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> |
8ea05e3a | 25 | #include <linux/uuid.h> |
79154b1b CM |
26 | #include "ctree.h" |
27 | #include "disk-io.h" | |
28 | #include "transaction.h" | |
925baedd | 29 | #include "locking.h" |
e02119d5 | 30 | #include "tree-log.h" |
581bb050 | 31 | #include "inode-map.h" |
733f4fbb | 32 | #include "volumes.h" |
79154b1b | 33 | |
0f7d52f4 CM |
34 | #define BTRFS_ROOT_TRANS_TAG 0 |
35 | ||
49b25e05 | 36 | void put_transaction(struct btrfs_transaction *transaction) |
79154b1b | 37 | { |
13c5a93e JB |
38 | WARN_ON(atomic_read(&transaction->use_count) == 0); |
39 | if (atomic_dec_and_test(&transaction->use_count)) { | |
a4abeea4 | 40 | BUG_ON(!list_empty(&transaction->list)); |
00f04b88 | 41 | WARN_ON(transaction->delayed_refs.root.rb_node); |
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; |
e4b50e14 DC |
103 | } else if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) { |
104 | spin_unlock(&fs_info->trans_lock); | |
7b8b92af JB |
105 | kmem_cache_free(btrfs_transaction_cachep, cur_trans); |
106 | return -EROFS; | |
79154b1b | 107 | } |
d43317dc | 108 | |
a4abeea4 JB |
109 | atomic_set(&cur_trans->num_writers, 1); |
110 | cur_trans->num_joined = 0; | |
111 | init_waitqueue_head(&cur_trans->writer_wait); | |
112 | init_waitqueue_head(&cur_trans->commit_wait); | |
113 | cur_trans->in_commit = 0; | |
114 | cur_trans->blocked = 0; | |
115 | /* | |
116 | * One for this trans handle, one so it will live on until we | |
117 | * commit the transaction. | |
118 | */ | |
119 | atomic_set(&cur_trans->use_count, 2); | |
120 | cur_trans->commit_done = 0; | |
121 | cur_trans->start_time = get_seconds(); | |
122 | ||
123 | cur_trans->delayed_refs.root = RB_ROOT; | |
124 | cur_trans->delayed_refs.num_entries = 0; | |
125 | cur_trans->delayed_refs.num_heads_ready = 0; | |
126 | cur_trans->delayed_refs.num_heads = 0; | |
127 | cur_trans->delayed_refs.flushing = 0; | |
128 | cur_trans->delayed_refs.run_delayed_start = 0; | |
20b297d6 JS |
129 | |
130 | /* | |
131 | * although the tree mod log is per file system and not per transaction, | |
132 | * the log must never go across transaction boundaries. | |
133 | */ | |
134 | smp_mb(); | |
135 | if (!list_empty(&fs_info->tree_mod_seq_list)) { | |
136 | printk(KERN_ERR "btrfs: tree_mod_seq_list not empty when " | |
137 | "creating a fresh transaction\n"); | |
138 | WARN_ON(1); | |
139 | } | |
140 | if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log)) { | |
141 | printk(KERN_ERR "btrfs: tree_mod_log rb tree not empty when " | |
142 | "creating a fresh transaction\n"); | |
143 | WARN_ON(1); | |
144 | } | |
145 | atomic_set(&fs_info->tree_mod_seq, 0); | |
146 | ||
a4abeea4 JB |
147 | spin_lock_init(&cur_trans->commit_lock); |
148 | spin_lock_init(&cur_trans->delayed_refs.lock); | |
149 | ||
150 | INIT_LIST_HEAD(&cur_trans->pending_snapshots); | |
19ae4e81 | 151 | list_add_tail(&cur_trans->list, &fs_info->trans_list); |
a4abeea4 | 152 | extent_io_tree_init(&cur_trans->dirty_pages, |
19ae4e81 JS |
153 | fs_info->btree_inode->i_mapping); |
154 | fs_info->generation++; | |
155 | cur_trans->transid = fs_info->generation; | |
156 | fs_info->running_transaction = cur_trans; | |
49b25e05 | 157 | cur_trans->aborted = 0; |
19ae4e81 | 158 | spin_unlock(&fs_info->trans_lock); |
15ee9bc7 | 159 | |
79154b1b CM |
160 | return 0; |
161 | } | |
162 | ||
d352ac68 | 163 | /* |
d397712b CM |
164 | * this does all the record keeping required to make sure that a reference |
165 | * counted root is properly recorded in a given transaction. This is required | |
166 | * to make sure the old root from before we joined the transaction is deleted | |
167 | * when the transaction commits | |
d352ac68 | 168 | */ |
7585717f | 169 | static int record_root_in_trans(struct btrfs_trans_handle *trans, |
a4abeea4 | 170 | struct btrfs_root *root) |
6702ed49 | 171 | { |
5d4f98a2 | 172 | if (root->ref_cows && root->last_trans < trans->transid) { |
6702ed49 | 173 | WARN_ON(root == root->fs_info->extent_root); |
5d4f98a2 YZ |
174 | WARN_ON(root->commit_root != root->node); |
175 | ||
7585717f CM |
176 | /* |
177 | * see below for in_trans_setup usage rules | |
178 | * we have the reloc mutex held now, so there | |
179 | * is only one writer in this function | |
180 | */ | |
181 | root->in_trans_setup = 1; | |
182 | ||
183 | /* make sure readers find in_trans_setup before | |
184 | * they find our root->last_trans update | |
185 | */ | |
186 | smp_wmb(); | |
187 | ||
a4abeea4 JB |
188 | spin_lock(&root->fs_info->fs_roots_radix_lock); |
189 | if (root->last_trans == trans->transid) { | |
190 | spin_unlock(&root->fs_info->fs_roots_radix_lock); | |
191 | return 0; | |
192 | } | |
5d4f98a2 YZ |
193 | radix_tree_tag_set(&root->fs_info->fs_roots_radix, |
194 | (unsigned long)root->root_key.objectid, | |
195 | BTRFS_ROOT_TRANS_TAG); | |
a4abeea4 | 196 | spin_unlock(&root->fs_info->fs_roots_radix_lock); |
7585717f CM |
197 | root->last_trans = trans->transid; |
198 | ||
199 | /* this is pretty tricky. We don't want to | |
200 | * take the relocation lock in btrfs_record_root_in_trans | |
201 | * unless we're really doing the first setup for this root in | |
202 | * this transaction. | |
203 | * | |
204 | * Normally we'd use root->last_trans as a flag to decide | |
205 | * if we want to take the expensive mutex. | |
206 | * | |
207 | * But, we have to set root->last_trans before we | |
208 | * init the relocation root, otherwise, we trip over warnings | |
209 | * in ctree.c. The solution used here is to flag ourselves | |
210 | * with root->in_trans_setup. When this is 1, we're still | |
211 | * fixing up the reloc trees and everyone must wait. | |
212 | * | |
213 | * When this is zero, they can trust root->last_trans and fly | |
214 | * through btrfs_record_root_in_trans without having to take the | |
215 | * lock. smp_wmb() makes sure that all the writes above are | |
216 | * done before we pop in the zero below | |
217 | */ | |
5d4f98a2 | 218 | btrfs_init_reloc_root(trans, root); |
7585717f CM |
219 | smp_wmb(); |
220 | root->in_trans_setup = 0; | |
5d4f98a2 YZ |
221 | } |
222 | return 0; | |
223 | } | |
bcc63abb | 224 | |
7585717f CM |
225 | |
226 | int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, | |
227 | struct btrfs_root *root) | |
228 | { | |
229 | if (!root->ref_cows) | |
230 | return 0; | |
231 | ||
232 | /* | |
233 | * see record_root_in_trans for comments about in_trans_setup usage | |
234 | * and barriers | |
235 | */ | |
236 | smp_rmb(); | |
237 | if (root->last_trans == trans->transid && | |
238 | !root->in_trans_setup) | |
239 | return 0; | |
240 | ||
241 | mutex_lock(&root->fs_info->reloc_mutex); | |
242 | record_root_in_trans(trans, root); | |
243 | mutex_unlock(&root->fs_info->reloc_mutex); | |
244 | ||
245 | return 0; | |
246 | } | |
247 | ||
d352ac68 CM |
248 | /* wait for commit against the current transaction to become unblocked |
249 | * when this is done, it is safe to start a new transaction, but the current | |
250 | * transaction might not be fully on disk. | |
251 | */ | |
37d1aeee | 252 | static void wait_current_trans(struct btrfs_root *root) |
79154b1b | 253 | { |
f9295749 | 254 | struct btrfs_transaction *cur_trans; |
79154b1b | 255 | |
a4abeea4 | 256 | spin_lock(&root->fs_info->trans_lock); |
f9295749 | 257 | cur_trans = root->fs_info->running_transaction; |
37d1aeee | 258 | if (cur_trans && cur_trans->blocked) { |
13c5a93e | 259 | atomic_inc(&cur_trans->use_count); |
a4abeea4 | 260 | spin_unlock(&root->fs_info->trans_lock); |
72d63ed6 LZ |
261 | |
262 | wait_event(root->fs_info->transaction_wait, | |
263 | !cur_trans->blocked); | |
f9295749 | 264 | put_transaction(cur_trans); |
a4abeea4 JB |
265 | } else { |
266 | spin_unlock(&root->fs_info->trans_lock); | |
f9295749 | 267 | } |
37d1aeee CM |
268 | } |
269 | ||
249ac1e5 JB |
