]>
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" |
fcebe456 | 34 | #include "qgroup.h" |
79154b1b | 35 | |
0f7d52f4 CM |
36 | #define BTRFS_ROOT_TRANS_TAG 0 |
37 | ||
e8c9f186 | 38 | static const unsigned int btrfs_blocked_trans_types[TRANS_STATE_MAX] = { |
4a9d8bde MX |
39 | [TRANS_STATE_RUNNING] = 0U, |
40 | [TRANS_STATE_BLOCKED] = (__TRANS_USERSPACE | | |
41 | __TRANS_START), | |
42 | [TRANS_STATE_COMMIT_START] = (__TRANS_USERSPACE | | |
43 | __TRANS_START | | |
44 | __TRANS_ATTACH), | |
45 | [TRANS_STATE_COMMIT_DOING] = (__TRANS_USERSPACE | | |
46 | __TRANS_START | | |
47 | __TRANS_ATTACH | | |
48 | __TRANS_JOIN), | |
49 | [TRANS_STATE_UNBLOCKED] = (__TRANS_USERSPACE | | |
50 | __TRANS_START | | |
51 | __TRANS_ATTACH | | |
52 | __TRANS_JOIN | | |
53 | __TRANS_JOIN_NOLOCK), | |
54 | [TRANS_STATE_COMPLETED] = (__TRANS_USERSPACE | | |
55 | __TRANS_START | | |
56 | __TRANS_ATTACH | | |
57 | __TRANS_JOIN | | |
58 | __TRANS_JOIN_NOLOCK), | |
59 | }; | |
60 | ||
724e2315 | 61 | void btrfs_put_transaction(struct btrfs_transaction *transaction) |
79154b1b | 62 | { |
13c5a93e JB |
63 | WARN_ON(atomic_read(&transaction->use_count) == 0); |
64 | if (atomic_dec_and_test(&transaction->use_count)) { | |
a4abeea4 | 65 | BUG_ON(!list_empty(&transaction->list)); |
c46effa6 | 66 | WARN_ON(!RB_EMPTY_ROOT(&transaction->delayed_refs.href_root)); |
1262133b JB |
67 | if (transaction->delayed_refs.pending_csums) |
68 | printk(KERN_ERR "pending csums is %llu\n", | |
69 | transaction->delayed_refs.pending_csums); | |
6df9a95e JB |
70 | while (!list_empty(&transaction->pending_chunks)) { |
71 | struct extent_map *em; | |
72 | ||
73 | em = list_first_entry(&transaction->pending_chunks, | |
74 | struct extent_map, list); | |
75 | list_del_init(&em->list); | |
76 | free_extent_map(em); | |
77 | } | |
2c90e5d6 | 78 | kmem_cache_free(btrfs_transaction_cachep, transaction); |
78fae27e | 79 | } |
79154b1b CM |
80 | } |
81 | ||
663dfbb0 FM |
82 | static void clear_btree_io_tree(struct extent_io_tree *tree) |
83 | { | |
84 | spin_lock(&tree->lock); | |
85 | while (!RB_EMPTY_ROOT(&tree->state)) { | |
86 | struct rb_node *node; | |
87 | struct extent_state *state; | |
88 | ||
89 | node = rb_first(&tree->state); | |
90 | state = rb_entry(node, struct extent_state, rb_node); | |
91 | rb_erase(&state->rb_node, &tree->state); | |
92 | RB_CLEAR_NODE(&state->rb_node); | |
93 | /* | |
94 | * btree io trees aren't supposed to have tasks waiting for | |
95 | * changes in the flags of extent states ever. | |
96 | */ | |
97 | ASSERT(!waitqueue_active(&state->wq)); | |
98 | free_extent_state(state); | |
351810c1 DS |
99 | |
100 | cond_resched_lock(&tree->lock); | |
663dfbb0 FM |
101 | } |
102 | spin_unlock(&tree->lock); | |
103 | } | |
104 | ||
9e351cc8 JB |
105 | static noinline void switch_commit_roots(struct btrfs_transaction *trans, |
106 | struct btrfs_fs_info *fs_info) | |
817d52f8 | 107 | { |
9e351cc8 JB |
108 | struct btrfs_root *root, *tmp; |
109 | ||
110 | down_write(&fs_info->commit_root_sem); | |
111 | list_for_each_entry_safe(root, tmp, &trans->switch_commits, | |
112 | dirty_list) { | |
113 | list_del_init(&root->dirty_list); | |
114 | free_extent_buffer(root->commit_root); | |
115 | root->commit_root = btrfs_root_node(root); | |
116 | if (is_fstree(root->objectid)) | |
117 | btrfs_unpin_free_ino(root); | |
663dfbb0 | 118 | clear_btree_io_tree(&root->dirty_log_pages); |
9e351cc8 JB |
119 | } |
120 | up_write(&fs_info->commit_root_sem); | |
817d52f8 JB |
121 | } |
122 | ||
0860adfd MX |
123 | static inline void extwriter_counter_inc(struct btrfs_transaction *trans, |
124 | unsigned int type) | |
125 | { | |
126 | if (type & TRANS_EXTWRITERS) | |
127 | atomic_inc(&trans->num_extwriters); | |
128 | } | |
129 | ||
130 | static inline void extwriter_counter_dec(struct btrfs_transaction *trans, | |
131 | unsigned int type) | |
132 | { | |
133 | if (type & TRANS_EXTWRITERS) | |
134 | atomic_dec(&trans->num_extwriters); | |
135 | } | |
136 | ||
137 | static inline void extwriter_counter_init(struct btrfs_transaction *trans, | |
138 | unsigned int type) | |
139 | { | |
140 | atomic_set(&trans->num_extwriters, ((type & TRANS_EXTWRITERS) ? 1 : 0)); | |
141 | } | |
142 | ||
143 | static inline int extwriter_counter_read(struct btrfs_transaction *trans) | |
144 | { | |
145 | return atomic_read(&trans->num_extwriters); | |
178260b2 MX |
146 | } |
147 | ||
d352ac68 CM |
148 | /* |
149 | * either allocate a new transaction or hop into the existing one | |
150 | */ | |
0860adfd | 151 | static noinline int join_transaction(struct btrfs_root *root, unsigned int type) |
79154b1b CM |
152 | { |
153 | struct btrfs_transaction *cur_trans; | |
19ae4e81 | 154 | struct btrfs_fs_info *fs_info = root->fs_info; |
a4abeea4 | 155 | |
19ae4e81 | 156 | spin_lock(&fs_info->trans_lock); |
d43317dc | 157 | loop: |
49b25e05 | 158 | /* The file system has been taken offline. No new transactions. */ |
87533c47 | 159 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { |
19ae4e81 | 160 | spin_unlock(&fs_info->trans_lock); |
49b25e05 JM |
161 | return -EROFS; |
162 | } | |
163 | ||
19ae4e81 | 164 | cur_trans = fs_info->running_transaction; |
a4abeea4 | 165 | if (cur_trans) { |
871383be | 166 | if (cur_trans->aborted) { |
19ae4e81 | 167 | spin_unlock(&fs_info->trans_lock); |
49b25e05 | 168 | return cur_trans->aborted; |
871383be | 169 | } |
4a9d8bde | 170 | if (btrfs_blocked_trans_types[cur_trans->state] & type) { |
178260b2 MX |
171 | spin_unlock(&fs_info->trans_lock); |
172 | return -EBUSY; | |
173 | } | |
a4abeea4 | 174 | atomic_inc(&cur_trans->use_count); |
13c5a93e | 175 | atomic_inc(&cur_trans->num_writers); |
0860adfd | 176 | extwriter_counter_inc(cur_trans, type); |
19ae4e81 | 177 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 | 178 | return 0; |
79154b1b | 179 | } |
19ae4e81 | 180 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 | 181 | |
354aa0fb MX |
182 | /* |
183 | * If we are ATTACH, we just want to catch the current transaction, | |
184 | * and commit it. If there is no transaction, just return ENOENT. | |
185 | */ | |
186 | if (type == TRANS_ATTACH) | |
187 | return -ENOENT; | |
188 | ||
4a9d8bde MX |
189 | /* |
190 | * JOIN_NOLOCK only happens during the transaction commit, so | |
191 | * it is impossible that ->running_transaction is NULL | |
192 | */ | |
193 | BUG_ON(type == TRANS_JOIN_NOLOCK); | |
194 | ||
a4abeea4 JB |
195 | cur_trans = kmem_cache_alloc(btrfs_transaction_cachep, GFP_NOFS); |
196 | if (!cur_trans) | |
197 | return -ENOMEM; | |
d43317dc | 198 | |
19ae4e81 JS |
199 | spin_lock(&fs_info->trans_lock); |
200 | if (fs_info->running_transaction) { | |
d43317dc CM |
201 | /* |
202 | * someone started a transaction after we unlocked. Make sure | |
4a9d8bde | 203 | * to redo the checks above |
d43317dc | 204 | */ |
a4abeea4 | 205 | kmem_cache_free(btrfs_transaction_cachep, cur_trans); |
d43317dc | 206 | goto loop; |
87533c47 | 207 | } else if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { |
e4b50e14 | 208 | spin_unlock(&fs_info->trans_lock); |
7b8b92af JB |
209 | kmem_cache_free(btrfs_transaction_cachep, cur_trans); |
210 | return -EROFS; | |
79154b1b | 211 | } |
d43317dc | 212 | |
a4abeea4 | 213 | atomic_set(&cur_trans->num_writers, 1); |
0860adfd | 214 | extwriter_counter_init(cur_trans, type); |
a4abeea4 JB |
215 | init_waitqueue_head(&cur_trans->writer_wait); |
216 | init_waitqueue_head(&cur_trans->commit_wait); | |
4a9d8bde | 217 | cur_trans->state = TRANS_STATE_RUNNING; |
a4abeea4 JB |
218 | /* |
219 | * One for this trans handle, one so it will live on until we | |
220 | * commit the transaction. | |
221 | */ | |
222 | atomic_set(&cur_trans->use_count, 2); | |
13212b54 | 223 | cur_trans->have_free_bgs = 0; |
a4abeea4 | 224 | cur_trans->start_time = get_seconds(); |
1bbc621e | 225 | cur_trans->dirty_bg_run = 0; |
a4abeea4 | 226 | |
c46effa6 | 227 | cur_trans->delayed_refs.href_root = RB_ROOT; |
3368d001 | 228 | cur_trans->delayed_refs.dirty_extent_root = RB_ROOT; |
d7df2c79 | 229 | atomic_set(&cur_trans->delayed_refs.num_entries, 0); |
a4abeea4 | 230 | cur_trans->delayed_refs.num_heads_ready = 0; |
1262133b | 231 | cur_trans->delayed_refs.pending_csums = 0; |
a4abeea4 JB |
232 | cur_trans->delayed_refs.num_heads = 0; |
233 | cur_trans->delayed_refs.flushing = 0; | |
234 | cur_trans->delayed_refs.run_delayed_start = 0; | |
9086db86 | 235 | cur_trans->delayed_refs.qgroup_to_skip = 0; |
20b297d6 JS |
236 | |
237 | /* | |
238 | * although the tree mod log is per file system and not per transaction, | |
239 | * the log must never go across transaction boundaries. | |
240 | */ | |
241 | smp_mb(); | |
31b1a2bd | 242 | if (!list_empty(&fs_info->tree_mod_seq_list)) |
efe120a0 | 243 | WARN(1, KERN_ERR "BTRFS: tree_mod_seq_list not empty when " |
20b297d6 | 244 | "creating a fresh transaction\n"); |
31b1a2bd | 245 | if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log)) |
efe120a0 | 246 | WARN(1, KERN_ERR "BTRFS: tree_mod_log rb tree not empty when " |
20b297d6 | 247 | "creating a fresh transaction\n"); |
fc36ed7e | 248 | atomic64_set(&fs_info->tree_mod_seq, 0); |
20b297d6 | 249 | |
a4abeea4 JB |
250 | spin_lock_init(&cur_trans->delayed_refs.lock); |
251 | ||
252 | INIT_LIST_HEAD(&cur_trans->pending_snapshots); | |
6df9a95e | 253 | INIT_LIST_HEAD(&cur_trans->pending_chunks); |
9e351cc8 | 254 | INIT_LIST_HEAD(&cur_trans->switch_commits); |
50d9aa99 | 255 | INIT_LIST_HEAD(&cur_trans->pending_ordered); |
ce93ec54 | 256 | INIT_LIST_HEAD(&cur_trans->dirty_bgs); |
1bbc621e CM |
257 | INIT_LIST_HEAD(&cur_trans->io_bgs); |
258 | mutex_init(&cur_trans->cache_write_mutex); | |
cb723e49 | 259 | cur_trans->num_dirty_bgs = 0; |
ce93ec54 | 260 | spin_lock_init(&cur_trans->dirty_bgs_lock); |
19ae4e81 | 261 | list_add_tail(&cur_trans->list, &fs_info->trans_list); |
a4abeea4 | 262 | extent_io_tree_init(&cur_trans->dirty_pages, |
19ae4e81 JS |
263 | fs_info->btree_inode->i_mapping); |
264 | fs_info->generation++; | |
265 | cur_trans->transid = fs_info->generation; | |
266 | fs_info->running_transaction = cur_trans; | |
49b25e05 | 267 | cur_trans->aborted = 0; |
19ae4e81 | 268 | spin_unlock(&fs_info->trans_lock); |
15ee9bc7 | 269 | |
79154b1b CM |
270 | return 0; |
271 | } | |
272 | ||
d352ac68 | 273 | /* |
d397712b CM |
274 | * this does all the record keeping required to make sure that a reference |
275 | * counted root is properly recorded in a given transaction. This is required | |
276 | * to make sure the old root from before we joined the transaction is deleted | |
277 | * when the transaction commits | |
d352ac68 | 278 | */ |
7585717f | 279 | static int record_root_in_trans(struct btrfs_trans_handle *trans, |
a4abeea4 | 280 | struct btrfs_root *root) |
6702ed49 | 281 | { |
27cdeb70 MX |
282 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
283 | root->last_trans < trans->transid) { | |
6702ed49 | 284 | WARN_ON(root == root->fs_info->extent_root); |
5d4f98a2 YZ |
285 | WARN_ON(root->commit_root != root->node); |
286 | ||
7585717f | 287 | /* |
27cdeb70 | 288 | * see below for IN_TRANS_SETUP usage rules |
7585717f CM |
289 | * we have the reloc mutex held now, so there |
290 | * is only one writer in this function | |
291 | */ | |
27cdeb70 | 292 | set_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); |
7585717f | 293 | |
27cdeb70 | 294 | /* make sure readers find IN_TRANS_SETUP before |
7585717f CM |
295 | * they find our root->last_trans update |
296 | */ | |
297 | smp_wmb(); | |
298 | ||
a4abeea4 JB |
