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; | |
06d5a589 | 184 | int retries = 0; |
37d1aeee | 185 | int ret; |
acce952b | 186 | |
187 | if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) | |
188 | return ERR_PTR(-EROFS); | |
a22285a6 YZ |
189 | again: |
190 | h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS); | |
191 | if (!h) | |
192 | return ERR_PTR(-ENOMEM); | |
37d1aeee | 193 | |
0af3d00b JB |
194 | if (type != TRANS_JOIN_NOLOCK) |
195 | mutex_lock(&root->fs_info->trans_mutex); | |
a22285a6 | 196 | if (may_wait_transaction(root, type)) |
37d1aeee | 197 | wait_current_trans(root); |
a22285a6 | 198 | |
79154b1b | 199 | ret = join_transaction(root); |
db5b493a | 200 | if (ret < 0) { |
6e8df2ae | 201 | kmem_cache_free(btrfs_trans_handle_cachep, h); |
db5b493a TI |
202 | if (type != TRANS_JOIN_NOLOCK) |
203 | mutex_unlock(&root->fs_info->trans_mutex); | |
204 | return ERR_PTR(ret); | |
205 | } | |
0f7d52f4 | 206 | |
a22285a6 YZ |
207 | cur_trans = root->fs_info->running_transaction; |
208 | cur_trans->use_count++; | |
0af3d00b JB |
209 | if (type != TRANS_JOIN_NOLOCK) |
210 | mutex_unlock(&root->fs_info->trans_mutex); | |
a22285a6 YZ |
211 | |
212 | h->transid = cur_trans->transid; | |
213 | h->transaction = cur_trans; | |
79154b1b | 214 | h->blocks_used = 0; |
d2fb3437 | 215 | h->block_group = 0; |
a22285a6 | 216 | h->bytes_reserved = 0; |
56bec294 | 217 | h->delayed_ref_updates = 0; |
f0486c68 | 218 | h->block_rsv = NULL; |
b7ec40d7 | 219 | |
a22285a6 YZ |
220 | smp_mb(); |
221 | if (cur_trans->blocked && may_wait_transaction(root, type)) { | |
222 | btrfs_commit_transaction(h, root); | |
223 | goto again; | |
224 | } | |
225 | ||
226 | if (num_items > 0) { | |
8bb8ab2e | 227 | ret = btrfs_trans_reserve_metadata(h, root, num_items); |
06d5a589 JB |
228 | if (ret == -EAGAIN && !retries) { |
229 | retries++; | |
a22285a6 YZ |
230 | btrfs_commit_transaction(h, root); |
231 | goto again; | |
06d5a589 JB |
232 | } else if (ret == -EAGAIN) { |
233 | /* | |
234 | * We have already retried and got EAGAIN, so really we | |
235 | * don't have space, so set ret to -ENOSPC. | |
236 | */ | |
237 | ret = -ENOSPC; | |
a22285a6 | 238 | } |
06d5a589 | 239 | |
a22285a6 YZ |
240 | if (ret < 0) { |
241 | btrfs_end_transaction(h, root); | |
242 | return ERR_PTR(ret); | |
243 | } | |
244 | } | |
9ed74f2d | 245 | |
0af3d00b JB |
246 | if (type != TRANS_JOIN_NOLOCK) |
247 | mutex_lock(&root->fs_info->trans_mutex); | |
5d4f98a2 | 248 | record_root_in_trans(h, root); |
0af3d00b JB |
249 | if (type != TRANS_JOIN_NOLOCK) |
250 | mutex_unlock(&root->fs_info->trans_mutex); | |
a22285a6 YZ |
251 | |
252 | if (!current->journal_info && type != TRANS_USERSPACE) | |
253 | current->journal_info = h; | |
79154b1b CM |
254 | return h; |
255 | } | |
256 | ||
f9295749 | 257 | struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, |
a22285a6 | 258 | int num_items) |
f9295749 | 259 | { |
a22285a6 | 260 | return start_transaction(root, num_items, TRANS_START); |
f9295749 CM |
261 | } |
262 | struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root, | |
263 | int num_blocks) | |
264 | { | |
a22285a6 | 265 | return start_transaction(root, 0, TRANS_JOIN); |
f9295749 CM |
266 | } |
267 | ||
0af3d00b JB |
268 | struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root, |
269 | int num_blocks) | |
270 | { | |
271 | return start_transaction(root, 0, TRANS_JOIN_NOLOCK); | |
272 | } | |
273 | ||
9ca9ee09 SW |
274 | struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *r, |
275 | int num_blocks) | |
276 | { | |
a22285a6 | 277 | return start_transaction(r, 0, TRANS_USERSPACE); |
9ca9ee09 SW |
278 | } |
279 | ||
d352ac68 | 280 | /* wait for a transaction commit to be fully complete */ |
89ce8a63 CM |
281 | static noinline int wait_for_commit(struct btrfs_root *root, |
282 | struct btrfs_transaction *commit) | |
283 | { | |
284 | DEFINE_WAIT(wait); | |
285 | mutex_lock(&root->fs_info->trans_mutex); | |
d397712b | 286 | while (!commit->commit_done) { |
89ce8a63 CM |
287 | prepare_to_wait(&commit->commit_wait, &wait, |
288 | TASK_UNINTERRUPTIBLE); | |
289 | if (commit->commit_done) | |
290 | break; | |
291 | mutex_unlock(&root->fs_info->trans_mutex); | |
292 | schedule(); | |
293 | mutex_lock(&root->fs_info->trans_mutex); | |
294 | } | |
295 | mutex_unlock(&root->fs_info->trans_mutex); | |
296 | finish_wait(&commit->commit_wait, &wait); | |
297 | return 0; | |
298 | } | |
299 | ||
46204592 SW |
300 | int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid) |
301 | { | |
302 | struct btrfs_transaction *cur_trans = NULL, *t; | |
303 | int ret; | |
304 | ||
305 | mutex_lock(&root->fs_info->trans_mutex); | |
306 | ||
307 | ret = 0; | |
308 | if (transid) { | |
309 | if (transid <= root->fs_info->last_trans_committed) | |
310 | goto out_unlock; | |
311 | ||
312 | /* find specified transaction */ | |
313 | list_for_each_entry(t, &root->fs_info->trans_list, list) { | |
314 | if (t->transid == transid) { | |
315 | cur_trans = t; | |
316 | break; | |
317 | } | |
318 | if (t->transid > transid) | |
319 | break; | |
320 | } | |
321 | ret = -EINVAL; | |
322 | if (!cur_trans) | |
323 | goto out_unlock; /* bad transid */ | |
324 | } else { | |
325 | /* find newest transaction that is committing | committed */ | |
326 | list_for_each_entry_reverse(t, &root->fs_info->trans_list, | |
327 | list) { | |
328 | if (t->in_commit) { | |
329 | if (t->commit_done) | |
330 | goto out_unlock; | |
331 | cur_trans = t; | |
332 | break; | |
333 | } | |
334 | } | |
335 | if (!cur_trans) | |
336 | goto out_unlock; /* nothing committing|committed */ | |
337 | } | |
338 | ||
339 | cur_trans->use_count++; | |
340 | mutex_unlock(&root->fs_info->trans_mutex); | |
341 | ||
342 | wait_for_commit(root, cur_trans); | |
343 | ||
344 | mutex_lock(&root->fs_info->trans_mutex); | |
345 | put_transaction(cur_trans); | |
346 | ret = 0; | |
347 | out_unlock: | |
348 | mutex_unlock(&root->fs_info->trans_mutex); | |
349 | return ret; | |
350 | } | |
351 | ||
5d4f98a2 | 352 | #if 0 |
d352ac68 | 353 | /* |
