]>
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 | n = root->rb_node; |
209 | entry = NULL; | |
56bec294 | 210 | while (n) { |
c46effa6 | 211 | entry = rb_entry(n, struct btrfs_delayed_ref_head, href_node); |
c3e69d58 CM |
212 | if (last) |
213 | *last = entry; | |
56bec294 | 214 | |
c46effa6 | 215 | if (bytenr < entry->node.bytenr) |
5d4f98a2 | 216 | cmp = -1; |
c46effa6 | 217 | else if (bytenr > entry->node.bytenr) |
5d4f98a2 YZ |
218 | cmp = 1; |
219 | else | |
220 | cmp = 0; | |
221 | ||
56bec294 CM |
222 | if (cmp < 0) |
223 | n = n->rb_left; | |
224 | else if (cmp > 0) | |
225 | n = n->rb_right; | |
226 | else | |
227 | return entry; | |
228 | } | |
d1270cd9 AJ |
229 | if (entry && return_bigger) { |
230 | if (cmp > 0) { | |
c46effa6 | 231 | n = rb_next(&entry->href_node); |
d1270cd9 AJ |
232 | if (!n) |
233 | n = rb_first(root); | |
c46effa6 LB |
234 | entry = rb_entry(n, struct btrfs_delayed_ref_head, |
235 | href_node); | |
6103fb43 FM |
236 | if (last) |
237 | *last = entry; | |
238 | return entry; | |
d1270cd9 AJ |
239 | } |
240 | return entry; | |
241 | } | |
56bec294 CM |
242 | return NULL; |
243 | } | |
244 | ||
c3e69d58 CM |
245 | int btrfs_delayed_ref_lock(struct btrfs_trans_handle *trans, |
246 | struct btrfs_delayed_ref_head *head) | |
56bec294 | 247 | { |
c3e69d58 CM |
248 | struct btrfs_delayed_ref_root *delayed_refs; |
249 | ||
250 | delayed_refs = &trans->transaction->delayed_refs; | |
251 | assert_spin_locked(&delayed_refs->lock); | |
252 | if (mutex_trylock(&head->mutex)) | |
253 | return 0; | |
254 | ||
255 | atomic_inc(&head->node.refs); | |
256 | spin_unlock(&delayed_refs->lock); | |
257 | ||
258 | mutex_lock(&head->mutex); | |
259 | spin_lock(&delayed_refs->lock); | |
260 | if (!head->node.in_tree) { | |
261 | mutex_unlock(&head->mutex); | |
262 | btrfs_put_delayed_ref(&head->node); | |
263 | return -EAGAIN; | |
264 | } | |
265 | btrfs_put_delayed_ref(&head->node); | |
266 | return 0; | |
267 | } | |
268 | ||
35a3621b | 269 | static inline void drop_delayed_ref(struct btrfs_trans_handle *trans, |
ae1e206b | 270 | struct btrfs_delayed_ref_root *delayed_refs, |
d7df2c79 | 271 | struct btrfs_delayed_ref_head *head, |
ae1e206b JB |
272 | struct btrfs_delayed_ref_node *ref) |
273 | { | |
c46effa6 | 274 | if (btrfs_delayed_ref_is_head(ref)) { |
c46effa6 LB |
275 | head = btrfs_delayed_node_to_head(ref); |
276 | rb_erase(&head->href_node, &delayed_refs->href_root); | |
d7df2c79 JB |
277 | } else { |
278 | assert_spin_locked(&head->lock); | |
279 | rb_erase(&ref->rb_node, &head->ref_root); | |
c46effa6 | 280 | } |
ae1e206b JB |
281 | ref->in_tree = 0; |
282 | btrfs_put_delayed_ref(ref); | |
d7df2c79 | 283 | atomic_dec(&delayed_refs->num_entries); |
ae1e206b JB |
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, | |
d7df2c79 | 290 | struct btrfs_delayed_ref_head *head, |
ae1e206b JB |
291 | struct btrfs_delayed_ref_node *ref, u64 seq) |
292 | { | |
293 | struct rb_node *node; | |
ae1e206b JB |
294 | int mod = 0; |
295 | int done = 0; | |
296 | ||
d7df2c79 JB |
297 | node = rb_next(&ref->rb_node); |
298 | while (!done && node) { | |
ae1e206b JB |
299 | struct btrfs_delayed_ref_node *next; |
300 | ||
301 | next = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); | |
d7df2c79 | 302 | node = rb_next(node); |
ae1e206b JB |
303 | if (seq && next->seq >= seq) |
304 | break; | |
305 | if (comp_entry(ref, next, 0)) | |
306 | continue; | |
307 | ||
308 | if (ref->action == next->action) { | |
309 | mod = next->ref_mod; | |
310 | } else { | |
311 | if (ref->ref_mod < next->ref_mod) { | |
312 | struct btrfs_delayed_ref_node *tmp; | |
