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_dir2.h"
  28 #include "xfs_dmapi.h"
  29 #include "xfs_mount.h"
  30 #include "xfs_da_btree.h"
  31 #include "xfs_bmap_btree.h"
  32 #include "xfs_ialloc_btree.h"
  33 #include "xfs_alloc_btree.h"
  34 #include "xfs_dir2_sf.h"
  35 #include "xfs_attr_sf.h"
  36 #include "xfs_dinode.h"
  37 #include "xfs_inode.h"
  38 #include "xfs_inode_item.h"
  39 #include "xfs_btree.h"
  40 #include "xfs_alloc.h"
  41 #include "xfs_ialloc.h"
  42 #include "xfs_quota.h"
  43 #include "xfs_error.h"
  44 #include "xfs_bmap.h"
  45 #include "xfs_rw.h"
  46 #include "xfs_refcache.h"
  47 #include "xfs_buf_item.h"
  48 #include "xfs_log_priv.h"
  49 #include "xfs_dir2_trace.h"
  50 #include "xfs_extfree_item.h"
  51 #include "xfs_acl.h"
  52 #include "xfs_attr.h"
  53 #include "xfs_clnt.h"
  54 #include "xfs_fsops.h"
  55
  56 STATIC int      xfs_sync(bhv_desc_t *, int, cred_t *);
  57
  58 int
  59 xfs_init(void)
  60 {
  61         extern kmem_zone_t      *xfs_bmap_free_item_zone;
  62         extern kmem_zone_t      *xfs_btree_cur_zone;
  63         extern kmem_zone_t      *xfs_trans_zone;
  64         extern kmem_zone_t      *xfs_buf_item_zone;
  65         extern kmem_zone_t      *xfs_dabuf_zone;
  66 #ifdef XFS_DABUF_DEBUG
  67         extern lock_t           xfs_dabuf_global_lock;
  68         spinlock_init(&xfs_dabuf_global_lock, "xfsda");
  69 #endif
  70
  71         /*
  72          * Initialize all of the zone allocators we use.
  73          */
  74         xfs_bmap_free_item_zone = kmem_zone_init(sizeof(xfs_bmap_free_item_t),
  75                                                  "xfs_bmap_free_item");
  76         xfs_btree_cur_zone = kmem_zone_init(sizeof(xfs_btree_cur_t),
  77                                             "xfs_btree_cur");
  78         xfs_trans_zone = kmem_zone_init(sizeof(xfs_trans_t), "xfs_trans");
  79         xfs_da_state_zone =
  80                 kmem_zone_init(sizeof(xfs_da_state_t), "xfs_da_state");
  81         xfs_dabuf_zone = kmem_zone_init(sizeof(xfs_dabuf_t), "xfs_dabuf");
  82         xfs_ifork_zone = kmem_zone_init(sizeof(xfs_ifork_t), "xfs_ifork");
  83         xfs_acl_zone_init(xfs_acl_zone, "xfs_acl");
  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 =
  96                 kmem_zone_init((sizeof(xfs_efd_log_item_t) +
  97                                ((XFS_EFD_MAX_FAST_EXTENTS - 1) *
  98                                  sizeof(xfs_extent_t))),
  99                                       "xfs_efd_item");
 100         xfs_efi_zone =
 101                 kmem_zone_init((sizeof(xfs_efi_log_item_t) +
 102                                ((XFS_EFI_MAX_FAST_EXTENTS - 1) *
 103                                  sizeof(xfs_extent_t))),
 104                                       "xfs_efi_item");
 105
 106         /*
 107          * These zones warrant special memory allocator hints
 108          */
 109         xfs_inode_zone =
 110                 kmem_zone_init_flags(sizeof(xfs_inode_t), "xfs_inode",
 111                                         KM_ZONE_HWALIGN | KM_ZONE_RECLAIM |
 112                                         KM_ZONE_SPREAD, NULL);
 113         xfs_ili_zone =
 114                 kmem_zone_init_flags(sizeof(xfs_inode_log_item_t), "xfs_ili",
 115                                         KM_ZONE_SPREAD, NULL);
 116         xfs_chashlist_zone =
 117                 kmem_zone_init_flags(sizeof(xfs_chashlist_t), "xfs_chashlist",
 118                                         KM_ZONE_SPREAD, NULL);
 119
 120         /*
 121          * Allocate global trace buffers.
 122          */
 123 #ifdef XFS_ALLOC_TRACE
 124         xfs_alloc_trace_buf = ktrace_alloc(XFS_ALLOC_TRACE_SIZE, KM_SLEEP);
 125 #endif
 126 #ifdef XFS_BMAP_TRACE
 127         xfs_bmap_trace_buf = ktrace_alloc(XFS_BMAP_TRACE_SIZE, KM_SLEEP);
 128 #endif
 129 #ifdef XFS_BMBT_TRACE
 130         xfs_bmbt_trace_buf = ktrace_alloc(XFS_BMBT_TRACE_SIZE, KM_SLEEP);
 131 #endif
 132 #ifdef XFS_ATTR_TRACE
 133         xfs_attr_trace_buf = ktrace_alloc(XFS_ATTR_TRACE_SIZE, KM_SLEEP);
 134 #endif
 135 #ifdef XFS_DIR2_TRACE
 136         xfs_dir2_trace_buf = ktrace_alloc(XFS_DIR2_GTRACE_SIZE, KM_SLEEP);
 137 #endif
 138
 139         xfs_dir_startup();
 140
 141 #if (defined(DEBUG) || defined(INDUCE_IO_ERROR))
 142         xfs_error_test_init();
 143 #endif /* DEBUG || INDUCE_IO_ERROR */
 144
 145         xfs_init_procfs();
 146         xfs_sysctl_register();
 147         return 0;
 148 }
 149
 150 void
 151 xfs_cleanup(void)
 152 {
 153         extern kmem_zone_t      *xfs_bmap_free_item_zone;
 154         extern kmem_zone_t      *xfs_btree_cur_zone;
 155         extern kmem_zone_t      *xfs_inode_zone;
 156         extern kmem_zone_t      *xfs_trans_zone;
 157         extern kmem_zone_t      *xfs_da_state_zone;
 158         extern kmem_zone_t      *xfs_dabuf_zone;
 159         extern kmem_zone_t      *xfs_efd_zone;
 160         extern kmem_zone_t      *xfs_efi_zone;
 161         extern kmem_zone_t      *xfs_buf_item_zone;
 162         extern kmem_zone_t      *xfs_chashlist_zone;
 163
 164         xfs_cleanup_procfs();
 165         xfs_sysctl_unregister();
 166         xfs_refcache_destroy();
 167         xfs_acl_zone_destroy(xfs_acl_zone);
 168
 169 #ifdef XFS_DIR2_TRACE
 170         ktrace_free(xfs_dir2_trace_buf);
 171 #endif
 172 #ifdef XFS_ATTR_TRACE
 173         ktrace_free(xfs_attr_trace_buf);
 174 #endif
 175 #ifdef XFS_BMBT_TRACE
 176         ktrace_free(xfs_bmbt_trace_buf);
 177 #endif
 178 #ifdef XFS_BMAP_TRACE
 179         ktrace_free(xfs_bmap_trace_buf);
 180 #endif
 181 #ifdef XFS_ALLOC_TRACE
 182         ktrace_free(xfs_alloc_trace_buf);
 183 #endif
 184
 185         kmem_zone_destroy(xfs_bmap_free_item_zone);
 186         kmem_zone_destroy(xfs_btree_cur_zone);
 187         kmem_zone_destroy(xfs_inode_zone);
 188         kmem_zone_destroy(xfs_trans_zone);
 189         kmem_zone_destroy(xfs_da_state_zone);
 190         kmem_zone_destroy(xfs_dabuf_zone);
 191         kmem_zone_destroy(xfs_buf_item_zone);
 192         kmem_zone_destroy(xfs_efd_zone);
 193         kmem_zone_destroy(xfs_efi_zone);
 194         kmem_zone_destroy(xfs_ifork_zone);
 195         kmem_zone_destroy(xfs_ili_zone);
 196         kmem_zone_destroy(xfs_chashlist_zone);
 197 }
 198
 199 /*
 200  * xfs_start_flags
 201  *
 202  * This function fills in xfs_mount_t fields based on mount args.
 203  * Note: the superblock has _not_ yet been read in.
 204  */
 205 STATIC int
 206 xfs_start_flags(
 207         struct bhv_vfs          *vfs,
 208         struct xfs_mount_args   *ap,
 209         struct xfs_mount        *mp)
 210 {
 211         /* Values are in BBs */
 212         if ((ap->flags & XFSMNT_NOALIGN) != XFSMNT_NOALIGN) {
 213                 /*
 214                  * At this point the superblock has not been read
 215                  * in, therefore we do not know the block size.
 216                  * Before the mount call ends we will convert
 217                  * these to FSBs.
