xfs_log.c 101 KB

1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162636465666768697071727374757677787980818283848586878889909192939495969798991001011021031041051061071081091101111121131141151161171181191201211221231241251261271281291301311321331341351361371381391401411421431441451461471481491501511521531541551561571581591601611621631641651661671681691701711721731741751761771781791801811821831841851861871881891901911921931941951961971981992002012022032042052062072082092102112122132142152162172182192202212222232242252262272282292302312322332342352362372382392402412422432442452462472482492502512522532542552562572582592602612622632642652662672682692702712722732742752762772782792802812822832842852862872882892902912922932942952962972982993003013023033043053063073083093103113123133143153163173183193203213223233243253263273283293303313323333343353363373383393403413423433443453463473483493503513523533543553563573583593603613623633643653663673683693703713723733743753763773783793803813823833843853863873883893903913923933943953963973983994004014024034044054064074084094104114124134144154164174184194204214224234244254264274284294304314324334344354364374384394404414424434444454464474484494504514524534544554564574584594604614624634644654664674684694704714724734744754764774784794804814824834844854864874884894904914924934944954964974984995005015025035045055065075085095105115125135145155165175185195205215225235245255265275285295305315325335345355365375385395405415425435445455465475485495505515525535545555565575585595605615625635645655665675685695705715725735745755765775785795805815825835845855865875885895905915925935945955965975985996006016026036046056066076086096106116126136146156166176186196206216226236246256266276286296306316326336346356366376386396406416426436446456466476486496506516526536546556566576586596606616626636646656666676686696706716726736746756766776786796806816826836846856866876886896906916926936946956966976986997007017027037047057067077087097107117127137147157167177187197207217227237247257267277287297307317327337347357367377387397407417427437447457467477487497507517527537547557567577587597607617627637647657667677687697707717727737747757767777787797807817827837847857867877887897907917927937947957967977987998008018028038048058068078088098108118128138148158168178188198208218228238248258268278288298308318328338348358368378388398408418428438448458468478488498508518528538548558568578588598608618628638648658668678688698708718728738748758768778788798808818828838848858868878888898908918928938948958968978988999009019029039049059069079089099109119129139149159169179189199209219229239249259269279289299309319329339349359369379389399409419429439449459469479489499509519529539549559569579589599609619629639649659669679689699709719729739749759769779789799809819829839849859869879889899909919929939949959969979989991000100110021003100410051006100710081009101010111012101310141015101610171018101910201021102210231024102510261027102810291030103110321033103410351036103710381039104010411042104310441045104610471048104910501051105210531054105510561057105810591060106110621063106410651066106710681069107010711072107310741075107610771078107910801081108210831084108510861087108810891090109110921093109410951096109710981099110011011102110311041105110611071108110911101111111211131114111511161117111811191120112111221123112411251126112711281129113011311132113311341135113611371138113911401141114211431144114511461147114811491150115111521153115411551156115711581159116011611162116311641165116611671168116911701171117211731174117511761177117811791180118111821183118411851186118711881189119011911192119311941195119611971198119912001201120212031204120512061207120812091210121112121213121412151216121712181219122012211222122312241225122612271228122912301231123212331234123512361237123812391240124112421243124412451246124712481249125012511252125312541255125612571258125912601261126212631264126512661267126812691270127112721273127412751276127712781279128012811282128312841285128612871288128912901291129212931294129512961297129812991300130113021303130413051306130713081309131013111312131313141315131613171318131913201321132213231324132513261327132813291330133113321333133413351336133713381339134013411342134313441345134613471348134913501351135213531354135513561357135813591360136113621363136413651366136713681369137013711372137313741375137613771378137913801381138213831384138513861387138813891390139113921393139413951396139713981399140014011402140314041405140614071408140914101411141214131414141514161417141814191420142114221423142414251426142714281429143014311432143314341435143614371438143914401441144214431444144514461447144814491450145114521453145414551456145714581459146014611462146314641465146614671468146914701471147214731474147514761477147814791480148114821483148414851486148714881489149014911492149314941495149614971498149915001501150215031504150515061507150815091510151115121513151415151516151715181519152015211522152315241525152615271528152915301531153215331534153515361537153815391540154115421543154415451546154715481549155015511552155315541555155615571558155915601561156215631564156515661567156815691570157115721573157415751576157715781579158015811582158315841585158615871588158915901591159215931594159515961597159815991600160116021603160416051606160716081609161016111612161316141615161616171618161916201621162216231624162516261627162816291630163116321633163416351636163716381639164016411642164316441645164616471648164916501651165216531654165516561657165816591660166116621663166416651666166716681669167016711672167316741675167616771678167916801681168216831684168516861687168816891690169116921693169416951696169716981699170017011702170317041705170617071708170917101711171217131714171517161717171817191720172117221723172417251726172717281729173017311732173317341735173617371738173917401741174217431744174517461747174817491750175117521753175417551756175717581759176017611762176317641765176617671768176917701771177217731774177517761777177817791780178117821783178417851786178717881789179017911792179317941795179617971798179918001801180218031804180518061807180818091810181118121813181418151816181718181819182018211822182318241825182618271828182918301831183218331834183518361837183818391840184118421843184418451846184718481849185018511852185318541855185618571858185918601861186218631864186518661867186818691870187118721873187418751876187718781879188018811882188318841885188618871888188918901891189218931894189518961897189818991900190119021903190419051906190719081909191019111912191319141915191619171918191919201921192219231924192519261927192819291930193119321933193419351936193719381939194019411942194319441945194619471948194919501951195219531954195519561957195819591960196119621963196419651966196719681969197019711972197319741975197619771978197919801981198219831984198519861987198819891990199119921993199419951996199719981999200020012002200320042005200620072008200920102011201220132014201520162017201820192020202120222023202420252026202720282029203020312032203320342035203620372038203920402041204220432044204520462047204820492050205120522053205420552056205720582059206020612062206320642065206620672068206920702071207220732074207520762077207820792080208120822083208420852086208720882089209020912092209320942095209620972098209921002101210221032104210521062107210821092110211121122113211421152116211721182119212021212122212321242125212621272128212921302131213221332134213521362137213821392140214121422143214421452146214721482149215021512152215321542155215621572158215921602161216221632164216521662167216821692170217121722173217421752176217721782179218021812182218321842185218621872188218921902191219221932194219521962197219821992200220122022203220422052206220722082209221022112212221322142215221622172218221922202221222222232224222522262227222822292230223122322233223422352236223722382239224022412242224322442245224622472248224922502251225222532254225522562257225822592260226122622263226422652266226722682269227022712272227322742275227622772278227922802281228222832284228522862287228822892290229122922293229422952296229722982299230023012302230323042305230623072308230923102311231223132314231523162317231823192320232123222323232423252326232723282329233023312332233323342335233623372338233923402341234223432344234523462347234823492350235123522353235423552356235723582359236023612362236323642365236623672368236923702371237223732374237523762377237823792380238123822383238423852386238723882389239023912392239323942395239623972398239924002401240224032404240524062407240824092410241124122413241424152416241724182419242024212422242324242425242624272428242924302431243224332434243524362437243824392440244124422443244424452446244724482449245024512452245324542455245624572458245924602461246224632464246524662467246824692470247124722473247424752476247724782479248024812482248324842485248624872488248924902491249224932494249524962497249824992500250125022503250425052506250725082509251025112512251325142515251625172518251925202521252225232524252525262527252825292530253125322533253425352536253725382539254025412542254325442545254625472548254925502551255225532554255525562557255825592560256125622563256425652566256725682569257025712572257325742575257625772578257925802581258225832584258525862587258825892590259125922593259425952596259725982599260026012602260326042605260626072608260926102611261226132614261526162617261826192620262126222623262426252626262726282629263026312632263326342635263626372638263926402641264226432644264526462647264826492650265126522653265426552656265726582659266026612662266326642665266626672668266926702671267226732674267526762677267826792680268126822683268426852686268726882689269026912692269326942695269626972698269927002701270227032704270527062707270827092710271127122713271427152716271727182719272027212722272327242725272627272728272927302731273227332734273527362737273827392740274127422743274427452746274727482749275027512752275327542755275627572758275927602761276227632764276527662767276827692770277127722773277427752776277727782779278027812782278327842785278627872788278927902791279227932794279527962797279827992800280128022803280428052806280728082809281028112812281328142815281628172818281928202821282228232824282528262827282828292830283128322833283428352836283728382839284028412842284328442845284628472848284928502851285228532854285528562857285828592860286128622863286428652866286728682869287028712872287328742875287628772878287928802881288228832884288528862887288828892890289128922893289428952896289728982899290029012902290329042905290629072908290929102911291229132914291529162917291829192920292129222923292429252926292729282929293029312932293329342935293629372938293929402941294229432944294529462947294829492950295129522953295429552956295729582959296029612962296329642965296629672968296929702971297229732974297529762977297829792980298129822983298429852986298729882989299029912992299329942995299629972998299930003001300230033004300530063007300830093010301130123013301430153016301730183019302030213022302330243025302630273028302930303031303230333034303530363037303830393040304130423043304430453046304730483049305030513052305330543055305630573058305930603061306230633064306530663067306830693070307130723073307430753076307730783079308030813082308330843085308630873088308930903091309230933094309530963097309830993100310131023103310431053106310731083109311031113112311331143115311631173118311931203121312231233124312531263127312831293130313131323133313431353136313731383139314031413142314331443145314631473148314931503151315231533154315531563157315831593160316131623163316431653166316731683169317031713172317331743175317631773178317931803181318231833184318531863187318831893190319131923193319431953196319731983199320032013202320332043205320632073208320932103211321232133214321532163217321832193220322132223223322432253226322732283229323032313232323332343235323632373238323932403241324232433244324532463247324832493250325132523253325432553256325732583259326032613262326332643265326632673268326932703271327232733274327532763277327832793280328132823283328432853286328732883289329032913292329332943295329632973298329933003301330233033304330533063307330833093310331133123313331433153316331733183319332033213322332333243325332633273328332933303331333233333334333533363337333833393340334133423343334433453346334733483349335033513352335333543355335633573358335933603361336233633364336533663367336833693370337133723373337433753376337733783379338033813382338333843385338633873388338933903391339233933394339533963397339833993400340134023403340434053406340734083409341034113412341334143415341634173418341934203421342234233424342534263427342834293430343134323433343434353436343734383439344034413442344334443445344634473448344934503451345234533454345534563457345834593460346134623463346434653466346734683469347034713472347334743475347634773478347934803481348234833484348534863487348834893490349134923493349434953496349734983499350035013502350335043505350635073508350935103511351235133514351535163517351835193520352135223523352435253526352735283529353035313532353335343535353635373538353935403541354235433544354535463547354835493550355135523553355435553556355735583559356035613562356335643565356635673568356935703571357235733574357535763577357835793580358135823583358435853586358735883589359035913592359335943595359635973598359936003601360236033604360536063607360836093610361136123613361436153616361736183619362036213622362336243625362636273628362936303631363236333634363536363637363836393640364136423643364436453646364736483649365036513652365336543655365636573658365936603661366236633664366536663667366836693670367136723673367436753676367736783679368036813682368336843685368636873688368936903691369236933694369536963697369836993700370137023703370437053706370737083709371037113712371337143715
  1. /*
  2. * Copyright (c) 2000-2005 Silicon Graphics, Inc.
  3. * All Rights Reserved.
  4. *
  5. * This program is free software; you can redistribute it and/or
  6. * modify it under the terms of the GNU General Public License as
  7. * published by the Free Software Foundation.
  8. *
  9. * This program is distributed in the hope that it would be useful,
  10. * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11. * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  12. * GNU General Public License for more details.
  13. *
  14. * You should have received a copy of the GNU General Public License
  15. * along with this program; if not, write the Free Software Foundation,
  16. * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
  17. */
  18. #include "xfs.h"
  19. #include "xfs_fs.h"
  20. #include "xfs_types.h"
  21. #include "xfs_log.h"
  22. #include "xfs_trans.h"
  23. #include "xfs_sb.h"
  24. #include "xfs_ag.h"
  25. #include "xfs_mount.h"
  26. #include "xfs_error.h"
  27. #include "xfs_log_priv.h"
  28. #include "xfs_buf_item.h"
  29. #include "xfs_bmap_btree.h"
  30. #include "xfs_alloc_btree.h"
  31. #include "xfs_ialloc_btree.h"
  32. #include "xfs_log_recover.h"
  33. #include "xfs_trans_priv.h"
  34. #include "xfs_dinode.h"
  35. #include "xfs_inode.h"
  36. #include "xfs_trace.h"
  37. kmem_zone_t *xfs_log_ticket_zone;
  38. /* Local miscellaneous function prototypes */
  39. STATIC int
  40. xlog_commit_record(
  41. struct xlog *log,
  42. struct xlog_ticket *ticket,
  43. struct xlog_in_core **iclog,
  44. xfs_lsn_t *commitlsnp);
  45. STATIC struct xlog *
  46. xlog_alloc_log(
  47. struct xfs_mount *mp,
  48. struct xfs_buftarg *log_target,
  49. xfs_daddr_t blk_offset,
  50. int num_bblks);
  51. STATIC int
  52. xlog_space_left(
  53. struct xlog *log,
  54. atomic64_t *head);
  55. STATIC int
  56. xlog_sync(
  57. struct xlog *log,
  58. struct xlog_in_core *iclog);
  59. STATIC void
  60. xlog_dealloc_log(
  61. struct xlog *log);
  62. /* local state machine functions */
  63. STATIC void xlog_state_done_syncing(xlog_in_core_t *iclog, int);
  64. STATIC void
  65. xlog_state_do_callback(
  66. struct xlog *log,
  67. int aborted,
  68. struct xlog_in_core *iclog);
  69. STATIC int
  70. xlog_state_get_iclog_space(
  71. struct xlog *log,
  72. int len,
  73. struct xlog_in_core **iclog,
  74. struct xlog_ticket *ticket,
  75. int *continued_write,
  76. int *logoffsetp);
  77. STATIC int
  78. xlog_state_release_iclog(
  79. struct xlog *log,
  80. struct xlog_in_core *iclog);
  81. STATIC void
  82. xlog_state_switch_iclogs(
  83. struct xlog *log,
  84. struct xlog_in_core *iclog,
  85. int eventual_size);
  86. STATIC void
  87. xlog_state_want_sync(
  88. struct xlog *log,
  89. struct xlog_in_core *iclog);
  90. STATIC void
  91. xlog_grant_push_ail(
  92. struct xlog *log,
  93. int need_bytes);
  94. STATIC void
  95. xlog_regrant_reserve_log_space(
  96. struct xlog *log,
  97. struct xlog_ticket *ticket);
  98. STATIC void
  99. xlog_ungrant_log_space(
  100. struct xlog *log,
  101. struct xlog_ticket *ticket);
  102. #if defined(DEBUG)
  103. STATIC void
  104. xlog_verify_dest_ptr(
  105. struct xlog *log,
  106. char *ptr);
  107. STATIC void
  108. xlog_verify_grant_tail(
  109. struct xlog *log);
  110. STATIC void
  111. xlog_verify_iclog(
  112. struct xlog *log,
  113. struct xlog_in_core *iclog,
  114. int count,
  115. boolean_t syncing);
  116. STATIC void
  117. xlog_verify_tail_lsn(
  118. struct xlog *log,
  119. struct xlog_in_core *iclog,
  120. xfs_lsn_t tail_lsn);
  121. #else
  122. #define xlog_verify_dest_ptr(a,b)
  123. #define xlog_verify_grant_tail(a)
  124. #define xlog_verify_iclog(a,b,c,d)
  125. #define xlog_verify_tail_lsn(a,b,c)
  126. #endif
  127. STATIC int
  128. xlog_iclogs_empty(
  129. struct xlog *log);
  130. static void
  131. xlog_grant_sub_space(
  132. struct xlog *log,
  133. atomic64_t *head,
  134. int bytes)
  135. {
  136. int64_t head_val = atomic64_read(head);
  137. int64_t new, old;
  138. do {
  139. int cycle, space;
  140. xlog_crack_grant_head_val(head_val, &cycle, &space);
  141. space -= bytes;
  142. if (space < 0) {
  143. space += log->l_logsize;
  144. cycle--;
  145. }
  146. old = head_val;
  147. new = xlog_assign_grant_head_val(cycle, space);
  148. head_val = atomic64_cmpxchg(head, old, new);
  149. } while (head_val != old);
  150. }
  151. static void
  152. xlog_grant_add_space(
  153. struct xlog *log,
  154. atomic64_t *head,
  155. int bytes)
  156. {
  157. int64_t head_val = atomic64_read(head);
  158. int64_t new, old;
  159. do {
  160. int tmp;
  161. int cycle, space;
  162. xlog_crack_grant_head_val(head_val, &cycle, &space);
  163. tmp = log->l_logsize - space;
  164. if (tmp > bytes)
  165. space += bytes;
  166. else {
  167. space = bytes - tmp;
  168. cycle++;
  169. }
  170. old = head_val;
  171. new = xlog_assign_grant_head_val(cycle, space);
  172. head_val = atomic64_cmpxchg(head, old, new);
  173. } while (head_val != old);
  174. }
  175. STATIC void
  176. xlog_grant_head_init(
  177. struct xlog_grant_head *head)
  178. {
  179. xlog_assign_grant_head(&head->grant, 1, 0);
  180. INIT_LIST_HEAD(&head->waiters);
  181. spin_lock_init(&head->lock);
  182. }
  183. STATIC void
  184. xlog_grant_head_wake_all(
  185. struct xlog_grant_head *head)
  186. {
  187. struct xlog_ticket *tic;
  188. spin_lock(&head->lock);
  189. list_for_each_entry(tic, &head->waiters, t_queue)
  190. wake_up_process(tic->t_task);
  191. spin_unlock(&head->lock);
  192. }
  193. static inline int
  194. xlog_ticket_reservation(
  195. struct xlog *log,
  196. struct xlog_grant_head *head,
  197. struct xlog_ticket *tic)
  198. {
  199. if (head == &log->l_write_head) {
  200. ASSERT(tic->t_flags & XLOG_TIC_PERM_RESERV);
  201. return tic->t_unit_res;
  202. } else {
  203. if (tic->t_flags & XLOG_TIC_PERM_RESERV)
  204. return tic->t_unit_res * tic->t_cnt;
  205. else
  206. return tic->t_unit_res;
  207. }
  208. }
  209. STATIC bool
  210. xlog_grant_head_wake(
  211. struct xlog *log,
  212. struct xlog_grant_head *head,
  213. int *free_bytes)
  214. {
  215. struct xlog_ticket *tic;
  216. int need_bytes;
  217. list_for_each_entry(tic, &head->waiters, t_queue) {
  218. need_bytes = xlog_ticket_reservation(log, head, tic);
  219. if (*free_bytes < need_bytes)
  220. return false;
  221. *free_bytes -= need_bytes;
  222. trace_xfs_log_grant_wake_up(log, tic);
  223. wake_up_process(tic->t_task);
  224. }
  225. return true;
  226. }
  227. STATIC int
  228. xlog_grant_head_wait(
  229. struct xlog *log,
  230. struct xlog_grant_head *head,
  231. struct xlog_ticket *tic,
  232. int need_bytes)
  233. {
  234. list_add_tail(&tic->t_queue, &head->waiters);
  235. do {
  236. if (XLOG_FORCED_SHUTDOWN(log))
  237. goto shutdown;
  238. xlog_grant_push_ail(log, need_bytes);
  239. __set_current_state(TASK_UNINTERRUPTIBLE);
  240. spin_unlock(&head->lock);
  241. XFS_STATS_INC(xs_sleep_logspace);
  242. trace_xfs_log_grant_sleep(log, tic);
  243. schedule();
  244. trace_xfs_log_grant_wake(log, tic);
  245. spin_lock(&head->lock);
  246. if (XLOG_FORCED_SHUTDOWN(log))
  247. goto shutdown;
  248. } while (xlog_space_left(log, &head->grant) < need_bytes);
  249. list_del_init(&tic->t_queue);
  250. return 0;
  251. shutdown:
  252. list_del_init(&tic->t_queue);
  253. return XFS_ERROR(EIO);
  254. }
  255. /*
  256. * Atomically get the log space required for a log ticket.
