Commit | Line | Data |
---|---|---|
c1d7c514 | 1 | // SPDX-License-Identifier: GPL-2.0 |
56bec294 CM |
2 | /* |
3 | * Copyright (C) 2009 Oracle. All rights reserved. | |
56bec294 CM |
4 | */ |
5 | ||
6 | #include <linux/sched.h> | |
5a0e3ad6 | 7 | #include <linux/slab.h> |
56bec294 | 8 | #include <linux/sort.h> |
56bec294 CM |
9 | #include "ctree.h" |
10 | #include "delayed-ref.h" | |
11 | #include "transaction.h" | |
3368d001 | 12 | #include "qgroup.h" |
56bec294 | 13 | |
78a6184a MX |
14 | struct kmem_cache *btrfs_delayed_ref_head_cachep; |
15 | struct kmem_cache *btrfs_delayed_tree_ref_cachep; | |
16 | struct kmem_cache *btrfs_delayed_data_ref_cachep; | |
17 | struct kmem_cache *btrfs_delayed_extent_op_cachep; | |
56bec294 CM |
18 | /* |
19 | * delayed back reference update tracking. For subvolume trees | |
20 | * we queue up extent allocations and backref maintenance for | |
21 | * delayed processing. This avoids deep call chains where we | |
22 | * add extents in the middle of btrfs_search_slot, and it allows | |
23 | * us to buffer up frequently modified backrefs in an rb tree instead | |
24 | * of hammering updates on the extent allocation tree. | |
56bec294 CM |
25 | */ |
26 | ||
27 | /* | |
5d4f98a2 YZ |
28 | * compare two delayed tree backrefs with same bytenr and type |
29 | */ | |
c7ad7c84 JB |
30 | static int comp_tree_refs(struct btrfs_delayed_tree_ref *ref1, |
31 | struct btrfs_delayed_tree_ref *ref2) | |
5d4f98a2 | 32 | { |
3b60d436 | 33 | if (ref1->node.type == BTRFS_TREE_BLOCK_REF_KEY) { |
41b0fc42 JB |
34 | if (ref1->root < ref2->root) |
35 | return -1; | |
36 | if (ref1->root > ref2->root) | |
37 | return 1; | |
38 | } else { | |
39 | if (ref1->parent < ref2->parent) | |
40 | return -1; | |
41 | if (ref1->parent > ref2->parent) | |
42 | return 1; | |
43 | } | |
5d4f98a2 YZ |
44 | return 0; |
45 | } | |
46 | ||
47 | /* | |
48 | * compare two delayed data backrefs with same bytenr and type | |
56bec294 | 49 | */ |
c7ad7c84 JB |
50 | static int comp_data_refs(struct btrfs_delayed_data_ref *ref1, |
51 | struct btrfs_delayed_data_ref *ref2) | |
56bec294 | 52 | { |
5d4f98a2 YZ |
53 | if (ref1->node.type == BTRFS_EXTENT_DATA_REF_KEY) { |
54 | if (ref1->root < ref2->root) | |
55 | return -1; | |
56 | if (ref1->root > ref2->root) | |
57 | return 1; | |
58 | if (ref1->objectid < ref2->objectid) | |
59 | return -1; | |
60 | if (ref1->objectid > ref2->objectid) | |
61 | return 1; | |
62 | if (ref1->offset < ref2->offset) | |
63 | return -1; | |
64 | if (ref1->offset > ref2->offset) | |
65 | return 1; | |
66 | } else { | |
67 | if (ref1->parent < ref2->parent) | |
68 | return -1; | |
69 | if (ref1->parent > ref2->parent) | |
70 | return 1; | |
71 | } | |
72 | return 0; | |
73 | } | |
74 | ||
1d148e59 JB |
75 | static int comp_refs(struct btrfs_delayed_ref_node *ref1, |
76 | struct btrfs_delayed_ref_node *ref2, | |
77 | bool check_seq) | |
78 | { | |
79 | int ret = 0; | |
80 | ||
81 | if (ref1->type < ref2->type) | |
82 | return -1; | |
83 | if (ref1->type > ref2->type) | |
84 | return 1; | |
85 | if (ref1->type == BTRFS_TREE_BLOCK_REF_KEY || | |
86 | ref1->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
87 | ret = comp_tree_refs(btrfs_delayed_node_to_tree_ref(ref1), | |
88 | btrfs_delayed_node_to_tree_ref(ref2)); | |
89 | else | |
90 | ret = comp_data_refs(btrfs_delayed_node_to_data_ref(ref1), | |
91 | btrfs_delayed_node_to_data_ref(ref2)); | |
92 | if (ret) | |
93 | return ret; | |
94 | if (check_seq) { | |
95 | if (ref1->seq < ref2->seq) | |
96 | return -1; | |
97 | if (ref1->seq > ref2->seq) | |
98 | return 1; | |
99 | } | |
100 | return 0; | |
101 | } | |
102 | ||
c46effa6 LB |
103 | /* insert a new ref to head ref rbtree */ |
104 | static struct btrfs_delayed_ref_head *htree_insert(struct rb_root *root, | |
105 | struct rb_node *node) | |
106 | { | |
107 | struct rb_node **p = &root->rb_node; | |
108 | struct rb_node *parent_node = NULL; | |
109 | struct btrfs_delayed_ref_head *entry; | |
110 | struct btrfs_delayed_ref_head *ins; | |
111 | u64 bytenr; | |
112 | ||
113 | ins = rb_entry(node, struct btrfs_delayed_ref_head, href_node); | |
d278850e | 114 | bytenr = ins->bytenr; |
c46effa6 LB |
115 | while (*p) { |
116 | parent_node = *p; | |
117 | entry = rb_entry(parent_node, struct btrfs_delayed_ref_head, | |
118 | href_node); | |
119 | ||
d278850e | 120 | if (bytenr < entry->bytenr) |
c46effa6 | 121 | p = &(*p)->rb_left; |
d278850e | 122 | else if (bytenr > entry->bytenr) |
c46effa6 LB |
123 | p = &(*p)->rb_right; |
124 | else | |
125 | return entry; | |
126 | } | |
127 | ||
128 | rb_link_node(node, parent_node, p); | |
