fs/xfs/xfs_vfsops.c

   1 /*
   2  * XFS filesystem operations.
   3  *
   4  * Copyright (c) 2000-2004 Silicon Graphics, Inc.  All Rights Reserved.
   5  *
   6  * This program is free software; you can redistribute it and/or modify it
   7  * under the terms of version 2 of the GNU General Public License as
   8  * published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it would be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
  13  *
  14  * Further, this software is distributed without any warranty that it is
  15  * free of the rightful claim of any third person regarding infringement
  16  * or the like.  Any license provided herein, whether implied or
  17  * otherwise, applies only to this software file.  Patent licenses, if
  18  * any, provided herein do not apply to combinations of this program with
  19  * other software, or any other product whatsoever.
  20  *
  21  * You should have received a copy of the GNU General Public License along
  22  * with this program; if not, write the Free Software Foundation, Inc., 59
  23  * Temple Place - Suite 330, Boston MA 02111-1307, USA.
  24  *
  25  * Contact information: Silicon Graphics, Inc., 1600 Amphitheatre Pkwy,
  26  * Mountain View, CA  94043, or:
  27  *
  28  * http://www.sgi.com
  29  *
  30  * For further information regarding this notice, see:
  31  *
  32  * http://oss.sgi.com/projects/GenInfo/SGIGPLNoticeExplan/
  33  */
  34
  35 #include "xfs.h"
  36 #include "xfs_macros.h"
  37 #include "xfs_types.h"
  38 #include "xfs_inum.h"
  39 #include "xfs_log.h"
  40 #include "xfs_trans.h"
  41 #include "xfs_sb.h"
  42 #include "xfs_dir.h"
  43 #include "xfs_dir2.h"
  44 #include "xfs_dmapi.h"
  45 #include "xfs_mount.h"
  46 #include "xfs_bmap_btree.h"
  47 #include "xfs_ialloc_btree.h"
  48 #include "xfs_alloc_btree.h"
  49 #include "xfs_btree.h"
  50 #include "xfs_alloc.h"
  51 #include "xfs_ialloc.h"
  52 #include "xfs_attr_sf.h"
  53 #include "xfs_dir_sf.h"
  54 #include "xfs_dir2_sf.h"
  55 #include "xfs_dinode.h"
  56 #include "xfs_inode_item.h"
  57 #include "xfs_inode.h"
  58 #include "xfs_ag.h"
  59 #include "xfs_error.h"
  60 #include "xfs_bmap.h"
  61 #include "xfs_da_btree.h"
  62 #include "xfs_rw.h"
  63 #include "xfs_refcache.h"
  64 #include "xfs_buf_item.h"
  65 #include "xfs_extfree_item.h"
  66 #include "xfs_quota.h"
  67 #include "xfs_dir2_trace.h"
  68 #include "xfs_acl.h"
  69 #include "xfs_attr.h"
  70 #include "xfs_clnt.h"
  71 #include "xfs_log_priv.h"
  72
  73 STATIC int xfs_sync(bhv_desc_t *, int, cred_t *);
  74
  75 int
  76 xfs_init(void)
  77 {
  78         extern kmem_zone_t      *xfs_bmap_free_item_zone;
  79         extern kmem_zone_t      *xfs_btree_cur_zone;
  80         extern kmem_zone_t      *xfs_trans_zone;
  81         extern kmem_zone_t      *xfs_buf_item_zone;
  82         extern kmem_zone_t      *xfs_dabuf_zone;
  83 #ifdef XFS_DABUF_DEBUG
  84         extern lock_t           xfs_dabuf_global_lock;
  85         spinlock_init(&xfs_dabuf_global_lock, "xfsda");
  86 #endif
  87
  88         /*
  89          * Initialize all of the zone allocators we use.
  90          */
  91         xfs_bmap_free_item_zone = kmem_zone_init(sizeof(xfs_bmap_free_item_t),
  92                                                  "xfs_bmap_free_item");
  93         xfs_btree_cur_zone = kmem_zone_init(sizeof(xfs_btree_cur_t),
  94                                             "xfs_btree_cur");
  95         xfs_inode_zone = kmem_zone_init(sizeof(xfs_inode_t), "xfs_inode");
  96         xfs_trans_zone = kmem_zone_init(sizeof(xfs_trans_t), "xfs_trans");
  97         xfs_da_state_zone =
  98                 kmem_zone_init(sizeof(xfs_da_state_t), "xfs_da_state");
  99         xfs_dabuf_zone = kmem_zone_init(sizeof(xfs_dabuf_t), "xfs_dabuf");
 100
 101         /*
 102          * The size of the zone allocated buf log item is the maximum
 103          * size possible under XFS.  This wastes a little bit of memory,
 104          * but it is much faster.
 105          */
 106         xfs_buf_item_zone =
 107                 kmem_zone_init((sizeof(xfs_buf_log_item_t) +
 108                                 (((XFS_MAX_BLOCKSIZE / XFS_BLI_CHUNK) /
 109                                   NBWORD) * sizeof(int))),
 110                                "xfs_buf_item");
 111         xfs_efd_zone = kmem_zone_init((sizeof(xfs_efd_log_item_t) +
 112                                        ((XFS_EFD_MAX_FAST_EXTENTS - 1) * sizeof(xfs_extent_t))),
 113                                       "xfs_efd_item");
 114         xfs_efi_zone = kmem_zone_init((sizeof(xfs_efi_log_item_t) +
 115                                        ((XFS_EFI_MAX_FAST_EXTENTS - 1) * sizeof(xfs_extent_t))),
 116                                       "xfs_efi_item");
 117         xfs_ifork_zone = kmem_zone_init(sizeof(xfs_ifork_t), "xfs_ifork");
 118         xfs_ili_zone = kmem_zone_init(sizeof(xfs_inode_log_item_t), "xfs_ili");
 119         xfs_chashlist_zone = kmem_zone_init(sizeof(xfs_chashlist_t),
 120                                             "xfs_chashlist");
 121         _ACL_ZONE_INIT(xfs_acl_zone, "xfs_acl");
 122
 123         /*
 124          * Allocate global trace buffers.
 125          */
 126 #ifdef XFS_ALLOC_TRACE
 127         xfs_alloc_trace_buf = ktrace_alloc(XFS_ALLOC_TRACE_SIZE, KM_SLEEP);
 128 #endif
 129 #ifdef XFS_BMAP_TRACE
 130         xfs_bmap_trace_buf = ktrace_alloc(XFS_BMAP_TRACE_SIZE, KM_SLEEP);
 131 #endif
 132 #ifdef XFS_BMBT_TRACE
 133         xfs_bmbt_trace_buf = ktrace_alloc(XFS_BMBT_TRACE_SIZE, KM_SLEEP);
 134 #endif
 135 #ifdef XFS_DIR_TRACE
 136         xfs_dir_trace_buf = ktrace_alloc(XFS_DIR_TRACE_SIZE, KM_SLEEP);
 137 #endif
 138 #ifdef XFS_ATTR_TRACE
 139         xfs_attr_trace_buf = ktrace_alloc(XFS_ATTR_TRACE_SIZE, KM_SLEEP);
 140 #endif
 141 #ifdef XFS_DIR2_TRACE
 142         xfs_dir2_trace_buf = ktrace_alloc(XFS_DIR2_GTRACE_SIZE, KM_SLEEP);
 143 #endif
 144
 145         xfs_dir_startup();
 146
 147 #if (defined(DEBUG) || defined(INDUCE_IO_ERROR))
 148         xfs_error_test_init();
 149 #endif /* DEBUG || INDUCE_IO_ERROR */
 150
 151         xfs_init_procfs();
 152         xfs_sysctl_register();
 153         return 0;
 154 }
 155
 156 void
 157 xfs_cleanup(void)
 158 {
 159         extern kmem_zone_t      *xfs_bmap_free_item_zone;
 160         extern kmem_zone_t      *xfs_btree_cur_zone;
 161         extern kmem_zone_t      *xfs_inode_zone;
 162         extern kmem_zone_t      *xfs_trans_zone;
 163         extern kmem_zone_t      *xfs_da_state_zone;
 164         extern kmem_zone_t      *xfs_dabuf_zone;
 165         extern kmem_zone_t      *xfs_efd_zone;
 166         extern kmem_zone_t      *xfs_efi_zone;
 167         extern kmem_zone_t      *xfs_buf_item_zone;
 168         extern kmem_zone_t      *xfs_chashlist_zone;
 169
 170         xfs_cleanup_procfs();
 171         xfs_sysctl_unregister();
 172         xfs_refcache_destroy();
 173
 174 #ifdef XFS_DIR2_TRACE
 175         ktrace_free(xfs_dir2_trace_buf);
 176 #endif
 177 #ifdef XFS_ATTR_TRACE
 178         ktrace_free(xfs_attr_trace_buf);
 179 #endif
 180 #ifdef XFS_DIR_TRACE
 181         ktrace_free(xfs_dir_trace_buf);
 182 #endif
 183 #ifdef XFS_BMBT_TRACE
 184         ktrace_free(xfs_bmbt_trace_buf);
 185 #endif
 186 #ifdef XFS_BMAP_TRACE
 187         ktrace_free(xfs_bmap_trace_buf);
 188 #endif
 189 #ifdef XFS_ALLOC_TRACE
 190         ktrace_free(xfs_alloc_trace_buf);
 191 #endif
 192
 193         kmem_cache_destroy(xfs_bmap_free_item_zone);
 194         kmem_cache_destroy(xfs_btree_cur_zone);
 195         kmem_cache_destroy(xfs_inode_zone);
 196         kmem_cache_destroy(xfs_trans_zone);
 197         kmem_cache_destroy(xfs_da_state_zone);
 198         kmem_cache_destroy(xfs_dabuf_zone);
 199         kmem_cache_destroy(xfs_buf_item_zone);
 200         kmem_cache_destroy(xfs_efd_zone);
 201         kmem_cache_destroy(xfs_efi_zone);
 202         kmem_cache_destroy(xfs_ifork_zone);
 203         kmem_cache_destroy(xfs_ili_zone);
 204         kmem_cache_destroy(xfs_chashlist_zone);
 205         _ACL_ZONE_DESTROY(xfs_acl_zone);
 206 }
 207
 208 /*
 209  * xfs_start_flags
 210  *
 211  * This function fills in xfs_mount_t fields based on mount args.
