disk-io.c 106 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919920921922923924925926927928929930931932933934935936937938939940941942943944945946947948949950951952953954955956957958959960961962963964965966967968969970971972973974975976977978979980981982983984985986987988989990991992993994995996997998999100010011002100310041005100610071008100910101011101210131014101510161017101810191020102110221023102410251026102710281029103010311032103310341035103610371038103910401041104210431044104510461047104810491050105110521053105410551056105710581059106010611062106310641065106610671068106910701071107210731074107510761077107810791080108110821083108410851086108710881089109010911092109310941095109610971098109911001101110211031104110511061107110811091110111111121113111411151116111711181119112011211122112311241125112611271128112911301131113211331134113511361137113811391140114111421143114411451146114711481149115011511152115311541155115611571158115911601161116211631164116511661167116811691170117111721173117411751176117711781179118011811182118311841185118611871188118911901191119211931194119511961197119811991200120112021203120412051206120712081209121012111212121312141215121612171218121912201221122212231224122512261227122812291230123112321233123412351236123712381239124012411242124312441245124612471248124912501251125212531254125512561257125812591260126112621263126412651266126712681269127012711272127312741275127612771278127912801281128212831284128512861287128812891290129112921293129412951296129712981299130013011302130313041305130613071308130913101311131213131314131513161317131813191320132113221323132413251326132713281329133013311332133313341335133613371338133913401341134213431344134513461347134813491350135113521353135413551356135713581359136013611362136313641365136613671368136913701371137213731374137513761377137813791380138113821383138413851386138713881389139013911392139313941395139613971398139914001401140214031404140514061407140814091410141114121413141414151416141714181419142014211422142314241425142614271428142914301431143214331434143514361437143814391440144114421443144414451446144714481449145014511452145314541455145614571458145914601461146214631464146514661467146814691470147114721473147414751476147714781479148014811482148314841485148614871488148914901491149214931494149514961497149814991500150115021503150415051506150715081509151015111512151315141515151615171518151915201521152215231524152515261527152815291530153115321533153415351536153715381539154015411542154315441545154615471548154915501551155215531554155515561557155815591560156115621563156415651566156715681569157015711572157315741575157615771578157915801581158215831584158515861587158815891590159115921593159415951596159715981599160016011602160316041605160616071608160916101611161216131614161516161617161816191620162116221623162416251626162716281629163016311632163316341635163616371638163916401641164216431644164516461647164816491650165116521653165416551656165716581659166016611662166316641665166616671668166916701671167216731674167516761677167816791680168116821683168416851686168716881689169016911692169316941695169616971698169917001701170217031704170517061707170817091710171117121713171417151716171717181719172017211722172317241725172617271728172917301731173217331734173517361737173817391740174117421743174417451746174717481749175017511752175317541755175617571758175917601761176217631764176517661767176817691770177117721773177417751776177717781779178017811782178317841785178617871788178917901791179217931794179517961797179817991800180118021803180418051806180718081809181018111812181318141815181618171818181918201821182218231824182518261827182818291830183118321833183418351836183718381839184018411842184318441845184618471848184918501851185218531854185518561857185818591860186118621863186418651866186718681869187018711872187318741875187618771878187918801881188218831884188518861887188818891890189118921893189418951896189718981899190019011902190319041905190619071908190919101911191219131914191519161917191819191920192119221923192419251926192719281929193019311932193319341935193619371938193919401941194219431944194519461947194819491950195119521953195419551956195719581959196019611962196319641965196619671968196919701971197219731974197519761977197819791980198119821983198419851986198719881989199019911992199319941995199619971998199920002001200220032004200520062007200820092010201120122013201420152016201720182019202020212022202320242025202620272028202920302031203220332034203520362037203820392040204120422043204420452046204720482049205020512052205320542055205620572058205920602061206220632064206520662067206820692070207120722073207420752076207720782079208020812082208320842085208620872088208920902091209220932094209520962097209820992100210121022103210421052106210721082109211021112112211321142115211621172118211921202121212221232124212521262127212821292130213121322133213421352136213721382139214021412142214321442145214621472148214921502151215221532154215521562157215821592160216121622163216421652166216721682169217021712172217321742175217621772178217921802181218221832184218521862187218821892190219121922193219421952196219721982199220022012202220322042205220622072208220922102211221222132214221522162217221822192220222122222223222422252226222722282229223022312232223322342235223622372238223922402241224222432244224522462247224822492250225122522253225422552256225722582259226022612262226322642265226622672268226922702271227222732274227522762277227822792280228122822283228422852286228722882289229022912292229322942295229622972298229923002301230223032304230523062307230823092310231123122313231423152316231723182319232023212322232323242325232623272328232923302331233223332334233523362337233823392340234123422343234423452346234723482349235023512352235323542355235623572358235923602361236223632364236523662367236823692370237123722373237423752376237723782379238023812382238323842385238623872388238923902391239223932394239523962397239823992400240124022403240424052406240724082409241024112412241324142415241624172418241924202421242224232424242524262427242824292430243124322433243424352436243724382439244024412442244324442445244624472448244924502451245224532454245524562457245824592460246124622463246424652466246724682469247024712472247324742475247624772478247924802481248224832484248524862487248824892490249124922493249424952496249724982499250025012502250325042505250625072508250925102511251225132514251525162517251825192520252125222523252425252526252725282529253025312532253325342535253625372538253925402541254225432544254525462547254825492550255125522553255425552556255725582559256025612562256325642565256625672568256925702571257225732574257525762577257825792580258125822583258425852586258725882589259025912592259325942595259625972598259926002601260226032604260526062607260826092610261126122613261426152616261726182619262026212622262326242625262626272628262926302631263226332634263526362637263826392640264126422643264426452646264726482649265026512652265326542655265626572658265926602661266226632664266526662667266826692670267126722673267426752676267726782679268026812682268326842685268626872688268926902691269226932694269526962697269826992700270127022703270427052706270727082709271027112712271327142715271627172718271927202721272227232724272527262727272827292730273127322733273427352736273727382739274027412742274327442745274627472748274927502751275227532754275527562757275827592760276127622763276427652766276727682769277027712772277327742775277627772778277927802781278227832784278527862787278827892790279127922793279427952796279727982799280028012802280328042805280628072808280928102811281228132814281528162817281828192820282128222823282428252826282728282829283028312832283328342835283628372838283928402841284228432844284528462847284828492850285128522853285428552856285728582859286028612862286328642865286628672868286928702871287228732874287528762877287828792880288128822883288428852886288728882889289028912892289328942895289628972898289929002901290229032904290529062907290829092910291129122913291429152916291729182919292029212922292329242925292629272928292929302931293229332934293529362937293829392940294129422943294429452946294729482949295029512952295329542955295629572958295929602961296229632964296529662967296829692970297129722973297429752976297729782979298029812982298329842985298629872988298929902991299229932994299529962997299829993000300130023003300430053006300730083009301030113012301330143015301630173018301930203021302230233024302530263027302830293030303130323033303430353036303730383039304030413042304330443045304630473048304930503051305230533054305530563057305830593060306130623063306430653066306730683069307030713072307330743075307630773078307930803081308230833084308530863087308830893090309130923093309430953096309730983099310031013102310331043105310631073108310931103111311231133114311531163117311831193120312131223123312431253126312731283129313031313132313331343135313631373138313931403141314231433144314531463147314831493150315131523153315431553156315731583159316031613162316331643165316631673168316931703171317231733174317531763177317831793180318131823183318431853186318731883189319031913192319331943195319631973198319932003201320232033204320532063207320832093210321132123213321432153216321732183219322032213222322332243225322632273228322932303231323232333234323532363237323832393240324132423243324432453246324732483249325032513252325332543255325632573258325932603261326232633264326532663267326832693270327132723273327432753276327732783279328032813282328332843285328632873288328932903291329232933294329532963297329832993300330133023303330433053306330733083309331033113312331333143315331633173318331933203321332233233324332533263327332833293330333133323333333433353336333733383339334033413342334333443345334633473348334933503351335233533354335533563357335833593360336133623363336433653366336733683369337033713372337333743375337633773378337933803381338233833384338533863387338833893390339133923393339433953396339733983399340034013402340334043405340634073408340934103411341234133414341534163417341834193420342134223423342434253426342734283429343034313432343334343435343634373438343934403441344234433444344534463447344834493450345134523453345434553456345734583459346034613462346334643465346634673468346934703471347234733474347534763477347834793480348134823483348434853486348734883489349034913492349334943495349634973498349935003501350235033504350535063507350835093510351135123513351435153516351735183519352035213522352335243525352635273528352935303531353235333534353535363537353835393540354135423543354435453546354735483549355035513552355335543555355635573558355935603561356235633564356535663567356835693570357135723573357435753576357735783579358035813582358335843585358635873588358935903591359235933594359535963597359835993600360136023603360436053606360736083609361036113612361336143615361636173618361936203621362236233624362536263627362836293630363136323633363436353636363736383639364036413642364336443645364636473648364936503651365236533654365536563657365836593660366136623663366436653666366736683669367036713672367336743675367636773678367936803681368236833684368536863687368836893690369136923693369436953696369736983699370037013702370337043705370637073708370937103711371237133714371537163717371837193720372137223723372437253726372737283729373037313732373337343735373637373738373937403741374237433744374537463747374837493750375137523753375437553756375737583759376037613762376337643765376637673768376937703771377237733774377537763777377837793780378137823783378437853786378737883789379037913792379337943795379637973798379938003801380238033804380538063807380838093810381138123813381438153816381738183819382038213822382338243825382638273828382938303831383238333834383538363837383838393840384138423843384438453846384738483849385038513852385338543855385638573858385938603861386238633864386538663867386838693870387138723873387438753876387738783879388038813882388338843885388638873888388938903891389238933894389538963897389838993900390139023903390439053906390739083909391039113912391339143915391639173918
  1. /*
  2. * Copyright (C) 2007 Oracle. All rights reserved.
  3. *
  4. * This program is free software; you can redistribute it and/or
  5. * modify it under the terms of the GNU General Public
  6. * License v2 as published by the Free Software Foundation.
  7. *
  8. * This program is distributed in the hope that it will be useful,
  9. * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10. * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  11. * General Public License for more details.
  12. *
  13. * You should have received a copy of the GNU General Public
  14. * License along with this program; if not, write to the
  15. * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16. * Boston, MA 021110-1307, USA.
  17. */
  18. #include <linux/fs.h>
  19. #include <linux/blkdev.h>
  20. #include <linux/scatterlist.h>
  21. #include <linux/swap.h>
  22. #include <linux/radix-tree.h>
  23. #include <linux/writeback.h>
  24. #include <linux/buffer_head.h>
  25. #include <linux/workqueue.h>
  26. #include <linux/kthread.h>
  27. #include <linux/freezer.h>
  28. #include <linux/crc32c.h>
  29. #include <linux/slab.h>
  30. #include <linux/migrate.h>
  31. #include <linux/ratelimit.h>
  32. #include <linux/uuid.h>
  33. #include <asm/unaligned.h>
  34. #include "compat.h"
  35. #include "ctree.h"
  36. #include "disk-io.h"
  37. #include "transaction.h"
  38. #include "btrfs_inode.h"
  39. #include "volumes.h"
  40. #include "print-tree.h"
  41. #include "async-thread.h"
  42. #include "locking.h"
  43. #include "tree-log.h"
  44. #include "free-space-cache.h"
  45. #include "inode-map.h"
  46. #include "check-integrity.h"
  47. #include "rcu-string.h"
  48. #include "dev-replace.h"
  49. #include "raid56.h"
  50. #ifdef CONFIG_X86
  51. #include <asm/cpufeature.h>
  52. #endif
  53. static struct extent_io_ops btree_extent_io_ops;
  54. static void end_workqueue_fn(struct btrfs_work *work);
  55. static void free_fs_root(struct btrfs_root *root);
  56. static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
  57. int read_only);
  58. static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
  59. struct btrfs_root *root);
  60. static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
  61. static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
  62. struct btrfs_root *root);
  63. static void btrfs_evict_pending_snapshots(struct btrfs_transaction *t);
  64. static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
  65. static int btrfs_destroy_marked_extents(struct btrfs_root *root,
  66. struct extent_io_tree *dirty_pages,
  67. int mark);
  68. static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
  69. struct extent_io_tree *pinned_extents);
  70. static int btrfs_cleanup_transaction(struct btrfs_root *root);
  71. static void btrfs_error_commit_super(struct btrfs_root *root);
  72. /*
  73. * end_io_wq structs are used to do processing in task context when an IO is
  74. * complete. This is used during reads to verify checksums, and it is used
  75. * by writes to insert metadata for new file extents after IO is complete.
  76. */
  77. struct end_io_wq {
  78. struct bio *bio;
  79. bio_end_io_t *end_io;
  80. void *private;
  81. struct btrfs_fs_info *info;
  82. int error;
  83. int metadata;
  84. struct list_head list;
  85. struct btrfs_work work;
  86. };
  87. /*
  88. * async submit bios are used to offload expensive checksumming
  89. * onto the worker threads. They checksum file and metadata bios
  90. * just before they are sent down the IO stack.
  91. */
  92. struct async_submit_bio {
  93. struct inode *inode;
  94. struct bio *bio;
  95. struct list_head list;
  96. extent_submit_bio_hook_t *submit_bio_start;
  97. extent_submit_bio_hook_t *submit_bio_done;
  98. int rw;
  99. int mirror_num;
  100. unsigned long bio_flags;
  101. /*
  102. * bio_offset is optional, can be used if the pages in the bio
  103. * can't tell us where in the file the bio should go
  104. */
  105. u64 bio_offset;
  106. struct btrfs_work work;
  107. int error;
  108. };
  109. /*
  110. * Lockdep class keys for extent_buffer->lock's in this root. For a given
  111. * eb, the lockdep key is determined by the btrfs_root it belongs to and
  112. * the level the eb occupies in the tree.
  113. *
  114. * Different roots are used for different purposes and may nest inside each
  115. * other and they require separate keysets. As lockdep keys should be
  116. * static, assign keysets according to the purpose of the root as indicated
  117. * by btrfs_root->objectid. This ensures that all special purpose roots
  118. * have separate keysets.
  119. *
  120. * Lock-nesting across peer nodes is always done with the immediate parent
  121. * node locked thus preventing deadlock. As lockdep doesn't know this, use
  122. * subclass to avoid triggering lockdep warning in such cases.
  123. *
  124. * The key is set by the readpage_end_io_hook after the buffer has passed
  125. * csum validation but before the pages are unlocked. It is also set by
  126. * btrfs_init_new_buffer on freshly allocated blocks.
  127. *
  128. * We also add a check to make sure the highest level of the tree is the
  129. * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
  130. * needs update as well.
  131. */
  132. #ifdef CONFIG_DEBUG_LOCK_ALLOC
  133. # if BTRFS_MAX_LEVEL != 8
  134. # error
  135. # endif
  136. static struct btrfs_lockdep_keyset {
  137. u64 id; /* root objectid */
  138. const char *name_stem; /* lock name stem */
  139. char names[BTRFS_MAX_LEVEL + 1][20];
  140. struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
  141. } btrfs_lockdep_keysets[] = {
  142. { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
  143. { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
  144. { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
  145. { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
  146. { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
  147. { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
  148. { .id = BTRFS_ORPHAN_OBJECTID, .name_stem = "orphan" },
  149. { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
  150. { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
  151. { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
  152. { .id = 0, .name_stem = "tree" },
  153. };
  154. void __init btrfs_init_lockdep(void)
  155. {
  156. int i, j;
  157. /* initialize lockdep class names */
  158. for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
  159. struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
  160. for (j = 0; j < ARRAY_SIZE(ks->names); j++)
  161. snprintf(ks->names[j], sizeof(ks->names[j]),
  162. "btrfs-%s-%02d", ks->name_stem, j);
  163. }
  164. }
  165. void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
  166. int level)
  167. {
  168. struct btrfs_lockdep_keyset *ks;
  169. BUG_ON(level >= ARRAY_SIZE(ks->keys));
  170. /* find the matching keyset, id 0 is the default entry */
  171. for (ks = btrfs_lockdep_keysets; ks->id; ks++)
  172. if (ks->id == objectid)
  173. break;
  174. lockdep_set_class_and_name(&eb->lock,
  175. &ks->keys[level], ks->names[level]);
  176. }
  177. #endif
  178. /*
  179. * extents on the btree inode are pretty simple, there's one extent
  180. * that covers the entire device
  181. */
  182. static struct extent_map *btree_get_extent(struct inode *inode,
  183. struct page *page, size_t pg_offset, u64 start, u64 len,
  184. int create)
  185. {
  186. struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
  187. struct extent_map *em;
  188. int ret;
  189. read_lock(&em_tree->lock);
  190. em = lookup_extent_mapping(em_tree, start, len);
  191. if (em) {
  192. em->bdev =
  193. BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
  194. read_unlock(&em_tree->lock);
  195. goto out;
  196. }
  197. read_unlock(&em_tree->lock);
  198. em = alloc_extent_map();
  199. if (!em) {
  200. em = ERR_PTR(-ENOMEM);
  201. goto out;
  202. }
  203. em->start = 0;
  204. em->len = (u64)-1;
  205. em->block_len = (u64)-1;
  206. em->block_start = 0;
  207. em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
  208. write_lock(&em_tree->lock);
  209. ret = add_extent_mapping(em_tree, em, 0);
  210. if (ret == -EEXIST) {
  211. free_extent_map(em);
  212. em = lookup_extent_mapping(em_tree, start, len);
  213. if (!em)
  214. em = ERR_PTR(-EIO);
  215. } else if (ret) {
  216. free_extent_map(em);
  217. em = ERR_PTR(ret);
  218. }
  219. write_unlock(&em_tree->lock);
  220. out:
  221. return em;
  222. }
  223. u32 btrfs_csum_data(char *data, u32 seed, size_t len)
  224. {
  225. return crc32c(seed, data, len);
  226. }
  227. void btrfs_csum_final(u32 crc, char *result)
  228. {
  229. put_unaligned_le32(~crc, result);
  230. }
  231. /*
  232. * compute the csum for a btree block, and either verify it or write it
  233. * into the csum field of the block.
