fs/xfs/xfs_vfsops.c

   1 /*
   2  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
   3  * All Rights Reserved.
   4  *
   5  * This program is free software; you can redistribute it and/or
   6  * modify it under the terms of the GNU General Public License as
   7  * published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it would be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write the Free Software Foundation,
  16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17  */
  18 #include "xfs.h"
  19 #include "xfs_fs.h"
  20 #include "xfs_types.h"
  21 #include "xfs_bit.h"
  22 #include "xfs_log.h"
  23 #include "xfs_inum.h"
  24 #include "xfs_trans.h"
  25 #include "xfs_sb.h"
  26 #include "xfs_ag.h"
  27 #include "xfs_dir.h"
  28 #include "xfs_dir2.h"
  29 #include "xfs_dmapi.h"
  30 #include "xfs_mount.h"
  31 #include "xfs_da_btree.h"
  32 #include "xfs_bmap_btree.h"
  33 #include "xfs_ialloc_btree.h"
  34 #include "xfs_alloc_btree.h"
  35 #include "xfs_dir_sf.h"
  36 #include "xfs_dir2_sf.h"
  37 #include "xfs_attr_sf.h"
  38 #include "xfs_dinode.h"
  39 #include "xfs_inode.h"
  40 #include "xfs_inode_item.h"
  41 #include "xfs_btree.h"
  42 #include "xfs_alloc.h"
  43 #include "xfs_ialloc.h"
  44 #include "xfs_quota.h"
  45 #include "xfs_error.h"
  46 #include "xfs_bmap.h"
  47 #include "xfs_rw.h"
  48 #include "xfs_refcache.h"
  49 #include "xfs_buf_item.h"
  50 #include "xfs_log_priv.h"
  51 #include "xfs_dir2_trace.h"
  52 #include "xfs_extfree_item.h"
  53 #include "xfs_acl.h"
  54 #include "xfs_attr.h"
  55 #include "xfs_clnt.h"
  56
  57 STATIC int xfs_sync(bhv_desc_t *, int, cred_t *);
  58
  59 int
  60 xfs_init(void)
  61 {
  62         extern kmem_zone_t      *xfs_bmap_free_item_zone;
  63         extern kmem_zone_t      *xfs_btree_cur_zone;
  64         extern kmem_zone_t      *xfs_trans_zone;
  65         extern kmem_zone_t      *xfs_buf_item_zone;
  66         extern kmem_zone_t      *xfs_dabuf_zone;
  67 #ifdef XFS_DABUF_DEBUG
  68         extern lock_t           xfs_dabuf_global_lock;
  69         spinlock_init(&xfs_dabuf_global_lock, "xfsda");
  70 #endif
  71
  72         /*
  73          * Initialize all of the zone allocators we use.
  74          */
  75         xfs_bmap_free_item_zone = kmem_zone_init(sizeof(xfs_bmap_free_item_t),
  76                                                  "xfs_bmap_free_item");
  77         xfs_btree_cur_zone = kmem_zone_init(sizeof(xfs_btree_cur_t),
  78                                             "xfs_btree_cur");
  79         xfs_inode_zone = kmem_zone_init(sizeof(xfs_inode_t), "xfs_inode");
  80         xfs_trans_zone = kmem_zone_init(sizeof(xfs_trans_t), "xfs_trans");
  81         xfs_da_state_zone =
  82                 kmem_zone_init(sizeof(xfs_da_state_t), "xfs_da_state");
  83         xfs_dabuf_zone = kmem_zone_init(sizeof(xfs_dabuf_t), "xfs_dabuf");
  84
  85         /*
  86          * The size of the zone allocated buf log item is the maximum
  87          * size possible under XFS.  This wastes a little bit of memory,
  88          * but it is much faster.
  89          */
  90         xfs_buf_item_zone =
  91                 kmem_zone_init((sizeof(xfs_buf_log_item_t) +
  92                                 (((XFS_MAX_BLOCKSIZE / XFS_BLI_CHUNK) /
  93                                   NBWORD) * sizeof(int))),
  94                                "xfs_buf_item");
  95         xfs_efd_zone = kmem_zone_init((sizeof(xfs_efd_log_item_t) +
  96                                        ((XFS_EFD_MAX_FAST_EXTENTS - 1) * sizeof(xfs_extent_t))),
  97                                       "xfs_efd_item");
  98         xfs_efi_zone = kmem_zone_init((sizeof(xfs_efi_log_item_t) +
  99                                        ((XFS_EFI_MAX_FAST_EXTENTS - 1) * sizeof(xfs_extent_t))),
 100                                       "xfs_efi_item");
 101         xfs_ifork_zone = kmem_zone_init(sizeof(xfs_ifork_t), "xfs_ifork");
 102         xfs_ili_zone = kmem_zone_init(sizeof(xfs_inode_log_item_t), "xfs_ili");
 103         xfs_chashlist_zone = kmem_zone_init(sizeof(xfs_chashlist_t),
 104                                             "xfs_chashlist");
 105         xfs_acl_zone_init(xfs_acl_zone, "xfs_acl");
 106
 107         /*
 108          * Allocate global trace buffers.
 109          */
 110 #ifdef XFS_ALLOC_TRACE
 111         xfs_alloc_trace_buf = ktrace_alloc(XFS_ALLOC_TRACE_SIZE, KM_SLEEP);
 112 #endif
 113 #ifdef XFS_BMAP_TRACE
 114         xfs_bmap_trace_buf = ktrace_alloc(XFS_BMAP_TRACE_SIZE, KM_SLEEP);
 115 #endif
 116 #ifdef XFS_BMBT_TRACE
 117         xfs_bmbt_trace_buf = ktrace_alloc(XFS_BMBT_TRACE_SIZE, KM_SLEEP);
 118 #endif
 119 #ifdef XFS_DIR_TRACE
 120         xfs_dir_trace_buf = ktrace_alloc(XFS_DIR_TRACE_SIZE, KM_SLEEP);
 121 #endif
 122 #ifdef XFS_ATTR_TRACE
 123         xfs_attr_trace_buf = ktrace_alloc(XFS_ATTR_TRACE_SIZE, KM_SLEEP);
 124 #endif
 125 #ifdef XFS_DIR2_TRACE
 126         xfs_dir2_trace_buf = ktrace_alloc(XFS_DIR2_GTRACE_SIZE, KM_SLEEP);
 127 #endif
 128
 129         xfs_dir_startup();
 130
 131 #if (defined(DEBUG) || defined(INDUCE_IO_ERROR))
 132         xfs_error_test_init();
 133 #endif /* DEBUG || INDUCE_IO_ERROR */
 134
 135         xfs_init_procfs();
 136         xfs_sysctl_register();
 137         return 0;
 138 }
 139
 140 void
 141 xfs_cleanup(void)
 142 {
 143         extern kmem_zone_t      *xfs_bmap_free_item_zone;
 144         extern kmem_zone_t      *xfs_btree_cur_zone;
 145         extern kmem_zone_t      *xfs_inode_zone;
 146         extern kmem_zone_t      *xfs_trans_zone;
 147         extern kmem_zone_t      *xfs_da_state_zone;
 148         extern kmem_zone_t      *xfs_dabuf_zone;
 149         extern kmem_zone_t      *xfs_efd_zone;
 150         extern kmem_zone_t      *xfs_efi_zone;
 151         extern kmem_zone_t      *xfs_buf_item_zone;
 152         extern kmem_zone_t      *xfs_chashlist_zone;
 153
 154         xfs_cleanup_procfs();
 155         xfs_sysctl_unregister();
 156         xfs_refcache_destroy();
 157         xfs_acl_zone_destroy(xfs_acl_zone);
 158
 159 #ifdef XFS_DIR2_TRACE
 160         ktrace_free(xfs_dir2_trace_buf);
 161 #endif
 162 #ifdef XFS_ATTR_TRACE
 163         ktrace_free(xfs_attr_trace_buf);
 164 #endif
 165 #ifdef XFS_DIR_TRACE
 166         ktrace_free(xfs_dir_trace_buf);
 167 #endif
 168 #ifdef XFS_BMBT_TRACE
 169         ktrace_free(xfs_bmbt_trace_buf);
 170 #endif
 171 #ifdef XFS_BMAP_TRACE
 172         ktrace_free(xfs_bmap_trace_buf);
 173 #endif
 174 #ifdef XFS_ALLOC_TRACE
 175         ktrace_free(xfs_alloc_trace_buf);
 176 #endif
 177
 178         kmem_cache_destroy(xfs_bmap_free_item_zone);
 179         kmem_cache_destroy(xfs_btree_cur_zone);
 180         kmem_cache_destroy(xfs_inode_zone);
 181         kmem_cache_destroy(xfs_trans_zone);
 182         kmem_cache_destroy(xfs_da_state_zone);
 183         kmem_cache_destroy(xfs_dabuf_zone);
 184         kmem_cache_destroy(xfs_buf_item_zone);
 185         kmem_cache_destroy(xfs_efd_zone);
 186         kmem_cache_destroy(xfs_efi_zone);
 187         kmem_cache_destroy(xfs_ifork_zone);
 188         kmem_cache_destroy(xfs_ili_zone);
 189         kmem_cache_destroy(xfs_chashlist_zone);
 190 }
 191
 192 /*
 193  * xfs_start_flags
 194  *
 195  * This function fills in xfs_mount_t fields based on mount args.
 196  * Note: the superblock has _not_ yet been read in.
 197  */
 198 STATIC int
 199 xfs_start_flags(
 200         struct vfs              *vfs,
 201         struct xfs_mount_args   *ap,
 202         struct xfs_mount        *mp)
 203 {
 204         /* Values are in BBs */
 205         if ((ap->flags & XFSMNT_NOALIGN) != XFSMNT_NOALIGN) {
 206                 /*
 207                  * At this point the superblock has not been read
 208                  * in, therefore we do not know the block size.
 209                  * Before the mount call ends we will convert
 210                  * these to FSBs.
