fs/xfs/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
  19 #include "xfs.h"
  20 #include "xfs_shared.h"
  21 #include "xfs_format.h"
  22 #include "xfs_log_format.h"
  23 #include "xfs_trans_resv.h"
  24 #include "xfs_sb.h"
  25 #include "xfs_mount.h"
  26 #include "xfs_da_format.h"
  27 #include "xfs_inode.h"
  28 #include "xfs_btree.h"
  29 #include "xfs_bmap.h"
  30 #include "xfs_alloc.h"
  31 #include "xfs_error.h"
  32 #include "xfs_fsops.h"
  33 #include "xfs_trans.h"
  34 #include "xfs_buf_item.h"
  35 #include "xfs_log.h"
  36 #include "xfs_log_priv.h"
  37 #include "xfs_da_btree.h"
  38 #include "xfs_dir2.h"
  39 #include "xfs_extfree_item.h"
  40 #include "xfs_mru_cache.h"
  41 #include "xfs_inode_item.h"
  42 #include "xfs_icache.h"
  43 #include "xfs_trace.h"
  44 #include "xfs_icreate_item.h"
  45 #include "xfs_filestream.h"
  46 #include "xfs_quota.h"
  47 #include "xfs_sysfs.h"
  48
  49 #include <linux/namei.h>
  50 #include <linux/init.h>
  51 #include <linux/slab.h>
  52 #include <linux/mount.h>
  53 #include <linux/mempool.h>
  54 #include <linux/writeback.h>
  55 #include <linux/kthread.h>
  56 #include <linux/freezer.h>
  57 #include <linux/parser.h>
  58
  59 static const struct super_operations xfs_super_operations;
  60 static kmem_zone_t *xfs_ioend_zone;
  61 mempool_t *xfs_ioend_pool;
  62
  63 static struct kset *xfs_kset;           /* top-level xfs sysfs dir */
  64 #ifdef DEBUG
  65 static struct xfs_kobj xfs_dbg_kobj;    /* global debug sysfs attrs */
  66 #endif
  67
  68 /*
  69  * Table driven mount option parser.
  70  */
  71 enum {
  72         Opt_logbufs, Opt_logbsize, Opt_logdev, Opt_rtdev, Opt_biosize,
  73         Opt_wsync, Opt_noalign, Opt_swalloc, Opt_sunit, Opt_swidth, Opt_nouuid,
  74         Opt_mtpt, Opt_grpid, Opt_nogrpid, Opt_bsdgroups, Opt_sysvgroups,
  75         Opt_allocsize, Opt_norecovery, Opt_barrier, Opt_nobarrier,
  76         Opt_inode64, Opt_inode32, Opt_ikeep, Opt_noikeep,
  77         Opt_largeio, Opt_nolargeio, Opt_attr2, Opt_noattr2, Opt_filestreams,
  78         Opt_quota, Opt_noquota, Opt_usrquota, Opt_grpquota, Opt_prjquota,
  79         Opt_uquota, Opt_gquota, Opt_pquota,
  80         Opt_uqnoenforce, Opt_gqnoenforce, Opt_pqnoenforce, Opt_qnoenforce,
  81         Opt_discard, Opt_nodiscard, Opt_dax, Opt_err,
  82 };
  83
  84 static const match_table_t tokens = {
  85         {Opt_logbufs,   "logbufs=%u"},  /* number of XFS log buffers */
  86         {Opt_logbsize,  "logbsize=%s"}, /* size of XFS log buffers */
  87         {Opt_logdev,    "logdev=%s"},   /* log device */
  88         {Opt_rtdev,     "rtdev=%s"},    /* realtime I/O device */
  89         {Opt_biosize,   "biosize=%u"},  /* log2 of preferred buffered io size */
  90         {Opt_wsync,     "wsync"},       /* safe-mode nfs compatible mount */
  91         {Opt_noalign,   "noalign"},     /* turn off stripe alignment */
  92         {Opt_swalloc,   "swalloc"},     /* turn on stripe width allocation */
  93         {Opt_sunit,     "sunit=%u"},    /* data volume stripe unit */
  94         {Opt_swidth,    "swidth=%u"},   /* data volume stripe width */
  95         {Opt_nouuid,    "nouuid"},      /* ignore filesystem UUID */
  96         {Opt_mtpt,      "mtpt"},        /* filesystem mount point */
  97         {Opt_grpid,     "grpid"},       /* group-ID from parent directory */
  98         {Opt_nogrpid,   "nogrpid"},     /* group-ID from current process */
  99         {Opt_bsdgroups, "bsdgroups"},   /* group-ID from parent directory */
 100         {Opt_sysvgroups,"sysvgroups"},  /* group-ID from current process */
 101         {Opt_allocsize, "allocsize=%s"},/* preferred allocation size */
 102         {Opt_norecovery,"norecovery"},  /* don't run XFS recovery */
 103         {Opt_barrier,   "barrier"},     /* use writer barriers for log write and
 104                                          * unwritten extent conversion */
 105         {Opt_nobarrier, "nobarrier"},   /* .. disable */
 106         {Opt_inode64,   "inode64"},     /* inodes can be allocated anywhere */
 107         {Opt_inode32,   "inode32"},     /* inode allocation limited to
 108                                          * XFS_MAXINUMBER_32 */
 109         {Opt_ikeep,     "ikeep"},       /* do not free empty inode clusters */
 110         {Opt_noikeep,   "noikeep"},     /* free empty inode clusters */
 111         {Opt_largeio,   "largeio"},     /* report large I/O sizes in stat() */
 112         {Opt_nolargeio, "nolargeio"},   /* do not report large I/O sizes
 113                                          * in stat(). */
 114         {Opt_attr2,     "attr2"},       /* do use attr2 attribute format */
 115         {Opt_noattr2,   "noattr2"},     /* do not use attr2 attribute format */
 116         {Opt_filestreams,"filestreams"},/* use filestreams allocator */
 117         {Opt_quota,     "quota"},       /* disk quotas (user) */
 118         {Opt_noquota,   "noquota"},     /* no quotas */
 119         {Opt_usrquota,  "usrquota"},    /* user quota enabled */
 120         {Opt_grpquota,  "grpquota"},    /* group quota enabled */
 121         {Opt_prjquota,  "prjquota"},    /* project quota enabled */
 122         {Opt_uquota,    "uquota"},      /* user quota (IRIX variant) */
 123         {Opt_gquota,    "gquota"},      /* group quota (IRIX variant) */
 124         {Opt_pquota,    "pquota"},      /* project quota (IRIX variant) */
 125         {Opt_uqnoenforce,"uqnoenforce"},/* user quota limit enforcement */
 126         {Opt_gqnoenforce,"gqnoenforce"},/* group quota limit enforcement */
 127         {Opt_pqnoenforce,"pqnoenforce"},/* project quota limit enforcement */
 128         {Opt_qnoenforce, "qnoenforce"}, /* same as uqnoenforce */
 129         {Opt_discard,   "discard"},     /* Discard unused blocks */
 130         {Opt_nodiscard, "nodiscard"},   /* Do not discard unused blocks */
 131
 132         {Opt_dax,       "dax"},         /* Enable direct access to bdev pages */
 133         {Opt_err,       NULL},
 134 };
 135
 136
 137 STATIC int
 138 suffix_kstrtoint(const substring_t *s, unsigned int base, int *res)
 139 {
 140         int     last, shift_left_factor = 0, _res;
 141         char    *value;
 142         int     ret = 0;
 143
 144         value = match_strdup(s);
 145         if (!value)
 146                 return -ENOMEM;
 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         if (kstrtoint(value, base, &_res))
 163                 ret = -EINVAL;
 164         kfree(value);
 165         *res = _res << shift_left_factor;
 166         return ret;
 167 }
 168
 169 /*
 170  * This function fills in xfs_mount_t fields based on mount args.
 171  * Note: the superblock has _not_ yet been read in.
 172  *
 173  * Note that this function leaks the various device name allocations on
 174  * failure.  The caller takes care of them.
 175  *
 176  * *sb is const because this is also used to test options on the remount
 177  * path, and we don't want this to have any side effects at remount time.
 178  * Today this function does not change *sb, but just to future-proof...
 179  */
 180 STATIC int
 181 xfs_parseargs(
 182         struct xfs_mount        *mp,
 183         char                    *options)
 184 {
 185         const struct super_block *sb = mp->m_super;
 186         char                    *p;
 187         substring_t             args[MAX_OPT_ARGS];
 188         int                     dsunit = 0;
 189         int                     dswidth = 0;
 190         int                     iosize = 0;
 191         __uint8_t               iosizelog = 0;
 192
 193         /*
 194          * set up the mount name first so all the errors will refer to the
 195          * correct device.
 196          */
 197         mp->m_fsname = kstrndup(sb->s_id, MAXNAMELEN, GFP_KERNEL);
 198         if (!mp->m_fsname)
 199                 return -ENOMEM;
 200         mp->m_fsname_len = strlen(mp->m_fsname) + 1;
 201
 202         /*
 203          * Copy binary VFS mount flags we are interested in.
 204          */
 205         if (sb->s_flags & MS_RDONLY)
 206                 mp->m_flags |= XFS_MOUNT_RDONLY;
 207         if (sb->s_flags & MS_DIRSYNC)
 208                 mp->m_flags |= XFS_MOUNT_DIRSYNC;
 209         if (sb->s_flags & MS_SYNCHRONOUS)
 210                 mp->m_flags |= XFS_MOUNT_WSYNC;
 211
 212         /*
 213          * Set some default flags that could be cleared by the mount option
 214          * parsing.
 215          */
 216         mp->m_flags |= XFS_MOUNT_BARRIER;
 217         mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE;
 218
 219         /*
 220          * These can be overridden by the mount option parsing.