299 | spin_lock(&root->fs_info->fs_roots_radix_lock); |
300 | if (root->last_trans == trans->transid) { | |
301 | spin_unlock(&root->fs_info->fs_roots_radix_lock); | |
302 | return 0; | |
303 | } | |
5d4f98a2 YZ |
304 | radix_tree_tag_set(&root->fs_info->fs_roots_radix, |
305 | (unsigned long)root->root_key.objectid, | |
306 | BTRFS_ROOT_TRANS_TAG); | |
a4abeea4 | 307 | spin_unlock(&root->fs_info->fs_roots_radix_lock); |
7585717f CM |
308 | root->last_trans = trans->transid; |
309 | ||
310 | /* this is pretty tricky. We don't want to | |
311 | * take the relocation lock in btrfs_record_root_in_trans | |
312 | * unless we're really doing the first setup for this root in | |
313 | * this transaction. | |
314 | * | |
315 | * Normally we'd use root->last_trans as a flag to decide | |
316 | * if we want to take the expensive mutex. | |
317 | * | |
318 | * But, we have to set root->last_trans before we | |
319 | * init the relocation root, otherwise, we trip over warnings | |
320 | * in ctree.c. The solution used here is to flag ourselves | |
27cdeb70 | 321 | * with root IN_TRANS_SETUP. When this is 1, we're still |
7585717f CM |
322 | * fixing up the reloc trees and everyone must wait. |
323 | * | |
324 | * When this is zero, they can trust root->last_trans and fly | |
325 | * through btrfs_record_root_in_trans without having to take the | |
326 | * lock. smp_wmb() makes sure that all the writes above are | |
327 | * done before we pop in the zero below | |
328 | */ | |
5d4f98a2 | 329 | btrfs_init_reloc_root(trans, root); |
c7548af6 | 330 | smp_mb__before_atomic(); |
27cdeb70 | 331 | clear_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); |
5d4f98a2 YZ |
332 | } |
333 | return 0; | |
334 | } | |
bcc63abb | 335 | |
7585717f CM |
336 | |
337 | int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, | |
338 | struct btrfs_root *root) | |
339 | { | |
27cdeb70 | 340 | if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state)) |
7585717f CM |
341 | return 0; |
342 | ||
343 | /* | |
27cdeb70 | 344 | * see record_root_in_trans for comments about IN_TRANS_SETUP usage |
7585717f CM |
345 | * and barriers |
346 | */ | |
347 | smp_rmb(); | |
348 | if (root->last_trans == trans->transid && | |
27cdeb70 | 349 | !test_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state)) |
7585717f CM |
350 | return 0; |
351 | ||
352 | mutex_lock(&root->fs_info->reloc_mutex); | |
353 | record_root_in_trans(trans, root); | |
354 | mutex_unlock(&root->fs_info->reloc_mutex); | |
355 | ||
356 | return 0; | |
357 | } | |
358 | ||
4a9d8bde MX |
359 | static inline int is_transaction_blocked(struct btrfs_transaction *trans) |
360 | { | |
361 | return (trans->state >= TRANS_STATE_BLOCKED && | |
501407aa JB |
362 | trans->state < TRANS_STATE_UNBLOCKED && |
363 | !trans->aborted); | |
4a9d8bde MX |
364 | } |
365 | ||
d352ac68 CM |
366 | /* wait for commit against the current transaction to become unblocked |
367 | * when this is done, it is safe to start a new transaction, but the current | |
368 | * transaction might not be fully on disk. | |
369 | */ | |
37d1aeee | 370 | static void wait_current_trans(struct btrfs_root *root) |
79154b1b | 371 | { |
f9295749 | 372 | struct btrfs_transaction *cur_trans; |
79154b1b | 373 | |
a4abeea4 | 374 | spin_lock(&root->fs_info->trans_lock); |
f9295749 | 375 | cur_trans = root->fs_info->running_transaction; |
4a9d8bde | 376 | if (cur_trans && is_transaction_blocked(cur_trans)) { |
13c5a93e | 377 | atomic_inc(&cur_trans->use_count); |
a4abeea4 | 378 | spin_unlock(&root->fs_info->trans_lock); |
72d63ed6 LZ |
379 | |
380 | wait_event(root->fs_info->transaction_wait, | |
501407aa JB |
381 | cur_trans->state >= TRANS_STATE_UNBLOCKED || |
382 | cur_trans->aborted); | |
724e2315 | 383 | btrfs_put_transaction(cur_trans); |
a4abeea4 JB |
384 | } else { |
385 | spin_unlock(&root->fs_info->trans_lock); | |
f9295749 | 386 | } |
37d1aeee CM |
387 | } |
388 | ||
a22285a6 YZ |
389 | static int may_wait_transaction(struct btrfs_root *root, int type) |
390 | { | |
a4abeea4 JB |
391 | if (root->fs_info->log_root_recovering) |
392 | return 0; | |
393 | ||
394 | if (type == TRANS_USERSPACE) | |
395 | return 1; | |
396 | ||
397 | if (type == TRANS_START && | |
398 | !atomic_read(&root->fs_info->open_ioctl_trans)) | |
a22285a6 | 399 | return 1; |
a4abeea4 | 400 | |
a22285a6 YZ |
401 | return 0; |
402 | } | |
403 | ||
20dd2cbf MX |
404 | static inline bool need_reserve_reloc_root(struct btrfs_root *root) |
405 | { | |
406 | if (!root->fs_info->reloc_ctl || | |
27cdeb70 | 407 | !test_bit(BTRFS_ROOT_REF_COWS, &root->state) || |
20dd2cbf MX |
408 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || |
409 | root->reloc_root) | |
410 | return false; | |
411 | ||
412 | return true; | |
413 | } | |
414 | ||
08e007d2 | 415 | static struct btrfs_trans_handle * |
0860adfd | 416 | start_transaction(struct btrfs_root *root, u64 num_items, unsigned int type, |
08e007d2 | 417 | enum btrfs_reserve_flush_enum flush) |
37d1aeee | 418 | { |
a22285a6 YZ |
419 | struct btrfs_trans_handle *h; |
420 | struct btrfs_transaction *cur_trans; | |
b5009945 | 421 | u64 num_bytes = 0; |
c5567237 | 422 | u64 qgroup_reserved = 0; |
20dd2cbf MX |
423 | bool reloc_reserved = false; |
424 | int ret; | |
acce952b | 425 | |
46c4e71e | 426 | /* Send isn't supposed to start transactions. */ |
2755a0de | 427 | ASSERT(current->journal_info != BTRFS_SEND_TRANS_STUB); |
46c4e71e | 428 | |
87533c47 | 429 | if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) |
acce952b | 430 | return ERR_PTR(-EROFS); |
2a1eb461 | 431 | |
46c4e71e | 432 | if (current->journal_info) { |
0860adfd | 433 | WARN_ON(type & TRANS_EXTWRITERS); |
2a1eb461 JB |
434 | h = current->journal_info; |
435 | h->use_count++; | |
b7d5b0a8 | 436 | WARN_ON(h->use_count > 2); |
2a1eb461 JB |
437 | h->orig_rsv = h->block_rsv; |
438 | h->block_rsv = NULL; | |
439 | goto got_it; | |
440 | } | |
b5009945 JB |
441 | |
442 | /* | |
443 | * Do the reservation before we join the transaction so we can do all | |
444 | * the appropriate flushing if need be. | |
445 | */ | |
446 | if (num_items > 0 && root != root->fs_info->chunk_root) { | |
c5567237 AJ |
447 | if (root->fs_info->quota_enabled && |
448 | is_fstree(root->root_key.objectid)) { | |
707e8a07 | 449 | qgroup_reserved = num_items * root->nodesize; |
c5567237 AJ |
450 | ret = btrfs_qgroup_reserve(root, qgroup_reserved); |
451 | if (ret) | |
452 | return ERR_PTR(ret); | |
453 | } | |
454 | ||
b5009945 | 455 | num_bytes = btrfs_calc_trans_metadata_size(root, num_items); |
20dd2cbf MX |
456 | /* |
457 | * Do the reservation for the relocation root creation | |
458 | */ | |
ee39b432 | 459 | if (need_reserve_reloc_root(root)) { |
20dd2cbf MX |
460 | num_bytes += root->nodesize; |
461 | reloc_reserved = true; | |
462 | } | |
463 | ||
08e007d2 MX |
464 | ret = btrfs_block_rsv_add(root, |
465 | &root->fs_info->trans_block_rsv, | |
466 | num_bytes, flush); | |
b5009945 | 467 | if (ret) |
843fcf35 | 468 | goto reserve_fail; |
b5009945 | 469 | } |
a22285a6 YZ |
470 | again: |
471 | h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS); | |
843fcf35 MX |
472 | if (!h) { |
473 | ret = -ENOMEM; | |
474 | goto alloc_fail; | |
475 | } | |
37d1aeee | 476 | |
98114659 JB |
477 | /* |
478 | * If we are JOIN_NOLOCK we're already committing a transaction and | |
479 | * waiting on this guy, so we don't need to do the sb_start_intwrite | |
480 | * because we're already holding a ref. We need this because we could | |
481 | * have raced in and did an fsync() on a file which can kick a commit | |
482 | * and then we deadlock with somebody doing a freeze. | |
354aa0fb MX |
483 | * |
484 | * If we are ATTACH, it means we just want to catch the current | |
485 | * transaction and commit it, so we needn't do sb_start_intwrite(). | |
98114659 | 486 | */ |
0860adfd | 487 | if (type & __TRANS_FREEZABLE) |
60376ce4 | 488 | sb_start_intwrite(root->fs_info->sb); |
b2b5ef5c | 489 | |
a22285a6 | 490 | if (may_wait_transaction(root, type)) |
37d1aeee | 491 | wait_current_trans(root); |
a22285a6 | 492 | |
a4abeea4 | 493 | do { |
354aa0fb | 494 | ret = join_transaction(root, type); |
178260b2 | 495 | if (ret == -EBUSY) { |
a4abeea4 | 496 | wait_current_trans(root); |
178260b2 MX |
497 | if (unlikely(type == TRANS_ATTACH)) |
498 | ret = -ENOENT; | |
499 | } | |
a4abeea4 JB |
500 | } while (ret == -EBUSY); |
501 | ||
db5b493a | 502 | if (ret < 0) { |
354aa0fb MX |
503 | /* We must get the transaction if we are JOIN_NOLOCK. */ |
504 | BUG_ON(type == TRANS_JOIN_NOLOCK); | |
843fcf35 | 505 | goto join_fail; |
db5b493a | 506 | } |
0f7d52f4 | 507 | |
a22285a6 | 508 | cur_trans = root->fs_info->running_transaction; |
a22285a6 YZ |
509 | |
510 | h->transid = cur_trans->transid; | |
511 | h->transaction = cur_trans; | |
79154b1b | 512 | h->blocks_used = 0; |
a22285a6 | 513 | h->bytes_reserved = 0; |
4fbcdf66 | 514 | h->chunk_bytes_reserved = 0; |
d13603ef | 515 | h->root = root; |
56bec294 | 516 | h->delayed_ref_updates = 0; |
2a1eb461 | 517 | h->use_count = 1; |
0e721106 | 518 | h->adding_csums = 0; |
f0486c68 | 519 | h->block_rsv = NULL; |
2a1eb461 | 520 | h->orig_rsv = NULL; |
49b25e05 | 521 | h->aborted = 0; |
4b824906 | 522 | h->qgroup_reserved = 0; |
bed92eae | 523 | h->delayed_ref_elem.seq = 0; |
a698d075 | 524 | h->type = type; |
c6b305a8 | 525 | h->allocating_chunk = false; |
20dd2cbf | 526 | h->reloc_reserved = false; |
5039eddc | 527 | h->sync = false; |
bed92eae | 528 | INIT_LIST_HEAD(&h->qgroup_ref_list); |
ea658bad | 529 | INIT_LIST_HEAD(&h->new_bgs); |
50d9aa99 | 530 | INIT_LIST_HEAD(&h->ordered); |
b7ec40d7 | 531 | |
a22285a6 | 532 | smp_mb(); |
4a9d8bde MX |
533 | if (cur_trans->state >= TRANS_STATE_BLOCKED && |
534 | may_wait_transaction(root, type)) { | |
abdd2e80 | 535 | current->journal_info = h; |
a22285a6 YZ |
536 | btrfs_commit_transaction(h, root); |
537 | goto again; | |
538 | } | |
539 | ||
b5009945 | 540 | if (num_bytes) { |
8c2a3ca2 | 541 | trace_btrfs_space_reservation(root->fs_info, "transaction", |
2bcc0328 | 542 | h->transid, num_bytes, 1); |
b5009945 JB |
543 | h->block_rsv = &root->fs_info->trans_block_rsv; |
544 | h->bytes_reserved = num_bytes; | |
20dd2cbf | 545 | h->reloc_reserved = reloc_reserved; |
a22285a6 | 546 | } |
4b824906 | 547 | h->qgroup_reserved = qgroup_reserved; |
9ed74f2d | 548 | |
2a1eb461 | 549 | got_it: |
a4abeea4 | 550 | btrfs_record_root_in_trans(h, root); |
a22285a6 YZ |
551 | |
552 | if (!current->journal_info && type != TRANS_USERSPACE) | |
553 | current->journal_info = h; | |
79154b1b | 554 | return h; |
843fcf35 MX |
555 | |
556 | join_fail: | |
0860adfd | 557 | if (type & __TRANS_FREEZABLE) |
843fcf35 MX |
558 | sb_end_intwrite(root->fs_info->sb); |
559 | kmem_cache_free(btrfs_trans_handle_cachep, h); | |
560 | alloc_fail: | |
561 | if (num_bytes) | |
562 | btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv, | |
563 | num_bytes); | |
564 | reserve_fail: | |
565 | if (qgroup_reserved) | |
566 | btrfs_qgroup_free(root, qgroup_reserved); | |
567 | return ERR_PTR(ret); | |
79154b1b CM |
568 | } |
569 | ||
f9295749 | 570 | struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, |
a22285a6 | 571 | int num_items) |
f9295749 | 572 | { |
08e007d2 MX |
573 | return start_transaction(root, num_items, TRANS_START, |
574 | BTRFS_RESERVE_FLUSH_ALL); | |
f9295749 | 575 | } |
8407aa46 | 576 | |
08e007d2 | 577 | struct btrfs_trans_handle *btrfs_start_transaction_lflush( |
8407aa46 MX |
578 | struct btrfs_root *root, int num_items) |
