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         xfs_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         xfs_acl_zone_destroy(xfs_acl_zone);
 174
 175 #ifdef XFS_DIR2_TRACE
 176         ktrace_free(xfs_dir2_trace_buf);
 177 #endif
 178 #ifdef XFS_ATTR_TRACE
 179         ktrace_free(xfs_attr_trace_buf);
 180 #endif
 181 #ifdef XFS_DIR_TRACE
 182         ktrace_free(xfs_dir_trace_buf);
 183 #endif
 184 #ifdef XFS_BMBT_TRACE
 185         ktrace_free(xfs_bmbt_trace_buf);
 186 #endif
 187 #ifdef XFS_BMAP_TRACE
 188         ktrace_free(xfs_bmap_trace_buf);
 189 #endif
 190 #ifdef XFS_ALLOC_TRACE
 191         ktrace_free(xfs_alloc_trace_buf);
 192 #endif
 193
 194         kmem_cache_destroy(xfs_bmap_free_item_zone);
 195         kmem_cache_destroy(xfs_btree_cur_zone);
 196         kmem_cache_destroy(xfs_inode_zone);
 197         kmem_cache_destroy(xfs_trans_zone);
 198         kmem_cache_destroy(xfs_da_state_zone);
 199         kmem_cache_destroy(xfs_dabuf_zone);
 200         kmem_cache_destroy(xfs_buf_item_zone);
 201         kmem_cache_destroy(xfs_efd_zone);
 202         kmem_cache_destroy(xfs_efi_zone);
 203         kmem_cache_destroy(xfs_ifork_zone);
 204         kmem_cache_destroy(xfs_ili_zone);
 205         kmem_cache_destroy(xfs_chashlist_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          * Setup xfs_mount function vectors from available behaviors
 435          */
 436         p = vfs_bhv_lookup(vfsp, VFS_POSITION_DM);
 437         mp->m_dm_ops = p ? *(xfs_dmops_t *) vfs_bhv_custom(p) : xfs_dmcore_stub;
 438         p = vfs_bhv_lookup(vfsp, VFS_POSITION_QM);
 439         mp->m_qm_ops = p ? *(xfs_qmops_t *) vfs_bhv_custom(p) : xfs_qmcore_stub;
 440         p = vfs_bhv_lookup(vfsp, VFS_POSITION_IO);
 441         mp->m_io_ops = p ? *(xfs_ioops_t *) vfs_bhv_custom(p) : xfs_iocore_xfs;
 442
 443         /*
 444          * Open real time and log devices - order is important.
 445          */
 446         if (args->logname[0]) {
 447                 error = xfs_blkdev_get(mp, args->logname, &logdev);
 448                 if (error)
 449                         return error;
 450         }
 451         if (args->rtname[0]) {
 452                 error = xfs_blkdev_get(mp, args->rtname, &rtdev);
 453                 if (error) {
 454                         xfs_blkdev_put(logdev);
 455                         return error;
 456                 }
 457
 458                 if (rtdev == ddev || rtdev == logdev) {
 459                         cmn_err(CE_WARN,
 460         "XFS: Cannot mount filesystem with identical rtdev and ddev/logdev.");
 461                         xfs_blkdev_put(logdev);
 462                         xfs_blkdev_put(rtdev);
 463                         return EINVAL;
 464                 }
 465         }
 466
 467         /*
 468          * Setup xfs_mount buffer target pointers
 469          */
 470         error = ENOMEM;
 471         mp->m_ddev_targp = xfs_alloc_buftarg(ddev);
 472         if (!mp->m_ddev_targp) {
 473                 xfs_blkdev_put(logdev);
 474                 xfs_blkdev_put(rtdev);
 475                 return error;
 476         }
 477         if (rtdev) {
 478                 mp->m_rtdev_targp = xfs_alloc_buftarg(rtdev);
 479                 if (!mp->m_rtdev_targp)
 480                         goto error0;
 481         }
 482         mp->m_logdev_targp = (logdev && logdev != ddev) ?
