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; | |
06d5a589 | 263 | int retries = 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 | } | |
a22285a6 YZ |
277 | again: |
278 | h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS); | |
279 | if (!h) | |
280 | return ERR_PTR(-ENOMEM); | |
37d1aeee | 281 | |
a22285a6 | 282 | if (may_wait_transaction(root, type)) |
37d1aeee | 283 | wait_current_trans(root); |
a22285a6 | 284 | |
a4abeea4 JB |
285 | do { |
286 | ret = join_transaction(root, type == TRANS_JOIN_NOLOCK); | |
287 | if (ret == -EBUSY) | |
288 | wait_current_trans(root); | |
289 | } while (ret == -EBUSY); | |
290 | ||
db5b493a | 291 | if (ret < 0) { |
6e8df2ae | 292 | kmem_cache_free(btrfs_trans_handle_cachep, h); |
db5b493a TI |
293 | return ERR_PTR(ret); |
294 | } | |
0f7d52f4 | 295 | |
a22285a6 | 296 | cur_trans = root->fs_info->running_transaction; |
a22285a6 YZ |
297 | |
298 | h->transid = cur_trans->transid; | |
299 | h->transaction = cur_trans; | |
79154b1b | 300 | h->blocks_used = 0; |
a22285a6 | 301 | h->bytes_reserved = 0; |
56bec294 | 302 | h->delayed_ref_updates = 0; |
2a1eb461 | 303 | h->use_count = 1; |
f0486c68 | 304 | h->block_rsv = NULL; |
2a1eb461 | 305 | h->orig_rsv = NULL; |
b7ec40d7 | 306 | |
a22285a6 YZ |
307 | smp_mb(); |
308 | if (cur_trans->blocked && may_wait_transaction(root, type)) { | |
309 | btrfs_commit_transaction(h, root); | |
310 | goto again; | |
311 | } | |
312 | ||
313 | if (num_items > 0) { | |
8bb8ab2e | 314 | ret = btrfs_trans_reserve_metadata(h, root, num_items); |
06d5a589 JB |
315 | if (ret == -EAGAIN && !retries) { |
316 | retries++; | |
a22285a6 YZ |
317 | btrfs_commit_transaction(h, root); |
318 | goto again; | |
06d5a589 JB |
319 | } else if (ret == -EAGAIN) { |
320 | /* | |
321 | * We have already retried and got EAGAIN, so really we | |
322 | * don't have space, so set ret to -ENOSPC. | |
323 | */ | |
324 | ret = -ENOSPC; | |
a22285a6 | 325 | } |
06d5a589 | 326 | |
a22285a6 YZ |
327 | if (ret < 0) { |
328 | btrfs_end_transaction(h, root); | |
329 | return ERR_PTR(ret); | |
330 | } | |
331 | } | |
9ed74f2d | 332 | |
2a1eb461 | 333 | got_it: |
a4abeea4 | 334 | btrfs_record_root_in_trans(h, root); |
a22285a6 YZ |
335 | |
336 | if (!current->journal_info && type != TRANS_USERSPACE) | |
337 | current->journal_info = h; | |
79154b1b CM |
338 | return h; |
339 | } | |
340 | ||
f9295749 | 341 | struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, |
a22285a6 | 342 | int num_items) |
f9295749 | 343 | { |
a22285a6 | 344 | return start_transaction(root, num_items, TRANS_START); |
f9295749 | 345 | } |
7a7eaa40 | 346 | struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root) |
f9295749 | 347 | { |
a22285a6 | 348 | return start_transaction(root, 0, TRANS_JOIN); |
f9295749 CM |
349 | } |
350 | ||
7a7eaa40 | 351 | struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root) |
0af3d00b JB |
352 | { |
353 | return start_transaction(root, 0, TRANS_JOIN_NOLOCK); | |
354 | } | |
355 | ||
7a7eaa40 | 356 | struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root) |
9ca9ee09 | 357 | { |
7a7eaa40 | 358 | return start_transaction(root, 0, TRANS_USERSPACE); |
9ca9ee09 SW |
359 | } |
360 | ||
d352ac68 | 361 | /* wait for a transaction commit to be fully complete */ |
89ce8a63 CM |
362 | static noinline int wait_for_commit(struct btrfs_root *root, |
363 | struct btrfs_transaction *commit) | |
364 | { | |
365 | DEFINE_WAIT(wait); | |
d397712b | 366 | while (!commit->commit_done) { |
89ce8a63 CM |
367 | prepare_to_wait(&commit->commit_wait, &wait, |
368 | TASK_UNINTERRUPTIBLE); | |
369 | if (commit->commit_done) | |
370 | break; | |
89ce8a63 | 371 | schedule(); |
89ce8a63 | 372 | } |
89ce8a63 CM |
373 | finish_wait(&commit->commit_wait, &wait); |
374 | return 0; | |
375 | } | |
376 | ||
46204592 SW |
377 | int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid) |
378 | { | |
379 | struct btrfs_transaction *cur_trans = NULL, *t; | |
380 | int ret; | |
381 | ||
46204592 SW |
382 | ret = 0; |
383 | if (transid) { | |
384 | if (transid <= root->fs_info->last_trans_committed) | |
a4abeea4 | 385 | goto out; |
46204592 SW |
386 | |
387 | /* find specified transaction */ | |
a4abeea4 | 388 | spin_lock(&root->fs_info->trans_lock); |
46204592 SW |
389 | list_for_each_entry(t, &root->fs_info->trans_list, list) { |
390 | if (t->transid == transid) { | |
391 | cur_trans = t; | |
a4abeea4 | 392 | atomic_inc(&cur_trans->use_count); |
46204592 SW |
393 | break; |
394 | } | |
395 | if (t->transid > transid) | |
396 | break; | |
397 | } | |
a4abeea4 | 398 | spin_unlock(&root->fs_info->trans_lock); |
46204592 SW |
399 | ret = -EINVAL; |
400 | if (!cur_trans) | |
a4abeea4 | 401 | goto out; /* bad transid */ |
46204592 SW |
402 | } else { |
403 | /* find newest transaction that is committing | committed */ | |
a4abeea4 | 404 | spin_lock(&root->fs_info->trans_lock); |
46204592 SW |
405 | list_for_each_entry_reverse(t, &root->fs_info->trans_list, |
406 | list) { | |
407 | if (t->in_commit) { | |
408 | if (t->commit_done) | |
3473f3c0 | 409 | break; |
46204592 | 410 | cur_trans = t; |
a4abeea4 | 411 | atomic_inc(&cur_trans->use_count); |
46204592 SW |
412 | break; |
413 | } | |
414 | } | |
a4abeea4 | 415 | spin_unlock(&root->fs_info->trans_lock); |
46204592 | 416 | if (!cur_trans) |
a4abeea4 | 417 | goto out; /* nothing committing|committed */ |
46204592 SW |
418 | } |
419 | ||
46204592 SW |
420 | wait_for_commit(root, cur_trans); |
421 | ||
46204592 SW |
422 | put_transaction(cur_trans); |
423 | ret = 0; | |
a4abeea4 | 424 | out: |
46204592 SW |
425 | return ret; |
426 | } | |
427 | ||
37d1aeee CM |
428 | void btrfs_throttle(struct btrfs_root *root) |
