fs/xfs/linux-2.6/xfs_super.c

   1 /*
   2  * Copyright (c) 2000-2006 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_bit.h"
  20 #include "xfs_log.h"
  21 #include "xfs_inum.h"
  22 #include "xfs_trans.h"
  23 #include "xfs_sb.h"
  24 #include "xfs_ag.h"
  25 #include "xfs_dir2.h"
  26 #include "xfs_alloc.h"
  27 #include "xfs_dmapi.h"
  28 #include "xfs_quota.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_btree_trace.h"
  39 #include "xfs_ialloc.h"
  40 #include "xfs_bmap.h"
  41 #include "xfs_rtalloc.h"
  42 #include "xfs_error.h"
  43 #include "xfs_itable.h"
  44 #include "xfs_fsops.h"
  45 #include "xfs_rw.h"
  46 #include "xfs_acl.h"
  47 #include "xfs_attr.h"
  48 #include "xfs_buf_item.h"
  49 #include "xfs_utils.h"
  50 #include "xfs_vnodeops.h"
  51 #include "xfs_vfsops.h"
  52 #include "xfs_version.h"
  53 #include "xfs_log_priv.h"
  54 #include "xfs_trans_priv.h"
  55 #include "xfs_filestream.h"
  56 #include "xfs_da_btree.h"
  57 #include "xfs_dir2_trace.h"
  58 #include "xfs_extfree_item.h"
  59 #include "xfs_mru_cache.h"
  60 #include "xfs_inode_item.h"
  61 #include "xfs_sync.h"
  62
  63 #include <linux/namei.h>
  64 #include <linux/init.h>
  65 #include <linux/mount.h>
  66 #include <linux/mempool.h>
  67 #include <linux/writeback.h>
  68 #include <linux/kthread.h>
  69 #include <linux/freezer.h>
  70 #include <linux/parser.h>
  71
  72 static struct quotactl_ops xfs_quotactl_operations;
  73 static struct super_operations xfs_super_operations;
  74 static kmem_zone_t *xfs_ioend_zone;
  75 mempool_t *xfs_ioend_pool;
  76
  77 #define MNTOPT_LOGBUFS  "logbufs"       /* number of XFS log buffers */
  78 #define MNTOPT_LOGBSIZE "logbsize"      /* size of XFS log buffers */
  79 #define MNTOPT_LOGDEV   "logdev"        /* log device */
  80 #define MNTOPT_RTDEV    "rtdev"         /* realtime I/O device */
  81 #define MNTOPT_BIOSIZE  "biosize"       /* log2 of preferred buffered io size */
  82 #define MNTOPT_WSYNC    "wsync"         /* safe-mode nfs compatible mount */
  83 #define MNTOPT_INO64    "ino64"         /* force inodes into 64-bit range */
  84 #define MNTOPT_NOALIGN  "noalign"       /* turn off stripe alignment */
  85 #define MNTOPT_SWALLOC  "swalloc"       /* turn on stripe width allocation */
  86 #define MNTOPT_SUNIT    "sunit"         /* data volume stripe unit */
  87 #define MNTOPT_SWIDTH   "swidth"        /* data volume stripe width */
  88 #define MNTOPT_NOUUID   "nouuid"        /* ignore filesystem UUID */
  89 #define MNTOPT_MTPT     "mtpt"          /* filesystem mount point */
  90 #define MNTOPT_GRPID    "grpid"         /* group-ID from parent directory */
  91 #define MNTOPT_NOGRPID  "nogrpid"       /* group-ID from current process */
  92 #define MNTOPT_BSDGROUPS    "bsdgroups"    /* group-ID from parent directory */
  93 #define MNTOPT_SYSVGROUPS   "sysvgroups"   /* group-ID from current process */
  94 #define MNTOPT_ALLOCSIZE    "allocsize"    /* preferred allocation size */
  95 #define MNTOPT_NORECOVERY   "norecovery"   /* don't run XFS recovery */
  96 #define MNTOPT_BARRIER  "barrier"       /* use writer barriers for log write and
  97                                          * unwritten extent conversion */
  98 #define MNTOPT_NOBARRIER "nobarrier"    /* .. disable */
  99 #define MNTOPT_OSYNCISOSYNC "osyncisosync" /* o_sync is REALLY o_sync */
 100 #define MNTOPT_64BITINODE   "inode64"   /* inodes can be allocated anywhere */
 101 #define MNTOPT_IKEEP    "ikeep"         /* do not free empty inode clusters */
 102 #define MNTOPT_NOIKEEP  "noikeep"       /* free empty inode clusters */
 103 #define MNTOPT_LARGEIO     "largeio"    /* report large I/O sizes in stat() */
 104 #define MNTOPT_NOLARGEIO   "nolargeio"  /* do not report large I/O sizes
 105                                          * in stat(). */
 106 #define MNTOPT_ATTR2    "attr2"         /* do use attr2 attribute format */
 107 #define MNTOPT_NOATTR2  "noattr2"       /* do not use attr2 attribute format */
 108 #define MNTOPT_FILESTREAM  "filestreams" /* use filestreams allocator */
 109 #define MNTOPT_QUOTA    "quota"         /* disk quotas (user) */
 110 #define MNTOPT_NOQUOTA  "noquota"       /* no quotas */
 111 #define MNTOPT_USRQUOTA "usrquota"      /* user quota enabled */
 112 #define MNTOPT_GRPQUOTA "grpquota"      /* group quota enabled */
 113 #define MNTOPT_PRJQUOTA "prjquota"      /* project quota enabled */
 114 #define MNTOPT_UQUOTA   "uquota"        /* user quota (IRIX variant) */
 115 #define MNTOPT_GQUOTA   "gquota"        /* group quota (IRIX variant) */
 116 #define MNTOPT_PQUOTA   "pquota"        /* project quota (IRIX variant) */
 117 #define MNTOPT_UQUOTANOENF "uqnoenforce"/* user quota limit enforcement */
 118 #define MNTOPT_GQUOTANOENF "gqnoenforce"/* group quota limit enforcement */
 119 #define MNTOPT_PQUOTANOENF "pqnoenforce"/* project quota limit enforcement */
 120 #define MNTOPT_QUOTANOENF  "qnoenforce" /* same as uqnoenforce */
 121 #define MNTOPT_DMAPI    "dmapi"         /* DMI enabled (DMAPI / XDSM) */
 122 #define MNTOPT_XDSM     "xdsm"          /* DMI enabled (DMAPI / XDSM) */
 123 #define MNTOPT_DMI      "dmi"           /* DMI enabled (DMAPI / XDSM) */
 124
 125 /*
 126  * Table driven mount option parser.
 127  *
 128  * Currently only used for remount, but it will be used for mount
 129  * in the future, too.
 130  */
 131 enum {
 132         Opt_barrier, Opt_nobarrier, Opt_err
 133 };
 134
 135 static const match_table_t tokens = {
 136         {Opt_barrier, "barrier"},
 137         {Opt_nobarrier, "nobarrier"},
 138         {Opt_err, NULL}
 139 };
 140
 141
 142 STATIC unsigned long
 143 suffix_strtoul(char *s, char **endp, unsigned int base)
 144 {
 145         int     last, shift_left_factor = 0;
 146         char    *value = s;
 147
 148         last = strlen(value) - 1;
 149         if (value[last] == 'K' || value[last] == 'k') {
 150                 shift_left_factor = 10;
 151                 value[last] = '\0';
 152         }
 153         if (value[last] == 'M' || value[last] == 'm') {
 154                 shift_left_factor = 20;
 155                 value[last] = '\0';
 156         }
 157         if (value[last] == 'G' || value[last] == 'g') {
 158                 shift_left_factor = 30;
 159                 value[last] = '\0';
 160         }
 161
 162         return simple_strtoul((const char *)s, endp, base) << shift_left_factor;
 163 }
 164
 165 /*
 166  * This function fills in xfs_mount_t fields based on mount args.
 167  * Note: the superblock has _not_ yet been read in.
 168  *
 169  * Note that this function leaks the various device name allocations on
 170  * failure.  The caller takes care of them.
 171  */
 172 STATIC int
 173 xfs_parseargs(
 174         struct xfs_mount        *mp,
 175         char                    *options,
 176         char                    **mtpt)
 177 {
 178         struct super_block      *sb = mp->m_super;
 179         char                    *this_char, *value, *eov;
 180         int                     dsunit = 0;
 181         int                     dswidth = 0;
 182         int                     iosize = 0;
 183         int                     dmapi_implies_ikeep = 1;
 184         uchar_t                 iosizelog = 0;
 185
 186         /*
 187          * Copy binary VFS mount flags we are interested in.
 188          */
 189         if (sb->s_flags & MS_RDONLY)
 190                 mp->m_flags |= XFS_MOUNT_RDONLY;
 191         if (sb->s_flags & MS_DIRSYNC)
 192                 mp->m_flags |= XFS_MOUNT_DIRSYNC;
 193         if (sb->s_flags & MS_SYNCHRONOUS)
 194                 mp->m_flags |= XFS_MOUNT_WSYNC;
 195
 196         /*
 197          * Set some default flags that could be cleared by the mount option
 198          * parsing.
 199          */
 200         mp->m_flags |= XFS_MOUNT_BARRIER;
 201         mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE;
 202         mp->m_flags |= XFS_MOUNT_SMALL_INUMS;
 203
 204         /*
 205          * These can be overridden by the mount option parsing.