129 | rb_insert_color(node, root); | |
130 | return NULL; | |
131 | } | |
132 | ||
0e0adbcf JB |
133 | static struct btrfs_delayed_ref_node* tree_insert(struct rb_root *root, |
134 | struct btrfs_delayed_ref_node *ins) | |
135 | { | |
136 | struct rb_node **p = &root->rb_node; | |
137 | struct rb_node *node = &ins->ref_node; | |
138 | struct rb_node *parent_node = NULL; | |
139 | struct btrfs_delayed_ref_node *entry; | |
140 | ||
141 | while (*p) { | |
142 | int comp; | |
143 | ||
144 | parent_node = *p; | |
145 | entry = rb_entry(parent_node, struct btrfs_delayed_ref_node, | |
146 | ref_node); | |
147 | comp = comp_refs(ins, entry, true); | |
148 | if (comp < 0) | |
149 | p = &(*p)->rb_left; | |
150 | else if (comp > 0) | |
151 | p = &(*p)->rb_right; | |
152 | else | |
153 | return entry; | |
154 | } | |
155 | ||
156 | rb_link_node(node, parent_node, p); | |
157 | rb_insert_color(node, root); | |
158 | return NULL; | |
159 | } | |
160 | ||
56bec294 | 161 | /* |
5d4f98a2 | 162 | * find an head entry based on bytenr. This returns the delayed ref |
d1270cd9 AJ |
163 | * head if it was able to find one, or NULL if nothing was in that spot. |
164 | * If return_bigger is given, the next bigger entry is returned if no exact | |
165 | * match is found. | |
56bec294 | 166 | */ |
c46effa6 LB |
167 | static struct btrfs_delayed_ref_head * |
168 | find_ref_head(struct rb_root *root, u64 bytenr, | |
85fdfdf6 | 169 | int return_bigger) |
56bec294 | 170 | { |
d1270cd9 | 171 | struct rb_node *n; |
c46effa6 | 172 | struct btrfs_delayed_ref_head *entry; |
56bec294 | 173 | |
d1270cd9 AJ |
174 | n = root->rb_node; |
175 | entry = NULL; | |
56bec294 | 176 | while (n) { |
c46effa6 | 177 | entry = rb_entry(n, struct btrfs_delayed_ref_head, href_node); |
56bec294 | 178 | |
d278850e | 179 | if (bytenr < entry->bytenr) |
56bec294 | 180 | n = n->rb_left; |
d278850e | 181 | else if (bytenr > entry->bytenr) |
56bec294 CM |
182 | n = n->rb_right; |
183 | else | |
184 | return entry; | |
185 | } | |
d1270cd9 | 186 | if (entry && return_bigger) { |
d278850e | 187 | if (bytenr > entry->bytenr) { |
c46effa6 | 188 | n = rb_next(&entry->href_node); |
d1270cd9 AJ |
189 | if (!n) |
190 | n = rb_first(root); | |
c46effa6 LB |
191 | entry = rb_entry(n, struct btrfs_delayed_ref_head, |
192 | href_node); | |
6103fb43 | 193 | return entry; |
d1270cd9 AJ |
194 | } |
195 | return entry; | |
196 | } | |
56bec294 CM |
197 | return NULL; |
198 | } | |
199 | ||
c3e69d58 CM |
200 | int btrfs_delayed_ref_lock(struct btrfs_trans_handle *trans, |
201 | struct btrfs_delayed_ref_head *head) | |
56bec294 | 202 | { |
c3e69d58 CM |
203 | struct btrfs_delayed_ref_root *delayed_refs; |
204 | ||
205 | delayed_refs = &trans->transaction->delayed_refs; | |
a4666e68 | 206 | lockdep_assert_held(&delayed_refs->lock); |
c3e69d58 CM |
207 | if (mutex_trylock(&head->mutex)) |
208 | return 0; | |
209 | ||
d278850e | 210 | refcount_inc(&head->refs); |
c3e69d58 CM |
211 | spin_unlock(&delayed_refs->lock); |
212 | ||
213 | mutex_lock(&head->mutex); | |
214 | spin_lock(&delayed_refs->lock); | |
d278850e | 215 | if (RB_EMPTY_NODE(&head->href_node)) { |
c3e69d58 | 216 | mutex_unlock(&head->mutex); |
d278850e | 217 | btrfs_put_delayed_ref_head(head); |
c3e69d58 CM |
218 | return -EAGAIN; |
219 | } | |
d278850e | 220 | btrfs_put_delayed_ref_head(head); |
c3e69d58 CM |
221 | return 0; |
222 | } | |
223 | ||
35a3621b | 224 | static inline void drop_delayed_ref(struct btrfs_trans_handle *trans, |
ae1e206b | 225 | struct btrfs_delayed_ref_root *delayed_refs, |
d7df2c79 | 226 | struct btrfs_delayed_ref_head *head, |
ae1e206b JB |
227 | struct btrfs_delayed_ref_node *ref) |
228 | { | |
a4666e68 | 229 | lockdep_assert_held(&head->lock); |
0e0adbcf JB |
230 | rb_erase(&ref->ref_node, &head->ref_tree); |
231 | RB_CLEAR_NODE(&ref->ref_node); | |
d278850e JB |
232 | if (!list_empty(&ref->add_list)) |
233 | list_del(&ref->add_list); | |
ae1e206b JB |
234 | ref->in_tree = 0; |
235 | btrfs_put_delayed_ref(ref); | |
d7df2c79 | 236 | atomic_dec(&delayed_refs->num_entries); |
ae1e206b JB |
237 | if (trans->delayed_ref_updates) |
238 | trans->delayed_ref_updates--; | |
239 | } | |
240 | ||
2c3cf7d5 FM |
241 | static bool merge_ref(struct btrfs_trans_handle *trans, |
242 | struct btrfs_delayed_ref_root *delayed_refs, | |
243 | struct btrfs_delayed_ref_head *head, | |
244 | struct btrfs_delayed_ref_node *ref, | |
245 | u64 seq) | |
246 | { | |
247 | struct btrfs_delayed_ref_node *next; | |
0e0adbcf | 248 | struct rb_node *node = rb_next(&ref->ref_node); |
2c3cf7d5 FM |
249 | bool done = false; |
250 | ||
0e0adbcf | 251 | while (!done && node) { |
2c3cf7d5 | 252 | int mod; |