270 | enum btrfs_trans_type { |
271 | TRANS_START, | |
272 | TRANS_JOIN, | |
273 | TRANS_USERSPACE, | |
0af3d00b | 274 | TRANS_JOIN_NOLOCK, |
249ac1e5 JB |
275 | }; |
276 | ||
a22285a6 YZ |
277 | static int may_wait_transaction(struct btrfs_root *root, int type) |
278 | { | |
a4abeea4 JB |
279 | if (root->fs_info->log_root_recovering) |
280 | return 0; | |
281 | ||
282 | if (type == TRANS_USERSPACE) | |
283 | return 1; | |
284 | ||
285 | if (type == TRANS_START && | |
286 | !atomic_read(&root->fs_info->open_ioctl_trans)) | |
a22285a6 | 287 | return 1; |
a4abeea4 | 288 | |
a22285a6 YZ |
289 | return 0; |
290 | } | |
291 | ||
e02119d5 | 292 | static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root, |
a22285a6 | 293 | u64 num_items, int type) |
37d1aeee | 294 | { |
a22285a6 YZ |
295 | struct btrfs_trans_handle *h; |
296 | struct btrfs_transaction *cur_trans; | |
b5009945 | 297 | u64 num_bytes = 0; |
37d1aeee | 298 | int ret; |
c5567237 | 299 | u64 qgroup_reserved = 0; |
acce952b | 300 | |
301 | if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) | |
302 | return ERR_PTR(-EROFS); | |
2a1eb461 JB |
303 | |
304 | if (current->journal_info) { | |
305 | WARN_ON(type != TRANS_JOIN && type != TRANS_JOIN_NOLOCK); | |
306 | h = current->journal_info; | |
307 | h->use_count++; | |
308 | h->orig_rsv = h->block_rsv; | |
309 | h->block_rsv = NULL; | |
310 | goto got_it; | |
311 | } | |
b5009945 JB |
312 | |
313 | /* | |
314 | * Do the reservation before we join the transaction so we can do all | |
315 | * the appropriate flushing if need be. | |
316 | */ | |
317 | if (num_items > 0 && root != root->fs_info->chunk_root) { | |
c5567237 AJ |
318 | if (root->fs_info->quota_enabled && |
319 | is_fstree(root->root_key.objectid)) { | |
320 | qgroup_reserved = num_items * root->leafsize; | |
321 | ret = btrfs_qgroup_reserve(root, qgroup_reserved); | |
322 | if (ret) | |
323 | return ERR_PTR(ret); | |
324 | } | |
325 | ||
b5009945 | 326 | num_bytes = btrfs_calc_trans_metadata_size(root, num_items); |
4a92b1b8 | 327 | ret = btrfs_block_rsv_add(root, |
b5009945 JB |
328 | &root->fs_info->trans_block_rsv, |
329 | num_bytes); | |
330 | if (ret) | |
331 | return ERR_PTR(ret); | |
332 | } | |
a22285a6 YZ |
333 | again: |
334 | h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS); | |
335 | if (!h) | |
336 | return ERR_PTR(-ENOMEM); | |
37d1aeee | 337 | |
b2b5ef5c JK |
338 | sb_start_intwrite(root->fs_info->sb); |
339 | ||
a22285a6 | 340 | if (may_wait_transaction(root, type)) |
37d1aeee | 341 | wait_current_trans(root); |
a22285a6 | 342 | |
a4abeea4 JB |
343 | do { |
344 | ret = join_transaction(root, type == TRANS_JOIN_NOLOCK); | |
345 | if (ret == -EBUSY) | |
346 | wait_current_trans(root); | |
347 | } while (ret == -EBUSY); | |
348 | ||
db5b493a | 349 | if (ret < 0) { |
b2b5ef5c | 350 | sb_end_intwrite(root->fs_info->sb); |
6e8df2ae | 351 | kmem_cache_free(btrfs_trans_handle_cachep, h); |
db5b493a TI |
352 | return ERR_PTR(ret); |
353 | } | |
0f7d52f4 | 354 | |
a22285a6 | 355 | cur_trans = root->fs_info->running_transaction; |
a22285a6 YZ |
356 | |
357 | h->transid = cur_trans->transid; | |
358 | h->transaction = cur_trans; | |
79154b1b | 359 | h->blocks_used = 0; |
a22285a6 | 360 | h->bytes_reserved = 0; |
d13603ef | 361 | h->root = root; |
56bec294 | 362 | h->delayed_ref_updates = 0; |
2a1eb461 | 363 | h->use_count = 1; |
0e721106 | 364 | h->adding_csums = 0; |
f0486c68 | 365 | h->block_rsv = NULL; |
2a1eb461 | 366 | h->orig_rsv = NULL; |
49b25e05 | 367 | h->aborted = 0; |
c5567237 | 368 | h->qgroup_reserved = qgroup_reserved; |
bed92eae AJ |
369 | h->delayed_ref_elem.seq = 0; |
370 | INIT_LIST_HEAD(&h->qgroup_ref_list); | |
b7ec40d7 | 371 | |
a22285a6 YZ |
372 | smp_mb(); |
373 | if (cur_trans->blocked && may_wait_transaction(root, type)) { | |
374 | btrfs_commit_transaction(h, root); | |
375 | goto again; | |
376 | } | |
377 | ||
b5009945 | 378 | if (num_bytes) { |
8c2a3ca2 | 379 | trace_btrfs_space_reservation(root->fs_info, "transaction", |
2bcc0328 | 380 | h->transid, num_bytes, 1); |
b5009945 JB |
381 | h->block_rsv = &root->fs_info->trans_block_rsv; |
382 | h->bytes_reserved = num_bytes; | |
a22285a6 | 383 | } |
9ed74f2d | 384 | |
2a1eb461 | 385 | got_it: |
a4abeea4 | 386 | btrfs_record_root_in_trans(h, root); |
a22285a6 YZ |
387 | |
388 | if (!current->journal_info && type != TRANS_USERSPACE) | |
389 | current->journal_info = h; | |
79154b1b CM |
390 | return h; |
391 | } | |
392 | ||
f9295749 | 393 | struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, |
a22285a6 | 394 | int num_items) |
f9295749 | 395 | { |
a22285a6 | 396 | return start_transaction(root, num_items, TRANS_START); |
f9295749 | 397 | } |
7a7eaa40 | 398 | struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root) |
f9295749 | 399 | { |
a22285a6 | 400 | return start_transaction(root, 0, TRANS_JOIN); |
f9295749 CM |
401 | } |
402 | ||
7a7eaa40 | 403 | struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root) |
0af3d00b JB |
404 | { |
405 | return start_transaction(root, 0, TRANS_JOIN_NOLOCK); | |
406 | } | |
407 | ||
7a7eaa40 | 408 | struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root) |
9ca9ee09 | 409 | { |
7a7eaa40 | 410 | return start_transaction(root, 0, TRANS_USERSPACE); |
9ca9ee09 SW |
411 | } |
412 | ||
d352ac68 | 413 | /* wait for a transaction commit to be fully complete */ |
b9c8300c | 414 | static noinline void wait_for_commit(struct btrfs_root *root, |
89ce8a63 CM |
415 | struct btrfs_transaction *commit) |
416 | { | |
72d63ed6 | 417 | wait_event(commit->commit_wait, commit->commit_done); |
89ce8a63 CM |
418 | } |
419 | ||
46204592 SW |
420 | int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid) |
421 | { | |
422 | struct btrfs_transaction *cur_trans = NULL, *t; | |
423 | int ret; | |
424 | ||
46204592 SW |
425 | ret = 0; |
426 | if (transid) { | |
427 | if (transid <= root->fs_info->last_trans_committed) | |
a4abeea4 | 428 | goto out; |
46204592 SW |
429 | |
430 | /* find specified transaction */ | |
a4abeea4 | 431 | spin_lock(&root->fs_info->trans_lock); |
46204592 SW |
432 | list_for_each_entry(t, &root->fs_info->trans_list, list) { |
433 | if (t->transid == transid) { | |
434 | cur_trans = t; | |
a4abeea4 | 435 | atomic_inc(&cur_trans->use_count); |
46204592 SW |
436 | break; |
437 | } | |
438 | if (t->transid > transid) | |
439 | break; | |
440 | } | |
a4abeea4 | 441 | spin_unlock(&root->fs_info->trans_lock); |
46204592 SW |
442 | ret = -EINVAL; |
443 | if (!cur_trans) | |
a4abeea4 | 444 | goto out; /* bad transid */ |
46204592 SW |
445 | } else { |
446 | /* find newest transaction that is committing | committed */ | |
a4abeea4 | 447 | spin_lock(&root->fs_info->trans_lock); |
46204592 SW |
448 | list_for_each_entry_reverse(t, &root->fs_info->trans_list, |
449 | list) { | |
450 | if (t->in_commit) { | |
451 | if (t->commit_done) | |
3473f3c0 | 452 | break; |
46204592 | 453 | cur_trans = t; |
a4abeea4 | 454 | atomic_inc(&cur_trans->use_count); |
46204592 SW |
455 | break; |
456 | } | |
457 | } | |
a4abeea4 | 458 | spin_unlock(&root->fs_info->trans_lock); |
46204592 | 459 | if (!cur_trans) |
a4abeea4 | 460 | goto out; /* nothing committing|committed */ |
46204592 SW |
461 | } |
462 | ||
46204592 SW |
463 | wait_for_commit(root, cur_trans); |
464 | ||
46204592 SW |
465 | put_transaction(cur_trans); |
466 | ret = 0; | |
a4abeea4 | 467 | out: |
46204592 SW |
468 | return ret; |
469 | } | |
470 | ||
37d1aeee CM |
471 | void btrfs_throttle(struct btrfs_root *root) |
472 | { | |
a4abeea4 | 473 | if (!atomic_read(&root->fs_info->open_ioctl_trans)) |
9ca9ee09 | 474 | wait_current_trans(root); |
37d1aeee CM |
475 | } |
476 | ||
8929ecfa YZ |
477 | static int should_end_transaction(struct btrfs_trans_handle *trans, |
478 | struct btrfs_root *root) | |
479 | { | |
480 | int ret; | |
36ba022a JB |