 206          */
 207         mp->m_logbufs = -1;
 208         mp->m_logbsize = -1;
 209
 210         if (!options)
 211                 goto done;
 212
 213         while ((this_char = strsep(&options, ",")) != NULL) {
 214                 if (!*this_char)
 215                         continue;
 216                 if ((value = strchr(this_char, '=')) != NULL)
 217                         *value++ = 0;
 218
 219                 if (!strcmp(this_char, MNTOPT_LOGBUFS)) {
 220                         if (!value || !*value) {
 221                                 cmn_err(CE_WARN,
 222                                         "XFS: %s option requires an argument",
 223                                         this_char);
 224                                 return EINVAL;
 225                         }
 226                         mp->m_logbufs = simple_strtoul(value, &eov, 10);
 227                 } else if (!strcmp(this_char, MNTOPT_LOGBSIZE)) {
 228                         if (!value || !*value) {
 229                                 cmn_err(CE_WARN,
 230                                         "XFS: %s option requires an argument",
 231                                         this_char);
 232                                 return EINVAL;
 233                         }
 234                         mp->m_logbsize = suffix_strtoul(value, &eov, 10);
 235                 } else if (!strcmp(this_char, MNTOPT_LOGDEV)) {
 236                         if (!value || !*value) {
 237                                 cmn_err(CE_WARN,
 238                                         "XFS: %s option requires an argument",
 239                                         this_char);
 240                                 return EINVAL;
 241                         }
 242                         mp->m_logname = kstrndup(value, MAXNAMELEN, GFP_KERNEL);
 243                         if (!mp->m_logname)
 244                                 return ENOMEM;
 245                 } else if (!strcmp(this_char, MNTOPT_MTPT)) {
 246                         if (!value || !*value) {
 247                                 cmn_err(CE_WARN,
 248                                         "XFS: %s option requires an argument",
 249                                         this_char);
 250                                 return EINVAL;
 251                         }
 252                         *mtpt = kstrndup(value, MAXNAMELEN, GFP_KERNEL);
 253                         if (!*mtpt)
 254                                 return ENOMEM;
 255                 } else if (!strcmp(this_char, MNTOPT_RTDEV)) {
 256                         if (!value || !*value) {
 257                                 cmn_err(CE_WARN,
 258                                         "XFS: %s option requires an argument",
 259                                         this_char);
 260                                 return EINVAL;
 261                         }
 262                         mp->m_rtname = kstrndup(value, MAXNAMELEN, GFP_KERNEL);
 263                         if (!mp->m_rtname)
 264                                 return ENOMEM;
 265                 } else if (!strcmp(this_char, MNTOPT_BIOSIZE)) {
 266                         if (!value || !*value) {
 267                                 cmn_err(CE_WARN,
 268                                         "XFS: %s option requires an argument",
 269                                         this_char);
 270                                 return EINVAL;
 271                         }
 272                         iosize = simple_strtoul(value, &eov, 10);
 273                         iosizelog = (uint8_t) iosize;
 274                 } else if (!strcmp(this_char, MNTOPT_ALLOCSIZE)) {
 275                         if (!value || !*value) {
 276                                 cmn_err(CE_WARN,
 277                                         "XFS: %s option requires an argument",
 278                                         this_char);
 279                                 return EINVAL;
 280                         }
 281                         iosize = suffix_strtoul(value, &eov, 10);
 282                         iosizelog = ffs(iosize) - 1;
 283                 } else if (!strcmp(this_char, MNTOPT_GRPID) ||
 284                            !strcmp(this_char, MNTOPT_BSDGROUPS)) {
 285                         mp->m_flags |= XFS_MOUNT_GRPID;
 286                 } else if (!strcmp(this_char, MNTOPT_NOGRPID) ||
 287                            !strcmp(this_char, MNTOPT_SYSVGROUPS)) {
 288                         mp->m_flags &= ~XFS_MOUNT_GRPID;
 289                 } else if (!strcmp(this_char, MNTOPT_WSYNC)) {
 290                         mp->m_flags |= XFS_MOUNT_WSYNC;
 291                 } else if (!strcmp(this_char, MNTOPT_OSYNCISOSYNC)) {
 292                         mp->m_flags |= XFS_MOUNT_OSYNCISOSYNC;
 293                 } else if (!strcmp(this_char, MNTOPT_NORECOVERY)) {
 294                         mp->m_flags |= XFS_MOUNT_NORECOVERY;
 295                 } else if (!strcmp(this_char, MNTOPT_INO64)) {
 296 #if XFS_BIG_INUMS
 297                         mp->m_flags |= XFS_MOUNT_INO64;
 298                         mp->m_inoadd = XFS_INO64_OFFSET;
 299 #else
 300                         cmn_err(CE_WARN,
 301                                 "XFS: %s option not allowed on this system",
 302                                 this_char);
 303                         return EINVAL;
 304 #endif
 305                 } else if (!strcmp(this_char, MNTOPT_NOALIGN)) {
 306                         mp->m_flags |= XFS_MOUNT_NOALIGN;
 307                 } else if (!strcmp(this_char, MNTOPT_SWALLOC)) {
 308                         mp->m_flags |= XFS_MOUNT_SWALLOC;
 309                 } else if (!strcmp(this_char, MNTOPT_SUNIT)) {
 310                         if (!value || !*value) {
 311                                 cmn_err(CE_WARN,
 312                                         "XFS: %s option requires an argument",
 313                                         this_char);
 314                                 return EINVAL;
 315                         }
 316                         dsunit = simple_strtoul(value, &eov, 10);
 317                 } else if (!strcmp(this_char, MNTOPT_SWIDTH)) {
 318                         if (!value || !*value) {
 319                                 cmn_err(CE_WARN,
 320                                         "XFS: %s option requires an argument",
 321                                         this_char);
 322                                 return EINVAL;
 323                         }
 324                         dswidth = simple_strtoul(value, &eov, 10);
 325                 } else if (!strcmp(this_char, MNTOPT_64BITINODE)) {
 326                         mp->m_flags &= ~XFS_MOUNT_SMALL_INUMS;
 327 #if !XFS_BIG_INUMS
 328                         cmn_err(CE_WARN,
 329                                 "XFS: %s option not allowed on this system",
 330                                 this_char);
 331                         return EINVAL;
 332 #endif
 333                 } else if (!strcmp(this_char, MNTOPT_NOUUID)) {
 334                         mp->m_flags |= XFS_MOUNT_NOUUID;
 335                 } else if (!strcmp(this_char, MNTOPT_BARRIER)) {
 336                         mp->m_flags |= XFS_MOUNT_BARRIER;
 337                 } else if (!strcmp(this_char, MNTOPT_NOBARRIER)) {
 338                         mp->m_flags &= ~XFS_MOUNT_BARRIER;
 339                 } else if (!strcmp(this_char, MNTOPT_IKEEP)) {
 340                         mp->m_flags |= XFS_MOUNT_IKEEP;
 341                 } else if (!strcmp(this_char, MNTOPT_NOIKEEP)) {
 342                         dmapi_implies_ikeep = 0;
 343                         mp->m_flags &= ~XFS_MOUNT_IKEEP;
 344                 } else if (!strcmp(this_char, MNTOPT_LARGEIO)) {
 345                         mp->m_flags &= ~XFS_MOUNT_COMPAT_IOSIZE;
 346                 } else if (!strcmp(this_char, MNTOPT_NOLARGEIO)) {
 347                         mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE;
 348                 } else if (!strcmp(this_char, MNTOPT_ATTR2)) {
 349                         mp->m_flags |= XFS_MOUNT_ATTR2;
 350                 } else if (!strcmp(this_char, MNTOPT_NOATTR2)) {
 351                         mp->m_flags &= ~XFS_MOUNT_ATTR2;
 352                         mp->m_flags |= XFS_MOUNT_NOATTR2;
 353                 } else if (!strcmp(this_char, MNTOPT_FILESTREAM)) {
 354                         mp->m_flags |= XFS_MOUNT_FILESTREAMS;
 355                 } else if (!strcmp(this_char, MNTOPT_NOQUOTA)) {
 356                         mp->m_qflags &= ~(XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE |
 357                                           XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE |
 358                                           XFS_UQUOTA_ENFD | XFS_OQUOTA_ENFD);
 359                 } else if (!strcmp(this_char, MNTOPT_QUOTA) ||
 360                            !strcmp(this_char, MNTOPT_UQUOTA) ||
 361                            !strcmp(this_char, MNTOPT_USRQUOTA)) {
 362                         mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE |
 363                                          XFS_UQUOTA_ENFD);
 364                 } else if (!strcmp(this_char, MNTOPT_QUOTANOENF) ||
 365                            !strcmp(this_char, MNTOPT_UQUOTANOENF)) {
 366                         mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE);
 367                         mp->m_qflags &= ~XFS_UQUOTA_ENFD;
 368                 } else if (!strcmp(this_char, MNTOPT_PQUOTA) ||
 369                            !strcmp(this_char, MNTOPT_PRJQUOTA)) {
 370                         mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE |
 371                                          XFS_OQUOTA_ENFD);
 372                 } else if (!strcmp(this_char, MNTOPT_PQUOTANOENF)) {
 373                         mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE);
 374                         mp->m_qflags &= ~XFS_OQUOTA_ENFD;
 375                 } else if (!strcmp(this_char, MNTOPT_GQUOTA) ||
 376                            !strcmp(this_char, MNTOPT_GRPQUOTA)) {
 377                         mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE |
 378                                          XFS_OQUOTA_ENFD);
 379                 } else if (!strcmp(this_char, MNTOPT_GQUOTANOENF)) {
 380                         mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE);
 381                         mp->m_qflags &= ~XFS_OQUOTA_ENFD;
 382                 } else if (!strcmp(this_char, MNTOPT_DMAPI)) {
 383                         mp->m_flags |= XFS_MOUNT_DMAPI;
 384                 } else if (!strcmp(this_char, MNTOPT_XDSM)) {
 385                         mp->m_flags |= XFS_MOUNT_DMAPI;
 386                 } else if (!strcmp(this_char, MNTOPT_DMI)) {
 387                         mp->m_flags |= XFS_MOUNT_DMAPI;
 388                 } else if (!strcmp(this_char, "ihashsize")) {
 389                         cmn_err(CE_WARN,
 390         "XFS: ihashsize no longer used, option is deprecated.");
 391                 } else if (!strcmp(this_char, "osyncisdsync")) {
 392                         /* no-op, this is now the default */
 393                         cmn_err(CE_WARN,
 394         "XFS: osyncisdsync is now the default, option is deprecated.");
 395                 } else if (!strcmp(this_char, "irixsgid")) {
 396                         cmn_err(CE_WARN,
 397         "XFS: irixsgid is now a sysctl(2) variable, option is deprecated.");
 398                 } else {
 399                         cmn_err(CE_WARN,
 400                                 "XFS: unknown mount option [%s].", this_char);
 401                         return EINVAL;
 402                 }
 403         }
 404
 405         /*
 406          * no recovery flag requires a read-only mount
 407          */
 408         if ((mp->m_flags & XFS_MOUNT_NORECOVERY) &&
 409             !(mp->m_flags & XFS_MOUNT_RDONLY)) {
 410                 cmn_err(CE_WARN, "XFS: no-recovery mounts must be read-only.");
 411                 return EINVAL;
 412         }
 413
 414         if ((mp->m_flags & XFS_MOUNT_NOALIGN) && (dsunit || dswidth)) {
 415                 cmn_err(CE_WARN,
 416         "XFS: sunit and swidth options incompatible with the noalign option");
 417                 return EINVAL;
 418         }
 419
 420         if ((mp->m_qflags & (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE)) &&
 421             (mp->m_qflags & (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE))) {
 422                 cmn_err(CE_WARN,
 423                         "XFS: cannot mount with both project and group quota");
 424                 return EINVAL;
 425         }
 426
 427         if ((mp->m_flags & XFS_MOUNT_DMAPI) && (!*mtpt || *mtpt[0] == '\0')) {
 428                 printk("XFS: %s option needs the mount point option as well\n",
 429                         MNTOPT_DMAPI);
 430                 return EINVAL;
 431         }
 432
 433         if ((dsunit && !dswidth) || (!dsunit && dswidth)) {
 434                 cmn_err(CE_WARN,
 435                         "XFS: sunit and swidth must be specified together");
 436                 return EINVAL;
 437         }
 438
 439         if (dsunit && (dswidth % dsunit != 0)) {
 440                 cmn_err(CE_WARN,
 441         "XFS: stripe width (%d) must be a multiple of the stripe unit (%d)",
 442                         dswidth, dsunit);
 443                 return EINVAL;
 444         }
 445
 446         /*
 447          * Applications using DMI filesystems often expect the
 448          * inode generation number to be monotonically increasing.