d397712b CM |
354 | * rate limit against the drop_snapshot code. This helps to slow down new |
355 | * operations if the drop_snapshot code isn't able to keep up. | |
d352ac68 | 356 | */ |
37d1aeee | 357 | static void throttle_on_drops(struct btrfs_root *root) |
ab78c84d CM |
358 | { |
359 | struct btrfs_fs_info *info = root->fs_info; | |
2dd3e67b | 360 | int harder_count = 0; |
ab78c84d | 361 | |
2dd3e67b | 362 | harder: |
ab78c84d CM |
363 | if (atomic_read(&info->throttles)) { |
364 | DEFINE_WAIT(wait); | |
365 | int thr; | |
ab78c84d CM |
366 | thr = atomic_read(&info->throttle_gen); |
367 | ||
368 | do { | |
369 | prepare_to_wait(&info->transaction_throttle, | |
370 | &wait, TASK_UNINTERRUPTIBLE); | |
371 | if (!atomic_read(&info->throttles)) { | |
372 | finish_wait(&info->transaction_throttle, &wait); | |
373 | break; | |
374 | } | |
375 | schedule(); | |
376 | finish_wait(&info->transaction_throttle, &wait); | |
377 | } while (thr == atomic_read(&info->throttle_gen)); | |
2dd3e67b CM |
378 | harder_count++; |
379 | ||
380 | if (root->fs_info->total_ref_cache_size > 1 * 1024 * 1024 && | |
381 | harder_count < 2) | |
382 | goto harder; | |
383 | ||
384 | if (root->fs_info->total_ref_cache_size > 5 * 1024 * 1024 && | |
385 | harder_count < 10) | |
386 | goto harder; | |
387 | ||
388 | if (root->fs_info->total_ref_cache_size > 10 * 1024 * 1024 && | |
389 | harder_count < 20) | |
390 | goto harder; | |
ab78c84d CM |
391 | } |
392 | } | |
5d4f98a2 | 393 | #endif |
ab78c84d | 394 | |
37d1aeee CM |
395 | void btrfs_throttle(struct btrfs_root *root) |
396 | { | |
397 | mutex_lock(&root->fs_info->trans_mutex); | |
9ca9ee09 SW |
398 | if (!root->fs_info->open_ioctl_trans) |
399 | wait_current_trans(root); | |
37d1aeee | 400 | mutex_unlock(&root->fs_info->trans_mutex); |
37d1aeee CM |
401 | } |
402 | ||
8929ecfa YZ |
403 | static int should_end_transaction(struct btrfs_trans_handle *trans, |
404 | struct btrfs_root *root) | |
405 | { | |
406 | int ret; | |
407 | ret = btrfs_block_rsv_check(trans, root, | |
408 | &root->fs_info->global_block_rsv, 0, 5); | |
409 | return ret ? 1 : 0; | |
410 | } | |
411 | ||
412 | int btrfs_should_end_transaction(struct btrfs_trans_handle *trans, | |
413 | struct btrfs_root *root) | |
414 | { | |
415 | struct btrfs_transaction *cur_trans = trans->transaction; | |
416 | int updates; | |
417 | ||
418 | if (cur_trans->blocked || cur_trans->delayed_refs.flushing) | |
419 | return 1; | |
420 | ||
421 | updates = trans->delayed_ref_updates; | |
422 | trans->delayed_ref_updates = 0; | |
423 | if (updates) | |
424 | btrfs_run_delayed_refs(trans, root, updates); | |
425 | ||
426 | return should_end_transaction(trans, root); | |
427 | } | |
428 | ||
89ce8a63 | 429 | static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, |
0af3d00b | 430 | struct btrfs_root *root, int throttle, int lock) |
79154b1b | 431 | { |
8929ecfa | 432 | struct btrfs_transaction *cur_trans = trans->transaction; |
ab78c84d | 433 | struct btrfs_fs_info *info = root->fs_info; |
c3e69d58 CM |
434 | int count = 0; |
435 | ||
436 | while (count < 4) { | |
437 | unsigned long cur = trans->delayed_ref_updates; | |
438 | trans->delayed_ref_updates = 0; | |
439 | if (cur && | |
440 | trans->transaction->delayed_refs.num_heads_ready > 64) { | |
441 | trans->delayed_ref_updates = 0; | |
b7ec40d7 CM |
442 | |
443 | /* | |
444 | * do a full flush if the transaction is trying | |
445 | * to close | |
446 | */ | |
447 | if (trans->transaction->delayed_refs.flushing) | |
448 | cur = 0; | |
c3e69d58 CM |
449 | btrfs_run_delayed_refs(trans, root, cur); |
450 | } else { | |
451 | break; | |
452 | } | |
453 | count++; | |
56bec294 CM |
454 | } |
455 | ||
a22285a6 YZ |
456 | btrfs_trans_release_metadata(trans, root); |
457 | ||
0af3d00b | 458 | if (lock && !root->fs_info->open_ioctl_trans && |
8929ecfa YZ |
459 | should_end_transaction(trans, root)) |
460 | trans->transaction->blocked = 1; | |
461 | ||
0af3d00b | 462 | if (lock && cur_trans->blocked && !cur_trans->in_commit) { |
8929ecfa YZ |
463 | if (throttle) |
464 | return btrfs_commit_transaction(trans, root); | |
465 | else | |
466 | wake_up_process(info->transaction_kthread); | |
467 | } | |
468 | ||
0af3d00b JB |
469 | if (lock) |
470 | mutex_lock(&info->trans_mutex); | |
8929ecfa | 471 | WARN_ON(cur_trans != info->running_transaction); |
d5719762 | 472 | WARN_ON(cur_trans->num_writers < 1); |
ccd467d6 | 473 | cur_trans->num_writers--; |
89ce8a63 | 474 | |
99d16cbc | 475 | smp_mb(); |
79154b1b CM |
476 | if (waitqueue_active(&cur_trans->writer_wait)) |
477 | wake_up(&cur_trans->writer_wait); | |
79154b1b | 478 | put_transaction(cur_trans); |
0af3d00b JB |
479 | if (lock) |
480 | mutex_unlock(&info->trans_mutex); | |
9ed74f2d JB |
481 | |
482 | if (current->journal_info == trans) | |
483 | current->journal_info = NULL; | |
d6025579 | 484 | memset(trans, 0, sizeof(*trans)); |
2c90e5d6 | 485 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
ab78c84d | 486 | |
24bbcf04 YZ |
487 | if (throttle) |
488 | btrfs_run_delayed_iputs(root); | |
489 | ||
79154b1b CM |
490 | return 0; |
491 | } | |
492 | ||
89ce8a63 CM |
493 | int btrfs_end_transaction(struct btrfs_trans_handle *trans, |
494 | struct btrfs_root *root) | |
495 | { | |
0af3d00b | 496 | return __btrfs_end_transaction(trans, root, 0, 1); |
89ce8a63 CM |
497 | } |
498 | ||
499 | int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans, | |
500 | struct btrfs_root *root) | |
501 | { | |
0af3d00b JB |
502 | return __btrfs_end_transaction(trans, root, 1, 1); |
503 | } | |
504 | ||
505 | int btrfs_end_transaction_nolock(struct btrfs_trans_handle *trans, | |
506 | struct btrfs_root *root) | |
507 | { | |
508 | return __btrfs_end_transaction(trans, root, 0, 0); | |
89ce8a63 CM |
509 | } |
510 | ||
d352ac68 CM |
511 | /* |
512 | * when btree blocks are allocated, they have some corresponding bits set for | |
513 | * them in one of two extent_io trees. This is used to make sure all of | |
690587d1 | 514 | * those extents are sent to disk but does not wait on them |
