xfs_mount.c 57 KB

1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162636465666768697071727374757677787980818283848586878889909192939495969798991001011021031041051061071081091101111121131141151161171181191201211221231241251261271281291301311321331341351361371381391401411421431441451461471481491501511521531541551561571581591601611621631641651661671681691701711721731741751761771781791801811821831841851861871881891901911921931941951961971981992002012022032042052062072082092102112122132142152162172182192202212222232242252262272282292302312322332342352362372382392402412422432442452462472482492502512522532542552562572582592602612622632642652662672682692702712722732742752762772782792802812822832842852862872882892902912922932942952962972982993003013023033043053063073083093103113123133143153163173183193203213223233243253263273283293303313323333343353363373383393403413423433443453463473483493503513523533543553563573583593603613623633643653663673683693703713723733743753763773783793803813823833843853863873883893903913923933943953963973983994004014024034044054064074084094104114124134144154164174184194204214224234244254264274284294304314324334344354364374384394404414424434444454464474484494504514524534544554564574584594604614624634644654664674684694704714724734744754764774784794804814824834844854864874884894904914924934944954964974984995005015025035045055065075085095105115125135145155165175185195205215225235245255265275285295305315325335345355365375385395405415425435445455465475485495505515525535545555565575585595605615625635645655665675685695705715725735745755765775785795805815825835845855865875885895905915925935945955965975985996006016026036046056066076086096106116126136146156166176186196206216226236246256266276286296306316326336346356366376386396406416426436446456466476486496506516526536546556566576586596606616626636646656666676686696706716726736746756766776786796806816826836846856866876886896906916926936946956966976986997007017027037047057067077087097107117127137147157167177187197207217227237247257267277287297307317327337347357367377387397407417427437447457467477487497507517527537547557567577587597607617627637647657667677687697707717727737747757767777787797807817827837847857867877887897907917927937947957967977987998008018028038048058068078088098108118128138148158168178188198208218228238248258268278288298308318328338348358368378388398408418428438448458468478488498508518528538548558568578588598608618628638648658668678688698708718728738748758768778788798808818828838848858868878888898908918928938948958968978988999009019029039049059069079089099109119129139149159169179189199209219229239249259269279289299309319329339349359369379389399409419429439449459469479489499509519529539549559569579589599609619629639649659669679689699709719729739749759769779789799809819829839849859869879889899909919929939949959969979989991000100110021003100410051006100710081009101010111012101310141015101610171018101910201021102210231024102510261027102810291030103110321033103410351036103710381039104010411042104310441045104610471048104910501051105210531054105510561057105810591060106110621063106410651066106710681069107010711072107310741075107610771078107910801081108210831084108510861087108810891090109110921093109410951096109710981099110011011102110311041105110611071108110911101111111211131114111511161117111811191120112111221123112411251126112711281129113011311132113311341135113611371138113911401141114211431144114511461147114811491150115111521153115411551156115711581159116011611162116311641165116611671168116911701171117211731174117511761177117811791180118111821183118411851186118711881189119011911192119311941195119611971198119912001201120212031204120512061207120812091210121112121213121412151216121712181219122012211222122312241225122612271228122912301231123212331234123512361237123812391240124112421243124412451246124712481249125012511252125312541255125612571258125912601261126212631264126512661267126812691270127112721273127412751276127712781279128012811282128312841285128612871288128912901291129212931294129512961297129812991300130113021303130413051306130713081309131013111312131313141315131613171318131913201321132213231324132513261327132813291330133113321333133413351336133713381339134013411342134313441345134613471348134913501351135213531354135513561357135813591360136113621363136413651366136713681369137013711372137313741375137613771378137913801381138213831384138513861387138813891390139113921393139413951396139713981399140014011402140314041405140614071408140914101411141214131414141514161417141814191420142114221423142414251426142714281429143014311432143314341435143614371438143914401441144214431444144514461447144814491450145114521453145414551456145714581459146014611462146314641465146614671468146914701471147214731474147514761477147814791480148114821483148414851486148714881489149014911492149314941495149614971498149915001501150215031504150515061507150815091510151115121513151415151516151715181519152015211522152315241525152615271528152915301531153215331534153515361537153815391540154115421543154415451546154715481549155015511552155315541555155615571558155915601561156215631564156515661567156815691570157115721573157415751576157715781579158015811582158315841585158615871588158915901591159215931594159515961597159815991600160116021603160416051606160716081609161016111612161316141615161616171618161916201621162216231624162516261627162816291630163116321633163416351636163716381639164016411642164316441645164616471648164916501651165216531654165516561657165816591660166116621663166416651666166716681669167016711672167316741675167616771678167916801681168216831684168516861687168816891690169116921693169416951696169716981699170017011702170317041705170617071708170917101711171217131714171517161717171817191720172117221723172417251726172717281729173017311732173317341735173617371738173917401741174217431744174517461747174817491750175117521753175417551756175717581759176017611762176317641765176617671768176917701771177217731774177517761777177817791780178117821783178417851786178717881789179017911792179317941795179617971798179918001801180218031804180518061807180818091810181118121813181418151816181718181819182018211822182318241825182618271828182918301831183218331834183518361837183818391840184118421843184418451846184718481849185018511852185318541855185618571858185918601861186218631864186518661867186818691870187118721873187418751876187718781879188018811882188318841885188618871888188918901891189218931894189518961897189818991900190119021903190419051906190719081909191019111912191319141915191619171918191919201921192219231924192519261927192819291930193119321933193419351936193719381939194019411942194319441945194619471948194919501951195219531954195519561957195819591960196119621963196419651966196719681969197019711972197319741975197619771978197919801981198219831984198519861987198819891990199119921993199419951996199719981999200020012002200320042005200620072008200920102011201220132014201520162017201820192020202120222023202420252026202720282029203020312032203320342035203620372038203920402041204220432044204520462047204820492050205120522053205420552056205720582059206020612062206320642065206620672068206920702071207220732074207520762077207820792080208120822083208420852086208720882089209020912092209320942095209620972098209921002101210221032104210521062107210821092110211121122113211421152116211721182119212021212122212321242125212621272128212921302131213221332134213521362137213821392140214121422143214421452146214721482149215021512152215321542155215621572158215921602161216221632164216521662167216821692170217121722173217421752176217721782179218021812182218321842185218621872188218921902191219221932194219521962197219821992200220122022203220422052206220722082209221022112212221322142215221622172218221922202221222222232224222522262227222822292230223122322233
  1. /*
  2. * Copyright (c) 2000-2005 Silicon Graphics, Inc.
  3. * All Rights Reserved.
  4. *
  5. * This program is free software; you can redistribute it and/or
  6. * modify it under the terms of the GNU General Public License as
  7. * published by the Free Software Foundation.
  8. *
  9. * This program is distributed in the hope that it would be useful,
  10. * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11. * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  12. * GNU General Public License for more details.
  13. *
  14. * You should have received a copy of the GNU General Public License
  15. * along with this program; if not, write the Free Software Foundation,
  16. * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
  17. */
  18. #include "xfs.h"
  19. #include "xfs_fs.h"
  20. #include "xfs_types.h"
  21. #include "xfs_bit.h"
  22. #include "xfs_log.h"
  23. #include "xfs_inum.h"
  24. #include "xfs_trans.h"
  25. #include "xfs_sb.h"
  26. #include "xfs_ag.h"
  27. #include "xfs_dir2.h"
  28. #include "xfs_dmapi.h"
  29. #include "xfs_mount.h"
  30. #include "xfs_bmap_btree.h"
  31. #include "xfs_alloc_btree.h"
  32. #include "xfs_ialloc_btree.h"
  33. #include "xfs_dir2_sf.h"
  34. #include "xfs_attr_sf.h"
  35. #include "xfs_dinode.h"
  36. #include "xfs_inode.h"
  37. #include "xfs_btree.h"
  38. #include "xfs_ialloc.h"
  39. #include "xfs_alloc.h"
  40. #include "xfs_rtalloc.h"
  41. #include "xfs_bmap.h"
  42. #include "xfs_error.h"
  43. #include "xfs_rw.h"
  44. #include "xfs_quota.h"
  45. #include "xfs_fsops.h"
  46. STATIC void xfs_mount_log_sbunit(xfs_mount_t *, __int64_t);
  47. STATIC int xfs_uuid_mount(xfs_mount_t *);
  48. STATIC void xfs_uuid_unmount(xfs_mount_t *mp);
  49. STATIC void xfs_unmountfs_wait(xfs_mount_t *);
  50. #ifdef HAVE_PERCPU_SB
  51. STATIC void xfs_icsb_destroy_counters(xfs_mount_t *);
  52. STATIC void xfs_icsb_balance_counter(xfs_mount_t *, xfs_sb_field_t,
  53. int, int);
  54. STATIC void xfs_icsb_sync_counters(xfs_mount_t *);
  55. STATIC int xfs_icsb_modify_counters(xfs_mount_t *, xfs_sb_field_t,
  56. int64_t, int);
  57. STATIC int xfs_icsb_disable_counter(xfs_mount_t *, xfs_sb_field_t);
  58. #else
  59. #define xfs_icsb_destroy_counters(mp) do { } while (0)
  60. #define xfs_icsb_balance_counter(mp, a, b, c) do { } while (0)
  61. #define xfs_icsb_sync_counters(mp) do { } while (0)
  62. #define xfs_icsb_modify_counters(mp, a, b, c) do { } while (0)
  63. #endif
  64. static const struct {
  65. short offset;
  66. short type; /* 0 = integer
  67. * 1 = binary / string (no translation)
  68. */
  69. } xfs_sb_info[] = {
  70. { offsetof(xfs_sb_t, sb_magicnum), 0 },
  71. { offsetof(xfs_sb_t, sb_blocksize), 0 },
  72. { offsetof(xfs_sb_t, sb_dblocks), 0 },
  73. { offsetof(xfs_sb_t, sb_rblocks), 0 },
  74. { offsetof(xfs_sb_t, sb_rextents), 0 },
  75. { offsetof(xfs_sb_t, sb_uuid), 1 },
  76. { offsetof(xfs_sb_t, sb_logstart), 0 },
  77. { offsetof(xfs_sb_t, sb_rootino), 0 },
  78. { offsetof(xfs_sb_t, sb_rbmino), 0 },
  79. { offsetof(xfs_sb_t, sb_rsumino), 0 },
  80. { offsetof(xfs_sb_t, sb_rextsize), 0 },
  81. { offsetof(xfs_sb_t, sb_agblocks), 0 },
  82. { offsetof(xfs_sb_t, sb_agcount), 0 },
  83. { offsetof(xfs_sb_t, sb_rbmblocks), 0 },
  84. { offsetof(xfs_sb_t, sb_logblocks), 0 },
  85. { offsetof(xfs_sb_t, sb_versionnum), 0 },
  86. { offsetof(xfs_sb_t, sb_sectsize), 0 },
  87. { offsetof(xfs_sb_t, sb_inodesize), 0 },
  88. { offsetof(xfs_sb_t, sb_inopblock), 0 },
  89. { offsetof(xfs_sb_t, sb_fname[0]), 1 },
  90. { offsetof(xfs_sb_t, sb_blocklog), 0 },
  91. { offsetof(xfs_sb_t, sb_sectlog), 0 },
  92. { offsetof(xfs_sb_t, sb_inodelog), 0 },
  93. { offsetof(xfs_sb_t, sb_inopblog), 0 },
  94. { offsetof(xfs_sb_t, sb_agblklog), 0 },
  95. { offsetof(xfs_sb_t, sb_rextslog), 0 },
  96. { offsetof(xfs_sb_t, sb_inprogress), 0 },
  97. { offsetof(xfs_sb_t, sb_imax_pct), 0 },
  98. { offsetof(xfs_sb_t, sb_icount), 0 },
  99. { offsetof(xfs_sb_t, sb_ifree), 0 },
  100. { offsetof(xfs_sb_t, sb_fdblocks), 0 },
  101. { offsetof(xfs_sb_t, sb_frextents), 0 },
  102. { offsetof(xfs_sb_t, sb_uquotino), 0 },
  103. { offsetof(xfs_sb_t, sb_gquotino), 0 },
  104. { offsetof(xfs_sb_t, sb_qflags), 0 },
  105. { offsetof(xfs_sb_t, sb_flags), 0 },
  106. { offsetof(xfs_sb_t, sb_shared_vn), 0 },
  107. { offsetof(xfs_sb_t, sb_inoalignmt), 0 },
  108. { offsetof(xfs_sb_t, sb_unit), 0 },
  109. { offsetof(xfs_sb_t, sb_width), 0 },
  110. { offsetof(xfs_sb_t, sb_dirblklog), 0 },
  111. { offsetof(xfs_sb_t, sb_logsectlog), 0 },
  112. { offsetof(xfs_sb_t, sb_logsectsize),0 },
  113. { offsetof(xfs_sb_t, sb_logsunit), 0 },
  114. { offsetof(xfs_sb_t, sb_features2), 0 },
  115. { sizeof(xfs_sb_t), 0 }
  116. };
  117. /*
  118. * Return a pointer to an initialized xfs_mount structure.
