Commit | Line | Data |
---|---|---|
6cbd5570 CM |
1 | /* |
2 | * Copyright (C) 2007 Oracle. All rights reserved. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public | |
6 | * License v2 as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, | |
9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public | |
14 | * License along with this program; if not, write to the | |
15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
16 | * Boston, MA 021110-1307, USA. | |
17 | */ | |
18 | ||
79154b1b | 19 | #include <linux/fs.h> |
5a0e3ad6 | 20 | #include <linux/slab.h> |
34088780 | 21 | #include <linux/sched.h> |
d3c2fdcf | 22 | #include <linux/writeback.h> |
5f39d397 | 23 | #include <linux/pagemap.h> |
5f2cc086 | 24 | #include <linux/blkdev.h> |
79154b1b CM |
25 | #include "ctree.h" |
26 | #include "disk-io.h" | |
27 | #include "transaction.h" | |
925baedd | 28 | #include "locking.h" |
e02119d5 | 29 | #include "tree-log.h" |
581bb050 | 30 | #include "inode-map.h" |
79154b1b | 31 | |
0f7d52f4 CM |
32 | #define BTRFS_ROOT_TRANS_TAG 0 |
33 | ||
80b6794d | 34 | static noinline void put_transaction(struct btrfs_transaction *transaction) |
79154b1b | 35 | { |
13c5a93e JB |
36 | WARN_ON(atomic_read(&transaction->use_count) == 0); |
37 | if (atomic_dec_and_test(&transaction->use_count)) { | |
a4abeea4 | 38 | BUG_ON(!list_empty(&transaction->list)); |
00f04b88 AJ |
39 | WARN_ON(transaction->delayed_refs.root.rb_node); |
40 | WARN_ON(!list_empty(&transaction->delayed_refs.seq_head)); | |
2c90e5d6 CM |
41 | memset(transaction, 0, sizeof(*transaction)); |
42 | kmem_cache_free(btrfs_transaction_cachep, transaction); | |
78fae27e | 43 | } |
79154b1b CM |
44 | } |
45 | ||
817d52f8 JB |
46 | static noinline void switch_commit_root(struct btrfs_root *root) |
47 | { | |
817d52f8 JB |
48 | free_extent_buffer(root->commit_root); |
49 | root->commit_root = btrfs_root_node(root); | |
817d52f8 JB |
50 | } |
51 | ||
d352ac68 CM |
52 | /* |
53 | * either allocate a new transaction or hop into the existing one | |
54 | */ | |
a4abeea4 | 55 | static noinline int join_transaction(struct btrfs_root *root, int nofail) |
79154b1b CM |
56 | { |
57 | struct btrfs_transaction *cur_trans; | |
a4abeea4 JB |
58 | |
59 | spin_lock(&root->fs_info->trans_lock); | |
d43317dc | 60 | loop: |
a4abeea4 JB |
61 | if (root->fs_info->trans_no_join) { |
62 | if (!nofail) { | |
63 | spin_unlock(&root->fs_info->trans_lock); | |
64 | return -EBUSY; | |
65 | } | |
66 | } | |
67 | ||
79154b1b | 68 | cur_trans = root->fs_info->running_transaction; |
a4abeea4 JB |
69 | if (cur_trans) { |
70 | atomic_inc(&cur_trans->use_count); | |
13c5a93e | 71 | atomic_inc(&cur_trans->num_writers); |
15ee9bc7 | 72 | cur_trans->num_joined++; |
a4abeea4 JB |
73 | spin_unlock(&root->fs_info->trans_lock); |
74 | return 0; | |
79154b1b | 75 | } |
a4abeea4 JB |
76 | spin_unlock(&root->fs_info->trans_lock); |
77 | ||
78 | cur_trans = kmem_cache_alloc(btrfs_transaction_cachep, GFP_NOFS); | |
79 | if (!cur_trans) | |
80 | return -ENOMEM; | |
d43317dc | 81 | |
a4abeea4 JB |
82 | spin_lock(&root->fs_info->trans_lock); |
83 | if (root->fs_info->running_transaction) { | |
d43317dc CM |
84 | /* |
85 | * someone started a transaction after we unlocked. Make sure | |
86 | * to redo the trans_no_join checks above | |
87 | */ | |
a4abeea4 JB |
88 | kmem_cache_free(btrfs_transaction_cachep, cur_trans); |
89 | cur_trans = root->fs_info->running_transaction; | |
d43317dc | 90 | goto loop; |
79154b1b | 91 | } |
d43317dc | 92 | |
a4abeea4 JB |
93 | atomic_set(&cur_trans->num_writers, 1); |
94 | cur_trans->num_joined = 0; | |
95 | init_waitqueue_head(&cur_trans->writer_wait); | |
96 | init_waitqueue_head(&cur_trans->commit_wait); | |
97 | cur_trans->in_commit = 0; | |
98 | cur_trans->blocked = 0; | |
99 | /* | |
100 | * One for this trans handle, one so it will live on until we | |
101 | * commit the transaction. | |
102 | */ | |
103 | atomic_set(&cur_trans->use_count, 2); | |
104 | cur_trans->commit_done = 0; | |
105 | cur_trans->start_time = get_seconds(); | |
106 | ||
107 | cur_trans->delayed_refs.root = RB_ROOT; | |
108 | cur_trans->delayed_refs.num_entries = 0; | |
109 | cur_trans->delayed_refs.num_heads_ready = 0; | |
110 | cur_trans->delayed_refs.num_heads = 0; | |
111 | cur_trans->delayed_refs.flushing = 0; | |
112 | cur_trans->delayed_refs.run_delayed_start = 0; | |
00f04b88 | 113 | cur_trans->delayed_refs.seq = 1; |
a168650c | 114 | init_waitqueue_head(&cur_trans->delayed_refs.seq_wait); |
a4abeea4 JB |
115 | spin_lock_init(&cur_trans->commit_lock); |
116 | spin_lock_init(&cur_trans->delayed_refs.lock); | |
00f04b88 | 117 | INIT_LIST_HEAD(&cur_trans->delayed_refs.seq_head); |
a4abeea4 JB |
118 | |
119 | INIT_LIST_HEAD(&cur_trans->pending_snapshots); | |
120 | list_add_tail(&cur_trans->list, &root->fs_info->trans_list); | |
121 | extent_io_tree_init(&cur_trans->dirty_pages, | |
ff5714cc | 122 | root->fs_info->btree_inode->i_mapping); |
a4abeea4 JB |
123 | root->fs_info->generation++; |
124 | cur_trans->transid = root->fs_info->generation; | |
125 | root->fs_info->running_transaction = cur_trans; | |
126 | spin_unlock(&root->fs_info->trans_lock); | |
15ee9bc7 | 127 | |
79154b1b CM |
128 | return 0; |
129 | } | |
130 | ||
d352ac68 | 131 | /* |
d397712b CM |
132 | * this does all the record keeping required to make sure that a reference |
133 | * counted root is properly recorded in a given transaction. This is required | |
134 | * to make sure the old root from before we joined the transaction is deleted | |
135 | * when the transaction commits | |
d352ac68 | 136 | */ |
7585717f | 137 | static int record_root_in_trans(struct btrfs_trans_handle *trans, |
a4abeea4 | 138 | struct btrfs_root *root) |
6702ed49 | 139 | { |
5d4f98a2 | 140 | if (root->ref_cows && root->last_trans < trans->transid) { |
6702ed49 | 141 | WARN_ON(root == root->fs_info->extent_root); |
5d4f98a2 YZ |
142 | WARN_ON(root->commit_root != root->node); |
143 | ||
7585717f CM |
144 | /* |
145 | * see below for in_trans_setup usage rules | |
146 | * we have the reloc mutex held now, so there | |
147 | * is only one writer in this function | |
148 | */ | |
149 | root->in_trans_setup = 1; | |
150 | ||
151 | /* make sure readers find in_trans_setup before | |
152 | * they find our root->last_trans update | |
153 | */ | |
154 | smp_wmb(); | |
155 | ||
a4abeea4 JB |
156 | spin_lock(&root->fs_info->fs_roots_radix_lock); |
157 | if (root->last_trans == trans->transid) { | |
158 | spin_unlock(&root->fs_info->fs_roots_radix_lock); | |
159 | return 0; | |
160 | } | |
5d4f98a2 YZ |
161 | radix_tree_tag_set(&root->fs_info->fs_roots_radix, |
162 | (unsigned long)root->root_key.objectid, | |
163 | BTRFS_ROOT_TRANS_TAG); | |
a4abeea4 | 164 | spin_unlock(&root->fs_info->fs_roots_radix_lock); |
7585717f CM |
165 | root->last_trans = trans->transid; |
166 | ||
167 | /* this is pretty tricky. We don't want to | |