 211                  */
 212                 mp->m_dalign = ap->sunit;
 213                 mp->m_swidth = ap->swidth;
 214         }
 215
 216         if (ap->logbufs != -1 &&
 217 #if defined(DEBUG) || defined(XLOG_NOLOG)
 218             ap->logbufs != 0 &&
 219 #endif
 220             (ap->logbufs < XLOG_MIN_ICLOGS ||
 221              ap->logbufs > XLOG_MAX_ICLOGS)) {
 222                 cmn_err(CE_WARN,
 223                         "XFS: invalid logbufs value: %d [not %d-%d]",
 224                         ap->logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
 225                 return XFS_ERROR(EINVAL);
 226         }
 227         mp->m_logbufs = ap->logbufs;
 228         if (ap->logbufsize != -1 &&
 229             ap->logbufsize != 16 * 1024 &&
 230             ap->logbufsize != 32 * 1024 &&
 231             ap->logbufsize != 64 * 1024 &&
 232             ap->logbufsize != 128 * 1024 &&
 233             ap->logbufsize != 256 * 1024) {
 234                 cmn_err(CE_WARN,
 235         "XFS: invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
 236                         ap->logbufsize);
 237                 return XFS_ERROR(EINVAL);
 238         }
 239         mp->m_ihsize = ap->ihashsize;
 240         mp->m_logbsize = ap->logbufsize;
 241         mp->m_fsname_len = strlen(ap->fsname) + 1;
 242         mp->m_fsname = kmem_alloc(mp->m_fsname_len, KM_SLEEP);
 243         strcpy(mp->m_fsname, ap->fsname);
 244         if (ap->rtname[0]) {
 245                 mp->m_rtname = kmem_alloc(strlen(ap->rtname) + 1, KM_SLEEP);
 246                 strcpy(mp->m_rtname, ap->rtname);
 247         }
 248         if (ap->logname[0]) {
 249                 mp->m_logname = kmem_alloc(strlen(ap->logname) + 1, KM_SLEEP);
 250                 strcpy(mp->m_logname, ap->logname);
 251         }
 252
 253         if (ap->flags & XFSMNT_WSYNC)
 254                 mp->m_flags |= XFS_MOUNT_WSYNC;
 255 #if XFS_BIG_INUMS
 256         if (ap->flags & XFSMNT_INO64) {
 257                 mp->m_flags |= XFS_MOUNT_INO64;
 258                 mp->m_inoadd = XFS_INO64_OFFSET;
 259         }
 260 #endif
 261         if (ap->flags & XFSMNT_NOATIME)
 262                 mp->m_flags |= XFS_MOUNT_NOATIME;
 263         if (ap->flags & XFSMNT_RETERR)
 264                 mp->m_flags |= XFS_MOUNT_RETERR;
 265         if (ap->flags & XFSMNT_NOALIGN)
 266                 mp->m_flags |= XFS_MOUNT_NOALIGN;
 267         if (ap->flags & XFSMNT_SWALLOC)
 268                 mp->m_flags |= XFS_MOUNT_SWALLOC;
 269         if (ap->flags & XFSMNT_OSYNCISOSYNC)
 270                 mp->m_flags |= XFS_MOUNT_OSYNCISOSYNC;
 271         if (ap->flags & XFSMNT_32BITINODES)
 272                 mp->m_flags |= XFS_MOUNT_32BITINODES;
 273
 274         if (ap->flags & XFSMNT_IOSIZE) {
 275                 if (ap->iosizelog > XFS_MAX_IO_LOG ||
 276                     ap->iosizelog < XFS_MIN_IO_LOG) {
 277                         cmn_err(CE_WARN,
 278                 "XFS: invalid log iosize: %d [not %d-%d]",
 279                                 ap->iosizelog, XFS_MIN_IO_LOG,
 280                                 XFS_MAX_IO_LOG);
 281                         return XFS_ERROR(EINVAL);
 282                 }
 283
 284                 mp->m_flags |= XFS_MOUNT_DFLT_IOSIZE;
 285                 mp->m_readio_log = mp->m_writeio_log = ap->iosizelog;
 286         }
 287
 288         if (ap->flags & XFSMNT_IHASHSIZE)
 289                 mp->m_flags |= XFS_MOUNT_IHASHSIZE;
 290         if (ap->flags & XFSMNT_IDELETE)
 291                 mp->m_flags |= XFS_MOUNT_IDELETE;
 292         if (ap->flags & XFSMNT_DIRSYNC)
 293                 mp->m_flags |= XFS_MOUNT_DIRSYNC;
 294         if (ap->flags & XFSMNT_COMPAT_ATTR)
 295                 mp->m_flags |= XFS_MOUNT_COMPAT_ATTR;
 296
 297         if (ap->flags2 & XFSMNT2_COMPAT_IOSIZE)
 298                 mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE;
 299
 300         /*
 301          * no recovery flag requires a read-only mount
 302          */
 303         if (ap->flags & XFSMNT_NORECOVERY) {
 304                 if (!(vfs->vfs_flag & VFS_RDONLY)) {
 305                         cmn_err(CE_WARN,
 306         "XFS: tried to mount a FS read-write without recovery!");
 307                         return XFS_ERROR(EINVAL);
 308                 }
 309                 mp->m_flags |= XFS_MOUNT_NORECOVERY;
 310         }
 311
 312         if (ap->flags & XFSMNT_NOUUID)
 313                 mp->m_flags |= XFS_MOUNT_NOUUID;
 314         if (ap->flags & XFSMNT_BARRIER)
 315                 mp->m_flags |= XFS_MOUNT_BARRIER;
 316
 317         return 0;
 318 }
 319
 320 /*
 321  * This function fills in xfs_mount_t fields based on mount args.
 322  * Note: the superblock _has_ now been read in.
 323  */
 324 STATIC int
 325 xfs_finish_flags(
 326         struct vfs              *vfs,
 327         struct xfs_mount_args   *ap,
 328         struct xfs_mount        *mp)
 329 {
 330         int                     ronly = (vfs->vfs_flag & VFS_RDONLY);
 331
 332         /* Fail a mount where the logbuf is smaller then the log stripe */
 333         if (XFS_SB_VERSION_HASLOGV2(&mp->m_sb)) {
 334                 if ((ap->logbufsize == -1) &&
 335                     (mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE)) {
 336                         mp->m_logbsize = mp->m_sb.sb_logsunit;
 337                 } else if (ap->logbufsize < mp->m_sb.sb_logsunit) {
 338                         cmn_err(CE_WARN,
 339         "XFS: logbuf size must be greater than or equal to log stripe size");
 340                         return XFS_ERROR(EINVAL);
 341                 }
 342         } else {
 343                 /* Fail a mount if the logbuf is larger than 32K */
 344                 if (ap->logbufsize > XLOG_BIG_RECORD_BSIZE) {
 345                         cmn_err(CE_WARN,
 346         "XFS: logbuf size for version 1 logs must be 16K or 32K");
 347                         return XFS_ERROR(EINVAL);
 348                 }
 349         }
 350
 351         /*
 352          * prohibit r/w mounts of read-only filesystems
 353          */
 354         if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !ronly) {
 355                 cmn_err(CE_WARN,
 356         "XFS: cannot mount a read-only filesystem as read-write");
 357                 return XFS_ERROR(EROFS);
 358         }
 359
 360         /*
 361          * check for shared mount.
 362          */
 363         if (ap->flags & XFSMNT_SHARED) {
 364                 if (!XFS_SB_VERSION_HASSHARED(&mp->m_sb))
 365                         return XFS_ERROR(EINVAL);
 366
 367                 /*
 368                  * For IRIX 6.5, shared mounts must have the shared
 369                  * version bit set, have the persistent readonly
 370                  * field set, must be version 0 and can only be mounted
 371                  * read-only.
 372                  */
 373                 if (!ronly || !(mp->m_sb.sb_flags & XFS_SBF_READONLY) ||
 374                      (mp->m_sb.sb_shared_vn != 0))
 375                         return XFS_ERROR(EINVAL);
 376
 377                 mp->m_flags |= XFS_MOUNT_SHARED;
 378
 379                 /*
 380                  * Shared XFS V0 can't deal with DMI.  Return EINVAL.
 381                  */
 382                 if (mp->m_sb.sb_shared_vn == 0 && (ap->flags & XFSMNT_DMAPI))
 383                         return XFS_ERROR(EINVAL);
 384         }
 385
 386         if (XFS_SB_VERSION_HASATTR2(&mp->m_sb)) {
 387                 mp->m_flags &= ~XFS_MOUNT_COMPAT_ATTR;
 388         }
 389
 390         return 0;
 391 }
 392
 393 /*
 394  * xfs_mount
 395  *
 396  * The file system configurations are:
 397  *      (1) device (partition) with data and internal log
 398  *      (2) logical volume with data and log subvolumes.
 399  *      (3) logical volume with data, log, and realtime subvolumes.
 400  *
 401  * We only have to handle opening the log and realtime volumes here if
 402  * they are present.  The data subvolume has already been opened by
 403  * get_sb_bdev() and is stored in vfsp->vfs_super->s_bdev.
 404  */
 405 STATIC int
 406 xfs_mount(
 407         struct bhv_desc         *bhvp,
 408         struct xfs_mount_args   *args,
 409         cred_t                  *credp)
 410 {
 411         struct vfs              *vfsp = bhvtovfs(bhvp);
 412         struct bhv_desc         *p;
 413         struct xfs_mount        *mp = XFS_BHVTOM(bhvp);
 414         struct block_device     *ddev, *logdev, *rtdev;
 415         int                     flags = 0, error;
 416
 417         ddev = vfsp->vfs_super->s_bdev;
 418         logdev = rtdev = NULL;
 419
 420         /*
 421          * Setup xfs_mount function vectors from available behaviors
 422          */
 423         p = vfs_bhv_lookup(vfsp, VFS_POSITION_DM);
 424         mp->m_dm_ops = p ? *(xfs_dmops_t *) vfs_bhv_custom(p) : xfs_dmcore_stub;
 425         p = vfs_bhv_lookup(vfsp, VFS_POSITION_QM);
 426         mp->m_qm_ops = p ? *(xfs_qmops_t *) vfs_bhv_custom(p) : xfs_qmcore_stub;
 427         p = vfs_bhv_lookup(vfsp, VFS_POSITION_IO);
 428         mp->m_io_ops = p ? *(xfs_ioops_t *) vfs_bhv_custom(p) : xfs_iocore_xfs;
 429
 430         /*
 431          * Open real time and log devices - order is important.
 432          */
 433         if (args->logname[0]) {
 434                 error = xfs_blkdev_get(mp, args->logname, &logdev);
 435                 if (error)
 436                         return error;
 437         }
 438         if (args->rtname[0]) {
 439                 error = xfs_blkdev_get(mp, args->rtname, &rtdev);
 440                 if (error) {
 441                         xfs_blkdev_put(logdev);
 442                         return error;
 443                 }
 444
 445                 if (rtdev == ddev || rtdev == logdev) {
 446                         cmn_err(CE_WARN,
 447         "XFS: Cannot mount filesystem with identical rtdev and ddev/logdev.");
 448                         xfs_blkdev_put(logdev);
 449                         xfs_blkdev_put(rtdev);
 450                         return EINVAL;
 451                 }
 452         }
 453
 454         /*
 455          * Setup xfs_mount buffer target pointers
 456          */
 457         error = ENOMEM;
 458         mp->m_ddev_targp = xfs_alloc_buftarg(ddev, 0);
 459         if (!mp->m_ddev_targp) {
 460                 xfs_blkdev_put(logdev);
 461                 xfs_blkdev_put(rtdev);
 462                 return error;
 463         }
 464         if (rtdev) {
 465                 mp->m_rtdev_targp = xfs_alloc_buftarg(rtdev, 1);
 466                 if (!mp->m_rtdev_targp)
 467                         goto error0;
 468         }
 469         mp->m_logdev_targp = (logdev && logdev != ddev) ?
