]>
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 | */ |
a4abeea4 JB |
129 | int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, |
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 | ||
a4abeea4 JB |
136 | spin_lock(&root->fs_info->fs_roots_radix_lock); |
137 | if (root->last_trans == trans->transid) { | |
138 | spin_unlock(&root->fs_info->fs_roots_radix_lock); | |
139 | return 0; | |
140 | } | |
141 | root->last_trans = trans->transid; | |
5d4f98a2 YZ |
142 | radix_tree_tag_set(&root->fs_info->fs_roots_radix, |
143 | (unsigned long)root->root_key.objectid, | |
144 | BTRFS_ROOT_TRANS_TAG); | |
a4abeea4 | 145 | spin_unlock(&root->fs_info->fs_roots_radix_lock); |
5d4f98a2 YZ |
146 | btrfs_init_reloc_root(trans, root); |
147 | } | |
148 | return 0; | |
149 | } | |
bcc63abb | 150 | |
d352ac68 CM |
151 | /* wait for commit against the current transaction to become unblocked |
152 | * when this is done, it is safe to start a new transaction, but the current | |
153 | * transaction might not be fully on disk. | |
154 | */ | |
37d1aeee | 155 | static void wait_current_trans(struct btrfs_root *root) |
79154b1b | 156 | { |
f9295749 | 157 | struct btrfs_transaction *cur_trans; |
79154b1b | 158 | |
a4abeea4 | 159 | spin_lock(&root->fs_info->trans_lock); |
f9295749 | 160 | cur_trans = root->fs_info->running_transaction; |
37d1aeee | 161 | if (cur_trans && cur_trans->blocked) { |
f9295749 | 162 | DEFINE_WAIT(wait); |
13c5a93e | 163 | atomic_inc(&cur_trans->use_count); |
a4abeea4 | 164 | spin_unlock(&root->fs_info->trans_lock); |
d397712b | 165 | while (1) { |
f9295749 CM |
166 | prepare_to_wait(&root->fs_info->transaction_wait, &wait, |
167 | TASK_UNINTERRUPTIBLE); | |
471fa17d | 168 | if (!cur_trans->blocked) |
f9295749 | 169 | break; |
471fa17d | 170 | schedule(); |
f9295749 | 171 | } |
471fa17d | 172 | finish_wait(&root->fs_info->transaction_wait, &wait); |
f9295749 | 173 | put_transaction(cur_trans); |
a4abeea4 JB |
174 | } else { |
175 | spin_unlock(&root->fs_info->trans_lock); | |
f9295749 | 176 | } |
37d1aeee CM |
177 | } |
178 | ||
249ac1e5 JB |
179 | enum btrfs_trans_type { |
180 | TRANS_START, | |
181 | TRANS_JOIN, | |
182 | TRANS_USERSPACE, | |
0af3d00b | 183 | TRANS_JOIN_NOLOCK, |
249ac1e5 JB |
184 | }; |
185 | ||
a22285a6 YZ |
186 | static int may_wait_transaction(struct btrfs_root *root, int type) |
187 | { | |
a4abeea4 JB |
188 | if (root->fs_info->log_root_recovering) |
189 | return 0; | |
190 | ||
191 | if (type == TRANS_USERSPACE) | |
192 | return 1; | |
193 | ||
194 | if (type == TRANS_START && | |
195 | !atomic_read(&root->fs_info->open_ioctl_trans)) | |
a22285a6 | 196 | return 1; |
a4abeea4 | 197 | |
a22285a6 YZ |
198 | return 0; |
199 | } | |
200 | ||
e02119d5 | 201 | static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root, |
a22285a6 | 202 | u64 num_items, int type) |
37d1aeee | 203 | { |
a22285a6 YZ |
204 | struct btrfs_trans_handle *h; |
205 | struct btrfs_transaction *cur_trans; | |
06d5a589 | 206 | int retries = 0; |
37d1aeee | 207 | int ret; |
acce952b | 208 | |
209 | if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) | |
210 | return ERR_PTR(-EROFS); | |
2a1eb461 JB |
211 | |
212 | if (current->journal_info) { | |
213 | WARN_ON(type != TRANS_JOIN && type != TRANS_JOIN_NOLOCK); | |
214 | h = current->journal_info; | |
215 | h->use_count++; | |
216 | h->orig_rsv = h->block_rsv; | |
217 | h->block_rsv = NULL; | |
218 | goto got_it; | |
219 | } | |
a22285a6 YZ |
220 | again: |
221 | h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS); | |
222 | if (!h) | |
223 | return ERR_PTR(-ENOMEM); | |
37d1aeee | 224 | |
a22285a6 | 225 | if (may_wait_transaction(root, type)) |
37d1aeee | 226 | wait_current_trans(root); |
a22285a6 | 227 | |
a4abeea4 JB |
228 | do { |
229 | ret = join_transaction(root, type == TRANS_JOIN_NOLOCK); | |
230 | if (ret == -EBUSY) | |
231 | wait_current_trans(root); | |
232 | } while (ret == -EBUSY); | |
233 | ||
db5b493a | 234 | if (ret < 0) { |
6e8df2ae | 235 | kmem_cache_free(btrfs_trans_handle_cachep, h); |
db5b493a TI |
236 | return ERR_PTR(ret); |
237 | } | |
0f7d52f4 | 238 | |
a22285a6 | 239 | cur_trans = root->fs_info->running_transaction; |
a22285a6 YZ |
240 | |
241 | h->transid = cur_trans->transid; | |
242 | h->transaction = cur_trans; | |
79154b1b | 243 | h->blocks_used = 0; |
a22285a6 | 244 | h->bytes_reserved = 0; |
56bec294 | 245 | h->delayed_ref_updates = 0; |
2a1eb461 | 246 | h->use_count = 1; |
f0486c68 | 247 | h->block_rsv = NULL; |
2a1eb461 | 248 | h->orig_rsv = NULL; |
b7ec40d7 | 249 | |
a22285a6 YZ |
250 | smp_mb(); |
251 | if (cur_trans->blocked && may_wait_transaction(root, type)) { | |
252 | btrfs_commit_transaction(h, root); | |
253 | goto again; | |
254 | } | |
255 | ||
256 | if (num_items > 0) { | |
8bb8ab2e | 257 | ret = btrfs_trans_reserve_metadata(h, root, num_items); |
06d5a589 JB |
258 | if (ret == -EAGAIN && !retries) { |
259 | retries++; | |
a22285a6 YZ |
260 | btrfs_commit_transaction(h, root); |
261 | goto again; | |
06d5a589 JB |
262 | } else if (ret == -EAGAIN) { |
263 | /* | |
264 | * We have already retried and got EAGAIN, so really we | |
265 | * don't have space, so set ret to -ENOSPC. | |
266 | */ | |
267 | ret = -ENOSPC; | |
a22285a6 | 268 | } |
06d5a589 | 269 | |
a22285a6 YZ |
270 | if (ret < 0) { |
271 | btrfs_end_transaction(h, root); | |
272 | return ERR_PTR(ret); | |
273 | } | |
274 | } | |
9ed74f2d | 275 | |
2a1eb461 | 276 | got_it: |
a4abeea4 | 277 | btrfs_record_root_in_trans(h, root); |
a22285a6 YZ |
278 | |
279 | if (!current->journal_info && type != TRANS_USERSPACE) | |
280 | current->journal_info = h; | |
79154b1b CM |
281 | return h; |
282 | } | |
283 | ||
f9295749 | 284 | struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, |
a22285a6 | 285 | int num_items) |
f9295749 | 286 | { |
a22285a6 | 287 | return start_transaction(root, num_items, TRANS_START); |
f9295749 | 288 | } |
7a7eaa40 | 289 | struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root) |
f9295749 | 290 | { |