168 | * take the relocation lock in btrfs_record_root_in_trans | |
169 | * unless we're really doing the first setup for this root in | |
170 | * this transaction. | |
171 | * | |
172 | * Normally we'd use root->last_trans as a flag to decide | |
173 | * if we want to take the expensive mutex. | |
174 | * | |
175 | * But, we have to set root->last_trans before we | |
176 | * init the relocation root, otherwise, we trip over warnings | |
177 | * in ctree.c. The solution used here is to flag ourselves | |
178 | * with root->in_trans_setup. When this is 1, we're still | |
179 | * fixing up the reloc trees and everyone must wait. | |
180 | * | |
181 | * When this is zero, they can trust root->last_trans and fly | |
182 | * through btrfs_record_root_in_trans without having to take the | |
183 | * lock. smp_wmb() makes sure that all the writes above are | |
184 | * done before we pop in the zero below | |
185 | */ | |
5d4f98a2 | 186 | btrfs_init_reloc_root(trans, root); |
7585717f CM |
187 | smp_wmb(); |
188 | root->in_trans_setup = 0; | |
5d4f98a2 YZ |
189 | } |
190 | return 0; | |
191 | } | |
bcc63abb | 192 | |
7585717f CM |
193 | |
194 | int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, | |
195 | struct btrfs_root *root) | |
196 | { | |
197 | if (!root->ref_cows) | |
198 | return 0; | |
199 | ||
200 | /* | |
201 | * see record_root_in_trans for comments about in_trans_setup usage | |
202 | * and barriers | |
203 | */ | |
204 | smp_rmb(); | |
205 | if (root->last_trans == trans->transid && | |
206 | !root->in_trans_setup) | |
207 | return 0; | |
208 | ||
209 | mutex_lock(&root->fs_info->reloc_mutex); | |
210 | record_root_in_trans(trans, root); | |
211 | mutex_unlock(&root->fs_info->reloc_mutex); | |
212 | ||
213 | return 0; | |
214 | } | |
215 | ||
d352ac68 CM |
216 | /* wait for commit against the current transaction to become unblocked |
217 | * when this is done, it is safe to start a new transaction, but the current | |
218 | * transaction might not be fully on disk. | |
219 | */ | |
37d1aeee | 220 | static void wait_current_trans(struct btrfs_root *root) |
79154b1b | 221 | { |
f9295749 | 222 | struct btrfs_transaction *cur_trans; |
79154b1b | 223 | |
a4abeea4 | 224 | spin_lock(&root->fs_info->trans_lock); |
f9295749 | 225 | cur_trans = root->fs_info->running_transaction; |
37d1aeee | 226 | if (cur_trans && cur_trans->blocked) { |
13c5a93e | 227 | atomic_inc(&cur_trans->use_count); |
a4abeea4 | 228 | spin_unlock(&root->fs_info->trans_lock); |
72d63ed6 LZ |
229 | |
230 | wait_event(root->fs_info->transaction_wait, | |
231 | !cur_trans->blocked); | |
f9295749 | 232 | put_transaction(cur_trans); |
a4abeea4 JB |
233 | } else { |
234 | spin_unlock(&root->fs_info->trans_lock); | |
f9295749 | 235 | } |
37d1aeee CM |
236 | } |
237 | ||
249ac1e5 JB |
238 | enum btrfs_trans_type { |
239 | TRANS_START, | |
240 | TRANS_JOIN, | |
241 | TRANS_USERSPACE, | |
0af3d00b | 242 | TRANS_JOIN_NOLOCK, |
249ac1e5 JB |
243 | }; |
244 | ||
a22285a6 YZ |
245 | static int may_wait_transaction(struct btrfs_root *root, int type) |
246 | { | |
a4abeea4 JB |
247 | if (root->fs_info->log_root_recovering) |
248 | return 0; | |
249 | ||
250 | if (type == TRANS_USERSPACE) | |
251 | return 1; | |
252 | ||
253 | if (type == TRANS_START && | |
254 | !atomic_read(&root->fs_info->open_ioctl_trans)) | |
a22285a6 | 255 | return 1; |
a4abeea4 | 256 | |
a22285a6 YZ |
257 | return 0; |
258 | } | |
259 | ||
e02119d5 | 260 | static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root, |
a22285a6 | 261 | u64 num_items, int type) |
37d1aeee | 262 | { |
a22285a6 YZ |
263 | struct btrfs_trans_handle *h; |
264 | struct btrfs_transaction *cur_trans; | |
b5009945 | 265 | u64 num_bytes = 0; |
37d1aeee | 266 | int ret; |
acce952b | 267 | |
268 | if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) | |
269 | return ERR_PTR(-EROFS); | |
2a1eb461 JB |
270 | |
271 | if (current->journal_info) { | |
272 | WARN_ON(type != TRANS_JOIN && type != TRANS_JOIN_NOLOCK); | |
273 | h = current->journal_info; | |
274 | h->use_count++; | |
275 | h->orig_rsv = h->block_rsv; | |
276 | h->block_rsv = NULL; | |
277 | goto got_it; | |
278 | } | |
b5009945 JB |
279 | |
280 | /* | |
281 | * Do the reservation before we join the transaction so we can do all | |
282 | * the appropriate flushing if need be. | |
283 | */ | |
284 | if (num_items > 0 && root != root->fs_info->chunk_root) { | |
285 | num_bytes = btrfs_calc_trans_metadata_size(root, num_items); | |
4a92b1b8 | 286 | ret = btrfs_block_rsv_add(root, |
b5009945 JB |
287 | &root->fs_info->trans_block_rsv, |
288 | num_bytes); | |
289 | if (ret) | |
290 | return ERR_PTR(ret); | |
291 | } | |
a22285a6 YZ |
292 | again: |
293 | h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS); | |
294 | if (!h) | |
295 | return ERR_PTR(-ENOMEM); | |
37d1aeee | 296 | |
a22285a6 | 297 | if (may_wait_transaction(root, type)) |
37d1aeee | 298 | wait_current_trans(root); |
a22285a6 | 299 | |
a4abeea4 JB |
300 | do { |
301 | ret = join_transaction(root, type == TRANS_JOIN_NOLOCK); | |
302 | if (ret == -EBUSY) | |
303 | wait_current_trans(root); | |
304 | } while (ret == -EBUSY); | |
305 | ||
db5b493a | 306 | if (ret < 0) { |
6e8df2ae | 307 | kmem_cache_free(btrfs_trans_handle_cachep, h); |
db5b493a TI |
308 | return ERR_PTR(ret); |
309 | } | |
0f7d52f4 | 310 | |
a22285a6 | 311 | cur_trans = root->fs_info->running_transaction; |
a22285a6 YZ |
312 | |
313 | h->transid = cur_trans->transid; | |
314 | h->transaction = cur_trans; | |
79154b1b | 315 | h->blocks_used = 0; |
a22285a6 | 316 | h->bytes_reserved = 0; |
56bec294 | 317 | h->delayed_ref_updates = 0; |
2a1eb461 | 318 | h->use_count = 1; |
f0486c68 | 319 | h->block_rsv = NULL; |
2a1eb461 | 320 | h->orig_rsv = NULL; |
b7ec40d7 | 321 | |
a22285a6 YZ |
322 | smp_mb(); |
323 | if (cur_trans->blocked && may_wait_transaction(root, type)) { | |
324 | btrfs_commit_transaction(h, root); | |
325 | goto again; | |
326 | } | |
327 | ||
b5009945 | 328 | if (num_bytes) { |
8c2a3ca2 JB |
329 | trace_btrfs_space_reservation(root->fs_info, "transaction", |
330 | (u64)h, num_bytes, 1); | |
b5009945 JB |
331 | h->block_rsv = &root->fs_info->trans_block_rsv; |
332 | h->bytes_reserved = num_bytes; | |
a22285a6 | 333 | } |
9ed74f2d | 334 | |
2a1eb461 | 335 | got_it: |
a4abeea4 | 336 | btrfs_record_root_in_trans(h, root); |
a22285a6 YZ |
337 | |
338 | if (!current->journal_info && type != TRANS_USERSPACE) | |
339 | current->journal_info = h; | |
79154b1b CM |
340 | return h; |
341 | } | |
342 | ||
f9295749 | 343 | struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, |
a22285a6 | 344 | int num_items) |
f9295749 | 345 | { |
a22285a6 | 346 | return start_transaction(root, num_items, TRANS_START); |
f9295749 | 347 | } |
7a7eaa40 | 348 | struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root) |
f9295749 | 349 | { |
a22285a6 | 350 | return start_transaction(root, 0, TRANS_JOIN); |
f9295749 CM |
351 | } |
