disk-io.c 112 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919920921922923924925926927928929930931932933934935936937938939940941942943944945946947948949950951952953954955956957958959960961962963964965966967968969970971972973974975976977978979980981982983984985986987988989990991992993994995996997998999100010011002100310041005100610071008100910101011101210131014101510161017101810191020102110221023102410251026102710281029103010311032103310341035103610371038103910401041104210431044104510461047104810491050105110521053105410551056105710581059106010611062106310641065106610671068106910701071107210731074107510761077107810791080108110821083108410851086108710881089109010911092109310941095109610971098109911001101110211031104110511061107110811091110111111121113111411151116111711181119112011211122112311241125112611271128112911301131113211331134113511361137113811391140114111421143114411451146114711481149115011511152115311541155115611571158115911601161116211631164116511661167116811691170117111721173117411751176117711781179118011811182118311841185118611871188118911901191119211931194119511961197119811991200120112021203120412051206120712081209121012111212121312141215121612171218121912201221122212231224122512261227122812291230123112321233123412351236123712381239124012411242124312441245124612471248124912501251125212531254125512561257125812591260126112621263126412651266126712681269127012711272127312741275127612771278127912801281128212831284128512861287128812891290129112921293129412951296129712981299130013011302130313041305130613071308130913101311131213131314131513161317131813191320132113221323132413251326132713281329133013311332133313341335133613371338133913401341134213431344134513461347134813491350135113521353135413551356135713581359136013611362136313641365136613671368136913701371137213731374137513761377137813791380138113821383138413851386138713881389139013911392139313941395139613971398139914001401140214031404140514061407140814091410141114121413141414151416141714181419142014211422142314241425142614271428142914301431143214331434143514361437143814391440144114421443144414451446144714481449145014511452145314541455145614571458145914601461146214631464146514661467146814691470147114721473147414751476147714781479148014811482148314841485148614871488148914901491149214931494149514961497149814991500150115021503150415051506150715081509151015111512151315141515151615171518151915201521152215231524152515261527152815291530153115321533153415351536153715381539154015411542154315441545154615471548154915501551155215531554155515561557155815591560156115621563156415651566156715681569157015711572157315741575157615771578157915801581158215831584158515861587158815891590159115921593159415951596159715981599160016011602160316041605160616071608160916101611161216131614161516161617161816191620162116221623162416251626162716281629163016311632163316341635163616371638163916401641164216431644164516461647164816491650165116521653165416551656165716581659166016611662166316641665166616671668166916701671167216731674167516761677167816791680168116821683168416851686168716881689169016911692169316941695169616971698169917001701170217031704170517061707170817091710171117121713171417151716171717181719172017211722172317241725172617271728172917301731173217331734173517361737173817391740174117421743174417451746174717481749175017511752175317541755175617571758175917601761176217631764176517661767176817691770177117721773177417751776177717781779178017811782178317841785178617871788178917901791179217931794179517961797179817991800180118021803180418051806180718081809181018111812181318141815181618171818181918201821182218231824182518261827182818291830183118321833183418351836183718381839184018411842184318441845184618471848184918501851185218531854185518561857185818591860186118621863186418651866186718681869187018711872187318741875187618771878187918801881188218831884188518861887188818891890189118921893189418951896189718981899190019011902190319041905190619071908190919101911191219131914191519161917191819191920192119221923192419251926192719281929193019311932193319341935193619371938193919401941194219431944194519461947194819491950195119521953195419551956195719581959196019611962196319641965196619671968196919701971197219731974197519761977197819791980198119821983198419851986198719881989199019911992199319941995199619971998199920002001200220032004200520062007200820092010201120122013201420152016201720182019202020212022202320242025202620272028202920302031203220332034203520362037203820392040204120422043204420452046204720482049205020512052205320542055205620572058205920602061206220632064206520662067206820692070207120722073207420752076207720782079208020812082208320842085208620872088208920902091209220932094209520962097209820992100210121022103210421052106210721082109211021112112211321142115211621172118211921202121212221232124212521262127212821292130213121322133213421352136213721382139214021412142214321442145214621472148214921502151215221532154215521562157215821592160216121622163216421652166216721682169217021712172217321742175217621772178217921802181218221832184218521862187218821892190219121922193219421952196219721982199220022012202220322042205220622072208220922102211221222132214221522162217221822192220222122222223222422252226222722282229223022312232223322342235223622372238223922402241224222432244224522462247224822492250225122522253225422552256225722582259226022612262226322642265226622672268226922702271227222732274227522762277227822792280228122822283228422852286228722882289229022912292229322942295229622972298229923002301230223032304230523062307230823092310231123122313231423152316231723182319232023212322232323242325232623272328232923302331233223332334233523362337233823392340234123422343234423452346234723482349235023512352235323542355235623572358235923602361236223632364236523662367236823692370237123722373237423752376237723782379238023812382238323842385238623872388238923902391239223932394239523962397239823992400240124022403240424052406240724082409241024112412241324142415241624172418241924202421242224232424242524262427242824292430243124322433243424352436243724382439244024412442244324442445244624472448244924502451245224532454245524562457245824592460246124622463246424652466246724682469247024712472247324742475247624772478247924802481248224832484248524862487248824892490249124922493249424952496249724982499250025012502250325042505250625072508250925102511251225132514251525162517251825192520252125222523252425252526252725282529253025312532253325342535253625372538253925402541254225432544254525462547254825492550255125522553255425552556255725582559256025612562256325642565256625672568256925702571257225732574257525762577257825792580258125822583258425852586258725882589259025912592259325942595259625972598259926002601260226032604260526062607260826092610261126122613261426152616261726182619262026212622262326242625262626272628262926302631263226332634263526362637263826392640264126422643264426452646264726482649265026512652265326542655265626572658265926602661266226632664266526662667266826692670267126722673267426752676267726782679268026812682268326842685268626872688268926902691269226932694269526962697269826992700270127022703270427052706270727082709271027112712271327142715271627172718271927202721272227232724272527262727272827292730273127322733273427352736273727382739274027412742274327442745274627472748274927502751275227532754275527562757275827592760276127622763276427652766276727682769277027712772277327742775277627772778277927802781278227832784278527862787278827892790279127922793279427952796279727982799280028012802280328042805280628072808280928102811281228132814281528162817281828192820282128222823282428252826282728282829283028312832283328342835283628372838283928402841284228432844284528462847284828492850285128522853285428552856285728582859286028612862286328642865286628672868286928702871287228732874287528762877287828792880288128822883288428852886288728882889289028912892289328942895289628972898289929002901290229032904290529062907290829092910291129122913291429152916291729182919292029212922292329242925292629272928292929302931293229332934293529362937293829392940294129422943294429452946294729482949295029512952295329542955295629572958295929602961296229632964296529662967296829692970297129722973297429752976297729782979298029812982298329842985298629872988298929902991299229932994299529962997299829993000300130023003300430053006300730083009301030113012301330143015301630173018301930203021302230233024302530263027302830293030303130323033303430353036303730383039304030413042304330443045304630473048304930503051305230533054305530563057305830593060306130623063306430653066306730683069307030713072307330743075307630773078307930803081308230833084308530863087308830893090309130923093309430953096309730983099310031013102310331043105310631073108310931103111311231133114311531163117311831193120312131223123312431253126312731283129313031313132313331343135313631373138313931403141314231433144314531463147314831493150315131523153315431553156315731583159316031613162316331643165316631673168316931703171317231733174317531763177317831793180318131823183318431853186318731883189319031913192319331943195319631973198319932003201320232033204320532063207320832093210321132123213321432153216321732183219322032213222322332243225322632273228322932303231323232333234323532363237323832393240324132423243324432453246324732483249325032513252325332543255325632573258325932603261326232633264326532663267326832693270327132723273327432753276327732783279328032813282328332843285328632873288328932903291329232933294329532963297329832993300330133023303330433053306330733083309331033113312331333143315331633173318331933203321332233233324332533263327332833293330333133323333333433353336333733383339334033413342334333443345334633473348334933503351335233533354335533563357335833593360336133623363336433653366336733683369337033713372337333743375337633773378337933803381338233833384338533863387338833893390339133923393339433953396339733983399340034013402340334043405340634073408340934103411341234133414341534163417341834193420342134223423342434253426342734283429343034313432343334343435343634373438343934403441344234433444344534463447344834493450345134523453345434553456345734583459346034613462346334643465346634673468346934703471347234733474347534763477347834793480348134823483348434853486348734883489349034913492349334943495349634973498349935003501350235033504350535063507350835093510351135123513351435153516351735183519352035213522352335243525352635273528352935303531353235333534353535363537353835393540354135423543354435453546354735483549355035513552355335543555355635573558355935603561356235633564356535663567356835693570357135723573357435753576357735783579358035813582358335843585358635873588358935903591359235933594359535963597359835993600360136023603360436053606360736083609361036113612361336143615361636173618361936203621362236233624362536263627362836293630363136323633363436353636363736383639364036413642364336443645364636473648364936503651365236533654365536563657365836593660366136623663366436653666366736683669367036713672367336743675367636773678367936803681368236833684368536863687368836893690369136923693369436953696369736983699370037013702370337043705370637073708370937103711371237133714371537163717371837193720372137223723372437253726372737283729373037313732373337343735373637373738373937403741374237433744374537463747374837493750375137523753375437553756375737583759376037613762376337643765376637673768376937703771377237733774377537763777377837793780378137823783378437853786378737883789379037913792379337943795379637973798379938003801380238033804380538063807380838093810381138123813381438153816381738183819382038213822382338243825382638273828382938303831383238333834383538363837383838393840384138423843384438453846384738483849385038513852385338543855385638573858385938603861386238633864386538663867386838693870387138723873387438753876387738783879388038813882388338843885388638873888388938903891389238933894389538963897389838993900390139023903390439053906390739083909391039113912391339143915391639173918391939203921392239233924392539263927392839293930393139323933393439353936393739383939394039413942394339443945394639473948394939503951395239533954395539563957395839593960396139623963396439653966396739683969397039713972397339743975397639773978397939803981398239833984398539863987398839893990399139923993399439953996399739983999400040014002400340044005400640074008400940104011401240134014401540164017401840194020402140224023402440254026402740284029403040314032403340344035403640374038403940404041404240434044404540464047404840494050405140524053405440554056405740584059406040614062406340644065406640674068406940704071407240734074407540764077407840794080408140824083408440854086408740884089409040914092409340944095409640974098409941004101410241034104410541064107410841094110411141124113411441154116411741184119412041214122412341244125412641274128412941304131413241334134413541364137413841394140414141424143414441454146414741484149415041514152415341544155415641574158415941604161416241634164
  1. /*
  2. * Copyright (C) 2007 Oracle. All rights reserved.
  3. *
  4. * This program is free software; you can redistribute it and/or
  5. * modify it under the terms of the GNU General Public
  6. * License v2 as published by the Free Software Foundation.
  7. *
  8. * This program is distributed in the hope that it will be useful,
  9. * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10. * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  11. * General Public License for more details.
  12. *
  13. * You should have received a copy of the GNU General Public
  14. * License along with this program; if not, write to the
  15. * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16. * Boston, MA 021110-1307, USA.
  17. */
  18. #include <linux/fs.h>
  19. #include <linux/blkdev.h>
  20. #include <linux/scatterlist.h>
  21. #include <linux/swap.h>
  22. #include <linux/radix-tree.h>
  23. #include <linux/writeback.h>
  24. #include <linux/buffer_head.h>
  25. #include <linux/workqueue.h>
  26. #include <linux/kthread.h>
  27. #include <linux/freezer.h>
  28. #include <linux/crc32c.h>
  29. #include <linux/slab.h>
  30. #include <linux/migrate.h>
  31. #include <linux/ratelimit.h>
  32. #include <linux/uuid.h>
  33. #include <linux/semaphore.h>
  34. #include <asm/unaligned.h>
  35. #include "compat.h"
  36. #include "ctree.h"
  37. #include "disk-io.h"
  38. #include "transaction.h"
  39. #include "btrfs_inode.h"
  40. #include "volumes.h"
  41. #include "print-tree.h"
  42. #include "async-thread.h"
  43. #include "locking.h"
  44. #include "tree-log.h"
  45. #include "free-space-cache.h"
  46. #include "inode-map.h"
  47. #include "check-integrity.h"
  48. #include "rcu-string.h"
  49. #include "dev-replace.h"
  50. #include "raid56.h"
  51. #ifdef CONFIG_X86
  52. #include <asm/cpufeature.h>
  53. #endif
  54. static struct extent_io_ops btree_extent_io_ops;
  55. static void end_workqueue_fn(struct btrfs_work *work);
  56. static void free_fs_root(struct btrfs_root *root);
  57. static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
  58. int read_only);
  59. static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
  60. struct btrfs_root *root);
  61. static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
  62. static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
  63. struct btrfs_root *root);
  64. static void btrfs_evict_pending_snapshots(struct btrfs_transaction *t);
  65. static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
  66. static int btrfs_destroy_marked_extents(struct btrfs_root *root,
  67. struct extent_io_tree *dirty_pages,
  68. int mark);
  69. static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
  70. struct extent_io_tree *pinned_extents);
  71. static int btrfs_cleanup_transaction(struct btrfs_root *root);
  72. static void btrfs_error_commit_super(struct btrfs_root *root);
  73. /*
  74. * end_io_wq structs are used to do processing in task context when an IO is
  75. * complete. This is used during reads to verify checksums, and it is used
  76. * by writes to insert metadata for new file extents after IO is complete.
  77. */
  78. struct end_io_wq {
  79. struct bio *bio;
  80. bio_end_io_t *end_io;
  81. void *private;
  82. struct btrfs_fs_info *info;
  83. int error;
  84. int metadata;
  85. struct list_head list;
  86. struct btrfs_work work;
  87. };
  88. /*
  89. * async submit bios are used to offload expensive checksumming
  90. * onto the worker threads. They checksum file and metadata bios
  91. * just before they are sent down the IO stack.
  92. */
  93. struct async_submit_bio {
  94. struct inode *inode;
  95. struct bio *bio;
  96. struct list_head list;
  97. extent_submit_bio_hook_t *submit_bio_start;
  98. extent_submit_bio_hook_t *submit_bio_done;
  99. int rw;
  100. int mirror_num;
  101. unsigned long bio_flags;
  102. /*
  103. * bio_offset is optional, can be used if the pages in the bio
  104. * can't tell us where in the file the bio should go
  105. */
  106. u64 bio_offset;
  107. struct btrfs_work work;
  108. int error;
  109. };
  110. /*
  111. * Lockdep class keys for extent_buffer->lock's in this root. For a given
  112. * eb, the lockdep key is determined by the btrfs_root it belongs to and
  113. * the level the eb occupies in the tree.
  114. *
  115. * Different roots are used for different purposes and may nest inside each
  116. * other and they require separate keysets. As lockdep keys should be
  117. * static, assign keysets according to the purpose of the root as indicated
  118. * by btrfs_root->objectid. This ensures that all special purpose roots
  119. * have separate keysets.
  120. *
  121. * Lock-nesting across peer nodes is always done with the immediate parent
  122. * node locked thus preventing deadlock. As lockdep doesn't know this, use
  123. * subclass to avoid triggering lockdep warning in such cases.
  124. *
  125. * The key is set by the readpage_end_io_hook after the buffer has passed
  126. * csum validation but before the pages are unlocked. It is also set by
  127. * btrfs_init_new_buffer on freshly allocated blocks.
  128. *
  129. * We also add a check to make sure the highest level of the tree is the
  130. * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
  131. * needs update as well.
  132. */
  133. #ifdef CONFIG_DEBUG_LOCK_ALLOC
  134. # if BTRFS_MAX_LEVEL != 8
  135. # error
  136. # endif
  137. static struct btrfs_lockdep_keyset {
  138. u64 id; /* root objectid */
  139. const char *name_stem; /* lock name stem */
  140. char names[BTRFS_MAX_LEVEL + 1][20];
  141. struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
  142. } btrfs_lockdep_keysets[] = {
  143. { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
  144. { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
  145. { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
  146. { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
  147. { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
  148. { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
  149. { .id = BTRFS_QUOTA_TREE_OBJECTID, .name_stem = "quota" },
  150. { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
  151. { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
  152. { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
  153. { .id = BTRFS_UUID_TREE_OBJECTID, .name_stem = "uuid" },
  154. { .id = 0, .name_stem = "tree" },
  155. };
  156. void __init btrfs_init_lockdep(void)
  157. {
  158. int i, j;
  159. /* initialize lockdep class names */
  160. for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
  161. struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
  162. for (j = 0; j < ARRAY_SIZE(ks->names); j++)
  163. snprintf(ks->names[j], sizeof(ks->names[j]),
  164. "btrfs-%s-%02d", ks->name_stem, j);
  165. }
  166. }
  167. void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
  168. int level)
  169. {
  170. struct btrfs_lockdep_keyset *ks;
  171. BUG_ON(level >= ARRAY_SIZE(ks->keys));
  172. /* find the matching keyset, id 0 is the default entry */
  173. for (ks = btrfs_lockdep_keysets; ks->id; ks++)
  174. if (ks->id == objectid)
  175. break;
  176. lockdep_set_class_and_name(&eb->lock,
  177. &ks->keys[level], ks->names[level]);
  178. }
  179. #endif
  180. /*
  181. * extents on the btree inode are pretty simple, there's one extent
  182. * that covers the entire device
  183. */
  184. static struct extent_map *btree_get_extent(struct inode *inode,
  185. struct page *page, size_t pg_offset, u64 start, u64 len,
  186. int create)
  187. {
  188. struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
  189. struct extent_map *em;
  190. int ret;
  191. read_lock(&em_tree->lock);
  192. em = lookup_extent_mapping(em_tree, start, len);
  193. if (em) {
  194. em->bdev =
  195. BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
  196. read_unlock(&em_tree->lock);
  197. goto out;
  198. }
  199. read_unlock(&em_tree->lock);
  200. em = alloc_extent_map();
  201. if (!em) {
  202. em = ERR_PTR(-ENOMEM);
  203. goto out;
  204. }
  205. em->start = 0;
  206. em->len = (u64)-1;
  207. em->block_len = (u64)-1;
  208. em->block_start = 0;
  209. em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
  210. write_lock(&em_tree->lock);
  211. ret = add_extent_mapping(em_tree, em, 0);
  212. if (ret == -EEXIST) {
  213. free_extent_map(em);
  214. em = lookup_extent_mapping(em_tree, start, len);
  215. if (!em)
  216. em = ERR_PTR(-EIO);
  217. } else if (ret) {
  218. free_extent_map(em);
  219. em = ERR_PTR(ret);
  220. }
  221. write_unlock(&em_tree->lock);
  222. out:
  223. return em;
  224. }
  225. u32 btrfs_csum_data(char *data, u32 seed, size_t len)
  226. {
  227. return crc32c(seed, data, len);
  228. }
  229. void btrfs_csum_final(u32 crc, char *result)
  230. {
  231. put_unaligned_le32(~crc, result);
  232. }
  233. /*
  234. * compute the csum for a btree block, and either verify it or write it
  235. * into the csum field of the block.