429 | { | |
a4abeea4 | 430 | if (!atomic_read(&root->fs_info->open_ioctl_trans)) |
9ca9ee09 | 431 | wait_current_trans(root); |
37d1aeee CM |
432 | } |
433 | ||
8929ecfa YZ |
434 | static int should_end_transaction(struct btrfs_trans_handle *trans, |
435 | struct btrfs_root *root) | |
436 | { | |
437 | int ret; | |
438 | ret = btrfs_block_rsv_check(trans, root, | |
439 | &root->fs_info->global_block_rsv, 0, 5); | |
440 | return ret ? 1 : 0; | |
441 | } | |
442 | ||
443 | int btrfs_should_end_transaction(struct btrfs_trans_handle *trans, | |
444 | struct btrfs_root *root) | |
445 | { | |
446 | struct btrfs_transaction *cur_trans = trans->transaction; | |
447 | int updates; | |
448 | ||
a4abeea4 | 449 | smp_mb(); |
8929ecfa YZ |
450 | if (cur_trans->blocked || cur_trans->delayed_refs.flushing) |
451 | return 1; | |
452 | ||
453 | updates = trans->delayed_ref_updates; | |
454 | trans->delayed_ref_updates = 0; | |
455 | if (updates) | |
456 | btrfs_run_delayed_refs(trans, root, updates); | |
457 | ||
458 | return should_end_transaction(trans, root); | |
459 | } | |
460 | ||
89ce8a63 | 461 | static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, |
0af3d00b | 462 | struct btrfs_root *root, int throttle, int lock) |
79154b1b | 463 | { |
8929ecfa | 464 | struct btrfs_transaction *cur_trans = trans->transaction; |
ab78c84d | 465 | struct btrfs_fs_info *info = root->fs_info; |
c3e69d58 CM |
466 | int count = 0; |
467 | ||
2a1eb461 JB |
468 | if (--trans->use_count) { |
469 | trans->block_rsv = trans->orig_rsv; | |
470 | return 0; | |
471 | } | |
472 | ||
c3e69d58 CM |
473 | while (count < 4) { |
474 | unsigned long cur = trans->delayed_ref_updates; | |
475 | trans->delayed_ref_updates = 0; | |
476 | if (cur && | |
477 | trans->transaction->delayed_refs.num_heads_ready > 64) { | |
478 | trans->delayed_ref_updates = 0; | |
b7ec40d7 CM |
479 | |
480 | /* | |
481 | * do a full flush if the transaction is trying | |
482 | * to close | |
483 | */ | |
484 | if (trans->transaction->delayed_refs.flushing) | |
485 | cur = 0; | |
c3e69d58 CM |
486 | btrfs_run_delayed_refs(trans, root, cur); |
487 | } else { | |
488 | break; | |
489 | } | |
490 | count++; | |
56bec294 CM |
491 | } |
492 | ||
a22285a6 YZ |
493 | btrfs_trans_release_metadata(trans, root); |
494 | ||
a4abeea4 JB |
495 | if (lock && !atomic_read(&root->fs_info->open_ioctl_trans) && |
496 | should_end_transaction(trans, root)) { | |
8929ecfa | 497 | trans->transaction->blocked = 1; |
a4abeea4 JB |
498 | smp_wmb(); |
499 | } | |
8929ecfa | 500 | |
0af3d00b | 501 | if (lock && cur_trans->blocked && !cur_trans->in_commit) { |
8929ecfa YZ |
502 | if (throttle) |
503 | return btrfs_commit_transaction(trans, root); | |
504 | else | |
505 | wake_up_process(info->transaction_kthread); | |
506 | } | |
507 | ||
8929ecfa | 508 | WARN_ON(cur_trans != info->running_transaction); |
13c5a93e JB |
509 | WARN_ON(atomic_read(&cur_trans->num_writers) < 1); |
510 | atomic_dec(&cur_trans->num_writers); | |
89ce8a63 | 511 | |
99d16cbc | 512 | smp_mb(); |
79154b1b CM |
513 | if (waitqueue_active(&cur_trans->writer_wait)) |
514 | wake_up(&cur_trans->writer_wait); | |
79154b1b | 515 | put_transaction(cur_trans); |
9ed74f2d JB |
516 | |
517 | if (current->journal_info == trans) | |
518 | current->journal_info = NULL; | |
d6025579 | 519 | memset(trans, 0, sizeof(*trans)); |
2c90e5d6 | 520 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
ab78c84d | 521 | |
24bbcf04 YZ |
522 | if (throttle) |
523 | btrfs_run_delayed_iputs(root); | |
524 | ||
79154b1b CM |
525 | return 0; |
526 | } | |
527 | ||
89ce8a63 CM |
528 | int btrfs_end_transaction(struct btrfs_trans_handle *trans, |
529 | struct btrfs_root *root) | |
530 | { | |
16cdcec7 MX |
531 | int ret; |
532 | ||
533 | ret = __btrfs_end_transaction(trans, root, 0, 1); | |
534 | if (ret) | |
535 | return ret; | |
536 | return 0; | |
89ce8a63 CM |
537 | } |
538 | ||
539 | int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans, | |
540 | struct btrfs_root *root) | |
541 | { | |
16cdcec7 MX |
542 | int ret; |
543 | ||
544 | ret = __btrfs_end_transaction(trans, root, 1, 1); | |
545 | if (ret) | |
546 | return ret; | |
547 | return 0; | |
0af3d00b JB |
548 | } |
549 | ||
550 | int btrfs_end_transaction_nolock(struct btrfs_trans_handle *trans, | |
551 | struct btrfs_root *root) | |
552 | { | |
16cdcec7 MX |
553 | int ret; |
554 | ||
555 | ret = __btrfs_end_transaction(trans, root, 0, 0); | |
556 | if (ret) | |
557 | return ret; | |
558 | return 0; | |
559 | } | |
560 | ||
561 | int btrfs_end_transaction_dmeta(struct btrfs_trans_handle *trans, | |
562 | struct btrfs_root *root) | |
563 | { | |
564 | return __btrfs_end_transaction(trans, root, 1, 1); | |
89ce8a63 CM |
565 | } |
566 | ||
d352ac68 CM |
567 | /* |
568 | * when btree blocks are allocated, they have some corresponding bits set for | |
569 | * them in one of two extent_io trees. This is used to make sure all of | |
690587d1 | 570 | * those extents are sent to disk but does not wait on them |
d352ac68 | 571 | */ |
690587d1 | 572 | int btrfs_write_marked_extents(struct btrfs_root *root, |
8cef4e16 | 573 | struct extent_io_tree *dirty_pages, int mark) |
79154b1b | 574 | { |
7c4452b9 | 575 | int ret; |
777e6bd7 | 576 | int err = 0; |
7c4452b9 CM |
577 | int werr = 0; |
578 | struct page *page; | |
7c4452b9 | 579 | struct inode *btree_inode = root->fs_info->btree_inode; |
777e6bd7 | 580 | u64 start = 0; |
5f39d397 CM |
581 | u64 end; |
582 | unsigned long index; | |
7c4452b9 | 583 | |
d397712b | 584 | while (1) { |
777e6bd7 | 585 | ret = find_first_extent_bit(dirty_pages, start, &start, &end, |
8cef4e16 | 586 | mark); |
5f39d397 | 587 | if (ret) |
