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 AJ |
40 | WARN_ON(transaction->delayed_refs.root.rb_node); |
41 | WARN_ON(!list_empty(&transaction->delayed_refs.seq_head)); | |
2c90e5d6 CM |
42 | memset(transaction, 0, sizeof(*transaction)); |
43 | kmem_cache_free(btrfs_transaction_cachep, transaction); | |
78fae27e | 44 | } |
79154b1b CM |
45 | } |
46 | ||
817d52f8 JB |
47 | static noinline void switch_commit_root(struct btrfs_root *root) |
48 | { | |
817d52f8 JB |
49 | free_extent_buffer(root->commit_root); |
50 | root->commit_root = btrfs_root_node(root); | |
817d52f8 JB |
51 | } |
52 | ||
d352ac68 CM |
53 | /* |
54 | * either allocate a new transaction or hop into the existing one | |
55 | */ | |
a4abeea4 | 56 | static noinline int join_transaction(struct btrfs_root *root, int nofail) |
79154b1b CM |
57 | { |
58 | struct btrfs_transaction *cur_trans; | |
19ae4e81 | 59 | struct btrfs_fs_info *fs_info = root->fs_info; |
a4abeea4 | 60 | |
19ae4e81 | 61 | spin_lock(&fs_info->trans_lock); |
d43317dc | 62 | loop: |
49b25e05 | 63 | /* The file system has been taken offline. No new transactions. */ |
19ae4e81 JS |
64 | if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) { |
65 | spin_unlock(&fs_info->trans_lock); | |
49b25e05 JM |
66 | return -EROFS; |
67 | } | |
68 | ||
19ae4e81 | 69 | if (fs_info->trans_no_join) { |
a4abeea4 | 70 | if (!nofail) { |
19ae4e81 | 71 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 JB |
72 | return -EBUSY; |
73 | } | |
74 | } | |
75 | ||
19ae4e81 | 76 | cur_trans = fs_info->running_transaction; |
a4abeea4 | 77 | if (cur_trans) { |
871383be | 78 | if (cur_trans->aborted) { |
19ae4e81 | 79 | spin_unlock(&fs_info->trans_lock); |
49b25e05 | 80 | return cur_trans->aborted; |
871383be | 81 | } |
a4abeea4 | 82 | atomic_inc(&cur_trans->use_count); |
13c5a93e | 83 | atomic_inc(&cur_trans->num_writers); |
15ee9bc7 | 84 | cur_trans->num_joined++; |
19ae4e81 | 85 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 | 86 | return 0; |
79154b1b | 87 | } |
19ae4e81 | 88 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 JB |
89 | |
90 | cur_trans = kmem_cache_alloc(btrfs_transaction_cachep, GFP_NOFS); | |
91 | if (!cur_trans) | |
92 | return -ENOMEM; | |
d43317dc | 93 | |
19ae4e81 JS |
94 | spin_lock(&fs_info->trans_lock); |
95 | if (fs_info->running_transaction) { | |
d43317dc CM |
96 | /* |
97 | * someone started a transaction after we unlocked. Make sure | |
98 | * to redo the trans_no_join checks above | |
99 | */ | |
a4abeea4 | 100 | kmem_cache_free(btrfs_transaction_cachep, cur_trans); |
19ae4e81 | 101 | cur_trans = fs_info->running_transaction; |
d43317dc | 102 | goto loop; |
79154b1b | 103 | } |
d43317dc | 104 | |
a4abeea4 JB |
105 | atomic_set(&cur_trans->num_writers, 1); |
106 | cur_trans->num_joined = 0; | |
107 | init_waitqueue_head(&cur_trans->writer_wait); | |
108 | init_waitqueue_head(&cur_trans->commit_wait); | |
109 | cur_trans->in_commit = 0; | |
110 | cur_trans->blocked = 0; | |
111 | /* | |
112 | * One for this trans handle, one so it will live on until we | |
113 | * commit the transaction. | |
114 | */ | |
115 | atomic_set(&cur_trans->use_count, 2); | |
116 | cur_trans->commit_done = 0; | |
117 | cur_trans->start_time = get_seconds(); | |
118 | ||
119 | cur_trans->delayed_refs.root = RB_ROOT; | |
120 | cur_trans->delayed_refs.num_entries = 0; | |
121 | cur_trans->delayed_refs.num_heads_ready = 0; | |
122 | cur_trans->delayed_refs.num_heads = 0; | |
123 | cur_trans->delayed_refs.flushing = 0; | |
124 | cur_trans->delayed_refs.run_delayed_start = 0; | |
00f04b88 | 125 | cur_trans->delayed_refs.seq = 1; |
20b297d6 JS |
126 | |
127 | /* | |
128 | * although the tree mod log is per file system and not per transaction, | |
129 | * the log must never go across transaction boundaries. | |
130 | */ | |
131 | smp_mb(); | |
132 | if (!list_empty(&fs_info->tree_mod_seq_list)) { | |
133 | printk(KERN_ERR "btrfs: tree_mod_seq_list not empty when " | |
134 | "creating a fresh transaction\n"); | |
135 | WARN_ON(1); | |
136 | } | |
137 | if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log)) { | |
138 | printk(KERN_ERR "btrfs: tree_mod_log rb tree not empty when " | |
139 | "creating a fresh transaction\n"); | |
140 | WARN_ON(1); | |
141 | } | |
142 | atomic_set(&fs_info->tree_mod_seq, 0); | |
143 | ||
a168650c | 144 | init_waitqueue_head(&cur_trans->delayed_refs.seq_wait); |
a4abeea4 JB |
145 | spin_lock_init(&cur_trans->commit_lock); |
146 | spin_lock_init(&cur_trans->delayed_refs.lock); | |
00f04b88 | 147 | INIT_LIST_HEAD(&cur_trans->delayed_refs.seq_head); |
a4abeea4 JB |
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; |
56bec294 | 348 | h->delayed_ref_updates = 0; |
2a1eb461 | 349 | h->use_count = 1; |
f0486c68 | 350 | h->block_rsv = NULL; |
2a1eb461 | 351 | h->orig_rsv = NULL; |
49b25e05 | 352 | h->aborted = 0; |
b7ec40d7 | 353 | |
a22285a6 YZ |
354 | smp_mb(); |
355 | if (cur_trans->blocked && may_wait_transaction(root, type)) { | |
356 | btrfs_commit_transaction(h, root); | |
357 | goto again; | |
358 | } | |
359 | ||
b5009945 | 360 | if (num_bytes) { |
8c2a3ca2 | 361 | trace_btrfs_space_reservation(root->fs_info, "transaction", |
2bcc0328 | 362 | h->transid, num_bytes, 1); |
b5009945 JB |
363 | h->block_rsv = &root->fs_info->trans_block_rsv; |
364 | h->bytes_reserved = num_bytes; | |
a22285a6 | 365 | } |
9ed74f2d | 366 | |
2a1eb461 | 367 | got_it: |
a4abeea4 | 368 | btrfs_record_root_in_trans(h, root); |
a22285a6 YZ |
369 | |
370 | if (!current->journal_info && type != TRANS_USERSPACE) | |
371 | current->journal_info = h; | |
79154b1b CM |
372 | return h; |
373 | } | |
374 | ||
f9295749 | 375 | struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, |
a22285a6 | 376 | int num_items) |
f9295749 | 377 | { |
a22285a6 | 378 | return start_transaction(root, num_items, TRANS_START); |
f9295749 | 379 | } |
7a7eaa40 | 380 | struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root) |
f9295749 | 381 | { |
a22285a6 | 382 | return start_transaction(root, 0, TRANS_JOIN); |
f9295749 CM |
383 | } |
384 | ||
7a7eaa40 | 385 | struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root) |
0af3d00b JB |
386 | { |
387 | return start_transaction(root, 0, TRANS_JOIN_NOLOCK); | |
388 | } | |
389 | ||
7a7eaa40 | 390 | struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root) |
9ca9ee09 | 391 | { |
7a7eaa40 | 392 | return start_transaction(root, 0, TRANS_USERSPACE); |
9ca9ee09 SW |
393 | } |
394 | ||
d352ac68 | 395 | /* wait for a transaction commit to be fully complete */ |