 470                                 xfs_alloc_buftarg(logdev, 1) : mp->m_ddev_targp;
 471         if (!mp->m_logdev_targp)
 472                 goto error0;
 473
 474         /*
 475          * Setup flags based on mount(2) options and then the superblock
 476          */
 477         error = xfs_start_flags(vfsp, args, mp);
 478         if (error)
 479                 goto error1;
 480         error = xfs_readsb(mp);
 481         if (error)
 482                 goto error1;
 483         error = xfs_finish_flags(vfsp, args, mp);
 484         if (error)
 485                 goto error2;
 486
 487         /*
 488          * Setup xfs_mount buffer target pointers based on superblock
 489          */
 490         error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_blocksize,
 491                                     mp->m_sb.sb_sectsize);
 492         if (!error && logdev && logdev != ddev) {
 493                 unsigned int    log_sector_size = BBSIZE;
 494
 495                 if (XFS_SB_VERSION_HASSECTOR(&mp->m_sb))
 496                         log_sector_size = mp->m_sb.sb_logsectsize;
 497                 error = xfs_setsize_buftarg(mp->m_logdev_targp,
 498                                             mp->m_sb.sb_blocksize,
 499                                             log_sector_size);
 500         }
 501         if (!error && rtdev)
 502                 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
 503                                             mp->m_sb.sb_blocksize,
 504                                             mp->m_sb.sb_sectsize);
 505         if (error)
 506                 goto error2;
 507
 508         error = XFS_IOINIT(vfsp, args, flags);
 509         if (error)
 510                 goto error2;
 511
 512         if ((args->flags & XFSMNT_BARRIER) &&
 513             !(XFS_MTOVFS(mp)->vfs_flag & VFS_RDONLY))
 514                 xfs_mountfs_check_barriers(mp);
 515         return 0;
 516
 517 error2:
 518         if (mp->m_sb_bp)
 519                 xfs_freesb(mp);
 520 error1:
 521         xfs_binval(mp->m_ddev_targp);
 522         if (logdev && logdev != ddev)
 523                 xfs_binval(mp->m_logdev_targp);
 524         if (rtdev)
 525                 xfs_binval(mp->m_rtdev_targp);
 526 error0:
 527         xfs_unmountfs_close(mp, credp);
 528         return error;
 529 }
 530
 531 STATIC int
 532 xfs_unmount(
 533         bhv_desc_t      *bdp,
 534         int             flags,
 535         cred_t          *credp)
 536 {
 537         struct vfs      *vfsp = bhvtovfs(bdp);
 538         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
 539         xfs_inode_t     *rip;
 540         vnode_t         *rvp;
 541         int             unmount_event_wanted = 0;
 542         int             unmount_event_flags = 0;
 543         int             xfs_unmountfs_needed = 0;
 544         int             error;
 545
 546         rip = mp->m_rootip;
 547         rvp = XFS_ITOV(rip);
 548
 549         if (vfsp->vfs_flag & VFS_DMI) {
 550                 error = XFS_SEND_PREUNMOUNT(mp, vfsp,
 551                                 rvp, DM_RIGHT_NULL, rvp, DM_RIGHT_NULL,
 552                                 NULL, NULL, 0, 0,
 553                                 (mp->m_dmevmask & (1<<DM_EVENT_PREUNMOUNT))?
 554                                         0:DM_FLAGS_UNWANTED);
 555                         if (error)
 556                                 return XFS_ERROR(error);
 557                 unmount_event_wanted = 1;
 558                 unmount_event_flags = (mp->m_dmevmask & (1<<DM_EVENT_UNMOUNT))?
 559                                         0 : DM_FLAGS_UNWANTED;
 560         }
 561
 562         /*
 563          * First blow any referenced inode from this file system
 564          * out of the reference cache, and delete the timer.
 565          */
 566         xfs_refcache_purge_mp(mp);
 567
 568         XFS_bflush(mp->m_ddev_targp);
 569         error = xfs_unmount_flush(mp, 0);
 570         if (error)
 571                 goto out;
 572
 573         ASSERT(vn_count(rvp) == 1);
 574
 575         /*
 576          * Drop the reference count
 577          */
 578         VN_RELE(rvp);
 579
 580         /*
 581          * If we're forcing a shutdown, typically because of a media error,
 582          * we want to make sure we invalidate dirty pages that belong to
 583          * referenced vnodes as well.
 584          */
 585         if (XFS_FORCED_SHUTDOWN(mp)) {
 586                 error = xfs_sync(&mp->m_bhv,
 587                          (SYNC_WAIT | SYNC_CLOSE), credp);
 588                 ASSERT(error != EFSCORRUPTED);
 589         }
 590         xfs_unmountfs_needed = 1;
 591
 592 out:
 593         /*      Send DMAPI event, if required.
 594          *      Then do xfs_unmountfs() if needed.
 595          *      Then return error (or zero).
 596          */
 597         if (unmount_event_wanted) {
 598                 /* Note: mp structure must still exist for
 599                  * XFS_SEND_UNMOUNT() call.
 600                  */
 601                 XFS_SEND_UNMOUNT(mp, vfsp, error == 0 ? rvp : NULL,
 602                         DM_RIGHT_NULL, 0, error, unmount_event_flags);
 603         }
 604         if (xfs_unmountfs_needed) {
 605                 /*
 606                  * Call common unmount function to flush to disk
 607                  * and free the super block buffer & mount structures.
 608                  */
 609                 xfs_unmountfs(mp, credp);
 610         }
 611
 612         return XFS_ERROR(error);
 613 }
 614
 615 STATIC int
 616 xfs_quiesce_fs(
 617         xfs_mount_t             *mp)
 618 {
 619         int                     count = 0, pincount;
 620
 621         xfs_refcache_purge_mp(mp);
 622         xfs_flush_buftarg(mp->m_ddev_targp, 0);
 623         xfs_finish_reclaim_all(mp, 0);
 624
 625         /* This loop must run at least twice.
 626          * The first instance of the loop will flush
 627          * most meta data but that will generate more
 628          * meta data (typically directory updates).
 629          * Which then must be flushed and logged before
 630          * we can write the unmount record.
 631          */
 632         do {
 633                 xfs_syncsub(mp, SYNC_REMOUNT|SYNC_ATTR|SYNC_WAIT, 0, NULL);
 634                 pincount = xfs_flush_buftarg(mp->m_ddev_targp, 1);
 635                 if (!pincount) {
 636                         delay(50);
 637                         count++;
 638                 }
 639         } while (count < 2);
 640
 641         return 0;
 642 }
 643
 644 STATIC int
 645 xfs_mntupdate(
 646         bhv_desc_t                      *bdp,
 647         int                             *flags,
 648         struct xfs_mount_args           *args)
 649 {
 650         struct vfs      *vfsp = bhvtovfs(bdp);
 651         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
 652         int             error;
 653
 654         if (args->flags & XFSMNT_NOATIME)
 655                 mp->m_flags |= XFS_MOUNT_NOATIME;
 656         else
 657                 mp->m_flags &= ~XFS_MOUNT_NOATIME;
 658
 659         if ((vfsp->vfs_flag & VFS_RDONLY) &&
 660             !(*flags & MS_RDONLY)) {
 661                 vfsp->vfs_flag &= ~VFS_RDONLY;
 662
 663                 if (args->flags & XFSMNT_BARRIER)
 664                         xfs_mountfs_check_barriers(mp);
 665         }
 666
 667         if (!(vfsp->vfs_flag & VFS_RDONLY) &&
 668             (*flags & MS_RDONLY)) {
 669                 VFS_SYNC(vfsp, SYNC_FSDATA|SYNC_BDFLUSH|SYNC_ATTR, NULL, error);
 670
 671                 xfs_quiesce_fs(mp);
 672
 673                 /* Ok now write out an unmount record */
 674                 xfs_log_unmount_write(mp);
 675                 xfs_unmountfs_writesb(mp);
 676                 vfsp->vfs_flag |= VFS_RDONLY;
 677         }
 678
 679         return 0;
 680 }
 681
 682 /*
 683  * xfs_unmount_flush implements a set of flush operation on special
 684  * inodes, which are needed as a separate set of operations so that
 685  * they can be called as part of relocation process.
 686  */
 687 int
 688 xfs_unmount_flush(
 689         xfs_mount_t     *mp,            /* Mount structure we are getting
 690                                            rid of. */
 691         int             relocation)     /* Called from vfs relocation. */
 692 {
 693         xfs_inode_t     *rip = mp->m_rootip;
 694         xfs_inode_t     *rbmip;
 695         xfs_inode_t     *rsumip = NULL;
 696         vnode_t         *rvp = XFS_ITOV(rip);
 697         int             error;
 698
 699         xfs_ilock(rip, XFS_ILOCK_EXCL);
 700         xfs_iflock(rip);
 701
 702         /*
 703          * Flush out the real time inodes.
 704          */
 705         if ((rbmip = mp->m_rbmip) != NULL) {
 706                 xfs_ilock(rbmip, XFS_ILOCK_EXCL);
 707                 xfs_iflock(rbmip);
 708                 error = xfs_iflush(rbmip, XFS_IFLUSH_SYNC);
 709                 xfs_iunlock(rbmip, XFS_ILOCK_EXCL);
 710
 711                 if (error == EFSCORRUPTED)
 712                         goto fscorrupt_out;
 713
 714                 ASSERT(vn_count(XFS_ITOV(rbmip)) == 1);
 715
 716                 rsumip = mp->m_rsumip;
 717                 xfs_ilock(rsumip, XFS_ILOCK_EXCL);
 718                 xfs_iflock(rsumip);
 719                 error = xfs_iflush(rsumip, XFS_IFLUSH_SYNC);
 720                 xfs_iunlock(rsumip, XFS_ILOCK_EXCL);
 721
 722                 if (error == EFSCORRUPTED)
 723                         goto fscorrupt_out;
 724
 725                 ASSERT(vn_count(XFS_ITOV(rsumip)) == 1);
 726         }
 727
 728         /*
 729          * Synchronously flush root inode to disk
 730          */
 731         error = xfs_iflush(rip, XFS_IFLUSH_SYNC);
 732         if (error == EFSCORRUPTED)
 733                 goto fscorrupt_out2;
 734
 735         if (vn_count(rvp) != 1 && !relocation) {
 736                 xfs_iunlock(rip, XFS_ILOCK_EXCL);
 737                 return XFS_ERROR(EBUSY);
 738         }
 739
 740         /*
 741          * Release dquot that rootinode, rbmino and rsumino might be holding,
 742          * flush and purge the quota inodes.