 218                  */
 219                 mp->m_dalign = ap->sunit;
 220                 mp->m_swidth = ap->swidth;
 221         }
 222
 223         if (ap->logbufs != -1 &&
 224             ap->logbufs != 0 &&
 225             (ap->logbufs < XLOG_MIN_ICLOGS ||
 226              ap->logbufs > XLOG_MAX_ICLOGS)) {
 227                 cmn_err(CE_WARN,
 228                         "XFS: invalid logbufs value: %d [not %d-%d]",
 229                         ap->logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
 230                 return XFS_ERROR(EINVAL);
 231         }
 232         mp->m_logbufs = ap->logbufs;
 233         if (ap->logbufsize != -1 &&
 234             ap->logbufsize !=  0 &&
 235             ap->logbufsize != 16 * 1024 &&
 236             ap->logbufsize != 32 * 1024 &&
 237             ap->logbufsize != 64 * 1024 &&
 238             ap->logbufsize != 128 * 1024 &&
 239             ap->logbufsize != 256 * 1024) {
 240                 cmn_err(CE_WARN,
 241         "XFS: invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
 242                         ap->logbufsize);
 243                 return XFS_ERROR(EINVAL);
 244         }
 245         mp->m_ihsize = ap->ihashsize;
 246         mp->m_logbsize = ap->logbufsize;
 247         mp->m_fsname_len = strlen(ap->fsname) + 1;
 248         mp->m_fsname = kmem_alloc(mp->m_fsname_len, KM_SLEEP);
 249         strcpy(mp->m_fsname, ap->fsname);
 250         if (ap->rtname[0]) {
 251                 mp->m_rtname = kmem_alloc(strlen(ap->rtname) + 1, KM_SLEEP);
 252                 strcpy(mp->m_rtname, ap->rtname);
 253         }
 254         if (ap->logname[0]) {
 255                 mp->m_logname = kmem_alloc(strlen(ap->logname) + 1, KM_SLEEP);
 256                 strcpy(mp->m_logname, ap->logname);
 257         }
 258
 259         if (ap->flags & XFSMNT_WSYNC)
 260                 mp->m_flags |= XFS_MOUNT_WSYNC;
 261 #if XFS_BIG_INUMS
 262         if (ap->flags & XFSMNT_INO64) {
 263                 mp->m_flags |= XFS_MOUNT_INO64;
 264                 mp->m_inoadd = XFS_INO64_OFFSET;
 265         }
 266 #endif
 267         if (ap->flags & XFSMNT_RETERR)
 268                 mp->m_flags |= XFS_MOUNT_RETERR;
 269         if (ap->flags & XFSMNT_NOALIGN)
 270                 mp->m_flags |= XFS_MOUNT_NOALIGN;
 271         if (ap->flags & XFSMNT_SWALLOC)
 272                 mp->m_flags |= XFS_MOUNT_SWALLOC;
 273         if (ap->flags & XFSMNT_OSYNCISOSYNC)
 274                 mp->m_flags |= XFS_MOUNT_OSYNCISOSYNC;
 275         if (ap->flags & XFSMNT_32BITINODES)
 276                 mp->m_flags |= XFS_MOUNT_32BITINODES;
 277
 278         if (ap->flags & XFSMNT_IOSIZE) {
 279                 if (ap->iosizelog > XFS_MAX_IO_LOG ||
 280                     ap->iosizelog < XFS_MIN_IO_LOG) {
 281                         cmn_err(CE_WARN,
 282                 "XFS: invalid log iosize: %d [not %d-%d]",
 283                                 ap->iosizelog, XFS_MIN_IO_LOG,
 284                                 XFS_MAX_IO_LOG);
 285                         return XFS_ERROR(EINVAL);
 286                 }
 287
 288                 mp->m_flags |= XFS_MOUNT_DFLT_IOSIZE;
 289                 mp->m_readio_log = mp->m_writeio_log = ap->iosizelog;
 290         }
 291
 292         if (ap->flags & XFSMNT_IHASHSIZE)
 293                 mp->m_flags |= XFS_MOUNT_IHASHSIZE;
 294         if (ap->flags & XFSMNT_IDELETE)
 295                 mp->m_flags |= XFS_MOUNT_IDELETE;
 296         if (ap->flags & XFSMNT_DIRSYNC)
 297                 mp->m_flags |= XFS_MOUNT_DIRSYNC;
 298         if (ap->flags & XFSMNT_ATTR2)
 299                 mp->m_flags |= XFS_MOUNT_ATTR2;
 300
 301         if (ap->flags2 & XFSMNT2_COMPAT_IOSIZE)
 302                 mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE;
 303
 304         /*
 305          * no recovery flag requires a read-only mount
 306          */
 307         if (ap->flags & XFSMNT_NORECOVERY) {
 308                 if (!(vfs->vfs_flag & VFS_RDONLY)) {
 309                         cmn_err(CE_WARN,
 310         "XFS: tried to mount a FS read-write without recovery!");
 311                         return XFS_ERROR(EINVAL);
 312                 }
 313                 mp->m_flags |= XFS_MOUNT_NORECOVERY;
 314         }
 315
 316         if (ap->flags & XFSMNT_NOUUID)
 317                 mp->m_flags |= XFS_MOUNT_NOUUID;
 318         if (ap->flags & XFSMNT_BARRIER)
 319                 mp->m_flags |= XFS_MOUNT_BARRIER;
 320         else
 321                 mp->m_flags &= ~XFS_MOUNT_BARRIER;
 322
 323         return 0;
 324 }
 325
 326 /*
 327  * This function fills in xfs_mount_t fields based on mount args.
 328  * Note: the superblock _has_ now been read in.
 329  */
 330 STATIC int
 331 xfs_finish_flags(
 332         struct bhv_vfs          *vfs,
 333         struct xfs_mount_args   *ap,
 334         struct xfs_mount        *mp)
 335 {
 336         int                     ronly = (vfs->vfs_flag & VFS_RDONLY);
 337
 338         /* Fail a mount where the logbuf is smaller then the log stripe */
 339         if (XFS_SB_VERSION_HASLOGV2(&mp->m_sb)) {
 340                 if ((ap->logbufsize <= 0) &&
 341                     (mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE)) {
 342                         mp->m_logbsize = mp->m_sb.sb_logsunit;
 343                 } else if (ap->logbufsize > 0 &&
 344                            ap->logbufsize < mp->m_sb.sb_logsunit) {
 345                         cmn_err(CE_WARN,
 346         "XFS: logbuf size must be greater than or equal to log stripe size");
 347                         return XFS_ERROR(EINVAL);
 348                 }
 349         } else {
 350                 /* Fail a mount if the logbuf is larger than 32K */
 351                 if (ap->logbufsize > XLOG_BIG_RECORD_BSIZE) {
 352                         cmn_err(CE_WARN,
 353         "XFS: logbuf size for version 1 logs must be 16K or 32K");
 354                         return XFS_ERROR(EINVAL);
 355                 }
 356         }
 357
 358         if (XFS_SB_VERSION_HASATTR2(&mp->m_sb)) {
 359                 mp->m_flags |= XFS_MOUNT_ATTR2;
 360         }
 361
 362         /*
 363          * prohibit r/w mounts of read-only filesystems
 364          */
 365         if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !ronly) {
 366                 cmn_err(CE_WARN,
 367         "XFS: cannot mount a read-only filesystem as read-write");
 368                 return XFS_ERROR(EROFS);
 369         }
 370
 371         /*
 372          * check for shared mount.
 373          */
 374         if (ap->flags & XFSMNT_SHARED) {
 375                 if (!XFS_SB_VERSION_HASSHARED(&mp->m_sb))
 376                         return XFS_ERROR(EINVAL);
 377
 378                 /*
 379                  * For IRIX 6.5, shared mounts must have the shared
 380                  * version bit set, have the persistent readonly
 381                  * field set, must be version 0 and can only be mounted
 382                  * read-only.
 383                  */
 384                 if (!ronly || !(mp->m_sb.sb_flags & XFS_SBF_READONLY) ||
 385                      (mp->m_sb.sb_shared_vn != 0))
 386                         return XFS_ERROR(EINVAL);
 387
 388                 mp->m_flags |= XFS_MOUNT_SHARED;
 389
 390                 /*
 391                  * Shared XFS V0 can't deal with DMI.  Return EINVAL.
 392                  */
 393                 if (mp->m_sb.sb_shared_vn == 0 && (ap->flags & XFSMNT_DMAPI))
 394                         return XFS_ERROR(EINVAL);
 395         }
 396
 397         return 0;
 398 }
 399
 400 /*
 401  * xfs_mount
 402  *
 403  * The file system configurations are:
 404  *      (1) device (partition) with data and internal log
 405  *      (2) logical volume with data and log subvolumes.
 406  *      (3) logical volume with data, log, and realtime subvolumes.
 407  *
 408  * We only have to handle opening the log and realtime volumes here if
 409  * they are present.  The data subvolume has already been opened by
 410  * get_sb_bdev() and is stored in vfsp->vfs_super->s_bdev.
 411  */
 412 STATIC int
 413 xfs_mount(
 414         struct bhv_desc         *bhvp,
 415         struct xfs_mount_args   *args,
 416         cred_t                  *credp)
 417 {
 418         struct bhv_vfs          *vfsp = bhvtovfs(bhvp);
 419         struct bhv_desc         *p;
 420         struct xfs_mount        *mp = XFS_BHVTOM(bhvp);
 421         struct block_device     *ddev, *logdev, *rtdev;
 422         int                     flags = 0, error;
 423
 424         ddev = vfsp->vfs_super->s_bdev;
 425         logdev = rtdev = NULL;
 426
 427         /*
 428          * Setup xfs_mount function vectors from available behaviors
 429          */
 430         p = vfs_bhv_lookup(vfsp, VFS_POSITION_DM);
 431         mp->m_dm_ops = p ? *(xfs_dmops_t *) vfs_bhv_custom(p) : xfs_dmcore_stub;
 432         p = vfs_bhv_lookup(vfsp, VFS_POSITION_QM);
 433         mp->m_qm_ops = p ? *(xfs_qmops_t *) vfs_bhv_custom(p) : xfs_qmcore_stub;
 434         p = vfs_bhv_lookup(vfsp, VFS_POSITION_IO);
 435         mp->m_io_ops = p ? *(xfs_ioops_t *) vfs_bhv_custom(p) : xfs_iocore_xfs;
 436
 437         if (args->flags & XFSMNT_QUIET)
 438                 flags |= XFS_MFSI_QUIET;
 439
 440         /*
 441          * Open real time and log devices - order is important.