313 | ||
314 | tmp = ref; | |
315 | ref = next; | |
316 | next = tmp; | |
317 | done = 1; | |
318 | } | |
319 | mod = -next->ref_mod; | |
320 | } | |
321 | ||
d7df2c79 | 322 | drop_delayed_ref(trans, delayed_refs, head, next); |
ae1e206b JB |
323 | ref->ref_mod += mod; |
324 | if (ref->ref_mod == 0) { | |
d7df2c79 JB |
325 | drop_delayed_ref(trans, delayed_refs, head, ref); |
326 | done = 1; | |
ae1e206b JB |
327 | } else { |
328 | /* | |
329 | * You can't have multiples of the same ref on a tree | |
330 | * block. | |
331 | */ | |
332 | WARN_ON(ref->type == BTRFS_TREE_BLOCK_REF_KEY || | |
333 | ref->type == BTRFS_SHARED_BLOCK_REF_KEY); | |
334 | } | |
ae1e206b | 335 | } |
d7df2c79 | 336 | return done; |
ae1e206b JB |
337 | } |
338 | ||
339 | void btrfs_merge_delayed_refs(struct btrfs_trans_handle *trans, | |
340 | struct btrfs_fs_info *fs_info, | |
341 | struct btrfs_delayed_ref_root *delayed_refs, | |
342 | struct btrfs_delayed_ref_head *head) | |
343 | { | |
344 | struct rb_node *node; | |
345 | u64 seq = 0; | |
346 | ||
d7df2c79 | 347 | assert_spin_locked(&head->lock); |
9e5ac13a LB |
348 | /* |
349 | * We don't have too much refs to merge in the case of delayed data | |
350 | * refs. | |
351 | */ | |
352 | if (head->is_data) | |
353 | return; | |
354 | ||
ae1e206b JB |
355 | spin_lock(&fs_info->tree_mod_seq_lock); |
356 | if (!list_empty(&fs_info->tree_mod_seq_list)) { | |
357 | struct seq_list *elem; | |
358 | ||
359 | elem = list_first_entry(&fs_info->tree_mod_seq_list, | |
360 | struct seq_list, list); | |
361 | seq = elem->seq; | |
362 | } | |
363 | spin_unlock(&fs_info->tree_mod_seq_lock); | |
364 | ||
d7df2c79 | 365 | node = rb_first(&head->ref_root); |
ae1e206b JB |
366 | while (node) { |
367 | struct btrfs_delayed_ref_node *ref; | |
368 | ||
369 | ref = rb_entry(node, struct btrfs_delayed_ref_node, | |
370 | rb_node); | |
ae1e206b JB |
371 | /* We can't merge refs that are outside of our seq count */ |
372 | if (seq && ref->seq >= seq) | |
373 | break; | |
d7df2c79 JB |
374 | if (merge_ref(trans, delayed_refs, head, ref, seq)) |
375 | node = rb_first(&head->ref_root); | |
ae1e206b | 376 | else |
d7df2c79 | 377 | node = rb_next(&ref->rb_node); |
ae1e206b JB |
378 | } |
379 | } | |
380 | ||
097b8a7c JS |
381 | int btrfs_check_delayed_seq(struct btrfs_fs_info *fs_info, |
382 | struct btrfs_delayed_ref_root *delayed_refs, | |
00f04b88 AJ |
383 | u64 seq) |
384 | { | |
385 | struct seq_list *elem; | |
097b8a7c JS |
386 | int ret = 0; |
387 | ||
388 | spin_lock(&fs_info->tree_mod_seq_lock); | |
389 | if (!list_empty(&fs_info->tree_mod_seq_list)) { | |
390 | elem = list_first_entry(&fs_info->tree_mod_seq_list, | |
391 | struct seq_list, list); | |
392 | if (seq >= elem->seq) { | |
fc36ed7e JS |
393 | pr_debug("holding back delayed_ref %#x.%x, lowest is %#x.%x (%p)\n", |
394 | (u32)(seq >> 32), (u32)seq, | |
395 | (u32)(elem->seq >> 32), (u32)elem->seq, | |
396 | delayed_refs); | |
097b8a7c JS |
397 | ret = 1; |
398 | } | |
00f04b88 | 399 | } |
097b8a7c JS |
400 | |
401 | spin_unlock(&fs_info->tree_mod_seq_lock); | |
402 | return ret; | |
00f04b88 AJ |
403 | } |
404 | ||
d7df2c79 JB |
405 | struct btrfs_delayed_ref_head * |
406 | btrfs_select_ref_head(struct btrfs_trans_handle *trans) | |
c3e69d58 | 407 | { |
c3e69d58 | 408 | struct btrfs_delayed_ref_root *delayed_refs; |
d7df2c79 JB |
409 | struct btrfs_delayed_ref_head *head; |
410 | u64 start; | |
411 | bool loop = false; | |
56bec294 | 412 | |
c3e69d58 | 413 | delayed_refs = &trans->transaction->delayed_refs; |
c46effa6 | 414 | |
c3e69d58 | 415 | again: |
d7df2c79 JB |