481 | |
482 | ret = btrfs_block_rsv_check(root, &root->fs_info->global_block_rsv, 5); | |
8929ecfa YZ |
483 | return ret ? 1 : 0; |
484 | } | |
485 | ||
486 | int btrfs_should_end_transaction(struct btrfs_trans_handle *trans, | |
487 | struct btrfs_root *root) | |
488 | { | |
489 | struct btrfs_transaction *cur_trans = trans->transaction; | |
490 | int updates; | |
49b25e05 | 491 | int err; |
8929ecfa | 492 | |
a4abeea4 | 493 | smp_mb(); |
8929ecfa YZ |
494 | if (cur_trans->blocked || cur_trans->delayed_refs.flushing) |
495 | return 1; | |
496 | ||
497 | updates = trans->delayed_ref_updates; | |
498 | trans->delayed_ref_updates = 0; | |
49b25e05 JM |
499 | if (updates) { |
500 | err = btrfs_run_delayed_refs(trans, root, updates); | |
501 | if (err) /* Error code will also eval true */ | |
502 | return err; | |
503 | } | |
8929ecfa YZ |
504 | |
505 | return should_end_transaction(trans, root); | |
506 | } | |
507 | ||
89ce8a63 | 508 | static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, |
0af3d00b | 509 | struct btrfs_root *root, int throttle, int lock) |
79154b1b | 510 | { |
8929ecfa | 511 | struct btrfs_transaction *cur_trans = trans->transaction; |
ab78c84d | 512 | struct btrfs_fs_info *info = root->fs_info; |
c3e69d58 | 513 | int count = 0; |
4edc2ca3 | 514 | int err = 0; |
c3e69d58 | 515 | |
2a1eb461 JB |
516 | if (--trans->use_count) { |
517 | trans->block_rsv = trans->orig_rsv; | |
518 | return 0; | |
519 | } | |
520 | ||
edf39272 JS |
521 | /* |
522 | * do the qgroup accounting as early as possible | |
523 | */ | |
524 | err = btrfs_delayed_refs_qgroup_accounting(trans, info); | |
525 | ||
b24e03db | 526 | btrfs_trans_release_metadata(trans, root); |
4c13d758 | 527 | trans->block_rsv = NULL; |
d13603ef AJ |
528 | /* |
529 | * the same root has to be passed to start_transaction and | |
530 | * end_transaction. Subvolume quota depends on this. | |
531 | */ | |
532 | WARN_ON(trans->root != root); | |
c5567237 AJ |
533 | |
534 | if (trans->qgroup_reserved) { | |
535 | btrfs_qgroup_free(root, trans->qgroup_reserved); | |
536 | trans->qgroup_reserved = 0; | |
537 | } | |
538 | ||
203bf287 | 539 | while (count < 2) { |
c3e69d58 CM |
540 | unsigned long cur = trans->delayed_ref_updates; |
541 | trans->delayed_ref_updates = 0; | |
542 | if (cur && | |
543 | trans->transaction->delayed_refs.num_heads_ready > 64) { | |
544 | trans->delayed_ref_updates = 0; | |
545 | btrfs_run_delayed_refs(trans, root, cur); | |
546 | } else { | |
547 | break; | |
548 | } | |
549 | count++; | |
56bec294 | 550 | } |
0e721106 JB |
551 | btrfs_trans_release_metadata(trans, root); |
552 | trans->block_rsv = NULL; | |
56bec294 | 553 | |
b2b5ef5c JK |
554 | sb_end_intwrite(root->fs_info->sb); |
555 | ||
a4abeea4 JB |
556 | if (lock && !atomic_read(&root->fs_info->open_ioctl_trans) && |
557 | should_end_transaction(trans, root)) { | |
8929ecfa | 558 | trans->transaction->blocked = 1; |
a4abeea4 JB |
559 | smp_wmb(); |
560 | } | |
8929ecfa | 561 | |
0af3d00b | 562 | if (lock && cur_trans->blocked && !cur_trans->in_commit) { |
81317fde JB |
563 | if (throttle) { |
564 | /* | |
565 | * We may race with somebody else here so end up having | |
566 | * to call end_transaction on ourselves again, so inc | |
567 | * our use_count. | |
568 | */ | |
569 | trans->use_count++; | |
8929ecfa | 570 | return btrfs_commit_transaction(trans, root); |
81317fde | 571 | } else { |
8929ecfa | 572 | wake_up_process(info->transaction_kthread); |
81317fde | 573 | } |
8929ecfa YZ |
574 | } |
575 | ||
8929ecfa | 576 | WARN_ON(cur_trans != info->running_transaction); |
13c5a93e JB |
577 | WARN_ON(atomic_read(&cur_trans->num_writers) < 1); |
578 | atomic_dec(&cur_trans->num_writers); | |
89ce8a63 | 579 | |
99d16cbc | 580 | smp_mb(); |
79154b1b CM |
581 | if (waitqueue_active(&cur_trans->writer_wait)) |
582 | wake_up(&cur_trans->writer_wait); | |
79154b1b | 583 | put_transaction(cur_trans); |
9ed74f2d JB |
584 | |
585 | if (current->journal_info == trans) | |
586 | current->journal_info = NULL; | |
ab78c84d | 587 | |
24bbcf04 YZ |
588 | if (throttle) |
589 | btrfs_run_delayed_iputs(root); | |
590 | ||
49b25e05 JM |
591 | if (trans->aborted || |
592 | root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) { | |
4edc2ca3 | 593 | err = -EIO; |
49b25e05 | 594 | } |
edf39272 | 595 | assert_qgroups_uptodate(trans); |
49b25e05 | 596 | |
4edc2ca3 DJ |
597 | memset(trans, 0, sizeof(*trans)); |
598 | kmem_cache_free(btrfs_trans_handle_cachep, trans); | |
599 | return err; | |
79154b1b CM |
600 | } |
601 | ||
89ce8a63 CM |
602 | int btrfs_end_transaction(struct btrfs_trans_handle *trans, |
603 | struct btrfs_root *root) | |
604 | { | |
16cdcec7 MX |
605 | int ret; |
606 | ||
607 | ret = __btrfs_end_transaction(trans, root, 0, 1); | |
608 | if (ret) | |
609 | return ret; | |
610 | return 0; | |
89ce8a63 CM |
611 | } |
612 | ||
613 | int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans, | |
614 | struct btrfs_root *root) | |
615 | { | |
16cdcec7 MX |
616 | int ret; |
617 | ||
618 | ret = __btrfs_end_transaction(trans, root, 1, 1); | |
619 | if (ret) | |
620 | return ret; | |
621 | return 0; | |
0af3d00b JB |
622 | } |
623 | ||
624 | int btrfs_end_transaction_nolock(struct btrfs_trans_handle *trans, | |
625 | struct btrfs_root *root) | |
626 | { | |
16cdcec7 MX |
627 | int ret; |
628 | ||
629 | ret = __btrfs_end_transaction(trans, root, 0, 0); | |
630 | if (ret) | |
631 | return ret; | |
632 | return 0; | |
633 | } | |
634 | ||
635 | int btrfs_end_transaction_dmeta(struct btrfs_trans_handle *trans, | |
636 | struct btrfs_root *root) | |
637 | { | |
638 | return __btrfs_end_transaction(trans, root, 1, 1); | |
89ce8a63 CM |
639 | } |
640 | ||
d352ac68 CM |
641 | /* |
642 | * when btree blocks are allocated, they have some corresponding bits set for | |
643 | * them in one of two extent_io trees. This is used to make sure all of | |
690587d1 | 644 | * those extents are sent to disk but does not wait on them |
d352ac68 | 645 | */ |
690587d1 | 646 | int btrfs_write_marked_extents(struct btrfs_root *root, |
8cef4e16 | 647 | struct extent_io_tree *dirty_pages, int mark) |
79154b1b | 648 | { |
777e6bd7 | 649 | int err = 0; |
7c4452b9 | 650 | int werr = 0; |
1728366e | 651 | struct address_space *mapping = root->fs_info->btree_inode->i_mapping; |
777e6bd7 | 652 | u64 start = 0; |
5f39d397 | 653 | u64 end; |
7c4452b9 | 654 | |
1728366e JB |
655 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
656 | mark)) { | |
657 | convert_extent_bit(dirty_pages, start, end, EXTENT_NEED_WAIT, mark, | |
658 | GFP_NOFS); | |
659 | err = filemap_fdatawrite_range(mapping, start, end); | |
660 | if (err) | |
661 | werr = err; | |
662 | cond_resched(); | |
663 | start = end + 1; | |
7c4452b9 | 664 | } |
690587d1 CM |
665 | if (err) |
666 | werr = err; | |
667 | return werr; | |
668 | } | |
669 | ||
670 | /* | |
671 | * when btree blocks are allocated, they have some corresponding bits set for | |
672 | * them in one of two extent_io trees. This is used to make sure all of | |
673 | * those extents are on disk for transaction or log commit. We wait | |
674 | * on all the pages and clear them from the dirty pages state tree | |
675 | */ | |
676 | int btrfs_wait_marked_extents(struct btrfs_root *root, | |
8cef4e16 | 677 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 | 678 | { |
690587d1 CM |
679 | int err = 0; |
680 | int werr = 0; | |
1728366e | 681 | struct address_space *mapping = root->fs_info->btree_inode->i_mapping; |
690587d1 CM |
682 | u64 start = 0; |
683 | u64 end; | |
777e6bd7 | 684 | |
1728366e JB |
685 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
686 | EXTENT_NEED_WAIT)) { | |
687 | clear_extent_bits(dirty_pages, start, end, EXTENT_NEED_WAIT, GFP_NOFS); | |
688 | err = filemap_fdatawait_range(mapping, start, end); | |
689 | if (err) | |
690 | werr = err; | |
