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 | { |
2c90e5d6 | 35 | WARN_ON(transaction->use_count == 0); |
79154b1b | 36 | transaction->use_count--; |
78fae27e | 37 | if (transaction->use_count == 0) { |
8fd17795 | 38 | list_del_init(&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 | */ | |
80b6794d | 53 | static noinline int join_transaction(struct btrfs_root *root) |
79154b1b CM |
54 | { |
55 | struct btrfs_transaction *cur_trans; | |
56 | cur_trans = root->fs_info->running_transaction; | |
57 | if (!cur_trans) { | |
2c90e5d6 CM |
58 | cur_trans = kmem_cache_alloc(btrfs_transaction_cachep, |
59 | GFP_NOFS); | |
79154b1b | 60 | BUG_ON(!cur_trans); |
0f7d52f4 | 61 | root->fs_info->generation++; |
15ee9bc7 JB |
62 | cur_trans->num_writers = 1; |
63 | cur_trans->num_joined = 0; | |
0f7d52f4 | 64 | cur_trans->transid = root->fs_info->generation; |
79154b1b CM |
65 | init_waitqueue_head(&cur_trans->writer_wait); |
66 | init_waitqueue_head(&cur_trans->commit_wait); | |
67 | cur_trans->in_commit = 0; | |
f9295749 | 68 | cur_trans->blocked = 0; |
d5719762 | 69 | cur_trans->use_count = 1; |
79154b1b | 70 | cur_trans->commit_done = 0; |
08607c1b | 71 | cur_trans->start_time = get_seconds(); |
56bec294 | 72 | |
6bef4d31 | 73 | cur_trans->delayed_refs.root = RB_ROOT; |
56bec294 | 74 | cur_trans->delayed_refs.num_entries = 0; |
c3e69d58 CM |
75 | cur_trans->delayed_refs.num_heads_ready = 0; |
76 | cur_trans->delayed_refs.num_heads = 0; | |
56bec294 | 77 | cur_trans->delayed_refs.flushing = 0; |
c3e69d58 | 78 | cur_trans->delayed_refs.run_delayed_start = 0; |
56bec294 CM |
79 | spin_lock_init(&cur_trans->delayed_refs.lock); |
80 | ||
3063d29f | 81 | INIT_LIST_HEAD(&cur_trans->pending_snapshots); |
8fd17795 | 82 | list_add_tail(&cur_trans->list, &root->fs_info->trans_list); |
d1310b2e | 83 | extent_io_tree_init(&cur_trans->dirty_pages, |
5f39d397 CM |
84 | root->fs_info->btree_inode->i_mapping, |
85 | GFP_NOFS); | |
48ec2cf8 CM |
86 | spin_lock(&root->fs_info->new_trans_lock); |
87 | root->fs_info->running_transaction = cur_trans; | |
88 | spin_unlock(&root->fs_info->new_trans_lock); | |
15ee9bc7 JB |
89 | } else { |
90 | cur_trans->num_writers++; | |
91 | cur_trans->num_joined++; | |
79154b1b | 92 | } |
15ee9bc7 | 93 | |
79154b1b CM |
94 | return 0; |
95 | } | |
96 | ||
d352ac68 | 97 | /* |
d397712b CM |
98 | * this does all the record keeping required to make sure that a reference |
99 | * counted root is properly recorded in a given transaction. This is required | |
100 | * to make sure the old root from before we joined the transaction is deleted | |
101 | * when the transaction commits | |
d352ac68 | 102 | */ |
5d4f98a2 YZ |
103 | static noinline int record_root_in_trans(struct btrfs_trans_handle *trans, |
104 | struct btrfs_root *root) | |
6702ed49 | 105 | { |
5d4f98a2 | 106 | if (root->ref_cows && root->last_trans < trans->transid) { |
6702ed49 | 107 | WARN_ON(root == root->fs_info->extent_root); |
5d4f98a2 YZ |
108 | WARN_ON(root->commit_root != root->node); |
109 | ||
110 | radix_tree_tag_set(&root->fs_info->fs_roots_radix, | |
111 | (unsigned long)root->root_key.objectid, | |
112 | BTRFS_ROOT_TRANS_TAG); | |
113 | root->last_trans = trans->transid; | |
114 | btrfs_init_reloc_root(trans, root); | |
115 | } | |
116 | return 0; | |
117 | } | |
bcc63abb | 118 | |
5d4f98a2 YZ |
119 | int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, |
120 | struct btrfs_root *root) | |
121 | { | |
122 | if (!root->ref_cows) | |
123 | return 0; | |
bcc63abb | 124 | |
5d4f98a2 YZ |
125 | mutex_lock(&root->fs_info->trans_mutex); |
126 | if (root->last_trans == trans->transid) { | |
127 | mutex_unlock(&root->fs_info->trans_mutex); | |
128 | return 0; | |
6702ed49 | 129 | } |
5d4f98a2 YZ |
130 | |
131 | record_root_in_trans(trans, root); | |
132 | mutex_unlock(&root->fs_info->trans_mutex); | |
6702ed49 CM |
133 | return 0; |
134 | } | |
135 | ||
d352ac68 CM |
136 | /* wait for commit against the current transaction to become unblocked |
137 | * when this is done, it is safe to start a new transaction, but the current | |
138 | * transaction might not be fully on disk. | |
139 | */ | |
37d1aeee | 140 | static void wait_current_trans(struct btrfs_root *root) |
79154b1b | 141 | { |
f9295749 | 142 | struct btrfs_transaction *cur_trans; |
79154b1b | 143 | |
f9295749 | 144 | cur_trans = root->fs_info->running_transaction; |
37d1aeee | 145 | if (cur_trans && cur_trans->blocked) { |
f9295749 CM |
146 | DEFINE_WAIT(wait); |
147 | cur_trans->use_count++; | |
d397712b | 148 | while (1) { |
f9295749 CM |
149 | prepare_to_wait(&root->fs_info->transaction_wait, &wait, |
150 | TASK_UNINTERRUPTIBLE); | |
471fa17d | 151 | if (!cur_trans->blocked) |
f9295749 | 152 | break; |
471fa17d ZL |
153 | mutex_unlock(&root->fs_info->trans_mutex); |
154 | schedule(); | |
155 | mutex_lock(&root->fs_info->trans_mutex); | |
f9295749 | 156 | } |
471fa17d | 157 | finish_wait(&root->fs_info->transaction_wait, &wait); |
f9295749 CM |
158 | put_transaction(cur_trans); |
159 | } | |
37d1aeee CM |
160 | } |
161 | ||
249ac1e5 JB |
162 | enum btrfs_trans_type { |
163 | TRANS_START, | |
164 | TRANS_JOIN, | |
165 | TRANS_USERSPACE, | |
0af3d00b | 166 | TRANS_JOIN_NOLOCK, |
249ac1e5 JB |
167 | }; |
168 | ||
a22285a6 YZ |
169 | static int may_wait_transaction(struct btrfs_root *root, int type) |
170 | { | |
171 | if (!root->fs_info->log_root_recovering && | |
172 | ((type == TRANS_START && !root->fs_info->open_ioctl_trans) || | |
173 | type == TRANS_USERSPACE)) | |
174 | return 1; | |
175 | return 0; | |
176 | } | |
177 | ||
e02119d5 | 178 | static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root, |
a22285a6 | 179 | u64 num_items, int type) |
37d1aeee | 180 | { |
a22285a6 YZ |
181 | struct btrfs_trans_handle *h; |
182 | struct btrfs_transaction *cur_trans; | |
183 | int retries = 0; | |
37d1aeee | 184 | int ret; |
