]>
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" | |
25 | ||
78a6184a MX |
26 | struct kmem_cache *btrfs_delayed_ref_head_cachep; |
27 | struct kmem_cache *btrfs_delayed_tree_ref_cachep; | |
28 | struct kmem_cache *btrfs_delayed_data_ref_cachep; | |
29 | struct kmem_cache *btrfs_delayed_extent_op_cachep; | |
56bec294 CM |
30 | /* |
31 | * delayed back reference update tracking. For subvolume trees | |
32 | * we queue up extent allocations and backref maintenance for | |
33 | * delayed processing. This avoids deep call chains where we | |
34 | * add extents in the middle of btrfs_search_slot, and it allows | |
35 | * us to buffer up frequently modified backrefs in an rb tree instead | |
36 | * of hammering updates on the extent allocation tree. | |
56bec294 CM |
37 | */ |
38 | ||
39 | /* | |
5d4f98a2 YZ |
40 | * compare two delayed tree backrefs with same bytenr and type |
41 | */ | |
42 | static int comp_tree_refs(struct btrfs_delayed_tree_ref *ref2, | |
41b0fc42 | 43 | struct btrfs_delayed_tree_ref *ref1, int type) |
5d4f98a2 | 44 | { |
41b0fc42 JB |
45 | if (type == BTRFS_TREE_BLOCK_REF_KEY) { |
46 | if (ref1->root < ref2->root) | |
47 | return -1; | |
48 | if (ref1->root > ref2->root) | |
49 | return 1; | |
50 | } else { | |
51 | if (ref1->parent < ref2->parent) | |
52 | return -1; | |
53 | if (ref1->parent > ref2->parent) | |
54 | return 1; | |
55 | } | |
5d4f98a2 YZ |
56 | return 0; |
57 | } | |
58 | ||
59 | /* | |
60 | * compare two delayed data backrefs with same bytenr and type | |
56bec294 | 61 | */ |
5d4f98a2 YZ |
62 | static int comp_data_refs(struct btrfs_delayed_data_ref *ref2, |
63 | struct btrfs_delayed_data_ref *ref1) | |
56bec294 | 64 | { |
5d4f98a2 YZ |
65 | if (ref1->node.type == BTRFS_EXTENT_DATA_REF_KEY) { |
66 | if (ref1->root < ref2->root) | |
67 | return -1; | |
68 | if (ref1->root > ref2->root) | |
69 | return 1; | |
70 | if (ref1->objectid < ref2->objectid) | |
71 | return -1; | |
72 | if (ref1->objectid > ref2->objectid) | |
73 | return 1; | |
74 | if (ref1->offset < ref2->offset) | |
75 | return -1; | |
76 | if (ref1->offset > ref2->offset) | |
77 | return 1; | |
78 | } else { | |
79 | if (ref1->parent < ref2->parent) | |
80 | return -1; | |
81 | if (ref1->parent > ref2->parent) | |
82 | return 1; | |
83 | } | |
84 | return 0; | |
85 | } | |
86 | ||
87 | /* | |
88 | * entries in the rb tree are ordered by the byte number of the extent, | |
89 | * type of the delayed backrefs and content of delayed backrefs. | |
90 | */ | |
91 | static int comp_entry(struct btrfs_delayed_ref_node *ref2, | |
ae1e206b JB |
92 | struct btrfs_delayed_ref_node *ref1, |
93 | bool compare_seq) | |
5d4f98a2 YZ |
94 | { |
95 | if (ref1->bytenr < ref2->bytenr) | |
56bec294 | 96 | return -1; |
5d4f98a2 | 97 | if (ref1->bytenr > ref2->bytenr) |
56bec294 | 98 | return 1; |
5d4f98a2 YZ |
99 | if (ref1->is_head && ref2->is_head) |
100 | return 0; | |
101 | if (ref2->is_head) | |
56bec294 | 102 | return -1; |
5d4f98a2 | 103 | if (ref1->is_head) |
56bec294 | 104 | return 1; |
5d4f98a2 YZ |
105 | if (ref1->type < ref2->type) |
106 | return -1; | |
107 | if (ref1->type > ref2->type) | |
108 | return 1; | |
00f04b88 | 109 | /* merging of sequenced refs is not allowed */ |
ae1e206b JB |
110 | if (compare_seq) { |
111 | if (ref1->seq < ref2->seq) | |
112 | return -1; | |
113 | if (ref1->seq > ref2->seq) | |
114 | return 1; | |
115 | } | |
5d4f98a2 YZ |
116 | if (ref1->type == BTRFS_TREE_BLOCK_REF_KEY || |
117 | ref1->type == BTRFS_SHARED_BLOCK_REF_KEY) { | |
118 | return comp_tree_refs(btrfs_delayed_node_to_tree_ref(ref2), | |
41b0fc42 JB |
119 | btrfs_delayed_node_to_tree_ref(ref1), |
120 | ref1->type); | |
5d4f98a2 YZ |
121 | } else if (ref1->type == BTRFS_EXTENT_DATA_REF_KEY || |
122 | ref1->type == BTRFS_SHARED_DATA_REF_KEY) { | |
123 | return comp_data_refs(btrfs_delayed_node_to_data_ref(ref2), | |
124 | btrfs_delayed_node_to_data_ref(ref1)); | |
125 | } | |
126 | BUG(); | |
56bec294 CM |
127 | return 0; |
128 | } | |
129 | ||
130 | /* | |
131 | * insert a new ref into the rbtree. This returns any existing refs | |
132 | * for the same (bytenr,parent) tuple, or NULL if the new node was properly | |
133 | * inserted. | |
134 | */ | |
135 | static struct btrfs_delayed_ref_node *tree_insert(struct rb_root *root, | |
56bec294 CM |
136 | struct rb_node *node) |
137 | { | |
138 | struct rb_node **p = &root->rb_node; | |
