btree.c 60 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919920921922923924925926927928929930931932933934935936937938939940941942943944945946947948949950951952953954955956957958959960961962963964965966967968969970971972973974975976977978979980981982983984985986987988989990991992993994995996997998999100010011002100310041005100610071008100910101011101210131014101510161017101810191020102110221023102410251026102710281029103010311032103310341035103610371038103910401041104210431044104510461047104810491050105110521053105410551056105710581059106010611062106310641065106610671068106910701071107210731074107510761077107810791080108110821083108410851086108710881089109010911092109310941095109610971098109911001101110211031104110511061107110811091110111111121113111411151116111711181119112011211122112311241125112611271128112911301131113211331134113511361137113811391140114111421143114411451146114711481149115011511152115311541155115611571158115911601161116211631164116511661167116811691170117111721173117411751176117711781179118011811182118311841185118611871188118911901191119211931194119511961197119811991200120112021203120412051206120712081209121012111212121312141215121612171218121912201221122212231224122512261227122812291230123112321233123412351236123712381239124012411242124312441245124612471248124912501251125212531254125512561257125812591260126112621263126412651266126712681269127012711272127312741275127612771278127912801281128212831284128512861287128812891290129112921293129412951296129712981299130013011302130313041305130613071308130913101311131213131314131513161317131813191320132113221323132413251326132713281329133013311332133313341335133613371338133913401341134213431344134513461347134813491350135113521353135413551356135713581359136013611362136313641365136613671368136913701371137213731374137513761377137813791380138113821383138413851386138713881389139013911392139313941395139613971398139914001401140214031404140514061407140814091410141114121413141414151416141714181419142014211422142314241425142614271428142914301431143214331434143514361437143814391440144114421443144414451446144714481449145014511452145314541455145614571458145914601461146214631464146514661467146814691470147114721473147414751476147714781479148014811482148314841485148614871488148914901491149214931494149514961497149814991500150115021503150415051506150715081509151015111512151315141515151615171518151915201521152215231524152515261527152815291530153115321533153415351536153715381539154015411542154315441545154615471548154915501551155215531554155515561557155815591560156115621563156415651566156715681569157015711572157315741575157615771578157915801581158215831584158515861587158815891590159115921593159415951596159715981599160016011602160316041605160616071608160916101611161216131614161516161617161816191620162116221623162416251626162716281629163016311632163316341635163616371638163916401641164216431644164516461647164816491650165116521653165416551656165716581659166016611662166316641665166616671668166916701671167216731674167516761677167816791680168116821683168416851686168716881689169016911692169316941695169616971698169917001701170217031704170517061707170817091710171117121713171417151716171717181719172017211722172317241725172617271728172917301731173217331734173517361737173817391740174117421743174417451746174717481749175017511752175317541755175617571758175917601761176217631764176517661767176817691770177117721773177417751776177717781779178017811782178317841785178617871788178917901791179217931794179517961797179817991800180118021803180418051806180718081809181018111812181318141815181618171818181918201821182218231824182518261827182818291830183118321833183418351836183718381839184018411842184318441845184618471848184918501851185218531854185518561857185818591860186118621863186418651866186718681869187018711872187318741875187618771878187918801881188218831884188518861887188818891890189118921893189418951896189718981899190019011902190319041905190619071908190919101911191219131914191519161917191819191920192119221923192419251926192719281929193019311932193319341935193619371938193919401941194219431944194519461947194819491950195119521953195419551956195719581959196019611962196319641965196619671968196919701971197219731974197519761977197819791980198119821983198419851986198719881989199019911992199319941995199619971998199920002001200220032004200520062007200820092010201120122013201420152016201720182019202020212022202320242025202620272028202920302031203220332034203520362037203820392040204120422043204420452046204720482049205020512052205320542055205620572058205920602061206220632064206520662067206820692070207120722073207420752076207720782079208020812082208320842085208620872088208920902091209220932094209520962097209820992100210121022103210421052106210721082109211021112112211321142115211621172118211921202121212221232124212521262127212821292130213121322133213421352136213721382139214021412142214321442145214621472148214921502151215221532154215521562157215821592160216121622163216421652166216721682169217021712172217321742175217621772178217921802181218221832184218521862187218821892190219121922193219421952196219721982199220022012202220322042205220622072208220922102211221222132214221522162217221822192220222122222223222422252226222722282229223022312232223322342235223622372238223922402241224222432244224522462247224822492250225122522253225422552256225722582259226022612262226322642265
  1. /*
  2. * btree.c - NILFS B-tree.
  3. *
  4. * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
  5. *
  6. * This program is free software; you can redistribute it and/or modify
  7. * it under the terms of the GNU General Public License as published by
  8. * the Free Software Foundation; either version 2 of the License, or
  9. * (at your option) any later version.
  10. *
  11. * This program is distributed in the hope that it will be useful,
  12. * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13. * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  14. * GNU General Public License for more details.
  15. *
  16. * You should have received a copy of the GNU General Public License
  17. * along with this program; if not, write to the Free Software
  18. * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
  19. *
  20. * Written by Koji Sato <koji@osrg.net>.
  21. */
  22. #include <linux/slab.h>
  23. #include <linux/string.h>
  24. #include <linux/errno.h>
  25. #include <linux/pagevec.h>
  26. #include "nilfs.h"
  27. #include "page.h"
  28. #include "btnode.h"
  29. #include "btree.h"
  30. #include "alloc.h"
  31. /**
  32. * struct nilfs_btree_path - A path on which B-tree operations are executed
  33. * @bp_bh: buffer head of node block
  34. * @bp_sib_bh: buffer head of sibling node block
  35. * @bp_index: index of child node
  36. * @bp_oldreq: ptr end request for old ptr
  37. * @bp_newreq: ptr alloc request for new ptr
  38. * @bp_op: rebalance operation
  39. */
  40. struct nilfs_btree_path {
  41. struct buffer_head *bp_bh;
  42. struct buffer_head *bp_sib_bh;
  43. int bp_index;
  44. union nilfs_bmap_ptr_req bp_oldreq;
  45. union nilfs_bmap_ptr_req bp_newreq;
  46. struct nilfs_btnode_chkey_ctxt bp_ctxt;
  47. void (*bp_op)(struct nilfs_btree *, struct nilfs_btree_path *,
  48. int, __u64 *, __u64 *);
  49. };
  50. /*
  51. * B-tree path operations
  52. */
  53. static struct kmem_cache *nilfs_btree_path_cache;
  54. int __init nilfs_btree_path_cache_init(void)
  55. {
  56. nilfs_btree_path_cache =
  57. kmem_cache_create("nilfs2_btree_path_cache",
  58. sizeof(struct nilfs_btree_path) *
  59. NILFS_BTREE_LEVEL_MAX, 0, 0, NULL);
  60. return (nilfs_btree_path_cache != NULL) ? 0 : -ENOMEM;
  61. }
  62. void nilfs_btree_path_cache_destroy(void)
  63. {
  64. kmem_cache_destroy(nilfs_btree_path_cache);
  65. }
  66. static inline struct nilfs_btree_path *
  67. nilfs_btree_alloc_path(const struct nilfs_btree *btree)
  68. {
  69. return (struct nilfs_btree_path *)
  70. kmem_cache_alloc(nilfs_btree_path_cache, GFP_NOFS);
  71. }
  72. static inline void nilfs_btree_free_path(const struct nilfs_btree *btree,
  73. struct nilfs_btree_path *path)
  74. {
  75. kmem_cache_free(nilfs_btree_path_cache, path);
  76. }
  77. static void nilfs_btree_init_path(const struct nilfs_btree *btree,
  78. struct nilfs_btree_path *path)
  79. {
  80. int level;
  81. for (level = NILFS_BTREE_LEVEL_DATA;
  82. level < NILFS_BTREE_LEVEL_MAX;
  83. level++) {
  84. path[level].bp_bh = NULL;
  85. path[level].bp_sib_bh = NULL;
  86. path[level].bp_index = 0;
  87. path[level].bp_oldreq.bpr_ptr = NILFS_BMAP_INVALID_PTR;
  88. path[level].bp_newreq.bpr_ptr = NILFS_BMAP_INVALID_PTR;
  89. path[level].bp_op = NULL;
  90. }
  91. }
  92. static void nilfs_btree_clear_path(const struct nilfs_btree *btree,
  93. struct nilfs_btree_path *path)
  94. {
  95. int level;
  96. for (level = NILFS_BTREE_LEVEL_DATA;
  97. level < NILFS_BTREE_LEVEL_MAX;
  98. level++) {
  99. if (path[level].bp_bh != NULL) {
  100. brelse(path[level].bp_bh);
  101. path[level].bp_bh = NULL;
  102. }
  103. /* sib_bh is released or deleted by prepare or commit
  104. * operations. */
  105. path[level].bp_sib_bh = NULL;
  106. path[level].bp_index = 0;
  107. path[level].bp_oldreq.bpr_ptr = NILFS_BMAP_INVALID_PTR;
  108. path[level].bp_newreq.bpr_ptr = NILFS_BMAP_INVALID_PTR;
  109. path[level].bp_op = NULL;
  110. }
  111. }
  112. /*
  113. * B-tree node operations
  114. */
  115. static int nilfs_btree_get_block(const struct nilfs_btree *btree, __u64 ptr,
  116. struct buffer_head **bhp)
  117. {
  118. struct address_space *btnc =
  119. &NILFS_BMAP_I((struct nilfs_bmap *)btree)->i_btnode_cache;
  120. return nilfs_btnode_get(btnc, ptr, 0, bhp, 0);
  121. }
  122. static int nilfs_btree_get_new_block(const struct nilfs_btree *btree,
  123. __u64 ptr, struct buffer_head **bhp)
  124. {
  125. struct address_space *btnc =
  126. &NILFS_BMAP_I((struct nilfs_bmap *)btree)->i_btnode_cache;
  127. int ret;
  128. ret = nilfs_btnode_get(btnc, ptr, 0, bhp, 1);
  129. if (!ret)
  130. set_buffer_nilfs_volatile(*bhp);
  131. return ret;
  132. }
  133. static inline int
  134. nilfs_btree_node_get_flags(const struct nilfs_btree *btree,
  135. const struct nilfs_btree_node *node)
  136. {
  137. return node->bn_flags;
  138. }
  139. static inline void
  140. nilfs_btree_node_set_flags(struct nilfs_btree *btree,
  141. struct nilfs_btree_node *node,
  142. int flags)
  143. {
  144. node->bn_flags = flags;
  145. }
  146. static inline int nilfs_btree_node_root(const struct nilfs_btree *btree,
  147. const struct nilfs_btree_node *node)
  148. {
  149. return nilfs_btree_node_get_flags(btree, node) & NILFS_BTREE_NODE_ROOT;
  150. }
  151. static inline int
  152. nilfs_btree_node_get_level(const struct nilfs_btree *btree,
  153. const struct nilfs_btree_node *node)
  154. {
  155. return node->bn_level;
  156. }
  157. static inline void
  158. nilfs_btree_node_set_level(struct nilfs_btree *btree,
  159. struct nilfs_btree_node *node,
  160. int level)
  161. {
  162. node->bn_level = level;
  163. }
  164. static inline int
  165. nilfs_btree_node_get_nchildren(const struct nilfs_btree *btree,
  166. const struct nilfs_btree_node *node)
  167. {
  168. return le16_to_cpu(node->bn_nchildren);
  169. }
  170. static inline void
  171. nilfs_btree_node_set_nchildren(struct nilfs_btree *btree,
  172. struct nilfs_btree_node *node,
  173. int nchildren)
  174. {
  175. node->bn_nchildren = cpu_to_le16(nchildren);
  176. }
  177. static inline int
  178. nilfs_btree_node_size(const struct nilfs_btree *btree)
  179. {
  180. return 1 << btree->bt_bmap.b_inode->i_blkbits;
  181. }
  182. static inline int
  183. nilfs_btree_node_nchildren_min(const struct nilfs_btree *btree,
  184. const struct nilfs_btree_node *node)
  185. {
  186. return nilfs_btree_node_root(btree, node) ?
