]>
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 | ||
26 | /* | |
27 | * delayed back reference update tracking. For subvolume trees | |
28 | * we queue up extent allocations and backref maintenance for | |
29 | * delayed processing. This avoids deep call chains where we | |
30 | * add extents in the middle of btrfs_search_slot, and it allows | |
31 | * us to buffer up frequently modified backrefs in an rb tree instead | |
32 | * of hammering updates on the extent allocation tree. | |
56bec294 CM |
33 | */ |
34 | ||
35 | /* | |
5d4f98a2 YZ |
36 | * compare two delayed tree backrefs with same bytenr and type |
37 | */ | |
38 | static int comp_tree_refs(struct btrfs_delayed_tree_ref *ref2, | |
39 | struct btrfs_delayed_tree_ref *ref1) | |
40 | { | |
41 | if (ref1->node.type == BTRFS_TREE_BLOCK_REF_KEY) { | |
42 | if (ref1->root < ref2->root) | |
43 | return -1; | |
44 | if (ref1->root > ref2->root) | |
45 | return 1; | |
46 | } else { | |
47 | if (ref1->parent < ref2->parent) | |
48 | return -1; | |
49 | if (ref1->parent > ref2->parent) | |
50 | return 1; | |
51 | } | |
52 | return 0; | |
53 | } | |
54 | ||
55 | /* | |
56 | * compare two delayed data backrefs with same bytenr and type | |
56bec294 | 57 | */ |
5d4f98a2 YZ |
58 | static int comp_data_refs(struct btrfs_delayed_data_ref *ref2, |
59 | struct btrfs_delayed_data_ref *ref1) | |
56bec294 | 60 | { |
5d4f98a2 YZ |
61 | if (ref1->node.type == BTRFS_EXTENT_DATA_REF_KEY) { |
62 | if (ref1->root < ref2->root) | |
63 | return -1; | |
64 | if (ref1->root > ref2->root) | |
65 | return 1; | |
66 | if (ref1->objectid < ref2->objectid) | |
67 | return -1; | |
68 | if (ref1->objectid > ref2->objectid) | |
69 | return 1; | |
70 | if (ref1->offset < ref2->offset) | |
71 | return -1; | |
72 | if (ref1->offset > ref2->offset) | |
73 | return 1; | |
74 | } else { | |
75 | if (ref1->parent < ref2->parent) | |
76 | return -1; | |
77 | if (ref1->parent > ref2->parent) | |
78 | return 1; | |
79 | } | |
80 | return 0; | |
81 | } | |
82 | ||
83 | /* | |
84 | * entries in the rb tree are ordered by the byte number of the extent, | |
85 | * type of the delayed backrefs and content of delayed backrefs. | |
86 | */ | |
87 | static int comp_entry(struct btrfs_delayed_ref_node *ref2, | |
88 | struct btrfs_delayed_ref_node *ref1) | |
89 | { | |
90 | if (ref1->bytenr < ref2->bytenr) | |
56bec294 | 91 | return -1; |
5d4f98a2 | 92 | if (ref1->bytenr > ref2->bytenr) |
56bec294 | 93 | return 1; |
5d4f98a2 YZ |
94 | if (ref1->is_head && ref2->is_head) |
95 | return 0; | |
96 | if (ref2->is_head) | |
56bec294 | 97 | return -1; |
5d4f98a2 | 98 | if (ref1->is_head) |
56bec294 | 99 | return 1; |
5d4f98a2 YZ |
100 | if (ref1->type < ref2->type) |
101 | return -1; | |
102 | if (ref1->type > ref2->type) | |
103 | return 1; | |
104 | if (ref1->type == BTRFS_TREE_BLOCK_REF_KEY || | |
105 | ref1->type == BTRFS_SHARED_BLOCK_REF_KEY) { | |
106 | return comp_tree_refs(btrfs_delayed_node_to_tree_ref(ref2), | |
107 | btrfs_delayed_node_to_tree_ref(ref1)); | |
108 | } else if (ref1->type == BTRFS_EXTENT_DATA_REF_KEY || | |
109 | ref1->type == BTRFS_SHARED_DATA_REF_KEY) { | |
110 | return comp_data_refs(btrfs_delayed_node_to_data_ref(ref2), | |
111 | btrfs_delayed_node_to_data_ref(ref1)); | |
112 | } | |
113 | BUG(); | |
56bec294 CM |
114 | return 0; |
115 | } | |
116 | ||
117 | /* | |
118 | * insert a new ref into the rbtree. This returns any existing refs | |
119 | * for the same (bytenr,parent) tuple, or NULL if the new node was properly | |
120 | * inserted. | |
121 | */ | |
122 | static struct btrfs_delayed_ref_node *tree_insert(struct rb_root *root, | |
56bec294 CM |
