ll_rw_blk.c 102 KB

12345678910111213141516171819202122232425262728293031323334353637383940414243444546474849505152535455565758596061626364656667686970717273747576777879808182838485868788899091929394959697989910010110210310410510610710810911011111211311411511611711811912012112212312412512612712812913013113213313413513613713813914014114214314414514614714814915015115215315415515615715815916016116216316416516616716816917017117217317417517617717817918018118218318418518618718818919019119219319419519619719819920020120220320420520620720820921021121221321421521621721821922022122222322422522622722822923023123223323423523623723823924024124224324424524624724824925025125225325425525625725825926026126226326426526626726826927027127227327427527627727827928028128228328428528628728828929029129229329429529629729829930030130230330430530630730830931031131231331431531631731831932032132232332432532632732832933033133233333433533633733833934034134234334434534634734834935035135235335435535635735835936036136236336436536636736836937037137237337437537637737837938038138238338438538638738838939039139239339439539639739839940040140240340440540640740840941041141241341441541641741841942042142242342442542642742842943043143243343443543643743843944044144244344444544644744844945045145245345445545645745845946046146246346446546646746846947047147247347447547647747847948048148248348448548648748848949049149249349449549649749849950050150250350450550650750850951051151251351451551651751851952052152252352452552652752852953053153253353453553653753853954054154254354454554654754854955055155255355455555655755855956056156256356456556656756856957057157257357457557657757857958058158258358458558658758858959059159259359459559659759859960060160260360460560660760860961061161261361461561661761861962062162262362462562662762862963063163263363463563663763863964064164264364464564664764864965065165265365465565665765865966066166266366466566666766866967067167267367467567667767867968068168268368468568668768868969069169269369469569669769869970070170270370470570670770870971071171271371471571671771871972072172272372472572672772872973073173273373473573673773873974074174274374474574674774874975075175275375475575675775875976076176276376476576676776876977077177277377477577677777877978078178278378478578678778878979079179279379479579679779879980080180280380480580680780880981081181281381481581681781881982082182282382482582682782882983083183283383483583683783883984084184284384484584684784884985085185285385485585685785885986086186286386486586686786886987087187287387487587687787887988088188288388488588688788888989089189289389489589689789889990090190290390490590690790890991091191291391491591691791891992092192292392492592692792892993093193293393493593693793893994094194294394494594694794894995095195295395495595695795895996096196296396496596696796896997097197297397497597697797897998098198298398498598698798898999099199299399499599699799899910001001100210031004100510061007100810091010101110121013101410151016101710181019102010211022102310241025102610271028102910301031103210331034103510361037103810391040104110421043104410451046104710481049105010511052105310541055105610571058105910601061106210631064106510661067106810691070107110721073107410751076107710781079108010811082108310841085108610871088108910901091109210931094109510961097109810991100110111021103110411051106110711081109111011111112111311141115111611171118111911201121112211231124112511261127112811291130113111321133113411351136113711381139114011411142114311441145114611471148114911501151115211531154115511561157115811591160116111621163116411651166116711681169117011711172117311741175117611771178117911801181118211831184118511861187118811891190119111921193119411951196119711981199120012011202120312041205120612071208120912101211121212131214121512161217121812191220122112221223122412251226122712281229123012311232123312341235123612371238123912401241124212431244124512461247124812491250125112521253125412551256125712581259126012611262126312641265126612671268126912701271127212731274127512761277127812791280128112821283128412851286128712881289129012911292129312941295129612971298129913001301130213031304130513061307130813091310131113121313131413151316131713181319132013211322132313241325132613271328132913301331133213331334133513361337133813391340134113421343134413451346134713481349135013511352135313541355135613571358135913601361136213631364136513661367136813691370137113721373137413751376137713781379138013811382138313841385138613871388138913901391139213931394139513961397139813991400140114021403140414051406140714081409141014111412141314141415141614171418141914201421142214231424142514261427142814291430143114321433143414351436143714381439144014411442144314441445144614471448144914501451145214531454145514561457145814591460146114621463146414651466146714681469147014711472147314741475147614771478147914801481148214831484148514861487148814891490149114921493149414951496149714981499150015011502150315041505150615071508150915101511151215131514151515161517151815191520152115221523152415251526152715281529153015311532153315341535153615371538153915401541154215431544154515461547154815491550155115521553155415551556155715581559156015611562156315641565156615671568156915701571157215731574157515761577157815791580158115821583158415851586158715881589159015911592159315941595159615971598159916001601160216031604160516061607160816091610161116121613161416151616161716181619162016211622162316241625162616271628162916301631163216331634163516361637163816391640164116421643164416451646164716481649165016511652165316541655165616571658165916601661166216631664166516661667166816691670167116721673167416751676167716781679168016811682168316841685168616871688168916901691169216931694169516961697169816991700170117021703170417051706170717081709171017111712171317141715171617171718171917201721172217231724172517261727172817291730173117321733173417351736173717381739174017411742174317441745174617471748174917501751175217531754175517561757175817591760176117621763176417651766176717681769177017711772177317741775177617771778177917801781178217831784178517861787178817891790179117921793179417951796179717981799180018011802180318041805180618071808180918101811181218131814181518161817181818191820182118221823182418251826182718281829183018311832183318341835183618371838183918401841184218431844184518461847184818491850185118521853185418551856185718581859186018611862186318641865186618671868186918701871187218731874187518761877187818791880188118821883188418851886188718881889189018911892189318941895189618971898189919001901190219031904190519061907190819091910191119121913191419151916191719181919192019211922192319241925192619271928192919301931193219331934193519361937193819391940194119421943194419451946194719481949195019511952195319541955195619571958195919601961196219631964196519661967196819691970197119721973197419751976197719781979198019811982198319841985198619871988198919901991199219931994199519961997199819992000200120022003200420052006200720082009201020112012201320142015201620172018201920202021202220232024202520262027202820292030203120322033203420352036203720382039204020412042204320442045204620472048204920502051205220532054205520562057205820592060206120622063206420652066206720682069207020712072207320742075207620772078207920802081208220832084208520862087208820892090209120922093209420952096209720982099210021012102210321042105210621072108210921102111211221132114211521162117211821192120212121222123212421252126212721282129213021312132213321342135213621372138213921402141214221432144214521462147214821492150215121522153215421552156215721582159216021612162216321642165216621672168216921702171217221732174217521762177217821792180218121822183218421852186218721882189219021912192219321942195219621972198219922002201220222032204220522062207220822092210221122122213221422152216221722182219222022212222222322242225222622272228222922302231223222332234223522362237223822392240224122422243224422452246224722482249225022512252225322542255225622572258225922602261226222632264226522662267226822692270227122722273227422752276227722782279228022812282228322842285228622872288228922902291229222932294229522962297229822992300230123022303230423052306230723082309231023112312231323142315231623172318231923202321232223232324232523262327232823292330233123322333233423352336233723382339234023412342234323442345234623472348234923502351235223532354235523562357235823592360236123622363236423652366236723682369237023712372237323742375237623772378237923802381238223832384238523862387238823892390239123922393239423952396239723982399240024012402240324042405240624072408240924102411241224132414241524162417241824192420242124222423242424252426242724282429243024312432243324342435243624372438243924402441244224432444244524462447244824492450245124522453245424552456245724582459246024612462246324642465246624672468246924702471247224732474247524762477247824792480248124822483248424852486248724882489249024912492249324942495249624972498249925002501250225032504250525062507250825092510251125122513251425152516251725182519252025212522252325242525252625272528252925302531253225332534253525362537253825392540254125422543254425452546254725482549255025512552255325542555255625572558255925602561256225632564256525662567256825692570257125722573257425752576257725782579258025812582258325842585258625872588258925902591259225932594259525962597259825992600260126022603260426052606260726082609261026112612261326142615261626172618261926202621262226232624262526262627262826292630263126322633263426352636263726382639264026412642264326442645264626472648264926502651265226532654265526562657265826592660266126622663266426652666266726682669267026712672267326742675267626772678267926802681268226832684268526862687268826892690269126922693269426952696269726982699270027012702270327042705270627072708270927102711271227132714271527162717271827192720272127222723272427252726272727282729273027312732273327342735273627372738273927402741274227432744274527462747274827492750275127522753275427552756275727582759276027612762276327642765276627672768276927702771277227732774277527762777277827792780278127822783278427852786278727882789279027912792279327942795279627972798279928002801280228032804280528062807280828092810281128122813281428152816281728182819282028212822282328242825282628272828282928302831283228332834283528362837283828392840284128422843284428452846284728482849285028512852285328542855285628572858285928602861286228632864286528662867286828692870287128722873287428752876287728782879288028812882288328842885288628872888288928902891289228932894289528962897289828992900290129022903290429052906290729082909291029112912291329142915291629172918291929202921292229232924292529262927292829292930293129322933293429352936293729382939294029412942294329442945294629472948294929502951295229532954295529562957295829592960296129622963296429652966296729682969297029712972297329742975297629772978297929802981298229832984298529862987298829892990299129922993299429952996299729982999300030013002300330043005300630073008300930103011301230133014301530163017301830193020302130223023302430253026302730283029303030313032303330343035303630373038303930403041304230433044304530463047304830493050305130523053305430553056305730583059306030613062306330643065306630673068306930703071307230733074307530763077307830793080308130823083308430853086308730883089309030913092309330943095309630973098309931003101310231033104310531063107310831093110311131123113311431153116311731183119312031213122312331243125312631273128312931303131313231333134313531363137313831393140314131423143314431453146314731483149315031513152315331543155315631573158315931603161316231633164316531663167316831693170317131723173317431753176317731783179318031813182318331843185318631873188318931903191319231933194319531963197319831993200320132023203320432053206320732083209321032113212321332143215321632173218321932203221322232233224322532263227322832293230323132323233323432353236323732383239324032413242324332443245324632473248324932503251325232533254325532563257325832593260326132623263326432653266326732683269327032713272327332743275327632773278327932803281328232833284328532863287328832893290329132923293329432953296329732983299330033013302330333043305330633073308330933103311331233133314331533163317331833193320332133223323332433253326332733283329333033313332333333343335333633373338333933403341334233433344334533463347334833493350335133523353335433553356335733583359336033613362336333643365336633673368336933703371337233733374337533763377337833793380338133823383338433853386338733883389339033913392339333943395339633973398339934003401340234033404340534063407340834093410341134123413341434153416341734183419342034213422342334243425342634273428342934303431343234333434343534363437343834393440344134423443344434453446344734483449345034513452345334543455345634573458345934603461346234633464346534663467346834693470347134723473347434753476347734783479348034813482348334843485348634873488348934903491349234933494349534963497349834993500350135023503350435053506350735083509351035113512351335143515351635173518351935203521352235233524352535263527352835293530353135323533353435353536353735383539354035413542354335443545354635473548354935503551355235533554355535563557355835593560356135623563356435653566356735683569357035713572357335743575357635773578357935803581358235833584358535863587358835893590359135923593359435953596359735983599360036013602360336043605360636073608360936103611361236133614361536163617361836193620362136223623362436253626362736283629363036313632363336343635363636373638363936403641364236433644364536463647364836493650365136523653365436553656365736583659366036613662366336643665366636673668366936703671367236733674367536763677367836793680368136823683368436853686368736883689369036913692369336943695369636973698369937003701370237033704370537063707370837093710371137123713371437153716371737183719372037213722372337243725372637273728372937303731373237333734373537363737373837393740374137423743374437453746374737483749375037513752375337543755375637573758375937603761376237633764376537663767376837693770377137723773377437753776377737783779378037813782378337843785378637873788378937903791379237933794379537963797379837993800380138023803380438053806380738083809381038113812381338143815381638173818381938203821382238233824382538263827382838293830383138323833383438353836383738383839384038413842384338443845384638473848384938503851385238533854385538563857385838593860386138623863386438653866386738683869387038713872387338743875387638773878387938803881388238833884388538863887388838893890389138923893389438953896389738983899390039013902390339043905390639073908390939103911391239133914391539163917391839193920392139223923392439253926392739283929393039313932393339343935393639373938393939403941394239433944394539463947394839493950395139523953395439553956395739583959396039613962396339643965396639673968396939703971397239733974397539763977397839793980398139823983398439853986398739883989399039913992399339943995399639973998399940004001400240034004400540064007400840094010401140124013401440154016401740184019402040214022402340244025402640274028402940304031403240334034403540364037403840394040404140424043404440454046404740484049405040514052405340544055
  1. /*
  2. * Copyright (C) 1991, 1992 Linus Torvalds
  3. * Copyright (C) 1994, Karl Keyte: Added support for disk statistics
  4. * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
  5. * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
  6. * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au> - July2000
  7. * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
  8. */
  9. /*
  10. * This handles all read/write requests to block devices
  11. */
  12. #include <linux/kernel.h>
  13. #include <linux/module.h>
  14. #include <linux/backing-dev.h>
  15. #include <linux/bio.h>
  16. #include <linux/blkdev.h>
  17. #include <linux/highmem.h>
  18. #include <linux/mm.h>
  19. #include <linux/kernel_stat.h>
  20. #include <linux/string.h>
  21. #include <linux/init.h>
  22. #include <linux/bootmem.h> /* for max_pfn/max_low_pfn */
  23. #include <linux/completion.h>
  24. #include <linux/slab.h>
  25. #include <linux/swap.h>
  26. #include <linux/writeback.h>
  27. #include <linux/task_io_accounting_ops.h>
  28. #include <linux/interrupt.h>
  29. #include <linux/cpu.h>
  30. #include <linux/blktrace_api.h>
  31. #include <linux/fault-inject.h>
  32. /*
  33. * for max sense size
  34. */
  35. #include <scsi/scsi_cmnd.h>
  36. static void blk_unplug_work(struct work_struct *work);
  37. static void blk_unplug_timeout(unsigned long data);
  38. static void drive_stat_acct(struct request *rq, int nr_sectors, int new_io);
  39. static void init_request_from_bio(struct request *req, struct bio *bio);
  40. static int __make_request(request_queue_t *q, struct bio *bio);
  41. static struct io_context *current_io_context(gfp_t gfp_flags, int node);
  42. /*
  43. * For the allocated request tables
  44. */
  45. static struct kmem_cache *request_cachep;
  46. /*
  47. * For queue allocation
  48. */
  49. static struct kmem_cache *requestq_cachep;
  50. /*
  51. * For io context allocations
  52. */
  53. static struct kmem_cache *iocontext_cachep;
  54. /*
  55. * Controlling structure to kblockd
  56. */
  57. static struct workqueue_struct *kblockd_workqueue;
  58. unsigned long blk_max_low_pfn, blk_max_pfn;
  59. EXPORT_SYMBOL(blk_max_low_pfn);
  60. EXPORT_SYMBOL(blk_max_pfn);
  61. static DEFINE_PER_CPU(struct list_head, blk_cpu_done);
  62. /* Amount of time in which a process may batch requests */
  63. #define BLK_BATCH_TIME (HZ/50UL)
  64. /* Number of requests a "batching" process may submit */
  65. #define BLK_BATCH_REQ 32
  66. /*
  67. * Return the threshold (number of used requests) at which the queue is
  68. * considered to be congested. It include a little hysteresis to keep the
  69. * context switch rate down.
  70. */
  71. static inline int queue_congestion_on_threshold(struct request_queue *q)
  72. {
  73. return q->nr_congestion_on;
  74. }
  75. /*
  76. * The threshold at which a queue is considered to be uncongested
  77. */
  78. static inline int queue_congestion_off_threshold(struct request_queue *q)
  79. {
  80. return q->nr_congestion_off;
  81. }
  82. static void blk_queue_congestion_threshold(struct request_queue *q)
  83. {
  84. int nr;
  85. nr = q->nr_requests - (q->nr_requests / 8) + 1;
  86. if (nr > q->nr_requests)
  87. nr = q->nr_requests;
  88. q->nr_congestion_on = nr;
  89. nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
  90. if (nr < 1)
  91. nr = 1;
  92. q->nr_congestion_off = nr;
  93. }
  94. /**
  95. * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
  96. * @bdev: device
  97. *
  98. * Locates the passed device's request queue and returns the address of its
  99. * backing_dev_info
  100. *
  101. * Will return NULL if the request queue cannot be located.
  102. */
  103. struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
  104. {
  105. struct backing_dev_info *ret = NULL;
  106. request_queue_t *q = bdev_get_queue(bdev);
  107. if (q)
  108. ret = &q->backing_dev_info;
  109. return ret;
  110. }
  111. EXPORT_SYMBOL(blk_get_backing_dev_info);
  112. /**
  113. * blk_queue_prep_rq - set a prepare_request function for queue
  114. * @q: queue
  115. * @pfn: prepare_request function
  116. *
  117. * It's possible for a queue to register a prepare_request callback which
  118. * is invoked before the request is handed to the request_fn. The goal of
  119. * the function is to prepare a request for I/O, it can be used to build a
  120. * cdb from the request data for instance.
  121. *
  122. */
  123. void blk_queue_prep_rq(request_queue_t *q, prep_rq_fn *pfn)
  124. {
  125. q->prep_rq_fn = pfn;
  126. }
  127. EXPORT_SYMBOL(blk_queue_prep_rq);
  128. /**
  129. * blk_queue_merge_bvec - set a merge_bvec function for queue
  130. * @q: queue
  131. * @mbfn: merge_bvec_fn
  132. *
  133. * Usually queues have static limitations on the max sectors or segments that
  134. * we can put in a request. Stacking drivers may have some settings that
  135. * are dynamic, and thus we have to query the queue whether it is ok to
  136. * add a new bio_vec to a bio at a given offset or not. If the block device
  137. * has such limitations, it needs to register a merge_bvec_fn to control
  138. * the size of bio's sent to it. Note that a block device *must* allow a
  139. * single page to be added to an empty bio. The block device driver may want
  140. * to use the bio_split() function to deal with these bio's. By default
  141. * no merge_bvec_fn is defined for a queue, and only the fixed limits are
  142. * honored.
  143. */
  144. void blk_queue_merge_bvec(request_queue_t *q, merge_bvec_fn *mbfn)
  145. {
  146. q->merge_bvec_fn = mbfn;
  147. }
  148. EXPORT_SYMBOL(blk_queue_merge_bvec);
  149. void blk_queue_softirq_done(request_queue_t *q, softirq_done_fn *fn)
  150. {
  151. q->softirq_done_fn = fn;
  152. }
  153. EXPORT_SYMBOL(blk_queue_softirq_done);
  154. /**
  155. * blk_queue_make_request - define an alternate make_request function for a device
  156. * @q: the request queue for the device to be affected
  157. * @mfn: the alternate make_request function
  158. *
  159. * Description:
  160. * The normal way for &struct bios to be passed to a device
  161. * driver is for them to be collected into requests on a request
  162. * queue, and then to allow the device driver to select requests
  163. * off that queue when it is ready. This works well for many block
  164. * devices. However some block devices (typically virtual devices
  165. * such as md or lvm) do not benefit from the processing on the
  166. * request queue, and are served best by having the requests passed
  167. * directly to them. This can be achieved by providing a function
  168. * to blk_queue_make_request().
  169. *
  170. * Caveat:
  171. * The driver that does this *must* be able to deal appropriately
  172. * with buffers in "highmemory". This can be accomplished by either calling
  173. * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling
  174. * blk_queue_bounce() to create a buffer in normal memory.