691 | cond_resched(); | |
692 | start = end + 1; | |
777e6bd7 | 693 | } |
7c4452b9 CM |
694 | if (err) |
695 | werr = err; | |
696 | return werr; | |
79154b1b CM |
697 | } |
698 | ||
690587d1 CM |
699 | /* |
700 | * when btree blocks are allocated, they have some corresponding bits set for | |
701 | * them in one of two extent_io trees. This is used to make sure all of | |
702 | * those extents are on disk for transaction or log commit | |
703 | */ | |
704 | int btrfs_write_and_wait_marked_extents(struct btrfs_root *root, | |
8cef4e16 | 705 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 CM |
706 | { |
707 | int ret; | |
708 | int ret2; | |
709 | ||
8cef4e16 YZ |
710 | ret = btrfs_write_marked_extents(root, dirty_pages, mark); |
711 | ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark); | |
bf0da8c1 CM |
712 | |
713 | if (ret) | |
714 | return ret; | |
715 | if (ret2) | |
716 | return ret2; | |
717 | return 0; | |
690587d1 CM |
718 | } |
719 | ||
d0c803c4 CM |
720 | int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans, |
721 | struct btrfs_root *root) | |
722 | { | |
723 | if (!trans || !trans->transaction) { | |
724 | struct inode *btree_inode; | |
725 | btree_inode = root->fs_info->btree_inode; | |
726 | return filemap_write_and_wait(btree_inode->i_mapping); | |
727 | } | |
728 | return btrfs_write_and_wait_marked_extents(root, | |
8cef4e16 YZ |
729 | &trans->transaction->dirty_pages, |
730 | EXTENT_DIRTY); | |
d0c803c4 CM |
731 | } |
732 | ||
d352ac68 CM |
733 | /* |
734 | * this is used to update the root pointer in the tree of tree roots. | |
735 | * | |
736 | * But, in the case of the extent allocation tree, updating the root | |
737 | * pointer may allocate blocks which may change the root of the extent | |
738 | * allocation tree. | |
739 | * | |
740 | * So, this loops and repeats and makes sure the cowonly root didn't | |
741 | * change while the root pointer was being updated in the metadata. | |
742 | */ | |
0b86a832 CM |
743 | static int update_cowonly_root(struct btrfs_trans_handle *trans, |
744 | struct btrfs_root *root) | |
79154b1b CM |
745 | { |
746 | int ret; | |
0b86a832 | 747 | u64 old_root_bytenr; |
86b9f2ec | 748 | u64 old_root_used; |
0b86a832 | 749 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
79154b1b | 750 | |
86b9f2ec | 751 | old_root_used = btrfs_root_used(&root->root_item); |
0b86a832 | 752 | btrfs_write_dirty_block_groups(trans, root); |
56bec294 | 753 | |
d397712b | 754 | while (1) { |
0b86a832 | 755 | old_root_bytenr = btrfs_root_bytenr(&root->root_item); |
86b9f2ec YZ |
756 | if (old_root_bytenr == root->node->start && |
757 | old_root_used == btrfs_root_used(&root->root_item)) | |
79154b1b | 758 | break; |
87ef2bb4 | 759 | |
5d4f98a2 | 760 | btrfs_set_root_node(&root->root_item, root->node); |
79154b1b | 761 | ret = btrfs_update_root(trans, tree_root, |
0b86a832 CM |
762 | &root->root_key, |
763 | &root->root_item); | |
49b25e05 JM |
764 | if (ret) |
765 | return ret; | |
56bec294 | 766 | |
86b9f2ec | 767 | old_root_used = btrfs_root_used(&root->root_item); |
4a8c9a62 | 768 | ret = btrfs_write_dirty_block_groups(trans, root); |
49b25e05 JM |
769 | if (ret) |
770 | return ret; | |
0b86a832 | 771 | } |
276e680d YZ |
772 | |
773 | if (root != root->fs_info->extent_root) | |
774 | switch_commit_root(root); | |
775 | ||
0b86a832 CM |
776 | return 0; |
777 | } | |
778 | ||
d352ac68 CM |
779 | /* |
780 | * update all the cowonly tree roots on disk | |
49b25e05 JM |
781 | * |
782 | * The error handling in this function may not be obvious. Any of the | |
783 | * failures will cause the file system to go offline. We still need | |
784 | * to clean up the delayed refs. | |
d352ac68 | 785 | */ |
5d4f98a2 YZ |
786 | static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans, |
787 | struct btrfs_root *root) | |
0b86a832 CM |
788 | { |
789 | struct btrfs_fs_info *fs_info = root->fs_info; | |
790 | struct list_head *next; | |
84234f3a | 791 | struct extent_buffer *eb; |
56bec294 | 792 | int ret; |
84234f3a | 793 | |
56bec294 | 794 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
49b25e05 JM |
795 | if (ret) |
796 | return ret; | |
87ef2bb4 | 797 | |
84234f3a | 798 | eb = btrfs_lock_root_node(fs_info->tree_root); |
49b25e05 JM |
799 | ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, |
800 | 0, &eb); | |
84234f3a YZ |
801 | btrfs_tree_unlock(eb); |
802 | free_extent_buffer(eb); | |
0b86a832 | 803 | |
49b25e05 JM |
804 | if (ret) |
805 | return ret; | |
806 | ||
56bec294 | 807 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
49b25e05 JM |
808 | if (ret) |
809 | return ret; | |
87ef2bb4 | 810 | |
733f4fbb SB |
811 | ret = btrfs_run_dev_stats(trans, root->fs_info); |
812 | BUG_ON(ret); | |
813 | ||
546adb0d JS |
814 | ret = btrfs_run_qgroups(trans, root->fs_info); |
815 | BUG_ON(ret); | |
816 | ||
817 | /* run_qgroups might have added some more refs */ | |
818 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); | |
819 | BUG_ON(ret); | |
820 | ||
d397712b | 821 | while (!list_empty(&fs_info->dirty_cowonly_roots)) { |
0b86a832 CM |
822 | next = fs_info->dirty_cowonly_roots.next; |
823 | list_del_init(next); | |
824 | root = list_entry(next, struct btrfs_root, dirty_list); | |
87ef2bb4 | 825 | |
49b25e05 JM |
826 | ret = update_cowonly_root(trans, root); |
827 | if (ret) | |
828 | return ret; | |
79154b1b | 829 | } |
276e680d YZ |
830 | |
831 | down_write(&fs_info->extent_commit_sem); | |
832 | switch_commit_root(fs_info->extent_root); | |
833 | up_write(&fs_info->extent_commit_sem); | |
834 | ||
79154b1b CM |
835 | return 0; |
836 | } | |
837 | ||
d352ac68 CM |
838 | /* |
839 | * dead roots are old snapshots that need to be deleted. This allocates | |
840 | * a dirty root struct and adds it into the list of dead roots that need to | |
841 | * be deleted | |
842 | */ | |
5d4f98a2 | 843 | int btrfs_add_dead_root(struct btrfs_root *root) |
5eda7b5e | 844 | { |
a4abeea4 | 845 | spin_lock(&root->fs_info->trans_lock); |
5d4f98a2 | 846 | list_add(&root->root_list, &root->fs_info->dead_roots); |
a4abeea4 | 847 | spin_unlock(&root->fs_info->trans_lock); |
5eda7b5e CM |
848 | return 0; |
849 | } | |
850 | ||
d352ac68 | 851 | /* |
5d4f98a2 | 852 | * update all the cowonly tree roots on disk |
d352ac68 | 853 | */ |
5d4f98a2 YZ |
854 | static noinline int commit_fs_roots(struct btrfs_trans_handle *trans, |
855 | struct btrfs_root *root) | |
0f7d52f4 | 856 | { |
0f7d52f4 | 857 | struct btrfs_root *gang[8]; |
5d4f98a2 | 858 | struct btrfs_fs_info *fs_info = root->fs_info; |
0f7d52f4 CM |
859 | int i; |
860 | int ret; | |
54aa1f4d CM |
861 | int err = 0; |
862 | ||
a4abeea4 | 863 | spin_lock(&fs_info->fs_roots_radix_lock); |
d397712b | 864 | while (1) { |
5d4f98a2 YZ |
865 | ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, |
866 | (void **)gang, 0, | |
0f7d52f4 CM |
867 | ARRAY_SIZE(gang), |
868 | BTRFS_ROOT_TRANS_TAG); | |
869 | if (ret == 0) | |
870 | break; | |
871 | for (i = 0; i < ret; i++) { | |
872 | root = gang[i]; | |
5d4f98a2 YZ |
873 | radix_tree_tag_clear(&fs_info->fs_roots_radix, |
874 | (unsigned long)root->root_key.objectid, | |
875 | BTRFS_ROOT_TRANS_TAG); | |
a4abeea4 | 876 | spin_unlock(&fs_info->fs_roots_radix_lock); |
31153d81 | 877 | |
e02119d5 | 878 | btrfs_free_log(trans, root); |
5d4f98a2 | 879 | btrfs_update_reloc_root(trans, root); |
d68fc57b | 880 | btrfs_orphan_commit_root(trans, root); |
bcc63abb | 881 | |
82d5902d LZ |
882 | btrfs_save_ino_cache(root, trans); |
883 | ||
f1ebcc74 LB |
884 | /* see comments in should_cow_block() */ |
885 | root->force_cow = 0; | |
886 | smp_wmb(); | |
887 | ||
978d910d | 888 | if (root->commit_root != root->node) { |
581bb050 | 889 | mutex_lock(&root->fs_commit_mutex); |
817d52f8 | 890 | switch_commit_root(root); |
581bb050 LZ |
891 | btrfs_unpin_free_ino(root); |