  119. */
  120. xfs_mount_t *
  121. xfs_mount_init(void)
  122. {
  123. xfs_mount_t *mp;
  124. mp = kmem_zalloc(sizeof(xfs_mount_t), KM_SLEEP);
  125. if (xfs_icsb_init_counters(mp)) {
  126. mp->m_flags |= XFS_MOUNT_NO_PERCPU_SB;
  127. }
  128. AIL_LOCKINIT(&mp->m_ail_lock, "xfs_ail");
  129. spinlock_init(&mp->m_sb_lock, "xfs_sb");
  130. mutex_init(&mp->m_ilock);
  131. initnsema(&mp->m_growlock, 1, "xfs_grow");
  132. /*
  133. * Initialize the AIL.
  134. */
  135. xfs_trans_ail_init(mp);
  136. atomic_set(&mp->m_active_trans, 0);
  137. return mp;
  138. }
  139. /*
  140. * Free up the resources associated with a mount structure. Assume that
  141. * the structure was initially zeroed, so we can tell which fields got
  142. * initialized.
  143. */
  144. void
  145. xfs_mount_free(
  146. xfs_mount_t *mp,
  147. int remove_bhv)
  148. {
  149. if (mp->m_ihash)
  150. xfs_ihash_free(mp);
  151. if (mp->m_chash)
  152. xfs_chash_free(mp);
  153. if (mp->m_perag) {
  154. int agno;
  155. for (agno = 0; agno < mp->m_maxagi; agno++)
  156. if (mp->m_perag[agno].pagb_list)
  157. kmem_free(mp->m_perag[agno].pagb_list,
  158. sizeof(xfs_perag_busy_t) *
  159. XFS_PAGB_NUM_SLOTS);
  160. kmem_free(mp->m_perag,
  161. sizeof(xfs_perag_t) * mp->m_sb.sb_agcount);
  162. }
  163. AIL_LOCK_DESTROY(&mp->m_ail_lock);
  164. spinlock_destroy(&mp->m_sb_lock);
  165. mutex_destroy(&mp->m_ilock);
  166. freesema(&mp->m_growlock);
  167. if (mp->m_quotainfo)
  168. XFS_QM_DONE(mp);
  169. if (mp->m_fsname != NULL)
  170. kmem_free(mp->m_fsname, mp->m_fsname_len);
  171. if (mp->m_rtname != NULL)
  172. kmem_free(mp->m_rtname, strlen(mp->m_rtname) + 1);
  173. if (mp->m_logname != NULL)
  174. kmem_free(mp->m_logname, strlen(mp->m_logname) + 1);
  175. if (remove_bhv) {
  176. struct bhv_vfs *vfsp = XFS_MTOVFS(mp);
  177. bhv_remove_all_vfsops(vfsp, 0);
  178. VFS_REMOVEBHV(vfsp, &mp->m_bhv);
  179. }
  180. xfs_icsb_destroy_counters(mp);
  181. kmem_free(mp, sizeof(xfs_mount_t));
  182. }
  183. /*
  184. * Check the validity of the SB found.
  185. */
  186. STATIC int
  187. xfs_mount_validate_sb(
  188. xfs_mount_t *mp,
  189. xfs_sb_t *sbp,
  190. int flags)
  191. {
  192. /*
  193. * If the log device and data device have the
  194. * same device number, the log is internal.
  195. * Consequently, the sb_logstart should be non-zero. If
  196. * we have a zero sb_logstart in this case, we may be trying to mount
  197. * a volume filesystem in a non-volume manner.
  198. */
  199. if (sbp->sb_magicnum != XFS_SB_MAGIC) {
  200. xfs_fs_mount_cmn_err(flags, "bad magic number");
  201. return XFS_ERROR(EWRONGFS);
  202. }
  203. if (!XFS_SB_GOOD_VERSION(sbp)) {
  204. xfs_fs_mount_cmn_err(flags, "bad version");
  205. return XFS_ERROR(EWRONGFS);
  206. }
  207. if (unlikely(
  208. sbp->sb_logstart == 0 && mp->m_logdev_targp == mp->m_ddev_targp)) {
  209. xfs_fs_mount_cmn_err(flags,
  210. "filesystem is marked as having an external log; "
  211. "specify logdev on the\nmount command line.");
  212. return XFS_ERROR(EINVAL);
  213. }
  214. if (unlikely(
  215. sbp->sb_logstart != 0 && mp->m_logdev_targp != mp->m_ddev_targp)) {
  216. xfs_fs_mount_cmn_err(flags,
  217. "filesystem is marked as having an internal log; "
  218. "do not specify logdev on\nthe mount command line.");
  219. return XFS_ERROR(EINVAL);
  220. }
  221. /*
  222. * More sanity checking. These were stolen directly from
  223. * xfs_repair.
  224. */
  225. if (unlikely(
  226. sbp->sb_agcount <= 0 ||
  227. sbp->sb_sectsize < XFS_MIN_SECTORSIZE ||
  228. sbp->sb_sectsize > XFS_MAX_SECTORSIZE ||
  229. sbp->sb_sectlog < XFS_MIN_SECTORSIZE_LOG ||
  230. sbp->sb_sectlog > XFS_MAX_SECTORSIZE_LOG ||
  231. sbp->sb_blocksize < XFS_MIN_BLOCKSIZE ||
  232. sbp->sb_blocksize > XFS_MAX_BLOCKSIZE ||
  233. sbp->sb_blocklog < XFS_MIN_BLOCKSIZE_LOG ||
  234. sbp->sb_blocklog > XFS_MAX_BLOCKSIZE_LOG ||
  235. sbp->sb_inodesize < XFS_DINODE_MIN_SIZE ||
  236. sbp->sb_inodesize > XFS_DINODE_MAX_SIZE ||
  237. sbp->sb_inodelog < XFS_DINODE_MIN_LOG ||
  238. sbp->sb_inodelog > XFS_DINODE_MAX_LOG ||
  239. (sbp->sb_blocklog - sbp->sb_inodelog != sbp->sb_inopblog) ||
  240. (sbp->sb_rextsize * sbp->sb_blocksize > XFS_MAX_RTEXTSIZE) ||
  241. (sbp->sb_rextsize * sbp->sb_blocksize < XFS_MIN_RTEXTSIZE) ||
  242. (sbp->sb_imax_pct > 100 /* zero sb_imax_pct is valid */))) {
  243. xfs_fs_mount_cmn_err(flags, "SB sanity check 1 failed");
  244. return XFS_ERROR(EFSCORRUPTED);
  245. }
  246. /*
  247. * Sanity check AG count, size fields against data size field
  248. */
  249. if (unlikely(
  250. sbp->sb_dblocks == 0 ||
  251. sbp->sb_dblocks >
  252. (xfs_drfsbno_t)sbp->sb_agcount * sbp->sb_agblocks ||
  253. sbp->sb_dblocks < (xfs_drfsbno_t)(sbp->sb_agcount - 1) *
  254. sbp->sb_agblocks + XFS_MIN_AG_BLOCKS)) {
  255. xfs_fs_mount_cmn_err(flags, "SB sanity check 2 failed");
  256. return XFS_ERROR(EFSCORRUPTED);
  257. }
  258. ASSERT(PAGE_SHIFT >= sbp->sb_blocklog);
  259. ASSERT(sbp->sb_blocklog >= BBSHIFT);
  260. #if XFS_BIG_BLKNOS /* Limited by ULONG_MAX of page cache index */
  261. if (unlikely(
  262. (sbp->sb_dblocks >> (PAGE_SHIFT - sbp->sb_blocklog)) > ULONG_MAX ||
  263. (sbp->sb_rblocks >> (PAGE_SHIFT - sbp->sb_blocklog)) > ULONG_MAX)) {
  264. #else /* Limited by UINT_MAX of sectors */
  265. if (unlikely(
  266. (sbp->sb_dblocks << (sbp->sb_blocklog - BBSHIFT)) > UINT_MAX ||
  267. (sbp->sb_rblocks << (sbp->sb_blocklog - BBSHIFT)) > UINT_MAX)) {
  268. #endif
  269. xfs_fs_mount_cmn_err(flags,
  270. "file system too large to be mounted on this system.");
  271. return XFS_ERROR(E2BIG);
  272. }
  273. if (unlikely(sbp->sb_inprogress)) {
  274. xfs_fs_mount_cmn_err(flags, "file system busy");
  275. return XFS_ERROR(EFSCORRUPTED);
  276. }
  277. /*
  278. * Version 1 directory format has never worked on Linux.
  279. */
  280. if (unlikely(!XFS_SB_VERSION_HASDIRV2(sbp))) {
  281. xfs_fs_mount_cmn_err(flags,
  282. "file system using version 1 directory format");
  283. return XFS_ERROR(ENOSYS);
  284. }
  285. /*
  286. * Until this is fixed only page-sized or smaller data blocks work.
  287. */
  288. if (unlikely(sbp->sb_blocksize > PAGE_SIZE)) {
  289. xfs_fs_mount_cmn_err(flags,
  290. "file system with blocksize %d bytes",
  291. sbp->sb_blocksize);
  292. xfs_fs_mount_cmn_err(flags,
  293. "only pagesize (%ld) or less will currently work.",
  294. PAGE_SIZE);
  295. return XFS_ERROR(ENOSYS);
  296. }
  297. return 0;
  298. }
  299. xfs_agnumber_t
  300. xfs_initialize_perag(
  301. bhv_vfs_t *vfs,
  302. xfs_mount_t *mp,
  303. xfs_agnumber_t agcount)
  304. {
  305. xfs_agnumber_t index, max_metadata;
  306. xfs_perag_t *pag;
  307. xfs_agino_t agino;
  308. xfs_ino_t ino;
  309. xfs_sb_t *sbp = &mp->m_sb;
  310. xfs_ino_t max_inum = XFS_MAXINUMBER_32;
  311. /* Check to see if the filesystem can overflow 32 bit inodes */
  312. agino = XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks - 1, 0);
  313. ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
  314. /* Clear the mount flag if no inode can overflow 32 bits
  315. * on this filesystem, or if specifically requested..
  316. */
  317. if ((vfs->vfs_flag & VFS_32BITINODES) && ino > max_inum) {
  318. mp->m_flags |= XFS_MOUNT_32BITINODES;
  319. } else {
  320. mp->m_flags &= ~XFS_MOUNT_32BITINODES;
  321. }
  322. /* If we can overflow then setup the ag headers accordingly */
  323. if (mp->m_flags & XFS_MOUNT_32BITINODES) {
  324. /* Calculate how much should be reserved for inodes to
  325. * meet the max inode percentage.