579 | { | |
08e007d2 MX |
580 | return start_transaction(root, num_items, TRANS_START, |
581 | BTRFS_RESERVE_FLUSH_LIMIT); | |
8407aa46 MX |
582 | } |
583 | ||
7a7eaa40 | 584 | struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root) |
f9295749 | 585 | { |
8407aa46 | 586 | return start_transaction(root, 0, TRANS_JOIN, 0); |
f9295749 CM |
587 | } |
588 | ||
7a7eaa40 | 589 | struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root) |
0af3d00b | 590 | { |
8407aa46 | 591 | return start_transaction(root, 0, TRANS_JOIN_NOLOCK, 0); |
0af3d00b JB |
592 | } |
593 | ||
7a7eaa40 | 594 | struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root) |
9ca9ee09 | 595 | { |
8407aa46 | 596 | return start_transaction(root, 0, TRANS_USERSPACE, 0); |
9ca9ee09 SW |
597 | } |
598 | ||
d4edf39b MX |
599 | /* |
600 | * btrfs_attach_transaction() - catch the running transaction | |
601 | * | |
602 | * It is used when we want to commit the current the transaction, but | |
603 | * don't want to start a new one. | |
604 | * | |
605 | * Note: If this function return -ENOENT, it just means there is no | |
606 | * running transaction. But it is possible that the inactive transaction | |
607 | * is still in the memory, not fully on disk. If you hope there is no | |
608 | * inactive transaction in the fs when -ENOENT is returned, you should | |
609 | * invoke | |
610 | * btrfs_attach_transaction_barrier() | |
611 | */ | |
354aa0fb | 612 | struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root) |
60376ce4 | 613 | { |
354aa0fb | 614 | return start_transaction(root, 0, TRANS_ATTACH, 0); |
60376ce4 JB |
615 | } |
616 | ||
d4edf39b | 617 | /* |
90b6d283 | 618 | * btrfs_attach_transaction_barrier() - catch the running transaction |
d4edf39b MX |
619 | * |
620 | * It is similar to the above function, the differentia is this one | |
621 | * will wait for all the inactive transactions until they fully | |
622 | * complete. | |
623 | */ | |
624 | struct btrfs_trans_handle * | |
625 | btrfs_attach_transaction_barrier(struct btrfs_root *root) | |
626 | { | |
627 | struct btrfs_trans_handle *trans; | |
628 | ||
629 | trans = start_transaction(root, 0, TRANS_ATTACH, 0); | |
630 | if (IS_ERR(trans) && PTR_ERR(trans) == -ENOENT) | |
631 | btrfs_wait_for_commit(root, 0); | |
632 | ||
633 | return trans; | |
634 | } | |
635 | ||
d352ac68 | 636 | /* wait for a transaction commit to be fully complete */ |
b9c8300c | 637 | static noinline void wait_for_commit(struct btrfs_root *root, |
89ce8a63 CM |
638 | struct btrfs_transaction *commit) |
639 | { | |
4a9d8bde | 640 | wait_event(commit->commit_wait, commit->state == TRANS_STATE_COMPLETED); |
89ce8a63 CM |
641 | } |
642 | ||
46204592 SW |
643 | int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid) |
644 | { | |
645 | struct btrfs_transaction *cur_trans = NULL, *t; | |
8cd2807f | 646 | int ret = 0; |
46204592 | 647 | |
46204592 SW |
648 | if (transid) { |
649 | if (transid <= root->fs_info->last_trans_committed) | |
a4abeea4 | 650 | goto out; |
46204592 SW |
651 | |
652 | /* find specified transaction */ | |
a4abeea4 | 653 | spin_lock(&root->fs_info->trans_lock); |
46204592 SW |
654 | list_for_each_entry(t, &root->fs_info->trans_list, list) { |
655 | if (t->transid == transid) { | |
656 | cur_trans = t; | |
a4abeea4 | 657 | atomic_inc(&cur_trans->use_count); |
8cd2807f | 658 | ret = 0; |
46204592 SW |
659 | break; |
660 | } | |
8cd2807f MX |
661 | if (t->transid > transid) { |
662 | ret = 0; | |
46204592 | 663 | break; |
8cd2807f | 664 | } |
46204592 | 665 | } |
a4abeea4 | 666 | spin_unlock(&root->fs_info->trans_lock); |
42383020 SW |
667 | |
668 | /* | |
669 | * The specified transaction doesn't exist, or we | |
670 | * raced with btrfs_commit_transaction | |
671 | */ | |
672 | if (!cur_trans) { | |
673 | if (transid > root->fs_info->last_trans_committed) | |
674 | ret = -EINVAL; | |
8cd2807f | 675 | goto out; |
42383020 | 676 | } |
46204592 SW |
677 | } else { |
678 | /* find newest transaction that is committing | committed */ | |
a4abeea4 | 679 | spin_lock(&root->fs_info->trans_lock); |
46204592 SW |
680 | list_for_each_entry_reverse(t, &root->fs_info->trans_list, |
681 | list) { | |
4a9d8bde MX |
682 | if (t->state >= TRANS_STATE_COMMIT_START) { |
683 | if (t->state == TRANS_STATE_COMPLETED) | |
3473f3c0 | 684 | break; |
46204592 | 685 | cur_trans = t; |
a4abeea4 | 686 | atomic_inc(&cur_trans->use_count); |
46204592 SW |
687 | break; |
688 | } | |
689 | } | |
a4abeea4 | 690 | spin_unlock(&root->fs_info->trans_lock); |
46204592 | 691 | if (!cur_trans) |
a4abeea4 | 692 | goto out; /* nothing committing|committed */ |
46204592 SW |
693 | } |
694 | ||
46204592 | 695 | wait_for_commit(root, cur_trans); |
724e2315 | 696 | btrfs_put_transaction(cur_trans); |
a4abeea4 | 697 | out: |
46204592 SW |
698 | return ret; |
699 | } | |
700 | ||
37d1aeee CM |
701 | void btrfs_throttle(struct btrfs_root *root) |
702 | { | |
a4abeea4 | 703 | if (!atomic_read(&root->fs_info->open_ioctl_trans)) |
9ca9ee09 | 704 | wait_current_trans(root); |
37d1aeee CM |
705 | } |
706 | ||
8929ecfa YZ |
707 | static int should_end_transaction(struct btrfs_trans_handle *trans, |
708 | struct btrfs_root *root) | |
709 | { | |
1be41b78 | 710 | if (root->fs_info->global_block_rsv.space_info->full && |
0a2b2a84 | 711 | btrfs_check_space_for_delayed_refs(trans, root)) |
1be41b78 | 712 | return 1; |
36ba022a | 713 | |
1be41b78 | 714 | return !!btrfs_block_rsv_check(root, &root->fs_info->global_block_rsv, 5); |
8929ecfa YZ |
715 | } |
716 | ||
717 | int btrfs_should_end_transaction(struct btrfs_trans_handle *trans, | |
718 | struct btrfs_root *root) | |
719 | { | |
720 | struct btrfs_transaction *cur_trans = trans->transaction; | |
721 | int updates; | |
49b25e05 | 722 | int err; |
8929ecfa | 723 | |
a4abeea4 | 724 | smp_mb(); |
4a9d8bde MX |
725 | if (cur_trans->state >= TRANS_STATE_BLOCKED || |
726 | cur_trans->delayed_refs.flushing) | |
8929ecfa YZ |
727 | return 1; |
728 | ||
729 | updates = trans->delayed_ref_updates; | |
730 | trans->delayed_ref_updates = 0; | |
49b25e05 | 731 | if (updates) { |
28ed1345 | 732 | err = btrfs_run_delayed_refs(trans, root, updates * 2); |
49b25e05 JM |
733 | if (err) /* Error code will also eval true */ |
734 | return err; | |
735 | } | |
8929ecfa YZ |
736 | |
737 | return should_end_transaction(trans, root); | |
738 | } | |
739 | ||
89ce8a63 | 740 | static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, |
a698d075 | 741 | struct btrfs_root *root, int throttle) |
79154b1b | 742 | { |
8929ecfa | 743 | struct btrfs_transaction *cur_trans = trans->transaction; |
ab78c84d | 744 | struct btrfs_fs_info *info = root->fs_info; |
1be41b78 | 745 | unsigned long cur = trans->delayed_ref_updates; |
a698d075 | 746 | int lock = (trans->type != TRANS_JOIN_NOLOCK); |
4edc2ca3 | 747 | int err = 0; |
a79b7d4b | 748 | int must_run_delayed_refs = 0; |
c3e69d58 | 749 | |
3bbb24b2 JB |
750 | if (trans->use_count > 1) { |
751 | trans->use_count--; | |
2a1eb461 JB |
752 | trans->block_rsv = trans->orig_rsv; |
753 | return 0; | |
754 | } | |
755 | ||
b24e03db | 756 | btrfs_trans_release_metadata(trans, root); |
4c13d758 | 757 | trans->block_rsv = NULL; |
c5567237 | 758 | |
ea658bad JB |
759 | if (!list_empty(&trans->new_bgs)) |
760 | btrfs_create_pending_block_groups(trans, root); | |
761 | ||
50d9aa99 JB |
762 | if (!list_empty(&trans->ordered)) { |
763 | spin_lock(&info->trans_lock); | |
d3efe084 | 764 | list_splice_init(&trans->ordered, &cur_trans->pending_ordered); |
50d9aa99 JB |
765 | spin_unlock(&info->trans_lock); |
766 | } | |
767 | ||
1be41b78 | 768 | trans->delayed_ref_updates = 0; |
a79b7d4b CM |
769 | if (!trans->sync) { |
770 | must_run_delayed_refs = | |
771 | btrfs_should_throttle_delayed_refs(trans, root); | |
0a2b2a84 | 772 | cur = max_t(unsigned long, cur, 32); |
a79b7d4b CM |
773 | |
774 | /* | |
775 | * don't make the caller wait if they are from a NOLOCK | |
776 | * or ATTACH transaction, it will deadlock with commit | |
777 | */ | |
778 | if (must_run_delayed_refs == 1 && | |
779 | (trans->type & (__TRANS_JOIN_NOLOCK | __TRANS_ATTACH))) | |
780 | must_run_delayed_refs = 2; | |
56bec294 | 781 | } |
bb721703 | 782 | |
fcebe456 JB |
783 | if (trans->qgroup_reserved) { |
784 | /* | |
785 | * the same root has to be passed here between start_transaction | |
786 | * and end_transaction. Subvolume quota depends on this. | |
787 | */ | |
788 | btrfs_qgroup_free(trans->root, trans->qgroup_reserved); | |
789 | trans->qgroup_reserved = 0; | |
790 | } | |
791 | ||
0e721106 JB |
792 | btrfs_trans_release_metadata(trans, root); |
793 | trans->block_rsv = NULL; | |
56bec294 | 794 | |
ea658bad JB |
795 | if (!list_empty(&trans->new_bgs)) |
796 | btrfs_create_pending_block_groups(trans, root); | |
797 | ||
4fbcdf66 FM |
798 | btrfs_trans_release_chunk_metadata(trans); |
799 | ||
a4abeea4 | 800 | if (lock && !atomic_read(&root->fs_info->open_ioctl_trans) && |
4a9d8bde MX |
801 | should_end_transaction(trans, root) && |
802 | ACCESS_ONCE(cur_trans->state) == TRANS_STATE_RUNNING) { | |
803 | spin_lock(&info->trans_lock); | |
804 | if (cur_trans->state == TRANS_STATE_RUNNING) | |
805 | cur_trans->state = TRANS_STATE_BLOCKED; | |
806 | spin_unlock(&info->trans_lock); | |
a4abeea4 | 807 | } |
8929ecfa | 808 | |
4a9d8bde | 809 | if (lock && ACCESS_ONCE(cur_trans->state) == TRANS_STATE_BLOCKED) { |
3bbb24b2 | 810 | if (throttle) |
8929ecfa | 811 | return btrfs_commit_transaction(trans, root); |
3bbb24b2 | 812 | else |
8929ecfa YZ |
813 | wake_up_process(info->transaction_kthread); |
814 | } | |
815 | ||
0860adfd | 816 | if (trans->type & __TRANS_FREEZABLE) |
98114659 | 817 | sb_end_intwrite(root->fs_info->sb); |
6df7881a | 818 | |
8929ecfa | 819 | WARN_ON(cur_trans != info->running_transaction); |
13c5a93e JB |
820 | WARN_ON(atomic_read(&cur_trans->num_writers) < 1); |
821 | atomic_dec(&cur_trans->num_writers); | |
0860adfd | 822 | extwriter_counter_dec(cur_trans, trans->type); |
89ce8a63 | 823 | |
99d16cbc | 824 | smp_mb(); |
79154b1b CM |
825 | if (waitqueue_active(&cur_trans->writer_wait)) |
826 | wake_up(&cur_trans->writer_wait); | |
724e2315 | 827 | btrfs_put_transaction(cur_trans); |
9ed74f2d JB |
828 | |
829 | if (current->journal_info == trans) | |
830 | current->journal_info = NULL; | |
ab78c84d | 831 | |
24bbcf04 YZ |
832 | if (throttle) |
833 | btrfs_run_delayed_iputs(root); | |
834 | ||
49b25e05 | 835 | if (trans->aborted || |
4e121c06 JB |
836 | test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) { |
837 | wake_up_process(info->transaction_kthread); | |
4edc2ca3 | 838 | err = -EIO; |
4e121c06 | 839 | } |
edf39272 | 840 | assert_qgroups_uptodate(trans); |
49b25e05 | 841 | |
4edc2ca3 | 842 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
a79b7d4b CM |
843 | if (must_run_delayed_refs) { |
844 | btrfs_async_run_delayed_refs(root, cur, | |
845 | must_run_delayed_refs == 1); | |
846 | } | |
4edc2ca3 | 847 | return err; |
79154b1b CM |
848 | } |
849 | ||
89ce8a63 CM |
850 | int btrfs_end_transaction(struct btrfs_trans_handle *trans, |
851 | struct btrfs_root *root) | |
852 | { | |
98ad43be | 853 | return __btrfs_end_transaction(trans, root, 0); |
89ce8a63 CM |
854 | } |
855 | ||
856 | int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans, | |
857 | struct btrfs_root *root) | |
858 | { | |
98ad43be | 859 | return __btrfs_end_transaction(trans, root, 1); |
16cdcec7 MX |
860 | } |
861 | ||
d352ac68 CM |
862 | /* |
863 | * when btree blocks are allocated, they have some corresponding bits set for | |