 449          * If we delete inode chunks we break this assumption, so
 450          * keep unused inode chunks on disk for DMI filesystems
 451          * until we come up with a better solution.
 452          * Note that if "ikeep" or "noikeep" mount options are
 453          * supplied, then they are honored.
 454          */
 455         if ((mp->m_flags & XFS_MOUNT_DMAPI) && dmapi_implies_ikeep)
 456                 mp->m_flags |= XFS_MOUNT_IKEEP;
 457
 458 done:
 459         if (!(mp->m_flags & XFS_MOUNT_NOALIGN)) {
 460                 /*
 461                  * At this point the superblock has not been read
 462                  * in, therefore we do not know the block size.
 463                  * Before the mount call ends we will convert
 464                  * these to FSBs.
 465                  */
 466                 if (dsunit) {
 467                         mp->m_dalign = dsunit;
 468                         mp->m_flags |= XFS_MOUNT_RETERR;
 469                 }
 470
 471                 if (dswidth)
 472                         mp->m_swidth = dswidth;
 473         }
 474
 475         if (mp->m_logbufs != -1 &&
 476             mp->m_logbufs != 0 &&
 477             (mp->m_logbufs < XLOG_MIN_ICLOGS ||
 478              mp->m_logbufs > XLOG_MAX_ICLOGS)) {
 479                 cmn_err(CE_WARN,
 480                         "XFS: invalid logbufs value: %d [not %d-%d]",
 481                         mp->m_logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
 482                 return XFS_ERROR(EINVAL);
 483         }
 484         if (mp->m_logbsize != -1 &&
 485             mp->m_logbsize !=  0 &&
 486             (mp->m_logbsize < XLOG_MIN_RECORD_BSIZE ||
 487              mp->m_logbsize > XLOG_MAX_RECORD_BSIZE ||
 488              !is_power_of_2(mp->m_logbsize))) {
 489                 cmn_err(CE_WARN,
 490         "XFS: invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
 491                         mp->m_logbsize);
 492                 return XFS_ERROR(EINVAL);
 493         }
 494
 495         mp->m_fsname = kstrndup(sb->s_id, MAXNAMELEN, GFP_KERNEL);
 496         if (!mp->m_fsname)
 497                 return ENOMEM;
 498         mp->m_fsname_len = strlen(mp->m_fsname) + 1;
 499
 500         if (iosizelog) {
 501                 if (iosizelog > XFS_MAX_IO_LOG ||
 502                     iosizelog < XFS_MIN_IO_LOG) {
 503                         cmn_err(CE_WARN,
 504                 "XFS: invalid log iosize: %d [not %d-%d]",
 505                                 iosizelog, XFS_MIN_IO_LOG,
 506                                 XFS_MAX_IO_LOG);
 507                         return XFS_ERROR(EINVAL);
 508                 }
 509
 510                 mp->m_flags |= XFS_MOUNT_DFLT_IOSIZE;
 511                 mp->m_readio_log = iosizelog;
 512                 mp->m_writeio_log = iosizelog;
 513         }
 514
 515         return 0;
 516 }
 517
 518 struct proc_xfs_info {
 519         int     flag;
 520         char    *str;
 521 };
 522
 523 STATIC int
 524 xfs_showargs(
 525         struct xfs_mount        *mp,
 526         struct seq_file         *m)
 527 {
 528         static struct proc_xfs_info xfs_info_set[] = {
 529                 /* the few simple ones we can get from the mount struct */
 530                 { XFS_MOUNT_IKEEP,              "," MNTOPT_IKEEP },
 531                 { XFS_MOUNT_WSYNC,              "," MNTOPT_WSYNC },
 532                 { XFS_MOUNT_INO64,              "," MNTOPT_INO64 },
 533                 { XFS_MOUNT_NOALIGN,            "," MNTOPT_NOALIGN },
 534                 { XFS_MOUNT_SWALLOC,            "," MNTOPT_SWALLOC },
 535                 { XFS_MOUNT_NOUUID,             "," MNTOPT_NOUUID },
 536                 { XFS_MOUNT_NORECOVERY,         "," MNTOPT_NORECOVERY },
 537                 { XFS_MOUNT_OSYNCISOSYNC,       "," MNTOPT_OSYNCISOSYNC },
 538                 { XFS_MOUNT_ATTR2,              "," MNTOPT_ATTR2 },
 539                 { XFS_MOUNT_FILESTREAMS,        "," MNTOPT_FILESTREAM },
 540                 { XFS_MOUNT_DMAPI,              "," MNTOPT_DMAPI },
 541                 { XFS_MOUNT_GRPID,              "," MNTOPT_GRPID },
 542                 { 0, NULL }
 543         };
 544         static struct proc_xfs_info xfs_info_unset[] = {
 545                 /* the few simple ones we can get from the mount struct */
 546                 { XFS_MOUNT_COMPAT_IOSIZE,      "," MNTOPT_LARGEIO },
 547                 { XFS_MOUNT_BARRIER,            "," MNTOPT_NOBARRIER },
 548                 { XFS_MOUNT_SMALL_INUMS,        "," MNTOPT_64BITINODE },
 549                 { 0, NULL }
 550         };
 551         struct proc_xfs_info    *xfs_infop;
 552
 553         for (xfs_infop = xfs_info_set; xfs_infop->flag; xfs_infop++) {
 554                 if (mp->m_flags & xfs_infop->flag)
 555                         seq_puts(m, xfs_infop->str);
 556         }
 557         for (xfs_infop = xfs_info_unset; xfs_infop->flag; xfs_infop++) {
 558                 if (!(mp->m_flags & xfs_infop->flag))
 559                         seq_puts(m, xfs_infop->str);
 560         }
 561
 562         if (mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)
 563                 seq_printf(m, "," MNTOPT_ALLOCSIZE "=%dk",
 564                                 (int)(1 << mp->m_writeio_log) >> 10);
 565
 566         if (mp->m_logbufs > 0)
 567                 seq_printf(m, "," MNTOPT_LOGBUFS "=%d", mp->m_logbufs);
 568         if (mp->m_logbsize > 0)
 569                 seq_printf(m, "," MNTOPT_LOGBSIZE "=%dk", mp->m_logbsize >> 10);
 570
 571         if (mp->m_logname)
 572                 seq_printf(m, "," MNTOPT_LOGDEV "=%s", mp->m_logname);
 573         if (mp->m_rtname)
 574                 seq_printf(m, "," MNTOPT_RTDEV "=%s", mp->m_rtname);
 575
 576         if (mp->m_dalign > 0)
 577                 seq_printf(m, "," MNTOPT_SUNIT "=%d",
 578                                 (int)XFS_FSB_TO_BB(mp, mp->m_dalign));
 579         if (mp->m_swidth > 0)
 580                 seq_printf(m, "," MNTOPT_SWIDTH "=%d",
 581                                 (int)XFS_FSB_TO_BB(mp, mp->m_swidth));
 582
 583         if (mp->m_qflags & (XFS_UQUOTA_ACCT|XFS_UQUOTA_ENFD))
 584                 seq_puts(m, "," MNTOPT_USRQUOTA);
 585         else if (mp->m_qflags & XFS_UQUOTA_ACCT)
 586                 seq_puts(m, "," MNTOPT_UQUOTANOENF);
 587
 588         if (mp->m_qflags & (XFS_PQUOTA_ACCT|XFS_OQUOTA_ENFD))
 589                 seq_puts(m, "," MNTOPT_PRJQUOTA);
 590         else if (mp->m_qflags & XFS_PQUOTA_ACCT)
 591                 seq_puts(m, "," MNTOPT_PQUOTANOENF);
 592
 593         if (mp->m_qflags & (XFS_GQUOTA_ACCT|XFS_OQUOTA_ENFD))
 594                 seq_puts(m, "," MNTOPT_GRPQUOTA);
 595         else if (mp->m_qflags & XFS_GQUOTA_ACCT)
 596                 seq_puts(m, "," MNTOPT_GQUOTANOENF);
 597
 598         if (!(mp->m_qflags & XFS_ALL_QUOTA_ACCT))
 599                 seq_puts(m, "," MNTOPT_NOQUOTA);
 600
 601         return 0;
 602 }
 603 __uint64_t
 604 xfs_max_file_offset(
 605         unsigned int            blockshift)
 606 {
 607         unsigned int            pagefactor = 1;
 608         unsigned int            bitshift = BITS_PER_LONG - 1;
 609
 610         /* Figure out maximum filesize, on Linux this can depend on
 611          * the filesystem blocksize (on 32 bit platforms).
