]>
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" |
8dabb742 | 33 | #include "dev-replace.h" |
79154b1b | 34 | |
0f7d52f4 CM |
35 | #define BTRFS_ROOT_TRANS_TAG 0 |
36 | ||
49b25e05 | 37 | void put_transaction(struct btrfs_transaction *transaction) |
79154b1b | 38 | { |
13c5a93e JB |
39 | WARN_ON(atomic_read(&transaction->use_count) == 0); |
40 | if (atomic_dec_and_test(&transaction->use_count)) { | |
a4abeea4 | 41 | BUG_ON(!list_empty(&transaction->list)); |
00f04b88 | 42 | WARN_ON(transaction->delayed_refs.root.rb_node); |
2c90e5d6 | 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 | */ | |
354aa0fb | 56 | static noinline int join_transaction(struct btrfs_root *root, int type) |
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) { |
354aa0fb MX |
70 | /* |
71 | * If we are JOIN_NOLOCK we're already committing a current | |
72 | * transaction, we just need a handle to deal with something | |
73 | * when committing the transaction, such as inode cache and | |
74 | * space cache. It is a special case. | |
75 | */ | |
76 | if (type != TRANS_JOIN_NOLOCK) { | |
19ae4e81 | 77 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 JB |
78 | return -EBUSY; |
79 | } | |
80 | } | |
81 | ||
19ae4e81 | 82 | cur_trans = fs_info->running_transaction; |
a4abeea4 | 83 | if (cur_trans) { |
871383be | 84 | if (cur_trans->aborted) { |
19ae4e81 | 85 | spin_unlock(&fs_info->trans_lock); |
49b25e05 | 86 | return cur_trans->aborted; |
871383be | 87 | } |
a4abeea4 | 88 | atomic_inc(&cur_trans->use_count); |
13c5a93e | 89 | atomic_inc(&cur_trans->num_writers); |
15ee9bc7 | 90 | cur_trans->num_joined++; |
19ae4e81 | 91 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 | 92 | return 0; |
79154b1b | 93 | } |
19ae4e81 | 94 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 | 95 | |
354aa0fb MX |
96 | /* |
97 | * If we are ATTACH, we just want to catch the current transaction, | |
98 | * and commit it. If there is no transaction, just return ENOENT. | |
99 | */ | |
100 | if (type == TRANS_ATTACH) | |
101 | return -ENOENT; | |
102 | ||
a4abeea4 JB |
103 | cur_trans = kmem_cache_alloc(btrfs_transaction_cachep, GFP_NOFS); |
104 | if (!cur_trans) | |
105 | return -ENOMEM; | |
d43317dc | 106 | |
19ae4e81 JS |
107 | spin_lock(&fs_info->trans_lock); |
108 | if (fs_info->running_transaction) { | |
d43317dc CM |
109 | /* |
110 | * someone started a transaction after we unlocked. Make sure | |
111 | * to redo the trans_no_join checks above | |
112 | */ | |
a4abeea4 | 113 | kmem_cache_free(btrfs_transaction_cachep, cur_trans); |
19ae4e81 | 114 | cur_trans = fs_info->running_transaction; |
d43317dc | 115 | goto loop; |
e4b50e14 DC |
116 | } else if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) { |
117 | spin_unlock(&fs_info->trans_lock); | |
7b8b92af JB |
118 | kmem_cache_free(btrfs_transaction_cachep, cur_trans); |
119 | return -EROFS; | |
79154b1b | 120 | } |
d43317dc | 121 | |
a4abeea4 JB |
122 | atomic_set(&cur_trans->num_writers, 1); |
123 | cur_trans->num_joined = 0; | |
124 | init_waitqueue_head(&cur_trans->writer_wait); | |
125 | init_waitqueue_head(&cur_trans->commit_wait); | |
126 | cur_trans->in_commit = 0; | |
127 | cur_trans->blocked = 0; | |
128 | /* | |
129 | * One for this trans handle, one so it will live on until we | |
130 | * commit the transaction. | |
131 | */ | |
132 | atomic_set(&cur_trans->use_count, 2); | |
133 | cur_trans->commit_done = 0; | |
134 | cur_trans->start_time = get_seconds(); | |
135 | ||
136 | cur_trans->delayed_refs.root = RB_ROOT; | |
137 | cur_trans->delayed_refs.num_entries = 0; | |
138 | cur_trans->delayed_refs.num_heads_ready = 0; | |
139 | cur_trans->delayed_refs.num_heads = 0; | |
140 | cur_trans->delayed_refs.flushing = 0; | |
141 | cur_trans->delayed_refs.run_delayed_start = 0; | |
20b297d6 JS |
142 | |
143 | /* | |
144 | * although the tree mod log is per file system and not per transaction, | |
145 | * the log must never go across transaction boundaries. | |
146 | */ | |
147 | smp_mb(); | |
31b1a2bd JL |
148 | if (!list_empty(&fs_info->tree_mod_seq_list)) |
149 | WARN(1, KERN_ERR "btrfs: tree_mod_seq_list not empty when " | |
20b297d6 | 150 | "creating a fresh transaction\n"); |
31b1a2bd JL |
151 | if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log)) |
152 | WARN(1, KERN_ERR "btrfs: tree_mod_log rb tree not empty when " | |
20b297d6 | 153 | "creating a fresh transaction\n"); |
20b297d6 JS |
154 | atomic_set(&fs_info->tree_mod_seq, 0); |
155 | ||
a4abeea4 JB |
156 | spin_lock_init(&cur_trans->commit_lock); |
157 | spin_lock_init(&cur_trans->delayed_refs.lock); | |
158 | ||
159 | INIT_LIST_HEAD(&cur_trans->pending_snapshots); | |
19ae4e81 | 160 | list_add_tail(&cur_trans->list, &fs_info->trans_list); |
a4abeea4 | 161 | extent_io_tree_init(&cur_trans->dirty_pages, |
19ae4e81 JS |
162 | fs_info->btree_inode->i_mapping); |
163 | fs_info->generation++; | |
164 | cur_trans->transid = fs_info->generation; | |
165 | fs_info->running_transaction = cur_trans; | |
49b25e05 | 166 | cur_trans->aborted = 0; |
19ae4e81 | 167 | spin_unlock(&fs_info->trans_lock); |
15ee9bc7 | 168 | |
79154b1b CM |
169 | return 0; |
170 | } | |
171 | ||
d352ac68 | 172 | /* |
d397712b CM |
173 | * this does all the record keeping required to make sure that a reference |
174 | * counted root is properly recorded in a given transaction. This is required | |
175 | * to make sure the old root from before we joined the transaction is deleted | |
176 | * when the transaction commits | |
d352ac68 | 177 | */ |
7585717f | 178 | static int record_root_in_trans(struct btrfs_trans_handle *trans, |
a4abeea4 | 179 | struct btrfs_root *root) |
6702ed49 | 180 | { |
5d4f98a2 | 181 | if (root->ref_cows && root->last_trans < trans->transid) { |
6702ed49 | 182 | WARN_ON(root == root->fs_info->extent_root); |
5d4f98a2 YZ |
183 | WARN_ON(root->commit_root != root->node); |
184 | ||
7585717f CM |
185 | /* |
186 | * see below for in_trans_setup usage rules | |
187 | * we have the reloc mutex held now, so there | |
188 | * is only one writer in this function | |
189 | */ | |
190 | root->in_trans_setup = 1; | |
191 | ||
192 | /* make sure readers find in_trans_setup before | |
193 | * they find our root->last_trans update | |
194 | */ | |
195 | smp_wmb(); | |
196 | ||
a4abeea4 JB |
197 | spin_lock(&root->fs_info->fs_roots_radix_lock); |
198 | if (root->last_trans == trans->transid) { | |
199 | spin_unlock(&root->fs_info->fs_roots_radix_lock); | |
200 | return 0; | |
201 | } | |
5d4f98a2 YZ |
202 | radix_tree_tag_set(&root->fs_info->fs_roots_radix, |
203 | (unsigned long)root->root_key.objectid, | |
204 | BTRFS_ROOT_TRANS_TAG); | |
a4abeea4 | 205 | spin_unlock(&root->fs_info->fs_roots_radix_lock); |
7585717f CM |
206 | root->last_trans = trans->transid; |
207 | ||
208 | /* this is pretty tricky. We don't want to | |
209 | * take the relocation lock in btrfs_record_root_in_trans | |
210 | * unless we're really doing the first setup for this root in | |
211 | * this transaction. | |
212 | * | |
213 | * Normally we'd use root->last_trans as a flag to decide | |
214 | * if we want to take the expensive mutex. | |
215 | * | |
216 | * But, we have to set root->last_trans before we | |
217 | * init the relocation root, otherwise, we trip over warnings | |
218 | * in ctree.c. The solution used here is to flag ourselves | |
219 | * with root->in_trans_setup. When this is 1, we're still | |
220 | * fixing up the reloc trees and everyone must wait. | |
221 | * | |
222 | * When this is zero, they can trust root->last_trans and fly | |
223 | * through btrfs_record_root_in_trans without having to take the | |
224 | * lock. smp_wmb() makes sure that all the writes above are | |
225 | * done before we pop in the zero below | |
226 | */ | |
5d4f98a2 | 227 | btrfs_init_reloc_root(trans, root); |
7585717f CM |
228 | smp_wmb(); |
229 | root->in_trans_setup = 0; | |
5d4f98a2 YZ |
230 | } |
231 | return 0; | |
232 | } | |
bcc63abb | 233 | |
7585717f CM |
234 | |
235 | int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, | |
236 | struct btrfs_root *root) | |
237 | { | |
238 | if (!root->ref_cows) | |
239 | return 0; | |
240 | ||
241 | /* | |
242 | * see record_root_in_trans for comments about in_trans_setup usage | |
243 | * and barriers | |
244 | */ | |
245 | smp_rmb(); | |
246 | if (root->last_trans == trans->transid && | |
247 | !root->in_trans_setup) | |
248 | return 0; | |
249 | ||
250 | mutex_lock(&root->fs_info->reloc_mutex); | |
251 | record_root_in_trans(trans, root); | |
252 | mutex_unlock(&root->fs_info->reloc_mutex); | |
253 | ||
254 | return 0; | |
255 | } | |
256 | ||
d352ac68 CM |
257 | /* wait for commit against the current transaction to become unblocked |
258 | * when this is done, it is safe to start a new transaction, but the current | |
259 | * transaction might not be fully on disk. | |
260 | */ | |
