Commit | Line | Data |
---|---|---|
6cbd5570 CM |
1 | /* |
2 | * Copyright (C) 2007 Oracle. All rights reserved. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public | |
6 | * License v2 as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, | |
9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public | |
14 | * License along with this program; if not, write to the | |
15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
16 | * Boston, MA 021110-1307, USA. | |
17 | */ | |
18 | ||
79154b1b | 19 | #include <linux/fs.h> |
5a0e3ad6 | 20 | #include <linux/slab.h> |
34088780 | 21 | #include <linux/sched.h> |
d3c2fdcf | 22 | #include <linux/writeback.h> |
5f39d397 | 23 | #include <linux/pagemap.h> |
5f2cc086 | 24 | #include <linux/blkdev.h> |
79154b1b CM |
25 | #include "ctree.h" |
26 | #include "disk-io.h" | |
27 | #include "transaction.h" | |
925baedd | 28 | #include "locking.h" |
e02119d5 | 29 | #include "tree-log.h" |
79154b1b | 30 | |
0f7d52f4 CM |
31 | #define BTRFS_ROOT_TRANS_TAG 0 |
32 | ||
80b6794d | 33 | static noinline void put_transaction(struct btrfs_transaction *transaction) |
79154b1b | 34 | { |
13c5a93e JB |
35 | WARN_ON(atomic_read(&transaction->use_count) == 0); |
36 | if (atomic_dec_and_test(&transaction->use_count)) { | |
2c90e5d6 CM |
37 | memset(transaction, 0, sizeof(*transaction)); |
38 | kmem_cache_free(btrfs_transaction_cachep, transaction); | |
78fae27e | 39 | } |
79154b1b CM |
40 | } |
41 | ||
817d52f8 JB |
42 | static noinline void switch_commit_root(struct btrfs_root *root) |
43 | { | |
817d52f8 JB |
44 | free_extent_buffer(root->commit_root); |
45 | root->commit_root = btrfs_root_node(root); | |
817d52f8 JB |
46 | } |
47 | ||
d352ac68 CM |
48 | /* |
49 | * either allocate a new transaction or hop into the existing one | |
50 | */ | |
80b6794d | 51 | static noinline int join_transaction(struct btrfs_root *root) |
79154b1b CM |
52 | { |
53 | struct btrfs_transaction *cur_trans; | |
54 | cur_trans = root->fs_info->running_transaction; | |
55 | if (!cur_trans) { | |
2c90e5d6 CM |
56 | cur_trans = kmem_cache_alloc(btrfs_transaction_cachep, |
57 | GFP_NOFS); | |
db5b493a TI |
58 | if (!cur_trans) |
59 | return -ENOMEM; | |
0f7d52f4 | 60 | root->fs_info->generation++; |
13c5a93e | 61 | atomic_set(&cur_trans->num_writers, 1); |
15ee9bc7 | 62 | cur_trans->num_joined = 0; |
0f7d52f4 | 63 | cur_trans->transid = root->fs_info->generation; |
79154b1b CM |
64 | init_waitqueue_head(&cur_trans->writer_wait); |
65 | init_waitqueue_head(&cur_trans->commit_wait); | |
66 | cur_trans->in_commit = 0; | |
f9295749 | 67 | cur_trans->blocked = 0; |
13c5a93e | 68 | atomic_set(&cur_trans->use_count, 1); |
79154b1b | 69 | cur_trans->commit_done = 0; |
08607c1b | 70 | cur_trans->start_time = get_seconds(); |
56bec294 | 71 | |
6bef4d31 | 72 | cur_trans->delayed_refs.root = RB_ROOT; |
56bec294 | 73 | cur_trans->delayed_refs.num_entries = 0; |
c3e69d58 CM |
74 | cur_trans->delayed_refs.num_heads_ready = 0; |
75 | cur_trans->delayed_refs.num_heads = 0; | |
56bec294 | 76 | cur_trans->delayed_refs.flushing = 0; |
c3e69d58 | 77 | cur_trans->delayed_refs.run_delayed_start = 0; |
56bec294 CM |
78 | spin_lock_init(&cur_trans->delayed_refs.lock); |
79 | ||
3063d29f | 80 | INIT_LIST_HEAD(&cur_trans->pending_snapshots); |
8fd17795 | 81 | list_add_tail(&cur_trans->list, &root->fs_info->trans_list); |
d1310b2e | 82 | extent_io_tree_init(&cur_trans->dirty_pages, |
5f39d397 CM |
83 | root->fs_info->btree_inode->i_mapping, |
84 | GFP_NOFS); | |
48ec2cf8 CM |
85 | spin_lock(&root->fs_info->new_trans_lock); |
86 | root->fs_info->running_transaction = cur_trans; | |
87 | spin_unlock(&root->fs_info->new_trans_lock); | |
15ee9bc7 | 88 | } else { |
13c5a93e | 89 | atomic_inc(&cur_trans->num_writers); |
15ee9bc7 | 90 | cur_trans->num_joined++; |
79154b1b | 91 | } |
15ee9bc7 | 92 | |
79154b1b CM |
93 | return 0; |
94 | } | |
95 | ||
d352ac68 | 96 | /* |
d397712b CM |
97 | * this does all the record keeping required to make sure that a reference |
98 | * counted root is properly recorded in a given transaction. This is required | |
99 | * to make sure the old root from before we joined the transaction is deleted | |
100 | * when the transaction commits | |
d352ac68 | 101 | */ |
5d4f98a2 YZ |
102 | static noinline int record_root_in_trans(struct btrfs_trans_handle *trans, |
103 | struct btrfs_root *root) | |
6702ed49 | 104 | { |
5d4f98a2 | 105 | if (root->ref_cows && root->last_trans < trans->transid) { |
6702ed49 | 106 | WARN_ON(root == root->fs_info->extent_root); |
5d4f98a2 YZ |
107 | WARN_ON(root->commit_root != root->node); |
108 | ||
109 | radix_tree_tag_set(&root->fs_info->fs_roots_radix, | |
110 | (unsigned long)root->root_key.objectid, | |
111 | BTRFS_ROOT_TRANS_TAG); | |
112 | root->last_trans = trans->transid; | |
113 | btrfs_init_reloc_root(trans, root); | |
114 | } | |
115 | return 0; | |
116 | } | |
bcc63abb | 117 | |
5d4f98a2 YZ |
118 | int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, |
119 | struct btrfs_root *root) | |
120 | { | |
121 | if (!root->ref_cows) | |
122 | return 0; | |
bcc63abb | 123 | |
5d4f98a2 YZ |
124 | mutex_lock(&root->fs_info->trans_mutex); |
125 | if (root->last_trans == trans->transid) { | |
126 | mutex_unlock(&root->fs_info->trans_mutex); | |
127 | return 0; | |
6702ed49 | 128 | } |
5d4f98a2 YZ |
129 | |
130 | record_root_in_trans(trans, root); | |
131 | mutex_unlock(&root->fs_info->trans_mutex); | |
6702ed49 CM |
132 | return 0; |
133 | } | |
134 | ||
d352ac68 CM |
135 | /* wait for commit against the current transaction to become unblocked |
136 | * when this is done, it is safe to start a new transaction, but the current | |
137 | * transaction might not be fully on disk. | |
138 | */ | |
37d1aeee | 139 | static void wait_current_trans(struct btrfs_root *root) |
79154b1b | 140 | { |
f9295749 | 141 | struct btrfs_transaction *cur_trans; |
79154b1b | 142 | |
f9295749 | 143 | cur_trans = root->fs_info->running_transaction; |
37d1aeee | 144 | if (cur_trans && cur_trans->blocked) { |
f9295749 | 145 | DEFINE_WAIT(wait); |
13c5a93e | 146 | atomic_inc(&cur_trans->use_count); |
d397712b | 147 | while (1) { |
f9295749 CM |
148 | prepare_to_wait(&root->fs_info->transaction_wait, &wait, |
149 | TASK_UNINTERRUPTIBLE); | |
471fa17d | 150 | if (!cur_trans->blocked) |
f9295749 | 151 | break; |
471fa17d ZL |
152 | mutex_unlock(&root->fs_info->trans_mutex); |
153 | schedule(); | |
154 | mutex_lock(&root->fs_info->trans_mutex); | |
f9295749 | 155 | } |
471fa17d | 156 | finish_wait(&root->fs_info->transaction_wait, &wait); |
f9295749 CM |
157 | put_transaction(cur_trans); |
158 | } | |
37d1aeee CM |
159 | } |
160 | ||
249ac1e5 JB |
161 | enum btrfs_trans_type { |
162 | TRANS_START, | |
163 | TRANS_JOIN, | |
164 | TRANS_USERSPACE, | |
0af3d00b | 165 | TRANS_JOIN_NOLOCK, |
249ac1e5 JB |
166 | }; |
167 | ||
a22285a6 YZ |
168 | static int may_wait_transaction(struct btrfs_root *root, int type) |
169 | { | |
170 | if (!root->fs_info->log_root_recovering && | |
171 | ((type == TRANS_START && !root->fs_info->open_ioctl_trans) || | |