 442          */
 443         if (args->logname[0]) {
 444                 error = xfs_blkdev_get(mp, args->logname, &logdev);
 445                 if (error)
 446                         return error;
 447         }
 448         if (args->rtname[0]) {
 449                 error = xfs_blkdev_get(mp, args->rtname, &rtdev);
 450                 if (error) {
 451                         xfs_blkdev_put(logdev);
 452                         return error;
 453                 }
 454
 455                 if (rtdev == ddev || rtdev == logdev) {
 456                         cmn_err(CE_WARN,
 457         "XFS: Cannot mount filesystem with identical rtdev and ddev/logdev.");
 458                         xfs_blkdev_put(logdev);
 459                         xfs_blkdev_put(rtdev);
 460                         return EINVAL;
 461                 }
 462         }
 463
 464         /*
 465          * Setup xfs_mount buffer target pointers
 466          */
 467         error = ENOMEM;
 468         mp->m_ddev_targp = xfs_alloc_buftarg(ddev, 0);
 469         if (!mp->m_ddev_targp) {
 470                 xfs_blkdev_put(logdev);
 471                 xfs_blkdev_put(rtdev);
 472                 return error;
 473         }
 474         if (rtdev) {
 475                 mp->m_rtdev_targp = xfs_alloc_buftarg(rtdev, 1);
 476                 if (!mp->m_rtdev_targp)
 477                         goto error0;
 478         }
 479         mp->m_logdev_targp = (logdev && logdev != ddev) ?
 480                                 xfs_alloc_buftarg(logdev, 1) : mp->m_ddev_targp;
 481         if (!mp->m_logdev_targp)
 482                 goto error0;
 483
 484         /*
 485          * Setup flags based on mount(2) options and then the superblock
 486          */
 487         error = xfs_start_flags(vfsp, args, mp);
 488         if (error)
 489                 goto error1;
 490         error = xfs_readsb(mp, flags);
 491         if (error)
 492                 goto error1;
 493         error = xfs_finish_flags(vfsp, args, mp);
 494         if (error)
 495                 goto error2;
 496
 497         /*
 498          * Setup xfs_mount buffer target pointers based on superblock
 499          */
 500         error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_blocksize,
 501                                     mp->m_sb.sb_sectsize);
 502         if (!error && logdev && logdev != ddev) {
 503                 unsigned int    log_sector_size = BBSIZE;
 504
 505                 if (XFS_SB_VERSION_HASSECTOR(&mp->m_sb))
 506                         log_sector_size = mp->m_sb.sb_logsectsize;
 507                 error = xfs_setsize_buftarg(mp->m_logdev_targp,
 508                                             mp->m_sb.sb_blocksize,
 509                                             log_sector_size);
 510         }
 511         if (!error && rtdev)
 512                 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
 513                                             mp->m_sb.sb_blocksize,
 514                                             mp->m_sb.sb_sectsize);
 515         if (error)
 516                 goto error2;
 517
 518         if (mp->m_flags & XFS_MOUNT_BARRIER)
 519                 xfs_mountfs_check_barriers(mp);
 520
 521         error = XFS_IOINIT(vfsp, args, flags);
 522         if (error)
 523                 goto error2;
 524
 525         return 0;
 526
 527 error2:
 528         if (mp->m_sb_bp)
 529                 xfs_freesb(mp);
 530 error1:
 531         xfs_binval(mp->m_ddev_targp);
 532         if (logdev && logdev != ddev)
 533                 xfs_binval(mp->m_logdev_targp);
 534         if (rtdev)
 535                 xfs_binval(mp->m_rtdev_targp);
 536 error0:
 537         xfs_unmountfs_close(mp, credp);
 538         return error;
 539 }
 540
 541 STATIC int
 542 xfs_unmount(
 543         bhv_desc_t      *bdp,
 544         int             flags,
 545         cred_t          *credp)
 546 {
 547         bhv_vfs_t       *vfsp = bhvtovfs(bdp);
 548         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
 549         xfs_inode_t     *rip;
 550         bhv_vnode_t     *rvp;
 551         int             unmount_event_wanted = 0;
 552         int             unmount_event_flags = 0;
 553         int             xfs_unmountfs_needed = 0;
 554         int             error;
 555
 556         rip = mp->m_rootip;
 557         rvp = XFS_ITOV(rip);
 558
 559         if (vfsp->vfs_flag & VFS_DMI) {
 560                 error = XFS_SEND_PREUNMOUNT(mp, vfsp,
 561                                 rvp, DM_RIGHT_NULL, rvp, DM_RIGHT_NULL,
 562                                 NULL, NULL, 0, 0,
 563                                 (mp->m_dmevmask & (1<<DM_EVENT_PREUNMOUNT))?
 564                                         0:DM_FLAGS_UNWANTED);
 565                         if (error)
 566                                 return XFS_ERROR(error);
 567                 unmount_event_wanted = 1;
 568                 unmount_event_flags = (mp->m_dmevmask & (1<<DM_EVENT_UNMOUNT))?
 569                                         0 : DM_FLAGS_UNWANTED;
 570         }
 571
 572         /*
 573          * First blow any referenced inode from this file system
 574          * out of the reference cache, and delete the timer.
 575          */
 576         xfs_refcache_purge_mp(mp);
 577
 578         XFS_bflush(mp->m_ddev_targp);
 579         error = xfs_unmount_flush(mp, 0);
 580         if (error)
 581                 goto out;
 582
 583         ASSERT(vn_count(rvp) == 1);
 584
 585         /*
 586          * Drop the reference count
 587          */
 588         VN_RELE(rvp);
 589
 590         /*
 591          * If we're forcing a shutdown, typically because of a media error,
 592          * we want to make sure we invalidate dirty pages that belong to
 593          * referenced vnodes as well.
 594          */
 595         if (XFS_FORCED_SHUTDOWN(mp)) {
 596                 error = xfs_sync(&mp->m_bhv,
 597                          (SYNC_WAIT | SYNC_CLOSE), credp);
 598                 ASSERT(error != EFSCORRUPTED);
 599         }
 600         xfs_unmountfs_needed = 1;
 601
 602 out:
 603         /*      Send DMAPI event, if required.
 604          *      Then do xfs_unmountfs() if needed.
 605          *      Then return error (or zero).
 606          */
 607         if (unmount_event_wanted) {
 608                 /* Note: mp structure must still exist for
 609                  * XFS_SEND_UNMOUNT() call.
 610                  */
 611                 XFS_SEND_UNMOUNT(mp, vfsp, error == 0 ? rvp : NULL,
 612                         DM_RIGHT_NULL, 0, error, unmount_event_flags);
 613         }
 614         if (xfs_unmountfs_needed) {
 615                 /*
 616                  * Call common unmount function to flush to disk
 617                  * and free the super block buffer & mount structures.
 618                  */
 619                 xfs_unmountfs(mp, credp);
 620         }
 621
 622         return XFS_ERROR(error);
 623 }
 624
 625 STATIC int
 626 xfs_quiesce_fs(
 627         xfs_mount_t             *mp)
 628 {
 629         int                     count = 0, pincount;
 630
 631         xfs_refcache_purge_mp(mp);
 632         xfs_flush_buftarg(mp->m_ddev_targp, 0);
 633         xfs_finish_reclaim_all(mp, 0);
 634
 635         /* This loop must run at least twice.
 636          * The first instance of the loop will flush
 637          * most meta data but that will generate more
 638          * meta data (typically directory updates).
 639          * Which then must be flushed and logged before
 640          * we can write the unmount record.
 641          */
 642         do {
 643                 xfs_syncsub(mp, SYNC_REMOUNT|SYNC_ATTR|SYNC_WAIT, 0, NULL);
 644                 pincount = xfs_flush_buftarg(mp->m_ddev_targp, 1);
 645                 if (!pincount) {
 646                         delay(50);
 647                         count++;
 648                 }
 649         } while (count < 2);
 650
 651         return 0;
 652 }
 653
 654 STATIC int
 655 xfs_mntupdate(
 656         bhv_desc_t                      *bdp,
 657         int                             *flags,
 658         struct xfs_mount_args           *args)
 659 {
 660         bhv_vfs_t       *vfsp = bhvtovfs(bdp);
 661         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
 662
 663         if (!(*flags & MS_RDONLY)) {                    /* rw/ro -> rw */
 664                 if (vfsp->vfs_flag & VFS_RDONLY)
 665                         vfsp->vfs_flag &= ~VFS_RDONLY;
 666                 if (args->flags & XFSMNT_BARRIER) {
 667                         mp->m_flags |= XFS_MOUNT_BARRIER;
 668                         xfs_mountfs_check_barriers(mp);
 669                 } else {
 670                         mp->m_flags &= ~XFS_MOUNT_BARRIER;
 671                 }
 672         } else if (!(vfsp->vfs_flag & VFS_RDONLY)) {    /* rw -> ro */
 673                 bhv_vfs_sync(vfsp, SYNC_FSDATA|SYNC_BDFLUSH|SYNC_ATTR, NULL);
 674                 xfs_quiesce_fs(mp);
 675                 xfs_log_unmount_write(mp);
 676                 xfs_unmountfs_writesb(mp);
 677                 vfsp->vfs_flag |= VFS_RDONLY;
 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         bhv_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         bhv_vnode_t     **vpp)
 776 {
 777         bhv_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         bhv_statvfs_t   *statp,
 796         bhv_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         xfs_icsb_sync_counters_lazy(mp);
 810         s = XFS_SB_LOCK(mp);
 811         statp->f_bsize = sbp->sb_blocksize;
 812         lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
 813         statp->f_blocks = sbp->sb_dblocks - lsize;
 814         statp->f_bfree = statp->f_bavail = sbp->sb_fdblocks;
 815         fakeinos = statp->f_bfree << sbp->sb_inopblog;
 816 #if XFS_BIG_INUMS
 817         fakeinos += mp->m_inoadd;
 818 #endif
 819         statp->f_files =
 820             MIN(sbp->sb_icount + fakeinos, (__uint64_t)XFS_MAXINUMBER);
 821         if (mp->m_maxicount)
 822 #if XFS_BIG_INUMS
 823                 if (!mp->m_inoadd)
 824 #endif
 825                         statp->f_files = min_t(typeof(statp->f_files),
 826                                                 statp->f_files,
 827                                                 mp->m_maxicount);
 828         statp->f_ffree = statp->f_files - (sbp->sb_icount - sbp->sb_ifree);
 829         XFS_SB_UNLOCK(mp, s);
 830
 831         xfs_statvfs_fsid(statp, mp);
 832         statp->f_namelen = MAXNAMELEN - 1;
 833
 834         return 0;
 835 }
 836
 837
 838 /*
 839  * xfs_sync flushes any pending I/O to file system vfsp.