  175. **/
  176. void blk_queue_make_request(request_queue_t * q, make_request_fn * mfn)
  177. {
  178. /*
  179. * set defaults
  180. */
  181. q->nr_requests = BLKDEV_MAX_RQ;
  182. blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS);
  183. blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS);
  184. q->make_request_fn = mfn;
  185. q->backing_dev_info.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
  186. q->backing_dev_info.state = 0;
  187. q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
  188. blk_queue_max_sectors(q, SAFE_MAX_SECTORS);
  189. blk_queue_hardsect_size(q, 512);
  190. blk_queue_dma_alignment(q, 511);
  191. blk_queue_congestion_threshold(q);
  192. q->nr_batching = BLK_BATCH_REQ;
  193. q->unplug_thresh = 4; /* hmm */
  194. q->unplug_delay = (3 * HZ) / 1000; /* 3 milliseconds */
  195. if (q->unplug_delay == 0)
  196. q->unplug_delay = 1;
  197. INIT_WORK(&q->unplug_work, blk_unplug_work);
  198. q->unplug_timer.function = blk_unplug_timeout;
  199. q->unplug_timer.data = (unsigned long)q;
  200. /*
  201. * by default assume old behaviour and bounce for any highmem page
  202. */
  203. blk_queue_bounce_limit(q, BLK_BOUNCE_HIGH);
  204. }
  205. EXPORT_SYMBOL(blk_queue_make_request);
  206. static void rq_init(request_queue_t *q, struct request *rq)
  207. {
  208. INIT_LIST_HEAD(&rq->queuelist);
  209. INIT_LIST_HEAD(&rq->donelist);
  210. rq->errors = 0;
  211. rq->bio = rq->biotail = NULL;
  212. INIT_HLIST_NODE(&rq->hash);
  213. RB_CLEAR_NODE(&rq->rb_node);
  214. rq->ioprio = 0;
  215. rq->buffer = NULL;
  216. rq->ref_count = 1;
  217. rq->q = q;
  218. rq->special = NULL;
  219. rq->data_len = 0;
  220. rq->data = NULL;
  221. rq->nr_phys_segments = 0;
  222. rq->sense = NULL;
  223. rq->end_io = NULL;
  224. rq->end_io_data = NULL;
  225. rq->completion_data = NULL;
  226. }
  227. /**
  228. * blk_queue_ordered - does this queue support ordered writes
  229. * @q: the request queue
  230. * @ordered: one of QUEUE_ORDERED_*
  231. * @prepare_flush_fn: rq setup helper for cache flush ordered writes
  232. *
  233. * Description:
  234. * For journalled file systems, doing ordered writes on a commit
  235. * block instead of explicitly doing wait_on_buffer (which is bad
  236. * for performance) can be a big win. Block drivers supporting this
  237. * feature should call this function and indicate so.
  238. *
  239. **/
  240. int blk_queue_ordered(request_queue_t *q, unsigned ordered,
  241. prepare_flush_fn *prepare_flush_fn)
  242. {
  243. if (ordered & (QUEUE_ORDERED_PREFLUSH | QUEUE_ORDERED_POSTFLUSH) &&
  244. prepare_flush_fn == NULL) {
  245. printk(KERN_ERR "blk_queue_ordered: prepare_flush_fn required\n");
  246. return -EINVAL;
  247. }
  248. if (ordered != QUEUE_ORDERED_NONE &&
  249. ordered != QUEUE_ORDERED_DRAIN &&
  250. ordered != QUEUE_ORDERED_DRAIN_FLUSH &&
  251. ordered != QUEUE_ORDERED_DRAIN_FUA &&
  252. ordered != QUEUE_ORDERED_TAG &&
  253. ordered != QUEUE_ORDERED_TAG_FLUSH &&
  254. ordered != QUEUE_ORDERED_TAG_FUA) {
  255. printk(KERN_ERR "blk_queue_ordered: bad value %d\n", ordered);
  256. return -EINVAL;
  257. }
  258. q->ordered = ordered;
  259. q->next_ordered = ordered;
  260. q->prepare_flush_fn = prepare_flush_fn;
  261. return 0;
  262. }
  263. EXPORT_SYMBOL(blk_queue_ordered);
  264. /**
  265. * blk_queue_issue_flush_fn - set function for issuing a flush
  266. * @q: the request queue
  267. * @iff: the function to be called issuing the flush
  268. *
  269. * Description:
  270. * If a driver supports issuing a flush command, the support is notified
  271. * to the block layer by defining it through this call.
  272. *
  273. **/
  274. void blk_queue_issue_flush_fn(request_queue_t *q, issue_flush_fn *iff)
  275. {
  276. q->issue_flush_fn = iff;
  277. }
  278. EXPORT_SYMBOL(blk_queue_issue_flush_fn);
  279. /*
  280. * Cache flushing for ordered writes handling
  281. */
  282. inline unsigned blk_ordered_cur_seq(request_queue_t *q)
  283. {
  284. if (!q->ordseq)
  285. return 0;
  286. return 1 << ffz(q->ordseq);
  287. }
  288. unsigned blk_ordered_req_seq(struct request *rq)
  289. {
  290. request_queue_t *q = rq->q;
  291. BUG_ON(q->ordseq == 0);
  292. if (rq == &q->pre_flush_rq)
  293. return QUEUE_ORDSEQ_PREFLUSH;
  294. if (rq == &q->bar_rq)
  295. return QUEUE_ORDSEQ_BAR;
  296. if (rq == &q->post_flush_rq)
  297. return QUEUE_ORDSEQ_POSTFLUSH;
  298. if ((rq->cmd_flags & REQ_ORDERED_COLOR) ==
  299. (q->orig_bar_rq->cmd_flags & REQ_ORDERED_COLOR))
  300. return QUEUE_ORDSEQ_DRAIN;
  301. else
  302. return QUEUE_ORDSEQ_DONE;
  303. }
  304. void blk_ordered_complete_seq(request_queue_t *q, unsigned seq, int error)
  305. {
  306. struct request *rq;
  307. int uptodate;
  308. if (error && !q->orderr)
  309. q->orderr = error;
  310. BUG_ON(q->ordseq & seq);
  311. q->ordseq |= seq;
  312. if (blk_ordered_cur_seq(q) != QUEUE_ORDSEQ_DONE)
  313. return;
  314. /*
  315. * Okay, sequence complete.
  316. */
  317. rq = q->orig_bar_rq;
  318. uptodate = q->orderr ? q->orderr : 1;
  319. q->ordseq = 0;
  320. end_that_request_first(rq, uptodate, rq->hard_nr_sectors);
  321. end_that_request_last(rq, uptodate);
  322. }
  323. static void pre_flush_end_io(struct request *rq, int error)
  324. {
  325. elv_completed_request(rq->q, rq);
  326. blk_ordered_complete_seq(rq->q, QUEUE_ORDSEQ_PREFLUSH, error);
  327. }
  328. static void bar_end_io(struct request *rq, int error)
  329. {
  330. elv_completed_request(rq->q, rq);
  331. blk_ordered_complete_seq(rq->q, QUEUE_ORDSEQ_BAR, error);
  332. }
  333. static void post_flush_end_io(struct request *rq, int error)
  334. {
  335. elv_completed_request(rq->q, rq);
  336. blk_ordered_complete_seq(rq->q, QUEUE_ORDSEQ_POSTFLUSH, error);
  337. }
  338. static void queue_flush(request_queue_t *q, unsigned which)
  339. {
  340. struct request *rq;
  341. rq_end_io_fn *end_io;
  342. if (which == QUEUE_ORDERED_PREFLUSH) {
  343. rq = &q->pre_flush_rq;
  344. end_io = pre_flush_end_io;
  345. } else {
  346. rq = &q->post_flush_rq;
  347. end_io = post_flush_end_io;
  348. }
  349. rq->cmd_flags = REQ_HARDBARRIER;
  350. rq_init(q, rq);
  351. rq->elevator_private = NULL;
  352. rq->elevator_private2 = NULL;
  353. rq->rq_disk = q->bar_rq.rq_disk;
  354. rq->end_io = end_io;
  355. q->prepare_flush_fn(q, rq);
  356. elv_insert(q, rq, ELEVATOR_INSERT_FRONT);
  357. }
  358. static inline struct request *start_ordered(request_queue_t *q,
  359. struct request *rq)
  360. {
  361. q->bi_size = 0;
  362. q->orderr = 0;
  363. q->ordered = q->next_ordered;
  364. q->ordseq |= QUEUE_ORDSEQ_STARTED;
  365. /*
  366. * Prep proxy barrier request.
  367. */
  368. blkdev_dequeue_request(rq);
  369. q->orig_bar_rq = rq;
  370. rq = &q->bar_rq;
  371. rq->cmd_flags = 0;
  372. rq_init(q, rq);
  373. if (bio_data_dir(q->orig_bar_rq->bio) == WRITE)
  374. rq->cmd_flags |= REQ_RW;
  375. rq->cmd_flags |= q->ordered & QUEUE_ORDERED_FUA ? REQ_FUA : 0;
  376. rq->elevator_private = NULL;
  377. rq->elevator_private2 = NULL;
  378. init_request_from_bio(rq, q->orig_bar_rq->bio);
  379. rq->end_io = bar_end_io;
  380. /*
  381. * Queue ordered sequence. As we stack them at the head, we
  382. * need to queue in reverse order. Note that we rely on that
  383. * no fs request uses ELEVATOR_INSERT_FRONT and thus no fs
  384. * request gets inbetween ordered sequence.
  385. */
  386. if (q->ordered & QUEUE_ORDERED_POSTFLUSH)
  387. queue_flush(q, QUEUE_ORDERED_POSTFLUSH);
  388. else
  389. q->ordseq |= QUEUE_ORDSEQ_POSTFLUSH;
  390. elv_insert(q, rq, ELEVATOR_INSERT_FRONT);
  391. if (q->ordered & QUEUE_ORDERED_PREFLUSH) {
  392. queue_flush(q, QUEUE_ORDERED_PREFLUSH);
  393. rq = &q->pre_flush_rq;
  394. } else
  395. q->ordseq |= QUEUE_ORDSEQ_PREFLUSH;
  396. if ((q->ordered & QUEUE_ORDERED_TAG) || q->in_flight == 0)
  397. q->ordseq |= QUEUE_ORDSEQ_DRAIN;
  398. else
  399. rq = NULL;
  400. return rq;
  401. }
  402. int blk_do_ordered(request_queue_t *q, struct request **rqp)
  403. {
  404. struct request *rq = *rqp;
  405. int is_barrier = blk_fs_request(rq) && blk_barrier_rq(rq);
  406. if (!q->ordseq) {
  407. if (!is_barrier)
  408. return 1;
  409. if (q->next_ordered != QUEUE_ORDERED_NONE) {
  410. *rqp = start_ordered(q, rq);
  411. return 1;
  412. } else {
  413. /*
  414. * This can happen when the queue switches to
  415. * ORDERED_NONE while this request is on it.
  416. */
  417. blkdev_dequeue_request(rq);
  418. end_that_request_first(rq, -EOPNOTSUPP,
  419. rq->hard_nr_sectors);
  420. end_that_request_last(rq, -EOPNOTSUPP);
  421. *rqp = NULL;
  422. return 0;
  423. }
  424. }
  425. /*
  426. * Ordered sequence in progress
  427. */
  428. /* Special requests are not subject to ordering rules. */
  429. if (!blk_fs_request(rq) &&
  430. rq != &q->pre_flush_rq && rq != &q->post_flush_rq)
  431. return 1;
  432. if (q->ordered & QUEUE_ORDERED_TAG) {
  433. /* Ordered by tag. Blocking the next barrier is enough. */
  434. if (is_barrier && rq != &q->bar_rq)
  435. *rqp = NULL;
  436. } else {
  437. /* Ordered by draining. Wait for turn. */
  438. WARN_ON(blk_ordered_req_seq(rq) < blk_ordered_cur_seq(q));
  439. if (blk_ordered_req_seq(rq) > blk_ordered_cur_seq(q))
  440. *rqp = NULL;
  441. }
  442. return 1;
  443. }
  444. static int flush_dry_bio_endio(struct bio *bio, unsigned int bytes, int error)
  445. {
  446. request_queue_t *q = bio->bi_private;
  447. struct bio_vec *bvec;
  448. int i;
  449. /*
  450. * This is dry run, restore bio_sector and size. We'll finish
  451. * this request again with the original bi_end_io after an
  452. * error occurs or post flush is complete.
  453. */
  454. q->bi_size += bytes;
  455. if (bio->bi_size)
  456. return 1;
  457. /* Rewind bvec's */
  458. bio->bi_idx = 0;
  459. bio_for_each_segment(bvec, bio, i) {
  460. bvec->bv_len += bvec->bv_offset;
  461. bvec->bv_offset = 0;
  462. }
  463. /* Reset bio */
  464. set_bit(BIO_UPTODATE, &bio->bi_flags);
  465. bio->bi_size = q->bi_size;
  466. bio->bi_sector -= (q->bi_size >> 9);
  467. q->bi_size = 0;
  468. return 0;
  469. }
  470. static int ordered_bio_endio(struct request *rq, struct bio *bio,
  471. unsigned int nbytes, int error)
  472. {
  473. request_queue_t *q = rq->q;
  474. bio_end_io_t *endio;
  475. void *private;
  476. if (&q->bar_rq != rq)
  477. return 0;
  478. /*
  479. * Okay, this is the barrier request in progress, dry finish it.
  480. */
  481. if (error && !q->orderr)
  482. q->orderr = error;
  483. endio = bio->bi_end_io;
  484. private = bio->bi_private;
  485. bio->bi_end_io = flush_dry_bio_endio;
  486. bio->bi_private = q;
  487. bio_endio(bio, nbytes, error);
  488. bio->bi_end_io = endio;
  489. bio->bi_private = private;
  490. return 1;
  491. }
  492. /**
  493. * blk_queue_bounce_limit - set bounce buffer limit for queue
  494. * @q: the request queue for the device
  495. * @dma_addr: bus address limit
  496. *
  497. * Description:
  498. * Different hardware can have different requirements as to what pages
  499. * it can do I/O directly to. A low level driver can call
  500. * blk_queue_bounce_limit to have lower memory pages allocated as bounce
  501. * buffers for doing I/O to pages residing above @page.
  502. **/
  503. void blk_queue_bounce_limit(request_queue_t *q, u64 dma_addr)
  504. {
  505. unsigned long bounce_pfn = dma_addr >> PAGE_SHIFT;
  506. int dma = 0;
  507. q->bounce_gfp = GFP_NOIO;
  508. #if BITS_PER_LONG == 64
  509. /* Assume anything <= 4GB can be handled by IOMMU.
  510. Actually some IOMMUs can handle everything, but I don't
  511. know of a way to test this here. */
  512. if (bounce_pfn < (min_t(u64,0xffffffff,BLK_BOUNCE_HIGH) >> PAGE_SHIFT))
  513. dma = 1;
  514. q->bounce_pfn = max_low_pfn;
  515. #else
  516. if (bounce_pfn < blk_max_low_pfn)
  517. dma = 1;
  518. q->bounce_pfn = bounce_pfn;
  519. #endif
  520. if (dma) {
  521. init_emergency_isa_pool();
  522. q->bounce_gfp = GFP_NOIO | GFP_DMA;
  523. q->bounce_pfn = bounce_pfn;
  524. }
  525. }
  526. EXPORT_SYMBOL(blk_queue_bounce_limit);
  527. /**
  528. * blk_queue_max_sectors - set max sectors for a request for this queue
  529. * @q: the request queue for the device
  530. * @max_sectors: max sectors in the usual 512b unit
  531. *
  532. * Description:
  533. * Enables a low level driver to set an upper limit on the size of
  534. * received requests.
  535. **/
  536. void blk_queue_max_sectors(request_queue_t *q, unsigned int max_sectors)
  537. {
  538. if ((max_sectors << 9) < PAGE_CACHE_SIZE) {
  539. max_sectors = 1 << (PAGE_CACHE_SHIFT - 9);
  540. printk("%s: set to minimum %d\n", __FUNCTION__, max_sectors);
  541. }
  542. if (BLK_DEF_MAX_SECTORS > max_sectors)
  543. q->max_hw_sectors = q->max_sectors = max_sectors;
  544. else {
  545. q->max_sectors = BLK_DEF_MAX_SECTORS;
  546. q->max_hw_sectors = max_sectors;
  547. }
  548. }
  549. EXPORT_SYMBOL(blk_queue_max_sectors);
  550. /**
  551. * blk_queue_max_phys_segments - set max phys segments for a request for this queue
  552. * @q: the request queue for the device
  553. * @max_segments: max number of segments
  554. *
  555. * Description:
  556. * Enables a low level driver to set an upper limit on the number of
  557. * physical data segments in a request. This would be the largest sized
  558. * scatter list the driver could handle.
  559. **/
  560. void blk_queue_max_phys_segments(request_queue_t *q, unsigned short max_segments)
  561. {
  562. if (!max_segments) {
  563. max_segments = 1;
  564. printk("%s: set to minimum %d\n", __FUNCTION__, max_segments);
  565. }
  566. q->max_phys_segments = max_segments;
  567. }
  568. EXPORT_SYMBOL(blk_queue_max_phys_segments);
  569. /**
  570. * blk_queue_max_hw_segments - set max hw segments for a request for this queue
  571. * @q: the request queue for the device
  572. * @max_segments: max number of segments
  573. *
  574. * Description:
  575. * Enables a low level driver to set an upper limit on the number of
  576. * hw data segments in a request. This would be the largest number of
  577. * address/length pairs the host adapter can actually give as once
  578. * to the device.
  579. **/
  580. void blk_queue_max_hw_segments(request_queue_t *q, unsigned short max_segments)
  581. {
  582. if (!max_segments) {
  583. max_segments = 1;
  584. printk("%s: set to minimum %d\n", __FUNCTION__, max_segments);
  585. }
  586. q->max_hw_segments = max_segments;
  587. }
  588. EXPORT_SYMBOL(blk_queue_max_hw_segments);
  589. /**
  590. * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
  591. * @q: the request queue for the device
  592. * @max_size: max size of segment in bytes
  593. *
  594. * Description:
  595. * Enables a low level driver to set an upper limit on the size of a
  596. * coalesced segment
  597. **/
  598. void blk_queue_max_segment_size(request_queue_t *q, unsigned int max_size)
  599. {
  600. if (max_size < PAGE_CACHE_SIZE) {
  601. max_size = PAGE_CACHE_SIZE;
  602. printk("%s: set to minimum %d\n", __FUNCTION__, max_size);
  603. }
  604. q->max_segment_size = max_size;
  605. }
  606. EXPORT_SYMBOL(blk_queue_max_segment_size);
  607. /**
  608. * blk_queue_hardsect_size - set hardware sector size for the queue
  609. * @q: the request queue for the device
  610. * @size: the hardware sector size, in bytes
  611. *
  612. * Description:
  613. * This should typically be set to the lowest possible sector size
  614. * that the hardware can operate on (possible without reverting to
  615. * even internal read-modify-write operations). Usually the default
  616. * of 512 covers most hardware.
  617. **/
  618. void blk_queue_hardsect_size(request_queue_t *q, unsigned short size)
  619. {
  620. q->hardsect_size = size;
  621. }
  622. EXPORT_SYMBOL(blk_queue_hardsect_size);
  623. /*
  624. * Returns the minimum that is _not_ zero, unless both are zero.
