753 | (unsigned long)root->root_key.objectid, | |
754 | BTRFS_ROOT_TRANS_TAG); | |
31153d81 | 755 | |
e02119d5 | 756 | btrfs_free_log(trans, root); |
5d4f98a2 | 757 | btrfs_update_reloc_root(trans, root); |
d68fc57b | 758 | btrfs_orphan_commit_root(trans, root); |
bcc63abb | 759 | |
978d910d | 760 | if (root->commit_root != root->node) { |
817d52f8 | 761 | switch_commit_root(root); |
978d910d YZ |
762 | btrfs_set_root_node(&root->root_item, |
763 | root->node); | |
764 | } | |
5d4f98a2 | 765 | |
5d4f98a2 | 766 | err = btrfs_update_root(trans, fs_info->tree_root, |
0f7d52f4 CM |
767 | &root->root_key, |
768 | &root->root_item); | |
54aa1f4d CM |
769 | if (err) |
770 | break; | |
0f7d52f4 CM |
771 | } |
772 | } | |
54aa1f4d | 773 | return err; |
0f7d52f4 CM |
774 | } |
775 | ||
d352ac68 CM |
776 | /* |
777 | * defrag a given btree. If cacheonly == 1, this won't read from the disk, | |
778 | * otherwise every leaf in the btree is read and defragged. | |
779 | */ | |
e9d0b13b CM |
780 | int btrfs_defrag_root(struct btrfs_root *root, int cacheonly) |
781 | { | |
782 | struct btrfs_fs_info *info = root->fs_info; | |
e9d0b13b | 783 | struct btrfs_trans_handle *trans; |
8929ecfa | 784 | int ret; |
d3c2fdcf | 785 | unsigned long nr; |
e9d0b13b | 786 | |
8929ecfa | 787 | if (xchg(&root->defrag_running, 1)) |
e9d0b13b | 788 | return 0; |
8929ecfa | 789 | |
6b80053d | 790 | while (1) { |
8929ecfa YZ |
791 | trans = btrfs_start_transaction(root, 0); |
792 | if (IS_ERR(trans)) | |
793 | return PTR_ERR(trans); | |
794 | ||
e9d0b13b | 795 | ret = btrfs_defrag_leaves(trans, root, cacheonly); |
8929ecfa | 796 | |
d3c2fdcf | 797 | nr = trans->blocks_used; |
e9d0b13b | 798 | btrfs_end_transaction(trans, root); |
d3c2fdcf | 799 | btrfs_btree_balance_dirty(info->tree_root, nr); |
e9d0b13b CM |
800 | cond_resched(); |
801 | ||
3f157a2f | 802 | if (root->fs_info->closing || ret != -EAGAIN) |
e9d0b13b CM |
803 | break; |
804 | } | |
805 | root->defrag_running = 0; | |
8929ecfa | 806 | return ret; |
e9d0b13b CM |
807 | } |
808 | ||
2c47e605 | 809 | #if 0 |
b7ec40d7 CM |
810 | /* |
811 | * when dropping snapshots, we generate a ton of delayed refs, and it makes | |
812 | * sense not to join the transaction while it is trying to flush the current | |
813 | * queue of delayed refs out. | |
814 | * | |
815 | * This is used by the drop snapshot code only | |
816 | */ | |
817 | static noinline int wait_transaction_pre_flush(struct btrfs_fs_info *info) | |
818 | { | |
819 | DEFINE_WAIT(wait); | |
820 | ||
821 | mutex_lock(&info->trans_mutex); | |
822 | while (info->running_transaction && | |
823 | info->running_transaction->delayed_refs.flushing) { | |
824 | prepare_to_wait(&info->transaction_wait, &wait, | |
825 | TASK_UNINTERRUPTIBLE); | |
826 | mutex_unlock(&info->trans_mutex); | |
59bc5c75 | 827 | |
b7ec40d7 | 828 | schedule(); |
59bc5c75 | 829 | |
b7ec40d7 CM |
830 | mutex_lock(&info->trans_mutex); |
831 | finish_wait(&info->transaction_wait, &wait); | |
832 | } | |
833 | mutex_unlock(&info->trans_mutex); | |
834 | return 0; | |
835 | } | |
836 | ||
d352ac68 CM |
837 | /* |
838 | * Given a list of roots that need to be deleted, call btrfs_drop_snapshot on | |
839 | * all of them | |
840 | */ | |
5d4f98a2 | 841 | int btrfs_drop_dead_root(struct btrfs_root *root) |
0f7d52f4 | 842 | { |
0f7d52f4 | 843 | struct btrfs_trans_handle *trans; |