37d1aeee | 261 | static void wait_current_trans(struct btrfs_root *root) |
79154b1b | 262 | { |
f9295749 | 263 | struct btrfs_transaction *cur_trans; |
79154b1b | 264 | |
a4abeea4 | 265 | spin_lock(&root->fs_info->trans_lock); |
f9295749 | 266 | cur_trans = root->fs_info->running_transaction; |
37d1aeee | 267 | if (cur_trans && cur_trans->blocked) { |
13c5a93e | 268 | atomic_inc(&cur_trans->use_count); |
a4abeea4 | 269 | spin_unlock(&root->fs_info->trans_lock); |
72d63ed6 LZ |
270 | |
271 | wait_event(root->fs_info->transaction_wait, | |
272 | !cur_trans->blocked); | |
f9295749 | 273 | put_transaction(cur_trans); |
a4abeea4 JB |
274 | } else { |
275 | spin_unlock(&root->fs_info->trans_lock); | |
f9295749 | 276 | } |
37d1aeee CM |
277 | } |
278 | ||
a22285a6 YZ |
279 | static int may_wait_transaction(struct btrfs_root *root, int type) |
280 | { | |
a4abeea4 JB |
281 | if (root->fs_info->log_root_recovering) |
282 | return 0; | |
283 | ||
284 | if (type == TRANS_USERSPACE) | |
285 | return 1; | |
286 | ||
287 | if (type == TRANS_START && | |
288 | !atomic_read(&root->fs_info->open_ioctl_trans)) | |
a22285a6 | 289 | return 1; |
a4abeea4 | 290 | |
a22285a6 YZ |
291 | return 0; |
292 | } | |
293 | ||
08e007d2 MX |
294 | static struct btrfs_trans_handle * |
295 | start_transaction(struct btrfs_root *root, u64 num_items, int type, | |
296 | enum btrfs_reserve_flush_enum flush) | |
37d1aeee | 297 | { |
a22285a6 YZ |
298 | struct btrfs_trans_handle *h; |
299 | struct btrfs_transaction *cur_trans; | |
b5009945 | 300 | u64 num_bytes = 0; |
37d1aeee | 301 | int ret; |
c5567237 | 302 | u64 qgroup_reserved = 0; |
acce952b | 303 | |
304 | if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) | |
305 | return ERR_PTR(-EROFS); | |
2a1eb461 JB |
306 | |
307 | if (current->journal_info) { | |
308 | WARN_ON(type != TRANS_JOIN && type != TRANS_JOIN_NOLOCK); | |
309 | h = current->journal_info; | |
310 | h->use_count++; | |
b7d5b0a8 | 311 | WARN_ON(h->use_count > 2); |
2a1eb461 JB |
312 | h->orig_rsv = h->block_rsv; |
313 | h->block_rsv = NULL; | |
314 | goto got_it; | |
315 | } | |
b5009945 JB |
316 | |
317 | /* | |
318 | * Do the reservation before we join the transaction so we can do all | |
319 | * the appropriate flushing if need be. | |
320 | */ | |
321 | if (num_items > 0 && root != root->fs_info->chunk_root) { | |
c5567237 AJ |
322 | if (root->fs_info->quota_enabled && |
323 | is_fstree(root->root_key.objectid)) { | |
324 | qgroup_reserved = num_items * root->leafsize; | |
325 | ret = btrfs_qgroup_reserve(root, qgroup_reserved); | |
326 | if (ret) | |
327 | return ERR_PTR(ret); | |
328 | } | |
329 | ||
b5009945 | 330 | num_bytes = btrfs_calc_trans_metadata_size(root, num_items); |
08e007d2 MX |
331 | ret = btrfs_block_rsv_add(root, |
332 | &root->fs_info->trans_block_rsv, | |
333 | num_bytes, flush); | |
b5009945 | 334 | if (ret) |
843fcf35 | 335 | goto reserve_fail; |
b5009945 | 336 | } |
a22285a6 YZ |
337 | again: |
338 | h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS); | |
843fcf35 MX |
339 | if (!h) { |
340 | ret = -ENOMEM; | |
341 | goto alloc_fail; | |
342 | } | |
37d1aeee | 343 | |
98114659 JB |
344 | /* |
345 | * If we are JOIN_NOLOCK we're already committing a transaction and | |
346 | * waiting on this guy, so we don't need to do the sb_start_intwrite | |
347 | * because we're already holding a ref. We need this because we could | |
348 | * have raced in and did an fsync() on a file which can kick a commit | |
349 | * and then we deadlock with somebody doing a freeze. | |
354aa0fb MX |
350 | * |
351 | * If we are ATTACH, it means we just want to catch the current | |
352 | * transaction and commit it, so we needn't do sb_start_intwrite(). | |
98114659 | 353 | */ |
354aa0fb | 354 | if (type < TRANS_JOIN_NOLOCK) |
60376ce4 | 355 | sb_start_intwrite(root->fs_info->sb); |
b2b5ef5c | 356 | |
a22285a6 | 357 | if (may_wait_transaction(root, type)) |
37d1aeee | 358 | wait_current_trans(root); |
a22285a6 | 359 | |
a4abeea4 | 360 | do { |
354aa0fb | 361 | ret = join_transaction(root, type); |
a4abeea4 JB |
362 | if (ret == -EBUSY) |
363 | wait_current_trans(root); | |
364 | } while (ret == -EBUSY); | |
365 | ||
db5b493a | 366 | if (ret < 0) { |
354aa0fb MX |
367 | /* We must get the transaction if we are JOIN_NOLOCK. */ |
368 | BUG_ON(type == TRANS_JOIN_NOLOCK); | |
843fcf35 | 369 | goto join_fail; |
db5b493a | 370 | } |
0f7d52f4 | 371 | |
a22285a6 | 372 | cur_trans = root->fs_info->running_transaction; |
a22285a6 YZ |
373 | |
374 | h->transid = cur_trans->transid; | |
375 | h->transaction = cur_trans; | |
79154b1b | 376 | h->blocks_used = 0; |
a22285a6 | 377 | h->bytes_reserved = 0; |
d13603ef | 378 | h->root = root; |
56bec294 | 379 | h->delayed_ref_updates = 0; |
2a1eb461 | 380 | h->use_count = 1; |
0e721106 | 381 | h->adding_csums = 0; |
f0486c68 | 382 | h->block_rsv = NULL; |
2a1eb461 | 383 | h->orig_rsv = NULL; |
49b25e05 | 384 | h->aborted = 0; |
c5567237 | 385 | h->qgroup_reserved = qgroup_reserved; |
bed92eae | 386 | h->delayed_ref_elem.seq = 0; |
a698d075 | 387 | h->type = type; |
c6b305a8 | 388 | h->allocating_chunk = false; |
bed92eae | 389 | INIT_LIST_HEAD(&h->qgroup_ref_list); |
ea658bad | 390 | INIT_LIST_HEAD(&h->new_bgs); |
b7ec40d7 | 391 | |
a22285a6 YZ |
392 | smp_mb(); |
393 | if (cur_trans->blocked && may_wait_transaction(root, type)) { | |
394 | btrfs_commit_transaction(h, root); | |
395 | goto again; | |
396 | } | |
397 | ||
b5009945 | 398 | if (num_bytes) { |
8c2a3ca2 | 399 | trace_btrfs_space_reservation(root->fs_info, "transaction", |
2bcc0328 | 400 | h->transid, num_bytes, 1); |
b5009945 JB |
401 | h->block_rsv = &root->fs_info->trans_block_rsv; |
402 | h->bytes_reserved = num_bytes; | |
a22285a6 | 403 | } |
9ed74f2d | 404 | |
2a1eb461 | 405 | got_it: |
a4abeea4 | 406 | btrfs_record_root_in_trans(h, root); |
a22285a6 YZ |
407 | |
408 | if (!current->journal_info && type != TRANS_USERSPACE) | |
409 | current->journal_info = h; | |
79154b1b | 410 | return h; |
843fcf35 MX |
411 | |
412 | join_fail: | |
413 | if (type < TRANS_JOIN_NOLOCK) | |
414 | sb_end_intwrite(root->fs_info->sb); | |
415 | kmem_cache_free(btrfs_trans_handle_cachep, h); | |
416 | alloc_fail: | |
417 | if (num_bytes) | |
418 | btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv, | |
419 | num_bytes); | |
420 | reserve_fail: | |
421 | if (qgroup_reserved) | |
422 | btrfs_qgroup_free(root, qgroup_reserved); | |
423 | return ERR_PTR(ret); | |
79154b1b CM |
424 | } |
425 | ||
f9295749 | 426 | struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, |
a22285a6 | 427 | int num_items) |
f9295749 | 428 | { |
08e007d2 MX |
429 | return start_transaction(root, num_items, TRANS_START, |
430 | BTRFS_RESERVE_FLUSH_ALL); | |
f9295749 | 431 | } |
8407aa46 | 432 | |
08e007d2 | 433 | struct btrfs_trans_handle *btrfs_start_transaction_lflush( |
8407aa46 MX |
434 | struct btrfs_root *root, int num_items) |
435 | { | |
08e007d2 MX |
436 | return start_transaction(root, num_items, TRANS_START, |
437 | BTRFS_RESERVE_FLUSH_LIMIT); | |
8407aa46 MX |
438 | } |
439 | ||
7a7eaa40 | 440 | struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root) |
f9295749 | 441 | { |
8407aa46 | 442 | return start_transaction(root, 0, TRANS_JOIN, 0); |
f9295749 CM |
443 | } |
444 | ||
7a7eaa40 | 445 | struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root) |
0af3d00b | 446 | { |
8407aa46 | 447 | return start_transaction(root, 0, TRANS_JOIN_NOLOCK, 0); |
0af3d00b JB |
448 | } |
449 | ||
7a7eaa40 | 450 | struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root) |
9ca9ee09 | 451 | { |
8407aa46 | 452 | return start_transaction(root, 0, TRANS_USERSPACE, 0); |
9ca9ee09 SW |
453 | } |
454 | ||
354aa0fb | 455 | struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root) |
60376ce4 | 456 | { |
354aa0fb | 457 | return start_transaction(root, 0, TRANS_ATTACH, 0); |
60376ce4 JB |
458 | } |
459 | ||
d352ac68 | 460 | /* wait for a transaction commit to be fully complete */ |
b9c8300c | 461 | static noinline void wait_for_commit(struct btrfs_root *root, |
89ce8a63 CM |
462 | struct btrfs_transaction *commit) |
463 | { | |
72d63ed6 | 464 | wait_event(commit->commit_wait, commit->commit_done); |
89ce8a63 CM |
465 | } |
466 | ||
46204592 SW |
467 | int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid) |
468 | { | |
469 | struct btrfs_transaction *cur_trans = NULL, *t; | |
8cd2807f | 470 | int ret = 0; |
46204592 | 471 | |
46204592 SW |
472 | if (transid) { |
473 | if (transid <= root->fs_info->last_trans_committed) | |
a4abeea4 | 474 | goto out; |
46204592 | 475 | |
8cd2807f | 476 | ret = -EINVAL; |
46204592 | 477 | /* find specified transaction */ |
a4abeea4 | 478 | spin_lock(&root->fs_info->trans_lock); |
46204592 SW |
479 | list_for_each_entry(t, &root->fs_info->trans_list, list) { |
480 | if (t->transid == transid) { | |
481 | cur_trans = t; | |
a4abeea4 | 482 | atomic_inc(&cur_trans->use_count); |
