raid56.c 66 KB

12345678910111213141516171819202122232425262728293031323334353637383940414243444546474849505152535455565758596061626364656667686970717273747576777879808182838485868788899091929394959697989910010110210310410510610710810911011111211311411511611711811912012112212312412512612712812913013113213313413513613713813914014114214314414514614714814915015115215315415515615715815916016116216316416516616716816917017117217317417517617717817918018118218318418518618718818919019119219319419519619719819920020120220320420520620720820921021121221321421521621721821922022122222322422522622722822923023123223323423523623723823924024124224324424524624724824925025125225325425525625725825926026126226326426526626726826927027127227327427527627727827928028128228328428528628728828929029129229329429529629729829930030130230330430530630730830931031131231331431531631731831932032132232332432532632732832933033133233333433533633733833934034134234334434534634734834935035135235335435535635735835936036136236336436536636736836937037137237337437537637737837938038138238338438538638738838939039139239339439539639739839940040140240340440540640740840941041141241341441541641741841942042142242342442542642742842943043143243343443543643743843944044144244344444544644744844945045145245345445545645745845946046146246346446546646746846947047147247347447547647747847948048148248348448548648748848949049149249349449549649749849950050150250350450550650750850951051151251351451551651751851952052152252352452552652752852953053153253353453553653753853954054154254354454554654754854955055155255355455555655755855956056156256356456556656756856957057157257357457557657757857958058158258358458558658758858959059159259359459559659759859960060160260360460560660760860961061161261361461561661761861962062162262362462562662762862963063163263363463563663763863964064164264364464564664764864965065165265365465565665765865966066166266366466566666766866967067167267367467567667767867968068168268368468568668768868969069169269369469569669769869970070170270370470570670770870971071171271371471571671771871972072172272372472572672772872973073173273373473573673773873974074174274374474574674774874975075175275375475575675775875976076176276376476576676776876977077177277377477577677777877978078178278378478578678778878979079179279379479579679779879980080180280380480580680780880981081181281381481581681781881982082182282382482582682782882983083183283383483583683783883984084184284384484584684784884985085185285385485585685785885986086186286386486586686786886987087187287387487587687787887988088188288388488588688788888989089189289389489589689789889990090190290390490590690790890991091191291391491591691791891992092192292392492592692792892993093193293393493593693793893994094194294394494594694794894995095195295395495595695795895996096196296396496596696796896997097197297397497597697797897998098198298398498598698798898999099199299399499599699799899910001001100210031004100510061007100810091010101110121013101410151016101710181019102010211022102310241025102610271028102910301031103210331034103510361037103810391040104110421043104410451046104710481049105010511052105310541055105610571058105910601061106210631064106510661067106810691070107110721073107410751076107710781079108010811082108310841085108610871088108910901091109210931094109510961097109810991100110111021103110411051106110711081109111011111112111311141115111611171118111911201121112211231124112511261127112811291130113111321133113411351136113711381139114011411142114311441145114611471148114911501151115211531154115511561157115811591160116111621163116411651166116711681169117011711172117311741175117611771178117911801181118211831184118511861187118811891190119111921193119411951196119711981199120012011202120312041205120612071208120912101211121212131214121512161217121812191220122112221223122412251226122712281229123012311232123312341235123612371238123912401241124212431244124512461247124812491250125112521253125412551256125712581259126012611262126312641265126612671268126912701271127212731274127512761277127812791280128112821283128412851286128712881289129012911292129312941295129612971298129913001301130213031304130513061307130813091310131113121313131413151316131713181319132013211322132313241325132613271328132913301331133213331334133513361337133813391340134113421343134413451346134713481349135013511352135313541355135613571358135913601361136213631364136513661367136813691370137113721373137413751376137713781379138013811382138313841385138613871388138913901391139213931394139513961397139813991400140114021403140414051406140714081409141014111412141314141415141614171418141914201421142214231424142514261427142814291430143114321433143414351436143714381439144014411442144314441445144614471448144914501451145214531454145514561457145814591460146114621463146414651466146714681469147014711472147314741475147614771478147914801481148214831484148514861487148814891490149114921493149414951496149714981499150015011502150315041505150615071508150915101511151215131514151515161517151815191520152115221523152415251526152715281529153015311532153315341535153615371538153915401541154215431544154515461547154815491550155115521553155415551556155715581559156015611562156315641565156615671568156915701571157215731574157515761577157815791580158115821583158415851586158715881589159015911592159315941595159615971598159916001601160216031604160516061607160816091610161116121613161416151616161716181619162016211622162316241625162616271628162916301631163216331634163516361637163816391640164116421643164416451646164716481649165016511652165316541655165616571658165916601661166216631664166516661667166816691670167116721673167416751676167716781679168016811682168316841685168616871688168916901691169216931694169516961697169816991700170117021703170417051706170717081709171017111712171317141715171617171718171917201721172217231724172517261727172817291730173117321733173417351736173717381739174017411742174317441745174617471748174917501751175217531754175517561757175817591760176117621763176417651766176717681769177017711772177317741775177617771778177917801781178217831784178517861787178817891790179117921793179417951796179717981799180018011802180318041805180618071808180918101811181218131814181518161817181818191820182118221823182418251826182718281829183018311832183318341835183618371838183918401841184218431844184518461847184818491850185118521853185418551856185718581859186018611862186318641865186618671868186918701871187218731874187518761877187818791880188118821883188418851886188718881889189018911892189318941895189618971898189919001901190219031904190519061907190819091910191119121913191419151916191719181919192019211922192319241925192619271928192919301931193219331934193519361937193819391940194119421943194419451946194719481949195019511952195319541955195619571958195919601961196219631964196519661967196819691970197119721973197419751976197719781979198019811982198319841985198619871988198919901991199219931994199519961997199819992000200120022003200420052006200720082009201020112012201320142015201620172018201920202021202220232024202520262027202820292030203120322033203420352036203720382039204020412042204320442045204620472048204920502051205220532054205520562057205820592060206120622063206420652066206720682069207020712072207320742075207620772078207920802081208220832084208520862087208820892090209120922093209420952096209720982099210021012102210321042105210621072108210921102111211221132114211521162117211821192120212121222123212421252126212721282129213021312132213321342135213621372138213921402141214221432144214521462147214821492150215121522153215421552156215721582159216021612162216321642165216621672168216921702171217221732174217521762177217821792180218121822183218421852186218721882189219021912192219321942195219621972198219922002201220222032204220522062207220822092210221122122213221422152216221722182219222022212222222322242225222622272228222922302231223222332234223522362237223822392240224122422243224422452246224722482249225022512252225322542255225622572258225922602261226222632264226522662267226822692270227122722273227422752276227722782279228022812282228322842285228622872288228922902291229222932294229522962297229822992300230123022303230423052306230723082309231023112312231323142315231623172318231923202321232223232324232523262327232823292330233123322333233423352336233723382339234023412342234323442345234623472348234923502351235223532354235523562357235823592360236123622363236423652366236723682369237023712372237323742375237623772378237923802381238223832384238523862387238823892390239123922393239423952396239723982399240024012402240324042405240624072408240924102411241224132414241524162417241824192420242124222423242424252426242724282429243024312432243324342435243624372438243924402441244224432444244524462447244824492450245124522453245424552456245724582459246024612462246324642465246624672468246924702471247224732474247524762477247824792480248124822483248424852486248724882489249024912492249324942495249624972498249925002501250225032504250525062507250825092510251125122513251425152516251725182519252025212522252325242525252625272528252925302531253225332534253525362537253825392540254125422543254425452546254725482549255025512552255325542555255625572558255925602561256225632564256525662567256825692570257125722573257425752576257725782579258025812582258325842585258625872588258925902591259225932594259525962597259825992600260126022603260426052606260726082609261026112612261326142615261626172618261926202621262226232624262526262627262826292630263126322633263426352636263726382639264026412642264326442645264626472648264926502651265226532654265526562657265826592660266126622663266426652666266726682669267026712672267326742675267626772678267926802681268226832684268526862687268826892690269126922693269426952696269726982699270027012702270327042705270627072708270927102711271227132714
  1. /*
  2. * Copyright (C) 2012 Fusion-io All rights reserved.
  3. * Copyright (C) 2012 Intel Corp. All rights reserved.
  4. *
  5. * This program is free software; you can redistribute it and/or
  6. * modify it under the terms of the GNU General Public
  7. * License v2 as published by the Free Software Foundation.
  8. *
  9. * This program is distributed in the hope that it will be useful,
  10. * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11. * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  12. * General Public License for more details.
  13. *
  14. * You should have received a copy of the GNU General Public
  15. * License along with this program; if not, write to the
  16. * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  17. * Boston, MA 021110-1307, USA.
  18. */
  19. #include <linux/sched.h>
  20. #include <linux/wait.h>
  21. #include <linux/bio.h>
  22. #include <linux/slab.h>
  23. #include <linux/buffer_head.h>
  24. #include <linux/blkdev.h>
  25. #include <linux/random.h>
  26. #include <linux/iocontext.h>
  27. #include <linux/capability.h>
  28. #include <linux/ratelimit.h>
  29. #include <linux/kthread.h>
  30. #include <linux/raid/pq.h>
  31. #include <linux/hash.h>
  32. #include <linux/list_sort.h>
  33. #include <linux/raid/xor.h>
  34. #include <linux/mm.h>
  35. #include <asm/div64.h>
  36. #include "ctree.h"
  37. #include "extent_map.h"
  38. #include "disk-io.h"
  39. #include "transaction.h"
  40. #include "print-tree.h"
  41. #include "volumes.h"
  42. #include "raid56.h"
  43. #include "async-thread.h"
  44. #include "check-integrity.h"
  45. #include "rcu-string.h"
  46. /* set when additional merges to this rbio are not allowed */
  47. #define RBIO_RMW_LOCKED_BIT 1
  48. /*
  49. * set when this rbio is sitting in the hash, but it is just a cache
  50. * of past RMW
  51. */
  52. #define RBIO_CACHE_BIT 2
  53. /*
  54. * set when it is safe to trust the stripe_pages for caching
  55. */
  56. #define RBIO_CACHE_READY_BIT 3
  57. #define RBIO_CACHE_SIZE 1024
  58. enum btrfs_rbio_ops {
  59. BTRFS_RBIO_WRITE,
  60. BTRFS_RBIO_READ_REBUILD,
  61. BTRFS_RBIO_PARITY_SCRUB,
  62. BTRFS_RBIO_REBUILD_MISSING,
  63. };
  64. struct btrfs_raid_bio {
  65. struct btrfs_fs_info *fs_info;
  66. struct btrfs_bio *bbio;
  67. /* while we're doing rmw on a stripe
  68. * we put it into a hash table so we can
  69. * lock the stripe and merge more rbios
  70. * into it.
  71. */
  72. struct list_head hash_list;
  73. /*
  74. * LRU list for the stripe cache
  75. */
  76. struct list_head stripe_cache;
  77. /*
  78. * for scheduling work in the helper threads
  79. */
  80. struct btrfs_work work;
  81. /*
  82. * bio list and bio_list_lock are used
  83. * to add more bios into the stripe
  84. * in hopes of avoiding the full rmw
  85. */
  86. struct bio_list bio_list;
  87. spinlock_t bio_list_lock;
  88. /* also protected by the bio_list_lock, the
  89. * plug list is used by the plugging code
  90. * to collect partial bios while plugged. The
  91. * stripe locking code also uses it to hand off
  92. * the stripe lock to the next pending IO
  93. */
  94. struct list_head plug_list;
  95. /*
  96. * flags that tell us if it is safe to
  97. * merge with this bio
  98. */
  99. unsigned long flags;
  100. /* size of each individual stripe on disk */
  101. int stripe_len;
  102. /* number of data stripes (no p/q) */
  103. int nr_data;
  104. int real_stripes;
  105. int stripe_npages;
  106. /*
  107. * set if we're doing a parity rebuild
  108. * for a read from higher up, which is handled
  109. * differently from a parity rebuild as part of
  110. * rmw
  111. */
  112. enum btrfs_rbio_ops operation;
  113. /* first bad stripe */
  114. int faila;
  115. /* second bad stripe (for raid6 use) */
  116. int failb;
  117. int scrubp;
  118. /*
  119. * number of pages needed to represent the full
  120. * stripe
  121. */
  122. int nr_pages;
  123. /*
  124. * size of all the bios in the bio_list. This
  125. * helps us decide if the rbio maps to a full
  126. * stripe or not
  127. */
  128. int bio_list_bytes;
  129. int generic_bio_cnt;
  130. refcount_t refs;
  131. atomic_t stripes_pending;
  132. atomic_t error;
  133. /*
  134. * these are two arrays of pointers. We allocate the
  135. * rbio big enough to hold them both and setup their
  136. * locations when the rbio is allocated
  137. */
  138. /* pointers to pages that we allocated for
  139. * reading/writing stripes directly from the disk (including P/Q)
  140. */
  141. struct page **stripe_pages;
  142. /*
  143. * pointers to the pages in the bio_list. Stored
  144. * here for faster lookup
  145. */
  146. struct page **bio_pages;
  147. /*
  148. * bitmap to record which horizontal stripe has data
  149. */
  150. unsigned long *dbitmap;
  151. };
  152. static int __raid56_parity_recover(struct btrfs_raid_bio *rbio);
  153. static noinline void finish_rmw(struct btrfs_raid_bio *rbio);
  154. static void rmw_work(struct btrfs_work *work);
  155. static void read_rebuild_work(struct btrfs_work *work);
  156. static void async_rmw_stripe(struct btrfs_raid_bio *rbio);
  157. static void async_read_rebuild(struct btrfs_raid_bio *rbio);
  158. static int fail_bio_stripe(struct btrfs_raid_bio *rbio, struct bio *bio);
  159. static int fail_rbio_index(struct btrfs_raid_bio *rbio, int failed);
  160. static void __free_raid_bio(struct btrfs_raid_bio *rbio);
  161. static void index_rbio_pages(struct btrfs_raid_bio *rbio);
  162. static int alloc_rbio_pages(struct btrfs_raid_bio *rbio);
  163. static noinline void finish_parity_scrub(struct btrfs_raid_bio *rbio,
  164. int need_check);
  165. static void async_scrub_parity(struct btrfs_raid_bio *rbio);
  166. /*
  167. * the stripe hash table is used for locking, and to collect
  168. * bios in hopes of making a full stripe
  169. */
  170. int btrfs_alloc_stripe_hash_table(struct btrfs_fs_info *info)
  171. {
  172. struct btrfs_stripe_hash_table *table;
  173. struct btrfs_stripe_hash_table *x;
  174. struct btrfs_stripe_hash *cur;
  175. struct btrfs_stripe_hash *h;
  176. int num_entries = 1 << BTRFS_STRIPE_HASH_TABLE_BITS;
  177. int i;
  178. int table_size;
  179. if (info->stripe_hash_table)
  180. return 0;
  181. /*
  182. * The table is large, starting with order 4 and can go as high as
  183. * order 7 in case lock debugging is turned on.
