]>
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) { |
13c5a93e | 219 | atomic_inc(&cur_trans->use_count); |
a4abeea4 | 220 | spin_unlock(&root->fs_info->trans_lock); |
72d63ed6 LZ |
221 | |
222 | wait_event(root->fs_info->transaction_wait, | |
223 | !cur_trans->blocked); | |
f9295749 | 224 | put_transaction(cur_trans); |
a4abeea4 JB |
225 | } else { |
226 | spin_unlock(&root->fs_info->trans_lock); | |
f9295749 | 227 | } |
37d1aeee CM |
228 | } |
229 | ||
249ac1e5 JB |
230 | enum btrfs_trans_type { |
231 | TRANS_START, | |
232 | TRANS_JOIN, | |
233 | TRANS_USERSPACE, | |
0af3d00b | 234 | TRANS_JOIN_NOLOCK, |
249ac1e5 JB |
235 | }; |
236 | ||
a22285a6 YZ |
237 | static int may_wait_transaction(struct btrfs_root *root, int type) |
238 | { | |
a4abeea4 JB |
239 | if (root->fs_info->log_root_recovering) |
240 | return 0; | |
241 | ||
242 | if (type == TRANS_USERSPACE) | |
243 | return 1; | |
244 | ||
245 | if (type == TRANS_START && | |
246 | !atomic_read(&root->fs_info->open_ioctl_trans)) | |
a22285a6 | 247 | return 1; |
a4abeea4 | 248 | |
a22285a6 YZ |
249 | return 0; |
250 | } | |
251 | ||
e02119d5 | 252 | static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root, |
a22285a6 | 253 | u64 num_items, int type) |
37d1aeee | 254 | { |
a22285a6 YZ |
255 | struct btrfs_trans_handle *h; |
256 | struct btrfs_transaction *cur_trans; | |
b5009945 | 257 | u64 num_bytes = 0; |
37d1aeee | 258 | int ret; |
acce952b | 259 | |
260 | if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) | |
261 | return ERR_PTR(-EROFS); | |
2a1eb461 JB |
262 | |
263 | if (current->journal_info) { | |
264 | WARN_ON(type != TRANS_JOIN && type != TRANS_JOIN_NOLOCK); | |
265 | h = current->journal_info; | |
266 | h->use_count++; | |
267 | h->orig_rsv = h->block_rsv; | |
268 | h->block_rsv = NULL; | |
269 | goto got_it; | |
270 | } | |
b5009945 JB |
271 | |
272 | /* | |
273 | * Do the reservation before we join the transaction so we can do all | |
274 | * the appropriate flushing if need be. | |
275 | */ | |
276 | if (num_items > 0 && root != root->fs_info->chunk_root) { | |
277 | num_bytes = btrfs_calc_trans_metadata_size(root, num_items); | |
278 | ret = btrfs_block_rsv_add(NULL, root, | |
279 | &root->fs_info->trans_block_rsv, | |
280 | num_bytes); | |
281 | if (ret) | |
282 | return ERR_PTR(ret); | |
283 | } | |
a22285a6 YZ |
284 | again: |
285 | h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS); | |
286 | if (!h) | |
287 | return ERR_PTR(-ENOMEM); | |
37d1aeee | 288 | |
a22285a6 | 289 | if (may_wait_transaction(root, type)) |
37d1aeee | 290 | wait_current_trans(root); |
a22285a6 | 291 | |
a4abeea4 JB |
292 | do { |
293 | ret = join_transaction(root, type == TRANS_JOIN_NOLOCK); | |
294 | if (ret == -EBUSY) | |
295 | wait_current_trans(root); | |
296 | } while (ret == -EBUSY); | |
297 | ||
db5b493a | 298 | if (ret < 0) { |
6e8df2ae | 299 | kmem_cache_free(btrfs_trans_handle_cachep, h); |
db5b493a TI |
300 | return ERR_PTR(ret); |
301 | } | |
0f7d52f4 | 302 | |
a22285a6 | 303 | cur_trans = root->fs_info->running_transaction; |
a22285a6 YZ |
304 | |
305 | h->transid = cur_trans->transid; | |
306 | h->transaction = cur_trans; | |
79154b1b | 307 | h->blocks_used = 0; |
a22285a6 | 308 | h->bytes_reserved = 0; |
56bec294 | 309 | h->delayed_ref_updates = 0; |
2a1eb461 | 310 | h->use_count = 1; |
f0486c68 | 311 | h->block_rsv = NULL; |
2a1eb461 | 312 | h->orig_rsv = NULL; |
b7ec40d7 | 313 | |
a22285a6 YZ |
314 | smp_mb(); |
315 | if (cur_trans->blocked && may_wait_transaction(root, type)) { | |
316 | btrfs_commit_transaction(h, root); | |
317 | goto again; | |
318 | } | |
319 | ||
b5009945 JB |
320 | if (num_bytes) { |
321 | h->block_rsv = &root->fs_info->trans_block_rsv; | |
322 | h->bytes_reserved = num_bytes; | |
a22285a6 | 323 | } |
9ed74f2d | 324 | |
2a1eb461 | 325 | got_it: |
a4abeea4 | 326 | btrfs_record_root_in_trans(h, root); |
a22285a6 YZ |
327 | |
328 | if (!current->journal_info && type != TRANS_USERSPACE) | |
329 | current->journal_info = h; | |
79154b1b CM |
330 | return h; |
331 | } | |
332 | ||
f9295749 | 333 | struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, |
a22285a6 | 334 | int num_items) |
f9295749 | 335 | { |
a22285a6 | 336 | return start_transaction(root, num_items, TRANS_START); |
f9295749 | 337 | } |
7a7eaa40 | 338 | struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root) |
f9295749 | 339 | { |
a22285a6 | 340 | return start_transaction(root, 0, TRANS_JOIN); |
f9295749 CM |
341 | } |
342 | ||
7a7eaa40 | 343 | struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root) |
0af3d00b JB |
344 | { |
345 | return start_transaction(root, 0, TRANS_JOIN_NOLOCK); | |
346 | } | |
347 | ||
7a7eaa40 | 348 | struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root) |
9ca9ee09 | 349 | { |
7a7eaa40 | 350 | return start_transaction(root, 0, TRANS_USERSPACE); |
9ca9ee09 SW |
351 | } |
352 | ||
d352ac68 | 353 | /* wait for a transaction commit to be fully complete */ |
b9c8300c | 354 | static noinline void wait_for_commit(struct btrfs_root *root, |
89ce8a63 CM |
355 | struct btrfs_transaction *commit) |
356 | { | |
72d63ed6 | 357 | wait_event(commit->commit_wait, commit->commit_done); |
89ce8a63 CM |
358 | } |
359 | ||
46204592 SW |
360 | int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid) |
361 | { | |
362 | struct btrfs_transaction *cur_trans = NULL, *t; | |
363 | int ret; | |
364 | ||
46204592 SW |
365 | ret = 0; |
366 | if (transid) { | |
367 | if (transid <= root->fs_info->last_trans_committed) | |
a4abeea4 | 368 | goto out; |
46204592 SW |
369 | |
370 | /* find specified transaction */ | |
a4abeea4 | 371 | spin_lock(&root->fs_info->trans_lock); |
46204592 SW |
372 | list_for_each_entry(t, &root->fs_info->trans_list, list) { |
373 | if (t->transid == transid) { | |
374 | cur_trans = t; | |
a4abeea4 | 375 | atomic_inc(&cur_trans->use_count); |
