volumes.c 86 KB

1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162636465666768697071727374757677787980818283848586878889909192939495969798991001011021031041051061071081091101111121131141151161171181191201211221231241251261271281291301311321331341351361371381391401411421431441451461471481491501511521531541551561571581591601611621631641651661671681691701711721731741751761771781791801811821831841851861871881891901911921931941951961971981992002012022032042052062072082092102112122132142152162172182192202212222232242252262272282292302312322332342352362372382392402412422432442452462472482492502512522532542552562572582592602612622632642652662672682692702712722732742752762772782792802812822832842852862872882892902912922932942952962972982993003013023033043053063073083093103113123133143153163173183193203213223233243253263273283293303313323333343353363373383393403413423433443453463473483493503513523533543553563573583593603613623633643653663673683693703713723733743753763773783793803813823833843853863873883893903913923933943953963973983994004014024034044054064074084094104114124134144154164174184194204214224234244254264274284294304314324334344354364374384394404414424434444454464474484494504514524534544554564574584594604614624634644654664674684694704714724734744754764774784794804814824834844854864874884894904914924934944954964974984995005015025035045055065075085095105115125135145155165175185195205215225235245255265275285295305315325335345355365375385395405415425435445455465475485495505515525535545555565575585595605615625635645655665675685695705715725735745755765775785795805815825835845855865875885895905915925935945955965975985996006016026036046056066076086096106116126136146156166176186196206216226236246256266276286296306316326336346356366376386396406416426436446456466476486496506516526536546556566576586596606616626636646656666676686696706716726736746756766776786796806816826836846856866876886896906916926936946956966976986997007017027037047057067077087097107117127137147157167177187197207217227237247257267277287297307317327337347357367377387397407417427437447457467477487497507517527537547557567577587597607617627637647657667677687697707717727737747757767777787797807817827837847857867877887897907917927937947957967977987998008018028038048058068078088098108118128138148158168178188198208218228238248258268278288298308318328338348358368378388398408418428438448458468478488498508518528538548558568578588598608618628638648658668678688698708718728738748758768778788798808818828838848858868878888898908918928938948958968978988999009019029039049059069079089099109119129139149159169179189199209219229239249259269279289299309319329339349359369379389399409419429439449459469479489499509519529539549559569579589599609619629639649659669679689699709719729739749759769779789799809819829839849859869879889899909919929939949959969979989991000100110021003100410051006100710081009101010111012101310141015101610171018101910201021102210231024102510261027102810291030103110321033103410351036103710381039104010411042104310441045104610471048104910501051105210531054105510561057105810591060106110621063106410651066106710681069107010711072107310741075107610771078107910801081108210831084108510861087108810891090109110921093109410951096109710981099110011011102110311041105110611071108110911101111111211131114111511161117111811191120112111221123112411251126112711281129113011311132113311341135113611371138113911401141114211431144114511461147114811491150115111521153115411551156115711581159116011611162116311641165116611671168116911701171117211731174117511761177117811791180118111821183118411851186118711881189119011911192119311941195119611971198119912001201120212031204120512061207120812091210121112121213121412151216121712181219122012211222122312241225122612271228122912301231123212331234123512361237123812391240124112421243124412451246124712481249125012511252125312541255125612571258125912601261126212631264126512661267126812691270127112721273127412751276127712781279128012811282128312841285128612871288128912901291129212931294129512961297129812991300130113021303130413051306130713081309131013111312131313141315131613171318131913201321132213231324132513261327132813291330133113321333133413351336133713381339134013411342134313441345134613471348134913501351135213531354135513561357135813591360136113621363136413651366136713681369137013711372137313741375137613771378137913801381138213831384138513861387138813891390139113921393139413951396139713981399140014011402140314041405140614071408140914101411141214131414141514161417141814191420142114221423142414251426142714281429143014311432143314341435143614371438143914401441144214431444144514461447144814491450145114521453145414551456145714581459146014611462146314641465146614671468146914701471147214731474147514761477147814791480148114821483148414851486148714881489149014911492149314941495149614971498149915001501150215031504150515061507150815091510151115121513151415151516151715181519152015211522152315241525152615271528152915301531153215331534153515361537153815391540154115421543154415451546154715481549155015511552155315541555155615571558155915601561156215631564156515661567156815691570157115721573157415751576157715781579158015811582158315841585158615871588158915901591159215931594159515961597159815991600160116021603160416051606160716081609161016111612161316141615161616171618161916201621162216231624162516261627162816291630163116321633163416351636163716381639164016411642164316441645164616471648164916501651165216531654165516561657165816591660166116621663166416651666166716681669167016711672167316741675167616771678167916801681168216831684168516861687168816891690169116921693169416951696169716981699170017011702170317041705170617071708170917101711171217131714171517161717171817191720172117221723172417251726172717281729173017311732173317341735173617371738173917401741174217431744174517461747174817491750175117521753175417551756175717581759176017611762176317641765176617671768176917701771177217731774177517761777177817791780178117821783178417851786178717881789179017911792179317941795179617971798179918001801180218031804180518061807180818091810181118121813181418151816181718181819182018211822182318241825182618271828182918301831183218331834183518361837183818391840184118421843184418451846184718481849185018511852185318541855185618571858185918601861186218631864186518661867186818691870187118721873187418751876187718781879188018811882188318841885188618871888188918901891189218931894189518961897189818991900190119021903190419051906190719081909191019111912191319141915191619171918191919201921192219231924192519261927192819291930193119321933193419351936193719381939194019411942194319441945194619471948194919501951195219531954195519561957195819591960196119621963196419651966196719681969197019711972197319741975197619771978197919801981198219831984198519861987198819891990199119921993199419951996199719981999200020012002200320042005200620072008200920102011201220132014201520162017201820192020202120222023202420252026202720282029203020312032203320342035203620372038203920402041204220432044204520462047204820492050205120522053205420552056205720582059206020612062206320642065206620672068206920702071207220732074207520762077207820792080208120822083208420852086208720882089209020912092209320942095209620972098209921002101210221032104210521062107210821092110211121122113211421152116211721182119212021212122212321242125212621272128212921302131213221332134213521362137213821392140214121422143214421452146214721482149215021512152215321542155215621572158215921602161216221632164216521662167216821692170217121722173217421752176217721782179218021812182218321842185218621872188218921902191219221932194219521962197219821992200220122022203220422052206220722082209221022112212221322142215221622172218221922202221222222232224222522262227222822292230223122322233223422352236223722382239224022412242224322442245224622472248224922502251225222532254225522562257225822592260226122622263226422652266226722682269227022712272227322742275227622772278227922802281228222832284228522862287228822892290229122922293229422952296229722982299230023012302230323042305230623072308230923102311231223132314231523162317231823192320232123222323232423252326232723282329233023312332233323342335233623372338233923402341234223432344234523462347234823492350235123522353235423552356235723582359236023612362236323642365236623672368236923702371237223732374237523762377237823792380238123822383238423852386238723882389239023912392239323942395239623972398239924002401240224032404240524062407240824092410241124122413241424152416241724182419242024212422242324242425242624272428242924302431243224332434243524362437243824392440244124422443244424452446244724482449245024512452245324542455245624572458245924602461246224632464246524662467246824692470247124722473247424752476247724782479248024812482248324842485248624872488248924902491249224932494249524962497249824992500250125022503250425052506250725082509251025112512251325142515251625172518251925202521252225232524252525262527252825292530253125322533253425352536253725382539254025412542254325442545254625472548254925502551255225532554255525562557255825592560256125622563256425652566256725682569257025712572257325742575257625772578257925802581258225832584258525862587258825892590259125922593259425952596259725982599260026012602260326042605260626072608260926102611261226132614261526162617261826192620262126222623262426252626262726282629263026312632263326342635263626372638263926402641264226432644264526462647264826492650265126522653265426552656265726582659266026612662266326642665266626672668266926702671267226732674267526762677267826792680268126822683268426852686268726882689269026912692269326942695269626972698269927002701270227032704270527062707270827092710271127122713271427152716271727182719272027212722272327242725272627272728272927302731273227332734273527362737273827392740274127422743274427452746274727482749275027512752275327542755275627572758275927602761276227632764276527662767276827692770277127722773277427752776277727782779278027812782278327842785278627872788278927902791279227932794279527962797279827992800280128022803280428052806280728082809281028112812281328142815281628172818281928202821282228232824282528262827282828292830283128322833283428352836283728382839284028412842284328442845284628472848284928502851285228532854285528562857285828592860286128622863286428652866286728682869287028712872287328742875287628772878287928802881288228832884288528862887288828892890289128922893289428952896289728982899290029012902290329042905290629072908290929102911291229132914291529162917291829192920292129222923292429252926292729282929293029312932293329342935293629372938293929402941294229432944294529462947294829492950295129522953295429552956295729582959296029612962296329642965296629672968296929702971297229732974297529762977297829792980298129822983298429852986298729882989299029912992299329942995299629972998299930003001300230033004300530063007300830093010301130123013301430153016301730183019302030213022302330243025302630273028302930303031303230333034303530363037303830393040304130423043304430453046304730483049305030513052305330543055305630573058305930603061306230633064306530663067306830693070307130723073307430753076307730783079308030813082308330843085308630873088308930903091309230933094309530963097309830993100310131023103310431053106310731083109311031113112311331143115311631173118311931203121312231233124312531263127312831293130313131323133313431353136313731383139314031413142314331443145314631473148314931503151315231533154315531563157315831593160316131623163316431653166316731683169317031713172317331743175317631773178317931803181318231833184318531863187318831893190319131923193319431953196319731983199320032013202320332043205320632073208320932103211321232133214321532163217321832193220322132223223322432253226322732283229323032313232323332343235323632373238323932403241324232433244324532463247324832493250325132523253325432553256325732583259326032613262326332643265326632673268326932703271327232733274327532763277327832793280328132823283328432853286328732883289329032913292329332943295329632973298329933003301330233033304330533063307330833093310331133123313331433153316331733183319332033213322332333243325332633273328332933303331333233333334333533363337333833393340334133423343334433453346334733483349335033513352335333543355335633573358335933603361336233633364336533663367336833693370337133723373337433753376337733783379338033813382338333843385338633873388338933903391339233933394339533963397339833993400340134023403340434053406340734083409341034113412341334143415341634173418341934203421342234233424342534263427342834293430343134323433343434353436
  1. /*
  2. * Copyright (C) 2007 Oracle. All rights reserved.
  3. *
  4. * This program is free software; you can redistribute it and/or
  5. * modify it under the terms of the GNU General Public
  6. * License v2 as published by the Free Software Foundation.
  7. *
  8. * This program is distributed in the hope that it will be useful,
  9. * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10. * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  11. * General Public License for more details.
  12. *
  13. * You should have received a copy of the GNU General Public
  14. * License along with this program; if not, write to the
  15. * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16. * Boston, MA 021110-1307, USA.
  17. */
  18. #include <linux/sched.h>
  19. #include <linux/bio.h>
  20. #include <linux/slab.h>
  21. #include <linux/buffer_head.h>
  22. #include <linux/blkdev.h>
  23. #include <linux/random.h>
  24. #include <linux/iocontext.h>
  25. #include <asm/div64.h>
  26. #include "compat.h"
  27. #include "ctree.h"
  28. #include "extent_map.h"
  29. #include "disk-io.h"
  30. #include "transaction.h"
  31. #include "print-tree.h"
  32. #include "volumes.h"
  33. #include "async-thread.h"
  34. struct map_lookup {
  35. u64 type;
  36. int io_align;
  37. int io_width;
  38. int stripe_len;
  39. int sector_size;
  40. int num_stripes;
  41. int sub_stripes;
  42. struct btrfs_bio_stripe stripes[];
  43. };
  44. static int init_first_rw_device(struct btrfs_trans_handle *trans,
  45. struct btrfs_root *root,
  46. struct btrfs_device *device);
  47. static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
  48. #define map_lookup_size(n) (sizeof(struct map_lookup) + \
  49. (sizeof(struct btrfs_bio_stripe) * (n)))
  50. static DEFINE_MUTEX(uuid_mutex);
  51. static LIST_HEAD(fs_uuids);
  52. void btrfs_lock_volumes(void)
  53. {
  54. mutex_lock(&uuid_mutex);
  55. }
  56. void btrfs_unlock_volumes(void)
  57. {
  58. mutex_unlock(&uuid_mutex);
  59. }
  60. static void lock_chunks(struct btrfs_root *root)
  61. {
  62. mutex_lock(&root->fs_info->chunk_mutex);
  63. }
  64. static void unlock_chunks(struct btrfs_root *root)
  65. {
  66. mutex_unlock(&root->fs_info->chunk_mutex);
  67. }
  68. static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
  69. {
  70. struct btrfs_device *device;
  71. WARN_ON(fs_devices->opened);
  72. while (!list_empty(&fs_devices->devices)) {
  73. device = list_entry(fs_devices->devices.next,
  74. struct btrfs_device, dev_list);
  75. list_del(&device->dev_list);
  76. kfree(device->name);
  77. kfree(device);
  78. }
  79. kfree(fs_devices);
  80. }
  81. int btrfs_cleanup_fs_uuids(void)
  82. {
  83. struct btrfs_fs_devices *fs_devices;
  84. while (!list_empty(&fs_uuids)) {
  85. fs_devices = list_entry(fs_uuids.next,
  86. struct btrfs_fs_devices, list);
  87. list_del(&fs_devices->list);
  88. free_fs_devices(fs_devices);
  89. }
  90. return 0;
  91. }
  92. static noinline struct btrfs_device *__find_device(struct list_head *head,
  93. u64 devid, u8 *uuid)
  94. {
  95. struct btrfs_device *dev;
  96. list_for_each_entry(dev, head, dev_list) {
  97. if (dev->devid == devid &&
  98. (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
  99. return dev;
  100. }
  101. }
  102. return NULL;
  103. }
  104. static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
  105. {
  106. struct btrfs_fs_devices *fs_devices;
  107. list_for_each_entry(fs_devices, &fs_uuids, list) {
  108. if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
  109. return fs_devices;
  110. }
  111. return NULL;
  112. }
  113. static void requeue_list(struct btrfs_pending_bios *pending_bios,
  114. struct bio *head, struct bio *tail)
  115. {
  116. struct bio *old_head;
  117. old_head = pending_bios->head;
  118. pending_bios->head = head;
  119. if (pending_bios->tail)
  120. tail->bi_next = old_head;
  121. else
  122. pending_bios->tail = tail;
  123. }
  124. /*
  125. * we try to collect pending bios for a device so we don't get a large
  126. * number of procs sending bios down to the same device. This greatly
  127. * improves the schedulers ability to collect and merge the bios.
  128. *
  129. * But, it also turns into a long list of bios to process and that is sure
  130. * to eventually make the worker thread block. The solution here is to
  131. * make some progress and then put this work struct back at the end of
  132. * the list if the block device is congested. This way, multiple devices
  133. * can make progress from a single worker thread.