  184. *
  185. * Try harder to allocate and fallback to vmalloc to lower the chance
  186. * of a failing mount.
  187. */
  188. table_size = sizeof(*table) + sizeof(*h) * num_entries;
  189. table = kvzalloc(table_size, GFP_KERNEL);
  190. if (!table)
  191. return -ENOMEM;
  192. spin_lock_init(&table->cache_lock);
  193. INIT_LIST_HEAD(&table->stripe_cache);
  194. h = table->table;
  195. for (i = 0; i < num_entries; i++) {
  196. cur = h + i;
  197. INIT_LIST_HEAD(&cur->hash_list);
  198. spin_lock_init(&cur->lock);
  199. init_waitqueue_head(&cur->wait);
  200. }
  201. x = cmpxchg(&info->stripe_hash_table, NULL, table);
  202. if (x)
  203. kvfree(x);
  204. return 0;
  205. }
  206. /*
  207. * caching an rbio means to copy anything from the
  208. * bio_pages array into the stripe_pages array. We
  209. * use the page uptodate bit in the stripe cache array
  210. * to indicate if it has valid data
  211. *
  212. * once the caching is done, we set the cache ready
  213. * bit.
  214. */
  215. static void cache_rbio_pages(struct btrfs_raid_bio *rbio)
  216. {
  217. int i;
  218. char *s;
  219. char *d;
  220. int ret;
  221. ret = alloc_rbio_pages(rbio);
  222. if (ret)
  223. return;
  224. for (i = 0; i < rbio->nr_pages; i++) {
  225. if (!rbio->bio_pages[i])
  226. continue;
  227. s = kmap(rbio->bio_pages[i]);
  228. d = kmap(rbio->stripe_pages[i]);
  229. memcpy(d, s, PAGE_SIZE);
  230. kunmap(rbio->bio_pages[i]);
  231. kunmap(rbio->stripe_pages[i]);
  232. SetPageUptodate(rbio->stripe_pages[i]);
  233. }
  234. set_bit(RBIO_CACHE_READY_BIT, &rbio->flags);
  235. }
  236. /*
  237. * we hash on the first logical address of the stripe
  238. */
  239. static int rbio_bucket(struct btrfs_raid_bio *rbio)
  240. {
  241. u64 num = rbio->bbio->raid_map[0];
  242. /*
  243. * we shift down quite a bit. We're using byte
  244. * addressing, and most of the lower bits are zeros.
  245. * This tends to upset hash_64, and it consistently
  246. * returns just one or two different values.
  247. *
  248. * shifting off the lower bits fixes things.
  249. */
  250. return hash_64(num >> 16, BTRFS_STRIPE_HASH_TABLE_BITS);
  251. }
  252. /*
  253. * stealing an rbio means taking all the uptodate pages from the stripe
  254. * array in the source rbio and putting them into the destination rbio
  255. */
  256. static void steal_rbio(struct btrfs_raid_bio *src, struct btrfs_raid_bio *dest)
  257. {
  258. int i;
  259. struct page *s;
  260. struct page *d;
  261. if (!test_bit(RBIO_CACHE_READY_BIT, &src->flags))
  262. return;
  263. for (i = 0; i < dest->nr_pages; i++) {
  264. s = src->stripe_pages[i];
  265. if (!s || !PageUptodate(s)) {
  266. continue;
  267. }
  268. d = dest->stripe_pages[i];
  269. if (d)
  270. __free_page(d);
  271. dest->stripe_pages[i] = s;
  272. src->stripe_pages[i] = NULL;
  273. }
  274. }
  275. /*
  276. * merging means we take the bio_list from the victim and
  277. * splice it into the destination. The victim should
  278. * be discarded afterwards.
  279. *
  280. * must be called with dest->rbio_list_lock held
  281. */
  282. static void merge_rbio(struct btrfs_raid_bio *dest,
  283. struct btrfs_raid_bio *victim)
  284. {
  285. bio_list_merge(&dest->bio_list, &victim->bio_list);
  286. dest->bio_list_bytes += victim->bio_list_bytes;
  287. dest->generic_bio_cnt += victim->generic_bio_cnt;
  288. bio_list_init(&victim->bio_list);
  289. }
  290. /*
  291. * used to prune items that are in the cache. The caller
  292. * must hold the hash table lock.
  293. */
  294. static void __remove_rbio_from_cache(struct btrfs_raid_bio *rbio)
  295. {
  296. int bucket = rbio_bucket(rbio);
  297. struct btrfs_stripe_hash_table *table;
  298. struct btrfs_stripe_hash *h;
  299. int freeit = 0;
  300. /*
  301. * check the bit again under the hash table lock.
  302. */
  303. if (!test_bit(RBIO_CACHE_BIT, &rbio->flags))
  304. return;
  305. table = rbio->fs_info->stripe_hash_table;
  306. h = table->table + bucket;
  307. /* hold the lock for the bucket because we may be
  308. * removing it from the hash table
  309. */
  310. spin_lock(&h->lock);
  311. /*
  312. * hold the lock for the bio list because we need
  313. * to make sure the bio list is empty
  314. */
  315. spin_lock(&rbio->bio_list_lock);
  316. if (test_and_clear_bit(RBIO_CACHE_BIT, &rbio->flags)) {
  317. list_del_init(&rbio->stripe_cache);
  318. table->cache_size -= 1;
  319. freeit = 1;
  320. /* if the bio list isn't empty, this rbio is
  321. * still involved in an IO. We take it out
  322. * of the cache list, and drop the ref that
  323. * was held for the list.
  324. *
  325. * If the bio_list was empty, we also remove
  326. * the rbio from the hash_table, and drop
  327. * the corresponding ref
  328. */
  329. if (bio_list_empty(&rbio->bio_list)) {
  330. if (!list_empty(&rbio->hash_list)) {
  331. list_del_init(&rbio->hash_list);
  332. refcount_dec(&rbio->refs);
  333. BUG_ON(!list_empty(&rbio->plug_list));
  334. }
  335. }
  336. }
  337. spin_unlock(&rbio->bio_list_lock);
  338. spin_unlock(&h->lock);
  339. if (freeit)
  340. __free_raid_bio(rbio);
  341. }
  342. /*
  343. * prune a given rbio from the cache
  344. */
  345. static void remove_rbio_from_cache(struct btrfs_raid_bio *rbio)
  346. {
  347. struct btrfs_stripe_hash_table *table;
  348. unsigned long flags;
  349. if (!test_bit(RBIO_CACHE_BIT, &rbio->flags))
  350. return;
  351. table = rbio->fs_info->stripe_hash_table;
  352. spin_lock_irqsave(&table->cache_lock, flags);
  353. __remove_rbio_from_cache(rbio);
  354. spin_unlock_irqrestore(&table->cache_lock, flags);
  355. }
  356. /*
  357. * remove everything in the cache
  358. */
  359. static void btrfs_clear_rbio_cache(struct btrfs_fs_info *info)
  360. {
  361. struct btrfs_stripe_hash_table *table;
  362. unsigned long flags;
  363. struct btrfs_raid_bio *rbio;
  364. table = info->stripe_hash_table;
  365. spin_lock_irqsave(&table->cache_lock, flags);
  366. while (!list_empty(&table->stripe_cache)) {
  367. rbio = list_entry(table->stripe_cache.next,
  368. struct btrfs_raid_bio,
  369. stripe_cache);
  370. __remove_rbio_from_cache(rbio);
  371. }
  372. spin_unlock_irqrestore(&table->cache_lock, flags);
  373. }
  374. /*
  375. * remove all cached entries and free the hash table
  376. * used by unmount
  377. */
  378. void btrfs_free_stripe_hash_table(struct btrfs_fs_info *info)
  379. {
  380. if (!info->stripe_hash_table)
  381. return;
  382. btrfs_clear_rbio_cache(info);
  383. kvfree(info->stripe_hash_table);
  384. info->stripe_hash_table = NULL;
  385. }
  386. /*
  387. * insert an rbio into the stripe cache. It
  388. * must have already been prepared by calling
  389. * cache_rbio_pages
  390. *
  391. * If this rbio was already cached, it gets
  392. * moved to the front of the lru.
  393. *
  394. * If the size of the rbio cache is too big, we
  395. * prune an item.
  396. */
  397. static void cache_rbio(struct btrfs_raid_bio *rbio)
  398. {
  399. struct btrfs_stripe_hash_table *table;
  400. unsigned long flags;
  401. if (!test_bit(RBIO_CACHE_READY_BIT, &rbio->flags))
  402. return;
  403. table = rbio->fs_info->stripe_hash_table;
  404. spin_lock_irqsave(&table->cache_lock, flags);
  405. spin_lock(&rbio->bio_list_lock);
  406. /* bump our ref if we were not in the list before */
  407. if (!test_and_set_bit(RBIO_CACHE_BIT, &rbio->flags))
  408. refcount_inc(&rbio->refs);
  409. if (!list_empty(&rbio->stripe_cache)){
  410. list_move(&rbio->stripe_cache, &table->stripe_cache);
  411. } else {
  412. list_add(&rbio->stripe_cache, &table->stripe_cache);
  413. table->cache_size += 1;
  414. }
  415. spin_unlock(&rbio->bio_list_lock);
  416. if (table->cache_size > RBIO_CACHE_SIZE) {
  417. struct btrfs_raid_bio *found;
  418. found = list_entry(table->stripe_cache.prev,
  419. struct btrfs_raid_bio,
  420. stripe_cache);
  421. if (found != rbio)
  422. __remove_rbio_from_cache(found);
  423. }
  424. spin_unlock_irqrestore(&table->cache_lock, flags);
  425. }
  426. /*
  427. * helper function to run the xor_blocks api. It is only
  428. * able to do MAX_XOR_BLOCKS at a time, so we need to
  429. * loop through.
  430. */
  431. static void run_xor(void **pages, int src_cnt, ssize_t len)
  432. {
  433. int src_off = 0;
  434. int xor_src_cnt = 0;
  435. void *dest = pages[src_cnt];
  436. while(src_cnt > 0) {
  437. xor_src_cnt = min(src_cnt, MAX_XOR_BLOCKS);
  438. xor_blocks(xor_src_cnt, len, dest, pages + src_off);
  439. src_cnt -= xor_src_cnt;
  440. src_off += xor_src_cnt;
  441. }
  442. }
  443. /*
  444. * returns true if the bio list inside this rbio
  445. * covers an entire stripe (no rmw required).
  446. * Must be called with the bio list lock held, or
  447. * at a time when you know it is impossible to add
  448. * new bios into the list
  449. */
  450. static int __rbio_is_full(struct btrfs_raid_bio *rbio)
  451. {
  452. unsigned long size = rbio->bio_list_bytes;
  453. int ret = 1;
  454. if (size != rbio->nr_data * rbio->stripe_len)
  455. ret = 0;
  456. BUG_ON(size > rbio->nr_data * rbio->stripe_len);
  457. return ret;
  458. }
  459. static int rbio_is_full(struct btrfs_raid_bio *rbio)
  460. {
  461. unsigned long flags;
  462. int ret;
  463. spin_lock_irqsave(&rbio->bio_list_lock, flags);
  464. ret = __rbio_is_full(rbio);
  465. spin_unlock_irqrestore(&rbio->bio_list_lock, flags);
  466. return ret;
  467. }
  468. /*
  469. * returns 1 if it is safe to merge two rbios together.