5d4f98a2 | 844 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
d3c2fdcf | 845 | unsigned long nr; |
5d4f98a2 | 846 | int ret; |
58176a96 | 847 | |
5d4f98a2 YZ |
848 | while (1) { |
849 | /* | |
850 | * we don't want to jump in and create a bunch of | |
851 | * delayed refs if the transaction is starting to close | |
852 | */ | |
853 | wait_transaction_pre_flush(tree_root->fs_info); | |
854 | trans = btrfs_start_transaction(tree_root, 1); | |
a2135011 | 855 | |
5d4f98a2 YZ |
856 | /* |
857 | * we've joined a transaction, make sure it isn't | |
858 | * closing right now | |
859 | */ | |
860 | if (trans->transaction->delayed_refs.flushing) { | |
861 | btrfs_end_transaction(trans, tree_root); | |
862 | continue; | |
9f3a7427 | 863 | } |
58176a96 | 864 | |
5d4f98a2 YZ |
865 | ret = btrfs_drop_snapshot(trans, root); |
866 | if (ret != -EAGAIN) | |
867 | break; | |
a2135011 | 868 | |
5d4f98a2 YZ |
869 | ret = btrfs_update_root(trans, tree_root, |
870 | &root->root_key, | |
871 | &root->root_item); | |
872 | if (ret) | |
54aa1f4d | 873 | break; |
bcc63abb | 874 | |
d3c2fdcf | 875 | nr = trans->blocks_used; |
0f7d52f4 CM |
876 | ret = btrfs_end_transaction(trans, tree_root); |
877 | BUG_ON(ret); | |
5eda7b5e | 878 | |
d3c2fdcf | 879 | btrfs_btree_balance_dirty(tree_root, nr); |
4dc11904 | 880 | cond_resched(); |
0f7d52f4 | 881 | } |
5d4f98a2 YZ |
882 | BUG_ON(ret); |
883 | ||
884 | ret = btrfs_del_root(trans, tree_root, &root->root_key); | |
885 | BUG_ON(ret); | |
886 | ||
887 | nr = trans->blocks_used; | |
888 | ret = btrfs_end_transaction(trans, tree_root); | |
889 | BUG_ON(ret); | |
890 | ||
891 | free_extent_buffer(root->node); | |
892 | free_extent_buffer(root->commit_root); | |
893 | kfree(root); | |
894 | ||
895 | btrfs_btree_balance_dirty(tree_root, nr); | |
54aa1f4d | 896 | return ret; |
0f7d52f4 | 897 | } |
2c47e605 | 898 | #endif |
0f7d52f4 | 899 | |
d352ac68 CM |
900 | /* |
901 | * new snapshots need to be created at a very specific time in the | |
902 | * transaction commit. This does the actual creation | |
903 | */ | |
80b6794d | 904 | static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, |
3063d29f CM |
905 | struct btrfs_fs_info *fs_info, |
906 | struct btrfs_pending_snapshot *pending) | |
907 | { | |
908 | struct btrfs_key key; | |
80b6794d | 909 | struct btrfs_root_item *new_root_item; |
3063d29f CM |
910 | struct btrfs_root *tree_root = fs_info->tree_root; |
911 | struct btrfs_root *root = pending->root; | |
6bdb72de SW |
912 | struct btrfs_root *parent_root; |
913 | struct inode *parent_inode; | |
6a912213 | 914 | struct dentry *parent; |
a22285a6 | 915 | struct dentry *dentry; |
3063d29f | 916 | struct extent_buffer *tmp; |
925baedd | 917 | struct extent_buffer *old; |
3063d29f | 918 | int ret; |
d68fc57b | 919 | u64 to_reserve = 0; |
6bdb72de | 920 | u64 index = 0; |
a22285a6 | 921 | u64 objectid; |
b83cc969 | 922 | u64 root_flags; |
3063d29f | 923 | |
80b6794d CM |
924 | new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS); |
925 | if (!new_root_item) { | |
a22285a6 | 926 | pending->error = -ENOMEM; |
80b6794d CM |
927 | goto fail; |
928 | } | |
a22285a6 | 929 | |
3063d29f | 930 | ret = btrfs_find_free_objectid(trans, tree_root, 0, &objectid); |
a22285a6 YZ |
931 | if (ret) { |
932 | pending->error = ret; | |
3063d29f | 933 | goto fail; |
a22285a6 | 934 | } |