b9c8300c | 396 | static noinline void wait_for_commit(struct btrfs_root *root, |
89ce8a63 CM |
397 | struct btrfs_transaction *commit) |
398 | { | |
72d63ed6 | 399 | wait_event(commit->commit_wait, commit->commit_done); |
89ce8a63 CM |
400 | } |
401 | ||
46204592 SW |
402 | int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid) |
403 | { | |
404 | struct btrfs_transaction *cur_trans = NULL, *t; | |
405 | int ret; | |
406 | ||
46204592 SW |
407 | ret = 0; |
408 | if (transid) { | |
409 | if (transid <= root->fs_info->last_trans_committed) | |
a4abeea4 | 410 | goto out; |
46204592 SW |
411 | |
412 | /* find specified transaction */ | |
a4abeea4 | 413 | spin_lock(&root->fs_info->trans_lock); |
46204592 SW |
414 | list_for_each_entry(t, &root->fs_info->trans_list, list) { |
415 | if (t->transid == transid) { | |
416 | cur_trans = t; | |
a4abeea4 | 417 | atomic_inc(&cur_trans->use_count); |
46204592 SW |
418 | break; |
419 | } | |
420 | if (t->transid > transid) | |
421 | break; | |
422 | } | |
a4abeea4 | 423 | spin_unlock(&root->fs_info->trans_lock); |
46204592 SW |
424 | ret = -EINVAL; |
425 | if (!cur_trans) | |
a4abeea4 | 426 | goto out; /* bad transid */ |
46204592 SW |
427 | } else { |
428 | /* find newest transaction that is committing | committed */ | |
a4abeea4 | 429 | spin_lock(&root->fs_info->trans_lock); |
46204592 SW |
430 | list_for_each_entry_reverse(t, &root->fs_info->trans_list, |
431 | list) { | |
432 | if (t->in_commit) { | |
433 | if (t->commit_done) | |
3473f3c0 | 434 | break; |
46204592 | 435 | cur_trans = t; |
a4abeea4 | 436 | atomic_inc(&cur_trans->use_count); |
46204592 SW |
437 | break; |
438 | } | |
439 | } | |
a4abeea4 | 440 | spin_unlock(&root->fs_info->trans_lock); |
46204592 | 441 | if (!cur_trans) |
a4abeea4 | 442 | goto out; /* nothing committing|committed */ |
46204592 SW |
443 | } |
444 | ||
46204592 SW |
445 | wait_for_commit(root, cur_trans); |
446 | ||
46204592 SW |
447 | put_transaction(cur_trans); |
448 | ret = 0; | |
a4abeea4 | 449 | out: |
46204592 SW |
450 | return ret; |
451 | } | |
452 | ||
37d1aeee CM |
453 | void btrfs_throttle(struct btrfs_root *root) |
454 | { | |
a4abeea4 | 455 | if (!atomic_read(&root->fs_info->open_ioctl_trans)) |
9ca9ee09 | 456 | wait_current_trans(root); |
37d1aeee CM |
457 | } |
458 | ||
8929ecfa YZ |
459 | static int should_end_transaction(struct btrfs_trans_handle *trans, |
460 | struct btrfs_root *root) | |
461 | { | |
462 | int ret; | |
36ba022a JB |
463 | |
464 | ret = btrfs_block_rsv_check(root, &root->fs_info->global_block_rsv, 5); | |
8929ecfa YZ |
465 | return ret ? 1 : 0; |
466 | } | |
467 | ||
468 | int btrfs_should_end_transaction(struct btrfs_trans_handle *trans, | |
469 | struct btrfs_root *root) | |
470 | { | |
471 | struct btrfs_transaction *cur_trans = trans->transaction; | |
9c8d86db | 472 | struct btrfs_block_rsv *rsv = trans->block_rsv; |
8929ecfa | 473 | int updates; |
49b25e05 | 474 | int err; |
8929ecfa | 475 | |
a4abeea4 | 476 | smp_mb(); |
8929ecfa YZ |
477 | if (cur_trans->blocked || cur_trans->delayed_refs.flushing) |
478 | return 1; | |
479 | ||
9c8d86db JB |
480 | /* |
481 | * We need to do this in case we're deleting csums so the global block | |
482 | * rsv get's used instead of the csum block rsv. | |
483 | */ | |
484 | trans->block_rsv = NULL; | |
485 | ||
8929ecfa YZ |
486 | updates = trans->delayed_ref_updates; |
487 | trans->delayed_ref_updates = 0; | |
49b25e05 JM |
488 | if (updates) { |
489 | err = btrfs_run_delayed_refs(trans, root, updates); | |
490 | if (err) /* Error code will also eval true */ | |
491 | return err; | |
492 | } | |
8929ecfa | 493 | |
9c8d86db JB |
494 | trans->block_rsv = rsv; |
495 | ||
8929ecfa YZ |
496 | return should_end_transaction(trans, root); |
497 | } | |
498 | ||
89ce8a63 | 499 | static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, |
0af3d00b | 500 | struct btrfs_root *root, int throttle, int lock) |
79154b1b | 501 | { |
8929ecfa | 502 | struct btrfs_transaction *cur_trans = trans->transaction; |
ab78c84d | 503 | struct btrfs_fs_info *info = root->fs_info; |
c3e69d58 | 504 | int count = 0; |
4edc2ca3 | 505 | int err = 0; |
c3e69d58 | 506 | |
2a1eb461 JB |
507 | if (--trans->use_count) { |
508 | trans->block_rsv = trans->orig_rsv; | |
509 | return 0; | |
510 | } | |
511 | ||
b24e03db | 512 | btrfs_trans_release_metadata(trans, root); |
4c13d758 | 513 | trans->block_rsv = NULL; |
203bf287 | 514 | while (count < 2) { |
c3e69d58 CM |
515 | unsigned long cur = trans->delayed_ref_updates; |
516 | trans->delayed_ref_updates = 0; | |
517 | if (cur && | |
518 | trans->transaction->delayed_refs.num_heads_ready > 64) { | |
519 | trans->delayed_ref_updates = 0; | |
520 | btrfs_run_delayed_refs(trans, root, cur); | |
521 | } else { | |
522 | break; | |
523 | } | |
524 | count++; | |
56bec294 CM |
525 | } |
526 | ||
a4abeea4 JB |
527 | if (lock && !atomic_read(&root->fs_info->open_ioctl_trans) && |
528 | should_end_transaction(trans, root)) { | |
8929ecfa | 529 | trans->transaction->blocked = 1; |
a4abeea4 JB |
530 | smp_wmb(); |
531 | } | |
8929ecfa | 532 | |
0af3d00b | 533 | if (lock && cur_trans->blocked && !cur_trans->in_commit) { |
81317fde JB |
534 | if (throttle) { |
535 | /* | |
536 | * We may race with somebody else here so end up having | |
537 | * to call end_transaction on ourselves again, so inc | |
538 | * our use_count. | |
539 | */ | |
540 | trans->use_count++; | |
8929ecfa | 541 | return btrfs_commit_transaction(trans, root); |
81317fde | 542 | } else { |
8929ecfa | 543 | wake_up_process(info->transaction_kthread); |
81317fde | 544 | } |
8929ecfa YZ |
545 | } |
546 | ||
8929ecfa | 547 | WARN_ON(cur_trans != info->running_transaction); |
13c5a93e JB |
548 | WARN_ON(atomic_read(&cur_trans->num_writers) < 1); |
549 | atomic_dec(&cur_trans->num_writers); | |
89ce8a63 | 550 | |
99d16cbc | 551 | smp_mb(); |
79154b1b CM |
552 | if (waitqueue_active(&cur_trans->writer_wait)) |
553 | wake_up(&cur_trans->writer_wait); | |
79154b1b | 554 | put_transaction(cur_trans); |
9ed74f2d JB |
555 | |
556 | if (current->journal_info == trans) | |
557 | current->journal_info = NULL; | |
ab78c84d | 558 | |
24bbcf04 YZ |
559 | if (throttle) |
560 | btrfs_run_delayed_iputs(root); | |
561 | ||
49b25e05 JM |
562 | if (trans->aborted || |
563 | root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) { | |
4edc2ca3 | 564 | err = -EIO; |
49b25e05 JM |
565 | } |
566 | ||
4edc2ca3 DJ |
567 | memset(trans, 0, sizeof(*trans)); |
568 | kmem_cache_free(btrfs_trans_handle_cachep, trans); | |
569 | return err; | |
