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" |
581bb050 | 30 | #include "inode-map.h" |
79154b1b | 31 | |
0f7d52f4 CM |
32 | #define BTRFS_ROOT_TRANS_TAG 0 |
33 | ||
80b6794d | 34 | static noinline void put_transaction(struct btrfs_transaction *transaction) |
79154b1b | 35 | { |
13c5a93e JB |
36 | WARN_ON(atomic_read(&transaction->use_count) == 0); |
37 | if (atomic_dec_and_test(&transaction->use_count)) { | |
a4abeea4 | 38 | BUG_ON(!list_empty(&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 | */ | |
a4abeea4 | 53 | static noinline int join_transaction(struct btrfs_root *root, int nofail) |
79154b1b CM |
54 | { |
55 | struct btrfs_transaction *cur_trans; | |
a4abeea4 JB |
56 | |
57 | spin_lock(&root->fs_info->trans_lock); | |
58 | if (root->fs_info->trans_no_join) { | |
59 | if (!nofail) { | |
60 | spin_unlock(&root->fs_info->trans_lock); | |
61 | return -EBUSY; | |
62 | } | |
63 | } | |
64 | ||
79154b1b | 65 | cur_trans = root->fs_info->running_transaction; |
a4abeea4 JB |
66 | if (cur_trans) { |
67 | atomic_inc(&cur_trans->use_count); | |
13c5a93e | 68 | atomic_inc(&cur_trans->num_writers); |
15ee9bc7 | 69 | cur_trans->num_joined++; |
a4abeea4 JB |
70 | spin_unlock(&root->fs_info->trans_lock); |
71 | return 0; | |
79154b1b | 72 | } |
a4abeea4 JB |
73 | spin_unlock(&root->fs_info->trans_lock); |
74 | ||
75 | cur_trans = kmem_cache_alloc(btrfs_transaction_cachep, GFP_NOFS); | |
76 | if (!cur_trans) | |
77 | return -ENOMEM; | |
78 | spin_lock(&root->fs_info->trans_lock); | |
79 | if (root->fs_info->running_transaction) { | |
80 | kmem_cache_free(btrfs_transaction_cachep, cur_trans); | |
81 | cur_trans = root->fs_info->running_transaction; | |
82 | atomic_inc(&cur_trans->use_count); | |
13c5a93e | 83 | atomic_inc(&cur_trans->num_writers); |
15ee9bc7 | 84 | cur_trans->num_joined++; |
a4abeea4 JB |
85 | spin_unlock(&root->fs_info->trans_lock); |
86 | return 0; | |
79154b1b | 87 | } |
a4abeea4 JB |
88 | atomic_set(&cur_trans->num_writers, 1); |
89 | cur_trans->num_joined = 0; | |
90 | init_waitqueue_head(&cur_trans->writer_wait); | |
91 | init_waitqueue_head(&cur_trans->commit_wait); | |
92 | cur_trans->in_commit = 0; | |
93 | cur_trans->blocked = 0; | |
94 | /* | |
95 | * One for this trans handle, one so it will live on until we | |
96 | * commit the transaction. | |
97 | */ | |
98 | atomic_set(&cur_trans->use_count, 2); | |
99 | cur_trans->commit_done = 0; | |
100 | cur_trans->start_time = get_seconds(); | |
101 | ||
102 | cur_trans->delayed_refs.root = RB_ROOT; | |
103 | cur_trans->delayed_refs.num_entries = 0; | |
104 | cur_trans->delayed_refs.num_heads_ready = 0; | |
105 | cur_trans->delayed_refs.num_heads = 0; | |
106 | cur_trans->delayed_refs.flushing = 0; | |
107 | cur_trans->delayed_refs.run_delayed_start = 0; | |
108 | spin_lock_init(&cur_trans->commit_lock); | |
109 | spin_lock_init(&cur_trans->delayed_refs.lock); | |
110 | ||
111 | INIT_LIST_HEAD(&cur_trans->pending_snapshots); | |
112 | list_add_tail(&cur_trans->list, &root->fs_info->trans_list); | |
113 | extent_io_tree_init(&cur_trans->dirty_pages, | |
ff5714cc | 114 | root->fs_info->btree_inode->i_mapping); |
a4abeea4 JB |
115 | root->fs_info->generation++; |
116 | cur_trans->transid = root->fs_info->generation; | |
117 | root->fs_info->running_transaction = cur_trans; | |
118 | spin_unlock(&root->fs_info->trans_lock); | |
15ee9bc7 | 119 | |
79154b1b CM |
120 | return 0; |
121 | } | |
122 | ||
d352ac68 | 123 | /* |
d397712b CM |
124 | * this does all the record keeping required to make sure that a reference |
125 | * counted root is properly recorded in a given transaction. This is required | |
126 | * to make sure the old root from before we joined the transaction is deleted | |
127 | * when the transaction commits | |
d352ac68 | 128 | */ |
7585717f | 129 | static int record_root_in_trans(struct btrfs_trans_handle *trans, |
a4abeea4 | 130 | struct btrfs_root *root) |
6702ed49 | 131 | { |
5d4f98a2 | 132 | if (root->ref_cows && root->last_trans < trans->transid) { |
6702ed49 | 133 | WARN_ON(root == root->fs_info->extent_root); |
5d4f98a2 YZ |
134 | WARN_ON(root->commit_root != root->node); |
135 | ||
7585717f CM |
136 | /* |
137 | * see below for in_trans_setup usage rules | |
138 | * we have the reloc mutex held now, so there | |
139 | * is only one writer in this function | |
140 | */ | |
141 | root->in_trans_setup = 1; | |
142 | ||
143 | /* make sure readers find in_trans_setup before | |
144 | * they find our root->last_trans update | |
145 | */ | |
146 | smp_wmb(); | |
147 | ||
a4abeea4 JB |
148 | spin_lock(&root->fs_info->fs_roots_radix_lock); |
149 | if (root->last_trans == trans->transid) { | |
150 | spin_unlock(&root->fs_info->fs_roots_radix_lock); | |
151 | return 0; | |
152 | } | |
5d4f98a2 YZ |
153 | radix_tree_tag_set(&root->fs_info->fs_roots_radix, |
154 | (unsigned long)root->root_key.objectid, | |
155 | BTRFS_ROOT_TRANS_TAG); | |
a4abeea4 | 156 | spin_unlock(&root->fs_info->fs_roots_radix_lock); |
7585717f CM |
157 | root->last_trans = trans->transid; |
158 | ||
159 | /* this is pretty tricky. We don't want to | |
160 | * take the relocation lock in btrfs_record_root_in_trans | |
161 | * unless we're really doing the first setup for this root in | |
162 | * this transaction. | |
163 | * | |
164 | * Normally we'd use root->last_trans as a flag to decide | |
165 | * if we want to take the expensive mutex. | |
166 | * | |
167 | * But, we have to set root->last_trans before we | |
168 | * init the relocation root, otherwise, we trip over warnings | |
169 | * in ctree.c. The solution used here is to flag ourselves | |
170 | * with root->in_trans_setup. When this is 1, we're still | |
171 | * fixing up the reloc trees and everyone must wait. | |
172 | * | |
173 | * When this is zero, they can trust root->last_trans and fly | |
174 | * through btrfs_record_root_in_trans without having to take the | |
175 | * lock. smp_wmb() makes sure that all the writes above are | |
176 | * done before we pop in the zero below | |
177 | */ | |
5d4f98a2 | 178 | btrfs_init_reloc_root(trans, root); |
7585717f CM |
179 | smp_wmb(); |
180 | root->in_trans_setup = 0; | |
5d4f98a2 YZ |
181 | } |
182 | return 0; | |
183 | } | |
bcc63abb | 184 | |
7585717f CM |
185 | |
186 | int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, | |
187 | struct btrfs_root *root) | |
188 | { | |
189 | if (!root->ref_cows) | |
190 | return 0; | |
191 | ||
192 | /* | |
193 | * see record_root_in_trans for comments about in_trans_setup usage | |
194 | * and barriers | |
195 | */ | |
196 | smp_rmb(); | |
197 | if (root->last_trans == trans->transid && | |
198 | !root->in_trans_setup) | |
199 | return 0; | |
200 | ||
201 | mutex_lock(&root->fs_info->reloc_mutex); | |
202 | record_root_in_trans(trans, root); | |
203 | mutex_unlock(&root->fs_info->reloc_mutex); | |
204 | ||
205 | return 0; | |
206 | } | |
207 | ||
d352ac68 CM |
208 | /* wait for commit against the current transaction to become unblocked |