172 | type == TRANS_USERSPACE)) | |
173 | return 1; | |
174 | return 0; | |
175 | } | |
176 | ||
e02119d5 | 177 | static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root, |
a22285a6 | 178 | u64 num_items, int type) |
37d1aeee | 179 | { |
a22285a6 YZ |
180 | struct btrfs_trans_handle *h; |
181 | struct btrfs_transaction *cur_trans; | |
06d5a589 | 182 | int retries = 0; |
37d1aeee | 183 | int ret; |
acce952b | 184 | |
185 | if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) | |
186 | return ERR_PTR(-EROFS); | |
2a1eb461 JB |
187 | |
188 | if (current->journal_info) { | |
189 | WARN_ON(type != TRANS_JOIN && type != TRANS_JOIN_NOLOCK); | |
190 | h = current->journal_info; | |
191 | h->use_count++; | |
192 | h->orig_rsv = h->block_rsv; | |
193 | h->block_rsv = NULL; | |
194 | goto got_it; | |
195 | } | |
a22285a6 YZ |
196 | again: |
197 | h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS); | |
198 | if (!h) | |
199 | return ERR_PTR(-ENOMEM); | |
37d1aeee | 200 | |
0af3d00b JB |
201 | if (type != TRANS_JOIN_NOLOCK) |
202 | mutex_lock(&root->fs_info->trans_mutex); | |
a22285a6 | 203 | if (may_wait_transaction(root, type)) |
37d1aeee | 204 | wait_current_trans(root); |
a22285a6 | 205 | |
79154b1b | 206 | ret = join_transaction(root); |
db5b493a | 207 | if (ret < 0) { |
6e8df2ae | 208 | kmem_cache_free(btrfs_trans_handle_cachep, h); |
db5b493a TI |
209 | if (type != TRANS_JOIN_NOLOCK) |
210 | mutex_unlock(&root->fs_info->trans_mutex); | |
211 | return ERR_PTR(ret); | |
212 | } | |
0f7d52f4 | 213 | |
a22285a6 | 214 | cur_trans = root->fs_info->running_transaction; |
13c5a93e | 215 | atomic_inc(&cur_trans->use_count); |
0af3d00b JB |
216 | if (type != TRANS_JOIN_NOLOCK) |
217 | mutex_unlock(&root->fs_info->trans_mutex); | |
a22285a6 YZ |
218 | |
219 | h->transid = cur_trans->transid; | |
220 | h->transaction = cur_trans; | |
79154b1b | 221 | h->blocks_used = 0; |
d2fb3437 | 222 | h->block_group = 0; |
a22285a6 | 223 | h->bytes_reserved = 0; |
56bec294 | 224 | h->delayed_ref_updates = 0; |
2a1eb461 | 225 | h->use_count = 1; |
f0486c68 | 226 | h->block_rsv = NULL; |
2a1eb461 | 227 | h->orig_rsv = NULL; |
b7ec40d7 | 228 | |
a22285a6 YZ |
229 | smp_mb(); |
230 | if (cur_trans->blocked && may_wait_transaction(root, type)) { | |
231 | btrfs_commit_transaction(h, root); | |
232 | goto again; | |
233 | } | |
234 | ||
235 | if (num_items > 0) { | |
8bb8ab2e | 236 | ret = btrfs_trans_reserve_metadata(h, root, num_items); |
06d5a589 JB |
237 | if (ret == -EAGAIN && !retries) { |
238 | retries++; | |
a22285a6 YZ |
239 | btrfs_commit_transaction(h, root); |
240 | goto again; | |
06d5a589 JB |
241 | } else if (ret == -EAGAIN) { |
242 | /* | |
243 | * We have already retried and got EAGAIN, so really we | |
244 | * don't have space, so set ret to -ENOSPC. | |
245 | */ | |
246 | ret = -ENOSPC; | |
a22285a6 | 247 | } |
06d5a589 | 248 | |
a22285a6 YZ |
249 | if (ret < 0) { |
250 | btrfs_end_transaction(h, root); | |
251 | return ERR_PTR(ret); | |
252 | } | |
253 | } | |
9ed74f2d | 254 | |
2a1eb461 | 255 | got_it: |
0af3d00b JB |
256 | if (type != TRANS_JOIN_NOLOCK) |
257 | mutex_lock(&root->fs_info->trans_mutex); | |
5d4f98a2 | 258 | record_root_in_trans(h, root); |
0af3d00b JB |
259 | if (type != TRANS_JOIN_NOLOCK) |
260 | mutex_unlock(&root->fs_info->trans_mutex); | |
a22285a6 YZ |
261 | |
262 | if (!current->journal_info && type != TRANS_USERSPACE) | |
263 | current->journal_info = h; | |
79154b1b CM |
264 | return h; |
265 | } | |
266 | ||
f9295749 | 267 | struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, |
a22285a6 | 268 | int num_items) |
f9295749 | 269 | { |
a22285a6 | 270 | return start_transaction(root, num_items, TRANS_START); |
f9295749 | 271 | } |
7a7eaa40 | 272 | struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root) |
f9295749 | 273 | { |
a22285a6 | 274 | return start_transaction(root, 0, TRANS_JOIN); |
f9295749 CM |
275 | } |
276 | ||
7a7eaa40 | 277 | struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root) |
0af3d00b JB |
278 | { |
279 | return start_transaction(root, 0, TRANS_JOIN_NOLOCK); | |
280 | } | |
281 | ||
7a7eaa40 | 282 | struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root) |
9ca9ee09 | 283 | { |
7a7eaa40 | 284 | return start_transaction(root, 0, TRANS_USERSPACE); |
9ca9ee09 SW |
285 | } |
286 | ||
d352ac68 | 287 | /* wait for a transaction commit to be fully complete */ |
89ce8a63 CM |
288 | static noinline int wait_for_commit(struct btrfs_root *root, |
289 | struct btrfs_transaction *commit) | |
290 | { | |
291 | DEFINE_WAIT(wait); | |
292 | mutex_lock(&root->fs_info->trans_mutex); | |
d397712b | 293 | while (!commit->commit_done) { |
89ce8a63 CM |
294 | prepare_to_wait(&commit->commit_wait, &wait, |
295 | TASK_UNINTERRUPTIBLE); | |
296 | if (commit->commit_done) | |
297 | break; | |
298 | mutex_unlock(&root->fs_info->trans_mutex); | |
299 | schedule(); | |
300 | mutex_lock(&root->fs_info->trans_mutex); | |
301 | } | |
302 | mutex_unlock(&root->fs_info->trans_mutex); | |
303 | finish_wait(&commit->commit_wait, &wait); | |
304 | return 0; | |
305 | } | |
306 | ||
46204592 SW |
307 | int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid) |
308 | { | |
309 | struct btrfs_transaction *cur_trans = NULL, *t; | |
310 | int ret; | |
311 | ||
312 | mutex_lock(&root->fs_info->trans_mutex); | |
313 | ||
314 | ret = 0; | |
315 | if (transid) { | |
316 | if (transid <= root->fs_info->last_trans_committed) | |
317 | goto out_unlock; | |
318 | ||
319 | /* find specified transaction */ | |
320 | list_for_each_entry(t, &root->fs_info->trans_list, list) { | |
321 | if (t->transid == transid) { | |
322 | cur_trans = t; | |
323 | break; | |
324 | } | |
325 | if (t->transid > transid) | |
326 | break; | |
327 | } | |
328 | ret = -EINVAL; | |
329 | if (!cur_trans) | |
330 | goto out_unlock; /* bad transid */ | |
331 | } else { | |
332 | /* find newest transaction that is committing | committed */ | |
333 | list_for_each_entry_reverse(t, &root->fs_info->trans_list, | |
334 | list) { | |
335 | if (t->in_commit) { | |
336 | if (t->commit_done) | |
337 | goto out_unlock; | |
338 | cur_trans = t; | |
339 | break; | |
340 | } | |
341 | } | |
342 | if (!cur_trans) | |
343 | goto out_unlock; /* nothing committing|committed */ | |
344 | } | |
345 | ||
13c5a93e | 346 | atomic_inc(&cur_trans->use_count); |
46204592 SW |
347 | mutex_unlock(&root->fs_info->trans_mutex); |
348 | ||
349 | wait_for_commit(root, cur_trans); | |
350 | ||
351 | mutex_lock(&root->fs_info->trans_mutex); | |
352 | put_transaction(cur_trans); | |
353 | ret = 0; | |
354 | out_unlock: | |
355 | mutex_unlock(&root->fs_info->trans_mutex); | |
356 | return ret; | |
357 | } | |
358 | ||
5d4f98a2 | 359 | #if 0 |
d352ac68 | 360 | /* |
d397712b CM |
361 | * rate limit against the drop_snapshot code. This helps to slow down new |
362 | * operations if the drop_snapshot code isn't able to keep up. | |
d352ac68 | 363 | */ |
37d1aeee | 364 | static void throttle_on_drops(struct btrfs_root *root) |
ab78c84d CM |