 221          */
 222         mp->m_logbufs = -1;
 223         mp->m_logbsize = -1;
 224
 225         if (!options)
 226                 goto done;
 227
 228         while ((p = strsep(&options, ",")) != NULL) {
 229                 int             token;
 230
 231                 if (!*p)
 232                         continue;
 233
 234                 token = match_token(p, tokens, args);
 235                 switch (token) {
 236                 case Opt_logbufs:
 237                         if (match_int(args, &mp->m_logbufs))
 238                                 return -EINVAL;
 239                         break;
 240                 case Opt_logbsize:
 241                         if (suffix_kstrtoint(args, 10, &mp->m_logbsize))
 242                                 return -EINVAL;
 243                         break;
 244                 case Opt_logdev:
 245                         mp->m_logname = match_strdup(args);
 246                         if (!mp->m_logname)
 247                                 return -ENOMEM;
 248                         break;
 249                 case Opt_mtpt:
 250                         xfs_warn(mp, "%s option not allowed on this system", p);
 251                         return -EINVAL;
 252                 case Opt_rtdev:
 253                         mp->m_rtname = match_strdup(args);
 254                         if (!mp->m_rtname)
 255                                 return -ENOMEM;
 256                         break;
 257                 case Opt_allocsize:
 258                 case Opt_biosize:
 259                         if (suffix_kstrtoint(args, 10, &iosize))
 260                                 return -EINVAL;
 261                         iosizelog = ffs(iosize) - 1;
 262                         break;
 263                 case Opt_grpid:
 264                 case Opt_bsdgroups:
 265                         mp->m_flags |= XFS_MOUNT_GRPID;
 266                         break;
 267                 case Opt_nogrpid:
 268                 case Opt_sysvgroups:
 269                         mp->m_flags &= ~XFS_MOUNT_GRPID;
 270                         break;
 271                 case Opt_wsync:
 272                         mp->m_flags |= XFS_MOUNT_WSYNC;
 273                         break;
 274                 case Opt_norecovery:
 275                         mp->m_flags |= XFS_MOUNT_NORECOVERY;
 276                         break;
 277                 case Opt_noalign:
 278                         mp->m_flags |= XFS_MOUNT_NOALIGN;
 279                         break;
 280                 case Opt_swalloc:
 281                         mp->m_flags |= XFS_MOUNT_SWALLOC;
 282                         break;
 283                 case Opt_sunit:
 284                         if (match_int(args, &dsunit))
 285                                 return -EINVAL;
 286                         break;
 287                 case Opt_swidth:
 288                         if (match_int(args, &dswidth))
 289                                 return -EINVAL;
 290                         break;
 291                 case Opt_inode32:
 292                         mp->m_flags |= XFS_MOUNT_SMALL_INUMS;
 293                         break;
 294                 case Opt_inode64:
 295                         mp->m_flags &= ~XFS_MOUNT_SMALL_INUMS;
 296                         break;
 297                 case Opt_nouuid:
 298                         mp->m_flags |= XFS_MOUNT_NOUUID;
 299                         break;
 300                 case Opt_barrier:
 301                         mp->m_flags |= XFS_MOUNT_BARRIER;
 302                         break;
 303                 case Opt_nobarrier:
 304                         mp->m_flags &= ~XFS_MOUNT_BARRIER;
 305                         break;
 306                 case Opt_ikeep:
 307                         mp->m_flags |= XFS_MOUNT_IKEEP;
 308                         break;
 309                 case Opt_noikeep:
 310                         mp->m_flags &= ~XFS_MOUNT_IKEEP;
 311                         break;
 312                 case Opt_largeio:
 313                         mp->m_flags &= ~XFS_MOUNT_COMPAT_IOSIZE;
 314                         break;
 315                 case Opt_nolargeio:
 316                         mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE;
 317                         break;
 318                 case Opt_attr2:
 319                         mp->m_flags |= XFS_MOUNT_ATTR2;
 320                         break;
 321                 case Opt_noattr2:
 322                         mp->m_flags &= ~XFS_MOUNT_ATTR2;
 323                         mp->m_flags |= XFS_MOUNT_NOATTR2;
 324                         break;
 325                 case Opt_filestreams:
 326                         mp->m_flags |= XFS_MOUNT_FILESTREAMS;
 327                         break;
 328                 case Opt_noquota:
 329                         mp->m_qflags &= ~XFS_ALL_QUOTA_ACCT;
 330                         mp->m_qflags &= ~XFS_ALL_QUOTA_ENFD;
 331                         mp->m_qflags &= ~XFS_ALL_QUOTA_ACTIVE;
 332                         break;
 333                 case Opt_quota:
 334                 case Opt_uquota:
 335                 case Opt_usrquota:
 336                         mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE |
 337                                          XFS_UQUOTA_ENFD);
 338                         break;
 339                 case Opt_qnoenforce:
 340                 case Opt_uqnoenforce:
 341                         mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE);
 342                         mp->m_qflags &= ~XFS_UQUOTA_ENFD;
 343                         break;
 344                 case Opt_pquota:
 345                 case Opt_prjquota:
 346                         mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE |
 347                                          XFS_PQUOTA_ENFD);
 348                         break;
 349                 case Opt_pqnoenforce:
 350                         mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE);
 351                         mp->m_qflags &= ~XFS_PQUOTA_ENFD;
 352                 case Opt_gquota:
 353                 case Opt_grpquota:
 354                         mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE |
 355                                          XFS_GQUOTA_ENFD);
 356                         break;
 357                 case Opt_gqnoenforce:
 358                         mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE);
 359                         mp->m_qflags &= ~XFS_GQUOTA_ENFD;
 360                         break;
 361                 case Opt_discard:
 362                         mp->m_flags |= XFS_MOUNT_DISCARD;
 363                         break;
 364                 case Opt_nodiscard:
 365                         mp->m_flags &= ~XFS_MOUNT_DISCARD;
 366                         break;
 367 #ifdef CONFIG_FS_DAX
 368                 case Opt_dax:
 369                         mp->m_flags |= XFS_MOUNT_DAX;
 370                         break;
 371 #endif
 372                 default:
 373                         xfs_warn(mp, "unknown mount option [%s].", p);
 374                         return -EINVAL;
 375                 }
 376         }
 377
 378         /*
 379          * no recovery flag requires a read-only mount
 380          */
 381         if ((mp->m_flags & XFS_MOUNT_NORECOVERY) &&
 382             !(mp->m_flags & XFS_MOUNT_RDONLY)) {
 383                 xfs_warn(mp, "no-recovery mounts must be read-only.");
 384                 return -EINVAL;
 385         }
 386
 387         if ((mp->m_flags & XFS_MOUNT_NOALIGN) && (dsunit || dswidth)) {
 388                 xfs_warn(mp,
 389         "sunit and swidth options incompatible with the noalign option");
 390                 return -EINVAL;
 391         }
 392
 393 #ifndef CONFIG_XFS_QUOTA
 394         if (XFS_IS_QUOTA_RUNNING(mp)) {
 395                 xfs_warn(mp, "quota support not available in this kernel.");
 396                 return -EINVAL;
 397         }
 398 #endif
 399
 400         if ((dsunit && !dswidth) || (!dsunit && dswidth)) {
 401                 xfs_warn(mp, "sunit and swidth must be specified together");
 402                 return -EINVAL;
 403         }
 404
 405         if (dsunit && (dswidth % dsunit != 0)) {
 406                 xfs_warn(mp,
 407         "stripe width (%d) must be a multiple of the stripe unit (%d)",
 408                         dswidth, dsunit);
 409                 return -EINVAL;
 410         }
 411
 412 done:
 413         if (dsunit && !(mp->m_flags & XFS_MOUNT_NOALIGN)) {
 414                 /*
 415                  * At this point the superblock has not been read
 416                  * in, therefore we do not know the block size.
 417                  * Before the mount call ends we will convert
 418                  * these to FSBs.
