]>
Commit | Line | Data |
---|---|---|
56bec294 CM |
1 | /* |
2 | * Copyright (C) 2009 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 | ||
19 | #include <linux/sched.h> | |
5a0e3ad6 | 20 | #include <linux/slab.h> |
56bec294 | 21 | #include <linux/sort.h> |
56bec294 CM |
22 | #include "ctree.h" |
23 | #include "delayed-ref.h" | |
24 | #include "transaction.h" | |
3368d001 | 25 | #include "qgroup.h" |
56bec294 | 26 | |
78a6184a MX |
27 | struct kmem_cache *btrfs_delayed_ref_head_cachep; |
28 | struct kmem_cache *btrfs_delayed_tree_ref_cachep; | |
29 | struct kmem_cache *btrfs_delayed_data_ref_cachep; | |
30 | struct kmem_cache *btrfs_delayed_extent_op_cachep; | |
56bec294 CM |
31 | /* |
32 | * delayed back reference update tracking. For subvolume trees | |
33 | * we queue up extent allocations and backref maintenance for | |
34 | * delayed processing. This avoids deep call chains where we | |
35 | * add extents in the middle of btrfs_search_slot, and it allows | |
36 | * us to buffer up frequently modified backrefs in an rb tree instead | |
37 | * of hammering updates on the extent allocation tree. | |
56bec294 CM |
38 | */ |
39 | ||
40 | /* | |
5d4f98a2 YZ |
41 | * compare two delayed tree backrefs with same bytenr and type |
42 | */ | |
43 | static int comp_tree_refs(struct btrfs_delayed_tree_ref *ref2, | |
41b0fc42 | 44 | struct btrfs_delayed_tree_ref *ref1, int type) |
5d4f98a2 | 45 | { |
41b0fc42 JB |
46 | if (type == BTRFS_TREE_BLOCK_REF_KEY) { |
47 | if (ref1->root < ref2->root) | |
48 | return -1; | |
49 | if (ref1->root > ref2->root) | |
50 | return 1; | |
51 | } else { | |
52 | if (ref1->parent < ref2->parent) | |
53 | return -1; | |
54 | if (ref1->parent > ref2->parent) | |
55 | return 1; | |
56 | } | |
5d4f98a2 YZ |
57 | return 0; |
58 | } | |
59 | ||
60 | /* | |
61 | * compare two delayed data backrefs with same bytenr and type | |
56bec294 | 62 | */ |
5d4f98a2 YZ |
63 | static int comp_data_refs(struct btrfs_delayed_data_ref *ref2, |
64 | struct btrfs_delayed_data_ref *ref1) | |
56bec294 | 65 | { |
5d4f98a2 YZ |
66 | if (ref1->node.type == BTRFS_EXTENT_DATA_REF_KEY) { |
67 | if (ref1->root < ref2->root) | |
68 | return -1; | |
69 | if (ref1->root > ref2->root) | |
70 | return 1; | |
71 | if (ref1->objectid < ref2->objectid) | |
72 | return -1; | |
73 | if (ref1->objectid > ref2->objectid) | |
74 | return 1; | |
75 | if (ref1->offset < ref2->offset) | |
76 | return -1; | |
77 | if (ref1->offset > ref2->offset) | |
78 | return 1; | |
79 | } else { | |
80 | if (ref1->parent < ref2->parent) | |
81 | return -1; | |
82 | if (ref1->parent > ref2->parent) | |
83 | return 1; | |
84 | } | |
85 | return 0; | |
86 | } | |
87 | ||
c46effa6 LB |
88 | /* insert a new ref to head ref rbtree */ |
89 | static struct btrfs_delayed_ref_head *htree_insert(struct rb_root *root, | |
90 | struct rb_node *node) | |
91 | { | |
92 | struct rb_node **p = &root->rb_node; | |
93 | struct rb_node *parent_node = NULL; | |
94 | struct btrfs_delayed_ref_head *entry; | |
95 | struct btrfs_delayed_ref_head *ins; | |
96 | u64 bytenr; | |
97 | ||
98 | ins = rb_entry(node, struct btrfs_delayed_ref_head, href_node); | |
99 | bytenr = ins->node.bytenr; | |
100 | while (*p) { | |
101 | parent_node = *p; | |
102 | entry = rb_entry(parent_node, struct btrfs_delayed_ref_head, | |
103 | href_node); | |
104 | ||
105 | if (bytenr < entry->node.bytenr) | |
106 | p = &(*p)->rb_left; | |
107 | else if (bytenr > entry->node.bytenr) | |
108 | p = &(*p)->rb_right; | |
109 | else | |
110 | return entry; | |
111 | } | |
112 | ||
113 | rb_link_node(node, parent_node, p); | |
114 | rb_insert_color(node, root); | |
115 | return NULL; | |
116 | } | |
117 | ||
56bec294 | 118 | /* |
5d4f98a2 | 119 | * find an head entry based on bytenr. This returns the delayed ref |
d1270cd9 AJ |
120 | * head if it was able to find one, or NULL if nothing was in that spot. |
121 | * If return_bigger is given, the next bigger entry is returned if no exact | |
122 | * match is found. | |
56bec294 | 123 | */ |
c46effa6 LB |
124 | static struct btrfs_delayed_ref_head * |
125 | find_ref_head(struct rb_root *root, u64 bytenr, | |
85fdfdf6 | 126 | int return_bigger) |
56bec294 | 127 | { |
d1270cd9 | 128 | struct rb_node *n; |
c46effa6 | 129 | struct btrfs_delayed_ref_head *entry; |
56bec294 | 130 | |
d1270cd9 AJ |
131 | n = root->rb_node; |
132 | entry = NULL; | |
56bec294 | 133 | while (n) { |
c46effa6 | 134 | entry = rb_entry(n, struct btrfs_delayed_ref_head, href_node); |
56bec294 | 135 | |
c46effa6 | 136 | if (bytenr < entry->node.bytenr) |