d352ac68 | 515 | */ |
690587d1 | 516 | int btrfs_write_marked_extents(struct btrfs_root *root, |
8cef4e16 | 517 | struct extent_io_tree *dirty_pages, int mark) |
79154b1b | 518 | { |
7c4452b9 | 519 | int ret; |
777e6bd7 | 520 | int err = 0; |
7c4452b9 CM |
521 | int werr = 0; |
522 | struct page *page; | |
7c4452b9 | 523 | struct inode *btree_inode = root->fs_info->btree_inode; |
777e6bd7 | 524 | u64 start = 0; |
5f39d397 CM |
525 | u64 end; |
526 | unsigned long index; | |
7c4452b9 | 527 | |
d397712b | 528 | while (1) { |
777e6bd7 | 529 | ret = find_first_extent_bit(dirty_pages, start, &start, &end, |
8cef4e16 | 530 | mark); |
5f39d397 | 531 | if (ret) |
7c4452b9 | 532 | break; |
d397712b | 533 | while (start <= end) { |
777e6bd7 CM |
534 | cond_resched(); |
535 | ||
5f39d397 | 536 | index = start >> PAGE_CACHE_SHIFT; |
35ebb934 | 537 | start = (u64)(index + 1) << PAGE_CACHE_SHIFT; |
4bef0848 | 538 | page = find_get_page(btree_inode->i_mapping, index); |
7c4452b9 CM |
539 | if (!page) |
540 | continue; | |
4bef0848 CM |
541 | |
542 | btree_lock_page_hook(page); | |
543 | if (!page->mapping) { | |
544 | unlock_page(page); | |
545 | page_cache_release(page); | |
546 | continue; | |
547 | } | |
548 | ||
6702ed49 CM |
549 | if (PageWriteback(page)) { |
550 | if (PageDirty(page)) | |
551 | wait_on_page_writeback(page); | |
552 | else { | |
553 | unlock_page(page); | |
554 | page_cache_release(page); | |
555 | continue; | |
556 | } | |
557 | } | |
7c4452b9 CM |
558 | err = write_one_page(page, 0); |
559 | if (err) | |
560 | werr = err; | |
561 | page_cache_release(page); | |
562 | } | |
563 | } | |
690587d1 CM |
564 | if (err) |
565 | werr = err; | |
566 | return werr; | |
567 | } | |
568 | ||
569 | /* | |
570 | * when btree blocks are allocated, they have some corresponding bits set for | |
571 | * them in one of two extent_io trees. This is used to make sure all of | |
572 | * those extents are on disk for transaction or log commit. We wait | |
573 | * on all the pages and clear them from the dirty pages state tree | |
574 | */ | |
575 | int btrfs_wait_marked_extents(struct btrfs_root *root, | |
8cef4e16 | 576 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 CM |
577 | { |
578 | int ret; | |
579 | int err = 0; | |
580 | int werr = 0; | |
581 | struct page *page; | |
582 | struct inode *btree_inode = root->fs_info->btree_inode; | |
583 | u64 start = 0; | |
584 | u64 end; | |
585 | unsigned long index; | |
586 | ||
d397712b | 587 | while (1) { |
8cef4e16 YZ |
588 | ret = find_first_extent_bit(dirty_pages, start, &start, &end, |
589 | mark); | |
777e6bd7 CM |
590 | if (ret) |
591 | break; | |
592 | ||
8cef4e16 | 593 | clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS); |
d397712b | 594 | while (start <= end) { |
777e6bd7 CM |
595 | index = start >> PAGE_CACHE_SHIFT; |
596 | start = (u64)(index + 1) << PAGE_CACHE_SHIFT; | |
597 | page = find_get_page(btree_inode->i_mapping, index); | |
598 | if (!page) | |
599 | continue; | |
600 | if (PageDirty(page)) { | |
4bef0848 CM |
601 | btree_lock_page_hook(page); |
602 | wait_on_page_writeback(page); | |
777e6bd7 CM |
603 | err = write_one_page(page, 0); |
604 | if (err) | |
605 | werr = err; | |
606 | } | |
105d931d | 607 | wait_on_page_writeback(page); |
777e6bd7 CM |
608 | page_cache_release(page); |
609 | cond_resched(); | |
610 | } | |
611 | } | |
7c4452b9 CM |
612 | if (err) |
613 | werr = err; | |
614 | return werr; | |
79154b1b CM |
615 | } |
616 | ||
690587d1 CM |
617 | /* |
618 | * when btree blocks are allocated, they have some corresponding bits set for | |
619 | * them in one of two extent_io trees. This is used to make sure all of | |
620 | * those extents are on disk for transaction or log commit | |
621 | */ | |
622 | int btrfs_write_and_wait_marked_extents(struct btrfs_root *root, | |
8cef4e16 | 623 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 CM |
624 | { |
625 | int ret; | |
626 | int ret2; | |
627 | ||
8cef4e16 YZ |
628 | ret = btrfs_write_marked_extents(root, dirty_pages, mark); |
629 | ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark); | |
690587d1 CM |
630 | return ret || ret2; |
631 | } | |
632 | ||
d0c803c4 CM |
633 | int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans, |
634 | struct btrfs_root *root) | |
635 | { | |
636 | if (!trans || !trans->transaction) { | |
637 | struct inode *btree_inode; | |
638 | btree_inode = root->fs_info->btree_inode; | |
639 | return filemap_write_and_wait(btree_inode->i_mapping); | |
640 | } | |
641 | return btrfs_write_and_wait_marked_extents(root, | |
8cef4e16 YZ |
642 | &trans->transaction->dirty_pages, |
643 | EXTENT_DIRTY); | |
d0c803c4 CM |
644 | } |
645 | ||
d352ac68 CM |
646 | /* |
647 | * this is used to update the root pointer in the tree of tree roots. | |
648 | * | |
649 | * But, in the case of the extent allocation tree, updating the root | |
650 | * pointer may allocate blocks which may change the root of the extent | |
651 | * allocation tree. | |
652 | * | |
653 | * So, this loops and repeats and makes sure the cowonly root didn't | |
654 | * change while the root pointer was being updated in the metadata. | |
655 | */ | |
0b86a832 CM |
656 | static int update_cowonly_root(struct btrfs_trans_handle *trans, |
657 | struct btrfs_root *root) | |
79154b1b CM |
658 | { |
659 | int ret; | |
0b86a832 | 660 | u64 old_root_bytenr; |
86b9f2ec | 661 | u64 old_root_used; |
0b86a832 | 662 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
79154b1b | 663 | |
86b9f2ec | 664 | old_root_used = btrfs_root_used(&root->root_item); |
0b86a832 | 665 | btrfs_write_dirty_block_groups(trans, root); |
56bec294 | 666 | |
d397712b | 667 | while (1) { |
0b86a832 | 668 | old_root_bytenr = btrfs_root_bytenr(&root->root_item); |
86b9f2ec YZ |
669 | if (old_root_bytenr == root->node->start && |
670 | old_root_used == btrfs_root_used(&root->root_item)) | |
79154b1b | 671 | break; |
87ef2bb4 | 672 | |
5d4f98a2 | 673 | btrfs_set_root_node(&root->root_item, root->node); |
79154b1b | 674 | ret = btrfs_update_root(trans, tree_root, |