  257. *
  258. * Once a ticket gets put onto head->waiters, it will only return after the
  259. * needed reservation is satisfied.
  260. *
  261. * This function is structured so that it has a lock free fast path. This is
  262. * necessary because every new transaction reservation will come through this
  263. * path. Hence any lock will be globally hot if we take it unconditionally on
  264. * every pass.
  265. *
  266. * As tickets are only ever moved on and off head->waiters under head->lock, we
  267. * only need to take that lock if we are going to add the ticket to the queue
  268. * and sleep. We can avoid taking the lock if the ticket was never added to
  269. * head->waiters because the t_queue list head will be empty and we hold the
  270. * only reference to it so it can safely be checked unlocked.
  271. */
  272. STATIC int
  273. xlog_grant_head_check(
  274. struct xlog *log,
  275. struct xlog_grant_head *head,
  276. struct xlog_ticket *tic,
  277. int *need_bytes)
  278. {
  279. int free_bytes;
  280. int error = 0;
  281. ASSERT(!(log->l_flags & XLOG_ACTIVE_RECOVERY));
  282. /*
  283. * If there are other waiters on the queue then give them a chance at
  284. * logspace before us. Wake up the first waiters, if we do not wake
  285. * up all the waiters then go to sleep waiting for more free space,
  286. * otherwise try to get some space for this transaction.
  287. */
  288. *need_bytes = xlog_ticket_reservation(log, head, tic);
  289. free_bytes = xlog_space_left(log, &head->grant);
  290. if (!list_empty_careful(&head->waiters)) {
  291. spin_lock(&head->lock);
  292. if (!xlog_grant_head_wake(log, head, &free_bytes) ||
  293. free_bytes < *need_bytes) {
  294. error = xlog_grant_head_wait(log, head, tic,
  295. *need_bytes);
  296. }
  297. spin_unlock(&head->lock);
  298. } else if (free_bytes < *need_bytes) {
  299. spin_lock(&head->lock);
  300. error = xlog_grant_head_wait(log, head, tic, *need_bytes);
  301. spin_unlock(&head->lock);
  302. }
  303. return error;
  304. }
  305. static void
  306. xlog_tic_reset_res(xlog_ticket_t *tic)
  307. {
  308. tic->t_res_num = 0;
  309. tic->t_res_arr_sum = 0;
  310. tic->t_res_num_ophdrs = 0;
  311. }
  312. static void
  313. xlog_tic_add_region(xlog_ticket_t *tic, uint len, uint type)
  314. {
  315. if (tic->t_res_num == XLOG_TIC_LEN_MAX) {
  316. /* add to overflow and start again */
  317. tic->t_res_o_flow += tic->t_res_arr_sum;
  318. tic->t_res_num = 0;
  319. tic->t_res_arr_sum = 0;
  320. }
  321. tic->t_res_arr[tic->t_res_num].r_len = len;
  322. tic->t_res_arr[tic->t_res_num].r_type = type;
  323. tic->t_res_arr_sum += len;
  324. tic->t_res_num++;
  325. }
  326. /*
  327. * Replenish the byte reservation required by moving the grant write head.
  328. */
  329. int
  330. xfs_log_regrant(
  331. struct xfs_mount *mp,
  332. struct xlog_ticket *tic)
  333. {
  334. struct xlog *log = mp->m_log;
  335. int need_bytes;
  336. int error = 0;
  337. if (XLOG_FORCED_SHUTDOWN(log))
  338. return XFS_ERROR(EIO);
  339. XFS_STATS_INC(xs_try_logspace);
  340. /*
  341. * This is a new transaction on the ticket, so we need to change the
  342. * transaction ID so that the next transaction has a different TID in
  343. * the log. Just add one to the existing tid so that we can see chains
  344. * of rolling transactions in the log easily.
  345. */
  346. tic->t_tid++;
  347. xlog_grant_push_ail(log, tic->t_unit_res);
  348. tic->t_curr_res = tic->t_unit_res;
  349. xlog_tic_reset_res(tic);
  350. if (tic->t_cnt > 0)
  351. return 0;
  352. trace_xfs_log_regrant(log, tic);
  353. error = xlog_grant_head_check(log, &log->l_write_head, tic,
  354. &need_bytes);
  355. if (error)
  356. goto out_error;
  357. xlog_grant_add_space(log, &log->l_write_head.grant, need_bytes);
  358. trace_xfs_log_regrant_exit(log, tic);
  359. xlog_verify_grant_tail(log);
  360. return 0;
  361. out_error:
  362. /*
  363. * If we are failing, make sure the ticket doesn't have any current
  364. * reservations. We don't want to add this back when the ticket/
  365. * transaction gets cancelled.
  366. */
  367. tic->t_curr_res = 0;
  368. tic->t_cnt = 0; /* ungrant will give back unit_res * t_cnt. */
  369. return error;
  370. }
  371. /*
  372. * Reserve log space and return a ticket corresponding the reservation.
  373. *
  374. * Each reservation is going to reserve extra space for a log record header.
  375. * When writes happen to the on-disk log, we don't subtract the length of the
  376. * log record header from any reservation. By wasting space in each
  377. * reservation, we prevent over allocation problems.
  378. */
  379. int
  380. xfs_log_reserve(
  381. struct xfs_mount *mp,
  382. int unit_bytes,
  383. int cnt,
  384. struct xlog_ticket **ticp,
  385. __uint8_t client,
  386. bool permanent,
  387. uint t_type)
  388. {
  389. struct xlog *log = mp->m_log;
  390. struct xlog_ticket *tic;
  391. int need_bytes;
  392. int error = 0;
  393. ASSERT(client == XFS_TRANSACTION || client == XFS_LOG);
  394. if (XLOG_FORCED_SHUTDOWN(log))
  395. return XFS_ERROR(EIO);
  396. XFS_STATS_INC(xs_try_logspace);
  397. ASSERT(*ticp == NULL);
  398. tic = xlog_ticket_alloc(log, unit_bytes, cnt, client, permanent,
  399. KM_SLEEP | KM_MAYFAIL);
  400. if (!tic)
  401. return XFS_ERROR(ENOMEM);
  402. tic->t_trans_type = t_type;
  403. *ticp = tic;
  404. xlog_grant_push_ail(log, tic->t_unit_res * tic->t_cnt);
  405. trace_xfs_log_reserve(log, tic);
  406. error = xlog_grant_head_check(log, &log->l_reserve_head, tic,
  407. &need_bytes);
  408. if (error)
  409. goto out_error;
  410. xlog_grant_add_space(log, &log->l_reserve_head.grant, need_bytes);
  411. xlog_grant_add_space(log, &log->l_write_head.grant, need_bytes);
  412. trace_xfs_log_reserve_exit(log, tic);
  413. xlog_verify_grant_tail(log);
  414. return 0;
  415. out_error:
  416. /*
  417. * If we are failing, make sure the ticket doesn't have any current
  418. * reservations. We don't want to add this back when the ticket/
  419. * transaction gets cancelled.
  420. */
  421. tic->t_curr_res = 0;
  422. tic->t_cnt = 0; /* ungrant will give back unit_res * t_cnt. */
  423. return error;
  424. }
  425. /*
  426. * NOTES:
  427. *
  428. * 1. currblock field gets updated at startup and after in-core logs
  429. * marked as with WANT_SYNC.
  430. */
  431. /*
  432. * This routine is called when a user of a log manager ticket is done with
  433. * the reservation. If the ticket was ever used, then a commit record for
  434. * the associated transaction is written out as a log operation header with
  435. * no data. The flag XLOG_TIC_INITED is set when the first write occurs with
  436. * a given ticket. If the ticket was one with a permanent reservation, then
  437. * a few operations are done differently. Permanent reservation tickets by
  438. * default don't release the reservation. They just commit the current
  439. * transaction with the belief that the reservation is still needed. A flag
  440. * must be passed in before permanent reservations are actually released.
  441. * When these type of tickets are not released, they need to be set into
  442. * the inited state again. By doing this, a start record will be written
  443. * out when the next write occurs.
  444. */
  445. xfs_lsn_t
  446. xfs_log_done(
  447. struct xfs_mount *mp,
  448. struct xlog_ticket *ticket,
  449. struct xlog_in_core **iclog,
  450. uint flags)
  451. {
  452. struct xlog *log = mp->m_log;
  453. xfs_lsn_t lsn = 0;
  454. if (XLOG_FORCED_SHUTDOWN(log) ||
  455. /*
  456. * If nothing was ever written, don't write out commit record.
  457. * If we get an error, just continue and give back the log ticket.
  458. */
  459. (((ticket->t_flags & XLOG_TIC_INITED) == 0) &&
  460. (xlog_commit_record(log, ticket, iclog, &lsn)))) {
  461. lsn = (xfs_lsn_t) -1;
  462. if (ticket->t_flags & XLOG_TIC_PERM_RESERV) {
  463. flags |= XFS_LOG_REL_PERM_RESERV;
  464. }
  465. }
  466. if ((ticket->t_flags & XLOG_TIC_PERM_RESERV) == 0 ||
  467. (flags & XFS_LOG_REL_PERM_RESERV)) {
  468. trace_xfs_log_done_nonperm(log, ticket);
  469. /*
  470. * Release ticket if not permanent reservation or a specific
  471. * request has been made to release a permanent reservation.
  472. */
  473. xlog_ungrant_log_space(log, ticket);
  474. xfs_log_ticket_put(ticket);
  475. } else {
  476. trace_xfs_log_done_perm(log, ticket);
  477. xlog_regrant_reserve_log_space(log, ticket);
  478. /* If this ticket was a permanent reservation and we aren't
  479. * trying to release it, reset the inited flags; so next time
  480. * we write, a start record will be written out.
  481. */
  482. ticket->t_flags |= XLOG_TIC_INITED;
  483. }
  484. return lsn;
  485. }
  486. /*
  487. * Attaches a new iclog I/O completion callback routine during
  488. * transaction commit. If the log is in error state, a non-zero
  489. * return code is handed back and the caller is responsible for
  490. * executing the callback at an appropriate time.
  491. */
  492. int
  493. xfs_log_notify(
  494. struct xfs_mount *mp,
  495. struct xlog_in_core *iclog,
  496. xfs_log_callback_t *cb)
  497. {
  498. int abortflg;
  499. spin_lock(&iclog->ic_callback_lock);
  500. abortflg = (iclog->ic_state & XLOG_STATE_IOERROR);
  501. if (!abortflg) {
  502. ASSERT_ALWAYS((iclog->ic_state == XLOG_STATE_ACTIVE) ||
  503. (iclog->ic_state == XLOG_STATE_WANT_SYNC));
  504. cb->cb_next = NULL;
  505. *(iclog->ic_callback_tail) = cb;
  506. iclog->ic_callback_tail = &(cb->cb_next);
  507. }
  508. spin_unlock(&iclog->ic_callback_lock);
  509. return abortflg;
  510. }
  511. int
  512. xfs_log_release_iclog(
  513. struct xfs_mount *mp,
  514. struct xlog_in_core *iclog)
  515. {
  516. if (xlog_state_release_iclog(mp->m_log, iclog)) {
  517. xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR);
  518. return EIO;
  519. }
  520. return 0;
  521. }
  522. /*
  523. * Mount a log filesystem
  524. *
  525. * mp - ubiquitous xfs mount point structure
  526. * log_target - buftarg of on-disk log device
  527. * blk_offset - Start block # where block size is 512 bytes (BBSIZE)
  528. * num_bblocks - Number of BBSIZE blocks in on-disk log
  529. *
  530. * Return error or zero.
  531. */
  532. int
  533. xfs_log_mount(
  534. xfs_mount_t *mp,
  535. xfs_buftarg_t *log_target,
  536. xfs_daddr_t blk_offset,
  537. int num_bblks)
  538. {
  539. int error;
  540. if (!(mp->m_flags & XFS_MOUNT_NORECOVERY))
  541. xfs_notice(mp, "Mounting Filesystem");
  542. else {
  543. xfs_notice(mp,
  544. "Mounting filesystem in no-recovery mode. Filesystem will be inconsistent.");
  545. ASSERT(mp->m_flags & XFS_MOUNT_RDONLY);
  546. }
  547. mp->m_log = xlog_alloc_log(mp, log_target, blk_offset, num_bblks);
  548. if (IS_ERR(mp->m_log)) {
  549. error = -PTR_ERR(mp->m_log);
  550. goto out;
  551. }
  552. /*
  553. * Initialize the AIL now we have a log.
  554. */
  555. error = xfs_trans_ail_init(mp);
  556. if (error) {
  557. xfs_warn(mp, "AIL initialisation failed: error %d", error);
  558. goto out_free_log;
  559. }
  560. mp->m_log->l_ailp = mp->m_ail;
  561. /*
  562. * skip log recovery on a norecovery mount. pretend it all
  563. * just worked.
  564. */
  565. if (!(mp->m_flags & XFS_MOUNT_NORECOVERY)) {
  566. int readonly = (mp->m_flags & XFS_MOUNT_RDONLY);
  567. if (readonly)
  568. mp->m_flags &= ~XFS_MOUNT_RDONLY;
  569. error = xlog_recover(mp->m_log);
  570. if (readonly)
  571. mp->m_flags |= XFS_MOUNT_RDONLY;
  572. if (error) {
  573. xfs_warn(mp, "log mount/recovery failed: error %d",
  574. error);
  575. goto out_destroy_ail;
  576. }
  577. }
  578. /* Normal transactions can now occur */
  579. mp->m_log->l_flags &= ~XLOG_ACTIVE_RECOVERY;
  580. /*
  581. * Now the log has been fully initialised and we know were our
  582. * space grant counters are, we can initialise the permanent ticket
  583. * needed for delayed logging to work.
  584. */
  585. xlog_cil_init_post_recovery(mp->m_log);
  586. return 0;
  587. out_destroy_ail:
  588. xfs_trans_ail_destroy(mp);
  589. out_free_log:
  590. xlog_dealloc_log(mp->m_log);
  591. out:
  592. return error;
  593. }
  594. /*
  595. * Finish the recovery of the file system. This is separate from
  596. * the xfs_log_mount() call, because it depends on the code in
  597. * xfs_mountfs() to read in the root and real-time bitmap inodes
  598. * between calling xfs_log_mount() and here.
  599. *
  600. * mp - ubiquitous xfs mount point structure
  601. */
  602. int
  603. xfs_log_mount_finish(xfs_mount_t *mp)
  604. {
  605. int error;
  606. if (!(mp->m_flags & XFS_MOUNT_NORECOVERY))
  607. error = xlog_recover_finish(mp->m_log);
  608. else {
  609. error = 0;
  610. ASSERT(mp->m_flags & XFS_MOUNT_RDONLY);
  611. }
  612. return error;
  613. }
  614. /*
  615. * Final log writes as part of unmount.
  616. *
  617. * Mark the filesystem clean as unmount happens. Note that during relocation
  618. * this routine needs to be executed as part of source-bag while the
  619. * deallocation must not be done until source-end.
  620. */
  621. /*
  622. * Unmount record used to have a string "Unmount filesystem--" in the
  623. * data section where the "Un" was really a magic number (XLOG_UNMOUNT_TYPE).
  624. * We just write the magic number now since that particular field isn't
  625. * currently architecture converted and "nUmount" is a bit foo.
  626. * As far as I know, there weren't any dependencies on the old behaviour.
  627. */
  628. int
  629. xfs_log_unmount_write(xfs_mount_t *mp)
  630. {
  631. struct xlog *log = mp->m_log;
  632. xlog_in_core_t *iclog;
  633. #ifdef DEBUG
  634. xlog_in_core_t *first_iclog;
  635. #endif
  636. xlog_ticket_t *tic = NULL;
  637. xfs_lsn_t lsn;
  638. int error;
  639. /*
  640. * Don't write out unmount record on read-only mounts.
  641. * Or, if we are doing a forced umount (typically because of IO errors).
  642. */
  643. if (mp->m_flags & XFS_MOUNT_RDONLY)
  644. return 0;
  645. error = _xfs_log_force(mp, XFS_LOG_SYNC, NULL);
  646. ASSERT(error || !(XLOG_FORCED_SHUTDOWN(log)));
  647. #ifdef DEBUG
  648. first_iclog = iclog = log->l_iclog;
  649. do {
  650. if (!(iclog->ic_state & XLOG_STATE_IOERROR)) {
  651. ASSERT(iclog->ic_state & XLOG_STATE_ACTIVE);
  652. ASSERT(iclog->ic_offset == 0);
  653. }
  654. iclog = iclog->ic_next;
  655. } while (iclog != first_iclog);
  656. #endif
  657. if (! (XLOG_FORCED_SHUTDOWN(log))) {
  658. error = xfs_log_reserve(mp, 600, 1, &tic,
  659. XFS_LOG, 0, XLOG_UNMOUNT_REC_TYPE);
  660. if (!error) {
  661. /* the data section must be 32 bit size aligned */
  662. struct {
  663. __uint16_t magic;
  664. __uint16_t pad1;
  665. __uint32_t pad2; /* may as well make it 64 bits */
  666. } magic = {
  667. .magic = XLOG_UNMOUNT_TYPE,
  668. };
  669. struct xfs_log_iovec reg = {
  670. .i_addr = &magic,
  671. .i_len = sizeof(magic),
  672. .i_type = XLOG_REG_TYPE_UNMOUNT,
  673. };
  674. struct xfs_log_vec vec = {
  675. .lv_niovecs = 1,
  676. .lv_iovecp = &reg,
  677. };
  678. /* remove inited flag, and account for space used */
  679. tic->t_flags = 0;
  680. tic->t_curr_res -= sizeof(magic);
  681. error = xlog_write(log, &vec, tic, &lsn,
  682. NULL, XLOG_UNMOUNT_TRANS);
  683. /*
  684. * At this point, we're umounting anyway,
  685. * so there's no point in transitioning log state
  686. * to IOERROR. Just continue...
  687. */
  688. }
  689. if (error)
  690. xfs_alert(mp, "%s: unmount record failed", __func__);
  691. spin_lock(&log->l_icloglock);
  692. iclog = log->l_iclog;
  693. atomic_inc(&iclog->ic_refcnt);
  694. xlog_state_want_sync(log, iclog);
  695. spin_unlock(&log->l_icloglock);
  696. error = xlog_state_release_iclog(log, iclog);
  697. spin_lock(&log->l_icloglock);
  698. if (!(iclog->ic_state == XLOG_STATE_ACTIVE ||
  699. iclog->ic_state == XLOG_STATE_DIRTY)) {
  700. if (!XLOG_FORCED_SHUTDOWN(log)) {
  701. xlog_wait(&iclog->ic_force_wait,
  702. &log->l_icloglock);
  703. } else {
  704. spin_unlock(&log->l_icloglock);
  705. }
  706. } else {
  707. spin_unlock(&log->l_icloglock);
  708. }
  709. if (tic) {
  710. trace_xfs_log_umount_write(log, tic);
  711. xlog_ungrant_log_space(log, tic);
  712. xfs_log_ticket_put(tic);
  713. }
  714. } else {
  715. /*
  716. * We're already in forced_shutdown mode, couldn't
  717. * even attempt to write out the unmount transaction.