a22285a6 YZ |
185 | again: |
186 | h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS); | |
187 | if (!h) | |
188 | return ERR_PTR(-ENOMEM); | |
37d1aeee | 189 | |
0af3d00b JB |
190 | if (type != TRANS_JOIN_NOLOCK) |
191 | mutex_lock(&root->fs_info->trans_mutex); | |
a22285a6 | 192 | if (may_wait_transaction(root, type)) |
37d1aeee | 193 | wait_current_trans(root); |
a22285a6 | 194 | |
79154b1b CM |
195 | ret = join_transaction(root); |
196 | BUG_ON(ret); | |
0f7d52f4 | 197 | |
a22285a6 YZ |
198 | cur_trans = root->fs_info->running_transaction; |
199 | cur_trans->use_count++; | |
0af3d00b JB |
200 | if (type != TRANS_JOIN_NOLOCK) |
201 | mutex_unlock(&root->fs_info->trans_mutex); | |
a22285a6 YZ |
202 | |
203 | h->transid = cur_trans->transid; | |
204 | h->transaction = cur_trans; | |
79154b1b | 205 | h->blocks_used = 0; |
d2fb3437 | 206 | h->block_group = 0; |
a22285a6 | 207 | h->bytes_reserved = 0; |
56bec294 | 208 | h->delayed_ref_updates = 0; |
f0486c68 | 209 | h->block_rsv = NULL; |
b7ec40d7 | 210 | |
a22285a6 YZ |
211 | smp_mb(); |
212 | if (cur_trans->blocked && may_wait_transaction(root, type)) { | |
213 | btrfs_commit_transaction(h, root); | |
214 | goto again; | |
215 | } | |
216 | ||
217 | if (num_items > 0) { | |
218 | ret = btrfs_trans_reserve_metadata(h, root, num_items, | |
219 | &retries); | |
220 | if (ret == -EAGAIN) { | |
221 | btrfs_commit_transaction(h, root); | |
222 | goto again; | |
223 | } | |
224 | if (ret < 0) { | |
225 | btrfs_end_transaction(h, root); | |
226 | return ERR_PTR(ret); | |
227 | } | |
228 | } | |
9ed74f2d | 229 | |
0af3d00b JB |
230 | if (type != TRANS_JOIN_NOLOCK) |
231 | mutex_lock(&root->fs_info->trans_mutex); | |
5d4f98a2 | 232 | record_root_in_trans(h, root); |
0af3d00b JB |
233 | if (type != TRANS_JOIN_NOLOCK) |
234 | mutex_unlock(&root->fs_info->trans_mutex); | |
a22285a6 YZ |
235 | |
236 | if (!current->journal_info && type != TRANS_USERSPACE) | |
237 | current->journal_info = h; | |
79154b1b CM |
238 | return h; |
239 | } | |
240 | ||
f9295749 | 241 | struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, |
a22285a6 | 242 | int num_items) |
f9295749 | 243 | { |
a22285a6 | 244 | return start_transaction(root, num_items, TRANS_START); |
f9295749 CM |
245 | } |
246 | struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root, | |
247 | int num_blocks) | |
248 | { | |
a22285a6 | 249 | return start_transaction(root, 0, TRANS_JOIN); |
f9295749 CM |
250 | } |
251 | ||
0af3d00b JB |
252 | struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root, |
253 | int num_blocks) | |
254 | { | |
255 | return start_transaction(root, 0, TRANS_JOIN_NOLOCK); | |
256 | } | |
257 | ||
9ca9ee09 SW |
258 | struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *r, |
259 | int num_blocks) | |
260 | { | |
a22285a6 | 261 | return start_transaction(r, 0, TRANS_USERSPACE); |
9ca9ee09 SW |
262 | } |
263 | ||
d352ac68 | 264 | /* wait for a transaction commit to be fully complete */ |
89ce8a63 CM |
265 | static noinline int wait_for_commit(struct btrfs_root *root, |
266 | struct btrfs_transaction *commit) | |
267 | { | |
268 | DEFINE_WAIT(wait); | |
269 | mutex_lock(&root->fs_info->trans_mutex); | |
d397712b | 270 | while (!commit->commit_done) { |
89ce8a63 CM |
271 | prepare_to_wait(&commit->commit_wait, &wait, |
272 | TASK_UNINTERRUPTIBLE); | |
273 | if (commit->commit_done) | |
274 | break; | |
275 | mutex_unlock(&root->fs_info->trans_mutex); | |
276 | schedule(); | |
277 | mutex_lock(&root->fs_info->trans_mutex); | |
278 | } | |
279 | mutex_unlock(&root->fs_info->trans_mutex); | |
280 | finish_wait(&commit->commit_wait, &wait); | |
281 | return 0; | |
282 | } | |
283 | ||
5d4f98a2 | 284 | #if 0 |
d352ac68 | 285 | /* |
d397712b CM |
286 | * rate limit against the drop_snapshot code. This helps to slow down new |
287 | * operations if the drop_snapshot code isn't able to keep up. | |
d352ac68 | 288 | */ |
37d1aeee | 289 | static void throttle_on_drops(struct btrfs_root *root) |
ab78c84d CM |
290 | { |
291 | struct btrfs_fs_info *info = root->fs_info; | |
2dd3e67b | 292 | int harder_count = 0; |
ab78c84d | 293 | |
2dd3e67b | 294 | harder: |
ab78c84d CM |
295 | if (atomic_read(&info->throttles)) { |
296 | DEFINE_WAIT(wait); | |
297 | int thr; | |
ab78c84d CM |
298 | thr = atomic_read(&info->throttle_gen); |
299 | ||
300 | do { | |
301 | prepare_to_wait(&info->transaction_throttle, | |
302 | &wait, TASK_UNINTERRUPTIBLE); | |
303 | if (!atomic_read(&info->throttles)) { | |
304 | finish_wait(&info->transaction_throttle, &wait); | |
305 | break; | |
306 | } | |
307 | schedule(); | |
308 | finish_wait(&info->transaction_throttle, &wait); | |
309 | } while (thr == atomic_read(&info->throttle_gen)); | |
2dd3e67b CM |
310 | harder_count++; |
311 | ||
312 | if (root->fs_info->total_ref_cache_size > 1 * 1024 * 1024 && | |
313 | harder_count < 2) | |
314 | goto harder; | |
315 | ||
316 | if (root->fs_info->total_ref_cache_size > 5 * 1024 * 1024 && | |
317 | harder_count < 10) | |
318 | goto harder; | |
319 | ||
320 | if (root->fs_info->total_ref_cache_size > 10 * 1024 * 1024 && | |
321 | harder_count < 20) | |
322 | goto harder; | |
ab78c84d CM |
323 | } |
324 | } | |
5d4f98a2 | 325 | #endif |
ab78c84d | 326 | |
37d1aeee CM |
327 | void btrfs_throttle(struct btrfs_root *root) |
328 | { | |
329 | mutex_lock(&root->fs_info->trans_mutex); | |
9ca9ee09 SW |
330 | if (!root->fs_info->open_ioctl_trans) |
331 | wait_current_trans(root); | |
37d1aeee | 332 | mutex_unlock(&root->fs_info->trans_mutex); |
37d1aeee CM |
333 | } |
334 | ||
8929ecfa YZ |
335 | static int should_end_transaction(struct btrfs_trans_handle *trans, |
336 | struct btrfs_root *root) | |
337 | { | |
338 | int ret; | |
339 | ret = btrfs_block_rsv_check(trans, root, | |
340 | &root->fs_info->global_block_rsv, 0, 5); | |
341 | return ret ? 1 : 0; | |
342 | } | |
343 | ||
