memory.c 105 KB

1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162636465666768697071727374757677787980818283848586878889909192939495969798991001011021031041051061071081091101111121131141151161171181191201211221231241251261271281291301311321331341351361371381391401411421431441451461471481491501511521531541551561571581591601611621631641651661671681691701711721731741751761771781791801811821831841851861871881891901911921931941951961971981992002012022032042052062072082092102112122132142152162172182192202212222232242252262272282292302312322332342352362372382392402412422432442452462472482492502512522532542552562572582592602612622632642652662672682692702712722732742752762772782792802812822832842852862872882892902912922932942952962972982993003013023033043053063073083093103113123133143153163173183193203213223233243253263273283293303313323333343353363373383393403413423433443453463473483493503513523533543553563573583593603613623633643653663673683693703713723733743753763773783793803813823833843853863873883893903913923933943953963973983994004014024034044054064074084094104114124134144154164174184194204214224234244254264274284294304314324334344354364374384394404414424434444454464474484494504514524534544554564574584594604614624634644654664674684694704714724734744754764774784794804814824834844854864874884894904914924934944954964974984995005015025035045055065075085095105115125135145155165175185195205215225235245255265275285295305315325335345355365375385395405415425435445455465475485495505515525535545555565575585595605615625635645655665675685695705715725735745755765775785795805815825835845855865875885895905915925935945955965975985996006016026036046056066076086096106116126136146156166176186196206216226236246256266276286296306316326336346356366376386396406416426436446456466476486496506516526536546556566576586596606616626636646656666676686696706716726736746756766776786796806816826836846856866876886896906916926936946956966976986997007017027037047057067077087097107117127137147157167177187197207217227237247257267277287297307317327337347357367377387397407417427437447457467477487497507517527537547557567577587597607617627637647657667677687697707717727737747757767777787797807817827837847857867877887897907917927937947957967977987998008018028038048058068078088098108118128138148158168178188198208218228238248258268278288298308318328338348358368378388398408418428438448458468478488498508518528538548558568578588598608618628638648658668678688698708718728738748758768778788798808818828838848858868878888898908918928938948958968978988999009019029039049059069079089099109119129139149159169179189199209219229239249259269279289299309319329339349359369379389399409419429439449459469479489499509519529539549559569579589599609619629639649659669679689699709719729739749759769779789799809819829839849859869879889899909919929939949959969979989991000100110021003100410051006100710081009101010111012101310141015101610171018101910201021102210231024102510261027102810291030103110321033103410351036103710381039104010411042104310441045104610471048104910501051105210531054105510561057105810591060106110621063106410651066106710681069107010711072107310741075107610771078107910801081108210831084108510861087108810891090109110921093109410951096109710981099110011011102110311041105110611071108110911101111111211131114111511161117111811191120112111221123112411251126112711281129113011311132113311341135113611371138113911401141114211431144114511461147114811491150115111521153115411551156115711581159116011611162116311641165116611671168116911701171117211731174117511761177117811791180118111821183118411851186118711881189119011911192119311941195119611971198119912001201120212031204120512061207120812091210121112121213121412151216121712181219122012211222122312241225122612271228122912301231123212331234123512361237123812391240124112421243124412451246124712481249125012511252125312541255125612571258125912601261126212631264126512661267126812691270127112721273127412751276127712781279128012811282128312841285128612871288128912901291129212931294129512961297129812991300130113021303130413051306130713081309131013111312131313141315131613171318131913201321132213231324132513261327132813291330133113321333133413351336133713381339134013411342134313441345134613471348134913501351135213531354135513561357135813591360136113621363136413651366136713681369137013711372137313741375137613771378137913801381138213831384138513861387138813891390139113921393139413951396139713981399140014011402140314041405140614071408140914101411141214131414141514161417141814191420142114221423142414251426142714281429143014311432143314341435143614371438143914401441144214431444144514461447144814491450145114521453145414551456145714581459146014611462146314641465146614671468146914701471147214731474147514761477147814791480148114821483148414851486148714881489149014911492149314941495149614971498149915001501150215031504150515061507150815091510151115121513151415151516151715181519152015211522152315241525152615271528152915301531153215331534153515361537153815391540154115421543154415451546154715481549155015511552155315541555155615571558155915601561156215631564156515661567156815691570157115721573157415751576157715781579158015811582158315841585158615871588158915901591159215931594159515961597159815991600160116021603160416051606160716081609161016111612161316141615161616171618161916201621162216231624162516261627162816291630163116321633163416351636163716381639164016411642164316441645164616471648164916501651165216531654165516561657165816591660166116621663166416651666166716681669167016711672167316741675167616771678167916801681168216831684168516861687168816891690169116921693169416951696169716981699170017011702170317041705170617071708170917101711171217131714171517161717171817191720172117221723172417251726172717281729173017311732173317341735173617371738173917401741174217431744174517461747174817491750175117521753175417551756175717581759176017611762176317641765176617671768176917701771177217731774177517761777177817791780178117821783178417851786178717881789179017911792179317941795179617971798179918001801180218031804180518061807180818091810181118121813181418151816181718181819182018211822182318241825182618271828182918301831183218331834183518361837183818391840184118421843184418451846184718481849185018511852185318541855185618571858185918601861186218631864186518661867186818691870187118721873187418751876187718781879188018811882188318841885188618871888188918901891189218931894189518961897189818991900190119021903190419051906190719081909191019111912191319141915191619171918191919201921192219231924192519261927192819291930193119321933193419351936193719381939194019411942194319441945194619471948194919501951195219531954195519561957195819591960196119621963196419651966196719681969197019711972197319741975197619771978197919801981198219831984198519861987198819891990199119921993199419951996199719981999200020012002200320042005200620072008200920102011201220132014201520162017201820192020202120222023202420252026202720282029203020312032203320342035203620372038203920402041204220432044204520462047204820492050205120522053205420552056205720582059206020612062206320642065206620672068206920702071207220732074207520762077207820792080208120822083208420852086208720882089209020912092209320942095209620972098209921002101210221032104210521062107210821092110211121122113211421152116211721182119212021212122212321242125212621272128212921302131213221332134213521362137213821392140214121422143214421452146214721482149215021512152215321542155215621572158215921602161216221632164216521662167216821692170217121722173217421752176217721782179218021812182218321842185218621872188218921902191219221932194219521962197219821992200220122022203220422052206220722082209221022112212221322142215221622172218221922202221222222232224222522262227222822292230223122322233223422352236223722382239224022412242224322442245224622472248224922502251225222532254225522562257225822592260226122622263226422652266226722682269227022712272227322742275227622772278227922802281228222832284228522862287228822892290229122922293229422952296229722982299230023012302230323042305230623072308230923102311231223132314231523162317231823192320232123222323232423252326232723282329233023312332233323342335233623372338233923402341234223432344234523462347234823492350235123522353235423552356235723582359236023612362236323642365236623672368236923702371237223732374237523762377237823792380238123822383238423852386238723882389239023912392239323942395239623972398239924002401240224032404240524062407240824092410241124122413241424152416241724182419242024212422242324242425242624272428242924302431243224332434243524362437243824392440244124422443244424452446244724482449245024512452245324542455245624572458245924602461246224632464246524662467246824692470247124722473247424752476247724782479248024812482248324842485248624872488248924902491249224932494249524962497249824992500250125022503250425052506250725082509251025112512251325142515251625172518251925202521252225232524252525262527252825292530253125322533253425352536253725382539254025412542254325442545254625472548254925502551255225532554255525562557255825592560256125622563256425652566256725682569257025712572257325742575257625772578257925802581258225832584258525862587258825892590259125922593259425952596259725982599260026012602260326042605260626072608260926102611261226132614261526162617261826192620262126222623262426252626262726282629263026312632263326342635263626372638263926402641264226432644264526462647264826492650265126522653265426552656265726582659266026612662266326642665266626672668266926702671267226732674267526762677267826792680268126822683268426852686268726882689269026912692269326942695269626972698269927002701270227032704270527062707270827092710271127122713271427152716271727182719272027212722272327242725272627272728272927302731273227332734273527362737273827392740274127422743274427452746274727482749275027512752275327542755275627572758275927602761276227632764276527662767276827692770277127722773277427752776277727782779278027812782278327842785278627872788278927902791279227932794279527962797279827992800280128022803280428052806280728082809281028112812281328142815281628172818281928202821282228232824282528262827282828292830283128322833283428352836283728382839284028412842284328442845284628472848284928502851285228532854285528562857285828592860286128622863286428652866286728682869287028712872287328742875287628772878287928802881288228832884288528862887288828892890289128922893289428952896289728982899290029012902290329042905290629072908290929102911291229132914291529162917291829192920292129222923292429252926292729282929293029312932293329342935293629372938293929402941294229432944294529462947294829492950295129522953295429552956295729582959296029612962296329642965296629672968296929702971297229732974297529762977297829792980298129822983298429852986298729882989299029912992299329942995299629972998299930003001300230033004300530063007300830093010301130123013301430153016301730183019302030213022302330243025302630273028302930303031303230333034303530363037303830393040304130423043304430453046304730483049305030513052305330543055305630573058305930603061306230633064306530663067306830693070307130723073307430753076307730783079308030813082308330843085308630873088308930903091309230933094309530963097309830993100310131023103310431053106310731083109311031113112311331143115311631173118311931203121312231233124312531263127312831293130313131323133313431353136313731383139314031413142314331443145314631473148314931503151315231533154315531563157315831593160316131623163316431653166316731683169317031713172317331743175317631773178317931803181318231833184318531863187318831893190319131923193319431953196319731983199320032013202320332043205320632073208320932103211321232133214321532163217321832193220322132223223322432253226322732283229323032313232323332343235323632373238323932403241324232433244324532463247324832493250325132523253325432553256325732583259326032613262326332643265326632673268326932703271327232733274327532763277327832793280328132823283328432853286328732883289329032913292329332943295329632973298329933003301330233033304330533063307330833093310331133123313331433153316331733183319332033213322332333243325332633273328332933303331333233333334333533363337333833393340334133423343334433453346334733483349335033513352335333543355335633573358335933603361336233633364336533663367336833693370337133723373337433753376337733783379338033813382338333843385338633873388338933903391339233933394339533963397339833993400340134023403340434053406340734083409341034113412341334143415341634173418341934203421342234233424342534263427342834293430343134323433343434353436343734383439344034413442344334443445344634473448344934503451345234533454345534563457345834593460346134623463346434653466346734683469347034713472347334743475347634773478347934803481348234833484348534863487348834893490349134923493349434953496349734983499350035013502350335043505350635073508350935103511351235133514351535163517351835193520352135223523352435253526352735283529353035313532353335343535353635373538353935403541354235433544354535463547354835493550355135523553355435553556355735583559356035613562356335643565356635673568356935703571357235733574357535763577357835793580358135823583358435853586358735883589359035913592359335943595359635973598359936003601360236033604360536063607360836093610361136123613361436153616361736183619362036213622362336243625362636273628362936303631363236333634363536363637363836393640364136423643364436453646364736483649365036513652365336543655365636573658365936603661366236633664366536663667366836693670367136723673367436753676367736783679368036813682368336843685368636873688368936903691369236933694369536963697369836993700370137023703370437053706370737083709371037113712371337143715371637173718371937203721372237233724372537263727372837293730373137323733373437353736373737383739374037413742374337443745374637473748374937503751375237533754375537563757375837593760376137623763376437653766376737683769377037713772377337743775377637773778377937803781378237833784378537863787378837893790379137923793379437953796379737983799380038013802380338043805380638073808380938103811381238133814381538163817381838193820382138223823382438253826382738283829383038313832383338343835383638373838383938403841384238433844384538463847384838493850385138523853385438553856385738583859386038613862386338643865386638673868386938703871387238733874387538763877387838793880388138823883388438853886388738883889389038913892389338943895389638973898389939003901390239033904390539063907390839093910391139123913391439153916391739183919392039213922392339243925392639273928
  1. /*
  2. * linux/mm/memory.c
  3. *
  4. * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
  5. */
  6. /*
  7. * demand-loading started 01.12.91 - seems it is high on the list of
  8. * things wanted, and it should be easy to implement. - Linus
  9. */
  10. /*
  11. * Ok, demand-loading was easy, shared pages a little bit tricker. Shared
  12. * pages started 02.12.91, seems to work. - Linus.
  13. *
  14. * Tested sharing by executing about 30 /bin/sh: under the old kernel it
  15. * would have taken more than the 6M I have free, but it worked well as
  16. * far as I could see.
  17. *
  18. * Also corrected some "invalidate()"s - I wasn't doing enough of them.
  19. */
  20. /*
  21. * Real VM (paging to/from disk) started 18.12.91. Much more work and
  22. * thought has to go into this. Oh, well..
  23. * 19.12.91 - works, somewhat. Sometimes I get faults, don't know why.
  24. * Found it. Everything seems to work now.
  25. * 20.12.91 - Ok, making the swap-device changeable like the root.
  26. */
  27. /*
  28. * 05.04.94 - Multi-page memory management added for v1.1.
  29. * Idea by Alex Bligh (alex@cconcepts.co.uk)
  30. *
  31. * 16.07.99 - Support of BIGMEM added by Gerhard Wichert, Siemens AG
  32. * (Gerhard.Wichert@pdb.siemens.de)
  33. *
  34. * Aug/Sep 2004 Changed to four level page tables (Andi Kleen)
  35. */
  36. #include <linux/kernel_stat.h>
  37. #include <linux/mm.h>
  38. #include <linux/hugetlb.h>
  39. #include <linux/mman.h>
  40. #include <linux/swap.h>
  41. #include <linux/highmem.h>
  42. #include <linux/pagemap.h>
  43. #include <linux/ksm.h>
  44. #include <linux/rmap.h>
  45. #include <linux/module.h>
  46. #include <linux/delayacct.h>
  47. #include <linux/init.h>
  48. #include <linux/writeback.h>
  49. #include <linux/memcontrol.h>
  50. #include <linux/mmu_notifier.h>
  51. #include <linux/kallsyms.h>
  52. #include <linux/swapops.h>
  53. #include <linux/elf.h>
  54. #include <linux/gfp.h>
  55. #include <asm/io.h>
  56. #include <asm/pgalloc.h>
  57. #include <asm/uaccess.h>
  58. #include <asm/tlb.h>
  59. #include <asm/tlbflush.h>
  60. #include <asm/pgtable.h>
  61. #include "internal.h"
  62. #ifndef CONFIG_NEED_MULTIPLE_NODES
  63. /* use the per-pgdat data instead for discontigmem - mbligh */
  64. unsigned long max_mapnr;
  65. struct page *mem_map;
  66. EXPORT_SYMBOL(max_mapnr);
  67. EXPORT_SYMBOL(mem_map);
  68. #endif
  69. unsigned long num_physpages;
  70. /*
  71. * A number of key systems in x86 including ioremap() rely on the assumption
  72. * that high_memory defines the upper bound on direct map memory, then end
  73. * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and
  74. * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL
  75. * and ZONE_HIGHMEM.
  76. */
  77. void * high_memory;
  78. EXPORT_SYMBOL(num_physpages);
  79. EXPORT_SYMBOL(high_memory);
  80. /*
  81. * Randomize the address space (stacks, mmaps, brk, etc.).
  82. *
  83. * ( When CONFIG_COMPAT_BRK=y we exclude brk from randomization,
  84. * as ancient (libc5 based) binaries can segfault. )
  85. */
  86. int randomize_va_space __read_mostly =
  87. #ifdef CONFIG_COMPAT_BRK
  88. 1;
  89. #else
  90. 2;
  91. #endif
  92. static int __init disable_randmaps(char *s)
  93. {
  94. randomize_va_space = 0;
  95. return 1;
  96. }
  97. __setup("norandmaps", disable_randmaps);
  98. unsigned long zero_pfn __read_mostly;
  99. unsigned long highest_memmap_pfn __read_mostly;
  100. /*
  101. * CONFIG_MMU architectures set up ZERO_PAGE in their paging_init()
  102. */
  103. static int __init init_zero_pfn(void)
  104. {
  105. zero_pfn = page_to_pfn(ZERO_PAGE(0));
  106. return 0;
  107. }
  108. core_initcall(init_zero_pfn);
  109. #if defined(SPLIT_RSS_COUNTING)
  110. static void __sync_task_rss_stat(struct task_struct *task, struct mm_struct *mm)
  111. {
  112. int i;
  113. for (i = 0; i < NR_MM_COUNTERS; i++) {
  114. if (task->rss_stat.count[i]) {
  115. add_mm_counter(mm, i, task->rss_stat.count[i]);
  116. task->rss_stat.count[i] = 0;
  117. }
  118. }
  119. task->rss_stat.events = 0;
  120. }
  121. static void add_mm_counter_fast(struct mm_struct *mm, int member, int val)
  122. {
  123. struct task_struct *task = current;
  124. if (likely(task->mm == mm))
  125. task->rss_stat.count[member] += val;
  126. else
  127. add_mm_counter(mm, member, val);
  128. }
  129. #define inc_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, 1)
  130. #define dec_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, -1)
  131. /* sync counter once per 64 page faults */
  132. #define TASK_RSS_EVENTS_THRESH (64)
  133. static void check_sync_rss_stat(struct task_struct *task)
  134. {
  135. if (unlikely(task != current))
  136. return;
  137. if (unlikely(task->rss_stat.events++ > TASK_RSS_EVENTS_THRESH))
  138. __sync_task_rss_stat(task, task->mm);
  139. }
  140. unsigned long get_mm_counter(struct mm_struct *mm, int member)
  141. {
  142. long val = 0;
  143. /*
  144. * Don't use task->mm here...for avoiding to use task_get_mm()..
  145. * The caller must guarantee task->mm is not invalid.
  146. */
  147. val = atomic_long_read(&mm->rss_stat.count[member]);
  148. /*
  149. * counter is updated in asynchronous manner and may go to minus.
  150. * But it's never be expected number for users.
  151. */
  152. if (val < 0)
  153. return 0;
  154. return (unsigned long)val;
  155. }
  156. void sync_mm_rss(struct task_struct *task, struct mm_struct *mm)
  157. {
  158. __sync_task_rss_stat(task, mm);
  159. }
  160. #else /* SPLIT_RSS_COUNTING */
  161. #define inc_mm_counter_fast(mm, member) inc_mm_counter(mm, member)
  162. #define dec_mm_counter_fast(mm, member) dec_mm_counter(mm, member)
  163. static void check_sync_rss_stat(struct task_struct *task)
  164. {
  165. }
  166. #endif /* SPLIT_RSS_COUNTING */
  167. #ifdef HAVE_GENERIC_MMU_GATHER
  168. static int tlb_next_batch(struct mmu_gather *tlb)
  169. {
  170. struct mmu_gather_batch *batch;
  171. batch = tlb->active;
  172. if (batch->next) {
  173. tlb->active = batch->next;
  174. return 1;
  175. }
  176. batch = (void *)__get_free_pages(GFP_NOWAIT | __GFP_NOWARN, 0);
  177. if (!batch)
  178. return 0;
  179. batch->next = NULL;
  180. batch->nr = 0;
  181. batch->max = MAX_GATHER_BATCH;
  182. tlb->active->next = batch;
  183. tlb->active = batch;
  184. return 1;
  185. }
  186. /* tlb_gather_mmu
  187. * Called to initialize an (on-stack) mmu_gather structure for page-table
  188. * tear-down from @mm. The @fullmm argument is used when @mm is without
  189. * users and we're going to destroy the full address space (exit/execve).