 483                                 xfs_alloc_buftarg(logdev) : mp->m_ddev_targp;
 484         if (!mp->m_logdev_targp)
 485                 goto error0;
 486
 487         /*
 488          * Setup flags based on mount(2) options and then the superblock
 489          */
 490         error = xfs_start_flags(vfsp, args, mp);
 491         if (error)
 492                 goto error1;
 493         error = xfs_readsb(mp);
 494         if (error)
 495                 goto error1;
 496         error = xfs_finish_flags(vfsp, args, mp);
 497         if (error)
 498                 goto error2;
 499
 500         /*
 501          * Setup xfs_mount buffer target pointers based on superblock
 502          */
 503         error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_blocksize,
 504                                     mp->m_sb.sb_sectsize);
 505         if (!error && logdev && logdev != ddev) {
 506                 unsigned int    log_sector_size = BBSIZE;
 507
 508                 if (XFS_SB_VERSION_HASSECTOR(&mp->m_sb))
 509                         log_sector_size = mp->m_sb.sb_logsectsize;
 510                 error = xfs_setsize_buftarg(mp->m_logdev_targp,
 511                                             mp->m_sb.sb_blocksize,
 512                                             log_sector_size);
 513         }
 514         if (!error && rtdev)
 515                 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
 516                                             mp->m_sb.sb_blocksize,
 517                                             mp->m_sb.sb_sectsize);
 518         if (error)
 519                 goto error2;
 520
 521         error = XFS_IOINIT(vfsp, args, flags);
 522         if (!error)
 523                 return 0;
 524 error2:
 525         if (mp->m_sb_bp)
 526                 xfs_freesb(mp);
 527 error1:
 528         xfs_binval(mp->m_ddev_targp);
 529         if (logdev && logdev != ddev)
 530                 xfs_binval(mp->m_logdev_targp);
 531         if (rtdev)
 532                 xfs_binval(mp->m_rtdev_targp);
 533 error0:
 534         xfs_unmountfs_close(mp, credp);
 535         return error;
 536 }
 537
 538 STATIC int
 539 xfs_unmount(
 540         bhv_desc_t      *bdp,
 541         int             flags,
 542         cred_t          *credp)
 543 {
 544         struct vfs      *vfsp = bhvtovfs(bdp);
 545         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
 546         xfs_inode_t     *rip;
 547         vnode_t         *rvp;
 548         int             unmount_event_wanted = 0;
 549         int             unmount_event_flags = 0;
 550         int             xfs_unmountfs_needed = 0;
 551         int             error;
 552
 553         rip = mp->m_rootip;
 554         rvp = XFS_ITOV(rip);
 555
 556         if (vfsp->vfs_flag & VFS_DMI) {
 557                 error = XFS_SEND_PREUNMOUNT(mp, vfsp,
 558                                 rvp, DM_RIGHT_NULL, rvp, DM_RIGHT_NULL,
 559                                 NULL, NULL, 0, 0,
 560                                 (mp->m_dmevmask & (1<<DM_EVENT_PREUNMOUNT))?
 561                                         0:DM_FLAGS_UNWANTED);
 562                         if (error)
 563                                 return XFS_ERROR(error);
 564                 unmount_event_wanted = 1;
 565                 unmount_event_flags = (mp->m_dmevmask & (1<<DM_EVENT_UNMOUNT))?
 566                                         0 : DM_FLAGS_UNWANTED;
 567         }
 568
 569         /*
 570          * First blow any referenced inode from this file system
 571          * out of the reference cache, and delete the timer.
 572          */
 573         xfs_refcache_purge_mp(mp);
 574
 575         XFS_bflush(mp->m_ddev_targp);
 576         error = xfs_unmount_flush(mp, 0);
 577         if (error)
 578                 goto out;
 579
 580         ASSERT(vn_count(rvp) == 1);
 581
 582         /*
 583          * Drop the reference count
 584          */
 585         VN_RELE(rvp);
 586
 587         /*
 588          * If we're forcing a shutdown, typically because of a media error,
 589          * we want to make sure we invalidate dirty pages that belong to
 590          * referenced vnodes as well.
 591          */
 592         if (XFS_FORCED_SHUTDOWN(mp)) {
 593                 error = xfs_sync(&mp->m_bhv,
 594                          (SYNC_WAIT | SYNC_CLOSE), credp);
 595                 ASSERT(error != EFSCORRUPTED);
 596         }
 597         xfs_unmountfs_needed = 1;
 598
 599 out:
 600         /*      Send DMAPI event, if required.
 601          *      Then do xfs_unmountfs() if needed.
 602          *      Then return error (or zero).
 603          */
 604         if (unmount_event_wanted) {
 605                 /* Note: mp structure must still exist for
 606                  * XFS_SEND_UNMOUNT() call.
 607                  */
 608                 XFS_SEND_UNMOUNT(mp, vfsp, error == 0 ? rvp : NULL,
 609                         DM_RIGHT_NULL, 0, error, unmount_event_flags);
 610         }
 611         if (xfs_unmountfs_needed) {
 612                 /*
 613                  * Call common unmount function to flush to disk
 614                  * and free the super block buffer & mount structures.
 615                  */
 616                 xfs_unmountfs(mp, credp);
 617         }
 618
 619         return XFS_ERROR(error);
 620 }
 621
 622 #define REMOUNT_READONLY_FLAGS  (SYNC_REMOUNT|SYNC_ATTR|SYNC_WAIT)
 623
 624 STATIC int
 625 xfs_mntupdate(
 626         bhv_desc_t                      *bdp,
 627         int                             *flags,
 628         struct xfs_mount_args           *args)
 629 {
 630         struct vfs      *vfsp = bhvtovfs(bdp);
 631         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
 632         int             pincount, error;
 633         int             count = 0;
 634
 635         if (args->flags & XFSMNT_NOATIME)
 636                 mp->m_flags |= XFS_MOUNT_NOATIME;
 637         else
 638                 mp->m_flags &= ~XFS_MOUNT_NOATIME;
 639
 640         if (!(vfsp->vfs_flag & VFS_RDONLY)) {
 641                 VFS_SYNC(vfsp, SYNC_FSDATA|SYNC_BDFLUSH|SYNC_ATTR, NULL, error);
 642         }
 643
 644         if (*flags & MS_RDONLY) {
 645                 xfs_refcache_purge_mp(mp);
 646                 xfs_flush_buftarg(mp->m_ddev_targp, 0);
 647                 xfs_finish_reclaim_all(mp, 0);
 648
 649                 /* This loop must run at least twice.