  326. */
  327. if (mp->m_maxicount) {
  328. __uint64_t icount;
  329. icount = sbp->sb_dblocks * sbp->sb_imax_pct;
  330. do_div(icount, 100);
  331. icount += sbp->sb_agblocks - 1;
  332. do_div(icount, sbp->sb_agblocks);
  333. max_metadata = icount;
  334. } else {
  335. max_metadata = agcount;
  336. }
  337. for (index = 0; index < agcount; index++) {
  338. ino = XFS_AGINO_TO_INO(mp, index, agino);
  339. if (ino > max_inum) {
  340. index++;
  341. break;
  342. }
  343. /* This ag is preferred for inodes */
  344. pag = &mp->m_perag[index];
  345. pag->pagi_inodeok = 1;
  346. if (index < max_metadata)
  347. pag->pagf_metadata = 1;
  348. }
  349. } else {
  350. /* Setup default behavior for smaller filesystems */
  351. for (index = 0; index < agcount; index++) {
  352. pag = &mp->m_perag[index];
  353. pag->pagi_inodeok = 1;
  354. }
  355. }
  356. return index;
  357. }
  358. /*
  359. * xfs_xlatesb
  360. *
  361. * data - on disk version of sb
  362. * sb - a superblock
  363. * dir - conversion direction: <0 - convert sb to buf
  364. * >0 - convert buf to sb
  365. * fields - which fields to copy (bitmask)
  366. */
  367. void
  368. xfs_xlatesb(
  369. void *data,
  370. xfs_sb_t *sb,
  371. int dir,
  372. __int64_t fields)
  373. {
  374. xfs_caddr_t buf_ptr;
  375. xfs_caddr_t mem_ptr;
  376. xfs_sb_field_t f;
  377. int first;
  378. int size;
  379. ASSERT(dir);
  380. ASSERT(fields);
  381. if (!fields)
  382. return;
  383. buf_ptr = (xfs_caddr_t)data;
  384. mem_ptr = (xfs_caddr_t)sb;
  385. while (fields) {
  386. f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
  387. first = xfs_sb_info[f].offset;
  388. size = xfs_sb_info[f + 1].offset - first;
  389. ASSERT(xfs_sb_info[f].type == 0 || xfs_sb_info[f].type == 1);
  390. if (size == 1 || xfs_sb_info[f].type == 1) {
  391. if (dir > 0) {
  392. memcpy(mem_ptr + first, buf_ptr + first, size);
  393. } else {
  394. memcpy(buf_ptr + first, mem_ptr + first, size);
  395. }
  396. } else {
  397. switch (size) {
  398. case 2:
  399. INT_XLATE(*(__uint16_t*)(buf_ptr+first),
  400. *(__uint16_t*)(mem_ptr+first),
  401. dir, ARCH_CONVERT);
  402. break;
  403. case 4:
  404. INT_XLATE(*(__uint32_t*)(buf_ptr+first),
  405. *(__uint32_t*)(mem_ptr+first),
  406. dir, ARCH_CONVERT);
  407. break;
  408. case 8:
  409. INT_XLATE(*(__uint64_t*)(buf_ptr+first),
  410. *(__uint64_t*)(mem_ptr+first), dir, ARCH_CONVERT);
  411. break;
  412. default:
  413. ASSERT(0);
  414. }
  415. }
  416. fields &= ~(1LL << f);
  417. }
  418. }
  419. /*
  420. * xfs_readsb
  421. *
  422. * Does the initial read of the superblock.
  423. */
  424. int
  425. xfs_readsb(xfs_mount_t *mp, int flags)
  426. {
  427. unsigned int sector_size;
  428. unsigned int extra_flags;
  429. xfs_buf_t *bp;
  430. xfs_sb_t *sbp;
  431. int error;
  432. ASSERT(mp->m_sb_bp == NULL);
  433. ASSERT(mp->m_ddev_targp != NULL);
  434. /*
  435. * Allocate a (locked) buffer to hold the superblock.
  436. * This will be kept around at all times to optimize
  437. * access to the superblock.
  438. */
  439. sector_size = xfs_getsize_buftarg(mp->m_ddev_targp);
  440. extra_flags = XFS_BUF_LOCK | XFS_BUF_MANAGE | XFS_BUF_MAPPED;
  441. bp = xfs_buf_read_flags(mp->m_ddev_targp, XFS_SB_DADDR,
  442. BTOBB(sector_size), extra_flags);
  443. if (!bp || XFS_BUF_ISERROR(bp)) {
  444. xfs_fs_mount_cmn_err(flags, "SB read failed");
  445. error = bp ? XFS_BUF_GETERROR(bp) : ENOMEM;
  446. goto fail;
  447. }
  448. ASSERT(XFS_BUF_ISBUSY(bp));
  449. ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
  450. /*
  451. * Initialize the mount structure from the superblock.
  452. * But first do some basic consistency checking.
  453. */
  454. sbp = XFS_BUF_TO_SBP(bp);
  455. xfs_xlatesb(XFS_BUF_PTR(bp), &(mp->m_sb), 1, XFS_SB_ALL_BITS);
  456. error = xfs_mount_validate_sb(mp, &(mp->m_sb), flags);
  457. if (error) {
  458. xfs_fs_mount_cmn_err(flags, "SB validate failed");
  459. goto fail;
  460. }
  461. /*
  462. * We must be able to do sector-sized and sector-aligned IO.
  463. */
  464. if (sector_size > mp->m_sb.sb_sectsize) {
  465. xfs_fs_mount_cmn_err(flags,
  466. "device supports only %u byte sectors (not %u)",
  467. sector_size, mp->m_sb.sb_sectsize);
  468. error = ENOSYS;
  469. goto fail;
  470. }
  471. /*
  472. * If device sector size is smaller than the superblock size,
  473. * re-read the superblock so the buffer is correctly sized.
  474. */
  475. if (sector_size < mp->m_sb.sb_sectsize) {
  476. XFS_BUF_UNMANAGE(bp);
  477. xfs_buf_relse(bp);
  478. sector_size = mp->m_sb.sb_sectsize;
  479. bp = xfs_buf_read_flags(mp->m_ddev_targp, XFS_SB_DADDR,
  480. BTOBB(sector_size), extra_flags);
  481. if (!bp || XFS_BUF_ISERROR(bp)) {
  482. xfs_fs_mount_cmn_err(flags, "SB re-read failed");
  483. error = bp ? XFS_BUF_GETERROR(bp) : ENOMEM;
  484. goto fail;
  485. }
  486. ASSERT(XFS_BUF_ISBUSY(bp));
  487. ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
  488. }
  489. /* Initialize per-cpu counters */
  490. xfs_icsb_reinit_counters(mp);
  491. mp->m_sb_bp = bp;
  492. xfs_buf_relse(bp);
  493. ASSERT(XFS_BUF_VALUSEMA(bp) > 0);
  494. return 0;
  495. fail:
  496. if (bp) {
  497. XFS_BUF_UNMANAGE(bp);
  498. xfs_buf_relse(bp);
  499. }
  500. return error;
  501. }
  502. /*
  503. * xfs_mount_common
  504. *
  505. * Mount initialization code establishing various mount
  506. * fields from the superblock associated with the given
  507. * mount structure
  508. */
  509. STATIC void
  510. xfs_mount_common(xfs_mount_t *mp, xfs_sb_t *sbp)
  511. {
  512. int i;
  513. mp->m_agfrotor = mp->m_agirotor = 0;
  514. spinlock_init(&mp->m_agirotor_lock, "m_agirotor_lock");
  515. mp->m_maxagi = mp->m_sb.sb_agcount;
  516. mp->m_blkbit_log = sbp->sb_blocklog + XFS_NBBYLOG;
  517. mp->m_blkbb_log = sbp->sb_blocklog - BBSHIFT;
  518. mp->m_sectbb_log = sbp->sb_sectlog - BBSHIFT;
  519. mp->m_agno_log = xfs_highbit32(sbp->sb_agcount - 1) + 1;
  520. mp->m_agino_log = sbp->sb_inopblog + sbp->sb_agblklog;
  521. mp->m_litino = sbp->sb_inodesize -
  522. ((uint)sizeof(xfs_dinode_core_t) + (uint)sizeof(xfs_agino_t));
  523. mp->m_blockmask = sbp->sb_blocksize - 1;
  524. mp->m_blockwsize = sbp->sb_blocksize >> XFS_WORDLOG;
  525. mp->m_blockwmask = mp->m_blockwsize - 1;
  526. INIT_LIST_HEAD(&mp->m_del_inodes);
  527. /*
  528. * Setup for attributes, in case they get created.
  529. * This value is for inodes getting attributes for the first time,
  530. * the per-inode value is for old attribute values.
  531. */
  532. ASSERT(sbp->sb_inodesize >= 256 && sbp->sb_inodesize <= 2048);
  533. switch (sbp->sb_inodesize) {
  534. case 256:
  535. mp->m_attroffset = XFS_LITINO(mp) -
  536. XFS_BMDR_SPACE_CALC(MINABTPTRS);
  537. break;
  538. case 512:
  539. case 1024:
  540. case 2048:
  541. mp->m_attroffset = XFS_BMDR_SPACE_CALC(6 * MINABTPTRS);
  542. break;
  543. default:
  544. ASSERT(0);
  545. }
  546. ASSERT(mp->m_attroffset < XFS_LITINO(mp));
  547. for (i = 0; i < 2; i++) {
  548. mp->m_alloc_mxr[i] = XFS_BTREE_BLOCK_MAXRECS(sbp->sb_blocksize,
  549. xfs_alloc, i == 0);
  550. mp->m_alloc_mnr[i] = XFS_BTREE_BLOCK_MINRECS(sbp->sb_blocksize,
  551. xfs_alloc, i == 0);
  552. }
  553. for (i = 0; i < 2; i++) {
  554. mp->m_bmap_dmxr[i] = XFS_BTREE_BLOCK_MAXRECS(sbp->sb_blocksize,
  555. xfs_bmbt, i == 0);
  556. mp->m_bmap_dmnr[i] = XFS_BTREE_BLOCK_MINRECS(sbp->sb_blocksize,
  557. xfs_bmbt, i == 0);
  558. }
  559. for (i = 0; i < 2; i++) {
  560. mp->m_inobt_mxr[i] = XFS_BTREE_BLOCK_MAXRECS(sbp->sb_blocksize,
  561. xfs_inobt, i == 0);
  562. mp->m_inobt_mnr[i] = XFS_BTREE_BLOCK_MINRECS(sbp->sb_blocksize,
  563. xfs_inobt, i == 0);
  564. }
  565. mp->m_bsize = XFS_FSB_TO_BB(mp, 1);
  566. mp->m_ialloc_inos = (int)MAX((__uint16_t)XFS_INODES_PER_CHUNK,
  567. sbp->sb_inopblock);
  568. mp->m_ialloc_blks = mp->m_ialloc_inos >> sbp->sb_inopblog;
  569. }
  570. /*
  571. * xfs_mountfs
  572. *
  573. * This function does the following on an initial mount of a file system:
  574. * - reads the superblock from disk and init the mount struct
  575. * - if we're a 32-bit kernel, do a size check on the superblock
  576. * so we don't mount terabyte filesystems
  577. * - init mount struct realtime fields
  578. * - allocate inode hash table for fs
  579. * - init directory manager
  580. * - perform recovery and init the log manager
  581. */
  582. int
  583. xfs_mountfs(
  584. bhv_vfs_t *vfsp,
  585. xfs_mount_t *mp,
  586. int mfsi_flags)
  587. {
  588. xfs_buf_t *bp;
  589. xfs_sb_t *sbp = &(mp->m_sb);
  590. xfs_inode_t *rip;
  591. bhv_vnode_t *rvp = NULL;
  592. int readio_log, writeio_log;
  593. xfs_daddr_t d;
  594. __uint64_t ret64;
  595. __int64_t update_flags;
  596. uint quotamount, quotaflags;
  597. int agno;
  598. int uuid_mounted = 0;
  599. int error = 0;
  600. if (mp->m_sb_bp == NULL) {
  601. if ((error = xfs_readsb(mp, mfsi_flags))) {
  602. return error;
  603. }
  604. }
  605. xfs_mount_common(mp, sbp);
  606. /*
  607. * Check if sb_agblocks is aligned at stripe boundary
  608. * If sb_agblocks is NOT aligned turn off m_dalign since
  609. * allocator alignment is within an ag, therefore ag has
  610. * to be aligned at stripe boundary.
