]>
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 | 103 | /* insert a new ref to head ref rbtree */ |
5c9d028b | 104 | static struct btrfs_delayed_ref_head *htree_insert(struct rb_root_cached *root, |
c46effa6 LB |
105 | struct rb_node *node) |
106 | { | |
5c9d028b | 107 | struct rb_node **p = &root->rb_root.rb_node; |
c46effa6 LB |
108 | struct rb_node *parent_node = NULL; |
109 | struct btrfs_delayed_ref_head *entry; | |
110 | struct btrfs_delayed_ref_head *ins; | |
111 | u64 bytenr; | |
5c9d028b | 112 | bool leftmost = true; |
c46effa6 LB |
113 | |
114 | ins = rb_entry(node, struct btrfs_delayed_ref_head, href_node); | |
d278850e | 115 | bytenr = ins->bytenr; |
c46effa6 LB |
116 | while (*p) { |
117 | parent_node = *p; | |
118 | entry = rb_entry(parent_node, struct btrfs_delayed_ref_head, | |
119 | href_node); | |
120 | ||
5c9d028b | 121 | if (bytenr < entry->bytenr) { |
c46effa6 | 122 | p = &(*p)->rb_left; |
5c9d028b | 123 | } else if (bytenr > entry->bytenr) { |
c46effa6 | 124 | p = &(*p)->rb_right; |
5c9d028b LB |
125 | leftmost = false; |
126 | } else { | |
c46effa6 | 127 | return entry; |
5c9d028b | 128 | } |
c46effa6 LB |
129 | } |
130 | ||
131 | rb_link_node(node, parent_node, p); | |
5c9d028b | 132 | rb_insert_color_cached(node, root, leftmost); |
c46effa6 LB |
133 | return NULL; |
134 | } | |
135 | ||
e3d03965 | 136 | static struct btrfs_delayed_ref_node* tree_insert(struct rb_root_cached *root, |
0e0adbcf JB |
137 | struct btrfs_delayed_ref_node *ins) |
138 | { | |
e3d03965 | 139 | struct rb_node **p = &root->rb_root.rb_node; |
0e0adbcf JB |
140 | struct rb_node *node = &ins->ref_node; |
141 | struct rb_node *parent_node = NULL; | |
142 | struct btrfs_delayed_ref_node *entry; | |
e3d03965 | 143 | bool leftmost = true; |
0e0adbcf JB |
144 | |
145 | while (*p) { | |
146 | int comp; | |
147 | ||
148 | parent_node = *p; | |
149 | entry = rb_entry(parent_node, struct btrfs_delayed_ref_node, | |
150 | ref_node); | |
151 | comp = comp_refs(ins, entry, true); | |
e3d03965 | 152 | if (comp < 0) { |
0e0adbcf | 153 | p = &(*p)->rb_left; |
e3d03965 | 154 | } else if (comp > 0) { |
0e0adbcf | 155 | p = &(*p)->rb_right; |
e3d03965 LB |
156 | leftmost = false; |
157 | } else { | |
0e0adbcf | 158 | return entry; |
e3d03965 | 159 | } |
0e0adbcf JB |
160 | } |
161 | ||
162 | rb_link_node(node, parent_node, p); | |
e3d03965 | 163 | rb_insert_color_cached(node, root, leftmost); |
0e0adbcf JB |
164 | return NULL; |
165 | } | |
166 | ||
56bec294 | 167 | /* |
5d4f98a2 | 168 | * find an head entry based on bytenr. This returns the delayed ref |
d1270cd9 AJ |
169 | * head if it was able to find one, or NULL if nothing was in that spot. |
170 | * If return_bigger is given, the next bigger entry is returned if no exact | |
171 | * match is found. | |
56bec294 | 172 | */ |
5c9d028b LB |
173 | static struct btrfs_delayed_ref_head* find_ref_head( |
174 | struct btrfs_delayed_ref_root *dr, u64 bytenr, | |
175 | int return_bigger) | |
56bec294 | 176 | { |
5c9d028b | 177 | struct rb_root *root = &dr->href_root.rb_root; |
d1270cd9 | 178 | struct rb_node *n; |
c46effa6 | 179 | struct btrfs_delayed_ref_head *entry; |
56bec294 | 180 | |
d1270cd9 AJ |
181 | n = root->rb_node; |
182 | entry = NULL; | |
56bec294 | 183 | while (n) { |
c46effa6 | 184 | entry = rb_entry(n, struct btrfs_delayed_ref_head, href_node); |
56bec294 | 185 | |
d278850e | 186 | if (bytenr < entry->bytenr) |
56bec294 | 187 | n = n->rb_left; |
d278850e | 188 | else if (bytenr > entry->bytenr) |
56bec294 CM |
189 | n = n->rb_right; |
190 | else | |
191 | return entry; | |
192 | } | |
d1270cd9 | 193 | if (entry && return_bigger) { |
d278850e | 194 | if (bytenr > entry->bytenr) { |
c46effa6 | 195 | n = rb_next(&entry->href_node); |
d1270cd9 | 196 | if (!n) |
5c9d028b | 197 | n = rb_first_cached(&dr->href_root); |
c46effa6 LB |
198 | entry = rb_entry(n, struct btrfs_delayed_ref_head, |
199 | href_node); | |
6103fb43 | 200 | return entry; |
d1270cd9 AJ |
201 | } |
202 | return entry; | |
203 | } | |
56bec294 CM |
204 | return NULL; |
205 | } | |
206 | ||
c3e69d58 CM |
207 | int btrfs_delayed_ref_lock(struct btrfs_trans_handle *trans, |
208 | struct btrfs_delayed_ref_head *head) | |
56bec294 | 209 | { |
c3e69d58 CM |