  187. NILFS_BTREE_ROOT_NCHILDREN_MIN :
  188. NILFS_BTREE_NODE_NCHILDREN_MIN(nilfs_btree_node_size(btree));
  189. }
  190. static inline int
  191. nilfs_btree_node_nchildren_max(const struct nilfs_btree *btree,
  192. const struct nilfs_btree_node *node)
  193. {
  194. return nilfs_btree_node_root(btree, node) ?
  195. NILFS_BTREE_ROOT_NCHILDREN_MAX :
  196. NILFS_BTREE_NODE_NCHILDREN_MAX(nilfs_btree_node_size(btree));
  197. }
  198. static inline __le64 *
  199. nilfs_btree_node_dkeys(const struct nilfs_btree *btree,
  200. const struct nilfs_btree_node *node)
  201. {
  202. return (__le64 *)((char *)(node + 1) +
  203. (nilfs_btree_node_root(btree, node) ?
  204. 0 : NILFS_BTREE_NODE_EXTRA_PAD_SIZE));
  205. }
  206. static inline __le64 *
  207. nilfs_btree_node_dptrs(const struct nilfs_btree *btree,
  208. const struct nilfs_btree_node *node)
  209. {
  210. return (__le64 *)(nilfs_btree_node_dkeys(btree, node) +
  211. nilfs_btree_node_nchildren_max(btree, node));
  212. }
  213. static inline __u64
  214. nilfs_btree_node_get_key(const struct nilfs_btree *btree,
  215. const struct nilfs_btree_node *node, int index)
  216. {
  217. return nilfs_bmap_dkey_to_key(*(nilfs_btree_node_dkeys(btree, node) +
  218. index));
  219. }
  220. static inline void
  221. nilfs_btree_node_set_key(struct nilfs_btree *btree,
  222. struct nilfs_btree_node *node, int index, __u64 key)
  223. {
  224. *(nilfs_btree_node_dkeys(btree, node) + index) =
  225. nilfs_bmap_key_to_dkey(key);
  226. }
  227. static inline __u64
  228. nilfs_btree_node_get_ptr(const struct nilfs_btree *btree,
  229. const struct nilfs_btree_node *node,
  230. int index)
  231. {
  232. return nilfs_bmap_dptr_to_ptr(*(nilfs_btree_node_dptrs(btree, node) +
  233. index));
  234. }
  235. static inline void
  236. nilfs_btree_node_set_ptr(struct nilfs_btree *btree,
  237. struct nilfs_btree_node *node,
  238. int index,
  239. __u64 ptr)
  240. {
  241. *(nilfs_btree_node_dptrs(btree, node) + index) =
  242. nilfs_bmap_ptr_to_dptr(ptr);
  243. }
  244. static void nilfs_btree_node_init(struct nilfs_btree *btree,
  245. struct nilfs_btree_node *node,
  246. int flags, int level, int nchildren,
  247. const __u64 *keys, const __u64 *ptrs)
  248. {
  249. __le64 *dkeys;
  250. __le64 *dptrs;
  251. int i;
  252. nilfs_btree_node_set_flags(btree, node, flags);
  253. nilfs_btree_node_set_level(btree, node, level);
  254. nilfs_btree_node_set_nchildren(btree, node, nchildren);
  255. dkeys = nilfs_btree_node_dkeys(btree, node);
  256. dptrs = nilfs_btree_node_dptrs(btree, node);
  257. for (i = 0; i < nchildren; i++) {
  258. dkeys[i] = nilfs_bmap_key_to_dkey(keys[i]);
  259. dptrs[i] = nilfs_bmap_ptr_to_dptr(ptrs[i]);
  260. }
  261. }
  262. /* Assume the buffer heads corresponding to left and right are locked. */
  263. static void nilfs_btree_node_move_left(struct nilfs_btree *btree,
  264. struct nilfs_btree_node *left,
  265. struct nilfs_btree_node *right,
  266. int n)
  267. {
  268. __le64 *ldkeys, *rdkeys;
  269. __le64 *ldptrs, *rdptrs;
  270. int lnchildren, rnchildren;
  271. ldkeys = nilfs_btree_node_dkeys(btree, left);
  272. ldptrs = nilfs_btree_node_dptrs(btree, left);
  273. lnchildren = nilfs_btree_node_get_nchildren(btree, left);
  274. rdkeys = nilfs_btree_node_dkeys(btree, right);
  275. rdptrs = nilfs_btree_node_dptrs(btree, right);
  276. rnchildren = nilfs_btree_node_get_nchildren(btree, right);
  277. memcpy(ldkeys + lnchildren, rdkeys, n * sizeof(*rdkeys));
  278. memcpy(ldptrs + lnchildren, rdptrs, n * sizeof(*rdptrs));
  279. memmove(rdkeys, rdkeys + n, (rnchildren - n) * sizeof(*rdkeys));
  280. memmove(rdptrs, rdptrs + n, (rnchildren - n) * sizeof(*rdptrs));
  281. lnchildren += n;
  282. rnchildren -= n;
  283. nilfs_btree_node_set_nchildren(btree, left, lnchildren);
  284. nilfs_btree_node_set_nchildren(btree, right, rnchildren);
  285. }
  286. /* Assume that the buffer heads corresponding to left and right are locked. */
  287. static void nilfs_btree_node_move_right(struct nilfs_btree *btree,
  288. struct nilfs_btree_node *left,
  289. struct nilfs_btree_node *right,
  290. int n)
  291. {
  292. __le64 *ldkeys, *rdkeys;
  293. __le64 *ldptrs, *rdptrs;
  294. int lnchildren, rnchildren;
  295. ldkeys = nilfs_btree_node_dkeys(btree, left);
  296. ldptrs = nilfs_btree_node_dptrs(btree, left);
  297. lnchildren = nilfs_btree_node_get_nchildren(btree, left);
  298. rdkeys = nilfs_btree_node_dkeys(btree, right);
  299. rdptrs = nilfs_btree_node_dptrs(btree, right);
  300. rnchildren = nilfs_btree_node_get_nchildren(btree, right);
  301. memmove(rdkeys + n, rdkeys, rnchildren * sizeof(*rdkeys));
  302. memmove(rdptrs + n, rdptrs, rnchildren * sizeof(*rdptrs));
  303. memcpy(rdkeys, ldkeys + lnchildren - n, n * sizeof(*rdkeys));
  304. memcpy(rdptrs, ldptrs + lnchildren - n, n * sizeof(*rdptrs));
  305. lnchildren -= n;
  306. rnchildren += n;
  307. nilfs_btree_node_set_nchildren(btree, left, lnchildren);
  308. nilfs_btree_node_set_nchildren(btree, right, rnchildren);
  309. }
  310. /* Assume that the buffer head corresponding to node is locked. */
  311. static void nilfs_btree_node_insert(struct nilfs_btree *btree,
  312. struct nilfs_btree_node *node,
  313. __u64 key, __u64 ptr, int index)
  314. {
  315. __le64 *dkeys;
  316. __le64 *dptrs;
  317. int nchildren;
  318. dkeys = nilfs_btree_node_dkeys(btree, node);
  319. dptrs = nilfs_btree_node_dptrs(btree, node);
  320. nchildren = nilfs_btree_node_get_nchildren(btree, node);
  321. if (index < nchildren) {
  322. memmove(dkeys + index + 1, dkeys + index,
  323. (nchildren - index) * sizeof(*dkeys));
  324. memmove(dptrs + index + 1, dptrs + index,
  325. (nchildren - index) * sizeof(*dptrs));
  326. }
  327. dkeys[index] = nilfs_bmap_key_to_dkey(key);
  328. dptrs[index] = nilfs_bmap_ptr_to_dptr(ptr);
  329. nchildren++;
  330. nilfs_btree_node_set_nchildren(btree, node, nchildren);
  331. }
  332. /* Assume that the buffer head corresponding to node is locked. */
  333. static void nilfs_btree_node_delete(struct nilfs_btree *btree,
  334. struct nilfs_btree_node *node,
  335. __u64 *keyp, __u64 *ptrp, int index)
  336. {
  337. __u64 key;
  338. __u64 ptr;
  339. __le64 *dkeys;
  340. __le64 *dptrs;
  341. int nchildren;
  342. dkeys = nilfs_btree_node_dkeys(btree, node);
  343. dptrs = nilfs_btree_node_dptrs(btree, node);
  344. key = nilfs_bmap_dkey_to_key(dkeys[index]);
  345. ptr = nilfs_bmap_dptr_to_ptr(dptrs[index]);
  346. nchildren = nilfs_btree_node_get_nchildren(btree, node);
  347. if (keyp != NULL)
  348. *keyp = key;
  349. if (ptrp != NULL)
  350. *ptrp = ptr;
  351. if (index < nchildren - 1) {
  352. memmove(dkeys + index, dkeys + index + 1,
  353. (nchildren - index - 1) * sizeof(*dkeys));
  354. memmove(dptrs + index, dptrs + index + 1,
  355. (nchildren - index - 1) * sizeof(*dptrs));
  356. }
  357. nchildren--;
  358. nilfs_btree_node_set_nchildren(btree, node, nchildren);
  359. }
  360. static int nilfs_btree_node_lookup(const struct nilfs_btree *btree,
  361. const struct nilfs_btree_node *node,
  362. __u64 key, int *indexp)
  363. {
  364. __u64 nkey;
  365. int index, low, high, s;
  366. /* binary search */
  367. low = 0;
  368. high = nilfs_btree_node_get_nchildren(btree, node) - 1;
  369. index = 0;
  370. s = 0;
  371. while (low <= high) {
  372. index = (low + high) / 2;
  373. nkey = nilfs_btree_node_get_key(btree, node, index);
  374. if (nkey == key) {
  375. s = 0;
  376. goto out;
  377. } else if (nkey < key) {
  378. low = index + 1;
  379. s = -1;
  380. } else {
  381. high = index - 1;
  382. s = 1;
  383. }
  384. }
  385. /* adjust index */
  386. if (nilfs_btree_node_get_level(btree, node) >
  387. NILFS_BTREE_LEVEL_NODE_MIN) {
  388. if ((s > 0) && (index > 0))
  389. index--;
  390. } else if (s < 0)
  391. index++;
  392. out:
  393. *indexp = index;
  394. return s == 0;
  395. }
  396. static inline struct nilfs_btree_node *
  397. nilfs_btree_get_root(const struct nilfs_btree *btree)
  398. {
  399. return (struct nilfs_btree_node *)btree->bt_bmap.b_u.u_data;
  400. }
  401. static inline struct nilfs_btree_node *
  402. nilfs_btree_get_nonroot_node(const struct nilfs_btree *btree,
  403. const struct nilfs_btree_path *path,
  404. int level)
  405. {
  406. return (struct nilfs_btree_node *)path[level].bp_bh->b_data;
  407. }
  408. static inline struct nilfs_btree_node *
  409. nilfs_btree_get_sib_node(const struct nilfs_btree *btree,
  410. const struct nilfs_btree_path *path,
  411. int level)
  412. {
  413. return (struct nilfs_btree_node *)path[level].bp_sib_bh->b_data;
  414. }
  415. static inline int nilfs_btree_height(const struct nilfs_btree *btree)
  416. {
  417. return nilfs_btree_node_get_level(btree, nilfs_btree_get_root(btree))
  418. + 1;
  419. }
  420. static inline struct nilfs_btree_node *
  421. nilfs_btree_get_node(const struct nilfs_btree *btree,
  422. const struct nilfs_btree_path *path,
  423. int level)
  424. {
  425. return (level == nilfs_btree_height(btree) - 1) ?