2c3cf7d5 | 253 | |
0e0adbcf JB |
254 | next = rb_entry(node, struct btrfs_delayed_ref_node, ref_node); |
255 | node = rb_next(node); | |
2c3cf7d5 | 256 | if (seq && next->seq >= seq) |
0e0adbcf | 257 | break; |
1d148e59 | 258 | if (comp_refs(ref, next, false)) |
0e0adbcf | 259 | break; |
2c3cf7d5 FM |
260 | |
261 | if (ref->action == next->action) { | |
262 | mod = next->ref_mod; | |
263 | } else { | |
264 | if (ref->ref_mod < next->ref_mod) { | |
265 | swap(ref, next); | |
266 | done = true; | |
267 | } | |
268 | mod = -next->ref_mod; | |
269 | } | |
270 | ||
271 | drop_delayed_ref(trans, delayed_refs, head, next); | |
272 | ref->ref_mod += mod; | |
273 | if (ref->ref_mod == 0) { | |
274 | drop_delayed_ref(trans, delayed_refs, head, ref); | |
275 | done = true; | |
276 | } else { | |
277 | /* | |
278 | * Can't have multiples of the same ref on a tree block. | |
279 | */ | |
280 | WARN_ON(ref->type == BTRFS_TREE_BLOCK_REF_KEY || | |
281 | ref->type == BTRFS_SHARED_BLOCK_REF_KEY); | |
282 | } | |
2c3cf7d5 FM |
283 | } |
284 | ||
285 | return done; | |
286 | } | |
287 | ||
288 | void btrfs_merge_delayed_refs(struct btrfs_trans_handle *trans, | |
2c3cf7d5 FM |
289 | struct btrfs_delayed_ref_root *delayed_refs, |
290 | struct btrfs_delayed_ref_head *head) | |
291 | { | |
be97f133 | 292 | struct btrfs_fs_info *fs_info = trans->fs_info; |
2c3cf7d5 | 293 | struct btrfs_delayed_ref_node *ref; |
0e0adbcf | 294 | struct rb_node *node; |
2c3cf7d5 FM |
295 | u64 seq = 0; |
296 | ||
a4666e68 | 297 | lockdep_assert_held(&head->lock); |
2c3cf7d5 | 298 | |
0e0adbcf | 299 | if (RB_EMPTY_ROOT(&head->ref_tree)) |
2c3cf7d5 FM |
300 | return; |
301 | ||
302 | /* We don't have too many refs to merge for data. */ | |
303 | if (head->is_data) | |
304 | return; | |
305 | ||
306 | spin_lock(&fs_info->tree_mod_seq_lock); | |
307 | if (!list_empty(&fs_info->tree_mod_seq_list)) { | |
308 | struct seq_list *elem; | |
309 | ||
310 | elem = list_first_entry(&fs_info->tree_mod_seq_list, | |
311 | struct seq_list, list); | |
312 | seq = elem->seq; | |
313 | } | |
314 | spin_unlock(&fs_info->tree_mod_seq_lock); | |
315 | ||
0e0adbcf JB |
316 | again: |
317 | for (node = rb_first(&head->ref_tree); node; node = rb_next(node)) { | |
318 | ref = rb_entry(node, struct btrfs_delayed_ref_node, ref_node); | |
2c3cf7d5 | 319 | if (seq && ref->seq >= seq) |
2c3cf7d5 | 320 | continue; |
0e0adbcf JB |
321 | if (merge_ref(trans, delayed_refs, head, ref, seq)) |
322 | goto again; | |
2c3cf7d5 FM |
323 | } |
324 | } | |
325 | ||
41d0bd3b | 326 | int btrfs_check_delayed_seq(struct btrfs_fs_info *fs_info, u64 seq) |
00f04b88 AJ |
327 | { |
328 | struct seq_list *elem; | |
097b8a7c JS |
329 | int ret = 0; |
330 | ||
331 | spin_lock(&fs_info->tree_mod_seq_lock); | |
332 | if (!list_empty(&fs_info->tree_mod_seq_list)) { | |
333 | elem = list_first_entry(&fs_info->tree_mod_seq_list, | |
334 | struct seq_list, list); | |
335 | if (seq >= elem->seq) { | |
ab8d0fc4 | 336 | btrfs_debug(fs_info, |
41d0bd3b | 337 | "holding back delayed_ref %#x.%x, lowest is %#x.%x", |
ab8d0fc4 | 338 | (u32)(seq >> 32), (u32)seq, |
41d0bd3b | 339 | (u32)(elem->seq >> 32), (u32)elem->seq); |
097b8a7c JS |
340 | ret = 1; |
341 | } | |
00f04b88 | 342 | } |
097b8a7c JS |
343 | |
344 | spin_unlock(&fs_info->tree_mod_seq_lock); | |
345 | return ret; | |
00f04b88 AJ |
346 | } |
347 | ||
d7df2c79 JB |
348 | struct btrfs_delayed_ref_head * |
349 | btrfs_select_ref_head(struct btrfs_trans_handle *trans) | |
c3e69d58 | 350 | { |
c3e69d58 | 351 | struct btrfs_delayed_ref_root *delayed_refs; |
d7df2c79 JB |
352 | struct btrfs_delayed_ref_head *head; |
353 | u64 start; | |
354 | bool loop = false; | |
56bec294 | 355 | |
c3e69d58 | 356 | delayed_refs = &trans->transaction->delayed_refs; |
c46effa6 | 357 | |
c3e69d58 | 358 | again: |
d7df2c79 | 359 | start = delayed_refs->run_delayed_start; |
85fdfdf6 | 360 | head = find_ref_head(&delayed_refs->href_root, start, 1); |
d7df2c79 JB |
361 | if (!head && !loop) { |
362 | delayed_refs->run_delayed_start = 0; | |
c3e69d58 | 363 | start = 0; |
d7df2c79 | 364 | loop = true; |
85fdfdf6 | 365 | head = find_ref_head(&delayed_refs->href_root, start, 1); |
d7df2c79 JB |
366 | if (!head) |
367 | return NULL; | |
368 | } else if (!head && loop) { | |
369 | return NULL; | |
c3e69d58 | 370 | } |
56bec294 | 371 | |
d7df2c79 JB |
372 | while (head->processing) { |
373 | struct rb_node *node; | |
374 | ||
375 | node = rb_next(&head->href_node); | |
376 | if (!node) { | |
377 | if (loop) | |
378 | return NULL; | |
379 | delayed_refs->run_delayed_start = 0; | |
380 | start = 0; | |
381 | loop = true; | |
