delayed-ref.c 26 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919920921922923924925926927928929930931932933934935936937938939940941942943944945946
  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. #include <linux/sched.h>
  19. #include <linux/slab.h>
  20. #include <linux/sort.h>
  21. #include "ctree.h"
  22. #include "delayed-ref.h"
  23. #include "transaction.h"
  24. #include "qgroup.h"
  25. struct kmem_cache *btrfs_delayed_ref_head_cachep;
  26. struct kmem_cache *btrfs_delayed_tree_ref_cachep;
  27. struct kmem_cache *btrfs_delayed_data_ref_cachep;
  28. struct kmem_cache *btrfs_delayed_extent_op_cachep;
  29. /*
  30. * delayed back reference update tracking. For subvolume trees
  31. * we queue up extent allocations and backref maintenance for
  32. * delayed processing. This avoids deep call chains where we
  33. * add extents in the middle of btrfs_search_slot, and it allows
  34. * us to buffer up frequently modified backrefs in an rb tree instead
  35. * of hammering updates on the extent allocation tree.
  36. */
  37. /*
  38. * compare two delayed tree backrefs with same bytenr and type
  39. */
  40. static int comp_tree_refs(struct btrfs_delayed_tree_ref *ref2,
  41. struct btrfs_delayed_tree_ref *ref1, int type)
  42. {
  43. if (type == BTRFS_TREE_BLOCK_REF_KEY) {
  44. if (ref1->root < ref2->root)
  45. return -1;
  46. if (ref1->root > ref2->root)
  47. return 1;
  48. } else {
  49. if (ref1->parent < ref2->parent)
  50. return -1;
  51. if (ref1->parent > ref2->parent)
  52. return 1;
  53. }
  54. return 0;
  55. }
  56. /*
  57. * compare two delayed data backrefs with same bytenr and type
  58. */
  59. static int comp_data_refs(struct btrfs_delayed_data_ref *ref2,
  60. struct btrfs_delayed_data_ref *ref1)
  61. {
  62. if (ref1->node.type == BTRFS_EXTENT_DATA_REF_KEY) {
  63. if (ref1->root < ref2->root)
  64. return -1;
  65. if (ref1->root > ref2->root)
  66. return 1;
  67. if (ref1->objectid < ref2->objectid)
  68. return -1;
  69. if (ref1->objectid > ref2->objectid)
  70. return 1;
  71. if (ref1->offset < ref2->offset)
  72. return -1;
  73. if (ref1->offset > ref2->offset)
  74. return 1;
  75. } else {
  76. if (ref1->parent < ref2->parent)
  77. return -1;
  78. if (ref1->parent > ref2->parent)
  79. return 1;
  80. }
  81. return 0;
  82. }
  83. /* insert a new ref to head ref rbtree */
  84. static struct btrfs_delayed_ref_head *htree_insert(struct rb_root *root,
  85. struct rb_node *node)
  86. {
  87. struct rb_node **p = &root->rb_node;
  88. struct rb_node *parent_node = NULL;
  89. struct btrfs_delayed_ref_head *entry;
  90. struct btrfs_delayed_ref_head *ins;
  91. u64 bytenr;
  92. ins = rb_entry(node, struct btrfs_delayed_ref_head, href_node);
  93. bytenr = ins->node.bytenr;
  94. while (*p) {
  95. parent_node = *p;
  96. entry = rb_entry(parent_node, struct btrfs_delayed_ref_head,
  97. href_node);
  98. if (bytenr < entry->node.bytenr)
  99. p = &(*p)->rb_left;
  100. else if (bytenr > entry->node.bytenr)
  101. p = &(*p)->rb_right;
  102. else
  103. return entry;
  104. }
  105. rb_link_node(node, parent_node, p);
  106. rb_insert_color(node, root);
  107. return NULL;
  108. }
  109. /*
  110. * find an head entry based on bytenr. This returns the delayed ref
  111. * head if it was able to find one, or NULL if nothing was in that spot.
  112. * If return_bigger is given, the next bigger entry is returned if no exact
  113. * match is found.