209 | * when this is done, it is safe to start a new transaction, but the current | |
210 | * transaction might not be fully on disk. | |
211 | */ | |
37d1aeee | 212 | static void wait_current_trans(struct btrfs_root *root) |
79154b1b | 213 | { |
f9295749 | 214 | struct btrfs_transaction *cur_trans; |
79154b1b | 215 | |
a4abeea4 | 216 | spin_lock(&root->fs_info->trans_lock); |
f9295749 | 217 | cur_trans = root->fs_info->running_transaction; |
37d1aeee | 218 | if (cur_trans && cur_trans->blocked) { |
f9295749 | 219 | DEFINE_WAIT(wait); |
13c5a93e | 220 | atomic_inc(&cur_trans->use_count); |
a4abeea4 | 221 | spin_unlock(&root->fs_info->trans_lock); |
d397712b | 222 | while (1) { |
f9295749 CM |
223 | prepare_to_wait(&root->fs_info->transaction_wait, &wait, |
224 | TASK_UNINTERRUPTIBLE); | |
471fa17d | 225 | if (!cur_trans->blocked) |
f9295749 | 226 | break; |
471fa17d | 227 | schedule(); |
f9295749 | 228 | } |
471fa17d | 229 | finish_wait(&root->fs_info->transaction_wait, &wait); |
f9295749 | 230 | put_transaction(cur_trans); |
a4abeea4 JB |
231 | } else { |
232 | spin_unlock(&root->fs_info->trans_lock); | |
f9295749 | 233 | } |
37d1aeee CM |
234 | } |
235 | ||
249ac1e5 JB |
236 | enum btrfs_trans_type { |
237 | TRANS_START, | |
238 | TRANS_JOIN, | |
239 | TRANS_USERSPACE, | |
0af3d00b | 240 | TRANS_JOIN_NOLOCK, |
249ac1e5 JB |
241 | }; |
242 | ||
a22285a6 YZ |
243 | static int may_wait_transaction(struct btrfs_root *root, int type) |
244 | { | |
a4abeea4 JB |
245 | if (root->fs_info->log_root_recovering) |
246 | return 0; | |
247 | ||
248 | if (type == TRANS_USERSPACE) | |
249 | return 1; | |
250 | ||
251 | if (type == TRANS_START && | |
252 | !atomic_read(&root->fs_info->open_ioctl_trans)) | |
a22285a6 | 253 | return 1; |
a4abeea4 | 254 | |
a22285a6 YZ |
255 | return 0; |
256 | } | |
257 | ||
e02119d5 | 258 | static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root, |
a22285a6 | 259 | u64 num_items, int type) |
37d1aeee | 260 | { |
a22285a6 YZ |
261 | struct btrfs_trans_handle *h; |
262 | struct btrfs_transaction *cur_trans; | |
b5009945 | 263 | u64 num_bytes = 0; |
37d1aeee | 264 | int ret; |
acce952b | 265 | |
266 | if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) | |
267 | return ERR_PTR(-EROFS); | |
2a1eb461 JB |
268 | |
269 | if (current->journal_info) { | |
270 | WARN_ON(type != TRANS_JOIN && type != TRANS_JOIN_NOLOCK); | |
271 | h = current->journal_info; | |
272 | h->use_count++; | |
273 | h->orig_rsv = h->block_rsv; | |
274 | h->block_rsv = NULL; | |
275 | goto got_it; | |
276 | } | |
b5009945 JB |
277 | |
278 | /* | |
279 | * Do the reservation before we join the transaction so we can do all | |
280 | * the appropriate flushing if need be. | |
281 | */ | |
282 | if (num_items > 0 && root != root->fs_info->chunk_root) { | |
283 | num_bytes = btrfs_calc_trans_metadata_size(root, num_items); | |
284 | ret = btrfs_block_rsv_add(NULL, root, | |
285 | &root->fs_info->trans_block_rsv, | |
286 | num_bytes); | |
287 | if (ret) | |
288 | return ERR_PTR(ret); | |
289 | } | |
a22285a6 YZ |
290 | again: |
291 | h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS); | |
292 | if (!h) | |
293 | return ERR_PTR(-ENOMEM); | |
37d1aeee | 294 | |
a22285a6 | 295 | if (may_wait_transaction(root, type)) |
37d1aeee | 296 | wait_current_trans(root); |
a22285a6 | 297 | |
a4abeea4 JB |
298 | do { |
299 | ret = join_transaction(root, type == TRANS_JOIN_NOLOCK); | |
300 | if (ret == -EBUSY) | |
301 | wait_current_trans(root); | |
302 | } while (ret == -EBUSY); | |
303 | ||
db5b493a | 304 | if (ret < 0) { |
6e8df2ae | 305 | kmem_cache_free(btrfs_trans_handle_cachep, h); |
db5b493a TI |
306 | return ERR_PTR(ret); |
307 | } | |
0f7d52f4 | 308 | |
a22285a6 | 309 | cur_trans = root->fs_info->running_transaction; |
a22285a6 YZ |
310 | |
311 | h->transid = cur_trans->transid; | |
312 | h->transaction = cur_trans; | |
79154b1b | 313 | h->blocks_used = 0; |
a22285a6 | 314 | h->bytes_reserved = 0; |
56bec294 | 315 | h->delayed_ref_updates = 0; |
2a1eb461 | 316 | h->use_count = 1; |
f0486c68 | 317 | h->block_rsv = NULL; |
2a1eb461 | 318 | h->orig_rsv = NULL; |
b7ec40d7 | 319 | |
a22285a6 YZ |
320 | smp_mb(); |
321 | if (cur_trans->blocked && may_wait_transaction(root, type)) { | |
322 | btrfs_commit_transaction(h, root); | |
323 | goto again; | |
324 | } | |
325 | ||
b5009945 JB |
326 | if (num_bytes) { |
327 | h->block_rsv = &root->fs_info->trans_block_rsv; | |
328 | h->bytes_reserved = num_bytes; | |
a22285a6 | 329 | } |
9ed74f2d | 330 | |
2a1eb461 | 331 | got_it: |
a4abeea4 | 332 | btrfs_record_root_in_trans(h, root); |
a22285a6 YZ |
333 | |
334 | if (!current->journal_info && type != TRANS_USERSPACE) | |
335 | current->journal_info = h; | |
79154b1b CM |
336 | return h; |
337 | } | |
338 | ||
f9295749 | 339 | struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, |
a22285a6 | 340 | int num_items) |
f9295749 | 341 | { |
a22285a6 | 342 | return start_transaction(root, num_items, TRANS_START); |
f9295749 | 343 | } |
7a7eaa40 | 344 | struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root) |
f9295749 | 345 | { |
a22285a6 | 346 | return start_transaction(root, 0, TRANS_JOIN); |
f9295749 CM |
347 | } |
348 | ||
7a7eaa40 | 349 | struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root) |
0af3d00b JB |
350 | { |
351 | return start_transaction(root, 0, TRANS_JOIN_NOLOCK); | |
352 | } | |
353 | ||
7a7eaa40 | 354 | struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root) |
9ca9ee09 | 355 | { |
7a7eaa40 | 356 | return start_transaction(root, 0, TRANS_USERSPACE); |
9ca9ee09 SW |
357 | } |
358 | ||
d352ac68 | 359 | /* wait for a transaction commit to be fully complete */ |
89ce8a63 CM |
360 | static noinline int wait_for_commit(struct btrfs_root *root, |
361 | struct btrfs_transaction *commit) | |
362 | { | |
363 | DEFINE_WAIT(wait); | |
d397712b | 364 | while (!commit->commit_done) { |
89ce8a63 CM |
365 | prepare_to_wait(&commit->commit_wait, &wait, |
366 | TASK_UNINTERRUPTIBLE); | |
367 | if (commit->commit_done) | |
368 | break; | |
89ce8a63 | 369 | schedule(); |
89ce8a63 | 370 | } |
89ce8a63 CM |
371 | finish_wait(&commit->commit_wait, &wait); |
372 | return 0; | |
373 | } | |
374 | ||
46204592 SW |
375 | int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid) |
376 | { | |
377 | struct btrfs_transaction *cur_trans = NULL, *t; | |
378 | int ret; | |
379 | ||
46204592 SW |
380 | ret = 0; |
381 | if (transid) { | |
382 | if (transid <= root->fs_info->last_trans_committed) | |
a4abeea4 | 383 | goto out; |
46204592 SW |
384 | |
385 | /* find specified transaction */ | |
a4abeea4 | 386 | spin_lock(&root->fs_info->trans_lock); |
46204592 SW |
387 | list_for_each_entry(t, &root->fs_info->trans_list, list) { |
388 | if (t->transid == transid) { | |
389 | cur_trans = t; | |
a4abeea4 | 390 | atomic_inc(&cur_trans->use_count); |
46204592 SW |
391 | break; |
392 | } | |
393 | if (t->transid > transid) | |