7c4452b9 | 588 | break; |
d397712b | 589 | while (start <= end) { |
777e6bd7 CM |
590 | cond_resched(); |
591 | ||
5f39d397 | 592 | index = start >> PAGE_CACHE_SHIFT; |
35ebb934 | 593 | start = (u64)(index + 1) << PAGE_CACHE_SHIFT; |
4bef0848 | 594 | page = find_get_page(btree_inode->i_mapping, index); |
7c4452b9 CM |
595 | if (!page) |
596 | continue; | |
4bef0848 CM |
597 | |
598 | btree_lock_page_hook(page); | |
599 | if (!page->mapping) { | |
600 | unlock_page(page); | |
601 | page_cache_release(page); | |
602 | continue; | |
603 | } | |
604 | ||
6702ed49 CM |
605 | if (PageWriteback(page)) { |
606 | if (PageDirty(page)) | |
607 | wait_on_page_writeback(page); | |
608 | else { | |
609 | unlock_page(page); | |
610 | page_cache_release(page); | |
611 | continue; | |
612 | } | |
613 | } | |
7c4452b9 CM |
614 | err = write_one_page(page, 0); |
615 | if (err) | |
616 | werr = err; | |
617 | page_cache_release(page); | |
618 | } | |
619 | } | |
690587d1 CM |
620 | if (err) |
621 | werr = err; | |
622 | return werr; | |
623 | } | |
624 | ||
625 | /* | |
626 | * when btree blocks are allocated, they have some corresponding bits set for | |
627 | * them in one of two extent_io trees. This is used to make sure all of | |
628 | * those extents are on disk for transaction or log commit. We wait | |
629 | * on all the pages and clear them from the dirty pages state tree | |
630 | */ | |
631 | int btrfs_wait_marked_extents(struct btrfs_root *root, | |
8cef4e16 | 632 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 CM |
633 | { |
634 | int ret; | |
635 | int err = 0; | |
636 | int werr = 0; | |
637 | struct page *page; | |
638 | struct inode *btree_inode = root->fs_info->btree_inode; | |
639 | u64 start = 0; | |
640 | u64 end; | |
641 | unsigned long index; | |
642 | ||
d397712b | 643 | while (1) { |
8cef4e16 YZ |
644 | ret = find_first_extent_bit(dirty_pages, start, &start, &end, |
645 | mark); | |
777e6bd7 CM |
646 | if (ret) |
647 | break; | |
648 | ||
8cef4e16 | 649 | clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS); |
d397712b | 650 | while (start <= end) { |
777e6bd7 CM |
651 | index = start >> PAGE_CACHE_SHIFT; |
652 | start = (u64)(index + 1) << PAGE_CACHE_SHIFT; | |
653 | page = find_get_page(btree_inode->i_mapping, index); | |
654 | if (!page) | |
655 | continue; | |
656 | if (PageDirty(page)) { | |
4bef0848 CM |
657 | btree_lock_page_hook(page); |
658 | wait_on_page_writeback(page); | |
777e6bd7 CM |
659 | err = write_one_page(page, 0); |
660 | if (err) | |
661 | werr = err; | |
662 | } | |
105d931d | 663 | wait_on_page_writeback(page); |
777e6bd7 CM |
664 | page_cache_release(page); |
665 | cond_resched(); | |
666 | } | |
667 | } | |
7c4452b9 CM |
668 | if (err) |
669 | werr = err; | |
670 | return werr; | |
79154b1b CM |
671 | } |
672 | ||
690587d1 CM |
673 | /* |
674 | * when btree blocks are allocated, they have some corresponding bits set for | |
675 | * them in one of two extent_io trees. This is used to make sure all of | |
676 | * those extents are on disk for transaction or log commit | |
677 | */ | |
678 | int btrfs_write_and_wait_marked_extents(struct btrfs_root *root, | |
8cef4e16 | 679 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 CM |
680 | { |
681 | int ret; | |
682 | int ret2; | |
683 | ||
8cef4e16 YZ |
684 | ret = btrfs_write_marked_extents(root, dirty_pages, mark); |
685 | ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark); | |
690587d1 CM |
686 | return ret || ret2; |
687 | } | |
688 | ||
d0c803c4 CM |
689 | int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans, |
690 | struct btrfs_root *root) | |
691 | { | |
692 | if (!trans || !trans->transaction) { | |
693 | struct inode *btree_inode; | |
694 | btree_inode = root->fs_info->btree_inode; | |
695 | return filemap_write_and_wait(btree_inode->i_mapping); | |
696 | } | |
697 | return btrfs_write_and_wait_marked_extents(root, | |
8cef4e16 YZ |
698 | &trans->transaction->dirty_pages, |
699 | EXTENT_DIRTY); | |
d0c803c4 CM |
700 | } |
701 | ||
d352ac68 CM |
702 | /* |
703 | * this is used to update the root pointer in the tree of tree roots. | |
704 | * | |
705 | * But, in the case of the extent allocation tree, updating the root | |
706 | * pointer may allocate blocks which may change the root of the extent | |
707 | * allocation tree. | |
708 | * | |
709 | * So, this loops and repeats and makes sure the cowonly root didn't | |
710 | * change while the root pointer was being updated in the metadata. | |
711 | */ | |
0b86a832 CM |
712 | static int update_cowonly_root(struct btrfs_trans_handle *trans, |
713 | struct btrfs_root *root) | |
79154b1b CM |
714 | { |
715 | int ret; | |
0b86a832 | 716 | u64 old_root_bytenr; |
86b9f2ec | 717 | u64 old_root_used; |
0b86a832 | 718 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
79154b1b | 719 | |
86b9f2ec | 720 | old_root_used = btrfs_root_used(&root->root_item); |
0b86a832 | 721 | btrfs_write_dirty_block_groups(trans, root); |
56bec294 | 722 | |
d397712b | 723 | while (1) { |
0b86a832 | 724 | old_root_bytenr = btrfs_root_bytenr(&root->root_item); |
86b9f2ec YZ |
725 | if (old_root_bytenr == root->node->start && |
726 | old_root_used == btrfs_root_used(&root->root_item)) | |
79154b1b | 727 | break; |
87ef2bb4 | 728 | |
5d4f98a2 | 729 | btrfs_set_root_node(&root->root_item, root->node); |
79154b1b | 730 | ret = btrfs_update_root(trans, tree_root, |
0b86a832 CM |
731 | &root->root_key, |
732 | &root->root_item); | |
79154b1b | 733 | BUG_ON(ret); |
56bec294 | 734 | |
86b9f2ec | 735 | old_root_used = btrfs_root_used(&root->root_item); |
4a8c9a62 | 736 | ret = btrfs_write_dirty_block_groups(trans, root); |