  134. */
  135. static noinline int run_scheduled_bios(struct btrfs_device *device)
  136. {
  137. struct bio *pending;
  138. struct backing_dev_info *bdi;
  139. struct btrfs_fs_info *fs_info;
  140. struct btrfs_pending_bios *pending_bios;
  141. struct bio *tail;
  142. struct bio *cur;
  143. int again = 0;
  144. unsigned long num_run;
  145. unsigned long num_sync_run;
  146. unsigned long batch_run = 0;
  147. unsigned long limit;
  148. unsigned long last_waited = 0;
  149. int force_reg = 0;
  150. bdi = blk_get_backing_dev_info(device->bdev);
  151. fs_info = device->dev_root->fs_info;
  152. limit = btrfs_async_submit_limit(fs_info);
  153. limit = limit * 2 / 3;
  154. /* we want to make sure that every time we switch from the sync
  155. * list to the normal list, we unplug
  156. */
  157. num_sync_run = 0;
  158. loop:
  159. spin_lock(&device->io_lock);
  160. loop_lock:
  161. num_run = 0;
  162. /* take all the bios off the list at once and process them
  163. * later on (without the lock held). But, remember the
  164. * tail and other pointers so the bios can be properly reinserted
  165. * into the list if we hit congestion
  166. */
  167. if (!force_reg && device->pending_sync_bios.head) {
  168. pending_bios = &device->pending_sync_bios;
  169. force_reg = 1;
  170. } else {
  171. pending_bios = &device->pending_bios;
  172. force_reg = 0;
  173. }
  174. pending = pending_bios->head;
  175. tail = pending_bios->tail;
  176. WARN_ON(pending && !tail);
  177. /*
  178. * if pending was null this time around, no bios need processing
  179. * at all and we can stop. Otherwise it'll loop back up again
  180. * and do an additional check so no bios are missed.
  181. *
  182. * device->running_pending is used to synchronize with the
  183. * schedule_bio code.
  184. */
  185. if (device->pending_sync_bios.head == NULL &&
  186. device->pending_bios.head == NULL) {
  187. again = 0;
  188. device->running_pending = 0;
  189. } else {
  190. again = 1;
  191. device->running_pending = 1;
  192. }
  193. pending_bios->head = NULL;
  194. pending_bios->tail = NULL;
  195. spin_unlock(&device->io_lock);
  196. /*
  197. * if we're doing the regular priority list, make sure we unplug
  198. * for any high prio bios we've sent down
  199. */
  200. if (pending_bios == &device->pending_bios && num_sync_run > 0) {
  201. num_sync_run = 0;
  202. blk_run_backing_dev(bdi, NULL);
  203. }
  204. while (pending) {
  205. rmb();
  206. /* we want to work on both lists, but do more bios on the
  207. * sync list than the regular list
  208. */
  209. if ((num_run > 32 &&
  210. pending_bios != &device->pending_sync_bios &&
  211. device->pending_sync_bios.head) ||
  212. (num_run > 64 && pending_bios == &device->pending_sync_bios &&
  213. device->pending_bios.head)) {
  214. spin_lock(&device->io_lock);
  215. requeue_list(pending_bios, pending, tail);
  216. goto loop_lock;
  217. }
  218. cur = pending;
  219. pending = pending->bi_next;
  220. cur->bi_next = NULL;
  221. atomic_dec(&fs_info->nr_async_bios);
  222. if (atomic_read(&fs_info->nr_async_bios) < limit &&
  223. waitqueue_active(&fs_info->async_submit_wait))
  224. wake_up(&fs_info->async_submit_wait);
  225. BUG_ON(atomic_read(&cur->bi_cnt) == 0);
  226. if (cur->bi_rw & REQ_SYNC)
  227. num_sync_run++;
  228. submit_bio(cur->bi_rw, cur);
  229. num_run++;
  230. batch_run++;
  231. if (need_resched()) {
  232. if (num_sync_run) {
  233. blk_run_backing_dev(bdi, NULL);
  234. num_sync_run = 0;
  235. }
  236. cond_resched();
  237. }
  238. /*
  239. * we made progress, there is more work to do and the bdi
  240. * is now congested. Back off and let other work structs
  241. * run instead
  242. */
  243. if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
  244. fs_info->fs_devices->open_devices > 1) {
  245. struct io_context *ioc;
  246. ioc = current->io_context;
  247. /*
  248. * the main goal here is that we don't want to
  249. * block if we're going to be able to submit
  250. * more requests without blocking.
  251. *
  252. * This code does two great things, it pokes into
  253. * the elevator code from a filesystem _and_
  254. * it makes assumptions about how batching works.
  255. */
  256. if (ioc && ioc->nr_batch_requests > 0 &&
  257. time_before(jiffies, ioc->last_waited + HZ/50UL) &&
  258. (last_waited == 0 ||
  259. ioc->last_waited == last_waited)) {
  260. /*
  261. * we want to go through our batch of
  262. * requests and stop. So, we copy out
  263. * the ioc->last_waited time and test
  264. * against it before looping
  265. */
  266. last_waited = ioc->last_waited;
  267. if (need_resched()) {
  268. if (num_sync_run) {
  269. blk_run_backing_dev(bdi, NULL);
  270. num_sync_run = 0;
  271. }
  272. cond_resched();
  273. }
  274. continue;
  275. }
  276. spin_lock(&device->io_lock);
  277. requeue_list(pending_bios, pending, tail);
  278. device->running_pending = 1;
  279. spin_unlock(&device->io_lock);
  280. btrfs_requeue_work(&device->work);
  281. goto done;
  282. }
  283. }
  284. if (num_sync_run) {
  285. num_sync_run = 0;
  286. blk_run_backing_dev(bdi, NULL);
  287. }
  288. /*
  289. * IO has already been through a long path to get here. Checksumming,
  290. * async helper threads, perhaps compression. We've done a pretty
  291. * good job of collecting a batch of IO and should just unplug
  292. * the device right away.
  293. *
  294. * This will help anyone who is waiting on the IO, they might have
  295. * already unplugged, but managed to do so before the bio they
  296. * cared about found its way down here.
  297. */
  298. blk_run_backing_dev(bdi, NULL);
  299. cond_resched();
  300. if (again)
  301. goto loop;
  302. spin_lock(&device->io_lock);
  303. if (device->pending_bios.head || device->pending_sync_bios.head)
  304. goto loop_lock;
  305. spin_unlock(&device->io_lock);
  306. done:
  307. return 0;
  308. }
  309. static void pending_bios_fn(struct btrfs_work *work)
  310. {
  311. struct btrfs_device *device;
  312. device = container_of(work, struct btrfs_device, work);
  313. run_scheduled_bios(device);
  314. }
  315. static noinline int device_list_add(const char *path,
  316. struct btrfs_super_block *disk_super,
  317. u64 devid, struct btrfs_fs_devices **fs_devices_ret)
  318. {
  319. struct btrfs_device *device;
  320. struct btrfs_fs_devices *fs_devices;
  321. u64 found_transid = btrfs_super_generation(disk_super);
  322. char *name;
  323. fs_devices = find_fsid(disk_super->fsid);
  324. if (!fs_devices) {
  325. fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
  326. if (!fs_devices)
  327. return -ENOMEM;
  328. INIT_LIST_HEAD(&fs_devices->devices);
  329. INIT_LIST_HEAD(&fs_devices->alloc_list);
  330. list_add(&fs_devices->list, &fs_uuids);
  331. memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
  332. fs_devices->latest_devid = devid;
  333. fs_devices->latest_trans = found_transid;
  334. mutex_init(&fs_devices->device_list_mutex);
  335. device = NULL;
  336. } else {
  337. device = __find_device(&fs_devices->devices, devid,
  338. disk_super->dev_item.uuid);
  339. }
  340. if (!device) {
  341. if (fs_devices->opened)
  342. return -EBUSY;
  343. device = kzalloc(sizeof(*device), GFP_NOFS);
  344. if (!device) {
  345. /* we can safely leave the fs_devices entry around */
  346. return -ENOMEM;
  347. }
  348. device->devid = devid;
  349. device->work.func = pending_bios_fn;
  350. memcpy(device->uuid, disk_super->dev_item.uuid,
  351. BTRFS_UUID_SIZE);
  352. device->barriers = 1;
  353. spin_lock_init(&device->io_lock);
  354. device->name = kstrdup(path, GFP_NOFS);
  355. if (!device->name) {
  356. kfree(device);
  357. return -ENOMEM;
  358. }
  359. INIT_LIST_HEAD(&device->dev_alloc_list);
  360. mutex_lock(&fs_devices->device_list_mutex);
  361. list_add(&device->dev_list, &fs_devices->devices);
  362. mutex_unlock(&fs_devices->device_list_mutex);
  363. device->fs_devices = fs_devices;
  364. fs_devices->num_devices++;
  365. } else if (strcmp(device->name, path)) {
  366. name = kstrdup(path, GFP_NOFS);
  367. if (!name)
  368. return -ENOMEM;
  369. kfree(device->name);
  370. device->name = name;
  371. }
  372. if (found_transid > fs_devices->latest_trans) {
  373. fs_devices->latest_devid = devid;
  374. fs_devices->latest_trans = found_transid;
  375. }
  376. *fs_devices_ret = fs_devices;
  377. return 0;
  378. }
  379. static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
  380. {
  381. struct btrfs_fs_devices *fs_devices;
  382. struct btrfs_device *device;
  383. struct btrfs_device *orig_dev;
  384. fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
  385. if (!fs_devices)
  386. return ERR_PTR(-ENOMEM);
  387. INIT_LIST_HEAD(&fs_devices->devices);
  388. INIT_LIST_HEAD(&fs_devices->alloc_list);
  389. INIT_LIST_HEAD(&fs_devices->list);
  390. mutex_init(&fs_devices->device_list_mutex);
  391. fs_devices->latest_devid = orig->latest_devid;
  392. fs_devices->latest_trans = orig->latest_trans;
  393. memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
  394. mutex_lock(&orig->device_list_mutex);
  395. list_for_each_entry(orig_dev, &orig->devices, dev_list) {
  396. device = kzalloc(sizeof(*device), GFP_NOFS);
  397. if (!device)
  398. goto error;
  399. device->name = kstrdup(orig_dev->name, GFP_NOFS);
  400. if (!device->name) {
  401. kfree(device);
  402. goto error;
  403. }
  404. device->devid = orig_dev->devid;
  405. device->work.func = pending_bios_fn;
  406. memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
  407. device->barriers = 1;
  408. spin_lock_init(&device->io_lock);
  409. INIT_LIST_HEAD(&device->dev_list);
  410. INIT_LIST_HEAD(&device->dev_alloc_list);
  411. list_add(&device->dev_list, &fs_devices->devices);
  412. device->fs_devices = fs_devices;
  413. fs_devices->num_devices++;
  414. }
  415. mutex_unlock(&orig->device_list_mutex);
  416. return fs_devices;
  417. error:
  418. mutex_unlock(&orig->device_list_mutex);
  419. free_fs_devices(fs_devices);
  420. return ERR_PTR(-ENOMEM);
  421. }
  422. int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
  423. {
  424. struct btrfs_device *device, *next;
  425. mutex_lock(&uuid_mutex);
  426. again:
  427. mutex_lock(&fs_devices->device_list_mutex);
  428. list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
  429. if (device->in_fs_metadata)
  430. continue;
  431. if (device->bdev) {
  432. close_bdev_exclusive(device->bdev, device->mode);
  433. device->bdev = NULL;
  434. fs_devices->open_devices--;
  435. }
  436. if (device->writeable) {
  437. list_del_init(&device->dev_alloc_list);
  438. device->writeable = 0;
  439. fs_devices->rw_devices--;
  440. }
  441. list_del_init(&device->dev_list);
  442. fs_devices->num_devices--;
  443. kfree(device->name);
  444. kfree(device);
  445. }
  446. mutex_unlock(&fs_devices->device_list_mutex);
  447. if (fs_devices->seed) {
  448. fs_devices = fs_devices->seed;
  449. goto again;
  450. }
  451. mutex_unlock(&uuid_mutex);
  452. return 0;
  453. }
  454. static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
  455. {
  456. struct btrfs_device *device;
  457. if (--fs_devices->opened > 0)
  458. return 0;
  459. list_for_each_entry(device, &fs_devices->devices, dev_list) {
  460. if (device->bdev) {
  461. close_bdev_exclusive(device->bdev, device->mode);
  462. fs_devices->open_devices--;
  463. }
  464. if (device->writeable) {
  465. list_del_init(&device->dev_alloc_list);
  466. fs_devices->rw_devices--;
  467. }
  468. device->bdev = NULL;
  469. device->writeable = 0;
  470. device->in_fs_metadata = 0;
  471. }
  472. WARN_ON(fs_devices->open_devices);
  473. WARN_ON(fs_devices->rw_devices);
  474. fs_devices->opened = 0;
  475. fs_devices->seeding = 0;
  476. return 0;
  477. }
  478. int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
  479. {
  480. struct btrfs_fs_devices *seed_devices = NULL;
  481. int ret;
  482. mutex_lock(&uuid_mutex);
  483. ret = __btrfs_close_devices(fs_devices);
  484. if (!fs_devices->opened) {
  485. seed_devices = fs_devices->seed;
  486. fs_devices->seed = NULL;
  487. }
  488. mutex_unlock(&uuid_mutex);
  489. while (seed_devices) {
  490. fs_devices = seed_devices;
  491. seed_devices = fs_devices->seed;
  492. __btrfs_close_devices(fs_devices);
  493. free_fs_devices(fs_devices);
  494. }
  495. return ret;
  496. }
  497. static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
  498. fmode_t flags, void *holder)
  499. {
  500. struct block_device *bdev;
  501. struct list_head *head = &fs_devices->devices;
  502. struct btrfs_device *device;
  503. struct block_device *latest_bdev = NULL;
  504. struct buffer_head *bh;
  505. struct btrfs_super_block *disk_super;
  506. u64 latest_devid = 0;
  507. u64 latest_transid = 0;
  508. u64 devid;
  509. int seeding = 1;
  510. int ret = 0;
  511. list_for_each_entry(device, head, dev_list) {
  512. if (device->bdev)
  513. continue;
  514. if (!device->name)
  515. continue;
  516. bdev = open_bdev_exclusive(device->name, flags, holder);
  517. if (IS_ERR(bdev)) {
  518. printk(KERN_INFO "open %s failed\n", device->name);
  519. goto error;
  520. }
  521. set_blocksize(bdev, 4096);
  522. bh = btrfs_read_dev_super(bdev);
  523. if (!bh)
  524. goto error_close;
  525. disk_super = (struct btrfs_super_block *)bh->b_data;
  526. devid = btrfs_stack_device_id(&disk_super->dev_item);
  527. if (devid != device->devid)
  528. goto error_brelse;
  529. if (memcmp(device->uuid, disk_super->dev_item.uuid,
  530. BTRFS_UUID_SIZE))
  531. goto error_brelse;
  532. device->generation = btrfs_super_generation(disk_super);
  533. if (!latest_transid || device->generation > latest_transid) {
  534. latest_devid = devid;
  535. latest_transid = device->generation;
  536. latest_bdev = bdev;
  537. }
  538. if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
  539. device->writeable = 0;
  540. } else {
  541. device->writeable = !bdev_read_only(bdev);
  542. seeding = 0;
  543. }
  544. device->bdev = bdev;
  545. device->in_fs_metadata = 0;
  546. device->mode = flags;
  547. if (!blk_queue_nonrot(bdev_get_queue(bdev)))
  548. fs_devices->rotating = 1;
  549. fs_devices->open_devices++;
  550. if (device->writeable) {
  551. fs_devices->rw_devices++;
  552. list_add(&device->dev_alloc_list,
  553. &fs_devices->alloc_list);
  554. }
  555. continue;
  556. error_brelse:
  557. brelse(bh);
  558. error_close:
  559. close_bdev_exclusive(bdev, FMODE_READ);
  560. error:
  561. continue;
  562. }
  563. if (fs_devices->open_devices == 0) {
  564. ret = -EIO;
  565. goto out;
  566. }
  567. fs_devices->seeding = seeding;
  568. fs_devices->opened = 1;
  569. fs_devices->latest_bdev = latest_bdev;
  570. fs_devices->latest_devid = latest_devid;
  571. fs_devices->latest_trans = latest_transid;
  572. fs_devices->total_rw_bytes = 0;
  573. out:
  574. return ret;
  575. }
  576. int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
  577. fmode_t flags, void *holder)
  578. {
  579. int ret;
  580. mutex_lock(&uuid_mutex);
  581. if (fs_devices->opened) {
  582. fs_devices->opened++;
  583. ret = 0;
  584. } else {
  585. ret = __btrfs_open_devices(fs_devices, flags, holder);
  586. }
  587. mutex_unlock(&uuid_mutex);
  588. return ret;
  589. }
  590. int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
  591. struct btrfs_fs_devices **fs_devices_ret)
  592. {
  593. struct btrfs_super_block *disk_super;
  594. struct block_device *bdev;
  595. struct buffer_head *bh;
  596. int ret;
  597. u64 devid;
  598. u64 transid;
  599. mutex_lock(&uuid_mutex);
  600. bdev = open_bdev_exclusive(path, flags, holder);
  601. if (IS_ERR(bdev)) {
  602. ret = PTR_ERR(bdev);