 612          * __block_prepare_write does this in an [unsigned] long...
 613          *      page->index << (PAGE_CACHE_SHIFT - bbits)
 614          * So, for page sized blocks (4K on 32 bit platforms),
 615          * this wraps at around 8Tb (hence MAX_LFS_FILESIZE which is
 616          *      (((u64)PAGE_CACHE_SIZE << (BITS_PER_LONG-1))-1)
 617          * but for smaller blocksizes it is less (bbits = log2 bsize).
 618          * Note1: get_block_t takes a long (implicit cast from above)
 619          * Note2: The Large Block Device (LBD and HAVE_SECTOR_T) patch
 620          * can optionally convert the [unsigned] long from above into
 621          * an [unsigned] long long.
 622          */
 623
 624 #if BITS_PER_LONG == 32
 625 # if defined(CONFIG_LBD)
 626         ASSERT(sizeof(sector_t) == 8);
 627         pagefactor = PAGE_CACHE_SIZE;
 628         bitshift = BITS_PER_LONG;
 629 # else
 630         pagefactor = PAGE_CACHE_SIZE >> (PAGE_CACHE_SHIFT - blockshift);
 631 # endif
 632 #endif
 633
 634         return (((__uint64_t)pagefactor) << bitshift) - 1;
 635 }
 636
 637 int
 638 xfs_blkdev_get(
 639         xfs_mount_t             *mp,
 640         const char              *name,
 641         struct block_device     **bdevp)
 642 {
 643         int                     error = 0;
 644
 645         *bdevp = open_bdev_exclusive(name, FMODE_READ|FMODE_WRITE, mp);
 646         if (IS_ERR(*bdevp)) {
 647                 error = PTR_ERR(*bdevp);
 648                 printk("XFS: Invalid device [%s], error=%d\n", name, error);
 649         }
 650
 651         return -error;
 652 }
 653
 654 void
 655 xfs_blkdev_put(
 656         struct block_device     *bdev)
 657 {
 658         if (bdev)
 659                 close_bdev_exclusive(bdev, FMODE_READ|FMODE_WRITE);
 660 }
 661
 662 /*
 663  * Try to write out the superblock using barriers.
 664  */
 665 STATIC int
 666 xfs_barrier_test(
 667         xfs_mount_t     *mp)
 668 {
 669         xfs_buf_t       *sbp = xfs_getsb(mp, 0);
 670         int             error;
 671
 672         XFS_BUF_UNDONE(sbp);
 673         XFS_BUF_UNREAD(sbp);
 674         XFS_BUF_UNDELAYWRITE(sbp);
 675         XFS_BUF_WRITE(sbp);
 676         XFS_BUF_UNASYNC(sbp);
 677         XFS_BUF_ORDERED(sbp);
 678
 679         xfsbdstrat(mp, sbp);
 680         error = xfs_iowait(sbp);
 681
 682         /*
 683          * Clear all the flags we set and possible error state in the
 684          * buffer.  We only did the write to try out whether barriers
 685          * worked and shouldn't leave any traces in the superblock
 686          * buffer.
 687          */
 688         XFS_BUF_DONE(sbp);
 689         XFS_BUF_ERROR(sbp, 0);
 690         XFS_BUF_UNORDERED(sbp);
 691
 692         xfs_buf_relse(sbp);
 693         return error;
 694 }
 695
 696 void
 697 xfs_mountfs_check_barriers(xfs_mount_t *mp)
 698 {
 699         int error;
 700
 701         if (mp->m_logdev_targp != mp->m_ddev_targp) {
 702                 xfs_fs_cmn_err(CE_NOTE, mp,
 703                   "Disabling barriers, not supported with external log device");
 704                 mp->m_flags &= ~XFS_MOUNT_BARRIER;
 705                 return;
 706         }
 707
 708         if (xfs_readonly_buftarg(mp->m_ddev_targp)) {
 709                 xfs_fs_cmn_err(CE_NOTE, mp,
 710                   "Disabling barriers, underlying device is readonly");
 711                 mp->m_flags &= ~XFS_MOUNT_BARRIER;
 712                 return;
 713         }
 714
 715         error = xfs_barrier_test(mp);
 716         if (error) {
 717                 xfs_fs_cmn_err(CE_NOTE, mp,
 718                   "Disabling barriers, trial barrier write failed");
 719                 mp->m_flags &= ~XFS_MOUNT_BARRIER;
 720                 return;
 721         }
 722 }
 723
 724 void
 725 xfs_blkdev_issue_flush(
 726         xfs_buftarg_t           *buftarg)
 727 {
 728         blkdev_issue_flush(buftarg->bt_bdev, NULL);
 729 }
 730
 731 STATIC void
 732 xfs_close_devices(
 733         struct xfs_mount        *mp)
 734 {
 735         if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
 736                 struct block_device *logdev = mp->m_logdev_targp->bt_bdev;
 737                 xfs_free_buftarg(mp->m_logdev_targp);
 738                 xfs_blkdev_put(logdev);
 739         }
 740         if (mp->m_rtdev_targp) {
 741                 struct block_device *rtdev = mp->m_rtdev_targp->bt_bdev;
 742                 xfs_free_buftarg(mp->m_rtdev_targp);
 743                 xfs_blkdev_put(rtdev);
 744         }
 745         xfs_free_buftarg(mp->m_ddev_targp);
 746 }
 747
 748 /*
 749  * The file system configurations are:
 750  *      (1) device (partition) with data and internal log
 751  *      (2) logical volume with data and log subvolumes.
 752  *      (3) logical volume with data, log, and realtime subvolumes.
 753  *
 754  * We only have to handle opening the log and realtime volumes here if
 755  * they are present.  The data subvolume has already been opened by
 756  * get_sb_bdev() and is stored in sb->s_bdev.
 757  */
 758 STATIC int
 759 xfs_open_devices(
 760         struct xfs_mount        *mp)
 761 {
 762         struct block_device     *ddev = mp->m_super->s_bdev;
 763         struct block_device     *logdev = NULL, *rtdev = NULL;
 764         int                     error;
 765
 766         /*
 767          * Open real time and log devices - order is important.
 768          */
 769         if (mp->m_logname) {
 770                 error = xfs_blkdev_get(mp, mp->m_logname, &logdev);
 771                 if (error)
 772                         goto out;
 773         }
 774
 775         if (mp->m_rtname) {
 776                 error = xfs_blkdev_get(mp, mp->m_rtname, &rtdev);
 777                 if (error)
 778                         goto out_close_logdev;
 779
 780                 if (rtdev == ddev || rtdev == logdev) {
 781                         cmn_err(CE_WARN,
 782         "XFS: Cannot mount filesystem with identical rtdev and ddev/logdev.");
 783                         error = EINVAL;
 784                         goto out_close_rtdev;
 785                 }
 786         }
 787
 788         /*
 789          * Setup xfs_mount buffer target pointers
 790          */
 791         error = ENOMEM;
 792         mp->m_ddev_targp = xfs_alloc_buftarg(ddev, 0);
 793         if (!mp->m_ddev_targp)
 794                 goto out_close_rtdev;
 795
 796         if (rtdev) {
 797                 mp->m_rtdev_targp = xfs_alloc_buftarg(rtdev, 1);
 798                 if (!mp->m_rtdev_targp)
 799                         goto out_free_ddev_targ;
 800         }
 801
 802         if (logdev && logdev != ddev) {
 803                 mp->m_logdev_targp = xfs_alloc_buftarg(logdev, 1);
 804                 if (!mp->m_logdev_targp)
 805                         goto out_free_rtdev_targ;
 806         } else {
 807                 mp->m_logdev_targp = mp->m_ddev_targp;
 808         }
 809
 810         return 0;
 811
 812  out_free_rtdev_targ:
 813         if (mp->m_rtdev_targp)
 814                 xfs_free_buftarg(mp->m_rtdev_targp);
 815  out_free_ddev_targ:
 816         xfs_free_buftarg(mp->m_ddev_targp);
 817  out_close_rtdev:
 818         if (rtdev)
 819                 xfs_blkdev_put(rtdev);
 820  out_close_logdev:
 821         if (logdev && logdev != ddev)
 822                 xfs_blkdev_put(logdev);
 823  out:
 824         return error;
 825 }
 826
 827 /*
 828  * Setup xfs_mount buffer target pointers based on superblock
 829  */
 830 STATIC int
 831 xfs_setup_devices(
 832         struct xfs_mount        *mp)
 833 {
 834         int                     error;
 835
 836         error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_blocksize,
 837                                     mp->m_sb.sb_sectsize);
 838         if (error)
 839                 return error;
 840
 841         if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
 842                 unsigned int    log_sector_size = BBSIZE;
 843
 844                 if (xfs_sb_version_hassector(&mp->m_sb))
 845                         log_sector_size = mp->m_sb.sb_logsectsize;
 846                 error = xfs_setsize_buftarg(mp->m_logdev_targp,
 847                                             mp->m_sb.sb_blocksize,
 848                                             log_sector_size);
 849                 if (error)
 850                         return error;
 851         }
 852         if (mp->m_rtdev_targp) {
 853                 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
 854                                             mp->m_sb.sb_blocksize,
 855                                             mp->m_sb.sb_sectsize);
 856                 if (error)
 857                         return error;
 858         }
 859
 860         return 0;
 861 }
 862
 863 /*
 864  * XFS AIL push thread support
 865  */
 866 void
 867 xfsaild_wakeup(
 868         struct xfs_ail          *ailp,
 869         xfs_lsn_t               threshold_lsn)
 870 {
 871         ailp->xa_target = threshold_lsn;
 872         wake_up_process(ailp->xa_task);
 873 }
 874
 875 int
 876 xfsaild(
 877         void    *data)
 878 {
 879         struct xfs_ail  *ailp = data;
 880         xfs_lsn_t       last_pushed_lsn = 0;
 881         long            tout = 0;
 882
 883         while (!kthread_should_stop()) {
 884                 if (tout)
 885                         schedule_timeout_interruptible(msecs_to_jiffies(tout));
 886                 tout = 1000;
 887
 888                 /* swsusp */
 889                 try_to_freeze();
 890
 891                 ASSERT(ailp->xa_mount->m_log);
 892                 if (XFS_FORCED_SHUTDOWN(ailp->xa_mount))
 893                         continue;
 894
 895                 tout = xfsaild_push(ailp, &last_pushed_lsn);
 896         }
 897
 898         return 0;
 899 }       /* xfsaild */
 900
 901 int
 902 xfsaild_start(
 903         struct xfs_ail  *ailp)
 904 {
 905         ailp->xa_target = 0;
 906         ailp->xa_task = kthread_run(xfsaild, ailp, "xfsaild");
 907         if (IS_ERR(ailp->xa_task))
 908                 return -PTR_ERR(ailp->xa_task);
 909         return 0;
 910 }
 911
 912 void
 913 xfsaild_stop(
 914         struct xfs_ail  *ailp)
 915 {
 916         kthread_stop(ailp->xa_task);
 917 }
 918
 919
 920 /* Catch misguided souls that try to use this interface on XFS */
 921 STATIC struct inode *
 922 xfs_fs_alloc_inode(
 923         struct super_block      *sb)
 924 {
 925         BUG();
 926         return NULL;
 927 }
 928
 929 /*
 930  * Now that the generic code is guaranteed not to be accessing
 931  * the linux inode, we can reclaim the inode.