 212  * Note: the superblock has _not_ yet been read in.
 213  */
 214 STATIC int
 215 xfs_start_flags(
 216         struct vfs              *vfs,
 217         struct xfs_mount_args   *ap,
 218         struct xfs_mount        *mp)
 219 {
 220         /* Values are in BBs */
 221         if ((ap->flags & XFSMNT_NOALIGN) != XFSMNT_NOALIGN) {
 222                 /*
 223                  * At this point the superblock has not been read
 224                  * in, therefore we do not know the block size.
 225                  * Before the mount call ends we will convert
 226                  * these to FSBs.
 227                  */
 228                 mp->m_dalign = ap->sunit;
 229                 mp->m_swidth = ap->swidth;
 230         }
 231
 232         if (ap->logbufs != -1 &&
 233 #if defined(DEBUG) || defined(XLOG_NOLOG)
 234             ap->logbufs != 0 &&
 235 #endif
 236             (ap->logbufs < XLOG_MIN_ICLOGS ||
 237              ap->logbufs > XLOG_MAX_ICLOGS)) {
 238                 cmn_err(CE_WARN,
 239                         "XFS: invalid logbufs value: %d [not %d-%d]",
 240                         ap->logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
 241                 return XFS_ERROR(EINVAL);
 242         }
 243         mp->m_logbufs = ap->logbufs;
 244         if (ap->logbufsize != -1 &&
 245             ap->logbufsize != 16 * 1024 &&
 246             ap->logbufsize != 32 * 1024 &&
 247             ap->logbufsize != 64 * 1024 &&
 248             ap->logbufsize != 128 * 1024 &&
 249             ap->logbufsize != 256 * 1024) {
 250                 cmn_err(CE_WARN,
 251         "XFS: invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
 252                         ap->logbufsize);
 253                 return XFS_ERROR(EINVAL);
 254         }
 255         mp->m_logbsize = ap->logbufsize;
 256         mp->m_fsname_len = strlen(ap->fsname) + 1;
 257         mp->m_fsname = kmem_alloc(mp->m_fsname_len, KM_SLEEP);
 258         strcpy(mp->m_fsname, ap->fsname);
 259
 260         /*
 261          * Pull in the 'wsync' and 'ino64' mount options before we do the real
 262          * work of mounting and recovery.  The arg pointer will
 263          * be NULL when we are being called from the root mount code.
 264          */
 265         if (ap->flags & XFSMNT_WSYNC)
 266                 mp->m_flags |= XFS_MOUNT_WSYNC;
 267 #if XFS_BIG_INUMS
 268         if (ap->flags & XFSMNT_INO64) {
 269                 mp->m_flags |= XFS_MOUNT_INO64;
 270                 mp->m_inoadd = XFS_INO64_OFFSET;
 271         }
 272 #endif
 273         if (ap->flags & XFSMNT_NOATIME)
 274                 mp->m_flags |= XFS_MOUNT_NOATIME;
 275
 276         if (ap->flags & XFSMNT_RETERR)
 277                 mp->m_flags |= XFS_MOUNT_RETERR;
 278
 279         if (ap->flags & XFSMNT_NOALIGN)
 280                 mp->m_flags |= XFS_MOUNT_NOALIGN;
 281
 282         if (ap->flags & XFSMNT_SWALLOC)
 283                 mp->m_flags |= XFS_MOUNT_SWALLOC;
 284
 285         if (ap->flags & XFSMNT_OSYNCISOSYNC)
 286                 mp->m_flags |= XFS_MOUNT_OSYNCISOSYNC;
 287
 288         if (ap->flags & XFSMNT_32BITINODES)
 289                 mp->m_flags |= (XFS_MOUNT_32BITINODES | XFS_MOUNT_32BITINOOPT);
 290
 291         if (ap->flags & XFSMNT_IOSIZE) {
 292                 if (ap->iosizelog > XFS_MAX_IO_LOG ||
 293                     ap->iosizelog < XFS_MIN_IO_LOG) {
 294                         cmn_err(CE_WARN,
 295                 "XFS: invalid log iosize: %d [not %d-%d]",
 296                                 ap->iosizelog, XFS_MIN_IO_LOG,
 297                                 XFS_MAX_IO_LOG);
 298                         return XFS_ERROR(EINVAL);
 299                 }
 300
 301                 mp->m_flags |= XFS_MOUNT_DFLT_IOSIZE;
 302                 mp->m_readio_log = mp->m_writeio_log = ap->iosizelog;
 303         }
 304         if (ap->flags & XFSMNT_IDELETE)
 305                 mp->m_flags |= XFS_MOUNT_IDELETE;
 306
 307         /*
 308          * no recovery flag requires a read-only mount
 309          */
 310         if (ap->flags & XFSMNT_NORECOVERY) {
 311                 if (!(vfs->vfs_flag & VFS_RDONLY)) {
 312                         cmn_err(CE_WARN,
 313         "XFS: tried to mount a FS read-write without recovery!");
 314                         return XFS_ERROR(EINVAL);
 315                 }
 316                 mp->m_flags |= XFS_MOUNT_NORECOVERY;
 317         }
 318
 319         if (ap->flags & XFSMNT_NOUUID)
 320                 mp->m_flags |= XFS_MOUNT_NOUUID;
 321         if (ap->flags & XFSMNT_NOLOGFLUSH)
 322                 mp->m_flags |= XFS_MOUNT_NOLOGFLUSH;
 323
 324         return 0;
 325 }
 326
 327 /*
 328  * This function fills in xfs_mount_t fields based on mount args.
 329  * Note: the superblock _has_ now been read in.
 330  */
 331 STATIC int
 332 xfs_finish_flags(
 333         struct vfs              *vfs,
 334         struct xfs_mount_args   *ap,
 335         struct xfs_mount        *mp)
 336 {
 337         int                     ronly = (vfs->vfs_flag & VFS_RDONLY);
 338
 339         /* Fail a mount where the logbuf is smaller then the log stripe */
 340         if (XFS_SB_VERSION_HASLOGV2(&mp->m_sb)) {
 341                 if ((ap->logbufsize == -1) &&
 342                     (mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE)) {
 343                         mp->m_logbsize = mp->m_sb.sb_logsunit;
 344                 } else if (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         /*
 359          * prohibit r/w mounts of read-only filesystems
 360          */
 361         if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !ronly) {
 362                 cmn_err(CE_WARN,
 363         "XFS: cannot mount a read-only filesystem as read-write");
 364                 return XFS_ERROR(EROFS);
 365         }
 366
 367         /*
 368          * disallow mount attempts with (IRIX) project quota enabled
 369          */
 370         if (XFS_SB_VERSION_HASQUOTA(&mp->m_sb) &&
 371             (mp->m_sb.sb_qflags & XFS_PQUOTA_ACCT)) {
 372                 cmn_err(CE_WARN,
 373         "XFS: cannot mount a filesystem with IRIX project quota enabled");
 374                 return XFS_ERROR(ENOSYS);
 375         }
 376
 377         /*
 378          * check for shared mount.
 379          */
 380         if (ap->flags & XFSMNT_SHARED) {
 381                 if (!XFS_SB_VERSION_HASSHARED(&mp->m_sb))
 382                         return XFS_ERROR(EINVAL);
 383
 384                 /*
 385                  * For IRIX 6.5, shared mounts must have the shared
 386                  * version bit set, have the persistent readonly
 387                  * field set, must be version 0 and can only be mounted
 388                  * read-only.
 389                  */
 390                 if (!ronly || !(mp->m_sb.sb_flags & XFS_SBF_READONLY) ||
 391                      (mp->m_sb.sb_shared_vn != 0))
 392                         return XFS_ERROR(EINVAL);
 393
 394                 mp->m_flags |= XFS_MOUNT_SHARED;
 395
 396                 /*
 397                  * Shared XFS V0 can't deal with DMI.  Return EINVAL.
 398                  */
 399                 if (mp->m_sb.sb_shared_vn == 0 && (ap->flags & XFSMNT_DMAPI))
 400                         return XFS_ERROR(EINVAL);
 401         }
 402
 403         return 0;
 404 }
 405
 406 /*
 407  * xfs_mount
 408  *
 409  * The file system configurations are:
 410  *      (1) device (partition) with data and internal log
 411  *      (2) logical volume with data and log subvolumes.
 412  *      (3) logical volume with data, log, and realtime subvolumes.
 413  *
 414  * We only have to handle opening the log and realtime volumes here if
 415  * they are present.  The data subvolume has already been opened by
 416  * get_sb_bdev() and is stored in vfsp->vfs_super->s_bdev.
 417  */
 418 STATIC int
 419 xfs_mount(
 420         struct bhv_desc         *bhvp,
 421         struct xfs_mount_args   *args,
 422         cred_t                  *credp)
 423 {
 424         struct vfs              *vfsp = bhvtovfs(bhvp);
 425         struct bhv_desc         *p;
 426         struct xfs_mount        *mp = XFS_BHVTOM(bhvp);
 427         struct block_device     *ddev, *logdev, *rtdev;
 428         int                     flags = 0, error;
 429
 430         ddev = vfsp->vfs_super->s_bdev;
 431         logdev = rtdev = NULL;
 432
 433         /*
 434          * Open real time and log devices - order is important.