 650                  * The first instance of the loop will flush
 651                  * most meta data but that will generate more
 652                  * meta data (typically directory updates).
 653                  * Which then must be flushed and logged before
 654                  * we can write the unmount record.
 655                  */
 656                 do {
 657                         VFS_SYNC(vfsp, REMOUNT_READONLY_FLAGS, NULL, error);
 658                         pincount = xfs_flush_buftarg(mp->m_ddev_targp, 1);
 659                         if (!pincount) {
 660                                 delay(50);
 661                                 count++;
 662                         }
 663                 } while (count < 2);
 664
 665                 /* Ok now write out an unmount record */
 666                 xfs_log_unmount_write(mp);
 667                 xfs_unmountfs_writesb(mp);
 668                 vfsp->vfs_flag |= VFS_RDONLY;
 669         } else {
 670                 vfsp->vfs_flag &= ~VFS_RDONLY;
 671         }
 672
 673         return 0;
 674 }
 675
 676 /*
 677  * xfs_unmount_flush implements a set of flush operation on special
 678  * inodes, which are needed as a separate set of operations so that
 679  * they can be called as part of relocation process.
 680  */
 681 int
 682 xfs_unmount_flush(
 683         xfs_mount_t     *mp,            /* Mount structure we are getting
 684                                            rid of. */
 685         int             relocation)     /* Called from vfs relocation. */
 686 {
 687         xfs_inode_t     *rip = mp->m_rootip;
 688         xfs_inode_t     *rbmip;
 689         xfs_inode_t     *rsumip = NULL;
 690         vnode_t         *rvp = XFS_ITOV(rip);
 691         int             error;
 692
 693         xfs_ilock(rip, XFS_ILOCK_EXCL);
 694         xfs_iflock(rip);
 695
 696         /*
 697          * Flush out the real time inodes.
 698          */
 699         if ((rbmip = mp->m_rbmip) != NULL) {
 700                 xfs_ilock(rbmip, XFS_ILOCK_EXCL);
 701                 xfs_iflock(rbmip);
 702                 error = xfs_iflush(rbmip, XFS_IFLUSH_SYNC);
 703                 xfs_iunlock(rbmip, XFS_ILOCK_EXCL);
 704
 705                 if (error == EFSCORRUPTED)
 706                         goto fscorrupt_out;
 707
 708                 ASSERT(vn_count(XFS_ITOV(rbmip)) == 1);
 709
 710                 rsumip = mp->m_rsumip;
 711                 xfs_ilock(rsumip, XFS_ILOCK_EXCL);
 712                 xfs_iflock(rsumip);
 713                 error = xfs_iflush(rsumip, XFS_IFLUSH_SYNC);
 714                 xfs_iunlock(rsumip, XFS_ILOCK_EXCL);
 715
 716                 if (error == EFSCORRUPTED)
 717                         goto fscorrupt_out;
 718
 719                 ASSERT(vn_count(XFS_ITOV(rsumip)) == 1);
 720         }
 721
 722         /*
 723          * Synchronously flush root inode to disk
 724          */
 725         error = xfs_iflush(rip, XFS_IFLUSH_SYNC);
 726         if (error == EFSCORRUPTED)
 727                 goto fscorrupt_out2;
 728
 729         if (vn_count(rvp) != 1 && !relocation) {
 730                 xfs_iunlock(rip, XFS_ILOCK_EXCL);
 731                 return XFS_ERROR(EBUSY);
 732         }
 733
 734         /*
 735          * Release dquot that rootinode, rbmino and rsumino might be holding,
 736          * flush and purge the quota inodes.
 737          */
 738         error = XFS_QM_UNMOUNT(mp);
 739         if (error == EFSCORRUPTED)
 740                 goto fscorrupt_out2;
 741
 742         if (rbmip) {
 743                 VN_RELE(XFS_ITOV(rbmip));
 744                 VN_RELE(XFS_ITOV(rsumip));
 745         }
 746
 747         xfs_iunlock(rip, XFS_ILOCK_EXCL);
 748         return 0;
 749
 750 fscorrupt_out:
 751         xfs_ifunlock(rip);
 752
 753 fscorrupt_out2:
 754         xfs_iunlock(rip, XFS_ILOCK_EXCL);
 755
 756         return XFS_ERROR(EFSCORRUPTED);
 757 }
 758
 759 /*
 760  * xfs_root extracts the root vnode from a vfs.
 761  *
 762  * vfsp -- the vfs struct for the desired file system
 763  * vpp  -- address of the caller's vnode pointer which should be
 764  *         set to the desired fs root vnode
 765  */
 766 STATIC int
 767 xfs_root(
 768         bhv_desc_t      *bdp,
 769         vnode_t         **vpp)
 770 {
 771         vnode_t         *vp;
 772
 773         vp = XFS_ITOV((XFS_BHVTOM(bdp))->m_rootip);
 774         VN_HOLD(vp);
 775         *vpp = vp;
 776         return 0;
 777 }
 778
 779 /*
 780  * xfs_statvfs
 781  *
 782  * Fill in the statvfs structure for the given file system.  We use
 783  * the superblock lock in the mount structure to ensure a consistent
 784  * snapshot of the counters returned.