  114. */
  115. static struct btrfs_delayed_ref_head *
  116. find_ref_head(struct rb_root *root, u64 bytenr,
  117. int return_bigger)
  118. {
  119. struct rb_node *n;
  120. struct btrfs_delayed_ref_head *entry;
  121. n = root->rb_node;
  122. entry = NULL;
  123. while (n) {
  124. entry = rb_entry(n, struct btrfs_delayed_ref_head, href_node);
  125. if (bytenr < entry->node.bytenr)
  126. n = n->rb_left;
  127. else if (bytenr > entry->node.bytenr)
  128. n = n->rb_right;
  129. else
  130. return entry;
  131. }
  132. if (entry && return_bigger) {
  133. if (bytenr > entry->node.bytenr) {
  134. n = rb_next(&entry->href_node);
  135. if (!n)
  136. n = rb_first(root);
  137. entry = rb_entry(n, struct btrfs_delayed_ref_head,
  138. href_node);
  139. return entry;
  140. }
  141. return entry;
  142. }
  143. return NULL;
  144. }
  145. int btrfs_delayed_ref_lock(struct btrfs_trans_handle *trans,
  146. struct btrfs_delayed_ref_head *head)
  147. {
  148. struct btrfs_delayed_ref_root *delayed_refs;
  149. delayed_refs = &trans->transaction->delayed_refs;
  150. assert_spin_locked(&delayed_refs->lock);
  151. if (mutex_trylock(&head->mutex))
  152. return 0;
  153. atomic_inc(&head->node.refs);
  154. spin_unlock(&delayed_refs->lock);
  155. mutex_lock(&head->mutex);
  156. spin_lock(&delayed_refs->lock);
  157. if (!head->node.in_tree) {
  158. mutex_unlock(&head->mutex);
  159. btrfs_put_delayed_ref(&head->node);
  160. return -EAGAIN;
  161. }
  162. btrfs_put_delayed_ref(&head->node);
  163. return 0;
  164. }
  165. static inline void drop_delayed_ref(struct btrfs_trans_handle *trans,
  166. struct btrfs_delayed_ref_root *delayed_refs,
  167. struct btrfs_delayed_ref_head *head,
  168. struct btrfs_delayed_ref_node *ref)
  169. {
  170. if (btrfs_delayed_ref_is_head(ref)) {
  171. head = btrfs_delayed_node_to_head(ref);
  172. rb_erase(&head->href_node, &delayed_refs->href_root);
  173. } else {
  174. assert_spin_locked(&head->lock);
  175. list_del(&ref->list);
  176. }
  177. ref->in_tree = 0;
  178. btrfs_put_delayed_ref(ref);
  179. atomic_dec(&delayed_refs->num_entries);
  180. if (trans->delayed_ref_updates)
  181. trans->delayed_ref_updates--;
  182. }
  183. static bool merge_ref(struct btrfs_trans_handle *trans,
  184. struct btrfs_delayed_ref_root *delayed_refs,
  185. struct btrfs_delayed_ref_head *head,
  186. struct btrfs_delayed_ref_node *ref,
  187. u64 seq)
  188. {
  189. struct btrfs_delayed_ref_node *next;
  190. bool done = false;
  191. next = list_first_entry(&head->ref_list, struct btrfs_delayed_ref_node,
  192. list);
  193. while (!done && &next->list != &head->ref_list) {
  194. int mod;
  195. struct btrfs_delayed_ref_node *next2;
  196. next2 = list_next_entry(next, list);
  197. if (next == ref)
  198. goto next;
  199. if (seq && next->seq >= seq)
  200. goto next;
  201. if (next->type != ref->type)
  202. goto next;
  203. if ((ref->type == BTRFS_TREE_BLOCK_REF_KEY ||
  204. ref->type == BTRFS_SHARED_BLOCK_REF_KEY) &&
  205. comp_tree_refs(btrfs_delayed_node_to_tree_ref(ref),
  206. btrfs_delayed_node_to_tree_ref(next),
  207. ref->type))
  208. goto next;
  209. if ((ref->type == BTRFS_EXTENT_DATA_REF_KEY ||
  210. ref->type == BTRFS_SHARED_DATA_REF_KEY) &&
  211. comp_data_refs(btrfs_delayed_node_to_data_ref(ref),
  212. btrfs_delayed_node_to_data_ref(next)))
  213. goto next;
  214. if (ref->action == next->action) {
  215. mod = next->ref_mod;
  216. } else {
  217. if (ref->ref_mod < next->ref_mod) {
  218. swap(ref, next);
  219. done = true;
  220. }
  221. mod = -next->ref_mod;
  222. }
  223. drop_delayed_ref(trans, delayed_refs, head, next);
  224. ref->ref_mod += mod;
  225. if (ref->ref_mod == 0) {
  226. drop_delayed_ref(trans, delayed_refs, head, ref);
  227. done = true;
  228. } else {
  229. /*
  230. * Can't have multiples of the same ref on a tree block.