  234. */
  235. static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
  236. int verify)
  237. {
  238. u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
  239. char *result = NULL;
  240. unsigned long len;
  241. unsigned long cur_len;
  242. unsigned long offset = BTRFS_CSUM_SIZE;
  243. char *kaddr;
  244. unsigned long map_start;
  245. unsigned long map_len;
  246. int err;
  247. u32 crc = ~(u32)0;
  248. unsigned long inline_result;
  249. len = buf->len - offset;
  250. while (len > 0) {
  251. err = map_private_extent_buffer(buf, offset, 32,
  252. &kaddr, &map_start, &map_len);
  253. if (err)
  254. return 1;
  255. cur_len = min(len, map_len - (offset - map_start));
  256. crc = btrfs_csum_data(kaddr + offset - map_start,
  257. crc, cur_len);
  258. len -= cur_len;
  259. offset += cur_len;
  260. }
  261. if (csum_size > sizeof(inline_result)) {
  262. result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
  263. if (!result)
  264. return 1;
  265. } else {
  266. result = (char *)&inline_result;
  267. }
  268. btrfs_csum_final(crc, result);
  269. if (verify) {
  270. if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
  271. u32 val;
  272. u32 found = 0;
  273. memcpy(&found, result, csum_size);
  274. read_extent_buffer(buf, &val, 0, csum_size);
  275. printk_ratelimited(KERN_INFO "btrfs: %s checksum verify "
  276. "failed on %llu wanted %X found %X "
  277. "level %d\n",
  278. root->fs_info->sb->s_id,
  279. (unsigned long long)buf->start, val, found,
  280. btrfs_header_level(buf));
  281. if (result != (char *)&inline_result)
  282. kfree(result);
  283. return 1;
  284. }
  285. } else {
  286. write_extent_buffer(buf, result, 0, csum_size);
  287. }
  288. if (result != (char *)&inline_result)
  289. kfree(result);
  290. return 0;
  291. }
  292. /*
  293. * we can't consider a given block up to date unless the transid of the
  294. * block matches the transid in the parent node's pointer. This is how we
  295. * detect blocks that either didn't get written at all or got written
  296. * in the wrong place.
  297. */
  298. static int verify_parent_transid(struct extent_io_tree *io_tree,
  299. struct extent_buffer *eb, u64 parent_transid,
  300. int atomic)
  301. {
  302. struct extent_state *cached_state = NULL;
  303. int ret;
  304. if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
  305. return 0;
  306. if (atomic)
  307. return -EAGAIN;
  308. lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
  309. 0, &cached_state);
  310. if (extent_buffer_uptodate(eb) &&
  311. btrfs_header_generation(eb) == parent_transid) {
  312. ret = 0;
  313. goto out;
  314. }
  315. printk_ratelimited("parent transid verify failed on %llu wanted %llu "
  316. "found %llu\n",
  317. (unsigned long long)eb->start,
  318. (unsigned long long)parent_transid,
  319. (unsigned long long)btrfs_header_generation(eb));
  320. ret = 1;
  321. clear_extent_buffer_uptodate(eb);
  322. out:
  323. unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
  324. &cached_state, GFP_NOFS);
  325. return ret;
  326. }
  327. /*
  328. * helper to read a given tree block, doing retries as required when
  329. * the checksums don't match and we have alternate mirrors to try.
  330. */
  331. static int btree_read_extent_buffer_pages(struct btrfs_root *root,
  332. struct extent_buffer *eb,
  333. u64 start, u64 parent_transid)
  334. {
  335. struct extent_io_tree *io_tree;
  336. int failed = 0;
  337. int ret;
  338. int num_copies = 0;
  339. int mirror_num = 0;
  340. int failed_mirror = 0;
  341. clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
  342. io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
  343. while (1) {
  344. ret = read_extent_buffer_pages(io_tree, eb, start,
  345. WAIT_COMPLETE,
  346. btree_get_extent, mirror_num);
  347. if (!ret) {
  348. if (!verify_parent_transid(io_tree, eb,
  349. parent_transid, 0))
  350. break;
  351. else
  352. ret = -EIO;
  353. }
  354. /*
  355. * This buffer's crc is fine, but its contents are corrupted, so
  356. * there is no reason to read the other copies, they won't be
  357. * any less wrong.
  358. */
  359. if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
  360. break;
  361. num_copies = btrfs_num_copies(root->fs_info,
  362. eb->start, eb->len);
  363. if (num_copies == 1)
  364. break;
  365. if (!failed_mirror) {
  366. failed = 1;
  367. failed_mirror = eb->read_mirror;
  368. }
  369. mirror_num++;
  370. if (mirror_num == failed_mirror)
  371. mirror_num++;
  372. if (mirror_num > num_copies)
  373. break;
  374. }
  375. if (failed && !ret && failed_mirror)
  376. repair_eb_io_failure(root, eb, failed_mirror);
  377. return ret;
  378. }
  379. /*
  380. * checksum a dirty tree block before IO. This has extra checks to make sure
  381. * we only fill in the checksum field in the first page of a multi-page block
  382. */
  383. static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
  384. {
  385. struct extent_io_tree *tree;
  386. u64 start = page_offset(page);
  387. u64 found_start;
  388. struct extent_buffer *eb;
  389. tree = &BTRFS_I(page->mapping->host)->io_tree;
  390. eb = (struct extent_buffer *)page->private;
  391. if (page != eb->pages[0])
  392. return 0;
  393. found_start = btrfs_header_bytenr(eb);
  394. if (found_start != start) {
  395. WARN_ON(1);
  396. return 0;
  397. }
  398. if (!PageUptodate(page)) {
  399. WARN_ON(1);
  400. return 0;
  401. }
  402. csum_tree_block(root, eb, 0);
  403. return 0;
  404. }
  405. static int check_tree_block_fsid(struct btrfs_root *root,
  406. struct extent_buffer *eb)
  407. {
  408. struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
  409. u8 fsid[BTRFS_UUID_SIZE];
  410. int ret = 1;
  411. read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
  412. BTRFS_FSID_SIZE);
  413. while (fs_devices) {
  414. if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
  415. ret = 0;
  416. break;
  417. }
  418. fs_devices = fs_devices->seed;
  419. }
  420. return ret;
  421. }
  422. #define CORRUPT(reason, eb, root, slot) \
  423. printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
  424. "root=%llu, slot=%d\n", reason, \
  425. (unsigned long long)btrfs_header_bytenr(eb), \
  426. (unsigned long long)root->objectid, slot)
  427. static noinline int check_leaf(struct btrfs_root *root,
  428. struct extent_buffer *leaf)
  429. {
  430. struct btrfs_key key;
  431. struct btrfs_key leaf_key;
  432. u32 nritems = btrfs_header_nritems(leaf);
  433. int slot;
  434. if (nritems == 0)
  435. return 0;
  436. /* Check the 0 item */
  437. if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
  438. BTRFS_LEAF_DATA_SIZE(root)) {
  439. CORRUPT("invalid item offset size pair", leaf, root, 0);
  440. return -EIO;
  441. }
  442. /*
  443. * Check to make sure each items keys are in the correct order and their
  444. * offsets make sense. We only have to loop through nritems-1 because
  445. * we check the current slot against the next slot, which verifies the
  446. * next slot's offset+size makes sense and that the current's slot
  447. * offset is correct.
  448. */
  449. for (slot = 0; slot < nritems - 1; slot++) {
  450. btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
  451. btrfs_item_key_to_cpu(leaf, &key, slot + 1);
  452. /* Make sure the keys are in the right order */
  453. if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
  454. CORRUPT("bad key order", leaf, root, slot);
  455. return -EIO;
  456. }
  457. /*
  458. * Make sure the offset and ends are right, remember that the
  459. * item data starts at the end of the leaf and grows towards the
  460. * front.
  461. */
  462. if (btrfs_item_offset_nr(leaf, slot) !=
  463. btrfs_item_end_nr(leaf, slot + 1)) {
  464. CORRUPT("slot offset bad", leaf, root, slot);
  465. return -EIO;
  466. }
  467. /*
  468. * Check to make sure that we don't point outside of the leaf,
  469. * just incase all the items are consistent to eachother, but
  470. * all point outside of the leaf.
  471. */
  472. if (btrfs_item_end_nr(leaf, slot) >
  473. BTRFS_LEAF_DATA_SIZE(root)) {
  474. CORRUPT("slot end outside of leaf", leaf, root, slot);
  475. return -EIO;
  476. }
  477. }
  478. return 0;
  479. }
  480. static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
  481. struct extent_state *state, int mirror)
  482. {
  483. struct extent_io_tree *tree;
  484. u64 found_start;
  485. int found_level;
  486. struct extent_buffer *eb;
  487. struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
  488. int ret = 0;
  489. int reads_done;
  490. if (!page->private)
  491. goto out;
  492. tree = &BTRFS_I(page->mapping->host)->io_tree;
  493. eb = (struct extent_buffer *)page->private;
  494. /* the pending IO might have been the only thing that kept this buffer
  495. * in memory. Make sure we have a ref for all this other checks
  496. */
  497. extent_buffer_get(eb);
  498. reads_done = atomic_dec_and_test(&eb->io_pages);
  499. if (!reads_done)
  500. goto err;
  501. eb->read_mirror = mirror;
  502. if (test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
  503. ret = -EIO;
  504. goto err;
  505. }
  506. found_start = btrfs_header_bytenr(eb);
  507. if (found_start != eb->start) {
  508. printk_ratelimited(KERN_INFO "btrfs bad tree block start "
  509. "%llu %llu\n",
  510. (unsigned long long)found_start,
  511. (unsigned long long)eb->start);
  512. ret = -EIO;
  513. goto err;
  514. }
  515. if (check_tree_block_fsid(root, eb)) {
  516. printk_ratelimited(KERN_INFO "btrfs bad fsid on block %llu\n",
  517. (unsigned long long)eb->start);
  518. ret = -EIO;
  519. goto err;
  520. }
  521. found_level = btrfs_header_level(eb);
  522. if (found_level >= BTRFS_MAX_LEVEL) {
  523. btrfs_info(root->fs_info, "bad tree block level %d\n",
  524. (int)btrfs_header_level(eb));
  525. ret = -EIO;
  526. goto err;
  527. }
  528. btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
  529. eb, found_level);
  530. ret = csum_tree_block(root, eb, 1);
  531. if (ret) {
  532. ret = -EIO;
  533. goto err;
  534. }
  535. /*
  536. * If this is a leaf block and it is corrupt, set the corrupt bit so
  537. * that we don't try and read the other copies of this block, just
  538. * return -EIO.
  539. */
  540. if (found_level == 0 && check_leaf(root, eb)) {
  541. set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
  542. ret = -EIO;
  543. }
  544. if (!ret)
  545. set_extent_buffer_uptodate(eb);
  546. err:
  547. if (reads_done &&
  548. test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
  549. btree_readahead_hook(root, eb, eb->start, ret);
  550. if (ret) {
  551. /*
  552. * our io error hook is going to dec the io pages
  553. * again, we have to make sure it has something
  554. * to decrement
  555. */
  556. atomic_inc(&eb->io_pages);
  557. clear_extent_buffer_uptodate(eb);
  558. }
  559. free_extent_buffer(eb);
  560. out:
  561. return ret;
  562. }
  563. static int btree_io_failed_hook(struct page *page, int failed_mirror)
  564. {
  565. struct extent_buffer *eb;
  566. struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
  567. eb = (struct extent_buffer *)page->private;
  568. set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
  569. eb->read_mirror = failed_mirror;
  570. atomic_dec(&eb->io_pages);
  571. if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
  572. btree_readahead_hook(root, eb, eb->start, -EIO);
  573. return -EIO; /* we fixed nothing */
  574. }
  575. static void end_workqueue_bio(struct bio *bio, int err)
  576. {
  577. struct end_io_wq *end_io_wq = bio->bi_private;
  578. struct btrfs_fs_info *fs_info;
  579. fs_info = end_io_wq->info;
  580. end_io_wq->error = err;
  581. end_io_wq->work.func = end_workqueue_fn;
  582. end_io_wq->work.flags = 0;
  583. if (bio->bi_rw & REQ_WRITE) {
  584. if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
  585. btrfs_queue_worker(&fs_info->endio_meta_write_workers,
  586. &end_io_wq->work);
  587. else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
  588. btrfs_queue_worker(&fs_info->endio_freespace_worker,
  589. &end_io_wq->work);
  590. else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
  591. btrfs_queue_worker(&fs_info->endio_raid56_workers,
  592. &end_io_wq->work);
  593. else
  594. btrfs_queue_worker(&fs_info->endio_write_workers,
  595. &end_io_wq->work);
  596. } else {
  597. if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
  598. btrfs_queue_worker(&fs_info->endio_raid56_workers,
  599. &end_io_wq->work);
  600. else if (end_io_wq->metadata)
  601. btrfs_queue_worker(&fs_info->endio_meta_workers,
  602. &end_io_wq->work);
  603. else
  604. btrfs_queue_worker(&fs_info->endio_workers,
  605. &end_io_wq->work);
  606. }
  607. }
  608. /*
  609. * For the metadata arg you want
  610. *
  611. * 0 - if data
  612. * 1 - if normal metadta
  613. * 2 - if writing to the free space cache area
  614. * 3 - raid parity work
  615. */
  616. int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
  617. int metadata)
  618. {
  619. struct end_io_wq *end_io_wq;
  620. end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
  621. if (!end_io_wq)
  622. return -ENOMEM;
  623. end_io_wq->private = bio->bi_private;
  624. end_io_wq->end_io = bio->bi_end_io;
  625. end_io_wq->info = info;
  626. end_io_wq->error = 0;
  627. end_io_wq->bio = bio;
  628. end_io_wq->metadata = metadata;
  629. bio->bi_private = end_io_wq;
  630. bio->bi_end_io = end_workqueue_bio;
  631. return 0;
  632. }
  633. unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
  634. {
  635. unsigned long limit = min_t(unsigned long,
  636. info->workers.max_workers,
  637. info->fs_devices->open_devices);
  638. return 256 * limit;
  639. }
  640. static void run_one_async_start(struct btrfs_work *work)
  641. {
  642. struct async_submit_bio *async;
  643. int ret;
  644. async = container_of(work, struct async_submit_bio, work);
  645. ret = async->submit_bio_start(async->inode, async->rw, async->bio,
  646. async->mirror_num, async->bio_flags,
  647. async->bio_offset);
  648. if (ret)
  649. async->error = ret;
  650. }
  651. static void run_one_async_done(struct btrfs_work *work)
  652. {
  653. struct btrfs_fs_info *fs_info;
  654. struct async_submit_bio *async;
  655. int limit;
  656. async = container_of(work, struct async_submit_bio, work);
  657. fs_info = BTRFS_I(async->inode)->root->fs_info;
  658. limit = btrfs_async_submit_limit(fs_info);
  659. limit = limit * 2 / 3;
  660. if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
  661. waitqueue_active(&fs_info->async_submit_wait))
  662. wake_up(&fs_info->async_submit_wait);
  663. /* If an error occured we just want to clean up the bio and move on */
  664. if (async->error) {
  665. bio_endio(async->bio, async->error);
  666. return;
  667. }
  668. async->submit_bio_done(async->inode, async->rw, async->bio,
  669. async->mirror_num, async->bio_flags,
  670. async->bio_offset);
  671. }
  672. static void run_one_async_free(struct btrfs_work *work)
  673. {
  674. struct async_submit_bio *async;
  675. async = container_of(work, struct async_submit_bio, work);
  676. kfree(async);
  677. }
  678. int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
  679. int rw, struct bio *bio, int mirror_num,
  680. unsigned long bio_flags,
  681. u64 bio_offset,
  682. extent_submit_bio_hook_t *submit_bio_start,
  683. extent_submit_bio_hook_t *submit_bio_done)
  684. {
  685. struct async_submit_bio *async;
  686. async = kmalloc(sizeof(*async), GFP_NOFS);
  687. if (!async)
  688. return -ENOMEM;
  689. async->inode = inode;
  690. async->rw = rw;
  691. async->bio = bio;
  692. async->mirror_num = mirror_num;
  693. async->submit_bio_start = submit_bio_start;
  694. async->submit_bio_done = submit_bio_done;
  695. async->work.func = run_one_async_start;
  696. async->work.ordered_func = run_one_async_done;
  697. async->work.ordered_free = run_one_async_free;
  698. async->work.flags = 0;
  699. async->bio_flags = bio_flags;
  700. async->bio_offset = bio_offset;
  701. async->error = 0;
  702. atomic_inc(&fs_info->nr_async_submits);
  703. if (rw & REQ_SYNC)
  704. btrfs_set_work_high_prio(&async->work);
  705. btrfs_queue_worker(&fs_info->workers, &async->work);
  706. while (atomic_read(&fs_info->async_submit_draining) &&
  707. atomic_read(&fs_info->nr_async_submits)) {
  708. wait_event(fs_info->async_submit_wait,
  709. (atomic_read(&fs_info->nr_async_submits) == 0));
  710. }
  711. return 0;
  712. }
  713. static int btree_csum_one_bio(struct bio *bio)
  714. {
  715. struct bio_vec *bvec = bio->bi_io_vec;
  716. int bio_index = 0;
  717. struct btrfs_root *root;
  718. int ret = 0;
  719. WARN_ON(bio->bi_vcnt <= 0);
  720. while (bio_index < bio->bi_vcnt) {
  721. root = BTRFS_I(bvec->bv_page->mapping->host)->root;
  722. ret = csum_dirty_buffer(root, bvec->bv_page);
  723. if (ret)
  724. break;
  725. bio_index++;
  726. bvec++;
  727. }
  728. return ret;
  729. }
  730. static int __btree_submit_bio_start(struct inode *inode, int rw,
  731. struct bio *bio, int mirror_num,
  732. unsigned long bio_flags,
  733. u64 bio_offset)
  734. {
  735. /*
  736. * when we're called for a write, we're already in the async
  737. * submission context. Just jump into btrfs_map_bio
  738. */
  739. return btree_csum_one_bio(bio);
  740. }
  741. static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
  742. int mirror_num, unsigned long bio_flags,
  743. u64 bio_offset)
  744. {
  745. int ret;
  746. /*
  747. * when we're called for a write, we're already in the async
  748. * submission context. Just jump into btrfs_map_bio
  749. */
  750. ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
  751. if (ret)
  752. bio_endio(bio, ret);
  753. return ret;
  754. }
  755. static int check_async_write(struct inode *inode, unsigned long bio_flags)
  756. {
  757. if (bio_flags & EXTENT_BIO_TREE_LOG)
  758. return 0;
  759. #ifdef CONFIG_X86
  760. if (cpu_has_xmm4_2)
  761. return 0;
  762. #endif
  763. return 1;
  764. }
  765. static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
  766. int mirror_num, unsigned long bio_flags,
  767. u64 bio_offset)
  768. {
  769. int async = check_async_write(inode, bio_flags);
  770. int ret;
  771. if (!(rw & REQ_WRITE)) {
  772. /*
  773. * called for a read, do the setup so that checksum validation
  774. * can happen in the async kernel threads
  775. */
  776. ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
  777. bio, 1);
  778. if (ret)
  779. goto out_w_error;
  780. ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
  781. mirror_num, 0);
  782. } else if (!async) {
  783. ret = btree_csum_one_bio(bio);
  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 {
  789. /*
  790. * kthread helpers are used to submit writes so that
  791. * checksumming can happen in parallel across all CPUs
  792. */
  793. ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
  794. inode, rw, bio, mirror_num, 0,
  795. bio_offset,
  796. __btree_submit_bio_start,
  797. __btree_submit_bio_done);
  798. }
  799. if (ret) {
  800. out_w_error:
  801. bio_endio(bio, ret);
  802. }
  803. return ret;
  804. }
  805. #ifdef CONFIG_MIGRATION
  806. static int btree_migratepage(struct address_space *mapping,
  807. struct page *newpage, struct page *page,
  808. enum migrate_mode mode)
  809. {
  810. /*
  811. * we can't safely write a btree page from here,
  812. * we haven't done the locking hook
  813. */
  814. if (PageDirty(page))
  815. return -EAGAIN;
  816. /*
  817. * Buffers may be managed in a filesystem specific way.