0b86a832 CM |
675 | &root->root_key, |
676 | &root->root_item); | |
79154b1b | 677 | BUG_ON(ret); |
56bec294 | 678 | |
86b9f2ec | 679 | old_root_used = btrfs_root_used(&root->root_item); |
4a8c9a62 | 680 | ret = btrfs_write_dirty_block_groups(trans, root); |
56bec294 | 681 | BUG_ON(ret); |
0b86a832 | 682 | } |
276e680d YZ |
683 | |
684 | if (root != root->fs_info->extent_root) | |
685 | switch_commit_root(root); | |
686 | ||
0b86a832 CM |
687 | return 0; |
688 | } | |
689 | ||
d352ac68 CM |
690 | /* |
691 | * update all the cowonly tree roots on disk | |
692 | */ | |
5d4f98a2 YZ |
693 | static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans, |
694 | struct btrfs_root *root) | |
0b86a832 CM |
695 | { |
696 | struct btrfs_fs_info *fs_info = root->fs_info; | |
697 | struct list_head *next; | |
84234f3a | 698 | struct extent_buffer *eb; |
56bec294 | 699 | int ret; |
84234f3a | 700 | |
56bec294 CM |
701 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
702 | BUG_ON(ret); | |
87ef2bb4 | 703 | |
84234f3a | 704 | eb = btrfs_lock_root_node(fs_info->tree_root); |
9fa8cfe7 | 705 | btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 0, &eb); |
84234f3a YZ |
706 | btrfs_tree_unlock(eb); |
707 | free_extent_buffer(eb); | |
0b86a832 | 708 | |
56bec294 CM |
709 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
710 | BUG_ON(ret); | |
87ef2bb4 | 711 | |
d397712b | 712 | while (!list_empty(&fs_info->dirty_cowonly_roots)) { |
0b86a832 CM |
713 | next = fs_info->dirty_cowonly_roots.next; |
714 | list_del_init(next); | |
715 | root = list_entry(next, struct btrfs_root, dirty_list); | |
87ef2bb4 | 716 | |
0b86a832 | 717 | update_cowonly_root(trans, root); |
79154b1b | 718 | } |
276e680d YZ |
719 | |
720 | down_write(&fs_info->extent_commit_sem); | |
721 | switch_commit_root(fs_info->extent_root); | |
722 | up_write(&fs_info->extent_commit_sem); | |
723 | ||
79154b1b CM |
724 | return 0; |
725 | } | |
726 | ||
d352ac68 CM |
727 | /* |
728 | * dead roots are old snapshots that need to be deleted. This allocates | |
729 | * a dirty root struct and adds it into the list of dead roots that need to | |
730 | * be deleted | |
731 | */ | |
5d4f98a2 | 732 | int btrfs_add_dead_root(struct btrfs_root *root) |
5eda7b5e | 733 | { |
b48652c1 | 734 | mutex_lock(&root->fs_info->trans_mutex); |
5d4f98a2 | 735 | list_add(&root->root_list, &root->fs_info->dead_roots); |
b48652c1 | 736 | mutex_unlock(&root->fs_info->trans_mutex); |
5eda7b5e CM |
737 | return 0; |
738 | } | |
739 | ||
d352ac68 | 740 | /* |
5d4f98a2 | 741 | * update all the cowonly tree roots on disk |
d352ac68 | 742 | */ |
5d4f98a2 YZ |
743 | static noinline int commit_fs_roots(struct btrfs_trans_handle *trans, |
744 | struct btrfs_root *root) | |
0f7d52f4 | 745 | { |
0f7d52f4 | 746 | struct btrfs_root *gang[8]; |
5d4f98a2 | 747 | struct btrfs_fs_info *fs_info = root->fs_info; |
0f7d52f4 CM |
748 | int i; |
749 | int ret; | |
54aa1f4d CM |
750 | int err = 0; |
751 | ||
d397712b | 752 | while (1) { |
5d4f98a2 YZ |
753 | ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, |
754 | (void **)gang, 0, | |
0f7d52f4 CM |
755 | ARRAY_SIZE(gang), |
756 | BTRFS_ROOT_TRANS_TAG); | |
757 | if (ret == 0) | |
758 | break; | |
759 | for (i = 0; i < ret; i++) { | |
760 | root = gang[i]; | |
5d4f98a2 YZ |
761 | radix_tree_tag_clear(&fs_info->fs_roots_radix, |
762 | (unsigned long)root->root_key.objectid, | |
763 | BTRFS_ROOT_TRANS_TAG); | |
31153d81 | 764 | |
e02119d5 | 765 | btrfs_free_log(trans, root); |
5d4f98a2 | 766 | btrfs_update_reloc_root(trans, root); |
d68fc57b | 767 | btrfs_orphan_commit_root(trans, root); |
bcc63abb | 768 | |
978d910d | 769 | if (root->commit_root != root->node) { |
817d52f8 | 770 | switch_commit_root(root); |
978d910d YZ |
771 | btrfs_set_root_node(&root->root_item, |
772 | root->node); | |
773 | } | |
5d4f98a2 | 774 | |
5d4f98a2 | 775 | err = btrfs_update_root(trans, fs_info->tree_root, |
0f7d52f4 CM |
776 | &root->root_key, |
777 | &root->root_item); | |
54aa1f4d CM |
778 | if (err) |
779 | break; | |
0f7d52f4 CM |
780 | } |
781 | } | |
54aa1f4d | 782 | return err; |
0f7d52f4 CM |
783 | } |
784 | ||
d352ac68 CM |
785 | /* |
786 | * defrag a given btree. If cacheonly == 1, this won't read from the disk, | |
787 | * otherwise every leaf in the btree is read and defragged. | |
788 | */ | |
e9d0b13b CM |
789 | int btrfs_defrag_root(struct btrfs_root *root, int cacheonly) |
790 | { | |
791 | struct btrfs_fs_info *info = root->fs_info; | |
e9d0b13b | 792 | struct btrfs_trans_handle *trans; |
8929ecfa | 793 | int ret; |
d3c2fdcf | 794 | unsigned long nr; |
e9d0b13b | 795 | |
8929ecfa | 796 | if (xchg(&root->defrag_running, 1)) |
e9d0b13b | 797 | return 0; |
8929ecfa | 798 | |
6b80053d | 799 | while (1) { |
8929ecfa YZ |
800 | trans = btrfs_start_transaction(root, 0); |
801 | if (IS_ERR(trans)) | |
802 | return PTR_ERR(trans); | |
803 | ||
e9d0b13b | 804 | ret = btrfs_defrag_leaves(trans, root, cacheonly); |
8929ecfa | 805 | |
d3c2fdcf | 806 | nr = trans->blocks_used; |
e9d0b13b | 807 | btrfs_end_transaction(trans, root); |
d3c2fdcf | 808 | btrfs_btree_balance_dirty(info->tree_root, nr); |
e9d0b13b CM |
809 | cond_resched(); |
810 | ||
3f157a2f | 811 | if (root->fs_info->closing || ret != -EAGAIN) |
e9d0b13b CM |
812 | break; |
813 | } | |
814 | root->defrag_running = 0; | |
8929ecfa | 815 | return ret; |
e9d0b13b CM |
816 | } |
817 | ||
2c47e605 | 818 | #if 0 |
b7ec40d7 CM |
819 | /* |
820 | * when dropping snapshots, we generate a ton of delayed refs, and it makes | |
821 | * sense not to join the transaction while it is trying to flush the current | |
822 | * queue of delayed refs out. | |
823 | * | |
824 | * This is used by the drop snapshot code only | |
825 | */ | |
826 | static noinline int wait_transaction_pre_flush(struct btrfs_fs_info *info) | |
827 | { | |
828 | DEFINE_WAIT(wait); | |
829 | ||
830 | mutex_lock(&info->trans_mutex); | |
831 | while (info->running_transaction && | |
832 | info->running_transaction->delayed_refs.flushing) { | |
833 | prepare_to_wait(&info->transaction_wait, &wait, | |