  236. */
  237. static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
  238. int verify)
  239. {
  240. u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
  241. char *result = NULL;
  242. unsigned long len;
  243. unsigned long cur_len;
  244. unsigned long offset = BTRFS_CSUM_SIZE;
  245. char *kaddr;
  246. unsigned long map_start;
  247. unsigned long map_len;
  248. int err;
  249. u32 crc = ~(u32)0;
  250. unsigned long inline_result;
  251. len = buf->len - offset;
  252. while (len > 0) {
  253. err = map_private_extent_buffer(buf, offset, 32,
  254. &kaddr, &map_start, &map_len);
  255. if (err)
  256. return 1;
  257. cur_len = min(len, map_len - (offset - map_start));
  258. crc = btrfs_csum_data(kaddr + offset - map_start,
  259. crc, cur_len);
  260. len -= cur_len;
  261. offset += cur_len;
  262. }
  263. if (csum_size > sizeof(inline_result)) {
  264. result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
  265. if (!result)
  266. return 1;
  267. } else {
  268. result = (char *)&inline_result;
  269. }
  270. btrfs_csum_final(crc, result);
  271. if (verify) {
  272. if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
  273. u32 val;
  274. u32 found = 0;
  275. memcpy(&found, result, csum_size);
  276. read_extent_buffer(buf, &val, 0, csum_size);
  277. printk_ratelimited(KERN_INFO "btrfs: %s checksum verify "
  278. "failed on %llu wanted %X found %X "
  279. "level %d\n",
  280. root->fs_info->sb->s_id, buf->start,
  281. val, found, btrfs_header_level(buf));
  282. if (result != (char *)&inline_result)
  283. kfree(result);
  284. return 1;
  285. }
  286. } else {
  287. write_extent_buffer(buf, result, 0, csum_size);
  288. }
  289. if (result != (char *)&inline_result)
  290. kfree(result);
  291. return 0;
  292. }
  293. /*
  294. * we can't consider a given block up to date unless the transid of the
  295. * block matches the transid in the parent node's pointer. This is how we
  296. * detect blocks that either didn't get written at all or got written
  297. * in the wrong place.
  298. */
  299. static int verify_parent_transid(struct extent_io_tree *io_tree,
  300. struct extent_buffer *eb, u64 parent_transid,
  301. int atomic)
  302. {
  303. struct extent_state *cached_state = NULL;
  304. int ret;
  305. if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
  306. return 0;
  307. if (atomic)
  308. return -EAGAIN;
  309. lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
  310. 0, &cached_state);
  311. if (extent_buffer_uptodate(eb) &&
  312. btrfs_header_generation(eb) == parent_transid) {
  313. ret = 0;
  314. goto out;
  315. }
  316. printk_ratelimited("parent transid verify failed on %llu wanted %llu "
  317. "found %llu\n",
  318. eb->start, parent_transid, btrfs_header_generation(eb));
  319. ret = 1;
  320. clear_extent_buffer_uptodate(eb);
  321. out:
  322. unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
  323. &cached_state, GFP_NOFS);
  324. return ret;
  325. }
  326. /*
  327. * Return 0 if the superblock checksum type matches the checksum value of that
  328. * algorithm. Pass the raw disk superblock data.
  329. */
  330. static int btrfs_check_super_csum(char *raw_disk_sb)
  331. {
  332. struct btrfs_super_block *disk_sb =
  333. (struct btrfs_super_block *)raw_disk_sb;
  334. u16 csum_type = btrfs_super_csum_type(disk_sb);
  335. int ret = 0;
  336. if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
  337. u32 crc = ~(u32)0;
  338. const int csum_size = sizeof(crc);
  339. char result[csum_size];
  340. /*
  341. * The super_block structure does not span the whole
  342. * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
  343. * is filled with zeros and is included in the checkum.
  344. */
  345. crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
  346. crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
  347. btrfs_csum_final(crc, result);
  348. if (memcmp(raw_disk_sb, result, csum_size))
  349. ret = 1;
  350. if (ret && btrfs_super_generation(disk_sb) < 10) {
  351. printk(KERN_WARNING "btrfs: super block crcs don't match, older mkfs detected\n");
  352. ret = 0;
  353. }
  354. }
  355. if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
  356. printk(KERN_ERR "btrfs: unsupported checksum algorithm %u\n",
  357. csum_type);
  358. ret = 1;
  359. }
  360. return ret;
  361. }
  362. /*
  363. * helper to read a given tree block, doing retries as required when
  364. * the checksums don't match and we have alternate mirrors to try.
  365. */
  366. static int btree_read_extent_buffer_pages(struct btrfs_root *root,
  367. struct extent_buffer *eb,
  368. u64 start, u64 parent_transid)
  369. {
  370. struct extent_io_tree *io_tree;
  371. int failed = 0;
  372. int ret;
  373. int num_copies = 0;
  374. int mirror_num = 0;
  375. int failed_mirror = 0;
  376. clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
  377. io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
  378. while (1) {
  379. ret = read_extent_buffer_pages(io_tree, eb, start,
  380. WAIT_COMPLETE,
  381. btree_get_extent, mirror_num);
  382. if (!ret) {
  383. if (!verify_parent_transid(io_tree, eb,
  384. parent_transid, 0))
  385. break;
  386. else
  387. ret = -EIO;
  388. }
  389. /*
  390. * This buffer's crc is fine, but its contents are corrupted, so
  391. * there is no reason to read the other copies, they won't be
  392. * any less wrong.
  393. */
  394. if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
  395. break;
  396. num_copies = btrfs_num_copies(root->fs_info,
  397. eb->start, eb->len);
  398. if (num_copies == 1)
  399. break;
  400. if (!failed_mirror) {
  401. failed = 1;
  402. failed_mirror = eb->read_mirror;
  403. }
  404. mirror_num++;
  405. if (mirror_num == failed_mirror)
  406. mirror_num++;
  407. if (mirror_num > num_copies)
  408. break;
  409. }
  410. if (failed && !ret && failed_mirror)
  411. repair_eb_io_failure(root, eb, failed_mirror);
  412. return ret;
  413. }
  414. /*
  415. * checksum a dirty tree block before IO. This has extra checks to make sure
  416. * we only fill in the checksum field in the first page of a multi-page block
  417. */
  418. static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
  419. {
  420. struct extent_io_tree *tree;
  421. u64 start = page_offset(page);
  422. u64 found_start;
  423. struct extent_buffer *eb;
  424. tree = &BTRFS_I(page->mapping->host)->io_tree;
  425. eb = (struct extent_buffer *)page->private;
  426. if (page != eb->pages[0])
  427. return 0;
  428. found_start = btrfs_header_bytenr(eb);
  429. if (found_start != start) {
  430. WARN_ON(1);
  431. return 0;
  432. }
  433. if (!PageUptodate(page)) {
  434. WARN_ON(1);
  435. return 0;
  436. }
  437. csum_tree_block(root, eb, 0);
  438. return 0;
  439. }
  440. static int check_tree_block_fsid(struct btrfs_root *root,
  441. struct extent_buffer *eb)
  442. {
  443. struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
  444. u8 fsid[BTRFS_UUID_SIZE];
  445. int ret = 1;
  446. read_extent_buffer(eb, fsid, btrfs_header_fsid(eb), BTRFS_FSID_SIZE);
  447. while (fs_devices) {
  448. if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
  449. ret = 0;
  450. break;
  451. }
  452. fs_devices = fs_devices->seed;
  453. }
  454. return ret;
  455. }
  456. #define CORRUPT(reason, eb, root, slot) \
  457. printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
  458. "root=%llu, slot=%d\n", reason, \
  459. btrfs_header_bytenr(eb), root->objectid, slot)
  460. static noinline int check_leaf(struct btrfs_root *root,
  461. struct extent_buffer *leaf)
  462. {
  463. struct btrfs_key key;
  464. struct btrfs_key leaf_key;
  465. u32 nritems = btrfs_header_nritems(leaf);
  466. int slot;
  467. if (nritems == 0)
  468. return 0;
  469. /* Check the 0 item */
  470. if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
  471. BTRFS_LEAF_DATA_SIZE(root)) {
  472. CORRUPT("invalid item offset size pair", leaf, root, 0);
  473. return -EIO;
  474. }
  475. /*
  476. * Check to make sure each items keys are in the correct order and their
  477. * offsets make sense. We only have to loop through nritems-1 because
  478. * we check the current slot against the next slot, which verifies the
  479. * next slot's offset+size makes sense and that the current's slot
  480. * offset is correct.
  481. */
  482. for (slot = 0; slot < nritems - 1; slot++) {
  483. btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
  484. btrfs_item_key_to_cpu(leaf, &key, slot + 1);
  485. /* Make sure the keys are in the right order */
  486. if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
  487. CORRUPT("bad key order", leaf, root, slot);
  488. return -EIO;
  489. }
  490. /*
  491. * Make sure the offset and ends are right, remember that the
  492. * item data starts at the end of the leaf and grows towards the
  493. * front.
  494. */
  495. if (btrfs_item_offset_nr(leaf, slot) !=
  496. btrfs_item_end_nr(leaf, slot + 1)) {
  497. CORRUPT("slot offset bad", leaf, root, slot);
  498. return -EIO;
  499. }
  500. /*
  501. * Check to make sure that we don't point outside of the leaf,
  502. * just incase all the items are consistent to eachother, but
  503. * all point outside of the leaf.
  504. */
  505. if (btrfs_item_end_nr(leaf, slot) >
  506. BTRFS_LEAF_DATA_SIZE(root)) {
  507. CORRUPT("slot end outside of leaf", leaf, root, slot);
  508. return -EIO;
  509. }
  510. }
  511. return 0;
  512. }
  513. static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
  514. u64 phy_offset, struct page *page,
  515. u64 start, u64 end, int mirror)
  516. {
  517. struct extent_io_tree *tree;
  518. u64 found_start;
  519. int found_level;
  520. struct extent_buffer *eb;
  521. struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
  522. int ret = 0;
  523. int reads_done;
  524. if (!page->private)
  525. goto out;
  526. tree = &BTRFS_I(page->mapping->host)->io_tree;
  527. eb = (struct extent_buffer *)page->private;
  528. /* the pending IO might have been the only thing that kept this buffer
  529. * in memory. Make sure we have a ref for all this other checks
  530. */
  531. extent_buffer_get(eb);
  532. reads_done = atomic_dec_and_test(&eb->io_pages);
  533. if (!reads_done)
  534. goto err;
  535. eb->read_mirror = mirror;
  536. if (test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
  537. ret = -EIO;
  538. goto err;
  539. }
  540. found_start = btrfs_header_bytenr(eb);
  541. if (found_start != eb->start) {
  542. printk_ratelimited(KERN_INFO "btrfs bad tree block start "
  543. "%llu %llu\n",
  544. found_start, eb->start);
  545. ret = -EIO;
  546. goto err;
  547. }
  548. if (check_tree_block_fsid(root, eb)) {
  549. printk_ratelimited(KERN_INFO "btrfs bad fsid on block %llu\n",
  550. eb->start);
  551. ret = -EIO;
  552. goto err;
  553. }
  554. found_level = btrfs_header_level(eb);
  555. if (found_level >= BTRFS_MAX_LEVEL) {
  556. btrfs_info(root->fs_info, "bad tree block level %d\n",
  557. (int)btrfs_header_level(eb));
  558. ret = -EIO;
  559. goto err;
  560. }
  561. btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
  562. eb, found_level);
  563. ret = csum_tree_block(root, eb, 1);
  564. if (ret) {
  565. ret = -EIO;
  566. goto err;
  567. }
  568. /*
  569. * If this is a leaf block and it is corrupt, set the corrupt bit so
  570. * that we don't try and read the other copies of this block, just
  571. * return -EIO.
  572. */
  573. if (found_level == 0 && check_leaf(root, eb)) {
  574. set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
  575. ret = -EIO;
  576. }
  577. if (!ret)
  578. set_extent_buffer_uptodate(eb);
  579. err:
  580. if (reads_done &&
  581. test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
  582. btree_readahead_hook(root, eb, eb->start, ret);
  583. if (ret) {
  584. /*
  585. * our io error hook is going to dec the io pages
  586. * again, we have to make sure it has something
  587. * to decrement
  588. */
  589. atomic_inc(&eb->io_pages);
  590. clear_extent_buffer_uptodate(eb);
  591. }
  592. free_extent_buffer(eb);
  593. out:
  594. return ret;
  595. }
  596. static int btree_io_failed_hook(struct page *page, int failed_mirror)
  597. {
  598. struct extent_buffer *eb;
  599. struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
  600. eb = (struct extent_buffer *)page->private;
  601. set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
  602. eb->read_mirror = failed_mirror;
  603. atomic_dec(&eb->io_pages);
  604. if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
  605. btree_readahead_hook(root, eb, eb->start, -EIO);
  606. return -EIO; /* we fixed nothing */
  607. }
  608. static void end_workqueue_bio(struct bio *bio, int err)
  609. {
  610. struct end_io_wq *end_io_wq = bio->bi_private;
  611. struct btrfs_fs_info *fs_info;
  612. fs_info = end_io_wq->info;
  613. end_io_wq->error = err;
  614. end_io_wq->work.func = end_workqueue_fn;
  615. end_io_wq->work.flags = 0;
  616. if (bio->bi_rw & REQ_WRITE) {
  617. if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
  618. btrfs_queue_worker(&fs_info->endio_meta_write_workers,
  619. &end_io_wq->work);
  620. else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
  621. btrfs_queue_worker(&fs_info->endio_freespace_worker,
  622. &end_io_wq->work);
  623. else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
  624. btrfs_queue_worker(&fs_info->endio_raid56_workers,
  625. &end_io_wq->work);
  626. else
  627. btrfs_queue_worker(&fs_info->endio_write_workers,
  628. &end_io_wq->work);
  629. } else {
  630. if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
  631. btrfs_queue_worker(&fs_info->endio_raid56_workers,
  632. &end_io_wq->work);
  633. else if (end_io_wq->metadata)
  634. btrfs_queue_worker(&fs_info->endio_meta_workers,
  635. &end_io_wq->work);
  636. else
  637. btrfs_queue_worker(&fs_info->endio_workers,
  638. &end_io_wq->work);
  639. }
  640. }
  641. /*
  642. * For the metadata arg you want
  643. *
  644. * 0 - if data
  645. * 1 - if normal metadta
  646. * 2 - if writing to the free space cache area
  647. * 3 - raid parity work
  648. */
  649. int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
  650. int metadata)
  651. {
  652. struct end_io_wq *end_io_wq;
  653. end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
  654. if (!end_io_wq)
  655. return -ENOMEM;
  656. end_io_wq->private = bio->bi_private;
  657. end_io_wq->end_io = bio->bi_end_io;
  658. end_io_wq->info = info;
  659. end_io_wq->error = 0;
  660. end_io_wq->bio = bio;
  661. end_io_wq->metadata = metadata;
  662. bio->bi_private = end_io_wq;
  663. bio->bi_end_io = end_workqueue_bio;
  664. return 0;
  665. }
  666. unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
  667. {
  668. unsigned long limit = min_t(unsigned long,
  669. info->workers.max_workers,
  670. info->fs_devices->open_devices);
  671. return 256 * limit;
  672. }
  673. static void run_one_async_start(struct btrfs_work *work)
  674. {
  675. struct async_submit_bio *async;
  676. int ret;
  677. async = container_of(work, struct async_submit_bio, work);
  678. ret = async->submit_bio_start(async->inode, async->rw, async->bio,
  679. async->mirror_num, async->bio_flags,
  680. async->bio_offset);
  681. if (ret)
  682. async->error = ret;
  683. }
  684. static void run_one_async_done(struct btrfs_work *work)
  685. {
  686. struct btrfs_fs_info *fs_info;
  687. struct async_submit_bio *async;
  688. int limit;
  689. async = container_of(work, struct async_submit_bio, work);
  690. fs_info = BTRFS_I(async->inode)->root->fs_info;
  691. limit = btrfs_async_submit_limit(fs_info);
  692. limit = limit * 2 / 3;
  693. if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
  694. waitqueue_active(&fs_info->async_submit_wait))
  695. wake_up(&fs_info->async_submit_wait);
  696. /* If an error occured we just want to clean up the bio and move on */
  697. if (async->error) {
  698. bio_endio(async->bio, async->error);
  699. return;
  700. }
  701. async->submit_bio_done(async->inode, async->rw, async->bio,
  702. async->mirror_num, async->bio_flags,
  703. async->bio_offset);
  704. }
  705. static void run_one_async_free(struct btrfs_work *work)
  706. {
  707. struct async_submit_bio *async;
  708. async = container_of(work, struct async_submit_bio, work);
  709. kfree(async);
  710. }
  711. int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
  712. int rw, struct bio *bio, int mirror_num,
  713. unsigned long bio_flags,
  714. u64 bio_offset,
  715. extent_submit_bio_hook_t *submit_bio_start,
  716. extent_submit_bio_hook_t *submit_bio_done)
  717. {
  718. struct async_submit_bio *async;
  719. async = kmalloc(sizeof(*async), GFP_NOFS);
  720. if (!async)
  721. return -ENOMEM;
  722. async->inode = inode;
  723. async->rw = rw;
  724. async->bio = bio;
  725. async->mirror_num = mirror_num;
  726. async->submit_bio_start = submit_bio_start;
  727. async->submit_bio_done = submit_bio_done;
  728. async->work.func = run_one_async_start;
  729. async->work.ordered_func = run_one_async_done;
  730. async->work.ordered_free = run_one_async_free;
  731. async->work.flags = 0;
  732. async->bio_flags = bio_flags;
  733. async->bio_offset = bio_offset;
  734. async->error = 0;
  735. atomic_inc(&fs_info->nr_async_submits);
  736. if (rw & REQ_SYNC)
  737. btrfs_set_work_high_prio(&async->work);
  738. btrfs_queue_worker(&fs_info->workers, &async->work);
  739. while (atomic_read(&fs_info->async_submit_draining) &&
  740. atomic_read(&fs_info->nr_async_submits)) {
  741. wait_event(fs_info->async_submit_wait,
  742. (atomic_read(&fs_info->nr_async_submits) == 0));
  743. }
  744. return 0;
  745. }
  746. static int btree_csum_one_bio(struct bio *bio)
  747. {
  748. struct bio_vec *bvec = bio->bi_io_vec;
  749. int bio_index = 0;
  750. struct btrfs_root *root;
  751. int ret = 0;
  752. WARN_ON(bio->bi_vcnt <= 0);
  753. while (bio_index < bio->bi_vcnt) {
  754. root = BTRFS_I(bvec->bv_page->mapping->host)->root;
  755. ret = csum_dirty_buffer(root, bvec->bv_page);
  756. if (ret)
  757. break;
  758. bio_index++;
  759. bvec++;
  760. }
  761. return ret;
  762. }
  763. static int __btree_submit_bio_start(struct inode *inode, int rw,
  764. struct bio *bio, int mirror_num,
  765. unsigned long bio_flags,
  766. u64 bio_offset)
  767. {
  768. /*
  769. * when we're called for a write, we're already in the async
  770. * submission context. Just jump into btrfs_map_bio
  771. */
  772. return btree_csum_one_bio(bio);
  773. }
  774. static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
  775. int mirror_num, unsigned long bio_flags,
  776. u64 bio_offset)
  777. {
  778. int ret;
  779. /*
  780. * when we're called for a write, we're already in the async
  781. * submission context. Just jump into btrfs_map_bio
  782. */
  783. ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
  784. if (ret)
  785. bio_endio(bio, ret);
  786. return ret;
  787. }
  788. static int check_async_write(struct inode *inode, unsigned long bio_flags)
  789. {
  790. if (bio_flags & EXTENT_BIO_TREE_LOG)
  791. return 0;
  792. #ifdef CONFIG_X86
  793. if (cpu_has_xmm4_2)
  794. return 0;
  795. #endif
  796. return 1;
  797. }
  798. static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
  799. int mirror_num, unsigned long bio_flags,
  800. u64 bio_offset)
  801. {
  802. int async = check_async_write(inode, bio_flags);
  803. int ret;
  804. if (!(rw & REQ_WRITE)) {
  805. /*
  806. * called for a read, do the setup so that checksum validation
  807. * can happen in the async kernel threads
  808. */
  809. ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
  810. bio, 1);
  811. if (ret)
  812. goto out_w_error;
  813. ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
  814. mirror_num, 0);
  815. } else if (!async) {
  816. ret = btree_csum_one_bio(bio);
  817. if (ret)
  818. goto out_w_error;
  819. ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
  820. mirror_num, 0);
  821. } else {
  822. /*
  823. * kthread helpers are used to submit writes so that
  824. * checksumming can happen in parallel across all CPUs
  825. */
  826. ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
  827. inode, rw, bio, mirror_num, 0,
  828. bio_offset,
  829. __btree_submit_bio_start,
  830. __btree_submit_bio_done);
  831. }
  832. if (ret) {
  833. out_w_error:
  834. bio_endio(bio, ret);
  835. }
  836. return ret;
  837. }
  838. #ifdef CONFIG_MIGRATION
  839. static int btree_migratepage(struct address_space *mapping,
  840. struct page *newpage, struct page *page,
  841. enum migrate_mode mode)
  842. {
  843. /*
  844. * we can't safely write a btree page from here,
  845. * we haven't done the locking hook
  846. */
  847. if (PageDirty(page))
  848. return -EAGAIN;
  849. /*
  850. * Buffers may be managed in a filesystem specific way.