  625. */
  626. #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
  627. /**
  628. * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers
  629. * @t: the stacking driver (top)
  630. * @b: the underlying device (bottom)
  631. **/
  632. void blk_queue_stack_limits(request_queue_t *t, request_queue_t *b)
  633. {
  634. /* zero is "infinity" */
  635. t->max_sectors = min_not_zero(t->max_sectors,b->max_sectors);
  636. t->max_hw_sectors = min_not_zero(t->max_hw_sectors,b->max_hw_sectors);
  637. t->max_phys_segments = min(t->max_phys_segments,b->max_phys_segments);
  638. t->max_hw_segments = min(t->max_hw_segments,b->max_hw_segments);
  639. t->max_segment_size = min(t->max_segment_size,b->max_segment_size);
  640. t->hardsect_size = max(t->hardsect_size,b->hardsect_size);
  641. if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags))
  642. clear_bit(QUEUE_FLAG_CLUSTER, &t->queue_flags);
  643. }
  644. EXPORT_SYMBOL(blk_queue_stack_limits);
  645. /**
  646. * blk_queue_segment_boundary - set boundary rules for segment merging
  647. * @q: the request queue for the device
  648. * @mask: the memory boundary mask
  649. **/
  650. void blk_queue_segment_boundary(request_queue_t *q, unsigned long mask)
  651. {
  652. if (mask < PAGE_CACHE_SIZE - 1) {
  653. mask = PAGE_CACHE_SIZE - 1;
  654. printk("%s: set to minimum %lx\n", __FUNCTION__, mask);
  655. }
  656. q->seg_boundary_mask = mask;
  657. }
  658. EXPORT_SYMBOL(blk_queue_segment_boundary);
  659. /**
  660. * blk_queue_dma_alignment - set dma length and memory alignment
  661. * @q: the request queue for the device
  662. * @mask: alignment mask
  663. *
  664. * description:
  665. * set required memory and length aligment for direct dma transactions.
  666. * this is used when buiding direct io requests for the queue.
  667. *
  668. **/
  669. void blk_queue_dma_alignment(request_queue_t *q, int mask)
  670. {
  671. q->dma_alignment = mask;
  672. }
  673. EXPORT_SYMBOL(blk_queue_dma_alignment);
  674. /**
  675. * blk_queue_find_tag - find a request by its tag and queue
  676. * @q: The request queue for the device
  677. * @tag: The tag of the request
  678. *
  679. * Notes:
  680. * Should be used when a device returns a tag and you want to match
  681. * it with a request.
  682. *
  683. * no locks need be held.
  684. **/
  685. struct request *blk_queue_find_tag(request_queue_t *q, int tag)
  686. {
  687. return blk_map_queue_find_tag(q->queue_tags, tag);
  688. }
  689. EXPORT_SYMBOL(blk_queue_find_tag);
  690. /**
  691. * __blk_free_tags - release a given set of tag maintenance info
  692. * @bqt: the tag map to free
  693. *
  694. * Tries to free the specified @bqt@. Returns true if it was
  695. * actually freed and false if there are still references using it
  696. */
  697. static int __blk_free_tags(struct blk_queue_tag *bqt)
  698. {
  699. int retval;
  700. retval = atomic_dec_and_test(&bqt->refcnt);
  701. if (retval) {
  702. BUG_ON(bqt->busy);
  703. BUG_ON(!list_empty(&bqt->busy_list));
  704. kfree(bqt->tag_index);
  705. bqt->tag_index = NULL;
  706. kfree(bqt->tag_map);
  707. bqt->tag_map = NULL;
  708. kfree(bqt);
  709. }
  710. return retval;
  711. }
  712. /**
  713. * __blk_queue_free_tags - release tag maintenance info
  714. * @q: the request queue for the device
  715. *
  716. * Notes:
  717. * blk_cleanup_queue() will take care of calling this function, if tagging
  718. * has been used. So there's no need to call this directly.
  719. **/
  720. static void __blk_queue_free_tags(request_queue_t *q)
  721. {
  722. struct blk_queue_tag *bqt = q->queue_tags;
  723. if (!bqt)
  724. return;
  725. __blk_free_tags(bqt);
  726. q->queue_tags = NULL;
  727. q->queue_flags &= ~(1 << QUEUE_FLAG_QUEUED);
  728. }
  729. /**
  730. * blk_free_tags - release a given set of tag maintenance info
  731. * @bqt: the tag map to free
  732. *
  733. * For externally managed @bqt@ frees the map. Callers of this
  734. * function must guarantee to have released all the queues that
  735. * might have been using this tag map.
  736. */
  737. void blk_free_tags(struct blk_queue_tag *bqt)
  738. {
  739. if (unlikely(!__blk_free_tags(bqt)))
  740. BUG();
  741. }
  742. EXPORT_SYMBOL(blk_free_tags);
  743. /**
  744. * blk_queue_free_tags - release tag maintenance info
  745. * @q: the request queue for the device
  746. *
  747. * Notes:
  748. * This is used to disabled tagged queuing to a device, yet leave
  749. * queue in function.
  750. **/
  751. void blk_queue_free_tags(request_queue_t *q)
  752. {
  753. clear_bit(QUEUE_FLAG_QUEUED, &q->queue_flags);
  754. }
  755. EXPORT_SYMBOL(blk_queue_free_tags);
  756. static int
  757. init_tag_map(request_queue_t *q, struct blk_queue_tag *tags, int depth)
  758. {
  759. struct request **tag_index;
  760. unsigned long *tag_map;
  761. int nr_ulongs;
  762. if (q && depth > q->nr_requests * 2) {
  763. depth = q->nr_requests * 2;
  764. printk(KERN_ERR "%s: adjusted depth to %d\n",
  765. __FUNCTION__, depth);
  766. }
  767. tag_index = kzalloc(depth * sizeof(struct request *), GFP_ATOMIC);
  768. if (!tag_index)
  769. goto fail;
  770. nr_ulongs = ALIGN(depth, BITS_PER_LONG) / BITS_PER_LONG;
  771. tag_map = kzalloc(nr_ulongs * sizeof(unsigned long), GFP_ATOMIC);
  772. if (!tag_map)
  773. goto fail;
  774. tags->real_max_depth = depth;
  775. tags->max_depth = depth;
  776. tags->tag_index = tag_index;
  777. tags->tag_map = tag_map;
  778. return 0;
  779. fail:
  780. kfree(tag_index);
  781. return -ENOMEM;
  782. }
  783. static struct blk_queue_tag *__blk_queue_init_tags(struct request_queue *q,
  784. int depth)
  785. {
  786. struct blk_queue_tag *tags;
  787. tags = kmalloc(sizeof(struct blk_queue_tag), GFP_ATOMIC);
  788. if (!tags)
  789. goto fail;
  790. if (init_tag_map(q, tags, depth))
  791. goto fail;
  792. INIT_LIST_HEAD(&tags->busy_list);
  793. tags->busy = 0;
  794. atomic_set(&tags->refcnt, 1);
  795. return tags;
  796. fail:
  797. kfree(tags);
  798. return NULL;
  799. }
  800. /**
  801. * blk_init_tags - initialize the tag info for an external tag map
  802. * @depth: the maximum queue depth supported
  803. * @tags: the tag to use
  804. **/
  805. struct blk_queue_tag *blk_init_tags(int depth)
  806. {
  807. return __blk_queue_init_tags(NULL, depth);
  808. }
  809. EXPORT_SYMBOL(blk_init_tags);
  810. /**
  811. * blk_queue_init_tags - initialize the queue tag info
  812. * @q: the request queue for the device
  813. * @depth: the maximum queue depth supported
  814. * @tags: the tag to use
  815. **/
  816. int blk_queue_init_tags(request_queue_t *q, int depth,
  817. struct blk_queue_tag *tags)
  818. {
  819. int rc;
  820. BUG_ON(tags && q->queue_tags && tags != q->queue_tags);
  821. if (!tags && !q->queue_tags) {
  822. tags = __blk_queue_init_tags(q, depth);
  823. if (!tags)
  824. goto fail;
  825. } else if (q->queue_tags) {
  826. if ((rc = blk_queue_resize_tags(q, depth)))
  827. return rc;
  828. set_bit(QUEUE_FLAG_QUEUED, &q->queue_flags);
  829. return 0;
  830. } else
  831. atomic_inc(&tags->refcnt);
  832. /*
  833. * assign it, all done
  834. */
  835. q->queue_tags = tags;
  836. q->queue_flags |= (1 << QUEUE_FLAG_QUEUED);
  837. return 0;
  838. fail:
  839. kfree(tags);
  840. return -ENOMEM;
  841. }
  842. EXPORT_SYMBOL(blk_queue_init_tags);
  843. /**
  844. * blk_queue_resize_tags - change the queueing depth
  845. * @q: the request queue for the device
  846. * @new_depth: the new max command queueing depth
  847. *
  848. * Notes:
  849. * Must be called with the queue lock held.
  850. **/
  851. int blk_queue_resize_tags(request_queue_t *q, int new_depth)
  852. {
  853. struct blk_queue_tag *bqt = q->queue_tags;
  854. struct request **tag_index;
  855. unsigned long *tag_map;
  856. int max_depth, nr_ulongs;
  857. if (!bqt)
  858. return -ENXIO;
  859. /*
  860. * if we already have large enough real_max_depth. just
  861. * adjust max_depth. *NOTE* as requests with tag value
  862. * between new_depth and real_max_depth can be in-flight, tag
  863. * map can not be shrunk blindly here.
  864. */
  865. if (new_depth <= bqt->real_max_depth) {
  866. bqt->max_depth = new_depth;
  867. return 0;
  868. }
  869. /*
  870. * Currently cannot replace a shared tag map with a new
  871. * one, so error out if this is the case
  872. */
  873. if (atomic_read(&bqt->refcnt) != 1)
  874. return -EBUSY;
  875. /*
  876. * save the old state info, so we can copy it back
  877. */
  878. tag_index = bqt->tag_index;
  879. tag_map = bqt->tag_map;
  880. max_depth = bqt->real_max_depth;
  881. if (init_tag_map(q, bqt, new_depth))
  882. return -ENOMEM;
  883. memcpy(bqt->tag_index, tag_index, max_depth * sizeof(struct request *));
  884. nr_ulongs = ALIGN(max_depth, BITS_PER_LONG) / BITS_PER_LONG;
  885. memcpy(bqt->tag_map, tag_map, nr_ulongs * sizeof(unsigned long));
  886. kfree(tag_index);
  887. kfree(tag_map);
  888. return 0;
  889. }
  890. EXPORT_SYMBOL(blk_queue_resize_tags);
  891. /**
  892. * blk_queue_end_tag - end tag operations for a request
  893. * @q: the request queue for the device
  894. * @rq: the request that has completed
  895. *
  896. * Description:
  897. * Typically called when end_that_request_first() returns 0, meaning
  898. * all transfers have been done for a request. It's important to call
  899. * this function before end_that_request_last(), as that will put the
  900. * request back on the free list thus corrupting the internal tag list.
  901. *
  902. * Notes:
  903. * queue lock must be held.
  904. **/
  905. void blk_queue_end_tag(request_queue_t *q, struct request *rq)
  906. {
  907. struct blk_queue_tag *bqt = q->queue_tags;
  908. int tag = rq->tag;
  909. BUG_ON(tag == -1);
  910. if (unlikely(tag >= bqt->real_max_depth))
  911. /*
  912. * This can happen after tag depth has been reduced.
  913. * FIXME: how about a warning or info message here?
  914. */
  915. return;
  916. if (unlikely(!__test_and_clear_bit(tag, bqt->tag_map))) {
  917. printk(KERN_ERR "%s: attempt to clear non-busy tag (%d)\n",
  918. __FUNCTION__, tag);
  919. return;
  920. }
  921. list_del_init(&rq->queuelist);
  922. rq->cmd_flags &= ~REQ_QUEUED;
  923. rq->tag = -1;
  924. if (unlikely(bqt->tag_index[tag] == NULL))
  925. printk(KERN_ERR "%s: tag %d is missing\n",
  926. __FUNCTION__, tag);
  927. bqt->tag_index[tag] = NULL;
  928. bqt->busy--;
  929. }
  930. EXPORT_SYMBOL(blk_queue_end_tag);
  931. /**
  932. * blk_queue_start_tag - find a free tag and assign it
  933. * @q: the request queue for the device
  934. * @rq: the block request that needs tagging
  935. *
  936. * Description:
  937. * This can either be used as a stand-alone helper, or possibly be
  938. * assigned as the queue &prep_rq_fn (in which case &struct request
  939. * automagically gets a tag assigned). Note that this function
  940. * assumes that any type of request can be queued! if this is not
  941. * true for your device, you must check the request type before
  942. * calling this function. The request will also be removed from
  943. * the request queue, so it's the drivers responsibility to readd
  944. * it if it should need to be restarted for some reason.
  945. *
  946. * Notes:
  947. * queue lock must be held.
  948. **/
  949. int blk_queue_start_tag(request_queue_t *q, struct request *rq)
  950. {
  951. struct blk_queue_tag *bqt = q->queue_tags;
  952. int tag;
  953. if (unlikely((rq->cmd_flags & REQ_QUEUED))) {
  954. printk(KERN_ERR
  955. "%s: request %p for device [%s] already tagged %d",
  956. __FUNCTION__, rq,
  957. rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->tag);
  958. BUG();
  959. }
  960. /*
  961. * Protect against shared tag maps, as we may not have exclusive
  962. * access to the tag map.
  963. */
  964. do {
  965. tag = find_first_zero_bit(bqt->tag_map, bqt->max_depth);
  966. if (tag >= bqt->max_depth)
  967. return 1;
  968. } while (test_and_set_bit(tag, bqt->tag_map));
  969. rq->cmd_flags |= REQ_QUEUED;
  970. rq->tag = tag;
  971. bqt->tag_index[tag] = rq;
  972. blkdev_dequeue_request(rq);
  973. list_add(&rq->queuelist, &bqt->busy_list);
  974. bqt->busy++;
  975. return 0;
  976. }
  977. EXPORT_SYMBOL(blk_queue_start_tag);
  978. /**
  979. * blk_queue_invalidate_tags - invalidate all pending tags
  980. * @q: the request queue for the device
  981. *
  982. * Description:
  983. * Hardware conditions may dictate a need to stop all pending requests.
  984. * In this case, we will safely clear the block side of the tag queue and
  985. * readd all requests to the request queue in the right order.
  986. *
  987. * Notes:
  988. * queue lock must be held.
  989. **/
  990. void blk_queue_invalidate_tags(request_queue_t *q)
  991. {
  992. struct blk_queue_tag *bqt = q->queue_tags;
  993. struct list_head *tmp, *n;
  994. struct request *rq;
  995. list_for_each_safe(tmp, n, &bqt->busy_list) {
  996. rq = list_entry_rq(tmp);
  997. if (rq->tag == -1) {
  998. printk(KERN_ERR
  999. "%s: bad tag found on list\n", __FUNCTION__);
  1000. list_del_init(&rq->queuelist);
  1001. rq->cmd_flags &= ~REQ_QUEUED;
  1002. } else
  1003. blk_queue_end_tag(q, rq);
  1004. rq->cmd_flags &= ~REQ_STARTED;
  1005. __elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 0);
  1006. }
  1007. }
  1008. EXPORT_SYMBOL(blk_queue_invalidate_tags);
  1009. void blk_dump_rq_flags(struct request *rq, char *msg)
  1010. {
  1011. int bit;
  1012. printk("%s: dev %s: type=%x, flags=%x\n", msg,
  1013. rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
  1014. rq->cmd_flags);
  1015. printk("\nsector %llu, nr/cnr %lu/%u\n", (unsigned long long)rq->sector,
  1016. rq->nr_sectors,
  1017. rq->current_nr_sectors);
  1018. printk("bio %p, biotail %p, buffer %p, data %p, len %u\n", rq->bio, rq->biotail, rq->buffer, rq->data, rq->data_len);
  1019. if (blk_pc_request(rq)) {
  1020. printk("cdb: ");
  1021. for (bit = 0; bit < sizeof(rq->cmd); bit++)
  1022. printk("%02x ", rq->cmd[bit]);
  1023. printk("\n");
  1024. }
  1025. }
  1026. EXPORT_SYMBOL(blk_dump_rq_flags);
  1027. void blk_recount_segments(request_queue_t *q, struct bio *bio)
  1028. {
  1029. struct bio_vec *bv, *bvprv = NULL;
  1030. int i, nr_phys_segs, nr_hw_segs, seg_size, hw_seg_size, cluster;
  1031. int high, highprv = 1;
  1032. if (unlikely(!bio->bi_io_vec))
  1033. return;
  1034. cluster = q->queue_flags & (1 << QUEUE_FLAG_CLUSTER);
  1035. hw_seg_size = seg_size = nr_phys_segs = nr_hw_segs = 0;
  1036. bio_for_each_segment(bv, bio, i) {
  1037. /*
  1038. * the trick here is making sure that a high page is never
  1039. * considered part of another segment, since that might
  1040. * change with the bounce page.
  1041. */
  1042. high = page_to_pfn(bv->bv_page) > q->bounce_pfn;
  1043. if (high || highprv)
  1044. goto new_hw_segment;
  1045. if (cluster) {
  1046. if (seg_size + bv->bv_len > q->max_segment_size)
  1047. goto new_segment;
  1048. if (!BIOVEC_PHYS_MERGEABLE(bvprv, bv))
  1049. goto new_segment;
  1050. if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bv))
  1051. goto new_segment;
  1052. if (BIOVEC_VIRT_OVERSIZE(hw_seg_size + bv->bv_len))
  1053. goto new_hw_segment;
  1054. seg_size += bv->bv_len;
  1055. hw_seg_size += bv->bv_len;
  1056. bvprv = bv;
  1057. continue;
  1058. }
  1059. new_segment:
  1060. if (BIOVEC_VIRT_MERGEABLE(bvprv, bv) &&
  1061. !BIOVEC_VIRT_OVERSIZE(hw_seg_size + bv->bv_len)) {
  1062. hw_seg_size += bv->bv_len;
  1063. } else {
  1064. new_hw_segment:
  1065. if (hw_seg_size > bio->bi_hw_front_size)
  1066. bio->bi_hw_front_size = hw_seg_size;
  1067. hw_seg_size = BIOVEC_VIRT_START_SIZE(bv) + bv->bv_len;
  1068. nr_hw_segs++;
  1069. }
  1070. nr_phys_segs++;
  1071. bvprv = bv;
  1072. seg_size = bv->bv_len;
  1073. highprv = high;
  1074. }
  1075. if (hw_seg_size > bio->bi_hw_back_size)
  1076. bio->bi_hw_back_size = hw_seg_size;
  1077. if (nr_hw_segs == 1 && hw_seg_size > bio->bi_hw_front_size)
  1078. bio->bi_hw_front_size = hw_seg_size;
  1079. bio->bi_phys_segments = nr_phys_segs;
  1080. bio->bi_hw_segments = nr_hw_segs;
  1081. bio->bi_flags |= (1 << BIO_SEG_VALID);
  1082. }
  1083. EXPORT_SYMBOL(blk_recount_segments);
  1084. static int blk_phys_contig_segment(request_queue_t *q, struct bio *bio,
  1085. struct bio *nxt)
  1086. {
  1087. if (!(q->queue_flags & (1 << QUEUE_FLAG_CLUSTER)))
  1088. return 0;
  1089. if (!BIOVEC_PHYS_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt)))
  1090. return 0;
  1091. if (bio->bi_size + nxt->bi_size > q->max_segment_size)
  1092. return 0;
  1093. /*
  1094. * bio and nxt are contigous in memory, check if the queue allows
  1095. * these two to be merged into one
  1096. */
  1097. if (BIO_SEG_BOUNDARY(q, bio, nxt))
  1098. return 1;
  1099. return 0;
  1100. }
  1101. static int blk_hw_contig_segment(request_queue_t *q, struct bio *bio,
  1102. struct bio *nxt)
  1103. {
  1104. if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
  1105. blk_recount_segments(q, bio);
  1106. if (unlikely(!bio_flagged(nxt, BIO_SEG_VALID)))
  1107. blk_recount_segments(q, nxt);
  1108. if (!BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt)) ||
  1109. BIOVEC_VIRT_OVERSIZE(bio->bi_hw_front_size + bio->bi_hw_back_size))
  1110. return 0;
  1111. if (bio->bi_size + nxt->bi_size > q->max_segment_size)
  1112. return 0;
  1113. return 1;
  1114. }
  1115. /*
  1116. * map a request to scatterlist, return number of sg entries setup. Caller
  1117. * must make sure sg can hold rq->nr_phys_segments entries
  1118. */
  1119. int blk_rq_map_sg(request_queue_t *q, struct request *rq, struct scatterlist *sg)
  1120. {
  1121. struct bio_vec *bvec, *bvprv;
  1122. struct bio *bio;
  1123. int nsegs, i, cluster;
  1124. nsegs = 0;
  1125. cluster = q->queue_flags & (1 << QUEUE_FLAG_CLUSTER);
  1126. /*
  1127. * for each bio in rq
  1128. */
  1129. bvprv = NULL;
  1130. rq_for_each_bio(bio, rq) {
  1131. /*
  1132. * for each segment in bio
  1133. */
  1134. bio_for_each_segment(bvec, bio, i) {
  1135. int nbytes = bvec->bv_len;
  1136. if (bvprv && cluster) {
  1137. if (sg[nsegs - 1].length + nbytes > q->max_segment_size)
  1138. goto new_segment;
  1139. if (!BIOVEC_PHYS_MERGEABLE(bvprv, bvec))
  1140. goto new_segment;
  1141. if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bvec))
  1142. goto new_segment;
  1143. sg[nsegs - 1].length += nbytes;
  1144. } else {
  1145. new_segment:
  1146. memset(&sg[nsegs],0,sizeof(struct scatterlist));
  1147. sg[nsegs].page = bvec->bv_page;
  1148. sg[nsegs].length = nbytes;
  1149. sg[nsegs].offset = bvec->bv_offset;
  1150. nsegs++;
  1151. }
  1152. bvprv = bvec;
  1153. } /* segments in bio */
  1154. } /* bios in rq */
  1155. return nsegs;
  1156. }
  1157. EXPORT_SYMBOL(blk_rq_map_sg);
  1158. /*
  1159. * the standard queue merge functions, can be overridden with device
  1160. * specific ones if so desired
  1161. */
  1162. static inline int ll_new_mergeable(request_queue_t *q,
  1163. struct request *req,
  1164. struct bio *bio)
  1165. {
  1166. int nr_phys_segs = bio_phys_segments(q, bio);
  1167. if (req->nr_phys_segments + nr_phys_segs > q->max_phys_segments) {
  1168. req->cmd_flags |= REQ_NOMERGE;
  1169. if (req == q->last_merge)
  1170. q->last_merge = NULL;
  1171. return 0;
  1172. }
  1173. /*
  1174. * A hw segment is just getting larger, bump just the phys
  1175. * counter.