  818. * We must have no buffers or drop them.
  819. */
  820. if (page_has_private(page) &&
  821. !try_to_release_page(page, GFP_KERNEL))
  822. return -EAGAIN;
  823. return migrate_page(mapping, newpage, page, mode);
  824. }
  825. #endif
  826. static int btree_writepages(struct address_space *mapping,
  827. struct writeback_control *wbc)
  828. {
  829. struct extent_io_tree *tree;
  830. struct btrfs_fs_info *fs_info;
  831. int ret;
  832. tree = &BTRFS_I(mapping->host)->io_tree;
  833. if (wbc->sync_mode == WB_SYNC_NONE) {
  834. if (wbc->for_kupdate)
  835. return 0;
  836. fs_info = BTRFS_I(mapping->host)->root->fs_info;
  837. /* this is a bit racy, but that's ok */
  838. ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
  839. BTRFS_DIRTY_METADATA_THRESH);
  840. if (ret < 0)
  841. return 0;
  842. }
  843. return btree_write_cache_pages(mapping, wbc);
  844. }
  845. static int btree_readpage(struct file *file, struct page *page)
  846. {
  847. struct extent_io_tree *tree;
  848. tree = &BTRFS_I(page->mapping->host)->io_tree;
  849. return extent_read_full_page(tree, page, btree_get_extent, 0);
  850. }
  851. static int btree_releasepage(struct page *page, gfp_t gfp_flags)
  852. {
  853. if (PageWriteback(page) || PageDirty(page))
  854. return 0;
  855. return try_release_extent_buffer(page);
  856. }
  857. static void btree_invalidatepage(struct page *page, unsigned long offset)
  858. {
  859. struct extent_io_tree *tree;
  860. tree = &BTRFS_I(page->mapping->host)->io_tree;
  861. extent_invalidatepage(tree, page, offset);
  862. btree_releasepage(page, GFP_NOFS);
  863. if (PagePrivate(page)) {
  864. printk(KERN_WARNING "btrfs warning page private not zero "
  865. "on page %llu\n", (unsigned long long)page_offset(page));
  866. ClearPagePrivate(page);
  867. set_page_private(page, 0);
  868. page_cache_release(page);
  869. }
  870. }
  871. static int btree_set_page_dirty(struct page *page)
  872. {
  873. #ifdef DEBUG
  874. struct extent_buffer *eb;
  875. BUG_ON(!PagePrivate(page));
  876. eb = (struct extent_buffer *)page->private;
  877. BUG_ON(!eb);
  878. BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
  879. BUG_ON(!atomic_read(&eb->refs));
  880. btrfs_assert_tree_locked(eb);
  881. #endif
  882. return __set_page_dirty_nobuffers(page);
  883. }
  884. static const struct address_space_operations btree_aops = {
  885. .readpage = btree_readpage,
  886. .writepages = btree_writepages,
  887. .releasepage = btree_releasepage,
  888. .invalidatepage = btree_invalidatepage,
  889. #ifdef CONFIG_MIGRATION
  890. .migratepage = btree_migratepage,
  891. #endif
  892. .set_page_dirty = btree_set_page_dirty,
  893. };
  894. int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
  895. u64 parent_transid)
  896. {
  897. struct extent_buffer *buf = NULL;
  898. struct inode *btree_inode = root->fs_info->btree_inode;
  899. int ret = 0;
  900. buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
  901. if (!buf)
  902. return 0;
  903. read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
  904. buf, 0, WAIT_NONE, btree_get_extent, 0);
  905. free_extent_buffer(buf);
  906. return ret;
  907. }
  908. int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
  909. int mirror_num, struct extent_buffer **eb)
  910. {
  911. struct extent_buffer *buf = NULL;
  912. struct inode *btree_inode = root->fs_info->btree_inode;
  913. struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
  914. int ret;
  915. buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
  916. if (!buf)
  917. return 0;
  918. set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
  919. ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
  920. btree_get_extent, mirror_num);
  921. if (ret) {
  922. free_extent_buffer(buf);
  923. return ret;
  924. }
  925. if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
  926. free_extent_buffer(buf);
  927. return -EIO;
  928. } else if (extent_buffer_uptodate(buf)) {
  929. *eb = buf;
  930. } else {
  931. free_extent_buffer(buf);
  932. }
  933. return 0;
  934. }
  935. struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
  936. u64 bytenr, u32 blocksize)
  937. {
  938. struct inode *btree_inode = root->fs_info->btree_inode;
  939. struct extent_buffer *eb;
  940. eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
  941. bytenr, blocksize);
  942. return eb;
  943. }
  944. struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
  945. u64 bytenr, u32 blocksize)
  946. {
  947. struct inode *btree_inode = root->fs_info->btree_inode;
  948. struct extent_buffer *eb;
  949. eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
  950. bytenr, blocksize);
  951. return eb;
  952. }
  953. int btrfs_write_tree_block(struct extent_buffer *buf)
  954. {
  955. return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
  956. buf->start + buf->len - 1);
  957. }
  958. int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
  959. {
  960. return filemap_fdatawait_range(buf->pages[0]->mapping,
  961. buf->start, buf->start + buf->len - 1);
  962. }
  963. struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
  964. u32 blocksize, u64 parent_transid)
  965. {
  966. struct extent_buffer *buf = NULL;
  967. int ret;
  968. buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
  969. if (!buf)
  970. return NULL;
  971. ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
  972. return buf;
  973. }
  974. void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
  975. struct extent_buffer *buf)
  976. {
  977. struct btrfs_fs_info *fs_info = root->fs_info;
  978. if (btrfs_header_generation(buf) ==
  979. fs_info->running_transaction->transid) {
  980. btrfs_assert_tree_locked(buf);
  981. if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
  982. __percpu_counter_add(&fs_info->dirty_metadata_bytes,
  983. -buf->len,
  984. fs_info->dirty_metadata_batch);
  985. /* ugh, clear_extent_buffer_dirty needs to lock the page */
  986. btrfs_set_lock_blocking(buf);
  987. clear_extent_buffer_dirty(buf);
  988. }
  989. }
  990. }
  991. static void __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
  992. u32 stripesize, struct btrfs_root *root,
  993. struct btrfs_fs_info *fs_info,
  994. u64 objectid)
  995. {
  996. root->node = NULL;
  997. root->commit_root = NULL;
  998. root->sectorsize = sectorsize;
  999. root->nodesize = nodesize;
  1000. root->leafsize = leafsize;
  1001. root->stripesize = stripesize;
  1002. root->ref_cows = 0;
  1003. root->track_dirty = 0;
  1004. root->in_radix = 0;
  1005. root->orphan_item_inserted = 0;
  1006. root->orphan_cleanup_state = 0;
  1007. root->objectid = objectid;
  1008. root->last_trans = 0;
  1009. root->highest_objectid = 0;
  1010. root->name = NULL;
  1011. root->inode_tree = RB_ROOT;
  1012. INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
  1013. root->block_rsv = NULL;
  1014. root->orphan_block_rsv = NULL;
  1015. INIT_LIST_HEAD(&root->dirty_list);
  1016. INIT_LIST_HEAD(&root->root_list);
  1017. INIT_LIST_HEAD(&root->logged_list[0]);
  1018. INIT_LIST_HEAD(&root->logged_list[1]);
  1019. spin_lock_init(&root->orphan_lock);
  1020. spin_lock_init(&root->inode_lock);
  1021. spin_lock_init(&root->accounting_lock);
  1022. spin_lock_init(&root->log_extents_lock[0]);
  1023. spin_lock_init(&root->log_extents_lock[1]);
  1024. mutex_init(&root->objectid_mutex);
  1025. mutex_init(&root->log_mutex);
  1026. init_waitqueue_head(&root->log_writer_wait);
  1027. init_waitqueue_head(&root->log_commit_wait[0]);
  1028. init_waitqueue_head(&root->log_commit_wait[1]);
  1029. atomic_set(&root->log_commit[0], 0);
  1030. atomic_set(&root->log_commit[1], 0);
  1031. atomic_set(&root->log_writers, 0);
  1032. atomic_set(&root->log_batch, 0);
  1033. atomic_set(&root->orphan_inodes, 0);
  1034. root->log_transid = 0;
  1035. root->last_log_commit = 0;
  1036. extent_io_tree_init(&root->dirty_log_pages,
  1037. fs_info->btree_inode->i_mapping);
  1038. memset(&root->root_key, 0, sizeof(root->root_key));
  1039. memset(&root->root_item, 0, sizeof(root->root_item));
  1040. memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
  1041. memset(&root->root_kobj, 0, sizeof(root->root_kobj));
  1042. root->defrag_trans_start = fs_info->generation;
  1043. init_completion(&root->kobj_unregister);
  1044. root->defrag_running = 0;
  1045. root->root_key.objectid = objectid;
  1046. root->anon_dev = 0;
  1047. spin_lock_init(&root->root_item_lock);
  1048. }
  1049. static int __must_check find_and_setup_root(struct btrfs_root *tree_root,
  1050. struct btrfs_fs_info *fs_info,
  1051. u64 objectid,
  1052. struct btrfs_root *root)
  1053. {
  1054. int ret;
  1055. u32 blocksize;
  1056. u64 generation;
  1057. __setup_root(tree_root->nodesize, tree_root->leafsize,
  1058. tree_root->sectorsize, tree_root->stripesize,
  1059. root, fs_info, objectid);
  1060. ret = btrfs_find_last_root(tree_root, objectid,
  1061. &root->root_item, &root->root_key);
  1062. if (ret > 0)
  1063. return -ENOENT;
  1064. else if (ret < 0)
  1065. return ret;
  1066. generation = btrfs_root_generation(&root->root_item);
  1067. blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
  1068. root->commit_root = NULL;
  1069. root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
  1070. blocksize, generation);
  1071. if (!root->node || !btrfs_buffer_uptodate(root->node, generation, 0)) {
  1072. free_extent_buffer(root->node);
  1073. root->node = NULL;
  1074. return -EIO;
  1075. }
  1076. root->commit_root = btrfs_root_node(root);
  1077. return 0;
  1078. }
  1079. static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
  1080. {
  1081. struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
  1082. if (root)
  1083. root->fs_info = fs_info;
  1084. return root;
  1085. }
  1086. struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
  1087. struct btrfs_fs_info *fs_info,
  1088. u64 objectid)
  1089. {
  1090. struct extent_buffer *leaf;
  1091. struct btrfs_root *tree_root = fs_info->tree_root;
  1092. struct btrfs_root *root;
  1093. struct btrfs_key key;
  1094. int ret = 0;
  1095. u64 bytenr;
  1096. uuid_le uuid;
  1097. root = btrfs_alloc_root(fs_info);
  1098. if (!root)
  1099. return ERR_PTR(-ENOMEM);
  1100. __setup_root(tree_root->nodesize, tree_root->leafsize,
  1101. tree_root->sectorsize, tree_root->stripesize,
  1102. root, fs_info, objectid);
  1103. root->root_key.objectid = objectid;
  1104. root->root_key.type = BTRFS_ROOT_ITEM_KEY;
  1105. root->root_key.offset = 0;
  1106. leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
  1107. 0, objectid, NULL, 0, 0, 0);
  1108. if (IS_ERR(leaf)) {
  1109. ret = PTR_ERR(leaf);
  1110. leaf = NULL;
  1111. goto fail;
  1112. }
  1113. bytenr = leaf->start;
  1114. memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
  1115. btrfs_set_header_bytenr(leaf, leaf->start);
  1116. btrfs_set_header_generation(leaf, trans->transid);
  1117. btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
  1118. btrfs_set_header_owner(leaf, objectid);
  1119. root->node = leaf;
  1120. write_extent_buffer(leaf, fs_info->fsid,
  1121. (unsigned long)btrfs_header_fsid(leaf),
  1122. BTRFS_FSID_SIZE);
  1123. write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
  1124. (unsigned long)btrfs_header_chunk_tree_uuid(leaf),
  1125. BTRFS_UUID_SIZE);
  1126. btrfs_mark_buffer_dirty(leaf);
  1127. root->commit_root = btrfs_root_node(root);
  1128. root->track_dirty = 1;
  1129. root->root_item.flags = 0;
  1130. root->root_item.byte_limit = 0;
  1131. btrfs_set_root_bytenr(&root->root_item, leaf->start);
  1132. btrfs_set_root_generation(&root->root_item, trans->transid);
  1133. btrfs_set_root_level(&root->root_item, 0);
  1134. btrfs_set_root_refs(&root->root_item, 1);
  1135. btrfs_set_root_used(&root->root_item, leaf->len);
  1136. btrfs_set_root_last_snapshot(&root->root_item, 0);
  1137. btrfs_set_root_dirid(&root->root_item, 0);
  1138. uuid_le_gen(&uuid);
  1139. memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
  1140. root->root_item.drop_level = 0;
  1141. key.objectid = objectid;
  1142. key.type = BTRFS_ROOT_ITEM_KEY;
  1143. key.offset = 0;
  1144. ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
  1145. if (ret)
  1146. goto fail;
  1147. btrfs_tree_unlock(leaf);
  1148. return root;
  1149. fail:
  1150. if (leaf) {
  1151. btrfs_tree_unlock(leaf);
  1152. free_extent_buffer(leaf);
  1153. }
  1154. kfree(root);
  1155. return ERR_PTR(ret);
  1156. }
  1157. static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
  1158. struct btrfs_fs_info *fs_info)
  1159. {
  1160. struct btrfs_root *root;
  1161. struct btrfs_root *tree_root = fs_info->tree_root;
  1162. struct extent_buffer *leaf;
  1163. root = btrfs_alloc_root(fs_info);
  1164. if (!root)
  1165. return ERR_PTR(-ENOMEM);
  1166. __setup_root(tree_root->nodesize, tree_root->leafsize,
  1167. tree_root->sectorsize, tree_root->stripesize,
  1168. root, fs_info, BTRFS_TREE_LOG_OBJECTID);
  1169. root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
  1170. root->root_key.type = BTRFS_ROOT_ITEM_KEY;
  1171. root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
  1172. /*
  1173. * log trees do not get reference counted because they go away
  1174. * before a real commit is actually done. They do store pointers
  1175. * to file data extents, and those reference counts still get
  1176. * updated (along with back refs to the log tree).