  851. * We must have no buffers or drop them.
  852. */
  853. if (page_has_private(page) &&
  854. !try_to_release_page(page, GFP_KERNEL))
  855. return -EAGAIN;
  856. return migrate_page(mapping, newpage, page, mode);
  857. }
  858. #endif
  859. static int btree_writepages(struct address_space *mapping,
  860. struct writeback_control *wbc)
  861. {
  862. struct extent_io_tree *tree;
  863. struct btrfs_fs_info *fs_info;
  864. int ret;
  865. tree = &BTRFS_I(mapping->host)->io_tree;
  866. if (wbc->sync_mode == WB_SYNC_NONE) {
  867. if (wbc->for_kupdate)
  868. return 0;
  869. fs_info = BTRFS_I(mapping->host)->root->fs_info;
  870. /* this is a bit racy, but that's ok */
  871. ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
  872. BTRFS_DIRTY_METADATA_THRESH);
  873. if (ret < 0)
  874. return 0;
  875. }
  876. return btree_write_cache_pages(mapping, wbc);
  877. }
  878. static int btree_readpage(struct file *file, struct page *page)
  879. {
  880. struct extent_io_tree *tree;
  881. tree = &BTRFS_I(page->mapping->host)->io_tree;
  882. return extent_read_full_page(tree, page, btree_get_extent, 0);
  883. }
  884. static int btree_releasepage(struct page *page, gfp_t gfp_flags)
  885. {
  886. if (PageWriteback(page) || PageDirty(page))
  887. return 0;
  888. return try_release_extent_buffer(page);
  889. }
  890. static void btree_invalidatepage(struct page *page, unsigned int offset,
  891. unsigned int length)
  892. {
  893. struct extent_io_tree *tree;
  894. tree = &BTRFS_I(page->mapping->host)->io_tree;
  895. extent_invalidatepage(tree, page, offset);
  896. btree_releasepage(page, GFP_NOFS);
  897. if (PagePrivate(page)) {
  898. printk(KERN_WARNING "btrfs warning page private not zero "
  899. "on page %llu\n", (unsigned long long)page_offset(page));
  900. ClearPagePrivate(page);
  901. set_page_private(page, 0);
  902. page_cache_release(page);
  903. }
  904. }
  905. static int btree_set_page_dirty(struct page *page)
  906. {
  907. #ifdef DEBUG
  908. struct extent_buffer *eb;
  909. BUG_ON(!PagePrivate(page));
  910. eb = (struct extent_buffer *)page->private;
  911. BUG_ON(!eb);
  912. BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
  913. BUG_ON(!atomic_read(&eb->refs));
  914. btrfs_assert_tree_locked(eb);
  915. #endif
  916. return __set_page_dirty_nobuffers(page);
  917. }
  918. static const struct address_space_operations btree_aops = {
  919. .readpage = btree_readpage,
  920. .writepages = btree_writepages,
  921. .releasepage = btree_releasepage,
  922. .invalidatepage = btree_invalidatepage,
  923. #ifdef CONFIG_MIGRATION
  924. .migratepage = btree_migratepage,
  925. #endif
  926. .set_page_dirty = btree_set_page_dirty,
  927. };
  928. int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
  929. u64 parent_transid)
  930. {
  931. struct extent_buffer *buf = NULL;
  932. struct inode *btree_inode = root->fs_info->btree_inode;
  933. int ret = 0;
  934. buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
  935. if (!buf)
  936. return 0;
  937. read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
  938. buf, 0, WAIT_NONE, btree_get_extent, 0);
  939. free_extent_buffer(buf);
  940. return ret;
  941. }
  942. int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
  943. int mirror_num, struct extent_buffer **eb)
  944. {
  945. struct extent_buffer *buf = NULL;
  946. struct inode *btree_inode = root->fs_info->btree_inode;
  947. struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
  948. int ret;
  949. buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
  950. if (!buf)
  951. return 0;
  952. set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
  953. ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
  954. btree_get_extent, mirror_num);
  955. if (ret) {
  956. free_extent_buffer(buf);
  957. return ret;
  958. }
  959. if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
  960. free_extent_buffer(buf);
  961. return -EIO;
  962. } else if (extent_buffer_uptodate(buf)) {
  963. *eb = buf;
  964. } else {
  965. free_extent_buffer(buf);
  966. }
  967. return 0;
  968. }
  969. struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
  970. u64 bytenr, u32 blocksize)
  971. {
  972. struct inode *btree_inode = root->fs_info->btree_inode;
  973. struct extent_buffer *eb;
  974. eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
  975. bytenr, blocksize);
  976. return eb;
  977. }
  978. struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
  979. u64 bytenr, u32 blocksize)
  980. {
  981. struct inode *btree_inode = root->fs_info->btree_inode;
  982. struct extent_buffer *eb;
  983. eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
  984. bytenr, blocksize);
  985. return eb;
  986. }
  987. int btrfs_write_tree_block(struct extent_buffer *buf)
  988. {
  989. return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
  990. buf->start + buf->len - 1);
  991. }
  992. int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
  993. {
  994. return filemap_fdatawait_range(buf->pages[0]->mapping,
  995. buf->start, buf->start + buf->len - 1);
  996. }
  997. struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
  998. u32 blocksize, u64 parent_transid)
  999. {
  1000. struct extent_buffer *buf = NULL;
  1001. int ret;
  1002. buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
  1003. if (!buf)
  1004. return NULL;
  1005. ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
  1006. if (ret) {
  1007. free_extent_buffer(buf);
  1008. return NULL;
  1009. }
  1010. return buf;
  1011. }
  1012. void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
  1013. struct extent_buffer *buf)
  1014. {
  1015. struct btrfs_fs_info *fs_info = root->fs_info;
  1016. if (btrfs_header_generation(buf) ==
  1017. fs_info->running_transaction->transid) {
  1018. btrfs_assert_tree_locked(buf);
  1019. if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
  1020. __percpu_counter_add(&fs_info->dirty_metadata_bytes,
  1021. -buf->len,
  1022. fs_info->dirty_metadata_batch);
  1023. /* ugh, clear_extent_buffer_dirty needs to lock the page */
  1024. btrfs_set_lock_blocking(buf);
  1025. clear_extent_buffer_dirty(buf);
  1026. }
  1027. }
  1028. }
  1029. static void __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
  1030. u32 stripesize, struct btrfs_root *root,
  1031. struct btrfs_fs_info *fs_info,
  1032. u64 objectid)
  1033. {
  1034. root->node = NULL;
  1035. root->commit_root = NULL;
  1036. root->sectorsize = sectorsize;
  1037. root->nodesize = nodesize;
  1038. root->leafsize = leafsize;
  1039. root->stripesize = stripesize;
  1040. root->ref_cows = 0;
  1041. root->track_dirty = 0;
  1042. root->in_radix = 0;
  1043. root->orphan_item_inserted = 0;
  1044. root->orphan_cleanup_state = 0;
  1045. root->objectid = objectid;
  1046. root->last_trans = 0;
  1047. root->highest_objectid = 0;
  1048. root->nr_delalloc_inodes = 0;
  1049. root->nr_ordered_extents = 0;
  1050. root->name = NULL;
  1051. root->inode_tree = RB_ROOT;
  1052. INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
  1053. root->block_rsv = NULL;
  1054. root->orphan_block_rsv = NULL;
  1055. INIT_LIST_HEAD(&root->dirty_list);
  1056. INIT_LIST_HEAD(&root->root_list);
  1057. INIT_LIST_HEAD(&root->delalloc_inodes);
  1058. INIT_LIST_HEAD(&root->delalloc_root);
  1059. INIT_LIST_HEAD(&root->ordered_extents);
  1060. INIT_LIST_HEAD(&root->ordered_root);
  1061. INIT_LIST_HEAD(&root->logged_list[0]);
  1062. INIT_LIST_HEAD(&root->logged_list[1]);
  1063. spin_lock_init(&root->orphan_lock);
  1064. spin_lock_init(&root->inode_lock);
  1065. spin_lock_init(&root->delalloc_lock);
  1066. spin_lock_init(&root->ordered_extent_lock);
  1067. spin_lock_init(&root->accounting_lock);
  1068. spin_lock_init(&root->log_extents_lock[0]);
  1069. spin_lock_init(&root->log_extents_lock[1]);
  1070. mutex_init(&root->objectid_mutex);
  1071. mutex_init(&root->log_mutex);
  1072. init_waitqueue_head(&root->log_writer_wait);
  1073. init_waitqueue_head(&root->log_commit_wait[0]);
  1074. init_waitqueue_head(&root->log_commit_wait[1]);
  1075. atomic_set(&root->log_commit[0], 0);
  1076. atomic_set(&root->log_commit[1], 0);
  1077. atomic_set(&root->log_writers, 0);
  1078. atomic_set(&root->log_batch, 0);
  1079. atomic_set(&root->orphan_inodes, 0);
  1080. atomic_set(&root->refs, 1);
  1081. root->log_transid = 0;
  1082. root->last_log_commit = 0;
  1083. if (fs_info)
  1084. extent_io_tree_init(&root->dirty_log_pages,
  1085. fs_info->btree_inode->i_mapping);
  1086. memset(&root->root_key, 0, sizeof(root->root_key));
  1087. memset(&root->root_item, 0, sizeof(root->root_item));
  1088. memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
  1089. memset(&root->root_kobj, 0, sizeof(root->root_kobj));
  1090. if (fs_info)
  1091. root->defrag_trans_start = fs_info->generation;
  1092. else
  1093. root->defrag_trans_start = 0;
  1094. init_completion(&root->kobj_unregister);
  1095. root->defrag_running = 0;
  1096. root->root_key.objectid = objectid;
  1097. root->anon_dev = 0;
  1098. spin_lock_init(&root->root_item_lock);
  1099. }
  1100. static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
  1101. {
  1102. struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
  1103. if (root)
  1104. root->fs_info = fs_info;
  1105. return root;
  1106. }
  1107. #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
  1108. /* Should only be used by the testing infrastructure */
  1109. struct btrfs_root *btrfs_alloc_dummy_root(void)
  1110. {
  1111. struct btrfs_root *root;
  1112. root = btrfs_alloc_root(NULL);
  1113. if (!root)
  1114. return ERR_PTR(-ENOMEM);
  1115. __setup_root(4096, 4096, 4096, 4096, root, NULL, 1);
  1116. root->dummy_root = 1;
  1117. return root;
  1118. }
  1119. #endif
  1120. struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
  1121. struct btrfs_fs_info *fs_info,
  1122. u64 objectid)
  1123. {
  1124. struct extent_buffer *leaf;
  1125. struct btrfs_root *tree_root = fs_info->tree_root;
  1126. struct btrfs_root *root;
  1127. struct btrfs_key key;
  1128. int ret = 0;
  1129. u64 bytenr;
  1130. uuid_le uuid;
  1131. root = btrfs_alloc_root(fs_info);
  1132. if (!root)
  1133. return ERR_PTR(-ENOMEM);
  1134. __setup_root(tree_root->nodesize, tree_root->leafsize,
  1135. tree_root->sectorsize, tree_root->stripesize,
  1136. root, fs_info, objectid);
  1137. root->root_key.objectid = objectid;
  1138. root->root_key.type = BTRFS_ROOT_ITEM_KEY;
  1139. root->root_key.offset = 0;
  1140. leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
  1141. 0, objectid, NULL, 0, 0, 0);
  1142. if (IS_ERR(leaf)) {
  1143. ret = PTR_ERR(leaf);
  1144. leaf = NULL;
  1145. goto fail;
  1146. }
  1147. bytenr = leaf->start;
  1148. memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
  1149. btrfs_set_header_bytenr(leaf, leaf->start);
  1150. btrfs_set_header_generation(leaf, trans->transid);
  1151. btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
  1152. btrfs_set_header_owner(leaf, objectid);
  1153. root->node = leaf;
  1154. write_extent_buffer(leaf, fs_info->fsid, btrfs_header_fsid(leaf),
  1155. BTRFS_FSID_SIZE);
  1156. write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
  1157. btrfs_header_chunk_tree_uuid(leaf),
  1158. BTRFS_UUID_SIZE);
  1159. btrfs_mark_buffer_dirty(leaf);
  1160. root->commit_root = btrfs_root_node(root);
  1161. root->track_dirty = 1;
  1162. root->root_item.flags = 0;
  1163. root->root_item.byte_limit = 0;
  1164. btrfs_set_root_bytenr(&root->root_item, leaf->start);
  1165. btrfs_set_root_generation(&root->root_item, trans->transid);
  1166. btrfs_set_root_level(&root->root_item, 0);
  1167. btrfs_set_root_refs(&root->root_item, 1);
  1168. btrfs_set_root_used(&root->root_item, leaf->len);
  1169. btrfs_set_root_last_snapshot(&root->root_item, 0);
  1170. btrfs_set_root_dirid(&root->root_item, 0);
  1171. uuid_le_gen(&uuid);
  1172. memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
  1173. root->root_item.drop_level = 0;
  1174. key.objectid = objectid;
  1175. key.type = BTRFS_ROOT_ITEM_KEY;
  1176. key.offset = 0;
  1177. ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
  1178. if (ret)
  1179. goto fail;
  1180. btrfs_tree_unlock(leaf);
  1181. return root;
  1182. fail:
  1183. if (leaf) {
  1184. btrfs_tree_unlock(leaf);
  1185. free_extent_buffer(leaf);
  1186. }
  1187. kfree(root);
  1188. return ERR_PTR(ret);
  1189. }
  1190. static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
  1191. struct btrfs_fs_info *fs_info)
  1192. {
  1193. struct btrfs_root *root;
  1194. struct btrfs_root *tree_root = fs_info->tree_root;
  1195. struct extent_buffer *leaf;
  1196. root = btrfs_alloc_root(fs_info);
  1197. if (!root)
  1198. return ERR_PTR(-ENOMEM);
  1199. __setup_root(tree_root->nodesize, tree_root->leafsize,
  1200. tree_root->sectorsize, tree_root->stripesize,
  1201. root, fs_info, BTRFS_TREE_LOG_OBJECTID);
  1202. root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
  1203. root->root_key.type = BTRFS_ROOT_ITEM_KEY;
  1204. root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
  1205. /*
  1206. * log trees do not get reference counted because they go away
  1207. * before a real commit is actually done. They do store pointers
  1208. * to file data extents, and those reference counts still get
  1209. * updated (along with back refs to the log tree).