365 | { |
366 | struct btrfs_fs_info *info = root->fs_info; | |
2dd3e67b | 367 | int harder_count = 0; |
ab78c84d | 368 | |
2dd3e67b | 369 | harder: |
ab78c84d CM |
370 | if (atomic_read(&info->throttles)) { |
371 | DEFINE_WAIT(wait); | |
372 | int thr; | |
ab78c84d CM |
373 | thr = atomic_read(&info->throttle_gen); |
374 | ||
375 | do { | |
376 | prepare_to_wait(&info->transaction_throttle, | |
377 | &wait, TASK_UNINTERRUPTIBLE); | |
378 | if (!atomic_read(&info->throttles)) { | |
379 | finish_wait(&info->transaction_throttle, &wait); | |
380 | break; | |
381 | } | |
382 | schedule(); | |
383 | finish_wait(&info->transaction_throttle, &wait); | |
384 | } while (thr == atomic_read(&info->throttle_gen)); | |
2dd3e67b CM |
385 | harder_count++; |
386 | ||
387 | if (root->fs_info->total_ref_cache_size > 1 * 1024 * 1024 && | |
388 | harder_count < 2) | |
389 | goto harder; | |
390 | ||
391 | if (root->fs_info->total_ref_cache_size > 5 * 1024 * 1024 && | |
392 | harder_count < 10) | |
393 | goto harder; | |
394 | ||
395 | if (root->fs_info->total_ref_cache_size > 10 * 1024 * 1024 && | |
396 | harder_count < 20) | |
397 | goto harder; | |
ab78c84d CM |
398 | } |
399 | } | |
5d4f98a2 | 400 | #endif |
ab78c84d | 401 | |
37d1aeee CM |
402 | void btrfs_throttle(struct btrfs_root *root) |
403 | { | |
404 | mutex_lock(&root->fs_info->trans_mutex); | |
9ca9ee09 SW |
405 | if (!root->fs_info->open_ioctl_trans) |
406 | wait_current_trans(root); | |
37d1aeee | 407 | mutex_unlock(&root->fs_info->trans_mutex); |
37d1aeee CM |
408 | } |
409 | ||
8929ecfa YZ |
410 | static int should_end_transaction(struct btrfs_trans_handle *trans, |
411 | struct btrfs_root *root) | |
412 | { | |
413 | int ret; | |
414 | ret = btrfs_block_rsv_check(trans, root, | |
415 | &root->fs_info->global_block_rsv, 0, 5); | |
416 | return ret ? 1 : 0; | |
417 | } | |
418 | ||
419 | int btrfs_should_end_transaction(struct btrfs_trans_handle *trans, | |
420 | struct btrfs_root *root) | |
421 | { | |
422 | struct btrfs_transaction *cur_trans = trans->transaction; | |
423 | int updates; | |
424 | ||
425 | if (cur_trans->blocked || cur_trans->delayed_refs.flushing) | |
426 | return 1; | |
427 | ||
428 | updates = trans->delayed_ref_updates; | |
429 | trans->delayed_ref_updates = 0; | |
430 | if (updates) | |
431 | btrfs_run_delayed_refs(trans, root, updates); | |
432 | ||
433 | return should_end_transaction(trans, root); | |
434 | } | |
435 | ||
89ce8a63 | 436 | static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, |
0af3d00b | 437 | struct btrfs_root *root, int throttle, int lock) |
79154b1b | 438 | { |
8929ecfa | 439 | struct btrfs_transaction *cur_trans = trans->transaction; |
ab78c84d | 440 | struct btrfs_fs_info *info = root->fs_info; |
c3e69d58 CM |
441 | int count = 0; |
442 | ||
2a1eb461 JB |
443 | if (--trans->use_count) { |
444 | trans->block_rsv = trans->orig_rsv; | |
445 | return 0; | |
446 | } | |
447 | ||
c3e69d58 CM |
448 | while (count < 4) { |
449 | unsigned long cur = trans->delayed_ref_updates; | |
450 | trans->delayed_ref_updates = 0; | |
451 | if (cur && | |
452 | trans->transaction->delayed_refs.num_heads_ready > 64) { | |
453 | trans->delayed_ref_updates = 0; | |
b7ec40d7 CM |
454 | |
455 | /* | |
456 | * do a full flush if the transaction is trying | |
457 | * to close | |
458 | */ | |
459 | if (trans->transaction->delayed_refs.flushing) | |
460 | cur = 0; | |
c3e69d58 CM |
461 | btrfs_run_delayed_refs(trans, root, cur); |
462 | } else { | |
463 | break; | |
464 | } | |
465 | count++; | |
56bec294 CM |
466 | } |
467 | ||
a22285a6 YZ |
468 | btrfs_trans_release_metadata(trans, root); |
469 | ||
0af3d00b | 470 | if (lock && !root->fs_info->open_ioctl_trans && |
8929ecfa YZ |
471 | should_end_transaction(trans, root)) |
472 | trans->transaction->blocked = 1; | |
473 | ||
0af3d00b | 474 | if (lock && cur_trans->blocked && !cur_trans->in_commit) { |
8929ecfa YZ |
475 | if (throttle) |
476 | return btrfs_commit_transaction(trans, root); | |
477 | else | |
478 | wake_up_process(info->transaction_kthread); | |
479 | } | |
480 | ||
8929ecfa | 481 | WARN_ON(cur_trans != info->running_transaction); |
13c5a93e JB |
482 | WARN_ON(atomic_read(&cur_trans->num_writers) < 1); |
483 | atomic_dec(&cur_trans->num_writers); | |
89ce8a63 | 484 | |
99d16cbc | 485 | smp_mb(); |
79154b1b CM |
486 | if (waitqueue_active(&cur_trans->writer_wait)) |
487 | wake_up(&cur_trans->writer_wait); | |
79154b1b | 488 | put_transaction(cur_trans); |
9ed74f2d JB |
489 | |
490 | if (current->journal_info == trans) | |
491 | current->journal_info = NULL; | |
d6025579 | 492 | memset(trans, 0, sizeof(*trans)); |
2c90e5d6 | 493 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
ab78c84d | 494 | |
24bbcf04 YZ |
495 | if (throttle) |
496 | btrfs_run_delayed_iputs(root); | |
497 | ||
79154b1b CM |
498 | return 0; |
499 | } | |
500 | ||
89ce8a63 CM |
501 | int btrfs_end_transaction(struct btrfs_trans_handle *trans, |
502 | struct btrfs_root *root) | |
503 | { | |
0af3d00b | 504 | return __btrfs_end_transaction(trans, root, 0, 1); |
89ce8a63 CM |
505 | } |
506 | ||
507 | int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans, | |
508 | struct btrfs_root *root) | |
509 | { | |
0af3d00b JB |
510 | return __btrfs_end_transaction(trans, root, 1, 1); |
511 | } | |
512 | ||
513 | int btrfs_end_transaction_nolock(struct btrfs_trans_handle *trans, | |
514 | struct btrfs_root *root) | |
515 | { | |
516 | return __btrfs_end_transaction(trans, root, 0, 0); | |
89ce8a63 CM |
517 | } |
518 | ||
d352ac68 CM |
519 | /* |
520 | * when btree blocks are allocated, they have some corresponding bits set for | |
521 | * them in one of two extent_io trees. This is used to make sure all of | |
690587d1 | 522 | * those extents are sent to disk but does not wait on them |
d352ac68 | 523 | */ |
690587d1 | 524 | int btrfs_write_marked_extents(struct btrfs_root *root, |
8cef4e16 | 525 | struct extent_io_tree *dirty_pages, int mark) |
79154b1b | 526 | { |
7c4452b9 | 527 | int ret; |
777e6bd7 | 528 | int err = 0; |
7c4452b9 CM |
529 | int werr = 0; |
530 | struct page *page; | |
7c4452b9 | 531 | struct inode *btree_inode = root->fs_info->btree_inode; |
777e6bd7 | 532 | u64 start = 0; |
5f39d397 CM |
533 | u64 end; |
534 | unsigned long index; | |
7c4452b9 | 535 | |
d397712b | 536 | while (1) { |
777e6bd7 | 537 | ret = find_first_extent_bit(dirty_pages, start, &start, &end, |
8cef4e16 | 538 | mark); |
5f39d397 | 539 | if (ret) |
7c4452b9 | 540 | break; |
d397712b | 541 | while (start <= end) { |
777e6bd7 CM |
542 | cond_resched(); |
543 | ||
5f39d397 | 544 | index = start >> PAGE_CACHE_SHIFT; |
35ebb934 | 545 | start = (u64)(index + 1) << PAGE_CACHE_SHIFT; |
4bef0848 | 546 | page = find_get_page(btree_inode->i_mapping, index); |
7c4452b9 CM |
547 | if (!page) |
548 | continue; | |
4bef0848 CM |
549 | |
550 | btree_lock_page_hook(page); | |
551 | if (!page->mapping) { | |
552 | unlock_page(page); | |
553 | page_cache_release(page); | |
554 | continue; | |
555 | } | |
556 | ||
6702ed49 CM |
557 | if (PageWriteback(page)) { |