  1176. */
  1177. req->nr_phys_segments += nr_phys_segs;
  1178. return 1;
  1179. }
  1180. static inline int ll_new_hw_segment(request_queue_t *q,
  1181. struct request *req,
  1182. struct bio *bio)
  1183. {
  1184. int nr_hw_segs = bio_hw_segments(q, bio);
  1185. int nr_phys_segs = bio_phys_segments(q, bio);
  1186. if (req->nr_hw_segments + nr_hw_segs > q->max_hw_segments
  1187. || req->nr_phys_segments + nr_phys_segs > q->max_phys_segments) {
  1188. req->cmd_flags |= REQ_NOMERGE;
  1189. if (req == q->last_merge)
  1190. q->last_merge = NULL;
  1191. return 0;
  1192. }
  1193. /*
  1194. * This will form the start of a new hw segment. Bump both
  1195. * counters.
  1196. */
  1197. req->nr_hw_segments += nr_hw_segs;
  1198. req->nr_phys_segments += nr_phys_segs;
  1199. return 1;
  1200. }
  1201. int ll_back_merge_fn(request_queue_t *q, struct request *req, struct bio *bio)
  1202. {
  1203. unsigned short max_sectors;
  1204. int len;
  1205. if (unlikely(blk_pc_request(req)))
  1206. max_sectors = q->max_hw_sectors;
  1207. else
  1208. max_sectors = q->max_sectors;
  1209. if (req->nr_sectors + bio_sectors(bio) > max_sectors) {
  1210. req->cmd_flags |= REQ_NOMERGE;
  1211. if (req == q->last_merge)
  1212. q->last_merge = NULL;
  1213. return 0;
  1214. }
  1215. if (unlikely(!bio_flagged(req->biotail, BIO_SEG_VALID)))
  1216. blk_recount_segments(q, req->biotail);
  1217. if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
  1218. blk_recount_segments(q, bio);
  1219. len = req->biotail->bi_hw_back_size + bio->bi_hw_front_size;
  1220. if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(req->biotail), __BVEC_START(bio)) &&
  1221. !BIOVEC_VIRT_OVERSIZE(len)) {
  1222. int mergeable = ll_new_mergeable(q, req, bio);
  1223. if (mergeable) {
  1224. if (req->nr_hw_segments == 1)
  1225. req->bio->bi_hw_front_size = len;
  1226. if (bio->bi_hw_segments == 1)
  1227. bio->bi_hw_back_size = len;
  1228. }
  1229. return mergeable;
  1230. }
  1231. return ll_new_hw_segment(q, req, bio);
  1232. }
  1233. EXPORT_SYMBOL(ll_back_merge_fn);
  1234. static int ll_front_merge_fn(request_queue_t *q, struct request *req,
  1235. struct bio *bio)
  1236. {
  1237. unsigned short max_sectors;
  1238. int len;
  1239. if (unlikely(blk_pc_request(req)))
  1240. max_sectors = q->max_hw_sectors;
  1241. else
  1242. max_sectors = q->max_sectors;
  1243. if (req->nr_sectors + bio_sectors(bio) > max_sectors) {
  1244. req->cmd_flags |= REQ_NOMERGE;
  1245. if (req == q->last_merge)
  1246. q->last_merge = NULL;
  1247. return 0;
  1248. }
  1249. len = bio->bi_hw_back_size + req->bio->bi_hw_front_size;
  1250. if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
  1251. blk_recount_segments(q, bio);
  1252. if (unlikely(!bio_flagged(req->bio, BIO_SEG_VALID)))
  1253. blk_recount_segments(q, req->bio);
  1254. if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio), __BVEC_START(req->bio)) &&
  1255. !BIOVEC_VIRT_OVERSIZE(len)) {
  1256. int mergeable = ll_new_mergeable(q, req, bio);
  1257. if (mergeable) {
  1258. if (bio->bi_hw_segments == 1)
  1259. bio->bi_hw_front_size = len;
  1260. if (req->nr_hw_segments == 1)
  1261. req->biotail->bi_hw_back_size = len;
  1262. }
  1263. return mergeable;
  1264. }
  1265. return ll_new_hw_segment(q, req, bio);
  1266. }
  1267. static int ll_merge_requests_fn(request_queue_t *q, struct request *req,
  1268. struct request *next)
  1269. {
  1270. int total_phys_segments;
  1271. int total_hw_segments;
  1272. /*
  1273. * First check if the either of the requests are re-queued
  1274. * requests. Can't merge them if they are.
  1275. */
  1276. if (req->special || next->special)
  1277. return 0;
  1278. /*
  1279. * Will it become too large?
  1280. */
  1281. if ((req->nr_sectors + next->nr_sectors) > q->max_sectors)
  1282. return 0;
  1283. total_phys_segments = req->nr_phys_segments + next->nr_phys_segments;
  1284. if (blk_phys_contig_segment(q, req->biotail, next->bio))
  1285. total_phys_segments--;
  1286. if (total_phys_segments > q->max_phys_segments)
  1287. return 0;
  1288. total_hw_segments = req->nr_hw_segments + next->nr_hw_segments;
  1289. if (blk_hw_contig_segment(q, req->biotail, next->bio)) {
  1290. int len = req->biotail->bi_hw_back_size + next->bio->bi_hw_front_size;
  1291. /*
  1292. * propagate the combined length to the end of the requests
  1293. */
  1294. if (req->nr_hw_segments == 1)
  1295. req->bio->bi_hw_front_size = len;
  1296. if (next->nr_hw_segments == 1)
  1297. next->biotail->bi_hw_back_size = len;
  1298. total_hw_segments--;
  1299. }
  1300. if (total_hw_segments > q->max_hw_segments)
  1301. return 0;
  1302. /* Merge is OK... */
  1303. req->nr_phys_segments = total_phys_segments;
  1304. req->nr_hw_segments = total_hw_segments;
  1305. return 1;
  1306. }
  1307. /*
  1308. * "plug" the device if there are no outstanding requests: this will
  1309. * force the transfer to start only after we have put all the requests
  1310. * on the list.
  1311. *
  1312. * This is called with interrupts off and no requests on the queue and
  1313. * with the queue lock held.
  1314. */
  1315. void blk_plug_device(request_queue_t *q)
  1316. {
  1317. WARN_ON(!irqs_disabled());
  1318. /*
  1319. * don't plug a stopped queue, it must be paired with blk_start_queue()
  1320. * which will restart the queueing
  1321. */
  1322. if (blk_queue_stopped(q))
  1323. return;
  1324. if (!test_and_set_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags)) {
  1325. mod_timer(&q->unplug_timer, jiffies + q->unplug_delay);
  1326. blk_add_trace_generic(q, NULL, 0, BLK_TA_PLUG);
  1327. }
  1328. }
  1329. EXPORT_SYMBOL(blk_plug_device);
  1330. /*
  1331. * remove the queue from the plugged list, if present. called with
  1332. * queue lock held and interrupts disabled.
  1333. */
  1334. int blk_remove_plug(request_queue_t *q)
  1335. {
  1336. WARN_ON(!irqs_disabled());
  1337. if (!test_and_clear_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags))
  1338. return 0;
  1339. del_timer(&q->unplug_timer);
  1340. return 1;
  1341. }
  1342. EXPORT_SYMBOL(blk_remove_plug);
  1343. /*
  1344. * remove the plug and let it rip..
  1345. */
  1346. void __generic_unplug_device(request_queue_t *q)
  1347. {
  1348. if (unlikely(blk_queue_stopped(q)))
  1349. return;
  1350. if (!blk_remove_plug(q))
  1351. return;
  1352. q->request_fn(q);
  1353. }
  1354. EXPORT_SYMBOL(__generic_unplug_device);
  1355. /**
  1356. * generic_unplug_device - fire a request queue
  1357. * @q: The &request_queue_t in question
  1358. *
  1359. * Description:
  1360. * Linux uses plugging to build bigger requests queues before letting
  1361. * the device have at them. If a queue is plugged, the I/O scheduler
  1362. * is still adding and merging requests on the queue. Once the queue
  1363. * gets unplugged, the request_fn defined for the queue is invoked and
  1364. * transfers started.
  1365. **/
  1366. void generic_unplug_device(request_queue_t *q)
  1367. {
  1368. spin_lock_irq(q->queue_lock);
  1369. __generic_unplug_device(q);
  1370. spin_unlock_irq(q->queue_lock);
  1371. }
  1372. EXPORT_SYMBOL(generic_unplug_device);
  1373. static void blk_backing_dev_unplug(struct backing_dev_info *bdi,
  1374. struct page *page)
  1375. {
  1376. request_queue_t *q = bdi->unplug_io_data;
  1377. /*
  1378. * devices don't necessarily have an ->unplug_fn defined
  1379. */
  1380. if (q->unplug_fn) {
  1381. blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL,
  1382. q->rq.count[READ] + q->rq.count[WRITE]);
  1383. q->unplug_fn(q);
  1384. }
  1385. }
  1386. static void blk_unplug_work(struct work_struct *work)
  1387. {
  1388. request_queue_t *q = container_of(work, request_queue_t, unplug_work);
  1389. blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL,
  1390. q->rq.count[READ] + q->rq.count[WRITE]);
  1391. q->unplug_fn(q);
  1392. }
  1393. static void blk_unplug_timeout(unsigned long data)
  1394. {
  1395. request_queue_t *q = (request_queue_t *)data;
  1396. blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_TIMER, NULL,
  1397. q->rq.count[READ] + q->rq.count[WRITE]);
  1398. kblockd_schedule_work(&q->unplug_work);
  1399. }
  1400. /**
  1401. * blk_start_queue - restart a previously stopped queue
  1402. * @q: The &request_queue_t in question
  1403. *
  1404. * Description:
  1405. * blk_start_queue() will clear the stop flag on the queue, and call
  1406. * the request_fn for the queue if it was in a stopped state when
  1407. * entered. Also see blk_stop_queue(). Queue lock must be held.
  1408. **/
  1409. void blk_start_queue(request_queue_t *q)
  1410. {
  1411. WARN_ON(!irqs_disabled());
  1412. clear_bit(QUEUE_FLAG_STOPPED, &q->queue_flags);
  1413. /*
  1414. * one level of recursion is ok and is much faster than kicking
  1415. * the unplug handling
  1416. */
  1417. if (!test_and_set_bit(QUEUE_FLAG_REENTER, &q->queue_flags)) {
  1418. q->request_fn(q);
  1419. clear_bit(QUEUE_FLAG_REENTER, &q->queue_flags);
  1420. } else {
  1421. blk_plug_device(q);
  1422. kblockd_schedule_work(&q->unplug_work);
  1423. }
  1424. }
  1425. EXPORT_SYMBOL(blk_start_queue);
  1426. /**
  1427. * blk_stop_queue - stop a queue
  1428. * @q: The &request_queue_t in question
  1429. *
  1430. * Description:
  1431. * The Linux block layer assumes that a block driver will consume all
  1432. * entries on the request queue when the request_fn strategy is called.
  1433. * Often this will not happen, because of hardware limitations (queue
  1434. * depth settings). If a device driver gets a 'queue full' response,
  1435. * or if it simply chooses not to queue more I/O at one point, it can
  1436. * call this function to prevent the request_fn from being called until
  1437. * the driver has signalled it's ready to go again. This happens by calling
  1438. * blk_start_queue() to restart queue operations. Queue lock must be held.
  1439. **/
  1440. void blk_stop_queue(request_queue_t *q)
  1441. {
  1442. blk_remove_plug(q);
  1443. set_bit(QUEUE_FLAG_STOPPED, &q->queue_flags);
  1444. }
  1445. EXPORT_SYMBOL(blk_stop_queue);
  1446. /**
  1447. * blk_sync_queue - cancel any pending callbacks on a queue
  1448. * @q: the queue
  1449. *
  1450. * Description:
  1451. * The block layer may perform asynchronous callback activity
  1452. * on a queue, such as calling the unplug function after a timeout.
  1453. * A block device may call blk_sync_queue to ensure that any
  1454. * such activity is cancelled, thus allowing it to release resources
  1455. * the the callbacks might use. The caller must already have made sure
  1456. * that its ->make_request_fn will not re-add plugging prior to calling
  1457. * this function.
  1458. *
  1459. */
  1460. void blk_sync_queue(struct request_queue *q)
  1461. {
  1462. del_timer_sync(&q->unplug_timer);
  1463. kblockd_flush();
  1464. }
  1465. EXPORT_SYMBOL(blk_sync_queue);
  1466. /**
  1467. * blk_run_queue - run a single device queue
  1468. * @q: The queue to run
  1469. */
  1470. void blk_run_queue(struct request_queue *q)
  1471. {
  1472. unsigned long flags;
  1473. spin_lock_irqsave(q->queue_lock, flags);
  1474. blk_remove_plug(q);
  1475. /*
  1476. * Only recurse once to avoid overrunning the stack, let the unplug
  1477. * handling reinvoke the handler shortly if we already got there.
  1478. */
  1479. if (!elv_queue_empty(q)) {
  1480. if (!test_and_set_bit(QUEUE_FLAG_REENTER, &q->queue_flags)) {
  1481. q->request_fn(q);
  1482. clear_bit(QUEUE_FLAG_REENTER, &q->queue_flags);
  1483. } else {
  1484. blk_plug_device(q);
  1485. kblockd_schedule_work(&q->unplug_work);
  1486. }
  1487. }
  1488. spin_unlock_irqrestore(q->queue_lock, flags);
  1489. }
  1490. EXPORT_SYMBOL(blk_run_queue);
  1491. /**
  1492. * blk_cleanup_queue: - release a &request_queue_t when it is no longer needed
  1493. * @kobj: the kobj belonging of the request queue to be released
  1494. *
  1495. * Description:
  1496. * blk_cleanup_queue is the pair to blk_init_queue() or
  1497. * blk_queue_make_request(). It should be called when a request queue is
  1498. * being released; typically when a block device is being de-registered.
  1499. * Currently, its primary task it to free all the &struct request
  1500. * structures that were allocated to the queue and the queue itself.
  1501. *
  1502. * Caveat:
  1503. * Hopefully the low level driver will have finished any
  1504. * outstanding requests first...
  1505. **/
  1506. static void blk_release_queue(struct kobject *kobj)
  1507. {
  1508. request_queue_t *q = container_of(kobj, struct request_queue, kobj);
  1509. struct request_list *rl = &q->rq;
  1510. blk_sync_queue(q);
  1511. if (rl->rq_pool)
  1512. mempool_destroy(rl->rq_pool);
  1513. if (q->queue_tags)
  1514. __blk_queue_free_tags(q);
  1515. blk_trace_shutdown(q);
  1516. kmem_cache_free(requestq_cachep, q);
  1517. }
  1518. void blk_put_queue(request_queue_t *q)
  1519. {
  1520. kobject_put(&q->kobj);
  1521. }
  1522. EXPORT_SYMBOL(blk_put_queue);
  1523. void blk_cleanup_queue(request_queue_t * q)
  1524. {
  1525. mutex_lock(&q->sysfs_lock);
  1526. set_bit(QUEUE_FLAG_DEAD, &q->queue_flags);
  1527. mutex_unlock(&q->sysfs_lock);
  1528. if (q->elevator)
  1529. elevator_exit(q->elevator);
  1530. blk_put_queue(q);
  1531. }
  1532. EXPORT_SYMBOL(blk_cleanup_queue);
  1533. static int blk_init_free_list(request_queue_t *q)
  1534. {
  1535. struct request_list *rl = &q->rq;
  1536. rl->count[READ] = rl->count[WRITE] = 0;
  1537. rl->starved[READ] = rl->starved[WRITE] = 0;
  1538. rl->elvpriv = 0;
  1539. init_waitqueue_head(&rl->wait[READ]);
  1540. init_waitqueue_head(&rl->wait[WRITE]);
  1541. rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
  1542. mempool_free_slab, request_cachep, q->node);
  1543. if (!rl->rq_pool)
  1544. return -ENOMEM;
  1545. return 0;
  1546. }
  1547. request_queue_t *blk_alloc_queue(gfp_t gfp_mask)
  1548. {
  1549. return blk_alloc_queue_node(gfp_mask, -1);
  1550. }
  1551. EXPORT_SYMBOL(blk_alloc_queue);
  1552. static struct kobj_type queue_ktype;
  1553. request_queue_t *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
  1554. {
  1555. request_queue_t *q;
  1556. q = kmem_cache_alloc_node(requestq_cachep, gfp_mask, node_id);
  1557. if (!q)
  1558. return NULL;
  1559. memset(q, 0, sizeof(*q));
  1560. init_timer(&q->unplug_timer);
  1561. snprintf(q->kobj.name, KOBJ_NAME_LEN, "%s", "queue");
  1562. q->kobj.ktype = &queue_ktype;
  1563. kobject_init(&q->kobj);
  1564. q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug;
  1565. q->backing_dev_info.unplug_io_data = q;
  1566. mutex_init(&q->sysfs_lock);
  1567. return q;
  1568. }
  1569. EXPORT_SYMBOL(blk_alloc_queue_node);
  1570. /**
  1571. * blk_init_queue - prepare a request queue for use with a block device
  1572. * @rfn: The function to be called to process requests that have been
  1573. * placed on the queue.