  470. * The merging is safe if the two rbios correspond to
  471. * the same stripe and if they are both going in the same
  472. * direction (read vs write), and if neither one is
  473. * locked for final IO
  474. *
  475. * The caller is responsible for locking such that
  476. * rmw_locked is safe to test
  477. */
  478. static int rbio_can_merge(struct btrfs_raid_bio *last,
  479. struct btrfs_raid_bio *cur)
  480. {
  481. if (test_bit(RBIO_RMW_LOCKED_BIT, &last->flags) ||
  482. test_bit(RBIO_RMW_LOCKED_BIT, &cur->flags))
  483. return 0;
  484. /*
  485. * we can't merge with cached rbios, since the
  486. * idea is that when we merge the destination
  487. * rbio is going to run our IO for us. We can
  488. * steal from cached rbios though, other functions
  489. * handle that.
  490. */
  491. if (test_bit(RBIO_CACHE_BIT, &last->flags) ||
  492. test_bit(RBIO_CACHE_BIT, &cur->flags))
  493. return 0;
  494. if (last->bbio->raid_map[0] !=
  495. cur->bbio->raid_map[0])
  496. return 0;
  497. /* we can't merge with different operations */
  498. if (last->operation != cur->operation)
  499. return 0;
  500. /*
  501. * We've need read the full stripe from the drive.
  502. * check and repair the parity and write the new results.
  503. *
  504. * We're not allowed to add any new bios to the
  505. * bio list here, anyone else that wants to
  506. * change this stripe needs to do their own rmw.
  507. */
  508. if (last->operation == BTRFS_RBIO_PARITY_SCRUB ||
  509. cur->operation == BTRFS_RBIO_PARITY_SCRUB)
  510. return 0;
  511. if (last->operation == BTRFS_RBIO_REBUILD_MISSING ||
  512. cur->operation == BTRFS_RBIO_REBUILD_MISSING)
  513. return 0;
  514. return 1;
  515. }
  516. static int rbio_stripe_page_index(struct btrfs_raid_bio *rbio, int stripe,
  517. int index)
  518. {
  519. return stripe * rbio->stripe_npages + index;
  520. }
  521. /*
  522. * these are just the pages from the rbio array, not from anything
  523. * the FS sent down to us
  524. */
  525. static struct page *rbio_stripe_page(struct btrfs_raid_bio *rbio, int stripe,
  526. int index)
  527. {
  528. return rbio->stripe_pages[rbio_stripe_page_index(rbio, stripe, index)];
  529. }
  530. /*
  531. * helper to index into the pstripe
  532. */
  533. static struct page *rbio_pstripe_page(struct btrfs_raid_bio *rbio, int index)
  534. {
  535. return rbio_stripe_page(rbio, rbio->nr_data, index);
  536. }
  537. /*
  538. * helper to index into the qstripe, returns null
  539. * if there is no qstripe
  540. */
  541. static struct page *rbio_qstripe_page(struct btrfs_raid_bio *rbio, int index)
  542. {
  543. if (rbio->nr_data + 1 == rbio->real_stripes)
  544. return NULL;
  545. return rbio_stripe_page(rbio, rbio->nr_data + 1, index);
  546. }
  547. /*
  548. * The first stripe in the table for a logical address
  549. * has the lock. rbios are added in one of three ways:
  550. *
  551. * 1) Nobody has the stripe locked yet. The rbio is given
  552. * the lock and 0 is returned. The caller must start the IO
  553. * themselves.
  554. *
  555. * 2) Someone has the stripe locked, but we're able to merge
  556. * with the lock owner. The rbio is freed and the IO will
  557. * start automatically along with the existing rbio. 1 is returned.
  558. *
  559. * 3) Someone has the stripe locked, but we're not able to merge.
  560. * The rbio is added to the lock owner's plug list, or merged into
  561. * an rbio already on the plug list. When the lock owner unlocks,
  562. * the next rbio on the list is run and the IO is started automatically.
  563. * 1 is returned
  564. *
  565. * If we return 0, the caller still owns the rbio and must continue with
  566. * IO submission. If we return 1, the caller must assume the rbio has
  567. * already been freed.
  568. */
  569. static noinline int lock_stripe_add(struct btrfs_raid_bio *rbio)
  570. {
  571. int bucket = rbio_bucket(rbio);
  572. struct btrfs_stripe_hash *h = rbio->fs_info->stripe_hash_table->table + bucket;
  573. struct btrfs_raid_bio *cur;
  574. struct btrfs_raid_bio *pending;
  575. unsigned long flags;
  576. DEFINE_WAIT(wait);
  577. struct btrfs_raid_bio *freeit = NULL;
  578. struct btrfs_raid_bio *cache_drop = NULL;
  579. int ret = 0;
  580. spin_lock_irqsave(&h->lock, flags);
  581. list_for_each_entry(cur, &h->hash_list, hash_list) {
  582. if (cur->bbio->raid_map[0] == rbio->bbio->raid_map[0]) {
  583. spin_lock(&cur->bio_list_lock);
  584. /* can we steal this cached rbio's pages? */
  585. if (bio_list_empty(&cur->bio_list) &&
  586. list_empty(&cur->plug_list) &&
  587. test_bit(RBIO_CACHE_BIT, &cur->flags) &&
  588. !test_bit(RBIO_RMW_LOCKED_BIT, &cur->flags)) {
  589. list_del_init(&cur->hash_list);
  590. refcount_dec(&cur->refs);
  591. steal_rbio(cur, rbio);
  592. cache_drop = cur;
  593. spin_unlock(&cur->bio_list_lock);
  594. goto lockit;
  595. }
  596. /* can we merge into the lock owner? */
  597. if (rbio_can_merge(cur, rbio)) {
  598. merge_rbio(cur, rbio);
  599. spin_unlock(&cur->bio_list_lock);
  600. freeit = rbio;
  601. ret = 1;
  602. goto out;
  603. }
  604. /*
  605. * we couldn't merge with the running
  606. * rbio, see if we can merge with the
  607. * pending ones. We don't have to
  608. * check for rmw_locked because there
  609. * is no way they are inside finish_rmw
  610. * right now
  611. */
  612. list_for_each_entry(pending, &cur->plug_list,
  613. plug_list) {
  614. if (rbio_can_merge(pending, rbio)) {
  615. merge_rbio(pending, rbio);
  616. spin_unlock(&cur->bio_list_lock);
  617. freeit = rbio;
  618. ret = 1;
  619. goto out;
  620. }
  621. }
  622. /* no merging, put us on the tail of the plug list,
  623. * our rbio will be started with the currently
  624. * running rbio unlocks
  625. */
  626. list_add_tail(&rbio->plug_list, &cur->plug_list);
  627. spin_unlock(&cur->bio_list_lock);
  628. ret = 1;
  629. goto out;
  630. }
  631. }
  632. lockit:
  633. refcount_inc(&rbio->refs);
  634. list_add(&rbio->hash_list, &h->hash_list);
  635. out:
  636. spin_unlock_irqrestore(&h->lock, flags);
  637. if (cache_drop)
  638. remove_rbio_from_cache(cache_drop);
  639. if (freeit)
  640. __free_raid_bio(freeit);
  641. return ret;
  642. }
  643. /*
  644. * called as rmw or parity rebuild is completed. If the plug list has more
  645. * rbios waiting for this stripe, the next one on the list will be started
  646. */
  647. static noinline void unlock_stripe(struct btrfs_raid_bio *rbio)
  648. {
  649. int bucket;
  650. struct btrfs_stripe_hash *h;
  651. unsigned long flags;
  652. int keep_cache = 0;
  653. bucket = rbio_bucket(rbio);
  654. h = rbio->fs_info->stripe_hash_table->table + bucket;
  655. if (list_empty(&rbio->plug_list))
  656. cache_rbio(rbio);
  657. spin_lock_irqsave(&h->lock, flags);
  658. spin_lock(&rbio->bio_list_lock);
  659. if (!list_empty(&rbio->hash_list)) {
  660. /*
  661. * if we're still cached and there is no other IO
  662. * to perform, just leave this rbio here for others
  663. * to steal from later
  664. */
  665. if (list_empty(&rbio->plug_list) &&
  666. test_bit(RBIO_CACHE_BIT, &rbio->flags)) {
  667. keep_cache = 1;
  668. clear_bit(RBIO_RMW_LOCKED_BIT, &rbio->flags);
  669. BUG_ON(!bio_list_empty(&rbio->bio_list));
  670. goto done;
  671. }
  672. list_del_init(&rbio->hash_list);
  673. refcount_dec(&rbio->refs);
  674. /*
  675. * we use the plug list to hold all the rbios
  676. * waiting for the chance to lock this stripe.
  677. * hand the lock over to one of them.
  678. */
  679. if (!list_empty(&rbio->plug_list)) {
  680. struct btrfs_raid_bio *next;
  681. struct list_head *head = rbio->plug_list.next;
  682. next = list_entry(head, struct btrfs_raid_bio,
  683. plug_list);
  684. list_del_init(&rbio->plug_list);
  685. list_add(&next->hash_list, &h->hash_list);
  686. refcount_inc(&next->refs);
  687. spin_unlock(&rbio->bio_list_lock);
  688. spin_unlock_irqrestore(&h->lock, flags);
  689. if (next->operation == BTRFS_RBIO_READ_REBUILD)
  690. async_read_rebuild(next);
  691. else if (next->operation == BTRFS_RBIO_REBUILD_MISSING) {
  692. steal_rbio(rbio, next);
  693. async_read_rebuild(next);
  694. } else if (next->operation == BTRFS_RBIO_WRITE) {
  695. steal_rbio(rbio, next);
  696. async_rmw_stripe(next);
  697. } else if (next->operation == BTRFS_RBIO_PARITY_SCRUB) {
  698. steal_rbio(rbio, next);
  699. async_scrub_parity(next);
  700. }
  701. goto done_nolock;
  702. /*
  703. * The barrier for this waitqueue_active is not needed,
  704. * we're protected by h->lock and can't miss a wakeup.
  705. */
  706. } else if (waitqueue_active(&h->wait)) {
  707. spin_unlock(&rbio->bio_list_lock);
  708. spin_unlock_irqrestore(&h->lock, flags);
  709. wake_up(&h->wait);
  710. goto done_nolock;
  711. }
  712. }
  713. done:
  714. spin_unlock(&rbio->bio_list_lock);
  715. spin_unlock_irqrestore(&h->lock, flags);
  716. done_nolock:
  717. if (!keep_cache)
  718. remove_rbio_from_cache(rbio);
  719. }
  720. static void __free_raid_bio(struct btrfs_raid_bio *rbio)
  721. {
  722. int i;
  723. if (!refcount_dec_and_test(&rbio->refs))
  724. return;
  725. WARN_ON(!list_empty(&rbio->stripe_cache));
  726. WARN_ON(!list_empty(&rbio->hash_list));
  727. WARN_ON(!bio_list_empty(&rbio->bio_list));
  728. for (i = 0; i < rbio->nr_pages; i++) {
  729. if (rbio->stripe_pages[i]) {
  730. __free_page(rbio->stripe_pages[i]);
  731. rbio->stripe_pages[i] = NULL;
  732. }
  733. }
  734. btrfs_put_bbio(rbio->bbio);
  735. kfree(rbio);
  736. }
  737. static void free_raid_bio(struct btrfs_raid_bio *rbio)
  738. {
  739. unlock_stripe(rbio);
  740. __free_raid_bio(rbio);
  741. }
  742. /*
  743. * this frees the rbio and runs through all the bios in the
  744. * bio_list and calls end_io on them
  745. */
  746. static void rbio_orig_end_io(struct btrfs_raid_bio *rbio, int err)
  747. {
  748. struct bio *cur = bio_list_get(&rbio->bio_list);
  749. struct bio *next;
  750. if (rbio->generic_bio_cnt)
  751. btrfs_bio_counter_sub(rbio->fs_info, rbio->generic_bio_cnt);
  752. free_raid_bio(rbio);
  753. while (cur) {
  754. next = cur->bi_next;
  755. cur->bi_next = NULL;
  756. cur->bi_error = err;
  757. bio_endio(cur);
  758. cur = next;
  759. }
  760. }
  761. /*
  762. * end io function used by finish_rmw. When we finally
  763. * get here, we've written a full stripe
  764. */
  765. static void raid_write_end_io(struct bio *bio)
  766. {
  767. struct btrfs_raid_bio *rbio = bio->bi_private;
  768. int err = bio->bi_error;
  769. int max_errors;
  770. if (err)
  771. fail_bio_stripe(rbio, bio);
  772. bio_put(bio);
  773. if (!atomic_dec_and_test(&rbio->stripes_pending))
  774. return;
  775. err = 0;
  776. /* OK, we have read all the stripes we need to. */
  777. max_errors = (rbio->operation == BTRFS_RBIO_PARITY_SCRUB) ?