  1177. */
  1178. root->ref_cows = 0;
  1179. leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
  1180. BTRFS_TREE_LOG_OBJECTID, NULL,
  1181. 0, 0, 0);
  1182. if (IS_ERR(leaf)) {
  1183. kfree(root);
  1184. return ERR_CAST(leaf);
  1185. }
  1186. memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
  1187. btrfs_set_header_bytenr(leaf, leaf->start);
  1188. btrfs_set_header_generation(leaf, trans->transid);
  1189. btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
  1190. btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
  1191. root->node = leaf;
  1192. write_extent_buffer(root->node, root->fs_info->fsid,
  1193. (unsigned long)btrfs_header_fsid(root->node),
  1194. BTRFS_FSID_SIZE);
  1195. btrfs_mark_buffer_dirty(root->node);
  1196. btrfs_tree_unlock(root->node);
  1197. return root;
  1198. }
  1199. int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
  1200. struct btrfs_fs_info *fs_info)
  1201. {
  1202. struct btrfs_root *log_root;
  1203. log_root = alloc_log_tree(trans, fs_info);
  1204. if (IS_ERR(log_root))
  1205. return PTR_ERR(log_root);
  1206. WARN_ON(fs_info->log_root_tree);
  1207. fs_info->log_root_tree = log_root;
  1208. return 0;
  1209. }
  1210. int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
  1211. struct btrfs_root *root)
  1212. {
  1213. struct btrfs_root *log_root;
  1214. struct btrfs_inode_item *inode_item;
  1215. log_root = alloc_log_tree(trans, root->fs_info);
  1216. if (IS_ERR(log_root))
  1217. return PTR_ERR(log_root);
  1218. log_root->last_trans = trans->transid;
  1219. log_root->root_key.offset = root->root_key.objectid;
  1220. inode_item = &log_root->root_item.inode;
  1221. inode_item->generation = cpu_to_le64(1);
  1222. inode_item->size = cpu_to_le64(3);
  1223. inode_item->nlink = cpu_to_le32(1);
  1224. inode_item->nbytes = cpu_to_le64(root->leafsize);
  1225. inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
  1226. btrfs_set_root_node(&log_root->root_item, log_root->node);
  1227. WARN_ON(root->log_root);
  1228. root->log_root = log_root;
  1229. root->log_transid = 0;
  1230. root->last_log_commit = 0;
  1231. return 0;
  1232. }
  1233. struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
  1234. struct btrfs_key *location)
  1235. {
  1236. struct btrfs_root *root;
  1237. struct btrfs_fs_info *fs_info = tree_root->fs_info;
  1238. struct btrfs_path *path;
  1239. struct extent_buffer *l;
  1240. u64 generation;
  1241. u32 blocksize;
  1242. int ret = 0;
  1243. int slot;
  1244. root = btrfs_alloc_root(fs_info);
  1245. if (!root)
  1246. return ERR_PTR(-ENOMEM);
  1247. if (location->offset == (u64)-1) {
  1248. ret = find_and_setup_root(tree_root, fs_info,
  1249. location->objectid, root);
  1250. if (ret) {
  1251. kfree(root);
  1252. return ERR_PTR(ret);
  1253. }
  1254. goto out;
  1255. }
  1256. __setup_root(tree_root->nodesize, tree_root->leafsize,
  1257. tree_root->sectorsize, tree_root->stripesize,
  1258. root, fs_info, location->objectid);
  1259. path = btrfs_alloc_path();
  1260. if (!path) {
  1261. kfree(root);
  1262. return ERR_PTR(-ENOMEM);
  1263. }
  1264. ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
  1265. if (ret == 0) {
  1266. l = path->nodes[0];
  1267. slot = path->slots[0];
  1268. btrfs_read_root_item(l, slot, &root->root_item);
  1269. memcpy(&root->root_key, location, sizeof(*location));
  1270. }
  1271. btrfs_free_path(path);
  1272. if (ret) {
  1273. kfree(root);
  1274. if (ret > 0)
  1275. ret = -ENOENT;
  1276. return ERR_PTR(ret);
  1277. }
  1278. generation = btrfs_root_generation(&root->root_item);
  1279. blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
  1280. root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
  1281. blocksize, generation);
  1282. if (!root->node || !extent_buffer_uptodate(root->node)) {
  1283. ret = (!root->node) ? -ENOMEM : -EIO;
  1284. free_extent_buffer(root->node);
  1285. kfree(root);
  1286. return ERR_PTR(ret);
  1287. }
  1288. root->commit_root = btrfs_root_node(root);
  1289. BUG_ON(!root->node); /* -ENOMEM */
  1290. out:
  1291. if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
  1292. root->ref_cows = 1;
  1293. btrfs_check_and_init_root_item(&root->root_item);
  1294. }
  1295. return root;
  1296. }
  1297. struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
  1298. struct btrfs_key *location)
  1299. {
  1300. struct btrfs_root *root;
  1301. int ret;
  1302. if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
  1303. return fs_info->tree_root;
  1304. if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
  1305. return fs_info->extent_root;
  1306. if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
  1307. return fs_info->chunk_root;
  1308. if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
  1309. return fs_info->dev_root;
  1310. if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
  1311. return fs_info->csum_root;
  1312. if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
  1313. return fs_info->quota_root ? fs_info->quota_root :
  1314. ERR_PTR(-ENOENT);
  1315. again:
  1316. spin_lock(&fs_info->fs_roots_radix_lock);
  1317. root = radix_tree_lookup(&fs_info->fs_roots_radix,
  1318. (unsigned long)location->objectid);
  1319. spin_unlock(&fs_info->fs_roots_radix_lock);
  1320. if (root)
  1321. return root;
  1322. root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
  1323. if (IS_ERR(root))
  1324. return root;
  1325. root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
  1326. root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
  1327. GFP_NOFS);
  1328. if (!root->free_ino_pinned || !root->free_ino_ctl) {
  1329. ret = -ENOMEM;
  1330. goto fail;
  1331. }
  1332. btrfs_init_free_ino_ctl(root);
  1333. mutex_init(&root->fs_commit_mutex);
  1334. spin_lock_init(&root->cache_lock);
  1335. init_waitqueue_head(&root->cache_wait);
  1336. ret = get_anon_bdev(&root->anon_dev);
  1337. if (ret)
  1338. goto fail;
  1339. if (btrfs_root_refs(&root->root_item) == 0) {
  1340. ret = -ENOENT;
  1341. goto fail;
  1342. }
  1343. ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
  1344. if (ret < 0)
  1345. goto fail;
  1346. if (ret == 0)
  1347. root->orphan_item_inserted = 1;
  1348. ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
  1349. if (ret)
  1350. goto fail;
  1351. spin_lock(&fs_info->fs_roots_radix_lock);
  1352. ret = radix_tree_insert(&fs_info->fs_roots_radix,
  1353. (unsigned long)root->root_key.objectid,
  1354. root);
  1355. if (ret == 0)
  1356. root->in_radix = 1;
  1357. spin_unlock(&fs_info->fs_roots_radix_lock);
  1358. radix_tree_preload_end();
  1359. if (ret) {
  1360. if (ret == -EEXIST) {
  1361. free_fs_root(root);
  1362. goto again;
  1363. }
  1364. goto fail;
  1365. }
  1366. ret = btrfs_find_dead_roots(fs_info->tree_root,
  1367. root->root_key.objectid);
  1368. WARN_ON(ret);
  1369. return root;
  1370. fail:
  1371. free_fs_root(root);
  1372. return ERR_PTR(ret);
  1373. }
  1374. static int btrfs_congested_fn(void *congested_data, int bdi_bits)
  1375. {
  1376. struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
  1377. int ret = 0;
  1378. struct btrfs_device *device;
  1379. struct backing_dev_info *bdi;
  1380. rcu_read_lock();
  1381. list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
  1382. if (!device->bdev)
  1383. continue;
  1384. bdi = blk_get_backing_dev_info(device->bdev);
  1385. if (bdi && bdi_congested(bdi, bdi_bits)) {
  1386. ret = 1;
  1387. break;
  1388. }
  1389. }
  1390. rcu_read_unlock();
  1391. return ret;
  1392. }
  1393. /*
  1394. * If this fails, caller must call bdi_destroy() to get rid of the
  1395. * bdi again.
  1396. */
  1397. static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
  1398. {
  1399. int err;
  1400. bdi->capabilities = BDI_CAP_MAP_COPY;
  1401. err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
  1402. if (err)
  1403. return err;
  1404. bdi->ra_pages = default_backing_dev_info.ra_pages;
  1405. bdi->congested_fn = btrfs_congested_fn;
  1406. bdi->congested_data = info;
  1407. return 0;
  1408. }
  1409. /*
  1410. * called by the kthread helper functions to finally call the bio end_io
  1411. * functions. This is where read checksum verification actually happens
  1412. */
  1413. static void end_workqueue_fn(struct btrfs_work *work)
  1414. {
  1415. struct bio *bio;
  1416. struct end_io_wq *end_io_wq;
  1417. struct btrfs_fs_info *fs_info;
  1418. int error;
  1419. end_io_wq = container_of(work, struct end_io_wq, work);
  1420. bio = end_io_wq->bio;
  1421. fs_info = end_io_wq->info;
  1422. error = end_io_wq->error;
  1423. bio->bi_private = end_io_wq->private;
  1424. bio->bi_end_io = end_io_wq->end_io;
  1425. kfree(end_io_wq);
  1426. bio_endio(bio, error);
  1427. }
  1428. static int cleaner_kthread(void *arg)
  1429. {
  1430. struct btrfs_root *root = arg;
  1431. do {
  1432. int again = 0;
  1433. if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
  1434. down_read_trylock(&root->fs_info->sb->s_umount)) {
  1435. if (mutex_trylock(&root->fs_info->cleaner_mutex)) {
  1436. btrfs_run_delayed_iputs(root);
  1437. again = btrfs_clean_one_deleted_snapshot(root);
  1438. mutex_unlock(&root->fs_info->cleaner_mutex);
  1439. }
  1440. btrfs_run_defrag_inodes(root->fs_info);
  1441. up_read(&root->fs_info->sb->s_umount);
  1442. }
  1443. if (!try_to_freeze() && !again) {
  1444. set_current_state(TASK_INTERRUPTIBLE);
  1445. if (!kthread_should_stop())
  1446. schedule();
  1447. __set_current_state(TASK_RUNNING);
  1448. }
  1449. } while (!kthread_should_stop());
  1450. return 0;
  1451. }
  1452. static int transaction_kthread(void *arg)
  1453. {
  1454. struct btrfs_root *root = arg;
  1455. struct btrfs_trans_handle *trans;
  1456. struct btrfs_transaction *cur;
  1457. u64 transid;
  1458. unsigned long now;
  1459. unsigned long delay;
  1460. bool cannot_commit;
  1461. do {
  1462. cannot_commit = false;
  1463. delay = HZ * 30;
  1464. mutex_lock(&root->fs_info->transaction_kthread_mutex);
  1465. spin_lock(&root->fs_info->trans_lock);
  1466. cur = root->fs_info->running_transaction;
  1467. if (!cur) {
  1468. spin_unlock(&root->fs_info->trans_lock);
  1469. goto sleep;
  1470. }
  1471. now = get_seconds();
  1472. if (!cur->blocked &&
  1473. (now < cur->start_time || now - cur->start_time < 30)) {
  1474. spin_unlock(&root->fs_info->trans_lock);
  1475. delay = HZ * 5;
  1476. goto sleep;
  1477. }
  1478. transid = cur->transid;
  1479. spin_unlock(&root->fs_info->trans_lock);
  1480. /* If the file system is aborted, this will always fail. */
  1481. trans = btrfs_attach_transaction(root);
  1482. if (IS_ERR(trans)) {
  1483. if (PTR_ERR(trans) != -ENOENT)
  1484. cannot_commit = true;
  1485. goto sleep;
  1486. }
  1487. if (transid == trans->transid) {
  1488. btrfs_commit_transaction(trans, root);
  1489. } else {
  1490. btrfs_end_transaction(trans, root);
  1491. }
  1492. sleep:
  1493. wake_up_process(root->fs_info->cleaner_kthread);
  1494. mutex_unlock(&root->fs_info->transaction_kthread_mutex);
  1495. if (!try_to_freeze()) {
  1496. set_current_state(TASK_INTERRUPTIBLE);
  1497. if (!kthread_should_stop() &&
  1498. (!btrfs_transaction_blocked(root->fs_info) ||
  1499. cannot_commit))
  1500. schedule_timeout(delay);
  1501. __set_current_state(TASK_RUNNING);
  1502. }
  1503. } while (!kthread_should_stop());
  1504. return 0;
  1505. }
  1506. /*
  1507. * this will find the highest generation in the array of
  1508. * root backups. The index of the highest array is returned,
  1509. * or -1 if we can't find anything.
  1510. *
  1511. * We check to make sure the array is valid by comparing the
  1512. * generation of the latest root in the array with the generation
  1513. * in the super block. If they don't match we pitch it.
  1514. */
  1515. static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
  1516. {
  1517. u64 cur;
  1518. int newest_index = -1;
  1519. struct btrfs_root_backup *root_backup;
  1520. int i;
  1521. for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
  1522. root_backup = info->super_copy->super_roots + i;
  1523. cur = btrfs_backup_tree_root_gen(root_backup);
  1524. if (cur == newest_gen)
  1525. newest_index = i;
  1526. }
  1527. /* check to see if we actually wrapped around */
  1528. if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
  1529. root_backup = info->super_copy->super_roots;
  1530. cur = btrfs_backup_tree_root_gen(root_backup);
  1531. if (cur == newest_gen)
  1532. newest_index = 0;
  1533. }
  1534. return newest_index;
  1535. }
  1536. /*
  1537. * find the oldest backup so we know where to store new entries
  1538. * in the backup array. This will set the backup_root_index
  1539. * field in the fs_info struct
  1540. */
  1541. static void find_oldest_super_backup(struct btrfs_fs_info *info,
  1542. u64 newest_gen)
  1543. {
  1544. int newest_index = -1;
  1545. newest_index = find_newest_super_backup(info, newest_gen);
  1546. /* if there was garbage in there, just move along */
  1547. if (newest_index == -1) {
  1548. info->backup_root_index = 0;
  1549. } else {
  1550. info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
  1551. }
  1552. }
  1553. /*
  1554. * copy all the root pointers into the super backup array.
  1555. * this will bump the backup pointer by one when it is
  1556. * done
  1557. */
  1558. static void backup_super_roots(struct btrfs_fs_info *info)
  1559. {
  1560. int next_backup;
  1561. struct btrfs_root_backup *root_backup;
  1562. int last_backup;
  1563. next_backup = info->backup_root_index;
  1564. last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
  1565. BTRFS_NUM_BACKUP_ROOTS;
  1566. /*
  1567. * just overwrite the last backup if we're at the same generation
  1568. * this happens only at umount
  1569. */
  1570. root_backup = info->super_for_commit->super_roots + last_backup;
  1571. if (btrfs_backup_tree_root_gen(root_backup) ==
  1572. btrfs_header_generation(info->tree_root->node))
  1573. next_backup = last_backup;
  1574. root_backup = info->super_for_commit->super_roots + next_backup;
  1575. /*
  1576. * make sure all of our padding and empty slots get zero filled
  1577. * regardless of which ones we use today
  1578. */
  1579. memset(root_backup, 0, sizeof(*root_backup));
  1580. info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
  1581. btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
  1582. btrfs_set_backup_tree_root_gen(root_backup,
  1583. btrfs_header_generation(info->tree_root->node));
  1584. btrfs_set_backup_tree_root_level(root_backup,
  1585. btrfs_header_level(info->tree_root->node));
  1586. btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
  1587. btrfs_set_backup_chunk_root_gen(root_backup,
  1588. btrfs_header_generation(info->chunk_root->node));
  1589. btrfs_set_backup_chunk_root_level(root_backup,
  1590. btrfs_header_level(info->chunk_root->node));
  1591. btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
  1592. btrfs_set_backup_extent_root_gen(root_backup,
  1593. btrfs_header_generation(info->extent_root->node));
  1594. btrfs_set_backup_extent_root_level(root_backup,
  1595. btrfs_header_level(info->extent_root->node));
  1596. /*
  1597. * we might commit during log recovery, which happens before we set
  1598. * the fs_root. Make sure it is valid before we fill it in.
  1599. */
  1600. if (info->fs_root && info->fs_root->node) {
  1601. btrfs_set_backup_fs_root(root_backup,
  1602. info->fs_root->node->start);
  1603. btrfs_set_backup_fs_root_gen(root_backup,
  1604. btrfs_header_generation(info->fs_root->node));
  1605. btrfs_set_backup_fs_root_level(root_backup,
  1606. btrfs_header_level(info->fs_root->node));
  1607. }
  1608. btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
  1609. btrfs_set_backup_dev_root_gen(root_backup,
  1610. btrfs_header_generation(info->dev_root->node));
  1611. btrfs_set_backup_dev_root_level(root_backup,
  1612. btrfs_header_level(info->dev_root->node));
  1613. btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
  1614. btrfs_set_backup_csum_root_gen(root_backup,
  1615. btrfs_header_generation(info->csum_root->node));
  1616. btrfs_set_backup_csum_root_level(root_backup,
  1617. btrfs_header_level(info->csum_root->node));
  1618. btrfs_set_backup_total_bytes(root_backup,
  1619. btrfs_super_total_bytes(info->super_copy));
  1620. btrfs_set_backup_bytes_used(root_backup,
  1621. btrfs_super_bytes_used(info->super_copy));
  1622. btrfs_set_backup_num_devices(root_backup,
  1623. btrfs_super_num_devices(info->super_copy));
  1624. /*
  1625. * if we don't copy this out to the super_copy, it won't get remembered
  1626. * for the next commit
  1627. */
  1628. memcpy(&info->super_copy->super_roots,
  1629. &info->super_for_commit->super_roots,
  1630. sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
  1631. }
  1632. /*
  1633. * this copies info out of the root backup array and back into
  1634. * the in-memory super block. It is meant to help iterate through
  1635. * the array, so you send it the number of backups you've already
  1636. * tried and the last backup index you used.