  603. goto error;
  604. }
  605. ret = set_blocksize(bdev, 4096);
  606. if (ret)
  607. goto error_close;
  608. bh = btrfs_read_dev_super(bdev);
  609. if (!bh) {
  610. ret = -EIO;
  611. goto error_close;
  612. }
  613. disk_super = (struct btrfs_super_block *)bh->b_data;
  614. devid = btrfs_stack_device_id(&disk_super->dev_item);
  615. transid = btrfs_super_generation(disk_super);
  616. if (disk_super->label[0])
  617. printk(KERN_INFO "device label %s ", disk_super->label);
  618. else {
  619. /* FIXME, make a readl uuid parser */
  620. printk(KERN_INFO "device fsid %llx-%llx ",
  621. *(unsigned long long *)disk_super->fsid,
  622. *(unsigned long long *)(disk_super->fsid + 8));
  623. }
  624. printk(KERN_CONT "devid %llu transid %llu %s\n",
  625. (unsigned long long)devid, (unsigned long long)transid, path);
  626. ret = device_list_add(path, disk_super, devid, fs_devices_ret);
  627. brelse(bh);
  628. error_close:
  629. close_bdev_exclusive(bdev, flags);
  630. error:
  631. mutex_unlock(&uuid_mutex);
  632. return ret;
  633. }
  634. /*
  635. * this uses a pretty simple search, the expectation is that it is
  636. * called very infrequently and that a given device has a small number
  637. * of extents
  638. */
  639. int find_free_dev_extent(struct btrfs_trans_handle *trans,
  640. struct btrfs_device *device, u64 num_bytes,
  641. u64 *start, u64 *max_avail)
  642. {
  643. struct btrfs_key key;
  644. struct btrfs_root *root = device->dev_root;
  645. struct btrfs_dev_extent *dev_extent = NULL;
  646. struct btrfs_path *path;
  647. u64 hole_size = 0;
  648. u64 last_byte = 0;
  649. u64 search_start = 0;
  650. u64 search_end = device->total_bytes;
  651. int ret;
  652. int slot = 0;
  653. int start_found;
  654. struct extent_buffer *l;
  655. path = btrfs_alloc_path();
  656. if (!path)
  657. return -ENOMEM;
  658. path->reada = 2;
  659. start_found = 0;
  660. /* FIXME use last free of some kind */
  661. /* we don't want to overwrite the superblock on the drive,
  662. * so we make sure to start at an offset of at least 1MB
  663. */
  664. search_start = max((u64)1024 * 1024, search_start);
  665. if (root->fs_info->alloc_start + num_bytes <= device->total_bytes)
  666. search_start = max(root->fs_info->alloc_start, search_start);
  667. key.objectid = device->devid;
  668. key.offset = search_start;
  669. key.type = BTRFS_DEV_EXTENT_KEY;
  670. ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
  671. if (ret < 0)
  672. goto error;
  673. if (ret > 0) {
  674. ret = btrfs_previous_item(root, path, key.objectid, key.type);
  675. if (ret < 0)
  676. goto error;
  677. if (ret > 0)
  678. start_found = 1;
  679. }
  680. l = path->nodes[0];
  681. btrfs_item_key_to_cpu(l, &key, path->slots[0]);
  682. while (1) {
  683. l = path->nodes[0];
  684. slot = path->slots[0];
  685. if (slot >= btrfs_header_nritems(l)) {
  686. ret = btrfs_next_leaf(root, path);
  687. if (ret == 0)
  688. continue;
  689. if (ret < 0)
  690. goto error;
  691. no_more_items:
  692. if (!start_found) {
  693. if (search_start >= search_end) {
  694. ret = -ENOSPC;
  695. goto error;
  696. }
  697. *start = search_start;
  698. start_found = 1;
  699. goto check_pending;
  700. }
  701. *start = last_byte > search_start ?
  702. last_byte : search_start;
  703. if (search_end <= *start) {
  704. ret = -ENOSPC;
  705. goto error;
  706. }
  707. goto check_pending;
  708. }
  709. btrfs_item_key_to_cpu(l, &key, slot);
  710. if (key.objectid < device->devid)
  711. goto next;
  712. if (key.objectid > device->devid)
  713. goto no_more_items;
  714. if (key.offset >= search_start && key.offset > last_byte &&
  715. start_found) {
  716. if (last_byte < search_start)
  717. last_byte = search_start;
  718. hole_size = key.offset - last_byte;
  719. if (hole_size > *max_avail)
  720. *max_avail = hole_size;
  721. if (key.offset > last_byte &&
  722. hole_size >= num_bytes) {
  723. *start = last_byte;
  724. goto check_pending;
  725. }
  726. }
  727. if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
  728. goto next;
  729. start_found = 1;
  730. dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
  731. last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
  732. next:
  733. path->slots[0]++;
  734. cond_resched();
  735. }
  736. check_pending:
  737. /* we have to make sure we didn't find an extent that has already
  738. * been allocated by the map tree or the original allocation
  739. */
  740. BUG_ON(*start < search_start);
  741. if (*start + num_bytes > search_end) {
  742. ret = -ENOSPC;
  743. goto error;
  744. }
  745. /* check for pending inserts here */
  746. ret = 0;
  747. error:
  748. btrfs_free_path(path);
  749. return ret;
  750. }
  751. static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
  752. struct btrfs_device *device,
  753. u64 start)
  754. {
  755. int ret;
  756. struct btrfs_path *path;
  757. struct btrfs_root *root = device->dev_root;
  758. struct btrfs_key key;
  759. struct btrfs_key found_key;
  760. struct extent_buffer *leaf = NULL;
  761. struct btrfs_dev_extent *extent = NULL;
  762. path = btrfs_alloc_path();
  763. if (!path)
  764. return -ENOMEM;
  765. key.objectid = device->devid;
  766. key.offset = start;
  767. key.type = BTRFS_DEV_EXTENT_KEY;
  768. ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
  769. if (ret > 0) {
  770. ret = btrfs_previous_item(root, path, key.objectid,
  771. BTRFS_DEV_EXTENT_KEY);
  772. BUG_ON(ret);
  773. leaf = path->nodes[0];
  774. btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
  775. extent = btrfs_item_ptr(leaf, path->slots[0],
  776. struct btrfs_dev_extent);
  777. BUG_ON(found_key.offset > start || found_key.offset +
  778. btrfs_dev_extent_length(leaf, extent) < start);
  779. ret = 0;
  780. } else if (ret == 0) {
  781. leaf = path->nodes[0];
  782. extent = btrfs_item_ptr(leaf, path->slots[0],
  783. struct btrfs_dev_extent);
  784. }
  785. BUG_ON(ret);
  786. if (device->bytes_used > 0)
  787. device->bytes_used -= btrfs_dev_extent_length(leaf, extent);
  788. ret = btrfs_del_item(trans, root, path);
  789. BUG_ON(ret);
  790. btrfs_free_path(path);
  791. return ret;
  792. }
  793. int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
  794. struct btrfs_device *device,
  795. u64 chunk_tree, u64 chunk_objectid,
  796. u64 chunk_offset, u64 start, u64 num_bytes)
  797. {
  798. int ret;
  799. struct btrfs_path *path;
  800. struct btrfs_root *root = device->dev_root;
  801. struct btrfs_dev_extent *extent;
  802. struct extent_buffer *leaf;
  803. struct btrfs_key key;
  804. WARN_ON(!device->in_fs_metadata);
  805. path = btrfs_alloc_path();
  806. if (!path)
  807. return -ENOMEM;
  808. key.objectid = device->devid;
  809. key.offset = start;
  810. key.type = BTRFS_DEV_EXTENT_KEY;
  811. ret = btrfs_insert_empty_item(trans, root, path, &key,
  812. sizeof(*extent));
  813. BUG_ON(ret);
  814. leaf = path->nodes[0];
  815. extent = btrfs_item_ptr(leaf, path->slots[0],
  816. struct btrfs_dev_extent);
  817. btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
  818. btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
  819. btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
  820. write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
  821. (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
  822. BTRFS_UUID_SIZE);
  823. btrfs_set_dev_extent_length(leaf, extent, num_bytes);
  824. btrfs_mark_buffer_dirty(leaf);
  825. btrfs_free_path(path);
  826. return ret;
  827. }
  828. static noinline int find_next_chunk(struct btrfs_root *root,
  829. u64 objectid, u64 *offset)
  830. {
  831. struct btrfs_path *path;
  832. int ret;
  833. struct btrfs_key key;
  834. struct btrfs_chunk *chunk;
  835. struct btrfs_key found_key;
  836. path = btrfs_alloc_path();
  837. BUG_ON(!path);
  838. key.objectid = objectid;
  839. key.offset = (u64)-1;
  840. key.type = BTRFS_CHUNK_ITEM_KEY;
  841. ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
  842. if (ret < 0)
  843. goto error;
  844. BUG_ON(ret == 0);
  845. ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
  846. if (ret) {
  847. *offset = 0;
  848. } else {
  849. btrfs_item_key_to_cpu(path->nodes[0], &found_key,
  850. path->slots[0]);
  851. if (found_key.objectid != objectid)
  852. *offset = 0;
  853. else {
  854. chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
  855. struct btrfs_chunk);
  856. *offset = found_key.offset +
  857. btrfs_chunk_length(path->nodes[0], chunk);
  858. }
  859. }
  860. ret = 0;
  861. error:
  862. btrfs_free_path(path);
  863. return ret;
  864. }
  865. static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
  866. {
  867. int ret;
  868. struct btrfs_key key;
  869. struct btrfs_key found_key;
  870. struct btrfs_path *path;
  871. root = root->fs_info->chunk_root;
  872. path = btrfs_alloc_path();
  873. if (!path)
  874. return -ENOMEM;
  875. key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
  876. key.type = BTRFS_DEV_ITEM_KEY;
  877. key.offset = (u64)-1;
  878. ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
  879. if (ret < 0)
  880. goto error;
  881. BUG_ON(ret == 0);
  882. ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
  883. BTRFS_DEV_ITEM_KEY);
  884. if (ret) {
  885. *objectid = 1;
  886. } else {
  887. btrfs_item_key_to_cpu(path->nodes[0], &found_key,
  888. path->slots[0]);
  889. *objectid = found_key.offset + 1;
  890. }
  891. ret = 0;
  892. error:
  893. btrfs_free_path(path);
  894. return ret;
  895. }
  896. /*
  897. * the device information is stored in the chunk root
  898. * the btrfs_device struct should be fully filled in
  899. */
  900. int btrfs_add_device(struct btrfs_trans_handle *trans,
  901. struct btrfs_root *root,
  902. struct btrfs_device *device)
  903. {
  904. int ret;
  905. struct btrfs_path *path;
  906. struct btrfs_dev_item *dev_item;
  907. struct extent_buffer *leaf;
  908. struct btrfs_key key;
  909. unsigned long ptr;
  910. root = root->fs_info->chunk_root;
  911. path = btrfs_alloc_path();
  912. if (!path)
  913. return -ENOMEM;
  914. key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
  915. key.type = BTRFS_DEV_ITEM_KEY;
  916. key.offset = device->devid;
  917. ret = btrfs_insert_empty_item(trans, root, path, &key,
  918. sizeof(*dev_item));
  919. if (ret)
  920. goto out;
  921. leaf = path->nodes[0];
  922. dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
  923. btrfs_set_device_id(leaf, dev_item, device->devid);
  924. btrfs_set_device_generation(leaf, dev_item, 0);
  925. btrfs_set_device_type(leaf, dev_item, device->type);
  926. btrfs_set_device_io_align(leaf, dev_item, device->io_align);
  927. btrfs_set_device_io_width(leaf, dev_item, device->io_width);
  928. btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
  929. btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
  930. btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
  931. btrfs_set_device_group(leaf, dev_item, 0);
  932. btrfs_set_device_seek_speed(leaf, dev_item, 0);
  933. btrfs_set_device_bandwidth(leaf, dev_item, 0);
  934. btrfs_set_device_start_offset(leaf, dev_item, 0);
  935. ptr = (unsigned long)btrfs_device_uuid(dev_item);
  936. write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
  937. ptr = (unsigned long)btrfs_device_fsid(dev_item);
  938. write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
  939. btrfs_mark_buffer_dirty(leaf);
  940. ret = 0;
  941. out:
  942. btrfs_free_path(path);
  943. return ret;
  944. }
  945. static int btrfs_rm_dev_item(struct btrfs_root *root,
  946. struct btrfs_device *device)
  947. {
  948. int ret;
  949. struct btrfs_path *path;
  950. struct btrfs_key key;
  951. struct btrfs_trans_handle *trans;
  952. root = root->fs_info->chunk_root;
  953. path = btrfs_alloc_path();
  954. if (!path)
  955. return -ENOMEM;
  956. trans = btrfs_start_transaction(root, 0);
  957. key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
  958. key.type = BTRFS_DEV_ITEM_KEY;
  959. key.offset = device->devid;
  960. lock_chunks(root);
  961. ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
  962. if (ret < 0)
  963. goto out;
  964. if (ret > 0) {
  965. ret = -ENOENT;
  966. goto out;
  967. }
  968. ret = btrfs_del_item(trans, root, path);
  969. if (ret)
  970. goto out;
  971. out:
  972. btrfs_free_path(path);
  973. unlock_chunks(root);
  974. btrfs_commit_transaction(trans, root);
  975. return ret;
  976. }
  977. int btrfs_rm_device(struct btrfs_root *root, char *device_path)
  978. {
  979. struct btrfs_device *device;
  980. struct btrfs_device *next_device;
  981. struct block_device *bdev;
  982. struct buffer_head *bh = NULL;
  983. struct btrfs_super_block *disk_super;
  984. u64 all_avail;
  985. u64 devid;
  986. u64 num_devices;
  987. u8 *dev_uuid;
  988. int ret = 0;
  989. mutex_lock(&uuid_mutex);
  990. mutex_lock(&root->fs_info->volume_mutex);
  991. all_avail = root->fs_info->avail_data_alloc_bits |
  992. root->fs_info->avail_system_alloc_bits |
  993. root->fs_info->avail_metadata_alloc_bits;
  994. if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
  995. root->fs_info->fs_devices->num_devices <= 4) {
  996. printk(KERN_ERR "btrfs: unable to go below four devices "
  997. "on raid10\n");
  998. ret = -EINVAL;
  999. goto out;
  1000. }
  1001. if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
  1002. root->fs_info->fs_devices->num_devices <= 2) {
  1003. printk(KERN_ERR "btrfs: unable to go below two "
  1004. "devices on raid1\n");
  1005. ret = -EINVAL;
  1006. goto out;
  1007. }
  1008. if (strcmp(device_path, "missing") == 0) {
  1009. struct list_head *devices;
  1010. struct btrfs_device *tmp;
  1011. device = NULL;
  1012. devices = &root->fs_info->fs_devices->devices;
  1013. mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
  1014. list_for_each_entry(tmp, devices, dev_list) {
  1015. if (tmp->in_fs_metadata && !tmp->bdev) {
  1016. device = tmp;
  1017. break;
  1018. }
  1019. }
  1020. mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
  1021. bdev = NULL;
  1022. bh = NULL;
  1023. disk_super = NULL;
  1024. if (!device) {
  1025. printk(KERN_ERR "btrfs: no missing devices found to "
  1026. "remove\n");
  1027. goto out;
  1028. }
  1029. } else {
  1030. bdev = open_bdev_exclusive(device_path, FMODE_READ,
  1031. root->fs_info->bdev_holder);
  1032. if (IS_ERR(bdev)) {
  1033. ret = PTR_ERR(bdev);
  1034. goto out;
  1035. }
  1036. set_blocksize(bdev, 4096);
  1037. bh = btrfs_read_dev_super(bdev);
  1038. if (!bh) {
  1039. ret = -EIO;
  1040. goto error_close;
  1041. }
  1042. disk_super = (struct btrfs_super_block *)bh->b_data;
  1043. devid = btrfs_stack_device_id(&disk_super->dev_item);
  1044. dev_uuid = disk_super->dev_item.uuid;
  1045. device = btrfs_find_device(root, devid, dev_uuid,
  1046. disk_super->fsid);
  1047. if (!device) {
  1048. ret = -ENOENT;
  1049. goto error_brelse;
  1050. }
  1051. }
  1052. if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
  1053. printk(KERN_ERR "btrfs: unable to remove the only writeable "
  1054. "device\n");
  1055. ret = -EINVAL;
  1056. goto error_brelse;
  1057. }
  1058. if (device->writeable) {
  1059. list_del_init(&device->dev_alloc_list);
  1060. root->fs_info->fs_devices->rw_devices--;
  1061. }
  1062. ret = btrfs_shrink_device(device, 0);
  1063. if (ret)
  1064. goto error_brelse;
  1065. ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
  1066. if (ret)
  1067. goto error_brelse;
  1068. device->in_fs_metadata = 0;
  1069. /*
  1070. * the device list mutex makes sure that we don't change
  1071. * the device list while someone else is writing out all
  1072. * the device supers.