  718. *
  719. * Go through the motions of sync'ing and releasing
  720. * the iclog, even though no I/O will actually happen,
  721. * we need to wait for other log I/Os that may already
  722. * be in progress. Do this as a separate section of
  723. * code so we'll know if we ever get stuck here that
  724. * we're in this odd situation of trying to unmount
  725. * a file system that went into forced_shutdown as
  726. * the result of an unmount..
  727. */
  728. spin_lock(&log->l_icloglock);
  729. iclog = log->l_iclog;
  730. atomic_inc(&iclog->ic_refcnt);
  731. xlog_state_want_sync(log, iclog);
  732. spin_unlock(&log->l_icloglock);
  733. error = xlog_state_release_iclog(log, iclog);
  734. spin_lock(&log->l_icloglock);
  735. if ( ! ( iclog->ic_state == XLOG_STATE_ACTIVE
  736. || iclog->ic_state == XLOG_STATE_DIRTY
  737. || iclog->ic_state == XLOG_STATE_IOERROR) ) {
  738. xlog_wait(&iclog->ic_force_wait,
  739. &log->l_icloglock);
  740. } else {
  741. spin_unlock(&log->l_icloglock);
  742. }
  743. }
  744. return error;
  745. } /* xfs_log_unmount_write */
  746. /*
  747. * Deallocate log structures for unmount/relocation.
  748. *
  749. * We need to stop the aild from running before we destroy
  750. * and deallocate the log as the aild references the log.
  751. */
  752. void
  753. xfs_log_unmount(xfs_mount_t *mp)
  754. {
  755. cancel_delayed_work_sync(&mp->m_sync_work);
  756. xfs_trans_ail_destroy(mp);
  757. xlog_dealloc_log(mp->m_log);
  758. }
  759. void
  760. xfs_log_item_init(
  761. struct xfs_mount *mp,
  762. struct xfs_log_item *item,
  763. int type,
  764. const struct xfs_item_ops *ops)
  765. {
  766. item->li_mountp = mp;
  767. item->li_ailp = mp->m_ail;
  768. item->li_type = type;
  769. item->li_ops = ops;
  770. item->li_lv = NULL;
  771. INIT_LIST_HEAD(&item->li_ail);
  772. INIT_LIST_HEAD(&item->li_cil);
  773. }
  774. /*
  775. * Wake up processes waiting for log space after we have moved the log tail.
  776. */
  777. void
  778. xfs_log_space_wake(
  779. struct xfs_mount *mp)
  780. {
  781. struct xlog *log = mp->m_log;
  782. int free_bytes;
  783. if (XLOG_FORCED_SHUTDOWN(log))
  784. return;
  785. if (!list_empty_careful(&log->l_write_head.waiters)) {
  786. ASSERT(!(log->l_flags & XLOG_ACTIVE_RECOVERY));
  787. spin_lock(&log->l_write_head.lock);
  788. free_bytes = xlog_space_left(log, &log->l_write_head.grant);
  789. xlog_grant_head_wake(log, &log->l_write_head, &free_bytes);
  790. spin_unlock(&log->l_write_head.lock);
  791. }
  792. if (!list_empty_careful(&log->l_reserve_head.waiters)) {
  793. ASSERT(!(log->l_flags & XLOG_ACTIVE_RECOVERY));
  794. spin_lock(&log->l_reserve_head.lock);
  795. free_bytes = xlog_space_left(log, &log->l_reserve_head.grant);
  796. xlog_grant_head_wake(log, &log->l_reserve_head, &free_bytes);
  797. spin_unlock(&log->l_reserve_head.lock);
  798. }
  799. }
  800. /*
  801. * Determine if we have a transaction that has gone to disk
  802. * that needs to be covered. To begin the transition to the idle state
  803. * firstly the log needs to be idle (no AIL and nothing in the iclogs).
  804. * If we are then in a state where covering is needed, the caller is informed
  805. * that dummy transactions are required to move the log into the idle state.
  806. *
  807. * Because this is called as part of the sync process, we should also indicate
  808. * that dummy transactions should be issued in anything but the covered or
  809. * idle states. This ensures that the log tail is accurately reflected in
  810. * the log at the end of the sync, hence if a crash occurrs avoids replay
  811. * of transactions where the metadata is already on disk.
  812. */
  813. int
  814. xfs_log_need_covered(xfs_mount_t *mp)
  815. {
  816. int needed = 0;
  817. struct xlog *log = mp->m_log;
  818. if (!xfs_fs_writable(mp))
  819. return 0;
  820. spin_lock(&log->l_icloglock);
  821. switch (log->l_covered_state) {
  822. case XLOG_STATE_COVER_DONE:
  823. case XLOG_STATE_COVER_DONE2:
  824. case XLOG_STATE_COVER_IDLE:
  825. break;
  826. case XLOG_STATE_COVER_NEED:
  827. case XLOG_STATE_COVER_NEED2:
  828. if (!xfs_ail_min_lsn(log->l_ailp) &&
  829. xlog_iclogs_empty(log)) {
  830. if (log->l_covered_state == XLOG_STATE_COVER_NEED)
  831. log->l_covered_state = XLOG_STATE_COVER_DONE;
  832. else
  833. log->l_covered_state = XLOG_STATE_COVER_DONE2;
  834. }
  835. /* FALLTHRU */
  836. default:
  837. needed = 1;
  838. break;
  839. }
  840. spin_unlock(&log->l_icloglock);
  841. return needed;
  842. }
  843. /*
  844. * We may be holding the log iclog lock upon entering this routine.
  845. */
  846. xfs_lsn_t
  847. xlog_assign_tail_lsn_locked(
  848. struct xfs_mount *mp)
  849. {
  850. struct xlog *log = mp->m_log;
  851. struct xfs_log_item *lip;
  852. xfs_lsn_t tail_lsn;
  853. assert_spin_locked(&mp->m_ail->xa_lock);
  854. /*
  855. * To make sure we always have a valid LSN for the log tail we keep
  856. * track of the last LSN which was committed in log->l_last_sync_lsn,
  857. * and use that when the AIL was empty.
  858. */
  859. lip = xfs_ail_min(mp->m_ail);
  860. if (lip)
  861. tail_lsn = lip->li_lsn;
  862. else
  863. tail_lsn = atomic64_read(&log->l_last_sync_lsn);
  864. atomic64_set(&log->l_tail_lsn, tail_lsn);
  865. return tail_lsn;
  866. }
  867. xfs_lsn_t
  868. xlog_assign_tail_lsn(
  869. struct xfs_mount *mp)
  870. {
  871. xfs_lsn_t tail_lsn;
  872. spin_lock(&mp->m_ail->xa_lock);
  873. tail_lsn = xlog_assign_tail_lsn_locked(mp);
  874. spin_unlock(&mp->m_ail->xa_lock);
  875. return tail_lsn;
  876. }
  877. /*
  878. * Return the space in the log between the tail and the head. The head
  879. * is passed in the cycle/bytes formal parms. In the special case where
  880. * the reserve head has wrapped passed the tail, this calculation is no
  881. * longer valid. In this case, just return 0 which means there is no space
  882. * in the log. This works for all places where this function is called
  883. * with the reserve head. Of course, if the write head were to ever
  884. * wrap the tail, we should blow up. Rather than catch this case here,
  885. * we depend on other ASSERTions in other parts of the code. XXXmiken
  886. *
  887. * This code also handles the case where the reservation head is behind
  888. * the tail. The details of this case are described below, but the end
  889. * result is that we return the size of the log as the amount of space left.
  890. */
  891. STATIC int
  892. xlog_space_left(
  893. struct xlog *log,
  894. atomic64_t *head)
  895. {
  896. int free_bytes;
  897. int tail_bytes;
  898. int tail_cycle;
  899. int head_cycle;
  900. int head_bytes;
  901. xlog_crack_grant_head(head, &head_cycle, &head_bytes);
  902. xlog_crack_atomic_lsn(&log->l_tail_lsn, &tail_cycle, &tail_bytes);
  903. tail_bytes = BBTOB(tail_bytes);
  904. if (tail_cycle == head_cycle && head_bytes >= tail_bytes)
  905. free_bytes = log->l_logsize - (head_bytes - tail_bytes);
  906. else if (tail_cycle + 1 < head_cycle)
  907. return 0;
  908. else if (tail_cycle < head_cycle) {
  909. ASSERT(tail_cycle == (head_cycle - 1));
  910. free_bytes = tail_bytes - head_bytes;
  911. } else {
  912. /*
  913. * The reservation head is behind the tail.
  914. * In this case we just want to return the size of the
  915. * log as the amount of space left.
  916. */
  917. xfs_alert(log->l_mp,
  918. "xlog_space_left: head behind tail\n"
  919. " tail_cycle = %d, tail_bytes = %d\n"
  920. " GH cycle = %d, GH bytes = %d",
  921. tail_cycle, tail_bytes, head_cycle, head_bytes);
  922. ASSERT(0);
  923. free_bytes = log->l_logsize;
  924. }
  925. return free_bytes;
  926. }
  927. /*
  928. * Log function which is called when an io completes.
  929. *
  930. * The log manager needs its own routine, in order to control what
  931. * happens with the buffer after the write completes.
  932. */
  933. void
  934. xlog_iodone(xfs_buf_t *bp)
  935. {
  936. struct xlog_in_core *iclog = bp->b_fspriv;
  937. struct xlog *l = iclog->ic_log;
  938. int aborted = 0;
  939. /*
  940. * Race to shutdown the filesystem if we see an error.
  941. */
  942. if (XFS_TEST_ERROR((xfs_buf_geterror(bp)), l->l_mp,
  943. XFS_ERRTAG_IODONE_IOERR, XFS_RANDOM_IODONE_IOERR)) {
  944. xfs_buf_ioerror_alert(bp, __func__);
  945. xfs_buf_stale(bp);
  946. xfs_force_shutdown(l->l_mp, SHUTDOWN_LOG_IO_ERROR);
  947. /*
  948. * This flag will be propagated to the trans-committed
  949. * callback routines to let them know that the log-commit
  950. * didn't succeed.
  951. */
  952. aborted = XFS_LI_ABORTED;
  953. } else if (iclog->ic_state & XLOG_STATE_IOERROR) {
  954. aborted = XFS_LI_ABORTED;
  955. }
  956. /* log I/O is always issued ASYNC */
  957. ASSERT(XFS_BUF_ISASYNC(bp));
  958. xlog_state_done_syncing(iclog, aborted);
  959. /*
  960. * do not reference the buffer (bp) here as we could race
  961. * with it being freed after writing the unmount record to the
  962. * log.
  963. */
  964. } /* xlog_iodone */
  965. /*
  966. * Return size of each in-core log record buffer.
  967. *
  968. * All machines get 8 x 32kB buffers by default, unless tuned otherwise.
  969. *
  970. * If the filesystem blocksize is too large, we may need to choose a
  971. * larger size since the directory code currently logs entire blocks.
  972. */
  973. STATIC void
  974. xlog_get_iclog_buffer_size(
  975. struct xfs_mount *mp,
  976. struct xlog *log)
  977. {
  978. int size;
  979. int xhdrs;
  980. if (mp->m_logbufs <= 0)
  981. log->l_iclog_bufs = XLOG_MAX_ICLOGS;
  982. else
  983. log->l_iclog_bufs = mp->m_logbufs;
  984. /*
  985. * Buffer size passed in from mount system call.
  986. */
  987. if (mp->m_logbsize > 0) {
  988. size = log->l_iclog_size = mp->m_logbsize;
  989. log->l_iclog_size_log = 0;
  990. while (size != 1) {
  991. log->l_iclog_size_log++;
  992. size >>= 1;
  993. }
  994. if (xfs_sb_version_haslogv2(&mp->m_sb)) {
  995. /* # headers = size / 32k
  996. * one header holds cycles from 32k of data
  997. */
  998. xhdrs = mp->m_logbsize / XLOG_HEADER_CYCLE_SIZE;
  999. if (mp->m_logbsize % XLOG_HEADER_CYCLE_SIZE)
  1000. xhdrs++;
  1001. log->l_iclog_hsize = xhdrs << BBSHIFT;
  1002. log->l_iclog_heads = xhdrs;
  1003. } else {
  1004. ASSERT(mp->m_logbsize <= XLOG_BIG_RECORD_BSIZE);
  1005. log->l_iclog_hsize = BBSIZE;
  1006. log->l_iclog_heads = 1;
  1007. }
  1008. goto done;
  1009. }
  1010. /* All machines use 32kB buffers by default. */
  1011. log->l_iclog_size = XLOG_BIG_RECORD_BSIZE;
  1012. log->l_iclog_size_log = XLOG_BIG_RECORD_BSHIFT;
  1013. /* the default log size is 16k or 32k which is one header sector */
  1014. log->l_iclog_hsize = BBSIZE;
  1015. log->l_iclog_heads = 1;
  1016. done:
  1017. /* are we being asked to make the sizes selected above visible? */
  1018. if (mp->m_logbufs == 0)
  1019. mp->m_logbufs = log->l_iclog_bufs;
  1020. if (mp->m_logbsize == 0)
  1021. mp->m_logbsize = log->l_iclog_size;
  1022. } /* xlog_get_iclog_buffer_size */
  1023. /*
  1024. * This routine initializes some of the log structure for a given mount point.
  1025. * Its primary purpose is to fill in enough, so recovery can occur. However,
  1026. * some other stuff may be filled in too.
  1027. */
  1028. STATIC struct xlog *
  1029. xlog_alloc_log(
  1030. struct xfs_mount *mp,
  1031. struct xfs_buftarg *log_target,
  1032. xfs_daddr_t blk_offset,
  1033. int num_bblks)
  1034. {
  1035. struct xlog *log;
  1036. xlog_rec_header_t *head;
  1037. xlog_in_core_t **iclogp;
  1038. xlog_in_core_t *iclog, *prev_iclog=NULL;
  1039. xfs_buf_t *bp;
  1040. int i;
  1041. int error = ENOMEM;
  1042. uint log2_size = 0;
  1043. log = kmem_zalloc(sizeof(struct xlog), KM_MAYFAIL);
  1044. if (!log) {
  1045. xfs_warn(mp, "Log allocation failed: No memory!");
  1046. goto out;
  1047. }
  1048. log->l_mp = mp;
  1049. log->l_targ = log_target;
  1050. log->l_logsize = BBTOB(num_bblks);
  1051. log->l_logBBstart = blk_offset;
  1052. log->l_logBBsize = num_bblks;
  1053. log->l_covered_state = XLOG_STATE_COVER_IDLE;
  1054. log->l_flags |= XLOG_ACTIVE_RECOVERY;
  1055. log->l_prev_block = -1;
  1056. /* log->l_tail_lsn = 0x100000000LL; cycle = 1; current block = 0 */
  1057. xlog_assign_atomic_lsn(&log->l_tail_lsn, 1, 0);
  1058. xlog_assign_atomic_lsn(&log->l_last_sync_lsn, 1, 0);
  1059. log->l_curr_cycle = 1; /* 0 is bad since this is initial value */
  1060. xlog_grant_head_init(&log->l_reserve_head);
  1061. xlog_grant_head_init(&log->l_write_head);
  1062. error = EFSCORRUPTED;
  1063. if (xfs_sb_version_hassector(&mp->m_sb)) {
  1064. log2_size = mp->m_sb.sb_logsectlog;
  1065. if (log2_size < BBSHIFT) {
  1066. xfs_warn(mp, "Log sector size too small (0x%x < 0x%x)",
  1067. log2_size, BBSHIFT);
  1068. goto out_free_log;
  1069. }
  1070. log2_size -= BBSHIFT;
  1071. if (log2_size > mp->m_sectbb_log) {
  1072. xfs_warn(mp, "Log sector size too large (0x%x > 0x%x)",
  1073. log2_size, mp->m_sectbb_log);
  1074. goto out_free_log;
  1075. }
  1076. /* for larger sector sizes, must have v2 or external log */
  1077. if (log2_size && log->l_logBBstart > 0 &&
  1078. !xfs_sb_version_haslogv2(&mp->m_sb)) {
  1079. xfs_warn(mp,
  1080. "log sector size (0x%x) invalid for configuration.",
  1081. log2_size);
  1082. goto out_free_log;
  1083. }
  1084. }
  1085. log->l_sectBBsize = 1 << log2_size;
  1086. xlog_get_iclog_buffer_size(mp, log);
  1087. error = ENOMEM;
  1088. bp = xfs_buf_alloc(mp->m_logdev_targp, 0, BTOBB(log->l_iclog_size), 0);
  1089. if (!bp)
  1090. goto out_free_log;
  1091. bp->b_iodone = xlog_iodone;
  1092. ASSERT(xfs_buf_islocked(bp));
  1093. log->l_xbuf = bp;
  1094. spin_lock_init(&log->l_icloglock);
  1095. init_waitqueue_head(&log->l_flush_wait);
  1096. iclogp = &log->l_iclog;
  1097. /*
  1098. * The amount of memory to allocate for the iclog structure is
  1099. * rather funky due to the way the structure is defined. It is
  1100. * done this way so that we can use different sizes for machines
  1101. * with different amounts of memory. See the definition of
  1102. * xlog_in_core_t in xfs_log_priv.h for details.
  1103. */
  1104. ASSERT(log->l_iclog_size >= 4096);
  1105. for (i=0; i < log->l_iclog_bufs; i++) {
  1106. *iclogp = kmem_zalloc(sizeof(xlog_in_core_t), KM_MAYFAIL);
  1107. if (!*iclogp)
  1108. goto out_free_iclog;
  1109. iclog = *iclogp;
  1110. iclog->ic_prev = prev_iclog;
  1111. prev_iclog = iclog;
  1112. bp = xfs_buf_get_uncached(mp->m_logdev_targp,
  1113. BTOBB(log->l_iclog_size), 0);
  1114. if (!bp)
  1115. goto out_free_iclog;
  1116. bp->b_iodone = xlog_iodone;
  1117. iclog->ic_bp = bp;
  1118. iclog->ic_data = bp->b_addr;
  1119. #ifdef DEBUG
  1120. log->l_iclog_bak[i] = (xfs_caddr_t)&(iclog->ic_header);
  1121. #endif
  1122. head = &iclog->ic_header;
  1123. memset(head, 0, sizeof(xlog_rec_header_t));
  1124. head->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM);
  1125. head->h_version = cpu_to_be32(
  1126. xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? 2 : 1);
  1127. head->h_size = cpu_to_be32(log->l_iclog_size);
  1128. /* new fields */
  1129. head->h_fmt = cpu_to_be32(XLOG_FMT);
  1130. memcpy(&head->h_fs_uuid, &mp->m_sb.sb_uuid, sizeof(uuid_t));
  1131. iclog->ic_size = BBTOB(bp->b_length) - log->l_iclog_hsize;
  1132. iclog->ic_state = XLOG_STATE_ACTIVE;
  1133. iclog->ic_log = log;
  1134. atomic_set(&iclog->ic_refcnt, 0);
  1135. spin_lock_init(&iclog->ic_callback_lock);
  1136. iclog->ic_callback_tail = &(iclog->ic_callback);
  1137. iclog->ic_datap = (char *)iclog->ic_data + log->l_iclog_hsize;
  1138. ASSERT(xfs_buf_islocked(iclog->ic_bp));
  1139. init_waitqueue_head(&iclog->ic_force_wait);
  1140. init_waitqueue_head(&iclog->ic_write_wait);
  1141. iclogp = &iclog->ic_next;
  1142. }
  1143. *iclogp = log->l_iclog; /* complete ring */
  1144. log->l_iclog->ic_prev = prev_iclog; /* re-write 1st prev ptr */
  1145. error = xlog_cil_init(log);
  1146. if (error)
  1147. goto out_free_iclog;
  1148. return log;
  1149. out_free_iclog:
  1150. for (iclog = log->l_iclog; iclog; iclog = prev_iclog) {
  1151. prev_iclog = iclog->ic_next;
  1152. if (iclog->ic_bp)
  1153. xfs_buf_free(iclog->ic_bp);
  1154. kmem_free(iclog);
  1155. }
  1156. spinlock_destroy(&log->l_icloglock);
  1157. xfs_buf_free(log->l_xbuf);
  1158. out_free_log:
  1159. kmem_free(log);
  1160. out:
  1161. return ERR_PTR(-error);
  1162. } /* xlog_alloc_log */
  1163. /*
  1164. * Write out the commit record of a transaction associated with the given
  1165. * ticket. Return the lsn of the commit record.
