recovery.c 43 KB

12345678910111213141516171819202122232425262728293031323334353637383940414243444546474849505152535455565758596061626364656667686970717273747576777879808182838485868788899091929394959697989910010110210310410510610710810911011111211311411511611711811912012112212312412512612712812913013113213313413513613713813914014114214314414514614714814915015115215315415515615715815916016116216316416516616716816917017117217317417517617717817918018118218318418518618718818919019119219319419519619719819920020120220320420520620720820921021121221321421521621721821922022122222322422522622722822923023123223323423523623723823924024124224324424524624724824925025125225325425525625725825926026126226326426526626726826927027127227327427527627727827928028128228328428528628728828929029129229329429529629729829930030130230330430530630730830931031131231331431531631731831932032132232332432532632732832933033133233333433533633733833934034134234334434534634734834935035135235335435535635735835936036136236336436536636736836937037137237337437537637737837938038138238338438538638738838939039139239339439539639739839940040140240340440540640740840941041141241341441541641741841942042142242342442542642742842943043143243343443543643743843944044144244344444544644744844945045145245345445545645745845946046146246346446546646746846947047147247347447547647747847948048148248348448548648748848949049149249349449549649749849950050150250350450550650750850951051151251351451551651751851952052152252352452552652752852953053153253353453553653753853954054154254354454554654754854955055155255355455555655755855956056156256356456556656756856957057157257357457557657757857958058158258358458558658758858959059159259359459559659759859960060160260360460560660760860961061161261361461561661761861962062162262362462562662762862963063163263363463563663763863964064164264364464564664764864965065165265365465565665765865966066166266366466566666766866967067167267367467567667767867968068168268368468568668768868969069169269369469569669769869970070170270370470570670770870971071171271371471571671771871972072172272372472572672772872973073173273373473573673773873974074174274374474574674774874975075175275375475575675775875976076176276376476576676776876977077177277377477577677777877978078178278378478578678778878979079179279379479579679779879980080180280380480580680780880981081181281381481581681781881982082182282382482582682782882983083183283383483583683783883984084184284384484584684784884985085185285385485585685785885986086186286386486586686786886987087187287387487587687787887988088188288388488588688788888989089189289389489589689789889990090190290390490590690790890991091191291391491591691791891992092192292392492592692792892993093193293393493593693793893994094194294394494594694794894995095195295395495595695795895996096196296396496596696796896997097197297397497597697797897998098198298398498598698798898999099199299399499599699799899910001001100210031004100510061007100810091010101110121013101410151016101710181019102010211022102310241025102610271028102910301031103210331034103510361037103810391040104110421043104410451046104710481049105010511052105310541055105610571058105910601061106210631064106510661067106810691070107110721073107410751076107710781079108010811082108310841085108610871088108910901091109210931094109510961097109810991100110111021103110411051106110711081109111011111112111311141115111611171118111911201121112211231124112511261127112811291130113111321133113411351136113711381139114011411142114311441145114611471148114911501151115211531154115511561157115811591160116111621163116411651166116711681169117011711172117311741175117611771178117911801181118211831184118511861187118811891190119111921193119411951196119711981199120012011202120312041205120612071208120912101211121212131214121512161217121812191220122112221223122412251226122712281229123012311232123312341235123612371238123912401241124212431244124512461247124812491250125112521253125412551256125712581259126012611262126312641265126612671268126912701271127212731274127512761277127812791280128112821283128412851286128712881289129012911292129312941295129612971298129913001301130213031304130513061307130813091310131113121313131413151316131713181319132013211322132313241325132613271328132913301331133213331334133513361337133813391340134113421343134413451346134713481349135013511352135313541355135613571358135913601361136213631364136513661367136813691370137113721373137413751376137713781379138013811382138313841385138613871388138913901391139213931394139513961397139813991400140114021403140414051406140714081409141014111412141314141415141614171418141914201421142214231424142514261427142814291430143114321433143414351436143714381439144014411442144314441445144614471448144914501451145214531454145514561457145814591460146114621463146414651466146714681469147014711472147314741475147614771478147914801481148214831484148514861487148814891490149114921493149414951496149714981499150015011502150315041505150615071508150915101511151215131514151515161517151815191520152115221523152415251526152715281529153015311532153315341535153615371538153915401541154215431544154515461547154815491550155115521553155415551556155715581559156015611562
  1. /*
  2. * This file is part of UBIFS.
  3. *
  4. * Copyright (C) 2006-2008 Nokia Corporation
  5. *
  6. * SPDX-License-Identifier: GPL-2.0+
  7. *
  8. * Authors: Adrian Hunter
  9. * Artem Bityutskiy (Битюцкий Артём)
  10. */
  11. /*
  12. * This file implements functions needed to recover from unclean un-mounts.
  13. * When UBIFS is mounted, it checks a flag on the master node to determine if
  14. * an un-mount was completed successfully. If not, the process of mounting
  15. * incorporates additional checking and fixing of on-flash data structures.
  16. * UBIFS always cleans away all remnants of an unclean un-mount, so that
  17. * errors do not accumulate. However UBIFS defers recovery if it is mounted
  18. * read-only, and the flash is not modified in that case.
  19. *
  20. * The general UBIFS approach to the recovery is that it recovers from
  21. * corruptions which could be caused by power cuts, but it refuses to recover
  22. * from corruption caused by other reasons. And UBIFS tries to distinguish
  23. * between these 2 reasons of corruptions and silently recover in the former
  24. * case and loudly complain in the latter case.
  25. *
  26. * UBIFS writes only to erased LEBs, so it writes only to the flash space
  27. * containing only 0xFFs. UBIFS also always writes strictly from the beginning
  28. * of the LEB to the end. And UBIFS assumes that the underlying flash media
  29. * writes in @c->max_write_size bytes at a time.
  30. *
  31. * Hence, if UBIFS finds a corrupted node at offset X, it expects only the min.
  32. * I/O unit corresponding to offset X to contain corrupted data, all the
  33. * following min. I/O units have to contain empty space (all 0xFFs). If this is
  34. * not true, the corruption cannot be the result of a power cut, and UBIFS
  35. * refuses to mount.
  36. */
  37. #ifndef __UBOOT__
  38. #include <linux/crc32.h>
  39. #include <linux/slab.h>
  40. #else
  41. #include <linux/err.h>
  42. #endif
  43. #include "ubifs.h"
  44. /**
  45. * is_empty - determine whether a buffer is empty (contains all 0xff).
  46. * @buf: buffer to clean
  47. * @len: length of buffer
  48. *
  49. * This function returns %1 if the buffer is empty (contains all 0xff) otherwise
  50. * %0 is returned.
  51. */
  52. static int is_empty(void *buf, int len)
  53. {
  54. uint8_t *p = buf;
  55. int i;
  56. for (i = 0; i < len; i++)
  57. if (*p++ != 0xff)
  58. return 0;
  59. return 1;
  60. }
  61. /**
  62. * first_non_ff - find offset of the first non-0xff byte.
  63. * @buf: buffer to search in
  64. * @len: length of buffer
  65. *
  66. * This function returns offset of the first non-0xff byte in @buf or %-1 if
  67. * the buffer contains only 0xff bytes.
  68. */
  69. static int first_non_ff(void *buf, int len)
  70. {
  71. uint8_t *p = buf;
  72. int i;
  73. for (i = 0; i < len; i++)
  74. if (*p++ != 0xff)
  75. return i;
  76. return -1;
  77. }
  78. /**
  79. * get_master_node - get the last valid master node allowing for corruption.
  80. * @c: UBIFS file-system description object
  81. * @lnum: LEB number
  82. * @pbuf: buffer containing the LEB read, is returned here
  83. * @mst: master node, if found, is returned here
  84. * @cor: corruption, if found, is returned here
  85. *
  86. * This function allocates a buffer, reads the LEB into it, and finds and
  87. * returns the last valid master node allowing for one area of corruption.