  426. nilfs_btree_get_root(btree) :
  427. nilfs_btree_get_nonroot_node(btree, path, level);
  428. }
  429. static int nilfs_btree_do_lookup(const struct nilfs_btree *btree,
  430. struct nilfs_btree_path *path,
  431. __u64 key, __u64 *ptrp, int minlevel)
  432. {
  433. struct nilfs_btree_node *node;
  434. __u64 ptr;
  435. int level, index, found, ret;
  436. node = nilfs_btree_get_root(btree);
  437. level = nilfs_btree_node_get_level(btree, node);
  438. if ((level < minlevel) ||
  439. (nilfs_btree_node_get_nchildren(btree, node) <= 0))
  440. return -ENOENT;
  441. found = nilfs_btree_node_lookup(btree, node, key, &index);
  442. ptr = nilfs_btree_node_get_ptr(btree, node, index);
  443. path[level].bp_bh = NULL;
  444. path[level].bp_index = index;
  445. for (level--; level >= minlevel; level--) {
  446. ret = nilfs_btree_get_block(btree, ptr, &path[level].bp_bh);
  447. if (ret < 0)
  448. return ret;
  449. node = nilfs_btree_get_nonroot_node(btree, path, level);
  450. BUG_ON(level != nilfs_btree_node_get_level(btree, node));
  451. if (!found)
  452. found = nilfs_btree_node_lookup(btree, node, key,
  453. &index);
  454. else
  455. index = 0;
  456. if (index < nilfs_btree_node_nchildren_max(btree, node))
  457. ptr = nilfs_btree_node_get_ptr(btree, node, index);
  458. else {
  459. WARN_ON(found || level != NILFS_BTREE_LEVEL_NODE_MIN);
  460. /* insert */
  461. ptr = NILFS_BMAP_INVALID_PTR;
  462. }
  463. path[level].bp_index = index;
  464. }
  465. if (!found)
  466. return -ENOENT;
  467. if (ptrp != NULL)
  468. *ptrp = ptr;
  469. return 0;
  470. }
  471. static int nilfs_btree_do_lookup_last(const struct nilfs_btree *btree,
  472. struct nilfs_btree_path *path,
  473. __u64 *keyp, __u64 *ptrp)
  474. {
  475. struct nilfs_btree_node *node;
  476. __u64 ptr;
  477. int index, level, ret;
  478. node = nilfs_btree_get_root(btree);
  479. index = nilfs_btree_node_get_nchildren(btree, node) - 1;
  480. if (index < 0)
  481. return -ENOENT;
  482. level = nilfs_btree_node_get_level(btree, node);
  483. ptr = nilfs_btree_node_get_ptr(btree, node, index);
  484. path[level].bp_bh = NULL;
  485. path[level].bp_index = index;
  486. for (level--; level > 0; level--) {
  487. ret = nilfs_btree_get_block(btree, ptr, &path[level].bp_bh);
  488. if (ret < 0)
  489. return ret;
  490. node = nilfs_btree_get_nonroot_node(btree, path, level);
  491. BUG_ON(level != nilfs_btree_node_get_level(btree, node));
  492. index = nilfs_btree_node_get_nchildren(btree, node) - 1;
  493. ptr = nilfs_btree_node_get_ptr(btree, node, index);
  494. path[level].bp_index = index;
  495. }
  496. if (keyp != NULL)
  497. *keyp = nilfs_btree_node_get_key(btree, node, index);
  498. if (ptrp != NULL)
  499. *ptrp = ptr;
  500. return 0;
  501. }
  502. static int nilfs_btree_lookup(const struct nilfs_bmap *bmap,
  503. __u64 key, int level, __u64 *ptrp)
  504. {
  505. struct nilfs_btree *btree;
  506. struct nilfs_btree_path *path;
  507. __u64 ptr;
  508. int ret;
  509. btree = (struct nilfs_btree *)bmap;
  510. path = nilfs_btree_alloc_path(btree);
  511. if (path == NULL)
  512. return -ENOMEM;
  513. nilfs_btree_init_path(btree, path);
  514. ret = nilfs_btree_do_lookup(btree, path, key, &ptr, level);
  515. if (ptrp != NULL)
  516. *ptrp = ptr;
  517. nilfs_btree_clear_path(btree, path);
  518. nilfs_btree_free_path(btree, path);
  519. return ret;
  520. }
  521. static void nilfs_btree_promote_key(struct nilfs_btree *btree,
  522. struct nilfs_btree_path *path,
  523. int level, __u64 key)
  524. {
  525. if (level < nilfs_btree_height(btree) - 1) {
  526. do {
  527. lock_buffer(path[level].bp_bh);
  528. nilfs_btree_node_set_key(
  529. btree,
  530. nilfs_btree_get_nonroot_node(
  531. btree, path, level),
  532. path[level].bp_index, key);
  533. if (!buffer_dirty(path[level].bp_bh))
  534. nilfs_btnode_mark_dirty(path[level].bp_bh);
  535. unlock_buffer(path[level].bp_bh);
  536. } while ((path[level].bp_index == 0) &&
  537. (++level < nilfs_btree_height(btree) - 1));
  538. }
  539. /* root */
  540. if (level == nilfs_btree_height(btree) - 1) {
  541. nilfs_btree_node_set_key(btree,
  542. nilfs_btree_get_root(btree),
  543. path[level].bp_index, key);
  544. }
  545. }
  546. static void nilfs_btree_do_insert(struct nilfs_btree *btree,
  547. struct nilfs_btree_path *path,
  548. int level, __u64 *keyp, __u64 *ptrp)
  549. {
  550. struct nilfs_btree_node *node;
  551. if (level < nilfs_btree_height(btree) - 1) {
  552. lock_buffer(path[level].bp_bh);
  553. node = nilfs_btree_get_nonroot_node(btree, path, level);
  554. nilfs_btree_node_insert(btree, node, *keyp, *ptrp,
  555. path[level].bp_index);
  556. if (!buffer_dirty(path[level].bp_bh))
  557. nilfs_btnode_mark_dirty(path[level].bp_bh);
  558. unlock_buffer(path[level].bp_bh);
  559. if (path[level].bp_index == 0)
  560. nilfs_btree_promote_key(btree, path, level + 1,
  561. nilfs_btree_node_get_key(
  562. btree, node, 0));
  563. } else {
  564. node = nilfs_btree_get_root(btree);
  565. nilfs_btree_node_insert(btree, node, *keyp, *ptrp,
  566. path[level].bp_index);
  567. }
  568. }
  569. static void nilfs_btree_carry_left(struct nilfs_btree *btree,
  570. struct nilfs_btree_path *path,
  571. int level, __u64 *keyp, __u64 *ptrp)
  572. {
  573. struct nilfs_btree_node *node, *left;
  574. int nchildren, lnchildren, n, move;
  575. lock_buffer(path[level].bp_bh);
  576. lock_buffer(path[level].bp_sib_bh);
  577. node = nilfs_btree_get_nonroot_node(btree, path, level);
  578. left = nilfs_btree_get_sib_node(btree, path, level);
  579. nchildren = nilfs_btree_node_get_nchildren(btree, node);
  580. lnchildren = nilfs_btree_node_get_nchildren(btree, left);
  581. move = 0;
  582. n = (nchildren + lnchildren + 1) / 2 - lnchildren;
  583. if (n > path[level].bp_index) {
  584. /* move insert point */
  585. n--;
  586. move = 1;
  587. }
  588. nilfs_btree_node_move_left(btree, left, node, n);
  589. if (!buffer_dirty(path[level].bp_bh))
  590. nilfs_btnode_mark_dirty(path[level].bp_bh);
  591. if (!buffer_dirty(path[level].bp_sib_bh))
  592. nilfs_btnode_mark_dirty(path[level].bp_sib_bh);
  593. unlock_buffer(path[level].bp_bh);
  594. unlock_buffer(path[level].bp_sib_bh);
  595. nilfs_btree_promote_key(btree, path, level + 1,
  596. nilfs_btree_node_get_key(btree, node, 0));
  597. if (move) {
  598. brelse(path[level].bp_bh);
  599. path[level].bp_bh = path[level].bp_sib_bh;
  600. path[level].bp_sib_bh = NULL;
  601. path[level].bp_index += lnchildren;
  602. path[level + 1].bp_index--;
  603. } else {
  604. brelse(path[level].bp_sib_bh);
  605. path[level].bp_sib_bh = NULL;
  606. path[level].bp_index -= n;
  607. }
  608. nilfs_btree_do_insert(btree, path, level, keyp, ptrp);
  609. }
  610. static void nilfs_btree_carry_right(struct nilfs_btree *btree,
  611. struct nilfs_btree_path *path,
  612. int level, __u64 *keyp, __u64 *ptrp)
  613. {
  614. struct nilfs_btree_node *node, *right;
  615. int nchildren, rnchildren, n, move;
  616. lock_buffer(path[level].bp_bh);
  617. lock_buffer(path[level].bp_sib_bh);
  618. node = nilfs_btree_get_nonroot_node(btree, path, level);
  619. right = nilfs_btree_get_sib_node(btree, path, level);
  620. nchildren = nilfs_btree_node_get_nchildren(btree, node);
  621. rnchildren = nilfs_btree_node_get_nchildren(btree, right);
  622. move = 0;
  623. n = (nchildren + rnchildren + 1) / 2 - rnchildren;
  624. if (n > nchildren - path[level].bp_index) {
  625. /* move insert point */
  626. n--;
  627. move = 1;
  628. }
  629. nilfs_btree_node_move_right(btree, node, right, n);
  630. if (!buffer_dirty(path[level].bp_bh))
  631. nilfs_btnode_mark_dirty(path[level].bp_bh);
  632. if (!buffer_dirty(path[level].bp_sib_bh))
  633. nilfs_btnode_mark_dirty(path[level].bp_sib_bh);
  634. unlock_buffer(path[level].bp_bh);
  635. unlock_buffer(path[level].bp_sib_bh);
  636. path[level + 1].bp_index++;
  637. nilfs_btree_promote_key(btree, path, level + 1,
  638. nilfs_btree_node_get_key(btree, right, 0));
  639. path[level + 1].bp_index--;
  640. if (move) {
  641. brelse(path[level].bp_bh);
  642. path[level].bp_bh = path[level].bp_sib_bh;
  643. path[level].bp_sib_bh = NULL;
  644. path[level].bp_index -=
  645. nilfs_btree_node_get_nchildren(btree, node);
  646. path[level + 1].bp_index++;
  647. } else {
  648. brelse(path[level].bp_sib_bh);
  649. path[level].bp_sib_bh = NULL;
  650. }
  651. nilfs_btree_do_insert(btree, path, level, keyp, ptrp);
  652. }
  653. static void nilfs_btree_split(struct nilfs_btree *btree,
  654. struct nilfs_btree_path *path,
  655. int level, __u64 *keyp, __u64 *ptrp)
  656. {
  657. struct nilfs_btree_node *node, *right;
  658. __u64 newkey;
  659. __u64 newptr;
  660. int nchildren, n, move;
  661. lock_buffer(path[level].bp_bh);
  662. lock_buffer(path[level].bp_sib_bh);
  663. node = nilfs_btree_get_nonroot_node(btree, path, level);
  664. right = nilfs_btree_get_sib_node(btree, path, level);
  665. nchildren = nilfs_btree_node_get_nchildren(btree, node);
  666. move = 0;
  667. n = (nchildren + 1) / 2;
  668. if (n > nchildren - path[level].bp_index) {
  669. n--;
  670. move = 1;
  671. }
  672. nilfs_btree_node_move_right(btree, node, right, n);
  673. if (!buffer_dirty(path[level].bp_bh))