3063d29f | 935 | |
3fd0a558 | 936 | btrfs_reloc_pre_snapshot(trans, pending, &to_reserve); |
d68fc57b YZ |
937 | btrfs_orphan_pre_snapshot(trans, pending, &to_reserve); |
938 | ||
939 | if (to_reserve > 0) { | |
940 | ret = btrfs_block_rsv_add(trans, root, &pending->block_rsv, | |
8bb8ab2e | 941 | to_reserve); |
d68fc57b YZ |
942 | if (ret) { |
943 | pending->error = ret; | |
944 | goto fail; | |
945 | } | |
946 | } | |
947 | ||
3063d29f | 948 | key.objectid = objectid; |
a22285a6 YZ |
949 | key.offset = (u64)-1; |
950 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3063d29f | 951 | |
a22285a6 | 952 | trans->block_rsv = &pending->block_rsv; |
3de4586c | 953 | |
a22285a6 | 954 | dentry = pending->dentry; |
6a912213 JB |
955 | parent = dget_parent(dentry); |
956 | parent_inode = parent->d_inode; | |
a22285a6 | 957 | parent_root = BTRFS_I(parent_inode)->root; |
6bdb72de | 958 | record_root_in_trans(trans, parent_root); |
a22285a6 | 959 | |
3063d29f CM |
960 | /* |
961 | * insert the directory item | |
962 | */ | |
3de4586c | 963 | ret = btrfs_set_inode_index(parent_inode, &index); |
6bdb72de | 964 | BUG_ON(ret); |
0660b5af | 965 | ret = btrfs_insert_dir_item(trans, parent_root, |
a22285a6 YZ |
966 | dentry->d_name.name, dentry->d_name.len, |
967 | parent_inode->i_ino, &key, | |
968 | BTRFS_FT_DIR, index); | |
6bdb72de | 969 | BUG_ON(ret); |
0660b5af | 970 | |
a22285a6 YZ |
971 | btrfs_i_size_write(parent_inode, parent_inode->i_size + |
972 | dentry->d_name.len * 2); | |
52c26179 YZ |
973 | ret = btrfs_update_inode(trans, parent_root, parent_inode); |
974 | BUG_ON(ret); | |
975 | ||
6bdb72de SW |
976 | record_root_in_trans(trans, root); |
977 | btrfs_set_root_last_snapshot(&root->root_item, trans->transid); | |
978 | memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); | |
08fe4db1 | 979 | btrfs_check_and_init_root_item(new_root_item); |
6bdb72de | 980 | |
b83cc969 LZ |
981 | root_flags = btrfs_root_flags(new_root_item); |
982 | if (pending->readonly) | |
983 | root_flags |= BTRFS_ROOT_SUBVOL_RDONLY; | |
984 | else | |
985 | root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY; | |
986 | btrfs_set_root_flags(new_root_item, root_flags); | |
987 | ||
6bdb72de SW |
988 | old = btrfs_lock_root_node(root); |
989 | btrfs_cow_block(trans, root, old, NULL, 0, &old); | |
990 | btrfs_set_lock_blocking(old); | |
991 | ||
992 | btrfs_copy_root(trans, root, old, &tmp, objectid); | |
993 | btrfs_tree_unlock(old); | |
994 | free_extent_buffer(old); | |
995 | ||
996 | btrfs_set_root_node(new_root_item, tmp); | |
a22285a6 YZ |
997 | /* record when the snapshot was created in key.offset */ |
998 | key.offset = trans->transid; | |
999 | ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); | |
6bdb72de SW |
1000 | btrfs_tree_unlock(tmp); |
1001 | free_extent_buffer(tmp); | |
a22285a6 | 1002 | BUG_ON(ret); |
6bdb72de | 1003 | |
a22285a6 YZ |
1004 | /* |
1005 | * insert root back/forward references | |
1006 | */ | |
1007 | ret = btrfs_add_root_ref(trans, tree_root, objectid, | |
0660b5af | 1008 | parent_root->root_key.objectid, |
a22285a6 YZ |
1009 | parent_inode->i_ino, index, |
1010 | dentry->d_name.name, dentry->d_name.len); | |
0660b5af | 1011 | BUG_ON(ret); |
6a912213 | 1012 | dput(parent); |
0660b5af | 1013 | |
a22285a6 YZ |
1014 | key.offset = (u64)-1; |
1015 | pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key); | |