  1166. */
  1167. STATIC int
  1168. xlog_commit_record(
  1169. struct xlog *log,
  1170. struct xlog_ticket *ticket,
  1171. struct xlog_in_core **iclog,
  1172. xfs_lsn_t *commitlsnp)
  1173. {
  1174. struct xfs_mount *mp = log->l_mp;
  1175. int error;
  1176. struct xfs_log_iovec reg = {
  1177. .i_addr = NULL,
  1178. .i_len = 0,
  1179. .i_type = XLOG_REG_TYPE_COMMIT,
  1180. };
  1181. struct xfs_log_vec vec = {
  1182. .lv_niovecs = 1,
  1183. .lv_iovecp = &reg,
  1184. };
  1185. ASSERT_ALWAYS(iclog);
  1186. error = xlog_write(log, &vec, ticket, commitlsnp, iclog,
  1187. XLOG_COMMIT_TRANS);
  1188. if (error)
  1189. xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR);
  1190. return error;
  1191. }
  1192. /*
  1193. * Push on the buffer cache code if we ever use more than 75% of the on-disk
  1194. * log space. This code pushes on the lsn which would supposedly free up
  1195. * the 25% which we want to leave free. We may need to adopt a policy which
  1196. * pushes on an lsn which is further along in the log once we reach the high
  1197. * water mark. In this manner, we would be creating a low water mark.
  1198. */
  1199. STATIC void
  1200. xlog_grant_push_ail(
  1201. struct xlog *log,
  1202. int need_bytes)
  1203. {
  1204. xfs_lsn_t threshold_lsn = 0;
  1205. xfs_lsn_t last_sync_lsn;
  1206. int free_blocks;
  1207. int free_bytes;
  1208. int threshold_block;
  1209. int threshold_cycle;
  1210. int free_threshold;
  1211. ASSERT(BTOBB(need_bytes) < log->l_logBBsize);
  1212. free_bytes = xlog_space_left(log, &log->l_reserve_head.grant);
  1213. free_blocks = BTOBBT(free_bytes);
  1214. /*
  1215. * Set the threshold for the minimum number of free blocks in the
  1216. * log to the maximum of what the caller needs, one quarter of the
  1217. * log, and 256 blocks.
  1218. */
  1219. free_threshold = BTOBB(need_bytes);
  1220. free_threshold = MAX(free_threshold, (log->l_logBBsize >> 2));
  1221. free_threshold = MAX(free_threshold, 256);
  1222. if (free_blocks >= free_threshold)
  1223. return;
  1224. xlog_crack_atomic_lsn(&log->l_tail_lsn, &threshold_cycle,
  1225. &threshold_block);
  1226. threshold_block += free_threshold;
  1227. if (threshold_block >= log->l_logBBsize) {
  1228. threshold_block -= log->l_logBBsize;
  1229. threshold_cycle += 1;
  1230. }
  1231. threshold_lsn = xlog_assign_lsn(threshold_cycle,
  1232. threshold_block);
  1233. /*
  1234. * Don't pass in an lsn greater than the lsn of the last
  1235. * log record known to be on disk. Use a snapshot of the last sync lsn
  1236. * so that it doesn't change between the compare and the set.
  1237. */
  1238. last_sync_lsn = atomic64_read(&log->l_last_sync_lsn);
  1239. if (XFS_LSN_CMP(threshold_lsn, last_sync_lsn) > 0)
  1240. threshold_lsn = last_sync_lsn;
  1241. /*
  1242. * Get the transaction layer to kick the dirty buffers out to
  1243. * disk asynchronously. No point in trying to do this if
  1244. * the filesystem is shutting down.
  1245. */
  1246. if (!XLOG_FORCED_SHUTDOWN(log))
  1247. xfs_ail_push(log->l_ailp, threshold_lsn);
  1248. }
  1249. /*
  1250. * The bdstrat callback function for log bufs. This gives us a central
  1251. * place to trap bufs in case we get hit by a log I/O error and need to
  1252. * shutdown. Actually, in practice, even when we didn't get a log error,
  1253. * we transition the iclogs to IOERROR state *after* flushing all existing
  1254. * iclogs to disk. This is because we don't want anymore new transactions to be
  1255. * started or completed afterwards.
  1256. */
  1257. STATIC int
  1258. xlog_bdstrat(
  1259. struct xfs_buf *bp)
  1260. {
  1261. struct xlog_in_core *iclog = bp->b_fspriv;
  1262. if (iclog->ic_state & XLOG_STATE_IOERROR) {
  1263. xfs_buf_ioerror(bp, EIO);
  1264. xfs_buf_stale(bp);
  1265. xfs_buf_ioend(bp, 0);
  1266. /*
  1267. * It would seem logical to return EIO here, but we rely on
  1268. * the log state machine to propagate I/O errors instead of
  1269. * doing it here.
  1270. */
  1271. return 0;
  1272. }
  1273. xfs_buf_iorequest(bp);
  1274. return 0;
  1275. }
  1276. /*
  1277. * Flush out the in-core log (iclog) to the on-disk log in an asynchronous
  1278. * fashion. Previously, we should have moved the current iclog
  1279. * ptr in the log to point to the next available iclog. This allows further
  1280. * write to continue while this code syncs out an iclog ready to go.
  1281. * Before an in-core log can be written out, the data section must be scanned
  1282. * to save away the 1st word of each BBSIZE block into the header. We replace
  1283. * it with the current cycle count. Each BBSIZE block is tagged with the
  1284. * cycle count because there in an implicit assumption that drives will
  1285. * guarantee that entire 512 byte blocks get written at once. In other words,
  1286. * we can't have part of a 512 byte block written and part not written. By
  1287. * tagging each block, we will know which blocks are valid when recovering
  1288. * after an unclean shutdown.
  1289. *
  1290. * This routine is single threaded on the iclog. No other thread can be in
  1291. * this routine with the same iclog. Changing contents of iclog can there-
  1292. * fore be done without grabbing the state machine lock. Updating the global
  1293. * log will require grabbing the lock though.
  1294. *
  1295. * The entire log manager uses a logical block numbering scheme. Only
  1296. * log_sync (and then only bwrite()) know about the fact that the log may
  1297. * not start with block zero on a given device. The log block start offset
  1298. * is added immediately before calling bwrite().
  1299. */
  1300. STATIC int
  1301. xlog_sync(
  1302. struct xlog *log,
  1303. struct xlog_in_core *iclog)
  1304. {
  1305. xfs_caddr_t dptr; /* pointer to byte sized element */
  1306. xfs_buf_t *bp;
  1307. int i;
  1308. uint count; /* byte count of bwrite */
  1309. uint count_init; /* initial count before roundup */
  1310. int roundoff; /* roundoff to BB or stripe */
  1311. int split = 0; /* split write into two regions */
  1312. int error;
  1313. int v2 = xfs_sb_version_haslogv2(&log->l_mp->m_sb);
  1314. XFS_STATS_INC(xs_log_writes);
  1315. ASSERT(atomic_read(&iclog->ic_refcnt) == 0);
  1316. /* Add for LR header */
  1317. count_init = log->l_iclog_hsize + iclog->ic_offset;
  1318. /* Round out the log write size */
  1319. if (v2 && log->l_mp->m_sb.sb_logsunit > 1) {
  1320. /* we have a v2 stripe unit to use */
  1321. count = XLOG_LSUNITTOB(log, XLOG_BTOLSUNIT(log, count_init));
  1322. } else {
  1323. count = BBTOB(BTOBB(count_init));
  1324. }
  1325. roundoff = count - count_init;
  1326. ASSERT(roundoff >= 0);
  1327. ASSERT((v2 && log->l_mp->m_sb.sb_logsunit > 1 &&
  1328. roundoff < log->l_mp->m_sb.sb_logsunit)
  1329. ||
  1330. (log->l_mp->m_sb.sb_logsunit <= 1 &&
  1331. roundoff < BBTOB(1)));
  1332. /* move grant heads by roundoff in sync */
  1333. xlog_grant_add_space(log, &log->l_reserve_head.grant, roundoff);
  1334. xlog_grant_add_space(log, &log->l_write_head.grant, roundoff);
  1335. /* put cycle number in every block */
  1336. xlog_pack_data(log, iclog, roundoff);
  1337. /* real byte length */
  1338. if (v2) {
  1339. iclog->ic_header.h_len =
  1340. cpu_to_be32(iclog->ic_offset + roundoff);
  1341. } else {
  1342. iclog->ic_header.h_len =
  1343. cpu_to_be32(iclog->ic_offset);
  1344. }
  1345. bp = iclog->ic_bp;
  1346. XFS_BUF_SET_ADDR(bp, BLOCK_LSN(be64_to_cpu(iclog->ic_header.h_lsn)));
  1347. XFS_STATS_ADD(xs_log_blocks, BTOBB(count));
  1348. /* Do we need to split this write into 2 parts? */
  1349. if (XFS_BUF_ADDR(bp) + BTOBB(count) > log->l_logBBsize) {
  1350. split = count - (BBTOB(log->l_logBBsize - XFS_BUF_ADDR(bp)));
  1351. count = BBTOB(log->l_logBBsize - XFS_BUF_ADDR(bp));
  1352. iclog->ic_bwritecnt = 2; /* split into 2 writes */
  1353. } else {
  1354. iclog->ic_bwritecnt = 1;
  1355. }
  1356. bp->b_io_length = BTOBB(count);
  1357. bp->b_fspriv = iclog;
  1358. XFS_BUF_ZEROFLAGS(bp);
  1359. XFS_BUF_ASYNC(bp);
  1360. bp->b_flags |= XBF_SYNCIO;
  1361. if (log->l_mp->m_flags & XFS_MOUNT_BARRIER) {
  1362. bp->b_flags |= XBF_FUA;
  1363. /*
  1364. * Flush the data device before flushing the log to make
  1365. * sure all meta data written back from the AIL actually made
  1366. * it to disk before stamping the new log tail LSN into the
  1367. * log buffer. For an external log we need to issue the
  1368. * flush explicitly, and unfortunately synchronously here;
  1369. * for an internal log we can simply use the block layer
  1370. * state machine for preflushes.
  1371. */
  1372. if (log->l_mp->m_logdev_targp != log->l_mp->m_ddev_targp)
  1373. xfs_blkdev_issue_flush(log->l_mp->m_ddev_targp);
  1374. else
  1375. bp->b_flags |= XBF_FLUSH;
  1376. }
  1377. ASSERT(XFS_BUF_ADDR(bp) <= log->l_logBBsize-1);
  1378. ASSERT(XFS_BUF_ADDR(bp) + BTOBB(count) <= log->l_logBBsize);
  1379. xlog_verify_iclog(log, iclog, count, B_TRUE);
  1380. /* account for log which doesn't start at block #0 */
  1381. XFS_BUF_SET_ADDR(bp, XFS_BUF_ADDR(bp) + log->l_logBBstart);
  1382. /*
  1383. * Don't call xfs_bwrite here. We do log-syncs even when the filesystem
  1384. * is shutting down.
  1385. */
  1386. XFS_BUF_WRITE(bp);
  1387. error = xlog_bdstrat(bp);
  1388. if (error) {
  1389. xfs_buf_ioerror_alert(bp, "xlog_sync");
  1390. return error;
  1391. }
  1392. if (split) {
  1393. bp = iclog->ic_log->l_xbuf;
  1394. XFS_BUF_SET_ADDR(bp, 0); /* logical 0 */
  1395. xfs_buf_associate_memory(bp,
  1396. (char *)&iclog->ic_header + count, split);
  1397. bp->b_fspriv = iclog;
  1398. XFS_BUF_ZEROFLAGS(bp);
  1399. XFS_BUF_ASYNC(bp);
  1400. bp->b_flags |= XBF_SYNCIO;
  1401. if (log->l_mp->m_flags & XFS_MOUNT_BARRIER)
  1402. bp->b_flags |= XBF_FUA;
  1403. dptr = bp->b_addr;
  1404. /*
  1405. * Bump the cycle numbers at the start of each block
  1406. * since this part of the buffer is at the start of
  1407. * a new cycle. Watch out for the header magic number
  1408. * case, though.
  1409. */
  1410. for (i = 0; i < split; i += BBSIZE) {
  1411. be32_add_cpu((__be32 *)dptr, 1);
  1412. if (be32_to_cpu(*(__be32 *)dptr) == XLOG_HEADER_MAGIC_NUM)
  1413. be32_add_cpu((__be32 *)dptr, 1);
  1414. dptr += BBSIZE;
  1415. }
  1416. ASSERT(XFS_BUF_ADDR(bp) <= log->l_logBBsize-1);
  1417. ASSERT(XFS_BUF_ADDR(bp) + BTOBB(count) <= log->l_logBBsize);
  1418. /* account for internal log which doesn't start at block #0 */
  1419. XFS_BUF_SET_ADDR(bp, XFS_BUF_ADDR(bp) + log->l_logBBstart);
  1420. XFS_BUF_WRITE(bp);
  1421. error = xlog_bdstrat(bp);
  1422. if (error) {
  1423. xfs_buf_ioerror_alert(bp, "xlog_sync (split)");
  1424. return error;
  1425. }
  1426. }
  1427. return 0;
  1428. } /* xlog_sync */
  1429. /*
  1430. * Deallocate a log structure
  1431. */
  1432. STATIC void
  1433. xlog_dealloc_log(
  1434. struct xlog *log)
  1435. {
  1436. xlog_in_core_t *iclog, *next_iclog;
  1437. int i;
  1438. xlog_cil_destroy(log);
  1439. /*
  1440. * always need to ensure that the extra buffer does not point to memory
  1441. * owned by another log buffer before we free it.
  1442. */
  1443. xfs_buf_set_empty(log->l_xbuf, BTOBB(log->l_iclog_size));
  1444. xfs_buf_free(log->l_xbuf);
  1445. iclog = log->l_iclog;
  1446. for (i=0; i<log->l_iclog_bufs; i++) {
  1447. xfs_buf_free(iclog->ic_bp);
  1448. next_iclog = iclog->ic_next;
  1449. kmem_free(iclog);
  1450. iclog = next_iclog;
  1451. }
  1452. spinlock_destroy(&log->l_icloglock);
  1453. log->l_mp->m_log = NULL;
  1454. kmem_free(log);
  1455. } /* xlog_dealloc_log */
  1456. /*
  1457. * Update counters atomically now that memcpy is done.
  1458. */
  1459. /* ARGSUSED */
  1460. static inline void
  1461. xlog_state_finish_copy(
  1462. struct xlog *log,
  1463. struct xlog_in_core *iclog,
  1464. int record_cnt,
  1465. int copy_bytes)
  1466. {
  1467. spin_lock(&log->l_icloglock);
  1468. be32_add_cpu(&iclog->ic_header.h_num_logops, record_cnt);
  1469. iclog->ic_offset += copy_bytes;
  1470. spin_unlock(&log->l_icloglock);
  1471. } /* xlog_state_finish_copy */
  1472. /*
  1473. * print out info relating to regions written which consume
  1474. * the reservation
  1475. */
  1476. void
  1477. xlog_print_tic_res(
  1478. struct xfs_mount *mp,
  1479. struct xlog_ticket *ticket)
  1480. {
  1481. uint i;
  1482. uint ophdr_spc = ticket->t_res_num_ophdrs * (uint)sizeof(xlog_op_header_t);
  1483. /* match with XLOG_REG_TYPE_* in xfs_log.h */
  1484. static char *res_type_str[XLOG_REG_TYPE_MAX] = {
  1485. "bformat",
  1486. "bchunk",
  1487. "efi_format",
  1488. "efd_format",
  1489. "iformat",
  1490. "icore",
  1491. "iext",
  1492. "ibroot",
  1493. "ilocal",
  1494. "iattr_ext",
  1495. "iattr_broot",
  1496. "iattr_local",
  1497. "qformat",
  1498. "dquot",
  1499. "quotaoff",
  1500. "LR header",
  1501. "unmount",
  1502. "commit",
  1503. "trans header"
  1504. };
  1505. static char *trans_type_str[XFS_TRANS_TYPE_MAX] = {
  1506. "SETATTR_NOT_SIZE",
  1507. "SETATTR_SIZE",
  1508. "INACTIVE",
  1509. "CREATE",
  1510. "CREATE_TRUNC",
  1511. "TRUNCATE_FILE",
  1512. "REMOVE",
  1513. "LINK",
  1514. "RENAME",
  1515. "MKDIR",
  1516. "RMDIR",
  1517. "SYMLINK",
  1518. "SET_DMATTRS",
  1519. "GROWFS",
  1520. "STRAT_WRITE",
  1521. "DIOSTRAT",
  1522. "WRITE_SYNC",
  1523. "WRITEID",
  1524. "ADDAFORK",
  1525. "ATTRINVAL",
  1526. "ATRUNCATE",
  1527. "ATTR_SET",
  1528. "ATTR_RM",
  1529. "ATTR_FLAG",
  1530. "CLEAR_AGI_BUCKET",
  1531. "QM_SBCHANGE",
  1532. "DUMMY1",
  1533. "DUMMY2",
  1534. "QM_QUOTAOFF",
  1535. "QM_DQALLOC",
  1536. "QM_SETQLIM",
  1537. "QM_DQCLUSTER",
  1538. "QM_QINOCREATE",
  1539. "QM_QUOTAOFF_END",
  1540. "SB_UNIT",
  1541. "FSYNC_TS",
  1542. "GROWFSRT_ALLOC",
  1543. "GROWFSRT_ZERO",
  1544. "GROWFSRT_FREE",
  1545. "SWAPEXT"
  1546. };
  1547. xfs_warn(mp,
  1548. "xlog_write: reservation summary:\n"
  1549. " trans type = %s (%u)\n"
  1550. " unit res = %d bytes\n"
  1551. " current res = %d bytes\n"
  1552. " total reg = %u bytes (o/flow = %u bytes)\n"
  1553. " ophdrs = %u (ophdr space = %u bytes)\n"
  1554. " ophdr + reg = %u bytes\n"
  1555. " num regions = %u\n",
  1556. ((ticket->t_trans_type <= 0 ||
  1557. ticket->t_trans_type > XFS_TRANS_TYPE_MAX) ?