  611. */
  612. update_flags = 0LL;
  613. if (mp->m_dalign && !(mfsi_flags & XFS_MFSI_SECOND)) {
  614. /*
  615. * If stripe unit and stripe width are not multiples
  616. * of the fs blocksize turn off alignment.
  617. */
  618. if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
  619. (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
  620. if (mp->m_flags & XFS_MOUNT_RETERR) {
  621. cmn_err(CE_WARN,
  622. "XFS: alignment check 1 failed");
  623. error = XFS_ERROR(EINVAL);
  624. goto error1;
  625. }
  626. mp->m_dalign = mp->m_swidth = 0;
  627. } else {
  628. /*
  629. * Convert the stripe unit and width to FSBs.
  630. */
  631. mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
  632. if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) {
  633. if (mp->m_flags & XFS_MOUNT_RETERR) {
  634. error = XFS_ERROR(EINVAL);
  635. goto error1;
  636. }
  637. xfs_fs_cmn_err(CE_WARN, mp,
  638. "stripe alignment turned off: sunit(%d)/swidth(%d) incompatible with agsize(%d)",
  639. mp->m_dalign, mp->m_swidth,
  640. sbp->sb_agblocks);
  641. mp->m_dalign = 0;
  642. mp->m_swidth = 0;
  643. } else if (mp->m_dalign) {
  644. mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
  645. } else {
  646. if (mp->m_flags & XFS_MOUNT_RETERR) {
  647. xfs_fs_cmn_err(CE_WARN, mp,
  648. "stripe alignment turned off: sunit(%d) less than bsize(%d)",
  649. mp->m_dalign,
  650. mp->m_blockmask +1);
  651. error = XFS_ERROR(EINVAL);
  652. goto error1;
  653. }
  654. mp->m_swidth = 0;
  655. }
  656. }
  657. /*
  658. * Update superblock with new values
  659. * and log changes
  660. */
  661. if (XFS_SB_VERSION_HASDALIGN(sbp)) {
  662. if (sbp->sb_unit != mp->m_dalign) {
  663. sbp->sb_unit = mp->m_dalign;
  664. update_flags |= XFS_SB_UNIT;
  665. }
  666. if (sbp->sb_width != mp->m_swidth) {
  667. sbp->sb_width = mp->m_swidth;
  668. update_flags |= XFS_SB_WIDTH;
  669. }
  670. }
  671. } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN &&
  672. XFS_SB_VERSION_HASDALIGN(&mp->m_sb)) {
  673. mp->m_dalign = sbp->sb_unit;
  674. mp->m_swidth = sbp->sb_width;
  675. }
  676. xfs_alloc_compute_maxlevels(mp);
  677. xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
  678. xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
  679. xfs_ialloc_compute_maxlevels(mp);
  680. if (sbp->sb_imax_pct) {
  681. __uint64_t icount;
  682. /* Make sure the maximum inode count is a multiple of the
  683. * units we allocate inodes in.
  684. */
  685. icount = sbp->sb_dblocks * sbp->sb_imax_pct;
  686. do_div(icount, 100);
  687. do_div(icount, mp->m_ialloc_blks);
  688. mp->m_maxicount = (icount * mp->m_ialloc_blks) <<
  689. sbp->sb_inopblog;
  690. } else
  691. mp->m_maxicount = 0;
  692. mp->m_maxioffset = xfs_max_file_offset(sbp->sb_blocklog);
  693. /*
  694. * XFS uses the uuid from the superblock as the unique
  695. * identifier for fsid. We can not use the uuid from the volume
  696. * since a single partition filesystem is identical to a single
  697. * partition volume/filesystem.
  698. */
  699. if ((mfsi_flags & XFS_MFSI_SECOND) == 0 &&
  700. (mp->m_flags & XFS_MOUNT_NOUUID) == 0) {
  701. if (xfs_uuid_mount(mp)) {
  702. error = XFS_ERROR(EINVAL);
  703. goto error1;
  704. }
  705. uuid_mounted=1;
  706. ret64 = uuid_hash64(&sbp->sb_uuid);
  707. memcpy(&vfsp->vfs_fsid, &ret64, sizeof(ret64));
  708. }
  709. /*
  710. * Set the default minimum read and write sizes unless
  711. * already specified in a mount option.
  712. * We use smaller I/O sizes when the file system
  713. * is being used for NFS service (wsync mount option).
  714. */
  715. if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) {
  716. if (mp->m_flags & XFS_MOUNT_WSYNC) {
  717. readio_log = XFS_WSYNC_READIO_LOG;
  718. writeio_log = XFS_WSYNC_WRITEIO_LOG;
  719. } else {
  720. readio_log = XFS_READIO_LOG_LARGE;
  721. writeio_log = XFS_WRITEIO_LOG_LARGE;
  722. }
  723. } else {
  724. readio_log = mp->m_readio_log;
  725. writeio_log = mp->m_writeio_log;
  726. }
  727. /*
  728. * Set the number of readahead buffers to use based on
  729. * physical memory size.
  730. */
  731. if (xfs_physmem <= 4096) /* <= 16MB */
  732. mp->m_nreadaheads = XFS_RW_NREADAHEAD_16MB;
  733. else if (xfs_physmem <= 8192) /* <= 32MB */
  734. mp->m_nreadaheads = XFS_RW_NREADAHEAD_32MB;
  735. else
  736. mp->m_nreadaheads = XFS_RW_NREADAHEAD_K32;
  737. if (sbp->sb_blocklog > readio_log) {
  738. mp->m_readio_log = sbp->sb_blocklog;
  739. } else {
  740. mp->m_readio_log = readio_log;
  741. }
  742. mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog);
  743. if (sbp->sb_blocklog > writeio_log) {
  744. mp->m_writeio_log = sbp->sb_blocklog;
  745. } else {
  746. mp->m_writeio_log = writeio_log;
  747. }
  748. mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog);
  749. /*
  750. * Set the inode cluster size based on the physical memory
  751. * size. This may still be overridden by the file system
  752. * block size if it is larger than the chosen cluster size.
  753. */
  754. if (xfs_physmem <= btoc(32 * 1024 * 1024)) { /* <= 32 MB */
  755. mp->m_inode_cluster_size = XFS_INODE_SMALL_CLUSTER_SIZE;
  756. } else {
  757. mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE;
  758. }
  759. /*
  760. * Set whether we're using inode alignment.
  761. */
  762. if (XFS_SB_VERSION_HASALIGN(&mp->m_sb) &&
  763. mp->m_sb.sb_inoalignmt >=
  764. XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size))
  765. mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1;
  766. else
  767. mp->m_inoalign_mask = 0;
  768. /*
  769. * If we are using stripe alignment, check whether
  770. * the stripe unit is a multiple of the inode alignment
  771. */
  772. if (mp->m_dalign && mp->m_inoalign_mask &&
  773. !(mp->m_dalign & mp->m_inoalign_mask))
  774. mp->m_sinoalign = mp->m_dalign;
  775. else
  776. mp->m_sinoalign = 0;
  777. /*
  778. * Check that the data (and log if separate) are an ok size.
  779. */
  780. d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
  781. if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
  782. cmn_err(CE_WARN, "XFS: size check 1 failed");
  783. error = XFS_ERROR(E2BIG);
  784. goto error1;
  785. }
  786. error = xfs_read_buf(mp, mp->m_ddev_targp,
  787. d - XFS_FSS_TO_BB(mp, 1),
  788. XFS_FSS_TO_BB(mp, 1), 0, &bp);
  789. if (!error) {
  790. xfs_buf_relse(bp);
  791. } else {
  792. cmn_err(CE_WARN, "XFS: size check 2 failed");
  793. if (error == ENOSPC) {
  794. error = XFS_ERROR(E2BIG);
  795. }
  796. goto error1;
  797. }
  798. if (((mfsi_flags & XFS_MFSI_CLIENT) == 0) &&
  799. mp->m_logdev_targp != mp->m_ddev_targp) {
  800. d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
  801. if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
  802. cmn_err(CE_WARN, "XFS: size check 3 failed");
  803. error = XFS_ERROR(E2BIG);
  804. goto error1;
  805. }
  806. error = xfs_read_buf(mp, mp->m_logdev_targp,
  807. d - XFS_FSB_TO_BB(mp, 1),
  808. XFS_FSB_TO_BB(mp, 1), 0, &bp);
  809. if (!error) {
  810. xfs_buf_relse(bp);
  811. } else {
  812. cmn_err(CE_WARN, "XFS: size check 3 failed");
  813. if (error == ENOSPC) {
  814. error = XFS_ERROR(E2BIG);
  815. }
  816. goto error1;
  817. }
  818. }
  819. /*
  820. * Initialize realtime fields in the mount structure
  821. */
  822. if ((error = xfs_rtmount_init(mp))) {
  823. cmn_err(CE_WARN, "XFS: RT mount failed");
  824. goto error1;
  825. }
  826. /*
  827. * For client case we are done now
  828. */
  829. if (mfsi_flags & XFS_MFSI_CLIENT) {
  830. return 0;
  831. }
  832. /*
  833. * Copies the low order bits of the timestamp and the randomly
  834. * set "sequence" number out of a UUID.
  835. */
  836. uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid);
  837. /*
  838. * The vfs structure needs to have a file system independent
  839. * way of checking for the invariant file system ID. Since it
  840. * can't look at mount structures it has a pointer to the data
  841. * in the mount structure.
  842. *
  843. * File systems that don't support user level file handles (i.e.
  844. * all of them except for XFS) will leave vfs_altfsid as NULL.
  845. */
  846. vfsp->vfs_altfsid = (xfs_fsid_t *)mp->m_fixedfsid;
  847. mp->m_dmevmask = 0; /* not persistent; set after each mount */
  848. xfs_dir_mount(mp);
  849. /*
  850. * Initialize the attribute manager's entries.
  851. */
  852. mp->m_attr_magicpct = (mp->m_sb.sb_blocksize * 37) / 100;
  853. /*
  854. * Initialize the precomputed transaction reservations values.
  855. */
  856. xfs_trans_init(mp);
  857. /*
  858. * Allocate and initialize the inode hash table for this
  859. * file system.
  860. */
  861. xfs_ihash_init(mp);
  862. xfs_chash_init(mp);
  863. /*
  864. * Allocate and initialize the per-ag data.
  865. */
  866. init_rwsem(&mp->m_peraglock);
  867. mp->m_perag =
  868. kmem_zalloc(sbp->sb_agcount * sizeof(xfs_perag_t), KM_SLEEP);
  869. mp->m_maxagi = xfs_initialize_perag(vfsp, mp, sbp->sb_agcount);
  870. /*
  871. * log's mount-time initialization. Perform 1st part recovery if needed
  872. */
  873. if (likely(sbp->sb_logblocks > 0)) { /* check for volume case */
  874. error = xfs_log_mount(mp, mp->m_logdev_targp,
  875. XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
  876. XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
  877. if (error) {
  878. cmn_err(CE_WARN, "XFS: log mount failed");
  879. goto error2;
  880. }
  881. } else { /* No log has been defined */
  882. cmn_err(CE_WARN, "XFS: no log defined");
  883. XFS_ERROR_REPORT("xfs_mountfs_int(1)", XFS_ERRLEVEL_LOW, mp);
  884. error = XFS_ERROR(EFSCORRUPTED);
  885. goto error2;
  886. }
  887. /*
  888. * Get and sanity-check the root inode.
  889. * Save the pointer to it in the mount structure.