56bec294 | 137 | n = n->rb_left; |
85fdfdf6 | 138 | else if (bytenr > entry->node.bytenr) |
56bec294 CM |
139 | n = n->rb_right; |
140 | else | |
141 | return entry; | |
142 | } | |
d1270cd9 | 143 | if (entry && return_bigger) { |
85fdfdf6 | 144 | if (bytenr > entry->node.bytenr) { |
c46effa6 | 145 | n = rb_next(&entry->href_node); |
d1270cd9 AJ |
146 | if (!n) |
147 | n = rb_first(root); | |
c46effa6 LB |
148 | entry = rb_entry(n, struct btrfs_delayed_ref_head, |
149 | href_node); | |
6103fb43 | 150 | return entry; |
d1270cd9 AJ |
151 | } |
152 | return entry; | |
153 | } | |
56bec294 CM |
154 | return NULL; |
155 | } | |
156 | ||
c3e69d58 CM |
157 | int btrfs_delayed_ref_lock(struct btrfs_trans_handle *trans, |
158 | struct btrfs_delayed_ref_head *head) | |
56bec294 | 159 | { |
c3e69d58 CM |
160 | struct btrfs_delayed_ref_root *delayed_refs; |
161 | ||
162 | delayed_refs = &trans->transaction->delayed_refs; | |
163 | assert_spin_locked(&delayed_refs->lock); | |
164 | if (mutex_trylock(&head->mutex)) | |
165 | return 0; | |
166 | ||
167 | atomic_inc(&head->node.refs); | |
168 | spin_unlock(&delayed_refs->lock); | |
169 | ||
170 | mutex_lock(&head->mutex); | |
171 | spin_lock(&delayed_refs->lock); | |
172 | if (!head->node.in_tree) { | |
173 | mutex_unlock(&head->mutex); | |
174 | btrfs_put_delayed_ref(&head->node); | |
175 | return -EAGAIN; | |
176 | } | |
177 | btrfs_put_delayed_ref(&head->node); | |
178 | return 0; | |
179 | } | |
180 | ||
35a3621b | 181 | static inline void drop_delayed_ref(struct btrfs_trans_handle *trans, |
ae1e206b | 182 | struct btrfs_delayed_ref_root *delayed_refs, |
d7df2c79 | 183 | struct btrfs_delayed_ref_head *head, |
ae1e206b JB |
184 | struct btrfs_delayed_ref_node *ref) |
185 | { | |
c46effa6 | 186 | if (btrfs_delayed_ref_is_head(ref)) { |
c46effa6 LB |
187 | head = btrfs_delayed_node_to_head(ref); |
188 | rb_erase(&head->href_node, &delayed_refs->href_root); | |
d7df2c79 JB |
189 | } else { |
190 | assert_spin_locked(&head->lock); | |
c6fc2454 | 191 | list_del(&ref->list); |
c46effa6 | 192 | } |
ae1e206b JB |
193 | ref->in_tree = 0; |
194 | btrfs_put_delayed_ref(ref); | |
d7df2c79 | 195 | atomic_dec(&delayed_refs->num_entries); |
ae1e206b JB |
196 | if (trans->delayed_ref_updates) |
197 | trans->delayed_ref_updates--; | |
198 | } | |
199 | ||
2c3cf7d5 FM |
200 | static bool merge_ref(struct btrfs_trans_handle *trans, |
201 | struct btrfs_delayed_ref_root *delayed_refs, | |
202 | struct btrfs_delayed_ref_head *head, | |
203 | struct btrfs_delayed_ref_node *ref, | |
204 | u64 seq) | |
205 | { | |
206 | struct btrfs_delayed_ref_node *next; | |
207 | bool done = false; | |
208 | ||
209 | next = list_first_entry(&head->ref_list, struct btrfs_delayed_ref_node, | |
210 | list); | |
211 | while (!done && &next->list != &head->ref_list) { | |
212 | int mod; | |
213 | struct btrfs_delayed_ref_node *next2; | |
214 | ||
215 | next2 = list_next_entry(next, list); | |
216 | ||
217 | if (next == ref) | |
218 | goto next; | |
219 | ||
220 | if (seq && next->seq >= seq) | |
221 | goto next; | |
222 | ||
b06c4bf5 | 223 | if (next->type != ref->type) |
2c3cf7d5 FM |
224 | goto next; |
225 | ||
226 | if ((ref->type == BTRFS_TREE_BLOCK_REF_KEY || | |
227 | ref->type == BTRFS_SHARED_BLOCK_REF_KEY) && | |
228 | comp_tree_refs(btrfs_delayed_node_to_tree_ref(ref), | |
229 | btrfs_delayed_node_to_tree_ref(next), | |
230 | ref->type)) | |
231 | goto next; | |
232 | if ((ref->type == BTRFS_EXTENT_DATA_REF_KEY || | |
233 | ref->type == BTRFS_SHARED_DATA_REF_KEY) && | |
234 | comp_data_refs(btrfs_delayed_node_to_data_ref(ref), | |
235 | btrfs_delayed_node_to_data_ref(next))) | |
236 | goto next; | |
237 | ||
238 | if (ref->action == next->action) { | |
239 | mod = next->ref_mod; | |
240 | } else { | |
241 | if (ref->ref_mod < next->ref_mod) { | |
242 | swap(ref, next); | |
243 | done = true; | |
244 | } | |
245 | mod = -next->ref_mod; | |
246 | } | |
247 | ||
248 | drop_delayed_ref(trans, delayed_refs, head, next); | |
249 | ref->ref_mod += mod; | |
250 | if (ref->ref_mod == 0) { | |
251 | drop_delayed_ref(trans, delayed_refs, head, ref); | |
252 | done = true; | |
253 | } else { | |
254 | /* | |
255 | * Can't have multiples of the same ref on a tree block. | |
256 | */ | |
257 | WARN_ON(ref->type == BTRFS_TREE_BLOCK_REF_KEY || | |
258 | ref->type == BTRFS_SHARED_BLOCK_REF_KEY); | |
259 | } | |
260 | next: | |
261 | next = next2; | |
262 | } | |
263 | ||
264 | return done; | |
265 | } | |
266 | ||
267 | void btrfs_merge_delayed_refs(struct btrfs_trans_handle *trans, | |