 435          */
 436         if (args->logname[0]) {
 437                 error = xfs_blkdev_get(mp, args->logname, &logdev);
 438                 if (error)
 439                         return error;
 440         }
 441         if (args->rtname[0]) {
 442                 error = xfs_blkdev_get(mp, args->rtname, &rtdev);
 443                 if (error) {
 444                         xfs_blkdev_put(logdev);
 445                         return error;
 446                 }
 447
 448                 if (rtdev == ddev || rtdev == logdev) {
 449                         cmn_err(CE_WARN,
 450         "XFS: Cannot mount filesystem with identical rtdev and ddev/logdev.");
 451                         xfs_blkdev_put(logdev);
 452                         xfs_blkdev_put(rtdev);
 453                         return EINVAL;
 454                 }
 455         }
 456
 457         /*
 458          * Setup xfs_mount function vectors from available behaviors
 459          */
 460         p = vfs_bhv_lookup(vfsp, VFS_POSITION_DM);
 461         mp->m_dm_ops = p ? *(xfs_dmops_t *) vfs_bhv_custom(p) : xfs_dmcore_stub;
 462         p = vfs_bhv_lookup(vfsp, VFS_POSITION_QM);
 463         mp->m_qm_ops = p ? *(xfs_qmops_t *) vfs_bhv_custom(p) : xfs_qmcore_stub;
 464         p = vfs_bhv_lookup(vfsp, VFS_POSITION_IO);
 465         mp->m_io_ops = p ? *(xfs_ioops_t *) vfs_bhv_custom(p) : xfs_iocore_xfs;
 466
 467         /*
 468          * Setup xfs_mount buffer target pointers
 469          */
 470         mp->m_ddev_targp = xfs_alloc_buftarg(ddev);
 471         if (rtdev)
 472                 mp->m_rtdev_targp = xfs_alloc_buftarg(rtdev);
 473         mp->m_logdev_targp = (logdev && logdev != ddev) ?
 474                                 xfs_alloc_buftarg(logdev) : mp->m_ddev_targp;
 475
 476         /*
 477          * Setup flags based on mount(2) options and then the superblock
 478          */
 479         error = xfs_start_flags(vfsp, args, mp);
 480         if (error)
 481                 goto error;
 482         error = xfs_readsb(mp);
 483         if (error)
 484                 goto error;
 485         error = xfs_finish_flags(vfsp, args, mp);
 486         if (error) {
 487                 xfs_freesb(mp);
 488                 goto error;
 489         }
 490
 491         /*
 492          * Setup xfs_mount buffer target pointers based on superblock
 493          */
 494         xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_blocksize,
 495                             mp->m_sb.sb_sectsize);
 496         if (logdev && logdev != ddev) {
 497                 unsigned int    log_sector_size = BBSIZE;
 498
 499                 if (XFS_SB_VERSION_HASSECTOR(&mp->m_sb))
 500                         log_sector_size = mp->m_sb.sb_logsectsize;
 501                 xfs_setsize_buftarg(mp->m_logdev_targp, mp->m_sb.sb_blocksize,
 502                                     log_sector_size);
 503         }
 504         if (rtdev)
 505                 xfs_setsize_buftarg(mp->m_rtdev_targp, mp->m_sb.sb_blocksize,
 506                                     mp->m_sb.sb_blocksize);
 507
 508         if (!(error = XFS_IOINIT(vfsp, args, flags)))
 509                 return 0;
 510
 511  error:
 512         xfs_binval(mp->m_ddev_targp);
 513         if (logdev != NULL && logdev != ddev) {
 514                 xfs_binval(mp->m_logdev_targp);
 515         }
 516         if (rtdev != NULL) {
 517                 xfs_binval(mp->m_rtdev_targp);
 518         }
 519         xfs_unmountfs_close(mp, NULL);
 520         return error;
 521 }
 522
 523 STATIC int
 524 xfs_unmount(
 525         bhv_desc_t      *bdp,
 526         int             flags,
 527         cred_t          *credp)
 528 {
 529         struct vfs      *vfsp = bhvtovfs(bdp);
 530         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
 531         xfs_inode_t     *rip;
 532         vnode_t         *rvp;
 533         int             unmount_event_wanted = 0;
 534         int             unmount_event_flags = 0;
 535         int             xfs_unmountfs_needed = 0;
 536         int             error;
 537
 538         rip = mp->m_rootip;
 539         rvp = XFS_ITOV(rip);
 540
 541         if (vfsp->vfs_flag & VFS_DMI) {
 542                 error = XFS_SEND_PREUNMOUNT(mp, vfsp,
 543                                 rvp, DM_RIGHT_NULL, rvp, DM_RIGHT_NULL,
 544                                 NULL, NULL, 0, 0,
 545                                 (mp->m_dmevmask & (1<<DM_EVENT_PREUNMOUNT))?
 546                                         0:DM_FLAGS_UNWANTED);
 547                         if (error)
 548                                 return XFS_ERROR(error);
 549                 unmount_event_wanted = 1;
 550                 unmount_event_flags = (mp->m_dmevmask & (1<<DM_EVENT_UNMOUNT))?
 551                                         0 : DM_FLAGS_UNWANTED;
 552         }
 553
 554         /*
 555          * First blow any referenced inode from this file system
 556          * out of the reference cache, and delete the timer.
 557          */
 558         xfs_refcache_purge_mp(mp);
 559
 560         XFS_bflush(mp->m_ddev_targp);
 561         error = xfs_unmount_flush(mp, 0);
 562         if (error)
 563                 goto out;
 564
 565         ASSERT(vn_count(rvp) == 1);
 566
 567         /*
 568          * Drop the reference count
 569          */
 570         VN_RELE(rvp);
 571
 572         /*
 573          * If we're forcing a shutdown, typically because of a media error,
 574          * we want to make sure we invalidate dirty pages that belong to
 575          * referenced vnodes as well.
 576          */
 577         if (XFS_FORCED_SHUTDOWN(mp)) {
 578                 error = xfs_sync(&mp->m_bhv,
 579                          (SYNC_WAIT | SYNC_CLOSE), credp);
 580                 ASSERT(error != EFSCORRUPTED);
 581         }
 582         xfs_unmountfs_needed = 1;
 583
 584 out:
 585         /*      Send DMAPI event, if required.
 586          *      Then do xfs_unmountfs() if needed.
 587          *      Then return error (or zero).
 588          */
 589         if (unmount_event_wanted) {
 590                 /* Note: mp structure must still exist for
 591                  * XFS_SEND_UNMOUNT() call.
 592                  */
 593                 XFS_SEND_UNMOUNT(mp, vfsp, error == 0 ? rvp : NULL,
 594                         DM_RIGHT_NULL, 0, error, unmount_event_flags);
 595         }
 596         if (xfs_unmountfs_needed) {
 597                 /*
 598                  * Call common unmount function to flush to disk
 599                  * and free the super block buffer & mount structures.
 600                  */
 601                 xfs_unmountfs(mp, credp);
 602         }
 603
 604         return XFS_ERROR(error);
 605 }
 606
 607 #define REMOUNT_READONLY_FLAGS  (SYNC_REMOUNT|SYNC_ATTR|SYNC_WAIT)
 608
 609 STATIC int
 610 xfs_mntupdate(
 611         bhv_desc_t                      *bdp,
 612         int                             *flags,
 613         struct xfs_mount_args           *args)
 614 {
 615         struct vfs      *vfsp = bhvtovfs(bdp);
 616         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
 617         int             pincount, error;
 618         int             count = 0;
 619
 620         if (args->flags & XFSMNT_NOATIME)
 621                 mp->m_flags |= XFS_MOUNT_NOATIME;
 622         else
 623                 mp->m_flags &= ~XFS_MOUNT_NOATIME;
 624
 625         if (!(vfsp->vfs_flag & VFS_RDONLY)) {
 626                 VFS_SYNC(vfsp, SYNC_FSDATA|SYNC_BDFLUSH|SYNC_ATTR, NULL, error);
 627         }
 628
 629         if (*flags & MS_RDONLY) {
 630                 xfs_refcache_purge_mp(mp);
 631                 xfs_flush_buftarg(mp->m_ddev_targp, 0);
 632                 xfs_finish_reclaim_all(mp, 0);
 633
 634                 /* This loop must run at least twice.
 635                  * The first instance of the loop will flush
 636                  * most meta data but that will generate more
 637                  * meta data (typically directory updates).
 638                  * Which then must be flushed and logged before
 639                  * we can write the unmount record.
 640                  */
 641                 do {
 642                         VFS_SYNC(vfsp, REMOUNT_READONLY_FLAGS, NULL, error);
 643                         pincount = xfs_flush_buftarg(mp->m_ddev_targp, 1);
 644                         if (!pincount) {
 645                                 delay(50);
 646                                 count++;
 647                         }
 648                 } while (count < 2);
 649
 650                 /* Ok now write out an unmount record */
 651                 xfs_log_unmount_write(mp);
 652                 xfs_unmountfs_writesb(mp);
 653                 vfsp->vfs_flag |= VFS_RDONLY;
 654         } else {
 655                 vfsp->vfs_flag &= ~VFS_RDONLY;
 656         }
 657
 658         return 0;
 659 }
 660
 661 /*
 662  * xfs_unmount_flush implements a set of flush operation on special
 663  * inodes, which are needed as a separate set of operations so that
 664  * they can be called as part of relocation process.
 665  */
 666 int
 667 xfs_unmount_flush(
 668         xfs_mount_t     *mp,            /* Mount structure we are getting
 669                                            rid of. */
 670         int             relocation)     /* Called from vfs relocation. */
 671 {
 672         xfs_inode_t     *rip = mp->m_rootip;
 673         xfs_inode_t     *rbmip;
 674         xfs_inode_t     *rsumip = NULL;
 675         vnode_t         *rvp = XFS_ITOV(rip);
 676         int             error;
 677
 678         xfs_ilock(rip, XFS_ILOCK_EXCL);
 679         xfs_iflock(rip);
 680
 681         /*
 682          * Flush out the real time inodes.