  1574. * @lock: Request queue spin lock
  1575. *
  1576. * Description:
  1577. * If a block device wishes to use the standard request handling procedures,
  1578. * which sorts requests and coalesces adjacent requests, then it must
  1579. * call blk_init_queue(). The function @rfn will be called when there
  1580. * are requests on the queue that need to be processed. If the device
  1581. * supports plugging, then @rfn may not be called immediately when requests
  1582. * are available on the queue, but may be called at some time later instead.
  1583. * Plugged queues are generally unplugged when a buffer belonging to one
  1584. * of the requests on the queue is needed, or due to memory pressure.
  1585. *
  1586. * @rfn is not required, or even expected, to remove all requests off the
  1587. * queue, but only as many as it can handle at a time. If it does leave
  1588. * requests on the queue, it is responsible for arranging that the requests
  1589. * get dealt with eventually.
  1590. *
  1591. * The queue spin lock must be held while manipulating the requests on the
  1592. * request queue; this lock will be taken also from interrupt context, so irq
  1593. * disabling is needed for it.
  1594. *
  1595. * Function returns a pointer to the initialized request queue, or NULL if
  1596. * it didn't succeed.
  1597. *
  1598. * Note:
  1599. * blk_init_queue() must be paired with a blk_cleanup_queue() call
  1600. * when the block device is deactivated (such as at module unload).
  1601. **/
  1602. request_queue_t *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
  1603. {
  1604. return blk_init_queue_node(rfn, lock, -1);
  1605. }
  1606. EXPORT_SYMBOL(blk_init_queue);
  1607. request_queue_t *
  1608. blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
  1609. {
  1610. request_queue_t *q = blk_alloc_queue_node(GFP_KERNEL, node_id);
  1611. if (!q)
  1612. return NULL;
  1613. q->node = node_id;
  1614. if (blk_init_free_list(q)) {
  1615. kmem_cache_free(requestq_cachep, q);
  1616. return NULL;
  1617. }
  1618. /*
  1619. * if caller didn't supply a lock, they get per-queue locking with
  1620. * our embedded lock
  1621. */
  1622. if (!lock) {
  1623. spin_lock_init(&q->__queue_lock);
  1624. lock = &q->__queue_lock;
  1625. }
  1626. q->request_fn = rfn;
  1627. q->prep_rq_fn = NULL;
  1628. q->unplug_fn = generic_unplug_device;
  1629. q->queue_flags = (1 << QUEUE_FLAG_CLUSTER);
  1630. q->queue_lock = lock;
  1631. blk_queue_segment_boundary(q, 0xffffffff);
  1632. blk_queue_make_request(q, __make_request);
  1633. blk_queue_max_segment_size(q, MAX_SEGMENT_SIZE);
  1634. blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS);
  1635. blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS);
  1636. /*
  1637. * all done
  1638. */
  1639. if (!elevator_init(q, NULL)) {
  1640. blk_queue_congestion_threshold(q);
  1641. return q;
  1642. }
  1643. blk_put_queue(q);
  1644. return NULL;
  1645. }
  1646. EXPORT_SYMBOL(blk_init_queue_node);
  1647. int blk_get_queue(request_queue_t *q)
  1648. {
  1649. if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) {
  1650. kobject_get(&q->kobj);
  1651. return 0;
  1652. }
  1653. return 1;
  1654. }
  1655. EXPORT_SYMBOL(blk_get_queue);
  1656. static inline void blk_free_request(request_queue_t *q, struct request *rq)
  1657. {
  1658. if (rq->cmd_flags & REQ_ELVPRIV)
  1659. elv_put_request(q, rq);
  1660. mempool_free(rq, q->rq.rq_pool);
  1661. }
  1662. static struct request *
  1663. blk_alloc_request(request_queue_t *q, int rw, int priv, gfp_t gfp_mask)
  1664. {
  1665. struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
  1666. if (!rq)
  1667. return NULL;
  1668. /*
  1669. * first three bits are identical in rq->cmd_flags and bio->bi_rw,
  1670. * see bio.h and blkdev.h
  1671. */
  1672. rq->cmd_flags = rw | REQ_ALLOCED;
  1673. if (priv) {
  1674. if (unlikely(elv_set_request(q, rq, gfp_mask))) {
  1675. mempool_free(rq, q->rq.rq_pool);
  1676. return NULL;
  1677. }
  1678. rq->cmd_flags |= REQ_ELVPRIV;
  1679. }
  1680. return rq;
  1681. }
  1682. /*
  1683. * ioc_batching returns true if the ioc is a valid batching request and
  1684. * should be given priority access to a request.
  1685. */
  1686. static inline int ioc_batching(request_queue_t *q, struct io_context *ioc)
  1687. {
  1688. if (!ioc)
  1689. return 0;
  1690. /*
  1691. * Make sure the process is able to allocate at least 1 request
  1692. * even if the batch times out, otherwise we could theoretically
  1693. * lose wakeups.
  1694. */
  1695. return ioc->nr_batch_requests == q->nr_batching ||
  1696. (ioc->nr_batch_requests > 0
  1697. && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
  1698. }
  1699. /*
  1700. * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
  1701. * will cause the process to be a "batcher" on all queues in the system. This
  1702. * is the behaviour we want though - once it gets a wakeup it should be given
  1703. * a nice run.
  1704. */
  1705. static void ioc_set_batching(request_queue_t *q, struct io_context *ioc)
  1706. {
  1707. if (!ioc || ioc_batching(q, ioc))
  1708. return;
  1709. ioc->nr_batch_requests = q->nr_batching;
  1710. ioc->last_waited = jiffies;
  1711. }
  1712. static void __freed_request(request_queue_t *q, int rw)
  1713. {
  1714. struct request_list *rl = &q->rq;
  1715. if (rl->count[rw] < queue_congestion_off_threshold(q))
  1716. blk_clear_queue_congested(q, rw);
  1717. if (rl->count[rw] + 1 <= q->nr_requests) {
  1718. if (waitqueue_active(&rl->wait[rw]))
  1719. wake_up(&rl->wait[rw]);
  1720. blk_clear_queue_full(q, rw);
  1721. }
  1722. }
  1723. /*
  1724. * A request has just been released. Account for it, update the full and
  1725. * congestion status, wake up any waiters. Called under q->queue_lock.
  1726. */
  1727. static void freed_request(request_queue_t *q, int rw, int priv)
  1728. {
  1729. struct request_list *rl = &q->rq;
  1730. rl->count[rw]--;
  1731. if (priv)
  1732. rl->elvpriv--;
  1733. __freed_request(q, rw);
  1734. if (unlikely(rl->starved[rw ^ 1]))
  1735. __freed_request(q, rw ^ 1);
  1736. }
  1737. #define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist)
  1738. /*
  1739. * Get a free request, queue_lock must be held.
  1740. * Returns NULL on failure, with queue_lock held.
  1741. * Returns !NULL on success, with queue_lock *not held*.
  1742. */
  1743. static struct request *get_request(request_queue_t *q, int rw_flags,
  1744. struct bio *bio, gfp_t gfp_mask)
  1745. {
  1746. struct request *rq = NULL;
  1747. struct request_list *rl = &q->rq;
  1748. struct io_context *ioc = NULL;
  1749. const int rw = rw_flags & 0x01;
  1750. int may_queue, priv;
  1751. may_queue = elv_may_queue(q, rw_flags);
  1752. if (may_queue == ELV_MQUEUE_NO)
  1753. goto rq_starved;
  1754. if (rl->count[rw]+1 >= queue_congestion_on_threshold(q)) {
  1755. if (rl->count[rw]+1 >= q->nr_requests) {
  1756. ioc = current_io_context(GFP_ATOMIC, q->node);
  1757. /*
  1758. * The queue will fill after this allocation, so set
  1759. * it as full, and mark this process as "batching".
  1760. * This process will be allowed to complete a batch of
  1761. * requests, others will be blocked.
  1762. */
  1763. if (!blk_queue_full(q, rw)) {
  1764. ioc_set_batching(q, ioc);
  1765. blk_set_queue_full(q, rw);
  1766. } else {
  1767. if (may_queue != ELV_MQUEUE_MUST
  1768. && !ioc_batching(q, ioc)) {
  1769. /*
  1770. * The queue is full and the allocating
  1771. * process is not a "batcher", and not
  1772. * exempted by the IO scheduler
  1773. */
  1774. goto out;
  1775. }
  1776. }
  1777. }
  1778. blk_set_queue_congested(q, rw);
  1779. }
  1780. /*
  1781. * Only allow batching queuers to allocate up to 50% over the defined
  1782. * limit of requests, otherwise we could have thousands of requests
  1783. * allocated with any setting of ->nr_requests
  1784. */
  1785. if (rl->count[rw] >= (3 * q->nr_requests / 2))
  1786. goto out;
  1787. rl->count[rw]++;
  1788. rl->starved[rw] = 0;
  1789. priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
  1790. if (priv)
  1791. rl->elvpriv++;
  1792. spin_unlock_irq(q->queue_lock);
  1793. rq = blk_alloc_request(q, rw_flags, priv, gfp_mask);
  1794. if (unlikely(!rq)) {
  1795. /*
  1796. * Allocation failed presumably due to memory. Undo anything
  1797. * we might have messed up.
  1798. *
  1799. * Allocating task should really be put onto the front of the
  1800. * wait queue, but this is pretty rare.
  1801. */
  1802. spin_lock_irq(q->queue_lock);
  1803. freed_request(q, rw, priv);
  1804. /*
  1805. * in the very unlikely event that allocation failed and no
  1806. * requests for this direction was pending, mark us starved
  1807. * so that freeing of a request in the other direction will
  1808. * notice us. another possible fix would be to split the
  1809. * rq mempool into READ and WRITE
  1810. */
  1811. rq_starved:
  1812. if (unlikely(rl->count[rw] == 0))
  1813. rl->starved[rw] = 1;
  1814. goto out;
  1815. }
  1816. /*
  1817. * ioc may be NULL here, and ioc_batching will be false. That's
  1818. * OK, if the queue is under the request limit then requests need
  1819. * not count toward the nr_batch_requests limit. There will always
  1820. * be some limit enforced by BLK_BATCH_TIME.
  1821. */
  1822. if (ioc_batching(q, ioc))
  1823. ioc->nr_batch_requests--;
  1824. rq_init(q, rq);
  1825. blk_add_trace_generic(q, bio, rw, BLK_TA_GETRQ);
  1826. out:
  1827. return rq;
  1828. }
  1829. /*
  1830. * No available requests for this queue, unplug the device and wait for some
  1831. * requests to become available.
  1832. *
  1833. * Called with q->queue_lock held, and returns with it unlocked.
  1834. */
  1835. static struct request *get_request_wait(request_queue_t *q, int rw_flags,
  1836. struct bio *bio)
  1837. {
  1838. const int rw = rw_flags & 0x01;
  1839. struct request *rq;
  1840. rq = get_request(q, rw_flags, bio, GFP_NOIO);
  1841. while (!rq) {
  1842. DEFINE_WAIT(wait);
  1843. struct request_list *rl = &q->rq;
  1844. prepare_to_wait_exclusive(&rl->wait[rw], &wait,
  1845. TASK_UNINTERRUPTIBLE);
  1846. rq = get_request(q, rw_flags, bio, GFP_NOIO);
  1847. if (!rq) {
  1848. struct io_context *ioc;
  1849. blk_add_trace_generic(q, bio, rw, BLK_TA_SLEEPRQ);
  1850. __generic_unplug_device(q);
  1851. spin_unlock_irq(q->queue_lock);
  1852. io_schedule();
  1853. /*
  1854. * After sleeping, we become a "batching" process and
  1855. * will be able to allocate at least one request, and
  1856. * up to a big batch of them for a small period time.
  1857. * See ioc_batching, ioc_set_batching
  1858. */
  1859. ioc = current_io_context(GFP_NOIO, q->node);
  1860. ioc_set_batching(q, ioc);
  1861. spin_lock_irq(q->queue_lock);
  1862. }
  1863. finish_wait(&rl->wait[rw], &wait);
  1864. }
  1865. return rq;
  1866. }
  1867. struct request *blk_get_request(request_queue_t *q, int rw, gfp_t gfp_mask)
  1868. {
  1869. struct request *rq;
  1870. BUG_ON(rw != READ && rw != WRITE);
  1871. spin_lock_irq(q->queue_lock);
  1872. if (gfp_mask & __GFP_WAIT) {
  1873. rq = get_request_wait(q, rw, NULL);
  1874. } else {
  1875. rq = get_request(q, rw, NULL, gfp_mask);
  1876. if (!rq)
  1877. spin_unlock_irq(q->queue_lock);
  1878. }
  1879. /* q->queue_lock is unlocked at this point */
  1880. return rq;
  1881. }
  1882. EXPORT_SYMBOL(blk_get_request);
  1883. /**
  1884. * blk_start_queueing - initiate dispatch of requests to device
  1885. * @q: request queue to kick into gear
  1886. *
  1887. * This is basically a helper to remove the need to know whether a queue
  1888. * is plugged or not if someone just wants to initiate dispatch of requests
  1889. * for this queue.
  1890. *
  1891. * The queue lock must be held with interrupts disabled.
  1892. */
  1893. void blk_start_queueing(request_queue_t *q)
  1894. {
  1895. if (!blk_queue_plugged(q))
  1896. q->request_fn(q);
  1897. else
  1898. __generic_unplug_device(q);
  1899. }
  1900. EXPORT_SYMBOL(blk_start_queueing);
  1901. /**
  1902. * blk_requeue_request - put a request back on queue
  1903. * @q: request queue where request should be inserted
  1904. * @rq: request to be inserted
  1905. *
  1906. * Description:
  1907. * Drivers often keep queueing requests until the hardware cannot accept
  1908. * more, when that condition happens we need to put the request back
  1909. * on the queue. Must be called with queue lock held.
  1910. */
  1911. void blk_requeue_request(request_queue_t *q, struct request *rq)
  1912. {
  1913. blk_add_trace_rq(q, rq, BLK_TA_REQUEUE);
  1914. if (blk_rq_tagged(rq))
  1915. blk_queue_end_tag(q, rq);
  1916. elv_requeue_request(q, rq);
  1917. }
  1918. EXPORT_SYMBOL(blk_requeue_request);
  1919. /**
  1920. * blk_insert_request - insert a special request in to a request queue
  1921. * @q: request queue where request should be inserted
  1922. * @rq: request to be inserted
  1923. * @at_head: insert request at head or tail of queue
  1924. * @data: private data
  1925. *
  1926. * Description:
  1927. * Many block devices need to execute commands asynchronously, so they don't
  1928. * block the whole kernel from preemption during request execution. This is
  1929. * accomplished normally by inserting aritficial requests tagged as
  1930. * REQ_SPECIAL in to the corresponding request queue, and letting them be
  1931. * scheduled for actual execution by the request queue.
  1932. *
  1933. * We have the option of inserting the head or the tail of the queue.
  1934. * Typically we use the tail for new ioctls and so forth. We use the head
  1935. * of the queue for things like a QUEUE_FULL message from a device, or a
  1936. * host that is unable to accept a particular command.
  1937. */
  1938. void blk_insert_request(request_queue_t *q, struct request *rq,
  1939. int at_head, void *data)
  1940. {
  1941. int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
  1942. unsigned long flags;
  1943. /*
  1944. * tell I/O scheduler that this isn't a regular read/write (ie it
  1945. * must not attempt merges on this) and that it acts as a soft
  1946. * barrier
  1947. */
  1948. rq->cmd_type = REQ_TYPE_SPECIAL;
  1949. rq->cmd_flags |= REQ_SOFTBARRIER;
  1950. rq->special = data;
  1951. spin_lock_irqsave(q->queue_lock, flags);
  1952. /*
  1953. * If command is tagged, release the tag
  1954. */
  1955. if (blk_rq_tagged(rq))
  1956. blk_queue_end_tag(q, rq);
  1957. drive_stat_acct(rq, rq->nr_sectors, 1);
  1958. __elv_add_request(q, rq, where, 0);
  1959. blk_start_queueing(q);
  1960. spin_unlock_irqrestore(q->queue_lock, flags);
  1961. }
  1962. EXPORT_SYMBOL(blk_insert_request);
  1963. static int __blk_rq_unmap_user(struct bio *bio)
  1964. {
  1965. int ret = 0;
  1966. if (bio) {
  1967. if (bio_flagged(bio, BIO_USER_MAPPED))
  1968. bio_unmap_user(bio);
  1969. else
  1970. ret = bio_uncopy_user(bio);
  1971. }
  1972. return ret;
  1973. }
  1974. static int __blk_rq_map_user(request_queue_t *q, struct request *rq,
  1975. void __user *ubuf, unsigned int len)
  1976. {
  1977. unsigned long uaddr;
  1978. struct bio *bio, *orig_bio;
  1979. int reading, ret;
  1980. reading = rq_data_dir(rq) == READ;
  1981. /*
  1982. * if alignment requirement is satisfied, map in user pages for
  1983. * direct dma. else, set up kernel bounce buffers
  1984. */
  1985. uaddr = (unsigned long) ubuf;
  1986. if (!(uaddr & queue_dma_alignment(q)) && !(len & queue_dma_alignment(q)))
  1987. bio = bio_map_user(q, NULL, uaddr, len, reading);
  1988. else
  1989. bio = bio_copy_user(q, uaddr, len, reading);
  1990. if (IS_ERR(bio))
  1991. return PTR_ERR(bio);
  1992. orig_bio = bio;
  1993. blk_queue_bounce(q, &bio);
  1994. /*
  1995. * We link the bounce buffer in and could have to traverse it
  1996. * later so we have to get a ref to prevent it from being freed
  1997. */
  1998. bio_get(bio);
  1999. if (!rq->bio)
  2000. blk_rq_bio_prep(q, rq, bio);
  2001. else if (!ll_back_merge_fn(q, rq, bio)) {
  2002. ret = -EINVAL;
  2003. goto unmap_bio;
  2004. } else {
  2005. rq->biotail->bi_next = bio;
  2006. rq->biotail = bio;
  2007. rq->data_len += bio->bi_size;
  2008. }
  2009. return bio->bi_size;
  2010. unmap_bio:
  2011. /* if it was boucned we must call the end io function */
  2012. bio_endio(bio, bio->bi_size, 0);
  2013. __blk_rq_unmap_user(orig_bio);
  2014. bio_put(bio);
  2015. return ret;
  2016. }
  2017. /**
  2018. * blk_rq_map_user - map user data to a request, for REQ_BLOCK_PC usage
  2019. * @q: request queue where request should be inserted
  2020. * @rq: request structure to fill
  2021. * @ubuf: the user buffer
  2022. * @len: length of user data
  2023. *
  2024. * Description:
  2025. * Data will be mapped directly for zero copy io, if possible. Otherwise
  2026. * a kernel bounce buffer is used.
  2027. *
  2028. * A matching blk_rq_unmap_user() must be issued at the end of io, while
  2029. * still in process context.
  2030. *
  2031. * Note: The mapped bio may need to be bounced through blk_queue_bounce()
  2032. * before being submitted to the device, as pages mapped may be out of
  2033. * reach. It's the callers responsibility to make sure this happens. The
  2034. * original bio must be passed back in to blk_rq_unmap_user() for proper
  2035. * unmapping.