268 | struct btrfs_fs_info *fs_info, | |
269 | struct btrfs_delayed_ref_root *delayed_refs, | |
270 | struct btrfs_delayed_ref_head *head) | |
271 | { | |
272 | struct btrfs_delayed_ref_node *ref; | |
273 | u64 seq = 0; | |
274 | ||
275 | assert_spin_locked(&head->lock); | |
276 | ||
277 | if (list_empty(&head->ref_list)) | |
278 | return; | |
279 | ||
280 | /* We don't have too many refs to merge for data. */ | |
281 | if (head->is_data) | |
282 | return; | |
283 | ||
284 | spin_lock(&fs_info->tree_mod_seq_lock); | |
285 | if (!list_empty(&fs_info->tree_mod_seq_list)) { | |
286 | struct seq_list *elem; | |
287 | ||
288 | elem = list_first_entry(&fs_info->tree_mod_seq_list, | |
289 | struct seq_list, list); | |
290 | seq = elem->seq; | |
291 | } | |
292 | spin_unlock(&fs_info->tree_mod_seq_lock); | |
293 | ||
294 | ref = list_first_entry(&head->ref_list, struct btrfs_delayed_ref_node, | |
295 | list); | |
296 | while (&ref->list != &head->ref_list) { | |
297 | if (seq && ref->seq >= seq) | |
298 | goto next; | |
299 | ||
300 | if (merge_ref(trans, delayed_refs, head, ref, seq)) { | |
301 | if (list_empty(&head->ref_list)) | |
302 | break; | |
303 | ref = list_first_entry(&head->ref_list, | |
304 | struct btrfs_delayed_ref_node, | |
305 | list); | |
306 | continue; | |
307 | } | |
308 | next: | |
309 | ref = list_next_entry(ref, list); | |
310 | } | |
311 | } | |
312 | ||
097b8a7c JS |
313 | int btrfs_check_delayed_seq(struct btrfs_fs_info *fs_info, |
314 | struct btrfs_delayed_ref_root *delayed_refs, | |
00f04b88 AJ |
315 | u64 seq) |
316 | { | |
317 | struct seq_list *elem; | |
097b8a7c JS |
318 | int ret = 0; |
319 | ||
320 | spin_lock(&fs_info->tree_mod_seq_lock); | |
321 | if (!list_empty(&fs_info->tree_mod_seq_list)) { | |
322 | elem = list_first_entry(&fs_info->tree_mod_seq_list, | |
323 | struct seq_list, list); | |
324 | if (seq >= elem->seq) { | |
fc36ed7e JS |
325 | pr_debug("holding back delayed_ref %#x.%x, lowest is %#x.%x (%p)\n", |
326 | (u32)(seq >> 32), (u32)seq, | |
327 | (u32)(elem->seq >> 32), (u32)elem->seq, | |
328 | delayed_refs); | |
097b8a7c JS |
329 | ret = 1; |
330 | } | |
00f04b88 | 331 | } |
097b8a7c JS |
332 | |
333 | spin_unlock(&fs_info->tree_mod_seq_lock); | |
334 | return ret; | |
00f04b88 AJ |
335 | } |
336 | ||
d7df2c79 JB |
337 | struct btrfs_delayed_ref_head * |
338 | btrfs_select_ref_head(struct btrfs_trans_handle *trans) | |
c3e69d58 | 339 | { |
c3e69d58 | 340 | struct btrfs_delayed_ref_root *delayed_refs; |
d7df2c79 JB |
341 | struct btrfs_delayed_ref_head *head; |
342 | u64 start; | |
343 | bool loop = false; | |
56bec294 | 344 | |
c3e69d58 | 345 | delayed_refs = &trans->transaction->delayed_refs; |
c46effa6 | 346 | |
c3e69d58 | 347 | again: |
d7df2c79 | 348 | start = delayed_refs->run_delayed_start; |
85fdfdf6 | 349 | head = find_ref_head(&delayed_refs->href_root, start, 1); |
d7df2c79 JB |
350 | if (!head && !loop) { |
351 | delayed_refs->run_delayed_start = 0; | |
c3e69d58 | 352 | start = 0; |
d7df2c79 | 353 | loop = true; |
85fdfdf6 | 354 | head = find_ref_head(&delayed_refs->href_root, start, 1); |
d7df2c79 JB |
355 | if (!head) |
356 | return NULL; | |
357 | } else if (!head && loop) { | |
358 | return NULL; | |
c3e69d58 | 359 | } |
56bec294 | 360 | |
d7df2c79 JB |
361 | while (head->processing) { |
362 | struct rb_node *node; | |
363 | ||
364 | node = rb_next(&head->href_node); | |
365 | if (!node) { | |
366 | if (loop) | |
367 | return NULL; | |
368 | delayed_refs->run_delayed_start = 0; | |
369 | start = 0; | |
370 | loop = true; | |
371 | goto again; | |
372 | } | |
373 | head = rb_entry(node, struct btrfs_delayed_ref_head, | |
374 | href_node); | |
375 | } | |
093486c4 | 376 | |
d7df2c79 JB |
377 | head->processing = 1; |
378 | WARN_ON(delayed_refs->num_heads_ready == 0); | |
379 | delayed_refs->num_heads_ready--; | |
380 | delayed_refs->run_delayed_start = head->node.bytenr + | |
381 | head->node.num_bytes; | |
382 | return head; | |
093486c4 MX |
383 | } |
384 | ||
c6fc2454 QW |
385 | /* |
386 | * Helper to insert the ref_node to the tail or merge with tail. | |
387 | * | |
388 | * Return 0 for insert. | |
389 | * Return >0 for merge. | |
390 | */ | |
391 | static int | |
392 | add_delayed_ref_tail_merge(struct btrfs_trans_handle *trans, | |
393 | struct btrfs_delayed_ref_root *root, | |
394 | struct btrfs_delayed_ref_head *href, | |
395 | struct btrfs_delayed_ref_node *ref) | |
396 | { | |
397 | struct btrfs_delayed_ref_node *exist; | |
398 | int mod; | |
399 | int ret = 0; | |
400 | ||
401 | spin_lock(&href->lock); | |