 683          */
 684         if ((rbmip = mp->m_rbmip) != NULL) {
 685                 xfs_ilock(rbmip, XFS_ILOCK_EXCL);
 686                 xfs_iflock(rbmip);
 687                 error = xfs_iflush(rbmip, XFS_IFLUSH_SYNC);
 688                 xfs_iunlock(rbmip, XFS_ILOCK_EXCL);
 689
 690                 if (error == EFSCORRUPTED)
 691                         goto fscorrupt_out;
 692
 693                 ASSERT(vn_count(XFS_ITOV(rbmip)) == 1);
 694
 695                 rsumip = mp->m_rsumip;
 696                 xfs_ilock(rsumip, XFS_ILOCK_EXCL);
 697                 xfs_iflock(rsumip);
 698                 error = xfs_iflush(rsumip, XFS_IFLUSH_SYNC);
 699                 xfs_iunlock(rsumip, XFS_ILOCK_EXCL);
 700
 701                 if (error == EFSCORRUPTED)
 702                         goto fscorrupt_out;
 703
 704                 ASSERT(vn_count(XFS_ITOV(rsumip)) == 1);
 705         }
 706
 707         /*
 708          * Synchronously flush root inode to disk
 709          */
 710         error = xfs_iflush(rip, XFS_IFLUSH_SYNC);
 711         if (error == EFSCORRUPTED)
 712                 goto fscorrupt_out2;
 713
 714         if (vn_count(rvp) != 1 && !relocation) {
 715                 xfs_iunlock(rip, XFS_ILOCK_EXCL);
 716                 return XFS_ERROR(EBUSY);
 717         }
 718
 719         /*
 720          * Release dquot that rootinode, rbmino and rsumino might be holding,
 721          * flush and purge the quota inodes.
 722          */
 723         error = XFS_QM_UNMOUNT(mp);
 724         if (error == EFSCORRUPTED)
 725                 goto fscorrupt_out2;
 726
 727         if (rbmip) {
 728                 VN_RELE(XFS_ITOV(rbmip));
 729                 VN_RELE(XFS_ITOV(rsumip));
 730         }
 731
 732         xfs_iunlock(rip, XFS_ILOCK_EXCL);
 733         return 0;
 734
 735 fscorrupt_out:
 736         xfs_ifunlock(rip);
 737
 738 fscorrupt_out2:
 739         xfs_iunlock(rip, XFS_ILOCK_EXCL);
 740
 741         return XFS_ERROR(EFSCORRUPTED);
 742 }
 743
 744 /*
 745  * xfs_root extracts the root vnode from a vfs.
 746  *
 747  * vfsp -- the vfs struct for the desired file system
 748  * vpp  -- address of the caller's vnode pointer which should be
 749  *         set to the desired fs root vnode
 750  */
 751 STATIC int
 752 xfs_root(
 753         bhv_desc_t      *bdp,
 754         vnode_t         **vpp)
 755 {
 756         vnode_t         *vp;
 757
 758         vp = XFS_ITOV((XFS_BHVTOM(bdp))->m_rootip);
 759         VN_HOLD(vp);
 760         *vpp = vp;
 761         return 0;
 762 }
 763
 764 /*
 765  * xfs_statvfs
 766  *
 767  * Fill in the statvfs structure for the given file system.  We use
 768  * the superblock lock in the mount structure to ensure a consistent
 769  * snapshot of the counters returned.
 770  */
 771 STATIC int
 772 xfs_statvfs(
 773         bhv_desc_t      *bdp,
 774         xfs_statfs_t    *statp,
 775         vnode_t         *vp)
 776 {
 777         __uint64_t      fakeinos;
 778         xfs_extlen_t    lsize;
 779         xfs_mount_t     *mp;
 780         xfs_sb_t        *sbp;
 781         unsigned long   s;
 782         u64 id;
 783
 784         mp = XFS_BHVTOM(bdp);
 785         sbp = &(mp->m_sb);
 786
 787         statp->f_type = XFS_SB_MAGIC;
 788
 789         s = XFS_SB_LOCK(mp);
 790         statp->f_bsize = sbp->sb_blocksize;
 791         lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
 792         statp->f_blocks = sbp->sb_dblocks - lsize;
 793         statp->f_bfree = statp->f_bavail = sbp->sb_fdblocks;
 794         fakeinos = statp->f_bfree << sbp->sb_inopblog;
 795 #if XFS_BIG_INUMS
 796         fakeinos += mp->m_inoadd;
 797 #endif
 798         statp->f_files =
 799             MIN(sbp->sb_icount + fakeinos, (__uint64_t)XFS_MAXINUMBER);
 800         if (mp->m_maxicount)
 801 #if XFS_BIG_INUMS
 802                 if (!mp->m_inoadd)
 803 #endif
 804                         statp->f_files = min_t(typeof(statp->f_files),
 805                                                 statp->f_files,
 806                                                 mp->m_maxicount);
 807         statp->f_ffree = statp->f_files - (sbp->sb_icount - sbp->sb_ifree);
 808         XFS_SB_UNLOCK(mp, s);
 809
 810         id = huge_encode_dev(mp->m_dev);
 811         statp->f_fsid.val[0] = (u32)id;
 812         statp->f_fsid.val[1] = (u32)(id >> 32);
 813         statp->f_namelen = MAXNAMELEN - 1;
 814
 815         return 0;
 816 }
 817
 818
 819 /*
 820  * xfs_sync flushes any pending I/O to file system vfsp.
 821  *
 822  * This routine is called by vfs_sync() to make sure that things make it
 823  * out to disk eventually, on sync() system calls to flush out everything,
 824  * and when the file system is unmounted.  For the vfs_sync() case, all
 825  * we really need to do is sync out the log to make all of our meta-data
 826  * updates permanent (except for timestamps).  For calls from pflushd(),
 827  * dirty pages are kept moving by calling pdflush() on the inodes
 828  * containing them.  We also flush the inodes that we can lock without
 829  * sleeping and the superblock if we can lock it without sleeping from
 830  * vfs_sync() so that items at the tail of the log are always moving out.
 831  *
 832  * Flags:
 833  *      SYNC_BDFLUSH - We're being called from vfs_sync() so we don't want
 834  *                     to sleep if we can help it.  All we really need
 835  *                     to do is ensure that the log is synced at least
 836  *                     periodically.  We also push the inodes and
 837  *                     superblock if we can lock them without sleeping
 838  *                      and they are not pinned.
 839  *      SYNC_ATTR    - We need to flush the inodes.  If SYNC_BDFLUSH is not
 840  *                     set, then we really want to lock each inode and flush
 841  *                     it.
 842  *      SYNC_WAIT    - All the flushes that take place in this call should
 843  *                     be synchronous.
 844  *      SYNC_DELWRI  - This tells us to push dirty pages associated with
 845  *                     inodes.  SYNC_WAIT and SYNC_BDFLUSH are used to
 846  *                     determine if they should be flushed sync, async, or
 847  *                     delwri.
 848  *      SYNC_CLOSE   - This flag is passed when the system is being
 849  *                     unmounted.  We should sync and invalidate everthing.
 850  *      SYNC_FSDATA  - This indicates that the caller would like to make
 851  *                     sure the superblock is safe on disk.  We can ensure
 852  *                     this by simply makeing sure the log gets flushed
 853  *                     if SYNC_BDFLUSH is set, and by actually writing it
 854  *                     out otherwise.
 855  *
 856  */
 857 /*ARGSUSED*/
 858 STATIC int
 859 xfs_sync(
 860         bhv_desc_t      *bdp,
 861         int             flags,
 862         cred_t          *credp)
 863 {
 864         xfs_mount_t     *mp;
 865
 866         mp = XFS_BHVTOM(bdp);
 867         return (xfs_syncsub(mp, flags, 0, NULL));
 868 }
 869
 870 /*
 871  * xfs sync routine for internal use
 872  *
 873  * This routine supports all of the flags defined for the generic VFS_SYNC
 874  * interface as explained above under xfs_sync.  In the interests of not
 875  * changing interfaces within the 6.5 family, additional internallly-
 876  * required functions are specified within a separate xflags parameter,
 877  * only available by calling this routine.
 878  *
 879  */
 880 STATIC int
 881 xfs_sync_inodes(
 882         xfs_mount_t     *mp,
 883         int             flags,
 884         int             xflags,
 885         int             *bypassed)
 886 {
 887         xfs_inode_t     *ip = NULL;
 888         xfs_inode_t     *ip_next;
 889         xfs_buf_t       *bp;
 890         vnode_t         *vp = NULL;
 891         vmap_t          vmap;
 892         int             error;
 893         int             last_error;
 894         uint64_t        fflag;
 895         uint            lock_flags;
 896         uint            base_lock_flags;
 897         boolean_t       mount_locked;
 898         boolean_t       vnode_refed;
 899         int             preempt;
 900         xfs_dinode_t    *dip;
 901         xfs_iptr_t      *ipointer;
 902 #ifdef DEBUG
 903         boolean_t       ipointer_in = B_FALSE;
 904
 905 #define IPOINTER_SET    ipointer_in = B_TRUE
 906 #define IPOINTER_CLR    ipointer_in = B_FALSE
 907 #else
 908 #define IPOINTER_SET
 909 #define IPOINTER_CLR
 910 #endif
 911
 912
 913 /* Insert a marker record into the inode list after inode ip. The list
 914  * must be locked when this is called. After the call the list will no
 915  * longer be locked.