 785  */
 786 STATIC int
 787 xfs_statvfs(
 788         bhv_desc_t      *bdp,
 789         xfs_statfs_t    *statp,
 790         vnode_t         *vp)
 791 {
 792         __uint64_t      fakeinos;
 793         xfs_extlen_t    lsize;
 794         xfs_mount_t     *mp;
 795         xfs_sb_t        *sbp;
 796         unsigned long   s;
 797         u64 id;
 798
 799         mp = XFS_BHVTOM(bdp);
 800         sbp = &(mp->m_sb);
 801
 802         statp->f_type = XFS_SB_MAGIC;
 803
 804         s = XFS_SB_LOCK(mp);
 805         statp->f_bsize = sbp->sb_blocksize;
 806         lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
 807         statp->f_blocks = sbp->sb_dblocks - lsize;
 808         statp->f_bfree = statp->f_bavail = sbp->sb_fdblocks;
 809         fakeinos = statp->f_bfree << sbp->sb_inopblog;
 810 #if XFS_BIG_INUMS
 811         fakeinos += mp->m_inoadd;
 812 #endif
 813         statp->f_files =
 814             MIN(sbp->sb_icount + fakeinos, (__uint64_t)XFS_MAXINUMBER);
 815         if (mp->m_maxicount)
 816 #if XFS_BIG_INUMS
 817                 if (!mp->m_inoadd)
 818 #endif
 819                         statp->f_files = min_t(typeof(statp->f_files),
 820                                                 statp->f_files,
 821                                                 mp->m_maxicount);
 822         statp->f_ffree = statp->f_files - (sbp->sb_icount - sbp->sb_ifree);
 823         XFS_SB_UNLOCK(mp, s);
 824
 825         id = huge_encode_dev(mp->m_dev);
 826         statp->f_fsid.val[0] = (u32)id;
 827         statp->f_fsid.val[1] = (u32)(id >> 32);
 828         statp->f_namelen = MAXNAMELEN - 1;
 829
 830         return 0;
 831 }
 832
 833
 834 /*
 835  * xfs_sync flushes any pending I/O to file system vfsp.
 836  *
 837  * This routine is called by vfs_sync() to make sure that things make it
 838  * out to disk eventually, on sync() system calls to flush out everything,
 839  * and when the file system is unmounted.  For the vfs_sync() case, all
 840  * we really need to do is sync out the log to make all of our meta-data
 841  * updates permanent (except for timestamps).  For calls from pflushd(),
 842  * dirty pages are kept moving by calling pdflush() on the inodes
 843  * containing them.  We also flush the inodes that we can lock without
 844  * sleeping and the superblock if we can lock it without sleeping from
 845  * vfs_sync() so that items at the tail of the log are always moving out.
 846  *
 847  * Flags:
 848  *      SYNC_BDFLUSH - We're being called from vfs_sync() so we don't want
 849  *                     to sleep if we can help it.  All we really need
 850  *                     to do is ensure that the log is synced at least
 851  *                     periodically.  We also push the inodes and
 852  *                     superblock if we can lock them without sleeping
 853  *                      and they are not pinned.
 854  *      SYNC_ATTR    - We need to flush the inodes.  If SYNC_BDFLUSH is not
 855  *                     set, then we really want to lock each inode and flush
 856  *                     it.
 857  *      SYNC_WAIT    - All the flushes that take place in this call should
 858  *                     be synchronous.
 859  *      SYNC_DELWRI  - This tells us to push dirty pages associated with
 860  *                     inodes.  SYNC_WAIT and SYNC_BDFLUSH are used to
 861  *                     determine if they should be flushed sync, async, or
 862  *                     delwri.
 863  *      SYNC_CLOSE   - This flag is passed when the system is being
 864  *                     unmounted.  We should sync and invalidate everthing.
 865  *      SYNC_FSDATA  - This indicates that the caller would like to make
 866  *                     sure the superblock is safe on disk.  We can ensure
 867  *                     this by simply makeing sure the log gets flushed
 868  *                     if SYNC_BDFLUSH is set, and by actually writing it
 869  *                     out otherwise.
 870  *
 871  */
 872 /*ARGSUSED*/
 873 STATIC int
 874 xfs_sync(
 875         bhv_desc_t      *bdp,
 876         int             flags,
 877         cred_t          *credp)
 878 {
 879         xfs_mount_t     *mp;
 880
 881         mp = XFS_BHVTOM(bdp);
 882         return (xfs_syncsub(mp, flags, 0, NULL));
 883 }
 884
 885 /*
 886  * xfs sync routine for internal use
 887  *
 888  * This routine supports all of the flags defined for the generic VFS_SYNC
 889  * interface as explained above under xfs_sync.  In the interests of not
 890  * changing interfaces within the 6.5 family, additional internallly-
 891  * required functions are specified within a separate xflags parameter,
 892  * only available by calling this routine.