  2036. */
  2037. int blk_rq_map_user(request_queue_t *q, struct request *rq, void __user *ubuf,
  2038. unsigned long len)
  2039. {
  2040. unsigned long bytes_read = 0;
  2041. struct bio *bio = NULL;
  2042. int ret;
  2043. if (len > (q->max_hw_sectors << 9))
  2044. return -EINVAL;
  2045. if (!len || !ubuf)
  2046. return -EINVAL;
  2047. while (bytes_read != len) {
  2048. unsigned long map_len, end, start;
  2049. map_len = min_t(unsigned long, len - bytes_read, BIO_MAX_SIZE);
  2050. end = ((unsigned long)ubuf + map_len + PAGE_SIZE - 1)
  2051. >> PAGE_SHIFT;
  2052. start = (unsigned long)ubuf >> PAGE_SHIFT;
  2053. /*
  2054. * A bad offset could cause us to require BIO_MAX_PAGES + 1
  2055. * pages. If this happens we just lower the requested
  2056. * mapping len by a page so that we can fit
  2057. */
  2058. if (end - start > BIO_MAX_PAGES)
  2059. map_len -= PAGE_SIZE;
  2060. ret = __blk_rq_map_user(q, rq, ubuf, map_len);
  2061. if (ret < 0)
  2062. goto unmap_rq;
  2063. if (!bio)
  2064. bio = rq->bio;
  2065. bytes_read += ret;
  2066. ubuf += ret;
  2067. }
  2068. rq->buffer = rq->data = NULL;
  2069. return 0;
  2070. unmap_rq:
  2071. blk_rq_unmap_user(bio);
  2072. return ret;
  2073. }
  2074. EXPORT_SYMBOL(blk_rq_map_user);
  2075. /**
  2076. * blk_rq_map_user_iov - map user data to a request, for REQ_BLOCK_PC usage
  2077. * @q: request queue where request should be inserted
  2078. * @rq: request to map data to
  2079. * @iov: pointer to the iovec
  2080. * @iov_count: number of elements in the iovec
  2081. * @len: I/O byte count
  2082. *
  2083. * Description:
  2084. * Data will be mapped directly for zero copy io, if possible. Otherwise
  2085. * a kernel bounce buffer is used.
  2086. *
  2087. * A matching blk_rq_unmap_user() must be issued at the end of io, while
  2088. * still in process context.
  2089. *
  2090. * Note: The mapped bio may need to be bounced through blk_queue_bounce()
  2091. * before being submitted to the device, as pages mapped may be out of
  2092. * reach. It's the callers responsibility to make sure this happens. The
  2093. * original bio must be passed back in to blk_rq_unmap_user() for proper
  2094. * unmapping.
  2095. */
  2096. int blk_rq_map_user_iov(request_queue_t *q, struct request *rq,
  2097. struct sg_iovec *iov, int iov_count, unsigned int len)
  2098. {
  2099. struct bio *bio;
  2100. if (!iov || iov_count <= 0)
  2101. return -EINVAL;
  2102. /* we don't allow misaligned data like bio_map_user() does. If the
  2103. * user is using sg, they're expected to know the alignment constraints
  2104. * and respect them accordingly */
  2105. bio = bio_map_user_iov(q, NULL, iov, iov_count, rq_data_dir(rq)== READ);
  2106. if (IS_ERR(bio))
  2107. return PTR_ERR(bio);
  2108. if (bio->bi_size != len) {
  2109. bio_endio(bio, bio->bi_size, 0);
  2110. bio_unmap_user(bio);
  2111. return -EINVAL;
  2112. }
  2113. bio_get(bio);
  2114. blk_rq_bio_prep(q, rq, bio);
  2115. rq->buffer = rq->data = NULL;
  2116. return 0;
  2117. }
  2118. EXPORT_SYMBOL(blk_rq_map_user_iov);
  2119. /**
  2120. * blk_rq_unmap_user - unmap a request with user data
  2121. * @bio: start of bio list
  2122. *
  2123. * Description:
  2124. * Unmap a rq previously mapped by blk_rq_map_user(). The caller must
  2125. * supply the original rq->bio from the blk_rq_map_user() return, since
  2126. * the io completion may have changed rq->bio.
  2127. */
  2128. int blk_rq_unmap_user(struct bio *bio)
  2129. {
  2130. struct bio *mapped_bio;
  2131. int ret = 0, ret2;
  2132. while (bio) {
  2133. mapped_bio = bio;
  2134. if (unlikely(bio_flagged(bio, BIO_BOUNCED)))
  2135. mapped_bio = bio->bi_private;
  2136. ret2 = __blk_rq_unmap_user(mapped_bio);
  2137. if (ret2 && !ret)
  2138. ret = ret2;
  2139. mapped_bio = bio;
  2140. bio = bio->bi_next;
  2141. bio_put(mapped_bio);
  2142. }
  2143. return ret;
  2144. }
  2145. EXPORT_SYMBOL(blk_rq_unmap_user);
  2146. /**
  2147. * blk_rq_map_kern - map kernel data to a request, for REQ_BLOCK_PC usage
  2148. * @q: request queue where request should be inserted
  2149. * @rq: request to fill
  2150. * @kbuf: the kernel buffer
  2151. * @len: length of user data
  2152. * @gfp_mask: memory allocation flags
  2153. */
  2154. int blk_rq_map_kern(request_queue_t *q, struct request *rq, void *kbuf,
  2155. unsigned int len, gfp_t gfp_mask)
  2156. {
  2157. struct bio *bio;
  2158. if (len > (q->max_hw_sectors << 9))
  2159. return -EINVAL;
  2160. if (!len || !kbuf)
  2161. return -EINVAL;
  2162. bio = bio_map_kern(q, kbuf, len, gfp_mask);
  2163. if (IS_ERR(bio))
  2164. return PTR_ERR(bio);
  2165. if (rq_data_dir(rq) == WRITE)
  2166. bio->bi_rw |= (1 << BIO_RW);
  2167. blk_rq_bio_prep(q, rq, bio);
  2168. rq->buffer = rq->data = NULL;
  2169. return 0;
  2170. }
  2171. EXPORT_SYMBOL(blk_rq_map_kern);
  2172. /**
  2173. * blk_execute_rq_nowait - insert a request into queue for execution
  2174. * @q: queue to insert the request in
  2175. * @bd_disk: matching gendisk
  2176. * @rq: request to insert
  2177. * @at_head: insert request at head or tail of queue
  2178. * @done: I/O completion handler
  2179. *
  2180. * Description:
  2181. * Insert a fully prepared request at the back of the io scheduler queue
  2182. * for execution. Don't wait for completion.
  2183. */
  2184. void blk_execute_rq_nowait(request_queue_t *q, struct gendisk *bd_disk,
  2185. struct request *rq, int at_head,
  2186. rq_end_io_fn *done)
  2187. {
  2188. int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
  2189. rq->rq_disk = bd_disk;
  2190. rq->cmd_flags |= REQ_NOMERGE;
  2191. rq->end_io = done;
  2192. WARN_ON(irqs_disabled());
  2193. spin_lock_irq(q->queue_lock);
  2194. __elv_add_request(q, rq, where, 1);
  2195. __generic_unplug_device(q);
  2196. spin_unlock_irq(q->queue_lock);
  2197. }
  2198. EXPORT_SYMBOL_GPL(blk_execute_rq_nowait);
  2199. /**
  2200. * blk_execute_rq - insert a request into queue for execution
  2201. * @q: queue to insert the request in
  2202. * @bd_disk: matching gendisk
  2203. * @rq: request to insert
  2204. * @at_head: insert request at head or tail of queue
  2205. *
  2206. * Description:
  2207. * Insert a fully prepared request at the back of the io scheduler queue
  2208. * for execution and wait for completion.
  2209. */
  2210. int blk_execute_rq(request_queue_t *q, struct gendisk *bd_disk,
  2211. struct request *rq, int at_head)
  2212. {
  2213. DECLARE_COMPLETION_ONSTACK(wait);
  2214. char sense[SCSI_SENSE_BUFFERSIZE];
  2215. int err = 0;
  2216. /*
  2217. * we need an extra reference to the request, so we can look at
  2218. * it after io completion
  2219. */
  2220. rq->ref_count++;
  2221. if (!rq->sense) {
  2222. memset(sense, 0, sizeof(sense));
  2223. rq->sense = sense;
  2224. rq->sense_len = 0;
  2225. }
  2226. rq->end_io_data = &wait;
  2227. blk_execute_rq_nowait(q, bd_disk, rq, at_head, blk_end_sync_rq);
  2228. wait_for_completion(&wait);
  2229. if (rq->errors)
  2230. err = -EIO;
  2231. return err;
  2232. }
  2233. EXPORT_SYMBOL(blk_execute_rq);
  2234. /**
  2235. * blkdev_issue_flush - queue a flush
  2236. * @bdev: blockdev to issue flush for
  2237. * @error_sector: error sector
  2238. *
  2239. * Description:
  2240. * Issue a flush for the block device in question. Caller can supply
  2241. * room for storing the error offset in case of a flush error, if they
  2242. * wish to. Caller must run wait_for_completion() on its own.
  2243. */
  2244. int blkdev_issue_flush(struct block_device *bdev, sector_t *error_sector)
  2245. {
  2246. request_queue_t *q;
  2247. if (bdev->bd_disk == NULL)
  2248. return -ENXIO;
  2249. q = bdev_get_queue(bdev);
  2250. if (!q)
  2251. return -ENXIO;
  2252. if (!q->issue_flush_fn)
  2253. return -EOPNOTSUPP;
  2254. return q->issue_flush_fn(q, bdev->bd_disk, error_sector);
  2255. }
  2256. EXPORT_SYMBOL(blkdev_issue_flush);
  2257. static void drive_stat_acct(struct request *rq, int nr_sectors, int new_io)
  2258. {
  2259. int rw = rq_data_dir(rq);
  2260. if (!blk_fs_request(rq) || !rq->rq_disk)
  2261. return;
  2262. if (!new_io) {
  2263. __disk_stat_inc(rq->rq_disk, merges[rw]);
  2264. } else {
  2265. disk_round_stats(rq->rq_disk);
  2266. rq->rq_disk->in_flight++;
  2267. }
  2268. }
  2269. /*
  2270. * add-request adds a request to the linked list.
  2271. * queue lock is held and interrupts disabled, as we muck with the
  2272. * request queue list.
  2273. */
  2274. static inline void add_request(request_queue_t * q, struct request * req)
  2275. {
  2276. drive_stat_acct(req, req->nr_sectors, 1);
  2277. /*
  2278. * elevator indicated where it wants this request to be
  2279. * inserted at elevator_merge time
  2280. */
  2281. __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0);
  2282. }
  2283. /*
  2284. * disk_round_stats() - Round off the performance stats on a struct
  2285. * disk_stats.
  2286. *
  2287. * The average IO queue length and utilisation statistics are maintained
  2288. * by observing the current state of the queue length and the amount of
  2289. * time it has been in this state for.
  2290. *
  2291. * Normally, that accounting is done on IO completion, but that can result
  2292. * in more than a second's worth of IO being accounted for within any one
  2293. * second, leading to >100% utilisation. To deal with that, we call this
  2294. * function to do a round-off before returning the results when reading
  2295. * /proc/diskstats. This accounts immediately for all queue usage up to
  2296. * the current jiffies and restarts the counters again.
  2297. */
  2298. void disk_round_stats(struct gendisk *disk)
  2299. {
  2300. unsigned long now = jiffies;
  2301. if (now == disk->stamp)
  2302. return;
  2303. if (disk->in_flight) {
  2304. __disk_stat_add(disk, time_in_queue,
  2305. disk->in_flight * (now - disk->stamp));
  2306. __disk_stat_add(disk, io_ticks, (now - disk->stamp));
  2307. }
  2308. disk->stamp = now;
  2309. }
  2310. EXPORT_SYMBOL_GPL(disk_round_stats);
  2311. /*
  2312. * queue lock must be held
  2313. */
  2314. void __blk_put_request(request_queue_t *q, struct request *req)
  2315. {
  2316. if (unlikely(!q))
  2317. return;
  2318. if (unlikely(--req->ref_count))
  2319. return;
  2320. elv_completed_request(q, req);
  2321. /*
  2322. * Request may not have originated from ll_rw_blk. if not,
  2323. * it didn't come out of our reserved rq pools
  2324. */
  2325. if (req->cmd_flags & REQ_ALLOCED) {
  2326. int rw = rq_data_dir(req);
  2327. int priv = req->cmd_flags & REQ_ELVPRIV;
  2328. BUG_ON(!list_empty(&req->queuelist));
  2329. BUG_ON(!hlist_unhashed(&req->hash));
  2330. blk_free_request(q, req);
  2331. freed_request(q, rw, priv);
  2332. }
  2333. }
  2334. EXPORT_SYMBOL_GPL(__blk_put_request);
  2335. void blk_put_request(struct request *req)
  2336. {
  2337. unsigned long flags;
  2338. request_queue_t *q = req->q;
  2339. /*
  2340. * Gee, IDE calls in w/ NULL q. Fix IDE and remove the
  2341. * following if (q) test.
  2342. */
  2343. if (q) {
  2344. spin_lock_irqsave(q->queue_lock, flags);
  2345. __blk_put_request(q, req);
  2346. spin_unlock_irqrestore(q->queue_lock, flags);
  2347. }
  2348. }
  2349. EXPORT_SYMBOL(blk_put_request);
  2350. /**
  2351. * blk_end_sync_rq - executes a completion event on a request
  2352. * @rq: request to complete
  2353. * @error: end io status of the request
  2354. */
  2355. void blk_end_sync_rq(struct request *rq, int error)
  2356. {
  2357. struct completion *waiting = rq->end_io_data;
  2358. rq->end_io_data = NULL;
  2359. __blk_put_request(rq->q, rq);
  2360. /*
  2361. * complete last, if this is a stack request the process (and thus
  2362. * the rq pointer) could be invalid right after this complete()
  2363. */
  2364. complete(waiting);
  2365. }
  2366. EXPORT_SYMBOL(blk_end_sync_rq);
  2367. /*
  2368. * Has to be called with the request spinlock acquired
  2369. */
  2370. static int attempt_merge(request_queue_t *q, struct request *req,
  2371. struct request *next)
  2372. {
  2373. if (!rq_mergeable(req) || !rq_mergeable(next))
  2374. return 0;
  2375. /*
  2376. * not contiguous
  2377. */
  2378. if (req->sector + req->nr_sectors != next->sector)
  2379. return 0;
  2380. if (rq_data_dir(req) != rq_data_dir(next)
  2381. || req->rq_disk != next->rq_disk
  2382. || next->special)
  2383. return 0;
  2384. /*
  2385. * If we are allowed to merge, then append bio list
  2386. * from next to rq and release next. merge_requests_fn
  2387. * will have updated segment counts, update sector
  2388. * counts here.
  2389. */
  2390. if (!ll_merge_requests_fn(q, req, next))
  2391. return 0;
  2392. /*
  2393. * At this point we have either done a back merge
  2394. * or front merge. We need the smaller start_time of
  2395. * the merged requests to be the current request
  2396. * for accounting purposes.
  2397. */
  2398. if (time_after(req->start_time, next->start_time))
  2399. req->start_time = next->start_time;
  2400. req->biotail->bi_next = next->bio;
  2401. req->biotail = next->biotail;
  2402. req->nr_sectors = req->hard_nr_sectors += next->hard_nr_sectors;
  2403. elv_merge_requests(q, req, next);
  2404. if (req->rq_disk) {
  2405. disk_round_stats(req->rq_disk);
  2406. req->rq_disk->in_flight--;
  2407. }
  2408. req->ioprio = ioprio_best(req->ioprio, next->ioprio);
  2409. __blk_put_request(q, next);
  2410. return 1;
  2411. }
  2412. static inline int attempt_back_merge(request_queue_t *q, struct request *rq)
  2413. {
  2414. struct request *next = elv_latter_request(q, rq);
  2415. if (next)
  2416. return attempt_merge(q, rq, next);
  2417. return 0;
  2418. }
  2419. static inline int attempt_front_merge(request_queue_t *q, struct request *rq)
  2420. {
  2421. struct request *prev = elv_former_request(q, rq);
  2422. if (prev)
  2423. return attempt_merge(q, prev, rq);
  2424. return 0;
  2425. }
  2426. static void init_request_from_bio(struct request *req, struct bio *bio)
  2427. {
  2428. req->cmd_type = REQ_TYPE_FS;
  2429. /*
  2430. * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST)
  2431. */
  2432. if (bio_rw_ahead(bio) || bio_failfast(bio))
  2433. req->cmd_flags |= REQ_FAILFAST;
  2434. /*
  2435. * REQ_BARRIER implies no merging, but lets make it explicit
  2436. */
  2437. if (unlikely(bio_barrier(bio)))
  2438. req->cmd_flags |= (REQ_HARDBARRIER | REQ_NOMERGE);
  2439. if (bio_sync(bio))
  2440. req->cmd_flags |= REQ_RW_SYNC;
  2441. if (bio_rw_meta(bio))
  2442. req->cmd_flags |= REQ_RW_META;
  2443. req->errors = 0;
  2444. req->hard_sector = req->sector = bio->bi_sector;
  2445. req->hard_nr_sectors = req->nr_sectors = bio_sectors(bio);
  2446. req->current_nr_sectors = req->hard_cur_sectors = bio_cur_sectors(bio);
  2447. req->nr_phys_segments = bio_phys_segments(req->q, bio);
  2448. req->nr_hw_segments = bio_hw_segments(req->q, bio);
  2449. req->buffer = bio_data(bio); /* see ->buffer comment above */
  2450. req->bio = req->biotail = bio;
  2451. req->ioprio = bio_prio(bio);
  2452. req->rq_disk = bio->bi_bdev->bd_disk;
  2453. req->start_time = jiffies;
  2454. }
  2455. static int __make_request(request_queue_t *q, struct bio *bio)
  2456. {
  2457. struct request *req;
  2458. int el_ret, nr_sectors, barrier, err;
  2459. const unsigned short prio = bio_prio(bio);
  2460. const int sync = bio_sync(bio);
  2461. int rw_flags;
  2462. nr_sectors = bio_sectors(bio);
  2463. /*
  2464. * low level driver can indicate that it wants pages above a
  2465. * certain limit bounced to low memory (ie for highmem, or even
  2466. * ISA dma in theory)
  2467. */
  2468. blk_queue_bounce(q, &bio);
  2469. barrier = bio_barrier(bio);
  2470. if (unlikely(barrier) && (q->next_ordered == QUEUE_ORDERED_NONE)) {
  2471. err = -EOPNOTSUPP;
  2472. goto end_io;
  2473. }
  2474. spin_lock_irq(q->queue_lock);
  2475. if (unlikely(barrier) || elv_queue_empty(q))
  2476. goto get_rq;
  2477. el_ret = elv_merge(q, &req, bio);
  2478. switch (el_ret) {
  2479. case ELEVATOR_BACK_MERGE:
  2480. BUG_ON(!rq_mergeable(req));
  2481. if (!ll_back_merge_fn(q, req, bio))
  2482. break;
  2483. blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE);
  2484. req->biotail->bi_next = bio;
  2485. req->biotail = bio;
  2486. req->nr_sectors = req->hard_nr_sectors += nr_sectors;
  2487. req->ioprio = ioprio_best(req->ioprio, prio);
  2488. drive_stat_acct(req, nr_sectors, 0);
  2489. if (!attempt_back_merge(q, req))
  2490. elv_merged_request(q, req, el_ret);
  2491. goto out;
  2492. case ELEVATOR_FRONT_MERGE:
  2493. BUG_ON(!rq_mergeable(req));
  2494. if (!ll_front_merge_fn(q, req, bio))
  2495. break;
  2496. blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE);
  2497. bio->bi_next = req->bio;
  2498. req->bio = bio;
  2499. /*
  2500. * may not be valid. if the low level driver said
  2501. * it didn't need a bounce buffer then it better
  2502. * not touch req->buffer either...