352 | ||
7a7eaa40 | 353 | struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root) |
0af3d00b JB |
354 | { |
355 | return start_transaction(root, 0, TRANS_JOIN_NOLOCK); | |
356 | } | |
357 | ||
7a7eaa40 | 358 | struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root) |
9ca9ee09 | 359 | { |
7a7eaa40 | 360 | return start_transaction(root, 0, TRANS_USERSPACE); |
9ca9ee09 SW |
361 | } |
362 | ||
d352ac68 | 363 | /* wait for a transaction commit to be fully complete */ |
b9c8300c | 364 | static noinline void wait_for_commit(struct btrfs_root *root, |
89ce8a63 CM |
365 | struct btrfs_transaction *commit) |
366 | { | |
72d63ed6 | 367 | wait_event(commit->commit_wait, commit->commit_done); |
89ce8a63 CM |
368 | } |
369 | ||
46204592 SW |
370 | int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid) |
371 | { | |
372 | struct btrfs_transaction *cur_trans = NULL, *t; | |
373 | int ret; | |
374 | ||
46204592 SW |
375 | ret = 0; |
376 | if (transid) { | |
377 | if (transid <= root->fs_info->last_trans_committed) | |
a4abeea4 | 378 | goto out; |
46204592 SW |
379 | |
380 | /* find specified transaction */ | |
a4abeea4 | 381 | spin_lock(&root->fs_info->trans_lock); |
46204592 SW |
382 | list_for_each_entry(t, &root->fs_info->trans_list, list) { |
383 | if (t->transid == transid) { | |
384 | cur_trans = t; | |
a4abeea4 | 385 | atomic_inc(&cur_trans->use_count); |
46204592 SW |
386 | break; |
387 | } | |
388 | if (t->transid > transid) | |
389 | break; | |
390 | } | |
a4abeea4 | 391 | spin_unlock(&root->fs_info->trans_lock); |
46204592 SW |
392 | ret = -EINVAL; |
393 | if (!cur_trans) | |
a4abeea4 | 394 | goto out; /* bad transid */ |
46204592 SW |
395 | } else { |
396 | /* find newest transaction that is committing | committed */ | |
a4abeea4 | 397 | spin_lock(&root->fs_info->trans_lock); |
46204592 SW |
398 | list_for_each_entry_reverse(t, &root->fs_info->trans_list, |
399 | list) { | |
400 | if (t->in_commit) { | |
401 | if (t->commit_done) | |
3473f3c0 | 402 | break; |
46204592 | 403 | cur_trans = t; |
a4abeea4 | 404 | atomic_inc(&cur_trans->use_count); |
46204592 SW |
405 | break; |
406 | } | |
407 | } | |
a4abeea4 | 408 | spin_unlock(&root->fs_info->trans_lock); |
46204592 | 409 | if (!cur_trans) |
a4abeea4 | 410 | goto out; /* nothing committing|committed */ |
46204592 SW |
411 | } |
412 | ||
46204592 SW |
413 | wait_for_commit(root, cur_trans); |
414 | ||
46204592 SW |
415 | put_transaction(cur_trans); |
416 | ret = 0; | |
a4abeea4 | 417 | out: |
46204592 SW |
418 | return ret; |
419 | } | |
420 | ||
37d1aeee CM |
421 | void btrfs_throttle(struct btrfs_root *root) |
422 | { | |
a4abeea4 | 423 | if (!atomic_read(&root->fs_info->open_ioctl_trans)) |
9ca9ee09 | 424 | wait_current_trans(root); |
37d1aeee CM |
425 | } |
426 | ||
8929ecfa YZ |
427 | static int should_end_transaction(struct btrfs_trans_handle *trans, |
428 | struct btrfs_root *root) | |
429 | { | |
430 | int ret; | |
36ba022a JB |
431 | |
432 | ret = btrfs_block_rsv_check(root, &root->fs_info->global_block_rsv, 5); | |
8929ecfa YZ |
433 | return ret ? 1 : 0; |
434 | } | |
435 | ||
436 | int btrfs_should_end_transaction(struct btrfs_trans_handle *trans, | |
437 | struct btrfs_root *root) | |
438 | { | |
439 | struct btrfs_transaction *cur_trans = trans->transaction; | |
9c8d86db | 440 | struct btrfs_block_rsv *rsv = trans->block_rsv; |
8929ecfa YZ |
441 | int updates; |
442 | ||
a4abeea4 | 443 | smp_mb(); |
8929ecfa YZ |
444 | if (cur_trans->blocked || cur_trans->delayed_refs.flushing) |
445 | return 1; | |
446 | ||
9c8d86db JB |
447 | /* |
448 | * We need to do this in case we're deleting csums so the global block | |
449 | * rsv get's used instead of the csum block rsv. | |
450 | */ | |
451 | trans->block_rsv = NULL; | |
452 | ||
8929ecfa YZ |
453 | updates = trans->delayed_ref_updates; |
454 | trans->delayed_ref_updates = 0; | |
455 | if (updates) | |
456 | btrfs_run_delayed_refs(trans, root, updates); | |
457 | ||
9c8d86db JB |
458 | trans->block_rsv = rsv; |
459 | ||
8929ecfa YZ |
460 | return should_end_transaction(trans, root); |
461 | } | |
462 | ||
89ce8a63 | 463 | static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, |
0af3d00b | 464 | struct btrfs_root *root, int throttle, int lock) |
79154b1b | 465 | { |
8929ecfa | 466 | struct btrfs_transaction *cur_trans = trans->transaction; |
ab78c84d | 467 | struct btrfs_fs_info *info = root->fs_info; |
c3e69d58 CM |
468 | int count = 0; |
469 | ||
2a1eb461 JB |
470 | if (--trans->use_count) { |
471 | trans->block_rsv = trans->orig_rsv; | |
472 | return 0; | |
473 | } | |
474 | ||
b24e03db | 475 | btrfs_trans_release_metadata(trans, root); |
4c13d758 | 476 | trans->block_rsv = NULL; |
203bf287 | 477 | while (count < 2) { |
c3e69d58 CM |
478 | unsigned long cur = trans->delayed_ref_updates; |
479 | trans->delayed_ref_updates = 0; | |
480 | if (cur && | |
481 | trans->transaction->delayed_refs.num_heads_ready > 64) { | |
482 | trans->delayed_ref_updates = 0; | |
483 | btrfs_run_delayed_refs(trans, root, cur); | |
484 | } else { | |
485 | break; | |
486 | } | |
487 | count++; | |
56bec294 CM |
488 | } |
489 | ||
a4abeea4 JB |
490 | if (lock && !atomic_read(&root->fs_info->open_ioctl_trans) && |
491 | should_end_transaction(trans, root)) { | |
8929ecfa | 492 | trans->transaction->blocked = 1; |
a4abeea4 JB |
493 | smp_wmb(); |
494 | } | |
8929ecfa | 495 | |
0af3d00b | 496 | if (lock && cur_trans->blocked && !cur_trans->in_commit) { |
81317fde JB |
497 | if (throttle) { |
498 | /* | |
499 | * We may race with somebody else here so end up having | |
500 | * to call end_transaction on ourselves again, so inc | |
501 | * our use_count. | |
502 | */ | |
503 | trans->use_count++; | |
8929ecfa | 504 | return btrfs_commit_transaction(trans, root); |
81317fde | 505 | } else { |
8929ecfa | 506 | wake_up_process(info->transaction_kthread); |
81317fde | 507 | } |
8929ecfa YZ |
508 | } |
509 | ||
8929ecfa | 510 | WARN_ON(cur_trans != info->running_transaction); |
13c5a93e JB |
511 | WARN_ON(atomic_read(&cur_trans->num_writers) < 1); |
512 | atomic_dec(&cur_trans->num_writers); | |
89ce8a63 | 513 | |
99d16cbc | 514 | smp_mb(); |
79154b1b CM |
515 | if (waitqueue_active(&cur_trans->writer_wait)) |
516 | wake_up(&cur_trans->writer_wait); | |
79154b1b | 517 | put_transaction(cur_trans); |
9ed74f2d JB |
518 | |
519 | if (current->journal_info == trans) | |
520 | current->journal_info = NULL; | |
d6025579 | 521 | memset(trans, 0, sizeof(*trans)); |
2c90e5d6 | 522 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
ab78c84d | 523 | |
24bbcf04 YZ |
524 | if (throttle) |
525 | btrfs_run_delayed_iputs(root); | |
526 | ||
79154b1b CM |
527 | return 0; |
528 | } | |
529 | ||
89ce8a63 CM |
530 | int btrfs_end_transaction(struct btrfs_trans_handle *trans, |
531 | struct btrfs_root *root) | |
532 | { | |
16cdcec7 MX |
533 | int ret; |
534 | ||
535 | ret = __btrfs_end_transaction(trans, root, 0, 1); | |