  231. */
  232. WARN_ON(ref->type == BTRFS_TREE_BLOCK_REF_KEY ||
  233. ref->type == BTRFS_SHARED_BLOCK_REF_KEY);
  234. }
  235. next:
  236. next = next2;
  237. }
  238. return done;
  239. }
  240. void btrfs_merge_delayed_refs(struct btrfs_trans_handle *trans,
  241. struct btrfs_fs_info *fs_info,
  242. struct btrfs_delayed_ref_root *delayed_refs,
  243. struct btrfs_delayed_ref_head *head)
  244. {
  245. struct btrfs_delayed_ref_node *ref;
  246. u64 seq = 0;
  247. assert_spin_locked(&head->lock);
  248. if (list_empty(&head->ref_list))
  249. return;
  250. /* We don't have too many refs to merge for data. */
  251. if (head->is_data)
  252. return;
  253. spin_lock(&fs_info->tree_mod_seq_lock);
  254. if (!list_empty(&fs_info->tree_mod_seq_list)) {
  255. struct seq_list *elem;
  256. elem = list_first_entry(&fs_info->tree_mod_seq_list,
  257. struct seq_list, list);
  258. seq = elem->seq;
  259. }
  260. spin_unlock(&fs_info->tree_mod_seq_lock);
  261. ref = list_first_entry(&head->ref_list, struct btrfs_delayed_ref_node,
  262. list);
  263. while (&ref->list != &head->ref_list) {
  264. if (seq && ref->seq >= seq)
  265. goto next;
  266. if (merge_ref(trans, delayed_refs, head, ref, seq)) {
  267. if (list_empty(&head->ref_list))
  268. break;
  269. ref = list_first_entry(&head->ref_list,
  270. struct btrfs_delayed_ref_node,
  271. list);
  272. continue;
  273. }
  274. next:
  275. ref = list_next_entry(ref, list);
  276. }
  277. }
  278. int btrfs_check_delayed_seq(struct btrfs_fs_info *fs_info,
  279. struct btrfs_delayed_ref_root *delayed_refs,
  280. u64 seq)
  281. {
  282. struct seq_list *elem;
  283. int ret = 0;
  284. spin_lock(&fs_info->tree_mod_seq_lock);
  285. if (!list_empty(&fs_info->tree_mod_seq_list)) {
  286. elem = list_first_entry(&fs_info->tree_mod_seq_list,
  287. struct seq_list, list);
  288. if (seq >= elem->seq) {
  289. btrfs_debug(fs_info,
  290. "holding back delayed_ref %#x.%x, lowest is %#x.%x (%p)",
  291. (u32)(seq >> 32), (u32)seq,
  292. (u32)(elem->seq >> 32), (u32)elem->seq,
  293. delayed_refs);
  294. ret = 1;
  295. }
  296. }
  297. spin_unlock(&fs_info->tree_mod_seq_lock);
  298. return ret;
  299. }
  300. struct btrfs_delayed_ref_head *
  301. btrfs_select_ref_head(struct btrfs_trans_handle *trans)
  302. {
  303. struct btrfs_delayed_ref_root *delayed_refs;
  304. struct btrfs_delayed_ref_head *head;
  305. u64 start;
  306. bool loop = false;
  307. delayed_refs = &trans->transaction->delayed_refs;
  308. again:
  309. start = delayed_refs->run_delayed_start;
  310. head = find_ref_head(&delayed_refs->href_root, start, 1);
  311. if (!head && !loop) {
  312. delayed_refs->run_delayed_start = 0;
  313. start = 0;
  314. loop = true;
  315. head = find_ref_head(&delayed_refs->href_root, start, 1);
  316. if (!head)
  317. return NULL;
  318. } else if (!head && loop) {
  319. return NULL;
  320. }
  321. while (head->processing) {
  322. struct rb_node *node;
  323. node = rb_next(&head->href_node);
  324. if (!node) {
  325. if (loop)
  326. return NULL;
  327. delayed_refs->run_delayed_start = 0;
  328. start = 0;
  329. loop = true;
  330. goto again;
  331. }
  332. head = rb_entry(node, struct btrfs_delayed_ref_head,
  333. href_node);
  334. }
  335. head->processing = 1;
  336. WARN_ON(delayed_refs->num_heads_ready == 0);
  337. delayed_refs->num_heads_ready--;
  338. delayed_refs->run_delayed_start = head->node.bytenr +
  339. head->node.num_bytes;
  340. return head;
  341. }
  342. /*
  343. * Helper to insert the ref_node to the tail or merge with tail.
  344. *
  345. * Return 0 for insert.
  346. * Return >0 for merge.