123 | struct rb_node *node) |
124 | { | |
125 | struct rb_node **p = &root->rb_node; | |
126 | struct rb_node *parent_node = NULL; | |
127 | struct btrfs_delayed_ref_node *entry; | |
5d4f98a2 | 128 | struct btrfs_delayed_ref_node *ins; |
56bec294 CM |
129 | int cmp; |
130 | ||
5d4f98a2 | 131 | ins = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); |
56bec294 CM |
132 | while (*p) { |
133 | parent_node = *p; | |
134 | entry = rb_entry(parent_node, struct btrfs_delayed_ref_node, | |
135 | rb_node); | |
136 | ||
5d4f98a2 | 137 | cmp = comp_entry(entry, ins); |
56bec294 CM |
138 | if (cmp < 0) |
139 | p = &(*p)->rb_left; | |
140 | else if (cmp > 0) | |
141 | p = &(*p)->rb_right; | |
142 | else | |
143 | return entry; | |
144 | } | |
145 | ||
56bec294 CM |
146 | rb_link_node(node, parent_node, p); |
147 | rb_insert_color(node, root); | |
148 | return NULL; | |
149 | } | |
150 | ||
151 | /* | |
5d4f98a2 YZ |
152 | * find an head entry based on bytenr. This returns the delayed ref |
153 | * head if it was able to find one, or NULL if nothing was in that spot | |
56bec294 | 154 | */ |
5d4f98a2 YZ |
155 | static struct btrfs_delayed_ref_node *find_ref_head(struct rb_root *root, |
156 | u64 bytenr, | |
c3e69d58 | 157 | struct btrfs_delayed_ref_node **last) |
56bec294 CM |
158 | { |
159 | struct rb_node *n = root->rb_node; | |
160 | struct btrfs_delayed_ref_node *entry; | |
161 | int cmp; | |
162 | ||
163 | while (n) { | |
164 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
165 | WARN_ON(!entry->in_tree); | |
c3e69d58 CM |
166 | if (last) |
167 | *last = entry; | |
56bec294 | 168 | |
5d4f98a2 YZ |
169 | if (bytenr < entry->bytenr) |
170 | cmp = -1; | |
171 | else if (bytenr > entry->bytenr) | |
172 | cmp = 1; | |
173 | else if (!btrfs_delayed_ref_is_head(entry)) | |
174 | cmp = 1; | |
175 | else | |
176 | cmp = 0; | |
177 | ||
56bec294 CM |
178 | if (cmp < 0) |
179 | n = n->rb_left; | |
180 | else if (cmp > 0) | |
181 | n = n->rb_right; | |
182 | else | |
183 | return entry; | |
184 | } | |
185 | return NULL; | |
186 | } | |
187 | ||
c3e69d58 CM |
188 | int btrfs_delayed_ref_lock(struct btrfs_trans_handle *trans, |
189 | struct btrfs_delayed_ref_head *head) | |
56bec294 | 190 | { |
c3e69d58 CM |
191 | struct btrfs_delayed_ref_root *delayed_refs; |
192 | ||
193 | delayed_refs = &trans->transaction->delayed_refs; | |
194 | assert_spin_locked(&delayed_refs->lock); | |
195 | if (mutex_trylock(&head->mutex)) | |
196 | return 0; | |
197 | ||
198 | atomic_inc(&head->node.refs); | |
199 | spin_unlock(&delayed_refs->lock); | |
200 | ||
201 | mutex_lock(&head->mutex); | |
202 | spin_lock(&delayed_refs->lock); | |
203 | if (!head->node.in_tree) { | |
204 | mutex_unlock(&head->mutex); | |
205 | btrfs_put_delayed_ref(&head->node); | |
206 | return -EAGAIN; | |
207 | } | |
208 | btrfs_put_delayed_ref(&head->node); | |
209 | return 0; | |
210 | } | |
211 | ||
212 | int btrfs_find_ref_cluster(struct btrfs_trans_handle *trans, | |
213 | struct list_head *cluster, u64 start) | |
214 | { | |
215 | int count = 0; | |
216 | struct btrfs_delayed_ref_root *delayed_refs; | |
56bec294 | 217 | struct rb_node *node; |
c3e69d58 | 218 | struct btrfs_delayed_ref_node *ref; |
56bec294 | 219 | struct btrfs_delayed_ref_head *head; |
56bec294 | 220 | |
c3e69d58 CM |
221 | delayed_refs = &trans->transaction->delayed_refs; |
222 | if (start == 0) { | |
223 | node = rb_first(&delayed_refs->root); | |
224 | } else { | |
225 | ref = NULL; | |
5d4f98a2 | 226 | find_ref_head(&delayed_refs->root, start, &ref); |
c3e69d58 CM |
227 | if (ref) { |
228 | struct btrfs_delayed_ref_node *tmp; | |
229 | ||
230 | node = rb_prev(&ref->rb_node); | |