139 | struct rb_node *parent_node = NULL; | |
140 | struct btrfs_delayed_ref_node *entry; | |
5d4f98a2 | 141 | struct btrfs_delayed_ref_node *ins; |
56bec294 CM |
142 | int cmp; |
143 | ||
5d4f98a2 | 144 | ins = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); |
56bec294 CM |
145 | while (*p) { |
146 | parent_node = *p; | |
147 | entry = rb_entry(parent_node, struct btrfs_delayed_ref_node, | |
148 | rb_node); | |
149 | ||
ae1e206b | 150 | cmp = comp_entry(entry, ins, 1); |
56bec294 CM |
151 | if (cmp < 0) |
152 | p = &(*p)->rb_left; | |
153 | else if (cmp > 0) | |
154 | p = &(*p)->rb_right; | |
155 | else | |
156 | return entry; | |
157 | } | |
158 | ||
56bec294 CM |
159 | rb_link_node(node, parent_node, p); |
160 | rb_insert_color(node, root); | |
161 | return NULL; | |
162 | } | |
163 | ||
c46effa6 LB |
164 | /* insert a new ref to head ref rbtree */ |
165 | static struct btrfs_delayed_ref_head *htree_insert(struct rb_root *root, | |
166 | struct rb_node *node) | |
167 | { | |
168 | struct rb_node **p = &root->rb_node; | |
169 | struct rb_node *parent_node = NULL; | |
170 | struct btrfs_delayed_ref_head *entry; | |
171 | struct btrfs_delayed_ref_head *ins; | |
172 | u64 bytenr; | |
173 | ||
174 | ins = rb_entry(node, struct btrfs_delayed_ref_head, href_node); | |
175 | bytenr = ins->node.bytenr; | |
176 | while (*p) { | |
177 | parent_node = *p; | |
178 | entry = rb_entry(parent_node, struct btrfs_delayed_ref_head, | |
179 | href_node); | |
180 | ||
181 | if (bytenr < entry->node.bytenr) | |
182 | p = &(*p)->rb_left; | |
183 | else if (bytenr > entry->node.bytenr) | |
184 | p = &(*p)->rb_right; | |
185 | else | |
186 | return entry; | |
187 | } | |
188 | ||
189 | rb_link_node(node, parent_node, p); | |
190 | rb_insert_color(node, root); | |
191 | return NULL; | |
192 | } | |
193 | ||
56bec294 | 194 | /* |
5d4f98a2 | 195 | * find an head entry based on bytenr. This returns the delayed ref |
d1270cd9 AJ |
196 | * head if it was able to find one, or NULL if nothing was in that spot. |
197 | * If return_bigger is given, the next bigger entry is returned if no exact | |
198 | * match is found. | |
56bec294 | 199 | */ |
c46effa6 LB |
200 | static struct btrfs_delayed_ref_head * |
201 | find_ref_head(struct rb_root *root, u64 bytenr, | |
202 | struct btrfs_delayed_ref_head **last, int return_bigger) | |
56bec294 | 203 | { |
d1270cd9 | 204 | struct rb_node *n; |
c46effa6 | 205 | struct btrfs_delayed_ref_head *entry; |
d1270cd9 | 206 | int cmp = 0; |
56bec294 | 207 | |
d1270cd9 AJ |
208 | again: |
209 | n = root->rb_node; | |
210 | entry = NULL; | |
56bec294 | 211 | while (n) { |
c46effa6 | 212 | entry = rb_entry(n, struct btrfs_delayed_ref_head, href_node); |
c3e69d58 CM |
213 | if (last) |
214 | *last = entry; | |
56bec294 | 215 | |
c46effa6 | 216 | if (bytenr < entry->node.bytenr) |
5d4f98a2 | 217 | cmp = -1; |
c46effa6 | 218 | else if (bytenr > entry->node.bytenr) |
5d4f98a2 YZ |
219 | cmp = 1; |
220 | else | |
221 | cmp = 0; | |
222 | ||
56bec294 CM |
223 | if (cmp < 0) |
224 | n = n->rb_left; | |
225 | else if (cmp > 0) | |
226 | n = n->rb_right; | |
227 | else | |
228 | return entry; | |
229 | } | |
d1270cd9 AJ |
230 | if (entry && return_bigger) { |
231 | if (cmp > 0) { | |
c46effa6 | 232 | n = rb_next(&entry->href_node); |
d1270cd9 AJ |
233 | if (!n) |
234 | n = rb_first(root); | |
c46effa6 LB |
235 | entry = rb_entry(n, struct btrfs_delayed_ref_head, |
236 | href_node); | |
237 | bytenr = entry->node.bytenr; | |
d1270cd9 AJ |
238 | return_bigger = 0; |
239 | goto again; | |
240 | } | |
241 | return entry; | |
242 | } | |
56bec294 CM |
243 | return NULL; |
244 | } | |
245 | ||
c3e69d58 CM |
246 | int btrfs_delayed_ref_lock(struct btrfs_trans_handle *trans, |
247 | struct btrfs_delayed_ref_head *head) | |
56bec294 | 248 | { |
c3e69d58 CM |
249 | struct btrfs_delayed_ref_root *delayed_refs; |
250 | ||
251 | delayed_refs = &trans->transaction->delayed_refs; | |
252 | assert_spin_locked(&delayed_refs->lock); | |
253 | if (mutex_trylock(&head->mutex)) | |
254 | return 0; | |
255 | ||
256 | atomic_inc(&head->node.refs); | |
257 | spin_unlock(&delayed_refs->lock); | |
258 | ||
259 | mutex_lock(&head->mutex); | |
260 | spin_lock(&delayed_refs->lock); | |
261 | if (!head->node.in_tree) { | |
262 | mutex_unlock(&head->mutex); | |
263 | btrfs_put_delayed_ref(&head->node); | |
264 | return -EAGAIN; | |
265 | } | |
266 | btrfs_put_delayed_ref(&head->node); | |