79154b1b CM |
570 | } |
571 | ||
89ce8a63 CM |
572 | int btrfs_end_transaction(struct btrfs_trans_handle *trans, |
573 | struct btrfs_root *root) | |
574 | { | |
16cdcec7 MX |
575 | int ret; |
576 | ||
577 | ret = __btrfs_end_transaction(trans, root, 0, 1); | |
578 | if (ret) | |
579 | return ret; | |
580 | return 0; | |
89ce8a63 CM |
581 | } |
582 | ||
583 | int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans, | |
584 | struct btrfs_root *root) | |
585 | { | |
16cdcec7 MX |
586 | int ret; |
587 | ||
588 | ret = __btrfs_end_transaction(trans, root, 1, 1); | |
589 | if (ret) | |
590 | return ret; | |
591 | return 0; | |
0af3d00b JB |
592 | } |
593 | ||
594 | int btrfs_end_transaction_nolock(struct btrfs_trans_handle *trans, | |
595 | struct btrfs_root *root) | |
596 | { | |
16cdcec7 MX |
597 | int ret; |
598 | ||
599 | ret = __btrfs_end_transaction(trans, root, 0, 0); | |
600 | if (ret) | |
601 | return ret; | |
602 | return 0; | |
603 | } | |
604 | ||
605 | int btrfs_end_transaction_dmeta(struct btrfs_trans_handle *trans, | |
606 | struct btrfs_root *root) | |
607 | { | |
608 | return __btrfs_end_transaction(trans, root, 1, 1); | |
89ce8a63 CM |
609 | } |
610 | ||
d352ac68 CM |
611 | /* |
612 | * when btree blocks are allocated, they have some corresponding bits set for | |
613 | * them in one of two extent_io trees. This is used to make sure all of | |
690587d1 | 614 | * those extents are sent to disk but does not wait on them |
d352ac68 | 615 | */ |
690587d1 | 616 | int btrfs_write_marked_extents(struct btrfs_root *root, |
8cef4e16 | 617 | struct extent_io_tree *dirty_pages, int mark) |
79154b1b | 618 | { |
777e6bd7 | 619 | int err = 0; |
7c4452b9 | 620 | int werr = 0; |
1728366e | 621 | struct address_space *mapping = root->fs_info->btree_inode->i_mapping; |
777e6bd7 | 622 | u64 start = 0; |
5f39d397 | 623 | u64 end; |
7c4452b9 | 624 | |
1728366e JB |
625 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
626 | mark)) { | |
627 | convert_extent_bit(dirty_pages, start, end, EXTENT_NEED_WAIT, mark, | |
628 | GFP_NOFS); | |
629 | err = filemap_fdatawrite_range(mapping, start, end); | |
630 | if (err) | |
631 | werr = err; | |
632 | cond_resched(); | |
633 | start = end + 1; | |
7c4452b9 | 634 | } |
690587d1 CM |
635 | if (err) |
636 | werr = err; | |
637 | return werr; | |
638 | } | |
639 | ||
640 | /* | |
641 | * when btree blocks are allocated, they have some corresponding bits set for | |
642 | * them in one of two extent_io trees. This is used to make sure all of | |
643 | * those extents are on disk for transaction or log commit. We wait | |
644 | * on all the pages and clear them from the dirty pages state tree | |
645 | */ | |
646 | int btrfs_wait_marked_extents(struct btrfs_root *root, | |
8cef4e16 | 647 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 | 648 | { |
690587d1 CM |
649 | int err = 0; |
650 | int werr = 0; | |
1728366e | 651 | struct address_space *mapping = root->fs_info->btree_inode->i_mapping; |
690587d1 CM |
652 | u64 start = 0; |
653 | u64 end; | |
777e6bd7 | 654 | |
1728366e JB |
655 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
656 | EXTENT_NEED_WAIT)) { | |
657 | clear_extent_bits(dirty_pages, start, end, EXTENT_NEED_WAIT, GFP_NOFS); | |
658 | err = filemap_fdatawait_range(mapping, start, end); | |
659 | if (err) | |
660 | werr = err; | |
661 | cond_resched(); | |
662 | start = end + 1; | |
777e6bd7 | 663 | } |
7c4452b9 CM |
664 | if (err) |
665 | werr = err; | |
666 | return werr; | |
79154b1b CM |
667 | } |
668 | ||
690587d1 CM |
669 | /* |
670 | * when btree blocks are allocated, they have some corresponding bits set for | |
671 | * them in one of two extent_io trees. This is used to make sure all of | |
672 | * those extents are on disk for transaction or log commit | |
673 | */ | |
674 | int btrfs_write_and_wait_marked_extents(struct btrfs_root *root, | |
8cef4e16 | 675 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 CM |
676 | { |
677 | int ret; | |
678 | int ret2; | |
679 | ||
8cef4e16 YZ |
680 | ret = btrfs_write_marked_extents(root, dirty_pages, mark); |
681 | ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark); | |
bf0da8c1 CM |
682 | |
683 | if (ret) | |
684 | return ret; | |
685 | if (ret2) | |
686 | return ret2; | |
687 | return 0; | |
690587d1 CM |
688 | } |
689 | ||
d0c803c4 CM |
690 | int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans, |
691 | struct btrfs_root *root) | |
692 | { | |
693 | if (!trans || !trans->transaction) { | |
694 | struct inode *btree_inode; | |
695 | btree_inode = root->fs_info->btree_inode; | |
696 | return filemap_write_and_wait(btree_inode->i_mapping); | |
697 | } | |
698 | return btrfs_write_and_wait_marked_extents(root, | |
8cef4e16 YZ |
699 | &trans->transaction->dirty_pages, |
700 | EXTENT_DIRTY); | |
d0c803c4 CM |
701 | } |
702 | ||
d352ac68 CM |
703 | /* |
704 | * this is used to update the root pointer in the tree of tree roots. | |
705 | * | |
706 | * But, in the case of the extent allocation tree, updating the root | |
707 | * pointer may allocate blocks which may change the root of the extent | |
708 | * allocation tree. | |
709 | * | |
710 | * So, this loops and repeats and makes sure the cowonly root didn't | |
711 | * change while the root pointer was being updated in the metadata. | |
712 | */ | |
0b86a832 CM |
713 | static int update_cowonly_root(struct btrfs_trans_handle *trans, |
714 | struct btrfs_root *root) | |
79154b1b CM |
715 | { |
716 | int ret; | |
0b86a832 | 717 | u64 old_root_bytenr; |
86b9f2ec | 718 | u64 old_root_used; |
0b86a832 | 719 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
79154b1b | 720 | |
86b9f2ec | 721 | old_root_used = btrfs_root_used(&root->root_item); |
0b86a832 | 722 | btrfs_write_dirty_block_groups(trans, root); |
56bec294 | 723 | |
d397712b | 724 | while (1) { |
0b86a832 | 725 | old_root_bytenr = btrfs_root_bytenr(&root->root_item); |
86b9f2ec YZ |
726 | if (old_root_bytenr == root->node->start && |
727 | old_root_used == btrfs_root_used(&root->root_item)) | |
79154b1b | 728 | break; |
87ef2bb4 | 729 | |
5d4f98a2 | 730 | btrfs_set_root_node(&root->root_item, root->node); |
79154b1b | 731 | ret = btrfs_update_root(trans, tree_root, |
0b86a832 CM |
732 | &root->root_key, |
733 | &root->root_item); | |
49b25e05 JM |
734 | if (ret) |
735 | return ret; | |
56bec294 | 736 | |
86b9f2ec | 737 | old_root_used = btrfs_root_used(&root->root_item); |
4a8c9a62 | 738 | ret = btrfs_write_dirty_block_groups(trans, root); |
49b25e05 JM |
739 | if (ret) |
740 | return ret; | |