231 | while (node) { | |
232 | tmp = rb_entry(node, | |
233 | struct btrfs_delayed_ref_node, | |
234 | rb_node); | |
235 | if (tmp->bytenr < start) | |
236 | break; | |
237 | ref = tmp; | |
238 | node = rb_prev(&ref->rb_node); | |
239 | } | |
240 | node = &ref->rb_node; | |
241 | } else | |
242 | node = rb_first(&delayed_refs->root); | |
243 | } | |
244 | again: | |
245 | while (node && count < 32) { | |
246 | ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); | |
56bec294 CM |
247 | if (btrfs_delayed_ref_is_head(ref)) { |
248 | head = btrfs_delayed_node_to_head(ref); | |
c3e69d58 CM |
249 | if (list_empty(&head->cluster)) { |
250 | list_add_tail(&head->cluster, cluster); | |
251 | delayed_refs->run_delayed_start = | |
252 | head->node.bytenr; | |
253 | count++; | |
254 | ||
255 | WARN_ON(delayed_refs->num_heads_ready == 0); | |
256 | delayed_refs->num_heads_ready--; | |
257 | } else if (count) { | |
258 | /* the goal of the clustering is to find extents | |
259 | * that are likely to end up in the same extent | |
260 | * leaf on disk. So, we don't want them spread | |
261 | * all over the tree. Stop now if we've hit | |
262 | * a head that was already in use | |
263 | */ | |
56bec294 CM |
264 | break; |
265 | } | |
266 | } | |
c3e69d58 | 267 | node = rb_next(node); |
56bec294 | 268 | } |
c3e69d58 CM |
269 | if (count) { |
270 | return 0; | |
271 | } else if (start) { | |
272 | /* | |
273 | * we've gone to the end of the rbtree without finding any | |
274 | * clusters. start from the beginning and try again | |
275 | */ | |
276 | start = 0; | |
277 | node = rb_first(&delayed_refs->root); | |
278 | goto again; | |
279 | } | |
280 | return 1; | |
56bec294 CM |
281 | } |
282 | ||
283 | /* | |
284 | * This checks to see if there are any delayed refs in the | |
285 | * btree for a given bytenr. It returns one if it finds any | |
286 | * and zero otherwise. | |
287 | * | |
288 | * If it only finds a head node, it returns 0. | |
289 | * | |
290 | * The idea is to use this when deciding if you can safely delete an | |
291 | * extent from the extent allocation tree. There may be a pending | |
292 | * ref in the rbtree that adds or removes references, so as long as this | |
293 | * returns one you need to leave the BTRFS_EXTENT_ITEM in the extent | |
294 | * allocation tree. | |
295 | */ | |
296 | int btrfs_delayed_ref_pending(struct btrfs_trans_handle *trans, u64 bytenr) | |
297 | { | |
298 | struct btrfs_delayed_ref_node *ref; | |
299 | struct btrfs_delayed_ref_root *delayed_refs; | |
300 | struct rb_node *prev_node; | |
301 | int ret = 0; | |
302 | ||
303 | delayed_refs = &trans->transaction->delayed_refs; | |
304 | spin_lock(&delayed_refs->lock); | |
305 | ||
5d4f98a2 | 306 | ref = find_ref_head(&delayed_refs->root, bytenr, NULL); |
56bec294 CM |
307 | if (ref) { |
308 | prev_node = rb_prev(&ref->rb_node); | |
309 | if (!prev_node) | |
310 | goto out; | |
311 | ref = rb_entry(prev_node, struct btrfs_delayed_ref_node, | |
312 | rb_node); | |
313 | if (ref->bytenr == bytenr) | |
314 | ret = 1; | |
315 | } | |
316 | out: | |
317 | spin_unlock(&delayed_refs->lock); | |
318 | return ret; | |
319 | } | |
320 | ||
56bec294 CM |
321 | /* |
322 | * helper function to update an extent delayed ref in the | |
323 | * rbtree. existing and update must both have the same | |
324 | * bytenr and parent | |
325 | * | |
326 | * This may free existing if the update cancels out whatever | |
327 | * operation it was doing. | |
328 | */ | |
329 | static noinline void | |
330 | update_existing_ref(struct btrfs_trans_handle *trans, | |
331 | struct btrfs_delayed_ref_root *delayed_refs, | |
332 | struct btrfs_delayed_ref_node *existing, | |