46204592 SW |
376 | break; |
377 | } | |
378 | if (t->transid > transid) | |
379 | break; | |
380 | } | |
a4abeea4 | 381 | spin_unlock(&root->fs_info->trans_lock); |
46204592 SW |
382 | ret = -EINVAL; |
383 | if (!cur_trans) | |
a4abeea4 | 384 | goto out; /* bad transid */ |
46204592 SW |
385 | } else { |
386 | /* find newest transaction that is committing | committed */ | |
a4abeea4 | 387 | spin_lock(&root->fs_info->trans_lock); |
46204592 SW |
388 | list_for_each_entry_reverse(t, &root->fs_info->trans_list, |
389 | list) { | |
390 | if (t->in_commit) { | |
391 | if (t->commit_done) | |
3473f3c0 | 392 | break; |
46204592 | 393 | cur_trans = t; |
a4abeea4 | 394 | atomic_inc(&cur_trans->use_count); |
46204592 SW |
395 | break; |
396 | } | |
397 | } | |
a4abeea4 | 398 | spin_unlock(&root->fs_info->trans_lock); |
46204592 | 399 | if (!cur_trans) |
a4abeea4 | 400 | goto out; /* nothing committing|committed */ |
46204592 SW |
401 | } |
402 | ||
46204592 SW |
403 | wait_for_commit(root, cur_trans); |
404 | ||
46204592 SW |
405 | put_transaction(cur_trans); |
406 | ret = 0; | |
a4abeea4 | 407 | out: |
46204592 SW |
408 | return ret; |
409 | } | |
410 | ||
37d1aeee CM |
411 | void btrfs_throttle(struct btrfs_root *root) |
412 | { | |
a4abeea4 | 413 | if (!atomic_read(&root->fs_info->open_ioctl_trans)) |
9ca9ee09 | 414 | wait_current_trans(root); |
37d1aeee CM |
415 | } |
416 | ||
8929ecfa YZ |
417 | static int should_end_transaction(struct btrfs_trans_handle *trans, |
418 | struct btrfs_root *root) | |
419 | { | |
420 | int ret; | |
421 | ret = btrfs_block_rsv_check(trans, root, | |
482e6dc5 | 422 | &root->fs_info->global_block_rsv, 0, 5, 0); |
8929ecfa YZ |
423 | return ret ? 1 : 0; |
424 | } | |
425 | ||
426 | int btrfs_should_end_transaction(struct btrfs_trans_handle *trans, | |
427 | struct btrfs_root *root) | |
428 | { | |
429 | struct btrfs_transaction *cur_trans = trans->transaction; | |
430 | int updates; | |
431 | ||
a4abeea4 | 432 | smp_mb(); |
8929ecfa YZ |
433 | if (cur_trans->blocked || cur_trans->delayed_refs.flushing) |
434 | return 1; | |
435 | ||
436 | updates = trans->delayed_ref_updates; | |
437 | trans->delayed_ref_updates = 0; | |
438 | if (updates) | |
439 | btrfs_run_delayed_refs(trans, root, updates); | |
440 | ||
441 | return should_end_transaction(trans, root); | |
442 | } | |
443 | ||
89ce8a63 | 444 | static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, |
0af3d00b | 445 | struct btrfs_root *root, int throttle, int lock) |
79154b1b | 446 | { |
8929ecfa | 447 | struct btrfs_transaction *cur_trans = trans->transaction; |
ab78c84d | 448 | struct btrfs_fs_info *info = root->fs_info; |
c3e69d58 CM |
449 | int count = 0; |
450 | ||
2a1eb461 JB |
451 | if (--trans->use_count) { |
452 | trans->block_rsv = trans->orig_rsv; | |
453 | return 0; | |
454 | } | |
455 | ||
c3e69d58 CM |
456 | while (count < 4) { |
457 | unsigned long cur = trans->delayed_ref_updates; | |
458 | trans->delayed_ref_updates = 0; | |
459 | if (cur && | |
460 | trans->transaction->delayed_refs.num_heads_ready > 64) { | |
461 | trans->delayed_ref_updates = 0; | |
b7ec40d7 CM |
462 | |
463 | /* | |
464 | * do a full flush if the transaction is trying | |
465 | * to close | |
466 | */ | |
467 | if (trans->transaction->delayed_refs.flushing) | |
468 | cur = 0; | |
c3e69d58 CM |
469 | btrfs_run_delayed_refs(trans, root, cur); |
470 | } else { | |
471 | break; | |
472 | } | |
473 | count++; | |
56bec294 CM |
474 | } |
475 | ||
a22285a6 YZ |
476 | btrfs_trans_release_metadata(trans, root); |
477 | ||
a4abeea4 JB |
478 | if (lock && !atomic_read(&root->fs_info->open_ioctl_trans) && |
479 | should_end_transaction(trans, root)) { | |
8929ecfa | 480 | trans->transaction->blocked = 1; |
a4abeea4 JB |
481 | smp_wmb(); |
482 | } | |
8929ecfa | 483 | |
0af3d00b | 484 | if (lock && cur_trans->blocked && !cur_trans->in_commit) { |
81317fde JB |
485 | if (throttle) { |
486 | /* | |
487 | * We may race with somebody else here so end up having | |
488 | * to call end_transaction on ourselves again, so inc | |
489 | * our use_count. | |
490 | */ | |
491 | trans->use_count++; | |
8929ecfa | 492 | return btrfs_commit_transaction(trans, root); |
81317fde | 493 | } else { |
8929ecfa | 494 | wake_up_process(info->transaction_kthread); |
81317fde | 495 | } |
8929ecfa YZ |
496 | } |
497 | ||
8929ecfa | 498 | WARN_ON(cur_trans != info->running_transaction); |
13c5a93e JB |
499 | WARN_ON(atomic_read(&cur_trans->num_writers) < 1); |
500 | atomic_dec(&cur_trans->num_writers); | |
89ce8a63 | 501 | |
99d16cbc | 502 | smp_mb(); |
79154b1b CM |
503 | if (waitqueue_active(&cur_trans->writer_wait)) |
504 | wake_up(&cur_trans->writer_wait); | |
79154b1b | 505 | put_transaction(cur_trans); |
9ed74f2d JB |
506 | |
507 | if (current->journal_info == trans) | |
508 | current->journal_info = NULL; | |
d6025579 | 509 | memset(trans, 0, sizeof(*trans)); |
2c90e5d6 | 510 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
ab78c84d | 511 | |
24bbcf04 YZ |
512 | if (throttle) |
513 | btrfs_run_delayed_iputs(root); | |
514 | ||
79154b1b CM |
515 | return 0; |
516 | } | |
517 | ||
89ce8a63 CM |
518 | int btrfs_end_transaction(struct btrfs_trans_handle *trans, |
519 | struct btrfs_root *root) | |
520 | { | |
16cdcec7 MX |
521 | int ret; |
522 | ||
523 | ret = __btrfs_end_transaction(trans, root, 0, 1); | |
524 | if (ret) | |
525 | return ret; | |
526 | return 0; | |
89ce8a63 CM |
527 | } |
528 | ||
529 | int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans, | |
530 | struct btrfs_root *root) | |
531 | { | |
16cdcec7 MX |
532 | int ret; |
533 | ||
534 | ret = __btrfs_end_transaction(trans, root, 1, 1); | |
535 | if (ret) | |
536 | return ret; | |
537 | return 0; | |
0af3d00b JB |
538 | } |
539 | ||
540 | int btrfs_end_transaction_nolock(struct btrfs_trans_handle *trans, | |