536 | if (ret) | |
537 | return ret; | |
538 | return 0; | |
89ce8a63 CM |
539 | } |
540 | ||
541 | int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans, | |
542 | struct btrfs_root *root) | |
543 | { | |
16cdcec7 MX |
544 | int ret; |
545 | ||
546 | ret = __btrfs_end_transaction(trans, root, 1, 1); | |
547 | if (ret) | |
548 | return ret; | |
549 | return 0; | |
0af3d00b JB |
550 | } |
551 | ||
552 | int btrfs_end_transaction_nolock(struct btrfs_trans_handle *trans, | |
553 | struct btrfs_root *root) | |
554 | { | |
16cdcec7 MX |
555 | int ret; |
556 | ||
557 | ret = __btrfs_end_transaction(trans, root, 0, 0); | |
558 | if (ret) | |
559 | return ret; | |
560 | return 0; | |
561 | } | |
562 | ||
563 | int btrfs_end_transaction_dmeta(struct btrfs_trans_handle *trans, | |
564 | struct btrfs_root *root) | |
565 | { | |
566 | return __btrfs_end_transaction(trans, root, 1, 1); | |
89ce8a63 CM |
567 | } |
568 | ||
d352ac68 CM |
569 | /* |
570 | * when btree blocks are allocated, they have some corresponding bits set for | |
571 | * them in one of two extent_io trees. This is used to make sure all of | |
690587d1 | 572 | * those extents are sent to disk but does not wait on them |
d352ac68 | 573 | */ |
690587d1 | 574 | int btrfs_write_marked_extents(struct btrfs_root *root, |
8cef4e16 | 575 | struct extent_io_tree *dirty_pages, int mark) |
79154b1b | 576 | { |
777e6bd7 | 577 | int err = 0; |
7c4452b9 | 578 | int werr = 0; |
1728366e | 579 | struct address_space *mapping = root->fs_info->btree_inode->i_mapping; |
777e6bd7 | 580 | u64 start = 0; |
5f39d397 | 581 | u64 end; |
7c4452b9 | 582 | |
1728366e JB |
583 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
584 | mark)) { | |
585 | convert_extent_bit(dirty_pages, start, end, EXTENT_NEED_WAIT, mark, | |
586 | GFP_NOFS); | |
587 | err = filemap_fdatawrite_range(mapping, start, end); | |
588 | if (err) | |
589 | werr = err; | |
590 | cond_resched(); | |
591 | start = end + 1; | |
7c4452b9 | 592 | } |
690587d1 CM |
593 | if (err) |
594 | werr = err; | |
595 | return werr; | |
596 | } | |
597 | ||
598 | /* | |
599 | * when btree blocks are allocated, they have some corresponding bits set for | |
600 | * them in one of two extent_io trees. This is used to make sure all of | |
601 | * those extents are on disk for transaction or log commit. We wait | |
602 | * on all the pages and clear them from the dirty pages state tree | |
603 | */ | |
604 | int btrfs_wait_marked_extents(struct btrfs_root *root, | |
8cef4e16 | 605 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 | 606 | { |
690587d1 CM |
607 | int err = 0; |
608 | int werr = 0; | |
1728366e | 609 | struct address_space *mapping = root->fs_info->btree_inode->i_mapping; |
690587d1 CM |
610 | u64 start = 0; |
611 | u64 end; | |
777e6bd7 | 612 | |
1728366e JB |
613 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
614 | EXTENT_NEED_WAIT)) { | |
615 | clear_extent_bits(dirty_pages, start, end, EXTENT_NEED_WAIT, GFP_NOFS); | |
616 | err = filemap_fdatawait_range(mapping, start, end); | |
617 | if (err) | |
618 | werr = err; | |
619 | cond_resched(); | |
620 | start = end + 1; | |
777e6bd7 | 621 | } |
7c4452b9 CM |
622 | if (err) |
623 | werr = err; | |
624 | return werr; | |
79154b1b CM |
625 | } |
626 | ||
690587d1 CM |
627 | /* |
628 | * when btree blocks are allocated, they have some corresponding bits set for | |
629 | * them in one of two extent_io trees. This is used to make sure all of | |
630 | * those extents are on disk for transaction or log commit | |
631 | */ | |
632 | int btrfs_write_and_wait_marked_extents(struct btrfs_root *root, | |
8cef4e16 | 633 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 CM |
634 | { |
635 | int ret; | |
636 | int ret2; | |
637 | ||
8cef4e16 YZ |
638 | ret = btrfs_write_marked_extents(root, dirty_pages, mark); |
639 | ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark); | |
bf0da8c1 CM |
640 | |
641 | if (ret) | |
642 | return ret; | |
643 | if (ret2) | |
644 | return ret2; | |
645 | return 0; | |
690587d1 CM |
646 | } |
647 | ||
d0c803c4 CM |
648 | int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans, |
649 | struct btrfs_root *root) | |
650 | { | |
651 | if (!trans || !trans->transaction) { | |
652 | struct inode *btree_inode; | |
653 | btree_inode = root->fs_info->btree_inode; | |
654 | return filemap_write_and_wait(btree_inode->i_mapping); | |
655 | } | |
656 | return btrfs_write_and_wait_marked_extents(root, | |
8cef4e16 YZ |
657 | &trans->transaction->dirty_pages, |
658 | EXTENT_DIRTY); | |
d0c803c4 CM |
659 | } |
660 | ||
d352ac68 CM |
661 | /* |
662 | * this is used to update the root pointer in the tree of tree roots. | |
663 | * | |
664 | * But, in the case of the extent allocation tree, updating the root | |
665 | * pointer may allocate blocks which may change the root of the extent | |
666 | * allocation tree. | |
667 | * | |
668 | * So, this loops and repeats and makes sure the cowonly root didn't | |
669 | * change while the root pointer was being updated in the metadata. | |
670 | */ | |
0b86a832 CM |
671 | static int update_cowonly_root(struct btrfs_trans_handle *trans, |
672 | struct btrfs_root *root) | |
79154b1b CM |
673 | { |
674 | int ret; | |
0b86a832 | 675 | u64 old_root_bytenr; |
86b9f2ec | 676 | u64 old_root_used; |
0b86a832 | 677 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
79154b1b | 678 | |
86b9f2ec | 679 | old_root_used = btrfs_root_used(&root->root_item); |
0b86a832 | 680 | btrfs_write_dirty_block_groups(trans, root); |
56bec294 | 681 | |
d397712b | 682 | while (1) { |
0b86a832 | 683 | old_root_bytenr = btrfs_root_bytenr(&root->root_item); |
86b9f2ec YZ |
684 | if (old_root_bytenr == root->node->start && |
685 | old_root_used == btrfs_root_used(&root->root_item)) | |
79154b1b | 686 | break; |
87ef2bb4 | 687 | |
5d4f98a2 | 688 | btrfs_set_root_node(&root->root_item, root->node); |
79154b1b | 689 | ret = btrfs_update_root(trans, tree_root, |
0b86a832 CM |
690 | &root->root_key, |
691 | &root->root_item); | |
79154b1b | 692 | BUG_ON(ret); |
56bec294 | 693 | |
86b9f2ec | 694 | old_root_used = btrfs_root_used(&root->root_item); |
4a8c9a62 | 695 | ret = btrfs_write_dirty_block_groups(trans, root); |
56bec294 | 696 | BUG_ON(ret); |
0b86a832 | 697 | } |
276e680d YZ |
698 | |
699 | if (root != root->fs_info->extent_root) | |
700 | switch_commit_root(root); | |
701 | ||
0b86a832 CM |
702 | return 0; |
703 | } | |
704 | ||
d352ac68 CM |
705 | /* |
706 | * update all the cowonly tree roots on disk | |
707 | */ | |
5d4f98a2 YZ |
708 | static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans, |
709 | struct btrfs_root *root) | |
0b86a832 CM |
710 | { |
711 | struct btrfs_fs_info *fs_info = root->fs_info; | |
712 | struct list_head *next; | |
84234f3a | 713 | struct extent_buffer *eb; |
56bec294 | 714 | int ret; |
84234f3a | 715 | |
56bec294 CM |
716 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