416 | start = delayed_refs->run_delayed_start; |
417 | head = find_ref_head(&delayed_refs->href_root, start, NULL, 1); | |
418 | if (!head && !loop) { | |
419 | delayed_refs->run_delayed_start = 0; | |
c3e69d58 | 420 | start = 0; |
d7df2c79 JB |
421 | loop = true; |
422 | head = find_ref_head(&delayed_refs->href_root, start, NULL, 1); | |
423 | if (!head) | |
424 | return NULL; | |
425 | } else if (!head && loop) { | |
426 | return NULL; | |
c3e69d58 | 427 | } |
56bec294 | 428 | |
d7df2c79 JB |
429 | while (head->processing) { |
430 | struct rb_node *node; | |
431 | ||
432 | node = rb_next(&head->href_node); | |
433 | if (!node) { | |
434 | if (loop) | |
435 | return NULL; | |
436 | delayed_refs->run_delayed_start = 0; | |
437 | start = 0; | |
438 | loop = true; | |
439 | goto again; | |
440 | } | |
441 | head = rb_entry(node, struct btrfs_delayed_ref_head, | |
442 | href_node); | |
443 | } | |
093486c4 | 444 | |
d7df2c79 JB |
445 | head->processing = 1; |
446 | WARN_ON(delayed_refs->num_heads_ready == 0); | |
447 | delayed_refs->num_heads_ready--; | |
448 | delayed_refs->run_delayed_start = head->node.bytenr + | |
449 | head->node.num_bytes; | |
450 | return head; | |
093486c4 MX |
451 | } |
452 | ||
56bec294 CM |
453 | /* |
454 | * helper function to update an extent delayed ref in the | |
455 | * rbtree. existing and update must both have the same | |
456 | * bytenr and parent | |
457 | * | |
458 | * This may free existing if the update cancels out whatever | |
459 | * operation it was doing. | |
460 | */ | |
461 | static noinline void | |
462 | update_existing_ref(struct btrfs_trans_handle *trans, | |
463 | struct btrfs_delayed_ref_root *delayed_refs, | |
d7df2c79 | 464 | struct btrfs_delayed_ref_head *head, |
56bec294 CM |
465 | struct btrfs_delayed_ref_node *existing, |
466 | struct btrfs_delayed_ref_node *update) | |
467 | { | |
5d4f98a2 | 468 | if (update->action != existing->action) { |
56bec294 CM |
469 | /* |
470 | * this is effectively undoing either an add or a | |
471 | * drop. We decrement the ref_mod, and if it goes | |
472 | * down to zero we just delete the entry without | |
473 | * every changing the extent allocation tree. | |
474 | */ | |
475 | existing->ref_mod--; | |
ae1e206b | 476 | if (existing->ref_mod == 0) |
d7df2c79 | 477 | drop_delayed_ref(trans, delayed_refs, head, existing); |
ae1e206b | 478 | else |
5d4f98a2 YZ |
479 | WARN_ON(existing->type == BTRFS_TREE_BLOCK_REF_KEY || |
480 | existing->type == BTRFS_SHARED_BLOCK_REF_KEY); | |
56bec294 | 481 | } else { |
5d4f98a2 YZ |
482 | WARN_ON(existing->type == BTRFS_TREE_BLOCK_REF_KEY || |
483 | existing->type == BTRFS_SHARED_BLOCK_REF_KEY); | |
56bec294 CM |
484 | /* |
485 | * the action on the existing ref matches | |
486 | * the action on the ref we're trying to add. | |
487 | * Bump the ref_mod by one so the backref that | |
488 | * is eventually added/removed has the correct | |
489 | * reference count | |
490 | */ | |
491 | existing->ref_mod += update->ref_mod; | |
492 | } | |
493 | } | |
494 | ||
495 | /* | |
496 | * helper function to update the accounting in the head ref | |
497 | * existing and update must have the same bytenr | |
498 | */ | |
499 | static noinline void | |
500 | update_existing_head_ref(struct btrfs_delayed_ref_node *existing, | |
501 | struct btrfs_delayed_ref_node *update) | |
502 | { | |
503 | struct btrfs_delayed_ref_head *existing_ref; | |
504 | struct btrfs_delayed_ref_head *ref; | |
505 | ||
506 | existing_ref = btrfs_delayed_node_to_head(existing); | |
507 | ref = btrfs_delayed_node_to_head(update); | |
5d4f98a2 | 508 | BUG_ON(existing_ref->is_data != ref->is_data); |
