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" |
733f4fbb | 31 | #include "volumes.h" |
79154b1b | 32 | |
0f7d52f4 CM |
33 | #define BTRFS_ROOT_TRANS_TAG 0 |
34 | ||
49b25e05 | 35 | void put_transaction(struct btrfs_transaction *transaction) |
79154b1b | 36 | { |
13c5a93e JB |
37 | WARN_ON(atomic_read(&transaction->use_count) == 0); |
38 | if (atomic_dec_and_test(&transaction->use_count)) { | |
a4abeea4 | 39 | BUG_ON(!list_empty(&transaction->list)); |
00f04b88 | 40 | WARN_ON(transaction->delayed_refs.root.rb_node); |
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; | |
19ae4e81 | 58 | struct btrfs_fs_info *fs_info = root->fs_info; |
a4abeea4 | 59 | |
19ae4e81 | 60 | spin_lock(&fs_info->trans_lock); |
d43317dc | 61 | loop: |
49b25e05 | 62 | /* The file system has been taken offline. No new transactions. */ |
19ae4e81 JS |
63 | if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) { |
64 | spin_unlock(&fs_info->trans_lock); | |
49b25e05 JM |
65 | return -EROFS; |
66 | } | |
67 | ||
19ae4e81 | 68 | if (fs_info->trans_no_join) { |
a4abeea4 | 69 | if (!nofail) { |
19ae4e81 | 70 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 JB |
71 | return -EBUSY; |
72 | } | |
73 | } | |
74 | ||
19ae4e81 | 75 | cur_trans = fs_info->running_transaction; |
a4abeea4 | 76 | if (cur_trans) { |
871383be | 77 | if (cur_trans->aborted) { |
19ae4e81 | 78 | spin_unlock(&fs_info->trans_lock); |
49b25e05 | 79 | return cur_trans->aborted; |
871383be | 80 | } |
a4abeea4 | 81 | atomic_inc(&cur_trans->use_count); |
13c5a93e | 82 | atomic_inc(&cur_trans->num_writers); |
15ee9bc7 | 83 | cur_trans->num_joined++; |
19ae4e81 | 84 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 | 85 | return 0; |
79154b1b | 86 | } |
19ae4e81 | 87 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 JB |
88 | |
89 | cur_trans = kmem_cache_alloc(btrfs_transaction_cachep, GFP_NOFS); | |
90 | if (!cur_trans) | |
91 | return -ENOMEM; | |
d43317dc | 92 | |
19ae4e81 JS |
93 | spin_lock(&fs_info->trans_lock); |
94 | if (fs_info->running_transaction) { | |
d43317dc CM |
95 | /* |
96 | * someone started a transaction after we unlocked. Make sure | |
97 | * to redo the trans_no_join checks above | |
98 | */ | |
a4abeea4 | 99 | kmem_cache_free(btrfs_transaction_cachep, cur_trans); |
19ae4e81 | 100 | cur_trans = fs_info->running_transaction; |
d43317dc | 101 | goto loop; |
7b8b92af JB |
102 | } else if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) { |
103 | spin_unlock(&root->fs_info->trans_lock); | |
104 | kmem_cache_free(btrfs_transaction_cachep, cur_trans); | |
105 | return -EROFS; | |
79154b1b | 106 | } |
d43317dc | 107 | |
a4abeea4 JB |
108 | atomic_set(&cur_trans->num_writers, 1); |
109 | cur_trans->num_joined = 0; | |
110 | init_waitqueue_head(&cur_trans->writer_wait); | |
111 | init_waitqueue_head(&cur_trans->commit_wait); | |
112 | cur_trans->in_commit = 0; | |
113 | cur_trans->blocked = 0; | |
114 | /* | |
115 | * One for this trans handle, one so it will live on until we | |
116 | * commit the transaction. | |
117 | */ | |
118 | atomic_set(&cur_trans->use_count, 2); | |
119 | cur_trans->commit_done = 0; | |
120 | cur_trans->start_time = get_seconds(); | |
121 | ||
122 | cur_trans->delayed_refs.root = RB_ROOT; | |
123 | cur_trans->delayed_refs.num_entries = 0; | |
124 | cur_trans->delayed_refs.num_heads_ready = 0; | |
125 | cur_trans->delayed_refs.num_heads = 0; | |
126 | cur_trans->delayed_refs.flushing = 0; | |
127 | cur_trans->delayed_refs.run_delayed_start = 0; | |
20b297d6 JS |
128 | |
129 | /* | |
130 | * although the tree mod log is per file system and not per transaction, | |
131 | * the log must never go across transaction boundaries. | |
132 | */ | |
133 | smp_mb(); | |
134 | if (!list_empty(&fs_info->tree_mod_seq_list)) { | |
135 | printk(KERN_ERR "btrfs: tree_mod_seq_list not empty when " | |
136 | "creating a fresh transaction\n"); | |
137 | WARN_ON(1); | |
138 | } | |
139 | if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log)) { | |
140 | printk(KERN_ERR "btrfs: tree_mod_log rb tree not empty when " | |
141 | "creating a fresh transaction\n"); | |
142 | WARN_ON(1); | |
143 | } | |
144 | atomic_set(&fs_info->tree_mod_seq, 0); | |
145 | ||
a4abeea4 JB |
146 | spin_lock_init(&cur_trans->commit_lock); |
147 | spin_lock_init(&cur_trans->delayed_refs.lock); | |
148 | ||
149 | INIT_LIST_HEAD(&cur_trans->pending_snapshots); | |
19ae4e81 | 150 | list_add_tail(&cur_trans->list, &fs_info->trans_list); |
a4abeea4 | 151 | extent_io_tree_init(&cur_trans->dirty_pages, |
19ae4e81 JS |
152 | fs_info->btree_inode->i_mapping); |
153 | fs_info->generation++; | |
154 | cur_trans->transid = fs_info->generation; | |
155 | fs_info->running_transaction = cur_trans; | |
49b25e05 | 156 | cur_trans->aborted = 0; |
19ae4e81 | 157 | spin_unlock(&fs_info->trans_lock); |
15ee9bc7 | 158 | |
79154b1b CM |
159 | return 0; |
160 | } | |
161 | ||
d352ac68 | 162 | /* |
d397712b CM |
163 | * this does all the record keeping required to make sure that a reference |
164 | * counted root is properly recorded in a given transaction. This is required | |
165 | * to make sure the old root from before we joined the transaction is deleted | |
166 | * when the transaction commits | |
d352ac68 | 167 | */ |
7585717f | 168 | static int record_root_in_trans(struct btrfs_trans_handle *trans, |
a4abeea4 | 169 | struct btrfs_root *root) |
6702ed49 | 170 | { |
5d4f98a2 | 171 | if (root->ref_cows && root->last_trans < trans->transid) { |
6702ed49 | 172 | WARN_ON(root == root->fs_info->extent_root); |
5d4f98a2 YZ |
173 | WARN_ON(root->commit_root != root->node); |
174 | ||
7585717f CM |
175 | /* |
176 | * see below for in_trans_setup usage rules | |
177 | * we have the reloc mutex held now, so there | |
178 | * is only one writer in this function | |
179 | */ | |
180 | root->in_trans_setup = 1; | |
181 | ||
182 | /* make sure readers find in_trans_setup before | |
183 | * they find our root->last_trans update | |
184 | */ | |
185 | smp_wmb(); | |
186 | ||
a4abeea4 JB |
187 | spin_lock(&root->fs_info->fs_roots_radix_lock); |
188 | if (root->last_trans == trans->transid) { | |
189 | spin_unlock(&root->fs_info->fs_roots_radix_lock); | |
190 | return 0; | |
191 | } | |
5d4f98a2 YZ |
192 | radix_tree_tag_set(&root->fs_info->fs_roots_radix, |
193 | (unsigned long)root->root_key.objectid, | |
194 | BTRFS_ROOT_TRANS_TAG); | |
a4abeea4 | 195 | spin_unlock(&root->fs_info->fs_roots_radix_lock); |
7585717f CM |
196 | root->last_trans = trans->transid; |
197 | ||
198 | /* this is pretty tricky. We don't want to | |
199 | * take the relocation lock in btrfs_record_root_in_trans | |
200 | * unless we're really doing the first setup for this root in | |
201 | * this transaction. | |
202 | * | |
203 | * Normally we'd use root->last_trans as a flag to decide | |
204 | * if we want to take the expensive mutex. | |
205 | * | |
206 | * But, we have to set root->last_trans before we | |
207 | * init the relocation root, otherwise, we trip over warnings | |
208 | * in ctree.c. The solution used here is to flag ourselves | |
209 | * with root->in_trans_setup. When this is 1, we're still | |
210 | * fixing up the reloc trees and everyone must wait. | |
211 | * | |
212 | * When this is zero, they can trust root->last_trans and fly | |
213 | * through btrfs_record_root_in_trans without having to take the | |
214 | * lock. smp_wmb() makes sure that all the writes above are | |
215 | * done before we pop in the zero below | |
216 | */ | |
5d4f98a2 | 217 | btrfs_init_reloc_root(trans, root); |
7585717f CM |
218 | smp_wmb(); |
219 | root->in_trans_setup = 0; | |
5d4f98a2 YZ |