 419                  */
 420                 mp->m_dalign = dsunit;
 421                 mp->m_swidth = dswidth;
 422         }
 423
 424         if (mp->m_logbufs != -1 &&
 425             mp->m_logbufs != 0 &&
 426             (mp->m_logbufs < XLOG_MIN_ICLOGS ||
 427              mp->m_logbufs > XLOG_MAX_ICLOGS)) {
 428                 xfs_warn(mp, "invalid logbufs value: %d [not %d-%d]",
 429                         mp->m_logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
 430                 return -EINVAL;
 431         }
 432         if (mp->m_logbsize != -1 &&
 433             mp->m_logbsize !=  0 &&
 434             (mp->m_logbsize < XLOG_MIN_RECORD_BSIZE ||
 435              mp->m_logbsize > XLOG_MAX_RECORD_BSIZE ||
 436              !is_power_of_2(mp->m_logbsize))) {
 437                 xfs_warn(mp,
 438                         "invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
 439                         mp->m_logbsize);
 440                 return -EINVAL;
 441         }
 442
 443         if (iosizelog) {
 444                 if (iosizelog > XFS_MAX_IO_LOG ||
 445                     iosizelog < XFS_MIN_IO_LOG) {
 446                         xfs_warn(mp, "invalid log iosize: %d [not %d-%d]",
 447                                 iosizelog, XFS_MIN_IO_LOG,
 448                                 XFS_MAX_IO_LOG);
 449                         return -EINVAL;
 450                 }
 451
 452                 mp->m_flags |= XFS_MOUNT_DFLT_IOSIZE;
 453                 mp->m_readio_log = iosizelog;
 454                 mp->m_writeio_log = iosizelog;
 455         }
 456
 457         return 0;
 458 }
 459
 460 struct proc_xfs_info {
 461         uint64_t        flag;
 462         char            *str;
 463 };
 464
 465 STATIC int
 466 xfs_showargs(
 467         struct xfs_mount        *mp,
 468         struct seq_file         *m)
 469 {
 470         static struct proc_xfs_info xfs_info_set[] = {
 471                 /* the few simple ones we can get from the mount struct */
 472                 { XFS_MOUNT_IKEEP,              ",ikeep" },
 473                 { XFS_MOUNT_WSYNC,              ",wsync" },
 474                 { XFS_MOUNT_NOALIGN,            ",noalign" },
 475                 { XFS_MOUNT_SWALLOC,            ",swalloc" },
 476                 { XFS_MOUNT_NOUUID,             ",nouuid" },
 477                 { XFS_MOUNT_NORECOVERY,         ",norecovery" },
 478                 { XFS_MOUNT_ATTR2,              ",attr2" },
 479                 { XFS_MOUNT_FILESTREAMS,        ",filestreams" },
 480                 { XFS_MOUNT_GRPID,              ",grpid" },
 481                 { XFS_MOUNT_DISCARD,            ",discard" },
 482                 { XFS_MOUNT_SMALL_INUMS,        ",inode32" },
 483                 { XFS_MOUNT_DAX,                ",dax" },
 484                 { 0, NULL }
 485         };
 486         static struct proc_xfs_info xfs_info_unset[] = {
 487                 /* the few simple ones we can get from the mount struct */
 488                 { XFS_MOUNT_COMPAT_IOSIZE,      ",largeio" },
 489                 { XFS_MOUNT_BARRIER,            ",nobarrier" },
 490                 { XFS_MOUNT_SMALL_INUMS,        ",inode64" },
 491                 { 0, NULL }
 492         };
 493         struct proc_xfs_info    *xfs_infop;
 494
 495         for (xfs_infop = xfs_info_set; xfs_infop->flag; xfs_infop++) {
 496                 if (mp->m_flags & xfs_infop->flag)
 497                         seq_puts(m, xfs_infop->str);
 498         }
 499         for (xfs_infop = xfs_info_unset; xfs_infop->flag; xfs_infop++) {
 500                 if (!(mp->m_flags & xfs_infop->flag))
 501                         seq_puts(m, xfs_infop->str);
 502         }
 503
 504         if (mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)
 505                 seq_printf(m, ",allocsize=%dk",
 506                                 (int)(1 << mp->m_writeio_log) >> 10);
 507
 508         if (mp->m_logbufs > 0)
 509                 seq_printf(m, ",logbufs=%d", mp->m_logbufs);
 510         if (mp->m_logbsize > 0)
 511                 seq_printf(m, ",logbsize=%dk", mp->m_logbsize >> 10);
 512
 513         if (mp->m_logname)
 514                 seq_show_option(m, "logdev", mp->m_logname);
 515         if (mp->m_rtname)
 516                 seq_show_option(m, "rtdev", mp->m_rtname);
 517
 518         if (mp->m_dalign > 0)
 519                 seq_printf(m, ",sunit=%d",
 520                                 (int)XFS_FSB_TO_BB(mp, mp->m_dalign));
 521         if (mp->m_swidth > 0)
 522                 seq_printf(m, ",swidth=%d",
 523                                 (int)XFS_FSB_TO_BB(mp, mp->m_swidth));
 524
 525         if (mp->m_qflags & (XFS_UQUOTA_ACCT|XFS_UQUOTA_ENFD))
 526                 seq_puts(m, ",usrquota");
 527         else if (mp->m_qflags & XFS_UQUOTA_ACCT)
 528                 seq_puts(m, ",uqnoenforce");
 529
 530         if (mp->m_qflags & XFS_PQUOTA_ACCT) {
 531                 if (mp->m_qflags & XFS_PQUOTA_ENFD)
 532                         seq_puts(m, ",prjquota");
 533                 else
 534                         seq_puts(m, ",pqnoenforce");
 535         }
 536         if (mp->m_qflags & XFS_GQUOTA_ACCT) {
 537                 if (mp->m_qflags & XFS_GQUOTA_ENFD)
 538                         seq_puts(m, ",grpquota");
 539                 else
 540                         seq_puts(m, ",gqnoenforce");
 541         }
 542
 543         if (!(mp->m_qflags & XFS_ALL_QUOTA_ACCT))
 544                 seq_puts(m, ",noquota");
 545
 546         return 0;
 547 }
 548 __uint64_t
 549 xfs_max_file_offset(
 550         unsigned int            blockshift)
 551 {
 552         unsigned int            pagefactor = 1;
 553         unsigned int            bitshift = BITS_PER_LONG - 1;
 554
 555         /* Figure out maximum filesize, on Linux this can depend on
 556          * the filesystem blocksize (on 32 bit platforms).
 557          * __block_write_begin does this in an [unsigned] long...
 558          *      page->index << (PAGE_CACHE_SHIFT - bbits)
 559          * So, for page sized blocks (4K on 32 bit platforms),
 560          * this wraps at around 8Tb (hence MAX_LFS_FILESIZE which is
 561          *      (((u64)PAGE_CACHE_SIZE << (BITS_PER_LONG-1))-1)
 562          * but for smaller blocksizes it is less (bbits = log2 bsize).
 563          * Note1: get_block_t takes a long (implicit cast from above)
 564          * Note2: The Large Block Device (LBD and HAVE_SECTOR_T) patch
 565          * can optionally convert the [unsigned] long from above into
 566          * an [unsigned] long long.
 567          */
 568
 569 #if BITS_PER_LONG == 32
 570 # if defined(CONFIG_LBDAF)
 571         ASSERT(sizeof(sector_t) == 8);
 572         pagefactor = PAGE_CACHE_SIZE;
 573         bitshift = BITS_PER_LONG;
 574 # else
 575         pagefactor = PAGE_CACHE_SIZE >> (PAGE_CACHE_SHIFT - blockshift);
 576 # endif
 577 #endif
 578
 579         return (((__uint64_t)pagefactor) << bitshift) - 1;
 580 }
 581
 582 /*
 583  * xfs_set_inode32() and xfs_set_inode64() are passed an agcount
 584  * because in the growfs case, mp->m_sb.sb_agcount is not updated
 585  * yet to the potentially higher ag count.
 586  */
 587 xfs_agnumber_t
 588 xfs_set_inode32(struct xfs_mount *mp, xfs_agnumber_t agcount)
 589 {
 590         xfs_agnumber_t  index = 0;
 591         xfs_agnumber_t  maxagi = 0;
 592         xfs_sb_t        *sbp = &mp->m_sb;
 593         xfs_agnumber_t  max_metadata;
 594         xfs_agino_t     agino;
 595         xfs_ino_t       ino;
 596         xfs_perag_t     *pag;
 597
 598         /* Calculate how much should be reserved for inodes to meet
 599          * the max inode percentage.
 600          */
 601         if (mp->m_maxicount) {
 602                 __uint64_t      icount;
 603
 604                 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
 605                 do_div(icount, 100);
 606                 icount += sbp->sb_agblocks - 1;
 607                 do_div(icount, sbp->sb_agblocks);
 608                 max_metadata = icount;
 609         } else {
 610                 max_metadata = agcount;
 611         }
 612
 613         agino = XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks - 1, 0);
 614
 615         for (index = 0; index < agcount; index++) {
 616                 ino = XFS_AGINO_TO_INO(mp, index, agino);
 617
 618                 if (ino > XFS_MAXINUMBER_32) {
 619                         pag = xfs_perag_get(mp, index);
 620                         pag->pagi_inodeok = 0;
 621                         pag->pagf_metadata = 0;
 622                         xfs_perag_put(pag);
 623                         continue;
 624                 }
 625
 626                 pag = xfs_perag_get(mp, index);
 627                 pag->pagi_inodeok = 1;
 628                 maxagi++;
 629                 if (index < max_metadata)
 630                         pag->pagf_metadata = 1;
 631                 xfs_perag_put(pag);
 632         }
 633         mp->m_flags |= (XFS_MOUNT_32BITINODES |
 634                         XFS_MOUNT_SMALL_INUMS);
 635
 636         return maxagi;
 637 }
 638
 639 xfs_agnumber_t
 640 xfs_set_inode64(struct xfs_mount *mp, xfs_agnumber_t agcount)
 641 {
 642         xfs_agnumber_t index = 0;
 643
 644         for (index = 0; index < agcount; index++) {
 645                 struct xfs_perag        *pag;
 646
 647                 pag = xfs_perag_get(mp, index);
 648                 pag->pagi_inodeok = 1;
 649                 pag->pagf_metadata = 0;
 650                 xfs_perag_put(pag);
 651         }
 652
 653         /* There is no need for lock protection on m_flags,
 654          * the rw_semaphore of the VFS superblock is locked
 655          * during mount/umount/remount operations, so this is
 656          * enough to avoid concurency on the m_flags field
 657          */
 658         mp->m_flags &= ~(XFS_MOUNT_32BITINODES |
 659                          XFS_MOUNT_SMALL_INUMS);
 660         return index;
 661 }
 662
 663 STATIC int
 664 xfs_blkdev_get(
 665         xfs_mount_t             *mp,
 666         const char              *name,
 667         struct block_device     **bdevp)
 668 {
 669         int                     error = 0;
 670
 671         *bdevp = blkdev_get_by_path(name, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
 672                                     mp);
 673         if (IS_ERR(*bdevp)) {
 674                 error = PTR_ERR(*bdevp);
 675                 xfs_warn(mp, "Invalid device [%s], error=%d", name, error);
 676         }
 677
 678         return error;
 679 }
 680
 681 STATIC void
 682 xfs_blkdev_put(
 683         struct block_device     *bdev)
 684 {
 685         if (bdev)
 686                 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 687 }
 688
 689 void
 690 xfs_blkdev_issue_flush(
 691         xfs_buftarg_t           *buftarg)
 692 {
 693         blkdev_issue_flush(buftarg->bt_bdev, GFP_NOFS, NULL);
 694 }
 695
 696 STATIC void
 697 xfs_close_devices(
 698         struct xfs_mount        *mp)
 699 {
 700         if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
 701                 struct block_device *logdev = mp->m_logdev_targp->bt_bdev;
 702                 xfs_free_buftarg(mp, mp->m_logdev_targp);
 703                 xfs_blkdev_put(logdev);
 704         }
 705         if (mp->m_rtdev_targp) {
 706                 struct block_device *rtdev = mp->m_rtdev_targp->bt_bdev;
 707                 xfs_free_buftarg(mp, mp->m_rtdev_targp);
 708                 xfs_blkdev_put(rtdev);
 709         }
 710         xfs_free_buftarg(mp, mp->m_ddev_targp);
 711 }
 712
 713 /*
 714  * The file system configurations are:
 715  *      (1) device (partition) with data and internal log
 716  *      (2) logical volume with data and log subvolumes.
