disk-io.c 106 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919920921922923924925926927928929930931932933934935936937938939940941942943944945946947948949950951952953954955956957958959960961962963964965966967968969970971972973974975976977978979980981982983984985986987988989990991992993994995996997998999100010011002100310041005100610071008100910101011101210131014101510161017101810191020102110221023102410251026102710281029103010311032103310341035103610371038103910401041104210431044104510461047104810491050105110521053105410551056105710581059106010611062106310641065106610671068106910701071107210731074107510761077107810791080108110821083108410851086108710881089109010911092109310941095109610971098109911001101110211031104110511061107110811091110111111121113111411151116111711181119112011211122112311241125112611271128112911301131113211331134113511361137113811391140114111421143114411451146114711481149115011511152115311541155115611571158115911601161116211631164116511661167116811691170117111721173117411751176117711781179118011811182118311841185118611871188118911901191119211931194119511961197119811991200120112021203120412051206120712081209121012111212121312141215121612171218121912201221122212231224122512261227122812291230123112321233123412351236123712381239124012411242124312441245124612471248124912501251125212531254125512561257125812591260126112621263126412651266126712681269127012711272127312741275127612771278127912801281128212831284128512861287128812891290129112921293129412951296129712981299130013011302130313041305130613071308130913101311131213131314131513161317131813191320132113221323132413251326132713281329133013311332133313341335133613371338133913401341134213431344134513461347134813491350135113521353135413551356135713581359136013611362136313641365136613671368136913701371137213731374137513761377137813791380138113821383138413851386138713881389139013911392139313941395139613971398139914001401140214031404140514061407140814091410141114121413141414151416141714181419142014211422142314241425142614271428142914301431143214331434143514361437143814391440144114421443144414451446144714481449145014511452145314541455145614571458145914601461146214631464146514661467146814691470147114721473147414751476147714781479148014811482148314841485148614871488148914901491149214931494149514961497149814991500150115021503150415051506150715081509151015111512151315141515151615171518151915201521152215231524152515261527152815291530153115321533153415351536153715381539154015411542154315441545154615471548154915501551155215531554155515561557155815591560156115621563156415651566156715681569157015711572157315741575157615771578157915801581158215831584158515861587158815891590159115921593159415951596159715981599160016011602160316041605160616071608160916101611161216131614161516161617161816191620162116221623162416251626162716281629163016311632163316341635163616371638163916401641164216431644164516461647164816491650165116521653165416551656165716581659166016611662166316641665166616671668166916701671167216731674167516761677167816791680168116821683168416851686168716881689169016911692169316941695169616971698169917001701170217031704170517061707170817091710171117121713171417151716171717181719172017211722172317241725172617271728172917301731173217331734173517361737173817391740174117421743174417451746174717481749175017511752175317541755175617571758175917601761176217631764176517661767176817691770177117721773177417751776177717781779178017811782178317841785178617871788178917901791179217931794179517961797179817991800180118021803180418051806180718081809181018111812181318141815181618171818181918201821182218231824182518261827182818291830183118321833183418351836183718381839184018411842184318441845184618471848184918501851185218531854185518561857185818591860186118621863186418651866186718681869187018711872187318741875187618771878187918801881188218831884188518861887188818891890189118921893189418951896189718981899190019011902190319041905190619071908190919101911191219131914191519161917191819191920192119221923192419251926192719281929193019311932193319341935193619371938193919401941194219431944194519461947194819491950195119521953195419551956195719581959196019611962196319641965196619671968196919701971197219731974197519761977197819791980198119821983198419851986198719881989199019911992199319941995199619971998199920002001200220032004200520062007200820092010201120122013201420152016201720182019202020212022202320242025202620272028202920302031203220332034203520362037203820392040204120422043204420452046204720482049205020512052205320542055205620572058205920602061206220632064206520662067206820692070207120722073207420752076207720782079208020812082208320842085208620872088208920902091209220932094209520962097209820992100210121022103210421052106210721082109211021112112211321142115211621172118211921202121212221232124212521262127212821292130213121322133213421352136213721382139214021412142214321442145214621472148214921502151215221532154215521562157215821592160216121622163216421652166216721682169217021712172217321742175217621772178217921802181218221832184218521862187218821892190219121922193219421952196219721982199220022012202220322042205220622072208220922102211221222132214221522162217221822192220222122222223222422252226222722282229223022312232223322342235223622372238223922402241224222432244224522462247224822492250225122522253225422552256225722582259226022612262226322642265226622672268226922702271227222732274227522762277227822792280228122822283228422852286228722882289229022912292229322942295229622972298229923002301230223032304230523062307230823092310231123122313231423152316231723182319232023212322232323242325232623272328232923302331233223332334233523362337233823392340234123422343234423452346234723482349235023512352235323542355235623572358235923602361236223632364236523662367236823692370237123722373237423752376237723782379238023812382238323842385238623872388238923902391239223932394239523962397239823992400240124022403240424052406240724082409241024112412241324142415241624172418241924202421242224232424242524262427242824292430243124322433243424352436243724382439244024412442244324442445244624472448244924502451245224532454245524562457245824592460246124622463246424652466246724682469247024712472247324742475247624772478247924802481248224832484248524862487248824892490249124922493249424952496249724982499250025012502250325042505250625072508250925102511251225132514251525162517251825192520252125222523252425252526252725282529253025312532253325342535253625372538253925402541254225432544254525462547254825492550255125522553255425552556255725582559256025612562256325642565256625672568256925702571257225732574257525762577257825792580258125822583258425852586258725882589259025912592259325942595259625972598259926002601260226032604260526062607260826092610261126122613261426152616261726182619262026212622262326242625262626272628262926302631263226332634263526362637263826392640264126422643264426452646264726482649265026512652265326542655265626572658265926602661266226632664266526662667266826692670267126722673267426752676267726782679268026812682268326842685268626872688268926902691269226932694269526962697269826992700270127022703270427052706270727082709271027112712271327142715271627172718271927202721272227232724272527262727272827292730273127322733273427352736273727382739274027412742274327442745274627472748274927502751275227532754275527562757275827592760276127622763276427652766276727682769277027712772277327742775277627772778277927802781278227832784278527862787278827892790279127922793279427952796279727982799280028012802280328042805280628072808280928102811281228132814281528162817281828192820282128222823282428252826282728282829283028312832283328342835283628372838283928402841284228432844284528462847284828492850285128522853285428552856285728582859286028612862286328642865286628672868286928702871287228732874287528762877287828792880288128822883288428852886288728882889289028912892289328942895289628972898289929002901290229032904290529062907290829092910291129122913291429152916291729182919292029212922292329242925292629272928292929302931293229332934293529362937293829392940294129422943294429452946294729482949295029512952295329542955295629572958295929602961296229632964296529662967296829692970297129722973297429752976297729782979298029812982298329842985298629872988298929902991299229932994299529962997299829993000300130023003300430053006300730083009301030113012301330143015301630173018301930203021302230233024302530263027302830293030303130323033303430353036303730383039304030413042304330443045304630473048304930503051305230533054305530563057305830593060306130623063306430653066306730683069307030713072307330743075307630773078307930803081308230833084308530863087308830893090309130923093309430953096309730983099310031013102310331043105310631073108310931103111311231133114311531163117311831193120312131223123312431253126312731283129313031313132313331343135313631373138313931403141314231433144314531463147314831493150315131523153315431553156315731583159316031613162316331643165316631673168316931703171317231733174317531763177317831793180318131823183318431853186318731883189319031913192319331943195319631973198319932003201320232033204320532063207320832093210321132123213321432153216321732183219322032213222322332243225322632273228322932303231323232333234323532363237323832393240324132423243324432453246324732483249325032513252325332543255325632573258325932603261326232633264326532663267326832693270327132723273327432753276327732783279328032813282328332843285328632873288328932903291329232933294329532963297329832993300330133023303330433053306330733083309331033113312331333143315331633173318331933203321332233233324332533263327332833293330333133323333333433353336333733383339334033413342334333443345334633473348334933503351335233533354335533563357335833593360336133623363336433653366336733683369337033713372337333743375337633773378337933803381338233833384338533863387338833893390339133923393339433953396339733983399340034013402340334043405340634073408340934103411341234133414341534163417341834193420342134223423342434253426342734283429343034313432343334343435343634373438343934403441344234433444344534463447344834493450345134523453345434553456345734583459346034613462346334643465346634673468346934703471347234733474347534763477347834793480348134823483348434853486348734883489349034913492349334943495349634973498349935003501350235033504350535063507350835093510351135123513351435153516351735183519352035213522352335243525352635273528352935303531353235333534353535363537353835393540354135423543354435453546354735483549355035513552355335543555355635573558355935603561356235633564356535663567356835693570357135723573357435753576357735783579358035813582358335843585358635873588358935903591359235933594359535963597359835993600360136023603360436053606360736083609361036113612361336143615361636173618361936203621362236233624362536263627362836293630363136323633363436353636363736383639364036413642364336443645364636473648364936503651365236533654365536563657365836593660366136623663366436653666366736683669367036713672367336743675367636773678367936803681368236833684368536863687368836893690369136923693369436953696369736983699370037013702370337043705370637073708370937103711371237133714371537163717371837193720372137223723372437253726372737283729373037313732373337343735373637373738373937403741374237433744374537463747374837493750375137523753375437553756375737583759376037613762376337643765376637673768376937703771377237733774377537763777377837793780378137823783378437853786378737883789379037913792379337943795379637973798379938003801380238033804380538063807380838093810381138123813381438153816381738183819382038213822382338243825382638273828382938303831383238333834383538363837383838393840384138423843384438453846384738483849385038513852385338543855385638573858385938603861386238633864386538663867386838693870387138723873387438753876387738783879388038813882388338843885388638873888388938903891389238933894389538963897
  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 <asm/unaligned.h>
  33. #include "compat.h"
  34. #include "ctree.h"
  35. #include "disk-io.h"
  36. #include "transaction.h"
  37. #include "btrfs_inode.h"
  38. #include "volumes.h"
  39. #include "print-tree.h"
  40. #include "async-thread.h"
  41. #include "locking.h"
  42. #include "tree-log.h"
  43. #include "free-space-cache.h"
  44. #include "inode-map.h"
  45. #include "check-integrity.h"
  46. #include "rcu-string.h"
  47. #include "dev-replace.h"
  48. #ifdef CONFIG_X86
  49. #include <asm/cpufeature.h>
  50. #endif
  51. static struct extent_io_ops btree_extent_io_ops;
  52. static void end_workqueue_fn(struct btrfs_work *work);
  53. static void free_fs_root(struct btrfs_root *root);
  54. static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
  55. int read_only);
  56. static void btrfs_destroy_ordered_operations(struct btrfs_root *root);
  57. static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
  58. static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
  59. struct btrfs_root *root);
  60. static void btrfs_destroy_pending_snapshots(struct btrfs_transaction *t);
  61. static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
  62. static int btrfs_destroy_marked_extents(struct btrfs_root *root,
  63. struct extent_io_tree *dirty_pages,
  64. int mark);
  65. static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
  66. struct extent_io_tree *pinned_extents);
  67. /*
  68. * end_io_wq structs are used to do processing in task context when an IO is
  69. * complete. This is used during reads to verify checksums, and it is used
  70. * by writes to insert metadata for new file extents after IO is complete.
  71. */
  72. struct end_io_wq {
  73. struct bio *bio;
  74. bio_end_io_t *end_io;
  75. void *private;
  76. struct btrfs_fs_info *info;
  77. int error;
  78. int metadata;
  79. struct list_head list;
  80. struct btrfs_work work;
  81. };
  82. /*
  83. * async submit bios are used to offload expensive checksumming
  84. * onto the worker threads. They checksum file and metadata bios
  85. * just before they are sent down the IO stack.
  86. */
  87. struct async_submit_bio {
  88. struct inode *inode;
  89. struct bio *bio;
  90. struct list_head list;
  91. extent_submit_bio_hook_t *submit_bio_start;
  92. extent_submit_bio_hook_t *submit_bio_done;
  93. int rw;
  94. int mirror_num;
  95. unsigned long bio_flags;
  96. /*
  97. * bio_offset is optional, can be used if the pages in the bio
  98. * can't tell us where in the file the bio should go
  99. */
  100. u64 bio_offset;
  101. struct btrfs_work work;
  102. int error;
  103. };
  104. /*
  105. * Lockdep class keys for extent_buffer->lock's in this root. For a given
  106. * eb, the lockdep key is determined by the btrfs_root it belongs to and
  107. * the level the eb occupies in the tree.
  108. *
  109. * Different roots are used for different purposes and may nest inside each
  110. * other and they require separate keysets. As lockdep keys should be
  111. * static, assign keysets according to the purpose of the root as indicated
  112. * by btrfs_root->objectid. This ensures that all special purpose roots
  113. * have separate keysets.
  114. *
  115. * Lock-nesting across peer nodes is always done with the immediate parent
  116. * node locked thus preventing deadlock. As lockdep doesn't know this, use
  117. * subclass to avoid triggering lockdep warning in such cases.
  118. *
  119. * The key is set by the readpage_end_io_hook after the buffer has passed
  120. * csum validation but before the pages are unlocked. It is also set by
  121. * btrfs_init_new_buffer on freshly allocated blocks.
  122. *
  123. * We also add a check to make sure the highest level of the tree is the
  124. * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
  125. * needs update as well.
  126. */
  127. #ifdef CONFIG_DEBUG_LOCK_ALLOC
  128. # if BTRFS_MAX_LEVEL != 8
  129. # error
  130. # endif
  131. static struct btrfs_lockdep_keyset {
  132. u64 id; /* root objectid */
  133. const char *name_stem; /* lock name stem */
  134. char names[BTRFS_MAX_LEVEL + 1][20];
  135. struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
  136. } btrfs_lockdep_keysets[] = {
  137. { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
  138. { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
  139. { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
  140. { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
  141. { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
  142. { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
  143. { .id = BTRFS_ORPHAN_OBJECTID, .name_stem = "orphan" },
  144. { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
  145. { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
  146. { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
  147. { .id = 0, .name_stem = "tree" },
  148. };
  149. void __init btrfs_init_lockdep(void)
  150. {
  151. int i, j;
  152. /* initialize lockdep class names */
  153. for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
  154. struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
  155. for (j = 0; j < ARRAY_SIZE(ks->names); j++)
  156. snprintf(ks->names[j], sizeof(ks->names[j]),
  157. "btrfs-%s-%02d", ks->name_stem, j);
  158. }
  159. }
  160. void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
  161. int level)
  162. {
  163. struct btrfs_lockdep_keyset *ks;
  164. BUG_ON(level >= ARRAY_SIZE(ks->keys));
  165. /* find the matching keyset, id 0 is the default entry */
  166. for (ks = btrfs_lockdep_keysets; ks->id; ks++)
  167. if (ks->id == objectid)
  168. break;
  169. lockdep_set_class_and_name(&eb->lock,
  170. &ks->keys[level], ks->names[level]);
  171. }
  172. #endif
  173. /*
  174. * extents on the btree inode are pretty simple, there's one extent
  175. * that covers the entire device
  176. */
  177. static struct extent_map *btree_get_extent(struct inode *inode,
  178. struct page *page, size_t pg_offset, u64 start, u64 len,
  179. int create)
  180. {
  181. struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
  182. struct extent_map *em;
  183. int ret;
  184. read_lock(&em_tree->lock);
  185. em = lookup_extent_mapping(em_tree, start, len);
  186. if (em) {
  187. em->bdev =
  188. BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
  189. read_unlock(&em_tree->lock);
  190. goto out;
  191. }
  192. read_unlock(&em_tree->lock);
  193. em = alloc_extent_map();
  194. if (!em) {
  195. em = ERR_PTR(-ENOMEM);
  196. goto out;
  197. }
  198. em->start = 0;
  199. em->len = (u64)-1;
  200. em->block_len = (u64)-1;
  201. em->block_start = 0;
  202. em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
  203. write_lock(&em_tree->lock);
  204. ret = add_extent_mapping(em_tree, em);
  205. if (ret == -EEXIST) {
  206. free_extent_map(em);
  207. em = lookup_extent_mapping(em_tree, start, len);
  208. if (!em)
  209. em = ERR_PTR(-EIO);
  210. } else if (ret) {
  211. free_extent_map(em);
  212. em = ERR_PTR(ret);
  213. }
  214. write_unlock(&em_tree->lock);
  215. out:
  216. return em;
  217. }
  218. u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
  219. {
  220. return crc32c(seed, data, len);
  221. }
  222. void btrfs_csum_final(u32 crc, char *result)
  223. {
  224. put_unaligned_le32(~crc, result);
  225. }
  226. /*
  227. * compute the csum for a btree block, and either verify it or write it
  228. * into the csum field of the block.
  229. */
  230. static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
  231. int verify)
  232. {
  233. u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
  234. char *result = NULL;
  235. unsigned long len;
  236. unsigned long cur_len;
  237. unsigned long offset = BTRFS_CSUM_SIZE;
  238. char *kaddr;
  239. unsigned long map_start;
  240. unsigned long map_len;
  241. int err;
  242. u32 crc = ~(u32)0;
  243. unsigned long inline_result;
  244. len = buf->len - offset;
  245. while (len > 0) {
  246. err = map_private_extent_buffer(buf, offset, 32,
  247. &kaddr, &map_start, &map_len);
  248. if (err)
  249. return 1;
  250. cur_len = min(len, map_len - (offset - map_start));
  251. crc = btrfs_csum_data(root, kaddr + offset - map_start,
  252. crc, cur_len);
  253. len -= cur_len;
  254. offset += cur_len;
  255. }
  256. if (csum_size > sizeof(inline_result)) {
  257. result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
  258. if (!result)
  259. return 1;
  260. } else {
  261. result = (char *)&inline_result;
  262. }
  263. btrfs_csum_final(crc, result);
  264. if (verify) {
  265. if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
  266. u32 val;
  267. u32 found = 0;
  268. memcpy(&found, result, csum_size);
  269. read_extent_buffer(buf, &val, 0, csum_size);
  270. printk_ratelimited(KERN_INFO "btrfs: %s checksum verify "
  271. "failed on %llu wanted %X found %X "
  272. "level %d\n",
  273. root->fs_info->sb->s_id,
  274. (unsigned long long)buf->start, val, found,
  275. btrfs_header_level(buf));
  276. if (result != (char *)&inline_result)
  277. kfree(result);
  278. return 1;
  279. }
  280. } else {
  281. write_extent_buffer(buf, result, 0, csum_size);
  282. }
  283. if (result != (char *)&inline_result)
  284. kfree(result);
  285. return 0;
  286. }
  287. /*
  288. * we can't consider a given block up to date unless the transid of the
  289. * block matches the transid in the parent node's pointer. This is how we
  290. * detect blocks that either didn't get written at all or got written
  291. * in the wrong place.