220 | } |
221 | return 0; | |
222 | } | |
bcc63abb | 223 | |
7585717f CM |
224 | |
225 | int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, | |
226 | struct btrfs_root *root) | |
227 | { | |
228 | if (!root->ref_cows) | |
229 | return 0; | |
230 | ||
231 | /* | |
232 | * see record_root_in_trans for comments about in_trans_setup usage | |
233 | * and barriers | |
234 | */ | |
235 | smp_rmb(); | |
236 | if (root->last_trans == trans->transid && | |
237 | !root->in_trans_setup) | |
238 | return 0; | |
239 | ||
240 | mutex_lock(&root->fs_info->reloc_mutex); | |
241 | record_root_in_trans(trans, root); | |
242 | mutex_unlock(&root->fs_info->reloc_mutex); | |
243 | ||
244 | return 0; | |
245 | } | |
246 | ||
d352ac68 CM |
247 | /* wait for commit against the current transaction to become unblocked |
248 | * when this is done, it is safe to start a new transaction, but the current | |
249 | * transaction might not be fully on disk. | |
250 | */ | |
37d1aeee | 251 | static void wait_current_trans(struct btrfs_root *root) |
79154b1b | 252 | { |
f9295749 | 253 | struct btrfs_transaction *cur_trans; |
79154b1b | 254 | |
a4abeea4 | 255 | spin_lock(&root->fs_info->trans_lock); |
f9295749 | 256 | cur_trans = root->fs_info->running_transaction; |
37d1aeee | 257 | if (cur_trans && cur_trans->blocked) { |
13c5a93e | 258 | atomic_inc(&cur_trans->use_count); |
a4abeea4 | 259 | spin_unlock(&root->fs_info->trans_lock); |
72d63ed6 LZ |
260 | |
261 | wait_event(root->fs_info->transaction_wait, | |
262 | !cur_trans->blocked); | |
f9295749 | 263 | put_transaction(cur_trans); |
a4abeea4 JB |
264 | } else { |
265 | spin_unlock(&root->fs_info->trans_lock); | |
f9295749 | 266 | } |
37d1aeee CM |
267 | } |
268 | ||
249ac1e5 JB |
269 | enum btrfs_trans_type { |
270 | TRANS_START, | |
271 | TRANS_JOIN, | |
272 | TRANS_USERSPACE, | |
0af3d00b | 273 | TRANS_JOIN_NOLOCK, |
249ac1e5 JB |
274 | }; |
275 | ||
a22285a6 YZ |
276 | static int may_wait_transaction(struct btrfs_root *root, int type) |
277 | { | |
a4abeea4 JB |
278 | if (root->fs_info->log_root_recovering) |
279 | return 0; | |
280 | ||
281 | if (type == TRANS_USERSPACE) | |
282 | return 1; | |
283 | ||
284 | if (type == TRANS_START && | |
285 | !atomic_read(&root->fs_info->open_ioctl_trans)) | |
a22285a6 | 286 | return 1; |
a4abeea4 | 287 | |
a22285a6 YZ |
288 | return 0; |
289 | } | |
290 | ||
e02119d5 | 291 | static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root, |
a22285a6 | 292 | u64 num_items, int type) |
37d1aeee | 293 | { |
a22285a6 YZ |
294 | struct btrfs_trans_handle *h; |
295 | struct btrfs_transaction *cur_trans; | |
b5009945 | 296 | u64 num_bytes = 0; |
37d1aeee | 297 | int ret; |
acce952b | 298 | |
299 | if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) | |
300 | return ERR_PTR(-EROFS); | |
2a1eb461 JB |
301 | |
302 | if (current->journal_info) { | |
303 | WARN_ON(type != TRANS_JOIN && type != TRANS_JOIN_NOLOCK); | |
304 | h = current->journal_info; | |
305 | h->use_count++; | |
306 | h->orig_rsv = h->block_rsv; | |
307 | h->block_rsv = NULL; | |
308 | goto got_it; | |
309 | } | |
b5009945 JB |
310 | |
311 | /* | |
312 | * Do the reservation before we join the transaction so we can do all | |
313 | * the appropriate flushing if need be. | |
314 | */ | |
315 | if (num_items > 0 && root != root->fs_info->chunk_root) { | |
316 | num_bytes = btrfs_calc_trans_metadata_size(root, num_items); | |
4a92b1b8 | 317 | ret = btrfs_block_rsv_add(root, |
b5009945 JB |
318 | &root->fs_info->trans_block_rsv, |
319 | num_bytes); | |
320 | if (ret) | |
321 | return ERR_PTR(ret); | |
322 | } | |
a22285a6 YZ |
323 | again: |
324 | h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS); | |
325 | if (!h) | |
326 | return ERR_PTR(-ENOMEM); | |
37d1aeee | 327 | |
a22285a6 | 328 | if (may_wait_transaction(root, type)) |
37d1aeee | 329 | wait_current_trans(root); |
a22285a6 | 330 | |
a4abeea4 JB |
331 | do { |
332 | ret = join_transaction(root, type == TRANS_JOIN_NOLOCK); | |
333 | if (ret == -EBUSY) | |
334 | wait_current_trans(root); | |
335 | } while (ret == -EBUSY); | |
336 | ||
db5b493a | 337 | if (ret < 0) { |
6e8df2ae | 338 | kmem_cache_free(btrfs_trans_handle_cachep, h); |
db5b493a TI |
339 | return ERR_PTR(ret); |
340 | } | |
0f7d52f4 | 341 | |
a22285a6 | 342 | cur_trans = root->fs_info->running_transaction; |
a22285a6 YZ |
343 | |
344 | h->transid = cur_trans->transid; | |
345 | h->transaction = cur_trans; | |
79154b1b | 346 | h->blocks_used = 0; |
a22285a6 | 347 | h->bytes_reserved = 0; |
d13603ef | 348 | h->root = root; |
56bec294 | 349 | h->delayed_ref_updates = 0; |
2a1eb461 | 350 | h->use_count = 1; |
f0486c68 | 351 | h->block_rsv = NULL; |
2a1eb461 | 352 | h->orig_rsv = NULL; |
49b25e05 | 353 | h->aborted = 0; |
bed92eae AJ |
354 | h->delayed_ref_elem.seq = 0; |
355 | INIT_LIST_HEAD(&h->qgroup_ref_list); | |
b7ec40d7 | 356 | |
a22285a6 YZ |
357 | smp_mb(); |
358 | if (cur_trans->blocked && may_wait_transaction(root, type)) { | |
359 | btrfs_commit_transaction(h, root); | |
360 | goto again; | |
361 | } | |
362 | ||
b5009945 | 363 | if (num_bytes) { |
8c2a3ca2 | 364 | trace_btrfs_space_reservation(root->fs_info, "transaction", |
2bcc0328 | 365 | h->transid, num_bytes, 1); |
b5009945 JB |
366 | h->block_rsv = &root->fs_info->trans_block_rsv; |
367 | h->bytes_reserved = num_bytes; | |
a22285a6 | 368 | } |
9ed74f2d | 369 | |
2a1eb461 | 370 | got_it: |
a4abeea4 | 371 | btrfs_record_root_in_trans(h, root); |
a22285a6 YZ |
372 | |
373 | if (!current->journal_info && type != TRANS_USERSPACE) | |
374 | current->journal_info = h; | |
79154b1b CM |
375 | return h; |
376 | } | |
377 | ||
f9295749 | 378 | struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, |
a22285a6 | 379 | int num_items) |
f9295749 | 380 | { |
a22285a6 | 381 | return start_transaction(root, num_items, TRANS_START); |
f9295749 | 382 | } |
7a7eaa40 | 383 | struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root) |
f9295749 | 384 | { |
a22285a6 | 385 | return start_transaction(root, 0, TRANS_JOIN); |
f9295749 CM |
386 | } |
387 | ||
7a7eaa40 | 388 | struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root) |
0af3d00b JB |
389 | { |
390 | return start_transaction(root, 0, TRANS_JOIN_NOLOCK); | |
391 | } | |
392 | ||
7a7eaa40 | 393 | struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root) |
9ca9ee09 | 394 | { |
7a7eaa40 | 395 | return start_transaction(root, 0, TRANS_USERSPACE); |
9ca9ee09 SW |
396 | } |
397 | ||
d352ac68 | 398 | /* wait for a transaction commit to be fully complete */ |
b9c8300c | 399 | static noinline void wait_for_commit(struct btrfs_root *root, |
89ce8a63 CM |
400 | struct btrfs_transaction *commit) |
401 | { | |
72d63ed6 | 402 | wait_event(commit->commit_wait, commit->commit_done); |
89ce8a63 CM |
403 | } |
404 | ||
46204592 SW |
405 | int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid) |
406 | { | |
407 | struct btrfs_transaction *cur_trans = NULL, *t; | |
408 | int ret; | |
409 | ||
46204592 SW |
410 | ret = 0; |
411 | if (transid) { | |
412 | if (transid <= root->fs_info->last_trans_committed) | |
a4abeea4 | 413 | goto out; |
46204592 SW |
414 | |
415 | /* find specified transaction */ | |
a4abeea4 | 416 | spin_lock(&root->fs_info->trans_lock); |
46204592 SW |
417 | list_for_each_entry(t, &root->fs_info->trans_list, list) { |
418 | if (t->transid == transid) { | |
419 | cur_trans = t; | |
a4abeea4 | 420 | atomic_inc(&cur_trans->use_count); |
46204592 SW |
421 | break; |
422 | } | |
423 | if (t->transid > transid) | |
424 | break; | |
425 | } | |
a4abeea4 | 426 | spin_unlock(&root->fs_info->trans_lock); |
46204592 SW |