0b86a832 | 741 | } |
276e680d YZ |
742 | |
743 | if (root != root->fs_info->extent_root) | |
744 | switch_commit_root(root); | |
745 | ||
0b86a832 CM |
746 | return 0; |
747 | } | |
748 | ||
d352ac68 CM |
749 | /* |
750 | * update all the cowonly tree roots on disk | |
49b25e05 JM |
751 | * |
752 | * The error handling in this function may not be obvious. Any of the | |
753 | * failures will cause the file system to go offline. We still need | |
754 | * to clean up the delayed refs. | |
d352ac68 | 755 | */ |
5d4f98a2 YZ |
756 | static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans, |
757 | struct btrfs_root *root) | |
0b86a832 CM |
758 | { |
759 | struct btrfs_fs_info *fs_info = root->fs_info; | |
760 | struct list_head *next; | |
84234f3a | 761 | struct extent_buffer *eb; |
56bec294 | 762 | int ret; |
84234f3a | 763 | |
56bec294 | 764 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
49b25e05 JM |
765 | if (ret) |
766 | return ret; | |
87ef2bb4 | 767 | |
84234f3a | 768 | eb = btrfs_lock_root_node(fs_info->tree_root); |
49b25e05 JM |
769 | ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, |
770 | 0, &eb); | |
84234f3a YZ |
771 | btrfs_tree_unlock(eb); |
772 | free_extent_buffer(eb); | |
0b86a832 | 773 | |
49b25e05 JM |
774 | if (ret) |
775 | return ret; | |
776 | ||
56bec294 | 777 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
49b25e05 JM |
778 | if (ret) |
779 | return ret; | |
87ef2bb4 | 780 | |
733f4fbb SB |
781 | ret = btrfs_run_dev_stats(trans, root->fs_info); |
782 | BUG_ON(ret); | |
783 | ||
d397712b | 784 | while (!list_empty(&fs_info->dirty_cowonly_roots)) { |
0b86a832 CM |
785 | next = fs_info->dirty_cowonly_roots.next; |
786 | list_del_init(next); | |
787 | root = list_entry(next, struct btrfs_root, dirty_list); | |
87ef2bb4 | 788 | |
49b25e05 JM |
789 | ret = update_cowonly_root(trans, root); |
790 | if (ret) | |
791 | return ret; | |
79154b1b | 792 | } |
276e680d YZ |
793 | |
794 | down_write(&fs_info->extent_commit_sem); | |
795 | switch_commit_root(fs_info->extent_root); | |
796 | up_write(&fs_info->extent_commit_sem); | |
797 | ||
79154b1b CM |
798 | return 0; |
799 | } | |
800 | ||
d352ac68 CM |
801 | /* |
802 | * dead roots are old snapshots that need to be deleted. This allocates | |
803 | * a dirty root struct and adds it into the list of dead roots that need to | |
804 | * be deleted | |
805 | */ | |
5d4f98a2 | 806 | int btrfs_add_dead_root(struct btrfs_root *root) |
5eda7b5e | 807 | { |
a4abeea4 | 808 | spin_lock(&root->fs_info->trans_lock); |
5d4f98a2 | 809 | list_add(&root->root_list, &root->fs_info->dead_roots); |
a4abeea4 | 810 | spin_unlock(&root->fs_info->trans_lock); |
5eda7b5e CM |
811 | return 0; |
812 | } | |
813 | ||
d352ac68 | 814 | /* |
5d4f98a2 | 815 | * update all the cowonly tree roots on disk |
d352ac68 | 816 | */ |
5d4f98a2 YZ |
817 | static noinline int commit_fs_roots(struct btrfs_trans_handle *trans, |
818 | struct btrfs_root *root) | |
0f7d52f4 | 819 | { |
0f7d52f4 | 820 | struct btrfs_root *gang[8]; |
5d4f98a2 | 821 | struct btrfs_fs_info *fs_info = root->fs_info; |
0f7d52f4 CM |
822 | int i; |
823 | int ret; | |
54aa1f4d CM |
824 | int err = 0; |
825 | ||
a4abeea4 | 826 | spin_lock(&fs_info->fs_roots_radix_lock); |
d397712b | 827 | while (1) { |
5d4f98a2 YZ |
828 | ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, |
829 | (void **)gang, 0, | |
0f7d52f4 CM |
830 | ARRAY_SIZE(gang), |
831 | BTRFS_ROOT_TRANS_TAG); | |
832 | if (ret == 0) | |
833 | break; | |
834 | for (i = 0; i < ret; i++) { | |
835 | root = gang[i]; | |
5d4f98a2 YZ |
836 | radix_tree_tag_clear(&fs_info->fs_roots_radix, |
837 | (unsigned long)root->root_key.objectid, | |
838 | BTRFS_ROOT_TRANS_TAG); | |
a4abeea4 | 839 | spin_unlock(&fs_info->fs_roots_radix_lock); |
31153d81 | 840 | |
e02119d5 | 841 | btrfs_free_log(trans, root); |
5d4f98a2 | 842 | btrfs_update_reloc_root(trans, root); |
d68fc57b | 843 | btrfs_orphan_commit_root(trans, root); |
bcc63abb | 844 | |
82d5902d LZ |
845 | btrfs_save_ino_cache(root, trans); |
846 | ||
f1ebcc74 LB |
847 | /* see comments in should_cow_block() */ |
848 | root->force_cow = 0; | |
849 | smp_wmb(); | |
850 | ||
978d910d | 851 | if (root->commit_root != root->node) { |
581bb050 | 852 | mutex_lock(&root->fs_commit_mutex); |
817d52f8 | 853 | switch_commit_root(root); |
581bb050 LZ |
854 | btrfs_unpin_free_ino(root); |
855 | mutex_unlock(&root->fs_commit_mutex); | |
856 | ||
978d910d YZ |
857 | btrfs_set_root_node(&root->root_item, |
858 | root->node); | |
859 | } | |
5d4f98a2 | 860 | |
5d4f98a2 | 861 | err = btrfs_update_root(trans, fs_info->tree_root, |
0f7d52f4 CM |
862 | &root->root_key, |
863 | &root->root_item); | |
a4abeea4 | 864 | spin_lock(&fs_info->fs_roots_radix_lock); |
54aa1f4d CM |
865 | if (err) |
866 | break; | |
0f7d52f4 CM |
867 | } |
868 | } | |
a4abeea4 | 869 | spin_unlock(&fs_info->fs_roots_radix_lock); |
54aa1f4d | 870 | return err; |
0f7d52f4 CM |
871 | } |
872 | ||
d352ac68 CM |
873 | /* |
874 | * defrag a given btree. If cacheonly == 1, this won't read from the disk, | |
875 | * otherwise every leaf in the btree is read and defragged. | |
876 | */ | |
e9d0b13b CM |
877 | int btrfs_defrag_root(struct btrfs_root *root, int cacheonly) |
878 | { | |
879 | struct btrfs_fs_info *info = root->fs_info; | |
e9d0b13b | 880 | struct btrfs_trans_handle *trans; |
8929ecfa | 881 | int ret; |
d3c2fdcf | 882 | unsigned long nr; |
e9d0b13b | 883 | |
8929ecfa | 884 | if (xchg(&root->defrag_running, 1)) |
e9d0b13b | 885 | return 0; |
8929ecfa | 886 | |
6b80053d | 887 | while (1) { |
8929ecfa YZ |
888 | trans = btrfs_start_transaction(root, 0); |
889 | if (IS_ERR(trans)) | |
890 | return PTR_ERR(trans); | |
891 | ||
e9d0b13b | 892 | ret = btrfs_defrag_leaves(trans, root, cacheonly); |
8929ecfa | 893 | |
d3c2fdcf | 894 | nr = trans->blocks_used; |
e9d0b13b | 895 | btrfs_end_transaction(trans, root); |
d3c2fdcf | 896 | btrfs_btree_balance_dirty(info->tree_root, nr); |
e9d0b13b CM |
897 | cond_resched(); |
898 | ||
7841cb28 | 899 | if (btrfs_fs_closing(root->fs_info) || ret != -EAGAIN) |
e9d0b13b CM |
900 | break; |
901 | } | |
902 | root->defrag_running = 0; | |
8929ecfa | 903 | return ret; |
e9d0b13b CM |
904 | } |
905 | ||
d352ac68 CM |
906 | /* |
907 | * new snapshots need to be created at a very specific time in the | |
908 | * transaction commit. This does the actual creation | |
909 | */ | |
80b6794d | 910 | static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, |
3063d29f CM |
911 | struct btrfs_fs_info *fs_info, |