717 | BUG_ON(ret); | |
87ef2bb4 | 718 | |
84234f3a | 719 | eb = btrfs_lock_root_node(fs_info->tree_root); |
9fa8cfe7 | 720 | btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 0, &eb); |
84234f3a YZ |
721 | btrfs_tree_unlock(eb); |
722 | free_extent_buffer(eb); | |
0b86a832 | 723 | |
56bec294 CM |
724 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
725 | BUG_ON(ret); | |
87ef2bb4 | 726 | |
d397712b | 727 | while (!list_empty(&fs_info->dirty_cowonly_roots)) { |
0b86a832 CM |
728 | next = fs_info->dirty_cowonly_roots.next; |
729 | list_del_init(next); | |
730 | root = list_entry(next, struct btrfs_root, dirty_list); | |
87ef2bb4 | 731 | |
0b86a832 | 732 | update_cowonly_root(trans, root); |
79154b1b | 733 | } |
276e680d YZ |
734 | |
735 | down_write(&fs_info->extent_commit_sem); | |
736 | switch_commit_root(fs_info->extent_root); | |
737 | up_write(&fs_info->extent_commit_sem); | |
738 | ||
79154b1b CM |
739 | return 0; |
740 | } | |
741 | ||
d352ac68 CM |
742 | /* |
743 | * dead roots are old snapshots that need to be deleted. This allocates | |
744 | * a dirty root struct and adds it into the list of dead roots that need to | |
745 | * be deleted | |
746 | */ | |
5d4f98a2 | 747 | int btrfs_add_dead_root(struct btrfs_root *root) |
5eda7b5e | 748 | { |
a4abeea4 | 749 | spin_lock(&root->fs_info->trans_lock); |
5d4f98a2 | 750 | list_add(&root->root_list, &root->fs_info->dead_roots); |
a4abeea4 | 751 | spin_unlock(&root->fs_info->trans_lock); |
5eda7b5e CM |
752 | return 0; |
753 | } | |
754 | ||
d352ac68 | 755 | /* |
5d4f98a2 | 756 | * update all the cowonly tree roots on disk |
d352ac68 | 757 | */ |
5d4f98a2 YZ |
758 | static noinline int commit_fs_roots(struct btrfs_trans_handle *trans, |
759 | struct btrfs_root *root) | |
0f7d52f4 | 760 | { |
0f7d52f4 | 761 | struct btrfs_root *gang[8]; |
5d4f98a2 | 762 | struct btrfs_fs_info *fs_info = root->fs_info; |
0f7d52f4 CM |
763 | int i; |
764 | int ret; | |
54aa1f4d CM |
765 | int err = 0; |
766 | ||
a4abeea4 | 767 | spin_lock(&fs_info->fs_roots_radix_lock); |
d397712b | 768 | while (1) { |
5d4f98a2 YZ |
769 | ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, |
770 | (void **)gang, 0, | |
0f7d52f4 CM |
771 | ARRAY_SIZE(gang), |
772 | BTRFS_ROOT_TRANS_TAG); | |
773 | if (ret == 0) | |
774 | break; | |
775 | for (i = 0; i < ret; i++) { | |
776 | root = gang[i]; | |
5d4f98a2 YZ |
777 | radix_tree_tag_clear(&fs_info->fs_roots_radix, |
778 | (unsigned long)root->root_key.objectid, | |
779 | BTRFS_ROOT_TRANS_TAG); | |
a4abeea4 | 780 | spin_unlock(&fs_info->fs_roots_radix_lock); |
31153d81 | 781 | |
e02119d5 | 782 | btrfs_free_log(trans, root); |
5d4f98a2 | 783 | btrfs_update_reloc_root(trans, root); |
d68fc57b | 784 | btrfs_orphan_commit_root(trans, root); |
bcc63abb | 785 | |
82d5902d LZ |
786 | btrfs_save_ino_cache(root, trans); |
787 | ||
f1ebcc74 LB |
788 | /* see comments in should_cow_block() */ |
789 | root->force_cow = 0; | |
790 | smp_wmb(); | |
791 | ||
978d910d | 792 | if (root->commit_root != root->node) { |
581bb050 | 793 | mutex_lock(&root->fs_commit_mutex); |
817d52f8 | 794 | switch_commit_root(root); |
581bb050 LZ |
795 | btrfs_unpin_free_ino(root); |
796 | mutex_unlock(&root->fs_commit_mutex); | |
797 | ||
978d910d YZ |
798 | btrfs_set_root_node(&root->root_item, |
799 | root->node); | |
800 | } | |
5d4f98a2 | 801 | |
5d4f98a2 | 802 | err = btrfs_update_root(trans, fs_info->tree_root, |
0f7d52f4 CM |
803 | &root->root_key, |
804 | &root->root_item); | |
a4abeea4 | 805 | spin_lock(&fs_info->fs_roots_radix_lock); |
54aa1f4d CM |
806 | if (err) |
807 | break; | |
0f7d52f4 CM |
808 | } |
809 | } | |
a4abeea4 | 810 | spin_unlock(&fs_info->fs_roots_radix_lock); |
54aa1f4d | 811 | return err; |
0f7d52f4 CM |
812 | } |
813 | ||
d352ac68 CM |
814 | /* |
815 | * defrag a given btree. If cacheonly == 1, this won't read from the disk, | |
816 | * otherwise every leaf in the btree is read and defragged. | |
817 | */ | |
e9d0b13b CM |
818 | int btrfs_defrag_root(struct btrfs_root *root, int cacheonly) |
819 | { | |
820 | struct btrfs_fs_info *info = root->fs_info; | |
e9d0b13b | 821 | struct btrfs_trans_handle *trans; |
8929ecfa | 822 | int ret; |
d3c2fdcf | 823 | unsigned long nr; |
e9d0b13b | 824 | |
8929ecfa | 825 | if (xchg(&root->defrag_running, 1)) |
e9d0b13b | 826 | return 0; |
8929ecfa | 827 | |
6b80053d | 828 | while (1) { |
8929ecfa YZ |
829 | trans = btrfs_start_transaction(root, 0); |
830 | if (IS_ERR(trans)) | |
831 | return PTR_ERR(trans); | |
832 | ||
e9d0b13b | 833 | ret = btrfs_defrag_leaves(trans, root, cacheonly); |
8929ecfa | 834 | |
d3c2fdcf | 835 | nr = trans->blocks_used; |
e9d0b13b | 836 | btrfs_end_transaction(trans, root); |
d3c2fdcf | 837 | btrfs_btree_balance_dirty(info->tree_root, nr); |
e9d0b13b CM |
838 | cond_resched(); |
839 | ||
7841cb28 | 840 | if (btrfs_fs_closing(root->fs_info) || ret != -EAGAIN) |
e9d0b13b CM |
841 | break; |
842 | } | |
843 | root->defrag_running = 0; | |
8929ecfa | 844 | return ret; |
e9d0b13b CM |
845 | } |
846 | ||
d352ac68 CM |
847 | /* |
848 | * new snapshots need to be created at a very specific time in the | |
849 | * transaction commit. This does the actual creation | |
850 | */ | |
80b6794d | 851 | static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, |
3063d29f CM |
852 | struct btrfs_fs_info *fs_info, |
853 | struct btrfs_pending_snapshot *pending) | |
854 | { | |
855 | struct btrfs_key key; | |
80b6794d | 856 | struct btrfs_root_item *new_root_item; |
3063d29f CM |
857 | struct btrfs_root *tree_root = fs_info->tree_root; |
858 | struct btrfs_root *root = pending->root; | |
6bdb72de | 859 | struct btrfs_root *parent_root; |
98c9942a | 860 | struct btrfs_block_rsv *rsv; |
6bdb72de | 861 | struct inode *parent_inode; |
6a912213 | 862 | struct dentry *parent; |
a22285a6 | 863 | struct dentry *dentry; |
3063d29f | 864 | struct extent_buffer *tmp; |
925baedd | 865 | struct extent_buffer *old; |
3063d29f | 866 | int ret; |
d68fc57b | 867 | u64 to_reserve = 0; |
6bdb72de | 868 | u64 index = 0; |
a22285a6 | 869 | u64 objectid; |
b83cc969 | 870 | u64 root_flags; |
3063d29f | 871 | |
98c9942a LB |
872 | rsv = trans->block_rsv; |
873 | ||
80b6794d CM |
874 | new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS); |
875 | if (!new_root_item) { | |
a22285a6 | 876 | pending->error = -ENOMEM; |
80b6794d CM |
877 | goto fail; |
878 | } | |
a22285a6 | 879 | |
581bb050 | 880 | ret = btrfs_find_free_objectid(tree_root, &objectid); |
a22285a6 YZ |
881 | if (ret) { |
882 | pending->error = ret; | |
3063d29f | 883 | goto fail; |
a22285a6 | 884 | } |
3063d29f | 885 | |
3fd0a558 | 886 | btrfs_reloc_pre_snapshot(trans, pending, &to_reserve); |
d68fc57b YZ |
887 | |
888 | if (to_reserve > 0) { | |
62f30c54 MX |
889 | ret = btrfs_block_rsv_add_noflush(root, &pending->block_rsv, |
890 | to_reserve); | |
d68fc57b YZ |
891 | if (ret) { |
892 | pending->error = ret; | |
893 | goto fail; | |