  1210. */
  1211. root->ref_cows = 0;
  1212. leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
  1213. BTRFS_TREE_LOG_OBJECTID, NULL,
  1214. 0, 0, 0);
  1215. if (IS_ERR(leaf)) {
  1216. kfree(root);
  1217. return ERR_CAST(leaf);
  1218. }
  1219. memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
  1220. btrfs_set_header_bytenr(leaf, leaf->start);
  1221. btrfs_set_header_generation(leaf, trans->transid);
  1222. btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
  1223. btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
  1224. root->node = leaf;
  1225. write_extent_buffer(root->node, root->fs_info->fsid,
  1226. btrfs_header_fsid(root->node), BTRFS_FSID_SIZE);
  1227. btrfs_mark_buffer_dirty(root->node);
  1228. btrfs_tree_unlock(root->node);
  1229. return root;
  1230. }
  1231. int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
  1232. struct btrfs_fs_info *fs_info)
  1233. {
  1234. struct btrfs_root *log_root;
  1235. log_root = alloc_log_tree(trans, fs_info);
  1236. if (IS_ERR(log_root))
  1237. return PTR_ERR(log_root);
  1238. WARN_ON(fs_info->log_root_tree);
  1239. fs_info->log_root_tree = log_root;
  1240. return 0;
  1241. }
  1242. int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
  1243. struct btrfs_root *root)
  1244. {
  1245. struct btrfs_root *log_root;
  1246. struct btrfs_inode_item *inode_item;
  1247. log_root = alloc_log_tree(trans, root->fs_info);
  1248. if (IS_ERR(log_root))
  1249. return PTR_ERR(log_root);
  1250. log_root->last_trans = trans->transid;
  1251. log_root->root_key.offset = root->root_key.objectid;
  1252. inode_item = &log_root->root_item.inode;
  1253. btrfs_set_stack_inode_generation(inode_item, 1);
  1254. btrfs_set_stack_inode_size(inode_item, 3);
  1255. btrfs_set_stack_inode_nlink(inode_item, 1);
  1256. btrfs_set_stack_inode_nbytes(inode_item, root->leafsize);
  1257. btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
  1258. btrfs_set_root_node(&log_root->root_item, log_root->node);
  1259. WARN_ON(root->log_root);
  1260. root->log_root = log_root;
  1261. root->log_transid = 0;
  1262. root->last_log_commit = 0;
  1263. return 0;
  1264. }
  1265. static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
  1266. struct btrfs_key *key)
  1267. {
  1268. struct btrfs_root *root;
  1269. struct btrfs_fs_info *fs_info = tree_root->fs_info;
  1270. struct btrfs_path *path;
  1271. u64 generation;
  1272. u32 blocksize;
  1273. int ret;
  1274. path = btrfs_alloc_path();
  1275. if (!path)
  1276. return ERR_PTR(-ENOMEM);
  1277. root = btrfs_alloc_root(fs_info);
  1278. if (!root) {
  1279. ret = -ENOMEM;
  1280. goto alloc_fail;
  1281. }
  1282. __setup_root(tree_root->nodesize, tree_root->leafsize,
  1283. tree_root->sectorsize, tree_root->stripesize,
  1284. root, fs_info, key->objectid);
  1285. ret = btrfs_find_root(tree_root, key, path,
  1286. &root->root_item, &root->root_key);
  1287. if (ret) {
  1288. if (ret > 0)
  1289. ret = -ENOENT;
  1290. goto find_fail;
  1291. }
  1292. generation = btrfs_root_generation(&root->root_item);
  1293. blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
  1294. root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
  1295. blocksize, generation);
  1296. if (!root->node) {
  1297. ret = -ENOMEM;
  1298. goto find_fail;
  1299. } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
  1300. ret = -EIO;
  1301. goto read_fail;
  1302. }
  1303. root->commit_root = btrfs_root_node(root);
  1304. out:
  1305. btrfs_free_path(path);
  1306. return root;
  1307. read_fail:
  1308. free_extent_buffer(root->node);
  1309. find_fail:
  1310. kfree(root);
  1311. alloc_fail:
  1312. root = ERR_PTR(ret);
  1313. goto out;
  1314. }
  1315. struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
  1316. struct btrfs_key *location)
  1317. {
  1318. struct btrfs_root *root;
  1319. root = btrfs_read_tree_root(tree_root, location);
  1320. if (IS_ERR(root))
  1321. return root;
  1322. if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
  1323. root->ref_cows = 1;
  1324. btrfs_check_and_init_root_item(&root->root_item);
  1325. }
  1326. return root;
  1327. }
  1328. int btrfs_init_fs_root(struct btrfs_root *root)
  1329. {
  1330. int ret;
  1331. root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
  1332. root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
  1333. GFP_NOFS);
  1334. if (!root->free_ino_pinned || !root->free_ino_ctl) {
  1335. ret = -ENOMEM;
  1336. goto fail;
  1337. }
  1338. btrfs_init_free_ino_ctl(root);
  1339. mutex_init(&root->fs_commit_mutex);
  1340. spin_lock_init(&root->cache_lock);
  1341. init_waitqueue_head(&root->cache_wait);
  1342. ret = get_anon_bdev(&root->anon_dev);
  1343. if (ret)
  1344. goto fail;
  1345. return 0;
  1346. fail:
  1347. kfree(root->free_ino_ctl);
  1348. kfree(root->free_ino_pinned);
  1349. return ret;
  1350. }
  1351. static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
  1352. u64 root_id)
  1353. {
  1354. struct btrfs_root *root;
  1355. spin_lock(&fs_info->fs_roots_radix_lock);
  1356. root = radix_tree_lookup(&fs_info->fs_roots_radix,
  1357. (unsigned long)root_id);
  1358. spin_unlock(&fs_info->fs_roots_radix_lock);
  1359. return root;
  1360. }
  1361. int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
  1362. struct btrfs_root *root)
  1363. {
  1364. int ret;
  1365. ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
  1366. if (ret)
  1367. return ret;
  1368. spin_lock(&fs_info->fs_roots_radix_lock);
  1369. ret = radix_tree_insert(&fs_info->fs_roots_radix,
  1370. (unsigned long)root->root_key.objectid,
  1371. root);
  1372. if (ret == 0)
  1373. root->in_radix = 1;
  1374. spin_unlock(&fs_info->fs_roots_radix_lock);
  1375. radix_tree_preload_end();
  1376. return ret;
  1377. }
  1378. struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
  1379. struct btrfs_key *location,
  1380. bool check_ref)
  1381. {
  1382. struct btrfs_root *root;
  1383. int ret;
  1384. if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
  1385. return fs_info->tree_root;
  1386. if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
  1387. return fs_info->extent_root;
  1388. if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
  1389. return fs_info->chunk_root;
  1390. if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
  1391. return fs_info->dev_root;
  1392. if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
  1393. return fs_info->csum_root;
  1394. if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
  1395. return fs_info->quota_root ? fs_info->quota_root :
  1396. ERR_PTR(-ENOENT);
  1397. if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
  1398. return fs_info->uuid_root ? fs_info->uuid_root :
  1399. ERR_PTR(-ENOENT);
  1400. again:
  1401. root = btrfs_lookup_fs_root(fs_info, location->objectid);
  1402. if (root) {
  1403. if (check_ref && btrfs_root_refs(&root->root_item) == 0)
  1404. return ERR_PTR(-ENOENT);
  1405. return root;
  1406. }
  1407. root = btrfs_read_fs_root(fs_info->tree_root, location);
  1408. if (IS_ERR(root))
  1409. return root;
  1410. if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
  1411. ret = -ENOENT;
  1412. goto fail;
  1413. }
  1414. ret = btrfs_init_fs_root(root);
  1415. if (ret)
  1416. goto fail;
  1417. ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
  1418. if (ret < 0)
  1419. goto fail;
  1420. if (ret == 0)
  1421. root->orphan_item_inserted = 1;
  1422. ret = btrfs_insert_fs_root(fs_info, root);
  1423. if (ret) {
  1424. if (ret == -EEXIST) {
  1425. free_fs_root(root);
  1426. goto again;
  1427. }
  1428. goto fail;
  1429. }
  1430. return root;
  1431. fail:
  1432. free_fs_root(root);
  1433. return ERR_PTR(ret);
  1434. }
  1435. static int btrfs_congested_fn(void *congested_data, int bdi_bits)
  1436. {
  1437. struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
  1438. int ret = 0;
  1439. struct btrfs_device *device;
  1440. struct backing_dev_info *bdi;
  1441. rcu_read_lock();
  1442. list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
  1443. if (!device->bdev)
  1444. continue;
  1445. bdi = blk_get_backing_dev_info(device->bdev);
  1446. if (bdi && bdi_congested(bdi, bdi_bits)) {
  1447. ret = 1;
  1448. break;
  1449. }
  1450. }
  1451. rcu_read_unlock();
  1452. return ret;
  1453. }
  1454. /*
  1455. * If this fails, caller must call bdi_destroy() to get rid of the
  1456. * bdi again.
  1457. */
  1458. static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
  1459. {
  1460. int err;
  1461. bdi->capabilities = BDI_CAP_MAP_COPY;
  1462. err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
  1463. if (err)
  1464. return err;
  1465. bdi->ra_pages = default_backing_dev_info.ra_pages;
  1466. bdi->congested_fn = btrfs_congested_fn;
  1467. bdi->congested_data = info;
  1468. return 0;
  1469. }
  1470. /*
  1471. * called by the kthread helper functions to finally call the bio end_io
  1472. * functions. This is where read checksum verification actually happens
  1473. */
  1474. static void end_workqueue_fn(struct btrfs_work *work)
  1475. {
  1476. struct bio *bio;
  1477. struct end_io_wq *end_io_wq;
  1478. struct btrfs_fs_info *fs_info;
  1479. int error;
  1480. end_io_wq = container_of(work, struct end_io_wq, work);
  1481. bio = end_io_wq->bio;
  1482. fs_info = end_io_wq->info;
  1483. error = end_io_wq->error;
  1484. bio->bi_private = end_io_wq->private;
  1485. bio->bi_end_io = end_io_wq->end_io;
  1486. kfree(end_io_wq);
  1487. bio_endio(bio, error);
  1488. }
  1489. static int cleaner_kthread(void *arg)
  1490. {
  1491. struct btrfs_root *root = arg;
  1492. int again;
  1493. do {
  1494. again = 0;
  1495. /* Make the cleaner go to sleep early. */
  1496. if (btrfs_need_cleaner_sleep(root))
  1497. goto sleep;
  1498. if (!mutex_trylock(&root->fs_info->cleaner_mutex))
  1499. goto sleep;
  1500. /*
  1501. * Avoid the problem that we change the status of the fs
  1502. * during the above check and trylock.
  1503. */
  1504. if (btrfs_need_cleaner_sleep(root)) {
  1505. mutex_unlock(&root->fs_info->cleaner_mutex);
  1506. goto sleep;
  1507. }
  1508. btrfs_run_delayed_iputs(root);
  1509. again = btrfs_clean_one_deleted_snapshot(root);
  1510. mutex_unlock(&root->fs_info->cleaner_mutex);
  1511. /*
  1512. * The defragger has dealt with the R/O remount and umount,
  1513. * needn't do anything special here.
  1514. */
  1515. btrfs_run_defrag_inodes(root->fs_info);
  1516. sleep:
  1517. if (!try_to_freeze() && !again) {
  1518. set_current_state(TASK_INTERRUPTIBLE);
  1519. if (!kthread_should_stop())
  1520. schedule();
  1521. __set_current_state(TASK_RUNNING);
  1522. }
  1523. } while (!kthread_should_stop());
  1524. return 0;
  1525. }
  1526. static int transaction_kthread(void *arg)
  1527. {
  1528. struct btrfs_root *root = arg;
  1529. struct btrfs_trans_handle *trans;
  1530. struct btrfs_transaction *cur;
  1531. u64 transid;
  1532. unsigned long now;
  1533. unsigned long delay;
  1534. bool cannot_commit;
  1535. do {
  1536. cannot_commit = false;
  1537. delay = HZ * root->fs_info->commit_interval;
  1538. mutex_lock(&root->fs_info->transaction_kthread_mutex);
  1539. spin_lock(&root->fs_info->trans_lock);
  1540. cur = root->fs_info->running_transaction;
  1541. if (!cur) {
  1542. spin_unlock(&root->fs_info->trans_lock);
  1543. goto sleep;
  1544. }
  1545. now = get_seconds();
  1546. if (cur->state < TRANS_STATE_BLOCKED &&
  1547. (now < cur->start_time ||
  1548. now - cur->start_time < root->fs_info->commit_interval)) {
  1549. spin_unlock(&root->fs_info->trans_lock);
  1550. delay = HZ * 5;
  1551. goto sleep;
  1552. }
  1553. transid = cur->transid;
  1554. spin_unlock(&root->fs_info->trans_lock);
  1555. /* If the file system is aborted, this will always fail. */
  1556. trans = btrfs_attach_transaction(root);
  1557. if (IS_ERR(trans)) {
  1558. if (PTR_ERR(trans) != -ENOENT)
  1559. cannot_commit = true;
  1560. goto sleep;
  1561. }
  1562. if (transid == trans->transid) {
  1563. btrfs_commit_transaction(trans, root);
  1564. } else {
  1565. btrfs_end_transaction(trans, root);
  1566. }
  1567. sleep:
  1568. wake_up_process(root->fs_info->cleaner_kthread);
  1569. mutex_unlock(&root->fs_info->transaction_kthread_mutex);
  1570. if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
  1571. &root->fs_info->fs_state)))
  1572. btrfs_cleanup_transaction(root);
  1573. if (!try_to_freeze()) {
  1574. set_current_state(TASK_INTERRUPTIBLE);
  1575. if (!kthread_should_stop() &&
  1576. (!btrfs_transaction_blocked(root->fs_info) ||
  1577. cannot_commit))
  1578. schedule_timeout(delay);
  1579. __set_current_state(TASK_RUNNING);
  1580. }
  1581. } while (!kthread_should_stop());
  1582. return 0;
  1583. }
  1584. /*
  1585. * this will find the highest generation in the array of
  1586. * root backups. The index of the highest array is returned,
  1587. * or -1 if we can't find anything.
  1588. *
  1589. * We check to make sure the array is valid by comparing the
  1590. * generation of the latest root in the array with the generation
  1591. * in the super block. If they don't match we pitch it.
  1592. */
  1593. static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
  1594. {
  1595. u64 cur;
  1596. int newest_index = -1;
  1597. struct btrfs_root_backup *root_backup;
  1598. int i;
  1599. for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
  1600. root_backup = info->super_copy->super_roots + i;
  1601. cur = btrfs_backup_tree_root_gen(root_backup);
  1602. if (cur == newest_gen)
  1603. newest_index = i;
  1604. }
  1605. /* check to see if we actually wrapped around */
  1606. if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
  1607. root_backup = info->super_copy->super_roots;
  1608. cur = btrfs_backup_tree_root_gen(root_backup);
  1609. if (cur == newest_gen)
  1610. newest_index = 0;
  1611. }
  1612. return newest_index;
  1613. }
  1614. /*
  1615. * find the oldest backup so we know where to store new entries
  1616. * in the backup array. This will set the backup_root_index
  1617. * field in the fs_info struct
  1618. */
  1619. static void find_oldest_super_backup(struct btrfs_fs_info *info,
  1620. u64 newest_gen)
  1621. {
  1622. int newest_index = -1;
  1623. newest_index = find_newest_super_backup(info, newest_gen);
  1624. /* if there was garbage in there, just move along */
  1625. if (newest_index == -1) {
  1626. info->backup_root_index = 0;
  1627. } else {
  1628. info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
  1629. }
  1630. }
  1631. /*
  1632. * copy all the root pointers into the super backup array.
  1633. * this will bump the backup pointer by one when it is
  1634. * done
  1635. */
  1636. static void backup_super_roots(struct btrfs_fs_info *info)
  1637. {
  1638. int next_backup;
  1639. struct btrfs_root_backup *root_backup;
  1640. int last_backup;
  1641. next_backup = info->backup_root_index;
  1642. last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
  1643. BTRFS_NUM_BACKUP_ROOTS;
  1644. /*
  1645. * just overwrite the last backup if we're at the same generation
  1646. * this happens only at umount
  1647. */
  1648. root_backup = info->super_for_commit->super_roots + last_backup;
  1649. if (btrfs_backup_tree_root_gen(root_backup) ==
  1650. btrfs_header_generation(info->tree_root->node))
  1651. next_backup = last_backup;
  1652. root_backup = info->super_for_commit->super_roots + next_backup;
  1653. /*
  1654. * make sure all of our padding and empty slots get zero filled
  1655. * regardless of which ones we use today
  1656. */
  1657. memset(root_backup, 0, sizeof(*root_backup));
  1658. info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
  1659. btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
  1660. btrfs_set_backup_tree_root_gen(root_backup,
  1661. btrfs_header_generation(info->tree_root->node));
  1662. btrfs_set_backup_tree_root_level(root_backup,
  1663. btrfs_header_level(info->tree_root->node));
  1664. btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
  1665. btrfs_set_backup_chunk_root_gen(root_backup,
  1666. btrfs_header_generation(info->chunk_root->node));
  1667. btrfs_set_backup_chunk_root_level(root_backup,
  1668. btrfs_header_level(info->chunk_root->node));
  1669. btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
  1670. btrfs_set_backup_extent_root_gen(root_backup,
  1671. btrfs_header_generation(info->extent_root->node));
  1672. btrfs_set_backup_extent_root_level(root_backup,
  1673. btrfs_header_level(info->extent_root->node));
  1674. /*
  1675. * we might commit during log recovery, which happens before we set
  1676. * the fs_root. Make sure it is valid before we fill it in.
  1677. */
  1678. if (info->fs_root && info->fs_root->node) {
  1679. btrfs_set_backup_fs_root(root_backup,
  1680. info->fs_root->node->start);
  1681. btrfs_set_backup_fs_root_gen(root_backup,
  1682. btrfs_header_generation(info->fs_root->node));
  1683. btrfs_set_backup_fs_root_level(root_backup,
  1684. btrfs_header_level(info->fs_root->node));
  1685. }
  1686. btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
  1687. btrfs_set_backup_dev_root_gen(root_backup,
  1688. btrfs_header_generation(info->dev_root->node));
  1689. btrfs_set_backup_dev_root_level(root_backup,
  1690. btrfs_header_level(info->dev_root->node));
  1691. btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
  1692. btrfs_set_backup_csum_root_gen(root_backup,
  1693. btrfs_header_generation(info->csum_root->node));
  1694. btrfs_set_backup_csum_root_level(root_backup,
  1695. btrfs_header_level(info->csum_root->node));
  1696. btrfs_set_backup_total_bytes(root_backup,
  1697. btrfs_super_total_bytes(info->super_copy));
  1698. btrfs_set_backup_bytes_used(root_backup,
  1699. btrfs_super_bytes_used(info->super_copy));
  1700. btrfs_set_backup_num_devices(root_backup,
  1701. btrfs_super_num_devices(info->super_copy));
  1702. /*
  1703. * if we don't copy this out to the super_copy, it won't get remembered
  1704. * for the next commit
  1705. */
  1706. memcpy(&info->super_copy->super_roots,
  1707. &info->super_for_commit->super_roots,
  1708. sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
  1709. }
  1710. /*
  1711. * this copies info out of the root backup array and back into
  1712. * the in-memory super block. It is meant to help iterate through
  1713. * the array, so you send it the number of backups you've already
  1714. * tried and the last backup index you used.