541 | struct btrfs_root *root) | |
542 | { | |
16cdcec7 MX |
543 | int ret; |
544 | ||
545 | ret = __btrfs_end_transaction(trans, root, 0, 0); | |
546 | if (ret) | |
547 | return ret; | |
548 | return 0; | |
549 | } | |
550 | ||
551 | int btrfs_end_transaction_dmeta(struct btrfs_trans_handle *trans, | |
552 | struct btrfs_root *root) | |
553 | { | |
554 | return __btrfs_end_transaction(trans, root, 1, 1); | |
89ce8a63 CM |
555 | } |
556 | ||
d352ac68 CM |
557 | /* |
558 | * when btree blocks are allocated, they have some corresponding bits set for | |
559 | * them in one of two extent_io trees. This is used to make sure all of | |
690587d1 | 560 | * those extents are sent to disk but does not wait on them |
d352ac68 | 561 | */ |
690587d1 | 562 | int btrfs_write_marked_extents(struct btrfs_root *root, |
8cef4e16 | 563 | struct extent_io_tree *dirty_pages, int mark) |
79154b1b | 564 | { |
7c4452b9 | 565 | int ret; |
777e6bd7 | 566 | int err = 0; |
7c4452b9 CM |
567 | int werr = 0; |
568 | struct page *page; | |
7c4452b9 | 569 | struct inode *btree_inode = root->fs_info->btree_inode; |
777e6bd7 | 570 | u64 start = 0; |
5f39d397 CM |
571 | u64 end; |
572 | unsigned long index; | |
7c4452b9 | 573 | |
d397712b | 574 | while (1) { |
777e6bd7 | 575 | ret = find_first_extent_bit(dirty_pages, start, &start, &end, |
8cef4e16 | 576 | mark); |
5f39d397 | 577 | if (ret) |
7c4452b9 | 578 | break; |
d397712b | 579 | while (start <= end) { |
777e6bd7 CM |
580 | cond_resched(); |
581 | ||
5f39d397 | 582 | index = start >> PAGE_CACHE_SHIFT; |
35ebb934 | 583 | start = (u64)(index + 1) << PAGE_CACHE_SHIFT; |
4bef0848 | 584 | page = find_get_page(btree_inode->i_mapping, index); |
7c4452b9 CM |
585 | if (!page) |
586 | continue; | |
4bef0848 CM |
587 | |
588 | btree_lock_page_hook(page); | |
589 | if (!page->mapping) { | |
590 | unlock_page(page); | |
591 | page_cache_release(page); | |
592 | continue; | |
593 | } | |
594 | ||
6702ed49 CM |
595 | if (PageWriteback(page)) { |
596 | if (PageDirty(page)) | |
597 | wait_on_page_writeback(page); | |
598 | else { | |
599 | unlock_page(page); | |
600 | page_cache_release(page); | |
601 | continue; | |
602 | } | |
603 | } | |
7c4452b9 CM |
604 | err = write_one_page(page, 0); |
605 | if (err) | |
606 | werr = err; | |
607 | page_cache_release(page); | |
608 | } | |
609 | } | |
690587d1 CM |
610 | if (err) |
611 | werr = err; | |
612 | return werr; | |
613 | } | |
614 | ||
615 | /* | |
616 | * when btree blocks are allocated, they have some corresponding bits set for | |
617 | * them in one of two extent_io trees. This is used to make sure all of | |
618 | * those extents are on disk for transaction or log commit. We wait | |
619 | * on all the pages and clear them from the dirty pages state tree | |
620 | */ | |
621 | int btrfs_wait_marked_extents(struct btrfs_root *root, | |
8cef4e16 | 622 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 CM |
623 | { |
624 | int ret; | |
625 | int err = 0; | |
626 | int werr = 0; | |
627 | struct page *page; | |
628 | struct inode *btree_inode = root->fs_info->btree_inode; | |
629 | u64 start = 0; | |
630 | u64 end; | |
631 | unsigned long index; | |
632 | ||
d397712b | 633 | while (1) { |
8cef4e16 YZ |
634 | ret = find_first_extent_bit(dirty_pages, start, &start, &end, |
635 | mark); | |
777e6bd7 CM |
636 | if (ret) |
637 | break; | |
638 | ||
8cef4e16 | 639 | clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS); |
d397712b | 640 | while (start <= end) { |
777e6bd7 CM |
641 | index = start >> PAGE_CACHE_SHIFT; |
642 | start = (u64)(index + 1) << PAGE_CACHE_SHIFT; | |
643 | page = find_get_page(btree_inode->i_mapping, index); | |
644 | if (!page) | |
645 | continue; | |
646 | if (PageDirty(page)) { | |
4bef0848 CM |
647 | btree_lock_page_hook(page); |
648 | wait_on_page_writeback(page); | |
777e6bd7 CM |
649 | err = write_one_page(page, 0); |
650 | if (err) | |
651 | werr = err; | |
652 | } | |
105d931d | 653 | wait_on_page_writeback(page); |
777e6bd7 CM |
654 | page_cache_release(page); |
655 | cond_resched(); | |
656 | } | |
657 | } | |
7c4452b9 CM |
658 | if (err) |
659 | werr = err; | |
660 | return werr; | |
79154b1b CM |
661 | } |
662 | ||
690587d1 CM |
663 | /* |
664 | * when btree blocks are allocated, they have some corresponding bits set for | |
665 | * them in one of two extent_io trees. This is used to make sure all of | |
666 | * those extents are on disk for transaction or log commit | |
667 | */ | |
668 | int btrfs_write_and_wait_marked_extents(struct btrfs_root *root, | |
8cef4e16 | 669 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 CM |
670 | { |
671 | int ret; | |
672 | int ret2; | |
673 | ||
8cef4e16 YZ |
674 | ret = btrfs_write_marked_extents(root, dirty_pages, mark); |
675 | ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark); | |
690587d1 CM |
676 | return ret || ret2; |
677 | } | |
678 | ||
d0c803c4 CM |
679 | int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans, |
680 | struct btrfs_root *root) | |
681 | { | |
682 | if (!trans || !trans->transaction) { | |
683 | struct inode *btree_inode; | |
684 | btree_inode = root->fs_info->btree_inode; | |
685 | return filemap_write_and_wait(btree_inode->i_mapping); | |
686 | } | |
687 | return btrfs_write_and_wait_marked_extents(root, | |
8cef4e16 YZ |
688 | &trans->transaction->dirty_pages, |
689 | EXTENT_DIRTY); | |
d0c803c4 CM |
690 | } |
691 | ||
d352ac68 CM |
692 | /* |
693 | * this is used to update the root pointer in the tree of tree roots. | |
694 | * | |
695 | * But, in the case of the extent allocation tree, updating the root | |
696 | * pointer may allocate blocks which may change the root of the extent | |
697 | * allocation tree. | |
698 | * | |
699 | * So, this loops and repeats and makes sure the cowonly root didn't | |
700 | * change while the root pointer was being updated in the metadata. | |