892 | mutex_unlock(&root->fs_commit_mutex); | |
893 | ||
978d910d YZ |
894 | btrfs_set_root_node(&root->root_item, |
895 | root->node); | |
896 | } | |
5d4f98a2 | 897 | |
5d4f98a2 | 898 | err = btrfs_update_root(trans, fs_info->tree_root, |
0f7d52f4 CM |
899 | &root->root_key, |
900 | &root->root_item); | |
a4abeea4 | 901 | spin_lock(&fs_info->fs_roots_radix_lock); |
54aa1f4d CM |
902 | if (err) |
903 | break; | |
0f7d52f4 CM |
904 | } |
905 | } | |
a4abeea4 | 906 | spin_unlock(&fs_info->fs_roots_radix_lock); |
54aa1f4d | 907 | return err; |
0f7d52f4 CM |
908 | } |
909 | ||
d352ac68 CM |
910 | /* |
911 | * defrag a given btree. If cacheonly == 1, this won't read from the disk, | |
912 | * otherwise every leaf in the btree is read and defragged. | |
913 | */ | |
e9d0b13b CM |
914 | int btrfs_defrag_root(struct btrfs_root *root, int cacheonly) |
915 | { | |
916 | struct btrfs_fs_info *info = root->fs_info; | |
e9d0b13b | 917 | struct btrfs_trans_handle *trans; |
8929ecfa | 918 | int ret; |
d3c2fdcf | 919 | unsigned long nr; |
e9d0b13b | 920 | |
8929ecfa | 921 | if (xchg(&root->defrag_running, 1)) |
e9d0b13b | 922 | return 0; |
8929ecfa | 923 | |
6b80053d | 924 | while (1) { |
8929ecfa YZ |
925 | trans = btrfs_start_transaction(root, 0); |
926 | if (IS_ERR(trans)) | |
927 | return PTR_ERR(trans); | |
928 | ||
e9d0b13b | 929 | ret = btrfs_defrag_leaves(trans, root, cacheonly); |
8929ecfa | 930 | |
d3c2fdcf | 931 | nr = trans->blocks_used; |
e9d0b13b | 932 | btrfs_end_transaction(trans, root); |
d3c2fdcf | 933 | btrfs_btree_balance_dirty(info->tree_root, nr); |
e9d0b13b CM |
934 | cond_resched(); |
935 | ||
7841cb28 | 936 | if (btrfs_fs_closing(root->fs_info) || ret != -EAGAIN) |
e9d0b13b CM |
937 | break; |
938 | } | |
939 | root->defrag_running = 0; | |
8929ecfa | 940 | return ret; |
e9d0b13b CM |
941 | } |
942 | ||
d352ac68 CM |
943 | /* |
944 | * new snapshots need to be created at a very specific time in the | |
945 | * transaction commit. This does the actual creation | |
946 | */ | |
80b6794d | 947 | static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, |
3063d29f CM |
948 | struct btrfs_fs_info *fs_info, |
949 | struct btrfs_pending_snapshot *pending) | |
950 | { | |
951 | struct btrfs_key key; | |
80b6794d | 952 | struct btrfs_root_item *new_root_item; |
3063d29f CM |
953 | struct btrfs_root *tree_root = fs_info->tree_root; |
954 | struct btrfs_root *root = pending->root; | |
6bdb72de | 955 | struct btrfs_root *parent_root; |
98c9942a | 956 | struct btrfs_block_rsv *rsv; |
6bdb72de | 957 | struct inode *parent_inode; |
6a912213 | 958 | struct dentry *parent; |
a22285a6 | 959 | struct dentry *dentry; |
3063d29f | 960 | struct extent_buffer *tmp; |
925baedd | 961 | struct extent_buffer *old; |
8ea05e3a | 962 | struct timespec cur_time = CURRENT_TIME; |
3063d29f | 963 | int ret; |
d68fc57b | 964 | u64 to_reserve = 0; |
6bdb72de | 965 | u64 index = 0; |
a22285a6 | 966 | u64 objectid; |
b83cc969 | 967 | u64 root_flags; |
8ea05e3a | 968 | uuid_le new_uuid; |
3063d29f | 969 | |
80b6794d CM |
970 | new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS); |
971 | if (!new_root_item) { | |
49b25e05 | 972 | ret = pending->error = -ENOMEM; |
6fa9700e | 973 | goto root_item_alloc_fail; |
80b6794d | 974 | } |
a22285a6 | 975 | |
581bb050 | 976 | ret = btrfs_find_free_objectid(tree_root, &objectid); |
a22285a6 YZ |
977 | if (ret) { |
978 | pending->error = ret; | |
6fa9700e | 979 | goto no_free_objectid; |
a22285a6 | 980 | } |
3063d29f | 981 | |
3fd0a558 | 982 | btrfs_reloc_pre_snapshot(trans, pending, &to_reserve); |
d68fc57b YZ |
983 | |
984 | if (to_reserve > 0) { | |
62f30c54 MX |
985 | ret = btrfs_block_rsv_add_noflush(root, &pending->block_rsv, |
986 | to_reserve); | |
d68fc57b YZ |
987 | if (ret) { |
988 | pending->error = ret; | |
6fa9700e | 989 | goto no_free_objectid; |
d68fc57b YZ |
990 | } |
991 | } | |
992 | ||
6f72c7e2 AJ |
993 | ret = btrfs_qgroup_inherit(trans, fs_info, root->root_key.objectid, |
994 | objectid, pending->inherit); | |
6f72c7e2 AJ |
995 | if (ret) { |
996 | pending->error = ret; | |
6fa9700e | 997 | goto no_free_objectid; |
6f72c7e2 AJ |
998 | } |
999 | ||
3063d29f | 1000 | key.objectid = objectid; |
a22285a6 YZ |
1001 | key.offset = (u64)-1; |
1002 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3063d29f | 1003 | |
6fa9700e | 1004 | rsv = trans->block_rsv; |
a22285a6 | 1005 | trans->block_rsv = &pending->block_rsv; |
3de4586c | 1006 | |
a22285a6 | 1007 | dentry = pending->dentry; |
6a912213 JB |
1008 | parent = dget_parent(dentry); |
1009 | parent_inode = parent->d_inode; | |
a22285a6 | 1010 | parent_root = BTRFS_I(parent_inode)->root; |
7585717f | 1011 | record_root_in_trans(trans, parent_root); |
a22285a6 | 1012 | |
3063d29f CM |
1013 | /* |
1014 | * insert the directory item | |
1015 | */ | |
3de4586c | 1016 | ret = btrfs_set_inode_index(parent_inode, &index); |
49b25e05 | 1017 | BUG_ON(ret); /* -ENOMEM */ |
0660b5af | 1018 | ret = btrfs_insert_dir_item(trans, parent_root, |
a22285a6 | 1019 | dentry->d_name.name, dentry->d_name.len, |
16cdcec7 | 1020 | parent_inode, &key, |
a22285a6 | 1021 | BTRFS_FT_DIR, index); |
79787eaa | 1022 | if (ret == -EEXIST) { |
fe66a05a | 1023 | pending->error = -EEXIST; |
fe66a05a | 1024 | goto fail; |
79787eaa | 1025 | } else if (ret) { |
6fa9700e | 1026 | goto abort_trans; |
79787eaa | 1027 | } |
0660b5af | 1028 | |
a22285a6 YZ |
1029 | btrfs_i_size_write(parent_inode, parent_inode->i_size + |
1030 | dentry->d_name.len * 2); | |
c0f62ded | 1031 | parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME; |
52c26179 | 1032 | ret = btrfs_update_inode(trans, parent_root, parent_inode); |
49b25e05 | 1033 | if (ret) |
6fa9700e | 1034 | goto abort_trans; |
52c26179 | 1035 | |
e999376f CM |
1036 | /* |
1037 | * pull in the delayed directory update | |
1038 | * and the delayed inode item | |
1039 | * otherwise we corrupt the FS during | |
1040 | * snapshot | |
1041 | */ | |
1042 | ret = btrfs_run_delayed_items(trans, root); | |
6fa9700e MX |
1043 | if (ret) /* Transaction aborted */ |
1044 | goto abort_trans; | |
e999376f | 1045 | |
7585717f | 1046 | record_root_in_trans(trans, root); |
6bdb72de SW |
1047 | btrfs_set_root_last_snapshot(&root->root_item, trans->transid); |
1048 | memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); | |
08fe4db1 | 1049 | btrfs_check_and_init_root_item(new_root_item); |
6bdb72de | 1050 | |
b83cc969 LZ |
1051 | root_flags = btrfs_root_flags(new_root_item); |
1052 | if (pending->readonly) | |
1053 | root_flags |= BTRFS_ROOT_SUBVOL_RDONLY; | |
1054 | else | |
1055 | root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY; | |
1056 | btrfs_set_root_flags(new_root_item, root_flags); | |
1057 | ||
8ea05e3a AB |
1058 | btrfs_set_root_generation_v2(new_root_item, |
1059 | trans->transid); | |
1060 | uuid_le_gen(&new_uuid); | |
1061 | memcpy(new_root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE); | |
1062 | memcpy(new_root_item->parent_uuid, root->root_item.uuid, | |
1063 | BTRFS_UUID_SIZE); | |
1064 | new_root_item->otime.sec = cpu_to_le64(cur_time.tv_sec); | |
dadd1105 | 1065 | new_root_item->otime.nsec = cpu_to_le32(cur_time.tv_nsec); |
8ea05e3a AB |
1066 | btrfs_set_root_otransid(new_root_item, trans->transid); |
1067 | memset(&new_root_item->stime, 0, sizeof(new_root_item->stime)); | |
1068 | memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime)); | |
1069 | btrfs_set_root_stransid(new_root_item, 0); | |
1070 | btrfs_set_root_rtransid(new_root_item, 0); | |
1071 | ||
6bdb72de | 1072 | old = btrfs_lock_root_node(root); |
49b25e05 | 1073 | ret = btrfs_cow_block(trans, root, old, NULL, 0, &old); |
79787eaa JM |
1074 | if (ret) { |
1075 | btrfs_tree_unlock(old); | |
1076 | free_extent_buffer(old); | |
6fa9700e | 1077 | goto abort_trans; |
79787eaa | 1078 | } |
49b25e05 | 1079 | |
6bdb72de SW |