 717  *      (3) logical volume with data, log, and realtime subvolumes.
 718  *
 719  * We only have to handle opening the log and realtime volumes here if
 720  * they are present.  The data subvolume has already been opened by
 721  * get_sb_bdev() and is stored in sb->s_bdev.
 722  */
 723 STATIC int
 724 xfs_open_devices(
 725         struct xfs_mount        *mp)
 726 {
 727         struct block_device     *ddev = mp->m_super->s_bdev;
 728         struct block_device     *logdev = NULL, *rtdev = NULL;
 729         int                     error;
 730
 731         /*
 732          * Open real time and log devices - order is important.
 733          */
 734         if (mp->m_logname) {
 735                 error = xfs_blkdev_get(mp, mp->m_logname, &logdev);
 736                 if (error)
 737                         goto out;
 738         }
 739
 740         if (mp->m_rtname) {
 741                 error = xfs_blkdev_get(mp, mp->m_rtname, &rtdev);
 742                 if (error)
 743                         goto out_close_logdev;
 744
 745                 if (rtdev == ddev || rtdev == logdev) {
 746                         xfs_warn(mp,
 747         "Cannot mount filesystem with identical rtdev and ddev/logdev.");
 748                         error = -EINVAL;
 749                         goto out_close_rtdev;
 750                 }
 751         }
 752
 753         /*
 754          * Setup xfs_mount buffer target pointers
 755          */
 756         error = -ENOMEM;
 757         mp->m_ddev_targp = xfs_alloc_buftarg(mp, ddev);
 758         if (!mp->m_ddev_targp)
 759                 goto out_close_rtdev;
 760
 761         if (rtdev) {
 762                 mp->m_rtdev_targp = xfs_alloc_buftarg(mp, rtdev);
 763                 if (!mp->m_rtdev_targp)
 764                         goto out_free_ddev_targ;
 765         }
 766
 767         if (logdev && logdev != ddev) {
 768                 mp->m_logdev_targp = xfs_alloc_buftarg(mp, logdev);
 769                 if (!mp->m_logdev_targp)
 770                         goto out_free_rtdev_targ;
 771         } else {
 772                 mp->m_logdev_targp = mp->m_ddev_targp;
 773         }
 774
 775         return 0;
 776
 777  out_free_rtdev_targ:
 778         if (mp->m_rtdev_targp)
 779                 xfs_free_buftarg(mp, mp->m_rtdev_targp);
 780  out_free_ddev_targ:
 781         xfs_free_buftarg(mp, mp->m_ddev_targp);
 782  out_close_rtdev:
 783         xfs_blkdev_put(rtdev);
 784  out_close_logdev:
 785         if (logdev && logdev != ddev)
 786                 xfs_blkdev_put(logdev);
 787  out:
 788         return error;
 789 }
 790
 791 /*
 792  * Setup xfs_mount buffer target pointers based on superblock
 793  */
 794 STATIC int
 795 xfs_setup_devices(
 796         struct xfs_mount        *mp)
 797 {
 798         int                     error;
 799
 800         error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_sectsize);
 801         if (error)
 802                 return error;
 803
 804         if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
 805                 unsigned int    log_sector_size = BBSIZE;
 806
 807                 if (xfs_sb_version_hassector(&mp->m_sb))
 808                         log_sector_size = mp->m_sb.sb_logsectsize;
 809                 error = xfs_setsize_buftarg(mp->m_logdev_targp,
 810                                             log_sector_size);
 811                 if (error)
 812                         return error;
 813         }
 814         if (mp->m_rtdev_targp) {
 815                 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
 816                                             mp->m_sb.sb_sectsize);
 817                 if (error)
 818                         return error;
 819         }
 820
 821         return 0;
 822 }
 823
 824 STATIC int
 825 xfs_init_mount_workqueues(
 826         struct xfs_mount        *mp)
 827 {
 828         mp->m_buf_workqueue = alloc_workqueue("xfs-buf/%s",
 829                         WQ_MEM_RECLAIM|WQ_FREEZABLE, 1, mp->m_fsname);
 830         if (!mp->m_buf_workqueue)
 831                 goto out;
 832
 833         mp->m_data_workqueue = alloc_workqueue("xfs-data/%s",
 834                         WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname);
 835         if (!mp->m_data_workqueue)
 836                 goto out_destroy_buf;
 837
 838         mp->m_unwritten_workqueue = alloc_workqueue("xfs-conv/%s",
 839                         WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname);
 840         if (!mp->m_unwritten_workqueue)
 841                 goto out_destroy_data_iodone_queue;
 842
 843         mp->m_cil_workqueue = alloc_workqueue("xfs-cil/%s",
 844                         WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname);
 845         if (!mp->m_cil_workqueue)
 846                 goto out_destroy_unwritten;
 847
 848         mp->m_reclaim_workqueue = alloc_workqueue("xfs-reclaim/%s",
 849                         WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname);
 850         if (!mp->m_reclaim_workqueue)
 851                 goto out_destroy_cil;
 852
 853         mp->m_log_workqueue = alloc_workqueue("xfs-log/%s",
 854                         WQ_MEM_RECLAIM|WQ_FREEZABLE|WQ_HIGHPRI, 0,
 855                         mp->m_fsname);
 856         if (!mp->m_log_workqueue)
 857                 goto out_destroy_reclaim;
 858
 859         mp->m_eofblocks_workqueue = alloc_workqueue("xfs-eofblocks/%s",
 860                         WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname);
 861         if (!mp->m_eofblocks_workqueue)
 862                 goto out_destroy_log;
 863
 864         return 0;
 865
 866 out_destroy_log:
 867         destroy_workqueue(mp->m_log_workqueue);
 868 out_destroy_reclaim:
 869         destroy_workqueue(mp->m_reclaim_workqueue);
 870 out_destroy_cil:
 871         destroy_workqueue(mp->m_cil_workqueue);
 872 out_destroy_unwritten:
 873         destroy_workqueue(mp->m_unwritten_workqueue);
 874 out_destroy_data_iodone_queue:
 875         destroy_workqueue(mp->m_data_workqueue);
 876 out_destroy_buf:
 877         destroy_workqueue(mp->m_buf_workqueue);
 878 out:
 879         return -ENOMEM;
 880 }
 881
 882 STATIC void
 883 xfs_destroy_mount_workqueues(
 884         struct xfs_mount        *mp)
 885 {
 886         destroy_workqueue(mp->m_eofblocks_workqueue);
 887         destroy_workqueue(mp->m_log_workqueue);
 888         destroy_workqueue(mp->m_reclaim_workqueue);
 889         destroy_workqueue(mp->m_cil_workqueue);
 890         destroy_workqueue(mp->m_data_workqueue);
 891         destroy_workqueue(mp->m_unwritten_workqueue);
 892         destroy_workqueue(mp->m_buf_workqueue);
 893 }
 894
 895 /*
 896  * Flush all dirty data to disk. Must not be called while holding an XFS_ILOCK
 897  * or a page lock. We use sync_inodes_sb() here to ensure we block while waiting
 898  * for IO to complete so that we effectively throttle multiple callers to the
 899  * rate at which IO is completing.
 900  */
 901 void
 902 xfs_flush_inodes(
 903         struct xfs_mount        *mp)
 904 {
 905         struct super_block      *sb = mp->m_super;
 906
 907         if (down_read_trylock(&sb->s_umount)) {
 908                 sync_inodes_sb(sb);
 909                 up_read(&sb->s_umount);
 910         }
 911 }
 912
 913 /* Catch misguided souls that try to use this interface on XFS */
 914 STATIC struct inode *
 915 xfs_fs_alloc_inode(
 916         struct super_block      *sb)
 917 {
 918         BUG();
 919         return NULL;
 920 }
 921
 922 /*
 923  * Now that the generic code is guaranteed not to be accessing
 924  * the linux inode, we can reclaim the inode.
 925  */
 926 STATIC void
 927 xfs_fs_destroy_inode(
 928         struct inode            *inode)
 929 {
 930         struct xfs_inode        *ip = XFS_I(inode);
 931
 932         trace_xfs_destroy_inode(ip);
 933
 934         XFS_STATS_INC(ip->i_mount, vn_reclaim);
 935
 936         ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0);
 937
 938         /*
 939          * We should never get here with one of the reclaim flags already set.
 940          */
 941         ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_IRECLAIMABLE));
 942         ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_IRECLAIM));
 943
 944         /*
 945          * We always use background reclaim here because even if the
 946          * inode is clean, it still may be under IO and hence we have
 947          * to take the flush lock. The background reclaim path handles
 948          * this more efficiently than we can here, so simply let background
 949          * reclaim tear down all inodes.
 950          */
 951         xfs_inode_set_reclaim_tag(ip);
 952 }
 953
 954 /*
 955  * Slab object creation initialisation for the XFS inode.