  778. 0 : rbio->bbio->max_errors;
  779. if (atomic_read(&rbio->error) > max_errors)
  780. err = -EIO;
  781. rbio_orig_end_io(rbio, err);
  782. }
  783. /*
  784. * the read/modify/write code wants to use the original bio for
  785. * any pages it included, and then use the rbio for everything
  786. * else. This function decides if a given index (stripe number)
  787. * and page number in that stripe fall inside the original bio
  788. * or the rbio.
  789. *
  790. * if you set bio_list_only, you'll get a NULL back for any ranges
  791. * that are outside the bio_list
  792. *
  793. * This doesn't take any refs on anything, you get a bare page pointer
  794. * and the caller must bump refs as required.
  795. *
  796. * You must call index_rbio_pages once before you can trust
  797. * the answers from this function.
  798. */
  799. static struct page *page_in_rbio(struct btrfs_raid_bio *rbio,
  800. int index, int pagenr, int bio_list_only)
  801. {
  802. int chunk_page;
  803. struct page *p = NULL;
  804. chunk_page = index * (rbio->stripe_len >> PAGE_SHIFT) + pagenr;
  805. spin_lock_irq(&rbio->bio_list_lock);
  806. p = rbio->bio_pages[chunk_page];
  807. spin_unlock_irq(&rbio->bio_list_lock);
  808. if (p || bio_list_only)
  809. return p;
  810. return rbio->stripe_pages[chunk_page];
  811. }
  812. /*
  813. * number of pages we need for the entire stripe across all the
  814. * drives
  815. */
  816. static unsigned long rbio_nr_pages(unsigned long stripe_len, int nr_stripes)
  817. {
  818. return DIV_ROUND_UP(stripe_len, PAGE_SIZE) * nr_stripes;
  819. }
  820. /*
  821. * allocation and initial setup for the btrfs_raid_bio. Not
  822. * this does not allocate any pages for rbio->pages.
  823. */
  824. static struct btrfs_raid_bio *alloc_rbio(struct btrfs_fs_info *fs_info,
  825. struct btrfs_bio *bbio,
  826. u64 stripe_len)
  827. {
  828. struct btrfs_raid_bio *rbio;
  829. int nr_data = 0;
  830. int real_stripes = bbio->num_stripes - bbio->num_tgtdevs;
  831. int num_pages = rbio_nr_pages(stripe_len, real_stripes);
  832. int stripe_npages = DIV_ROUND_UP(stripe_len, PAGE_SIZE);
  833. void *p;
  834. rbio = kzalloc(sizeof(*rbio) + num_pages * sizeof(struct page *) * 2 +
  835. DIV_ROUND_UP(stripe_npages, BITS_PER_LONG) *
  836. sizeof(long), GFP_NOFS);
  837. if (!rbio)
  838. return ERR_PTR(-ENOMEM);
  839. bio_list_init(&rbio->bio_list);
  840. INIT_LIST_HEAD(&rbio->plug_list);
  841. spin_lock_init(&rbio->bio_list_lock);
  842. INIT_LIST_HEAD(&rbio->stripe_cache);
  843. INIT_LIST_HEAD(&rbio->hash_list);
  844. rbio->bbio = bbio;
  845. rbio->fs_info = fs_info;
  846. rbio->stripe_len = stripe_len;
  847. rbio->nr_pages = num_pages;
  848. rbio->real_stripes = real_stripes;
  849. rbio->stripe_npages = stripe_npages;
  850. rbio->faila = -1;
  851. rbio->failb = -1;
  852. refcount_set(&rbio->refs, 1);
  853. atomic_set(&rbio->error, 0);
  854. atomic_set(&rbio->stripes_pending, 0);
  855. /*
  856. * the stripe_pages and bio_pages array point to the extra
  857. * memory we allocated past the end of the rbio
  858. */
  859. p = rbio + 1;
  860. rbio->stripe_pages = p;
  861. rbio->bio_pages = p + sizeof(struct page *) * num_pages;
  862. rbio->dbitmap = p + sizeof(struct page *) * num_pages * 2;
  863. if (bbio->map_type & BTRFS_BLOCK_GROUP_RAID5)
  864. nr_data = real_stripes - 1;
  865. else if (bbio->map_type & BTRFS_BLOCK_GROUP_RAID6)
  866. nr_data = real_stripes - 2;
  867. else
  868. BUG();
  869. rbio->nr_data = nr_data;
  870. return rbio;
  871. }
  872. /* allocate pages for all the stripes in the bio, including parity */
  873. static int alloc_rbio_pages(struct btrfs_raid_bio *rbio)
  874. {
  875. int i;
  876. struct page *page;
  877. for (i = 0; i < rbio->nr_pages; i++) {
  878. if (rbio->stripe_pages[i])
  879. continue;
  880. page = alloc_page(GFP_NOFS | __GFP_HIGHMEM);
  881. if (!page)
  882. return -ENOMEM;
  883. rbio->stripe_pages[i] = page;
  884. }
  885. return 0;
  886. }
  887. /* only allocate pages for p/q stripes */
  888. static int alloc_rbio_parity_pages(struct btrfs_raid_bio *rbio)
  889. {
  890. int i;
  891. struct page *page;
  892. i = rbio_stripe_page_index(rbio, rbio->nr_data, 0);
  893. for (; i < rbio->nr_pages; i++) {
  894. if (rbio->stripe_pages[i])
  895. continue;
  896. page = alloc_page(GFP_NOFS | __GFP_HIGHMEM);
  897. if (!page)
  898. return -ENOMEM;
  899. rbio->stripe_pages[i] = page;
  900. }
  901. return 0;
  902. }
  903. /*
  904. * add a single page from a specific stripe into our list of bios for IO
  905. * this will try to merge into existing bios if possible, and returns
  906. * zero if all went well.
  907. */
  908. static int rbio_add_io_page(struct btrfs_raid_bio *rbio,
  909. struct bio_list *bio_list,
  910. struct page *page,
  911. int stripe_nr,
  912. unsigned long page_index,
  913. unsigned long bio_max_len)
  914. {
  915. struct bio *last = bio_list->tail;
  916. u64 last_end = 0;
  917. int ret;
  918. struct bio *bio;
  919. struct btrfs_bio_stripe *stripe;
  920. u64 disk_start;
  921. stripe = &rbio->bbio->stripes[stripe_nr];
  922. disk_start = stripe->physical + (page_index << PAGE_SHIFT);
  923. /* if the device is missing, just fail this stripe */
  924. if (!stripe->dev->bdev)
  925. return fail_rbio_index(rbio, stripe_nr);
  926. /* see if we can add this page onto our existing bio */
  927. if (last) {
  928. last_end = (u64)last->bi_iter.bi_sector << 9;
  929. last_end += last->bi_iter.bi_size;
  930. /*
  931. * we can't merge these if they are from different
  932. * devices or if they are not contiguous
  933. */
  934. if (last_end == disk_start && stripe->dev->bdev &&
  935. !last->bi_error &&
  936. last->bi_bdev == stripe->dev->bdev) {
  937. ret = bio_add_page(last, page, PAGE_SIZE, 0);
  938. if (ret == PAGE_SIZE)
  939. return 0;
  940. }
  941. }
  942. /* put a new bio on the list */
  943. bio = btrfs_io_bio_alloc(GFP_NOFS, bio_max_len >> PAGE_SHIFT?:1);
  944. if (!bio)
  945. return -ENOMEM;
  946. bio->bi_iter.bi_size = 0;
  947. bio->bi_bdev = stripe->dev->bdev;
  948. bio->bi_iter.bi_sector = disk_start >> 9;
  949. bio_add_page(bio, page, PAGE_SIZE, 0);
  950. bio_list_add(bio_list, bio);
  951. return 0;
  952. }
  953. /*
  954. * while we're doing the read/modify/write cycle, we could
  955. * have errors in reading pages off the disk. This checks
  956. * for errors and if we're not able to read the page it'll
  957. * trigger parity reconstruction. The rmw will be finished
  958. * after we've reconstructed the failed stripes
  959. */
  960. static void validate_rbio_for_rmw(struct btrfs_raid_bio *rbio)
  961. {
  962. if (rbio->faila >= 0 || rbio->failb >= 0) {
  963. BUG_ON(rbio->faila == rbio->real_stripes - 1);
  964. __raid56_parity_recover(rbio);
  965. } else {
  966. finish_rmw(rbio);
  967. }
  968. }
  969. /*
  970. * helper function to walk our bio list and populate the bio_pages array with
  971. * the result. This seems expensive, but it is faster than constantly
  972. * searching through the bio list as we setup the IO in finish_rmw or stripe
  973. * reconstruction.
  974. *
  975. * This must be called before you trust the answers from page_in_rbio
  976. */
  977. static void index_rbio_pages(struct btrfs_raid_bio *rbio)
  978. {
  979. struct bio *bio;
  980. struct bio_vec *bvec;
  981. u64 start;
  982. unsigned long stripe_offset;
  983. unsigned long page_index;
  984. int i;
  985. spin_lock_irq(&rbio->bio_list_lock);
  986. bio_list_for_each(bio, &rbio->bio_list) {
  987. start = (u64)bio->bi_iter.bi_sector << 9;
  988. stripe_offset = start - rbio->bbio->raid_map[0];
  989. page_index = stripe_offset >> PAGE_SHIFT;
  990. bio_for_each_segment_all(bvec, bio, i)
  991. rbio->bio_pages[page_index + i] = bvec->bv_page;
  992. }
  993. spin_unlock_irq(&rbio->bio_list_lock);
  994. }
  995. /*
  996. * this is called from one of two situations. We either
  997. * have a full stripe from the higher layers, or we've read all
  998. * the missing bits off disk.
  999. *
  1000. * This will calculate the parity and then send down any
  1001. * changed blocks.
  1002. */
  1003. static noinline void finish_rmw(struct btrfs_raid_bio *rbio)
  1004. {
  1005. struct btrfs_bio *bbio = rbio->bbio;
  1006. void *pointers[rbio->real_stripes];
  1007. int nr_data = rbio->nr_data;
  1008. int stripe;
  1009. int pagenr;
  1010. int p_stripe = -1;
  1011. int q_stripe = -1;
  1012. struct bio_list bio_list;
  1013. struct bio *bio;
  1014. int ret;
  1015. bio_list_init(&bio_list);
  1016. if (rbio->real_stripes - rbio->nr_data == 1) {
  1017. p_stripe = rbio->real_stripes - 1;
  1018. } else if (rbio->real_stripes - rbio->nr_data == 2) {
  1019. p_stripe = rbio->real_stripes - 2;
  1020. q_stripe = rbio->real_stripes - 1;
  1021. } else {
  1022. BUG();
  1023. }
  1024. /* at this point we either have a full stripe,
  1025. * or we've read the full stripe from the drive.
  1026. * recalculate the parity and write the new results.
  1027. *
  1028. * We're not allowed to add any new bios to the
  1029. * bio list here, anyone else that wants to
  1030. * change this stripe needs to do their own rmw.
  1031. */
  1032. spin_lock_irq(&rbio->bio_list_lock);
  1033. set_bit(RBIO_RMW_LOCKED_BIT, &rbio->flags);
  1034. spin_unlock_irq(&rbio->bio_list_lock);
  1035. atomic_set(&rbio->error, 0);
  1036. /*
  1037. * now that we've set rmw_locked, run through the
  1038. * bio list one last time and map the page pointers
  1039. *
  1040. * We don't cache full rbios because we're assuming
  1041. * the higher layers are unlikely to use this area of
  1042. * the disk again soon. If they do use it again,
  1043. * hopefully they will send another full bio.
  1044. */
  1045. index_rbio_pages(rbio);
  1046. if (!rbio_is_full(rbio))
  1047. cache_rbio_pages(rbio);
  1048. else
  1049. clear_bit(RBIO_CACHE_READY_BIT, &rbio->flags);
  1050. for (pagenr = 0; pagenr < rbio->stripe_npages; pagenr++) {
  1051. struct page *p;
  1052. /* first collect one page from each data stripe */
  1053. for (stripe = 0; stripe < nr_data; stripe++) {
  1054. p = page_in_rbio(rbio, stripe, pagenr, 0);
  1055. pointers[stripe] = kmap(p);
  1056. }
  1057. /* then add the parity stripe */
  1058. p = rbio_pstripe_page(rbio, pagenr);
  1059. SetPageUptodate(p);
  1060. pointers[stripe++] = kmap(p);
  1061. if (q_stripe != -1) {
  1062. /*
  1063. * raid6, add the qstripe and call the
  1064. * library function to fill in our p/q
  1065. */
  1066. p = rbio_qstripe_page(rbio, pagenr);
  1067. SetPageUptodate(p);
  1068. pointers[stripe++] = kmap(p);
  1069. raid6_call.gen_syndrome(rbio->real_stripes, PAGE_SIZE,
  1070. pointers);
  1071. } else {
  1072. /* raid5 */
  1073. memcpy(pointers[nr_data], pointers[0], PAGE_SIZE);
  1074. run_xor(pointers + 1, nr_data - 1, PAGE_SIZE);
  1075. }
  1076. for (stripe = 0; stripe < rbio->real_stripes; stripe++)
  1077. kunmap(page_in_rbio(rbio, stripe, pagenr, 0));
  1078. }
  1079. /*
  1080. * time to start writing. Make bios for everything from the
  1081. * higher layers (the bio_list in our rbio) and our p/q. Ignore
  1082. * everything else.