344 | int btrfs_should_end_transaction(struct btrfs_trans_handle *trans, | |
345 | struct btrfs_root *root) | |
346 | { | |
347 | struct btrfs_transaction *cur_trans = trans->transaction; | |
348 | int updates; | |
349 | ||
350 | if (cur_trans->blocked || cur_trans->delayed_refs.flushing) | |
351 | return 1; | |
352 | ||
353 | updates = trans->delayed_ref_updates; | |
354 | trans->delayed_ref_updates = 0; | |
355 | if (updates) | |
356 | btrfs_run_delayed_refs(trans, root, updates); | |
357 | ||
358 | return should_end_transaction(trans, root); | |
359 | } | |
360 | ||
89ce8a63 | 361 | static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, |
0af3d00b | 362 | struct btrfs_root *root, int throttle, int lock) |
79154b1b | 363 | { |
8929ecfa | 364 | struct btrfs_transaction *cur_trans = trans->transaction; |
ab78c84d | 365 | struct btrfs_fs_info *info = root->fs_info; |
c3e69d58 CM |
366 | int count = 0; |
367 | ||
368 | while (count < 4) { | |
369 | unsigned long cur = trans->delayed_ref_updates; | |
370 | trans->delayed_ref_updates = 0; | |
371 | if (cur && | |
372 | trans->transaction->delayed_refs.num_heads_ready > 64) { | |
373 | trans->delayed_ref_updates = 0; | |
b7ec40d7 CM |
374 | |
375 | /* | |
376 | * do a full flush if the transaction is trying | |
377 | * to close | |
378 | */ | |
379 | if (trans->transaction->delayed_refs.flushing) | |
380 | cur = 0; | |
c3e69d58 CM |
381 | btrfs_run_delayed_refs(trans, root, cur); |
382 | } else { | |
383 | break; | |
384 | } | |
385 | count++; | |
56bec294 CM |
386 | } |
387 | ||
a22285a6 YZ |
388 | btrfs_trans_release_metadata(trans, root); |
389 | ||
0af3d00b | 390 | if (lock && !root->fs_info->open_ioctl_trans && |
8929ecfa YZ |
391 | should_end_transaction(trans, root)) |
392 | trans->transaction->blocked = 1; | |
393 | ||
0af3d00b | 394 | if (lock && cur_trans->blocked && !cur_trans->in_commit) { |
8929ecfa YZ |
395 | if (throttle) |
396 | return btrfs_commit_transaction(trans, root); | |
397 | else | |
398 | wake_up_process(info->transaction_kthread); | |
399 | } | |
400 | ||
0af3d00b JB |
401 | if (lock) |
402 | mutex_lock(&info->trans_mutex); | |
8929ecfa | 403 | WARN_ON(cur_trans != info->running_transaction); |
d5719762 | 404 | WARN_ON(cur_trans->num_writers < 1); |
ccd467d6 | 405 | cur_trans->num_writers--; |
89ce8a63 | 406 | |
79154b1b CM |
407 | if (waitqueue_active(&cur_trans->writer_wait)) |
408 | wake_up(&cur_trans->writer_wait); | |
79154b1b | 409 | put_transaction(cur_trans); |
0af3d00b JB |
410 | if (lock) |
411 | mutex_unlock(&info->trans_mutex); | |
9ed74f2d JB |
412 | |
413 | if (current->journal_info == trans) | |
414 | current->journal_info = NULL; | |
d6025579 | 415 | memset(trans, 0, sizeof(*trans)); |
2c90e5d6 | 416 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
ab78c84d | 417 | |
24bbcf04 YZ |
418 | if (throttle) |
419 | btrfs_run_delayed_iputs(root); | |
420 | ||
79154b1b CM |
421 | return 0; |
422 | } | |
423 | ||
89ce8a63 CM |
424 | int btrfs_end_transaction(struct btrfs_trans_handle *trans, |
425 | struct btrfs_root *root) | |
426 | { | |
0af3d00b | 427 | return __btrfs_end_transaction(trans, root, 0, 1); |
89ce8a63 CM |
428 | } |
429 | ||
430 | int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans, | |
431 | struct btrfs_root *root) | |
432 | { | |
0af3d00b JB |
433 | return __btrfs_end_transaction(trans, root, 1, 1); |
434 | } | |
435 | ||
436 | int btrfs_end_transaction_nolock(struct btrfs_trans_handle *trans, | |
437 | struct btrfs_root *root) | |
438 | { | |
439 | return __btrfs_end_transaction(trans, root, 0, 0); | |
89ce8a63 CM |
440 | } |
441 | ||
d352ac68 CM |
442 | /* |
443 | * when btree blocks are allocated, they have some corresponding bits set for | |
444 | * them in one of two extent_io trees. This is used to make sure all of | |
690587d1 | 445 | * those extents are sent to disk but does not wait on them |
d352ac68 | 446 | */ |
690587d1 | 447 | int btrfs_write_marked_extents(struct btrfs_root *root, |
8cef4e16 | 448 | struct extent_io_tree *dirty_pages, int mark) |
79154b1b | 449 | { |
7c4452b9 | 450 | int ret; |
777e6bd7 | 451 | int err = 0; |
7c4452b9 CM |
452 | int werr = 0; |
453 | struct page *page; | |
7c4452b9 | 454 | struct inode *btree_inode = root->fs_info->btree_inode; |
777e6bd7 | 455 | u64 start = 0; |
5f39d397 CM |
456 | u64 end; |
457 | unsigned long index; | |
7c4452b9 | 458 | |
d397712b | 459 | while (1) { |
777e6bd7 | 460 | ret = find_first_extent_bit(dirty_pages, start, &start, &end, |
8cef4e16 | 461 | mark); |
5f39d397 | 462 | if (ret) |
7c4452b9 | 463 | break; |
d397712b | 464 | while (start <= end) { |
777e6bd7 CM |
465 | cond_resched(); |
466 | ||
5f39d397 | 467 | index = start >> PAGE_CACHE_SHIFT; |
35ebb934 | 468 | start = (u64)(index + 1) << PAGE_CACHE_SHIFT; |
4bef0848 | 469 | page = find_get_page(btree_inode->i_mapping, index); |
7c4452b9 CM |
470 | if (!page) |
471 | continue; | |
4bef0848 CM |
472 | |
473 | btree_lock_page_hook(page); | |
474 | if (!page->mapping) { | |
475 | unlock_page(page); | |
476 | page_cache_release(page); | |
477 | continue; | |
478 | } | |
479 | ||
6702ed49 CM |
480 | if (PageWriteback(page)) { |
481 | if (PageDirty(page)) | |
482 | wait_on_page_writeback(page); | |
483 | else { | |
484 | unlock_page(page); | |
485 | page_cache_release(page); | |
486 | continue; | |
487 | } | |
488 | } | |
7c4452b9 CM |
489 | err = write_one_page(page, 0); |
490 | if (err) | |
491 | werr = err; | |
492 | page_cache_release(page); | |
493 | } | |
494 | } | |
690587d1 CM |
495 | if (err) |
496 | werr = err; | |
497 | return werr; | |
498 | } | |
499 | ||
500 | /* | |
501 | * when btree blocks are allocated, they have some corresponding bits set for | |
502 | * them in one of two extent_io trees. This is used to make sure all of | |
503 | * those extents are on disk for transaction or log commit. We wait | |
504 | * on all the pages and clear them from the dirty pages state tree | |