56bec294 | 737 | BUG_ON(ret); |
0b86a832 | 738 | } |
276e680d YZ |
739 | |
740 | if (root != root->fs_info->extent_root) | |
741 | switch_commit_root(root); | |
742 | ||
0b86a832 CM |
743 | return 0; |
744 | } | |
745 | ||
d352ac68 CM |
746 | /* |
747 | * update all the cowonly tree roots on disk | |
748 | */ | |
5d4f98a2 YZ |
749 | static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans, |
750 | struct btrfs_root *root) | |
0b86a832 CM |
751 | { |
752 | struct btrfs_fs_info *fs_info = root->fs_info; | |
753 | struct list_head *next; | |
84234f3a | 754 | struct extent_buffer *eb; |
56bec294 | 755 | int ret; |
84234f3a | 756 | |
56bec294 CM |
757 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
758 | BUG_ON(ret); | |
87ef2bb4 | 759 | |
84234f3a | 760 | eb = btrfs_lock_root_node(fs_info->tree_root); |
9fa8cfe7 | 761 | btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 0, &eb); |
84234f3a YZ |
762 | btrfs_tree_unlock(eb); |
763 | free_extent_buffer(eb); | |
0b86a832 | 764 | |
56bec294 CM |
765 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
766 | BUG_ON(ret); | |
87ef2bb4 | 767 | |
d397712b | 768 | while (!list_empty(&fs_info->dirty_cowonly_roots)) { |
0b86a832 CM |
769 | next = fs_info->dirty_cowonly_roots.next; |
770 | list_del_init(next); | |
771 | root = list_entry(next, struct btrfs_root, dirty_list); | |
87ef2bb4 | 772 | |
0b86a832 | 773 | update_cowonly_root(trans, root); |
79154b1b | 774 | } |
276e680d YZ |
775 | |
776 | down_write(&fs_info->extent_commit_sem); | |
777 | switch_commit_root(fs_info->extent_root); | |
778 | up_write(&fs_info->extent_commit_sem); | |
779 | ||
79154b1b CM |
780 | return 0; |
781 | } | |
782 | ||
d352ac68 CM |
783 | /* |
784 | * dead roots are old snapshots that need to be deleted. This allocates | |
785 | * a dirty root struct and adds it into the list of dead roots that need to | |
786 | * be deleted | |
787 | */ | |
5d4f98a2 | 788 | int btrfs_add_dead_root(struct btrfs_root *root) |
5eda7b5e | 789 | { |
a4abeea4 | 790 | spin_lock(&root->fs_info->trans_lock); |
5d4f98a2 | 791 | list_add(&root->root_list, &root->fs_info->dead_roots); |
a4abeea4 | 792 | spin_unlock(&root->fs_info->trans_lock); |
5eda7b5e CM |
793 | return 0; |
794 | } | |
795 | ||
d352ac68 | 796 | /* |
5d4f98a2 | 797 | * update all the cowonly tree roots on disk |
d352ac68 | 798 | */ |
5d4f98a2 YZ |
799 | static noinline int commit_fs_roots(struct btrfs_trans_handle *trans, |
800 | struct btrfs_root *root) | |
0f7d52f4 | 801 | { |
0f7d52f4 | 802 | struct btrfs_root *gang[8]; |
5d4f98a2 | 803 | struct btrfs_fs_info *fs_info = root->fs_info; |
0f7d52f4 CM |
804 | int i; |
805 | int ret; | |
54aa1f4d CM |
806 | int err = 0; |
807 | ||
a4abeea4 | 808 | spin_lock(&fs_info->fs_roots_radix_lock); |
d397712b | 809 | while (1) { |
5d4f98a2 YZ |
810 | ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, |
811 | (void **)gang, 0, | |
0f7d52f4 CM |
812 | ARRAY_SIZE(gang), |
813 | BTRFS_ROOT_TRANS_TAG); | |
814 | if (ret == 0) | |
815 | break; | |
816 | for (i = 0; i < ret; i++) { | |
817 | root = gang[i]; | |
5d4f98a2 YZ |
818 | radix_tree_tag_clear(&fs_info->fs_roots_radix, |
819 | (unsigned long)root->root_key.objectid, | |
820 | BTRFS_ROOT_TRANS_TAG); | |
a4abeea4 | 821 | spin_unlock(&fs_info->fs_roots_radix_lock); |
31153d81 | 822 | |
e02119d5 | 823 | btrfs_free_log(trans, root); |
5d4f98a2 | 824 | btrfs_update_reloc_root(trans, root); |
d68fc57b | 825 | btrfs_orphan_commit_root(trans, root); |
bcc63abb | 826 | |
82d5902d LZ |
827 | btrfs_save_ino_cache(root, trans); |
828 | ||
978d910d | 829 | if (root->commit_root != root->node) { |
581bb050 | 830 | mutex_lock(&root->fs_commit_mutex); |
817d52f8 | 831 | switch_commit_root(root); |
581bb050 LZ |
832 | btrfs_unpin_free_ino(root); |
833 | mutex_unlock(&root->fs_commit_mutex); | |
834 | ||
978d910d YZ |
835 | btrfs_set_root_node(&root->root_item, |
836 | root->node); | |
837 | } | |
5d4f98a2 | 838 | |
5d4f98a2 | 839 | err = btrfs_update_root(trans, fs_info->tree_root, |
0f7d52f4 CM |
840 | &root->root_key, |
841 | &root->root_item); | |
a4abeea4 | 842 | spin_lock(&fs_info->fs_roots_radix_lock); |
54aa1f4d CM |
843 | if (err) |
844 | break; | |
0f7d52f4 CM |
845 | } |
846 | } | |
a4abeea4 | 847 | spin_unlock(&fs_info->fs_roots_radix_lock); |
54aa1f4d | 848 | return err; |
0f7d52f4 CM |
849 | } |
850 | ||
d352ac68 CM |
851 | /* |
852 | * defrag a given btree. If cacheonly == 1, this won't read from the disk, | |
853 | * otherwise every leaf in the btree is read and defragged. | |
854 | */ | |
e9d0b13b CM |
855 | int btrfs_defrag_root(struct btrfs_root *root, int cacheonly) |
856 | { | |
857 | struct btrfs_fs_info *info = root->fs_info; | |
e9d0b13b | 858 | struct btrfs_trans_handle *trans; |
8929ecfa | 859 | int ret; |
d3c2fdcf | 860 | unsigned long nr; |
e9d0b13b | 861 | |
8929ecfa | 862 | if (xchg(&root->defrag_running, 1)) |
e9d0b13b | 863 | return 0; |
8929ecfa | 864 | |
6b80053d | 865 | while (1) { |
8929ecfa YZ |
866 | trans = btrfs_start_transaction(root, 0); |
867 | if (IS_ERR(trans)) | |
868 | return PTR_ERR(trans); | |
869 | ||
e9d0b13b | 870 | ret = btrfs_defrag_leaves(trans, root, cacheonly); |
8929ecfa | 871 | |
d3c2fdcf | 872 | nr = trans->blocks_used; |
e9d0b13b | 873 | btrfs_end_transaction(trans, root); |
d3c2fdcf | 874 | btrfs_btree_balance_dirty(info->tree_root, nr); |
e9d0b13b CM |
875 | cond_resched(); |
876 | ||
7841cb28 | 877 | if (btrfs_fs_closing(root->fs_info) || ret != -EAGAIN) |
e9d0b13b CM |
878 | break; |
879 | } | |
880 | root->defrag_running = 0; | |
8929ecfa | 881 | return ret; |
e9d0b13b CM |
882 | } |
883 | ||
d352ac68 CM |
884 | /* |
885 | * new snapshots need to be created at a very specific time in the | |