394 | break; | |
395 | } | |
a4abeea4 | 396 | spin_unlock(&root->fs_info->trans_lock); |
46204592 SW |
397 | ret = -EINVAL; |
398 | if (!cur_trans) | |
a4abeea4 | 399 | goto out; /* bad transid */ |
46204592 SW |
400 | } else { |
401 | /* find newest transaction that is committing | committed */ | |
a4abeea4 | 402 | spin_lock(&root->fs_info->trans_lock); |
46204592 SW |
403 | list_for_each_entry_reverse(t, &root->fs_info->trans_list, |
404 | list) { | |
405 | if (t->in_commit) { | |
406 | if (t->commit_done) | |
3473f3c0 | 407 | break; |
46204592 | 408 | cur_trans = t; |
a4abeea4 | 409 | atomic_inc(&cur_trans->use_count); |
46204592 SW |
410 | break; |
411 | } | |
412 | } | |
a4abeea4 | 413 | spin_unlock(&root->fs_info->trans_lock); |
46204592 | 414 | if (!cur_trans) |
a4abeea4 | 415 | goto out; /* nothing committing|committed */ |
46204592 SW |
416 | } |
417 | ||
46204592 SW |
418 | wait_for_commit(root, cur_trans); |
419 | ||
46204592 SW |
420 | put_transaction(cur_trans); |
421 | ret = 0; | |
a4abeea4 | 422 | out: |
46204592 SW |
423 | return ret; |
424 | } | |
425 | ||
37d1aeee CM |
426 | void btrfs_throttle(struct btrfs_root *root) |
427 | { | |
a4abeea4 | 428 | if (!atomic_read(&root->fs_info->open_ioctl_trans)) |
9ca9ee09 | 429 | wait_current_trans(root); |
37d1aeee CM |
430 | } |
431 | ||
8929ecfa YZ |
432 | static int should_end_transaction(struct btrfs_trans_handle *trans, |
433 | struct btrfs_root *root) | |
434 | { | |
435 | int ret; | |
436 | ret = btrfs_block_rsv_check(trans, root, | |
437 | &root->fs_info->global_block_rsv, 0, 5); | |
438 | return ret ? 1 : 0; | |
439 | } | |
440 | ||
441 | int btrfs_should_end_transaction(struct btrfs_trans_handle *trans, | |
442 | struct btrfs_root *root) | |
443 | { | |
444 | struct btrfs_transaction *cur_trans = trans->transaction; | |
445 | int updates; | |
446 | ||
a4abeea4 | 447 | smp_mb(); |
8929ecfa YZ |
448 | if (cur_trans->blocked || cur_trans->delayed_refs.flushing) |
449 | return 1; | |
450 | ||
451 | updates = trans->delayed_ref_updates; | |
452 | trans->delayed_ref_updates = 0; | |
453 | if (updates) | |
454 | btrfs_run_delayed_refs(trans, root, updates); | |
455 | ||
456 | return should_end_transaction(trans, root); | |
457 | } | |
458 | ||
89ce8a63 | 459 | static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, |
0af3d00b | 460 | struct btrfs_root *root, int throttle, int lock) |
79154b1b | 461 | { |
8929ecfa | 462 | struct btrfs_transaction *cur_trans = trans->transaction; |
ab78c84d | 463 | struct btrfs_fs_info *info = root->fs_info; |
c3e69d58 CM |
464 | int count = 0; |
465 | ||
2a1eb461 JB |
466 | if (--trans->use_count) { |
467 | trans->block_rsv = trans->orig_rsv; | |
468 | return 0; | |
469 | } | |
470 | ||
c3e69d58 CM |
471 | while (count < 4) { |
472 | unsigned long cur = trans->delayed_ref_updates; | |
473 | trans->delayed_ref_updates = 0; | |
474 | if (cur && | |
475 | trans->transaction->delayed_refs.num_heads_ready > 64) { | |
476 | trans->delayed_ref_updates = 0; | |
b7ec40d7 CM |
477 | |
478 | /* | |
479 | * do a full flush if the transaction is trying | |
480 | * to close | |
481 | */ | |
482 | if (trans->transaction->delayed_refs.flushing) | |
483 | cur = 0; | |
c3e69d58 CM |
484 | btrfs_run_delayed_refs(trans, root, cur); |
485 | } else { | |
486 | break; | |
487 | } | |
488 | count++; | |
56bec294 CM |
489 | } |
490 | ||
a22285a6 YZ |
491 | btrfs_trans_release_metadata(trans, root); |
492 | ||
a4abeea4 JB |
493 | if (lock && !atomic_read(&root->fs_info->open_ioctl_trans) && |
494 | should_end_transaction(trans, root)) { | |
8929ecfa | 495 | trans->transaction->blocked = 1; |
a4abeea4 JB |
496 | smp_wmb(); |
497 | } | |
8929ecfa | 498 | |
0af3d00b | 499 | if (lock && cur_trans->blocked && !cur_trans->in_commit) { |
81317fde JB |
500 | if (throttle) { |
501 | /* | |
502 | * We may race with somebody else here so end up having | |
503 | * to call end_transaction on ourselves again, so inc | |
504 | * our use_count. | |
505 | */ | |
506 | trans->use_count++; | |
8929ecfa | 507 | return btrfs_commit_transaction(trans, root); |
81317fde | 508 | } else { |
8929ecfa | 509 | wake_up_process(info->transaction_kthread); |
81317fde | 510 | } |
8929ecfa YZ |
511 | } |
512 | ||
8929ecfa | 513 | WARN_ON(cur_trans != info->running_transaction); |
13c5a93e JB |
514 | WARN_ON(atomic_read(&cur_trans->num_writers) < 1); |
515 | atomic_dec(&cur_trans->num_writers); | |
89ce8a63 | 516 | |
99d16cbc | 517 | smp_mb(); |
79154b1b CM |
518 | if (waitqueue_active(&cur_trans->writer_wait)) |
519 | wake_up(&cur_trans->writer_wait); | |
79154b1b | 520 | put_transaction(cur_trans); |
9ed74f2d JB |
521 | |
522 | if (current->journal_info == trans) | |
523 | current->journal_info = NULL; | |
d6025579 | 524 | memset(trans, 0, sizeof(*trans)); |
2c90e5d6 | 525 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
ab78c84d | 526 | |
24bbcf04 YZ |
527 | if (throttle) |
528 | btrfs_run_delayed_iputs(root); | |
529 | ||
79154b1b CM |
530 | return 0; |
531 | } | |
532 | ||
89ce8a63 CM |
533 | int btrfs_end_transaction(struct btrfs_trans_handle *trans, |
534 | struct btrfs_root *root) | |
535 | { | |
16cdcec7 MX |
536 | int ret; |
537 | ||
538 | ret = __btrfs_end_transaction(trans, root, 0, 1); | |
539 | if (ret) | |
540 | return ret; | |
541 | return 0; | |
89ce8a63 CM |
542 | } |
543 | ||
544 | int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans, | |
545 | struct btrfs_root *root) | |
546 | { | |
16cdcec7 MX |
547 | int ret; |
548 | ||
549 | ret = __btrfs_end_transaction(trans, root, 1, 1); | |
550 | if (ret) | |
551 | return ret; | |
552 | return 0; | |
0af3d00b JB |
553 | } |
554 | ||
555 | int btrfs_end_transaction_nolock(struct btrfs_trans_handle *trans, | |
556 | struct btrfs_root *root) | |
557 | { | |
16cdcec7 MX |
558 | int ret; |
559 | ||
560 | ret = __btrfs_end_transaction(trans, root, 0, 0); | |
561 | if (ret) | |
562 | return ret; | |
563 | return 0; | |
564 | } | |
565 | ||
566 | int btrfs_end_transaction_dmeta(struct btrfs_trans_handle *trans, | |
567 | struct btrfs_root *root) | |
568 | { | |
569 | return __btrfs_end_transaction(trans, root, 1, 1); | |
89ce8a63 CM |
570 | } |
571 | ||
d352ac68 CM |
572 | /* |
573 | * when btree blocks are allocated, they have some corresponding bits set for | |
574 | * them in one of two extent_io trees. This is used to make sure all of | |
690587d1 | 575 | * those extents are sent to disk but does not wait on them |
d352ac68 | 576 | */ |
690587d1 | 577 | int btrfs_write_marked_extents(struct btrfs_root *root, |
8cef4e16 | 578 | struct extent_io_tree *dirty_pages, int mark) |
79154b1b | 579 | { |
7c4452b9 | 580 | int ret; |
777e6bd7 | 581 | int err = 0; |
7c4452b9 CM |
582 | int werr = 0; |
583 | struct page *page; | |
7c4452b9 | 584 | struct inode *btree_inode = root->fs_info->btree_inode; |
777e6bd7 | 585 | u64 start = 0; |
5f39d397 CM |
586 | u64 end; |
587 | unsigned long index; | |
7c4452b9 | 588 | |