 893  *
 894  */
 895 STATIC int
 896 xfs_sync_inodes(
 897         xfs_mount_t     *mp,
 898         int             flags,
 899         int             xflags,
 900         int             *bypassed)
 901 {
 902         xfs_inode_t     *ip = NULL;
 903         xfs_inode_t     *ip_next;
 904         xfs_buf_t       *bp;
 905         vnode_t         *vp = NULL;
 906         vmap_t          vmap;
 907         int             error;
 908         int             last_error;
 909         uint64_t        fflag;
 910         uint            lock_flags;
 911         uint            base_lock_flags;
 912         boolean_t       mount_locked;
 913         boolean_t       vnode_refed;
 914         int             preempt;
 915         xfs_dinode_t    *dip;
 916         xfs_iptr_t      *ipointer;
 917 #ifdef DEBUG
 918         boolean_t       ipointer_in = B_FALSE;
 919
 920 #define IPOINTER_SET    ipointer_in = B_TRUE
 921 #define IPOINTER_CLR    ipointer_in = B_FALSE
 922 #else
 923 #define IPOINTER_SET
 924 #define IPOINTER_CLR
 925 #endif
 926
 927
 928 /* Insert a marker record into the inode list after inode ip. The list
 929  * must be locked when this is called. After the call the list will no
 930  * longer be locked.
 931  */
 932 #define IPOINTER_INSERT(ip, mp) { \
 933                 ASSERT(ipointer_in == B_FALSE); \
 934                 ipointer->ip_mnext = ip->i_mnext; \
 935                 ipointer->ip_mprev = ip; \
 936                 ip->i_mnext = (xfs_inode_t *)ipointer; \
 937                 ipointer->ip_mnext->i_mprev = (xfs_inode_t *)ipointer; \
 938                 preempt = 0; \
 939                 XFS_MOUNT_IUNLOCK(mp); \
 940                 mount_locked = B_FALSE; \
 941                 IPOINTER_SET; \
 942         }
 943
 944 /* Remove the marker from the inode list. If the marker was the only item
 945  * in the list then there are no remaining inodes and we should zero out
 946  * the whole list. If we are the current head of the list then move the head
 947  * past us.
 948  */
 949 #define IPOINTER_REMOVE(ip, mp) { \
 950                 ASSERT(ipointer_in == B_TRUE); \
 951                 if (ipointer->ip_mnext != (xfs_inode_t *)ipointer) { \
 952                         ip = ipointer->ip_mnext; \
 953                         ip->i_mprev = ipointer->ip_mprev; \
 954                         ipointer->ip_mprev->i_mnext = ip; \
 955                         if (mp->m_inodes == (xfs_inode_t *)ipointer) { \
 956                                 mp->m_inodes = ip; \
 957                         } \
 958                 } else { \
 959                         ASSERT(mp->m_inodes == (xfs_inode_t *)ipointer); \
 960                         mp->m_inodes = NULL; \
 961                         ip = NULL; \
 962                 } \
 963                 IPOINTER_CLR; \
 964         }
 965
 966 #define XFS_PREEMPT_MASK        0x7f
 967
 968         if (bypassed)
 969                 *bypassed = 0;
 970         if (XFS_MTOVFS(mp)->vfs_flag & VFS_RDONLY)
 971                 return 0;
 972         error = 0;
 973         last_error = 0;
 974         preempt = 0;
 975
 976         /* Allocate a reference marker */
 977         ipointer = (xfs_iptr_t *)kmem_zalloc(sizeof(xfs_iptr_t), KM_SLEEP);
 978
 979         fflag = XFS_B_ASYNC;            /* default is don't wait */
 980         if (flags & SYNC_BDFLUSH)
 981                 fflag = XFS_B_DELWRI;
 982         if (flags & SYNC_WAIT)
 983                 fflag = 0;              /* synchronous overrides all */
 984
 985         base_lock_flags = XFS_ILOCK_SHARED;
 986         if (flags & (SYNC_DELWRI | SYNC_CLOSE)) {
 987                 /*
 988                  * We need the I/O lock if we're going to call any of
 989                  * the flush/inval routines.
 990                  */
 991                 base_lock_flags |= XFS_IOLOCK_SHARED;
 992         }
 993
 994         XFS_MOUNT_ILOCK(mp);
 995
 996         ip = mp->m_inodes;
 997
 998         mount_locked = B_TRUE;
 999         vnode_refed  = B_FALSE;
1000
1001         IPOINTER_CLR;
1002
1003         do {
1004                 ASSERT(ipointer_in == B_FALSE);
1005                 ASSERT(vnode_refed == B_FALSE);
1006
1007                 lock_flags = base_lock_flags;
1008
1009                 /*
1010                  * There were no inodes in the list, just break out
1011                  * of the loop.
1012                  */
1013                 if (ip == NULL) {
1014                         break;
1015                 }
1016
1017                 /*
1018                  * We found another sync thread marker - skip it
1019                  */
1020                 if (ip->i_mount == NULL) {
1021                         ip = ip->i_mnext;
1022                         continue;
1023                 }
1024
1025                 vp = XFS_ITOV_NULL(ip);
1026
1027                 /*
1028                  * If the vnode is gone then this is being torn down,
1029                  * call reclaim if it is flushed, else let regular flush
1030                  * code deal with it later in the loop.