8cd2807f | 483 | ret = 0; |
46204592 SW |
484 | break; |
485 | } | |
8cd2807f MX |
486 | if (t->transid > transid) { |
487 | ret = 0; | |
46204592 | 488 | break; |
8cd2807f | 489 | } |
46204592 | 490 | } |
a4abeea4 | 491 | spin_unlock(&root->fs_info->trans_lock); |
8cd2807f | 492 | /* The specified transaction doesn't exist */ |
46204592 | 493 | if (!cur_trans) |
8cd2807f | 494 | goto out; |
46204592 SW |
495 | } else { |
496 | /* find newest transaction that is committing | committed */ | |
a4abeea4 | 497 | spin_lock(&root->fs_info->trans_lock); |
46204592 SW |
498 | list_for_each_entry_reverse(t, &root->fs_info->trans_list, |
499 | list) { | |
500 | if (t->in_commit) { | |
501 | if (t->commit_done) | |
3473f3c0 | 502 | break; |
46204592 | 503 | cur_trans = t; |
a4abeea4 | 504 | atomic_inc(&cur_trans->use_count); |
46204592 SW |
505 | break; |
506 | } | |
507 | } | |
a4abeea4 | 508 | spin_unlock(&root->fs_info->trans_lock); |
46204592 | 509 | if (!cur_trans) |
a4abeea4 | 510 | goto out; /* nothing committing|committed */ |
46204592 SW |
511 | } |
512 | ||
46204592 | 513 | wait_for_commit(root, cur_trans); |
46204592 | 514 | put_transaction(cur_trans); |
a4abeea4 | 515 | out: |
46204592 SW |
516 | return ret; |
517 | } | |
518 | ||
37d1aeee CM |
519 | void btrfs_throttle(struct btrfs_root *root) |
520 | { | |
a4abeea4 | 521 | if (!atomic_read(&root->fs_info->open_ioctl_trans)) |
9ca9ee09 | 522 | wait_current_trans(root); |
37d1aeee CM |
523 | } |
524 | ||
8929ecfa YZ |
525 | static int should_end_transaction(struct btrfs_trans_handle *trans, |
526 | struct btrfs_root *root) | |
527 | { | |
528 | int ret; | |
36ba022a JB |
529 | |
530 | ret = btrfs_block_rsv_check(root, &root->fs_info->global_block_rsv, 5); | |
8929ecfa YZ |
531 | return ret ? 1 : 0; |
532 | } | |
533 | ||
534 | int btrfs_should_end_transaction(struct btrfs_trans_handle *trans, | |
535 | struct btrfs_root *root) | |
536 | { | |
537 | struct btrfs_transaction *cur_trans = trans->transaction; | |
538 | int updates; | |
49b25e05 | 539 | int err; |
8929ecfa | 540 | |
a4abeea4 | 541 | smp_mb(); |
8929ecfa YZ |
542 | if (cur_trans->blocked || cur_trans->delayed_refs.flushing) |
543 | return 1; | |
544 | ||
545 | updates = trans->delayed_ref_updates; | |
546 | trans->delayed_ref_updates = 0; | |
49b25e05 JM |
547 | if (updates) { |
548 | err = btrfs_run_delayed_refs(trans, root, updates); | |
549 | if (err) /* Error code will also eval true */ | |
550 | return err; | |
551 | } | |
8929ecfa YZ |
552 | |
553 | return should_end_transaction(trans, root); | |
554 | } | |
555 | ||
89ce8a63 | 556 | static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, |
a698d075 | 557 | struct btrfs_root *root, int throttle) |
79154b1b | 558 | { |
8929ecfa | 559 | struct btrfs_transaction *cur_trans = trans->transaction; |
ab78c84d | 560 | struct btrfs_fs_info *info = root->fs_info; |
c3e69d58 | 561 | int count = 0; |
a698d075 | 562 | int lock = (trans->type != TRANS_JOIN_NOLOCK); |
4edc2ca3 | 563 | int err = 0; |
c3e69d58 | 564 | |
2a1eb461 JB |
565 | if (--trans->use_count) { |
566 | trans->block_rsv = trans->orig_rsv; | |
567 | return 0; | |
568 | } | |
569 | ||
edf39272 JS |
570 | /* |
571 | * do the qgroup accounting as early as possible | |
572 | */ | |
573 | err = btrfs_delayed_refs_qgroup_accounting(trans, info); | |
574 | ||
b24e03db | 575 | btrfs_trans_release_metadata(trans, root); |
4c13d758 | 576 | trans->block_rsv = NULL; |
d13603ef AJ |
577 | /* |
578 | * the same root has to be passed to start_transaction and | |
579 | * end_transaction. Subvolume quota depends on this. | |
580 | */ | |
581 | WARN_ON(trans->root != root); | |
c5567237 AJ |
582 | |
583 | if (trans->qgroup_reserved) { | |
584 | btrfs_qgroup_free(root, trans->qgroup_reserved); | |
585 | trans->qgroup_reserved = 0; | |
586 | } | |
587 | ||
ea658bad JB |
588 | if (!list_empty(&trans->new_bgs)) |
589 | btrfs_create_pending_block_groups(trans, root); | |
590 | ||
203bf287 | 591 | while (count < 2) { |
c3e69d58 CM |
592 | unsigned long cur = trans->delayed_ref_updates; |
593 | trans->delayed_ref_updates = 0; | |
594 | if (cur && | |
595 | trans->transaction->delayed_refs.num_heads_ready > 64) { | |
596 | trans->delayed_ref_updates = 0; | |
597 | btrfs_run_delayed_refs(trans, root, cur); | |
598 | } else { | |
599 | break; | |
600 | } | |
601 | count++; | |
56bec294 | 602 | } |
0e721106 JB |
603 | btrfs_trans_release_metadata(trans, root); |
604 | trans->block_rsv = NULL; | |
56bec294 | 605 | |
ea658bad JB |
606 | if (!list_empty(&trans->new_bgs)) |
607 | btrfs_create_pending_block_groups(trans, root); | |
608 | ||
a4abeea4 JB |
609 | if (lock && !atomic_read(&root->fs_info->open_ioctl_trans) && |
610 | should_end_transaction(trans, root)) { | |
8929ecfa | 611 | trans->transaction->blocked = 1; |
a4abeea4 JB |
612 | smp_wmb(); |
613 | } | |
8929ecfa | 614 | |
0af3d00b | 615 | if (lock && cur_trans->blocked && !cur_trans->in_commit) { |
81317fde JB |
616 | if (throttle) { |
617 | /* | |
618 | * We may race with somebody else here so end up having | |
619 | * to call end_transaction on ourselves again, so inc | |
620 | * our use_count. | |
621 | */ | |
622 | trans->use_count++; | |
8929ecfa | 623 | return btrfs_commit_transaction(trans, root); |
81317fde | 624 | } else { |
8929ecfa | 625 | wake_up_process(info->transaction_kthread); |
81317fde | 626 | } |
8929ecfa YZ |
627 | } |
628 | ||
354aa0fb | 629 | if (trans->type < TRANS_JOIN_NOLOCK) |
98114659 | 630 | sb_end_intwrite(root->fs_info->sb); |
6df7881a | 631 | |
8929ecfa | 632 | WARN_ON(cur_trans != info->running_transaction); |
13c5a93e JB |
633 | WARN_ON(atomic_read(&cur_trans->num_writers) < 1); |
634 | atomic_dec(&cur_trans->num_writers); | |
89ce8a63 | 635 | |
99d16cbc | 636 | smp_mb(); |
79154b1b CM |
637 | if (waitqueue_active(&cur_trans->writer_wait)) |
638 | wake_up(&cur_trans->writer_wait); | |
79154b1b | 639 | put_transaction(cur_trans); |
9ed74f2d JB |
640 | |
641 | if (current->journal_info == trans) | |
642 | current->journal_info = NULL; | |
ab78c84d | 643 | |
24bbcf04 YZ |
644 | if (throttle) |
645 | btrfs_run_delayed_iputs(root); | |
646 | ||
49b25e05 JM |
647 | if (trans->aborted || |
648 | root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) { | |
4edc2ca3 | 649 | err = -EIO; |
49b25e05 | 650 | } |
edf39272 | 651 | assert_qgroups_uptodate(trans); |
49b25e05 | 652 | |
4edc2ca3 DJ |
653 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
654 | return err; | |
79154b1b CM |
655 | } |
656 | ||
89ce8a63 CM |
657 | int btrfs_end_transaction(struct btrfs_trans_handle *trans, |
658 | struct btrfs_root *root) | |
659 | { | |
16cdcec7 MX |
660 | int ret; |
661 | ||
a698d075 | 662 | ret = __btrfs_end_transaction(trans, root, 0); |
16cdcec7 MX |
663 | if (ret) |
664 | return ret; | |
665 | return 0; | |
89ce8a63 CM |
666 | } |
667 | ||
668 | int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans, | |
669 | struct btrfs_root *root) | |
670 | { | |
16cdcec7 MX |
671 | int ret; |
672 | ||
a698d075 | 673 | ret = __btrfs_end_transaction(trans, root, 1); |
16cdcec7 MX |
674 | if (ret) |
675 | return ret; | |
676 | return 0; | |
677 | } | |
678 | ||
679 | int btrfs_end_transaction_dmeta(struct btrfs_trans_handle *trans, | |
680 | struct btrfs_root *root) | |
681 | { | |
a698d075 | 682 | return __btrfs_end_transaction(trans, root, 1); |
89ce8a63 CM |
683 | } |
684 | ||
d352ac68 CM |
685 | /* |
686 | * when btree blocks are allocated, they have some corresponding bits set for | |
687 | * them in one of two extent_io trees. This is used to make sure all of | |
690587d1 | 688 | * those extents are sent to disk but does not wait on them |
d352ac68 | 689 | */ |
690587d1 | 690 | int btrfs_write_marked_extents(struct btrfs_root *root, |
8cef4e16 | 691 | struct extent_io_tree *dirty_pages, int mark) |
79154b1b | 692 | { |
777e6bd7 | 693 | int err = 0; |
7c4452b9 | 694 | int werr = 0; |
1728366e | 695 | struct address_space *mapping = root->fs_info->btree_inode->i_mapping; |
e6138876 | 696 | struct extent_state *cached_state = NULL; |
777e6bd7 | 697 | u64 start = 0; |
5f39d397 | 698 | u64 end; |
7c4452b9 | 699 | |
1728366e | 700 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
e6138876 JB |
701 | mark, &cached_state)) { |
702 | convert_extent_bit(dirty_pages, start, end, EXTENT_NEED_WAIT, | |
703 | mark, &cached_state, GFP_NOFS); | |
704 | cached_state = NULL; | |
1728366e JB |
705 | err = filemap_fdatawrite_range(mapping, start, end); |
706 | if (err) | |
707 | werr = err; | |
708 | cond_resched(); | |
709 | start = end + 1; | |
7c4452b9 | 710 | } |
690587d1 CM |
711 | if (err) |
712 | werr = err; | |
713 | return werr; | |
714 | } | |
715 | ||
716 | /* | |
717 | * when btree blocks are allocated, they have some corresponding bits set for | |
718 | * them in one of two extent_io trees. This is used to make sure all of | |