210 | struct btrfs_delayed_ref_root *delayed_refs; |
211 | ||
212 | delayed_refs = &trans->transaction->delayed_refs; | |
a4666e68 | 213 | lockdep_assert_held(&delayed_refs->lock); |
c3e69d58 CM |
214 | if (mutex_trylock(&head->mutex)) |
215 | return 0; | |
216 | ||
d278850e | 217 | refcount_inc(&head->refs); |
c3e69d58 CM |
218 | spin_unlock(&delayed_refs->lock); |
219 | ||
220 | mutex_lock(&head->mutex); | |
221 | spin_lock(&delayed_refs->lock); | |
d278850e | 222 | if (RB_EMPTY_NODE(&head->href_node)) { |
c3e69d58 | 223 | mutex_unlock(&head->mutex); |
d278850e | 224 | btrfs_put_delayed_ref_head(head); |
c3e69d58 CM |
225 | return -EAGAIN; |
226 | } | |
d278850e | 227 | btrfs_put_delayed_ref_head(head); |
c3e69d58 CM |
228 | return 0; |
229 | } | |
230 | ||
35a3621b | 231 | static inline void drop_delayed_ref(struct btrfs_trans_handle *trans, |
ae1e206b | 232 | struct btrfs_delayed_ref_root *delayed_refs, |
d7df2c79 | 233 | struct btrfs_delayed_ref_head *head, |
ae1e206b JB |
234 | struct btrfs_delayed_ref_node *ref) |
235 | { | |
a4666e68 | 236 | lockdep_assert_held(&head->lock); |
e3d03965 | 237 | rb_erase_cached(&ref->ref_node, &head->ref_tree); |
0e0adbcf | 238 | RB_CLEAR_NODE(&ref->ref_node); |
d278850e JB |
239 | if (!list_empty(&ref->add_list)) |
240 | list_del(&ref->add_list); | |
ae1e206b JB |
241 | ref->in_tree = 0; |
242 | btrfs_put_delayed_ref(ref); | |
d7df2c79 | 243 | atomic_dec(&delayed_refs->num_entries); |
ae1e206b JB |
244 | if (trans->delayed_ref_updates) |
245 | trans->delayed_ref_updates--; | |
246 | } | |
247 | ||
2c3cf7d5 FM |
248 | static bool merge_ref(struct btrfs_trans_handle *trans, |
249 | struct btrfs_delayed_ref_root *delayed_refs, | |
250 | struct btrfs_delayed_ref_head *head, | |
251 | struct btrfs_delayed_ref_node *ref, | |
252 | u64 seq) | |
253 | { | |
254 | struct btrfs_delayed_ref_node *next; | |
0e0adbcf | 255 | struct rb_node *node = rb_next(&ref->ref_node); |
2c3cf7d5 FM |
256 | bool done = false; |
257 | ||
0e0adbcf | 258 | while (!done && node) { |
2c3cf7d5 | 259 | int mod; |
2c3cf7d5 | 260 | |
0e0adbcf JB |
261 | next = rb_entry(node, struct btrfs_delayed_ref_node, ref_node); |
262 | node = rb_next(node); | |
2c3cf7d5 | 263 | if (seq && next->seq >= seq) |
0e0adbcf | 264 | break; |
1d148e59 | 265 | if (comp_refs(ref, next, false)) |
0e0adbcf | 266 | break; |
2c3cf7d5 FM |
267 | |
268 | if (ref->action == next->action) { | |
269 | mod = next->ref_mod; | |
270 | } else { | |
271 | if (ref->ref_mod < next->ref_mod) { | |
272 | swap(ref, next); | |
273 | done = true; | |
274 | } | |
275 | mod = -next->ref_mod; | |
276 | } | |
277 | ||
278 | drop_delayed_ref(trans, delayed_refs, head, next); | |
279 | ref->ref_mod += mod; | |
280 | if (ref->ref_mod == 0) { | |
281 | drop_delayed_ref(trans, delayed_refs, head, ref); | |
282 | done = true; | |
283 | } else { | |
284 | /* | |
285 | * Can't have multiples of the same ref on a tree block. | |
286 | */ | |
287 | WARN_ON(ref->type == BTRFS_TREE_BLOCK_REF_KEY || | |
288 | ref->type == BTRFS_SHARED_BLOCK_REF_KEY); | |
289 | } | |
2c3cf7d5 FM |
290 | } |
291 | ||
292 | return done; | |
293 | } | |
294 | ||
295 | void btrfs_merge_delayed_refs(struct btrfs_trans_handle *trans, | |
2c3cf7d5 FM |
296 | struct btrfs_delayed_ref_root *delayed_refs, |
297 | struct btrfs_delayed_ref_head *head) | |
298 | { | |
be97f133 | 299 | struct btrfs_fs_info *fs_info = trans->fs_info; |
2c3cf7d5 | 300 | struct btrfs_delayed_ref_node *ref; |
0e0adbcf | 301 | struct rb_node *node; |
2c3cf7d5 FM |
302 | u64 seq = 0; |
303 | ||
a4666e68 | 304 | lockdep_assert_held(&head->lock); |
2c3cf7d5 | 305 | |
e3d03965 | 306 | if (RB_EMPTY_ROOT(&head->ref_tree.rb_root)) |
2c3cf7d5 FM |
307 | return; |
308 | ||
309 | /* We don't have too many refs to merge for data. */ | |
310 | if (head->is_data) | |
311 | return; | |
312 | ||
313 | spin_lock(&fs_info->tree_mod_seq_lock); | |
314 | if (!list_empty(&fs_info->tree_mod_seq_list)) { | |
315 | struct seq_list *elem; | |
316 | ||
317 | elem = list_first_entry(&fs_info->tree_mod_seq_list, | |
318 | struct seq_list, list); | |
319 | seq = elem->seq; | |
320 | } | |
321 | spin_unlock(&fs_info->tree_mod_seq_lock); | |
322 | ||
0e0adbcf | 323 | again: |
e3d03965 LB |
324 | for (node = rb_first_cached(&head->ref_tree); node; |