701 | */ | |
0b86a832 CM |
702 | static int update_cowonly_root(struct btrfs_trans_handle *trans, |
703 | struct btrfs_root *root) | |
79154b1b CM |
704 | { |
705 | int ret; | |
0b86a832 | 706 | u64 old_root_bytenr; |
86b9f2ec | 707 | u64 old_root_used; |
0b86a832 | 708 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
79154b1b | 709 | |
86b9f2ec | 710 | old_root_used = btrfs_root_used(&root->root_item); |
0b86a832 | 711 | btrfs_write_dirty_block_groups(trans, root); |
56bec294 | 712 | |
d397712b | 713 | while (1) { |
0b86a832 | 714 | old_root_bytenr = btrfs_root_bytenr(&root->root_item); |
86b9f2ec YZ |
715 | if (old_root_bytenr == root->node->start && |
716 | old_root_used == btrfs_root_used(&root->root_item)) | |
79154b1b | 717 | break; |
87ef2bb4 | 718 | |
5d4f98a2 | 719 | btrfs_set_root_node(&root->root_item, root->node); |
79154b1b | 720 | ret = btrfs_update_root(trans, tree_root, |
0b86a832 CM |
721 | &root->root_key, |
722 | &root->root_item); | |
79154b1b | 723 | BUG_ON(ret); |
56bec294 | 724 | |
86b9f2ec | 725 | old_root_used = btrfs_root_used(&root->root_item); |
4a8c9a62 | 726 | ret = btrfs_write_dirty_block_groups(trans, root); |
56bec294 | 727 | BUG_ON(ret); |
0b86a832 | 728 | } |
276e680d YZ |
729 | |
730 | if (root != root->fs_info->extent_root) | |
731 | switch_commit_root(root); | |
732 | ||
0b86a832 CM |
733 | return 0; |
734 | } | |
735 | ||
d352ac68 CM |
736 | /* |
737 | * update all the cowonly tree roots on disk | |
738 | */ | |
5d4f98a2 YZ |
739 | static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans, |
740 | struct btrfs_root *root) | |
0b86a832 CM |
741 | { |
742 | struct btrfs_fs_info *fs_info = root->fs_info; | |
743 | struct list_head *next; | |
84234f3a | 744 | struct extent_buffer *eb; |
56bec294 | 745 | int ret; |
84234f3a | 746 | |
56bec294 CM |
747 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
748 | BUG_ON(ret); | |
87ef2bb4 | 749 | |
84234f3a | 750 | eb = btrfs_lock_root_node(fs_info->tree_root); |
9fa8cfe7 | 751 | btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 0, &eb); |
84234f3a YZ |
752 | btrfs_tree_unlock(eb); |
753 | free_extent_buffer(eb); | |
0b86a832 | 754 | |
56bec294 CM |
755 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
756 | BUG_ON(ret); | |
87ef2bb4 | 757 | |
d397712b | 758 | while (!list_empty(&fs_info->dirty_cowonly_roots)) { |
0b86a832 CM |
759 | next = fs_info->dirty_cowonly_roots.next; |
760 | list_del_init(next); | |
761 | root = list_entry(next, struct btrfs_root, dirty_list); | |
87ef2bb4 | 762 | |
0b86a832 | 763 | update_cowonly_root(trans, root); |
79154b1b | 764 | } |
276e680d YZ |
765 | |
766 | down_write(&fs_info->extent_commit_sem); | |
767 | switch_commit_root(fs_info->extent_root); | |
768 | up_write(&fs_info->extent_commit_sem); | |
769 | ||
79154b1b CM |
770 | return 0; |
771 | } | |
772 | ||
d352ac68 CM |
773 | /* |
774 | * dead roots are old snapshots that need to be deleted. This allocates | |
775 | * a dirty root struct and adds it into the list of dead roots that need to | |
776 | * be deleted | |
777 | */ | |
5d4f98a2 | 778 | int btrfs_add_dead_root(struct btrfs_root *root) |
5eda7b5e | 779 | { |
a4abeea4 | 780 | spin_lock(&root->fs_info->trans_lock); |
5d4f98a2 | 781 | list_add(&root->root_list, &root->fs_info->dead_roots); |
a4abeea4 | 782 | spin_unlock(&root->fs_info->trans_lock); |
5eda7b5e CM |
783 | return 0; |
784 | } | |
785 | ||
d352ac68 | 786 | /* |
5d4f98a2 | 787 | * update all the cowonly tree roots on disk |
d352ac68 | 788 | */ |
5d4f98a2 YZ |
789 | static noinline int commit_fs_roots(struct btrfs_trans_handle *trans, |
790 | struct btrfs_root *root) | |
0f7d52f4 | 791 | { |
0f7d52f4 | 792 | struct btrfs_root *gang[8]; |
5d4f98a2 | 793 | struct btrfs_fs_info *fs_info = root->fs_info; |
0f7d52f4 CM |
794 | int i; |
795 | int ret; | |
54aa1f4d CM |
796 | int err = 0; |
797 | ||
a4abeea4 | 798 | spin_lock(&fs_info->fs_roots_radix_lock); |
d397712b | 799 | while (1) { |
5d4f98a2 YZ |
800 | ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, |
801 | (void **)gang, 0, | |
0f7d52f4 CM |
802 | ARRAY_SIZE(gang), |
803 | BTRFS_ROOT_TRANS_TAG); | |
804 | if (ret == 0) | |
805 | break; | |
806 | for (i = 0; i < ret; i++) { | |
807 | root = gang[i]; | |
5d4f98a2 YZ |
808 | radix_tree_tag_clear(&fs_info->fs_roots_radix, |
809 | (unsigned long)root->root_key.objectid, | |
810 | BTRFS_ROOT_TRANS_TAG); | |
a4abeea4 | 811 | spin_unlock(&fs_info->fs_roots_radix_lock); |
31153d81 | 812 | |
e02119d5 | 813 | btrfs_free_log(trans, root); |
5d4f98a2 | 814 | btrfs_update_reloc_root(trans, root); |
d68fc57b | 815 | btrfs_orphan_commit_root(trans, root); |
bcc63abb | 816 | |
82d5902d LZ |
817 | btrfs_save_ino_cache(root, trans); |
818 | ||
978d910d | 819 | if (root->commit_root != root->node) { |
581bb050 | 820 | mutex_lock(&root->fs_commit_mutex); |
817d52f8 | 821 | switch_commit_root(root); |
581bb050 LZ |
822 | btrfs_unpin_free_ino(root); |
823 | mutex_unlock(&root->fs_commit_mutex); | |
824 | ||
978d910d YZ |
825 | btrfs_set_root_node(&root->root_item, |
826 | root->node); | |
827 | } | |
5d4f98a2 | 828 | |
5d4f98a2 | 829 | err = btrfs_update_root(trans, fs_info->tree_root, |
0f7d52f4 CM |
830 | &root->root_key, |
831 | &root->root_item); | |
a4abeea4 | 832 | spin_lock(&fs_info->fs_roots_radix_lock); |
54aa1f4d CM |
833 | if (err) |
834 | break; | |
0f7d52f4 CM |
835 | } |
836 | } | |
a4abeea4 | 837 | spin_unlock(&fs_info->fs_roots_radix_lock); |
54aa1f4d | 838 | return err; |
0f7d52f4 CM |
839 | } |
840 | ||
d352ac68 CM |
841 | /* |
842 | * defrag a given btree. If cacheonly == 1, this won't read from the disk, | |
843 | * otherwise every leaf in the btree is read and defragged. | |
844 | */ | |
e9d0b13b CM |
845 | int btrfs_defrag_root(struct btrfs_root *root, int cacheonly) |
846 | { | |
847 | struct btrfs_fs_info *info = root->fs_info; | |