427 | ret = -EINVAL; |
428 | if (!cur_trans) | |
a4abeea4 | 429 | goto out; /* bad transid */ |
46204592 SW |
430 | } else { |
431 | /* find newest transaction that is committing | committed */ | |
a4abeea4 | 432 | spin_lock(&root->fs_info->trans_lock); |
46204592 SW |
433 | list_for_each_entry_reverse(t, &root->fs_info->trans_list, |
434 | list) { | |
435 | if (t->in_commit) { | |
436 | if (t->commit_done) | |
3473f3c0 | 437 | break; |
46204592 | 438 | cur_trans = t; |
a4abeea4 | 439 | atomic_inc(&cur_trans->use_count); |
46204592 SW |
440 | break; |
441 | } | |
442 | } | |
a4abeea4 | 443 | spin_unlock(&root->fs_info->trans_lock); |
46204592 | 444 | if (!cur_trans) |
a4abeea4 | 445 | goto out; /* nothing committing|committed */ |
46204592 SW |
446 | } |
447 | ||
46204592 SW |
448 | wait_for_commit(root, cur_trans); |
449 | ||
46204592 SW |
450 | put_transaction(cur_trans); |
451 | ret = 0; | |
a4abeea4 | 452 | out: |
46204592 SW |
453 | return ret; |
454 | } | |
455 | ||
37d1aeee CM |
456 | void btrfs_throttle(struct btrfs_root *root) |
457 | { | |
a4abeea4 | 458 | if (!atomic_read(&root->fs_info->open_ioctl_trans)) |
9ca9ee09 | 459 | wait_current_trans(root); |
37d1aeee CM |
460 | } |
461 | ||
8929ecfa YZ |
462 | static int should_end_transaction(struct btrfs_trans_handle *trans, |
463 | struct btrfs_root *root) | |
464 | { | |
465 | int ret; | |
36ba022a JB |
466 | |
467 | ret = btrfs_block_rsv_check(root, &root->fs_info->global_block_rsv, 5); | |
8929ecfa YZ |
468 | return ret ? 1 : 0; |
469 | } | |
470 | ||
471 | int btrfs_should_end_transaction(struct btrfs_trans_handle *trans, | |
472 | struct btrfs_root *root) | |
473 | { | |
474 | struct btrfs_transaction *cur_trans = trans->transaction; | |
9c8d86db | 475 | struct btrfs_block_rsv *rsv = trans->block_rsv; |
8929ecfa | 476 | int updates; |
49b25e05 | 477 | int err; |
8929ecfa | 478 | |
a4abeea4 | 479 | smp_mb(); |
8929ecfa YZ |
480 | if (cur_trans->blocked || cur_trans->delayed_refs.flushing) |
481 | return 1; | |
482 | ||
9c8d86db JB |
483 | /* |
484 | * We need to do this in case we're deleting csums so the global block | |
485 | * rsv get's used instead of the csum block rsv. | |
486 | */ | |
487 | trans->block_rsv = NULL; | |
488 | ||
8929ecfa YZ |
489 | updates = trans->delayed_ref_updates; |
490 | trans->delayed_ref_updates = 0; | |
49b25e05 JM |
491 | if (updates) { |
492 | err = btrfs_run_delayed_refs(trans, root, updates); | |
493 | if (err) /* Error code will also eval true */ | |
494 | return err; | |
495 | } | |
8929ecfa | 496 | |
9c8d86db JB |
497 | trans->block_rsv = rsv; |
498 | ||
8929ecfa YZ |
499 | return should_end_transaction(trans, root); |
500 | } | |
501 | ||
89ce8a63 | 502 | static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, |
0af3d00b | 503 | struct btrfs_root *root, int throttle, int lock) |
79154b1b | 504 | { |
8929ecfa | 505 | struct btrfs_transaction *cur_trans = trans->transaction; |
ab78c84d | 506 | struct btrfs_fs_info *info = root->fs_info; |
c3e69d58 | 507 | int count = 0; |
4edc2ca3 | 508 | int err = 0; |
c3e69d58 | 509 | |
2a1eb461 JB |
510 | if (--trans->use_count) { |
511 | trans->block_rsv = trans->orig_rsv; | |
512 | return 0; | |
513 | } | |
514 | ||
edf39272 JS |
515 | /* |
516 | * do the qgroup accounting as early as possible | |
517 | */ | |
518 | err = btrfs_delayed_refs_qgroup_accounting(trans, info); | |
519 | ||
b24e03db | 520 | btrfs_trans_release_metadata(trans, root); |
4c13d758 | 521 | trans->block_rsv = NULL; |
d13603ef AJ |
522 | /* |
523 | * the same root has to be passed to start_transaction and | |
524 | * end_transaction. Subvolume quota depends on this. | |
525 | */ | |
526 | WARN_ON(trans->root != root); | |
203bf287 | 527 | while (count < 2) { |
c3e69d58 CM |
528 | unsigned long cur = trans->delayed_ref_updates; |
529 | trans->delayed_ref_updates = 0; | |
530 | if (cur && | |
531 | trans->transaction->delayed_refs.num_heads_ready > 64) { | |
532 | trans->delayed_ref_updates = 0; | |
533 | btrfs_run_delayed_refs(trans, root, cur); | |
534 | } else { | |
535 | break; | |
536 | } | |
537 | count++; | |
56bec294 CM |
538 | } |
539 | ||
a4abeea4 JB |
540 | if (lock && !atomic_read(&root->fs_info->open_ioctl_trans) && |
541 | should_end_transaction(trans, root)) { | |
8929ecfa | 542 | trans->transaction->blocked = 1; |
a4abeea4 JB |
543 | smp_wmb(); |
544 | } | |
8929ecfa | 545 | |
0af3d00b | 546 | if (lock && cur_trans->blocked && !cur_trans->in_commit) { |
81317fde JB |
547 | if (throttle) { |
548 | /* | |
549 | * We may race with somebody else here so end up having | |
550 | * to call end_transaction on ourselves again, so inc | |
551 | * our use_count. | |
552 | */ | |
553 | trans->use_count++; | |
8929ecfa | 554 | return btrfs_commit_transaction(trans, root); |
81317fde | 555 | } else { |
8929ecfa | 556 | wake_up_process(info->transaction_kthread); |
81317fde | 557 | } |
8929ecfa YZ |
558 | } |
559 | ||
8929ecfa | 560 | WARN_ON(cur_trans != info->running_transaction); |
13c5a93e JB |
561 | WARN_ON(atomic_read(&cur_trans->num_writers) < 1); |
562 | atomic_dec(&cur_trans->num_writers); | |
89ce8a63 | 563 | |
99d16cbc | 564 | smp_mb(); |
79154b1b CM |
565 | if (waitqueue_active(&cur_trans->writer_wait)) |
566 | wake_up(&cur_trans->writer_wait); | |
79154b1b | 567 | put_transaction(cur_trans); |
9ed74f2d JB |
568 | |
569 | if (current->journal_info == trans) | |
570 | current->journal_info = NULL; | |
ab78c84d | 571 | |
24bbcf04 YZ |
572 | if (throttle) |
573 | btrfs_run_delayed_iputs(root); | |
574 | ||
49b25e05 JM |
575 | if (trans->aborted || |
576 | root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) { | |
4edc2ca3 | 577 | err = -EIO; |
49b25e05 | 578 | } |
edf39272 | 579 | assert_qgroups_uptodate(trans); |
49b25e05 | 580 | |
4edc2ca3 DJ |
581 | memset(trans, 0, sizeof(*trans)); |
582 | kmem_cache_free(btrfs_trans_handle_cachep, trans); | |
583 | return err; | |
79154b1b CM |
584 | } |
585 | ||
89ce8a63 CM |
586 | int btrfs_end_transaction(struct btrfs_trans_handle *trans, |
587 | struct btrfs_root *root) | |
588 | { | |
16cdcec7 MX |
589 | int ret; |
590 | ||
591 | ret = __btrfs_end_transaction(trans, root, 0, 1); | |
592 | if (ret) | |
593 | return ret; | |
594 | return 0; | |
89ce8a63 CM |
595 | } |
596 | ||
597 | int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans, | |
598 | struct btrfs_root *root) | |
599 | { | |
16cdcec7 MX |
600 | int ret; |
601 | ||
602 | ret = __btrfs_end_transaction(trans, root, 1, 1); | |
603 | if (ret) | |
604 | return ret; | |
605 | return 0; | |
0af3d00b JB |
606 | } |
607 | ||
608 | int btrfs_end_transaction_nolock(struct btrfs_trans_handle *trans, | |
609 | struct btrfs_root *root) | |
610 | { | |
16cdcec7 MX |
611 | int ret; |
612 | ||
613 | ret = __btrfs_end_transaction(trans, root, 0, 0); | |
614 | if (ret) | |
615 | return ret; | |
616 | return 0; | |
617 | } | |
618 | ||
619 | int btrfs_end_transaction_dmeta(struct btrfs_trans_handle *trans, | |
620 | struct btrfs_root *root) | |
621 | { | |
622 | return __btrfs_end_transaction(trans, root, 1, 1); | |
89ce8a63 CM |
623 | } |
624 | ||
d352ac68 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 | |
690587d1 | 628 | * those extents are sent to disk but does not wait on them |
d352ac68 | 629 | */ |
690587d1 | 630 | int btrfs_write_marked_extents(struct btrfs_root *root, |
8cef4e16 | 631 | struct extent_io_tree *dirty_pages, int mark) |
79154b1b | 632 | { |
777e6bd7 | 633 | int err = 0; |
7c4452b9 | 634 | int werr = 0; |
1728366e | 635 | struct address_space *mapping = root->fs_info->btree_inode->i_mapping; |
777e6bd7 | 636 | u64 start = 0; |
5f39d397 | 637 | u64 end; |