  190. */
  191. void tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm, bool fullmm)
  192. {
  193. tlb->mm = mm;
  194. tlb->fullmm = fullmm;
  195. tlb->need_flush = 0;
  196. tlb->fast_mode = (num_possible_cpus() == 1);
  197. tlb->local.next = NULL;
  198. tlb->local.nr = 0;
  199. tlb->local.max = ARRAY_SIZE(tlb->__pages);
  200. tlb->active = &tlb->local;
  201. #ifdef CONFIG_HAVE_RCU_TABLE_FREE
  202. tlb->batch = NULL;
  203. #endif
  204. }
  205. void tlb_flush_mmu(struct mmu_gather *tlb)
  206. {
  207. struct mmu_gather_batch *batch;
  208. if (!tlb->need_flush)
  209. return;
  210. tlb->need_flush = 0;
  211. tlb_flush(tlb);
  212. #ifdef CONFIG_HAVE_RCU_TABLE_FREE
  213. tlb_table_flush(tlb);
  214. #endif
  215. if (tlb_fast_mode(tlb))
  216. return;
  217. for (batch = &tlb->local; batch; batch = batch->next) {
  218. free_pages_and_swap_cache(batch->pages, batch->nr);
  219. batch->nr = 0;
  220. }
  221. tlb->active = &tlb->local;
  222. }
  223. /* tlb_finish_mmu
  224. * Called at the end of the shootdown operation to free up any resources
  225. * that were required.
  226. */
  227. void tlb_finish_mmu(struct mmu_gather *tlb, unsigned long start, unsigned long end)
  228. {
  229. struct mmu_gather_batch *batch, *next;
  230. tlb_flush_mmu(tlb);
  231. /* keep the page table cache within bounds */
  232. check_pgt_cache();
  233. for (batch = tlb->local.next; batch; batch = next) {
  234. next = batch->next;
  235. free_pages((unsigned long)batch, 0);
  236. }
  237. tlb->local.next = NULL;
  238. }
  239. /* __tlb_remove_page
  240. * Must perform the equivalent to __free_pte(pte_get_and_clear(ptep)), while
  241. * handling the additional races in SMP caused by other CPUs caching valid
  242. * mappings in their TLBs. Returns the number of free page slots left.
  243. * When out of page slots we must call tlb_flush_mmu().
  244. */
  245. int __tlb_remove_page(struct mmu_gather *tlb, struct page *page)
  246. {
  247. struct mmu_gather_batch *batch;
  248. tlb->need_flush = 1;
  249. if (tlb_fast_mode(tlb)) {
  250. free_page_and_swap_cache(page);
  251. return 1; /* avoid calling tlb_flush_mmu() */
  252. }
  253. batch = tlb->active;
  254. batch->pages[batch->nr++] = page;
  255. if (batch->nr == batch->max) {
  256. if (!tlb_next_batch(tlb))
  257. return 0;
  258. }
  259. VM_BUG_ON(batch->nr > batch->max);
  260. return batch->max - batch->nr;
  261. }
  262. #endif /* HAVE_GENERIC_MMU_GATHER */
  263. #ifdef CONFIG_HAVE_RCU_TABLE_FREE
  264. /*
  265. * See the comment near struct mmu_table_batch.
  266. */
  267. static void tlb_remove_table_smp_sync(void *arg)
  268. {
  269. /* Simply deliver the interrupt */
  270. }
  271. static void tlb_remove_table_one(void *table)
  272. {
  273. /*
  274. * This isn't an RCU grace period and hence the page-tables cannot be
  275. * assumed to be actually RCU-freed.
  276. *
  277. * It is however sufficient for software page-table walkers that rely on
  278. * IRQ disabling. See the comment near struct mmu_table_batch.
  279. */
  280. smp_call_function(tlb_remove_table_smp_sync, NULL, 1);
  281. __tlb_remove_table(table);
  282. }
  283. static void tlb_remove_table_rcu(struct rcu_head *head)
  284. {
  285. struct mmu_table_batch *batch;
  286. int i;
  287. batch = container_of(head, struct mmu_table_batch, rcu);
  288. for (i = 0; i < batch->nr; i++)
  289. __tlb_remove_table(batch->tables[i]);
  290. free_page((unsigned long)batch);
  291. }
  292. void tlb_table_flush(struct mmu_gather *tlb)
  293. {
  294. struct mmu_table_batch **batch = &tlb->batch;
  295. if (*batch) {
  296. call_rcu_sched(&(*batch)->rcu, tlb_remove_table_rcu);
  297. *batch = NULL;
  298. }
  299. }
  300. void tlb_remove_table(struct mmu_gather *tlb, void *table)
  301. {
  302. struct mmu_table_batch **batch = &tlb->batch;
  303. tlb->need_flush = 1;
  304. /*
  305. * When there's less then two users of this mm there cannot be a
  306. * concurrent page-table walk.
  307. */
  308. if (atomic_read(&tlb->mm->mm_users) < 2) {
  309. __tlb_remove_table(table);
  310. return;
  311. }
  312. if (*batch == NULL) {
  313. *batch = (struct mmu_table_batch *)__get_free_page(GFP_NOWAIT | __GFP_NOWARN);
  314. if (*batch == NULL) {
  315. tlb_remove_table_one(table);
  316. return;
  317. }
  318. (*batch)->nr = 0;
  319. }
  320. (*batch)->tables[(*batch)->nr++] = table;
  321. if ((*batch)->nr == MAX_TABLE_BATCH)
  322. tlb_table_flush(tlb);
  323. }
  324. #endif /* CONFIG_HAVE_RCU_TABLE_FREE */
  325. /*
  326. * If a p?d_bad entry is found while walking page tables, report
  327. * the error, before resetting entry to p?d_none. Usually (but
  328. * very seldom) called out from the p?d_none_or_clear_bad macros.
  329. */
  330. void pgd_clear_bad(pgd_t *pgd)
  331. {
  332. pgd_ERROR(*pgd);
  333. pgd_clear(pgd);
  334. }
  335. void pud_clear_bad(pud_t *pud)
  336. {
  337. pud_ERROR(*pud);
  338. pud_clear(pud);
  339. }
  340. void pmd_clear_bad(pmd_t *pmd)
  341. {
  342. pmd_ERROR(*pmd);
  343. pmd_clear(pmd);
  344. }
  345. /*
  346. * Note: this doesn't free the actual pages themselves. That
  347. * has been handled earlier when unmapping all the memory regions.
  348. */
  349. static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd,
  350. unsigned long addr)
  351. {
  352. pgtable_t token = pmd_pgtable(*pmd);
  353. pmd_clear(pmd);
  354. pte_free_tlb(tlb, token, addr);
  355. tlb->mm->nr_ptes--;
  356. }
  357. static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud,
  358. unsigned long addr, unsigned long end,
  359. unsigned long floor, unsigned long ceiling)
  360. {
  361. pmd_t *pmd;
  362. unsigned long next;
  363. unsigned long start;
  364. start = addr;
  365. pmd = pmd_offset(pud, addr);
  366. do {
  367. next = pmd_addr_end(addr, end);
  368. if (pmd_none_or_clear_bad(pmd))
  369. continue;
  370. free_pte_range(tlb, pmd, addr);
  371. } while (pmd++, addr = next, addr != end);
  372. start &= PUD_MASK;
  373. if (start < floor)
  374. return;
  375. if (ceiling) {
  376. ceiling &= PUD_MASK;
  377. if (!ceiling)
  378. return;
  379. }
  380. if (end - 1 > ceiling - 1)
  381. return;
  382. pmd = pmd_offset(pud, start);
  383. pud_clear(pud);
  384. pmd_free_tlb(tlb, pmd, start);
  385. }
  386. static inline void free_pud_range(struct mmu_gather *tlb, pgd_t *pgd,
  387. unsigned long addr, unsigned long end,
  388. unsigned long floor, unsigned long ceiling)
  389. {
  390. pud_t *pud;
  391. unsigned long next;
  392. unsigned long start;
  393. start = addr;
  394. pud = pud_offset(pgd, addr);
  395. do {
  396. next = pud_addr_end(addr, end);
  397. if (pud_none_or_clear_bad(pud))
  398. continue;
  399. free_pmd_range(tlb, pud, addr, next, floor, ceiling);
  400. } while (pud++, addr = next, addr != end);
  401. start &= PGDIR_MASK;
  402. if (start < floor)
  403. return;
  404. if (ceiling) {
  405. ceiling &= PGDIR_MASK;
  406. if (!ceiling)
  407. return;
  408. }
  409. if (end - 1 > ceiling - 1)
  410. return;
  411. pud = pud_offset(pgd, start);
  412. pgd_clear(pgd);
  413. pud_free_tlb(tlb, pud, start);
  414. }
  415. /*
  416. * This function frees user-level page tables of a process.
  417. *
  418. * Must be called with pagetable lock held.
  419. */
  420. void free_pgd_range(struct mmu_gather *tlb,
  421. unsigned long addr, unsigned long end,
  422. unsigned long floor, unsigned long ceiling)
  423. {
  424. pgd_t *pgd;
  425. unsigned long next;
  426. /*
  427. * The next few lines have given us lots of grief...
  428. *
  429. * Why are we testing PMD* at this top level? Because often
  430. * there will be no work to do at all, and we'd prefer not to
  431. * go all the way down to the bottom just to discover that.
  432. *
  433. * Why all these "- 1"s? Because 0 represents both the bottom
  434. * of the address space and the top of it (using -1 for the
  435. * top wouldn't help much: the masks would do the wrong thing).
  436. * The rule is that addr 0 and floor 0 refer to the bottom of
  437. * the address space, but end 0 and ceiling 0 refer to the top
  438. * Comparisons need to use "end - 1" and "ceiling - 1" (though
  439. * that end 0 case should be mythical).
  440. *
  441. * Wherever addr is brought up or ceiling brought down, we must
  442. * be careful to reject "the opposite 0" before it confuses the
  443. * subsequent tests. But what about where end is brought down
  444. * by PMD_SIZE below? no, end can't go down to 0 there.
  445. *
  446. * Whereas we round start (addr) and ceiling down, by different
  447. * masks at different levels, in order to test whether a table
  448. * now has no other vmas using it, so can be freed, we don't
  449. * bother to round floor or end up - the tests don't need that.
  450. */
  451. addr &= PMD_MASK;
  452. if (addr < floor) {
  453. addr += PMD_SIZE;
  454. if (!addr)
  455. return;
  456. }
  457. if (ceiling) {
  458. ceiling &= PMD_MASK;
  459. if (!ceiling)
  460. return;
  461. }
  462. if (end - 1 > ceiling - 1)
  463. end -= PMD_SIZE;
  464. if (addr > end - 1)
  465. return;
  466. pgd = pgd_offset(tlb->mm, addr);
  467. do {
  468. next = pgd_addr_end(addr, end);
  469. if (pgd_none_or_clear_bad(pgd))
  470. continue;
  471. free_pud_range(tlb, pgd, addr, next, floor, ceiling);
  472. } while (pgd++, addr = next, addr != end);
  473. }
  474. void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *vma,
  475. unsigned long floor, unsigned long ceiling)
  476. {
  477. while (vma) {
  478. struct vm_area_struct *next = vma->vm_next;
  479. unsigned long addr = vma->vm_start;
  480. /*
  481. * Hide vma from rmap and truncate_pagecache before freeing
  482. * pgtables
  483. */
  484. unlink_anon_vmas(vma);
  485. unlink_file_vma(vma);
  486. if (is_vm_hugetlb_page(vma)) {
  487. hugetlb_free_pgd_range(tlb, addr, vma->vm_end,
  488. floor, next? next->vm_start: ceiling);
  489. } else {
  490. /*
  491. * Optimization: gather nearby vmas into one call down
  492. */
  493. while (next && next->vm_start <= vma->vm_end + PMD_SIZE
  494. && !is_vm_hugetlb_page(next)) {
  495. vma = next;
  496. next = vma->vm_next;
  497. unlink_anon_vmas(vma);
  498. unlink_file_vma(vma);
  499. }
  500. free_pgd_range(tlb, addr, vma->vm_end,
  501. floor, next? next->vm_start: ceiling);
  502. }
  503. vma = next;
  504. }
  505. }
  506. int __pte_alloc(struct mm_struct *mm, struct vm_area_struct *vma,
  507. pmd_t *pmd, unsigned long address)
  508. {
  509. pgtable_t new = pte_alloc_one(mm, address);
  510. int wait_split_huge_page;
  511. if (!new)
  512. return -ENOMEM;
  513. /*
  514. * Ensure all pte setup (eg. pte page lock and page clearing) are
  515. * visible before the pte is made visible to other CPUs by being
  516. * put into page tables.
  517. *
  518. * The other side of the story is the pointer chasing in the page
  519. * table walking code (when walking the page table without locking;
  520. * ie. most of the time). Fortunately, these data accesses consist
  521. * of a chain of data-dependent loads, meaning most CPUs (alpha
  522. * being the notable exception) will already guarantee loads are
  523. * seen in-order. See the alpha page table accessors for the
  524. * smp_read_barrier_depends() barriers in page table walking code.
  525. */
  526. smp_wmb(); /* Could be smp_wmb__xxx(before|after)_spin_lock */
  527. spin_lock(&mm->page_table_lock);
  528. wait_split_huge_page = 0;
  529. if (likely(pmd_none(*pmd))) { /* Has another populated it ? */
  530. mm->nr_ptes++;
  531. pmd_populate(mm, pmd, new);
  532. new = NULL;
  533. } else if (unlikely(pmd_trans_splitting(*pmd)))
  534. wait_split_huge_page = 1;
  535. spin_unlock(&mm->page_table_lock);
  536. if (new)
  537. pte_free(mm, new);
  538. if (wait_split_huge_page)
  539. wait_split_huge_page(vma->anon_vma, pmd);
  540. return 0;
  541. }
  542. int __pte_alloc_kernel(pmd_t *pmd, unsigned long address)
  543. {
  544. pte_t *new = pte_alloc_one_kernel(&init_mm, address);
  545. if (!new)
  546. return -ENOMEM;
  547. smp_wmb(); /* See comment in __pte_alloc */
  548. spin_lock(&init_mm.page_table_lock);
  549. if (likely(pmd_none(*pmd))) { /* Has another populated it ? */
  550. pmd_populate_kernel(&init_mm, pmd, new);
  551. new = NULL;
  552. } else
  553. VM_BUG_ON(pmd_trans_splitting(*pmd));
  554. spin_unlock(&init_mm.page_table_lock);
  555. if (new)
  556. pte_free_kernel(&init_mm, new);
  557. return 0;
  558. }
  559. static inline void init_rss_vec(int *rss)
  560. {
  561. memset(rss, 0, sizeof(int) * NR_MM_COUNTERS);
  562. }
  563. static inline void add_mm_rss_vec(struct mm_struct *mm, int *rss)
  564. {
  565. int i;
  566. if (current->mm == mm)
  567. sync_mm_rss(current, mm);
  568. for (i = 0; i < NR_MM_COUNTERS; i++)
  569. if (rss[i])
  570. add_mm_counter(mm, i, rss[i]);
  571. }
  572. /*
  573. * This function is called to print an error when a bad pte
  574. * is found. For example, we might have a PFN-mapped pte in
  575. * a region that doesn't allow it.
  576. *
  577. * The calling function must still handle the error.
  578. */
  579. static void print_bad_pte(struct vm_area_struct *vma, unsigned long addr,
  580. pte_t pte, struct page *page)
  581. {
  582. pgd_t *pgd = pgd_offset(vma->vm_mm, addr);
  583. pud_t *pud = pud_offset(pgd, addr);
  584. pmd_t *pmd = pmd_offset(pud, addr);
  585. struct address_space *mapping;
  586. pgoff_t index;
  587. static unsigned long resume;
  588. static unsigned long nr_shown;
  589. static unsigned long nr_unshown;
  590. /*
  591. * Allow a burst of 60 reports, then keep quiet for that minute;
  592. * or allow a steady drip of one report per second.
  593. */
  594. if (nr_shown == 60) {
  595. if (time_before(jiffies, resume)) {
  596. nr_unshown++;
  597. return;
  598. }
  599. if (nr_unshown) {
  600. printk(KERN_ALERT
  601. "BUG: Bad page map: %lu messages suppressed\n",
  602. nr_unshown);
  603. nr_unshown = 0;
  604. }
  605. nr_shown = 0;
  606. }
  607. if (nr_shown++ == 0)
  608. resume = jiffies + 60 * HZ;
  609. mapping = vma->vm_file ? vma->vm_file->f_mapping : NULL;
  610. index = linear_page_index(vma, addr);
  611. printk(KERN_ALERT
  612. "BUG: Bad page map in process %s pte:%08llx pmd:%08llx\n",
  613. current->comm,
  614. (long long)pte_val(pte), (long long)pmd_val(*pmd));
  615. if (page)
  616. dump_page(page);
  617. printk(KERN_ALERT
  618. "addr:%p vm_flags:%08lx anon_vma:%p mapping:%p index:%lx\n",
  619. (void *)addr, vma->vm_flags, vma->anon_vma, mapping, index);
  620. /*
  621. * Choose text because data symbols depend on CONFIG_KALLSYMS_ALL=y
  622. */
  623. if (vma->vm_ops)
  624. print_symbol(KERN_ALERT "vma->vm_ops->fault: %s\n",
  625. (unsigned long)vma->vm_ops->fault);
  626. if (vma->vm_file && vma->vm_file->f_op)
  627. print_symbol(KERN_ALERT "vma->vm_file->f_op->mmap: %s\n",
  628. (unsigned long)vma->vm_file->f_op->mmap);
  629. dump_stack();
  630. add_taint(TAINT_BAD_PAGE);
  631. }
  632. static inline int is_cow_mapping(vm_flags_t flags)
  633. {
  634. return (flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE;
  635. }
  636. #ifndef is_zero_pfn
  637. static inline int is_zero_pfn(unsigned long pfn)
  638. {
  639. return pfn == zero_pfn;
  640. }
  641. #endif
  642. #ifndef my_zero_pfn
  643. static inline unsigned long my_zero_pfn(unsigned long addr)
  644. {
  645. return zero_pfn;
  646. }
  647. #endif
  648. /*
  649. * vm_normal_page -- This function gets the "struct page" associated with a pte.
  650. *
  651. * "Special" mappings do not wish to be associated with a "struct page" (either
  652. * it doesn't exist, or it exists but they don't want to touch it). In this
  653. * case, NULL is returned here. "Normal" mappings do have a struct page.
  654. *
  655. * There are 2 broad cases. Firstly, an architecture may define a pte_special()
  656. * pte bit, in which case this function is trivial. Secondly, an architecture
  657. * may not have a spare pte bit, which requires a more complicated scheme,
  658. * described below.
  659. *
  660. * A raw VM_PFNMAP mapping (ie. one that is not COWed) is always considered a
  661. * special mapping (even if there are underlying and valid "struct pages").
  662. * COWed pages of a VM_PFNMAP are always normal.
  663. *
  664. * The way we recognize COWed pages within VM_PFNMAP mappings is through the
  665. * rules set up by "remap_pfn_range()": the vma will have the VM_PFNMAP bit
  666. * set, and the vm_pgoff will point to the first PFN mapped: thus every special
  667. * mapping will always honor the rule
  668. *
  669. * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT)
  670. *
  671. * And for normal mappings this is false.
  672. *
  673. * This restricts such mappings to be a linear translation from virtual address
  674. * to pfn. To get around this restriction, we allow arbitrary mappings so long
  675. * as the vma is not a COW mapping; in that case, we know that all ptes are
  676. * special (because none can have been COWed).
  677. *
  678. *
  679. * In order to support COW of arbitrary special mappings, we have VM_MIXEDMAP.
  680. *
  681. * VM_MIXEDMAP mappings can likewise contain memory with or without "struct
  682. * page" backing, however the difference is that _all_ pages with a struct
  683. * page (that is, those where pfn_valid is true) are refcounted and considered
  684. * normal pages by the VM. The disadvantage is that pages are refcounted
  685. * (which can be slower and simply not an option for some PFNMAP users). The
  686. * advantage is that we don't have to follow the strict linearity rule of
  687. * PFNMAP mappings in order to support COWable mappings.