  1558. "bad-trans-type" : trans_type_str[ticket->t_trans_type-1]),
  1559. ticket->t_trans_type,
  1560. ticket->t_unit_res,
  1561. ticket->t_curr_res,
  1562. ticket->t_res_arr_sum, ticket->t_res_o_flow,
  1563. ticket->t_res_num_ophdrs, ophdr_spc,
  1564. ticket->t_res_arr_sum +
  1565. ticket->t_res_o_flow + ophdr_spc,
  1566. ticket->t_res_num);
  1567. for (i = 0; i < ticket->t_res_num; i++) {
  1568. uint r_type = ticket->t_res_arr[i].r_type;
  1569. xfs_warn(mp, "region[%u]: %s - %u bytes\n", i,
  1570. ((r_type <= 0 || r_type > XLOG_REG_TYPE_MAX) ?
  1571. "bad-rtype" : res_type_str[r_type-1]),
  1572. ticket->t_res_arr[i].r_len);
  1573. }
  1574. xfs_alert_tag(mp, XFS_PTAG_LOGRES,
  1575. "xlog_write: reservation ran out. Need to up reservation");
  1576. xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
  1577. }
  1578. /*
  1579. * Calculate the potential space needed by the log vector. Each region gets
  1580. * its own xlog_op_header_t and may need to be double word aligned.
  1581. */
  1582. static int
  1583. xlog_write_calc_vec_length(
  1584. struct xlog_ticket *ticket,
  1585. struct xfs_log_vec *log_vector)
  1586. {
  1587. struct xfs_log_vec *lv;
  1588. int headers = 0;
  1589. int len = 0;
  1590. int i;
  1591. /* acct for start rec of xact */
  1592. if (ticket->t_flags & XLOG_TIC_INITED)
  1593. headers++;
  1594. for (lv = log_vector; lv; lv = lv->lv_next) {
  1595. headers += lv->lv_niovecs;
  1596. for (i = 0; i < lv->lv_niovecs; i++) {
  1597. struct xfs_log_iovec *vecp = &lv->lv_iovecp[i];
  1598. len += vecp->i_len;
  1599. xlog_tic_add_region(ticket, vecp->i_len, vecp->i_type);
  1600. }
  1601. }
  1602. ticket->t_res_num_ophdrs += headers;
  1603. len += headers * sizeof(struct xlog_op_header);
  1604. return len;
  1605. }
  1606. /*
  1607. * If first write for transaction, insert start record We can't be trying to
  1608. * commit if we are inited. We can't have any "partial_copy" if we are inited.
  1609. */
  1610. static int
  1611. xlog_write_start_rec(
  1612. struct xlog_op_header *ophdr,
  1613. struct xlog_ticket *ticket)
  1614. {
  1615. if (!(ticket->t_flags & XLOG_TIC_INITED))
  1616. return 0;
  1617. ophdr->oh_tid = cpu_to_be32(ticket->t_tid);
  1618. ophdr->oh_clientid = ticket->t_clientid;
  1619. ophdr->oh_len = 0;
  1620. ophdr->oh_flags = XLOG_START_TRANS;
  1621. ophdr->oh_res2 = 0;
  1622. ticket->t_flags &= ~XLOG_TIC_INITED;
  1623. return sizeof(struct xlog_op_header);
  1624. }
  1625. static xlog_op_header_t *
  1626. xlog_write_setup_ophdr(
  1627. struct xlog *log,
  1628. struct xlog_op_header *ophdr,
  1629. struct xlog_ticket *ticket,
  1630. uint flags)
  1631. {
  1632. ophdr->oh_tid = cpu_to_be32(ticket->t_tid);
  1633. ophdr->oh_clientid = ticket->t_clientid;
  1634. ophdr->oh_res2 = 0;
  1635. /* are we copying a commit or unmount record? */
  1636. ophdr->oh_flags = flags;
  1637. /*
  1638. * We've seen logs corrupted with bad transaction client ids. This
  1639. * makes sure that XFS doesn't generate them on. Turn this into an EIO
  1640. * and shut down the filesystem.
  1641. */
  1642. switch (ophdr->oh_clientid) {
  1643. case XFS_TRANSACTION:
  1644. case XFS_VOLUME:
  1645. case XFS_LOG:
  1646. break;
  1647. default:
  1648. xfs_warn(log->l_mp,
  1649. "Bad XFS transaction clientid 0x%x in ticket 0x%p",
  1650. ophdr->oh_clientid, ticket);
  1651. return NULL;
  1652. }
  1653. return ophdr;
  1654. }
  1655. /*
  1656. * Set up the parameters of the region copy into the log. This has
  1657. * to handle region write split across multiple log buffers - this
  1658. * state is kept external to this function so that this code can
  1659. * can be written in an obvious, self documenting manner.
  1660. */
  1661. static int
  1662. xlog_write_setup_copy(
  1663. struct xlog_ticket *ticket,
  1664. struct xlog_op_header *ophdr,
  1665. int space_available,
  1666. int space_required,
  1667. int *copy_off,
  1668. int *copy_len,
  1669. int *last_was_partial_copy,
  1670. int *bytes_consumed)
  1671. {
  1672. int still_to_copy;
  1673. still_to_copy = space_required - *bytes_consumed;
  1674. *copy_off = *bytes_consumed;
  1675. if (still_to_copy <= space_available) {
  1676. /* write of region completes here */
  1677. *copy_len = still_to_copy;
  1678. ophdr->oh_len = cpu_to_be32(*copy_len);
  1679. if (*last_was_partial_copy)
  1680. ophdr->oh_flags |= (XLOG_END_TRANS|XLOG_WAS_CONT_TRANS);
  1681. *last_was_partial_copy = 0;
  1682. *bytes_consumed = 0;
  1683. return 0;
  1684. }
  1685. /* partial write of region, needs extra log op header reservation */
  1686. *copy_len = space_available;
  1687. ophdr->oh_len = cpu_to_be32(*copy_len);
  1688. ophdr->oh_flags |= XLOG_CONTINUE_TRANS;
  1689. if (*last_was_partial_copy)
  1690. ophdr->oh_flags |= XLOG_WAS_CONT_TRANS;
  1691. *bytes_consumed += *copy_len;
  1692. (*last_was_partial_copy)++;
  1693. /* account for new log op header */
  1694. ticket->t_curr_res -= sizeof(struct xlog_op_header);
  1695. ticket->t_res_num_ophdrs++;
  1696. return sizeof(struct xlog_op_header);
  1697. }
  1698. static int
  1699. xlog_write_copy_finish(
  1700. struct xlog *log,
  1701. struct xlog_in_core *iclog,
  1702. uint flags,
  1703. int *record_cnt,
  1704. int *data_cnt,
  1705. int *partial_copy,
  1706. int *partial_copy_len,
  1707. int log_offset,
  1708. struct xlog_in_core **commit_iclog)
  1709. {
  1710. if (*partial_copy) {
  1711. /*
  1712. * This iclog has already been marked WANT_SYNC by
  1713. * xlog_state_get_iclog_space.
  1714. */
  1715. xlog_state_finish_copy(log, iclog, *record_cnt, *data_cnt);
  1716. *record_cnt = 0;
  1717. *data_cnt = 0;
  1718. return xlog_state_release_iclog(log, iclog);
  1719. }
  1720. *partial_copy = 0;
  1721. *partial_copy_len = 0;
  1722. if (iclog->ic_size - log_offset <= sizeof(xlog_op_header_t)) {
  1723. /* no more space in this iclog - push it. */
  1724. xlog_state_finish_copy(log, iclog, *record_cnt, *data_cnt);
  1725. *record_cnt = 0;
  1726. *data_cnt = 0;
  1727. spin_lock(&log->l_icloglock);
  1728. xlog_state_want_sync(log, iclog);
  1729. spin_unlock(&log->l_icloglock);
  1730. if (!commit_iclog)
  1731. return xlog_state_release_iclog(log, iclog);
  1732. ASSERT(flags & XLOG_COMMIT_TRANS);
  1733. *commit_iclog = iclog;
  1734. }
  1735. return 0;
  1736. }
  1737. /*
  1738. * Write some region out to in-core log
  1739. *
  1740. * This will be called when writing externally provided regions or when
  1741. * writing out a commit record for a given transaction.
  1742. *
  1743. * General algorithm:
  1744. * 1. Find total length of this write. This may include adding to the
  1745. * lengths passed in.
  1746. * 2. Check whether we violate the tickets reservation.
  1747. * 3. While writing to this iclog
  1748. * A. Reserve as much space in this iclog as can get
  1749. * B. If this is first write, save away start lsn
  1750. * C. While writing this region:
  1751. * 1. If first write of transaction, write start record
  1752. * 2. Write log operation header (header per region)
  1753. * 3. Find out if we can fit entire region into this iclog
  1754. * 4. Potentially, verify destination memcpy ptr
  1755. * 5. Memcpy (partial) region
  1756. * 6. If partial copy, release iclog; otherwise, continue
  1757. * copying more regions into current iclog
  1758. * 4. Mark want sync bit (in simulation mode)
  1759. * 5. Release iclog for potential flush to on-disk log.
  1760. *
  1761. * ERRORS:
  1762. * 1. Panic if reservation is overrun. This should never happen since
  1763. * reservation amounts are generated internal to the filesystem.
  1764. * NOTES:
  1765. * 1. Tickets are single threaded data structures.
  1766. * 2. The XLOG_END_TRANS & XLOG_CONTINUE_TRANS flags are passed down to the
  1767. * syncing routine. When a single log_write region needs to span
  1768. * multiple in-core logs, the XLOG_CONTINUE_TRANS bit should be set
  1769. * on all log operation writes which don't contain the end of the
  1770. * region. The XLOG_END_TRANS bit is used for the in-core log
  1771. * operation which contains the end of the continued log_write region.
  1772. * 3. When xlog_state_get_iclog_space() grabs the rest of the current iclog,
  1773. * we don't really know exactly how much space will be used. As a result,
  1774. * we don't update ic_offset until the end when we know exactly how many
  1775. * bytes have been written out.
  1776. */
  1777. int
  1778. xlog_write(
  1779. struct xlog *log,
  1780. struct xfs_log_vec *log_vector,
  1781. struct xlog_ticket *ticket,
  1782. xfs_lsn_t *start_lsn,
  1783. struct xlog_in_core **commit_iclog,
  1784. uint flags)
  1785. {
  1786. struct xlog_in_core *iclog = NULL;
  1787. struct xfs_log_iovec *vecp;
  1788. struct xfs_log_vec *lv;
  1789. int len;
  1790. int index;
  1791. int partial_copy = 0;
  1792. int partial_copy_len = 0;
  1793. int contwr = 0;
  1794. int record_cnt = 0;
  1795. int data_cnt = 0;
  1796. int error;
  1797. *start_lsn = 0;
  1798. len = xlog_write_calc_vec_length(ticket, log_vector);
  1799. /*
  1800. * Region headers and bytes are already accounted for.
  1801. * We only need to take into account start records and
  1802. * split regions in this function.
  1803. */
  1804. if (ticket->t_flags & XLOG_TIC_INITED)
  1805. ticket->t_curr_res -= sizeof(xlog_op_header_t);
  1806. /*
  1807. * Commit record headers need to be accounted for. These
  1808. * come in as separate writes so are easy to detect.
  1809. */
  1810. if (flags & (XLOG_COMMIT_TRANS | XLOG_UNMOUNT_TRANS))
  1811. ticket->t_curr_res -= sizeof(xlog_op_header_t);
  1812. if (ticket->t_curr_res < 0)
  1813. xlog_print_tic_res(log->l_mp, ticket);
  1814. index = 0;
  1815. lv = log_vector;
  1816. vecp = lv->lv_iovecp;
  1817. while (lv && index < lv->lv_niovecs) {
  1818. void *ptr;
  1819. int log_offset;
  1820. error = xlog_state_get_iclog_space(log, len, &iclog, ticket,
  1821. &contwr, &log_offset);
  1822. if (error)
  1823. return error;
  1824. ASSERT(log_offset <= iclog->ic_size - 1);
  1825. ptr = iclog->ic_datap + log_offset;
  1826. /* start_lsn is the first lsn written to. That's all we need. */
  1827. if (!*start_lsn)
  1828. *start_lsn = be64_to_cpu(iclog->ic_header.h_lsn);
  1829. /*
  1830. * This loop writes out as many regions as can fit in the amount
  1831. * of space which was allocated by xlog_state_get_iclog_space().
  1832. */
  1833. while (lv && index < lv->lv_niovecs) {
  1834. struct xfs_log_iovec *reg = &vecp[index];
  1835. struct xlog_op_header *ophdr;
  1836. int start_rec_copy;
  1837. int copy_len;
  1838. int copy_off;
  1839. ASSERT(reg->i_len % sizeof(__int32_t) == 0);
  1840. ASSERT((unsigned long)ptr % sizeof(__int32_t) == 0);
  1841. start_rec_copy = xlog_write_start_rec(ptr, ticket);
  1842. if (start_rec_copy) {
  1843. record_cnt++;
  1844. xlog_write_adv_cnt(&ptr, &len, &log_offset,
  1845. start_rec_copy);
  1846. }
  1847. ophdr = xlog_write_setup_ophdr(log, ptr, ticket, flags);
  1848. if (!ophdr)
  1849. return XFS_ERROR(EIO);
  1850. xlog_write_adv_cnt(&ptr, &len, &log_offset,
  1851. sizeof(struct xlog_op_header));
  1852. len += xlog_write_setup_copy(ticket, ophdr,
  1853. iclog->ic_size-log_offset,
  1854. reg->i_len,
  1855. &copy_off, &copy_len,
  1856. &partial_copy,
  1857. &partial_copy_len);
  1858. xlog_verify_dest_ptr(log, ptr);
  1859. /* copy region */
  1860. ASSERT(copy_len >= 0);
  1861. memcpy(ptr, reg->i_addr + copy_off, copy_len);
  1862. xlog_write_adv_cnt(&ptr, &len, &log_offset, copy_len);
  1863. copy_len += start_rec_copy + sizeof(xlog_op_header_t);
  1864. record_cnt++;
  1865. data_cnt += contwr ? copy_len : 0;
  1866. error = xlog_write_copy_finish(log, iclog, flags,
  1867. &record_cnt, &data_cnt,
  1868. &partial_copy,
  1869. &partial_copy_len,
  1870. log_offset,
  1871. commit_iclog);
  1872. if (error)
  1873. return error;
  1874. /*
  1875. * if we had a partial copy, we need to get more iclog
  1876. * space but we don't want to increment the region
  1877. * index because there is still more is this region to
  1878. * write.
  1879. *
  1880. * If we completed writing this region, and we flushed
  1881. * the iclog (indicated by resetting of the record
  1882. * count), then we also need to get more log space. If
  1883. * this was the last record, though, we are done and
  1884. * can just return.
  1885. */
  1886. if (partial_copy)
  1887. break;
  1888. if (++index == lv->lv_niovecs) {
  1889. lv = lv->lv_next;
  1890. index = 0;
  1891. if (lv)
  1892. vecp = lv->lv_iovecp;
  1893. }
  1894. if (record_cnt == 0) {
  1895. if (!lv)
  1896. return 0;
  1897. break;
  1898. }
  1899. }
  1900. }
  1901. ASSERT(len == 0);
  1902. xlog_state_finish_copy(log, iclog, record_cnt, data_cnt);
  1903. if (!commit_iclog)
  1904. return xlog_state_release_iclog(log, iclog);
  1905. ASSERT(flags & XLOG_COMMIT_TRANS);
  1906. *commit_iclog = iclog;
  1907. return 0;
  1908. }
  1909. /*****************************************************************************
  1910. *
  1911. * State Machine functions
  1912. *
  1913. *****************************************************************************
  1914. */
  1915. /* Clean iclogs starting from the head. This ordering must be
  1916. * maintained, so an iclog doesn't become ACTIVE beyond one that
  1917. * is SYNCING. This is also required to maintain the notion that we use
  1918. * a ordered wait queue to hold off would be writers to the log when every
  1919. * iclog is trying to sync to disk.
  1920. *
  1921. * State Change: DIRTY -> ACTIVE
  1922. */
  1923. STATIC void
  1924. xlog_state_clean_log(
  1925. struct xlog *log)
  1926. {
  1927. xlog_in_core_t *iclog;
  1928. int changed = 0;
  1929. iclog = log->l_iclog;
  1930. do {
  1931. if (iclog->ic_state == XLOG_STATE_DIRTY) {
  1932. iclog->ic_state = XLOG_STATE_ACTIVE;
  1933. iclog->ic_offset = 0;
  1934. ASSERT(iclog->ic_callback == NULL);
  1935. /*
  1936. * If the number of ops in this iclog indicate it just
  1937. * contains the dummy transaction, we can
  1938. * change state into IDLE (the second time around).
  1939. * Otherwise we should change the state into
  1940. * NEED a dummy.
  1941. * We don't need to cover the dummy.
  1942. */
  1943. if (!changed &&
  1944. (be32_to_cpu(iclog->ic_header.h_num_logops) ==
  1945. XLOG_COVER_OPS)) {
  1946. changed = 1;
  1947. } else {
  1948. /*
  1949. * We have two dirty iclogs so start over
  1950. * This could also be num of ops indicates
  1951. * this is not the dummy going out.
  1952. */
  1953. changed = 2;
  1954. }
  1955. iclog->ic_header.h_num_logops = 0;
  1956. memset(iclog->ic_header.h_cycle_data, 0,
  1957. sizeof(iclog->ic_header.h_cycle_data));
  1958. iclog->ic_header.h_lsn = 0;
  1959. } else if (iclog->ic_state == XLOG_STATE_ACTIVE)
  1960. /* do nothing */;
  1961. else
  1962. break; /* stop cleaning */
  1963. iclog = iclog->ic_next;
  1964. } while (iclog != log->l_iclog);
  1965. /* log is locked when we are called */
  1966. /*
  1967. * Change state for the dummy log recording.
  1968. * We usually go to NEED. But we go to NEED2 if the changed indicates
  1969. * we are done writing the dummy record.
  1970. * If we are done with the second dummy recored (DONE2), then
  1971. * we go to IDLE.