  1083. */
  1084. for (stripe = 0; stripe < rbio->real_stripes; stripe++) {
  1085. for (pagenr = 0; pagenr < rbio->stripe_npages; pagenr++) {
  1086. struct page *page;
  1087. if (stripe < rbio->nr_data) {
  1088. page = page_in_rbio(rbio, stripe, pagenr, 1);
  1089. if (!page)
  1090. continue;
  1091. } else {
  1092. page = rbio_stripe_page(rbio, stripe, pagenr);
  1093. }
  1094. ret = rbio_add_io_page(rbio, &bio_list,
  1095. page, stripe, pagenr, rbio->stripe_len);
  1096. if (ret)
  1097. goto cleanup;
  1098. }
  1099. }
  1100. if (likely(!bbio->num_tgtdevs))
  1101. goto write_data;
  1102. for (stripe = 0; stripe < rbio->real_stripes; stripe++) {
  1103. if (!bbio->tgtdev_map[stripe])
  1104. continue;
  1105. for (pagenr = 0; pagenr < rbio->stripe_npages; pagenr++) {
  1106. struct page *page;
  1107. if (stripe < rbio->nr_data) {
  1108. page = page_in_rbio(rbio, stripe, pagenr, 1);
  1109. if (!page)
  1110. continue;
  1111. } else {
  1112. page = rbio_stripe_page(rbio, stripe, pagenr);
  1113. }
  1114. ret = rbio_add_io_page(rbio, &bio_list, page,
  1115. rbio->bbio->tgtdev_map[stripe],
  1116. pagenr, rbio->stripe_len);
  1117. if (ret)
  1118. goto cleanup;
  1119. }
  1120. }
  1121. write_data:
  1122. atomic_set(&rbio->stripes_pending, bio_list_size(&bio_list));
  1123. BUG_ON(atomic_read(&rbio->stripes_pending) == 0);
  1124. while (1) {
  1125. bio = bio_list_pop(&bio_list);
  1126. if (!bio)
  1127. break;
  1128. bio->bi_private = rbio;
  1129. bio->bi_end_io = raid_write_end_io;
  1130. bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
  1131. submit_bio(bio);
  1132. }
  1133. return;
  1134. cleanup:
  1135. rbio_orig_end_io(rbio, -EIO);
  1136. }
  1137. /*
  1138. * helper to find the stripe number for a given bio. Used to figure out which
  1139. * stripe has failed. This expects the bio to correspond to a physical disk,
  1140. * so it looks up based on physical sector numbers.
  1141. */
  1142. static int find_bio_stripe(struct btrfs_raid_bio *rbio,
  1143. struct bio *bio)
  1144. {
  1145. u64 physical = bio->bi_iter.bi_sector;
  1146. u64 stripe_start;
  1147. int i;
  1148. struct btrfs_bio_stripe *stripe;
  1149. physical <<= 9;
  1150. for (i = 0; i < rbio->bbio->num_stripes; i++) {
  1151. stripe = &rbio->bbio->stripes[i];
  1152. stripe_start = stripe->physical;
  1153. if (physical >= stripe_start &&
  1154. physical < stripe_start + rbio->stripe_len &&
  1155. bio->bi_bdev == stripe->dev->bdev) {
  1156. return i;
  1157. }
  1158. }
  1159. return -1;
  1160. }
  1161. /*
  1162. * helper to find the stripe number for a given
  1163. * bio (before mapping). Used to figure out which stripe has
  1164. * failed. This looks up based on logical block numbers.
  1165. */
  1166. static int find_logical_bio_stripe(struct btrfs_raid_bio *rbio,
  1167. struct bio *bio)
  1168. {
  1169. u64 logical = bio->bi_iter.bi_sector;
  1170. u64 stripe_start;
  1171. int i;
  1172. logical <<= 9;
  1173. for (i = 0; i < rbio->nr_data; i++) {
  1174. stripe_start = rbio->bbio->raid_map[i];
  1175. if (logical >= stripe_start &&
  1176. logical < stripe_start + rbio->stripe_len) {
  1177. return i;
  1178. }
  1179. }
  1180. return -1;
  1181. }
  1182. /*
  1183. * returns -EIO if we had too many failures
  1184. */
  1185. static int fail_rbio_index(struct btrfs_raid_bio *rbio, int failed)
  1186. {
  1187. unsigned long flags;
  1188. int ret = 0;
  1189. spin_lock_irqsave(&rbio->bio_list_lock, flags);
  1190. /* we already know this stripe is bad, move on */
  1191. if (rbio->faila == failed || rbio->failb == failed)
  1192. goto out;
  1193. if (rbio->faila == -1) {
  1194. /* first failure on this rbio */
  1195. rbio->faila = failed;
  1196. atomic_inc(&rbio->error);
  1197. } else if (rbio->failb == -1) {
  1198. /* second failure on this rbio */
  1199. rbio->failb = failed;
  1200. atomic_inc(&rbio->error);
  1201. } else {
  1202. ret = -EIO;
  1203. }
  1204. out:
  1205. spin_unlock_irqrestore(&rbio->bio_list_lock, flags);
  1206. return ret;
  1207. }
  1208. /*
  1209. * helper to fail a stripe based on a physical disk
  1210. * bio.
  1211. */
  1212. static int fail_bio_stripe(struct btrfs_raid_bio *rbio,
  1213. struct bio *bio)
  1214. {
  1215. int failed = find_bio_stripe(rbio, bio);
  1216. if (failed < 0)
  1217. return -EIO;
  1218. return fail_rbio_index(rbio, failed);
  1219. }
  1220. /*
  1221. * this sets each page in the bio uptodate. It should only be used on private
  1222. * rbio pages, nothing that comes in from the higher layers
  1223. */
  1224. static void set_bio_pages_uptodate(struct bio *bio)
  1225. {
  1226. struct bio_vec *bvec;
  1227. int i;
  1228. bio_for_each_segment_all(bvec, bio, i)
  1229. SetPageUptodate(bvec->bv_page);
  1230. }
  1231. /*
  1232. * end io for the read phase of the rmw cycle. All the bios here are physical
  1233. * stripe bios we've read from the disk so we can recalculate the parity of the
  1234. * stripe.
  1235. *
  1236. * This will usually kick off finish_rmw once all the bios are read in, but it
  1237. * may trigger parity reconstruction if we had any errors along the way
  1238. */
  1239. static void raid_rmw_end_io(struct bio *bio)
  1240. {
  1241. struct btrfs_raid_bio *rbio = bio->bi_private;
  1242. if (bio->bi_error)
  1243. fail_bio_stripe(rbio, bio);
  1244. else
  1245. set_bio_pages_uptodate(bio);
  1246. bio_put(bio);
  1247. if (!atomic_dec_and_test(&rbio->stripes_pending))
  1248. return;
  1249. if (atomic_read(&rbio->error) > rbio->bbio->max_errors)
  1250. goto cleanup;
  1251. /*
  1252. * this will normally call finish_rmw to start our write
  1253. * but if there are any failed stripes we'll reconstruct
  1254. * from parity first
  1255. */
  1256. validate_rbio_for_rmw(rbio);
  1257. return;
  1258. cleanup:
  1259. rbio_orig_end_io(rbio, -EIO);
  1260. }
  1261. static void async_rmw_stripe(struct btrfs_raid_bio *rbio)
  1262. {
  1263. btrfs_init_work(&rbio->work, btrfs_rmw_helper, rmw_work, NULL, NULL);
  1264. btrfs_queue_work(rbio->fs_info->rmw_workers, &rbio->work);
  1265. }
  1266. static void async_read_rebuild(struct btrfs_raid_bio *rbio)
  1267. {
  1268. btrfs_init_work(&rbio->work, btrfs_rmw_helper,
  1269. read_rebuild_work, NULL, NULL);
  1270. btrfs_queue_work(rbio->fs_info->rmw_workers, &rbio->work);
  1271. }
  1272. /*
  1273. * the stripe must be locked by the caller. It will
  1274. * unlock after all the writes are done
  1275. */
  1276. static int raid56_rmw_stripe(struct btrfs_raid_bio *rbio)
  1277. {
  1278. int bios_to_read = 0;
  1279. struct bio_list bio_list;
  1280. int ret;
  1281. int pagenr;
  1282. int stripe;
  1283. struct bio *bio;
  1284. bio_list_init(&bio_list);
  1285. ret = alloc_rbio_pages(rbio);
  1286. if (ret)
  1287. goto cleanup;
  1288. index_rbio_pages(rbio);
  1289. atomic_set(&rbio->error, 0);
  1290. /*
  1291. * build a list of bios to read all the missing parts of this
  1292. * stripe
  1293. */
  1294. for (stripe = 0; stripe < rbio->nr_data; stripe++) {
  1295. for (pagenr = 0; pagenr < rbio->stripe_npages; pagenr++) {
  1296. struct page *page;
  1297. /*
  1298. * we want to find all the pages missing from
  1299. * the rbio and read them from the disk. If
  1300. * page_in_rbio finds a page in the bio list
  1301. * we don't need to read it off the stripe.
  1302. */
  1303. page = page_in_rbio(rbio, stripe, pagenr, 1);
  1304. if (page)
  1305. continue;
  1306. page = rbio_stripe_page(rbio, stripe, pagenr);
  1307. /*
  1308. * the bio cache may have handed us an uptodate
  1309. * page. If so, be happy and use it
  1310. */
  1311. if (PageUptodate(page))
  1312. continue;
  1313. ret = rbio_add_io_page(rbio, &bio_list, page,
  1314. stripe, pagenr, rbio->stripe_len);
  1315. if (ret)
  1316. goto cleanup;
  1317. }
  1318. }
  1319. bios_to_read = bio_list_size(&bio_list);
  1320. if (!bios_to_read) {
  1321. /*
  1322. * this can happen if others have merged with
  1323. * us, it means there is nothing left to read.
  1324. * But if there are missing devices it may not be
  1325. * safe to do the full stripe write yet.
  1326. */
  1327. goto finish;
  1328. }
  1329. /*
  1330. * the bbio may be freed once we submit the last bio. Make sure
  1331. * not to touch it after that
  1332. */
  1333. atomic_set(&rbio->stripes_pending, bios_to_read);
  1334. while (1) {
  1335. bio = bio_list_pop(&bio_list);
  1336. if (!bio)
  1337. break;
  1338. bio->bi_private = rbio;
  1339. bio->bi_end_io = raid_rmw_end_io;
  1340. bio_set_op_attrs(bio, REQ_OP_READ, 0);
  1341. btrfs_bio_wq_end_io(rbio->fs_info, bio, BTRFS_WQ_ENDIO_RAID56);
  1342. submit_bio(bio);
  1343. }
  1344. /* the actual write will happen once the reads are done */
  1345. return 0;
  1346. cleanup:
  1347. rbio_orig_end_io(rbio, -EIO);
  1348. return -EIO;
  1349. finish:
  1350. validate_rbio_for_rmw(rbio);
  1351. return 0;
  1352. }
  1353. /*
  1354. * if the upper layers pass in a full stripe, we thank them by only allocating
  1355. * enough pages to hold the parity, and sending it all down quickly.
  1356. */
  1357. static int full_stripe_write(struct btrfs_raid_bio *rbio)
  1358. {
  1359. int ret;
  1360. ret = alloc_rbio_parity_pages(rbio);
  1361. if (ret) {
  1362. __free_raid_bio(rbio);
  1363. return ret;
  1364. }
  1365. ret = lock_stripe_add(rbio);
  1366. if (ret == 0)
  1367. finish_rmw(rbio);
  1368. return 0;
  1369. }
  1370. /*
  1371. * partial stripe writes get handed over to async helpers.
  1372. * We're really hoping to merge a few more writes into this
  1373. * rbio before calculating new parity
  1374. */
  1375. static int partial_stripe_write(struct btrfs_raid_bio *rbio)
  1376. {
  1377. int ret;
  1378. ret = lock_stripe_add(rbio);
  1379. if (ret == 0)
  1380. async_rmw_stripe(rbio);
  1381. return 0;
  1382. }
  1383. /*
  1384. * sometimes while we were reading from the drive to
  1385. * recalculate parity, enough new bios come into create
  1386. * a full stripe. So we do a check here to see if we can
  1387. * go directly to finish_rmw
  1388. */
  1389. static int __raid56_parity_write(struct btrfs_raid_bio *rbio)
  1390. {
  1391. /* head off into rmw land if we don't have a full stripe */
  1392. if (!rbio_is_full(rbio))
  1393. return partial_stripe_write(rbio);
  1394. return full_stripe_write(rbio);
  1395. }
  1396. /*
  1397. * We use plugging call backs to collect full stripes.
  1398. * Any time we get a partial stripe write while plugged
  1399. * we collect it into a list. When the unplug comes down,
  1400. * we sort the list by logical block number and merge
  1401. * everything we can into the same rbios
  1402. */
  1403. struct btrfs_plug_cb {
  1404. struct blk_plug_cb cb;
  1405. struct btrfs_fs_info *info;
  1406. struct list_head rbio_list;
  1407. struct btrfs_work work;
  1408. };
  1409. /*
  1410. * rbios on the plug list are sorted for easier merging.