56bec294 CM |
509 | |
510 | if (ref->must_insert_reserved) { | |
511 | /* if the extent was freed and then | |
512 | * reallocated before the delayed ref | |
513 | * entries were processed, we can end up | |
514 | * with an existing head ref without | |
515 | * the must_insert_reserved flag set. | |
516 | * Set it again here | |
517 | */ | |
518 | existing_ref->must_insert_reserved = ref->must_insert_reserved; | |
519 | ||
520 | /* | |
521 | * update the num_bytes so we make sure the accounting | |
522 | * is done correctly | |
523 | */ | |
524 | existing->num_bytes = update->num_bytes; | |
525 | ||
526 | } | |
527 | ||
5d4f98a2 YZ |
528 | if (ref->extent_op) { |
529 | if (!existing_ref->extent_op) { | |
530 | existing_ref->extent_op = ref->extent_op; | |
531 | } else { | |
532 | if (ref->extent_op->update_key) { | |
533 | memcpy(&existing_ref->extent_op->key, | |
534 | &ref->extent_op->key, | |
535 | sizeof(ref->extent_op->key)); | |
536 | existing_ref->extent_op->update_key = 1; | |
537 | } | |
538 | if (ref->extent_op->update_flags) { | |
539 | existing_ref->extent_op->flags_to_set |= | |
540 | ref->extent_op->flags_to_set; | |
541 | existing_ref->extent_op->update_flags = 1; | |
542 | } | |
78a6184a | 543 | btrfs_free_delayed_extent_op(ref->extent_op); |
5d4f98a2 YZ |
544 | } |
545 | } | |
56bec294 | 546 | /* |
d7df2c79 JB |
547 | * update the reference mod on the head to reflect this new operation, |
548 | * only need the lock for this case cause we could be processing it | |
549 | * currently, for refs we just added we know we're a-ok. | |
56bec294 | 550 | */ |
d7df2c79 | 551 | spin_lock(&existing_ref->lock); |
56bec294 | 552 | existing->ref_mod += update->ref_mod; |
d7df2c79 | 553 | spin_unlock(&existing_ref->lock); |
56bec294 CM |
554 | } |
555 | ||
556 | /* | |
5d4f98a2 | 557 | * helper function to actually insert a head node into the rbtree. |
56bec294 | 558 | * this does all the dirty work in terms of maintaining the correct |
5d4f98a2 | 559 | * overall modification count. |
56bec294 | 560 | */ |
d7df2c79 JB |
561 | static noinline struct btrfs_delayed_ref_head * |
562 | add_delayed_ref_head(struct btrfs_fs_info *fs_info, | |
563 | struct btrfs_trans_handle *trans, | |
564 | struct btrfs_delayed_ref_node *ref, u64 bytenr, | |
565 | u64 num_bytes, int action, int is_data) | |
56bec294 | 566 | { |
d7df2c79 | 567 | struct btrfs_delayed_ref_head *existing; |
c3e69d58 | 568 | struct btrfs_delayed_ref_head *head_ref = NULL; |
56bec294 CM |
569 | struct btrfs_delayed_ref_root *delayed_refs; |
570 | int count_mod = 1; | |
571 | int must_insert_reserved = 0; | |
572 | ||
573 | /* | |
574 | * the head node stores the sum of all the mods, so dropping a ref | |
575 | * should drop the sum in the head node by one. | |
576 | */ | |
5d4f98a2 YZ |
577 | if (action == BTRFS_UPDATE_DELAYED_HEAD) |
578 | count_mod = 0; | |
579 | else if (action == BTRFS_DROP_DELAYED_REF) | |
580 | count_mod = -1; | |
56bec294 CM |
581 | |
582 | /* | |
583 | * BTRFS_ADD_DELAYED_EXTENT means that we need to update | |
584 | * the reserved accounting when the extent is finally added, or | |
585 | * if a later modification deletes the delayed ref without ever | |
586 | * inserting the extent into the extent allocation tree. | |
587 | * ref->must_insert_reserved is the flag used to record | |
588 | * that accounting mods are required. | |
589 | * | |
590 | * Once we record must_insert_reserved, switch the action to | |
591 | * BTRFS_ADD_DELAYED_REF because other special casing is not required. | |
592 | */ | |
5d4f98a2 | 593 | if (action == BTRFS_ADD_DELAYED_EXTENT) |