382 | goto again; | |
383 | } | |
384 | head = rb_entry(node, struct btrfs_delayed_ref_head, | |
385 | href_node); | |
386 | } | |
093486c4 | 387 | |
d7df2c79 JB |
388 | head->processing = 1; |
389 | WARN_ON(delayed_refs->num_heads_ready == 0); | |
390 | delayed_refs->num_heads_ready--; | |
d278850e JB |
391 | delayed_refs->run_delayed_start = head->bytenr + |
392 | head->num_bytes; | |
d7df2c79 | 393 | return head; |
093486c4 MX |
394 | } |
395 | ||
c6fc2454 QW |
396 | /* |
397 | * Helper to insert the ref_node to the tail or merge with tail. | |
398 | * | |
399 | * Return 0 for insert. | |
400 | * Return >0 for merge. | |
401 | */ | |
0e0adbcf JB |
402 | static int insert_delayed_ref(struct btrfs_trans_handle *trans, |
403 | struct btrfs_delayed_ref_root *root, | |
404 | struct btrfs_delayed_ref_head *href, | |
405 | struct btrfs_delayed_ref_node *ref) | |
c6fc2454 QW |
406 | { |
407 | struct btrfs_delayed_ref_node *exist; | |
408 | int mod; | |
409 | int ret = 0; | |
410 | ||
411 | spin_lock(&href->lock); | |
0e0adbcf JB |
412 | exist = tree_insert(&href->ref_tree, ref); |
413 | if (!exist) | |
414 | goto inserted; | |
c6fc2454 QW |
415 | |
416 | /* Now we are sure we can merge */ | |
417 | ret = 1; | |
418 | if (exist->action == ref->action) { | |
419 | mod = ref->ref_mod; | |
420 | } else { | |
421 | /* Need to change action */ | |
422 | if (exist->ref_mod < ref->ref_mod) { | |
423 | exist->action = ref->action; | |
424 | mod = -exist->ref_mod; | |
425 | exist->ref_mod = ref->ref_mod; | |
1d57ee94 WX |
426 | if (ref->action == BTRFS_ADD_DELAYED_REF) |
427 | list_add_tail(&exist->add_list, | |
428 | &href->ref_add_list); | |
429 | else if (ref->action == BTRFS_DROP_DELAYED_REF) { | |
430 | ASSERT(!list_empty(&exist->add_list)); | |
431 | list_del(&exist->add_list); | |
432 | } else { | |
433 | ASSERT(0); | |
434 | } | |
c6fc2454 QW |
435 | } else |
436 | mod = -ref->ref_mod; | |
437 | } | |
438 | exist->ref_mod += mod; | |
439 | ||
440 | /* remove existing tail if its ref_mod is zero */ | |
441 | if (exist->ref_mod == 0) | |
442 | drop_delayed_ref(trans, root, href, exist); | |
443 | spin_unlock(&href->lock); | |
444 | return ret; | |
0e0adbcf | 445 | inserted: |
1d57ee94 WX |
446 | if (ref->action == BTRFS_ADD_DELAYED_REF) |
447 | list_add_tail(&ref->add_list, &href->ref_add_list); | |
c6fc2454 QW |
448 | atomic_inc(&root->num_entries); |
449 | trans->delayed_ref_updates++; | |
450 | spin_unlock(&href->lock); | |
451 | return ret; | |
452 | } | |
453 | ||
56bec294 CM |
454 | /* |
455 | * helper function to update the accounting in the head ref | |
456 | * existing and update must have the same bytenr | |
457 | */ | |
458 | static noinline void | |
1262133b | 459 | update_existing_head_ref(struct btrfs_delayed_ref_root *delayed_refs, |
d278850e JB |
460 | struct btrfs_delayed_ref_head *existing, |
461 | struct btrfs_delayed_ref_head *update, | |
7be07912 | 462 | int *old_ref_mod_ret) |
56bec294 | 463 | { |
1262133b | 464 | int old_ref_mod; |
56bec294 | 465 | |
d278850e | 466 | BUG_ON(existing->is_data != update->is_data); |
56bec294 | 467 | |
d278850e JB |
468 | spin_lock(&existing->lock); |
469 | if (update->must_insert_reserved) { | |
56bec294 CM |
470 | /* if the extent was freed and then |
471 | * reallocated before the delayed ref | |
472 | * entries were processed, we can end up | |
473 | * with an existing head ref without | |
474 | * the must_insert_reserved flag set. | |
475 | * Set it again here | |
476 | */ | |
d278850e | 477 | existing->must_insert_reserved = update->must_insert_reserved; |
56bec294 CM |
478 | |
479 | /* | |
480 | * update the num_bytes so we make sure the accounting | |
481 | * is done correctly | |
482 | */ | |
483 | existing->num_bytes = update->num_bytes; | |
484 | ||
485 | } | |
486 | ||
d278850e JB |
487 | if (update->extent_op) { |
488 | if (!existing->extent_op) { | |
489 | existing->extent_op = update->extent_op; | |
5d4f98a2 | 490 | } else { |
d278850e JB |
491 | if (update->extent_op->update_key) { |
492 | memcpy(&existing->extent_op->key, | |
493 | &update->extent_op->key, | |
494 | sizeof(update->extent_op->key)); | |
495 | existing->extent_op->update_key = true; | |
5d4f98a2 | 496 | } |
d278850e JB |
497 | if (update->extent_op->update_flags) { |
498 | existing->extent_op->flags_to_set |= | |
499 | update->extent_op->flags_to_set; | |
500 | existing->extent_op->update_flags = true; | |
5d4f98a2 | 501 | } |
d278850e | 502 | btrfs_free_delayed_extent_op(update->extent_op); |
5d4f98a2 YZ |
503 | } |
504 | } | |
56bec294 | 505 | /* |
d7df2c79 JB |
506 | * update the reference mod on the head to reflect this new operation, |