  88. * The corrupt area, if there is one, must be consistent with the assumption
  89. * that it is the result of an unclean unmount while the master node was being
  90. * written. Under those circumstances, it is valid to use the previously written
  91. * master node.
  92. *
  93. * This function returns %0 on success and a negative error code on failure.
  94. */
  95. static int get_master_node(const struct ubifs_info *c, int lnum, void **pbuf,
  96. struct ubifs_mst_node **mst, void **cor)
  97. {
  98. const int sz = c->mst_node_alsz;
  99. int err, offs, len;
  100. void *sbuf, *buf;
  101. sbuf = vmalloc(c->leb_size);
  102. if (!sbuf)
  103. return -ENOMEM;
  104. err = ubifs_leb_read(c, lnum, sbuf, 0, c->leb_size, 0);
  105. if (err && err != -EBADMSG)
  106. goto out_free;
  107. /* Find the first position that is definitely not a node */
  108. offs = 0;
  109. buf = sbuf;
  110. len = c->leb_size;
  111. while (offs + UBIFS_MST_NODE_SZ <= c->leb_size) {
  112. struct ubifs_ch *ch = buf;
  113. if (le32_to_cpu(ch->magic) != UBIFS_NODE_MAGIC)
  114. break;
  115. offs += sz;
  116. buf += sz;
  117. len -= sz;
  118. }
  119. /* See if there was a valid master node before that */
  120. if (offs) {
  121. int ret;
  122. offs -= sz;
  123. buf -= sz;
  124. len += sz;
  125. ret = ubifs_scan_a_node(c, buf, len, lnum, offs, 1);
  126. if (ret != SCANNED_A_NODE && offs) {
  127. /* Could have been corruption so check one place back */
  128. offs -= sz;
  129. buf -= sz;
  130. len += sz;
  131. ret = ubifs_scan_a_node(c, buf, len, lnum, offs, 1);
  132. if (ret != SCANNED_A_NODE)
  133. /*
  134. * We accept only one area of corruption because
  135. * we are assuming that it was caused while
  136. * trying to write a master node.
  137. */
  138. goto out_err;
  139. }
  140. if (ret == SCANNED_A_NODE) {
  141. struct ubifs_ch *ch = buf;
  142. if (ch->node_type != UBIFS_MST_NODE)
  143. goto out_err;
  144. dbg_rcvry("found a master node at %d:%d", lnum, offs);
  145. *mst = buf;
  146. offs += sz;
  147. buf += sz;
  148. len -= sz;
  149. }
  150. }
  151. /* Check for corruption */
  152. if (offs < c->leb_size) {
  153. if (!is_empty(buf, min_t(int, len, sz))) {
  154. *cor = buf;
  155. dbg_rcvry("found corruption at %d:%d", lnum, offs);
  156. }
  157. offs += sz;
  158. buf += sz;
  159. len -= sz;
  160. }
  161. /* Check remaining empty space */
  162. if (offs < c->leb_size)
  163. if (!is_empty(buf, len))
  164. goto out_err;
  165. *pbuf = sbuf;
  166. return 0;
  167. out_err:
  168. err = -EINVAL;
  169. out_free:
  170. vfree(sbuf);
  171. *mst = NULL;
  172. *cor = NULL;
  173. return err;
  174. }
  175. /**
  176. * write_rcvrd_mst_node - write recovered master node.
  177. * @c: UBIFS file-system description object
  178. * @mst: master node
  179. *
  180. * This function returns %0 on success and a negative error code on failure.
  181. */
  182. static int write_rcvrd_mst_node(struct ubifs_info *c,
  183. struct ubifs_mst_node *mst)
  184. {
  185. int err = 0, lnum = UBIFS_MST_LNUM, sz = c->mst_node_alsz;
  186. __le32 save_flags;
  187. dbg_rcvry("recovery");
  188. save_flags = mst->flags;
  189. mst->flags |= cpu_to_le32(UBIFS_MST_RCVRY);
  190. ubifs_prepare_node(c, mst, UBIFS_MST_NODE_SZ, 1);
  191. err = ubifs_leb_change(c, lnum, mst, sz);
  192. if (err)
  193. goto out;
  194. err = ubifs_leb_change(c, lnum + 1, mst, sz);
  195. if (err)
  196. goto out;
  197. out:
  198. mst->flags = save_flags;
  199. return err;
  200. }
  201. /**
  202. * ubifs_recover_master_node - recover the master node.
  203. * @c: UBIFS file-system description object
  204. *
  205. * This function recovers the master node from corruption that may occur due to
  206. * an unclean unmount.
  207. *
  208. * This function returns %0 on success and a negative error code on failure.
  209. */
  210. int ubifs_recover_master_node(struct ubifs_info *c)
  211. {
  212. void *buf1 = NULL, *buf2 = NULL, *cor1 = NULL, *cor2 = NULL;
  213. struct ubifs_mst_node *mst1 = NULL, *mst2 = NULL, *mst;
  214. const int sz = c->mst_node_alsz;
  215. int err, offs1, offs2;
  216. dbg_rcvry("recovery");
  217. err = get_master_node(c, UBIFS_MST_LNUM, &buf1, &mst1, &cor1);
  218. if (err)
  219. goto out_free;
  220. err = get_master_node(c, UBIFS_MST_LNUM + 1, &buf2, &mst2, &cor2);
  221. if (err)
  222. goto out_free;
  223. if (mst1) {
  224. offs1 = (void *)mst1 - buf1;
  225. if ((le32_to_cpu(mst1->flags) & UBIFS_MST_RCVRY) &&
  226. (offs1 == 0 && !cor1)) {
  227. /*
  228. * mst1 was written by recovery at offset 0 with no
  229. * corruption.
  230. */
  231. dbg_rcvry("recovery recovery");
  232. mst = mst1;
  233. } else if (mst2) {
  234. offs2 = (void *)mst2 - buf2;
  235. if (offs1 == offs2) {
  236. /* Same offset, so must be the same */
  237. if (memcmp((void *)mst1 + UBIFS_CH_SZ,
  238. (void *)mst2 + UBIFS_CH_SZ,
  239. UBIFS_MST_NODE_SZ - UBIFS_CH_SZ))
  240. goto out_err;
  241. mst = mst1;
  242. } else if (offs2 + sz == offs1) {
  243. /* 1st LEB was written, 2nd was not */
  244. if (cor1)
  245. goto out_err;
  246. mst = mst1;
  247. } else if (offs1 == 0 &&
  248. c->leb_size - offs2 - sz < sz) {
  249. /* 1st LEB was unmapped and written, 2nd not */
  250. if (cor1)
  251. goto out_err;
  252. mst = mst1;
  253. } else
  254. goto out_err;
  255. } else {
  256. /*
  257. * 2nd LEB was unmapped and about to be written, so
  258. * there must be only one master node in the first LEB
  259. * and no corruption.
  260. */
  261. if (offs1 != 0 || cor1)
  262. goto out_err;
  263. mst = mst1;
  264. }
  265. } else {
  266. if (!mst2)
  267. goto out_err;
  268. /*
  269. * 1st LEB was unmapped and about to be written, so there must
  270. * be no room left in 2nd LEB.
  271. */
  272. offs2 = (void *)mst2 - buf2;
  273. if (offs2 + sz + sz <= c->leb_size)
  274. goto out_err;
  275. mst = mst2;
  276. }
  277. ubifs_msg("recovered master node from LEB %d",
  278. (mst == mst1 ? UBIFS_MST_LNUM : UBIFS_MST_LNUM + 1));
  279. memcpy(c->mst_node, mst, UBIFS_MST_NODE_SZ);
  280. if (c->ro_mount) {
  281. /* Read-only mode. Keep a copy for switching to rw mode */
  282. c->rcvrd_mst_node = kmalloc(sz, GFP_KERNEL);
  283. if (!c->rcvrd_mst_node) {
  284. err = -ENOMEM;
  285. goto out_free;
  286. }
  287. memcpy(c->rcvrd_mst_node, c->mst_node, UBIFS_MST_NODE_SZ);
  288. /*
  289. * We had to recover the master node, which means there was an
  290. * unclean reboot. However, it is possible that the master node
  291. * is clean at this point, i.e., %UBIFS_MST_DIRTY is not set.