 956  * This covers only the idempotent fields in the XFS inode;
 957  * all other fields need to be initialised on allocation
 958  * from the slab. This avoids the need to repeatedly initialise
 959  * fields in the xfs inode that left in the initialise state
 960  * when freeing the inode.
 961  */
 962 STATIC void
 963 xfs_fs_inode_init_once(
 964         void                    *inode)
 965 {
 966         struct xfs_inode        *ip = inode;
 967
 968         memset(ip, 0, sizeof(struct xfs_inode));
 969
 970         /* vfs inode */
 971         inode_init_once(VFS_I(ip));
 972
 973         /* xfs inode */
 974         atomic_set(&ip->i_pincount, 0);
 975         spin_lock_init(&ip->i_flags_lock);
 976
 977         mrlock_init(&ip->i_mmaplock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
 978                      "xfsino", ip->i_ino);
 979         mrlock_init(&ip->i_lock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
 980                      "xfsino", ip->i_ino);
 981 }
 982
 983 STATIC void
 984 xfs_fs_evict_inode(
 985         struct inode            *inode)
 986 {
 987         xfs_inode_t             *ip = XFS_I(inode);
 988
 989         ASSERT(!rwsem_is_locked(&ip->i_iolock.mr_lock));
 990
 991         trace_xfs_evict_inode(ip);
 992
 993         truncate_inode_pages_final(&inode->i_data);
 994         clear_inode(inode);
 995         XFS_STATS_INC(ip->i_mount, vn_rele);
 996         XFS_STATS_INC(ip->i_mount, vn_remove);
 997
 998         xfs_inactive(ip);
 999 }
1000
1001 /*
1002  * We do an unlocked check for XFS_IDONTCACHE here because we are already
1003  * serialised against cache hits here via the inode->i_lock and igrab() in
1004  * xfs_iget_cache_hit(). Hence a lookup that might clear this flag will not be
1005  * racing with us, and it avoids needing to grab a spinlock here for every inode
1006  * we drop the final reference on.
1007  */
1008 STATIC int
1009 xfs_fs_drop_inode(
1010         struct inode            *inode)
1011 {
1012         struct xfs_inode        *ip = XFS_I(inode);
1013
1014         return generic_drop_inode(inode) || (ip->i_flags & XFS_IDONTCACHE);
1015 }
1016
1017 STATIC void
1018 xfs_free_fsname(
1019         struct xfs_mount        *mp)
1020 {
1021         kfree(mp->m_fsname);
1022         kfree(mp->m_rtname);
1023         kfree(mp->m_logname);
1024 }
1025
1026 STATIC int
1027 xfs_fs_sync_fs(
1028         struct super_block      *sb,
1029         int                     wait)
1030 {
1031         struct xfs_mount        *mp = XFS_M(sb);
1032
1033         /*
1034          * Doing anything during the async pass would be counterproductive.
1035          */
1036         if (!wait)
1037                 return 0;
1038
1039         xfs_log_force(mp, XFS_LOG_SYNC);
1040         if (laptop_mode) {
1041                 /*
1042                  * The disk must be active because we're syncing.
1043                  * We schedule log work now (now that the disk is
1044                  * active) instead of later (when it might not be).
1045                  */
1046                 flush_delayed_work(&mp->m_log->l_work);
1047         }
1048
1049         return 0;
1050 }
1051
1052 STATIC int
1053 xfs_fs_statfs(
1054         struct dentry           *dentry,
1055         struct kstatfs          *statp)
1056 {
1057         struct xfs_mount        *mp = XFS_M(dentry->d_sb);
1058         xfs_sb_t                *sbp = &mp->m_sb;
1059         struct xfs_inode        *ip = XFS_I(d_inode(dentry));
1060         __uint64_t              fakeinos, id;
1061         __uint64_t              icount;
1062         __uint64_t              ifree;
1063         __uint64_t              fdblocks;
1064         xfs_extlen_t            lsize;
1065         __int64_t               ffree;
1066
1067         statp->f_type = XFS_SB_MAGIC;
1068         statp->f_namelen = MAXNAMELEN - 1;
1069
1070         id = huge_encode_dev(mp->m_ddev_targp->bt_dev);
1071         statp->f_fsid.val[0] = (u32)id;
1072         statp->f_fsid.val[1] = (u32)(id >> 32);
1073
1074         icount = percpu_counter_sum(&mp->m_icount);
1075         ifree = percpu_counter_sum(&mp->m_ifree);
1076         fdblocks = percpu_counter_sum(&mp->m_fdblocks);
1077
1078         spin_lock(&mp->m_sb_lock);
1079         statp->f_bsize = sbp->sb_blocksize;
1080         lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
1081         statp->f_blocks = sbp->sb_dblocks - lsize;
1082         spin_unlock(&mp->m_sb_lock);
1083
1084         statp->f_bfree = fdblocks - XFS_ALLOC_SET_ASIDE(mp);
1085         statp->f_bavail = statp->f_bfree;
1086
1087         fakeinos = statp->f_bfree << sbp->sb_inopblog;
1088         statp->f_files = MIN(icount + fakeinos, (__uint64_t)XFS_MAXINUMBER);
1089         if (mp->m_maxicount)
1090                 statp->f_files = min_t(typeof(statp->f_files),
1091                                         statp->f_files,
1092                                         mp->m_maxicount);
1093
1094         /* If sb_icount overshot maxicount, report actual allocation */
1095         statp->f_files = max_t(typeof(statp->f_files),
1096                                         statp->f_files,
1097                                         sbp->sb_icount);
1098
1099         /* make sure statp->f_ffree does not underflow */
1100         ffree = statp->f_files - (icount - ifree);
1101         statp->f_ffree = max_t(__int64_t, ffree, 0);
1102
1103
1104         if ((ip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
1105             ((mp->m_qflags & (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))) ==
1106                               (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))
1107                 xfs_qm_statvfs(ip, statp);
1108         return 0;
1109 }
1110
1111 STATIC void
1112 xfs_save_resvblks(struct xfs_mount *mp)
1113 {
1114         __uint64_t resblks = 0;
1115
1116         mp->m_resblks_save = mp->m_resblks;
1117         xfs_reserve_blocks(mp, &resblks, NULL);
1118 }
1119
1120 STATIC void
1121 xfs_restore_resvblks(struct xfs_mount *mp)
1122 {
1123         __uint64_t resblks;
1124
1125         if (mp->m_resblks_save) {
1126                 resblks = mp->m_resblks_save;
1127                 mp->m_resblks_save = 0;
1128         } else
1129                 resblks = xfs_default_resblks(mp);
1130
1131         xfs_reserve_blocks(mp, &resblks, NULL);
1132 }
1133
1134 /*
1135  * Trigger writeback of all the dirty metadata in the file system.
1136  *
1137  * This ensures that the metadata is written to their location on disk rather
1138  * than just existing in transactions in the log. This means after a quiesce
1139  * there is no log replay required to write the inodes to disk - this is the
1140  * primary difference between a sync and a quiesce.
1141  *
1142  * Note: xfs_log_quiesce() stops background log work - the callers must ensure
1143  * it is started again when appropriate.
1144  */
1145 static void
1146 xfs_quiesce_attr(
1147         struct xfs_mount        *mp)
1148 {
1149         int     error = 0;
1150
1151         /* wait for all modifications to complete */
1152         while (atomic_read(&mp->m_active_trans) > 0)
1153                 delay(100);
1154
1155         /* force the log to unpin objects from the now complete transactions */
1156         xfs_log_force(mp, XFS_LOG_SYNC);
1157
1158         /* reclaim inodes to do any IO before the freeze completes */
1159         xfs_reclaim_inodes(mp, 0);
1160         xfs_reclaim_inodes(mp, SYNC_WAIT);
1161
1162         /* Push the superblock and write an unmount record */
1163         error = xfs_log_sbcount(mp);
1164         if (error)
1165                 xfs_warn(mp, "xfs_attr_quiesce: failed to log sb changes. "
1166                                 "Frozen image may not be consistent.");
1167         /*
1168          * Just warn here till VFS can correctly support
1169          * read-only remount without racing.
1170          */
1171         WARN_ON(atomic_read(&mp->m_active_trans) != 0);
1172
1173         xfs_log_quiesce(mp);
1174 }
1175
1176 STATIC int
1177 xfs_test_remount_options(
1178         struct super_block      *sb,
1179         struct xfs_mount        *mp,
1180         char                    *options)
1181 {
1182         int                     error = 0;
1183         struct xfs_mount        *tmp_mp;
1184
1185         tmp_mp = kmem_zalloc(sizeof(*tmp_mp), KM_MAYFAIL);
1186         if (!tmp_mp)
1187                 return -ENOMEM;
1188
1189         tmp_mp->m_super = sb;
1190         error = xfs_parseargs(tmp_mp, options);
1191         xfs_free_fsname(tmp_mp);
1192         kfree(tmp_mp);
1193
1194         return error;
1195 }
1196
1197 STATIC int
1198 xfs_fs_remount(
1199         struct super_block      *sb,
1200         int                     *flags,
1201         char                    *options)
1202 {
1203         struct xfs_mount        *mp = XFS_M(sb);
1204         xfs_sb_t                *sbp = &mp->m_sb;
1205         substring_t             args[MAX_OPT_ARGS];
1206         char                    *p;
1207         int                     error;
1208
1209         /* First, check for complete junk; i.e. invalid options */
1210         error = xfs_test_remount_options(sb, mp, options);
1211         if (error)
1212                 return error;
1213
1214         sync_filesystem(sb);
1215         while ((p = strsep(&options, ",")) != NULL) {
1216                 int token;
1217
1218                 if (!*p)
1219                         continue;
1220
1221                 token = match_token(p, tokens, args);
1222                 switch (token) {
1223                 case Opt_barrier:
1224                         mp->m_flags |= XFS_MOUNT_BARRIER;
1225                         break;
1226                 case Opt_nobarrier:
1227                         mp->m_flags &= ~XFS_MOUNT_BARRIER;
1228                         break;
1229                 case Opt_inode64:
1230                         mp->m_maxagi = xfs_set_inode64(mp, sbp->sb_agcount);
1231                         break;
1232                 case Opt_inode32:
1233                         mp->m_maxagi = xfs_set_inode32(mp, sbp->sb_agcount);
1234                         break;
1235                 default:
1236                         /*
1237                          * Logically we would return an error here to prevent
1238                          * users from believing they might have changed
1239                          * mount options using remount which can't be changed.