 743          */
 744         error = XFS_QM_UNMOUNT(mp);
 745         if (error == EFSCORRUPTED)
 746                 goto fscorrupt_out2;
 747
 748         if (rbmip) {
 749                 VN_RELE(XFS_ITOV(rbmip));
 750                 VN_RELE(XFS_ITOV(rsumip));
 751         }
 752
 753         xfs_iunlock(rip, XFS_ILOCK_EXCL);
 754         return 0;
 755
 756 fscorrupt_out:
 757         xfs_ifunlock(rip);
 758
 759 fscorrupt_out2:
 760         xfs_iunlock(rip, XFS_ILOCK_EXCL);
 761
 762         return XFS_ERROR(EFSCORRUPTED);
 763 }
 764
 765 /*
 766  * xfs_root extracts the root vnode from a vfs.
 767  *
 768  * vfsp -- the vfs struct for the desired file system
 769  * vpp  -- address of the caller's vnode pointer which should be
 770  *         set to the desired fs root vnode
 771  */
 772 STATIC int
 773 xfs_root(
 774         bhv_desc_t      *bdp,
 775         vnode_t         **vpp)
 776 {
 777         vnode_t         *vp;
 778
 779         vp = XFS_ITOV((XFS_BHVTOM(bdp))->m_rootip);
 780         VN_HOLD(vp);
 781         *vpp = vp;
 782         return 0;
 783 }
 784
 785 /*
 786  * xfs_statvfs
 787  *
 788  * Fill in the statvfs structure for the given file system.  We use
 789  * the superblock lock in the mount structure to ensure a consistent
 790  * snapshot of the counters returned.
 791  */
 792 STATIC int
 793 xfs_statvfs(
 794         bhv_desc_t      *bdp,
 795         xfs_statfs_t    *statp,
 796         vnode_t         *vp)
 797 {
 798         __uint64_t      fakeinos;
 799         xfs_extlen_t    lsize;
 800         xfs_mount_t     *mp;
 801         xfs_sb_t        *sbp;
 802         unsigned long   s;
 803
 804         mp = XFS_BHVTOM(bdp);
 805         sbp = &(mp->m_sb);
 806
 807         statp->f_type = XFS_SB_MAGIC;
 808
 809         s = XFS_SB_LOCK(mp);
 810         statp->f_bsize = sbp->sb_blocksize;
 811         lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
 812         statp->f_blocks = sbp->sb_dblocks - lsize;
 813         statp->f_bfree = statp->f_bavail = sbp->sb_fdblocks;
 814         fakeinos = statp->f_bfree << sbp->sb_inopblog;
 815 #if XFS_BIG_INUMS
 816         fakeinos += mp->m_inoadd;
 817 #endif
 818         statp->f_files =
 819             MIN(sbp->sb_icount + fakeinos, (__uint64_t)XFS_MAXINUMBER);
 820         if (mp->m_maxicount)
 821 #if XFS_BIG_INUMS
 822                 if (!mp->m_inoadd)
 823 #endif
 824                         statp->f_files = min_t(typeof(statp->f_files),
 825                                                 statp->f_files,
 826                                                 mp->m_maxicount);
 827         statp->f_ffree = statp->f_files - (sbp->sb_icount - sbp->sb_ifree);
 828         XFS_SB_UNLOCK(mp, s);
 829
 830         xfs_statvfs_fsid(statp, mp);
 831         statp->f_namelen = MAXNAMELEN - 1;
 832
 833         return 0;
 834 }
 835
 836
 837 /*
 838  * xfs_sync flushes any pending I/O to file system vfsp.
 839  *
 840  * This routine is called by vfs_sync() to make sure that things make it
 841  * out to disk eventually, on sync() system calls to flush out everything,
 842  * and when the file system is unmounted.  For the vfs_sync() case, all
 843  * we really need to do is sync out the log to make all of our meta-data
 844  * updates permanent (except for timestamps).  For calls from pflushd(),
 845  * dirty pages are kept moving by calling pdflush() on the inodes
 846  * containing them.  We also flush the inodes that we can lock without
 847  * sleeping and the superblock if we can lock it without sleeping from
 848  * vfs_sync() so that items at the tail of the log are always moving out.
 849  *
 850  * Flags:
 851  *      SYNC_BDFLUSH - We're being called from vfs_sync() so we don't want
 852  *                     to sleep if we can help it.  All we really need
 853  *                     to do is ensure that the log is synced at least
 854  *                     periodically.  We also push the inodes and
 855  *                     superblock if we can lock them without sleeping
 856  *                      and they are not pinned.
 857  *      SYNC_ATTR    - We need to flush the inodes.  If SYNC_BDFLUSH is not
 858  *                     set, then we really want to lock each inode and flush
 859  *                     it.
 860  *      SYNC_WAIT    - All the flushes that take place in this call should
 861  *                     be synchronous.
 862  *      SYNC_DELWRI  - This tells us to push dirty pages associated with
 863  *                     inodes.  SYNC_WAIT and SYNC_BDFLUSH are used to
 864  *                     determine if they should be flushed sync, async, or
 865  *                     delwri.
 866  *      SYNC_CLOSE   - This flag is passed when the system is being
 867  *                     unmounted.  We should sync and invalidate everthing.
 868  *      SYNC_FSDATA  - This indicates that the caller would like to make
 869  *                     sure the superblock is safe on disk.  We can ensure
 870  *                     this by simply makeing sure the log gets flushed
 871  *                     if SYNC_BDFLUSH is set, and by actually writing it
 872  *                     out otherwise.
 873  *
 874  */
 875 /*ARGSUSED*/
 876 STATIC int
 877 xfs_sync(
 878         bhv_desc_t      *bdp,
 879         int             flags,
 880         cred_t          *credp)
 881 {
 882         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
 883
 884         if (unlikely(flags == SYNC_QUIESCE))
 885                 return xfs_quiesce_fs(mp);
 886         else
 887                 return xfs_syncsub(mp, flags, 0, NULL);
 888 }
 889
 890 /*
 891  * xfs sync routine for internal use
 892  *
 893  * This routine supports all of the flags defined for the generic VFS_SYNC
 894  * interface as explained above under xfs_sync.  In the interests of not
 895  * changing interfaces within the 6.5 family, additional internallly-
 896  * required functions are specified within a separate xflags parameter,
 897  * only available by calling this routine.
 898  *
 899  */
 900 int
 901 xfs_sync_inodes(
 902         xfs_mount_t     *mp,
 903         int             flags,
 904         int             xflags,
 905         int             *bypassed)
 906 {
 907         xfs_inode_t     *ip = NULL;
 908         xfs_inode_t     *ip_next;
 909         xfs_buf_t       *bp;
 910         vnode_t         *vp = NULL;
 911         int             error;
 912         int             last_error;
 913         uint64_t        fflag;
 914         uint            lock_flags;
 915         uint            base_lock_flags;
 916         boolean_t       mount_locked;
 917         boolean_t       vnode_refed;
 918         int             preempt;
 919         xfs_dinode_t    *dip;
 920         xfs_iptr_t      *ipointer;
 921 #ifdef DEBUG
 922         boolean_t       ipointer_in = B_FALSE;
 923
 924 #define IPOINTER_SET    ipointer_in = B_TRUE
 925 #define IPOINTER_CLR    ipointer_in = B_FALSE
 926 #else
 927 #define IPOINTER_SET
 928 #define IPOINTER_CLR
 929 #endif
 930
 931
 932 /* Insert a marker record into the inode list after inode ip. The list
 933  * must be locked when this is called. After the call the list will no
 934  * longer be locked.
 935  */
 936 #define IPOINTER_INSERT(ip, mp) { \
 937                 ASSERT(ipointer_in == B_FALSE); \
 938                 ipointer->ip_mnext = ip->i_mnext; \
 939                 ipointer->ip_mprev = ip; \
 940                 ip->i_mnext = (xfs_inode_t *)ipointer; \
 941                 ipointer->ip_mnext->i_mprev = (xfs_inode_t *)ipointer; \
 942                 preempt = 0; \
 943                 XFS_MOUNT_IUNLOCK(mp); \
 944                 mount_locked = B_FALSE; \
 945                 IPOINTER_SET; \
 946         }
 947
 948 /* Remove the marker from the inode list. If the marker was the only item
 949  * in the list then there are no remaining inodes and we should zero out
 950  * the whole list. If we are the current head of the list then move the head
 951  * past us.
 952  */
 953 #define IPOINTER_REMOVE(ip, mp) { \
 954                 ASSERT(ipointer_in == B_TRUE); \
 955                 if (ipointer->ip_mnext != (xfs_inode_t *)ipointer) { \
 956                         ip = ipointer->ip_mnext; \
 957                         ip->i_mprev = ipointer->ip_mprev; \
 958                         ipointer->ip_mprev->i_mnext = ip; \
 959                         if (mp->m_inodes == (xfs_inode_t *)ipointer) { \
 960                                 mp->m_inodes = ip; \
 961                         } \
 962                 } else { \
 963                         ASSERT(mp->m_inodes == (xfs_inode_t *)ipointer); \
 964                         mp->m_inodes = NULL; \
 965                         ip = NULL; \
 966                 } \
 967                 IPOINTER_CLR; \
 968         }
 969
 970 #define XFS_PREEMPT_MASK        0x7f
 971
 972         if (bypassed)
 973                 *bypassed = 0;
 974         if (XFS_MTOVFS(mp)->vfs_flag & VFS_RDONLY)
 975                 return 0;
 976         error = 0;
 977         last_error = 0;
 978         preempt = 0;
 979
 980         /* Allocate a reference marker */
 981         ipointer = (xfs_iptr_t *)kmem_zalloc(sizeof(xfs_iptr_t), KM_SLEEP);
 982
 983         fflag = XFS_B_ASYNC;            /* default is don't wait */
 984         if (flags & (SYNC_BDFLUSH | SYNC_DELWRI))
 985                 fflag = XFS_B_DELWRI;
 986         if (flags & SYNC_WAIT)
 987                 fflag = 0;              /* synchronous overrides all */
 988
 989         base_lock_flags = XFS_ILOCK_SHARED;
 990         if (flags & (SYNC_DELWRI | SYNC_CLOSE)) {
 991                 /*
 992                  * We need the I/O lock if we're going to call any of
 993                  * the flush/inval routines.