  347. */
  348. static int
  349. add_delayed_ref_tail_merge(struct btrfs_trans_handle *trans,
  350. struct btrfs_delayed_ref_root *root,
  351. struct btrfs_delayed_ref_head *href,
  352. struct btrfs_delayed_ref_node *ref)
  353. {
  354. struct btrfs_delayed_ref_node *exist;
  355. int mod;
  356. int ret = 0;
  357. spin_lock(&href->lock);
  358. /* Check whether we can merge the tail node with ref */
  359. if (list_empty(&href->ref_list))
  360. goto add_tail;
  361. exist = list_entry(href->ref_list.prev, struct btrfs_delayed_ref_node,
  362. list);
  363. /* No need to compare bytenr nor is_head */
  364. if (exist->type != ref->type || exist->seq != ref->seq)
  365. goto add_tail;
  366. if ((exist->type == BTRFS_TREE_BLOCK_REF_KEY ||
  367. exist->type == BTRFS_SHARED_BLOCK_REF_KEY) &&
  368. comp_tree_refs(btrfs_delayed_node_to_tree_ref(exist),
  369. btrfs_delayed_node_to_tree_ref(ref),
  370. ref->type))
  371. goto add_tail;
  372. if ((exist->type == BTRFS_EXTENT_DATA_REF_KEY ||
  373. exist->type == BTRFS_SHARED_DATA_REF_KEY) &&
  374. comp_data_refs(btrfs_delayed_node_to_data_ref(exist),
  375. btrfs_delayed_node_to_data_ref(ref)))
  376. goto add_tail;
  377. /* Now we are sure we can merge */
  378. ret = 1;
  379. if (exist->action == ref->action) {
  380. mod = ref->ref_mod;
  381. } else {
  382. /* Need to change action */
  383. if (exist->ref_mod < ref->ref_mod) {
  384. exist->action = ref->action;
  385. mod = -exist->ref_mod;
  386. exist->ref_mod = ref->ref_mod;
  387. } else
  388. mod = -ref->ref_mod;
  389. }
  390. exist->ref_mod += mod;
  391. /* remove existing tail if its ref_mod is zero */
  392. if (exist->ref_mod == 0)
  393. drop_delayed_ref(trans, root, href, exist);
  394. spin_unlock(&href->lock);
  395. return ret;
  396. add_tail:
  397. list_add_tail(&ref->list, &href->ref_list);
  398. atomic_inc(&root->num_entries);
  399. trans->delayed_ref_updates++;
  400. spin_unlock(&href->lock);
  401. return ret;
  402. }
  403. /*
  404. * helper function to update the accounting in the head ref
  405. * existing and update must have the same bytenr
  406. */
  407. static noinline void
  408. update_existing_head_ref(struct btrfs_delayed_ref_root *delayed_refs,
  409. struct btrfs_delayed_ref_node *existing,
  410. struct btrfs_delayed_ref_node *update)
  411. {
  412. struct btrfs_delayed_ref_head *existing_ref;
  413. struct btrfs_delayed_ref_head *ref;
  414. int old_ref_mod;
  415. existing_ref = btrfs_delayed_node_to_head(existing);
  416. ref = btrfs_delayed_node_to_head(update);
  417. BUG_ON(existing_ref->is_data != ref->is_data);
  418. spin_lock(&existing_ref->lock);
  419. if (ref->must_insert_reserved) {
  420. /* if the extent was freed and then
  421. * reallocated before the delayed ref
  422. * entries were processed, we can end up
  423. * with an existing head ref without
  424. * the must_insert_reserved flag set.
  425. * Set it again here
  426. */
  427. existing_ref->must_insert_reserved = ref->must_insert_reserved;
  428. /*
  429. * update the num_bytes so we make sure the accounting
  430. * is done correctly
  431. */
  432. existing->num_bytes = update->num_bytes;
  433. }
  434. if (ref->extent_op) {
  435. if (!existing_ref->extent_op) {
  436. existing_ref->extent_op = ref->extent_op;
  437. } else {
  438. if (ref->extent_op->update_key) {
  439. memcpy(&existing_ref->extent_op->key,
  440. &ref->extent_op->key,
  441. sizeof(ref->extent_op->key));
  442. existing_ref->extent_op->update_key = true;
  443. }
  444. if (ref->extent_op->update_flags) {
  445. existing_ref->extent_op->flags_to_set |=
  446. ref->extent_op->flags_to_set;
  447. existing_ref->extent_op->update_flags = true;
  448. }
  449. btrfs_free_delayed_extent_op(ref->extent_op);
  450. }
  451. }
  452. /*
  453. * update the reference mod on the head to reflect this new operation,
  454. * only need the lock for this case cause we could be processing it
  455. * currently, for refs we just added we know we're a-ok.