  1715. *
  1716. * this returns -1 when it has tried all the backups
  1717. */
  1718. static noinline int next_root_backup(struct btrfs_fs_info *info,
  1719. struct btrfs_super_block *super,
  1720. int *num_backups_tried, int *backup_index)
  1721. {
  1722. struct btrfs_root_backup *root_backup;
  1723. int newest = *backup_index;
  1724. if (*num_backups_tried == 0) {
  1725. u64 gen = btrfs_super_generation(super);
  1726. newest = find_newest_super_backup(info, gen);
  1727. if (newest == -1)
  1728. return -1;
  1729. *backup_index = newest;
  1730. *num_backups_tried = 1;
  1731. } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
  1732. /* we've tried all the backups, all done */
  1733. return -1;
  1734. } else {
  1735. /* jump to the next oldest backup */
  1736. newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
  1737. BTRFS_NUM_BACKUP_ROOTS;
  1738. *backup_index = newest;
  1739. *num_backups_tried += 1;
  1740. }
  1741. root_backup = super->super_roots + newest;
  1742. btrfs_set_super_generation(super,
  1743. btrfs_backup_tree_root_gen(root_backup));
  1744. btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
  1745. btrfs_set_super_root_level(super,
  1746. btrfs_backup_tree_root_level(root_backup));
  1747. btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
  1748. /*
  1749. * fixme: the total bytes and num_devices need to match or we should
  1750. * need a fsck
  1751. */
  1752. btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
  1753. btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
  1754. return 0;
  1755. }
  1756. /* helper to cleanup workers */
  1757. static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
  1758. {
  1759. btrfs_stop_workers(&fs_info->generic_worker);
  1760. btrfs_stop_workers(&fs_info->fixup_workers);
  1761. btrfs_stop_workers(&fs_info->delalloc_workers);
  1762. btrfs_stop_workers(&fs_info->workers);
  1763. btrfs_stop_workers(&fs_info->endio_workers);
  1764. btrfs_stop_workers(&fs_info->endio_meta_workers);
  1765. btrfs_stop_workers(&fs_info->endio_raid56_workers);
  1766. btrfs_stop_workers(&fs_info->rmw_workers);
  1767. btrfs_stop_workers(&fs_info->endio_meta_write_workers);
  1768. btrfs_stop_workers(&fs_info->endio_write_workers);
  1769. btrfs_stop_workers(&fs_info->endio_freespace_worker);
  1770. btrfs_stop_workers(&fs_info->submit_workers);
  1771. btrfs_stop_workers(&fs_info->delayed_workers);
  1772. btrfs_stop_workers(&fs_info->caching_workers);
  1773. btrfs_stop_workers(&fs_info->readahead_workers);
  1774. btrfs_stop_workers(&fs_info->flush_workers);
  1775. btrfs_stop_workers(&fs_info->qgroup_rescan_workers);
  1776. }
  1777. /* helper to cleanup tree roots */
  1778. static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
  1779. {
  1780. free_extent_buffer(info->tree_root->node);
  1781. free_extent_buffer(info->tree_root->commit_root);
  1782. info->tree_root->node = NULL;
  1783. info->tree_root->commit_root = NULL;
  1784. if (info->dev_root) {
  1785. free_extent_buffer(info->dev_root->node);
  1786. free_extent_buffer(info->dev_root->commit_root);
  1787. info->dev_root->node = NULL;
  1788. info->dev_root->commit_root = NULL;
  1789. }
  1790. if (info->extent_root) {
  1791. free_extent_buffer(info->extent_root->node);
  1792. free_extent_buffer(info->extent_root->commit_root);
  1793. info->extent_root->node = NULL;
  1794. info->extent_root->commit_root = NULL;
  1795. }
  1796. if (info->csum_root) {
  1797. free_extent_buffer(info->csum_root->node);
  1798. free_extent_buffer(info->csum_root->commit_root);
  1799. info->csum_root->node = NULL;
  1800. info->csum_root->commit_root = NULL;
  1801. }
  1802. if (info->quota_root) {
  1803. free_extent_buffer(info->quota_root->node);
  1804. free_extent_buffer(info->quota_root->commit_root);
  1805. info->quota_root->node = NULL;
  1806. info->quota_root->commit_root = NULL;
  1807. }
  1808. if (info->uuid_root) {
  1809. free_extent_buffer(info->uuid_root->node);
  1810. free_extent_buffer(info->uuid_root->commit_root);
  1811. info->uuid_root->node = NULL;
  1812. info->uuid_root->commit_root = NULL;
  1813. }
  1814. if (chunk_root) {
  1815. free_extent_buffer(info->chunk_root->node);
  1816. free_extent_buffer(info->chunk_root->commit_root);
  1817. info->chunk_root->node = NULL;
  1818. info->chunk_root->commit_root = NULL;
  1819. }
  1820. }
  1821. static void del_fs_roots(struct btrfs_fs_info *fs_info)
  1822. {
  1823. int ret;
  1824. struct btrfs_root *gang[8];
  1825. int i;
  1826. while (!list_empty(&fs_info->dead_roots)) {
  1827. gang[0] = list_entry(fs_info->dead_roots.next,
  1828. struct btrfs_root, root_list);
  1829. list_del(&gang[0]->root_list);
  1830. if (gang[0]->in_radix) {
  1831. btrfs_drop_and_free_fs_root(fs_info, gang[0]);
  1832. } else {
  1833. free_extent_buffer(gang[0]->node);
  1834. free_extent_buffer(gang[0]->commit_root);
  1835. btrfs_put_fs_root(gang[0]);
  1836. }
  1837. }
  1838. while (1) {
  1839. ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
  1840. (void **)gang, 0,
  1841. ARRAY_SIZE(gang));
  1842. if (!ret)
  1843. break;
  1844. for (i = 0; i < ret; i++)
  1845. btrfs_drop_and_free_fs_root(fs_info, gang[i]);
  1846. }
  1847. }
  1848. int open_ctree(struct super_block *sb,
  1849. struct btrfs_fs_devices *fs_devices,
  1850. char *options)
  1851. {
  1852. u32 sectorsize;
  1853. u32 nodesize;
  1854. u32 leafsize;
  1855. u32 blocksize;
  1856. u32 stripesize;
  1857. u64 generation;
  1858. u64 features;
  1859. struct btrfs_key location;
  1860. struct buffer_head *bh;
  1861. struct btrfs_super_block *disk_super;
  1862. struct btrfs_fs_info *fs_info = btrfs_sb(sb);
  1863. struct btrfs_root *tree_root;
  1864. struct btrfs_root *extent_root;
  1865. struct btrfs_root *csum_root;
  1866. struct btrfs_root *chunk_root;
  1867. struct btrfs_root *dev_root;
  1868. struct btrfs_root *quota_root;
  1869. struct btrfs_root *uuid_root;
  1870. struct btrfs_root *log_tree_root;
  1871. int ret;
  1872. int err = -EINVAL;
  1873. int num_backups_tried = 0;
  1874. int backup_index = 0;
  1875. bool create_uuid_tree;
  1876. bool check_uuid_tree;
  1877. tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
  1878. chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
  1879. if (!tree_root || !chunk_root) {
  1880. err = -ENOMEM;
  1881. goto fail;
  1882. }
  1883. ret = init_srcu_struct(&fs_info->subvol_srcu);
  1884. if (ret) {
  1885. err = ret;
  1886. goto fail;
  1887. }
  1888. ret = setup_bdi(fs_info, &fs_info->bdi);
  1889. if (ret) {
  1890. err = ret;
  1891. goto fail_srcu;
  1892. }
  1893. ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0);
  1894. if (ret) {
  1895. err = ret;
  1896. goto fail_bdi;
  1897. }
  1898. fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
  1899. (1 + ilog2(nr_cpu_ids));
  1900. ret = percpu_counter_init(&fs_info->delalloc_bytes, 0);
  1901. if (ret) {
  1902. err = ret;
  1903. goto fail_dirty_metadata_bytes;
  1904. }
  1905. fs_info->btree_inode = new_inode(sb);
  1906. if (!fs_info->btree_inode) {
  1907. err = -ENOMEM;
  1908. goto fail_delalloc_bytes;
  1909. }
  1910. mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
  1911. INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
  1912. INIT_LIST_HEAD(&fs_info->trans_list);
  1913. INIT_LIST_HEAD(&fs_info->dead_roots);
  1914. INIT_LIST_HEAD(&fs_info->delayed_iputs);
  1915. INIT_LIST_HEAD(&fs_info->delalloc_roots);
  1916. INIT_LIST_HEAD(&fs_info->caching_block_groups);
  1917. spin_lock_init(&fs_info->delalloc_root_lock);
  1918. spin_lock_init(&fs_info->trans_lock);
  1919. spin_lock_init(&fs_info->fs_roots_radix_lock);
  1920. spin_lock_init(&fs_info->delayed_iput_lock);
  1921. spin_lock_init(&fs_info->defrag_inodes_lock);
  1922. spin_lock_init(&fs_info->free_chunk_lock);
  1923. spin_lock_init(&fs_info->tree_mod_seq_lock);
  1924. spin_lock_init(&fs_info->super_lock);
  1925. rwlock_init(&fs_info->tree_mod_log_lock);
  1926. mutex_init(&fs_info->reloc_mutex);
  1927. seqlock_init(&fs_info->profiles_lock);
  1928. init_completion(&fs_info->kobj_unregister);
  1929. INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
  1930. INIT_LIST_HEAD(&fs_info->space_info);
  1931. INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
  1932. btrfs_mapping_init(&fs_info->mapping_tree);
  1933. btrfs_init_block_rsv(&fs_info->global_block_rsv,
  1934. BTRFS_BLOCK_RSV_GLOBAL);
  1935. btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
  1936. BTRFS_BLOCK_RSV_DELALLOC);
  1937. btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
  1938. btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
  1939. btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
  1940. btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
  1941. BTRFS_BLOCK_RSV_DELOPS);
  1942. atomic_set(&fs_info->nr_async_submits, 0);
  1943. atomic_set(&fs_info->async_delalloc_pages, 0);
  1944. atomic_set(&fs_info->async_submit_draining, 0);
  1945. atomic_set(&fs_info->nr_async_bios, 0);
  1946. atomic_set(&fs_info->defrag_running, 0);
  1947. atomic64_set(&fs_info->tree_mod_seq, 0);
  1948. fs_info->sb = sb;
  1949. fs_info->max_inline = 8192 * 1024;
  1950. fs_info->metadata_ratio = 0;
  1951. fs_info->defrag_inodes = RB_ROOT;
  1952. fs_info->free_chunk_space = 0;
  1953. fs_info->tree_mod_log = RB_ROOT;
  1954. fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
  1955. /* readahead state */
  1956. INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
  1957. spin_lock_init(&fs_info->reada_lock);
  1958. fs_info->thread_pool_size = min_t(unsigned long,
  1959. num_online_cpus() + 2, 8);
  1960. INIT_LIST_HEAD(&fs_info->ordered_roots);
  1961. spin_lock_init(&fs_info->ordered_root_lock);
  1962. fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
  1963. GFP_NOFS);
  1964. if (!fs_info->delayed_root) {
  1965. err = -ENOMEM;
  1966. goto fail_iput;
  1967. }
  1968. btrfs_init_delayed_root(fs_info->delayed_root);
  1969. mutex_init(&fs_info->scrub_lock);
  1970. atomic_set(&fs_info->scrubs_running, 0);
  1971. atomic_set(&fs_info->scrub_pause_req, 0);
  1972. atomic_set(&fs_info->scrubs_paused, 0);
  1973. atomic_set(&fs_info->scrub_cancel_req, 0);
  1974. init_waitqueue_head(&fs_info->scrub_pause_wait);
  1975. init_rwsem(&fs_info->scrub_super_lock);
  1976. fs_info->scrub_workers_refcnt = 0;
  1977. #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
  1978. fs_info->check_integrity_print_mask = 0;
  1979. #endif
  1980. spin_lock_init(&fs_info->balance_lock);
  1981. mutex_init(&fs_info->balance_mutex);
  1982. atomic_set(&fs_info->balance_running, 0);
  1983. atomic_set(&fs_info->balance_pause_req, 0);
  1984. atomic_set(&fs_info->balance_cancel_req, 0);
  1985. fs_info->balance_ctl = NULL;
  1986. init_waitqueue_head(&fs_info->balance_wait_q);
  1987. sb->s_blocksize = 4096;
  1988. sb->s_blocksize_bits = blksize_bits(4096);
  1989. sb->s_bdi = &fs_info->bdi;
  1990. fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
  1991. set_nlink(fs_info->btree_inode, 1);
  1992. /*
  1993. * we set the i_size on the btree inode to the max possible int.
  1994. * the real end of the address space is determined by all of
  1995. * the devices in the system
  1996. */
  1997. fs_info->btree_inode->i_size = OFFSET_MAX;
  1998. fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
  1999. fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
  2000. RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
  2001. extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
  2002. fs_info->btree_inode->i_mapping);
  2003. BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
  2004. extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
  2005. BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
  2006. BTRFS_I(fs_info->btree_inode)->root = tree_root;
  2007. memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
  2008. sizeof(struct btrfs_key));
  2009. set_bit(BTRFS_INODE_DUMMY,
  2010. &BTRFS_I(fs_info->btree_inode)->runtime_flags);
  2011. insert_inode_hash(fs_info->btree_inode);
  2012. spin_lock_init(&fs_info->block_group_cache_lock);
  2013. fs_info->block_group_cache_tree = RB_ROOT;
  2014. fs_info->first_logical_byte = (u64)-1;
  2015. extent_io_tree_init(&fs_info->freed_extents[0],
  2016. fs_info->btree_inode->i_mapping);
  2017. extent_io_tree_init(&fs_info->freed_extents[1],
  2018. fs_info->btree_inode->i_mapping);
  2019. fs_info->pinned_extents = &fs_info->freed_extents[0];
  2020. fs_info->do_barriers = 1;
  2021. mutex_init(&fs_info->ordered_operations_mutex);
  2022. mutex_init(&fs_info->ordered_extent_flush_mutex);
  2023. mutex_init(&fs_info->tree_log_mutex);
  2024. mutex_init(&fs_info->chunk_mutex);
  2025. mutex_init(&fs_info->transaction_kthread_mutex);
  2026. mutex_init(&fs_info->cleaner_mutex);
  2027. mutex_init(&fs_info->volume_mutex);
  2028. init_rwsem(&fs_info->extent_commit_sem);
  2029. init_rwsem(&fs_info->cleanup_work_sem);
  2030. init_rwsem(&fs_info->subvol_sem);
  2031. sema_init(&fs_info->uuid_tree_rescan_sem, 1);
  2032. fs_info->dev_replace.lock_owner = 0;
  2033. atomic_set(&fs_info->dev_replace.nesting_level, 0);
  2034. mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
  2035. mutex_init(&fs_info->dev_replace.lock_management_lock);
  2036. mutex_init(&fs_info->dev_replace.lock);
  2037. spin_lock_init(&fs_info->qgroup_lock);
  2038. mutex_init(&fs_info->qgroup_ioctl_lock);
  2039. fs_info->qgroup_tree = RB_ROOT;
  2040. INIT_LIST_HEAD(&fs_info->dirty_qgroups);
  2041. fs_info->qgroup_seq = 1;
  2042. fs_info->quota_enabled = 0;
  2043. fs_info->pending_quota_state = 0;
  2044. fs_info->qgroup_ulist = NULL;
  2045. mutex_init(&fs_info->qgroup_rescan_lock);
  2046. btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
  2047. btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
  2048. init_waitqueue_head(&fs_info->transaction_throttle);
  2049. init_waitqueue_head(&fs_info->transaction_wait);
  2050. init_waitqueue_head(&fs_info->transaction_blocked_wait);
  2051. init_waitqueue_head(&fs_info->async_submit_wait);
  2052. ret = btrfs_alloc_stripe_hash_table(fs_info);
  2053. if (ret) {
  2054. err = ret;
  2055. goto fail_alloc;
  2056. }
  2057. __setup_root(4096, 4096, 4096, 4096, tree_root,
  2058. fs_info, BTRFS_ROOT_TREE_OBJECTID);
  2059. invalidate_bdev(fs_devices->latest_bdev);
  2060. /*
  2061. * Read super block and check the signature bytes only
  2062. */
  2063. bh = btrfs_read_dev_super(fs_devices->latest_bdev);
  2064. if (!bh) {
  2065. err = -EINVAL;
  2066. goto fail_alloc;
  2067. }
  2068. /*
  2069. * We want to check superblock checksum, the type is stored inside.
  2070. * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
  2071. */
  2072. if (btrfs_check_super_csum(bh->b_data)) {
  2073. printk(KERN_ERR "btrfs: superblock checksum mismatch\n");
  2074. err = -EINVAL;
  2075. goto fail_alloc;
  2076. }
  2077. /*
  2078. * super_copy is zeroed at allocation time and we never touch the
  2079. * following bytes up to INFO_SIZE, the checksum is calculated from
  2080. * the whole block of INFO_SIZE
  2081. */
  2082. memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
  2083. memcpy(fs_info->super_for_commit, fs_info->super_copy,
  2084. sizeof(*fs_info->super_for_commit));
  2085. brelse(bh);
  2086. memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
  2087. ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
  2088. if (ret) {
  2089. printk(KERN_ERR "btrfs: superblock contains fatal errors\n");
  2090. err = -EINVAL;
  2091. goto fail_alloc;
  2092. }
  2093. disk_super = fs_info->super_copy;
  2094. if (!btrfs_super_root(disk_super))
  2095. goto fail_alloc;
  2096. /* check FS state, whether FS is broken. */
  2097. if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
  2098. set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
  2099. /*
  2100. * run through our array of backup supers and setup
  2101. * our ring pointer to the oldest one
  2102. */
  2103. generation = btrfs_super_generation(disk_super);
  2104. find_oldest_super_backup(fs_info, generation);
  2105. /*
  2106. * In the long term, we'll store the compression type in the super
  2107. * block, and it'll be used for per file compression control.
  2108. */
  2109. fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
  2110. ret = btrfs_parse_options(tree_root, options);
  2111. if (ret) {
  2112. err = ret;
  2113. goto fail_alloc;
  2114. }
  2115. features = btrfs_super_incompat_flags(disk_super) &
  2116. ~BTRFS_FEATURE_INCOMPAT_SUPP;
  2117. if (features) {
  2118. printk(KERN_ERR "BTRFS: couldn't mount because of "
  2119. "unsupported optional features (%Lx).\n",
  2120. features);
  2121. err = -EINVAL;
  2122. goto fail_alloc;
  2123. }
  2124. if (btrfs_super_leafsize(disk_super) !=
  2125. btrfs_super_nodesize(disk_super)) {
  2126. printk(KERN_ERR "BTRFS: couldn't mount because metadata "
  2127. "blocksizes don't match. node %d leaf %d\n",
  2128. btrfs_super_nodesize(disk_super),
  2129. btrfs_super_leafsize(disk_super));
  2130. err = -EINVAL;
  2131. goto fail_alloc;
  2132. }
  2133. if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
  2134. printk(KERN_ERR "BTRFS: couldn't mount because metadata "
  2135. "blocksize (%d) was too large\n",
  2136. btrfs_super_leafsize(disk_super));
  2137. err = -EINVAL;
  2138. goto fail_alloc;
  2139. }
  2140. features = btrfs_super_incompat_flags(disk_super);
  2141. features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
  2142. if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
  2143. features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
  2144. if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
  2145. printk(KERN_ERR "btrfs: has skinny extents\n");
  2146. /*
  2147. * flag our filesystem as having big metadata blocks if
  2148. * they are bigger than the page size
  2149. */
  2150. if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) {
  2151. if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
  2152. printk(KERN_INFO "btrfs flagging fs with big metadata feature\n");
  2153. features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
  2154. }
  2155. nodesize = btrfs_super_nodesize(disk_super);
  2156. leafsize = btrfs_super_leafsize(disk_super);
  2157. sectorsize = btrfs_super_sectorsize(disk_super);
  2158. stripesize = btrfs_super_stripesize(disk_super);
  2159. fs_info->dirty_metadata_batch = leafsize * (1 + ilog2(nr_cpu_ids));
  2160. fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
  2161. /*
  2162. * mixed block groups end up with duplicate but slightly offset
  2163. * extent buffers for the same range. It leads to corruptions
  2164. */
  2165. if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
  2166. (sectorsize != leafsize)) {
  2167. printk(KERN_WARNING "btrfs: unequal leaf/node/sector sizes "
  2168. "are not allowed for mixed block groups on %s\n",
  2169. sb->s_id);
  2170. goto fail_alloc;
  2171. }
  2172. /*
  2173. * Needn't use the lock because there is no other task which will
  2174. * update the flag.