  1972. */
  1973. if (changed) {
  1974. switch (log->l_covered_state) {
  1975. case XLOG_STATE_COVER_IDLE:
  1976. case XLOG_STATE_COVER_NEED:
  1977. case XLOG_STATE_COVER_NEED2:
  1978. log->l_covered_state = XLOG_STATE_COVER_NEED;
  1979. break;
  1980. case XLOG_STATE_COVER_DONE:
  1981. if (changed == 1)
  1982. log->l_covered_state = XLOG_STATE_COVER_NEED2;
  1983. else
  1984. log->l_covered_state = XLOG_STATE_COVER_NEED;
  1985. break;
  1986. case XLOG_STATE_COVER_DONE2:
  1987. if (changed == 1)
  1988. log->l_covered_state = XLOG_STATE_COVER_IDLE;
  1989. else
  1990. log->l_covered_state = XLOG_STATE_COVER_NEED;
  1991. break;
  1992. default:
  1993. ASSERT(0);
  1994. }
  1995. }
  1996. } /* xlog_state_clean_log */
  1997. STATIC xfs_lsn_t
  1998. xlog_get_lowest_lsn(
  1999. struct xlog *log)
  2000. {
  2001. xlog_in_core_t *lsn_log;
  2002. xfs_lsn_t lowest_lsn, lsn;
  2003. lsn_log = log->l_iclog;
  2004. lowest_lsn = 0;
  2005. do {
  2006. if (!(lsn_log->ic_state & (XLOG_STATE_ACTIVE|XLOG_STATE_DIRTY))) {
  2007. lsn = be64_to_cpu(lsn_log->ic_header.h_lsn);
  2008. if ((lsn && !lowest_lsn) ||
  2009. (XFS_LSN_CMP(lsn, lowest_lsn) < 0)) {
  2010. lowest_lsn = lsn;
  2011. }
  2012. }
  2013. lsn_log = lsn_log->ic_next;
  2014. } while (lsn_log != log->l_iclog);
  2015. return lowest_lsn;
  2016. }
  2017. STATIC void
  2018. xlog_state_do_callback(
  2019. struct xlog *log,
  2020. int aborted,
  2021. struct xlog_in_core *ciclog)
  2022. {
  2023. xlog_in_core_t *iclog;
  2024. xlog_in_core_t *first_iclog; /* used to know when we've
  2025. * processed all iclogs once */
  2026. xfs_log_callback_t *cb, *cb_next;
  2027. int flushcnt = 0;
  2028. xfs_lsn_t lowest_lsn;
  2029. int ioerrors; /* counter: iclogs with errors */
  2030. int loopdidcallbacks; /* flag: inner loop did callbacks*/
  2031. int funcdidcallbacks; /* flag: function did callbacks */
  2032. int repeats; /* for issuing console warnings if
  2033. * looping too many times */
  2034. int wake = 0;
  2035. spin_lock(&log->l_icloglock);
  2036. first_iclog = iclog = log->l_iclog;
  2037. ioerrors = 0;
  2038. funcdidcallbacks = 0;
  2039. repeats = 0;
  2040. do {
  2041. /*
  2042. * Scan all iclogs starting with the one pointed to by the
  2043. * log. Reset this starting point each time the log is
  2044. * unlocked (during callbacks).
  2045. *
  2046. * Keep looping through iclogs until one full pass is made
  2047. * without running any callbacks.
  2048. */
  2049. first_iclog = log->l_iclog;
  2050. iclog = log->l_iclog;
  2051. loopdidcallbacks = 0;
  2052. repeats++;
  2053. do {
  2054. /* skip all iclogs in the ACTIVE & DIRTY states */
  2055. if (iclog->ic_state &
  2056. (XLOG_STATE_ACTIVE|XLOG_STATE_DIRTY)) {
  2057. iclog = iclog->ic_next;
  2058. continue;
  2059. }
  2060. /*
  2061. * Between marking a filesystem SHUTDOWN and stopping
  2062. * the log, we do flush all iclogs to disk (if there
  2063. * wasn't a log I/O error). So, we do want things to
  2064. * go smoothly in case of just a SHUTDOWN w/o a
  2065. * LOG_IO_ERROR.
  2066. */
  2067. if (!(iclog->ic_state & XLOG_STATE_IOERROR)) {
  2068. /*
  2069. * Can only perform callbacks in order. Since
  2070. * this iclog is not in the DONE_SYNC/
  2071. * DO_CALLBACK state, we skip the rest and
  2072. * just try to clean up. If we set our iclog
  2073. * to DO_CALLBACK, we will not process it when
  2074. * we retry since a previous iclog is in the
  2075. * CALLBACK and the state cannot change since
  2076. * we are holding the l_icloglock.
  2077. */
  2078. if (!(iclog->ic_state &
  2079. (XLOG_STATE_DONE_SYNC |
  2080. XLOG_STATE_DO_CALLBACK))) {
  2081. if (ciclog && (ciclog->ic_state ==
  2082. XLOG_STATE_DONE_SYNC)) {
  2083. ciclog->ic_state = XLOG_STATE_DO_CALLBACK;
  2084. }
  2085. break;
  2086. }
  2087. /*
  2088. * We now have an iclog that is in either the
  2089. * DO_CALLBACK or DONE_SYNC states. The other
  2090. * states (WANT_SYNC, SYNCING, or CALLBACK were
  2091. * caught by the above if and are going to
  2092. * clean (i.e. we aren't doing their callbacks)
  2093. * see the above if.
  2094. */
  2095. /*
  2096. * We will do one more check here to see if we
  2097. * have chased our tail around.
  2098. */
  2099. lowest_lsn = xlog_get_lowest_lsn(log);
  2100. if (lowest_lsn &&
  2101. XFS_LSN_CMP(lowest_lsn,
  2102. be64_to_cpu(iclog->ic_header.h_lsn)) < 0) {
  2103. iclog = iclog->ic_next;
  2104. continue; /* Leave this iclog for
  2105. * another thread */
  2106. }
  2107. iclog->ic_state = XLOG_STATE_CALLBACK;
  2108. /*
  2109. * Completion of a iclog IO does not imply that
  2110. * a transaction has completed, as transactions
  2111. * can be large enough to span many iclogs. We
  2112. * cannot change the tail of the log half way
  2113. * through a transaction as this may be the only
  2114. * transaction in the log and moving th etail to
  2115. * point to the middle of it will prevent
  2116. * recovery from finding the start of the
  2117. * transaction. Hence we should only update the
  2118. * last_sync_lsn if this iclog contains
  2119. * transaction completion callbacks on it.
  2120. *
  2121. * We have to do this before we drop the
  2122. * icloglock to ensure we are the only one that
  2123. * can update it.
  2124. */
  2125. ASSERT(XFS_LSN_CMP(atomic64_read(&log->l_last_sync_lsn),
  2126. be64_to_cpu(iclog->ic_header.h_lsn)) <= 0);
  2127. if (iclog->ic_callback)
  2128. atomic64_set(&log->l_last_sync_lsn,
  2129. be64_to_cpu(iclog->ic_header.h_lsn));
  2130. } else
  2131. ioerrors++;
  2132. spin_unlock(&log->l_icloglock);
  2133. /*
  2134. * Keep processing entries in the callback list until
  2135. * we come around and it is empty. We need to
  2136. * atomically see that the list is empty and change the
  2137. * state to DIRTY so that we don't miss any more
  2138. * callbacks being added.
  2139. */
  2140. spin_lock(&iclog->ic_callback_lock);
  2141. cb = iclog->ic_callback;
  2142. while (cb) {
  2143. iclog->ic_callback_tail = &(iclog->ic_callback);
  2144. iclog->ic_callback = NULL;
  2145. spin_unlock(&iclog->ic_callback_lock);
  2146. /* perform callbacks in the order given */
  2147. for (; cb; cb = cb_next) {
  2148. cb_next = cb->cb_next;
  2149. cb->cb_func(cb->cb_arg, aborted);
  2150. }
  2151. spin_lock(&iclog->ic_callback_lock);
  2152. cb = iclog->ic_callback;
  2153. }
  2154. loopdidcallbacks++;
  2155. funcdidcallbacks++;
  2156. spin_lock(&log->l_icloglock);
  2157. ASSERT(iclog->ic_callback == NULL);
  2158. spin_unlock(&iclog->ic_callback_lock);
  2159. if (!(iclog->ic_state & XLOG_STATE_IOERROR))
  2160. iclog->ic_state = XLOG_STATE_DIRTY;
  2161. /*
  2162. * Transition from DIRTY to ACTIVE if applicable.
  2163. * NOP if STATE_IOERROR.
  2164. */
  2165. xlog_state_clean_log(log);
  2166. /* wake up threads waiting in xfs_log_force() */
  2167. wake_up_all(&iclog->ic_force_wait);
  2168. iclog = iclog->ic_next;
  2169. } while (first_iclog != iclog);
  2170. if (repeats > 5000) {
  2171. flushcnt += repeats;
  2172. repeats = 0;
  2173. xfs_warn(log->l_mp,
  2174. "%s: possible infinite loop (%d iterations)",
  2175. __func__, flushcnt);
  2176. }
  2177. } while (!ioerrors && loopdidcallbacks);
  2178. /*
  2179. * make one last gasp attempt to see if iclogs are being left in
  2180. * limbo..
  2181. */
  2182. #ifdef DEBUG
  2183. if (funcdidcallbacks) {
  2184. first_iclog = iclog = log->l_iclog;
  2185. do {
  2186. ASSERT(iclog->ic_state != XLOG_STATE_DO_CALLBACK);
  2187. /*
  2188. * Terminate the loop if iclogs are found in states
  2189. * which will cause other threads to clean up iclogs.
  2190. *
  2191. * SYNCING - i/o completion will go through logs
  2192. * DONE_SYNC - interrupt thread should be waiting for
  2193. * l_icloglock
  2194. * IOERROR - give up hope all ye who enter here
  2195. */
  2196. if (iclog->ic_state == XLOG_STATE_WANT_SYNC ||
  2197. iclog->ic_state == XLOG_STATE_SYNCING ||
  2198. iclog->ic_state == XLOG_STATE_DONE_SYNC ||
  2199. iclog->ic_state == XLOG_STATE_IOERROR )
  2200. break;
  2201. iclog = iclog->ic_next;
  2202. } while (first_iclog != iclog);
  2203. }
  2204. #endif
  2205. if (log->l_iclog->ic_state & (XLOG_STATE_ACTIVE|XLOG_STATE_IOERROR))
  2206. wake = 1;
  2207. spin_unlock(&log->l_icloglock);
  2208. if (wake)
  2209. wake_up_all(&log->l_flush_wait);
  2210. }
  2211. /*
  2212. * Finish transitioning this iclog to the dirty state.
  2213. *
  2214. * Make sure that we completely execute this routine only when this is
  2215. * the last call to the iclog. There is a good chance that iclog flushes,
  2216. * when we reach the end of the physical log, get turned into 2 separate
  2217. * calls to bwrite. Hence, one iclog flush could generate two calls to this
  2218. * routine. By using the reference count bwritecnt, we guarantee that only
  2219. * the second completion goes through.
  2220. *
  2221. * Callbacks could take time, so they are done outside the scope of the
  2222. * global state machine log lock.
  2223. */
  2224. STATIC void
  2225. xlog_state_done_syncing(
  2226. xlog_in_core_t *iclog,
  2227. int aborted)
  2228. {
  2229. struct xlog *log = iclog->ic_log;
  2230. spin_lock(&log->l_icloglock);
  2231. ASSERT(iclog->ic_state == XLOG_STATE_SYNCING ||
  2232. iclog->ic_state == XLOG_STATE_IOERROR);
  2233. ASSERT(atomic_read(&iclog->ic_refcnt) == 0);
  2234. ASSERT(iclog->ic_bwritecnt == 1 || iclog->ic_bwritecnt == 2);
  2235. /*
  2236. * If we got an error, either on the first buffer, or in the case of
  2237. * split log writes, on the second, we mark ALL iclogs STATE_IOERROR,
  2238. * and none should ever be attempted to be written to disk
  2239. * again.
  2240. */
  2241. if (iclog->ic_state != XLOG_STATE_IOERROR) {
  2242. if (--iclog->ic_bwritecnt == 1) {
  2243. spin_unlock(&log->l_icloglock);
  2244. return;
  2245. }
  2246. iclog->ic_state = XLOG_STATE_DONE_SYNC;
  2247. }
  2248. /*
  2249. * Someone could be sleeping prior to writing out the next
  2250. * iclog buffer, we wake them all, one will get to do the
  2251. * I/O, the others get to wait for the result.
  2252. */
  2253. wake_up_all(&iclog->ic_write_wait);
  2254. spin_unlock(&log->l_icloglock);
  2255. xlog_state_do_callback(log, aborted, iclog); /* also cleans log */
  2256. } /* xlog_state_done_syncing */
  2257. /*
  2258. * If the head of the in-core log ring is not (ACTIVE or DIRTY), then we must
  2259. * sleep. We wait on the flush queue on the head iclog as that should be
  2260. * the first iclog to complete flushing. Hence if all iclogs are syncing,
  2261. * we will wait here and all new writes will sleep until a sync completes.
  2262. *
  2263. * The in-core logs are used in a circular fashion. They are not used
  2264. * out-of-order even when an iclog past the head is free.
  2265. *
  2266. * return:
  2267. * * log_offset where xlog_write() can start writing into the in-core
  2268. * log's data space.
  2269. * * in-core log pointer to which xlog_write() should write.
  2270. * * boolean indicating this is a continued write to an in-core log.
  2271. * If this is the last write, then the in-core log's offset field
  2272. * needs to be incremented, depending on the amount of data which
  2273. * is copied.
  2274. */
  2275. STATIC int
  2276. xlog_state_get_iclog_space(
  2277. struct xlog *log,
  2278. int len,
  2279. struct xlog_in_core **iclogp,
  2280. struct xlog_ticket *ticket,
  2281. int *continued_write,
  2282. int *logoffsetp)
  2283. {
  2284. int log_offset;
  2285. xlog_rec_header_t *head;
  2286. xlog_in_core_t *iclog;
  2287. int error;
  2288. restart:
  2289. spin_lock(&log->l_icloglock);
  2290. if (XLOG_FORCED_SHUTDOWN(log)) {
  2291. spin_unlock(&log->l_icloglock);
  2292. return XFS_ERROR(EIO);
  2293. }
  2294. iclog = log->l_iclog;
  2295. if (iclog->ic_state != XLOG_STATE_ACTIVE) {
  2296. XFS_STATS_INC(xs_log_noiclogs);
  2297. /* Wait for log writes to have flushed */
  2298. xlog_wait(&log->l_flush_wait, &log->l_icloglock);
  2299. goto restart;
  2300. }
  2301. head = &iclog->ic_header;
  2302. atomic_inc(&iclog->ic_refcnt); /* prevents sync */
  2303. log_offset = iclog->ic_offset;
  2304. /* On the 1st write to an iclog, figure out lsn. This works
  2305. * if iclogs marked XLOG_STATE_WANT_SYNC always write out what they are
  2306. * committing to. If the offset is set, that's how many blocks
  2307. * must be written.
  2308. */
  2309. if (log_offset == 0) {
  2310. ticket->t_curr_res -= log->l_iclog_hsize;
  2311. xlog_tic_add_region(ticket,
  2312. log->l_iclog_hsize,
  2313. XLOG_REG_TYPE_LRHEADER);
  2314. head->h_cycle = cpu_to_be32(log->l_curr_cycle);
  2315. head->h_lsn = cpu_to_be64(
  2316. xlog_assign_lsn(log->l_curr_cycle, log->l_curr_block));
  2317. ASSERT(log->l_curr_block >= 0);
  2318. }
  2319. /* If there is enough room to write everything, then do it. Otherwise,
  2320. * claim the rest of the region and make sure the XLOG_STATE_WANT_SYNC
  2321. * bit is on, so this will get flushed out. Don't update ic_offset
  2322. * until you know exactly how many bytes get copied. Therefore, wait
  2323. * until later to update ic_offset.
  2324. *
  2325. * xlog_write() algorithm assumes that at least 2 xlog_op_header_t's
  2326. * can fit into remaining data section.
  2327. */
  2328. if (iclog->ic_size - iclog->ic_offset < 2*sizeof(xlog_op_header_t)) {
  2329. xlog_state_switch_iclogs(log, iclog, iclog->ic_size);
  2330. /*
  2331. * If I'm the only one writing to this iclog, sync it to disk.
  2332. * We need to do an atomic compare and decrement here to avoid
  2333. * racing with concurrent atomic_dec_and_lock() calls in
  2334. * xlog_state_release_iclog() when there is more than one
  2335. * reference to the iclog.
  2336. */
  2337. if (!atomic_add_unless(&iclog->ic_refcnt, -1, 1)) {
  2338. /* we are the only one */
  2339. spin_unlock(&log->l_icloglock);
  2340. error = xlog_state_release_iclog(log, iclog);
  2341. if (error)
  2342. return error;
  2343. } else {
  2344. spin_unlock(&log->l_icloglock);
  2345. }
  2346. goto restart;
  2347. }
  2348. /* Do we have enough room to write the full amount in the remainder
  2349. * of this iclog? Or must we continue a write on the next iclog and
  2350. * mark this iclog as completely taken? In the case where we switch
  2351. * iclogs (to mark it taken), this particular iclog will release/sync
  2352. * to disk in xlog_write().
  2353. */
  2354. if (len <= iclog->ic_size - iclog->ic_offset) {
  2355. *continued_write = 0;
  2356. iclog->ic_offset += len;
  2357. } else {
  2358. *continued_write = 1;
  2359. xlog_state_switch_iclogs(log, iclog, iclog->ic_size);
  2360. }
  2361. *iclogp = iclog;
  2362. ASSERT(iclog->ic_offset <= iclog->ic_size);
  2363. spin_unlock(&log->l_icloglock);
  2364. *logoffsetp = log_offset;
  2365. return 0;
  2366. } /* xlog_state_get_iclog_space */
  2367. /* The first cnt-1 times through here we don't need to
  2368. * move the grant write head because the permanent
  2369. * reservation has reserved cnt times the unit amount.
  2370. * Release part of current permanent unit reservation and
  2371. * reset current reservation to be one units worth. Also
  2372. * move grant reservation head forward.
  2373. */
  2374. STATIC void
  2375. xlog_regrant_reserve_log_space(
  2376. struct xlog *log,
  2377. struct xlog_ticket *ticket)
  2378. {
  2379. trace_xfs_log_regrant_reserve_enter(log, ticket);
  2380. if (ticket->t_cnt > 0)
  2381. ticket->t_cnt--;
  2382. xlog_grant_sub_space(log, &log->l_reserve_head.grant,
  2383. ticket->t_curr_res);
  2384. xlog_grant_sub_space(log, &log->l_write_head.grant,
  2385. ticket->t_curr_res);
  2386. ticket->t_curr_res = ticket->t_unit_res;
  2387. xlog_tic_reset_res(ticket);
  2388. trace_xfs_log_regrant_reserve_sub(log, ticket);
  2389. /* just return if we still have some of the pre-reserved space */
  2390. if (ticket->t_cnt > 0)
  2391. return;
  2392. xlog_grant_add_space(log, &log->l_reserve_head.grant,
  2393. ticket->t_unit_res);
  2394. trace_xfs_log_regrant_reserve_exit(log, ticket);
  2395. ticket->t_curr_res = ticket->t_unit_res;
  2396. xlog_tic_reset_res(ticket);
  2397. } /* xlog_regrant_reserve_log_space */
  2398. /*
  2399. * Give back the space left from a reservation.
  2400. *
  2401. * All the information we need to make a correct determination of space left
  2402. * is present. For non-permanent reservations, things are quite easy. The
  2403. * count should have been decremented to zero. We only need to deal with the
  2404. * space remaining in the current reservation part of the ticket. If the
  2405. * ticket contains a permanent reservation, there may be left over space which
  2406. * needs to be released. A count of N means that N-1 refills of the current
  2407. * reservation can be done before we need to ask for more space. The first
  2408. * one goes to fill up the first current reservation. Once we run out of
  2409. * space, the count will stay at zero and the only space remaining will be
  2410. * in the current reservation field.
  2411. */
  2412. STATIC void
  2413. xlog_ungrant_log_space(
  2414. struct xlog *log,
  2415. struct xlog_ticket *ticket)
  2416. {
  2417. int bytes;
  2418. if (ticket->t_cnt > 0)
  2419. ticket->t_cnt--;
  2420. trace_xfs_log_ungrant_enter(log, ticket);
  2421. trace_xfs_log_ungrant_sub(log, ticket);
  2422. /*
  2423. * If this is a permanent reservation ticket, we may be able to free
  2424. * up more space based on the remaining count.