  456. */
  457. old_ref_mod = existing_ref->total_ref_mod;
  458. existing->ref_mod += update->ref_mod;
  459. existing_ref->total_ref_mod += update->ref_mod;
  460. /*
  461. * If we are going to from a positive ref mod to a negative or vice
  462. * versa we need to make sure to adjust pending_csums accordingly.
  463. */
  464. if (existing_ref->is_data) {
  465. if (existing_ref->total_ref_mod >= 0 && old_ref_mod < 0)
  466. delayed_refs->pending_csums -= existing->num_bytes;
  467. if (existing_ref->total_ref_mod < 0 && old_ref_mod >= 0)
  468. delayed_refs->pending_csums += existing->num_bytes;
  469. }
  470. spin_unlock(&existing_ref->lock);
  471. }
  472. /*
  473. * helper function to actually insert a head node into the rbtree.
  474. * this does all the dirty work in terms of maintaining the correct
  475. * overall modification count.
  476. */
  477. static noinline struct btrfs_delayed_ref_head *
  478. add_delayed_ref_head(struct btrfs_fs_info *fs_info,
  479. struct btrfs_trans_handle *trans,
  480. struct btrfs_delayed_ref_node *ref,
  481. struct btrfs_qgroup_extent_record *qrecord,
  482. u64 bytenr, u64 num_bytes, u64 ref_root, u64 reserved,
  483. int action, int is_data)
  484. {
  485. struct btrfs_delayed_ref_head *existing;
  486. struct btrfs_delayed_ref_head *head_ref = NULL;
  487. struct btrfs_delayed_ref_root *delayed_refs;
  488. int count_mod = 1;
  489. int must_insert_reserved = 0;
  490. /* If reserved is provided, it must be a data extent. */
  491. BUG_ON(!is_data && reserved);
  492. /*
  493. * the head node stores the sum of all the mods, so dropping a ref
  494. * should drop the sum in the head node by one.
  495. */
  496. if (action == BTRFS_UPDATE_DELAYED_HEAD)
  497. count_mod = 0;
  498. else if (action == BTRFS_DROP_DELAYED_REF)
  499. count_mod = -1;
  500. /*
  501. * BTRFS_ADD_DELAYED_EXTENT means that we need to update
  502. * the reserved accounting when the extent is finally added, or
  503. * if a later modification deletes the delayed ref without ever
  504. * inserting the extent into the extent allocation tree.
  505. * ref->must_insert_reserved is the flag used to record
  506. * that accounting mods are required.
  507. *
  508. * Once we record must_insert_reserved, switch the action to
  509. * BTRFS_ADD_DELAYED_REF because other special casing is not required.
  510. */
  511. if (action == BTRFS_ADD_DELAYED_EXTENT)
  512. must_insert_reserved = 1;
  513. else
  514. must_insert_reserved = 0;
  515. delayed_refs = &trans->transaction->delayed_refs;
  516. /* first set the basic ref node struct up */
  517. atomic_set(&ref->refs, 1);
  518. ref->bytenr = bytenr;
  519. ref->num_bytes = num_bytes;
  520. ref->ref_mod = count_mod;
  521. ref->type = 0;
  522. ref->action = 0;
  523. ref->is_head = 1;
  524. ref->in_tree = 1;
  525. ref->seq = 0;
  526. head_ref = btrfs_delayed_node_to_head(ref);
  527. head_ref->must_insert_reserved = must_insert_reserved;
  528. head_ref->is_data = is_data;
  529. INIT_LIST_HEAD(&head_ref->ref_list);
  530. head_ref->processing = 0;
  531. head_ref->total_ref_mod = count_mod;
  532. head_ref->qgroup_reserved = 0;
  533. head_ref->qgroup_ref_root = 0;
  534. /* Record qgroup extent info if provided */
  535. if (qrecord) {
  536. if (ref_root && reserved) {
  537. head_ref->qgroup_ref_root = ref_root;
  538. head_ref->qgroup_reserved = reserved;
  539. }
  540. qrecord->bytenr = bytenr;
  541. qrecord->num_bytes = num_bytes;
  542. qrecord->old_roots = NULL;
  543. if(btrfs_qgroup_insert_dirty_extent_nolock(fs_info,
  544. delayed_refs, qrecord))
  545. kfree(qrecord);
  546. }
  547. spin_lock_init(&head_ref->lock);
  548. mutex_init(&head_ref->mutex);
  549. trace_add_delayed_ref_head(fs_info, ref, head_ref, action);
  550. existing = htree_insert(&delayed_refs->href_root,
  551. &head_ref->href_node);
  552. if (existing) {
  553. WARN_ON(ref_root && reserved && existing->qgroup_ref_root
  554. && existing->qgroup_reserved);
  555. update_existing_head_ref(delayed_refs, &existing->node, ref);
  556. /*
  557. * we've updated the existing ref, free the newly
  558. * allocated ref
  559. */
  560. kmem_cache_free(btrfs_delayed_ref_head_cachep, head_ref);
  561. head_ref = existing;
  562. } else {
  563. if (is_data && count_mod < 0)
  564. delayed_refs->pending_csums += num_bytes;
  565. delayed_refs->num_heads++;
  566. delayed_refs->num_heads_ready++;
  567. atomic_inc(&delayed_refs->num_entries);
  568. trans->delayed_ref_updates++;
  569. }
  570. return head_ref;
  571. }
  572. /*
  573. * helper to insert a delayed tree ref into the rbtree.
