Commit | Line | Data |
---|---|---|
6cbd5570 CM |
1 | /* |
2 | * Copyright (C) 2007 Oracle. All rights reserved. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public | |
6 | * License v2 as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, | |
9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public | |
14 | * License along with this program; if not, write to the | |
15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
16 | * Boston, MA 021110-1307, USA. | |
17 | */ | |
18 | ||
79154b1b | 19 | #include <linux/fs.h> |
5a0e3ad6 | 20 | #include <linux/slab.h> |
34088780 | 21 | #include <linux/sched.h> |
d3c2fdcf | 22 | #include <linux/writeback.h> |
5f39d397 | 23 | #include <linux/pagemap.h> |
5f2cc086 | 24 | #include <linux/blkdev.h> |
79154b1b CM |
25 | #include "ctree.h" |
26 | #include "disk-io.h" | |
27 | #include "transaction.h" | |
925baedd | 28 | #include "locking.h" |
e02119d5 | 29 | #include "tree-log.h" |
581bb050 | 30 | #include "inode-map.h" |
79154b1b | 31 | |
0f7d52f4 CM |
32 | #define BTRFS_ROOT_TRANS_TAG 0 |
33 | ||
80b6794d | 34 | static noinline void put_transaction(struct btrfs_transaction *transaction) |
79154b1b | 35 | { |
13c5a93e JB |
36 | WARN_ON(atomic_read(&transaction->use_count) == 0); |
37 | if (atomic_dec_and_test(&transaction->use_count)) { | |
a4abeea4 | 38 | BUG_ON(!list_empty(&transaction->list)); |
00f04b88 AJ |
39 | WARN_ON(transaction->delayed_refs.root.rb_node); |
40 | WARN_ON(!list_empty(&transaction->delayed_refs.seq_head)); | |
2c90e5d6 CM |
41 | memset(transaction, 0, sizeof(*transaction)); |
42 | kmem_cache_free(btrfs_transaction_cachep, transaction); | |
78fae27e | 43 | } |
79154b1b CM |
44 | } |
45 | ||
817d52f8 JB |
46 | static noinline void switch_commit_root(struct btrfs_root *root) |
47 | { | |
817d52f8 JB |
48 | free_extent_buffer(root->commit_root); |
49 | root->commit_root = btrfs_root_node(root); | |
817d52f8 JB |
50 | } |
51 | ||
d352ac68 CM |
52 | /* |
53 | * either allocate a new transaction or hop into the existing one | |
54 | */ | |
a4abeea4 | 55 | static noinline int join_transaction(struct btrfs_root *root, int nofail) |
79154b1b CM |
56 | { |
57 | struct btrfs_transaction *cur_trans; | |
a4abeea4 JB |
58 | |
59 | spin_lock(&root->fs_info->trans_lock); | |
d43317dc | 60 | loop: |
a4abeea4 JB |
61 | if (root->fs_info->trans_no_join) { |
62 | if (!nofail) { | |
63 | spin_unlock(&root->fs_info->trans_lock); | |
64 | return -EBUSY; | |
65 | } | |
66 | } | |
67 | ||
79154b1b | 68 | cur_trans = root->fs_info->running_transaction; |
a4abeea4 JB |
69 | if (cur_trans) { |
70 | atomic_inc(&cur_trans->use_count); | |
13c5a93e | 71 | atomic_inc(&cur_trans->num_writers); |
15ee9bc7 | 72 | cur_trans->num_joined++; |
a4abeea4 JB |
73 | spin_unlock(&root->fs_info->trans_lock); |
74 | return 0; | |
79154b1b | 75 | } |
a4abeea4 JB |
76 | spin_unlock(&root->fs_info->trans_lock); |
77 | ||
78 | cur_trans = kmem_cache_alloc(btrfs_transaction_cachep, GFP_NOFS); | |
79 | if (!cur_trans) | |
80 | return -ENOMEM; | |
d43317dc | 81 | |
a4abeea4 JB |
82 | spin_lock(&root->fs_info->trans_lock); |
83 | if (root->fs_info->running_transaction) { | |
d43317dc CM |
84 | /* |
85 | * someone started a transaction after we unlocked. Make sure | |
86 | * to redo the trans_no_join checks above | |
87 | */ | |
a4abeea4 JB |
88 | kmem_cache_free(btrfs_transaction_cachep, cur_trans); |
89 | cur_trans = root->fs_info->running_transaction; | |
d43317dc | 90 | goto loop; |
79154b1b | 91 | } |
d43317dc | 92 | |
a4abeea4 JB |
93 | atomic_set(&cur_trans->num_writers, 1); |
94 | cur_trans->num_joined = 0; | |
95 | init_waitqueue_head(&cur_trans->writer_wait); | |
96 | init_waitqueue_head(&cur_trans->commit_wait); | |
97 | cur_trans->in_commit = 0; | |
98 | cur_trans->blocked = 0; | |
99 | /* | |
100 | * One for this trans handle, one so it will live on until we | |
101 | * commit the transaction. | |
102 | */ | |
103 | atomic_set(&cur_trans->use_count, 2); | |
104 | cur_trans->commit_done = 0; | |
105 | cur_trans->start_time = get_seconds(); | |
106 | ||
107 | cur_trans->delayed_refs.root = RB_ROOT; | |
108 | cur_trans->delayed_refs.num_entries = 0; | |
109 | cur_trans->delayed_refs.num_heads_ready = 0; | |
110 | cur_trans->delayed_refs.num_heads = 0; | |
111 | cur_trans->delayed_refs.flushing = 0; | |
112 | cur_trans->delayed_refs.run_delayed_start = 0; | |
00f04b88 | 113 | cur_trans->delayed_refs.seq = 1; |
a168650c | 114 | init_waitqueue_head(&cur_trans->delayed_refs.seq_wait); |
a4abeea4 JB |
115 | spin_lock_init(&cur_trans->commit_lock); |
116 | spin_lock_init(&cur_trans->delayed_refs.lock); | |
00f04b88 | 117 | INIT_LIST_HEAD(&cur_trans->delayed_refs.seq_head); |
a4abeea4 JB |
118 | |
119 | INIT_LIST_HEAD(&cur_trans->pending_snapshots); | |
120 | list_add_tail(&cur_trans->list, &root->fs_info->trans_list); | |
121 | extent_io_tree_init(&cur_trans->dirty_pages, | |
ff5714cc | 122 | root->fs_info->btree_inode->i_mapping); |
a4abeea4 JB |
123 | root->fs_info->generation++; |
124 | cur_trans->transid = root->fs_info->generation; | |
125 | root->fs_info->running_transaction = cur_trans; | |
126 | spin_unlock(&root->fs_info->trans_lock); | |
15ee9bc7 | 127 | |
79154b1b CM |
128 | return 0; |
129 | } | |
130 | ||
d352ac68 | 131 | /* |
d397712b CM |
132 | * this does all the record keeping required to make sure that a reference |
133 | * counted root is properly recorded in a given transaction. This is required | |
134 | * to make sure the old root from before we joined the transaction is deleted | |
135 | * when the transaction commits | |
d352ac68 | 136 | */ |
7585717f | 137 | static int record_root_in_trans(struct btrfs_trans_handle *trans, |
a4abeea4 | 138 | struct btrfs_root *root) |
6702ed49 | 139 | { |
5d4f98a2 | 140 | if (root->ref_cows && root->last_trans < trans->transid) { |
6702ed49 | 141 | WARN_ON(root == root->fs_info->extent_root); |
5d4f98a2 YZ |
142 | WARN_ON(root->commit_root != root->node); |
143 | ||
7585717f CM |
144 | /* |
145 | * see below for in_trans_setup usage rules | |
146 | * we have the reloc mutex held now, so there | |
147 | * is only one writer in this function | |
148 | */ | |
149 | root->in_trans_setup = 1; | |
150 | ||
151 | /* make sure readers find in_trans_setup before | |
152 | * they find our root->last_trans update | |
153 | */ | |
154 | smp_wmb(); | |
155 | ||
a4abeea4 JB |
156 | spin_lock(&root->fs_info->fs_roots_radix_lock); |
157 | if (root->last_trans == trans->transid) { | |
158 | spin_unlock(&root->fs_info->fs_roots_radix_lock); | |
159 | return 0; | |
160 | } | |
5d4f98a2 YZ |
161 | radix_tree_tag_set(&root->fs_info->fs_roots_radix, |
162 | (unsigned long)root->root_key.objectid, | |
163 | BTRFS_ROOT_TRANS_TAG); | |
a4abeea4 | 164 | spin_unlock(&root->fs_info->fs_roots_radix_lock); |
7585717f CM |
165 | root->last_trans = trans->transid; |