  1637. *
  1638. * this returns -1 when it has tried all the backups
  1639. */
  1640. static noinline int next_root_backup(struct btrfs_fs_info *info,
  1641. struct btrfs_super_block *super,
  1642. int *num_backups_tried, int *backup_index)
  1643. {
  1644. struct btrfs_root_backup *root_backup;
  1645. int newest = *backup_index;
  1646. if (*num_backups_tried == 0) {
  1647. u64 gen = btrfs_super_generation(super);
  1648. newest = find_newest_super_backup(info, gen);
  1649. if (newest == -1)
  1650. return -1;
  1651. *backup_index = newest;
  1652. *num_backups_tried = 1;
  1653. } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
  1654. /* we've tried all the backups, all done */
  1655. return -1;
  1656. } else {
  1657. /* jump to the next oldest backup */
  1658. newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
  1659. BTRFS_NUM_BACKUP_ROOTS;
  1660. *backup_index = newest;
  1661. *num_backups_tried += 1;
  1662. }
  1663. root_backup = super->super_roots + newest;
  1664. btrfs_set_super_generation(super,
  1665. btrfs_backup_tree_root_gen(root_backup));
  1666. btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
  1667. btrfs_set_super_root_level(super,
  1668. btrfs_backup_tree_root_level(root_backup));
  1669. btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
  1670. /*
  1671. * fixme: the total bytes and num_devices need to match or we should
  1672. * need a fsck
  1673. */
  1674. btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
  1675. btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
  1676. return 0;
  1677. }
  1678. /* helper to cleanup workers */
  1679. static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
  1680. {
  1681. btrfs_stop_workers(&fs_info->generic_worker);
  1682. btrfs_stop_workers(&fs_info->fixup_workers);
  1683. btrfs_stop_workers(&fs_info->delalloc_workers);
  1684. btrfs_stop_workers(&fs_info->workers);
  1685. btrfs_stop_workers(&fs_info->endio_workers);
  1686. btrfs_stop_workers(&fs_info->endio_meta_workers);
  1687. btrfs_stop_workers(&fs_info->endio_raid56_workers);
  1688. btrfs_stop_workers(&fs_info->rmw_workers);
  1689. btrfs_stop_workers(&fs_info->endio_meta_write_workers);
  1690. btrfs_stop_workers(&fs_info->endio_write_workers);
  1691. btrfs_stop_workers(&fs_info->endio_freespace_worker);
  1692. btrfs_stop_workers(&fs_info->submit_workers);
  1693. btrfs_stop_workers(&fs_info->delayed_workers);
  1694. btrfs_stop_workers(&fs_info->caching_workers);
  1695. btrfs_stop_workers(&fs_info->readahead_workers);
  1696. btrfs_stop_workers(&fs_info->flush_workers);
  1697. btrfs_stop_workers(&fs_info->qgroup_rescan_workers);
  1698. }
  1699. /* helper to cleanup tree roots */
  1700. static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
  1701. {
  1702. free_extent_buffer(info->tree_root->node);
  1703. free_extent_buffer(info->tree_root->commit_root);
  1704. free_extent_buffer(info->dev_root->node);
  1705. free_extent_buffer(info->dev_root->commit_root);
  1706. free_extent_buffer(info->extent_root->node);
  1707. free_extent_buffer(info->extent_root->commit_root);
  1708. free_extent_buffer(info->csum_root->node);
  1709. free_extent_buffer(info->csum_root->commit_root);
  1710. if (info->quota_root) {
  1711. free_extent_buffer(info->quota_root->node);
  1712. free_extent_buffer(info->quota_root->commit_root);
  1713. }
  1714. info->tree_root->node = NULL;
  1715. info->tree_root->commit_root = NULL;
  1716. info->dev_root->node = NULL;
  1717. info->dev_root->commit_root = NULL;
  1718. info->extent_root->node = NULL;
  1719. info->extent_root->commit_root = NULL;
  1720. info->csum_root->node = NULL;
  1721. info->csum_root->commit_root = NULL;
  1722. if (info->quota_root) {
  1723. info->quota_root->node = NULL;
  1724. info->quota_root->commit_root = NULL;
  1725. }
  1726. if (chunk_root) {
  1727. free_extent_buffer(info->chunk_root->node);
  1728. free_extent_buffer(info->chunk_root->commit_root);
  1729. info->chunk_root->node = NULL;
  1730. info->chunk_root->commit_root = NULL;
  1731. }
  1732. }
  1733. static void del_fs_roots(struct btrfs_fs_info *fs_info)
  1734. {
  1735. int ret;
  1736. struct btrfs_root *gang[8];
  1737. int i;
  1738. while (!list_empty(&fs_info->dead_roots)) {
  1739. gang[0] = list_entry(fs_info->dead_roots.next,
  1740. struct btrfs_root, root_list);
  1741. list_del(&gang[0]->root_list);
  1742. if (gang[0]->in_radix) {
  1743. btrfs_free_fs_root(fs_info, gang[0]);
  1744. } else {
  1745. free_extent_buffer(gang[0]->node);
  1746. free_extent_buffer(gang[0]->commit_root);
  1747. kfree(gang[0]);
  1748. }
  1749. }
  1750. while (1) {
  1751. ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
  1752. (void **)gang, 0,
  1753. ARRAY_SIZE(gang));
  1754. if (!ret)
  1755. break;
  1756. for (i = 0; i < ret; i++)
  1757. btrfs_free_fs_root(fs_info, gang[i]);
  1758. }
  1759. }
  1760. int open_ctree(struct super_block *sb,
  1761. struct btrfs_fs_devices *fs_devices,
  1762. char *options)
  1763. {
  1764. u32 sectorsize;
  1765. u32 nodesize;
  1766. u32 leafsize;
  1767. u32 blocksize;
  1768. u32 stripesize;
  1769. u64 generation;
  1770. u64 features;
  1771. struct btrfs_key location;
  1772. struct buffer_head *bh;
  1773. struct btrfs_super_block *disk_super;
  1774. struct btrfs_fs_info *fs_info = btrfs_sb(sb);
  1775. struct btrfs_root *tree_root;
  1776. struct btrfs_root *extent_root;
  1777. struct btrfs_root *csum_root;
  1778. struct btrfs_root *chunk_root;
  1779. struct btrfs_root *dev_root;
  1780. struct btrfs_root *quota_root;
  1781. struct btrfs_root *log_tree_root;
  1782. int ret;
  1783. int err = -EINVAL;
  1784. int num_backups_tried = 0;
  1785. int backup_index = 0;
  1786. tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
  1787. extent_root = fs_info->extent_root = btrfs_alloc_root(fs_info);
  1788. csum_root = fs_info->csum_root = btrfs_alloc_root(fs_info);
  1789. chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
  1790. dev_root = fs_info->dev_root = btrfs_alloc_root(fs_info);
  1791. quota_root = fs_info->quota_root = btrfs_alloc_root(fs_info);
  1792. if (!tree_root || !extent_root || !csum_root ||
  1793. !chunk_root || !dev_root || !quota_root) {
  1794. err = -ENOMEM;
  1795. goto fail;
  1796. }
  1797. ret = init_srcu_struct(&fs_info->subvol_srcu);
  1798. if (ret) {
  1799. err = ret;
  1800. goto fail;
  1801. }
  1802. ret = setup_bdi(fs_info, &fs_info->bdi);
  1803. if (ret) {
  1804. err = ret;
  1805. goto fail_srcu;
  1806. }
  1807. ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0);
  1808. if (ret) {
  1809. err = ret;
  1810. goto fail_bdi;
  1811. }
  1812. fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
  1813. (1 + ilog2(nr_cpu_ids));
  1814. ret = percpu_counter_init(&fs_info->delalloc_bytes, 0);
  1815. if (ret) {
  1816. err = ret;
  1817. goto fail_dirty_metadata_bytes;
  1818. }
  1819. fs_info->btree_inode = new_inode(sb);
  1820. if (!fs_info->btree_inode) {
  1821. err = -ENOMEM;
  1822. goto fail_delalloc_bytes;
  1823. }
  1824. mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
  1825. INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
  1826. INIT_LIST_HEAD(&fs_info->trans_list);
  1827. INIT_LIST_HEAD(&fs_info->dead_roots);
  1828. INIT_LIST_HEAD(&fs_info->delayed_iputs);
  1829. INIT_LIST_HEAD(&fs_info->delalloc_inodes);
  1830. INIT_LIST_HEAD(&fs_info->caching_block_groups);
  1831. spin_lock_init(&fs_info->delalloc_lock);
  1832. spin_lock_init(&fs_info->trans_lock);
  1833. spin_lock_init(&fs_info->fs_roots_radix_lock);
  1834. spin_lock_init(&fs_info->delayed_iput_lock);
  1835. spin_lock_init(&fs_info->defrag_inodes_lock);
  1836. spin_lock_init(&fs_info->free_chunk_lock);
  1837. spin_lock_init(&fs_info->tree_mod_seq_lock);
  1838. spin_lock_init(&fs_info->super_lock);
  1839. rwlock_init(&fs_info->tree_mod_log_lock);
  1840. mutex_init(&fs_info->reloc_mutex);
  1841. seqlock_init(&fs_info->profiles_lock);
  1842. init_completion(&fs_info->kobj_unregister);
  1843. INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
  1844. INIT_LIST_HEAD(&fs_info->space_info);
  1845. INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
  1846. btrfs_mapping_init(&fs_info->mapping_tree);
  1847. btrfs_init_block_rsv(&fs_info->global_block_rsv,
  1848. BTRFS_BLOCK_RSV_GLOBAL);
  1849. btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
  1850. BTRFS_BLOCK_RSV_DELALLOC);
  1851. btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
  1852. btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
  1853. btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
  1854. btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
  1855. BTRFS_BLOCK_RSV_DELOPS);
  1856. atomic_set(&fs_info->nr_async_submits, 0);
  1857. atomic_set(&fs_info->async_delalloc_pages, 0);
  1858. atomic_set(&fs_info->async_submit_draining, 0);
  1859. atomic_set(&fs_info->nr_async_bios, 0);
  1860. atomic_set(&fs_info->defrag_running, 0);
  1861. atomic64_set(&fs_info->tree_mod_seq, 0);
  1862. fs_info->sb = sb;
  1863. fs_info->max_inline = 8192 * 1024;
  1864. fs_info->metadata_ratio = 0;
  1865. fs_info->defrag_inodes = RB_ROOT;
  1866. fs_info->trans_no_join = 0;
  1867. fs_info->free_chunk_space = 0;
  1868. fs_info->tree_mod_log = RB_ROOT;
  1869. /* readahead state */
  1870. INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
  1871. spin_lock_init(&fs_info->reada_lock);
  1872. fs_info->thread_pool_size = min_t(unsigned long,
  1873. num_online_cpus() + 2, 8);
  1874. INIT_LIST_HEAD(&fs_info->ordered_extents);
  1875. spin_lock_init(&fs_info->ordered_extent_lock);
  1876. fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
  1877. GFP_NOFS);
  1878. if (!fs_info->delayed_root) {
  1879. err = -ENOMEM;
  1880. goto fail_iput;
  1881. }
  1882. btrfs_init_delayed_root(fs_info->delayed_root);
  1883. mutex_init(&fs_info->scrub_lock);
  1884. atomic_set(&fs_info->scrubs_running, 0);
  1885. atomic_set(&fs_info->scrub_pause_req, 0);
  1886. atomic_set(&fs_info->scrubs_paused, 0);
  1887. atomic_set(&fs_info->scrub_cancel_req, 0);
  1888. init_waitqueue_head(&fs_info->scrub_pause_wait);
  1889. init_rwsem(&fs_info->scrub_super_lock);
  1890. fs_info->scrub_workers_refcnt = 0;
  1891. #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
  1892. fs_info->check_integrity_print_mask = 0;
  1893. #endif
  1894. spin_lock_init(&fs_info->balance_lock);
  1895. mutex_init(&fs_info->balance_mutex);
  1896. atomic_set(&fs_info->balance_running, 0);
  1897. atomic_set(&fs_info->balance_pause_req, 0);
  1898. atomic_set(&fs_info->balance_cancel_req, 0);
  1899. fs_info->balance_ctl = NULL;
  1900. init_waitqueue_head(&fs_info->balance_wait_q);
  1901. sb->s_blocksize = 4096;
  1902. sb->s_blocksize_bits = blksize_bits(4096);
  1903. sb->s_bdi = &fs_info->bdi;
  1904. fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
  1905. set_nlink(fs_info->btree_inode, 1);
  1906. /*
  1907. * we set the i_size on the btree inode to the max possible int.
  1908. * the real end of the address space is determined by all of
  1909. * the devices in the system
  1910. */
  1911. fs_info->btree_inode->i_size = OFFSET_MAX;
  1912. fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
  1913. fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
  1914. RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
  1915. extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
  1916. fs_info->btree_inode->i_mapping);
  1917. BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
  1918. extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
  1919. BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
  1920. BTRFS_I(fs_info->btree_inode)->root = tree_root;
  1921. memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
  1922. sizeof(struct btrfs_key));
  1923. set_bit(BTRFS_INODE_DUMMY,
  1924. &BTRFS_I(fs_info->btree_inode)->runtime_flags);
  1925. insert_inode_hash(fs_info->btree_inode);
  1926. spin_lock_init(&fs_info->block_group_cache_lock);
  1927. fs_info->block_group_cache_tree = RB_ROOT;
  1928. fs_info->first_logical_byte = (u64)-1;
  1929. extent_io_tree_init(&fs_info->freed_extents[0],
  1930. fs_info->btree_inode->i_mapping);
  1931. extent_io_tree_init(&fs_info->freed_extents[1],
  1932. fs_info->btree_inode->i_mapping);
  1933. fs_info->pinned_extents = &fs_info->freed_extents[0];
  1934. fs_info->do_barriers = 1;
  1935. mutex_init(&fs_info->ordered_operations_mutex);
  1936. mutex_init(&fs_info->tree_log_mutex);
  1937. mutex_init(&fs_info->chunk_mutex);
  1938. mutex_init(&fs_info->transaction_kthread_mutex);
  1939. mutex_init(&fs_info->cleaner_mutex);
  1940. mutex_init(&fs_info->volume_mutex);
  1941. init_rwsem(&fs_info->extent_commit_sem);
  1942. init_rwsem(&fs_info->cleanup_work_sem);
  1943. init_rwsem(&fs_info->subvol_sem);
  1944. fs_info->dev_replace.lock_owner = 0;
  1945. atomic_set(&fs_info->dev_replace.nesting_level, 0);
  1946. mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
  1947. mutex_init(&fs_info->dev_replace.lock_management_lock);
  1948. mutex_init(&fs_info->dev_replace.lock);
  1949. spin_lock_init(&fs_info->qgroup_lock);
  1950. mutex_init(&fs_info->qgroup_ioctl_lock);
  1951. fs_info->qgroup_tree = RB_ROOT;
  1952. INIT_LIST_HEAD(&fs_info->dirty_qgroups);
  1953. fs_info->qgroup_seq = 1;
  1954. fs_info->quota_enabled = 0;
  1955. fs_info->pending_quota_state = 0;
  1956. mutex_init(&fs_info->qgroup_rescan_lock);
  1957. btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
  1958. btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
  1959. init_waitqueue_head(&fs_info->transaction_throttle);
  1960. init_waitqueue_head(&fs_info->transaction_wait);
  1961. init_waitqueue_head(&fs_info->transaction_blocked_wait);
  1962. init_waitqueue_head(&fs_info->async_submit_wait);
  1963. ret = btrfs_alloc_stripe_hash_table(fs_info);
  1964. if (ret) {
  1965. err = ret;
  1966. goto fail_alloc;
  1967. }
  1968. __setup_root(4096, 4096, 4096, 4096, tree_root,
  1969. fs_info, BTRFS_ROOT_TREE_OBJECTID);
  1970. invalidate_bdev(fs_devices->latest_bdev);
  1971. bh = btrfs_read_dev_super(fs_devices->latest_bdev);
  1972. if (!bh) {
  1973. err = -EINVAL;
  1974. goto fail_alloc;
  1975. }
  1976. memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
  1977. memcpy(fs_info->super_for_commit, fs_info->super_copy,
  1978. sizeof(*fs_info->super_for_commit));
  1979. brelse(bh);
  1980. memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
  1981. disk_super = fs_info->super_copy;
  1982. if (!btrfs_super_root(disk_super))
  1983. goto fail_alloc;
  1984. /* check FS state, whether FS is broken. */
  1985. if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
  1986. set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
  1987. ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
  1988. if (ret) {
  1989. printk(KERN_ERR "btrfs: superblock contains fatal errors\n");
  1990. err = ret;
  1991. goto fail_alloc;
  1992. }
  1993. /*
  1994. * run through our array of backup supers and setup
  1995. * our ring pointer to the oldest one
  1996. */
  1997. generation = btrfs_super_generation(disk_super);
  1998. find_oldest_super_backup(fs_info, generation);
  1999. /*
  2000. * In the long term, we'll store the compression type in the super
  2001. * block, and it'll be used for per file compression control.
  2002. */
  2003. fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
  2004. ret = btrfs_parse_options(tree_root, options);
  2005. if (ret) {
  2006. err = ret;
  2007. goto fail_alloc;
  2008. }
  2009. features = btrfs_super_incompat_flags(disk_super) &
  2010. ~BTRFS_FEATURE_INCOMPAT_SUPP;
  2011. if (features) {
  2012. printk(KERN_ERR "BTRFS: couldn't mount because of "
  2013. "unsupported optional features (%Lx).\n",
  2014. (unsigned long long)features);
  2015. err = -EINVAL;
  2016. goto fail_alloc;
  2017. }
  2018. if (btrfs_super_leafsize(disk_super) !=
  2019. btrfs_super_nodesize(disk_super)) {
  2020. printk(KERN_ERR "BTRFS: couldn't mount because metadata "
  2021. "blocksizes don't match. node %d leaf %d\n",
  2022. btrfs_super_nodesize(disk_super),
  2023. btrfs_super_leafsize(disk_super));
  2024. err = -EINVAL;
  2025. goto fail_alloc;
  2026. }
  2027. if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
  2028. printk(KERN_ERR "BTRFS: couldn't mount because metadata "
  2029. "blocksize (%d) was too large\n",
  2030. btrfs_super_leafsize(disk_super));
  2031. err = -EINVAL;
  2032. goto fail_alloc;
  2033. }
  2034. features = btrfs_super_incompat_flags(disk_super);
  2035. features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
  2036. if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
  2037. features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
  2038. if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
  2039. printk(KERN_ERR "btrfs: has skinny extents\n");
  2040. /*
  2041. * flag our filesystem as having big metadata blocks if
  2042. * they are bigger than the page size
  2043. */
  2044. if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) {
  2045. if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
  2046. printk(KERN_INFO "btrfs flagging fs with big metadata feature\n");
  2047. features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
  2048. }
  2049. nodesize = btrfs_super_nodesize(disk_super);
  2050. leafsize = btrfs_super_leafsize(disk_super);
  2051. sectorsize = btrfs_super_sectorsize(disk_super);
  2052. stripesize = btrfs_super_stripesize(disk_super);
  2053. fs_info->dirty_metadata_batch = leafsize * (1 + ilog2(nr_cpu_ids));
  2054. fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
  2055. /*
  2056. * mixed block groups end up with duplicate but slightly offset
  2057. * extent buffers for the same range. It leads to corruptions
  2058. */
  2059. if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
  2060. (sectorsize != leafsize)) {
  2061. printk(KERN_WARNING "btrfs: unequal leaf/node/sector sizes "
  2062. "are not allowed for mixed block groups on %s\n",
  2063. sb->s_id);
  2064. goto fail_alloc;
  2065. }
  2066. /*
  2067. * Needn't use the lock because there is no other task which will
  2068. * update the flag.