1031                  */
1032
1033                 if (vp == NULL) {
1034                         /* Skip ones already in reclaim */
1035                         if (ip->i_flags & XFS_IRECLAIM) {
1036                                 ip = ip->i_mnext;
1037                                 continue;
1038                         }
1039                         if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0) {
1040                                 ip = ip->i_mnext;
1041                         } else if ((xfs_ipincount(ip) == 0) &&
1042                                     xfs_iflock_nowait(ip)) {
1043                                 IPOINTER_INSERT(ip, mp);
1044
1045                                 xfs_finish_reclaim(ip, 1,
1046                                                 XFS_IFLUSH_DELWRI_ELSE_ASYNC);
1047
1048                                 XFS_MOUNT_ILOCK(mp);
1049                                 mount_locked = B_TRUE;
1050                                 IPOINTER_REMOVE(ip, mp);
1051                         } else {
1052                                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1053                                 ip = ip->i_mnext;
1054                         }
1055                         continue;
1056                 }
1057
1058                 if (VN_BAD(vp)) {
1059                         ip = ip->i_mnext;
1060                         continue;
1061                 }
1062
1063                 if (XFS_FORCED_SHUTDOWN(mp) && !(flags & SYNC_CLOSE)) {
1064                         XFS_MOUNT_IUNLOCK(mp);
1065                         kmem_free(ipointer, sizeof(xfs_iptr_t));
1066                         return 0;
1067                 }
1068
1069                 /*
1070                  * If this is just vfs_sync() or pflushd() calling
1071                  * then we can skip inodes for which it looks like
1072                  * there is nothing to do.  Since we don't have the
1073                  * inode locked this is racey, but these are periodic
1074                  * calls so it doesn't matter.  For the others we want
1075                  * to know for sure, so we at least try to lock them.
1076                  */
1077                 if (flags & SYNC_BDFLUSH) {
1078                         if (((ip->i_itemp == NULL) ||
1079                              !(ip->i_itemp->ili_format.ilf_fields &
1080                                XFS_ILOG_ALL)) &&
1081                             (ip->i_update_core == 0)) {
1082                                 ip = ip->i_mnext;
1083                                 continue;
1084                         }
1085                 }
1086
1087                 /*
1088                  * Try to lock without sleeping.  We're out of order with
1089                  * the inode list lock here, so if we fail we need to drop
1090                  * the mount lock and try again.  If we're called from
1091                  * bdflush() here, then don't bother.
1092                  *
1093                  * The inode lock here actually coordinates with the
1094                  * almost spurious inode lock in xfs_ireclaim() to prevent
1095                  * the vnode we handle here without a reference from
1096                  * being freed while we reference it.  If we lock the inode
1097                  * while it's on the mount list here, then the spurious inode
1098                  * lock in xfs_ireclaim() after the inode is pulled from
1099                  * the mount list will sleep until we release it here.
1100                  * This keeps the vnode from being freed while we reference
1101                  * it.  It is also cheaper and simpler than actually doing
1102                  * a vn_get() for every inode we touch here.
1103                  */
1104                 if (xfs_ilock_nowait(ip, lock_flags) == 0) {
1105
1106                         if ((flags & SYNC_BDFLUSH) || (vp == NULL)) {
1107                                 ip = ip->i_mnext;
1108                                 continue;
1109                         }
1110
1111                         /*
1112                          * We need to unlock the inode list lock in order
1113                          * to lock the inode. Insert a marker record into
1114                          * the inode list to remember our position, dropping
1115                          * the lock is now done inside the IPOINTER_INSERT
1116                          * macro.
1117                          *
1118                          * We also use the inode list lock to protect us
1119                          * in taking a snapshot of the vnode version number
1120                          * for use in calling vn_get().
1121                          */
1122                         VMAP(vp, vmap);
1123                         IPOINTER_INSERT(ip, mp);
1124
1125                         vp = vn_get(vp, &vmap);
1126                         if (vp == NULL) {
1127                                 /*
1128                                  * The vnode was reclaimed once we let go
1129                                  * of the inode list lock.  Skip to the
1130                                  * next list entry. Remove the marker.
1131                                  */
1132
1133                                 XFS_MOUNT_ILOCK(mp);
1134
1135                                 mount_locked = B_TRUE;
1136                                 vnode_refed  = B_FALSE;
1137
1138                                 IPOINTER_REMOVE(ip, mp);
1139
1140                                 continue;
1141                         }
1142
1143                         xfs_ilock(ip, lock_flags);
1144
1145                         ASSERT(vp == XFS_ITOV(ip));
1146                         ASSERT(ip->i_mount == mp);
1147
1148                         vnode_refed = B_TRUE;
1149                 }
1150
1151                 /* From here on in the loop we may have a marker record
1152                  * in the inode list.
1153                  */
1154
1155                 if ((flags & SYNC_CLOSE)  && (vp != NULL)) {
1156                         /*
1157                          * This is the shutdown case.  We just need to
1158                          * flush and invalidate all the pages associated
1159                          * with the inode.  Drop the inode lock since
1160                          * we can't hold it across calls to the buffer
1161                          * cache.
1162                          *
1163                          * We don't set the VREMAPPING bit in the vnode
1164                          * here, because we don't hold the vnode lock
1165                          * exclusively.  It doesn't really matter, though,
1166                          * because we only come here when we're shutting
1167                          * down anyway.