  2425. */
  2426. bytes = ticket->t_curr_res;
  2427. if (ticket->t_cnt > 0) {
  2428. ASSERT(ticket->t_flags & XLOG_TIC_PERM_RESERV);
  2429. bytes += ticket->t_unit_res*ticket->t_cnt;
  2430. }
  2431. xlog_grant_sub_space(log, &log->l_reserve_head.grant, bytes);
  2432. xlog_grant_sub_space(log, &log->l_write_head.grant, bytes);
  2433. trace_xfs_log_ungrant_exit(log, ticket);
  2434. xfs_log_space_wake(log->l_mp);
  2435. }
  2436. /*
  2437. * Flush iclog to disk if this is the last reference to the given iclog and
  2438. * the WANT_SYNC bit is set.
  2439. *
  2440. * When this function is entered, the iclog is not necessarily in the
  2441. * WANT_SYNC state. It may be sitting around waiting to get filled.
  2442. *
  2443. *
  2444. */
  2445. STATIC int
  2446. xlog_state_release_iclog(
  2447. struct xlog *log,
  2448. struct xlog_in_core *iclog)
  2449. {
  2450. int sync = 0; /* do we sync? */
  2451. if (iclog->ic_state & XLOG_STATE_IOERROR)
  2452. return XFS_ERROR(EIO);
  2453. ASSERT(atomic_read(&iclog->ic_refcnt) > 0);
  2454. if (!atomic_dec_and_lock(&iclog->ic_refcnt, &log->l_icloglock))
  2455. return 0;
  2456. if (iclog->ic_state & XLOG_STATE_IOERROR) {
  2457. spin_unlock(&log->l_icloglock);
  2458. return XFS_ERROR(EIO);
  2459. }
  2460. ASSERT(iclog->ic_state == XLOG_STATE_ACTIVE ||
  2461. iclog->ic_state == XLOG_STATE_WANT_SYNC);
  2462. if (iclog->ic_state == XLOG_STATE_WANT_SYNC) {
  2463. /* update tail before writing to iclog */
  2464. xfs_lsn_t tail_lsn = xlog_assign_tail_lsn(log->l_mp);
  2465. sync++;
  2466. iclog->ic_state = XLOG_STATE_SYNCING;
  2467. iclog->ic_header.h_tail_lsn = cpu_to_be64(tail_lsn);
  2468. xlog_verify_tail_lsn(log, iclog, tail_lsn);
  2469. /* cycle incremented when incrementing curr_block */
  2470. }
  2471. spin_unlock(&log->l_icloglock);
  2472. /*
  2473. * We let the log lock go, so it's possible that we hit a log I/O
  2474. * error or some other SHUTDOWN condition that marks the iclog
  2475. * as XLOG_STATE_IOERROR before the bwrite. However, we know that
  2476. * this iclog has consistent data, so we ignore IOERROR
  2477. * flags after this point.
  2478. */
  2479. if (sync)
  2480. return xlog_sync(log, iclog);
  2481. return 0;
  2482. } /* xlog_state_release_iclog */
  2483. /*
  2484. * This routine will mark the current iclog in the ring as WANT_SYNC
  2485. * and move the current iclog pointer to the next iclog in the ring.
  2486. * When this routine is called from xlog_state_get_iclog_space(), the
  2487. * exact size of the iclog has not yet been determined. All we know is
  2488. * that every data block. We have run out of space in this log record.
  2489. */
  2490. STATIC void
  2491. xlog_state_switch_iclogs(
  2492. struct xlog *log,
  2493. struct xlog_in_core *iclog,
  2494. int eventual_size)
  2495. {
  2496. ASSERT(iclog->ic_state == XLOG_STATE_ACTIVE);
  2497. if (!eventual_size)
  2498. eventual_size = iclog->ic_offset;
  2499. iclog->ic_state = XLOG_STATE_WANT_SYNC;
  2500. iclog->ic_header.h_prev_block = cpu_to_be32(log->l_prev_block);
  2501. log->l_prev_block = log->l_curr_block;
  2502. log->l_prev_cycle = log->l_curr_cycle;
  2503. /* roll log?: ic_offset changed later */
  2504. log->l_curr_block += BTOBB(eventual_size)+BTOBB(log->l_iclog_hsize);
  2505. /* Round up to next log-sunit */
  2506. if (xfs_sb_version_haslogv2(&log->l_mp->m_sb) &&
  2507. log->l_mp->m_sb.sb_logsunit > 1) {
  2508. __uint32_t sunit_bb = BTOBB(log->l_mp->m_sb.sb_logsunit);
  2509. log->l_curr_block = roundup(log->l_curr_block, sunit_bb);
  2510. }
  2511. if (log->l_curr_block >= log->l_logBBsize) {
  2512. log->l_curr_cycle++;
  2513. if (log->l_curr_cycle == XLOG_HEADER_MAGIC_NUM)
  2514. log->l_curr_cycle++;
  2515. log->l_curr_block -= log->l_logBBsize;
  2516. ASSERT(log->l_curr_block >= 0);
  2517. }
  2518. ASSERT(iclog == log->l_iclog);
  2519. log->l_iclog = iclog->ic_next;
  2520. } /* xlog_state_switch_iclogs */
  2521. /*
  2522. * Write out all data in the in-core log as of this exact moment in time.
  2523. *
  2524. * Data may be written to the in-core log during this call. However,
  2525. * we don't guarantee this data will be written out. A change from past
  2526. * implementation means this routine will *not* write out zero length LRs.
  2527. *
  2528. * Basically, we try and perform an intelligent scan of the in-core logs.
  2529. * If we determine there is no flushable data, we just return. There is no
  2530. * flushable data if:
  2531. *
  2532. * 1. the current iclog is active and has no data; the previous iclog
  2533. * is in the active or dirty state.
  2534. * 2. the current iclog is drity, and the previous iclog is in the
  2535. * active or dirty state.
  2536. *
  2537. * We may sleep if:
  2538. *
  2539. * 1. the current iclog is not in the active nor dirty state.
  2540. * 2. the current iclog dirty, and the previous iclog is not in the
  2541. * active nor dirty state.
  2542. * 3. the current iclog is active, and there is another thread writing
  2543. * to this particular iclog.
  2544. * 4. a) the current iclog is active and has no other writers
  2545. * b) when we return from flushing out this iclog, it is still
  2546. * not in the active nor dirty state.
  2547. */
  2548. int
  2549. _xfs_log_force(
  2550. struct xfs_mount *mp,
  2551. uint flags,
  2552. int *log_flushed)
  2553. {
  2554. struct xlog *log = mp->m_log;
  2555. struct xlog_in_core *iclog;
  2556. xfs_lsn_t lsn;
  2557. XFS_STATS_INC(xs_log_force);
  2558. xlog_cil_force(log);
  2559. spin_lock(&log->l_icloglock);
  2560. iclog = log->l_iclog;
  2561. if (iclog->ic_state & XLOG_STATE_IOERROR) {
  2562. spin_unlock(&log->l_icloglock);
  2563. return XFS_ERROR(EIO);
  2564. }
  2565. /* If the head iclog is not active nor dirty, we just attach
  2566. * ourselves to the head and go to sleep.
  2567. */
  2568. if (iclog->ic_state == XLOG_STATE_ACTIVE ||
  2569. iclog->ic_state == XLOG_STATE_DIRTY) {
  2570. /*
  2571. * If the head is dirty or (active and empty), then
  2572. * we need to look at the previous iclog. If the previous
  2573. * iclog is active or dirty we are done. There is nothing
  2574. * to sync out. Otherwise, we attach ourselves to the
  2575. * previous iclog and go to sleep.
  2576. */
  2577. if (iclog->ic_state == XLOG_STATE_DIRTY ||
  2578. (atomic_read(&iclog->ic_refcnt) == 0
  2579. && iclog->ic_offset == 0)) {
  2580. iclog = iclog->ic_prev;
  2581. if (iclog->ic_state == XLOG_STATE_ACTIVE ||
  2582. iclog->ic_state == XLOG_STATE_DIRTY)
  2583. goto no_sleep;
  2584. else
  2585. goto maybe_sleep;
  2586. } else {
  2587. if (atomic_read(&iclog->ic_refcnt) == 0) {
  2588. /* We are the only one with access to this
  2589. * iclog. Flush it out now. There should
  2590. * be a roundoff of zero to show that someone
  2591. * has already taken care of the roundoff from
  2592. * the previous sync.
  2593. */
  2594. atomic_inc(&iclog->ic_refcnt);
  2595. lsn = be64_to_cpu(iclog->ic_header.h_lsn);
  2596. xlog_state_switch_iclogs(log, iclog, 0);
  2597. spin_unlock(&log->l_icloglock);
  2598. if (xlog_state_release_iclog(log, iclog))
  2599. return XFS_ERROR(EIO);
  2600. if (log_flushed)
  2601. *log_flushed = 1;
  2602. spin_lock(&log->l_icloglock);
  2603. if (be64_to_cpu(iclog->ic_header.h_lsn) == lsn &&
  2604. iclog->ic_state != XLOG_STATE_DIRTY)
  2605. goto maybe_sleep;
  2606. else
  2607. goto no_sleep;
  2608. } else {
  2609. /* Someone else is writing to this iclog.
  2610. * Use its call to flush out the data. However,
  2611. * the other thread may not force out this LR,
  2612. * so we mark it WANT_SYNC.
  2613. */
  2614. xlog_state_switch_iclogs(log, iclog, 0);
  2615. goto maybe_sleep;
  2616. }
  2617. }
  2618. }
  2619. /* By the time we come around again, the iclog could've been filled
  2620. * which would give it another lsn. If we have a new lsn, just
  2621. * return because the relevant data has been flushed.
  2622. */
  2623. maybe_sleep:
  2624. if (flags & XFS_LOG_SYNC) {
  2625. /*
  2626. * We must check if we're shutting down here, before
  2627. * we wait, while we're holding the l_icloglock.
  2628. * Then we check again after waking up, in case our
  2629. * sleep was disturbed by a bad news.
  2630. */
  2631. if (iclog->ic_state & XLOG_STATE_IOERROR) {
  2632. spin_unlock(&log->l_icloglock);
  2633. return XFS_ERROR(EIO);
  2634. }
  2635. XFS_STATS_INC(xs_log_force_sleep);
  2636. xlog_wait(&iclog->ic_force_wait, &log->l_icloglock);
  2637. /*
  2638. * No need to grab the log lock here since we're
  2639. * only deciding whether or not to return EIO
  2640. * and the memory read should be atomic.
  2641. */
  2642. if (iclog->ic_state & XLOG_STATE_IOERROR)
  2643. return XFS_ERROR(EIO);
  2644. if (log_flushed)
  2645. *log_flushed = 1;
  2646. } else {
  2647. no_sleep:
  2648. spin_unlock(&log->l_icloglock);
  2649. }
  2650. return 0;
  2651. }
  2652. /*
  2653. * Wrapper for _xfs_log_force(), to be used when caller doesn't care
  2654. * about errors or whether the log was flushed or not. This is the normal
  2655. * interface to use when trying to unpin items or move the log forward.
  2656. */
  2657. void
  2658. xfs_log_force(
  2659. xfs_mount_t *mp,
  2660. uint flags)
  2661. {
  2662. int error;
  2663. trace_xfs_log_force(mp, 0);
  2664. error = _xfs_log_force(mp, flags, NULL);
  2665. if (error)
  2666. xfs_warn(mp, "%s: error %d returned.", __func__, error);
  2667. }
  2668. /*
  2669. * Force the in-core log to disk for a specific LSN.
  2670. *
  2671. * Find in-core log with lsn.
  2672. * If it is in the DIRTY state, just return.
  2673. * If it is in the ACTIVE state, move the in-core log into the WANT_SYNC
  2674. * state and go to sleep or return.
  2675. * If it is in any other state, go to sleep or return.
  2676. *
  2677. * Synchronous forces are implemented with a signal variable. All callers
  2678. * to force a given lsn to disk will wait on a the sv attached to the
  2679. * specific in-core log. When given in-core log finally completes its
  2680. * write to disk, that thread will wake up all threads waiting on the
  2681. * sv.
  2682. */
  2683. int
  2684. _xfs_log_force_lsn(
  2685. struct xfs_mount *mp,
  2686. xfs_lsn_t lsn,
  2687. uint flags,
  2688. int *log_flushed)
  2689. {
  2690. struct xlog *log = mp->m_log;
  2691. struct xlog_in_core *iclog;
  2692. int already_slept = 0;
  2693. ASSERT(lsn != 0);
  2694. XFS_STATS_INC(xs_log_force);
  2695. lsn = xlog_cil_force_lsn(log, lsn);
  2696. if (lsn == NULLCOMMITLSN)
  2697. return 0;
  2698. try_again:
  2699. spin_lock(&log->l_icloglock);
  2700. iclog = log->l_iclog;
  2701. if (iclog->ic_state & XLOG_STATE_IOERROR) {
  2702. spin_unlock(&log->l_icloglock);
  2703. return XFS_ERROR(EIO);
  2704. }
  2705. do {
  2706. if (be64_to_cpu(iclog->ic_header.h_lsn) != lsn) {
  2707. iclog = iclog->ic_next;
  2708. continue;
  2709. }
  2710. if (iclog->ic_state == XLOG_STATE_DIRTY) {
  2711. spin_unlock(&log->l_icloglock);
  2712. return 0;
  2713. }
  2714. if (iclog->ic_state == XLOG_STATE_ACTIVE) {
  2715. /*
  2716. * We sleep here if we haven't already slept (e.g.
  2717. * this is the first time we've looked at the correct
  2718. * iclog buf) and the buffer before us is going to
  2719. * be sync'ed. The reason for this is that if we
  2720. * are doing sync transactions here, by waiting for
  2721. * the previous I/O to complete, we can allow a few
  2722. * more transactions into this iclog before we close
  2723. * it down.
  2724. *
  2725. * Otherwise, we mark the buffer WANT_SYNC, and bump
  2726. * up the refcnt so we can release the log (which
  2727. * drops the ref count). The state switch keeps new
  2728. * transaction commits from using this buffer. When
  2729. * the current commits finish writing into the buffer,
  2730. * the refcount will drop to zero and the buffer will
  2731. * go out then.
  2732. */
  2733. if (!already_slept &&
  2734. (iclog->ic_prev->ic_state &
  2735. (XLOG_STATE_WANT_SYNC | XLOG_STATE_SYNCING))) {
  2736. ASSERT(!(iclog->ic_state & XLOG_STATE_IOERROR));
  2737. XFS_STATS_INC(xs_log_force_sleep);
  2738. xlog_wait(&iclog->ic_prev->ic_write_wait,
  2739. &log->l_icloglock);
  2740. if (log_flushed)
  2741. *log_flushed = 1;
  2742. already_slept = 1;
  2743. goto try_again;
  2744. }
  2745. atomic_inc(&iclog->ic_refcnt);
  2746. xlog_state_switch_iclogs(log, iclog, 0);
  2747. spin_unlock(&log->l_icloglock);
  2748. if (xlog_state_release_iclog(log, iclog))
  2749. return XFS_ERROR(EIO);
  2750. if (log_flushed)
  2751. *log_flushed = 1;
  2752. spin_lock(&log->l_icloglock);
  2753. }
  2754. if ((flags & XFS_LOG_SYNC) && /* sleep */
  2755. !(iclog->ic_state &
  2756. (XLOG_STATE_ACTIVE | XLOG_STATE_DIRTY))) {
  2757. /*
  2758. * Don't wait on completion if we know that we've
  2759. * gotten a log write error.
  2760. */
  2761. if (iclog->ic_state & XLOG_STATE_IOERROR) {
  2762. spin_unlock(&log->l_icloglock);
  2763. return XFS_ERROR(EIO);
  2764. }
  2765. XFS_STATS_INC(xs_log_force_sleep);
  2766. xlog_wait(&iclog->ic_force_wait, &log->l_icloglock);
  2767. /*
  2768. * No need to grab the log lock here since we're
  2769. * only deciding whether or not to return EIO
  2770. * and the memory read should be atomic.
  2771. */
  2772. if (iclog->ic_state & XLOG_STATE_IOERROR)
  2773. return XFS_ERROR(EIO);
  2774. if (log_flushed)
  2775. *log_flushed = 1;
  2776. } else { /* just return */
  2777. spin_unlock(&log->l_icloglock);
  2778. }
  2779. return 0;
  2780. } while (iclog != log->l_iclog);
  2781. spin_unlock(&log->l_icloglock);
  2782. return 0;
  2783. }
  2784. /*
  2785. * Wrapper for _xfs_log_force_lsn(), to be used when caller doesn't care
  2786. * about errors or whether the log was flushed or not. This is the normal
  2787. * interface to use when trying to unpin items or move the log forward.
  2788. */
  2789. void
  2790. xfs_log_force_lsn(
  2791. xfs_mount_t *mp,
  2792. xfs_lsn_t lsn,
  2793. uint flags)
  2794. {
  2795. int error;
  2796. trace_xfs_log_force(mp, lsn);
  2797. error = _xfs_log_force_lsn(mp, lsn, flags, NULL);
  2798. if (error)
  2799. xfs_warn(mp, "%s: error %d returned.", __func__, error);
  2800. }
  2801. /*
  2802. * Called when we want to mark the current iclog as being ready to sync to
  2803. * disk.
  2804. */
  2805. STATIC void
  2806. xlog_state_want_sync(
  2807. struct xlog *log,
  2808. struct xlog_in_core *iclog)
  2809. {
  2810. assert_spin_locked(&log->l_icloglock);
  2811. if (iclog->ic_state == XLOG_STATE_ACTIVE) {
  2812. xlog_state_switch_iclogs(log, iclog, 0);
  2813. } else {
  2814. ASSERT(iclog->ic_state &
  2815. (XLOG_STATE_WANT_SYNC|XLOG_STATE_IOERROR));
  2816. }
  2817. }
  2818. /*****************************************************************************
  2819. *
  2820. * TICKET functions
  2821. *
  2822. *****************************************************************************
  2823. */
  2824. /*
  2825. * Free a used ticket when its refcount falls to zero.
  2826. */
  2827. void
  2828. xfs_log_ticket_put(
  2829. xlog_ticket_t *ticket)
  2830. {
  2831. ASSERT(atomic_read(&ticket->t_ref) > 0);
  2832. if (atomic_dec_and_test(&ticket->t_ref))
  2833. kmem_zone_free(xfs_log_ticket_zone, ticket);
  2834. }
  2835. xlog_ticket_t *
  2836. xfs_log_ticket_get(
  2837. xlog_ticket_t *ticket)
  2838. {
  2839. ASSERT(atomic_read(&ticket->t_ref) > 0);
  2840. atomic_inc(&ticket->t_ref);
  2841. return ticket;
  2842. }
  2843. /*
  2844. * Allocate and initialise a new log ticket.
  2845. */
  2846. struct xlog_ticket *
  2847. xlog_ticket_alloc(
  2848. struct xlog *log,
  2849. int unit_bytes,
  2850. int cnt,
  2851. char client,
  2852. bool permanent,
  2853. xfs_km_flags_t alloc_flags)
  2854. {
  2855. struct xlog_ticket *tic;
  2856. uint num_headers;
  2857. int iclog_space;
  2858. tic = kmem_zone_zalloc(xfs_log_ticket_zone, alloc_flags);
  2859. if (!tic)
  2860. return NULL;
  2861. /*
  2862. * Permanent reservations have up to 'cnt'-1 active log operations
  2863. * in the log. A unit in this case is the amount of space for one
  2864. * of these log operations. Normal reservations have a cnt of 1
  2865. * and their unit amount is the total amount of space required.
  2866. *
  2867. * The following lines of code account for non-transaction data
  2868. * which occupy space in the on-disk log.
  2869. *
  2870. * Normal form of a transaction is:
  2871. * <oph><trans-hdr><start-oph><reg1-oph><reg1><reg2-oph>...<commit-oph>
  2872. * and then there are LR hdrs, split-recs and roundoff at end of syncs.