507 | * only need the lock for this case cause we could be processing it | |
508 | * currently, for refs we just added we know we're a-ok. | |
56bec294 | 509 | */ |
d278850e | 510 | old_ref_mod = existing->total_ref_mod; |
7be07912 OS |
511 | if (old_ref_mod_ret) |
512 | *old_ref_mod_ret = old_ref_mod; | |
56bec294 | 513 | existing->ref_mod += update->ref_mod; |
d278850e | 514 | existing->total_ref_mod += update->ref_mod; |
1262133b JB |
515 | |
516 | /* | |
517 | * If we are going to from a positive ref mod to a negative or vice | |
518 | * versa we need to make sure to adjust pending_csums accordingly. | |
519 | */ | |
d278850e JB |
520 | if (existing->is_data) { |
521 | if (existing->total_ref_mod >= 0 && old_ref_mod < 0) | |
1262133b | 522 | delayed_refs->pending_csums -= existing->num_bytes; |
d278850e | 523 | if (existing->total_ref_mod < 0 && old_ref_mod >= 0) |
1262133b JB |
524 | delayed_refs->pending_csums += existing->num_bytes; |
525 | } | |
d278850e | 526 | spin_unlock(&existing->lock); |
56bec294 CM |
527 | } |
528 | ||
a2e569b3 NB |
529 | static void init_delayed_ref_head(struct btrfs_delayed_ref_head *head_ref, |
530 | struct btrfs_qgroup_extent_record *qrecord, | |
531 | u64 bytenr, u64 num_bytes, u64 ref_root, | |
532 | u64 reserved, int action, bool is_data, | |
533 | bool is_system) | |
534 | { | |
535 | int count_mod = 1; | |
536 | int must_insert_reserved = 0; | |
537 | ||
538 | /* If reserved is provided, it must be a data extent. */ | |
539 | BUG_ON(!is_data && reserved); | |
540 | ||
541 | /* | |
542 | * The head node stores the sum of all the mods, so dropping a ref | |
543 | * should drop the sum in the head node by one. | |
544 | */ | |
545 | if (action == BTRFS_UPDATE_DELAYED_HEAD) | |
546 | count_mod = 0; | |
547 | else if (action == BTRFS_DROP_DELAYED_REF) | |
548 | count_mod = -1; | |
549 | ||
550 | /* | |
551 | * BTRFS_ADD_DELAYED_EXTENT means that we need to update the reserved | |
552 | * accounting when the extent is finally added, or if a later | |
553 | * modification deletes the delayed ref without ever inserting the | |
554 | * extent into the extent allocation tree. ref->must_insert_reserved | |
555 | * is the flag used to record that accounting mods are required. | |
556 | * | |
557 | * Once we record must_insert_reserved, switch the action to | |
558 | * BTRFS_ADD_DELAYED_REF because other special casing is not required. | |
559 | */ | |
560 | if (action == BTRFS_ADD_DELAYED_EXTENT) | |
561 | must_insert_reserved = 1; | |
562 | else | |
563 | must_insert_reserved = 0; | |
564 | ||
565 | refcount_set(&head_ref->refs, 1); | |
566 | head_ref->bytenr = bytenr; | |
567 | head_ref->num_bytes = num_bytes; | |
568 | head_ref->ref_mod = count_mod; | |
569 | head_ref->must_insert_reserved = must_insert_reserved; | |
570 | head_ref->is_data = is_data; | |
571 | head_ref->is_system = is_system; | |
572 | head_ref->ref_tree = RB_ROOT; | |
573 | INIT_LIST_HEAD(&head_ref->ref_add_list); | |
574 | RB_CLEAR_NODE(&head_ref->href_node); | |
575 | head_ref->processing = 0; | |
576 | head_ref->total_ref_mod = count_mod; | |
577 | head_ref->qgroup_reserved = 0; | |
578 | head_ref->qgroup_ref_root = 0; | |
579 | spin_lock_init(&head_ref->lock); | |
580 | mutex_init(&head_ref->mutex); | |
581 | ||
582 | if (qrecord) { | |
583 | if (ref_root && reserved) { | |
584 | head_ref->qgroup_ref_root = ref_root; | |
585 | head_ref->qgroup_reserved = reserved; | |
586 | } | |
587 | ||
588 | qrecord->bytenr = bytenr; | |
589 | qrecord->num_bytes = num_bytes; | |
590 | qrecord->old_roots = NULL; | |
591 | } | |
592 | } | |
593 | ||
56bec294 | 594 | /* |
5d4f98a2 | 595 | * helper function to actually insert a head node into the rbtree. |
56bec294 | 596 | * this does all the dirty work in terms of maintaining the correct |
5d4f98a2 | 597 | * overall modification count. |
56bec294 | 598 | */ |
d7df2c79 | 599 | static noinline struct btrfs_delayed_ref_head * |
1acda0c2 | 600 | add_delayed_ref_head(struct btrfs_trans_handle *trans, |
d278850e | 601 | struct btrfs_delayed_ref_head *head_ref, |
3368d001 | 602 | struct btrfs_qgroup_extent_record *qrecord, |
2335efaf | 603 | int action, int *qrecord_inserted_ret, |
7be07912 | 604 | int *old_ref_mod, int *new_ref_mod) |
56bec294 | 605 | { |
d7df2c79 | 606 | struct btrfs_delayed_ref_head *existing; |
56bec294 | 607 | struct btrfs_delayed_ref_root *delayed_refs; |
fb235dc0 | 608 | int qrecord_inserted = 0; |
56bec294 | 609 | |
56bec294 | 610 | delayed_refs = &trans->transaction->delayed_refs; |
2335efaf | 611 | |
3368d001 QW |
612 | /* Record qgroup extent info if provided */ |
613 | if (qrecord) { | |