886 | * transaction commit. This does the actual creation | |
887 | */ | |
80b6794d | 888 | static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, |
3063d29f CM |
889 | struct btrfs_fs_info *fs_info, |
890 | struct btrfs_pending_snapshot *pending) | |
891 | { | |
892 | struct btrfs_key key; | |
80b6794d | 893 | struct btrfs_root_item *new_root_item; |
3063d29f CM |
894 | struct btrfs_root *tree_root = fs_info->tree_root; |
895 | struct btrfs_root *root = pending->root; | |
6bdb72de SW |
896 | struct btrfs_root *parent_root; |
897 | struct inode *parent_inode; | |
6a912213 | 898 | struct dentry *parent; |
a22285a6 | 899 | struct dentry *dentry; |
3063d29f | 900 | struct extent_buffer *tmp; |
925baedd | 901 | struct extent_buffer *old; |
3063d29f | 902 | int ret; |
d68fc57b | 903 | u64 to_reserve = 0; |
6bdb72de | 904 | u64 index = 0; |
a22285a6 | 905 | u64 objectid; |
b83cc969 | 906 | u64 root_flags; |
3063d29f | 907 | |
80b6794d CM |
908 | new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS); |
909 | if (!new_root_item) { | |
a22285a6 | 910 | pending->error = -ENOMEM; |
80b6794d CM |
911 | goto fail; |
912 | } | |
a22285a6 | 913 | |
581bb050 | 914 | ret = btrfs_find_free_objectid(tree_root, &objectid); |
a22285a6 YZ |
915 | if (ret) { |
916 | pending->error = ret; | |
3063d29f | 917 | goto fail; |
a22285a6 | 918 | } |
3063d29f | 919 | |
3fd0a558 | 920 | btrfs_reloc_pre_snapshot(trans, pending, &to_reserve); |
d68fc57b YZ |
921 | btrfs_orphan_pre_snapshot(trans, pending, &to_reserve); |
922 | ||
923 | if (to_reserve > 0) { | |
924 | ret = btrfs_block_rsv_add(trans, root, &pending->block_rsv, | |
8bb8ab2e | 925 | to_reserve); |
d68fc57b YZ |
926 | if (ret) { |
927 | pending->error = ret; | |
928 | goto fail; | |
929 | } | |
930 | } | |
931 | ||
3063d29f | 932 | key.objectid = objectid; |
a22285a6 YZ |
933 | key.offset = (u64)-1; |
934 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3063d29f | 935 | |
a22285a6 | 936 | trans->block_rsv = &pending->block_rsv; |
3de4586c | 937 | |
a22285a6 | 938 | dentry = pending->dentry; |
6a912213 JB |
939 | parent = dget_parent(dentry); |
940 | parent_inode = parent->d_inode; | |
a22285a6 | 941 | parent_root = BTRFS_I(parent_inode)->root; |
7585717f | 942 | record_root_in_trans(trans, parent_root); |
a22285a6 | 943 | |
3063d29f CM |
944 | /* |
945 | * insert the directory item | |
946 | */ | |
3de4586c | 947 | ret = btrfs_set_inode_index(parent_inode, &index); |
6bdb72de | 948 | BUG_ON(ret); |
0660b5af | 949 | ret = btrfs_insert_dir_item(trans, parent_root, |
a22285a6 | 950 | dentry->d_name.name, dentry->d_name.len, |
16cdcec7 | 951 | parent_inode, &key, |
a22285a6 | 952 | BTRFS_FT_DIR, index); |
6bdb72de | 953 | BUG_ON(ret); |
0660b5af | 954 | |
a22285a6 YZ |
955 | btrfs_i_size_write(parent_inode, parent_inode->i_size + |
956 | dentry->d_name.len * 2); | |
52c26179 YZ |
957 | ret = btrfs_update_inode(trans, parent_root, parent_inode); |
958 | BUG_ON(ret); | |
959 | ||
7585717f | 960 | record_root_in_trans(trans, root); |
6bdb72de SW |
961 | btrfs_set_root_last_snapshot(&root->root_item, trans->transid); |
962 | memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); | |
08fe4db1 | 963 | btrfs_check_and_init_root_item(new_root_item); |
6bdb72de | 964 | |
b83cc969 LZ |
965 | root_flags = btrfs_root_flags(new_root_item); |
966 | if (pending->readonly) | |
967 | root_flags |= BTRFS_ROOT_SUBVOL_RDONLY; | |
968 | else | |
969 | root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY; | |
970 | btrfs_set_root_flags(new_root_item, root_flags); | |
971 | ||
6bdb72de SW |
972 | old = btrfs_lock_root_node(root); |
973 | btrfs_cow_block(trans, root, old, NULL, 0, &old); | |
974 | btrfs_set_lock_blocking(old); | |
975 | ||
976 | btrfs_copy_root(trans, root, old, &tmp, objectid); | |
977 | btrfs_tree_unlock(old); | |
978 | free_extent_buffer(old); | |
979 | ||
980 | btrfs_set_root_node(new_root_item, tmp); | |
a22285a6 YZ |
981 | /* record when the snapshot was created in key.offset */ |
982 | key.offset = trans->transid; | |
983 | ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); | |
6bdb72de SW |
984 | btrfs_tree_unlock(tmp); |
985 | free_extent_buffer(tmp); | |
a22285a6 | 986 | BUG_ON(ret); |
6bdb72de | 987 | |
a22285a6 YZ |
988 | /* |
989 | * insert root back/forward references | |
990 | */ | |
991 | ret = btrfs_add_root_ref(trans, tree_root, objectid, | |
0660b5af | 992 | parent_root->root_key.objectid, |
33345d01 | 993 | btrfs_ino(parent_inode), index, |
a22285a6 | 994 | dentry->d_name.name, dentry->d_name.len); |
0660b5af | 995 | BUG_ON(ret); |
6a912213 | 996 | dput(parent); |
0660b5af | 997 | |
a22285a6 YZ |
998 | key.offset = (u64)-1; |
999 | pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key); | |
1000 | BUG_ON(IS_ERR(pending->snap)); | |
d68fc57b | 1001 | |
3fd0a558 | 1002 | btrfs_reloc_post_snapshot(trans, pending); |
d68fc57b | 1003 | btrfs_orphan_post_snapshot(trans, pending); |
3063d29f | 1004 | fail: |
6bdb72de | 1005 | kfree(new_root_item); |
a22285a6 YZ |
1006 | btrfs_block_rsv_release(root, &pending->block_rsv, (u64)-1); |
1007 | return 0; | |
3063d29f CM |
1008 | } |
1009 | ||
d352ac68 CM |
1010 | /* |
1011 | * create all the snapshots we've scheduled for creation | |
1012 | */ | |
80b6794d CM |
1013 | static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans, |
1014 | struct btrfs_fs_info *fs_info) | |
3de4586c CM |
1015 | { |
1016 | struct btrfs_pending_snapshot *pending; | |
1017 | struct list_head *head = &trans->transaction->pending_snapshots; | |
3de4586c CM |
1018 | int ret; |
1019 | ||
c6e30871 | 1020 | list_for_each_entry(pending, head, list) { |
16cdcec7 MX |