e9d0b13b | 848 | struct btrfs_trans_handle *trans; |
8929ecfa | 849 | int ret; |
d3c2fdcf | 850 | unsigned long nr; |
e9d0b13b | 851 | |
8929ecfa | 852 | if (xchg(&root->defrag_running, 1)) |
e9d0b13b | 853 | return 0; |
8929ecfa | 854 | |
6b80053d | 855 | while (1) { |
8929ecfa YZ |
856 | trans = btrfs_start_transaction(root, 0); |
857 | if (IS_ERR(trans)) | |
858 | return PTR_ERR(trans); | |
859 | ||
e9d0b13b | 860 | ret = btrfs_defrag_leaves(trans, root, cacheonly); |
8929ecfa | 861 | |
d3c2fdcf | 862 | nr = trans->blocks_used; |
e9d0b13b | 863 | btrfs_end_transaction(trans, root); |
d3c2fdcf | 864 | btrfs_btree_balance_dirty(info->tree_root, nr); |
e9d0b13b CM |
865 | cond_resched(); |
866 | ||
7841cb28 | 867 | if (btrfs_fs_closing(root->fs_info) || ret != -EAGAIN) |
e9d0b13b CM |
868 | break; |
869 | } | |
870 | root->defrag_running = 0; | |
8929ecfa | 871 | return ret; |
e9d0b13b CM |
872 | } |
873 | ||
d352ac68 CM |
874 | /* |
875 | * new snapshots need to be created at a very specific time in the | |
876 | * transaction commit. This does the actual creation | |
877 | */ | |
80b6794d | 878 | static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, |
3063d29f CM |
879 | struct btrfs_fs_info *fs_info, |
880 | struct btrfs_pending_snapshot *pending) | |
881 | { | |
882 | struct btrfs_key key; | |
80b6794d | 883 | struct btrfs_root_item *new_root_item; |
3063d29f CM |
884 | struct btrfs_root *tree_root = fs_info->tree_root; |
885 | struct btrfs_root *root = pending->root; | |
6bdb72de | 886 | struct btrfs_root *parent_root; |
98c9942a | 887 | struct btrfs_block_rsv *rsv; |
6bdb72de | 888 | struct inode *parent_inode; |
6a912213 | 889 | struct dentry *parent; |
a22285a6 | 890 | struct dentry *dentry; |
3063d29f | 891 | struct extent_buffer *tmp; |
925baedd | 892 | struct extent_buffer *old; |
3063d29f | 893 | int ret; |
d68fc57b | 894 | u64 to_reserve = 0; |
6bdb72de | 895 | u64 index = 0; |
a22285a6 | 896 | u64 objectid; |
b83cc969 | 897 | u64 root_flags; |
3063d29f | 898 | |
98c9942a LB |
899 | rsv = trans->block_rsv; |
900 | ||
80b6794d CM |
901 | new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS); |
902 | if (!new_root_item) { | |
a22285a6 | 903 | pending->error = -ENOMEM; |
80b6794d CM |
904 | goto fail; |
905 | } | |
a22285a6 | 906 | |
581bb050 | 907 | ret = btrfs_find_free_objectid(tree_root, &objectid); |
a22285a6 YZ |
908 | if (ret) { |
909 | pending->error = ret; | |
3063d29f | 910 | goto fail; |
a22285a6 | 911 | } |
3063d29f | 912 | |
3fd0a558 | 913 | btrfs_reloc_pre_snapshot(trans, pending, &to_reserve); |
d68fc57b YZ |
914 | |
915 | if (to_reserve > 0) { | |
916 | ret = btrfs_block_rsv_add(trans, root, &pending->block_rsv, | |
8bb8ab2e | 917 | to_reserve); |
d68fc57b YZ |
918 | if (ret) { |
919 | pending->error = ret; | |
920 | goto fail; | |
921 | } | |
922 | } | |
923 | ||
3063d29f | 924 | key.objectid = objectid; |
a22285a6 YZ |
925 | key.offset = (u64)-1; |
926 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3063d29f | 927 | |
a22285a6 | 928 | trans->block_rsv = &pending->block_rsv; |
3de4586c | 929 | |
a22285a6 | 930 | dentry = pending->dentry; |
6a912213 JB |
931 | parent = dget_parent(dentry); |
932 | parent_inode = parent->d_inode; | |
a22285a6 | 933 | parent_root = BTRFS_I(parent_inode)->root; |
7585717f | 934 | record_root_in_trans(trans, parent_root); |
a22285a6 | 935 | |
3063d29f CM |
936 | /* |
937 | * insert the directory item | |
938 | */ | |
3de4586c | 939 | ret = btrfs_set_inode_index(parent_inode, &index); |
6bdb72de | 940 | BUG_ON(ret); |
0660b5af | 941 | ret = btrfs_insert_dir_item(trans, parent_root, |
a22285a6 | 942 | dentry->d_name.name, dentry->d_name.len, |
16cdcec7 | 943 | parent_inode, &key, |
a22285a6 | 944 | BTRFS_FT_DIR, index); |
6bdb72de | 945 | BUG_ON(ret); |
0660b5af | 946 | |
a22285a6 YZ |
947 | btrfs_i_size_write(parent_inode, parent_inode->i_size + |
948 | dentry->d_name.len * 2); | |
52c26179 YZ |
949 | ret = btrfs_update_inode(trans, parent_root, parent_inode); |
950 | BUG_ON(ret); | |
951 | ||
e999376f CM |
952 | /* |
953 | * pull in the delayed directory update | |
954 | * and the delayed inode item | |
955 | * otherwise we corrupt the FS during | |
956 | * snapshot | |
957 | */ | |
958 | ret = btrfs_run_delayed_items(trans, root); | |
959 | BUG_ON(ret); | |
960 | ||
7585717f | 961 | record_root_in_trans(trans, root); |
6bdb72de SW |
962 | btrfs_set_root_last_snapshot(&root->root_item, trans->transid); |
963 | memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); | |
08fe4db1 | 964 | btrfs_check_and_init_root_item(new_root_item); |
6bdb72de | 965 | |
b83cc969 LZ |
966 | root_flags = btrfs_root_flags(new_root_item); |
967 | if (pending->readonly) | |
968 | root_flags |= BTRFS_ROOT_SUBVOL_RDONLY; | |
969 | else | |
970 | root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY; | |
971 | btrfs_set_root_flags(new_root_item, root_flags); | |
972 | ||
6bdb72de SW |
973 | old = btrfs_lock_root_node(root); |
974 | btrfs_cow_block(trans, root, old, NULL, 0, &old); | |
975 | btrfs_set_lock_blocking(old); | |
976 | ||
977 | btrfs_copy_root(trans, root, old, &tmp, objectid); | |
978 | btrfs_tree_unlock(old); | |
979 | free_extent_buffer(old); | |
980 | ||
981 | btrfs_set_root_node(new_root_item, tmp); | |
a22285a6 YZ |
982 | /* record when the snapshot was created in key.offset */ |
983 | key.offset = trans->transid; | |
984 | ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); | |
6bdb72de SW |
985 | btrfs_tree_unlock(tmp); |
986 | free_extent_buffer(tmp); | |
a22285a6 | 987 | BUG_ON(ret); |
6bdb72de | 988 | |
a22285a6 YZ |
989 | /* |
990 | * insert root back/forward references | |
991 | */ | |
992 | ret = btrfs_add_root_ref(trans, tree_root, objectid, | |
0660b5af | 993 | parent_root->root_key.objectid, |
33345d01 | 994 | btrfs_ino(parent_inode), index, |