166 | ||
167 | /* this is pretty tricky. We don't want to | |
168 | * take the relocation lock in btrfs_record_root_in_trans | |
169 | * unless we're really doing the first setup for this root in | |
170 | * this transaction. | |
171 | * | |
172 | * Normally we'd use root->last_trans as a flag to decide | |
173 | * if we want to take the expensive mutex. | |
174 | * | |
175 | * But, we have to set root->last_trans before we | |
176 | * init the relocation root, otherwise, we trip over warnings | |
177 | * in ctree.c. The solution used here is to flag ourselves | |
178 | * with root->in_trans_setup. When this is 1, we're still | |
179 | * fixing up the reloc trees and everyone must wait. | |
180 | * | |
181 | * When this is zero, they can trust root->last_trans and fly | |
182 | * through btrfs_record_root_in_trans without having to take the | |
183 | * lock. smp_wmb() makes sure that all the writes above are | |
184 | * done before we pop in the zero below | |
185 | */ | |
5d4f98a2 | 186 | btrfs_init_reloc_root(trans, root); |
7585717f CM |
187 | smp_wmb(); |
188 | root->in_trans_setup = 0; | |
5d4f98a2 YZ |
189 | } |
190 | return 0; | |
191 | } | |
bcc63abb | 192 | |
7585717f CM |
193 | |
194 | int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, | |
195 | struct btrfs_root *root) | |
196 | { | |
197 | if (!root->ref_cows) | |
198 | return 0; | |
199 | ||
200 | /* | |
201 | * see record_root_in_trans for comments about in_trans_setup usage | |
202 | * and barriers | |
203 | */ | |
204 | smp_rmb(); | |
205 | if (root->last_trans == trans->transid && | |
206 | !root->in_trans_setup) | |
207 | return 0; | |
208 | ||
209 | mutex_lock(&root->fs_info->reloc_mutex); | |
210 | record_root_in_trans(trans, root); | |
211 | mutex_unlock(&root->fs_info->reloc_mutex); | |
212 | ||
213 | return 0; | |
214 | } | |
215 | ||
d352ac68 CM |
216 | /* wait for commit against the current transaction to become unblocked |
217 | * when this is done, it is safe to start a new transaction, but the current | |
218 | * transaction might not be fully on disk. | |
219 | */ | |
37d1aeee | 220 | static void wait_current_trans(struct btrfs_root *root) |
79154b1b | 221 | { |
f9295749 | 222 | struct btrfs_transaction *cur_trans; |
79154b1b | 223 | |
a4abeea4 | 224 | spin_lock(&root->fs_info->trans_lock); |
f9295749 | 225 | cur_trans = root->fs_info->running_transaction; |
37d1aeee | 226 | if (cur_trans && cur_trans->blocked) { |
13c5a93e | 227 | atomic_inc(&cur_trans->use_count); |
a4abeea4 | 228 | spin_unlock(&root->fs_info->trans_lock); |
72d63ed6 LZ |
229 | |
230 | wait_event(root->fs_info->transaction_wait, | |
231 | !cur_trans->blocked); | |
f9295749 | 232 | put_transaction(cur_trans); |
a4abeea4 JB |
233 | } else { |
234 | spin_unlock(&root->fs_info->trans_lock); | |
f9295749 | 235 | } |
37d1aeee CM |
236 | } |
237 | ||
249ac1e5 JB |
238 | enum btrfs_trans_type { |
239 | TRANS_START, | |
240 | TRANS_JOIN, | |
241 | TRANS_USERSPACE, | |
0af3d00b | 242 | TRANS_JOIN_NOLOCK, |
249ac1e5 JB |
243 | }; |
244 | ||
a22285a6 YZ |
245 | static int may_wait_transaction(struct btrfs_root *root, int type) |
246 | { | |
a4abeea4 JB |
247 | if (root->fs_info->log_root_recovering) |
248 | return 0; | |
249 | ||
250 | if (type == TRANS_USERSPACE) | |
251 | return 1; | |
252 | ||
253 | if (type == TRANS_START && | |
254 | !atomic_read(&root->fs_info->open_ioctl_trans)) | |
a22285a6 | 255 | return 1; |
a4abeea4 | 256 | |
a22285a6 YZ |
257 | return 0; |
258 | } | |
259 | ||
e02119d5 | 260 | static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root, |
a22285a6 | 261 | u64 num_items, int type) |
37d1aeee | 262 | { |
a22285a6 YZ |
263 | struct btrfs_trans_handle *h; |
264 | struct btrfs_transaction *cur_trans; | |
b5009945 | 265 | u64 num_bytes = 0; |
37d1aeee | 266 | int ret; |
acce952b | 267 | |
268 | if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) | |
269 | return ERR_PTR(-EROFS); | |
2a1eb461 JB |
270 | |
271 | if (current->journal_info) { | |
272 | WARN_ON(type != TRANS_JOIN && type != TRANS_JOIN_NOLOCK); | |
273 | h = current->journal_info; | |
274 | h->use_count++; | |
275 | h->orig_rsv = h->block_rsv; | |
276 | h->block_rsv = NULL; | |
277 | goto got_it; | |
278 | } | |
b5009945 JB |
279 | |
280 | /* | |
281 | * Do the reservation before we join the transaction so we can do all | |
282 | * the appropriate flushing if need be. | |
283 | */ | |
284 | if (num_items > 0 && root != root->fs_info->chunk_root) { | |
285 | num_bytes = btrfs_calc_trans_metadata_size(root, num_items); | |
4a92b1b8 | 286 | ret = btrfs_block_rsv_add(root, |
b5009945 JB |
287 | &root->fs_info->trans_block_rsv, |
288 | num_bytes); | |
289 | if (ret) | |
290 | return ERR_PTR(ret); | |
291 | } | |
a22285a6 YZ |
292 | again: |
293 | h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS); | |
294 | if (!h) | |
295 | return ERR_PTR(-ENOMEM); | |
37d1aeee | 296 | |
a22285a6 | 297 | if (may_wait_transaction(root, type)) |
37d1aeee | 298 | wait_current_trans(root); |
a22285a6 | 299 | |
a4abeea4 JB |
300 | do { |
301 | ret = join_transaction(root, type == TRANS_JOIN_NOLOCK); | |
302 | if (ret == -EBUSY) | |
303 | wait_current_trans(root); | |
304 | } while (ret == -EBUSY); | |
305 | ||
db5b493a | 306 | if (ret < 0) { |
6e8df2ae | 307 | kmem_cache_free(btrfs_trans_handle_cachep, h); |
db5b493a TI |
308 | return ERR_PTR(ret); |
309 | } | |
0f7d52f4 | 310 | |
a22285a6 | 311 | cur_trans = root->fs_info->running_transaction; |
a22285a6 YZ |
312 | |
313 | h->transid = cur_trans->transid; | |
314 | h->transaction = cur_trans; | |
79154b1b | 315 | h->blocks_used = 0; |
a22285a6 | 316 | h->bytes_reserved = 0; |
56bec294 | 317 | h->delayed_ref_updates = 0; |
2a1eb461 | 318 | h->use_count = 1; |
f0486c68 | 319 | h->block_rsv = NULL; |
2a1eb461 | 320 | h->orig_rsv = NULL; |
b7ec40d7 | 321 | |
a22285a6 YZ |
322 | smp_mb(); |
323 | if (cur_trans->blocked && may_wait_transaction(root, type)) { | |
324 | btrfs_commit_transaction(h, root); | |
325 | goto again; | |
326 | } | |
327 | ||
b5009945 JB |
328 | if (num_bytes) { |
329 | h->block_rsv = &root->fs_info->trans_block_rsv; | |
330 | h->bytes_reserved = num_bytes; | |
a22285a6 | 331 | } |
9ed74f2d | 332 | |
2a1eb461 | 333 | got_it: |
a4abeea4 | 334 | btrfs_record_root_in_trans(h, root); |
a22285a6 YZ |
335 | |
336 | if (!current->journal_info && type != TRANS_USERSPACE) | |
337 | current->journal_info = h; | |
79154b1b CM |
338 | return h; |
339 | } | |
340 | ||
f9295749 | 341 | struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, |
a22285a6 | 342 | int num_items) |
f9295749 | 343 | { |
a22285a6 | 344 | return start_transaction(root, num_items, TRANS_START); |
f9295749 | 345 | } |
7a7eaa40 | 346 | struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root) |
f9295749 | 347 | { |