eb86ec73 | 614 | if (btrfs_qgroup_trace_extent_nolock(trans->fs_info, |
cb93b52c | 615 | delayed_refs, qrecord)) |
3368d001 | 616 | kfree(qrecord); |
fb235dc0 QW |
617 | else |
618 | qrecord_inserted = 1; | |
3368d001 QW |
619 | } |
620 | ||
1acda0c2 | 621 | trace_add_delayed_ref_head(trans->fs_info, head_ref, action); |
1abe9b8a | 622 | |
d7df2c79 JB |
623 | existing = htree_insert(&delayed_refs->href_root, |
624 | &head_ref->href_node); | |
5d4f98a2 | 625 | if (existing) { |
2335efaf NB |
626 | WARN_ON(qrecord && head_ref->qgroup_ref_root |
627 | && head_ref->qgroup_reserved | |
628 | && existing->qgroup_ref_root | |
5846a3c2 | 629 | && existing->qgroup_reserved); |
d278850e | 630 | update_existing_head_ref(delayed_refs, existing, head_ref, |
7be07912 | 631 | old_ref_mod); |
5d4f98a2 YZ |
632 | /* |
633 | * we've updated the existing ref, free the newly | |
634 | * allocated ref | |
635 | */ | |
78a6184a | 636 | kmem_cache_free(btrfs_delayed_ref_head_cachep, head_ref); |
d7df2c79 | 637 | head_ref = existing; |
5d4f98a2 | 638 | } else { |
7be07912 OS |
639 | if (old_ref_mod) |
640 | *old_ref_mod = 0; | |
2335efaf NB |
641 | if (head_ref->is_data && head_ref->ref_mod < 0) |
642 | delayed_refs->pending_csums += head_ref->num_bytes; | |
5d4f98a2 YZ |
643 | delayed_refs->num_heads++; |
644 | delayed_refs->num_heads_ready++; | |
d7df2c79 | 645 | atomic_inc(&delayed_refs->num_entries); |
5d4f98a2 YZ |
646 | trans->delayed_ref_updates++; |
647 | } | |
fb235dc0 QW |
648 | if (qrecord_inserted_ret) |
649 | *qrecord_inserted_ret = qrecord_inserted; | |
7be07912 OS |
650 | if (new_ref_mod) |
651 | *new_ref_mod = head_ref->total_ref_mod; | |
2335efaf | 652 | |
d7df2c79 | 653 | return head_ref; |
5d4f98a2 YZ |
654 | } |
655 | ||
cb49a87b NB |
656 | /* |
657 | * init_delayed_ref_common - Initialize the structure which represents a | |
658 | * modification to a an extent. | |
659 | * | |
660 | * @fs_info: Internal to the mounted filesystem mount structure. | |
661 | * | |
662 | * @ref: The structure which is going to be initialized. | |
663 | * | |
664 | * @bytenr: The logical address of the extent for which a modification is | |
665 | * going to be recorded. | |
666 | * | |
667 | * @num_bytes: Size of the extent whose modification is being recorded. | |
668 | * | |
669 | * @ref_root: The id of the root where this modification has originated, this | |
670 | * can be either one of the well-known metadata trees or the | |
671 | * subvolume id which references this extent. | |
672 | * | |
673 | * @action: Can be one of BTRFS_ADD_DELAYED_REF/BTRFS_DROP_DELAYED_REF or | |
674 | * BTRFS_ADD_DELAYED_EXTENT | |
675 | * | |
676 | * @ref_type: Holds the type of the extent which is being recorded, can be | |
677 | * one of BTRFS_SHARED_BLOCK_REF_KEY/BTRFS_TREE_BLOCK_REF_KEY | |
678 | * when recording a metadata extent or BTRFS_SHARED_DATA_REF_KEY/ | |
679 | * BTRFS_EXTENT_DATA_REF_KEY when recording data extent | |
680 | */ | |
681 | static void init_delayed_ref_common(struct btrfs_fs_info *fs_info, | |
682 | struct btrfs_delayed_ref_node *ref, | |
683 | u64 bytenr, u64 num_bytes, u64 ref_root, | |
684 | int action, u8 ref_type) | |
685 | { | |
686 | u64 seq = 0; | |
687 | ||
688 | if (action == BTRFS_ADD_DELAYED_EXTENT) | |
689 | action = BTRFS_ADD_DELAYED_REF; | |
690 | ||
691 | if (is_fstree(ref_root)) | |
692 | seq = atomic64_read(&fs_info->tree_mod_seq); | |
693 | ||
694 | refcount_set(&ref->refs, 1); | |
695 | ref->bytenr = bytenr; | |
696 | ref->num_bytes = num_bytes; | |
697 | ref->ref_mod = 1; | |
698 | ref->action = action; | |
699 | ref->is_head = 0; | |
700 | ref->in_tree = 1; | |
701 | ref->seq = seq; | |
702 | ref->type = ref_type; | |
703 | RB_CLEAR_NODE(&ref->ref_node); | |
704 | INIT_LIST_HEAD(&ref->add_list); | |
705 | } | |
706 | ||
56bec294 | 707 | /* |
5d4f98a2 | 708 | * add a delayed tree ref. This does all of the accounting required |
56bec294 CM |
709 | * to make sure the delayed ref is eventually processed before this |
710 | * transaction commits. | |
711 | */ | |
44e1c47d | 712 | int btrfs_add_delayed_tree_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
713 | u64 bytenr, u64 num_bytes, u64 parent, |
714 | u64 ref_root, int level, int action, | |
7be07912 OS |
715 | struct btrfs_delayed_extent_op *extent_op, |
716 | int *old_ref_mod, int *new_ref_mod) | |
56bec294 | 717 | { |
44e1c47d | 718 | struct btrfs_fs_info *fs_info = trans->fs_info; |
5d4f98a2 | 719 | struct btrfs_delayed_tree_ref *ref; |
56bec294 CM |
720 | struct btrfs_delayed_ref_head *head_ref; |
721 | struct btrfs_delayed_ref_root *delayed_refs; | |
3368d001 | 722 | struct btrfs_qgroup_extent_record *record = NULL; |