1080 | btrfs_set_lock_blocking(old); |
1081 | ||
49b25e05 | 1082 | ret = btrfs_copy_root(trans, root, old, &tmp, objectid); |
79787eaa | 1083 | /* clean up in any case */ |
6bdb72de SW |
1084 | btrfs_tree_unlock(old); |
1085 | free_extent_buffer(old); | |
79787eaa | 1086 | if (ret) |
6fa9700e | 1087 | goto abort_trans; |
6bdb72de | 1088 | |
f1ebcc74 LB |
1089 | /* see comments in should_cow_block() */ |
1090 | root->force_cow = 1; | |
1091 | smp_wmb(); | |
1092 | ||
6bdb72de | 1093 | btrfs_set_root_node(new_root_item, tmp); |
a22285a6 YZ |
1094 | /* record when the snapshot was created in key.offset */ |
1095 | key.offset = trans->transid; | |
1096 | ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); | |
6bdb72de SW |
1097 | btrfs_tree_unlock(tmp); |
1098 | free_extent_buffer(tmp); | |
49b25e05 | 1099 | if (ret) |
6fa9700e | 1100 | goto abort_trans; |
6bdb72de | 1101 | |
a22285a6 YZ |
1102 | /* |
1103 | * insert root back/forward references | |
1104 | */ | |
1105 | ret = btrfs_add_root_ref(trans, tree_root, objectid, | |
0660b5af | 1106 | parent_root->root_key.objectid, |
33345d01 | 1107 | btrfs_ino(parent_inode), index, |
a22285a6 | 1108 | dentry->d_name.name, dentry->d_name.len); |
49b25e05 | 1109 | if (ret) |
6fa9700e | 1110 | goto abort_trans; |
0660b5af | 1111 | |
a22285a6 YZ |
1112 | key.offset = (u64)-1; |
1113 | pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key); | |
79787eaa JM |
1114 | if (IS_ERR(pending->snap)) { |
1115 | ret = PTR_ERR(pending->snap); | |
49b25e05 | 1116 | goto abort_trans; |
79787eaa | 1117 | } |
d68fc57b | 1118 | |
49b25e05 JM |
1119 | ret = btrfs_reloc_post_snapshot(trans, pending); |
1120 | if (ret) | |
1121 | goto abort_trans; | |
361048f5 MX |
1122 | |
1123 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); | |
1124 | if (ret) | |
1125 | goto abort_trans; | |
3063d29f | 1126 | fail: |
6fa9700e | 1127 | dput(parent); |
98c9942a | 1128 | trans->block_rsv = rsv; |
6fa9700e MX |
1129 | no_free_objectid: |
1130 | kfree(new_root_item); | |
1131 | root_item_alloc_fail: | |
a22285a6 | 1132 | btrfs_block_rsv_release(root, &pending->block_rsv, (u64)-1); |
49b25e05 JM |
1133 | return ret; |
1134 | ||
1135 | abort_trans: | |
1136 | btrfs_abort_transaction(trans, root, ret); | |
1137 | goto fail; | |
3063d29f CM |
1138 | } |
1139 | ||
d352ac68 CM |
1140 | /* |
1141 | * create all the snapshots we've scheduled for creation | |
1142 | */ | |
80b6794d CM |
1143 | static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans, |
1144 | struct btrfs_fs_info *fs_info) | |
3de4586c CM |
1145 | { |
1146 | struct btrfs_pending_snapshot *pending; | |
1147 | struct list_head *head = &trans->transaction->pending_snapshots; | |
3de4586c | 1148 | |
fe66a05a CM |
1149 | list_for_each_entry(pending, head, list) |
1150 | create_pending_snapshot(trans, fs_info, pending); | |
3de4586c CM |
1151 | return 0; |
1152 | } | |
1153 | ||
5d4f98a2 YZ |
1154 | static void update_super_roots(struct btrfs_root *root) |
1155 | { | |
1156 | struct btrfs_root_item *root_item; | |
1157 | struct btrfs_super_block *super; | |
1158 | ||
6c41761f | 1159 | super = root->fs_info->super_copy; |
5d4f98a2 YZ |
1160 | |
1161 | root_item = &root->fs_info->chunk_root->root_item; | |
1162 | super->chunk_root = root_item->bytenr; | |
1163 | super->chunk_root_generation = root_item->generation; | |
1164 | super->chunk_root_level = root_item->level; | |
1165 | ||
1166 | root_item = &root->fs_info->tree_root->root_item; | |
1167 | super->root = root_item->bytenr; | |
1168 | super->generation = root_item->generation; | |
1169 | super->root_level = root_item->level; | |
73bc1876 | 1170 | if (btrfs_test_opt(root, SPACE_CACHE)) |
0af3d00b | 1171 | super->cache_generation = root_item->generation; |
5d4f98a2 YZ |
1172 | } |
1173 | ||
f36f3042 CM |
1174 | int btrfs_transaction_in_commit(struct btrfs_fs_info *info) |
1175 | { | |
1176 | int ret = 0; | |
a4abeea4 | 1177 | spin_lock(&info->trans_lock); |
f36f3042 CM |
1178 | if (info->running_transaction) |
1179 | ret = info->running_transaction->in_commit; | |
a4abeea4 | 1180 | spin_unlock(&info->trans_lock); |
f36f3042 CM |
1181 | return ret; |
1182 | } | |
1183 | ||
8929ecfa YZ |
1184 | int btrfs_transaction_blocked(struct btrfs_fs_info *info) |
1185 | { | |
1186 | int ret = 0; | |
a4abeea4 | 1187 | spin_lock(&info->trans_lock); |
8929ecfa YZ |
1188 | if (info->running_transaction) |
1189 | ret = info->running_transaction->blocked; | |
a4abeea4 | 1190 | spin_unlock(&info->trans_lock); |
8929ecfa YZ |
1191 | return ret; |
1192 | } | |
1193 | ||
bb9c12c9 SW |
1194 | /* |
1195 | * wait for the current transaction commit to start and block subsequent | |
1196 | * transaction joins | |
1197 | */ | |
1198 | static void wait_current_trans_commit_start(struct btrfs_root *root, | |
1199 | struct btrfs_transaction *trans) | |
1200 | { | |
72d63ed6 | 1201 | wait_event(root->fs_info->transaction_blocked_wait, trans->in_commit); |
bb9c12c9 SW |
1202 | } |
1203 | ||
1204 | /* | |
1205 | * wait for the current transaction to start and then become unblocked. | |
1206 | * caller holds ref. | |
1207 | */ | |
1208 | static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root, | |
1209 | struct btrfs_transaction *trans) | |
1210 | { | |
72d63ed6 LZ |
1211 | wait_event(root->fs_info->transaction_wait, |
1212 | trans->commit_done || (trans->in_commit && !trans->blocked)); | |
bb9c12c9 SW |
1213 | } |
1214 | ||
1215 | /* | |
1216 | * commit transactions asynchronously. once btrfs_commit_transaction_async | |
1217 | * returns, any subsequent transaction will not be allowed to join. | |
1218 | */ | |
1219 | struct btrfs_async_commit { | |
1220 | struct btrfs_trans_handle *newtrans; | |
1221 | struct btrfs_root *root; | |
1222 | struct delayed_work work; | |
1223 | }; | |
1224 | ||
1225 | static void do_async_commit(struct work_struct *work) | |
1226 | { | |
1227 | struct btrfs_async_commit *ac = | |
1228 | container_of(work, struct btrfs_async_commit, work.work); | |
1229 | ||
6fc4e354 SW |
1230 | /* |
1231 | * We've got freeze protection passed with the transaction. | |
1232 | * Tell lockdep about it. | |
1233 | */ | |
1234 | rwsem_acquire_read( | |
1235 | &ac->root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1], | |
1236 | 0, 1, _THIS_IP_); | |
1237 | ||
e209db7a SW |
1238 | current->journal_info = ac->newtrans; |
1239 | ||
bb9c12c9 SW |
1240 | btrfs_commit_transaction(ac->newtrans, ac->root); |
1241 | kfree(ac); | |
1242 | } | |
1243 | ||
1244 | int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans, | |
1245 | struct btrfs_root *root, | |
1246 | int wait_for_unblock) | |
1247 | { | |
1248 | struct btrfs_async_commit *ac; | |
1249 | struct btrfs_transaction *cur_trans; | |
1250 | ||
1251 | ac = kmalloc(sizeof(*ac), GFP_NOFS); | |
db5b493a TI |
1252 | if (!ac) |
1253 | return -ENOMEM; | |
bb9c12c9 SW |
1254 | |
1255 | INIT_DELAYED_WORK(&ac->work, do_async_commit); | |
1256 | ac->root = root; | |
7a7eaa40 | 1257 | ac->newtrans = btrfs_join_transaction(root); |
3612b495 TI |
1258 | if (IS_ERR(ac->newtrans)) { |
1259 | int err = PTR_ERR(ac->newtrans); | |
1260 | kfree(ac); | |
1261 | return err; | |
1262 | } | |
bb9c12c9 SW |
1263 | |
1264 | /* take transaction reference */ | |
bb9c12c9 | 1265 | cur_trans = trans->transaction; |
13c5a93e | 1266 | atomic_inc(&cur_trans->use_count); |
bb9c12c9 SW |
1267 | |
1268 | btrfs_end_transaction(trans, root); | |
6fc4e354 SW |
1269 | |
1270 | /* | |
1271 | * Tell lockdep we've released the freeze rwsem, since the | |
1272 | * async commit thread will be the one to unlock it. | |
1273 | */ | |
1274 | rwsem_release(&root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1], | |
1275 | 1, _THIS_IP_); | |
1276 | ||
bb9c12c9 SW |
1277 | schedule_delayed_work(&ac->work, 0); |
1278 | ||
1279 | /* wait for transaction to start and unblock */ | |
bb9c12c9 SW |