402 | /* Check whether we can merge the tail node with ref */ | |
403 | if (list_empty(&href->ref_list)) | |
404 | goto add_tail; | |
405 | exist = list_entry(href->ref_list.prev, struct btrfs_delayed_ref_node, | |
406 | list); | |
407 | /* No need to compare bytenr nor is_head */ | |
b06c4bf5 | 408 | if (exist->type != ref->type || exist->seq != ref->seq) |
c6fc2454 QW |
409 | goto add_tail; |
410 | ||
411 | if ((exist->type == BTRFS_TREE_BLOCK_REF_KEY || | |
412 | exist->type == BTRFS_SHARED_BLOCK_REF_KEY) && | |
413 | comp_tree_refs(btrfs_delayed_node_to_tree_ref(exist), | |
414 | btrfs_delayed_node_to_tree_ref(ref), | |
415 | ref->type)) | |
416 | goto add_tail; | |
417 | if ((exist->type == BTRFS_EXTENT_DATA_REF_KEY || | |
418 | exist->type == BTRFS_SHARED_DATA_REF_KEY) && | |
419 | comp_data_refs(btrfs_delayed_node_to_data_ref(exist), | |
420 | btrfs_delayed_node_to_data_ref(ref))) | |
421 | goto add_tail; | |
422 | ||
423 | /* Now we are sure we can merge */ | |
424 | ret = 1; | |
425 | if (exist->action == ref->action) { | |
426 | mod = ref->ref_mod; | |
427 | } else { | |
428 | /* Need to change action */ | |
429 | if (exist->ref_mod < ref->ref_mod) { | |
430 | exist->action = ref->action; | |
431 | mod = -exist->ref_mod; | |
432 | exist->ref_mod = ref->ref_mod; | |
433 | } else | |
434 | mod = -ref->ref_mod; | |
435 | } | |
436 | exist->ref_mod += mod; | |
437 | ||
438 | /* remove existing tail if its ref_mod is zero */ | |
439 | if (exist->ref_mod == 0) | |
440 | drop_delayed_ref(trans, root, href, exist); | |
441 | spin_unlock(&href->lock); | |
442 | return ret; | |
443 | ||
444 | add_tail: | |
445 | list_add_tail(&ref->list, &href->ref_list); | |
446 | atomic_inc(&root->num_entries); | |
447 | trans->delayed_ref_updates++; | |
448 | spin_unlock(&href->lock); | |
449 | return ret; | |
450 | } | |
451 | ||
56bec294 CM |
452 | /* |
453 | * helper function to update the accounting in the head ref | |
454 | * existing and update must have the same bytenr | |
455 | */ | |
456 | static noinline void | |
1262133b JB |
457 | update_existing_head_ref(struct btrfs_delayed_ref_root *delayed_refs, |
458 | struct btrfs_delayed_ref_node *existing, | |
56bec294 CM |
459 | struct btrfs_delayed_ref_node *update) |
460 | { | |
461 | struct btrfs_delayed_ref_head *existing_ref; | |
462 | struct btrfs_delayed_ref_head *ref; | |
1262133b | 463 | int old_ref_mod; |
56bec294 CM |
464 | |
465 | existing_ref = btrfs_delayed_node_to_head(existing); | |
466 | ref = btrfs_delayed_node_to_head(update); | |
5d4f98a2 | 467 | BUG_ON(existing_ref->is_data != ref->is_data); |
56bec294 | 468 | |
21543bad | 469 | spin_lock(&existing_ref->lock); |
56bec294 CM |
470 | if (ref->must_insert_reserved) { |
471 | /* if the extent was freed and then | |
472 | * reallocated before the delayed ref | |
473 | * entries were processed, we can end up | |
474 | * with an existing head ref without | |
475 | * the must_insert_reserved flag set. | |
476 | * Set it again here | |
477 | */ | |
478 | existing_ref->must_insert_reserved = ref->must_insert_reserved; | |
479 | ||
480 | /* | |
481 | * update the num_bytes so we make sure the accounting | |
482 | * is done correctly | |
483 | */ | |
484 | existing->num_bytes = update->num_bytes; | |
485 | ||
486 | } | |
487 | ||
5d4f98a2 YZ |
488 | if (ref->extent_op) { |
489 | if (!existing_ref->extent_op) { | |
490 | existing_ref->extent_op = ref->extent_op; | |
491 | } else { | |
492 | if (ref->extent_op->update_key) { | |
493 | memcpy(&existing_ref->extent_op->key, | |
494 | &ref->extent_op->key, | |
495 | sizeof(ref->extent_op->key)); | |
496 | existing_ref->extent_op->update_key = 1; | |
497 | } | |
498 | if (ref->extent_op->update_flags) { | |
499 | existing_ref->extent_op->flags_to_set |= | |
500 | ref->extent_op->flags_to_set; | |
501 | existing_ref->extent_op->update_flags = 1; | |
502 | } | |
78a6184a | 503 | btrfs_free_delayed_extent_op(ref->extent_op); |
5d4f98a2 YZ |
504 | } |
505 | } | |
56bec294 | 506 | /* |
d7df2c79 JB |
507 | * update the reference mod on the head to reflect this new operation, |
508 | * only need the lock for this case cause we could be processing it | |
509 | * currently, for refs we just added we know we're a-ok. | |
56bec294 | 510 | */ |
1262133b | 511 | old_ref_mod = existing_ref->total_ref_mod; |
56bec294 | 512 | existing->ref_mod += update->ref_mod; |
1262133b JB |
513 | existing_ref->total_ref_mod += update->ref_mod; |
514 | ||
515 | /* | |
516 | * If we are going to from a positive ref mod to a negative or vice | |