  890. */
  891. error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip, 0);
  892. if (error) {
  893. cmn_err(CE_WARN, "XFS: failed to read root inode");
  894. goto error3;
  895. }
  896. ASSERT(rip != NULL);
  897. rvp = XFS_ITOV(rip);
  898. if (unlikely((rip->i_d.di_mode & S_IFMT) != S_IFDIR)) {
  899. cmn_err(CE_WARN, "XFS: corrupted root inode");
  900. cmn_err(CE_WARN, "Device %s - root %llu is not a directory",
  901. XFS_BUFTARG_NAME(mp->m_ddev_targp),
  902. (unsigned long long)rip->i_ino);
  903. xfs_iunlock(rip, XFS_ILOCK_EXCL);
  904. XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW,
  905. mp);
  906. error = XFS_ERROR(EFSCORRUPTED);
  907. goto error4;
  908. }
  909. mp->m_rootip = rip; /* save it */
  910. xfs_iunlock(rip, XFS_ILOCK_EXCL);
  911. /*
  912. * Initialize realtime inode pointers in the mount structure
  913. */
  914. if ((error = xfs_rtmount_inodes(mp))) {
  915. /*
  916. * Free up the root inode.
  917. */
  918. cmn_err(CE_WARN, "XFS: failed to read RT inodes");
  919. goto error4;
  920. }
  921. /*
  922. * If fs is not mounted readonly, then update the superblock
  923. * unit and width changes.
  924. */
  925. if (update_flags && !(vfsp->vfs_flag & VFS_RDONLY))
  926. xfs_mount_log_sbunit(mp, update_flags);
  927. /*
  928. * Initialise the XFS quota management subsystem for this mount
  929. */
  930. if ((error = XFS_QM_INIT(mp, &quotamount, &quotaflags)))
  931. goto error4;
  932. /*
  933. * Finish recovering the file system. This part needed to be
  934. * delayed until after the root and real-time bitmap inodes
  935. * were consistently read in.
  936. */
  937. error = xfs_log_mount_finish(mp, mfsi_flags);
  938. if (error) {
  939. cmn_err(CE_WARN, "XFS: log mount finish failed");
  940. goto error4;
  941. }
  942. /*
  943. * Complete the quota initialisation, post-log-replay component.
  944. */
  945. if ((error = XFS_QM_MOUNT(mp, quotamount, quotaflags, mfsi_flags)))
  946. goto error4;
  947. return 0;
  948. error4:
  949. /*
  950. * Free up the root inode.
  951. */
  952. VN_RELE(rvp);
  953. error3:
  954. xfs_log_unmount_dealloc(mp);
  955. error2:
  956. xfs_ihash_free(mp);
  957. xfs_chash_free(mp);
  958. for (agno = 0; agno < sbp->sb_agcount; agno++)
  959. if (mp->m_perag[agno].pagb_list)
  960. kmem_free(mp->m_perag[agno].pagb_list,
  961. sizeof(xfs_perag_busy_t) * XFS_PAGB_NUM_SLOTS);
  962. kmem_free(mp->m_perag, sbp->sb_agcount * sizeof(xfs_perag_t));
  963. mp->m_perag = NULL;
  964. /* FALLTHROUGH */
  965. error1:
  966. if (uuid_mounted)
  967. xfs_uuid_unmount(mp);
  968. xfs_freesb(mp);
  969. return error;
  970. }
  971. /*
  972. * xfs_unmountfs
  973. *
  974. * This flushes out the inodes,dquots and the superblock, unmounts the
  975. * log and makes sure that incore structures are freed.
  976. */
  977. int
  978. xfs_unmountfs(xfs_mount_t *mp, struct cred *cr)
  979. {
  980. struct bhv_vfs *vfsp = XFS_MTOVFS(mp);
  981. #if defined(DEBUG) || defined(INDUCE_IO_ERROR)
  982. int64_t fsid;
  983. #endif
  984. xfs_iflush_all(mp);
  985. XFS_QM_DQPURGEALL(mp, XFS_QMOPT_QUOTALL | XFS_QMOPT_UMOUNTING);
  986. /*
  987. * Flush out the log synchronously so that we know for sure
  988. * that nothing is pinned. This is important because bflush()
  989. * will skip pinned buffers.
  990. */
  991. xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE | XFS_LOG_SYNC);
  992. xfs_binval(mp->m_ddev_targp);
  993. if (mp->m_rtdev_targp) {
  994. xfs_binval(mp->m_rtdev_targp);
  995. }
  996. xfs_unmountfs_writesb(mp);
  997. xfs_unmountfs_wait(mp); /* wait for async bufs */
  998. xfs_log_unmount(mp); /* Done! No more fs ops. */
  999. xfs_freesb(mp);
  1000. /*
  1001. * All inodes from this mount point should be freed.
  1002. */
  1003. ASSERT(mp->m_inodes == NULL);
  1004. xfs_unmountfs_close(mp, cr);
  1005. if ((mp->m_flags & XFS_MOUNT_NOUUID) == 0)
  1006. xfs_uuid_unmount(mp);
  1007. #if defined(DEBUG) || defined(INDUCE_IO_ERROR)
  1008. /*
  1009. * clear all error tags on this filesystem
  1010. */
  1011. memcpy(&fsid, &vfsp->vfs_fsid, sizeof(int64_t));
  1012. xfs_errortag_clearall_umount(fsid, mp->m_fsname, 0);
  1013. #endif
  1014. XFS_IODONE(vfsp);
  1015. xfs_mount_free(mp, 1);
  1016. return 0;
  1017. }
  1018. void
  1019. xfs_unmountfs_close(xfs_mount_t *mp, struct cred *cr)
  1020. {
  1021. if (mp->m_logdev_targp != mp->m_ddev_targp)
  1022. xfs_free_buftarg(mp->m_logdev_targp, 1);
  1023. if (mp->m_rtdev_targp)
  1024. xfs_free_buftarg(mp->m_rtdev_targp, 1);
  1025. xfs_free_buftarg(mp->m_ddev_targp, 0);
  1026. }
  1027. STATIC void
  1028. xfs_unmountfs_wait(xfs_mount_t *mp)
  1029. {
  1030. if (mp->m_logdev_targp != mp->m_ddev_targp)
  1031. xfs_wait_buftarg(mp->m_logdev_targp);
  1032. if (mp->m_rtdev_targp)
  1033. xfs_wait_buftarg(mp->m_rtdev_targp);
  1034. xfs_wait_buftarg(mp->m_ddev_targp);
  1035. }
  1036. int
  1037. xfs_unmountfs_writesb(xfs_mount_t *mp)
  1038. {
  1039. xfs_buf_t *sbp;
  1040. xfs_sb_t *sb;
  1041. int error = 0;
  1042. /*
  1043. * skip superblock write if fs is read-only, or
  1044. * if we are doing a forced umount.
  1045. */
  1046. sbp = xfs_getsb(mp, 0);
  1047. if (!(XFS_MTOVFS(mp)->vfs_flag & VFS_RDONLY ||
  1048. XFS_FORCED_SHUTDOWN(mp))) {
  1049. xfs_icsb_sync_counters(mp);
  1050. /*
  1051. * mark shared-readonly if desired
  1052. */
  1053. sb = XFS_BUF_TO_SBP(sbp);
  1054. if (mp->m_mk_sharedro) {
  1055. if (!(sb->sb_flags & XFS_SBF_READONLY))
  1056. sb->sb_flags |= XFS_SBF_READONLY;
  1057. if (!XFS_SB_VERSION_HASSHARED(sb))
  1058. XFS_SB_VERSION_ADDSHARED(sb);
  1059. xfs_fs_cmn_err(CE_NOTE, mp,
  1060. "Unmounting, marking shared read-only");
  1061. }
  1062. XFS_BUF_UNDONE(sbp);
  1063. XFS_BUF_UNREAD(sbp);
  1064. XFS_BUF_UNDELAYWRITE(sbp);
  1065. XFS_BUF_WRITE(sbp);
  1066. XFS_BUF_UNASYNC(sbp);
  1067. ASSERT(XFS_BUF_TARGET(sbp) == mp->m_ddev_targp);
  1068. xfsbdstrat(mp, sbp);
  1069. /* Nevermind errors we might get here. */
  1070. error = xfs_iowait(sbp);
  1071. if (error)
  1072. xfs_ioerror_alert("xfs_unmountfs_writesb",
  1073. mp, sbp, XFS_BUF_ADDR(sbp));
  1074. if (error && mp->m_mk_sharedro)
  1075. xfs_fs_cmn_err(CE_ALERT, mp, "Superblock write error detected while unmounting. Filesystem may not be marked shared readonly");
  1076. }
  1077. xfs_buf_relse(sbp);
  1078. return error;
  1079. }
  1080. /*
  1081. * xfs_mod_sb() can be used to copy arbitrary changes to the
  1082. * in-core superblock into the superblock buffer to be logged.
  1083. * It does not provide the higher level of locking that is
  1084. * needed to protect the in-core superblock from concurrent
  1085. * access.
  1086. */
  1087. void
  1088. xfs_mod_sb(xfs_trans_t *tp, __int64_t fields)
  1089. {
  1090. xfs_buf_t *bp;
  1091. int first;
  1092. int last;
  1093. xfs_mount_t *mp;
  1094. xfs_sb_t *sbp;
  1095. xfs_sb_field_t f;
  1096. ASSERT(fields);
  1097. if (!fields)
  1098. return;
  1099. mp = tp->t_mountp;
  1100. bp = xfs_trans_getsb(tp, mp, 0);
  1101. sbp = XFS_BUF_TO_SBP(bp);
  1102. first = sizeof(xfs_sb_t);
  1103. last = 0;
  1104. /* translate/copy */
  1105. xfs_xlatesb(XFS_BUF_PTR(bp), &(mp->m_sb), -1, fields);
  1106. /* find modified range */
  1107. f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
  1108. ASSERT((1LL << f) & XFS_SB_MOD_BITS);
  1109. first = xfs_sb_info[f].offset;
  1110. f = (xfs_sb_field_t)xfs_highbit64((__uint64_t)fields);
  1111. ASSERT((1LL << f) & XFS_SB_MOD_BITS);
  1112. last = xfs_sb_info[f + 1].offset - 1;
  1113. xfs_trans_log_buf(tp, bp, first, last);
  1114. }
  1115. /*
  1116. * xfs_mod_incore_sb_unlocked() is a utility routine common used to apply
  1117. * a delta to a specified field in the in-core superblock. Simply
  1118. * switch on the field indicated and apply the delta to that field.
  1119. * Fields are not allowed to dip below zero, so if the delta would
  1120. * do this do not apply it and return EINVAL.
  1121. *
  1122. * The SB_LOCK must be held when this routine is called.
  1123. */
  1124. int
  1125. xfs_mod_incore_sb_unlocked(
  1126. xfs_mount_t *mp,
  1127. xfs_sb_field_t field,
  1128. int64_t delta,
  1129. int rsvd)
  1130. {
  1131. int scounter; /* short counter for 32 bit fields */
  1132. long long lcounter; /* long counter for 64 bit fields */
  1133. long long res_used, rem;
  1134. /*
  1135. * With the in-core superblock spin lock held, switch
  1136. * on the indicated field. Apply the delta to the
  1137. * proper field. If the fields value would dip below
  1138. * 0, then do not apply the delta and return EINVAL.