1280 | if (wait_for_unblock) |
1281 | wait_current_trans_commit_start_and_unblock(root, cur_trans); | |
1282 | else | |
1283 | wait_current_trans_commit_start(root, cur_trans); | |
bb9c12c9 | 1284 | |
38e88054 SW |
1285 | if (current->journal_info == trans) |
1286 | current->journal_info = NULL; | |
1287 | ||
1288 | put_transaction(cur_trans); | |
bb9c12c9 SW |
1289 | return 0; |
1290 | } | |
1291 | ||
49b25e05 JM |
1292 | |
1293 | static void cleanup_transaction(struct btrfs_trans_handle *trans, | |
7b8b92af | 1294 | struct btrfs_root *root, int err) |
49b25e05 JM |
1295 | { |
1296 | struct btrfs_transaction *cur_trans = trans->transaction; | |
1297 | ||
1298 | WARN_ON(trans->use_count > 1); | |
1299 | ||
7b8b92af JB |
1300 | btrfs_abort_transaction(trans, root, err); |
1301 | ||
49b25e05 JM |
1302 | spin_lock(&root->fs_info->trans_lock); |
1303 | list_del_init(&cur_trans->list); | |
d7096fc3 JB |
1304 | if (cur_trans == root->fs_info->running_transaction) { |
1305 | root->fs_info->running_transaction = NULL; | |
1306 | root->fs_info->trans_no_join = 0; | |
1307 | } | |
49b25e05 JM |
1308 | spin_unlock(&root->fs_info->trans_lock); |
1309 | ||
1310 | btrfs_cleanup_one_transaction(trans->transaction, root); | |
1311 | ||
1312 | put_transaction(cur_trans); | |
1313 | put_transaction(cur_trans); | |
1314 | ||
1315 | trace_btrfs_transaction_commit(root); | |
1316 | ||
1317 | btrfs_scrub_continue(root); | |
1318 | ||
1319 | if (current->journal_info == trans) | |
1320 | current->journal_info = NULL; | |
1321 | ||
1322 | kmem_cache_free(btrfs_trans_handle_cachep, trans); | |
1323 | } | |
1324 | ||
bb9c12c9 SW |
1325 | /* |
1326 | * btrfs_transaction state sequence: | |
1327 | * in_commit = 0, blocked = 0 (initial) | |
1328 | * in_commit = 1, blocked = 1 | |
1329 | * blocked = 0 | |
1330 | * commit_done = 1 | |
1331 | */ | |
79154b1b CM |
1332 | int btrfs_commit_transaction(struct btrfs_trans_handle *trans, |
1333 | struct btrfs_root *root) | |
1334 | { | |
15ee9bc7 | 1335 | unsigned long joined = 0; |
49b25e05 | 1336 | struct btrfs_transaction *cur_trans = trans->transaction; |
8fd17795 | 1337 | struct btrfs_transaction *prev_trans = NULL; |
79154b1b | 1338 | DEFINE_WAIT(wait); |
49b25e05 | 1339 | int ret = -EIO; |
89573b9c CM |
1340 | int should_grow = 0; |
1341 | unsigned long now = get_seconds(); | |
dccae999 | 1342 | int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT); |
79154b1b | 1343 | |
5a3f23d5 CM |
1344 | btrfs_run_ordered_operations(root, 0); |
1345 | ||
49b25e05 JM |
1346 | if (cur_trans->aborted) |
1347 | goto cleanup_transaction; | |
1348 | ||
56bec294 CM |
1349 | /* make a pass through all the delayed refs we have so far |
1350 | * any runnings procs may add more while we are here | |
1351 | */ | |
1352 | ret = btrfs_run_delayed_refs(trans, root, 0); | |
49b25e05 JM |
1353 | if (ret) |
1354 | goto cleanup_transaction; | |
56bec294 | 1355 | |
0e721106 JB |
1356 | btrfs_trans_release_metadata(trans, root); |
1357 | trans->block_rsv = NULL; | |
1358 | ||
b7ec40d7 | 1359 | cur_trans = trans->transaction; |
49b25e05 | 1360 | |
56bec294 CM |
1361 | /* |
1362 | * set the flushing flag so procs in this transaction have to | |
1363 | * start sending their work down. | |
1364 | */ | |
b7ec40d7 | 1365 | cur_trans->delayed_refs.flushing = 1; |
56bec294 | 1366 | |
c3e69d58 | 1367 | ret = btrfs_run_delayed_refs(trans, root, 0); |
49b25e05 JM |
1368 | if (ret) |
1369 | goto cleanup_transaction; | |
56bec294 | 1370 | |
a4abeea4 | 1371 | spin_lock(&cur_trans->commit_lock); |
b7ec40d7 | 1372 | if (cur_trans->in_commit) { |
a4abeea4 | 1373 | spin_unlock(&cur_trans->commit_lock); |
13c5a93e | 1374 | atomic_inc(&cur_trans->use_count); |
49b25e05 | 1375 | ret = btrfs_end_transaction(trans, root); |
ccd467d6 | 1376 | |
b9c8300c | 1377 | wait_for_commit(root, cur_trans); |
15ee9bc7 | 1378 | |
79154b1b | 1379 | put_transaction(cur_trans); |
15ee9bc7 | 1380 | |
49b25e05 | 1381 | return ret; |
79154b1b | 1382 | } |
4313b399 | 1383 | |
2c90e5d6 | 1384 | trans->transaction->in_commit = 1; |
f9295749 | 1385 | trans->transaction->blocked = 1; |
a4abeea4 | 1386 | spin_unlock(&cur_trans->commit_lock); |
bb9c12c9 SW |
1387 | wake_up(&root->fs_info->transaction_blocked_wait); |
1388 | ||
a4abeea4 | 1389 | spin_lock(&root->fs_info->trans_lock); |
ccd467d6 CM |
1390 | if (cur_trans->list.prev != &root->fs_info->trans_list) { |
1391 | prev_trans = list_entry(cur_trans->list.prev, | |
1392 | struct btrfs_transaction, list); | |
1393 | if (!prev_trans->commit_done) { | |
13c5a93e | 1394 | atomic_inc(&prev_trans->use_count); |
a4abeea4 | 1395 | spin_unlock(&root->fs_info->trans_lock); |
ccd467d6 CM |
1396 | |
1397 | wait_for_commit(root, prev_trans); | |
ccd467d6 | 1398 | |
15ee9bc7 | 1399 | put_transaction(prev_trans); |
a4abeea4 JB |
1400 | } else { |
1401 | spin_unlock(&root->fs_info->trans_lock); | |
ccd467d6 | 1402 | } |
a4abeea4 JB |
1403 | } else { |
1404 | spin_unlock(&root->fs_info->trans_lock); | |
ccd467d6 | 1405 | } |
15ee9bc7 | 1406 | |
e39e64ac CM |
1407 | if (!btrfs_test_opt(root, SSD) && |
1408 | (now < cur_trans->start_time || now - cur_trans->start_time < 1)) | |
89573b9c CM |
1409 | should_grow = 1; |
1410 | ||
15ee9bc7 | 1411 | do { |
7ea394f1 | 1412 | int snap_pending = 0; |
a4abeea4 | 1413 | |
15ee9bc7 | 1414 | joined = cur_trans->num_joined; |
7ea394f1 YZ |
1415 | if (!list_empty(&trans->transaction->pending_snapshots)) |
1416 | snap_pending = 1; | |
1417 | ||
2c90e5d6 | 1418 | WARN_ON(cur_trans != trans->transaction); |
15ee9bc7 | 1419 | |
0bdb1db2 | 1420 | if (flush_on_commit || snap_pending) { |
24bbcf04 | 1421 | btrfs_start_delalloc_inodes(root, 1); |
143bede5 | 1422 | btrfs_wait_ordered_extents(root, 0, 1); |
7ea394f1 YZ |
1423 | } |
1424 | ||
16cdcec7 | 1425 | ret = btrfs_run_delayed_items(trans, root); |
49b25e05 JM |
1426 | if (ret) |
1427 | goto cleanup_transaction; | |
16cdcec7 | 1428 | |
edf39272 JS |
1429 | /* |
1430 | * running the delayed items may have added new refs. account | |
1431 | * them now so that they hinder processing of more delayed refs | |
1432 | * as little as possible. | |
1433 | */ | |
1434 | btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info); | |
1435 | ||
5a3f23d5 CM |
1436 | /* |
1437 | * rename don't use btrfs_join_transaction, so, once we | |
1438 | * set the transaction to blocked above, we aren't going | |
1439 | * to get any new ordered operations. We can safely run | |
1440 | * it here and no for sure that nothing new will be added | |
1441 | * to the list | |
1442 | */ | |
1443 | btrfs_run_ordered_operations(root, 1); | |
1444 | ||
ed3b3d31 CM |
1445 | prepare_to_wait(&cur_trans->writer_wait, &wait, |
1446 | TASK_UNINTERRUPTIBLE); | |
1447 | ||
13c5a93e | 1448 | if (atomic_read(&cur_trans->num_writers) > 1) |
99d16cbc SW |
1449 | schedule_timeout(MAX_SCHEDULE_TIMEOUT); |
1450 | else if (should_grow) | |
1451 | schedule_timeout(1); | |
15ee9bc7 | 1452 | |
15ee9bc7 | 1453 | finish_wait(&cur_trans->writer_wait, &wait); |
13c5a93e | 1454 | } while (atomic_read(&cur_trans->num_writers) > 1 || |
89573b9c | 1455 | (should_grow && cur_trans->num_joined != joined)); |
15ee9bc7 | 1456 | |
ed0ca140 JB |
1457 | /* |
1458 | * Ok now we need to make sure to block out any other joins while we | |
1459 | * commit the transaction. We could have started a join before setting | |
1460 | * no_join so make sure to wait for num_writers to == 1 again. | |
1461 | */ | |
1462 | spin_lock(&root->fs_info->trans_lock); | |
1463 | root->fs_info->trans_no_join = 1; | |
1464 | spin_unlock(&root->fs_info->trans_lock); | |
1465 | wait_event(cur_trans->writer_wait, | |
1466 | atomic_read(&cur_trans->num_writers) == 1); | |
1467 | ||
7585717f CM |
1468 | /* |
1469 | * the reloc mutex makes sure that we stop | |
1470 | * the balancing code from coming in and moving | |