d397712b | 589 | while (1) { |
777e6bd7 | 590 | ret = find_first_extent_bit(dirty_pages, start, &start, &end, |
8cef4e16 | 591 | mark); |
5f39d397 | 592 | if (ret) |
7c4452b9 | 593 | break; |
d397712b | 594 | while (start <= end) { |
777e6bd7 CM |
595 | cond_resched(); |
596 | ||
5f39d397 | 597 | index = start >> PAGE_CACHE_SHIFT; |
35ebb934 | 598 | start = (u64)(index + 1) << PAGE_CACHE_SHIFT; |
4bef0848 | 599 | page = find_get_page(btree_inode->i_mapping, index); |
7c4452b9 CM |
600 | if (!page) |
601 | continue; | |
4bef0848 CM |
602 | |
603 | btree_lock_page_hook(page); | |
604 | if (!page->mapping) { | |
605 | unlock_page(page); | |
606 | page_cache_release(page); | |
607 | continue; | |
608 | } | |
609 | ||
6702ed49 CM |
610 | if (PageWriteback(page)) { |
611 | if (PageDirty(page)) | |
612 | wait_on_page_writeback(page); | |
613 | else { | |
614 | unlock_page(page); | |
615 | page_cache_release(page); | |
616 | continue; | |
617 | } | |
618 | } | |
7c4452b9 CM |
619 | err = write_one_page(page, 0); |
620 | if (err) | |
621 | werr = err; | |
622 | page_cache_release(page); | |
623 | } | |
624 | } | |
690587d1 CM |
625 | if (err) |
626 | werr = err; | |
627 | return werr; | |
628 | } | |
629 | ||
630 | /* | |
631 | * when btree blocks are allocated, they have some corresponding bits set for | |
632 | * them in one of two extent_io trees. This is used to make sure all of | |
633 | * those extents are on disk for transaction or log commit. We wait | |
634 | * on all the pages and clear them from the dirty pages state tree | |
635 | */ | |
636 | int btrfs_wait_marked_extents(struct btrfs_root *root, | |
8cef4e16 | 637 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 CM |
638 | { |
639 | int ret; | |
640 | int err = 0; | |
641 | int werr = 0; | |
642 | struct page *page; | |
643 | struct inode *btree_inode = root->fs_info->btree_inode; | |
644 | u64 start = 0; | |
645 | u64 end; | |
646 | unsigned long index; | |
647 | ||
d397712b | 648 | while (1) { |
8cef4e16 YZ |
649 | ret = find_first_extent_bit(dirty_pages, start, &start, &end, |
650 | mark); | |
777e6bd7 CM |
651 | if (ret) |
652 | break; | |
653 | ||
8cef4e16 | 654 | clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS); |
d397712b | 655 | while (start <= end) { |
777e6bd7 CM |
656 | index = start >> PAGE_CACHE_SHIFT; |
657 | start = (u64)(index + 1) << PAGE_CACHE_SHIFT; | |
658 | page = find_get_page(btree_inode->i_mapping, index); | |
659 | if (!page) | |
660 | continue; | |
661 | if (PageDirty(page)) { | |
4bef0848 CM |
662 | btree_lock_page_hook(page); |
663 | wait_on_page_writeback(page); | |
777e6bd7 CM |
664 | err = write_one_page(page, 0); |
665 | if (err) | |
666 | werr = err; | |
667 | } | |
105d931d | 668 | wait_on_page_writeback(page); |
777e6bd7 CM |
669 | page_cache_release(page); |
670 | cond_resched(); | |
671 | } | |
672 | } | |
7c4452b9 CM |
673 | if (err) |
674 | werr = err; | |
675 | return werr; | |
79154b1b CM |
676 | } |
677 | ||
690587d1 CM |
678 | /* |
679 | * when btree blocks are allocated, they have some corresponding bits set for | |
680 | * them in one of two extent_io trees. This is used to make sure all of | |
681 | * those extents are on disk for transaction or log commit | |
682 | */ | |
683 | int btrfs_write_and_wait_marked_extents(struct btrfs_root *root, | |
8cef4e16 | 684 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 CM |
685 | { |
686 | int ret; | |
687 | int ret2; | |
688 | ||
8cef4e16 YZ |
689 | ret = btrfs_write_marked_extents(root, dirty_pages, mark); |
690 | ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark); | |
690587d1 CM |
691 | return ret || ret2; |
692 | } | |
693 | ||
d0c803c4 CM |
694 | int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans, |
695 | struct btrfs_root *root) | |
696 | { | |
697 | if (!trans || !trans->transaction) { | |
698 | struct inode *btree_inode; | |
699 | btree_inode = root->fs_info->btree_inode; | |
700 | return filemap_write_and_wait(btree_inode->i_mapping); | |
701 | } | |
702 | return btrfs_write_and_wait_marked_extents(root, | |
8cef4e16 YZ |
703 | &trans->transaction->dirty_pages, |
704 | EXTENT_DIRTY); | |
d0c803c4 CM |
705 | } |
706 | ||
d352ac68 CM |
707 | /* |
708 | * this is used to update the root pointer in the tree of tree roots. | |
709 | * | |
710 | * But, in the case of the extent allocation tree, updating the root | |
711 | * pointer may allocate blocks which may change the root of the extent | |
712 | * allocation tree. | |
713 | * | |
714 | * So, this loops and repeats and makes sure the cowonly root didn't | |
715 | * change while the root pointer was being updated in the metadata. | |
716 | */ | |
0b86a832 CM |
717 | static int update_cowonly_root(struct btrfs_trans_handle *trans, |
718 | struct btrfs_root *root) | |
79154b1b CM |
719 | { |
720 | int ret; | |
0b86a832 | 721 | u64 old_root_bytenr; |
86b9f2ec | 722 | u64 old_root_used; |
0b86a832 | 723 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
79154b1b | 724 | |
86b9f2ec | 725 | old_root_used = btrfs_root_used(&root->root_item); |
0b86a832 | 726 | btrfs_write_dirty_block_groups(trans, root); |
56bec294 | 727 | |
d397712b | 728 | while (1) { |
0b86a832 | 729 | old_root_bytenr = btrfs_root_bytenr(&root->root_item); |
86b9f2ec YZ |
730 | if (old_root_bytenr == root->node->start && |
731 | old_root_used == btrfs_root_used(&root->root_item)) | |
79154b1b | 732 | break; |
87ef2bb4 | 733 | |
5d4f98a2 | 734 | btrfs_set_root_node(&root->root_item, root->node); |
79154b1b | 735 | ret = btrfs_update_root(trans, tree_root, |
0b86a832 CM |
736 | &root->root_key, |
737 | &root->root_item); | |
79154b1b | 738 | BUG_ON(ret); |
56bec294 | 739 | |
86b9f2ec | 740 | old_root_used = btrfs_root_used(&root->root_item); |
4a8c9a62 | 741 | ret = btrfs_write_dirty_block_groups(trans, root); |
56bec294 | 742 | BUG_ON(ret); |
0b86a832 | 743 | } |
276e680d YZ |
744 | |
745 | if (root != root->fs_info->extent_root) | |
746 | switch_commit_root(root); | |
747 | ||
0b86a832 CM |
748 | return 0; |
749 | } | |
750 | ||
d352ac68 CM |
751 | /* |
752 | * update all the cowonly tree roots on disk | |
753 | */ | |
5d4f98a2 YZ |
754 | static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans, |
755 | struct btrfs_root *root) | |
0b86a832 CM |
756 | { |
757 | struct btrfs_fs_info *fs_info = root->fs_info; | |
758 | struct list_head *next; | |
84234f3a | 759 | struct extent_buffer *eb; |
56bec294 | 760 | int ret; |
84234f3a | 761 | |
56bec294 CM |
762 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
763 | BUG_ON(ret); | |
87ef2bb4 | 764 | |
84234f3a | 765 | eb = btrfs_lock_root_node(fs_info->tree_root); |
9fa8cfe7 | 766 | btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 0, &eb); |
84234f3a YZ |
767 | btrfs_tree_unlock(eb); |
768 | free_extent_buffer(eb); | |
0b86a832 | 769 | |
56bec294 CM |
770 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
771 | BUG_ON(ret); | |
87ef2bb4 | 772 | |