834 | TASK_UNINTERRUPTIBLE); | |
835 | mutex_unlock(&info->trans_mutex); | |
59bc5c75 | 836 | |
b7ec40d7 | 837 | schedule(); |
59bc5c75 | 838 | |
b7ec40d7 CM |
839 | mutex_lock(&info->trans_mutex); |
840 | finish_wait(&info->transaction_wait, &wait); | |
841 | } | |
842 | mutex_unlock(&info->trans_mutex); | |
843 | return 0; | |
844 | } | |
845 | ||
d352ac68 CM |
846 | /* |
847 | * Given a list of roots that need to be deleted, call btrfs_drop_snapshot on | |
848 | * all of them | |
849 | */ | |
5d4f98a2 | 850 | int btrfs_drop_dead_root(struct btrfs_root *root) |
0f7d52f4 | 851 | { |
0f7d52f4 | 852 | struct btrfs_trans_handle *trans; |
5d4f98a2 | 853 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
d3c2fdcf | 854 | unsigned long nr; |
5d4f98a2 | 855 | int ret; |
58176a96 | 856 | |
5d4f98a2 YZ |
857 | while (1) { |
858 | /* | |
859 | * we don't want to jump in and create a bunch of | |
860 | * delayed refs if the transaction is starting to close | |
861 | */ | |
862 | wait_transaction_pre_flush(tree_root->fs_info); | |
863 | trans = btrfs_start_transaction(tree_root, 1); | |
a2135011 | 864 | |
5d4f98a2 YZ |
865 | /* |
866 | * we've joined a transaction, make sure it isn't | |
867 | * closing right now | |
868 | */ | |
869 | if (trans->transaction->delayed_refs.flushing) { | |
870 | btrfs_end_transaction(trans, tree_root); | |
871 | continue; | |
9f3a7427 | 872 | } |
58176a96 | 873 | |
5d4f98a2 YZ |
874 | ret = btrfs_drop_snapshot(trans, root); |
875 | if (ret != -EAGAIN) | |
876 | break; | |
a2135011 | 877 | |
5d4f98a2 YZ |
878 | ret = btrfs_update_root(trans, tree_root, |
879 | &root->root_key, | |
880 | &root->root_item); | |
881 | if (ret) | |
54aa1f4d | 882 | break; |
bcc63abb | 883 | |
d3c2fdcf | 884 | nr = trans->blocks_used; |
0f7d52f4 CM |
885 | ret = btrfs_end_transaction(trans, tree_root); |
886 | BUG_ON(ret); | |
5eda7b5e | 887 | |
d3c2fdcf | 888 | btrfs_btree_balance_dirty(tree_root, nr); |
4dc11904 | 889 | cond_resched(); |
0f7d52f4 | 890 | } |
5d4f98a2 YZ |
891 | BUG_ON(ret); |
892 | ||
893 | ret = btrfs_del_root(trans, tree_root, &root->root_key); | |
894 | BUG_ON(ret); | |
895 | ||
896 | nr = trans->blocks_used; | |
897 | ret = btrfs_end_transaction(trans, tree_root); | |
898 | BUG_ON(ret); | |
899 | ||
900 | free_extent_buffer(root->node); | |
901 | free_extent_buffer(root->commit_root); | |
902 | kfree(root); | |
903 | ||
904 | btrfs_btree_balance_dirty(tree_root, nr); | |
54aa1f4d | 905 | return ret; |
0f7d52f4 | 906 | } |
2c47e605 | 907 | #endif |
0f7d52f4 | 908 | |
d352ac68 CM |
909 | /* |
910 | * new snapshots need to be created at a very specific time in the | |
911 | * transaction commit. This does the actual creation | |
912 | */ | |
80b6794d | 913 | static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, |
3063d29f CM |
914 | struct btrfs_fs_info *fs_info, |
915 | struct btrfs_pending_snapshot *pending) | |
916 | { | |
917 | struct btrfs_key key; | |
80b6794d | 918 | struct btrfs_root_item *new_root_item; |
3063d29f CM |
919 | struct btrfs_root *tree_root = fs_info->tree_root; |
920 | struct btrfs_root *root = pending->root; | |
6bdb72de SW |
921 | struct btrfs_root *parent_root; |
922 | struct inode *parent_inode; | |
6a912213 | 923 | struct dentry *parent; |
a22285a6 | 924 | struct dentry *dentry; |
3063d29f | 925 | struct extent_buffer *tmp; |
925baedd | 926 | struct extent_buffer *old; |
3063d29f | 927 | int ret; |
d68fc57b | 928 | u64 to_reserve = 0; |
6bdb72de | 929 | u64 index = 0; |
a22285a6 | 930 | u64 objectid; |
b83cc969 | 931 | u64 root_flags; |
3063d29f | 932 | |
80b6794d CM |
933 | new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS); |
934 | if (!new_root_item) { | |
a22285a6 | 935 | pending->error = -ENOMEM; |
80b6794d CM |
936 | goto fail; |
937 | } | |
a22285a6 | 938 | |
3063d29f | 939 | ret = btrfs_find_free_objectid(trans, tree_root, 0, &objectid); |
a22285a6 YZ |
940 | if (ret) { |
941 | pending->error = ret; | |
3063d29f | 942 | goto fail; |
a22285a6 | 943 | } |
3063d29f | 944 | |
3fd0a558 | 945 | btrfs_reloc_pre_snapshot(trans, pending, &to_reserve); |
d68fc57b YZ |
946 | btrfs_orphan_pre_snapshot(trans, pending, &to_reserve); |
947 | ||
948 | if (to_reserve > 0) { | |
949 | ret = btrfs_block_rsv_add(trans, root, &pending->block_rsv, | |
8bb8ab2e | 950 | to_reserve); |
d68fc57b YZ |
951 | if (ret) { |
952 | pending->error = ret; | |
953 | goto fail; | |
954 | } | |
955 | } | |
956 | ||
3063d29f | 957 | key.objectid = objectid; |
a22285a6 YZ |
958 | key.offset = (u64)-1; |
959 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3063d29f | 960 | |
a22285a6 | 961 | trans->block_rsv = &pending->block_rsv; |
3de4586c | 962 | |
a22285a6 | 963 | dentry = pending->dentry; |
6a912213 JB |
964 | parent = dget_parent(dentry); |
965 | parent_inode = parent->d_inode; | |
a22285a6 | 966 | parent_root = BTRFS_I(parent_inode)->root; |
6bdb72de | 967 | record_root_in_trans(trans, parent_root); |
a22285a6 | 968 | |
3063d29f CM |
969 | /* |
970 | * insert the directory item | |
971 | */ | |
3de4586c | 972 | ret = btrfs_set_inode_index(parent_inode, &index); |
6bdb72de | 973 | BUG_ON(ret); |
0660b5af | 974 | ret = btrfs_insert_dir_item(trans, parent_root, |
a22285a6 YZ |
975 | dentry->d_name.name, dentry->d_name.len, |
976 | parent_inode->i_ino, &key, | |
977 | BTRFS_FT_DIR, index); | |
6bdb72de | 978 | BUG_ON(ret); |
0660b5af | 979 | |
a22285a6 YZ |
980 | btrfs_i_size_write(parent_inode, parent_inode->i_size + |
981 | dentry->d_name.len * 2); | |
52c26179 YZ |
982 | ret = btrfs_update_inode(trans, parent_root, parent_inode); |
983 | BUG_ON(ret); | |
984 | ||
6bdb72de SW |
985 | record_root_in_trans(trans, root); |
986 | btrfs_set_root_last_snapshot(&root->root_item, trans->transid); | |
987 | memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); | |
08fe4db1 | 988 | btrfs_check_and_init_root_item(new_root_item); |
6bdb72de | 989 | |
b83cc969 LZ |
990 | root_flags = btrfs_root_flags(new_root_item); |
991 | if (pending->readonly) | |
992 | root_flags |= BTRFS_ROOT_SUBVOL_RDONLY; | |