 840  *
 841  * This routine is called by vfs_sync() to make sure that things make it
 842  * out to disk eventually, on sync() system calls to flush out everything,
 843  * and when the file system is unmounted.  For the vfs_sync() case, all
 844  * we really need to do is sync out the log to make all of our meta-data
 845  * updates permanent (except for timestamps).  For calls from pflushd(),
 846  * dirty pages are kept moving by calling pdflush() on the inodes
 847  * containing them.  We also flush the inodes that we can lock without
 848  * sleeping and the superblock if we can lock it without sleeping from
 849  * vfs_sync() so that items at the tail of the log are always moving out.
 850  *
 851  * Flags:
 852  *      SYNC_BDFLUSH - We're being called from vfs_sync() so we don't want
 853  *                     to sleep if we can help it.  All we really need
 854  *                     to do is ensure that the log is synced at least
 855  *                     periodically.  We also push the inodes and
 856  *                     superblock if we can lock them without sleeping
 857  *                      and they are not pinned.
 858  *      SYNC_ATTR    - We need to flush the inodes.  If SYNC_BDFLUSH is not
 859  *                     set, then we really want to lock each inode and flush
 860  *                     it.
 861  *      SYNC_WAIT    - All the flushes that take place in this call should
 862  *                     be synchronous.
 863  *      SYNC_DELWRI  - This tells us to push dirty pages associated with
 864  *                     inodes.  SYNC_WAIT and SYNC_BDFLUSH are used to
 865  *                     determine if they should be flushed sync, async, or
 866  *                     delwri.
 867  *      SYNC_CLOSE   - This flag is passed when the system is being
 868  *                     unmounted.  We should sync and invalidate everything.
 869  *      SYNC_FSDATA  - This indicates that the caller would like to make
 870  *                     sure the superblock is safe on disk.  We can ensure
 871  *                     this by simply making sure the log gets flushed
 872  *                     if SYNC_BDFLUSH is set, and by actually writing it
 873  *                     out otherwise.
 874  *
 875  */
 876 /*ARGSUSED*/
 877 STATIC int
 878 xfs_sync(
 879         bhv_desc_t      *bdp,
 880         int             flags,
 881         cred_t          *credp)
 882 {
 883         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
 884
 885         if (unlikely(flags == SYNC_QUIESCE))
 886                 return xfs_quiesce_fs(mp);
 887         else
 888                 return xfs_syncsub(mp, flags, 0, NULL);
 889 }
 890
 891 /*
 892  * xfs sync routine for internal use
 893  *
 894  * This routine supports all of the flags defined for the generic vfs_sync
 895  * interface as explained above under xfs_sync.  In the interests of not
 896  * changing interfaces within the 6.5 family, additional internally-
 897  * required functions are specified within a separate xflags parameter,
 898  * only available by calling this routine.
 899  *
 900  */
 901 int
 902 xfs_sync_inodes(
 903         xfs_mount_t     *mp,
 904         int             flags,
 905         int             xflags,
 906         int             *bypassed)
 907 {
 908         xfs_inode_t     *ip = NULL;
 909         xfs_inode_t     *ip_next;
 910         xfs_buf_t       *bp;
 911         bhv_vnode_t     *vp = NULL;
 912         int             error;
 913         int             last_error;
 914         uint64_t        fflag;
 915         uint            lock_flags;
 916         uint            base_lock_flags;
 917         boolean_t       mount_locked;
 918         boolean_t       vnode_refed;
 919         int             preempt;
 920         xfs_dinode_t    *dip;
 921         xfs_iptr_t      *ipointer;
 922 #ifdef DEBUG
 923         boolean_t       ipointer_in = B_FALSE;
 924
 925 #define IPOINTER_SET    ipointer_in = B_TRUE
 926 #define IPOINTER_CLR    ipointer_in = B_FALSE
 927 #else
 928 #define IPOINTER_SET
 929 #define IPOINTER_CLR
 930 #endif
 931
 932
 933 /* Insert a marker record into the inode list after inode ip. The list
 934  * must be locked when this is called. After the call the list will no
 935  * longer be locked.
 936  */
 937 #define IPOINTER_INSERT(ip, mp) { \
 938                 ASSERT(ipointer_in == B_FALSE); \
 939                 ipointer->ip_mnext = ip->i_mnext; \
 940                 ipointer->ip_mprev = ip; \
 941                 ip->i_mnext = (xfs_inode_t *)ipointer; \
 942                 ipointer->ip_mnext->i_mprev = (xfs_inode_t *)ipointer; \
 943                 preempt = 0; \
 944                 XFS_MOUNT_IUNLOCK(mp); \
 945                 mount_locked = B_FALSE; \
 946                 IPOINTER_SET; \
 947         }
 948
 949 /* Remove the marker from the inode list. If the marker was the only item
 950  * in the list then there are no remaining inodes and we should zero out
 951  * the whole list. If we are the current head of the list then move the head
 952  * past us.
 953  */
 954 #define IPOINTER_REMOVE(ip, mp) { \
 955                 ASSERT(ipointer_in == B_TRUE); \
 956                 if (ipointer->ip_mnext != (xfs_inode_t *)ipointer) { \
 957                         ip = ipointer->ip_mnext; \
 958                         ip->i_mprev = ipointer->ip_mprev; \
 959                         ipointer->ip_mprev->i_mnext = ip; \
 960                         if (mp->m_inodes == (xfs_inode_t *)ipointer) { \
 961                                 mp->m_inodes = ip; \
 962                         } \
 963                 } else { \
 964                         ASSERT(mp->m_inodes == (xfs_inode_t *)ipointer); \
 965                         mp->m_inodes = NULL; \
 966                         ip = NULL; \
 967                 } \
 968                 IPOINTER_CLR; \
 969         }
 970
 971 #define XFS_PREEMPT_MASK        0x7f
 972
 973         if (bypassed)
 974                 *bypassed = 0;
 975         if (XFS_MTOVFS(mp)->vfs_flag & VFS_RDONLY)
 976                 return 0;
 977         error = 0;
 978         last_error = 0;
 979         preempt = 0;
 980
 981         /* Allocate a reference marker */
 982         ipointer = (xfs_iptr_t *)kmem_zalloc(sizeof(xfs_iptr_t), KM_SLEEP);
 983
 984         fflag = XFS_B_ASYNC;            /* default is don't wait */
 985         if (flags & (SYNC_BDFLUSH | SYNC_DELWRI))
 986                 fflag = XFS_B_DELWRI;
 987         if (flags & SYNC_WAIT)
 988                 fflag = 0;              /* synchronous overrides all */
 989
 990         base_lock_flags = XFS_ILOCK_SHARED;
 991         if (flags & (SYNC_DELWRI | SYNC_CLOSE)) {
 992                 /*
 993                  * We need the I/O lock if we're going to call any of
 994                  * the flush/inval routines.
 995                  */
 996                 base_lock_flags |= XFS_IOLOCK_SHARED;
 997         }
 998
 999         XFS_MOUNT_ILOCK(mp);
1000
1001         ip = mp->m_inodes;
1002
1003         mount_locked = B_TRUE;
1004         vnode_refed  = B_FALSE;
1005
1006         IPOINTER_CLR;
1007
1008         do {
1009                 ASSERT(ipointer_in == B_FALSE);
1010                 ASSERT(vnode_refed == B_FALSE);
1011
1012                 lock_flags = base_lock_flags;
1013
1014                 /*
1015                  * There were no inodes in the list, just break out
1016                  * of the loop.
1017                  */
1018                 if (ip == NULL) {
1019                         break;
1020                 }
1021
1022                 /*
1023                  * We found another sync thread marker - skip it
1024                  */
1025                 if (ip->i_mount == NULL) {
1026                         ip = ip->i_mnext;
1027                         continue;
1028                 }
1029
1030                 vp = XFS_ITOV_NULL(ip);
1031
1032                 /*
1033                  * If the vnode is gone then this is being torn down,
1034                  * call reclaim if it is flushed, else let regular flush
1035                  * code deal with it later in the loop.