  1073. */
  1074. mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
  1075. list_del_init(&device->dev_list);
  1076. mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
  1077. device->fs_devices->num_devices--;
  1078. next_device = list_entry(root->fs_info->fs_devices->devices.next,
  1079. struct btrfs_device, dev_list);
  1080. if (device->bdev == root->fs_info->sb->s_bdev)
  1081. root->fs_info->sb->s_bdev = next_device->bdev;
  1082. if (device->bdev == root->fs_info->fs_devices->latest_bdev)
  1083. root->fs_info->fs_devices->latest_bdev = next_device->bdev;
  1084. if (device->bdev) {
  1085. close_bdev_exclusive(device->bdev, device->mode);
  1086. device->bdev = NULL;
  1087. device->fs_devices->open_devices--;
  1088. }
  1089. num_devices = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
  1090. btrfs_set_super_num_devices(&root->fs_info->super_copy, num_devices);
  1091. if (device->fs_devices->open_devices == 0) {
  1092. struct btrfs_fs_devices *fs_devices;
  1093. fs_devices = root->fs_info->fs_devices;
  1094. while (fs_devices) {
  1095. if (fs_devices->seed == device->fs_devices)
  1096. break;
  1097. fs_devices = fs_devices->seed;
  1098. }
  1099. fs_devices->seed = device->fs_devices->seed;
  1100. device->fs_devices->seed = NULL;
  1101. __btrfs_close_devices(device->fs_devices);
  1102. free_fs_devices(device->fs_devices);
  1103. }
  1104. /*
  1105. * at this point, the device is zero sized. We want to
  1106. * remove it from the devices list and zero out the old super
  1107. */
  1108. if (device->writeable) {
  1109. /* make sure this device isn't detected as part of
  1110. * the FS anymore
  1111. */
  1112. memset(&disk_super->magic, 0, sizeof(disk_super->magic));
  1113. set_buffer_dirty(bh);
  1114. sync_dirty_buffer(bh);
  1115. }
  1116. kfree(device->name);
  1117. kfree(device);
  1118. ret = 0;
  1119. error_brelse:
  1120. brelse(bh);
  1121. error_close:
  1122. if (bdev)
  1123. close_bdev_exclusive(bdev, FMODE_READ);
  1124. out:
  1125. mutex_unlock(&root->fs_info->volume_mutex);
  1126. mutex_unlock(&uuid_mutex);
  1127. return ret;
  1128. }
  1129. /*
  1130. * does all the dirty work required for changing file system's UUID.
  1131. */
  1132. static int btrfs_prepare_sprout(struct btrfs_trans_handle *trans,
  1133. struct btrfs_root *root)
  1134. {
  1135. struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
  1136. struct btrfs_fs_devices *old_devices;
  1137. struct btrfs_fs_devices *seed_devices;
  1138. struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
  1139. struct btrfs_device *device;
  1140. u64 super_flags;
  1141. BUG_ON(!mutex_is_locked(&uuid_mutex));
  1142. if (!fs_devices->seeding)
  1143. return -EINVAL;
  1144. seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
  1145. if (!seed_devices)
  1146. return -ENOMEM;
  1147. old_devices = clone_fs_devices(fs_devices);
  1148. if (IS_ERR(old_devices)) {
  1149. kfree(seed_devices);
  1150. return PTR_ERR(old_devices);
  1151. }
  1152. list_add(&old_devices->list, &fs_uuids);
  1153. memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
  1154. seed_devices->opened = 1;
  1155. INIT_LIST_HEAD(&seed_devices->devices);
  1156. INIT_LIST_HEAD(&seed_devices->alloc_list);
  1157. mutex_init(&seed_devices->device_list_mutex);
  1158. list_splice_init(&fs_devices->devices, &seed_devices->devices);
  1159. list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
  1160. list_for_each_entry(device, &seed_devices->devices, dev_list) {
  1161. device->fs_devices = seed_devices;
  1162. }
  1163. fs_devices->seeding = 0;
  1164. fs_devices->num_devices = 0;
  1165. fs_devices->open_devices = 0;
  1166. fs_devices->seed = seed_devices;
  1167. generate_random_uuid(fs_devices->fsid);
  1168. memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
  1169. memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
  1170. super_flags = btrfs_super_flags(disk_super) &
  1171. ~BTRFS_SUPER_FLAG_SEEDING;
  1172. btrfs_set_super_flags(disk_super, super_flags);
  1173. return 0;
  1174. }
  1175. /*
  1176. * strore the expected generation for seed devices in device items.
  1177. */
  1178. static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
  1179. struct btrfs_root *root)
  1180. {
  1181. struct btrfs_path *path;
  1182. struct extent_buffer *leaf;
  1183. struct btrfs_dev_item *dev_item;
  1184. struct btrfs_device *device;
  1185. struct btrfs_key key;
  1186. u8 fs_uuid[BTRFS_UUID_SIZE];
  1187. u8 dev_uuid[BTRFS_UUID_SIZE];
  1188. u64 devid;
  1189. int ret;
  1190. path = btrfs_alloc_path();
  1191. if (!path)
  1192. return -ENOMEM;
  1193. root = root->fs_info->chunk_root;
  1194. key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
  1195. key.offset = 0;
  1196. key.type = BTRFS_DEV_ITEM_KEY;
  1197. while (1) {
  1198. ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
  1199. if (ret < 0)
  1200. goto error;
  1201. leaf = path->nodes[0];
  1202. next_slot:
  1203. if (path->slots[0] >= btrfs_header_nritems(leaf)) {
  1204. ret = btrfs_next_leaf(root, path);
  1205. if (ret > 0)
  1206. break;
  1207. if (ret < 0)
  1208. goto error;
  1209. leaf = path->nodes[0];
  1210. btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
  1211. btrfs_release_path(root, path);
  1212. continue;
  1213. }
  1214. btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
  1215. if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
  1216. key.type != BTRFS_DEV_ITEM_KEY)
  1217. break;
  1218. dev_item = btrfs_item_ptr(leaf, path->slots[0],
  1219. struct btrfs_dev_item);
  1220. devid = btrfs_device_id(leaf, dev_item);
  1221. read_extent_buffer(leaf, dev_uuid,
  1222. (unsigned long)btrfs_device_uuid(dev_item),
  1223. BTRFS_UUID_SIZE);
  1224. read_extent_buffer(leaf, fs_uuid,
  1225. (unsigned long)btrfs_device_fsid(dev_item),
  1226. BTRFS_UUID_SIZE);
  1227. device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
  1228. BUG_ON(!device);
  1229. if (device->fs_devices->seeding) {
  1230. btrfs_set_device_generation(leaf, dev_item,
  1231. device->generation);
  1232. btrfs_mark_buffer_dirty(leaf);
  1233. }
  1234. path->slots[0]++;
  1235. goto next_slot;
  1236. }
  1237. ret = 0;
  1238. error:
  1239. btrfs_free_path(path);
  1240. return ret;
  1241. }
  1242. int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
  1243. {
  1244. struct btrfs_trans_handle *trans;
  1245. struct btrfs_device *device;
  1246. struct block_device *bdev;
  1247. struct list_head *devices;
  1248. struct super_block *sb = root->fs_info->sb;
  1249. u64 total_bytes;
  1250. int seeding_dev = 0;
  1251. int ret = 0;
  1252. if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
  1253. return -EINVAL;
  1254. bdev = open_bdev_exclusive(device_path, 0, root->fs_info->bdev_holder);
  1255. if (IS_ERR(bdev))
  1256. return PTR_ERR(bdev);
  1257. if (root->fs_info->fs_devices->seeding) {
  1258. seeding_dev = 1;
  1259. down_write(&sb->s_umount);
  1260. mutex_lock(&uuid_mutex);
  1261. }
  1262. filemap_write_and_wait(bdev->bd_inode->i_mapping);
  1263. mutex_lock(&root->fs_info->volume_mutex);
  1264. devices = &root->fs_info->fs_devices->devices;
  1265. /*
  1266. * we have the volume lock, so we don't need the extra
  1267. * device list mutex while reading the list here.