  674. nilfs_btnode_mark_dirty(path[level].bp_bh);
  675. if (!buffer_dirty(path[level].bp_sib_bh))
  676. nilfs_btnode_mark_dirty(path[level].bp_sib_bh);
  677. unlock_buffer(path[level].bp_bh);
  678. unlock_buffer(path[level].bp_sib_bh);
  679. newkey = nilfs_btree_node_get_key(btree, right, 0);
  680. newptr = path[level].bp_newreq.bpr_ptr;
  681. if (move) {
  682. path[level].bp_index -=
  683. nilfs_btree_node_get_nchildren(btree, node);
  684. nilfs_btree_node_insert(btree, right, *keyp, *ptrp,
  685. path[level].bp_index);
  686. *keyp = nilfs_btree_node_get_key(btree, right, 0);
  687. *ptrp = path[level].bp_newreq.bpr_ptr;
  688. brelse(path[level].bp_bh);
  689. path[level].bp_bh = path[level].bp_sib_bh;
  690. path[level].bp_sib_bh = NULL;
  691. } else {
  692. nilfs_btree_do_insert(btree, path, level, keyp, ptrp);
  693. *keyp = nilfs_btree_node_get_key(btree, right, 0);
  694. *ptrp = path[level].bp_newreq.bpr_ptr;
  695. brelse(path[level].bp_sib_bh);
  696. path[level].bp_sib_bh = NULL;
  697. }
  698. path[level + 1].bp_index++;
  699. }
  700. static void nilfs_btree_grow(struct nilfs_btree *btree,
  701. struct nilfs_btree_path *path,
  702. int level, __u64 *keyp, __u64 *ptrp)
  703. {
  704. struct nilfs_btree_node *root, *child;
  705. int n;
  706. lock_buffer(path[level].bp_sib_bh);
  707. root = nilfs_btree_get_root(btree);
  708. child = nilfs_btree_get_sib_node(btree, path, level);
  709. n = nilfs_btree_node_get_nchildren(btree, root);
  710. nilfs_btree_node_move_right(btree, root, child, n);
  711. nilfs_btree_node_set_level(btree, root, level + 1);
  712. if (!buffer_dirty(path[level].bp_sib_bh))
  713. nilfs_btnode_mark_dirty(path[level].bp_sib_bh);
  714. unlock_buffer(path[level].bp_sib_bh);
  715. path[level].bp_bh = path[level].bp_sib_bh;
  716. path[level].bp_sib_bh = NULL;
  717. nilfs_btree_do_insert(btree, path, level, keyp, ptrp);
  718. *keyp = nilfs_btree_node_get_key(btree, child, 0);
  719. *ptrp = path[level].bp_newreq.bpr_ptr;
  720. }
  721. static __u64 nilfs_btree_find_near(const struct nilfs_btree *btree,
  722. const struct nilfs_btree_path *path)
  723. {
  724. struct nilfs_btree_node *node;
  725. int level;
  726. if (path == NULL)
  727. return NILFS_BMAP_INVALID_PTR;
  728. /* left sibling */
  729. level = NILFS_BTREE_LEVEL_NODE_MIN;
  730. if (path[level].bp_index > 0) {
  731. node = nilfs_btree_get_node(btree, path, level);
  732. return nilfs_btree_node_get_ptr(btree, node,
  733. path[level].bp_index - 1);
  734. }
  735. /* parent */
  736. level = NILFS_BTREE_LEVEL_NODE_MIN + 1;
  737. if (level <= nilfs_btree_height(btree) - 1) {
  738. node = nilfs_btree_get_node(btree, path, level);
  739. return nilfs_btree_node_get_ptr(btree, node,
  740. path[level].bp_index);
  741. }
  742. return NILFS_BMAP_INVALID_PTR;
  743. }
  744. static __u64 nilfs_btree_find_target_v(const struct nilfs_btree *btree,
  745. const struct nilfs_btree_path *path,
  746. __u64 key)
  747. {
  748. __u64 ptr;
  749. ptr = nilfs_bmap_find_target_seq(&btree->bt_bmap, key);
  750. if (ptr != NILFS_BMAP_INVALID_PTR)
  751. /* sequential access */
  752. return ptr;
  753. else {
  754. ptr = nilfs_btree_find_near(btree, path);
  755. if (ptr != NILFS_BMAP_INVALID_PTR)
  756. /* near */
  757. return ptr;
  758. }
  759. /* block group */
  760. return nilfs_bmap_find_target_in_group(&btree->bt_bmap);
  761. }
  762. static void nilfs_btree_set_target_v(struct nilfs_btree *btree, __u64 key,
  763. __u64 ptr)
  764. {
  765. btree->bt_bmap.b_last_allocated_key = key;
  766. btree->bt_bmap.b_last_allocated_ptr = ptr;
  767. }
  768. static int nilfs_btree_prepare_insert(struct nilfs_btree *btree,
  769. struct nilfs_btree_path *path,
  770. int *levelp, __u64 key, __u64 ptr,
  771. struct nilfs_bmap_stats *stats)
  772. {
  773. struct buffer_head *bh;
  774. struct nilfs_btree_node *node, *parent, *sib;
  775. __u64 sibptr;
  776. int pindex, level, ret;
  777. stats->bs_nblocks = 0;
  778. level = NILFS_BTREE_LEVEL_DATA;
  779. /* allocate a new ptr for data block */
  780. if (btree->bt_ops->btop_find_target != NULL)
  781. path[level].bp_newreq.bpr_ptr =
  782. btree->bt_ops->btop_find_target(btree, path, key);
  783. ret = btree->bt_bmap.b_pops->bpop_prepare_alloc_ptr(
  784. &btree->bt_bmap, &path[level].bp_newreq);
  785. if (ret < 0)
  786. goto err_out_data;
  787. for (level = NILFS_BTREE_LEVEL_NODE_MIN;
  788. level < nilfs_btree_height(btree) - 1;
  789. level++) {
  790. node = nilfs_btree_get_nonroot_node(btree, path, level);
  791. if (nilfs_btree_node_get_nchildren(btree, node) <
  792. nilfs_btree_node_nchildren_max(btree, node)) {
  793. path[level].bp_op = nilfs_btree_do_insert;
  794. stats->bs_nblocks++;
  795. goto out;
  796. }
  797. parent = nilfs_btree_get_node(btree, path, level + 1);
  798. pindex = path[level + 1].bp_index;
  799. /* left sibling */
  800. if (pindex > 0) {
  801. sibptr = nilfs_btree_node_get_ptr(btree, parent,
  802. pindex - 1);
  803. ret = nilfs_btree_get_block(btree, sibptr, &bh);
  804. if (ret < 0)
  805. goto err_out_child_node;
  806. sib = (struct nilfs_btree_node *)bh->b_data;
  807. if (nilfs_btree_node_get_nchildren(btree, sib) <
  808. nilfs_btree_node_nchildren_max(btree, sib)) {
  809. path[level].bp_sib_bh = bh;
  810. path[level].bp_op = nilfs_btree_carry_left;
  811. stats->bs_nblocks++;
  812. goto out;
  813. } else
  814. brelse(bh);
  815. }
  816. /* right sibling */
  817. if (pindex <
  818. nilfs_btree_node_get_nchildren(btree, parent) - 1) {
  819. sibptr = nilfs_btree_node_get_ptr(btree, parent,
  820. pindex + 1);
  821. ret = nilfs_btree_get_block(btree, sibptr, &bh);
  822. if (ret < 0)
  823. goto err_out_child_node;
  824. sib = (struct nilfs_btree_node *)bh->b_data;
  825. if (nilfs_btree_node_get_nchildren(btree, sib) <
  826. nilfs_btree_node_nchildren_max(btree, sib)) {
  827. path[level].bp_sib_bh = bh;
  828. path[level].bp_op = nilfs_btree_carry_right;
  829. stats->bs_nblocks++;
  830. goto out;
  831. } else
  832. brelse(bh);
  833. }
  834. /* split */
  835. path[level].bp_newreq.bpr_ptr =
  836. path[level - 1].bp_newreq.bpr_ptr + 1;
  837. ret = btree->bt_bmap.b_pops->bpop_prepare_alloc_ptr(
  838. &btree->bt_bmap, &path[level].bp_newreq);
  839. if (ret < 0)
  840. goto err_out_child_node;
  841. ret = nilfs_btree_get_new_block(btree,
  842. path[level].bp_newreq.bpr_ptr,
  843. &bh);
  844. if (ret < 0)
  845. goto err_out_curr_node;
  846. stats->bs_nblocks++;
  847. lock_buffer(bh);
  848. nilfs_btree_node_init(btree,
  849. (struct nilfs_btree_node *)bh->b_data,
  850. 0, level, 0, NULL, NULL);
  851. unlock_buffer(bh);
  852. path[level].bp_sib_bh = bh;
  853. path[level].bp_op = nilfs_btree_split;
  854. }
  855. /* root */
  856. node = nilfs_btree_get_root(btree);
  857. if (nilfs_btree_node_get_nchildren(btree, node) <
  858. nilfs_btree_node_nchildren_max(btree, node)) {
  859. path[level].bp_op = nilfs_btree_do_insert;
  860. stats->bs_nblocks++;
  861. goto out;
  862. }
  863. /* grow */
  864. path[level].bp_newreq.bpr_ptr = path[level - 1].bp_newreq.bpr_ptr + 1;
  865. ret = btree->bt_bmap.b_pops->bpop_prepare_alloc_ptr(
  866. &btree->bt_bmap, &path[level].bp_newreq);
  867. if (ret < 0)
  868. goto err_out_child_node;
  869. ret = nilfs_btree_get_new_block(btree, path[level].bp_newreq.bpr_ptr,
  870. &bh);
  871. if (ret < 0)
  872. goto err_out_curr_node;
  873. lock_buffer(bh);
  874. nilfs_btree_node_init(btree, (struct nilfs_btree_node *)bh->b_data,
  875. 0, level, 0, NULL, NULL);
  876. unlock_buffer(bh);
  877. path[level].bp_sib_bh = bh;
  878. path[level].bp_op = nilfs_btree_grow;
  879. level++;
  880. path[level].bp_op = nilfs_btree_do_insert;
  881. /* a newly-created node block and a data block are added */
  882. stats->bs_nblocks += 2;
  883. /* success */
  884. out:
  885. *levelp = level;
  886. return ret;
  887. /* error */
  888. err_out_curr_node:
  889. btree->bt_bmap.b_pops->bpop_abort_alloc_ptr(&btree->bt_bmap,
  890. &path[level].bp_newreq);
  891. err_out_child_node:
  892. for (level--; level > NILFS_BTREE_LEVEL_DATA; level--) {
  893. nilfs_btnode_delete(path[level].bp_sib_bh);
  894. btree->bt_bmap.b_pops->bpop_abort_alloc_ptr(
  895. &btree->bt_bmap, &path[level].bp_newreq);
  896. }
  897. btree->bt_bmap.b_pops->bpop_abort_alloc_ptr(&btree->bt_bmap,
  898. &path[level].bp_newreq);
  899. err_out_data:
  900. *levelp = level;
  901. stats->bs_nblocks = 0;
  902. return ret;
  903. }
  904. static void nilfs_btree_commit_insert(struct nilfs_btree *btree,
  905. struct nilfs_btree_path *path,
  906. int maxlevel, __u64 key, __u64 ptr)
  907. {
  908. int level;
  909. set_buffer_nilfs_volatile((struct buffer_head *)((unsigned long)ptr));
  910. ptr = path[NILFS_BTREE_LEVEL_DATA].bp_newreq.bpr_ptr;
  911. if (btree->bt_ops->btop_set_target != NULL)
  912. btree->bt_ops->btop_set_target(btree, key, ptr);
  913. for (level = NILFS_BTREE_LEVEL_NODE_MIN; level <= maxlevel; level++) {
  914. btree->bt_bmap.b_pops->bpop_commit_alloc_ptr(
  915. &btree->bt_bmap, &path[level - 1].bp_newreq);
  916. path[level].bp_op(btree, path, level, &key, &ptr);
  917. }
  918. if (!nilfs_bmap_dirty(&btree->bt_bmap))
  919. nilfs_bmap_set_dirty(&btree->bt_bmap);