  2503. */
  2504. req->buffer = bio_data(bio);
  2505. req->current_nr_sectors = bio_cur_sectors(bio);
  2506. req->hard_cur_sectors = req->current_nr_sectors;
  2507. req->sector = req->hard_sector = bio->bi_sector;
  2508. req->nr_sectors = req->hard_nr_sectors += nr_sectors;
  2509. req->ioprio = ioprio_best(req->ioprio, prio);
  2510. drive_stat_acct(req, nr_sectors, 0);
  2511. if (!attempt_front_merge(q, req))
  2512. elv_merged_request(q, req, el_ret);
  2513. goto out;
  2514. /* ELV_NO_MERGE: elevator says don't/can't merge. */
  2515. default:
  2516. ;
  2517. }
  2518. get_rq:
  2519. /*
  2520. * This sync check and mask will be re-done in init_request_from_bio(),
  2521. * but we need to set it earlier to expose the sync flag to the
  2522. * rq allocator and io schedulers.
  2523. */
  2524. rw_flags = bio_data_dir(bio);
  2525. if (sync)
  2526. rw_flags |= REQ_RW_SYNC;
  2527. /*
  2528. * Grab a free request. This is might sleep but can not fail.
  2529. * Returns with the queue unlocked.
  2530. */
  2531. req = get_request_wait(q, rw_flags, bio);
  2532. /*
  2533. * After dropping the lock and possibly sleeping here, our request
  2534. * may now be mergeable after it had proven unmergeable (above).
  2535. * We don't worry about that case for efficiency. It won't happen
  2536. * often, and the elevators are able to handle it.
  2537. */
  2538. init_request_from_bio(req, bio);
  2539. spin_lock_irq(q->queue_lock);
  2540. if (elv_queue_empty(q))
  2541. blk_plug_device(q);
  2542. add_request(q, req);
  2543. out:
  2544. if (sync)
  2545. __generic_unplug_device(q);
  2546. spin_unlock_irq(q->queue_lock);
  2547. return 0;
  2548. end_io:
  2549. bio_endio(bio, nr_sectors << 9, err);
  2550. return 0;
  2551. }
  2552. /*
  2553. * If bio->bi_dev is a partition, remap the location
  2554. */
  2555. static inline void blk_partition_remap(struct bio *bio)
  2556. {
  2557. struct block_device *bdev = bio->bi_bdev;
  2558. if (bdev != bdev->bd_contains) {
  2559. struct hd_struct *p = bdev->bd_part;
  2560. const int rw = bio_data_dir(bio);
  2561. p->sectors[rw] += bio_sectors(bio);
  2562. p->ios[rw]++;
  2563. bio->bi_sector += p->start_sect;
  2564. bio->bi_bdev = bdev->bd_contains;
  2565. }
  2566. }
  2567. static void handle_bad_sector(struct bio *bio)
  2568. {
  2569. char b[BDEVNAME_SIZE];
  2570. printk(KERN_INFO "attempt to access beyond end of device\n");
  2571. printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
  2572. bdevname(bio->bi_bdev, b),
  2573. bio->bi_rw,
  2574. (unsigned long long)bio->bi_sector + bio_sectors(bio),
  2575. (long long)(bio->bi_bdev->bd_inode->i_size >> 9));
  2576. set_bit(BIO_EOF, &bio->bi_flags);
  2577. }
  2578. #ifdef CONFIG_FAIL_MAKE_REQUEST
  2579. static DECLARE_FAULT_ATTR(fail_make_request);
  2580. static int __init setup_fail_make_request(char *str)
  2581. {
  2582. return setup_fault_attr(&fail_make_request, str);
  2583. }
  2584. __setup("fail_make_request=", setup_fail_make_request);
  2585. static int should_fail_request(struct bio *bio)
  2586. {
  2587. if ((bio->bi_bdev->bd_disk->flags & GENHD_FL_FAIL) ||
  2588. (bio->bi_bdev->bd_part && bio->bi_bdev->bd_part->make_it_fail))
  2589. return should_fail(&fail_make_request, bio->bi_size);
  2590. return 0;
  2591. }
  2592. static int __init fail_make_request_debugfs(void)
  2593. {
  2594. return init_fault_attr_dentries(&fail_make_request,
  2595. "fail_make_request");
  2596. }
  2597. late_initcall(fail_make_request_debugfs);
  2598. #else /* CONFIG_FAIL_MAKE_REQUEST */
  2599. static inline int should_fail_request(struct bio *bio)
  2600. {
  2601. return 0;
  2602. }
  2603. #endif /* CONFIG_FAIL_MAKE_REQUEST */
  2604. /**
  2605. * generic_make_request: hand a buffer to its device driver for I/O
  2606. * @bio: The bio describing the location in memory and on the device.
  2607. *
  2608. * generic_make_request() is used to make I/O requests of block
  2609. * devices. It is passed a &struct bio, which describes the I/O that needs
  2610. * to be done.
  2611. *
  2612. * generic_make_request() does not return any status. The
  2613. * success/failure status of the request, along with notification of
  2614. * completion, is delivered asynchronously through the bio->bi_end_io
  2615. * function described (one day) else where.
  2616. *
  2617. * The caller of generic_make_request must make sure that bi_io_vec
  2618. * are set to describe the memory buffer, and that bi_dev and bi_sector are
  2619. * set to describe the device address, and the
  2620. * bi_end_io and optionally bi_private are set to describe how
  2621. * completion notification should be signaled.
  2622. *
  2623. * generic_make_request and the drivers it calls may use bi_next if this
  2624. * bio happens to be merged with someone else, and may change bi_dev and
  2625. * bi_sector for remaps as it sees fit. So the values of these fields
  2626. * should NOT be depended on after the call to generic_make_request.
  2627. */
  2628. void generic_make_request(struct bio *bio)
  2629. {
  2630. request_queue_t *q;
  2631. sector_t maxsector;
  2632. sector_t old_sector;
  2633. int ret, nr_sectors = bio_sectors(bio);
  2634. dev_t old_dev;
  2635. might_sleep();
  2636. /* Test device or partition size, when known. */
  2637. maxsector = bio->bi_bdev->bd_inode->i_size >> 9;
  2638. if (maxsector) {
  2639. sector_t sector = bio->bi_sector;
  2640. if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
  2641. /*
  2642. * This may well happen - the kernel calls bread()
  2643. * without checking the size of the device, e.g., when
  2644. * mounting a device.
  2645. */
  2646. handle_bad_sector(bio);
  2647. goto end_io;
  2648. }
  2649. }
  2650. /*
  2651. * Resolve the mapping until finished. (drivers are
  2652. * still free to implement/resolve their own stacking
  2653. * by explicitly returning 0)
  2654. *
  2655. * NOTE: we don't repeat the blk_size check for each new device.
  2656. * Stacking drivers are expected to know what they are doing.
  2657. */
  2658. old_sector = -1;
  2659. old_dev = 0;
  2660. do {
  2661. char b[BDEVNAME_SIZE];
  2662. q = bdev_get_queue(bio->bi_bdev);
  2663. if (!q) {
  2664. printk(KERN_ERR
  2665. "generic_make_request: Trying to access "
  2666. "nonexistent block-device %s (%Lu)\n",
  2667. bdevname(bio->bi_bdev, b),
  2668. (long long) bio->bi_sector);
  2669. end_io:
  2670. bio_endio(bio, bio->bi_size, -EIO);
  2671. break;
  2672. }
  2673. if (unlikely(bio_sectors(bio) > q->max_hw_sectors)) {
  2674. printk("bio too big device %s (%u > %u)\n",
  2675. bdevname(bio->bi_bdev, b),
  2676. bio_sectors(bio),
  2677. q->max_hw_sectors);
  2678. goto end_io;
  2679. }
  2680. if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
  2681. goto end_io;
  2682. if (should_fail_request(bio))
  2683. goto end_io;
  2684. /*
  2685. * If this device has partitions, remap block n
  2686. * of partition p to block n+start(p) of the disk.
  2687. */
  2688. blk_partition_remap(bio);
  2689. if (old_sector != -1)
  2690. blk_add_trace_remap(q, bio, old_dev, bio->bi_sector,
  2691. old_sector);
  2692. blk_add_trace_bio(q, bio, BLK_TA_QUEUE);
  2693. old_sector = bio->bi_sector;
  2694. old_dev = bio->bi_bdev->bd_dev;
  2695. maxsector = bio->bi_bdev->bd_inode->i_size >> 9;
  2696. if (maxsector) {
  2697. sector_t sector = bio->bi_sector;
  2698. if (maxsector < nr_sectors ||
  2699. maxsector - nr_sectors < sector) {
  2700. /*
  2701. * This may well happen - partitions are not
  2702. * checked to make sure they are within the size
  2703. * of the whole device.
  2704. */
  2705. handle_bad_sector(bio);
  2706. goto end_io;
  2707. }
  2708. }
  2709. ret = q->make_request_fn(q, bio);
  2710. } while (ret);
  2711. }
  2712. EXPORT_SYMBOL(generic_make_request);
  2713. /**
  2714. * submit_bio: submit a bio to the block device layer for I/O
  2715. * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
  2716. * @bio: The &struct bio which describes the I/O
  2717. *
  2718. * submit_bio() is very similar in purpose to generic_make_request(), and
  2719. * uses that function to do most of the work. Both are fairly rough
  2720. * interfaces, @bio must be presetup and ready for I/O.
  2721. *
  2722. */
  2723. void submit_bio(int rw, struct bio *bio)
  2724. {
  2725. int count = bio_sectors(bio);
  2726. BIO_BUG_ON(!bio->bi_size);
  2727. BIO_BUG_ON(!bio->bi_io_vec);
  2728. bio->bi_rw |= rw;
  2729. if (rw & WRITE) {
  2730. count_vm_events(PGPGOUT, count);
  2731. } else {
  2732. task_io_account_read(bio->bi_size);
  2733. count_vm_events(PGPGIN, count);
  2734. }
  2735. if (unlikely(block_dump)) {
  2736. char b[BDEVNAME_SIZE];
  2737. printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n",
  2738. current->comm, current->pid,
  2739. (rw & WRITE) ? "WRITE" : "READ",
  2740. (unsigned long long)bio->bi_sector,
  2741. bdevname(bio->bi_bdev,b));
  2742. }
  2743. generic_make_request(bio);
  2744. }
  2745. EXPORT_SYMBOL(submit_bio);
  2746. static void blk_recalc_rq_segments(struct request *rq)
  2747. {
  2748. struct bio *bio, *prevbio = NULL;
  2749. int nr_phys_segs, nr_hw_segs;
  2750. unsigned int phys_size, hw_size;
  2751. request_queue_t *q = rq->q;
  2752. if (!rq->bio)
  2753. return;
  2754. phys_size = hw_size = nr_phys_segs = nr_hw_segs = 0;
  2755. rq_for_each_bio(bio, rq) {
  2756. /* Force bio hw/phys segs to be recalculated. */
  2757. bio->bi_flags &= ~(1 << BIO_SEG_VALID);
  2758. nr_phys_segs += bio_phys_segments(q, bio);
  2759. nr_hw_segs += bio_hw_segments(q, bio);
  2760. if (prevbio) {
  2761. int pseg = phys_size + prevbio->bi_size + bio->bi_size;
  2762. int hseg = hw_size + prevbio->bi_size + bio->bi_size;
  2763. if (blk_phys_contig_segment(q, prevbio, bio) &&
  2764. pseg <= q->max_segment_size) {
  2765. nr_phys_segs--;
  2766. phys_size += prevbio->bi_size + bio->bi_size;
  2767. } else
  2768. phys_size = 0;
  2769. if (blk_hw_contig_segment(q, prevbio, bio) &&
  2770. hseg <= q->max_segment_size) {
  2771. nr_hw_segs--;
  2772. hw_size += prevbio->bi_size + bio->bi_size;
  2773. } else
  2774. hw_size = 0;
  2775. }
  2776. prevbio = bio;
  2777. }
  2778. rq->nr_phys_segments = nr_phys_segs;
  2779. rq->nr_hw_segments = nr_hw_segs;
  2780. }
  2781. static void blk_recalc_rq_sectors(struct request *rq, int nsect)
  2782. {
  2783. if (blk_fs_request(rq)) {
  2784. rq->hard_sector += nsect;
  2785. rq->hard_nr_sectors -= nsect;
  2786. /*
  2787. * Move the I/O submission pointers ahead if required.
  2788. */
  2789. if ((rq->nr_sectors >= rq->hard_nr_sectors) &&
  2790. (rq->sector <= rq->hard_sector)) {
  2791. rq->sector = rq->hard_sector;
  2792. rq->nr_sectors = rq->hard_nr_sectors;
  2793. rq->hard_cur_sectors = bio_cur_sectors(rq->bio);
  2794. rq->current_nr_sectors = rq->hard_cur_sectors;
  2795. rq->buffer = bio_data(rq->bio);
  2796. }
  2797. /*
  2798. * if total number of sectors is less than the first segment
  2799. * size, something has gone terribly wrong
  2800. */
  2801. if (rq->nr_sectors < rq->current_nr_sectors) {
  2802. printk("blk: request botched\n");
  2803. rq->nr_sectors = rq->current_nr_sectors;
  2804. }
  2805. }
  2806. }
  2807. static int __end_that_request_first(struct request *req, int uptodate,
  2808. int nr_bytes)
  2809. {
  2810. int total_bytes, bio_nbytes, error, next_idx = 0;
  2811. struct bio *bio;
  2812. blk_add_trace_rq(req->q, req, BLK_TA_COMPLETE);
  2813. /*
  2814. * extend uptodate bool to allow < 0 value to be direct io error
  2815. */
  2816. error = 0;
  2817. if (end_io_error(uptodate))
  2818. error = !uptodate ? -EIO : uptodate;
  2819. /*
  2820. * for a REQ_BLOCK_PC request, we want to carry any eventual
  2821. * sense key with us all the way through
  2822. */
  2823. if (!blk_pc_request(req))
  2824. req->errors = 0;
  2825. if (!uptodate) {
  2826. if (blk_fs_request(req) && !(req->cmd_flags & REQ_QUIET))
  2827. printk("end_request: I/O error, dev %s, sector %llu\n",
  2828. req->rq_disk ? req->rq_disk->disk_name : "?",
  2829. (unsigned long long)req->sector);
  2830. }
  2831. if (blk_fs_request(req) && req->rq_disk) {
  2832. const int rw = rq_data_dir(req);
  2833. disk_stat_add(req->rq_disk, sectors[rw], nr_bytes >> 9);
  2834. }
  2835. total_bytes = bio_nbytes = 0;
  2836. while ((bio = req->bio) != NULL) {
  2837. int nbytes;
  2838. if (nr_bytes >= bio->bi_size) {
  2839. req->bio = bio->bi_next;
  2840. nbytes = bio->bi_size;
  2841. if (!ordered_bio_endio(req, bio, nbytes, error))
  2842. bio_endio(bio, nbytes, error);
  2843. next_idx = 0;
  2844. bio_nbytes = 0;
  2845. } else {
  2846. int idx = bio->bi_idx + next_idx;
  2847. if (unlikely(bio->bi_idx >= bio->bi_vcnt)) {
  2848. blk_dump_rq_flags(req, "__end_that");
  2849. printk("%s: bio idx %d >= vcnt %d\n",
  2850. __FUNCTION__,
  2851. bio->bi_idx, bio->bi_vcnt);
  2852. break;
  2853. }
  2854. nbytes = bio_iovec_idx(bio, idx)->bv_len;
  2855. BIO_BUG_ON(nbytes > bio->bi_size);
  2856. /*
  2857. * not a complete bvec done
  2858. */
  2859. if (unlikely(nbytes > nr_bytes)) {
  2860. bio_nbytes += nr_bytes;
  2861. total_bytes += nr_bytes;
  2862. break;
  2863. }
  2864. /*
  2865. * advance to the next vector
  2866. */
  2867. next_idx++;
  2868. bio_nbytes += nbytes;
  2869. }
  2870. total_bytes += nbytes;
  2871. nr_bytes -= nbytes;
  2872. if ((bio = req->bio)) {
  2873. /*
  2874. * end more in this run, or just return 'not-done'
  2875. */
  2876. if (unlikely(nr_bytes <= 0))
  2877. break;
  2878. }
  2879. }
  2880. /*
  2881. * completely done
  2882. */
  2883. if (!req->bio)
  2884. return 0;
  2885. /*
  2886. * if the request wasn't completed, update state
  2887. */
  2888. if (bio_nbytes) {
  2889. if (!ordered_bio_endio(req, bio, bio_nbytes, error))
  2890. bio_endio(bio, bio_nbytes, error);
  2891. bio->bi_idx += next_idx;
  2892. bio_iovec(bio)->bv_offset += nr_bytes;
  2893. bio_iovec(bio)->bv_len -= nr_bytes;
  2894. }
  2895. blk_recalc_rq_sectors(req, total_bytes >> 9);
  2896. blk_recalc_rq_segments(req);
  2897. return 1;
  2898. }
  2899. /**
  2900. * end_that_request_first - end I/O on a request
  2901. * @req: the request being processed
  2902. * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
  2903. * @nr_sectors: number of sectors to end I/O on
  2904. *
  2905. * Description:
  2906. * Ends I/O on a number of sectors attached to @req, and sets it up
  2907. * for the next range of segments (if any) in the cluster.
  2908. *
  2909. * Return:
  2910. * 0 - we are done with this request, call end_that_request_last()
  2911. * 1 - still buffers pending for this request
  2912. **/
  2913. int end_that_request_first(struct request *req, int uptodate, int nr_sectors)
  2914. {
  2915. return __end_that_request_first(req, uptodate, nr_sectors << 9);
  2916. }
  2917. EXPORT_SYMBOL(end_that_request_first);
  2918. /**
  2919. * end_that_request_chunk - end I/O on a request
  2920. * @req: the request being processed
  2921. * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
  2922. * @nr_bytes: number of bytes to complete
  2923. *
  2924. * Description:
  2925. * Ends I/O on a number of bytes attached to @req, and sets it up
  2926. * for the next range of segments (if any). Like end_that_request_first(),
  2927. * but deals with bytes instead of sectors.
  2928. *
  2929. * Return:
  2930. * 0 - we are done with this request, call end_that_request_last()
  2931. * 1 - still buffers pending for this request
  2932. **/
  2933. int end_that_request_chunk(struct request *req, int uptodate, int nr_bytes)
  2934. {
  2935. return __end_that_request_first(req, uptodate, nr_bytes);
  2936. }
  2937. EXPORT_SYMBOL(end_that_request_chunk);
  2938. /*
  2939. * splice the completion data to a local structure and hand off to
  2940. * process_completion_queue() to complete the requests
  2941. */
  2942. static void blk_done_softirq(struct softirq_action *h)
  2943. {
  2944. struct list_head *cpu_list, local_list;
  2945. local_irq_disable();
  2946. cpu_list = &__get_cpu_var(blk_cpu_done);
  2947. list_replace_init(cpu_list, &local_list);
  2948. local_irq_enable();
  2949. while (!list_empty(&local_list)) {
  2950. struct request *rq = list_entry(local_list.next, struct request, donelist);
  2951. list_del_init(&rq->donelist);
  2952. rq->q->softirq_done_fn(rq);
  2953. }
  2954. }
  2955. static int blk_cpu_notify(struct notifier_block *self, unsigned long action,
  2956. void *hcpu)
  2957. {
  2958. /*
  2959. * If a CPU goes away, splice its entries to the current CPU
  2960. * and trigger a run of the softirq
  2961. */
  2962. if (action == CPU_DEAD) {
  2963. int cpu = (unsigned long) hcpu;
  2964. local_irq_disable();
  2965. list_splice_init(&per_cpu(blk_cpu_done, cpu),
  2966. &__get_cpu_var(blk_cpu_done));
  2967. raise_softirq_irqoff(BLOCK_SOFTIRQ);
  2968. local_irq_enable();
  2969. }
  2970. return NOTIFY_OK;
  2971. }
  2972. static struct notifier_block __devinitdata blk_cpu_notifier = {
  2973. .notifier_call = blk_cpu_notify,
  2974. };
  2975. /**
  2976. * blk_complete_request - end I/O on a request
  2977. * @req: the request being processed
  2978. *
  2979. * Description:
  2980. * Ends all I/O on a request. It does not handle partial completions,
  2981. * unless the driver actually implements this in its completion callback
  2982. * through requeueing. Theh actual completion happens out-of-order,
  2983. * through a softirq handler. The user must have registered a completion
  2984. * callback through blk_queue_softirq_done().