558 | if (PageDirty(page)) | |
559 | wait_on_page_writeback(page); | |
560 | else { | |
561 | unlock_page(page); | |
562 | page_cache_release(page); | |
563 | continue; | |
564 | } | |
565 | } | |
7c4452b9 CM |
566 | err = write_one_page(page, 0); |
567 | if (err) | |
568 | werr = err; | |
569 | page_cache_release(page); | |
570 | } | |
571 | } | |
690587d1 CM |
572 | if (err) |
573 | werr = err; | |
574 | return werr; | |
575 | } | |
576 | ||
577 | /* | |
578 | * when btree blocks are allocated, they have some corresponding bits set for | |
579 | * them in one of two extent_io trees. This is used to make sure all of | |
580 | * those extents are on disk for transaction or log commit. We wait | |
581 | * on all the pages and clear them from the dirty pages state tree | |
582 | */ | |
583 | int btrfs_wait_marked_extents(struct btrfs_root *root, | |
8cef4e16 | 584 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 CM |
585 | { |
586 | int ret; | |
587 | int err = 0; | |
588 | int werr = 0; | |
589 | struct page *page; | |
590 | struct inode *btree_inode = root->fs_info->btree_inode; | |
591 | u64 start = 0; | |
592 | u64 end; | |
593 | unsigned long index; | |
594 | ||
d397712b | 595 | while (1) { |
8cef4e16 YZ |
596 | ret = find_first_extent_bit(dirty_pages, start, &start, &end, |
597 | mark); | |
777e6bd7 CM |
598 | if (ret) |
599 | break; | |
600 | ||
8cef4e16 | 601 | clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS); |
d397712b | 602 | while (start <= end) { |
777e6bd7 CM |
603 | index = start >> PAGE_CACHE_SHIFT; |
604 | start = (u64)(index + 1) << PAGE_CACHE_SHIFT; | |
605 | page = find_get_page(btree_inode->i_mapping, index); | |
606 | if (!page) | |
607 | continue; | |
608 | if (PageDirty(page)) { | |
4bef0848 CM |
609 | btree_lock_page_hook(page); |
610 | wait_on_page_writeback(page); | |
777e6bd7 CM |
611 | err = write_one_page(page, 0); |
612 | if (err) | |
613 | werr = err; | |
614 | } | |
105d931d | 615 | wait_on_page_writeback(page); |
777e6bd7 CM |
616 | page_cache_release(page); |
617 | cond_resched(); | |
618 | } | |
619 | } | |
7c4452b9 CM |
620 | if (err) |
621 | werr = err; | |
622 | return werr; | |
79154b1b CM |
623 | } |
624 | ||
690587d1 CM |
625 | /* |
626 | * when btree blocks are allocated, they have some corresponding bits set for | |
627 | * them in one of two extent_io trees. This is used to make sure all of | |
628 | * those extents are on disk for transaction or log commit | |
629 | */ | |
630 | int btrfs_write_and_wait_marked_extents(struct btrfs_root *root, | |
8cef4e16 | 631 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 CM |
632 | { |
633 | int ret; | |
634 | int ret2; | |
635 | ||
8cef4e16 YZ |
636 | ret = btrfs_write_marked_extents(root, dirty_pages, mark); |
637 | ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark); | |
690587d1 CM |
638 | return ret || ret2; |
639 | } | |
640 | ||
d0c803c4 CM |
641 | int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans, |
642 | struct btrfs_root *root) | |
643 | { | |
644 | if (!trans || !trans->transaction) { | |
645 | struct inode *btree_inode; | |
646 | btree_inode = root->fs_info->btree_inode; | |
647 | return filemap_write_and_wait(btree_inode->i_mapping); | |
648 | } | |
649 | return btrfs_write_and_wait_marked_extents(root, | |
8cef4e16 YZ |
650 | &trans->transaction->dirty_pages, |
651 | EXTENT_DIRTY); | |
d0c803c4 CM |
652 | } |
653 | ||
d352ac68 CM |
654 | /* |
655 | * this is used to update the root pointer in the tree of tree roots. | |
656 | * | |
657 | * But, in the case of the extent allocation tree, updating the root | |
658 | * pointer may allocate blocks which may change the root of the extent | |
659 | * allocation tree. | |
660 | * | |
661 | * So, this loops and repeats and makes sure the cowonly root didn't | |
662 | * change while the root pointer was being updated in the metadata. | |
663 | */ | |
0b86a832 CM |
664 | static int update_cowonly_root(struct btrfs_trans_handle *trans, |
665 | struct btrfs_root *root) | |
79154b1b CM |
666 | { |
667 | int ret; | |
0b86a832 | 668 | u64 old_root_bytenr; |
86b9f2ec | 669 | u64 old_root_used; |
0b86a832 | 670 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
79154b1b | 671 | |
86b9f2ec | 672 | old_root_used = btrfs_root_used(&root->root_item); |
0b86a832 | 673 | btrfs_write_dirty_block_groups(trans, root); |
56bec294 | 674 | |
d397712b | 675 | while (1) { |
0b86a832 | 676 | old_root_bytenr = btrfs_root_bytenr(&root->root_item); |
86b9f2ec YZ |
677 | if (old_root_bytenr == root->node->start && |
678 | old_root_used == btrfs_root_used(&root->root_item)) | |
79154b1b | 679 | break; |
87ef2bb4 | 680 | |
5d4f98a2 | 681 | btrfs_set_root_node(&root->root_item, root->node); |
79154b1b | 682 | ret = btrfs_update_root(trans, tree_root, |
0b86a832 CM |
683 | &root->root_key, |
684 | &root->root_item); | |
79154b1b | 685 | BUG_ON(ret); |
56bec294 | 686 | |
86b9f2ec | 687 | old_root_used = btrfs_root_used(&root->root_item); |
4a8c9a62 | 688 | ret = btrfs_write_dirty_block_groups(trans, root); |
56bec294 | 689 | BUG_ON(ret); |
0b86a832 | 690 | } |
276e680d YZ |
691 | |
692 | if (root != root->fs_info->extent_root) | |
693 | switch_commit_root(root); | |
694 | ||
0b86a832 CM |
695 | return 0; |
696 | } | |
697 | ||
d352ac68 CM |
698 | /* |
699 | * update all the cowonly tree roots on disk | |
700 | */ | |
5d4f98a2 YZ |
701 | static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans, |
702 | struct btrfs_root *root) | |
0b86a832 CM |
703 | { |
704 | struct btrfs_fs_info *fs_info = root->fs_info; | |
705 | struct list_head *next; | |
84234f3a | 706 | struct extent_buffer *eb; |
56bec294 | 707 | int ret; |
84234f3a | 708 | |
56bec294 CM |
709 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
710 | BUG_ON(ret); | |
87ef2bb4 | 711 | |
84234f3a | 712 | eb = btrfs_lock_root_node(fs_info->tree_root); |
9fa8cfe7 | 713 | btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 0, &eb); |
84234f3a YZ |
714 | btrfs_tree_unlock(eb); |
715 | free_extent_buffer(eb); | |
0b86a832 | 716 | |
56bec294 CM |
717 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
718 | BUG_ON(ret); | |
87ef2bb4 | 719 | |
d397712b | 720 | while (!list_empty(&fs_info->dirty_cowonly_roots)) { |
0b86a832 CM |
721 | next = fs_info->dirty_cowonly_roots.next; |
722 | list_del_init(next); | |
723 | root = list_entry(next, struct btrfs_root, dirty_list); | |
87ef2bb4 | 724 | |
0b86a832 | 725 | update_cowonly_root(trans, root); |
79154b1b | 726 | } |
276e680d YZ |
727 | |
728 | down_write(&fs_info->extent_commit_sem); | |
729 | switch_commit_root(fs_info->extent_root); | |
730 | up_write(&fs_info->extent_commit_sem); | |
731 | ||
79154b1b CM |
732 | return 0; |
733 | } | |
734 | ||
d352ac68 CM |
735 | /* |
736 | * dead roots are old snapshots that need to be deleted. This allocates | |
737 | * a dirty root struct and adds it into the list of dead roots that need to | |
738 | * be deleted | |
739 | */ | |
5d4f98a2 | 740 | int btrfs_add_dead_root(struct btrfs_root *root) |
5eda7b5e | 741 | { |