  292. */
  293. static int verify_parent_transid(struct extent_io_tree *io_tree,
  294. struct extent_buffer *eb, u64 parent_transid,
  295. int atomic)
  296. {
  297. struct extent_state *cached_state = NULL;
  298. int ret;
  299. if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
  300. return 0;
  301. if (atomic)
  302. return -EAGAIN;
  303. lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
  304. 0, &cached_state);
  305. if (extent_buffer_uptodate(eb) &&
  306. btrfs_header_generation(eb) == parent_transid) {
  307. ret = 0;
  308. goto out;
  309. }
  310. printk_ratelimited("parent transid verify failed on %llu wanted %llu "
  311. "found %llu\n",
  312. (unsigned long long)eb->start,
  313. (unsigned long long)parent_transid,
  314. (unsigned long long)btrfs_header_generation(eb));
  315. ret = 1;
  316. clear_extent_buffer_uptodate(eb);
  317. out:
  318. unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
  319. &cached_state, GFP_NOFS);
  320. return ret;
  321. }
  322. /*
  323. * helper to read a given tree block, doing retries as required when
  324. * the checksums don't match and we have alternate mirrors to try.
  325. */
  326. static int btree_read_extent_buffer_pages(struct btrfs_root *root,
  327. struct extent_buffer *eb,
  328. u64 start, u64 parent_transid)
  329. {
  330. struct extent_io_tree *io_tree;
  331. int failed = 0;
  332. int ret;
  333. int num_copies = 0;
  334. int mirror_num = 0;
  335. int failed_mirror = 0;
  336. clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
  337. io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
  338. while (1) {
  339. ret = read_extent_buffer_pages(io_tree, eb, start,
  340. WAIT_COMPLETE,
  341. btree_get_extent, mirror_num);
  342. if (!ret) {
  343. if (!verify_parent_transid(io_tree, eb,
  344. parent_transid, 0))
  345. break;
  346. else
  347. ret = -EIO;
  348. }
  349. /*
  350. * This buffer's crc is fine, but its contents are corrupted, so
  351. * there is no reason to read the other copies, they won't be
  352. * any less wrong.
  353. */
  354. if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
  355. break;
  356. num_copies = btrfs_num_copies(root->fs_info,
  357. eb->start, eb->len);
  358. if (num_copies == 1)
  359. break;
  360. if (!failed_mirror) {
  361. failed = 1;
  362. failed_mirror = eb->read_mirror;
  363. }
  364. mirror_num++;
  365. if (mirror_num == failed_mirror)
  366. mirror_num++;
  367. if (mirror_num > num_copies)
  368. break;
  369. }
  370. if (failed && !ret && failed_mirror)
  371. repair_eb_io_failure(root, eb, failed_mirror);
  372. return ret;
  373. }
  374. /*
  375. * checksum a dirty tree block before IO. This has extra checks to make sure
  376. * we only fill in the checksum field in the first page of a multi-page block
  377. */
  378. static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
  379. {
  380. struct extent_io_tree *tree;
  381. u64 start = page_offset(page);
  382. u64 found_start;
  383. struct extent_buffer *eb;
  384. tree = &BTRFS_I(page->mapping->host)->io_tree;
  385. eb = (struct extent_buffer *)page->private;
  386. if (page != eb->pages[0])
  387. return 0;
  388. found_start = btrfs_header_bytenr(eb);
  389. if (found_start != start) {
  390. WARN_ON(1);
  391. return 0;
  392. }
  393. if (!PageUptodate(page)) {
  394. WARN_ON(1);
  395. return 0;
  396. }
  397. csum_tree_block(root, eb, 0);
  398. return 0;
  399. }
  400. static int check_tree_block_fsid(struct btrfs_root *root,
  401. struct extent_buffer *eb)
  402. {
  403. struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
  404. u8 fsid[BTRFS_UUID_SIZE];
  405. int ret = 1;
  406. read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
  407. BTRFS_FSID_SIZE);
  408. while (fs_devices) {
  409. if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
  410. ret = 0;
  411. break;
  412. }
  413. fs_devices = fs_devices->seed;
  414. }
  415. return ret;
  416. }
  417. #define CORRUPT(reason, eb, root, slot) \
  418. printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
  419. "root=%llu, slot=%d\n", reason, \
  420. (unsigned long long)btrfs_header_bytenr(eb), \
  421. (unsigned long long)root->objectid, slot)
  422. static noinline int check_leaf(struct btrfs_root *root,
  423. struct extent_buffer *leaf)
  424. {
  425. struct btrfs_key key;
  426. struct btrfs_key leaf_key;
  427. u32 nritems = btrfs_header_nritems(leaf);
  428. int slot;
  429. if (nritems == 0)
  430. return 0;
  431. /* Check the 0 item */
  432. if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
  433. BTRFS_LEAF_DATA_SIZE(root)) {
  434. CORRUPT("invalid item offset size pair", leaf, root, 0);
  435. return -EIO;
  436. }
  437. /*
  438. * Check to make sure each items keys are in the correct order and their
  439. * offsets make sense. We only have to loop through nritems-1 because
  440. * we check the current slot against the next slot, which verifies the
  441. * next slot's offset+size makes sense and that the current's slot
  442. * offset is correct.
  443. */
  444. for (slot = 0; slot < nritems - 1; slot++) {
  445. btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
  446. btrfs_item_key_to_cpu(leaf, &key, slot + 1);
  447. /* Make sure the keys are in the right order */
  448. if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
  449. CORRUPT("bad key order", leaf, root, slot);
  450. return -EIO;
  451. }
  452. /*
  453. * Make sure the offset and ends are right, remember that the
  454. * item data starts at the end of the leaf and grows towards the
  455. * front.
  456. */
  457. if (btrfs_item_offset_nr(leaf, slot) !=
  458. btrfs_item_end_nr(leaf, slot + 1)) {
  459. CORRUPT("slot offset bad", leaf, root, slot);
  460. return -EIO;
  461. }
  462. /*
  463. * Check to make sure that we don't point outside of the leaf,
  464. * just incase all the items are consistent to eachother, but
  465. * all point outside of the leaf.
  466. */
  467. if (btrfs_item_end_nr(leaf, slot) >
  468. BTRFS_LEAF_DATA_SIZE(root)) {
  469. CORRUPT("slot end outside of leaf", leaf, root, slot);
  470. return -EIO;
  471. }
  472. }
  473. return 0;
  474. }
  475. struct extent_buffer *find_eb_for_page(struct extent_io_tree *tree,
  476. struct page *page, int max_walk)
  477. {
  478. struct extent_buffer *eb;
  479. u64 start = page_offset(page);
  480. u64 target = start;
  481. u64 min_start;
  482. if (start < max_walk)
  483. min_start = 0;
  484. else
  485. min_start = start - max_walk;
  486. while (start >= min_start) {
  487. eb = find_extent_buffer(tree, start, 0);
  488. if (eb) {
  489. /*
  490. * we found an extent buffer and it contains our page
  491. * horray!
  492. */
  493. if (eb->start <= target &&
  494. eb->start + eb->len > target)
  495. return eb;
  496. /* we found an extent buffer that wasn't for us */
  497. free_extent_buffer(eb);
  498. return NULL;
  499. }
  500. if (start == 0)
  501. break;
  502. start -= PAGE_CACHE_SIZE;
  503. }
  504. return NULL;
  505. }
  506. static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
  507. struct extent_state *state, int mirror)
  508. {
  509. struct extent_io_tree *tree;
  510. u64 found_start;
  511. int found_level;
  512. struct extent_buffer *eb;
  513. struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
  514. int ret = 0;
  515. int reads_done;
  516. if (!page->private)
  517. goto out;
  518. tree = &BTRFS_I(page->mapping->host)->io_tree;
  519. eb = (struct extent_buffer *)page->private;
  520. /* the pending IO might have been the only thing that kept this buffer
  521. * in memory. Make sure we have a ref for all this other checks
  522. */
  523. extent_buffer_get(eb);
  524. reads_done = atomic_dec_and_test(&eb->io_pages);
  525. if (!reads_done)
  526. goto err;
  527. eb->read_mirror = mirror;
  528. if (test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
  529. ret = -EIO;
  530. goto err;
  531. }
  532. found_start = btrfs_header_bytenr(eb);
  533. if (found_start != eb->start) {
  534. printk_ratelimited(KERN_INFO "btrfs bad tree block start "
  535. "%llu %llu\n",
  536. (unsigned long long)found_start,
  537. (unsigned long long)eb->start);
  538. ret = -EIO;
  539. goto err;
  540. }
  541. if (check_tree_block_fsid(root, eb)) {
  542. printk_ratelimited(KERN_INFO "btrfs bad fsid on block %llu\n",
  543. (unsigned long long)eb->start);
  544. ret = -EIO;
  545. goto err;
  546. }
  547. found_level = btrfs_header_level(eb);
  548. btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
  549. eb, found_level);
  550. ret = csum_tree_block(root, eb, 1);
  551. if (ret) {
  552. ret = -EIO;
  553. goto err;
  554. }
  555. /*
  556. * If this is a leaf block and it is corrupt, set the corrupt bit so
  557. * that we don't try and read the other copies of this block, just
  558. * return -EIO.
  559. */
  560. if (found_level == 0 && check_leaf(root, eb)) {
  561. set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
  562. ret = -EIO;
  563. }
  564. if (!ret)
  565. set_extent_buffer_uptodate(eb);
  566. err:
  567. if (test_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) {
  568. clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags);
  569. btree_readahead_hook(root, eb, eb->start, ret);
  570. }
  571. if (ret)
  572. clear_extent_buffer_uptodate(eb);
  573. free_extent_buffer(eb);
  574. out:
  575. return ret;
  576. }
  577. static int btree_io_failed_hook(struct page *page, int failed_mirror)
  578. {
  579. struct extent_buffer *eb;
  580. struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
  581. eb = (struct extent_buffer *)page->private;
  582. set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
  583. eb->read_mirror = failed_mirror;
  584. if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
  585. btree_readahead_hook(root, eb, eb->start, -EIO);
  586. return -EIO; /* we fixed nothing */
  587. }
  588. static void end_workqueue_bio(struct bio *bio, int err)
  589. {
  590. struct end_io_wq *end_io_wq = bio->bi_private;
  591. struct btrfs_fs_info *fs_info;
  592. fs_info = end_io_wq->info;
  593. end_io_wq->error = err;
  594. end_io_wq->work.func = end_workqueue_fn;
  595. end_io_wq->work.flags = 0;
  596. if (bio->bi_rw & REQ_WRITE) {
  597. if (end_io_wq->metadata == 1)
  598. btrfs_queue_worker(&fs_info->endio_meta_write_workers,
  599. &end_io_wq->work);
  600. else if (end_io_wq->metadata == 2)
  601. btrfs_queue_worker(&fs_info->endio_freespace_worker,
  602. &end_io_wq->work);
  603. else
  604. btrfs_queue_worker(&fs_info->endio_write_workers,
  605. &end_io_wq->work);
  606. } else {
  607. if (end_io_wq->metadata)
  608. btrfs_queue_worker(&fs_info->endio_meta_workers,
  609. &end_io_wq->work);
  610. else
  611. btrfs_queue_worker(&fs_info->endio_workers,
  612. &end_io_wq->work);
  613. }
  614. }
  615. /*
  616. * For the metadata arg you want
  617. *
  618. * 0 - if data
  619. * 1 - if normal metadta
  620. * 2 - if writing to the free space cache area
  621. */
  622. int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
  623. int metadata)
  624. {
  625. struct end_io_wq *end_io_wq;
  626. end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
  627. if (!end_io_wq)
  628. return -ENOMEM;
  629. end_io_wq->private = bio->bi_private;
  630. end_io_wq->end_io = bio->bi_end_io;
  631. end_io_wq->info = info;
  632. end_io_wq->error = 0;
  633. end_io_wq->bio = bio;
  634. end_io_wq->metadata = metadata;
  635. bio->bi_private = end_io_wq;
  636. bio->bi_end_io = end_workqueue_bio;
  637. return 0;
  638. }
  639. unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
  640. {
  641. unsigned long limit = min_t(unsigned long,
  642. info->workers.max_workers,
  643. info->fs_devices->open_devices);
  644. return 256 * limit;
  645. }
  646. static void run_one_async_start(struct btrfs_work *work)
  647. {
  648. struct async_submit_bio *async;
  649. int ret;
  650. async = container_of(work, struct async_submit_bio, work);
  651. ret = async->submit_bio_start(async->inode, async->rw, async->bio,
  652. async->mirror_num, async->bio_flags,
  653. async->bio_offset);
  654. if (ret)
  655. async->error = ret;
  656. }
  657. static void run_one_async_done(struct btrfs_work *work)
  658. {
  659. struct btrfs_fs_info *fs_info;
  660. struct async_submit_bio *async;
  661. int limit;
  662. async = container_of(work, struct async_submit_bio, work);
  663. fs_info = BTRFS_I(async->inode)->root->fs_info;
  664. limit = btrfs_async_submit_limit(fs_info);
  665. limit = limit * 2 / 3;
  666. if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
  667. waitqueue_active(&fs_info->async_submit_wait))
  668. wake_up(&fs_info->async_submit_wait);
  669. /* If an error occured we just want to clean up the bio and move on */
  670. if (async->error) {
  671. bio_endio(async->bio, async->error);
  672. return;
  673. }
  674. async->submit_bio_done(async->inode, async->rw, async->bio,
  675. async->mirror_num, async->bio_flags,
  676. async->bio_offset);
  677. }
  678. static void run_one_async_free(struct btrfs_work *work)
  679. {
  680. struct async_submit_bio *async;
  681. async = container_of(work, struct async_submit_bio, work);
  682. kfree(async);
  683. }
  684. int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
  685. int rw, struct bio *bio, int mirror_num,
  686. unsigned long bio_flags,
  687. u64 bio_offset,
  688. extent_submit_bio_hook_t *submit_bio_start,
  689. extent_submit_bio_hook_t *submit_bio_done)
  690. {
  691. struct async_submit_bio *async;
  692. async = kmalloc(sizeof(*async), GFP_NOFS);
  693. if (!async)
  694. return -ENOMEM;
  695. async->inode = inode;
  696. async->rw = rw;
  697. async->bio = bio;
  698. async->mirror_num = mirror_num;
  699. async->submit_bio_start = submit_bio_start;
  700. async->submit_bio_done = submit_bio_done;
  701. async->work.func = run_one_async_start;
  702. async->work.ordered_func = run_one_async_done;
  703. async->work.ordered_free = run_one_async_free;
  704. async->work.flags = 0;
  705. async->bio_flags = bio_flags;
  706. async->bio_offset = bio_offset;
  707. async->error = 0;
  708. atomic_inc(&fs_info->nr_async_submits);
  709. if (rw & REQ_SYNC)
  710. btrfs_set_work_high_prio(&async->work);
  711. btrfs_queue_worker(&fs_info->workers, &async->work);
  712. while (atomic_read(&fs_info->async_submit_draining) &&
  713. atomic_read(&fs_info->nr_async_submits)) {
  714. wait_event(fs_info->async_submit_wait,
  715. (atomic_read(&fs_info->nr_async_submits) == 0));
  716. }
  717. return 0;
  718. }
  719. static int btree_csum_one_bio(struct bio *bio)
  720. {
  721. struct bio_vec *bvec = bio->bi_io_vec;
  722. int bio_index = 0;
  723. struct btrfs_root *root;
  724. int ret = 0;
  725. WARN_ON(bio->bi_vcnt <= 0);
  726. while (bio_index < bio->bi_vcnt) {
  727. root = BTRFS_I(bvec->bv_page->mapping->host)->root;
  728. ret = csum_dirty_buffer(root, bvec->bv_page);
  729. if (ret)
  730. break;
  731. bio_index++;
  732. bvec++;
  733. }
  734. return ret;
  735. }
  736. static int __btree_submit_bio_start(struct inode *inode, int rw,
  737. struct bio *bio, int mirror_num,
  738. unsigned long bio_flags,
  739. u64 bio_offset)
  740. {
  741. /*
  742. * when we're called for a write, we're already in the async
  743. * submission context. Just jump into btrfs_map_bio
  744. */
  745. return btree_csum_one_bio(bio);
  746. }
  747. static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
  748. int mirror_num, unsigned long bio_flags,
  749. u64 bio_offset)
  750. {
  751. int ret;
  752. /*
  753. * when we're called for a write, we're already in the async
  754. * submission context. Just jump into btrfs_map_bio
  755. */
  756. ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
  757. if (ret)
  758. bio_endio(bio, ret);
  759. return ret;
  760. }
  761. static int check_async_write(struct inode *inode, unsigned long bio_flags)
  762. {
  763. if (bio_flags & EXTENT_BIO_TREE_LOG)
  764. return 0;
  765. #ifdef CONFIG_X86
  766. if (cpu_has_xmm4_2)
  767. return 0;
  768. #endif
  769. return 1;
  770. }
  771. static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
  772. int mirror_num, unsigned long bio_flags,
  773. u64 bio_offset)
  774. {
  775. int async = check_async_write(inode, bio_flags);
  776. int ret;
  777. if (!(rw & REQ_WRITE)) {
  778. /*
  779. * called for a read, do the setup so that checksum validation
  780. * can happen in the async kernel threads
  781. */
  782. ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
  783. bio, 1);
  784. if (ret)
  785. goto out_w_error;
  786. ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
  787. mirror_num, 0);
  788. } else if (!async) {
  789. ret = btree_csum_one_bio(bio);
  790. if (ret)
  791. goto out_w_error;
  792. ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
  793. mirror_num, 0);
  794. } else {
  795. /*
  796. * kthread helpers are used to submit writes so that
  797. * checksumming can happen in parallel across all CPUs
  798. */
  799. ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
  800. inode, rw, bio, mirror_num, 0,
  801. bio_offset,
  802. __btree_submit_bio_start,
  803. __btree_submit_bio_done);
  804. }
  805. if (ret) {
  806. out_w_error:
  807. bio_endio(bio, ret);
  808. }
  809. return ret;
  810. }
  811. #ifdef CONFIG_MIGRATION
  812. static int btree_migratepage(struct address_space *mapping,
  813. struct page *newpage, struct page *page,
  814. enum migrate_mode mode)
  815. {
  816. /*
  817. * we can't safely write a btree page from here,
  818. * we haven't done the locking hook
  819. */
  820. if (PageDirty(page))
  821. return -EAGAIN;
  822. /*
  823. * Buffers may be managed in a filesystem specific way.