1168                          */
1169                         xfs_iunlock(ip, XFS_ILOCK_SHARED);
1170
1171                         if (XFS_FORCED_SHUTDOWN(mp)) {
1172                                 VOP_TOSS_PAGES(vp, 0, -1, FI_REMAPF);
1173                         } else {
1174                                 VOP_FLUSHINVAL_PAGES(vp, 0, -1, FI_REMAPF);
1175                         }
1176
1177                         xfs_ilock(ip, XFS_ILOCK_SHARED);
1178
1179                 } else if ((flags & SYNC_DELWRI) && (vp != NULL)) {
1180                         if (VN_DIRTY(vp)) {
1181                                 /* We need to have dropped the lock here,
1182                                  * so insert a marker if we have not already
1183                                  * done so.
1184                                  */
1185                                 if (mount_locked) {
1186                                         IPOINTER_INSERT(ip, mp);
1187                                 }
1188
1189                                 /*
1190                                  * Drop the inode lock since we can't hold it
1191                                  * across calls to the buffer cache.
1192                                  */
1193                                 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1194                                 VOP_FLUSH_PAGES(vp, (xfs_off_t)0, -1,
1195                                                         fflag, FI_NONE, error);
1196                                 xfs_ilock(ip, XFS_ILOCK_SHARED);
1197                         }
1198
1199                 }
1200
1201                 if (flags & SYNC_BDFLUSH) {
1202                         if ((flags & SYNC_ATTR) &&
1203                             ((ip->i_update_core) ||
1204                              ((ip->i_itemp != NULL) &&
1205                               (ip->i_itemp->ili_format.ilf_fields != 0)))) {
1206
1207                                 /* Insert marker and drop lock if not already
1208                                  * done.
1209                                  */
1210                                 if (mount_locked) {
1211                                         IPOINTER_INSERT(ip, mp);
1212                                 }
1213
1214                                 /*
1215                                  * We don't want the periodic flushing of the
1216                                  * inodes by vfs_sync() to interfere with
1217                                  * I/O to the file, especially read I/O
1218                                  * where it is only the access time stamp
1219                                  * that is being flushed out.  To prevent
1220                                  * long periods where we have both inode
1221                                  * locks held shared here while reading the
1222                                  * inode's buffer in from disk, we drop the
1223                                  * inode lock while reading in the inode
1224                                  * buffer.  We have to release the buffer
1225                                  * and reacquire the inode lock so that they
1226                                  * are acquired in the proper order (inode
1227                                  * locks first).  The buffer will go at the
1228                                  * end of the lru chain, though, so we can
1229                                  * expect it to still be there when we go
1230                                  * for it again in xfs_iflush().
1231                                  */
1232                                 if ((xfs_ipincount(ip) == 0) &&
1233                                     xfs_iflock_nowait(ip)) {
1234
1235                                         xfs_ifunlock(ip);
1236                                         xfs_iunlock(ip, XFS_ILOCK_SHARED);
1237
1238                                         error = xfs_itobp(mp, NULL, ip,
1239                                                           &dip, &bp, 0);
1240                                         if (!error) {
1241                                                 xfs_buf_relse(bp);
1242                                         } else {
1243                                                 /* Bailing out, remove the
1244                                                  * marker and free it.
1245                                                  */
1246                                                 XFS_MOUNT_ILOCK(mp);
1247
1248                                                 IPOINTER_REMOVE(ip, mp);
1249
1250                                                 XFS_MOUNT_IUNLOCK(mp);
1251
1252                                                 ASSERT(!(lock_flags &
1253                                                         XFS_IOLOCK_SHARED));
1254
1255                                                 kmem_free(ipointer,
1256                                                         sizeof(xfs_iptr_t));
1257                                                 return (0);
1258                                         }
1259
1260                                         /*
1261                                          * Since we dropped the inode lock,
1262                                          * the inode may have been reclaimed.
1263                                          * Therefore, we reacquire the mount
1264                                          * lock and check to see if we were the
1265                                          * inode reclaimed. If this happened
1266                                          * then the ipointer marker will no
1267                                          * longer point back at us. In this
1268                                          * case, move ip along to the inode
1269                                          * after the marker, remove the marker
1270                                          * and continue.
1271                                          */
1272                                         XFS_MOUNT_ILOCK(mp);
1273                                         mount_locked = B_TRUE;
1274
1275                                         if (ip != ipointer->ip_mprev) {
1276                                                 IPOINTER_REMOVE(ip, mp);
1277
1278                                                 ASSERT(!vnode_refed);
1279                                                 ASSERT(!(lock_flags &
1280                                                         XFS_IOLOCK_SHARED));
1281                                                 continue;
1282                                         }
1283
1284                                         ASSERT(ip->i_mount == mp);
1285
1286                                         if (xfs_ilock_nowait(ip,
1287                                                     XFS_ILOCK_SHARED) == 0) {
1288                                                 ASSERT(ip->i_mount == mp);
1289                                                 /*
1290                                                  * We failed to reacquire
1291                                                  * the inode lock without
1292                                                  * sleeping, so just skip
1293                                                  * the inode for now.  We
1294                                                  * clear the ILOCK bit from
1295                                                  * the lock_flags so that we
1296                                                  * won't try to drop a lock
1297                                                  * we don't hold below.