b48652c1 | 742 | mutex_lock(&root->fs_info->trans_mutex); |
5d4f98a2 | 743 | list_add(&root->root_list, &root->fs_info->dead_roots); |
b48652c1 | 744 | mutex_unlock(&root->fs_info->trans_mutex); |
5eda7b5e CM |
745 | return 0; |
746 | } | |
747 | ||
d352ac68 | 748 | /* |
5d4f98a2 | 749 | * update all the cowonly tree roots on disk |
d352ac68 | 750 | */ |
5d4f98a2 YZ |
751 | static noinline int commit_fs_roots(struct btrfs_trans_handle *trans, |
752 | struct btrfs_root *root) | |
0f7d52f4 | 753 | { |
0f7d52f4 | 754 | struct btrfs_root *gang[8]; |
5d4f98a2 | 755 | struct btrfs_fs_info *fs_info = root->fs_info; |
0f7d52f4 CM |
756 | int i; |
757 | int ret; | |
54aa1f4d CM |
758 | int err = 0; |
759 | ||
d397712b | 760 | while (1) { |
5d4f98a2 YZ |
761 | ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, |
762 | (void **)gang, 0, | |
0f7d52f4 CM |
763 | ARRAY_SIZE(gang), |
764 | BTRFS_ROOT_TRANS_TAG); | |
765 | if (ret == 0) | |
766 | break; | |
767 | for (i = 0; i < ret; i++) { | |
768 | root = gang[i]; | |
5d4f98a2 YZ |
769 | radix_tree_tag_clear(&fs_info->fs_roots_radix, |
770 | (unsigned long)root->root_key.objectid, | |
771 | BTRFS_ROOT_TRANS_TAG); | |
31153d81 | 772 | |
e02119d5 | 773 | btrfs_free_log(trans, root); |
5d4f98a2 | 774 | btrfs_update_reloc_root(trans, root); |
d68fc57b | 775 | btrfs_orphan_commit_root(trans, root); |
bcc63abb | 776 | |
978d910d | 777 | if (root->commit_root != root->node) { |
817d52f8 | 778 | switch_commit_root(root); |
978d910d YZ |
779 | btrfs_set_root_node(&root->root_item, |
780 | root->node); | |
781 | } | |
5d4f98a2 | 782 | |
5d4f98a2 | 783 | err = btrfs_update_root(trans, fs_info->tree_root, |
0f7d52f4 CM |
784 | &root->root_key, |
785 | &root->root_item); | |
54aa1f4d CM |
786 | if (err) |
787 | break; | |
0f7d52f4 CM |
788 | } |
789 | } | |
54aa1f4d | 790 | return err; |
0f7d52f4 CM |
791 | } |
792 | ||
d352ac68 CM |
793 | /* |
794 | * defrag a given btree. If cacheonly == 1, this won't read from the disk, | |
795 | * otherwise every leaf in the btree is read and defragged. | |
796 | */ | |
e9d0b13b CM |
797 | int btrfs_defrag_root(struct btrfs_root *root, int cacheonly) |
798 | { | |
799 | struct btrfs_fs_info *info = root->fs_info; | |
e9d0b13b | 800 | struct btrfs_trans_handle *trans; |
8929ecfa | 801 | int ret; |
d3c2fdcf | 802 | unsigned long nr; |
e9d0b13b | 803 | |
8929ecfa | 804 | if (xchg(&root->defrag_running, 1)) |
e9d0b13b | 805 | return 0; |
8929ecfa | 806 | |
6b80053d | 807 | while (1) { |
8929ecfa YZ |
808 | trans = btrfs_start_transaction(root, 0); |
809 | if (IS_ERR(trans)) | |
810 | return PTR_ERR(trans); | |
811 | ||
e9d0b13b | 812 | ret = btrfs_defrag_leaves(trans, root, cacheonly); |
8929ecfa | 813 | |
d3c2fdcf | 814 | nr = trans->blocks_used; |
e9d0b13b | 815 | btrfs_end_transaction(trans, root); |
d3c2fdcf | 816 | btrfs_btree_balance_dirty(info->tree_root, nr); |
e9d0b13b CM |
817 | cond_resched(); |
818 | ||
3f157a2f | 819 | if (root->fs_info->closing || ret != -EAGAIN) |
e9d0b13b CM |
820 | break; |
821 | } | |
822 | root->defrag_running = 0; | |
8929ecfa | 823 | return ret; |
e9d0b13b CM |
824 | } |
825 | ||
2c47e605 | 826 | #if 0 |
b7ec40d7 CM |
827 | /* |
828 | * when dropping snapshots, we generate a ton of delayed refs, and it makes | |
829 | * sense not to join the transaction while it is trying to flush the current | |
830 | * queue of delayed refs out. | |
831 | * | |
832 | * This is used by the drop snapshot code only | |
833 | */ | |
834 | static noinline int wait_transaction_pre_flush(struct btrfs_fs_info *info) | |
835 | { | |
836 | DEFINE_WAIT(wait); | |
837 | ||
838 | mutex_lock(&info->trans_mutex); | |
839 | while (info->running_transaction && | |
840 | info->running_transaction->delayed_refs.flushing) { | |
841 | prepare_to_wait(&info->transaction_wait, &wait, | |
842 | TASK_UNINTERRUPTIBLE); | |
843 | mutex_unlock(&info->trans_mutex); | |
59bc5c75 | 844 | |
b7ec40d7 | 845 | schedule(); |
59bc5c75 | 846 | |
b7ec40d7 CM |
847 | mutex_lock(&info->trans_mutex); |
848 | finish_wait(&info->transaction_wait, &wait); | |
849 | } | |
850 | mutex_unlock(&info->trans_mutex); | |
851 | return 0; | |
852 | } | |
853 | ||
d352ac68 CM |
854 | /* |
855 | * Given a list of roots that need to be deleted, call btrfs_drop_snapshot on | |
856 | * all of them | |
857 | */ | |
5d4f98a2 | 858 | int btrfs_drop_dead_root(struct btrfs_root *root) |
0f7d52f4 | 859 | { |
0f7d52f4 | 860 | struct btrfs_trans_handle *trans; |
5d4f98a2 | 861 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
d3c2fdcf | 862 | unsigned long nr; |
5d4f98a2 | 863 | int ret; |
58176a96 | 864 | |
5d4f98a2 YZ |
865 | while (1) { |
866 | /* | |
867 | * we don't want to jump in and create a bunch of | |
868 | * delayed refs if the transaction is starting to close | |
869 | */ | |
870 | wait_transaction_pre_flush(tree_root->fs_info); | |
871 | trans = btrfs_start_transaction(tree_root, 1); | |
a2135011 | 872 | |
5d4f98a2 YZ |
873 | /* |
874 | * we've joined a transaction, make sure it isn't | |
875 | * closing right now | |
876 | */ | |
877 | if (trans->transaction->delayed_refs.flushing) { | |
878 | btrfs_end_transaction(trans, tree_root); | |
879 | continue; | |
9f3a7427 | 880 | } |
58176a96 | 881 | |
5d4f98a2 YZ |
882 | ret = btrfs_drop_snapshot(trans, root); |
883 | if (ret != -EAGAIN) | |
884 | break; | |
a2135011 | 885 | |
5d4f98a2 YZ |
886 | ret = btrfs_update_root(trans, tree_root, |
887 | &root->root_key, | |
888 | &root->root_item); | |
889 | if (ret) | |
54aa1f4d | 890 | break; |
bcc63abb | 891 | |
d3c2fdcf | 892 | nr = trans->blocks_used; |
0f7d52f4 CM |
893 | ret = btrfs_end_transaction(trans, tree_root); |
894 | BUG_ON(ret); | |
5eda7b5e | 895 | |
d3c2fdcf | 896 | btrfs_btree_balance_dirty(tree_root, nr); |
4dc11904 | 897 | cond_resched(); |
0f7d52f4 | 898 | } |
5d4f98a2 YZ |
899 | BUG_ON(ret); |
900 | ||
901 | ret = btrfs_del_root(trans, tree_root, &root->root_key); | |
902 | BUG_ON(ret); | |
903 | ||
904 | nr = trans->blocks_used; | |
905 | ret = btrfs_end_transaction(trans, tree_root); | |
906 | BUG_ON(ret); | |
907 | ||
908 | free_extent_buffer(root->node); | |
909 | free_extent_buffer(root->commit_root); | |
910 | kfree(root); | |
911 | ||
912 | btrfs_btree_balance_dirty(tree_root, nr); | |
54aa1f4d | 913 | return ret; |
0f7d52f4 | 914 | } |
2c47e605 | 915 | #endif |
0f7d52f4 | 916 | |
d352ac68 CM |
917 | /* |
918 | * new snapshots need to be created at a very specific time in the | |
919 | * transaction commit. This does the actual creation | |
920 | */ | |
80b6794d | 921 | static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, |
3063d29f CM |
922 | struct btrfs_fs_info *fs_info, |
923 | struct btrfs_pending_snapshot *pending) | |
924 | { | |
925 | struct btrfs_key key; | |
80b6794d | 926 | struct btrfs_root_item *new_root_item; |
3063d29f CM |
927 | struct btrfs_root *tree_root = fs_info->tree_root; |
928 | struct btrfs_root *root = pending->root; | |
6bdb72de SW |
929 | struct btrfs_root *parent_root; |
930 | struct inode *parent_inode; | |