a22285a6 | 348 | return start_transaction(root, 0, TRANS_JOIN); |
f9295749 CM |
349 | } |
350 | ||
7a7eaa40 | 351 | struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root) |
0af3d00b JB |
352 | { |
353 | return start_transaction(root, 0, TRANS_JOIN_NOLOCK); | |
354 | } | |
355 | ||
7a7eaa40 | 356 | struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root) |
9ca9ee09 | 357 | { |
7a7eaa40 | 358 | return start_transaction(root, 0, TRANS_USERSPACE); |
9ca9ee09 SW |
359 | } |
360 | ||
d352ac68 | 361 | /* wait for a transaction commit to be fully complete */ |
b9c8300c | 362 | static noinline void wait_for_commit(struct btrfs_root *root, |
89ce8a63 CM |
363 | struct btrfs_transaction *commit) |
364 | { | |
72d63ed6 | 365 | wait_event(commit->commit_wait, commit->commit_done); |
89ce8a63 CM |
366 | } |
367 | ||
46204592 SW |
368 | int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid) |
369 | { | |
370 | struct btrfs_transaction *cur_trans = NULL, *t; | |
371 | int ret; | |
372 | ||
46204592 SW |
373 | ret = 0; |
374 | if (transid) { | |
375 | if (transid <= root->fs_info->last_trans_committed) | |
a4abeea4 | 376 | goto out; |
46204592 SW |
377 | |
378 | /* find specified transaction */ | |
a4abeea4 | 379 | spin_lock(&root->fs_info->trans_lock); |
46204592 SW |
380 | list_for_each_entry(t, &root->fs_info->trans_list, list) { |
381 | if (t->transid == transid) { | |
382 | cur_trans = t; | |
a4abeea4 | 383 | atomic_inc(&cur_trans->use_count); |
46204592 SW |
384 | break; |
385 | } | |
386 | if (t->transid > transid) | |
387 | break; | |
388 | } | |
a4abeea4 | 389 | spin_unlock(&root->fs_info->trans_lock); |
46204592 SW |
390 | ret = -EINVAL; |
391 | if (!cur_trans) | |
a4abeea4 | 392 | goto out; /* bad transid */ |
46204592 SW |
393 | } else { |
394 | /* find newest transaction that is committing | committed */ | |
a4abeea4 | 395 | spin_lock(&root->fs_info->trans_lock); |
46204592 SW |
396 | list_for_each_entry_reverse(t, &root->fs_info->trans_list, |
397 | list) { | |
398 | if (t->in_commit) { | |
399 | if (t->commit_done) | |
3473f3c0 | 400 | break; |
46204592 | 401 | cur_trans = t; |
a4abeea4 | 402 | atomic_inc(&cur_trans->use_count); |
46204592 SW |
403 | break; |
404 | } | |
405 | } | |
a4abeea4 | 406 | spin_unlock(&root->fs_info->trans_lock); |
46204592 | 407 | if (!cur_trans) |
a4abeea4 | 408 | goto out; /* nothing committing|committed */ |
46204592 SW |
409 | } |
410 | ||
46204592 SW |
411 | wait_for_commit(root, cur_trans); |
412 | ||
46204592 SW |
413 | put_transaction(cur_trans); |
414 | ret = 0; | |
a4abeea4 | 415 | out: |
46204592 SW |
416 | return ret; |
417 | } | |
418 | ||
37d1aeee CM |
419 | void btrfs_throttle(struct btrfs_root *root) |
420 | { | |
a4abeea4 | 421 | if (!atomic_read(&root->fs_info->open_ioctl_trans)) |
9ca9ee09 | 422 | wait_current_trans(root); |
37d1aeee CM |
423 | } |
424 | ||
8929ecfa YZ |
425 | static int should_end_transaction(struct btrfs_trans_handle *trans, |
426 | struct btrfs_root *root) | |
427 | { | |
428 | int ret; | |
36ba022a JB |
429 | |
430 | ret = btrfs_block_rsv_check(root, &root->fs_info->global_block_rsv, 5); | |
8929ecfa YZ |
431 | return ret ? 1 : 0; |
432 | } | |
433 | ||
434 | int btrfs_should_end_transaction(struct btrfs_trans_handle *trans, | |
435 | struct btrfs_root *root) | |
436 | { | |
437 | struct btrfs_transaction *cur_trans = trans->transaction; | |
9c8d86db | 438 | struct btrfs_block_rsv *rsv = trans->block_rsv; |
8929ecfa YZ |
439 | int updates; |
440 | ||
a4abeea4 | 441 | smp_mb(); |
8929ecfa YZ |
442 | if (cur_trans->blocked || cur_trans->delayed_refs.flushing) |
443 | return 1; | |
444 | ||
9c8d86db JB |
445 | /* |
446 | * We need to do this in case we're deleting csums so the global block | |
447 | * rsv get's used instead of the csum block rsv. | |
448 | */ | |
449 | trans->block_rsv = NULL; | |
450 | ||
8929ecfa YZ |
451 | updates = trans->delayed_ref_updates; |
452 | trans->delayed_ref_updates = 0; | |
453 | if (updates) | |
454 | btrfs_run_delayed_refs(trans, root, updates); | |
455 | ||
9c8d86db JB |
456 | trans->block_rsv = rsv; |
457 | ||
8929ecfa YZ |
458 | return should_end_transaction(trans, root); |
459 | } | |
460 | ||
89ce8a63 | 461 | static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, |
0af3d00b | 462 | struct btrfs_root *root, int throttle, int lock) |
79154b1b | 463 | { |
8929ecfa | 464 | struct btrfs_transaction *cur_trans = trans->transaction; |
ab78c84d | 465 | struct btrfs_fs_info *info = root->fs_info; |
c3e69d58 CM |
466 | int count = 0; |
467 | ||
2a1eb461 JB |
468 | if (--trans->use_count) { |
469 | trans->block_rsv = trans->orig_rsv; | |
470 | return 0; | |
471 | } | |
472 | ||
b24e03db | 473 | btrfs_trans_release_metadata(trans, root); |
4c13d758 | 474 | trans->block_rsv = NULL; |
203bf287 | 475 | while (count < 2) { |
c3e69d58 CM |
476 | unsigned long cur = trans->delayed_ref_updates; |
477 | trans->delayed_ref_updates = 0; | |
478 | if (cur && | |
479 | trans->transaction->delayed_refs.num_heads_ready > 64) { | |
480 | trans->delayed_ref_updates = 0; | |
481 | btrfs_run_delayed_refs(trans, root, cur); | |
482 | } else { | |
483 | break; | |
484 | } | |
485 | count++; | |
56bec294 CM |
486 | } |
487 | ||
a4abeea4 JB |
488 | if (lock && !atomic_read(&root->fs_info->open_ioctl_trans) && |
489 | should_end_transaction(trans, root)) { | |
8929ecfa | 490 | trans->transaction->blocked = 1; |
a4abeea4 JB |
491 | smp_wmb(); |
492 | } | |
8929ecfa | 493 | |
0af3d00b | 494 | if (lock && cur_trans->blocked && !cur_trans->in_commit) { |
81317fde JB |
495 | if (throttle) { |
496 | /* | |
497 | * We may race with somebody else here so end up having | |
498 | * to call end_transaction on ourselves again, so inc | |
499 | * our use_count. | |
500 | */ | |
501 | trans->use_count++; | |
8929ecfa | 502 | return btrfs_commit_transaction(trans, root); |
81317fde | 503 | } else { |
8929ecfa | 504 | wake_up_process(info->transaction_kthread); |
81317fde | 505 | } |
8929ecfa YZ |
506 | } |
507 | ||
8929ecfa | 508 | WARN_ON(cur_trans != info->running_transaction); |
13c5a93e JB |
509 | WARN_ON(atomic_read(&cur_trans->num_writers) < 1); |
510 | atomic_dec(&cur_trans->num_writers); | |
89ce8a63 | 511 | |
99d16cbc | 512 | smp_mb(); |
79154b1b CM |
513 | if (waitqueue_active(&cur_trans->writer_wait)) |
514 | wake_up(&cur_trans->writer_wait); | |
79154b1b | 515 | put_transaction(cur_trans); |
9ed74f2d JB |
516 | |
517 | if (current->journal_info == trans) | |
518 | current->journal_info = NULL; | |
d6025579 | 519 | memset(trans, 0, sizeof(*trans)); |
2c90e5d6 | 520 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
ab78c84d | 521 | |
24bbcf04 YZ |
522 | if (throttle) |
523 | btrfs_run_delayed_iputs(root); | |
524 | ||
79154b1b CM |
525 | return 0; |
526 | } | |
527 | ||
89ce8a63 CM |