325 | node = rb_next(node)) { | |
0e0adbcf | 326 | ref = rb_entry(node, struct btrfs_delayed_ref_node, ref_node); |
2c3cf7d5 | 327 | if (seq && ref->seq >= seq) |
2c3cf7d5 | 328 | continue; |
0e0adbcf JB |
329 | if (merge_ref(trans, delayed_refs, head, ref, seq)) |
330 | goto again; | |
2c3cf7d5 FM |
331 | } |
332 | } | |
333 | ||
41d0bd3b | 334 | int btrfs_check_delayed_seq(struct btrfs_fs_info *fs_info, u64 seq) |
00f04b88 AJ |
335 | { |
336 | struct seq_list *elem; | |
097b8a7c JS |
337 | int ret = 0; |
338 | ||
339 | spin_lock(&fs_info->tree_mod_seq_lock); | |
340 | if (!list_empty(&fs_info->tree_mod_seq_list)) { | |
341 | elem = list_first_entry(&fs_info->tree_mod_seq_list, | |
342 | struct seq_list, list); | |
343 | if (seq >= elem->seq) { | |
ab8d0fc4 | 344 | btrfs_debug(fs_info, |
41d0bd3b | 345 | "holding back delayed_ref %#x.%x, lowest is %#x.%x", |
ab8d0fc4 | 346 | (u32)(seq >> 32), (u32)seq, |
41d0bd3b | 347 | (u32)(elem->seq >> 32), (u32)elem->seq); |
097b8a7c JS |
348 | ret = 1; |
349 | } | |
00f04b88 | 350 | } |
097b8a7c JS |
351 | |
352 | spin_unlock(&fs_info->tree_mod_seq_lock); | |
353 | return ret; | |
00f04b88 AJ |
354 | } |
355 | ||
5637c74b LF |
356 | struct btrfs_delayed_ref_head *btrfs_select_ref_head( |
357 | struct btrfs_delayed_ref_root *delayed_refs) | |
c3e69d58 | 358 | { |
d7df2c79 JB |
359 | struct btrfs_delayed_ref_head *head; |
360 | u64 start; | |
361 | bool loop = false; | |
56bec294 | 362 | |
c3e69d58 | 363 | again: |
d7df2c79 | 364 | start = delayed_refs->run_delayed_start; |
5c9d028b | 365 | head = find_ref_head(delayed_refs, start, 1); |
d7df2c79 JB |
366 | if (!head && !loop) { |
367 | delayed_refs->run_delayed_start = 0; | |
c3e69d58 | 368 | start = 0; |
d7df2c79 | 369 | loop = true; |
5c9d028b | 370 | head = find_ref_head(delayed_refs, start, 1); |
d7df2c79 JB |
371 | if (!head) |
372 | return NULL; | |
373 | } else if (!head && loop) { | |
374 | return NULL; | |
c3e69d58 | 375 | } |
56bec294 | 376 | |
d7df2c79 JB |
377 | while (head->processing) { |
378 | struct rb_node *node; | |
379 | ||
380 | node = rb_next(&head->href_node); | |
381 | if (!node) { | |
382 | if (loop) | |
383 | return NULL; | |
384 | delayed_refs->run_delayed_start = 0; | |
385 | start = 0; | |
386 | loop = true; | |
387 | goto again; | |
388 | } | |
389 | head = rb_entry(node, struct btrfs_delayed_ref_head, | |
390 | href_node); | |
391 | } | |
093486c4 | 392 | |
d7df2c79 JB |
393 | head->processing = 1; |
394 | WARN_ON(delayed_refs->num_heads_ready == 0); | |
395 | delayed_refs->num_heads_ready--; | |
d278850e JB |
396 | delayed_refs->run_delayed_start = head->bytenr + |
397 | head->num_bytes; | |
d7df2c79 | 398 | return head; |
093486c4 MX |
399 | } |
400 | ||
c6fc2454 QW |
401 | /* |
402 | * Helper to insert the ref_node to the tail or merge with tail. | |
403 | * | |
404 | * Return 0 for insert. | |
405 | * Return >0 for merge. | |
406 | */ | |
0e0adbcf JB |
407 | static int insert_delayed_ref(struct btrfs_trans_handle *trans, |
408 | struct btrfs_delayed_ref_root *root, | |
409 | struct btrfs_delayed_ref_head *href, | |
410 | struct btrfs_delayed_ref_node *ref) | |
c6fc2454 QW |
411 | { |
412 | struct btrfs_delayed_ref_node *exist; | |
413 | int mod; | |
414 | int ret = 0; | |
415 | ||
416 | spin_lock(&href->lock); | |
0e0adbcf JB |
417 | exist = tree_insert(&href->ref_tree, ref); |
418 | if (!exist) | |
419 | goto inserted; | |
c6fc2454 QW |
420 | |
421 | /* Now we are sure we can merge */ | |
422 | ret = 1; | |
423 | if (exist->action == ref->action) { | |
424 | mod = ref->ref_mod; | |
425 | } else { | |
426 | /* Need to change action */ | |
427 | if (exist->ref_mod < ref->ref_mod) { | |
428 | exist->action = ref->action; | |
429 | mod = -exist->ref_mod; | |
430 | exist->ref_mod = ref->ref_mod; | |
1d57ee94 WX |
431 | if (ref->action == BTRFS_ADD_DELAYED_REF) |
432 | list_add_tail(&exist->add_list, | |
433 | &href->ref_add_list); | |
434 | else if (ref->action == BTRFS_DROP_DELAYED_REF) { | |
435 | ASSERT(!list_empty(&exist->add_list)); | |
436 | list_del(&exist->add_list); | |
437 | } else { | |
438 | ASSERT(0); | |
439 | } | |
c6fc2454 QW |
440 | } else |
441 | mod = -ref->ref_mod; | |
442 | } | |
443 | exist->ref_mod += mod; | |