  688. *
  689. */
  690. #ifdef __HAVE_ARCH_PTE_SPECIAL
  691. # define HAVE_PTE_SPECIAL 1
  692. #else
  693. # define HAVE_PTE_SPECIAL 0
  694. #endif
  695. struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr,
  696. pte_t pte)
  697. {
  698. unsigned long pfn = pte_pfn(pte);
  699. if (HAVE_PTE_SPECIAL) {
  700. if (likely(!pte_special(pte)))
  701. goto check_pfn;
  702. if (vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP))
  703. return NULL;
  704. if (!is_zero_pfn(pfn))
  705. print_bad_pte(vma, addr, pte, NULL);
  706. return NULL;
  707. }
  708. /* !HAVE_PTE_SPECIAL case follows: */
  709. if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) {
  710. if (vma->vm_flags & VM_MIXEDMAP) {
  711. if (!pfn_valid(pfn))
  712. return NULL;
  713. goto out;
  714. } else {
  715. unsigned long off;
  716. off = (addr - vma->vm_start) >> PAGE_SHIFT;
  717. if (pfn == vma->vm_pgoff + off)
  718. return NULL;
  719. if (!is_cow_mapping(vma->vm_flags))
  720. return NULL;
  721. }
  722. }
  723. if (is_zero_pfn(pfn))
  724. return NULL;
  725. check_pfn:
  726. if (unlikely(pfn > highest_memmap_pfn)) {
  727. print_bad_pte(vma, addr, pte, NULL);
  728. return NULL;
  729. }
  730. /*
  731. * NOTE! We still have PageReserved() pages in the page tables.
  732. * eg. VDSO mappings can cause them to exist.
  733. */
  734. out:
  735. return pfn_to_page(pfn);
  736. }
  737. /*
  738. * copy one vm_area from one task to the other. Assumes the page tables
  739. * already present in the new task to be cleared in the whole range
  740. * covered by this vma.
  741. */
  742. static inline unsigned long
  743. copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm,
  744. pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma,
  745. unsigned long addr, int *rss)
  746. {
  747. unsigned long vm_flags = vma->vm_flags;
  748. pte_t pte = *src_pte;
  749. struct page *page;
  750. /* pte contains position in swap or file, so copy. */
  751. if (unlikely(!pte_present(pte))) {
  752. if (!pte_file(pte)) {
  753. swp_entry_t entry = pte_to_swp_entry(pte);
  754. if (swap_duplicate(entry) < 0)
  755. return entry.val;
  756. /* make sure dst_mm is on swapoff's mmlist. */
  757. if (unlikely(list_empty(&dst_mm->mmlist))) {
  758. spin_lock(&mmlist_lock);
  759. if (list_empty(&dst_mm->mmlist))
  760. list_add(&dst_mm->mmlist,
  761. &src_mm->mmlist);
  762. spin_unlock(&mmlist_lock);
  763. }
  764. if (likely(!non_swap_entry(entry)))
  765. rss[MM_SWAPENTS]++;
  766. else if (is_write_migration_entry(entry) &&
  767. is_cow_mapping(vm_flags)) {
  768. /*
  769. * COW mappings require pages in both parent
  770. * and child to be set to read.
  771. */
  772. make_migration_entry_read(&entry);
  773. pte = swp_entry_to_pte(entry);
  774. set_pte_at(src_mm, addr, src_pte, pte);
  775. }
  776. }
  777. goto out_set_pte;
  778. }
  779. /*
  780. * If it's a COW mapping, write protect it both
  781. * in the parent and the child
  782. */
  783. if (is_cow_mapping(vm_flags)) {
  784. ptep_set_wrprotect(src_mm, addr, src_pte);
  785. pte = pte_wrprotect(pte);
  786. }
  787. /*
  788. * If it's a shared mapping, mark it clean in
  789. * the child
  790. */
  791. if (vm_flags & VM_SHARED)
  792. pte = pte_mkclean(pte);
  793. pte = pte_mkold(pte);
  794. page = vm_normal_page(vma, addr, pte);
  795. if (page) {
  796. get_page(page);
  797. page_dup_rmap(page);
  798. if (PageAnon(page))
  799. rss[MM_ANONPAGES]++;
  800. else
  801. rss[MM_FILEPAGES]++;
  802. }
  803. out_set_pte:
  804. set_pte_at(dst_mm, addr, dst_pte, pte);
  805. return 0;
  806. }
  807. int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
  808. pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma,
  809. unsigned long addr, unsigned long end)
  810. {
  811. pte_t *orig_src_pte, *orig_dst_pte;
  812. pte_t *src_pte, *dst_pte;
  813. spinlock_t *src_ptl, *dst_ptl;
  814. int progress = 0;
  815. int rss[NR_MM_COUNTERS];
  816. swp_entry_t entry = (swp_entry_t){0};
  817. again:
  818. init_rss_vec(rss);
  819. dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl);
  820. if (!dst_pte)
  821. return -ENOMEM;
  822. src_pte = pte_offset_map(src_pmd, addr);
  823. src_ptl = pte_lockptr(src_mm, src_pmd);
  824. spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING);
  825. orig_src_pte = src_pte;
  826. orig_dst_pte = dst_pte;
  827. arch_enter_lazy_mmu_mode();
  828. do {
  829. /*
  830. * We are holding two locks at this point - either of them
  831. * could generate latencies in another task on another CPU.
  832. */
  833. if (progress >= 32) {
  834. progress = 0;
  835. if (need_resched() ||
  836. spin_needbreak(src_ptl) || spin_needbreak(dst_ptl))
  837. break;
  838. }
  839. if (pte_none(*src_pte)) {
  840. progress++;
  841. continue;
  842. }
  843. entry.val = copy_one_pte(dst_mm, src_mm, dst_pte, src_pte,
  844. vma, addr, rss);
  845. if (entry.val)
  846. break;
  847. progress += 8;
  848. } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end);
  849. arch_leave_lazy_mmu_mode();
  850. spin_unlock(src_ptl);
  851. pte_unmap(orig_src_pte);
  852. add_mm_rss_vec(dst_mm, rss);
  853. pte_unmap_unlock(orig_dst_pte, dst_ptl);
  854. cond_resched();
  855. if (entry.val) {
  856. if (add_swap_count_continuation(entry, GFP_KERNEL) < 0)
  857. return -ENOMEM;
  858. progress = 0;
  859. }
  860. if (addr != end)
  861. goto again;
  862. return 0;
  863. }
  864. static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
  865. pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma,
  866. unsigned long addr, unsigned long end)
  867. {
  868. pmd_t *src_pmd, *dst_pmd;
  869. unsigned long next;
  870. dst_pmd = pmd_alloc(dst_mm, dst_pud, addr);
  871. if (!dst_pmd)
  872. return -ENOMEM;
  873. src_pmd = pmd_offset(src_pud, addr);
  874. do {
  875. next = pmd_addr_end(addr, end);
  876. if (pmd_trans_huge(*src_pmd)) {
  877. int err;
  878. VM_BUG_ON(next-addr != HPAGE_PMD_SIZE);
  879. err = copy_huge_pmd(dst_mm, src_mm,
  880. dst_pmd, src_pmd, addr, vma);
  881. if (err == -ENOMEM)
  882. return -ENOMEM;
  883. if (!err)
  884. continue;
  885. /* fall through */
  886. }
  887. if (pmd_none_or_clear_bad(src_pmd))
  888. continue;
  889. if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd,
  890. vma, addr, next))
  891. return -ENOMEM;
  892. } while (dst_pmd++, src_pmd++, addr = next, addr != end);
  893. return 0;
  894. }
  895. static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
  896. pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma,
  897. unsigned long addr, unsigned long end)
  898. {
  899. pud_t *src_pud, *dst_pud;
  900. unsigned long next;
  901. dst_pud = pud_alloc(dst_mm, dst_pgd, addr);
  902. if (!dst_pud)
  903. return -ENOMEM;
  904. src_pud = pud_offset(src_pgd, addr);
  905. do {
  906. next = pud_addr_end(addr, end);
  907. if (pud_none_or_clear_bad(src_pud))
  908. continue;
  909. if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud,
  910. vma, addr, next))
  911. return -ENOMEM;
  912. } while (dst_pud++, src_pud++, addr = next, addr != end);
  913. return 0;
  914. }
  915. int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
  916. struct vm_area_struct *vma)
  917. {
  918. pgd_t *src_pgd, *dst_pgd;
  919. unsigned long next;
  920. unsigned long addr = vma->vm_start;
  921. unsigned long end = vma->vm_end;
  922. int ret;
  923. /*
  924. * Don't copy ptes where a page fault will fill them correctly.
  925. * Fork becomes much lighter when there are big shared or private
  926. * readonly mappings. The tradeoff is that copy_page_range is more
  927. * efficient than faulting.
  928. */
  929. if (!(vma->vm_flags & (VM_HUGETLB|VM_NONLINEAR|VM_PFNMAP|VM_INSERTPAGE))) {
  930. if (!vma->anon_vma)
  931. return 0;
  932. }
  933. if (is_vm_hugetlb_page(vma))
  934. return copy_hugetlb_page_range(dst_mm, src_mm, vma);
  935. if (unlikely(is_pfn_mapping(vma))) {
  936. /*
  937. * We do not free on error cases below as remove_vma
  938. * gets called on error from higher level routine
  939. */
  940. ret = track_pfn_vma_copy(vma);
  941. if (ret)
  942. return ret;
  943. }
  944. /*
  945. * We need to invalidate the secondary MMU mappings only when
  946. * there could be a permission downgrade on the ptes of the
  947. * parent mm. And a permission downgrade will only happen if
  948. * is_cow_mapping() returns true.
  949. */
  950. if (is_cow_mapping(vma->vm_flags))
  951. mmu_notifier_invalidate_range_start(src_mm, addr, end);
  952. ret = 0;
  953. dst_pgd = pgd_offset(dst_mm, addr);
  954. src_pgd = pgd_offset(src_mm, addr);
  955. do {
  956. next = pgd_addr_end(addr, end);
  957. if (pgd_none_or_clear_bad(src_pgd))
  958. continue;
  959. if (unlikely(copy_pud_range(dst_mm, src_mm, dst_pgd, src_pgd,
  960. vma, addr, next))) {
  961. ret = -ENOMEM;
  962. break;
  963. }
  964. } while (dst_pgd++, src_pgd++, addr = next, addr != end);
  965. if (is_cow_mapping(vma->vm_flags))
  966. mmu_notifier_invalidate_range_end(src_mm,
  967. vma->vm_start, end);
  968. return ret;
  969. }
  970. static unsigned long zap_pte_range(struct mmu_gather *tlb,
  971. struct vm_area_struct *vma, pmd_t *pmd,
  972. unsigned long addr, unsigned long end,
  973. struct zap_details *details)
  974. {
  975. struct mm_struct *mm = tlb->mm;
  976. int force_flush = 0;
  977. int rss[NR_MM_COUNTERS];
  978. spinlock_t *ptl;
  979. pte_t *start_pte;
  980. pte_t *pte;
  981. again:
  982. init_rss_vec(rss);
  983. start_pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
  984. pte = start_pte;
  985. arch_enter_lazy_mmu_mode();
  986. do {
  987. pte_t ptent = *pte;
  988. if (pte_none(ptent)) {
  989. continue;
  990. }
  991. if (pte_present(ptent)) {
  992. struct page *page;
  993. page = vm_normal_page(vma, addr, ptent);
  994. if (unlikely(details) && page) {
  995. /*
  996. * unmap_shared_mapping_pages() wants to
  997. * invalidate cache without truncating:
  998. * unmap shared but keep private pages.
  999. */
  1000. if (details->check_mapping &&
  1001. details->check_mapping != page->mapping)
  1002. continue;
  1003. /*
  1004. * Each page->index must be checked when
  1005. * invalidating or truncating nonlinear.
  1006. */
  1007. if (details->nonlinear_vma &&
  1008. (page->index < details->first_index ||
  1009. page->index > details->last_index))
  1010. continue;
  1011. }
  1012. ptent = ptep_get_and_clear_full(mm, addr, pte,
  1013. tlb->fullmm);
  1014. tlb_remove_tlb_entry(tlb, pte, addr);
  1015. if (unlikely(!page))
  1016. continue;
  1017. if (unlikely(details) && details->nonlinear_vma
  1018. && linear_page_index(details->nonlinear_vma,
  1019. addr) != page->index)
  1020. set_pte_at(mm, addr, pte,
  1021. pgoff_to_pte(page->index));
  1022. if (PageAnon(page))
  1023. rss[MM_ANONPAGES]--;
  1024. else {
  1025. if (pte_dirty(ptent))
  1026. set_page_dirty(page);
  1027. if (pte_young(ptent) &&
  1028. likely(!VM_SequentialReadHint(vma)))
  1029. mark_page_accessed(page);
  1030. rss[MM_FILEPAGES]--;
  1031. }
  1032. page_remove_rmap(page);
  1033. if (unlikely(page_mapcount(page) < 0))
  1034. print_bad_pte(vma, addr, ptent, page);
  1035. force_flush = !__tlb_remove_page(tlb, page);
  1036. if (force_flush)
  1037. break;
  1038. continue;
  1039. }
  1040. /*
  1041. * If details->check_mapping, we leave swap entries;
  1042. * if details->nonlinear_vma, we leave file entries.
  1043. */
  1044. if (unlikely(details))
  1045. continue;
  1046. if (pte_file(ptent)) {
  1047. if (unlikely(!(vma->vm_flags & VM_NONLINEAR)))
  1048. print_bad_pte(vma, addr, ptent, NULL);
  1049. } else {
  1050. swp_entry_t entry = pte_to_swp_entry(ptent);
  1051. if (!non_swap_entry(entry))
  1052. rss[MM_SWAPENTS]--;
  1053. if (unlikely(!free_swap_and_cache(entry)))
  1054. print_bad_pte(vma, addr, ptent, NULL);
  1055. }
  1056. pte_clear_not_present_full(mm, addr, pte, tlb->fullmm);
  1057. } while (pte++, addr += PAGE_SIZE, addr != end);
  1058. add_mm_rss_vec(mm, rss);
  1059. arch_leave_lazy_mmu_mode();
  1060. pte_unmap_unlock(start_pte, ptl);
  1061. /*
  1062. * mmu_gather ran out of room to batch pages, we break out of
  1063. * the PTE lock to avoid doing the potential expensive TLB invalidate
  1064. * and page-free while holding it.
  1065. */
  1066. if (force_flush) {
  1067. force_flush = 0;
  1068. tlb_flush_mmu(tlb);
  1069. if (addr != end)
  1070. goto again;
  1071. }
  1072. return addr;
  1073. }
  1074. static inline unsigned long zap_pmd_range(struct mmu_gather *tlb,
  1075. struct vm_area_struct *vma, pud_t *pud,
  1076. unsigned long addr, unsigned long end,
  1077. struct zap_details *details)
  1078. {
  1079. pmd_t *pmd;
  1080. unsigned long next;
  1081. pmd = pmd_offset(pud, addr);
  1082. do {
  1083. next = pmd_addr_end(addr, end);
  1084. if (pmd_trans_huge(*pmd)) {
  1085. if (next-addr != HPAGE_PMD_SIZE) {
  1086. VM_BUG_ON(!rwsem_is_locked(&tlb->mm->mmap_sem));
  1087. split_huge_page_pmd(vma->vm_mm, pmd);
  1088. } else if (zap_huge_pmd(tlb, vma, pmd))
  1089. continue;
  1090. /* fall through */
  1091. }
  1092. if (pmd_none_or_clear_bad(pmd))
  1093. continue;
  1094. next = zap_pte_range(tlb, vma, pmd, addr, next, details);
  1095. cond_resched();
  1096. } while (pmd++, addr = next, addr != end);
  1097. return addr;
  1098. }
  1099. static inline unsigned long zap_pud_range(struct mmu_gather *tlb,
  1100. struct vm_area_struct *vma, pgd_t *pgd,
  1101. unsigned long addr, unsigned long end,
  1102. struct zap_details *details)
  1103. {
  1104. pud_t *pud;
  1105. unsigned long next;
  1106. pud = pud_offset(pgd, addr);
  1107. do {
  1108. next = pud_addr_end(addr, end);
  1109. if (pud_none_or_clear_bad(pud))
  1110. continue;
  1111. next = zap_pmd_range(tlb, vma, pud, addr, next, details);
  1112. } while (pud++, addr = next, addr != end);
  1113. return addr;
  1114. }
  1115. static unsigned long unmap_page_range(struct mmu_gather *tlb,
  1116. struct vm_area_struct *vma,
  1117. unsigned long addr, unsigned long end,
  1118. struct zap_details *details)
  1119. {
  1120. pgd_t *pgd;
  1121. unsigned long next;
  1122. if (details && !details->check_mapping && !details->nonlinear_vma)
  1123. details = NULL;
  1124. BUG_ON(addr >= end);
  1125. mem_cgroup_uncharge_start();
  1126. tlb_start_vma(tlb, vma);
  1127. pgd = pgd_offset(vma->vm_mm, addr);
  1128. do {
  1129. next = pgd_addr_end(addr, end);
  1130. if (pgd_none_or_clear_bad(pgd))
  1131. continue;
  1132. next = zap_pud_range(tlb, vma, pgd, addr, next, details);
  1133. } while (pgd++, addr = next, addr != end);
  1134. tlb_end_vma(tlb, vma);
  1135. mem_cgroup_uncharge_end();
  1136. return addr;
  1137. }
  1138. #ifdef CONFIG_PREEMPT
  1139. # define ZAP_BLOCK_SIZE (8 * PAGE_SIZE)
  1140. #else
  1141. /* No preempt: go for improved straight-line efficiency */
  1142. # define ZAP_BLOCK_SIZE (1024 * PAGE_SIZE)
  1143. #endif
  1144. /**
  1145. * unmap_vmas - unmap a range of memory covered by a list of vma's
  1146. * @tlb: address of the caller's struct mmu_gather
  1147. * @vma: the starting vma
  1148. * @start_addr: virtual address at which to start unmapping
  1149. * @end_addr: virtual address at which to end unmapping
  1150. * @nr_accounted: Place number of unmapped pages in vm-accountable vma's here
  1151. * @details: details of nonlinear truncation or shared cache invalidation
  1152. *
  1153. * Returns the end address of the unmapping (restart addr if interrupted).
  1154. *
  1155. * Unmap all pages in the vma list.
  1156. *
  1157. * We aim to not hold locks for too long (for scheduling latency reasons).
  1158. * So zap pages in ZAP_BLOCK_SIZE bytecounts. This means we need to
  1159. * return the ending mmu_gather to the caller.
  1160. *
  1161. * Only addresses between `start' and `end' will be unmapped.
  1162. *
  1163. * The VMA list must be sorted in ascending virtual address order.
  1164. *
  1165. * unmap_vmas() assumes that the caller will flush the whole unmapped address
  1166. * range after unmap_vmas() returns. So the only responsibility here is to
  1167. * ensure that any thus-far unmapped pages are flushed before unmap_vmas()
  1168. * drops the lock and schedules.
  1169. */
  1170. unsigned long unmap_vmas(struct mmu_gather *tlb,
  1171. struct vm_area_struct *vma, unsigned long start_addr,
  1172. unsigned long end_addr, unsigned long *nr_accounted,
  1173. struct zap_details *details)
  1174. {
  1175. unsigned long start = start_addr;
  1176. struct mm_struct *mm = vma->vm_mm;
  1177. mmu_notifier_invalidate_range_start(mm, start_addr, end_addr);
  1178. for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) {
  1179. unsigned long end;
  1180. start = max(vma->vm_start, start_addr);
  1181. if (start >= vma->vm_end)
  1182. continue;
  1183. end = min(vma->vm_end, end_addr);
  1184. if (end <= vma->vm_start)
  1185. continue;
  1186. if (vma->vm_flags & VM_ACCOUNT)
  1187. *nr_accounted += (end - start) >> PAGE_SHIFT;
  1188. if (unlikely(is_pfn_mapping(vma)))
  1189. untrack_pfn_vma(vma, 0, 0);
  1190. while (start != end) {
  1191. if (unlikely(is_vm_hugetlb_page(vma))) {
  1192. /*
  1193. * It is undesirable to test vma->vm_file as it
  1194. * should be non-null for valid hugetlb area.