  824. * We must have no buffers or drop them.
  825. */
  826. if (page_has_private(page) &&
  827. !try_to_release_page(page, GFP_KERNEL))
  828. return -EAGAIN;
  829. return migrate_page(mapping, newpage, page, mode);
  830. }
  831. #endif
  832. static int btree_writepages(struct address_space *mapping,
  833. struct writeback_control *wbc)
  834. {
  835. struct extent_io_tree *tree;
  836. tree = &BTRFS_I(mapping->host)->io_tree;
  837. if (wbc->sync_mode == WB_SYNC_NONE) {
  838. struct btrfs_root *root = BTRFS_I(mapping->host)->root;
  839. u64 num_dirty;
  840. unsigned long thresh = 32 * 1024 * 1024;
  841. if (wbc->for_kupdate)
  842. return 0;
  843. /* this is a bit racy, but that's ok */
  844. num_dirty = root->fs_info->dirty_metadata_bytes;
  845. if (num_dirty < thresh)
  846. return 0;
  847. }
  848. return btree_write_cache_pages(mapping, wbc);
  849. }
  850. static int btree_readpage(struct file *file, struct page *page)
  851. {
  852. struct extent_io_tree *tree;
  853. tree = &BTRFS_I(page->mapping->host)->io_tree;
  854. return extent_read_full_page(tree, page, btree_get_extent, 0);
  855. }
  856. static int btree_releasepage(struct page *page, gfp_t gfp_flags)
  857. {
  858. if (PageWriteback(page) || PageDirty(page))
  859. return 0;
  860. /*
  861. * We need to mask out eg. __GFP_HIGHMEM and __GFP_DMA32 as we're doing
  862. * slab allocation from alloc_extent_state down the callchain where
  863. * it'd hit a BUG_ON as those flags are not allowed.
  864. */
  865. gfp_flags &= ~GFP_SLAB_BUG_MASK;
  866. return try_release_extent_buffer(page, gfp_flags);
  867. }
  868. static void btree_invalidatepage(struct page *page, unsigned long offset)
  869. {
  870. struct extent_io_tree *tree;
  871. tree = &BTRFS_I(page->mapping->host)->io_tree;
  872. extent_invalidatepage(tree, page, offset);
  873. btree_releasepage(page, GFP_NOFS);
  874. if (PagePrivate(page)) {
  875. printk(KERN_WARNING "btrfs warning page private not zero "
  876. "on page %llu\n", (unsigned long long)page_offset(page));
  877. ClearPagePrivate(page);
  878. set_page_private(page, 0);
  879. page_cache_release(page);
  880. }
  881. }
  882. static int btree_set_page_dirty(struct page *page)
  883. {
  884. #ifdef DEBUG
  885. struct extent_buffer *eb;
  886. BUG_ON(!PagePrivate(page));
  887. eb = (struct extent_buffer *)page->private;
  888. BUG_ON(!eb);
  889. BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
  890. BUG_ON(!atomic_read(&eb->refs));
  891. btrfs_assert_tree_locked(eb);
  892. #endif
  893. return __set_page_dirty_nobuffers(page);
  894. }
  895. static const struct address_space_operations btree_aops = {
  896. .readpage = btree_readpage,
  897. .writepages = btree_writepages,
  898. .releasepage = btree_releasepage,
  899. .invalidatepage = btree_invalidatepage,
  900. #ifdef CONFIG_MIGRATION
  901. .migratepage = btree_migratepage,
  902. #endif
  903. .set_page_dirty = btree_set_page_dirty,
  904. };
  905. int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
  906. u64 parent_transid)
  907. {
  908. struct extent_buffer *buf = NULL;
  909. struct inode *btree_inode = root->fs_info->btree_inode;
  910. int ret = 0;
  911. buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
  912. if (!buf)
  913. return 0;
  914. read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
  915. buf, 0, WAIT_NONE, btree_get_extent, 0);
  916. free_extent_buffer(buf);
  917. return ret;
  918. }
  919. int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
  920. int mirror_num, struct extent_buffer **eb)
  921. {
  922. struct extent_buffer *buf = NULL;
  923. struct inode *btree_inode = root->fs_info->btree_inode;
  924. struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
  925. int ret;
  926. buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
  927. if (!buf)
  928. return 0;
  929. set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
  930. ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
  931. btree_get_extent, mirror_num);
  932. if (ret) {
  933. free_extent_buffer(buf);
  934. return ret;
  935. }
  936. if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
  937. free_extent_buffer(buf);
  938. return -EIO;
  939. } else if (extent_buffer_uptodate(buf)) {
  940. *eb = buf;
  941. } else {
  942. free_extent_buffer(buf);
  943. }
  944. return 0;
  945. }
  946. struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
  947. u64 bytenr, u32 blocksize)
  948. {
  949. struct inode *btree_inode = root->fs_info->btree_inode;
  950. struct extent_buffer *eb;
  951. eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
  952. bytenr, blocksize);
  953. return eb;
  954. }
  955. struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
  956. u64 bytenr, u32 blocksize)
  957. {
  958. struct inode *btree_inode = root->fs_info->btree_inode;
  959. struct extent_buffer *eb;
  960. eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
  961. bytenr, blocksize);
  962. return eb;
  963. }
  964. int btrfs_write_tree_block(struct extent_buffer *buf)
  965. {
  966. return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
  967. buf->start + buf->len - 1);
  968. }
  969. int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
  970. {
  971. return filemap_fdatawait_range(buf->pages[0]->mapping,
  972. buf->start, buf->start + buf->len - 1);
  973. }
  974. struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
  975. u32 blocksize, u64 parent_transid)
  976. {
  977. struct extent_buffer *buf = NULL;
  978. int ret;
  979. buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
  980. if (!buf)
  981. return NULL;
  982. ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
  983. return buf;
  984. }
  985. void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
  986. struct extent_buffer *buf)
  987. {
  988. if (btrfs_header_generation(buf) ==
  989. root->fs_info->running_transaction->transid) {
  990. btrfs_assert_tree_locked(buf);
  991. if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
  992. spin_lock(&root->fs_info->delalloc_lock);
  993. if (root->fs_info->dirty_metadata_bytes >= buf->len)
  994. root->fs_info->dirty_metadata_bytes -= buf->len;
  995. else {
  996. spin_unlock(&root->fs_info->delalloc_lock);
  997. btrfs_panic(root->fs_info, -EOVERFLOW,
  998. "Can't clear %lu bytes from "
  999. " dirty_mdatadata_bytes (%llu)",
  1000. buf->len,
  1001. root->fs_info->dirty_metadata_bytes);
  1002. }
  1003. spin_unlock(&root->fs_info->delalloc_lock);
  1004. /* ugh, clear_extent_buffer_dirty needs to lock the page */
  1005. btrfs_set_lock_blocking(buf);
  1006. clear_extent_buffer_dirty(buf);
  1007. }
  1008. }
  1009. }
  1010. static void __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
  1011. u32 stripesize, struct btrfs_root *root,
  1012. struct btrfs_fs_info *fs_info,
  1013. u64 objectid)
  1014. {
  1015. root->node = NULL;
  1016. root->commit_root = NULL;
  1017. root->sectorsize = sectorsize;
  1018. root->nodesize = nodesize;
  1019. root->leafsize = leafsize;
  1020. root->stripesize = stripesize;
  1021. root->ref_cows = 0;
  1022. root->track_dirty = 0;
  1023. root->in_radix = 0;
  1024. root->orphan_item_inserted = 0;
  1025. root->orphan_cleanup_state = 0;
  1026. root->objectid = objectid;
  1027. root->last_trans = 0;
  1028. root->highest_objectid = 0;
  1029. root->name = NULL;
  1030. root->inode_tree = RB_ROOT;
  1031. INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
  1032. root->block_rsv = NULL;
  1033. root->orphan_block_rsv = NULL;
  1034. INIT_LIST_HEAD(&root->dirty_list);
  1035. INIT_LIST_HEAD(&root->root_list);
  1036. INIT_LIST_HEAD(&root->logged_list[0]);
  1037. INIT_LIST_HEAD(&root->logged_list[1]);
  1038. spin_lock_init(&root->orphan_lock);
  1039. spin_lock_init(&root->inode_lock);
  1040. spin_lock_init(&root->accounting_lock);
  1041. spin_lock_init(&root->log_extents_lock[0]);
  1042. spin_lock_init(&root->log_extents_lock[1]);
  1043. mutex_init(&root->objectid_mutex);
  1044. mutex_init(&root->log_mutex);
  1045. init_waitqueue_head(&root->log_writer_wait);
  1046. init_waitqueue_head(&root->log_commit_wait[0]);
  1047. init_waitqueue_head(&root->log_commit_wait[1]);
  1048. atomic_set(&root->log_commit[0], 0);
  1049. atomic_set(&root->log_commit[1], 0);
  1050. atomic_set(&root->log_writers, 0);
  1051. atomic_set(&root->log_batch, 0);
  1052. atomic_set(&root->orphan_inodes, 0);
  1053. root->log_transid = 0;
  1054. root->last_log_commit = 0;
  1055. extent_io_tree_init(&root->dirty_log_pages,
  1056. fs_info->btree_inode->i_mapping);
  1057. memset(&root->root_key, 0, sizeof(root->root_key));
  1058. memset(&root->root_item, 0, sizeof(root->root_item));
  1059. memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
  1060. memset(&root->root_kobj, 0, sizeof(root->root_kobj));
  1061. root->defrag_trans_start = fs_info->generation;
  1062. init_completion(&root->kobj_unregister);
  1063. root->defrag_running = 0;
  1064. root->root_key.objectid = objectid;
  1065. root->anon_dev = 0;
  1066. spin_lock_init(&root->root_item_lock);
  1067. }
  1068. static int __must_check find_and_setup_root(struct btrfs_root *tree_root,
  1069. struct btrfs_fs_info *fs_info,
  1070. u64 objectid,
  1071. struct btrfs_root *root)
  1072. {
  1073. int ret;
  1074. u32 blocksize;
  1075. u64 generation;
  1076. __setup_root(tree_root->nodesize, tree_root->leafsize,
  1077. tree_root->sectorsize, tree_root->stripesize,
  1078. root, fs_info, objectid);
  1079. ret = btrfs_find_last_root(tree_root, objectid,
  1080. &root->root_item, &root->root_key);
  1081. if (ret > 0)
  1082. return -ENOENT;
  1083. else if (ret < 0)
  1084. return ret;
  1085. generation = btrfs_root_generation(&root->root_item);
  1086. blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
  1087. root->commit_root = NULL;
  1088. root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
  1089. blocksize, generation);
  1090. if (!root->node || !btrfs_buffer_uptodate(root->node, generation, 0)) {
  1091. free_extent_buffer(root->node);
  1092. root->node = NULL;
  1093. return -EIO;
  1094. }
  1095. root->commit_root = btrfs_root_node(root);
  1096. return 0;
  1097. }
  1098. static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
  1099. {
  1100. struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
  1101. if (root)
  1102. root->fs_info = fs_info;
  1103. return root;
  1104. }
  1105. struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
  1106. struct btrfs_fs_info *fs_info,
  1107. u64 objectid)
  1108. {
  1109. struct extent_buffer *leaf;
  1110. struct btrfs_root *tree_root = fs_info->tree_root;
  1111. struct btrfs_root *root;
  1112. struct btrfs_key key;
  1113. int ret = 0;
  1114. u64 bytenr;
  1115. root = btrfs_alloc_root(fs_info);
  1116. if (!root)
  1117. return ERR_PTR(-ENOMEM);
  1118. __setup_root(tree_root->nodesize, tree_root->leafsize,
  1119. tree_root->sectorsize, tree_root->stripesize,
  1120. root, fs_info, objectid);
  1121. root->root_key.objectid = objectid;
  1122. root->root_key.type = BTRFS_ROOT_ITEM_KEY;
  1123. root->root_key.offset = 0;
  1124. leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
  1125. 0, objectid, NULL, 0, 0, 0);
  1126. if (IS_ERR(leaf)) {
  1127. ret = PTR_ERR(leaf);
  1128. goto fail;
  1129. }
  1130. bytenr = leaf->start;
  1131. memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
  1132. btrfs_set_header_bytenr(leaf, leaf->start);
  1133. btrfs_set_header_generation(leaf, trans->transid);
  1134. btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
  1135. btrfs_set_header_owner(leaf, objectid);
  1136. root->node = leaf;
  1137. write_extent_buffer(leaf, fs_info->fsid,
  1138. (unsigned long)btrfs_header_fsid(leaf),
  1139. BTRFS_FSID_SIZE);
  1140. write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
  1141. (unsigned long)btrfs_header_chunk_tree_uuid(leaf),
  1142. BTRFS_UUID_SIZE);
  1143. btrfs_mark_buffer_dirty(leaf);
  1144. root->commit_root = btrfs_root_node(root);
  1145. root->track_dirty = 1;
  1146. root->root_item.flags = 0;
  1147. root->root_item.byte_limit = 0;
  1148. btrfs_set_root_bytenr(&root->root_item, leaf->start);
  1149. btrfs_set_root_generation(&root->root_item, trans->transid);
  1150. btrfs_set_root_level(&root->root_item, 0);
  1151. btrfs_set_root_refs(&root->root_item, 1);
  1152. btrfs_set_root_used(&root->root_item, leaf->len);
  1153. btrfs_set_root_last_snapshot(&root->root_item, 0);
  1154. btrfs_set_root_dirid(&root->root_item, 0);
  1155. root->root_item.drop_level = 0;
  1156. key.objectid = objectid;
  1157. key.type = BTRFS_ROOT_ITEM_KEY;
  1158. key.offset = 0;
  1159. ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
  1160. if (ret)
  1161. goto fail;
  1162. btrfs_tree_unlock(leaf);
  1163. fail:
  1164. if (ret)
  1165. return ERR_PTR(ret);
  1166. return root;
  1167. }
  1168. static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
  1169. struct btrfs_fs_info *fs_info)
  1170. {
  1171. struct btrfs_root *root;
  1172. struct btrfs_root *tree_root = fs_info->tree_root;
  1173. struct extent_buffer *leaf;
  1174. root = btrfs_alloc_root(fs_info);
  1175. if (!root)
  1176. return ERR_PTR(-ENOMEM);
  1177. __setup_root(tree_root->nodesize, tree_root->leafsize,
  1178. tree_root->sectorsize, tree_root->stripesize,
  1179. root, fs_info, BTRFS_TREE_LOG_OBJECTID);
  1180. root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
  1181. root->root_key.type = BTRFS_ROOT_ITEM_KEY;
  1182. root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
  1183. /*
  1184. * log trees do not get reference counted because they go away
  1185. * before a real commit is actually done. They do store pointers
  1186. * to file data extents, and those reference counts still get
  1187. * updated (along with back refs to the log tree).