517 | * versa we need to make sure to adjust pending_csums accordingly. | |
518 | */ | |
519 | if (existing_ref->is_data) { | |
520 | if (existing_ref->total_ref_mod >= 0 && old_ref_mod < 0) | |
521 | delayed_refs->pending_csums -= existing->num_bytes; | |
522 | if (existing_ref->total_ref_mod < 0 && old_ref_mod >= 0) | |
523 | delayed_refs->pending_csums += existing->num_bytes; | |
524 | } | |
d7df2c79 | 525 | spin_unlock(&existing_ref->lock); |
56bec294 CM |
526 | } |
527 | ||
528 | /* | |
5d4f98a2 | 529 | * helper function to actually insert a head node into the rbtree. |
56bec294 | 530 | * this does all the dirty work in terms of maintaining the correct |
5d4f98a2 | 531 | * overall modification count. |
56bec294 | 532 | */ |
d7df2c79 JB |
533 | static noinline struct btrfs_delayed_ref_head * |
534 | add_delayed_ref_head(struct btrfs_fs_info *fs_info, | |
535 | struct btrfs_trans_handle *trans, | |
3368d001 QW |
536 | struct btrfs_delayed_ref_node *ref, |
537 | struct btrfs_qgroup_extent_record *qrecord, | |
538 | u64 bytenr, u64 num_bytes, int action, int is_data) | |
56bec294 | 539 | { |
d7df2c79 | 540 | struct btrfs_delayed_ref_head *existing; |
c3e69d58 | 541 | struct btrfs_delayed_ref_head *head_ref = NULL; |
56bec294 | 542 | struct btrfs_delayed_ref_root *delayed_refs; |
3368d001 | 543 | struct btrfs_qgroup_extent_record *qexisting; |
56bec294 CM |
544 | int count_mod = 1; |
545 | int must_insert_reserved = 0; | |
546 | ||
547 | /* | |
548 | * the head node stores the sum of all the mods, so dropping a ref | |
549 | * should drop the sum in the head node by one. | |
550 | */ | |
5d4f98a2 YZ |
551 | if (action == BTRFS_UPDATE_DELAYED_HEAD) |
552 | count_mod = 0; | |
553 | else if (action == BTRFS_DROP_DELAYED_REF) | |
554 | count_mod = -1; | |
56bec294 CM |
555 | |
556 | /* | |
557 | * BTRFS_ADD_DELAYED_EXTENT means that we need to update | |
558 | * the reserved accounting when the extent is finally added, or | |
559 | * if a later modification deletes the delayed ref without ever | |
560 | * inserting the extent into the extent allocation tree. | |
561 | * ref->must_insert_reserved is the flag used to record | |
562 | * that accounting mods are required. | |
563 | * | |
564 | * Once we record must_insert_reserved, switch the action to | |
565 | * BTRFS_ADD_DELAYED_REF because other special casing is not required. | |
566 | */ | |
5d4f98a2 | 567 | if (action == BTRFS_ADD_DELAYED_EXTENT) |
56bec294 | 568 | must_insert_reserved = 1; |
5d4f98a2 | 569 | else |
56bec294 | 570 | must_insert_reserved = 0; |
56bec294 CM |
571 | |
572 | delayed_refs = &trans->transaction->delayed_refs; | |
573 | ||
574 | /* first set the basic ref node struct up */ | |
575 | atomic_set(&ref->refs, 1); | |
576 | ref->bytenr = bytenr; | |
5d4f98a2 | 577 | ref->num_bytes = num_bytes; |
56bec294 | 578 | ref->ref_mod = count_mod; |
5d4f98a2 YZ |
579 | ref->type = 0; |
580 | ref->action = 0; | |
581 | ref->is_head = 1; | |
56bec294 | 582 | ref->in_tree = 1; |
00f04b88 | 583 | ref->seq = 0; |
5d4f98a2 YZ |
584 | |
585 | head_ref = btrfs_delayed_node_to_head(ref); | |
586 | head_ref->must_insert_reserved = must_insert_reserved; | |
587 | head_ref->is_data = is_data; | |
c6fc2454 | 588 | INIT_LIST_HEAD(&head_ref->ref_list); |
d7df2c79 | 589 | head_ref->processing = 0; |
1262133b | 590 | head_ref->total_ref_mod = count_mod; |
f64d5ca8 QW |
591 | head_ref->qgroup_reserved = 0; |
592 | head_ref->qgroup_ref_root = 0; | |
5d4f98a2 | 593 | |
3368d001 QW |
594 | /* Record qgroup extent info if provided */ |
595 | if (qrecord) { | |
596 | qrecord->bytenr = bytenr; | |
597 | qrecord->num_bytes = num_bytes; | |
598 | qrecord->old_roots = NULL; | |
599 | ||
600 | qexisting = btrfs_qgroup_insert_dirty_extent(delayed_refs, | |
601 | qrecord); | |
602 | if (qexisting) | |
603 | kfree(qrecord); | |
604 | } | |
605 | ||
d7df2c79 | 606 | spin_lock_init(&head_ref->lock); |
5d4f98a2 YZ |
607 | mutex_init(&head_ref->mutex); |
608 | ||
599c75ec | 609 | trace_add_delayed_ref_head(ref, head_ref, action); |
1abe9b8a | 610 | |
d7df2c79 JB |
611 | existing = htree_insert(&delayed_refs->href_root, |
612 | &head_ref->href_node); | |
5d4f98a2 | 613 | if (existing) { |
1262133b | 614 | update_existing_head_ref(delayed_refs, &existing->node, ref); |
5d4f98a2 YZ |
615 | /* |
616 | * we've updated the existing ref, free the newly | |
617 | * allocated ref | |
618 | */ | |
78a6184a | 619 | kmem_cache_free(btrfs_delayed_ref_head_cachep, head_ref); |
d7df2c79 | 620 | head_ref = existing; |
5d4f98a2 | 621 | } else { |
1262133b JB |
622 | if (is_data && count_mod < 0) |