528 | int btrfs_end_transaction(struct btrfs_trans_handle *trans, |
529 | struct btrfs_root *root) | |
530 | { | |
16cdcec7 MX |
531 | int ret; |
532 | ||
533 | ret = __btrfs_end_transaction(trans, root, 0, 1); | |
534 | if (ret) | |
535 | return ret; | |
536 | return 0; | |
89ce8a63 CM |
537 | } |
538 | ||
539 | int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans, | |
540 | struct btrfs_root *root) | |
541 | { | |
16cdcec7 MX |
542 | int ret; |
543 | ||
544 | ret = __btrfs_end_transaction(trans, root, 1, 1); | |
545 | if (ret) | |
546 | return ret; | |
547 | return 0; | |
0af3d00b JB |
548 | } |
549 | ||
550 | int btrfs_end_transaction_nolock(struct btrfs_trans_handle *trans, | |
551 | struct btrfs_root *root) | |
552 | { | |
16cdcec7 MX |
553 | int ret; |
554 | ||
555 | ret = __btrfs_end_transaction(trans, root, 0, 0); | |
556 | if (ret) | |
557 | return ret; | |
558 | return 0; | |
559 | } | |
560 | ||
561 | int btrfs_end_transaction_dmeta(struct btrfs_trans_handle *trans, | |
562 | struct btrfs_root *root) | |
563 | { | |
564 | return __btrfs_end_transaction(trans, root, 1, 1); | |
89ce8a63 CM |
565 | } |
566 | ||
d352ac68 CM |
567 | /* |
568 | * when btree blocks are allocated, they have some corresponding bits set for | |
569 | * them in one of two extent_io trees. This is used to make sure all of | |
690587d1 | 570 | * those extents are sent to disk but does not wait on them |
d352ac68 | 571 | */ |
690587d1 | 572 | int btrfs_write_marked_extents(struct btrfs_root *root, |
8cef4e16 | 573 | struct extent_io_tree *dirty_pages, int mark) |
79154b1b | 574 | { |
777e6bd7 | 575 | int err = 0; |
7c4452b9 | 576 | int werr = 0; |
1728366e | 577 | struct address_space *mapping = root->fs_info->btree_inode->i_mapping; |
777e6bd7 | 578 | u64 start = 0; |
5f39d397 | 579 | u64 end; |
7c4452b9 | 580 | |
1728366e JB |
581 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
582 | mark)) { | |
583 | convert_extent_bit(dirty_pages, start, end, EXTENT_NEED_WAIT, mark, | |
584 | GFP_NOFS); | |
585 | err = filemap_fdatawrite_range(mapping, start, end); | |
586 | if (err) | |
587 | werr = err; | |
588 | cond_resched(); | |
589 | start = end + 1; | |
7c4452b9 | 590 | } |
690587d1 CM |
591 | if (err) |
592 | werr = err; | |
593 | return werr; | |
594 | } | |
595 | ||
596 | /* | |
597 | * when btree blocks are allocated, they have some corresponding bits set for | |
598 | * them in one of two extent_io trees. This is used to make sure all of | |
599 | * those extents are on disk for transaction or log commit. We wait | |
600 | * on all the pages and clear them from the dirty pages state tree | |
601 | */ | |
602 | int btrfs_wait_marked_extents(struct btrfs_root *root, | |
8cef4e16 | 603 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 | 604 | { |
690587d1 CM |
605 | int err = 0; |
606 | int werr = 0; | |
1728366e | 607 | struct address_space *mapping = root->fs_info->btree_inode->i_mapping; |
690587d1 CM |
608 | u64 start = 0; |
609 | u64 end; | |
777e6bd7 | 610 | |
1728366e JB |
611 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
612 | EXTENT_NEED_WAIT)) { | |
613 | clear_extent_bits(dirty_pages, start, end, EXTENT_NEED_WAIT, GFP_NOFS); | |
614 | err = filemap_fdatawait_range(mapping, start, end); | |
615 | if (err) | |
616 | werr = err; | |
617 | cond_resched(); | |
618 | start = end + 1; | |
777e6bd7 | 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); | |
bf0da8c1 CM |
638 | |
639 | if (ret) | |
640 | return ret; | |
641 | if (ret2) | |
642 | return ret2; | |
643 | return 0; | |
690587d1 CM |
644 | } |
645 | ||
d0c803c4 CM |
646 | int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans, |
647 | struct btrfs_root *root) | |
648 | { | |
649 | if (!trans || !trans->transaction) { | |
650 | struct inode *btree_inode; | |
651 | btree_inode = root->fs_info->btree_inode; | |
652 | return filemap_write_and_wait(btree_inode->i_mapping); | |
653 | } | |
654 | return btrfs_write_and_wait_marked_extents(root, | |
8cef4e16 YZ |
655 | &trans->transaction->dirty_pages, |
656 | EXTENT_DIRTY); | |
d0c803c4 CM |
657 | } |
658 | ||
d352ac68 CM |
659 | /* |
660 | * this is used to update the root pointer in the tree of tree roots. | |
661 | * | |
662 | * But, in the case of the extent allocation tree, updating the root | |
663 | * pointer may allocate blocks which may change the root of the extent | |
664 | * allocation tree. | |
665 | * | |
666 | * So, this loops and repeats and makes sure the cowonly root didn't | |
667 | * change while the root pointer was being updated in the metadata. | |
668 | */ | |
0b86a832 CM |
669 | static int update_cowonly_root(struct btrfs_trans_handle *trans, |
670 | struct btrfs_root *root) | |
79154b1b CM |
671 | { |
672 | int ret; | |
0b86a832 | 673 | u64 old_root_bytenr; |
86b9f2ec | 674 | u64 old_root_used; |
0b86a832 | 675 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
79154b1b | 676 | |
86b9f2ec | 677 | old_root_used = btrfs_root_used(&root->root_item); |
0b86a832 | 678 | btrfs_write_dirty_block_groups(trans, root); |
56bec294 | 679 | |
d397712b | 680 | while (1) { |
0b86a832 | 681 | old_root_bytenr = btrfs_root_bytenr(&root->root_item); |
86b9f2ec YZ |
682 | if (old_root_bytenr == root->node->start && |
683 | old_root_used == btrfs_root_used(&root->root_item)) | |
79154b1b | 684 | break; |
87ef2bb4 | 685 | |
5d4f98a2 | 686 | btrfs_set_root_node(&root->root_item, root->node); |
79154b1b | 687 | ret = btrfs_update_root(trans, tree_root, |
0b86a832 CM |
688 | &root->root_key, |
689 | &root->root_item); | |
79154b1b | 690 | BUG_ON(ret); |
56bec294 | 691 | |
86b9f2ec | 692 | old_root_used = btrfs_root_used(&root->root_item); |
4a8c9a62 | 693 | ret = btrfs_write_dirty_block_groups(trans, root); |
56bec294 | 694 | BUG_ON(ret); |
0b86a832 | 695 | } |
276e680d YZ |
696 | |
697 | if (root != root->fs_info->extent_root) | |
698 | switch_commit_root(root); | |
699 | ||
0b86a832 CM |
700 | return 0; |
701 | } | |
702 | ||
d352ac68 CM |
703 | /* |
704 | * update all the cowonly tree roots on disk | |
705 | */ | |
5d4f98a2 YZ |
706 | static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans, |
707 | struct btrfs_root *root) | |
0b86a832 CM |
708 | { |
709 | struct btrfs_fs_info *fs_info = root->fs_info; | |
710 | struct list_head *next; | |
84234f3a | 711 | struct extent_buffer *eb; |
56bec294 | 712 | int ret; |
84234f3a | 713 | |
56bec294 CM |
714 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
715 | BUG_ON(ret); | |
87ef2bb4 | 716 | |
84234f3a | 717 | eb = btrfs_lock_root_node(fs_info->tree_root); |
9fa8cfe7 | 718 | btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 0, &eb); |
84234f3a YZ |
719 | btrfs_tree_unlock(eb); |
720 | free_extent_buffer(eb); | |
0b86a832 | 721 | |
56bec294 CM |
722 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
723 | BUG_ON(ret); | |
87ef2bb4 | 724 | |