  1195. * However, vm_file will be NULL in the error
  1196. * cleanup path of do_mmap_pgoff. When
  1197. * hugetlbfs ->mmap method fails,
  1198. * do_mmap_pgoff() nullifies vma->vm_file
  1199. * before calling this function to clean up.
  1200. * Since no pte has actually been setup, it is
  1201. * safe to do nothing in this case.
  1202. */
  1203. if (vma->vm_file)
  1204. unmap_hugepage_range(vma, start, end, NULL);
  1205. start = end;
  1206. } else
  1207. start = unmap_page_range(tlb, vma, start, end, details);
  1208. }
  1209. }
  1210. mmu_notifier_invalidate_range_end(mm, start_addr, end_addr);
  1211. return start; /* which is now the end (or restart) address */
  1212. }
  1213. /**
  1214. * zap_page_range - remove user pages in a given range
  1215. * @vma: vm_area_struct holding the applicable pages
  1216. * @address: starting address of pages to zap
  1217. * @size: number of bytes to zap
  1218. * @details: details of nonlinear truncation or shared cache invalidation
  1219. */
  1220. unsigned long zap_page_range(struct vm_area_struct *vma, unsigned long address,
  1221. unsigned long size, struct zap_details *details)
  1222. {
  1223. struct mm_struct *mm = vma->vm_mm;
  1224. struct mmu_gather tlb;
  1225. unsigned long end = address + size;
  1226. unsigned long nr_accounted = 0;
  1227. lru_add_drain();
  1228. tlb_gather_mmu(&tlb, mm, 0);
  1229. update_hiwater_rss(mm);
  1230. end = unmap_vmas(&tlb, vma, address, end, &nr_accounted, details);
  1231. tlb_finish_mmu(&tlb, address, end);
  1232. return end;
  1233. }
  1234. /**
  1235. * zap_vma_ptes - remove ptes mapping the vma
  1236. * @vma: vm_area_struct holding ptes to be zapped
  1237. * @address: starting address of pages to zap
  1238. * @size: number of bytes to zap
  1239. *
  1240. * This function only unmaps ptes assigned to VM_PFNMAP vmas.
  1241. *
  1242. * The entire address range must be fully contained within the vma.
  1243. *
  1244. * Returns 0 if successful.
  1245. */
  1246. int zap_vma_ptes(struct vm_area_struct *vma, unsigned long address,
  1247. unsigned long size)
  1248. {
  1249. if (address < vma->vm_start || address + size > vma->vm_end ||
  1250. !(vma->vm_flags & VM_PFNMAP))
  1251. return -1;
  1252. zap_page_range(vma, address, size, NULL);
  1253. return 0;
  1254. }
  1255. EXPORT_SYMBOL_GPL(zap_vma_ptes);
  1256. /**
  1257. * follow_page - look up a page descriptor from a user-virtual address
  1258. * @vma: vm_area_struct mapping @address
  1259. * @address: virtual address to look up
  1260. * @flags: flags modifying lookup behaviour
  1261. *
  1262. * @flags can have FOLL_ flags set, defined in <linux/mm.h>
  1263. *
  1264. * Returns the mapped (struct page *), %NULL if no mapping exists, or
  1265. * an error pointer if there is a mapping to something not represented
  1266. * by a page descriptor (see also vm_normal_page()).
  1267. */
  1268. struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
  1269. unsigned int flags)
  1270. {
  1271. pgd_t *pgd;
  1272. pud_t *pud;
  1273. pmd_t *pmd;
  1274. pte_t *ptep, pte;
  1275. spinlock_t *ptl;
  1276. struct page *page;
  1277. struct mm_struct *mm = vma->vm_mm;
  1278. page = follow_huge_addr(mm, address, flags & FOLL_WRITE);
  1279. if (!IS_ERR(page)) {
  1280. BUG_ON(flags & FOLL_GET);
  1281. goto out;
  1282. }
  1283. page = NULL;
  1284. pgd = pgd_offset(mm, address);
  1285. if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
  1286. goto no_page_table;
  1287. pud = pud_offset(pgd, address);
  1288. if (pud_none(*pud))
  1289. goto no_page_table;
  1290. if (pud_huge(*pud) && vma->vm_flags & VM_HUGETLB) {
  1291. BUG_ON(flags & FOLL_GET);
  1292. page = follow_huge_pud(mm, address, pud, flags & FOLL_WRITE);
  1293. goto out;
  1294. }
  1295. if (unlikely(pud_bad(*pud)))
  1296. goto no_page_table;
  1297. pmd = pmd_offset(pud, address);
  1298. if (pmd_none(*pmd))
  1299. goto no_page_table;
  1300. if (pmd_huge(*pmd) && vma->vm_flags & VM_HUGETLB) {
  1301. BUG_ON(flags & FOLL_GET);
  1302. page = follow_huge_pmd(mm, address, pmd, flags & FOLL_WRITE);
  1303. goto out;
  1304. }
  1305. if (pmd_trans_huge(*pmd)) {
  1306. if (flags & FOLL_SPLIT) {
  1307. split_huge_page_pmd(mm, pmd);
  1308. goto split_fallthrough;
  1309. }
  1310. spin_lock(&mm->page_table_lock);
  1311. if (likely(pmd_trans_huge(*pmd))) {
  1312. if (unlikely(pmd_trans_splitting(*pmd))) {
  1313. spin_unlock(&mm->page_table_lock);
  1314. wait_split_huge_page(vma->anon_vma, pmd);
  1315. } else {
  1316. page = follow_trans_huge_pmd(mm, address,
  1317. pmd, flags);
  1318. spin_unlock(&mm->page_table_lock);
  1319. goto out;
  1320. }
  1321. } else
  1322. spin_unlock(&mm->page_table_lock);
  1323. /* fall through */
  1324. }
  1325. split_fallthrough:
  1326. if (unlikely(pmd_bad(*pmd)))
  1327. goto no_page_table;
  1328. ptep = pte_offset_map_lock(mm, pmd, address, &ptl);
  1329. pte = *ptep;
  1330. if (!pte_present(pte))
  1331. goto no_page;
  1332. if ((flags & FOLL_WRITE) && !pte_write(pte))
  1333. goto unlock;
  1334. page = vm_normal_page(vma, address, pte);
  1335. if (unlikely(!page)) {
  1336. if ((flags & FOLL_DUMP) ||
  1337. !is_zero_pfn(pte_pfn(pte)))
  1338. goto bad_page;
  1339. page = pte_page(pte);
  1340. }
  1341. if (flags & FOLL_GET)
  1342. get_page(page);
  1343. if (flags & FOLL_TOUCH) {
  1344. if ((flags & FOLL_WRITE) &&
  1345. !pte_dirty(pte) && !PageDirty(page))
  1346. set_page_dirty(page);
  1347. /*
  1348. * pte_mkyoung() would be more correct here, but atomic care
  1349. * is needed to avoid losing the dirty bit: it is easier to use
  1350. * mark_page_accessed().
  1351. */
  1352. mark_page_accessed(page);
  1353. }
  1354. if ((flags & FOLL_MLOCK) && (vma->vm_flags & VM_LOCKED)) {
  1355. /*
  1356. * The preliminary mapping check is mainly to avoid the
  1357. * pointless overhead of lock_page on the ZERO_PAGE
  1358. * which might bounce very badly if there is contention.
  1359. *
  1360. * If the page is already locked, we don't need to
  1361. * handle it now - vmscan will handle it later if and
  1362. * when it attempts to reclaim the page.
  1363. */
  1364. if (page->mapping && trylock_page(page)) {
  1365. lru_add_drain(); /* push cached pages to LRU */
  1366. /*
  1367. * Because we lock page here and migration is
  1368. * blocked by the pte's page reference, we need
  1369. * only check for file-cache page truncation.
  1370. */
  1371. if (page->mapping)
  1372. mlock_vma_page(page);
  1373. unlock_page(page);
  1374. }
  1375. }
  1376. unlock:
  1377. pte_unmap_unlock(ptep, ptl);
  1378. out:
  1379. return page;
  1380. bad_page:
  1381. pte_unmap_unlock(ptep, ptl);
  1382. return ERR_PTR(-EFAULT);
  1383. no_page:
  1384. pte_unmap_unlock(ptep, ptl);
  1385. if (!pte_none(pte))
  1386. return page;
  1387. no_page_table:
  1388. /*
  1389. * When core dumping an enormous anonymous area that nobody
  1390. * has touched so far, we don't want to allocate unnecessary pages or
  1391. * page tables. Return error instead of NULL to skip handle_mm_fault,
  1392. * then get_dump_page() will return NULL to leave a hole in the dump.
  1393. * But we can only make this optimization where a hole would surely
  1394. * be zero-filled if handle_mm_fault() actually did handle it.
  1395. */
  1396. if ((flags & FOLL_DUMP) &&
  1397. (!vma->vm_ops || !vma->vm_ops->fault))
  1398. return ERR_PTR(-EFAULT);
  1399. return page;
  1400. }
  1401. static inline int stack_guard_page(struct vm_area_struct *vma, unsigned long addr)
  1402. {
  1403. return stack_guard_page_start(vma, addr) ||
  1404. stack_guard_page_end(vma, addr+PAGE_SIZE);
  1405. }
  1406. /**
  1407. * __get_user_pages() - pin user pages in memory
  1408. * @tsk: task_struct of target task
  1409. * @mm: mm_struct of target mm
  1410. * @start: starting user address
  1411. * @nr_pages: number of pages from start to pin
  1412. * @gup_flags: flags modifying pin behaviour
  1413. * @pages: array that receives pointers to the pages pinned.
  1414. * Should be at least nr_pages long. Or NULL, if caller
  1415. * only intends to ensure the pages are faulted in.
  1416. * @vmas: array of pointers to vmas corresponding to each page.
  1417. * Or NULL if the caller does not require them.
  1418. * @nonblocking: whether waiting for disk IO or mmap_sem contention
  1419. *
  1420. * Returns number of pages pinned. This may be fewer than the number
  1421. * requested. If nr_pages is 0 or negative, returns 0. If no pages
  1422. * were pinned, returns -errno. Each page returned must be released
  1423. * with a put_page() call when it is finished with. vmas will only
  1424. * remain valid while mmap_sem is held.
  1425. *
  1426. * Must be called with mmap_sem held for read or write.
  1427. *
  1428. * __get_user_pages walks a process's page tables and takes a reference to
  1429. * each struct page that each user address corresponds to at a given
  1430. * instant. That is, it takes the page that would be accessed if a user
  1431. * thread accesses the given user virtual address at that instant.
  1432. *
  1433. * This does not guarantee that the page exists in the user mappings when
  1434. * __get_user_pages returns, and there may even be a completely different
  1435. * page there in some cases (eg. if mmapped pagecache has been invalidated
  1436. * and subsequently re faulted). However it does guarantee that the page
  1437. * won't be freed completely. And mostly callers simply care that the page
  1438. * contains data that was valid *at some point in time*. Typically, an IO
  1439. * or similar operation cannot guarantee anything stronger anyway because
  1440. * locks can't be held over the syscall boundary.
  1441. *
  1442. * If @gup_flags & FOLL_WRITE == 0, the page must not be written to. If
  1443. * the page is written to, set_page_dirty (or set_page_dirty_lock, as
  1444. * appropriate) must be called after the page is finished with, and
  1445. * before put_page is called.
  1446. *
  1447. * If @nonblocking != NULL, __get_user_pages will not wait for disk IO
  1448. * or mmap_sem contention, and if waiting is needed to pin all pages,
  1449. * *@nonblocking will be set to 0.
  1450. *
  1451. * In most cases, get_user_pages or get_user_pages_fast should be used
  1452. * instead of __get_user_pages. __get_user_pages should be used only if
  1453. * you need some special @gup_flags.
  1454. */
  1455. int __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
  1456. unsigned long start, int nr_pages, unsigned int gup_flags,
  1457. struct page **pages, struct vm_area_struct **vmas,
  1458. int *nonblocking)
  1459. {
  1460. int i;
  1461. unsigned long vm_flags;
  1462. if (nr_pages <= 0)
  1463. return 0;
  1464. VM_BUG_ON(!!pages != !!(gup_flags & FOLL_GET));
  1465. /*
  1466. * Require read or write permissions.
  1467. * If FOLL_FORCE is set, we only require the "MAY" flags.
  1468. */
  1469. vm_flags = (gup_flags & FOLL_WRITE) ?
  1470. (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
  1471. vm_flags &= (gup_flags & FOLL_FORCE) ?
  1472. (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
  1473. i = 0;
  1474. do {
  1475. struct vm_area_struct *vma;
  1476. vma = find_extend_vma(mm, start);
  1477. if (!vma && in_gate_area(mm, start)) {
  1478. unsigned long pg = start & PAGE_MASK;
  1479. pgd_t *pgd;
  1480. pud_t *pud;
  1481. pmd_t *pmd;
  1482. pte_t *pte;
  1483. /* user gate pages are read-only */
  1484. if (gup_flags & FOLL_WRITE)
  1485. return i ? : -EFAULT;
  1486. if (pg > TASK_SIZE)
  1487. pgd = pgd_offset_k(pg);
  1488. else
  1489. pgd = pgd_offset_gate(mm, pg);
  1490. BUG_ON(pgd_none(*pgd));
  1491. pud = pud_offset(pgd, pg);
  1492. BUG_ON(pud_none(*pud));
  1493. pmd = pmd_offset(pud, pg);
  1494. if (pmd_none(*pmd))
  1495. return i ? : -EFAULT;
  1496. VM_BUG_ON(pmd_trans_huge(*pmd));
  1497. pte = pte_offset_map(pmd, pg);
  1498. if (pte_none(*pte)) {
  1499. pte_unmap(pte);
  1500. return i ? : -EFAULT;
  1501. }
  1502. vma = get_gate_vma(mm);
  1503. if (pages) {
  1504. struct page *page;
  1505. page = vm_normal_page(vma, start, *pte);
  1506. if (!page) {
  1507. if (!(gup_flags & FOLL_DUMP) &&
  1508. is_zero_pfn(pte_pfn(*pte)))
  1509. page = pte_page(*pte);
  1510. else {
  1511. pte_unmap(pte);
  1512. return i ? : -EFAULT;
  1513. }
  1514. }
  1515. pages[i] = page;
  1516. get_page(page);
  1517. }
  1518. pte_unmap(pte);
  1519. goto next_page;
  1520. }
  1521. if (!vma ||
  1522. (vma->vm_flags & (VM_IO | VM_PFNMAP)) ||
  1523. !(vm_flags & vma->vm_flags))
  1524. return i ? : -EFAULT;
  1525. if (is_vm_hugetlb_page(vma)) {
  1526. i = follow_hugetlb_page(mm, vma, pages, vmas,
  1527. &start, &nr_pages, i, gup_flags);
  1528. continue;
  1529. }
  1530. do {
  1531. struct page *page;
  1532. unsigned int foll_flags = gup_flags;
  1533. /*
  1534. * If we have a pending SIGKILL, don't keep faulting
  1535. * pages and potentially allocating memory.
  1536. */
  1537. if (unlikely(fatal_signal_pending(current)))
  1538. return i ? i : -ERESTARTSYS;
  1539. cond_resched();
  1540. while (!(page = follow_page(vma, start, foll_flags))) {
  1541. int ret;
  1542. unsigned int fault_flags = 0;
  1543. /* For mlock, just skip the stack guard page. */
  1544. if (foll_flags & FOLL_MLOCK) {
  1545. if (stack_guard_page(vma, start))
  1546. goto next_page;
  1547. }
  1548. if (foll_flags & FOLL_WRITE)
  1549. fault_flags |= FAULT_FLAG_WRITE;
  1550. if (nonblocking)
  1551. fault_flags |= FAULT_FLAG_ALLOW_RETRY;
  1552. if (foll_flags & FOLL_NOWAIT)
  1553. fault_flags |= (FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_RETRY_NOWAIT);
  1554. ret = handle_mm_fault(mm, vma, start,
  1555. fault_flags);
  1556. if (ret & VM_FAULT_ERROR) {
  1557. if (ret & VM_FAULT_OOM)
  1558. return i ? i : -ENOMEM;
  1559. if (ret & (VM_FAULT_HWPOISON |
  1560. VM_FAULT_HWPOISON_LARGE)) {
  1561. if (i)
  1562. return i;
  1563. else if (gup_flags & FOLL_HWPOISON)
  1564. return -EHWPOISON;
  1565. else
  1566. return -EFAULT;
  1567. }
  1568. if (ret & VM_FAULT_SIGBUS)
  1569. return i ? i : -EFAULT;
  1570. BUG();
  1571. }
  1572. if (tsk) {
  1573. if (ret & VM_FAULT_MAJOR)
  1574. tsk->maj_flt++;
  1575. else
  1576. tsk->min_flt++;
  1577. }
  1578. if (ret & VM_FAULT_RETRY) {
  1579. if (nonblocking)
  1580. *nonblocking = 0;
  1581. return i;
  1582. }
  1583. /*
  1584. * The VM_FAULT_WRITE bit tells us that
  1585. * do_wp_page has broken COW when necessary,
  1586. * even if maybe_mkwrite decided not to set
  1587. * pte_write. We can thus safely do subsequent
  1588. * page lookups as if they were reads. But only
  1589. * do so when looping for pte_write is futile:
  1590. * in some cases userspace may also be wanting
  1591. * to write to the gotten user page, which a
  1592. * read fault here might prevent (a readonly
  1593. * page might get reCOWed by userspace write).
  1594. */
  1595. if ((ret & VM_FAULT_WRITE) &&
  1596. !(vma->vm_flags & VM_WRITE))
  1597. foll_flags &= ~FOLL_WRITE;
  1598. cond_resched();
  1599. }
  1600. if (IS_ERR(page))
  1601. return i ? i : PTR_ERR(page);
  1602. if (pages) {
  1603. pages[i] = page;
  1604. flush_anon_page(vma, page, start);
  1605. flush_dcache_page(page);
  1606. }
  1607. next_page:
  1608. if (vmas)
  1609. vmas[i] = vma;
  1610. i++;
  1611. start += PAGE_SIZE;
  1612. nr_pages--;
  1613. } while (nr_pages && start < vma->vm_end);
  1614. } while (nr_pages);
  1615. return i;
  1616. }
  1617. EXPORT_SYMBOL(__get_user_pages);
  1618. /**
  1619. * get_user_pages() - pin user pages in memory
  1620. * @tsk: the task_struct to use for page fault accounting, or
  1621. * NULL if faults are not to be recorded.
  1622. * @mm: mm_struct of target mm
  1623. * @start: starting user address
  1624. * @nr_pages: number of pages from start to pin
  1625. * @write: whether pages will be written to by the caller
  1626. * @force: whether to force write access even if user mapping is
  1627. * readonly. This will result in the page being COWed even
  1628. * in MAP_SHARED mappings. You do not want this.
  1629. * @pages: array that receives pointers to the pages pinned.
  1630. * Should be at least nr_pages long. Or NULL, if caller
  1631. * only intends to ensure the pages are faulted in.
  1632. * @vmas: array of pointers to vmas corresponding to each page.
  1633. * Or NULL if the caller does not require them.
  1634. *
  1635. * Returns number of pages pinned. This may be fewer than the number
  1636. * requested. If nr_pages is 0 or negative, returns 0. If no pages
  1637. * were pinned, returns -errno. Each page returned must be released
  1638. * with a put_page() call when it is finished with. vmas will only
  1639. * remain valid while mmap_sem is held.
  1640. *
  1641. * Must be called with mmap_sem held for read or write.