267 | return 0; | |
268 | } | |
269 | ||
35a3621b | 270 | static inline void drop_delayed_ref(struct btrfs_trans_handle *trans, |
ae1e206b JB |
271 | struct btrfs_delayed_ref_root *delayed_refs, |
272 | struct btrfs_delayed_ref_node *ref) | |
273 | { | |
274 | rb_erase(&ref->rb_node, &delayed_refs->root); | |
c46effa6 LB |
275 | if (btrfs_delayed_ref_is_head(ref)) { |
276 | struct btrfs_delayed_ref_head *head; | |
277 | ||
278 | head = btrfs_delayed_node_to_head(ref); | |
279 | rb_erase(&head->href_node, &delayed_refs->href_root); | |
280 | } | |
ae1e206b JB |
281 | ref->in_tree = 0; |
282 | btrfs_put_delayed_ref(ref); | |
283 | delayed_refs->num_entries--; | |
284 | if (trans->delayed_ref_updates) | |
285 | trans->delayed_ref_updates--; | |
286 | } | |
287 | ||
288 | static int merge_ref(struct btrfs_trans_handle *trans, | |
289 | struct btrfs_delayed_ref_root *delayed_refs, | |
290 | struct btrfs_delayed_ref_node *ref, u64 seq) | |
291 | { | |
292 | struct rb_node *node; | |
293 | int merged = 0; | |
294 | int mod = 0; | |
295 | int done = 0; | |
296 | ||
297 | node = rb_prev(&ref->rb_node); | |
298 | while (node) { | |
299 | struct btrfs_delayed_ref_node *next; | |
300 | ||
301 | next = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); | |
302 | node = rb_prev(node); | |
303 | if (next->bytenr != ref->bytenr) | |
304 | break; | |
305 | if (seq && next->seq >= seq) | |
306 | break; | |
307 | if (comp_entry(ref, next, 0)) | |
308 | continue; | |
309 | ||
310 | if (ref->action == next->action) { | |
311 | mod = next->ref_mod; | |
312 | } else { | |
313 | if (ref->ref_mod < next->ref_mod) { | |
314 | struct btrfs_delayed_ref_node *tmp; | |
315 | ||
316 | tmp = ref; | |
317 | ref = next; | |
318 | next = tmp; | |
319 | done = 1; | |
320 | } | |
321 | mod = -next->ref_mod; | |
322 | } | |
323 | ||
324 | merged++; | |
325 | drop_delayed_ref(trans, delayed_refs, next); | |
326 | ref->ref_mod += mod; | |
327 | if (ref->ref_mod == 0) { | |
328 | drop_delayed_ref(trans, delayed_refs, ref); | |
329 | break; | |
330 | } else { | |
331 | /* | |
332 | * You can't have multiples of the same ref on a tree | |
333 | * block. | |
334 | */ | |
335 | WARN_ON(ref->type == BTRFS_TREE_BLOCK_REF_KEY || | |
336 | ref->type == BTRFS_SHARED_BLOCK_REF_KEY); | |
337 | } | |
338 | ||
339 | if (done) | |
340 | break; | |
341 | node = rb_prev(&ref->rb_node); | |
342 | } | |
343 | ||
344 | return merged; | |
345 | } | |
346 | ||
347 | void btrfs_merge_delayed_refs(struct btrfs_trans_handle *trans, | |
348 | struct btrfs_fs_info *fs_info, | |
349 | struct btrfs_delayed_ref_root *delayed_refs, | |
350 | struct btrfs_delayed_ref_head *head) | |
351 | { | |
352 | struct rb_node *node; | |
353 | u64 seq = 0; | |
354 | ||
9e5ac13a LB |
355 | /* |
356 | * We don't have too much refs to merge in the case of delayed data | |
357 | * refs. | |
358 | */ | |
359 | if (head->is_data) | |
360 | return; | |
361 | ||
ae1e206b JB |
362 | spin_lock(&fs_info->tree_mod_seq_lock); |
363 | if (!list_empty(&fs_info->tree_mod_seq_list)) { | |
364 | struct seq_list *elem; | |
365 | ||
366 | elem = list_first_entry(&fs_info->tree_mod_seq_list, | |
367 | struct seq_list, list); | |
368 | seq = elem->seq; | |
369 | } | |
370 | spin_unlock(&fs_info->tree_mod_seq_lock); | |
371 | ||
372 | node = rb_prev(&head->node.rb_node); | |
373 | while (node) { | |
374 | struct btrfs_delayed_ref_node *ref; | |
375 | ||
376 | ref = rb_entry(node, struct btrfs_delayed_ref_node, | |
377 | rb_node); | |
378 | if (ref->bytenr != head->node.bytenr) | |
379 | break; | |
380 | ||
381 | /* We can't merge refs that are outside of our seq count */ | |
382 | if (seq && ref->seq >= seq) | |
383 | break; | |
384 | if (merge_ref(trans, delayed_refs, ref, seq)) | |
385 | node = rb_prev(&head->node.rb_node); | |
386 | else | |
387 | node = rb_prev(node); | |
388 | } | |
389 | } | |
390 | ||
097b8a7c JS |
391 | int btrfs_check_delayed_seq(struct btrfs_fs_info *fs_info, |
392 | struct btrfs_delayed_ref_root *delayed_refs, | |
00f04b88 AJ |
393 | u64 seq) |
394 | { | |
395 | struct seq_list *elem; | |
097b8a7c JS |
396 | int ret = 0; |
397 | ||
398 | spin_lock(&fs_info->tree_mod_seq_lock); | |
399 | if (!list_empty(&fs_info->tree_mod_seq_list)) { | |
400 | elem = list_first_entry(&fs_info->tree_mod_seq_list, | |
401 | struct seq_list, list); | |
402 | if (seq >= elem->seq) { | |
fc36ed7e JS |