  2873. *
  2874. * We need to account for all the leadup data and trailer data
  2875. * around the transaction data.
  2876. * And then we need to account for the worst case in terms of using
  2877. * more space.
  2878. * The worst case will happen if:
  2879. * - the placement of the transaction happens to be such that the
  2880. * roundoff is at its maximum
  2881. * - the transaction data is synced before the commit record is synced
  2882. * i.e. <transaction-data><roundoff> | <commit-rec><roundoff>
  2883. * Therefore the commit record is in its own Log Record.
  2884. * This can happen as the commit record is called with its
  2885. * own region to xlog_write().
  2886. * This then means that in the worst case, roundoff can happen for
  2887. * the commit-rec as well.
  2888. * The commit-rec is smaller than padding in this scenario and so it is
  2889. * not added separately.
  2890. */
  2891. /* for trans header */
  2892. unit_bytes += sizeof(xlog_op_header_t);
  2893. unit_bytes += sizeof(xfs_trans_header_t);
  2894. /* for start-rec */
  2895. unit_bytes += sizeof(xlog_op_header_t);
  2896. /*
  2897. * for LR headers - the space for data in an iclog is the size minus
  2898. * the space used for the headers. If we use the iclog size, then we
  2899. * undercalculate the number of headers required.
  2900. *
  2901. * Furthermore - the addition of op headers for split-recs might
  2902. * increase the space required enough to require more log and op
  2903. * headers, so take that into account too.
  2904. *
  2905. * IMPORTANT: This reservation makes the assumption that if this
  2906. * transaction is the first in an iclog and hence has the LR headers
  2907. * accounted to it, then the remaining space in the iclog is
  2908. * exclusively for this transaction. i.e. if the transaction is larger
  2909. * than the iclog, it will be the only thing in that iclog.
  2910. * Fundamentally, this means we must pass the entire log vector to
  2911. * xlog_write to guarantee this.
  2912. */
  2913. iclog_space = log->l_iclog_size - log->l_iclog_hsize;
  2914. num_headers = howmany(unit_bytes, iclog_space);
  2915. /* for split-recs - ophdrs added when data split over LRs */
  2916. unit_bytes += sizeof(xlog_op_header_t) * num_headers;
  2917. /* add extra header reservations if we overrun */
  2918. while (!num_headers ||
  2919. howmany(unit_bytes, iclog_space) > num_headers) {
  2920. unit_bytes += sizeof(xlog_op_header_t);
  2921. num_headers++;
  2922. }
  2923. unit_bytes += log->l_iclog_hsize * num_headers;
  2924. /* for commit-rec LR header - note: padding will subsume the ophdr */
  2925. unit_bytes += log->l_iclog_hsize;
  2926. /* for roundoff padding for transaction data and one for commit record */
  2927. if (xfs_sb_version_haslogv2(&log->l_mp->m_sb) &&
  2928. log->l_mp->m_sb.sb_logsunit > 1) {
  2929. /* log su roundoff */
  2930. unit_bytes += 2*log->l_mp->m_sb.sb_logsunit;
  2931. } else {
  2932. /* BB roundoff */
  2933. unit_bytes += 2*BBSIZE;
  2934. }
  2935. atomic_set(&tic->t_ref, 1);
  2936. tic->t_task = current;
  2937. INIT_LIST_HEAD(&tic->t_queue);
  2938. tic->t_unit_res = unit_bytes;
  2939. tic->t_curr_res = unit_bytes;
  2940. tic->t_cnt = cnt;
  2941. tic->t_ocnt = cnt;
  2942. tic->t_tid = random32();
  2943. tic->t_clientid = client;
  2944. tic->t_flags = XLOG_TIC_INITED;
  2945. tic->t_trans_type = 0;
  2946. if (permanent)
  2947. tic->t_flags |= XLOG_TIC_PERM_RESERV;
  2948. xlog_tic_reset_res(tic);
  2949. return tic;
  2950. }
  2951. /******************************************************************************
  2952. *
  2953. * Log debug routines
  2954. *
  2955. ******************************************************************************
  2956. */
  2957. #if defined(DEBUG)
  2958. /*
  2959. * Make sure that the destination ptr is within the valid data region of
  2960. * one of the iclogs. This uses backup pointers stored in a different
  2961. * part of the log in case we trash the log structure.
  2962. */
  2963. void
  2964. xlog_verify_dest_ptr(
  2965. struct xlog *log,
  2966. char *ptr)
  2967. {
  2968. int i;
  2969. int good_ptr = 0;
  2970. for (i = 0; i < log->l_iclog_bufs; i++) {
  2971. if (ptr >= log->l_iclog_bak[i] &&
  2972. ptr <= log->l_iclog_bak[i] + log->l_iclog_size)
  2973. good_ptr++;
  2974. }
  2975. if (!good_ptr)
  2976. xfs_emerg(log->l_mp, "%s: invalid ptr", __func__);
  2977. }
  2978. /*
  2979. * Check to make sure the grant write head didn't just over lap the tail. If
  2980. * the cycles are the same, we can't be overlapping. Otherwise, make sure that
  2981. * the cycles differ by exactly one and check the byte count.
  2982. *
  2983. * This check is run unlocked, so can give false positives. Rather than assert
  2984. * on failures, use a warn-once flag and a panic tag to allow the admin to
  2985. * determine if they want to panic the machine when such an error occurs. For
  2986. * debug kernels this will have the same effect as using an assert but, unlinke
  2987. * an assert, it can be turned off at runtime.
  2988. */
  2989. STATIC void
  2990. xlog_verify_grant_tail(
  2991. struct xlog *log)
  2992. {
  2993. int tail_cycle, tail_blocks;
  2994. int cycle, space;
  2995. xlog_crack_grant_head(&log->l_write_head.grant, &cycle, &space);
  2996. xlog_crack_atomic_lsn(&log->l_tail_lsn, &tail_cycle, &tail_blocks);
  2997. if (tail_cycle != cycle) {
  2998. if (cycle - 1 != tail_cycle &&
  2999. !(log->l_flags & XLOG_TAIL_WARN)) {
  3000. xfs_alert_tag(log->l_mp, XFS_PTAG_LOGRES,
  3001. "%s: cycle - 1 != tail_cycle", __func__);
  3002. log->l_flags |= XLOG_TAIL_WARN;
  3003. }
  3004. if (space > BBTOB(tail_blocks) &&
  3005. !(log->l_flags & XLOG_TAIL_WARN)) {
  3006. xfs_alert_tag(log->l_mp, XFS_PTAG_LOGRES,
  3007. "%s: space > BBTOB(tail_blocks)", __func__);
  3008. log->l_flags |= XLOG_TAIL_WARN;
  3009. }
  3010. }
  3011. }
  3012. /* check if it will fit */
  3013. STATIC void
  3014. xlog_verify_tail_lsn(
  3015. struct xlog *log,
  3016. struct xlog_in_core *iclog,
  3017. xfs_lsn_t tail_lsn)
  3018. {
  3019. int blocks;
  3020. if (CYCLE_LSN(tail_lsn) == log->l_prev_cycle) {
  3021. blocks =
  3022. log->l_logBBsize - (log->l_prev_block - BLOCK_LSN(tail_lsn));
  3023. if (blocks < BTOBB(iclog->ic_offset)+BTOBB(log->l_iclog_hsize))
  3024. xfs_emerg(log->l_mp, "%s: ran out of log space", __func__);
  3025. } else {
  3026. ASSERT(CYCLE_LSN(tail_lsn)+1 == log->l_prev_cycle);
  3027. if (BLOCK_LSN(tail_lsn) == log->l_prev_block)
  3028. xfs_emerg(log->l_mp, "%s: tail wrapped", __func__);
  3029. blocks = BLOCK_LSN(tail_lsn) - log->l_prev_block;
  3030. if (blocks < BTOBB(iclog->ic_offset) + 1)
  3031. xfs_emerg(log->l_mp, "%s: ran out of log space", __func__);
  3032. }
  3033. } /* xlog_verify_tail_lsn */
  3034. /*
  3035. * Perform a number of checks on the iclog before writing to disk.
  3036. *
  3037. * 1. Make sure the iclogs are still circular
  3038. * 2. Make sure we have a good magic number
  3039. * 3. Make sure we don't have magic numbers in the data
  3040. * 4. Check fields of each log operation header for:
  3041. * A. Valid client identifier
  3042. * B. tid ptr value falls in valid ptr space (user space code)
  3043. * C. Length in log record header is correct according to the
  3044. * individual operation headers within record.
  3045. * 5. When a bwrite will occur within 5 blocks of the front of the physical
  3046. * log, check the preceding blocks of the physical log to make sure all
  3047. * the cycle numbers agree with the current cycle number.
  3048. */
  3049. STATIC void
  3050. xlog_verify_iclog(
  3051. struct xlog *log,
  3052. struct xlog_in_core *iclog,
  3053. int count,
  3054. boolean_t syncing)
  3055. {
  3056. xlog_op_header_t *ophead;
  3057. xlog_in_core_t *icptr;
  3058. xlog_in_core_2_t *xhdr;
  3059. xfs_caddr_t ptr;
  3060. xfs_caddr_t base_ptr;
  3061. __psint_t field_offset;
  3062. __uint8_t clientid;
  3063. int len, i, j, k, op_len;
  3064. int idx;
  3065. /* check validity of iclog pointers */
  3066. spin_lock(&log->l_icloglock);
  3067. icptr = log->l_iclog;
  3068. for (i=0; i < log->l_iclog_bufs; i++) {
  3069. if (icptr == NULL)
  3070. xfs_emerg(log->l_mp, "%s: invalid ptr", __func__);
  3071. icptr = icptr->ic_next;
  3072. }
  3073. if (icptr != log->l_iclog)
  3074. xfs_emerg(log->l_mp, "%s: corrupt iclog ring", __func__);
  3075. spin_unlock(&log->l_icloglock);
  3076. /* check log magic numbers */
  3077. if (iclog->ic_header.h_magicno != cpu_to_be32(XLOG_HEADER_MAGIC_NUM))
  3078. xfs_emerg(log->l_mp, "%s: invalid magic num", __func__);
  3079. ptr = (xfs_caddr_t) &iclog->ic_header;
  3080. for (ptr += BBSIZE; ptr < ((xfs_caddr_t)&iclog->ic_header) + count;
  3081. ptr += BBSIZE) {
  3082. if (*(__be32 *)ptr == cpu_to_be32(XLOG_HEADER_MAGIC_NUM))
  3083. xfs_emerg(log->l_mp, "%s: unexpected magic num",
  3084. __func__);
  3085. }
  3086. /* check fields */
  3087. len = be32_to_cpu(iclog->ic_header.h_num_logops);
  3088. ptr = iclog->ic_datap;
  3089. base_ptr = ptr;
  3090. ophead = (xlog_op_header_t *)ptr;
  3091. xhdr = iclog->ic_data;
  3092. for (i = 0; i < len; i++) {
  3093. ophead = (xlog_op_header_t *)ptr;
  3094. /* clientid is only 1 byte */
  3095. field_offset = (__psint_t)
  3096. ((xfs_caddr_t)&(ophead->oh_clientid) - base_ptr);
  3097. if (syncing == B_FALSE || (field_offset & 0x1ff)) {
  3098. clientid = ophead->oh_clientid;
  3099. } else {
  3100. idx = BTOBBT((xfs_caddr_t)&(ophead->oh_clientid) - iclog->ic_datap);
  3101. if (idx >= (XLOG_HEADER_CYCLE_SIZE / BBSIZE)) {
  3102. j = idx / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
  3103. k = idx % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
  3104. clientid = xlog_get_client_id(
  3105. xhdr[j].hic_xheader.xh_cycle_data[k]);
  3106. } else {
  3107. clientid = xlog_get_client_id(
  3108. iclog->ic_header.h_cycle_data[idx]);
  3109. }
  3110. }
  3111. if (clientid != XFS_TRANSACTION && clientid != XFS_LOG)
  3112. xfs_warn(log->l_mp,
  3113. "%s: invalid clientid %d op 0x%p offset 0x%lx",
  3114. __func__, clientid, ophead,
  3115. (unsigned long)field_offset);
  3116. /* check length */
  3117. field_offset = (__psint_t)
  3118. ((xfs_caddr_t)&(ophead->oh_len) - base_ptr);
  3119. if (syncing == B_FALSE || (field_offset & 0x1ff)) {
  3120. op_len = be32_to_cpu(ophead->oh_len);
  3121. } else {
  3122. idx = BTOBBT((__psint_t)&ophead->oh_len -
  3123. (__psint_t)iclog->ic_datap);
  3124. if (idx >= (XLOG_HEADER_CYCLE_SIZE / BBSIZE)) {
  3125. j = idx / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
  3126. k = idx % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
  3127. op_len = be32_to_cpu(xhdr[j].hic_xheader.xh_cycle_data[k]);
  3128. } else {
  3129. op_len = be32_to_cpu(iclog->ic_header.h_cycle_data[idx]);
  3130. }
  3131. }
  3132. ptr += sizeof(xlog_op_header_t) + op_len;
  3133. }
  3134. } /* xlog_verify_iclog */
  3135. #endif
  3136. /*
  3137. * Mark all iclogs IOERROR. l_icloglock is held by the caller.
  3138. */
  3139. STATIC int
  3140. xlog_state_ioerror(
  3141. struct xlog *log)
  3142. {
  3143. xlog_in_core_t *iclog, *ic;
  3144. iclog = log->l_iclog;
  3145. if (! (iclog->ic_state & XLOG_STATE_IOERROR)) {
  3146. /*
  3147. * Mark all the incore logs IOERROR.
  3148. * From now on, no log flushes will result.
  3149. */
  3150. ic = iclog;
  3151. do {
  3152. ic->ic_state = XLOG_STATE_IOERROR;
  3153. ic = ic->ic_next;
  3154. } while (ic != iclog);
  3155. return 0;
  3156. }
  3157. /*
  3158. * Return non-zero, if state transition has already happened.
  3159. */
  3160. return 1;
  3161. }
  3162. /*
  3163. * This is called from xfs_force_shutdown, when we're forcibly
  3164. * shutting down the filesystem, typically because of an IO error.
  3165. * Our main objectives here are to make sure that:
  3166. * a. the filesystem gets marked 'SHUTDOWN' for all interested
  3167. * parties to find out, 'atomically'.
  3168. * b. those who're sleeping on log reservations, pinned objects and
  3169. * other resources get woken up, and be told the bad news.
  3170. * c. nothing new gets queued up after (a) and (b) are done.
  3171. * d. if !logerror, flush the iclogs to disk, then seal them off
  3172. * for business.
  3173. *
  3174. * Note: for delayed logging the !logerror case needs to flush the regions
  3175. * held in memory out to the iclogs before flushing them to disk. This needs
  3176. * to be done before the log is marked as shutdown, otherwise the flush to the
  3177. * iclogs will fail.
  3178. */
  3179. int
  3180. xfs_log_force_umount(
  3181. struct xfs_mount *mp,
  3182. int logerror)
  3183. {
  3184. struct xlog *log;
  3185. int retval;
  3186. log = mp->m_log;
  3187. /*
  3188. * If this happens during log recovery, don't worry about
  3189. * locking; the log isn't open for business yet.
  3190. */
  3191. if (!log ||
  3192. log->l_flags & XLOG_ACTIVE_RECOVERY) {
  3193. mp->m_flags |= XFS_MOUNT_FS_SHUTDOWN;
  3194. if (mp->m_sb_bp)
  3195. XFS_BUF_DONE(mp->m_sb_bp);
  3196. return 0;
  3197. }
  3198. /*
  3199. * Somebody could've already done the hard work for us.
  3200. * No need to get locks for this.
  3201. */
  3202. if (logerror && log->l_iclog->ic_state & XLOG_STATE_IOERROR) {
  3203. ASSERT(XLOG_FORCED_SHUTDOWN(log));
  3204. return 1;
  3205. }
  3206. retval = 0;
  3207. /*
  3208. * Flush the in memory commit item list before marking the log as
  3209. * being shut down. We need to do it in this order to ensure all the
  3210. * completed transactions are flushed to disk with the xfs_log_force()
  3211. * call below.
  3212. */
  3213. if (!logerror)
  3214. xlog_cil_force(log);
  3215. /*
  3216. * mark the filesystem and the as in a shutdown state and wake
  3217. * everybody up to tell them the bad news.
  3218. */
  3219. spin_lock(&log->l_icloglock);
  3220. mp->m_flags |= XFS_MOUNT_FS_SHUTDOWN;
  3221. if (mp->m_sb_bp)
  3222. XFS_BUF_DONE(mp->m_sb_bp);
  3223. /*
  3224. * This flag is sort of redundant because of the mount flag, but
  3225. * it's good to maintain the separation between the log and the rest
  3226. * of XFS.
  3227. */
  3228. log->l_flags |= XLOG_IO_ERROR;
  3229. /*
  3230. * If we hit a log error, we want to mark all the iclogs IOERROR
  3231. * while we're still holding the loglock.
  3232. */
  3233. if (logerror)
  3234. retval = xlog_state_ioerror(log);
  3235. spin_unlock(&log->l_icloglock);
  3236. /*
  3237. * We don't want anybody waiting for log reservations after this. That
  3238. * means we have to wake up everybody queued up on reserveq as well as
  3239. * writeq. In addition, we make sure in xlog_{re}grant_log_space that
  3240. * we don't enqueue anything once the SHUTDOWN flag is set, and this
  3241. * action is protected by the grant locks.
  3242. */
  3243. xlog_grant_head_wake_all(&log->l_reserve_head);
  3244. xlog_grant_head_wake_all(&log->l_write_head);
  3245. if (!(log->l_iclog->ic_state & XLOG_STATE_IOERROR)) {
  3246. ASSERT(!logerror);
  3247. /*
  3248. * Force the incore logs to disk before shutting the
  3249. * log down completely.
  3250. */
  3251. _xfs_log_force(mp, XFS_LOG_SYNC, NULL);
  3252. spin_lock(&log->l_icloglock);
  3253. retval = xlog_state_ioerror(log);
  3254. spin_unlock(&log->l_icloglock);
  3255. }
  3256. /*
  3257. * Wake up everybody waiting on xfs_log_force.
  3258. * Callback all log item committed functions as if the
  3259. * log writes were completed.
  3260. */
  3261. xlog_state_do_callback(log, XFS_LI_ABORTED, NULL);
  3262. #ifdef XFSERRORDEBUG
  3263. {
  3264. xlog_in_core_t *iclog;
  3265. spin_lock(&log->l_icloglock);
  3266. iclog = log->l_iclog;
  3267. do {
  3268. ASSERT(iclog->ic_callback == 0);
  3269. iclog = iclog->ic_next;
  3270. } while (iclog != log->l_iclog);
  3271. spin_unlock(&log->l_icloglock);
  3272. }
  3273. #endif
  3274. /* return non-zero if log IOERROR transition had already happened */
  3275. return retval;
  3276. }
  3277. STATIC int
  3278. xlog_iclogs_empty(
  3279. struct xlog *log)
  3280. {
  3281. xlog_in_core_t *iclog;
  3282. iclog = log->l_iclog;
  3283. do {
  3284. /* endianness does not matter here, zero is zero in
  3285. * any language.
  3286. */
  3287. if (iclog->ic_header.h_num_logops)
  3288. return 0;
  3289. iclog = iclog->ic_next;
  3290. } while (iclog != log->l_iclog);
  3291. return 1;
  3292. }