1240                          *
1241                          * But unfortunately mount(8) adds all options from
1242                          * mtab and fstab to the mount arguments in some cases
1243                          * so we can't blindly reject options, but have to
1244                          * check for each specified option if it actually
1245                          * differs from the currently set option and only
1246                          * reject it if that's the case.
1247                          *
1248                          * Until that is implemented we return success for
1249                          * every remount request, and silently ignore all
1250                          * options that we can't actually change.
1251                          */
1252 #if 0
1253                         xfs_info(mp,
1254                 "mount option \"%s\" not supported for remount", p);
1255                         return -EINVAL;
1256 #else
1257                         break;
1258 #endif
1259                 }
1260         }
1261
1262         /* ro -> rw */
1263         if ((mp->m_flags & XFS_MOUNT_RDONLY) && !(*flags & MS_RDONLY)) {
1264                 if (mp->m_flags & XFS_MOUNT_NORECOVERY) {
1265                         xfs_warn(mp,
1266                 "ro->rw transition prohibited on norecovery mount");
1267                         return -EINVAL;
1268                 }
1269
1270                 mp->m_flags &= ~XFS_MOUNT_RDONLY;
1271
1272                 /*
1273                  * If this is the first remount to writeable state we
1274                  * might have some superblock changes to update.
1275                  */
1276                 if (mp->m_update_sb) {
1277                         error = xfs_sync_sb(mp, false);
1278                         if (error) {
1279                                 xfs_warn(mp, "failed to write sb changes");
1280                                 return error;
1281                         }
1282                         mp->m_update_sb = false;
1283                 }
1284
1285                 /*
1286                  * Fill out the reserve pool if it is empty. Use the stashed
1287                  * value if it is non-zero, otherwise go with the default.
1288                  */
1289                 xfs_restore_resvblks(mp);
1290                 xfs_log_work_queue(mp);
1291         }
1292
1293         /* rw -> ro */
1294         if (!(mp->m_flags & XFS_MOUNT_RDONLY) && (*flags & MS_RDONLY)) {
1295                 /*
1296                  * Before we sync the metadata, we need to free up the reserve
1297                  * block pool so that the used block count in the superblock on
1298                  * disk is correct at the end of the remount. Stash the current
1299                  * reserve pool size so that if we get remounted rw, we can
1300                  * return it to the same size.
1301                  */
1302                 xfs_save_resvblks(mp);
1303                 xfs_quiesce_attr(mp);
1304                 mp->m_flags |= XFS_MOUNT_RDONLY;
1305         }
1306
1307         return 0;
1308 }
1309
1310 /*
1311  * Second stage of a freeze. The data is already frozen so we only
1312  * need to take care of the metadata. Once that's done sync the superblock
1313  * to the log to dirty it in case of a crash while frozen. This ensures that we
1314  * will recover the unlinked inode lists on the next mount.
1315  */
1316 STATIC int
1317 xfs_fs_freeze(
1318         struct super_block      *sb)
1319 {
1320         struct xfs_mount        *mp = XFS_M(sb);
1321
1322         xfs_save_resvblks(mp);
1323         xfs_quiesce_attr(mp);
1324         return xfs_sync_sb(mp, true);
1325 }
1326
1327 STATIC int
1328 xfs_fs_unfreeze(
1329         struct super_block      *sb)
1330 {
1331         struct xfs_mount        *mp = XFS_M(sb);
1332
1333         xfs_restore_resvblks(mp);
1334         xfs_log_work_queue(mp);
1335         return 0;
1336 }
1337
1338 STATIC int
1339 xfs_fs_show_options(
1340         struct seq_file         *m,
1341         struct dentry           *root)
1342 {
1343         return xfs_showargs(XFS_M(root->d_sb), m);
1344 }
1345
1346 /*
1347  * This function fills in xfs_mount_t fields based on mount args.
1348  * Note: the superblock _has_ now been read in.
1349  */
1350 STATIC int
1351 xfs_finish_flags(
1352         struct xfs_mount        *mp)
1353 {
1354         int                     ronly = (mp->m_flags & XFS_MOUNT_RDONLY);
1355
1356         /* Fail a mount where the logbuf is smaller than the log stripe */
1357         if (xfs_sb_version_haslogv2(&mp->m_sb)) {
1358                 if (mp->m_logbsize <= 0 &&
1359                     mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE) {
1360                         mp->m_logbsize = mp->m_sb.sb_logsunit;
1361                 } else if (mp->m_logbsize > 0 &&
1362                            mp->m_logbsize < mp->m_sb.sb_logsunit) {
1363                         xfs_warn(mp,
1364                 "logbuf size must be greater than or equal to log stripe size");
1365                         return -EINVAL;
1366                 }
1367         } else {
1368                 /* Fail a mount if the logbuf is larger than 32K */
1369                 if (mp->m_logbsize > XLOG_BIG_RECORD_BSIZE) {
1370                         xfs_warn(mp,
1371                 "logbuf size for version 1 logs must be 16K or 32K");
1372                         return -EINVAL;
1373                 }
1374         }
1375
1376         /*
1377          * V5 filesystems always use attr2 format for attributes.
1378          */
1379         if (xfs_sb_version_hascrc(&mp->m_sb) &&
1380             (mp->m_flags & XFS_MOUNT_NOATTR2)) {
1381                 xfs_warn(mp, "Cannot mount a V5 filesystem as noattr2. "
1382                              "attr2 is always enabled for V5 filesystems.");
1383                 return -EINVAL;
1384         }
1385
1386         /*
1387          * mkfs'ed attr2 will turn on attr2 mount unless explicitly
1388          * told by noattr2 to turn it off
1389          */
1390         if (xfs_sb_version_hasattr2(&mp->m_sb) &&
1391             !(mp->m_flags & XFS_MOUNT_NOATTR2))
1392                 mp->m_flags |= XFS_MOUNT_ATTR2;
1393
1394         /*
1395          * prohibit r/w mounts of read-only filesystems
1396          */
1397         if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !ronly) {
1398                 xfs_warn(mp,
1399                         "cannot mount a read-only filesystem as read-write");
1400                 return -EROFS;
1401         }
1402
1403         if ((mp->m_qflags & (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE)) &&
1404             (mp->m_qflags & (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE)) &&
1405             !xfs_sb_version_has_pquotino(&mp->m_sb)) {
1406                 xfs_warn(mp,
1407                   "Super block does not support project and group quota together");
1408                 return -EINVAL;
1409         }
1410
1411         return 0;
1412 }
1413
1414 static int
1415 xfs_init_percpu_counters(
1416         struct xfs_mount        *mp)
1417 {
1418         int             error;
1419
1420         error = percpu_counter_init(&mp->m_icount, 0, GFP_KERNEL);
1421         if (error)
1422                 return -ENOMEM;
1423
1424         error = percpu_counter_init(&mp->m_ifree, 0, GFP_KERNEL);
1425         if (error)
1426                 goto free_icount;
1427
1428         error = percpu_counter_init(&mp->m_fdblocks, 0, GFP_KERNEL);
1429         if (error)
1430                 goto free_ifree;
1431
1432         return 0;
1433
1434 free_ifree:
1435         percpu_counter_destroy(&mp->m_ifree);
1436 free_icount:
1437         percpu_counter_destroy(&mp->m_icount);
1438         return -ENOMEM;
1439 }
1440
1441 void
1442 xfs_reinit_percpu_counters(
1443         struct xfs_mount        *mp)
1444 {
1445         percpu_counter_set(&mp->m_icount, mp->m_sb.sb_icount);
1446         percpu_counter_set(&mp->m_ifree, mp->m_sb.sb_ifree);
1447         percpu_counter_set(&mp->m_fdblocks, mp->m_sb.sb_fdblocks);
1448 }
1449
1450 static void
1451 xfs_destroy_percpu_counters(
1452         struct xfs_mount        *mp)
1453 {
1454         percpu_counter_destroy(&mp->m_icount);
1455         percpu_counter_destroy(&mp->m_ifree);
1456         percpu_counter_destroy(&mp->m_fdblocks);
1457 }
1458
1459 STATIC int
1460 xfs_fs_fill_super(
1461         struct super_block      *sb,
1462         void                    *data,
1463         int                     silent)
1464 {
1465         struct inode            *root;
1466         struct xfs_mount        *mp = NULL;
1467         int                     flags = 0, error = -ENOMEM;
1468
1469         mp = kzalloc(sizeof(struct xfs_mount), GFP_KERNEL);
1470         if (!mp)
1471                 goto out;
1472
1473         spin_lock_init(&mp->m_sb_lock);
1474         mutex_init(&mp->m_growlock);
1475         atomic_set(&mp->m_active_trans, 0);
1476         INIT_DELAYED_WORK(&mp->m_reclaim_work, xfs_reclaim_worker);
1477         INIT_DELAYED_WORK(&mp->m_eofblocks_work, xfs_eofblocks_worker);
1478         mp->m_kobj.kobject.kset = xfs_kset;
1479
1480         mp->m_super = sb;
1481         sb->s_fs_info = mp;
1482
1483         error = xfs_parseargs(mp, (char *)data);
1484         if (error)
1485                 goto out_free_fsname;
1486
1487         sb_min_blocksize(sb, BBSIZE);
1488         sb->s_xattr = xfs_xattr_handlers;
1489         sb->s_export_op = &xfs_export_operations;
1490 #ifdef CONFIG_XFS_QUOTA
1491         sb->s_qcop = &xfs_quotactl_operations;
1492         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
1493 #endif
1494         sb->s_op = &xfs_super_operations;
1495
1496         if (silent)
1497                 flags |= XFS_MFSI_QUIET;
1498
1499         error = xfs_open_devices(mp);
1500         if (error)
1501                 goto out_free_fsname;
1502
1503         error = xfs_init_mount_workqueues(mp);
1504         if (error)
1505                 goto out_close_devices;
1506
1507         error = xfs_init_percpu_counters(mp);
1508         if (error)
1509                 goto out_destroy_workqueues;
1510
1511         /* Allocate stats memory before we do operations that might use it */
1512         mp->m_stats.xs_stats = alloc_percpu(struct xfsstats);
1513         if (!mp->m_stats.xs_stats) {
1514                 error = -ENOMEM;
1515                 goto out_destroy_counters;
1516         }
1517
1518         error = xfs_readsb(mp, flags);
1519         if (error)
1520                 goto out_free_stats;
1521
1522         error = xfs_finish_flags(mp);
1523         if (error)
1524                 goto out_free_sb;
1525
1526         error = xfs_setup_devices(mp);
1527         if (error)
1528                 goto out_free_sb;
1529
1530         error = xfs_filestream_mount(mp);
1531         if (error)
1532                 goto out_free_sb;
1533
1534         /*
1535          * we must configure the block size in the superblock before we run the
1536          * full mount process as the mount process can lookup and cache inodes.