 916  */
 917 #define IPOINTER_INSERT(ip, mp) { \
 918                 ASSERT(ipointer_in == B_FALSE); \
 919                 ipointer->ip_mnext = ip->i_mnext; \
 920                 ipointer->ip_mprev = ip; \
 921                 ip->i_mnext = (xfs_inode_t *)ipointer; \
 922                 ipointer->ip_mnext->i_mprev = (xfs_inode_t *)ipointer; \
 923                 preempt = 0; \
 924                 XFS_MOUNT_IUNLOCK(mp); \
 925                 mount_locked = B_FALSE; \
 926                 IPOINTER_SET; \
 927         }
 928
 929 /* Remove the marker from the inode list. If the marker was the only item
 930  * in the list then there are no remaining inodes and we should zero out
 931  * the whole list. If we are the current head of the list then move the head
 932  * past us.
 933  */
 934 #define IPOINTER_REMOVE(ip, mp) { \
 935                 ASSERT(ipointer_in == B_TRUE); \
 936                 if (ipointer->ip_mnext != (xfs_inode_t *)ipointer) { \
 937                         ip = ipointer->ip_mnext; \
 938                         ip->i_mprev = ipointer->ip_mprev; \
 939                         ipointer->ip_mprev->i_mnext = ip; \
 940                         if (mp->m_inodes == (xfs_inode_t *)ipointer) { \
 941                                 mp->m_inodes = ip; \
 942                         } \
 943                 } else { \
 944                         ASSERT(mp->m_inodes == (xfs_inode_t *)ipointer); \
 945                         mp->m_inodes = NULL; \
 946                         ip = NULL; \
 947                 } \
 948                 IPOINTER_CLR; \
 949         }
 950
 951 #define XFS_PREEMPT_MASK        0x7f
 952
 953         if (bypassed)
 954                 *bypassed = 0;
 955         if (XFS_MTOVFS(mp)->vfs_flag & VFS_RDONLY)
 956                 return 0;
 957         error = 0;
 958         last_error = 0;
 959         preempt = 0;
 960
 961         /* Allocate a reference marker */
 962         ipointer = (xfs_iptr_t *)kmem_zalloc(sizeof(xfs_iptr_t), KM_SLEEP);
 963
 964         fflag = XFS_B_ASYNC;            /* default is don't wait */
 965         if (flags & SYNC_BDFLUSH)
 966                 fflag = XFS_B_DELWRI;
 967         if (flags & SYNC_WAIT)
 968                 fflag = 0;              /* synchronous overrides all */
 969
 970         base_lock_flags = XFS_ILOCK_SHARED;
 971         if (flags & (SYNC_DELWRI | SYNC_CLOSE)) {
 972                 /*
 973                  * We need the I/O lock if we're going to call any of
 974                  * the flush/inval routines.
 975                  */
 976                 base_lock_flags |= XFS_IOLOCK_SHARED;
 977         }
 978
 979         XFS_MOUNT_ILOCK(mp);
 980
 981         ip = mp->m_inodes;
 982
 983         mount_locked = B_TRUE;
 984         vnode_refed  = B_FALSE;
 985
 986         IPOINTER_CLR;
 987
 988         do {
 989                 ASSERT(ipointer_in == B_FALSE);
 990                 ASSERT(vnode_refed == B_FALSE);
 991
 992                 lock_flags = base_lock_flags;
 993
 994                 /*
 995                  * There were no inodes in the list, just break out
 996                  * of the loop.
 997                  */
 998                 if (ip == NULL) {
 999                         break;
1000                 }
1001
1002                 /*
1003                  * We found another sync thread marker - skip it
1004                  */
1005                 if (ip->i_mount == NULL) {
1006                         ip = ip->i_mnext;
1007                         continue;
1008                 }
1009
1010                 vp = XFS_ITOV_NULL(ip);
1011
1012                 /*
1013                  * If the vnode is gone then this is being torn down,
1014                  * call reclaim if it is flushed, else let regular flush
1015                  * code deal with it later in the loop.
1016                  */
1017
1018                 if (vp == NULL) {
1019                         /* Skip ones already in reclaim */
1020                         if (ip->i_flags & XFS_IRECLAIM) {
1021                                 ip = ip->i_mnext;
1022                                 continue;
1023                         }
1024                         if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0) {
1025                                 ip = ip->i_mnext;
1026                         } else if ((xfs_ipincount(ip) == 0) &&
1027                                     xfs_iflock_nowait(ip)) {
1028                                 IPOINTER_INSERT(ip, mp);
1029
1030                                 xfs_finish_reclaim(ip, 1,
1031                                                 XFS_IFLUSH_DELWRI_ELSE_ASYNC);
1032
1033                                 XFS_MOUNT_ILOCK(mp);
1034                                 mount_locked = B_TRUE;
1035                                 IPOINTER_REMOVE(ip, mp);
1036                         } else {
1037                                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1038                                 ip = ip->i_mnext;
1039                         }
1040                         continue;
1041                 }
1042
1043                 if (XFS_FORCED_SHUTDOWN(mp) && !(flags & SYNC_CLOSE)) {
1044                         XFS_MOUNT_IUNLOCK(mp);
1045                         kmem_free(ipointer, sizeof(xfs_iptr_t));
1046                         return 0;
1047                 }
1048
1049                 /*
1050                  * If this is just vfs_sync() or pflushd() calling
1051                  * then we can skip inodes for which it looks like
1052                  * there is nothing to do.  Since we don't have the
1053                  * inode locked this is racey, but these are periodic
1054                  * calls so it doesn't matter.  For the others we want
1055                  * to know for sure, so we at least try to lock them.
1056                  */
1057                 if (flags & SYNC_BDFLUSH) {
1058                         if (((ip->i_itemp == NULL) ||
1059                              !(ip->i_itemp->ili_format.ilf_fields &
1060                                XFS_ILOG_ALL)) &&
1061                             (ip->i_update_core == 0)) {
1062                                 ip = ip->i_mnext;
1063                                 continue;
1064                         }
1065                 }
1066
1067                 /*
1068                  * Try to lock without sleeping.  We're out of order with
1069                  * the inode list lock here, so if we fail we need to drop
1070                  * the mount lock and try again.  If we're called from
1071                  * bdflush() here, then don't bother.
1072                  *
1073                  * The inode lock here actually coordinates with the
1074                  * almost spurious inode lock in xfs_ireclaim() to prevent
1075                  * the vnode we handle here without a reference from
1076                  * being freed while we reference it.  If we lock the inode
1077                  * while it's on the mount list here, then the spurious inode
1078                  * lock in xfs_ireclaim() after the inode is pulled from
1079                  * the mount list will sleep until we release it here.
1080                  * This keeps the vnode from being freed while we reference
1081                  * it.  It is also cheaper and simpler than actually doing
1082                  * a vn_get() for every inode we touch here.
1083                  */
1084                 if (xfs_ilock_nowait(ip, lock_flags) == 0) {
1085
1086                         if ((flags & SYNC_BDFLUSH) || (vp == NULL)) {
1087                                 ip = ip->i_mnext;
1088                                 continue;
1089                         }
1090
1091                         /*
1092                          * We need to unlock the inode list lock in order
1093                          * to lock the inode. Insert a marker record into
1094                          * the inode list to remember our position, dropping
1095                          * the lock is now done inside the IPOINTER_INSERT
1096                          * macro.
1097                          *
1098                          * We also use the inode list lock to protect us
1099                          * in taking a snapshot of the vnode version number
1100                          * for use in calling vn_get().
1101                          */
1102                         VMAP(vp, vmap);
1103                         IPOINTER_INSERT(ip, mp);
1104
1105                         vp = vn_get(vp, &vmap);
1106                         if (vp == NULL) {
1107                                 /*
1108                                  * The vnode was reclaimed once we let go
1109                                  * of the inode list lock.  Skip to the
1110                                  * next list entry. Remove the marker.
1111                                  */
1112
1113                                 XFS_MOUNT_ILOCK(mp);
1114
1115                                 mount_locked = B_TRUE;
1116                                 vnode_refed  = B_FALSE;
1117
1118                                 IPOINTER_REMOVE(ip, mp);
1119
1120                                 continue;
1121                         }
1122
1123                         xfs_ilock(ip, lock_flags);
1124
1125                         ASSERT(vp == XFS_ITOV(ip));
1126                         ASSERT(ip->i_mount == mp);
1127
1128                         vnode_refed = B_TRUE;
1129                 }
1130
1131                 /* From here on in the loop we may have a marker record
1132                  * in the inode list.
1133                  */
1134
1135                 if ((flags & SYNC_CLOSE)  && (vp != NULL)) {
1136                         /*
1137                          * This is the shutdown case.  We just need to
1138                          * flush and invalidate all the pages associated
1139                          * with the inode.  Drop the inode lock since
1140                          * we can't hold it across calls to the buffer
1141                          * cache.
1142                          *
1143                          * We don't set the VREMAPPING bit in the vnode
1144                          * here, because we don't hold the vnode lock
1145                          * exclusively.  It doesn't really matter, though,
1146                          * because we only come here when we're shutting
1147                          * down anyway.
1148                          */
1149                         xfs_iunlock(ip, XFS_ILOCK_SHARED);
1150
1151                         if (XFS_FORCED_SHUTDOWN(mp)) {
1152                                 VOP_TOSS_PAGES(vp, 0, -1, FI_REMAPF);
1153                         } else {
1154                                 VOP_FLUSHINVAL_PAGES(vp, 0, -1, FI_REMAPF);
1155                         }
1156
1157                         xfs_ilock(ip, XFS_ILOCK_SHARED);
1158
1159                 } else if ((flags & SYNC_DELWRI) && (vp != NULL)) {
1160                         if (VN_DIRTY(vp)) {
1161                                 /* We need to have dropped the lock here,
1162                                  * so insert a marker if we have not already
1163                                  * done so.
1164                                  */
1165                                 if (mount_locked) {
1166                                         IPOINTER_INSERT(ip, mp);
1167                                 }
1168
1169                                 /*
1170                                  * Drop the inode lock since we can't hold it
1171                                  * across calls to the buffer cache.