a22285a6 | 291 | return start_transaction(root, 0, TRANS_JOIN); |
f9295749 CM |
292 | } |
293 | ||
7a7eaa40 | 294 | struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root) |
0af3d00b JB |
295 | { |
296 | return start_transaction(root, 0, TRANS_JOIN_NOLOCK); | |
297 | } | |
298 | ||
7a7eaa40 | 299 | struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root) |
9ca9ee09 | 300 | { |
7a7eaa40 | 301 | return start_transaction(root, 0, TRANS_USERSPACE); |
9ca9ee09 SW |
302 | } |
303 | ||
d352ac68 | 304 | /* wait for a transaction commit to be fully complete */ |
89ce8a63 CM |
305 | static noinline int wait_for_commit(struct btrfs_root *root, |
306 | struct btrfs_transaction *commit) | |
307 | { | |
308 | DEFINE_WAIT(wait); | |
d397712b | 309 | while (!commit->commit_done) { |
89ce8a63 CM |
310 | prepare_to_wait(&commit->commit_wait, &wait, |
311 | TASK_UNINTERRUPTIBLE); | |
312 | if (commit->commit_done) | |
313 | break; | |
89ce8a63 | 314 | schedule(); |
89ce8a63 | 315 | } |
89ce8a63 CM |
316 | finish_wait(&commit->commit_wait, &wait); |
317 | return 0; | |
318 | } | |
319 | ||
46204592 SW |
320 | int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid) |
321 | { | |
322 | struct btrfs_transaction *cur_trans = NULL, *t; | |
323 | int ret; | |
324 | ||
46204592 SW |
325 | ret = 0; |
326 | if (transid) { | |
327 | if (transid <= root->fs_info->last_trans_committed) | |
a4abeea4 | 328 | goto out; |
46204592 SW |
329 | |
330 | /* find specified transaction */ | |
a4abeea4 | 331 | spin_lock(&root->fs_info->trans_lock); |
46204592 SW |
332 | list_for_each_entry(t, &root->fs_info->trans_list, list) { |
333 | if (t->transid == transid) { | |
334 | cur_trans = t; | |
a4abeea4 | 335 | atomic_inc(&cur_trans->use_count); |
46204592 SW |
336 | break; |
337 | } | |
338 | if (t->transid > transid) | |
339 | break; | |
340 | } | |
a4abeea4 | 341 | spin_unlock(&root->fs_info->trans_lock); |
46204592 SW |
342 | ret = -EINVAL; |
343 | if (!cur_trans) | |
a4abeea4 | 344 | goto out; /* bad transid */ |
46204592 SW |
345 | } else { |
346 | /* find newest transaction that is committing | committed */ | |
a4abeea4 | 347 | spin_lock(&root->fs_info->trans_lock); |
46204592 SW |
348 | list_for_each_entry_reverse(t, &root->fs_info->trans_list, |
349 | list) { | |
350 | if (t->in_commit) { | |
351 | if (t->commit_done) | |
3473f3c0 | 352 | break; |
46204592 | 353 | cur_trans = t; |
a4abeea4 | 354 | atomic_inc(&cur_trans->use_count); |
46204592 SW |
355 | break; |
356 | } | |
357 | } | |
a4abeea4 | 358 | spin_unlock(&root->fs_info->trans_lock); |
46204592 | 359 | if (!cur_trans) |
a4abeea4 | 360 | goto out; /* nothing committing|committed */ |
46204592 SW |
361 | } |
362 | ||
46204592 SW |
363 | wait_for_commit(root, cur_trans); |
364 | ||
46204592 SW |
365 | put_transaction(cur_trans); |
366 | ret = 0; | |
a4abeea4 | 367 | out: |
46204592 SW |
368 | return ret; |
369 | } | |
370 | ||
37d1aeee CM |
371 | void btrfs_throttle(struct btrfs_root *root) |
372 | { | |
a4abeea4 | 373 | if (!atomic_read(&root->fs_info->open_ioctl_trans)) |
9ca9ee09 | 374 | wait_current_trans(root); |
37d1aeee CM |
375 | } |
376 | ||
8929ecfa YZ |
377 | static int should_end_transaction(struct btrfs_trans_handle *trans, |
378 | struct btrfs_root *root) | |
379 | { | |
380 | int ret; | |
381 | ret = btrfs_block_rsv_check(trans, root, | |
382 | &root->fs_info->global_block_rsv, 0, 5); | |
383 | return ret ? 1 : 0; | |
384 | } | |
385 | ||
386 | int btrfs_should_end_transaction(struct btrfs_trans_handle *trans, | |
387 | struct btrfs_root *root) | |
388 | { | |
389 | struct btrfs_transaction *cur_trans = trans->transaction; | |
390 | int updates; | |
391 | ||
a4abeea4 | 392 | smp_mb(); |
8929ecfa YZ |
393 | if (cur_trans->blocked || cur_trans->delayed_refs.flushing) |
394 | return 1; | |
395 | ||
396 | updates = trans->delayed_ref_updates; | |
397 | trans->delayed_ref_updates = 0; | |
398 | if (updates) | |
399 | btrfs_run_delayed_refs(trans, root, updates); | |
400 | ||
401 | return should_end_transaction(trans, root); | |
402 | } | |
403 | ||
89ce8a63 | 404 | static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, |
0af3d00b | 405 | struct btrfs_root *root, int throttle, int lock) |
79154b1b | 406 | { |
8929ecfa | 407 | struct btrfs_transaction *cur_trans = trans->transaction; |
ab78c84d | 408 | struct btrfs_fs_info *info = root->fs_info; |
c3e69d58 CM |
409 | int count = 0; |
410 | ||
2a1eb461 JB |
411 | if (--trans->use_count) { |
412 | trans->block_rsv = trans->orig_rsv; | |
413 | return 0; | |
414 | } | |
415 | ||
c3e69d58 CM |
416 | while (count < 4) { |
417 | unsigned long cur = trans->delayed_ref_updates; | |
418 | trans->delayed_ref_updates = 0; | |
419 | if (cur && | |
420 | trans->transaction->delayed_refs.num_heads_ready > 64) { | |
421 | trans->delayed_ref_updates = 0; | |
b7ec40d7 CM |
422 | |
423 | /* | |
424 | * do a full flush if the transaction is trying | |
425 | * to close | |
426 | */ | |
427 | if (trans->transaction->delayed_refs.flushing) | |
428 | cur = 0; | |
c3e69d58 CM |
429 | btrfs_run_delayed_refs(trans, root, cur); |
430 | } else { | |
431 | break; | |
432 | } | |
433 | count++; | |
56bec294 CM |
434 | } |
435 | ||
a22285a6 YZ |
436 | btrfs_trans_release_metadata(trans, root); |
437 | ||
a4abeea4 JB |
438 | if (lock && !atomic_read(&root->fs_info->open_ioctl_trans) && |
439 | should_end_transaction(trans, root)) { | |
8929ecfa | 440 | trans->transaction->blocked = 1; |
a4abeea4 JB |
441 | smp_wmb(); |
442 | } | |
8929ecfa | 443 | |
0af3d00b | 444 | if (lock && cur_trans->blocked && !cur_trans->in_commit) { |
8929ecfa YZ |
445 | if (throttle) |
446 | return btrfs_commit_transaction(trans, root); | |
447 | else | |
448 | wake_up_process(info->transaction_kthread); | |
449 | } | |
450 | ||
8929ecfa | 451 | WARN_ON(cur_trans != info->running_transaction); |
13c5a93e JB |
452 | WARN_ON(atomic_read(&cur_trans->num_writers) < 1); |
453 | atomic_dec(&cur_trans->num_writers); | |
89ce8a63 | 454 | |
99d16cbc | 455 | smp_mb(); |
79154b1b CM |