1537          */
1538         sb->s_magic = XFS_SB_MAGIC;
1539         sb->s_blocksize = mp->m_sb.sb_blocksize;
1540         sb->s_blocksize_bits = ffs(sb->s_blocksize) - 1;
1541         sb->s_maxbytes = xfs_max_file_offset(sb->s_blocksize_bits);
1542         sb->s_max_links = XFS_MAXLINK;
1543         sb->s_time_gran = 1;
1544         set_posix_acl_flag(sb);
1545
1546         /* version 5 superblocks support inode version counters. */
1547         if (XFS_SB_VERSION_NUM(&mp->m_sb) == XFS_SB_VERSION_5)
1548                 sb->s_flags |= MS_I_VERSION;
1549
1550         if (mp->m_flags & XFS_MOUNT_DAX) {
1551                 xfs_warn(mp,
1552         "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
1553                 if (sb->s_blocksize != PAGE_SIZE) {
1554                         xfs_alert(mp,
1555                 "Filesystem block size invalid for DAX Turning DAX off.");
1556                         mp->m_flags &= ~XFS_MOUNT_DAX;
1557                 } else if (!sb->s_bdev->bd_disk->fops->direct_access) {
1558                         xfs_alert(mp,
1559                 "Block device does not support DAX Turning DAX off.");
1560                         mp->m_flags &= ~XFS_MOUNT_DAX;
1561                 }
1562         }
1563
1564         if (xfs_sb_version_hassparseinodes(&mp->m_sb))
1565                 xfs_alert(mp,
1566         "EXPERIMENTAL sparse inode feature enabled. Use at your own risk!");
1567
1568         error = xfs_mountfs(mp);
1569         if (error)
1570                 goto out_filestream_unmount;
1571
1572         root = igrab(VFS_I(mp->m_rootip));
1573         if (!root) {
1574                 error = -ENOENT;
1575                 goto out_unmount;
1576         }
1577         sb->s_root = d_make_root(root);
1578         if (!sb->s_root) {
1579                 error = -ENOMEM;
1580                 goto out_unmount;
1581         }
1582
1583         return 0;
1584
1585  out_filestream_unmount:
1586         xfs_filestream_unmount(mp);
1587  out_free_sb:
1588         xfs_freesb(mp);
1589  out_free_stats:
1590         free_percpu(mp->m_stats.xs_stats);
1591  out_destroy_counters:
1592         xfs_destroy_percpu_counters(mp);
1593  out_destroy_workqueues:
1594         xfs_destroy_mount_workqueues(mp);
1595  out_close_devices:
1596         xfs_close_devices(mp);
1597  out_free_fsname:
1598         xfs_free_fsname(mp);
1599         kfree(mp);
1600  out:
1601         return error;
1602
1603  out_unmount:
1604         xfs_filestream_unmount(mp);
1605         xfs_unmountfs(mp);
1606         goto out_free_sb;
1607 }
1608
1609 STATIC void
1610 xfs_fs_put_super(
1611         struct super_block      *sb)
1612 {
1613         struct xfs_mount        *mp = XFS_M(sb);
1614
1615         xfs_notice(mp, "Unmounting Filesystem");
1616         xfs_filestream_unmount(mp);
1617         xfs_unmountfs(mp);
1618
1619         xfs_freesb(mp);
1620         free_percpu(mp->m_stats.xs_stats);
1621         xfs_destroy_percpu_counters(mp);
1622         xfs_destroy_mount_workqueues(mp);
1623         xfs_close_devices(mp);
1624         xfs_free_fsname(mp);
1625         kfree(mp);
1626 }
1627
1628 STATIC struct dentry *
1629 xfs_fs_mount(
1630         struct file_system_type *fs_type,
1631         int                     flags,
1632         const char              *dev_name,
1633         void                    *data)
1634 {
1635         return mount_bdev(fs_type, flags, dev_name, data, xfs_fs_fill_super);
1636 }
1637
1638 static long
1639 xfs_fs_nr_cached_objects(
1640         struct super_block      *sb,
1641         struct shrink_control   *sc)
1642 {
1643         return xfs_reclaim_inodes_count(XFS_M(sb));
1644 }
1645
1646 static long
1647 xfs_fs_free_cached_objects(
1648         struct super_block      *sb,
1649         struct shrink_control   *sc)
1650 {
1651         return xfs_reclaim_inodes_nr(XFS_M(sb), sc->nr_to_scan);
1652 }
1653
1654 static const struct super_operations xfs_super_operations = {
1655         .alloc_inode            = xfs_fs_alloc_inode,
1656         .destroy_inode          = xfs_fs_destroy_inode,
1657         .evict_inode            = xfs_fs_evict_inode,
1658         .drop_inode             = xfs_fs_drop_inode,
1659         .put_super              = xfs_fs_put_super,
1660         .sync_fs                = xfs_fs_sync_fs,
1661         .freeze_fs              = xfs_fs_freeze,
1662         .unfreeze_fs            = xfs_fs_unfreeze,
1663         .statfs                 = xfs_fs_statfs,
1664         .remount_fs             = xfs_fs_remount,
1665         .show_options           = xfs_fs_show_options,
1666         .nr_cached_objects      = xfs_fs_nr_cached_objects,
1667         .free_cached_objects    = xfs_fs_free_cached_objects,
1668 };
1669
1670 static struct file_system_type xfs_fs_type = {
1671         .owner                  = THIS_MODULE,
1672         .name                   = "xfs",
1673         .mount                  = xfs_fs_mount,
1674         .kill_sb                = kill_block_super,
1675         .fs_flags               = FS_REQUIRES_DEV,
1676 };
1677 MODULE_ALIAS_FS("xfs");
1678
1679 STATIC int __init
1680 xfs_init_zones(void)
1681 {
1682
1683         xfs_ioend_zone = kmem_zone_init(sizeof(xfs_ioend_t), "xfs_ioend");
1684         if (!xfs_ioend_zone)
1685                 goto out;
1686
1687         xfs_ioend_pool = mempool_create_slab_pool(4 * MAX_BUF_PER_PAGE,
1688                                                   xfs_ioend_zone);
1689         if (!xfs_ioend_pool)
1690                 goto out_destroy_ioend_zone;
1691
1692         xfs_log_ticket_zone = kmem_zone_init(sizeof(xlog_ticket_t),
1693                                                 "xfs_log_ticket");
1694         if (!xfs_log_ticket_zone)
1695                 goto out_destroy_ioend_pool;
1696
1697         xfs_bmap_free_item_zone = kmem_zone_init(sizeof(xfs_bmap_free_item_t),
1698                                                 "xfs_bmap_free_item");
1699         if (!xfs_bmap_free_item_zone)
1700                 goto out_destroy_log_ticket_zone;
1701
1702         xfs_btree_cur_zone = kmem_zone_init(sizeof(xfs_btree_cur_t),
1703                                                 "xfs_btree_cur");
1704         if (!xfs_btree_cur_zone)
1705                 goto out_destroy_bmap_free_item_zone;
1706
1707         xfs_da_state_zone = kmem_zone_init(sizeof(xfs_da_state_t),
1708                                                 "xfs_da_state");
1709         if (!xfs_da_state_zone)
1710                 goto out_destroy_btree_cur_zone;
1711
1712         xfs_ifork_zone = kmem_zone_init(sizeof(xfs_ifork_t), "xfs_ifork");
1713         if (!xfs_ifork_zone)
1714                 goto out_destroy_da_state_zone;
1715
1716         xfs_trans_zone = kmem_zone_init(sizeof(xfs_trans_t), "xfs_trans");
1717         if (!xfs_trans_zone)
1718                 goto out_destroy_ifork_zone;
1719
1720         xfs_log_item_desc_zone =
1721                 kmem_zone_init(sizeof(struct xfs_log_item_desc),
1722                                "xfs_log_item_desc");
1723         if (!xfs_log_item_desc_zone)
1724                 goto out_destroy_trans_zone;
1725
1726         /*
1727          * The size of the zone allocated buf log item is the maximum
1728          * size possible under XFS.  This wastes a little bit of memory,
1729          * but it is much faster.