 994                  */
 995                 base_lock_flags |= XFS_IOLOCK_SHARED;
 996         }
 997
 998         XFS_MOUNT_ILOCK(mp);
 999
1000         ip = mp->m_inodes;
1001
1002         mount_locked = B_TRUE;
1003         vnode_refed  = B_FALSE;
1004
1005         IPOINTER_CLR;
1006
1007         do {
1008                 ASSERT(ipointer_in == B_FALSE);
1009                 ASSERT(vnode_refed == B_FALSE);
1010
1011                 lock_flags = base_lock_flags;
1012
1013                 /*
1014                  * There were no inodes in the list, just break out
1015                  * of the loop.
1016                  */
1017                 if (ip == NULL) {
1018                         break;
1019                 }
1020
1021                 /*
1022                  * We found another sync thread marker - skip it
1023                  */
1024                 if (ip->i_mount == NULL) {
1025                         ip = ip->i_mnext;
1026                         continue;
1027                 }
1028
1029                 vp = XFS_ITOV_NULL(ip);
1030
1031                 /*
1032                  * If the vnode is gone then this is being torn down,
1033                  * call reclaim if it is flushed, else let regular flush
1034                  * code deal with it later in the loop.
1035                  */
1036
1037                 if (vp == NULL) {
1038                         /* Skip ones already in reclaim */
1039                         if (ip->i_flags & XFS_IRECLAIM) {
1040                                 ip = ip->i_mnext;
1041                                 continue;
1042                         }
1043                         if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0) {
1044                                 ip = ip->i_mnext;
1045                         } else if ((xfs_ipincount(ip) == 0) &&
1046                                     xfs_iflock_nowait(ip)) {
1047                                 IPOINTER_INSERT(ip, mp);
1048
1049                                 xfs_finish_reclaim(ip, 1,
1050                                                 XFS_IFLUSH_DELWRI_ELSE_ASYNC);
1051
1052                                 XFS_MOUNT_ILOCK(mp);
1053                                 mount_locked = B_TRUE;
1054                                 IPOINTER_REMOVE(ip, mp);
1055                         } else {
1056                                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1057                                 ip = ip->i_mnext;
1058                         }
1059                         continue;
1060                 }
1061
1062                 if (VN_BAD(vp)) {
1063                         ip = ip->i_mnext;
1064                         continue;
1065                 }
1066
1067                 if (XFS_FORCED_SHUTDOWN(mp) && !(flags & SYNC_CLOSE)) {
1068                         XFS_MOUNT_IUNLOCK(mp);
1069                         kmem_free(ipointer, sizeof(xfs_iptr_t));
1070                         return 0;
1071                 }
1072
1073                 /*
1074                  * If this is just vfs_sync() or pflushd() calling
1075                  * then we can skip inodes for which it looks like
1076                  * there is nothing to do.  Since we don't have the
1077                  * inode locked this is racey, but these are periodic
1078                  * calls so it doesn't matter.  For the others we want
1079                  * to know for sure, so we at least try to lock them.
1080                  */
1081                 if (flags & SYNC_BDFLUSH) {
1082                         if (((ip->i_itemp == NULL) ||
1083                              !(ip->i_itemp->ili_format.ilf_fields &
1084                                XFS_ILOG_ALL)) &&
1085                             (ip->i_update_core == 0)) {
1086                                 ip = ip->i_mnext;
1087                                 continue;
1088                         }
1089                 }
1090
1091                 /*
1092                  * Try to lock without sleeping.  We're out of order with
1093                  * the inode list lock here, so if we fail we need to drop
1094                  * the mount lock and try again.  If we're called from
1095                  * bdflush() here, then don't bother.
1096                  *
1097                  * The inode lock here actually coordinates with the
1098                  * almost spurious inode lock in xfs_ireclaim() to prevent
1099                  * the vnode we handle here without a reference from
1100                  * being freed while we reference it.  If we lock the inode
1101                  * while it's on the mount list here, then the spurious inode
1102                  * lock in xfs_ireclaim() after the inode is pulled from
1103                  * the mount list will sleep until we release it here.
1104                  * This keeps the vnode from being freed while we reference
1105                  * it.
1106                  */
1107                 if (xfs_ilock_nowait(ip, lock_flags) == 0) {
1108                         if ((flags & SYNC_BDFLUSH) || (vp == NULL)) {
1109                                 ip = ip->i_mnext;
1110                                 continue;
1111                         }
1112
1113                         vp = vn_grab(vp);
1114                         if (vp == NULL) {
1115                                 ip = ip->i_mnext;
1116                                 continue;
1117                         }
1118
1119                         IPOINTER_INSERT(ip, mp);
1120                         xfs_ilock(ip, lock_flags);
1121
1122                         ASSERT(vp == XFS_ITOV(ip));
1123                         ASSERT(ip->i_mount == mp);
1124
1125                         vnode_refed = B_TRUE;
1126                 }
1127
1128                 /* From here on in the loop we may have a marker record
1129                  * in the inode list.
1130                  */
1131
1132                 if ((flags & SYNC_CLOSE)  && (vp != NULL)) {
1133                         /*
1134                          * This is the shutdown case.  We just need to
1135                          * flush and invalidate all the pages associated
1136                          * with the inode.  Drop the inode lock since
1137                          * we can't hold it across calls to the buffer
1138                          * cache.
1139                          *
1140                          * We don't set the VREMAPPING bit in the vnode
1141                          * here, because we don't hold the vnode lock
1142                          * exclusively.  It doesn't really matter, though,
1143                          * because we only come here when we're shutting
1144                          * down anyway.
1145                          */
1146                         xfs_iunlock(ip, XFS_ILOCK_SHARED);
1147
1148                         if (XFS_FORCED_SHUTDOWN(mp)) {
1149                                 VOP_TOSS_PAGES(vp, 0, -1, FI_REMAPF);
1150                         } else {
1151                                 VOP_FLUSHINVAL_PAGES(vp, 0, -1, FI_REMAPF);
1152                         }
1153
1154                         xfs_ilock(ip, XFS_ILOCK_SHARED);
1155
1156                 } else if ((flags & SYNC_DELWRI) && (vp != NULL)) {
1157                         if (VN_DIRTY(vp)) {
1158                                 /* We need to have dropped the lock here,
1159                                  * so insert a marker if we have not already
1160                                  * done so.
1161                                  */
1162                                 if (mount_locked) {
1163                                         IPOINTER_INSERT(ip, mp);
1164                                 }
1165
1166                                 /*
1167                                  * Drop the inode lock since we can't hold it
1168                                  * across calls to the buffer cache.
1169                                  */
1170                                 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1171                                 VOP_FLUSH_PAGES(vp, (xfs_off_t)0, -1,
1172                                                         fflag, FI_NONE, error);
1173                                 xfs_ilock(ip, XFS_ILOCK_SHARED);
1174                         }
1175
1176                 }
1177
1178                 if (flags & SYNC_BDFLUSH) {
1179                         if ((flags & SYNC_ATTR) &&
1180                             ((ip->i_update_core) ||
1181                              ((ip->i_itemp != NULL) &&
1182                               (ip->i_itemp->ili_format.ilf_fields != 0)))) {
1183
1184                                 /* Insert marker and drop lock if not already
1185                                  * done.
1186                                  */
1187                                 if (mount_locked) {
1188                                         IPOINTER_INSERT(ip, mp);
1189                                 }
1190
1191                                 /*
1192                                  * We don't want the periodic flushing of the
1193                                  * inodes by vfs_sync() to interfere with
1194                                  * I/O to the file, especially read I/O
1195                                  * where it is only the access time stamp
1196                                  * that is being flushed out.  To prevent
1197                                  * long periods where we have both inode
1198                                  * locks held shared here while reading the
1199                                  * inode's buffer in from disk, we drop the
1200                                  * inode lock while reading in the inode
1201                                  * buffer.  We have to release the buffer
1202                                  * and reacquire the inode lock so that they
1203                                  * are acquired in the proper order (inode
1204                                  * locks first).  The buffer will go at the
1205                                  * end of the lru chain, though, so we can
1206                                  * expect it to still be there when we go
1207                                  * for it again in xfs_iflush().
1208                                  */
1209                                 if ((xfs_ipincount(ip) == 0) &&
1210                                     xfs_iflock_nowait(ip)) {
1211
1212                                         xfs_ifunlock(ip);
1213                                         xfs_iunlock(ip, XFS_ILOCK_SHARED);
1214
1215                                         error = xfs_itobp(mp, NULL, ip,
1216                                                           &dip, &bp, 0);
1217                                         if (!error) {
1218                                                 xfs_buf_relse(bp);
1219                                         } else {
1220                                                 /* Bailing out, remove the
1221                                                  * marker and free it.
1222                                                  */
1223                                                 XFS_MOUNT_ILOCK(mp);
1224
1225                                                 IPOINTER_REMOVE(ip, mp);
1226
1227                                                 XFS_MOUNT_IUNLOCK(mp);
1228
1229                                                 ASSERT(!(lock_flags &
1230                                                         XFS_IOLOCK_SHARED));
1231
1232                                                 kmem_free(ipointer,
1233                                                         sizeof(xfs_iptr_t));
1234                                                 return (0);
1235                                         }
1236
1237                                         /*
1238                                          * Since we dropped the inode lock,
1239                                          * the inode may have been reclaimed.
1240                                          * Therefore, we reacquire the mount
1241                                          * lock and check to see if we were the
1242                                          * inode reclaimed. If this happened
1243                                          * then the ipointer marker will no
1244                                          * longer point back at us. In this
1245                                          * case, move ip along to the inode
1246                                          * after the marker, remove the marker
1247                                          * and continue.
1248                                          */
1249                                         XFS_MOUNT_ILOCK(mp);
1250                                         mount_locked = B_TRUE;
1251
1252                                         if (ip != ipointer->ip_mprev) {
1253                                                 IPOINTER_REMOVE(ip, mp);
1254
1255                                                 ASSERT(!vnode_refed);
1256                                                 ASSERT(!(lock_flags &
1257                                                         XFS_IOLOCK_SHARED));
1258                                                 continue;
1259                                         }
1260
1261                                         ASSERT(ip->i_mount == mp);
1262
1263                                         if (xfs_ilock_nowait(ip,
1264                                                     XFS_ILOCK_SHARED) == 0) {
1265                                                 ASSERT(ip->i_mount == mp);
1266                                                 /*
1267                                                  * We failed to reacquire
1268                                                  * the inode lock without
1269                                                  * sleeping, so just skip
1270                                                  * the inode for now.  We
1271                                                  * clear the ILOCK bit from
1272                                                  * the lock_flags so that we
1273                                                  * won't try to drop a lock
1274                                                  * we don't hold below.