444 | ||
445 | /* remove existing tail if its ref_mod is zero */ | |
446 | if (exist->ref_mod == 0) | |
447 | drop_delayed_ref(trans, root, href, exist); | |
448 | spin_unlock(&href->lock); | |
449 | return ret; | |
0e0adbcf | 450 | inserted: |
1d57ee94 WX |
451 | if (ref->action == BTRFS_ADD_DELAYED_REF) |
452 | list_add_tail(&ref->add_list, &href->ref_add_list); | |
c6fc2454 QW |
453 | atomic_inc(&root->num_entries); |
454 | trans->delayed_ref_updates++; | |
455 | spin_unlock(&href->lock); | |
456 | return ret; | |
457 | } | |
458 | ||
56bec294 CM |
459 | /* |
460 | * helper function to update the accounting in the head ref | |
461 | * existing and update must have the same bytenr | |
462 | */ | |
463 | static noinline void | |
1262133b | 464 | update_existing_head_ref(struct btrfs_delayed_ref_root *delayed_refs, |
d278850e JB |
465 | struct btrfs_delayed_ref_head *existing, |
466 | struct btrfs_delayed_ref_head *update, | |
7be07912 | 467 | int *old_ref_mod_ret) |
56bec294 | 468 | { |
1262133b | 469 | int old_ref_mod; |
56bec294 | 470 | |
d278850e | 471 | BUG_ON(existing->is_data != update->is_data); |
56bec294 | 472 | |
d278850e JB |
473 | spin_lock(&existing->lock); |
474 | if (update->must_insert_reserved) { | |
56bec294 CM |
475 | /* if the extent was freed and then |
476 | * reallocated before the delayed ref | |
477 | * entries were processed, we can end up | |
478 | * with an existing head ref without | |
479 | * the must_insert_reserved flag set. | |
480 | * Set it again here | |
481 | */ | |
d278850e | 482 | existing->must_insert_reserved = update->must_insert_reserved; |
56bec294 CM |
483 | |
484 | /* | |
485 | * update the num_bytes so we make sure the accounting | |
486 | * is done correctly | |
487 | */ | |
488 | existing->num_bytes = update->num_bytes; | |
489 | ||
490 | } | |
491 | ||
d278850e JB |
492 | if (update->extent_op) { |
493 | if (!existing->extent_op) { | |
494 | existing->extent_op = update->extent_op; | |
5d4f98a2 | 495 | } else { |
d278850e JB |
496 | if (update->extent_op->update_key) { |
497 | memcpy(&existing->extent_op->key, | |
498 | &update->extent_op->key, | |
499 | sizeof(update->extent_op->key)); | |
500 | existing->extent_op->update_key = true; | |
5d4f98a2 | 501 | } |
d278850e JB |
502 | if (update->extent_op->update_flags) { |
503 | existing->extent_op->flags_to_set |= | |
504 | update->extent_op->flags_to_set; | |
505 | existing->extent_op->update_flags = true; | |
5d4f98a2 | 506 | } |
d278850e | 507 | btrfs_free_delayed_extent_op(update->extent_op); |
5d4f98a2 YZ |
508 | } |
509 | } | |
56bec294 | 510 | /* |
d7df2c79 JB |
511 | * update the reference mod on the head to reflect this new operation, |
512 | * only need the lock for this case cause we could be processing it | |
513 | * currently, for refs we just added we know we're a-ok. | |
56bec294 | 514 | */ |
d278850e | 515 | old_ref_mod = existing->total_ref_mod; |
7be07912 OS |
516 | if (old_ref_mod_ret) |
517 | *old_ref_mod_ret = old_ref_mod; | |
56bec294 | 518 | existing->ref_mod += update->ref_mod; |
d278850e | 519 | existing->total_ref_mod += update->ref_mod; |
1262133b JB |
520 | |
521 | /* | |
522 | * If we are going to from a positive ref mod to a negative or vice | |
523 | * versa we need to make sure to adjust pending_csums accordingly. | |
524 | */ | |
d278850e JB |
525 | if (existing->is_data) { |
526 | if (existing->total_ref_mod >= 0 && old_ref_mod < 0) | |
1262133b | 527 | delayed_refs->pending_csums -= existing->num_bytes; |
d278850e | 528 | if (existing->total_ref_mod < 0 && old_ref_mod >= 0) |
1262133b JB |
529 | delayed_refs->pending_csums += existing->num_bytes; |
530 | } | |
d278850e | 531 | spin_unlock(&existing->lock); |
56bec294 CM |
532 | } |
533 | ||
a2e569b3 NB |
534 | static void init_delayed_ref_head(struct btrfs_delayed_ref_head *head_ref, |
535 | struct btrfs_qgroup_extent_record *qrecord, | |
536 | u64 bytenr, u64 num_bytes, u64 ref_root, | |
537 | u64 reserved, int action, bool is_data, | |
538 | bool is_system) | |
539 | { | |
540 | int count_mod = 1; | |
541 | int must_insert_reserved = 0; | |
542 | ||
543 | /* If reserved is provided, it must be a data extent. */ | |
544 | BUG_ON(!is_data && reserved); | |
545 | ||
546 | /* | |