456 | if (waitqueue_active(&cur_trans->writer_wait)) |
457 | wake_up(&cur_trans->writer_wait); | |
79154b1b | 458 | put_transaction(cur_trans); |
9ed74f2d JB |
459 | |
460 | if (current->journal_info == trans) | |
461 | current->journal_info = NULL; | |
d6025579 | 462 | memset(trans, 0, sizeof(*trans)); |
2c90e5d6 | 463 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
ab78c84d | 464 | |
24bbcf04 YZ |
465 | if (throttle) |
466 | btrfs_run_delayed_iputs(root); | |
467 | ||
79154b1b CM |
468 | return 0; |
469 | } | |
470 | ||
89ce8a63 CM |
471 | int btrfs_end_transaction(struct btrfs_trans_handle *trans, |
472 | struct btrfs_root *root) | |
473 | { | |
16cdcec7 MX |
474 | int ret; |
475 | ||
476 | ret = __btrfs_end_transaction(trans, root, 0, 1); | |
477 | if (ret) | |
478 | return ret; | |
479 | return 0; | |
89ce8a63 CM |
480 | } |
481 | ||
482 | int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans, | |
483 | struct btrfs_root *root) | |
484 | { | |
16cdcec7 MX |
485 | int ret; |
486 | ||
487 | ret = __btrfs_end_transaction(trans, root, 1, 1); | |
488 | if (ret) | |
489 | return ret; | |
490 | return 0; | |
0af3d00b JB |
491 | } |
492 | ||
493 | int btrfs_end_transaction_nolock(struct btrfs_trans_handle *trans, | |
494 | struct btrfs_root *root) | |
495 | { | |
16cdcec7 MX |
496 | int ret; |
497 | ||
498 | ret = __btrfs_end_transaction(trans, root, 0, 0); | |
499 | if (ret) | |
500 | return ret; | |
501 | return 0; | |
502 | } | |
503 | ||
504 | int btrfs_end_transaction_dmeta(struct btrfs_trans_handle *trans, | |
505 | struct btrfs_root *root) | |
506 | { | |
507 | return __btrfs_end_transaction(trans, root, 1, 1); | |
89ce8a63 CM |
508 | } |
509 | ||
d352ac68 CM |
510 | /* |
511 | * when btree blocks are allocated, they have some corresponding bits set for | |
512 | * them in one of two extent_io trees. This is used to make sure all of | |
690587d1 | 513 | * those extents are sent to disk but does not wait on them |
d352ac68 | 514 | */ |
690587d1 | 515 | int btrfs_write_marked_extents(struct btrfs_root *root, |
8cef4e16 | 516 | struct extent_io_tree *dirty_pages, int mark) |
79154b1b | 517 | { |
7c4452b9 | 518 | int ret; |
777e6bd7 | 519 | int err = 0; |
7c4452b9 CM |
520 | int werr = 0; |
521 | struct page *page; | |
7c4452b9 | 522 | struct inode *btree_inode = root->fs_info->btree_inode; |
777e6bd7 | 523 | u64 start = 0; |
5f39d397 CM |
524 | u64 end; |
525 | unsigned long index; | |
7c4452b9 | 526 | |
d397712b | 527 | while (1) { |
777e6bd7 | 528 | ret = find_first_extent_bit(dirty_pages, start, &start, &end, |
8cef4e16 | 529 | mark); |
5f39d397 | 530 | if (ret) |
7c4452b9 | 531 | break; |
d397712b | 532 | while (start <= end) { |
777e6bd7 CM |
533 | cond_resched(); |
534 | ||
5f39d397 | 535 | index = start >> PAGE_CACHE_SHIFT; |
35ebb934 | 536 | start = (u64)(index + 1) << PAGE_CACHE_SHIFT; |
4bef0848 | 537 | page = find_get_page(btree_inode->i_mapping, index); |
7c4452b9 CM |
538 | if (!page) |
539 | continue; | |
4bef0848 CM |
540 | |
541 | btree_lock_page_hook(page); | |
542 | if (!page->mapping) { | |
543 | unlock_page(page); | |
544 | page_cache_release(page); | |
545 | continue; | |
546 | } | |
547 | ||
6702ed49 CM |
548 | if (PageWriteback(page)) { |
549 | if (PageDirty(page)) | |
550 | wait_on_page_writeback(page); | |
551 | else { | |
552 | unlock_page(page); | |
553 | page_cache_release(page); | |
554 | continue; | |
555 | } | |
556 | } | |
7c4452b9 CM |
557 | err = write_one_page(page, 0); |
558 | if (err) | |
559 | werr = err; | |
560 | page_cache_release(page); | |
561 | } | |
562 | } | |
690587d1 CM |
563 | if (err) |
564 | werr = err; | |
565 | return werr; | |
566 | } | |
567 | ||
568 | /* | |
569 | * when btree blocks are allocated, they have some corresponding bits set for | |
570 | * them in one of two extent_io trees. This is used to make sure all of | |
571 | * those extents are on disk for transaction or log commit. We wait | |
572 | * on all the pages and clear them from the dirty pages state tree | |
573 | */ | |
574 | int btrfs_wait_marked_extents(struct btrfs_root *root, | |
8cef4e16 | 575 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 CM |
576 | { |
577 | int ret; | |
578 | int err = 0; | |
579 | int werr = 0; | |
580 | struct page *page; | |
581 | struct inode *btree_inode = root->fs_info->btree_inode; | |
582 | u64 start = 0; | |
583 | u64 end; | |
584 | unsigned long index; | |
585 | ||
d397712b | 586 | while (1) { |
8cef4e16 YZ |
587 | ret = find_first_extent_bit(dirty_pages, start, &start, &end, |
588 | mark); | |
777e6bd7 CM |
589 | if (ret) |
590 | break; | |
591 | ||
8cef4e16 | 592 | clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS); |
d397712b | 593 | while (start <= end) { |
777e6bd7 CM |
594 | index = start >> PAGE_CACHE_SHIFT; |
595 | start = (u64)(index + 1) << PAGE_CACHE_SHIFT; | |
596 | page = find_get_page(btree_inode->i_mapping, index); | |
597 | if (!page) | |
598 | continue; | |
599 | if (PageDirty(page)) { | |
4bef0848 CM |
600 | btree_lock_page_hook(page); |
601 | wait_on_page_writeback(page); | |
777e6bd7 CM |
602 | err = write_one_page(page, 0); |
603 | if (err) | |
604 | werr = err; | |
605 | } | |
105d931d | 606 | wait_on_page_writeback(page); |
777e6bd7 CM |
607 | page_cache_release(page); |
608 | cond_resched(); | |
609 | } | |
610 | } | |
7c4452b9 CM |
611 | if (err) |
612 | werr = err; | |
613 | return werr; | |
79154b1b CM |
614 | } |
615 | ||
690587d1 CM |
616 | /* |
617 | * when btree blocks are allocated, they have some corresponding bits set for | |
618 | * them in one of two extent_io trees. This is used to make sure all of | |
619 | * those extents are on disk for transaction or log commit | |
620 | */ | |
621 | int btrfs_write_and_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 ret2; | |
626 | ||
8cef4e16 YZ |
627 | ret = btrfs_write_marked_extents(root, dirty_pages, mark); |
628 | ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark); | |
690587d1 CM |
629 | return ret || ret2; |
630 | } | |
631 | ||
d0c803c4 CM |