1021 | /* |
1022 | * We must deal with the delayed items before creating | |
1023 | * snapshots, or we will create a snapthot with inconsistent | |
1024 | * information. | |
1025 | */ | |
1026 | ret = btrfs_run_delayed_items(trans, fs_info->fs_root); | |
1027 | BUG_ON(ret); | |
1028 | ||
3de4586c CM |
1029 | ret = create_pending_snapshot(trans, fs_info, pending); |
1030 | BUG_ON(ret); | |
1031 | } | |
1032 | return 0; | |
1033 | } | |
1034 | ||
5d4f98a2 YZ |
1035 | static void update_super_roots(struct btrfs_root *root) |
1036 | { | |
1037 | struct btrfs_root_item *root_item; | |
1038 | struct btrfs_super_block *super; | |
1039 | ||
1040 | super = &root->fs_info->super_copy; | |
1041 | ||
1042 | root_item = &root->fs_info->chunk_root->root_item; | |
1043 | super->chunk_root = root_item->bytenr; | |
1044 | super->chunk_root_generation = root_item->generation; | |
1045 | super->chunk_root_level = root_item->level; | |
1046 | ||
1047 | root_item = &root->fs_info->tree_root->root_item; | |
1048 | super->root = root_item->bytenr; | |
1049 | super->generation = root_item->generation; | |
1050 | super->root_level = root_item->level; | |
0af3d00b JB |
1051 | if (super->cache_generation != 0 || btrfs_test_opt(root, SPACE_CACHE)) |
1052 | super->cache_generation = root_item->generation; | |
5d4f98a2 YZ |
1053 | } |
1054 | ||
f36f3042 CM |
1055 | int btrfs_transaction_in_commit(struct btrfs_fs_info *info) |
1056 | { | |
1057 | int ret = 0; | |
a4abeea4 | 1058 | spin_lock(&info->trans_lock); |
f36f3042 CM |
1059 | if (info->running_transaction) |
1060 | ret = info->running_transaction->in_commit; | |
a4abeea4 | 1061 | spin_unlock(&info->trans_lock); |
f36f3042 CM |
1062 | return ret; |
1063 | } | |
1064 | ||
8929ecfa YZ |
1065 | int btrfs_transaction_blocked(struct btrfs_fs_info *info) |
1066 | { | |
1067 | int ret = 0; | |
a4abeea4 | 1068 | spin_lock(&info->trans_lock); |
8929ecfa YZ |
1069 | if (info->running_transaction) |
1070 | ret = info->running_transaction->blocked; | |
a4abeea4 | 1071 | spin_unlock(&info->trans_lock); |
8929ecfa YZ |
1072 | return ret; |
1073 | } | |
1074 | ||
bb9c12c9 SW |
1075 | /* |
1076 | * wait for the current transaction commit to start and block subsequent | |
1077 | * transaction joins | |
1078 | */ | |
1079 | static void wait_current_trans_commit_start(struct btrfs_root *root, | |
1080 | struct btrfs_transaction *trans) | |
1081 | { | |
1082 | DEFINE_WAIT(wait); | |
1083 | ||
1084 | if (trans->in_commit) | |
1085 | return; | |
1086 | ||
1087 | while (1) { | |
1088 | prepare_to_wait(&root->fs_info->transaction_blocked_wait, &wait, | |
1089 | TASK_UNINTERRUPTIBLE); | |
1090 | if (trans->in_commit) { | |
1091 | finish_wait(&root->fs_info->transaction_blocked_wait, | |
1092 | &wait); | |
1093 | break; | |
1094 | } | |
bb9c12c9 | 1095 | schedule(); |
bb9c12c9 SW |
1096 | finish_wait(&root->fs_info->transaction_blocked_wait, &wait); |
1097 | } | |
1098 | } | |
1099 | ||
1100 | /* | |
1101 | * wait for the current transaction to start and then become unblocked. | |
1102 | * caller holds ref. | |
1103 | */ | |
1104 | static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root, | |
1105 | struct btrfs_transaction *trans) | |
1106 | { | |
1107 | DEFINE_WAIT(wait); | |
1108 | ||
1109 | if (trans->commit_done || (trans->in_commit && !trans->blocked)) | |
1110 | return; | |
1111 | ||
1112 | while (1) { | |
1113 | prepare_to_wait(&root->fs_info->transaction_wait, &wait, | |
1114 | TASK_UNINTERRUPTIBLE); | |
1115 | if (trans->commit_done || | |
1116 | (trans->in_commit && !trans->blocked)) { | |
1117 | finish_wait(&root->fs_info->transaction_wait, | |
1118 | &wait); | |
1119 | break; | |
1120 | } | |
bb9c12c9 | 1121 | schedule(); |
bb9c12c9 SW |
1122 | finish_wait(&root->fs_info->transaction_wait, |
1123 | &wait); | |
1124 | } | |
1125 | } | |
1126 | ||
1127 | /* | |
1128 | * commit transactions asynchronously. once btrfs_commit_transaction_async | |
1129 | * returns, any subsequent transaction will not be allowed to join. | |
1130 | */ | |
1131 | struct btrfs_async_commit { | |
1132 | struct btrfs_trans_handle *newtrans; | |
1133 | struct btrfs_root *root; | |
1134 | struct delayed_work work; | |
1135 | }; | |
1136 | ||
1137 | static void do_async_commit(struct work_struct *work) | |
1138 | { | |
1139 | struct btrfs_async_commit *ac = | |
1140 | container_of(work, struct btrfs_async_commit, work.work); | |
1141 | ||
1142 | btrfs_commit_transaction(ac->newtrans, ac->root); | |
1143 | kfree(ac); | |
1144 | } | |
1145 | ||
1146 | int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans, | |
1147 | struct btrfs_root *root, | |
1148 | int wait_for_unblock) | |
1149 | { | |
1150 | struct btrfs_async_commit *ac; | |
1151 | struct btrfs_transaction *cur_trans; | |
1152 | ||
1153 | ac = kmalloc(sizeof(*ac), GFP_NOFS); | |
db5b493a TI |
1154 | if (!ac) |
1155 | return -ENOMEM; | |
bb9c12c9 SW |
1156 | |
1157 | INIT_DELAYED_WORK(&ac->work, do_async_commit); | |
1158 | ac->root = root; | |
7a7eaa40 | 1159 | ac->newtrans = btrfs_join_transaction(root); |
3612b495 TI |
1160 | if (IS_ERR(ac->newtrans)) { |
1161 | int err = PTR_ERR(ac->newtrans); | |
1162 | kfree(ac); | |
1163 | return err; | |
1164 | } | |
bb9c12c9 SW |
1165 | |
1166 | /* take transaction reference */ | |
bb9c12c9 | 1167 | cur_trans = trans->transaction; |
13c5a93e | 1168 | atomic_inc(&cur_trans->use_count); |
bb9c12c9 SW |
1169 | |
1170 | btrfs_end_transaction(trans, root); | |
1171 | schedule_delayed_work(&ac->work, 0); | |
1172 | ||
1173 | /* wait for transaction to start and unblock */ | |
bb9c12c9 SW |
1174 | if (wait_for_unblock) |
1175 | wait_current_trans_commit_start_and_unblock(root, cur_trans); | |