719 | * those extents are on disk for transaction or log commit. We wait | |
720 | * on all the pages and clear them from the dirty pages state tree | |
721 | */ | |
722 | int btrfs_wait_marked_extents(struct btrfs_root *root, | |
8cef4e16 | 723 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 | 724 | { |
690587d1 CM |
725 | int err = 0; |
726 | int werr = 0; | |
1728366e | 727 | struct address_space *mapping = root->fs_info->btree_inode->i_mapping; |
e6138876 | 728 | struct extent_state *cached_state = NULL; |
690587d1 CM |
729 | u64 start = 0; |
730 | u64 end; | |
777e6bd7 | 731 | |
1728366e | 732 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
e6138876 JB |
733 | EXTENT_NEED_WAIT, &cached_state)) { |
734 | clear_extent_bit(dirty_pages, start, end, EXTENT_NEED_WAIT, | |
735 | 0, 0, &cached_state, GFP_NOFS); | |
1728366e JB |
736 | err = filemap_fdatawait_range(mapping, start, end); |
737 | if (err) | |
738 | werr = err; | |
739 | cond_resched(); | |
740 | start = end + 1; | |
777e6bd7 | 741 | } |
7c4452b9 CM |
742 | if (err) |
743 | werr = err; | |
744 | return werr; | |
79154b1b CM |
745 | } |
746 | ||
690587d1 CM |
747 | /* |
748 | * when btree blocks are allocated, they have some corresponding bits set for | |
749 | * them in one of two extent_io trees. This is used to make sure all of | |
750 | * those extents are on disk for transaction or log commit | |
751 | */ | |
752 | int btrfs_write_and_wait_marked_extents(struct btrfs_root *root, | |
8cef4e16 | 753 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 CM |
754 | { |
755 | int ret; | |
756 | int ret2; | |
757 | ||
8cef4e16 YZ |
758 | ret = btrfs_write_marked_extents(root, dirty_pages, mark); |
759 | ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark); | |
bf0da8c1 CM |
760 | |
761 | if (ret) | |
762 | return ret; | |
763 | if (ret2) | |
764 | return ret2; | |
765 | return 0; | |
690587d1 CM |
766 | } |
767 | ||
d0c803c4 CM |
768 | int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans, |
769 | struct btrfs_root *root) | |
770 | { | |
771 | if (!trans || !trans->transaction) { | |
772 | struct inode *btree_inode; | |
773 | btree_inode = root->fs_info->btree_inode; | |
774 | return filemap_write_and_wait(btree_inode->i_mapping); | |
775 | } | |
776 | return btrfs_write_and_wait_marked_extents(root, | |
8cef4e16 YZ |
777 | &trans->transaction->dirty_pages, |
778 | EXTENT_DIRTY); | |
d0c803c4 CM |
779 | } |
780 | ||
d352ac68 CM |
781 | /* |
782 | * this is used to update the root pointer in the tree of tree roots. | |
783 | * | |
784 | * But, in the case of the extent allocation tree, updating the root | |
785 | * pointer may allocate blocks which may change the root of the extent | |
786 | * allocation tree. | |
787 | * | |
788 | * So, this loops and repeats and makes sure the cowonly root didn't | |
789 | * change while the root pointer was being updated in the metadata. | |
790 | */ | |
0b86a832 CM |
791 | static int update_cowonly_root(struct btrfs_trans_handle *trans, |
792 | struct btrfs_root *root) | |
79154b1b CM |
793 | { |
794 | int ret; | |
0b86a832 | 795 | u64 old_root_bytenr; |
86b9f2ec | 796 | u64 old_root_used; |
0b86a832 | 797 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
79154b1b | 798 | |
86b9f2ec | 799 | old_root_used = btrfs_root_used(&root->root_item); |
0b86a832 | 800 | btrfs_write_dirty_block_groups(trans, root); |
56bec294 | 801 | |
d397712b | 802 | while (1) { |
0b86a832 | 803 | old_root_bytenr = btrfs_root_bytenr(&root->root_item); |
86b9f2ec YZ |
804 | if (old_root_bytenr == root->node->start && |
805 | old_root_used == btrfs_root_used(&root->root_item)) | |
79154b1b | 806 | break; |
87ef2bb4 | 807 | |
5d4f98a2 | 808 | btrfs_set_root_node(&root->root_item, root->node); |
79154b1b | 809 | ret = btrfs_update_root(trans, tree_root, |
0b86a832 CM |
810 | &root->root_key, |
811 | &root->root_item); | |
49b25e05 JM |
812 | if (ret) |
813 | return ret; | |
56bec294 | 814 | |
86b9f2ec | 815 | old_root_used = btrfs_root_used(&root->root_item); |
4a8c9a62 | 816 | ret = btrfs_write_dirty_block_groups(trans, root); |
49b25e05 JM |
817 | if (ret) |
818 | return ret; | |
0b86a832 | 819 | } |
276e680d YZ |
820 | |
821 | if (root != root->fs_info->extent_root) | |
822 | switch_commit_root(root); | |
823 | ||
0b86a832 CM |
824 | return 0; |
825 | } | |
826 | ||
d352ac68 CM |
827 | /* |
828 | * update all the cowonly tree roots on disk | |
49b25e05 JM |
829 | * |
830 | * The error handling in this function may not be obvious. Any of the | |
831 | * failures will cause the file system to go offline. We still need | |
832 | * to clean up the delayed refs. | |
d352ac68 | 833 | */ |
5d4f98a2 YZ |
834 | static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans, |
835 | struct btrfs_root *root) | |
0b86a832 CM |
836 | { |
837 | struct btrfs_fs_info *fs_info = root->fs_info; | |
838 | struct list_head *next; | |
84234f3a | 839 | struct extent_buffer *eb; |
56bec294 | 840 | int ret; |
84234f3a | 841 | |
56bec294 | 842 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
49b25e05 JM |
843 | if (ret) |
844 | return ret; | |
87ef2bb4 | 845 | |
84234f3a | 846 | eb = btrfs_lock_root_node(fs_info->tree_root); |
49b25e05 JM |
847 | ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, |
848 | 0, &eb); | |
84234f3a YZ |
849 | btrfs_tree_unlock(eb); |
850 | free_extent_buffer(eb); | |
0b86a832 | 851 | |
49b25e05 JM |
852 | if (ret) |
853 | return ret; | |
854 | ||
56bec294 | 855 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
49b25e05 JM |
856 | if (ret) |
857 | return ret; | |
87ef2bb4 | 858 | |
733f4fbb | 859 | ret = btrfs_run_dev_stats(trans, root->fs_info); |
8dabb742 SB |
860 | WARN_ON(ret); |
861 | ret = btrfs_run_dev_replace(trans, root->fs_info); | |
862 | WARN_ON(ret); | |
733f4fbb | 863 | |
546adb0d JS |
864 | ret = btrfs_run_qgroups(trans, root->fs_info); |
865 | BUG_ON(ret); | |
866 | ||
867 | /* run_qgroups might have added some more refs */ | |
868 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); | |
869 | BUG_ON(ret); | |
870 | ||
d397712b | 871 | while (!list_empty(&fs_info->dirty_cowonly_roots)) { |
0b86a832 CM |
872 | next = fs_info->dirty_cowonly_roots.next; |
873 | list_del_init(next); | |
874 | root = list_entry(next, struct btrfs_root, dirty_list); | |
87ef2bb4 | 875 | |
49b25e05 JM |
876 | ret = update_cowonly_root(trans, root); |
877 | if (ret) | |
878 | return ret; | |
79154b1b | 879 | } |
276e680d YZ |
880 | |
881 | down_write(&fs_info->extent_commit_sem); | |
882 | switch_commit_root(fs_info->extent_root); | |
883 | up_write(&fs_info->extent_commit_sem); | |
884 | ||
8dabb742 SB |
885 | btrfs_after_dev_replace_commit(fs_info); |
886 | ||
79154b1b CM |
887 | return 0; |
888 | } | |
889 | ||
d352ac68 CM |
890 | /* |
891 | * dead roots are old snapshots that need to be deleted. This allocates | |
892 | * a dirty root struct and adds it into the list of dead roots that need to | |
893 | * be deleted | |
894 | */ | |
5d4f98a2 | 895 | int btrfs_add_dead_root(struct btrfs_root *root) |
5eda7b5e | 896 | { |
a4abeea4 | 897 | spin_lock(&root->fs_info->trans_lock); |
5d4f98a2 | 898 | list_add(&root->root_list, &root->fs_info->dead_roots); |
a4abeea4 | 899 | spin_unlock(&root->fs_info->trans_lock); |
5eda7b5e CM |
900 | return 0; |
901 | } | |
902 | ||
d352ac68 | 903 | /* |
5d4f98a2 | 904 | * update all the cowonly tree roots on disk |
d352ac68 | 905 | */ |
5d4f98a2 YZ |
906 | static noinline int commit_fs_roots(struct btrfs_trans_handle *trans, |
907 | struct btrfs_root *root) | |
0f7d52f4 | 908 | { |
0f7d52f4 | 909 | struct btrfs_root *gang[8]; |
5d4f98a2 | 910 | struct btrfs_fs_info *fs_info = root->fs_info; |
0f7d52f4 CM |
911 | int i; |
912 | int ret; | |
54aa1f4d CM |
913 | int err = 0; |
914 | ||
a4abeea4 | 915 | spin_lock(&fs_info->fs_roots_radix_lock); |
d397712b | 916 | while (1) { |
5d4f98a2 YZ |
917 | ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, |
918 | (void **)gang, 0, | |
0f7d52f4 CM |
919 | ARRAY_SIZE(gang), |
920 | BTRFS_ROOT_TRANS_TAG); | |
921 | if (ret == 0) | |
922 | break; | |
923 | for (i = 0; i < ret; i++) { | |
924 | root = gang[i]; | |
5d4f98a2 YZ |
925 | radix_tree_tag_clear(&fs_info->fs_roots_radix, |
926 | (unsigned long)root->root_key.objectid, | |
927 | BTRFS_ROOT_TRANS_TAG); | |
a4abeea4 | 928 | spin_unlock(&fs_info->fs_roots_radix_lock); |
31153d81 | 929 | |
e02119d5 | 930 | btrfs_free_log(trans, root); |
5d4f98a2 | 931 | btrfs_update_reloc_root(trans, root); |
d68fc57b | 932 | btrfs_orphan_commit_root(trans, root); |
bcc63abb | 933 | |
82d5902d LZ |
934 | btrfs_save_ino_cache(root, trans); |
935 | ||
f1ebcc74 LB |
936 | /* see comments in should_cow_block() */ |
937 | root->force_cow = 0; | |
938 | smp_wmb(); | |
939 | ||
978d910d | 940 | if (root->commit_root != root->node) { |
581bb050 | 941 | mutex_lock(&root->fs_commit_mutex); |
817d52f8 | 942 | switch_commit_root(root); |