6a912213 | 931 | struct dentry *parent; |
a22285a6 | 932 | struct dentry *dentry; |
3063d29f | 933 | struct extent_buffer *tmp; |
925baedd | 934 | struct extent_buffer *old; |
3063d29f | 935 | int ret; |
d68fc57b | 936 | u64 to_reserve = 0; |
6bdb72de | 937 | u64 index = 0; |
a22285a6 | 938 | u64 objectid; |
b83cc969 | 939 | u64 root_flags; |
3063d29f | 940 | |
80b6794d CM |
941 | new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS); |
942 | if (!new_root_item) { | |
a22285a6 | 943 | pending->error = -ENOMEM; |
80b6794d CM |
944 | goto fail; |
945 | } | |
a22285a6 | 946 | |
3063d29f | 947 | ret = btrfs_find_free_objectid(trans, tree_root, 0, &objectid); |
a22285a6 YZ |
948 | if (ret) { |
949 | pending->error = ret; | |
3063d29f | 950 | goto fail; |
a22285a6 | 951 | } |
3063d29f | 952 | |
3fd0a558 | 953 | btrfs_reloc_pre_snapshot(trans, pending, &to_reserve); |
d68fc57b YZ |
954 | btrfs_orphan_pre_snapshot(trans, pending, &to_reserve); |
955 | ||
956 | if (to_reserve > 0) { | |
957 | ret = btrfs_block_rsv_add(trans, root, &pending->block_rsv, | |
8bb8ab2e | 958 | to_reserve); |
d68fc57b YZ |
959 | if (ret) { |
960 | pending->error = ret; | |
961 | goto fail; | |
962 | } | |
963 | } | |
964 | ||
3063d29f | 965 | key.objectid = objectid; |
a22285a6 YZ |
966 | key.offset = (u64)-1; |
967 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3063d29f | 968 | |
a22285a6 | 969 | trans->block_rsv = &pending->block_rsv; |
3de4586c | 970 | |
a22285a6 | 971 | dentry = pending->dentry; |
6a912213 JB |
972 | parent = dget_parent(dentry); |
973 | parent_inode = parent->d_inode; | |
a22285a6 | 974 | parent_root = BTRFS_I(parent_inode)->root; |
6bdb72de | 975 | record_root_in_trans(trans, parent_root); |
a22285a6 | 976 | |
3063d29f CM |
977 | /* |
978 | * insert the directory item | |
979 | */ | |
3de4586c | 980 | ret = btrfs_set_inode_index(parent_inode, &index); |
6bdb72de | 981 | BUG_ON(ret); |
0660b5af | 982 | ret = btrfs_insert_dir_item(trans, parent_root, |
a22285a6 YZ |
983 | dentry->d_name.name, dentry->d_name.len, |
984 | parent_inode->i_ino, &key, | |
985 | BTRFS_FT_DIR, index); | |
6bdb72de | 986 | BUG_ON(ret); |
0660b5af | 987 | |
a22285a6 YZ |
988 | btrfs_i_size_write(parent_inode, parent_inode->i_size + |
989 | dentry->d_name.len * 2); | |
52c26179 YZ |
990 | ret = btrfs_update_inode(trans, parent_root, parent_inode); |
991 | BUG_ON(ret); | |
992 | ||
6bdb72de SW |
993 | record_root_in_trans(trans, root); |
994 | btrfs_set_root_last_snapshot(&root->root_item, trans->transid); | |
995 | memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); | |
08fe4db1 | 996 | btrfs_check_and_init_root_item(new_root_item); |
6bdb72de | 997 | |
b83cc969 LZ |
998 | root_flags = btrfs_root_flags(new_root_item); |
999 | if (pending->readonly) | |
1000 | root_flags |= BTRFS_ROOT_SUBVOL_RDONLY; | |
1001 | else | |
1002 | root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY; | |
1003 | btrfs_set_root_flags(new_root_item, root_flags); | |
1004 | ||
6bdb72de SW |
1005 | old = btrfs_lock_root_node(root); |
1006 | btrfs_cow_block(trans, root, old, NULL, 0, &old); | |
1007 | btrfs_set_lock_blocking(old); | |
1008 | ||
1009 | btrfs_copy_root(trans, root, old, &tmp, objectid); | |
1010 | btrfs_tree_unlock(old); | |
1011 | free_extent_buffer(old); | |
1012 | ||
1013 | btrfs_set_root_node(new_root_item, tmp); | |
a22285a6 YZ |
1014 | /* record when the snapshot was created in key.offset */ |
1015 | key.offset = trans->transid; | |
1016 | ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); | |
6bdb72de SW |
1017 | btrfs_tree_unlock(tmp); |
1018 | free_extent_buffer(tmp); | |
a22285a6 | 1019 | BUG_ON(ret); |
6bdb72de | 1020 | |
a22285a6 YZ |
1021 | /* |
1022 | * insert root back/forward references | |
1023 | */ | |
1024 | ret = btrfs_add_root_ref(trans, tree_root, objectid, | |
0660b5af | 1025 | parent_root->root_key.objectid, |
a22285a6 YZ |
1026 | parent_inode->i_ino, index, |
1027 | dentry->d_name.name, dentry->d_name.len); | |
0660b5af | 1028 | BUG_ON(ret); |
6a912213 | 1029 | dput(parent); |
0660b5af | 1030 | |
a22285a6 YZ |
1031 | key.offset = (u64)-1; |
1032 | pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key); | |
1033 | BUG_ON(IS_ERR(pending->snap)); | |
d68fc57b | 1034 | |
3fd0a558 | 1035 | btrfs_reloc_post_snapshot(trans, pending); |
d68fc57b | 1036 | btrfs_orphan_post_snapshot(trans, pending); |
3063d29f | 1037 | fail: |
6bdb72de | 1038 | kfree(new_root_item); |
a22285a6 YZ |
1039 | btrfs_block_rsv_release(root, &pending->block_rsv, (u64)-1); |
1040 | return 0; | |
3063d29f CM |
1041 | } |
1042 | ||
d352ac68 CM |
1043 | /* |
1044 | * create all the snapshots we've scheduled for creation | |
1045 | */ | |
80b6794d CM |
1046 | static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans, |
1047 | struct btrfs_fs_info *fs_info) | |
3de4586c CM |
1048 | { |
1049 | struct btrfs_pending_snapshot *pending; | |
1050 | struct list_head *head = &trans->transaction->pending_snapshots; | |
3de4586c CM |
1051 | int ret; |
1052 | ||
c6e30871 | 1053 | list_for_each_entry(pending, head, list) { |
3de4586c CM |
1054 | ret = create_pending_snapshot(trans, fs_info, pending); |
1055 | BUG_ON(ret); | |
1056 | } | |
1057 | return 0; | |
1058 | } | |
1059 | ||
5d4f98a2 YZ |
1060 | static void update_super_roots(struct btrfs_root *root) |
1061 | { | |
1062 | struct btrfs_root_item *root_item; | |
1063 | struct btrfs_super_block *super; | |
1064 | ||
1065 | super = &root->fs_info->super_copy; | |
1066 | ||
1067 | root_item = &root->fs_info->chunk_root->root_item; | |
1068 | super->chunk_root = root_item->bytenr; | |
1069 | super->chunk_root_generation = root_item->generation; | |
1070 | super->chunk_root_level = root_item->level; | |
1071 | ||
1072 | root_item = &root->fs_info->tree_root->root_item; | |
1073 | super->root = root_item->bytenr; | |
1074 | super->generation = root_item->generation; | |
1075 | super->root_level = root_item->level; | |
0af3d00b JB |
1076 | if (super->cache_generation != 0 || btrfs_test_opt(root, SPACE_CACHE)) |
1077 | super->cache_generation = root_item->generation; | |
5d4f98a2 YZ |
1078 | } |
1079 | ||
f36f3042 CM |
1080 | int btrfs_transaction_in_commit(struct btrfs_fs_info *info) |
1081 | { | |
1082 | int ret = 0; | |
1083 | spin_lock(&info->new_trans_lock); | |
1084 | if (info->running_transaction) | |
1085 | ret = info->running_transaction->in_commit; | |
1086 | spin_unlock(&info->new_trans_lock); | |
1087 | return ret; | |
1088 | } | |
1089 | ||
8929ecfa YZ |
1090 | int btrfs_transaction_blocked(struct btrfs_fs_info *info) |
1091 | { | |
1092 | int ret = 0; | |
1093 | spin_lock(&info->new_trans_lock); | |
1094 | if (info->running_transaction) | |
1095 | ret = info->running_transaction->blocked; | |
1096 | spin_unlock(&info->new_trans_lock); | |
1097 | return ret; | |
1098 | } | |
1099 | ||
bb9c12c9 SW |
1100 | /* |
1101 | * wait for the current transaction commit to start and block subsequent | |
1102 | * transaction joins | |
1103 | */ | |
1104 | static void wait_current_trans_commit_start(struct btrfs_root *root, | |