  292. * E.g., consider the following chain of events:
  293. *
  294. * 1. UBIFS was cleanly unmounted, so the master node is clean
  295. * 2. UBIFS is being mounted R/W and starts changing the master
  296. * node in the first (%UBIFS_MST_LNUM). A power cut happens,
  297. * so this LEB ends up with some amount of garbage at the
  298. * end.
  299. * 3. UBIFS is being mounted R/O. We reach this place and
  300. * recover the master node from the second LEB
  301. * (%UBIFS_MST_LNUM + 1). But we cannot update the media
  302. * because we are being mounted R/O. We have to defer the
  303. * operation.
  304. * 4. However, this master node (@c->mst_node) is marked as
  305. * clean (since the step 1). And if we just return, the
  306. * mount code will be confused and won't recover the master
  307. * node when it is re-mounter R/W later.
  308. *
  309. * Thus, to force the recovery by marking the master node as
  310. * dirty.
  311. */
  312. c->mst_node->flags |= cpu_to_le32(UBIFS_MST_DIRTY);
  313. #ifndef __UBOOT__
  314. } else {
  315. /* Write the recovered master node */
  316. c->max_sqnum = le64_to_cpu(mst->ch.sqnum) - 1;
  317. err = write_rcvrd_mst_node(c, c->mst_node);
  318. if (err)
  319. goto out_free;
  320. #endif
  321. }
  322. vfree(buf2);
  323. vfree(buf1);
  324. return 0;
  325. out_err:
  326. err = -EINVAL;
  327. out_free:
  328. ubifs_err("failed to recover master node");
  329. if (mst1) {
  330. ubifs_err("dumping first master node");
  331. ubifs_dump_node(c, mst1);
  332. }
  333. if (mst2) {
  334. ubifs_err("dumping second master node");
  335. ubifs_dump_node(c, mst2);
  336. }
  337. vfree(buf2);
  338. vfree(buf1);
  339. return err;
  340. }
  341. /**
  342. * ubifs_write_rcvrd_mst_node - write the recovered master node.
  343. * @c: UBIFS file-system description object
  344. *
  345. * This function writes the master node that was recovered during mounting in
  346. * read-only mode and must now be written because we are remounting rw.
  347. *
  348. * This function returns %0 on success and a negative error code on failure.
  349. */
  350. int ubifs_write_rcvrd_mst_node(struct ubifs_info *c)
  351. {
  352. int err;
  353. if (!c->rcvrd_mst_node)
  354. return 0;
  355. c->rcvrd_mst_node->flags |= cpu_to_le32(UBIFS_MST_DIRTY);
  356. c->mst_node->flags |= cpu_to_le32(UBIFS_MST_DIRTY);
  357. err = write_rcvrd_mst_node(c, c->rcvrd_mst_node);
  358. if (err)
  359. return err;
  360. kfree(c->rcvrd_mst_node);
  361. c->rcvrd_mst_node = NULL;
  362. return 0;
  363. }
  364. /**
  365. * is_last_write - determine if an offset was in the last write to a LEB.
  366. * @c: UBIFS file-system description object
  367. * @buf: buffer to check
  368. * @offs: offset to check
  369. *
  370. * This function returns %1 if @offs was in the last write to the LEB whose data
  371. * is in @buf, otherwise %0 is returned. The determination is made by checking
  372. * for subsequent empty space starting from the next @c->max_write_size
  373. * boundary.
  374. */
  375. static int is_last_write(const struct ubifs_info *c, void *buf, int offs)
  376. {
  377. int empty_offs, check_len;
  378. uint8_t *p;
  379. /*
  380. * Round up to the next @c->max_write_size boundary i.e. @offs is in
  381. * the last wbuf written. After that should be empty space.
  382. */
  383. empty_offs = ALIGN(offs + 1, c->max_write_size);
  384. check_len = c->leb_size - empty_offs;
  385. p = buf + empty_offs - offs;
  386. return is_empty(p, check_len);
  387. }
  388. /**
  389. * clean_buf - clean the data from an LEB sitting in a buffer.
  390. * @c: UBIFS file-system description object
  391. * @buf: buffer to clean
  392. * @lnum: LEB number to clean
  393. * @offs: offset from which to clean
  394. * @len: length of buffer
  395. *
  396. * This function pads up to the next min_io_size boundary (if there is one) and
  397. * sets empty space to all 0xff. @buf, @offs and @len are updated to the next
  398. * @c->min_io_size boundary.
  399. */
  400. static void clean_buf(const struct ubifs_info *c, void **buf, int lnum,
  401. int *offs, int *len)
  402. {
  403. int empty_offs, pad_len;
  404. lnum = lnum;
  405. dbg_rcvry("cleaning corruption at %d:%d", lnum, *offs);
  406. ubifs_assert(!(*offs & 7));
  407. empty_offs = ALIGN(*offs, c->min_io_size);
  408. pad_len = empty_offs - *offs;
  409. ubifs_pad(c, *buf, pad_len);
  410. *offs += pad_len;
  411. *buf += pad_len;
  412. *len -= pad_len;
  413. memset(*buf, 0xff, c->leb_size - empty_offs);
  414. }
  415. /**
  416. * no_more_nodes - determine if there are no more nodes in a buffer.
  417. * @c: UBIFS file-system description object
  418. * @buf: buffer to check
  419. * @len: length of buffer
  420. * @lnum: LEB number of the LEB from which @buf was read
  421. * @offs: offset from which @buf was read
  422. *
  423. * This function ensures that the corrupted node at @offs is the last thing
  424. * written to a LEB. This function returns %1 if more data is not found and
  425. * %0 if more data is found.
  426. */
  427. static int no_more_nodes(const struct ubifs_info *c, void *buf, int len,
  428. int lnum, int offs)
  429. {
  430. struct ubifs_ch *ch = buf;
  431. int skip, dlen = le32_to_cpu(ch->len);
  432. /* Check for empty space after the corrupt node's common header */
  433. skip = ALIGN(offs + UBIFS_CH_SZ, c->max_write_size) - offs;
  434. if (is_empty(buf + skip, len - skip))
  435. return 1;
  436. /*
  437. * The area after the common header size is not empty, so the common
  438. * header must be intact. Check it.
  439. */
  440. if (ubifs_check_node(c, buf, lnum, offs, 1, 0) != -EUCLEAN) {
  441. dbg_rcvry("unexpected bad common header at %d:%d", lnum, offs);
  442. return 0;
  443. }
  444. /* Now we know the corrupt node's length we can skip over it */
  445. skip = ALIGN(offs + dlen, c->max_write_size) - offs;
  446. /* After which there should be empty space */
  447. if (is_empty(buf + skip, len - skip))
  448. return 1;
  449. dbg_rcvry("unexpected data at %d:%d", lnum, offs + skip);
  450. return 0;
  451. }
  452. /**
  453. * fix_unclean_leb - fix an unclean LEB.