632 | int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans, |
633 | struct btrfs_root *root) | |
634 | { | |
635 | if (!trans || !trans->transaction) { | |
636 | struct inode *btree_inode; | |
637 | btree_inode = root->fs_info->btree_inode; | |
638 | return filemap_write_and_wait(btree_inode->i_mapping); | |
639 | } | |
640 | return btrfs_write_and_wait_marked_extents(root, | |
8cef4e16 YZ |
641 | &trans->transaction->dirty_pages, |
642 | EXTENT_DIRTY); | |
d0c803c4 CM |
643 | } |
644 | ||
d352ac68 CM |
645 | /* |
646 | * this is used to update the root pointer in the tree of tree roots. | |
647 | * | |
648 | * But, in the case of the extent allocation tree, updating the root | |
649 | * pointer may allocate blocks which may change the root of the extent | |
650 | * allocation tree. | |
651 | * | |
652 | * So, this loops and repeats and makes sure the cowonly root didn't | |
653 | * change while the root pointer was being updated in the metadata. | |
654 | */ | |
0b86a832 CM |
655 | static int update_cowonly_root(struct btrfs_trans_handle *trans, |
656 | struct btrfs_root *root) | |
79154b1b CM |
657 | { |
658 | int ret; | |
0b86a832 | 659 | u64 old_root_bytenr; |
86b9f2ec | 660 | u64 old_root_used; |
0b86a832 | 661 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
79154b1b | 662 | |
86b9f2ec | 663 | old_root_used = btrfs_root_used(&root->root_item); |
0b86a832 | 664 | btrfs_write_dirty_block_groups(trans, root); |
56bec294 | 665 | |
d397712b | 666 | while (1) { |
0b86a832 | 667 | old_root_bytenr = btrfs_root_bytenr(&root->root_item); |
86b9f2ec YZ |
668 | if (old_root_bytenr == root->node->start && |
669 | old_root_used == btrfs_root_used(&root->root_item)) | |
79154b1b | 670 | break; |
87ef2bb4 | 671 | |
5d4f98a2 | 672 | btrfs_set_root_node(&root->root_item, root->node); |
79154b1b | 673 | ret = btrfs_update_root(trans, tree_root, |
0b86a832 CM |
674 | &root->root_key, |
675 | &root->root_item); | |
79154b1b | 676 | BUG_ON(ret); |
56bec294 | 677 | |
86b9f2ec | 678 | old_root_used = btrfs_root_used(&root->root_item); |
4a8c9a62 | 679 | ret = btrfs_write_dirty_block_groups(trans, root); |
56bec294 | 680 | BUG_ON(ret); |
0b86a832 | 681 | } |
276e680d YZ |
682 | |
683 | if (root != root->fs_info->extent_root) | |
684 | switch_commit_root(root); | |
685 | ||
0b86a832 CM |
686 | return 0; |
687 | } | |
688 | ||
d352ac68 CM |
689 | /* |
690 | * update all the cowonly tree roots on disk | |
691 | */ | |
5d4f98a2 YZ |
692 | static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans, |
693 | struct btrfs_root *root) | |
0b86a832 CM |
694 | { |
695 | struct btrfs_fs_info *fs_info = root->fs_info; | |
696 | struct list_head *next; | |
84234f3a | 697 | struct extent_buffer *eb; |
56bec294 | 698 | int ret; |
84234f3a | 699 | |
56bec294 CM |
700 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
701 | BUG_ON(ret); | |
87ef2bb4 | 702 | |
84234f3a | 703 | eb = btrfs_lock_root_node(fs_info->tree_root); |
9fa8cfe7 | 704 | btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 0, &eb); |
84234f3a YZ |
705 | btrfs_tree_unlock(eb); |
706 | free_extent_buffer(eb); | |
0b86a832 | 707 | |
56bec294 CM |
708 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
709 | BUG_ON(ret); | |
87ef2bb4 | 710 | |
d397712b | 711 | while (!list_empty(&fs_info->dirty_cowonly_roots)) { |
0b86a832 CM |
712 | next = fs_info->dirty_cowonly_roots.next; |
713 | list_del_init(next); | |
714 | root = list_entry(next, struct btrfs_root, dirty_list); | |
87ef2bb4 | 715 | |
0b86a832 | 716 | update_cowonly_root(trans, root); |
79154b1b | 717 | } |
276e680d YZ |
718 | |
719 | down_write(&fs_info->extent_commit_sem); | |
720 | switch_commit_root(fs_info->extent_root); | |
721 | up_write(&fs_info->extent_commit_sem); | |
722 | ||
79154b1b CM |
723 | return 0; |
724 | } | |
725 | ||
d352ac68 CM |
726 | /* |
727 | * dead roots are old snapshots that need to be deleted. This allocates | |
728 | * a dirty root struct and adds it into the list of dead roots that need to | |
729 | * be deleted | |
730 | */ | |
5d4f98a2 | 731 | int btrfs_add_dead_root(struct btrfs_root *root) |
5eda7b5e | 732 | { |
a4abeea4 | 733 | spin_lock(&root->fs_info->trans_lock); |
5d4f98a2 | 734 | list_add(&root->root_list, &root->fs_info->dead_roots); |
a4abeea4 | 735 | spin_unlock(&root->fs_info->trans_lock); |
5eda7b5e CM |
736 | return 0; |
737 | } | |
738 | ||
d352ac68 | 739 | /* |
5d4f98a2 | 740 | * update all the cowonly tree roots on disk |
d352ac68 | 741 | */ |
5d4f98a2 YZ |
742 | static noinline int commit_fs_roots(struct btrfs_trans_handle *trans, |
743 | struct btrfs_root *root) | |
0f7d52f4 | 744 | { |
0f7d52f4 | 745 | struct btrfs_root *gang[8]; |
5d4f98a2 | 746 | struct btrfs_fs_info *fs_info = root->fs_info; |
0f7d52f4 CM |
747 | int i; |
748 | int ret; | |
54aa1f4d CM |
749 | int err = 0; |
750 | ||
a4abeea4 | 751 | spin_lock(&fs_info->fs_roots_radix_lock); |
d397712b | 752 | while (1) { |
5d4f98a2 YZ |
753 | ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, |
754 | (void **)gang, 0, | |
0f7d52f4 CM |
755 | ARRAY_SIZE(gang), |
756 | BTRFS_ROOT_TRANS_TAG); | |
757 | if (ret == 0) | |
758 | break; | |
759 | for (i = 0; i < ret; i++) { | |
760 | root = gang[i]; | |
5d4f98a2 YZ |
761 | radix_tree_tag_clear(&fs_info->fs_roots_radix, |
762 | (unsigned long)root->root_key.objectid, | |
763 | BTRFS_ROOT_TRANS_TAG); | |
a4abeea4 | 764 | spin_unlock(&fs_info->fs_roots_radix_lock); |
31153d81 | 765 | |
e02119d5 | 766 | btrfs_free_log(trans, root); |
5d4f98a2 | 767 | btrfs_update_reloc_root(trans, root); |
d68fc57b | 768 | btrfs_orphan_commit_root(trans, root); |
bcc63abb | 769 | |
82d5902d LZ |
770 | btrfs_save_ino_cache(root, trans); |
771 | ||
978d910d | 772 | if (root->commit_root != root->node) { |
581bb050 | 773 | mutex_lock(&root->fs_commit_mutex); |
817d52f8 | 774 | switch_commit_root(root); |
581bb050 LZ |
775 | btrfs_unpin_free_ino(root); |
776 | mutex_unlock(&root->fs_commit_mutex); | |
777 | ||
978d910d YZ |
778 | btrfs_set_root_node(&root->root_item, |
779 | root->node); | |