  1411. */
  1412. static int plug_cmp(void *priv, struct list_head *a, struct list_head *b)
  1413. {
  1414. struct btrfs_raid_bio *ra = container_of(a, struct btrfs_raid_bio,
  1415. plug_list);
  1416. struct btrfs_raid_bio *rb = container_of(b, struct btrfs_raid_bio,
  1417. plug_list);
  1418. u64 a_sector = ra->bio_list.head->bi_iter.bi_sector;
  1419. u64 b_sector = rb->bio_list.head->bi_iter.bi_sector;
  1420. if (a_sector < b_sector)
  1421. return -1;
  1422. if (a_sector > b_sector)
  1423. return 1;
  1424. return 0;
  1425. }
  1426. static void run_plug(struct btrfs_plug_cb *plug)
  1427. {
  1428. struct btrfs_raid_bio *cur;
  1429. struct btrfs_raid_bio *last = NULL;
  1430. /*
  1431. * sort our plug list then try to merge
  1432. * everything we can in hopes of creating full
  1433. * stripes.
  1434. */
  1435. list_sort(NULL, &plug->rbio_list, plug_cmp);
  1436. while (!list_empty(&plug->rbio_list)) {
  1437. cur = list_entry(plug->rbio_list.next,
  1438. struct btrfs_raid_bio, plug_list);
  1439. list_del_init(&cur->plug_list);
  1440. if (rbio_is_full(cur)) {
  1441. /* we have a full stripe, send it down */
  1442. full_stripe_write(cur);
  1443. continue;
  1444. }
  1445. if (last) {
  1446. if (rbio_can_merge(last, cur)) {
  1447. merge_rbio(last, cur);
  1448. __free_raid_bio(cur);
  1449. continue;
  1450. }
  1451. __raid56_parity_write(last);
  1452. }
  1453. last = cur;
  1454. }
  1455. if (last) {
  1456. __raid56_parity_write(last);
  1457. }
  1458. kfree(plug);
  1459. }
  1460. /*
  1461. * if the unplug comes from schedule, we have to push the
  1462. * work off to a helper thread
  1463. */
  1464. static void unplug_work(struct btrfs_work *work)
  1465. {
  1466. struct btrfs_plug_cb *plug;
  1467. plug = container_of(work, struct btrfs_plug_cb, work);
  1468. run_plug(plug);
  1469. }
  1470. static void btrfs_raid_unplug(struct blk_plug_cb *cb, bool from_schedule)
  1471. {
  1472. struct btrfs_plug_cb *plug;
  1473. plug = container_of(cb, struct btrfs_plug_cb, cb);
  1474. if (from_schedule) {
  1475. btrfs_init_work(&plug->work, btrfs_rmw_helper,
  1476. unplug_work, NULL, NULL);
  1477. btrfs_queue_work(plug->info->rmw_workers,
  1478. &plug->work);
  1479. return;
  1480. }
  1481. run_plug(plug);
  1482. }
  1483. /*
  1484. * our main entry point for writes from the rest of the FS.
  1485. */
  1486. int raid56_parity_write(struct btrfs_fs_info *fs_info, struct bio *bio,
  1487. struct btrfs_bio *bbio, u64 stripe_len)
  1488. {
  1489. struct btrfs_raid_bio *rbio;
  1490. struct btrfs_plug_cb *plug = NULL;
  1491. struct blk_plug_cb *cb;
  1492. int ret;
  1493. rbio = alloc_rbio(fs_info, bbio, stripe_len);
  1494. if (IS_ERR(rbio)) {
  1495. btrfs_put_bbio(bbio);
  1496. return PTR_ERR(rbio);
  1497. }
  1498. bio_list_add(&rbio->bio_list, bio);
  1499. rbio->bio_list_bytes = bio->bi_iter.bi_size;
  1500. rbio->operation = BTRFS_RBIO_WRITE;
  1501. btrfs_bio_counter_inc_noblocked(fs_info);
  1502. rbio->generic_bio_cnt = 1;
  1503. /*
  1504. * don't plug on full rbios, just get them out the door
  1505. * as quickly as we can
  1506. */
  1507. if (rbio_is_full(rbio)) {
  1508. ret = full_stripe_write(rbio);
  1509. if (ret)
  1510. btrfs_bio_counter_dec(fs_info);
  1511. return ret;
  1512. }
  1513. cb = blk_check_plugged(btrfs_raid_unplug, fs_info, sizeof(*plug));
  1514. if (cb) {
  1515. plug = container_of(cb, struct btrfs_plug_cb, cb);
  1516. if (!plug->info) {
  1517. plug->info = fs_info;
  1518. INIT_LIST_HEAD(&plug->rbio_list);
  1519. }
  1520. list_add_tail(&rbio->plug_list, &plug->rbio_list);
  1521. ret = 0;
  1522. } else {
  1523. ret = __raid56_parity_write(rbio);
  1524. if (ret)
  1525. btrfs_bio_counter_dec(fs_info);
  1526. }
  1527. return ret;
  1528. }
  1529. /*
  1530. * all parity reconstruction happens here. We've read in everything
  1531. * we can find from the drives and this does the heavy lifting of
  1532. * sorting the good from the bad.
  1533. */
  1534. static void __raid_recover_end_io(struct btrfs_raid_bio *rbio)
  1535. {
  1536. int pagenr, stripe;
  1537. void **pointers;
  1538. int faila = -1, failb = -1;
  1539. struct page *page;
  1540. int err;
  1541. int i;
  1542. pointers = kcalloc(rbio->real_stripes, sizeof(void *), GFP_NOFS);
  1543. if (!pointers) {
  1544. err = -ENOMEM;
  1545. goto cleanup_io;
  1546. }
  1547. faila = rbio->faila;
  1548. failb = rbio->failb;
  1549. if (rbio->operation == BTRFS_RBIO_READ_REBUILD ||
  1550. rbio->operation == BTRFS_RBIO_REBUILD_MISSING) {
  1551. spin_lock_irq(&rbio->bio_list_lock);
  1552. set_bit(RBIO_RMW_LOCKED_BIT, &rbio->flags);
  1553. spin_unlock_irq(&rbio->bio_list_lock);
  1554. }
  1555. index_rbio_pages(rbio);
  1556. for (pagenr = 0; pagenr < rbio->stripe_npages; pagenr++) {
  1557. /*
  1558. * Now we just use bitmap to mark the horizontal stripes in
  1559. * which we have data when doing parity scrub.
  1560. */
  1561. if (rbio->operation == BTRFS_RBIO_PARITY_SCRUB &&
  1562. !test_bit(pagenr, rbio->dbitmap))
  1563. continue;
  1564. /* setup our array of pointers with pages
  1565. * from each stripe
  1566. */
  1567. for (stripe = 0; stripe < rbio->real_stripes; stripe++) {
  1568. /*
  1569. * if we're rebuilding a read, we have to use
  1570. * pages from the bio list
  1571. */
  1572. if ((rbio->operation == BTRFS_RBIO_READ_REBUILD ||
  1573. rbio->operation == BTRFS_RBIO_REBUILD_MISSING) &&
  1574. (stripe == faila || stripe == failb)) {
  1575. page = page_in_rbio(rbio, stripe, pagenr, 0);
  1576. } else {
  1577. page = rbio_stripe_page(rbio, stripe, pagenr);
  1578. }
  1579. pointers[stripe] = kmap(page);
  1580. }
  1581. /* all raid6 handling here */
  1582. if (rbio->bbio->map_type & BTRFS_BLOCK_GROUP_RAID6) {
  1583. /*
  1584. * single failure, rebuild from parity raid5
  1585. * style
  1586. */
  1587. if (failb < 0) {
  1588. if (faila == rbio->nr_data) {
  1589. /*
  1590. * Just the P stripe has failed, without
  1591. * a bad data or Q stripe.
  1592. * TODO, we should redo the xor here.
  1593. */
  1594. err = -EIO;
  1595. goto cleanup;
  1596. }
  1597. /*
  1598. * a single failure in raid6 is rebuilt
  1599. * in the pstripe code below
  1600. */
  1601. goto pstripe;
  1602. }
  1603. /* make sure our ps and qs are in order */
  1604. if (faila > failb) {
  1605. int tmp = failb;
  1606. failb = faila;
  1607. faila = tmp;
  1608. }
  1609. /* if the q stripe is failed, do a pstripe reconstruction
  1610. * from the xors.
  1611. * If both the q stripe and the P stripe are failed, we're
  1612. * here due to a crc mismatch and we can't give them the
  1613. * data they want
  1614. */
  1615. if (rbio->bbio->raid_map[failb] == RAID6_Q_STRIPE) {
  1616. if (rbio->bbio->raid_map[faila] ==
  1617. RAID5_P_STRIPE) {
  1618. err = -EIO;
  1619. goto cleanup;
  1620. }
  1621. /*
  1622. * otherwise we have one bad data stripe and
  1623. * a good P stripe. raid5!
  1624. */
  1625. goto pstripe;
  1626. }
  1627. if (rbio->bbio->raid_map[failb] == RAID5_P_STRIPE) {
  1628. raid6_datap_recov(rbio->real_stripes,
  1629. PAGE_SIZE, faila, pointers);
  1630. } else {
  1631. raid6_2data_recov(rbio->real_stripes,
  1632. PAGE_SIZE, faila, failb,
  1633. pointers);
  1634. }
  1635. } else {
  1636. void *p;
  1637. /* rebuild from P stripe here (raid5 or raid6) */
  1638. BUG_ON(failb != -1);
  1639. pstripe:
  1640. /* Copy parity block into failed block to start with */
  1641. memcpy(pointers[faila],
  1642. pointers[rbio->nr_data],
  1643. PAGE_SIZE);
  1644. /* rearrange the pointer array */
  1645. p = pointers[faila];
  1646. for (stripe = faila; stripe < rbio->nr_data - 1; stripe++)
  1647. pointers[stripe] = pointers[stripe + 1];
  1648. pointers[rbio->nr_data - 1] = p;
  1649. /* xor in the rest */
  1650. run_xor(pointers, rbio->nr_data - 1, PAGE_SIZE);
  1651. }
  1652. /* if we're doing this rebuild as part of an rmw, go through
  1653. * and set all of our private rbio pages in the
  1654. * failed stripes as uptodate. This way finish_rmw will
  1655. * know they can be trusted. If this was a read reconstruction,
  1656. * other endio functions will fiddle the uptodate bits
  1657. */
  1658. if (rbio->operation == BTRFS_RBIO_WRITE) {
  1659. for (i = 0; i < rbio->stripe_npages; i++) {
  1660. if (faila != -1) {
  1661. page = rbio_stripe_page(rbio, faila, i);
  1662. SetPageUptodate(page);
  1663. }
  1664. if (failb != -1) {
  1665. page = rbio_stripe_page(rbio, failb, i);
  1666. SetPageUptodate(page);
  1667. }
  1668. }
  1669. }
  1670. for (stripe = 0; stripe < rbio->real_stripes; stripe++) {
  1671. /*
  1672. * if we're rebuilding a read, we have to use
  1673. * pages from the bio list
  1674. */
  1675. if ((rbio->operation == BTRFS_RBIO_READ_REBUILD ||
  1676. rbio->operation == BTRFS_RBIO_REBUILD_MISSING) &&
  1677. (stripe == faila || stripe == failb)) {
  1678. page = page_in_rbio(rbio, stripe, pagenr, 0);
  1679. } else {
  1680. page = rbio_stripe_page(rbio, stripe, pagenr);
  1681. }
  1682. kunmap(page);
  1683. }
  1684. }
  1685. err = 0;
  1686. cleanup:
  1687. kfree(pointers);
  1688. cleanup_io:
  1689. if (rbio->operation == BTRFS_RBIO_READ_REBUILD) {
  1690. if (err == 0)
  1691. cache_rbio_pages(rbio);
  1692. else
  1693. clear_bit(RBIO_CACHE_READY_BIT, &rbio->flags);
  1694. rbio_orig_end_io(rbio, err);
  1695. } else if (rbio->operation == BTRFS_RBIO_REBUILD_MISSING) {
  1696. rbio_orig_end_io(rbio, err);
  1697. } else if (err == 0) {
  1698. rbio->faila = -1;
  1699. rbio->failb = -1;
  1700. if (rbio->operation == BTRFS_RBIO_WRITE)
  1701. finish_rmw(rbio);
  1702. else if (rbio->operation == BTRFS_RBIO_PARITY_SCRUB)
  1703. finish_parity_scrub(rbio, 0);
  1704. else
  1705. BUG();
  1706. } else {
  1707. rbio_orig_end_io(rbio, err);
  1708. }
  1709. }
  1710. /*
  1711. * This is called only for stripes we've read from disk to
  1712. * reconstruct the parity.