7c4452b9 | 638 | |
1728366e JB |
639 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
640 | mark)) { | |
641 | convert_extent_bit(dirty_pages, start, end, EXTENT_NEED_WAIT, mark, | |
642 | GFP_NOFS); | |
643 | err = filemap_fdatawrite_range(mapping, start, end); | |
644 | if (err) | |
645 | werr = err; | |
646 | cond_resched(); | |
647 | start = end + 1; | |
7c4452b9 | 648 | } |
690587d1 CM |
649 | if (err) |
650 | werr = err; | |
651 | return werr; | |
652 | } | |
653 | ||
654 | /* | |
655 | * when btree blocks are allocated, they have some corresponding bits set for | |
656 | * them in one of two extent_io trees. This is used to make sure all of | |
657 | * those extents are on disk for transaction or log commit. We wait | |
658 | * on all the pages and clear them from the dirty pages state tree | |
659 | */ | |
660 | int btrfs_wait_marked_extents(struct btrfs_root *root, | |
8cef4e16 | 661 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 | 662 | { |
690587d1 CM |
663 | int err = 0; |
664 | int werr = 0; | |
1728366e | 665 | struct address_space *mapping = root->fs_info->btree_inode->i_mapping; |
690587d1 CM |
666 | u64 start = 0; |
667 | u64 end; | |
777e6bd7 | 668 | |
1728366e JB |
669 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
670 | EXTENT_NEED_WAIT)) { | |
671 | clear_extent_bits(dirty_pages, start, end, EXTENT_NEED_WAIT, GFP_NOFS); | |
672 | err = filemap_fdatawait_range(mapping, start, end); | |
673 | if (err) | |
674 | werr = err; | |
675 | cond_resched(); | |
676 | start = end + 1; | |
777e6bd7 | 677 | } |
7c4452b9 CM |
678 | if (err) |
679 | werr = err; | |
680 | return werr; | |
79154b1b CM |
681 | } |
682 | ||
690587d1 CM |
683 | /* |
684 | * when btree blocks are allocated, they have some corresponding bits set for | |
685 | * them in one of two extent_io trees. This is used to make sure all of | |
686 | * those extents are on disk for transaction or log commit | |
687 | */ | |
688 | int btrfs_write_and_wait_marked_extents(struct btrfs_root *root, | |
8cef4e16 | 689 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 CM |
690 | { |
691 | int ret; | |
692 | int ret2; | |
693 | ||
8cef4e16 YZ |
694 | ret = btrfs_write_marked_extents(root, dirty_pages, mark); |
695 | ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark); | |
bf0da8c1 CM |
696 | |
697 | if (ret) | |
698 | return ret; | |
699 | if (ret2) | |
700 | return ret2; | |
701 | return 0; | |
690587d1 CM |
702 | } |
703 | ||
d0c803c4 CM |
704 | int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans, |
705 | struct btrfs_root *root) | |
706 | { | |
707 | if (!trans || !trans->transaction) { | |
708 | struct inode *btree_inode; | |
709 | btree_inode = root->fs_info->btree_inode; | |
710 | return filemap_write_and_wait(btree_inode->i_mapping); | |
711 | } | |
712 | return btrfs_write_and_wait_marked_extents(root, | |
8cef4e16 YZ |
713 | &trans->transaction->dirty_pages, |
714 | EXTENT_DIRTY); | |
d0c803c4 CM |
715 | } |
716 | ||
d352ac68 CM |
717 | /* |
718 | * this is used to update the root pointer in the tree of tree roots. | |
719 | * | |
720 | * But, in the case of the extent allocation tree, updating the root | |
721 | * pointer may allocate blocks which may change the root of the extent | |
722 | * allocation tree. | |
723 | * | |
724 | * So, this loops and repeats and makes sure the cowonly root didn't | |
725 | * change while the root pointer was being updated in the metadata. | |
726 | */ | |
0b86a832 CM |
727 | static int update_cowonly_root(struct btrfs_trans_handle *trans, |
728 | struct btrfs_root *root) | |
79154b1b CM |
729 | { |
730 | int ret; | |
0b86a832 | 731 | u64 old_root_bytenr; |
86b9f2ec | 732 | u64 old_root_used; |
0b86a832 | 733 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
79154b1b | 734 | |
86b9f2ec | 735 | old_root_used = btrfs_root_used(&root->root_item); |
0b86a832 | 736 | btrfs_write_dirty_block_groups(trans, root); |
56bec294 | 737 | |
d397712b | 738 | while (1) { |
0b86a832 | 739 | old_root_bytenr = btrfs_root_bytenr(&root->root_item); |
86b9f2ec YZ |
740 | if (old_root_bytenr == root->node->start && |
741 | old_root_used == btrfs_root_used(&root->root_item)) | |
79154b1b | 742 | break; |
87ef2bb4 | 743 | |
5d4f98a2 | 744 | btrfs_set_root_node(&root->root_item, root->node); |
79154b1b | 745 | ret = btrfs_update_root(trans, tree_root, |
0b86a832 CM |
746 | &root->root_key, |
747 | &root->root_item); | |
49b25e05 JM |
748 | if (ret) |
749 | return ret; | |
56bec294 | 750 | |
86b9f2ec | 751 | old_root_used = btrfs_root_used(&root->root_item); |
4a8c9a62 | 752 | ret = btrfs_write_dirty_block_groups(trans, root); |
49b25e05 JM |
753 | if (ret) |
754 | return ret; | |
0b86a832 | 755 | } |
276e680d YZ |
756 | |
757 | if (root != root->fs_info->extent_root) | |
758 | switch_commit_root(root); | |
759 | ||
0b86a832 CM |
760 | return 0; |
761 | } | |
762 | ||
d352ac68 CM |
763 | /* |
764 | * update all the cowonly tree roots on disk | |
49b25e05 JM |
765 | * |
766 | * The error handling in this function may not be obvious. Any of the | |
767 | * failures will cause the file system to go offline. We still need | |
768 | * to clean up the delayed refs. | |
d352ac68 | 769 | */ |
5d4f98a2 YZ |
770 | static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans, |
771 | struct btrfs_root *root) | |
0b86a832 CM |
772 | { |
773 | struct btrfs_fs_info *fs_info = root->fs_info; | |
774 | struct list_head *next; | |
84234f3a | 775 | struct extent_buffer *eb; |
56bec294 | 776 | int ret; |
84234f3a | 777 | |
56bec294 | 778 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
49b25e05 JM |
779 | if (ret) |
780 | return ret; | |
87ef2bb4 | 781 | |
84234f3a | 782 | eb = btrfs_lock_root_node(fs_info->tree_root); |
49b25e05 JM |
783 | ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, |
784 | 0, &eb); | |
84234f3a YZ |
785 | btrfs_tree_unlock(eb); |
786 | free_extent_buffer(eb); | |
0b86a832 | 787 | |
49b25e05 JM |
788 | if (ret) |
789 | return ret; | |
790 | ||
56bec294 | 791 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
49b25e05 JM |
792 | if (ret) |
793 | return ret; | |
87ef2bb4 | 794 | |
733f4fbb SB |
795 | ret = btrfs_run_dev_stats(trans, root->fs_info); |
796 | BUG_ON(ret); | |
797 | ||
d397712b | 798 | while (!list_empty(&fs_info->dirty_cowonly_roots)) { |
0b86a832 CM |
799 | next = fs_info->dirty_cowonly_roots.next; |
800 | list_del_init(next); | |
801 | root = list_entry(next, struct btrfs_root, dirty_list); | |
87ef2bb4 | 802 | |
49b25e05 JM |
803 | ret = update_cowonly_root(trans, root); |
804 | if (ret) | |
805 | return ret; | |
79154b1b | 806 | } |
276e680d YZ |
807 | |
808 | down_write(&fs_info->extent_commit_sem); | |
809 | switch_commit_root(fs_info->extent_root); | |
810 | up_write(&fs_info->extent_commit_sem); | |
811 | ||
79154b1b CM |
812 | return 0; |
813 | } | |
814 | ||
d352ac68 CM |
815 | /* |
816 | * dead roots are old snapshots that need to be deleted. This allocates | |
817 | * a dirty root struct and adds it into the list of dead roots that need to | |
818 | * be deleted | |
819 | */ | |
5d4f98a2 | 820 | int btrfs_add_dead_root(struct btrfs_root *root) |
5eda7b5e | 821 | { |
a4abeea4 | 822 | spin_lock(&root->fs_info->trans_lock); |
5d4f98a2 | 823 | list_add(&root->root_list, &root->fs_info->dead_roots); |
a4abeea4 | 824 | spin_unlock(&root->fs_info->trans_lock); |
5eda7b5e CM |
825 | return 0; |
826 | } | |
827 | ||
d352ac68 | 828 | /* |
5d4f98a2 | 829 | * update all the cowonly tree roots on disk |
d352ac68 | 830 | */ |
5d4f98a2 YZ |
831 | static noinline int commit_fs_roots(struct btrfs_trans_handle *trans, |
832 | struct btrfs_root *root) | |