780 | } | |
5d4f98a2 | 781 | |
5d4f98a2 | 782 | err = btrfs_update_root(trans, fs_info->tree_root, |
0f7d52f4 CM |
783 | &root->root_key, |
784 | &root->root_item); | |
a4abeea4 | 785 | spin_lock(&fs_info->fs_roots_radix_lock); |
54aa1f4d CM |
786 | if (err) |
787 | break; | |
0f7d52f4 CM |
788 | } |
789 | } | |
a4abeea4 | 790 | spin_unlock(&fs_info->fs_roots_radix_lock); |
54aa1f4d | 791 | return err; |
0f7d52f4 CM |
792 | } |
793 | ||
d352ac68 CM |
794 | /* |
795 | * defrag a given btree. If cacheonly == 1, this won't read from the disk, | |
796 | * otherwise every leaf in the btree is read and defragged. | |
797 | */ | |
e9d0b13b CM |
798 | int btrfs_defrag_root(struct btrfs_root *root, int cacheonly) |
799 | { | |
800 | struct btrfs_fs_info *info = root->fs_info; | |
e9d0b13b | 801 | struct btrfs_trans_handle *trans; |
8929ecfa | 802 | int ret; |
d3c2fdcf | 803 | unsigned long nr; |
e9d0b13b | 804 | |
8929ecfa | 805 | if (xchg(&root->defrag_running, 1)) |
e9d0b13b | 806 | return 0; |
8929ecfa | 807 | |
6b80053d | 808 | while (1) { |
8929ecfa YZ |
809 | trans = btrfs_start_transaction(root, 0); |
810 | if (IS_ERR(trans)) | |
811 | return PTR_ERR(trans); | |
812 | ||
e9d0b13b | 813 | ret = btrfs_defrag_leaves(trans, root, cacheonly); |
8929ecfa | 814 | |
d3c2fdcf | 815 | nr = trans->blocks_used; |
e9d0b13b | 816 | btrfs_end_transaction(trans, root); |
d3c2fdcf | 817 | btrfs_btree_balance_dirty(info->tree_root, nr); |
e9d0b13b CM |
818 | cond_resched(); |
819 | ||
7841cb28 | 820 | if (btrfs_fs_closing(root->fs_info) || ret != -EAGAIN) |
e9d0b13b CM |
821 | break; |
822 | } | |
823 | root->defrag_running = 0; | |
8929ecfa | 824 | return ret; |
e9d0b13b CM |
825 | } |
826 | ||
d352ac68 CM |
827 | /* |
828 | * new snapshots need to be created at a very specific time in the | |
829 | * transaction commit. This does the actual creation | |
830 | */ | |
80b6794d | 831 | static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, |
3063d29f CM |
832 | struct btrfs_fs_info *fs_info, |
833 | struct btrfs_pending_snapshot *pending) | |
834 | { | |
835 | struct btrfs_key key; | |
80b6794d | 836 | struct btrfs_root_item *new_root_item; |
3063d29f CM |
837 | struct btrfs_root *tree_root = fs_info->tree_root; |
838 | struct btrfs_root *root = pending->root; | |
6bdb72de SW |
839 | struct btrfs_root *parent_root; |
840 | struct inode *parent_inode; | |
6a912213 | 841 | struct dentry *parent; |
a22285a6 | 842 | struct dentry *dentry; |
3063d29f | 843 | struct extent_buffer *tmp; |
925baedd | 844 | struct extent_buffer *old; |
3063d29f | 845 | int ret; |
d68fc57b | 846 | u64 to_reserve = 0; |
6bdb72de | 847 | u64 index = 0; |
a22285a6 | 848 | u64 objectid; |
b83cc969 | 849 | u64 root_flags; |
3063d29f | 850 | |
80b6794d CM |
851 | new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS); |
852 | if (!new_root_item) { | |
a22285a6 | 853 | pending->error = -ENOMEM; |
80b6794d CM |
854 | goto fail; |
855 | } | |
a22285a6 | 856 | |
581bb050 | 857 | ret = btrfs_find_free_objectid(tree_root, &objectid); |
a22285a6 YZ |
858 | if (ret) { |
859 | pending->error = ret; | |
3063d29f | 860 | goto fail; |
a22285a6 | 861 | } |
3063d29f | 862 | |
3fd0a558 | 863 | btrfs_reloc_pre_snapshot(trans, pending, &to_reserve); |
d68fc57b YZ |
864 | btrfs_orphan_pre_snapshot(trans, pending, &to_reserve); |
865 | ||
866 | if (to_reserve > 0) { | |
867 | ret = btrfs_block_rsv_add(trans, root, &pending->block_rsv, | |
8bb8ab2e | 868 | to_reserve); |
d68fc57b YZ |
869 | if (ret) { |
870 | pending->error = ret; | |
871 | goto fail; | |
872 | } | |
873 | } | |
874 | ||
3063d29f | 875 | key.objectid = objectid; |
a22285a6 YZ |
876 | key.offset = (u64)-1; |
877 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3063d29f | 878 | |
a22285a6 | 879 | trans->block_rsv = &pending->block_rsv; |
3de4586c | 880 | |
a22285a6 | 881 | dentry = pending->dentry; |
6a912213 JB |
882 | parent = dget_parent(dentry); |
883 | parent_inode = parent->d_inode; | |
a22285a6 | 884 | parent_root = BTRFS_I(parent_inode)->root; |
a4abeea4 | 885 | btrfs_record_root_in_trans(trans, parent_root); |
a22285a6 | 886 | |
3063d29f CM |
887 | /* |
888 | * insert the directory item | |
889 | */ | |
3de4586c | 890 | ret = btrfs_set_inode_index(parent_inode, &index); |
6bdb72de | 891 | BUG_ON(ret); |
0660b5af | 892 | ret = btrfs_insert_dir_item(trans, parent_root, |
a22285a6 | 893 | dentry->d_name.name, dentry->d_name.len, |
16cdcec7 | 894 | parent_inode, &key, |
a22285a6 | 895 | BTRFS_FT_DIR, index); |
6bdb72de | 896 | BUG_ON(ret); |
0660b5af | 897 | |
a22285a6 YZ |
898 | btrfs_i_size_write(parent_inode, parent_inode->i_size + |
899 | dentry->d_name.len * 2); | |
52c26179 YZ |
900 | ret = btrfs_update_inode(trans, parent_root, parent_inode); |
901 | BUG_ON(ret); | |
902 | ||
a4abeea4 | 903 | btrfs_record_root_in_trans(trans, root); |
6bdb72de SW |
904 | btrfs_set_root_last_snapshot(&root->root_item, trans->transid); |
905 | memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); | |
08fe4db1 | 906 | btrfs_check_and_init_root_item(new_root_item); |
6bdb72de | 907 | |
b83cc969 LZ |
908 | root_flags = btrfs_root_flags(new_root_item); |
909 | if (pending->readonly) | |
910 | root_flags |= BTRFS_ROOT_SUBVOL_RDONLY; | |
911 | else | |
912 | root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY; | |
913 | btrfs_set_root_flags(new_root_item, root_flags); | |
914 | ||
6bdb72de SW |
915 | old = btrfs_lock_root_node(root); |
916 | btrfs_cow_block(trans, root, old, NULL, 0, &old); | |
917 | btrfs_set_lock_blocking(old); | |
918 | ||
919 | btrfs_copy_root(trans, root, old, &tmp, objectid); | |
920 | btrfs_tree_unlock(old); | |
921 | free_extent_buffer(old); | |
922 | ||
923 | btrfs_set_root_node(new_root_item, tmp); | |
a22285a6 YZ |
924 | /* record when the snapshot was created in key.offset */ |
925 | key.offset = trans->transid; | |
926 | ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); | |
6bdb72de SW |
927 | btrfs_tree_unlock(tmp); |
928 | free_extent_buffer(tmp); | |
a22285a6 | 929 | BUG_ON(ret); |