993 | else | |
994 | root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY; | |
995 | btrfs_set_root_flags(new_root_item, root_flags); | |
996 | ||
6bdb72de SW |
997 | old = btrfs_lock_root_node(root); |
998 | btrfs_cow_block(trans, root, old, NULL, 0, &old); | |
999 | btrfs_set_lock_blocking(old); | |
1000 | ||
1001 | btrfs_copy_root(trans, root, old, &tmp, objectid); | |
1002 | btrfs_tree_unlock(old); | |
1003 | free_extent_buffer(old); | |
1004 | ||
1005 | btrfs_set_root_node(new_root_item, tmp); | |
a22285a6 YZ |
1006 | /* record when the snapshot was created in key.offset */ |
1007 | key.offset = trans->transid; | |
1008 | ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); | |
6bdb72de SW |
1009 | btrfs_tree_unlock(tmp); |
1010 | free_extent_buffer(tmp); | |
a22285a6 | 1011 | BUG_ON(ret); |
6bdb72de | 1012 | |
a22285a6 YZ |
1013 | /* |
1014 | * insert root back/forward references | |
1015 | */ | |
1016 | ret = btrfs_add_root_ref(trans, tree_root, objectid, | |
0660b5af | 1017 | parent_root->root_key.objectid, |
a22285a6 YZ |
1018 | parent_inode->i_ino, index, |
1019 | dentry->d_name.name, dentry->d_name.len); | |
0660b5af | 1020 | BUG_ON(ret); |
6a912213 | 1021 | dput(parent); |
0660b5af | 1022 | |
a22285a6 YZ |
1023 | key.offset = (u64)-1; |
1024 | pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key); | |
1025 | BUG_ON(IS_ERR(pending->snap)); | |
d68fc57b | 1026 | |
3fd0a558 | 1027 | btrfs_reloc_post_snapshot(trans, pending); |
d68fc57b | 1028 | btrfs_orphan_post_snapshot(trans, pending); |
3063d29f | 1029 | fail: |
6bdb72de | 1030 | kfree(new_root_item); |
a22285a6 YZ |
1031 | btrfs_block_rsv_release(root, &pending->block_rsv, (u64)-1); |
1032 | return 0; | |
3063d29f CM |
1033 | } |
1034 | ||
d352ac68 CM |
1035 | /* |
1036 | * create all the snapshots we've scheduled for creation | |
1037 | */ | |
80b6794d CM |
1038 | static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans, |
1039 | struct btrfs_fs_info *fs_info) | |
3de4586c CM |
1040 | { |
1041 | struct btrfs_pending_snapshot *pending; | |
1042 | struct list_head *head = &trans->transaction->pending_snapshots; | |
3de4586c CM |
1043 | int ret; |
1044 | ||
c6e30871 | 1045 | list_for_each_entry(pending, head, list) { |
3de4586c CM |
1046 | ret = create_pending_snapshot(trans, fs_info, pending); |
1047 | BUG_ON(ret); | |
1048 | } | |
1049 | return 0; | |
1050 | } | |
1051 | ||
5d4f98a2 YZ |
1052 | static void update_super_roots(struct btrfs_root *root) |
1053 | { | |
1054 | struct btrfs_root_item *root_item; | |
1055 | struct btrfs_super_block *super; | |
1056 | ||
1057 | super = &root->fs_info->super_copy; | |
1058 | ||
1059 | root_item = &root->fs_info->chunk_root->root_item; | |
1060 | super->chunk_root = root_item->bytenr; | |
1061 | super->chunk_root_generation = root_item->generation; | |
1062 | super->chunk_root_level = root_item->level; | |
1063 | ||
1064 | root_item = &root->fs_info->tree_root->root_item; | |
1065 | super->root = root_item->bytenr; | |
1066 | super->generation = root_item->generation; | |
1067 | super->root_level = root_item->level; | |
0af3d00b JB |
1068 | if (super->cache_generation != 0 || btrfs_test_opt(root, SPACE_CACHE)) |
1069 | super->cache_generation = root_item->generation; | |
5d4f98a2 YZ |
1070 | } |
1071 | ||
f36f3042 CM |
1072 | int btrfs_transaction_in_commit(struct btrfs_fs_info *info) |
1073 | { | |
1074 | int ret = 0; | |
1075 | spin_lock(&info->new_trans_lock); | |
1076 | if (info->running_transaction) | |
1077 | ret = info->running_transaction->in_commit; | |
1078 | spin_unlock(&info->new_trans_lock); | |
1079 | return ret; | |
1080 | } | |
1081 | ||
8929ecfa YZ |
1082 | int btrfs_transaction_blocked(struct btrfs_fs_info *info) |
1083 | { | |
1084 | int ret = 0; | |
1085 | spin_lock(&info->new_trans_lock); | |
1086 | if (info->running_transaction) | |
1087 | ret = info->running_transaction->blocked; | |
1088 | spin_unlock(&info->new_trans_lock); | |
1089 | return ret; | |
1090 | } | |
1091 | ||
bb9c12c9 SW |
1092 | /* |
1093 | * wait for the current transaction commit to start and block subsequent | |
1094 | * transaction joins | |
1095 | */ | |
1096 | static void wait_current_trans_commit_start(struct btrfs_root *root, | |
1097 | struct btrfs_transaction *trans) | |
1098 | { | |
1099 | DEFINE_WAIT(wait); | |
1100 | ||
1101 | if (trans->in_commit) | |
1102 | return; | |
1103 | ||
1104 | while (1) { | |
1105 | prepare_to_wait(&root->fs_info->transaction_blocked_wait, &wait, | |
1106 | TASK_UNINTERRUPTIBLE); | |
1107 | if (trans->in_commit) { | |
1108 | finish_wait(&root->fs_info->transaction_blocked_wait, | |
1109 | &wait); | |
1110 | break; | |
1111 | } | |
1112 | mutex_unlock(&root->fs_info->trans_mutex); | |
1113 | schedule(); | |
1114 | mutex_lock(&root->fs_info->trans_mutex); | |
1115 | finish_wait(&root->fs_info->transaction_blocked_wait, &wait); | |
1116 | } | |
1117 | } | |
1118 | ||
1119 | /* | |
1120 | * wait for the current transaction to start and then become unblocked. | |
1121 | * caller holds ref. | |
1122 | */ | |
1123 | static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root, | |
1124 | struct btrfs_transaction *trans) | |
1125 | { | |
1126 | DEFINE_WAIT(wait); | |
1127 | ||
1128 | if (trans->commit_done || (trans->in_commit && !trans->blocked)) | |
1129 | return; | |
1130 | ||
1131 | while (1) { | |
1132 | prepare_to_wait(&root->fs_info->transaction_wait, &wait, | |
1133 | TASK_UNINTERRUPTIBLE); | |
1134 | if (trans->commit_done || | |
1135 | (trans->in_commit && !trans->blocked)) { | |
1136 | finish_wait(&root->fs_info->transaction_wait, | |
1137 | &wait); | |
1138 | break; | |
1139 | } | |
1140 | mutex_unlock(&root->fs_info->trans_mutex); | |
1141 | schedule(); | |
1142 | mutex_lock(&root->fs_info->trans_mutex); | |
1143 | finish_wait(&root->fs_info->transaction_wait, | |
1144 | &wait); | |
1145 | } | |
1146 | } | |
1147 | ||
1148 | /* | |
1149 | * commit transactions asynchronously. once btrfs_commit_transaction_async | |
1150 | * returns, any subsequent transaction will not be allowed to join. | |