0f7d52f4 | 833 | { |
0f7d52f4 | 834 | struct btrfs_root *gang[8]; |
5d4f98a2 | 835 | struct btrfs_fs_info *fs_info = root->fs_info; |
0f7d52f4 CM |
836 | int i; |
837 | int ret; | |
54aa1f4d CM |
838 | int err = 0; |
839 | ||
a4abeea4 | 840 | spin_lock(&fs_info->fs_roots_radix_lock); |
d397712b | 841 | while (1) { |
5d4f98a2 YZ |
842 | ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, |
843 | (void **)gang, 0, | |
0f7d52f4 CM |
844 | ARRAY_SIZE(gang), |
845 | BTRFS_ROOT_TRANS_TAG); | |
846 | if (ret == 0) | |
847 | break; | |
848 | for (i = 0; i < ret; i++) { | |
849 | root = gang[i]; | |
5d4f98a2 YZ |
850 | radix_tree_tag_clear(&fs_info->fs_roots_radix, |
851 | (unsigned long)root->root_key.objectid, | |
852 | BTRFS_ROOT_TRANS_TAG); | |
a4abeea4 | 853 | spin_unlock(&fs_info->fs_roots_radix_lock); |
31153d81 | 854 | |
e02119d5 | 855 | btrfs_free_log(trans, root); |
5d4f98a2 | 856 | btrfs_update_reloc_root(trans, root); |
d68fc57b | 857 | btrfs_orphan_commit_root(trans, root); |
bcc63abb | 858 | |
82d5902d LZ |
859 | btrfs_save_ino_cache(root, trans); |
860 | ||
f1ebcc74 LB |
861 | /* see comments in should_cow_block() */ |
862 | root->force_cow = 0; | |
863 | smp_wmb(); | |
864 | ||
978d910d | 865 | if (root->commit_root != root->node) { |
581bb050 | 866 | mutex_lock(&root->fs_commit_mutex); |
817d52f8 | 867 | switch_commit_root(root); |
581bb050 LZ |
868 | btrfs_unpin_free_ino(root); |
869 | mutex_unlock(&root->fs_commit_mutex); | |
870 | ||
978d910d YZ |
871 | btrfs_set_root_node(&root->root_item, |
872 | root->node); | |
873 | } | |
5d4f98a2 | 874 | |
5d4f98a2 | 875 | err = btrfs_update_root(trans, fs_info->tree_root, |
0f7d52f4 CM |
876 | &root->root_key, |
877 | &root->root_item); | |
a4abeea4 | 878 | spin_lock(&fs_info->fs_roots_radix_lock); |
54aa1f4d CM |
879 | if (err) |
880 | break; | |
0f7d52f4 CM |
881 | } |
882 | } | |
a4abeea4 | 883 | spin_unlock(&fs_info->fs_roots_radix_lock); |
54aa1f4d | 884 | return err; |
0f7d52f4 CM |
885 | } |
886 | ||
d352ac68 CM |
887 | /* |
888 | * defrag a given btree. If cacheonly == 1, this won't read from the disk, | |
889 | * otherwise every leaf in the btree is read and defragged. | |
890 | */ | |
e9d0b13b CM |
891 | int btrfs_defrag_root(struct btrfs_root *root, int cacheonly) |
892 | { | |
893 | struct btrfs_fs_info *info = root->fs_info; | |
e9d0b13b | 894 | struct btrfs_trans_handle *trans; |
8929ecfa | 895 | int ret; |
d3c2fdcf | 896 | unsigned long nr; |
e9d0b13b | 897 | |
8929ecfa | 898 | if (xchg(&root->defrag_running, 1)) |
e9d0b13b | 899 | return 0; |
8929ecfa | 900 | |
6b80053d | 901 | while (1) { |
8929ecfa YZ |
902 | trans = btrfs_start_transaction(root, 0); |
903 | if (IS_ERR(trans)) | |
904 | return PTR_ERR(trans); | |
905 | ||
e9d0b13b | 906 | ret = btrfs_defrag_leaves(trans, root, cacheonly); |
8929ecfa | 907 | |
d3c2fdcf | 908 | nr = trans->blocks_used; |
e9d0b13b | 909 | btrfs_end_transaction(trans, root); |
d3c2fdcf | 910 | btrfs_btree_balance_dirty(info->tree_root, nr); |
e9d0b13b CM |
911 | cond_resched(); |
912 | ||
7841cb28 | 913 | if (btrfs_fs_closing(root->fs_info) || ret != -EAGAIN) |
e9d0b13b CM |
914 | break; |
915 | } | |
916 | root->defrag_running = 0; | |
8929ecfa | 917 | return ret; |
e9d0b13b CM |
918 | } |
919 | ||
d352ac68 CM |
920 | /* |
921 | * new snapshots need to be created at a very specific time in the | |
922 | * transaction commit. This does the actual creation | |
923 | */ | |
80b6794d | 924 | static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, |
3063d29f CM |
925 | struct btrfs_fs_info *fs_info, |
926 | struct btrfs_pending_snapshot *pending) | |
927 | { | |
928 | struct btrfs_key key; | |
80b6794d | 929 | struct btrfs_root_item *new_root_item; |
3063d29f CM |
930 | struct btrfs_root *tree_root = fs_info->tree_root; |
931 | struct btrfs_root *root = pending->root; | |
6bdb72de | 932 | struct btrfs_root *parent_root; |
98c9942a | 933 | struct btrfs_block_rsv *rsv; |
6bdb72de | 934 | struct inode *parent_inode; |
6a912213 | 935 | struct dentry *parent; |
a22285a6 | 936 | struct dentry *dentry; |
3063d29f | 937 | struct extent_buffer *tmp; |
925baedd | 938 | struct extent_buffer *old; |
3063d29f | 939 | int ret; |
d68fc57b | 940 | u64 to_reserve = 0; |
6bdb72de | 941 | u64 index = 0; |
a22285a6 | 942 | u64 objectid; |
b83cc969 | 943 | u64 root_flags; |
3063d29f | 944 | |
98c9942a LB |
945 | rsv = trans->block_rsv; |
946 | ||
80b6794d CM |
947 | new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS); |
948 | if (!new_root_item) { | |
49b25e05 | 949 | ret = pending->error = -ENOMEM; |
80b6794d CM |
950 | goto fail; |
951 | } | |
a22285a6 | 952 | |
581bb050 | 953 | ret = btrfs_find_free_objectid(tree_root, &objectid); |
a22285a6 YZ |
954 | if (ret) { |
955 | pending->error = ret; | |
3063d29f | 956 | goto fail; |
a22285a6 | 957 | } |
3063d29f | 958 | |
3fd0a558 | 959 | btrfs_reloc_pre_snapshot(trans, pending, &to_reserve); |
d68fc57b YZ |
960 | |
961 | if (to_reserve > 0) { | |
62f30c54 MX |
962 | ret = btrfs_block_rsv_add_noflush(root, &pending->block_rsv, |
963 | to_reserve); | |
d68fc57b YZ |
964 | if (ret) { |
965 | pending->error = ret; | |
966 | goto fail; | |
967 | } | |
968 | } | |
969 | ||
3063d29f | 970 | key.objectid = objectid; |
a22285a6 YZ |
971 | key.offset = (u64)-1; |
972 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3063d29f | 973 | |
a22285a6 | 974 | trans->block_rsv = &pending->block_rsv; |
3de4586c | 975 | |
a22285a6 | 976 | dentry = pending->dentry; |
6a912213 JB |
977 | parent = dget_parent(dentry); |
978 | parent_inode = parent->d_inode; | |
a22285a6 | 979 | parent_root = BTRFS_I(parent_inode)->root; |
7585717f | 980 | record_root_in_trans(trans, parent_root); |
a22285a6 | 981 | |
3063d29f CM |
982 | /* |
983 | * insert the directory item | |
984 | */ | |
3de4586c | 985 | ret = btrfs_set_inode_index(parent_inode, &index); |
49b25e05 | 986 | BUG_ON(ret); /* -ENOMEM */ |
0660b5af | 987 | ret = btrfs_insert_dir_item(trans, parent_root, |
a22285a6 | 988 | dentry->d_name.name, dentry->d_name.len, |
16cdcec7 | 989 | parent_inode, &key, |
a22285a6 | 990 | BTRFS_FT_DIR, index); |
79787eaa | 991 | if (ret == -EEXIST) { |
fe66a05a CM |
992 | pending->error = -EEXIST; |
993 | dput(parent); | |
994 | goto fail; | |
79787eaa JM |
995 | } else if (ret) { |
996 | goto abort_trans_dput; | |
997 | } | |
0660b5af | 998 | |
a22285a6 YZ |
999 | btrfs_i_size_write(parent_inode, parent_inode->i_size + |
1000 | dentry->d_name.len * 2); | |
52c26179 | 1001 | ret = btrfs_update_inode(trans, parent_root, parent_inode); |
49b25e05 | 1002 | if (ret) |
79787eaa | 1003 | goto abort_trans_dput; |
52c26179 | 1004 | |
e999376f CM |
1005 | /* |
1006 | * pull in the delayed directory update | |
1007 | * and the delayed inode item | |
1008 | * otherwise we corrupt the FS during | |
1009 | * snapshot | |
1010 | */ | |
1011 | ret = btrfs_run_delayed_items(trans, root); | |
79787eaa JM |
1012 | if (ret) { /* Transaction aborted */ |
1013 | dput(parent); | |
49b25e05 | 1014 | goto fail; |
79787eaa | 1015 | } |
e999376f | 1016 | |
7585717f | 1017 | record_root_in_trans(trans, root); |
6bdb72de SW |
1018 | btrfs_set_root_last_snapshot(&root->root_item, trans->transid); |
1019 | memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); | |
08fe4db1 | 1020 | btrfs_check_and_init_root_item(new_root_item); |
6bdb72de | 1021 | |
b83cc969 LZ |
1022 | root_flags = btrfs_root_flags(new_root_item); |
1023 | if (pending->readonly) | |
1024 | root_flags |= BTRFS_ROOT_SUBVOL_RDONLY; | |
1025 | else | |
1026 | root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY; | |