1730          */
1731         xfs_buf_item_zone = kmem_zone_init(sizeof(struct xfs_buf_log_item),
1732                                            "xfs_buf_item");
1733         if (!xfs_buf_item_zone)
1734                 goto out_destroy_log_item_desc_zone;
1735
1736         xfs_efd_zone = kmem_zone_init((sizeof(xfs_efd_log_item_t) +
1737                         ((XFS_EFD_MAX_FAST_EXTENTS - 1) *
1738                                  sizeof(xfs_extent_t))), "xfs_efd_item");
1739         if (!xfs_efd_zone)
1740                 goto out_destroy_buf_item_zone;
1741
1742         xfs_efi_zone = kmem_zone_init((sizeof(xfs_efi_log_item_t) +
1743                         ((XFS_EFI_MAX_FAST_EXTENTS - 1) *
1744                                 sizeof(xfs_extent_t))), "xfs_efi_item");
1745         if (!xfs_efi_zone)
1746                 goto out_destroy_efd_zone;
1747
1748         xfs_inode_zone =
1749                 kmem_zone_init_flags(sizeof(xfs_inode_t), "xfs_inode",
1750                         KM_ZONE_HWALIGN | KM_ZONE_RECLAIM | KM_ZONE_SPREAD |
1751                         KM_ZONE_ACCOUNT, xfs_fs_inode_init_once);
1752         if (!xfs_inode_zone)
1753                 goto out_destroy_efi_zone;
1754
1755         xfs_ili_zone =
1756                 kmem_zone_init_flags(sizeof(xfs_inode_log_item_t), "xfs_ili",
1757                                         KM_ZONE_SPREAD, NULL);
1758         if (!xfs_ili_zone)
1759                 goto out_destroy_inode_zone;
1760         xfs_icreate_zone = kmem_zone_init(sizeof(struct xfs_icreate_item),
1761                                         "xfs_icr");
1762         if (!xfs_icreate_zone)
1763                 goto out_destroy_ili_zone;
1764
1765         return 0;
1766
1767  out_destroy_ili_zone:
1768         kmem_zone_destroy(xfs_ili_zone);
1769  out_destroy_inode_zone:
1770         kmem_zone_destroy(xfs_inode_zone);
1771  out_destroy_efi_zone:
1772         kmem_zone_destroy(xfs_efi_zone);
1773  out_destroy_efd_zone:
1774         kmem_zone_destroy(xfs_efd_zone);
1775  out_destroy_buf_item_zone:
1776         kmem_zone_destroy(xfs_buf_item_zone);
1777  out_destroy_log_item_desc_zone:
1778         kmem_zone_destroy(xfs_log_item_desc_zone);
1779  out_destroy_trans_zone:
1780         kmem_zone_destroy(xfs_trans_zone);
1781  out_destroy_ifork_zone:
1782         kmem_zone_destroy(xfs_ifork_zone);
1783  out_destroy_da_state_zone:
1784         kmem_zone_destroy(xfs_da_state_zone);
1785  out_destroy_btree_cur_zone:
1786         kmem_zone_destroy(xfs_btree_cur_zone);
1787  out_destroy_bmap_free_item_zone:
1788         kmem_zone_destroy(xfs_bmap_free_item_zone);
1789  out_destroy_log_ticket_zone:
1790         kmem_zone_destroy(xfs_log_ticket_zone);
1791  out_destroy_ioend_pool:
1792         mempool_destroy(xfs_ioend_pool);
1793  out_destroy_ioend_zone:
1794         kmem_zone_destroy(xfs_ioend_zone);
1795  out:
1796         return -ENOMEM;
1797 }
1798
1799 STATIC void
1800 xfs_destroy_zones(void)
1801 {
1802         /*
1803          * Make sure all delayed rcu free are flushed before we
1804          * destroy caches.
1805          */
1806         rcu_barrier();
1807         kmem_zone_destroy(xfs_icreate_zone);
1808         kmem_zone_destroy(xfs_ili_zone);
1809         kmem_zone_destroy(xfs_inode_zone);
1810         kmem_zone_destroy(xfs_efi_zone);
1811         kmem_zone_destroy(xfs_efd_zone);
1812         kmem_zone_destroy(xfs_buf_item_zone);
1813         kmem_zone_destroy(xfs_log_item_desc_zone);
1814         kmem_zone_destroy(xfs_trans_zone);
1815         kmem_zone_destroy(xfs_ifork_zone);
1816         kmem_zone_destroy(xfs_da_state_zone);
1817         kmem_zone_destroy(xfs_btree_cur_zone);
1818         kmem_zone_destroy(xfs_bmap_free_item_zone);
1819         kmem_zone_destroy(xfs_log_ticket_zone);
1820         mempool_destroy(xfs_ioend_pool);
1821         kmem_zone_destroy(xfs_ioend_zone);
1822
1823 }
1824
1825 STATIC int __init
1826 xfs_init_workqueues(void)
1827 {
1828         /*
1829          * The allocation workqueue can be used in memory reclaim situations
1830          * (writepage path), and parallelism is only limited by the number of
1831          * AGs in all the filesystems mounted. Hence use the default large
1832          * max_active value for this workqueue.
1833          */
1834         xfs_alloc_wq = alloc_workqueue("xfsalloc",
1835                         WQ_MEM_RECLAIM|WQ_FREEZABLE, 0);
1836         if (!xfs_alloc_wq)
1837                 return -ENOMEM;
1838
1839         return 0;
1840 }
1841
1842 STATIC void
1843 xfs_destroy_workqueues(void)
1844 {
1845         destroy_workqueue(xfs_alloc_wq);
1846 }
1847
1848 STATIC int __init
1849 init_xfs_fs(void)
1850 {
1851         int                     error;
1852
1853         printk(KERN_INFO XFS_VERSION_STRING " with "
1854                          XFS_BUILD_OPTIONS " enabled\n");
1855
1856         xfs_dir_startup();
1857
1858         error = xfs_init_zones();
1859         if (error)
1860                 goto out;
1861
1862         error = xfs_init_workqueues();
1863         if (error)
1864                 goto out_destroy_zones;
1865
1866         error = xfs_mru_cache_init();
1867         if (error)
1868                 goto out_destroy_wq;
1869
1870         error = xfs_buf_init();
1871         if (error)
1872                 goto out_mru_cache_uninit;
1873
1874         error = xfs_init_procfs();
1875         if (error)
1876                 goto out_buf_terminate;
1877
1878         error = xfs_sysctl_register();
1879         if (error)
1880                 goto out_cleanup_procfs;
1881
1882         xfs_kset = kset_create_and_add("xfs", NULL, fs_kobj);
1883         if (!xfs_kset) {
1884                 error = -ENOMEM;
1885                 goto out_sysctl_unregister;
1886         }
1887
1888         xfsstats.xs_kobj.kobject.kset = xfs_kset;
1889
1890         xfsstats.xs_stats = alloc_percpu(struct xfsstats);
1891         if (!xfsstats.xs_stats) {
1892                 error = -ENOMEM;
1893                 goto out_kset_unregister;
1894         }
1895
1896         error = xfs_sysfs_init(&xfsstats.xs_kobj, &xfs_stats_ktype, NULL,
1897                                "stats");
1898         if (error)
1899                 goto out_free_stats;
1900
1901 #ifdef DEBUG
1902         xfs_dbg_kobj.kobject.kset = xfs_kset;
1903         error = xfs_sysfs_init(&xfs_dbg_kobj, &xfs_dbg_ktype, NULL, "debug");
1904         if (error)
1905                 goto out_remove_stats_kobj;
1906 #endif
1907
1908         error = xfs_qm_init();
1909         if (error)
1910                 goto out_remove_dbg_kobj;
1911
1912         error = register_filesystem(&xfs_fs_type);
1913         if (error)
1914                 goto out_qm_exit;
1915         return 0;
1916
1917  out_qm_exit:
1918         xfs_qm_exit();
1919  out_remove_dbg_kobj:
1920 #ifdef DEBUG
1921         xfs_sysfs_del(&xfs_dbg_kobj);
1922  out_remove_stats_kobj:
1923 #endif
1924         xfs_sysfs_del(&xfsstats.xs_kobj);
1925  out_free_stats:
1926         free_percpu(xfsstats.xs_stats);
1927  out_kset_unregister:
1928         kset_unregister(xfs_kset);
1929  out_sysctl_unregister:
1930         xfs_sysctl_unregister();
1931  out_cleanup_procfs:
1932         xfs_cleanup_procfs();
1933  out_buf_terminate:
1934         xfs_buf_terminate();
1935  out_mru_cache_uninit:
1936         xfs_mru_cache_uninit();
1937  out_destroy_wq:
1938         xfs_destroy_workqueues();
1939  out_destroy_zones:
1940         xfs_destroy_zones();
1941  out:
1942         return error;
1943 }
1944
1945 STATIC void __exit
1946 exit_xfs_fs(void)
1947 {
1948         xfs_qm_exit();
1949         unregister_filesystem(&xfs_fs_type);
1950 #ifdef DEBUG
1951         xfs_sysfs_del(&xfs_dbg_kobj);
1952 #endif
1953         xfs_sysfs_del(&xfsstats.xs_kobj);
1954         free_percpu(xfsstats.xs_stats);
1955         kset_unregister(xfs_kset);
1956         xfs_sysctl_unregister();
1957         xfs_cleanup_procfs();
1958         xfs_buf_terminate();
1959         xfs_mru_cache_uninit();
1960         xfs_destroy_workqueues();
1961         xfs_destroy_zones();
1962         xfs_uuid_table_free();
1963 }
1964
1965 module_init(init_xfs_fs);
1966 module_exit(exit_xfs_fs);
1967
1968 MODULE_AUTHOR("Silicon Graphics, Inc.");
1969 MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled");
1970 MODULE_LICENSE("GPL");