  2069. */
  2070. btrfs_set_super_incompat_flags(disk_super, features);
  2071. features = btrfs_super_compat_ro_flags(disk_super) &
  2072. ~BTRFS_FEATURE_COMPAT_RO_SUPP;
  2073. if (!(sb->s_flags & MS_RDONLY) && features) {
  2074. printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
  2075. "unsupported option features (%Lx).\n",
  2076. (unsigned long long)features);
  2077. err = -EINVAL;
  2078. goto fail_alloc;
  2079. }
  2080. btrfs_init_workers(&fs_info->generic_worker,
  2081. "genwork", 1, NULL);
  2082. btrfs_init_workers(&fs_info->workers, "worker",
  2083. fs_info->thread_pool_size,
  2084. &fs_info->generic_worker);
  2085. btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
  2086. fs_info->thread_pool_size,
  2087. &fs_info->generic_worker);
  2088. btrfs_init_workers(&fs_info->flush_workers, "flush_delalloc",
  2089. fs_info->thread_pool_size,
  2090. &fs_info->generic_worker);
  2091. btrfs_init_workers(&fs_info->submit_workers, "submit",
  2092. min_t(u64, fs_devices->num_devices,
  2093. fs_info->thread_pool_size),
  2094. &fs_info->generic_worker);
  2095. btrfs_init_workers(&fs_info->caching_workers, "cache",
  2096. 2, &fs_info->generic_worker);
  2097. /* a higher idle thresh on the submit workers makes it much more
  2098. * likely that bios will be send down in a sane order to the
  2099. * devices
  2100. */
  2101. fs_info->submit_workers.idle_thresh = 64;
  2102. fs_info->workers.idle_thresh = 16;
  2103. fs_info->workers.ordered = 1;
  2104. fs_info->delalloc_workers.idle_thresh = 2;
  2105. fs_info->delalloc_workers.ordered = 1;
  2106. btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
  2107. &fs_info->generic_worker);
  2108. btrfs_init_workers(&fs_info->endio_workers, "endio",
  2109. fs_info->thread_pool_size,
  2110. &fs_info->generic_worker);
  2111. btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
  2112. fs_info->thread_pool_size,
  2113. &fs_info->generic_worker);
  2114. btrfs_init_workers(&fs_info->endio_meta_write_workers,
  2115. "endio-meta-write", fs_info->thread_pool_size,
  2116. &fs_info->generic_worker);
  2117. btrfs_init_workers(&fs_info->endio_raid56_workers,
  2118. "endio-raid56", fs_info->thread_pool_size,
  2119. &fs_info->generic_worker);
  2120. btrfs_init_workers(&fs_info->rmw_workers,
  2121. "rmw", fs_info->thread_pool_size,
  2122. &fs_info->generic_worker);
  2123. btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
  2124. fs_info->thread_pool_size,
  2125. &fs_info->generic_worker);
  2126. btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
  2127. 1, &fs_info->generic_worker);
  2128. btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
  2129. fs_info->thread_pool_size,
  2130. &fs_info->generic_worker);
  2131. btrfs_init_workers(&fs_info->readahead_workers, "readahead",
  2132. fs_info->thread_pool_size,
  2133. &fs_info->generic_worker);
  2134. btrfs_init_workers(&fs_info->qgroup_rescan_workers, "qgroup-rescan", 1,
  2135. &fs_info->generic_worker);
  2136. /*
  2137. * endios are largely parallel and should have a very
  2138. * low idle thresh
  2139. */
  2140. fs_info->endio_workers.idle_thresh = 4;
  2141. fs_info->endio_meta_workers.idle_thresh = 4;
  2142. fs_info->endio_raid56_workers.idle_thresh = 4;
  2143. fs_info->rmw_workers.idle_thresh = 2;
  2144. fs_info->endio_write_workers.idle_thresh = 2;
  2145. fs_info->endio_meta_write_workers.idle_thresh = 2;
  2146. fs_info->readahead_workers.idle_thresh = 2;
  2147. /*
  2148. * btrfs_start_workers can really only fail because of ENOMEM so just
  2149. * return -ENOMEM if any of these fail.
  2150. */
  2151. ret = btrfs_start_workers(&fs_info->workers);
  2152. ret |= btrfs_start_workers(&fs_info->generic_worker);
  2153. ret |= btrfs_start_workers(&fs_info->submit_workers);
  2154. ret |= btrfs_start_workers(&fs_info->delalloc_workers);
  2155. ret |= btrfs_start_workers(&fs_info->fixup_workers);
  2156. ret |= btrfs_start_workers(&fs_info->endio_workers);
  2157. ret |= btrfs_start_workers(&fs_info->endio_meta_workers);
  2158. ret |= btrfs_start_workers(&fs_info->rmw_workers);
  2159. ret |= btrfs_start_workers(&fs_info->endio_raid56_workers);
  2160. ret |= btrfs_start_workers(&fs_info->endio_meta_write_workers);
  2161. ret |= btrfs_start_workers(&fs_info->endio_write_workers);
  2162. ret |= btrfs_start_workers(&fs_info->endio_freespace_worker);
  2163. ret |= btrfs_start_workers(&fs_info->delayed_workers);
  2164. ret |= btrfs_start_workers(&fs_info->caching_workers);
  2165. ret |= btrfs_start_workers(&fs_info->readahead_workers);
  2166. ret |= btrfs_start_workers(&fs_info->flush_workers);
  2167. ret |= btrfs_start_workers(&fs_info->qgroup_rescan_workers);
  2168. if (ret) {
  2169. err = -ENOMEM;
  2170. goto fail_sb_buffer;
  2171. }
  2172. fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
  2173. fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
  2174. 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
  2175. tree_root->nodesize = nodesize;
  2176. tree_root->leafsize = leafsize;
  2177. tree_root->sectorsize = sectorsize;
  2178. tree_root->stripesize = stripesize;
  2179. sb->s_blocksize = sectorsize;
  2180. sb->s_blocksize_bits = blksize_bits(sectorsize);
  2181. if (disk_super->magic != cpu_to_le64(BTRFS_MAGIC)) {
  2182. printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
  2183. goto fail_sb_buffer;
  2184. }
  2185. if (sectorsize != PAGE_SIZE) {
  2186. printk(KERN_WARNING "btrfs: Incompatible sector size(%lu) "
  2187. "found on %s\n", (unsigned long)sectorsize, sb->s_id);
  2188. goto fail_sb_buffer;
  2189. }
  2190. mutex_lock(&fs_info->chunk_mutex);
  2191. ret = btrfs_read_sys_array(tree_root);
  2192. mutex_unlock(&fs_info->chunk_mutex);
  2193. if (ret) {
  2194. printk(KERN_WARNING "btrfs: failed to read the system "
  2195. "array on %s\n", sb->s_id);
  2196. goto fail_sb_buffer;
  2197. }
  2198. blocksize = btrfs_level_size(tree_root,
  2199. btrfs_super_chunk_root_level(disk_super));
  2200. generation = btrfs_super_chunk_root_generation(disk_super);
  2201. __setup_root(nodesize, leafsize, sectorsize, stripesize,
  2202. chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
  2203. chunk_root->node = read_tree_block(chunk_root,
  2204. btrfs_super_chunk_root(disk_super),
  2205. blocksize, generation);
  2206. if (!chunk_root->node ||
  2207. !test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
  2208. printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
  2209. sb->s_id);
  2210. goto fail_tree_roots;
  2211. }
  2212. btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
  2213. chunk_root->commit_root = btrfs_root_node(chunk_root);
  2214. read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
  2215. (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
  2216. BTRFS_UUID_SIZE);
  2217. ret = btrfs_read_chunk_tree(chunk_root);
  2218. if (ret) {
  2219. printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
  2220. sb->s_id);
  2221. goto fail_tree_roots;
  2222. }
  2223. /*
  2224. * keep the device that is marked to be the target device for the
  2225. * dev_replace procedure
  2226. */
  2227. btrfs_close_extra_devices(fs_info, fs_devices, 0);
  2228. if (!fs_devices->latest_bdev) {
  2229. printk(KERN_CRIT "btrfs: failed to read devices on %s\n",
  2230. sb->s_id);
  2231. goto fail_tree_roots;
  2232. }
  2233. retry_root_backup:
  2234. blocksize = btrfs_level_size(tree_root,
  2235. btrfs_super_root_level(disk_super));
  2236. generation = btrfs_super_generation(disk_super);
  2237. tree_root->node = read_tree_block(tree_root,
  2238. btrfs_super_root(disk_super),
  2239. blocksize, generation);
  2240. if (!tree_root->node ||
  2241. !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
  2242. printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
  2243. sb->s_id);
  2244. goto recovery_tree_root;
  2245. }
  2246. btrfs_set_root_node(&tree_root->root_item, tree_root->node);
  2247. tree_root->commit_root = btrfs_root_node(tree_root);
  2248. ret = find_and_setup_root(tree_root, fs_info,
  2249. BTRFS_EXTENT_TREE_OBJECTID, extent_root);
  2250. if (ret)
  2251. goto recovery_tree_root;
  2252. extent_root->track_dirty = 1;
  2253. ret = find_and_setup_root(tree_root, fs_info,
  2254. BTRFS_DEV_TREE_OBJECTID, dev_root);
  2255. if (ret)
  2256. goto recovery_tree_root;
  2257. dev_root->track_dirty = 1;
  2258. ret = find_and_setup_root(tree_root, fs_info,
  2259. BTRFS_CSUM_TREE_OBJECTID, csum_root);
  2260. if (ret)
  2261. goto recovery_tree_root;
  2262. csum_root->track_dirty = 1;
  2263. ret = find_and_setup_root(tree_root, fs_info,
  2264. BTRFS_QUOTA_TREE_OBJECTID, quota_root);
  2265. if (ret) {
  2266. kfree(quota_root);
  2267. quota_root = fs_info->quota_root = NULL;
  2268. } else {
  2269. quota_root->track_dirty = 1;
  2270. fs_info->quota_enabled = 1;
  2271. fs_info->pending_quota_state = 1;
  2272. }
  2273. fs_info->generation = generation;
  2274. fs_info->last_trans_committed = generation;
  2275. ret = btrfs_recover_balance(fs_info);
  2276. if (ret) {
  2277. printk(KERN_WARNING "btrfs: failed to recover balance\n");
  2278. goto fail_block_groups;
  2279. }
  2280. ret = btrfs_init_dev_stats(fs_info);
  2281. if (ret) {
  2282. printk(KERN_ERR "btrfs: failed to init dev_stats: %d\n",
  2283. ret);
  2284. goto fail_block_groups;
  2285. }
  2286. ret = btrfs_init_dev_replace(fs_info);
  2287. if (ret) {
  2288. pr_err("btrfs: failed to init dev_replace: %d\n", ret);
  2289. goto fail_block_groups;
  2290. }
  2291. btrfs_close_extra_devices(fs_info, fs_devices, 1);
  2292. ret = btrfs_init_space_info(fs_info);
  2293. if (ret) {
  2294. printk(KERN_ERR "Failed to initial space info: %d\n", ret);
  2295. goto fail_block_groups;
  2296. }
  2297. ret = btrfs_read_block_groups(extent_root);
  2298. if (ret) {
  2299. printk(KERN_ERR "Failed to read block groups: %d\n", ret);
  2300. goto fail_block_groups;
  2301. }
  2302. fs_info->num_tolerated_disk_barrier_failures =
  2303. btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
  2304. if (fs_info->fs_devices->missing_devices >
  2305. fs_info->num_tolerated_disk_barrier_failures &&
  2306. !(sb->s_flags & MS_RDONLY)) {
  2307. printk(KERN_WARNING
  2308. "Btrfs: too many missing devices, writeable mount is not allowed\n");
  2309. goto fail_block_groups;
  2310. }
  2311. fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
  2312. "btrfs-cleaner");
  2313. if (IS_ERR(fs_info->cleaner_kthread))
  2314. goto fail_block_groups;
  2315. fs_info->transaction_kthread = kthread_run(transaction_kthread,
  2316. tree_root,
  2317. "btrfs-transaction");
  2318. if (IS_ERR(fs_info->transaction_kthread))
  2319. goto fail_cleaner;
  2320. if (!btrfs_test_opt(tree_root, SSD) &&
  2321. !btrfs_test_opt(tree_root, NOSSD) &&
  2322. !fs_info->fs_devices->rotating) {
  2323. printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
  2324. "mode\n");
  2325. btrfs_set_opt(fs_info->mount_opt, SSD);
  2326. }
  2327. #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
  2328. if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
  2329. ret = btrfsic_mount(tree_root, fs_devices,
  2330. btrfs_test_opt(tree_root,
  2331. CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
  2332. 1 : 0,
  2333. fs_info->check_integrity_print_mask);
  2334. if (ret)
  2335. printk(KERN_WARNING "btrfs: failed to initialize"
  2336. " integrity check module %s\n", sb->s_id);
  2337. }
  2338. #endif
  2339. ret = btrfs_read_qgroup_config(fs_info);
  2340. if (ret)
  2341. goto fail_trans_kthread;
  2342. /* do not make disk changes in broken FS */
  2343. if (btrfs_super_log_root(disk_super) != 0) {
  2344. u64 bytenr = btrfs_super_log_root(disk_super);
  2345. if (fs_devices->rw_devices == 0) {
  2346. printk(KERN_WARNING "Btrfs log replay required "
  2347. "on RO media\n");
  2348. err = -EIO;
  2349. goto fail_qgroup;
  2350. }
  2351. blocksize =
  2352. btrfs_level_size(tree_root,
  2353. btrfs_super_log_root_level(disk_super));
  2354. log_tree_root = btrfs_alloc_root(fs_info);
  2355. if (!log_tree_root) {
  2356. err = -ENOMEM;
  2357. goto fail_qgroup;
  2358. }
  2359. __setup_root(nodesize, leafsize, sectorsize, stripesize,
  2360. log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
  2361. log_tree_root->node = read_tree_block(tree_root, bytenr,
  2362. blocksize,
  2363. generation + 1);
  2364. if (!log_tree_root->node ||
  2365. !extent_buffer_uptodate(log_tree_root->node)) {
  2366. printk(KERN_ERR "btrfs: failed to read log tree\n");
  2367. free_extent_buffer(log_tree_root->node);
  2368. kfree(log_tree_root);
  2369. goto fail_trans_kthread;
  2370. }
  2371. /* returns with log_tree_root freed on success */
  2372. ret = btrfs_recover_log_trees(log_tree_root);
  2373. if (ret) {
  2374. btrfs_error(tree_root->fs_info, ret,
  2375. "Failed to recover log tree");
  2376. free_extent_buffer(log_tree_root->node);
  2377. kfree(log_tree_root);
  2378. goto fail_trans_kthread;
  2379. }
  2380. if (sb->s_flags & MS_RDONLY) {
  2381. ret = btrfs_commit_super(tree_root);
  2382. if (ret)
  2383. goto fail_trans_kthread;
  2384. }
  2385. }
  2386. ret = btrfs_find_orphan_roots(tree_root);
  2387. if (ret)
  2388. goto fail_trans_kthread;
  2389. if (!(sb->s_flags & MS_RDONLY)) {
  2390. ret = btrfs_cleanup_fs_roots(fs_info);
  2391. if (ret)
  2392. goto fail_trans_kthread;
  2393. ret = btrfs_recover_relocation(tree_root);
  2394. if (ret < 0) {
  2395. printk(KERN_WARNING
  2396. "btrfs: failed to recover relocation\n");
  2397. err = -EINVAL;
  2398. goto fail_qgroup;
  2399. }
  2400. }
  2401. location.objectid = BTRFS_FS_TREE_OBJECTID;
  2402. location.type = BTRFS_ROOT_ITEM_KEY;
  2403. location.offset = (u64)-1;
  2404. fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
  2405. if (!fs_info->fs_root)
  2406. goto fail_qgroup;
  2407. if (IS_ERR(fs_info->fs_root)) {
  2408. err = PTR_ERR(fs_info->fs_root);
  2409. goto fail_qgroup;
  2410. }
  2411. if (sb->s_flags & MS_RDONLY)
  2412. return 0;
  2413. down_read(&fs_info->cleanup_work_sem);
  2414. if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
  2415. (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
  2416. up_read(&fs_info->cleanup_work_sem);
  2417. close_ctree(tree_root);
  2418. return ret;
  2419. }
  2420. up_read(&fs_info->cleanup_work_sem);
  2421. ret = btrfs_resume_balance_async(fs_info);
  2422. if (ret) {
  2423. printk(KERN_WARNING "btrfs: failed to resume balance\n");
  2424. close_ctree(tree_root);
  2425. return ret;
  2426. }
  2427. ret = btrfs_resume_dev_replace_async(fs_info);
  2428. if (ret) {
  2429. pr_warn("btrfs: failed to resume dev_replace\n");
  2430. close_ctree(tree_root);
  2431. return ret;
  2432. }
  2433. return 0;
  2434. fail_qgroup:
  2435. btrfs_free_qgroup_config(fs_info);
  2436. fail_trans_kthread:
  2437. kthread_stop(fs_info->transaction_kthread);
  2438. del_fs_roots(fs_info);
  2439. btrfs_cleanup_transaction(fs_info->tree_root);
  2440. fail_cleaner:
  2441. kthread_stop(fs_info->cleaner_kthread);
  2442. /*
  2443. * make sure we're done with the btree inode before we stop our
  2444. * kthreads
  2445. */
  2446. filemap_write_and_wait(fs_info->btree_inode->i_mapping);
  2447. fail_block_groups:
  2448. btrfs_put_block_group_cache(fs_info);
  2449. btrfs_free_block_groups(fs_info);
  2450. fail_tree_roots:
  2451. free_root_pointers(fs_info, 1);
  2452. invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
  2453. fail_sb_buffer:
  2454. btrfs_stop_all_workers(fs_info);
  2455. fail_alloc:
  2456. fail_iput:
  2457. btrfs_mapping_tree_free(&fs_info->mapping_tree);
  2458. iput(fs_info->btree_inode);
  2459. fail_delalloc_bytes:
  2460. percpu_counter_destroy(&fs_info->delalloc_bytes);
  2461. fail_dirty_metadata_bytes:
  2462. percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
  2463. fail_bdi:
  2464. bdi_destroy(&fs_info->bdi);
  2465. fail_srcu:
  2466. cleanup_srcu_struct(&fs_info->subvol_srcu);
  2467. fail:
  2468. btrfs_free_stripe_hash_table(fs_info);
  2469. btrfs_close_devices(fs_info->fs_devices);
  2470. return err;
  2471. recovery_tree_root:
  2472. if (!btrfs_test_opt(tree_root, RECOVERY))
  2473. goto fail_tree_roots;
  2474. free_root_pointers(fs_info, 0);
  2475. /* don't use the log in recovery mode, it won't be valid */
  2476. btrfs_set_super_log_root(disk_super, 0);
  2477. /* we can't trust the free space cache either */
  2478. btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
  2479. ret = next_root_backup(fs_info, fs_info->super_copy,
  2480. &num_backups_tried, &backup_index);
  2481. if (ret == -1)
  2482. goto fail_block_groups;
  2483. goto retry_root_backup;
  2484. }
  2485. static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
  2486. {
  2487. if (uptodate) {
  2488. set_buffer_uptodate(bh);
  2489. } else {
  2490. struct btrfs_device *device = (struct btrfs_device *)
  2491. bh->b_private;
  2492. printk_ratelimited_in_rcu(KERN_WARNING "lost page write due to "
  2493. "I/O error on %s\n",
  2494. rcu_str_deref(device->name));
  2495. /* note, we dont' set_buffer_write_io_error because we have
  2496. * our own ways of dealing with the IO errors
  2497. */
  2498. clear_buffer_uptodate(bh);
  2499. btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
  2500. }
  2501. unlock_buffer(bh);
  2502. put_bh(bh);
  2503. }
  2504. struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
  2505. {
  2506. struct buffer_head *bh;
  2507. struct buffer_head *latest = NULL;
  2508. struct btrfs_super_block *super;
  2509. int i;
  2510. u64 transid = 0;
  2511. u64 bytenr;
  2512. /* we would like to check all the supers, but that would make
  2513. * a btrfs mount succeed after a mkfs from a different FS.