581bb050 LZ |
943 | btrfs_unpin_free_ino(root); |
944 | mutex_unlock(&root->fs_commit_mutex); | |
945 | ||
978d910d YZ |
946 | btrfs_set_root_node(&root->root_item, |
947 | root->node); | |
948 | } | |
5d4f98a2 | 949 | |
5d4f98a2 | 950 | err = btrfs_update_root(trans, fs_info->tree_root, |
0f7d52f4 CM |
951 | &root->root_key, |
952 | &root->root_item); | |
a4abeea4 | 953 | spin_lock(&fs_info->fs_roots_radix_lock); |
54aa1f4d CM |
954 | if (err) |
955 | break; | |
0f7d52f4 CM |
956 | } |
957 | } | |
a4abeea4 | 958 | spin_unlock(&fs_info->fs_roots_radix_lock); |
54aa1f4d | 959 | return err; |
0f7d52f4 CM |
960 | } |
961 | ||
d352ac68 CM |
962 | /* |
963 | * defrag a given btree. If cacheonly == 1, this won't read from the disk, | |
964 | * otherwise every leaf in the btree is read and defragged. | |
965 | */ | |
e9d0b13b CM |
966 | int btrfs_defrag_root(struct btrfs_root *root, int cacheonly) |
967 | { | |
968 | struct btrfs_fs_info *info = root->fs_info; | |
e9d0b13b | 969 | struct btrfs_trans_handle *trans; |
8929ecfa | 970 | int ret; |
e9d0b13b | 971 | |
8929ecfa | 972 | if (xchg(&root->defrag_running, 1)) |
e9d0b13b | 973 | return 0; |
8929ecfa | 974 | |
6b80053d | 975 | while (1) { |
8929ecfa YZ |
976 | trans = btrfs_start_transaction(root, 0); |
977 | if (IS_ERR(trans)) | |
978 | return PTR_ERR(trans); | |
979 | ||
e9d0b13b | 980 | ret = btrfs_defrag_leaves(trans, root, cacheonly); |
8929ecfa | 981 | |
e9d0b13b | 982 | btrfs_end_transaction(trans, root); |
b53d3f5d | 983 | btrfs_btree_balance_dirty(info->tree_root); |
e9d0b13b CM |
984 | cond_resched(); |
985 | ||
7841cb28 | 986 | if (btrfs_fs_closing(root->fs_info) || ret != -EAGAIN) |
e9d0b13b CM |
987 | break; |
988 | } | |
989 | root->defrag_running = 0; | |
8929ecfa | 990 | return ret; |
e9d0b13b CM |
991 | } |
992 | ||
d352ac68 CM |
993 | /* |
994 | * new snapshots need to be created at a very specific time in the | |
995 | * transaction commit. This does the actual creation | |
996 | */ | |
80b6794d | 997 | static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, |
3063d29f CM |
998 | struct btrfs_fs_info *fs_info, |
999 | struct btrfs_pending_snapshot *pending) | |
1000 | { | |
1001 | struct btrfs_key key; | |
80b6794d | 1002 | struct btrfs_root_item *new_root_item; |
3063d29f CM |
1003 | struct btrfs_root *tree_root = fs_info->tree_root; |
1004 | struct btrfs_root *root = pending->root; | |
6bdb72de | 1005 | struct btrfs_root *parent_root; |
98c9942a | 1006 | struct btrfs_block_rsv *rsv; |
6bdb72de | 1007 | struct inode *parent_inode; |
42874b3d MX |
1008 | struct btrfs_path *path; |
1009 | struct btrfs_dir_item *dir_item; | |
6a912213 | 1010 | struct dentry *parent; |
a22285a6 | 1011 | struct dentry *dentry; |
3063d29f | 1012 | struct extent_buffer *tmp; |
925baedd | 1013 | struct extent_buffer *old; |
8ea05e3a | 1014 | struct timespec cur_time = CURRENT_TIME; |
3063d29f | 1015 | int ret; |
d68fc57b | 1016 | u64 to_reserve = 0; |
6bdb72de | 1017 | u64 index = 0; |
a22285a6 | 1018 | u64 objectid; |
b83cc969 | 1019 | u64 root_flags; |
8ea05e3a | 1020 | uuid_le new_uuid; |
3063d29f | 1021 | |
42874b3d MX |
1022 | path = btrfs_alloc_path(); |
1023 | if (!path) { | |
1024 | ret = pending->error = -ENOMEM; | |
1025 | goto path_alloc_fail; | |
1026 | } | |
1027 | ||
80b6794d CM |
1028 | new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS); |
1029 | if (!new_root_item) { | |
49b25e05 | 1030 | ret = pending->error = -ENOMEM; |
6fa9700e | 1031 | goto root_item_alloc_fail; |
80b6794d | 1032 | } |
a22285a6 | 1033 | |
581bb050 | 1034 | ret = btrfs_find_free_objectid(tree_root, &objectid); |
a22285a6 YZ |
1035 | if (ret) { |
1036 | pending->error = ret; | |
6fa9700e | 1037 | goto no_free_objectid; |
a22285a6 | 1038 | } |
3063d29f | 1039 | |
3fd0a558 | 1040 | btrfs_reloc_pre_snapshot(trans, pending, &to_reserve); |
d68fc57b YZ |
1041 | |
1042 | if (to_reserve > 0) { | |
08e007d2 MX |
1043 | ret = btrfs_block_rsv_add(root, &pending->block_rsv, |
1044 | to_reserve, | |
1045 | BTRFS_RESERVE_NO_FLUSH); | |
d68fc57b YZ |
1046 | if (ret) { |
1047 | pending->error = ret; | |
6fa9700e | 1048 | goto no_free_objectid; |
d68fc57b YZ |
1049 | } |
1050 | } | |
1051 | ||
6f72c7e2 AJ |
1052 | ret = btrfs_qgroup_inherit(trans, fs_info, root->root_key.objectid, |
1053 | objectid, pending->inherit); | |
6f72c7e2 AJ |
1054 | if (ret) { |
1055 | pending->error = ret; | |
6fa9700e | 1056 | goto no_free_objectid; |
6f72c7e2 AJ |
1057 | } |
1058 | ||
3063d29f | 1059 | key.objectid = objectid; |
a22285a6 YZ |
1060 | key.offset = (u64)-1; |
1061 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3063d29f | 1062 | |
6fa9700e | 1063 | rsv = trans->block_rsv; |
a22285a6 | 1064 | trans->block_rsv = &pending->block_rsv; |
3de4586c | 1065 | |
a22285a6 | 1066 | dentry = pending->dentry; |
6a912213 JB |
1067 | parent = dget_parent(dentry); |
1068 | parent_inode = parent->d_inode; | |
a22285a6 | 1069 | parent_root = BTRFS_I(parent_inode)->root; |
7585717f | 1070 | record_root_in_trans(trans, parent_root); |
a22285a6 | 1071 | |
3063d29f CM |
1072 | /* |
1073 | * insert the directory item | |
1074 | */ | |
3de4586c | 1075 | ret = btrfs_set_inode_index(parent_inode, &index); |
49b25e05 | 1076 | BUG_ON(ret); /* -ENOMEM */ |
42874b3d MX |
1077 | |
1078 | /* check if there is a file/dir which has the same name. */ | |
1079 | dir_item = btrfs_lookup_dir_item(NULL, parent_root, path, | |
1080 | btrfs_ino(parent_inode), | |
1081 | dentry->d_name.name, | |
1082 | dentry->d_name.len, 0); | |
1083 | if (dir_item != NULL && !IS_ERR(dir_item)) { | |
fe66a05a | 1084 | pending->error = -EEXIST; |
fe66a05a | 1085 | goto fail; |
42874b3d MX |
1086 | } else if (IS_ERR(dir_item)) { |
1087 | ret = PTR_ERR(dir_item); | |
8732d44f MX |
1088 | btrfs_abort_transaction(trans, root, ret); |
1089 | goto fail; | |
79787eaa | 1090 | } |
42874b3d | 1091 | btrfs_release_path(path); |
52c26179 | 1092 | |
e999376f CM |
1093 | /* |
1094 | * pull in the delayed directory update | |
1095 | * and the delayed inode item | |
1096 | * otherwise we corrupt the FS during | |
1097 | * snapshot | |
1098 | */ | |
1099 | ret = btrfs_run_delayed_items(trans, root); | |
8732d44f MX |
1100 | if (ret) { /* Transaction aborted */ |
1101 | btrfs_abort_transaction(trans, root, ret); | |
1102 | goto fail; | |
1103 | } | |
e999376f | 1104 | |
7585717f | 1105 | record_root_in_trans(trans, root); |
6bdb72de SW |
1106 | btrfs_set_root_last_snapshot(&root->root_item, trans->transid); |
1107 | memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); | |
08fe4db1 | 1108 | btrfs_check_and_init_root_item(new_root_item); |
6bdb72de | 1109 | |
b83cc969 LZ |
1110 | root_flags = btrfs_root_flags(new_root_item); |
1111 | if (pending->readonly) | |
1112 | root_flags |= BTRFS_ROOT_SUBVOL_RDONLY; | |
1113 | else | |
1114 | root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY; | |
1115 | btrfs_set_root_flags(new_root_item, root_flags); | |
1116 | ||
8ea05e3a AB |
1117 | btrfs_set_root_generation_v2(new_root_item, |
1118 | trans->transid); | |
1119 | uuid_le_gen(&new_uuid); | |
1120 | memcpy(new_root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE); | |
1121 | memcpy(new_root_item->parent_uuid, root->root_item.uuid, | |
1122 | BTRFS_UUID_SIZE); | |
1123 | new_root_item->otime.sec = cpu_to_le64(cur_time.tv_sec); | |
dadd1105 | 1124 | new_root_item->otime.nsec = cpu_to_le32(cur_time.tv_nsec); |
8ea05e3a AB |
1125 | btrfs_set_root_otransid(new_root_item, trans->transid); |
1126 | memset(&new_root_item->stime, 0, sizeof(new_root_item->stime)); | |
1127 | memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime)); | |
1128 | btrfs_set_root_stransid(new_root_item, 0); | |
1129 | btrfs_set_root_rtransid(new_root_item, 0); | |
1130 | ||
6bdb72de | 1131 | old = btrfs_lock_root_node(root); |
49b25e05 | 1132 | ret = btrfs_cow_block(trans, root, old, NULL, 0, &old); |
79787eaa JM |
1133 | if (ret) { |
1134 | btrfs_tree_unlock(old); | |
1135 | free_extent_buffer(old); | |
8732d44f MX |
1136 | btrfs_abort_transaction(trans, root, ret); |
1137 | goto fail; | |
79787eaa | 1138 | } |
49b25e05 | 1139 | |
6bdb72de SW |
1140 | btrfs_set_lock_blocking(old); |
1141 | ||
49b25e05 | 1142 | ret = btrfs_copy_root(trans, root, old, &tmp, objectid); |
79787eaa | 1143 | /* clean up in any case */ |
6bdb72de SW |
1144 | btrfs_tree_unlock(old); |
1145 | free_extent_buffer(old); | |
8732d44f MX |
1146 | if (ret) { |
1147 | btrfs_abort_transaction(trans, root, ret); | |
1148 | goto fail; | |
1149 | } | |
6bdb72de | 1150 | |
f1ebcc74 LB |
1151 | /* see comments in should_cow_block() */ |
1152 | root->force_cow = 1; | |
1153 | smp_wmb(); | |
1154 | ||
6bdb72de | 1155 | btrfs_set_root_node(new_root_item, tmp); |
a22285a6 YZ |
1156 | /* record when the snapshot was created in key.offset */ |
1157 | key.offset = trans->transid; | |
1158 | ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); | |