1016 | BUG_ON(IS_ERR(pending->snap)); | |
d68fc57b | 1017 | |
3fd0a558 | 1018 | btrfs_reloc_post_snapshot(trans, pending); |
d68fc57b | 1019 | btrfs_orphan_post_snapshot(trans, pending); |
3063d29f | 1020 | fail: |
6bdb72de | 1021 | kfree(new_root_item); |
a22285a6 YZ |
1022 | btrfs_block_rsv_release(root, &pending->block_rsv, (u64)-1); |
1023 | return 0; | |
3063d29f CM |
1024 | } |
1025 | ||
d352ac68 CM |
1026 | /* |
1027 | * create all the snapshots we've scheduled for creation | |
1028 | */ | |
80b6794d CM |
1029 | static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans, |
1030 | struct btrfs_fs_info *fs_info) | |
3de4586c CM |
1031 | { |
1032 | struct btrfs_pending_snapshot *pending; | |
1033 | struct list_head *head = &trans->transaction->pending_snapshots; | |
3de4586c CM |
1034 | int ret; |
1035 | ||
c6e30871 | 1036 | list_for_each_entry(pending, head, list) { |
3de4586c CM |
1037 | ret = create_pending_snapshot(trans, fs_info, pending); |
1038 | BUG_ON(ret); | |
1039 | } | |
1040 | return 0; | |
1041 | } | |
1042 | ||
5d4f98a2 YZ |
1043 | static void update_super_roots(struct btrfs_root *root) |
1044 | { | |
1045 | struct btrfs_root_item *root_item; | |
1046 | struct btrfs_super_block *super; | |
1047 | ||
1048 | super = &root->fs_info->super_copy; | |
1049 | ||
1050 | root_item = &root->fs_info->chunk_root->root_item; | |
1051 | super->chunk_root = root_item->bytenr; | |
1052 | super->chunk_root_generation = root_item->generation; | |
1053 | super->chunk_root_level = root_item->level; | |
1054 | ||
1055 | root_item = &root->fs_info->tree_root->root_item; | |
1056 | super->root = root_item->bytenr; | |
1057 | super->generation = root_item->generation; | |
1058 | super->root_level = root_item->level; | |
0af3d00b JB |
1059 | if (super->cache_generation != 0 || btrfs_test_opt(root, SPACE_CACHE)) |
1060 | super->cache_generation = root_item->generation; | |
5d4f98a2 YZ |
1061 | } |
1062 | ||
f36f3042 CM |
1063 | int btrfs_transaction_in_commit(struct btrfs_fs_info *info) |
1064 | { | |
1065 | int ret = 0; | |
1066 | spin_lock(&info->new_trans_lock); | |
1067 | if (info->running_transaction) | |
1068 | ret = info->running_transaction->in_commit; | |
1069 | spin_unlock(&info->new_trans_lock); | |
1070 | return ret; | |
1071 | } | |
1072 | ||
8929ecfa YZ |
1073 | int btrfs_transaction_blocked(struct btrfs_fs_info *info) |
1074 | { | |
1075 | int ret = 0; | |
1076 | spin_lock(&info->new_trans_lock); | |
1077 | if (info->running_transaction) | |
1078 | ret = info->running_transaction->blocked; | |
1079 | spin_unlock(&info->new_trans_lock); | |
1080 | return ret; | |
1081 | } | |
1082 | ||
bb9c12c9 SW |
1083 | /* |
1084 | * wait for the current transaction commit to start and block subsequent | |
1085 | * transaction joins | |
1086 | */ | |
1087 | static void wait_current_trans_commit_start(struct btrfs_root *root, | |
1088 | struct btrfs_transaction *trans) | |
1089 | { | |
1090 | DEFINE_WAIT(wait); | |
1091 | ||
1092 | if (trans->in_commit) | |
1093 | return; | |
1094 | ||
1095 | while (1) { | |
1096 | prepare_to_wait(&root->fs_info->transaction_blocked_wait, &wait, | |
1097 | TASK_UNINTERRUPTIBLE); | |
1098 | if (trans->in_commit) { | |
1099 | finish_wait(&root->fs_info->transaction_blocked_wait, | |
1100 | &wait); | |
1101 | break; | |
1102 | } | |