  1268. */
  1269. list_for_each_entry(device, devices, dev_list) {
  1270. if (device->bdev == bdev) {
  1271. ret = -EEXIST;
  1272. goto error;
  1273. }
  1274. }
  1275. device = kzalloc(sizeof(*device), GFP_NOFS);
  1276. if (!device) {
  1277. /* we can safely leave the fs_devices entry around */
  1278. ret = -ENOMEM;
  1279. goto error;
  1280. }
  1281. device->name = kstrdup(device_path, GFP_NOFS);
  1282. if (!device->name) {
  1283. kfree(device);
  1284. ret = -ENOMEM;
  1285. goto error;
  1286. }
  1287. ret = find_next_devid(root, &device->devid);
  1288. if (ret) {
  1289. kfree(device);
  1290. goto error;
  1291. }
  1292. trans = btrfs_start_transaction(root, 0);
  1293. lock_chunks(root);
  1294. device->barriers = 1;
  1295. device->writeable = 1;
  1296. device->work.func = pending_bios_fn;
  1297. generate_random_uuid(device->uuid);
  1298. spin_lock_init(&device->io_lock);
  1299. device->generation = trans->transid;
  1300. device->io_width = root->sectorsize;
  1301. device->io_align = root->sectorsize;
  1302. device->sector_size = root->sectorsize;
  1303. device->total_bytes = i_size_read(bdev->bd_inode);
  1304. device->disk_total_bytes = device->total_bytes;
  1305. device->dev_root = root->fs_info->dev_root;
  1306. device->bdev = bdev;
  1307. device->in_fs_metadata = 1;
  1308. device->mode = 0;
  1309. set_blocksize(device->bdev, 4096);
  1310. if (seeding_dev) {
  1311. sb->s_flags &= ~MS_RDONLY;
  1312. ret = btrfs_prepare_sprout(trans, root);
  1313. BUG_ON(ret);
  1314. }
  1315. device->fs_devices = root->fs_info->fs_devices;
  1316. /*
  1317. * we don't want write_supers to jump in here with our device
  1318. * half setup
  1319. */
  1320. mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
  1321. list_add(&device->dev_list, &root->fs_info->fs_devices->devices);
  1322. list_add(&device->dev_alloc_list,
  1323. &root->fs_info->fs_devices->alloc_list);
  1324. root->fs_info->fs_devices->num_devices++;
  1325. root->fs_info->fs_devices->open_devices++;
  1326. root->fs_info->fs_devices->rw_devices++;
  1327. root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
  1328. if (!blk_queue_nonrot(bdev_get_queue(bdev)))
  1329. root->fs_info->fs_devices->rotating = 1;
  1330. total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
  1331. btrfs_set_super_total_bytes(&root->fs_info->super_copy,
  1332. total_bytes + device->total_bytes);
  1333. total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
  1334. btrfs_set_super_num_devices(&root->fs_info->super_copy,
  1335. total_bytes + 1);
  1336. mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
  1337. if (seeding_dev) {
  1338. ret = init_first_rw_device(trans, root, device);
  1339. BUG_ON(ret);
  1340. ret = btrfs_finish_sprout(trans, root);
  1341. BUG_ON(ret);
  1342. } else {
  1343. ret = btrfs_add_device(trans, root, device);
  1344. }
  1345. /*
  1346. * we've got more storage, clear any full flags on the space
  1347. * infos
  1348. */
  1349. btrfs_clear_space_info_full(root->fs_info);
  1350. unlock_chunks(root);
  1351. btrfs_commit_transaction(trans, root);
  1352. if (seeding_dev) {
  1353. mutex_unlock(&uuid_mutex);
  1354. up_write(&sb->s_umount);
  1355. ret = btrfs_relocate_sys_chunks(root);
  1356. BUG_ON(ret);
  1357. }
  1358. out:
  1359. mutex_unlock(&root->fs_info->volume_mutex);
  1360. return ret;
  1361. error:
  1362. close_bdev_exclusive(bdev, 0);
  1363. if (seeding_dev) {
  1364. mutex_unlock(&uuid_mutex);
  1365. up_write(&sb->s_umount);
  1366. }
  1367. goto out;
  1368. }
  1369. static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
  1370. struct btrfs_device *device)
  1371. {
  1372. int ret;
  1373. struct btrfs_path *path;
  1374. struct btrfs_root *root;
  1375. struct btrfs_dev_item *dev_item;
  1376. struct extent_buffer *leaf;
  1377. struct btrfs_key key;
  1378. root = device->dev_root->fs_info->chunk_root;
  1379. path = btrfs_alloc_path();
  1380. if (!path)
  1381. return -ENOMEM;
  1382. key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
  1383. key.type = BTRFS_DEV_ITEM_KEY;
  1384. key.offset = device->devid;
  1385. ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
  1386. if (ret < 0)
  1387. goto out;
  1388. if (ret > 0) {
  1389. ret = -ENOENT;
  1390. goto out;
  1391. }
  1392. leaf = path->nodes[0];
  1393. dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
  1394. btrfs_set_device_id(leaf, dev_item, device->devid);
  1395. btrfs_set_device_type(leaf, dev_item, device->type);
  1396. btrfs_set_device_io_align(leaf, dev_item, device->io_align);
  1397. btrfs_set_device_io_width(leaf, dev_item, device->io_width);
  1398. btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
  1399. btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
  1400. btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
  1401. btrfs_mark_buffer_dirty(leaf);
  1402. out:
  1403. btrfs_free_path(path);
  1404. return ret;
  1405. }
  1406. static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
  1407. struct btrfs_device *device, u64 new_size)
  1408. {
  1409. struct btrfs_super_block *super_copy =
  1410. &device->dev_root->fs_info->super_copy;
  1411. u64 old_total = btrfs_super_total_bytes(super_copy);
  1412. u64 diff = new_size - device->total_bytes;
  1413. if (!device->writeable)
  1414. return -EACCES;
  1415. if (new_size <= device->total_bytes)
  1416. return -EINVAL;
  1417. btrfs_set_super_total_bytes(super_copy, old_total + diff);
  1418. device->fs_devices->total_rw_bytes += diff;
  1419. device->total_bytes = new_size;
  1420. device->disk_total_bytes = new_size;
  1421. btrfs_clear_space_info_full(device->dev_root->fs_info);
  1422. return btrfs_update_device(trans, device);
  1423. }
  1424. int btrfs_grow_device(struct btrfs_trans_handle *trans,
  1425. struct btrfs_device *device, u64 new_size)
  1426. {
  1427. int ret;
  1428. lock_chunks(device->dev_root);
  1429. ret = __btrfs_grow_device(trans, device, new_size);
  1430. unlock_chunks(device->dev_root);
  1431. return ret;
  1432. }
  1433. static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
  1434. struct btrfs_root *root,
  1435. u64 chunk_tree, u64 chunk_objectid,
  1436. u64 chunk_offset)
  1437. {
  1438. int ret;
  1439. struct btrfs_path *path;
  1440. struct btrfs_key key;
  1441. root = root->fs_info->chunk_root;
  1442. path = btrfs_alloc_path();
  1443. if (!path)
  1444. return -ENOMEM;
  1445. key.objectid = chunk_objectid;
  1446. key.offset = chunk_offset;
  1447. key.type = BTRFS_CHUNK_ITEM_KEY;
  1448. ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
  1449. BUG_ON(ret);
  1450. ret = btrfs_del_item(trans, root, path);
  1451. BUG_ON(ret);
  1452. btrfs_free_path(path);
  1453. return 0;
  1454. }
  1455. static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
  1456. chunk_offset)
  1457. {
  1458. struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
  1459. struct btrfs_disk_key *disk_key;
  1460. struct btrfs_chunk *chunk;
  1461. u8 *ptr;
  1462. int ret = 0;
  1463. u32 num_stripes;
  1464. u32 array_size;
  1465. u32 len = 0;
  1466. u32 cur;
  1467. struct btrfs_key key;
  1468. array_size = btrfs_super_sys_array_size(super_copy);
  1469. ptr = super_copy->sys_chunk_array;
  1470. cur = 0;
  1471. while (cur < array_size) {
  1472. disk_key = (struct btrfs_disk_key *)ptr;
  1473. btrfs_disk_key_to_cpu(&key, disk_key);
  1474. len = sizeof(*disk_key);
  1475. if (key.type == BTRFS_CHUNK_ITEM_KEY) {
  1476. chunk = (struct btrfs_chunk *)(ptr + len);
  1477. num_stripes = btrfs_stack_chunk_num_stripes(chunk);
  1478. len += btrfs_chunk_item_size(num_stripes);
  1479. } else {
  1480. ret = -EIO;
  1481. break;
  1482. }
  1483. if (key.objectid == chunk_objectid &&
  1484. key.offset == chunk_offset) {
  1485. memmove(ptr, ptr + len, array_size - (cur + len));
  1486. array_size -= len;
  1487. btrfs_set_super_sys_array_size(super_copy, array_size);
  1488. } else {
  1489. ptr += len;
  1490. cur += len;
  1491. }
  1492. }
  1493. return ret;
  1494. }
  1495. static int btrfs_relocate_chunk(struct btrfs_root *root,
  1496. u64 chunk_tree, u64 chunk_objectid,
  1497. u64 chunk_offset)
  1498. {
  1499. struct extent_map_tree *em_tree;
  1500. struct btrfs_root *extent_root;
  1501. struct btrfs_trans_handle *trans;
  1502. struct extent_map *em;
  1503. struct map_lookup *map;
  1504. int ret;
  1505. int i;
  1506. root = root->fs_info->chunk_root;
  1507. extent_root = root->fs_info->extent_root;
  1508. em_tree = &root->fs_info->mapping_tree.map_tree;
  1509. ret = btrfs_can_relocate(extent_root, chunk_offset);
  1510. if (ret)
  1511. return -ENOSPC;
  1512. /* step one, relocate all the extents inside this chunk */
  1513. ret = btrfs_relocate_block_group(extent_root, chunk_offset);
  1514. if (ret)
  1515. return ret;
  1516. trans = btrfs_start_transaction(root, 0);
  1517. BUG_ON(!trans);
  1518. lock_chunks(root);
  1519. /*
  1520. * step two, delete the device extents and the
  1521. * chunk tree entries
  1522. */
  1523. read_lock(&em_tree->lock);
  1524. em = lookup_extent_mapping(em_tree, chunk_offset, 1);
  1525. read_unlock(&em_tree->lock);
  1526. BUG_ON(em->start > chunk_offset ||
  1527. em->start + em->len < chunk_offset);
  1528. map = (struct map_lookup *)em->bdev;
  1529. for (i = 0; i < map->num_stripes; i++) {
  1530. ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
  1531. map->stripes[i].physical);
  1532. BUG_ON(ret);
  1533. if (map->stripes[i].dev) {
  1534. ret = btrfs_update_device(trans, map->stripes[i].dev);
  1535. BUG_ON(ret);
  1536. }
  1537. }
  1538. ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
  1539. chunk_offset);
  1540. BUG_ON(ret);
  1541. if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
  1542. ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
  1543. BUG_ON(ret);
  1544. }
  1545. ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
  1546. BUG_ON(ret);
  1547. write_lock(&em_tree->lock);
  1548. remove_extent_mapping(em_tree, em);
  1549. write_unlock(&em_tree->lock);
  1550. kfree(map);
  1551. em->bdev = NULL;
  1552. /* once for the tree */
  1553. free_extent_map(em);
  1554. /* once for us */
  1555. free_extent_map(em);
  1556. unlock_chunks(root);
  1557. btrfs_end_transaction(trans, root);
  1558. return 0;
  1559. }
  1560. static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
  1561. {
  1562. struct btrfs_root *chunk_root = root->fs_info->chunk_root;
  1563. struct btrfs_path *path;
  1564. struct extent_buffer *leaf;
  1565. struct btrfs_chunk *chunk;
  1566. struct btrfs_key key;
  1567. struct btrfs_key found_key;
  1568. u64 chunk_tree = chunk_root->root_key.objectid;
  1569. u64 chunk_type;
  1570. bool retried = false;
  1571. int failed = 0;
  1572. int ret;
  1573. path = btrfs_alloc_path();
  1574. if (!path)
  1575. return -ENOMEM;
  1576. again:
  1577. key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
  1578. key.offset = (u64)-1;
  1579. key.type = BTRFS_CHUNK_ITEM_KEY;
  1580. while (1) {
  1581. ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
  1582. if (ret < 0)
  1583. goto error;
  1584. BUG_ON(ret == 0);
  1585. ret = btrfs_previous_item(chunk_root, path, key.objectid,
  1586. key.type);
  1587. if (ret < 0)
  1588. goto error;
  1589. if (ret > 0)
  1590. break;
  1591. leaf = path->nodes[0];
  1592. btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
  1593. chunk = btrfs_item_ptr(leaf, path->slots[0],
  1594. struct btrfs_chunk);
  1595. chunk_type = btrfs_chunk_type(leaf, chunk);
  1596. btrfs_release_path(chunk_root, path);
  1597. if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
  1598. ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
  1599. found_key.objectid,
  1600. found_key.offset);
  1601. if (ret == -ENOSPC)
  1602. failed++;
  1603. else if (ret)
  1604. BUG();
  1605. }
  1606. if (found_key.offset == 0)
  1607. break;
  1608. key.offset = found_key.offset - 1;
  1609. }
  1610. ret = 0;
  1611. if (failed && !retried) {
  1612. failed = 0;
  1613. retried = true;
  1614. goto again;
  1615. } else if (failed && retried) {
  1616. WARN_ON(1);
  1617. ret = -ENOSPC;
  1618. }
  1619. error:
  1620. btrfs_free_path(path);
  1621. return ret;
  1622. }
  1623. static u64 div_factor(u64 num, int factor)
  1624. {
  1625. if (factor == 10)
  1626. return num;
  1627. num *= factor;
  1628. do_div(num, 10);
  1629. return num;
  1630. }
  1631. int btrfs_balance(struct btrfs_root *dev_root)
  1632. {
  1633. int ret;
  1634. struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
  1635. struct btrfs_device *device;
  1636. u64 old_size;
  1637. u64 size_to_free;
  1638. struct btrfs_path *path;
  1639. struct btrfs_key key;
  1640. struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
  1641. struct btrfs_trans_handle *trans;
  1642. struct btrfs_key found_key;
  1643. if (dev_root->fs_info->sb->s_flags & MS_RDONLY)
  1644. return -EROFS;
  1645. mutex_lock(&dev_root->fs_info->volume_mutex);
  1646. dev_root = dev_root->fs_info->dev_root;
  1647. /* step one make some room on all the devices */
  1648. list_for_each_entry(device, devices, dev_list) {
  1649. old_size = device->total_bytes;
  1650. size_to_free = div_factor(old_size, 1);
  1651. size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
  1652. if (!device->writeable ||
  1653. device->total_bytes - device->bytes_used > size_to_free)
  1654. continue;
  1655. ret = btrfs_shrink_device(device, old_size - size_to_free);
  1656. if (ret == -ENOSPC)
  1657. break;
  1658. BUG_ON(ret);
  1659. trans = btrfs_start_transaction(dev_root, 0);
  1660. BUG_ON(!trans);
  1661. ret = btrfs_grow_device(trans, device, old_size);
  1662. BUG_ON(ret);
  1663. btrfs_end_transaction(trans, dev_root);
  1664. }
  1665. /* step two, relocate all the chunks */
  1666. path = btrfs_alloc_path();
  1667. BUG_ON(!path);
  1668. key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
  1669. key.offset = (u64)-1;
  1670. key.type = BTRFS_CHUNK_ITEM_KEY;
  1671. while (1) {
  1672. ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
  1673. if (ret < 0)
  1674. goto error;
  1675. /*
  1676. * this shouldn't happen, it means the last relocate
  1677. * failed
  1678. */
  1679. if (ret == 0)
  1680. break;
  1681. ret = btrfs_previous_item(chunk_root, path, 0,
  1682. BTRFS_CHUNK_ITEM_KEY);
  1683. if (ret)
  1684. break;
  1685. btrfs_item_key_to_cpu(path->nodes[0], &found_key,
  1686. path->slots[0]);
  1687. if (found_key.objectid != key.objectid)
  1688. break;
  1689. /* chunk zero is special */
  1690. if (found_key.offset == 0)
  1691. break;
  1692. btrfs_release_path(chunk_root, path);
  1693. ret = btrfs_relocate_chunk(chunk_root,
  1694. chunk_root->root_key.objectid,
  1695. found_key.objectid,
  1696. found_key.offset);
  1697. BUG_ON(ret && ret != -ENOSPC);
  1698. key.offset = found_key.offset - 1;
  1699. }
  1700. ret = 0;
  1701. error:
  1702. btrfs_free_path(path);
  1703. mutex_unlock(&dev_root->fs_info->volume_mutex);
  1704. return ret;
  1705. }
  1706. /*
  1707. * shrinking a device means finding all of the device extents past
  1708. * the new size, and then following the back refs to the chunks.