403 | pr_debug("holding back delayed_ref %#x.%x, lowest is %#x.%x (%p)\n", |
404 | (u32)(seq >> 32), (u32)seq, | |
405 | (u32)(elem->seq >> 32), (u32)elem->seq, | |
406 | delayed_refs); | |
097b8a7c JS |
407 | ret = 1; |
408 | } | |
00f04b88 | 409 | } |
097b8a7c JS |
410 | |
411 | spin_unlock(&fs_info->tree_mod_seq_lock); | |
412 | return ret; | |
00f04b88 AJ |
413 | } |
414 | ||
c3e69d58 CM |
415 | int btrfs_find_ref_cluster(struct btrfs_trans_handle *trans, |
416 | struct list_head *cluster, u64 start) | |
417 | { | |
418 | int count = 0; | |
419 | struct btrfs_delayed_ref_root *delayed_refs; | |
56bec294 | 420 | struct rb_node *node; |
c46effa6 | 421 | struct btrfs_delayed_ref_head *head = NULL; |
56bec294 | 422 | |
c3e69d58 | 423 | delayed_refs = &trans->transaction->delayed_refs; |
c46effa6 LB |
424 | node = rb_first(&delayed_refs->href_root); |
425 | ||
426 | if (start) { | |
427 | find_ref_head(&delayed_refs->href_root, start + 1, &head, 1); | |
428 | if (head) | |
429 | node = &head->href_node; | |
c3e69d58 CM |
430 | } |
431 | again: | |
432 | while (node && count < 32) { | |
c46effa6 LB |
433 | head = rb_entry(node, struct btrfs_delayed_ref_head, href_node); |
434 | if (list_empty(&head->cluster)) { | |
435 | list_add_tail(&head->cluster, cluster); | |
436 | delayed_refs->run_delayed_start = | |
437 | head->node.bytenr; | |
438 | count++; | |
439 | ||
440 | WARN_ON(delayed_refs->num_heads_ready == 0); | |
441 | delayed_refs->num_heads_ready--; | |
442 | } else if (count) { | |
443 | /* the goal of the clustering is to find extents | |
444 | * that are likely to end up in the same extent | |
445 | * leaf on disk. So, we don't want them spread | |
446 | * all over the tree. Stop now if we've hit | |
447 | * a head that was already in use | |
448 | */ | |
449 | break; | |
56bec294 | 450 | } |
c3e69d58 | 451 | node = rb_next(node); |
56bec294 | 452 | } |
c3e69d58 CM |
453 | if (count) { |
454 | return 0; | |
455 | } else if (start) { | |
456 | /* | |
457 | * we've gone to the end of the rbtree without finding any | |
458 | * clusters. start from the beginning and try again | |
459 | */ | |
460 | start = 0; | |
c46effa6 | 461 | node = rb_first(&delayed_refs->href_root); |
c3e69d58 CM |
462 | goto again; |
463 | } | |
464 | return 1; | |
56bec294 CM |
465 | } |
466 | ||
093486c4 MX |
467 | void btrfs_release_ref_cluster(struct list_head *cluster) |
468 | { | |
469 | struct list_head *pos, *q; | |
470 | ||
471 | list_for_each_safe(pos, q, cluster) | |
472 | list_del_init(pos); | |
473 | } | |
474 | ||
56bec294 CM |
475 | /* |
476 | * helper function to update an extent delayed ref in the | |
477 | * rbtree. existing and update must both have the same | |
478 | * bytenr and parent | |
479 | * | |
480 | * This may free existing if the update cancels out whatever | |
481 | * operation it was doing. | |
482 | */ | |
483 | static noinline void | |
484 | update_existing_ref(struct btrfs_trans_handle *trans, | |
485 | struct btrfs_delayed_ref_root *delayed_refs, | |
486 | struct btrfs_delayed_ref_node *existing, | |
487 | struct btrfs_delayed_ref_node *update) | |
488 | { | |
5d4f98a2 | 489 | if (update->action != existing->action) { |
56bec294 CM |
490 | /* |
491 | * this is effectively undoing either an add or a | |
492 | * drop. We decrement the ref_mod, and if it goes | |
493 | * down to zero we just delete the entry without | |
494 | * every changing the extent allocation tree. | |
495 | */ | |
496 | existing->ref_mod--; | |
ae1e206b JB |
497 | if (existing->ref_mod == 0) |
498 | drop_delayed_ref(trans, delayed_refs, existing); | |
499 | else | |
5d4f98a2 YZ |
500 | WARN_ON(existing->type == BTRFS_TREE_BLOCK_REF_KEY || |
501 | existing->type == BTRFS_SHARED_BLOCK_REF_KEY); | |
56bec294 | 502 | } else { |
5d4f98a2 YZ |
503 | WARN_ON(existing->type == BTRFS_TREE_BLOCK_REF_KEY || |
504 | existing->type == BTRFS_SHARED_BLOCK_REF_KEY); | |
56bec294 CM |
505 | /* |
506 | * the action on the existing ref matches | |
507 | * the action on the ref we're trying to add. | |
508 | * Bump the ref_mod by one so the backref that | |
509 | * is eventually added/removed has the correct | |
510 | * reference count | |
511 | */ | |
512 | existing->ref_mod += update->ref_mod; | |
513 | } | |
514 | } | |
515 | ||
516 | /* | |
517 | * helper function to update the accounting in the head ref | |
518 | * existing and update must have the same bytenr | |
519 | */ | |
520 | static noinline void | |