  1139. */
  1140. switch (field) {
  1141. case XFS_SBS_ICOUNT:
  1142. lcounter = (long long)mp->m_sb.sb_icount;
  1143. lcounter += delta;
  1144. if (lcounter < 0) {
  1145. ASSERT(0);
  1146. return XFS_ERROR(EINVAL);
  1147. }
  1148. mp->m_sb.sb_icount = lcounter;
  1149. return 0;
  1150. case XFS_SBS_IFREE:
  1151. lcounter = (long long)mp->m_sb.sb_ifree;
  1152. lcounter += delta;
  1153. if (lcounter < 0) {
  1154. ASSERT(0);
  1155. return XFS_ERROR(EINVAL);
  1156. }
  1157. mp->m_sb.sb_ifree = lcounter;
  1158. return 0;
  1159. case XFS_SBS_FDBLOCKS:
  1160. lcounter = (long long)
  1161. mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
  1162. res_used = (long long)(mp->m_resblks - mp->m_resblks_avail);
  1163. if (delta > 0) { /* Putting blocks back */
  1164. if (res_used > delta) {
  1165. mp->m_resblks_avail += delta;
  1166. } else {
  1167. rem = delta - res_used;
  1168. mp->m_resblks_avail = mp->m_resblks;
  1169. lcounter += rem;
  1170. }
  1171. } else { /* Taking blocks away */
  1172. lcounter += delta;
  1173. /*
  1174. * If were out of blocks, use any available reserved blocks if
  1175. * were allowed to.
  1176. */
  1177. if (lcounter < 0) {
  1178. if (rsvd) {
  1179. lcounter = (long long)mp->m_resblks_avail + delta;
  1180. if (lcounter < 0) {
  1181. return XFS_ERROR(ENOSPC);
  1182. }
  1183. mp->m_resblks_avail = lcounter;
  1184. return 0;
  1185. } else { /* not reserved */
  1186. return XFS_ERROR(ENOSPC);
  1187. }
  1188. }
  1189. }
  1190. mp->m_sb.sb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp);
  1191. return 0;
  1192. case XFS_SBS_FREXTENTS:
  1193. lcounter = (long long)mp->m_sb.sb_frextents;
  1194. lcounter += delta;
  1195. if (lcounter < 0) {
  1196. return XFS_ERROR(ENOSPC);
  1197. }
  1198. mp->m_sb.sb_frextents = lcounter;
  1199. return 0;
  1200. case XFS_SBS_DBLOCKS:
  1201. lcounter = (long long)mp->m_sb.sb_dblocks;
  1202. lcounter += delta;
  1203. if (lcounter < 0) {
  1204. ASSERT(0);
  1205. return XFS_ERROR(EINVAL);
  1206. }
  1207. mp->m_sb.sb_dblocks = lcounter;
  1208. return 0;
  1209. case XFS_SBS_AGCOUNT:
  1210. scounter = mp->m_sb.sb_agcount;
  1211. scounter += delta;
  1212. if (scounter < 0) {
  1213. ASSERT(0);
  1214. return XFS_ERROR(EINVAL);
  1215. }
  1216. mp->m_sb.sb_agcount = scounter;
  1217. return 0;
  1218. case XFS_SBS_IMAX_PCT:
  1219. scounter = mp->m_sb.sb_imax_pct;
  1220. scounter += delta;
  1221. if (scounter < 0) {
  1222. ASSERT(0);
  1223. return XFS_ERROR(EINVAL);
  1224. }
  1225. mp->m_sb.sb_imax_pct = scounter;
  1226. return 0;
  1227. case XFS_SBS_REXTSIZE:
  1228. scounter = mp->m_sb.sb_rextsize;
  1229. scounter += delta;
  1230. if (scounter < 0) {
  1231. ASSERT(0);
  1232. return XFS_ERROR(EINVAL);
  1233. }
  1234. mp->m_sb.sb_rextsize = scounter;
  1235. return 0;
  1236. case XFS_SBS_RBMBLOCKS:
  1237. scounter = mp->m_sb.sb_rbmblocks;
  1238. scounter += delta;
  1239. if (scounter < 0) {
  1240. ASSERT(0);
  1241. return XFS_ERROR(EINVAL);
  1242. }
  1243. mp->m_sb.sb_rbmblocks = scounter;
  1244. return 0;
  1245. case XFS_SBS_RBLOCKS:
  1246. lcounter = (long long)mp->m_sb.sb_rblocks;
  1247. lcounter += delta;
  1248. if (lcounter < 0) {
  1249. ASSERT(0);
  1250. return XFS_ERROR(EINVAL);
  1251. }
  1252. mp->m_sb.sb_rblocks = lcounter;
  1253. return 0;
  1254. case XFS_SBS_REXTENTS:
  1255. lcounter = (long long)mp->m_sb.sb_rextents;
  1256. lcounter += delta;
  1257. if (lcounter < 0) {
  1258. ASSERT(0);
  1259. return XFS_ERROR(EINVAL);
  1260. }
  1261. mp->m_sb.sb_rextents = lcounter;
  1262. return 0;
  1263. case XFS_SBS_REXTSLOG:
  1264. scounter = mp->m_sb.sb_rextslog;
  1265. scounter += delta;
  1266. if (scounter < 0) {
  1267. ASSERT(0);
  1268. return XFS_ERROR(EINVAL);
  1269. }
  1270. mp->m_sb.sb_rextslog = scounter;
  1271. return 0;
  1272. default:
  1273. ASSERT(0);
  1274. return XFS_ERROR(EINVAL);
  1275. }
  1276. }
  1277. /*
  1278. * xfs_mod_incore_sb() is used to change a field in the in-core
  1279. * superblock structure by the specified delta. This modification
  1280. * is protected by the SB_LOCK. Just use the xfs_mod_incore_sb_unlocked()
  1281. * routine to do the work.
  1282. */
  1283. int
  1284. xfs_mod_incore_sb(
  1285. xfs_mount_t *mp,
  1286. xfs_sb_field_t field,
  1287. int64_t delta,
  1288. int rsvd)
  1289. {
  1290. unsigned long s;
  1291. int status;
  1292. /* check for per-cpu counters */
  1293. switch (field) {
  1294. #ifdef HAVE_PERCPU_SB
  1295. case XFS_SBS_ICOUNT:
  1296. case XFS_SBS_IFREE:
  1297. case XFS_SBS_FDBLOCKS:
  1298. if (!(mp->m_flags & XFS_MOUNT_NO_PERCPU_SB)) {
  1299. status = xfs_icsb_modify_counters(mp, field,
  1300. delta, rsvd);
  1301. break;
  1302. }
  1303. /* FALLTHROUGH */
  1304. #endif
  1305. default:
  1306. s = XFS_SB_LOCK(mp);
  1307. status = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
  1308. XFS_SB_UNLOCK(mp, s);
  1309. break;
  1310. }
  1311. return status;
  1312. }
  1313. /*
  1314. * xfs_mod_incore_sb_batch() is used to change more than one field
  1315. * in the in-core superblock structure at a time. This modification
  1316. * is protected by a lock internal to this module. The fields and
  1317. * changes to those fields are specified in the array of xfs_mod_sb
  1318. * structures passed in.
  1319. *
  1320. * Either all of the specified deltas will be applied or none of
  1321. * them will. If any modified field dips below 0, then all modifications
  1322. * will be backed out and EINVAL will be returned.
  1323. */
  1324. int
  1325. xfs_mod_incore_sb_batch(xfs_mount_t *mp, xfs_mod_sb_t *msb, uint nmsb, int rsvd)
  1326. {
  1327. unsigned long s;
  1328. int status=0;
  1329. xfs_mod_sb_t *msbp;
  1330. /*
  1331. * Loop through the array of mod structures and apply each
  1332. * individually. If any fail, then back out all those
  1333. * which have already been applied. Do all of this within
  1334. * the scope of the SB_LOCK so that all of the changes will
  1335. * be atomic.
  1336. */
  1337. s = XFS_SB_LOCK(mp);
  1338. msbp = &msb[0];
  1339. for (msbp = &msbp[0]; msbp < (msb + nmsb); msbp++) {
  1340. /*
  1341. * Apply the delta at index n. If it fails, break
  1342. * from the loop so we'll fall into the undo loop
  1343. * below.
  1344. */
  1345. switch (msbp->msb_field) {
  1346. #ifdef HAVE_PERCPU_SB
  1347. case XFS_SBS_ICOUNT:
  1348. case XFS_SBS_IFREE:
  1349. case XFS_SBS_FDBLOCKS:
  1350. if (!(mp->m_flags & XFS_MOUNT_NO_PERCPU_SB)) {
  1351. XFS_SB_UNLOCK(mp, s);
  1352. status = xfs_icsb_modify_counters(mp,
  1353. msbp->msb_field,
  1354. msbp->msb_delta, rsvd);
  1355. s = XFS_SB_LOCK(mp);
  1356. break;
  1357. }
  1358. /* FALLTHROUGH */
  1359. #endif
  1360. default:
  1361. status = xfs_mod_incore_sb_unlocked(mp,
  1362. msbp->msb_field,
  1363. msbp->msb_delta, rsvd);
  1364. break;
  1365. }
  1366. if (status != 0) {
  1367. break;
  1368. }
  1369. }
  1370. /*
  1371. * If we didn't complete the loop above, then back out
  1372. * any changes made to the superblock. If you add code
  1373. * between the loop above and here, make sure that you
  1374. * preserve the value of status. Loop back until
  1375. * we step below the beginning of the array. Make sure
  1376. * we don't touch anything back there.
  1377. */
  1378. if (status != 0) {
  1379. msbp--;
  1380. while (msbp >= msb) {
  1381. switch (msbp->msb_field) {
  1382. #ifdef HAVE_PERCPU_SB
  1383. case XFS_SBS_ICOUNT:
  1384. case XFS_SBS_IFREE:
  1385. case XFS_SBS_FDBLOCKS:
  1386. if (!(mp->m_flags & XFS_MOUNT_NO_PERCPU_SB)) {
  1387. XFS_SB_UNLOCK(mp, s);
  1388. status = xfs_icsb_modify_counters(mp,
  1389. msbp->msb_field,
  1390. -(msbp->msb_delta),
  1391. rsvd);
  1392. s = XFS_SB_LOCK(mp);
  1393. break;
  1394. }
  1395. /* FALLTHROUGH */
  1396. #endif
  1397. default:
  1398. status = xfs_mod_incore_sb_unlocked(mp,
  1399. msbp->msb_field,
  1400. -(msbp->msb_delta),
  1401. rsvd);
  1402. break;
  1403. }
  1404. ASSERT(status == 0);
  1405. msbp--;
  1406. }
  1407. }
  1408. XFS_SB_UNLOCK(mp, s);
  1409. return status;
  1410. }
  1411. /*
  1412. * xfs_getsb() is called to obtain the buffer for the superblock.
  1413. * The buffer is returned locked and read in from disk.
  1414. * The buffer should be released with a call to xfs_brelse().
  1415. *
  1416. * If the flags parameter is BUF_TRYLOCK, then we'll only return
  1417. * the superblock buffer if it can be locked without sleeping.
  1418. * If it can't then we'll return NULL.
  1419. */
  1420. xfs_buf_t *
  1421. xfs_getsb(
  1422. xfs_mount_t *mp,
  1423. int flags)
  1424. {
  1425. xfs_buf_t *bp;
  1426. ASSERT(mp->m_sb_bp != NULL);
  1427. bp = mp->m_sb_bp;
  1428. if (flags & XFS_BUF_TRYLOCK) {
  1429. if (!XFS_BUF_CPSEMA(bp)) {
  1430. return NULL;
  1431. }
  1432. } else {
  1433. XFS_BUF_PSEMA(bp, PRIBIO);
  1434. }
  1435. XFS_BUF_HOLD(bp);
  1436. ASSERT(XFS_BUF_ISDONE(bp));
  1437. return bp;
  1438. }
  1439. /*
  1440. * Used to free the superblock along various error paths.
  1441. */
  1442. void
  1443. xfs_freesb(
  1444. xfs_mount_t *mp)
  1445. {
  1446. xfs_buf_t *bp;
  1447. /*
  1448. * Use xfs_getsb() so that the buffer will be locked
  1449. * when we call xfs_buf_relse().
  1450. */
  1451. bp = xfs_getsb(mp, 0);
  1452. XFS_BUF_UNMANAGE(bp);
  1453. xfs_buf_relse(bp);
  1454. mp->m_sb_bp = NULL;
  1455. }
  1456. /*
  1457. * See if the UUID is unique among mounted XFS filesystems.