d397712b | 773 | while (!list_empty(&fs_info->dirty_cowonly_roots)) { |
0b86a832 CM |
774 | next = fs_info->dirty_cowonly_roots.next; |
775 | list_del_init(next); | |
776 | root = list_entry(next, struct btrfs_root, dirty_list); | |
87ef2bb4 | 777 | |
0b86a832 | 778 | update_cowonly_root(trans, root); |
79154b1b | 779 | } |
276e680d YZ |
780 | |
781 | down_write(&fs_info->extent_commit_sem); | |
782 | switch_commit_root(fs_info->extent_root); | |
783 | up_write(&fs_info->extent_commit_sem); | |
784 | ||
79154b1b CM |
785 | return 0; |
786 | } | |
787 | ||
d352ac68 CM |
788 | /* |
789 | * dead roots are old snapshots that need to be deleted. This allocates | |
790 | * a dirty root struct and adds it into the list of dead roots that need to | |
791 | * be deleted | |
792 | */ | |
5d4f98a2 | 793 | int btrfs_add_dead_root(struct btrfs_root *root) |
5eda7b5e | 794 | { |
a4abeea4 | 795 | spin_lock(&root->fs_info->trans_lock); |
5d4f98a2 | 796 | list_add(&root->root_list, &root->fs_info->dead_roots); |
a4abeea4 | 797 | spin_unlock(&root->fs_info->trans_lock); |
5eda7b5e CM |
798 | return 0; |
799 | } | |
800 | ||
d352ac68 | 801 | /* |
5d4f98a2 | 802 | * update all the cowonly tree roots on disk |
d352ac68 | 803 | */ |
5d4f98a2 YZ |
804 | static noinline int commit_fs_roots(struct btrfs_trans_handle *trans, |
805 | struct btrfs_root *root) | |
0f7d52f4 | 806 | { |
0f7d52f4 | 807 | struct btrfs_root *gang[8]; |
5d4f98a2 | 808 | struct btrfs_fs_info *fs_info = root->fs_info; |
0f7d52f4 CM |
809 | int i; |
810 | int ret; | |
54aa1f4d CM |
811 | int err = 0; |
812 | ||
a4abeea4 | 813 | spin_lock(&fs_info->fs_roots_radix_lock); |
d397712b | 814 | while (1) { |
5d4f98a2 YZ |
815 | ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, |
816 | (void **)gang, 0, | |
0f7d52f4 CM |
817 | ARRAY_SIZE(gang), |
818 | BTRFS_ROOT_TRANS_TAG); | |
819 | if (ret == 0) | |
820 | break; | |
821 | for (i = 0; i < ret; i++) { | |
822 | root = gang[i]; | |
5d4f98a2 YZ |
823 | radix_tree_tag_clear(&fs_info->fs_roots_radix, |
824 | (unsigned long)root->root_key.objectid, | |
825 | BTRFS_ROOT_TRANS_TAG); | |
a4abeea4 | 826 | spin_unlock(&fs_info->fs_roots_radix_lock); |
31153d81 | 827 | |
e02119d5 | 828 | btrfs_free_log(trans, root); |
5d4f98a2 | 829 | btrfs_update_reloc_root(trans, root); |
d68fc57b | 830 | btrfs_orphan_commit_root(trans, root); |
bcc63abb | 831 | |
82d5902d LZ |
832 | btrfs_save_ino_cache(root, trans); |
833 | ||
978d910d | 834 | if (root->commit_root != root->node) { |
581bb050 | 835 | mutex_lock(&root->fs_commit_mutex); |
817d52f8 | 836 | switch_commit_root(root); |
581bb050 LZ |
837 | btrfs_unpin_free_ino(root); |
838 | mutex_unlock(&root->fs_commit_mutex); | |
839 | ||
978d910d YZ |
840 | btrfs_set_root_node(&root->root_item, |
841 | root->node); | |
842 | } | |
5d4f98a2 | 843 | |
5d4f98a2 | 844 | err = btrfs_update_root(trans, fs_info->tree_root, |
0f7d52f4 CM |
845 | &root->root_key, |
846 | &root->root_item); | |
a4abeea4 | 847 | spin_lock(&fs_info->fs_roots_radix_lock); |
54aa1f4d CM |
848 | if (err) |
849 | break; | |
0f7d52f4 CM |
850 | } |
851 | } | |
a4abeea4 | 852 | spin_unlock(&fs_info->fs_roots_radix_lock); |
54aa1f4d | 853 | return err; |
0f7d52f4 CM |
854 | } |
855 | ||
d352ac68 CM |
856 | /* |
857 | * defrag a given btree. If cacheonly == 1, this won't read from the disk, | |
858 | * otherwise every leaf in the btree is read and defragged. | |
859 | */ | |
e9d0b13b CM |
860 | int btrfs_defrag_root(struct btrfs_root *root, int cacheonly) |
861 | { | |
862 | struct btrfs_fs_info *info = root->fs_info; | |
e9d0b13b | 863 | struct btrfs_trans_handle *trans; |
8929ecfa | 864 | int ret; |
d3c2fdcf | 865 | unsigned long nr; |
e9d0b13b | 866 | |
8929ecfa | 867 | if (xchg(&root->defrag_running, 1)) |
e9d0b13b | 868 | return 0; |
8929ecfa | 869 | |
6b80053d | 870 | while (1) { |
8929ecfa YZ |
871 | trans = btrfs_start_transaction(root, 0); |
872 | if (IS_ERR(trans)) | |
873 | return PTR_ERR(trans); | |
874 | ||
e9d0b13b | 875 | ret = btrfs_defrag_leaves(trans, root, cacheonly); |
8929ecfa | 876 | |
d3c2fdcf | 877 | nr = trans->blocks_used; |
e9d0b13b | 878 | btrfs_end_transaction(trans, root); |
d3c2fdcf | 879 | btrfs_btree_balance_dirty(info->tree_root, nr); |
e9d0b13b CM |
880 | cond_resched(); |
881 | ||
7841cb28 | 882 | if (btrfs_fs_closing(root->fs_info) || ret != -EAGAIN) |
e9d0b13b CM |
883 | break; |
884 | } | |
885 | root->defrag_running = 0; | |
8929ecfa | 886 | return ret; |
e9d0b13b CM |
887 | } |
888 | ||
d352ac68 CM |
889 | /* |
890 | * new snapshots need to be created at a very specific time in the | |
891 | * transaction commit. This does the actual creation | |
892 | */ | |
80b6794d | 893 | static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, |
3063d29f CM |
894 | struct btrfs_fs_info *fs_info, |
895 | struct btrfs_pending_snapshot *pending) | |
896 | { | |
897 | struct btrfs_key key; | |
80b6794d | 898 | struct btrfs_root_item *new_root_item; |
3063d29f CM |
899 | struct btrfs_root *tree_root = fs_info->tree_root; |
900 | struct btrfs_root *root = pending->root; | |
6bdb72de SW |
901 | struct btrfs_root *parent_root; |
902 | struct inode *parent_inode; | |
6a912213 | 903 | struct dentry *parent; |
a22285a6 | 904 | struct dentry *dentry; |
3063d29f | 905 | struct extent_buffer *tmp; |
925baedd | 906 | struct extent_buffer *old; |
3063d29f | 907 | int ret; |
d68fc57b | 908 | u64 to_reserve = 0; |
6bdb72de | 909 | u64 index = 0; |
a22285a6 | 910 | u64 objectid; |
b83cc969 | 911 | u64 root_flags; |
3063d29f | 912 | |
80b6794d CM |
913 | new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS); |
914 | if (!new_root_item) { | |
a22285a6 | 915 | pending->error = -ENOMEM; |
80b6794d CM |
916 | goto fail; |
917 | } | |
a22285a6 | 918 | |
581bb050 | 919 | ret = btrfs_find_free_objectid(tree_root, &objectid); |
a22285a6 YZ |
920 | if (ret) { |
921 | pending->error = ret; | |
3063d29f | 922 | goto fail; |
a22285a6 | 923 | } |
3063d29f | 924 | |
3fd0a558 | 925 | btrfs_reloc_pre_snapshot(trans, pending, &to_reserve); |
d68fc57b YZ |
926 | btrfs_orphan_pre_snapshot(trans, pending, &to_reserve); |
927 | ||
928 | if (to_reserve > 0) { | |
929 | ret = btrfs_block_rsv_add(trans, root, &pending->block_rsv, | |
8bb8ab2e | 930 | to_reserve); |
d68fc57b YZ |
931 | if (ret) { |
932 | pending->error = ret; | |
933 | goto fail; | |
934 | } | |
935 | } | |
936 | ||
3063d29f | 937 | key.objectid = objectid; |
a22285a6 YZ |
938 | key.offset = (u64)-1; |
939 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3063d29f | 940 | |
a22285a6 | 941 | trans->block_rsv = &pending->block_rsv; |
3de4586c | 942 | |
a22285a6 | 943 | dentry = pending->dentry; |
6a912213 JB |
944 | parent = dget_parent(dentry); |
945 | parent_inode = parent->d_inode; | |
a22285a6 | 946 | parent_root = BTRFS_I(parent_inode)->root; |
7585717f | 947 | record_root_in_trans(trans, parent_root); |
a22285a6 | 948 | |
3063d29f CM |
949 | /* |
950 | * insert the directory item | |
951 | */ | |