  2175. */
  2176. btrfs_set_super_incompat_flags(disk_super, features);
  2177. features = btrfs_super_compat_ro_flags(disk_super) &
  2178. ~BTRFS_FEATURE_COMPAT_RO_SUPP;
  2179. if (!(sb->s_flags & MS_RDONLY) && features) {
  2180. printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
  2181. "unsupported option features (%Lx).\n",
  2182. features);
  2183. err = -EINVAL;
  2184. goto fail_alloc;
  2185. }
  2186. btrfs_init_workers(&fs_info->generic_worker,
  2187. "genwork", 1, NULL);
  2188. btrfs_init_workers(&fs_info->workers, "worker",
  2189. fs_info->thread_pool_size,
  2190. &fs_info->generic_worker);
  2191. btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
  2192. fs_info->thread_pool_size, NULL);
  2193. btrfs_init_workers(&fs_info->flush_workers, "flush_delalloc",
  2194. fs_info->thread_pool_size, NULL);
  2195. btrfs_init_workers(&fs_info->submit_workers, "submit",
  2196. min_t(u64, fs_devices->num_devices,
  2197. fs_info->thread_pool_size), NULL);
  2198. btrfs_init_workers(&fs_info->caching_workers, "cache",
  2199. fs_info->thread_pool_size, NULL);
  2200. /* a higher idle thresh on the submit workers makes it much more
  2201. * likely that bios will be send down in a sane order to the
  2202. * devices
  2203. */
  2204. fs_info->submit_workers.idle_thresh = 64;
  2205. fs_info->workers.idle_thresh = 16;
  2206. fs_info->workers.ordered = 1;
  2207. fs_info->delalloc_workers.idle_thresh = 2;
  2208. fs_info->delalloc_workers.ordered = 1;
  2209. btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
  2210. &fs_info->generic_worker);
  2211. btrfs_init_workers(&fs_info->endio_workers, "endio",
  2212. fs_info->thread_pool_size,
  2213. &fs_info->generic_worker);
  2214. btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
  2215. fs_info->thread_pool_size,
  2216. &fs_info->generic_worker);
  2217. btrfs_init_workers(&fs_info->endio_meta_write_workers,
  2218. "endio-meta-write", fs_info->thread_pool_size,
  2219. &fs_info->generic_worker);
  2220. btrfs_init_workers(&fs_info->endio_raid56_workers,
  2221. "endio-raid56", fs_info->thread_pool_size,
  2222. &fs_info->generic_worker);
  2223. btrfs_init_workers(&fs_info->rmw_workers,
  2224. "rmw", fs_info->thread_pool_size,
  2225. &fs_info->generic_worker);
  2226. btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
  2227. fs_info->thread_pool_size,
  2228. &fs_info->generic_worker);
  2229. btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
  2230. 1, &fs_info->generic_worker);
  2231. btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
  2232. fs_info->thread_pool_size,
  2233. &fs_info->generic_worker);
  2234. btrfs_init_workers(&fs_info->readahead_workers, "readahead",
  2235. fs_info->thread_pool_size,
  2236. &fs_info->generic_worker);
  2237. btrfs_init_workers(&fs_info->qgroup_rescan_workers, "qgroup-rescan", 1,
  2238. &fs_info->generic_worker);
  2239. /*
  2240. * endios are largely parallel and should have a very
  2241. * low idle thresh
  2242. */
  2243. fs_info->endio_workers.idle_thresh = 4;
  2244. fs_info->endio_meta_workers.idle_thresh = 4;
  2245. fs_info->endio_raid56_workers.idle_thresh = 4;
  2246. fs_info->rmw_workers.idle_thresh = 2;
  2247. fs_info->endio_write_workers.idle_thresh = 2;
  2248. fs_info->endio_meta_write_workers.idle_thresh = 2;
  2249. fs_info->readahead_workers.idle_thresh = 2;
  2250. /*
  2251. * btrfs_start_workers can really only fail because of ENOMEM so just
  2252. * return -ENOMEM if any of these fail.
  2253. */
  2254. ret = btrfs_start_workers(&fs_info->workers);
  2255. ret |= btrfs_start_workers(&fs_info->generic_worker);
  2256. ret |= btrfs_start_workers(&fs_info->submit_workers);
  2257. ret |= btrfs_start_workers(&fs_info->delalloc_workers);
  2258. ret |= btrfs_start_workers(&fs_info->fixup_workers);
  2259. ret |= btrfs_start_workers(&fs_info->endio_workers);
  2260. ret |= btrfs_start_workers(&fs_info->endio_meta_workers);
  2261. ret |= btrfs_start_workers(&fs_info->rmw_workers);
  2262. ret |= btrfs_start_workers(&fs_info->endio_raid56_workers);
  2263. ret |= btrfs_start_workers(&fs_info->endio_meta_write_workers);
  2264. ret |= btrfs_start_workers(&fs_info->endio_write_workers);
  2265. ret |= btrfs_start_workers(&fs_info->endio_freespace_worker);
  2266. ret |= btrfs_start_workers(&fs_info->delayed_workers);
  2267. ret |= btrfs_start_workers(&fs_info->caching_workers);
  2268. ret |= btrfs_start_workers(&fs_info->readahead_workers);
  2269. ret |= btrfs_start_workers(&fs_info->flush_workers);
  2270. ret |= btrfs_start_workers(&fs_info->qgroup_rescan_workers);
  2271. if (ret) {
  2272. err = -ENOMEM;
  2273. goto fail_sb_buffer;
  2274. }
  2275. fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
  2276. fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
  2277. 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
  2278. tree_root->nodesize = nodesize;
  2279. tree_root->leafsize = leafsize;
  2280. tree_root->sectorsize = sectorsize;
  2281. tree_root->stripesize = stripesize;
  2282. sb->s_blocksize = sectorsize;
  2283. sb->s_blocksize_bits = blksize_bits(sectorsize);
  2284. if (btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
  2285. printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
  2286. goto fail_sb_buffer;
  2287. }
  2288. if (sectorsize != PAGE_SIZE) {
  2289. printk(KERN_WARNING "btrfs: Incompatible sector size(%lu) "
  2290. "found on %s\n", (unsigned long)sectorsize, sb->s_id);
  2291. goto fail_sb_buffer;
  2292. }
  2293. mutex_lock(&fs_info->chunk_mutex);
  2294. ret = btrfs_read_sys_array(tree_root);
  2295. mutex_unlock(&fs_info->chunk_mutex);
  2296. if (ret) {
  2297. printk(KERN_WARNING "btrfs: failed to read the system "
  2298. "array on %s\n", sb->s_id);
  2299. goto fail_sb_buffer;
  2300. }
  2301. blocksize = btrfs_level_size(tree_root,
  2302. btrfs_super_chunk_root_level(disk_super));
  2303. generation = btrfs_super_chunk_root_generation(disk_super);
  2304. __setup_root(nodesize, leafsize, sectorsize, stripesize,
  2305. chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
  2306. chunk_root->node = read_tree_block(chunk_root,
  2307. btrfs_super_chunk_root(disk_super),
  2308. blocksize, generation);
  2309. if (!chunk_root->node ||
  2310. !test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
  2311. printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
  2312. sb->s_id);
  2313. goto fail_tree_roots;
  2314. }
  2315. btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
  2316. chunk_root->commit_root = btrfs_root_node(chunk_root);
  2317. read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
  2318. btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
  2319. ret = btrfs_read_chunk_tree(chunk_root);
  2320. if (ret) {
  2321. printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
  2322. sb->s_id);
  2323. goto fail_tree_roots;
  2324. }
  2325. /*
  2326. * keep the device that is marked to be the target device for the
  2327. * dev_replace procedure
  2328. */
  2329. btrfs_close_extra_devices(fs_info, fs_devices, 0);
  2330. if (!fs_devices->latest_bdev) {
  2331. printk(KERN_CRIT "btrfs: failed to read devices on %s\n",
  2332. sb->s_id);
  2333. goto fail_tree_roots;
  2334. }
  2335. retry_root_backup:
  2336. blocksize = btrfs_level_size(tree_root,
  2337. btrfs_super_root_level(disk_super));
  2338. generation = btrfs_super_generation(disk_super);
  2339. tree_root->node = read_tree_block(tree_root,
  2340. btrfs_super_root(disk_super),
  2341. blocksize, generation);
  2342. if (!tree_root->node ||
  2343. !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
  2344. printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
  2345. sb->s_id);
  2346. goto recovery_tree_root;
  2347. }
  2348. btrfs_set_root_node(&tree_root->root_item, tree_root->node);
  2349. tree_root->commit_root = btrfs_root_node(tree_root);
  2350. btrfs_set_root_refs(&tree_root->root_item, 1);
  2351. location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
  2352. location.type = BTRFS_ROOT_ITEM_KEY;
  2353. location.offset = 0;
  2354. extent_root = btrfs_read_tree_root(tree_root, &location);
  2355. if (IS_ERR(extent_root)) {
  2356. ret = PTR_ERR(extent_root);
  2357. goto recovery_tree_root;
  2358. }
  2359. extent_root->track_dirty = 1;
  2360. fs_info->extent_root = extent_root;
  2361. location.objectid = BTRFS_DEV_TREE_OBJECTID;
  2362. dev_root = btrfs_read_tree_root(tree_root, &location);
  2363. if (IS_ERR(dev_root)) {
  2364. ret = PTR_ERR(dev_root);
  2365. goto recovery_tree_root;
  2366. }
  2367. dev_root->track_dirty = 1;
  2368. fs_info->dev_root = dev_root;
  2369. btrfs_init_devices_late(fs_info);
  2370. location.objectid = BTRFS_CSUM_TREE_OBJECTID;
  2371. csum_root = btrfs_read_tree_root(tree_root, &location);
  2372. if (IS_ERR(csum_root)) {
  2373. ret = PTR_ERR(csum_root);
  2374. goto recovery_tree_root;
  2375. }
  2376. csum_root->track_dirty = 1;
  2377. fs_info->csum_root = csum_root;
  2378. location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
  2379. quota_root = btrfs_read_tree_root(tree_root, &location);
  2380. if (!IS_ERR(quota_root)) {
  2381. quota_root->track_dirty = 1;
  2382. fs_info->quota_enabled = 1;
  2383. fs_info->pending_quota_state = 1;
  2384. fs_info->quota_root = quota_root;
  2385. }
  2386. location.objectid = BTRFS_UUID_TREE_OBJECTID;
  2387. uuid_root = btrfs_read_tree_root(tree_root, &location);
  2388. if (IS_ERR(uuid_root)) {
  2389. ret = PTR_ERR(uuid_root);
  2390. if (ret != -ENOENT)
  2391. goto recovery_tree_root;
  2392. create_uuid_tree = true;
  2393. check_uuid_tree = false;
  2394. } else {
  2395. uuid_root->track_dirty = 1;
  2396. fs_info->uuid_root = uuid_root;
  2397. create_uuid_tree = false;
  2398. check_uuid_tree =
  2399. generation != btrfs_super_uuid_tree_generation(disk_super);
  2400. }
  2401. fs_info->generation = generation;
  2402. fs_info->last_trans_committed = generation;
  2403. ret = btrfs_recover_balance(fs_info);
  2404. if (ret) {
  2405. printk(KERN_WARNING "btrfs: failed to recover balance\n");
  2406. goto fail_block_groups;
  2407. }
  2408. ret = btrfs_init_dev_stats(fs_info);
  2409. if (ret) {
  2410. printk(KERN_ERR "btrfs: failed to init dev_stats: %d\n",
  2411. ret);
  2412. goto fail_block_groups;
  2413. }
  2414. ret = btrfs_init_dev_replace(fs_info);
  2415. if (ret) {
  2416. pr_err("btrfs: failed to init dev_replace: %d\n", ret);
  2417. goto fail_block_groups;
  2418. }
  2419. btrfs_close_extra_devices(fs_info, fs_devices, 1);
  2420. ret = btrfs_init_space_info(fs_info);
  2421. if (ret) {
  2422. printk(KERN_ERR "Failed to initial space info: %d\n", ret);
  2423. goto fail_block_groups;
  2424. }
  2425. ret = btrfs_read_block_groups(extent_root);
  2426. if (ret) {
  2427. printk(KERN_ERR "Failed to read block groups: %d\n", ret);
  2428. goto fail_block_groups;
  2429. }
  2430. fs_info->num_tolerated_disk_barrier_failures =
  2431. btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
  2432. if (fs_info->fs_devices->missing_devices >
  2433. fs_info->num_tolerated_disk_barrier_failures &&
  2434. !(sb->s_flags & MS_RDONLY)) {
  2435. printk(KERN_WARNING
  2436. "Btrfs: too many missing devices, writeable mount is not allowed\n");
  2437. goto fail_block_groups;
  2438. }
  2439. fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
  2440. "btrfs-cleaner");
  2441. if (IS_ERR(fs_info->cleaner_kthread))
  2442. goto fail_block_groups;
  2443. fs_info->transaction_kthread = kthread_run(transaction_kthread,
  2444. tree_root,
  2445. "btrfs-transaction");
  2446. if (IS_ERR(fs_info->transaction_kthread))
  2447. goto fail_cleaner;
  2448. if (!btrfs_test_opt(tree_root, SSD) &&
  2449. !btrfs_test_opt(tree_root, NOSSD) &&
  2450. !fs_info->fs_devices->rotating) {
  2451. printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
  2452. "mode\n");
  2453. btrfs_set_opt(fs_info->mount_opt, SSD);
  2454. }
  2455. #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
  2456. if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
  2457. ret = btrfsic_mount(tree_root, fs_devices,
  2458. btrfs_test_opt(tree_root,
  2459. CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
  2460. 1 : 0,
  2461. fs_info->check_integrity_print_mask);
  2462. if (ret)
  2463. printk(KERN_WARNING "btrfs: failed to initialize"
  2464. " integrity check module %s\n", sb->s_id);
  2465. }
  2466. #endif
  2467. ret = btrfs_read_qgroup_config(fs_info);
  2468. if (ret)
  2469. goto fail_trans_kthread;
  2470. /* do not make disk changes in broken FS */
  2471. if (btrfs_super_log_root(disk_super) != 0) {
  2472. u64 bytenr = btrfs_super_log_root(disk_super);
  2473. if (fs_devices->rw_devices == 0) {
  2474. printk(KERN_WARNING "Btrfs log replay required "
  2475. "on RO media\n");
  2476. err = -EIO;
  2477. goto fail_qgroup;
  2478. }
  2479. blocksize =
  2480. btrfs_level_size(tree_root,
  2481. btrfs_super_log_root_level(disk_super));
  2482. log_tree_root = btrfs_alloc_root(fs_info);
  2483. if (!log_tree_root) {
  2484. err = -ENOMEM;
  2485. goto fail_qgroup;
  2486. }
  2487. __setup_root(nodesize, leafsize, sectorsize, stripesize,
  2488. log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
  2489. log_tree_root->node = read_tree_block(tree_root, bytenr,
  2490. blocksize,
  2491. generation + 1);
  2492. if (!log_tree_root->node ||
  2493. !extent_buffer_uptodate(log_tree_root->node)) {
  2494. printk(KERN_ERR "btrfs: failed to read log tree\n");
  2495. free_extent_buffer(log_tree_root->node);
  2496. kfree(log_tree_root);
  2497. goto fail_trans_kthread;
  2498. }
  2499. /* returns with log_tree_root freed on success */
  2500. ret = btrfs_recover_log_trees(log_tree_root);
  2501. if (ret) {
  2502. btrfs_error(tree_root->fs_info, ret,
  2503. "Failed to recover log tree");
  2504. free_extent_buffer(log_tree_root->node);
  2505. kfree(log_tree_root);
  2506. goto fail_trans_kthread;
  2507. }
  2508. if (sb->s_flags & MS_RDONLY) {
  2509. ret = btrfs_commit_super(tree_root);
  2510. if (ret)
  2511. goto fail_trans_kthread;
  2512. }
  2513. }
  2514. ret = btrfs_find_orphan_roots(tree_root);
  2515. if (ret)
  2516. goto fail_trans_kthread;
  2517. if (!(sb->s_flags & MS_RDONLY)) {
  2518. ret = btrfs_cleanup_fs_roots(fs_info);
  2519. if (ret)
  2520. goto fail_trans_kthread;
  2521. ret = btrfs_recover_relocation(tree_root);
  2522. if (ret < 0) {
  2523. printk(KERN_WARNING
  2524. "btrfs: failed to recover relocation\n");
  2525. err = -EINVAL;
  2526. goto fail_qgroup;
  2527. }
  2528. }
  2529. location.objectid = BTRFS_FS_TREE_OBJECTID;
  2530. location.type = BTRFS_ROOT_ITEM_KEY;
  2531. location.offset = 0;
  2532. fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
  2533. if (IS_ERR(fs_info->fs_root)) {
  2534. err = PTR_ERR(fs_info->fs_root);
  2535. goto fail_qgroup;
  2536. }
  2537. if (sb->s_flags & MS_RDONLY)
  2538. return 0;
  2539. down_read(&fs_info->cleanup_work_sem);
  2540. if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
  2541. (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
  2542. up_read(&fs_info->cleanup_work_sem);
  2543. close_ctree(tree_root);
  2544. return ret;
  2545. }
  2546. up_read(&fs_info->cleanup_work_sem);
  2547. ret = btrfs_resume_balance_async(fs_info);
  2548. if (ret) {
  2549. printk(KERN_WARNING "btrfs: failed to resume balance\n");
  2550. close_ctree(tree_root);
  2551. return ret;
  2552. }
  2553. ret = btrfs_resume_dev_replace_async(fs_info);
  2554. if (ret) {
  2555. pr_warn("btrfs: failed to resume dev_replace\n");
  2556. close_ctree(tree_root);
  2557. return ret;
  2558. }
  2559. btrfs_qgroup_rescan_resume(fs_info);
  2560. if (create_uuid_tree) {
  2561. pr_info("btrfs: creating UUID tree\n");
  2562. ret = btrfs_create_uuid_tree(fs_info);
  2563. if (ret) {
  2564. pr_warn("btrfs: failed to create the UUID tree %d\n",
  2565. ret);
  2566. close_ctree(tree_root);
  2567. return ret;
  2568. }
  2569. } else if (check_uuid_tree ||
  2570. btrfs_test_opt(tree_root, RESCAN_UUID_TREE)) {
  2571. pr_info("btrfs: checking UUID tree\n");
  2572. ret = btrfs_check_uuid_tree(fs_info);
  2573. if (ret) {
  2574. pr_warn("btrfs: failed to check the UUID tree %d\n",
  2575. ret);
  2576. close_ctree(tree_root);
  2577. return ret;
  2578. }
  2579. } else {
  2580. fs_info->update_uuid_tree_gen = 1;
  2581. }
  2582. return 0;
  2583. fail_qgroup:
  2584. btrfs_free_qgroup_config(fs_info);
  2585. fail_trans_kthread:
  2586. kthread_stop(fs_info->transaction_kthread);
  2587. btrfs_cleanup_transaction(fs_info->tree_root);
  2588. del_fs_roots(fs_info);
  2589. fail_cleaner:
  2590. kthread_stop(fs_info->cleaner_kthread);
  2591. /*
  2592. * make sure we're done with the btree inode before we stop our
  2593. * kthreads
  2594. */
  2595. filemap_write_and_wait(fs_info->btree_inode->i_mapping);
  2596. fail_block_groups:
  2597. btrfs_put_block_group_cache(fs_info);
  2598. btrfs_free_block_groups(fs_info);
  2599. fail_tree_roots:
  2600. free_root_pointers(fs_info, 1);
  2601. invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
  2602. fail_sb_buffer:
  2603. btrfs_stop_all_workers(fs_info);
  2604. fail_alloc:
  2605. fail_iput:
  2606. btrfs_mapping_tree_free(&fs_info->mapping_tree);
  2607. iput(fs_info->btree_inode);
  2608. fail_delalloc_bytes:
  2609. percpu_counter_destroy(&fs_info->delalloc_bytes);
  2610. fail_dirty_metadata_bytes:
  2611. percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
  2612. fail_bdi:
  2613. bdi_destroy(&fs_info->bdi);
  2614. fail_srcu:
  2615. cleanup_srcu_struct(&fs_info->subvol_srcu);
  2616. fail:
  2617. btrfs_free_stripe_hash_table(fs_info);
  2618. btrfs_close_devices(fs_info->fs_devices);
  2619. return err;
  2620. recovery_tree_root:
  2621. if (!btrfs_test_opt(tree_root, RECOVERY))
  2622. goto fail_tree_roots;
  2623. free_root_pointers(fs_info, 0);
  2624. /* don't use the log in recovery mode, it won't be valid */
  2625. btrfs_set_super_log_root(disk_super, 0);
  2626. /* we can't trust the free space cache either */
  2627. btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
  2628. ret = next_root_backup(fs_info, fs_info->super_copy,
  2629. &num_backups_tried, &backup_index);
  2630. if (ret == -1)
  2631. goto fail_block_groups;
  2632. goto retry_root_backup;
  2633. }
  2634. static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
  2635. {
  2636. if (uptodate) {
  2637. set_buffer_uptodate(bh);
  2638. } else {
  2639. struct btrfs_device *device = (struct btrfs_device *)
  2640. bh->b_private;
  2641. printk_ratelimited_in_rcu(KERN_WARNING "lost page write due to "
  2642. "I/O error on %s\n",
  2643. rcu_str_deref(device->name));
  2644. /* note, we dont' set_buffer_write_io_error because we have
  2645. * our own ways of dealing with the IO errors
  2646. */
  2647. clear_buffer_uptodate(bh);
  2648. btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
  2649. }
  2650. unlock_buffer(bh);
  2651. put_bh(bh);
  2652. }
  2653. struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
  2654. {
  2655. struct buffer_head *bh;
  2656. struct buffer_head *latest = NULL;
  2657. struct btrfs_super_block *super;
  2658. int i;
  2659. u64 transid = 0;
  2660. u64 bytenr;
  2661. /* we would like to check all the supers, but that would make
  2662. * a btrfs mount succeed after a mkfs from a different FS.