  454. * @c: UBIFS file-system description object
  455. * @sleb: scanned LEB information
  456. * @start: offset where scan started
  457. */
  458. static int fix_unclean_leb(struct ubifs_info *c, struct ubifs_scan_leb *sleb,
  459. int start)
  460. {
  461. int lnum = sleb->lnum, endpt = start;
  462. /* Get the end offset of the last node we are keeping */
  463. if (!list_empty(&sleb->nodes)) {
  464. struct ubifs_scan_node *snod;
  465. snod = list_entry(sleb->nodes.prev,
  466. struct ubifs_scan_node, list);
  467. endpt = snod->offs + snod->len;
  468. }
  469. if (c->ro_mount && !c->remounting_rw) {
  470. /* Add to recovery list */
  471. struct ubifs_unclean_leb *ucleb;
  472. dbg_rcvry("need to fix LEB %d start %d endpt %d",
  473. lnum, start, sleb->endpt);
  474. ucleb = kzalloc(sizeof(struct ubifs_unclean_leb), GFP_NOFS);
  475. if (!ucleb)
  476. return -ENOMEM;
  477. ucleb->lnum = lnum;
  478. ucleb->endpt = endpt;
  479. list_add_tail(&ucleb->list, &c->unclean_leb_list);
  480. #ifndef __UBOOT__
  481. } else {
  482. /* Write the fixed LEB back to flash */
  483. int err;
  484. dbg_rcvry("fixing LEB %d start %d endpt %d",
  485. lnum, start, sleb->endpt);
  486. if (endpt == 0) {
  487. err = ubifs_leb_unmap(c, lnum);
  488. if (err)
  489. return err;
  490. } else {
  491. int len = ALIGN(endpt, c->min_io_size);
  492. if (start) {
  493. err = ubifs_leb_read(c, lnum, sleb->buf, 0,
  494. start, 1);
  495. if (err)
  496. return err;
  497. }
  498. /* Pad to min_io_size */
  499. if (len > endpt) {
  500. int pad_len = len - ALIGN(endpt, 8);
  501. if (pad_len > 0) {
  502. void *buf = sleb->buf + len - pad_len;
  503. ubifs_pad(c, buf, pad_len);
  504. }
  505. }
  506. err = ubifs_leb_change(c, lnum, sleb->buf, len);
  507. if (err)
  508. return err;
  509. }
  510. #endif
  511. }
  512. return 0;
  513. }
  514. /**
  515. * drop_last_group - drop the last group of nodes.
  516. * @sleb: scanned LEB information
  517. * @offs: offset of dropped nodes is returned here
  518. *
  519. * This is a helper function for 'ubifs_recover_leb()' which drops the last
  520. * group of nodes of the scanned LEB.
  521. */
  522. static void drop_last_group(struct ubifs_scan_leb *sleb, int *offs)
  523. {
  524. while (!list_empty(&sleb->nodes)) {
  525. struct ubifs_scan_node *snod;
  526. struct ubifs_ch *ch;
  527. snod = list_entry(sleb->nodes.prev, struct ubifs_scan_node,
  528. list);
  529. ch = snod->node;
  530. if (ch->group_type != UBIFS_IN_NODE_GROUP)
  531. break;
  532. dbg_rcvry("dropping grouped node at %d:%d",
  533. sleb->lnum, snod->offs);
  534. *offs = snod->offs;
  535. list_del(&snod->list);
  536. kfree(snod);
  537. sleb->nodes_cnt -= 1;
  538. }
  539. }
  540. /**
  541. * drop_last_node - drop the last node.
  542. * @sleb: scanned LEB information
  543. * @offs: offset of dropped nodes is returned here
  544. * @grouped: non-zero if whole group of nodes have to be dropped
  545. *
  546. * This is a helper function for 'ubifs_recover_leb()' which drops the last
  547. * node of the scanned LEB.
  548. */
  549. static void drop_last_node(struct ubifs_scan_leb *sleb, int *offs)
  550. {
  551. struct ubifs_scan_node *snod;
  552. if (!list_empty(&sleb->nodes)) {
  553. snod = list_entry(sleb->nodes.prev, struct ubifs_scan_node,
  554. list);
  555. dbg_rcvry("dropping last node at %d:%d",
  556. sleb->lnum, snod->offs);
  557. *offs = snod->offs;
  558. list_del(&snod->list);
  559. kfree(snod);
  560. sleb->nodes_cnt -= 1;
  561. }
  562. }
  563. /**
  564. * ubifs_recover_leb - scan and recover a LEB.
  565. * @c: UBIFS file-system description object
  566. * @lnum: LEB number
  567. * @offs: offset
  568. * @sbuf: LEB-sized buffer to use
  569. * @jhead: journal head number this LEB belongs to (%-1 if the LEB does not
  570. * belong to any journal head)
  571. *
  572. * This function does a scan of a LEB, but caters for errors that might have
  573. * been caused by the unclean unmount from which we are attempting to recover.
  574. * Returns %0 in case of success, %-EUCLEAN if an unrecoverable corruption is
  575. * found, and a negative error code in case of failure.
  576. */
  577. struct ubifs_scan_leb *ubifs_recover_leb(struct ubifs_info *c, int lnum,
  578. int offs, void *sbuf, int jhead)
  579. {
  580. int ret = 0, err, len = c->leb_size - offs, start = offs, min_io_unit;
  581. int grouped = jhead == -1 ? 0 : c->jheads[jhead].grouped;
  582. struct ubifs_scan_leb *sleb;
  583. void *buf = sbuf + offs;
  584. dbg_rcvry("%d:%d, jhead %d, grouped %d", lnum, offs, jhead, grouped);
  585. sleb = ubifs_start_scan(c, lnum, offs, sbuf);
  586. if (IS_ERR(sleb))
  587. return sleb;
  588. ubifs_assert(len >= 8);
  589. while (len >= 8) {
  590. dbg_scan("look at LEB %d:%d (%d bytes left)",
  591. lnum, offs, len);
  592. cond_resched();
  593. /*
  594. * Scan quietly until there is an error from which we cannot
  595. * recover
  596. */
  597. ret = ubifs_scan_a_node(c, buf, len, lnum, offs, 1);
  598. if (ret == SCANNED_A_NODE) {
  599. /* A valid node, and not a padding node */
  600. struct ubifs_ch *ch = buf;
  601. int node_len;
  602. err = ubifs_add_snod(c, sleb, buf, offs);
  603. if (err)
  604. goto error;
  605. node_len = ALIGN(le32_to_cpu(ch->len), 8);
  606. offs += node_len;
  607. buf += node_len;
  608. len -= node_len;
  609. } else if (ret > 0) {
  610. /* Padding bytes or a valid padding node */
  611. offs += ret;
  612. buf += ret;
  613. len -= ret;
  614. } else if (ret == SCANNED_EMPTY_SPACE ||
  615. ret == SCANNED_GARBAGE ||
  616. ret == SCANNED_A_BAD_PAD_NODE ||
  617. ret == SCANNED_A_CORRUPT_NODE) {
  618. dbg_rcvry("found corruption (%d) at %d:%d",
  619. ret, lnum, offs);
  620. break;
  621. } else {
  622. ubifs_err("unexpected return value %d", ret);
  623. err = -EINVAL;
  624. goto error;
  625. }
  626. }
  627. if (ret == SCANNED_GARBAGE || ret == SCANNED_A_BAD_PAD_NODE) {
  628. if (!is_last_write(c, buf, offs))
  629. goto corrupted_rescan;
  630. } else if (ret == SCANNED_A_CORRUPT_NODE) {
  631. if (!no_more_nodes(c, buf, len, lnum, offs))
  632. goto corrupted_rescan;
  633. } else if (!is_empty(buf, len)) {
  634. if (!is_last_write(c, buf, offs)) {
  635. int corruption = first_non_ff(buf, len);
  636. /*
  637. * See header comment for this file for more
  638. * explanations about the reasons we have this check.
  639. */
  640. ubifs_err("corrupt empty space LEB %d:%d, corruption starts at %d",
  641. lnum, offs, corruption);
  642. /* Make sure we dump interesting non-0xFF data */
  643. offs += corruption;
  644. buf += corruption;
  645. goto corrupted;
  646. }
  647. }
  648. min_io_unit = round_down(offs, c->min_io_size);
  649. if (grouped)
  650. /*
  651. * If nodes are grouped, always drop the incomplete group at
  652. * the end.
  653. */
  654. drop_last_group(sleb, &offs);
  655. if (jhead == GCHD) {
  656. /*
  657. * If this LEB belongs to the GC head then while we are in the
  658. * middle of the same min. I/O unit keep dropping nodes. So
  659. * basically, what we want is to make sure that the last min.
  660. * I/O unit where we saw the corruption is dropped completely
  661. * with all the uncorrupted nodes which may possibly sit there.