  1642. *
  1643. * get_user_pages walks a process's page tables and takes a reference to
  1644. * each struct page that each user address corresponds to at a given
  1645. * instant. That is, it takes the page that would be accessed if a user
  1646. * thread accesses the given user virtual address at that instant.
  1647. *
  1648. * This does not guarantee that the page exists in the user mappings when
  1649. * get_user_pages returns, and there may even be a completely different
  1650. * page there in some cases (eg. if mmapped pagecache has been invalidated
  1651. * and subsequently re faulted). However it does guarantee that the page
  1652. * won't be freed completely. And mostly callers simply care that the page
  1653. * contains data that was valid *at some point in time*. Typically, an IO
  1654. * or similar operation cannot guarantee anything stronger anyway because
  1655. * locks can't be held over the syscall boundary.
  1656. *
  1657. * If write=0, the page must not be written to. If the page is written to,
  1658. * set_page_dirty (or set_page_dirty_lock, as appropriate) must be called
  1659. * after the page is finished with, and before put_page is called.
  1660. *
  1661. * get_user_pages is typically used for fewer-copy IO operations, to get a
  1662. * handle on the memory by some means other than accesses via the user virtual
  1663. * addresses. The pages may be submitted for DMA to devices or accessed via
  1664. * their kernel linear mapping (via the kmap APIs). Care should be taken to
  1665. * use the correct cache flushing APIs.
  1666. *
  1667. * See also get_user_pages_fast, for performance critical applications.
  1668. */
  1669. int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
  1670. unsigned long start, int nr_pages, int write, int force,
  1671. struct page **pages, struct vm_area_struct **vmas)
  1672. {
  1673. int flags = FOLL_TOUCH;
  1674. if (pages)
  1675. flags |= FOLL_GET;
  1676. if (write)
  1677. flags |= FOLL_WRITE;
  1678. if (force)
  1679. flags |= FOLL_FORCE;
  1680. return __get_user_pages(tsk, mm, start, nr_pages, flags, pages, vmas,
  1681. NULL);
  1682. }
  1683. EXPORT_SYMBOL(get_user_pages);
  1684. /**
  1685. * get_dump_page() - pin user page in memory while writing it to core dump
  1686. * @addr: user address
  1687. *
  1688. * Returns struct page pointer of user page pinned for dump,
  1689. * to be freed afterwards by page_cache_release() or put_page().
  1690. *
  1691. * Returns NULL on any kind of failure - a hole must then be inserted into
  1692. * the corefile, to preserve alignment with its headers; and also returns
  1693. * NULL wherever the ZERO_PAGE, or an anonymous pte_none, has been found -
  1694. * allowing a hole to be left in the corefile to save diskspace.
  1695. *
  1696. * Called without mmap_sem, but after all other threads have been killed.
  1697. */
  1698. #ifdef CONFIG_ELF_CORE
  1699. struct page *get_dump_page(unsigned long addr)
  1700. {
  1701. struct vm_area_struct *vma;
  1702. struct page *page;
  1703. if (__get_user_pages(current, current->mm, addr, 1,
  1704. FOLL_FORCE | FOLL_DUMP | FOLL_GET, &page, &vma,
  1705. NULL) < 1)
  1706. return NULL;
  1707. flush_cache_page(vma, addr, page_to_pfn(page));
  1708. return page;
  1709. }
  1710. #endif /* CONFIG_ELF_CORE */
  1711. pte_t *__get_locked_pte(struct mm_struct *mm, unsigned long addr,
  1712. spinlock_t **ptl)
  1713. {
  1714. pgd_t * pgd = pgd_offset(mm, addr);
  1715. pud_t * pud = pud_alloc(mm, pgd, addr);
  1716. if (pud) {
  1717. pmd_t * pmd = pmd_alloc(mm, pud, addr);
  1718. if (pmd) {
  1719. VM_BUG_ON(pmd_trans_huge(*pmd));
  1720. return pte_alloc_map_lock(mm, pmd, addr, ptl);
  1721. }
  1722. }
  1723. return NULL;
  1724. }
  1725. /*
  1726. * This is the old fallback for page remapping.
  1727. *
  1728. * For historical reasons, it only allows reserved pages. Only
  1729. * old drivers should use this, and they needed to mark their
  1730. * pages reserved for the old functions anyway.
  1731. */
  1732. static int insert_page(struct vm_area_struct *vma, unsigned long addr,
  1733. struct page *page, pgprot_t prot)
  1734. {
  1735. struct mm_struct *mm = vma->vm_mm;
  1736. int retval;
  1737. pte_t *pte;
  1738. spinlock_t *ptl;
  1739. retval = -EINVAL;
  1740. if (PageAnon(page))
  1741. goto out;
  1742. retval = -ENOMEM;
  1743. flush_dcache_page(page);
  1744. pte = get_locked_pte(mm, addr, &ptl);
  1745. if (!pte)
  1746. goto out;
  1747. retval = -EBUSY;
  1748. if (!pte_none(*pte))
  1749. goto out_unlock;
  1750. /* Ok, finally just insert the thing.. */
  1751. get_page(page);
  1752. inc_mm_counter_fast(mm, MM_FILEPAGES);
  1753. page_add_file_rmap(page);
  1754. set_pte_at(mm, addr, pte, mk_pte(page, prot));
  1755. retval = 0;
  1756. pte_unmap_unlock(pte, ptl);
  1757. return retval;
  1758. out_unlock:
  1759. pte_unmap_unlock(pte, ptl);
  1760. out:
  1761. return retval;
  1762. }
  1763. /**
  1764. * vm_insert_page - insert single page into user vma
  1765. * @vma: user vma to map to
  1766. * @addr: target user address of this page
  1767. * @page: source kernel page
  1768. *
  1769. * This allows drivers to insert individual pages they've allocated
  1770. * into a user vma.
  1771. *
  1772. * The page has to be a nice clean _individual_ kernel allocation.
  1773. * If you allocate a compound page, you need to have marked it as
  1774. * such (__GFP_COMP), or manually just split the page up yourself
  1775. * (see split_page()).
  1776. *
  1777. * NOTE! Traditionally this was done with "remap_pfn_range()" which
  1778. * took an arbitrary page protection parameter. This doesn't allow
  1779. * that. Your vma protection will have to be set up correctly, which
  1780. * means that if you want a shared writable mapping, you'd better
  1781. * ask for a shared writable mapping!
  1782. *
  1783. * The page does not need to be reserved.
  1784. */
  1785. int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
  1786. struct page *page)
  1787. {
  1788. if (addr < vma->vm_start || addr >= vma->vm_end)
  1789. return -EFAULT;
  1790. if (!page_count(page))
  1791. return -EINVAL;
  1792. vma->vm_flags |= VM_INSERTPAGE;
  1793. return insert_page(vma, addr, page, vma->vm_page_prot);
  1794. }
  1795. EXPORT_SYMBOL(vm_insert_page);
  1796. static int insert_pfn(struct vm_area_struct *vma, unsigned long addr,
  1797. unsigned long pfn, pgprot_t prot)
  1798. {
  1799. struct mm_struct *mm = vma->vm_mm;
  1800. int retval;
  1801. pte_t *pte, entry;
  1802. spinlock_t *ptl;
  1803. retval = -ENOMEM;
  1804. pte = get_locked_pte(mm, addr, &ptl);
  1805. if (!pte)
  1806. goto out;
  1807. retval = -EBUSY;
  1808. if (!pte_none(*pte))
  1809. goto out_unlock;
  1810. /* Ok, finally just insert the thing.. */
  1811. entry = pte_mkspecial(pfn_pte(pfn, prot));
  1812. set_pte_at(mm, addr, pte, entry);
  1813. update_mmu_cache(vma, addr, pte); /* XXX: why not for insert_page? */
  1814. retval = 0;
  1815. out_unlock:
  1816. pte_unmap_unlock(pte, ptl);
  1817. out:
  1818. return retval;
  1819. }
  1820. /**
  1821. * vm_insert_pfn - insert single pfn into user vma
  1822. * @vma: user vma to map to
  1823. * @addr: target user address of this page
  1824. * @pfn: source kernel pfn
  1825. *
  1826. * Similar to vm_inert_page, this allows drivers to insert individual pages
  1827. * they've allocated into a user vma. Same comments apply.
  1828. *
  1829. * This function should only be called from a vm_ops->fault handler, and
  1830. * in that case the handler should return NULL.
  1831. *
  1832. * vma cannot be a COW mapping.
  1833. *
  1834. * As this is called only for pages that do not currently exist, we
  1835. * do not need to flush old virtual caches or the TLB.
  1836. */
  1837. int vm_insert_pfn(struct vm_area_struct *vma, unsigned long addr,
  1838. unsigned long pfn)
  1839. {
  1840. int ret;
  1841. pgprot_t pgprot = vma->vm_page_prot;
  1842. /*
  1843. * Technically, architectures with pte_special can avoid all these
  1844. * restrictions (same for remap_pfn_range). However we would like
  1845. * consistency in testing and feature parity among all, so we should
  1846. * try to keep these invariants in place for everybody.
  1847. */
  1848. BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)));
  1849. BUG_ON((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) ==
  1850. (VM_PFNMAP|VM_MIXEDMAP));
  1851. BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags));
  1852. BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn));
  1853. if (addr < vma->vm_start || addr >= vma->vm_end)
  1854. return -EFAULT;
  1855. if (track_pfn_vma_new(vma, &pgprot, pfn, PAGE_SIZE))
  1856. return -EINVAL;
  1857. ret = insert_pfn(vma, addr, pfn, pgprot);
  1858. if (ret)
  1859. untrack_pfn_vma(vma, pfn, PAGE_SIZE);
  1860. return ret;
  1861. }
  1862. EXPORT_SYMBOL(vm_insert_pfn);
  1863. int vm_insert_mixed(struct vm_area_struct *vma, unsigned long addr,
  1864. unsigned long pfn)
  1865. {
  1866. BUG_ON(!(vma->vm_flags & VM_MIXEDMAP));
  1867. if (addr < vma->vm_start || addr >= vma->vm_end)
  1868. return -EFAULT;
  1869. /*
  1870. * If we don't have pte special, then we have to use the pfn_valid()
  1871. * based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must*
  1872. * refcount the page if pfn_valid is true (hence insert_page rather
  1873. * than insert_pfn). If a zero_pfn were inserted into a VM_MIXEDMAP
  1874. * without pte special, it would there be refcounted as a normal page.
  1875. */
  1876. if (!HAVE_PTE_SPECIAL && pfn_valid(pfn)) {
  1877. struct page *page;
  1878. page = pfn_to_page(pfn);
  1879. return insert_page(vma, addr, page, vma->vm_page_prot);
  1880. }
  1881. return insert_pfn(vma, addr, pfn, vma->vm_page_prot);
  1882. }
  1883. EXPORT_SYMBOL(vm_insert_mixed);
  1884. /*
  1885. * maps a range of physical memory into the requested pages. the old
  1886. * mappings are removed. any references to nonexistent pages results
  1887. * in null mappings (currently treated as "copy-on-access")
  1888. */
  1889. static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd,
  1890. unsigned long addr, unsigned long end,
  1891. unsigned long pfn, pgprot_t prot)
  1892. {
  1893. pte_t *pte;
  1894. spinlock_t *ptl;
  1895. pte = pte_alloc_map_lock(mm, pmd, addr, &ptl);
  1896. if (!pte)
  1897. return -ENOMEM;
  1898. arch_enter_lazy_mmu_mode();
  1899. do {
  1900. BUG_ON(!pte_none(*pte));
  1901. set_pte_at(mm, addr, pte, pte_mkspecial(pfn_pte(pfn, prot)));
  1902. pfn++;
  1903. } while (pte++, addr += PAGE_SIZE, addr != end);
  1904. arch_leave_lazy_mmu_mode();
  1905. pte_unmap_unlock(pte - 1, ptl);
  1906. return 0;
  1907. }
  1908. static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud,
  1909. unsigned long addr, unsigned long end,
  1910. unsigned long pfn, pgprot_t prot)
  1911. {
  1912. pmd_t *pmd;
  1913. unsigned long next;
  1914. pfn -= addr >> PAGE_SHIFT;
  1915. pmd = pmd_alloc(mm, pud, addr);
  1916. if (!pmd)
  1917. return -ENOMEM;
  1918. VM_BUG_ON(pmd_trans_huge(*pmd));
  1919. do {
  1920. next = pmd_addr_end(addr, end);
  1921. if (remap_pte_range(mm, pmd, addr, next,
  1922. pfn + (addr >> PAGE_SHIFT), prot))
  1923. return -ENOMEM;
  1924. } while (pmd++, addr = next, addr != end);
  1925. return 0;
  1926. }
  1927. static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd,
  1928. unsigned long addr, unsigned long end,
  1929. unsigned long pfn, pgprot_t prot)
  1930. {
  1931. pud_t *pud;
  1932. unsigned long next;
  1933. pfn -= addr >> PAGE_SHIFT;
  1934. pud = pud_alloc(mm, pgd, addr);
  1935. if (!pud)
  1936. return -ENOMEM;
  1937. do {
  1938. next = pud_addr_end(addr, end);
  1939. if (remap_pmd_range(mm, pud, addr, next,
  1940. pfn + (addr >> PAGE_SHIFT), prot))
  1941. return -ENOMEM;
  1942. } while (pud++, addr = next, addr != end);
  1943. return 0;
  1944. }
  1945. /**
  1946. * remap_pfn_range - remap kernel memory to userspace
  1947. * @vma: user vma to map to
  1948. * @addr: target user address to start at
  1949. * @pfn: physical address of kernel memory
  1950. * @size: size of map area
  1951. * @prot: page protection flags for this mapping
  1952. *
  1953. * Note: this is only safe if the mm semaphore is held when called.
  1954. */
  1955. int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
  1956. unsigned long pfn, unsigned long size, pgprot_t prot)
  1957. {
  1958. pgd_t *pgd;
  1959. unsigned long next;
  1960. unsigned long end = addr + PAGE_ALIGN(size);
  1961. struct mm_struct *mm = vma->vm_mm;
  1962. int err;
  1963. /*
  1964. * Physically remapped pages are special. Tell the
  1965. * rest of the world about it:
  1966. * VM_IO tells people not to look at these pages
  1967. * (accesses can have side effects).
  1968. * VM_RESERVED is specified all over the place, because
  1969. * in 2.4 it kept swapout's vma scan off this vma; but
  1970. * in 2.6 the LRU scan won't even find its pages, so this
  1971. * flag means no more than count its pages in reserved_vm,
  1972. * and omit it from core dump, even when VM_IO turned off.
  1973. * VM_PFNMAP tells the core MM that the base pages are just
  1974. * raw PFN mappings, and do not have a "struct page" associated
  1975. * with them.
  1976. *
  1977. * There's a horrible special case to handle copy-on-write
  1978. * behaviour that some programs depend on. We mark the "original"
  1979. * un-COW'ed pages by matching them up with "vma->vm_pgoff".
  1980. */
  1981. if (addr == vma->vm_start && end == vma->vm_end) {
  1982. vma->vm_pgoff = pfn;
  1983. vma->vm_flags |= VM_PFN_AT_MMAP;
  1984. } else if (is_cow_mapping(vma->vm_flags))
  1985. return -EINVAL;
  1986. vma->vm_flags |= VM_IO | VM_RESERVED | VM_PFNMAP;
  1987. err = track_pfn_vma_new(vma, &prot, pfn, PAGE_ALIGN(size));
  1988. if (err) {
  1989. /*
  1990. * To indicate that track_pfn related cleanup is not
  1991. * needed from higher level routine calling unmap_vmas
  1992. */
  1993. vma->vm_flags &= ~(VM_IO | VM_RESERVED | VM_PFNMAP);
  1994. vma->vm_flags &= ~VM_PFN_AT_MMAP;
  1995. return -EINVAL;
  1996. }
  1997. BUG_ON(addr >= end);
  1998. pfn -= addr >> PAGE_SHIFT;
  1999. pgd = pgd_offset(mm, addr);
  2000. flush_cache_range(vma, addr, end);
  2001. do {
  2002. next = pgd_addr_end(addr, end);
  2003. err = remap_pud_range(mm, pgd, addr, next,
  2004. pfn + (addr >> PAGE_SHIFT), prot);
  2005. if (err)
  2006. break;
  2007. } while (pgd++, addr = next, addr != end);
  2008. if (err)
  2009. untrack_pfn_vma(vma, pfn, PAGE_ALIGN(size));
  2010. return err;
  2011. }
  2012. EXPORT_SYMBOL(remap_pfn_range);
  2013. static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd,
  2014. unsigned long addr, unsigned long end,
  2015. pte_fn_t fn, void *data)
  2016. {
  2017. pte_t *pte;
  2018. int err;
  2019. pgtable_t token;
  2020. spinlock_t *uninitialized_var(ptl);
  2021. pte = (mm == &init_mm) ?
  2022. pte_alloc_kernel(pmd, addr) :
  2023. pte_alloc_map_lock(mm, pmd, addr, &ptl);
  2024. if (!pte)
  2025. return -ENOMEM;
  2026. BUG_ON(pmd_huge(*pmd));
  2027. arch_enter_lazy_mmu_mode();
  2028. token = pmd_pgtable(*pmd);
  2029. do {
  2030. err = fn(pte++, token, addr, data);
  2031. if (err)
  2032. break;
  2033. } while (addr += PAGE_SIZE, addr != end);
  2034. arch_leave_lazy_mmu_mode();
  2035. if (mm != &init_mm)
  2036. pte_unmap_unlock(pte-1, ptl);
  2037. return err;
  2038. }
  2039. static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud,
  2040. unsigned long addr, unsigned long end,
  2041. pte_fn_t fn, void *data)
  2042. {
  2043. pmd_t *pmd;
  2044. unsigned long next;
  2045. int err;
  2046. BUG_ON(pud_huge(*pud));
  2047. pmd = pmd_alloc(mm, pud, addr);
  2048. if (!pmd)
  2049. return -ENOMEM;
  2050. do {
  2051. next = pmd_addr_end(addr, end);
  2052. err = apply_to_pte_range(mm, pmd, addr, next, fn, data);
  2053. if (err)
  2054. break;
  2055. } while (pmd++, addr = next, addr != end);
  2056. return err;
  2057. }
  2058. static int apply_to_pud_range(struct mm_struct *mm, pgd_t *pgd,
  2059. unsigned long addr, unsigned long end,
  2060. pte_fn_t fn, void *data)
  2061. {
  2062. pud_t *pud;
  2063. unsigned long next;
  2064. int err;
  2065. pud = pud_alloc(mm, pgd, addr);
  2066. if (!pud)
  2067. return -ENOMEM;
  2068. do {
  2069. next = pud_addr_end(addr, end);
  2070. err = apply_to_pmd_range(mm, pud, addr, next, fn, data);
  2071. if (err)
  2072. break;
  2073. } while (pud++, addr = next, addr != end);
  2074. return err;
  2075. }
  2076. /*
  2077. * Scan a region of virtual memory, filling in page tables as necessary
  2078. * and calling a provided function on each leaf page table.
  2079. */
  2080. int apply_to_page_range(struct mm_struct *mm, unsigned long addr,
  2081. unsigned long size, pte_fn_t fn, void *data)
  2082. {
  2083. pgd_t *pgd;
  2084. unsigned long next;
  2085. unsigned long end = addr + size;
  2086. int err;
  2087. BUG_ON(addr >= end);
  2088. pgd = pgd_offset(mm, addr);
  2089. do {
  2090. next = pgd_addr_end(addr, end);
  2091. err = apply_to_pud_range(mm, pgd, addr, next, fn, data);
  2092. if (err)
  2093. break;
  2094. } while (pgd++, addr = next, addr != end);
  2095. return err;
  2096. }
  2097. EXPORT_SYMBOL_GPL(apply_to_page_range);
  2098. /*
  2099. * handle_pte_fault chooses page fault handler according to an entry
  2100. * which was read non-atomically. Before making any commitment, on
  2101. * those architectures or configurations (e.g. i386 with PAE) which
  2102. * might give a mix of unmatched parts, do_swap_page and do_nonlinear_fault
  2103. * must check under lock before unmapping the pte and proceeding
  2104. * (but do_wp_page is only called after already making such a check;
  2105. * and do_anonymous_page can safely check later on).