 932  */
 933 STATIC void
 934 xfs_fs_destroy_inode(
 935         struct inode    *inode)
 936 {
 937         xfs_inode_t             *ip = XFS_I(inode);
 938
 939         XFS_STATS_INC(vn_reclaim);
 940         if (xfs_reclaim(ip))
 941                 panic("%s: cannot reclaim 0x%p\n", __func__, inode);
 942 }
 943
 944 /*
 945  * Slab object creation initialisation for the XFS inode.
 946  * This covers only the idempotent fields in the XFS inode;
 947  * all other fields need to be initialised on allocation
 948  * from the slab. This avoids the need to repeatedly intialise
 949  * fields in the xfs inode that left in the initialise state
 950  * when freeing the inode.
 951  */
 952 STATIC void
 953 xfs_fs_inode_init_once(
 954         void                    *inode)
 955 {
 956         struct xfs_inode        *ip = inode;
 957
 958         memset(ip, 0, sizeof(struct xfs_inode));
 959
 960         /* vfs inode */
 961         inode_init_once(VFS_I(ip));
 962
 963         /* xfs inode */
 964         atomic_set(&ip->i_iocount, 0);
 965         atomic_set(&ip->i_pincount, 0);
 966         spin_lock_init(&ip->i_flags_lock);
 967         init_waitqueue_head(&ip->i_ipin_wait);
 968         /*
 969          * Because we want to use a counting completion, complete
 970          * the flush completion once to allow a single access to
 971          * the flush completion without blocking.
 972          */
 973         init_completion(&ip->i_flush);
 974         complete(&ip->i_flush);
 975
 976         mrlock_init(&ip->i_lock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
 977                      "xfsino", ip->i_ino);
 978         mrlock_init(&ip->i_iolock, MRLOCK_BARRIER, "xfsio", ip->i_ino);
 979 }
 980
 981 /*
 982  * Attempt to flush the inode, this will actually fail
 983  * if the inode is pinned, but we dirty the inode again
 984  * at the point when it is unpinned after a log write,
 985  * since this is when the inode itself becomes flushable.
 986  */
 987 STATIC int
 988 xfs_fs_write_inode(
 989         struct inode            *inode,
 990         int                     sync)
 991 {
 992         int                     error = 0;
 993         int                     flags = 0;
 994
 995         xfs_itrace_entry(XFS_I(inode));
 996         if (sync) {
 997                 filemap_fdatawait(inode->i_mapping);
 998                 flags |= FLUSH_SYNC;
 999         }
1000         error = xfs_inode_flush(XFS_I(inode), flags);
1001         /*
1002          * if we failed to write out the inode then mark
1003          * it dirty again so we'll try again later.
1004          */
1005         if (error)
1006                 xfs_mark_inode_dirty_sync(XFS_I(inode));
1007
1008         return -error;
1009 }
1010
1011 STATIC void
1012 xfs_fs_clear_inode(
1013         struct inode            *inode)
1014 {
1015         xfs_inode_t             *ip = XFS_I(inode);
1016
1017         xfs_itrace_entry(ip);
1018         XFS_STATS_INC(vn_rele);
1019         XFS_STATS_INC(vn_remove);
1020         XFS_STATS_DEC(vn_active);
1021
1022         xfs_inactive(ip);
1023         xfs_iflags_clear(ip, XFS_IMODIFIED);
1024 }
1025
1026 STATIC void
1027 xfs_free_fsname(
1028         struct xfs_mount        *mp)
1029 {
1030         kfree(mp->m_fsname);
1031         kfree(mp->m_rtname);
1032         kfree(mp->m_logname);
1033 }
1034
1035 STATIC void
1036 xfs_fs_put_super(
1037         struct super_block      *sb)
1038 {
1039         struct xfs_mount        *mp = XFS_M(sb);
1040         struct xfs_inode        *rip = mp->m_rootip;
1041         int                     unmount_event_flags = 0;
1042         int                     error;
1043
1044         xfs_syncd_stop(mp);
1045         xfs_sync_inodes(mp, SYNC_ATTR|SYNC_DELWRI);
1046
1047 #ifdef HAVE_DMAPI
1048         if (mp->m_flags & XFS_MOUNT_DMAPI) {
1049                 unmount_event_flags =
1050                         (mp->m_dmevmask & (1 << DM_EVENT_UNMOUNT)) ?
1051                                 0 : DM_FLAGS_UNWANTED;
1052                 /*
1053                  * Ignore error from dmapi here, first unmount is not allowed
1054                  * to fail anyway, and second we wouldn't want to fail a
1055                  * unmount because of dmapi.
1056                  */
1057                 XFS_SEND_PREUNMOUNT(mp, rip, DM_RIGHT_NULL, rip, DM_RIGHT_NULL,
1058                                 NULL, NULL, 0, 0, unmount_event_flags);
1059         }
1060 #endif
1061
1062         /*
1063          * Blow away any referenced inode in the filestreams cache.
1064          * This can and will cause log traffic as inodes go inactive
1065          * here.
1066          */
1067         xfs_filestream_unmount(mp);
1068
1069         XFS_bflush(mp->m_ddev_targp);
1070         error = xfs_unmount_flush(mp, 0);
1071         WARN_ON(error);
1072
1073         if (mp->m_flags & XFS_MOUNT_DMAPI) {
1074                 XFS_SEND_UNMOUNT(mp, rip, DM_RIGHT_NULL, 0, 0,
1075                                 unmount_event_flags);
1076         }
1077
1078         xfs_unmountfs(mp);
1079         xfs_freesb(mp);
1080         xfs_icsb_destroy_counters(mp);
1081         xfs_close_devices(mp);
1082         xfs_qmops_put(mp);
1083         xfs_dmops_put(mp);
1084         xfs_free_fsname(mp);
1085         kfree(mp);
1086 }
1087
1088 STATIC void
1089 xfs_fs_write_super(
1090         struct super_block      *sb)
1091 {
1092         if (!(sb->s_flags & MS_RDONLY))
1093                 xfs_sync_fsdata(XFS_M(sb), 0);
1094         sb->s_dirt = 0;
1095 }
1096
1097 STATIC int
1098 xfs_fs_sync_super(
1099         struct super_block      *sb,
1100         int                     wait)
1101 {
1102         struct xfs_mount        *mp = XFS_M(sb);
1103         int                     error;
1104
1105         /*
1106          * Treat a sync operation like a freeze.  This is to work
1107          * around a race in sync_inodes() which works in two phases
1108          * - an asynchronous flush, which can write out an inode
1109          * without waiting for file size updates to complete, and a
1110          * synchronous flush, which wont do anything because the
1111          * async flush removed the inode's dirty flag.  Also
1112          * sync_inodes() will not see any files that just have
1113          * outstanding transactions to be flushed because we don't
1114          * dirty the Linux inode until after the transaction I/O
1115          * completes.
1116          */
1117         if (wait || unlikely(sb->s_frozen == SB_FREEZE_WRITE))
1118                 error = xfs_quiesce_data(mp);
1119         else
1120                 error = xfs_sync_fsdata(mp, 0);
1121         sb->s_dirt = 0;
1122
1123         if (unlikely(laptop_mode)) {
1124                 int     prev_sync_seq = mp->m_sync_seq;
1125
1126                 /*
1127                  * The disk must be active because we're syncing.
1128                  * We schedule xfssyncd now (now that the disk is
1129                  * active) instead of later (when it might not be).
1130                  */
1131                 wake_up_process(mp->m_sync_task);
1132                 /*
1133                  * We have to wait for the sync iteration to complete.
1134                  * If we don't, the disk activity caused by the sync
1135                  * will come after the sync is completed, and that
1136                  * triggers another sync from laptop mode.