a22285a6 | 995 | dentry->d_name.name, dentry->d_name.len); |
0660b5af | 996 | BUG_ON(ret); |
6a912213 | 997 | dput(parent); |
0660b5af | 998 | |
a22285a6 YZ |
999 | key.offset = (u64)-1; |
1000 | pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key); | |
1001 | BUG_ON(IS_ERR(pending->snap)); | |
d68fc57b | 1002 | |
3fd0a558 | 1003 | btrfs_reloc_post_snapshot(trans, pending); |
3063d29f | 1004 | fail: |
6bdb72de | 1005 | kfree(new_root_item); |
98c9942a | 1006 | trans->block_rsv = rsv; |
a22285a6 YZ |
1007 | btrfs_block_rsv_release(root, &pending->block_rsv, (u64)-1); |
1008 | return 0; | |
3063d29f CM |
1009 | } |
1010 | ||
d352ac68 CM |
1011 | /* |
1012 | * create all the snapshots we've scheduled for creation | |
1013 | */ | |
80b6794d CM |
1014 | static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans, |
1015 | struct btrfs_fs_info *fs_info) | |
3de4586c CM |
1016 | { |
1017 | struct btrfs_pending_snapshot *pending; | |
1018 | struct list_head *head = &trans->transaction->pending_snapshots; | |
3de4586c CM |
1019 | int ret; |
1020 | ||
c6e30871 | 1021 | list_for_each_entry(pending, head, list) { |
3de4586c CM |
1022 | ret = create_pending_snapshot(trans, fs_info, pending); |
1023 | BUG_ON(ret); | |
1024 | } | |
1025 | return 0; | |
1026 | } | |
1027 | ||
5d4f98a2 YZ |
1028 | static void update_super_roots(struct btrfs_root *root) |
1029 | { | |
1030 | struct btrfs_root_item *root_item; | |
1031 | struct btrfs_super_block *super; | |
1032 | ||
1033 | super = &root->fs_info->super_copy; | |
1034 | ||
1035 | root_item = &root->fs_info->chunk_root->root_item; | |
1036 | super->chunk_root = root_item->bytenr; | |
1037 | super->chunk_root_generation = root_item->generation; | |
1038 | super->chunk_root_level = root_item->level; | |
1039 | ||
1040 | root_item = &root->fs_info->tree_root->root_item; | |
1041 | super->root = root_item->bytenr; | |
1042 | super->generation = root_item->generation; | |
1043 | super->root_level = root_item->level; | |
0af3d00b JB |
1044 | if (super->cache_generation != 0 || btrfs_test_opt(root, SPACE_CACHE)) |
1045 | super->cache_generation = root_item->generation; | |
5d4f98a2 YZ |
1046 | } |
1047 | ||
f36f3042 CM |
1048 | int btrfs_transaction_in_commit(struct btrfs_fs_info *info) |
1049 | { | |
1050 | int ret = 0; | |
a4abeea4 | 1051 | spin_lock(&info->trans_lock); |
f36f3042 CM |
1052 | if (info->running_transaction) |
1053 | ret = info->running_transaction->in_commit; | |
a4abeea4 | 1054 | spin_unlock(&info->trans_lock); |
f36f3042 CM |
1055 | return ret; |
1056 | } | |
1057 | ||
8929ecfa YZ |
1058 | int btrfs_transaction_blocked(struct btrfs_fs_info *info) |
1059 | { | |
1060 | int ret = 0; | |
a4abeea4 | 1061 | spin_lock(&info->trans_lock); |
8929ecfa YZ |
1062 | if (info->running_transaction) |
1063 | ret = info->running_transaction->blocked; | |
a4abeea4 | 1064 | spin_unlock(&info->trans_lock); |
8929ecfa YZ |
1065 | return ret; |
1066 | } | |
1067 | ||
bb9c12c9 SW |
1068 | /* |
1069 | * wait for the current transaction commit to start and block subsequent | |
1070 | * transaction joins | |
1071 | */ | |
1072 | static void wait_current_trans_commit_start(struct btrfs_root *root, | |
1073 | struct btrfs_transaction *trans) | |
1074 | { | |
72d63ed6 | 1075 | wait_event(root->fs_info->transaction_blocked_wait, trans->in_commit); |
bb9c12c9 SW |
1076 | } |
1077 | ||
1078 | /* | |
1079 | * wait for the current transaction to start and then become unblocked. | |
1080 | * caller holds ref. | |
1081 | */ | |
1082 | static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root, | |
1083 | struct btrfs_transaction *trans) | |
1084 | { | |
72d63ed6 LZ |
1085 | wait_event(root->fs_info->transaction_wait, |
1086 | trans->commit_done || (trans->in_commit && !trans->blocked)); | |
bb9c12c9 SW |
1087 | } |
1088 | ||
1089 | /* | |
1090 | * commit transactions asynchronously. once btrfs_commit_transaction_async | |
1091 | * returns, any subsequent transaction will not be allowed to join. | |
1092 | */ | |
1093 | struct btrfs_async_commit { | |
1094 | struct btrfs_trans_handle *newtrans; | |
1095 | struct btrfs_root *root; | |
1096 | struct delayed_work work; | |
1097 | }; | |
1098 | ||
1099 | static void do_async_commit(struct work_struct *work) | |
1100 | { | |
1101 | struct btrfs_async_commit *ac = | |
1102 | container_of(work, struct btrfs_async_commit, work.work); | |
1103 | ||
1104 | btrfs_commit_transaction(ac->newtrans, ac->root); | |
1105 | kfree(ac); | |
1106 | } | |
1107 | ||
1108 | int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans, | |
1109 | struct btrfs_root *root, | |
1110 | int wait_for_unblock) | |
1111 | { | |
1112 | struct btrfs_async_commit *ac; | |
1113 | struct btrfs_transaction *cur_trans; | |
1114 | ||
1115 | ac = kmalloc(sizeof(*ac), GFP_NOFS); | |
db5b493a TI |
1116 | if (!ac) |
1117 | return -ENOMEM; | |
bb9c12c9 SW |
1118 | |
1119 | INIT_DELAYED_WORK(&ac->work, do_async_commit); | |
1120 | ac->root = root; | |
7a7eaa40 | 1121 | ac->newtrans = btrfs_join_transaction(root); |
3612b495 TI |
1122 | if (IS_ERR(ac->newtrans)) { |
1123 | int err = PTR_ERR(ac->newtrans); | |
1124 | kfree(ac); | |
1125 | return err; | |
1126 | } | |
bb9c12c9 SW |
1127 | |
1128 | /* take transaction reference */ | |
bb9c12c9 | 1129 | cur_trans = trans->transaction; |
13c5a93e | 1130 | atomic_inc(&cur_trans->use_count); |
bb9c12c9 SW |
1131 | |
1132 | btrfs_end_transaction(trans, root); | |
1133 | schedule_delayed_work(&ac->work, 0); | |
1134 | ||
1135 | /* wait for transaction to start and unblock */ | |
bb9c12c9 SW |
1136 | if (wait_for_unblock) |
1137 | wait_current_trans_commit_start_and_unblock(root, cur_trans); | |
1138 | else | |
1139 | wait_current_trans_commit_start(root, cur_trans); | |
bb9c12c9 | 1140 | |
38e88054 SW |
1141 | if (current->journal_info == trans) |
1142 | current->journal_info = NULL; | |