1036                  */
1037
1038                 if (vp == NULL) {
1039                         /* Skip ones already in reclaim */
1040                         if (ip->i_flags & XFS_IRECLAIM) {
1041                                 ip = ip->i_mnext;
1042                                 continue;
1043                         }
1044                         if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0) {
1045                                 ip = ip->i_mnext;
1046                         } else if ((xfs_ipincount(ip) == 0) &&
1047                                     xfs_iflock_nowait(ip)) {
1048                                 IPOINTER_INSERT(ip, mp);
1049
1050                                 xfs_finish_reclaim(ip, 1,
1051                                                 XFS_IFLUSH_DELWRI_ELSE_ASYNC);
1052
1053                                 XFS_MOUNT_ILOCK(mp);
1054                                 mount_locked = B_TRUE;
1055                                 IPOINTER_REMOVE(ip, mp);
1056                         } else {
1057                                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1058                                 ip = ip->i_mnext;
1059                         }
1060                         continue;
1061                 }
1062
1063                 if (VN_BAD(vp)) {
1064                         ip = ip->i_mnext;
1065                         continue;
1066                 }
1067
1068                 if (XFS_FORCED_SHUTDOWN(mp) && !(flags & SYNC_CLOSE)) {
1069                         XFS_MOUNT_IUNLOCK(mp);
1070                         kmem_free(ipointer, sizeof(xfs_iptr_t));
1071                         return 0;
1072                 }
1073
1074                 /*
1075                  * If this is just vfs_sync() or pflushd() calling
1076                  * then we can skip inodes for which it looks like
1077                  * there is nothing to do.  Since we don't have the
1078                  * inode locked this is racy, but these are periodic
1079                  * calls so it doesn't matter.  For the others we want
1080                  * to know for sure, so we at least try to lock them.
1081                  */
1082                 if (flags & SYNC_BDFLUSH) {
1083                         if (((ip->i_itemp == NULL) ||
1084                              !(ip->i_itemp->ili_format.ilf_fields &
1085                                XFS_ILOG_ALL)) &&
1086                             (ip->i_update_core == 0)) {
1087                                 ip = ip->i_mnext;
1088                                 continue;
1089                         }
1090                 }
1091
1092                 /*
1093                  * Try to lock without sleeping.  We're out of order with
1094                  * the inode list lock here, so if we fail we need to drop
1095                  * the mount lock and try again.  If we're called from
1096                  * bdflush() here, then don't bother.
1097                  *
1098                  * The inode lock here actually coordinates with the
1099                  * almost spurious inode lock in xfs_ireclaim() to prevent
1100                  * the vnode we handle here without a reference from
1101                  * being freed while we reference it.  If we lock the inode
1102                  * while it's on the mount list here, then the spurious inode
1103                  * lock in xfs_ireclaim() after the inode is pulled from
1104                  * the mount list will sleep until we release it here.
1105                  * This keeps the vnode from being freed while we reference
1106                  * it.
1107                  */
1108                 if (xfs_ilock_nowait(ip, lock_flags) == 0) {
1109                         if ((flags & SYNC_BDFLUSH) || (vp == NULL)) {
1110                                 ip = ip->i_mnext;
1111                                 continue;
1112                         }
1113
1114                         vp = vn_grab(vp);
1115                         if (vp == NULL) {
1116                                 ip = ip->i_mnext;
1117                                 continue;
1118                         }
1119
1120                         IPOINTER_INSERT(ip, mp);
1121                         xfs_ilock(ip, lock_flags);
1122
1123                         ASSERT(vp == XFS_ITOV(ip));
1124                         ASSERT(ip->i_mount == mp);
1125
1126                         vnode_refed = B_TRUE;
1127                 }
1128
1129                 /* From here on in the loop we may have a marker record
1130                  * in the inode list.
1131                  */
1132
1133                 if ((flags & SYNC_CLOSE)  && (vp != NULL)) {
1134                         /*
1135                          * This is the shutdown case.  We just need to
1136                          * flush and invalidate all the pages associated
1137                          * with the inode.  Drop the inode lock since
1138                          * we can't hold it across calls to the buffer
1139                          * cache.
1140                          *
1141                          * We don't set the VREMAPPING bit in the vnode
1142                          * here, because we don't hold the vnode lock
1143                          * exclusively.  It doesn't really matter, though,
1144                          * because we only come here when we're shutting
1145                          * down anyway.
1146                          */
1147                         xfs_iunlock(ip, XFS_ILOCK_SHARED);
1148
1149                         if (XFS_FORCED_SHUTDOWN(mp)) {
1150                                 bhv_vop_toss_pages(vp, 0, -1, FI_REMAPF);
1151                         } else {
1152                                 bhv_vop_flushinval_pages(vp, 0, -1, FI_REMAPF);
1153                         }
1154
1155                         xfs_ilock(ip, XFS_ILOCK_SHARED);
1156
1157                 } else if ((flags & SYNC_DELWRI) && (vp != NULL)) {
1158                         if (VN_DIRTY(vp)) {
1159                                 /* We need to have dropped the lock here,
1160                                  * so insert a marker if we have not already
1161                                  * done so.
1162                                  */
1163                                 if (mount_locked) {
1164                                         IPOINTER_INSERT(ip, mp);
1165                                 }
1166
1167                                 /*
1168                                  * Drop the inode lock since we can't hold it
1169                                  * across calls to the buffer cache.
1170                                  */
1171                                 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1172                                 error = bhv_vop_flush_pages(vp, (xfs_off_t)0,
1173                                                         -1, fflag, FI_NONE);
1174                                 xfs_ilock(ip, XFS_ILOCK_SHARED);
1175                         }
1176
1177                 }
1178
1179                 if (flags & SYNC_BDFLUSH) {
1180                         if ((flags & SYNC_ATTR) &&
1181                             ((ip->i_update_core) ||
1182                              ((ip->i_itemp != NULL) &&
1183                               (ip->i_itemp->ili_format.ilf_fields != 0)))) {
1184
1185                                 /* Insert marker and drop lock if not already
1186                                  * done.
1187                                  */
1188                                 if (mount_locked) {
1189                                         IPOINTER_INSERT(ip, mp);
1190                                 }
1191
1192                                 /*
1193                                  * We don't want the periodic flushing of the
1194                                  * inodes by vfs_sync() to interfere with
1195                                  * I/O to the file, especially read I/O
1196                                  * where it is only the access time stamp
1197                                  * that is being flushed out.  To prevent
1198                                  * long periods where we have both inode
1199                                  * locks held shared here while reading the
1200                                  * inode's buffer in from disk, we drop the
1201                                  * inode lock while reading in the inode
1202                                  * buffer.  We have to release the buffer
1203                                  * and reacquire the inode lock so that they
1204                                  * are acquired in the proper order (inode
1205                                  * locks first).  The buffer will go at the
1206                                  * end of the lru chain, though, so we can
1207                                  * expect it to still be there when we go
1208                                  * for it again in xfs_iflush().
1209                                  */
1210                                 if ((xfs_ipincount(ip) == 0) &&
1211                                     xfs_iflock_nowait(ip)) {
1212
1213                                         xfs_ifunlock(ip);
1214                                         xfs_iunlock(ip, XFS_ILOCK_SHARED);
1215
1216                                         error = xfs_itobp(mp, NULL, ip,
1217                                                           &dip, &bp, 0, 0);
1218                                         if (!error) {
1219                                                 xfs_buf_relse(bp);
1220                                         } else {
1221                                                 /* Bailing out, remove the
1222                                                  * marker and free it.
1223                                                  */
1224                                                 XFS_MOUNT_ILOCK(mp);
1225                                                 IPOINTER_REMOVE(ip, mp);
1226                                                 XFS_MOUNT_IUNLOCK(mp);
1227
1228                                                 ASSERT(!(lock_flags &
1229                                                         XFS_IOLOCK_SHARED));
1230
1231                                                 kmem_free(ipointer,
1232                                                         sizeof(xfs_iptr_t));
1233                                                 return (0);
1234                                         }
1235
1236                                         /*
1237                                          * Since we dropped the inode lock,
1238                                          * the inode may have been reclaimed.
1239                                          * Therefore, we reacquire the mount
1240                                          * lock and check to see if we were the
1241                                          * inode reclaimed. If this happened
1242                                          * then the ipointer marker will no
1243                                          * longer point back at us. In this
1244                                          * case, move ip along to the inode
1245                                          * after the marker, remove the marker
1246                                          * and continue.
1247                                          */
1248                                         XFS_MOUNT_ILOCK(mp);
1249                                         mount_locked = B_TRUE;
1250
1251                                         if (ip != ipointer->ip_mprev) {
1252                                                 IPOINTER_REMOVE(ip, mp);
1253
1254                                                 ASSERT(!vnode_refed);
1255                                                 ASSERT(!(lock_flags &
1256                                                         XFS_IOLOCK_SHARED));
1257                                                 continue;
1258                                         }
1259
1260                                         ASSERT(ip->i_mount == mp);
1261
1262                                         if (xfs_ilock_nowait(ip,
1263                                                     XFS_ILOCK_SHARED) == 0) {
1264                                                 ASSERT(ip->i_mount == mp);
1265                                                 /*
1266                                                  * We failed to reacquire
1267                                                  * the inode lock without
1268                                                  * sleeping, so just skip
1269                                                  * the inode for now.  We
1270                                                  * clear the ILOCK bit from
1271                                                  * the lock_flags so that we
1272                                                  * won't try to drop a lock
1273                                                  * we don't hold below.
1274                                                  */
1275                                                 lock_flags &= ~XFS_ILOCK_SHARED;
1276                                                 IPOINTER_REMOVE(ip_next, mp);
1277                                         } else if ((xfs_ipincount(ip) == 0) &&
1278                                                    xfs_iflock_nowait(ip)) {
1279                                                 ASSERT(ip->i_mount == mp);
1280                                                 /*
1281                                                  * Since this is vfs_sync()
1282                                                  * calling we only flush the
1283                                                  * inode out if we can lock
1284                                                  * it without sleeping and
1285                                                  * it is not pinned.  Drop
1286                                                  * the mount lock here so
1287                                                  * that we don't hold it for
1288                                                  * too long. We already have
1289                                                  * a marker in the list here.