  2106. */
  2107. static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd,
  2108. pte_t *page_table, pte_t orig_pte)
  2109. {
  2110. int same = 1;
  2111. #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT)
  2112. if (sizeof(pte_t) > sizeof(unsigned long)) {
  2113. spinlock_t *ptl = pte_lockptr(mm, pmd);
  2114. spin_lock(ptl);
  2115. same = pte_same(*page_table, orig_pte);
  2116. spin_unlock(ptl);
  2117. }
  2118. #endif
  2119. pte_unmap(page_table);
  2120. return same;
  2121. }
  2122. static inline void cow_user_page(struct page *dst, struct page *src, unsigned long va, struct vm_area_struct *vma)
  2123. {
  2124. /*
  2125. * If the source page was a PFN mapping, we don't have
  2126. * a "struct page" for it. We do a best-effort copy by
  2127. * just copying from the original user address. If that
  2128. * fails, we just zero-fill it. Live with it.
  2129. */
  2130. if (unlikely(!src)) {
  2131. void *kaddr = kmap_atomic(dst, KM_USER0);
  2132. void __user *uaddr = (void __user *)(va & PAGE_MASK);
  2133. /*
  2134. * This really shouldn't fail, because the page is there
  2135. * in the page tables. But it might just be unreadable,
  2136. * in which case we just give up and fill the result with
  2137. * zeroes.
  2138. */
  2139. if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE))
  2140. clear_page(kaddr);
  2141. kunmap_atomic(kaddr, KM_USER0);
  2142. flush_dcache_page(dst);
  2143. } else
  2144. copy_user_highpage(dst, src, va, vma);
  2145. }
  2146. /*
  2147. * This routine handles present pages, when users try to write
  2148. * to a shared page. It is done by copying the page to a new address
  2149. * and decrementing the shared-page counter for the old page.
  2150. *
  2151. * Note that this routine assumes that the protection checks have been
  2152. * done by the caller (the low-level page fault routine in most cases).
  2153. * Thus we can safely just mark it writable once we've done any necessary
  2154. * COW.
  2155. *
  2156. * We also mark the page dirty at this point even though the page will
  2157. * change only once the write actually happens. This avoids a few races,
  2158. * and potentially makes it more efficient.
  2159. *
  2160. * We enter with non-exclusive mmap_sem (to exclude vma changes,
  2161. * but allow concurrent faults), with pte both mapped and locked.
  2162. * We return with mmap_sem still held, but pte unmapped and unlocked.
  2163. */
  2164. static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma,
  2165. unsigned long address, pte_t *page_table, pmd_t *pmd,
  2166. spinlock_t *ptl, pte_t orig_pte)
  2167. __releases(ptl)
  2168. {
  2169. struct page *old_page, *new_page;
  2170. pte_t entry;
  2171. int ret = 0;
  2172. int page_mkwrite = 0;
  2173. struct page *dirty_page = NULL;
  2174. old_page = vm_normal_page(vma, address, orig_pte);
  2175. if (!old_page) {
  2176. /*
  2177. * VM_MIXEDMAP !pfn_valid() case
  2178. *
  2179. * We should not cow pages in a shared writeable mapping.
  2180. * Just mark the pages writable as we can't do any dirty
  2181. * accounting on raw pfn maps.
  2182. */
  2183. if ((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==
  2184. (VM_WRITE|VM_SHARED))
  2185. goto reuse;
  2186. goto gotten;
  2187. }
  2188. /*
  2189. * Take out anonymous pages first, anonymous shared vmas are
  2190. * not dirty accountable.
  2191. */
  2192. if (PageAnon(old_page) && !PageKsm(old_page)) {
  2193. if (!trylock_page(old_page)) {
  2194. page_cache_get(old_page);
  2195. pte_unmap_unlock(page_table, ptl);
  2196. lock_page(old_page);
  2197. page_table = pte_offset_map_lock(mm, pmd, address,
  2198. &ptl);
  2199. if (!pte_same(*page_table, orig_pte)) {
  2200. unlock_page(old_page);
  2201. goto unlock;
  2202. }
  2203. page_cache_release(old_page);
  2204. }
  2205. if (reuse_swap_page(old_page)) {
  2206. /*
  2207. * The page is all ours. Move it to our anon_vma so
  2208. * the rmap code will not search our parent or siblings.
  2209. * Protected against the rmap code by the page lock.
  2210. */
  2211. page_move_anon_rmap(old_page, vma, address);
  2212. unlock_page(old_page);
  2213. goto reuse;
  2214. }
  2215. unlock_page(old_page);
  2216. } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==
  2217. (VM_WRITE|VM_SHARED))) {
  2218. /*
  2219. * Only catch write-faults on shared writable pages,
  2220. * read-only shared pages can get COWed by
  2221. * get_user_pages(.write=1, .force=1).
  2222. */
  2223. if (vma->vm_ops && vma->vm_ops->page_mkwrite) {
  2224. struct vm_fault vmf;
  2225. int tmp;
  2226. vmf.virtual_address = (void __user *)(address &
  2227. PAGE_MASK);
  2228. vmf.pgoff = old_page->index;
  2229. vmf.flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE;
  2230. vmf.page = old_page;
  2231. /*
  2232. * Notify the address space that the page is about to
  2233. * become writable so that it can prohibit this or wait
  2234. * for the page to get into an appropriate state.
  2235. *
  2236. * We do this without the lock held, so that it can
  2237. * sleep if it needs to.
  2238. */
  2239. page_cache_get(old_page);
  2240. pte_unmap_unlock(page_table, ptl);
  2241. tmp = vma->vm_ops->page_mkwrite(vma, &vmf);
  2242. if (unlikely(tmp &
  2243. (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) {
  2244. ret = tmp;
  2245. goto unwritable_page;
  2246. }
  2247. if (unlikely(!(tmp & VM_FAULT_LOCKED))) {
  2248. lock_page(old_page);
  2249. if (!old_page->mapping) {
  2250. ret = 0; /* retry the fault */
  2251. unlock_page(old_page);
  2252. goto unwritable_page;
  2253. }
  2254. } else
  2255. VM_BUG_ON(!PageLocked(old_page));
  2256. /*
  2257. * Since we dropped the lock we need to revalidate
  2258. * the PTE as someone else may have changed it. If
  2259. * they did, we just return, as we can count on the
  2260. * MMU to tell us if they didn't also make it writable.
  2261. */
  2262. page_table = pte_offset_map_lock(mm, pmd, address,
  2263. &ptl);
  2264. if (!pte_same(*page_table, orig_pte)) {
  2265. unlock_page(old_page);
  2266. goto unlock;
  2267. }
  2268. page_mkwrite = 1;
  2269. }
  2270. dirty_page = old_page;
  2271. get_page(dirty_page);
  2272. reuse:
  2273. flush_cache_page(vma, address, pte_pfn(orig_pte));
  2274. entry = pte_mkyoung(orig_pte);
  2275. entry = maybe_mkwrite(pte_mkdirty(entry), vma);
  2276. if (ptep_set_access_flags(vma, address, page_table, entry,1))
  2277. update_mmu_cache(vma, address, page_table);
  2278. pte_unmap_unlock(page_table, ptl);
  2279. ret |= VM_FAULT_WRITE;
  2280. if (!dirty_page)
  2281. return ret;
  2282. /*
  2283. * Yes, Virginia, this is actually required to prevent a race
  2284. * with clear_page_dirty_for_io() from clearing the page dirty
  2285. * bit after it clear all dirty ptes, but before a racing
  2286. * do_wp_page installs a dirty pte.
  2287. *
  2288. * __do_fault is protected similarly.
  2289. */
  2290. if (!page_mkwrite) {
  2291. wait_on_page_locked(dirty_page);
  2292. set_page_dirty_balance(dirty_page, page_mkwrite);
  2293. }
  2294. put_page(dirty_page);
  2295. if (page_mkwrite) {
  2296. struct address_space *mapping = dirty_page->mapping;
  2297. set_page_dirty(dirty_page);
  2298. unlock_page(dirty_page);
  2299. page_cache_release(dirty_page);
  2300. if (mapping) {
  2301. /*
  2302. * Some device drivers do not set page.mapping
  2303. * but still dirty their pages
  2304. */
  2305. balance_dirty_pages_ratelimited(mapping);
  2306. }
  2307. }
  2308. /* file_update_time outside page_lock */
  2309. if (vma->vm_file)
  2310. file_update_time(vma->vm_file);
  2311. return ret;
  2312. }
  2313. /*
  2314. * Ok, we need to copy. Oh, well..
  2315. */
  2316. page_cache_get(old_page);
  2317. gotten:
  2318. pte_unmap_unlock(page_table, ptl);
  2319. if (unlikely(anon_vma_prepare(vma)))
  2320. goto oom;
  2321. if (is_zero_pfn(pte_pfn(orig_pte))) {
  2322. new_page = alloc_zeroed_user_highpage_movable(vma, address);
  2323. if (!new_page)
  2324. goto oom;
  2325. } else {
  2326. new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address);
  2327. if (!new_page)
  2328. goto oom;
  2329. cow_user_page(new_page, old_page, address, vma);
  2330. }
  2331. __SetPageUptodate(new_page);
  2332. if (mem_cgroup_newpage_charge(new_page, mm, GFP_KERNEL))
  2333. goto oom_free_new;
  2334. /*
  2335. * Re-check the pte - we dropped the lock
  2336. */
  2337. page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
  2338. if (likely(pte_same(*page_table, orig_pte))) {
  2339. if (old_page) {
  2340. if (!PageAnon(old_page)) {
  2341. dec_mm_counter_fast(mm, MM_FILEPAGES);
  2342. inc_mm_counter_fast(mm, MM_ANONPAGES);
  2343. }
  2344. } else
  2345. inc_mm_counter_fast(mm, MM_ANONPAGES);
  2346. flush_cache_page(vma, address, pte_pfn(orig_pte));
  2347. entry = mk_pte(new_page, vma->vm_page_prot);
  2348. entry = maybe_mkwrite(pte_mkdirty(entry), vma);
  2349. /*
  2350. * Clear the pte entry and flush it first, before updating the
  2351. * pte with the new entry. This will avoid a race condition
  2352. * seen in the presence of one thread doing SMC and another
  2353. * thread doing COW.
  2354. */
  2355. ptep_clear_flush(vma, address, page_table);
  2356. page_add_new_anon_rmap(new_page, vma, address);
  2357. /*
  2358. * We call the notify macro here because, when using secondary
  2359. * mmu page tables (such as kvm shadow page tables), we want the
  2360. * new page to be mapped directly into the secondary page table.
  2361. */
  2362. set_pte_at_notify(mm, address, page_table, entry);
  2363. update_mmu_cache(vma, address, page_table);
  2364. if (old_page) {
  2365. /*
  2366. * Only after switching the pte to the new page may
  2367. * we remove the mapcount here. Otherwise another
  2368. * process may come and find the rmap count decremented
  2369. * before the pte is switched to the new page, and
  2370. * "reuse" the old page writing into it while our pte
  2371. * here still points into it and can be read by other
  2372. * threads.
  2373. *
  2374. * The critical issue is to order this
  2375. * page_remove_rmap with the ptp_clear_flush above.
  2376. * Those stores are ordered by (if nothing else,)
  2377. * the barrier present in the atomic_add_negative
  2378. * in page_remove_rmap.
  2379. *
  2380. * Then the TLB flush in ptep_clear_flush ensures that
  2381. * no process can access the old page before the
  2382. * decremented mapcount is visible. And the old page
  2383. * cannot be reused until after the decremented
  2384. * mapcount is visible. So transitively, TLBs to
  2385. * old page will be flushed before it can be reused.
  2386. */
  2387. page_remove_rmap(old_page);
  2388. }
  2389. /* Free the old page.. */
  2390. new_page = old_page;
  2391. ret |= VM_FAULT_WRITE;
  2392. } else
  2393. mem_cgroup_uncharge_page(new_page);
  2394. if (new_page)
  2395. page_cache_release(new_page);
  2396. unlock:
  2397. pte_unmap_unlock(page_table, ptl);
  2398. if (old_page) {
  2399. /*
  2400. * Don't let another task, with possibly unlocked vma,
  2401. * keep the mlocked page.
  2402. */
  2403. if ((ret & VM_FAULT_WRITE) && (vma->vm_flags & VM_LOCKED)) {
  2404. lock_page(old_page); /* LRU manipulation */
  2405. munlock_vma_page(old_page);
  2406. unlock_page(old_page);
  2407. }
  2408. page_cache_release(old_page);
  2409. }
  2410. return ret;
  2411. oom_free_new:
  2412. page_cache_release(new_page);
  2413. oom:
  2414. if (old_page) {
  2415. if (page_mkwrite) {
  2416. unlock_page(old_page);
  2417. page_cache_release(old_page);
  2418. }
  2419. page_cache_release(old_page);
  2420. }
  2421. return VM_FAULT_OOM;
  2422. unwritable_page:
  2423. page_cache_release(old_page);
  2424. return ret;
  2425. }
  2426. static void unmap_mapping_range_vma(struct vm_area_struct *vma,
  2427. unsigned long start_addr, unsigned long end_addr,
  2428. struct zap_details *details)
  2429. {
  2430. zap_page_range(vma, start_addr, end_addr - start_addr, details);
  2431. }
  2432. static inline void unmap_mapping_range_tree(struct prio_tree_root *root,
  2433. struct zap_details *details)
  2434. {
  2435. struct vm_area_struct *vma;
  2436. struct prio_tree_iter iter;
  2437. pgoff_t vba, vea, zba, zea;
  2438. vma_prio_tree_foreach(vma, &iter, root,
  2439. details->first_index, details->last_index) {
  2440. vba = vma->vm_pgoff;
  2441. vea = vba + ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT) - 1;
  2442. /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */
  2443. zba = details->first_index;
  2444. if (zba < vba)
  2445. zba = vba;
  2446. zea = details->last_index;
  2447. if (zea > vea)
  2448. zea = vea;
  2449. unmap_mapping_range_vma(vma,
  2450. ((zba - vba) << PAGE_SHIFT) + vma->vm_start,
  2451. ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start,
  2452. details);
  2453. }
  2454. }
  2455. static inline void unmap_mapping_range_list(struct list_head *head,
  2456. struct zap_details *details)
  2457. {
  2458. struct vm_area_struct *vma;
  2459. /*
  2460. * In nonlinear VMAs there is no correspondence between virtual address
  2461. * offset and file offset. So we must perform an exhaustive search
  2462. * across *all* the pages in each nonlinear VMA, not just the pages
  2463. * whose virtual address lies outside the file truncation point.
  2464. */
  2465. list_for_each_entry(vma, head, shared.vm_set.list) {
  2466. details->nonlinear_vma = vma;
  2467. unmap_mapping_range_vma(vma, vma->vm_start, vma->vm_end, details);
  2468. }
  2469. }
  2470. /**
  2471. * unmap_mapping_range - unmap the portion of all mmaps in the specified address_space corresponding to the specified page range in the underlying file.
  2472. * @mapping: the address space containing mmaps to be unmapped.
  2473. * @holebegin: byte in first page to unmap, relative to the start of
  2474. * the underlying file. This will be rounded down to a PAGE_SIZE
  2475. * boundary. Note that this is different from truncate_pagecache(), which
  2476. * must keep the partial page. In contrast, we must get rid of
  2477. * partial pages.
  2478. * @holelen: size of prospective hole in bytes. This will be rounded
  2479. * up to a PAGE_SIZE boundary. A holelen of zero truncates to the
  2480. * end of the file.
  2481. * @even_cows: 1 when truncating a file, unmap even private COWed pages;
  2482. * but 0 when invalidating pagecache, don't throw away private data.
  2483. */
  2484. void unmap_mapping_range(struct address_space *mapping,
  2485. loff_t const holebegin, loff_t const holelen, int even_cows)
  2486. {
  2487. struct zap_details details;
  2488. pgoff_t hba = holebegin >> PAGE_SHIFT;
  2489. pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
  2490. /* Check for overflow. */
  2491. if (sizeof(holelen) > sizeof(hlen)) {
  2492. long long holeend =
  2493. (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
  2494. if (holeend & ~(long long)ULONG_MAX)
  2495. hlen = ULONG_MAX - hba + 1;
  2496. }
  2497. details.check_mapping = even_cows? NULL: mapping;
  2498. details.nonlinear_vma = NULL;
  2499. details.first_index = hba;
  2500. details.last_index = hba + hlen - 1;
  2501. if (details.last_index < details.first_index)
  2502. details.last_index = ULONG_MAX;
  2503. mutex_lock(&mapping->i_mmap_mutex);
  2504. if (unlikely(!prio_tree_empty(&mapping->i_mmap)))
  2505. unmap_mapping_range_tree(&mapping->i_mmap, &details);
  2506. if (unlikely(!list_empty(&mapping->i_mmap_nonlinear)))
  2507. unmap_mapping_range_list(&mapping->i_mmap_nonlinear, &details);
  2508. mutex_unlock(&mapping->i_mmap_mutex);
  2509. }
  2510. EXPORT_SYMBOL(unmap_mapping_range);
  2511. /*
  2512. * We enter with non-exclusive mmap_sem (to exclude vma changes,
  2513. * but allow concurrent faults), and pte mapped but not yet locked.
  2514. * We return with mmap_sem still held, but pte unmapped and unlocked.
  2515. */
  2516. static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma,
  2517. unsigned long address, pte_t *page_table, pmd_t *pmd,
  2518. unsigned int flags, pte_t orig_pte)
  2519. {
  2520. spinlock_t *ptl;
  2521. struct page *page, *swapcache = NULL;
  2522. swp_entry_t entry;
  2523. pte_t pte;
  2524. int locked;
  2525. struct mem_cgroup *ptr;
  2526. int exclusive = 0;
  2527. int ret = 0;
  2528. if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
  2529. goto out;
  2530. entry = pte_to_swp_entry(orig_pte);
  2531. if (unlikely(non_swap_entry(entry))) {
  2532. if (is_migration_entry(entry)) {
  2533. migration_entry_wait(mm, pmd, address);
  2534. } else if (is_hwpoison_entry(entry)) {
  2535. ret = VM_FAULT_HWPOISON;
  2536. } else {
  2537. print_bad_pte(vma, address, orig_pte, NULL);
  2538. ret = VM_FAULT_SIGBUS;
  2539. }
  2540. goto out;
  2541. }
  2542. delayacct_set_flag(DELAYACCT_PF_SWAPIN);
  2543. page = lookup_swap_cache(entry);
  2544. if (!page) {
  2545. grab_swap_token(mm); /* Contend for token _before_ read-in */
  2546. page = swapin_readahead(entry,
  2547. GFP_HIGHUSER_MOVABLE, vma, address);
  2548. if (!page) {
  2549. /*
  2550. * Back out if somebody else faulted in this pte
  2551. * while we released the pte lock.
  2552. */
  2553. page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
  2554. if (likely(pte_same(*page_table, orig_pte)))
  2555. ret = VM_FAULT_OOM;
  2556. delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
  2557. goto unlock;
  2558. }
  2559. /* Had to read the page from swap area: Major fault */
  2560. ret = VM_FAULT_MAJOR;
  2561. count_vm_event(PGMAJFAULT);
  2562. mem_cgroup_count_vm_event(mm, PGMAJFAULT);
  2563. } else if (PageHWPoison(page)) {
  2564. /*
  2565. * hwpoisoned dirty swapcache pages are kept for killing
  2566. * owner processes (which may be unknown at hwpoison time)
  2567. */
  2568. ret = VM_FAULT_HWPOISON;
  2569. delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
  2570. goto out_release;
  2571. }
  2572. locked = lock_page_or_retry(page, mm, flags);
  2573. delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
  2574. if (!locked) {
  2575. ret |= VM_FAULT_RETRY;
  2576. goto out_release;
  2577. }
  2578. /*
  2579. * Make sure try_to_free_swap or reuse_swap_page or swapoff did not
  2580. * release the swapcache from under us. The page pin, and pte_same
  2581. * test below, are not enough to exclude that. Even if it is still
  2582. * swapcache, we need to check that the page's swap has not changed.