6bdb72de | 930 | |
a22285a6 YZ |
931 | /* |
932 | * insert root back/forward references | |
933 | */ | |
934 | ret = btrfs_add_root_ref(trans, tree_root, objectid, | |
0660b5af | 935 | parent_root->root_key.objectid, |
33345d01 | 936 | btrfs_ino(parent_inode), index, |
a22285a6 | 937 | dentry->d_name.name, dentry->d_name.len); |
0660b5af | 938 | BUG_ON(ret); |
6a912213 | 939 | dput(parent); |
0660b5af | 940 | |
a22285a6 YZ |
941 | key.offset = (u64)-1; |
942 | pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key); | |
943 | BUG_ON(IS_ERR(pending->snap)); | |
d68fc57b | 944 | |
3fd0a558 | 945 | btrfs_reloc_post_snapshot(trans, pending); |
d68fc57b | 946 | btrfs_orphan_post_snapshot(trans, pending); |
3063d29f | 947 | fail: |
6bdb72de | 948 | kfree(new_root_item); |
a22285a6 YZ |
949 | btrfs_block_rsv_release(root, &pending->block_rsv, (u64)-1); |
950 | return 0; | |
3063d29f CM |
951 | } |
952 | ||
d352ac68 CM |
953 | /* |
954 | * create all the snapshots we've scheduled for creation | |
955 | */ | |
80b6794d CM |
956 | static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans, |
957 | struct btrfs_fs_info *fs_info) | |
3de4586c CM |
958 | { |
959 | struct btrfs_pending_snapshot *pending; | |
960 | struct list_head *head = &trans->transaction->pending_snapshots; | |
3de4586c CM |
961 | int ret; |
962 | ||
c6e30871 | 963 | list_for_each_entry(pending, head, list) { |
16cdcec7 MX |
964 | /* |
965 | * We must deal with the delayed items before creating | |
966 | * snapshots, or we will create a snapthot with inconsistent | |
967 | * information. | |
968 | */ | |
969 | ret = btrfs_run_delayed_items(trans, fs_info->fs_root); | |
970 | BUG_ON(ret); | |
971 | ||
3de4586c CM |
972 | ret = create_pending_snapshot(trans, fs_info, pending); |
973 | BUG_ON(ret); | |
974 | } | |
975 | return 0; | |
976 | } | |
977 | ||
5d4f98a2 YZ |
978 | static void update_super_roots(struct btrfs_root *root) |
979 | { | |
980 | struct btrfs_root_item *root_item; | |
981 | struct btrfs_super_block *super; | |
982 | ||
983 | super = &root->fs_info->super_copy; | |
984 | ||
985 | root_item = &root->fs_info->chunk_root->root_item; | |
986 | super->chunk_root = root_item->bytenr; | |
987 | super->chunk_root_generation = root_item->generation; | |
988 | super->chunk_root_level = root_item->level; | |
989 | ||
990 | root_item = &root->fs_info->tree_root->root_item; | |
991 | super->root = root_item->bytenr; | |
992 | super->generation = root_item->generation; | |
993 | super->root_level = root_item->level; | |
0af3d00b JB |
994 | if (super->cache_generation != 0 || btrfs_test_opt(root, SPACE_CACHE)) |
995 | super->cache_generation = root_item->generation; | |
5d4f98a2 YZ |
996 | } |
997 | ||
f36f3042 CM |
998 | int btrfs_transaction_in_commit(struct btrfs_fs_info *info) |
999 | { | |
1000 | int ret = 0; | |
a4abeea4 | 1001 | spin_lock(&info->trans_lock); |
f36f3042 CM |
1002 | if (info->running_transaction) |
1003 | ret = info->running_transaction->in_commit; | |
a4abeea4 | 1004 | spin_unlock(&info->trans_lock); |
f36f3042 CM |
1005 | return ret; |
1006 | } | |
1007 | ||
8929ecfa YZ |
1008 | int btrfs_transaction_blocked(struct btrfs_fs_info *info) |
1009 | { | |
1010 | int ret = 0; | |
a4abeea4 | 1011 | spin_lock(&info->trans_lock); |
8929ecfa YZ |
1012 | if (info->running_transaction) |
1013 | ret = info->running_transaction->blocked; | |
a4abeea4 | 1014 | spin_unlock(&info->trans_lock); |
8929ecfa YZ |
1015 | return ret; |
1016 | } | |
1017 | ||
bb9c12c9 SW |
1018 | /* |
1019 | * wait for the current transaction commit to start and block subsequent | |
1020 | * transaction joins | |
1021 | */ | |
1022 | static void wait_current_trans_commit_start(struct btrfs_root *root, | |
1023 | struct btrfs_transaction *trans) | |
1024 | { | |
1025 | DEFINE_WAIT(wait); | |
1026 | ||
1027 | if (trans->in_commit) | |
1028 | return; | |
1029 | ||
1030 | while (1) { | |
1031 | prepare_to_wait(&root->fs_info->transaction_blocked_wait, &wait, | |
1032 | TASK_UNINTERRUPTIBLE); | |
1033 | if (trans->in_commit) { | |
1034 | finish_wait(&root->fs_info->transaction_blocked_wait, | |
1035 | &wait); | |
1036 | break; | |
1037 | } | |
bb9c12c9 | 1038 | schedule(); |
bb9c12c9 SW |
1039 | finish_wait(&root->fs_info->transaction_blocked_wait, &wait); |
1040 | } | |
1041 | } | |
1042 | ||
1043 | /* | |
1044 | * wait for the current transaction to start and then become unblocked. | |
1045 | * caller holds ref. | |
1046 | */ | |
1047 | static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root, | |
1048 | struct btrfs_transaction *trans) | |
1049 | { | |
1050 | DEFINE_WAIT(wait); | |
1051 | ||
1052 | if (trans->commit_done || (trans->in_commit && !trans->blocked)) | |
1053 | return; | |
1054 | ||
1055 | while (1) { | |
1056 | prepare_to_wait(&root->fs_info->transaction_wait, &wait, | |
1057 | TASK_UNINTERRUPTIBLE); | |
1058 | if (trans->commit_done || | |
1059 | (trans->in_commit && !trans->blocked)) { | |
1060 | finish_wait(&root->fs_info->transaction_wait, | |
1061 | &wait); | |
1062 | break; | |
1063 | } | |
bb9c12c9 | 1064 | schedule(); |
bb9c12c9 SW |
1065 | finish_wait(&root->fs_info->transaction_wait, |
1066 | &wait); | |
1067 | } | |
1068 | } | |
1069 | ||
1070 | /* | |
1071 | * commit transactions asynchronously. once btrfs_commit_transaction_async | |
1072 | * returns, any subsequent transaction will not be allowed to join. | |
1073 | */ | |
1074 | struct btrfs_async_commit { | |
1075 | struct btrfs_trans_handle *newtrans; | |
1076 | struct btrfs_root *root; | |
1077 | struct delayed_work work; | |
1078 | }; | |
1079 | ||
1080 | static void do_async_commit(struct work_struct *work) | |
1081 | { | |
1082 | struct btrfs_async_commit *ac = | |
1083 | container_of(work, struct btrfs_async_commit, work.work); | |
1084 | ||
1085 | btrfs_commit_transaction(ac->newtrans, ac->root); | |
1086 | kfree(ac); | |
1087 | } | |
1088 | ||
1089 | int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans, | |
1090 | struct btrfs_root *root, | |
1091 | int wait_for_unblock) | |