521 | update_existing_head_ref(struct btrfs_delayed_ref_node *existing, | |
522 | struct btrfs_delayed_ref_node *update) | |
523 | { | |
524 | struct btrfs_delayed_ref_head *existing_ref; | |
525 | struct btrfs_delayed_ref_head *ref; | |
526 | ||
527 | existing_ref = btrfs_delayed_node_to_head(existing); | |
528 | ref = btrfs_delayed_node_to_head(update); | |
5d4f98a2 | 529 | BUG_ON(existing_ref->is_data != ref->is_data); |
56bec294 CM |
530 | |
531 | if (ref->must_insert_reserved) { | |
532 | /* if the extent was freed and then | |
533 | * reallocated before the delayed ref | |
534 | * entries were processed, we can end up | |
535 | * with an existing head ref without | |
536 | * the must_insert_reserved flag set. | |
537 | * Set it again here | |
538 | */ | |
539 | existing_ref->must_insert_reserved = ref->must_insert_reserved; | |
540 | ||
541 | /* | |
542 | * update the num_bytes so we make sure the accounting | |
543 | * is done correctly | |
544 | */ | |
545 | existing->num_bytes = update->num_bytes; | |
546 | ||
547 | } | |
548 | ||
5d4f98a2 YZ |
549 | if (ref->extent_op) { |
550 | if (!existing_ref->extent_op) { | |
551 | existing_ref->extent_op = ref->extent_op; | |
552 | } else { | |
553 | if (ref->extent_op->update_key) { | |
554 | memcpy(&existing_ref->extent_op->key, | |
555 | &ref->extent_op->key, | |
556 | sizeof(ref->extent_op->key)); | |
557 | existing_ref->extent_op->update_key = 1; | |
558 | } | |
559 | if (ref->extent_op->update_flags) { | |
560 | existing_ref->extent_op->flags_to_set |= | |
561 | ref->extent_op->flags_to_set; | |
562 | existing_ref->extent_op->update_flags = 1; | |
563 | } | |
78a6184a | 564 | btrfs_free_delayed_extent_op(ref->extent_op); |
5d4f98a2 YZ |
565 | } |
566 | } | |
56bec294 CM |
567 | /* |
568 | * update the reference mod on the head to reflect this new operation | |
569 | */ | |
570 | existing->ref_mod += update->ref_mod; | |
571 | } | |
572 | ||
573 | /* | |
5d4f98a2 | 574 | * helper function to actually insert a head node into the rbtree. |
56bec294 | 575 | * this does all the dirty work in terms of maintaining the correct |
5d4f98a2 | 576 | * overall modification count. |
56bec294 | 577 | */ |
143bede5 | 578 | static noinline void add_delayed_ref_head(struct btrfs_fs_info *fs_info, |
66d7e7f0 | 579 | struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
580 | struct btrfs_delayed_ref_node *ref, |
581 | u64 bytenr, u64 num_bytes, | |
582 | int action, int is_data) | |
56bec294 CM |
583 | { |
584 | struct btrfs_delayed_ref_node *existing; | |
c3e69d58 | 585 | struct btrfs_delayed_ref_head *head_ref = NULL; |
56bec294 CM |
586 | struct btrfs_delayed_ref_root *delayed_refs; |
587 | int count_mod = 1; | |
588 | int must_insert_reserved = 0; | |
589 | ||
590 | /* | |
591 | * the head node stores the sum of all the mods, so dropping a ref | |
592 | * should drop the sum in the head node by one. | |
593 | */ | |
5d4f98a2 YZ |
594 | if (action == BTRFS_UPDATE_DELAYED_HEAD) |
595 | count_mod = 0; | |
596 | else if (action == BTRFS_DROP_DELAYED_REF) | |
597 | count_mod = -1; | |
56bec294 CM |
598 | |
599 | /* | |
600 | * BTRFS_ADD_DELAYED_EXTENT means that we need to update | |
601 | * the reserved accounting when the extent is finally added, or | |
602 | * if a later modification deletes the delayed ref without ever | |
603 | * inserting the extent into the extent allocation tree. | |
604 | * ref->must_insert_reserved is the flag used to record | |
605 | * that accounting mods are required. | |
606 | * | |
607 | * Once we record must_insert_reserved, switch the action to | |
608 | * BTRFS_ADD_DELAYED_REF because other special casing is not required. | |
609 | */ | |
5d4f98a2 | 610 | if (action == BTRFS_ADD_DELAYED_EXTENT) |
56bec294 | 611 | must_insert_reserved = 1; |
5d4f98a2 | 612 | else |
56bec294 | 613 | must_insert_reserved = 0; |
56bec294 CM |
614 | |
615 | delayed_refs = &trans->transaction->delayed_refs; | |
616 | ||
617 | /* first set the basic ref node struct up */ | |
618 | atomic_set(&ref->refs, 1); | |
619 | ref->bytenr = bytenr; | |
5d4f98a2 | 620 | ref->num_bytes = num_bytes; |
56bec294 | 621 | ref->ref_mod = count_mod; |
5d4f98a2 YZ |
622 | ref->type = 0; |
623 | ref->action = 0; | |
624 | ref->is_head = 1; | |
56bec294 | 625 | ref->in_tree = 1; |
00f04b88 | 626 | ref->seq = 0; |
5d4f98a2 YZ |
627 | |
628 | head_ref = btrfs_delayed_node_to_head(ref); | |