505 | */ | |
506 | int btrfs_wait_marked_extents(struct btrfs_root *root, | |
8cef4e16 | 507 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 CM |
508 | { |
509 | int ret; | |
510 | int err = 0; | |
511 | int werr = 0; | |
512 | struct page *page; | |
513 | struct inode *btree_inode = root->fs_info->btree_inode; | |
514 | u64 start = 0; | |
515 | u64 end; | |
516 | unsigned long index; | |
517 | ||
d397712b | 518 | while (1) { |
8cef4e16 YZ |
519 | ret = find_first_extent_bit(dirty_pages, start, &start, &end, |
520 | mark); | |
777e6bd7 CM |
521 | if (ret) |
522 | break; | |
523 | ||
8cef4e16 | 524 | clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS); |
d397712b | 525 | while (start <= end) { |
777e6bd7 CM |
526 | index = start >> PAGE_CACHE_SHIFT; |
527 | start = (u64)(index + 1) << PAGE_CACHE_SHIFT; | |
528 | page = find_get_page(btree_inode->i_mapping, index); | |
529 | if (!page) | |
530 | continue; | |
531 | if (PageDirty(page)) { | |
4bef0848 CM |
532 | btree_lock_page_hook(page); |
533 | wait_on_page_writeback(page); | |
777e6bd7 CM |
534 | err = write_one_page(page, 0); |
535 | if (err) | |
536 | werr = err; | |
537 | } | |
105d931d | 538 | wait_on_page_writeback(page); |
777e6bd7 CM |
539 | page_cache_release(page); |
540 | cond_resched(); | |
541 | } | |
542 | } | |
7c4452b9 CM |
543 | if (err) |
544 | werr = err; | |
545 | return werr; | |
79154b1b CM |
546 | } |
547 | ||
690587d1 CM |
548 | /* |
549 | * when btree blocks are allocated, they have some corresponding bits set for | |
550 | * them in one of two extent_io trees. This is used to make sure all of | |
551 | * those extents are on disk for transaction or log commit | |
552 | */ | |
553 | int btrfs_write_and_wait_marked_extents(struct btrfs_root *root, | |
8cef4e16 | 554 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 CM |
555 | { |
556 | int ret; | |
557 | int ret2; | |
558 | ||
8cef4e16 YZ |
559 | ret = btrfs_write_marked_extents(root, dirty_pages, mark); |
560 | ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark); | |
690587d1 CM |
561 | return ret || ret2; |
562 | } | |
563 | ||
d0c803c4 CM |
564 | int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans, |
565 | struct btrfs_root *root) | |
566 | { | |
567 | if (!trans || !trans->transaction) { | |
568 | struct inode *btree_inode; | |
569 | btree_inode = root->fs_info->btree_inode; | |
570 | return filemap_write_and_wait(btree_inode->i_mapping); | |
571 | } | |
572 | return btrfs_write_and_wait_marked_extents(root, | |
8cef4e16 YZ |
573 | &trans->transaction->dirty_pages, |
574 | EXTENT_DIRTY); | |
d0c803c4 CM |
575 | } |
576 | ||
d352ac68 CM |
577 | /* |
578 | * this is used to update the root pointer in the tree of tree roots. | |
579 | * | |
580 | * But, in the case of the extent allocation tree, updating the root | |
581 | * pointer may allocate blocks which may change the root of the extent | |
582 | * allocation tree. | |
583 | * | |
584 | * So, this loops and repeats and makes sure the cowonly root didn't | |
585 | * change while the root pointer was being updated in the metadata. | |
586 | */ | |
0b86a832 CM |
587 | static int update_cowonly_root(struct btrfs_trans_handle *trans, |
588 | struct btrfs_root *root) | |
79154b1b CM |
589 | { |
590 | int ret; | |
0b86a832 | 591 | u64 old_root_bytenr; |
86b9f2ec | 592 | u64 old_root_used; |
0b86a832 | 593 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
79154b1b | 594 | |
86b9f2ec | 595 | old_root_used = btrfs_root_used(&root->root_item); |
0b86a832 | 596 | btrfs_write_dirty_block_groups(trans, root); |
56bec294 | 597 | |
d397712b | 598 | while (1) { |
0b86a832 | 599 | old_root_bytenr = btrfs_root_bytenr(&root->root_item); |
86b9f2ec YZ |
600 | if (old_root_bytenr == root->node->start && |
601 | old_root_used == btrfs_root_used(&root->root_item)) | |
79154b1b | 602 | break; |
87ef2bb4 | 603 | |
5d4f98a2 | 604 | btrfs_set_root_node(&root->root_item, root->node); |
79154b1b | 605 | ret = btrfs_update_root(trans, tree_root, |
0b86a832 CM |
606 | &root->root_key, |
607 | &root->root_item); | |
79154b1b | 608 | BUG_ON(ret); |
56bec294 | 609 | |
86b9f2ec | 610 | old_root_used = btrfs_root_used(&root->root_item); |
4a8c9a62 | 611 | ret = btrfs_write_dirty_block_groups(trans, root); |
56bec294 | 612 | BUG_ON(ret); |
0b86a832 | 613 | } |
276e680d YZ |
614 | |
615 | if (root != root->fs_info->extent_root) | |
616 | switch_commit_root(root); | |
617 | ||
0b86a832 CM |
618 | return 0; |
619 | } | |
620 | ||
d352ac68 CM |
621 | /* |
622 | * update all the cowonly tree roots on disk | |
623 | */ | |
5d4f98a2 YZ |
624 | static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans, |
625 | struct btrfs_root *root) | |
0b86a832 CM |
626 | { |
627 | struct btrfs_fs_info *fs_info = root->fs_info; | |
628 | struct list_head *next; | |
84234f3a | 629 | struct extent_buffer *eb; |
56bec294 | 630 | int ret; |
84234f3a | 631 | |
56bec294 CM |
632 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
633 | BUG_ON(ret); | |
87ef2bb4 | 634 | |
84234f3a | 635 | eb = btrfs_lock_root_node(fs_info->tree_root); |
9fa8cfe7 | 636 | btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 0, &eb); |
84234f3a YZ |
637 | btrfs_tree_unlock(eb); |
638 | free_extent_buffer(eb); | |
0b86a832 | 639 | |
56bec294 CM |
640 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
641 | BUG_ON(ret); | |
87ef2bb4 | 642 | |
d397712b | 643 | while (!list_empty(&fs_info->dirty_cowonly_roots)) { |
0b86a832 CM |
644 | next = fs_info->dirty_cowonly_roots.next; |
645 | list_del_init(next); | |
646 | root = list_entry(next, struct btrfs_root, dirty_list); | |
87ef2bb4 | 647 | |
0b86a832 | 648 | update_cowonly_root(trans, root); |
79154b1b | 649 | } |
276e680d YZ |
650 | |
651 | down_write(&fs_info->extent_commit_sem); | |
652 | switch_commit_root(fs_info->extent_root); | |