d397712b | 725 | while (!list_empty(&fs_info->dirty_cowonly_roots)) { |
0b86a832 CM |
726 | next = fs_info->dirty_cowonly_roots.next; |
727 | list_del_init(next); | |
728 | root = list_entry(next, struct btrfs_root, dirty_list); | |
87ef2bb4 | 729 | |
0b86a832 | 730 | update_cowonly_root(trans, root); |
79154b1b | 731 | } |
276e680d YZ |
732 | |
733 | down_write(&fs_info->extent_commit_sem); | |
734 | switch_commit_root(fs_info->extent_root); | |
735 | up_write(&fs_info->extent_commit_sem); | |
736 | ||
79154b1b CM |
737 | return 0; |
738 | } | |
739 | ||
d352ac68 CM |
740 | /* |
741 | * dead roots are old snapshots that need to be deleted. This allocates | |
742 | * a dirty root struct and adds it into the list of dead roots that need to | |
743 | * be deleted | |
744 | */ | |
5d4f98a2 | 745 | int btrfs_add_dead_root(struct btrfs_root *root) |
5eda7b5e | 746 | { |
a4abeea4 | 747 | spin_lock(&root->fs_info->trans_lock); |
5d4f98a2 | 748 | list_add(&root->root_list, &root->fs_info->dead_roots); |
a4abeea4 | 749 | spin_unlock(&root->fs_info->trans_lock); |
5eda7b5e CM |
750 | return 0; |
751 | } | |
752 | ||
d352ac68 | 753 | /* |
5d4f98a2 | 754 | * update all the cowonly tree roots on disk |
d352ac68 | 755 | */ |
5d4f98a2 YZ |
756 | static noinline int commit_fs_roots(struct btrfs_trans_handle *trans, |
757 | struct btrfs_root *root) | |
0f7d52f4 | 758 | { |
0f7d52f4 | 759 | struct btrfs_root *gang[8]; |
5d4f98a2 | 760 | struct btrfs_fs_info *fs_info = root->fs_info; |
0f7d52f4 CM |
761 | int i; |
762 | int ret; | |
54aa1f4d CM |
763 | int err = 0; |
764 | ||
a4abeea4 | 765 | spin_lock(&fs_info->fs_roots_radix_lock); |
d397712b | 766 | while (1) { |
5d4f98a2 YZ |
767 | ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, |
768 | (void **)gang, 0, | |
0f7d52f4 CM |
769 | ARRAY_SIZE(gang), |
770 | BTRFS_ROOT_TRANS_TAG); | |
771 | if (ret == 0) | |
772 | break; | |
773 | for (i = 0; i < ret; i++) { | |
774 | root = gang[i]; | |
5d4f98a2 YZ |
775 | radix_tree_tag_clear(&fs_info->fs_roots_radix, |
776 | (unsigned long)root->root_key.objectid, | |
777 | BTRFS_ROOT_TRANS_TAG); | |
a4abeea4 | 778 | spin_unlock(&fs_info->fs_roots_radix_lock); |
31153d81 | 779 | |
e02119d5 | 780 | btrfs_free_log(trans, root); |
5d4f98a2 | 781 | btrfs_update_reloc_root(trans, root); |
d68fc57b | 782 | btrfs_orphan_commit_root(trans, root); |
bcc63abb | 783 | |
82d5902d LZ |
784 | btrfs_save_ino_cache(root, trans); |
785 | ||
f1ebcc74 LB |
786 | /* see comments in should_cow_block() */ |
787 | root->force_cow = 0; | |
788 | smp_wmb(); | |
789 | ||
978d910d | 790 | if (root->commit_root != root->node) { |
581bb050 | 791 | mutex_lock(&root->fs_commit_mutex); |
817d52f8 | 792 | switch_commit_root(root); |
581bb050 LZ |
793 | btrfs_unpin_free_ino(root); |
794 | mutex_unlock(&root->fs_commit_mutex); | |
795 | ||
978d910d YZ |
796 | btrfs_set_root_node(&root->root_item, |
797 | root->node); | |
798 | } | |
5d4f98a2 | 799 | |
5d4f98a2 | 800 | err = btrfs_update_root(trans, fs_info->tree_root, |
0f7d52f4 CM |
801 | &root->root_key, |
802 | &root->root_item); | |
a4abeea4 | 803 | spin_lock(&fs_info->fs_roots_radix_lock); |
54aa1f4d CM |
804 | if (err) |
805 | break; | |
0f7d52f4 CM |
806 | } |
807 | } | |
a4abeea4 | 808 | spin_unlock(&fs_info->fs_roots_radix_lock); |
54aa1f4d | 809 | return err; |
0f7d52f4 CM |
810 | } |
811 | ||
d352ac68 CM |
812 | /* |
813 | * defrag a given btree. If cacheonly == 1, this won't read from the disk, | |
814 | * otherwise every leaf in the btree is read and defragged. | |
815 | */ | |
e9d0b13b CM |
816 | int btrfs_defrag_root(struct btrfs_root *root, int cacheonly) |
817 | { | |
818 | struct btrfs_fs_info *info = root->fs_info; | |
e9d0b13b | 819 | struct btrfs_trans_handle *trans; |
8929ecfa | 820 | int ret; |
d3c2fdcf | 821 | unsigned long nr; |
e9d0b13b | 822 | |
8929ecfa | 823 | if (xchg(&root->defrag_running, 1)) |
e9d0b13b | 824 | return 0; |
8929ecfa | 825 | |
6b80053d | 826 | while (1) { |
8929ecfa YZ |
827 | trans = btrfs_start_transaction(root, 0); |
828 | if (IS_ERR(trans)) | |
829 | return PTR_ERR(trans); | |
830 | ||
e9d0b13b | 831 | ret = btrfs_defrag_leaves(trans, root, cacheonly); |
8929ecfa | 832 | |
d3c2fdcf | 833 | nr = trans->blocks_used; |
e9d0b13b | 834 | btrfs_end_transaction(trans, root); |
d3c2fdcf | 835 | btrfs_btree_balance_dirty(info->tree_root, nr); |
e9d0b13b CM |
836 | cond_resched(); |
837 | ||
7841cb28 | 838 | if (btrfs_fs_closing(root->fs_info) || ret != -EAGAIN) |
e9d0b13b CM |
839 | break; |
840 | } | |
841 | root->defrag_running = 0; | |
8929ecfa | 842 | return ret; |
e9d0b13b CM |
843 | } |
844 | ||
d352ac68 CM |
845 | /* |
846 | * new snapshots need to be created at a very specific time in the | |
847 | * transaction commit. This does the actual creation | |
848 | */ | |
80b6794d | 849 | static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, |
3063d29f CM |
850 | struct btrfs_fs_info *fs_info, |
851 | struct btrfs_pending_snapshot *pending) | |
852 | { | |
853 | struct btrfs_key key; | |
80b6794d | 854 | struct btrfs_root_item *new_root_item; |
3063d29f CM |
855 | struct btrfs_root *tree_root = fs_info->tree_root; |
856 | struct btrfs_root *root = pending->root; | |
6bdb72de | 857 | struct btrfs_root *parent_root; |
98c9942a | 858 | struct btrfs_block_rsv *rsv; |
6bdb72de | 859 | struct inode *parent_inode; |
6a912213 | 860 | struct dentry *parent; |
a22285a6 | 861 | struct dentry *dentry; |
3063d29f | 862 | struct extent_buffer *tmp; |
925baedd | 863 | struct extent_buffer *old; |
3063d29f | 864 | int ret; |
d68fc57b | 865 | u64 to_reserve = 0; |
6bdb72de | 866 | u64 index = 0; |
a22285a6 | 867 | u64 objectid; |
b83cc969 | 868 | u64 root_flags; |
3063d29f | 869 | |
98c9942a LB |
870 | rsv = trans->block_rsv; |
871 | ||
80b6794d CM |
872 | new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS); |
873 | if (!new_root_item) { | |
a22285a6 | 874 | pending->error = -ENOMEM; |
80b6794d CM |
875 | goto fail; |
876 | } | |
a22285a6 | 877 | |
581bb050 | 878 | ret = btrfs_find_free_objectid(tree_root, &objectid); |
a22285a6 YZ |
879 | if (ret) { |
880 | pending->error = ret; | |
3063d29f | 881 | goto fail; |
a22285a6 | 882 | } |
3063d29f | 883 | |
3fd0a558 | 884 | btrfs_reloc_pre_snapshot(trans, pending, &to_reserve); |
d68fc57b YZ |
885 | |
886 | if (to_reserve > 0) { | |
62f30c54 MX |
887 | ret = btrfs_block_rsv_add_noflush(root, &pending->block_rsv, |
888 | to_reserve); | |
d68fc57b YZ |
889 | if (ret) { |
890 | pending->error = ret; | |
891 | goto fail; | |
892 | } | |
893 | } | |
894 | ||
3063d29f | 895 | key.objectid = objectid; |
a22285a6 YZ |
896 | key.offset = (u64)-1; |
897 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3063d29f | 898 | |
a22285a6 | 899 | trans->block_rsv = &pending->block_rsv; |
3de4586c | 900 | |
a22285a6 | 901 | dentry = pending->dentry; |