1172                                  */
1173                                 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1174                                 VOP_FLUSH_PAGES(vp, (xfs_off_t)0, -1,
1175                                                         fflag, FI_NONE, error);
1176                                 xfs_ilock(ip, XFS_ILOCK_SHARED);
1177                         }
1178
1179                 }
1180
1181                 if (flags & SYNC_BDFLUSH) {
1182                         if ((flags & SYNC_ATTR) &&
1183                             ((ip->i_update_core) ||
1184                              ((ip->i_itemp != NULL) &&
1185                               (ip->i_itemp->ili_format.ilf_fields != 0)))) {
1186
1187                                 /* Insert marker and drop lock if not already
1188                                  * done.
1189                                  */
1190                                 if (mount_locked) {
1191                                         IPOINTER_INSERT(ip, mp);
1192                                 }
1193
1194                                 /*
1195                                  * We don't want the periodic flushing of the
1196                                  * inodes by vfs_sync() to interfere with
1197                                  * I/O to the file, especially read I/O
1198                                  * where it is only the access time stamp
1199                                  * that is being flushed out.  To prevent
1200                                  * long periods where we have both inode
1201                                  * locks held shared here while reading the
1202                                  * inode's buffer in from disk, we drop the
1203                                  * inode lock while reading in the inode
1204                                  * buffer.  We have to release the buffer
1205                                  * and reacquire the inode lock so that they
1206                                  * are acquired in the proper order (inode
1207                                  * locks first).  The buffer will go at the
1208                                  * end of the lru chain, though, so we can
1209                                  * expect it to still be there when we go
1210                                  * for it again in xfs_iflush().
1211                                  */
1212                                 if ((xfs_ipincount(ip) == 0) &&
1213                                     xfs_iflock_nowait(ip)) {
1214
1215                                         xfs_ifunlock(ip);
1216                                         xfs_iunlock(ip, XFS_ILOCK_SHARED);
1217
1218                                         error = xfs_itobp(mp, NULL, ip,
1219                                                           &dip, &bp, 0);
1220                                         if (!error) {
1221                                                 xfs_buf_relse(bp);
1222                                         } else {
1223                                                 /* Bailing out, remove the
1224                                                  * marker and free it.
1225                                                  */
1226                                                 XFS_MOUNT_ILOCK(mp);
1227
1228                                                 IPOINTER_REMOVE(ip, mp);
1229
1230                                                 XFS_MOUNT_IUNLOCK(mp);
1231
1232                                                 ASSERT(!(lock_flags &
1233                                                         XFS_IOLOCK_SHARED));
1234
1235                                                 kmem_free(ipointer,
1236                                                         sizeof(xfs_iptr_t));
1237                                                 return (0);
1238                                         }
1239
1240                                         /*
1241                                          * Since we dropped the inode lock,
1242                                          * the inode may have been reclaimed.
1243                                          * Therefore, we reacquire the mount
1244                                          * lock and check to see if we were the
1245                                          * inode reclaimed. If this happened
1246                                          * then the ipointer marker will no
1247                                          * longer point back at us. In this
1248                                          * case, move ip along to the inode
1249                                          * after the marker, remove the marker
1250                                          * and continue.
1251                                          */
1252                                         XFS_MOUNT_ILOCK(mp);
1253                                         mount_locked = B_TRUE;
1254
1255                                         if (ip != ipointer->ip_mprev) {
1256                                                 IPOINTER_REMOVE(ip, mp);
1257
1258                                                 ASSERT(!vnode_refed);
1259                                                 ASSERT(!(lock_flags &
1260                                                         XFS_IOLOCK_SHARED));
1261                                                 continue;
1262                                         }
1263
1264                                         ASSERT(ip->i_mount == mp);
1265
1266                                         if (xfs_ilock_nowait(ip,
1267                                                     XFS_ILOCK_SHARED) == 0) {
1268                                                 ASSERT(ip->i_mount == mp);
1269                                                 /*
1270                                                  * We failed to reacquire
1271                                                  * the inode lock without
1272                                                  * sleeping, so just skip
1273                                                  * the inode for now.  We
1274                                                  * clear the ILOCK bit from
1275                                                  * the lock_flags so that we
1276                                                  * won't try to drop a lock
1277                                                  * we don't hold below.
1278                                                  */
1279                                                 lock_flags &= ~XFS_ILOCK_SHARED;
1280                                                 IPOINTER_REMOVE(ip_next, mp);
1281                                         } else if ((xfs_ipincount(ip) == 0) &&
1282                                                    xfs_iflock_nowait(ip)) {
1283                                                 ASSERT(ip->i_mount == mp);
1284                                                 /*
1285                                                  * Since this is vfs_sync()
1286                                                  * calling we only flush the
1287                                                  * inode out if we can lock
1288                                                  * it without sleeping and
1289                                                  * it is not pinned.  Drop
1290                                                  * the mount lock here so
1291                                                  * that we don't hold it for
1292                                                  * too long. We already have
1293                                                  * a marker in the list here.
1294                                                  */
1295                                                 XFS_MOUNT_IUNLOCK(mp);
1296                                                 mount_locked = B_FALSE;
1297                                                 error = xfs_iflush(ip,
1298                                                            XFS_IFLUSH_DELWRI);
1299                                         } else {
1300                                                 ASSERT(ip->i_mount == mp);
1301                                                 IPOINTER_REMOVE(ip_next, mp);
1302                                         }
1303                                 }
1304
1305                         }
1306
1307                 } else {
1308                         if ((flags & SYNC_ATTR) &&
1309                             ((ip->i_update_core) ||
1310                              ((ip->i_itemp != NULL) &&
1311                               (ip->i_itemp->ili_format.ilf_fields != 0)))) {
1312                                 if (mount_locked) {
1313                                         IPOINTER_INSERT(ip, mp);
1314                                 }
1315
1316                                 if (flags & SYNC_WAIT) {
1317                                         xfs_iflock(ip);
1318                                         error = xfs_iflush(ip,
1319                                                            XFS_IFLUSH_SYNC);
1320                                 } else {
1321                                         /*
1322                                          * If we can't acquire the flush
1323                                          * lock, then the inode is already
1324                                          * being flushed so don't bother
1325                                          * waiting.  If we can lock it then
1326                                          * do a delwri flush so we can
1327                                          * combine multiple inode flushes
1328                                          * in each disk write.
1329                                          */
1330                                         if (xfs_iflock_nowait(ip)) {
1331                                                 error = xfs_iflush(ip,
1332                                                            XFS_IFLUSH_DELWRI);
1333                                         }
1334                                         else if (bypassed)
1335                                                 (*bypassed)++;
1336                                 }
1337                         }
1338                 }
1339
1340                 if (lock_flags != 0) {
1341                         xfs_iunlock(ip, lock_flags);
1342                 }
1343
1344                 if (vnode_refed) {
1345                         /*
1346                          * If we had to take a reference on the vnode
1347                          * above, then wait until after we've unlocked
1348                          * the inode to release the reference.  This is
1349                          * because we can be already holding the inode
1350                          * lock when VN_RELE() calls xfs_inactive().
1351                          *
1352                          * Make sure to drop the mount lock before calling
1353                          * VN_RELE() so that we don't trip over ourselves if
1354                          * we have to go for the mount lock again in the
1355                          * inactive code.
1356                          */
1357                         if (mount_locked) {
1358                                 IPOINTER_INSERT(ip, mp);
1359                         }
1360
1361                         VN_RELE(vp);
1362
1363                         vnode_refed = B_FALSE;
1364                 }
1365
1366                 if (error) {
1367                         last_error = error;
1368                 }
1369
1370                 /*
1371                  * bail out if the filesystem is corrupted.
1372                  */
1373                 if (error == EFSCORRUPTED)  {
1374                         if (!mount_locked) {
1375                                 XFS_MOUNT_ILOCK(mp);
1376                                 IPOINTER_REMOVE(ip, mp);
1377                         }
1378                         XFS_MOUNT_IUNLOCK(mp);
1379                         ASSERT(ipointer_in == B_FALSE);
1380                         kmem_free(ipointer, sizeof(xfs_iptr_t));
1381                         return XFS_ERROR(error);
1382                 }
1383
1384                 /* Let other threads have a chance at the mount lock
1385                  * if we have looped many times without dropping the
1386                  * lock.
1387                  */
1388                 if ((++preempt & XFS_PREEMPT_MASK) == 0) {
1389                         if (mount_locked) {
1390                                 IPOINTER_INSERT(ip, mp);
1391                         }
1392                 }
1393
1394                 if (mount_locked == B_FALSE) {
1395                         XFS_MOUNT_ILOCK(mp);
1396                         mount_locked = B_TRUE;
1397                         IPOINTER_REMOVE(ip, mp);
1398                         continue;
1399                 }
1400
1401                 ASSERT(ipointer_in == B_FALSE);
1402                 ip = ip->i_mnext;
1403
1404         } while (ip != mp->m_inodes);
1405
1406         XFS_MOUNT_IUNLOCK(mp);
1407
1408         ASSERT(ipointer_in == B_FALSE);
1409
1410         kmem_free(ipointer, sizeof(xfs_iptr_t));
1411         return XFS_ERROR(last_error);
1412 }
1413
1414 /*
1415  * xfs sync routine for internal use
1416  *
1417  * This routine supports all of the flags defined for the generic VFS_SYNC
1418  * interface as explained above under xfs_sync.  In the interests of not
1419  * changing interfaces within the 6.5 family, additional internallly-
1420  * required functions are specified within a separate xflags parameter,
1421  * only available by calling this routine.