1471 | * extents around in the middle of the commit | |
1472 | */ | |
1473 | mutex_lock(&root->fs_info->reloc_mutex); | |
1474 | ||
e999376f | 1475 | ret = btrfs_run_delayed_items(trans, root); |
49b25e05 JM |
1476 | if (ret) { |
1477 | mutex_unlock(&root->fs_info->reloc_mutex); | |
1478 | goto cleanup_transaction; | |
1479 | } | |
3063d29f | 1480 | |
e999376f | 1481 | ret = create_pending_snapshots(trans, root->fs_info); |
49b25e05 JM |
1482 | if (ret) { |
1483 | mutex_unlock(&root->fs_info->reloc_mutex); | |
1484 | goto cleanup_transaction; | |
1485 | } | |
16cdcec7 | 1486 | |
56bec294 | 1487 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
49b25e05 JM |
1488 | if (ret) { |
1489 | mutex_unlock(&root->fs_info->reloc_mutex); | |
1490 | goto cleanup_transaction; | |
1491 | } | |
56bec294 | 1492 | |
e999376f CM |
1493 | /* |
1494 | * make sure none of the code above managed to slip in a | |
1495 | * delayed item | |
1496 | */ | |
1497 | btrfs_assert_delayed_root_empty(root); | |
1498 | ||
2c90e5d6 | 1499 | WARN_ON(cur_trans != trans->transaction); |
dc17ff8f | 1500 | |
a2de733c | 1501 | btrfs_scrub_pause(root); |
e02119d5 CM |
1502 | /* btrfs_commit_tree_roots is responsible for getting the |
1503 | * various roots consistent with each other. Every pointer | |
1504 | * in the tree of tree roots has to point to the most up to date | |
1505 | * root for every subvolume and other tree. So, we have to keep | |
1506 | * the tree logging code from jumping in and changing any | |
1507 | * of the trees. | |
1508 | * | |
1509 | * At this point in the commit, there can't be any tree-log | |
1510 | * writers, but a little lower down we drop the trans mutex | |
1511 | * and let new people in. By holding the tree_log_mutex | |
1512 | * from now until after the super is written, we avoid races | |
1513 | * with the tree-log code. | |
1514 | */ | |
1515 | mutex_lock(&root->fs_info->tree_log_mutex); | |
1516 | ||
5d4f98a2 | 1517 | ret = commit_fs_roots(trans, root); |
49b25e05 JM |
1518 | if (ret) { |
1519 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
871383be | 1520 | mutex_unlock(&root->fs_info->reloc_mutex); |
49b25e05 JM |
1521 | goto cleanup_transaction; |
1522 | } | |
54aa1f4d | 1523 | |
5d4f98a2 | 1524 | /* commit_fs_roots gets rid of all the tree log roots, it is now |
e02119d5 CM |
1525 | * safe to free the root of tree log roots |
1526 | */ | |
1527 | btrfs_free_log_root_tree(trans, root->fs_info); | |
1528 | ||
5d4f98a2 | 1529 | ret = commit_cowonly_roots(trans, root); |
49b25e05 JM |
1530 | if (ret) { |
1531 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
871383be | 1532 | mutex_unlock(&root->fs_info->reloc_mutex); |
49b25e05 JM |
1533 | goto cleanup_transaction; |
1534 | } | |
54aa1f4d | 1535 | |
11833d66 YZ |
1536 | btrfs_prepare_extent_commit(trans, root); |
1537 | ||
78fae27e | 1538 | cur_trans = root->fs_info->running_transaction; |
5d4f98a2 YZ |
1539 | |
1540 | btrfs_set_root_node(&root->fs_info->tree_root->root_item, | |
1541 | root->fs_info->tree_root->node); | |
817d52f8 | 1542 | switch_commit_root(root->fs_info->tree_root); |
5d4f98a2 YZ |
1543 | |
1544 | btrfs_set_root_node(&root->fs_info->chunk_root->root_item, | |
1545 | root->fs_info->chunk_root->node); | |
817d52f8 | 1546 | switch_commit_root(root->fs_info->chunk_root); |
5d4f98a2 | 1547 | |
edf39272 | 1548 | assert_qgroups_uptodate(trans); |
5d4f98a2 | 1549 | update_super_roots(root); |
e02119d5 CM |
1550 | |
1551 | if (!root->fs_info->log_root_recovering) { | |
6c41761f DS |
1552 | btrfs_set_super_log_root(root->fs_info->super_copy, 0); |
1553 | btrfs_set_super_log_root_level(root->fs_info->super_copy, 0); | |
e02119d5 CM |
1554 | } |
1555 | ||
6c41761f DS |
1556 | memcpy(root->fs_info->super_for_commit, root->fs_info->super_copy, |
1557 | sizeof(*root->fs_info->super_copy)); | |
ccd467d6 | 1558 | |
f9295749 | 1559 | trans->transaction->blocked = 0; |
a4abeea4 JB |
1560 | spin_lock(&root->fs_info->trans_lock); |
1561 | root->fs_info->running_transaction = NULL; | |
1562 | root->fs_info->trans_no_join = 0; | |
1563 | spin_unlock(&root->fs_info->trans_lock); | |
7585717f | 1564 | mutex_unlock(&root->fs_info->reloc_mutex); |
b7ec40d7 | 1565 | |
f9295749 | 1566 | wake_up(&root->fs_info->transaction_wait); |
e6dcd2dc | 1567 | |
79154b1b | 1568 | ret = btrfs_write_and_wait_transaction(trans, root); |
49b25e05 JM |
1569 | if (ret) { |
1570 | btrfs_error(root->fs_info, ret, | |
1571 | "Error while writing out transaction."); | |
1572 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
1573 | goto cleanup_transaction; | |
1574 | } | |
1575 | ||
1576 | ret = write_ctree_super(trans, root, 0); | |
1577 | if (ret) { | |
1578 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
1579 | goto cleanup_transaction; | |
1580 | } | |
4313b399 | 1581 | |
e02119d5 CM |
1582 | /* |
1583 | * the super is written, we can safely allow the tree-loggers | |
1584 | * to go about their business | |
1585 | */ | |
1586 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
1587 | ||
11833d66 | 1588 | btrfs_finish_extent_commit(trans, root); |
4313b399 | 1589 | |
2c90e5d6 | 1590 | cur_trans->commit_done = 1; |
b7ec40d7 | 1591 | |
15ee9bc7 | 1592 | root->fs_info->last_trans_committed = cur_trans->transid; |
817d52f8 | 1593 | |
2c90e5d6 | 1594 | wake_up(&cur_trans->commit_wait); |
3de4586c | 1595 | |
a4abeea4 | 1596 | spin_lock(&root->fs_info->trans_lock); |
13c5a93e | 1597 | list_del_init(&cur_trans->list); |
a4abeea4 JB |
1598 | spin_unlock(&root->fs_info->trans_lock); |
1599 | ||
78fae27e | 1600 | put_transaction(cur_trans); |
79154b1b | 1601 | put_transaction(cur_trans); |
58176a96 | 1602 | |
b2b5ef5c JK |
1603 | sb_end_intwrite(root->fs_info->sb); |
1604 | ||
1abe9b8a | 1605 | trace_btrfs_transaction_commit(root); |
1606 | ||
a2de733c AJ |
1607 | btrfs_scrub_continue(root); |
1608 | ||
9ed74f2d JB |
1609 | if (current->journal_info == trans) |
1610 | current->journal_info = NULL; | |
1611 | ||
2c90e5d6 | 1612 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
24bbcf04 YZ |
1613 | |
1614 | if (current != root->fs_info->transaction_kthread) | |
1615 | btrfs_run_delayed_iputs(root); | |
1616 | ||
79154b1b | 1617 | return ret; |
49b25e05 JM |
1618 | |
1619 | cleanup_transaction: | |
0e721106 JB |
1620 | btrfs_trans_release_metadata(trans, root); |
1621 | trans->block_rsv = NULL; | |
49b25e05 JM |
1622 | btrfs_printk(root->fs_info, "Skipping commit of aborted transaction.\n"); |
1623 | // WARN_ON(1); | |
1624 | if (current->journal_info == trans) | |
1625 | current->journal_info = NULL; | |
7b8b92af | 1626 | cleanup_transaction(trans, root, ret); |
49b25e05 JM |
1627 | |
1628 | return ret; | |
79154b1b CM |
1629 | } |
1630 | ||
d352ac68 CM |
1631 | /* |
1632 | * interface function to delete all the snapshots we have scheduled for deletion | |
1633 | */ | |
e9d0b13b CM |
1634 | int btrfs_clean_old_snapshots(struct btrfs_root *root) |
1635 | { | |
5d4f98a2 YZ |
1636 | LIST_HEAD(list); |
1637 | struct btrfs_fs_info *fs_info = root->fs_info; | |
1638 | ||
a4abeea4 | 1639 | spin_lock(&fs_info->trans_lock); |
5d4f98a2 | 1640 | list_splice_init(&fs_info->dead_roots, &list); |
a4abeea4 | 1641 | spin_unlock(&fs_info->trans_lock); |
e9d0b13b | 1642 | |
5d4f98a2 | 1643 | while (!list_empty(&list)) { |
2c536799 JM |
1644 | int ret; |
1645 | ||
5d4f98a2 | 1646 | root = list_entry(list.next, struct btrfs_root, root_list); |
76dda93c YZ |
1647 | list_del(&root->root_list); |
1648 | ||
16cdcec7 MX |
1649 | btrfs_kill_all_delayed_nodes(root); |
1650 | ||
76dda93c YZ |
1651 | if (btrfs_header_backref_rev(root->node) < |
1652 | BTRFS_MIXED_BACKREF_REV) | |
2c536799 | 1653 | ret = btrfs_drop_snapshot(root, NULL, 0, 0); |
76dda93c | 1654 | else |
2c536799 JM |
1655 | ret =btrfs_drop_snapshot(root, NULL, 1, 0); |
1656 | BUG_ON(ret < 0); | |
e9d0b13b CM |
1657 | } |
1658 | return 0; | |
1659 | } |