  1188. */
  1189. root->ref_cows = 0;
  1190. leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
  1191. BTRFS_TREE_LOG_OBJECTID, NULL,
  1192. 0, 0, 0);
  1193. if (IS_ERR(leaf)) {
  1194. kfree(root);
  1195. return ERR_CAST(leaf);
  1196. }
  1197. memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
  1198. btrfs_set_header_bytenr(leaf, leaf->start);
  1199. btrfs_set_header_generation(leaf, trans->transid);
  1200. btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
  1201. btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
  1202. root->node = leaf;
  1203. write_extent_buffer(root->node, root->fs_info->fsid,
  1204. (unsigned long)btrfs_header_fsid(root->node),
  1205. BTRFS_FSID_SIZE);
  1206. btrfs_mark_buffer_dirty(root->node);
  1207. btrfs_tree_unlock(root->node);
  1208. return root;
  1209. }
  1210. int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
  1211. struct btrfs_fs_info *fs_info)
  1212. {
  1213. struct btrfs_root *log_root;
  1214. log_root = alloc_log_tree(trans, fs_info);
  1215. if (IS_ERR(log_root))
  1216. return PTR_ERR(log_root);
  1217. WARN_ON(fs_info->log_root_tree);
  1218. fs_info->log_root_tree = log_root;
  1219. return 0;
  1220. }
  1221. int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
  1222. struct btrfs_root *root)
  1223. {
  1224. struct btrfs_root *log_root;
  1225. struct btrfs_inode_item *inode_item;
  1226. log_root = alloc_log_tree(trans, root->fs_info);
  1227. if (IS_ERR(log_root))
  1228. return PTR_ERR(log_root);
  1229. log_root->last_trans = trans->transid;
  1230. log_root->root_key.offset = root->root_key.objectid;
  1231. inode_item = &log_root->root_item.inode;
  1232. inode_item->generation = cpu_to_le64(1);
  1233. inode_item->size = cpu_to_le64(3);
  1234. inode_item->nlink = cpu_to_le32(1);
  1235. inode_item->nbytes = cpu_to_le64(root->leafsize);
  1236. inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
  1237. btrfs_set_root_node(&log_root->root_item, log_root->node);
  1238. WARN_ON(root->log_root);
  1239. root->log_root = log_root;
  1240. root->log_transid = 0;
  1241. root->last_log_commit = 0;
  1242. return 0;
  1243. }
  1244. struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
  1245. struct btrfs_key *location)
  1246. {
  1247. struct btrfs_root *root;
  1248. struct btrfs_fs_info *fs_info = tree_root->fs_info;
  1249. struct btrfs_path *path;
  1250. struct extent_buffer *l;
  1251. u64 generation;
  1252. u32 blocksize;
  1253. int ret = 0;
  1254. int slot;
  1255. root = btrfs_alloc_root(fs_info);
  1256. if (!root)
  1257. return ERR_PTR(-ENOMEM);
  1258. if (location->offset == (u64)-1) {
  1259. ret = find_and_setup_root(tree_root, fs_info,
  1260. location->objectid, root);
  1261. if (ret) {
  1262. kfree(root);
  1263. return ERR_PTR(ret);
  1264. }
  1265. goto out;
  1266. }
  1267. __setup_root(tree_root->nodesize, tree_root->leafsize,
  1268. tree_root->sectorsize, tree_root->stripesize,
  1269. root, fs_info, location->objectid);
  1270. path = btrfs_alloc_path();
  1271. if (!path) {
  1272. kfree(root);
  1273. return ERR_PTR(-ENOMEM);
  1274. }
  1275. ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
  1276. if (ret == 0) {
  1277. l = path->nodes[0];
  1278. slot = path->slots[0];
  1279. btrfs_read_root_item(tree_root, l, slot, &root->root_item);
  1280. memcpy(&root->root_key, location, sizeof(*location));
  1281. }
  1282. btrfs_free_path(path);
  1283. if (ret) {
  1284. kfree(root);
  1285. if (ret > 0)
  1286. ret = -ENOENT;
  1287. return ERR_PTR(ret);
  1288. }
  1289. generation = btrfs_root_generation(&root->root_item);
  1290. blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
  1291. root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
  1292. blocksize, generation);
  1293. root->commit_root = btrfs_root_node(root);
  1294. BUG_ON(!root->node); /* -ENOMEM */
  1295. out:
  1296. if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
  1297. root->ref_cows = 1;
  1298. btrfs_check_and_init_root_item(&root->root_item);
  1299. }
  1300. return root;
  1301. }
  1302. struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
  1303. struct btrfs_key *location)
  1304. {
  1305. struct btrfs_root *root;
  1306. int ret;
  1307. if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
  1308. return fs_info->tree_root;
  1309. if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
  1310. return fs_info->extent_root;
  1311. if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
  1312. return fs_info->chunk_root;
  1313. if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
  1314. return fs_info->dev_root;
  1315. if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
  1316. return fs_info->csum_root;
  1317. if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
  1318. return fs_info->quota_root ? fs_info->quota_root :
  1319. ERR_PTR(-ENOENT);
  1320. again:
  1321. spin_lock(&fs_info->fs_roots_radix_lock);
  1322. root = radix_tree_lookup(&fs_info->fs_roots_radix,
  1323. (unsigned long)location->objectid);
  1324. spin_unlock(&fs_info->fs_roots_radix_lock);
  1325. if (root)
  1326. return root;
  1327. root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
  1328. if (IS_ERR(root))
  1329. return root;
  1330. root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
  1331. root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
  1332. GFP_NOFS);
  1333. if (!root->free_ino_pinned || !root->free_ino_ctl) {
  1334. ret = -ENOMEM;
  1335. goto fail;
  1336. }
  1337. btrfs_init_free_ino_ctl(root);
  1338. mutex_init(&root->fs_commit_mutex);
  1339. spin_lock_init(&root->cache_lock);
  1340. init_waitqueue_head(&root->cache_wait);
  1341. ret = get_anon_bdev(&root->anon_dev);
  1342. if (ret)
  1343. goto fail;
  1344. if (btrfs_root_refs(&root->root_item) == 0) {
  1345. ret = -ENOENT;
  1346. goto fail;
  1347. }
  1348. ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
  1349. if (ret < 0)
  1350. goto fail;
  1351. if (ret == 0)
  1352. root->orphan_item_inserted = 1;
  1353. ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
  1354. if (ret)
  1355. goto fail;
  1356. spin_lock(&fs_info->fs_roots_radix_lock);
  1357. ret = radix_tree_insert(&fs_info->fs_roots_radix,
  1358. (unsigned long)root->root_key.objectid,
  1359. root);
  1360. if (ret == 0)
  1361. root->in_radix = 1;
  1362. spin_unlock(&fs_info->fs_roots_radix_lock);
  1363. radix_tree_preload_end();
  1364. if (ret) {
  1365. if (ret == -EEXIST) {
  1366. free_fs_root(root);
  1367. goto again;
  1368. }
  1369. goto fail;
  1370. }
  1371. ret = btrfs_find_dead_roots(fs_info->tree_root,
  1372. root->root_key.objectid);
  1373. WARN_ON(ret);
  1374. return root;
  1375. fail:
  1376. free_fs_root(root);
  1377. return ERR_PTR(ret);
  1378. }
  1379. static int btrfs_congested_fn(void *congested_data, int bdi_bits)
  1380. {
  1381. struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
  1382. int ret = 0;
  1383. struct btrfs_device *device;
  1384. struct backing_dev_info *bdi;
  1385. rcu_read_lock();
  1386. list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
  1387. if (!device->bdev)
  1388. continue;
  1389. bdi = blk_get_backing_dev_info(device->bdev);
  1390. if (bdi && bdi_congested(bdi, bdi_bits)) {
  1391. ret = 1;
  1392. break;
  1393. }
  1394. }
  1395. rcu_read_unlock();
  1396. return ret;
  1397. }
  1398. /*
  1399. * If this fails, caller must call bdi_destroy() to get rid of the
  1400. * bdi again.
  1401. */
  1402. static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
  1403. {
  1404. int err;
  1405. bdi->capabilities = BDI_CAP_MAP_COPY;
  1406. err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
  1407. if (err)
  1408. return err;
  1409. bdi->ra_pages = default_backing_dev_info.ra_pages;
  1410. bdi->congested_fn = btrfs_congested_fn;
  1411. bdi->congested_data = info;
  1412. return 0;
  1413. }
  1414. /*
  1415. * called by the kthread helper functions to finally call the bio end_io
  1416. * functions. This is where read checksum verification actually happens
  1417. */
  1418. static void end_workqueue_fn(struct btrfs_work *work)
  1419. {
  1420. struct bio *bio;
  1421. struct end_io_wq *end_io_wq;
  1422. struct btrfs_fs_info *fs_info;
  1423. int error;
  1424. end_io_wq = container_of(work, struct end_io_wq, work);
  1425. bio = end_io_wq->bio;
  1426. fs_info = end_io_wq->info;
  1427. error = end_io_wq->error;
  1428. bio->bi_private = end_io_wq->private;
  1429. bio->bi_end_io = end_io_wq->end_io;
  1430. kfree(end_io_wq);
  1431. bio_endio(bio, error);
  1432. }
  1433. static int cleaner_kthread(void *arg)
  1434. {
  1435. struct btrfs_root *root = arg;
  1436. do {
  1437. if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
  1438. mutex_trylock(&root->fs_info->cleaner_mutex)) {
  1439. btrfs_run_delayed_iputs(root);
  1440. btrfs_clean_old_snapshots(root);
  1441. mutex_unlock(&root->fs_info->cleaner_mutex);
  1442. btrfs_run_defrag_inodes(root->fs_info);
  1443. }
  1444. if (!try_to_freeze()) {
  1445. set_current_state(TASK_INTERRUPTIBLE);
  1446. if (!kthread_should_stop())
  1447. schedule();
  1448. __set_current_state(TASK_RUNNING);
  1449. }
  1450. } while (!kthread_should_stop());
  1451. return 0;
  1452. }
  1453. static int transaction_kthread(void *arg)
  1454. {
  1455. struct btrfs_root *root = arg;
  1456. struct btrfs_trans_handle *trans;
  1457. struct btrfs_transaction *cur;
  1458. u64 transid;
  1459. unsigned long now;
  1460. unsigned long delay;
  1461. bool cannot_commit;
  1462. do {
  1463. cannot_commit = false;
  1464. delay = HZ * 30;
  1465. mutex_lock(&root->fs_info->transaction_kthread_mutex);
  1466. spin_lock(&root->fs_info->trans_lock);
  1467. cur = root->fs_info->running_transaction;
  1468. if (!cur) {
  1469. spin_unlock(&root->fs_info->trans_lock);
  1470. goto sleep;
  1471. }
  1472. now = get_seconds();
  1473. if (!cur->blocked &&
  1474. (now < cur->start_time || now - cur->start_time < 30)) {
  1475. spin_unlock(&root->fs_info->trans_lock);
  1476. delay = HZ * 5;
  1477. goto sleep;
  1478. }
  1479. transid = cur->transid;
  1480. spin_unlock(&root->fs_info->trans_lock);
  1481. /* If the file system is aborted, this will always fail. */
  1482. trans = btrfs_attach_transaction(root);
  1483. if (IS_ERR(trans)) {
  1484. if (PTR_ERR(trans) != -ENOENT)
  1485. cannot_commit = true;
  1486. goto sleep;
  1487. }
  1488. if (transid == trans->transid) {
  1489. btrfs_commit_transaction(trans, root);
  1490. } else {
  1491. btrfs_end_transaction(trans, root);
  1492. }
  1493. sleep:
  1494. wake_up_process(root->fs_info->cleaner_kthread);
  1495. mutex_unlock(&root->fs_info->transaction_kthread_mutex);
  1496. if (!try_to_freeze()) {
  1497. set_current_state(TASK_INTERRUPTIBLE);
  1498. if (!kthread_should_stop() &&
  1499. (!btrfs_transaction_blocked(root->fs_info) ||
  1500. cannot_commit))
  1501. schedule_timeout(delay);
  1502. __set_current_state(TASK_RUNNING);
  1503. }
  1504. } while (!kthread_should_stop());
  1505. return 0;
  1506. }
  1507. /*
  1508. * this will find the highest generation in the array of
  1509. * root backups. The index of the highest array is returned,
  1510. * or -1 if we can't find anything.
  1511. *
  1512. * We check to make sure the array is valid by comparing the
  1513. * generation of the latest root in the array with the generation
  1514. * in the super block. If they don't match we pitch it.
  1515. */
  1516. static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
  1517. {
  1518. u64 cur;
  1519. int newest_index = -1;
  1520. struct btrfs_root_backup *root_backup;
  1521. int i;
  1522. for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
  1523. root_backup = info->super_copy->super_roots + i;
  1524. cur = btrfs_backup_tree_root_gen(root_backup);
  1525. if (cur == newest_gen)
  1526. newest_index = i;
  1527. }
  1528. /* check to see if we actually wrapped around */
  1529. if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
  1530. root_backup = info->super_copy->super_roots;
  1531. cur = btrfs_backup_tree_root_gen(root_backup);
  1532. if (cur == newest_gen)
  1533. newest_index = 0;
  1534. }
  1535. return newest_index;
  1536. }
  1537. /*
  1538. * find the oldest backup so we know where to store new entries
  1539. * in the backup array. This will set the backup_root_index
  1540. * field in the fs_info struct
  1541. */
  1542. static void find_oldest_super_backup(struct btrfs_fs_info *info,
  1543. u64 newest_gen)
  1544. {
  1545. int newest_index = -1;
  1546. newest_index = find_newest_super_backup(info, newest_gen);
  1547. /* if there was garbage in there, just move along */
  1548. if (newest_index == -1) {
  1549. info->backup_root_index = 0;
  1550. } else {
  1551. info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
  1552. }
  1553. }
  1554. /*
  1555. * copy all the root pointers into the super backup array.
  1556. * this will bump the backup pointer by one when it is
  1557. * done
  1558. */
  1559. static void backup_super_roots(struct btrfs_fs_info *info)
  1560. {
  1561. int next_backup;
  1562. struct btrfs_root_backup *root_backup;
  1563. int last_backup;
  1564. next_backup = info->backup_root_index;
  1565. last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
  1566. BTRFS_NUM_BACKUP_ROOTS;
  1567. /*
  1568. * just overwrite the last backup if we're at the same generation
  1569. * this happens only at umount
  1570. */
  1571. root_backup = info->super_for_commit->super_roots + last_backup;
  1572. if (btrfs_backup_tree_root_gen(root_backup) ==
  1573. btrfs_header_generation(info->tree_root->node))
  1574. next_backup = last_backup;
  1575. root_backup = info->super_for_commit->super_roots + next_backup;
  1576. /*
  1577. * make sure all of our padding and empty slots get zero filled
  1578. * regardless of which ones we use today
  1579. */
  1580. memset(root_backup, 0, sizeof(*root_backup));
  1581. info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
  1582. btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
  1583. btrfs_set_backup_tree_root_gen(root_backup,
  1584. btrfs_header_generation(info->tree_root->node));
  1585. btrfs_set_backup_tree_root_level(root_backup,
  1586. btrfs_header_level(info->tree_root->node));
  1587. btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
  1588. btrfs_set_backup_chunk_root_gen(root_backup,
  1589. btrfs_header_generation(info->chunk_root->node));
  1590. btrfs_set_backup_chunk_root_level(root_backup,
  1591. btrfs_header_level(info->chunk_root->node));
  1592. btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
  1593. btrfs_set_backup_extent_root_gen(root_backup,
  1594. btrfs_header_generation(info->extent_root->node));
  1595. btrfs_set_backup_extent_root_level(root_backup,
  1596. btrfs_header_level(info->extent_root->node));
  1597. /*
  1598. * we might commit during log recovery, which happens before we set
  1599. * the fs_root. Make sure it is valid before we fill it in.
  1600. */
  1601. if (info->fs_root && info->fs_root->node) {
  1602. btrfs_set_backup_fs_root(root_backup,
  1603. info->fs_root->node->start);
  1604. btrfs_set_backup_fs_root_gen(root_backup,
  1605. btrfs_header_generation(info->fs_root->node));
  1606. btrfs_set_backup_fs_root_level(root_backup,
  1607. btrfs_header_level(info->fs_root->node));
  1608. }
  1609. btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
  1610. btrfs_set_backup_dev_root_gen(root_backup,
  1611. btrfs_header_generation(info->dev_root->node));
  1612. btrfs_set_backup_dev_root_level(root_backup,
  1613. btrfs_header_level(info->dev_root->node));
  1614. btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
  1615. btrfs_set_backup_csum_root_gen(root_backup,
  1616. btrfs_header_generation(info->csum_root->node));
  1617. btrfs_set_backup_csum_root_level(root_backup,
  1618. btrfs_header_level(info->csum_root->node));
  1619. btrfs_set_backup_total_bytes(root_backup,
  1620. btrfs_super_total_bytes(info->super_copy));
  1621. btrfs_set_backup_bytes_used(root_backup,
  1622. btrfs_super_bytes_used(info->super_copy));
  1623. btrfs_set_backup_num_devices(root_backup,
  1624. btrfs_super_num_devices(info->super_copy));
  1625. /*
  1626. * if we don't copy this out to the super_copy, it won't get remembered
  1627. * for the next commit
  1628. */
  1629. memcpy(&info->super_copy->super_roots,
  1630. &info->super_for_commit->super_roots,
  1631. sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
  1632. }
  1633. /*
  1634. * this copies info out of the root backup array and back into
  1635. * the in-memory super block. It is meant to help iterate through
  1636. * the array, so you send it the number of backups you've already
  1637. * tried and the last backup index you used.