  574. */
  575. static noinline void
  576. add_delayed_tree_ref(struct btrfs_fs_info *fs_info,
  577. struct btrfs_trans_handle *trans,
  578. struct btrfs_delayed_ref_head *head_ref,
  579. struct btrfs_delayed_ref_node *ref, u64 bytenr,
  580. u64 num_bytes, u64 parent, u64 ref_root, int level,
  581. int action)
  582. {
  583. struct btrfs_delayed_tree_ref *full_ref;
  584. struct btrfs_delayed_ref_root *delayed_refs;
  585. u64 seq = 0;
  586. int ret;
  587. if (action == BTRFS_ADD_DELAYED_EXTENT)
  588. action = BTRFS_ADD_DELAYED_REF;
  589. if (is_fstree(ref_root))
  590. seq = atomic64_read(&fs_info->tree_mod_seq);
  591. delayed_refs = &trans->transaction->delayed_refs;
  592. /* first set the basic ref node struct up */
  593. atomic_set(&ref->refs, 1);
  594. ref->bytenr = bytenr;
  595. ref->num_bytes = num_bytes;
  596. ref->ref_mod = 1;
  597. ref->action = action;
  598. ref->is_head = 0;
  599. ref->in_tree = 1;
  600. ref->seq = seq;
  601. full_ref = btrfs_delayed_node_to_tree_ref(ref);
  602. full_ref->parent = parent;
  603. full_ref->root = ref_root;
  604. if (parent)
  605. ref->type = BTRFS_SHARED_BLOCK_REF_KEY;
  606. else
  607. ref->type = BTRFS_TREE_BLOCK_REF_KEY;
  608. full_ref->level = level;
  609. trace_add_delayed_tree_ref(fs_info, ref, full_ref, action);
  610. ret = add_delayed_ref_tail_merge(trans, delayed_refs, head_ref, ref);
  611. /*
  612. * XXX: memory should be freed at the same level allocated.
  613. * But bad practice is anywhere... Follow it now. Need cleanup.
  614. */
  615. if (ret > 0)
  616. kmem_cache_free(btrfs_delayed_tree_ref_cachep, full_ref);
  617. }
  618. /*
  619. * helper to insert a delayed data ref into the rbtree.
  620. */
  621. static noinline void
  622. add_delayed_data_ref(struct btrfs_fs_info *fs_info,
  623. struct btrfs_trans_handle *trans,
  624. struct btrfs_delayed_ref_head *head_ref,
  625. struct btrfs_delayed_ref_node *ref, u64 bytenr,
  626. u64 num_bytes, u64 parent, u64 ref_root, u64 owner,
  627. u64 offset, int action)
  628. {
  629. struct btrfs_delayed_data_ref *full_ref;
  630. struct btrfs_delayed_ref_root *delayed_refs;
  631. u64 seq = 0;
  632. int ret;
  633. if (action == BTRFS_ADD_DELAYED_EXTENT)
  634. action = BTRFS_ADD_DELAYED_REF;
  635. delayed_refs = &trans->transaction->delayed_refs;
  636. if (is_fstree(ref_root))
  637. seq = atomic64_read(&fs_info->tree_mod_seq);
  638. /* first set the basic ref node struct up */
  639. atomic_set(&ref->refs, 1);
  640. ref->bytenr = bytenr;
  641. ref->num_bytes = num_bytes;
  642. ref->ref_mod = 1;
  643. ref->action = action;
  644. ref->is_head = 0;
  645. ref->in_tree = 1;
  646. ref->seq = seq;
  647. full_ref = btrfs_delayed_node_to_data_ref(ref);
  648. full_ref->parent = parent;
  649. full_ref->root = ref_root;
  650. if (parent)
  651. ref->type = BTRFS_SHARED_DATA_REF_KEY;
  652. else
  653. ref->type = BTRFS_EXTENT_DATA_REF_KEY;
  654. full_ref->objectid = owner;
  655. full_ref->offset = offset;
  656. trace_add_delayed_data_ref(fs_info, ref, full_ref, action);
  657. ret = add_delayed_ref_tail_merge(trans, delayed_refs, head_ref, ref);
  658. if (ret > 0)
  659. kmem_cache_free(btrfs_delayed_data_ref_cachep, full_ref);
  660. }
  661. /*
  662. * add a delayed tree ref. This does all of the accounting required
  663. * to make sure the delayed ref is eventually processed before this
  664. * transaction commits.