  1713. */
  1714. static void raid_recover_end_io(struct bio *bio)
  1715. {
  1716. struct btrfs_raid_bio *rbio = bio->bi_private;
  1717. /*
  1718. * we only read stripe pages off the disk, set them
  1719. * up to date if there were no errors
  1720. */
  1721. if (bio->bi_error)
  1722. fail_bio_stripe(rbio, bio);
  1723. else
  1724. set_bio_pages_uptodate(bio);
  1725. bio_put(bio);
  1726. if (!atomic_dec_and_test(&rbio->stripes_pending))
  1727. return;
  1728. if (atomic_read(&rbio->error) > rbio->bbio->max_errors)
  1729. rbio_orig_end_io(rbio, -EIO);
  1730. else
  1731. __raid_recover_end_io(rbio);
  1732. }
  1733. /*
  1734. * reads everything we need off the disk to reconstruct
  1735. * the parity. endio handlers trigger final reconstruction
  1736. * when the IO is done.
  1737. *
  1738. * This is used both for reads from the higher layers and for
  1739. * parity construction required to finish a rmw cycle.
  1740. */
  1741. static int __raid56_parity_recover(struct btrfs_raid_bio *rbio)
  1742. {
  1743. int bios_to_read = 0;
  1744. struct bio_list bio_list;
  1745. int ret;
  1746. int pagenr;
  1747. int stripe;
  1748. struct bio *bio;
  1749. bio_list_init(&bio_list);
  1750. ret = alloc_rbio_pages(rbio);
  1751. if (ret)
  1752. goto cleanup;
  1753. atomic_set(&rbio->error, 0);
  1754. /*
  1755. * read everything that hasn't failed. Thanks to the
  1756. * stripe cache, it is possible that some or all of these
  1757. * pages are going to be uptodate.
  1758. */
  1759. for (stripe = 0; stripe < rbio->real_stripes; stripe++) {
  1760. if (rbio->faila == stripe || rbio->failb == stripe) {
  1761. atomic_inc(&rbio->error);
  1762. continue;
  1763. }
  1764. for (pagenr = 0; pagenr < rbio->stripe_npages; pagenr++) {
  1765. struct page *p;
  1766. /*
  1767. * the rmw code may have already read this
  1768. * page in
  1769. */
  1770. p = rbio_stripe_page(rbio, stripe, pagenr);
  1771. if (PageUptodate(p))
  1772. continue;
  1773. ret = rbio_add_io_page(rbio, &bio_list,
  1774. rbio_stripe_page(rbio, stripe, pagenr),
  1775. stripe, pagenr, rbio->stripe_len);
  1776. if (ret < 0)
  1777. goto cleanup;
  1778. }
  1779. }
  1780. bios_to_read = bio_list_size(&bio_list);
  1781. if (!bios_to_read) {
  1782. /*
  1783. * we might have no bios to read just because the pages
  1784. * were up to date, or we might have no bios to read because
  1785. * the devices were gone.
  1786. */
  1787. if (atomic_read(&rbio->error) <= rbio->bbio->max_errors) {
  1788. __raid_recover_end_io(rbio);
  1789. goto out;
  1790. } else {
  1791. goto cleanup;
  1792. }
  1793. }
  1794. /*
  1795. * the bbio may be freed once we submit the last bio. Make sure
  1796. * not to touch it after that
  1797. */
  1798. atomic_set(&rbio->stripes_pending, bios_to_read);
  1799. while (1) {
  1800. bio = bio_list_pop(&bio_list);
  1801. if (!bio)
  1802. break;
  1803. bio->bi_private = rbio;
  1804. bio->bi_end_io = raid_recover_end_io;
  1805. bio_set_op_attrs(bio, REQ_OP_READ, 0);
  1806. btrfs_bio_wq_end_io(rbio->fs_info, bio, BTRFS_WQ_ENDIO_RAID56);
  1807. submit_bio(bio);
  1808. }
  1809. out:
  1810. return 0;
  1811. cleanup:
  1812. if (rbio->operation == BTRFS_RBIO_READ_REBUILD ||
  1813. rbio->operation == BTRFS_RBIO_REBUILD_MISSING)
  1814. rbio_orig_end_io(rbio, -EIO);
  1815. return -EIO;
  1816. }
  1817. /*
  1818. * the main entry point for reads from the higher layers. This
  1819. * is really only called when the normal read path had a failure,
  1820. * so we assume the bio they send down corresponds to a failed part
  1821. * of the drive.
  1822. */
  1823. int raid56_parity_recover(struct btrfs_fs_info *fs_info, struct bio *bio,
  1824. struct btrfs_bio *bbio, u64 stripe_len,
  1825. int mirror_num, int generic_io)
  1826. {
  1827. struct btrfs_raid_bio *rbio;
  1828. int ret;
  1829. if (generic_io) {
  1830. ASSERT(bbio->mirror_num == mirror_num);
  1831. btrfs_io_bio(bio)->mirror_num = mirror_num;
  1832. }
  1833. rbio = alloc_rbio(fs_info, bbio, stripe_len);
  1834. if (IS_ERR(rbio)) {
  1835. if (generic_io)
  1836. btrfs_put_bbio(bbio);
  1837. return PTR_ERR(rbio);
  1838. }
  1839. rbio->operation = BTRFS_RBIO_READ_REBUILD;
  1840. bio_list_add(&rbio->bio_list, bio);
  1841. rbio->bio_list_bytes = bio->bi_iter.bi_size;
  1842. rbio->faila = find_logical_bio_stripe(rbio, bio);
  1843. if (rbio->faila == -1) {
  1844. btrfs_warn(fs_info,
  1845. "%s could not find the bad stripe in raid56 so that we cannot recover any more (bio has logical %llu len %llu, bbio has map_type %llu)",
  1846. __func__, (u64)bio->bi_iter.bi_sector << 9,
  1847. (u64)bio->bi_iter.bi_size, bbio->map_type);
  1848. if (generic_io)
  1849. btrfs_put_bbio(bbio);
  1850. kfree(rbio);
  1851. return -EIO;
  1852. }
  1853. if (generic_io) {
  1854. btrfs_bio_counter_inc_noblocked(fs_info);
  1855. rbio->generic_bio_cnt = 1;
  1856. } else {
  1857. btrfs_get_bbio(bbio);
  1858. }
  1859. /*
  1860. * reconstruct from the q stripe if they are
  1861. * asking for mirror 3
  1862. */
  1863. if (mirror_num == 3)
  1864. rbio->failb = rbio->real_stripes - 2;
  1865. ret = lock_stripe_add(rbio);
  1866. /*
  1867. * __raid56_parity_recover will end the bio with
  1868. * any errors it hits. We don't want to return
  1869. * its error value up the stack because our caller
  1870. * will end up calling bio_endio with any nonzero
  1871. * return
  1872. */
  1873. if (ret == 0)
  1874. __raid56_parity_recover(rbio);
  1875. /*
  1876. * our rbio has been added to the list of
  1877. * rbios that will be handled after the
  1878. * currently lock owner is done
  1879. */
  1880. return 0;
  1881. }
  1882. static void rmw_work(struct btrfs_work *work)
  1883. {
  1884. struct btrfs_raid_bio *rbio;
  1885. rbio = container_of(work, struct btrfs_raid_bio, work);
  1886. raid56_rmw_stripe(rbio);
  1887. }
  1888. static void read_rebuild_work(struct btrfs_work *work)
  1889. {
  1890. struct btrfs_raid_bio *rbio;
  1891. rbio = container_of(work, struct btrfs_raid_bio, work);
  1892. __raid56_parity_recover(rbio);
  1893. }
  1894. /*
  1895. * The following code is used to scrub/replace the parity stripe
  1896. *
  1897. * Caller must have already increased bio_counter for getting @bbio.
  1898. *
  1899. * Note: We need make sure all the pages that add into the scrub/replace
  1900. * raid bio are correct and not be changed during the scrub/replace. That
  1901. * is those pages just hold metadata or file data with checksum.
  1902. */
  1903. struct btrfs_raid_bio *
  1904. raid56_parity_alloc_scrub_rbio(struct btrfs_fs_info *fs_info, struct bio *bio,
  1905. struct btrfs_bio *bbio, u64 stripe_len,
  1906. struct btrfs_device *scrub_dev,
  1907. unsigned long *dbitmap, int stripe_nsectors)
  1908. {
  1909. struct btrfs_raid_bio *rbio;
  1910. int i;
  1911. rbio = alloc_rbio(fs_info, bbio, stripe_len);
  1912. if (IS_ERR(rbio))
  1913. return NULL;
  1914. bio_list_add(&rbio->bio_list, bio);
  1915. /*
  1916. * This is a special bio which is used to hold the completion handler
  1917. * and make the scrub rbio is similar to the other types
  1918. */
  1919. ASSERT(!bio->bi_iter.bi_size);
  1920. rbio->operation = BTRFS_RBIO_PARITY_SCRUB;
  1921. for (i = 0; i < rbio->real_stripes; i++) {
  1922. if (bbio->stripes[i].dev == scrub_dev) {
  1923. rbio->scrubp = i;
  1924. break;
  1925. }
  1926. }
  1927. /* Now we just support the sectorsize equals to page size */
  1928. ASSERT(fs_info->sectorsize == PAGE_SIZE);
  1929. ASSERT(rbio->stripe_npages == stripe_nsectors);
  1930. bitmap_copy(rbio->dbitmap, dbitmap, stripe_nsectors);
  1931. /*
  1932. * We have already increased bio_counter when getting bbio, record it
  1933. * so we can free it at rbio_orig_end_io().
  1934. */
  1935. rbio->generic_bio_cnt = 1;
  1936. return rbio;
  1937. }
  1938. /* Used for both parity scrub and missing. */
  1939. void raid56_add_scrub_pages(struct btrfs_raid_bio *rbio, struct page *page,
  1940. u64 logical)
  1941. {
  1942. int stripe_offset;
  1943. int index;
  1944. ASSERT(logical >= rbio->bbio->raid_map[0]);
  1945. ASSERT(logical + PAGE_SIZE <= rbio->bbio->raid_map[0] +
  1946. rbio->stripe_len * rbio->nr_data);
  1947. stripe_offset = (int)(logical - rbio->bbio->raid_map[0]);
  1948. index = stripe_offset >> PAGE_SHIFT;
  1949. rbio->bio_pages[index] = page;
  1950. }
  1951. /*
  1952. * We just scrub the parity that we have correct data on the same horizontal,
  1953. * so we needn't allocate all pages for all the stripes.
  1954. */
  1955. static int alloc_rbio_essential_pages(struct btrfs_raid_bio *rbio)
  1956. {
  1957. int i;
  1958. int bit;
  1959. int index;
  1960. struct page *page;
  1961. for_each_set_bit(bit, rbio->dbitmap, rbio->stripe_npages) {
  1962. for (i = 0; i < rbio->real_stripes; i++) {
  1963. index = i * rbio->stripe_npages + bit;
  1964. if (rbio->stripe_pages[index])
  1965. continue;
  1966. page = alloc_page(GFP_NOFS | __GFP_HIGHMEM);
  1967. if (!page)
  1968. return -ENOMEM;
  1969. rbio->stripe_pages[index] = page;
  1970. }
  1971. }
  1972. return 0;
  1973. }
  1974. static noinline void finish_parity_scrub(struct btrfs_raid_bio *rbio,
  1975. int need_check)
  1976. {
  1977. struct btrfs_bio *bbio = rbio->bbio;
  1978. void *pointers[rbio->real_stripes];
  1979. DECLARE_BITMAP(pbitmap, rbio->stripe_npages);
  1980. int nr_data = rbio->nr_data;
  1981. int stripe;
  1982. int pagenr;
  1983. int p_stripe = -1;
  1984. int q_stripe = -1;
  1985. struct page *p_page = NULL;
  1986. struct page *q_page = NULL;
  1987. struct bio_list bio_list;
  1988. struct bio *bio;
  1989. int is_replace = 0;
  1990. int ret;
  1991. bio_list_init(&bio_list);
  1992. if (rbio->real_stripes - rbio->nr_data == 1) {
  1993. p_stripe = rbio->real_stripes - 1;
  1994. } else if (rbio->real_stripes - rbio->nr_data == 2) {
  1995. p_stripe = rbio->real_stripes - 2;
  1996. q_stripe = rbio->real_stripes - 1;
  1997. } else {
  1998. BUG();
  1999. }
  2000. if (bbio->num_tgtdevs && bbio->tgtdev_map[rbio->scrubp]) {
  2001. is_replace = 1;
  2002. bitmap_copy(pbitmap, rbio->dbitmap, rbio->stripe_npages);
  2003. }
  2004. /*
  2005. * Because the higher layers(scrubber) are unlikely to
  2006. * use this area of the disk again soon, so don't cache
  2007. * it.