  920. }
  921. static int nilfs_btree_insert(struct nilfs_bmap *bmap, __u64 key, __u64 ptr)
  922. {
  923. struct nilfs_btree *btree;
  924. struct nilfs_btree_path *path;
  925. struct nilfs_bmap_stats stats;
  926. int level, ret;
  927. btree = (struct nilfs_btree *)bmap;
  928. path = nilfs_btree_alloc_path(btree);
  929. if (path == NULL)
  930. return -ENOMEM;
  931. nilfs_btree_init_path(btree, path);
  932. ret = nilfs_btree_do_lookup(btree, path, key, NULL,
  933. NILFS_BTREE_LEVEL_NODE_MIN);
  934. if (ret != -ENOENT) {
  935. if (ret == 0)
  936. ret = -EEXIST;
  937. goto out;
  938. }
  939. ret = nilfs_btree_prepare_insert(btree, path, &level, key, ptr, &stats);
  940. if (ret < 0)
  941. goto out;
  942. nilfs_btree_commit_insert(btree, path, level, key, ptr);
  943. nilfs_bmap_add_blocks(bmap, stats.bs_nblocks);
  944. out:
  945. nilfs_btree_clear_path(btree, path);
  946. nilfs_btree_free_path(btree, path);
  947. return ret;
  948. }
  949. static void nilfs_btree_do_delete(struct nilfs_btree *btree,
  950. struct nilfs_btree_path *path,
  951. int level, __u64 *keyp, __u64 *ptrp)
  952. {
  953. struct nilfs_btree_node *node;
  954. if (level < nilfs_btree_height(btree) - 1) {
  955. lock_buffer(path[level].bp_bh);
  956. node = nilfs_btree_get_nonroot_node(btree, path, level);
  957. nilfs_btree_node_delete(btree, node, keyp, ptrp,
  958. path[level].bp_index);
  959. if (!buffer_dirty(path[level].bp_bh))
  960. nilfs_btnode_mark_dirty(path[level].bp_bh);
  961. unlock_buffer(path[level].bp_bh);
  962. if (path[level].bp_index == 0)
  963. nilfs_btree_promote_key(btree, path, level + 1,
  964. nilfs_btree_node_get_key(btree, node, 0));
  965. } else {
  966. node = nilfs_btree_get_root(btree);
  967. nilfs_btree_node_delete(btree, node, keyp, ptrp,
  968. path[level].bp_index);
  969. }
  970. }
  971. static void nilfs_btree_borrow_left(struct nilfs_btree *btree,
  972. struct nilfs_btree_path *path,
  973. int level, __u64 *keyp, __u64 *ptrp)
  974. {
  975. struct nilfs_btree_node *node, *left;
  976. int nchildren, lnchildren, n;
  977. nilfs_btree_do_delete(btree, path, level, keyp, ptrp);
  978. lock_buffer(path[level].bp_bh);
  979. lock_buffer(path[level].bp_sib_bh);
  980. node = nilfs_btree_get_nonroot_node(btree, path, level);
  981. left = nilfs_btree_get_sib_node(btree, path, level);
  982. nchildren = nilfs_btree_node_get_nchildren(btree, node);
  983. lnchildren = nilfs_btree_node_get_nchildren(btree, left);
  984. n = (nchildren + lnchildren) / 2 - nchildren;
  985. nilfs_btree_node_move_right(btree, left, node, n);
  986. if (!buffer_dirty(path[level].bp_bh))
  987. nilfs_btnode_mark_dirty(path[level].bp_bh);
  988. if (!buffer_dirty(path[level].bp_sib_bh))
  989. nilfs_btnode_mark_dirty(path[level].bp_sib_bh);
  990. unlock_buffer(path[level].bp_bh);
  991. unlock_buffer(path[level].bp_sib_bh);
  992. nilfs_btree_promote_key(btree, path, level + 1,
  993. nilfs_btree_node_get_key(btree, node, 0));
  994. brelse(path[level].bp_sib_bh);
  995. path[level].bp_sib_bh = NULL;
  996. path[level].bp_index += n;
  997. }
  998. static void nilfs_btree_borrow_right(struct nilfs_btree *btree,
  999. struct nilfs_btree_path *path,
  1000. int level, __u64 *keyp, __u64 *ptrp)
  1001. {
  1002. struct nilfs_btree_node *node, *right;
  1003. int nchildren, rnchildren, n;
  1004. nilfs_btree_do_delete(btree, path, level, keyp, ptrp);
  1005. lock_buffer(path[level].bp_bh);
  1006. lock_buffer(path[level].bp_sib_bh);
  1007. node = nilfs_btree_get_nonroot_node(btree, path, level);
  1008. right = nilfs_btree_get_sib_node(btree, path, level);
  1009. nchildren = nilfs_btree_node_get_nchildren(btree, node);
  1010. rnchildren = nilfs_btree_node_get_nchildren(btree, right);
  1011. n = (nchildren + rnchildren) / 2 - nchildren;
  1012. nilfs_btree_node_move_left(btree, node, right, n);
  1013. if (!buffer_dirty(path[level].bp_bh))
  1014. nilfs_btnode_mark_dirty(path[level].bp_bh);
  1015. if (!buffer_dirty(path[level].bp_sib_bh))
  1016. nilfs_btnode_mark_dirty(path[level].bp_sib_bh);
  1017. unlock_buffer(path[level].bp_bh);
  1018. unlock_buffer(path[level].bp_sib_bh);
  1019. path[level + 1].bp_index++;
  1020. nilfs_btree_promote_key(btree, path, level + 1,
  1021. nilfs_btree_node_get_key(btree, right, 0));
  1022. path[level + 1].bp_index--;
  1023. brelse(path[level].bp_sib_bh);
  1024. path[level].bp_sib_bh = NULL;
  1025. }
  1026. static void nilfs_btree_concat_left(struct nilfs_btree *btree,
  1027. struct nilfs_btree_path *path,
  1028. int level, __u64 *keyp, __u64 *ptrp)
  1029. {
  1030. struct nilfs_btree_node *node, *left;
  1031. int n;
  1032. nilfs_btree_do_delete(btree, path, level, keyp, ptrp);
  1033. lock_buffer(path[level].bp_bh);
  1034. lock_buffer(path[level].bp_sib_bh);
  1035. node = nilfs_btree_get_nonroot_node(btree, path, level);
  1036. left = nilfs_btree_get_sib_node(btree, path, level);
  1037. n = nilfs_btree_node_get_nchildren(btree, node);
  1038. nilfs_btree_node_move_left(btree, left, node, n);
  1039. if (!buffer_dirty(path[level].bp_sib_bh))
  1040. nilfs_btnode_mark_dirty(path[level].bp_sib_bh);
  1041. unlock_buffer(path[level].bp_bh);
  1042. unlock_buffer(path[level].bp_sib_bh);
  1043. nilfs_btnode_delete(path[level].bp_bh);
  1044. path[level].bp_bh = path[level].bp_sib_bh;
  1045. path[level].bp_sib_bh = NULL;
  1046. path[level].bp_index += nilfs_btree_node_get_nchildren(btree, left);
  1047. }
  1048. static void nilfs_btree_concat_right(struct nilfs_btree *btree,
  1049. struct nilfs_btree_path *path,
  1050. int level, __u64 *keyp, __u64 *ptrp)
  1051. {
  1052. struct nilfs_btree_node *node, *right;
  1053. int n;
  1054. nilfs_btree_do_delete(btree, path, level, keyp, ptrp);
  1055. lock_buffer(path[level].bp_bh);
  1056. lock_buffer(path[level].bp_sib_bh);
  1057. node = nilfs_btree_get_nonroot_node(btree, path, level);
  1058. right = nilfs_btree_get_sib_node(btree, path, level);
  1059. n = nilfs_btree_node_get_nchildren(btree, right);
  1060. nilfs_btree_node_move_left(btree, node, right, n);
  1061. if (!buffer_dirty(path[level].bp_bh))
  1062. nilfs_btnode_mark_dirty(path[level].bp_bh);
  1063. unlock_buffer(path[level].bp_bh);
  1064. unlock_buffer(path[level].bp_sib_bh);
  1065. nilfs_btnode_delete(path[level].bp_sib_bh);
  1066. path[level].bp_sib_bh = NULL;
  1067. path[level + 1].bp_index++;
  1068. }
  1069. static void nilfs_btree_shrink(struct nilfs_btree *btree,
  1070. struct nilfs_btree_path *path,
  1071. int level, __u64 *keyp, __u64 *ptrp)
  1072. {
  1073. struct nilfs_btree_node *root, *child;
  1074. int n;
  1075. nilfs_btree_do_delete(btree, path, level, keyp, ptrp);
  1076. lock_buffer(path[level].bp_bh);
  1077. root = nilfs_btree_get_root(btree);
  1078. child = nilfs_btree_get_nonroot_node(btree, path, level);
  1079. nilfs_btree_node_delete(btree, root, NULL, NULL, 0);
  1080. nilfs_btree_node_set_level(btree, root, level);
  1081. n = nilfs_btree_node_get_nchildren(btree, child);
  1082. nilfs_btree_node_move_left(btree, root, child, n);
  1083. unlock_buffer(path[level].bp_bh);
  1084. nilfs_btnode_delete(path[level].bp_bh);
  1085. path[level].bp_bh = NULL;
  1086. }
  1087. static int nilfs_btree_prepare_delete(struct nilfs_btree *btree,
  1088. struct nilfs_btree_path *path,
  1089. int *levelp,
  1090. struct nilfs_bmap_stats *stats)
  1091. {
  1092. struct buffer_head *bh;
  1093. struct nilfs_btree_node *node, *parent, *sib;
  1094. __u64 sibptr;
  1095. int pindex, level, ret;
  1096. ret = 0;
  1097. stats->bs_nblocks = 0;
  1098. for (level = NILFS_BTREE_LEVEL_NODE_MIN;
  1099. level < nilfs_btree_height(btree) - 1;
  1100. level++) {
  1101. node = nilfs_btree_get_nonroot_node(btree, path, level);
  1102. path[level].bp_oldreq.bpr_ptr =
  1103. nilfs_btree_node_get_ptr(btree, node,
  1104. path[level].bp_index);
  1105. if (btree->bt_bmap.b_pops->bpop_prepare_end_ptr != NULL) {
  1106. ret = btree->bt_bmap.b_pops->bpop_prepare_end_ptr(
  1107. &btree->bt_bmap, &path[level].bp_oldreq);
  1108. if (ret < 0)
  1109. goto err_out_child_node;
  1110. }
  1111. if (nilfs_btree_node_get_nchildren(btree, node) >
  1112. nilfs_btree_node_nchildren_min(btree, node)) {
  1113. path[level].bp_op = nilfs_btree_do_delete;
  1114. stats->bs_nblocks++;
  1115. goto out;
  1116. }
  1117. parent = nilfs_btree_get_node(btree, path, level + 1);
  1118. pindex = path[level + 1].bp_index;
  1119. if (pindex > 0) {
  1120. /* left sibling */
  1121. sibptr = nilfs_btree_node_get_ptr(btree, parent,
  1122. pindex - 1);
  1123. ret = nilfs_btree_get_block(btree, sibptr, &bh);
  1124. if (ret < 0)
  1125. goto err_out_curr_node;
  1126. sib = (struct nilfs_btree_node *)bh->b_data;
  1127. if (nilfs_btree_node_get_nchildren(btree, sib) >
  1128. nilfs_btree_node_nchildren_min(btree, sib)) {
  1129. path[level].bp_sib_bh = bh;
  1130. path[level].bp_op = nilfs_btree_borrow_left;
  1131. stats->bs_nblocks++;
  1132. goto out;
  1133. } else {
  1134. path[level].bp_sib_bh = bh;
  1135. path[level].bp_op = nilfs_btree_concat_left;
  1136. stats->bs_nblocks++;
  1137. /* continue; */
  1138. }
  1139. } else if (pindex <
  1140. nilfs_btree_node_get_nchildren(btree, parent) - 1) {
  1141. /* right sibling */