623 | delayed_refs->pending_csums += num_bytes; | |
5d4f98a2 YZ |
624 | delayed_refs->num_heads++; |
625 | delayed_refs->num_heads_ready++; | |
d7df2c79 | 626 | atomic_inc(&delayed_refs->num_entries); |
5d4f98a2 YZ |
627 | trans->delayed_ref_updates++; |
628 | } | |
d7df2c79 | 629 | return head_ref; |
5d4f98a2 YZ |
630 | } |
631 | ||
632 | /* | |
633 | * helper to insert a delayed tree ref into the rbtree. | |
634 | */ | |
d7df2c79 JB |
635 | static noinline void |
636 | add_delayed_tree_ref(struct btrfs_fs_info *fs_info, | |
637 | struct btrfs_trans_handle *trans, | |
638 | struct btrfs_delayed_ref_head *head_ref, | |
639 | struct btrfs_delayed_ref_node *ref, u64 bytenr, | |
640 | u64 num_bytes, u64 parent, u64 ref_root, int level, | |
b06c4bf5 | 641 | int action) |
5d4f98a2 | 642 | { |
5d4f98a2 YZ |
643 | struct btrfs_delayed_tree_ref *full_ref; |
644 | struct btrfs_delayed_ref_root *delayed_refs; | |
00f04b88 | 645 | u64 seq = 0; |
c6fc2454 | 646 | int ret; |
5d4f98a2 YZ |
647 | |
648 | if (action == BTRFS_ADD_DELAYED_EXTENT) | |
649 | action = BTRFS_ADD_DELAYED_REF; | |
650 | ||
fcebe456 JB |
651 | if (is_fstree(ref_root)) |
652 | seq = atomic64_read(&fs_info->tree_mod_seq); | |
5d4f98a2 YZ |
653 | delayed_refs = &trans->transaction->delayed_refs; |
654 | ||
655 | /* first set the basic ref node struct up */ | |
656 | atomic_set(&ref->refs, 1); | |
657 | ref->bytenr = bytenr; | |
56bec294 | 658 | ref->num_bytes = num_bytes; |
5d4f98a2 YZ |
659 | ref->ref_mod = 1; |
660 | ref->action = action; | |
661 | ref->is_head = 0; | |
662 | ref->in_tree = 1; | |
00f04b88 AJ |
663 | ref->seq = seq; |
664 | ||
5d4f98a2 | 665 | full_ref = btrfs_delayed_node_to_tree_ref(ref); |
eebe063b AJ |
666 | full_ref->parent = parent; |
667 | full_ref->root = ref_root; | |
668 | if (parent) | |
5d4f98a2 | 669 | ref->type = BTRFS_SHARED_BLOCK_REF_KEY; |
eebe063b | 670 | else |
5d4f98a2 | 671 | ref->type = BTRFS_TREE_BLOCK_REF_KEY; |
5d4f98a2 | 672 | full_ref->level = level; |
56bec294 | 673 | |
599c75ec | 674 | trace_add_delayed_tree_ref(ref, full_ref, action); |
1abe9b8a | 675 | |
c6fc2454 QW |
676 | ret = add_delayed_ref_tail_merge(trans, delayed_refs, head_ref, ref); |
677 | ||
678 | /* | |
679 | * XXX: memory should be freed at the same level allocated. | |
680 | * But bad practice is anywhere... Follow it now. Need cleanup. | |
681 | */ | |
682 | if (ret > 0) | |
78a6184a | 683 | kmem_cache_free(btrfs_delayed_tree_ref_cachep, full_ref); |
5d4f98a2 YZ |
684 | } |
685 | ||
686 | /* | |
687 | * helper to insert a delayed data ref into the rbtree. | |
688 | */ | |
d7df2c79 JB |
689 | static noinline void |
690 | add_delayed_data_ref(struct btrfs_fs_info *fs_info, | |
691 | struct btrfs_trans_handle *trans, | |
692 | struct btrfs_delayed_ref_head *head_ref, | |
693 | struct btrfs_delayed_ref_node *ref, u64 bytenr, | |
694 | u64 num_bytes, u64 parent, u64 ref_root, u64 owner, | |
b06c4bf5 | 695 | u64 offset, int action) |
5d4f98a2 | 696 | { |
5d4f98a2 YZ |
697 | struct btrfs_delayed_data_ref *full_ref; |
698 | struct btrfs_delayed_ref_root *delayed_refs; | |
00f04b88 | 699 | u64 seq = 0; |
c6fc2454 | 700 | int ret; |
5d4f98a2 YZ |
701 | |
702 | if (action == BTRFS_ADD_DELAYED_EXTENT) | |
703 | action = BTRFS_ADD_DELAYED_REF; | |
704 | ||
705 | delayed_refs = &trans->transaction->delayed_refs; | |
706 | ||
fcebe456 JB |
707 | if (is_fstree(ref_root)) |
708 | seq = atomic64_read(&fs_info->tree_mod_seq); | |
709 | ||
5d4f98a2 YZ |
710 | /* first set the basic ref node struct up */ |
711 | atomic_set(&ref->refs, 1); | |
712 | ref->bytenr = bytenr; | |
713 | ref->num_bytes = num_bytes; | |
714 | ref->ref_mod = 1; | |
715 | ref->action = action; | |
716 | ref->is_head = 0; | |
717 | ref->in_tree = 1; | |
00f04b88 AJ |
718 | ref->seq = seq; |
719 | ||
5d4f98a2 | 720 | full_ref = btrfs_delayed_node_to_data_ref(ref); |
eebe063b AJ |
721 | full_ref->parent = parent; |
722 | full_ref->root = ref_root; | |
723 | if (parent) | |
5d4f98a2 | 724 | ref->type = BTRFS_SHARED_DATA_REF_KEY; |
eebe063b | 725 | else |
5d4f98a2 | 726 | ref->type = BTRFS_EXTENT_DATA_REF_KEY; |
66d7e7f0 | 727 | |
5d4f98a2 YZ |
728 | full_ref->objectid = owner; |
729 | full_ref->offset = offset; | |
56bec294 | 730 | |
599c75ec | 731 | trace_add_delayed_data_ref(ref, full_ref, action); |
1abe9b8a | 732 | |
c6fc2454 QW |
733 | ret = add_delayed_ref_tail_merge(trans, delayed_refs, head_ref, ref); |
734 | ||
735 | if (ret > 0) | |
78a6184a | 736 | kmem_cache_free(btrfs_delayed_data_ref_cachep, full_ref); |
56bec294 CM |
737 | } |
738 | ||
739 | /* | |