653 | up_write(&fs_info->extent_commit_sem); | |
654 | ||
79154b1b CM |
655 | return 0; |
656 | } | |
657 | ||
d352ac68 CM |
658 | /* |
659 | * dead roots are old snapshots that need to be deleted. This allocates | |
660 | * a dirty root struct and adds it into the list of dead roots that need to | |
661 | * be deleted | |
662 | */ | |
5d4f98a2 | 663 | int btrfs_add_dead_root(struct btrfs_root *root) |
5eda7b5e | 664 | { |
b48652c1 | 665 | mutex_lock(&root->fs_info->trans_mutex); |
5d4f98a2 | 666 | list_add(&root->root_list, &root->fs_info->dead_roots); |
b48652c1 | 667 | mutex_unlock(&root->fs_info->trans_mutex); |
5eda7b5e CM |
668 | return 0; |
669 | } | |
670 | ||
d352ac68 | 671 | /* |
5d4f98a2 | 672 | * update all the cowonly tree roots on disk |
d352ac68 | 673 | */ |
5d4f98a2 YZ |
674 | static noinline int commit_fs_roots(struct btrfs_trans_handle *trans, |
675 | struct btrfs_root *root) | |
0f7d52f4 | 676 | { |
0f7d52f4 | 677 | struct btrfs_root *gang[8]; |
5d4f98a2 | 678 | struct btrfs_fs_info *fs_info = root->fs_info; |
0f7d52f4 CM |
679 | int i; |
680 | int ret; | |
54aa1f4d CM |
681 | int err = 0; |
682 | ||
d397712b | 683 | while (1) { |
5d4f98a2 YZ |
684 | ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, |
685 | (void **)gang, 0, | |
0f7d52f4 CM |
686 | ARRAY_SIZE(gang), |
687 | BTRFS_ROOT_TRANS_TAG); | |
688 | if (ret == 0) | |
689 | break; | |
690 | for (i = 0; i < ret; i++) { | |
691 | root = gang[i]; | |
5d4f98a2 YZ |
692 | radix_tree_tag_clear(&fs_info->fs_roots_radix, |
693 | (unsigned long)root->root_key.objectid, | |
694 | BTRFS_ROOT_TRANS_TAG); | |
31153d81 | 695 | |
e02119d5 | 696 | btrfs_free_log(trans, root); |
5d4f98a2 | 697 | btrfs_update_reloc_root(trans, root); |
d68fc57b | 698 | btrfs_orphan_commit_root(trans, root); |
bcc63abb | 699 | |
978d910d | 700 | if (root->commit_root != root->node) { |
817d52f8 | 701 | switch_commit_root(root); |
978d910d YZ |
702 | btrfs_set_root_node(&root->root_item, |
703 | root->node); | |
704 | } | |
5d4f98a2 | 705 | |
5d4f98a2 | 706 | err = btrfs_update_root(trans, fs_info->tree_root, |
0f7d52f4 CM |
707 | &root->root_key, |
708 | &root->root_item); | |
54aa1f4d CM |
709 | if (err) |
710 | break; | |
0f7d52f4 CM |
711 | } |
712 | } | |
54aa1f4d | 713 | return err; |
0f7d52f4 CM |
714 | } |
715 | ||
d352ac68 CM |
716 | /* |
717 | * defrag a given btree. If cacheonly == 1, this won't read from the disk, | |
718 | * otherwise every leaf in the btree is read and defragged. | |
719 | */ | |
e9d0b13b CM |
720 | int btrfs_defrag_root(struct btrfs_root *root, int cacheonly) |
721 | { | |
722 | struct btrfs_fs_info *info = root->fs_info; | |
e9d0b13b | 723 | struct btrfs_trans_handle *trans; |
8929ecfa | 724 | int ret; |
d3c2fdcf | 725 | unsigned long nr; |
e9d0b13b | 726 | |
8929ecfa | 727 | if (xchg(&root->defrag_running, 1)) |
e9d0b13b | 728 | return 0; |
8929ecfa | 729 | |
6b80053d | 730 | while (1) { |
8929ecfa YZ |
731 | trans = btrfs_start_transaction(root, 0); |
732 | if (IS_ERR(trans)) | |
733 | return PTR_ERR(trans); | |
734 | ||
e9d0b13b | 735 | ret = btrfs_defrag_leaves(trans, root, cacheonly); |
8929ecfa | 736 | |
d3c2fdcf | 737 | nr = trans->blocks_used; |
e9d0b13b | 738 | btrfs_end_transaction(trans, root); |
d3c2fdcf | 739 | btrfs_btree_balance_dirty(info->tree_root, nr); |
e9d0b13b CM |
740 | cond_resched(); |
741 | ||
3f157a2f | 742 | if (root->fs_info->closing || ret != -EAGAIN) |
e9d0b13b CM |
743 | break; |
744 | } | |
745 | root->defrag_running = 0; | |
8929ecfa | 746 | return ret; |
e9d0b13b CM |
747 | } |
748 | ||
2c47e605 | 749 | #if 0 |
b7ec40d7 CM |
750 | /* |
751 | * when dropping snapshots, we generate a ton of delayed refs, and it makes | |
752 | * sense not to join the transaction while it is trying to flush the current | |
753 | * queue of delayed refs out. | |
754 | * | |
755 | * This is used by the drop snapshot code only | |
756 | */ | |
757 | static noinline int wait_transaction_pre_flush(struct btrfs_fs_info *info) | |
758 | { | |
759 | DEFINE_WAIT(wait); | |
760 | ||
761 | mutex_lock(&info->trans_mutex); | |
762 | while (info->running_transaction && | |
763 | info->running_transaction->delayed_refs.flushing) { | |
764 | prepare_to_wait(&info->transaction_wait, &wait, | |
765 | TASK_UNINTERRUPTIBLE); | |
766 | mutex_unlock(&info->trans_mutex); | |
59bc5c75 | 767 | |
b7ec40d7 | 768 | schedule(); |
59bc5c75 | 769 | |
b7ec40d7 CM |
770 | mutex_lock(&info->trans_mutex); |
771 | finish_wait(&info->transaction_wait, &wait); | |
772 | } | |
773 | mutex_unlock(&info->trans_mutex); | |
774 | return 0; | |
775 | } | |
776 | ||
d352ac68 CM |
777 | /* |
778 | * Given a list of roots that need to be deleted, call btrfs_drop_snapshot on | |
779 | * all of them | |
780 | */ | |
5d4f98a2 | 781 | int btrfs_drop_dead_root(struct btrfs_root *root) |
0f7d52f4 | 782 | { |
0f7d52f4 | 783 | struct btrfs_trans_handle *trans; |
5d4f98a2 | 784 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
d3c2fdcf | 785 | unsigned long nr; |
5d4f98a2 | 786 | int ret; |
58176a96 | 787 | |
5d4f98a2 YZ |
788 | while (1) { |
789 | /* | |
790 | * we don't want to jump in and create a bunch of | |
791 | * delayed refs if the transaction is starting to close | |
792 | */ | |
793 | wait_transaction_pre_flush(tree_root->fs_info); | |
794 | trans = btrfs_start_transaction(tree_root, 1); | |
a2135011 | 795 | |
5d4f98a2 YZ |
796 | /* |
797 | * we've joined a transaction, make sure it isn't | |
798 | * closing right now | |
799 | */ | |
800 | if (trans->transaction->delayed_refs.flushing) { | |
801 | btrfs_end_transaction(trans, tree_root); | |
802 | continue; | |
9f3a7427 | 803 | } |
58176a96 | 804 | |
5d4f98a2 YZ |
805 | ret = btrfs_drop_snapshot(trans, root); |
806 | if (ret != -EAGAIN) | |
807 | break; | |
a2135011 | 808 | |
5d4f98a2 YZ |
809 | ret = btrfs_update_root(trans, tree_root, |
810 | &root->root_key, | |
811 | &root->root_item); | |