56bec294 | 594 | must_insert_reserved = 1; |
5d4f98a2 | 595 | else |
56bec294 | 596 | must_insert_reserved = 0; |
56bec294 CM |
597 | |
598 | delayed_refs = &trans->transaction->delayed_refs; | |
599 | ||
600 | /* first set the basic ref node struct up */ | |
601 | atomic_set(&ref->refs, 1); | |
602 | ref->bytenr = bytenr; | |
5d4f98a2 | 603 | ref->num_bytes = num_bytes; |
56bec294 | 604 | ref->ref_mod = count_mod; |
5d4f98a2 YZ |
605 | ref->type = 0; |
606 | ref->action = 0; | |
607 | ref->is_head = 1; | |
56bec294 | 608 | ref->in_tree = 1; |
00f04b88 | 609 | ref->seq = 0; |
5d4f98a2 YZ |
610 | |
611 | head_ref = btrfs_delayed_node_to_head(ref); | |
612 | head_ref->must_insert_reserved = must_insert_reserved; | |
613 | head_ref->is_data = is_data; | |
d7df2c79 JB |
614 | head_ref->ref_root = RB_ROOT; |
615 | head_ref->processing = 0; | |
5d4f98a2 | 616 | |
d7df2c79 | 617 | spin_lock_init(&head_ref->lock); |
5d4f98a2 YZ |
618 | mutex_init(&head_ref->mutex); |
619 | ||
599c75ec | 620 | trace_add_delayed_ref_head(ref, head_ref, action); |
1abe9b8a | 621 | |
d7df2c79 JB |
622 | existing = htree_insert(&delayed_refs->href_root, |
623 | &head_ref->href_node); | |
5d4f98a2 | 624 | if (existing) { |
d7df2c79 | 625 | update_existing_head_ref(&existing->node, ref); |
5d4f98a2 YZ |
626 | /* |
627 | * we've updated the existing ref, free the newly | |
628 | * allocated ref | |
629 | */ | |
78a6184a | 630 | kmem_cache_free(btrfs_delayed_ref_head_cachep, head_ref); |
d7df2c79 | 631 | head_ref = existing; |
5d4f98a2 YZ |
632 | } else { |
633 | delayed_refs->num_heads++; | |
634 | delayed_refs->num_heads_ready++; | |
d7df2c79 | 635 | atomic_inc(&delayed_refs->num_entries); |
5d4f98a2 YZ |
636 | trans->delayed_ref_updates++; |
637 | } | |
d7df2c79 | 638 | return head_ref; |
5d4f98a2 YZ |
639 | } |
640 | ||
641 | /* | |
642 | * helper to insert a delayed tree ref into the rbtree. | |
643 | */ | |
d7df2c79 JB |
644 | static noinline void |
645 | add_delayed_tree_ref(struct btrfs_fs_info *fs_info, | |
646 | struct btrfs_trans_handle *trans, | |
647 | struct btrfs_delayed_ref_head *head_ref, | |
648 | struct btrfs_delayed_ref_node *ref, u64 bytenr, | |
649 | u64 num_bytes, u64 parent, u64 ref_root, int level, | |
650 | int action, int for_cow) | |
5d4f98a2 YZ |
651 | { |
652 | struct btrfs_delayed_ref_node *existing; | |
653 | struct btrfs_delayed_tree_ref *full_ref; | |
654 | struct btrfs_delayed_ref_root *delayed_refs; | |
00f04b88 | 655 | u64 seq = 0; |
5d4f98a2 YZ |
656 | |
657 | if (action == BTRFS_ADD_DELAYED_EXTENT) | |
658 | action = BTRFS_ADD_DELAYED_REF; | |
659 | ||
660 | delayed_refs = &trans->transaction->delayed_refs; | |
661 | ||
662 | /* first set the basic ref node struct up */ | |
663 | atomic_set(&ref->refs, 1); | |
664 | ref->bytenr = bytenr; | |
56bec294 | 665 | ref->num_bytes = num_bytes; |
5d4f98a2 YZ |
666 | ref->ref_mod = 1; |
667 | ref->action = action; | |
668 | ref->is_head = 0; | |
669 | ref->in_tree = 1; | |
56bec294 | 670 | |
546adb0d JS |
671 | if (need_ref_seq(for_cow, ref_root)) |
672 | seq = btrfs_get_tree_mod_seq(fs_info, &trans->delayed_ref_elem); | |
00f04b88 AJ |
673 | ref->seq = seq; |
674 | ||
5d4f98a2 | 675 | full_ref = btrfs_delayed_node_to_tree_ref(ref); |
eebe063b AJ |
676 | full_ref->parent = parent; |
677 | full_ref->root = ref_root; | |
678 | if (parent) | |
5d4f98a2 | 679 | ref->type = BTRFS_SHARED_BLOCK_REF_KEY; |
eebe063b | 680 | else |
5d4f98a2 | 681 | ref->type = BTRFS_TREE_BLOCK_REF_KEY; |
5d4f98a2 | 682 | full_ref->level = level; |
56bec294 | 683 | |
599c75ec | 684 | trace_add_delayed_tree_ref(ref, full_ref, action); |