333 | struct btrfs_delayed_ref_node *update) | |
334 | { | |
5d4f98a2 | 335 | if (update->action != existing->action) { |
56bec294 CM |
336 | /* |
337 | * this is effectively undoing either an add or a | |
338 | * drop. We decrement the ref_mod, and if it goes | |
339 | * down to zero we just delete the entry without | |
340 | * every changing the extent allocation tree. | |
341 | */ | |
342 | existing->ref_mod--; | |
343 | if (existing->ref_mod == 0) { | |
344 | rb_erase(&existing->rb_node, | |
345 | &delayed_refs->root); | |
346 | existing->in_tree = 0; | |
347 | btrfs_put_delayed_ref(existing); | |
348 | delayed_refs->num_entries--; | |
349 | if (trans->delayed_ref_updates) | |
350 | trans->delayed_ref_updates--; | |
5d4f98a2 YZ |
351 | } else { |
352 | WARN_ON(existing->type == BTRFS_TREE_BLOCK_REF_KEY || | |
353 | existing->type == BTRFS_SHARED_BLOCK_REF_KEY); | |
56bec294 CM |
354 | } |
355 | } else { | |
5d4f98a2 YZ |
356 | WARN_ON(existing->type == BTRFS_TREE_BLOCK_REF_KEY || |
357 | existing->type == BTRFS_SHARED_BLOCK_REF_KEY); | |
56bec294 CM |
358 | /* |
359 | * the action on the existing ref matches | |
360 | * the action on the ref we're trying to add. | |
361 | * Bump the ref_mod by one so the backref that | |
362 | * is eventually added/removed has the correct | |
363 | * reference count | |
364 | */ | |
365 | existing->ref_mod += update->ref_mod; | |
366 | } | |
367 | } | |
368 | ||
369 | /* | |
370 | * helper function to update the accounting in the head ref | |
371 | * existing and update must have the same bytenr | |
372 | */ | |
373 | static noinline void | |
374 | update_existing_head_ref(struct btrfs_delayed_ref_node *existing, | |
375 | struct btrfs_delayed_ref_node *update) | |
376 | { | |
377 | struct btrfs_delayed_ref_head *existing_ref; | |
378 | struct btrfs_delayed_ref_head *ref; | |
379 | ||
380 | existing_ref = btrfs_delayed_node_to_head(existing); | |
381 | ref = btrfs_delayed_node_to_head(update); | |
5d4f98a2 | 382 | BUG_ON(existing_ref->is_data != ref->is_data); |
56bec294 CM |
383 | |
384 | if (ref->must_insert_reserved) { | |
385 | /* if the extent was freed and then | |
386 | * reallocated before the delayed ref | |
387 | * entries were processed, we can end up | |
388 | * with an existing head ref without | |
389 | * the must_insert_reserved flag set. | |
390 | * Set it again here | |
391 | */ | |
392 | existing_ref->must_insert_reserved = ref->must_insert_reserved; | |
393 | ||
394 | /* | |
395 | * update the num_bytes so we make sure the accounting | |
396 | * is done correctly | |
397 | */ | |
398 | existing->num_bytes = update->num_bytes; | |
399 | ||
400 | } | |
401 | ||
5d4f98a2 YZ |
402 | if (ref->extent_op) { |
403 | if (!existing_ref->extent_op) { | |
404 | existing_ref->extent_op = ref->extent_op; | |
405 | } else { | |
406 | if (ref->extent_op->update_key) { | |
407 | memcpy(&existing_ref->extent_op->key, | |
408 | &ref->extent_op->key, | |
409 | sizeof(ref->extent_op->key)); | |
410 | existing_ref->extent_op->update_key = 1; | |
411 | } | |
412 | if (ref->extent_op->update_flags) { | |
413 | existing_ref->extent_op->flags_to_set |= | |
414 | ref->extent_op->flags_to_set; | |
415 | existing_ref->extent_op->update_flags = 1; | |
416 | } | |
417 | kfree(ref->extent_op); | |
418 | } | |
419 | } | |
56bec294 CM |
420 | /* |
421 | * update the reference mod on the head to reflect this new operation | |
422 | */ | |
423 | existing->ref_mod += update->ref_mod; | |
424 | } | |
425 | ||
426 | /* | |
5d4f98a2 | 427 | * helper function to actually insert a head node into the rbtree. |
56bec294 | 428 | * this does all the dirty work in terms of maintaining the correct |