1105 | struct btrfs_transaction *trans) | |
1106 | { | |
1107 | DEFINE_WAIT(wait); | |
1108 | ||
1109 | if (trans->in_commit) | |
1110 | return; | |
1111 | ||
1112 | while (1) { | |
1113 | prepare_to_wait(&root->fs_info->transaction_blocked_wait, &wait, | |
1114 | TASK_UNINTERRUPTIBLE); | |
1115 | if (trans->in_commit) { | |
1116 | finish_wait(&root->fs_info->transaction_blocked_wait, | |
1117 | &wait); | |
1118 | break; | |
1119 | } | |
1120 | mutex_unlock(&root->fs_info->trans_mutex); | |
1121 | schedule(); | |
1122 | mutex_lock(&root->fs_info->trans_mutex); | |
1123 | finish_wait(&root->fs_info->transaction_blocked_wait, &wait); | |
1124 | } | |
1125 | } | |
1126 | ||
1127 | /* | |
1128 | * wait for the current transaction to start and then become unblocked. | |
1129 | * caller holds ref. | |
1130 | */ | |
1131 | static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root, | |
1132 | struct btrfs_transaction *trans) | |
1133 | { | |
1134 | DEFINE_WAIT(wait); | |
1135 | ||
1136 | if (trans->commit_done || (trans->in_commit && !trans->blocked)) | |
1137 | return; | |
1138 | ||
1139 | while (1) { | |
1140 | prepare_to_wait(&root->fs_info->transaction_wait, &wait, | |
1141 | TASK_UNINTERRUPTIBLE); | |
1142 | if (trans->commit_done || | |
1143 | (trans->in_commit && !trans->blocked)) { | |
1144 | finish_wait(&root->fs_info->transaction_wait, | |
1145 | &wait); | |
1146 | break; | |
1147 | } | |
1148 | mutex_unlock(&root->fs_info->trans_mutex); | |
1149 | schedule(); | |
1150 | mutex_lock(&root->fs_info->trans_mutex); | |
1151 | finish_wait(&root->fs_info->transaction_wait, | |
1152 | &wait); | |
1153 | } | |
1154 | } | |
1155 | ||
1156 | /* | |
1157 | * commit transactions asynchronously. once btrfs_commit_transaction_async | |
1158 | * returns, any subsequent transaction will not be allowed to join. | |
1159 | */ | |
1160 | struct btrfs_async_commit { | |
1161 | struct btrfs_trans_handle *newtrans; | |
1162 | struct btrfs_root *root; | |
1163 | struct delayed_work work; | |
1164 | }; | |
1165 | ||
1166 | static void do_async_commit(struct work_struct *work) | |
1167 | { | |
1168 | struct btrfs_async_commit *ac = | |
1169 | container_of(work, struct btrfs_async_commit, work.work); | |
1170 | ||
1171 | btrfs_commit_transaction(ac->newtrans, ac->root); | |
1172 | kfree(ac); | |
1173 | } | |
1174 | ||
1175 | int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans, | |
1176 | struct btrfs_root *root, | |
1177 | int wait_for_unblock) | |
1178 | { | |
1179 | struct btrfs_async_commit *ac; | |
1180 | struct btrfs_transaction *cur_trans; | |
1181 | ||
1182 | ac = kmalloc(sizeof(*ac), GFP_NOFS); | |
db5b493a TI |
1183 | if (!ac) |
1184 | return -ENOMEM; | |
bb9c12c9 SW |
1185 | |
1186 | INIT_DELAYED_WORK(&ac->work, do_async_commit); | |
1187 | ac->root = root; | |
7a7eaa40 | 1188 | ac->newtrans = btrfs_join_transaction(root); |
3612b495 TI |
1189 | if (IS_ERR(ac->newtrans)) { |
1190 | int err = PTR_ERR(ac->newtrans); | |
1191 | kfree(ac); | |
1192 | return err; | |
1193 | } | |
bb9c12c9 SW |
1194 | |
1195 | /* take transaction reference */ | |
1196 | mutex_lock(&root->fs_info->trans_mutex); | |
1197 | cur_trans = trans->transaction; | |
13c5a93e | 1198 | atomic_inc(&cur_trans->use_count); |
bb9c12c9 SW |
1199 | mutex_unlock(&root->fs_info->trans_mutex); |
1200 | ||
1201 | btrfs_end_transaction(trans, root); | |
1202 | schedule_delayed_work(&ac->work, 0); | |
1203 | ||
1204 | /* wait for transaction to start and unblock */ | |
1205 | mutex_lock(&root->fs_info->trans_mutex); | |
1206 | if (wait_for_unblock) | |
1207 | wait_current_trans_commit_start_and_unblock(root, cur_trans); | |
1208 | else | |
1209 | wait_current_trans_commit_start(root, cur_trans); | |
1210 | put_transaction(cur_trans); | |
1211 | mutex_unlock(&root->fs_info->trans_mutex); | |
1212 | ||
1213 | return 0; | |
1214 | } | |
1215 | ||
1216 | /* | |
1217 | * btrfs_transaction state sequence: | |
1218 | * in_commit = 0, blocked = 0 (initial) | |
1219 | * in_commit = 1, blocked = 1 | |
1220 | * blocked = 0 | |
1221 | * commit_done = 1 | |
1222 | */ | |
79154b1b CM |
1223 | int btrfs_commit_transaction(struct btrfs_trans_handle *trans, |
1224 | struct btrfs_root *root) | |
1225 | { | |
15ee9bc7 | 1226 | unsigned long joined = 0; |
79154b1b | 1227 | struct btrfs_transaction *cur_trans; |
8fd17795 | 1228 | struct btrfs_transaction *prev_trans = NULL; |
79154b1b | 1229 | DEFINE_WAIT(wait); |
15ee9bc7 | 1230 | int ret; |
89573b9c CM |
1231 | int should_grow = 0; |
1232 | unsigned long now = get_seconds(); | |
dccae999 | 1233 | int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT); |
79154b1b | 1234 | |
5a3f23d5 CM |
1235 | btrfs_run_ordered_operations(root, 0); |
1236 | ||
56bec294 CM |
1237 | /* make a pass through all the delayed refs we have so far |
1238 | * any runnings procs may add more while we are here | |
1239 | */ | |
1240 | ret = btrfs_run_delayed_refs(trans, root, 0); | |
1241 | BUG_ON(ret); | |
1242 | ||
a22285a6 YZ |
1243 | btrfs_trans_release_metadata(trans, root); |
1244 | ||
b7ec40d7 | 1245 | cur_trans = trans->transaction; |
56bec294 CM |
1246 | /* |
1247 | * set the flushing flag so procs in this transaction have to | |
1248 | * start sending their work down. | |
1249 | */ | |
b7ec40d7 | 1250 | cur_trans->delayed_refs.flushing = 1; |
56bec294 | 1251 | |
c3e69d58 | 1252 | ret = btrfs_run_delayed_refs(trans, root, 0); |
56bec294 CM |
1253 | BUG_ON(ret); |
1254 | ||
79154b1b | 1255 | mutex_lock(&root->fs_info->trans_mutex); |
b7ec40d7 | 1256 | if (cur_trans->in_commit) { |
13c5a93e | 1257 | atomic_inc(&cur_trans->use_count); |
ccd467d6 | 1258 | mutex_unlock(&root->fs_info->trans_mutex); |
79154b1b | 1259 | btrfs_end_transaction(trans, root); |
ccd467d6 | 1260 | |
79154b1b CM |
1261 | ret = wait_for_commit(root, cur_trans); |
1262 | BUG_ON(ret); | |
15ee9bc7 JB |
1263 | |
1264 | mutex_lock(&root->fs_info->trans_mutex); | |
79154b1b | 1265 | put_transaction(cur_trans); |
15ee9bc7 JB |
1266 | mutex_unlock(&root->fs_info->trans_mutex); |
1267 | ||
79154b1b CM |
1268 | return 0; |
1269 | } | |
4313b399 | 1270 | |
2c90e5d6 | 1271 | trans->transaction->in_commit = 1; |
f9295749 | 1272 | trans->transaction->blocked = 1; |
bb9c12c9 SW |
1273 | wake_up(&root->fs_info->transaction_blocked_wait); |
1274 | ||
ccd467d6 CM |
1275 | if (cur_trans->list.prev != &root->fs_info->trans_list) { |
1276 | prev_trans = list_entry(cur_trans->list.prev, | |
1277 | struct btrfs_transaction, list); | |
1278 | if (!prev_trans->commit_done) { | |
13c5a93e | 1279 | atomic_inc(&prev_trans->use_count); |
ccd467d6 CM |
1280 | mutex_unlock(&root->fs_info->trans_mutex); |
1281 | ||
1282 | wait_for_commit(root, prev_trans); | |
ccd467d6 | 1283 | |
ccd467d6 | 1284 | mutex_lock(&root->fs_info->trans_mutex); |
15ee9bc7 | 1285 | put_transaction(prev_trans); |
ccd467d6 CM |
1286 | } |
1287 | } | |
15ee9bc7 | 1288 | |
89573b9c CM |
1289 | if (now < cur_trans->start_time || now - cur_trans->start_time < 1) |
1290 | should_grow = 1; | |
1291 | ||
15ee9bc7 | 1292 | do { |