6bdb72de SW |
1159 | btrfs_tree_unlock(tmp); |
1160 | free_extent_buffer(tmp); | |
8732d44f MX |
1161 | if (ret) { |
1162 | btrfs_abort_transaction(trans, root, ret); | |
1163 | goto fail; | |
1164 | } | |
6bdb72de | 1165 | |
a22285a6 YZ |
1166 | /* |
1167 | * insert root back/forward references | |
1168 | */ | |
1169 | ret = btrfs_add_root_ref(trans, tree_root, objectid, | |
0660b5af | 1170 | parent_root->root_key.objectid, |
33345d01 | 1171 | btrfs_ino(parent_inode), index, |
a22285a6 | 1172 | dentry->d_name.name, dentry->d_name.len); |
8732d44f MX |
1173 | if (ret) { |
1174 | btrfs_abort_transaction(trans, root, ret); | |
1175 | goto fail; | |
1176 | } | |
0660b5af | 1177 | |
a22285a6 YZ |
1178 | key.offset = (u64)-1; |
1179 | pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key); | |
79787eaa JM |
1180 | if (IS_ERR(pending->snap)) { |
1181 | ret = PTR_ERR(pending->snap); | |
8732d44f MX |
1182 | btrfs_abort_transaction(trans, root, ret); |
1183 | goto fail; | |
79787eaa | 1184 | } |
d68fc57b | 1185 | |
49b25e05 | 1186 | ret = btrfs_reloc_post_snapshot(trans, pending); |
8732d44f MX |
1187 | if (ret) { |
1188 | btrfs_abort_transaction(trans, root, ret); | |
1189 | goto fail; | |
1190 | } | |
361048f5 MX |
1191 | |
1192 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); | |
8732d44f MX |
1193 | if (ret) { |
1194 | btrfs_abort_transaction(trans, root, ret); | |
1195 | goto fail; | |
1196 | } | |
42874b3d MX |
1197 | |
1198 | ret = btrfs_insert_dir_item(trans, parent_root, | |
1199 | dentry->d_name.name, dentry->d_name.len, | |
1200 | parent_inode, &key, | |
1201 | BTRFS_FT_DIR, index); | |
1202 | /* We have check then name at the beginning, so it is impossible. */ | |
9c52057c | 1203 | BUG_ON(ret == -EEXIST || ret == -EOVERFLOW); |
8732d44f MX |
1204 | if (ret) { |
1205 | btrfs_abort_transaction(trans, root, ret); | |
1206 | goto fail; | |
1207 | } | |
42874b3d MX |
1208 | |
1209 | btrfs_i_size_write(parent_inode, parent_inode->i_size + | |
1210 | dentry->d_name.len * 2); | |
1211 | parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME; | |
be6aef60 | 1212 | ret = btrfs_update_inode_fallback(trans, parent_root, parent_inode); |
42874b3d | 1213 | if (ret) |
8732d44f | 1214 | btrfs_abort_transaction(trans, root, ret); |
3063d29f | 1215 | fail: |
6fa9700e | 1216 | dput(parent); |
98c9942a | 1217 | trans->block_rsv = rsv; |
6fa9700e MX |
1218 | no_free_objectid: |
1219 | kfree(new_root_item); | |
1220 | root_item_alloc_fail: | |
42874b3d MX |
1221 | btrfs_free_path(path); |
1222 | path_alloc_fail: | |
a22285a6 | 1223 | btrfs_block_rsv_release(root, &pending->block_rsv, (u64)-1); |
49b25e05 | 1224 | return ret; |
3063d29f CM |
1225 | } |
1226 | ||
d352ac68 CM |
1227 | /* |
1228 | * create all the snapshots we've scheduled for creation | |
1229 | */ | |
80b6794d CM |
1230 | static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans, |
1231 | struct btrfs_fs_info *fs_info) | |
3de4586c CM |
1232 | { |
1233 | struct btrfs_pending_snapshot *pending; | |
1234 | struct list_head *head = &trans->transaction->pending_snapshots; | |
3de4586c | 1235 | |
fe66a05a CM |
1236 | list_for_each_entry(pending, head, list) |
1237 | create_pending_snapshot(trans, fs_info, pending); | |
3de4586c CM |
1238 | return 0; |
1239 | } | |
1240 | ||
5d4f98a2 YZ |
1241 | static void update_super_roots(struct btrfs_root *root) |
1242 | { | |
1243 | struct btrfs_root_item *root_item; | |
1244 | struct btrfs_super_block *super; | |
1245 | ||
6c41761f | 1246 | super = root->fs_info->super_copy; |
5d4f98a2 YZ |
1247 | |
1248 | root_item = &root->fs_info->chunk_root->root_item; | |
1249 | super->chunk_root = root_item->bytenr; | |
1250 | super->chunk_root_generation = root_item->generation; | |
1251 | super->chunk_root_level = root_item->level; | |
1252 | ||
1253 | root_item = &root->fs_info->tree_root->root_item; | |
1254 | super->root = root_item->bytenr; | |
1255 | super->generation = root_item->generation; | |
1256 | super->root_level = root_item->level; | |
73bc1876 | 1257 | if (btrfs_test_opt(root, SPACE_CACHE)) |
0af3d00b | 1258 | super->cache_generation = root_item->generation; |
5d4f98a2 YZ |
1259 | } |
1260 | ||
f36f3042 CM |
1261 | int btrfs_transaction_in_commit(struct btrfs_fs_info *info) |
1262 | { | |
1263 | int ret = 0; | |
a4abeea4 | 1264 | spin_lock(&info->trans_lock); |
f36f3042 CM |
1265 | if (info->running_transaction) |
1266 | ret = info->running_transaction->in_commit; | |
a4abeea4 | 1267 | spin_unlock(&info->trans_lock); |
f36f3042 CM |
1268 | return ret; |
1269 | } | |
1270 | ||
8929ecfa YZ |
1271 | int btrfs_transaction_blocked(struct btrfs_fs_info *info) |
1272 | { | |
1273 | int ret = 0; | |
a4abeea4 | 1274 | spin_lock(&info->trans_lock); |
8929ecfa YZ |
1275 | if (info->running_transaction) |
1276 | ret = info->running_transaction->blocked; | |
a4abeea4 | 1277 | spin_unlock(&info->trans_lock); |
8929ecfa YZ |
1278 | return ret; |
1279 | } | |
1280 | ||
bb9c12c9 SW |
1281 | /* |
1282 | * wait for the current transaction commit to start and block subsequent | |
1283 | * transaction joins | |
1284 | */ | |
1285 | static void wait_current_trans_commit_start(struct btrfs_root *root, | |
1286 | struct btrfs_transaction *trans) | |
1287 | { | |
72d63ed6 | 1288 | wait_event(root->fs_info->transaction_blocked_wait, trans->in_commit); |
bb9c12c9 SW |
1289 | } |
1290 | ||
1291 | /* | |
1292 | * wait for the current transaction to start and then become unblocked. | |
1293 | * caller holds ref. | |
1294 | */ | |
1295 | static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root, | |
1296 | struct btrfs_transaction *trans) | |
1297 | { | |
72d63ed6 LZ |
1298 | wait_event(root->fs_info->transaction_wait, |
1299 | trans->commit_done || (trans->in_commit && !trans->blocked)); | |
bb9c12c9 SW |
1300 | } |
1301 | ||
1302 | /* | |
1303 | * commit transactions asynchronously. once btrfs_commit_transaction_async | |
1304 | * returns, any subsequent transaction will not be allowed to join. | |
1305 | */ | |
1306 | struct btrfs_async_commit { | |
1307 | struct btrfs_trans_handle *newtrans; | |
1308 | struct btrfs_root *root; | |
7892b5af | 1309 | struct work_struct work; |
bb9c12c9 SW |
1310 | }; |
1311 | ||
1312 | static void do_async_commit(struct work_struct *work) | |
1313 | { | |
1314 | struct btrfs_async_commit *ac = | |
7892b5af | 1315 | container_of(work, struct btrfs_async_commit, work); |
bb9c12c9 | 1316 | |
6fc4e354 SW |
1317 | /* |
1318 | * We've got freeze protection passed with the transaction. | |
1319 | * Tell lockdep about it. | |
1320 | */ | |
ff7c1d33 MX |
1321 | if (ac->newtrans->type < TRANS_JOIN_NOLOCK) |
1322 | rwsem_acquire_read( | |
1323 | &ac->root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1], | |
1324 | 0, 1, _THIS_IP_); | |
6fc4e354 | 1325 | |
e209db7a SW |
1326 | current->journal_info = ac->newtrans; |
1327 | ||
bb9c12c9 SW |
1328 | btrfs_commit_transaction(ac->newtrans, ac->root); |
1329 | kfree(ac); | |
1330 | } | |
1331 | ||
1332 | int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans, | |
1333 | struct btrfs_root *root, | |
1334 | int wait_for_unblock) | |
1335 | { | |
1336 | struct btrfs_async_commit *ac; | |
1337 | struct btrfs_transaction *cur_trans; | |
1338 | ||
1339 | ac = kmalloc(sizeof(*ac), GFP_NOFS); | |
db5b493a TI |
1340 | if (!ac) |
1341 | return -ENOMEM; | |
bb9c12c9 | 1342 | |
7892b5af | 1343 | INIT_WORK(&ac->work, do_async_commit); |
bb9c12c9 | 1344 | ac->root = root; |
7a7eaa40 | 1345 | ac->newtrans = btrfs_join_transaction(root); |
3612b495 TI |
1346 | if (IS_ERR(ac->newtrans)) { |
1347 | int err = PTR_ERR(ac->newtrans); | |
1348 | kfree(ac); | |
1349 | return err; | |
1350 | } | |
bb9c12c9 SW |
1351 | |
1352 | /* take transaction reference */ | |
bb9c12c9 | 1353 | cur_trans = trans->transaction; |
13c5a93e | 1354 | atomic_inc(&cur_trans->use_count); |
bb9c12c9 SW |
1355 | |
1356 | btrfs_end_transaction(trans, root); | |
6fc4e354 SW |
1357 | |
1358 | /* | |
1359 | * Tell lockdep we've released the freeze rwsem, since the | |
1360 | * async commit thread will be the one to unlock it. | |
1361 | */ | |
ff7c1d33 MX |
1362 | if (trans->type < TRANS_JOIN_NOLOCK) |
1363 | rwsem_release( | |
1364 | &root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1], | |
1365 | 1, _THIS_IP_); | |
6fc4e354 | 1366 | |
7892b5af | 1367 | schedule_work(&ac->work); |
bb9c12c9 SW |
1368 | |
1369 | /* wait for transaction to start and unblock */ | |
bb9c12c9 SW |
1370 | if (wait_for_unblock) |
1371 | wait_current_trans_commit_start_and_unblock(root, cur_trans); | |
1372 | else | |
1373 | wait_current_trans_commit_start(root, cur_trans); | |
bb9c12c9 | 1374 | |
38e88054 SW |
1375 | if (current->journal_info == trans) |
1376 | current->journal_info = NULL; | |
1377 | ||
1378 | put_transaction(cur_trans); | |
bb9c12c9 SW |
1379 | return 0; |
1380 | } | |
1381 | ||
49b25e05 JM |
1382 | |