  1458. * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
  1459. */
  1460. STATIC int
  1461. xfs_uuid_mount(
  1462. xfs_mount_t *mp)
  1463. {
  1464. if (uuid_is_nil(&mp->m_sb.sb_uuid)) {
  1465. cmn_err(CE_WARN,
  1466. "XFS: Filesystem %s has nil UUID - can't mount",
  1467. mp->m_fsname);
  1468. return -1;
  1469. }
  1470. if (!uuid_table_insert(&mp->m_sb.sb_uuid)) {
  1471. cmn_err(CE_WARN,
  1472. "XFS: Filesystem %s has duplicate UUID - can't mount",
  1473. mp->m_fsname);
  1474. return -1;
  1475. }
  1476. return 0;
  1477. }
  1478. /*
  1479. * Remove filesystem from the UUID table.
  1480. */
  1481. STATIC void
  1482. xfs_uuid_unmount(
  1483. xfs_mount_t *mp)
  1484. {
  1485. uuid_table_remove(&mp->m_sb.sb_uuid);
  1486. }
  1487. /*
  1488. * Used to log changes to the superblock unit and width fields which could
  1489. * be altered by the mount options. Only the first superblock is updated.
  1490. */
  1491. STATIC void
  1492. xfs_mount_log_sbunit(
  1493. xfs_mount_t *mp,
  1494. __int64_t fields)
  1495. {
  1496. xfs_trans_t *tp;
  1497. ASSERT(fields & (XFS_SB_UNIT|XFS_SB_WIDTH|XFS_SB_UUID));
  1498. tp = xfs_trans_alloc(mp, XFS_TRANS_SB_UNIT);
  1499. if (xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0,
  1500. XFS_DEFAULT_LOG_COUNT)) {
  1501. xfs_trans_cancel(tp, 0);
  1502. return;
  1503. }
  1504. xfs_mod_sb(tp, fields);
  1505. xfs_trans_commit(tp, 0, NULL);
  1506. }
  1507. #ifdef HAVE_PERCPU_SB
  1508. /*
  1509. * Per-cpu incore superblock counters
  1510. *
  1511. * Simple concept, difficult implementation
  1512. *
  1513. * Basically, replace the incore superblock counters with a distributed per cpu
  1514. * counter for contended fields (e.g. free block count).
  1515. *
  1516. * Difficulties arise in that the incore sb is used for ENOSPC checking, and
  1517. * hence needs to be accurately read when we are running low on space. Hence
  1518. * there is a method to enable and disable the per-cpu counters based on how
  1519. * much "stuff" is available in them.
  1520. *
  1521. * Basically, a counter is enabled if there is enough free resource to justify
  1522. * running a per-cpu fast-path. If the per-cpu counter runs out (i.e. a local
  1523. * ENOSPC), then we disable the counters to synchronise all callers and
  1524. * re-distribute the available resources.
  1525. *
  1526. * If, once we redistributed the available resources, we still get a failure,
  1527. * we disable the per-cpu counter and go through the slow path.
  1528. *
  1529. * The slow path is the current xfs_mod_incore_sb() function. This means that
  1530. * when we disable a per-cpu counter, we need to drain it's resources back to
  1531. * the global superblock. We do this after disabling the counter to prevent
  1532. * more threads from queueing up on the counter.
  1533. *
  1534. * Essentially, this means that we still need a lock in the fast path to enable
  1535. * synchronisation between the global counters and the per-cpu counters. This
  1536. * is not a problem because the lock will be local to a CPU almost all the time
  1537. * and have little contention except when we get to ENOSPC conditions.
  1538. *
  1539. * Basically, this lock becomes a barrier that enables us to lock out the fast
  1540. * path while we do things like enabling and disabling counters and
  1541. * synchronising the counters.
  1542. *
  1543. * Locking rules:
  1544. *
  1545. * 1. XFS_SB_LOCK() before picking up per-cpu locks
  1546. * 2. per-cpu locks always picked up via for_each_online_cpu() order
  1547. * 3. accurate counter sync requires XFS_SB_LOCK + per cpu locks
  1548. * 4. modifying per-cpu counters requires holding per-cpu lock
  1549. * 5. modifying global counters requires holding XFS_SB_LOCK
  1550. * 6. enabling or disabling a counter requires holding the XFS_SB_LOCK
  1551. * and _none_ of the per-cpu locks.
  1552. *
  1553. * Disabled counters are only ever re-enabled by a balance operation
  1554. * that results in more free resources per CPU than a given threshold.
  1555. * To ensure counters don't remain disabled, they are rebalanced when
  1556. * the global resource goes above a higher threshold (i.e. some hysteresis
  1557. * is present to prevent thrashing).
  1558. */
  1559. #ifdef CONFIG_HOTPLUG_CPU
  1560. /*
  1561. * hot-plug CPU notifier support.
  1562. *
  1563. * We need a notifier per filesystem as we need to be able to identify
  1564. * the filesystem to balance the counters out. This is achieved by
  1565. * having a notifier block embedded in the xfs_mount_t and doing pointer
  1566. * magic to get the mount pointer from the notifier block address.
  1567. */
  1568. STATIC int
  1569. xfs_icsb_cpu_notify(
  1570. struct notifier_block *nfb,
  1571. unsigned long action,
  1572. void *hcpu)
  1573. {
  1574. xfs_icsb_cnts_t *cntp;
  1575. xfs_mount_t *mp;
  1576. int s;
  1577. mp = (xfs_mount_t *)container_of(nfb, xfs_mount_t, m_icsb_notifier);
  1578. cntp = (xfs_icsb_cnts_t *)
  1579. per_cpu_ptr(mp->m_sb_cnts, (unsigned long)hcpu);
  1580. switch (action) {
  1581. case CPU_UP_PREPARE:
  1582. /* Easy Case - initialize the area and locks, and
  1583. * then rebalance when online does everything else for us. */
  1584. memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
  1585. break;
  1586. case CPU_ONLINE:
  1587. xfs_icsb_lock(mp);
  1588. xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0, 0);
  1589. xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0, 0);
  1590. xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0, 0);
  1591. xfs_icsb_unlock(mp);
  1592. break;
  1593. case CPU_DEAD:
  1594. /* Disable all the counters, then fold the dead cpu's
  1595. * count into the total on the global superblock and
  1596. * re-enable the counters. */
  1597. xfs_icsb_lock(mp);
  1598. s = XFS_SB_LOCK(mp);
  1599. xfs_icsb_disable_counter(mp, XFS_SBS_ICOUNT);
  1600. xfs_icsb_disable_counter(mp, XFS_SBS_IFREE);
  1601. xfs_icsb_disable_counter(mp, XFS_SBS_FDBLOCKS);
  1602. mp->m_sb.sb_icount += cntp->icsb_icount;
  1603. mp->m_sb.sb_ifree += cntp->icsb_ifree;
  1604. mp->m_sb.sb_fdblocks += cntp->icsb_fdblocks;
  1605. memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
  1606. xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT,
  1607. XFS_ICSB_SB_LOCKED, 0);
  1608. xfs_icsb_balance_counter(mp, XFS_SBS_IFREE,
  1609. XFS_ICSB_SB_LOCKED, 0);
  1610. xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS,
  1611. XFS_ICSB_SB_LOCKED, 0);
  1612. XFS_SB_UNLOCK(mp, s);
  1613. xfs_icsb_unlock(mp);
  1614. break;
  1615. }
  1616. return NOTIFY_OK;
  1617. }
  1618. #endif /* CONFIG_HOTPLUG_CPU */
  1619. int
  1620. xfs_icsb_init_counters(
  1621. xfs_mount_t *mp)
  1622. {
  1623. xfs_icsb_cnts_t *cntp;
  1624. int i;
  1625. mp->m_sb_cnts = alloc_percpu(xfs_icsb_cnts_t);
  1626. if (mp->m_sb_cnts == NULL)
  1627. return -ENOMEM;
  1628. #ifdef CONFIG_HOTPLUG_CPU
  1629. mp->m_icsb_notifier.notifier_call = xfs_icsb_cpu_notify;
  1630. mp->m_icsb_notifier.priority = 0;
  1631. register_hotcpu_notifier(&mp->m_icsb_notifier);
  1632. #endif /* CONFIG_HOTPLUG_CPU */
  1633. for_each_online_cpu(i) {
  1634. cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
  1635. memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
  1636. }
  1637. mutex_init(&mp->m_icsb_mutex);
  1638. /*
  1639. * start with all counters disabled so that the
  1640. * initial balance kicks us off correctly
  1641. */
  1642. mp->m_icsb_counters = -1;
  1643. return 0;
  1644. }
  1645. void
  1646. xfs_icsb_reinit_counters(
  1647. xfs_mount_t *mp)
  1648. {
  1649. xfs_icsb_lock(mp);
  1650. /*
  1651. * start with all counters disabled so that the
  1652. * initial balance kicks us off correctly
  1653. */
  1654. mp->m_icsb_counters = -1;
  1655. xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0, 0);
  1656. xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0, 0);
  1657. xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0, 0);
  1658. xfs_icsb_unlock(mp);
  1659. }
  1660. STATIC void
  1661. xfs_icsb_destroy_counters(
  1662. xfs_mount_t *mp)
  1663. {
  1664. if (mp->m_sb_cnts) {
  1665. unregister_hotcpu_notifier(&mp->m_icsb_notifier);
  1666. free_percpu(mp->m_sb_cnts);
  1667. }
  1668. mutex_destroy(&mp->m_icsb_mutex);
  1669. }
  1670. STATIC_INLINE void
  1671. xfs_icsb_lock_cntr(
  1672. xfs_icsb_cnts_t *icsbp)
  1673. {
  1674. while (test_and_set_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags)) {
  1675. ndelay(1000);
  1676. }
  1677. }
  1678. STATIC_INLINE void
  1679. xfs_icsb_unlock_cntr(
  1680. xfs_icsb_cnts_t *icsbp)
  1681. {
  1682. clear_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags);
  1683. }
  1684. STATIC_INLINE void
  1685. xfs_icsb_lock_all_counters(
  1686. xfs_mount_t *mp)
  1687. {
  1688. xfs_icsb_cnts_t *cntp;
  1689. int i;
  1690. for_each_online_cpu(i) {
  1691. cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
  1692. xfs_icsb_lock_cntr(cntp);
  1693. }
  1694. }
  1695. STATIC_INLINE void
  1696. xfs_icsb_unlock_all_counters(
  1697. xfs_mount_t *mp)
  1698. {
  1699. xfs_icsb_cnts_t *cntp;
  1700. int i;
  1701. for_each_online_cpu(i) {
  1702. cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
  1703. xfs_icsb_unlock_cntr(cntp);
  1704. }
  1705. }
  1706. STATIC void
  1707. xfs_icsb_count(
  1708. xfs_mount_t *mp,
  1709. xfs_icsb_cnts_t *cnt,
  1710. int flags)
  1711. {
  1712. xfs_icsb_cnts_t *cntp;
  1713. int i;
  1714. memset(cnt, 0, sizeof(xfs_icsb_cnts_t));
  1715. if (!(flags & XFS_ICSB_LAZY_COUNT))
  1716. xfs_icsb_lock_all_counters(mp);
  1717. for_each_online_cpu(i) {
  1718. cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
  1719. cnt->icsb_icount += cntp->icsb_icount;
  1720. cnt->icsb_ifree += cntp->icsb_ifree;
  1721. cnt->icsb_fdblocks += cntp->icsb_fdblocks;
  1722. }
  1723. if (!(flags & XFS_ICSB_LAZY_COUNT))
  1724. xfs_icsb_unlock_all_counters(mp);
  1725. }
  1726. STATIC int
  1727. xfs_icsb_counter_disabled(
  1728. xfs_mount_t *mp,
  1729. xfs_sb_field_t field)
  1730. {
  1731. ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
  1732. return test_bit(field, &mp->m_icsb_counters);
  1733. }
  1734. STATIC int
  1735. xfs_icsb_disable_counter(
  1736. xfs_mount_t *mp,
  1737. xfs_sb_field_t field)
  1738. {
  1739. xfs_icsb_cnts_t cnt;
  1740. ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
  1741. /*
  1742. * If we are already disabled, then there is nothing to do
  1743. * here. We check before locking all the counters to avoid
  1744. * the expensive lock operation when being called in the
  1745. * slow path and the counter is already disabled. This is
  1746. * safe because the only time we set or clear this state is under
  1747. * the m_icsb_mutex.