5d4f98a2 | 429 | * overall modification count. |
56bec294 | 430 | */ |
5d4f98a2 YZ |
431 | static noinline int add_delayed_ref_head(struct btrfs_trans_handle *trans, |
432 | struct btrfs_delayed_ref_node *ref, | |
433 | u64 bytenr, u64 num_bytes, | |
434 | int action, int is_data) | |
56bec294 CM |
435 | { |
436 | struct btrfs_delayed_ref_node *existing; | |
c3e69d58 | 437 | struct btrfs_delayed_ref_head *head_ref = NULL; |
56bec294 CM |
438 | struct btrfs_delayed_ref_root *delayed_refs; |
439 | int count_mod = 1; | |
440 | int must_insert_reserved = 0; | |
441 | ||
442 | /* | |
443 | * the head node stores the sum of all the mods, so dropping a ref | |
444 | * should drop the sum in the head node by one. | |
445 | */ | |
5d4f98a2 YZ |
446 | if (action == BTRFS_UPDATE_DELAYED_HEAD) |
447 | count_mod = 0; | |
448 | else if (action == BTRFS_DROP_DELAYED_REF) | |
449 | count_mod = -1; | |
56bec294 CM |
450 | |
451 | /* | |
452 | * BTRFS_ADD_DELAYED_EXTENT means that we need to update | |
453 | * the reserved accounting when the extent is finally added, or | |
454 | * if a later modification deletes the delayed ref without ever | |
455 | * inserting the extent into the extent allocation tree. | |
456 | * ref->must_insert_reserved is the flag used to record | |
457 | * that accounting mods are required. | |
458 | * | |
459 | * Once we record must_insert_reserved, switch the action to | |
460 | * BTRFS_ADD_DELAYED_REF because other special casing is not required. | |
461 | */ | |
5d4f98a2 | 462 | if (action == BTRFS_ADD_DELAYED_EXTENT) |
56bec294 | 463 | must_insert_reserved = 1; |
5d4f98a2 | 464 | else |
56bec294 | 465 | must_insert_reserved = 0; |
56bec294 CM |
466 | |
467 | delayed_refs = &trans->transaction->delayed_refs; | |
468 | ||
469 | /* first set the basic ref node struct up */ | |
470 | atomic_set(&ref->refs, 1); | |
471 | ref->bytenr = bytenr; | |
5d4f98a2 | 472 | ref->num_bytes = num_bytes; |
56bec294 | 473 | ref->ref_mod = count_mod; |
5d4f98a2 YZ |
474 | ref->type = 0; |
475 | ref->action = 0; | |
476 | ref->is_head = 1; | |
56bec294 | 477 | ref->in_tree = 1; |
5d4f98a2 YZ |
478 | |
479 | head_ref = btrfs_delayed_node_to_head(ref); | |
480 | head_ref->must_insert_reserved = must_insert_reserved; | |
481 | head_ref->is_data = is_data; | |
482 | ||
483 | INIT_LIST_HEAD(&head_ref->cluster); | |
484 | mutex_init(&head_ref->mutex); | |
485 | ||
1abe9b8a | 486 | trace_btrfs_delayed_ref_head(ref, head_ref, action); |
487 | ||
5d4f98a2 YZ |
488 | existing = tree_insert(&delayed_refs->root, &ref->rb_node); |
489 | ||
490 | if (existing) { | |
491 | update_existing_head_ref(existing, ref); | |
492 | /* | |
493 | * we've updated the existing ref, free the newly | |
494 | * allocated ref | |
495 | */ | |
496 | kfree(ref); | |
497 | } else { | |
498 | delayed_refs->num_heads++; | |
499 | delayed_refs->num_heads_ready++; | |
500 | delayed_refs->num_entries++; | |
501 | trans->delayed_ref_updates++; | |
502 | } | |
503 | return 0; | |
504 | } | |
505 | ||
506 | /* | |
507 | * helper to insert a delayed tree ref into the rbtree. | |
508 | */ | |
509 | static noinline int add_delayed_tree_ref(struct btrfs_trans_handle *trans, | |
510 | struct btrfs_delayed_ref_node *ref, | |
511 | u64 bytenr, u64 num_bytes, u64 parent, | |
512 | u64 ref_root, int level, int action) | |
513 | { | |
514 | struct btrfs_delayed_ref_node *existing; | |
515 | struct btrfs_delayed_tree_ref *full_ref; | |
516 | struct btrfs_delayed_ref_root *delayed_refs; | |
517 | ||
518 | if (action == BTRFS_ADD_DELAYED_EXTENT) | |
519 | action = BTRFS_ADD_DELAYED_REF; | |
520 | ||