629 | head_ref->must_insert_reserved = must_insert_reserved; | |
630 | head_ref->is_data = is_data; | |
631 | ||
632 | INIT_LIST_HEAD(&head_ref->cluster); | |
633 | mutex_init(&head_ref->mutex); | |
634 | ||
599c75ec | 635 | trace_add_delayed_ref_head(ref, head_ref, action); |
1abe9b8a | 636 | |
5d4f98a2 YZ |
637 | existing = tree_insert(&delayed_refs->root, &ref->rb_node); |
638 | ||
639 | if (existing) { | |
640 | update_existing_head_ref(existing, ref); | |
641 | /* | |
642 | * we've updated the existing ref, free the newly | |
643 | * allocated ref | |
644 | */ | |
78a6184a | 645 | kmem_cache_free(btrfs_delayed_ref_head_cachep, head_ref); |
5d4f98a2 | 646 | } else { |
c46effa6 | 647 | htree_insert(&delayed_refs->href_root, &head_ref->href_node); |
5d4f98a2 YZ |
648 | delayed_refs->num_heads++; |
649 | delayed_refs->num_heads_ready++; | |
650 | delayed_refs->num_entries++; | |
651 | trans->delayed_ref_updates++; | |
652 | } | |
5d4f98a2 YZ |
653 | } |
654 | ||
655 | /* | |
656 | * helper to insert a delayed tree ref into the rbtree. | |
657 | */ | |
143bede5 | 658 | static noinline void add_delayed_tree_ref(struct btrfs_fs_info *fs_info, |
66d7e7f0 | 659 | struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
660 | struct btrfs_delayed_ref_node *ref, |
661 | u64 bytenr, u64 num_bytes, u64 parent, | |
66d7e7f0 AJ |
662 | u64 ref_root, int level, int action, |
663 | int for_cow) | |
5d4f98a2 YZ |
664 | { |
665 | struct btrfs_delayed_ref_node *existing; | |
666 | struct btrfs_delayed_tree_ref *full_ref; | |
667 | struct btrfs_delayed_ref_root *delayed_refs; | |
00f04b88 | 668 | u64 seq = 0; |
5d4f98a2 YZ |
669 | |
670 | if (action == BTRFS_ADD_DELAYED_EXTENT) | |
671 | action = BTRFS_ADD_DELAYED_REF; | |
672 | ||
673 | delayed_refs = &trans->transaction->delayed_refs; | |
674 | ||
675 | /* first set the basic ref node struct up */ | |
676 | atomic_set(&ref->refs, 1); | |
677 | ref->bytenr = bytenr; | |
56bec294 | 678 | ref->num_bytes = num_bytes; |
5d4f98a2 YZ |
679 | ref->ref_mod = 1; |
680 | ref->action = action; | |
681 | ref->is_head = 0; | |
682 | ref->in_tree = 1; | |
56bec294 | 683 | |
546adb0d JS |
684 | if (need_ref_seq(for_cow, ref_root)) |
685 | seq = btrfs_get_tree_mod_seq(fs_info, &trans->delayed_ref_elem); | |
00f04b88 AJ |
686 | ref->seq = seq; |
687 | ||
5d4f98a2 | 688 | full_ref = btrfs_delayed_node_to_tree_ref(ref); |
eebe063b AJ |
689 | full_ref->parent = parent; |
690 | full_ref->root = ref_root; | |
691 | if (parent) | |
5d4f98a2 | 692 | ref->type = BTRFS_SHARED_BLOCK_REF_KEY; |
eebe063b | 693 | else |
5d4f98a2 | 694 | ref->type = BTRFS_TREE_BLOCK_REF_KEY; |
5d4f98a2 | 695 | full_ref->level = level; |
56bec294 | 696 | |
599c75ec | 697 | trace_add_delayed_tree_ref(ref, full_ref, action); |
1abe9b8a | 698 | |
5d4f98a2 | 699 | existing = tree_insert(&delayed_refs->root, &ref->rb_node); |
56bec294 CM |
700 | |
701 | if (existing) { | |
5d4f98a2 YZ |
702 | update_existing_ref(trans, delayed_refs, existing, ref); |
703 | /* | |
704 | * we've updated the existing ref, free the newly | |
705 | * allocated ref | |
706 | */ | |
78a6184a | 707 | kmem_cache_free(btrfs_delayed_tree_ref_cachep, full_ref); |
5d4f98a2 YZ |
708 | } else { |
709 | delayed_refs->num_entries++; | |
710 | trans->delayed_ref_updates++; | |
711 | } | |
5d4f98a2 YZ |
712 | } |
713 | ||
714 | /* | |
715 | * helper to insert a delayed data ref into the rbtree. | |
716 | */ | |
143bede5 | 717 | static noinline void add_delayed_data_ref(struct btrfs_fs_info *fs_info, |
66d7e7f0 | 718 | struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
719 | struct btrfs_delayed_ref_node *ref, |
720 | u64 bytenr, u64 num_bytes, u64 parent, | |
721 | u64 ref_root, u64 owner, u64 offset, | |
66d7e7f0 | 722 | int action, int for_cow) |
5d4f98a2 YZ |
723 | { |
724 | struct btrfs_delayed_ref_node *existing; | |
725 | struct btrfs_delayed_data_ref *full_ref; | |
726 | struct btrfs_delayed_ref_root *delayed_refs; | |
00f04b88 | 727 | u64 seq = 0; |
5d4f98a2 YZ |
728 | |
729 | if (action == BTRFS_ADD_DELAYED_EXTENT) | |
730 | action = BTRFS_ADD_DELAYED_REF; | |
731 | ||
732 | delayed_refs = &trans->transaction->delayed_refs; | |
733 | ||
734 | /* first set the basic ref node struct up */ | |
735 | atomic_set(&ref->refs, 1); | |
736 | ref->bytenr = bytenr; | |
737 | ref->num_bytes = num_bytes; | |
738 | ref->ref_mod = 1; | |
739 | ref->action = action; | |
740 | ref->is_head = 0; | |