  1709. * The chunk relocation code actually frees the device extent
  1710. */
  1711. int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
  1712. {
  1713. struct btrfs_trans_handle *trans;
  1714. struct btrfs_root *root = device->dev_root;
  1715. struct btrfs_dev_extent *dev_extent = NULL;
  1716. struct btrfs_path *path;
  1717. u64 length;
  1718. u64 chunk_tree;
  1719. u64 chunk_objectid;
  1720. u64 chunk_offset;
  1721. int ret;
  1722. int slot;
  1723. int failed = 0;
  1724. bool retried = false;
  1725. struct extent_buffer *l;
  1726. struct btrfs_key key;
  1727. struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
  1728. u64 old_total = btrfs_super_total_bytes(super_copy);
  1729. u64 old_size = device->total_bytes;
  1730. u64 diff = device->total_bytes - new_size;
  1731. if (new_size >= device->total_bytes)
  1732. return -EINVAL;
  1733. path = btrfs_alloc_path();
  1734. if (!path)
  1735. return -ENOMEM;
  1736. path->reada = 2;
  1737. lock_chunks(root);
  1738. device->total_bytes = new_size;
  1739. if (device->writeable)
  1740. device->fs_devices->total_rw_bytes -= diff;
  1741. unlock_chunks(root);
  1742. again:
  1743. key.objectid = device->devid;
  1744. key.offset = (u64)-1;
  1745. key.type = BTRFS_DEV_EXTENT_KEY;
  1746. while (1) {
  1747. ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
  1748. if (ret < 0)
  1749. goto done;
  1750. ret = btrfs_previous_item(root, path, 0, key.type);
  1751. if (ret < 0)
  1752. goto done;
  1753. if (ret) {
  1754. ret = 0;
  1755. btrfs_release_path(root, path);
  1756. break;
  1757. }
  1758. l = path->nodes[0];
  1759. slot = path->slots[0];
  1760. btrfs_item_key_to_cpu(l, &key, path->slots[0]);
  1761. if (key.objectid != device->devid) {
  1762. btrfs_release_path(root, path);
  1763. break;
  1764. }
  1765. dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
  1766. length = btrfs_dev_extent_length(l, dev_extent);
  1767. if (key.offset + length <= new_size) {
  1768. btrfs_release_path(root, path);
  1769. break;
  1770. }
  1771. chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
  1772. chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
  1773. chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
  1774. btrfs_release_path(root, path);
  1775. ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
  1776. chunk_offset);
  1777. if (ret && ret != -ENOSPC)
  1778. goto done;
  1779. if (ret == -ENOSPC)
  1780. failed++;
  1781. key.offset -= 1;
  1782. }
  1783. if (failed && !retried) {
  1784. failed = 0;
  1785. retried = true;
  1786. goto again;
  1787. } else if (failed && retried) {
  1788. ret = -ENOSPC;
  1789. lock_chunks(root);
  1790. device->total_bytes = old_size;
  1791. if (device->writeable)
  1792. device->fs_devices->total_rw_bytes += diff;
  1793. unlock_chunks(root);
  1794. goto done;
  1795. }
  1796. /* Shrinking succeeded, else we would be at "done". */
  1797. trans = btrfs_start_transaction(root, 0);
  1798. lock_chunks(root);
  1799. device->disk_total_bytes = new_size;
  1800. /* Now btrfs_update_device() will change the on-disk size. */
  1801. ret = btrfs_update_device(trans, device);
  1802. if (ret) {
  1803. unlock_chunks(root);
  1804. btrfs_end_transaction(trans, root);
  1805. goto done;
  1806. }
  1807. WARN_ON(diff > old_total);
  1808. btrfs_set_super_total_bytes(super_copy, old_total - diff);
  1809. unlock_chunks(root);
  1810. btrfs_end_transaction(trans, root);
  1811. done:
  1812. btrfs_free_path(path);
  1813. return ret;
  1814. }
  1815. static int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
  1816. struct btrfs_root *root,
  1817. struct btrfs_key *key,
  1818. struct btrfs_chunk *chunk, int item_size)
  1819. {
  1820. struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
  1821. struct btrfs_disk_key disk_key;
  1822. u32 array_size;
  1823. u8 *ptr;
  1824. array_size = btrfs_super_sys_array_size(super_copy);
  1825. if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
  1826. return -EFBIG;
  1827. ptr = super_copy->sys_chunk_array + array_size;
  1828. btrfs_cpu_key_to_disk(&disk_key, key);
  1829. memcpy(ptr, &disk_key, sizeof(disk_key));
  1830. ptr += sizeof(disk_key);
  1831. memcpy(ptr, chunk, item_size);
  1832. item_size += sizeof(disk_key);
  1833. btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
  1834. return 0;
  1835. }
  1836. static noinline u64 chunk_bytes_by_type(u64 type, u64 calc_size,
  1837. int num_stripes, int sub_stripes)
  1838. {
  1839. if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
  1840. return calc_size;
  1841. else if (type & BTRFS_BLOCK_GROUP_RAID10)
  1842. return calc_size * (num_stripes / sub_stripes);
  1843. else
  1844. return calc_size * num_stripes;
  1845. }
  1846. static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
  1847. struct btrfs_root *extent_root,
  1848. struct map_lookup **map_ret,
  1849. u64 *num_bytes, u64 *stripe_size,
  1850. u64 start, u64 type)
  1851. {
  1852. struct btrfs_fs_info *info = extent_root->fs_info;
  1853. struct btrfs_device *device = NULL;
  1854. struct btrfs_fs_devices *fs_devices = info->fs_devices;
  1855. struct list_head *cur;
  1856. struct map_lookup *map = NULL;
  1857. struct extent_map_tree *em_tree;
  1858. struct extent_map *em;
  1859. struct list_head private_devs;
  1860. int min_stripe_size = 1 * 1024 * 1024;
  1861. u64 calc_size = 1024 * 1024 * 1024;
  1862. u64 max_chunk_size = calc_size;
  1863. u64 min_free;
  1864. u64 avail;
  1865. u64 max_avail = 0;
  1866. u64 dev_offset;
  1867. int num_stripes = 1;
  1868. int min_stripes = 1;
  1869. int sub_stripes = 0;
  1870. int looped = 0;
  1871. int ret;
  1872. int index;
  1873. int stripe_len = 64 * 1024;
  1874. if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
  1875. (type & BTRFS_BLOCK_GROUP_DUP)) {
  1876. WARN_ON(1);
  1877. type &= ~BTRFS_BLOCK_GROUP_DUP;
  1878. }
  1879. if (list_empty(&fs_devices->alloc_list))
  1880. return -ENOSPC;
  1881. if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
  1882. num_stripes = fs_devices->rw_devices;
  1883. min_stripes = 2;
  1884. }
  1885. if (type & (BTRFS_BLOCK_GROUP_DUP)) {
  1886. num_stripes = 2;
  1887. min_stripes = 2;
  1888. }
  1889. if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
  1890. if (fs_devices->rw_devices < 2)
  1891. return -ENOSPC;
  1892. num_stripes = 2;
  1893. min_stripes = 2;
  1894. }
  1895. if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
  1896. num_stripes = fs_devices->rw_devices;
  1897. if (num_stripes < 4)
  1898. return -ENOSPC;
  1899. num_stripes &= ~(u32)1;
  1900. sub_stripes = 2;
  1901. min_stripes = 4;
  1902. }
  1903. if (type & BTRFS_BLOCK_GROUP_DATA) {
  1904. max_chunk_size = 10 * calc_size;
  1905. min_stripe_size = 64 * 1024 * 1024;
  1906. } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
  1907. max_chunk_size = 256 * 1024 * 1024;
  1908. min_stripe_size = 32 * 1024 * 1024;
  1909. } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
  1910. calc_size = 8 * 1024 * 1024;
  1911. max_chunk_size = calc_size * 2;
  1912. min_stripe_size = 1 * 1024 * 1024;
  1913. }
  1914. /* we don't want a chunk larger than 10% of writeable space */
  1915. max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
  1916. max_chunk_size);
  1917. again:
  1918. max_avail = 0;
  1919. if (!map || map->num_stripes != num_stripes) {
  1920. kfree(map);
  1921. map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
  1922. if (!map)
  1923. return -ENOMEM;
  1924. map->num_stripes = num_stripes;
  1925. }
  1926. if (calc_size * num_stripes > max_chunk_size) {
  1927. calc_size = max_chunk_size;
  1928. do_div(calc_size, num_stripes);
  1929. do_div(calc_size, stripe_len);
  1930. calc_size *= stripe_len;
  1931. }
  1932. /* we don't want tiny stripes */
  1933. if (!looped)
  1934. calc_size = max_t(u64, min_stripe_size, calc_size);
  1935. /*
  1936. * we're about to do_div by the stripe_len so lets make sure
  1937. * we end up with something bigger than a stripe
  1938. */
  1939. calc_size = max_t(u64, calc_size, stripe_len * 4);
  1940. do_div(calc_size, stripe_len);
  1941. calc_size *= stripe_len;
  1942. cur = fs_devices->alloc_list.next;
  1943. index = 0;
  1944. if (type & BTRFS_BLOCK_GROUP_DUP)
  1945. min_free = calc_size * 2;
  1946. else
  1947. min_free = calc_size;
  1948. /*
  1949. * we add 1MB because we never use the first 1MB of the device, unless
  1950. * we've looped, then we are likely allocating the maximum amount of
  1951. * space left already
  1952. */
  1953. if (!looped)
  1954. min_free += 1024 * 1024;
  1955. INIT_LIST_HEAD(&private_devs);
  1956. while (index < num_stripes) {
  1957. device = list_entry(cur, struct btrfs_device, dev_alloc_list);
  1958. BUG_ON(!device->writeable);
  1959. if (device->total_bytes > device->bytes_used)
  1960. avail = device->total_bytes - device->bytes_used;
  1961. else
  1962. avail = 0;
  1963. cur = cur->next;
  1964. if (device->in_fs_metadata && avail >= min_free) {
  1965. ret = find_free_dev_extent(trans, device,
  1966. min_free, &dev_offset,
  1967. &max_avail);
  1968. if (ret == 0) {
  1969. list_move_tail(&device->dev_alloc_list,
  1970. &private_devs);
  1971. map->stripes[index].dev = device;
  1972. map->stripes[index].physical = dev_offset;
  1973. index++;
  1974. if (type & BTRFS_BLOCK_GROUP_DUP) {
  1975. map->stripes[index].dev = device;
  1976. map->stripes[index].physical =
  1977. dev_offset + calc_size;
  1978. index++;
  1979. }
  1980. }
  1981. } else if (device->in_fs_metadata && avail > max_avail)
  1982. max_avail = avail;
  1983. if (cur == &fs_devices->alloc_list)
  1984. break;
  1985. }
  1986. list_splice(&private_devs, &fs_devices->alloc_list);
  1987. if (index < num_stripes) {
  1988. if (index >= min_stripes) {
  1989. num_stripes = index;
  1990. if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
  1991. num_stripes /= sub_stripes;
  1992. num_stripes *= sub_stripes;
  1993. }
  1994. looped = 1;
  1995. goto again;
  1996. }
  1997. if (!looped && max_avail > 0) {
  1998. looped = 1;
  1999. calc_size = max_avail;
  2000. goto again;
  2001. }
  2002. kfree(map);
  2003. return -ENOSPC;
  2004. }
  2005. map->sector_size = extent_root->sectorsize;
  2006. map->stripe_len = stripe_len;
  2007. map->io_align = stripe_len;
  2008. map->io_width = stripe_len;
  2009. map->type = type;
  2010. map->num_stripes = num_stripes;
  2011. map->sub_stripes = sub_stripes;
  2012. *map_ret = map;
  2013. *stripe_size = calc_size;
  2014. *num_bytes = chunk_bytes_by_type(type, calc_size,
  2015. num_stripes, sub_stripes);
  2016. em = alloc_extent_map(GFP_NOFS);
  2017. if (!em) {
  2018. kfree(map);
  2019. return -ENOMEM;
  2020. }
  2021. em->bdev = (struct block_device *)map;
  2022. em->start = start;
  2023. em->len = *num_bytes;
  2024. em->block_start = 0;
  2025. em->block_len = em->len;
  2026. em_tree = &extent_root->fs_info->mapping_tree.map_tree;
  2027. write_lock(&em_tree->lock);
  2028. ret = add_extent_mapping(em_tree, em);
  2029. write_unlock(&em_tree->lock);
  2030. BUG_ON(ret);
  2031. free_extent_map(em);
  2032. ret = btrfs_make_block_group(trans, extent_root, 0, type,
  2033. BTRFS_FIRST_CHUNK_TREE_OBJECTID,
  2034. start, *num_bytes);
  2035. BUG_ON(ret);
  2036. index = 0;
  2037. while (index < map->num_stripes) {
  2038. device = map->stripes[index].dev;
  2039. dev_offset = map->stripes[index].physical;
  2040. ret = btrfs_alloc_dev_extent(trans, device,
  2041. info->chunk_root->root_key.objectid,
  2042. BTRFS_FIRST_CHUNK_TREE_OBJECTID,
  2043. start, dev_offset, calc_size);
  2044. BUG_ON(ret);
  2045. index++;
  2046. }
  2047. return 0;
  2048. }
  2049. static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
  2050. struct btrfs_root *extent_root,
  2051. struct map_lookup *map, u64 chunk_offset,
  2052. u64 chunk_size, u64 stripe_size)
  2053. {
  2054. u64 dev_offset;
  2055. struct btrfs_key key;
  2056. struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
  2057. struct btrfs_device *device;
  2058. struct btrfs_chunk *chunk;
  2059. struct btrfs_stripe *stripe;
  2060. size_t item_size = btrfs_chunk_item_size(map->num_stripes);
  2061. int index = 0;
  2062. int ret;
  2063. chunk = kzalloc(item_size, GFP_NOFS);
  2064. if (!chunk)
  2065. return -ENOMEM;
  2066. index = 0;
  2067. while (index < map->num_stripes) {
  2068. device = map->stripes[index].dev;
  2069. device->bytes_used += stripe_size;
  2070. ret = btrfs_update_device(trans, device);
  2071. BUG_ON(ret);
  2072. index++;
  2073. }
  2074. index = 0;
  2075. stripe = &chunk->stripe;
  2076. while (index < map->num_stripes) {
  2077. device = map->stripes[index].dev;
  2078. dev_offset = map->stripes[index].physical;
  2079. btrfs_set_stack_stripe_devid(stripe, device->devid);
  2080. btrfs_set_stack_stripe_offset(stripe, dev_offset);
  2081. memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
  2082. stripe++;
  2083. index++;
  2084. }
  2085. btrfs_set_stack_chunk_length(chunk, chunk_size);
  2086. btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
  2087. btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
  2088. btrfs_set_stack_chunk_type(chunk, map->type);
  2089. btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
  2090. btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
  2091. btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
  2092. btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
  2093. btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
  2094. key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
  2095. key.type = BTRFS_CHUNK_ITEM_KEY;
  2096. key.offset = chunk_offset;
  2097. ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
  2098. BUG_ON(ret);
  2099. if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
  2100. ret = btrfs_add_system_chunk(trans, chunk_root, &key, chunk,
  2101. item_size);
  2102. BUG_ON(ret);
  2103. }
  2104. kfree(chunk);
  2105. return 0;
  2106. }
  2107. /*
  2108. * Chunk allocation falls into two parts. The first part does works
  2109. * that make the new allocated chunk useable, but not do any operation
  2110. * that modifies the chunk tree. The second part does the works that
  2111. * require modifying the chunk tree. This division is important for the
  2112. * bootstrap process of adding storage to a seed btrfs.
  2113. */
  2114. int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
  2115. struct btrfs_root *extent_root, u64 type)
  2116. {
  2117. u64 chunk_offset;
  2118. u64 chunk_size;
  2119. u64 stripe_size;
  2120. struct map_lookup *map;
  2121. struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
  2122. int ret;
  2123. ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
  2124. &chunk_offset);
  2125. if (ret)
  2126. return ret;
  2127. ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
  2128. &stripe_size, chunk_offset, type);
  2129. if (ret)
  2130. return ret;
  2131. ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
  2132. chunk_size, stripe_size);
  2133. BUG_ON(ret);
  2134. return 0;
  2135. }
  2136. static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
  2137. struct btrfs_root *root,
  2138. struct btrfs_device *device)
  2139. {
  2140. u64 chunk_offset;
  2141. u64 sys_chunk_offset;
  2142. u64 chunk_size;
  2143. u64 sys_chunk_size;
  2144. u64 stripe_size;
  2145. u64 sys_stripe_size;
  2146. u64 alloc_profile;
  2147. struct map_lookup *map;
  2148. struct map_lookup *sys_map;
  2149. struct btrfs_fs_info *fs_info = root->fs_info;
  2150. struct btrfs_root *extent_root = fs_info->extent_root;
  2151. int ret;
  2152. ret = find_next_chunk(fs_info->chunk_root,
  2153. BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
  2154. BUG_ON(ret);
  2155. alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
  2156. (fs_info->metadata_alloc_profile &
  2157. fs_info->avail_metadata_alloc_bits);
  2158. alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
  2159. ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
  2160. &stripe_size, chunk_offset, alloc_profile);
  2161. BUG_ON(ret);
  2162. sys_chunk_offset = chunk_offset + chunk_size;
  2163. alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
  2164. (fs_info->system_alloc_profile &
  2165. fs_info->avail_system_alloc_bits);
  2166. alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
  2167. ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
  2168. &sys_chunk_size, &sys_stripe_size,
  2169. sys_chunk_offset, alloc_profile);
  2170. BUG_ON(ret);
  2171. ret = btrfs_add_device(trans, fs_info->chunk_root, device);
  2172. BUG_ON(ret);
  2173. /*
  2174. * Modifying chunk tree needs allocating new blocks from both
  2175. * system block group and metadata block group. So we only can
  2176. * do operations require modifying the chunk tree after both
  2177. * block groups were created.