1abe9b8a | 685 | |
d7df2c79 JB |
686 | spin_lock(&head_ref->lock); |
687 | existing = tree_insert(&head_ref->ref_root, &ref->rb_node); | |
56bec294 | 688 | if (existing) { |
d7df2c79 JB |
689 | update_existing_ref(trans, delayed_refs, head_ref, existing, |
690 | ref); | |
5d4f98a2 YZ |
691 | /* |
692 | * we've updated the existing ref, free the newly | |
693 | * allocated ref | |
694 | */ | |
78a6184a | 695 | kmem_cache_free(btrfs_delayed_tree_ref_cachep, full_ref); |
5d4f98a2 | 696 | } else { |
d7df2c79 | 697 | atomic_inc(&delayed_refs->num_entries); |
5d4f98a2 YZ |
698 | trans->delayed_ref_updates++; |
699 | } | |
d7df2c79 | 700 | spin_unlock(&head_ref->lock); |
5d4f98a2 YZ |
701 | } |
702 | ||
703 | /* | |
704 | * helper to insert a delayed data ref into the rbtree. | |
705 | */ | |
d7df2c79 JB |
706 | static noinline void |
707 | add_delayed_data_ref(struct btrfs_fs_info *fs_info, | |
708 | struct btrfs_trans_handle *trans, | |
709 | struct btrfs_delayed_ref_head *head_ref, | |
710 | struct btrfs_delayed_ref_node *ref, u64 bytenr, | |
711 | u64 num_bytes, u64 parent, u64 ref_root, u64 owner, | |
712 | u64 offset, int action, int for_cow) | |
5d4f98a2 YZ |
713 | { |
714 | struct btrfs_delayed_ref_node *existing; | |
715 | struct btrfs_delayed_data_ref *full_ref; | |
716 | struct btrfs_delayed_ref_root *delayed_refs; | |
00f04b88 | 717 | u64 seq = 0; |
5d4f98a2 YZ |
718 | |
719 | if (action == BTRFS_ADD_DELAYED_EXTENT) | |
720 | action = BTRFS_ADD_DELAYED_REF; | |
721 | ||
722 | delayed_refs = &trans->transaction->delayed_refs; | |
723 | ||
724 | /* first set the basic ref node struct up */ | |
725 | atomic_set(&ref->refs, 1); | |
726 | ref->bytenr = bytenr; | |
727 | ref->num_bytes = num_bytes; | |
728 | ref->ref_mod = 1; | |
729 | ref->action = action; | |
730 | ref->is_head = 0; | |
731 | ref->in_tree = 1; | |
732 | ||
546adb0d JS |
733 | if (need_ref_seq(for_cow, ref_root)) |
734 | seq = btrfs_get_tree_mod_seq(fs_info, &trans->delayed_ref_elem); | |
00f04b88 AJ |
735 | ref->seq = seq; |
736 | ||
5d4f98a2 | 737 | full_ref = btrfs_delayed_node_to_data_ref(ref); |
eebe063b AJ |
738 | full_ref->parent = parent; |
739 | full_ref->root = ref_root; | |
740 | if (parent) | |
5d4f98a2 | 741 | ref->type = BTRFS_SHARED_DATA_REF_KEY; |
eebe063b | 742 | else |
5d4f98a2 | 743 | ref->type = BTRFS_EXTENT_DATA_REF_KEY; |
66d7e7f0 | 744 | |
5d4f98a2 YZ |
745 | full_ref->objectid = owner; |
746 | full_ref->offset = offset; | |
56bec294 | 747 | |
599c75ec | 748 | trace_add_delayed_data_ref(ref, full_ref, action); |
1abe9b8a | 749 | |
d7df2c79 JB |
750 | spin_lock(&head_ref->lock); |
751 | existing = tree_insert(&head_ref->ref_root, &ref->rb_node); | |
5d4f98a2 | 752 | if (existing) { |
d7df2c79 JB |
753 | update_existing_ref(trans, delayed_refs, head_ref, existing, |
754 | ref); | |
56bec294 CM |
755 | /* |
756 | * we've updated the existing ref, free the newly | |
757 | * allocated ref | |
758 | */ | |
78a6184a | 759 | kmem_cache_free(btrfs_delayed_data_ref_cachep, full_ref); |
56bec294 | 760 | } else { |
d7df2c79 | 761 | atomic_inc(&delayed_refs->num_entries); |
56bec294 CM |
762 | trans->delayed_ref_updates++; |
763 | } | |
d7df2c79 | 764 | spin_unlock(&head_ref->lock); |
56bec294 CM |
765 | } |
766 | ||
767 | /* | |
5d4f98a2 | 768 | * add a delayed tree ref. This does all of the accounting required |
56bec294 CM |
769 | * to make sure the delayed ref is eventually processed before this |
770 | * transaction commits. | |
771 | */ | |
66d7e7f0 AJ |
772 | int btrfs_add_delayed_tree_ref(struct btrfs_fs_info *fs_info, |
773 | struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