7ea394f1 | 1293 | int snap_pending = 0; |
15ee9bc7 | 1294 | joined = cur_trans->num_joined; |
7ea394f1 YZ |
1295 | if (!list_empty(&trans->transaction->pending_snapshots)) |
1296 | snap_pending = 1; | |
1297 | ||
2c90e5d6 | 1298 | WARN_ON(cur_trans != trans->transaction); |
79154b1b | 1299 | mutex_unlock(&root->fs_info->trans_mutex); |
15ee9bc7 | 1300 | |
0bdb1db2 | 1301 | if (flush_on_commit || snap_pending) { |
24bbcf04 YZ |
1302 | btrfs_start_delalloc_inodes(root, 1); |
1303 | ret = btrfs_wait_ordered_extents(root, 0, 1); | |
ebecd3d9 | 1304 | BUG_ON(ret); |
7ea394f1 YZ |
1305 | } |
1306 | ||
5a3f23d5 CM |
1307 | /* |
1308 | * rename don't use btrfs_join_transaction, so, once we | |
1309 | * set the transaction to blocked above, we aren't going | |
1310 | * to get any new ordered operations. We can safely run | |
1311 | * it here and no for sure that nothing new will be added | |
1312 | * to the list | |
1313 | */ | |
1314 | btrfs_run_ordered_operations(root, 1); | |
1315 | ||
ed3b3d31 CM |
1316 | prepare_to_wait(&cur_trans->writer_wait, &wait, |
1317 | TASK_UNINTERRUPTIBLE); | |
1318 | ||
89573b9c | 1319 | smp_mb(); |
13c5a93e | 1320 | if (atomic_read(&cur_trans->num_writers) > 1) |
99d16cbc SW |
1321 | schedule_timeout(MAX_SCHEDULE_TIMEOUT); |
1322 | else if (should_grow) | |
1323 | schedule_timeout(1); | |
15ee9bc7 | 1324 | |
79154b1b | 1325 | mutex_lock(&root->fs_info->trans_mutex); |
15ee9bc7 | 1326 | finish_wait(&cur_trans->writer_wait, &wait); |
13c5a93e | 1327 | } while (atomic_read(&cur_trans->num_writers) > 1 || |
89573b9c | 1328 | (should_grow && cur_trans->num_joined != joined)); |
15ee9bc7 | 1329 | |
3063d29f CM |
1330 | ret = create_pending_snapshots(trans, root->fs_info); |
1331 | BUG_ON(ret); | |
1332 | ||
56bec294 CM |
1333 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
1334 | BUG_ON(ret); | |
1335 | ||
2c90e5d6 | 1336 | WARN_ON(cur_trans != trans->transaction); |
dc17ff8f | 1337 | |
e02119d5 CM |
1338 | /* btrfs_commit_tree_roots is responsible for getting the |
1339 | * various roots consistent with each other. Every pointer | |
1340 | * in the tree of tree roots has to point to the most up to date | |
1341 | * root for every subvolume and other tree. So, we have to keep | |
1342 | * the tree logging code from jumping in and changing any | |
1343 | * of the trees. | |
1344 | * | |
1345 | * At this point in the commit, there can't be any tree-log | |
1346 | * writers, but a little lower down we drop the trans mutex | |
1347 | * and let new people in. By holding the tree_log_mutex | |
1348 | * from now until after the super is written, we avoid races | |
1349 | * with the tree-log code. | |
1350 | */ | |
1351 | mutex_lock(&root->fs_info->tree_log_mutex); | |
1352 | ||
5d4f98a2 | 1353 | ret = commit_fs_roots(trans, root); |
54aa1f4d CM |
1354 | BUG_ON(ret); |
1355 | ||
5d4f98a2 | 1356 | /* commit_fs_roots gets rid of all the tree log roots, it is now |
e02119d5 CM |
1357 | * safe to free the root of tree log roots |
1358 | */ | |
1359 | btrfs_free_log_root_tree(trans, root->fs_info); | |
1360 | ||
5d4f98a2 | 1361 | ret = commit_cowonly_roots(trans, root); |
79154b1b | 1362 | BUG_ON(ret); |
54aa1f4d | 1363 | |
11833d66 YZ |
1364 | btrfs_prepare_extent_commit(trans, root); |
1365 | ||
78fae27e | 1366 | cur_trans = root->fs_info->running_transaction; |
cee36a03 | 1367 | spin_lock(&root->fs_info->new_trans_lock); |
78fae27e | 1368 | root->fs_info->running_transaction = NULL; |
cee36a03 | 1369 | spin_unlock(&root->fs_info->new_trans_lock); |
5d4f98a2 YZ |
1370 | |
1371 | btrfs_set_root_node(&root->fs_info->tree_root->root_item, | |
1372 | root->fs_info->tree_root->node); | |
817d52f8 | 1373 | switch_commit_root(root->fs_info->tree_root); |
5d4f98a2 YZ |
1374 | |
1375 | btrfs_set_root_node(&root->fs_info->chunk_root->root_item, | |
1376 | root->fs_info->chunk_root->node); | |
817d52f8 | 1377 | switch_commit_root(root->fs_info->chunk_root); |
5d4f98a2 YZ |
1378 | |
1379 | update_super_roots(root); | |
e02119d5 CM |
1380 | |
1381 | if (!root->fs_info->log_root_recovering) { | |
1382 | btrfs_set_super_log_root(&root->fs_info->super_copy, 0); | |
1383 | btrfs_set_super_log_root_level(&root->fs_info->super_copy, 0); | |
1384 | } | |
1385 | ||
a061fc8d CM |
1386 | memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy, |
1387 | sizeof(root->fs_info->super_copy)); | |
ccd467d6 | 1388 | |
f9295749 | 1389 | trans->transaction->blocked = 0; |
b7ec40d7 | 1390 | |
f9295749 | 1391 | wake_up(&root->fs_info->transaction_wait); |
e6dcd2dc | 1392 | |
78fae27e | 1393 | mutex_unlock(&root->fs_info->trans_mutex); |
79154b1b CM |
1394 | ret = btrfs_write_and_wait_transaction(trans, root); |
1395 | BUG_ON(ret); | |
a512bbf8 | 1396 | write_ctree_super(trans, root, 0); |
4313b399 | 1397 | |
e02119d5 CM |
1398 | /* |
1399 | * the super is written, we can safely allow the tree-loggers | |
1400 | * to go about their business | |
1401 | */ | |
1402 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
1403 | ||
11833d66 | 1404 | btrfs_finish_extent_commit(trans, root); |
4313b399 | 1405 | |
1a40e23b ZY |
1406 | mutex_lock(&root->fs_info->trans_mutex); |
1407 | ||
2c90e5d6 | 1408 | cur_trans->commit_done = 1; |
b7ec40d7 | 1409 | |
15ee9bc7 | 1410 | root->fs_info->last_trans_committed = cur_trans->transid; |
817d52f8 | 1411 | |
2c90e5d6 | 1412 | wake_up(&cur_trans->commit_wait); |
3de4586c | 1413 | |
13c5a93e | 1414 | list_del_init(&cur_trans->list); |
78fae27e | 1415 | put_transaction(cur_trans); |
79154b1b | 1416 | put_transaction(cur_trans); |
58176a96 | 1417 | |
1abe9b8a | 1418 | trace_btrfs_transaction_commit(root); |
1419 | ||
78fae27e | 1420 | mutex_unlock(&root->fs_info->trans_mutex); |
3de4586c | 1421 | |
9ed74f2d JB |
1422 | if (current->journal_info == trans) |
1423 | current->journal_info = NULL; | |
1424 | ||
2c90e5d6 | 1425 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
24bbcf04 YZ |
1426 | |
1427 | if (current != root->fs_info->transaction_kthread) | |
1428 | btrfs_run_delayed_iputs(root); | |
1429 | ||
79154b1b CM |
1430 | return ret; |
1431 | } | |
1432 | ||
d352ac68 CM |
1433 | /* |
1434 | * interface function to delete all the snapshots we have scheduled for deletion | |
1435 | */ | |
e9d0b13b CM |
1436 | int btrfs_clean_old_snapshots(struct btrfs_root *root) |
1437 | { | |
5d4f98a2 YZ |
1438 | LIST_HEAD(list); |
1439 | struct btrfs_fs_info *fs_info = root->fs_info; | |
1440 | ||
1441 | mutex_lock(&fs_info->trans_mutex); | |
1442 | list_splice_init(&fs_info->dead_roots, &list); | |
1443 | mutex_unlock(&fs_info->trans_mutex); | |
e9d0b13b | 1444 | |
5d4f98a2 YZ |
1445 | while (!list_empty(&list)) { |
1446 | root = list_entry(list.next, struct btrfs_root, root_list); | |
76dda93c YZ |
1447 | list_del(&root->root_list); |
1448 | ||
1449 | if (btrfs_header_backref_rev(root->node) < | |
1450 | BTRFS_MIXED_BACKREF_REV) | |
3fd0a558 | 1451 | btrfs_drop_snapshot(root, NULL, 0); |
76dda93c | 1452 | else |
3fd0a558 | 1453 | btrfs_drop_snapshot(root, NULL, 1); |
e9d0b13b CM |
1454 | } |
1455 | return 0; | |
1456 | } |