1290                                                  */
1291                                                 XFS_MOUNT_IUNLOCK(mp);
1292                                                 mount_locked = B_FALSE;
1293                                                 error = xfs_iflush(ip,
1294                                                            XFS_IFLUSH_DELWRI);
1295                                         } else {
1296                                                 ASSERT(ip->i_mount == mp);
1297                                                 IPOINTER_REMOVE(ip_next, mp);
1298                                         }
1299                                 }
1300
1301                         }
1302
1303                 } else {
1304                         if ((flags & SYNC_ATTR) &&
1305                             ((ip->i_update_core) ||
1306                              ((ip->i_itemp != NULL) &&
1307                               (ip->i_itemp->ili_format.ilf_fields != 0)))) {
1308                                 if (mount_locked) {
1309                                         IPOINTER_INSERT(ip, mp);
1310                                 }
1311
1312                                 if (flags & SYNC_WAIT) {
1313                                         xfs_iflock(ip);
1314                                         error = xfs_iflush(ip,
1315                                                            XFS_IFLUSH_SYNC);
1316                                 } else {
1317                                         /*
1318                                          * If we can't acquire the flush
1319                                          * lock, then the inode is already
1320                                          * being flushed so don't bother
1321                                          * waiting.  If we can lock it then
1322                                          * do a delwri flush so we can
1323                                          * combine multiple inode flushes
1324                                          * in each disk write.
1325                                          */
1326                                         if (xfs_iflock_nowait(ip)) {
1327                                                 error = xfs_iflush(ip,
1328                                                            XFS_IFLUSH_DELWRI);
1329                                         }
1330                                         else if (bypassed)
1331                                                 (*bypassed)++;
1332                                 }
1333                         }
1334                 }
1335
1336                 if (lock_flags != 0) {
1337                         xfs_iunlock(ip, lock_flags);
1338                 }
1339
1340                 if (vnode_refed) {
1341                         /*
1342                          * If we had to take a reference on the vnode
1343                          * above, then wait until after we've unlocked
1344                          * the inode to release the reference.  This is
1345                          * because we can be already holding the inode
1346                          * lock when VN_RELE() calls xfs_inactive().
1347                          *
1348                          * Make sure to drop the mount lock before calling
1349                          * VN_RELE() so that we don't trip over ourselves if
1350                          * we have to go for the mount lock again in the
1351                          * inactive code.
1352                          */
1353                         if (mount_locked) {
1354                                 IPOINTER_INSERT(ip, mp);
1355                         }
1356
1357                         VN_RELE(vp);
1358
1359                         vnode_refed = B_FALSE;
1360                 }
1361
1362                 if (error) {
1363                         last_error = error;
1364                 }
1365
1366                 /*
1367                  * bail out if the filesystem is corrupted.
1368                  */
1369                 if (error == EFSCORRUPTED)  {
1370                         if (!mount_locked) {
1371                                 XFS_MOUNT_ILOCK(mp);
1372                                 IPOINTER_REMOVE(ip, mp);
1373                         }
1374                         XFS_MOUNT_IUNLOCK(mp);
1375                         ASSERT(ipointer_in == B_FALSE);
1376                         kmem_free(ipointer, sizeof(xfs_iptr_t));
1377                         return XFS_ERROR(error);
1378                 }
1379
1380                 /* Let other threads have a chance at the mount lock
1381                  * if we have looped many times without dropping the
1382                  * lock.
1383                  */
1384                 if ((++preempt & XFS_PREEMPT_MASK) == 0) {
1385                         if (mount_locked) {
1386                                 IPOINTER_INSERT(ip, mp);
1387                         }
1388                 }
1389
1390                 if (mount_locked == B_FALSE) {
1391                         XFS_MOUNT_ILOCK(mp);
1392                         mount_locked = B_TRUE;
1393                         IPOINTER_REMOVE(ip, mp);
1394                         continue;
1395                 }
1396
1397                 ASSERT(ipointer_in == B_FALSE);
1398                 ip = ip->i_mnext;
1399
1400         } while (ip != mp->m_inodes);
1401
1402         XFS_MOUNT_IUNLOCK(mp);
1403
1404         ASSERT(ipointer_in == B_FALSE);
1405
1406         kmem_free(ipointer, sizeof(xfs_iptr_t));
1407         return XFS_ERROR(last_error);
1408 }
1409
1410 /*
1411  * xfs sync routine for internal use
1412  *
1413  * This routine supports all of the flags defined for the generic vfs_sync
1414  * interface as explained above under xfs_sync.  In the interests of not
1415  * changing interfaces within the 6.5 family, additional internally-
1416  * required functions are specified within a separate xflags parameter,
1417  * only available by calling this routine.
1418  *
1419  */
1420 int
1421 xfs_syncsub(
1422         xfs_mount_t     *mp,
1423         int             flags,
1424         int             xflags,
1425         int             *bypassed)
1426 {
1427         int             error = 0;
1428         int             last_error = 0;
1429         uint            log_flags = XFS_LOG_FORCE;
1430         xfs_buf_t       *bp;
1431         xfs_buf_log_item_t      *bip;
1432
1433         /*
1434          * Sync out the log.  This ensures that the log is periodically
1435          * flushed even if there is not enough activity to fill it up.
1436          */
1437         if (flags & SYNC_WAIT)
1438                 log_flags |= XFS_LOG_SYNC;
1439
1440         xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1441
1442         if (flags & (SYNC_ATTR|SYNC_DELWRI)) {
1443                 if (flags & SYNC_BDFLUSH)
1444                         xfs_finish_reclaim_all(mp, 1);
1445                 else
1446                         error = xfs_sync_inodes(mp, flags, xflags, bypassed);
1447         }
1448
1449         /*
1450          * Flushing out dirty data above probably generated more
1451          * log activity, so if this isn't vfs_sync() then flush
1452          * the log again.
1453          */
1454         if (flags & SYNC_DELWRI) {
1455                 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1456         }
1457
1458         if (flags & SYNC_FSDATA) {
1459                 /*
1460                  * If this is vfs_sync() then only sync the superblock
1461                  * if we can lock it without sleeping and it is not pinned.
1462                  */
1463                 if (flags & SYNC_BDFLUSH) {
1464                         bp = xfs_getsb(mp, XFS_BUF_TRYLOCK);
1465                         if (bp != NULL) {
1466                                 bip = XFS_BUF_FSPRIVATE(bp,xfs_buf_log_item_t*);
1467                                 if ((bip != NULL) &&
1468                                     xfs_buf_item_dirty(bip)) {
1469                                         if (!(XFS_BUF_ISPINNED(bp))) {
1470                                                 XFS_BUF_ASYNC(bp);
1471                                                 error = xfs_bwrite(mp, bp);
1472                                         } else {
1473                                                 xfs_buf_relse(bp);
1474                                         }
1475                                 } else {
1476                                         xfs_buf_relse(bp);
1477                                 }
1478                         }
1479                 } else {
1480                         bp = xfs_getsb(mp, 0);
1481                         /*
1482                          * If the buffer is pinned then push on the log so
1483                          * we won't get stuck waiting in the write for
1484                          * someone, maybe ourselves, to flush the log.
1485                          * Even though we just pushed the log above, we
1486                          * did not have the superblock buffer locked at
1487                          * that point so it can become pinned in between
1488                          * there and here.
1489                          */
1490                         if (XFS_BUF_ISPINNED(bp))
1491                                 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
1492                         if (flags & SYNC_WAIT)
1493                                 XFS_BUF_UNASYNC(bp);
1494                         else
1495                                 XFS_BUF_ASYNC(bp);
1496                         error = xfs_bwrite(mp, bp);
1497                 }
1498                 if (error) {
1499                         last_error = error;
1500                 }
1501         }
1502
1503         /*
1504          * If this is the periodic sync, then kick some entries out of
1505          * the reference cache.  This ensures that idle entries are
1506          * eventually kicked out of the cache.
1507          */
1508         if (flags & SYNC_REFCACHE) {
1509                 if (flags & SYNC_WAIT)
1510                         xfs_refcache_purge_mp(mp);
1511                 else
1512                         xfs_refcache_purge_some(mp);
1513         }
1514
1515         /*
1516          * Now check to see if the log needs a "dummy" transaction.
1517          */
1518
1519         if (!(flags & SYNC_REMOUNT) && xfs_log_need_covered(mp)) {
1520                 xfs_trans_t *tp;
1521                 xfs_inode_t *ip;
1522
1523                 /*
1524                  * Put a dummy transaction in the log to tell
1525                  * recovery that all others are OK.
1526                  */
1527                 tp = xfs_trans_alloc(mp, XFS_TRANS_DUMMY1);
1528                 if ((error = xfs_trans_reserve(tp, 0,
1529                                 XFS_ICHANGE_LOG_RES(mp),
1530                                 0, 0, 0)))  {
1531                         xfs_trans_cancel(tp, 0);
1532                         return error;
1533                 }
1534
1535                 ip = mp->m_rootip;
1536                 xfs_ilock(ip, XFS_ILOCK_EXCL);
1537
1538                 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1539                 xfs_trans_ihold(tp, ip);
1540                 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1541                 error = xfs_trans_commit(tp, 0, NULL);
1542                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1543                 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1544         }
1545
1546         /*
1547          * When shutting down, we need to insure that the AIL is pushed
1548          * to disk or the filesystem can appear corrupt from the PROM.
1549          */
1550         if ((flags & (SYNC_CLOSE|SYNC_WAIT)) == (SYNC_CLOSE|SYNC_WAIT)) {
1551                 XFS_bflush(mp->m_ddev_targp);
1552                 if (mp->m_rtdev_targp) {
1553                         XFS_bflush(mp->m_rtdev_targp);
1554                 }
1555         }
1556
1557         return XFS_ERROR(last_error);
1558 }
1559
1560 /*
1561  * xfs_vget - called by DMAPI and NFSD to get vnode from file handle
1562  */
1563 STATIC int
1564 xfs_vget(
1565         bhv_desc_t      *bdp,
1566         bhv_vnode_t     **vpp,
1567         fid_t           *fidp)
1568 {
1569         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
1570         xfs_fid_t       *xfid = (struct xfs_fid *)fidp;
1571         xfs_inode_t     *ip;
1572         int             error;
1573         xfs_ino_t       ino;
1574         unsigned int    igen;
1575
1576         /*
1577          * Invalid.  Since handles can be created in user space and passed in
1578          * via gethandle(), this is not cause for a panic.