547 | * The head node stores the sum of all the mods, so dropping a ref | |
548 | * should drop the sum in the head node by one. | |
549 | */ | |
550 | if (action == BTRFS_UPDATE_DELAYED_HEAD) | |
551 | count_mod = 0; | |
552 | else if (action == BTRFS_DROP_DELAYED_REF) | |
553 | count_mod = -1; | |
554 | ||
555 | /* | |
556 | * BTRFS_ADD_DELAYED_EXTENT means that we need to update the reserved | |
557 | * accounting when the extent is finally added, or if a later | |
558 | * modification deletes the delayed ref without ever inserting the | |
559 | * extent into the extent allocation tree. ref->must_insert_reserved | |
560 | * is the flag used to record that accounting mods are required. | |
561 | * | |
562 | * Once we record must_insert_reserved, switch the action to | |
563 | * BTRFS_ADD_DELAYED_REF because other special casing is not required. | |
564 | */ | |
565 | if (action == BTRFS_ADD_DELAYED_EXTENT) | |
566 | must_insert_reserved = 1; | |
567 | else | |
568 | must_insert_reserved = 0; | |
569 | ||
570 | refcount_set(&head_ref->refs, 1); | |
571 | head_ref->bytenr = bytenr; | |
572 | head_ref->num_bytes = num_bytes; | |
573 | head_ref->ref_mod = count_mod; | |
574 | head_ref->must_insert_reserved = must_insert_reserved; | |
575 | head_ref->is_data = is_data; | |
576 | head_ref->is_system = is_system; | |
e3d03965 | 577 | head_ref->ref_tree = RB_ROOT_CACHED; |
a2e569b3 NB |
578 | INIT_LIST_HEAD(&head_ref->ref_add_list); |
579 | RB_CLEAR_NODE(&head_ref->href_node); | |
580 | head_ref->processing = 0; | |
581 | head_ref->total_ref_mod = count_mod; | |
582 | head_ref->qgroup_reserved = 0; | |
583 | head_ref->qgroup_ref_root = 0; | |
584 | spin_lock_init(&head_ref->lock); | |
585 | mutex_init(&head_ref->mutex); | |
586 | ||
587 | if (qrecord) { | |
588 | if (ref_root && reserved) { | |
589 | head_ref->qgroup_ref_root = ref_root; | |
590 | head_ref->qgroup_reserved = reserved; | |
591 | } | |
592 | ||
593 | qrecord->bytenr = bytenr; | |
594 | qrecord->num_bytes = num_bytes; | |
595 | qrecord->old_roots = NULL; | |
596 | } | |
597 | } | |
598 | ||
56bec294 | 599 | /* |
5d4f98a2 | 600 | * helper function to actually insert a head node into the rbtree. |
56bec294 | 601 | * this does all the dirty work in terms of maintaining the correct |
5d4f98a2 | 602 | * overall modification count. |
56bec294 | 603 | */ |
d7df2c79 | 604 | static noinline struct btrfs_delayed_ref_head * |
1acda0c2 | 605 | add_delayed_ref_head(struct btrfs_trans_handle *trans, |
d278850e | 606 | struct btrfs_delayed_ref_head *head_ref, |
3368d001 | 607 | struct btrfs_qgroup_extent_record *qrecord, |
2335efaf | 608 | int action, int *qrecord_inserted_ret, |
7be07912 | 609 | int *old_ref_mod, int *new_ref_mod) |
56bec294 | 610 | { |
d7df2c79 | 611 | struct btrfs_delayed_ref_head *existing; |
56bec294 | 612 | struct btrfs_delayed_ref_root *delayed_refs; |
fb235dc0 | 613 | int qrecord_inserted = 0; |
56bec294 | 614 | |
56bec294 | 615 | delayed_refs = &trans->transaction->delayed_refs; |
2335efaf | 616 | |
3368d001 QW |
617 | /* Record qgroup extent info if provided */ |
618 | if (qrecord) { | |
eb86ec73 | 619 | if (btrfs_qgroup_trace_extent_nolock(trans->fs_info, |
cb93b52c | 620 | delayed_refs, qrecord)) |
3368d001 | 621 | kfree(qrecord); |
fb235dc0 QW |
622 | else |
623 | qrecord_inserted = 1; | |
3368d001 QW |
624 | } |
625 | ||
1acda0c2 | 626 | trace_add_delayed_ref_head(trans->fs_info, head_ref, action); |
1abe9b8a | 627 | |
d7df2c79 JB |
628 | existing = htree_insert(&delayed_refs->href_root, |
629 | &head_ref->href_node); | |
5d4f98a2 | 630 | if (existing) { |
2335efaf NB |
631 | WARN_ON(qrecord && head_ref->qgroup_ref_root |
632 | && head_ref->qgroup_reserved | |
633 | && existing->qgroup_ref_root | |
5846a3c2 | 634 | && existing->qgroup_reserved); |
d278850e | 635 | update_existing_head_ref(delayed_refs, existing, head_ref, |
7be07912 | 636 | old_ref_mod); |
5d4f98a2 YZ |
637 | /* |
638 | * we've updated the existing ref, free the newly | |
639 | * allocated ref | |
640 | */ | |
78a6184a | 641 | kmem_cache_free(btrfs_delayed_ref_head_cachep, head_ref); |
d7df2c79 | 642 | head_ref = existing; |
5d4f98a2 | 643 | } else { |
7be07912 OS |
644 | if (old_ref_mod) |
645 | *old_ref_mod = 0; | |
2335efaf NB |