  1638. *
  1639. * this returns -1 when it has tried all the backups
  1640. */
  1641. static noinline int next_root_backup(struct btrfs_fs_info *info,
  1642. struct btrfs_super_block *super,
  1643. int *num_backups_tried, int *backup_index)
  1644. {
  1645. struct btrfs_root_backup *root_backup;
  1646. int newest = *backup_index;
  1647. if (*num_backups_tried == 0) {
  1648. u64 gen = btrfs_super_generation(super);
  1649. newest = find_newest_super_backup(info, gen);
  1650. if (newest == -1)
  1651. return -1;
  1652. *backup_index = newest;
  1653. *num_backups_tried = 1;
  1654. } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
  1655. /* we've tried all the backups, all done */
  1656. return -1;
  1657. } else {
  1658. /* jump to the next oldest backup */
  1659. newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
  1660. BTRFS_NUM_BACKUP_ROOTS;
  1661. *backup_index = newest;
  1662. *num_backups_tried += 1;
  1663. }
  1664. root_backup = super->super_roots + newest;
  1665. btrfs_set_super_generation(super,
  1666. btrfs_backup_tree_root_gen(root_backup));
  1667. btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
  1668. btrfs_set_super_root_level(super,
  1669. btrfs_backup_tree_root_level(root_backup));
  1670. btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
  1671. /*
  1672. * fixme: the total bytes and num_devices need to match or we should
  1673. * need a fsck
  1674. */
  1675. btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
  1676. btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
  1677. return 0;
  1678. }
  1679. /* helper to cleanup tree roots */
  1680. static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
  1681. {
  1682. free_extent_buffer(info->tree_root->node);
  1683. free_extent_buffer(info->tree_root->commit_root);
  1684. free_extent_buffer(info->dev_root->node);
  1685. free_extent_buffer(info->dev_root->commit_root);
  1686. free_extent_buffer(info->extent_root->node);
  1687. free_extent_buffer(info->extent_root->commit_root);
  1688. free_extent_buffer(info->csum_root->node);
  1689. free_extent_buffer(info->csum_root->commit_root);
  1690. if (info->quota_root) {
  1691. free_extent_buffer(info->quota_root->node);
  1692. free_extent_buffer(info->quota_root->commit_root);
  1693. }
  1694. info->tree_root->node = NULL;
  1695. info->tree_root->commit_root = NULL;
  1696. info->dev_root->node = NULL;
  1697. info->dev_root->commit_root = NULL;
  1698. info->extent_root->node = NULL;
  1699. info->extent_root->commit_root = NULL;
  1700. info->csum_root->node = NULL;
  1701. info->csum_root->commit_root = NULL;
  1702. if (info->quota_root) {
  1703. info->quota_root->node = NULL;
  1704. info->quota_root->commit_root = NULL;
  1705. }
  1706. if (chunk_root) {
  1707. free_extent_buffer(info->chunk_root->node);
  1708. free_extent_buffer(info->chunk_root->commit_root);
  1709. info->chunk_root->node = NULL;
  1710. info->chunk_root->commit_root = NULL;
  1711. }
  1712. }
  1713. int open_ctree(struct super_block *sb,
  1714. struct btrfs_fs_devices *fs_devices,
  1715. char *options)
  1716. {
  1717. u32 sectorsize;
  1718. u32 nodesize;
  1719. u32 leafsize;
  1720. u32 blocksize;
  1721. u32 stripesize;
  1722. u64 generation;
  1723. u64 features;
  1724. struct btrfs_key location;
  1725. struct buffer_head *bh;
  1726. struct btrfs_super_block *disk_super;
  1727. struct btrfs_fs_info *fs_info = btrfs_sb(sb);
  1728. struct btrfs_root *tree_root;
  1729. struct btrfs_root *extent_root;
  1730. struct btrfs_root *csum_root;
  1731. struct btrfs_root *chunk_root;
  1732. struct btrfs_root *dev_root;
  1733. struct btrfs_root *quota_root;
  1734. struct btrfs_root *log_tree_root;
  1735. int ret;
  1736. int err = -EINVAL;
  1737. int num_backups_tried = 0;
  1738. int backup_index = 0;
  1739. tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
  1740. extent_root = fs_info->extent_root = btrfs_alloc_root(fs_info);
  1741. csum_root = fs_info->csum_root = btrfs_alloc_root(fs_info);
  1742. chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
  1743. dev_root = fs_info->dev_root = btrfs_alloc_root(fs_info);
  1744. quota_root = fs_info->quota_root = btrfs_alloc_root(fs_info);
  1745. if (!tree_root || !extent_root || !csum_root ||
  1746. !chunk_root || !dev_root || !quota_root) {
  1747. err = -ENOMEM;
  1748. goto fail;
  1749. }
  1750. ret = init_srcu_struct(&fs_info->subvol_srcu);
  1751. if (ret) {
  1752. err = ret;
  1753. goto fail;
  1754. }
  1755. ret = setup_bdi(fs_info, &fs_info->bdi);
  1756. if (ret) {
  1757. err = ret;
  1758. goto fail_srcu;
  1759. }
  1760. fs_info->btree_inode = new_inode(sb);
  1761. if (!fs_info->btree_inode) {
  1762. err = -ENOMEM;
  1763. goto fail_bdi;
  1764. }
  1765. mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
  1766. INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
  1767. INIT_LIST_HEAD(&fs_info->trans_list);
  1768. INIT_LIST_HEAD(&fs_info->dead_roots);
  1769. INIT_LIST_HEAD(&fs_info->delayed_iputs);
  1770. INIT_LIST_HEAD(&fs_info->delalloc_inodes);
  1771. INIT_LIST_HEAD(&fs_info->ordered_operations);
  1772. INIT_LIST_HEAD(&fs_info->caching_block_groups);
  1773. spin_lock_init(&fs_info->delalloc_lock);
  1774. spin_lock_init(&fs_info->trans_lock);
  1775. spin_lock_init(&fs_info->fs_roots_radix_lock);
  1776. spin_lock_init(&fs_info->delayed_iput_lock);
  1777. spin_lock_init(&fs_info->defrag_inodes_lock);
  1778. spin_lock_init(&fs_info->free_chunk_lock);
  1779. spin_lock_init(&fs_info->tree_mod_seq_lock);
  1780. rwlock_init(&fs_info->tree_mod_log_lock);
  1781. mutex_init(&fs_info->reloc_mutex);
  1782. init_completion(&fs_info->kobj_unregister);
  1783. INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
  1784. INIT_LIST_HEAD(&fs_info->space_info);
  1785. INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
  1786. btrfs_mapping_init(&fs_info->mapping_tree);
  1787. btrfs_init_block_rsv(&fs_info->global_block_rsv,
  1788. BTRFS_BLOCK_RSV_GLOBAL);
  1789. btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
  1790. BTRFS_BLOCK_RSV_DELALLOC);
  1791. btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
  1792. btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
  1793. btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
  1794. btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
  1795. BTRFS_BLOCK_RSV_DELOPS);
  1796. atomic_set(&fs_info->nr_async_submits, 0);
  1797. atomic_set(&fs_info->async_delalloc_pages, 0);
  1798. atomic_set(&fs_info->async_submit_draining, 0);
  1799. atomic_set(&fs_info->nr_async_bios, 0);
  1800. atomic_set(&fs_info->defrag_running, 0);
  1801. atomic_set(&fs_info->tree_mod_seq, 0);
  1802. fs_info->sb = sb;
  1803. fs_info->max_inline = 8192 * 1024;
  1804. fs_info->metadata_ratio = 0;
  1805. fs_info->defrag_inodes = RB_ROOT;
  1806. fs_info->trans_no_join = 0;
  1807. fs_info->free_chunk_space = 0;
  1808. fs_info->tree_mod_log = RB_ROOT;
  1809. /* readahead state */
  1810. INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
  1811. spin_lock_init(&fs_info->reada_lock);
  1812. fs_info->thread_pool_size = min_t(unsigned long,
  1813. num_online_cpus() + 2, 8);
  1814. INIT_LIST_HEAD(&fs_info->ordered_extents);
  1815. spin_lock_init(&fs_info->ordered_extent_lock);
  1816. fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
  1817. GFP_NOFS);
  1818. if (!fs_info->delayed_root) {
  1819. err = -ENOMEM;
  1820. goto fail_iput;
  1821. }
  1822. btrfs_init_delayed_root(fs_info->delayed_root);
  1823. mutex_init(&fs_info->scrub_lock);
  1824. atomic_set(&fs_info->scrubs_running, 0);
  1825. atomic_set(&fs_info->scrub_pause_req, 0);
  1826. atomic_set(&fs_info->scrubs_paused, 0);
  1827. atomic_set(&fs_info->scrub_cancel_req, 0);
  1828. init_waitqueue_head(&fs_info->scrub_pause_wait);
  1829. init_rwsem(&fs_info->scrub_super_lock);
  1830. fs_info->scrub_workers_refcnt = 0;
  1831. #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
  1832. fs_info->check_integrity_print_mask = 0;
  1833. #endif
  1834. spin_lock_init(&fs_info->balance_lock);
  1835. mutex_init(&fs_info->balance_mutex);
  1836. atomic_set(&fs_info->balance_running, 0);
  1837. atomic_set(&fs_info->balance_pause_req, 0);
  1838. atomic_set(&fs_info->balance_cancel_req, 0);
  1839. fs_info->balance_ctl = NULL;
  1840. init_waitqueue_head(&fs_info->balance_wait_q);
  1841. sb->s_blocksize = 4096;
  1842. sb->s_blocksize_bits = blksize_bits(4096);
  1843. sb->s_bdi = &fs_info->bdi;
  1844. fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
  1845. set_nlink(fs_info->btree_inode, 1);
  1846. /*
  1847. * we set the i_size on the btree inode to the max possible int.
  1848. * the real end of the address space is determined by all of
  1849. * the devices in the system
  1850. */
  1851. fs_info->btree_inode->i_size = OFFSET_MAX;
  1852. fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
  1853. fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
  1854. RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
  1855. extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
  1856. fs_info->btree_inode->i_mapping);
  1857. BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
  1858. extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
  1859. BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
  1860. BTRFS_I(fs_info->btree_inode)->root = tree_root;
  1861. memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
  1862. sizeof(struct btrfs_key));
  1863. set_bit(BTRFS_INODE_DUMMY,
  1864. &BTRFS_I(fs_info->btree_inode)->runtime_flags);
  1865. insert_inode_hash(fs_info->btree_inode);
  1866. spin_lock_init(&fs_info->block_group_cache_lock);
  1867. fs_info->block_group_cache_tree = RB_ROOT;
  1868. extent_io_tree_init(&fs_info->freed_extents[0],
  1869. fs_info->btree_inode->i_mapping);
  1870. extent_io_tree_init(&fs_info->freed_extents[1],
  1871. fs_info->btree_inode->i_mapping);
  1872. fs_info->pinned_extents = &fs_info->freed_extents[0];
  1873. fs_info->do_barriers = 1;
  1874. mutex_init(&fs_info->ordered_operations_mutex);
  1875. mutex_init(&fs_info->tree_log_mutex);
  1876. mutex_init(&fs_info->chunk_mutex);
  1877. mutex_init(&fs_info->transaction_kthread_mutex);
  1878. mutex_init(&fs_info->cleaner_mutex);
  1879. mutex_init(&fs_info->volume_mutex);
  1880. init_rwsem(&fs_info->extent_commit_sem);
  1881. init_rwsem(&fs_info->cleanup_work_sem);
  1882. init_rwsem(&fs_info->subvol_sem);
  1883. fs_info->dev_replace.lock_owner = 0;
  1884. atomic_set(&fs_info->dev_replace.nesting_level, 0);
  1885. mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
  1886. mutex_init(&fs_info->dev_replace.lock_management_lock);
  1887. mutex_init(&fs_info->dev_replace.lock);
  1888. spin_lock_init(&fs_info->qgroup_lock);
  1889. fs_info->qgroup_tree = RB_ROOT;
  1890. INIT_LIST_HEAD(&fs_info->dirty_qgroups);
  1891. fs_info->qgroup_seq = 1;
  1892. fs_info->quota_enabled = 0;
  1893. fs_info->pending_quota_state = 0;
  1894. btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
  1895. btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
  1896. init_waitqueue_head(&fs_info->transaction_throttle);
  1897. init_waitqueue_head(&fs_info->transaction_wait);
  1898. init_waitqueue_head(&fs_info->transaction_blocked_wait);
  1899. init_waitqueue_head(&fs_info->async_submit_wait);
  1900. __setup_root(4096, 4096, 4096, 4096, tree_root,
  1901. fs_info, BTRFS_ROOT_TREE_OBJECTID);
  1902. invalidate_bdev(fs_devices->latest_bdev);
  1903. bh = btrfs_read_dev_super(fs_devices->latest_bdev);
  1904. if (!bh) {
  1905. err = -EINVAL;
  1906. goto fail_alloc;
  1907. }
  1908. memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
  1909. memcpy(fs_info->super_for_commit, fs_info->super_copy,
  1910. sizeof(*fs_info->super_for_commit));
  1911. brelse(bh);
  1912. memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
  1913. disk_super = fs_info->super_copy;
  1914. if (!btrfs_super_root(disk_super))
  1915. goto fail_alloc;
  1916. /* check FS state, whether FS is broken. */
  1917. fs_info->fs_state |= btrfs_super_flags(disk_super);
  1918. ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
  1919. if (ret) {
  1920. printk(KERN_ERR "btrfs: superblock contains fatal errors\n");
  1921. err = ret;
  1922. goto fail_alloc;
  1923. }
  1924. /*
  1925. * run through our array of backup supers and setup
  1926. * our ring pointer to the oldest one
  1927. */
  1928. generation = btrfs_super_generation(disk_super);
  1929. find_oldest_super_backup(fs_info, generation);
  1930. /*
  1931. * In the long term, we'll store the compression type in the super
  1932. * block, and it'll be used for per file compression control.
  1933. */
  1934. fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
  1935. ret = btrfs_parse_options(tree_root, options);
  1936. if (ret) {
  1937. err = ret;
  1938. goto fail_alloc;
  1939. }
  1940. features = btrfs_super_incompat_flags(disk_super) &
  1941. ~BTRFS_FEATURE_INCOMPAT_SUPP;
  1942. if (features) {
  1943. printk(KERN_ERR "BTRFS: couldn't mount because of "
  1944. "unsupported optional features (%Lx).\n",
  1945. (unsigned long long)features);
  1946. err = -EINVAL;
  1947. goto fail_alloc;
  1948. }
  1949. if (btrfs_super_leafsize(disk_super) !=
  1950. btrfs_super_nodesize(disk_super)) {
  1951. printk(KERN_ERR "BTRFS: couldn't mount because metadata "
  1952. "blocksizes don't match. node %d leaf %d\n",
  1953. btrfs_super_nodesize(disk_super),
  1954. btrfs_super_leafsize(disk_super));
  1955. err = -EINVAL;
  1956. goto fail_alloc;
  1957. }
  1958. if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
  1959. printk(KERN_ERR "BTRFS: couldn't mount because metadata "
  1960. "blocksize (%d) was too large\n",
  1961. btrfs_super_leafsize(disk_super));
  1962. err = -EINVAL;
  1963. goto fail_alloc;
  1964. }
  1965. features = btrfs_super_incompat_flags(disk_super);
  1966. features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
  1967. if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
  1968. features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
  1969. /*
  1970. * flag our filesystem as having big metadata blocks if
  1971. * they are bigger than the page size
  1972. */
  1973. if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) {
  1974. if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
  1975. printk(KERN_INFO "btrfs flagging fs with big metadata feature\n");
  1976. features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
  1977. }
  1978. nodesize = btrfs_super_nodesize(disk_super);
  1979. leafsize = btrfs_super_leafsize(disk_super);
  1980. sectorsize = btrfs_super_sectorsize(disk_super);
  1981. stripesize = btrfs_super_stripesize(disk_super);
  1982. /*
  1983. * mixed block groups end up with duplicate but slightly offset
  1984. * extent buffers for the same range. It leads to corruptions
  1985. */
  1986. if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
  1987. (sectorsize != leafsize)) {
  1988. printk(KERN_WARNING "btrfs: unequal leaf/node/sector sizes "
  1989. "are not allowed for mixed block groups on %s\n",
  1990. sb->s_id);
  1991. goto fail_alloc;
  1992. }
  1993. btrfs_set_super_incompat_flags(disk_super, features);
  1994. features = btrfs_super_compat_ro_flags(disk_super) &
  1995. ~BTRFS_FEATURE_COMPAT_RO_SUPP;
  1996. if (!(sb->s_flags & MS_RDONLY) && features) {
  1997. printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
  1998. "unsupported option features (%Lx).\n",
  1999. (unsigned long long)features);
  2000. err = -EINVAL;
  2001. goto fail_alloc;
  2002. }
  2003. btrfs_init_workers(&fs_info->generic_worker,
  2004. "genwork", 1, NULL);
  2005. btrfs_init_workers(&fs_info->workers, "worker",
  2006. fs_info->thread_pool_size,
  2007. &fs_info->generic_worker);
  2008. btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
  2009. fs_info->thread_pool_size,
  2010. &fs_info->generic_worker);
  2011. btrfs_init_workers(&fs_info->flush_workers, "flush_delalloc",
  2012. fs_info->thread_pool_size,
  2013. &fs_info->generic_worker);
  2014. btrfs_init_workers(&fs_info->submit_workers, "submit",
  2015. min_t(u64, fs_devices->num_devices,
  2016. fs_info->thread_pool_size),
  2017. &fs_info->generic_worker);
  2018. btrfs_init_workers(&fs_info->caching_workers, "cache",
  2019. 2, &fs_info->generic_worker);
  2020. /* a higher idle thresh on the submit workers makes it much more
  2021. * likely that bios will be send down in a sane order to the
  2022. * devices
  2023. */
  2024. fs_info->submit_workers.idle_thresh = 64;
  2025. fs_info->workers.idle_thresh = 16;
  2026. fs_info->workers.ordered = 1;
  2027. fs_info->delalloc_workers.idle_thresh = 2;
  2028. fs_info->delalloc_workers.ordered = 1;
  2029. btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
  2030. &fs_info->generic_worker);
  2031. btrfs_init_workers(&fs_info->endio_workers, "endio",
  2032. fs_info->thread_pool_size,
  2033. &fs_info->generic_worker);
  2034. btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
  2035. fs_info->thread_pool_size,
  2036. &fs_info->generic_worker);
  2037. btrfs_init_workers(&fs_info->endio_meta_write_workers,
  2038. "endio-meta-write", fs_info->thread_pool_size,
  2039. &fs_info->generic_worker);
  2040. btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
  2041. fs_info->thread_pool_size,
  2042. &fs_info->generic_worker);
  2043. btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
  2044. 1, &fs_info->generic_worker);
  2045. btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
  2046. fs_info->thread_pool_size,
  2047. &fs_info->generic_worker);
  2048. btrfs_init_workers(&fs_info->readahead_workers, "readahead",
  2049. fs_info->thread_pool_size,
  2050. &fs_info->generic_worker);
  2051. /*
  2052. * endios are largely parallel and should have a very
  2053. * low idle thresh
  2054. */
  2055. fs_info->endio_workers.idle_thresh = 4;
  2056. fs_info->endio_meta_workers.idle_thresh = 4;
  2057. fs_info->endio_write_workers.idle_thresh = 2;
  2058. fs_info->endio_meta_write_workers.idle_thresh = 2;
  2059. fs_info->readahead_workers.idle_thresh = 2;
  2060. /*
  2061. * btrfs_start_workers can really only fail because of ENOMEM so just
  2062. * return -ENOMEM if any of these fail.