1422  *
1423  */
1424 int
1425 xfs_syncsub(
1426         xfs_mount_t     *mp,
1427         int             flags,
1428         int             xflags,
1429         int             *bypassed)
1430 {
1431         int             error = 0;
1432         int             last_error = 0;
1433         uint            log_flags = XFS_LOG_FORCE;
1434         xfs_buf_t       *bp;
1435         xfs_buf_log_item_t      *bip;
1436
1437         /*
1438          * Sync out the log.  This ensures that the log is periodically
1439          * flushed even if there is not enough activity to fill it up.
1440          */
1441         if (flags & SYNC_WAIT)
1442                 log_flags |= XFS_LOG_SYNC;
1443
1444         xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1445
1446         if (flags & (SYNC_ATTR|SYNC_DELWRI)) {
1447                 if (flags & SYNC_BDFLUSH)
1448                         xfs_finish_reclaim_all(mp, 1);
1449                 else
1450                         error = xfs_sync_inodes(mp, flags, xflags, bypassed);
1451         }
1452
1453         /*
1454          * Flushing out dirty data above probably generated more
1455          * log activity, so if this isn't vfs_sync() then flush
1456          * the log again.
1457          */
1458         if (flags & SYNC_DELWRI) {
1459                 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1460         }
1461
1462         if (flags & SYNC_FSDATA) {
1463                 /*
1464                  * If this is vfs_sync() then only sync the superblock
1465                  * if we can lock it without sleeping and it is not pinned.
1466                  */
1467                 if (flags & SYNC_BDFLUSH) {
1468                         bp = xfs_getsb(mp, XFS_BUF_TRYLOCK);
1469                         if (bp != NULL) {
1470                                 bip = XFS_BUF_FSPRIVATE(bp,xfs_buf_log_item_t*);
1471                                 if ((bip != NULL) &&
1472                                     xfs_buf_item_dirty(bip)) {
1473                                         if (!(XFS_BUF_ISPINNED(bp))) {
1474                                                 XFS_BUF_ASYNC(bp);
1475                                                 error = xfs_bwrite(mp, bp);
1476                                         } else {
1477                                                 xfs_buf_relse(bp);
1478                                         }
1479                                 } else {
1480                                         xfs_buf_relse(bp);
1481                                 }
1482                         }
1483                 } else {
1484                         bp = xfs_getsb(mp, 0);
1485                         /*
1486                          * If the buffer is pinned then push on the log so
1487                          * we won't get stuck waiting in the write for
1488                          * someone, maybe ourselves, to flush the log.
1489                          * Even though we just pushed the log above, we
1490                          * did not have the superblock buffer locked at
1491                          * that point so it can become pinned in between
1492                          * there and here.
1493                          */
1494                         if (XFS_BUF_ISPINNED(bp))
1495                                 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
1496                         if (flags & SYNC_WAIT)
1497                                 XFS_BUF_UNASYNC(bp);
1498                         else
1499                                 XFS_BUF_ASYNC(bp);
1500                         error = xfs_bwrite(mp, bp);
1501                 }
1502                 if (error) {
1503                         last_error = error;
1504                 }
1505         }
1506
1507         /*
1508          * If this is the periodic sync, then kick some entries out of
1509          * the reference cache.  This ensures that idle entries are
1510          * eventually kicked out of the cache.
1511          */
1512         if (flags & SYNC_REFCACHE) {
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 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_fid_t       *xfid;
1571         xfs_inode_t     *ip;
1572         int             error;
1573         xfs_ino_t       ino;
1574         unsigned int    igen;
1575         xfs_mount_t     *mp;
1576
1577         xfid  = (struct xfs_fid *)fidp;
1578         if (xfid->xfs_fid_len == sizeof(*xfid) - sizeof(xfid->xfs_fid_len)) {
1579                 ino  = xfid->xfs_fid_ino;
1580                 igen = xfid->xfs_fid_gen;
1581         } else {
1582                 /*
1583                  * Invalid.  Since handles can be created in user space
1584                  * and passed in via gethandle(), this is not cause for
1585                  * a panic.
1586                  */
1587                 return XFS_ERROR(EINVAL);
1588         }
1589         mp = XFS_BHVTOM(bdp);
1590         error = xfs_iget(mp, NULL, ino, XFS_ILOCK_SHARED, &ip, 0);
1591         if (error) {
1592                 *vpp = NULL;
1593                 return error;
1594         }
1595         if (ip == NULL) {
1596                 *vpp = NULL;
1597                 return XFS_ERROR(EIO);
1598         }
1599
1600         if (ip->i_d.di_mode == 0 || (igen && (ip->i_d.di_gen != igen))) {
1601                 xfs_iput_new(ip, XFS_ILOCK_SHARED);
1602                 *vpp = NULL;
1603                 return XFS_ERROR(ENOENT);
1604         }
1605
1606         *vpp = XFS_ITOV(ip);
1607         xfs_iunlock(ip, XFS_ILOCK_SHARED);
1608         return 0;
1609 }
1610
1611
1612 #define MNTOPT_LOGBUFS  "logbufs"       /* number of XFS log buffers */
1613 #define MNTOPT_LOGBSIZE "logbsize"      /* size of XFS log buffers */
1614 #define MNTOPT_LOGDEV   "logdev"        /* log device */
1615 #define MNTOPT_RTDEV    "rtdev"         /* realtime I/O device */
1616 #define MNTOPT_BIOSIZE  "biosize"       /* log2 of preferred buffered io size */
1617 #define MNTOPT_WSYNC    "wsync"         /* safe-mode nfs compatible mount */
1618 #define MNTOPT_INO64    "ino64"         /* force inodes into 64-bit range */
1619 #define MNTOPT_NOALIGN  "noalign"       /* turn off stripe alignment */
1620 #define MNTOPT_SWALLOC  "swalloc"       /* turn on stripe width allocation */
1621 #define MNTOPT_SUNIT    "sunit"         /* data volume stripe unit */
1622 #define MNTOPT_SWIDTH   "swidth"        /* data volume stripe width */
1623 #define MNTOPT_NOUUID   "nouuid"        /* ignore filesystem UUID */
1624 #define MNTOPT_MTPT     "mtpt"          /* filesystem mount point */
1625 #define MNTOPT_NORECOVERY   "norecovery"   /* don't run XFS recovery */
1626 #define MNTOPT_NOLOGFLUSH   "nologflush"   /* don't hard flush on log writes */
1627 #define MNTOPT_OSYNCISOSYNC "osyncisosync" /* o_sync is REALLY o_sync */
1628 #define MNTOPT_64BITINODE   "inode64"   /* inodes can be allocated anywhere */
1629 #define MNTOPT_IKEEP    "ikeep"         /* do not free empty inode clusters */
1630 #define MNTOPT_NOIKEEP  "noikeep"       /* free empty inode clusters */
1631
1632
1633 int
1634 xfs_parseargs(
1635         struct bhv_desc         *bhv,
1636         char                    *options,
1637         struct xfs_mount_args   *args,
1638         int                     update)
1639 {
1640         struct vfs              *vfsp = bhvtovfs(bhv);
1641         char                    *this_char, *value, *eov;
1642         int                     dsunit, dswidth, vol_dsunit, vol_dswidth;
1643         int                     iosize;
1644
1645 #if 0   /* XXX: off by default, until some remaining issues ironed out */
1646         args->flags |= XFSMNT_IDELETE; /* default to on */
1647 #endif
1648
1649         if (!options)
1650                 return 0;
1651
1652         iosize = dsunit = dswidth = vol_dsunit = vol_dswidth = 0;
1653
1654         while ((this_char = strsep(&options, ",")) != NULL) {
1655                 if (!*this_char)
1656                         continue;
1657                 if ((value = strchr(this_char, '=')) != NULL)
1658                         *value++ = 0;
1659
1660                 if (!strcmp(this_char, MNTOPT_LOGBUFS)) {
1661                         if (!value || !*value) {
1662                                 printk("XFS: %s option requires an argument\n",
1663                                         MNTOPT_LOGBUFS);
1664                                 return EINVAL;
1665                         }
1666                         args->logbufs = simple_strtoul(value, &eov, 10);
1667                 } else if (!strcmp(this_char, MNTOPT_LOGBSIZE)) {
1668                         int     last, in_kilobytes = 0;
1669
1670                         if (!value || !*value) {
1671                                 printk("XFS: %s option requires an argument\n",
1672                                         MNTOPT_LOGBSIZE);
1673                                 return EINVAL;
1674                         }
1675                         last = strlen(value) - 1;
1676                         if (value[last] == 'K' || value[last] == 'k') {
1677                                 in_kilobytes = 1;
1678                                 value[last] = '\0';
1679                         }
1680                         args->logbufsize = simple_strtoul(value, &eov, 10);
1681                         if (in_kilobytes)
1682                                 args->logbufsize <<= 10;
1683                 } else if (!strcmp(this_char, MNTOPT_LOGDEV)) {
1684                         if (!value || !*value) {
1685                                 printk("XFS: %s option requires an argument\n",
1686                                         MNTOPT_LOGDEV);
1687                                 return EINVAL;
1688                         }
1689                         strncpy(args->logname, value, MAXNAMELEN);
1690                 } else if (!strcmp(this_char, MNTOPT_MTPT)) {
1691                         if (!value || !*value) {
1692                                 printk("XFS: %s option requires an argument\n",