1092 | { | |
1093 | struct btrfs_async_commit *ac; | |
1094 | struct btrfs_transaction *cur_trans; | |
1095 | ||
1096 | ac = kmalloc(sizeof(*ac), GFP_NOFS); | |
db5b493a TI |
1097 | if (!ac) |
1098 | return -ENOMEM; | |
bb9c12c9 SW |
1099 | |
1100 | INIT_DELAYED_WORK(&ac->work, do_async_commit); | |
1101 | ac->root = root; | |
7a7eaa40 | 1102 | ac->newtrans = btrfs_join_transaction(root); |
3612b495 TI |
1103 | if (IS_ERR(ac->newtrans)) { |
1104 | int err = PTR_ERR(ac->newtrans); | |
1105 | kfree(ac); | |
1106 | return err; | |
1107 | } | |
bb9c12c9 SW |
1108 | |
1109 | /* take transaction reference */ | |
bb9c12c9 | 1110 | cur_trans = trans->transaction; |
13c5a93e | 1111 | atomic_inc(&cur_trans->use_count); |
bb9c12c9 SW |
1112 | |
1113 | btrfs_end_transaction(trans, root); | |
1114 | schedule_delayed_work(&ac->work, 0); | |
1115 | ||
1116 | /* wait for transaction to start and unblock */ | |
bb9c12c9 SW |
1117 | if (wait_for_unblock) |
1118 | wait_current_trans_commit_start_and_unblock(root, cur_trans); | |
1119 | else | |
1120 | wait_current_trans_commit_start(root, cur_trans); | |
bb9c12c9 | 1121 | |
38e88054 SW |
1122 | if (current->journal_info == trans) |
1123 | current->journal_info = NULL; | |
1124 | ||
1125 | put_transaction(cur_trans); | |
bb9c12c9 SW |
1126 | return 0; |
1127 | } | |
1128 | ||
1129 | /* | |
1130 | * btrfs_transaction state sequence: | |
1131 | * in_commit = 0, blocked = 0 (initial) | |
1132 | * in_commit = 1, blocked = 1 | |
1133 | * blocked = 0 | |
1134 | * commit_done = 1 | |
1135 | */ | |
79154b1b CM |
1136 | int btrfs_commit_transaction(struct btrfs_trans_handle *trans, |
1137 | struct btrfs_root *root) | |
1138 | { | |
15ee9bc7 | 1139 | unsigned long joined = 0; |
79154b1b | 1140 | struct btrfs_transaction *cur_trans; |
8fd17795 | 1141 | struct btrfs_transaction *prev_trans = NULL; |
79154b1b | 1142 | DEFINE_WAIT(wait); |
15ee9bc7 | 1143 | int ret; |
89573b9c CM |
1144 | int should_grow = 0; |
1145 | unsigned long now = get_seconds(); | |
dccae999 | 1146 | int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT); |
79154b1b | 1147 | |
5a3f23d5 CM |
1148 | btrfs_run_ordered_operations(root, 0); |
1149 | ||
56bec294 CM |
1150 | /* make a pass through all the delayed refs we have so far |
1151 | * any runnings procs may add more while we are here | |
1152 | */ | |
1153 | ret = btrfs_run_delayed_refs(trans, root, 0); | |
1154 | BUG_ON(ret); | |
1155 | ||
a22285a6 YZ |
1156 | btrfs_trans_release_metadata(trans, root); |
1157 | ||
b7ec40d7 | 1158 | cur_trans = trans->transaction; |
56bec294 CM |
1159 | /* |
1160 | * set the flushing flag so procs in this transaction have to | |
1161 | * start sending their work down. | |
1162 | */ | |
b7ec40d7 | 1163 | cur_trans->delayed_refs.flushing = 1; |
56bec294 | 1164 | |
c3e69d58 | 1165 | ret = btrfs_run_delayed_refs(trans, root, 0); |
56bec294 CM |
1166 | BUG_ON(ret); |
1167 | ||
a4abeea4 | 1168 | spin_lock(&cur_trans->commit_lock); |
b7ec40d7 | 1169 | if (cur_trans->in_commit) { |
a4abeea4 | 1170 | spin_unlock(&cur_trans->commit_lock); |
13c5a93e | 1171 | atomic_inc(&cur_trans->use_count); |
79154b1b | 1172 | btrfs_end_transaction(trans, root); |
ccd467d6 | 1173 | |
79154b1b CM |
1174 | ret = wait_for_commit(root, cur_trans); |
1175 | BUG_ON(ret); | |
15ee9bc7 | 1176 | |
79154b1b | 1177 | put_transaction(cur_trans); |
15ee9bc7 | 1178 | |
79154b1b CM |
1179 | return 0; |
1180 | } | |
4313b399 | 1181 | |
2c90e5d6 | 1182 | trans->transaction->in_commit = 1; |
f9295749 | 1183 | trans->transaction->blocked = 1; |
a4abeea4 | 1184 | spin_unlock(&cur_trans->commit_lock); |
bb9c12c9 SW |
1185 | wake_up(&root->fs_info->transaction_blocked_wait); |
1186 | ||
a4abeea4 | 1187 | spin_lock(&root->fs_info->trans_lock); |
ccd467d6 CM |
1188 | if (cur_trans->list.prev != &root->fs_info->trans_list) { |
1189 | prev_trans = list_entry(cur_trans->list.prev, | |
1190 | struct btrfs_transaction, list); | |
1191 | if (!prev_trans->commit_done) { | |
13c5a93e | 1192 | atomic_inc(&prev_trans->use_count); |
a4abeea4 | 1193 | spin_unlock(&root->fs_info->trans_lock); |
ccd467d6 CM |
1194 | |
1195 | wait_for_commit(root, prev_trans); | |
ccd467d6 | 1196 | |
15ee9bc7 | 1197 | put_transaction(prev_trans); |
a4abeea4 JB |
1198 | } else { |
1199 | spin_unlock(&root->fs_info->trans_lock); | |
ccd467d6 | 1200 | } |
a4abeea4 JB |
1201 | } else { |
1202 | spin_unlock(&root->fs_info->trans_lock); | |
ccd467d6 | 1203 | } |
15ee9bc7 | 1204 | |
89573b9c CM |
1205 | if (now < cur_trans->start_time || now - cur_trans->start_time < 1) |
1206 | should_grow = 1; | |
1207 | ||
15ee9bc7 | 1208 | do { |
7ea394f1 | 1209 | int snap_pending = 0; |
a4abeea4 | 1210 | |
15ee9bc7 | 1211 | joined = cur_trans->num_joined; |
7ea394f1 YZ |
1212 | if (!list_empty(&trans->transaction->pending_snapshots)) |
1213 | snap_pending = 1; | |
1214 | ||
2c90e5d6 | 1215 | WARN_ON(cur_trans != trans->transaction); |
15ee9bc7 | 1216 | |
0bdb1db2 | 1217 | if (flush_on_commit || snap_pending) { |
24bbcf04 YZ |
1218 | btrfs_start_delalloc_inodes(root, 1); |
1219 | ret = btrfs_wait_ordered_extents(root, 0, 1); | |
ebecd3d9 | 1220 | BUG_ON(ret); |
7ea394f1 YZ |
1221 | } |
1222 | ||
16cdcec7 MX |
1223 | ret = btrfs_run_delayed_items(trans, root); |
1224 | BUG_ON(ret); | |
1225 | ||
5a3f23d5 CM |
1226 | /* |
1227 | * rename don't use btrfs_join_transaction, so, once we | |
1228 | * set the transaction to blocked above, we aren't going | |
1229 | * to get any new ordered operations. We can safely run | |
1230 | * it here and no for sure that nothing new will be added | |
1231 | * to the list | |
1232 | */ | |
1233 | btrfs_run_ordered_operations(root, 1); | |
1234 | ||
ed3b3d31 CM |
1235 | prepare_to_wait(&cur_trans->writer_wait, &wait, |
1236 | TASK_UNINTERRUPTIBLE); | |
1237 | ||
13c5a93e | 1238 | if (atomic_read(&cur_trans->num_writers) > 1) |
99d16cbc SW |
1239 | schedule_timeout(MAX_SCHEDULE_TIMEOUT); |
1240 | else if (should_grow) | |
1241 | schedule_timeout(1); | |
15ee9bc7 | 1242 | |