646 | if (head_ref->is_data && head_ref->ref_mod < 0) |
647 | delayed_refs->pending_csums += head_ref->num_bytes; | |
5d4f98a2 YZ |
648 | delayed_refs->num_heads++; |
649 | delayed_refs->num_heads_ready++; | |
d7df2c79 | 650 | atomic_inc(&delayed_refs->num_entries); |
5d4f98a2 YZ |
651 | trans->delayed_ref_updates++; |
652 | } | |
fb235dc0 QW |
653 | if (qrecord_inserted_ret) |
654 | *qrecord_inserted_ret = qrecord_inserted; | |
7be07912 OS |
655 | if (new_ref_mod) |
656 | *new_ref_mod = head_ref->total_ref_mod; | |
2335efaf | 657 | |
d7df2c79 | 658 | return head_ref; |
5d4f98a2 YZ |
659 | } |
660 | ||
cb49a87b NB |
661 | /* |
662 | * init_delayed_ref_common - Initialize the structure which represents a | |
663 | * modification to a an extent. | |
664 | * | |
665 | * @fs_info: Internal to the mounted filesystem mount structure. | |
666 | * | |
667 | * @ref: The structure which is going to be initialized. | |
668 | * | |
669 | * @bytenr: The logical address of the extent for which a modification is | |
670 | * going to be recorded. | |
671 | * | |
672 | * @num_bytes: Size of the extent whose modification is being recorded. | |
673 | * | |
674 | * @ref_root: The id of the root where this modification has originated, this | |
675 | * can be either one of the well-known metadata trees or the | |
676 | * subvolume id which references this extent. | |
677 | * | |
678 | * @action: Can be one of BTRFS_ADD_DELAYED_REF/BTRFS_DROP_DELAYED_REF or | |
679 | * BTRFS_ADD_DELAYED_EXTENT | |
680 | * | |
681 | * @ref_type: Holds the type of the extent which is being recorded, can be | |
682 | * one of BTRFS_SHARED_BLOCK_REF_KEY/BTRFS_TREE_BLOCK_REF_KEY | |
683 | * when recording a metadata extent or BTRFS_SHARED_DATA_REF_KEY/ | |
684 | * BTRFS_EXTENT_DATA_REF_KEY when recording data extent | |
685 | */ | |
686 | static void init_delayed_ref_common(struct btrfs_fs_info *fs_info, | |
687 | struct btrfs_delayed_ref_node *ref, | |
688 | u64 bytenr, u64 num_bytes, u64 ref_root, | |
689 | int action, u8 ref_type) | |
690 | { | |
691 | u64 seq = 0; | |
692 | ||
693 | if (action == BTRFS_ADD_DELAYED_EXTENT) | |
694 | action = BTRFS_ADD_DELAYED_REF; | |
695 | ||
696 | if (is_fstree(ref_root)) | |
697 | seq = atomic64_read(&fs_info->tree_mod_seq); | |
698 | ||
699 | refcount_set(&ref->refs, 1); | |
700 | ref->bytenr = bytenr; | |
701 | ref->num_bytes = num_bytes; | |
702 | ref->ref_mod = 1; | |
703 | ref->action = action; | |
704 | ref->is_head = 0; | |
705 | ref->in_tree = 1; | |
706 | ref->seq = seq; | |
707 | ref->type = ref_type; | |
708 | RB_CLEAR_NODE(&ref->ref_node); | |
709 | INIT_LIST_HEAD(&ref->add_list); | |
710 | } | |
711 | ||
56bec294 | 712 | /* |
5d4f98a2 | 713 | * add a delayed tree ref. This does all of the accounting required |
56bec294 CM |
714 | * to make sure the delayed ref is eventually processed before this |
715 | * transaction commits. | |
716 | */ | |
44e1c47d | 717 | int btrfs_add_delayed_tree_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
718 | u64 bytenr, u64 num_bytes, u64 parent, |
719 | u64 ref_root, int level, int action, | |
7be07912 OS |
720 | struct btrfs_delayed_extent_op *extent_op, |
721 | int *old_ref_mod, int *new_ref_mod) | |
56bec294 | 722 | { |
44e1c47d | 723 | struct btrfs_fs_info *fs_info = trans->fs_info; |
5d4f98a2 | 724 | struct btrfs_delayed_tree_ref *ref; |
56bec294 CM |
725 | struct btrfs_delayed_ref_head *head_ref; |
726 | struct btrfs_delayed_ref_root *delayed_refs; | |
3368d001 | 727 | struct btrfs_qgroup_extent_record *record = NULL; |
fb235dc0 | 728 | int qrecord_inserted; |
2335efaf | 729 | bool is_system = (ref_root == BTRFS_CHUNK_TREE_OBJECTID); |
70d64000 NB |
730 | int ret; |
731 | u8 ref_type; | |
56bec294 | 732 | |
5d4f98a2 | 733 | BUG_ON(extent_op && extent_op->is_data); |
78a6184a | 734 | ref = kmem_cache_alloc(btrfs_delayed_tree_ref_cachep, GFP_NOFS); |
56bec294 CM |
735 | if (!ref) |
736 | return -ENOMEM; | |
737 | ||
7b4284de NB |
738 | head_ref = kmem_cache_alloc(btrfs_delayed_ref_head_cachep, GFP_NOFS); |
739 | if (!head_ref) { | |
740 | kmem_cache_free(btrfs_delayed_tree_ref_cachep, ref); | |
741 | return -ENOMEM; | |
742 | } | |
743 | ||
744 | if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) && | |
745 | is_fstree(ref_root)) { | |