1275                                                  */
1276                                                 lock_flags &= ~XFS_ILOCK_SHARED;
1277                                                 IPOINTER_REMOVE(ip_next, mp);
1278                                         } else if ((xfs_ipincount(ip) == 0) &&
1279                                                    xfs_iflock_nowait(ip)) {
1280                                                 ASSERT(ip->i_mount == mp);
1281                                                 /*
1282                                                  * Since this is vfs_sync()
1283                                                  * calling we only flush the
1284                                                  * inode out if we can lock
1285                                                  * it without sleeping and
1286                                                  * it is not pinned.  Drop
1287                                                  * the mount lock here so
1288                                                  * that we don't hold it for
1289                                                  * too long. We already have
1290                                                  * a marker in the list here.
1291                                                  */
1292                                                 XFS_MOUNT_IUNLOCK(mp);
1293                                                 mount_locked = B_FALSE;
1294                                                 error = xfs_iflush(ip,
1295                                                            XFS_IFLUSH_DELWRI);
1296                                         } else {
1297                                                 ASSERT(ip->i_mount == mp);
1298                                                 IPOINTER_REMOVE(ip_next, mp);
1299                                         }
1300                                 }
1301
1302                         }
1303
1304                 } else {
1305                         if ((flags & SYNC_ATTR) &&
1306                             ((ip->i_update_core) ||
1307                              ((ip->i_itemp != NULL) &&
1308                               (ip->i_itemp->ili_format.ilf_fields != 0)))) {
1309                                 if (mount_locked) {
1310                                         IPOINTER_INSERT(ip, mp);
1311                                 }
1312
1313                                 if (flags & SYNC_WAIT) {
1314                                         xfs_iflock(ip);
1315                                         error = xfs_iflush(ip,
1316                                                            XFS_IFLUSH_SYNC);
1317                                 } else {
1318                                         /*
1319                                          * If we can't acquire the flush
1320                                          * lock, then the inode is already
1321                                          * being flushed so don't bother
1322                                          * waiting.  If we can lock it then
1323                                          * do a delwri flush so we can
1324                                          * combine multiple inode flushes
1325                                          * in each disk write.
1326                                          */
1327                                         if (xfs_iflock_nowait(ip)) {
1328                                                 error = xfs_iflush(ip,
1329                                                            XFS_IFLUSH_DELWRI);
1330                                         }
1331                                         else if (bypassed)
1332                                                 (*bypassed)++;
1333                                 }
1334                         }
1335                 }
1336
1337                 if (lock_flags != 0) {
1338                         xfs_iunlock(ip, lock_flags);
1339                 }
1340
1341                 if (vnode_refed) {
1342                         /*
1343                          * If we had to take a reference on the vnode
1344                          * above, then wait until after we've unlocked
1345                          * the inode to release the reference.  This is
1346                          * because we can be already holding the inode
1347                          * lock when VN_RELE() calls xfs_inactive().
1348                          *
1349                          * Make sure to drop the mount lock before calling
1350                          * VN_RELE() so that we don't trip over ourselves if
1351                          * we have to go for the mount lock again in the
1352                          * inactive code.
1353                          */
1354                         if (mount_locked) {
1355                                 IPOINTER_INSERT(ip, mp);
1356                         }
1357
1358                         VN_RELE(vp);
1359
1360                         vnode_refed = B_FALSE;
1361                 }
1362
1363                 if (error) {
1364                         last_error = error;
1365                 }
1366
1367                 /*
1368                  * bail out if the filesystem is corrupted.
1369                  */
1370                 if (error == EFSCORRUPTED)  {
1371                         if (!mount_locked) {
1372                                 XFS_MOUNT_ILOCK(mp);
1373                                 IPOINTER_REMOVE(ip, mp);
1374                         }
1375                         XFS_MOUNT_IUNLOCK(mp);
1376                         ASSERT(ipointer_in == B_FALSE);
1377                         kmem_free(ipointer, sizeof(xfs_iptr_t));
1378                         return XFS_ERROR(error);
1379                 }
1380
1381                 /* Let other threads have a chance at the mount lock
1382                  * if we have looped many times without dropping the
1383                  * lock.
1384                  */
1385                 if ((++preempt & XFS_PREEMPT_MASK) == 0) {
1386                         if (mount_locked) {
1387                                 IPOINTER_INSERT(ip, mp);
1388                         }
1389                 }
1390
1391                 if (mount_locked == B_FALSE) {
1392                         XFS_MOUNT_ILOCK(mp);
1393                         mount_locked = B_TRUE;
1394                         IPOINTER_REMOVE(ip, mp);
1395                         continue;
1396                 }
1397
1398                 ASSERT(ipointer_in == B_FALSE);
1399                 ip = ip->i_mnext;
1400
1401         } while (ip != mp->m_inodes);
1402
1403         XFS_MOUNT_IUNLOCK(mp);
1404
1405         ASSERT(ipointer_in == B_FALSE);
1406
1407         kmem_free(ipointer, sizeof(xfs_iptr_t));
1408         return XFS_ERROR(last_error);
1409 }
1410
1411 /*
1412  * xfs sync routine for internal use
1413  *
1414  * This routine supports all of the flags defined for the generic VFS_SYNC
1415  * interface as explained above under xfs_sync.  In the interests of not
1416  * changing interfaces within the 6.5 family, additional internallly-
1417  * required functions are specified within a separate xflags parameter,
1418  * only available by calling this routine.
1419  *
1420  */
1421 int
1422 xfs_syncsub(
1423         xfs_mount_t     *mp,
1424         int             flags,
1425         int             xflags,
1426         int             *bypassed)
1427 {
1428         int             error = 0;
1429         int             last_error = 0;
1430         uint            log_flags = XFS_LOG_FORCE;
1431         xfs_buf_t       *bp;
1432         xfs_buf_log_item_t      *bip;
1433
1434         /*
1435          * Sync out the log.  This ensures that the log is periodically
1436          * flushed even if there is not enough activity to fill it up.
1437          */
1438         if (flags & SYNC_WAIT)
1439                 log_flags |= XFS_LOG_SYNC;
1440
1441         xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1442
1443         if (flags & (SYNC_ATTR|SYNC_DELWRI)) {
1444                 if (flags & SYNC_BDFLUSH)
1445                         xfs_finish_reclaim_all(mp, 1);
1446                 else
1447                         error = xfs_sync_inodes(mp, flags, xflags, bypassed);
1448         }
1449
1450         /*
1451          * Flushing out dirty data above probably generated more
1452          * log activity, so if this isn't vfs_sync() then flush
1453          * the log again.
1454          */
1455         if (flags & SYNC_DELWRI) {
1456                 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1457         }
1458
1459         if (flags & SYNC_FSDATA) {
1460                 /*
1461                  * If this is vfs_sync() then only sync the superblock
1462                  * if we can lock it without sleeping and it is not pinned.
1463                  */
1464                 if (flags & SYNC_BDFLUSH) {
1465                         bp = xfs_getsb(mp, XFS_BUF_TRYLOCK);
1466                         if (bp != NULL) {
1467                                 bip = XFS_BUF_FSPRIVATE(bp,xfs_buf_log_item_t*);
1468                                 if ((bip != NULL) &&
1469                                     xfs_buf_item_dirty(bip)) {
1470                                         if (!(XFS_BUF_ISPINNED(bp))) {
1471                                                 XFS_BUF_ASYNC(bp);
1472                                                 error = xfs_bwrite(mp, bp);
1473                                         } else {
1474                                                 xfs_buf_relse(bp);
1475                                         }
1476                                 } else {
1477                                         xfs_buf_relse(bp);
1478                                 }
1479                         }
1480                 } else {
1481                         bp = xfs_getsb(mp, 0);
1482                         /*
1483                          * If the buffer is pinned then push on the log so
1484                          * we won't get stuck waiting in the write for
1485                          * someone, maybe ourselves, to flush the log.
1486                          * Even though we just pushed the log above, we
1487                          * did not have the superblock buffer locked at
1488                          * that point so it can become pinned in between
1489                          * there and here.
1490                          */
1491                         if (XFS_BUF_ISPINNED(bp))
1492                                 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
1493                         if (flags & SYNC_WAIT)
1494                                 XFS_BUF_UNASYNC(bp);
1495                         else
1496                                 XFS_BUF_ASYNC(bp);
1497                         error = xfs_bwrite(mp, bp);
1498                 }
1499                 if (error) {
1500                         last_error = error;
1501                 }
1502         }
1503
1504         /*
1505          * If this is the periodic sync, then kick some entries out of
1506          * the reference cache.  This ensures that idle entries are
1507          * eventually kicked out of the cache.
1508          */
1509         if (flags & SYNC_REFCACHE) {
1510                 if (flags & SYNC_WAIT)
1511                         xfs_refcache_purge_mp(mp);
1512                 else
1513                         xfs_refcache_purge_some(mp);
1514         }
1515
1516         /*
1517          * Now check to see if the log needs a "dummy" transaction.
1518          */
1519
1520         if (!(flags & SYNC_REMOUNT) && xfs_log_need_covered(mp)) {
1521                 xfs_trans_t *tp;
1522                 xfs_inode_t *ip;
1523
1524                 /*
1525                  * Put a dummy transaction in the log to tell
1526                  * recovery that all others are OK.
1527                  */
1528                 tp = xfs_trans_alloc(mp, XFS_TRANS_DUMMY1);
1529                 if ((error = xfs_trans_reserve(tp, 0,
1530                                 XFS_ICHANGE_LOG_RES(mp),
1531                                 0, 0, 0)))  {
1532                         xfs_trans_cancel(tp, 0);
1533                         return error;
1534                 }
1535
1536                 ip = mp->m_rootip;
1537                 xfs_ilock(ip, XFS_ILOCK_EXCL);
1538
1539                 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1540                 xfs_trans_ihold(tp, ip);
1541                 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1542                 error = xfs_trans_commit(tp, 0, NULL);
1543                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1544                 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1545         }
1546
1547         /*
1548          * When shutting down, we need to insure that the AIL is pushed
1549          * to disk or the filesystem can appear corrupt from the PROM.