5d4f98a2 | 740 | * add a delayed tree ref. This does all of the accounting required |
56bec294 CM |
741 | * to make sure the delayed ref is eventually processed before this |
742 | * transaction commits. | |
743 | */ | |
66d7e7f0 AJ |
744 | int btrfs_add_delayed_tree_ref(struct btrfs_fs_info *fs_info, |
745 | struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
746 | u64 bytenr, u64 num_bytes, u64 parent, |
747 | u64 ref_root, int level, int action, | |
b06c4bf5 | 748 | struct btrfs_delayed_extent_op *extent_op) |
56bec294 | 749 | { |
5d4f98a2 | 750 | struct btrfs_delayed_tree_ref *ref; |
56bec294 CM |
751 | struct btrfs_delayed_ref_head *head_ref; |
752 | struct btrfs_delayed_ref_root *delayed_refs; | |
3368d001 | 753 | struct btrfs_qgroup_extent_record *record = NULL; |
56bec294 | 754 | |
5d4f98a2 | 755 | BUG_ON(extent_op && extent_op->is_data); |
78a6184a | 756 | ref = kmem_cache_alloc(btrfs_delayed_tree_ref_cachep, GFP_NOFS); |
56bec294 CM |
757 | if (!ref) |
758 | return -ENOMEM; | |
759 | ||
78a6184a | 760 | head_ref = kmem_cache_alloc(btrfs_delayed_ref_head_cachep, GFP_NOFS); |
5a5003df DC |
761 | if (!head_ref) |
762 | goto free_ref; | |
5d4f98a2 | 763 | |
3368d001 QW |
764 | if (fs_info->quota_enabled && is_fstree(ref_root)) { |
765 | record = kmalloc(sizeof(*record), GFP_NOFS); | |
5a5003df DC |
766 | if (!record) |
767 | goto free_head_ref; | |
3368d001 QW |
768 | } |
769 | ||
5d4f98a2 YZ |
770 | head_ref->extent_op = extent_op; |
771 | ||
772 | delayed_refs = &trans->transaction->delayed_refs; | |
773 | spin_lock(&delayed_refs->lock); | |
774 | ||
56bec294 | 775 | /* |
5d4f98a2 YZ |
776 | * insert both the head node and the new ref without dropping |
777 | * the spin lock | |
56bec294 | 778 | */ |
3368d001 | 779 | head_ref = add_delayed_ref_head(fs_info, trans, &head_ref->node, record, |
d7df2c79 | 780 | bytenr, num_bytes, action, 0); |
5d4f98a2 | 781 | |
d7df2c79 | 782 | add_delayed_tree_ref(fs_info, trans, head_ref, &ref->node, bytenr, |
b06c4bf5 | 783 | num_bytes, parent, ref_root, level, action); |
5d4f98a2 | 784 | spin_unlock(&delayed_refs->lock); |
95a06077 | 785 | |
5d4f98a2 | 786 | return 0; |
5a5003df DC |
787 | |
788 | free_head_ref: | |
789 | kmem_cache_free(btrfs_delayed_ref_head_cachep, head_ref); | |
790 | free_ref: | |
791 | kmem_cache_free(btrfs_delayed_tree_ref_cachep, ref); | |
792 | ||
793 | return -ENOMEM; | |
5d4f98a2 YZ |
794 | } |
795 | ||
796 | /* | |
797 | * add a delayed data ref. it's similar to btrfs_add_delayed_tree_ref. | |
798 | */ | |
66d7e7f0 AJ |
799 | int btrfs_add_delayed_data_ref(struct btrfs_fs_info *fs_info, |
800 | struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
801 | u64 bytenr, u64 num_bytes, |
802 | u64 parent, u64 ref_root, | |
803 | u64 owner, u64 offset, int action, | |
b06c4bf5 | 804 | struct btrfs_delayed_extent_op *extent_op) |
5d4f98a2 YZ |
805 | { |
806 | struct btrfs_delayed_data_ref *ref; | |
807 | struct btrfs_delayed_ref_head *head_ref; | |
808 | struct btrfs_delayed_ref_root *delayed_refs; | |
3368d001 | 809 | struct btrfs_qgroup_extent_record *record = NULL; |
5d4f98a2 YZ |
810 | |
811 | BUG_ON(extent_op && !extent_op->is_data); | |
78a6184a | 812 | ref = kmem_cache_alloc(btrfs_delayed_data_ref_cachep, GFP_NOFS); |
5d4f98a2 YZ |
813 | if (!ref) |
814 | return -ENOMEM; | |
56bec294 | 815 | |
78a6184a | 816 | head_ref = kmem_cache_alloc(btrfs_delayed_ref_head_cachep, GFP_NOFS); |
56bec294 | 817 | if (!head_ref) { |
78a6184a | 818 | kmem_cache_free(btrfs_delayed_data_ref_cachep, ref); |
56bec294 CM |
819 | return -ENOMEM; |
820 | } | |
5d4f98a2 | 821 | |
3368d001 QW |
822 | if (fs_info->quota_enabled && is_fstree(ref_root)) { |
823 | record = kmalloc(sizeof(*record), GFP_NOFS); | |
824 | if (!record) { | |
825 | kmem_cache_free(btrfs_delayed_data_ref_cachep, ref); | |
826 | kmem_cache_free(btrfs_delayed_ref_head_cachep, | |
827 | head_ref); | |
828 | return -ENOMEM; | |
829 | } | |
830 | } | |
831 | ||
5d4f98a2 YZ |
832 | head_ref->extent_op = extent_op; |
833 | ||
56bec294 CM |
834 | delayed_refs = &trans->transaction->delayed_refs; |
835 | spin_lock(&delayed_refs->lock); | |
836 | ||
837 | /* | |
838 | * insert both the head node and the new ref without dropping | |
839 | * the spin lock | |
840 | */ | |
3368d001 | 841 | head_ref = add_delayed_ref_head(fs_info, trans, &head_ref->node, record, |
d7df2c79 | 842 | bytenr, num_bytes, action, 1); |
56bec294 | 843 | |
d7df2c79 | 844 | add_delayed_data_ref(fs_info, trans, head_ref, &ref->node, bytenr, |
66d7e7f0 | 845 | num_bytes, parent, ref_root, owner, offset, |