1143 | ||
1144 | put_transaction(cur_trans); | |
bb9c12c9 SW |
1145 | return 0; |
1146 | } | |
1147 | ||
1148 | /* | |
1149 | * btrfs_transaction state sequence: | |
1150 | * in_commit = 0, blocked = 0 (initial) | |
1151 | * in_commit = 1, blocked = 1 | |
1152 | * blocked = 0 | |
1153 | * commit_done = 1 | |
1154 | */ | |
79154b1b CM |
1155 | int btrfs_commit_transaction(struct btrfs_trans_handle *trans, |
1156 | struct btrfs_root *root) | |
1157 | { | |
15ee9bc7 | 1158 | unsigned long joined = 0; |
79154b1b | 1159 | struct btrfs_transaction *cur_trans; |
8fd17795 | 1160 | struct btrfs_transaction *prev_trans = NULL; |
79154b1b | 1161 | DEFINE_WAIT(wait); |
15ee9bc7 | 1162 | int ret; |
89573b9c CM |
1163 | int should_grow = 0; |
1164 | unsigned long now = get_seconds(); | |
dccae999 | 1165 | int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT); |
79154b1b | 1166 | |
5a3f23d5 CM |
1167 | btrfs_run_ordered_operations(root, 0); |
1168 | ||
56bec294 CM |
1169 | /* make a pass through all the delayed refs we have so far |
1170 | * any runnings procs may add more while we are here | |
1171 | */ | |
1172 | ret = btrfs_run_delayed_refs(trans, root, 0); | |
1173 | BUG_ON(ret); | |
1174 | ||
a22285a6 YZ |
1175 | btrfs_trans_release_metadata(trans, root); |
1176 | ||
b7ec40d7 | 1177 | cur_trans = trans->transaction; |
56bec294 CM |
1178 | /* |
1179 | * set the flushing flag so procs in this transaction have to | |
1180 | * start sending their work down. | |
1181 | */ | |
b7ec40d7 | 1182 | cur_trans->delayed_refs.flushing = 1; |
56bec294 | 1183 | |
c3e69d58 | 1184 | ret = btrfs_run_delayed_refs(trans, root, 0); |
56bec294 CM |
1185 | BUG_ON(ret); |
1186 | ||
a4abeea4 | 1187 | spin_lock(&cur_trans->commit_lock); |
b7ec40d7 | 1188 | if (cur_trans->in_commit) { |
a4abeea4 | 1189 | spin_unlock(&cur_trans->commit_lock); |
13c5a93e | 1190 | atomic_inc(&cur_trans->use_count); |
79154b1b | 1191 | btrfs_end_transaction(trans, root); |
ccd467d6 | 1192 | |
b9c8300c | 1193 | wait_for_commit(root, cur_trans); |
15ee9bc7 | 1194 | |
79154b1b | 1195 | put_transaction(cur_trans); |
15ee9bc7 | 1196 | |
79154b1b CM |
1197 | return 0; |
1198 | } | |
4313b399 | 1199 | |
2c90e5d6 | 1200 | trans->transaction->in_commit = 1; |
f9295749 | 1201 | trans->transaction->blocked = 1; |
a4abeea4 | 1202 | spin_unlock(&cur_trans->commit_lock); |
bb9c12c9 SW |
1203 | wake_up(&root->fs_info->transaction_blocked_wait); |
1204 | ||
a4abeea4 | 1205 | spin_lock(&root->fs_info->trans_lock); |
ccd467d6 CM |
1206 | if (cur_trans->list.prev != &root->fs_info->trans_list) { |
1207 | prev_trans = list_entry(cur_trans->list.prev, | |
1208 | struct btrfs_transaction, list); | |
1209 | if (!prev_trans->commit_done) { | |
13c5a93e | 1210 | atomic_inc(&prev_trans->use_count); |
a4abeea4 | 1211 | spin_unlock(&root->fs_info->trans_lock); |
ccd467d6 CM |
1212 | |
1213 | wait_for_commit(root, prev_trans); | |
ccd467d6 | 1214 | |
15ee9bc7 | 1215 | put_transaction(prev_trans); |
a4abeea4 JB |
1216 | } else { |
1217 | spin_unlock(&root->fs_info->trans_lock); | |
ccd467d6 | 1218 | } |
a4abeea4 JB |
1219 | } else { |
1220 | spin_unlock(&root->fs_info->trans_lock); | |
ccd467d6 | 1221 | } |
15ee9bc7 | 1222 | |
89573b9c CM |
1223 | if (now < cur_trans->start_time || now - cur_trans->start_time < 1) |
1224 | should_grow = 1; | |
1225 | ||
15ee9bc7 | 1226 | do { |
7ea394f1 | 1227 | int snap_pending = 0; |
a4abeea4 | 1228 | |
15ee9bc7 | 1229 | joined = cur_trans->num_joined; |
7ea394f1 YZ |
1230 | if (!list_empty(&trans->transaction->pending_snapshots)) |
1231 | snap_pending = 1; | |
1232 | ||
2c90e5d6 | 1233 | WARN_ON(cur_trans != trans->transaction); |
15ee9bc7 | 1234 | |
0bdb1db2 | 1235 | if (flush_on_commit || snap_pending) { |
24bbcf04 YZ |
1236 | btrfs_start_delalloc_inodes(root, 1); |
1237 | ret = btrfs_wait_ordered_extents(root, 0, 1); | |
ebecd3d9 | 1238 | BUG_ON(ret); |
7ea394f1 YZ |
1239 | } |
1240 | ||
16cdcec7 MX |
1241 | ret = btrfs_run_delayed_items(trans, root); |
1242 | BUG_ON(ret); | |
1243 | ||
5a3f23d5 CM |
1244 | /* |
1245 | * rename don't use btrfs_join_transaction, so, once we | |
1246 | * set the transaction to blocked above, we aren't going | |
1247 | * to get any new ordered operations. We can safely run | |
1248 | * it here and no for sure that nothing new will be added | |
1249 | * to the list | |
1250 | */ | |
1251 | btrfs_run_ordered_operations(root, 1); | |
1252 | ||
ed3b3d31 CM |
1253 | prepare_to_wait(&cur_trans->writer_wait, &wait, |
1254 | TASK_UNINTERRUPTIBLE); | |
1255 | ||
13c5a93e | 1256 | if (atomic_read(&cur_trans->num_writers) > 1) |
99d16cbc SW |
1257 | schedule_timeout(MAX_SCHEDULE_TIMEOUT); |
1258 | else if (should_grow) | |
1259 | schedule_timeout(1); | |
15ee9bc7 | 1260 | |
15ee9bc7 | 1261 | finish_wait(&cur_trans->writer_wait, &wait); |
13c5a93e | 1262 | } while (atomic_read(&cur_trans->num_writers) > 1 || |
89573b9c | 1263 | (should_grow && cur_trans->num_joined != joined)); |
15ee9bc7 | 1264 | |
ed0ca140 JB |
1265 | /* |
1266 | * Ok now we need to make sure to block out any other joins while we | |
1267 | * commit the transaction. We could have started a join before setting | |
1268 | * no_join so make sure to wait for num_writers to == 1 again. | |
1269 | */ | |
1270 | spin_lock(&root->fs_info->trans_lock); | |
1271 | root->fs_info->trans_no_join = 1; | |
1272 | spin_unlock(&root->fs_info->trans_lock); | |
1273 | wait_event(cur_trans->writer_wait, | |
1274 | atomic_read(&cur_trans->num_writers) == 1); | |
1275 | ||
7585717f CM |
1276 | /* |
1277 | * the reloc mutex makes sure that we stop | |
1278 | * the balancing code from coming in and moving | |
1279 | * extents around in the middle of the commit | |
1280 | */ | |
1281 | mutex_lock(&root->fs_info->reloc_mutex); | |
1282 | ||
e999376f | 1283 | ret = btrfs_run_delayed_items(trans, root); |
3063d29f CM |