521 | delayed_refs = &trans->transaction->delayed_refs; | |
522 | ||
523 | /* first set the basic ref node struct up */ | |
524 | atomic_set(&ref->refs, 1); | |
525 | ref->bytenr = bytenr; | |
56bec294 | 526 | ref->num_bytes = num_bytes; |
5d4f98a2 YZ |
527 | ref->ref_mod = 1; |
528 | ref->action = action; | |
529 | ref->is_head = 0; | |
530 | ref->in_tree = 1; | |
56bec294 | 531 | |
5d4f98a2 YZ |
532 | full_ref = btrfs_delayed_node_to_tree_ref(ref); |
533 | if (parent) { | |
534 | full_ref->parent = parent; | |
535 | ref->type = BTRFS_SHARED_BLOCK_REF_KEY; | |
56bec294 | 536 | } else { |
56bec294 | 537 | full_ref->root = ref_root; |
5d4f98a2 | 538 | ref->type = BTRFS_TREE_BLOCK_REF_KEY; |
56bec294 | 539 | } |
5d4f98a2 | 540 | full_ref->level = level; |
56bec294 | 541 | |
1abe9b8a | 542 | trace_btrfs_delayed_tree_ref(ref, full_ref, action); |
543 | ||
5d4f98a2 | 544 | existing = tree_insert(&delayed_refs->root, &ref->rb_node); |
56bec294 CM |
545 | |
546 | if (existing) { | |
5d4f98a2 YZ |
547 | update_existing_ref(trans, delayed_refs, existing, ref); |
548 | /* | |
549 | * we've updated the existing ref, free the newly | |
550 | * allocated ref | |
551 | */ | |
552 | kfree(ref); | |
553 | } else { | |
554 | delayed_refs->num_entries++; | |
555 | trans->delayed_ref_updates++; | |
556 | } | |
557 | return 0; | |
558 | } | |
559 | ||
560 | /* | |
561 | * helper to insert a delayed data ref into the rbtree. | |
562 | */ | |
563 | static noinline int add_delayed_data_ref(struct btrfs_trans_handle *trans, | |
564 | struct btrfs_delayed_ref_node *ref, | |
565 | u64 bytenr, u64 num_bytes, u64 parent, | |
566 | u64 ref_root, u64 owner, u64 offset, | |
567 | int action) | |
568 | { | |
569 | struct btrfs_delayed_ref_node *existing; | |
570 | struct btrfs_delayed_data_ref *full_ref; | |
571 | struct btrfs_delayed_ref_root *delayed_refs; | |
572 | ||
573 | if (action == BTRFS_ADD_DELAYED_EXTENT) | |
574 | action = BTRFS_ADD_DELAYED_REF; | |
575 | ||
576 | delayed_refs = &trans->transaction->delayed_refs; | |
577 | ||
578 | /* first set the basic ref node struct up */ | |
579 | atomic_set(&ref->refs, 1); | |
580 | ref->bytenr = bytenr; | |
581 | ref->num_bytes = num_bytes; | |
582 | ref->ref_mod = 1; | |
583 | ref->action = action; | |
584 | ref->is_head = 0; | |
585 | ref->in_tree = 1; | |
586 | ||
587 | full_ref = btrfs_delayed_node_to_data_ref(ref); | |
588 | if (parent) { | |
589 | full_ref->parent = parent; | |
590 | ref->type = BTRFS_SHARED_DATA_REF_KEY; | |
591 | } else { | |
592 | full_ref->root = ref_root; | |
593 | ref->type = BTRFS_EXTENT_DATA_REF_KEY; | |
594 | } | |
595 | full_ref->objectid = owner; | |
596 | full_ref->offset = offset; | |
56bec294 | 597 | |
1abe9b8a | 598 | trace_btrfs_delayed_data_ref(ref, full_ref, action); |
599 | ||
5d4f98a2 YZ |
600 | existing = tree_insert(&delayed_refs->root, &ref->rb_node); |
601 | ||
602 | if (existing) { | |
603 | update_existing_ref(trans, delayed_refs, existing, ref); | |
56bec294 CM |
604 | /* |
605 | * we've updated the existing ref, free the newly | |
606 | * allocated ref | |
607 | */ | |
608 | kfree(ref); | |
609 | } else { | |
610 | delayed_refs->num_entries++; | |
611 | trans->delayed_ref_updates++; | |
612 | } | |
613 | return 0; | |
614 | } | |
615 | ||
616 | /* | |
5d4f98a2 | 617 | * add a delayed tree ref. This does all of the accounting required |
56bec294 CM |
618 | * to make sure the delayed ref is eventually processed before this |
619 | * transaction commits. | |
620 | */ | |
5d4f98a2 YZ |
621 | int btrfs_add_delayed_tree_ref(struct btrfs_trans_handle *trans, |
622 | u64 bytenr, u64 num_bytes, u64 parent, | |
623 | u64 ref_root, int level, int action, | |