1176 | else | |
1177 | wait_current_trans_commit_start(root, cur_trans); | |
bb9c12c9 | 1178 | |
38e88054 SW |
1179 | if (current->journal_info == trans) |
1180 | current->journal_info = NULL; | |
1181 | ||
1182 | put_transaction(cur_trans); | |
bb9c12c9 SW |
1183 | return 0; |
1184 | } | |
1185 | ||
1186 | /* | |
1187 | * btrfs_transaction state sequence: | |
1188 | * in_commit = 0, blocked = 0 (initial) | |
1189 | * in_commit = 1, blocked = 1 | |
1190 | * blocked = 0 | |
1191 | * commit_done = 1 | |
1192 | */ | |
79154b1b CM |
1193 | int btrfs_commit_transaction(struct btrfs_trans_handle *trans, |
1194 | struct btrfs_root *root) | |
1195 | { | |
15ee9bc7 | 1196 | unsigned long joined = 0; |
79154b1b | 1197 | struct btrfs_transaction *cur_trans; |
8fd17795 | 1198 | struct btrfs_transaction *prev_trans = NULL; |
79154b1b | 1199 | DEFINE_WAIT(wait); |
15ee9bc7 | 1200 | int ret; |
89573b9c CM |
1201 | int should_grow = 0; |
1202 | unsigned long now = get_seconds(); | |
dccae999 | 1203 | int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT); |
79154b1b | 1204 | |
5a3f23d5 CM |
1205 | btrfs_run_ordered_operations(root, 0); |
1206 | ||
56bec294 CM |
1207 | /* make a pass through all the delayed refs we have so far |
1208 | * any runnings procs may add more while we are here | |
1209 | */ | |
1210 | ret = btrfs_run_delayed_refs(trans, root, 0); | |
1211 | BUG_ON(ret); | |
1212 | ||
a22285a6 YZ |
1213 | btrfs_trans_release_metadata(trans, root); |
1214 | ||
b7ec40d7 | 1215 | cur_trans = trans->transaction; |
56bec294 CM |
1216 | /* |
1217 | * set the flushing flag so procs in this transaction have to | |
1218 | * start sending their work down. | |
1219 | */ | |
b7ec40d7 | 1220 | cur_trans->delayed_refs.flushing = 1; |
56bec294 | 1221 | |
c3e69d58 | 1222 | ret = btrfs_run_delayed_refs(trans, root, 0); |
56bec294 CM |
1223 | BUG_ON(ret); |
1224 | ||
a4abeea4 | 1225 | spin_lock(&cur_trans->commit_lock); |
b7ec40d7 | 1226 | if (cur_trans->in_commit) { |
a4abeea4 | 1227 | spin_unlock(&cur_trans->commit_lock); |
13c5a93e | 1228 | atomic_inc(&cur_trans->use_count); |
79154b1b | 1229 | btrfs_end_transaction(trans, root); |
ccd467d6 | 1230 | |
79154b1b CM |
1231 | ret = wait_for_commit(root, cur_trans); |
1232 | BUG_ON(ret); | |
15ee9bc7 | 1233 | |
79154b1b | 1234 | put_transaction(cur_trans); |
15ee9bc7 | 1235 | |
79154b1b CM |
1236 | return 0; |
1237 | } | |
4313b399 | 1238 | |
2c90e5d6 | 1239 | trans->transaction->in_commit = 1; |
f9295749 | 1240 | trans->transaction->blocked = 1; |
a4abeea4 | 1241 | spin_unlock(&cur_trans->commit_lock); |
bb9c12c9 SW |
1242 | wake_up(&root->fs_info->transaction_blocked_wait); |
1243 | ||
a4abeea4 | 1244 | spin_lock(&root->fs_info->trans_lock); |
ccd467d6 CM |
1245 | if (cur_trans->list.prev != &root->fs_info->trans_list) { |
1246 | prev_trans = list_entry(cur_trans->list.prev, | |
1247 | struct btrfs_transaction, list); | |
1248 | if (!prev_trans->commit_done) { | |
13c5a93e | 1249 | atomic_inc(&prev_trans->use_count); |
a4abeea4 | 1250 | spin_unlock(&root->fs_info->trans_lock); |
ccd467d6 CM |
1251 | |
1252 | wait_for_commit(root, prev_trans); | |
ccd467d6 | 1253 | |
15ee9bc7 | 1254 | put_transaction(prev_trans); |
a4abeea4 JB |
1255 | } else { |
1256 | spin_unlock(&root->fs_info->trans_lock); | |
ccd467d6 | 1257 | } |
a4abeea4 JB |
1258 | } else { |
1259 | spin_unlock(&root->fs_info->trans_lock); | |
ccd467d6 | 1260 | } |
15ee9bc7 | 1261 | |
89573b9c CM |
1262 | if (now < cur_trans->start_time || now - cur_trans->start_time < 1) |
1263 | should_grow = 1; | |
1264 | ||
15ee9bc7 | 1265 | do { |
7ea394f1 | 1266 | int snap_pending = 0; |
a4abeea4 | 1267 | |
15ee9bc7 | 1268 | joined = cur_trans->num_joined; |
7ea394f1 YZ |
1269 | if (!list_empty(&trans->transaction->pending_snapshots)) |
1270 | snap_pending = 1; | |
1271 | ||
2c90e5d6 | 1272 | WARN_ON(cur_trans != trans->transaction); |
15ee9bc7 | 1273 | |
0bdb1db2 | 1274 | if (flush_on_commit || snap_pending) { |
24bbcf04 YZ |
1275 | btrfs_start_delalloc_inodes(root, 1); |
1276 | ret = btrfs_wait_ordered_extents(root, 0, 1); | |
ebecd3d9 | 1277 | BUG_ON(ret); |
7ea394f1 YZ |
1278 | } |
1279 | ||
16cdcec7 MX |
1280 | ret = btrfs_run_delayed_items(trans, root); |
1281 | BUG_ON(ret); | |
1282 | ||
5a3f23d5 CM |
1283 | /* |
1284 | * rename don't use btrfs_join_transaction, so, once we | |
1285 | * set the transaction to blocked above, we aren't going | |
1286 | * to get any new ordered operations. We can safely run | |
1287 | * it here and no for sure that nothing new will be added | |
1288 | * to the list | |
1289 | */ | |
1290 | btrfs_run_ordered_operations(root, 1); | |
1291 | ||
ed3b3d31 CM |
1292 | prepare_to_wait(&cur_trans->writer_wait, &wait, |
1293 | TASK_UNINTERRUPTIBLE); | |
1294 | ||
13c5a93e | 1295 | if (atomic_read(&cur_trans->num_writers) > 1) |
99d16cbc SW |
1296 | schedule_timeout(MAX_SCHEDULE_TIMEOUT); |
1297 | else if (should_grow) | |
1298 | schedule_timeout(1); | |
15ee9bc7 | 1299 | |
15ee9bc7 | 1300 | finish_wait(&cur_trans->writer_wait, &wait); |
a4abeea4 JB |
1301 | spin_lock(&root->fs_info->trans_lock); |
1302 | root->fs_info->trans_no_join = 1; | |
1303 | spin_unlock(&root->fs_info->trans_lock); | |
13c5a93e | 1304 | } while (atomic_read(&cur_trans->num_writers) > 1 || |
89573b9c | 1305 | (should_grow && cur_trans->num_joined != joined)); |
15ee9bc7 | 1306 | |
7585717f CM |
1307 | /* |
1308 | * the reloc mutex makes sure that we stop | |
1309 | * the balancing code from coming in and moving | |
1310 | * extents around in the middle of the commit | |
1311 | */ | |
1312 | mutex_lock(&root->fs_info->reloc_mutex); | |
1313 | ||
3063d29f CM |
1314 | ret = create_pending_snapshots(trans, root->fs_info); |