6a912213 JB |
902 | parent = dget_parent(dentry); |
903 | parent_inode = parent->d_inode; | |
a22285a6 | 904 | parent_root = BTRFS_I(parent_inode)->root; |
7585717f | 905 | record_root_in_trans(trans, parent_root); |
a22285a6 | 906 | |
3063d29f CM |
907 | /* |
908 | * insert the directory item | |
909 | */ | |
3de4586c | 910 | ret = btrfs_set_inode_index(parent_inode, &index); |
6bdb72de | 911 | BUG_ON(ret); |
0660b5af | 912 | ret = btrfs_insert_dir_item(trans, parent_root, |
a22285a6 | 913 | dentry->d_name.name, dentry->d_name.len, |
16cdcec7 | 914 | parent_inode, &key, |
a22285a6 | 915 | BTRFS_FT_DIR, index); |
6bdb72de | 916 | BUG_ON(ret); |
0660b5af | 917 | |
a22285a6 YZ |
918 | btrfs_i_size_write(parent_inode, parent_inode->i_size + |
919 | dentry->d_name.len * 2); | |
52c26179 YZ |
920 | ret = btrfs_update_inode(trans, parent_root, parent_inode); |
921 | BUG_ON(ret); | |
922 | ||
e999376f CM |
923 | /* |
924 | * pull in the delayed directory update | |
925 | * and the delayed inode item | |
926 | * otherwise we corrupt the FS during | |
927 | * snapshot | |
928 | */ | |
929 | ret = btrfs_run_delayed_items(trans, root); | |
930 | BUG_ON(ret); | |
931 | ||
7585717f | 932 | record_root_in_trans(trans, root); |
6bdb72de SW |
933 | btrfs_set_root_last_snapshot(&root->root_item, trans->transid); |
934 | memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); | |
08fe4db1 | 935 | btrfs_check_and_init_root_item(new_root_item); |
6bdb72de | 936 | |
b83cc969 LZ |
937 | root_flags = btrfs_root_flags(new_root_item); |
938 | if (pending->readonly) | |
939 | root_flags |= BTRFS_ROOT_SUBVOL_RDONLY; | |
940 | else | |
941 | root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY; | |
942 | btrfs_set_root_flags(new_root_item, root_flags); | |
943 | ||
6bdb72de SW |
944 | old = btrfs_lock_root_node(root); |
945 | btrfs_cow_block(trans, root, old, NULL, 0, &old); | |
946 | btrfs_set_lock_blocking(old); | |
947 | ||
948 | btrfs_copy_root(trans, root, old, &tmp, objectid); | |
949 | btrfs_tree_unlock(old); | |
950 | free_extent_buffer(old); | |
951 | ||
f1ebcc74 LB |
952 | /* see comments in should_cow_block() */ |
953 | root->force_cow = 1; | |
954 | smp_wmb(); | |
955 | ||
6bdb72de | 956 | btrfs_set_root_node(new_root_item, tmp); |
a22285a6 YZ |
957 | /* record when the snapshot was created in key.offset */ |
958 | key.offset = trans->transid; | |
959 | ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); | |
6bdb72de SW |
960 | btrfs_tree_unlock(tmp); |
961 | free_extent_buffer(tmp); | |
a22285a6 | 962 | BUG_ON(ret); |
6bdb72de | 963 | |
a22285a6 YZ |
964 | /* |
965 | * insert root back/forward references | |
966 | */ | |
967 | ret = btrfs_add_root_ref(trans, tree_root, objectid, | |
0660b5af | 968 | parent_root->root_key.objectid, |
33345d01 | 969 | btrfs_ino(parent_inode), index, |
a22285a6 | 970 | dentry->d_name.name, dentry->d_name.len); |
0660b5af | 971 | BUG_ON(ret); |
6a912213 | 972 | dput(parent); |
0660b5af | 973 | |
a22285a6 YZ |
974 | key.offset = (u64)-1; |
975 | pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key); | |
976 | BUG_ON(IS_ERR(pending->snap)); | |
d68fc57b | 977 | |
3fd0a558 | 978 | btrfs_reloc_post_snapshot(trans, pending); |
3063d29f | 979 | fail: |
6bdb72de | 980 | kfree(new_root_item); |
98c9942a | 981 | trans->block_rsv = rsv; |
a22285a6 YZ |
982 | btrfs_block_rsv_release(root, &pending->block_rsv, (u64)-1); |
983 | return 0; | |
3063d29f CM |
984 | } |
985 | ||
d352ac68 CM |
986 | /* |
987 | * create all the snapshots we've scheduled for creation | |
988 | */ | |
80b6794d CM |
989 | static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans, |
990 | struct btrfs_fs_info *fs_info) | |
3de4586c CM |
991 | { |
992 | struct btrfs_pending_snapshot *pending; | |
993 | struct list_head *head = &trans->transaction->pending_snapshots; | |
3de4586c CM |
994 | int ret; |
995 | ||
c6e30871 | 996 | list_for_each_entry(pending, head, list) { |
3de4586c CM |
997 | ret = create_pending_snapshot(trans, fs_info, pending); |
998 | BUG_ON(ret); | |
999 | } | |
1000 | return 0; | |
1001 | } | |
1002 | ||
5d4f98a2 YZ |
1003 | static void update_super_roots(struct btrfs_root *root) |
1004 | { | |
1005 | struct btrfs_root_item *root_item; | |
1006 | struct btrfs_super_block *super; | |
1007 | ||
6c41761f | 1008 | super = root->fs_info->super_copy; |
5d4f98a2 YZ |
1009 | |
1010 | root_item = &root->fs_info->chunk_root->root_item; | |
1011 | super->chunk_root = root_item->bytenr; | |
1012 | super->chunk_root_generation = root_item->generation; | |
1013 | super->chunk_root_level = root_item->level; | |
1014 | ||
1015 | root_item = &root->fs_info->tree_root->root_item; | |
1016 | super->root = root_item->bytenr; | |
1017 | super->generation = root_item->generation; | |
1018 | super->root_level = root_item->level; | |
73bc1876 | 1019 | if (btrfs_test_opt(root, SPACE_CACHE)) |
0af3d00b | 1020 | super->cache_generation = root_item->generation; |
5d4f98a2 YZ |
1021 | } |
1022 | ||
f36f3042 CM |
1023 | int btrfs_transaction_in_commit(struct btrfs_fs_info *info) |
1024 | { | |
1025 | int ret = 0; | |
a4abeea4 | 1026 | spin_lock(&info->trans_lock); |
f36f3042 CM |
1027 | if (info->running_transaction) |
1028 | ret = info->running_transaction->in_commit; | |
a4abeea4 | 1029 | spin_unlock(&info->trans_lock); |
f36f3042 CM |
1030 | return ret; |
1031 | } | |
1032 | ||
8929ecfa YZ |
1033 | int btrfs_transaction_blocked(struct btrfs_fs_info *info) |
1034 | { | |
1035 | int ret = 0; | |
a4abeea4 | 1036 | spin_lock(&info->trans_lock); |
8929ecfa YZ |
1037 | if (info->running_transaction) |
1038 | ret = info->running_transaction->blocked; | |
a4abeea4 | 1039 | spin_unlock(&info->trans_lock); |
8929ecfa YZ |
1040 | return ret; |
1041 | } | |
1042 | ||
bb9c12c9 SW |
1043 | /* |
1044 | * wait for the current transaction commit to start and block subsequent | |
1045 | * transaction joins | |
1046 | */ | |
1047 | static void wait_current_trans_commit_start(struct btrfs_root *root, | |
1048 | struct btrfs_transaction *trans) | |
1049 | { | |
72d63ed6 | 1050 | wait_event(root->fs_info->transaction_blocked_wait, trans->in_commit); |
bb9c12c9 SW |
1051 | } |
1052 | ||
1053 | /* | |
1054 | * wait for the current transaction to start and then become unblocked. | |
1055 | * caller holds ref. | |
1056 | */ | |
1057 | static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root, | |
1058 | struct btrfs_transaction *trans) | |
1059 | { | |
72d63ed6 LZ |
1060 | wait_event(root->fs_info->transaction_wait, |
1061 | trans->commit_done || (trans->in_commit && !trans->blocked)); | |
bb9c12c9 SW |
1062 | } |
1063 | ||
1064 | /* | |
1065 | * commit transactions asynchronously. once btrfs_commit_transaction_async | |
1066 | * returns, any subsequent transaction will not be allowed to join. | |