1137                  */
1138                 wait_event(mp->m_wait_single_sync_task,
1139                                 mp->m_sync_seq != prev_sync_seq);
1140         }
1141
1142         return -error;
1143 }
1144
1145 STATIC int
1146 xfs_fs_statfs(
1147         struct dentry           *dentry,
1148         struct kstatfs          *statp)
1149 {
1150         struct xfs_mount        *mp = XFS_M(dentry->d_sb);
1151         xfs_sb_t                *sbp = &mp->m_sb;
1152         __uint64_t              fakeinos, id;
1153         xfs_extlen_t            lsize;
1154
1155         statp->f_type = XFS_SB_MAGIC;
1156         statp->f_namelen = MAXNAMELEN - 1;
1157
1158         id = huge_encode_dev(mp->m_ddev_targp->bt_dev);
1159         statp->f_fsid.val[0] = (u32)id;
1160         statp->f_fsid.val[1] = (u32)(id >> 32);
1161
1162         xfs_icsb_sync_counters(mp, XFS_ICSB_LAZY_COUNT);
1163
1164         spin_lock(&mp->m_sb_lock);
1165         statp->f_bsize = sbp->sb_blocksize;
1166         lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
1167         statp->f_blocks = sbp->sb_dblocks - lsize;
1168         statp->f_bfree = statp->f_bavail =
1169                                 sbp->sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
1170         fakeinos = statp->f_bfree << sbp->sb_inopblog;
1171 #if XFS_BIG_INUMS
1172         fakeinos += mp->m_inoadd;
1173 #endif
1174         statp->f_files =
1175             MIN(sbp->sb_icount + fakeinos, (__uint64_t)XFS_MAXINUMBER);
1176         if (mp->m_maxicount)
1177 #if XFS_BIG_INUMS
1178                 if (!mp->m_inoadd)
1179 #endif
1180                         statp->f_files = min_t(typeof(statp->f_files),
1181                                                 statp->f_files,
1182                                                 mp->m_maxicount);
1183         statp->f_ffree = statp->f_files - (sbp->sb_icount - sbp->sb_ifree);
1184         spin_unlock(&mp->m_sb_lock);
1185
1186         XFS_QM_DQSTATVFS(XFS_I(dentry->d_inode), statp);
1187         return 0;
1188 }
1189
1190 STATIC int
1191 xfs_fs_remount(
1192         struct super_block      *sb,
1193         int                     *flags,
1194         char                    *options)
1195 {
1196         struct xfs_mount        *mp = XFS_M(sb);
1197         substring_t             args[MAX_OPT_ARGS];
1198         char                    *p;
1199
1200         while ((p = strsep(&options, ",")) != NULL) {
1201                 int token;
1202
1203                 if (!*p)
1204                         continue;
1205
1206                 token = match_token(p, tokens, args);
1207                 switch (token) {
1208                 case Opt_barrier:
1209                         mp->m_flags |= XFS_MOUNT_BARRIER;
1210
1211                         /*
1212                          * Test if barriers are actually working if we can,
1213                          * else delay this check until the filesystem is
1214                          * marked writeable.
1215                          */
1216                         if (!(mp->m_flags & XFS_MOUNT_RDONLY))
1217                                 xfs_mountfs_check_barriers(mp);
1218                         break;
1219                 case Opt_nobarrier:
1220                         mp->m_flags &= ~XFS_MOUNT_BARRIER;
1221                         break;
1222                 default:
1223                         /*
1224                          * Logically we would return an error here to prevent
1225                          * users from believing they might have changed
1226                          * mount options using remount which can't be changed.
1227                          *
1228                          * But unfortunately mount(8) adds all options from
1229                          * mtab and fstab to the mount arguments in some cases
1230                          * so we can't blindly reject options, but have to
1231                          * check for each specified option if it actually
1232                          * differs from the currently set option and only
1233                          * reject it if that's the case.
1234                          *
1235                          * Until that is implemented we return success for
1236                          * every remount request, and silently ignore all
1237                          * options that we can't actually change.
1238                          */
1239 #if 0
1240                         printk(KERN_INFO
1241         "XFS: mount option \"%s\" not supported for remount\n", p);
1242                         return -EINVAL;
1243 #else
1244                         break;
1245 #endif
1246                 }
1247         }
1248
1249         /* rw/ro -> rw */
1250         if ((mp->m_flags & XFS_MOUNT_RDONLY) && !(*flags & MS_RDONLY)) {
1251                 mp->m_flags &= ~XFS_MOUNT_RDONLY;
1252                 if (mp->m_flags & XFS_MOUNT_BARRIER)
1253                         xfs_mountfs_check_barriers(mp);
1254         }
1255
1256         /* rw -> ro */
1257         if (!(mp->m_flags & XFS_MOUNT_RDONLY) && (*flags & MS_RDONLY)) {
1258                 xfs_quiesce_data(mp);
1259                 xfs_quiesce_attr(mp);
1260                 mp->m_flags |= XFS_MOUNT_RDONLY;
1261         }
1262
1263         return 0;
1264 }
1265
1266 /*
1267  * Second stage of a freeze. The data is already frozen so we only
1268  * need to take care of the metadata. Once that's done write a dummy
1269  * record to dirty the log in case of a crash while frozen.
1270  */
1271 STATIC void
1272 xfs_fs_lockfs(
1273         struct super_block      *sb)
1274 {
1275         struct xfs_mount        *mp = XFS_M(sb);
1276
1277         xfs_quiesce_attr(mp);
1278         xfs_fs_log_dummy(mp);
1279 }
1280
1281 STATIC int
1282 xfs_fs_show_options(
1283         struct seq_file         *m,
1284         struct vfsmount         *mnt)
1285 {
1286         return -xfs_showargs(XFS_M(mnt->mnt_sb), m);
1287 }
1288
1289 STATIC int
1290 xfs_fs_quotasync(
1291         struct super_block      *sb,
1292         int                     type)
1293 {
1294         return -XFS_QM_QUOTACTL(XFS_M(sb), Q_XQUOTASYNC, 0, NULL);
1295 }
1296
1297 STATIC int
1298 xfs_fs_getxstate(
1299         struct super_block      *sb,
1300         struct fs_quota_stat    *fqs)
1301 {
1302         return -XFS_QM_QUOTACTL(XFS_M(sb), Q_XGETQSTAT, 0, (caddr_t)fqs);
1303 }
1304
1305 STATIC int
1306 xfs_fs_setxstate(
1307         struct super_block      *sb,
1308         unsigned int            flags,
1309         int                     op)
1310 {
1311         return -XFS_QM_QUOTACTL(XFS_M(sb), op, 0, (caddr_t)&flags);
1312 }
1313
1314 STATIC int
1315 xfs_fs_getxquota(
1316         struct super_block      *sb,
1317         int                     type,
1318         qid_t                   id,
1319         struct fs_disk_quota    *fdq)
1320 {
1321         return -XFS_QM_QUOTACTL(XFS_M(sb),
1322                                  (type == USRQUOTA) ? Q_XGETQUOTA :
1323                                   ((type == GRPQUOTA) ? Q_XGETGQUOTA :
1324                                    Q_XGETPQUOTA), id, (caddr_t)fdq);
1325 }
1326
1327 STATIC int
1328 xfs_fs_setxquota(
1329         struct super_block      *sb,
1330         int                     type,
1331         qid_t                   id,
1332         struct fs_disk_quota    *fdq)
1333 {
1334         return -XFS_QM_QUOTACTL(XFS_M(sb),
1335                                  (type == USRQUOTA) ? Q_XSETQLIM :
1336                                   ((type == GRPQUOTA) ? Q_XSETGQLIM :
1337                                    Q_XSETPQLIM), id, (caddr_t)fdq);
1338 }
1339
1340 /*
1341  * This function fills in xfs_mount_t fields based on mount args.
1342  * Note: the superblock _has_ now been read in.
1343  */
1344 STATIC int
1345 xfs_finish_flags(
1346         struct xfs_mount        *mp)
1347 {
1348         int                     ronly = (mp->m_flags & XFS_MOUNT_RDONLY);
1349
1350         /* Fail a mount where the logbuf is smaller then the log stripe */
1351         if (xfs_sb_version_haslogv2(&mp->m_sb)) {
1352                 if (mp->m_logbsize <= 0 &&
1353                     mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE) {
1354                         mp->m_logbsize = mp->m_sb.sb_logsunit;
1355                 } else if (mp->m_logbsize > 0 &&
1356                            mp->m_logbsize < mp->m_sb.sb_logsunit) {
1357                         cmn_err(CE_WARN,
1358         "XFS: logbuf size must be greater than or equal to log stripe size");
1359                         return XFS_ERROR(EINVAL);
1360                 }
1361         } else {
1362                 /* Fail a mount if the logbuf is larger than 32K */
1363                 if (mp->m_logbsize > XLOG_BIG_RECORD_BSIZE) {
1364                         cmn_err(CE_WARN,
1365         "XFS: logbuf size for version 1 logs must be 16K or 32K");
1366                         return XFS_ERROR(EINVAL);
1367                 }
1368         }
1369
1370         /*
1371          * mkfs'ed attr2 will turn on attr2 mount unless explicitly
1372          * told by noattr2 to turn it off
1373          */
1374         if (xfs_sb_version_hasattr2(&mp->m_sb) &&
1375             !(mp->m_flags & XFS_MOUNT_NOATTR2))
1376                 mp->m_flags |= XFS_MOUNT_ATTR2;
1377
1378         /*
1379          * prohibit r/w mounts of read-only filesystems
1380          */
1381         if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !ronly) {
1382                 cmn_err(CE_WARN,
1383         "XFS: cannot mount a read-only filesystem as read-write");
1384                 return XFS_ERROR(EROFS);
1385         }
1386
1387 #if 0 /* shared mounts were never supported on Linux */
1388         /*
1389          * check for shared mount.
1390          */
1391         if (ap->flags & XFSMNT_SHARED) {
1392                 if (!xfs_sb_version_hasshared(&mp->m_sb))
1393                         return XFS_ERROR(EINVAL);
1394
1395                 /*
1396                  * For IRIX 6.5, shared mounts must have the shared
1397                  * version bit set, have the persistent readonly
1398                  * field set, must be version 0 and can only be mounted
1399                  * read-only.
1400                  */
1401                 if (!ronly || !(mp->m_sb.sb_flags & XFS_SBF_READONLY) ||
1402                      (mp->m_sb.sb_shared_vn != 0))
1403                         return XFS_ERROR(EINVAL);
1404
1405                 mp->m_flags |= XFS_MOUNT_SHARED;
1406
1407                 /*
1408                  * Shared XFS V0 can't deal with DMI.  Return EINVAL.