  2178. */
  2179. ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
  2180. chunk_size, stripe_size);
  2181. BUG_ON(ret);
  2182. ret = __finish_chunk_alloc(trans, extent_root, sys_map,
  2183. sys_chunk_offset, sys_chunk_size,
  2184. sys_stripe_size);
  2185. BUG_ON(ret);
  2186. return 0;
  2187. }
  2188. int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
  2189. {
  2190. struct extent_map *em;
  2191. struct map_lookup *map;
  2192. struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
  2193. int readonly = 0;
  2194. int i;
  2195. read_lock(&map_tree->map_tree.lock);
  2196. em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
  2197. read_unlock(&map_tree->map_tree.lock);
  2198. if (!em)
  2199. return 1;
  2200. if (btrfs_test_opt(root, DEGRADED)) {
  2201. free_extent_map(em);
  2202. return 0;
  2203. }
  2204. map = (struct map_lookup *)em->bdev;
  2205. for (i = 0; i < map->num_stripes; i++) {
  2206. if (!map->stripes[i].dev->writeable) {
  2207. readonly = 1;
  2208. break;
  2209. }
  2210. }
  2211. free_extent_map(em);
  2212. return readonly;
  2213. }
  2214. void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
  2215. {
  2216. extent_map_tree_init(&tree->map_tree, GFP_NOFS);
  2217. }
  2218. void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
  2219. {
  2220. struct extent_map *em;
  2221. while (1) {
  2222. write_lock(&tree->map_tree.lock);
  2223. em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
  2224. if (em)
  2225. remove_extent_mapping(&tree->map_tree, em);
  2226. write_unlock(&tree->map_tree.lock);
  2227. if (!em)
  2228. break;
  2229. kfree(em->bdev);
  2230. /* once for us */
  2231. free_extent_map(em);
  2232. /* once for the tree */
  2233. free_extent_map(em);
  2234. }
  2235. }
  2236. int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
  2237. {
  2238. struct extent_map *em;
  2239. struct map_lookup *map;
  2240. struct extent_map_tree *em_tree = &map_tree->map_tree;
  2241. int ret;
  2242. read_lock(&em_tree->lock);
  2243. em = lookup_extent_mapping(em_tree, logical, len);
  2244. read_unlock(&em_tree->lock);
  2245. BUG_ON(!em);
  2246. BUG_ON(em->start > logical || em->start + em->len < logical);
  2247. map = (struct map_lookup *)em->bdev;
  2248. if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
  2249. ret = map->num_stripes;
  2250. else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
  2251. ret = map->sub_stripes;
  2252. else
  2253. ret = 1;
  2254. free_extent_map(em);
  2255. return ret;
  2256. }
  2257. static int find_live_mirror(struct map_lookup *map, int first, int num,
  2258. int optimal)
  2259. {
  2260. int i;
  2261. if (map->stripes[optimal].dev->bdev)
  2262. return optimal;
  2263. for (i = first; i < first + num; i++) {
  2264. if (map->stripes[i].dev->bdev)
  2265. return i;
  2266. }
  2267. /* we couldn't find one that doesn't fail. Just return something
  2268. * and the io error handling code will clean up eventually
  2269. */
  2270. return optimal;
  2271. }
  2272. static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
  2273. u64 logical, u64 *length,
  2274. struct btrfs_multi_bio **multi_ret,
  2275. int mirror_num, struct page *unplug_page)
  2276. {
  2277. struct extent_map *em;
  2278. struct map_lookup *map;
  2279. struct extent_map_tree *em_tree = &map_tree->map_tree;
  2280. u64 offset;
  2281. u64 stripe_offset;
  2282. u64 stripe_nr;
  2283. int stripes_allocated = 8;
  2284. int stripes_required = 1;
  2285. int stripe_index;
  2286. int i;
  2287. int num_stripes;
  2288. int max_errors = 0;
  2289. struct btrfs_multi_bio *multi = NULL;
  2290. if (multi_ret && !(rw & REQ_WRITE))
  2291. stripes_allocated = 1;
  2292. again:
  2293. if (multi_ret) {
  2294. multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
  2295. GFP_NOFS);
  2296. if (!multi)
  2297. return -ENOMEM;
  2298. atomic_set(&multi->error, 0);
  2299. }
  2300. read_lock(&em_tree->lock);
  2301. em = lookup_extent_mapping(em_tree, logical, *length);
  2302. read_unlock(&em_tree->lock);
  2303. if (!em && unplug_page) {
  2304. kfree(multi);
  2305. return 0;
  2306. }
  2307. if (!em) {
  2308. printk(KERN_CRIT "unable to find logical %llu len %llu\n",
  2309. (unsigned long long)logical,
  2310. (unsigned long long)*length);
  2311. BUG();
  2312. }
  2313. BUG_ON(em->start > logical || em->start + em->len < logical);
  2314. map = (struct map_lookup *)em->bdev;
  2315. offset = logical - em->start;
  2316. if (mirror_num > map->num_stripes)
  2317. mirror_num = 0;
  2318. /* if our multi bio struct is too small, back off and try again */
  2319. if (rw & REQ_WRITE) {
  2320. if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
  2321. BTRFS_BLOCK_GROUP_DUP)) {
  2322. stripes_required = map->num_stripes;
  2323. max_errors = 1;
  2324. } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
  2325. stripes_required = map->sub_stripes;
  2326. max_errors = 1;
  2327. }
  2328. }
  2329. if (multi_ret && (rw & REQ_WRITE) &&
  2330. stripes_allocated < stripes_required) {
  2331. stripes_allocated = map->num_stripes;
  2332. free_extent_map(em);
  2333. kfree(multi);
  2334. goto again;
  2335. }
  2336. stripe_nr = offset;
  2337. /*
  2338. * stripe_nr counts the total number of stripes we have to stride
  2339. * to get to this block
  2340. */
  2341. do_div(stripe_nr, map->stripe_len);
  2342. stripe_offset = stripe_nr * map->stripe_len;
  2343. BUG_ON(offset < stripe_offset);
  2344. /* stripe_offset is the offset of this block in its stripe*/
  2345. stripe_offset = offset - stripe_offset;
  2346. if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
  2347. BTRFS_BLOCK_GROUP_RAID10 |
  2348. BTRFS_BLOCK_GROUP_DUP)) {
  2349. /* we limit the length of each bio to what fits in a stripe */
  2350. *length = min_t(u64, em->len - offset,
  2351. map->stripe_len - stripe_offset);
  2352. } else {
  2353. *length = em->len - offset;
  2354. }
  2355. if (!multi_ret && !unplug_page)
  2356. goto out;
  2357. num_stripes = 1;
  2358. stripe_index = 0;
  2359. if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
  2360. if (unplug_page || (rw & REQ_WRITE))
  2361. num_stripes = map->num_stripes;
  2362. else if (mirror_num)
  2363. stripe_index = mirror_num - 1;
  2364. else {
  2365. stripe_index = find_live_mirror(map, 0,
  2366. map->num_stripes,
  2367. current->pid % map->num_stripes);
  2368. }
  2369. } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
  2370. if (rw & REQ_WRITE)
  2371. num_stripes = map->num_stripes;
  2372. else if (mirror_num)
  2373. stripe_index = mirror_num - 1;
  2374. } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
  2375. int factor = map->num_stripes / map->sub_stripes;
  2376. stripe_index = do_div(stripe_nr, factor);
  2377. stripe_index *= map->sub_stripes;
  2378. if (unplug_page || (rw & REQ_WRITE))
  2379. num_stripes = map->sub_stripes;
  2380. else if (mirror_num)
  2381. stripe_index += mirror_num - 1;
  2382. else {
  2383. stripe_index = find_live_mirror(map, stripe_index,
  2384. map->sub_stripes, stripe_index +
  2385. current->pid % map->sub_stripes);
  2386. }
  2387. } else {
  2388. /*
  2389. * after this do_div call, stripe_nr is the number of stripes
  2390. * on this device we have to walk to find the data, and
  2391. * stripe_index is the number of our device in the stripe array
  2392. */
  2393. stripe_index = do_div(stripe_nr, map->num_stripes);
  2394. }
  2395. BUG_ON(stripe_index >= map->num_stripes);
  2396. for (i = 0; i < num_stripes; i++) {
  2397. if (unplug_page) {
  2398. struct btrfs_device *device;
  2399. struct backing_dev_info *bdi;
  2400. device = map->stripes[stripe_index].dev;
  2401. if (device->bdev) {
  2402. bdi = blk_get_backing_dev_info(device->bdev);
  2403. if (bdi->unplug_io_fn)
  2404. bdi->unplug_io_fn(bdi, unplug_page);
  2405. }
  2406. } else {
  2407. multi->stripes[i].physical =
  2408. map->stripes[stripe_index].physical +
  2409. stripe_offset + stripe_nr * map->stripe_len;
  2410. multi->stripes[i].dev = map->stripes[stripe_index].dev;
  2411. }
  2412. stripe_index++;
  2413. }
  2414. if (multi_ret) {
  2415. *multi_ret = multi;
  2416. multi->num_stripes = num_stripes;
  2417. multi->max_errors = max_errors;
  2418. }
  2419. out:
  2420. free_extent_map(em);
  2421. return 0;
  2422. }
  2423. int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
  2424. u64 logical, u64 *length,
  2425. struct btrfs_multi_bio **multi_ret, int mirror_num)
  2426. {
  2427. return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
  2428. mirror_num, NULL);
  2429. }
  2430. int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
  2431. u64 chunk_start, u64 physical, u64 devid,
  2432. u64 **logical, int *naddrs, int *stripe_len)
  2433. {
  2434. struct extent_map_tree *em_tree = &map_tree->map_tree;
  2435. struct extent_map *em;
  2436. struct map_lookup *map;
  2437. u64 *buf;
  2438. u64 bytenr;
  2439. u64 length;
  2440. u64 stripe_nr;
  2441. int i, j, nr = 0;
  2442. read_lock(&em_tree->lock);
  2443. em = lookup_extent_mapping(em_tree, chunk_start, 1);
  2444. read_unlock(&em_tree->lock);
  2445. BUG_ON(!em || em->start != chunk_start);
  2446. map = (struct map_lookup *)em->bdev;
  2447. length = em->len;
  2448. if (map->type & BTRFS_BLOCK_GROUP_RAID10)
  2449. do_div(length, map->num_stripes / map->sub_stripes);
  2450. else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
  2451. do_div(length, map->num_stripes);
  2452. buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
  2453. BUG_ON(!buf);
  2454. for (i = 0; i < map->num_stripes; i++) {
  2455. if (devid && map->stripes[i].dev->devid != devid)
  2456. continue;
  2457. if (map->stripes[i].physical > physical ||
  2458. map->stripes[i].physical + length <= physical)
  2459. continue;
  2460. stripe_nr = physical - map->stripes[i].physical;
  2461. do_div(stripe_nr, map->stripe_len);
  2462. if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
  2463. stripe_nr = stripe_nr * map->num_stripes + i;
  2464. do_div(stripe_nr, map->sub_stripes);
  2465. } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
  2466. stripe_nr = stripe_nr * map->num_stripes + i;
  2467. }
  2468. bytenr = chunk_start + stripe_nr * map->stripe_len;
  2469. WARN_ON(nr >= map->num_stripes);
  2470. for (j = 0; j < nr; j++) {
  2471. if (buf[j] == bytenr)
  2472. break;
  2473. }
  2474. if (j == nr) {
  2475. WARN_ON(nr >= map->num_stripes);
  2476. buf[nr++] = bytenr;
  2477. }
  2478. }
  2479. *logical = buf;
  2480. *naddrs = nr;
  2481. *stripe_len = map->stripe_len;
  2482. free_extent_map(em);
  2483. return 0;
  2484. }
  2485. int btrfs_unplug_page(struct btrfs_mapping_tree *map_tree,
  2486. u64 logical, struct page *page)
  2487. {
  2488. u64 length = PAGE_CACHE_SIZE;
  2489. return __btrfs_map_block(map_tree, READ, logical, &length,
  2490. NULL, 0, page);
  2491. }
  2492. static void end_bio_multi_stripe(struct bio *bio, int err)
  2493. {
  2494. struct btrfs_multi_bio *multi = bio->bi_private;
  2495. int is_orig_bio = 0;
  2496. if (err)
  2497. atomic_inc(&multi->error);
  2498. if (bio == multi->orig_bio)
  2499. is_orig_bio = 1;
  2500. if (atomic_dec_and_test(&multi->stripes_pending)) {
  2501. if (!is_orig_bio) {
  2502. bio_put(bio);
  2503. bio = multi->orig_bio;
  2504. }
  2505. bio->bi_private = multi->private;
  2506. bio->bi_end_io = multi->end_io;
  2507. /* only send an error to the higher layers if it is
  2508. * beyond the tolerance of the multi-bio
  2509. */
  2510. if (atomic_read(&multi->error) > multi->max_errors) {
  2511. err = -EIO;
  2512. } else if (err) {
  2513. /*
  2514. * this bio is actually up to date, we didn't
  2515. * go over the max number of errors
  2516. */
  2517. set_bit(BIO_UPTODATE, &bio->bi_flags);
  2518. err = 0;
  2519. }
  2520. kfree(multi);
  2521. bio_endio(bio, err);
  2522. } else if (!is_orig_bio) {
  2523. bio_put(bio);
  2524. }
  2525. }
  2526. struct async_sched {
  2527. struct bio *bio;
  2528. int rw;
  2529. struct btrfs_fs_info *info;
  2530. struct btrfs_work work;
  2531. };
  2532. /*
  2533. * see run_scheduled_bios for a description of why bios are collected for
  2534. * async submit.
  2535. *
  2536. * This will add one bio to the pending list for a device and make sure
  2537. * the work struct is scheduled.