  2663. * So, we need to add a special mount option to scan for
  2664. * later supers, using BTRFS_SUPER_MIRROR_MAX instead
  2665. */
  2666. for (i = 0; i < 1; i++) {
  2667. bytenr = btrfs_sb_offset(i);
  2668. if (bytenr + BTRFS_SUPER_INFO_SIZE >=
  2669. i_size_read(bdev->bd_inode))
  2670. break;
  2671. bh = __bread(bdev, bytenr / 4096,
  2672. BTRFS_SUPER_INFO_SIZE);
  2673. if (!bh)
  2674. continue;
  2675. super = (struct btrfs_super_block *)bh->b_data;
  2676. if (btrfs_super_bytenr(super) != bytenr ||
  2677. btrfs_super_magic(super) != BTRFS_MAGIC) {
  2678. brelse(bh);
  2679. continue;
  2680. }
  2681. if (!latest || btrfs_super_generation(super) > transid) {
  2682. brelse(latest);
  2683. latest = bh;
  2684. transid = btrfs_super_generation(super);
  2685. } else {
  2686. brelse(bh);
  2687. }
  2688. }
  2689. return latest;
  2690. }
  2691. /*
  2692. * this should be called twice, once with wait == 0 and
  2693. * once with wait == 1. When wait == 0 is done, all the buffer heads
  2694. * we write are pinned.
  2695. *
  2696. * They are released when wait == 1 is done.
  2697. * max_mirrors must be the same for both runs, and it indicates how
  2698. * many supers on this one device should be written.
  2699. *
  2700. * max_mirrors == 0 means to write them all.
  2701. */
  2702. static int write_dev_supers(struct btrfs_device *device,
  2703. struct btrfs_super_block *sb,
  2704. int do_barriers, int wait, int max_mirrors)
  2705. {
  2706. struct buffer_head *bh;
  2707. int i;
  2708. int ret;
  2709. int errors = 0;
  2710. u32 crc;
  2711. u64 bytenr;
  2712. if (max_mirrors == 0)
  2713. max_mirrors = BTRFS_SUPER_MIRROR_MAX;
  2714. for (i = 0; i < max_mirrors; i++) {
  2715. bytenr = btrfs_sb_offset(i);
  2716. if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
  2717. break;
  2718. if (wait) {
  2719. bh = __find_get_block(device->bdev, bytenr / 4096,
  2720. BTRFS_SUPER_INFO_SIZE);
  2721. if (!bh) {
  2722. errors++;
  2723. continue;
  2724. }
  2725. wait_on_buffer(bh);
  2726. if (!buffer_uptodate(bh))
  2727. errors++;
  2728. /* drop our reference */
  2729. brelse(bh);
  2730. /* drop the reference from the wait == 0 run */
  2731. brelse(bh);
  2732. continue;
  2733. } else {
  2734. btrfs_set_super_bytenr(sb, bytenr);
  2735. crc = ~(u32)0;
  2736. crc = btrfs_csum_data((char *)sb +
  2737. BTRFS_CSUM_SIZE, crc,
  2738. BTRFS_SUPER_INFO_SIZE -
  2739. BTRFS_CSUM_SIZE);
  2740. btrfs_csum_final(crc, sb->csum);
  2741. /*
  2742. * one reference for us, and we leave it for the
  2743. * caller
  2744. */
  2745. bh = __getblk(device->bdev, bytenr / 4096,
  2746. BTRFS_SUPER_INFO_SIZE);
  2747. if (!bh) {
  2748. printk(KERN_ERR "btrfs: couldn't get super "
  2749. "buffer head for bytenr %Lu\n", bytenr);
  2750. errors++;
  2751. continue;
  2752. }
  2753. memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
  2754. /* one reference for submit_bh */
  2755. get_bh(bh);
  2756. set_buffer_uptodate(bh);
  2757. lock_buffer(bh);
  2758. bh->b_end_io = btrfs_end_buffer_write_sync;
  2759. bh->b_private = device;
  2760. }
  2761. /*
  2762. * we fua the first super. The others we allow
  2763. * to go down lazy.
  2764. */
  2765. ret = btrfsic_submit_bh(WRITE_FUA, bh);
  2766. if (ret)
  2767. errors++;
  2768. }
  2769. return errors < i ? 0 : -1;
  2770. }
  2771. /*
  2772. * endio for the write_dev_flush, this will wake anyone waiting
  2773. * for the barrier when it is done
  2774. */
  2775. static void btrfs_end_empty_barrier(struct bio *bio, int err)
  2776. {
  2777. if (err) {
  2778. if (err == -EOPNOTSUPP)
  2779. set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
  2780. clear_bit(BIO_UPTODATE, &bio->bi_flags);
  2781. }
  2782. if (bio->bi_private)
  2783. complete(bio->bi_private);
  2784. bio_put(bio);
  2785. }
  2786. /*
  2787. * trigger flushes for one the devices. If you pass wait == 0, the flushes are
  2788. * sent down. With wait == 1, it waits for the previous flush.
  2789. *
  2790. * any device where the flush fails with eopnotsupp are flagged as not-barrier
  2791. * capable
  2792. */
  2793. static int write_dev_flush(struct btrfs_device *device, int wait)
  2794. {
  2795. struct bio *bio;
  2796. int ret = 0;
  2797. if (device->nobarriers)
  2798. return 0;
  2799. if (wait) {
  2800. bio = device->flush_bio;
  2801. if (!bio)
  2802. return 0;
  2803. wait_for_completion(&device->flush_wait);
  2804. if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
  2805. printk_in_rcu("btrfs: disabling barriers on dev %s\n",
  2806. rcu_str_deref(device->name));
  2807. device->nobarriers = 1;
  2808. } else if (!bio_flagged(bio, BIO_UPTODATE)) {
  2809. ret = -EIO;
  2810. btrfs_dev_stat_inc_and_print(device,
  2811. BTRFS_DEV_STAT_FLUSH_ERRS);
  2812. }
  2813. /* drop the reference from the wait == 0 run */
  2814. bio_put(bio);
  2815. device->flush_bio = NULL;
  2816. return ret;
  2817. }
  2818. /*
  2819. * one reference for us, and we leave it for the
  2820. * caller
  2821. */
  2822. device->flush_bio = NULL;
  2823. bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
  2824. if (!bio)
  2825. return -ENOMEM;
  2826. bio->bi_end_io = btrfs_end_empty_barrier;
  2827. bio->bi_bdev = device->bdev;
  2828. init_completion(&device->flush_wait);
  2829. bio->bi_private = &device->flush_wait;
  2830. device->flush_bio = bio;
  2831. bio_get(bio);
  2832. btrfsic_submit_bio(WRITE_FLUSH, bio);
  2833. return 0;
  2834. }
  2835. /*
  2836. * send an empty flush down to each device in parallel,
  2837. * then wait for them
  2838. */
  2839. static int barrier_all_devices(struct btrfs_fs_info *info)
  2840. {
  2841. struct list_head *head;
  2842. struct btrfs_device *dev;
  2843. int errors_send = 0;
  2844. int errors_wait = 0;
  2845. int ret;
  2846. /* send down all the barriers */
  2847. head = &info->fs_devices->devices;
  2848. list_for_each_entry_rcu(dev, head, dev_list) {
  2849. if (!dev->bdev) {
  2850. errors_send++;
  2851. continue;
  2852. }
  2853. if (!dev->in_fs_metadata || !dev->writeable)
  2854. continue;
  2855. ret = write_dev_flush(dev, 0);
  2856. if (ret)
  2857. errors_send++;
  2858. }
  2859. /* wait for all the barriers */
  2860. list_for_each_entry_rcu(dev, head, dev_list) {
  2861. if (!dev->bdev) {
  2862. errors_wait++;
  2863. continue;
  2864. }
  2865. if (!dev->in_fs_metadata || !dev->writeable)
  2866. continue;
  2867. ret = write_dev_flush(dev, 1);
  2868. if (ret)
  2869. errors_wait++;
  2870. }
  2871. if (errors_send > info->num_tolerated_disk_barrier_failures ||
  2872. errors_wait > info->num_tolerated_disk_barrier_failures)
  2873. return -EIO;
  2874. return 0;
  2875. }
  2876. int btrfs_calc_num_tolerated_disk_barrier_failures(
  2877. struct btrfs_fs_info *fs_info)
  2878. {
  2879. struct btrfs_ioctl_space_info space;
  2880. struct btrfs_space_info *sinfo;
  2881. u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
  2882. BTRFS_BLOCK_GROUP_SYSTEM,
  2883. BTRFS_BLOCK_GROUP_METADATA,
  2884. BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
  2885. int num_types = 4;
  2886. int i;
  2887. int c;
  2888. int num_tolerated_disk_barrier_failures =
  2889. (int)fs_info->fs_devices->num_devices;
  2890. for (i = 0; i < num_types; i++) {
  2891. struct btrfs_space_info *tmp;
  2892. sinfo = NULL;
  2893. rcu_read_lock();
  2894. list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
  2895. if (tmp->flags == types[i]) {
  2896. sinfo = tmp;
  2897. break;
  2898. }
  2899. }
  2900. rcu_read_unlock();
  2901. if (!sinfo)
  2902. continue;
  2903. down_read(&sinfo->groups_sem);
  2904. for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
  2905. if (!list_empty(&sinfo->block_groups[c])) {
  2906. u64 flags;
  2907. btrfs_get_block_group_info(
  2908. &sinfo->block_groups[c], &space);
  2909. if (space.total_bytes == 0 ||
  2910. space.used_bytes == 0)
  2911. continue;
  2912. flags = space.flags;
  2913. /*
  2914. * return
  2915. * 0: if dup, single or RAID0 is configured for
  2916. * any of metadata, system or data, else
  2917. * 1: if RAID5 is configured, or if RAID1 or
  2918. * RAID10 is configured and only two mirrors
  2919. * are used, else
  2920. * 2: if RAID6 is configured, else
  2921. * num_mirrors - 1: if RAID1 or RAID10 is
  2922. * configured and more than
  2923. * 2 mirrors are used.
  2924. */
  2925. if (num_tolerated_disk_barrier_failures > 0 &&
  2926. ((flags & (BTRFS_BLOCK_GROUP_DUP |
  2927. BTRFS_BLOCK_GROUP_RAID0)) ||
  2928. ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)
  2929. == 0)))
  2930. num_tolerated_disk_barrier_failures = 0;
  2931. else if (num_tolerated_disk_barrier_failures > 1) {
  2932. if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
  2933. BTRFS_BLOCK_GROUP_RAID5 |
  2934. BTRFS_BLOCK_GROUP_RAID10)) {
  2935. num_tolerated_disk_barrier_failures = 1;
  2936. } else if (flags &
  2937. BTRFS_BLOCK_GROUP_RAID6) {
  2938. num_tolerated_disk_barrier_failures = 2;
  2939. }
  2940. }
  2941. }
  2942. }
  2943. up_read(&sinfo->groups_sem);
  2944. }
  2945. return num_tolerated_disk_barrier_failures;
  2946. }
  2947. static int write_all_supers(struct btrfs_root *root, int max_mirrors)
  2948. {
  2949. struct list_head *head;
  2950. struct btrfs_device *dev;
  2951. struct btrfs_super_block *sb;
  2952. struct btrfs_dev_item *dev_item;
  2953. int ret;
  2954. int do_barriers;
  2955. int max_errors;
  2956. int total_errors = 0;
  2957. u64 flags;
  2958. do_barriers = !btrfs_test_opt(root, NOBARRIER);
  2959. backup_super_roots(root->fs_info);
  2960. sb = root->fs_info->super_for_commit;
  2961. dev_item = &sb->dev_item;
  2962. mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
  2963. head = &root->fs_info->fs_devices->devices;
  2964. max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
  2965. if (do_barriers) {
  2966. ret = barrier_all_devices(root->fs_info);
  2967. if (ret) {
  2968. mutex_unlock(
  2969. &root->fs_info->fs_devices->device_list_mutex);
  2970. btrfs_error(root->fs_info, ret,
  2971. "errors while submitting device barriers.");
  2972. return ret;
  2973. }
  2974. }
  2975. list_for_each_entry_rcu(dev, head, dev_list) {
  2976. if (!dev->bdev) {
  2977. total_errors++;
  2978. continue;
  2979. }
  2980. if (!dev->in_fs_metadata || !dev->writeable)
  2981. continue;
  2982. btrfs_set_stack_device_generation(dev_item, 0);
  2983. btrfs_set_stack_device_type(dev_item, dev->type);
  2984. btrfs_set_stack_device_id(dev_item, dev->devid);
  2985. btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
  2986. btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
  2987. btrfs_set_stack_device_io_align(dev_item, dev->io_align);
  2988. btrfs_set_stack_device_io_width(dev_item, dev->io_width);
  2989. btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
  2990. memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
  2991. memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
  2992. flags = btrfs_super_flags(sb);
  2993. btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
  2994. ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
  2995. if (ret)
  2996. total_errors++;
  2997. }
  2998. if (total_errors > max_errors) {
  2999. printk(KERN_ERR "btrfs: %d errors while writing supers\n",
  3000. total_errors);
  3001. mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
  3002. /* FUA is masked off if unsupported and can't be the reason */
  3003. btrfs_error(root->fs_info, -EIO,
  3004. "%d errors while writing supers", total_errors);
  3005. return -EIO;
  3006. }
  3007. total_errors = 0;
  3008. list_for_each_entry_rcu(dev, head, dev_list) {
  3009. if (!dev->bdev)
  3010. continue;
  3011. if (!dev->in_fs_metadata || !dev->writeable)
  3012. continue;
  3013. ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
  3014. if (ret)
  3015. total_errors++;
  3016. }
  3017. mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
  3018. if (total_errors > max_errors) {
  3019. btrfs_error(root->fs_info, -EIO,
  3020. "%d errors while writing supers", total_errors);
  3021. return -EIO;
  3022. }
  3023. return 0;
  3024. }
  3025. int write_ctree_super(struct btrfs_trans_handle *trans,
  3026. struct btrfs_root *root, int max_mirrors)
  3027. {
  3028. int ret;
  3029. ret = write_all_supers(root, max_mirrors);
  3030. return ret;
  3031. }
  3032. /* Drop a fs root from the radix tree and free it. */
  3033. void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
  3034. struct btrfs_root *root)
  3035. {
  3036. spin_lock(&fs_info->fs_roots_radix_lock);
  3037. radix_tree_delete(&fs_info->fs_roots_radix,
  3038. (unsigned long)root->root_key.objectid);
  3039. spin_unlock(&fs_info->fs_roots_radix_lock);
  3040. if (btrfs_root_refs(&root->root_item) == 0)
  3041. synchronize_srcu(&fs_info->subvol_srcu);
  3042. if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
  3043. btrfs_free_log(NULL, root);
  3044. btrfs_free_log_root_tree(NULL, fs_info);
  3045. }
  3046. __btrfs_remove_free_space_cache(root->free_ino_pinned);
  3047. __btrfs_remove_free_space_cache(root->free_ino_ctl);
  3048. free_fs_root(root);
  3049. }
  3050. static void free_fs_root(struct btrfs_root *root)
  3051. {
  3052. iput(root->cache_inode);
  3053. WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
  3054. btrfs_free_block_rsv(root, root->orphan_block_rsv);
  3055. root->orphan_block_rsv = NULL;
  3056. if (root->anon_dev)
  3057. free_anon_bdev(root->anon_dev);
  3058. free_extent_buffer(root->node);
  3059. free_extent_buffer(root->commit_root);
  3060. kfree(root->free_ino_ctl);
  3061. kfree(root->free_ino_pinned);
  3062. kfree(root->name);
  3063. btrfs_put_fs_root(root);
  3064. }
  3065. void btrfs_free_fs_root(struct btrfs_root *root)
  3066. {
  3067. free_fs_root(root);
  3068. }
  3069. int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
  3070. {
  3071. u64 root_objectid = 0;
  3072. struct btrfs_root *gang[8];
  3073. int i;
  3074. int ret;
  3075. while (1) {
  3076. ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
  3077. (void **)gang, root_objectid,
  3078. ARRAY_SIZE(gang));
  3079. if (!ret)
  3080. break;
  3081. root_objectid = gang[ret - 1]->root_key.objectid + 1;
  3082. for (i = 0; i < ret; i++) {
  3083. int err;
  3084. root_objectid = gang[i]->root_key.objectid;
  3085. err = btrfs_orphan_cleanup(gang[i]);
  3086. if (err)
  3087. return err;
  3088. }
  3089. root_objectid++;
  3090. }
  3091. return 0;
  3092. }
  3093. int btrfs_commit_super(struct btrfs_root *root)
  3094. {
  3095. struct btrfs_trans_handle *trans;
  3096. int ret;
  3097. mutex_lock(&root->fs_info->cleaner_mutex);
  3098. btrfs_run_delayed_iputs(root);
  3099. mutex_unlock(&root->fs_info->cleaner_mutex);
  3100. wake_up_process(root->fs_info->cleaner_kthread);
  3101. /* wait until ongoing cleanup work done */
  3102. down_write(&root->fs_info->cleanup_work_sem);
  3103. up_write(&root->fs_info->cleanup_work_sem);
  3104. trans = btrfs_join_transaction(root);
  3105. if (IS_ERR(trans))
  3106. return PTR_ERR(trans);
  3107. ret = btrfs_commit_transaction(trans, root);
  3108. if (ret)
  3109. return ret;
  3110. /* run commit again to drop the original snapshot */
  3111. trans = btrfs_join_transaction(root);
  3112. if (IS_ERR(trans))
  3113. return PTR_ERR(trans);
  3114. ret = btrfs_commit_transaction(trans, root);
  3115. if (ret)
  3116. return ret;
  3117. ret = btrfs_write_and_wait_transaction(NULL, root);
  3118. if (ret) {
  3119. btrfs_error(root->fs_info, ret,
  3120. "Failed to sync btree inode to disk.");
  3121. return ret;
  3122. }
  3123. ret = write_ctree_super(NULL, root, 0);
  3124. return ret;
  3125. }
  3126. int close_ctree(struct btrfs_root *root)
  3127. {
  3128. struct btrfs_fs_info *fs_info = root->fs_info;
  3129. int ret;
  3130. fs_info->closing = 1;
  3131. smp_mb();
  3132. /* wait for the uuid_scan task to finish */
  3133. down(&fs_info->uuid_tree_rescan_sem);
  3134. /* avoid complains from lockdep et al., set sem back to initial state */
  3135. up(&fs_info->uuid_tree_rescan_sem);
  3136. /* pause restriper - we want to resume on mount */
  3137. btrfs_pause_balance(fs_info);
  3138. btrfs_dev_replace_suspend_for_unmount(fs_info);
  3139. btrfs_scrub_cancel(fs_info);
  3140. /* wait for any defraggers to finish */
  3141. wait_event(fs_info->transaction_wait,
  3142. (atomic_read(&fs_info->defrag_running) == 0));
  3143. /* clear out the rbtree of defraggable inodes */
  3144. btrfs_cleanup_defrag_inodes(fs_info);
  3145. if (!(fs_info->sb->s_flags & MS_RDONLY)) {
  3146. ret = btrfs_commit_super(root);
  3147. if (ret)
  3148. printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
  3149. }
  3150. if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
  3151. btrfs_error_commit_super(root);
  3152. btrfs_put_block_group_cache(fs_info);
  3153. kthread_stop(fs_info->transaction_kthread);
  3154. kthread_stop(fs_info->cleaner_kthread);
  3155. fs_info->closing = 2;
  3156. smp_mb();
  3157. btrfs_free_qgroup_config(root->fs_info);
  3158. if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
  3159. printk(KERN_INFO "btrfs: at unmount delalloc count %lld\n",
  3160. percpu_counter_sum(&fs_info->delalloc_bytes));
  3161. }
  3162. btrfs_free_block_groups(fs_info);
  3163. btrfs_stop_all_workers(fs_info);
  3164. del_fs_roots(fs_info);
  3165. free_root_pointers(fs_info, 1);
  3166. iput(fs_info->btree_inode);
  3167. #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
  3168. if (btrfs_test_opt(root, CHECK_INTEGRITY))
  3169. btrfsic_unmount(root, fs_info->fs_devices);
  3170. #endif
  3171. btrfs_close_devices(fs_info->fs_devices);
  3172. btrfs_mapping_tree_free(&fs_info->mapping_tree);
  3173. percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
  3174. percpu_counter_destroy(&fs_info->delalloc_bytes);
  3175. bdi_destroy(&fs_info->bdi);
  3176. cleanup_srcu_struct(&fs_info->subvol_srcu);
  3177. btrfs_free_stripe_hash_table(fs_info);
  3178. btrfs_free_block_rsv(root, root->orphan_block_rsv);
  3179. root->orphan_block_rsv = NULL;
  3180. return 0;
  3181. }
  3182. int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
  3183. int atomic)
  3184. {
  3185. int ret;
  3186. struct inode *btree_inode = buf->pages[0]->mapping->host;
  3187. ret = extent_buffer_uptodate(buf);
  3188. if (!ret)
  3189. return ret;
  3190. ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
  3191. parent_transid, atomic);
  3192. if (ret == -EAGAIN)
  3193. return ret;
  3194. return !ret;
  3195. }
  3196. int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
  3197. {
  3198. return set_extent_buffer_uptodate(buf);
  3199. }
  3200. void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
  3201. {
  3202. struct btrfs_root *root;
  3203. u64 transid = btrfs_header_generation(buf);
  3204. int was_dirty;
  3205. #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
  3206. /*
  3207. * This is a fast path so only do this check if we have sanity tests
  3208. * enabled. Normal people shouldn't be marking dummy buffers as dirty
  3209. * outside of the sanity tests.