864 | * them in one of two extent_io trees. This is used to make sure all of | |
690587d1 | 865 | * those extents are sent to disk but does not wait on them |
d352ac68 | 866 | */ |
690587d1 | 867 | int btrfs_write_marked_extents(struct btrfs_root *root, |
8cef4e16 | 868 | struct extent_io_tree *dirty_pages, int mark) |
79154b1b | 869 | { |
777e6bd7 | 870 | int err = 0; |
7c4452b9 | 871 | int werr = 0; |
1728366e | 872 | struct address_space *mapping = root->fs_info->btree_inode->i_mapping; |
e6138876 | 873 | struct extent_state *cached_state = NULL; |
777e6bd7 | 874 | u64 start = 0; |
5f39d397 | 875 | u64 end; |
7c4452b9 | 876 | |
1728366e | 877 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
e6138876 | 878 | mark, &cached_state)) { |
663dfbb0 FM |
879 | bool wait_writeback = false; |
880 | ||
881 | err = convert_extent_bit(dirty_pages, start, end, | |
882 | EXTENT_NEED_WAIT, | |
883 | mark, &cached_state, GFP_NOFS); | |
884 | /* | |
885 | * convert_extent_bit can return -ENOMEM, which is most of the | |
886 | * time a temporary error. So when it happens, ignore the error | |
887 | * and wait for writeback of this range to finish - because we | |
888 | * failed to set the bit EXTENT_NEED_WAIT for the range, a call | |
889 | * to btrfs_wait_marked_extents() would not know that writeback | |
890 | * for this range started and therefore wouldn't wait for it to | |
891 | * finish - we don't want to commit a superblock that points to | |
892 | * btree nodes/leafs for which writeback hasn't finished yet | |
893 | * (and without errors). | |
894 | * We cleanup any entries left in the io tree when committing | |
895 | * the transaction (through clear_btree_io_tree()). | |
896 | */ | |
897 | if (err == -ENOMEM) { | |
898 | err = 0; | |
899 | wait_writeback = true; | |
900 | } | |
901 | if (!err) | |
902 | err = filemap_fdatawrite_range(mapping, start, end); | |
1728366e JB |
903 | if (err) |
904 | werr = err; | |
663dfbb0 FM |
905 | else if (wait_writeback) |
906 | werr = filemap_fdatawait_range(mapping, start, end); | |
e38e2ed7 | 907 | free_extent_state(cached_state); |
663dfbb0 | 908 | cached_state = NULL; |
1728366e JB |
909 | cond_resched(); |
910 | start = end + 1; | |
7c4452b9 | 911 | } |
690587d1 CM |
912 | return werr; |
913 | } | |
914 | ||
915 | /* | |
916 | * when btree blocks are allocated, they have some corresponding bits set for | |
917 | * them in one of two extent_io trees. This is used to make sure all of | |
918 | * those extents are on disk for transaction or log commit. We wait | |
919 | * on all the pages and clear them from the dirty pages state tree | |
920 | */ | |
921 | int btrfs_wait_marked_extents(struct btrfs_root *root, | |
8cef4e16 | 922 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 | 923 | { |
690587d1 CM |
924 | int err = 0; |
925 | int werr = 0; | |
1728366e | 926 | struct address_space *mapping = root->fs_info->btree_inode->i_mapping; |
e6138876 | 927 | struct extent_state *cached_state = NULL; |
690587d1 CM |
928 | u64 start = 0; |
929 | u64 end; | |
656f30db FM |
930 | struct btrfs_inode *btree_ino = BTRFS_I(root->fs_info->btree_inode); |
931 | bool errors = false; | |
777e6bd7 | 932 | |
1728366e | 933 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
e6138876 | 934 | EXTENT_NEED_WAIT, &cached_state)) { |
663dfbb0 FM |
935 | /* |
936 | * Ignore -ENOMEM errors returned by clear_extent_bit(). | |
937 | * When committing the transaction, we'll remove any entries | |
938 | * left in the io tree. For a log commit, we don't remove them | |
939 | * after committing the log because the tree can be accessed | |
940 | * concurrently - we do it only at transaction commit time when | |
941 | * it's safe to do it (through clear_btree_io_tree()). | |
942 | */ | |
943 | err = clear_extent_bit(dirty_pages, start, end, | |
944 | EXTENT_NEED_WAIT, | |
945 | 0, 0, &cached_state, GFP_NOFS); | |
946 | if (err == -ENOMEM) | |
947 | err = 0; | |
948 | if (!err) | |
949 | err = filemap_fdatawait_range(mapping, start, end); | |
1728366e JB |
950 | if (err) |
951 | werr = err; | |
e38e2ed7 FM |
952 | free_extent_state(cached_state); |
953 | cached_state = NULL; | |
1728366e JB |
954 | cond_resched(); |
955 | start = end + 1; | |
777e6bd7 | 956 | } |
7c4452b9 CM |
957 | if (err) |
958 | werr = err; | |
656f30db FM |
959 | |
960 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { | |
961 | if ((mark & EXTENT_DIRTY) && | |
962 | test_and_clear_bit(BTRFS_INODE_BTREE_LOG1_ERR, | |
963 | &btree_ino->runtime_flags)) | |
964 | errors = true; | |
965 | ||
966 | if ((mark & EXTENT_NEW) && | |
967 | test_and_clear_bit(BTRFS_INODE_BTREE_LOG2_ERR, | |
968 | &btree_ino->runtime_flags)) | |
969 | errors = true; | |
970 | } else { | |
971 | if (test_and_clear_bit(BTRFS_INODE_BTREE_ERR, | |
972 | &btree_ino->runtime_flags)) | |
973 | errors = true; | |
974 | } | |
975 | ||
976 | if (errors && !werr) | |
977 | werr = -EIO; | |
978 | ||
7c4452b9 | 979 | return werr; |
79154b1b CM |
980 | } |
981 | ||
690587d1 CM |
982 | /* |
983 | * when btree blocks are allocated, they have some corresponding bits set for | |
984 | * them in one of two extent_io trees. This is used to make sure all of | |
985 | * those extents are on disk for transaction or log commit | |
986 | */ | |
171170c1 | 987 | static int btrfs_write_and_wait_marked_extents(struct btrfs_root *root, |
8cef4e16 | 988 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 CM |
989 | { |
990 | int ret; | |
991 | int ret2; | |
c6adc9cc | 992 | struct blk_plug plug; |
690587d1 | 993 | |
c6adc9cc | 994 | blk_start_plug(&plug); |
8cef4e16 | 995 | ret = btrfs_write_marked_extents(root, dirty_pages, mark); |
c6adc9cc | 996 | blk_finish_plug(&plug); |
8cef4e16 | 997 | ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark); |
bf0da8c1 CM |
998 | |
999 | if (ret) | |
1000 | return ret; | |
1001 | if (ret2) | |
1002 | return ret2; | |
1003 | return 0; | |
690587d1 CM |
1004 | } |
1005 | ||
663dfbb0 | 1006 | static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans, |
d0c803c4 CM |
1007 | struct btrfs_root *root) |
1008 | { | |
663dfbb0 FM |
1009 | int ret; |
1010 | ||
1011 | ret = btrfs_write_and_wait_marked_extents(root, | |
8cef4e16 YZ |
1012 | &trans->transaction->dirty_pages, |
1013 | EXTENT_DIRTY); | |
663dfbb0 FM |
1014 | clear_btree_io_tree(&trans->transaction->dirty_pages); |
1015 | ||
1016 | return ret; | |
d0c803c4 CM |
1017 | } |
1018 | ||
d352ac68 CM |
1019 | /* |
1020 | * this is used to update the root pointer in the tree of tree roots. | |
1021 | * | |
1022 | * But, in the case of the extent allocation tree, updating the root | |
1023 | * pointer may allocate blocks which may change the root of the extent | |
1024 | * allocation tree. | |
1025 | * | |
1026 | * So, this loops and repeats and makes sure the cowonly root didn't | |
1027 | * change while the root pointer was being updated in the metadata. | |
1028 | */ | |
0b86a832 CM |
1029 | static int update_cowonly_root(struct btrfs_trans_handle *trans, |
1030 | struct btrfs_root *root) | |
79154b1b CM |
1031 | { |
1032 | int ret; | |
0b86a832 | 1033 | u64 old_root_bytenr; |
86b9f2ec | 1034 | u64 old_root_used; |
0b86a832 | 1035 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
79154b1b | 1036 | |
86b9f2ec | 1037 | old_root_used = btrfs_root_used(&root->root_item); |
56bec294 | 1038 | |
d397712b | 1039 | while (1) { |
0b86a832 | 1040 | old_root_bytenr = btrfs_root_bytenr(&root->root_item); |
86b9f2ec | 1041 | if (old_root_bytenr == root->node->start && |
ea526d18 | 1042 | old_root_used == btrfs_root_used(&root->root_item)) |
79154b1b | 1043 | break; |
87ef2bb4 | 1044 | |
5d4f98a2 | 1045 | btrfs_set_root_node(&root->root_item, root->node); |
79154b1b | 1046 | ret = btrfs_update_root(trans, tree_root, |
0b86a832 CM |
1047 | &root->root_key, |
1048 | &root->root_item); | |
49b25e05 JM |
1049 | if (ret) |
1050 | return ret; | |
56bec294 | 1051 | |
86b9f2ec | 1052 | old_root_used = btrfs_root_used(&root->root_item); |
0b86a832 | 1053 | } |
276e680d | 1054 | |
0b86a832 CM |
1055 | return 0; |
1056 | } | |
1057 | ||
d352ac68 CM |
1058 | /* |
1059 | * update all the cowonly tree roots on disk | |
49b25e05 JM |
1060 | * |
1061 | * The error handling in this function may not be obvious. Any of the | |
1062 | * failures will cause the file system to go offline. We still need | |
1063 | * to clean up the delayed refs. | |
d352ac68 | 1064 | */ |
5d4f98a2 YZ |
1065 | static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans, |
1066 | struct btrfs_root *root) | |
0b86a832 CM |
1067 | { |
1068 | struct btrfs_fs_info *fs_info = root->fs_info; | |
ea526d18 | 1069 | struct list_head *dirty_bgs = &trans->transaction->dirty_bgs; |
1bbc621e | 1070 | struct list_head *io_bgs = &trans->transaction->io_bgs; |
0b86a832 | 1071 | struct list_head *next; |
84234f3a | 1072 | struct extent_buffer *eb; |
56bec294 | 1073 | int ret; |
84234f3a YZ |
1074 | |
1075 | eb = btrfs_lock_root_node(fs_info->tree_root); | |
49b25e05 JM |
1076 | ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, |
1077 | 0, &eb); | |
84234f3a YZ |
1078 | btrfs_tree_unlock(eb); |
1079 | free_extent_buffer(eb); | |
0b86a832 | 1080 | |
49b25e05 JM |
1081 | if (ret) |
1082 | return ret; | |
1083 | ||
56bec294 | 1084 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
49b25e05 JM |
1085 | if (ret) |
1086 | return ret; | |
87ef2bb4 | 1087 | |
733f4fbb | 1088 | ret = btrfs_run_dev_stats(trans, root->fs_info); |
c16ce190 JB |
1089 | if (ret) |
1090 | return ret; | |
8dabb742 | 1091 | ret = btrfs_run_dev_replace(trans, root->fs_info); |
c16ce190 JB |
1092 | if (ret) |
1093 | return ret; | |
546adb0d | 1094 | ret = btrfs_run_qgroups(trans, root->fs_info); |
c16ce190 JB |
1095 | if (ret) |
1096 | return ret; | |
546adb0d | 1097 | |
dcdf7f6d JB |
1098 | ret = btrfs_setup_space_cache(trans, root); |
1099 | if (ret) | |
1100 | return ret; | |
1101 | ||
546adb0d JS |
1102 | /* run_qgroups might have added some more refs */ |
1103 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); | |
c16ce190 JB |
1104 | if (ret) |
1105 | return ret; | |
ea526d18 | 1106 | again: |
d397712b | 1107 | while (!list_empty(&fs_info->dirty_cowonly_roots)) { |
0b86a832 CM |
1108 | next = fs_info->dirty_cowonly_roots.next; |
1109 | list_del_init(next); | |
1110 | root = list_entry(next, struct btrfs_root, dirty_list); | |
e7070be1 | 1111 | clear_bit(BTRFS_ROOT_DIRTY, &root->state); |
87ef2bb4 | 1112 | |
9e351cc8 JB |
1113 | if (root != fs_info->extent_root) |
1114 | list_add_tail(&root->dirty_list, | |
1115 | &trans->transaction->switch_commits); | |
49b25e05 JM |
1116 | ret = update_cowonly_root(trans, root); |
1117 | if (ret) | |
1118 | return ret; | |
ea526d18 JB |
1119 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
1120 | if (ret) | |
1121 | return ret; | |
79154b1b | 1122 | } |
276e680d | 1123 | |
1bbc621e | 1124 | while (!list_empty(dirty_bgs) || !list_empty(io_bgs)) { |
ea526d18 JB |
1125 | ret = btrfs_write_dirty_block_groups(trans, root); |
1126 | if (ret) | |
1127 | return ret; | |
1128 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); | |
1129 | if (ret) | |
1130 | return ret; | |
1131 | } | |
1132 | ||
1133 | if (!list_empty(&fs_info->dirty_cowonly_roots)) | |
1134 | goto again; | |
1135 | ||
9e351cc8 JB |
1136 | list_add_tail(&fs_info->extent_root->dirty_list, |
1137 | &trans->transaction->switch_commits); | |
8dabb742 SB |
1138 | btrfs_after_dev_replace_commit(fs_info); |
1139 | ||
79154b1b CM |
1140 | return 0; |