  1748. */
  1749. if (xfs_icsb_counter_disabled(mp, field))
  1750. return 0;
  1751. xfs_icsb_lock_all_counters(mp);
  1752. if (!test_and_set_bit(field, &mp->m_icsb_counters)) {
  1753. /* drain back to superblock */
  1754. xfs_icsb_count(mp, &cnt, XFS_ICSB_SB_LOCKED|XFS_ICSB_LAZY_COUNT);
  1755. switch(field) {
  1756. case XFS_SBS_ICOUNT:
  1757. mp->m_sb.sb_icount = cnt.icsb_icount;
  1758. break;
  1759. case XFS_SBS_IFREE:
  1760. mp->m_sb.sb_ifree = cnt.icsb_ifree;
  1761. break;
  1762. case XFS_SBS_FDBLOCKS:
  1763. mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
  1764. break;
  1765. default:
  1766. BUG();
  1767. }
  1768. }
  1769. xfs_icsb_unlock_all_counters(mp);
  1770. return 0;
  1771. }
  1772. STATIC void
  1773. xfs_icsb_enable_counter(
  1774. xfs_mount_t *mp,
  1775. xfs_sb_field_t field,
  1776. uint64_t count,
  1777. uint64_t resid)
  1778. {
  1779. xfs_icsb_cnts_t *cntp;
  1780. int i;
  1781. ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
  1782. xfs_icsb_lock_all_counters(mp);
  1783. for_each_online_cpu(i) {
  1784. cntp = per_cpu_ptr(mp->m_sb_cnts, i);
  1785. switch (field) {
  1786. case XFS_SBS_ICOUNT:
  1787. cntp->icsb_icount = count + resid;
  1788. break;
  1789. case XFS_SBS_IFREE:
  1790. cntp->icsb_ifree = count + resid;
  1791. break;
  1792. case XFS_SBS_FDBLOCKS:
  1793. cntp->icsb_fdblocks = count + resid;
  1794. break;
  1795. default:
  1796. BUG();
  1797. break;
  1798. }
  1799. resid = 0;
  1800. }
  1801. clear_bit(field, &mp->m_icsb_counters);
  1802. xfs_icsb_unlock_all_counters(mp);
  1803. }
  1804. void
  1805. xfs_icsb_sync_counters_flags(
  1806. xfs_mount_t *mp,
  1807. int flags)
  1808. {
  1809. xfs_icsb_cnts_t cnt;
  1810. int s;
  1811. /* Pass 1: lock all counters */
  1812. if ((flags & XFS_ICSB_SB_LOCKED) == 0)
  1813. s = XFS_SB_LOCK(mp);
  1814. xfs_icsb_count(mp, &cnt, flags);
  1815. /* Step 3: update mp->m_sb fields */
  1816. if (!xfs_icsb_counter_disabled(mp, XFS_SBS_ICOUNT))
  1817. mp->m_sb.sb_icount = cnt.icsb_icount;
  1818. if (!xfs_icsb_counter_disabled(mp, XFS_SBS_IFREE))
  1819. mp->m_sb.sb_ifree = cnt.icsb_ifree;
  1820. if (!xfs_icsb_counter_disabled(mp, XFS_SBS_FDBLOCKS))
  1821. mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
  1822. if ((flags & XFS_ICSB_SB_LOCKED) == 0)
  1823. XFS_SB_UNLOCK(mp, s);
  1824. }
  1825. /*
  1826. * Accurate update of per-cpu counters to incore superblock
  1827. */
  1828. STATIC void
  1829. xfs_icsb_sync_counters(
  1830. xfs_mount_t *mp)
  1831. {
  1832. xfs_icsb_sync_counters_flags(mp, 0);
  1833. }
  1834. /*
  1835. * Balance and enable/disable counters as necessary.
  1836. *
  1837. * Thresholds for re-enabling counters are somewhat magic. inode counts are
  1838. * chosen to be the same number as single on disk allocation chunk per CPU, and
  1839. * free blocks is something far enough zero that we aren't going thrash when we
  1840. * get near ENOSPC. We also need to supply a minimum we require per cpu to
  1841. * prevent looping endlessly when xfs_alloc_space asks for more than will
  1842. * be distributed to a single CPU but each CPU has enough blocks to be
  1843. * reenabled.
  1844. *
  1845. * Note that we can be called when counters are already disabled.
  1846. * xfs_icsb_disable_counter() optimises the counter locking in this case to
  1847. * prevent locking every per-cpu counter needlessly.
  1848. */
  1849. #define XFS_ICSB_INO_CNTR_REENABLE (uint64_t)64
  1850. #define XFS_ICSB_FDBLK_CNTR_REENABLE(mp) \
  1851. (uint64_t)(512 + XFS_ALLOC_SET_ASIDE(mp))
  1852. STATIC void
  1853. xfs_icsb_balance_counter(
  1854. xfs_mount_t *mp,
  1855. xfs_sb_field_t field,
  1856. int flags,
  1857. int min_per_cpu)
  1858. {
  1859. uint64_t count, resid;
  1860. int weight = num_online_cpus();
  1861. int s;
  1862. uint64_t min = (uint64_t)min_per_cpu;
  1863. if (!(flags & XFS_ICSB_SB_LOCKED))
  1864. s = XFS_SB_LOCK(mp);
  1865. /* disable counter and sync counter */
  1866. xfs_icsb_disable_counter(mp, field);
  1867. /* update counters - first CPU gets residual*/
  1868. switch (field) {
  1869. case XFS_SBS_ICOUNT:
  1870. count = mp->m_sb.sb_icount;
  1871. resid = do_div(count, weight);
  1872. if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE))
  1873. goto out;
  1874. break;
  1875. case XFS_SBS_IFREE:
  1876. count = mp->m_sb.sb_ifree;
  1877. resid = do_div(count, weight);
  1878. if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE))
  1879. goto out;
  1880. break;
  1881. case XFS_SBS_FDBLOCKS:
  1882. count = mp->m_sb.sb_fdblocks;
  1883. resid = do_div(count, weight);
  1884. if (count < max(min, XFS_ICSB_FDBLK_CNTR_REENABLE(mp)))
  1885. goto out;
  1886. break;
  1887. default:
  1888. BUG();
  1889. count = resid = 0; /* quiet, gcc */
  1890. break;
  1891. }
  1892. xfs_icsb_enable_counter(mp, field, count, resid);
  1893. out:
  1894. if (!(flags & XFS_ICSB_SB_LOCKED))
  1895. XFS_SB_UNLOCK(mp, s);
  1896. }
  1897. int
  1898. xfs_icsb_modify_counters(
  1899. xfs_mount_t *mp,
  1900. xfs_sb_field_t field,
  1901. int64_t delta,
  1902. int rsvd)
  1903. {
  1904. xfs_icsb_cnts_t *icsbp;
  1905. long long lcounter; /* long counter for 64 bit fields */
  1906. int cpu, ret = 0, s;
  1907. might_sleep();
  1908. again:
  1909. cpu = get_cpu();
  1910. icsbp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, cpu);
  1911. /*
  1912. * if the counter is disabled, go to slow path
  1913. */
  1914. if (unlikely(xfs_icsb_counter_disabled(mp, field)))
  1915. goto slow_path;
  1916. xfs_icsb_lock_cntr(icsbp);
  1917. if (unlikely(xfs_icsb_counter_disabled(mp, field))) {
  1918. xfs_icsb_unlock_cntr(icsbp);
  1919. goto slow_path;
  1920. }
  1921. switch (field) {
  1922. case XFS_SBS_ICOUNT:
  1923. lcounter = icsbp->icsb_icount;
  1924. lcounter += delta;
  1925. if (unlikely(lcounter < 0))
  1926. goto balance_counter;
  1927. icsbp->icsb_icount = lcounter;
  1928. break;
  1929. case XFS_SBS_IFREE:
  1930. lcounter = icsbp->icsb_ifree;
  1931. lcounter += delta;
  1932. if (unlikely(lcounter < 0))
  1933. goto balance_counter;
  1934. icsbp->icsb_ifree = lcounter;
  1935. break;
  1936. case XFS_SBS_FDBLOCKS:
  1937. BUG_ON((mp->m_resblks - mp->m_resblks_avail) != 0);
  1938. lcounter = icsbp->icsb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
  1939. lcounter += delta;
  1940. if (unlikely(lcounter < 0))
  1941. goto balance_counter;
  1942. icsbp->icsb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp);
  1943. break;
  1944. default:
  1945. BUG();
  1946. break;
  1947. }
  1948. xfs_icsb_unlock_cntr(icsbp);
  1949. put_cpu();
  1950. return 0;
  1951. slow_path:
  1952. put_cpu();
  1953. /*
  1954. * serialise with a mutex so we don't burn lots of cpu on
  1955. * the superblock lock. We still need to hold the superblock
  1956. * lock, however, when we modify the global structures.
  1957. */
  1958. xfs_icsb_lock(mp);
  1959. /*
  1960. * Now running atomically.
  1961. *
  1962. * If the counter is enabled, someone has beaten us to rebalancing.
  1963. * Drop the lock and try again in the fast path....
  1964. */
  1965. if (!(xfs_icsb_counter_disabled(mp, field))) {
  1966. xfs_icsb_unlock(mp);
  1967. goto again;
  1968. }
  1969. /*
  1970. * The counter is currently disabled. Because we are
  1971. * running atomically here, we know a rebalance cannot
  1972. * be in progress. Hence we can go straight to operating
  1973. * on the global superblock. We do not call xfs_mod_incore_sb()
  1974. * here even though we need to get the SB_LOCK. Doing so
  1975. * will cause us to re-enter this function and deadlock.
  1976. * Hence we get the SB_LOCK ourselves and then call
  1977. * xfs_mod_incore_sb_unlocked() as the unlocked path operates
  1978. * directly on the global counters.
  1979. */
  1980. s = XFS_SB_LOCK(mp);
  1981. ret = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
  1982. XFS_SB_UNLOCK(mp, s);
  1983. /*
  1984. * Now that we've modified the global superblock, we
  1985. * may be able to re-enable the distributed counters
  1986. * (e.g. lots of space just got freed). After that
  1987. * we are done.
  1988. */
  1989. if (ret != ENOSPC)
  1990. xfs_icsb_balance_counter(mp, field, 0, 0);
  1991. xfs_icsb_unlock(mp);
  1992. return ret;
  1993. balance_counter:
  1994. xfs_icsb_unlock_cntr(icsbp);
  1995. put_cpu();
  1996. /*
  1997. * We may have multiple threads here if multiple per-cpu
  1998. * counters run dry at the same time. This will mean we can
  1999. * do more balances than strictly necessary but it is not
  2000. * the common slowpath case.
  2001. */
  2002. xfs_icsb_lock(mp);
  2003. /*
  2004. * running atomically.
  2005. *
  2006. * This will leave the counter in the correct state for future
  2007. * accesses. After the rebalance, we simply try again and our retry
  2008. * will either succeed through the fast path or slow path without
  2009. * another balance operation being required.
  2010. */
  2011. xfs_icsb_balance_counter(mp, field, 0, delta);
  2012. xfs_icsb_unlock(mp);
  2013. goto again;
  2014. }
  2015. #endif