1067 | */ | |
1068 | struct btrfs_async_commit { | |
1069 | struct btrfs_trans_handle *newtrans; | |
1070 | struct btrfs_root *root; | |
1071 | struct delayed_work work; | |
1072 | }; | |
1073 | ||
1074 | static void do_async_commit(struct work_struct *work) | |
1075 | { | |
1076 | struct btrfs_async_commit *ac = | |
1077 | container_of(work, struct btrfs_async_commit, work.work); | |
1078 | ||
1079 | btrfs_commit_transaction(ac->newtrans, ac->root); | |
1080 | kfree(ac); | |
1081 | } | |
1082 | ||
1083 | int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans, | |
1084 | struct btrfs_root *root, | |
1085 | int wait_for_unblock) | |
1086 | { | |
1087 | struct btrfs_async_commit *ac; | |
1088 | struct btrfs_transaction *cur_trans; | |
1089 | ||
1090 | ac = kmalloc(sizeof(*ac), GFP_NOFS); | |
db5b493a TI |
1091 | if (!ac) |
1092 | return -ENOMEM; | |
bb9c12c9 SW |
1093 | |
1094 | INIT_DELAYED_WORK(&ac->work, do_async_commit); | |
1095 | ac->root = root; | |
7a7eaa40 | 1096 | ac->newtrans = btrfs_join_transaction(root); |
3612b495 TI |
1097 | if (IS_ERR(ac->newtrans)) { |
1098 | int err = PTR_ERR(ac->newtrans); | |
1099 | kfree(ac); | |
1100 | return err; | |
1101 | } | |
bb9c12c9 SW |
1102 | |
1103 | /* take transaction reference */ | |
bb9c12c9 | 1104 | cur_trans = trans->transaction; |
13c5a93e | 1105 | atomic_inc(&cur_trans->use_count); |
bb9c12c9 SW |
1106 | |
1107 | btrfs_end_transaction(trans, root); | |
1108 | schedule_delayed_work(&ac->work, 0); | |
1109 | ||
1110 | /* wait for transaction to start and unblock */ | |
bb9c12c9 SW |
1111 | if (wait_for_unblock) |
1112 | wait_current_trans_commit_start_and_unblock(root, cur_trans); | |
1113 | else | |
1114 | wait_current_trans_commit_start(root, cur_trans); | |
bb9c12c9 | 1115 | |
38e88054 SW |
1116 | if (current->journal_info == trans) |
1117 | current->journal_info = NULL; | |
1118 | ||
1119 | put_transaction(cur_trans); | |
bb9c12c9 SW |
1120 | return 0; |
1121 | } | |
1122 | ||
1123 | /* | |
1124 | * btrfs_transaction state sequence: | |
1125 | * in_commit = 0, blocked = 0 (initial) | |
1126 | * in_commit = 1, blocked = 1 | |
1127 | * blocked = 0 | |
1128 | * commit_done = 1 | |
1129 | */ | |
79154b1b CM |
1130 | int btrfs_commit_transaction(struct btrfs_trans_handle *trans, |
1131 | struct btrfs_root *root) | |
1132 | { | |
15ee9bc7 | 1133 | unsigned long joined = 0; |
79154b1b | 1134 | struct btrfs_transaction *cur_trans; |
8fd17795 | 1135 | struct btrfs_transaction *prev_trans = NULL; |
79154b1b | 1136 | DEFINE_WAIT(wait); |
15ee9bc7 | 1137 | int ret; |
89573b9c CM |
1138 | int should_grow = 0; |
1139 | unsigned long now = get_seconds(); | |
dccae999 | 1140 | int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT); |
79154b1b | 1141 | |
5a3f23d5 CM |
1142 | btrfs_run_ordered_operations(root, 0); |
1143 | ||
b24e03db | 1144 | btrfs_trans_release_metadata(trans, root); |
9c8d86db JB |
1145 | trans->block_rsv = NULL; |
1146 | ||
56bec294 CM |
1147 | /* make a pass through all the delayed refs we have so far |
1148 | * any runnings procs may add more while we are here | |
1149 | */ | |
1150 | ret = btrfs_run_delayed_refs(trans, root, 0); | |
1151 | BUG_ON(ret); | |
1152 | ||
b7ec40d7 | 1153 | cur_trans = trans->transaction; |
56bec294 CM |
1154 | /* |
1155 | * set the flushing flag so procs in this transaction have to | |
1156 | * start sending their work down. | |
1157 | */ | |
b7ec40d7 | 1158 | cur_trans->delayed_refs.flushing = 1; |
56bec294 | 1159 | |
c3e69d58 | 1160 | ret = btrfs_run_delayed_refs(trans, root, 0); |
56bec294 CM |
1161 | BUG_ON(ret); |
1162 | ||
a4abeea4 | 1163 | spin_lock(&cur_trans->commit_lock); |
b7ec40d7 | 1164 | if (cur_trans->in_commit) { |
a4abeea4 | 1165 | spin_unlock(&cur_trans->commit_lock); |
13c5a93e | 1166 | atomic_inc(&cur_trans->use_count); |
79154b1b | 1167 | btrfs_end_transaction(trans, root); |
ccd467d6 | 1168 | |
b9c8300c | 1169 | wait_for_commit(root, cur_trans); |
15ee9bc7 | 1170 | |
79154b1b | 1171 | put_transaction(cur_trans); |
15ee9bc7 | 1172 | |
79154b1b CM |
1173 | return 0; |
1174 | } | |
4313b399 | 1175 | |
2c90e5d6 | 1176 | trans->transaction->in_commit = 1; |
f9295749 | 1177 | trans->transaction->blocked = 1; |
a4abeea4 | 1178 | spin_unlock(&cur_trans->commit_lock); |
bb9c12c9 SW |
1179 | wake_up(&root->fs_info->transaction_blocked_wait); |
1180 | ||
a4abeea4 | 1181 | spin_lock(&root->fs_info->trans_lock); |
ccd467d6 CM |
1182 | if (cur_trans->list.prev != &root->fs_info->trans_list) { |
1183 | prev_trans = list_entry(cur_trans->list.prev, | |
1184 | struct btrfs_transaction, list); | |
1185 | if (!prev_trans->commit_done) { | |
13c5a93e | 1186 | atomic_inc(&prev_trans->use_count); |
a4abeea4 | 1187 | spin_unlock(&root->fs_info->trans_lock); |
ccd467d6 CM |
1188 | |
1189 | wait_for_commit(root, prev_trans); | |
ccd467d6 | 1190 | |
15ee9bc7 | 1191 | put_transaction(prev_trans); |
a4abeea4 JB |
1192 | } else { |
1193 | spin_unlock(&root->fs_info->trans_lock); | |
ccd467d6 | 1194 | } |
a4abeea4 JB |
1195 | } else { |
1196 | spin_unlock(&root->fs_info->trans_lock); | |
ccd467d6 | 1197 | } |
15ee9bc7 | 1198 | |
89573b9c CM |
1199 | if (now < cur_trans->start_time || now - cur_trans->start_time < 1) |
1200 | should_grow = 1; | |
1201 | ||
15ee9bc7 | 1202 | do { |
7ea394f1 | 1203 | int snap_pending = 0; |
a4abeea4 | 1204 | |
15ee9bc7 | 1205 | joined = cur_trans->num_joined; |
7ea394f1 YZ |
1206 | if (!list_empty(&trans->transaction->pending_snapshots)) |
1207 | snap_pending = 1; | |
1208 | ||
2c90e5d6 | 1209 | WARN_ON(cur_trans != trans->transaction); |
15ee9bc7 | 1210 | |
0bdb1db2 | 1211 | if (flush_on_commit || snap_pending) { |
24bbcf04 YZ |
1212 | btrfs_start_delalloc_inodes(root, 1); |
1213 | ret = btrfs_wait_ordered_extents(root, 0, 1); | |
ebecd3d9 | 1214 | BUG_ON(ret); |
7ea394f1 YZ |
1215 | } |
1216 | ||
16cdcec7 MX |
1217 | ret = btrfs_run_delayed_items(trans, root); |
1218 | BUG_ON(ret); | |
1219 | ||
5a3f23d5 CM |
1220 | /* |
1221 | * rename don't use btrfs_join_transaction, so, once we | |
1222 | * set the transaction to blocked above, we aren't going | |
1223 | * to get any new ordered operations. We can safely run | |
1224 | * it here and no for sure that nothing new will be added | |
1225 | * to the list | |
1226 | */ | |
1227 | btrfs_run_ordered_operations(root, 1); | |
1228 | ||
ed3b3d31 CM |
1229 | prepare_to_wait(&cur_trans->writer_wait, &wait, |
1230 | TASK_UNINTERRUPTIBLE); | |
1231 | ||
13c5a93e | 1232 | if (atomic_read(&cur_trans->num_writers) > 1) |
99d16cbc SW |
1233 | schedule_timeout(MAX_SCHEDULE_TIMEOUT); |
1234 | else if (should_grow) | |
1235 | schedule_timeout(1); | |
15ee9bc7 | 1236 | |
15ee9bc7 | 1237 | finish_wait(&cur_trans->writer_wait, &wait); |
13c5a93e | 1238 | } while (atomic_read(&cur_trans->num_writers) > 1 || |
89573b9c | 1239 | (should_grow && cur_trans->num_joined != joined)); |