  1142. sibptr = nilfs_btree_node_get_ptr(btree, parent,
  1143. pindex + 1);
  1144. ret = nilfs_btree_get_block(btree, sibptr, &bh);
  1145. if (ret < 0)
  1146. goto err_out_curr_node;
  1147. sib = (struct nilfs_btree_node *)bh->b_data;
  1148. if (nilfs_btree_node_get_nchildren(btree, sib) >
  1149. nilfs_btree_node_nchildren_min(btree, sib)) {
  1150. path[level].bp_sib_bh = bh;
  1151. path[level].bp_op = nilfs_btree_borrow_right;
  1152. stats->bs_nblocks++;
  1153. goto out;
  1154. } else {
  1155. path[level].bp_sib_bh = bh;
  1156. path[level].bp_op = nilfs_btree_concat_right;
  1157. stats->bs_nblocks++;
  1158. /* continue; */
  1159. }
  1160. } else {
  1161. /* no siblings */
  1162. /* the only child of the root node */
  1163. WARN_ON(level != nilfs_btree_height(btree) - 2);
  1164. if (nilfs_btree_node_get_nchildren(btree, node) - 1 <=
  1165. NILFS_BTREE_ROOT_NCHILDREN_MAX) {
  1166. path[level].bp_op = nilfs_btree_shrink;
  1167. stats->bs_nblocks += 2;
  1168. } else {
  1169. path[level].bp_op = nilfs_btree_do_delete;
  1170. stats->bs_nblocks++;
  1171. }
  1172. goto out;
  1173. }
  1174. }
  1175. node = nilfs_btree_get_root(btree);
  1176. path[level].bp_oldreq.bpr_ptr =
  1177. nilfs_btree_node_get_ptr(btree, node, path[level].bp_index);
  1178. if (btree->bt_bmap.b_pops->bpop_prepare_end_ptr != NULL) {
  1179. ret = btree->bt_bmap.b_pops->bpop_prepare_end_ptr(
  1180. &btree->bt_bmap, &path[level].bp_oldreq);
  1181. if (ret < 0)
  1182. goto err_out_child_node;
  1183. }
  1184. /* child of the root node is deleted */
  1185. path[level].bp_op = nilfs_btree_do_delete;
  1186. stats->bs_nblocks++;
  1187. /* success */
  1188. out:
  1189. *levelp = level;
  1190. return ret;
  1191. /* error */
  1192. err_out_curr_node:
  1193. if (btree->bt_bmap.b_pops->bpop_abort_end_ptr != NULL)
  1194. btree->bt_bmap.b_pops->bpop_abort_end_ptr(
  1195. &btree->bt_bmap, &path[level].bp_oldreq);
  1196. err_out_child_node:
  1197. for (level--; level >= NILFS_BTREE_LEVEL_NODE_MIN; level--) {
  1198. brelse(path[level].bp_sib_bh);
  1199. if (btree->bt_bmap.b_pops->bpop_abort_end_ptr != NULL)
  1200. btree->bt_bmap.b_pops->bpop_abort_end_ptr(
  1201. &btree->bt_bmap, &path[level].bp_oldreq);
  1202. }
  1203. *levelp = level;
  1204. stats->bs_nblocks = 0;
  1205. return ret;
  1206. }
  1207. static void nilfs_btree_commit_delete(struct nilfs_btree *btree,
  1208. struct nilfs_btree_path *path,
  1209. int maxlevel)
  1210. {
  1211. int level;
  1212. for (level = NILFS_BTREE_LEVEL_NODE_MIN; level <= maxlevel; level++) {
  1213. if (btree->bt_bmap.b_pops->bpop_commit_end_ptr != NULL)
  1214. btree->bt_bmap.b_pops->bpop_commit_end_ptr(
  1215. &btree->bt_bmap, &path[level].bp_oldreq);
  1216. path[level].bp_op(btree, path, level, NULL, NULL);
  1217. }
  1218. if (!nilfs_bmap_dirty(&btree->bt_bmap))
  1219. nilfs_bmap_set_dirty(&btree->bt_bmap);
  1220. }
  1221. static int nilfs_btree_delete(struct nilfs_bmap *bmap, __u64 key)
  1222. {
  1223. struct nilfs_btree *btree;
  1224. struct nilfs_btree_path *path;
  1225. struct nilfs_bmap_stats stats;
  1226. int level, ret;
  1227. btree = (struct nilfs_btree *)bmap;
  1228. path = nilfs_btree_alloc_path(btree);
  1229. if (path == NULL)
  1230. return -ENOMEM;
  1231. nilfs_btree_init_path(btree, path);
  1232. ret = nilfs_btree_do_lookup(btree, path, key, NULL,
  1233. NILFS_BTREE_LEVEL_NODE_MIN);
  1234. if (ret < 0)
  1235. goto out;
  1236. ret = nilfs_btree_prepare_delete(btree, path, &level, &stats);
  1237. if (ret < 0)
  1238. goto out;
  1239. nilfs_btree_commit_delete(btree, path, level);
  1240. nilfs_bmap_sub_blocks(bmap, stats.bs_nblocks);
  1241. out:
  1242. nilfs_btree_clear_path(btree, path);
  1243. nilfs_btree_free_path(btree, path);
  1244. return ret;
  1245. }
  1246. static int nilfs_btree_last_key(const struct nilfs_bmap *bmap, __u64 *keyp)
  1247. {
  1248. struct nilfs_btree *btree;
  1249. struct nilfs_btree_path *path;
  1250. int ret;
  1251. btree = (struct nilfs_btree *)bmap;
  1252. path = nilfs_btree_alloc_path(btree);
  1253. if (path == NULL)
  1254. return -ENOMEM;
  1255. nilfs_btree_init_path(btree, path);
  1256. ret = nilfs_btree_do_lookup_last(btree, path, keyp, NULL);
  1257. nilfs_btree_clear_path(btree, path);
  1258. nilfs_btree_free_path(btree, path);
  1259. return ret;
  1260. }
  1261. static int nilfs_btree_check_delete(struct nilfs_bmap *bmap, __u64 key)
  1262. {
  1263. struct buffer_head *bh;
  1264. struct nilfs_btree *btree;
  1265. struct nilfs_btree_node *root, *node;
  1266. __u64 maxkey, nextmaxkey;
  1267. __u64 ptr;
  1268. int nchildren, ret;
  1269. btree = (struct nilfs_btree *)bmap;
  1270. root = nilfs_btree_get_root(btree);
  1271. switch (nilfs_btree_height(btree)) {
  1272. case 2:
  1273. bh = NULL;
  1274. node = root;
  1275. break;
  1276. case 3:
  1277. nchildren = nilfs_btree_node_get_nchildren(btree, root);
  1278. if (nchildren > 1)
  1279. return 0;
  1280. ptr = nilfs_btree_node_get_ptr(btree, root, nchildren - 1);
  1281. ret = nilfs_btree_get_block(btree, ptr, &bh);
  1282. if (ret < 0)
  1283. return ret;
  1284. node = (struct nilfs_btree_node *)bh->b_data;
  1285. break;
  1286. default:
  1287. return 0;
  1288. }
  1289. nchildren = nilfs_btree_node_get_nchildren(btree, node);
  1290. maxkey = nilfs_btree_node_get_key(btree, node, nchildren - 1);
  1291. nextmaxkey = (nchildren > 1) ?
  1292. nilfs_btree_node_get_key(btree, node, nchildren - 2) : 0;
  1293. if (bh != NULL)
  1294. brelse(bh);
  1295. return (maxkey == key) && (nextmaxkey < NILFS_BMAP_LARGE_LOW);
  1296. }
  1297. static int nilfs_btree_gather_data(struct nilfs_bmap *bmap,
  1298. __u64 *keys, __u64 *ptrs, int nitems)
  1299. {
  1300. struct buffer_head *bh;
  1301. struct nilfs_btree *btree;
  1302. struct nilfs_btree_node *node, *root;
  1303. __le64 *dkeys;
  1304. __le64 *dptrs;
  1305. __u64 ptr;
  1306. int nchildren, i, ret;
  1307. btree = (struct nilfs_btree *)bmap;
  1308. root = nilfs_btree_get_root(btree);
  1309. switch (nilfs_btree_height(btree)) {
  1310. case 2:
  1311. bh = NULL;
  1312. node = root;
  1313. break;
  1314. case 3:
  1315. nchildren = nilfs_btree_node_get_nchildren(btree, root);
  1316. WARN_ON(nchildren > 1);
  1317. ptr = nilfs_btree_node_get_ptr(btree, root, nchildren - 1);
  1318. ret = nilfs_btree_get_block(btree, ptr, &bh);
  1319. if (ret < 0)
  1320. return ret;
  1321. node = (struct nilfs_btree_node *)bh->b_data;
  1322. break;
  1323. default:
  1324. node = NULL;
  1325. return -EINVAL;
  1326. }
  1327. nchildren = nilfs_btree_node_get_nchildren(btree, node);
  1328. if (nchildren < nitems)
  1329. nitems = nchildren;
  1330. dkeys = nilfs_btree_node_dkeys(btree, node);
  1331. dptrs = nilfs_btree_node_dptrs(btree, node);
  1332. for (i = 0; i < nitems; i++) {
  1333. keys[i] = nilfs_bmap_dkey_to_key(dkeys[i]);
  1334. ptrs[i] = nilfs_bmap_dptr_to_ptr(dptrs[i]);
  1335. }
  1336. if (bh != NULL)
  1337. brelse(bh);
  1338. return nitems;
  1339. }
  1340. static int
  1341. nilfs_btree_prepare_convert_and_insert(struct nilfs_bmap *bmap, __u64 key,
  1342. union nilfs_bmap_ptr_req *dreq,
  1343. union nilfs_bmap_ptr_req *nreq,
  1344. struct buffer_head **bhp,
  1345. struct nilfs_bmap_stats *stats)
  1346. {
  1347. struct buffer_head *bh;
  1348. struct nilfs_btree *btree;
  1349. int ret;
  1350. btree = (struct nilfs_btree *)bmap;
  1351. stats->bs_nblocks = 0;
  1352. /* for data */
  1353. /* cannot find near ptr */
  1354. if (btree->bt_ops->btop_find_target != NULL)
  1355. dreq->bpr_ptr
  1356. = btree->bt_ops->btop_find_target(btree, NULL, key);
  1357. ret = bmap->b_pops->bpop_prepare_alloc_ptr(bmap, dreq);
  1358. if (ret < 0)
  1359. return ret;
  1360. *bhp = NULL;
  1361. stats->bs_nblocks++;
  1362. if (nreq != NULL) {
  1363. nreq->bpr_ptr = dreq->bpr_ptr + 1;
  1364. ret = bmap->b_pops->bpop_prepare_alloc_ptr(bmap, nreq);
  1365. if (ret < 0)
  1366. goto err_out_dreq;
  1367. ret = nilfs_btree_get_new_block(btree, nreq->bpr_ptr, &bh);
  1368. if (ret < 0)
  1369. goto err_out_nreq;
  1370. *bhp = bh;
  1371. stats->bs_nblocks++;
  1372. }
  1373. /* success */
  1374. return 0;
  1375. /* error */
  1376. err_out_nreq:
  1377. bmap->b_pops->bpop_abort_alloc_ptr(bmap, nreq);
  1378. err_out_dreq:
  1379. bmap->b_pops->bpop_abort_alloc_ptr(bmap, dreq);
  1380. stats->bs_nblocks = 0;
  1381. return ret;
  1382. }
  1383. static void
  1384. nilfs_btree_commit_convert_and_insert(struct nilfs_bmap *bmap,
  1385. __u64 key, __u64 ptr,
  1386. const __u64 *keys, const __u64 *ptrs,
  1387. int n,
  1388. union nilfs_bmap_ptr_req *dreq,
  1389. union nilfs_bmap_ptr_req *nreq,
  1390. struct buffer_head *bh)
  1391. {
  1392. struct nilfs_btree *btree;
  1393. struct nilfs_btree_node *node;
  1394. __u64 tmpptr;
  1395. /* free resources */
  1396. if (bmap->b_ops->bop_clear != NULL)
  1397. bmap->b_ops->bop_clear(bmap);
  1398. /* ptr must be a pointer to a buffer head. */
  1399. set_buffer_nilfs_volatile((struct buffer_head *)((unsigned long)ptr));
  1400. /* convert and insert */
  1401. btree = (struct nilfs_btree *)bmap;
  1402. nilfs_btree_init(bmap);
  1403. if (nreq != NULL) {
  1404. bmap->b_pops->bpop_commit_alloc_ptr(bmap, dreq);