  662. *
  663. * In other words, let's name the min. I/O unit where the
  664. * corruption starts B, and the previous min. I/O unit A. The
  665. * below code tries to deal with a situation when half of B
  666. * contains valid nodes or the end of a valid node, and the
  667. * second half of B contains corrupted data or garbage. This
  668. * means that UBIFS had been writing to B just before the power
  669. * cut happened. I do not know how realistic is this scenario
  670. * that half of the min. I/O unit had been written successfully
  671. * and the other half not, but this is possible in our 'failure
  672. * mode emulation' infrastructure at least.
  673. *
  674. * So what is the problem, why we need to drop those nodes? Why
  675. * can't we just clean-up the second half of B by putting a
  676. * padding node there? We can, and this works fine with one
  677. * exception which was reproduced with power cut emulation
  678. * testing and happens extremely rarely.
  679. *
  680. * Imagine the file-system is full, we run GC which starts
  681. * moving valid nodes from LEB X to LEB Y (obviously, LEB Y is
  682. * the current GC head LEB). The @c->gc_lnum is -1, which means
  683. * that GC will retain LEB X and will try to continue. Imagine
  684. * that LEB X is currently the dirtiest LEB, and the amount of
  685. * used space in LEB Y is exactly the same as amount of free
  686. * space in LEB X.
  687. *
  688. * And a power cut happens when nodes are moved from LEB X to
  689. * LEB Y. We are here trying to recover LEB Y which is the GC
  690. * head LEB. We find the min. I/O unit B as described above.
  691. * Then we clean-up LEB Y by padding min. I/O unit. And later
  692. * 'ubifs_rcvry_gc_commit()' function fails, because it cannot
  693. * find a dirty LEB which could be GC'd into LEB Y! Even LEB X
  694. * does not match because the amount of valid nodes there does
  695. * not fit the free space in LEB Y any more! And this is
  696. * because of the padding node which we added to LEB Y. The
  697. * user-visible effect of this which I once observed and
  698. * analysed is that we cannot mount the file-system with
  699. * -ENOSPC error.
  700. *
  701. * So obviously, to make sure that situation does not happen we
  702. * should free min. I/O unit B in LEB Y completely and the last
  703. * used min. I/O unit in LEB Y should be A. This is basically
  704. * what the below code tries to do.
  705. */
  706. while (offs > min_io_unit)
  707. drop_last_node(sleb, &offs);
  708. }
  709. buf = sbuf + offs;
  710. len = c->leb_size - offs;
  711. clean_buf(c, &buf, lnum, &offs, &len);
  712. ubifs_end_scan(c, sleb, lnum, offs);
  713. err = fix_unclean_leb(c, sleb, start);
  714. if (err)
  715. goto error;
  716. return sleb;
  717. corrupted_rescan:
  718. /* Re-scan the corrupted data with verbose messages */
  719. ubifs_err("corruption %d", ret);
  720. ubifs_scan_a_node(c, buf, len, lnum, offs, 1);
  721. corrupted:
  722. ubifs_scanned_corruption(c, lnum, offs, buf);
  723. err = -EUCLEAN;
  724. error:
  725. ubifs_err("LEB %d scanning failed", lnum);
  726. ubifs_scan_destroy(sleb);
  727. return ERR_PTR(err);
  728. }
  729. /**
  730. * get_cs_sqnum - get commit start sequence number.
  731. * @c: UBIFS file-system description object
  732. * @lnum: LEB number of commit start node
  733. * @offs: offset of commit start node
  734. * @cs_sqnum: commit start sequence number is returned here
  735. *
  736. * This function returns %0 on success and a negative error code on failure.
  737. */
  738. static int get_cs_sqnum(struct ubifs_info *c, int lnum, int offs,
  739. unsigned long long *cs_sqnum)
  740. {
  741. struct ubifs_cs_node *cs_node = NULL;
  742. int err, ret;
  743. dbg_rcvry("at %d:%d", lnum, offs);
  744. cs_node = kmalloc(UBIFS_CS_NODE_SZ, GFP_KERNEL);
  745. if (!cs_node)
  746. return -ENOMEM;
  747. if (c->leb_size - offs < UBIFS_CS_NODE_SZ)
  748. goto out_err;
  749. err = ubifs_leb_read(c, lnum, (void *)cs_node, offs,
  750. UBIFS_CS_NODE_SZ, 0);
  751. if (err && err != -EBADMSG)
  752. goto out_free;
  753. ret = ubifs_scan_a_node(c, cs_node, UBIFS_CS_NODE_SZ, lnum, offs, 0);
  754. if (ret != SCANNED_A_NODE) {
  755. ubifs_err("Not a valid node");
  756. goto out_err;
  757. }
  758. if (cs_node->ch.node_type != UBIFS_CS_NODE) {
  759. ubifs_err("Node a CS node, type is %d", cs_node->ch.node_type);
  760. goto out_err;
  761. }
  762. if (le64_to_cpu(cs_node->cmt_no) != c->cmt_no) {
  763. ubifs_err("CS node cmt_no %llu != current cmt_no %llu",
  764. (unsigned long long)le64_to_cpu(cs_node->cmt_no),
  765. c->cmt_no);
  766. goto out_err;
  767. }
  768. *cs_sqnum = le64_to_cpu(cs_node->ch.sqnum);
  769. dbg_rcvry("commit start sqnum %llu", *cs_sqnum);
  770. kfree(cs_node);
  771. return 0;
  772. out_err:
  773. err = -EINVAL;
  774. out_free:
  775. ubifs_err("failed to get CS sqnum");
  776. kfree(cs_node);
  777. return err;
  778. }
  779. /**
  780. * ubifs_recover_log_leb - scan and recover a log LEB.
  781. * @c: UBIFS file-system description object
  782. * @lnum: LEB number
  783. * @offs: offset
  784. * @sbuf: LEB-sized buffer to use
  785. *
  786. * This function does a scan of a LEB, but caters for errors that might have
  787. * been caused by unclean reboots from which we are attempting to recover
  788. * (assume that only the last log LEB can be corrupted by an unclean reboot).
  789. *
  790. * This function returns %0 on success and a negative error code on failure.
  791. */
  792. struct ubifs_scan_leb *ubifs_recover_log_leb(struct ubifs_info *c, int lnum,
  793. int offs, void *sbuf)
  794. {
  795. struct ubifs_scan_leb *sleb;
  796. int next_lnum;
  797. dbg_rcvry("LEB %d", lnum);
  798. next_lnum = lnum + 1;
  799. if (next_lnum >= UBIFS_LOG_LNUM + c->log_lebs)
  800. next_lnum = UBIFS_LOG_LNUM;
  801. if (next_lnum != c->ltail_lnum) {
  802. /*
  803. * We can only recover at the end of the log, so check that the
  804. * next log LEB is empty or out of date.
  805. */
  806. sleb = ubifs_scan(c, next_lnum, 0, sbuf, 0);
  807. if (IS_ERR(sleb))
  808. return sleb;
  809. if (sleb->nodes_cnt) {
  810. struct ubifs_scan_node *snod;
  811. unsigned long long cs_sqnum = c->cs_sqnum;
  812. snod = list_entry(sleb->nodes.next,
  813. struct ubifs_scan_node, list);
  814. if (cs_sqnum == 0) {
  815. int err;
  816. err = get_cs_sqnum(c, lnum, offs, &cs_sqnum);
  817. if (err) {
  818. ubifs_scan_destroy(sleb);
  819. return ERR_PTR(err);
  820. }
  821. }
  822. if (snod->sqnum > cs_sqnum) {
  823. ubifs_err("unrecoverable log corruption in LEB %d",
  824. lnum);
  825. ubifs_scan_destroy(sleb);
  826. return ERR_PTR(-EUCLEAN);
  827. }
  828. }
  829. ubifs_scan_destroy(sleb);
  830. }
  831. return ubifs_recover_leb(c, lnum, offs, sbuf, -1);
  832. }
  833. /**
  834. * recover_head - recover a head.