  2985. **/
  2986. void blk_complete_request(struct request *req)
  2987. {
  2988. struct list_head *cpu_list;
  2989. unsigned long flags;
  2990. BUG_ON(!req->q->softirq_done_fn);
  2991. local_irq_save(flags);
  2992. cpu_list = &__get_cpu_var(blk_cpu_done);
  2993. list_add_tail(&req->donelist, cpu_list);
  2994. raise_softirq_irqoff(BLOCK_SOFTIRQ);
  2995. local_irq_restore(flags);
  2996. }
  2997. EXPORT_SYMBOL(blk_complete_request);
  2998. /*
  2999. * queue lock must be held
  3000. */
  3001. void end_that_request_last(struct request *req, int uptodate)
  3002. {
  3003. struct gendisk *disk = req->rq_disk;
  3004. int error;
  3005. /*
  3006. * extend uptodate bool to allow < 0 value to be direct io error
  3007. */
  3008. error = 0;
  3009. if (end_io_error(uptodate))
  3010. error = !uptodate ? -EIO : uptodate;
  3011. if (unlikely(laptop_mode) && blk_fs_request(req))
  3012. laptop_io_completion();
  3013. /*
  3014. * Account IO completion. bar_rq isn't accounted as a normal
  3015. * IO on queueing nor completion. Accounting the containing
  3016. * request is enough.
  3017. */
  3018. if (disk && blk_fs_request(req) && req != &req->q->bar_rq) {
  3019. unsigned long duration = jiffies - req->start_time;
  3020. const int rw = rq_data_dir(req);
  3021. __disk_stat_inc(disk, ios[rw]);
  3022. __disk_stat_add(disk, ticks[rw], duration);
  3023. disk_round_stats(disk);
  3024. disk->in_flight--;
  3025. }
  3026. if (req->end_io)
  3027. req->end_io(req, error);
  3028. else
  3029. __blk_put_request(req->q, req);
  3030. }
  3031. EXPORT_SYMBOL(end_that_request_last);
  3032. void end_request(struct request *req, int uptodate)
  3033. {
  3034. if (!end_that_request_first(req, uptodate, req->hard_cur_sectors)) {
  3035. add_disk_randomness(req->rq_disk);
  3036. blkdev_dequeue_request(req);
  3037. end_that_request_last(req, uptodate);
  3038. }
  3039. }
  3040. EXPORT_SYMBOL(end_request);
  3041. void blk_rq_bio_prep(request_queue_t *q, struct request *rq, struct bio *bio)
  3042. {
  3043. /* first two bits are identical in rq->cmd_flags and bio->bi_rw */
  3044. rq->cmd_flags |= (bio->bi_rw & 3);
  3045. rq->nr_phys_segments = bio_phys_segments(q, bio);
  3046. rq->nr_hw_segments = bio_hw_segments(q, bio);
  3047. rq->current_nr_sectors = bio_cur_sectors(bio);
  3048. rq->hard_cur_sectors = rq->current_nr_sectors;
  3049. rq->hard_nr_sectors = rq->nr_sectors = bio_sectors(bio);
  3050. rq->buffer = bio_data(bio);
  3051. rq->data_len = bio->bi_size;
  3052. rq->bio = rq->biotail = bio;
  3053. }
  3054. EXPORT_SYMBOL(blk_rq_bio_prep);
  3055. int kblockd_schedule_work(struct work_struct *work)
  3056. {
  3057. return queue_work(kblockd_workqueue, work);
  3058. }
  3059. EXPORT_SYMBOL(kblockd_schedule_work);
  3060. void kblockd_flush(void)
  3061. {
  3062. flush_workqueue(kblockd_workqueue);
  3063. }
  3064. EXPORT_SYMBOL(kblockd_flush);
  3065. int __init blk_dev_init(void)
  3066. {
  3067. int i;
  3068. kblockd_workqueue = create_workqueue("kblockd");
  3069. if (!kblockd_workqueue)
  3070. panic("Failed to create kblockd\n");
  3071. request_cachep = kmem_cache_create("blkdev_requests",
  3072. sizeof(struct request), 0, SLAB_PANIC, NULL, NULL);
  3073. requestq_cachep = kmem_cache_create("blkdev_queue",
  3074. sizeof(request_queue_t), 0, SLAB_PANIC, NULL, NULL);
  3075. iocontext_cachep = kmem_cache_create("blkdev_ioc",
  3076. sizeof(struct io_context), 0, SLAB_PANIC, NULL, NULL);
  3077. for_each_possible_cpu(i)
  3078. INIT_LIST_HEAD(&per_cpu(blk_cpu_done, i));
  3079. open_softirq(BLOCK_SOFTIRQ, blk_done_softirq, NULL);
  3080. register_hotcpu_notifier(&blk_cpu_notifier);
  3081. blk_max_low_pfn = max_low_pfn - 1;
  3082. blk_max_pfn = max_pfn - 1;
  3083. return 0;
  3084. }
  3085. /*
  3086. * IO Context helper functions
  3087. */
  3088. void put_io_context(struct io_context *ioc)
  3089. {
  3090. if (ioc == NULL)
  3091. return;
  3092. BUG_ON(atomic_read(&ioc->refcount) == 0);
  3093. if (atomic_dec_and_test(&ioc->refcount)) {
  3094. struct cfq_io_context *cic;
  3095. rcu_read_lock();
  3096. if (ioc->aic && ioc->aic->dtor)
  3097. ioc->aic->dtor(ioc->aic);
  3098. if (ioc->cic_root.rb_node != NULL) {
  3099. struct rb_node *n = rb_first(&ioc->cic_root);
  3100. cic = rb_entry(n, struct cfq_io_context, rb_node);
  3101. cic->dtor(ioc);
  3102. }
  3103. rcu_read_unlock();
  3104. kmem_cache_free(iocontext_cachep, ioc);
  3105. }
  3106. }
  3107. EXPORT_SYMBOL(put_io_context);
  3108. /* Called by the exitting task */
  3109. void exit_io_context(void)
  3110. {
  3111. struct io_context *ioc;
  3112. struct cfq_io_context *cic;
  3113. task_lock(current);
  3114. ioc = current->io_context;
  3115. current->io_context = NULL;
  3116. task_unlock(current);
  3117. ioc->task = NULL;
  3118. if (ioc->aic && ioc->aic->exit)
  3119. ioc->aic->exit(ioc->aic);
  3120. if (ioc->cic_root.rb_node != NULL) {
  3121. cic = rb_entry(rb_first(&ioc->cic_root), struct cfq_io_context, rb_node);
  3122. cic->exit(ioc);
  3123. }
  3124. put_io_context(ioc);
  3125. }
  3126. /*
  3127. * If the current task has no IO context then create one and initialise it.
  3128. * Otherwise, return its existing IO context.
  3129. *
  3130. * This returned IO context doesn't have a specifically elevated refcount,
  3131. * but since the current task itself holds a reference, the context can be
  3132. * used in general code, so long as it stays within `current` context.
  3133. */
  3134. static struct io_context *current_io_context(gfp_t gfp_flags, int node)
  3135. {
  3136. struct task_struct *tsk = current;
  3137. struct io_context *ret;
  3138. ret = tsk->io_context;
  3139. if (likely(ret))
  3140. return ret;
  3141. ret = kmem_cache_alloc_node(iocontext_cachep, gfp_flags, node);
  3142. if (ret) {
  3143. atomic_set(&ret->refcount, 1);
  3144. ret->task = current;
  3145. ret->ioprio_changed = 0;
  3146. ret->last_waited = jiffies; /* doesn't matter... */
  3147. ret->nr_batch_requests = 0; /* because this is 0 */
  3148. ret->aic = NULL;
  3149. ret->cic_root.rb_node = NULL;
  3150. /* make sure set_task_ioprio() sees the settings above */
  3151. smp_wmb();
  3152. tsk->io_context = ret;
  3153. }
  3154. return ret;
  3155. }
  3156. EXPORT_SYMBOL(current_io_context);
  3157. /*
  3158. * If the current task has no IO context then create one and initialise it.
  3159. * If it does have a context, take a ref on it.
  3160. *
  3161. * This is always called in the context of the task which submitted the I/O.
  3162. */
  3163. struct io_context *get_io_context(gfp_t gfp_flags, int node)
  3164. {
  3165. struct io_context *ret;
  3166. ret = current_io_context(gfp_flags, node);
  3167. if (likely(ret))
  3168. atomic_inc(&ret->refcount);
  3169. return ret;
  3170. }
  3171. EXPORT_SYMBOL(get_io_context);
  3172. void copy_io_context(struct io_context **pdst, struct io_context **psrc)
  3173. {
  3174. struct io_context *src = *psrc;
  3175. struct io_context *dst = *pdst;
  3176. if (src) {
  3177. BUG_ON(atomic_read(&src->refcount) == 0);
  3178. atomic_inc(&src->refcount);
  3179. put_io_context(dst);
  3180. *pdst = src;
  3181. }
  3182. }
  3183. EXPORT_SYMBOL(copy_io_context);
  3184. void swap_io_context(struct io_context **ioc1, struct io_context **ioc2)
  3185. {
  3186. struct io_context *temp;
  3187. temp = *ioc1;
  3188. *ioc1 = *ioc2;
  3189. *ioc2 = temp;
  3190. }
  3191. EXPORT_SYMBOL(swap_io_context);
  3192. /*
  3193. * sysfs parts below
  3194. */
  3195. struct queue_sysfs_entry {
  3196. struct attribute attr;
  3197. ssize_t (*show)(struct request_queue *, char *);
  3198. ssize_t (*store)(struct request_queue *, const char *, size_t);
  3199. };
  3200. static ssize_t
  3201. queue_var_show(unsigned int var, char *page)
  3202. {
  3203. return sprintf(page, "%d\n", var);
  3204. }
  3205. static ssize_t
  3206. queue_var_store(unsigned long *var, const char *page, size_t count)
  3207. {
  3208. char *p = (char *) page;
  3209. *var = simple_strtoul(p, &p, 10);
  3210. return count;
  3211. }
  3212. static ssize_t queue_requests_show(struct request_queue *q, char *page)
  3213. {
  3214. return queue_var_show(q->nr_requests, (page));
  3215. }
  3216. static ssize_t
  3217. queue_requests_store(struct request_queue *q, const char *page, size_t count)
  3218. {
  3219. struct request_list *rl = &q->rq;
  3220. unsigned long nr;
  3221. int ret = queue_var_store(&nr, page, count);
  3222. if (nr < BLKDEV_MIN_RQ)
  3223. nr = BLKDEV_MIN_RQ;
  3224. spin_lock_irq(q->queue_lock);
  3225. q->nr_requests = nr;
  3226. blk_queue_congestion_threshold(q);
  3227. if (rl->count[READ] >= queue_congestion_on_threshold(q))
  3228. blk_set_queue_congested(q, READ);
  3229. else if (rl->count[READ] < queue_congestion_off_threshold(q))
  3230. blk_clear_queue_congested(q, READ);
  3231. if (rl->count[WRITE] >= queue_congestion_on_threshold(q))
  3232. blk_set_queue_congested(q, WRITE);
  3233. else if (rl->count[WRITE] < queue_congestion_off_threshold(q))
  3234. blk_clear_queue_congested(q, WRITE);
  3235. if (rl->count[READ] >= q->nr_requests) {
  3236. blk_set_queue_full(q, READ);
  3237. } else if (rl->count[READ]+1 <= q->nr_requests) {
  3238. blk_clear_queue_full(q, READ);
  3239. wake_up(&rl->wait[READ]);
  3240. }
  3241. if (rl->count[WRITE] >= q->nr_requests) {
  3242. blk_set_queue_full(q, WRITE);
  3243. } else if (rl->count[WRITE]+1 <= q->nr_requests) {
  3244. blk_clear_queue_full(q, WRITE);
  3245. wake_up(&rl->wait[WRITE]);
  3246. }
  3247. spin_unlock_irq(q->queue_lock);
  3248. return ret;
  3249. }
  3250. static ssize_t queue_ra_show(struct request_queue *q, char *page)
  3251. {
  3252. int ra_kb = q->backing_dev_info.ra_pages << (PAGE_CACHE_SHIFT - 10);
  3253. return queue_var_show(ra_kb, (page));
  3254. }
  3255. static ssize_t
  3256. queue_ra_store(struct request_queue *q, const char *page, size_t count)
  3257. {
  3258. unsigned long ra_kb;
  3259. ssize_t ret = queue_var_store(&ra_kb, page, count);
  3260. spin_lock_irq(q->queue_lock);
  3261. q->backing_dev_info.ra_pages = ra_kb >> (PAGE_CACHE_SHIFT - 10);
  3262. spin_unlock_irq(q->queue_lock);
  3263. return ret;
  3264. }
  3265. static ssize_t queue_max_sectors_show(struct request_queue *q, char *page)
  3266. {
  3267. int max_sectors_kb = q->max_sectors >> 1;
  3268. return queue_var_show(max_sectors_kb, (page));
  3269. }
  3270. static ssize_t
  3271. queue_max_sectors_store(struct request_queue *q, const char *page, size_t count)
  3272. {
  3273. unsigned long max_sectors_kb,
  3274. max_hw_sectors_kb = q->max_hw_sectors >> 1,
  3275. page_kb = 1 << (PAGE_CACHE_SHIFT - 10);
  3276. ssize_t ret = queue_var_store(&max_sectors_kb, page, count);
  3277. int ra_kb;
  3278. if (max_sectors_kb > max_hw_sectors_kb || max_sectors_kb < page_kb)
  3279. return -EINVAL;
  3280. /*
  3281. * Take the queue lock to update the readahead and max_sectors
  3282. * values synchronously:
  3283. */
  3284. spin_lock_irq(q->queue_lock);
  3285. /*
  3286. * Trim readahead window as well, if necessary:
  3287. */
  3288. ra_kb = q->backing_dev_info.ra_pages << (PAGE_CACHE_SHIFT - 10);
  3289. if (ra_kb > max_sectors_kb)
  3290. q->backing_dev_info.ra_pages =
  3291. max_sectors_kb >> (PAGE_CACHE_SHIFT - 10);
  3292. q->max_sectors = max_sectors_kb << 1;
  3293. spin_unlock_irq(q->queue_lock);
  3294. return ret;
  3295. }
  3296. static ssize_t queue_max_hw_sectors_show(struct request_queue *q, char *page)
  3297. {
  3298. int max_hw_sectors_kb = q->max_hw_sectors >> 1;
  3299. return queue_var_show(max_hw_sectors_kb, (page));
  3300. }
  3301. static struct queue_sysfs_entry queue_requests_entry = {
  3302. .attr = {.name = "nr_requests", .mode = S_IRUGO | S_IWUSR },
  3303. .show = queue_requests_show,
  3304. .store = queue_requests_store,
  3305. };
  3306. static struct queue_sysfs_entry queue_ra_entry = {
  3307. .attr = {.name = "read_ahead_kb", .mode = S_IRUGO | S_IWUSR },
  3308. .show = queue_ra_show,
  3309. .store = queue_ra_store,
  3310. };
  3311. static struct queue_sysfs_entry queue_max_sectors_entry = {
  3312. .attr = {.name = "max_sectors_kb", .mode = S_IRUGO | S_IWUSR },
  3313. .show = queue_max_sectors_show,
  3314. .store = queue_max_sectors_store,
  3315. };
  3316. static struct queue_sysfs_entry queue_max_hw_sectors_entry = {
  3317. .attr = {.name = "max_hw_sectors_kb", .mode = S_IRUGO },
  3318. .show = queue_max_hw_sectors_show,
  3319. };
  3320. static struct queue_sysfs_entry queue_iosched_entry = {
  3321. .attr = {.name = "scheduler", .mode = S_IRUGO | S_IWUSR },
  3322. .show = elv_iosched_show,
  3323. .store = elv_iosched_store,
  3324. };
  3325. static struct attribute *default_attrs[] = {
  3326. &queue_requests_entry.attr,
  3327. &queue_ra_entry.attr,
  3328. &queue_max_hw_sectors_entry.attr,
  3329. &queue_max_sectors_entry.attr,
  3330. &queue_iosched_entry.attr,
  3331. NULL,
  3332. };
  3333. #define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr)
  3334. static ssize_t
  3335. queue_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
  3336. {
  3337. struct queue_sysfs_entry *entry = to_queue(attr);
  3338. request_queue_t *q = container_of(kobj, struct request_queue, kobj);
  3339. ssize_t res;
  3340. if (!entry->show)
  3341. return -EIO;
  3342. mutex_lock(&q->sysfs_lock);
  3343. if (test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)) {
  3344. mutex_unlock(&q->sysfs_lock);
  3345. return -ENOENT;
  3346. }
  3347. res = entry->show(q, page);
  3348. mutex_unlock(&q->sysfs_lock);
  3349. return res;
  3350. }
  3351. static ssize_t
  3352. queue_attr_store(struct kobject *kobj, struct attribute *attr,
  3353. const char *page, size_t length)
  3354. {
  3355. struct queue_sysfs_entry *entry = to_queue(attr);
  3356. request_queue_t *q = container_of(kobj, struct request_queue, kobj);
  3357. ssize_t res;
  3358. if (!entry->store)
  3359. return -EIO;
  3360. mutex_lock(&q->sysfs_lock);
  3361. if (test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)) {
  3362. mutex_unlock(&q->sysfs_lock);
  3363. return -ENOENT;
  3364. }
  3365. res = entry->store(q, page, length);
  3366. mutex_unlock(&q->sysfs_lock);
  3367. return res;
  3368. }
  3369. static struct sysfs_ops queue_sysfs_ops = {
  3370. .show = queue_attr_show,
  3371. .store = queue_attr_store,
  3372. };
  3373. static struct kobj_type queue_ktype = {
  3374. .sysfs_ops = &queue_sysfs_ops,
  3375. .default_attrs = default_attrs,
  3376. .release = blk_release_queue,
  3377. };
  3378. int blk_register_queue(struct gendisk *disk)
  3379. {
  3380. int ret;
  3381. request_queue_t *q = disk->queue;
  3382. if (!q || !q->request_fn)
  3383. return -ENXIO;
  3384. q->kobj.parent = kobject_get(&disk->kobj);
  3385. ret = kobject_add(&q->kobj);
  3386. if (ret < 0)
  3387. return ret;
  3388. kobject_uevent(&q->kobj, KOBJ_ADD);
  3389. ret = elv_register_queue(q);
  3390. if (ret) {
  3391. kobject_uevent(&q->kobj, KOBJ_REMOVE);
  3392. kobject_del(&q->kobj);
  3393. return ret;
  3394. }
  3395. return 0;
  3396. }
  3397. void blk_unregister_queue(struct gendisk *disk)
  3398. {
  3399. request_queue_t *q = disk->queue;
  3400. if (q && q->request_fn) {
  3401. elv_unregister_queue(q);
  3402. kobject_uevent(&q->kobj, KOBJ_REMOVE);
  3403. kobject_del(&q->kobj);
  3404. kobject_put(&disk->kobj);
  3405. }
  3406. }