774 | u64 bytenr, u64 num_bytes, u64 parent, |
775 | u64 ref_root, int level, int action, | |
66d7e7f0 AJ |
776 | struct btrfs_delayed_extent_op *extent_op, |
777 | int for_cow) | |
56bec294 | 778 | { |
5d4f98a2 | 779 | struct btrfs_delayed_tree_ref *ref; |
56bec294 CM |
780 | struct btrfs_delayed_ref_head *head_ref; |
781 | struct btrfs_delayed_ref_root *delayed_refs; | |
56bec294 | 782 | |
5d4f98a2 | 783 | BUG_ON(extent_op && extent_op->is_data); |
78a6184a | 784 | ref = kmem_cache_alloc(btrfs_delayed_tree_ref_cachep, GFP_NOFS); |
56bec294 CM |
785 | if (!ref) |
786 | return -ENOMEM; | |
787 | ||
78a6184a | 788 | head_ref = kmem_cache_alloc(btrfs_delayed_ref_head_cachep, GFP_NOFS); |
5d4f98a2 | 789 | if (!head_ref) { |
78a6184a | 790 | kmem_cache_free(btrfs_delayed_tree_ref_cachep, ref); |
5d4f98a2 YZ |
791 | return -ENOMEM; |
792 | } | |
793 | ||
794 | head_ref->extent_op = extent_op; | |
795 | ||
796 | delayed_refs = &trans->transaction->delayed_refs; | |
797 | spin_lock(&delayed_refs->lock); | |
798 | ||
56bec294 | 799 | /* |
5d4f98a2 YZ |
800 | * insert both the head node and the new ref without dropping |
801 | * the spin lock | |
56bec294 | 802 | */ |
d7df2c79 JB |
803 | head_ref = add_delayed_ref_head(fs_info, trans, &head_ref->node, |
804 | bytenr, num_bytes, action, 0); | |
5d4f98a2 | 805 | |
d7df2c79 | 806 | add_delayed_tree_ref(fs_info, trans, head_ref, &ref->node, bytenr, |
66d7e7f0 AJ |
807 | num_bytes, parent, ref_root, level, action, |
808 | for_cow); | |
5d4f98a2 | 809 | spin_unlock(&delayed_refs->lock); |
546adb0d JS |
810 | if (need_ref_seq(for_cow, ref_root)) |
811 | btrfs_qgroup_record_ref(trans, &ref->node, extent_op); | |
95a06077 | 812 | |
5d4f98a2 YZ |
813 | return 0; |
814 | } | |
815 | ||
816 | /* | |
817 | * add a delayed data ref. it's similar to btrfs_add_delayed_tree_ref. | |
818 | */ | |
66d7e7f0 AJ |
819 | int btrfs_add_delayed_data_ref(struct btrfs_fs_info *fs_info, |
820 | struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
821 | u64 bytenr, u64 num_bytes, |
822 | u64 parent, u64 ref_root, | |
823 | u64 owner, u64 offset, int action, | |
66d7e7f0 AJ |
824 | struct btrfs_delayed_extent_op *extent_op, |
825 | int for_cow) | |
5d4f98a2 YZ |
826 | { |
827 | struct btrfs_delayed_data_ref *ref; | |
828 | struct btrfs_delayed_ref_head *head_ref; | |
829 | struct btrfs_delayed_ref_root *delayed_refs; | |
5d4f98a2 YZ |
830 | |
831 | BUG_ON(extent_op && !extent_op->is_data); | |
78a6184a | 832 | ref = kmem_cache_alloc(btrfs_delayed_data_ref_cachep, GFP_NOFS); |
5d4f98a2 YZ |
833 | if (!ref) |
834 | return -ENOMEM; | |
56bec294 | 835 | |
78a6184a | 836 | head_ref = kmem_cache_alloc(btrfs_delayed_ref_head_cachep, GFP_NOFS); |
56bec294 | 837 | if (!head_ref) { |
78a6184a | 838 | kmem_cache_free(btrfs_delayed_data_ref_cachep, ref); |
56bec294 CM |
839 | return -ENOMEM; |
840 | } | |
5d4f98a2 YZ |
841 | |
842 | head_ref->extent_op = extent_op; | |
843 | ||
56bec294 CM |
844 | delayed_refs = &trans->transaction->delayed_refs; |
845 | spin_lock(&delayed_refs->lock); | |
846 | ||
847 | /* | |
848 | * insert both the head node and the new ref without dropping | |
849 | * the spin lock | |
850 | */ | |
d7df2c79 JB |
851 | head_ref = add_delayed_ref_head(fs_info, trans, &head_ref->node, |
852 | bytenr, num_bytes, action, 1); | |
56bec294 | 853 | |
d7df2c79 | 854 | add_delayed_data_ref(fs_info, trans, head_ref, &ref->node, bytenr, |
66d7e7f0 AJ |
855 | num_bytes, parent, ref_root, owner, offset, |
856 | action, for_cow); | |
5d4f98a2 | 857 | spin_unlock(&delayed_refs->lock); |
546adb0d JS |