912 | struct btrfs_pending_snapshot *pending) | |
913 | { | |
914 | struct btrfs_key key; | |
80b6794d | 915 | struct btrfs_root_item *new_root_item; |
3063d29f CM |
916 | struct btrfs_root *tree_root = fs_info->tree_root; |
917 | struct btrfs_root *root = pending->root; | |
6bdb72de | 918 | struct btrfs_root *parent_root; |
98c9942a | 919 | struct btrfs_block_rsv *rsv; |
6bdb72de | 920 | struct inode *parent_inode; |
6a912213 | 921 | struct dentry *parent; |
a22285a6 | 922 | struct dentry *dentry; |
3063d29f | 923 | struct extent_buffer *tmp; |
925baedd | 924 | struct extent_buffer *old; |
3063d29f | 925 | int ret; |
d68fc57b | 926 | u64 to_reserve = 0; |
6bdb72de | 927 | u64 index = 0; |
a22285a6 | 928 | u64 objectid; |
b83cc969 | 929 | u64 root_flags; |
3063d29f | 930 | |
98c9942a LB |
931 | rsv = trans->block_rsv; |
932 | ||
80b6794d CM |
933 | new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS); |
934 | if (!new_root_item) { | |
49b25e05 | 935 | ret = pending->error = -ENOMEM; |
80b6794d CM |
936 | goto fail; |
937 | } | |
a22285a6 | 938 | |
581bb050 | 939 | ret = btrfs_find_free_objectid(tree_root, &objectid); |
a22285a6 YZ |
940 | if (ret) { |
941 | pending->error = ret; | |
3063d29f | 942 | goto fail; |
a22285a6 | 943 | } |
3063d29f | 944 | |
3fd0a558 | 945 | btrfs_reloc_pre_snapshot(trans, pending, &to_reserve); |
d68fc57b YZ |
946 | |
947 | if (to_reserve > 0) { | |
62f30c54 MX |
948 | ret = btrfs_block_rsv_add_noflush(root, &pending->block_rsv, |
949 | to_reserve); | |
d68fc57b YZ |
950 | if (ret) { |
951 | pending->error = ret; | |
952 | goto fail; | |
953 | } | |
954 | } | |
955 | ||
3063d29f | 956 | key.objectid = objectid; |
a22285a6 YZ |
957 | key.offset = (u64)-1; |
958 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3063d29f | 959 | |
a22285a6 | 960 | trans->block_rsv = &pending->block_rsv; |
3de4586c | 961 | |
a22285a6 | 962 | dentry = pending->dentry; |
6a912213 JB |
963 | parent = dget_parent(dentry); |
964 | parent_inode = parent->d_inode; | |
a22285a6 | 965 | parent_root = BTRFS_I(parent_inode)->root; |
7585717f | 966 | record_root_in_trans(trans, parent_root); |
a22285a6 | 967 | |
3063d29f CM |
968 | /* |
969 | * insert the directory item | |
970 | */ | |
3de4586c | 971 | ret = btrfs_set_inode_index(parent_inode, &index); |
49b25e05 | 972 | BUG_ON(ret); /* -ENOMEM */ |
0660b5af | 973 | ret = btrfs_insert_dir_item(trans, parent_root, |
a22285a6 | 974 | dentry->d_name.name, dentry->d_name.len, |
16cdcec7 | 975 | parent_inode, &key, |
a22285a6 | 976 | BTRFS_FT_DIR, index); |
79787eaa | 977 | if (ret == -EEXIST) { |
fe66a05a CM |
978 | pending->error = -EEXIST; |
979 | dput(parent); | |
980 | goto fail; | |
79787eaa JM |
981 | } else if (ret) { |
982 | goto abort_trans_dput; | |
983 | } | |
0660b5af | 984 | |
a22285a6 YZ |
985 | btrfs_i_size_write(parent_inode, parent_inode->i_size + |
986 | dentry->d_name.len * 2); | |
52c26179 | 987 | ret = btrfs_update_inode(trans, parent_root, parent_inode); |
49b25e05 | 988 | if (ret) |
79787eaa | 989 | goto abort_trans_dput; |
52c26179 | 990 | |
e999376f CM |
991 | /* |
992 | * pull in the delayed directory update | |
993 | * and the delayed inode item | |
994 | * otherwise we corrupt the FS during | |
995 | * snapshot | |
996 | */ | |
997 | ret = btrfs_run_delayed_items(trans, root); | |
79787eaa JM |
998 | if (ret) { /* Transaction aborted */ |
999 | dput(parent); | |
49b25e05 | 1000 | goto fail; |
79787eaa | 1001 | } |
e999376f | 1002 | |
7585717f | 1003 | record_root_in_trans(trans, root); |
6bdb72de SW |
1004 | btrfs_set_root_last_snapshot(&root->root_item, trans->transid); |
1005 | memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); | |
08fe4db1 | 1006 | btrfs_check_and_init_root_item(new_root_item); |
6bdb72de | 1007 | |
b83cc969 LZ |
1008 | root_flags = btrfs_root_flags(new_root_item); |
1009 | if (pending->readonly) | |
1010 | root_flags |= BTRFS_ROOT_SUBVOL_RDONLY; | |
1011 | else | |
1012 | root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY; | |
1013 | btrfs_set_root_flags(new_root_item, root_flags); | |
1014 | ||
6bdb72de | 1015 | old = btrfs_lock_root_node(root); |
49b25e05 | 1016 | ret = btrfs_cow_block(trans, root, old, NULL, 0, &old); |
79787eaa JM |
1017 | if (ret) { |
1018 | btrfs_tree_unlock(old); | |
1019 | free_extent_buffer(old); | |
1020 | goto abort_trans_dput; | |
1021 | } | |
49b25e05 | 1022 | |
6bdb72de SW |
1023 | btrfs_set_lock_blocking(old); |
1024 | ||
49b25e05 | 1025 | ret = btrfs_copy_root(trans, root, old, &tmp, objectid); |
79787eaa | 1026 | /* clean up in any case */ |
6bdb72de SW |
1027 | btrfs_tree_unlock(old); |
1028 | free_extent_buffer(old); | |
79787eaa JM |
1029 | if (ret) |
1030 | goto abort_trans_dput; | |
6bdb72de | 1031 | |
f1ebcc74 LB |
1032 | /* see comments in should_cow_block() */ |
1033 | root->force_cow = 1; | |
1034 | smp_wmb(); | |
1035 | ||
6bdb72de | 1036 | btrfs_set_root_node(new_root_item, tmp); |
a22285a6 YZ |
1037 | /* record when the snapshot was created in key.offset */ |
1038 | key.offset = trans->transid; | |
1039 | ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); | |
6bdb72de SW |
1040 | btrfs_tree_unlock(tmp); |
1041 | free_extent_buffer(tmp); | |
49b25e05 | 1042 | if (ret) |
79787eaa | 1043 | goto abort_trans_dput; |
6bdb72de | 1044 | |
a22285a6 YZ |
1045 | /* |
1046 | * insert root back/forward references | |
1047 | */ | |
1048 | ret = btrfs_add_root_ref(trans, tree_root, objectid, | |
0660b5af | 1049 | parent_root->root_key.objectid, |
33345d01 | 1050 | btrfs_ino(parent_inode), index, |
a22285a6 | 1051 | dentry->d_name.name, dentry->d_name.len); |
79787eaa | 1052 | dput(parent); |
49b25e05 JM |
1053 | if (ret) |
1054 | goto fail; | |
0660b5af | 1055 | |
a22285a6 YZ |
1056 | key.offset = (u64)-1; |
1057 | pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key); | |
79787eaa JM |
1058 | if (IS_ERR(pending->snap)) { |
1059 | ret = PTR_ERR(pending->snap); | |
49b25e05 | 1060 | goto abort_trans; |
79787eaa | 1061 | } |
d68fc57b | 1062 | |
49b25e05 JM |
1063 | ret = btrfs_reloc_post_snapshot(trans, pending); |
1064 | if (ret) | |
1065 | goto abort_trans; | |
1066 | ret = 0; | |
3063d29f | 1067 | fail: |
6bdb72de | 1068 | kfree(new_root_item); |
98c9942a | 1069 | trans->block_rsv = rsv; |
a22285a6 | 1070 | btrfs_block_rsv_release(root, &pending->block_rsv, (u64)-1); |
49b25e05 JM |
1071 | return ret; |
1072 | ||
79787eaa JM |
1073 | abort_trans_dput: |
1074 | dput(parent); | |
49b25e05 JM |
1075 | abort_trans: |
1076 | btrfs_abort_transaction(trans, root, ret); | |
1077 | goto fail; | |