  2514. * So, we need to add a special mount option to scan for
  2515. * later supers, using BTRFS_SUPER_MIRROR_MAX instead
  2516. */
  2517. for (i = 0; i < 1; i++) {
  2518. bytenr = btrfs_sb_offset(i);
  2519. if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
  2520. break;
  2521. bh = __bread(bdev, bytenr / 4096, 4096);
  2522. if (!bh)
  2523. continue;
  2524. super = (struct btrfs_super_block *)bh->b_data;
  2525. if (btrfs_super_bytenr(super) != bytenr ||
  2526. super->magic != cpu_to_le64(BTRFS_MAGIC)) {
  2527. brelse(bh);
  2528. continue;
  2529. }
  2530. if (!latest || btrfs_super_generation(super) > transid) {
  2531. brelse(latest);
  2532. latest = bh;
  2533. transid = btrfs_super_generation(super);
  2534. } else {
  2535. brelse(bh);
  2536. }
  2537. }
  2538. return latest;
  2539. }
  2540. /*
  2541. * this should be called twice, once with wait == 0 and
  2542. * once with wait == 1. When wait == 0 is done, all the buffer heads
  2543. * we write are pinned.
  2544. *
  2545. * They are released when wait == 1 is done.
  2546. * max_mirrors must be the same for both runs, and it indicates how
  2547. * many supers on this one device should be written.
  2548. *
  2549. * max_mirrors == 0 means to write them all.
  2550. */
  2551. static int write_dev_supers(struct btrfs_device *device,
  2552. struct btrfs_super_block *sb,
  2553. int do_barriers, int wait, int max_mirrors)
  2554. {
  2555. struct buffer_head *bh;
  2556. int i;
  2557. int ret;
  2558. int errors = 0;
  2559. u32 crc;
  2560. u64 bytenr;
  2561. if (max_mirrors == 0)
  2562. max_mirrors = BTRFS_SUPER_MIRROR_MAX;
  2563. for (i = 0; i < max_mirrors; i++) {
  2564. bytenr = btrfs_sb_offset(i);
  2565. if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
  2566. break;
  2567. if (wait) {
  2568. bh = __find_get_block(device->bdev, bytenr / 4096,
  2569. BTRFS_SUPER_INFO_SIZE);
  2570. if (!bh) {
  2571. errors++;
  2572. continue;
  2573. }
  2574. wait_on_buffer(bh);
  2575. if (!buffer_uptodate(bh))
  2576. errors++;
  2577. /* drop our reference */
  2578. brelse(bh);
  2579. /* drop the reference from the wait == 0 run */
  2580. brelse(bh);
  2581. continue;
  2582. } else {
  2583. btrfs_set_super_bytenr(sb, bytenr);
  2584. crc = ~(u32)0;
  2585. crc = btrfs_csum_data((char *)sb +
  2586. BTRFS_CSUM_SIZE, crc,
  2587. BTRFS_SUPER_INFO_SIZE -
  2588. BTRFS_CSUM_SIZE);
  2589. btrfs_csum_final(crc, sb->csum);
  2590. /*
  2591. * one reference for us, and we leave it for the
  2592. * caller
  2593. */
  2594. bh = __getblk(device->bdev, bytenr / 4096,
  2595. BTRFS_SUPER_INFO_SIZE);
  2596. if (!bh) {
  2597. printk(KERN_ERR "btrfs: couldn't get super "
  2598. "buffer head for bytenr %Lu\n", bytenr);
  2599. errors++;
  2600. continue;
  2601. }
  2602. memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
  2603. /* one reference for submit_bh */
  2604. get_bh(bh);
  2605. set_buffer_uptodate(bh);
  2606. lock_buffer(bh);
  2607. bh->b_end_io = btrfs_end_buffer_write_sync;
  2608. bh->b_private = device;
  2609. }
  2610. /*
  2611. * we fua the first super. The others we allow
  2612. * to go down lazy.
  2613. */
  2614. ret = btrfsic_submit_bh(WRITE_FUA, bh);
  2615. if (ret)
  2616. errors++;
  2617. }
  2618. return errors < i ? 0 : -1;
  2619. }
  2620. /*
  2621. * endio for the write_dev_flush, this will wake anyone waiting
  2622. * for the barrier when it is done
  2623. */
  2624. static void btrfs_end_empty_barrier(struct bio *bio, int err)
  2625. {
  2626. if (err) {
  2627. if (err == -EOPNOTSUPP)
  2628. set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
  2629. clear_bit(BIO_UPTODATE, &bio->bi_flags);
  2630. }
  2631. if (bio->bi_private)
  2632. complete(bio->bi_private);
  2633. bio_put(bio);
  2634. }
  2635. /*
  2636. * trigger flushes for one the devices. If you pass wait == 0, the flushes are
  2637. * sent down. With wait == 1, it waits for the previous flush.
  2638. *
  2639. * any device where the flush fails with eopnotsupp are flagged as not-barrier
  2640. * capable
  2641. */
  2642. static int write_dev_flush(struct btrfs_device *device, int wait)
  2643. {
  2644. struct bio *bio;
  2645. int ret = 0;
  2646. if (device->nobarriers)
  2647. return 0;
  2648. if (wait) {
  2649. bio = device->flush_bio;
  2650. if (!bio)
  2651. return 0;
  2652. wait_for_completion(&device->flush_wait);
  2653. if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
  2654. printk_in_rcu("btrfs: disabling barriers on dev %s\n",
  2655. rcu_str_deref(device->name));
  2656. device->nobarriers = 1;
  2657. } else if (!bio_flagged(bio, BIO_UPTODATE)) {
  2658. ret = -EIO;
  2659. btrfs_dev_stat_inc_and_print(device,
  2660. BTRFS_DEV_STAT_FLUSH_ERRS);
  2661. }
  2662. /* drop the reference from the wait == 0 run */
  2663. bio_put(bio);
  2664. device->flush_bio = NULL;
  2665. return ret;
  2666. }
  2667. /*
  2668. * one reference for us, and we leave it for the
  2669. * caller
  2670. */
  2671. device->flush_bio = NULL;
  2672. bio = bio_alloc(GFP_NOFS, 0);
  2673. if (!bio)
  2674. return -ENOMEM;
  2675. bio->bi_end_io = btrfs_end_empty_barrier;
  2676. bio->bi_bdev = device->bdev;
  2677. init_completion(&device->flush_wait);
  2678. bio->bi_private = &device->flush_wait;
  2679. device->flush_bio = bio;
  2680. bio_get(bio);
  2681. btrfsic_submit_bio(WRITE_FLUSH, bio);
  2682. return 0;
  2683. }
  2684. /*
  2685. * send an empty flush down to each device in parallel,
  2686. * then wait for them
  2687. */
  2688. static int barrier_all_devices(struct btrfs_fs_info *info)
  2689. {
  2690. struct list_head *head;
  2691. struct btrfs_device *dev;
  2692. int errors_send = 0;
  2693. int errors_wait = 0;
  2694. int ret;
  2695. /* send down all the barriers */
  2696. head = &info->fs_devices->devices;
  2697. list_for_each_entry_rcu(dev, head, dev_list) {
  2698. if (!dev->bdev) {
  2699. errors_send++;
  2700. continue;
  2701. }
  2702. if (!dev->in_fs_metadata || !dev->writeable)
  2703. continue;
  2704. ret = write_dev_flush(dev, 0);
  2705. if (ret)
  2706. errors_send++;
  2707. }
  2708. /* wait for all the barriers */
  2709. list_for_each_entry_rcu(dev, head, dev_list) {
  2710. if (!dev->bdev) {
  2711. errors_wait++;
  2712. continue;
  2713. }
  2714. if (!dev->in_fs_metadata || !dev->writeable)
  2715. continue;
  2716. ret = write_dev_flush(dev, 1);
  2717. if (ret)
  2718. errors_wait++;
  2719. }
  2720. if (errors_send > info->num_tolerated_disk_barrier_failures ||
  2721. errors_wait > info->num_tolerated_disk_barrier_failures)
  2722. return -EIO;
  2723. return 0;
  2724. }
  2725. int btrfs_calc_num_tolerated_disk_barrier_failures(
  2726. struct btrfs_fs_info *fs_info)
  2727. {
  2728. struct btrfs_ioctl_space_info space;
  2729. struct btrfs_space_info *sinfo;
  2730. u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
  2731. BTRFS_BLOCK_GROUP_SYSTEM,
  2732. BTRFS_BLOCK_GROUP_METADATA,
  2733. BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
  2734. int num_types = 4;
  2735. int i;
  2736. int c;
  2737. int num_tolerated_disk_barrier_failures =
  2738. (int)fs_info->fs_devices->num_devices;
  2739. for (i = 0; i < num_types; i++) {
  2740. struct btrfs_space_info *tmp;
  2741. sinfo = NULL;
  2742. rcu_read_lock();
  2743. list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
  2744. if (tmp->flags == types[i]) {
  2745. sinfo = tmp;
  2746. break;
  2747. }
  2748. }
  2749. rcu_read_unlock();
  2750. if (!sinfo)
  2751. continue;
  2752. down_read(&sinfo->groups_sem);
  2753. for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
  2754. if (!list_empty(&sinfo->block_groups[c])) {
  2755. u64 flags;
  2756. btrfs_get_block_group_info(
  2757. &sinfo->block_groups[c], &space);
  2758. if (space.total_bytes == 0 ||
  2759. space.used_bytes == 0)
  2760. continue;
  2761. flags = space.flags;
  2762. /*
  2763. * return
  2764. * 0: if dup, single or RAID0 is configured for
  2765. * any of metadata, system or data, else
  2766. * 1: if RAID5 is configured, or if RAID1 or
  2767. * RAID10 is configured and only two mirrors
  2768. * are used, else
  2769. * 2: if RAID6 is configured, else
  2770. * num_mirrors - 1: if RAID1 or RAID10 is
  2771. * configured and more than
  2772. * 2 mirrors are used.
  2773. */
  2774. if (num_tolerated_disk_barrier_failures > 0 &&
  2775. ((flags & (BTRFS_BLOCK_GROUP_DUP |
  2776. BTRFS_BLOCK_GROUP_RAID0)) ||
  2777. ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)
  2778. == 0)))
  2779. num_tolerated_disk_barrier_failures = 0;
  2780. else if (num_tolerated_disk_barrier_failures > 1) {
  2781. if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
  2782. BTRFS_BLOCK_GROUP_RAID5 |
  2783. BTRFS_BLOCK_GROUP_RAID10)) {
  2784. num_tolerated_disk_barrier_failures = 1;
  2785. } else if (flags &
  2786. BTRFS_BLOCK_GROUP_RAID5) {
  2787. num_tolerated_disk_barrier_failures = 2;
  2788. }
  2789. }
  2790. }
  2791. }
  2792. up_read(&sinfo->groups_sem);
  2793. }
  2794. return num_tolerated_disk_barrier_failures;
  2795. }
  2796. static int write_all_supers(struct btrfs_root *root, int max_mirrors)
  2797. {
  2798. struct list_head *head;
  2799. struct btrfs_device *dev;
  2800. struct btrfs_super_block *sb;
  2801. struct btrfs_dev_item *dev_item;
  2802. int ret;
  2803. int do_barriers;
  2804. int max_errors;
  2805. int total_errors = 0;
  2806. u64 flags;
  2807. max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
  2808. do_barriers = !btrfs_test_opt(root, NOBARRIER);
  2809. backup_super_roots(root->fs_info);
  2810. sb = root->fs_info->super_for_commit;
  2811. dev_item = &sb->dev_item;
  2812. mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
  2813. head = &root->fs_info->fs_devices->devices;
  2814. if (do_barriers) {
  2815. ret = barrier_all_devices(root->fs_info);
  2816. if (ret) {
  2817. mutex_unlock(
  2818. &root->fs_info->fs_devices->device_list_mutex);
  2819. btrfs_error(root->fs_info, ret,
  2820. "errors while submitting device barriers.");
  2821. return ret;
  2822. }
  2823. }
  2824. list_for_each_entry_rcu(dev, head, dev_list) {
  2825. if (!dev->bdev) {
  2826. total_errors++;
  2827. continue;
  2828. }
  2829. if (!dev->in_fs_metadata || !dev->writeable)
  2830. continue;
  2831. btrfs_set_stack_device_generation(dev_item, 0);
  2832. btrfs_set_stack_device_type(dev_item, dev->type);
  2833. btrfs_set_stack_device_id(dev_item, dev->devid);
  2834. btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
  2835. btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
  2836. btrfs_set_stack_device_io_align(dev_item, dev->io_align);
  2837. btrfs_set_stack_device_io_width(dev_item, dev->io_width);
  2838. btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
  2839. memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
  2840. memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
  2841. flags = btrfs_super_flags(sb);
  2842. btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
  2843. ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
  2844. if (ret)
  2845. total_errors++;
  2846. }
  2847. if (total_errors > max_errors) {
  2848. printk(KERN_ERR "btrfs: %d errors while writing supers\n",
  2849. total_errors);
  2850. /* This shouldn't happen. FUA is masked off if unsupported */
  2851. BUG();
  2852. }
  2853. total_errors = 0;
  2854. list_for_each_entry_rcu(dev, head, dev_list) {
  2855. if (!dev->bdev)
  2856. continue;
  2857. if (!dev->in_fs_metadata || !dev->writeable)
  2858. continue;
  2859. ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
  2860. if (ret)
  2861. total_errors++;
  2862. }
  2863. mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
  2864. if (total_errors > max_errors) {
  2865. btrfs_error(root->fs_info, -EIO,
  2866. "%d errors while writing supers", total_errors);
  2867. return -EIO;
  2868. }
  2869. return 0;
  2870. }
  2871. int write_ctree_super(struct btrfs_trans_handle *trans,
  2872. struct btrfs_root *root, int max_mirrors)
  2873. {
  2874. int ret;
  2875. ret = write_all_supers(root, max_mirrors);
  2876. return ret;
  2877. }
  2878. void btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
  2879. {
  2880. spin_lock(&fs_info->fs_roots_radix_lock);
  2881. radix_tree_delete(&fs_info->fs_roots_radix,
  2882. (unsigned long)root->root_key.objectid);
  2883. spin_unlock(&fs_info->fs_roots_radix_lock);
  2884. if (btrfs_root_refs(&root->root_item) == 0)
  2885. synchronize_srcu(&fs_info->subvol_srcu);
  2886. if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
  2887. btrfs_free_log(NULL, root);
  2888. btrfs_free_log_root_tree(NULL, fs_info);
  2889. }
  2890. __btrfs_remove_free_space_cache(root->free_ino_pinned);
  2891. __btrfs_remove_free_space_cache(root->free_ino_ctl);
  2892. free_fs_root(root);
  2893. }
  2894. static void free_fs_root(struct btrfs_root *root)
  2895. {
  2896. iput(root->cache_inode);
  2897. WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
  2898. if (root->anon_dev)
  2899. free_anon_bdev(root->anon_dev);
  2900. free_extent_buffer(root->node);
  2901. free_extent_buffer(root->commit_root);
  2902. kfree(root->free_ino_ctl);
  2903. kfree(root->free_ino_pinned);
  2904. kfree(root->name);
  2905. kfree(root);
  2906. }
  2907. int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
  2908. {
  2909. u64 root_objectid = 0;
  2910. struct btrfs_root *gang[8];
  2911. int i;
  2912. int ret;
  2913. while (1) {
  2914. ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
  2915. (void **)gang, root_objectid,
  2916. ARRAY_SIZE(gang));
  2917. if (!ret)
  2918. break;
  2919. root_objectid = gang[ret - 1]->root_key.objectid + 1;
  2920. for (i = 0; i < ret; i++) {
  2921. int err;
  2922. root_objectid = gang[i]->root_key.objectid;
  2923. err = btrfs_orphan_cleanup(gang[i]);
  2924. if (err)
  2925. return err;
  2926. }
  2927. root_objectid++;
  2928. }
  2929. return 0;
  2930. }
  2931. int btrfs_commit_super(struct btrfs_root *root)
  2932. {
  2933. struct btrfs_trans_handle *trans;
  2934. int ret;
  2935. mutex_lock(&root->fs_info->cleaner_mutex);
  2936. btrfs_run_delayed_iputs(root);
  2937. mutex_unlock(&root->fs_info->cleaner_mutex);
  2938. wake_up_process(root->fs_info->cleaner_kthread);
  2939. /* wait until ongoing cleanup work done */
  2940. down_write(&root->fs_info->cleanup_work_sem);
  2941. up_write(&root->fs_info->cleanup_work_sem);
  2942. trans = btrfs_join_transaction(root);
  2943. if (IS_ERR(trans))
  2944. return PTR_ERR(trans);
  2945. ret = btrfs_commit_transaction(trans, root);
  2946. if (ret)
  2947. return ret;
  2948. /* run commit again to drop the original snapshot */
  2949. trans = btrfs_join_transaction(root);
  2950. if (IS_ERR(trans))
  2951. return PTR_ERR(trans);
  2952. ret = btrfs_commit_transaction(trans, root);
  2953. if (ret)
  2954. return ret;
  2955. ret = btrfs_write_and_wait_transaction(NULL, root);
  2956. if (ret) {
  2957. btrfs_error(root->fs_info, ret,
  2958. "Failed to sync btree inode to disk.");
  2959. return ret;
  2960. }
  2961. ret = write_ctree_super(NULL, root, 0);