3de4586c | 952 | ret = btrfs_set_inode_index(parent_inode, &index); |
6bdb72de | 953 | BUG_ON(ret); |
0660b5af | 954 | ret = btrfs_insert_dir_item(trans, parent_root, |
a22285a6 | 955 | dentry->d_name.name, dentry->d_name.len, |
16cdcec7 | 956 | parent_inode, &key, |
a22285a6 | 957 | BTRFS_FT_DIR, index); |
6bdb72de | 958 | BUG_ON(ret); |
0660b5af | 959 | |
a22285a6 YZ |
960 | btrfs_i_size_write(parent_inode, parent_inode->i_size + |
961 | dentry->d_name.len * 2); | |
52c26179 YZ |
962 | ret = btrfs_update_inode(trans, parent_root, parent_inode); |
963 | BUG_ON(ret); | |
964 | ||
e999376f CM |
965 | /* |
966 | * pull in the delayed directory update | |
967 | * and the delayed inode item | |
968 | * otherwise we corrupt the FS during | |
969 | * snapshot | |
970 | */ | |
971 | ret = btrfs_run_delayed_items(trans, root); | |
972 | BUG_ON(ret); | |
973 | ||
7585717f | 974 | record_root_in_trans(trans, root); |
6bdb72de SW |
975 | btrfs_set_root_last_snapshot(&root->root_item, trans->transid); |
976 | memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); | |
08fe4db1 | 977 | btrfs_check_and_init_root_item(new_root_item); |
6bdb72de | 978 | |
b83cc969 LZ |
979 | root_flags = btrfs_root_flags(new_root_item); |
980 | if (pending->readonly) | |
981 | root_flags |= BTRFS_ROOT_SUBVOL_RDONLY; | |
982 | else | |
983 | root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY; | |
984 | btrfs_set_root_flags(new_root_item, root_flags); | |
985 | ||
6bdb72de SW |
986 | old = btrfs_lock_root_node(root); |
987 | btrfs_cow_block(trans, root, old, NULL, 0, &old); | |
988 | btrfs_set_lock_blocking(old); | |
989 | ||
990 | btrfs_copy_root(trans, root, old, &tmp, objectid); | |
991 | btrfs_tree_unlock(old); | |
992 | free_extent_buffer(old); | |
993 | ||
994 | btrfs_set_root_node(new_root_item, tmp); | |
a22285a6 YZ |
995 | /* record when the snapshot was created in key.offset */ |
996 | key.offset = trans->transid; | |
997 | ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); | |
6bdb72de SW |
998 | btrfs_tree_unlock(tmp); |
999 | free_extent_buffer(tmp); | |
a22285a6 | 1000 | BUG_ON(ret); |
6bdb72de | 1001 | |
a22285a6 YZ |
1002 | /* |
1003 | * insert root back/forward references | |
1004 | */ | |
1005 | ret = btrfs_add_root_ref(trans, tree_root, objectid, | |
0660b5af | 1006 | parent_root->root_key.objectid, |
33345d01 | 1007 | btrfs_ino(parent_inode), index, |
a22285a6 | 1008 | dentry->d_name.name, dentry->d_name.len); |
0660b5af | 1009 | BUG_ON(ret); |
6a912213 | 1010 | dput(parent); |
0660b5af | 1011 | |
a22285a6 YZ |
1012 | key.offset = (u64)-1; |
1013 | pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key); | |
1014 | BUG_ON(IS_ERR(pending->snap)); | |
d68fc57b | 1015 | |
3fd0a558 | 1016 | btrfs_reloc_post_snapshot(trans, pending); |
d68fc57b | 1017 | btrfs_orphan_post_snapshot(trans, pending); |
3063d29f | 1018 | fail: |
6bdb72de | 1019 | kfree(new_root_item); |
a22285a6 YZ |
1020 | btrfs_block_rsv_release(root, &pending->block_rsv, (u64)-1); |
1021 | return 0; | |
3063d29f CM |
1022 | } |
1023 | ||
d352ac68 CM |
1024 | /* |
1025 | * create all the snapshots we've scheduled for creation | |
1026 | */ | |
80b6794d CM |
1027 | static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans, |
1028 | struct btrfs_fs_info *fs_info) | |
3de4586c CM |
1029 | { |
1030 | struct btrfs_pending_snapshot *pending; | |
1031 | struct list_head *head = &trans->transaction->pending_snapshots; | |
3de4586c CM |
1032 | int ret; |
1033 | ||
c6e30871 | 1034 | list_for_each_entry(pending, head, list) { |
3de4586c CM |
1035 | ret = create_pending_snapshot(trans, fs_info, pending); |
1036 | BUG_ON(ret); | |
1037 | } | |
1038 | return 0; | |
1039 | } | |
1040 | ||
5d4f98a2 YZ |
1041 | static void update_super_roots(struct btrfs_root *root) |
1042 | { | |
1043 | struct btrfs_root_item *root_item; | |
1044 | struct btrfs_super_block *super; | |
1045 | ||
1046 | super = &root->fs_info->super_copy; | |
1047 | ||
1048 | root_item = &root->fs_info->chunk_root->root_item; | |
1049 | super->chunk_root = root_item->bytenr; | |
1050 | super->chunk_root_generation = root_item->generation; | |
1051 | super->chunk_root_level = root_item->level; | |
1052 | ||
1053 | root_item = &root->fs_info->tree_root->root_item; | |
1054 | super->root = root_item->bytenr; | |
1055 | super->generation = root_item->generation; | |
1056 | super->root_level = root_item->level; | |
0af3d00b JB |
1057 | if (super->cache_generation != 0 || btrfs_test_opt(root, SPACE_CACHE)) |
1058 | super->cache_generation = root_item->generation; | |
5d4f98a2 YZ |
1059 | } |
1060 | ||
f36f3042 CM |
1061 | int btrfs_transaction_in_commit(struct btrfs_fs_info *info) |
1062 | { | |
1063 | int ret = 0; | |
a4abeea4 | 1064 | spin_lock(&info->trans_lock); |
f36f3042 CM |
1065 | if (info->running_transaction) |
1066 | ret = info->running_transaction->in_commit; | |
a4abeea4 | 1067 | spin_unlock(&info->trans_lock); |
f36f3042 CM |
1068 | return ret; |
1069 | } | |
1070 | ||
8929ecfa YZ |
1071 | int btrfs_transaction_blocked(struct btrfs_fs_info *info) |
1072 | { | |
1073 | int ret = 0; | |
a4abeea4 | 1074 | spin_lock(&info->trans_lock); |
8929ecfa YZ |
1075 | if (info->running_transaction) |
1076 | ret = info->running_transaction->blocked; | |
a4abeea4 | 1077 | spin_unlock(&info->trans_lock); |
8929ecfa YZ |
1078 | return ret; |
1079 | } | |
1080 | ||
bb9c12c9 SW |
1081 | /* |
1082 | * wait for the current transaction commit to start and block subsequent | |
1083 | * transaction joins | |
1084 | */ | |
1085 | static void wait_current_trans_commit_start(struct btrfs_root *root, | |
1086 | struct btrfs_transaction *trans) | |
1087 | { | |
1088 | DEFINE_WAIT(wait); | |
1089 | ||
1090 | if (trans->in_commit) | |
1091 | return; | |
1092 | ||
1093 | while (1) { | |
1094 | prepare_to_wait(&root->fs_info->transaction_blocked_wait, &wait, | |
1095 | TASK_UNINTERRUPTIBLE); | |
1096 | if (trans->in_commit) { | |
1097 | finish_wait(&root->fs_info->transaction_blocked_wait, | |
1098 | &wait); | |
1099 | break; | |
1100 | } | |
bb9c12c9 | 1101 | schedule(); |
bb9c12c9 SW |
1102 | finish_wait(&root->fs_info->transaction_blocked_wait, &wait); |
1103 | } | |
1104 | } | |
1105 | ||
1106 | /* | |
1107 | * wait for the current transaction to start and then become unblocked. | |
1108 | * caller holds ref. | |
1109 | */ | |
1110 | static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root, | |
1111 | struct btrfs_transaction *trans) | |
1112 | { | |
1113 | DEFINE_WAIT(wait); | |
1114 | ||
1115 | if (trans->commit_done || (trans->in_commit && !trans->blocked)) | |
1116 | return; | |
1117 | ||
1118 | while (1) { | |
1119 | prepare_to_wait(&root->fs_info->transaction_wait, &wait, | |
1120 | TASK_UNINTERRUPTIBLE); | |
1121 | if (trans->commit_done || | |
1122 | (trans->in_commit && !trans->blocked)) { | |
1123 | finish_wait(&root->fs_info->transaction_wait, | |
1124 | &wait); | |
1125 | break; | |
1126 | } | |
bb9c12c9 | 1127 | schedule(); |