1141 | } | |
1142 | ||
d352ac68 CM |
1143 | /* |
1144 | * dead roots are old snapshots that need to be deleted. This allocates | |
1145 | * a dirty root struct and adds it into the list of dead roots that need to | |
1146 | * be deleted | |
1147 | */ | |
cfad392b | 1148 | void btrfs_add_dead_root(struct btrfs_root *root) |
5eda7b5e | 1149 | { |
a4abeea4 | 1150 | spin_lock(&root->fs_info->trans_lock); |
cfad392b JB |
1151 | if (list_empty(&root->root_list)) |
1152 | list_add_tail(&root->root_list, &root->fs_info->dead_roots); | |
a4abeea4 | 1153 | spin_unlock(&root->fs_info->trans_lock); |
5eda7b5e CM |
1154 | } |
1155 | ||
d352ac68 | 1156 | /* |
5d4f98a2 | 1157 | * update all the cowonly tree roots on disk |
d352ac68 | 1158 | */ |
5d4f98a2 YZ |
1159 | static noinline int commit_fs_roots(struct btrfs_trans_handle *trans, |
1160 | struct btrfs_root *root) | |
0f7d52f4 | 1161 | { |
0f7d52f4 | 1162 | struct btrfs_root *gang[8]; |
5d4f98a2 | 1163 | struct btrfs_fs_info *fs_info = root->fs_info; |
0f7d52f4 CM |
1164 | int i; |
1165 | int ret; | |
54aa1f4d CM |
1166 | int err = 0; |
1167 | ||
a4abeea4 | 1168 | spin_lock(&fs_info->fs_roots_radix_lock); |
d397712b | 1169 | while (1) { |
5d4f98a2 YZ |
1170 | ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, |
1171 | (void **)gang, 0, | |
0f7d52f4 CM |
1172 | ARRAY_SIZE(gang), |
1173 | BTRFS_ROOT_TRANS_TAG); | |
1174 | if (ret == 0) | |
1175 | break; | |
1176 | for (i = 0; i < ret; i++) { | |
1177 | root = gang[i]; | |
5d4f98a2 YZ |
1178 | radix_tree_tag_clear(&fs_info->fs_roots_radix, |
1179 | (unsigned long)root->root_key.objectid, | |
1180 | BTRFS_ROOT_TRANS_TAG); | |
a4abeea4 | 1181 | spin_unlock(&fs_info->fs_roots_radix_lock); |
31153d81 | 1182 | |
e02119d5 | 1183 | btrfs_free_log(trans, root); |
5d4f98a2 | 1184 | btrfs_update_reloc_root(trans, root); |
d68fc57b | 1185 | btrfs_orphan_commit_root(trans, root); |
bcc63abb | 1186 | |
82d5902d LZ |
1187 | btrfs_save_ino_cache(root, trans); |
1188 | ||
f1ebcc74 | 1189 | /* see comments in should_cow_block() */ |
27cdeb70 | 1190 | clear_bit(BTRFS_ROOT_FORCE_COW, &root->state); |
c7548af6 | 1191 | smp_mb__after_atomic(); |
f1ebcc74 | 1192 | |
978d910d | 1193 | if (root->commit_root != root->node) { |
9e351cc8 JB |
1194 | list_add_tail(&root->dirty_list, |
1195 | &trans->transaction->switch_commits); | |
978d910d YZ |
1196 | btrfs_set_root_node(&root->root_item, |
1197 | root->node); | |
1198 | } | |
5d4f98a2 | 1199 | |
5d4f98a2 | 1200 | err = btrfs_update_root(trans, fs_info->tree_root, |
0f7d52f4 CM |
1201 | &root->root_key, |
1202 | &root->root_item); | |
a4abeea4 | 1203 | spin_lock(&fs_info->fs_roots_radix_lock); |
54aa1f4d CM |
1204 | if (err) |
1205 | break; | |
0f7d52f4 CM |
1206 | } |
1207 | } | |
a4abeea4 | 1208 | spin_unlock(&fs_info->fs_roots_radix_lock); |
54aa1f4d | 1209 | return err; |
0f7d52f4 CM |
1210 | } |
1211 | ||
d352ac68 | 1212 | /* |
de78b51a ES |
1213 | * defrag a given btree. |
1214 | * Every leaf in the btree is read and defragged. | |
d352ac68 | 1215 | */ |
de78b51a | 1216 | int btrfs_defrag_root(struct btrfs_root *root) |
e9d0b13b CM |
1217 | { |
1218 | struct btrfs_fs_info *info = root->fs_info; | |
e9d0b13b | 1219 | struct btrfs_trans_handle *trans; |
8929ecfa | 1220 | int ret; |
e9d0b13b | 1221 | |
27cdeb70 | 1222 | if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state)) |
e9d0b13b | 1223 | return 0; |
8929ecfa | 1224 | |
6b80053d | 1225 | while (1) { |
8929ecfa YZ |
1226 | trans = btrfs_start_transaction(root, 0); |
1227 | if (IS_ERR(trans)) | |
1228 | return PTR_ERR(trans); | |
1229 | ||
de78b51a | 1230 | ret = btrfs_defrag_leaves(trans, root); |
8929ecfa | 1231 | |
e9d0b13b | 1232 | btrfs_end_transaction(trans, root); |
b53d3f5d | 1233 | btrfs_btree_balance_dirty(info->tree_root); |
e9d0b13b CM |
1234 | cond_resched(); |
1235 | ||
7841cb28 | 1236 | if (btrfs_fs_closing(root->fs_info) || ret != -EAGAIN) |
e9d0b13b | 1237 | break; |
210549eb DS |
1238 | |
1239 | if (btrfs_defrag_cancelled(root->fs_info)) { | |
efe120a0 | 1240 | pr_debug("BTRFS: defrag_root cancelled\n"); |
210549eb DS |
1241 | ret = -EAGAIN; |
1242 | break; | |
1243 | } | |
e9d0b13b | 1244 | } |
27cdeb70 | 1245 | clear_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state); |
8929ecfa | 1246 | return ret; |
e9d0b13b CM |
1247 | } |
1248 | ||
d352ac68 CM |
1249 | /* |
1250 | * new snapshots need to be created at a very specific time in the | |
aec8030a MX |
1251 | * transaction commit. This does the actual creation. |
1252 | * | |
1253 | * Note: | |
1254 | * If the error which may affect the commitment of the current transaction | |
1255 | * happens, we should return the error number. If the error which just affect | |
1256 | * the creation of the pending snapshots, just return 0. | |
d352ac68 | 1257 | */ |
80b6794d | 1258 | static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, |
3063d29f CM |
1259 | struct btrfs_fs_info *fs_info, |
1260 | struct btrfs_pending_snapshot *pending) | |
1261 | { | |
1262 | struct btrfs_key key; | |
80b6794d | 1263 | struct btrfs_root_item *new_root_item; |
3063d29f CM |
1264 | struct btrfs_root *tree_root = fs_info->tree_root; |
1265 | struct btrfs_root *root = pending->root; | |
6bdb72de | 1266 | struct btrfs_root *parent_root; |
98c9942a | 1267 | struct btrfs_block_rsv *rsv; |
6bdb72de | 1268 | struct inode *parent_inode; |
42874b3d MX |
1269 | struct btrfs_path *path; |
1270 | struct btrfs_dir_item *dir_item; | |
a22285a6 | 1271 | struct dentry *dentry; |
3063d29f | 1272 | struct extent_buffer *tmp; |
925baedd | 1273 | struct extent_buffer *old; |
8ea05e3a | 1274 | struct timespec cur_time = CURRENT_TIME; |
aec8030a | 1275 | int ret = 0; |
d68fc57b | 1276 | u64 to_reserve = 0; |
6bdb72de | 1277 | u64 index = 0; |
a22285a6 | 1278 | u64 objectid; |
b83cc969 | 1279 | u64 root_flags; |
8ea05e3a | 1280 | uuid_le new_uuid; |
3063d29f | 1281 | |
42874b3d MX |
1282 | path = btrfs_alloc_path(); |
1283 | if (!path) { | |
aec8030a MX |
1284 | pending->error = -ENOMEM; |
1285 | return 0; | |
42874b3d MX |
1286 | } |
1287 | ||
80b6794d CM |
1288 | new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS); |
1289 | if (!new_root_item) { | |
aec8030a | 1290 | pending->error = -ENOMEM; |
6fa9700e | 1291 | goto root_item_alloc_fail; |
80b6794d | 1292 | } |
a22285a6 | 1293 | |
aec8030a MX |
1294 | pending->error = btrfs_find_free_objectid(tree_root, &objectid); |
1295 | if (pending->error) | |
6fa9700e | 1296 | goto no_free_objectid; |
3063d29f | 1297 | |
d6726335 QW |
1298 | /* |
1299 | * Make qgroup to skip current new snapshot's qgroupid, as it is | |
1300 | * accounted by later btrfs_qgroup_inherit(). | |
1301 | */ | |
1302 | btrfs_set_skip_qgroup(trans, objectid); | |
1303 | ||
3fd0a558 | 1304 | btrfs_reloc_pre_snapshot(trans, pending, &to_reserve); |
d68fc57b YZ |
1305 | |
1306 | if (to_reserve > 0) { | |
aec8030a MX |
1307 | pending->error = btrfs_block_rsv_add(root, |
1308 | &pending->block_rsv, | |
1309 | to_reserve, | |
1310 | BTRFS_RESERVE_NO_FLUSH); | |
1311 | if (pending->error) | |
d6726335 | 1312 | goto clear_skip_qgroup; |
d68fc57b YZ |
1313 | } |
1314 | ||
3063d29f | 1315 | key.objectid = objectid; |
a22285a6 YZ |
1316 | key.offset = (u64)-1; |
1317 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3063d29f | 1318 | |
6fa9700e | 1319 | rsv = trans->block_rsv; |
a22285a6 | 1320 | trans->block_rsv = &pending->block_rsv; |
2382c5cc | 1321 | trans->bytes_reserved = trans->block_rsv->reserved; |
3de4586c | 1322 | |
a22285a6 | 1323 | dentry = pending->dentry; |
e9662f70 | 1324 | parent_inode = pending->dir; |
a22285a6 | 1325 | parent_root = BTRFS_I(parent_inode)->root; |
7585717f | 1326 | record_root_in_trans(trans, parent_root); |
a22285a6 | 1327 | |
3063d29f CM |
1328 | /* |
1329 | * insert the directory item | |
1330 | */ | |
3de4586c | 1331 | ret = btrfs_set_inode_index(parent_inode, &index); |
49b25e05 | 1332 | BUG_ON(ret); /* -ENOMEM */ |
42874b3d MX |
1333 | |
1334 | /* check if there is a file/dir which has the same name. */ | |
1335 | dir_item = btrfs_lookup_dir_item(NULL, parent_root, path, | |
1336 | btrfs_ino(parent_inode), | |
1337 | dentry->d_name.name, | |
1338 | dentry->d_name.len, 0); | |
1339 | if (dir_item != NULL && !IS_ERR(dir_item)) { | |
fe66a05a | 1340 | pending->error = -EEXIST; |
aec8030a | 1341 | goto dir_item_existed; |
42874b3d MX |
1342 | } else if (IS_ERR(dir_item)) { |
1343 | ret = PTR_ERR(dir_item); | |
8732d44f MX |
1344 | btrfs_abort_transaction(trans, root, ret); |
1345 | goto fail; | |
79787eaa | 1346 | } |
42874b3d | 1347 | btrfs_release_path(path); |
52c26179 | 1348 | |
e999376f CM |
1349 | /* |
1350 | * pull in the delayed directory update | |
1351 | * and the delayed inode item | |
1352 | * otherwise we corrupt the FS during | |
1353 | * snapshot | |
1354 | */ | |
1355 | ret = btrfs_run_delayed_items(trans, root); | |
8732d44f MX |
1356 | if (ret) { /* Transaction aborted */ |
1357 | btrfs_abort_transaction(trans, root, ret); | |
1358 | goto fail; | |
1359 | } | |
e999376f | 1360 | |
7585717f | 1361 | record_root_in_trans(trans, root); |
6bdb72de SW |
1362 | btrfs_set_root_last_snapshot(&root->root_item, trans->transid); |
1363 | memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); | |
08fe4db1 | 1364 | btrfs_check_and_init_root_item(new_root_item); |
6bdb72de | 1365 | |
b83cc969 LZ |
1366 | root_flags = btrfs_root_flags(new_root_item); |
1367 | if (pending->readonly) | |
1368 | root_flags |= BTRFS_ROOT_SUBVOL_RDONLY; | |
1369 | else | |
1370 | root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY; | |
1371 | btrfs_set_root_flags(new_root_item, root_flags); | |
1372 | ||
8ea05e3a AB |
1373 | btrfs_set_root_generation_v2(new_root_item, |
1374 | trans->transid); | |
1375 | uuid_le_gen(&new_uuid); | |
1376 | memcpy(new_root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE); | |
1377 | memcpy(new_root_item->parent_uuid, root->root_item.uuid, | |
1378 | BTRFS_UUID_SIZE); | |
70023da2 SB |
1379 | if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) { |
1380 | memset(new_root_item->received_uuid, 0, | |
1381 | sizeof(new_root_item->received_uuid)); | |
1382 | memset(&new_root_item->stime, 0, sizeof(new_root_item->stime)); | |
1383 | memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime)); | |
1384 | btrfs_set_root_stransid(new_root_item, 0); | |
1385 | btrfs_set_root_rtransid(new_root_item, 0); | |
1386 | } | |
3cae210f QW |
1387 | btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec); |
1388 | btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec); | |
8ea05e3a | 1389 | btrfs_set_root_otransid(new_root_item, trans->transid); |
8ea05e3a | 1390 | |
6bdb72de | 1391 | old = btrfs_lock_root_node(root); |
49b25e05 | 1392 | ret = btrfs_cow_block(trans, root, old, NULL, 0, &old); |
79787eaa JM |
1393 | if (ret) { |
1394 | btrfs_tree_unlock(old); | |
1395 | free_extent_buffer(old); | |
8732d44f MX |
1396 | btrfs_abort_transaction(trans, root, ret); |
1397 | goto fail; | |
79787eaa | 1398 | } |
49b25e05 | 1399 | |
6bdb72de SW |
1400 | btrfs_set_lock_blocking(old); |
1401 | ||
49b25e05 | 1402 | ret = btrfs_copy_root(trans, root, old, &tmp, objectid); |
79787eaa | 1403 | /* clean up in any case */ |
6bdb72de SW |
1404 | btrfs_tree_unlock(old); |
1405 | free_extent_buffer(old); | |
8732d44f MX |
1406 | if (ret) { |
1407 | btrfs_abort_transaction(trans, root, ret); | |
1408 | goto fail; | |
1409 | } | |
f1ebcc74 | 1410 | /* see comments in should_cow_block() */ |
27cdeb70 | 1411 | set_bit(BTRFS_ROOT_FORCE_COW, &root->state); |