1550          */
1551         if ((flags & (SYNC_CLOSE|SYNC_WAIT)) == (SYNC_CLOSE|SYNC_WAIT)) {
1552                 XFS_bflush(mp->m_ddev_targp);
1553                 if (mp->m_rtdev_targp) {
1554                         XFS_bflush(mp->m_rtdev_targp);
1555                 }
1556         }
1557
1558         return XFS_ERROR(last_error);
1559 }
1560
1561 /*
1562  * xfs_vget - called by DMAPI and NFSD to get vnode from file handle
1563  */
1564 STATIC int
1565 xfs_vget(
1566         bhv_desc_t      *bdp,
1567         vnode_t         **vpp,
1568         fid_t           *fidp)
1569 {
1570         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
1571         xfs_fid_t       *xfid = (struct xfs_fid *)fidp;
1572         xfs_inode_t     *ip;
1573         int             error;
1574         xfs_ino_t       ino;
1575         unsigned int    igen;
1576
1577         /*
1578          * Invalid.  Since handles can be created in user space and passed in
1579          * via gethandle(), this is not cause for a panic.
1580          */
1581         if (xfid->xfs_fid_len != sizeof(*xfid) - sizeof(xfid->xfs_fid_len))
1582                 return XFS_ERROR(EINVAL);
1583
1584         ino  = xfid->xfs_fid_ino;
1585         igen = xfid->xfs_fid_gen;
1586
1587         /*
1588          * NFS can sometimes send requests for ino 0.  Fail them gracefully.
1589          */
1590         if (ino == 0)
1591                 return XFS_ERROR(ESTALE);
1592
1593         error = xfs_iget(mp, NULL, ino, 0, XFS_ILOCK_SHARED, &ip, 0);
1594         if (error) {
1595                 *vpp = NULL;
1596                 return error;
1597         }
1598
1599         if (ip == NULL) {
1600                 *vpp = NULL;
1601                 return XFS_ERROR(EIO);
1602         }
1603
1604         if (ip->i_d.di_mode == 0 || ip->i_d.di_gen != igen) {
1605                 xfs_iput_new(ip, XFS_ILOCK_SHARED);
1606                 *vpp = NULL;
1607                 return XFS_ERROR(ENOENT);
1608         }
1609
1610         *vpp = XFS_ITOV(ip);
1611         xfs_iunlock(ip, XFS_ILOCK_SHARED);
1612         return 0;
1613 }
1614
1615
1616 #define MNTOPT_LOGBUFS  "logbufs"       /* number of XFS log buffers */
1617 #define MNTOPT_LOGBSIZE "logbsize"      /* size of XFS log buffers */
1618 #define MNTOPT_LOGDEV   "logdev"        /* log device */
1619 #define MNTOPT_RTDEV    "rtdev"         /* realtime I/O device */
1620 #define MNTOPT_BIOSIZE  "biosize"       /* log2 of preferred buffered io size */
1621 #define MNTOPT_WSYNC    "wsync"         /* safe-mode nfs compatible mount */
1622 #define MNTOPT_INO64    "ino64"         /* force inodes into 64-bit range */
1623 #define MNTOPT_NOALIGN  "noalign"       /* turn off stripe alignment */
1624 #define MNTOPT_SWALLOC  "swalloc"       /* turn on stripe width allocation */
1625 #define MNTOPT_SUNIT    "sunit"         /* data volume stripe unit */
1626 #define MNTOPT_SWIDTH   "swidth"        /* data volume stripe width */
1627 #define MNTOPT_NOUUID   "nouuid"        /* ignore filesystem UUID */
1628 #define MNTOPT_MTPT     "mtpt"          /* filesystem mount point */
1629 #define MNTOPT_GRPID    "grpid"         /* group-ID from parent directory */
1630 #define MNTOPT_NOGRPID  "nogrpid"       /* group-ID from current process */
1631 #define MNTOPT_BSDGROUPS    "bsdgroups"    /* group-ID from parent directory */
1632 #define MNTOPT_SYSVGROUPS   "sysvgroups"   /* group-ID from current process */
1633 #define MNTOPT_ALLOCSIZE    "allocsize"    /* preferred allocation size */
1634 #define MNTOPT_IHASHSIZE    "ihashsize"    /* size of inode hash table */
1635 #define MNTOPT_NORECOVERY   "norecovery"   /* don't run XFS recovery */
1636 #define MNTOPT_BARRIER  "barrier"       /* use writer barriers for log write and
1637                                          * unwritten extent conversion */
1638 #define MNTOPT_OSYNCISOSYNC "osyncisosync" /* o_sync is REALLY o_sync */
1639 #define MNTOPT_64BITINODE   "inode64"   /* inodes can be allocated anywhere */
1640 #define MNTOPT_IKEEP    "ikeep"         /* do not free empty inode clusters */
1641 #define MNTOPT_NOIKEEP  "noikeep"       /* free empty inode clusters */
1642 #define MNTOPT_LARGEIO     "largeio"    /* report large I/O sizes in stat() */
1643 #define MNTOPT_NOLARGEIO   "nolargeio"  /* do not report large I/O sizes
1644                                          * in stat(). */
1645 #define MNTOPT_ATTR2    "attr2"         /* do use attr2 attribute format */
1646 #define MNTOPT_NOATTR2  "noattr2"       /* do not use attr2 attribute format */
1647
1648 STATIC unsigned long
1649 suffix_strtoul(const char *cp, char **endp, unsigned int base)
1650 {
1651         int     last, shift_left_factor = 0;
1652         char    *value = (char *)cp;
1653
1654         last = strlen(value) - 1;
1655         if (value[last] == 'K' || value[last] == 'k') {
1656                 shift_left_factor = 10;
1657                 value[last] = '\0';
1658         }
1659         if (value[last] == 'M' || value[last] == 'm') {
1660                 shift_left_factor = 20;
1661                 value[last] = '\0';
1662         }
1663         if (value[last] == 'G' || value[last] == 'g') {
1664                 shift_left_factor = 30;
1665                 value[last] = '\0';
1666         }
1667
1668         return simple_strtoul(cp, endp, base) << shift_left_factor;
1669 }
1670
1671 STATIC int
1672 xfs_parseargs(
1673         struct bhv_desc         *bhv,
1674         char                    *options,
1675         struct xfs_mount_args   *args,
1676         int                     update)
1677 {
1678         struct vfs              *vfsp = bhvtovfs(bhv);
1679         char                    *this_char, *value, *eov;
1680         int                     dsunit, dswidth, vol_dsunit, vol_dswidth;
1681         int                     iosize;
1682
1683         args->flags2 |= XFSMNT2_COMPAT_IOSIZE;
1684         args->flags |= XFSMNT_COMPAT_ATTR;
1685
1686 #if 0   /* XXX: off by default, until some remaining issues ironed out */
1687         args->flags |= XFSMNT_IDELETE; /* default to on */
1688 #endif
1689
1690         if (!options)
1691                 return 0;
1692
1693         iosize = dsunit = dswidth = vol_dsunit = vol_dswidth = 0;
1694
1695         while ((this_char = strsep(&options, ",")) != NULL) {
1696                 if (!*this_char)
1697                         continue;
1698                 if ((value = strchr(this_char, '=')) != NULL)
1699                         *value++ = 0;
1700
1701                 if (!strcmp(this_char, MNTOPT_LOGBUFS)) {
1702                         if (!value || !*value) {
1703                                 printk("XFS: %s option requires an argument\n",
1704                                         this_char);
1705                                 return EINVAL;
1706                         }
1707                         args->logbufs = simple_strtoul(value, &eov, 10);
1708                 } else if (!strcmp(this_char, MNTOPT_LOGBSIZE)) {
1709                         if (!value || !*value) {
1710                                 printk("XFS: %s option requires an argument\n",
1711                                         this_char);
1712                                 return EINVAL;
1713                         }
1714                         args->logbufsize = suffix_strtoul(value, &eov, 10);
1715                 } else if (!strcmp(this_char, MNTOPT_LOGDEV)) {
1716                         if (!value || !*value) {
1717                                 printk("XFS: %s option requires an argument\n",
1718                                         this_char);
1719                                 return EINVAL;
1720                         }
1721                         strncpy(args->logname, value, MAXNAMELEN);
1722                 } else if (!strcmp(this_char, MNTOPT_MTPT)) {
1723                         if (!value || !*value) {
1724                                 printk("XFS: %s option requires an argument\n",
1725                                         this_char);
1726                                 return EINVAL;
1727                         }
1728                         strncpy(args->mtpt, value, MAXNAMELEN);
1729                 } else if (!strcmp(this_char, MNTOPT_RTDEV)) {
1730                         if (!value || !*value) {
1731                                 printk("XFS: %s option requires an argument\n",
1732                                         this_char);
1733                                 return EINVAL;
1734                         }
1735                         strncpy(args->rtname, value, MAXNAMELEN);
1736                 } else if (!strcmp(this_char, MNTOPT_BIOSIZE)) {
1737                         if (!value || !*value) {
1738                                 printk("XFS: %s option requires an argument\n",
1739                                         this_char);
1740                                 return EINVAL;
1741                         }
1742                         iosize = simple_strtoul(value, &eov, 10);
1743                         args->flags |= XFSMNT_IOSIZE;
1744                         args->iosizelog = (uint8_t) iosize;
1745                 } else if (!strcmp(this_char, MNTOPT_ALLOCSIZE)) {
1746                         if (!value || !*value) {
1747                                 printk("XFS: %s option requires an argument\n",
1748                                         this_char);
1749                                 return EINVAL;
1750                         }
1751                         iosize = suffix_strtoul(value, &eov, 10);
1752                         args->flags |= XFSMNT_IOSIZE;
1753                         args->iosizelog = ffs(iosize) - 1;
1754                 } else if (!strcmp(this_char, MNTOPT_IHASHSIZE)) {
1755                         if (!value || !*value) {
1756                                 printk("XFS: %s option requires an argument\n",
1757                                         this_char);
1758                                 return EINVAL;
1759                         }
1760                         args->flags |= XFSMNT_IHASHSIZE;
1761                         args->ihashsize = simple_strtoul(value, &eov, 10);
1762                 } else if (!strcmp(this_char, MNTOPT_GRPID) ||