  835. * @c: UBIFS file-system description object
  836. * @lnum: LEB number of head to recover
  837. * @offs: offset of head to recover
  838. * @sbuf: LEB-sized buffer to use
  839. *
  840. * This function ensures that there is no data on the flash at a head location.
  841. *
  842. * This function returns %0 on success and a negative error code on failure.
  843. */
  844. static int recover_head(struct ubifs_info *c, int lnum, int offs, void *sbuf)
  845. {
  846. int len = c->max_write_size, err;
  847. if (offs + len > c->leb_size)
  848. len = c->leb_size - offs;
  849. if (!len)
  850. return 0;
  851. /* Read at the head location and check it is empty flash */
  852. err = ubifs_leb_read(c, lnum, sbuf, offs, len, 1);
  853. if (err || !is_empty(sbuf, len)) {
  854. dbg_rcvry("cleaning head at %d:%d", lnum, offs);
  855. if (offs == 0)
  856. return ubifs_leb_unmap(c, lnum);
  857. err = ubifs_leb_read(c, lnum, sbuf, 0, offs, 1);
  858. if (err)
  859. return err;
  860. return ubifs_leb_change(c, lnum, sbuf, offs);
  861. }
  862. return 0;
  863. }
  864. /**
  865. * ubifs_recover_inl_heads - recover index and LPT heads.
  866. * @c: UBIFS file-system description object
  867. * @sbuf: LEB-sized buffer to use
  868. *
  869. * This function ensures that there is no data on the flash at the index and
  870. * LPT head locations.
  871. *
  872. * This deals with the recovery of a half-completed journal commit. UBIFS is
  873. * careful never to overwrite the last version of the index or the LPT. Because
  874. * the index and LPT are wandering trees, data from a half-completed commit will
  875. * not be referenced anywhere in UBIFS. The data will be either in LEBs that are
  876. * assumed to be empty and will be unmapped anyway before use, or in the index
  877. * and LPT heads.
  878. *
  879. * This function returns %0 on success and a negative error code on failure.
  880. */
  881. int ubifs_recover_inl_heads(struct ubifs_info *c, void *sbuf)
  882. {
  883. int err;
  884. ubifs_assert(!c->ro_mount || c->remounting_rw);
  885. dbg_rcvry("checking index head at %d:%d", c->ihead_lnum, c->ihead_offs);
  886. err = recover_head(c, c->ihead_lnum, c->ihead_offs, sbuf);
  887. if (err)
  888. return err;
  889. dbg_rcvry("checking LPT head at %d:%d", c->nhead_lnum, c->nhead_offs);
  890. err = recover_head(c, c->nhead_lnum, c->nhead_offs, sbuf);
  891. if (err)
  892. return err;
  893. return 0;
  894. }
  895. /**
  896. * clean_an_unclean_leb - read and write a LEB to remove corruption.
  897. * @c: UBIFS file-system description object
  898. * @ucleb: unclean LEB information
  899. * @sbuf: LEB-sized buffer to use
  900. *
  901. * This function reads a LEB up to a point pre-determined by the mount recovery,
  902. * checks the nodes, and writes the result back to the flash, thereby cleaning
  903. * off any following corruption, or non-fatal ECC errors.
  904. *
  905. * This function returns %0 on success and a negative error code on failure.
  906. */
  907. static int clean_an_unclean_leb(struct ubifs_info *c,
  908. struct ubifs_unclean_leb *ucleb, void *sbuf)
  909. {
  910. int err, lnum = ucleb->lnum, offs = 0, len = ucleb->endpt, quiet = 1;
  911. void *buf = sbuf;
  912. dbg_rcvry("LEB %d len %d", lnum, len);
  913. if (len == 0) {
  914. /* Nothing to read, just unmap it */
  915. err = ubifs_leb_unmap(c, lnum);
  916. if (err)
  917. return err;
  918. return 0;
  919. }
  920. err = ubifs_leb_read(c, lnum, buf, offs, len, 0);
  921. if (err && err != -EBADMSG)
  922. return err;
  923. while (len >= 8) {
  924. int ret;
  925. cond_resched();
  926. /* Scan quietly until there is an error */
  927. ret = ubifs_scan_a_node(c, buf, len, lnum, offs, quiet);
  928. if (ret == SCANNED_A_NODE) {
  929. /* A valid node, and not a padding node */
  930. struct ubifs_ch *ch = buf;
  931. int node_len;
  932. node_len = ALIGN(le32_to_cpu(ch->len), 8);
  933. offs += node_len;
  934. buf += node_len;
  935. len -= node_len;
  936. continue;
  937. }
  938. if (ret > 0) {
  939. /* Padding bytes or a valid padding node */
  940. offs += ret;
  941. buf += ret;
  942. len -= ret;
  943. continue;
  944. }
  945. if (ret == SCANNED_EMPTY_SPACE) {
  946. ubifs_err("unexpected empty space at %d:%d",
  947. lnum, offs);
  948. return -EUCLEAN;
  949. }
  950. if (quiet) {
  951. /* Redo the last scan but noisily */
  952. quiet = 0;
  953. continue;
  954. }
  955. ubifs_scanned_corruption(c, lnum, offs, buf);
  956. return -EUCLEAN;
  957. }
  958. /* Pad to min_io_size */
  959. len = ALIGN(ucleb->endpt, c->min_io_size);
  960. if (len > ucleb->endpt) {
  961. int pad_len = len - ALIGN(ucleb->endpt, 8);
  962. if (pad_len > 0) {
  963. buf = c->sbuf + len - pad_len;
  964. ubifs_pad(c, buf, pad_len);
  965. }
  966. }
  967. /* Write back the LEB atomically */
  968. err = ubifs_leb_change(c, lnum, sbuf, len);
  969. if (err)
  970. return err;
  971. dbg_rcvry("cleaned LEB %d", lnum);
  972. return 0;
  973. }
  974. /**
  975. * ubifs_clean_lebs - clean LEBs recovered during read-only mount.
  976. * @c: UBIFS file-system description object
  977. * @sbuf: LEB-sized buffer to use
  978. *
  979. * This function cleans a LEB identified during recovery that needs to be
  980. * written but was not because UBIFS was mounted read-only. This happens when
  981. * remounting to read-write mode.
  982. *
  983. * This function returns %0 on success and a negative error code on failure.
  984. */
  985. int ubifs_clean_lebs(struct ubifs_info *c, void *sbuf)
  986. {
  987. dbg_rcvry("recovery");
  988. while (!list_empty(&c->unclean_leb_list)) {
  989. struct ubifs_unclean_leb *ucleb;
  990. int err;
  991. ucleb = list_entry(c->unclean_leb_list.next,
  992. struct ubifs_unclean_leb, list);
  993. err = clean_an_unclean_leb(c, ucleb, sbuf);
  994. if (err)
  995. return err;
  996. list_del(&ucleb->list);
  997. kfree(ucleb);
  998. }
  999. return 0;
  1000. }
  1001. #ifndef __UBOOT__
  1002. /**
  1003. * grab_empty_leb - grab an empty LEB to use as GC LEB and run commit.
  1004. * @c: UBIFS file-system description object
  1005. *
  1006. * This is a helper function for 'ubifs_rcvry_gc_commit()' which grabs an empty
  1007. * LEB to be used as GC LEB (@c->gc_lnum), and then runs the commit. Returns
  1008. * zero in case of success and a negative error code in case of failure.
  1009. */
  1010. static int grab_empty_leb(struct ubifs_info *c)
  1011. {
  1012. int lnum, err;
  1013. /*
  1014. * Note, it is very important to first search for an empty LEB and then
  1015. * run the commit, not vice-versa. The reason is that there might be
  1016. * only one empty LEB at the moment, the one which has been the
  1017. * @c->gc_lnum just before the power cut happened. During the regular
  1018. * UBIFS operation (not now) @c->gc_lnum is marked as "taken", so no
  1019. * one but GC can grab it. But at this moment this single empty LEB is
  1020. * not marked as taken, so if we run commit - what happens? Right, the
  1021. * commit will grab it and write the index there. Remember that the
  1022. * index always expands as long as there is free space, and it only
  1023. * starts consolidating when we run out of space.