f1ebcc74 LB |
1412 | smp_wmb(); |
1413 | ||
6bdb72de | 1414 | btrfs_set_root_node(new_root_item, tmp); |
a22285a6 YZ |
1415 | /* record when the snapshot was created in key.offset */ |
1416 | key.offset = trans->transid; | |
1417 | ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); | |
6bdb72de SW |
1418 | btrfs_tree_unlock(tmp); |
1419 | free_extent_buffer(tmp); | |
8732d44f MX |
1420 | if (ret) { |
1421 | btrfs_abort_transaction(trans, root, ret); | |
1422 | goto fail; | |
1423 | } | |
6bdb72de | 1424 | |
a22285a6 YZ |
1425 | /* |
1426 | * insert root back/forward references | |
1427 | */ | |
1428 | ret = btrfs_add_root_ref(trans, tree_root, objectid, | |
0660b5af | 1429 | parent_root->root_key.objectid, |
33345d01 | 1430 | btrfs_ino(parent_inode), index, |
a22285a6 | 1431 | dentry->d_name.name, dentry->d_name.len); |
8732d44f MX |
1432 | if (ret) { |
1433 | btrfs_abort_transaction(trans, root, ret); | |
1434 | goto fail; | |
1435 | } | |
0660b5af | 1436 | |
a22285a6 YZ |
1437 | key.offset = (u64)-1; |
1438 | pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key); | |
79787eaa JM |
1439 | if (IS_ERR(pending->snap)) { |
1440 | ret = PTR_ERR(pending->snap); | |
8732d44f MX |
1441 | btrfs_abort_transaction(trans, root, ret); |
1442 | goto fail; | |
79787eaa | 1443 | } |
d68fc57b | 1444 | |
49b25e05 | 1445 | ret = btrfs_reloc_post_snapshot(trans, pending); |
8732d44f MX |
1446 | if (ret) { |
1447 | btrfs_abort_transaction(trans, root, ret); | |
1448 | goto fail; | |
1449 | } | |
361048f5 MX |
1450 | |
1451 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); | |
8732d44f MX |
1452 | if (ret) { |
1453 | btrfs_abort_transaction(trans, root, ret); | |
1454 | goto fail; | |
1455 | } | |
42874b3d MX |
1456 | |
1457 | ret = btrfs_insert_dir_item(trans, parent_root, | |
1458 | dentry->d_name.name, dentry->d_name.len, | |
1459 | parent_inode, &key, | |
1460 | BTRFS_FT_DIR, index); | |
1461 | /* We have check then name at the beginning, so it is impossible. */ | |
9c52057c | 1462 | BUG_ON(ret == -EEXIST || ret == -EOVERFLOW); |
8732d44f MX |
1463 | if (ret) { |
1464 | btrfs_abort_transaction(trans, root, ret); | |
1465 | goto fail; | |
1466 | } | |
42874b3d MX |
1467 | |
1468 | btrfs_i_size_write(parent_inode, parent_inode->i_size + | |
1469 | dentry->d_name.len * 2); | |
1470 | parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME; | |
be6aef60 | 1471 | ret = btrfs_update_inode_fallback(trans, parent_root, parent_inode); |
dd5f9615 SB |
1472 | if (ret) { |
1473 | btrfs_abort_transaction(trans, root, ret); | |
1474 | goto fail; | |
1475 | } | |
1476 | ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root, new_uuid.b, | |
1477 | BTRFS_UUID_KEY_SUBVOL, objectid); | |
1478 | if (ret) { | |
8732d44f | 1479 | btrfs_abort_transaction(trans, root, ret); |
dd5f9615 SB |
1480 | goto fail; |
1481 | } | |
1482 | if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) { | |
1483 | ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root, | |
1484 | new_root_item->received_uuid, | |
1485 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, | |
1486 | objectid); | |
1487 | if (ret && ret != -EEXIST) { | |
1488 | btrfs_abort_transaction(trans, root, ret); | |
1489 | goto fail; | |
1490 | } | |
1491 | } | |
d6726335 QW |
1492 | |
1493 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); | |
1494 | if (ret) { | |
1495 | btrfs_abort_transaction(trans, root, ret); | |
1496 | goto fail; | |
1497 | } | |
1498 | ||
1499 | /* | |
1500 | * account qgroup counters before qgroup_inherit() | |
1501 | */ | |
1502 | ret = btrfs_qgroup_prepare_account_extents(trans, fs_info); | |
1503 | if (ret) | |
1504 | goto fail; | |
1505 | ret = btrfs_qgroup_account_extents(trans, fs_info); | |
1506 | if (ret) | |
1507 | goto fail; | |
1508 | ret = btrfs_qgroup_inherit(trans, fs_info, | |
1509 | root->root_key.objectid, | |
1510 | objectid, pending->inherit); | |
1511 | if (ret) { | |
1512 | btrfs_abort_transaction(trans, root, ret); | |
1513 | goto fail; | |
1514 | } | |
1515 | ||
3063d29f | 1516 | fail: |
aec8030a MX |
1517 | pending->error = ret; |
1518 | dir_item_existed: | |
98c9942a | 1519 | trans->block_rsv = rsv; |
2382c5cc | 1520 | trans->bytes_reserved = 0; |
d6726335 QW |
1521 | clear_skip_qgroup: |
1522 | btrfs_clear_skip_qgroup(trans); | |
6fa9700e MX |
1523 | no_free_objectid: |
1524 | kfree(new_root_item); | |
1525 | root_item_alloc_fail: | |
42874b3d | 1526 | btrfs_free_path(path); |
49b25e05 | 1527 | return ret; |
3063d29f CM |
1528 | } |
1529 | ||
d352ac68 CM |
1530 | /* |
1531 | * create all the snapshots we've scheduled for creation | |
1532 | */ | |
80b6794d CM |
1533 | static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans, |
1534 | struct btrfs_fs_info *fs_info) | |
3de4586c | 1535 | { |
aec8030a | 1536 | struct btrfs_pending_snapshot *pending, *next; |
3de4586c | 1537 | struct list_head *head = &trans->transaction->pending_snapshots; |
aec8030a | 1538 | int ret = 0; |
3de4586c | 1539 | |
aec8030a MX |
1540 | list_for_each_entry_safe(pending, next, head, list) { |
1541 | list_del(&pending->list); | |
1542 | ret = create_pending_snapshot(trans, fs_info, pending); | |
1543 | if (ret) | |
1544 | break; | |
1545 | } | |
1546 | return ret; | |
3de4586c CM |
1547 | } |
1548 | ||
5d4f98a2 YZ |
1549 | static void update_super_roots(struct btrfs_root *root) |
1550 | { | |
1551 | struct btrfs_root_item *root_item; | |
1552 | struct btrfs_super_block *super; | |
1553 | ||
6c41761f | 1554 | super = root->fs_info->super_copy; |
5d4f98a2 YZ |
1555 | |
1556 | root_item = &root->fs_info->chunk_root->root_item; | |
1557 | super->chunk_root = root_item->bytenr; | |
1558 | super->chunk_root_generation = root_item->generation; | |
1559 | super->chunk_root_level = root_item->level; | |
1560 | ||
1561 | root_item = &root->fs_info->tree_root->root_item; | |
1562 | super->root = root_item->bytenr; | |
1563 | super->generation = root_item->generation; | |
1564 | super->root_level = root_item->level; | |
73bc1876 | 1565 | if (btrfs_test_opt(root, SPACE_CACHE)) |
0af3d00b | 1566 | super->cache_generation = root_item->generation; |
70f80175 SB |
1567 | if (root->fs_info->update_uuid_tree_gen) |
1568 | super->uuid_tree_generation = root_item->generation; | |
5d4f98a2 YZ |
1569 | } |
1570 | ||
f36f3042 CM |
1571 | int btrfs_transaction_in_commit(struct btrfs_fs_info *info) |
1572 | { | |
4a9d8bde | 1573 | struct btrfs_transaction *trans; |
f36f3042 | 1574 | int ret = 0; |
4a9d8bde | 1575 | |
a4abeea4 | 1576 | spin_lock(&info->trans_lock); |
4a9d8bde MX |
1577 | trans = info->running_transaction; |
1578 | if (trans) | |
1579 | ret = (trans->state >= TRANS_STATE_COMMIT_START); | |
a4abeea4 | 1580 | spin_unlock(&info->trans_lock); |
f36f3042 CM |
1581 | return ret; |
1582 | } | |
1583 | ||
8929ecfa YZ |
1584 | int btrfs_transaction_blocked(struct btrfs_fs_info *info) |
1585 | { | |
4a9d8bde | 1586 | struct btrfs_transaction *trans; |
8929ecfa | 1587 | int ret = 0; |
4a9d8bde | 1588 | |
a4abeea4 | 1589 | spin_lock(&info->trans_lock); |
4a9d8bde MX |
1590 | trans = info->running_transaction; |
1591 | if (trans) | |
1592 | ret = is_transaction_blocked(trans); | |
a4abeea4 | 1593 | spin_unlock(&info->trans_lock); |
8929ecfa YZ |
1594 | return ret; |
1595 | } | |
1596 | ||
bb9c12c9 SW |
1597 | /* |
1598 | * wait for the current transaction commit to start and block subsequent | |
1599 | * transaction joins | |
1600 | */ | |
1601 | static void wait_current_trans_commit_start(struct btrfs_root *root, | |
1602 | struct btrfs_transaction *trans) | |
1603 | { | |
4a9d8bde | 1604 | wait_event(root->fs_info->transaction_blocked_wait, |
501407aa JB |
1605 | trans->state >= TRANS_STATE_COMMIT_START || |
1606 | trans->aborted); | |
bb9c12c9 SW |
1607 | } |
1608 | ||
1609 | /* | |
1610 | * wait for the current transaction to start and then become unblocked. | |
1611 | * caller holds ref. | |
1612 | */ | |
1613 | static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root, | |
1614 | struct btrfs_transaction *trans) | |
1615 | { | |
72d63ed6 | 1616 | wait_event(root->fs_info->transaction_wait, |
501407aa JB |
1617 | trans->state >= TRANS_STATE_UNBLOCKED || |
1618 | trans->aborted); | |
bb9c12c9 SW |
1619 | } |
1620 | ||
1621 | /* | |
1622 | * commit transactions asynchronously. once btrfs_commit_transaction_async | |
1623 | * returns, any subsequent transaction will not be allowed to join. | |
1624 | */ | |
1625 | struct btrfs_async_commit { | |
1626 | struct btrfs_trans_handle *newtrans; | |
1627 | struct btrfs_root *root; | |
7892b5af | 1628 | struct work_struct work; |
bb9c12c9 SW |
1629 | }; |
1630 | ||
1631 | static void do_async_commit(struct work_struct *work) | |
1632 | { | |
1633 | struct btrfs_async_commit *ac = | |
7892b5af | 1634 | container_of(work, struct btrfs_async_commit, work); |
bb9c12c9 | 1635 | |
6fc4e354 SW |
1636 | /* |
1637 | * We've got freeze protection passed with the transaction. | |
1638 | * Tell lockdep about it. | |
1639 | */ | |
b1a06a4b | 1640 | if (ac->newtrans->type & __TRANS_FREEZABLE) |
ff7c1d33 MX |
1641 | rwsem_acquire_read( |
1642 | &ac->root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1], | |
1643 | 0, 1, _THIS_IP_); | |
6fc4e354 | 1644 | |
e209db7a SW |
1645 | current->journal_info = ac->newtrans; |
1646 | ||
bb9c12c9 SW |
1647 | btrfs_commit_transaction(ac->newtrans, ac->root); |
1648 | kfree(ac); | |
1649 | } | |
1650 | ||
1651 | int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans, | |
1652 | struct btrfs_root *root, | |
1653 | int wait_for_unblock) | |
1654 | { | |
1655 | struct btrfs_async_commit *ac; | |
1656 | struct btrfs_transaction *cur_trans; | |
1657 | ||
1658 | ac = kmalloc(sizeof(*ac), GFP_NOFS); | |
db5b493a TI |
1659 | if (!ac) |
1660 | return -ENOMEM; | |
bb9c12c9 | 1661 | |
7892b5af | 1662 | INIT_WORK(&ac->work, do_async_commit); |
bb9c12c9 | 1663 | ac->root = root; |
7a7eaa40 | 1664 | ac->newtrans = btrfs_join_transaction(root); |
3612b495 TI |
1665 | if (IS_ERR(ac->newtrans)) { |
1666 | int err = PTR_ERR(ac->newtrans); | |
1667 | kfree(ac); | |
1668 | return err; | |
1669 | } | |
bb9c12c9 SW |
1670 | |
1671 | /* take transaction reference */ | |
bb9c12c9 | 1672 | cur_trans = trans->transaction; |
13c5a93e | 1673 | atomic_inc(&cur_trans->use_count); |
bb9c12c9 SW |
1674 | |
1675 | btrfs_end_transaction(trans, root); | |
6fc4e354 SW |
1676 | |
1677 | /* | |
1678 | * Tell lockdep we've released the freeze rwsem, since the | |
1679 | * async commit thread will be the one to unlock it. | |
1680 | */ | |
b1a06a4b | 1681 | if (ac->newtrans->type & __TRANS_FREEZABLE) |
ff7c1d33 MX |
1682 | rwsem_release( |
1683 | &root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1], | |
1684 | 1, _THIS_IP_); | |
6fc4e354 | 1685 | |
7892b5af | 1686 | schedule_work(&ac->work); |
bb9c12c9 SW |
1687 | |
1688 | /* wait for transaction to start and unblock */ | |
bb9c12c9 SW |
1689 | if (wait_for_unblock) |
1690 | wait_current_trans_commit_start_and_unblock(root, cur_trans); | |
1691 | else | |
1692 | wait_current_trans_commit_start(root, cur_trans); | |
bb9c12c9 | 1693 | |
38e88054 SW |
1694 | if (current->journal_info == trans) |
1695 | current->journal_info = NULL; | |
1696 | ||
724e2315 | 1697 | btrfs_put_transaction(cur_trans); |
bb9c12c9 SW |
1698 | return 0; |
1699 | } | |
1700 | ||
49b25e05 JM |
1701 | |
1702 | static void cleanup_transaction(struct btrfs_trans_handle *trans, | |
7b8b92af | 1703 | struct btrfs_root *root, int err) |
49b25e05 JM |