894 | } | |
895 | } | |
896 | ||
3063d29f | 897 | key.objectid = objectid; |
a22285a6 YZ |
898 | key.offset = (u64)-1; |
899 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3063d29f | 900 | |
a22285a6 | 901 | trans->block_rsv = &pending->block_rsv; |
3de4586c | 902 | |
a22285a6 | 903 | dentry = pending->dentry; |
6a912213 JB |
904 | parent = dget_parent(dentry); |
905 | parent_inode = parent->d_inode; | |
a22285a6 | 906 | parent_root = BTRFS_I(parent_inode)->root; |
7585717f | 907 | record_root_in_trans(trans, parent_root); |
a22285a6 | 908 | |
3063d29f CM |
909 | /* |
910 | * insert the directory item | |
911 | */ | |
3de4586c | 912 | ret = btrfs_set_inode_index(parent_inode, &index); |
6bdb72de | 913 | BUG_ON(ret); |
0660b5af | 914 | ret = btrfs_insert_dir_item(trans, parent_root, |
a22285a6 | 915 | dentry->d_name.name, dentry->d_name.len, |
16cdcec7 | 916 | parent_inode, &key, |
a22285a6 | 917 | BTRFS_FT_DIR, index); |
6bdb72de | 918 | BUG_ON(ret); |
0660b5af | 919 | |
a22285a6 YZ |
920 | btrfs_i_size_write(parent_inode, parent_inode->i_size + |
921 | dentry->d_name.len * 2); | |
52c26179 YZ |
922 | ret = btrfs_update_inode(trans, parent_root, parent_inode); |
923 | BUG_ON(ret); | |
924 | ||
e999376f CM |
925 | /* |
926 | * pull in the delayed directory update | |
927 | * and the delayed inode item | |
928 | * otherwise we corrupt the FS during | |
929 | * snapshot | |
930 | */ | |
931 | ret = btrfs_run_delayed_items(trans, root); | |
932 | BUG_ON(ret); | |
933 | ||
7585717f | 934 | record_root_in_trans(trans, root); |
6bdb72de SW |
935 | btrfs_set_root_last_snapshot(&root->root_item, trans->transid); |
936 | memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); | |
08fe4db1 | 937 | btrfs_check_and_init_root_item(new_root_item); |
6bdb72de | 938 | |
b83cc969 LZ |
939 | root_flags = btrfs_root_flags(new_root_item); |
940 | if (pending->readonly) | |
941 | root_flags |= BTRFS_ROOT_SUBVOL_RDONLY; | |
942 | else | |
943 | root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY; | |
944 | btrfs_set_root_flags(new_root_item, root_flags); | |
945 | ||
6bdb72de SW |
946 | old = btrfs_lock_root_node(root); |
947 | btrfs_cow_block(trans, root, old, NULL, 0, &old); | |
948 | btrfs_set_lock_blocking(old); | |
949 | ||
950 | btrfs_copy_root(trans, root, old, &tmp, objectid); | |
951 | btrfs_tree_unlock(old); | |
952 | free_extent_buffer(old); | |
953 | ||
f1ebcc74 LB |
954 | /* see comments in should_cow_block() */ |
955 | root->force_cow = 1; | |
956 | smp_wmb(); | |
957 | ||
6bdb72de | 958 | btrfs_set_root_node(new_root_item, tmp); |
a22285a6 YZ |
959 | /* record when the snapshot was created in key.offset */ |
960 | key.offset = trans->transid; | |
961 | ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); | |
6bdb72de SW |
962 | btrfs_tree_unlock(tmp); |
963 | free_extent_buffer(tmp); | |
a22285a6 | 964 | BUG_ON(ret); |
6bdb72de | 965 | |
a22285a6 YZ |
966 | /* |
967 | * insert root back/forward references | |
968 | */ | |
969 | ret = btrfs_add_root_ref(trans, tree_root, objectid, | |
0660b5af | 970 | parent_root->root_key.objectid, |
33345d01 | 971 | btrfs_ino(parent_inode), index, |
a22285a6 | 972 | dentry->d_name.name, dentry->d_name.len); |
0660b5af | 973 | BUG_ON(ret); |
6a912213 | 974 | dput(parent); |
0660b5af | 975 | |
a22285a6 YZ |
976 | key.offset = (u64)-1; |
977 | pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key); | |
978 | BUG_ON(IS_ERR(pending->snap)); | |
d68fc57b | 979 | |
3fd0a558 | 980 | btrfs_reloc_post_snapshot(trans, pending); |
3063d29f | 981 | fail: |
6bdb72de | 982 | kfree(new_root_item); |
98c9942a | 983 | trans->block_rsv = rsv; |
a22285a6 YZ |
984 | btrfs_block_rsv_release(root, &pending->block_rsv, (u64)-1); |
985 | return 0; | |
3063d29f CM |
986 | } |
987 | ||
d352ac68 CM |
988 | /* |
989 | * create all the snapshots we've scheduled for creation | |
990 | */ | |
80b6794d CM |
991 | static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans, |
992 | struct btrfs_fs_info *fs_info) | |
3de4586c CM |
993 | { |
994 | struct btrfs_pending_snapshot *pending; | |
995 | struct list_head *head = &trans->transaction->pending_snapshots; | |
3de4586c CM |
996 | int ret; |
997 | ||
c6e30871 | 998 | list_for_each_entry(pending, head, list) { |
3de4586c CM |
999 | ret = create_pending_snapshot(trans, fs_info, pending); |
1000 | BUG_ON(ret); | |
1001 | } | |
1002 | return 0; | |
1003 | } | |
1004 | ||
5d4f98a2 YZ |
1005 | static void update_super_roots(struct btrfs_root *root) |
1006 | { | |
1007 | struct btrfs_root_item *root_item; | |
1008 | struct btrfs_super_block *super; | |
1009 | ||
6c41761f | 1010 | super = root->fs_info->super_copy; |
5d4f98a2 YZ |
1011 | |
1012 | root_item = &root->fs_info->chunk_root->root_item; | |
1013 | super->chunk_root = root_item->bytenr; | |
1014 | super->chunk_root_generation = root_item->generation; | |
1015 | super->chunk_root_level = root_item->level; | |
1016 | ||
1017 | root_item = &root->fs_info->tree_root->root_item; | |
1018 | super->root = root_item->bytenr; | |
1019 | super->generation = root_item->generation; | |
1020 | super->root_level = root_item->level; | |
73bc1876 | 1021 | if (btrfs_test_opt(root, SPACE_CACHE)) |
0af3d00b | 1022 | super->cache_generation = root_item->generation; |
5d4f98a2 YZ |
1023 | } |
1024 | ||
f36f3042 CM |
1025 | int btrfs_transaction_in_commit(struct btrfs_fs_info *info) |
1026 | { | |
1027 | int ret = 0; | |
a4abeea4 | 1028 | spin_lock(&info->trans_lock); |
f36f3042 CM |
1029 | if (info->running_transaction) |
1030 | ret = info->running_transaction->in_commit; | |
a4abeea4 | 1031 | spin_unlock(&info->trans_lock); |
f36f3042 CM |
1032 | return ret; |
1033 | } | |
1034 | ||
8929ecfa YZ |
1035 | int btrfs_transaction_blocked(struct btrfs_fs_info *info) |
1036 | { | |
1037 | int ret = 0; | |
a4abeea4 | 1038 | spin_lock(&info->trans_lock); |
8929ecfa YZ |
1039 | if (info->running_transaction) |
1040 | ret = info->running_transaction->blocked; | |
a4abeea4 | 1041 | spin_unlock(&info->trans_lock); |
8929ecfa YZ |
1042 | return ret; |
1043 | } | |
1044 | ||
bb9c12c9 SW |
1045 | /* |
1046 | * wait for the current transaction commit to start and block subsequent | |
1047 | * transaction joins | |
1048 | */ | |
1049 | static void wait_current_trans_commit_start(struct btrfs_root *root, | |
1050 | struct btrfs_transaction *trans) | |
1051 | { | |
72d63ed6 | 1052 | wait_event(root->fs_info->transaction_blocked_wait, trans->in_commit); |
bb9c12c9 SW |
1053 | } |
1054 | ||
1055 | /* | |
1056 | * wait for the current transaction to start and then become unblocked. | |
1057 | * caller holds ref. | |
1058 | */ | |
1059 | static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root, | |
1060 | struct btrfs_transaction *trans) | |
1061 | { | |
72d63ed6 LZ |
1062 | wait_event(root->fs_info->transaction_wait, |
1063 | trans->commit_done || (trans->in_commit && !trans->blocked)); | |
bb9c12c9 SW |