746 | record = kmalloc(sizeof(*record), GFP_NOFS); | |
747 | if (!record) { | |
748 | kmem_cache_free(btrfs_delayed_tree_ref_cachep, ref); | |
749 | kmem_cache_free(btrfs_delayed_ref_head_cachep, head_ref); | |
750 | return -ENOMEM; | |
751 | } | |
752 | } | |
753 | ||
70d64000 NB |
754 | if (parent) |
755 | ref_type = BTRFS_SHARED_BLOCK_REF_KEY; | |
756 | else | |
757 | ref_type = BTRFS_TREE_BLOCK_REF_KEY; | |
7b4284de | 758 | |
70d64000 NB |
759 | init_delayed_ref_common(fs_info, &ref->node, bytenr, num_bytes, |
760 | ref_root, action, ref_type); | |
761 | ref->root = ref_root; | |
762 | ref->parent = parent; | |
763 | ref->level = level; | |
764 | ||
2335efaf NB |
765 | init_delayed_ref_head(head_ref, record, bytenr, num_bytes, |
766 | ref_root, 0, action, false, is_system); | |
5d4f98a2 YZ |
767 | head_ref->extent_op = extent_op; |
768 | ||
769 | delayed_refs = &trans->transaction->delayed_refs; | |
770 | spin_lock(&delayed_refs->lock); | |
771 | ||
56bec294 | 772 | /* |
5d4f98a2 YZ |
773 | * insert both the head node and the new ref without dropping |
774 | * the spin lock | |
56bec294 | 775 | */ |
2335efaf NB |
776 | head_ref = add_delayed_ref_head(trans, head_ref, record, |
777 | action, &qrecord_inserted, | |
5e388e95 | 778 | old_ref_mod, new_ref_mod); |
5d4f98a2 | 779 | |
70d64000 | 780 | ret = insert_delayed_ref(trans, delayed_refs, head_ref, &ref->node); |
5d4f98a2 | 781 | spin_unlock(&delayed_refs->lock); |
95a06077 | 782 | |
70d64000 NB |
783 | trace_add_delayed_tree_ref(fs_info, &ref->node, ref, |
784 | action == BTRFS_ADD_DELAYED_EXTENT ? | |
785 | BTRFS_ADD_DELAYED_REF : action); | |
786 | if (ret > 0) | |
787 | kmem_cache_free(btrfs_delayed_tree_ref_cachep, ref); | |
788 | ||
fb235dc0 | 789 | if (qrecord_inserted) |
952bd3db NB |
790 | btrfs_qgroup_trace_extent_post(fs_info, record); |
791 | ||
5d4f98a2 YZ |
792 | return 0; |
793 | } | |
794 | ||
795 | /* | |
796 | * add a delayed data ref. it's similar to btrfs_add_delayed_tree_ref. | |
797 | */ | |
88a979c6 | 798 | int btrfs_add_delayed_data_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
799 | u64 bytenr, u64 num_bytes, |
800 | u64 parent, u64 ref_root, | |
7be07912 OS |
801 | u64 owner, u64 offset, u64 reserved, int action, |
802 | int *old_ref_mod, int *new_ref_mod) | |
5d4f98a2 | 803 | { |
88a979c6 | 804 | struct btrfs_fs_info *fs_info = trans->fs_info; |
5d4f98a2 YZ |
805 | struct btrfs_delayed_data_ref *ref; |
806 | struct btrfs_delayed_ref_head *head_ref; | |
807 | struct btrfs_delayed_ref_root *delayed_refs; | |
3368d001 | 808 | struct btrfs_qgroup_extent_record *record = NULL; |
fb235dc0 | 809 | int qrecord_inserted; |
cd7f9699 NB |
810 | int ret; |
811 | u8 ref_type; | |
5d4f98a2 | 812 | |
78a6184a | 813 | ref = kmem_cache_alloc(btrfs_delayed_data_ref_cachep, GFP_NOFS); |
5d4f98a2 YZ |
814 | if (!ref) |
815 | return -ENOMEM; | |
56bec294 | 816 | |
cd7f9699 NB |
817 | if (parent) |
818 | ref_type = BTRFS_SHARED_DATA_REF_KEY; | |
819 | else | |
820 | ref_type = BTRFS_EXTENT_DATA_REF_KEY; | |
821 | init_delayed_ref_common(fs_info, &ref->node, bytenr, num_bytes, | |
822 | ref_root, action, ref_type); | |
823 | ref->root = ref_root; | |
824 | ref->parent = parent; | |
825 | ref->objectid = owner; | |
826 | ref->offset = offset; | |
827 | ||
828 | ||
78a6184a | 829 | head_ref = kmem_cache_alloc(btrfs_delayed_ref_head_cachep, GFP_NOFS); |
56bec294 | 830 | if (!head_ref) { |
78a6184a | 831 | kmem_cache_free(btrfs_delayed_data_ref_cachep, ref); |
56bec294 CM |
832 | return -ENOMEM; |
833 | } | |
5d4f98a2 | 834 | |
afcdd129 JB |
835 | if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) && |
836 | is_fstree(ref_root)) { | |
3368d001 QW |
837 | record = kmalloc(sizeof(*record), GFP_NOFS); |
838 | if (!record) { | |
839 | kmem_cache_free(btrfs_delayed_data_ref_cachep, ref); | |
840 | kmem_cache_free(btrfs_delayed_ref_head_cachep, | |
841 | head_ref); | |
842 | return -ENOMEM; | |
843 | } | |
844 | } | |
845 | ||
2335efaf NB |
846 | init_delayed_ref_head(head_ref, record, bytenr, num_bytes, ref_root, |
847 | reserved, action, true, false); | |
fef394f7 | 848 | head_ref->extent_op = NULL; |
5d4f98a2 | 849 | |
56bec294 CM |
850 | delayed_refs = &trans->transaction->delayed_refs; |