bb9c12c9 SW |
1128 | finish_wait(&root->fs_info->transaction_wait, |
1129 | &wait); | |
1130 | } | |
1131 | } | |
1132 | ||
1133 | /* | |
1134 | * commit transactions asynchronously. once btrfs_commit_transaction_async | |
1135 | * returns, any subsequent transaction will not be allowed to join. | |
1136 | */ | |
1137 | struct btrfs_async_commit { | |
1138 | struct btrfs_trans_handle *newtrans; | |
1139 | struct btrfs_root *root; | |
1140 | struct delayed_work work; | |
1141 | }; | |
1142 | ||
1143 | static void do_async_commit(struct work_struct *work) | |
1144 | { | |
1145 | struct btrfs_async_commit *ac = | |
1146 | container_of(work, struct btrfs_async_commit, work.work); | |
1147 | ||
1148 | btrfs_commit_transaction(ac->newtrans, ac->root); | |
1149 | kfree(ac); | |
1150 | } | |
1151 | ||
1152 | int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans, | |
1153 | struct btrfs_root *root, | |
1154 | int wait_for_unblock) | |
1155 | { | |
1156 | struct btrfs_async_commit *ac; | |
1157 | struct btrfs_transaction *cur_trans; | |
1158 | ||
1159 | ac = kmalloc(sizeof(*ac), GFP_NOFS); | |
db5b493a TI |
1160 | if (!ac) |
1161 | return -ENOMEM; | |
bb9c12c9 SW |
1162 | |
1163 | INIT_DELAYED_WORK(&ac->work, do_async_commit); | |
1164 | ac->root = root; | |
7a7eaa40 | 1165 | ac->newtrans = btrfs_join_transaction(root); |
3612b495 TI |
1166 | if (IS_ERR(ac->newtrans)) { |
1167 | int err = PTR_ERR(ac->newtrans); | |
1168 | kfree(ac); | |
1169 | return err; | |
1170 | } | |
bb9c12c9 SW |
1171 | |
1172 | /* take transaction reference */ | |
bb9c12c9 | 1173 | cur_trans = trans->transaction; |
13c5a93e | 1174 | atomic_inc(&cur_trans->use_count); |
bb9c12c9 SW |
1175 | |
1176 | btrfs_end_transaction(trans, root); | |
1177 | schedule_delayed_work(&ac->work, 0); | |
1178 | ||
1179 | /* wait for transaction to start and unblock */ | |
bb9c12c9 SW |
1180 | if (wait_for_unblock) |
1181 | wait_current_trans_commit_start_and_unblock(root, cur_trans); | |
1182 | else | |
1183 | wait_current_trans_commit_start(root, cur_trans); | |
bb9c12c9 | 1184 | |
38e88054 SW |
1185 | if (current->journal_info == trans) |
1186 | current->journal_info = NULL; | |
1187 | ||
1188 | put_transaction(cur_trans); | |
bb9c12c9 SW |
1189 | return 0; |
1190 | } | |
1191 | ||
1192 | /* | |
1193 | * btrfs_transaction state sequence: | |
1194 | * in_commit = 0, blocked = 0 (initial) | |
1195 | * in_commit = 1, blocked = 1 | |
1196 | * blocked = 0 | |
1197 | * commit_done = 1 | |
1198 | */ | |
79154b1b CM |
1199 | int btrfs_commit_transaction(struct btrfs_trans_handle *trans, |
1200 | struct btrfs_root *root) | |
1201 | { | |
15ee9bc7 | 1202 | unsigned long joined = 0; |
79154b1b | 1203 | struct btrfs_transaction *cur_trans; |
8fd17795 | 1204 | struct btrfs_transaction *prev_trans = NULL; |
79154b1b | 1205 | DEFINE_WAIT(wait); |
15ee9bc7 | 1206 | int ret; |
89573b9c CM |
1207 | int should_grow = 0; |
1208 | unsigned long now = get_seconds(); | |
dccae999 | 1209 | int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT); |
79154b1b | 1210 | |
5a3f23d5 CM |
1211 | btrfs_run_ordered_operations(root, 0); |
1212 | ||
56bec294 CM |
1213 | /* make a pass through all the delayed refs we have so far |
1214 | * any runnings procs may add more while we are here | |
1215 | */ | |
1216 | ret = btrfs_run_delayed_refs(trans, root, 0); | |
1217 | BUG_ON(ret); | |
1218 | ||
a22285a6 YZ |
1219 | btrfs_trans_release_metadata(trans, root); |
1220 | ||
b7ec40d7 | 1221 | cur_trans = trans->transaction; |
56bec294 CM |
1222 | /* |
1223 | * set the flushing flag so procs in this transaction have to | |
1224 | * start sending their work down. | |
1225 | */ | |
b7ec40d7 | 1226 | cur_trans->delayed_refs.flushing = 1; |
56bec294 | 1227 | |
c3e69d58 | 1228 | ret = btrfs_run_delayed_refs(trans, root, 0); |
56bec294 CM |
1229 | BUG_ON(ret); |
1230 | ||
a4abeea4 | 1231 | spin_lock(&cur_trans->commit_lock); |
b7ec40d7 | 1232 | if (cur_trans->in_commit) { |
a4abeea4 | 1233 | spin_unlock(&cur_trans->commit_lock); |
13c5a93e | 1234 | atomic_inc(&cur_trans->use_count); |
79154b1b | 1235 | btrfs_end_transaction(trans, root); |
ccd467d6 | 1236 | |
79154b1b CM |
1237 | ret = wait_for_commit(root, cur_trans); |
1238 | BUG_ON(ret); | |
15ee9bc7 | 1239 | |
79154b1b | 1240 | put_transaction(cur_trans); |
15ee9bc7 | 1241 | |
79154b1b CM |
1242 | return 0; |
1243 | } | |
4313b399 | 1244 | |
2c90e5d6 | 1245 | trans->transaction->in_commit = 1; |
f9295749 | 1246 | trans->transaction->blocked = 1; |
a4abeea4 | 1247 | spin_unlock(&cur_trans->commit_lock); |
bb9c12c9 SW |
1248 | wake_up(&root->fs_info->transaction_blocked_wait); |
1249 | ||
a4abeea4 | 1250 | spin_lock(&root->fs_info->trans_lock); |
ccd467d6 CM |
1251 | if (cur_trans->list.prev != &root->fs_info->trans_list) { |
1252 | prev_trans = list_entry(cur_trans->list.prev, | |
1253 | struct btrfs_transaction, list); | |
1254 | if (!prev_trans->commit_done) { | |
13c5a93e | 1255 | atomic_inc(&prev_trans->use_count); |
a4abeea4 | 1256 | spin_unlock(&root->fs_info->trans_lock); |
ccd467d6 CM |
1257 | |
1258 | wait_for_commit(root, prev_trans); | |
ccd467d6 | 1259 | |
15ee9bc7 | 1260 | put_transaction(prev_trans); |
a4abeea4 JB |
1261 | } else { |
1262 | spin_unlock(&root->fs_info->trans_lock); | |
ccd467d6 | 1263 | } |
a4abeea4 JB |
1264 | } else { |
1265 | spin_unlock(&root->fs_info->trans_lock); | |
ccd467d6 | 1266 | } |
15ee9bc7 | 1267 | |
89573b9c CM |
1268 | if (now < cur_trans->start_time || now - cur_trans->start_time < 1) |
1269 | should_grow = 1; | |
1270 | ||
15ee9bc7 | 1271 | do { |
7ea394f1 | 1272 | int snap_pending = 0; |
a4abeea4 | 1273 | |
15ee9bc7 | 1274 | joined = cur_trans->num_joined; |
7ea394f1 YZ |
1275 | if (!list_empty(&trans->transaction->pending_snapshots)) |
1276 | snap_pending = 1; | |
1277 | ||
2c90e5d6 | 1278 | WARN_ON(cur_trans != trans->transaction); |
15ee9bc7 | 1279 | |
0bdb1db2 | 1280 | if (flush_on_commit || snap_pending) { |
24bbcf04 YZ |
1281 | btrfs_start_delalloc_inodes(root, 1); |
1282 | ret = btrfs_wait_ordered_extents(root, 0, 1); | |
ebecd3d9 | 1283 | BUG_ON(ret); |
7ea394f1 YZ |
1284 | } |
1285 | ||
16cdcec7 MX |
1286 | ret = btrfs_run_delayed_items(trans, root); |
1287 | BUG_ON(ret); | |
1288 | ||
5a3f23d5 CM |
1289 | /* |
1290 | * rename don't use btrfs_join_transaction, so, once we | |
1291 | * set the transaction to blocked above, we aren't going | |
1292 | * to get any new ordered operations. We can safely run | |
1293 | * it here and no for sure that nothing new will be added | |
1294 | * to the list | |
1295 | */ | |
1296 | btrfs_run_ordered_operations(root, 1); | |
1297 | ||
ed3b3d31 CM |
1298 | prepare_to_wait(&cur_trans->writer_wait, &wait, |
1299 | TASK_UNINTERRUPTIBLE); | |
1300 | ||
13c5a93e | 1301 | if (atomic_read(&cur_trans->num_writers) > 1) |
99d16cbc SW |
1302 | schedule_timeout(MAX_SCHEDULE_TIMEOUT); |
1303 | else if (should_grow) | |
1304 | schedule_timeout(1); | |
15ee9bc7 | 1305 | |