1383 | static void cleanup_transaction(struct btrfs_trans_handle *trans, | |
7b8b92af | 1384 | struct btrfs_root *root, int err) |
49b25e05 JM |
1385 | { |
1386 | struct btrfs_transaction *cur_trans = trans->transaction; | |
1387 | ||
1388 | WARN_ON(trans->use_count > 1); | |
1389 | ||
7b8b92af JB |
1390 | btrfs_abort_transaction(trans, root, err); |
1391 | ||
49b25e05 JM |
1392 | spin_lock(&root->fs_info->trans_lock); |
1393 | list_del_init(&cur_trans->list); | |
d7096fc3 JB |
1394 | if (cur_trans == root->fs_info->running_transaction) { |
1395 | root->fs_info->running_transaction = NULL; | |
1396 | root->fs_info->trans_no_join = 0; | |
1397 | } | |
49b25e05 JM |
1398 | spin_unlock(&root->fs_info->trans_lock); |
1399 | ||
1400 | btrfs_cleanup_one_transaction(trans->transaction, root); | |
1401 | ||
1402 | put_transaction(cur_trans); | |
1403 | put_transaction(cur_trans); | |
1404 | ||
1405 | trace_btrfs_transaction_commit(root); | |
1406 | ||
1407 | btrfs_scrub_continue(root); | |
1408 | ||
1409 | if (current->journal_info == trans) | |
1410 | current->journal_info = NULL; | |
1411 | ||
1412 | kmem_cache_free(btrfs_trans_handle_cachep, trans); | |
1413 | } | |
1414 | ||
ca469637 MX |
1415 | static int btrfs_flush_all_pending_stuffs(struct btrfs_trans_handle *trans, |
1416 | struct btrfs_root *root) | |
1417 | { | |
1418 | int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT); | |
1419 | int snap_pending = 0; | |
1420 | int ret; | |
1421 | ||
1422 | if (!flush_on_commit) { | |
1423 | spin_lock(&root->fs_info->trans_lock); | |
1424 | if (!list_empty(&trans->transaction->pending_snapshots)) | |
1425 | snap_pending = 1; | |
1426 | spin_unlock(&root->fs_info->trans_lock); | |
1427 | } | |
1428 | ||
1429 | if (flush_on_commit || snap_pending) { | |
3edb2a68 MX |
1430 | ret = btrfs_start_delalloc_inodes(root, 1); |
1431 | if (ret) | |
1432 | return ret; | |
ca469637 MX |
1433 | btrfs_wait_ordered_extents(root, 1); |
1434 | } | |
1435 | ||
1436 | ret = btrfs_run_delayed_items(trans, root); | |
1437 | if (ret) | |
1438 | return ret; | |
1439 | ||
1440 | /* | |
1441 | * running the delayed items may have added new refs. account | |
1442 | * them now so that they hinder processing of more delayed refs | |
1443 | * as little as possible. | |
1444 | */ | |
1445 | btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info); | |
1446 | ||
1447 | /* | |
1448 | * rename don't use btrfs_join_transaction, so, once we | |
1449 | * set the transaction to blocked above, we aren't going | |
1450 | * to get any new ordered operations. We can safely run | |
1451 | * it here and no for sure that nothing new will be added | |
1452 | * to the list | |
1453 | */ | |
eebc6084 | 1454 | ret = btrfs_run_ordered_operations(root, 1); |
ca469637 | 1455 | |
eebc6084 | 1456 | return ret; |
ca469637 MX |
1457 | } |
1458 | ||
bb9c12c9 SW |
1459 | /* |
1460 | * btrfs_transaction state sequence: | |
1461 | * in_commit = 0, blocked = 0 (initial) | |
1462 | * in_commit = 1, blocked = 1 | |
1463 | * blocked = 0 | |
1464 | * commit_done = 1 | |
1465 | */ | |
79154b1b CM |
1466 | int btrfs_commit_transaction(struct btrfs_trans_handle *trans, |
1467 | struct btrfs_root *root) | |
1468 | { | |
15ee9bc7 | 1469 | unsigned long joined = 0; |
49b25e05 | 1470 | struct btrfs_transaction *cur_trans = trans->transaction; |
8fd17795 | 1471 | struct btrfs_transaction *prev_trans = NULL; |
79154b1b | 1472 | DEFINE_WAIT(wait); |
25287e0a | 1473 | int ret; |
89573b9c CM |
1474 | int should_grow = 0; |
1475 | unsigned long now = get_seconds(); | |
79154b1b | 1476 | |
25287e0a MX |
1477 | ret = btrfs_run_ordered_operations(root, 0); |
1478 | if (ret) { | |
1479 | btrfs_abort_transaction(trans, root, ret); | |
1480 | goto cleanup_transaction; | |
1481 | } | |
5a3f23d5 | 1482 | |
8d25a086 MX |
1483 | /* Stop the commit early if ->aborted is set */ |
1484 | if (unlikely(ACCESS_ONCE(cur_trans->aborted))) { | |
25287e0a | 1485 | ret = cur_trans->aborted; |
49b25e05 | 1486 | goto cleanup_transaction; |
25287e0a | 1487 | } |
49b25e05 | 1488 | |
56bec294 CM |
1489 | /* make a pass through all the delayed refs we have so far |
1490 | * any runnings procs may add more while we are here | |
1491 | */ | |
1492 | ret = btrfs_run_delayed_refs(trans, root, 0); | |
49b25e05 JM |
1493 | if (ret) |
1494 | goto cleanup_transaction; | |
56bec294 | 1495 | |
0e721106 JB |
1496 | btrfs_trans_release_metadata(trans, root); |
1497 | trans->block_rsv = NULL; | |
1498 | ||
b7ec40d7 | 1499 | cur_trans = trans->transaction; |
49b25e05 | 1500 | |
56bec294 CM |
1501 | /* |
1502 | * set the flushing flag so procs in this transaction have to | |
1503 | * start sending their work down. | |
1504 | */ | |
b7ec40d7 | 1505 | cur_trans->delayed_refs.flushing = 1; |
56bec294 | 1506 | |
ea658bad JB |
1507 | if (!list_empty(&trans->new_bgs)) |
1508 | btrfs_create_pending_block_groups(trans, root); | |
1509 | ||
c3e69d58 | 1510 | ret = btrfs_run_delayed_refs(trans, root, 0); |
49b25e05 JM |
1511 | if (ret) |
1512 | goto cleanup_transaction; | |
56bec294 | 1513 | |
a4abeea4 | 1514 | spin_lock(&cur_trans->commit_lock); |
b7ec40d7 | 1515 | if (cur_trans->in_commit) { |
a4abeea4 | 1516 | spin_unlock(&cur_trans->commit_lock); |
13c5a93e | 1517 | atomic_inc(&cur_trans->use_count); |
49b25e05 | 1518 | ret = btrfs_end_transaction(trans, root); |
ccd467d6 | 1519 | |
b9c8300c | 1520 | wait_for_commit(root, cur_trans); |
15ee9bc7 | 1521 | |
79154b1b | 1522 | put_transaction(cur_trans); |
15ee9bc7 | 1523 | |
49b25e05 | 1524 | return ret; |
79154b1b | 1525 | } |
4313b399 | 1526 | |
2c90e5d6 | 1527 | trans->transaction->in_commit = 1; |
f9295749 | 1528 | trans->transaction->blocked = 1; |
a4abeea4 | 1529 | spin_unlock(&cur_trans->commit_lock); |
bb9c12c9 SW |
1530 | wake_up(&root->fs_info->transaction_blocked_wait); |
1531 | ||
a4abeea4 | 1532 | spin_lock(&root->fs_info->trans_lock); |
ccd467d6 CM |
1533 | if (cur_trans->list.prev != &root->fs_info->trans_list) { |
1534 | prev_trans = list_entry(cur_trans->list.prev, | |
1535 | struct btrfs_transaction, list); | |
1536 | if (!prev_trans->commit_done) { | |
13c5a93e | 1537 | atomic_inc(&prev_trans->use_count); |
a4abeea4 | 1538 | spin_unlock(&root->fs_info->trans_lock); |
ccd467d6 CM |
1539 | |
1540 | wait_for_commit(root, prev_trans); | |
ccd467d6 | 1541 | |
15ee9bc7 | 1542 | put_transaction(prev_trans); |
a4abeea4 JB |
1543 | } else { |
1544 | spin_unlock(&root->fs_info->trans_lock); | |
ccd467d6 | 1545 | } |
a4abeea4 JB |
1546 | } else { |
1547 | spin_unlock(&root->fs_info->trans_lock); | |
ccd467d6 | 1548 | } |
15ee9bc7 | 1549 | |
e39e64ac CM |
1550 | if (!btrfs_test_opt(root, SSD) && |
1551 | (now < cur_trans->start_time || now - cur_trans->start_time < 1)) | |
89573b9c CM |
1552 | should_grow = 1; |
1553 | ||
15ee9bc7 JB |
1554 | do { |
1555 | joined = cur_trans->num_joined; | |
7ea394f1 | 1556 | |
2c90e5d6 | 1557 | WARN_ON(cur_trans != trans->transaction); |
15ee9bc7 | 1558 | |
ca469637 | 1559 | ret = btrfs_flush_all_pending_stuffs(trans, root); |
49b25e05 JM |
1560 | if (ret) |
1561 | goto cleanup_transaction; | |
16cdcec7 | 1562 | |
ed3b3d31 CM |
1563 | prepare_to_wait(&cur_trans->writer_wait, &wait, |
1564 | TASK_UNINTERRUPTIBLE); | |
1565 | ||
13c5a93e | 1566 | if (atomic_read(&cur_trans->num_writers) > 1) |
99d16cbc SW |
1567 | schedule_timeout(MAX_SCHEDULE_TIMEOUT); |
1568 | else if (should_grow) | |
1569 | schedule_timeout(1); | |
15ee9bc7 | 1570 | |
15ee9bc7 | 1571 | finish_wait(&cur_trans->writer_wait, &wait); |
13c5a93e | 1572 | } while (atomic_read(&cur_trans->num_writers) > 1 || |
89573b9c | 1573 | (should_grow && cur_trans->num_joined != joined)); |
15ee9bc7 | 1574 | |
ca469637 MX |
1575 | ret = btrfs_flush_all_pending_stuffs(trans, root); |
1576 | if (ret) | |
1577 | goto cleanup_transaction; | |
1578 | ||
ed0ca140 JB |
1579 | /* |
1580 | * Ok now we need to make sure to block out any other joins while we | |
1581 | * commit the transaction. We could have started a join before setting | |
1582 | * no_join so make sure to wait for num_writers to == 1 again. | |
1583 | */ | |
1584 | spin_lock(&root->fs_info->trans_lock); | |
1585 | root->fs_info->trans_no_join = 1; | |
1586 | spin_unlock(&root->fs_info->trans_lock); | |
1587 | wait_event(cur_trans->writer_wait, | |
1588 | atomic_read(&cur_trans->num_writers) == 1); | |
1589 | ||
2cba30f1 MX |
1590 | /* ->aborted might be set after the previous check, so check it */ |
1591 | if (unlikely(ACCESS_ONCE(cur_trans->aborted))) { | |
1592 | ret = cur_trans->aborted; | |
1593 | goto cleanup_transaction; | |
1594 | } | |
7585717f CM |
1595 | /* |
1596 | * the reloc mutex makes sure that we stop | |
1597 | * the balancing code from coming in and moving | |
1598 | * extents around in the middle of the commit | |
1599 | */ | |
1600 | mutex_lock(&root->fs_info->reloc_mutex); | |
1601 | ||
42874b3d MX |
1602 | /* |