1027 | btrfs_set_root_flags(new_root_item, root_flags); | |
1028 | ||
6bdb72de | 1029 | old = btrfs_lock_root_node(root); |
49b25e05 | 1030 | ret = btrfs_cow_block(trans, root, old, NULL, 0, &old); |
79787eaa JM |
1031 | if (ret) { |
1032 | btrfs_tree_unlock(old); | |
1033 | free_extent_buffer(old); | |
1034 | goto abort_trans_dput; | |
1035 | } | |
49b25e05 | 1036 | |
6bdb72de SW |
1037 | btrfs_set_lock_blocking(old); |
1038 | ||
49b25e05 | 1039 | ret = btrfs_copy_root(trans, root, old, &tmp, objectid); |
79787eaa | 1040 | /* clean up in any case */ |
6bdb72de SW |
1041 | btrfs_tree_unlock(old); |
1042 | free_extent_buffer(old); | |
79787eaa JM |
1043 | if (ret) |
1044 | goto abort_trans_dput; | |
6bdb72de | 1045 | |
f1ebcc74 LB |
1046 | /* see comments in should_cow_block() */ |
1047 | root->force_cow = 1; | |
1048 | smp_wmb(); | |
1049 | ||
6bdb72de | 1050 | btrfs_set_root_node(new_root_item, tmp); |
a22285a6 YZ |
1051 | /* record when the snapshot was created in key.offset */ |
1052 | key.offset = trans->transid; | |
1053 | ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); | |
6bdb72de SW |
1054 | btrfs_tree_unlock(tmp); |
1055 | free_extent_buffer(tmp); | |
49b25e05 | 1056 | if (ret) |
79787eaa | 1057 | goto abort_trans_dput; |
6bdb72de | 1058 | |
a22285a6 YZ |
1059 | /* |
1060 | * insert root back/forward references | |
1061 | */ | |
1062 | ret = btrfs_add_root_ref(trans, tree_root, objectid, | |
0660b5af | 1063 | parent_root->root_key.objectid, |
33345d01 | 1064 | btrfs_ino(parent_inode), index, |
a22285a6 | 1065 | dentry->d_name.name, dentry->d_name.len); |
79787eaa | 1066 | dput(parent); |
49b25e05 JM |
1067 | if (ret) |
1068 | goto fail; | |
0660b5af | 1069 | |
a22285a6 YZ |
1070 | key.offset = (u64)-1; |
1071 | pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key); | |
79787eaa JM |
1072 | if (IS_ERR(pending->snap)) { |
1073 | ret = PTR_ERR(pending->snap); | |
49b25e05 | 1074 | goto abort_trans; |
79787eaa | 1075 | } |
d68fc57b | 1076 | |
49b25e05 JM |
1077 | ret = btrfs_reloc_post_snapshot(trans, pending); |
1078 | if (ret) | |
1079 | goto abort_trans; | |
1080 | ret = 0; | |
3063d29f | 1081 | fail: |
6bdb72de | 1082 | kfree(new_root_item); |
98c9942a | 1083 | trans->block_rsv = rsv; |
a22285a6 | 1084 | btrfs_block_rsv_release(root, &pending->block_rsv, (u64)-1); |
49b25e05 JM |
1085 | return ret; |
1086 | ||
79787eaa JM |
1087 | abort_trans_dput: |
1088 | dput(parent); | |
49b25e05 JM |
1089 | abort_trans: |
1090 | btrfs_abort_transaction(trans, root, ret); | |
1091 | goto fail; | |
3063d29f CM |
1092 | } |
1093 | ||
d352ac68 CM |
1094 | /* |
1095 | * create all the snapshots we've scheduled for creation | |
1096 | */ | |
80b6794d CM |
1097 | static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans, |
1098 | struct btrfs_fs_info *fs_info) | |
3de4586c CM |
1099 | { |
1100 | struct btrfs_pending_snapshot *pending; | |
1101 | struct list_head *head = &trans->transaction->pending_snapshots; | |
3de4586c | 1102 | |
fe66a05a CM |
1103 | list_for_each_entry(pending, head, list) |
1104 | create_pending_snapshot(trans, fs_info, pending); | |
3de4586c CM |
1105 | return 0; |
1106 | } | |
1107 | ||
5d4f98a2 YZ |
1108 | static void update_super_roots(struct btrfs_root *root) |
1109 | { | |
1110 | struct btrfs_root_item *root_item; | |
1111 | struct btrfs_super_block *super; | |
1112 | ||
6c41761f | 1113 | super = root->fs_info->super_copy; |
5d4f98a2 YZ |
1114 | |
1115 | root_item = &root->fs_info->chunk_root->root_item; | |
1116 | super->chunk_root = root_item->bytenr; | |
1117 | super->chunk_root_generation = root_item->generation; | |
1118 | super->chunk_root_level = root_item->level; | |
1119 | ||
1120 | root_item = &root->fs_info->tree_root->root_item; | |
1121 | super->root = root_item->bytenr; | |
1122 | super->generation = root_item->generation; | |
1123 | super->root_level = root_item->level; | |
73bc1876 | 1124 | if (btrfs_test_opt(root, SPACE_CACHE)) |
0af3d00b | 1125 | super->cache_generation = root_item->generation; |
5d4f98a2 YZ |
1126 | } |
1127 | ||
f36f3042 CM |
1128 | int btrfs_transaction_in_commit(struct btrfs_fs_info *info) |
1129 | { | |
1130 | int ret = 0; | |
a4abeea4 | 1131 | spin_lock(&info->trans_lock); |
f36f3042 CM |
1132 | if (info->running_transaction) |
1133 | ret = info->running_transaction->in_commit; | |
a4abeea4 | 1134 | spin_unlock(&info->trans_lock); |
f36f3042 CM |
1135 | return ret; |
1136 | } | |
1137 | ||
8929ecfa YZ |
1138 | int btrfs_transaction_blocked(struct btrfs_fs_info *info) |
1139 | { | |
1140 | int ret = 0; | |
a4abeea4 | 1141 | spin_lock(&info->trans_lock); |
8929ecfa YZ |
1142 | if (info->running_transaction) |
1143 | ret = info->running_transaction->blocked; | |
a4abeea4 | 1144 | spin_unlock(&info->trans_lock); |
8929ecfa YZ |
1145 | return ret; |
1146 | } | |
1147 | ||
bb9c12c9 SW |
1148 | /* |
1149 | * wait for the current transaction commit to start and block subsequent | |
1150 | * transaction joins | |
1151 | */ | |
1152 | static void wait_current_trans_commit_start(struct btrfs_root *root, | |
1153 | struct btrfs_transaction *trans) | |
1154 | { | |
72d63ed6 | 1155 | wait_event(root->fs_info->transaction_blocked_wait, trans->in_commit); |
bb9c12c9 SW |
1156 | } |
1157 | ||
1158 | /* | |
1159 | * wait for the current transaction to start and then become unblocked. | |
1160 | * caller holds ref. | |
1161 | */ | |
1162 | static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root, | |
1163 | struct btrfs_transaction *trans) | |
1164 | { | |
72d63ed6 LZ |
1165 | wait_event(root->fs_info->transaction_wait, |
1166 | trans->commit_done || (trans->in_commit && !trans->blocked)); | |
bb9c12c9 SW |
1167 | } |
1168 | ||
1169 | /* | |
1170 | * commit transactions asynchronously. once btrfs_commit_transaction_async | |
1171 | * returns, any subsequent transaction will not be allowed to join. | |
1172 | */ | |
1173 | struct btrfs_async_commit { | |
1174 | struct btrfs_trans_handle *newtrans; | |
1175 | struct btrfs_root *root; | |
1176 | struct delayed_work work; | |
1177 | }; | |
1178 | ||
1179 | static void do_async_commit(struct work_struct *work) | |
1180 | { | |
1181 | struct btrfs_async_commit *ac = | |
1182 | container_of(work, struct btrfs_async_commit, work.work); | |
1183 | ||
1184 | btrfs_commit_transaction(ac->newtrans, ac->root); | |
1185 | kfree(ac); | |
1186 | } | |
1187 | ||
1188 | int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans, | |
1189 | struct btrfs_root *root, | |
1190 | int wait_for_unblock) | |
1191 | { | |
1192 | struct btrfs_async_commit *ac; | |
1193 | struct btrfs_transaction *cur_trans; | |
1194 | ||
1195 | ac = kmalloc(sizeof(*ac), GFP_NOFS); | |
db5b493a TI |
1196 | if (!ac) |
1197 | return -ENOMEM; | |
bb9c12c9 SW |
1198 | |
1199 | INIT_DELAYED_WORK(&ac->work, do_async_commit); | |
1200 | ac->root = root; | |
7a7eaa40 | 1201 | ac->newtrans = btrfs_join_transaction(root); |
3612b495 TI |
1202 | if (IS_ERR(ac->newtrans)) { |
1203 | int err = PTR_ERR(ac->newtrans); | |
1204 | kfree(ac); | |
1205 | return err; | |
1206 | } | |
bb9c12c9 SW |
1207 | |
1208 | /* take transaction reference */ | |
bb9c12c9 | 1209 | cur_trans = trans->transaction; |
13c5a93e | 1210 | atomic_inc(&cur_trans->use_count); |
bb9c12c9 SW |
1211 | |
1212 | btrfs_end_transaction(trans, root); | |
1213 | schedule_delayed_work(&ac->work, 0); | |
1214 | ||
1215 | /* wait for transaction to start and unblock */ | |
bb9c12c9 SW |
1216 | if (wait_for_unblock) |
1217 | wait_current_trans_commit_start_and_unblock(root, cur_trans); | |
1218 | else | |
1219 | wait_current_trans_commit_start(root, cur_trans); | |
bb9c12c9 | 1220 | |
38e88054 SW |