  1024. *
  1025. * IOW, if we run commit now, we might not be able to find a free LEB
  1026. * after this.
  1027. */
  1028. lnum = ubifs_find_free_leb_for_idx(c);
  1029. if (lnum < 0) {
  1030. ubifs_err("could not find an empty LEB");
  1031. ubifs_dump_lprops(c);
  1032. ubifs_dump_budg(c, &c->bi);
  1033. return lnum;
  1034. }
  1035. /* Reset the index flag */
  1036. err = ubifs_change_one_lp(c, lnum, LPROPS_NC, LPROPS_NC, 0,
  1037. LPROPS_INDEX, 0);
  1038. if (err)
  1039. return err;
  1040. c->gc_lnum = lnum;
  1041. dbg_rcvry("found empty LEB %d, run commit", lnum);
  1042. return ubifs_run_commit(c);
  1043. }
  1044. /**
  1045. * ubifs_rcvry_gc_commit - recover the GC LEB number and run the commit.
  1046. * @c: UBIFS file-system description object
  1047. *
  1048. * Out-of-place garbage collection requires always one empty LEB with which to
  1049. * start garbage collection. The LEB number is recorded in c->gc_lnum and is
  1050. * written to the master node on unmounting. In the case of an unclean unmount
  1051. * the value of gc_lnum recorded in the master node is out of date and cannot
  1052. * be used. Instead, recovery must allocate an empty LEB for this purpose.
  1053. * However, there may not be enough empty space, in which case it must be
  1054. * possible to GC the dirtiest LEB into the GC head LEB.
  1055. *
  1056. * This function also runs the commit which causes the TNC updates from
  1057. * size-recovery and orphans to be written to the flash. That is important to
  1058. * ensure correct replay order for subsequent mounts.
  1059. *
  1060. * This function returns %0 on success and a negative error code on failure.
  1061. */
  1062. int ubifs_rcvry_gc_commit(struct ubifs_info *c)
  1063. {
  1064. struct ubifs_wbuf *wbuf = &c->jheads[GCHD].wbuf;
  1065. struct ubifs_lprops lp;
  1066. int err;
  1067. dbg_rcvry("GC head LEB %d, offs %d", wbuf->lnum, wbuf->offs);
  1068. c->gc_lnum = -1;
  1069. if (wbuf->lnum == -1 || wbuf->offs == c->leb_size)
  1070. return grab_empty_leb(c);
  1071. err = ubifs_find_dirty_leb(c, &lp, wbuf->offs, 2);
  1072. if (err) {
  1073. if (err != -ENOSPC)
  1074. return err;
  1075. dbg_rcvry("could not find a dirty LEB");
  1076. return grab_empty_leb(c);
  1077. }
  1078. ubifs_assert(!(lp.flags & LPROPS_INDEX));
  1079. ubifs_assert(lp.free + lp.dirty >= wbuf->offs);
  1080. /*
  1081. * We run the commit before garbage collection otherwise subsequent
  1082. * mounts will see the GC and orphan deletion in a different order.
  1083. */
  1084. dbg_rcvry("committing");
  1085. err = ubifs_run_commit(c);
  1086. if (err)
  1087. return err;
  1088. dbg_rcvry("GC'ing LEB %d", lp.lnum);
  1089. mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
  1090. err = ubifs_garbage_collect_leb(c, &lp);
  1091. if (err >= 0) {
  1092. int err2 = ubifs_wbuf_sync_nolock(wbuf);
  1093. if (err2)
  1094. err = err2;
  1095. }
  1096. mutex_unlock(&wbuf->io_mutex);
  1097. if (err < 0) {
  1098. ubifs_err("GC failed, error %d", err);
  1099. if (err == -EAGAIN)
  1100. err = -EINVAL;
  1101. return err;
  1102. }
  1103. ubifs_assert(err == LEB_RETAINED);
  1104. if (err != LEB_RETAINED)
  1105. return -EINVAL;
  1106. err = ubifs_leb_unmap(c, c->gc_lnum);
  1107. if (err)
  1108. return err;
  1109. dbg_rcvry("allocated LEB %d for GC", lp.lnum);
  1110. return 0;
  1111. }
  1112. #else
  1113. int ubifs_rcvry_gc_commit(struct ubifs_info *c)
  1114. {
  1115. return 0;
  1116. }
  1117. #endif
  1118. /**
  1119. * struct size_entry - inode size information for recovery.
  1120. * @rb: link in the RB-tree of sizes
  1121. * @inum: inode number
  1122. * @i_size: size on inode
  1123. * @d_size: maximum size based on data nodes
  1124. * @exists: indicates whether the inode exists
  1125. * @inode: inode if pinned in memory awaiting rw mode to fix it
  1126. */
  1127. struct size_entry {
  1128. struct rb_node rb;
  1129. ino_t inum;
  1130. loff_t i_size;
  1131. loff_t d_size;
  1132. int exists;
  1133. struct inode *inode;
  1134. };
  1135. /**
  1136. * add_ino - add an entry to the size tree.
  1137. * @c: UBIFS file-system description object
  1138. * @inum: inode number
  1139. * @i_size: size on inode
  1140. * @d_size: maximum size based on data nodes
  1141. * @exists: indicates whether the inode exists
  1142. */
  1143. static int add_ino(struct ubifs_info *c, ino_t inum, loff_t i_size,
  1144. loff_t d_size, int exists)
  1145. {
  1146. struct rb_node **p = &c->size_tree.rb_node, *parent = NULL;
  1147. struct size_entry *e;
  1148. while (*p) {
  1149. parent = *p;
  1150. e = rb_entry(parent, struct size_entry, rb);
  1151. if (inum < e->inum)
  1152. p = &(*p)->rb_left;
  1153. else
  1154. p = &(*p)->rb_right;
  1155. }
  1156. e = kzalloc(sizeof(struct size_entry), GFP_KERNEL);
  1157. if (!e)
  1158. return -ENOMEM;
  1159. e->inum = inum;
  1160. e->i_size = i_size;
  1161. e->d_size = d_size;
  1162. e->exists = exists;
  1163. rb_link_node(&e->rb, parent, p);
  1164. rb_insert_color(&e->rb, &c->size_tree);
  1165. return 0;
  1166. }
  1167. /**
  1168. * find_ino - find an entry on the size tree.
  1169. * @c: UBIFS file-system description object
  1170. * @inum: inode number
  1171. */
  1172. static struct size_entry *find_ino(struct ubifs_info *c, ino_t inum)
  1173. {
  1174. struct rb_node *p = c->size_tree.rb_node;
  1175. struct size_entry *e;
  1176. while (p) {
  1177. e = rb_entry(p, struct size_entry, rb);
  1178. if (inum < e->inum)
  1179. p = p->rb_left;
  1180. else if (inum > e->inum)
  1181. p = p->rb_right;
  1182. else
  1183. return e;
  1184. }
  1185. return NULL;
  1186. }
  1187. /**
  1188. * remove_ino - remove an entry from the size tree.
  1189. * @c: UBIFS file-system description object
  1190. * @inum: inode number
  1191. */
  1192. static void remove_ino(struct ubifs_info *c, ino_t inum)
  1193. {
  1194. struct size_entry *e = find_ino(c, inum);
  1195. if (!e)
  1196. return;
  1197. rb_erase(&e->rb, &c->size_tree);
  1198. kfree(e);
  1199. }
  1200. /**
  1201. * ubifs_destroy_size_tree - free resources related to the size tree.
  1202. * @c: UBIFS file-system description object
  1203. */
  1204. void ubifs_destroy_size_tree(struct ubifs_info *c)
  1205. {
  1206. struct size_entry *e, *n;
  1207. rbtree_postorder_for_each_entry_safe(e, n, &c->size_tree, rb) {
  1208. if (e->inode)
  1209. iput(e->inode);
  1210. kfree(e);
  1211. }
  1212. c->size_tree = RB_ROOT;
  1213. }
  1214. /**
  1215. * ubifs_recover_size_accum - accumulate inode sizes for recovery.