812 | if (ret) | |
54aa1f4d | 813 | break; |
bcc63abb | 814 | |
d3c2fdcf | 815 | nr = trans->blocks_used; |
0f7d52f4 CM |
816 | ret = btrfs_end_transaction(trans, tree_root); |
817 | BUG_ON(ret); | |
5eda7b5e | 818 | |
d3c2fdcf | 819 | btrfs_btree_balance_dirty(tree_root, nr); |
4dc11904 | 820 | cond_resched(); |
0f7d52f4 | 821 | } |
5d4f98a2 YZ |
822 | BUG_ON(ret); |
823 | ||
824 | ret = btrfs_del_root(trans, tree_root, &root->root_key); | |
825 | BUG_ON(ret); | |
826 | ||
827 | nr = trans->blocks_used; | |
828 | ret = btrfs_end_transaction(trans, tree_root); | |
829 | BUG_ON(ret); | |
830 | ||
831 | free_extent_buffer(root->node); | |
832 | free_extent_buffer(root->commit_root); | |
833 | kfree(root); | |
834 | ||
835 | btrfs_btree_balance_dirty(tree_root, nr); | |
54aa1f4d | 836 | return ret; |
0f7d52f4 | 837 | } |
2c47e605 | 838 | #endif |
0f7d52f4 | 839 | |
d352ac68 CM |
840 | /* |
841 | * new snapshots need to be created at a very specific time in the | |
842 | * transaction commit. This does the actual creation | |
843 | */ | |
80b6794d | 844 | static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, |
3063d29f CM |
845 | struct btrfs_fs_info *fs_info, |
846 | struct btrfs_pending_snapshot *pending) | |
847 | { | |
848 | struct btrfs_key key; | |
80b6794d | 849 | struct btrfs_root_item *new_root_item; |
3063d29f CM |
850 | struct btrfs_root *tree_root = fs_info->tree_root; |
851 | struct btrfs_root *root = pending->root; | |
6bdb72de SW |
852 | struct btrfs_root *parent_root; |
853 | struct inode *parent_inode; | |
a22285a6 | 854 | struct dentry *dentry; |
3063d29f | 855 | struct extent_buffer *tmp; |
925baedd | 856 | struct extent_buffer *old; |
3063d29f | 857 | int ret; |
d68fc57b YZ |
858 | int retries = 0; |
859 | u64 to_reserve = 0; | |
6bdb72de | 860 | u64 index = 0; |
a22285a6 | 861 | u64 objectid; |
3063d29f | 862 | |
80b6794d CM |
863 | new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS); |
864 | if (!new_root_item) { | |
a22285a6 | 865 | pending->error = -ENOMEM; |
80b6794d CM |
866 | goto fail; |
867 | } | |
a22285a6 | 868 | |
3063d29f | 869 | ret = btrfs_find_free_objectid(trans, tree_root, 0, &objectid); |
a22285a6 YZ |
870 | if (ret) { |
871 | pending->error = ret; | |
3063d29f | 872 | goto fail; |
a22285a6 | 873 | } |
3063d29f | 874 | |
3fd0a558 | 875 | btrfs_reloc_pre_snapshot(trans, pending, &to_reserve); |
d68fc57b YZ |
876 | btrfs_orphan_pre_snapshot(trans, pending, &to_reserve); |
877 | ||
878 | if (to_reserve > 0) { | |
879 | ret = btrfs_block_rsv_add(trans, root, &pending->block_rsv, | |
880 | to_reserve, &retries); | |
881 | if (ret) { | |
882 | pending->error = ret; | |
883 | goto fail; | |
884 | } | |
885 | } | |
886 | ||
3063d29f | 887 | key.objectid = objectid; |
a22285a6 YZ |
888 | key.offset = (u64)-1; |
889 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3063d29f | 890 | |
a22285a6 | 891 | trans->block_rsv = &pending->block_rsv; |
3de4586c | 892 | |
a22285a6 YZ |
893 | dentry = pending->dentry; |
894 | parent_inode = dentry->d_parent->d_inode; | |
895 | parent_root = BTRFS_I(parent_inode)->root; | |
6bdb72de | 896 | record_root_in_trans(trans, parent_root); |
a22285a6 | 897 | |
3063d29f CM |
898 | /* |
899 | * insert the directory item | |
900 | */ | |
3de4586c | 901 | ret = btrfs_set_inode_index(parent_inode, &index); |
6bdb72de | 902 | BUG_ON(ret); |
0660b5af | 903 | ret = btrfs_insert_dir_item(trans, parent_root, |
a22285a6 YZ |
904 | dentry->d_name.name, dentry->d_name.len, |
905 | parent_inode->i_ino, &key, | |
906 | BTRFS_FT_DIR, index); | |
6bdb72de | 907 | BUG_ON(ret); |
0660b5af | 908 | |
a22285a6 YZ |
909 | btrfs_i_size_write(parent_inode, parent_inode->i_size + |
910 | dentry->d_name.len * 2); | |
52c26179 YZ |
911 | ret = btrfs_update_inode(trans, parent_root, parent_inode); |
912 | BUG_ON(ret); | |
913 | ||
6bdb72de SW |
914 | record_root_in_trans(trans, root); |
915 | btrfs_set_root_last_snapshot(&root->root_item, trans->transid); | |
916 | memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); | |
917 | ||
918 | old = btrfs_lock_root_node(root); | |
919 | btrfs_cow_block(trans, root, old, NULL, 0, &old); | |
920 | btrfs_set_lock_blocking(old); | |
921 | ||
922 | btrfs_copy_root(trans, root, old, &tmp, objectid); | |
923 | btrfs_tree_unlock(old); | |
924 | free_extent_buffer(old); | |
925 | ||
926 | btrfs_set_root_node(new_root_item, tmp); | |
a22285a6 YZ |
927 | /* record when the snapshot was created in key.offset */ |
928 | key.offset = trans->transid; | |
929 | ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); | |
6bdb72de SW |
930 | btrfs_tree_unlock(tmp); |
931 | free_extent_buffer(tmp); | |
a22285a6 | 932 | BUG_ON(ret); |
6bdb72de | 933 | |
a22285a6 YZ |
934 | /* |
935 | * insert root back/forward references | |
936 | */ | |
937 | ret = btrfs_add_root_ref(trans, tree_root, objectid, | |
0660b5af | 938 | parent_root->root_key.objectid, |
a22285a6 YZ |
939 | parent_inode->i_ino, index, |
940 | dentry->d_name.name, dentry->d_name.len); | |
0660b5af CM |
941 | BUG_ON(ret); |
942 | ||
a22285a6 YZ |
943 | key.offset = (u64)-1; |
944 | pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key); | |
945 | BUG_ON(IS_ERR(pending->snap)); | |
d68fc57b | 946 | |
3fd0a558 | 947 | btrfs_reloc_post_snapshot(trans, pending); |
d68fc57b | 948 | btrfs_orphan_post_snapshot(trans, pending); |
3063d29f | 949 | fail: |
6bdb72de | 950 | kfree(new_root_item); |
a22285a6 YZ |
951 | btrfs_block_rsv_release(root, &pending->block_rsv, (u64)-1); |
952 | return 0; | |
3063d29f CM |
953 | } |
954 | ||
d352ac68 CM |
955 | /* |
956 | * create all the snapshots we've scheduled for creation | |
957 | */ | |
80b6794d CM |
958 | static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans, |
959 | struct btrfs_fs_info *fs_info) | |
3de4586c CM |
960 | { |
961 | struct btrfs_pending_snapshot *pending; | |