1103 | mutex_unlock(&root->fs_info->trans_mutex); | |
1104 | schedule(); | |
1105 | mutex_lock(&root->fs_info->trans_mutex); | |
1106 | finish_wait(&root->fs_info->transaction_blocked_wait, &wait); | |
1107 | } | |
1108 | } | |
1109 | ||
1110 | /* | |
1111 | * wait for the current transaction to start and then become unblocked. | |
1112 | * caller holds ref. | |
1113 | */ | |
1114 | static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root, | |
1115 | struct btrfs_transaction *trans) | |
1116 | { | |
1117 | DEFINE_WAIT(wait); | |
1118 | ||
1119 | if (trans->commit_done || (trans->in_commit && !trans->blocked)) | |
1120 | return; | |
1121 | ||
1122 | while (1) { | |
1123 | prepare_to_wait(&root->fs_info->transaction_wait, &wait, | |
1124 | TASK_UNINTERRUPTIBLE); | |
1125 | if (trans->commit_done || | |
1126 | (trans->in_commit && !trans->blocked)) { | |
1127 | finish_wait(&root->fs_info->transaction_wait, | |
1128 | &wait); | |
1129 | break; | |
1130 | } | |
1131 | mutex_unlock(&root->fs_info->trans_mutex); | |
1132 | schedule(); | |
1133 | mutex_lock(&root->fs_info->trans_mutex); | |
1134 | finish_wait(&root->fs_info->transaction_wait, | |
1135 | &wait); | |
1136 | } | |
1137 | } | |
1138 | ||
1139 | /* | |
1140 | * commit transactions asynchronously. once btrfs_commit_transaction_async | |
1141 | * returns, any subsequent transaction will not be allowed to join. | |
1142 | */ | |
1143 | struct btrfs_async_commit { | |
1144 | struct btrfs_trans_handle *newtrans; | |
1145 | struct btrfs_root *root; | |
1146 | struct delayed_work work; | |
1147 | }; | |
1148 | ||
1149 | static void do_async_commit(struct work_struct *work) | |
1150 | { | |
1151 | struct btrfs_async_commit *ac = | |
1152 | container_of(work, struct btrfs_async_commit, work.work); | |
1153 | ||
1154 | btrfs_commit_transaction(ac->newtrans, ac->root); | |
1155 | kfree(ac); | |
1156 | } | |
1157 | ||
1158 | int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans, | |
1159 | struct btrfs_root *root, | |
1160 | int wait_for_unblock) | |
1161 | { | |
1162 | struct btrfs_async_commit *ac; | |
1163 | struct btrfs_transaction *cur_trans; | |
1164 | ||
1165 | ac = kmalloc(sizeof(*ac), GFP_NOFS); | |
db5b493a TI |
1166 | if (!ac) |
1167 | return -ENOMEM; | |
bb9c12c9 SW |
1168 | |
1169 | INIT_DELAYED_WORK(&ac->work, do_async_commit); | |
1170 | ac->root = root; | |
1171 | ac->newtrans = btrfs_join_transaction(root, 0); | |
3612b495 TI |
1172 | if (IS_ERR(ac->newtrans)) { |
1173 | int err = PTR_ERR(ac->newtrans); | |
1174 | kfree(ac); | |
1175 | return err; | |
1176 | } | |
bb9c12c9 SW |
1177 | |
1178 | /* take transaction reference */ | |
1179 | mutex_lock(&root->fs_info->trans_mutex); | |
1180 | cur_trans = trans->transaction; | |
1181 | cur_trans->use_count++; | |
1182 | mutex_unlock(&root->fs_info->trans_mutex); | |
1183 | ||
1184 | btrfs_end_transaction(trans, root); | |
1185 | schedule_delayed_work(&ac->work, 0); | |
1186 | ||
1187 | /* wait for transaction to start and unblock */ | |
1188 | mutex_lock(&root->fs_info->trans_mutex); | |
1189 | if (wait_for_unblock) | |
1190 | wait_current_trans_commit_start_and_unblock(root, cur_trans); | |
1191 | else | |
1192 | wait_current_trans_commit_start(root, cur_trans); | |
1193 | put_transaction(cur_trans); | |
1194 | mutex_unlock(&root->fs_info->trans_mutex); | |
1195 | ||
1196 | return 0; | |
1197 | } | |
1198 | ||