1603 | * We needn't worry about the delayed items because we will | |
1604 | * deal with them in create_pending_snapshot(), which is the | |
1605 | * core function of the snapshot creation. | |
1606 | */ | |
1607 | ret = create_pending_snapshots(trans, root->fs_info); | |
49b25e05 JM |
1608 | if (ret) { |
1609 | mutex_unlock(&root->fs_info->reloc_mutex); | |
1610 | goto cleanup_transaction; | |
1611 | } | |
3063d29f | 1612 | |
42874b3d MX |
1613 | /* |
1614 | * We insert the dir indexes of the snapshots and update the inode | |
1615 | * of the snapshots' parents after the snapshot creation, so there | |
1616 | * are some delayed items which are not dealt with. Now deal with | |
1617 | * them. | |
1618 | * | |
1619 | * We needn't worry that this operation will corrupt the snapshots, | |
1620 | * because all the tree which are snapshoted will be forced to COW | |
1621 | * the nodes and leaves. | |
1622 | */ | |
1623 | ret = btrfs_run_delayed_items(trans, root); | |
49b25e05 JM |
1624 | if (ret) { |
1625 | mutex_unlock(&root->fs_info->reloc_mutex); | |
1626 | goto cleanup_transaction; | |
1627 | } | |
16cdcec7 | 1628 | |
56bec294 | 1629 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
49b25e05 JM |
1630 | if (ret) { |
1631 | mutex_unlock(&root->fs_info->reloc_mutex); | |
1632 | goto cleanup_transaction; | |
1633 | } | |
56bec294 | 1634 | |
e999376f CM |
1635 | /* |
1636 | * make sure none of the code above managed to slip in a | |
1637 | * delayed item | |
1638 | */ | |
1639 | btrfs_assert_delayed_root_empty(root); | |
1640 | ||
2c90e5d6 | 1641 | WARN_ON(cur_trans != trans->transaction); |
dc17ff8f | 1642 | |
a2de733c | 1643 | btrfs_scrub_pause(root); |
e02119d5 CM |
1644 | /* btrfs_commit_tree_roots is responsible for getting the |
1645 | * various roots consistent with each other. Every pointer | |
1646 | * in the tree of tree roots has to point to the most up to date | |
1647 | * root for every subvolume and other tree. So, we have to keep | |
1648 | * the tree logging code from jumping in and changing any | |
1649 | * of the trees. | |
1650 | * | |
1651 | * At this point in the commit, there can't be any tree-log | |
1652 | * writers, but a little lower down we drop the trans mutex | |
1653 | * and let new people in. By holding the tree_log_mutex | |
1654 | * from now until after the super is written, we avoid races | |
1655 | * with the tree-log code. | |
1656 | */ | |
1657 | mutex_lock(&root->fs_info->tree_log_mutex); | |
1658 | ||
5d4f98a2 | 1659 | ret = commit_fs_roots(trans, root); |
49b25e05 JM |
1660 | if (ret) { |
1661 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
871383be | 1662 | mutex_unlock(&root->fs_info->reloc_mutex); |
49b25e05 JM |
1663 | goto cleanup_transaction; |
1664 | } | |
54aa1f4d | 1665 | |
5d4f98a2 | 1666 | /* commit_fs_roots gets rid of all the tree log roots, it is now |
e02119d5 CM |
1667 | * safe to free the root of tree log roots |
1668 | */ | |
1669 | btrfs_free_log_root_tree(trans, root->fs_info); | |
1670 | ||
5d4f98a2 | 1671 | ret = commit_cowonly_roots(trans, root); |
49b25e05 JM |
1672 | if (ret) { |
1673 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
871383be | 1674 | mutex_unlock(&root->fs_info->reloc_mutex); |
49b25e05 JM |
1675 | goto cleanup_transaction; |
1676 | } | |
54aa1f4d | 1677 | |
2cba30f1 MX |
1678 | /* |
1679 | * The tasks which save the space cache and inode cache may also | |
1680 | * update ->aborted, check it. | |
1681 | */ | |
1682 | if (unlikely(ACCESS_ONCE(cur_trans->aborted))) { | |
1683 | ret = cur_trans->aborted; | |
1684 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
1685 | mutex_unlock(&root->fs_info->reloc_mutex); | |
1686 | goto cleanup_transaction; | |
1687 | } | |
1688 | ||
11833d66 YZ |
1689 | btrfs_prepare_extent_commit(trans, root); |
1690 | ||
78fae27e | 1691 | cur_trans = root->fs_info->running_transaction; |
5d4f98a2 YZ |
1692 | |
1693 | btrfs_set_root_node(&root->fs_info->tree_root->root_item, | |
1694 | root->fs_info->tree_root->node); | |
817d52f8 | 1695 | switch_commit_root(root->fs_info->tree_root); |
5d4f98a2 YZ |
1696 | |
1697 | btrfs_set_root_node(&root->fs_info->chunk_root->root_item, | |
1698 | root->fs_info->chunk_root->node); | |
817d52f8 | 1699 | switch_commit_root(root->fs_info->chunk_root); |
5d4f98a2 | 1700 | |
edf39272 | 1701 | assert_qgroups_uptodate(trans); |
5d4f98a2 | 1702 | update_super_roots(root); |
e02119d5 CM |
1703 | |
1704 | if (!root->fs_info->log_root_recovering) { | |
6c41761f DS |
1705 | btrfs_set_super_log_root(root->fs_info->super_copy, 0); |
1706 | btrfs_set_super_log_root_level(root->fs_info->super_copy, 0); | |
e02119d5 CM |
1707 | } |
1708 | ||
6c41761f DS |
1709 | memcpy(root->fs_info->super_for_commit, root->fs_info->super_copy, |
1710 | sizeof(*root->fs_info->super_copy)); | |
ccd467d6 | 1711 | |
f9295749 | 1712 | trans->transaction->blocked = 0; |
a4abeea4 JB |
1713 | spin_lock(&root->fs_info->trans_lock); |
1714 | root->fs_info->running_transaction = NULL; | |
1715 | root->fs_info->trans_no_join = 0; | |
1716 | spin_unlock(&root->fs_info->trans_lock); | |
7585717f | 1717 | mutex_unlock(&root->fs_info->reloc_mutex); |
b7ec40d7 | 1718 | |
f9295749 | 1719 | wake_up(&root->fs_info->transaction_wait); |
e6dcd2dc | 1720 | |
79154b1b | 1721 | ret = btrfs_write_and_wait_transaction(trans, root); |
49b25e05 JM |
1722 | if (ret) { |
1723 | btrfs_error(root->fs_info, ret, | |
1724 | "Error while writing out transaction."); | |
1725 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
1726 | goto cleanup_transaction; | |
1727 | } | |
1728 | ||
1729 | ret = write_ctree_super(trans, root, 0); | |
1730 | if (ret) { | |
1731 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
1732 | goto cleanup_transaction; | |
1733 | } | |
4313b399 | 1734 | |
e02119d5 CM |
1735 | /* |
1736 | * the super is written, we can safely allow the tree-loggers | |
1737 | * to go about their business | |
1738 | */ | |
1739 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
1740 | ||
11833d66 | 1741 | btrfs_finish_extent_commit(trans, root); |
4313b399 | 1742 | |
2c90e5d6 | 1743 | cur_trans->commit_done = 1; |
b7ec40d7 | 1744 | |
15ee9bc7 | 1745 | root->fs_info->last_trans_committed = cur_trans->transid; |
817d52f8 | 1746 | |
2c90e5d6 | 1747 | wake_up(&cur_trans->commit_wait); |
3de4586c | 1748 | |
a4abeea4 | 1749 | spin_lock(&root->fs_info->trans_lock); |
13c5a93e | 1750 | list_del_init(&cur_trans->list); |
a4abeea4 JB |
1751 | spin_unlock(&root->fs_info->trans_lock); |
1752 | ||
78fae27e | 1753 | put_transaction(cur_trans); |
79154b1b | 1754 | put_transaction(cur_trans); |
58176a96 | 1755 | |
354aa0fb MX |
1756 | if (trans->type < TRANS_JOIN_NOLOCK) |
1757 | sb_end_intwrite(root->fs_info->sb); | |
b2b5ef5c | 1758 | |
1abe9b8a | 1759 | trace_btrfs_transaction_commit(root); |
1760 | ||
a2de733c AJ |
1761 | btrfs_scrub_continue(root); |
1762 | ||
9ed74f2d JB |
1763 | if (current->journal_info == trans) |
1764 | current->journal_info = NULL; | |
1765 | ||
2c90e5d6 | 1766 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
24bbcf04 YZ |
1767 | |
1768 | if (current != root->fs_info->transaction_kthread) | |
1769 | btrfs_run_delayed_iputs(root); | |
1770 | ||
79154b1b | 1771 | return ret; |
49b25e05 JM |
1772 | |
1773 | cleanup_transaction: | |
0e721106 JB |
1774 | btrfs_trans_release_metadata(trans, root); |
1775 | trans->block_rsv = NULL; | |
49b25e05 JM |
1776 | btrfs_printk(root->fs_info, "Skipping commit of aborted transaction.\n"); |
1777 | // WARN_ON(1); | |
1778 | if (current->journal_info == trans) | |
1779 | current->journal_info = NULL; | |
7b8b92af | 1780 | cleanup_transaction(trans, root, ret); |
49b25e05 JM |
1781 | |
1782 | return ret; | |
79154b1b CM |
1783 | } |
1784 | ||
d352ac68 CM |
1785 | /* |
1786 | * interface function to delete all the snapshots we have scheduled for deletion | |
1787 | */ | |
e9d0b13b CM |
1788 | int btrfs_clean_old_snapshots(struct btrfs_root *root) |
1789 | { | |
5d4f98a2 YZ |
1790 | LIST_HEAD(list); |
1791 | struct btrfs_fs_info *fs_info = root->fs_info; | |
1792 | ||
a4abeea4 | 1793 | spin_lock(&fs_info->trans_lock); |
5d4f98a2 | 1794 | list_splice_init(&fs_info->dead_roots, &list); |
a4abeea4 | 1795 | spin_unlock(&fs_info->trans_lock); |
e9d0b13b | 1796 | |
5d4f98a2 | 1797 | while (!list_empty(&list)) { |
2c536799 JM |
1798 | int ret; |
1799 | ||
5d4f98a2 | 1800 | root = list_entry(list.next, struct btrfs_root, root_list); |
76dda93c YZ |
1801 | list_del(&root->root_list); |
1802 | ||
16cdcec7 MX |
1803 | btrfs_kill_all_delayed_nodes(root); |
1804 | ||
76dda93c YZ |
1805 | if (btrfs_header_backref_rev(root->node) < |
1806 | BTRFS_MIXED_BACKREF_REV) | |
2c536799 | 1807 | ret = btrfs_drop_snapshot(root, NULL, 0, 0); |
76dda93c | 1808 | else |
2c536799 JM |
1809 | ret =btrfs_drop_snapshot(root, NULL, 1, 0); |
1810 | BUG_ON(ret < 0); | |
e9d0b13b CM |
1811 | } |
1812 | return 0; | |
1813 | } |