962 | struct list_head *head = &trans->transaction->pending_snapshots; | |
3de4586c CM |
963 | int ret; |
964 | ||
c6e30871 | 965 | list_for_each_entry(pending, head, list) { |
3de4586c CM |
966 | ret = create_pending_snapshot(trans, fs_info, pending); |
967 | BUG_ON(ret); | |
968 | } | |
969 | return 0; | |
970 | } | |
971 | ||
5d4f98a2 YZ |
972 | static void update_super_roots(struct btrfs_root *root) |
973 | { | |
974 | struct btrfs_root_item *root_item; | |
975 | struct btrfs_super_block *super; | |
976 | ||
977 | super = &root->fs_info->super_copy; | |
978 | ||
979 | root_item = &root->fs_info->chunk_root->root_item; | |
980 | super->chunk_root = root_item->bytenr; | |
981 | super->chunk_root_generation = root_item->generation; | |
982 | super->chunk_root_level = root_item->level; | |
983 | ||
984 | root_item = &root->fs_info->tree_root->root_item; | |
985 | super->root = root_item->bytenr; | |
986 | super->generation = root_item->generation; | |
987 | super->root_level = root_item->level; | |
0af3d00b JB |
988 | if (super->cache_generation != 0 || btrfs_test_opt(root, SPACE_CACHE)) |
989 | super->cache_generation = root_item->generation; | |
5d4f98a2 YZ |
990 | } |
991 | ||
f36f3042 CM |
992 | int btrfs_transaction_in_commit(struct btrfs_fs_info *info) |
993 | { | |
994 | int ret = 0; | |
995 | spin_lock(&info->new_trans_lock); | |
996 | if (info->running_transaction) | |
997 | ret = info->running_transaction->in_commit; | |
998 | spin_unlock(&info->new_trans_lock); | |
999 | return ret; | |
1000 | } | |
1001 | ||
8929ecfa YZ |
1002 | int btrfs_transaction_blocked(struct btrfs_fs_info *info) |
1003 | { | |
1004 | int ret = 0; | |
1005 | spin_lock(&info->new_trans_lock); | |
1006 | if (info->running_transaction) | |
1007 | ret = info->running_transaction->blocked; | |
1008 | spin_unlock(&info->new_trans_lock); | |
1009 | return ret; | |
1010 | } | |
1011 | ||
79154b1b CM |
1012 | int btrfs_commit_transaction(struct btrfs_trans_handle *trans, |
1013 | struct btrfs_root *root) | |
1014 | { | |
15ee9bc7 JB |
1015 | unsigned long joined = 0; |
1016 | unsigned long timeout = 1; | |
79154b1b | 1017 | struct btrfs_transaction *cur_trans; |
8fd17795 | 1018 | struct btrfs_transaction *prev_trans = NULL; |
79154b1b | 1019 | DEFINE_WAIT(wait); |
15ee9bc7 | 1020 | int ret; |
89573b9c CM |
1021 | int should_grow = 0; |
1022 | unsigned long now = get_seconds(); | |
dccae999 | 1023 | int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT); |
79154b1b | 1024 | |
5a3f23d5 CM |
1025 | btrfs_run_ordered_operations(root, 0); |
1026 | ||
56bec294 CM |
1027 | /* make a pass through all the delayed refs we have so far |
1028 | * any runnings procs may add more while we are here | |
1029 | */ | |
1030 | ret = btrfs_run_delayed_refs(trans, root, 0); | |
1031 | BUG_ON(ret); | |
1032 | ||
a22285a6 YZ |
1033 | btrfs_trans_release_metadata(trans, root); |
1034 | ||
b7ec40d7 | 1035 | cur_trans = trans->transaction; |
56bec294 CM |
1036 | /* |
1037 | * set the flushing flag so procs in this transaction have to | |
1038 | * start sending their work down. | |
1039 | */ | |
b7ec40d7 | 1040 | cur_trans->delayed_refs.flushing = 1; |
56bec294 | 1041 | |
c3e69d58 | 1042 | ret = btrfs_run_delayed_refs(trans, root, 0); |
56bec294 CM |
1043 | BUG_ON(ret); |
1044 | ||
79154b1b | 1045 | mutex_lock(&root->fs_info->trans_mutex); |
b7ec40d7 CM |
1046 | if (cur_trans->in_commit) { |
1047 | cur_trans->use_count++; | |
ccd467d6 | 1048 | mutex_unlock(&root->fs_info->trans_mutex); |
79154b1b | 1049 | btrfs_end_transaction(trans, root); |
ccd467d6 | 1050 | |
79154b1b CM |
1051 | ret = wait_for_commit(root, cur_trans); |
1052 | BUG_ON(ret); | |
15ee9bc7 JB |
1053 | |
1054 | mutex_lock(&root->fs_info->trans_mutex); | |
79154b1b | 1055 | put_transaction(cur_trans); |
15ee9bc7 JB |
1056 | mutex_unlock(&root->fs_info->trans_mutex); |
1057 | ||
79154b1b CM |
1058 | return 0; |
1059 | } | |
4313b399 | 1060 | |
2c90e5d6 | 1061 | trans->transaction->in_commit = 1; |
f9295749 | 1062 | trans->transaction->blocked = 1; |
ccd467d6 CM |
1063 | if (cur_trans->list.prev != &root->fs_info->trans_list) { |
1064 | prev_trans = list_entry(cur_trans->list.prev, | |
1065 | struct btrfs_transaction, list); | |
1066 | if (!prev_trans->commit_done) { | |
1067 | prev_trans->use_count++; | |
ccd467d6 CM |
1068 | mutex_unlock(&root->fs_info->trans_mutex); |
1069 | ||
1070 | wait_for_commit(root, prev_trans); | |
ccd467d6 | 1071 | |
ccd467d6 | 1072 | mutex_lock(&root->fs_info->trans_mutex); |
15ee9bc7 | 1073 | put_transaction(prev_trans); |
ccd467d6 CM |
1074 | } |
1075 | } | |
15ee9bc7 | 1076 | |
89573b9c CM |
1077 | if (now < cur_trans->start_time || now - cur_trans->start_time < 1) |
1078 | should_grow = 1; | |
1079 | ||
15ee9bc7 | 1080 | do { |
7ea394f1 | 1081 | int snap_pending = 0; |
15ee9bc7 | 1082 | joined = cur_trans->num_joined; |
7ea394f1 YZ |
1083 | if (!list_empty(&trans->transaction->pending_snapshots)) |
1084 | snap_pending = 1; | |
1085 | ||
2c90e5d6 | 1086 | WARN_ON(cur_trans != trans->transaction); |
15ee9bc7 JB |
1087 | if (cur_trans->num_writers > 1) |
1088 | timeout = MAX_SCHEDULE_TIMEOUT; | |
89573b9c | 1089 | else if (should_grow) |
15ee9bc7 JB |
1090 | timeout = 1; |
1091 | ||
79154b1b | 1092 | mutex_unlock(&root->fs_info->trans_mutex); |
15ee9bc7 | 1093 | |
0bdb1db2 | 1094 | if (flush_on_commit || snap_pending) { |
24bbcf04 YZ |
1095 | btrfs_start_delalloc_inodes(root, 1); |
1096 | ret = btrfs_wait_ordered_extents(root, 0, 1); | |
ebecd3d9 | 1097 | BUG_ON(ret); |
7ea394f1 YZ |
1098 | } |
1099 | ||
5a3f23d5 CM |
1100 | /* |
1101 | * rename don't use btrfs_join_transaction, so, once we | |
1102 | * set the transaction to blocked above, we aren't going | |
1103 | * to get any new ordered operations. We can safely run | |
1104 | * it here and no for sure that nothing new will be added | |
1105 | * to the list | |
1106 | */ | |