15ee9bc7 | 1243 | finish_wait(&cur_trans->writer_wait, &wait); |
a4abeea4 JB |
1244 | spin_lock(&root->fs_info->trans_lock); |
1245 | root->fs_info->trans_no_join = 1; | |
1246 | spin_unlock(&root->fs_info->trans_lock); | |
13c5a93e | 1247 | } while (atomic_read(&cur_trans->num_writers) > 1 || |
89573b9c | 1248 | (should_grow && cur_trans->num_joined != joined)); |
15ee9bc7 | 1249 | |
3063d29f CM |
1250 | ret = create_pending_snapshots(trans, root->fs_info); |
1251 | BUG_ON(ret); | |
1252 | ||
16cdcec7 MX |
1253 | ret = btrfs_run_delayed_items(trans, root); |
1254 | BUG_ON(ret); | |
1255 | ||
56bec294 CM |
1256 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
1257 | BUG_ON(ret); | |
1258 | ||
2c90e5d6 | 1259 | WARN_ON(cur_trans != trans->transaction); |
dc17ff8f | 1260 | |
a2de733c | 1261 | btrfs_scrub_pause(root); |
e02119d5 CM |
1262 | /* btrfs_commit_tree_roots is responsible for getting the |
1263 | * various roots consistent with each other. Every pointer | |
1264 | * in the tree of tree roots has to point to the most up to date | |
1265 | * root for every subvolume and other tree. So, we have to keep | |
1266 | * the tree logging code from jumping in and changing any | |
1267 | * of the trees. | |
1268 | * | |
1269 | * At this point in the commit, there can't be any tree-log | |
1270 | * writers, but a little lower down we drop the trans mutex | |
1271 | * and let new people in. By holding the tree_log_mutex | |
1272 | * from now until after the super is written, we avoid races | |
1273 | * with the tree-log code. | |
1274 | */ | |
1275 | mutex_lock(&root->fs_info->tree_log_mutex); | |
1276 | ||
5d4f98a2 | 1277 | ret = commit_fs_roots(trans, root); |
54aa1f4d CM |
1278 | BUG_ON(ret); |
1279 | ||
5d4f98a2 | 1280 | /* commit_fs_roots gets rid of all the tree log roots, it is now |
e02119d5 CM |
1281 | * safe to free the root of tree log roots |
1282 | */ | |
1283 | btrfs_free_log_root_tree(trans, root->fs_info); | |
1284 | ||
5d4f98a2 | 1285 | ret = commit_cowonly_roots(trans, root); |
79154b1b | 1286 | BUG_ON(ret); |
54aa1f4d | 1287 | |
11833d66 YZ |
1288 | btrfs_prepare_extent_commit(trans, root); |
1289 | ||
78fae27e | 1290 | cur_trans = root->fs_info->running_transaction; |
5d4f98a2 YZ |
1291 | |
1292 | btrfs_set_root_node(&root->fs_info->tree_root->root_item, | |
1293 | root->fs_info->tree_root->node); | |
817d52f8 | 1294 | switch_commit_root(root->fs_info->tree_root); |
5d4f98a2 YZ |
1295 | |
1296 | btrfs_set_root_node(&root->fs_info->chunk_root->root_item, | |
1297 | root->fs_info->chunk_root->node); | |
817d52f8 | 1298 | switch_commit_root(root->fs_info->chunk_root); |
5d4f98a2 YZ |
1299 | |
1300 | update_super_roots(root); | |
e02119d5 CM |
1301 | |
1302 | if (!root->fs_info->log_root_recovering) { | |
1303 | btrfs_set_super_log_root(&root->fs_info->super_copy, 0); | |
1304 | btrfs_set_super_log_root_level(&root->fs_info->super_copy, 0); | |
1305 | } | |
1306 | ||
a061fc8d CM |
1307 | memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy, |
1308 | sizeof(root->fs_info->super_copy)); | |
ccd467d6 | 1309 | |
f9295749 | 1310 | trans->transaction->blocked = 0; |
a4abeea4 JB |
1311 | spin_lock(&root->fs_info->trans_lock); |
1312 | root->fs_info->running_transaction = NULL; | |
1313 | root->fs_info->trans_no_join = 0; | |
1314 | spin_unlock(&root->fs_info->trans_lock); | |
b7ec40d7 | 1315 | |
f9295749 | 1316 | wake_up(&root->fs_info->transaction_wait); |
e6dcd2dc | 1317 | |
79154b1b CM |
1318 | ret = btrfs_write_and_wait_transaction(trans, root); |
1319 | BUG_ON(ret); | |
a512bbf8 | 1320 | write_ctree_super(trans, root, 0); |
4313b399 | 1321 | |
e02119d5 CM |
1322 | /* |
1323 | * the super is written, we can safely allow the tree-loggers | |
1324 | * to go about their business | |
1325 | */ | |
1326 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
1327 | ||
11833d66 | 1328 | btrfs_finish_extent_commit(trans, root); |
4313b399 | 1329 | |
2c90e5d6 | 1330 | cur_trans->commit_done = 1; |
b7ec40d7 | 1331 | |
15ee9bc7 | 1332 | root->fs_info->last_trans_committed = cur_trans->transid; |
817d52f8 | 1333 | |
2c90e5d6 | 1334 | wake_up(&cur_trans->commit_wait); |
3de4586c | 1335 | |
a4abeea4 | 1336 | spin_lock(&root->fs_info->trans_lock); |
13c5a93e | 1337 | list_del_init(&cur_trans->list); |
a4abeea4 JB |
1338 | spin_unlock(&root->fs_info->trans_lock); |
1339 | ||
78fae27e | 1340 | put_transaction(cur_trans); |
79154b1b | 1341 | put_transaction(cur_trans); |
58176a96 | 1342 | |
1abe9b8a | 1343 | trace_btrfs_transaction_commit(root); |
1344 | ||
a2de733c AJ |
1345 | btrfs_scrub_continue(root); |
1346 | ||
9ed74f2d JB |
1347 | if (current->journal_info == trans) |
1348 | current->journal_info = NULL; | |
1349 | ||
2c90e5d6 | 1350 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
24bbcf04 YZ |
1351 | |
1352 | if (current != root->fs_info->transaction_kthread) | |
1353 | btrfs_run_delayed_iputs(root); | |
1354 | ||
79154b1b CM |
1355 | return ret; |
1356 | } | |
1357 | ||
d352ac68 CM |
1358 | /* |
1359 | * interface function to delete all the snapshots we have scheduled for deletion | |
1360 | */ | |
e9d0b13b CM |
1361 | int btrfs_clean_old_snapshots(struct btrfs_root *root) |
1362 | { | |
5d4f98a2 YZ |
1363 | LIST_HEAD(list); |
1364 | struct btrfs_fs_info *fs_info = root->fs_info; | |
1365 | ||
a4abeea4 | 1366 | spin_lock(&fs_info->trans_lock); |
5d4f98a2 | 1367 | list_splice_init(&fs_info->dead_roots, &list); |
a4abeea4 | 1368 | spin_unlock(&fs_info->trans_lock); |
e9d0b13b | 1369 | |
5d4f98a2 YZ |
1370 | while (!list_empty(&list)) { |
1371 | root = list_entry(list.next, struct btrfs_root, root_list); | |
76dda93c YZ |
1372 | list_del(&root->root_list); |
1373 | ||
16cdcec7 MX |
1374 | btrfs_kill_all_delayed_nodes(root); |
1375 | ||
76dda93c YZ |
1376 | if (btrfs_header_backref_rev(root->node) < |
1377 | BTRFS_MIXED_BACKREF_REV) | |
3fd0a558 | 1378 | btrfs_drop_snapshot(root, NULL, 0); |
76dda93c | 1379 | else |
3fd0a558 | 1380 | btrfs_drop_snapshot(root, NULL, 1); |
e9d0b13b CM |
1381 | } |
1382 | return 0; | |
1383 | } |