1221 | if (current->journal_info == trans) |
1222 | current->journal_info = NULL; | |
1223 | ||
1224 | put_transaction(cur_trans); | |
bb9c12c9 SW |
1225 | return 0; |
1226 | } | |
1227 | ||
49b25e05 JM |
1228 | |
1229 | static void cleanup_transaction(struct btrfs_trans_handle *trans, | |
7b8b92af | 1230 | struct btrfs_root *root, int err) |
49b25e05 JM |
1231 | { |
1232 | struct btrfs_transaction *cur_trans = trans->transaction; | |
1233 | ||
1234 | WARN_ON(trans->use_count > 1); | |
1235 | ||
7b8b92af JB |
1236 | btrfs_abort_transaction(trans, root, err); |
1237 | ||
49b25e05 JM |
1238 | spin_lock(&root->fs_info->trans_lock); |
1239 | list_del_init(&cur_trans->list); | |
d7096fc3 JB |
1240 | if (cur_trans == root->fs_info->running_transaction) { |
1241 | root->fs_info->running_transaction = NULL; | |
1242 | root->fs_info->trans_no_join = 0; | |
1243 | } | |
49b25e05 JM |
1244 | spin_unlock(&root->fs_info->trans_lock); |
1245 | ||
1246 | btrfs_cleanup_one_transaction(trans->transaction, root); | |
1247 | ||
1248 | put_transaction(cur_trans); | |
1249 | put_transaction(cur_trans); | |
1250 | ||
1251 | trace_btrfs_transaction_commit(root); | |
1252 | ||
1253 | btrfs_scrub_continue(root); | |
1254 | ||
1255 | if (current->journal_info == trans) | |
1256 | current->journal_info = NULL; | |
1257 | ||
1258 | kmem_cache_free(btrfs_trans_handle_cachep, trans); | |
1259 | } | |
1260 | ||
bb9c12c9 SW |
1261 | /* |
1262 | * btrfs_transaction state sequence: | |
1263 | * in_commit = 0, blocked = 0 (initial) | |
1264 | * in_commit = 1, blocked = 1 | |
1265 | * blocked = 0 | |
1266 | * commit_done = 1 | |
1267 | */ | |
79154b1b CM |
1268 | int btrfs_commit_transaction(struct btrfs_trans_handle *trans, |
1269 | struct btrfs_root *root) | |
1270 | { | |
15ee9bc7 | 1271 | unsigned long joined = 0; |
49b25e05 | 1272 | struct btrfs_transaction *cur_trans = trans->transaction; |
8fd17795 | 1273 | struct btrfs_transaction *prev_trans = NULL; |
79154b1b | 1274 | DEFINE_WAIT(wait); |
49b25e05 | 1275 | int ret = -EIO; |
89573b9c CM |
1276 | int should_grow = 0; |
1277 | unsigned long now = get_seconds(); | |
dccae999 | 1278 | int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT); |
79154b1b | 1279 | |
5a3f23d5 CM |
1280 | btrfs_run_ordered_operations(root, 0); |
1281 | ||
b24e03db | 1282 | btrfs_trans_release_metadata(trans, root); |
9c8d86db JB |
1283 | trans->block_rsv = NULL; |
1284 | ||
49b25e05 JM |
1285 | if (cur_trans->aborted) |
1286 | goto cleanup_transaction; | |
1287 | ||
56bec294 CM |
1288 | /* make a pass through all the delayed refs we have so far |
1289 | * any runnings procs may add more while we are here | |
1290 | */ | |
1291 | ret = btrfs_run_delayed_refs(trans, root, 0); | |
49b25e05 JM |
1292 | if (ret) |
1293 | goto cleanup_transaction; | |
56bec294 | 1294 | |
b7ec40d7 | 1295 | cur_trans = trans->transaction; |
49b25e05 | 1296 | |
56bec294 CM |
1297 | /* |
1298 | * set the flushing flag so procs in this transaction have to | |
1299 | * start sending their work down. | |
1300 | */ | |
b7ec40d7 | 1301 | cur_trans->delayed_refs.flushing = 1; |
56bec294 | 1302 | |
c3e69d58 | 1303 | ret = btrfs_run_delayed_refs(trans, root, 0); |
49b25e05 JM |
1304 | if (ret) |
1305 | goto cleanup_transaction; | |
56bec294 | 1306 | |
a4abeea4 | 1307 | spin_lock(&cur_trans->commit_lock); |
b7ec40d7 | 1308 | if (cur_trans->in_commit) { |
a4abeea4 | 1309 | spin_unlock(&cur_trans->commit_lock); |
13c5a93e | 1310 | atomic_inc(&cur_trans->use_count); |
49b25e05 | 1311 | ret = btrfs_end_transaction(trans, root); |
ccd467d6 | 1312 | |
b9c8300c | 1313 | wait_for_commit(root, cur_trans); |
15ee9bc7 | 1314 | |
79154b1b | 1315 | put_transaction(cur_trans); |
15ee9bc7 | 1316 | |
49b25e05 | 1317 | return ret; |
79154b1b | 1318 | } |
4313b399 | 1319 | |
2c90e5d6 | 1320 | trans->transaction->in_commit = 1; |
f9295749 | 1321 | trans->transaction->blocked = 1; |
a4abeea4 | 1322 | spin_unlock(&cur_trans->commit_lock); |
bb9c12c9 SW |
1323 | wake_up(&root->fs_info->transaction_blocked_wait); |
1324 | ||
a4abeea4 | 1325 | spin_lock(&root->fs_info->trans_lock); |
ccd467d6 CM |
1326 | if (cur_trans->list.prev != &root->fs_info->trans_list) { |
1327 | prev_trans = list_entry(cur_trans->list.prev, | |
1328 | struct btrfs_transaction, list); | |
1329 | if (!prev_trans->commit_done) { | |
13c5a93e | 1330 | atomic_inc(&prev_trans->use_count); |
a4abeea4 | 1331 | spin_unlock(&root->fs_info->trans_lock); |
ccd467d6 CM |
1332 | |
1333 | wait_for_commit(root, prev_trans); | |
ccd467d6 | 1334 | |
15ee9bc7 | 1335 | put_transaction(prev_trans); |
a4abeea4 JB |
1336 | } else { |
1337 | spin_unlock(&root->fs_info->trans_lock); | |
ccd467d6 | 1338 | } |
a4abeea4 JB |
1339 | } else { |
1340 | spin_unlock(&root->fs_info->trans_lock); | |
ccd467d6 | 1341 | } |
15ee9bc7 | 1342 | |
89573b9c CM |
1343 | if (now < cur_trans->start_time || now - cur_trans->start_time < 1) |
1344 | should_grow = 1; | |
1345 | ||
15ee9bc7 | 1346 | do { |
7ea394f1 | 1347 | int snap_pending = 0; |
a4abeea4 | 1348 | |
15ee9bc7 | 1349 | joined = cur_trans->num_joined; |
7ea394f1 YZ |
1350 | if (!list_empty(&trans->transaction->pending_snapshots)) |
1351 | snap_pending = 1; | |
1352 | ||
2c90e5d6 | 1353 | WARN_ON(cur_trans != trans->transaction); |
15ee9bc7 | 1354 | |
0bdb1db2 | 1355 | if (flush_on_commit || snap_pending) { |
24bbcf04 | 1356 | btrfs_start_delalloc_inodes(root, 1); |
143bede5 | 1357 | btrfs_wait_ordered_extents(root, 0, 1); |
7ea394f1 YZ |
1358 | } |
1359 | ||
16cdcec7 | 1360 | ret = btrfs_run_delayed_items(trans, root); |
49b25e05 JM |
1361 | if (ret) |
1362 | goto cleanup_transaction; | |
16cdcec7 | 1363 | |
edf39272 JS |
1364 | /* |
1365 | * running the delayed items may have added new refs. account | |
1366 | * them now so that they hinder processing of more delayed refs | |
1367 | * as little as possible. | |
1368 | */ | |
1369 | btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info); | |
1370 | ||
5a3f23d5 CM |
1371 | /* |
1372 | * rename don't use btrfs_join_transaction, so, once we | |
1373 | * set the transaction to blocked above, we aren't going | |
1374 | * to get any new ordered operations. We can safely run | |
1375 | * it here and no for sure that nothing new will be added | |
1376 | * to the list | |
1377 | */ | |
1378 | btrfs_run_ordered_operations(root, 1); | |
1379 | ||
ed3b3d31 CM |
1380 | prepare_to_wait(&cur_trans->writer_wait, &wait, |
1381 | TASK_UNINTERRUPTIBLE); | |
1382 | ||
13c5a93e | 1383 | if (atomic_read(&cur_trans->num_writers) > 1) |
99d16cbc SW |
1384 | schedule_timeout(MAX_SCHEDULE_TIMEOUT); |
1385 | else if (should_grow) | |
1386 | schedule_timeout(1); | |
15ee9bc7 | 1387 | |
15ee9bc7 | 1388 | finish_wait(&cur_trans->writer_wait, &wait); |
13c5a93e | 1389 | } while (atomic_read(&cur_trans->num_writers) > 1 || |
89573b9c | 1390 | (should_grow && cur_trans->num_joined != joined)); |
15ee9bc7 | 1391 | |
ed0ca140 JB |
1392 | /* |
1393 | * Ok now we need to make sure to block out any other joins while we | |
1394 | * commit the transaction. We could have started a join before setting | |
1395 | * no_join so make sure to wait for num_writers to == 1 again. | |
1396 | */ | |
1397 | spin_lock(&root->fs_info->trans_lock); | |
1398 | root->fs_info->trans_no_join = 1; | |
1399 | spin_unlock(&root->fs_info->trans_lock); | |
1400 | wait_event(cur_trans->writer_wait, | |
1401 | atomic_read(&cur_trans->num_writers) == 1); | |
1402 | ||
7585717f CM |
1403 | /* |
1404 | * the reloc mutex makes sure that we stop | |
1405 | * the balancing code from coming in and moving | |
1406 | * extents around in the middle of the commit | |