1693                                         MNTOPT_MTPT);
1694                                 return EINVAL;
1695                         }
1696                         strncpy(args->mtpt, value, MAXNAMELEN);
1697                 } else if (!strcmp(this_char, MNTOPT_RTDEV)) {
1698                         if (!value || !*value) {
1699                                 printk("XFS: %s option requires an argument\n",
1700                                         MNTOPT_RTDEV);
1701                                 return EINVAL;
1702                         }
1703                         strncpy(args->rtname, value, MAXNAMELEN);
1704                 } else if (!strcmp(this_char, MNTOPT_BIOSIZE)) {
1705                         if (!value || !*value) {
1706                                 printk("XFS: %s option requires an argument\n",
1707                                         MNTOPT_BIOSIZE);
1708                                 return EINVAL;
1709                         }
1710                         iosize = simple_strtoul(value, &eov, 10);
1711                         args->flags |= XFSMNT_IOSIZE;
1712                         args->iosizelog = (uint8_t) iosize;
1713                 } else if (!strcmp(this_char, MNTOPT_WSYNC)) {
1714                         args->flags |= XFSMNT_WSYNC;
1715                 } else if (!strcmp(this_char, MNTOPT_OSYNCISOSYNC)) {
1716                         args->flags |= XFSMNT_OSYNCISOSYNC;
1717                 } else if (!strcmp(this_char, MNTOPT_NORECOVERY)) {
1718                         args->flags |= XFSMNT_NORECOVERY;
1719                 } else if (!strcmp(this_char, MNTOPT_INO64)) {
1720                         args->flags |= XFSMNT_INO64;
1721 #if !XFS_BIG_INUMS
1722                         printk("XFS: %s option not allowed on this system\n",
1723                                 MNTOPT_INO64);
1724                         return EINVAL;
1725 #endif
1726                 } else if (!strcmp(this_char, MNTOPT_NOALIGN)) {
1727                         args->flags |= XFSMNT_NOALIGN;
1728                 } else if (!strcmp(this_char, MNTOPT_SWALLOC)) {
1729                         args->flags |= XFSMNT_SWALLOC;
1730                 } else if (!strcmp(this_char, MNTOPT_SUNIT)) {
1731                         if (!value || !*value) {
1732                                 printk("XFS: %s option requires an argument\n",
1733                                         MNTOPT_SUNIT);
1734                                 return EINVAL;
1735                         }
1736                         dsunit = simple_strtoul(value, &eov, 10);
1737                 } else if (!strcmp(this_char, MNTOPT_SWIDTH)) {
1738                         if (!value || !*value) {
1739                                 printk("XFS: %s option requires an argument\n",
1740                                         MNTOPT_SWIDTH);
1741                                 return EINVAL;
1742                         }
1743                         dswidth = simple_strtoul(value, &eov, 10);
1744                 } else if (!strcmp(this_char, MNTOPT_64BITINODE)) {
1745                         args->flags &= ~XFSMNT_32BITINODES;
1746 #if !XFS_BIG_INUMS
1747                         printk("XFS: %s option not allowed on this system\n",
1748                                 MNTOPT_64BITINODE);
1749                         return EINVAL;
1750 #endif
1751                 } else if (!strcmp(this_char, MNTOPT_NOUUID)) {
1752                         args->flags |= XFSMNT_NOUUID;
1753                 } else if (!strcmp(this_char, MNTOPT_NOLOGFLUSH)) {
1754                         args->flags |= XFSMNT_NOLOGFLUSH;
1755                 } else if (!strcmp(this_char, MNTOPT_IKEEP)) {
1756                         args->flags &= ~XFSMNT_IDELETE;
1757                 } else if (!strcmp(this_char, MNTOPT_NOIKEEP)) {
1758                         args->flags |= XFSMNT_IDELETE;
1759                 } else if (!strcmp(this_char, "osyncisdsync")) {
1760                         /* no-op, this is now the default */
1761 printk("XFS: osyncisdsync is now the default, option is deprecated.\n");
1762                 } else if (!strcmp(this_char, "irixsgid")) {
1763 printk("XFS: irixsgid is now a sysctl(2) variable, option is deprecated.\n");
1764                 } else {
1765                         printk("XFS: unknown mount option [%s].\n", this_char);
1766                         return EINVAL;
1767                 }
1768         }
1769
1770         if (args->flags & XFSMNT_NORECOVERY) {
1771                 if ((vfsp->vfs_flag & VFS_RDONLY) == 0) {
1772                         printk("XFS: no-recovery mounts must be read-only.\n");
1773                         return EINVAL;
1774                 }
1775         }
1776
1777         if ((args->flags & XFSMNT_NOALIGN) && (dsunit || dswidth)) {
1778                 printk(
1779         "XFS: sunit and swidth options incompatible with the noalign option\n");
1780                 return EINVAL;
1781         }
1782
1783         if ((dsunit && !dswidth) || (!dsunit && dswidth)) {
1784                 printk("XFS: sunit and swidth must be specified together\n");
1785                 return EINVAL;
1786         }
1787
1788         if (dsunit && (dswidth % dsunit != 0)) {
1789                 printk(
1790         "XFS: stripe width (%d) must be a multiple of the stripe unit (%d)\n",
1791                         dswidth, dsunit);
1792                 return EINVAL;
1793         }
1794
1795         if ((args->flags & XFSMNT_NOALIGN) != XFSMNT_NOALIGN) {
1796                 if (dsunit) {
1797                         args->sunit = dsunit;
1798                         args->flags |= XFSMNT_RETERR;
1799                 } else {
1800                         args->sunit = vol_dsunit;
1801                 }
1802                 dswidth ? (args->swidth = dswidth) :
1803                           (args->swidth = vol_dswidth);
1804         } else {
1805                 args->sunit = args->swidth = 0;
1806         }
1807
1808         return 0;
1809 }
1810
1811 int
1812 xfs_showargs(
1813         struct bhv_desc         *bhv,
1814         struct seq_file         *m)
1815 {
1816         static struct proc_xfs_info {
1817                 int     flag;
1818                 char    *str;
1819         } xfs_info[] = {
1820                 /* the few simple ones we can get from the mount struct */
1821                 { XFS_MOUNT_WSYNC,              "," MNTOPT_WSYNC },
1822                 { XFS_MOUNT_INO64,              "," MNTOPT_INO64 },
1823                 { XFS_MOUNT_NOALIGN,            "," MNTOPT_NOALIGN },
1824                 { XFS_MOUNT_SWALLOC,            "," MNTOPT_SWALLOC },
1825                 { XFS_MOUNT_NOUUID,             "," MNTOPT_NOUUID },
1826                 { XFS_MOUNT_NORECOVERY,         "," MNTOPT_NORECOVERY },
1827                 { XFS_MOUNT_OSYNCISOSYNC,       "," MNTOPT_OSYNCISOSYNC },
1828                 { XFS_MOUNT_NOLOGFLUSH,         "," MNTOPT_NOLOGFLUSH },
1829                 { XFS_MOUNT_IDELETE,            "," MNTOPT_NOIKEEP },
1830                 { 0, NULL }
1831         };
1832         struct proc_xfs_info    *xfs_infop;
1833         struct xfs_mount        *mp = XFS_BHVTOM(bhv);
1834
1835         for (xfs_infop = xfs_info; xfs_infop->flag; xfs_infop++) {
1836                 if (mp->m_flags & xfs_infop->flag)
1837                         seq_puts(m, xfs_infop->str);
1838         }
1839
1840         if (mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)
1841                 seq_printf(m, "," MNTOPT_BIOSIZE "=%d", mp->m_writeio_log);
1842
1843         if (mp->m_logbufs > 0)
1844                 seq_printf(m, "," MNTOPT_LOGBUFS "=%d", mp->m_logbufs);
1845
1846         if (mp->m_logbsize > 0)
1847                 seq_printf(m, "," MNTOPT_LOGBSIZE "=%d", mp->m_logbsize);
1848
1849         if (mp->m_ddev_targp != mp->m_logdev_targp)
1850                 seq_printf(m, "," MNTOPT_LOGDEV "=%s",
1851                                 XFS_BUFTARG_NAME(mp->m_logdev_targp));
1852
1853         if (mp->m_rtdev_targp && mp->m_ddev_targp != mp->m_rtdev_targp)
1854                 seq_printf(m, "," MNTOPT_RTDEV "=%s",
1855                                 XFS_BUFTARG_NAME(mp->m_rtdev_targp));
1856
1857         if (mp->m_dalign > 0)
1858                 seq_printf(m, "," MNTOPT_SUNIT "=%d",
1859                                 (int)XFS_FSB_TO_BB(mp, mp->m_dalign));
1860
1861         if (mp->m_swidth > 0)
1862                 seq_printf(m, "," MNTOPT_SWIDTH "=%d",
1863                                 (int)XFS_FSB_TO_BB(mp, mp->m_swidth));
1864
1865         if (!(mp->m_flags & XFS_MOUNT_32BITINOOPT))
1866                 seq_printf(m, "," MNTOPT_64BITINODE);
1867
1868         return 0;
1869 }
1870
1871 STATIC void
1872 xfs_freeze(
1873         bhv_desc_t      *bdp)
1874 {
1875         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
1876
1877         while (atomic_read(&mp->m_active_trans) > 0)
1878                 delay(100);
1879
1880         /* Push the superblock and write an unmount record */
1881         xfs_log_unmount_write(mp);
1882         xfs_unmountfs_writesb(mp);
1883 }
1884
1885
1886 vfsops_t xfs_vfsops = {
1887         BHV_IDENTITY_INIT(VFS_BHV_XFS,VFS_POSITION_XFS),
1888         .vfs_parseargs          = xfs_parseargs,
1889         .vfs_showargs           = xfs_showargs,
1890         .vfs_mount              = xfs_mount,
1891         .vfs_unmount            = xfs_unmount,
1892         .vfs_mntupdate          = xfs_mntupdate,
1893         .vfs_root               = xfs_root,
1894         .vfs_statvfs            = xfs_statvfs,
1895         .vfs_sync               = xfs_sync,
1896         .vfs_vget               = xfs_vget,
1897         .vfs_dmapiops           = (vfs_dmapiops_t)fs_nosys,
1898         .vfs_quotactl           = (vfs_quotactl_t)fs_nosys,
1899         .vfs_init_vnode         = xfs_initialize_vnode,
1900         .vfs_force_shutdown     = xfs_do_force_shutdown,
1901         .vfs_freeze             = xfs_freeze,
1902 };