  1216. * @c: UBIFS file-system description object
  1217. * @key: node key
  1218. * @deletion: node is for a deletion
  1219. * @new_size: inode size
  1220. *
  1221. * This function has two purposes:
  1222. * 1) to ensure there are no data nodes that fall outside the inode size
  1223. * 2) to ensure there are no data nodes for inodes that do not exist
  1224. * To accomplish those purposes, a rb-tree is constructed containing an entry
  1225. * for each inode number in the journal that has not been deleted, and recording
  1226. * the size from the inode node, the maximum size of any data node (also altered
  1227. * by truncations) and a flag indicating a inode number for which no inode node
  1228. * was present in the journal.
  1229. *
  1230. * Note that there is still the possibility that there are data nodes that have
  1231. * been committed that are beyond the inode size, however the only way to find
  1232. * them would be to scan the entire index. Alternatively, some provision could
  1233. * be made to record the size of inodes at the start of commit, which would seem
  1234. * very cumbersome for a scenario that is quite unlikely and the only negative
  1235. * consequence of which is wasted space.
  1236. *
  1237. * This functions returns %0 on success and a negative error code on failure.
  1238. */
  1239. int ubifs_recover_size_accum(struct ubifs_info *c, union ubifs_key *key,
  1240. int deletion, loff_t new_size)
  1241. {
  1242. ino_t inum = key_inum(c, key);
  1243. struct size_entry *e;
  1244. int err;
  1245. switch (key_type(c, key)) {
  1246. case UBIFS_INO_KEY:
  1247. if (deletion)
  1248. remove_ino(c, inum);
  1249. else {
  1250. e = find_ino(c, inum);
  1251. if (e) {
  1252. e->i_size = new_size;
  1253. e->exists = 1;
  1254. } else {
  1255. err = add_ino(c, inum, new_size, 0, 1);
  1256. if (err)
  1257. return err;
  1258. }
  1259. }
  1260. break;
  1261. case UBIFS_DATA_KEY:
  1262. e = find_ino(c, inum);
  1263. if (e) {
  1264. if (new_size > e->d_size)
  1265. e->d_size = new_size;
  1266. } else {
  1267. err = add_ino(c, inum, 0, new_size, 0);
  1268. if (err)
  1269. return err;
  1270. }
  1271. break;
  1272. case UBIFS_TRUN_KEY:
  1273. e = find_ino(c, inum);
  1274. if (e)
  1275. e->d_size = new_size;
  1276. break;
  1277. }
  1278. return 0;
  1279. }
  1280. #ifndef __UBOOT__
  1281. /**
  1282. * fix_size_in_place - fix inode size in place on flash.
  1283. * @c: UBIFS file-system description object
  1284. * @e: inode size information for recovery
  1285. */
  1286. static int fix_size_in_place(struct ubifs_info *c, struct size_entry *e)
  1287. {
  1288. struct ubifs_ino_node *ino = c->sbuf;
  1289. unsigned char *p;
  1290. union ubifs_key key;
  1291. int err, lnum, offs, len;
  1292. loff_t i_size;
  1293. uint32_t crc;
  1294. /* Locate the inode node LEB number and offset */
  1295. ino_key_init(c, &key, e->inum);
  1296. err = ubifs_tnc_locate(c, &key, ino, &lnum, &offs);
  1297. if (err)
  1298. goto out;
  1299. /*
  1300. * If the size recorded on the inode node is greater than the size that
  1301. * was calculated from nodes in the journal then don't change the inode.
  1302. */
  1303. i_size = le64_to_cpu(ino->size);
  1304. if (i_size >= e->d_size)
  1305. return 0;
  1306. /* Read the LEB */
  1307. err = ubifs_leb_read(c, lnum, c->sbuf, 0, c->leb_size, 1);
  1308. if (err)
  1309. goto out;
  1310. /* Change the size field and recalculate the CRC */
  1311. ino = c->sbuf + offs;
  1312. ino->size = cpu_to_le64(e->d_size);
  1313. len = le32_to_cpu(ino->ch.len);
  1314. crc = crc32(UBIFS_CRC32_INIT, (void *)ino + 8, len - 8);
  1315. ino->ch.crc = cpu_to_le32(crc);
  1316. /* Work out where data in the LEB ends and free space begins */
  1317. p = c->sbuf;
  1318. len = c->leb_size - 1;
  1319. while (p[len] == 0xff)
  1320. len -= 1;
  1321. len = ALIGN(len + 1, c->min_io_size);
  1322. /* Atomically write the fixed LEB back again */
  1323. err = ubifs_leb_change(c, lnum, c->sbuf, len);
  1324. if (err)
  1325. goto out;
  1326. dbg_rcvry("inode %lu at %d:%d size %lld -> %lld",
  1327. (unsigned long)e->inum, lnum, offs, i_size, e->d_size);
  1328. return 0;
  1329. out:
  1330. ubifs_warn("inode %lu failed to fix size %lld -> %lld error %d",
  1331. (unsigned long)e->inum, e->i_size, e->d_size, err);
  1332. return err;
  1333. }
  1334. #endif
  1335. /**
  1336. * ubifs_recover_size - recover inode size.
  1337. * @c: UBIFS file-system description object
  1338. *
  1339. * This function attempts to fix inode size discrepancies identified by the
  1340. * 'ubifs_recover_size_accum()' function.
  1341. *
  1342. * This functions returns %0 on success and a negative error code on failure.
  1343. */
  1344. int ubifs_recover_size(struct ubifs_info *c)
  1345. {
  1346. struct rb_node *this = rb_first(&c->size_tree);
  1347. while (this) {
  1348. struct size_entry *e;
  1349. int err;
  1350. e = rb_entry(this, struct size_entry, rb);
  1351. if (!e->exists) {
  1352. union ubifs_key key;
  1353. ino_key_init(c, &key, e->inum);
  1354. err = ubifs_tnc_lookup(c, &key, c->sbuf);
  1355. if (err && err != -ENOENT)
  1356. return err;
  1357. if (err == -ENOENT) {
  1358. /* Remove data nodes that have no inode */
  1359. dbg_rcvry("removing ino %lu",
  1360. (unsigned long)e->inum);
  1361. err = ubifs_tnc_remove_ino(c, e->inum);
  1362. if (err)
  1363. return err;
  1364. } else {
  1365. struct ubifs_ino_node *ino = c->sbuf;
  1366. e->exists = 1;
  1367. e->i_size = le64_to_cpu(ino->size);
  1368. }
  1369. }
  1370. if (e->exists && e->i_size < e->d_size) {
  1371. if (c->ro_mount) {
  1372. /* Fix the inode size and pin it in memory */
  1373. struct inode *inode;
  1374. struct ubifs_inode *ui;
  1375. ubifs_assert(!e->inode);
  1376. inode = ubifs_iget(c->vfs_sb, e->inum);
  1377. if (IS_ERR(inode))
  1378. return PTR_ERR(inode);
  1379. ui = ubifs_inode(inode);
  1380. if (inode->i_size < e->d_size) {
  1381. dbg_rcvry("ino %lu size %lld -> %lld",
  1382. (unsigned long)e->inum,
  1383. inode->i_size, e->d_size);
  1384. inode->i_size = e->d_size;
  1385. ui->ui_size = e->d_size;
  1386. ui->synced_i_size = e->d_size;
  1387. e->inode = inode;
  1388. this = rb_next(this);
  1389. continue;
  1390. }
  1391. iput(inode);
  1392. #ifndef __UBOOT__
  1393. } else {
  1394. /* Fix the size in place */
  1395. err = fix_size_in_place(c, e);
  1396. if (err)
  1397. return err;
  1398. if (e->inode)
  1399. iput(e->inode);
  1400. #endif
  1401. }
  1402. }
  1403. this = rb_next(this);
  1404. rb_erase(&e->rb, &c->size_tree);
  1405. kfree(e);
  1406. }
  1407. return 0;
  1408. }