15ee9bc7 | 1240 | |
ed0ca140 JB |
1241 | /* |
1242 | * Ok now we need to make sure to block out any other joins while we | |
1243 | * commit the transaction. We could have started a join before setting | |
1244 | * no_join so make sure to wait for num_writers to == 1 again. | |
1245 | */ | |
1246 | spin_lock(&root->fs_info->trans_lock); | |
1247 | root->fs_info->trans_no_join = 1; | |
1248 | spin_unlock(&root->fs_info->trans_lock); | |
1249 | wait_event(cur_trans->writer_wait, | |
1250 | atomic_read(&cur_trans->num_writers) == 1); | |
1251 | ||
7585717f CM |
1252 | /* |
1253 | * the reloc mutex makes sure that we stop | |
1254 | * the balancing code from coming in and moving | |
1255 | * extents around in the middle of the commit | |
1256 | */ | |
1257 | mutex_lock(&root->fs_info->reloc_mutex); | |
1258 | ||
e999376f | 1259 | ret = btrfs_run_delayed_items(trans, root); |
3063d29f CM |
1260 | BUG_ON(ret); |
1261 | ||
e999376f | 1262 | ret = create_pending_snapshots(trans, root->fs_info); |
16cdcec7 MX |
1263 | BUG_ON(ret); |
1264 | ||
56bec294 CM |
1265 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
1266 | BUG_ON(ret); | |
1267 | ||
e999376f CM |
1268 | /* |
1269 | * make sure none of the code above managed to slip in a | |
1270 | * delayed item | |
1271 | */ | |
1272 | btrfs_assert_delayed_root_empty(root); | |
1273 | ||
2c90e5d6 | 1274 | WARN_ON(cur_trans != trans->transaction); |
dc17ff8f | 1275 | |
a2de733c | 1276 | btrfs_scrub_pause(root); |
e02119d5 CM |
1277 | /* btrfs_commit_tree_roots is responsible for getting the |
1278 | * various roots consistent with each other. Every pointer | |
1279 | * in the tree of tree roots has to point to the most up to date | |
1280 | * root for every subvolume and other tree. So, we have to keep | |
1281 | * the tree logging code from jumping in and changing any | |
1282 | * of the trees. | |
1283 | * | |
1284 | * At this point in the commit, there can't be any tree-log | |
1285 | * writers, but a little lower down we drop the trans mutex | |
1286 | * and let new people in. By holding the tree_log_mutex | |
1287 | * from now until after the super is written, we avoid races | |
1288 | * with the tree-log code. | |
1289 | */ | |
1290 | mutex_lock(&root->fs_info->tree_log_mutex); | |
1291 | ||
5d4f98a2 | 1292 | ret = commit_fs_roots(trans, root); |
54aa1f4d CM |
1293 | BUG_ON(ret); |
1294 | ||
5d4f98a2 | 1295 | /* commit_fs_roots gets rid of all the tree log roots, it is now |
e02119d5 CM |
1296 | * safe to free the root of tree log roots |
1297 | */ | |
1298 | btrfs_free_log_root_tree(trans, root->fs_info); | |
1299 | ||
5d4f98a2 | 1300 | ret = commit_cowonly_roots(trans, root); |
79154b1b | 1301 | BUG_ON(ret); |
54aa1f4d | 1302 | |
11833d66 YZ |
1303 | btrfs_prepare_extent_commit(trans, root); |
1304 | ||
78fae27e | 1305 | cur_trans = root->fs_info->running_transaction; |
5d4f98a2 YZ |
1306 | |
1307 | btrfs_set_root_node(&root->fs_info->tree_root->root_item, | |
1308 | root->fs_info->tree_root->node); | |
817d52f8 | 1309 | switch_commit_root(root->fs_info->tree_root); |
5d4f98a2 YZ |
1310 | |
1311 | btrfs_set_root_node(&root->fs_info->chunk_root->root_item, | |
1312 | root->fs_info->chunk_root->node); | |
817d52f8 | 1313 | switch_commit_root(root->fs_info->chunk_root); |
5d4f98a2 YZ |
1314 | |
1315 | update_super_roots(root); | |
e02119d5 CM |
1316 | |
1317 | if (!root->fs_info->log_root_recovering) { | |
6c41761f DS |
1318 | btrfs_set_super_log_root(root->fs_info->super_copy, 0); |
1319 | btrfs_set_super_log_root_level(root->fs_info->super_copy, 0); | |
e02119d5 CM |
1320 | } |
1321 | ||
6c41761f DS |
1322 | memcpy(root->fs_info->super_for_commit, root->fs_info->super_copy, |
1323 | sizeof(*root->fs_info->super_copy)); | |
ccd467d6 | 1324 | |
f9295749 | 1325 | trans->transaction->blocked = 0; |
a4abeea4 JB |
1326 | spin_lock(&root->fs_info->trans_lock); |
1327 | root->fs_info->running_transaction = NULL; | |
1328 | root->fs_info->trans_no_join = 0; | |
1329 | spin_unlock(&root->fs_info->trans_lock); | |
7585717f | 1330 | mutex_unlock(&root->fs_info->reloc_mutex); |
b7ec40d7 | 1331 | |
f9295749 | 1332 | wake_up(&root->fs_info->transaction_wait); |
e6dcd2dc | 1333 | |
79154b1b CM |
1334 | ret = btrfs_write_and_wait_transaction(trans, root); |
1335 | BUG_ON(ret); | |
a512bbf8 | 1336 | write_ctree_super(trans, root, 0); |
4313b399 | 1337 | |
e02119d5 CM |
1338 | /* |
1339 | * the super is written, we can safely allow the tree-loggers | |
1340 | * to go about their business | |
1341 | */ | |
1342 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
1343 | ||
11833d66 | 1344 | btrfs_finish_extent_commit(trans, root); |
4313b399 | 1345 | |
2c90e5d6 | 1346 | cur_trans->commit_done = 1; |
b7ec40d7 | 1347 | |
15ee9bc7 | 1348 | root->fs_info->last_trans_committed = cur_trans->transid; |
817d52f8 | 1349 | |
2c90e5d6 | 1350 | wake_up(&cur_trans->commit_wait); |
3de4586c | 1351 | |
a4abeea4 | 1352 | spin_lock(&root->fs_info->trans_lock); |
13c5a93e | 1353 | list_del_init(&cur_trans->list); |
a4abeea4 JB |
1354 | spin_unlock(&root->fs_info->trans_lock); |
1355 | ||
78fae27e | 1356 | put_transaction(cur_trans); |
79154b1b | 1357 | put_transaction(cur_trans); |
58176a96 | 1358 | |
1abe9b8a | 1359 | trace_btrfs_transaction_commit(root); |
1360 | ||
a2de733c AJ |
1361 | btrfs_scrub_continue(root); |
1362 | ||
9ed74f2d JB |
1363 | if (current->journal_info == trans) |
1364 | current->journal_info = NULL; | |
1365 | ||
2c90e5d6 | 1366 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
24bbcf04 YZ |
1367 | |
1368 | if (current != root->fs_info->transaction_kthread) | |
1369 | btrfs_run_delayed_iputs(root); | |
1370 | ||
79154b1b CM |
1371 | return ret; |
1372 | } | |
1373 | ||
d352ac68 CM |
1374 | /* |
1375 | * interface function to delete all the snapshots we have scheduled for deletion | |
1376 | */ | |
e9d0b13b CM |
1377 | int btrfs_clean_old_snapshots(struct btrfs_root *root) |
1378 | { | |
5d4f98a2 YZ |
1379 | LIST_HEAD(list); |
1380 | struct btrfs_fs_info *fs_info = root->fs_info; | |
1381 | ||
a4abeea4 | 1382 | spin_lock(&fs_info->trans_lock); |
5d4f98a2 | 1383 | list_splice_init(&fs_info->dead_roots, &list); |
a4abeea4 | 1384 | spin_unlock(&fs_info->trans_lock); |
e9d0b13b | 1385 | |
5d4f98a2 YZ |
1386 | while (!list_empty(&list)) { |
1387 | root = list_entry(list.next, struct btrfs_root, root_list); | |
76dda93c YZ |
1388 | list_del(&root->root_list); |
1389 | ||
16cdcec7 MX |
1390 | btrfs_kill_all_delayed_nodes(root); |
1391 | ||
76dda93c YZ |
1392 | if (btrfs_header_backref_rev(root->node) < |
1393 | BTRFS_MIXED_BACKREF_REV) | |
66d7e7f0 | 1394 | btrfs_drop_snapshot(root, NULL, 0, 0); |
76dda93c | 1395 | else |
66d7e7f0 | 1396 | btrfs_drop_snapshot(root, NULL, 1, 0); |
e9d0b13b CM |
1397 | } |
1398 | return 0; | |
1399 | } |