15ee9bc7 | 1306 | finish_wait(&cur_trans->writer_wait, &wait); |
13c5a93e | 1307 | } while (atomic_read(&cur_trans->num_writers) > 1 || |
89573b9c | 1308 | (should_grow && cur_trans->num_joined != joined)); |
15ee9bc7 | 1309 | |
ed0ca140 JB |
1310 | /* |
1311 | * Ok now we need to make sure to block out any other joins while we | |
1312 | * commit the transaction. We could have started a join before setting | |
1313 | * no_join so make sure to wait for num_writers to == 1 again. | |
1314 | */ | |
1315 | spin_lock(&root->fs_info->trans_lock); | |
1316 | root->fs_info->trans_no_join = 1; | |
1317 | spin_unlock(&root->fs_info->trans_lock); | |
1318 | wait_event(cur_trans->writer_wait, | |
1319 | atomic_read(&cur_trans->num_writers) == 1); | |
1320 | ||
7585717f CM |
1321 | /* |
1322 | * the reloc mutex makes sure that we stop | |
1323 | * the balancing code from coming in and moving | |
1324 | * extents around in the middle of the commit | |
1325 | */ | |
1326 | mutex_lock(&root->fs_info->reloc_mutex); | |
1327 | ||
e999376f | 1328 | ret = btrfs_run_delayed_items(trans, root); |
3063d29f CM |
1329 | BUG_ON(ret); |
1330 | ||
e999376f | 1331 | ret = create_pending_snapshots(trans, root->fs_info); |
16cdcec7 MX |
1332 | BUG_ON(ret); |
1333 | ||
56bec294 CM |
1334 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
1335 | BUG_ON(ret); | |
1336 | ||
e999376f CM |
1337 | /* |
1338 | * make sure none of the code above managed to slip in a | |
1339 | * delayed item | |
1340 | */ | |
1341 | btrfs_assert_delayed_root_empty(root); | |
1342 | ||
2c90e5d6 | 1343 | WARN_ON(cur_trans != trans->transaction); |
dc17ff8f | 1344 | |
a2de733c | 1345 | btrfs_scrub_pause(root); |
e02119d5 CM |
1346 | /* btrfs_commit_tree_roots is responsible for getting the |
1347 | * various roots consistent with each other. Every pointer | |
1348 | * in the tree of tree roots has to point to the most up to date | |
1349 | * root for every subvolume and other tree. So, we have to keep | |
1350 | * the tree logging code from jumping in and changing any | |
1351 | * of the trees. | |
1352 | * | |
1353 | * At this point in the commit, there can't be any tree-log | |
1354 | * writers, but a little lower down we drop the trans mutex | |
1355 | * and let new people in. By holding the tree_log_mutex | |
1356 | * from now until after the super is written, we avoid races | |
1357 | * with the tree-log code. | |
1358 | */ | |
1359 | mutex_lock(&root->fs_info->tree_log_mutex); | |
1360 | ||
5d4f98a2 | 1361 | ret = commit_fs_roots(trans, root); |
54aa1f4d CM |
1362 | BUG_ON(ret); |
1363 | ||
5d4f98a2 | 1364 | /* commit_fs_roots gets rid of all the tree log roots, it is now |
e02119d5 CM |
1365 | * safe to free the root of tree log roots |
1366 | */ | |
1367 | btrfs_free_log_root_tree(trans, root->fs_info); | |
1368 | ||
5d4f98a2 | 1369 | ret = commit_cowonly_roots(trans, root); |
79154b1b | 1370 | BUG_ON(ret); |
54aa1f4d | 1371 | |
11833d66 YZ |
1372 | btrfs_prepare_extent_commit(trans, root); |
1373 | ||
78fae27e | 1374 | cur_trans = root->fs_info->running_transaction; |
5d4f98a2 YZ |
1375 | |
1376 | btrfs_set_root_node(&root->fs_info->tree_root->root_item, | |
1377 | root->fs_info->tree_root->node); | |
817d52f8 | 1378 | switch_commit_root(root->fs_info->tree_root); |
5d4f98a2 YZ |
1379 | |
1380 | btrfs_set_root_node(&root->fs_info->chunk_root->root_item, | |
1381 | root->fs_info->chunk_root->node); | |
817d52f8 | 1382 | switch_commit_root(root->fs_info->chunk_root); |
5d4f98a2 YZ |
1383 | |
1384 | update_super_roots(root); | |
e02119d5 CM |
1385 | |
1386 | if (!root->fs_info->log_root_recovering) { | |
1387 | btrfs_set_super_log_root(&root->fs_info->super_copy, 0); | |
1388 | btrfs_set_super_log_root_level(&root->fs_info->super_copy, 0); | |
1389 | } | |
1390 | ||
a061fc8d CM |
1391 | memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy, |
1392 | sizeof(root->fs_info->super_copy)); | |
ccd467d6 | 1393 | |
f9295749 | 1394 | trans->transaction->blocked = 0; |
a4abeea4 JB |
1395 | spin_lock(&root->fs_info->trans_lock); |
1396 | root->fs_info->running_transaction = NULL; | |
1397 | root->fs_info->trans_no_join = 0; | |
1398 | spin_unlock(&root->fs_info->trans_lock); | |
7585717f | 1399 | mutex_unlock(&root->fs_info->reloc_mutex); |
b7ec40d7 | 1400 | |
f9295749 | 1401 | wake_up(&root->fs_info->transaction_wait); |
e6dcd2dc | 1402 | |
79154b1b CM |
1403 | ret = btrfs_write_and_wait_transaction(trans, root); |
1404 | BUG_ON(ret); | |
a512bbf8 | 1405 | write_ctree_super(trans, root, 0); |
4313b399 | 1406 | |
e02119d5 CM |
1407 | /* |
1408 | * the super is written, we can safely allow the tree-loggers | |
1409 | * to go about their business | |
1410 | */ | |
1411 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
1412 | ||
11833d66 | 1413 | btrfs_finish_extent_commit(trans, root); |
4313b399 | 1414 | |
2c90e5d6 | 1415 | cur_trans->commit_done = 1; |
b7ec40d7 | 1416 | |
15ee9bc7 | 1417 | root->fs_info->last_trans_committed = cur_trans->transid; |
817d52f8 | 1418 | |
2c90e5d6 | 1419 | wake_up(&cur_trans->commit_wait); |
3de4586c | 1420 | |
a4abeea4 | 1421 | spin_lock(&root->fs_info->trans_lock); |
13c5a93e | 1422 | list_del_init(&cur_trans->list); |
a4abeea4 JB |
1423 | spin_unlock(&root->fs_info->trans_lock); |
1424 | ||
78fae27e | 1425 | put_transaction(cur_trans); |
79154b1b | 1426 | put_transaction(cur_trans); |
58176a96 | 1427 | |
1abe9b8a | 1428 | trace_btrfs_transaction_commit(root); |
1429 | ||
a2de733c AJ |
1430 | btrfs_scrub_continue(root); |
1431 | ||
9ed74f2d JB |
1432 | if (current->journal_info == trans) |
1433 | current->journal_info = NULL; | |
1434 | ||
2c90e5d6 | 1435 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
24bbcf04 YZ |
1436 | |
1437 | if (current != root->fs_info->transaction_kthread) | |
1438 | btrfs_run_delayed_iputs(root); | |
1439 | ||
79154b1b CM |
1440 | return ret; |
1441 | } | |
1442 | ||
d352ac68 CM |
1443 | /* |
1444 | * interface function to delete all the snapshots we have scheduled for deletion | |
1445 | */ | |
e9d0b13b CM |
1446 | int btrfs_clean_old_snapshots(struct btrfs_root *root) |
1447 | { | |
5d4f98a2 YZ |
1448 | LIST_HEAD(list); |
1449 | struct btrfs_fs_info *fs_info = root->fs_info; | |
1450 | ||
a4abeea4 | 1451 | spin_lock(&fs_info->trans_lock); |
5d4f98a2 | 1452 | list_splice_init(&fs_info->dead_roots, &list); |
a4abeea4 | 1453 | spin_unlock(&fs_info->trans_lock); |
e9d0b13b | 1454 | |
5d4f98a2 YZ |
1455 | while (!list_empty(&list)) { |
1456 | root = list_entry(list.next, struct btrfs_root, root_list); | |
76dda93c YZ |
1457 | list_del(&root->root_list); |
1458 | ||
16cdcec7 MX |
1459 | btrfs_kill_all_delayed_nodes(root); |
1460 | ||
76dda93c YZ |
1461 | if (btrfs_header_backref_rev(root->node) < |
1462 | BTRFS_MIXED_BACKREF_REV) | |
3fd0a558 | 1463 | btrfs_drop_snapshot(root, NULL, 0); |
76dda93c | 1464 | else |
3fd0a558 | 1465 | btrfs_drop_snapshot(root, NULL, 1); |
e9d0b13b CM |
1466 | } |
1467 | return 0; | |
1468 | } |