  1405. bmap->b_pops->bpop_commit_alloc_ptr(bmap, nreq);
  1406. /* create child node at level 1 */
  1407. lock_buffer(bh);
  1408. node = (struct nilfs_btree_node *)bh->b_data;
  1409. nilfs_btree_node_init(btree, node, 0, 1, n, keys, ptrs);
  1410. nilfs_btree_node_insert(btree, node,
  1411. key, dreq->bpr_ptr, n);
  1412. if (!buffer_dirty(bh))
  1413. nilfs_btnode_mark_dirty(bh);
  1414. if (!nilfs_bmap_dirty(bmap))
  1415. nilfs_bmap_set_dirty(bmap);
  1416. unlock_buffer(bh);
  1417. brelse(bh);
  1418. /* create root node at level 2 */
  1419. node = nilfs_btree_get_root(btree);
  1420. tmpptr = nreq->bpr_ptr;
  1421. nilfs_btree_node_init(btree, node, NILFS_BTREE_NODE_ROOT,
  1422. 2, 1, &keys[0], &tmpptr);
  1423. } else {
  1424. bmap->b_pops->bpop_commit_alloc_ptr(bmap, dreq);
  1425. /* create root node at level 1 */
  1426. node = nilfs_btree_get_root(btree);
  1427. nilfs_btree_node_init(btree, node, NILFS_BTREE_NODE_ROOT,
  1428. 1, n, keys, ptrs);
  1429. nilfs_btree_node_insert(btree, node,
  1430. key, dreq->bpr_ptr, n);
  1431. if (!nilfs_bmap_dirty(bmap))
  1432. nilfs_bmap_set_dirty(bmap);
  1433. }
  1434. if (btree->bt_ops->btop_set_target != NULL)
  1435. btree->bt_ops->btop_set_target(btree, key, dreq->bpr_ptr);
  1436. }
  1437. /**
  1438. * nilfs_btree_convert_and_insert -
  1439. * @bmap:
  1440. * @key:
  1441. * @ptr:
  1442. * @keys:
  1443. * @ptrs:
  1444. * @n:
  1445. */
  1446. int nilfs_btree_convert_and_insert(struct nilfs_bmap *bmap,
  1447. __u64 key, __u64 ptr,
  1448. const __u64 *keys, const __u64 *ptrs, int n)
  1449. {
  1450. struct buffer_head *bh;
  1451. union nilfs_bmap_ptr_req dreq, nreq, *di, *ni;
  1452. struct nilfs_bmap_stats stats;
  1453. int ret;
  1454. if (n + 1 <= NILFS_BTREE_ROOT_NCHILDREN_MAX) {
  1455. di = &dreq;
  1456. ni = NULL;
  1457. } else if ((n + 1) <= NILFS_BTREE_NODE_NCHILDREN_MAX(
  1458. 1 << bmap->b_inode->i_blkbits)) {
  1459. di = &dreq;
  1460. ni = &nreq;
  1461. } else {
  1462. di = NULL;
  1463. ni = NULL;
  1464. BUG();
  1465. }
  1466. ret = nilfs_btree_prepare_convert_and_insert(bmap, key, di, ni, &bh,
  1467. &stats);
  1468. if (ret < 0)
  1469. return ret;
  1470. nilfs_btree_commit_convert_and_insert(bmap, key, ptr, keys, ptrs, n,
  1471. di, ni, bh);
  1472. nilfs_bmap_add_blocks(bmap, stats.bs_nblocks);
  1473. return 0;
  1474. }
  1475. static int nilfs_btree_propagate_p(struct nilfs_btree *btree,
  1476. struct nilfs_btree_path *path,
  1477. int level,
  1478. struct buffer_head *bh)
  1479. {
  1480. while ((++level < nilfs_btree_height(btree) - 1) &&
  1481. !buffer_dirty(path[level].bp_bh))
  1482. nilfs_btnode_mark_dirty(path[level].bp_bh);
  1483. return 0;
  1484. }
  1485. static int nilfs_btree_prepare_update_v(struct nilfs_btree *btree,
  1486. struct nilfs_btree_path *path,
  1487. int level)
  1488. {
  1489. struct nilfs_btree_node *parent;
  1490. int ret;
  1491. parent = nilfs_btree_get_node(btree, path, level + 1);
  1492. path[level].bp_oldreq.bpr_ptr =
  1493. nilfs_btree_node_get_ptr(btree, parent,
  1494. path[level + 1].bp_index);
  1495. path[level].bp_newreq.bpr_ptr = path[level].bp_oldreq.bpr_ptr + 1;
  1496. ret = nilfs_bmap_prepare_update(&btree->bt_bmap,
  1497. &path[level].bp_oldreq,
  1498. &path[level].bp_newreq);
  1499. if (ret < 0)
  1500. return ret;
  1501. if (buffer_nilfs_node(path[level].bp_bh)) {
  1502. path[level].bp_ctxt.oldkey = path[level].bp_oldreq.bpr_ptr;
  1503. path[level].bp_ctxt.newkey = path[level].bp_newreq.bpr_ptr;
  1504. path[level].bp_ctxt.bh = path[level].bp_bh;
  1505. ret = nilfs_btnode_prepare_change_key(
  1506. &NILFS_BMAP_I(&btree->bt_bmap)->i_btnode_cache,
  1507. &path[level].bp_ctxt);
  1508. if (ret < 0) {
  1509. nilfs_bmap_abort_update(&btree->bt_bmap,
  1510. &path[level].bp_oldreq,
  1511. &path[level].bp_newreq);
  1512. return ret;
  1513. }
  1514. }
  1515. return 0;
  1516. }
  1517. static void nilfs_btree_commit_update_v(struct nilfs_btree *btree,
  1518. struct nilfs_btree_path *path,
  1519. int level)
  1520. {
  1521. struct nilfs_btree_node *parent;
  1522. nilfs_bmap_commit_update(&btree->bt_bmap,
  1523. &path[level].bp_oldreq,
  1524. &path[level].bp_newreq);
  1525. if (buffer_nilfs_node(path[level].bp_bh)) {
  1526. nilfs_btnode_commit_change_key(
  1527. &NILFS_BMAP_I(&btree->bt_bmap)->i_btnode_cache,
  1528. &path[level].bp_ctxt);
  1529. path[level].bp_bh = path[level].bp_ctxt.bh;
  1530. }
  1531. set_buffer_nilfs_volatile(path[level].bp_bh);
  1532. parent = nilfs_btree_get_node(btree, path, level + 1);
  1533. nilfs_btree_node_set_ptr(btree, parent, path[level + 1].bp_index,
  1534. path[level].bp_newreq.bpr_ptr);
  1535. }
  1536. static void nilfs_btree_abort_update_v(struct nilfs_btree *btree,
  1537. struct nilfs_btree_path *path,
  1538. int level)
  1539. {
  1540. nilfs_bmap_abort_update(&btree->bt_bmap,
  1541. &path[level].bp_oldreq,
  1542. &path[level].bp_newreq);
  1543. if (buffer_nilfs_node(path[level].bp_bh))
  1544. nilfs_btnode_abort_change_key(
  1545. &NILFS_BMAP_I(&btree->bt_bmap)->i_btnode_cache,
  1546. &path[level].bp_ctxt);
  1547. }
  1548. static int nilfs_btree_prepare_propagate_v(struct nilfs_btree *btree,
  1549. struct nilfs_btree_path *path,
  1550. int minlevel,
  1551. int *maxlevelp)
  1552. {
  1553. int level, ret;
  1554. level = minlevel;
  1555. if (!buffer_nilfs_volatile(path[level].bp_bh)) {
  1556. ret = nilfs_btree_prepare_update_v(btree, path, level);
  1557. if (ret < 0)
  1558. return ret;
  1559. }
  1560. while ((++level < nilfs_btree_height(btree) - 1) &&
  1561. !buffer_dirty(path[level].bp_bh)) {
  1562. WARN_ON(buffer_nilfs_volatile(path[level].bp_bh));
  1563. ret = nilfs_btree_prepare_update_v(btree, path, level);
  1564. if (ret < 0)
  1565. goto out;
  1566. }
  1567. /* success */
  1568. *maxlevelp = level - 1;
  1569. return 0;
  1570. /* error */
  1571. out:
  1572. while (--level > minlevel)
  1573. nilfs_btree_abort_update_v(btree, path, level);
  1574. if (!buffer_nilfs_volatile(path[level].bp_bh))
  1575. nilfs_btree_abort_update_v(btree, path, level);
  1576. return ret;
  1577. }
  1578. static void nilfs_btree_commit_propagate_v(struct nilfs_btree *btree,
  1579. struct nilfs_btree_path *path,
  1580. int minlevel,
  1581. int maxlevel,
  1582. struct buffer_head *bh)
  1583. {
  1584. int level;
  1585. if (!buffer_nilfs_volatile(path[minlevel].bp_bh))
  1586. nilfs_btree_commit_update_v(btree, path, minlevel);
  1587. for (level = minlevel + 1; level <= maxlevel; level++)
  1588. nilfs_btree_commit_update_v(btree, path, level);
  1589. }
  1590. static int nilfs_btree_propagate_v(struct nilfs_btree *btree,
  1591. struct nilfs_btree_path *path,
  1592. int level,
  1593. struct buffer_head *bh)
  1594. {
  1595. int maxlevel, ret;
  1596. struct nilfs_btree_node *parent;
  1597. __u64 ptr;
  1598. get_bh(bh);
  1599. path[level].bp_bh = bh;
  1600. ret = nilfs_btree_prepare_propagate_v(btree, path, level, &maxlevel);
  1601. if (ret < 0)
  1602. goto out;
  1603. if (buffer_nilfs_volatile(path[level].bp_bh)) {
  1604. parent = nilfs_btree_get_node(btree, path, level + 1);
  1605. ptr = nilfs_btree_node_get_ptr(btree, parent,
  1606. path[level + 1].bp_index);
  1607. ret = nilfs_bmap_mark_dirty(&btree->bt_bmap, ptr);
  1608. if (ret < 0)
  1609. goto out;
  1610. }
  1611. nilfs_btree_commit_propagate_v(btree, path, level, maxlevel, bh);
  1612. out:
  1613. brelse(path[level].bp_bh);
  1614. path[level].bp_bh = NULL;
  1615. return ret;
  1616. }
  1617. static int nilfs_btree_propagate(const struct nilfs_bmap *bmap,
  1618. struct buffer_head *bh)
  1619. {
  1620. struct nilfs_btree *btree;
  1621. struct nilfs_btree_path *path;
  1622. struct nilfs_btree_node *node;
  1623. __u64 key;
  1624. int level, ret;
  1625. WARN_ON(!buffer_dirty(bh));
  1626. btree = (struct nilfs_btree *)bmap;
  1627. path = nilfs_btree_alloc_path(btree);
  1628. if (path == NULL)
  1629. return -ENOMEM;
  1630. nilfs_btree_init_path(btree, path);
  1631. if (buffer_nilfs_node(bh)) {
  1632. node = (struct nilfs_btree_node *)bh->b_data;
  1633. key = nilfs_btree_node_get_key(btree, node, 0);
  1634. level = nilfs_btree_node_get_level(btree, node);
  1635. } else {
  1636. key = nilfs_bmap_data_get_key(bmap, bh);
  1637. level = NILFS_BTREE_LEVEL_DATA;
  1638. }
  1639. ret = nilfs_btree_do_lookup(btree, path, key, NULL, level + 1);
  1640. if (ret < 0) {
  1641. if (unlikely(ret == -ENOENT))
  1642. printk(KERN_CRIT "%s: key = %llu, level == %d\n",
  1643. __func__, (unsigned long long)key, level);
  1644. goto out;
  1645. }
  1646. ret = btree->bt_ops->btop_propagate(btree, path, level, bh);
  1647. out:
  1648. nilfs_btree_clear_path(btree, path);
  1649. nilfs_btree_free_path(btree, path);
  1650. return ret;
  1651. }
  1652. static int nilfs_btree_propagate_gc(const struct nilfs_bmap *bmap,
  1653. struct buffer_head *bh)
  1654. {
  1655. return nilfs_bmap_mark_dirty(bmap, bh->b_blocknr);
  1656. }
  1657. static void nilfs_btree_add_dirty_buffer(struct nilfs_btree *btree,
  1658. struct list_head *lists,
  1659. struct buffer_head *bh)
  1660. {
  1661. struct list_head *head;
  1662. struct buffer_head *cbh;
  1663. struct nilfs_btree_node *node, *cnode;
  1664. __u64 key, ckey;