1064 | } |
1065 | ||
1066 | /* | |
1067 | * commit transactions asynchronously. once btrfs_commit_transaction_async | |
1068 | * returns, any subsequent transaction will not be allowed to join. | |
1069 | */ | |
1070 | struct btrfs_async_commit { | |
1071 | struct btrfs_trans_handle *newtrans; | |
1072 | struct btrfs_root *root; | |
1073 | struct delayed_work work; | |
1074 | }; | |
1075 | ||
1076 | static void do_async_commit(struct work_struct *work) | |
1077 | { | |
1078 | struct btrfs_async_commit *ac = | |
1079 | container_of(work, struct btrfs_async_commit, work.work); | |
1080 | ||
1081 | btrfs_commit_transaction(ac->newtrans, ac->root); | |
1082 | kfree(ac); | |
1083 | } | |
1084 | ||
1085 | int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans, | |
1086 | struct btrfs_root *root, | |
1087 | int wait_for_unblock) | |
1088 | { | |
1089 | struct btrfs_async_commit *ac; | |
1090 | struct btrfs_transaction *cur_trans; | |
1091 | ||
1092 | ac = kmalloc(sizeof(*ac), GFP_NOFS); | |
db5b493a TI |
1093 | if (!ac) |
1094 | return -ENOMEM; | |
bb9c12c9 SW |
1095 | |
1096 | INIT_DELAYED_WORK(&ac->work, do_async_commit); | |
1097 | ac->root = root; | |
7a7eaa40 | 1098 | ac->newtrans = btrfs_join_transaction(root); |
3612b495 TI |
1099 | if (IS_ERR(ac->newtrans)) { |
1100 | int err = PTR_ERR(ac->newtrans); | |
1101 | kfree(ac); | |
1102 | return err; | |
1103 | } | |
bb9c12c9 SW |
1104 | |
1105 | /* take transaction reference */ | |
bb9c12c9 | 1106 | cur_trans = trans->transaction; |
13c5a93e | 1107 | atomic_inc(&cur_trans->use_count); |
bb9c12c9 SW |
1108 | |
1109 | btrfs_end_transaction(trans, root); | |
1110 | schedule_delayed_work(&ac->work, 0); | |
1111 | ||
1112 | /* wait for transaction to start and unblock */ | |
bb9c12c9 SW |
1113 | if (wait_for_unblock) |
1114 | wait_current_trans_commit_start_and_unblock(root, cur_trans); | |
1115 | else | |
1116 | wait_current_trans_commit_start(root, cur_trans); | |
bb9c12c9 | 1117 | |
38e88054 SW |
1118 | if (current->journal_info == trans) |
1119 | current->journal_info = NULL; | |
1120 | ||
1121 | put_transaction(cur_trans); | |
bb9c12c9 SW |
1122 | return 0; |
1123 | } | |
1124 | ||
1125 | /* | |
1126 | * btrfs_transaction state sequence: | |
1127 | * in_commit = 0, blocked = 0 (initial) | |
1128 | * in_commit = 1, blocked = 1 | |
1129 | * blocked = 0 | |
1130 | * commit_done = 1 | |
1131 | */ | |
79154b1b CM |
1132 | int btrfs_commit_transaction(struct btrfs_trans_handle *trans, |
1133 | struct btrfs_root *root) | |
1134 | { | |
15ee9bc7 | 1135 | unsigned long joined = 0; |
79154b1b | 1136 | struct btrfs_transaction *cur_trans; |
8fd17795 | 1137 | struct btrfs_transaction *prev_trans = NULL; |
79154b1b | 1138 | DEFINE_WAIT(wait); |
15ee9bc7 | 1139 | int ret; |
89573b9c CM |
1140 | int should_grow = 0; |
1141 | unsigned long now = get_seconds(); | |
dccae999 | 1142 | int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT); |
79154b1b | 1143 | |
5a3f23d5 CM |
1144 | btrfs_run_ordered_operations(root, 0); |
1145 | ||
b24e03db | 1146 | btrfs_trans_release_metadata(trans, root); |
9c8d86db JB |
1147 | trans->block_rsv = NULL; |
1148 | ||
56bec294 CM |
1149 | /* make a pass through all the delayed refs we have so far |
1150 | * any runnings procs may add more while we are here | |
1151 | */ | |
1152 | ret = btrfs_run_delayed_refs(trans, root, 0); | |
1153 | BUG_ON(ret); | |
1154 | ||
b7ec40d7 | 1155 | cur_trans = trans->transaction; |
56bec294 CM |
1156 | /* |
1157 | * set the flushing flag so procs in this transaction have to | |
1158 | * start sending their work down. | |
1159 | */ | |
b7ec40d7 | 1160 | cur_trans->delayed_refs.flushing = 1; |
56bec294 | 1161 | |
c3e69d58 | 1162 | ret = btrfs_run_delayed_refs(trans, root, 0); |
56bec294 CM |
1163 | BUG_ON(ret); |
1164 | ||
a4abeea4 | 1165 | spin_lock(&cur_trans->commit_lock); |
b7ec40d7 | 1166 | if (cur_trans->in_commit) { |
a4abeea4 | 1167 | spin_unlock(&cur_trans->commit_lock); |
13c5a93e | 1168 | atomic_inc(&cur_trans->use_count); |
79154b1b | 1169 | btrfs_end_transaction(trans, root); |
ccd467d6 | 1170 | |
b9c8300c | 1171 | wait_for_commit(root, cur_trans); |
15ee9bc7 | 1172 | |
79154b1b | 1173 | put_transaction(cur_trans); |
15ee9bc7 | 1174 | |
79154b1b CM |
1175 | return 0; |
1176 | } | |
4313b399 | 1177 | |
2c90e5d6 | 1178 | trans->transaction->in_commit = 1; |
f9295749 | 1179 | trans->transaction->blocked = 1; |
a4abeea4 | 1180 | spin_unlock(&cur_trans->commit_lock); |
bb9c12c9 SW |
1181 | wake_up(&root->fs_info->transaction_blocked_wait); |
1182 | ||
a4abeea4 | 1183 | spin_lock(&root->fs_info->trans_lock); |
ccd467d6 CM |
1184 | if (cur_trans->list.prev != &root->fs_info->trans_list) { |
1185 | prev_trans = list_entry(cur_trans->list.prev, | |
1186 | struct btrfs_transaction, list); | |
1187 | if (!prev_trans->commit_done) { | |
13c5a93e | 1188 | atomic_inc(&prev_trans->use_count); |
a4abeea4 | 1189 | spin_unlock(&root->fs_info->trans_lock); |
ccd467d6 CM |
1190 | |
1191 | wait_for_commit(root, prev_trans); | |
ccd467d6 | 1192 | |
15ee9bc7 | 1193 | put_transaction(prev_trans); |
a4abeea4 JB |
1194 | } else { |
1195 | spin_unlock(&root->fs_info->trans_lock); | |
ccd467d6 | 1196 | } |
a4abeea4 JB |
1197 | } else { |
1198 | spin_unlock(&root->fs_info->trans_lock); | |
ccd467d6 | 1199 | } |
15ee9bc7 | 1200 | |
89573b9c CM |
1201 | if (now < cur_trans->start_time || now - cur_trans->start_time < 1) |
1202 | should_grow = 1; | |
1203 | ||
15ee9bc7 | 1204 | do { |
7ea394f1 | 1205 | int snap_pending = 0; |
a4abeea4 | 1206 | |
15ee9bc7 | 1207 | joined = cur_trans->num_joined; |
7ea394f1 YZ |
1208 | if (!list_empty(&trans->transaction->pending_snapshots)) |
1209 | snap_pending = 1; | |
1210 | ||
2c90e5d6 | 1211 | WARN_ON(cur_trans != trans->transaction); |
15ee9bc7 | 1212 | |
0bdb1db2 | 1213 | if (flush_on_commit || snap_pending) { |
24bbcf04 YZ |
1214 | btrfs_start_delalloc_inodes(root, 1); |
1215 | ret = btrfs_wait_ordered_extents(root, 0, 1); | |
ebecd3d9 | 1216 | BUG_ON(ret); |
7ea394f1 YZ |
1217 | } |
1218 | ||
16cdcec7 MX |
1219 | ret = btrfs_run_delayed_items(trans, root); |
1220 | BUG_ON(ret); | |
1221 | ||
5a3f23d5 CM |
1222 | /* |
1223 | * rename don't use btrfs_join_transaction, so, once we | |
1224 | * set the transaction to blocked above, we aren't going | |
1225 | * to get any new ordered operations. We can safely run | |
1226 | * it here and no for sure that nothing new will be added | |
1227 | * to the list | |
1228 | */ | |
1229 | btrfs_run_ordered_operations(root, 1); | |
1230 | ||
ed3b3d31 CM |
1231 | prepare_to_wait(&cur_trans->writer_wait, &wait, |
1232 | TASK_UNINTERRUPTIBLE); | |
1233 | ||
13c5a93e | 1234 | if (atomic_read(&cur_trans->num_writers) > 1) |
99d16cbc SW |
1235 | schedule_timeout(MAX_SCHEDULE_TIMEOUT); |
1236 | else if (should_grow) | |
1237 | schedule_timeout(1); | |
15ee9bc7 | 1238 | |
15ee9bc7 | 1239 | finish_wait(&cur_trans->writer_wait, &wait); |