3063d29f CM |
1078 | } |
1079 | ||
d352ac68 CM |
1080 | /* |
1081 | * create all the snapshots we've scheduled for creation | |
1082 | */ | |
80b6794d CM |
1083 | static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans, |
1084 | struct btrfs_fs_info *fs_info) | |
3de4586c CM |
1085 | { |
1086 | struct btrfs_pending_snapshot *pending; | |
1087 | struct list_head *head = &trans->transaction->pending_snapshots; | |
3de4586c | 1088 | |
fe66a05a CM |
1089 | list_for_each_entry(pending, head, list) |
1090 | create_pending_snapshot(trans, fs_info, pending); | |
3de4586c CM |
1091 | return 0; |
1092 | } | |
1093 | ||
5d4f98a2 YZ |
1094 | static void update_super_roots(struct btrfs_root *root) |
1095 | { | |
1096 | struct btrfs_root_item *root_item; | |
1097 | struct btrfs_super_block *super; | |
1098 | ||
6c41761f | 1099 | super = root->fs_info->super_copy; |
5d4f98a2 YZ |
1100 | |
1101 | root_item = &root->fs_info->chunk_root->root_item; | |
1102 | super->chunk_root = root_item->bytenr; | |
1103 | super->chunk_root_generation = root_item->generation; | |
1104 | super->chunk_root_level = root_item->level; | |
1105 | ||
1106 | root_item = &root->fs_info->tree_root->root_item; | |
1107 | super->root = root_item->bytenr; | |
1108 | super->generation = root_item->generation; | |
1109 | super->root_level = root_item->level; | |
73bc1876 | 1110 | if (btrfs_test_opt(root, SPACE_CACHE)) |
0af3d00b | 1111 | super->cache_generation = root_item->generation; |
5d4f98a2 YZ |
1112 | } |
1113 | ||
f36f3042 CM |
1114 | int btrfs_transaction_in_commit(struct btrfs_fs_info *info) |
1115 | { | |
1116 | int ret = 0; | |
a4abeea4 | 1117 | spin_lock(&info->trans_lock); |
f36f3042 CM |
1118 | if (info->running_transaction) |
1119 | ret = info->running_transaction->in_commit; | |
a4abeea4 | 1120 | spin_unlock(&info->trans_lock); |
f36f3042 CM |
1121 | return ret; |
1122 | } | |
1123 | ||
8929ecfa YZ |
1124 | int btrfs_transaction_blocked(struct btrfs_fs_info *info) |
1125 | { | |
1126 | int ret = 0; | |
a4abeea4 | 1127 | spin_lock(&info->trans_lock); |
8929ecfa YZ |
1128 | if (info->running_transaction) |
1129 | ret = info->running_transaction->blocked; | |
a4abeea4 | 1130 | spin_unlock(&info->trans_lock); |
8929ecfa YZ |
1131 | return ret; |
1132 | } | |
1133 | ||
bb9c12c9 SW |
1134 | /* |
1135 | * wait for the current transaction commit to start and block subsequent | |
1136 | * transaction joins | |
1137 | */ | |
1138 | static void wait_current_trans_commit_start(struct btrfs_root *root, | |
1139 | struct btrfs_transaction *trans) | |
1140 | { | |
72d63ed6 | 1141 | wait_event(root->fs_info->transaction_blocked_wait, trans->in_commit); |
bb9c12c9 SW |
1142 | } |
1143 | ||
1144 | /* | |
1145 | * wait for the current transaction to start and then become unblocked. | |
1146 | * caller holds ref. | |
1147 | */ | |
1148 | static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root, | |
1149 | struct btrfs_transaction *trans) | |
1150 | { | |
72d63ed6 LZ |
1151 | wait_event(root->fs_info->transaction_wait, |
1152 | trans->commit_done || (trans->in_commit && !trans->blocked)); | |
bb9c12c9 SW |
1153 | } |
1154 | ||
1155 | /* | |
1156 | * commit transactions asynchronously. once btrfs_commit_transaction_async | |
1157 | * returns, any subsequent transaction will not be allowed to join. | |
1158 | */ | |
1159 | struct btrfs_async_commit { | |
1160 | struct btrfs_trans_handle *newtrans; | |
1161 | struct btrfs_root *root; | |
1162 | struct delayed_work work; | |
1163 | }; | |
1164 | ||
1165 | static void do_async_commit(struct work_struct *work) | |
1166 | { | |
1167 | struct btrfs_async_commit *ac = | |
1168 | container_of(work, struct btrfs_async_commit, work.work); | |
1169 | ||
1170 | btrfs_commit_transaction(ac->newtrans, ac->root); | |
1171 | kfree(ac); | |
1172 | } | |
1173 | ||
1174 | int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans, | |
1175 | struct btrfs_root *root, | |
1176 | int wait_for_unblock) | |
1177 | { | |
1178 | struct btrfs_async_commit *ac; | |
1179 | struct btrfs_transaction *cur_trans; | |
1180 | ||
1181 | ac = kmalloc(sizeof(*ac), GFP_NOFS); | |
db5b493a TI |
1182 | if (!ac) |
1183 | return -ENOMEM; | |
bb9c12c9 SW |
1184 | |
1185 | INIT_DELAYED_WORK(&ac->work, do_async_commit); | |
1186 | ac->root = root; | |
7a7eaa40 | 1187 | ac->newtrans = btrfs_join_transaction(root); |
3612b495 TI |
1188 | if (IS_ERR(ac->newtrans)) { |
1189 | int err = PTR_ERR(ac->newtrans); | |
1190 | kfree(ac); | |
1191 | return err; | |
1192 | } | |
bb9c12c9 SW |
1193 | |
1194 | /* take transaction reference */ | |
bb9c12c9 | 1195 | cur_trans = trans->transaction; |
13c5a93e | 1196 | atomic_inc(&cur_trans->use_count); |
bb9c12c9 SW |
1197 | |
1198 | btrfs_end_transaction(trans, root); | |
1199 | schedule_delayed_work(&ac->work, 0); | |
1200 | ||
1201 | /* wait for transaction to start and unblock */ | |
bb9c12c9 SW |
1202 | if (wait_for_unblock) |
1203 | wait_current_trans_commit_start_and_unblock(root, cur_trans); | |
1204 | else | |
1205 | wait_current_trans_commit_start(root, cur_trans); | |
bb9c12c9 | 1206 | |
38e88054 SW |
1207 | if (current->journal_info == trans) |
1208 | current->journal_info = NULL; | |
1209 | ||
1210 | put_transaction(cur_trans); | |
bb9c12c9 SW |
1211 | return 0; |
1212 | } | |
1213 | ||
49b25e05 JM |
1214 | |
1215 | static void cleanup_transaction(struct btrfs_trans_handle *trans, | |
1216 | struct btrfs_root *root) | |
1217 | { | |
1218 | struct btrfs_transaction *cur_trans = trans->transaction; | |
1219 | ||
1220 | WARN_ON(trans->use_count > 1); | |
1221 | ||
1222 | spin_lock(&root->fs_info->trans_lock); | |
1223 | list_del_init(&cur_trans->list); | |
1224 | spin_unlock(&root->fs_info->trans_lock); | |
1225 | ||
1226 | btrfs_cleanup_one_transaction(trans->transaction, root); | |
1227 | ||
1228 | put_transaction(cur_trans); | |
1229 | put_transaction(cur_trans); | |
1230 | ||
1231 | trace_btrfs_transaction_commit(root); | |
1232 | ||
1233 | btrfs_scrub_continue(root); | |
1234 | ||
1235 | if (current->journal_info == trans) | |
1236 | current->journal_info = NULL; | |
1237 | ||
1238 | kmem_cache_free(btrfs_trans_handle_cachep, trans); | |
1239 | } | |
1240 | ||
bb9c12c9 SW |
1241 | /* |
1242 | * btrfs_transaction state sequence: | |
1243 | * in_commit = 0, blocked = 0 (initial) | |
1244 | * in_commit = 1, blocked = 1 | |
1245 | * blocked = 0 | |
1246 | * commit_done = 1 | |
1247 | */ | |
79154b1b CM |
1248 | int btrfs_commit_transaction(struct btrfs_trans_handle *trans, |
1249 | struct btrfs_root *root) | |
1250 | { | |