1579          */
1580         if (xfid->xfs_fid_len != sizeof(*xfid) - sizeof(xfid->xfs_fid_len))
1581                 return XFS_ERROR(EINVAL);
1582
1583         ino  = xfid->xfs_fid_ino;
1584         igen = xfid->xfs_fid_gen;
1585
1586         /*
1587          * NFS can sometimes send requests for ino 0.  Fail them gracefully.
1588          */
1589         if (ino == 0)
1590                 return XFS_ERROR(ESTALE);
1591
1592         error = xfs_iget(mp, NULL, ino, 0, XFS_ILOCK_SHARED, &ip, 0);
1593         if (error) {
1594                 *vpp = NULL;
1595                 return error;
1596         }
1597
1598         if (ip == NULL) {
1599                 *vpp = NULL;
1600                 return XFS_ERROR(EIO);
1601         }
1602
1603         if (ip->i_d.di_mode == 0 || ip->i_d.di_gen != igen) {
1604                 xfs_iput_new(ip, XFS_ILOCK_SHARED);
1605                 *vpp = NULL;
1606                 return XFS_ERROR(ENOENT);
1607         }
1608
1609         *vpp = XFS_ITOV(ip);
1610         xfs_iunlock(ip, XFS_ILOCK_SHARED);
1611         return 0;
1612 }
1613
1614
1615 #define MNTOPT_LOGBUFS  "logbufs"       /* number of XFS log buffers */
1616 #define MNTOPT_LOGBSIZE "logbsize"      /* size of XFS log buffers */
1617 #define MNTOPT_LOGDEV   "logdev"        /* log device */
1618 #define MNTOPT_RTDEV    "rtdev"         /* realtime I/O device */
1619 #define MNTOPT_BIOSIZE  "biosize"       /* log2 of preferred buffered io size */
1620 #define MNTOPT_WSYNC    "wsync"         /* safe-mode nfs compatible mount */
1621 #define MNTOPT_INO64    "ino64"         /* force inodes into 64-bit range */
1622 #define MNTOPT_NOALIGN  "noalign"       /* turn off stripe alignment */
1623 #define MNTOPT_SWALLOC  "swalloc"       /* turn on stripe width allocation */
1624 #define MNTOPT_SUNIT    "sunit"         /* data volume stripe unit */
1625 #define MNTOPT_SWIDTH   "swidth"        /* data volume stripe width */
1626 #define MNTOPT_NOUUID   "nouuid"        /* ignore filesystem UUID */
1627 #define MNTOPT_MTPT     "mtpt"          /* filesystem mount point */
1628 #define MNTOPT_GRPID    "grpid"         /* group-ID from parent directory */
1629 #define MNTOPT_NOGRPID  "nogrpid"       /* group-ID from current process */
1630 #define MNTOPT_BSDGROUPS    "bsdgroups"    /* group-ID from parent directory */
1631 #define MNTOPT_SYSVGROUPS   "sysvgroups"   /* group-ID from current process */
1632 #define MNTOPT_ALLOCSIZE    "allocsize"    /* preferred allocation size */
1633 #define MNTOPT_IHASHSIZE    "ihashsize"    /* size of inode hash table */
1634 #define MNTOPT_NORECOVERY   "norecovery"   /* don't run XFS recovery */
1635 #define MNTOPT_BARRIER  "barrier"       /* use writer barriers for log write and
1636                                          * unwritten extent conversion */
1637 #define MNTOPT_NOBARRIER "nobarrier"    /* .. disable */
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(char *s, char **endp, unsigned int base)
1650 {
1651         int     last, shift_left_factor = 0;
1652         char    *value = s;
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((const char *)s, 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         bhv_vfs_t               *vfsp = bhvtovfs(bhv);
1679         char                    *this_char, *value, *eov;
1680         int                     dsunit, dswidth, vol_dsunit, vol_dswidth;
1681         int                     iosize;
1682
1683         args->flags |= XFSMNT_IDELETE;
1684         args->flags |= XFSMNT_BARRIER;
1685         args->flags2 |= XFSMNT2_COMPAT_IOSIZE;
1686
1687         if (!options)
1688                 goto done;
1689
1690         iosize = dsunit = dswidth = vol_dsunit = vol_dswidth = 0;
1691
1692         while ((this_char = strsep(&options, ",")) != NULL) {
1693                 if (!*this_char)
1694                         continue;
1695                 if ((value = strchr(this_char, '=')) != NULL)
1696                         *value++ = 0;
1697
1698                 if (!strcmp(this_char, MNTOPT_LOGBUFS)) {
1699                         if (!value || !*value) {
1700                                 cmn_err(CE_WARN,
1701                                         "XFS: %s option requires an argument",
1702                                         this_char);
1703                                 return EINVAL;
1704                         }
1705                         args->logbufs = simple_strtoul(value, &eov, 10);
1706                 } else if (!strcmp(this_char, MNTOPT_LOGBSIZE)) {
1707                         if (!value || !*value) {
1708                                 cmn_err(CE_WARN,
1709                                         "XFS: %s option requires an argument",
1710                                         this_char);
1711                                 return EINVAL;
1712                         }
1713                         args->logbufsize = suffix_strtoul(value, &eov, 10);
1714                 } else if (!strcmp(this_char, MNTOPT_LOGDEV)) {
1715                         if (!value || !*value) {
1716                                 cmn_err(CE_WARN,
1717                                         "XFS: %s option requires an argument",
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                                 cmn_err(CE_WARN,
1725                                         "XFS: %s option requires an argument",
1726                                         this_char);
1727                                 return EINVAL;
1728                         }
1729                         strncpy(args->mtpt, value, MAXNAMELEN);
1730                 } else if (!strcmp(this_char, MNTOPT_RTDEV)) {
1731                         if (!value || !*value) {
1732                                 cmn_err(CE_WARN,
1733                                         "XFS: %s option requires an argument",
1734                                         this_char);
1735                                 return EINVAL;
1736                         }
1737                         strncpy(args->rtname, value, MAXNAMELEN);
1738                 } else if (!strcmp(this_char, MNTOPT_BIOSIZE)) {
1739                         if (!value || !*value) {
1740                                 cmn_err(CE_WARN,
1741                                         "XFS: %s option requires an argument",
1742                                         this_char);
1743                                 return EINVAL;
1744                         }
1745                         iosize = simple_strtoul(value, &eov, 10);
1746                         args->flags |= XFSMNT_IOSIZE;
1747                         args->iosizelog = (uint8_t) iosize;
1748                 } else if (!strcmp(this_char, MNTOPT_ALLOCSIZE)) {
1749                         if (!value || !*value) {
1750                                 cmn_err(CE_WARN,
1751                                         "XFS: %s option requires an argument",
1752                                         this_char);
1753                                 return EINVAL;
1754                         }
1755                         iosize = suffix_strtoul(value, &eov, 10);
1756                         args->flags |= XFSMNT_IOSIZE;
1757                         args->iosizelog = ffs(iosize) - 1;
1758                 } else if (!strcmp(this_char, MNTOPT_IHASHSIZE)) {
1759                         if (!value || !*value) {
1760                                 cmn_err(CE_WARN,
1761                                         "XFS: %s option requires an argument",
1762                                         this_char);
1763                                 return EINVAL;
1764                         }
1765                         args->flags |= XFSMNT_IHASHSIZE;
1766                         args->ihashsize = simple_strtoul(value, &eov, 10);
1767                 } else if (!strcmp(this_char, MNTOPT_GRPID) ||
1768                            !strcmp(this_char, MNTOPT_BSDGROUPS)) {
1769                         vfsp->vfs_flag |= VFS_GRPID;
1770                 } else if (!strcmp(this_char, MNTOPT_NOGRPID) ||
1771                            !strcmp(this_char, MNTOPT_SYSVGROUPS)) {
1772                         vfsp->vfs_flag &= ~VFS_GRPID;
1773                 } else if (!strcmp(this_char, MNTOPT_WSYNC)) {
1774                         args->flags |= XFSMNT_WSYNC;
1775                 } else if (!strcmp(this_char, MNTOPT_OSYNCISOSYNC)) {
1776                         args->flags |= XFSMNT_OSYNCISOSYNC;
1777                 } else if (!strcmp(this_char, MNTOPT_NORECOVERY)) {
1778                         args->flags |= XFSMNT_NORECOVERY;
1779                 } else if (!strcmp(this_char, MNTOPT_INO64)) {
1780                         args->flags |= XFSMNT_INO64;
1781 #if !XFS_BIG_INUMS
1782                         cmn_err(CE_WARN,
1783                                 "XFS: %s option not allowed on this system",
1784                                 this_char);
1785                         return EINVAL;
1786 #endif
1787                 } else if (!strcmp(this_char, MNTOPT_NOALIGN)) {
1788                         args->flags |= XFSMNT_NOALIGN;
1789                 } else if (!strcmp(this_char, MNTOPT_SWALLOC)) {
1790                         args->flags |= XFSMNT_SWALLOC;
1791                 } else if (!strcmp(this_char, MNTOPT_SUNIT)) {