741 | ref->in_tree = 1; | |
742 | ||
546adb0d JS |
743 | if (need_ref_seq(for_cow, ref_root)) |
744 | seq = btrfs_get_tree_mod_seq(fs_info, &trans->delayed_ref_elem); | |
00f04b88 AJ |
745 | ref->seq = seq; |
746 | ||
5d4f98a2 | 747 | full_ref = btrfs_delayed_node_to_data_ref(ref); |
eebe063b AJ |
748 | full_ref->parent = parent; |
749 | full_ref->root = ref_root; | |
750 | if (parent) | |
5d4f98a2 | 751 | ref->type = BTRFS_SHARED_DATA_REF_KEY; |
eebe063b | 752 | else |
5d4f98a2 | 753 | ref->type = BTRFS_EXTENT_DATA_REF_KEY; |
66d7e7f0 | 754 | |
5d4f98a2 YZ |
755 | full_ref->objectid = owner; |
756 | full_ref->offset = offset; | |
56bec294 | 757 | |
599c75ec | 758 | trace_add_delayed_data_ref(ref, full_ref, action); |
1abe9b8a | 759 | |
5d4f98a2 YZ |
760 | existing = tree_insert(&delayed_refs->root, &ref->rb_node); |
761 | ||
762 | if (existing) { | |
763 | update_existing_ref(trans, delayed_refs, existing, ref); | |
56bec294 CM |
764 | /* |
765 | * we've updated the existing ref, free the newly | |
766 | * allocated ref | |
767 | */ | |
78a6184a | 768 | kmem_cache_free(btrfs_delayed_data_ref_cachep, full_ref); |
56bec294 CM |
769 | } else { |
770 | delayed_refs->num_entries++; | |
771 | trans->delayed_ref_updates++; | |
772 | } | |
56bec294 CM |
773 | } |
774 | ||
775 | /* | |
5d4f98a2 | 776 | * add a delayed tree ref. This does all of the accounting required |
56bec294 CM |
777 | * to make sure the delayed ref is eventually processed before this |
778 | * transaction commits. | |
779 | */ | |
66d7e7f0 AJ |
780 | int btrfs_add_delayed_tree_ref(struct btrfs_fs_info *fs_info, |
781 | struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
782 | u64 bytenr, u64 num_bytes, u64 parent, |
783 | u64 ref_root, int level, int action, | |
66d7e7f0 AJ |
784 | struct btrfs_delayed_extent_op *extent_op, |
785 | int for_cow) | |
56bec294 | 786 | { |
5d4f98a2 | 787 | struct btrfs_delayed_tree_ref *ref; |
56bec294 CM |
788 | struct btrfs_delayed_ref_head *head_ref; |
789 | struct btrfs_delayed_ref_root *delayed_refs; | |
56bec294 | 790 | |
5d4f98a2 | 791 | BUG_ON(extent_op && extent_op->is_data); |
78a6184a | 792 | ref = kmem_cache_alloc(btrfs_delayed_tree_ref_cachep, GFP_NOFS); |
56bec294 CM |
793 | if (!ref) |
794 | return -ENOMEM; | |
795 | ||
78a6184a | 796 | head_ref = kmem_cache_alloc(btrfs_delayed_ref_head_cachep, GFP_NOFS); |
5d4f98a2 | 797 | if (!head_ref) { |
78a6184a | 798 | kmem_cache_free(btrfs_delayed_tree_ref_cachep, ref); |
5d4f98a2 YZ |
799 | return -ENOMEM; |
800 | } | |
801 | ||
802 | head_ref->extent_op = extent_op; | |
803 | ||
804 | delayed_refs = &trans->transaction->delayed_refs; | |
805 | spin_lock(&delayed_refs->lock); | |
806 | ||
56bec294 | 807 | /* |
5d4f98a2 YZ |
808 | * insert both the head node and the new ref without dropping |
809 | * the spin lock | |
56bec294 | 810 | */ |
143bede5 | 811 | add_delayed_ref_head(fs_info, trans, &head_ref->node, bytenr, |
66d7e7f0 | 812 | num_bytes, action, 0); |
5d4f98a2 | 813 | |
143bede5 | 814 | add_delayed_tree_ref(fs_info, trans, &ref->node, bytenr, |
66d7e7f0 AJ |
815 | num_bytes, parent, ref_root, level, action, |
816 | for_cow); | |
5d4f98a2 | 817 | spin_unlock(&delayed_refs->lock); |
546adb0d JS |
818 | if (need_ref_seq(for_cow, ref_root)) |
819 | btrfs_qgroup_record_ref(trans, &ref->node, extent_op); | |
95a06077 | 820 | |
5d4f98a2 YZ |
821 | return 0; |
822 | } | |
823 | ||
824 | /* | |
825 | * add a delayed data ref. it's similar to btrfs_add_delayed_tree_ref. | |
826 | */ | |
66d7e7f0 AJ |
827 | int btrfs_add_delayed_data_ref(struct btrfs_fs_info *fs_info, |
828 | struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
829 | u64 bytenr, u64 num_bytes, |
830 | u64 parent, u64 ref_root, | |
831 | u64 owner, u64 offset, int action, | |
66d7e7f0 AJ |
832 | struct btrfs_delayed_extent_op *extent_op, |
833 | int for_cow) | |
5d4f98a2 YZ |
834 | { |
835 | struct btrfs_delayed_data_ref *ref; | |
836 | struct btrfs_delayed_ref_head *head_ref; | |
837 | struct btrfs_delayed_ref_root *delayed_refs; | |
5d4f98a2 YZ |
838 | |
839 | BUG_ON(extent_op && !extent_op->is_data); | |
78a6184a | 840 | ref = kmem_cache_alloc(btrfs_delayed_data_ref_cachep, GFP_NOFS); |
5d4f98a2 YZ |
841 | if (!ref) |
842 | return -ENOMEM; | |
56bec294 | 843 | |
78a6184a | 844 | head_ref = kmem_cache_alloc(btrfs_delayed_ref_head_cachep, GFP_NOFS); |
56bec294 | 845 | if (!head_ref) { |