  2063. */
  2064. ret = btrfs_start_workers(&fs_info->workers);
  2065. ret |= btrfs_start_workers(&fs_info->generic_worker);
  2066. ret |= btrfs_start_workers(&fs_info->submit_workers);
  2067. ret |= btrfs_start_workers(&fs_info->delalloc_workers);
  2068. ret |= btrfs_start_workers(&fs_info->fixup_workers);
  2069. ret |= btrfs_start_workers(&fs_info->endio_workers);
  2070. ret |= btrfs_start_workers(&fs_info->endio_meta_workers);
  2071. ret |= btrfs_start_workers(&fs_info->endio_meta_write_workers);
  2072. ret |= btrfs_start_workers(&fs_info->endio_write_workers);
  2073. ret |= btrfs_start_workers(&fs_info->endio_freespace_worker);
  2074. ret |= btrfs_start_workers(&fs_info->delayed_workers);
  2075. ret |= btrfs_start_workers(&fs_info->caching_workers);
  2076. ret |= btrfs_start_workers(&fs_info->readahead_workers);
  2077. ret |= btrfs_start_workers(&fs_info->flush_workers);
  2078. if (ret) {
  2079. err = -ENOMEM;
  2080. goto fail_sb_buffer;
  2081. }
  2082. fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
  2083. fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
  2084. 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
  2085. tree_root->nodesize = nodesize;
  2086. tree_root->leafsize = leafsize;
  2087. tree_root->sectorsize = sectorsize;
  2088. tree_root->stripesize = stripesize;
  2089. sb->s_blocksize = sectorsize;
  2090. sb->s_blocksize_bits = blksize_bits(sectorsize);
  2091. if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
  2092. sizeof(disk_super->magic))) {
  2093. printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
  2094. goto fail_sb_buffer;
  2095. }
  2096. if (sectorsize != PAGE_SIZE) {
  2097. printk(KERN_WARNING "btrfs: Incompatible sector size(%lu) "
  2098. "found on %s\n", (unsigned long)sectorsize, sb->s_id);
  2099. goto fail_sb_buffer;
  2100. }
  2101. mutex_lock(&fs_info->chunk_mutex);
  2102. ret = btrfs_read_sys_array(tree_root);
  2103. mutex_unlock(&fs_info->chunk_mutex);
  2104. if (ret) {
  2105. printk(KERN_WARNING "btrfs: failed to read the system "
  2106. "array on %s\n", sb->s_id);
  2107. goto fail_sb_buffer;
  2108. }
  2109. blocksize = btrfs_level_size(tree_root,
  2110. btrfs_super_chunk_root_level(disk_super));
  2111. generation = btrfs_super_chunk_root_generation(disk_super);
  2112. __setup_root(nodesize, leafsize, sectorsize, stripesize,
  2113. chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
  2114. chunk_root->node = read_tree_block(chunk_root,
  2115. btrfs_super_chunk_root(disk_super),
  2116. blocksize, generation);
  2117. BUG_ON(!chunk_root->node); /* -ENOMEM */
  2118. if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
  2119. printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
  2120. sb->s_id);
  2121. goto fail_tree_roots;
  2122. }
  2123. btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
  2124. chunk_root->commit_root = btrfs_root_node(chunk_root);
  2125. read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
  2126. (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
  2127. BTRFS_UUID_SIZE);
  2128. ret = btrfs_read_chunk_tree(chunk_root);
  2129. if (ret) {
  2130. printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
  2131. sb->s_id);
  2132. goto fail_tree_roots;
  2133. }
  2134. /*
  2135. * keep the device that is marked to be the target device for the
  2136. * dev_replace procedure
  2137. */
  2138. btrfs_close_extra_devices(fs_info, fs_devices, 0);
  2139. if (!fs_devices->latest_bdev) {
  2140. printk(KERN_CRIT "btrfs: failed to read devices on %s\n",
  2141. sb->s_id);
  2142. goto fail_tree_roots;
  2143. }
  2144. retry_root_backup:
  2145. blocksize = btrfs_level_size(tree_root,
  2146. btrfs_super_root_level(disk_super));
  2147. generation = btrfs_super_generation(disk_super);
  2148. tree_root->node = read_tree_block(tree_root,
  2149. btrfs_super_root(disk_super),
  2150. blocksize, generation);
  2151. if (!tree_root->node ||
  2152. !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
  2153. printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
  2154. sb->s_id);
  2155. goto recovery_tree_root;
  2156. }
  2157. btrfs_set_root_node(&tree_root->root_item, tree_root->node);
  2158. tree_root->commit_root = btrfs_root_node(tree_root);
  2159. ret = find_and_setup_root(tree_root, fs_info,
  2160. BTRFS_EXTENT_TREE_OBJECTID, extent_root);
  2161. if (ret)
  2162. goto recovery_tree_root;
  2163. extent_root->track_dirty = 1;
  2164. ret = find_and_setup_root(tree_root, fs_info,
  2165. BTRFS_DEV_TREE_OBJECTID, dev_root);
  2166. if (ret)
  2167. goto recovery_tree_root;
  2168. dev_root->track_dirty = 1;
  2169. ret = find_and_setup_root(tree_root, fs_info,
  2170. BTRFS_CSUM_TREE_OBJECTID, csum_root);
  2171. if (ret)
  2172. goto recovery_tree_root;
  2173. csum_root->track_dirty = 1;
  2174. ret = find_and_setup_root(tree_root, fs_info,
  2175. BTRFS_QUOTA_TREE_OBJECTID, quota_root);
  2176. if (ret) {
  2177. kfree(quota_root);
  2178. quota_root = fs_info->quota_root = NULL;
  2179. } else {
  2180. quota_root->track_dirty = 1;
  2181. fs_info->quota_enabled = 1;
  2182. fs_info->pending_quota_state = 1;
  2183. }
  2184. fs_info->generation = generation;
  2185. fs_info->last_trans_committed = generation;
  2186. ret = btrfs_recover_balance(fs_info);
  2187. if (ret) {
  2188. printk(KERN_WARNING "btrfs: failed to recover balance\n");
  2189. goto fail_block_groups;
  2190. }
  2191. ret = btrfs_init_dev_stats(fs_info);
  2192. if (ret) {
  2193. printk(KERN_ERR "btrfs: failed to init dev_stats: %d\n",
  2194. ret);
  2195. goto fail_block_groups;
  2196. }
  2197. ret = btrfs_init_dev_replace(fs_info);
  2198. if (ret) {
  2199. pr_err("btrfs: failed to init dev_replace: %d\n", ret);
  2200. goto fail_block_groups;
  2201. }
  2202. btrfs_close_extra_devices(fs_info, fs_devices, 1);
  2203. ret = btrfs_init_space_info(fs_info);
  2204. if (ret) {
  2205. printk(KERN_ERR "Failed to initial space info: %d\n", ret);
  2206. goto fail_block_groups;
  2207. }
  2208. ret = btrfs_read_block_groups(extent_root);
  2209. if (ret) {
  2210. printk(KERN_ERR "Failed to read block groups: %d\n", ret);
  2211. goto fail_block_groups;
  2212. }
  2213. fs_info->num_tolerated_disk_barrier_failures =
  2214. btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
  2215. if (fs_info->fs_devices->missing_devices >
  2216. fs_info->num_tolerated_disk_barrier_failures &&
  2217. !(sb->s_flags & MS_RDONLY)) {
  2218. printk(KERN_WARNING
  2219. "Btrfs: too many missing devices, writeable mount is not allowed\n");
  2220. goto fail_block_groups;
  2221. }
  2222. fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
  2223. "btrfs-cleaner");
  2224. if (IS_ERR(fs_info->cleaner_kthread))
  2225. goto fail_block_groups;
  2226. fs_info->transaction_kthread = kthread_run(transaction_kthread,
  2227. tree_root,
  2228. "btrfs-transaction");
  2229. if (IS_ERR(fs_info->transaction_kthread))
  2230. goto fail_cleaner;
  2231. if (!btrfs_test_opt(tree_root, SSD) &&
  2232. !btrfs_test_opt(tree_root, NOSSD) &&
  2233. !fs_info->fs_devices->rotating) {
  2234. printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
  2235. "mode\n");
  2236. btrfs_set_opt(fs_info->mount_opt, SSD);
  2237. }
  2238. #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
  2239. if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
  2240. ret = btrfsic_mount(tree_root, fs_devices,
  2241. btrfs_test_opt(tree_root,
  2242. CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
  2243. 1 : 0,
  2244. fs_info->check_integrity_print_mask);
  2245. if (ret)
  2246. printk(KERN_WARNING "btrfs: failed to initialize"
  2247. " integrity check module %s\n", sb->s_id);
  2248. }
  2249. #endif
  2250. ret = btrfs_read_qgroup_config(fs_info);
  2251. if (ret)
  2252. goto fail_trans_kthread;
  2253. /* do not make disk changes in broken FS */
  2254. if (btrfs_super_log_root(disk_super) != 0) {
  2255. u64 bytenr = btrfs_super_log_root(disk_super);
  2256. if (fs_devices->rw_devices == 0) {
  2257. printk(KERN_WARNING "Btrfs log replay required "
  2258. "on RO media\n");
  2259. err = -EIO;
  2260. goto fail_qgroup;
  2261. }
  2262. blocksize =
  2263. btrfs_level_size(tree_root,
  2264. btrfs_super_log_root_level(disk_super));
  2265. log_tree_root = btrfs_alloc_root(fs_info);
  2266. if (!log_tree_root) {
  2267. err = -ENOMEM;
  2268. goto fail_qgroup;
  2269. }
  2270. __setup_root(nodesize, leafsize, sectorsize, stripesize,
  2271. log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
  2272. log_tree_root->node = read_tree_block(tree_root, bytenr,
  2273. blocksize,
  2274. generation + 1);
  2275. /* returns with log_tree_root freed on success */
  2276. ret = btrfs_recover_log_trees(log_tree_root);
  2277. if (ret) {
  2278. btrfs_error(tree_root->fs_info, ret,
  2279. "Failed to recover log tree");
  2280. free_extent_buffer(log_tree_root->node);
  2281. kfree(log_tree_root);
  2282. goto fail_trans_kthread;
  2283. }
  2284. if (sb->s_flags & MS_RDONLY) {
  2285. ret = btrfs_commit_super(tree_root);
  2286. if (ret)
  2287. goto fail_trans_kthread;
  2288. }
  2289. }
  2290. ret = btrfs_find_orphan_roots(tree_root);
  2291. if (ret)
  2292. goto fail_trans_kthread;
  2293. if (!(sb->s_flags & MS_RDONLY)) {
  2294. ret = btrfs_cleanup_fs_roots(fs_info);
  2295. if (ret)
  2296. goto fail_trans_kthread;
  2297. ret = btrfs_recover_relocation(tree_root);
  2298. if (ret < 0) {
  2299. printk(KERN_WARNING
  2300. "btrfs: failed to recover relocation\n");
  2301. err = -EINVAL;
  2302. goto fail_qgroup;
  2303. }
  2304. }
  2305. location.objectid = BTRFS_FS_TREE_OBJECTID;
  2306. location.type = BTRFS_ROOT_ITEM_KEY;
  2307. location.offset = (u64)-1;
  2308. fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
  2309. if (!fs_info->fs_root)
  2310. goto fail_qgroup;
  2311. if (IS_ERR(fs_info->fs_root)) {
  2312. err = PTR_ERR(fs_info->fs_root);
  2313. goto fail_qgroup;
  2314. }
  2315. if (sb->s_flags & MS_RDONLY)
  2316. return 0;
  2317. down_read(&fs_info->cleanup_work_sem);
  2318. if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
  2319. (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
  2320. up_read(&fs_info->cleanup_work_sem);
  2321. close_ctree(tree_root);
  2322. return ret;
  2323. }
  2324. up_read(&fs_info->cleanup_work_sem);
  2325. ret = btrfs_resume_balance_async(fs_info);
  2326. if (ret) {
  2327. printk(KERN_WARNING "btrfs: failed to resume balance\n");
  2328. close_ctree(tree_root);
  2329. return ret;
  2330. }
  2331. ret = btrfs_resume_dev_replace_async(fs_info);
  2332. if (ret) {
  2333. pr_warn("btrfs: failed to resume dev_replace\n");
  2334. close_ctree(tree_root);
  2335. return ret;
  2336. }
  2337. return 0;
  2338. fail_qgroup:
  2339. btrfs_free_qgroup_config(fs_info);
  2340. fail_trans_kthread:
  2341. kthread_stop(fs_info->transaction_kthread);
  2342. fail_cleaner:
  2343. kthread_stop(fs_info->cleaner_kthread);
  2344. /*
  2345. * make sure we're done with the btree inode before we stop our
  2346. * kthreads
  2347. */
  2348. filemap_write_and_wait(fs_info->btree_inode->i_mapping);
  2349. invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
  2350. fail_block_groups:
  2351. btrfs_free_block_groups(fs_info);
  2352. fail_tree_roots:
  2353. free_root_pointers(fs_info, 1);
  2354. fail_sb_buffer:
  2355. btrfs_stop_workers(&fs_info->generic_worker);
  2356. btrfs_stop_workers(&fs_info->readahead_workers);
  2357. btrfs_stop_workers(&fs_info->fixup_workers);
  2358. btrfs_stop_workers(&fs_info->delalloc_workers);
  2359. btrfs_stop_workers(&fs_info->workers);
  2360. btrfs_stop_workers(&fs_info->endio_workers);
  2361. btrfs_stop_workers(&fs_info->endio_meta_workers);
  2362. btrfs_stop_workers(&fs_info->endio_meta_write_workers);
  2363. btrfs_stop_workers(&fs_info->endio_write_workers);
  2364. btrfs_stop_workers(&fs_info->endio_freespace_worker);
  2365. btrfs_stop_workers(&fs_info->submit_workers);
  2366. btrfs_stop_workers(&fs_info->delayed_workers);
  2367. btrfs_stop_workers(&fs_info->caching_workers);
  2368. btrfs_stop_workers(&fs_info->flush_workers);
  2369. fail_alloc:
  2370. fail_iput:
  2371. btrfs_mapping_tree_free(&fs_info->mapping_tree);
  2372. invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
  2373. iput(fs_info->btree_inode);
  2374. fail_bdi:
  2375. bdi_destroy(&fs_info->bdi);
  2376. fail_srcu:
  2377. cleanup_srcu_struct(&fs_info->subvol_srcu);
  2378. fail:
  2379. btrfs_close_devices(fs_info->fs_devices);
  2380. return err;
  2381. recovery_tree_root:
  2382. if (!btrfs_test_opt(tree_root, RECOVERY))
  2383. goto fail_tree_roots;
  2384. free_root_pointers(fs_info, 0);
  2385. /* don't use the log in recovery mode, it won't be valid */
  2386. btrfs_set_super_log_root(disk_super, 0);
  2387. /* we can't trust the free space cache either */
  2388. btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
  2389. ret = next_root_backup(fs_info, fs_info->super_copy,
  2390. &num_backups_tried, &backup_index);
  2391. if (ret == -1)
  2392. goto fail_block_groups;
  2393. goto retry_root_backup;
  2394. }
  2395. static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
  2396. {
  2397. if (uptodate) {
  2398. set_buffer_uptodate(bh);
  2399. } else {
  2400. struct btrfs_device *device = (struct btrfs_device *)
  2401. bh->b_private;
  2402. printk_ratelimited_in_rcu(KERN_WARNING "lost page write due to "
  2403. "I/O error on %s\n",
  2404. rcu_str_deref(device->name));
  2405. /* note, we dont' set_buffer_write_io_error because we have
  2406. * our own ways of dealing with the IO errors
  2407. */
  2408. clear_buffer_uptodate(bh);
  2409. btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
  2410. }
  2411. unlock_buffer(bh);
  2412. put_bh(bh);
  2413. }
  2414. struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
  2415. {
  2416. struct buffer_head *bh;
  2417. struct buffer_head *latest = NULL;
  2418. struct btrfs_super_block *super;
  2419. int i;
  2420. u64 transid = 0;
  2421. u64 bytenr;
  2422. /* we would like to check all the supers, but that would make
  2423. * a btrfs mount succeed after a mkfs from a different FS.
  2424. * So, we need to add a special mount option to scan for
  2425. * later supers, using BTRFS_SUPER_MIRROR_MAX instead
  2426. */
  2427. for (i = 0; i < 1; i++) {
  2428. bytenr = btrfs_sb_offset(i);
  2429. if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
  2430. break;
  2431. bh = __bread(bdev, bytenr / 4096, 4096);
  2432. if (!bh)
  2433. continue;
  2434. super = (struct btrfs_super_block *)bh->b_data;
  2435. if (btrfs_super_bytenr(super) != bytenr ||
  2436. strncmp((char *)(&super->magic), BTRFS_MAGIC,
  2437. sizeof(super->magic))) {
  2438. brelse(bh);
  2439. continue;
  2440. }
  2441. if (!latest || btrfs_super_generation(super) > transid) {
  2442. brelse(latest);
  2443. latest = bh;
  2444. transid = btrfs_super_generation(super);
  2445. } else {
  2446. brelse(bh);
  2447. }
  2448. }
  2449. return latest;
  2450. }
  2451. /*
  2452. * this should be called twice, once with wait == 0 and
  2453. * once with wait == 1. When wait == 0 is done, all the buffer heads
  2454. * we write are pinned.
  2455. *
  2456. * They are released when wait == 1 is done.
  2457. * max_mirrors must be the same for both runs, and it indicates how
  2458. * many supers on this one device should be written.
  2459. *
  2460. * max_mirrors == 0 means to write them all.