  3210. */
  3211. if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
  3212. return;
  3213. #endif
  3214. root = BTRFS_I(buf->pages[0]->mapping->host)->root;
  3215. btrfs_assert_tree_locked(buf);
  3216. if (transid != root->fs_info->generation)
  3217. WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
  3218. "found %llu running %llu\n",
  3219. buf->start, transid, root->fs_info->generation);
  3220. was_dirty = set_extent_buffer_dirty(buf);
  3221. if (!was_dirty)
  3222. __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
  3223. buf->len,
  3224. root->fs_info->dirty_metadata_batch);
  3225. }
  3226. static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
  3227. int flush_delayed)
  3228. {
  3229. /*
  3230. * looks as though older kernels can get into trouble with
  3231. * this code, they end up stuck in balance_dirty_pages forever
  3232. */
  3233. int ret;
  3234. if (current->flags & PF_MEMALLOC)
  3235. return;
  3236. if (flush_delayed)
  3237. btrfs_balance_delayed_items(root);
  3238. ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
  3239. BTRFS_DIRTY_METADATA_THRESH);
  3240. if (ret > 0) {
  3241. balance_dirty_pages_ratelimited(
  3242. root->fs_info->btree_inode->i_mapping);
  3243. }
  3244. return;
  3245. }
  3246. void btrfs_btree_balance_dirty(struct btrfs_root *root)
  3247. {
  3248. __btrfs_btree_balance_dirty(root, 1);
  3249. }
  3250. void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
  3251. {
  3252. __btrfs_btree_balance_dirty(root, 0);
  3253. }
  3254. int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
  3255. {
  3256. struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
  3257. return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
  3258. }
  3259. static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
  3260. int read_only)
  3261. {
  3262. /*
  3263. * Placeholder for checks
  3264. */
  3265. return 0;
  3266. }
  3267. static void btrfs_error_commit_super(struct btrfs_root *root)
  3268. {
  3269. mutex_lock(&root->fs_info->cleaner_mutex);
  3270. btrfs_run_delayed_iputs(root);
  3271. mutex_unlock(&root->fs_info->cleaner_mutex);
  3272. down_write(&root->fs_info->cleanup_work_sem);
  3273. up_write(&root->fs_info->cleanup_work_sem);
  3274. /* cleanup FS via transaction */
  3275. btrfs_cleanup_transaction(root);
  3276. }
  3277. static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
  3278. struct btrfs_root *root)
  3279. {
  3280. struct btrfs_inode *btrfs_inode;
  3281. struct list_head splice;
  3282. INIT_LIST_HEAD(&splice);
  3283. mutex_lock(&root->fs_info->ordered_operations_mutex);
  3284. spin_lock(&root->fs_info->ordered_root_lock);
  3285. list_splice_init(&t->ordered_operations, &splice);
  3286. while (!list_empty(&splice)) {
  3287. btrfs_inode = list_entry(splice.next, struct btrfs_inode,
  3288. ordered_operations);
  3289. list_del_init(&btrfs_inode->ordered_operations);
  3290. spin_unlock(&root->fs_info->ordered_root_lock);
  3291. btrfs_invalidate_inodes(btrfs_inode->root);
  3292. spin_lock(&root->fs_info->ordered_root_lock);
  3293. }
  3294. spin_unlock(&root->fs_info->ordered_root_lock);
  3295. mutex_unlock(&root->fs_info->ordered_operations_mutex);
  3296. }
  3297. static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
  3298. {
  3299. struct btrfs_ordered_extent *ordered;
  3300. spin_lock(&root->ordered_extent_lock);
  3301. /*
  3302. * This will just short circuit the ordered completion stuff which will
  3303. * make sure the ordered extent gets properly cleaned up.
  3304. */
  3305. list_for_each_entry(ordered, &root->ordered_extents,
  3306. root_extent_list)
  3307. set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
  3308. spin_unlock(&root->ordered_extent_lock);
  3309. }
  3310. static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
  3311. {
  3312. struct btrfs_root *root;
  3313. struct list_head splice;
  3314. INIT_LIST_HEAD(&splice);
  3315. spin_lock(&fs_info->ordered_root_lock);
  3316. list_splice_init(&fs_info->ordered_roots, &splice);
  3317. while (!list_empty(&splice)) {
  3318. root = list_first_entry(&splice, struct btrfs_root,
  3319. ordered_root);
  3320. list_del_init(&root->ordered_root);
  3321. btrfs_destroy_ordered_extents(root);
  3322. cond_resched_lock(&fs_info->ordered_root_lock);
  3323. }
  3324. spin_unlock(&fs_info->ordered_root_lock);
  3325. }
  3326. static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
  3327. struct btrfs_root *root)
  3328. {
  3329. struct rb_node *node;
  3330. struct btrfs_delayed_ref_root *delayed_refs;
  3331. struct btrfs_delayed_ref_node *ref;
  3332. int ret = 0;
  3333. delayed_refs = &trans->delayed_refs;
  3334. spin_lock(&delayed_refs->lock);
  3335. if (delayed_refs->num_entries == 0) {
  3336. spin_unlock(&delayed_refs->lock);
  3337. printk(KERN_INFO "delayed_refs has NO entry\n");
  3338. return ret;
  3339. }
  3340. while ((node = rb_first(&delayed_refs->root)) != NULL) {
  3341. struct btrfs_delayed_ref_head *head = NULL;
  3342. bool pin_bytes = false;
  3343. ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
  3344. atomic_set(&ref->refs, 1);
  3345. if (btrfs_delayed_ref_is_head(ref)) {
  3346. head = btrfs_delayed_node_to_head(ref);
  3347. if (!mutex_trylock(&head->mutex)) {
  3348. atomic_inc(&ref->refs);
  3349. spin_unlock(&delayed_refs->lock);
  3350. /* Need to wait for the delayed ref to run */
  3351. mutex_lock(&head->mutex);
  3352. mutex_unlock(&head->mutex);
  3353. btrfs_put_delayed_ref(ref);
  3354. spin_lock(&delayed_refs->lock);
  3355. continue;
  3356. }
  3357. if (head->must_insert_reserved)
  3358. pin_bytes = true;
  3359. btrfs_free_delayed_extent_op(head->extent_op);
  3360. delayed_refs->num_heads--;
  3361. if (list_empty(&head->cluster))
  3362. delayed_refs->num_heads_ready--;
  3363. list_del_init(&head->cluster);
  3364. }
  3365. ref->in_tree = 0;
  3366. rb_erase(&ref->rb_node, &delayed_refs->root);
  3367. delayed_refs->num_entries--;
  3368. spin_unlock(&delayed_refs->lock);
  3369. if (head) {
  3370. if (pin_bytes)
  3371. btrfs_pin_extent(root, ref->bytenr,
  3372. ref->num_bytes, 1);
  3373. mutex_unlock(&head->mutex);
  3374. }
  3375. btrfs_put_delayed_ref(ref);
  3376. cond_resched();
  3377. spin_lock(&delayed_refs->lock);
  3378. }
  3379. spin_unlock(&delayed_refs->lock);
  3380. return ret;
  3381. }
  3382. static void btrfs_evict_pending_snapshots(struct btrfs_transaction *t)
  3383. {
  3384. struct btrfs_pending_snapshot *snapshot;
  3385. struct list_head splice;
  3386. INIT_LIST_HEAD(&splice);
  3387. list_splice_init(&t->pending_snapshots, &splice);
  3388. while (!list_empty(&splice)) {
  3389. snapshot = list_entry(splice.next,
  3390. struct btrfs_pending_snapshot,
  3391. list);
  3392. snapshot->error = -ECANCELED;
  3393. list_del_init(&snapshot->list);
  3394. }
  3395. }
  3396. static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
  3397. {
  3398. struct btrfs_inode *btrfs_inode;
  3399. struct list_head splice;
  3400. INIT_LIST_HEAD(&splice);
  3401. spin_lock(&root->delalloc_lock);
  3402. list_splice_init(&root->delalloc_inodes, &splice);
  3403. while (!list_empty(&splice)) {
  3404. btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
  3405. delalloc_inodes);
  3406. list_del_init(&btrfs_inode->delalloc_inodes);
  3407. clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
  3408. &btrfs_inode->runtime_flags);
  3409. spin_unlock(&root->delalloc_lock);
  3410. btrfs_invalidate_inodes(btrfs_inode->root);
  3411. spin_lock(&root->delalloc_lock);
  3412. }
  3413. spin_unlock(&root->delalloc_lock);
  3414. }
  3415. static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
  3416. {
  3417. struct btrfs_root *root;
  3418. struct list_head splice;
  3419. INIT_LIST_HEAD(&splice);
  3420. spin_lock(&fs_info->delalloc_root_lock);
  3421. list_splice_init(&fs_info->delalloc_roots, &splice);
  3422. while (!list_empty(&splice)) {
  3423. root = list_first_entry(&splice, struct btrfs_root,
  3424. delalloc_root);
  3425. list_del_init(&root->delalloc_root);
  3426. root = btrfs_grab_fs_root(root);
  3427. BUG_ON(!root);
  3428. spin_unlock(&fs_info->delalloc_root_lock);
  3429. btrfs_destroy_delalloc_inodes(root);
  3430. btrfs_put_fs_root(root);
  3431. spin_lock(&fs_info->delalloc_root_lock);
  3432. }
  3433. spin_unlock(&fs_info->delalloc_root_lock);
  3434. }
  3435. static int btrfs_destroy_marked_extents(struct btrfs_root *root,
  3436. struct extent_io_tree *dirty_pages,
  3437. int mark)
  3438. {
  3439. int ret;
  3440. struct extent_buffer *eb;
  3441. u64 start = 0;
  3442. u64 end;
  3443. while (1) {
  3444. ret = find_first_extent_bit(dirty_pages, start, &start, &end,
  3445. mark, NULL);
  3446. if (ret)
  3447. break;
  3448. clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
  3449. while (start <= end) {
  3450. eb = btrfs_find_tree_block(root, start,
  3451. root->leafsize);
  3452. start += root->leafsize;
  3453. if (!eb)
  3454. continue;
  3455. wait_on_extent_buffer_writeback(eb);
  3456. if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
  3457. &eb->bflags))
  3458. clear_extent_buffer_dirty(eb);
  3459. free_extent_buffer_stale(eb);
  3460. }
  3461. }
  3462. return ret;
  3463. }
  3464. static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
  3465. struct extent_io_tree *pinned_extents)
  3466. {
  3467. struct extent_io_tree *unpin;
  3468. u64 start;
  3469. u64 end;
  3470. int ret;
  3471. bool loop = true;
  3472. unpin = pinned_extents;
  3473. again:
  3474. while (1) {
  3475. ret = find_first_extent_bit(unpin, 0, &start, &end,
  3476. EXTENT_DIRTY, NULL);
  3477. if (ret)
  3478. break;
  3479. /* opt_discard */
  3480. if (btrfs_test_opt(root, DISCARD))
  3481. ret = btrfs_error_discard_extent(root, start,
  3482. end + 1 - start,
  3483. NULL);
  3484. clear_extent_dirty(unpin, start, end, GFP_NOFS);
  3485. btrfs_error_unpin_extent_range(root, start, end);
  3486. cond_resched();
  3487. }
  3488. if (loop) {
  3489. if (unpin == &root->fs_info->freed_extents[0])
  3490. unpin = &root->fs_info->freed_extents[1];
  3491. else
  3492. unpin = &root->fs_info->freed_extents[0];
  3493. loop = false;
  3494. goto again;
  3495. }
  3496. return 0;
  3497. }
  3498. void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
  3499. struct btrfs_root *root)
  3500. {
  3501. btrfs_destroy_delayed_refs(cur_trans, root);
  3502. btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv,
  3503. cur_trans->dirty_pages.dirty_bytes);
  3504. cur_trans->state = TRANS_STATE_COMMIT_START;
  3505. wake_up(&root->fs_info->transaction_blocked_wait);
  3506. btrfs_evict_pending_snapshots(cur_trans);
  3507. cur_trans->state = TRANS_STATE_UNBLOCKED;
  3508. wake_up(&root->fs_info->transaction_wait);
  3509. btrfs_destroy_delayed_inodes(root);
  3510. btrfs_assert_delayed_root_empty(root);
  3511. btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
  3512. EXTENT_DIRTY);
  3513. btrfs_destroy_pinned_extent(root,
  3514. root->fs_info->pinned_extents);
  3515. cur_trans->state =TRANS_STATE_COMPLETED;
  3516. wake_up(&cur_trans->commit_wait);
  3517. /*
  3518. memset(cur_trans, 0, sizeof(*cur_trans));
  3519. kmem_cache_free(btrfs_transaction_cachep, cur_trans);
  3520. */
  3521. }
  3522. static int btrfs_cleanup_transaction(struct btrfs_root *root)
  3523. {
  3524. struct btrfs_transaction *t;
  3525. LIST_HEAD(list);
  3526. mutex_lock(&root->fs_info->transaction_kthread_mutex);
  3527. spin_lock(&root->fs_info->trans_lock);
  3528. list_splice_init(&root->fs_info->trans_list, &list);
  3529. root->fs_info->running_transaction = NULL;
  3530. spin_unlock(&root->fs_info->trans_lock);
  3531. while (!list_empty(&list)) {
  3532. t = list_entry(list.next, struct btrfs_transaction, list);
  3533. btrfs_destroy_ordered_operations(t, root);
  3534. btrfs_destroy_all_ordered_extents(root->fs_info);
  3535. btrfs_destroy_delayed_refs(t, root);
  3536. /*
  3537. * FIXME: cleanup wait for commit
  3538. * We needn't acquire the lock here, because we are during
  3539. * the umount, there is no other task which will change it.
  3540. */
  3541. t->state = TRANS_STATE_COMMIT_START;
  3542. smp_mb();
  3543. if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
  3544. wake_up(&root->fs_info->transaction_blocked_wait);
  3545. btrfs_evict_pending_snapshots(t);
  3546. t->state = TRANS_STATE_UNBLOCKED;
  3547. smp_mb();
  3548. if (waitqueue_active(&root->fs_info->transaction_wait))
  3549. wake_up(&root->fs_info->transaction_wait);
  3550. btrfs_destroy_delayed_inodes(root);
  3551. btrfs_assert_delayed_root_empty(root);
  3552. btrfs_destroy_all_delalloc_inodes(root->fs_info);
  3553. btrfs_destroy_marked_extents(root, &t->dirty_pages,
  3554. EXTENT_DIRTY);
  3555. btrfs_destroy_pinned_extent(root,
  3556. root->fs_info->pinned_extents);
  3557. t->state = TRANS_STATE_COMPLETED;
  3558. smp_mb();
  3559. if (waitqueue_active(&t->commit_wait))
  3560. wake_up(&t->commit_wait);
  3561. atomic_set(&t->use_count, 0);
  3562. list_del_init(&t->list);
  3563. memset(t, 0, sizeof(*t));
  3564. kmem_cache_free(btrfs_transaction_cachep, t);
  3565. }
  3566. mutex_unlock(&root->fs_info->transaction_kthread_mutex);
  3567. return 0;
  3568. }
  3569. static struct extent_io_ops btree_extent_io_ops = {
  3570. .readpage_end_io_hook = btree_readpage_end_io_hook,
  3571. .readpage_io_failed_hook = btree_io_failed_hook,
  3572. .submit_bio_hook = btree_submit_bio_hook,
  3573. /* note we're sharing with inode.c for the merge bio hook */
  3574. .merge_bio_hook = btrfs_merge_bio_hook,
  3575. };