1409                  */
1410                 if (mp->m_sb.sb_shared_vn == 0 &&
1411                     (mp->m_flags & XFS_MOUNT_DMAPI))
1412                         return XFS_ERROR(EINVAL);
1413         }
1414 #endif
1415
1416         return 0;
1417 }
1418
1419 STATIC int
1420 xfs_fs_fill_super(
1421         struct super_block      *sb,
1422         void                    *data,
1423         int                     silent)
1424 {
1425         struct inode            *root;
1426         struct xfs_mount        *mp = NULL;
1427         int                     flags = 0, error = ENOMEM;
1428         char                    *mtpt = NULL;
1429
1430         mp = kzalloc(sizeof(struct xfs_mount), GFP_KERNEL);
1431         if (!mp)
1432                 goto out;
1433
1434         spin_lock_init(&mp->m_sb_lock);
1435         mutex_init(&mp->m_growlock);
1436         atomic_set(&mp->m_active_trans, 0);
1437         INIT_LIST_HEAD(&mp->m_sync_list);
1438         spin_lock_init(&mp->m_sync_lock);
1439         init_waitqueue_head(&mp->m_wait_single_sync_task);
1440
1441         mp->m_super = sb;
1442         sb->s_fs_info = mp;
1443
1444         error = xfs_parseargs(mp, (char *)data, &mtpt);
1445         if (error)
1446                 goto out_free_fsname;
1447
1448         sb_min_blocksize(sb, BBSIZE);
1449         sb->s_xattr = xfs_xattr_handlers;
1450         sb->s_export_op = &xfs_export_operations;
1451         sb->s_qcop = &xfs_quotactl_operations;
1452         sb->s_op = &xfs_super_operations;
1453
1454         error = xfs_dmops_get(mp);
1455         if (error)
1456                 goto out_free_fsname;
1457         error = xfs_qmops_get(mp);
1458         if (error)
1459                 goto out_put_dmops;
1460
1461         if (silent)
1462                 flags |= XFS_MFSI_QUIET;
1463
1464         error = xfs_open_devices(mp);
1465         if (error)
1466                 goto out_put_qmops;
1467
1468         if (xfs_icsb_init_counters(mp))
1469                 mp->m_flags |= XFS_MOUNT_NO_PERCPU_SB;
1470
1471         error = xfs_readsb(mp, flags);
1472         if (error)
1473                 goto out_destroy_counters;
1474
1475         error = xfs_finish_flags(mp);
1476         if (error)
1477                 goto out_free_sb;
1478
1479         error = xfs_setup_devices(mp);
1480         if (error)
1481                 goto out_free_sb;
1482
1483         if (mp->m_flags & XFS_MOUNT_BARRIER)
1484                 xfs_mountfs_check_barriers(mp);
1485
1486         error = xfs_filestream_mount(mp);
1487         if (error)
1488                 goto out_free_sb;
1489
1490         error = xfs_mountfs(mp);
1491         if (error)
1492                 goto out_filestream_unmount;
1493
1494         XFS_SEND_MOUNT(mp, DM_RIGHT_NULL, mtpt, mp->m_fsname);
1495
1496         sb->s_dirt = 1;
1497         sb->s_magic = XFS_SB_MAGIC;
1498         sb->s_blocksize = mp->m_sb.sb_blocksize;
1499         sb->s_blocksize_bits = ffs(sb->s_blocksize) - 1;
1500         sb->s_maxbytes = xfs_max_file_offset(sb->s_blocksize_bits);
1501         sb->s_time_gran = 1;
1502         set_posix_acl_flag(sb);
1503
1504         root = igrab(VFS_I(mp->m_rootip));
1505         if (!root) {
1506                 error = ENOENT;
1507                 goto fail_unmount;
1508         }
1509         if (is_bad_inode(root)) {
1510                 error = EINVAL;
1511                 goto fail_vnrele;
1512         }
1513         sb->s_root = d_alloc_root(root);
1514         if (!sb->s_root) {
1515                 error = ENOMEM;
1516                 goto fail_vnrele;
1517         }
1518
1519         error = xfs_syncd_init(mp);
1520         if (error)
1521                 goto fail_vnrele;
1522
1523         kfree(mtpt);
1524
1525         xfs_itrace_exit(XFS_I(sb->s_root->d_inode));
1526         return 0;
1527
1528  out_filestream_unmount:
1529         xfs_filestream_unmount(mp);
1530  out_free_sb:
1531         xfs_freesb(mp);
1532  out_destroy_counters:
1533         xfs_icsb_destroy_counters(mp);
1534         xfs_close_devices(mp);
1535  out_put_qmops:
1536         xfs_qmops_put(mp);
1537  out_put_dmops:
1538         xfs_dmops_put(mp);
1539  out_free_fsname:
1540         xfs_free_fsname(mp);
1541         kfree(mtpt);
1542         kfree(mp);
1543  out:
1544         return -error;
1545
1546  fail_vnrele:
1547         if (sb->s_root) {
1548                 dput(sb->s_root);
1549                 sb->s_root = NULL;
1550         } else {
1551                 iput(root);
1552         }
1553
1554  fail_unmount:
1555         /*
1556          * Blow away any referenced inode in the filestreams cache.
1557          * This can and will cause log traffic as inodes go inactive
1558          * here.
1559          */
1560         xfs_filestream_unmount(mp);
1561
1562         XFS_bflush(mp->m_ddev_targp);
1563         error = xfs_unmount_flush(mp, 0);
1564         WARN_ON(error);
1565
1566         xfs_unmountfs(mp);
1567         goto out_free_sb;
1568 }
1569
1570 STATIC int
1571 xfs_fs_get_sb(
1572         struct file_system_type *fs_type,
1573         int                     flags,
1574         const char              *dev_name,
1575         void                    *data,
1576         struct vfsmount         *mnt)
1577 {
1578         return get_sb_bdev(fs_type, flags, dev_name, data, xfs_fs_fill_super,
1579                            mnt);
1580 }
1581
1582 static struct super_operations xfs_super_operations = {
1583         .alloc_inode            = xfs_fs_alloc_inode,
1584         .destroy_inode          = xfs_fs_destroy_inode,
1585         .write_inode            = xfs_fs_write_inode,
1586         .clear_inode            = xfs_fs_clear_inode,
1587         .put_super              = xfs_fs_put_super,
1588         .write_super            = xfs_fs_write_super,
1589         .sync_fs                = xfs_fs_sync_super,
1590         .write_super_lockfs     = xfs_fs_lockfs,
1591         .statfs                 = xfs_fs_statfs,
1592         .remount_fs             = xfs_fs_remount,
1593         .show_options           = xfs_fs_show_options,
1594 };
1595
1596 static struct quotactl_ops xfs_quotactl_operations = {
1597         .quota_sync             = xfs_fs_quotasync,
1598         .get_xstate             = xfs_fs_getxstate,
1599         .set_xstate             = xfs_fs_setxstate,
1600         .get_xquota             = xfs_fs_getxquota,
1601         .set_xquota             = xfs_fs_setxquota,
1602 };
1603
1604 static struct file_system_type xfs_fs_type = {
1605         .owner                  = THIS_MODULE,
1606         .name                   = "xfs",
1607         .get_sb                 = xfs_fs_get_sb,
1608         .kill_sb                = kill_block_super,
1609         .fs_flags               = FS_REQUIRES_DEV,
1610 };
1611
1612 STATIC int __init
1613 xfs_alloc_trace_bufs(void)
1614 {
1615 #ifdef XFS_ALLOC_TRACE
1616         xfs_alloc_trace_buf = ktrace_alloc(XFS_ALLOC_TRACE_SIZE, KM_MAYFAIL);
1617         if (!xfs_alloc_trace_buf)
1618                 goto out;
1619 #endif
1620 #ifdef XFS_BMAP_TRACE
1621         xfs_bmap_trace_buf = ktrace_alloc(XFS_BMAP_TRACE_SIZE, KM_MAYFAIL);
1622         if (!xfs_bmap_trace_buf)
1623                 goto out_free_alloc_trace;
1624 #endif
1625 #ifdef XFS_BTREE_TRACE
1626         xfs_allocbt_trace_buf = ktrace_alloc(XFS_ALLOCBT_TRACE_SIZE,
1627                                              KM_MAYFAIL);
1628         if (!xfs_allocbt_trace_buf)
1629                 goto out_free_bmap_trace;
1630
1631         xfs_inobt_trace_buf = ktrace_alloc(XFS_INOBT_TRACE_SIZE, KM_MAYFAIL);
1632         if (!xfs_inobt_trace_buf)
1633                 goto out_free_allocbt_trace;
1634
1635         xfs_bmbt_trace_buf = ktrace_alloc(XFS_BMBT_TRACE_SIZE, KM_MAYFAIL);
1636         if (!xfs_bmbt_trace_buf)
1637                 goto out_free_inobt_trace;
1638 #endif
1639 #ifdef XFS_ATTR_TRACE
1640         xfs_attr_trace_buf = ktrace_alloc(XFS_ATTR_TRACE_SIZE, KM_MAYFAIL);
1641         if (!xfs_attr_trace_buf)
1642                 goto out_free_bmbt_trace;
1643 #endif
1644 #ifdef XFS_DIR2_TRACE
1645         xfs_dir2_trace_buf = ktrace_alloc(XFS_DIR2_GTRACE_SIZE, KM_MAYFAIL);
1646         if (!xfs_dir2_trace_buf)
1647                 goto out_free_attr_trace;
1648 #endif
1649
1650         return 0;
1651
1652 #ifdef XFS_DIR2_TRACE
1653  out_free_attr_trace:
1654 #endif
1655 #ifdef XFS_ATTR_TRACE
1656         ktrace_free(xfs_attr_trace_buf);
1657  out_free_bmbt_trace:
1658 #endif
1659 #ifdef XFS_BTREE_TRACE
1660         ktrace_free(xfs_bmbt_trace_buf);
1661  out_free_inobt_trace:
1662         ktrace_free(xfs_inobt_trace_buf);
1663  out_free_allocbt_trace:
1664         ktrace_free(xfs_allocbt_trace_buf);
1665  out_free_bmap_trace:
1666 #endif
1667 #ifdef XFS_BMAP_TRACE
1668         ktrace_free(xfs_bmap_trace_buf);
1669  out_free_alloc_trace:
1670 #endif
1671 #ifdef XFS_ALLOC_TRACE
1672         ktrace_free(xfs_alloc_trace_buf);
1673  out:
1674 #endif
1675         return -ENOMEM;
1676 }
1677
1678 STATIC void
1679 xfs_free_trace_bufs(void)
1680 {
1681 #ifdef XFS_DIR2_TRACE
1682         ktrace_free(xfs_dir2_trace_buf);
1683 #endif
1684 #ifdef XFS_ATTR_TRACE
1685         ktrace_free(xfs_attr_trace_buf);
1686 #endif
1687 #ifdef XFS_BTREE_TRACE
1688         ktrace_free(xfs_bmbt_trace_buf);
1689         ktrace_free(xfs_inobt_trace_buf);
1690         ktrace_free(xfs_allocbt_trace_buf);
1691 #endif
1692 #ifdef XFS_BMAP_TRACE
1693         ktrace_free(xfs_bmap_trace_buf);
1694 #endif
1695 #ifdef XFS_ALLOC_TRACE
1696         ktrace_free(xfs_alloc_trace_buf);
1697 #endif
1698 }
1699
1700 STATIC int __init
1701 xfs_init_zones(void)
1702 {
1703
1704         xfs_ioend_zone = kmem_zone_init(sizeof(xfs_ioend_t), "xfs_ioend");
1705         if (!xfs_ioend_zone)
1706                 goto out;
1707
1708         xfs_ioend_pool = mempool_create_slab_pool(4 * MAX_BUF_PER_PAGE,
1709                                                   xfs_ioend_zone);
1710         if (!xfs_ioend_pool)
1711                 goto out_destroy_ioend_zone;
1712
1713         xfs_log_ticket_zone = kmem_zone_init(sizeof(xlog_ticket_t),
1714                                                 "xfs_log_ticket");
1715         if (!xfs_log_ticket_zone)
1716                 goto out_destroy_ioend_pool;
1717
1718         xfs_bmap_free_item_zone = kmem_zone_init(sizeof(xfs_bmap_free_item_t),
1719                                                 "xfs_bmap_free_item");
1720         if (!xfs_bmap_free_item_zone)
1721                 goto out_destroy_log_ticket_zone;
1722
1723         xfs_btree_cur_zone = kmem_zone_init(sizeof(xfs_btree_cur_t),
1724                                                 "xfs_btree_cur");
1725         if (!xfs_btree_cur_zone)
1726                 goto out_destroy_bmap_free_item_zone;
1727
1728         xfs_da_state_zone = kmem_zone_init(sizeof(xfs_da_state_t),
1729                                                 "xfs_da_state");
1730         if (!xfs_da_state_zone)
1731                 goto out_destroy_btree_cur_zone;
1732
1733         xfs_dabuf_zone = kmem_zone_init(sizeof(xfs_dabuf_t), "xfs_dabuf");
1734         if (!xfs_dabuf_zone)
1735                 goto out_destroy_da_state_zone;
1736
1737         xfs_ifork_zone = kmem_zone_init(sizeof(xfs_ifork_t), "xfs_ifork");
1738         if (!xfs_ifork_zone)
1739                 goto out_destroy_dabuf_zone;
1740
1741         xfs_trans_zone = kmem_zone_init(sizeof(xfs_trans_t), "xfs_trans");
1742         if (!xfs_trans_zone)
1743                 goto out_destroy_ifork_zone;
1744
1745         /*
1746          * The size of the zone allocated buf log item is the maximum
1747          * size possible under XFS.  This wastes a little bit of memory,
1748          * but it is much faster.