  2008. */
  2009. clear_bit(RBIO_CACHE_READY_BIT, &rbio->flags);
  2010. if (!need_check)
  2011. goto writeback;
  2012. p_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM);
  2013. if (!p_page)
  2014. goto cleanup;
  2015. SetPageUptodate(p_page);
  2016. if (q_stripe != -1) {
  2017. q_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM);
  2018. if (!q_page) {
  2019. __free_page(p_page);
  2020. goto cleanup;
  2021. }
  2022. SetPageUptodate(q_page);
  2023. }
  2024. atomic_set(&rbio->error, 0);
  2025. for_each_set_bit(pagenr, rbio->dbitmap, rbio->stripe_npages) {
  2026. struct page *p;
  2027. void *parity;
  2028. /* first collect one page from each data stripe */
  2029. for (stripe = 0; stripe < nr_data; stripe++) {
  2030. p = page_in_rbio(rbio, stripe, pagenr, 0);
  2031. pointers[stripe] = kmap(p);
  2032. }
  2033. /* then add the parity stripe */
  2034. pointers[stripe++] = kmap(p_page);
  2035. if (q_stripe != -1) {
  2036. /*
  2037. * raid6, add the qstripe and call the
  2038. * library function to fill in our p/q
  2039. */
  2040. pointers[stripe++] = kmap(q_page);
  2041. raid6_call.gen_syndrome(rbio->real_stripes, PAGE_SIZE,
  2042. pointers);
  2043. } else {
  2044. /* raid5 */
  2045. memcpy(pointers[nr_data], pointers[0], PAGE_SIZE);
  2046. run_xor(pointers + 1, nr_data - 1, PAGE_SIZE);
  2047. }
  2048. /* Check scrubbing parity and repair it */
  2049. p = rbio_stripe_page(rbio, rbio->scrubp, pagenr);
  2050. parity = kmap(p);
  2051. if (memcmp(parity, pointers[rbio->scrubp], PAGE_SIZE))
  2052. memcpy(parity, pointers[rbio->scrubp], PAGE_SIZE);
  2053. else
  2054. /* Parity is right, needn't writeback */
  2055. bitmap_clear(rbio->dbitmap, pagenr, 1);
  2056. kunmap(p);
  2057. for (stripe = 0; stripe < rbio->real_stripes; stripe++)
  2058. kunmap(page_in_rbio(rbio, stripe, pagenr, 0));
  2059. }
  2060. __free_page(p_page);
  2061. if (q_page)
  2062. __free_page(q_page);
  2063. writeback:
  2064. /*
  2065. * time to start writing. Make bios for everything from the
  2066. * higher layers (the bio_list in our rbio) and our p/q. Ignore
  2067. * everything else.
  2068. */
  2069. for_each_set_bit(pagenr, rbio->dbitmap, rbio->stripe_npages) {
  2070. struct page *page;
  2071. page = rbio_stripe_page(rbio, rbio->scrubp, pagenr);
  2072. ret = rbio_add_io_page(rbio, &bio_list,
  2073. page, rbio->scrubp, pagenr, rbio->stripe_len);
  2074. if (ret)
  2075. goto cleanup;
  2076. }
  2077. if (!is_replace)
  2078. goto submit_write;
  2079. for_each_set_bit(pagenr, pbitmap, rbio->stripe_npages) {
  2080. struct page *page;
  2081. page = rbio_stripe_page(rbio, rbio->scrubp, pagenr);
  2082. ret = rbio_add_io_page(rbio, &bio_list, page,
  2083. bbio->tgtdev_map[rbio->scrubp],
  2084. pagenr, rbio->stripe_len);
  2085. if (ret)
  2086. goto cleanup;
  2087. }
  2088. submit_write:
  2089. nr_data = bio_list_size(&bio_list);
  2090. if (!nr_data) {
  2091. /* Every parity is right */
  2092. rbio_orig_end_io(rbio, 0);
  2093. return;
  2094. }
  2095. atomic_set(&rbio->stripes_pending, nr_data);
  2096. while (1) {
  2097. bio = bio_list_pop(&bio_list);
  2098. if (!bio)
  2099. break;
  2100. bio->bi_private = rbio;
  2101. bio->bi_end_io = raid_write_end_io;
  2102. bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
  2103. submit_bio(bio);
  2104. }
  2105. return;
  2106. cleanup:
  2107. rbio_orig_end_io(rbio, -EIO);
  2108. }
  2109. static inline int is_data_stripe(struct btrfs_raid_bio *rbio, int stripe)
  2110. {
  2111. if (stripe >= 0 && stripe < rbio->nr_data)
  2112. return 1;
  2113. return 0;
  2114. }
  2115. /*
  2116. * While we're doing the parity check and repair, we could have errors
  2117. * in reading pages off the disk. This checks for errors and if we're
  2118. * not able to read the page it'll trigger parity reconstruction. The
  2119. * parity scrub will be finished after we've reconstructed the failed
  2120. * stripes
  2121. */
  2122. static void validate_rbio_for_parity_scrub(struct btrfs_raid_bio *rbio)
  2123. {
  2124. if (atomic_read(&rbio->error) > rbio->bbio->max_errors)
  2125. goto cleanup;
  2126. if (rbio->faila >= 0 || rbio->failb >= 0) {
  2127. int dfail = 0, failp = -1;
  2128. if (is_data_stripe(rbio, rbio->faila))
  2129. dfail++;
  2130. else if (is_parity_stripe(rbio->faila))
  2131. failp = rbio->faila;
  2132. if (is_data_stripe(rbio, rbio->failb))
  2133. dfail++;
  2134. else if (is_parity_stripe(rbio->failb))
  2135. failp = rbio->failb;
  2136. /*
  2137. * Because we can not use a scrubbing parity to repair
  2138. * the data, so the capability of the repair is declined.
  2139. * (In the case of RAID5, we can not repair anything)
  2140. */
  2141. if (dfail > rbio->bbio->max_errors - 1)
  2142. goto cleanup;
  2143. /*
  2144. * If all data is good, only parity is correctly, just
  2145. * repair the parity.
  2146. */
  2147. if (dfail == 0) {
  2148. finish_parity_scrub(rbio, 0);
  2149. return;
  2150. }
  2151. /*
  2152. * Here means we got one corrupted data stripe and one
  2153. * corrupted parity on RAID6, if the corrupted parity
  2154. * is scrubbing parity, luckily, use the other one to repair
  2155. * the data, or we can not repair the data stripe.
  2156. */
  2157. if (failp != rbio->scrubp)
  2158. goto cleanup;
  2159. __raid_recover_end_io(rbio);
  2160. } else {
  2161. finish_parity_scrub(rbio, 1);
  2162. }
  2163. return;
  2164. cleanup:
  2165. rbio_orig_end_io(rbio, -EIO);
  2166. }
  2167. /*
  2168. * end io for the read phase of the rmw cycle. All the bios here are physical
  2169. * stripe bios we've read from the disk so we can recalculate the parity of the
  2170. * stripe.
  2171. *
  2172. * This will usually kick off finish_rmw once all the bios are read in, but it
  2173. * may trigger parity reconstruction if we had any errors along the way
  2174. */
  2175. static void raid56_parity_scrub_end_io(struct bio *bio)
  2176. {
  2177. struct btrfs_raid_bio *rbio = bio->bi_private;
  2178. if (bio->bi_error)
  2179. fail_bio_stripe(rbio, bio);
  2180. else
  2181. set_bio_pages_uptodate(bio);
  2182. bio_put(bio);
  2183. if (!atomic_dec_and_test(&rbio->stripes_pending))
  2184. return;
  2185. /*
  2186. * this will normally call finish_rmw to start our write
  2187. * but if there are any failed stripes we'll reconstruct
  2188. * from parity first
  2189. */
  2190. validate_rbio_for_parity_scrub(rbio);
  2191. }
  2192. static void raid56_parity_scrub_stripe(struct btrfs_raid_bio *rbio)
  2193. {
  2194. int bios_to_read = 0;
  2195. struct bio_list bio_list;
  2196. int ret;
  2197. int pagenr;
  2198. int stripe;
  2199. struct bio *bio;
  2200. ret = alloc_rbio_essential_pages(rbio);
  2201. if (ret)
  2202. goto cleanup;
  2203. bio_list_init(&bio_list);
  2204. atomic_set(&rbio->error, 0);
  2205. /*
  2206. * build a list of bios to read all the missing parts of this
  2207. * stripe
  2208. */
  2209. for (stripe = 0; stripe < rbio->real_stripes; stripe++) {
  2210. for_each_set_bit(pagenr, rbio->dbitmap, rbio->stripe_npages) {
  2211. struct page *page;
  2212. /*
  2213. * we want to find all the pages missing from
  2214. * the rbio and read them from the disk. If
  2215. * page_in_rbio finds a page in the bio list
  2216. * we don't need to read it off the stripe.
  2217. */
  2218. page = page_in_rbio(rbio, stripe, pagenr, 1);
  2219. if (page)
  2220. continue;
  2221. page = rbio_stripe_page(rbio, stripe, pagenr);
  2222. /*
  2223. * the bio cache may have handed us an uptodate
  2224. * page. If so, be happy and use it
  2225. */
  2226. if (PageUptodate(page))
  2227. continue;
  2228. ret = rbio_add_io_page(rbio, &bio_list, page,
  2229. stripe, pagenr, rbio->stripe_len);
  2230. if (ret)
  2231. goto cleanup;
  2232. }
  2233. }
  2234. bios_to_read = bio_list_size(&bio_list);
  2235. if (!bios_to_read) {
  2236. /*
  2237. * this can happen if others have merged with
  2238. * us, it means there is nothing left to read.
  2239. * But if there are missing devices it may not be
  2240. * safe to do the full stripe write yet.
  2241. */
  2242. goto finish;
  2243. }
  2244. /*
  2245. * the bbio may be freed once we submit the last bio. Make sure
  2246. * not to touch it after that
  2247. */
  2248. atomic_set(&rbio->stripes_pending, bios_to_read);
  2249. while (1) {
  2250. bio = bio_list_pop(&bio_list);
  2251. if (!bio)
  2252. break;
  2253. bio->bi_private = rbio;
  2254. bio->bi_end_io = raid56_parity_scrub_end_io;
  2255. bio_set_op_attrs(bio, REQ_OP_READ, 0);
  2256. btrfs_bio_wq_end_io(rbio->fs_info, bio, BTRFS_WQ_ENDIO_RAID56);
  2257. submit_bio(bio);
  2258. }
  2259. /* the actual write will happen once the reads are done */
  2260. return;
  2261. cleanup:
  2262. rbio_orig_end_io(rbio, -EIO);
  2263. return;
  2264. finish:
  2265. validate_rbio_for_parity_scrub(rbio);
  2266. }
  2267. static void scrub_parity_work(struct btrfs_work *work)
  2268. {
  2269. struct btrfs_raid_bio *rbio;
  2270. rbio = container_of(work, struct btrfs_raid_bio, work);
  2271. raid56_parity_scrub_stripe(rbio);
  2272. }
  2273. static void async_scrub_parity(struct btrfs_raid_bio *rbio)
  2274. {
  2275. btrfs_init_work(&rbio->work, btrfs_rmw_helper,
  2276. scrub_parity_work, NULL, NULL);
  2277. btrfs_queue_work(rbio->fs_info->rmw_workers, &rbio->work);
  2278. }
  2279. void raid56_parity_submit_scrub_rbio(struct btrfs_raid_bio *rbio)
  2280. {
  2281. if (!lock_stripe_add(rbio))
  2282. async_scrub_parity(rbio);
  2283. }
  2284. /* The following code is used for dev replace of a missing RAID 5/6 device. */
  2285. struct btrfs_raid_bio *
  2286. raid56_alloc_missing_rbio(struct btrfs_fs_info *fs_info, struct bio *bio,
  2287. struct btrfs_bio *bbio, u64 length)
  2288. {
  2289. struct btrfs_raid_bio *rbio;
  2290. rbio = alloc_rbio(fs_info, bbio, length);
  2291. if (IS_ERR(rbio))
  2292. return NULL;
  2293. rbio->operation = BTRFS_RBIO_REBUILD_MISSING;
  2294. bio_list_add(&rbio->bio_list, bio);
  2295. /*
  2296. * This is a special bio which is used to hold the completion handler
  2297. * and make the scrub rbio is similar to the other types
  2298. */
  2299. ASSERT(!bio->bi_iter.bi_size);
  2300. rbio->faila = find_logical_bio_stripe(rbio, bio);
  2301. if (rbio->faila == -1) {
  2302. BUG();
  2303. kfree(rbio);
  2304. return NULL;
  2305. }
  2306. /*
  2307. * When we get bbio, we have already increased bio_counter, record it
  2308. * so we can free it at rbio_orig_end_io()
  2309. */
  2310. rbio->generic_bio_cnt = 1;
  2311. return rbio;
  2312. }
  2313. static void missing_raid56_work(struct btrfs_work *work)
  2314. {
  2315. struct btrfs_raid_bio *rbio;
  2316. rbio = container_of(work, struct btrfs_raid_bio, work);
  2317. __raid56_parity_recover(rbio);
  2318. }
  2319. static void async_missing_raid56(struct btrfs_raid_bio *rbio)
  2320. {
  2321. btrfs_init_work(&rbio->work, btrfs_rmw_helper,
  2322. missing_raid56_work, NULL, NULL);
  2323. btrfs_queue_work(rbio->fs_info->rmw_workers, &rbio->work);
  2324. }
  2325. void raid56_submit_missing_rbio(struct btrfs_raid_bio *rbio)
  2326. {
  2327. if (!lock_stripe_add(rbio))
  2328. async_missing_raid56(rbio);
  2329. }