  2583. */
  2584. if (unlikely(!PageSwapCache(page) || page_private(page) != entry.val))
  2585. goto out_page;
  2586. if (ksm_might_need_to_copy(page, vma, address)) {
  2587. swapcache = page;
  2588. page = ksm_does_need_to_copy(page, vma, address);
  2589. if (unlikely(!page)) {
  2590. ret = VM_FAULT_OOM;
  2591. page = swapcache;
  2592. swapcache = NULL;
  2593. goto out_page;
  2594. }
  2595. }
  2596. if (mem_cgroup_try_charge_swapin(mm, page, GFP_KERNEL, &ptr)) {
  2597. ret = VM_FAULT_OOM;
  2598. goto out_page;
  2599. }
  2600. /*
  2601. * Back out if somebody else already faulted in this pte.
  2602. */
  2603. page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
  2604. if (unlikely(!pte_same(*page_table, orig_pte)))
  2605. goto out_nomap;
  2606. if (unlikely(!PageUptodate(page))) {
  2607. ret = VM_FAULT_SIGBUS;
  2608. goto out_nomap;
  2609. }
  2610. /*
  2611. * The page isn't present yet, go ahead with the fault.
  2612. *
  2613. * Be careful about the sequence of operations here.
  2614. * To get its accounting right, reuse_swap_page() must be called
  2615. * while the page is counted on swap but not yet in mapcount i.e.
  2616. * before page_add_anon_rmap() and swap_free(); try_to_free_swap()
  2617. * must be called after the swap_free(), or it will never succeed.
  2618. * Because delete_from_swap_page() may be called by reuse_swap_page(),
  2619. * mem_cgroup_commit_charge_swapin() may not be able to find swp_entry
  2620. * in page->private. In this case, a record in swap_cgroup is silently
  2621. * discarded at swap_free().
  2622. */
  2623. inc_mm_counter_fast(mm, MM_ANONPAGES);
  2624. dec_mm_counter_fast(mm, MM_SWAPENTS);
  2625. pte = mk_pte(page, vma->vm_page_prot);
  2626. if ((flags & FAULT_FLAG_WRITE) && reuse_swap_page(page)) {
  2627. pte = maybe_mkwrite(pte_mkdirty(pte), vma);
  2628. flags &= ~FAULT_FLAG_WRITE;
  2629. ret |= VM_FAULT_WRITE;
  2630. exclusive = 1;
  2631. }
  2632. flush_icache_page(vma, page);
  2633. set_pte_at(mm, address, page_table, pte);
  2634. do_page_add_anon_rmap(page, vma, address, exclusive);
  2635. /* It's better to call commit-charge after rmap is established */
  2636. mem_cgroup_commit_charge_swapin(page, ptr);
  2637. swap_free(entry);
  2638. if (vm_swap_full() || (vma->vm_flags & VM_LOCKED) || PageMlocked(page))
  2639. try_to_free_swap(page);
  2640. unlock_page(page);
  2641. if (swapcache) {
  2642. /*
  2643. * Hold the lock to avoid the swap entry to be reused
  2644. * until we take the PT lock for the pte_same() check
  2645. * (to avoid false positives from pte_same). For
  2646. * further safety release the lock after the swap_free
  2647. * so that the swap count won't change under a
  2648. * parallel locked swapcache.
  2649. */
  2650. unlock_page(swapcache);
  2651. page_cache_release(swapcache);
  2652. }
  2653. if (flags & FAULT_FLAG_WRITE) {
  2654. ret |= do_wp_page(mm, vma, address, page_table, pmd, ptl, pte);
  2655. if (ret & VM_FAULT_ERROR)
  2656. ret &= VM_FAULT_ERROR;
  2657. goto out;
  2658. }
  2659. /* No need to invalidate - it was non-present before */
  2660. update_mmu_cache(vma, address, page_table);
  2661. unlock:
  2662. pte_unmap_unlock(page_table, ptl);
  2663. out:
  2664. return ret;
  2665. out_nomap:
  2666. mem_cgroup_cancel_charge_swapin(ptr);
  2667. pte_unmap_unlock(page_table, ptl);
  2668. out_page:
  2669. unlock_page(page);
  2670. out_release:
  2671. page_cache_release(page);
  2672. if (swapcache) {
  2673. unlock_page(swapcache);
  2674. page_cache_release(swapcache);
  2675. }
  2676. return ret;
  2677. }
  2678. /*
  2679. * This is like a special single-page "expand_{down|up}wards()",
  2680. * except we must first make sure that 'address{-|+}PAGE_SIZE'
  2681. * doesn't hit another vma.
  2682. */
  2683. static inline int check_stack_guard_page(struct vm_area_struct *vma, unsigned long address)
  2684. {
  2685. address &= PAGE_MASK;
  2686. if ((vma->vm_flags & VM_GROWSDOWN) && address == vma->vm_start) {
  2687. struct vm_area_struct *prev = vma->vm_prev;
  2688. /*
  2689. * Is there a mapping abutting this one below?
  2690. *
  2691. * That's only ok if it's the same stack mapping
  2692. * that has gotten split..
  2693. */
  2694. if (prev && prev->vm_end == address)
  2695. return prev->vm_flags & VM_GROWSDOWN ? 0 : -ENOMEM;
  2696. expand_downwards(vma, address - PAGE_SIZE);
  2697. }
  2698. if ((vma->vm_flags & VM_GROWSUP) && address + PAGE_SIZE == vma->vm_end) {
  2699. struct vm_area_struct *next = vma->vm_next;
  2700. /* As VM_GROWSDOWN but s/below/above/ */
  2701. if (next && next->vm_start == address + PAGE_SIZE)
  2702. return next->vm_flags & VM_GROWSUP ? 0 : -ENOMEM;
  2703. expand_upwards(vma, address + PAGE_SIZE);
  2704. }
  2705. return 0;
  2706. }
  2707. /*
  2708. * We enter with non-exclusive mmap_sem (to exclude vma changes,
  2709. * but allow concurrent faults), and pte mapped but not yet locked.
  2710. * We return with mmap_sem still held, but pte unmapped and unlocked.
  2711. */
  2712. static int do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma,
  2713. unsigned long address, pte_t *page_table, pmd_t *pmd,
  2714. unsigned int flags)
  2715. {
  2716. struct page *page;
  2717. spinlock_t *ptl;
  2718. pte_t entry;
  2719. pte_unmap(page_table);
  2720. /* Check if we need to add a guard page to the stack */
  2721. if (check_stack_guard_page(vma, address) < 0)
  2722. return VM_FAULT_SIGBUS;
  2723. /* Use the zero-page for reads */
  2724. if (!(flags & FAULT_FLAG_WRITE)) {
  2725. entry = pte_mkspecial(pfn_pte(my_zero_pfn(address),
  2726. vma->vm_page_prot));
  2727. page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
  2728. if (!pte_none(*page_table))
  2729. goto unlock;
  2730. goto setpte;
  2731. }
  2732. /* Allocate our own private page. */
  2733. if (unlikely(anon_vma_prepare(vma)))
  2734. goto oom;
  2735. page = alloc_zeroed_user_highpage_movable(vma, address);
  2736. if (!page)
  2737. goto oom;
  2738. __SetPageUptodate(page);
  2739. if (mem_cgroup_newpage_charge(page, mm, GFP_KERNEL))
  2740. goto oom_free_page;
  2741. entry = mk_pte(page, vma->vm_page_prot);
  2742. if (vma->vm_flags & VM_WRITE)
  2743. entry = pte_mkwrite(pte_mkdirty(entry));
  2744. page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
  2745. if (!pte_none(*page_table))
  2746. goto release;
  2747. inc_mm_counter_fast(mm, MM_ANONPAGES);
  2748. page_add_new_anon_rmap(page, vma, address);
  2749. setpte:
  2750. set_pte_at(mm, address, page_table, entry);
  2751. /* No need to invalidate - it was non-present before */
  2752. update_mmu_cache(vma, address, page_table);
  2753. unlock:
  2754. pte_unmap_unlock(page_table, ptl);
  2755. return 0;
  2756. release:
  2757. mem_cgroup_uncharge_page(page);
  2758. page_cache_release(page);
  2759. goto unlock;
  2760. oom_free_page:
  2761. page_cache_release(page);
  2762. oom:
  2763. return VM_FAULT_OOM;
  2764. }
  2765. /*
  2766. * __do_fault() tries to create a new page mapping. It aggressively
  2767. * tries to share with existing pages, but makes a separate copy if
  2768. * the FAULT_FLAG_WRITE is set in the flags parameter in order to avoid
  2769. * the next page fault.
  2770. *
  2771. * As this is called only for pages that do not currently exist, we
  2772. * do not need to flush old virtual caches or the TLB.
  2773. *
  2774. * We enter with non-exclusive mmap_sem (to exclude vma changes,
  2775. * but allow concurrent faults), and pte neither mapped nor locked.
  2776. * We return with mmap_sem still held, but pte unmapped and unlocked.
  2777. */
  2778. static int __do_fault(struct mm_struct *mm, struct vm_area_struct *vma,
  2779. unsigned long address, pmd_t *pmd,
  2780. pgoff_t pgoff, unsigned int flags, pte_t orig_pte)
  2781. {
  2782. pte_t *page_table;
  2783. spinlock_t *ptl;
  2784. struct page *page;
  2785. pte_t entry;
  2786. int anon = 0;
  2787. int charged = 0;
  2788. struct page *dirty_page = NULL;
  2789. struct vm_fault vmf;
  2790. int ret;
  2791. int page_mkwrite = 0;
  2792. vmf.virtual_address = (void __user *)(address & PAGE_MASK);
  2793. vmf.pgoff = pgoff;
  2794. vmf.flags = flags;
  2795. vmf.page = NULL;
  2796. ret = vma->vm_ops->fault(vma, &vmf);
  2797. if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE |
  2798. VM_FAULT_RETRY)))
  2799. return ret;
  2800. if (unlikely(PageHWPoison(vmf.page))) {
  2801. if (ret & VM_FAULT_LOCKED)
  2802. unlock_page(vmf.page);
  2803. return VM_FAULT_HWPOISON;
  2804. }
  2805. /*
  2806. * For consistency in subsequent calls, make the faulted page always
  2807. * locked.
  2808. */
  2809. if (unlikely(!(ret & VM_FAULT_LOCKED)))
  2810. lock_page(vmf.page);
  2811. else
  2812. VM_BUG_ON(!PageLocked(vmf.page));
  2813. /*
  2814. * Should we do an early C-O-W break?
  2815. */
  2816. page = vmf.page;
  2817. if (flags & FAULT_FLAG_WRITE) {
  2818. if (!(vma->vm_flags & VM_SHARED)) {
  2819. anon = 1;
  2820. if (unlikely(anon_vma_prepare(vma))) {
  2821. ret = VM_FAULT_OOM;
  2822. goto out;
  2823. }
  2824. page = alloc_page_vma(GFP_HIGHUSER_MOVABLE,
  2825. vma, address);
  2826. if (!page) {
  2827. ret = VM_FAULT_OOM;
  2828. goto out;
  2829. }
  2830. if (mem_cgroup_newpage_charge(page, mm, GFP_KERNEL)) {
  2831. ret = VM_FAULT_OOM;
  2832. page_cache_release(page);
  2833. goto out;
  2834. }
  2835. charged = 1;
  2836. copy_user_highpage(page, vmf.page, address, vma);
  2837. __SetPageUptodate(page);
  2838. } else {
  2839. /*
  2840. * If the page will be shareable, see if the backing
  2841. * address space wants to know that the page is about
  2842. * to become writable
  2843. */
  2844. if (vma->vm_ops->page_mkwrite) {
  2845. int tmp;
  2846. unlock_page(page);
  2847. vmf.flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE;
  2848. tmp = vma->vm_ops->page_mkwrite(vma, &vmf);
  2849. if (unlikely(tmp &
  2850. (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) {
  2851. ret = tmp;
  2852. goto unwritable_page;
  2853. }
  2854. if (unlikely(!(tmp & VM_FAULT_LOCKED))) {
  2855. lock_page(page);
  2856. if (!page->mapping) {
  2857. ret = 0; /* retry the fault */
  2858. unlock_page(page);
  2859. goto unwritable_page;
  2860. }
  2861. } else
  2862. VM_BUG_ON(!PageLocked(page));
  2863. page_mkwrite = 1;
  2864. }
  2865. }
  2866. }
  2867. page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
  2868. /*
  2869. * This silly early PAGE_DIRTY setting removes a race
  2870. * due to the bad i386 page protection. But it's valid
  2871. * for other architectures too.
  2872. *
  2873. * Note that if FAULT_FLAG_WRITE is set, we either now have
  2874. * an exclusive copy of the page, or this is a shared mapping,
  2875. * so we can make it writable and dirty to avoid having to
  2876. * handle that later.
  2877. */
  2878. /* Only go through if we didn't race with anybody else... */
  2879. if (likely(pte_same(*page_table, orig_pte))) {
  2880. flush_icache_page(vma, page);
  2881. entry = mk_pte(page, vma->vm_page_prot);
  2882. if (flags & FAULT_FLAG_WRITE)
  2883. entry = maybe_mkwrite(pte_mkdirty(entry), vma);
  2884. if (anon) {
  2885. inc_mm_counter_fast(mm, MM_ANONPAGES);
  2886. page_add_new_anon_rmap(page, vma, address);
  2887. } else {
  2888. inc_mm_counter_fast(mm, MM_FILEPAGES);
  2889. page_add_file_rmap(page);
  2890. if (flags & FAULT_FLAG_WRITE) {
  2891. dirty_page = page;
  2892. get_page(dirty_page);
  2893. }
  2894. }
  2895. set_pte_at(mm, address, page_table, entry);
  2896. /* no need to invalidate: a not-present page won't be cached */
  2897. update_mmu_cache(vma, address, page_table);
  2898. } else {
  2899. if (charged)
  2900. mem_cgroup_uncharge_page(page);
  2901. if (anon)
  2902. page_cache_release(page);
  2903. else
  2904. anon = 1; /* no anon but release faulted_page */
  2905. }
  2906. pte_unmap_unlock(page_table, ptl);
  2907. out:
  2908. if (dirty_page) {
  2909. struct address_space *mapping = page->mapping;
  2910. if (set_page_dirty(dirty_page))
  2911. page_mkwrite = 1;
  2912. unlock_page(dirty_page);
  2913. put_page(dirty_page);
  2914. if (page_mkwrite && mapping) {
  2915. /*
  2916. * Some device drivers do not set page.mapping but still
  2917. * dirty their pages
  2918. */
  2919. balance_dirty_pages_ratelimited(mapping);
  2920. }
  2921. /* file_update_time outside page_lock */
  2922. if (vma->vm_file)
  2923. file_update_time(vma->vm_file);
  2924. } else {
  2925. unlock_page(vmf.page);
  2926. if (anon)
  2927. page_cache_release(vmf.page);
  2928. }
  2929. return ret;
  2930. unwritable_page:
  2931. page_cache_release(page);
  2932. return ret;
  2933. }
  2934. static int do_linear_fault(struct mm_struct *mm, struct vm_area_struct *vma,
  2935. unsigned long address, pte_t *page_table, pmd_t *pmd,
  2936. unsigned int flags, pte_t orig_pte)
  2937. {
  2938. pgoff_t pgoff = (((address & PAGE_MASK)
  2939. - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
  2940. pte_unmap(page_table);
  2941. return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte);
  2942. }
  2943. /*
  2944. * Fault of a previously existing named mapping. Repopulate the pte
  2945. * from the encoded file_pte if possible. This enables swappable
  2946. * nonlinear vmas.
  2947. *
  2948. * We enter with non-exclusive mmap_sem (to exclude vma changes,
  2949. * but allow concurrent faults), and pte mapped but not yet locked.
  2950. * We return with mmap_sem still held, but pte unmapped and unlocked.
  2951. */
  2952. static int do_nonlinear_fault(struct mm_struct *mm, struct vm_area_struct *vma,
  2953. unsigned long address, pte_t *page_table, pmd_t *pmd,
  2954. unsigned int flags, pte_t orig_pte)
  2955. {
  2956. pgoff_t pgoff;
  2957. flags |= FAULT_FLAG_NONLINEAR;
  2958. if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
  2959. return 0;
  2960. if (unlikely(!(vma->vm_flags & VM_NONLINEAR))) {
  2961. /*
  2962. * Page table corrupted: show pte and kill process.
  2963. */
  2964. print_bad_pte(vma, address, orig_pte, NULL);
  2965. return VM_FAULT_SIGBUS;
  2966. }
  2967. pgoff = pte_to_pgoff(orig_pte);
  2968. return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte);
  2969. }
  2970. /*
  2971. * These routines also need to handle stuff like marking pages dirty
  2972. * and/or accessed for architectures that don't do it in hardware (most
  2973. * RISC architectures). The early dirtying is also good on the i386.
  2974. *
  2975. * There is also a hook called "update_mmu_cache()" that architectures
  2976. * with external mmu caches can use to update those (ie the Sparc or
  2977. * PowerPC hashed page tables that act as extended TLBs).
  2978. *
  2979. * We enter with non-exclusive mmap_sem (to exclude vma changes,
  2980. * but allow concurrent faults), and pte mapped but not yet locked.
  2981. * We return with mmap_sem still held, but pte unmapped and unlocked.
  2982. */
  2983. int handle_pte_fault(struct mm_struct *mm,
  2984. struct vm_area_struct *vma, unsigned long address,
  2985. pte_t *pte, pmd_t *pmd, unsigned int flags)
  2986. {
  2987. pte_t entry;
  2988. spinlock_t *ptl;
  2989. entry = *pte;
  2990. if (!pte_present(entry)) {
  2991. if (pte_none(entry)) {
  2992. if (vma->vm_ops) {
  2993. if (likely(vma->vm_ops->fault))
  2994. return do_linear_fault(mm, vma, address,
  2995. pte, pmd, flags, entry);
  2996. }
  2997. return do_anonymous_page(mm, vma, address,
  2998. pte, pmd, flags);
  2999. }
  3000. if (pte_file(entry))
  3001. return do_nonlinear_fault(mm, vma, address,
  3002. pte, pmd, flags, entry);
  3003. return do_swap_page(mm, vma, address,
  3004. pte, pmd, flags, entry);
  3005. }
  3006. ptl = pte_lockptr(mm, pmd);
  3007. spin_lock(ptl);
  3008. if (unlikely(!pte_same(*pte, entry)))
  3009. goto unlock;
  3010. if (flags & FAULT_FLAG_WRITE) {
  3011. if (!pte_write(entry))
  3012. return do_wp_page(mm, vma, address,
  3013. pte, pmd, ptl, entry);
  3014. entry = pte_mkdirty(entry);
  3015. }
  3016. entry = pte_mkyoung(entry);
  3017. if (ptep_set_access_flags(vma, address, pte, entry, flags & FAULT_FLAG_WRITE)) {
  3018. update_mmu_cache(vma, address, pte);
  3019. } else {
  3020. /*
  3021. * This is needed only for protection faults but the arch code
  3022. * is not yet telling us if this is a protection fault or not.
  3023. * This still avoids useless tlb flushes for .text page faults
  3024. * with threads.