1284 | BUG_ON(ret); |
1285 | ||
e999376f | 1286 | ret = create_pending_snapshots(trans, root->fs_info); |
16cdcec7 MX |
1287 | BUG_ON(ret); |
1288 | ||
56bec294 CM |
1289 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
1290 | BUG_ON(ret); | |
1291 | ||
e999376f CM |
1292 | /* |
1293 | * make sure none of the code above managed to slip in a | |
1294 | * delayed item | |
1295 | */ | |
1296 | btrfs_assert_delayed_root_empty(root); | |
1297 | ||
2c90e5d6 | 1298 | WARN_ON(cur_trans != trans->transaction); |
dc17ff8f | 1299 | |
a2de733c | 1300 | btrfs_scrub_pause(root); |
e02119d5 CM |
1301 | /* btrfs_commit_tree_roots is responsible for getting the |
1302 | * various roots consistent with each other. Every pointer | |
1303 | * in the tree of tree roots has to point to the most up to date | |
1304 | * root for every subvolume and other tree. So, we have to keep | |
1305 | * the tree logging code from jumping in and changing any | |
1306 | * of the trees. | |
1307 | * | |
1308 | * At this point in the commit, there can't be any tree-log | |
1309 | * writers, but a little lower down we drop the trans mutex | |
1310 | * and let new people in. By holding the tree_log_mutex | |
1311 | * from now until after the super is written, we avoid races | |
1312 | * with the tree-log code. | |
1313 | */ | |
1314 | mutex_lock(&root->fs_info->tree_log_mutex); | |
1315 | ||
5d4f98a2 | 1316 | ret = commit_fs_roots(trans, root); |
54aa1f4d CM |
1317 | BUG_ON(ret); |
1318 | ||
5d4f98a2 | 1319 | /* commit_fs_roots gets rid of all the tree log roots, it is now |
e02119d5 CM |
1320 | * safe to free the root of tree log roots |
1321 | */ | |
1322 | btrfs_free_log_root_tree(trans, root->fs_info); | |
1323 | ||
5d4f98a2 | 1324 | ret = commit_cowonly_roots(trans, root); |
79154b1b | 1325 | BUG_ON(ret); |
54aa1f4d | 1326 | |
11833d66 YZ |
1327 | btrfs_prepare_extent_commit(trans, root); |
1328 | ||
78fae27e | 1329 | cur_trans = root->fs_info->running_transaction; |
5d4f98a2 YZ |
1330 | |
1331 | btrfs_set_root_node(&root->fs_info->tree_root->root_item, | |
1332 | root->fs_info->tree_root->node); | |
817d52f8 | 1333 | switch_commit_root(root->fs_info->tree_root); |
5d4f98a2 YZ |
1334 | |
1335 | btrfs_set_root_node(&root->fs_info->chunk_root->root_item, | |
1336 | root->fs_info->chunk_root->node); | |
817d52f8 | 1337 | switch_commit_root(root->fs_info->chunk_root); |
5d4f98a2 YZ |
1338 | |
1339 | update_super_roots(root); | |
e02119d5 CM |
1340 | |
1341 | if (!root->fs_info->log_root_recovering) { | |
1342 | btrfs_set_super_log_root(&root->fs_info->super_copy, 0); | |
1343 | btrfs_set_super_log_root_level(&root->fs_info->super_copy, 0); | |
1344 | } | |
1345 | ||
a061fc8d CM |
1346 | memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy, |
1347 | sizeof(root->fs_info->super_copy)); | |
ccd467d6 | 1348 | |
f9295749 | 1349 | trans->transaction->blocked = 0; |
a4abeea4 JB |
1350 | spin_lock(&root->fs_info->trans_lock); |
1351 | root->fs_info->running_transaction = NULL; | |
1352 | root->fs_info->trans_no_join = 0; | |
1353 | spin_unlock(&root->fs_info->trans_lock); | |
7585717f | 1354 | mutex_unlock(&root->fs_info->reloc_mutex); |
b7ec40d7 | 1355 | |
f9295749 | 1356 | wake_up(&root->fs_info->transaction_wait); |
e6dcd2dc | 1357 | |
79154b1b CM |
1358 | ret = btrfs_write_and_wait_transaction(trans, root); |
1359 | BUG_ON(ret); | |
a512bbf8 | 1360 | write_ctree_super(trans, root, 0); |
4313b399 | 1361 | |
e02119d5 CM |
1362 | /* |
1363 | * the super is written, we can safely allow the tree-loggers | |
1364 | * to go about their business | |
1365 | */ | |
1366 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
1367 | ||
11833d66 | 1368 | btrfs_finish_extent_commit(trans, root); |
4313b399 | 1369 | |
2c90e5d6 | 1370 | cur_trans->commit_done = 1; |
b7ec40d7 | 1371 | |
15ee9bc7 | 1372 | root->fs_info->last_trans_committed = cur_trans->transid; |
817d52f8 | 1373 | |
2c90e5d6 | 1374 | wake_up(&cur_trans->commit_wait); |
3de4586c | 1375 | |
a4abeea4 | 1376 | spin_lock(&root->fs_info->trans_lock); |
13c5a93e | 1377 | list_del_init(&cur_trans->list); |
a4abeea4 JB |
1378 | spin_unlock(&root->fs_info->trans_lock); |
1379 | ||
78fae27e | 1380 | put_transaction(cur_trans); |
79154b1b | 1381 | put_transaction(cur_trans); |
58176a96 | 1382 | |
1abe9b8a | 1383 | trace_btrfs_transaction_commit(root); |
1384 | ||
a2de733c AJ |
1385 | btrfs_scrub_continue(root); |
1386 | ||
9ed74f2d JB |
1387 | if (current->journal_info == trans) |
1388 | current->journal_info = NULL; | |
1389 | ||
2c90e5d6 | 1390 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
24bbcf04 YZ |
1391 | |
1392 | if (current != root->fs_info->transaction_kthread) | |
1393 | btrfs_run_delayed_iputs(root); | |
1394 | ||
79154b1b CM |
1395 | return ret; |
1396 | } | |
1397 | ||
d352ac68 CM |
1398 | /* |
1399 | * interface function to delete all the snapshots we have scheduled for deletion | |
1400 | */ | |
e9d0b13b CM |
1401 | int btrfs_clean_old_snapshots(struct btrfs_root *root) |
1402 | { | |
5d4f98a2 YZ |
1403 | LIST_HEAD(list); |
1404 | struct btrfs_fs_info *fs_info = root->fs_info; | |
1405 | ||
a4abeea4 | 1406 | spin_lock(&fs_info->trans_lock); |
5d4f98a2 | 1407 | list_splice_init(&fs_info->dead_roots, &list); |
a4abeea4 | 1408 | spin_unlock(&fs_info->trans_lock); |
e9d0b13b | 1409 | |
5d4f98a2 YZ |
1410 | while (!list_empty(&list)) { |
1411 | root = list_entry(list.next, struct btrfs_root, root_list); | |
76dda93c YZ |
1412 | list_del(&root->root_list); |
1413 | ||
16cdcec7 MX |
1414 | btrfs_kill_all_delayed_nodes(root); |
1415 | ||
76dda93c YZ |
1416 | if (btrfs_header_backref_rev(root->node) < |
1417 | BTRFS_MIXED_BACKREF_REV) | |
3fd0a558 | 1418 | btrfs_drop_snapshot(root, NULL, 0); |
76dda93c | 1419 | else |
3fd0a558 | 1420 | btrfs_drop_snapshot(root, NULL, 1); |
e9d0b13b CM |
1421 | } |
1422 | return 0; | |
1423 | } |