1151 | */ | |
1152 | struct btrfs_async_commit { | |
1153 | struct btrfs_trans_handle *newtrans; | |
1154 | struct btrfs_root *root; | |
1155 | struct delayed_work work; | |
1156 | }; | |
1157 | ||
1158 | static void do_async_commit(struct work_struct *work) | |
1159 | { | |
1160 | struct btrfs_async_commit *ac = | |
1161 | container_of(work, struct btrfs_async_commit, work.work); | |
1162 | ||
1163 | btrfs_commit_transaction(ac->newtrans, ac->root); | |
1164 | kfree(ac); | |
1165 | } | |
1166 | ||
1167 | int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans, | |
1168 | struct btrfs_root *root, | |
1169 | int wait_for_unblock) | |
1170 | { | |
1171 | struct btrfs_async_commit *ac; | |
1172 | struct btrfs_transaction *cur_trans; | |
1173 | ||
1174 | ac = kmalloc(sizeof(*ac), GFP_NOFS); | |
db5b493a TI |
1175 | if (!ac) |
1176 | return -ENOMEM; | |
bb9c12c9 SW |
1177 | |
1178 | INIT_DELAYED_WORK(&ac->work, do_async_commit); | |
1179 | ac->root = root; | |
1180 | ac->newtrans = btrfs_join_transaction(root, 0); | |
3612b495 TI |
1181 | if (IS_ERR(ac->newtrans)) { |
1182 | int err = PTR_ERR(ac->newtrans); | |
1183 | kfree(ac); | |
1184 | return err; | |
1185 | } | |
bb9c12c9 SW |
1186 | |
1187 | /* take transaction reference */ | |
1188 | mutex_lock(&root->fs_info->trans_mutex); | |
1189 | cur_trans = trans->transaction; | |
1190 | cur_trans->use_count++; | |
1191 | mutex_unlock(&root->fs_info->trans_mutex); | |
1192 | ||
1193 | btrfs_end_transaction(trans, root); | |
1194 | schedule_delayed_work(&ac->work, 0); | |
1195 | ||
1196 | /* wait for transaction to start and unblock */ | |
1197 | mutex_lock(&root->fs_info->trans_mutex); | |
1198 | if (wait_for_unblock) | |
1199 | wait_current_trans_commit_start_and_unblock(root, cur_trans); | |
1200 | else | |
1201 | wait_current_trans_commit_start(root, cur_trans); | |
1202 | put_transaction(cur_trans); | |
1203 | mutex_unlock(&root->fs_info->trans_mutex); | |
1204 | ||
1205 | return 0; | |
1206 | } | |
1207 | ||
1208 | /* | |
1209 | * btrfs_transaction state sequence: | |
1210 | * in_commit = 0, blocked = 0 (initial) | |
1211 | * in_commit = 1, blocked = 1 | |
1212 | * blocked = 0 | |
1213 | * commit_done = 1 | |
1214 | */ | |
79154b1b CM |
1215 | int btrfs_commit_transaction(struct btrfs_trans_handle *trans, |
1216 | struct btrfs_root *root) | |
1217 | { | |
15ee9bc7 | 1218 | unsigned long joined = 0; |
79154b1b | 1219 | struct btrfs_transaction *cur_trans; |
8fd17795 | 1220 | struct btrfs_transaction *prev_trans = NULL; |
79154b1b | 1221 | DEFINE_WAIT(wait); |
15ee9bc7 | 1222 | int ret; |
89573b9c CM |
1223 | int should_grow = 0; |
1224 | unsigned long now = get_seconds(); | |
dccae999 | 1225 | int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT); |
79154b1b | 1226 | |
5a3f23d5 CM |
1227 | btrfs_run_ordered_operations(root, 0); |
1228 | ||
56bec294 CM |
1229 | /* make a pass through all the delayed refs we have so far |
1230 | * any runnings procs may add more while we are here | |
1231 | */ | |
1232 | ret = btrfs_run_delayed_refs(trans, root, 0); | |
1233 | BUG_ON(ret); | |
1234 | ||
a22285a6 YZ |
1235 | btrfs_trans_release_metadata(trans, root); |
1236 | ||
b7ec40d7 | 1237 | cur_trans = trans->transaction; |
56bec294 CM |
1238 | /* |
1239 | * set the flushing flag so procs in this transaction have to | |
1240 | * start sending their work down. | |
1241 | */ | |
b7ec40d7 | 1242 | cur_trans->delayed_refs.flushing = 1; |
56bec294 | 1243 | |
c3e69d58 | 1244 | ret = btrfs_run_delayed_refs(trans, root, 0); |
56bec294 CM |
1245 | BUG_ON(ret); |
1246 | ||
79154b1b | 1247 | mutex_lock(&root->fs_info->trans_mutex); |
b7ec40d7 CM |
1248 | if (cur_trans->in_commit) { |
1249 | cur_trans->use_count++; | |
ccd467d6 | 1250 | mutex_unlock(&root->fs_info->trans_mutex); |
79154b1b | 1251 | btrfs_end_transaction(trans, root); |
ccd467d6 | 1252 | |
79154b1b CM |
1253 | ret = wait_for_commit(root, cur_trans); |
1254 | BUG_ON(ret); | |
15ee9bc7 JB |
1255 | |
1256 | mutex_lock(&root->fs_info->trans_mutex); | |
79154b1b | 1257 | put_transaction(cur_trans); |
15ee9bc7 JB |
1258 | mutex_unlock(&root->fs_info->trans_mutex); |
1259 | ||
79154b1b CM |
1260 | return 0; |
1261 | } | |
4313b399 | 1262 | |
2c90e5d6 | 1263 | trans->transaction->in_commit = 1; |
f9295749 | 1264 | trans->transaction->blocked = 1; |
bb9c12c9 SW |
1265 | wake_up(&root->fs_info->transaction_blocked_wait); |
1266 | ||
ccd467d6 CM |
1267 | if (cur_trans->list.prev != &root->fs_info->trans_list) { |
1268 | prev_trans = list_entry(cur_trans->list.prev, | |
1269 | struct btrfs_transaction, list); | |
1270 | if (!prev_trans->commit_done) { | |
1271 | prev_trans->use_count++; | |
ccd467d6 CM |
1272 | mutex_unlock(&root->fs_info->trans_mutex); |
1273 | ||
1274 | wait_for_commit(root, prev_trans); | |
ccd467d6 | 1275 | |
ccd467d6 | 1276 | mutex_lock(&root->fs_info->trans_mutex); |
15ee9bc7 | 1277 | put_transaction(prev_trans); |
ccd467d6 CM |
1278 | } |
1279 | } | |
15ee9bc7 | 1280 | |
89573b9c CM |
1281 | if (now < cur_trans->start_time || now - cur_trans->start_time < 1) |
1282 | should_grow = 1; | |
1283 | ||
15ee9bc7 | 1284 | do { |
7ea394f1 | 1285 | int snap_pending = 0; |
15ee9bc7 | 1286 | joined = cur_trans->num_joined; |
7ea394f1 YZ |
1287 | if (!list_empty(&trans->transaction->pending_snapshots)) |
1288 | snap_pending = 1; | |
1289 | ||
2c90e5d6 | 1290 | WARN_ON(cur_trans != trans->transaction); |
79154b1b | 1291 | mutex_unlock(&root->fs_info->trans_mutex); |
15ee9bc7 | 1292 | |
0bdb1db2 | 1293 | if (flush_on_commit || snap_pending) { |
24bbcf04 YZ |
1294 | btrfs_start_delalloc_inodes(root, 1); |
1295 | ret = btrfs_wait_ordered_extents(root, 0, 1); | |
ebecd3d9 | 1296 | BUG_ON(ret); |
7ea394f1 YZ |
1297 | } |
1298 | ||
5a3f23d5 CM |
1299 | /* |
1300 | * rename don't use btrfs_join_transaction, so, once we | |
1301 | * set the transaction to blocked above, we aren't going | |
1302 | * to get any new ordered operations. We can safely run | |
1303 | * it here and no for sure that nothing new will be added | |