851 | spin_lock(&delayed_refs->lock); | |
852 | ||
853 | /* | |
854 | * insert both the head node and the new ref without dropping | |
855 | * the spin lock | |
856 | */ | |
2335efaf NB |
857 | head_ref = add_delayed_ref_head(trans, head_ref, record, |
858 | action, &qrecord_inserted, | |
7be07912 | 859 | old_ref_mod, new_ref_mod); |
56bec294 | 860 | |
cd7f9699 | 861 | ret = insert_delayed_ref(trans, delayed_refs, head_ref, &ref->node); |
5d4f98a2 | 862 | spin_unlock(&delayed_refs->lock); |
95a06077 | 863 | |
cd7f9699 NB |
864 | trace_add_delayed_data_ref(trans->fs_info, &ref->node, ref, |
865 | action == BTRFS_ADD_DELAYED_EXTENT ? | |
866 | BTRFS_ADD_DELAYED_REF : action); | |
867 | if (ret > 0) | |
868 | kmem_cache_free(btrfs_delayed_data_ref_cachep, ref); | |
869 | ||
870 | ||
fb235dc0 QW |
871 | if (qrecord_inserted) |
872 | return btrfs_qgroup_trace_extent_post(fs_info, record); | |
5d4f98a2 YZ |
873 | return 0; |
874 | } | |
875 | ||
66d7e7f0 AJ |
876 | int btrfs_add_delayed_extent_op(struct btrfs_fs_info *fs_info, |
877 | struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
878 | u64 bytenr, u64 num_bytes, |
879 | struct btrfs_delayed_extent_op *extent_op) | |
880 | { | |
881 | struct btrfs_delayed_ref_head *head_ref; | |
882 | struct btrfs_delayed_ref_root *delayed_refs; | |
5d4f98a2 | 883 | |
78a6184a | 884 | head_ref = kmem_cache_alloc(btrfs_delayed_ref_head_cachep, GFP_NOFS); |
5d4f98a2 YZ |
885 | if (!head_ref) |
886 | return -ENOMEM; | |
887 | ||
2335efaf NB |
888 | init_delayed_ref_head(head_ref, NULL, bytenr, num_bytes, 0, 0, |
889 | BTRFS_UPDATE_DELAYED_HEAD, extent_op->is_data, | |
890 | false); | |
5d4f98a2 YZ |
891 | head_ref->extent_op = extent_op; |
892 | ||
893 | delayed_refs = &trans->transaction->delayed_refs; | |
894 | spin_lock(&delayed_refs->lock); | |
895 | ||
2335efaf NB |
896 | add_delayed_ref_head(trans, head_ref, NULL, BTRFS_UPDATE_DELAYED_HEAD, |
897 | NULL, NULL, NULL); | |
5d4f98a2 | 898 | |
56bec294 CM |
899 | spin_unlock(&delayed_refs->lock); |
900 | return 0; | |
901 | } | |
902 | ||
1887be66 CM |
903 | /* |
904 | * this does a simple search for the head node for a given extent. | |
905 | * It must be called with the delayed ref spinlock held, and it returns | |
906 | * the head node if any where found, or NULL if not. | |
907 | */ | |
908 | struct btrfs_delayed_ref_head * | |
f72ad18e | 909 | btrfs_find_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs, u64 bytenr) |
1887be66 | 910 | { |
5c9d028b | 911 | return find_ref_head(delayed_refs, bytenr, 0); |
1887be66 | 912 | } |
78a6184a | 913 | |
e67c718b | 914 | void __cold btrfs_delayed_ref_exit(void) |
78a6184a | 915 | { |
5598e900 KM |
916 | kmem_cache_destroy(btrfs_delayed_ref_head_cachep); |
917 | kmem_cache_destroy(btrfs_delayed_tree_ref_cachep); | |
918 | kmem_cache_destroy(btrfs_delayed_data_ref_cachep); | |
919 | kmem_cache_destroy(btrfs_delayed_extent_op_cachep); | |
78a6184a MX |
920 | } |
921 | ||
f5c29bd9 | 922 | int __init btrfs_delayed_ref_init(void) |
78a6184a MX |
923 | { |
924 | btrfs_delayed_ref_head_cachep = kmem_cache_create( | |
925 | "btrfs_delayed_ref_head", | |
926 | sizeof(struct btrfs_delayed_ref_head), 0, | |
fba4b697 | 927 | SLAB_MEM_SPREAD, NULL); |
78a6184a MX |
928 | if (!btrfs_delayed_ref_head_cachep) |
929 | goto fail; | |
930 | ||
931 | btrfs_delayed_tree_ref_cachep = kmem_cache_create( | |
932 | "btrfs_delayed_tree_ref", | |
933 | sizeof(struct btrfs_delayed_tree_ref), 0, | |
fba4b697 | 934 | SLAB_MEM_SPREAD, NULL); |
78a6184a MX |
935 | if (!btrfs_delayed_tree_ref_cachep) |
936 | goto fail; | |
937 | ||
938 | btrfs_delayed_data_ref_cachep = kmem_cache_create( | |
939 | "btrfs_delayed_data_ref", | |
940 | sizeof(struct btrfs_delayed_data_ref), 0, | |
fba4b697 | 941 | SLAB_MEM_SPREAD, NULL); |
78a6184a MX |
942 | if (!btrfs_delayed_data_ref_cachep) |
943 | goto fail; | |
944 | ||
945 | btrfs_delayed_extent_op_cachep = kmem_cache_create( | |
946 | "btrfs_delayed_extent_op", | |
947 | sizeof(struct btrfs_delayed_extent_op), 0, | |
fba4b697 | 948 | SLAB_MEM_SPREAD, NULL); |
78a6184a MX |
949 | if (!btrfs_delayed_extent_op_cachep) |
950 | goto fail; | |
951 | ||
952 | return 0; | |
953 | fail: | |
954 | btrfs_delayed_ref_exit(); | |
955 | return -ENOMEM; | |
956 | } |