fb235dc0 | 723 | int qrecord_inserted; |
2335efaf | 724 | bool is_system = (ref_root == BTRFS_CHUNK_TREE_OBJECTID); |
70d64000 NB |
725 | int ret; |
726 | u8 ref_type; | |
56bec294 | 727 | |
5d4f98a2 | 728 | BUG_ON(extent_op && extent_op->is_data); |
78a6184a | 729 | ref = kmem_cache_alloc(btrfs_delayed_tree_ref_cachep, GFP_NOFS); |
56bec294 CM |
730 | if (!ref) |
731 | return -ENOMEM; | |
732 | ||
7b4284de NB |
733 | head_ref = kmem_cache_alloc(btrfs_delayed_ref_head_cachep, GFP_NOFS); |
734 | if (!head_ref) { | |
735 | kmem_cache_free(btrfs_delayed_tree_ref_cachep, ref); | |
736 | return -ENOMEM; | |
737 | } | |
738 | ||
739 | if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) && | |
740 | is_fstree(ref_root)) { | |
741 | record = kmalloc(sizeof(*record), GFP_NOFS); | |
742 | if (!record) { | |
743 | kmem_cache_free(btrfs_delayed_tree_ref_cachep, ref); | |
744 | kmem_cache_free(btrfs_delayed_ref_head_cachep, head_ref); | |
745 | return -ENOMEM; | |
746 | } | |
747 | } | |
748 | ||
70d64000 NB |
749 | if (parent) |
750 | ref_type = BTRFS_SHARED_BLOCK_REF_KEY; | |
751 | else | |
752 | ref_type = BTRFS_TREE_BLOCK_REF_KEY; | |
7b4284de | 753 | |
70d64000 NB |
754 | init_delayed_ref_common(fs_info, &ref->node, bytenr, num_bytes, |
755 | ref_root, action, ref_type); | |
756 | ref->root = ref_root; | |
757 | ref->parent = parent; | |
758 | ref->level = level; | |
759 | ||
2335efaf NB |
760 | init_delayed_ref_head(head_ref, record, bytenr, num_bytes, |
761 | ref_root, 0, action, false, is_system); | |
5d4f98a2 YZ |
762 | head_ref->extent_op = extent_op; |
763 | ||
764 | delayed_refs = &trans->transaction->delayed_refs; | |
765 | spin_lock(&delayed_refs->lock); | |
766 | ||
56bec294 | 767 | /* |
5d4f98a2 YZ |
768 | * insert both the head node and the new ref without dropping |
769 | * the spin lock | |
56bec294 | 770 | */ |
2335efaf NB |
771 | head_ref = add_delayed_ref_head(trans, head_ref, record, |
772 | action, &qrecord_inserted, | |
5e388e95 | 773 | old_ref_mod, new_ref_mod); |
5d4f98a2 | 774 | |
70d64000 | 775 | ret = insert_delayed_ref(trans, delayed_refs, head_ref, &ref->node); |
5d4f98a2 | 776 | spin_unlock(&delayed_refs->lock); |
95a06077 | 777 | |
70d64000 NB |
778 | trace_add_delayed_tree_ref(fs_info, &ref->node, ref, |
779 | action == BTRFS_ADD_DELAYED_EXTENT ? | |
780 | BTRFS_ADD_DELAYED_REF : action); | |
781 | if (ret > 0) | |
782 | kmem_cache_free(btrfs_delayed_tree_ref_cachep, ref); | |
783 | ||
fb235dc0 | 784 | if (qrecord_inserted) |
952bd3db NB |
785 | btrfs_qgroup_trace_extent_post(fs_info, record); |
786 | ||
5d4f98a2 YZ |
787 | return 0; |
788 | } | |
789 | ||
790 | /* | |
791 | * add a delayed data ref. it's similar to btrfs_add_delayed_tree_ref. | |
792 | */ | |
88a979c6 | 793 | int btrfs_add_delayed_data_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
794 | u64 bytenr, u64 num_bytes, |
795 | u64 parent, u64 ref_root, | |
7be07912 OS |
796 | u64 owner, u64 offset, u64 reserved, int action, |
797 | int *old_ref_mod, int *new_ref_mod) | |
5d4f98a2 | 798 | { |
88a979c6 | 799 | struct btrfs_fs_info *fs_info = trans->fs_info; |
5d4f98a2 YZ |
800 | struct btrfs_delayed_data_ref *ref; |
801 | struct btrfs_delayed_ref_head *head_ref; | |
802 | struct btrfs_delayed_ref_root *delayed_refs; | |
3368d001 | 803 | struct btrfs_qgroup_extent_record *record = NULL; |
fb235dc0 | 804 | int qrecord_inserted; |
cd7f9699 NB |
805 | int ret; |
806 | u8 ref_type; | |
5d4f98a2 | 807 | |
78a6184a | 808 | ref = kmem_cache_alloc(btrfs_delayed_data_ref_cachep, GFP_NOFS); |
5d4f98a2 YZ |
809 | if (!ref) |
810 | return -ENOMEM; | |
56bec294 | 811 | |
cd7f9699 NB |
812 | if (parent) |
813 | ref_type = BTRFS_SHARED_DATA_REF_KEY; | |
814 | else | |
815 | ref_type = BTRFS_EXTENT_DATA_REF_KEY; | |
816 | init_delayed_ref_common(fs_info, &ref->node, bytenr, num_bytes, | |
817 | ref_root, action, ref_type); | |
818 | ref->root = ref_root; | |
819 | ref->parent = parent; | |
820 | ref->objectid = owner; | |
821 | ref->offset = offset; | |
822 | ||
823 | ||
78a6184a | 824 | head_ref = kmem_cache_alloc(btrfs_delayed_ref_head_cachep, GFP_NOFS); |
56bec294 | 825 | if (!head_ref) { |
78a6184a | 826 | kmem_cache_free(btrfs_delayed_data_ref_cachep, ref); |
56bec294 CM |
827 | return -ENOMEM; |
828 | } | |
5d4f98a2 | 829 | |
afcdd129 JB |
830 | if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) && |
831 | is_fstree(ref_root)) { | |
3368d001 QW |
832 | record = kmalloc(sizeof(*record), GFP_NOFS); |
833 | if (!record) { | |
834 | kmem_cache_free(btrfs_delayed_data_ref_cachep, ref); | |
835 | kmem_cache_free(btrfs_delayed_ref_head_cachep, | |