1304 | * to the list | |
1305 | */ | |
1306 | btrfs_run_ordered_operations(root, 1); | |
1307 | ||
ed3b3d31 CM |
1308 | prepare_to_wait(&cur_trans->writer_wait, &wait, |
1309 | TASK_UNINTERRUPTIBLE); | |
1310 | ||
89573b9c | 1311 | smp_mb(); |
99d16cbc SW |
1312 | if (cur_trans->num_writers > 1) |
1313 | schedule_timeout(MAX_SCHEDULE_TIMEOUT); | |
1314 | else if (should_grow) | |
1315 | schedule_timeout(1); | |
15ee9bc7 | 1316 | |
79154b1b | 1317 | mutex_lock(&root->fs_info->trans_mutex); |
15ee9bc7 JB |
1318 | finish_wait(&cur_trans->writer_wait, &wait); |
1319 | } while (cur_trans->num_writers > 1 || | |
89573b9c | 1320 | (should_grow && cur_trans->num_joined != joined)); |
15ee9bc7 | 1321 | |
3063d29f CM |
1322 | ret = create_pending_snapshots(trans, root->fs_info); |
1323 | BUG_ON(ret); | |
1324 | ||
56bec294 CM |
1325 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
1326 | BUG_ON(ret); | |
1327 | ||
2c90e5d6 | 1328 | WARN_ON(cur_trans != trans->transaction); |
dc17ff8f | 1329 | |
e02119d5 CM |
1330 | /* btrfs_commit_tree_roots is responsible for getting the |
1331 | * various roots consistent with each other. Every pointer | |
1332 | * in the tree of tree roots has to point to the most up to date | |
1333 | * root for every subvolume and other tree. So, we have to keep | |
1334 | * the tree logging code from jumping in and changing any | |
1335 | * of the trees. | |
1336 | * | |
1337 | * At this point in the commit, there can't be any tree-log | |
1338 | * writers, but a little lower down we drop the trans mutex | |
1339 | * and let new people in. By holding the tree_log_mutex | |
1340 | * from now until after the super is written, we avoid races | |
1341 | * with the tree-log code. | |
1342 | */ | |
1343 | mutex_lock(&root->fs_info->tree_log_mutex); | |
1344 | ||
5d4f98a2 | 1345 | ret = commit_fs_roots(trans, root); |
54aa1f4d CM |
1346 | BUG_ON(ret); |
1347 | ||
5d4f98a2 | 1348 | /* commit_fs_roots gets rid of all the tree log roots, it is now |
e02119d5 CM |
1349 | * safe to free the root of tree log roots |
1350 | */ | |
1351 | btrfs_free_log_root_tree(trans, root->fs_info); | |
1352 | ||
5d4f98a2 | 1353 | ret = commit_cowonly_roots(trans, root); |
79154b1b | 1354 | BUG_ON(ret); |
54aa1f4d | 1355 | |
11833d66 YZ |
1356 | btrfs_prepare_extent_commit(trans, root); |
1357 | ||
78fae27e | 1358 | cur_trans = root->fs_info->running_transaction; |
cee36a03 | 1359 | spin_lock(&root->fs_info->new_trans_lock); |
78fae27e | 1360 | root->fs_info->running_transaction = NULL; |
cee36a03 | 1361 | spin_unlock(&root->fs_info->new_trans_lock); |
5d4f98a2 YZ |
1362 | |
1363 | btrfs_set_root_node(&root->fs_info->tree_root->root_item, | |
1364 | root->fs_info->tree_root->node); | |
817d52f8 | 1365 | switch_commit_root(root->fs_info->tree_root); |
5d4f98a2 YZ |
1366 | |
1367 | btrfs_set_root_node(&root->fs_info->chunk_root->root_item, | |
1368 | root->fs_info->chunk_root->node); | |
817d52f8 | 1369 | switch_commit_root(root->fs_info->chunk_root); |
5d4f98a2 YZ |
1370 | |
1371 | update_super_roots(root); | |
e02119d5 CM |
1372 | |
1373 | if (!root->fs_info->log_root_recovering) { | |
1374 | btrfs_set_super_log_root(&root->fs_info->super_copy, 0); | |
1375 | btrfs_set_super_log_root_level(&root->fs_info->super_copy, 0); | |
1376 | } | |
1377 | ||
a061fc8d CM |
1378 | memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy, |
1379 | sizeof(root->fs_info->super_copy)); | |
ccd467d6 | 1380 | |
f9295749 | 1381 | trans->transaction->blocked = 0; |
b7ec40d7 | 1382 | |
f9295749 | 1383 | wake_up(&root->fs_info->transaction_wait); |
e6dcd2dc | 1384 | |
78fae27e | 1385 | mutex_unlock(&root->fs_info->trans_mutex); |
79154b1b CM |
1386 | ret = btrfs_write_and_wait_transaction(trans, root); |
1387 | BUG_ON(ret); | |
a512bbf8 | 1388 | write_ctree_super(trans, root, 0); |
4313b399 | 1389 | |
e02119d5 CM |
1390 | /* |
1391 | * the super is written, we can safely allow the tree-loggers | |
1392 | * to go about their business | |
1393 | */ | |
1394 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
1395 | ||
11833d66 | 1396 | btrfs_finish_extent_commit(trans, root); |
4313b399 | 1397 | |
1a40e23b ZY |
1398 | mutex_lock(&root->fs_info->trans_mutex); |
1399 | ||
2c90e5d6 | 1400 | cur_trans->commit_done = 1; |
b7ec40d7 | 1401 | |
15ee9bc7 | 1402 | root->fs_info->last_trans_committed = cur_trans->transid; |
817d52f8 | 1403 | |
2c90e5d6 | 1404 | wake_up(&cur_trans->commit_wait); |
3de4586c | 1405 | |
78fae27e | 1406 | put_transaction(cur_trans); |
79154b1b | 1407 | put_transaction(cur_trans); |
58176a96 | 1408 | |
1abe9b8a | 1409 | trace_btrfs_transaction_commit(root); |
1410 | ||
78fae27e | 1411 | mutex_unlock(&root->fs_info->trans_mutex); |
3de4586c | 1412 | |
9ed74f2d JB |
1413 | if (current->journal_info == trans) |
1414 | current->journal_info = NULL; | |
1415 | ||
2c90e5d6 | 1416 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
24bbcf04 YZ |
1417 | |
1418 | if (current != root->fs_info->transaction_kthread) | |
1419 | btrfs_run_delayed_iputs(root); | |
1420 | ||
79154b1b CM |
1421 | return ret; |
1422 | } | |
1423 | ||
d352ac68 CM |
1424 | /* |
1425 | * interface function to delete all the snapshots we have scheduled for deletion | |
1426 | */ | |
e9d0b13b CM |
1427 | int btrfs_clean_old_snapshots(struct btrfs_root *root) |
1428 | { | |
5d4f98a2 YZ |
1429 | LIST_HEAD(list); |
1430 | struct btrfs_fs_info *fs_info = root->fs_info; | |
1431 | ||
1432 | mutex_lock(&fs_info->trans_mutex); | |
1433 | list_splice_init(&fs_info->dead_roots, &list); | |
1434 | mutex_unlock(&fs_info->trans_mutex); | |
e9d0b13b | 1435 | |
5d4f98a2 YZ |
1436 | while (!list_empty(&list)) { |
1437 | root = list_entry(list.next, struct btrfs_root, root_list); | |
76dda93c YZ |
1438 | list_del(&root->root_list); |
1439 | ||
1440 | if (btrfs_header_backref_rev(root->node) < | |
1441 | BTRFS_MIXED_BACKREF_REV) | |
3fd0a558 | 1442 | btrfs_drop_snapshot(root, NULL, 0); |
76dda93c | 1443 | else |
3fd0a558 | 1444 | btrfs_drop_snapshot(root, NULL, 1); |
e9d0b13b CM |
1445 | } |
1446 | return 0; | |
1447 | } |