  2461. */
  2462. static int write_dev_supers(struct btrfs_device *device,
  2463. struct btrfs_super_block *sb,
  2464. int do_barriers, int wait, int max_mirrors)
  2465. {
  2466. struct buffer_head *bh;
  2467. int i;
  2468. int ret;
  2469. int errors = 0;
  2470. u32 crc;
  2471. u64 bytenr;
  2472. if (max_mirrors == 0)
  2473. max_mirrors = BTRFS_SUPER_MIRROR_MAX;
  2474. for (i = 0; i < max_mirrors; i++) {
  2475. bytenr = btrfs_sb_offset(i);
  2476. if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
  2477. break;
  2478. if (wait) {
  2479. bh = __find_get_block(device->bdev, bytenr / 4096,
  2480. BTRFS_SUPER_INFO_SIZE);
  2481. BUG_ON(!bh);
  2482. wait_on_buffer(bh);
  2483. if (!buffer_uptodate(bh))
  2484. errors++;
  2485. /* drop our reference */
  2486. brelse(bh);
  2487. /* drop the reference from the wait == 0 run */
  2488. brelse(bh);
  2489. continue;
  2490. } else {
  2491. btrfs_set_super_bytenr(sb, bytenr);
  2492. crc = ~(u32)0;
  2493. crc = btrfs_csum_data(NULL, (char *)sb +
  2494. BTRFS_CSUM_SIZE, crc,
  2495. BTRFS_SUPER_INFO_SIZE -
  2496. BTRFS_CSUM_SIZE);
  2497. btrfs_csum_final(crc, sb->csum);
  2498. /*
  2499. * one reference for us, and we leave it for the
  2500. * caller
  2501. */
  2502. bh = __getblk(device->bdev, bytenr / 4096,
  2503. BTRFS_SUPER_INFO_SIZE);
  2504. memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
  2505. /* one reference for submit_bh */
  2506. get_bh(bh);
  2507. set_buffer_uptodate(bh);
  2508. lock_buffer(bh);
  2509. bh->b_end_io = btrfs_end_buffer_write_sync;
  2510. bh->b_private = device;
  2511. }
  2512. /*
  2513. * we fua the first super. The others we allow
  2514. * to go down lazy.
  2515. */
  2516. ret = btrfsic_submit_bh(WRITE_FUA, bh);
  2517. if (ret)
  2518. errors++;
  2519. }
  2520. return errors < i ? 0 : -1;
  2521. }
  2522. /*
  2523. * endio for the write_dev_flush, this will wake anyone waiting
  2524. * for the barrier when it is done
  2525. */
  2526. static void btrfs_end_empty_barrier(struct bio *bio, int err)
  2527. {
  2528. if (err) {
  2529. if (err == -EOPNOTSUPP)
  2530. set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
  2531. clear_bit(BIO_UPTODATE, &bio->bi_flags);
  2532. }
  2533. if (bio->bi_private)
  2534. complete(bio->bi_private);
  2535. bio_put(bio);
  2536. }
  2537. /*
  2538. * trigger flushes for one the devices. If you pass wait == 0, the flushes are
  2539. * sent down. With wait == 1, it waits for the previous flush.
  2540. *
  2541. * any device where the flush fails with eopnotsupp are flagged as not-barrier
  2542. * capable
  2543. */
  2544. static int write_dev_flush(struct btrfs_device *device, int wait)
  2545. {
  2546. struct bio *bio;
  2547. int ret = 0;
  2548. if (device->nobarriers)
  2549. return 0;
  2550. if (wait) {
  2551. bio = device->flush_bio;
  2552. if (!bio)
  2553. return 0;
  2554. wait_for_completion(&device->flush_wait);
  2555. if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
  2556. printk_in_rcu("btrfs: disabling barriers on dev %s\n",
  2557. rcu_str_deref(device->name));
  2558. device->nobarriers = 1;
  2559. } else if (!bio_flagged(bio, BIO_UPTODATE)) {
  2560. ret = -EIO;
  2561. btrfs_dev_stat_inc_and_print(device,
  2562. BTRFS_DEV_STAT_FLUSH_ERRS);
  2563. }
  2564. /* drop the reference from the wait == 0 run */
  2565. bio_put(bio);
  2566. device->flush_bio = NULL;
  2567. return ret;
  2568. }
  2569. /*
  2570. * one reference for us, and we leave it for the
  2571. * caller
  2572. */
  2573. device->flush_bio = NULL;
  2574. bio = bio_alloc(GFP_NOFS, 0);
  2575. if (!bio)
  2576. return -ENOMEM;
  2577. bio->bi_end_io = btrfs_end_empty_barrier;
  2578. bio->bi_bdev = device->bdev;
  2579. init_completion(&device->flush_wait);
  2580. bio->bi_private = &device->flush_wait;
  2581. device->flush_bio = bio;
  2582. bio_get(bio);
  2583. btrfsic_submit_bio(WRITE_FLUSH, bio);
  2584. return 0;
  2585. }
  2586. /*
  2587. * send an empty flush down to each device in parallel,
  2588. * then wait for them
  2589. */
  2590. static int barrier_all_devices(struct btrfs_fs_info *info)
  2591. {
  2592. struct list_head *head;
  2593. struct btrfs_device *dev;
  2594. int errors_send = 0;
  2595. int errors_wait = 0;
  2596. int ret;
  2597. /* send down all the barriers */
  2598. head = &info->fs_devices->devices;
  2599. list_for_each_entry_rcu(dev, head, dev_list) {
  2600. if (!dev->bdev) {
  2601. errors_send++;
  2602. continue;
  2603. }
  2604. if (!dev->in_fs_metadata || !dev->writeable)
  2605. continue;
  2606. ret = write_dev_flush(dev, 0);
  2607. if (ret)
  2608. errors_send++;
  2609. }
  2610. /* wait for all the barriers */
  2611. list_for_each_entry_rcu(dev, head, dev_list) {
  2612. if (!dev->bdev) {
  2613. errors_wait++;
  2614. continue;
  2615. }
  2616. if (!dev->in_fs_metadata || !dev->writeable)
  2617. continue;
  2618. ret = write_dev_flush(dev, 1);
  2619. if (ret)
  2620. errors_wait++;
  2621. }
  2622. if (errors_send > info->num_tolerated_disk_barrier_failures ||
  2623. errors_wait > info->num_tolerated_disk_barrier_failures)
  2624. return -EIO;
  2625. return 0;
  2626. }
  2627. int btrfs_calc_num_tolerated_disk_barrier_failures(
  2628. struct btrfs_fs_info *fs_info)
  2629. {
  2630. struct btrfs_ioctl_space_info space;
  2631. struct btrfs_space_info *sinfo;
  2632. u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
  2633. BTRFS_BLOCK_GROUP_SYSTEM,
  2634. BTRFS_BLOCK_GROUP_METADATA,
  2635. BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
  2636. int num_types = 4;
  2637. int i;
  2638. int c;
  2639. int num_tolerated_disk_barrier_failures =
  2640. (int)fs_info->fs_devices->num_devices;
  2641. for (i = 0; i < num_types; i++) {
  2642. struct btrfs_space_info *tmp;
  2643. sinfo = NULL;
  2644. rcu_read_lock();
  2645. list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
  2646. if (tmp->flags == types[i]) {
  2647. sinfo = tmp;
  2648. break;
  2649. }
  2650. }
  2651. rcu_read_unlock();
  2652. if (!sinfo)
  2653. continue;
  2654. down_read(&sinfo->groups_sem);
  2655. for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
  2656. if (!list_empty(&sinfo->block_groups[c])) {
  2657. u64 flags;
  2658. btrfs_get_block_group_info(
  2659. &sinfo->block_groups[c], &space);
  2660. if (space.total_bytes == 0 ||
  2661. space.used_bytes == 0)
  2662. continue;
  2663. flags = space.flags;
  2664. /*
  2665. * return
  2666. * 0: if dup, single or RAID0 is configured for
  2667. * any of metadata, system or data, else
  2668. * 1: if RAID5 is configured, or if RAID1 or
  2669. * RAID10 is configured and only two mirrors
  2670. * are used, else
  2671. * 2: if RAID6 is configured, else
  2672. * num_mirrors - 1: if RAID1 or RAID10 is
  2673. * configured and more than
  2674. * 2 mirrors are used.
  2675. */
  2676. if (num_tolerated_disk_barrier_failures > 0 &&
  2677. ((flags & (BTRFS_BLOCK_GROUP_DUP |
  2678. BTRFS_BLOCK_GROUP_RAID0)) ||
  2679. ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)
  2680. == 0)))
  2681. num_tolerated_disk_barrier_failures = 0;
  2682. else if (num_tolerated_disk_barrier_failures > 1
  2683. &&
  2684. (flags & (BTRFS_BLOCK_GROUP_RAID1 |
  2685. BTRFS_BLOCK_GROUP_RAID10)))
  2686. num_tolerated_disk_barrier_failures = 1;
  2687. }
  2688. }
  2689. up_read(&sinfo->groups_sem);
  2690. }
  2691. return num_tolerated_disk_barrier_failures;
  2692. }
  2693. int write_all_supers(struct btrfs_root *root, int max_mirrors)
  2694. {
  2695. struct list_head *head;
  2696. struct btrfs_device *dev;
  2697. struct btrfs_super_block *sb;
  2698. struct btrfs_dev_item *dev_item;
  2699. int ret;
  2700. int do_barriers;
  2701. int max_errors;
  2702. int total_errors = 0;
  2703. u64 flags;
  2704. max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
  2705. do_barriers = !btrfs_test_opt(root, NOBARRIER);
  2706. backup_super_roots(root->fs_info);
  2707. sb = root->fs_info->super_for_commit;
  2708. dev_item = &sb->dev_item;
  2709. mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
  2710. head = &root->fs_info->fs_devices->devices;
  2711. if (do_barriers) {
  2712. ret = barrier_all_devices(root->fs_info);
  2713. if (ret) {
  2714. mutex_unlock(
  2715. &root->fs_info->fs_devices->device_list_mutex);
  2716. btrfs_error(root->fs_info, ret,
  2717. "errors while submitting device barriers.");
  2718. return ret;
  2719. }
  2720. }
  2721. list_for_each_entry_rcu(dev, head, dev_list) {
  2722. if (!dev->bdev) {
  2723. total_errors++;
  2724. continue;
  2725. }
  2726. if (!dev->in_fs_metadata || !dev->writeable)
  2727. continue;
  2728. btrfs_set_stack_device_generation(dev_item, 0);
  2729. btrfs_set_stack_device_type(dev_item, dev->type);
  2730. btrfs_set_stack_device_id(dev_item, dev->devid);
  2731. btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
  2732. btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
  2733. btrfs_set_stack_device_io_align(dev_item, dev->io_align);
  2734. btrfs_set_stack_device_io_width(dev_item, dev->io_width);
  2735. btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
  2736. memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
  2737. memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
  2738. flags = btrfs_super_flags(sb);
  2739. btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
  2740. ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
  2741. if (ret)
  2742. total_errors++;
  2743. }
  2744. if (total_errors > max_errors) {
  2745. printk(KERN_ERR "btrfs: %d errors while writing supers\n",
  2746. total_errors);
  2747. /* This shouldn't happen. FUA is masked off if unsupported */
  2748. BUG();
  2749. }
  2750. total_errors = 0;
  2751. list_for_each_entry_rcu(dev, head, dev_list) {
  2752. if (!dev->bdev)
  2753. continue;
  2754. if (!dev->in_fs_metadata || !dev->writeable)
  2755. continue;
  2756. ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
  2757. if (ret)
  2758. total_errors++;
  2759. }
  2760. mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
  2761. if (total_errors > max_errors) {
  2762. btrfs_error(root->fs_info, -EIO,
  2763. "%d errors while writing supers", total_errors);
  2764. return -EIO;
  2765. }
  2766. return 0;
  2767. }
  2768. int write_ctree_super(struct btrfs_trans_handle *trans,
  2769. struct btrfs_root *root, int max_mirrors)
  2770. {
  2771. int ret;
  2772. ret = write_all_supers(root, max_mirrors);
  2773. return ret;
  2774. }
  2775. void btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
  2776. {
  2777. spin_lock(&fs_info->fs_roots_radix_lock);
  2778. radix_tree_delete(&fs_info->fs_roots_radix,
  2779. (unsigned long)root->root_key.objectid);
  2780. spin_unlock(&fs_info->fs_roots_radix_lock);
  2781. if (btrfs_root_refs(&root->root_item) == 0)
  2782. synchronize_srcu(&fs_info->subvol_srcu);
  2783. __btrfs_remove_free_space_cache(root->free_ino_pinned);
  2784. __btrfs_remove_free_space_cache(root->free_ino_ctl);
  2785. free_fs_root(root);
  2786. }
  2787. static void free_fs_root(struct btrfs_root *root)
  2788. {
  2789. iput(root->cache_inode);
  2790. WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
  2791. if (root->anon_dev)
  2792. free_anon_bdev(root->anon_dev);
  2793. free_extent_buffer(root->node);
  2794. free_extent_buffer(root->commit_root);
  2795. kfree(root->free_ino_ctl);
  2796. kfree(root->free_ino_pinned);
  2797. kfree(root->name);
  2798. kfree(root);
  2799. }
  2800. static void del_fs_roots(struct btrfs_fs_info *fs_info)
  2801. {
  2802. int ret;
  2803. struct btrfs_root *gang[8];
  2804. int i;
  2805. while (!list_empty(&fs_info->dead_roots)) {
  2806. gang[0] = list_entry(fs_info->dead_roots.next,
  2807. struct btrfs_root, root_list);
  2808. list_del(&gang[0]->root_list);
  2809. if (gang[0]->in_radix) {
  2810. btrfs_free_fs_root(fs_info, gang[0]);
  2811. } else {
  2812. free_extent_buffer(gang[0]->node);
  2813. free_extent_buffer(gang[0]->commit_root);
  2814. kfree(gang[0]);
  2815. }
  2816. }
  2817. while (1) {
  2818. ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
  2819. (void **)gang, 0,
  2820. ARRAY_SIZE(gang));
  2821. if (!ret)
  2822. break;
  2823. for (i = 0; i < ret; i++)
  2824. btrfs_free_fs_root(fs_info, gang[i]);
  2825. }
  2826. }
  2827. int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
  2828. {
  2829. u64 root_objectid = 0;
  2830. struct btrfs_root *gang[8];
  2831. int i;
  2832. int ret;
  2833. while (1) {
  2834. ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
  2835. (void **)gang, root_objectid,
  2836. ARRAY_SIZE(gang));
  2837. if (!ret)
  2838. break;
  2839. root_objectid = gang[ret - 1]->root_key.objectid + 1;
  2840. for (i = 0; i < ret; i++) {
  2841. int err;
  2842. root_objectid = gang[i]->root_key.objectid;
  2843. err = btrfs_orphan_cleanup(gang[i]);
  2844. if (err)
  2845. return err;
  2846. }
  2847. root_objectid++;
  2848. }
  2849. return 0;
  2850. }
  2851. int btrfs_commit_super(struct btrfs_root *root)
  2852. {
  2853. struct btrfs_trans_handle *trans;
  2854. int ret;
  2855. mutex_lock(&root->fs_info->cleaner_mutex);
  2856. btrfs_run_delayed_iputs(root);
  2857. btrfs_clean_old_snapshots(root);
  2858. mutex_unlock(&root->fs_info->cleaner_mutex);
  2859. /* wait until ongoing cleanup work done */
  2860. down_write(&root->fs_info->cleanup_work_sem);
  2861. up_write(&root->fs_info->cleanup_work_sem);
  2862. trans = btrfs_join_transaction(root);
  2863. if (IS_ERR(trans))
  2864. return PTR_ERR(trans);
  2865. ret = btrfs_commit_transaction(trans, root);
  2866. if (ret)
  2867. return ret;
  2868. /* run commit again to drop the original snapshot */
  2869. trans = btrfs_join_transaction(root);
  2870. if (IS_ERR(trans))
  2871. return PTR_ERR(trans);
  2872. ret = btrfs_commit_transaction(trans, root);
  2873. if (ret)
  2874. return ret;
  2875. ret = btrfs_write_and_wait_transaction(NULL, root);
  2876. if (ret) {
  2877. btrfs_error(root->fs_info, ret,
  2878. "Failed to sync btree inode to disk.");
  2879. return ret;
  2880. }
  2881. ret = write_ctree_super(NULL, root, 0);
  2882. return ret;
  2883. }
  2884. int close_ctree(struct btrfs_root *root)
  2885. {
  2886. struct btrfs_fs_info *fs_info = root->fs_info;
  2887. int ret;
  2888. fs_info->closing = 1;
  2889. smp_mb();
  2890. /* pause restriper - we want to resume on mount */
  2891. btrfs_pause_balance(fs_info);
  2892. btrfs_dev_replace_suspend_for_unmount(fs_info);
  2893. btrfs_scrub_cancel(fs_info);
  2894. /* wait for any defraggers to finish */
  2895. wait_event(fs_info->transaction_wait,
  2896. (atomic_read(&fs_info->defrag_running) == 0));
  2897. /* clear out the rbtree of defraggable inodes */
  2898. btrfs_cleanup_defrag_inodes(fs_info);
  2899. if (!(fs_info->sb->s_flags & MS_RDONLY)) {
  2900. ret = btrfs_commit_super(root);
  2901. if (ret)
  2902. printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
  2903. }
  2904. if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
  2905. btrfs_error_commit_super(root);
  2906. btrfs_put_block_group_cache(fs_info);
  2907. kthread_stop(fs_info->transaction_kthread);
  2908. kthread_stop(fs_info->cleaner_kthread);
  2909. fs_info->closing = 2;
  2910. smp_mb();
  2911. btrfs_free_qgroup_config(root->fs_info);
  2912. if (fs_info->delalloc_bytes) {
  2913. printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
  2914. (unsigned long long)fs_info->delalloc_bytes);
  2915. }
  2916. free_extent_buffer(fs_info->extent_root->node);
  2917. free_extent_buffer(fs_info->extent_root->commit_root);
  2918. free_extent_buffer(fs_info->tree_root->node);
  2919. free_extent_buffer(fs_info->tree_root->commit_root);
  2920. free_extent_buffer(fs_info->chunk_root->node);
  2921. free_extent_buffer(fs_info->chunk_root->commit_root);