1315 | BUG_ON(ret); | |
1316 | ||
16cdcec7 MX |
1317 | ret = btrfs_run_delayed_items(trans, root); |
1318 | BUG_ON(ret); | |
1319 | ||
56bec294 CM |
1320 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
1321 | BUG_ON(ret); | |
1322 | ||
2c90e5d6 | 1323 | WARN_ON(cur_trans != trans->transaction); |
dc17ff8f | 1324 | |
a2de733c | 1325 | btrfs_scrub_pause(root); |
e02119d5 CM |
1326 | /* btrfs_commit_tree_roots is responsible for getting the |
1327 | * various roots consistent with each other. Every pointer | |
1328 | * in the tree of tree roots has to point to the most up to date | |
1329 | * root for every subvolume and other tree. So, we have to keep | |
1330 | * the tree logging code from jumping in and changing any | |
1331 | * of the trees. | |
1332 | * | |
1333 | * At this point in the commit, there can't be any tree-log | |
1334 | * writers, but a little lower down we drop the trans mutex | |
1335 | * and let new people in. By holding the tree_log_mutex | |
1336 | * from now until after the super is written, we avoid races | |
1337 | * with the tree-log code. | |
1338 | */ | |
1339 | mutex_lock(&root->fs_info->tree_log_mutex); | |
1340 | ||
5d4f98a2 | 1341 | ret = commit_fs_roots(trans, root); |
54aa1f4d CM |
1342 | BUG_ON(ret); |
1343 | ||
5d4f98a2 | 1344 | /* commit_fs_roots gets rid of all the tree log roots, it is now |
e02119d5 CM |
1345 | * safe to free the root of tree log roots |
1346 | */ | |
1347 | btrfs_free_log_root_tree(trans, root->fs_info); | |
1348 | ||
5d4f98a2 | 1349 | ret = commit_cowonly_roots(trans, root); |
79154b1b | 1350 | BUG_ON(ret); |
54aa1f4d | 1351 | |
11833d66 YZ |
1352 | btrfs_prepare_extent_commit(trans, root); |
1353 | ||
78fae27e | 1354 | cur_trans = root->fs_info->running_transaction; |
5d4f98a2 YZ |
1355 | |
1356 | btrfs_set_root_node(&root->fs_info->tree_root->root_item, | |
1357 | root->fs_info->tree_root->node); | |
817d52f8 | 1358 | switch_commit_root(root->fs_info->tree_root); |
5d4f98a2 YZ |
1359 | |
1360 | btrfs_set_root_node(&root->fs_info->chunk_root->root_item, | |
1361 | root->fs_info->chunk_root->node); | |
817d52f8 | 1362 | switch_commit_root(root->fs_info->chunk_root); |
5d4f98a2 YZ |
1363 | |
1364 | update_super_roots(root); | |
e02119d5 CM |
1365 | |
1366 | if (!root->fs_info->log_root_recovering) { | |
1367 | btrfs_set_super_log_root(&root->fs_info->super_copy, 0); | |
1368 | btrfs_set_super_log_root_level(&root->fs_info->super_copy, 0); | |
1369 | } | |
1370 | ||
a061fc8d CM |
1371 | memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy, |
1372 | sizeof(root->fs_info->super_copy)); | |
ccd467d6 | 1373 | |
f9295749 | 1374 | trans->transaction->blocked = 0; |
a4abeea4 JB |
1375 | spin_lock(&root->fs_info->trans_lock); |
1376 | root->fs_info->running_transaction = NULL; | |
1377 | root->fs_info->trans_no_join = 0; | |
1378 | spin_unlock(&root->fs_info->trans_lock); | |
7585717f | 1379 | mutex_unlock(&root->fs_info->reloc_mutex); |
b7ec40d7 | 1380 | |
f9295749 | 1381 | wake_up(&root->fs_info->transaction_wait); |
e6dcd2dc | 1382 | |
79154b1b CM |
1383 | ret = btrfs_write_and_wait_transaction(trans, root); |
1384 | BUG_ON(ret); | |
a512bbf8 | 1385 | write_ctree_super(trans, root, 0); |
4313b399 | 1386 | |
e02119d5 CM |
1387 | /* |
1388 | * the super is written, we can safely allow the tree-loggers | |
1389 | * to go about their business | |
1390 | */ | |
1391 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
1392 | ||
11833d66 | 1393 | btrfs_finish_extent_commit(trans, root); |
4313b399 | 1394 | |
2c90e5d6 | 1395 | cur_trans->commit_done = 1; |
b7ec40d7 | 1396 | |
15ee9bc7 | 1397 | root->fs_info->last_trans_committed = cur_trans->transid; |
817d52f8 | 1398 | |
2c90e5d6 | 1399 | wake_up(&cur_trans->commit_wait); |
3de4586c | 1400 | |
a4abeea4 | 1401 | spin_lock(&root->fs_info->trans_lock); |
13c5a93e | 1402 | list_del_init(&cur_trans->list); |
a4abeea4 JB |
1403 | spin_unlock(&root->fs_info->trans_lock); |
1404 | ||
78fae27e | 1405 | put_transaction(cur_trans); |
79154b1b | 1406 | put_transaction(cur_trans); |
58176a96 | 1407 | |
1abe9b8a | 1408 | trace_btrfs_transaction_commit(root); |
1409 | ||
a2de733c AJ |
1410 | btrfs_scrub_continue(root); |
1411 | ||
9ed74f2d JB |
1412 | if (current->journal_info == trans) |
1413 | current->journal_info = NULL; | |
1414 | ||
2c90e5d6 | 1415 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
24bbcf04 YZ |
1416 | |
1417 | if (current != root->fs_info->transaction_kthread) | |
1418 | btrfs_run_delayed_iputs(root); | |
1419 | ||
79154b1b CM |
1420 | return ret; |
1421 | } | |
1422 | ||
d352ac68 CM |
1423 | /* |
1424 | * interface function to delete all the snapshots we have scheduled for deletion | |
1425 | */ | |
e9d0b13b CM |
1426 | int btrfs_clean_old_snapshots(struct btrfs_root *root) |
1427 | { | |
5d4f98a2 YZ |
1428 | LIST_HEAD(list); |
1429 | struct btrfs_fs_info *fs_info = root->fs_info; | |
1430 | ||
a4abeea4 | 1431 | spin_lock(&fs_info->trans_lock); |
5d4f98a2 | 1432 | list_splice_init(&fs_info->dead_roots, &list); |
a4abeea4 | 1433 | spin_unlock(&fs_info->trans_lock); |
e9d0b13b | 1434 | |
5d4f98a2 YZ |
1435 | while (!list_empty(&list)) { |
1436 | root = list_entry(list.next, struct btrfs_root, root_list); | |
76dda93c YZ |
1437 | list_del(&root->root_list); |
1438 | ||
16cdcec7 MX |
1439 | btrfs_kill_all_delayed_nodes(root); |
1440 | ||
76dda93c YZ |
1441 | if (btrfs_header_backref_rev(root->node) < |
1442 | BTRFS_MIXED_BACKREF_REV) | |
3fd0a558 | 1443 | btrfs_drop_snapshot(root, NULL, 0); |
76dda93c | 1444 | else |
3fd0a558 | 1445 | btrfs_drop_snapshot(root, NULL, 1); |
e9d0b13b CM |
1446 | } |
1447 | return 0; | |
1448 | } |