1704 | { |
1705 | struct btrfs_transaction *cur_trans = trans->transaction; | |
f094ac32 | 1706 | DEFINE_WAIT(wait); |
49b25e05 JM |
1707 | |
1708 | WARN_ON(trans->use_count > 1); | |
1709 | ||
7b8b92af JB |
1710 | btrfs_abort_transaction(trans, root, err); |
1711 | ||
49b25e05 | 1712 | spin_lock(&root->fs_info->trans_lock); |
66b6135b | 1713 | |
25d8c284 MX |
1714 | /* |
1715 | * If the transaction is removed from the list, it means this | |
1716 | * transaction has been committed successfully, so it is impossible | |
1717 | * to call the cleanup function. | |
1718 | */ | |
1719 | BUG_ON(list_empty(&cur_trans->list)); | |
66b6135b | 1720 | |
49b25e05 | 1721 | list_del_init(&cur_trans->list); |
d7096fc3 | 1722 | if (cur_trans == root->fs_info->running_transaction) { |
4a9d8bde | 1723 | cur_trans->state = TRANS_STATE_COMMIT_DOING; |
f094ac32 LB |
1724 | spin_unlock(&root->fs_info->trans_lock); |
1725 | wait_event(cur_trans->writer_wait, | |
1726 | atomic_read(&cur_trans->num_writers) == 1); | |
1727 | ||
1728 | spin_lock(&root->fs_info->trans_lock); | |
d7096fc3 | 1729 | } |
49b25e05 JM |
1730 | spin_unlock(&root->fs_info->trans_lock); |
1731 | ||
1732 | btrfs_cleanup_one_transaction(trans->transaction, root); | |
1733 | ||
4a9d8bde MX |
1734 | spin_lock(&root->fs_info->trans_lock); |
1735 | if (cur_trans == root->fs_info->running_transaction) | |
1736 | root->fs_info->running_transaction = NULL; | |
1737 | spin_unlock(&root->fs_info->trans_lock); | |
1738 | ||
e0228285 JB |
1739 | if (trans->type & __TRANS_FREEZABLE) |
1740 | sb_end_intwrite(root->fs_info->sb); | |
724e2315 JB |
1741 | btrfs_put_transaction(cur_trans); |
1742 | btrfs_put_transaction(cur_trans); | |
49b25e05 JM |
1743 | |
1744 | trace_btrfs_transaction_commit(root); | |
1745 | ||
49b25e05 JM |
1746 | if (current->journal_info == trans) |
1747 | current->journal_info = NULL; | |
c0af8f0b | 1748 | btrfs_scrub_cancel(root->fs_info); |
49b25e05 JM |
1749 | |
1750 | kmem_cache_free(btrfs_trans_handle_cachep, trans); | |
1751 | } | |
1752 | ||
82436617 MX |
1753 | static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info) |
1754 | { | |
1755 | if (btrfs_test_opt(fs_info->tree_root, FLUSHONCOMMIT)) | |
6c255e67 | 1756 | return btrfs_start_delalloc_roots(fs_info, 1, -1); |
82436617 MX |
1757 | return 0; |
1758 | } | |
1759 | ||
1760 | static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info) | |
1761 | { | |
1762 | if (btrfs_test_opt(fs_info->tree_root, FLUSHONCOMMIT)) | |
b0244199 | 1763 | btrfs_wait_ordered_roots(fs_info, -1); |
82436617 MX |
1764 | } |
1765 | ||
50d9aa99 JB |
1766 | static inline void |
1767 | btrfs_wait_pending_ordered(struct btrfs_transaction *cur_trans, | |
1768 | struct btrfs_fs_info *fs_info) | |
1769 | { | |
1770 | struct btrfs_ordered_extent *ordered; | |
1771 | ||
1772 | spin_lock(&fs_info->trans_lock); | |
1773 | while (!list_empty(&cur_trans->pending_ordered)) { | |
1774 | ordered = list_first_entry(&cur_trans->pending_ordered, | |
1775 | struct btrfs_ordered_extent, | |
1776 | trans_list); | |
1777 | list_del_init(&ordered->trans_list); | |
1778 | spin_unlock(&fs_info->trans_lock); | |
1779 | ||
1780 | wait_event(ordered->wait, test_bit(BTRFS_ORDERED_COMPLETE, | |
1781 | &ordered->flags)); | |
1782 | btrfs_put_ordered_extent(ordered); | |
1783 | spin_lock(&fs_info->trans_lock); | |
1784 | } | |
1785 | spin_unlock(&fs_info->trans_lock); | |
1786 | } | |
1787 | ||
79154b1b CM |
1788 | int btrfs_commit_transaction(struct btrfs_trans_handle *trans, |
1789 | struct btrfs_root *root) | |
1790 | { | |
49b25e05 | 1791 | struct btrfs_transaction *cur_trans = trans->transaction; |
8fd17795 | 1792 | struct btrfs_transaction *prev_trans = NULL; |
656f30db | 1793 | struct btrfs_inode *btree_ino = BTRFS_I(root->fs_info->btree_inode); |
25287e0a | 1794 | int ret; |
79154b1b | 1795 | |
8d25a086 MX |
1796 | /* Stop the commit early if ->aborted is set */ |
1797 | if (unlikely(ACCESS_ONCE(cur_trans->aborted))) { | |
25287e0a | 1798 | ret = cur_trans->aborted; |
e4a2bcac JB |
1799 | btrfs_end_transaction(trans, root); |
1800 | return ret; | |
25287e0a | 1801 | } |
49b25e05 | 1802 | |
56bec294 CM |
1803 | /* make a pass through all the delayed refs we have so far |
1804 | * any runnings procs may add more while we are here | |
1805 | */ | |
1806 | ret = btrfs_run_delayed_refs(trans, root, 0); | |
e4a2bcac JB |
1807 | if (ret) { |
1808 | btrfs_end_transaction(trans, root); | |
1809 | return ret; | |
1810 | } | |
56bec294 | 1811 | |
0e721106 JB |
1812 | btrfs_trans_release_metadata(trans, root); |
1813 | trans->block_rsv = NULL; | |
272d26d0 MX |
1814 | if (trans->qgroup_reserved) { |
1815 | btrfs_qgroup_free(root, trans->qgroup_reserved); | |
1816 | trans->qgroup_reserved = 0; | |
1817 | } | |
0e721106 | 1818 | |
b7ec40d7 | 1819 | cur_trans = trans->transaction; |
49b25e05 | 1820 | |
56bec294 CM |
1821 | /* |
1822 | * set the flushing flag so procs in this transaction have to | |
1823 | * start sending their work down. | |
1824 | */ | |
b7ec40d7 | 1825 | cur_trans->delayed_refs.flushing = 1; |
1be41b78 | 1826 | smp_wmb(); |
56bec294 | 1827 | |
ea658bad JB |
1828 | if (!list_empty(&trans->new_bgs)) |
1829 | btrfs_create_pending_block_groups(trans, root); | |
1830 | ||
c3e69d58 | 1831 | ret = btrfs_run_delayed_refs(trans, root, 0); |
e4a2bcac JB |
1832 | if (ret) { |
1833 | btrfs_end_transaction(trans, root); | |
1834 | return ret; | |
1835 | } | |
56bec294 | 1836 | |
1bbc621e CM |
1837 | if (!cur_trans->dirty_bg_run) { |
1838 | int run_it = 0; | |
1839 | ||
1840 | /* this mutex is also taken before trying to set | |
1841 | * block groups readonly. We need to make sure | |
1842 | * that nobody has set a block group readonly | |
1843 | * after a extents from that block group have been | |
1844 | * allocated for cache files. btrfs_set_block_group_ro | |
1845 | * will wait for the transaction to commit if it | |
1846 | * finds dirty_bg_run = 1 | |
1847 | * | |
1848 | * The dirty_bg_run flag is also used to make sure only | |
1849 | * one process starts all the block group IO. It wouldn't | |
1850 | * hurt to have more than one go through, but there's no | |
1851 | * real advantage to it either. | |
1852 | */ | |
1853 | mutex_lock(&root->fs_info->ro_block_group_mutex); | |
1854 | if (!cur_trans->dirty_bg_run) { | |
1855 | run_it = 1; | |
1856 | cur_trans->dirty_bg_run = 1; | |
1857 | } | |
1858 | mutex_unlock(&root->fs_info->ro_block_group_mutex); | |
1859 | ||
1860 | if (run_it) | |
1861 | ret = btrfs_start_dirty_block_groups(trans, root); | |
1862 | } | |
1863 | if (ret) { | |
1864 | btrfs_end_transaction(trans, root); | |
1865 | return ret; | |
1866 | } | |
1867 | ||
4a9d8bde | 1868 | spin_lock(&root->fs_info->trans_lock); |
d3efe084 | 1869 | list_splice_init(&trans->ordered, &cur_trans->pending_ordered); |
4a9d8bde MX |
1870 | if (cur_trans->state >= TRANS_STATE_COMMIT_START) { |
1871 | spin_unlock(&root->fs_info->trans_lock); | |
13c5a93e | 1872 | atomic_inc(&cur_trans->use_count); |
49b25e05 | 1873 | ret = btrfs_end_transaction(trans, root); |
ccd467d6 | 1874 | |
b9c8300c | 1875 | wait_for_commit(root, cur_trans); |
15ee9bc7 | 1876 | |
b4924a0f LB |
1877 | if (unlikely(cur_trans->aborted)) |
1878 | ret = cur_trans->aborted; | |
1879 | ||
724e2315 | 1880 | btrfs_put_transaction(cur_trans); |
15ee9bc7 | 1881 | |
49b25e05 | 1882 | return ret; |
79154b1b | 1883 | } |
4313b399 | 1884 | |
4a9d8bde | 1885 | cur_trans->state = TRANS_STATE_COMMIT_START; |
bb9c12c9 SW |
1886 | wake_up(&root->fs_info->transaction_blocked_wait); |
1887 | ||
ccd467d6 CM |
1888 | if (cur_trans->list.prev != &root->fs_info->trans_list) { |
1889 | prev_trans = list_entry(cur_trans->list.prev, | |
1890 | struct btrfs_transaction, list); | |
4a9d8bde | 1891 | if (prev_trans->state != TRANS_STATE_COMPLETED) { |
13c5a93e | 1892 | atomic_inc(&prev_trans->use_count); |
a4abeea4 | 1893 | spin_unlock(&root->fs_info->trans_lock); |
ccd467d6 CM |
1894 | |
1895 | wait_for_commit(root, prev_trans); | |
ccd467d6 | 1896 | |
724e2315 | 1897 | btrfs_put_transaction(prev_trans); |
a4abeea4 JB |
1898 | } else { |
1899 | spin_unlock(&root->fs_info->trans_lock); | |
ccd467d6 | 1900 | } |
a4abeea4 JB |
1901 | } else { |
1902 | spin_unlock(&root->fs_info->trans_lock); | |
ccd467d6 | 1903 | } |
15ee9bc7 | 1904 | |
0860adfd MX |
1905 | extwriter_counter_dec(cur_trans, trans->type); |
1906 | ||
82436617 MX |
1907 | ret = btrfs_start_delalloc_flush(root->fs_info); |
1908 | if (ret) | |
1909 | goto cleanup_transaction; | |
1910 | ||
8d875f95 | 1911 | ret = btrfs_run_delayed_items(trans, root); |
581227d0 MX |
1912 | if (ret) |
1913 | goto cleanup_transaction; | |
15ee9bc7 | 1914 | |
581227d0 MX |
1915 | wait_event(cur_trans->writer_wait, |
1916 | extwriter_counter_read(cur_trans) == 0); | |
15ee9bc7 | 1917 | |
581227d0 | 1918 | /* some pending stuffs might be added after the previous flush. */ |
8d875f95 | 1919 | ret = btrfs_run_delayed_items(trans, root); |
ca469637 MX |
1920 | if (ret) |
1921 | goto cleanup_transaction; | |
1922 | ||
82436617 | 1923 | btrfs_wait_delalloc_flush(root->fs_info); |
cb7ab021 | 1924 | |
50d9aa99 JB |
1925 | btrfs_wait_pending_ordered(cur_trans, root->fs_info); |
1926 | ||
cb7ab021 | 1927 | btrfs_scrub_pause(root); |
ed0ca140 JB |
1928 | /* |
1929 | * Ok now we need to make sure to block out any other joins while we | |
1930 | * commit the transaction. We could have started a join before setting | |
4a9d8bde | 1931 | * COMMIT_DOING so make sure to wait for num_writers to == 1 again. |
ed0ca140 JB |
1932 | */ |
1933 | spin_lock(&root->fs_info->trans_lock); | |
4a9d8bde | 1934 | cur_trans->state = TRANS_STATE_COMMIT_DOING; |
ed0ca140 JB |
1935 | spin_unlock(&root->fs_info->trans_lock); |
1936 | wait_event(cur_trans->writer_wait, | |
1937 | atomic_read(&cur_trans->num_writers) == 1); | |
1938 | ||
2cba30f1 MX |
1939 | /* ->aborted might be set after the previous check, so check it */ |
1940 | if (unlikely(ACCESS_ONCE(cur_trans->aborted))) { | |
1941 | ret = cur_trans->aborted; | |
6cf7f77e | 1942 | goto scrub_continue; |
2cba30f1 | 1943 | } |
7585717f CM |
1944 | /* |
1945 | * the reloc mutex makes sure that we stop | |
1946 | * the balancing code from coming in and moving | |
1947 | * extents around in the middle of the commit | |
1948 | */ | |
1949 | mutex_lock(&root->fs_info->reloc_mutex); | |
1950 | ||
42874b3d MX |
1951 | /* |
1952 | * We needn't worry about the delayed items because we will | |
1953 | * deal with them in create_pending_snapshot(), which is the | |
1954 | * core function of the snapshot creation. | |
1955 | */ | |
1956 | ret = create_pending_snapshots(trans, root->fs_info); | |
49b25e05 JM |
1957 | if (ret) { |
1958 | mutex_unlock(&root->fs_info->reloc_mutex); | |
6cf7f77e | 1959 | goto scrub_continue; |
49b25e05 | 1960 | } |
3063d29f | 1961 | |
42874b3d MX |
1962 | /* |
1963 | * We insert the dir indexes of the snapshots and update the inode | |
1964 | * of the snapshots' parents after the snapshot creation, so there | |
1965 | * are some delayed items which are not dealt with. Now deal with | |
1966 | * them. | |
1967 | * | |
1968 | * We needn't worry that this operation will corrupt the snapshots, | |
1969 | * because all the tree which are snapshoted will be forced to COW | |
1970 | * the nodes and leaves. | |
1971 | */ | |
1972 | ret = btrfs_run_delayed_items(trans, root); | |
49b25e05 JM |
1973 | if (ret) { |
1974 | mutex_unlock(&root->fs_info->reloc_mutex); | |
6cf7f77e | 1975 | goto scrub_continue; |
49b25e05 | 1976 | } |