1792                         if (!value || !*value) {
1793                                 cmn_err(CE_WARN,
1794                                         "XFS: %s option requires an argument",
1795                                         this_char);
1796                                 return EINVAL;
1797                         }
1798                         dsunit = simple_strtoul(value, &eov, 10);
1799                 } else if (!strcmp(this_char, MNTOPT_SWIDTH)) {
1800                         if (!value || !*value) {
1801                                 cmn_err(CE_WARN,
1802                                         "XFS: %s option requires an argument",
1803                                         this_char);
1804                                 return EINVAL;
1805                         }
1806                         dswidth = simple_strtoul(value, &eov, 10);
1807                 } else if (!strcmp(this_char, MNTOPT_64BITINODE)) {
1808                         args->flags &= ~XFSMNT_32BITINODES;
1809 #if !XFS_BIG_INUMS
1810                         cmn_err(CE_WARN,
1811                                 "XFS: %s option not allowed on this system",
1812                                 this_char);
1813                         return EINVAL;
1814 #endif
1815                 } else if (!strcmp(this_char, MNTOPT_NOUUID)) {
1816                         args->flags |= XFSMNT_NOUUID;
1817                 } else if (!strcmp(this_char, MNTOPT_BARRIER)) {
1818                         args->flags |= XFSMNT_BARRIER;
1819                 } else if (!strcmp(this_char, MNTOPT_NOBARRIER)) {
1820                         args->flags &= ~XFSMNT_BARRIER;
1821                 } else if (!strcmp(this_char, MNTOPT_IKEEP)) {
1822                         args->flags &= ~XFSMNT_IDELETE;
1823                 } else if (!strcmp(this_char, MNTOPT_NOIKEEP)) {
1824                         args->flags |= XFSMNT_IDELETE;
1825                 } else if (!strcmp(this_char, MNTOPT_LARGEIO)) {
1826                         args->flags2 &= ~XFSMNT2_COMPAT_IOSIZE;
1827                 } else if (!strcmp(this_char, MNTOPT_NOLARGEIO)) {
1828                         args->flags2 |= XFSMNT2_COMPAT_IOSIZE;
1829                 } else if (!strcmp(this_char, MNTOPT_ATTR2)) {
1830                         args->flags |= XFSMNT_ATTR2;
1831                 } else if (!strcmp(this_char, MNTOPT_NOATTR2)) {
1832                         args->flags &= ~XFSMNT_ATTR2;
1833                 } else if (!strcmp(this_char, "osyncisdsync")) {
1834                         /* no-op, this is now the default */
1835                         cmn_err(CE_WARN,
1836         "XFS: osyncisdsync is now the default, option is deprecated.");
1837                 } else if (!strcmp(this_char, "irixsgid")) {
1838                         cmn_err(CE_WARN,
1839         "XFS: irixsgid is now a sysctl(2) variable, option is deprecated.");
1840                 } else {
1841                         cmn_err(CE_WARN,
1842                                 "XFS: unknown mount option [%s].", this_char);
1843                         return EINVAL;
1844                 }
1845         }
1846
1847         if (args->flags & XFSMNT_NORECOVERY) {
1848                 if ((vfsp->vfs_flag & VFS_RDONLY) == 0) {
1849                         cmn_err(CE_WARN,
1850                                 "XFS: no-recovery mounts must be read-only.");
1851                         return EINVAL;
1852                 }
1853         }
1854
1855         if ((args->flags & XFSMNT_NOALIGN) && (dsunit || dswidth)) {
1856                 cmn_err(CE_WARN,
1857         "XFS: sunit and swidth options incompatible with the noalign option");
1858                 return EINVAL;
1859         }
1860
1861         if ((dsunit && !dswidth) || (!dsunit && dswidth)) {
1862                 cmn_err(CE_WARN,
1863                         "XFS: sunit and swidth must be specified together");
1864                 return EINVAL;
1865         }
1866
1867         if (dsunit && (dswidth % dsunit != 0)) {
1868                 cmn_err(CE_WARN,
1869         "XFS: stripe width (%d) must be a multiple of the stripe unit (%d)",
1870                         dswidth, dsunit);
1871                 return EINVAL;
1872         }
1873
1874         if ((args->flags & XFSMNT_NOALIGN) != XFSMNT_NOALIGN) {
1875                 if (dsunit) {
1876                         args->sunit = dsunit;
1877                         args->flags |= XFSMNT_RETERR;
1878                 } else {
1879                         args->sunit = vol_dsunit;
1880                 }
1881                 dswidth ? (args->swidth = dswidth) :
1882                           (args->swidth = vol_dswidth);
1883         } else {
1884                 args->sunit = args->swidth = 0;
1885         }
1886
1887 done:
1888         if (args->flags & XFSMNT_32BITINODES)
1889                 vfsp->vfs_flag |= VFS_32BITINODES;
1890         if (args->flags2)
1891                 args->flags |= XFSMNT_FLAGS2;
1892         return 0;
1893 }
1894
1895 STATIC int
1896 xfs_showargs(
1897         struct bhv_desc         *bhv,
1898         struct seq_file         *m)
1899 {
1900         static struct proc_xfs_info {
1901                 int     flag;
1902                 char    *str;
1903         } xfs_info[] = {
1904                 /* the few simple ones we can get from the mount struct */
1905                 { XFS_MOUNT_WSYNC,              "," MNTOPT_WSYNC },
1906                 { XFS_MOUNT_INO64,              "," MNTOPT_INO64 },
1907                 { XFS_MOUNT_NOALIGN,            "," MNTOPT_NOALIGN },
1908                 { XFS_MOUNT_SWALLOC,            "," MNTOPT_SWALLOC },
1909                 { XFS_MOUNT_NOUUID,             "," MNTOPT_NOUUID },
1910                 { XFS_MOUNT_NORECOVERY,         "," MNTOPT_NORECOVERY },
1911                 { XFS_MOUNT_OSYNCISOSYNC,       "," MNTOPT_OSYNCISOSYNC },
1912                 { 0, NULL }
1913         };
1914         struct proc_xfs_info    *xfs_infop;
1915         struct xfs_mount        *mp = XFS_BHVTOM(bhv);
1916         struct bhv_vfs          *vfsp = XFS_MTOVFS(mp);
1917
1918         for (xfs_infop = xfs_info; xfs_infop->flag; xfs_infop++) {
1919                 if (mp->m_flags & xfs_infop->flag)
1920                         seq_puts(m, xfs_infop->str);
1921         }
1922
1923         if (mp->m_flags & XFS_MOUNT_IHASHSIZE)
1924                 seq_printf(m, "," MNTOPT_IHASHSIZE "=%d", mp->m_ihsize);
1925
1926         if (mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)
1927                 seq_printf(m, "," MNTOPT_ALLOCSIZE "=%dk",
1928                                 (int)(1 << mp->m_writeio_log) >> 10);
1929
1930         if (mp->m_logbufs > 0)
1931                 seq_printf(m, "," MNTOPT_LOGBUFS "=%d", mp->m_logbufs);
1932         if (mp->m_logbsize > 0)
1933                 seq_printf(m, "," MNTOPT_LOGBSIZE "=%dk", mp->m_logbsize >> 10);
1934
1935         if (mp->m_logname)
1936                 seq_printf(m, "," MNTOPT_LOGDEV "=%s", mp->m_logname);
1937         if (mp->m_rtname)
1938                 seq_printf(m, "," MNTOPT_RTDEV "=%s", mp->m_rtname);
1939
1940         if (mp->m_dalign > 0)
1941                 seq_printf(m, "," MNTOPT_SUNIT "=%d",
1942                                 (int)XFS_FSB_TO_BB(mp, mp->m_dalign));
1943         if (mp->m_swidth > 0)
1944                 seq_printf(m, "," MNTOPT_SWIDTH "=%d",
1945                                 (int)XFS_FSB_TO_BB(mp, mp->m_swidth));
1946
1947         if (!(mp->m_flags & XFS_MOUNT_IDELETE))
1948                 seq_printf(m, "," MNTOPT_IKEEP);
1949         if (!(mp->m_flags & XFS_MOUNT_COMPAT_IOSIZE))
1950                 seq_printf(m, "," MNTOPT_LARGEIO);
1951
1952         if (!(vfsp->vfs_flag & VFS_32BITINODES))
1953                 seq_printf(m, "," MNTOPT_64BITINODE);
1954         if (vfsp->vfs_flag & VFS_GRPID)
1955                 seq_printf(m, "," MNTOPT_GRPID);
1956
1957         return 0;
1958 }
1959
1960 STATIC void
1961 xfs_freeze(
1962         bhv_desc_t      *bdp)
1963 {
1964         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
1965
1966         while (atomic_read(&mp->m_active_trans) > 0)
1967                 delay(100);
1968
1969         /* Push the superblock and write an unmount record */
1970         xfs_log_unmount_write(mp);
1971         xfs_unmountfs_writesb(mp);
1972         xfs_fs_log_dummy(mp);
1973 }
1974
1975
1976 bhv_vfsops_t xfs_vfsops = {
1977         BHV_IDENTITY_INIT(VFS_BHV_XFS,VFS_POSITION_XFS),
1978         .vfs_parseargs          = xfs_parseargs,
1979         .vfs_showargs           = xfs_showargs,
1980         .vfs_mount              = xfs_mount,
1981         .vfs_unmount            = xfs_unmount,
1982         .vfs_mntupdate          = xfs_mntupdate,
1983         .vfs_root               = xfs_root,
1984         .vfs_statvfs            = xfs_statvfs,
1985         .vfs_sync               = xfs_sync,
1986         .vfs_vget               = xfs_vget,
1987         .vfs_dmapiops           = (vfs_dmapiops_t)fs_nosys,
1988         .vfs_quotactl           = (vfs_quotactl_t)fs_nosys,
1989         .vfs_init_vnode         = xfs_initialize_vnode,
1990         .vfs_force_shutdown     = xfs_do_force_shutdown,
1991         .vfs_freeze             = xfs_freeze,
1992 };