  665. */
  666. int btrfs_add_delayed_tree_ref(struct btrfs_fs_info *fs_info,
  667. struct btrfs_trans_handle *trans,
  668. u64 bytenr, u64 num_bytes, u64 parent,
  669. u64 ref_root, int level, int action,
  670. struct btrfs_delayed_extent_op *extent_op)
  671. {
  672. struct btrfs_delayed_tree_ref *ref;
  673. struct btrfs_delayed_ref_head *head_ref;
  674. struct btrfs_delayed_ref_root *delayed_refs;
  675. struct btrfs_qgroup_extent_record *record = NULL;
  676. BUG_ON(extent_op && extent_op->is_data);
  677. ref = kmem_cache_alloc(btrfs_delayed_tree_ref_cachep, GFP_NOFS);
  678. if (!ref)
  679. return -ENOMEM;
  680. head_ref = kmem_cache_alloc(btrfs_delayed_ref_head_cachep, GFP_NOFS);
  681. if (!head_ref)
  682. goto free_ref;
  683. if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) &&
  684. is_fstree(ref_root)) {
  685. record = kmalloc(sizeof(*record), GFP_NOFS);
  686. if (!record)
  687. goto free_head_ref;
  688. }
  689. head_ref->extent_op = extent_op;
  690. delayed_refs = &trans->transaction->delayed_refs;
  691. spin_lock(&delayed_refs->lock);
  692. /*
  693. * insert both the head node and the new ref without dropping
  694. * the spin lock
  695. */
  696. head_ref = add_delayed_ref_head(fs_info, trans, &head_ref->node, record,
  697. bytenr, num_bytes, 0, 0, action, 0);
  698. add_delayed_tree_ref(fs_info, trans, head_ref, &ref->node, bytenr,
  699. num_bytes, parent, ref_root, level, action);
  700. spin_unlock(&delayed_refs->lock);
  701. return 0;
  702. free_head_ref:
  703. kmem_cache_free(btrfs_delayed_ref_head_cachep, head_ref);
  704. free_ref:
  705. kmem_cache_free(btrfs_delayed_tree_ref_cachep, ref);
  706. return -ENOMEM;
  707. }
  708. /*
  709. * add a delayed data ref. it's similar to btrfs_add_delayed_tree_ref.