15ee9bc7 | 1251 | unsigned long joined = 0; |
49b25e05 | 1252 | struct btrfs_transaction *cur_trans = trans->transaction; |
8fd17795 | 1253 | struct btrfs_transaction *prev_trans = NULL; |
79154b1b | 1254 | DEFINE_WAIT(wait); |
49b25e05 | 1255 | int ret = -EIO; |
89573b9c CM |
1256 | int should_grow = 0; |
1257 | unsigned long now = get_seconds(); | |
dccae999 | 1258 | int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT); |
79154b1b | 1259 | |
5a3f23d5 CM |
1260 | btrfs_run_ordered_operations(root, 0); |
1261 | ||
b24e03db | 1262 | btrfs_trans_release_metadata(trans, root); |
9c8d86db JB |
1263 | trans->block_rsv = NULL; |
1264 | ||
49b25e05 JM |
1265 | if (cur_trans->aborted) |
1266 | goto cleanup_transaction; | |
1267 | ||
56bec294 CM |
1268 | /* make a pass through all the delayed refs we have so far |
1269 | * any runnings procs may add more while we are here | |
1270 | */ | |
1271 | ret = btrfs_run_delayed_refs(trans, root, 0); | |
49b25e05 JM |
1272 | if (ret) |
1273 | goto cleanup_transaction; | |
56bec294 | 1274 | |
b7ec40d7 | 1275 | cur_trans = trans->transaction; |
49b25e05 | 1276 | |
56bec294 CM |
1277 | /* |
1278 | * set the flushing flag so procs in this transaction have to | |
1279 | * start sending their work down. | |
1280 | */ | |
b7ec40d7 | 1281 | cur_trans->delayed_refs.flushing = 1; |
56bec294 | 1282 | |
c3e69d58 | 1283 | ret = btrfs_run_delayed_refs(trans, root, 0); |
49b25e05 JM |
1284 | if (ret) |
1285 | goto cleanup_transaction; | |
56bec294 | 1286 | |
a4abeea4 | 1287 | spin_lock(&cur_trans->commit_lock); |
b7ec40d7 | 1288 | if (cur_trans->in_commit) { |
a4abeea4 | 1289 | spin_unlock(&cur_trans->commit_lock); |
13c5a93e | 1290 | atomic_inc(&cur_trans->use_count); |
49b25e05 | 1291 | ret = btrfs_end_transaction(trans, root); |
ccd467d6 | 1292 | |
b9c8300c | 1293 | wait_for_commit(root, cur_trans); |
15ee9bc7 | 1294 | |
79154b1b | 1295 | put_transaction(cur_trans); |
15ee9bc7 | 1296 | |
49b25e05 | 1297 | return ret; |
79154b1b | 1298 | } |
4313b399 | 1299 | |
2c90e5d6 | 1300 | trans->transaction->in_commit = 1; |
f9295749 | 1301 | trans->transaction->blocked = 1; |
a4abeea4 | 1302 | spin_unlock(&cur_trans->commit_lock); |
bb9c12c9 SW |
1303 | wake_up(&root->fs_info->transaction_blocked_wait); |
1304 | ||
a4abeea4 | 1305 | spin_lock(&root->fs_info->trans_lock); |
ccd467d6 CM |
1306 | if (cur_trans->list.prev != &root->fs_info->trans_list) { |
1307 | prev_trans = list_entry(cur_trans->list.prev, | |
1308 | struct btrfs_transaction, list); | |
1309 | if (!prev_trans->commit_done) { | |
13c5a93e | 1310 | atomic_inc(&prev_trans->use_count); |
a4abeea4 | 1311 | spin_unlock(&root->fs_info->trans_lock); |
ccd467d6 CM |
1312 | |
1313 | wait_for_commit(root, prev_trans); | |
ccd467d6 | 1314 | |
15ee9bc7 | 1315 | put_transaction(prev_trans); |
a4abeea4 JB |
1316 | } else { |
1317 | spin_unlock(&root->fs_info->trans_lock); | |
ccd467d6 | 1318 | } |
a4abeea4 JB |
1319 | } else { |
1320 | spin_unlock(&root->fs_info->trans_lock); | |
ccd467d6 | 1321 | } |
15ee9bc7 | 1322 | |
89573b9c CM |
1323 | if (now < cur_trans->start_time || now - cur_trans->start_time < 1) |
1324 | should_grow = 1; | |
1325 | ||
15ee9bc7 | 1326 | do { |
7ea394f1 | 1327 | int snap_pending = 0; |
a4abeea4 | 1328 | |
15ee9bc7 | 1329 | joined = cur_trans->num_joined; |
7ea394f1 YZ |
1330 | if (!list_empty(&trans->transaction->pending_snapshots)) |
1331 | snap_pending = 1; | |
1332 | ||
2c90e5d6 | 1333 | WARN_ON(cur_trans != trans->transaction); |
15ee9bc7 | 1334 | |
0bdb1db2 | 1335 | if (flush_on_commit || snap_pending) { |
24bbcf04 | 1336 | btrfs_start_delalloc_inodes(root, 1); |
143bede5 | 1337 | btrfs_wait_ordered_extents(root, 0, 1); |
7ea394f1 YZ |
1338 | } |
1339 | ||
16cdcec7 | 1340 | ret = btrfs_run_delayed_items(trans, root); |
49b25e05 JM |
1341 | if (ret) |
1342 | goto cleanup_transaction; | |
16cdcec7 | 1343 | |
5a3f23d5 CM |
1344 | /* |
1345 | * rename don't use btrfs_join_transaction, so, once we | |
1346 | * set the transaction to blocked above, we aren't going | |
1347 | * to get any new ordered operations. We can safely run | |
1348 | * it here and no for sure that nothing new will be added | |
1349 | * to the list | |
1350 | */ | |
1351 | btrfs_run_ordered_operations(root, 1); | |
1352 | ||
ed3b3d31 CM |
1353 | prepare_to_wait(&cur_trans->writer_wait, &wait, |
1354 | TASK_UNINTERRUPTIBLE); | |
1355 | ||
13c5a93e | 1356 | if (atomic_read(&cur_trans->num_writers) > 1) |
99d16cbc SW |
1357 | schedule_timeout(MAX_SCHEDULE_TIMEOUT); |
1358 | else if (should_grow) | |
1359 | schedule_timeout(1); | |
15ee9bc7 | 1360 | |
15ee9bc7 | 1361 | finish_wait(&cur_trans->writer_wait, &wait); |
13c5a93e | 1362 | } while (atomic_read(&cur_trans->num_writers) > 1 || |
89573b9c | 1363 | (should_grow && cur_trans->num_joined != joined)); |
15ee9bc7 | 1364 | |
ed0ca140 JB |
1365 | /* |
1366 | * Ok now we need to make sure to block out any other joins while we | |
1367 | * commit the transaction. We could have started a join before setting | |
1368 | * no_join so make sure to wait for num_writers to == 1 again. | |
1369 | */ | |
1370 | spin_lock(&root->fs_info->trans_lock); | |
1371 | root->fs_info->trans_no_join = 1; | |
1372 | spin_unlock(&root->fs_info->trans_lock); | |
1373 | wait_event(cur_trans->writer_wait, | |
1374 | atomic_read(&cur_trans->num_writers) == 1); | |
1375 | ||
7585717f CM |
1376 | /* |
1377 | * the reloc mutex makes sure that we stop | |
1378 | * the balancing code from coming in and moving | |
1379 | * extents around in the middle of the commit | |
1380 | */ | |
1381 | mutex_lock(&root->fs_info->reloc_mutex); | |
1382 | ||
e999376f | 1383 | ret = btrfs_run_delayed_items(trans, root); |
49b25e05 JM |
1384 | if (ret) { |
1385 | mutex_unlock(&root->fs_info->reloc_mutex); | |
1386 | goto cleanup_transaction; | |
1387 | } | |
3063d29f | 1388 | |
e999376f | 1389 | ret = create_pending_snapshots(trans, root->fs_info); |
49b25e05 JM |
1390 | if (ret) { |
1391 | mutex_unlock(&root->fs_info->reloc_mutex); | |
1392 | goto cleanup_transaction; | |
1393 | } | |
16cdcec7 | 1394 | |
56bec294 | 1395 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
49b25e05 JM |
1396 | if (ret) { |
1397 | mutex_unlock(&root->fs_info->reloc_mutex); | |
1398 | goto cleanup_transaction; | |
1399 | } | |
56bec294 | 1400 | |
e999376f CM |
1401 | /* |
1402 | * make sure none of the code above managed to slip in a | |
1403 | * delayed item | |
1404 | */ | |
1405 | btrfs_assert_delayed_root_empty(root); | |
1406 | ||
2c90e5d6 | 1407 | WARN_ON(cur_trans != trans->transaction); |