624 | struct btrfs_delayed_extent_op *extent_op) | |
56bec294 | 625 | { |
5d4f98a2 | 626 | struct btrfs_delayed_tree_ref *ref; |
56bec294 CM |
627 | struct btrfs_delayed_ref_head *head_ref; |
628 | struct btrfs_delayed_ref_root *delayed_refs; | |
629 | int ret; | |
630 | ||
5d4f98a2 | 631 | BUG_ON(extent_op && extent_op->is_data); |
56bec294 CM |
632 | ref = kmalloc(sizeof(*ref), GFP_NOFS); |
633 | if (!ref) | |
634 | return -ENOMEM; | |
635 | ||
5d4f98a2 YZ |
636 | head_ref = kmalloc(sizeof(*head_ref), GFP_NOFS); |
637 | if (!head_ref) { | |
638 | kfree(ref); | |
639 | return -ENOMEM; | |
640 | } | |
641 | ||
642 | head_ref->extent_op = extent_op; | |
643 | ||
644 | delayed_refs = &trans->transaction->delayed_refs; | |
645 | spin_lock(&delayed_refs->lock); | |
646 | ||
56bec294 | 647 | /* |
5d4f98a2 YZ |
648 | * insert both the head node and the new ref without dropping |
649 | * the spin lock | |
56bec294 | 650 | */ |
5d4f98a2 YZ |
651 | ret = add_delayed_ref_head(trans, &head_ref->node, bytenr, num_bytes, |
652 | action, 0); | |
653 | BUG_ON(ret); | |
654 | ||
655 | ret = add_delayed_tree_ref(trans, &ref->node, bytenr, num_bytes, | |
656 | parent, ref_root, level, action); | |
657 | BUG_ON(ret); | |
658 | spin_unlock(&delayed_refs->lock); | |
659 | return 0; | |
660 | } | |
661 | ||
662 | /* | |
663 | * add a delayed data ref. it's similar to btrfs_add_delayed_tree_ref. | |
664 | */ | |
665 | int btrfs_add_delayed_data_ref(struct btrfs_trans_handle *trans, | |
666 | u64 bytenr, u64 num_bytes, | |
667 | u64 parent, u64 ref_root, | |
668 | u64 owner, u64 offset, int action, | |
669 | struct btrfs_delayed_extent_op *extent_op) | |
670 | { | |
671 | struct btrfs_delayed_data_ref *ref; | |
672 | struct btrfs_delayed_ref_head *head_ref; | |
673 | struct btrfs_delayed_ref_root *delayed_refs; | |
674 | int ret; | |
675 | ||
676 | BUG_ON(extent_op && !extent_op->is_data); | |
677 | ref = kmalloc(sizeof(*ref), GFP_NOFS); | |
678 | if (!ref) | |
679 | return -ENOMEM; | |
56bec294 CM |
680 | |
681 | head_ref = kmalloc(sizeof(*head_ref), GFP_NOFS); | |
682 | if (!head_ref) { | |
683 | kfree(ref); | |
684 | return -ENOMEM; | |
685 | } | |
5d4f98a2 YZ |
686 | |
687 | head_ref->extent_op = extent_op; | |
688 | ||
56bec294 CM |
689 | delayed_refs = &trans->transaction->delayed_refs; |
690 | spin_lock(&delayed_refs->lock); | |
691 | ||
692 | /* | |
693 | * insert both the head node and the new ref without dropping | |
694 | * the spin lock | |
695 | */ | |
5d4f98a2 YZ |
696 | ret = add_delayed_ref_head(trans, &head_ref->node, bytenr, num_bytes, |
697 | action, 1); | |
56bec294 CM |
698 | BUG_ON(ret); |
699 | ||
5d4f98a2 YZ |
700 | ret = add_delayed_data_ref(trans, &ref->node, bytenr, num_bytes, |
701 | parent, ref_root, owner, offset, action); | |
702 | BUG_ON(ret); | |
703 | spin_unlock(&delayed_refs->lock); | |
704 | return 0; | |
705 | } | |
706 | ||
707 | int btrfs_add_delayed_extent_op(struct btrfs_trans_handle *trans, | |
708 | u64 bytenr, u64 num_bytes, | |
709 | struct btrfs_delayed_extent_op *extent_op) | |
710 | { | |
711 | struct btrfs_delayed_ref_head *head_ref; | |
712 | struct btrfs_delayed_ref_root *delayed_refs; | |
713 | int ret; | |
714 | ||
715 | head_ref = kmalloc(sizeof(*head_ref), GFP_NOFS); | |
716 | if (!head_ref) | |
717 | return -ENOMEM; | |
718 | ||
719 | head_ref->extent_op = extent_op; | |
720 | ||
721 | delayed_refs = &trans->transaction->delayed_refs; | |
722 | spin_lock(&delayed_refs->lock); | |
723 | ||
724 | ret = add_delayed_ref_head(trans, &head_ref->node, bytenr, | |
725 | num_bytes, BTRFS_UPDATE_DELAYED_HEAD, | |
726 | extent_op->is_data); | |
56bec294 | 727 | BUG_ON(ret); |