1763                            !strcmp(this_char, MNTOPT_BSDGROUPS)) {
1764                         vfsp->vfs_flag |= VFS_GRPID;
1765                 } else if (!strcmp(this_char, MNTOPT_NOGRPID) ||
1766                            !strcmp(this_char, MNTOPT_SYSVGROUPS)) {
1767                         vfsp->vfs_flag &= ~VFS_GRPID;
1768                 } else if (!strcmp(this_char, MNTOPT_WSYNC)) {
1769                         args->flags |= XFSMNT_WSYNC;
1770                 } else if (!strcmp(this_char, MNTOPT_OSYNCISOSYNC)) {
1771                         args->flags |= XFSMNT_OSYNCISOSYNC;
1772                 } else if (!strcmp(this_char, MNTOPT_NORECOVERY)) {
1773                         args->flags |= XFSMNT_NORECOVERY;
1774                 } else if (!strcmp(this_char, MNTOPT_INO64)) {
1775                         args->flags |= XFSMNT_INO64;
1776 #if !XFS_BIG_INUMS
1777                         printk("XFS: %s option not allowed on this system\n",
1778                                 this_char);
1779                         return EINVAL;
1780 #endif
1781                 } else if (!strcmp(this_char, MNTOPT_NOALIGN)) {
1782                         args->flags |= XFSMNT_NOALIGN;
1783                 } else if (!strcmp(this_char, MNTOPT_SWALLOC)) {
1784                         args->flags |= XFSMNT_SWALLOC;
1785                 } else if (!strcmp(this_char, MNTOPT_SUNIT)) {
1786                         if (!value || !*value) {
1787                                 printk("XFS: %s option requires an argument\n",
1788                                         this_char);
1789                                 return EINVAL;
1790                         }
1791                         dsunit = simple_strtoul(value, &eov, 10);
1792                 } else if (!strcmp(this_char, MNTOPT_SWIDTH)) {
1793                         if (!value || !*value) {
1794                                 printk("XFS: %s option requires an argument\n",
1795                                         this_char);
1796                                 return EINVAL;
1797                         }
1798                         dswidth = simple_strtoul(value, &eov, 10);
1799                 } else if (!strcmp(this_char, MNTOPT_64BITINODE)) {
1800                         args->flags &= ~XFSMNT_32BITINODES;
1801 #if !XFS_BIG_INUMS
1802                         printk("XFS: %s option not allowed on this system\n",
1803                                 this_char);
1804                         return EINVAL;
1805 #endif
1806                 } else if (!strcmp(this_char, MNTOPT_NOUUID)) {
1807                         args->flags |= XFSMNT_NOUUID;
1808                 } else if (!strcmp(this_char, MNTOPT_BARRIER)) {
1809                         args->flags |= XFSMNT_BARRIER;
1810                 } else if (!strcmp(this_char, MNTOPT_IKEEP)) {
1811                         args->flags &= ~XFSMNT_IDELETE;
1812                 } else if (!strcmp(this_char, MNTOPT_NOIKEEP)) {
1813                         args->flags |= XFSMNT_IDELETE;
1814                 } else if (!strcmp(this_char, MNTOPT_LARGEIO)) {
1815                         args->flags2 &= ~XFSMNT2_COMPAT_IOSIZE;
1816                 } else if (!strcmp(this_char, MNTOPT_NOLARGEIO)) {
1817                         args->flags2 |= XFSMNT2_COMPAT_IOSIZE;
1818                 } else if (!strcmp(this_char, MNTOPT_ATTR2)) {
1819                         args->flags &= ~XFSMNT_COMPAT_ATTR;
1820                 } else if (!strcmp(this_char, MNTOPT_NOATTR2)) {
1821                         args->flags |= XFSMNT_COMPAT_ATTR;
1822                 } else if (!strcmp(this_char, "osyncisdsync")) {
1823                         /* no-op, this is now the default */
1824 printk("XFS: osyncisdsync is now the default, option is deprecated.\n");
1825                 } else if (!strcmp(this_char, "irixsgid")) {
1826 printk("XFS: irixsgid is now a sysctl(2) variable, option is deprecated.\n");
1827                 } else {
1828                         printk("XFS: unknown mount option [%s].\n", this_char);
1829                         return EINVAL;
1830                 }
1831         }
1832
1833         if (args->flags & XFSMNT_NORECOVERY) {
1834                 if ((vfsp->vfs_flag & VFS_RDONLY) == 0) {
1835                         printk("XFS: no-recovery mounts must be read-only.\n");
1836                         return EINVAL;
1837                 }
1838         }
1839
1840         if ((args->flags & XFSMNT_NOALIGN) && (dsunit || dswidth)) {
1841                 printk(
1842         "XFS: sunit and swidth options incompatible with the noalign option\n");
1843                 return EINVAL;
1844         }
1845
1846         if ((dsunit && !dswidth) || (!dsunit && dswidth)) {
1847                 printk("XFS: sunit and swidth must be specified together\n");
1848                 return EINVAL;
1849         }
1850
1851         if (dsunit && (dswidth % dsunit != 0)) {
1852                 printk(
1853         "XFS: stripe width (%d) must be a multiple of the stripe unit (%d)\n",
1854                         dswidth, dsunit);
1855                 return EINVAL;
1856         }
1857
1858         if ((args->flags & XFSMNT_NOALIGN) != XFSMNT_NOALIGN) {
1859                 if (dsunit) {
1860                         args->sunit = dsunit;
1861                         args->flags |= XFSMNT_RETERR;
1862                 } else {
1863                         args->sunit = vol_dsunit;
1864                 }
1865                 dswidth ? (args->swidth = dswidth) :
1866                           (args->swidth = vol_dswidth);
1867         } else {
1868                 args->sunit = args->swidth = 0;
1869         }
1870
1871         if (args->flags & XFSMNT_32BITINODES)
1872                 vfsp->vfs_flag |= VFS_32BITINODES;
1873
1874         if (args->flags2)
1875                 args->flags |= XFSMNT_FLAGS2;
1876         return 0;
1877 }
1878
1879 STATIC int
1880 xfs_showargs(
1881         struct bhv_desc         *bhv,
1882         struct seq_file         *m)
1883 {
1884         static struct proc_xfs_info {
1885                 int     flag;
1886                 char    *str;
1887         } xfs_info[] = {
1888                 /* the few simple ones we can get from the mount struct */
1889                 { XFS_MOUNT_WSYNC,              "," MNTOPT_WSYNC },
1890                 { XFS_MOUNT_INO64,              "," MNTOPT_INO64 },
1891                 { XFS_MOUNT_NOALIGN,            "," MNTOPT_NOALIGN },
1892                 { XFS_MOUNT_SWALLOC,            "," MNTOPT_SWALLOC },
1893                 { XFS_MOUNT_NOUUID,             "," MNTOPT_NOUUID },
1894                 { XFS_MOUNT_NORECOVERY,         "," MNTOPT_NORECOVERY },
1895                 { XFS_MOUNT_OSYNCISOSYNC,       "," MNTOPT_OSYNCISOSYNC },
1896                 { XFS_MOUNT_BARRIER,            "," MNTOPT_BARRIER },
1897                 { XFS_MOUNT_IDELETE,            "," MNTOPT_NOIKEEP },
1898                 { 0, NULL }
1899         };
1900         struct proc_xfs_info    *xfs_infop;
1901         struct xfs_mount        *mp = XFS_BHVTOM(bhv);
1902         struct vfs              *vfsp = XFS_MTOVFS(mp);
1903
1904         for (xfs_infop = xfs_info; xfs_infop->flag; xfs_infop++) {
1905                 if (mp->m_flags & xfs_infop->flag)
1906                         seq_puts(m, xfs_infop->str);
1907         }
1908
1909         if (mp->m_flags & XFS_MOUNT_IHASHSIZE)
1910                 seq_printf(m, "," MNTOPT_IHASHSIZE "=%d", mp->m_ihsize);
1911
1912         if (mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)
1913                 seq_printf(m, "," MNTOPT_ALLOCSIZE "=%d", 1<<mp->m_writeio_log);
1914
1915         if (mp->m_logbufs > 0)
1916                 seq_printf(m, "," MNTOPT_LOGBUFS "=%d", mp->m_logbufs);
1917
1918         if (mp->m_logbsize > 0)
1919                 seq_printf(m, "," MNTOPT_LOGBSIZE "=%d", mp->m_logbsize);
1920
1921         if (mp->m_logname)
1922                 seq_printf(m, "," MNTOPT_LOGDEV "=%s", mp->m_logname);
1923
1924         if (mp->m_rtname)
1925                 seq_printf(m, "," MNTOPT_RTDEV "=%s", mp->m_rtname);
1926
1927         if (mp->m_dalign > 0)
1928                 seq_printf(m, "," MNTOPT_SUNIT "=%d",
1929                                 (int)XFS_FSB_TO_BB(mp, mp->m_dalign));
1930
1931         if (mp->m_swidth > 0)
1932                 seq_printf(m, "," MNTOPT_SWIDTH "=%d",
1933                                 (int)XFS_FSB_TO_BB(mp, mp->m_swidth));
1934
1935         if (!(mp->m_flags & XFS_MOUNT_COMPAT_ATTR))
1936                 seq_printf(m, "," MNTOPT_ATTR2);
1937
1938         if (!(mp->m_flags & XFS_MOUNT_COMPAT_IOSIZE))
1939                 seq_printf(m, "," MNTOPT_LARGEIO);
1940
1941         if (!(vfsp->vfs_flag & VFS_32BITINODES))
1942                 seq_printf(m, "," MNTOPT_64BITINODE);
1943
1944         if (vfsp->vfs_flag & VFS_GRPID)
1945                 seq_printf(m, "," MNTOPT_GRPID);
1946
1947         return 0;
1948 }
1949
1950 STATIC void
1951 xfs_freeze(
1952         bhv_desc_t      *bdp)
1953 {
1954         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
1955
1956         while (atomic_read(&mp->m_active_trans) > 0)
1957                 delay(100);
1958
1959         /* Push the superblock and write an unmount record */
1960         xfs_log_unmount_write(mp);
1961         xfs_unmountfs_writesb(mp);
1962 }
1963
1964
1965 vfsops_t xfs_vfsops = {
1966         BHV_IDENTITY_INIT(VFS_BHV_XFS,VFS_POSITION_XFS),
1967         .vfs_parseargs          = xfs_parseargs,
1968         .vfs_showargs           = xfs_showargs,
1969         .vfs_mount              = xfs_mount,
1970         .vfs_unmount            = xfs_unmount,
1971         .vfs_mntupdate          = xfs_mntupdate,
1972         .vfs_root               = xfs_root,
1973         .vfs_statvfs            = xfs_statvfs,
1974         .vfs_sync               = xfs_sync,
1975         .vfs_vget               = xfs_vget,
1976         .vfs_dmapiops           = (vfs_dmapiops_t)fs_nosys,
1977         .vfs_quotactl           = (vfs_quotactl_t)fs_nosys,
1978         .vfs_init_vnode         = xfs_initialize_vnode,
1979         .vfs_force_shutdown     = xfs_do_force_shutdown,
1980         .vfs_freeze             = xfs_freeze,
1981 };