  1665. int level;
  1666. get_bh(bh);
  1667. node = (struct nilfs_btree_node *)bh->b_data;
  1668. key = nilfs_btree_node_get_key(btree, node, 0);
  1669. level = nilfs_btree_node_get_level(btree, node);
  1670. list_for_each(head, &lists[level]) {
  1671. cbh = list_entry(head, struct buffer_head, b_assoc_buffers);
  1672. cnode = (struct nilfs_btree_node *)cbh->b_data;
  1673. ckey = nilfs_btree_node_get_key(btree, cnode, 0);
  1674. if (key < ckey)
  1675. break;
  1676. }
  1677. list_add_tail(&bh->b_assoc_buffers, head);
  1678. }
  1679. static void nilfs_btree_lookup_dirty_buffers(struct nilfs_bmap *bmap,
  1680. struct list_head *listp)
  1681. {
  1682. struct nilfs_btree *btree = (struct nilfs_btree *)bmap;
  1683. struct address_space *btcache = &NILFS_BMAP_I(bmap)->i_btnode_cache;
  1684. struct list_head lists[NILFS_BTREE_LEVEL_MAX];
  1685. struct pagevec pvec;
  1686. struct buffer_head *bh, *head;
  1687. pgoff_t index = 0;
  1688. int level, i;
  1689. for (level = NILFS_BTREE_LEVEL_NODE_MIN;
  1690. level < NILFS_BTREE_LEVEL_MAX;
  1691. level++)
  1692. INIT_LIST_HEAD(&lists[level]);
  1693. pagevec_init(&pvec, 0);
  1694. while (pagevec_lookup_tag(&pvec, btcache, &index, PAGECACHE_TAG_DIRTY,
  1695. PAGEVEC_SIZE)) {
  1696. for (i = 0; i < pagevec_count(&pvec); i++) {
  1697. bh = head = page_buffers(pvec.pages[i]);
  1698. do {
  1699. if (buffer_dirty(bh))
  1700. nilfs_btree_add_dirty_buffer(btree,
  1701. lists, bh);
  1702. } while ((bh = bh->b_this_page) != head);
  1703. }
  1704. pagevec_release(&pvec);
  1705. cond_resched();
  1706. }
  1707. for (level = NILFS_BTREE_LEVEL_NODE_MIN;
  1708. level < NILFS_BTREE_LEVEL_MAX;
  1709. level++)
  1710. list_splice(&lists[level], listp->prev);
  1711. }
  1712. static int nilfs_btree_assign_p(struct nilfs_btree *btree,
  1713. struct nilfs_btree_path *path,
  1714. int level,
  1715. struct buffer_head **bh,
  1716. sector_t blocknr,
  1717. union nilfs_binfo *binfo)
  1718. {
  1719. struct nilfs_btree_node *parent;
  1720. __u64 key;
  1721. __u64 ptr;
  1722. int ret;
  1723. parent = nilfs_btree_get_node(btree, path, level + 1);
  1724. ptr = nilfs_btree_node_get_ptr(btree, parent,
  1725. path[level + 1].bp_index);
  1726. if (buffer_nilfs_node(*bh)) {
  1727. path[level].bp_ctxt.oldkey = ptr;
  1728. path[level].bp_ctxt.newkey = blocknr;
  1729. path[level].bp_ctxt.bh = *bh;
  1730. ret = nilfs_btnode_prepare_change_key(
  1731. &NILFS_BMAP_I(&btree->bt_bmap)->i_btnode_cache,
  1732. &path[level].bp_ctxt);
  1733. if (ret < 0)
  1734. return ret;
  1735. nilfs_btnode_commit_change_key(
  1736. &NILFS_BMAP_I(&btree->bt_bmap)->i_btnode_cache,
  1737. &path[level].bp_ctxt);
  1738. *bh = path[level].bp_ctxt.bh;
  1739. }
  1740. nilfs_btree_node_set_ptr(btree, parent,
  1741. path[level + 1].bp_index, blocknr);
  1742. key = nilfs_btree_node_get_key(btree, parent,
  1743. path[level + 1].bp_index);
  1744. /* on-disk format */
  1745. binfo->bi_dat.bi_blkoff = nilfs_bmap_key_to_dkey(key);
  1746. binfo->bi_dat.bi_level = level;
  1747. return 0;
  1748. }
  1749. static int nilfs_btree_assign_v(struct nilfs_btree *btree,
  1750. struct nilfs_btree_path *path,
  1751. int level,
  1752. struct buffer_head **bh,
  1753. sector_t blocknr,
  1754. union nilfs_binfo *binfo)
  1755. {
  1756. struct nilfs_btree_node *parent;
  1757. __u64 key;
  1758. __u64 ptr;
  1759. union nilfs_bmap_ptr_req req;
  1760. int ret;
  1761. parent = nilfs_btree_get_node(btree, path, level + 1);
  1762. ptr = nilfs_btree_node_get_ptr(btree, parent,
  1763. path[level + 1].bp_index);
  1764. req.bpr_ptr = ptr;
  1765. ret = nilfs_bmap_start_v(&btree->bt_bmap, &req, blocknr);
  1766. if (unlikely(ret < 0))
  1767. return ret;
  1768. key = nilfs_btree_node_get_key(btree, parent,
  1769. path[level + 1].bp_index);
  1770. /* on-disk format */
  1771. binfo->bi_v.bi_vblocknr = nilfs_bmap_ptr_to_dptr(ptr);
  1772. binfo->bi_v.bi_blkoff = nilfs_bmap_key_to_dkey(key);
  1773. return 0;
  1774. }
  1775. static int nilfs_btree_assign(struct nilfs_bmap *bmap,
  1776. struct buffer_head **bh,
  1777. sector_t blocknr,
  1778. union nilfs_binfo *binfo)
  1779. {
  1780. struct nilfs_btree *btree;
  1781. struct nilfs_btree_path *path;
  1782. struct nilfs_btree_node *node;
  1783. __u64 key;
  1784. int level, ret;
  1785. btree = (struct nilfs_btree *)bmap;
  1786. path = nilfs_btree_alloc_path(btree);
  1787. if (path == NULL)
  1788. return -ENOMEM;
  1789. nilfs_btree_init_path(btree, path);
  1790. if (buffer_nilfs_node(*bh)) {
  1791. node = (struct nilfs_btree_node *)(*bh)->b_data;
  1792. key = nilfs_btree_node_get_key(btree, node, 0);
  1793. level = nilfs_btree_node_get_level(btree, node);
  1794. } else {
  1795. key = nilfs_bmap_data_get_key(bmap, *bh);
  1796. level = NILFS_BTREE_LEVEL_DATA;
  1797. }
  1798. ret = nilfs_btree_do_lookup(btree, path, key, NULL, level + 1);
  1799. if (ret < 0) {
  1800. WARN_ON(ret == -ENOENT);
  1801. goto out;
  1802. }
  1803. ret = btree->bt_ops->btop_assign(btree, path, level, bh,
  1804. blocknr, binfo);
  1805. out:
  1806. nilfs_btree_clear_path(btree, path);
  1807. nilfs_btree_free_path(btree, path);
  1808. return ret;
  1809. }
  1810. static int nilfs_btree_assign_gc(struct nilfs_bmap *bmap,
  1811. struct buffer_head **bh,
  1812. sector_t blocknr,
  1813. union nilfs_binfo *binfo)
  1814. {
  1815. struct nilfs_btree *btree;
  1816. struct nilfs_btree_node *node;
  1817. __u64 key;
  1818. int ret;
  1819. btree = (struct nilfs_btree *)bmap;
  1820. ret = nilfs_bmap_move_v(bmap, (*bh)->b_blocknr, blocknr);
  1821. if (ret < 0)
  1822. return ret;
  1823. if (buffer_nilfs_node(*bh)) {
  1824. node = (struct nilfs_btree_node *)(*bh)->b_data;
  1825. key = nilfs_btree_node_get_key(btree, node, 0);
  1826. } else
  1827. key = nilfs_bmap_data_get_key(bmap, *bh);
  1828. /* on-disk format */
  1829. binfo->bi_v.bi_vblocknr = cpu_to_le64((*bh)->b_blocknr);
  1830. binfo->bi_v.bi_blkoff = nilfs_bmap_key_to_dkey(key);
  1831. return 0;
  1832. }
  1833. static int nilfs_btree_mark(struct nilfs_bmap *bmap, __u64 key, int level)
  1834. {
  1835. struct buffer_head *bh;
  1836. struct nilfs_btree *btree;
  1837. struct nilfs_btree_path *path;
  1838. __u64 ptr;
  1839. int ret;
  1840. btree = (struct nilfs_btree *)bmap;
  1841. path = nilfs_btree_alloc_path(btree);
  1842. if (path == NULL)
  1843. return -ENOMEM;
  1844. nilfs_btree_init_path(btree, path);
  1845. ret = nilfs_btree_do_lookup(btree, path, key, &ptr, level + 1);
  1846. if (ret < 0) {
  1847. WARN_ON(ret == -ENOENT);
  1848. goto out;
  1849. }
  1850. ret = nilfs_btree_get_block(btree, ptr, &bh);
  1851. if (ret < 0) {
  1852. WARN_ON(ret == -ENOENT);
  1853. goto out;
  1854. }
  1855. if (!buffer_dirty(bh))
  1856. nilfs_btnode_mark_dirty(bh);
  1857. brelse(bh);
  1858. if (!nilfs_bmap_dirty(&btree->bt_bmap))
  1859. nilfs_bmap_set_dirty(&btree->bt_bmap);
  1860. out:
  1861. nilfs_btree_clear_path(btree, path);
  1862. nilfs_btree_free_path(btree, path);
  1863. return ret;
  1864. }
  1865. static const struct nilfs_bmap_operations nilfs_btree_ops = {
  1866. .bop_lookup = nilfs_btree_lookup,
  1867. .bop_insert = nilfs_btree_insert,
  1868. .bop_delete = nilfs_btree_delete,
  1869. .bop_clear = NULL,
  1870. .bop_propagate = nilfs_btree_propagate,
  1871. .bop_lookup_dirty_buffers = nilfs_btree_lookup_dirty_buffers,
  1872. .bop_assign = nilfs_btree_assign,
  1873. .bop_mark = nilfs_btree_mark,
  1874. .bop_last_key = nilfs_btree_last_key,
  1875. .bop_check_insert = NULL,
  1876. .bop_check_delete = nilfs_btree_check_delete,
  1877. .bop_gather_data = nilfs_btree_gather_data,
  1878. };
  1879. static const struct nilfs_bmap_operations nilfs_btree_ops_gc = {
  1880. .bop_lookup = NULL,
  1881. .bop_insert = NULL,
  1882. .bop_delete = NULL,
  1883. .bop_clear = NULL,
  1884. .bop_propagate = nilfs_btree_propagate_gc,
  1885. .bop_lookup_dirty_buffers = nilfs_btree_lookup_dirty_buffers,
  1886. .bop_assign = nilfs_btree_assign_gc,
  1887. .bop_mark = NULL,
  1888. .bop_last_key = NULL,
  1889. .bop_check_insert = NULL,
  1890. .bop_check_delete = NULL,
  1891. .bop_gather_data = NULL,
  1892. };
  1893. static const struct nilfs_btree_operations nilfs_btree_ops_v = {
  1894. .btop_find_target = nilfs_btree_find_target_v,
  1895. .btop_set_target = nilfs_btree_set_target_v,
  1896. .btop_propagate = nilfs_btree_propagate_v,
  1897. .btop_assign = nilfs_btree_assign_v,
  1898. };
  1899. static const struct nilfs_btree_operations nilfs_btree_ops_p = {
  1900. .btop_find_target = NULL,
  1901. .btop_set_target = NULL,
  1902. .btop_propagate = nilfs_btree_propagate_p,
  1903. .btop_assign = nilfs_btree_assign_p,
  1904. };
  1905. int nilfs_btree_init(struct nilfs_bmap *bmap)
  1906. {
  1907. struct nilfs_btree *btree = (struct nilfs_btree *)bmap;
  1908. bmap->b_ops = &nilfs_btree_ops;
  1909. switch (bmap->b_inode->i_ino) {
  1910. case NILFS_DAT_INO:
  1911. btree->bt_ops = &nilfs_btree_ops_p;
  1912. break;
  1913. default:
  1914. btree->bt_ops = &nilfs_btree_ops_v;
  1915. break;
  1916. }
  1917. return 0;
  1918. }
  1919. void nilfs_btree_init_gc(struct nilfs_bmap *bmap)
  1920. {
  1921. bmap->b_ops = &nilfs_btree_ops_gc;
  1922. }