1749          */
1750         xfs_buf_item_zone = kmem_zone_init((sizeof(xfs_buf_log_item_t) +
1751                                 (((XFS_MAX_BLOCKSIZE / XFS_BLI_CHUNK) /
1752                                   NBWORD) * sizeof(int))), "xfs_buf_item");
1753         if (!xfs_buf_item_zone)
1754                 goto out_destroy_trans_zone;
1755
1756         xfs_efd_zone = kmem_zone_init((sizeof(xfs_efd_log_item_t) +
1757                         ((XFS_EFD_MAX_FAST_EXTENTS - 1) *
1758                                  sizeof(xfs_extent_t))), "xfs_efd_item");
1759         if (!xfs_efd_zone)
1760                 goto out_destroy_buf_item_zone;
1761
1762         xfs_efi_zone = kmem_zone_init((sizeof(xfs_efi_log_item_t) +
1763                         ((XFS_EFI_MAX_FAST_EXTENTS - 1) *
1764                                 sizeof(xfs_extent_t))), "xfs_efi_item");
1765         if (!xfs_efi_zone)
1766                 goto out_destroy_efd_zone;
1767
1768         xfs_inode_zone =
1769                 kmem_zone_init_flags(sizeof(xfs_inode_t), "xfs_inode",
1770                         KM_ZONE_HWALIGN | KM_ZONE_RECLAIM | KM_ZONE_SPREAD,
1771                         xfs_fs_inode_init_once);
1772         if (!xfs_inode_zone)
1773                 goto out_destroy_efi_zone;
1774
1775         xfs_ili_zone =
1776                 kmem_zone_init_flags(sizeof(xfs_inode_log_item_t), "xfs_ili",
1777                                         KM_ZONE_SPREAD, NULL);
1778         if (!xfs_ili_zone)
1779                 goto out_destroy_inode_zone;
1780
1781 #ifdef CONFIG_XFS_POSIX_ACL
1782         xfs_acl_zone = kmem_zone_init(sizeof(xfs_acl_t), "xfs_acl");
1783         if (!xfs_acl_zone)
1784                 goto out_destroy_ili_zone;
1785 #endif
1786
1787         return 0;
1788
1789 #ifdef CONFIG_XFS_POSIX_ACL
1790  out_destroy_ili_zone:
1791 #endif
1792         kmem_zone_destroy(xfs_ili_zone);
1793  out_destroy_inode_zone:
1794         kmem_zone_destroy(xfs_inode_zone);
1795  out_destroy_efi_zone:
1796         kmem_zone_destroy(xfs_efi_zone);
1797  out_destroy_efd_zone:
1798         kmem_zone_destroy(xfs_efd_zone);
1799  out_destroy_buf_item_zone:
1800         kmem_zone_destroy(xfs_buf_item_zone);
1801  out_destroy_trans_zone:
1802         kmem_zone_destroy(xfs_trans_zone);
1803  out_destroy_ifork_zone:
1804         kmem_zone_destroy(xfs_ifork_zone);
1805  out_destroy_dabuf_zone:
1806         kmem_zone_destroy(xfs_dabuf_zone);
1807  out_destroy_da_state_zone:
1808         kmem_zone_destroy(xfs_da_state_zone);
1809  out_destroy_btree_cur_zone:
1810         kmem_zone_destroy(xfs_btree_cur_zone);
1811  out_destroy_bmap_free_item_zone:
1812         kmem_zone_destroy(xfs_bmap_free_item_zone);
1813  out_destroy_log_ticket_zone:
1814         kmem_zone_destroy(xfs_log_ticket_zone);
1815  out_destroy_ioend_pool:
1816         mempool_destroy(xfs_ioend_pool);
1817  out_destroy_ioend_zone:
1818         kmem_zone_destroy(xfs_ioend_zone);
1819  out:
1820         return -ENOMEM;
1821 }
1822
1823 STATIC void
1824 xfs_destroy_zones(void)
1825 {
1826 #ifdef CONFIG_XFS_POSIX_ACL
1827         kmem_zone_destroy(xfs_acl_zone);
1828 #endif
1829         kmem_zone_destroy(xfs_ili_zone);
1830         kmem_zone_destroy(xfs_inode_zone);
1831         kmem_zone_destroy(xfs_efi_zone);
1832         kmem_zone_destroy(xfs_efd_zone);
1833         kmem_zone_destroy(xfs_buf_item_zone);
1834         kmem_zone_destroy(xfs_trans_zone);
1835         kmem_zone_destroy(xfs_ifork_zone);
1836         kmem_zone_destroy(xfs_dabuf_zone);
1837         kmem_zone_destroy(xfs_da_state_zone);
1838         kmem_zone_destroy(xfs_btree_cur_zone);
1839         kmem_zone_destroy(xfs_bmap_free_item_zone);
1840         kmem_zone_destroy(xfs_log_ticket_zone);
1841         mempool_destroy(xfs_ioend_pool);
1842         kmem_zone_destroy(xfs_ioend_zone);
1843
1844 }
1845
1846 STATIC int __init
1847 init_xfs_fs(void)
1848 {
1849         int                     error;
1850         static char             message[] __initdata = KERN_INFO \
1851                 XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled\n";
1852
1853         printk(message);
1854
1855         ktrace_init(64);
1856         vn_init();
1857         xfs_dir_startup();
1858
1859         error = xfs_init_zones();
1860         if (error)
1861                 goto out;
1862
1863         error = xfs_alloc_trace_bufs();
1864         if (error)
1865                 goto out_destroy_zones;
1866
1867         error = xfs_mru_cache_init();
1868         if (error)
1869                 goto out_free_trace_buffers;
1870
1871         error = xfs_filestream_init();
1872         if (error)
1873                 goto out_mru_cache_uninit;
1874
1875         error = xfs_buf_init();
1876         if (error)
1877                 goto out_filestream_uninit;
1878
1879         error = xfs_init_procfs();
1880         if (error)
1881                 goto out_buf_terminate;
1882
1883         error = xfs_sysctl_register();
1884         if (error)
1885                 goto out_cleanup_procfs;
1886
1887         vfs_initquota();
1888
1889         error = register_filesystem(&xfs_fs_type);
1890         if (error)
1891                 goto out_sysctl_unregister;
1892         return 0;
1893
1894  out_sysctl_unregister:
1895         xfs_sysctl_unregister();
1896  out_cleanup_procfs:
1897         xfs_cleanup_procfs();
1898  out_buf_terminate:
1899         xfs_buf_terminate();
1900  out_filestream_uninit:
1901         xfs_filestream_uninit();
1902  out_mru_cache_uninit:
1903         xfs_mru_cache_uninit();
1904  out_free_trace_buffers:
1905         xfs_free_trace_bufs();
1906  out_destroy_zones:
1907         xfs_destroy_zones();
1908  out:
1909         return error;
1910 }
1911
1912 STATIC void __exit
1913 exit_xfs_fs(void)
1914 {
1915         vfs_exitquota();
1916         unregister_filesystem(&xfs_fs_type);
1917         xfs_sysctl_unregister();
1918         xfs_cleanup_procfs();
1919         xfs_buf_terminate();
1920         xfs_filestream_uninit();
1921         xfs_mru_cache_uninit();
1922         xfs_free_trace_bufs();
1923         xfs_destroy_zones();
1924         ktrace_uninit();
1925 }
1926
1927 module_init(init_xfs_fs);
1928 module_exit(exit_xfs_fs);
1929
1930 MODULE_AUTHOR("Silicon Graphics, Inc.");
1931 MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled");
1932 MODULE_LICENSE("GPL");