16cdcec7 | 1977 | |
56bec294 | 1978 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
49b25e05 JM |
1979 | if (ret) { |
1980 | mutex_unlock(&root->fs_info->reloc_mutex); | |
6cf7f77e | 1981 | goto scrub_continue; |
49b25e05 | 1982 | } |
56bec294 | 1983 | |
0ed4792a QW |
1984 | /* Reocrd old roots for later qgroup accounting */ |
1985 | ret = btrfs_qgroup_prepare_account_extents(trans, root->fs_info); | |
1986 | if (ret) { | |
1987 | mutex_unlock(&root->fs_info->reloc_mutex); | |
1988 | goto scrub_continue; | |
1989 | } | |
1990 | ||
e999376f CM |
1991 | /* |
1992 | * make sure none of the code above managed to slip in a | |
1993 | * delayed item | |
1994 | */ | |
1995 | btrfs_assert_delayed_root_empty(root); | |
1996 | ||
2c90e5d6 | 1997 | WARN_ON(cur_trans != trans->transaction); |
dc17ff8f | 1998 | |
e02119d5 CM |
1999 | /* btrfs_commit_tree_roots is responsible for getting the |
2000 | * various roots consistent with each other. Every pointer | |
2001 | * in the tree of tree roots has to point to the most up to date | |
2002 | * root for every subvolume and other tree. So, we have to keep | |
2003 | * the tree logging code from jumping in and changing any | |
2004 | * of the trees. | |
2005 | * | |
2006 | * At this point in the commit, there can't be any tree-log | |
2007 | * writers, but a little lower down we drop the trans mutex | |
2008 | * and let new people in. By holding the tree_log_mutex | |
2009 | * from now until after the super is written, we avoid races | |
2010 | * with the tree-log code. | |
2011 | */ | |
2012 | mutex_lock(&root->fs_info->tree_log_mutex); | |
2013 | ||
5d4f98a2 | 2014 | ret = commit_fs_roots(trans, root); |
49b25e05 JM |
2015 | if (ret) { |
2016 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
871383be | 2017 | mutex_unlock(&root->fs_info->reloc_mutex); |
6cf7f77e | 2018 | goto scrub_continue; |
49b25e05 | 2019 | } |
54aa1f4d | 2020 | |
3818aea2 | 2021 | /* |
7e1876ac DS |
2022 | * Since the transaction is done, we can apply the pending changes |
2023 | * before the next transaction. | |
3818aea2 | 2024 | */ |
572d9ab7 | 2025 | btrfs_apply_pending_changes(root->fs_info); |
3818aea2 | 2026 | |
5d4f98a2 | 2027 | /* commit_fs_roots gets rid of all the tree log roots, it is now |
e02119d5 CM |
2028 | * safe to free the root of tree log roots |
2029 | */ | |
2030 | btrfs_free_log_root_tree(trans, root->fs_info); | |
2031 | ||
0ed4792a QW |
2032 | /* |
2033 | * Since fs roots are all committed, we can get a quite accurate | |
2034 | * new_roots. So let's do quota accounting. | |
2035 | */ | |
2036 | ret = btrfs_qgroup_account_extents(trans, root->fs_info); | |
2037 | if (ret < 0) { | |
2038 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
2039 | mutex_unlock(&root->fs_info->reloc_mutex); | |
2040 | goto scrub_continue; | |
2041 | } | |
2042 | ||
5d4f98a2 | 2043 | ret = commit_cowonly_roots(trans, root); |
49b25e05 JM |
2044 | if (ret) { |
2045 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
871383be | 2046 | mutex_unlock(&root->fs_info->reloc_mutex); |
6cf7f77e | 2047 | goto scrub_continue; |
49b25e05 | 2048 | } |
54aa1f4d | 2049 | |
2cba30f1 MX |
2050 | /* |
2051 | * The tasks which save the space cache and inode cache may also | |
2052 | * update ->aborted, check it. | |
2053 | */ | |
2054 | if (unlikely(ACCESS_ONCE(cur_trans->aborted))) { | |
2055 | ret = cur_trans->aborted; | |
2056 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
2057 | mutex_unlock(&root->fs_info->reloc_mutex); | |
6cf7f77e | 2058 | goto scrub_continue; |
2cba30f1 MX |
2059 | } |
2060 | ||
11833d66 YZ |
2061 | btrfs_prepare_extent_commit(trans, root); |
2062 | ||
78fae27e | 2063 | cur_trans = root->fs_info->running_transaction; |
5d4f98a2 YZ |
2064 | |
2065 | btrfs_set_root_node(&root->fs_info->tree_root->root_item, | |
2066 | root->fs_info->tree_root->node); | |
9e351cc8 JB |
2067 | list_add_tail(&root->fs_info->tree_root->dirty_list, |
2068 | &cur_trans->switch_commits); | |
5d4f98a2 YZ |
2069 | |
2070 | btrfs_set_root_node(&root->fs_info->chunk_root->root_item, | |
2071 | root->fs_info->chunk_root->node); | |
9e351cc8 JB |
2072 | list_add_tail(&root->fs_info->chunk_root->dirty_list, |
2073 | &cur_trans->switch_commits); | |
2074 | ||
2075 | switch_commit_roots(cur_trans, root->fs_info); | |
5d4f98a2 | 2076 | |
edf39272 | 2077 | assert_qgroups_uptodate(trans); |
ce93ec54 | 2078 | ASSERT(list_empty(&cur_trans->dirty_bgs)); |
1bbc621e | 2079 | ASSERT(list_empty(&cur_trans->io_bgs)); |
5d4f98a2 | 2080 | update_super_roots(root); |
e02119d5 | 2081 | |
60e7cd3a JB |
2082 | btrfs_set_super_log_root(root->fs_info->super_copy, 0); |
2083 | btrfs_set_super_log_root_level(root->fs_info->super_copy, 0); | |
6c41761f DS |
2084 | memcpy(root->fs_info->super_for_commit, root->fs_info->super_copy, |
2085 | sizeof(*root->fs_info->super_copy)); | |
ccd467d6 | 2086 | |
935e5cc9 | 2087 | btrfs_update_commit_device_size(root->fs_info); |
ce7213c7 | 2088 | btrfs_update_commit_device_bytes_used(root, cur_trans); |
935e5cc9 | 2089 | |
656f30db FM |
2090 | clear_bit(BTRFS_INODE_BTREE_LOG1_ERR, &btree_ino->runtime_flags); |
2091 | clear_bit(BTRFS_INODE_BTREE_LOG2_ERR, &btree_ino->runtime_flags); | |
2092 | ||
4fbcdf66 FM |
2093 | btrfs_trans_release_chunk_metadata(trans); |
2094 | ||
a4abeea4 | 2095 | spin_lock(&root->fs_info->trans_lock); |
4a9d8bde | 2096 | cur_trans->state = TRANS_STATE_UNBLOCKED; |
a4abeea4 | 2097 | root->fs_info->running_transaction = NULL; |
a4abeea4 | 2098 | spin_unlock(&root->fs_info->trans_lock); |
7585717f | 2099 | mutex_unlock(&root->fs_info->reloc_mutex); |
b7ec40d7 | 2100 | |
f9295749 | 2101 | wake_up(&root->fs_info->transaction_wait); |
e6dcd2dc | 2102 | |
79154b1b | 2103 | ret = btrfs_write_and_wait_transaction(trans, root); |
49b25e05 JM |
2104 | if (ret) { |
2105 | btrfs_error(root->fs_info, ret, | |
08748810 | 2106 | "Error while writing out transaction"); |
49b25e05 | 2107 | mutex_unlock(&root->fs_info->tree_log_mutex); |
6cf7f77e | 2108 | goto scrub_continue; |
49b25e05 JM |
2109 | } |
2110 | ||
2111 | ret = write_ctree_super(trans, root, 0); | |
2112 | if (ret) { | |
2113 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
6cf7f77e | 2114 | goto scrub_continue; |
49b25e05 | 2115 | } |
4313b399 | 2116 | |
e02119d5 CM |
2117 | /* |
2118 | * the super is written, we can safely allow the tree-loggers | |
2119 | * to go about their business | |
2120 | */ | |
2121 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
2122 | ||
11833d66 | 2123 | btrfs_finish_extent_commit(trans, root); |
4313b399 | 2124 | |
13212b54 ZL |
2125 | if (cur_trans->have_free_bgs) |
2126 | btrfs_clear_space_info_full(root->fs_info); | |
2127 | ||
15ee9bc7 | 2128 | root->fs_info->last_trans_committed = cur_trans->transid; |
4a9d8bde MX |
2129 | /* |
2130 | * We needn't acquire the lock here because there is no other task | |
2131 | * which can change it. | |
2132 | */ | |
2133 | cur_trans->state = TRANS_STATE_COMPLETED; | |
2c90e5d6 | 2134 | wake_up(&cur_trans->commit_wait); |
3de4586c | 2135 | |
a4abeea4 | 2136 | spin_lock(&root->fs_info->trans_lock); |
13c5a93e | 2137 | list_del_init(&cur_trans->list); |
a4abeea4 JB |
2138 | spin_unlock(&root->fs_info->trans_lock); |
2139 | ||
724e2315 JB |
2140 | btrfs_put_transaction(cur_trans); |
2141 | btrfs_put_transaction(cur_trans); | |
58176a96 | 2142 | |
0860adfd | 2143 | if (trans->type & __TRANS_FREEZABLE) |
354aa0fb | 2144 | sb_end_intwrite(root->fs_info->sb); |
b2b5ef5c | 2145 | |
1abe9b8a | 2146 | trace_btrfs_transaction_commit(root); |
2147 | ||
a2de733c AJ |
2148 | btrfs_scrub_continue(root); |
2149 | ||
9ed74f2d JB |
2150 | if (current->journal_info == trans) |
2151 | current->journal_info = NULL; | |
2152 | ||
2c90e5d6 | 2153 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
24bbcf04 | 2154 | |
8a733013 ZL |
2155 | if (current != root->fs_info->transaction_kthread && |
2156 | current != root->fs_info->cleaner_kthread) | |
24bbcf04 YZ |
2157 | btrfs_run_delayed_iputs(root); |
2158 | ||
79154b1b | 2159 | return ret; |
49b25e05 | 2160 | |
6cf7f77e WS |
2161 | scrub_continue: |
2162 | btrfs_scrub_continue(root); | |
49b25e05 | 2163 | cleanup_transaction: |
0e721106 | 2164 | btrfs_trans_release_metadata(trans, root); |
4fbcdf66 | 2165 | btrfs_trans_release_chunk_metadata(trans); |
0e721106 | 2166 | trans->block_rsv = NULL; |
272d26d0 MX |
2167 | if (trans->qgroup_reserved) { |
2168 | btrfs_qgroup_free(root, trans->qgroup_reserved); | |
2169 | trans->qgroup_reserved = 0; | |
2170 | } | |
c2cf52eb | 2171 | btrfs_warn(root->fs_info, "Skipping commit of aborted transaction."); |
49b25e05 JM |
2172 | if (current->journal_info == trans) |
2173 | current->journal_info = NULL; | |
7b8b92af | 2174 | cleanup_transaction(trans, root, ret); |
49b25e05 JM |
2175 | |
2176 | return ret; | |
79154b1b CM |
2177 | } |
2178 | ||
d352ac68 | 2179 | /* |
9d1a2a3a DS |
2180 | * return < 0 if error |
2181 | * 0 if there are no more dead_roots at the time of call | |
2182 | * 1 there are more to be processed, call me again | |
2183 | * | |
2184 | * The return value indicates there are certainly more snapshots to delete, but | |
2185 | * if there comes a new one during processing, it may return 0. We don't mind, | |
2186 | * because btrfs_commit_super will poke cleaner thread and it will process it a | |
2187 | * few seconds later. | |
d352ac68 | 2188 | */ |
9d1a2a3a | 2189 | int btrfs_clean_one_deleted_snapshot(struct btrfs_root *root) |
e9d0b13b | 2190 | { |
9d1a2a3a | 2191 | int ret; |
5d4f98a2 YZ |
2192 | struct btrfs_fs_info *fs_info = root->fs_info; |
2193 | ||
a4abeea4 | 2194 | spin_lock(&fs_info->trans_lock); |
9d1a2a3a DS |
2195 | if (list_empty(&fs_info->dead_roots)) { |
2196 | spin_unlock(&fs_info->trans_lock); | |
2197 | return 0; | |
2198 | } | |
2199 | root = list_first_entry(&fs_info->dead_roots, | |
2200 | struct btrfs_root, root_list); | |
cfad392b | 2201 | list_del_init(&root->root_list); |
a4abeea4 | 2202 | spin_unlock(&fs_info->trans_lock); |
e9d0b13b | 2203 | |
efe120a0 | 2204 | pr_debug("BTRFS: cleaner removing %llu\n", root->objectid); |
76dda93c | 2205 | |
9d1a2a3a | 2206 | btrfs_kill_all_delayed_nodes(root); |
16cdcec7 | 2207 | |
9d1a2a3a DS |
2208 | if (btrfs_header_backref_rev(root->node) < |
2209 | BTRFS_MIXED_BACKREF_REV) | |
2210 | ret = btrfs_drop_snapshot(root, NULL, 0, 0); | |
2211 | else | |
2212 | ret = btrfs_drop_snapshot(root, NULL, 1, 0); | |
32471dc2 | 2213 | |
6596a928 | 2214 | return (ret < 0) ? 0 : 1; |
e9d0b13b | 2215 | } |
572d9ab7 DS |
2216 | |
2217 | void btrfs_apply_pending_changes(struct btrfs_fs_info *fs_info) | |
2218 | { | |
2219 | unsigned long prev; | |
2220 | unsigned long bit; | |
2221 | ||
6c9fe14f | 2222 | prev = xchg(&fs_info->pending_changes, 0); |
572d9ab7 DS |
2223 | if (!prev) |
2224 | return; | |
2225 | ||
7e1876ac DS |
2226 | bit = 1 << BTRFS_PENDING_SET_INODE_MAP_CACHE; |
2227 | if (prev & bit) | |
2228 | btrfs_set_opt(fs_info->mount_opt, INODE_MAP_CACHE); | |
2229 | prev &= ~bit; | |
2230 | ||
2231 | bit = 1 << BTRFS_PENDING_CLEAR_INODE_MAP_CACHE; | |
2232 | if (prev & bit) | |
2233 | btrfs_clear_opt(fs_info->mount_opt, INODE_MAP_CACHE); | |
2234 | prev &= ~bit; | |
2235 | ||
d51033d0 DS |
2236 | bit = 1 << BTRFS_PENDING_COMMIT; |
2237 | if (prev & bit) | |
2238 | btrfs_debug(fs_info, "pending commit done"); | |
2239 | prev &= ~bit; | |
2240 | ||
572d9ab7 DS |
2241 | if (prev) |
2242 | btrfs_warn(fs_info, | |
2243 | "unknown pending changes left 0x%lx, ignoring", prev); | |
2244 | } |