1407 | */ | |
1408 | mutex_lock(&root->fs_info->reloc_mutex); | |
1409 | ||
e999376f | 1410 | ret = btrfs_run_delayed_items(trans, root); |
49b25e05 JM |
1411 | if (ret) { |
1412 | mutex_unlock(&root->fs_info->reloc_mutex); | |
1413 | goto cleanup_transaction; | |
1414 | } | |
3063d29f | 1415 | |
e999376f | 1416 | ret = create_pending_snapshots(trans, root->fs_info); |
49b25e05 JM |
1417 | if (ret) { |
1418 | mutex_unlock(&root->fs_info->reloc_mutex); | |
1419 | goto cleanup_transaction; | |
1420 | } | |
16cdcec7 | 1421 | |
56bec294 | 1422 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
49b25e05 JM |
1423 | if (ret) { |
1424 | mutex_unlock(&root->fs_info->reloc_mutex); | |
1425 | goto cleanup_transaction; | |
1426 | } | |
56bec294 | 1427 | |
e999376f CM |
1428 | /* |
1429 | * make sure none of the code above managed to slip in a | |
1430 | * delayed item | |
1431 | */ | |
1432 | btrfs_assert_delayed_root_empty(root); | |
1433 | ||
2c90e5d6 | 1434 | WARN_ON(cur_trans != trans->transaction); |
dc17ff8f | 1435 | |
a2de733c | 1436 | btrfs_scrub_pause(root); |
e02119d5 CM |
1437 | /* btrfs_commit_tree_roots is responsible for getting the |
1438 | * various roots consistent with each other. Every pointer | |
1439 | * in the tree of tree roots has to point to the most up to date | |
1440 | * root for every subvolume and other tree. So, we have to keep | |
1441 | * the tree logging code from jumping in and changing any | |
1442 | * of the trees. | |
1443 | * | |
1444 | * At this point in the commit, there can't be any tree-log | |
1445 | * writers, but a little lower down we drop the trans mutex | |
1446 | * and let new people in. By holding the tree_log_mutex | |
1447 | * from now until after the super is written, we avoid races | |
1448 | * with the tree-log code. | |
1449 | */ | |
1450 | mutex_lock(&root->fs_info->tree_log_mutex); | |
1451 | ||
5d4f98a2 | 1452 | ret = commit_fs_roots(trans, root); |
49b25e05 JM |
1453 | if (ret) { |
1454 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
871383be | 1455 | mutex_unlock(&root->fs_info->reloc_mutex); |
49b25e05 JM |
1456 | goto cleanup_transaction; |
1457 | } | |
54aa1f4d | 1458 | |
5d4f98a2 | 1459 | /* commit_fs_roots gets rid of all the tree log roots, it is now |
e02119d5 CM |
1460 | * safe to free the root of tree log roots |
1461 | */ | |
1462 | btrfs_free_log_root_tree(trans, root->fs_info); | |
1463 | ||
5d4f98a2 | 1464 | ret = commit_cowonly_roots(trans, root); |
49b25e05 JM |
1465 | if (ret) { |
1466 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
871383be | 1467 | mutex_unlock(&root->fs_info->reloc_mutex); |
49b25e05 JM |
1468 | goto cleanup_transaction; |
1469 | } | |
54aa1f4d | 1470 | |
11833d66 YZ |
1471 | btrfs_prepare_extent_commit(trans, root); |
1472 | ||
78fae27e | 1473 | cur_trans = root->fs_info->running_transaction; |
5d4f98a2 YZ |
1474 | |
1475 | btrfs_set_root_node(&root->fs_info->tree_root->root_item, | |
1476 | root->fs_info->tree_root->node); | |
817d52f8 | 1477 | switch_commit_root(root->fs_info->tree_root); |
5d4f98a2 YZ |
1478 | |
1479 | btrfs_set_root_node(&root->fs_info->chunk_root->root_item, | |
1480 | root->fs_info->chunk_root->node); | |
817d52f8 | 1481 | switch_commit_root(root->fs_info->chunk_root); |
5d4f98a2 | 1482 | |
edf39272 | 1483 | assert_qgroups_uptodate(trans); |
5d4f98a2 | 1484 | update_super_roots(root); |
e02119d5 CM |
1485 | |
1486 | if (!root->fs_info->log_root_recovering) { | |
6c41761f DS |
1487 | btrfs_set_super_log_root(root->fs_info->super_copy, 0); |
1488 | btrfs_set_super_log_root_level(root->fs_info->super_copy, 0); | |
e02119d5 CM |
1489 | } |
1490 | ||
6c41761f DS |
1491 | memcpy(root->fs_info->super_for_commit, root->fs_info->super_copy, |
1492 | sizeof(*root->fs_info->super_copy)); | |
ccd467d6 | 1493 | |
f9295749 | 1494 | trans->transaction->blocked = 0; |
a4abeea4 JB |
1495 | spin_lock(&root->fs_info->trans_lock); |
1496 | root->fs_info->running_transaction = NULL; | |
1497 | root->fs_info->trans_no_join = 0; | |
1498 | spin_unlock(&root->fs_info->trans_lock); | |
7585717f | 1499 | mutex_unlock(&root->fs_info->reloc_mutex); |
b7ec40d7 | 1500 | |
f9295749 | 1501 | wake_up(&root->fs_info->transaction_wait); |
e6dcd2dc | 1502 | |
79154b1b | 1503 | ret = btrfs_write_and_wait_transaction(trans, root); |
49b25e05 JM |
1504 | if (ret) { |
1505 | btrfs_error(root->fs_info, ret, | |
1506 | "Error while writing out transaction."); | |
1507 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
1508 | goto cleanup_transaction; | |
1509 | } | |
1510 | ||
1511 | ret = write_ctree_super(trans, root, 0); | |
1512 | if (ret) { | |
1513 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
1514 | goto cleanup_transaction; | |
1515 | } | |
4313b399 | 1516 | |
e02119d5 CM |
1517 | /* |
1518 | * the super is written, we can safely allow the tree-loggers | |
1519 | * to go about their business | |
1520 | */ | |
1521 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
1522 | ||
11833d66 | 1523 | btrfs_finish_extent_commit(trans, root); |
4313b399 | 1524 | |
2c90e5d6 | 1525 | cur_trans->commit_done = 1; |
b7ec40d7 | 1526 | |
15ee9bc7 | 1527 | root->fs_info->last_trans_committed = cur_trans->transid; |
817d52f8 | 1528 | |
2c90e5d6 | 1529 | wake_up(&cur_trans->commit_wait); |
3de4586c | 1530 | |
a4abeea4 | 1531 | spin_lock(&root->fs_info->trans_lock); |
13c5a93e | 1532 | list_del_init(&cur_trans->list); |
a4abeea4 JB |
1533 | spin_unlock(&root->fs_info->trans_lock); |
1534 | ||
78fae27e | 1535 | put_transaction(cur_trans); |
79154b1b | 1536 | put_transaction(cur_trans); |
58176a96 | 1537 | |
1abe9b8a | 1538 | trace_btrfs_transaction_commit(root); |
1539 | ||
a2de733c AJ |
1540 | btrfs_scrub_continue(root); |
1541 | ||
9ed74f2d JB |
1542 | if (current->journal_info == trans) |
1543 | current->journal_info = NULL; | |
1544 | ||
2c90e5d6 | 1545 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
24bbcf04 YZ |
1546 | |
1547 | if (current != root->fs_info->transaction_kthread) | |
1548 | btrfs_run_delayed_iputs(root); | |
1549 | ||
79154b1b | 1550 | return ret; |
49b25e05 JM |
1551 | |
1552 | cleanup_transaction: | |
1553 | btrfs_printk(root->fs_info, "Skipping commit of aborted transaction.\n"); | |
1554 | // WARN_ON(1); | |
1555 | if (current->journal_info == trans) | |
1556 | current->journal_info = NULL; | |
7b8b92af | 1557 | cleanup_transaction(trans, root, ret); |
49b25e05 JM |
1558 | |
1559 | return ret; | |
79154b1b CM |
1560 | } |
1561 | ||
d352ac68 CM |
1562 | /* |
1563 | * interface function to delete all the snapshots we have scheduled for deletion | |
1564 | */ | |
e9d0b13b CM |
1565 | int btrfs_clean_old_snapshots(struct btrfs_root *root) |
1566 | { | |
5d4f98a2 YZ |
1567 | LIST_HEAD(list); |
1568 | struct btrfs_fs_info *fs_info = root->fs_info; | |
1569 | ||
a4abeea4 | 1570 | spin_lock(&fs_info->trans_lock); |
5d4f98a2 | 1571 | list_splice_init(&fs_info->dead_roots, &list); |
a4abeea4 | 1572 | spin_unlock(&fs_info->trans_lock); |
e9d0b13b | 1573 | |
5d4f98a2 | 1574 | while (!list_empty(&list)) { |
2c536799 JM |
1575 | int ret; |
1576 | ||
5d4f98a2 | 1577 | root = list_entry(list.next, struct btrfs_root, root_list); |
76dda93c YZ |
1578 | list_del(&root->root_list); |
1579 | ||
16cdcec7 MX |
1580 | btrfs_kill_all_delayed_nodes(root); |
1581 | ||
76dda93c YZ |
1582 | if (btrfs_header_backref_rev(root->node) < |
1583 | BTRFS_MIXED_BACKREF_REV) | |
2c536799 | 1584 | ret = btrfs_drop_snapshot(root, NULL, 0, 0); |
76dda93c | 1585 | else |
2c536799 JM |
1586 | ret =btrfs_drop_snapshot(root, NULL, 1, 0); |
1587 | BUG_ON(ret < 0); | |
e9d0b13b CM |
1588 | } |
1589 | return 0; | |
1590 | } |