836 | head_ref); | |
837 | return -ENOMEM; | |
838 | } | |
839 | } | |
840 | ||
2335efaf NB |
841 | init_delayed_ref_head(head_ref, record, bytenr, num_bytes, ref_root, |
842 | reserved, action, true, false); | |
fef394f7 | 843 | head_ref->extent_op = NULL; |
5d4f98a2 | 844 | |
56bec294 CM |
845 | delayed_refs = &trans->transaction->delayed_refs; |
846 | spin_lock(&delayed_refs->lock); | |
847 | ||
848 | /* | |
849 | * insert both the head node and the new ref without dropping | |
850 | * the spin lock | |
851 | */ | |
2335efaf NB |
852 | head_ref = add_delayed_ref_head(trans, head_ref, record, |
853 | action, &qrecord_inserted, | |
7be07912 | 854 | old_ref_mod, new_ref_mod); |
56bec294 | 855 | |
cd7f9699 | 856 | ret = insert_delayed_ref(trans, delayed_refs, head_ref, &ref->node); |
5d4f98a2 | 857 | spin_unlock(&delayed_refs->lock); |
95a06077 | 858 | |
cd7f9699 NB |
859 | trace_add_delayed_data_ref(trans->fs_info, &ref->node, ref, |
860 | action == BTRFS_ADD_DELAYED_EXTENT ? | |
861 | BTRFS_ADD_DELAYED_REF : action); | |
862 | if (ret > 0) | |
863 | kmem_cache_free(btrfs_delayed_data_ref_cachep, ref); | |
864 | ||
865 | ||
fb235dc0 QW |
866 | if (qrecord_inserted) |
867 | return btrfs_qgroup_trace_extent_post(fs_info, record); | |
5d4f98a2 YZ |
868 | return 0; |
869 | } | |
870 | ||
66d7e7f0 AJ |
871 | int btrfs_add_delayed_extent_op(struct btrfs_fs_info *fs_info, |
872 | struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
873 | u64 bytenr, u64 num_bytes, |
874 | struct btrfs_delayed_extent_op *extent_op) | |
875 | { | |
876 | struct btrfs_delayed_ref_head *head_ref; | |
877 | struct btrfs_delayed_ref_root *delayed_refs; | |
5d4f98a2 | 878 | |
78a6184a | 879 | head_ref = kmem_cache_alloc(btrfs_delayed_ref_head_cachep, GFP_NOFS); |
5d4f98a2 YZ |
880 | if (!head_ref) |
881 | return -ENOMEM; | |
882 | ||
2335efaf NB |
883 | init_delayed_ref_head(head_ref, NULL, bytenr, num_bytes, 0, 0, |
884 | BTRFS_UPDATE_DELAYED_HEAD, extent_op->is_data, | |
885 | false); | |
5d4f98a2 YZ |
886 | head_ref->extent_op = extent_op; |
887 | ||
888 | delayed_refs = &trans->transaction->delayed_refs; | |
889 | spin_lock(&delayed_refs->lock); | |
890 | ||
2335efaf NB |
891 | add_delayed_ref_head(trans, head_ref, NULL, BTRFS_UPDATE_DELAYED_HEAD, |
892 | NULL, NULL, NULL); | |
5d4f98a2 | 893 | |
56bec294 CM |
894 | spin_unlock(&delayed_refs->lock); |
895 | return 0; | |
896 | } | |
897 | ||
1887be66 CM |
898 | /* |
899 | * this does a simple search for the head node for a given extent. | |
900 | * It must be called with the delayed ref spinlock held, and it returns | |
901 | * the head node if any where found, or NULL if not. | |
902 | */ | |
903 | struct btrfs_delayed_ref_head * | |
f72ad18e | 904 | btrfs_find_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs, u64 bytenr) |
1887be66 | 905 | { |
85fdfdf6 | 906 | return find_ref_head(&delayed_refs->href_root, bytenr, 0); |
1887be66 | 907 | } |
78a6184a | 908 | |
e67c718b | 909 | void __cold btrfs_delayed_ref_exit(void) |
78a6184a | 910 | { |
5598e900 KM |
911 | kmem_cache_destroy(btrfs_delayed_ref_head_cachep); |
912 | kmem_cache_destroy(btrfs_delayed_tree_ref_cachep); | |
913 | kmem_cache_destroy(btrfs_delayed_data_ref_cachep); | |
914 | kmem_cache_destroy(btrfs_delayed_extent_op_cachep); | |
78a6184a MX |
915 | } |
916 | ||
f5c29bd9 | 917 | int __init btrfs_delayed_ref_init(void) |
78a6184a MX |
918 | { |
919 | btrfs_delayed_ref_head_cachep = kmem_cache_create( | |
920 | "btrfs_delayed_ref_head", | |
921 | sizeof(struct btrfs_delayed_ref_head), 0, | |
fba4b697 | 922 | SLAB_MEM_SPREAD, NULL); |
78a6184a MX |
923 | if (!btrfs_delayed_ref_head_cachep) |
924 | goto fail; | |
925 | ||
926 | btrfs_delayed_tree_ref_cachep = kmem_cache_create( | |
927 | "btrfs_delayed_tree_ref", | |
928 | sizeof(struct btrfs_delayed_tree_ref), 0, | |
fba4b697 | 929 | SLAB_MEM_SPREAD, NULL); |
78a6184a MX |
930 | if (!btrfs_delayed_tree_ref_cachep) |
931 | goto fail; | |
932 | ||
933 | btrfs_delayed_data_ref_cachep = kmem_cache_create( | |
934 | "btrfs_delayed_data_ref", | |
935 | sizeof(struct btrfs_delayed_data_ref), 0, | |
fba4b697 | 936 | SLAB_MEM_SPREAD, NULL); |
78a6184a MX |
937 | if (!btrfs_delayed_data_ref_cachep) |
938 | goto fail; | |
939 | ||
940 | btrfs_delayed_extent_op_cachep = kmem_cache_create( | |
941 | "btrfs_delayed_extent_op", | |
942 | sizeof(struct btrfs_delayed_extent_op), 0, | |
fba4b697 | 943 | SLAB_MEM_SPREAD, NULL); |
78a6184a MX |
944 | if (!btrfs_delayed_extent_op_cachep) |
945 | goto fail; | |
946 | ||
947 | return 0; | |
948 | fail: | |
949 | btrfs_delayed_ref_exit(); | |
950 | return -ENOMEM; | |
951 | } |