1199 | /* | |
1200 | * btrfs_transaction state sequence: | |
1201 | * in_commit = 0, blocked = 0 (initial) | |
1202 | * in_commit = 1, blocked = 1 | |
1203 | * blocked = 0 | |
1204 | * commit_done = 1 | |
1205 | */ | |
79154b1b CM |
1206 | int btrfs_commit_transaction(struct btrfs_trans_handle *trans, |
1207 | struct btrfs_root *root) | |
1208 | { | |
15ee9bc7 | 1209 | unsigned long joined = 0; |
79154b1b | 1210 | struct btrfs_transaction *cur_trans; |
8fd17795 | 1211 | struct btrfs_transaction *prev_trans = NULL; |
79154b1b | 1212 | DEFINE_WAIT(wait); |
15ee9bc7 | 1213 | int ret; |
89573b9c CM |
1214 | int should_grow = 0; |
1215 | unsigned long now = get_seconds(); | |
dccae999 | 1216 | int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT); |
79154b1b | 1217 | |
5a3f23d5 CM |
1218 | btrfs_run_ordered_operations(root, 0); |
1219 | ||
56bec294 CM |
1220 | /* make a pass through all the delayed refs we have so far |
1221 | * any runnings procs may add more while we are here | |
1222 | */ | |
1223 | ret = btrfs_run_delayed_refs(trans, root, 0); | |
1224 | BUG_ON(ret); | |
1225 | ||
a22285a6 YZ |
1226 | btrfs_trans_release_metadata(trans, root); |
1227 | ||
b7ec40d7 | 1228 | cur_trans = trans->transaction; |
56bec294 CM |
1229 | /* |
1230 | * set the flushing flag so procs in this transaction have to | |
1231 | * start sending their work down. | |
1232 | */ | |
b7ec40d7 | 1233 | cur_trans->delayed_refs.flushing = 1; |
56bec294 | 1234 | |
c3e69d58 | 1235 | ret = btrfs_run_delayed_refs(trans, root, 0); |
56bec294 CM |
1236 | BUG_ON(ret); |
1237 | ||
79154b1b | 1238 | mutex_lock(&root->fs_info->trans_mutex); |
b7ec40d7 CM |
1239 | if (cur_trans->in_commit) { |
1240 | cur_trans->use_count++; | |
ccd467d6 | 1241 | mutex_unlock(&root->fs_info->trans_mutex); |
79154b1b | 1242 | btrfs_end_transaction(trans, root); |
ccd467d6 | 1243 | |
79154b1b CM |
1244 | ret = wait_for_commit(root, cur_trans); |
1245 | BUG_ON(ret); | |
15ee9bc7 JB |
1246 | |
1247 | mutex_lock(&root->fs_info->trans_mutex); | |
79154b1b | 1248 | put_transaction(cur_trans); |
15ee9bc7 JB |
1249 | mutex_unlock(&root->fs_info->trans_mutex); |
1250 | ||
79154b1b CM |
1251 | return 0; |
1252 | } | |
4313b399 | 1253 | |
2c90e5d6 | 1254 | trans->transaction->in_commit = 1; |
f9295749 | 1255 | trans->transaction->blocked = 1; |
bb9c12c9 SW |
1256 | wake_up(&root->fs_info->transaction_blocked_wait); |
1257 | ||
ccd467d6 CM |
1258 | if (cur_trans->list.prev != &root->fs_info->trans_list) { |
1259 | prev_trans = list_entry(cur_trans->list.prev, | |
1260 | struct btrfs_transaction, list); | |
1261 | if (!prev_trans->commit_done) { | |
1262 | prev_trans->use_count++; | |
ccd467d6 CM |
1263 | mutex_unlock(&root->fs_info->trans_mutex); |
1264 | ||
1265 | wait_for_commit(root, prev_trans); | |
ccd467d6 | 1266 | |
ccd467d6 | 1267 | mutex_lock(&root->fs_info->trans_mutex); |
15ee9bc7 | 1268 | put_transaction(prev_trans); |
ccd467d6 CM |
1269 | } |
1270 | } | |
15ee9bc7 | 1271 | |
89573b9c CM |
1272 | if (now < cur_trans->start_time || now - cur_trans->start_time < 1) |
1273 | should_grow = 1; | |
1274 | ||
15ee9bc7 | 1275 | do { |
7ea394f1 | 1276 | int snap_pending = 0; |
15ee9bc7 | 1277 | joined = cur_trans->num_joined; |
7ea394f1 YZ |
1278 | if (!list_empty(&trans->transaction->pending_snapshots)) |
1279 | snap_pending = 1; | |
1280 | ||