b06c4bf5 | 846 | action); |
5d4f98a2 | 847 | spin_unlock(&delayed_refs->lock); |
95a06077 | 848 | |
5d4f98a2 YZ |
849 | return 0; |
850 | } | |
851 | ||
f64d5ca8 QW |
852 | int btrfs_add_delayed_qgroup_reserve(struct btrfs_fs_info *fs_info, |
853 | struct btrfs_trans_handle *trans, | |
854 | u64 ref_root, u64 bytenr, u64 num_bytes) | |
855 | { | |
856 | struct btrfs_delayed_ref_root *delayed_refs; | |
857 | struct btrfs_delayed_ref_head *ref_head; | |
858 | int ret = 0; | |
859 | ||
860 | if (!fs_info->quota_enabled || !is_fstree(ref_root)) | |
861 | return 0; | |
862 | ||
863 | delayed_refs = &trans->transaction->delayed_refs; | |
864 | ||
865 | spin_lock(&delayed_refs->lock); | |
866 | ref_head = find_ref_head(&delayed_refs->href_root, bytenr, 0); | |
867 | if (!ref_head) { | |
868 | ret = -ENOENT; | |
869 | goto out; | |
870 | } | |
871 | WARN_ON(ref_head->qgroup_reserved || ref_head->qgroup_ref_root); | |
872 | ref_head->qgroup_ref_root = ref_root; | |
873 | ref_head->qgroup_reserved = num_bytes; | |
874 | out: | |
875 | spin_unlock(&delayed_refs->lock); | |
876 | return ret; | |
877 | } | |
878 | ||
66d7e7f0 AJ |
879 | int btrfs_add_delayed_extent_op(struct btrfs_fs_info *fs_info, |
880 | struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
881 | u64 bytenr, u64 num_bytes, |
882 | struct btrfs_delayed_extent_op *extent_op) | |
883 | { | |
884 | struct btrfs_delayed_ref_head *head_ref; | |
885 | struct btrfs_delayed_ref_root *delayed_refs; | |
5d4f98a2 | 886 | |
78a6184a | 887 | head_ref = kmem_cache_alloc(btrfs_delayed_ref_head_cachep, GFP_NOFS); |
5d4f98a2 YZ |
888 | if (!head_ref) |
889 | return -ENOMEM; | |
890 | ||
891 | head_ref->extent_op = extent_op; | |
892 | ||
893 | delayed_refs = &trans->transaction->delayed_refs; | |
894 | spin_lock(&delayed_refs->lock); | |
895 | ||
3368d001 QW |
896 | add_delayed_ref_head(fs_info, trans, &head_ref->node, NULL, bytenr, |
897 | num_bytes, BTRFS_UPDATE_DELAYED_HEAD, | |
898 | extent_op->is_data); | |
5d4f98a2 | 899 | |
56bec294 CM |
900 | spin_unlock(&delayed_refs->lock); |
901 | return 0; | |
902 | } | |
903 | ||
1887be66 CM |
904 | /* |
905 | * this does a simple search for the head node for a given extent. | |
906 | * It must be called with the delayed ref spinlock held, and it returns | |
907 | * the head node if any where found, or NULL if not. | |
908 | */ | |
909 | struct btrfs_delayed_ref_head * | |
910 | btrfs_find_delayed_ref_head(struct btrfs_trans_handle *trans, u64 bytenr) | |
911 | { | |
1887be66 CM |
912 | struct btrfs_delayed_ref_root *delayed_refs; |
913 | ||
914 | delayed_refs = &trans->transaction->delayed_refs; | |
85fdfdf6 | 915 | return find_ref_head(&delayed_refs->href_root, bytenr, 0); |
1887be66 | 916 | } |
78a6184a MX |
917 | |
918 | void btrfs_delayed_ref_exit(void) | |
919 | { | |
920 | if (btrfs_delayed_ref_head_cachep) | |
921 | kmem_cache_destroy(btrfs_delayed_ref_head_cachep); | |
922 | if (btrfs_delayed_tree_ref_cachep) | |
923 | kmem_cache_destroy(btrfs_delayed_tree_ref_cachep); | |
924 | if (btrfs_delayed_data_ref_cachep) | |
925 | kmem_cache_destroy(btrfs_delayed_data_ref_cachep); | |
926 | if (btrfs_delayed_extent_op_cachep) | |
927 | kmem_cache_destroy(btrfs_delayed_extent_op_cachep); | |
928 | } | |
929 | ||
930 | int btrfs_delayed_ref_init(void) | |
931 | { | |
932 | btrfs_delayed_ref_head_cachep = kmem_cache_create( | |
933 | "btrfs_delayed_ref_head", | |
934 | sizeof(struct btrfs_delayed_ref_head), 0, | |
935 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
936 | if (!btrfs_delayed_ref_head_cachep) | |
937 | goto fail; | |
938 | ||
939 | btrfs_delayed_tree_ref_cachep = kmem_cache_create( | |
940 | "btrfs_delayed_tree_ref", | |
941 | sizeof(struct btrfs_delayed_tree_ref), 0, | |
942 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
943 | if (!btrfs_delayed_tree_ref_cachep) | |
944 | goto fail; | |
945 | ||
946 | btrfs_delayed_data_ref_cachep = kmem_cache_create( | |
947 | "btrfs_delayed_data_ref", | |
948 | sizeof(struct btrfs_delayed_data_ref), 0, | |
949 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
950 | if (!btrfs_delayed_data_ref_cachep) | |
951 | goto fail; | |
952 | ||
953 | btrfs_delayed_extent_op_cachep = kmem_cache_create( | |
954 | "btrfs_delayed_extent_op", | |
955 | sizeof(struct btrfs_delayed_extent_op), 0, | |
956 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
957 | if (!btrfs_delayed_extent_op_cachep) | |
958 | goto fail; | |
959 | ||
960 | return 0; | |
961 | fail: | |
962 | btrfs_delayed_ref_exit(); | |
963 | return -ENOMEM; | |
964 | } |