  2962. return ret;
  2963. }
  2964. int close_ctree(struct btrfs_root *root)
  2965. {
  2966. struct btrfs_fs_info *fs_info = root->fs_info;
  2967. int ret;
  2968. fs_info->closing = 1;
  2969. smp_mb();
  2970. /* pause restriper - we want to resume on mount */
  2971. btrfs_pause_balance(fs_info);
  2972. btrfs_dev_replace_suspend_for_unmount(fs_info);
  2973. btrfs_scrub_cancel(fs_info);
  2974. /* wait for any defraggers to finish */
  2975. wait_event(fs_info->transaction_wait,
  2976. (atomic_read(&fs_info->defrag_running) == 0));
  2977. /* clear out the rbtree of defraggable inodes */
  2978. btrfs_cleanup_defrag_inodes(fs_info);
  2979. if (!(fs_info->sb->s_flags & MS_RDONLY)) {
  2980. ret = btrfs_commit_super(root);
  2981. if (ret)
  2982. printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
  2983. }
  2984. if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
  2985. btrfs_error_commit_super(root);
  2986. btrfs_put_block_group_cache(fs_info);
  2987. kthread_stop(fs_info->transaction_kthread);
  2988. kthread_stop(fs_info->cleaner_kthread);
  2989. fs_info->closing = 2;
  2990. smp_mb();
  2991. btrfs_free_qgroup_config(root->fs_info);
  2992. if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
  2993. printk(KERN_INFO "btrfs: at unmount delalloc count %lld\n",
  2994. percpu_counter_sum(&fs_info->delalloc_bytes));
  2995. }
  2996. free_root_pointers(fs_info, 1);
  2997. btrfs_free_block_groups(fs_info);
  2998. del_fs_roots(fs_info);
  2999. iput(fs_info->btree_inode);
  3000. btrfs_stop_all_workers(fs_info);
  3001. #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
  3002. if (btrfs_test_opt(root, CHECK_INTEGRITY))
  3003. btrfsic_unmount(root, fs_info->fs_devices);
  3004. #endif
  3005. btrfs_close_devices(fs_info->fs_devices);
  3006. btrfs_mapping_tree_free(&fs_info->mapping_tree);
  3007. percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
  3008. percpu_counter_destroy(&fs_info->delalloc_bytes);
  3009. bdi_destroy(&fs_info->bdi);
  3010. cleanup_srcu_struct(&fs_info->subvol_srcu);
  3011. btrfs_free_stripe_hash_table(fs_info);
  3012. return 0;
  3013. }
  3014. int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
  3015. int atomic)
  3016. {
  3017. int ret;
  3018. struct inode *btree_inode = buf->pages[0]->mapping->host;
  3019. ret = extent_buffer_uptodate(buf);
  3020. if (!ret)
  3021. return ret;
  3022. ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
  3023. parent_transid, atomic);
  3024. if (ret == -EAGAIN)
  3025. return ret;
  3026. return !ret;
  3027. }
  3028. int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
  3029. {
  3030. return set_extent_buffer_uptodate(buf);
  3031. }
  3032. void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
  3033. {
  3034. struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
  3035. u64 transid = btrfs_header_generation(buf);
  3036. int was_dirty;
  3037. btrfs_assert_tree_locked(buf);
  3038. if (transid != root->fs_info->generation)
  3039. WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
  3040. "found %llu running %llu\n",
  3041. (unsigned long long)buf->start,
  3042. (unsigned long long)transid,
  3043. (unsigned long long)root->fs_info->generation);
  3044. was_dirty = set_extent_buffer_dirty(buf);
  3045. if (!was_dirty)
  3046. __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
  3047. buf->len,
  3048. root->fs_info->dirty_metadata_batch);
  3049. }
  3050. static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
  3051. int flush_delayed)
  3052. {
  3053. /*
  3054. * looks as though older kernels can get into trouble with
  3055. * this code, they end up stuck in balance_dirty_pages forever
  3056. */
  3057. int ret;
  3058. if (current->flags & PF_MEMALLOC)
  3059. return;
  3060. if (flush_delayed)
  3061. btrfs_balance_delayed_items(root);
  3062. ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
  3063. BTRFS_DIRTY_METADATA_THRESH);
  3064. if (ret > 0) {
  3065. balance_dirty_pages_ratelimited(
  3066. root->fs_info->btree_inode->i_mapping);
  3067. }
  3068. return;
  3069. }
  3070. void btrfs_btree_balance_dirty(struct btrfs_root *root)
  3071. {
  3072. __btrfs_btree_balance_dirty(root, 1);
  3073. }
  3074. void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
  3075. {
  3076. __btrfs_btree_balance_dirty(root, 0);
  3077. }
  3078. int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
  3079. {
  3080. struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
  3081. return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
  3082. }
  3083. static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
  3084. int read_only)
  3085. {
  3086. if (btrfs_super_csum_type(fs_info->super_copy) >= ARRAY_SIZE(btrfs_csum_sizes)) {
  3087. printk(KERN_ERR "btrfs: unsupported checksum algorithm\n");
  3088. return -EINVAL;
  3089. }
  3090. if (read_only)
  3091. return 0;
  3092. return 0;
  3093. }
  3094. static void btrfs_error_commit_super(struct btrfs_root *root)
  3095. {
  3096. mutex_lock(&root->fs_info->cleaner_mutex);
  3097. btrfs_run_delayed_iputs(root);
  3098. mutex_unlock(&root->fs_info->cleaner_mutex);
  3099. down_write(&root->fs_info->cleanup_work_sem);
  3100. up_write(&root->fs_info->cleanup_work_sem);
  3101. /* cleanup FS via transaction */
  3102. btrfs_cleanup_transaction(root);
  3103. }
  3104. static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
  3105. struct btrfs_root *root)
  3106. {
  3107. struct btrfs_inode *btrfs_inode;
  3108. struct list_head splice;
  3109. INIT_LIST_HEAD(&splice);
  3110. mutex_lock(&root->fs_info->ordered_operations_mutex);
  3111. spin_lock(&root->fs_info->ordered_extent_lock);
  3112. list_splice_init(&t->ordered_operations, &splice);
  3113. while (!list_empty(&splice)) {
  3114. btrfs_inode = list_entry(splice.next, struct btrfs_inode,
  3115. ordered_operations);
  3116. list_del_init(&btrfs_inode->ordered_operations);
  3117. btrfs_invalidate_inodes(btrfs_inode->root);
  3118. }
  3119. spin_unlock(&root->fs_info->ordered_extent_lock);
  3120. mutex_unlock(&root->fs_info->ordered_operations_mutex);
  3121. }
  3122. static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
  3123. {
  3124. struct btrfs_ordered_extent *ordered;
  3125. spin_lock(&root->fs_info->ordered_extent_lock);
  3126. /*
  3127. * This will just short circuit the ordered completion stuff which will
  3128. * make sure the ordered extent gets properly cleaned up.
  3129. */
  3130. list_for_each_entry(ordered, &root->fs_info->ordered_extents,
  3131. root_extent_list)
  3132. set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
  3133. spin_unlock(&root->fs_info->ordered_extent_lock);
  3134. }
  3135. int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
  3136. struct btrfs_root *root)
  3137. {
  3138. struct rb_node *node;
  3139. struct btrfs_delayed_ref_root *delayed_refs;
  3140. struct btrfs_delayed_ref_node *ref;
  3141. int ret = 0;
  3142. delayed_refs = &trans->delayed_refs;
  3143. spin_lock(&delayed_refs->lock);
  3144. if (delayed_refs->num_entries == 0) {
  3145. spin_unlock(&delayed_refs->lock);
  3146. printk(KERN_INFO "delayed_refs has NO entry\n");
  3147. return ret;
  3148. }
  3149. while ((node = rb_first(&delayed_refs->root)) != NULL) {
  3150. struct btrfs_delayed_ref_head *head = NULL;
  3151. ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
  3152. atomic_set(&ref->refs, 1);
  3153. if (btrfs_delayed_ref_is_head(ref)) {
  3154. head = btrfs_delayed_node_to_head(ref);
  3155. if (!mutex_trylock(&head->mutex)) {
  3156. atomic_inc(&ref->refs);
  3157. spin_unlock(&delayed_refs->lock);
  3158. /* Need to wait for the delayed ref to run */
  3159. mutex_lock(&head->mutex);
  3160. mutex_unlock(&head->mutex);
  3161. btrfs_put_delayed_ref(ref);
  3162. spin_lock(&delayed_refs->lock);
  3163. continue;
  3164. }
  3165. if (head->must_insert_reserved)
  3166. btrfs_pin_extent(root, ref->bytenr,
  3167. ref->num_bytes, 1);
  3168. btrfs_free_delayed_extent_op(head->extent_op);
  3169. delayed_refs->num_heads--;
  3170. if (list_empty(&head->cluster))
  3171. delayed_refs->num_heads_ready--;
  3172. list_del_init(&head->cluster);
  3173. }
  3174. ref->in_tree = 0;
  3175. rb_erase(&ref->rb_node, &delayed_refs->root);
  3176. delayed_refs->num_entries--;
  3177. if (head)
  3178. mutex_unlock(&head->mutex);
  3179. spin_unlock(&delayed_refs->lock);
  3180. btrfs_put_delayed_ref(ref);
  3181. cond_resched();
  3182. spin_lock(&delayed_refs->lock);
  3183. }
  3184. spin_unlock(&delayed_refs->lock);
  3185. return ret;
  3186. }
  3187. static void btrfs_evict_pending_snapshots(struct btrfs_transaction *t)
  3188. {
  3189. struct btrfs_pending_snapshot *snapshot;
  3190. struct list_head splice;
  3191. INIT_LIST_HEAD(&splice);
  3192. list_splice_init(&t->pending_snapshots, &splice);
  3193. while (!list_empty(&splice)) {
  3194. snapshot = list_entry(splice.next,
  3195. struct btrfs_pending_snapshot,
  3196. list);
  3197. snapshot->error = -ECANCELED;
  3198. list_del_init(&snapshot->list);
  3199. }
  3200. }
  3201. static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
  3202. {
  3203. struct btrfs_inode *btrfs_inode;
  3204. struct list_head splice;
  3205. INIT_LIST_HEAD(&splice);
  3206. spin_lock(&root->fs_info->delalloc_lock);
  3207. list_splice_init(&root->fs_info->delalloc_inodes, &splice);
  3208. while (!list_empty(&splice)) {
  3209. btrfs_inode = list_entry(splice.next, struct btrfs_inode,
  3210. delalloc_inodes);
  3211. list_del_init(&btrfs_inode->delalloc_inodes);
  3212. clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
  3213. &btrfs_inode->runtime_flags);
  3214. btrfs_invalidate_inodes(btrfs_inode->root);
  3215. }
  3216. spin_unlock(&root->fs_info->delalloc_lock);
  3217. }
  3218. static int btrfs_destroy_marked_extents(struct btrfs_root *root,
  3219. struct extent_io_tree *dirty_pages,
  3220. int mark)
  3221. {
  3222. int ret;
  3223. struct extent_buffer *eb;
  3224. u64 start = 0;
  3225. u64 end;
  3226. while (1) {
  3227. ret = find_first_extent_bit(dirty_pages, start, &start, &end,
  3228. mark, NULL);
  3229. if (ret)
  3230. break;
  3231. clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
  3232. while (start <= end) {
  3233. eb = btrfs_find_tree_block(root, start,
  3234. root->leafsize);
  3235. start += eb->len;
  3236. if (!eb)
  3237. continue;
  3238. wait_on_extent_buffer_writeback(eb);
  3239. if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
  3240. &eb->bflags))
  3241. clear_extent_buffer_dirty(eb);
  3242. free_extent_buffer_stale(eb);
  3243. }
  3244. }
  3245. return ret;
  3246. }
  3247. static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
  3248. struct extent_io_tree *pinned_extents)
  3249. {
  3250. struct extent_io_tree *unpin;
  3251. u64 start;
  3252. u64 end;
  3253. int ret;
  3254. bool loop = true;
  3255. unpin = pinned_extents;
  3256. again:
  3257. while (1) {
  3258. ret = find_first_extent_bit(unpin, 0, &start, &end,
  3259. EXTENT_DIRTY, NULL);
  3260. if (ret)
  3261. break;
  3262. /* opt_discard */
  3263. if (btrfs_test_opt(root, DISCARD))
  3264. ret = btrfs_error_discard_extent(root, start,
  3265. end + 1 - start,
  3266. NULL);
  3267. clear_extent_dirty(unpin, start, end, GFP_NOFS);
  3268. btrfs_error_unpin_extent_range(root, start, end);
  3269. cond_resched();
  3270. }
  3271. if (loop) {
  3272. if (unpin == &root->fs_info->freed_extents[0])
  3273. unpin = &root->fs_info->freed_extents[1];
  3274. else
  3275. unpin = &root->fs_info->freed_extents[0];
  3276. loop = false;
  3277. goto again;
  3278. }
  3279. return 0;
  3280. }
  3281. void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
  3282. struct btrfs_root *root)
  3283. {
  3284. btrfs_destroy_delayed_refs(cur_trans, root);
  3285. btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv,
  3286. cur_trans->dirty_pages.dirty_bytes);
  3287. /* FIXME: cleanup wait for commit */
  3288. cur_trans->in_commit = 1;
  3289. cur_trans->blocked = 1;
  3290. wake_up(&root->fs_info->transaction_blocked_wait);
  3291. btrfs_evict_pending_snapshots(cur_trans);
  3292. cur_trans->blocked = 0;
  3293. wake_up(&root->fs_info->transaction_wait);
  3294. cur_trans->commit_done = 1;
  3295. wake_up(&cur_trans->commit_wait);
  3296. btrfs_destroy_delayed_inodes(root);
  3297. btrfs_assert_delayed_root_empty(root);
  3298. btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
  3299. EXTENT_DIRTY);
  3300. btrfs_destroy_pinned_extent(root,
  3301. root->fs_info->pinned_extents);
  3302. /*
  3303. memset(cur_trans, 0, sizeof(*cur_trans));
  3304. kmem_cache_free(btrfs_transaction_cachep, cur_trans);
  3305. */
  3306. }
  3307. static int btrfs_cleanup_transaction(struct btrfs_root *root)
  3308. {
  3309. struct btrfs_transaction *t;
  3310. LIST_HEAD(list);
  3311. mutex_lock(&root->fs_info->transaction_kthread_mutex);
  3312. spin_lock(&root->fs_info->trans_lock);
  3313. list_splice_init(&root->fs_info->trans_list, &list);
  3314. root->fs_info->trans_no_join = 1;
  3315. spin_unlock(&root->fs_info->trans_lock);
  3316. while (!list_empty(&list)) {
  3317. t = list_entry(list.next, struct btrfs_transaction, list);
  3318. btrfs_destroy_ordered_operations(t, root);
  3319. btrfs_destroy_ordered_extents(root);
  3320. btrfs_destroy_delayed_refs(t, root);
  3321. /* FIXME: cleanup wait for commit */
  3322. t->in_commit = 1;
  3323. t->blocked = 1;
  3324. smp_mb();
  3325. if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
  3326. wake_up(&root->fs_info->transaction_blocked_wait);
  3327. btrfs_evict_pending_snapshots(t);
  3328. t->blocked = 0;
  3329. smp_mb();
  3330. if (waitqueue_active(&root->fs_info->transaction_wait))
  3331. wake_up(&root->fs_info->transaction_wait);
  3332. t->commit_done = 1;
  3333. smp_mb();
  3334. if (waitqueue_active(&t->commit_wait))
  3335. wake_up(&t->commit_wait);
  3336. btrfs_destroy_delayed_inodes(root);
  3337. btrfs_assert_delayed_root_empty(root);
  3338. btrfs_destroy_delalloc_inodes(root);
  3339. spin_lock(&root->fs_info->trans_lock);
  3340. root->fs_info->running_transaction = NULL;
  3341. spin_unlock(&root->fs_info->trans_lock);
  3342. btrfs_destroy_marked_extents(root, &t->dirty_pages,
  3343. EXTENT_DIRTY);
  3344. btrfs_destroy_pinned_extent(root,
  3345. root->fs_info->pinned_extents);
  3346. atomic_set(&t->use_count, 0);
  3347. list_del_init(&t->list);
  3348. memset(t, 0, sizeof(*t));
  3349. kmem_cache_free(btrfs_transaction_cachep, t);
  3350. }
  3351. spin_lock(&root->fs_info->trans_lock);
  3352. root->fs_info->trans_no_join = 0;
  3353. spin_unlock(&root->fs_info->trans_lock);
  3354. mutex_unlock(&root->fs_info->transaction_kthread_mutex);
  3355. return 0;
  3356. }
  3357. static struct extent_io_ops btree_extent_io_ops = {
  3358. .readpage_end_io_hook = btree_readpage_end_io_hook,
  3359. .readpage_io_failed_hook = btree_io_failed_hook,
  3360. .submit_bio_hook = btree_submit_bio_hook,
  3361. /* note we're sharing with inode.c for the merge bio hook */
  3362. .merge_bio_hook = btrfs_merge_bio_hook,
  3363. };