2c90e5d6 | 1281 | WARN_ON(cur_trans != trans->transaction); |
79154b1b | 1282 | mutex_unlock(&root->fs_info->trans_mutex); |
15ee9bc7 | 1283 | |
0bdb1db2 | 1284 | if (flush_on_commit || snap_pending) { |
24bbcf04 YZ |
1285 | btrfs_start_delalloc_inodes(root, 1); |
1286 | ret = btrfs_wait_ordered_extents(root, 0, 1); | |
ebecd3d9 | 1287 | BUG_ON(ret); |
7ea394f1 YZ |
1288 | } |
1289 | ||
5a3f23d5 CM |
1290 | /* |
1291 | * rename don't use btrfs_join_transaction, so, once we | |
1292 | * set the transaction to blocked above, we aren't going | |
1293 | * to get any new ordered operations. We can safely run | |
1294 | * it here and no for sure that nothing new will be added | |
1295 | * to the list | |
1296 | */ | |
1297 | btrfs_run_ordered_operations(root, 1); | |
1298 | ||
ed3b3d31 CM |
1299 | prepare_to_wait(&cur_trans->writer_wait, &wait, |
1300 | TASK_UNINTERRUPTIBLE); | |
1301 | ||
89573b9c | 1302 | smp_mb(); |
99d16cbc SW |
1303 | if (cur_trans->num_writers > 1) |
1304 | schedule_timeout(MAX_SCHEDULE_TIMEOUT); | |
1305 | else if (should_grow) | |
1306 | schedule_timeout(1); | |
15ee9bc7 | 1307 | |
79154b1b | 1308 | mutex_lock(&root->fs_info->trans_mutex); |
15ee9bc7 JB |
1309 | finish_wait(&cur_trans->writer_wait, &wait); |
1310 | } while (cur_trans->num_writers > 1 || | |
89573b9c | 1311 | (should_grow && cur_trans->num_joined != joined)); |
15ee9bc7 | 1312 | |
3063d29f CM |
1313 | ret = create_pending_snapshots(trans, root->fs_info); |
1314 | BUG_ON(ret); | |
1315 | ||
56bec294 CM |
1316 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
1317 | BUG_ON(ret); | |
1318 | ||
2c90e5d6 | 1319 | WARN_ON(cur_trans != trans->transaction); |
dc17ff8f | 1320 | |
e02119d5 CM |
1321 | /* btrfs_commit_tree_roots is responsible for getting the |
1322 | * various roots consistent with each other. Every pointer | |
1323 | * in the tree of tree roots has to point to the most up to date | |
1324 | * root for every subvolume and other tree. So, we have to keep | |
1325 | * the tree logging code from jumping in and changing any | |
1326 | * of the trees. | |
1327 | * | |
1328 | * At this point in the commit, there can't be any tree-log | |
1329 | * writers, but a little lower down we drop the trans mutex | |
1330 | * and let new people in. By holding the tree_log_mutex | |
1331 | * from now until after the super is written, we avoid races | |
1332 | * with the tree-log code. | |
1333 | */ | |
1334 | mutex_lock(&root->fs_info->tree_log_mutex); | |
1335 | ||
5d4f98a2 | 1336 | ret = commit_fs_roots(trans, root); |
54aa1f4d CM |
1337 | BUG_ON(ret); |
1338 | ||
5d4f98a2 | 1339 | /* commit_fs_roots gets rid of all the tree log roots, it is now |
e02119d5 CM |
1340 | * safe to free the root of tree log roots |
1341 | */ | |
1342 | btrfs_free_log_root_tree(trans, root->fs_info); | |
1343 | ||
5d4f98a2 | 1344 | ret = commit_cowonly_roots(trans, root); |
79154b1b | 1345 | BUG_ON(ret); |
54aa1f4d | 1346 | |
11833d66 YZ |
1347 | btrfs_prepare_extent_commit(trans, root); |
1348 | ||
78fae27e | 1349 | cur_trans = root->fs_info->running_transaction; |
cee36a03 | 1350 | spin_lock(&root->fs_info->new_trans_lock); |
78fae27e | 1351 | root->fs_info->running_transaction = NULL; |
cee36a03 | 1352 | spin_unlock(&root->fs_info->new_trans_lock); |
5d4f98a2 YZ |
1353 | |
1354 | btrfs_set_root_node(&root->fs_info->tree_root->root_item, | |
1355 | root->fs_info->tree_root->node); | |