dc17ff8f | 1408 | |
a2de733c | 1409 | btrfs_scrub_pause(root); |
e02119d5 CM |
1410 | /* btrfs_commit_tree_roots is responsible for getting the |
1411 | * various roots consistent with each other. Every pointer | |
1412 | * in the tree of tree roots has to point to the most up to date | |
1413 | * root for every subvolume and other tree. So, we have to keep | |
1414 | * the tree logging code from jumping in and changing any | |
1415 | * of the trees. | |
1416 | * | |
1417 | * At this point in the commit, there can't be any tree-log | |
1418 | * writers, but a little lower down we drop the trans mutex | |
1419 | * and let new people in. By holding the tree_log_mutex | |
1420 | * from now until after the super is written, we avoid races | |
1421 | * with the tree-log code. | |
1422 | */ | |
1423 | mutex_lock(&root->fs_info->tree_log_mutex); | |
1424 | ||
5d4f98a2 | 1425 | ret = commit_fs_roots(trans, root); |
49b25e05 JM |
1426 | if (ret) { |
1427 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
871383be | 1428 | mutex_unlock(&root->fs_info->reloc_mutex); |
49b25e05 JM |
1429 | goto cleanup_transaction; |
1430 | } | |
54aa1f4d | 1431 | |
5d4f98a2 | 1432 | /* commit_fs_roots gets rid of all the tree log roots, it is now |
e02119d5 CM |
1433 | * safe to free the root of tree log roots |
1434 | */ | |
1435 | btrfs_free_log_root_tree(trans, root->fs_info); | |
1436 | ||
5d4f98a2 | 1437 | ret = commit_cowonly_roots(trans, root); |
49b25e05 JM |
1438 | if (ret) { |
1439 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
871383be | 1440 | mutex_unlock(&root->fs_info->reloc_mutex); |
49b25e05 JM |
1441 | goto cleanup_transaction; |
1442 | } | |
54aa1f4d | 1443 | |
11833d66 YZ |
1444 | btrfs_prepare_extent_commit(trans, root); |
1445 | ||
78fae27e | 1446 | cur_trans = root->fs_info->running_transaction; |
5d4f98a2 YZ |
1447 | |
1448 | btrfs_set_root_node(&root->fs_info->tree_root->root_item, | |
1449 | root->fs_info->tree_root->node); | |
817d52f8 | 1450 | switch_commit_root(root->fs_info->tree_root); |
5d4f98a2 YZ |
1451 | |
1452 | btrfs_set_root_node(&root->fs_info->chunk_root->root_item, | |
1453 | root->fs_info->chunk_root->node); | |
817d52f8 | 1454 | switch_commit_root(root->fs_info->chunk_root); |
5d4f98a2 YZ |
1455 | |
1456 | update_super_roots(root); | |
e02119d5 CM |
1457 | |
1458 | if (!root->fs_info->log_root_recovering) { | |
6c41761f DS |
1459 | btrfs_set_super_log_root(root->fs_info->super_copy, 0); |
1460 | btrfs_set_super_log_root_level(root->fs_info->super_copy, 0); | |
e02119d5 CM |
1461 | } |
1462 | ||
6c41761f DS |
1463 | memcpy(root->fs_info->super_for_commit, root->fs_info->super_copy, |
1464 | sizeof(*root->fs_info->super_copy)); | |
ccd467d6 | 1465 | |
f9295749 | 1466 | trans->transaction->blocked = 0; |
a4abeea4 JB |
1467 | spin_lock(&root->fs_info->trans_lock); |
1468 | root->fs_info->running_transaction = NULL; | |
1469 | root->fs_info->trans_no_join = 0; | |
1470 | spin_unlock(&root->fs_info->trans_lock); | |
7585717f | 1471 | mutex_unlock(&root->fs_info->reloc_mutex); |
b7ec40d7 | 1472 | |
f9295749 | 1473 | wake_up(&root->fs_info->transaction_wait); |
e6dcd2dc | 1474 | |
79154b1b | 1475 | ret = btrfs_write_and_wait_transaction(trans, root); |
49b25e05 JM |
1476 | if (ret) { |
1477 | btrfs_error(root->fs_info, ret, | |
1478 | "Error while writing out transaction."); | |
1479 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
1480 | goto cleanup_transaction; | |
1481 | } | |
1482 | ||
1483 | ret = write_ctree_super(trans, root, 0); | |
1484 | if (ret) { | |
1485 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
1486 | goto cleanup_transaction; | |
1487 | } | |
4313b399 | 1488 | |
e02119d5 CM |
1489 | /* |
1490 | * the super is written, we can safely allow the tree-loggers | |
1491 | * to go about their business | |
1492 | */ | |
1493 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
1494 | ||
11833d66 | 1495 | btrfs_finish_extent_commit(trans, root); |
4313b399 | 1496 | |
2c90e5d6 | 1497 | cur_trans->commit_done = 1; |
b7ec40d7 | 1498 | |
15ee9bc7 | 1499 | root->fs_info->last_trans_committed = cur_trans->transid; |
817d52f8 | 1500 | |
2c90e5d6 | 1501 | wake_up(&cur_trans->commit_wait); |
3de4586c | 1502 | |
a4abeea4 | 1503 | spin_lock(&root->fs_info->trans_lock); |
13c5a93e | 1504 | list_del_init(&cur_trans->list); |
a4abeea4 JB |
1505 | spin_unlock(&root->fs_info->trans_lock); |
1506 | ||
78fae27e | 1507 | put_transaction(cur_trans); |
79154b1b | 1508 | put_transaction(cur_trans); |
58176a96 | 1509 | |
1abe9b8a | 1510 | trace_btrfs_transaction_commit(root); |
1511 | ||
a2de733c AJ |
1512 | btrfs_scrub_continue(root); |
1513 | ||
9ed74f2d JB |
1514 | if (current->journal_info == trans) |
1515 | current->journal_info = NULL; | |
1516 | ||
2c90e5d6 | 1517 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
24bbcf04 YZ |
1518 | |
1519 | if (current != root->fs_info->transaction_kthread) | |
1520 | btrfs_run_delayed_iputs(root); | |
1521 | ||
79154b1b | 1522 | return ret; |
49b25e05 JM |
1523 | |
1524 | cleanup_transaction: | |
1525 | btrfs_printk(root->fs_info, "Skipping commit of aborted transaction.\n"); | |
1526 | // WARN_ON(1); | |
1527 | if (current->journal_info == trans) | |
1528 | current->journal_info = NULL; | |
1529 | cleanup_transaction(trans, root); | |
1530 | ||
1531 | return ret; | |
79154b1b CM |
1532 | } |
1533 | ||
d352ac68 CM |
1534 | /* |
1535 | * interface function to delete all the snapshots we have scheduled for deletion | |
1536 | */ | |
e9d0b13b CM |
1537 | int btrfs_clean_old_snapshots(struct btrfs_root *root) |
1538 | { | |
5d4f98a2 YZ |
1539 | LIST_HEAD(list); |
1540 | struct btrfs_fs_info *fs_info = root->fs_info; | |
1541 | ||
a4abeea4 | 1542 | spin_lock(&fs_info->trans_lock); |
5d4f98a2 | 1543 | list_splice_init(&fs_info->dead_roots, &list); |
a4abeea4 | 1544 | spin_unlock(&fs_info->trans_lock); |
e9d0b13b | 1545 | |
5d4f98a2 | 1546 | while (!list_empty(&list)) { |
2c536799 JM |
1547 | int ret; |
1548 | ||
5d4f98a2 | 1549 | root = list_entry(list.next, struct btrfs_root, root_list); |
76dda93c YZ |
1550 | list_del(&root->root_list); |
1551 | ||
16cdcec7 MX |
1552 | btrfs_kill_all_delayed_nodes(root); |
1553 | ||
76dda93c YZ |
1554 | if (btrfs_header_backref_rev(root->node) < |
1555 | BTRFS_MIXED_BACKREF_REV) | |
2c536799 | 1556 | ret = btrfs_drop_snapshot(root, NULL, 0, 0); |
76dda93c | 1557 | else |
2c536799 JM |
1558 | ret =btrfs_drop_snapshot(root, NULL, 1, 0); |
1559 | BUG_ON(ret < 0); | |
e9d0b13b CM |
1560 | } |
1561 | return 0; | |
1562 | } |