  710. */
  711. int btrfs_add_delayed_data_ref(struct btrfs_fs_info *fs_info,
  712. struct btrfs_trans_handle *trans,
  713. u64 bytenr, u64 num_bytes,
  714. u64 parent, u64 ref_root,
  715. u64 owner, u64 offset, u64 reserved, int action,
  716. struct btrfs_delayed_extent_op *extent_op)
  717. {
  718. struct btrfs_delayed_data_ref *ref;
  719. struct btrfs_delayed_ref_head *head_ref;
  720. struct btrfs_delayed_ref_root *delayed_refs;
  721. struct btrfs_qgroup_extent_record *record = NULL;
  722. BUG_ON(extent_op && !extent_op->is_data);
  723. ref = kmem_cache_alloc(btrfs_delayed_data_ref_cachep, GFP_NOFS);
  724. if (!ref)
  725. return -ENOMEM;
  726. head_ref = kmem_cache_alloc(btrfs_delayed_ref_head_cachep, GFP_NOFS);
  727. if (!head_ref) {
  728. kmem_cache_free(btrfs_delayed_data_ref_cachep, ref);
  729. return -ENOMEM;
  730. }
  731. if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) &&
  732. is_fstree(ref_root)) {
  733. record = kmalloc(sizeof(*record), GFP_NOFS);
  734. if (!record) {
  735. kmem_cache_free(btrfs_delayed_data_ref_cachep, ref);
  736. kmem_cache_free(btrfs_delayed_ref_head_cachep,
  737. head_ref);
  738. return -ENOMEM;
  739. }
  740. }
  741. head_ref->extent_op = extent_op;
  742. delayed_refs = &trans->transaction->delayed_refs;
  743. spin_lock(&delayed_refs->lock);
  744. /*
  745. * insert both the head node and the new ref without dropping
  746. * the spin lock
  747. */
  748. head_ref = add_delayed_ref_head(fs_info, trans, &head_ref->node, record,
  749. bytenr, num_bytes, ref_root, reserved,
  750. action, 1);
  751. add_delayed_data_ref(fs_info, trans, head_ref, &ref->node, bytenr,
  752. num_bytes, parent, ref_root, owner, offset,
  753. action);
  754. spin_unlock(&delayed_refs->lock);
  755. return 0;
  756. }
  757. int btrfs_add_delayed_extent_op(struct btrfs_fs_info *fs_info,
  758. struct btrfs_trans_handle *trans,
  759. u64 bytenr, u64 num_bytes,
  760. struct btrfs_delayed_extent_op *extent_op)
  761. {
  762. struct btrfs_delayed_ref_head *head_ref;
  763. struct btrfs_delayed_ref_root *delayed_refs;
  764. head_ref = kmem_cache_alloc(btrfs_delayed_ref_head_cachep, GFP_NOFS);
  765. if (!head_ref)
  766. return -ENOMEM;
  767. head_ref->extent_op = extent_op;
  768. delayed_refs = &trans->transaction->delayed_refs;
  769. spin_lock(&delayed_refs->lock);
  770. add_delayed_ref_head(fs_info, trans, &head_ref->node, NULL, bytenr,
  771. num_bytes, 0, 0, BTRFS_UPDATE_DELAYED_HEAD,
  772. extent_op->is_data);
  773. spin_unlock(&delayed_refs->lock);
  774. return 0;
  775. }
  776. /*
  777. * this does a simple search for the head node for a given extent.
  778. * It must be called with the delayed ref spinlock held, and it returns
  779. * the head node if any where found, or NULL if not.
  780. */
  781. struct btrfs_delayed_ref_head *
  782. btrfs_find_delayed_ref_head(struct btrfs_trans_handle *trans, u64 bytenr)
  783. {
  784. struct btrfs_delayed_ref_root *delayed_refs;
  785. delayed_refs = &trans->transaction->delayed_refs;
  786. return find_ref_head(&delayed_refs->href_root, bytenr, 0);
  787. }
  788. void btrfs_delayed_ref_exit(void)
  789. {
  790. kmem_cache_destroy(btrfs_delayed_ref_head_cachep);
  791. kmem_cache_destroy(btrfs_delayed_tree_ref_cachep);
  792. kmem_cache_destroy(btrfs_delayed_data_ref_cachep);
  793. kmem_cache_destroy(btrfs_delayed_extent_op_cachep);
  794. }
  795. int btrfs_delayed_ref_init(void)
  796. {
  797. btrfs_delayed_ref_head_cachep = kmem_cache_create(
  798. "btrfs_delayed_ref_head",
  799. sizeof(struct btrfs_delayed_ref_head), 0,
  800. SLAB_MEM_SPREAD, NULL);
  801. if (!btrfs_delayed_ref_head_cachep)
  802. goto fail;
  803. btrfs_delayed_tree_ref_cachep = kmem_cache_create(
  804. "btrfs_delayed_tree_ref",
  805. sizeof(struct btrfs_delayed_tree_ref), 0,
  806. SLAB_MEM_SPREAD, NULL);
  807. if (!btrfs_delayed_tree_ref_cachep)
  808. goto fail;
  809. btrfs_delayed_data_ref_cachep = kmem_cache_create(
  810. "btrfs_delayed_data_ref",
  811. sizeof(struct btrfs_delayed_data_ref), 0,
  812. SLAB_MEM_SPREAD, NULL);
  813. if (!btrfs_delayed_data_ref_cachep)
  814. goto fail;
  815. btrfs_delayed_extent_op_cachep = kmem_cache_create(
  816. "btrfs_delayed_extent_op",
  817. sizeof(struct btrfs_delayed_extent_op), 0,
  818. SLAB_MEM_SPREAD, NULL);
  819. if (!btrfs_delayed_extent_op_cachep)
  820. goto fail;
  821. return 0;
  822. fail:
  823. btrfs_delayed_ref_exit();
  824. return -ENOMEM;
  825. }