  3025. */
  3026. if (flags & FAULT_FLAG_WRITE)
  3027. flush_tlb_fix_spurious_fault(vma, address);
  3028. }
  3029. unlock:
  3030. pte_unmap_unlock(pte, ptl);
  3031. return 0;
  3032. }
  3033. /*
  3034. * By the time we get here, we already hold the mm semaphore
  3035. */
  3036. int handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma,
  3037. unsigned long address, unsigned int flags)
  3038. {
  3039. pgd_t *pgd;
  3040. pud_t *pud;
  3041. pmd_t *pmd;
  3042. pte_t *pte;
  3043. __set_current_state(TASK_RUNNING);
  3044. count_vm_event(PGFAULT);
  3045. mem_cgroup_count_vm_event(mm, PGFAULT);
  3046. /* do counter updates before entering really critical section. */
  3047. check_sync_rss_stat(current);
  3048. if (unlikely(is_vm_hugetlb_page(vma)))
  3049. return hugetlb_fault(mm, vma, address, flags);
  3050. pgd = pgd_offset(mm, address);
  3051. pud = pud_alloc(mm, pgd, address);
  3052. if (!pud)
  3053. return VM_FAULT_OOM;
  3054. pmd = pmd_alloc(mm, pud, address);
  3055. if (!pmd)
  3056. return VM_FAULT_OOM;
  3057. if (pmd_none(*pmd) && transparent_hugepage_enabled(vma)) {
  3058. if (!vma->vm_ops)
  3059. return do_huge_pmd_anonymous_page(mm, vma, address,
  3060. pmd, flags);
  3061. } else {
  3062. pmd_t orig_pmd = *pmd;
  3063. barrier();
  3064. if (pmd_trans_huge(orig_pmd)) {
  3065. if (flags & FAULT_FLAG_WRITE &&
  3066. !pmd_write(orig_pmd) &&
  3067. !pmd_trans_splitting(orig_pmd))
  3068. return do_huge_pmd_wp_page(mm, vma, address,
  3069. pmd, orig_pmd);
  3070. return 0;
  3071. }
  3072. }
  3073. /*
  3074. * Use __pte_alloc instead of pte_alloc_map, because we can't
  3075. * run pte_offset_map on the pmd, if an huge pmd could
  3076. * materialize from under us from a different thread.
  3077. */
  3078. if (unlikely(pmd_none(*pmd)) && __pte_alloc(mm, vma, pmd, address))
  3079. return VM_FAULT_OOM;
  3080. /* if an huge pmd materialized from under us just retry later */
  3081. if (unlikely(pmd_trans_huge(*pmd)))
  3082. return 0;
  3083. /*
  3084. * A regular pmd is established and it can't morph into a huge pmd
  3085. * from under us anymore at this point because we hold the mmap_sem
  3086. * read mode and khugepaged takes it in write mode. So now it's
  3087. * safe to run pte_offset_map().
  3088. */
  3089. pte = pte_offset_map(pmd, address);
  3090. return handle_pte_fault(mm, vma, address, pte, pmd, flags);
  3091. }
  3092. #ifndef __PAGETABLE_PUD_FOLDED
  3093. /*
  3094. * Allocate page upper directory.
  3095. * We've already handled the fast-path in-line.
  3096. */
  3097. int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address)
  3098. {
  3099. pud_t *new = pud_alloc_one(mm, address);
  3100. if (!new)
  3101. return -ENOMEM;
  3102. smp_wmb(); /* See comment in __pte_alloc */
  3103. spin_lock(&mm->page_table_lock);
  3104. if (pgd_present(*pgd)) /* Another has populated it */
  3105. pud_free(mm, new);
  3106. else
  3107. pgd_populate(mm, pgd, new);
  3108. spin_unlock(&mm->page_table_lock);
  3109. return 0;
  3110. }
  3111. #endif /* __PAGETABLE_PUD_FOLDED */
  3112. #ifndef __PAGETABLE_PMD_FOLDED
  3113. /*
  3114. * Allocate page middle directory.
  3115. * We've already handled the fast-path in-line.
  3116. */
  3117. int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address)
  3118. {
  3119. pmd_t *new = pmd_alloc_one(mm, address);
  3120. if (!new)
  3121. return -ENOMEM;
  3122. smp_wmb(); /* See comment in __pte_alloc */
  3123. spin_lock(&mm->page_table_lock);
  3124. #ifndef __ARCH_HAS_4LEVEL_HACK
  3125. if (pud_present(*pud)) /* Another has populated it */
  3126. pmd_free(mm, new);
  3127. else
  3128. pud_populate(mm, pud, new);
  3129. #else
  3130. if (pgd_present(*pud)) /* Another has populated it */
  3131. pmd_free(mm, new);
  3132. else
  3133. pgd_populate(mm, pud, new);
  3134. #endif /* __ARCH_HAS_4LEVEL_HACK */
  3135. spin_unlock(&mm->page_table_lock);
  3136. return 0;
  3137. }
  3138. #endif /* __PAGETABLE_PMD_FOLDED */
  3139. int make_pages_present(unsigned long addr, unsigned long end)
  3140. {
  3141. int ret, len, write;
  3142. struct vm_area_struct * vma;
  3143. vma = find_vma(current->mm, addr);
  3144. if (!vma)
  3145. return -ENOMEM;
  3146. /*
  3147. * We want to touch writable mappings with a write fault in order
  3148. * to break COW, except for shared mappings because these don't COW
  3149. * and we would not want to dirty them for nothing.
  3150. */
  3151. write = (vma->vm_flags & (VM_WRITE | VM_SHARED)) == VM_WRITE;
  3152. BUG_ON(addr >= end);
  3153. BUG_ON(end > vma->vm_end);
  3154. len = DIV_ROUND_UP(end, PAGE_SIZE) - addr/PAGE_SIZE;
  3155. ret = get_user_pages(current, current->mm, addr,
  3156. len, write, 0, NULL, NULL);
  3157. if (ret < 0)
  3158. return ret;
  3159. return ret == len ? 0 : -EFAULT;
  3160. }
  3161. #if !defined(__HAVE_ARCH_GATE_AREA)
  3162. #if defined(AT_SYSINFO_EHDR)
  3163. static struct vm_area_struct gate_vma;
  3164. static int __init gate_vma_init(void)
  3165. {
  3166. gate_vma.vm_mm = NULL;
  3167. gate_vma.vm_start = FIXADDR_USER_START;
  3168. gate_vma.vm_end = FIXADDR_USER_END;
  3169. gate_vma.vm_flags = VM_READ | VM_MAYREAD | VM_EXEC | VM_MAYEXEC;
  3170. gate_vma.vm_page_prot = __P101;
  3171. /*
  3172. * Make sure the vDSO gets into every core dump.
  3173. * Dumping its contents makes post-mortem fully interpretable later
  3174. * without matching up the same kernel and hardware config to see
  3175. * what PC values meant.
  3176. */
  3177. gate_vma.vm_flags |= VM_ALWAYSDUMP;
  3178. return 0;
  3179. }
  3180. __initcall(gate_vma_init);
  3181. #endif
  3182. struct vm_area_struct *get_gate_vma(struct mm_struct *mm)
  3183. {
  3184. #ifdef AT_SYSINFO_EHDR
  3185. return &gate_vma;
  3186. #else
  3187. return NULL;
  3188. #endif
  3189. }
  3190. int in_gate_area_no_mm(unsigned long addr)
  3191. {
  3192. #ifdef AT_SYSINFO_EHDR
  3193. if ((addr >= FIXADDR_USER_START) && (addr < FIXADDR_USER_END))
  3194. return 1;
  3195. #endif
  3196. return 0;
  3197. }
  3198. #endif /* __HAVE_ARCH_GATE_AREA */
  3199. static int __follow_pte(struct mm_struct *mm, unsigned long address,
  3200. pte_t **ptepp, spinlock_t **ptlp)
  3201. {
  3202. pgd_t *pgd;
  3203. pud_t *pud;
  3204. pmd_t *pmd;
  3205. pte_t *ptep;
  3206. pgd = pgd_offset(mm, address);
  3207. if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
  3208. goto out;
  3209. pud = pud_offset(pgd, address);
  3210. if (pud_none(*pud) || unlikely(pud_bad(*pud)))
  3211. goto out;
  3212. pmd = pmd_offset(pud, address);
  3213. VM_BUG_ON(pmd_trans_huge(*pmd));
  3214. if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd)))
  3215. goto out;
  3216. /* We cannot handle huge page PFN maps. Luckily they don't exist. */
  3217. if (pmd_huge(*pmd))
  3218. goto out;
  3219. ptep = pte_offset_map_lock(mm, pmd, address, ptlp);
  3220. if (!ptep)
  3221. goto out;
  3222. if (!pte_present(*ptep))
  3223. goto unlock;
  3224. *ptepp = ptep;
  3225. return 0;
  3226. unlock:
  3227. pte_unmap_unlock(ptep, *ptlp);
  3228. out:
  3229. return -EINVAL;
  3230. }
  3231. static inline int follow_pte(struct mm_struct *mm, unsigned long address,
  3232. pte_t **ptepp, spinlock_t **ptlp)
  3233. {
  3234. int res;
  3235. /* (void) is needed to make gcc happy */
  3236. (void) __cond_lock(*ptlp,
  3237. !(res = __follow_pte(mm, address, ptepp, ptlp)));
  3238. return res;
  3239. }
  3240. /**
  3241. * follow_pfn - look up PFN at a user virtual address
  3242. * @vma: memory mapping
  3243. * @address: user virtual address
  3244. * @pfn: location to store found PFN
  3245. *
  3246. * Only IO mappings and raw PFN mappings are allowed.
  3247. *
  3248. * Returns zero and the pfn at @pfn on success, -ve otherwise.
  3249. */
  3250. int follow_pfn(struct vm_area_struct *vma, unsigned long address,
  3251. unsigned long *pfn)
  3252. {
  3253. int ret = -EINVAL;
  3254. spinlock_t *ptl;
  3255. pte_t *ptep;
  3256. if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
  3257. return ret;
  3258. ret = follow_pte(vma->vm_mm, address, &ptep, &ptl);
  3259. if (ret)
  3260. return ret;
  3261. *pfn = pte_pfn(*ptep);
  3262. pte_unmap_unlock(ptep, ptl);
  3263. return 0;
  3264. }
  3265. EXPORT_SYMBOL(follow_pfn);
  3266. #ifdef CONFIG_HAVE_IOREMAP_PROT
  3267. int follow_phys(struct vm_area_struct *vma,
  3268. unsigned long address, unsigned int flags,
  3269. unsigned long *prot, resource_size_t *phys)
  3270. {
  3271. int ret = -EINVAL;
  3272. pte_t *ptep, pte;
  3273. spinlock_t *ptl;
  3274. if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
  3275. goto out;
  3276. if (follow_pte(vma->vm_mm, address, &ptep, &ptl))
  3277. goto out;
  3278. pte = *ptep;
  3279. if ((flags & FOLL_WRITE) && !pte_write(pte))
  3280. goto unlock;
  3281. *prot = pgprot_val(pte_pgprot(pte));
  3282. *phys = (resource_size_t)pte_pfn(pte) << PAGE_SHIFT;
  3283. ret = 0;
  3284. unlock:
  3285. pte_unmap_unlock(ptep, ptl);
  3286. out:
  3287. return ret;
  3288. }
  3289. int generic_access_phys(struct vm_area_struct *vma, unsigned long addr,
  3290. void *buf, int len, int write)
  3291. {
  3292. resource_size_t phys_addr;
  3293. unsigned long prot = 0;
  3294. void __iomem *maddr;
  3295. int offset = addr & (PAGE_SIZE-1);
  3296. if (follow_phys(vma, addr, write, &prot, &phys_addr))
  3297. return -EINVAL;
  3298. maddr = ioremap_prot(phys_addr, PAGE_SIZE, prot);
  3299. if (write)
  3300. memcpy_toio(maddr + offset, buf, len);
  3301. else
  3302. memcpy_fromio(buf, maddr + offset, len);
  3303. iounmap(maddr);
  3304. return len;
  3305. }
  3306. #endif
  3307. /*
  3308. * Access another process' address space as given in mm. If non-NULL, use the
  3309. * given task for page fault accounting.
  3310. */
  3311. static int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm,
  3312. unsigned long addr, void *buf, int len, int write)
  3313. {
  3314. struct vm_area_struct *vma;
  3315. void *old_buf = buf;
  3316. down_read(&mm->mmap_sem);
  3317. /* ignore errors, just check how much was successfully transferred */
  3318. while (len) {
  3319. int bytes, ret, offset;
  3320. void *maddr;
  3321. struct page *page = NULL;
  3322. ret = get_user_pages(tsk, mm, addr, 1,
  3323. write, 1, &page, &vma);
  3324. if (ret <= 0) {
  3325. /*
  3326. * Check if this is a VM_IO | VM_PFNMAP VMA, which
  3327. * we can access using slightly different code.
  3328. */
  3329. #ifdef CONFIG_HAVE_IOREMAP_PROT
  3330. vma = find_vma(mm, addr);
  3331. if (!vma || vma->vm_start > addr)
  3332. break;
  3333. if (vma->vm_ops && vma->vm_ops->access)
  3334. ret = vma->vm_ops->access(vma, addr, buf,
  3335. len, write);
  3336. if (ret <= 0)
  3337. #endif
  3338. break;
  3339. bytes = ret;
  3340. } else {
  3341. bytes = len;
  3342. offset = addr & (PAGE_SIZE-1);
  3343. if (bytes > PAGE_SIZE-offset)
  3344. bytes = PAGE_SIZE-offset;
  3345. maddr = kmap(page);
  3346. if (write) {
  3347. copy_to_user_page(vma, page, addr,
  3348. maddr + offset, buf, bytes);
  3349. set_page_dirty_lock(page);
  3350. } else {
  3351. copy_from_user_page(vma, page, addr,
  3352. buf, maddr + offset, bytes);
  3353. }
  3354. kunmap(page);
  3355. page_cache_release(page);
  3356. }
  3357. len -= bytes;
  3358. buf += bytes;
  3359. addr += bytes;
  3360. }
  3361. up_read(&mm->mmap_sem);
  3362. return buf - old_buf;
  3363. }
  3364. /**
  3365. * access_remote_vm - access another process' address space
  3366. * @mm: the mm_struct of the target address space
  3367. * @addr: start address to access
  3368. * @buf: source or destination buffer
  3369. * @len: number of bytes to transfer
  3370. * @write: whether the access is a write
  3371. *
  3372. * The caller must hold a reference on @mm.
  3373. */
  3374. int access_remote_vm(struct mm_struct *mm, unsigned long addr,
  3375. void *buf, int len, int write)
  3376. {
  3377. return __access_remote_vm(NULL, mm, addr, buf, len, write);
  3378. }
  3379. /*
  3380. * Access another process' address space.
  3381. * Source/target buffer must be kernel space,
  3382. * Do not walk the page table directly, use get_user_pages
  3383. */
  3384. int access_process_vm(struct task_struct *tsk, unsigned long addr,
  3385. void *buf, int len, int write)
  3386. {
  3387. struct mm_struct *mm;
  3388. int ret;
  3389. mm = get_task_mm(tsk);
  3390. if (!mm)
  3391. return 0;
  3392. ret = __access_remote_vm(tsk, mm, addr, buf, len, write);
  3393. mmput(mm);
  3394. return ret;
  3395. }
  3396. /*
  3397. * Print the name of a VMA.
  3398. */
  3399. void print_vma_addr(char *prefix, unsigned long ip)
  3400. {
  3401. struct mm_struct *mm = current->mm;
  3402. struct vm_area_struct *vma;
  3403. /*
  3404. * Do not print if we are in atomic
  3405. * contexts (in exception stacks, etc.):
  3406. */
  3407. if (preempt_count())
  3408. return;
  3409. down_read(&mm->mmap_sem);
  3410. vma = find_vma(mm, ip);
  3411. if (vma && vma->vm_file) {
  3412. struct file *f = vma->vm_file;
  3413. char *buf = (char *)__get_free_page(GFP_KERNEL);
  3414. if (buf) {
  3415. char *p, *s;
  3416. p = d_path(&f->f_path, buf, PAGE_SIZE);
  3417. if (IS_ERR(p))
  3418. p = "?";
  3419. s = strrchr(p, '/');
  3420. if (s)
  3421. p = s+1;
  3422. printk("%s%s[%lx+%lx]", prefix, p,
  3423. vma->vm_start,
  3424. vma->vm_end - vma->vm_start);
  3425. free_page((unsigned long)buf);
  3426. }
  3427. }
  3428. up_read(&current->mm->mmap_sem);
  3429. }
  3430. #ifdef CONFIG_PROVE_LOCKING
  3431. void might_fault(void)
  3432. {
  3433. /*
  3434. * Some code (nfs/sunrpc) uses socket ops on kernel memory while
  3435. * holding the mmap_sem, this is safe because kernel memory doesn't
  3436. * get paged out, therefore we'll never actually fault, and the
  3437. * below annotations will generate false positives.
  3438. */
  3439. if (segment_eq(get_fs(), KERNEL_DS))
  3440. return;
  3441. might_sleep();
  3442. /*
  3443. * it would be nicer only to annotate paths which are not under
  3444. * pagefault_disable, however that requires a larger audit and
  3445. * providing helpers like get_user_atomic.
  3446. */
  3447. if (!in_atomic() && current->mm)
  3448. might_lock_read(&current->mm->mmap_sem);
  3449. }
  3450. EXPORT_SYMBOL(might_fault);
  3451. #endif
  3452. #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLBFS)
  3453. static void clear_gigantic_page(struct page *page,
  3454. unsigned long addr,
  3455. unsigned int pages_per_huge_page)
  3456. {
  3457. int i;
  3458. struct page *p = page;
  3459. might_sleep();
  3460. for (i = 0; i < pages_per_huge_page;
  3461. i++, p = mem_map_next(p, page, i)) {
  3462. cond_resched();
  3463. clear_user_highpage(p, addr + i * PAGE_SIZE);
  3464. }
  3465. }
  3466. void clear_huge_page(struct page *page,
  3467. unsigned long addr, unsigned int pages_per_huge_page)
  3468. {
  3469. int i;
  3470. if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) {
  3471. clear_gigantic_page(page, addr, pages_per_huge_page);
  3472. return;
  3473. }
  3474. might_sleep();
  3475. for (i = 0; i < pages_per_huge_page; i++) {
  3476. cond_resched();
  3477. clear_user_highpage(page + i, addr + i * PAGE_SIZE);
  3478. }
  3479. }
  3480. static void copy_user_gigantic_page(struct page *dst, struct page *src,
  3481. unsigned long addr,
  3482. struct vm_area_struct *vma,
  3483. unsigned int pages_per_huge_page)
  3484. {
  3485. int i;
  3486. struct page *dst_base = dst;
  3487. struct page *src_base = src;
  3488. for (i = 0; i < pages_per_huge_page; ) {
  3489. cond_resched();
  3490. copy_user_highpage(dst, src, addr + i*PAGE_SIZE, vma);
  3491. i++;
  3492. dst = mem_map_next(dst, dst_base, i);
  3493. src = mem_map_next(src, src_base, i);
  3494. }
  3495. }
  3496. void copy_user_huge_page(struct page *dst, struct page *src,
  3497. unsigned long addr, struct vm_area_struct *vma,
  3498. unsigned int pages_per_huge_page)
  3499. {
  3500. int i;
  3501. if (unlikely(pages_per_huge_page > MAX_ORDER_NR_PAGES)) {
  3502. copy_user_gigantic_page(dst, src, addr, vma,
  3503. pages_per_huge_page);
  3504. return;
  3505. }
  3506. might_sleep();
  3507. for (i = 0; i < pages_per_huge_page; i++) {
  3508. cond_resched();
  3509. copy_user_highpage(dst + i, src + i, addr + i*PAGE_SIZE, vma);
  3510. }
  3511. }
  3512. #endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_HUGETLBFS */