5d4f98a2 | 728 | |
56bec294 CM |
729 | spin_unlock(&delayed_refs->lock); |
730 | return 0; | |
731 | } | |
732 | ||
1887be66 CM |
733 | /* |
734 | * this does a simple search for the head node for a given extent. | |
735 | * It must be called with the delayed ref spinlock held, and it returns | |
736 | * the head node if any where found, or NULL if not. | |
737 | */ | |
738 | struct btrfs_delayed_ref_head * | |
739 | btrfs_find_delayed_ref_head(struct btrfs_trans_handle *trans, u64 bytenr) | |
740 | { | |
741 | struct btrfs_delayed_ref_node *ref; | |
742 | struct btrfs_delayed_ref_root *delayed_refs; | |
743 | ||
744 | delayed_refs = &trans->transaction->delayed_refs; | |
5d4f98a2 | 745 | ref = find_ref_head(&delayed_refs->root, bytenr, NULL); |
1887be66 CM |
746 | if (ref) |
747 | return btrfs_delayed_node_to_head(ref); | |
748 | return NULL; | |
749 | } | |
750 | ||
56bec294 CM |
751 | /* |
752 | * add a delayed ref to the tree. This does all of the accounting required | |
753 | * to make sure the delayed ref is eventually processed before this | |
754 | * transaction commits. | |
755 | * | |
756 | * The main point of this call is to add and remove a backreference in a single | |
757 | * shot, taking the lock only once, and only searching for the head node once. | |
758 | * | |
759 | * It is the same as doing a ref add and delete in two separate calls. | |
760 | */ | |
5d4f98a2 | 761 | #if 0 |
56bec294 CM |
762 | int btrfs_update_delayed_ref(struct btrfs_trans_handle *trans, |
763 | u64 bytenr, u64 num_bytes, u64 orig_parent, | |
764 | u64 parent, u64 orig_ref_root, u64 ref_root, | |
765 | u64 orig_ref_generation, u64 ref_generation, | |
766 | u64 owner_objectid, int pin) | |
767 | { | |
768 | struct btrfs_delayed_ref *ref; | |
769 | struct btrfs_delayed_ref *old_ref; | |
770 | struct btrfs_delayed_ref_head *head_ref; | |
771 | struct btrfs_delayed_ref_root *delayed_refs; | |
772 | int ret; | |
773 | ||
774 | ref = kmalloc(sizeof(*ref), GFP_NOFS); | |
775 | if (!ref) | |
776 | return -ENOMEM; | |
777 | ||
778 | old_ref = kmalloc(sizeof(*old_ref), GFP_NOFS); | |
779 | if (!old_ref) { | |
780 | kfree(ref); | |
781 | return -ENOMEM; | |
782 | } | |
783 | ||
784 | /* | |
785 | * the parent = 0 case comes from cases where we don't actually | |
786 | * know the parent yet. It will get updated later via a add/drop | |
787 | * pair. | |
788 | */ | |
789 | if (parent == 0) | |
790 | parent = bytenr; | |
791 | if (orig_parent == 0) | |
792 | orig_parent = bytenr; | |
793 | ||
794 | head_ref = kmalloc(sizeof(*head_ref), GFP_NOFS); | |
795 | if (!head_ref) { | |
796 | kfree(ref); | |
797 | kfree(old_ref); | |
798 | return -ENOMEM; | |
799 | } | |
800 | delayed_refs = &trans->transaction->delayed_refs; | |
801 | spin_lock(&delayed_refs->lock); | |
802 | ||
803 | /* | |
804 | * insert both the head node and the new ref without dropping | |
805 | * the spin lock | |
806 | */ | |
807 | ret = __btrfs_add_delayed_ref(trans, &head_ref->node, bytenr, num_bytes, | |
808 | (u64)-1, 0, 0, 0, | |
1a81af4d | 809 | BTRFS_UPDATE_DELAYED_HEAD, 0); |
56bec294 CM |
810 | BUG_ON(ret); |
811 | ||
812 | ret = __btrfs_add_delayed_ref(trans, &ref->node, bytenr, num_bytes, | |
813 | parent, ref_root, ref_generation, | |
814 | owner_objectid, BTRFS_ADD_DELAYED_REF, 0); | |
815 | BUG_ON(ret); | |
816 | ||
817 | ret = __btrfs_add_delayed_ref(trans, &old_ref->node, bytenr, num_bytes, | |
818 | orig_parent, orig_ref_root, | |
819 | orig_ref_generation, owner_objectid, | |
820 | BTRFS_DROP_DELAYED_REF, pin); | |
821 | BUG_ON(ret); | |
822 | spin_unlock(&delayed_refs->lock); | |
823 | return 0; | |
824 | } | |
5d4f98a2 | 825 | #endif |