  2922. free_extent_buffer(fs_info->dev_root->node);
  2923. free_extent_buffer(fs_info->dev_root->commit_root);
  2924. free_extent_buffer(fs_info->csum_root->node);
  2925. free_extent_buffer(fs_info->csum_root->commit_root);
  2926. if (fs_info->quota_root) {
  2927. free_extent_buffer(fs_info->quota_root->node);
  2928. free_extent_buffer(fs_info->quota_root->commit_root);
  2929. }
  2930. btrfs_free_block_groups(fs_info);
  2931. del_fs_roots(fs_info);
  2932. iput(fs_info->btree_inode);
  2933. btrfs_stop_workers(&fs_info->generic_worker);
  2934. btrfs_stop_workers(&fs_info->fixup_workers);
  2935. btrfs_stop_workers(&fs_info->delalloc_workers);
  2936. btrfs_stop_workers(&fs_info->workers);
  2937. btrfs_stop_workers(&fs_info->endio_workers);
  2938. btrfs_stop_workers(&fs_info->endio_meta_workers);
  2939. btrfs_stop_workers(&fs_info->endio_meta_write_workers);
  2940. btrfs_stop_workers(&fs_info->endio_write_workers);
  2941. btrfs_stop_workers(&fs_info->endio_freespace_worker);
  2942. btrfs_stop_workers(&fs_info->submit_workers);
  2943. btrfs_stop_workers(&fs_info->delayed_workers);
  2944. btrfs_stop_workers(&fs_info->caching_workers);
  2945. btrfs_stop_workers(&fs_info->readahead_workers);
  2946. btrfs_stop_workers(&fs_info->flush_workers);
  2947. #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
  2948. if (btrfs_test_opt(root, CHECK_INTEGRITY))
  2949. btrfsic_unmount(root, fs_info->fs_devices);
  2950. #endif
  2951. btrfs_close_devices(fs_info->fs_devices);
  2952. btrfs_mapping_tree_free(&fs_info->mapping_tree);
  2953. bdi_destroy(&fs_info->bdi);
  2954. cleanup_srcu_struct(&fs_info->subvol_srcu);
  2955. return 0;
  2956. }
  2957. int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
  2958. int atomic)
  2959. {
  2960. int ret;
  2961. struct inode *btree_inode = buf->pages[0]->mapping->host;
  2962. ret = extent_buffer_uptodate(buf);
  2963. if (!ret)
  2964. return ret;
  2965. ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
  2966. parent_transid, atomic);
  2967. if (ret == -EAGAIN)
  2968. return ret;
  2969. return !ret;
  2970. }
  2971. int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
  2972. {
  2973. return set_extent_buffer_uptodate(buf);
  2974. }
  2975. void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
  2976. {
  2977. struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
  2978. u64 transid = btrfs_header_generation(buf);
  2979. int was_dirty;
  2980. btrfs_assert_tree_locked(buf);
  2981. if (transid != root->fs_info->generation)
  2982. WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
  2983. "found %llu running %llu\n",
  2984. (unsigned long long)buf->start,
  2985. (unsigned long long)transid,
  2986. (unsigned long long)root->fs_info->generation);
  2987. was_dirty = set_extent_buffer_dirty(buf);
  2988. if (!was_dirty) {
  2989. spin_lock(&root->fs_info->delalloc_lock);
  2990. root->fs_info->dirty_metadata_bytes += buf->len;
  2991. spin_unlock(&root->fs_info->delalloc_lock);
  2992. }
  2993. }
  2994. static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
  2995. int flush_delayed)
  2996. {
  2997. /*
  2998. * looks as though older kernels can get into trouble with
  2999. * this code, they end up stuck in balance_dirty_pages forever
  3000. */
  3001. u64 num_dirty;
  3002. unsigned long thresh = 32 * 1024 * 1024;
  3003. if (current->flags & PF_MEMALLOC)
  3004. return;
  3005. if (flush_delayed)
  3006. btrfs_balance_delayed_items(root);
  3007. num_dirty = root->fs_info->dirty_metadata_bytes;
  3008. if (num_dirty > thresh) {
  3009. balance_dirty_pages_ratelimited_nr(
  3010. root->fs_info->btree_inode->i_mapping, 1);
  3011. }
  3012. return;
  3013. }
  3014. void btrfs_btree_balance_dirty(struct btrfs_root *root)
  3015. {
  3016. __btrfs_btree_balance_dirty(root, 1);
  3017. }
  3018. void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
  3019. {
  3020. __btrfs_btree_balance_dirty(root, 0);
  3021. }
  3022. int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
  3023. {
  3024. struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
  3025. return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
  3026. }
  3027. static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
  3028. int read_only)
  3029. {
  3030. if (btrfs_super_csum_type(fs_info->super_copy) >= ARRAY_SIZE(btrfs_csum_sizes)) {
  3031. printk(KERN_ERR "btrfs: unsupported checksum algorithm\n");
  3032. return -EINVAL;
  3033. }
  3034. if (read_only)
  3035. return 0;
  3036. return 0;
  3037. }
  3038. void btrfs_error_commit_super(struct btrfs_root *root)
  3039. {
  3040. mutex_lock(&root->fs_info->cleaner_mutex);
  3041. btrfs_run_delayed_iputs(root);
  3042. mutex_unlock(&root->fs_info->cleaner_mutex);
  3043. down_write(&root->fs_info->cleanup_work_sem);
  3044. up_write(&root->fs_info->cleanup_work_sem);
  3045. /* cleanup FS via transaction */
  3046. btrfs_cleanup_transaction(root);
  3047. }
  3048. static void btrfs_destroy_ordered_operations(struct btrfs_root *root)
  3049. {
  3050. struct btrfs_inode *btrfs_inode;
  3051. struct list_head splice;
  3052. INIT_LIST_HEAD(&splice);
  3053. mutex_lock(&root->fs_info->ordered_operations_mutex);
  3054. spin_lock(&root->fs_info->ordered_extent_lock);
  3055. list_splice_init(&root->fs_info->ordered_operations, &splice);
  3056. while (!list_empty(&splice)) {
  3057. btrfs_inode = list_entry(splice.next, struct btrfs_inode,
  3058. ordered_operations);
  3059. list_del_init(&btrfs_inode->ordered_operations);
  3060. btrfs_invalidate_inodes(btrfs_inode->root);
  3061. }
  3062. spin_unlock(&root->fs_info->ordered_extent_lock);
  3063. mutex_unlock(&root->fs_info->ordered_operations_mutex);
  3064. }
  3065. static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
  3066. {
  3067. struct list_head splice;
  3068. struct btrfs_ordered_extent *ordered;
  3069. struct inode *inode;
  3070. INIT_LIST_HEAD(&splice);
  3071. spin_lock(&root->fs_info->ordered_extent_lock);
  3072. list_splice_init(&root->fs_info->ordered_extents, &splice);
  3073. while (!list_empty(&splice)) {
  3074. ordered = list_entry(splice.next, struct btrfs_ordered_extent,
  3075. root_extent_list);
  3076. list_del_init(&ordered->root_extent_list);
  3077. atomic_inc(&ordered->refs);
  3078. /* the inode may be getting freed (in sys_unlink path). */
  3079. inode = igrab(ordered->inode);
  3080. spin_unlock(&root->fs_info->ordered_extent_lock);
  3081. if (inode)
  3082. iput(inode);
  3083. atomic_set(&ordered->refs, 1);
  3084. btrfs_put_ordered_extent(ordered);
  3085. spin_lock(&root->fs_info->ordered_extent_lock);
  3086. }
  3087. spin_unlock(&root->fs_info->ordered_extent_lock);
  3088. }
  3089. int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
  3090. struct btrfs_root *root)
  3091. {
  3092. struct rb_node *node;
  3093. struct btrfs_delayed_ref_root *delayed_refs;
  3094. struct btrfs_delayed_ref_node *ref;
  3095. int ret = 0;
  3096. delayed_refs = &trans->delayed_refs;
  3097. spin_lock(&delayed_refs->lock);
  3098. if (delayed_refs->num_entries == 0) {
  3099. spin_unlock(&delayed_refs->lock);
  3100. printk(KERN_INFO "delayed_refs has NO entry\n");
  3101. return ret;
  3102. }
  3103. while ((node = rb_first(&delayed_refs->root)) != NULL) {
  3104. ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
  3105. atomic_set(&ref->refs, 1);
  3106. if (btrfs_delayed_ref_is_head(ref)) {
  3107. struct btrfs_delayed_ref_head *head;
  3108. head = btrfs_delayed_node_to_head(ref);
  3109. if (!mutex_trylock(&head->mutex)) {
  3110. atomic_inc(&ref->refs);
  3111. spin_unlock(&delayed_refs->lock);
  3112. /* Need to wait for the delayed ref to run */
  3113. mutex_lock(&head->mutex);
  3114. mutex_unlock(&head->mutex);
  3115. btrfs_put_delayed_ref(ref);
  3116. spin_lock(&delayed_refs->lock);
  3117. continue;
  3118. }
  3119. btrfs_free_delayed_extent_op(head->extent_op);
  3120. delayed_refs->num_heads--;
  3121. if (list_empty(&head->cluster))
  3122. delayed_refs->num_heads_ready--;
  3123. list_del_init(&head->cluster);
  3124. }
  3125. ref->in_tree = 0;
  3126. rb_erase(&ref->rb_node, &delayed_refs->root);
  3127. delayed_refs->num_entries--;
  3128. spin_unlock(&delayed_refs->lock);
  3129. btrfs_put_delayed_ref(ref);
  3130. cond_resched();
  3131. spin_lock(&delayed_refs->lock);
  3132. }
  3133. spin_unlock(&delayed_refs->lock);
  3134. return ret;
  3135. }
  3136. static void btrfs_destroy_pending_snapshots(struct btrfs_transaction *t)
  3137. {
  3138. struct btrfs_pending_snapshot *snapshot;
  3139. struct list_head splice;
  3140. INIT_LIST_HEAD(&splice);
  3141. list_splice_init(&t->pending_snapshots, &splice);
  3142. while (!list_empty(&splice)) {
  3143. snapshot = list_entry(splice.next,
  3144. struct btrfs_pending_snapshot,
  3145. list);
  3146. list_del_init(&snapshot->list);
  3147. kfree(snapshot);
  3148. }
  3149. }
  3150. static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
  3151. {
  3152. struct btrfs_inode *btrfs_inode;
  3153. struct list_head splice;
  3154. INIT_LIST_HEAD(&splice);
  3155. spin_lock(&root->fs_info->delalloc_lock);
  3156. list_splice_init(&root->fs_info->delalloc_inodes, &splice);
  3157. while (!list_empty(&splice)) {
  3158. btrfs_inode = list_entry(splice.next, struct btrfs_inode,
  3159. delalloc_inodes);
  3160. list_del_init(&btrfs_inode->delalloc_inodes);
  3161. btrfs_invalidate_inodes(btrfs_inode->root);
  3162. }
  3163. spin_unlock(&root->fs_info->delalloc_lock);
  3164. }
  3165. static int btrfs_destroy_marked_extents(struct btrfs_root *root,
  3166. struct extent_io_tree *dirty_pages,
  3167. int mark)
  3168. {
  3169. int ret;
  3170. struct page *page;
  3171. struct inode *btree_inode = root->fs_info->btree_inode;
  3172. struct extent_buffer *eb;
  3173. u64 start = 0;
  3174. u64 end;
  3175. u64 offset;
  3176. unsigned long index;
  3177. while (1) {
  3178. ret = find_first_extent_bit(dirty_pages, start, &start, &end,
  3179. mark, NULL);
  3180. if (ret)
  3181. break;
  3182. clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
  3183. while (start <= end) {
  3184. index = start >> PAGE_CACHE_SHIFT;
  3185. start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
  3186. page = find_get_page(btree_inode->i_mapping, index);
  3187. if (!page)
  3188. continue;
  3189. offset = page_offset(page);
  3190. spin_lock(&dirty_pages->buffer_lock);
  3191. eb = radix_tree_lookup(
  3192. &(&BTRFS_I(page->mapping->host)->io_tree)->buffer,
  3193. offset >> PAGE_CACHE_SHIFT);
  3194. spin_unlock(&dirty_pages->buffer_lock);
  3195. if (eb)
  3196. ret = test_and_clear_bit(EXTENT_BUFFER_DIRTY,
  3197. &eb->bflags);
  3198. if (PageWriteback(page))
  3199. end_page_writeback(page);
  3200. lock_page(page);
  3201. if (PageDirty(page)) {
  3202. clear_page_dirty_for_io(page);
  3203. spin_lock_irq(&page->mapping->tree_lock);
  3204. radix_tree_tag_clear(&page->mapping->page_tree,
  3205. page_index(page),
  3206. PAGECACHE_TAG_DIRTY);
  3207. spin_unlock_irq(&page->mapping->tree_lock);
  3208. }
  3209. unlock_page(page);
  3210. page_cache_release(page);
  3211. }
  3212. }
  3213. return ret;
  3214. }
  3215. static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
  3216. struct extent_io_tree *pinned_extents)
  3217. {
  3218. struct extent_io_tree *unpin;
  3219. u64 start;
  3220. u64 end;
  3221. int ret;
  3222. bool loop = true;
  3223. unpin = pinned_extents;
  3224. again:
  3225. while (1) {
  3226. ret = find_first_extent_bit(unpin, 0, &start, &end,
  3227. EXTENT_DIRTY, NULL);
  3228. if (ret)
  3229. break;
  3230. /* opt_discard */
  3231. if (btrfs_test_opt(root, DISCARD))
  3232. ret = btrfs_error_discard_extent(root, start,
  3233. end + 1 - start,
  3234. NULL);
  3235. clear_extent_dirty(unpin, start, end, GFP_NOFS);
  3236. btrfs_error_unpin_extent_range(root, start, end);
  3237. cond_resched();
  3238. }
  3239. if (loop) {
  3240. if (unpin == &root->fs_info->freed_extents[0])
  3241. unpin = &root->fs_info->freed_extents[1];
  3242. else
  3243. unpin = &root->fs_info->freed_extents[0];
  3244. loop = false;
  3245. goto again;
  3246. }
  3247. return 0;
  3248. }
  3249. void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
  3250. struct btrfs_root *root)
  3251. {
  3252. btrfs_destroy_delayed_refs(cur_trans, root);
  3253. btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv,
  3254. cur_trans->dirty_pages.dirty_bytes);
  3255. /* FIXME: cleanup wait for commit */
  3256. cur_trans->in_commit = 1;
  3257. cur_trans->blocked = 1;
  3258. wake_up(&root->fs_info->transaction_blocked_wait);
  3259. cur_trans->blocked = 0;
  3260. wake_up(&root->fs_info->transaction_wait);
  3261. cur_trans->commit_done = 1;
  3262. wake_up(&cur_trans->commit_wait);
  3263. btrfs_destroy_delayed_inodes(root);
  3264. btrfs_assert_delayed_root_empty(root);
  3265. btrfs_destroy_pending_snapshots(cur_trans);
  3266. btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
  3267. EXTENT_DIRTY);
  3268. btrfs_destroy_pinned_extent(root,
  3269. root->fs_info->pinned_extents);
  3270. /*
  3271. memset(cur_trans, 0, sizeof(*cur_trans));
  3272. kmem_cache_free(btrfs_transaction_cachep, cur_trans);
  3273. */
  3274. }
  3275. int btrfs_cleanup_transaction(struct btrfs_root *root)
  3276. {
  3277. struct btrfs_transaction *t;
  3278. LIST_HEAD(list);
  3279. mutex_lock(&root->fs_info->transaction_kthread_mutex);
  3280. spin_lock(&root->fs_info->trans_lock);
  3281. list_splice_init(&root->fs_info->trans_list, &list);
  3282. root->fs_info->trans_no_join = 1;
  3283. spin_unlock(&root->fs_info->trans_lock);
  3284. while (!list_empty(&list)) {
  3285. t = list_entry(list.next, struct btrfs_transaction, list);
  3286. if (!t)
  3287. break;
  3288. btrfs_destroy_ordered_operations(root);
  3289. btrfs_destroy_ordered_extents(root);
  3290. btrfs_destroy_delayed_refs(t, root);
  3291. btrfs_block_rsv_release(root,
  3292. &root->fs_info->trans_block_rsv,
  3293. t->dirty_pages.dirty_bytes);
  3294. /* FIXME: cleanup wait for commit */
  3295. t->in_commit = 1;
  3296. t->blocked = 1;
  3297. smp_mb();
  3298. if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
  3299. wake_up(&root->fs_info->transaction_blocked_wait);
  3300. t->blocked = 0;
  3301. smp_mb();
  3302. if (waitqueue_active(&root->fs_info->transaction_wait))
  3303. wake_up(&root->fs_info->transaction_wait);
  3304. t->commit_done = 1;
  3305. smp_mb();
  3306. if (waitqueue_active(&t->commit_wait))
  3307. wake_up(&t->commit_wait);
  3308. btrfs_destroy_delayed_inodes(root);
  3309. btrfs_assert_delayed_root_empty(root);
  3310. btrfs_destroy_pending_snapshots(t);
  3311. btrfs_destroy_delalloc_inodes(root);
  3312. spin_lock(&root->fs_info->trans_lock);
  3313. root->fs_info->running_transaction = NULL;
  3314. spin_unlock(&root->fs_info->trans_lock);
  3315. btrfs_destroy_marked_extents(root, &t->dirty_pages,
  3316. EXTENT_DIRTY);
  3317. btrfs_destroy_pinned_extent(root,
  3318. root->fs_info->pinned_extents);
  3319. atomic_set(&t->use_count, 0);
  3320. list_del_init(&t->list);
  3321. memset(t, 0, sizeof(*t));
  3322. kmem_cache_free(btrfs_transaction_cachep, t);
  3323. }
  3324. spin_lock(&root->fs_info->trans_lock);
  3325. root->fs_info->trans_no_join = 0;
  3326. spin_unlock(&root->fs_info->trans_lock);
  3327. mutex_unlock(&root->fs_info->transaction_kthread_mutex);
  3328. return 0;
  3329. }
  3330. static struct extent_io_ops btree_extent_io_ops = {
  3331. .readpage_end_io_hook = btree_readpage_end_io_hook,
  3332. .readpage_io_failed_hook = btree_io_failed_hook,
  3333. .submit_bio_hook = btree_submit_bio_hook,
  3334. /* note we're sharing with inode.c for the merge bio hook */
  3335. .merge_bio_hook = btrfs_merge_bio_hook,
  3336. };