78a6184a | 846 | kmem_cache_free(btrfs_delayed_data_ref_cachep, ref); |
56bec294 CM |
847 | return -ENOMEM; |
848 | } | |
5d4f98a2 YZ |
849 | |
850 | head_ref->extent_op = extent_op; | |
851 | ||
56bec294 CM |
852 | delayed_refs = &trans->transaction->delayed_refs; |
853 | spin_lock(&delayed_refs->lock); | |
854 | ||
855 | /* | |
856 | * insert both the head node and the new ref without dropping | |
857 | * the spin lock | |
858 | */ | |
143bede5 | 859 | add_delayed_ref_head(fs_info, trans, &head_ref->node, bytenr, |
66d7e7f0 | 860 | num_bytes, action, 1); |
56bec294 | 861 | |
143bede5 | 862 | add_delayed_data_ref(fs_info, trans, &ref->node, bytenr, |
66d7e7f0 AJ |
863 | num_bytes, parent, ref_root, owner, offset, |
864 | action, for_cow); | |
5d4f98a2 | 865 | spin_unlock(&delayed_refs->lock); |
546adb0d JS |
866 | if (need_ref_seq(for_cow, ref_root)) |
867 | btrfs_qgroup_record_ref(trans, &ref->node, extent_op); | |
95a06077 | 868 | |
5d4f98a2 YZ |
869 | return 0; |
870 | } | |
871 | ||
66d7e7f0 AJ |
872 | int btrfs_add_delayed_extent_op(struct btrfs_fs_info *fs_info, |
873 | struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
874 | u64 bytenr, u64 num_bytes, |
875 | struct btrfs_delayed_extent_op *extent_op) | |
876 | { | |
877 | struct btrfs_delayed_ref_head *head_ref; | |
878 | struct btrfs_delayed_ref_root *delayed_refs; | |
5d4f98a2 | 879 | |
78a6184a | 880 | head_ref = kmem_cache_alloc(btrfs_delayed_ref_head_cachep, GFP_NOFS); |
5d4f98a2 YZ |
881 | if (!head_ref) |
882 | return -ENOMEM; | |
883 | ||
884 | head_ref->extent_op = extent_op; | |
885 | ||
886 | delayed_refs = &trans->transaction->delayed_refs; | |
887 | spin_lock(&delayed_refs->lock); | |
888 | ||
143bede5 | 889 | add_delayed_ref_head(fs_info, trans, &head_ref->node, bytenr, |
5d4f98a2 YZ |
890 | num_bytes, BTRFS_UPDATE_DELAYED_HEAD, |
891 | extent_op->is_data); | |
5d4f98a2 | 892 | |
56bec294 CM |
893 | spin_unlock(&delayed_refs->lock); |
894 | return 0; | |
895 | } | |
896 | ||
1887be66 CM |
897 | /* |
898 | * this does a simple search for the head node for a given extent. | |
899 | * It must be called with the delayed ref spinlock held, and it returns | |
900 | * the head node if any where found, or NULL if not. | |
901 | */ | |
902 | struct btrfs_delayed_ref_head * | |
903 | btrfs_find_delayed_ref_head(struct btrfs_trans_handle *trans, u64 bytenr) | |
904 | { | |
1887be66 CM |
905 | struct btrfs_delayed_ref_root *delayed_refs; |
906 | ||
907 | delayed_refs = &trans->transaction->delayed_refs; | |
c46effa6 | 908 | return find_ref_head(&delayed_refs->href_root, bytenr, NULL, 0); |
1887be66 | 909 | } |
78a6184a MX |
910 | |
911 | void btrfs_delayed_ref_exit(void) | |
912 | { | |
913 | if (btrfs_delayed_ref_head_cachep) | |
914 | kmem_cache_destroy(btrfs_delayed_ref_head_cachep); | |
915 | if (btrfs_delayed_tree_ref_cachep) | |
916 | kmem_cache_destroy(btrfs_delayed_tree_ref_cachep); | |
917 | if (btrfs_delayed_data_ref_cachep) | |
918 | kmem_cache_destroy(btrfs_delayed_data_ref_cachep); | |
919 | if (btrfs_delayed_extent_op_cachep) | |
920 | kmem_cache_destroy(btrfs_delayed_extent_op_cachep); | |
921 | } | |
922 | ||
923 | int btrfs_delayed_ref_init(void) | |
924 | { | |
925 | btrfs_delayed_ref_head_cachep = kmem_cache_create( | |
926 | "btrfs_delayed_ref_head", | |
927 | sizeof(struct btrfs_delayed_ref_head), 0, | |
928 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
929 | if (!btrfs_delayed_ref_head_cachep) | |
930 | goto fail; | |
931 | ||
932 | btrfs_delayed_tree_ref_cachep = kmem_cache_create( | |
933 | "btrfs_delayed_tree_ref", | |
934 | sizeof(struct btrfs_delayed_tree_ref), 0, | |
935 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
936 | if (!btrfs_delayed_tree_ref_cachep) | |
937 | goto fail; | |
938 | ||
939 | btrfs_delayed_data_ref_cachep = kmem_cache_create( | |
940 | "btrfs_delayed_data_ref", | |
941 | sizeof(struct btrfs_delayed_data_ref), 0, | |
942 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
943 | if (!btrfs_delayed_data_ref_cachep) | |
944 | goto fail; | |
945 | ||
946 | btrfs_delayed_extent_op_cachep = kmem_cache_create( | |
947 | "btrfs_delayed_extent_op", | |
948 | sizeof(struct btrfs_delayed_extent_op), 0, | |
949 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
950 | if (!btrfs_delayed_extent_op_cachep) | |
951 | goto fail; | |
952 | ||
953 | return 0; | |
954 | fail: | |
955 | btrfs_delayed_ref_exit(); | |
956 | return -ENOMEM; | |
957 | } |