858 | if (need_ref_seq(for_cow, ref_root)) |
859 | btrfs_qgroup_record_ref(trans, &ref->node, extent_op); | |
95a06077 | 860 | |
5d4f98a2 YZ |
861 | return 0; |
862 | } | |
863 | ||
66d7e7f0 AJ |
864 | int btrfs_add_delayed_extent_op(struct btrfs_fs_info *fs_info, |
865 | struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
866 | u64 bytenr, u64 num_bytes, |
867 | struct btrfs_delayed_extent_op *extent_op) | |
868 | { | |
869 | struct btrfs_delayed_ref_head *head_ref; | |
870 | struct btrfs_delayed_ref_root *delayed_refs; | |
5d4f98a2 | 871 | |
78a6184a | 872 | head_ref = kmem_cache_alloc(btrfs_delayed_ref_head_cachep, GFP_NOFS); |
5d4f98a2 YZ |
873 | if (!head_ref) |
874 | return -ENOMEM; | |
875 | ||
876 | head_ref->extent_op = extent_op; | |
877 | ||
878 | delayed_refs = &trans->transaction->delayed_refs; | |
879 | spin_lock(&delayed_refs->lock); | |
880 | ||
143bede5 | 881 | add_delayed_ref_head(fs_info, trans, &head_ref->node, bytenr, |
5d4f98a2 YZ |
882 | num_bytes, BTRFS_UPDATE_DELAYED_HEAD, |
883 | extent_op->is_data); | |
5d4f98a2 | 884 | |
56bec294 CM |
885 | spin_unlock(&delayed_refs->lock); |
886 | return 0; | |
887 | } | |
888 | ||
1887be66 CM |
889 | /* |
890 | * this does a simple search for the head node for a given extent. | |
891 | * It must be called with the delayed ref spinlock held, and it returns | |
892 | * the head node if any where found, or NULL if not. | |
893 | */ | |
894 | struct btrfs_delayed_ref_head * | |
895 | btrfs_find_delayed_ref_head(struct btrfs_trans_handle *trans, u64 bytenr) | |
896 | { | |
1887be66 CM |
897 | struct btrfs_delayed_ref_root *delayed_refs; |
898 | ||
899 | delayed_refs = &trans->transaction->delayed_refs; | |
c46effa6 | 900 | return find_ref_head(&delayed_refs->href_root, bytenr, NULL, 0); |
1887be66 | 901 | } |
78a6184a MX |
902 | |
903 | void btrfs_delayed_ref_exit(void) | |
904 | { | |
905 | if (btrfs_delayed_ref_head_cachep) | |
906 | kmem_cache_destroy(btrfs_delayed_ref_head_cachep); | |
907 | if (btrfs_delayed_tree_ref_cachep) | |
908 | kmem_cache_destroy(btrfs_delayed_tree_ref_cachep); | |
909 | if (btrfs_delayed_data_ref_cachep) | |
910 | kmem_cache_destroy(btrfs_delayed_data_ref_cachep); | |
911 | if (btrfs_delayed_extent_op_cachep) | |
912 | kmem_cache_destroy(btrfs_delayed_extent_op_cachep); | |
913 | } | |
914 | ||
915 | int btrfs_delayed_ref_init(void) | |
916 | { | |
917 | btrfs_delayed_ref_head_cachep = kmem_cache_create( | |
918 | "btrfs_delayed_ref_head", | |
919 | sizeof(struct btrfs_delayed_ref_head), 0, | |
920 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
921 | if (!btrfs_delayed_ref_head_cachep) | |
922 | goto fail; | |
923 | ||
924 | btrfs_delayed_tree_ref_cachep = kmem_cache_create( | |
925 | "btrfs_delayed_tree_ref", | |
926 | sizeof(struct btrfs_delayed_tree_ref), 0, | |
927 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
928 | if (!btrfs_delayed_tree_ref_cachep) | |
929 | goto fail; | |
930 | ||
931 | btrfs_delayed_data_ref_cachep = kmem_cache_create( | |
932 | "btrfs_delayed_data_ref", | |
933 | sizeof(struct btrfs_delayed_data_ref), 0, | |
934 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
935 | if (!btrfs_delayed_data_ref_cachep) | |
936 | goto fail; | |
937 | ||
938 | btrfs_delayed_extent_op_cachep = kmem_cache_create( | |
939 | "btrfs_delayed_extent_op", | |
940 | sizeof(struct btrfs_delayed_extent_op), 0, | |
941 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
942 | if (!btrfs_delayed_extent_op_cachep) | |
943 | goto fail; | |
944 | ||
945 | return 0; | |
946 | fail: | |
947 | btrfs_delayed_ref_exit(); | |
948 | return -ENOMEM; | |
949 | } |