13c5a93e | 1240 | } while (atomic_read(&cur_trans->num_writers) > 1 || |
89573b9c | 1241 | (should_grow && cur_trans->num_joined != joined)); |
15ee9bc7 | 1242 | |
ed0ca140 JB |
1243 | /* |
1244 | * Ok now we need to make sure to block out any other joins while we | |
1245 | * commit the transaction. We could have started a join before setting | |
1246 | * no_join so make sure to wait for num_writers to == 1 again. | |
1247 | */ | |
1248 | spin_lock(&root->fs_info->trans_lock); | |
1249 | root->fs_info->trans_no_join = 1; | |
1250 | spin_unlock(&root->fs_info->trans_lock); | |
1251 | wait_event(cur_trans->writer_wait, | |
1252 | atomic_read(&cur_trans->num_writers) == 1); | |
1253 | ||
7585717f CM |
1254 | /* |
1255 | * the reloc mutex makes sure that we stop | |
1256 | * the balancing code from coming in and moving | |
1257 | * extents around in the middle of the commit | |
1258 | */ | |
1259 | mutex_lock(&root->fs_info->reloc_mutex); | |
1260 | ||
e999376f | 1261 | ret = btrfs_run_delayed_items(trans, root); |
3063d29f CM |
1262 | BUG_ON(ret); |
1263 | ||
e999376f | 1264 | ret = create_pending_snapshots(trans, root->fs_info); |
16cdcec7 MX |
1265 | BUG_ON(ret); |
1266 | ||
56bec294 CM |
1267 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
1268 | BUG_ON(ret); | |
1269 | ||
e999376f CM |
1270 | /* |
1271 | * make sure none of the code above managed to slip in a | |
1272 | * delayed item | |
1273 | */ | |
1274 | btrfs_assert_delayed_root_empty(root); | |
1275 | ||
2c90e5d6 | 1276 | WARN_ON(cur_trans != trans->transaction); |
dc17ff8f | 1277 | |
a2de733c | 1278 | btrfs_scrub_pause(root); |
e02119d5 CM |
1279 | /* btrfs_commit_tree_roots is responsible for getting the |
1280 | * various roots consistent with each other. Every pointer | |
1281 | * in the tree of tree roots has to point to the most up to date | |
1282 | * root for every subvolume and other tree. So, we have to keep | |
1283 | * the tree logging code from jumping in and changing any | |
1284 | * of the trees. | |
1285 | * | |
1286 | * At this point in the commit, there can't be any tree-log | |
1287 | * writers, but a little lower down we drop the trans mutex | |
1288 | * and let new people in. By holding the tree_log_mutex | |
1289 | * from now until after the super is written, we avoid races | |
1290 | * with the tree-log code. | |
1291 | */ | |
1292 | mutex_lock(&root->fs_info->tree_log_mutex); | |
1293 | ||
5d4f98a2 | 1294 | ret = commit_fs_roots(trans, root); |
54aa1f4d CM |
1295 | BUG_ON(ret); |
1296 | ||
5d4f98a2 | 1297 | /* commit_fs_roots gets rid of all the tree log roots, it is now |
e02119d5 CM |
1298 | * safe to free the root of tree log roots |
1299 | */ | |
1300 | btrfs_free_log_root_tree(trans, root->fs_info); | |
1301 | ||
5d4f98a2 | 1302 | ret = commit_cowonly_roots(trans, root); |
79154b1b | 1303 | BUG_ON(ret); |
54aa1f4d | 1304 | |
11833d66 YZ |
1305 | btrfs_prepare_extent_commit(trans, root); |
1306 | ||
78fae27e | 1307 | cur_trans = root->fs_info->running_transaction; |
5d4f98a2 YZ |
1308 | |
1309 | btrfs_set_root_node(&root->fs_info->tree_root->root_item, | |
1310 | root->fs_info->tree_root->node); | |
817d52f8 | 1311 | switch_commit_root(root->fs_info->tree_root); |
5d4f98a2 YZ |
1312 | |
1313 | btrfs_set_root_node(&root->fs_info->chunk_root->root_item, | |
1314 | root->fs_info->chunk_root->node); | |
817d52f8 | 1315 | switch_commit_root(root->fs_info->chunk_root); |
5d4f98a2 YZ |
1316 | |
1317 | update_super_roots(root); | |
e02119d5 CM |
1318 | |
1319 | if (!root->fs_info->log_root_recovering) { | |
6c41761f DS |
1320 | btrfs_set_super_log_root(root->fs_info->super_copy, 0); |
1321 | btrfs_set_super_log_root_level(root->fs_info->super_copy, 0); | |
e02119d5 CM |
1322 | } |
1323 | ||
6c41761f DS |
1324 | memcpy(root->fs_info->super_for_commit, root->fs_info->super_copy, |
1325 | sizeof(*root->fs_info->super_copy)); | |
ccd467d6 | 1326 | |
f9295749 | 1327 | trans->transaction->blocked = 0; |
a4abeea4 JB |
1328 | spin_lock(&root->fs_info->trans_lock); |
1329 | root->fs_info->running_transaction = NULL; | |
1330 | root->fs_info->trans_no_join = 0; | |
1331 | spin_unlock(&root->fs_info->trans_lock); | |
7585717f | 1332 | mutex_unlock(&root->fs_info->reloc_mutex); |
b7ec40d7 | 1333 | |
f9295749 | 1334 | wake_up(&root->fs_info->transaction_wait); |
e6dcd2dc | 1335 | |
79154b1b CM |
1336 | ret = btrfs_write_and_wait_transaction(trans, root); |
1337 | BUG_ON(ret); | |
a512bbf8 | 1338 | write_ctree_super(trans, root, 0); |
4313b399 | 1339 | |
e02119d5 CM |
1340 | /* |
1341 | * the super is written, we can safely allow the tree-loggers | |
1342 | * to go about their business | |
1343 | */ | |
1344 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
1345 | ||
11833d66 | 1346 | btrfs_finish_extent_commit(trans, root); |
4313b399 | 1347 | |
2c90e5d6 | 1348 | cur_trans->commit_done = 1; |
b7ec40d7 | 1349 | |
15ee9bc7 | 1350 | root->fs_info->last_trans_committed = cur_trans->transid; |
817d52f8 | 1351 | |
2c90e5d6 | 1352 | wake_up(&cur_trans->commit_wait); |
3de4586c | 1353 | |
a4abeea4 | 1354 | spin_lock(&root->fs_info->trans_lock); |
13c5a93e | 1355 | list_del_init(&cur_trans->list); |
a4abeea4 JB |
1356 | spin_unlock(&root->fs_info->trans_lock); |
1357 | ||
78fae27e | 1358 | put_transaction(cur_trans); |
79154b1b | 1359 | put_transaction(cur_trans); |
58176a96 | 1360 | |
1abe9b8a | 1361 | trace_btrfs_transaction_commit(root); |
1362 | ||
a2de733c AJ |
1363 | btrfs_scrub_continue(root); |
1364 | ||
9ed74f2d JB |
1365 | if (current->journal_info == trans) |
1366 | current->journal_info = NULL; | |
1367 | ||
2c90e5d6 | 1368 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
24bbcf04 YZ |
1369 | |
1370 | if (current != root->fs_info->transaction_kthread) | |
1371 | btrfs_run_delayed_iputs(root); | |
1372 | ||
79154b1b CM |
1373 | return ret; |
1374 | } | |
1375 | ||
d352ac68 CM |
1376 | /* |
1377 | * interface function to delete all the snapshots we have scheduled for deletion | |
1378 | */ | |
e9d0b13b CM |
1379 | int btrfs_clean_old_snapshots(struct btrfs_root *root) |
1380 | { | |
5d4f98a2 YZ |
1381 | LIST_HEAD(list); |
1382 | struct btrfs_fs_info *fs_info = root->fs_info; | |
1383 | ||
a4abeea4 | 1384 | spin_lock(&fs_info->trans_lock); |
5d4f98a2 | 1385 | list_splice_init(&fs_info->dead_roots, &list); |
a4abeea4 | 1386 | spin_unlock(&fs_info->trans_lock); |
e9d0b13b | 1387 | |
5d4f98a2 YZ |
1388 | while (!list_empty(&list)) { |
1389 | root = list_entry(list.next, struct btrfs_root, root_list); | |
76dda93c YZ |
1390 | list_del(&root->root_list); |
1391 | ||
16cdcec7 MX |
1392 | btrfs_kill_all_delayed_nodes(root); |
1393 | ||
76dda93c YZ |
1394 | if (btrfs_header_backref_rev(root->node) < |
1395 | BTRFS_MIXED_BACKREF_REV) | |
66d7e7f0 | 1396 | btrfs_drop_snapshot(root, NULL, 0, 0); |
76dda93c | 1397 | else |
66d7e7f0 | 1398 | btrfs_drop_snapshot(root, NULL, 1, 0); |
e9d0b13b CM |
1399 | } |
1400 | return 0; | |
1401 | } |