1107 | btrfs_run_ordered_operations(root, 1); | |
1108 | ||
ed3b3d31 CM |
1109 | prepare_to_wait(&cur_trans->writer_wait, &wait, |
1110 | TASK_UNINTERRUPTIBLE); | |
1111 | ||
89573b9c CM |
1112 | smp_mb(); |
1113 | if (cur_trans->num_writers > 1 || should_grow) | |
1114 | schedule_timeout(timeout); | |
15ee9bc7 | 1115 | |
79154b1b | 1116 | mutex_lock(&root->fs_info->trans_mutex); |
15ee9bc7 JB |
1117 | finish_wait(&cur_trans->writer_wait, &wait); |
1118 | } while (cur_trans->num_writers > 1 || | |
89573b9c | 1119 | (should_grow && cur_trans->num_joined != joined)); |
15ee9bc7 | 1120 | |
3063d29f CM |
1121 | ret = create_pending_snapshots(trans, root->fs_info); |
1122 | BUG_ON(ret); | |
1123 | ||
56bec294 CM |
1124 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
1125 | BUG_ON(ret); | |
1126 | ||
2c90e5d6 | 1127 | WARN_ON(cur_trans != trans->transaction); |
dc17ff8f | 1128 | |
e02119d5 CM |
1129 | /* btrfs_commit_tree_roots is responsible for getting the |
1130 | * various roots consistent with each other. Every pointer | |
1131 | * in the tree of tree roots has to point to the most up to date | |
1132 | * root for every subvolume and other tree. So, we have to keep | |
1133 | * the tree logging code from jumping in and changing any | |
1134 | * of the trees. | |
1135 | * | |
1136 | * At this point in the commit, there can't be any tree-log | |
1137 | * writers, but a little lower down we drop the trans mutex | |
1138 | * and let new people in. By holding the tree_log_mutex | |
1139 | * from now until after the super is written, we avoid races | |
1140 | * with the tree-log code. | |
1141 | */ | |
1142 | mutex_lock(&root->fs_info->tree_log_mutex); | |
1143 | ||
5d4f98a2 | 1144 | ret = commit_fs_roots(trans, root); |
54aa1f4d CM |
1145 | BUG_ON(ret); |
1146 | ||
5d4f98a2 | 1147 | /* commit_fs_roots gets rid of all the tree log roots, it is now |
e02119d5 CM |
1148 | * safe to free the root of tree log roots |
1149 | */ | |
1150 | btrfs_free_log_root_tree(trans, root->fs_info); | |
1151 | ||
5d4f98a2 | 1152 | ret = commit_cowonly_roots(trans, root); |
79154b1b | 1153 | BUG_ON(ret); |
54aa1f4d | 1154 | |
11833d66 YZ |
1155 | btrfs_prepare_extent_commit(trans, root); |
1156 | ||
78fae27e | 1157 | cur_trans = root->fs_info->running_transaction; |
cee36a03 | 1158 | spin_lock(&root->fs_info->new_trans_lock); |
78fae27e | 1159 | root->fs_info->running_transaction = NULL; |
cee36a03 | 1160 | spin_unlock(&root->fs_info->new_trans_lock); |
5d4f98a2 YZ |
1161 | |
1162 | btrfs_set_root_node(&root->fs_info->tree_root->root_item, | |
1163 | root->fs_info->tree_root->node); | |
817d52f8 | 1164 | switch_commit_root(root->fs_info->tree_root); |
5d4f98a2 YZ |
1165 | |
1166 | btrfs_set_root_node(&root->fs_info->chunk_root->root_item, | |
1167 | root->fs_info->chunk_root->node); | |
817d52f8 | 1168 | switch_commit_root(root->fs_info->chunk_root); |
5d4f98a2 YZ |
1169 | |
1170 | update_super_roots(root); | |
e02119d5 CM |
1171 | |
1172 | if (!root->fs_info->log_root_recovering) { | |
1173 | btrfs_set_super_log_root(&root->fs_info->super_copy, 0); | |
1174 | btrfs_set_super_log_root_level(&root->fs_info->super_copy, 0); | |
1175 | } | |
1176 | ||
a061fc8d CM |
1177 | memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy, |
1178 | sizeof(root->fs_info->super_copy)); | |
ccd467d6 | 1179 | |
f9295749 | 1180 | trans->transaction->blocked = 0; |
b7ec40d7 | 1181 | |
f9295749 | 1182 | wake_up(&root->fs_info->transaction_wait); |
e6dcd2dc | 1183 | |
78fae27e | 1184 | mutex_unlock(&root->fs_info->trans_mutex); |
79154b1b CM |
1185 | ret = btrfs_write_and_wait_transaction(trans, root); |
1186 | BUG_ON(ret); | |
a512bbf8 | 1187 | write_ctree_super(trans, root, 0); |
4313b399 | 1188 | |
e02119d5 CM |
1189 | /* |
1190 | * the super is written, we can safely allow the tree-loggers | |
1191 | * to go about their business | |
1192 | */ | |
1193 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
1194 | ||
11833d66 | 1195 | btrfs_finish_extent_commit(trans, root); |
4313b399 | 1196 | |
1a40e23b ZY |
1197 | mutex_lock(&root->fs_info->trans_mutex); |
1198 | ||
2c90e5d6 | 1199 | cur_trans->commit_done = 1; |
b7ec40d7 | 1200 | |
15ee9bc7 | 1201 | root->fs_info->last_trans_committed = cur_trans->transid; |
817d52f8 | 1202 | |
2c90e5d6 | 1203 | wake_up(&cur_trans->commit_wait); |
3de4586c | 1204 | |
78fae27e | 1205 | put_transaction(cur_trans); |
79154b1b | 1206 | put_transaction(cur_trans); |
58176a96 | 1207 | |
78fae27e | 1208 | mutex_unlock(&root->fs_info->trans_mutex); |
3de4586c | 1209 | |
9ed74f2d JB |
1210 | if (current->journal_info == trans) |
1211 | current->journal_info = NULL; | |
1212 | ||
2c90e5d6 | 1213 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
24bbcf04 YZ |
1214 | |
1215 | if (current != root->fs_info->transaction_kthread) | |
1216 | btrfs_run_delayed_iputs(root); | |
1217 | ||
79154b1b CM |
1218 | return ret; |
1219 | } | |
1220 | ||
d352ac68 CM |
1221 | /* |
1222 | * interface function to delete all the snapshots we have scheduled for deletion | |
1223 | */ | |
e9d0b13b CM |
1224 | int btrfs_clean_old_snapshots(struct btrfs_root *root) |
1225 | { | |
5d4f98a2 YZ |
1226 | LIST_HEAD(list); |
1227 | struct btrfs_fs_info *fs_info = root->fs_info; | |
1228 | ||
1229 | mutex_lock(&fs_info->trans_mutex); | |
1230 | list_splice_init(&fs_info->dead_roots, &list); | |
1231 | mutex_unlock(&fs_info->trans_mutex); | |
e9d0b13b | 1232 | |
5d4f98a2 YZ |
1233 | while (!list_empty(&list)) { |
1234 | root = list_entry(list.next, struct btrfs_root, root_list); | |
76dda93c YZ |
1235 | list_del(&root->root_list); |
1236 | ||
1237 | if (btrfs_header_backref_rev(root->node) < | |
1238 | BTRFS_MIXED_BACKREF_REV) | |
3fd0a558 | 1239 | btrfs_drop_snapshot(root, NULL, 0); |
76dda93c | 1240 | else |
3fd0a558 | 1241 | btrfs_drop_snapshot(root, NULL, 1); |
e9d0b13b CM |
1242 | } |
1243 | return 0; | |
1244 | } |