817d52f8 | 1356 | switch_commit_root(root->fs_info->tree_root); |
5d4f98a2 YZ |
1357 | |
1358 | btrfs_set_root_node(&root->fs_info->chunk_root->root_item, | |
1359 | root->fs_info->chunk_root->node); | |
817d52f8 | 1360 | switch_commit_root(root->fs_info->chunk_root); |
5d4f98a2 YZ |
1361 | |
1362 | update_super_roots(root); | |
e02119d5 CM |
1363 | |
1364 | if (!root->fs_info->log_root_recovering) { | |
1365 | btrfs_set_super_log_root(&root->fs_info->super_copy, 0); | |
1366 | btrfs_set_super_log_root_level(&root->fs_info->super_copy, 0); | |
1367 | } | |
1368 | ||
a061fc8d CM |
1369 | memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy, |
1370 | sizeof(root->fs_info->super_copy)); | |
ccd467d6 | 1371 | |
f9295749 | 1372 | trans->transaction->blocked = 0; |
b7ec40d7 | 1373 | |
f9295749 | 1374 | wake_up(&root->fs_info->transaction_wait); |
e6dcd2dc | 1375 | |
78fae27e | 1376 | mutex_unlock(&root->fs_info->trans_mutex); |
79154b1b CM |
1377 | ret = btrfs_write_and_wait_transaction(trans, root); |
1378 | BUG_ON(ret); | |
a512bbf8 | 1379 | write_ctree_super(trans, root, 0); |
4313b399 | 1380 | |
e02119d5 CM |
1381 | /* |
1382 | * the super is written, we can safely allow the tree-loggers | |
1383 | * to go about their business | |
1384 | */ | |
1385 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
1386 | ||
11833d66 | 1387 | btrfs_finish_extent_commit(trans, root); |
4313b399 | 1388 | |
1a40e23b ZY |
1389 | mutex_lock(&root->fs_info->trans_mutex); |
1390 | ||
2c90e5d6 | 1391 | cur_trans->commit_done = 1; |
b7ec40d7 | 1392 | |
15ee9bc7 | 1393 | root->fs_info->last_trans_committed = cur_trans->transid; |
817d52f8 | 1394 | |
2c90e5d6 | 1395 | wake_up(&cur_trans->commit_wait); |
3de4586c | 1396 | |
78fae27e | 1397 | put_transaction(cur_trans); |
79154b1b | 1398 | put_transaction(cur_trans); |
58176a96 | 1399 | |
1abe9b8a | 1400 | trace_btrfs_transaction_commit(root); |
1401 | ||
78fae27e | 1402 | mutex_unlock(&root->fs_info->trans_mutex); |
3de4586c | 1403 | |
9ed74f2d JB |
1404 | if (current->journal_info == trans) |
1405 | current->journal_info = NULL; | |
1406 | ||
2c90e5d6 | 1407 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
24bbcf04 YZ |
1408 | |
1409 | if (current != root->fs_info->transaction_kthread) | |
1410 | btrfs_run_delayed_iputs(root); | |
1411 | ||
79154b1b CM |
1412 | return ret; |
1413 | } | |
1414 | ||
d352ac68 CM |
1415 | /* |
1416 | * interface function to delete all the snapshots we have scheduled for deletion | |
1417 | */ | |
e9d0b13b CM |
1418 | int btrfs_clean_old_snapshots(struct btrfs_root *root) |
1419 | { | |
5d4f98a2 YZ |
1420 | LIST_HEAD(list); |
1421 | struct btrfs_fs_info *fs_info = root->fs_info; | |
1422 | ||
1423 | mutex_lock(&fs_info->trans_mutex); | |
1424 | list_splice_init(&fs_info->dead_roots, &list); | |
1425 | mutex_unlock(&fs_info->trans_mutex); | |
e9d0b13b | 1426 | |
5d4f98a2 YZ |
1427 | while (!list_empty(&list)) { |
1428 | root = list_entry(list.next, struct btrfs_root, root_list); | |
76dda93c YZ |
1429 | list_del(&root->root_list); |
1430 | ||
1431 | if (btrfs_header_backref_rev(root->node) < | |
1432 | BTRFS_MIXED_BACKREF_REV) | |
3fd0a558 | 1433 | btrfs_drop_snapshot(root, NULL, 0); |
76dda93c | 1434 | else |
3fd0a558 | 1435 | btrfs_drop_snapshot(root, NULL, 1); |
e9d0b13b CM |
1436 | } |
1437 | return 0; | |
1438 | } |