  2538. */
  2539. static noinline int schedule_bio(struct btrfs_root *root,
  2540. struct btrfs_device *device,
  2541. int rw, struct bio *bio)
  2542. {
  2543. int should_queue = 1;
  2544. struct btrfs_pending_bios *pending_bios;
  2545. /* don't bother with additional async steps for reads, right now */
  2546. if (!(rw & REQ_WRITE)) {
  2547. bio_get(bio);
  2548. submit_bio(rw, bio);
  2549. bio_put(bio);
  2550. return 0;
  2551. }
  2552. /*
  2553. * nr_async_bios allows us to reliably return congestion to the
  2554. * higher layers. Otherwise, the async bio makes it appear we have
  2555. * made progress against dirty pages when we've really just put it
  2556. * on a queue for later
  2557. */
  2558. atomic_inc(&root->fs_info->nr_async_bios);
  2559. WARN_ON(bio->bi_next);
  2560. bio->bi_next = NULL;
  2561. bio->bi_rw |= rw;
  2562. spin_lock(&device->io_lock);
  2563. if (bio->bi_rw & REQ_SYNC)
  2564. pending_bios = &device->pending_sync_bios;
  2565. else
  2566. pending_bios = &device->pending_bios;
  2567. if (pending_bios->tail)
  2568. pending_bios->tail->bi_next = bio;
  2569. pending_bios->tail = bio;
  2570. if (!pending_bios->head)
  2571. pending_bios->head = bio;
  2572. if (device->running_pending)
  2573. should_queue = 0;
  2574. spin_unlock(&device->io_lock);
  2575. if (should_queue)
  2576. btrfs_queue_worker(&root->fs_info->submit_workers,
  2577. &device->work);
  2578. return 0;
  2579. }
  2580. int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
  2581. int mirror_num, int async_submit)
  2582. {
  2583. struct btrfs_mapping_tree *map_tree;
  2584. struct btrfs_device *dev;
  2585. struct bio *first_bio = bio;
  2586. u64 logical = (u64)bio->bi_sector << 9;
  2587. u64 length = 0;
  2588. u64 map_length;
  2589. struct btrfs_multi_bio *multi = NULL;
  2590. int ret;
  2591. int dev_nr = 0;
  2592. int total_devs = 1;
  2593. length = bio->bi_size;
  2594. map_tree = &root->fs_info->mapping_tree;
  2595. map_length = length;
  2596. ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
  2597. mirror_num);
  2598. BUG_ON(ret);
  2599. total_devs = multi->num_stripes;
  2600. if (map_length < length) {
  2601. printk(KERN_CRIT "mapping failed logical %llu bio len %llu "
  2602. "len %llu\n", (unsigned long long)logical,
  2603. (unsigned long long)length,
  2604. (unsigned long long)map_length);
  2605. BUG();
  2606. }
  2607. multi->end_io = first_bio->bi_end_io;
  2608. multi->private = first_bio->bi_private;
  2609. multi->orig_bio = first_bio;
  2610. atomic_set(&multi->stripes_pending, multi->num_stripes);
  2611. while (dev_nr < total_devs) {
  2612. if (total_devs > 1) {
  2613. if (dev_nr < total_devs - 1) {
  2614. bio = bio_clone(first_bio, GFP_NOFS);
  2615. BUG_ON(!bio);
  2616. } else {
  2617. bio = first_bio;
  2618. }
  2619. bio->bi_private = multi;
  2620. bio->bi_end_io = end_bio_multi_stripe;
  2621. }
  2622. bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
  2623. dev = multi->stripes[dev_nr].dev;
  2624. if (dev && dev->bdev && (rw != WRITE || dev->writeable)) {
  2625. bio->bi_bdev = dev->bdev;
  2626. if (async_submit)
  2627. schedule_bio(root, dev, rw, bio);
  2628. else
  2629. submit_bio(rw, bio);
  2630. } else {
  2631. bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
  2632. bio->bi_sector = logical >> 9;
  2633. bio_endio(bio, -EIO);
  2634. }
  2635. dev_nr++;
  2636. }
  2637. if (total_devs == 1)
  2638. kfree(multi);
  2639. return 0;
  2640. }
  2641. struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
  2642. u8 *uuid, u8 *fsid)
  2643. {
  2644. struct btrfs_device *device;
  2645. struct btrfs_fs_devices *cur_devices;
  2646. cur_devices = root->fs_info->fs_devices;
  2647. while (cur_devices) {
  2648. if (!fsid ||
  2649. !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
  2650. device = __find_device(&cur_devices->devices,
  2651. devid, uuid);
  2652. if (device)
  2653. return device;
  2654. }
  2655. cur_devices = cur_devices->seed;
  2656. }
  2657. return NULL;
  2658. }
  2659. static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
  2660. u64 devid, u8 *dev_uuid)
  2661. {
  2662. struct btrfs_device *device;
  2663. struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
  2664. device = kzalloc(sizeof(*device), GFP_NOFS);
  2665. if (!device)
  2666. return NULL;
  2667. list_add(&device->dev_list,
  2668. &fs_devices->devices);
  2669. device->barriers = 1;
  2670. device->dev_root = root->fs_info->dev_root;
  2671. device->devid = devid;
  2672. device->work.func = pending_bios_fn;
  2673. device->fs_devices = fs_devices;
  2674. fs_devices->num_devices++;
  2675. spin_lock_init(&device->io_lock);
  2676. INIT_LIST_HEAD(&device->dev_alloc_list);
  2677. memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
  2678. return device;
  2679. }
  2680. static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
  2681. struct extent_buffer *leaf,
  2682. struct btrfs_chunk *chunk)
  2683. {
  2684. struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
  2685. struct map_lookup *map;
  2686. struct extent_map *em;
  2687. u64 logical;
  2688. u64 length;
  2689. u64 devid;
  2690. u8 uuid[BTRFS_UUID_SIZE];
  2691. int num_stripes;
  2692. int ret;
  2693. int i;
  2694. logical = key->offset;
  2695. length = btrfs_chunk_length(leaf, chunk);
  2696. read_lock(&map_tree->map_tree.lock);
  2697. em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
  2698. read_unlock(&map_tree->map_tree.lock);
  2699. /* already mapped? */
  2700. if (em && em->start <= logical && em->start + em->len > logical) {
  2701. free_extent_map(em);
  2702. return 0;
  2703. } else if (em) {
  2704. free_extent_map(em);
  2705. }
  2706. em = alloc_extent_map(GFP_NOFS);
  2707. if (!em)
  2708. return -ENOMEM;
  2709. num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
  2710. map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
  2711. if (!map) {
  2712. free_extent_map(em);
  2713. return -ENOMEM;
  2714. }
  2715. em->bdev = (struct block_device *)map;
  2716. em->start = logical;
  2717. em->len = length;
  2718. em->block_start = 0;
  2719. em->block_len = em->len;
  2720. map->num_stripes = num_stripes;
  2721. map->io_width = btrfs_chunk_io_width(leaf, chunk);
  2722. map->io_align = btrfs_chunk_io_align(leaf, chunk);
  2723. map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
  2724. map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
  2725. map->type = btrfs_chunk_type(leaf, chunk);
  2726. map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
  2727. for (i = 0; i < num_stripes; i++) {
  2728. map->stripes[i].physical =
  2729. btrfs_stripe_offset_nr(leaf, chunk, i);
  2730. devid = btrfs_stripe_devid_nr(leaf, chunk, i);
  2731. read_extent_buffer(leaf, uuid, (unsigned long)
  2732. btrfs_stripe_dev_uuid_nr(chunk, i),
  2733. BTRFS_UUID_SIZE);
  2734. map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
  2735. NULL);
  2736. if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
  2737. kfree(map);
  2738. free_extent_map(em);
  2739. return -EIO;
  2740. }
  2741. if (!map->stripes[i].dev) {
  2742. map->stripes[i].dev =
  2743. add_missing_dev(root, devid, uuid);
  2744. if (!map->stripes[i].dev) {
  2745. kfree(map);
  2746. free_extent_map(em);
  2747. return -EIO;
  2748. }
  2749. }
  2750. map->stripes[i].dev->in_fs_metadata = 1;
  2751. }
  2752. write_lock(&map_tree->map_tree.lock);
  2753. ret = add_extent_mapping(&map_tree->map_tree, em);
  2754. write_unlock(&map_tree->map_tree.lock);
  2755. BUG_ON(ret);
  2756. free_extent_map(em);
  2757. return 0;
  2758. }
  2759. static int fill_device_from_item(struct extent_buffer *leaf,
  2760. struct btrfs_dev_item *dev_item,
  2761. struct btrfs_device *device)
  2762. {
  2763. unsigned long ptr;
  2764. device->devid = btrfs_device_id(leaf, dev_item);
  2765. device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
  2766. device->total_bytes = device->disk_total_bytes;
  2767. device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
  2768. device->type = btrfs_device_type(leaf, dev_item);
  2769. device->io_align = btrfs_device_io_align(leaf, dev_item);
  2770. device->io_width = btrfs_device_io_width(leaf, dev_item);
  2771. device->sector_size = btrfs_device_sector_size(leaf, dev_item);
  2772. ptr = (unsigned long)btrfs_device_uuid(dev_item);
  2773. read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
  2774. return 0;
  2775. }
  2776. static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
  2777. {
  2778. struct btrfs_fs_devices *fs_devices;
  2779. int ret;
  2780. mutex_lock(&uuid_mutex);
  2781. fs_devices = root->fs_info->fs_devices->seed;
  2782. while (fs_devices) {
  2783. if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
  2784. ret = 0;
  2785. goto out;
  2786. }
  2787. fs_devices = fs_devices->seed;
  2788. }
  2789. fs_devices = find_fsid(fsid);
  2790. if (!fs_devices) {
  2791. ret = -ENOENT;
  2792. goto out;
  2793. }
  2794. fs_devices = clone_fs_devices(fs_devices);
  2795. if (IS_ERR(fs_devices)) {
  2796. ret = PTR_ERR(fs_devices);
  2797. goto out;
  2798. }
  2799. ret = __btrfs_open_devices(fs_devices, FMODE_READ,
  2800. root->fs_info->bdev_holder);
  2801. if (ret)
  2802. goto out;
  2803. if (!fs_devices->seeding) {
  2804. __btrfs_close_devices(fs_devices);
  2805. free_fs_devices(fs_devices);
  2806. ret = -EINVAL;
  2807. goto out;
  2808. }
  2809. fs_devices->seed = root->fs_info->fs_devices->seed;
  2810. root->fs_info->fs_devices->seed = fs_devices;
  2811. out:
  2812. mutex_unlock(&uuid_mutex);
  2813. return ret;
  2814. }
  2815. static int read_one_dev(struct btrfs_root *root,
  2816. struct extent_buffer *leaf,
  2817. struct btrfs_dev_item *dev_item)
  2818. {
  2819. struct btrfs_device *device;
  2820. u64 devid;
  2821. int ret;
  2822. u8 fs_uuid[BTRFS_UUID_SIZE];
  2823. u8 dev_uuid[BTRFS_UUID_SIZE];
  2824. devid = btrfs_device_id(leaf, dev_item);
  2825. read_extent_buffer(leaf, dev_uuid,
  2826. (unsigned long)btrfs_device_uuid(dev_item),
  2827. BTRFS_UUID_SIZE);
  2828. read_extent_buffer(leaf, fs_uuid,
  2829. (unsigned long)btrfs_device_fsid(dev_item),
  2830. BTRFS_UUID_SIZE);
  2831. if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
  2832. ret = open_seed_devices(root, fs_uuid);
  2833. if (ret && !btrfs_test_opt(root, DEGRADED))
  2834. return ret;
  2835. }
  2836. device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
  2837. if (!device || !device->bdev) {
  2838. if (!btrfs_test_opt(root, DEGRADED))
  2839. return -EIO;
  2840. if (!device) {
  2841. printk(KERN_WARNING "warning devid %llu missing\n",
  2842. (unsigned long long)devid);
  2843. device = add_missing_dev(root, devid, dev_uuid);
  2844. if (!device)
  2845. return -ENOMEM;
  2846. }
  2847. }
  2848. if (device->fs_devices != root->fs_info->fs_devices) {
  2849. BUG_ON(device->writeable);
  2850. if (device->generation !=
  2851. btrfs_device_generation(leaf, dev_item))
  2852. return -EINVAL;
  2853. }
  2854. fill_device_from_item(leaf, dev_item, device);
  2855. device->dev_root = root->fs_info->dev_root;
  2856. device->in_fs_metadata = 1;
  2857. if (device->writeable)
  2858. device->fs_devices->total_rw_bytes += device->total_bytes;
  2859. ret = 0;
  2860. return ret;
  2861. }
  2862. int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf)
  2863. {
  2864. struct btrfs_dev_item *dev_item;
  2865. dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
  2866. dev_item);
  2867. return read_one_dev(root, buf, dev_item);
  2868. }
  2869. int btrfs_read_sys_array(struct btrfs_root *root)
  2870. {
  2871. struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
  2872. struct extent_buffer *sb;
  2873. struct btrfs_disk_key *disk_key;
  2874. struct btrfs_chunk *chunk;
  2875. u8 *ptr;
  2876. unsigned long sb_ptr;
  2877. int ret = 0;
  2878. u32 num_stripes;
  2879. u32 array_size;
  2880. u32 len = 0;
  2881. u32 cur;
  2882. struct btrfs_key key;
  2883. sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
  2884. BTRFS_SUPER_INFO_SIZE);
  2885. if (!sb)
  2886. return -ENOMEM;
  2887. btrfs_set_buffer_uptodate(sb);
  2888. btrfs_set_buffer_lockdep_class(sb, 0);
  2889. write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
  2890. array_size = btrfs_super_sys_array_size(super_copy);
  2891. ptr = super_copy->sys_chunk_array;
  2892. sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
  2893. cur = 0;
  2894. while (cur < array_size) {
  2895. disk_key = (struct btrfs_disk_key *)ptr;
  2896. btrfs_disk_key_to_cpu(&key, disk_key);
  2897. len = sizeof(*disk_key); ptr += len;
  2898. sb_ptr += len;
  2899. cur += len;
  2900. if (key.type == BTRFS_CHUNK_ITEM_KEY) {
  2901. chunk = (struct btrfs_chunk *)sb_ptr;
  2902. ret = read_one_chunk(root, &key, sb, chunk);
  2903. if (ret)
  2904. break;
  2905. num_stripes = btrfs_chunk_num_stripes(sb, chunk);
  2906. len = btrfs_chunk_item_size(num_stripes);
  2907. } else {
  2908. ret = -EIO;
  2909. break;
  2910. }
  2911. ptr += len;
  2912. sb_ptr += len;
  2913. cur += len;
  2914. }
  2915. free_extent_buffer(sb);
  2916. return ret;
  2917. }
  2918. int btrfs_read_chunk_tree(struct btrfs_root *root)
  2919. {
  2920. struct btrfs_path *path;
  2921. struct extent_buffer *leaf;
  2922. struct btrfs_key key;
  2923. struct btrfs_key found_key;
  2924. int ret;
  2925. int slot;
  2926. root = root->fs_info->chunk_root;
  2927. path = btrfs_alloc_path();
  2928. if (!path)
  2929. return -ENOMEM;
  2930. /* first we search for all of the device items, and then we
  2931. * read in all of the chunk items. This way we can create chunk
  2932. * mappings that reference all of the devices that are afound
  2933. */
  2934. key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
  2935. key.offset = 0;
  2936. key.type = 0;
  2937. again:
  2938. ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
  2939. if (ret < 0)
  2940. goto error;
  2941. while (1) {
  2942. leaf = path->nodes[0];
  2943. slot = path->slots[0];
  2944. if (slot >= btrfs_header_nritems(leaf)) {
  2945. ret = btrfs_next_leaf(root, path);
  2946. if (ret == 0)
  2947. continue;
  2948. if (ret < 0)
  2949. goto error;
  2950. break;
  2951. }
  2952. btrfs_item_key_to_cpu(leaf, &found_key, slot);
  2953. if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
  2954. if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
  2955. break;
  2956. if (found_key.type == BTRFS_DEV_ITEM_KEY) {
  2957. struct btrfs_dev_item *dev_item;
  2958. dev_item = btrfs_item_ptr(leaf, slot,
  2959. struct btrfs_dev_item);
  2960. ret = read_one_dev(root, leaf, dev_item);
  2961. if (ret)
  2962. goto error;
  2963. }
  2964. } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
  2965. struct btrfs_chunk *chunk;
  2966. chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
  2967. ret = read_one_chunk(root, &found_key, leaf, chunk);
  2968. if (ret)
  2969. goto error;
  2970. }
  2971. path->slots[0]++;
  2972. }
  2973. if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
  2974. key.objectid = 0;
  2975. btrfs_release_path(root, path);
  2976. goto again;
  2977. }
  2978. ret = 0;
  2979. error:
  2980. btrfs_free_path(path);
  2981. return ret;
  2982. }