1298                                                  */
1299                                                 lock_flags &= ~XFS_ILOCK_SHARED;
1300                                                 IPOINTER_REMOVE(ip_next, mp);
1301                                         } else if ((xfs_ipincount(ip) == 0) &&
1302                                                    xfs_iflock_nowait(ip)) {
1303                                                 ASSERT(ip->i_mount == mp);
1304                                                 /*
1305                                                  * Since this is vfs_sync()
1306                                                  * calling we only flush the
1307                                                  * inode out if we can lock
1308                                                  * it without sleeping and
1309                                                  * it is not pinned.  Drop
1310                                                  * the mount lock here so
1311                                                  * that we don't hold it for
1312                                                  * too long. We already have
1313                                                  * a marker in the list here.
1314                                                  */
1315                                                 XFS_MOUNT_IUNLOCK(mp);
1316                                                 mount_locked = B_FALSE;
1317                                                 error = xfs_iflush(ip,
1318                                                            XFS_IFLUSH_DELWRI);
1319                                         } else {
1320                                                 ASSERT(ip->i_mount == mp);
1321                                                 IPOINTER_REMOVE(ip_next, mp);
1322                                         }
1323                                 }
1324
1325                         }
1326
1327                 } else {
1328                         if ((flags & SYNC_ATTR) &&
1329                             ((ip->i_update_core) ||
1330                              ((ip->i_itemp != NULL) &&
1331                               (ip->i_itemp->ili_format.ilf_fields != 0)))) {
1332                                 if (mount_locked) {
1333                                         IPOINTER_INSERT(ip, mp);
1334                                 }
1335
1336                                 if (flags & SYNC_WAIT) {
1337                                         xfs_iflock(ip);
1338                                         error = xfs_iflush(ip,
1339                                                            XFS_IFLUSH_SYNC);
1340                                 } else {
1341                                         /*
1342                                          * If we can't acquire the flush
1343                                          * lock, then the inode is already
1344                                          * being flushed so don't bother
1345                                          * waiting.  If we can lock it then
1346                                          * do a delwri flush so we can
1347                                          * combine multiple inode flushes
1348                                          * in each disk write.
1349                                          */
1350                                         if (xfs_iflock_nowait(ip)) {
1351                                                 error = xfs_iflush(ip,
1352                                                            XFS_IFLUSH_DELWRI);
1353                                         }
1354                                         else if (bypassed)
1355                                                 (*bypassed)++;
1356                                 }
1357                         }
1358                 }
1359
1360                 if (lock_flags != 0) {
1361                         xfs_iunlock(ip, lock_flags);
1362                 }
1363
1364                 if (vnode_refed) {
1365                         /*
1366                          * If we had to take a reference on the vnode
1367                          * above, then wait until after we've unlocked
1368                          * the inode to release the reference.  This is
1369                          * because we can be already holding the inode
1370                          * lock when VN_RELE() calls xfs_inactive().
1371                          *
1372                          * Make sure to drop the mount lock before calling
1373                          * VN_RELE() so that we don't trip over ourselves if
1374                          * we have to go for the mount lock again in the
1375                          * inactive code.
1376                          */
1377                         if (mount_locked) {
1378                                 IPOINTER_INSERT(ip, mp);
1379                         }
1380
1381                         VN_RELE(vp);
1382
1383                         vnode_refed = B_FALSE;
1384                 }
1385
1386                 if (error) {
1387                         last_error = error;
1388                 }
1389
1390                 /*
1391                  * bail out if the filesystem is corrupted.
1392                  */
1393                 if (error == EFSCORRUPTED)  {
1394                         if (!mount_locked) {
1395                                 XFS_MOUNT_ILOCK(mp);
1396                                 IPOINTER_REMOVE(ip, mp);
1397                         }
1398                         XFS_MOUNT_IUNLOCK(mp);
1399                         ASSERT(ipointer_in == B_FALSE);
1400                         kmem_free(ipointer, sizeof(xfs_iptr_t));
1401                         return XFS_ERROR(error);
1402                 }
1403
1404                 /* Let other threads have a chance at the mount lock
1405                  * if we have looped many times without dropping the
1406                  * lock.
1407                  */
1408                 if ((++preempt & XFS_PREEMPT_MASK) == 0) {
1409                         if (mount_locked) {
1410                                 IPOINTER_INSERT(ip, mp);
1411                         }
1412                 }
1413
1414                 if (mount_locked == B_FALSE) {
1415                         XFS_MOUNT_ILOCK(mp);
1416                         mount_locked = B_TRUE;
1417                         IPOINTER_REMOVE(ip, mp);
1418                         continue;
1419                 }
1420
1421                 ASSERT(ipointer_in == B_FALSE);
1422                 ip = ip->i_mnext;
1423
1424         } while (ip != mp->m_inodes);
1425
1426         XFS_MOUNT_IUNLOCK(mp);
1427
1428         ASSERT(ipointer_in == B_FALSE);
1429
1430         kmem_free(ipointer, sizeof(xfs_iptr_t));
1431         return XFS_ERROR(last_error);
1432 }
1433
1434 /*
1435  * xfs sync routine for internal use
1436  *
1437  * This routine supports all of the flags defined for the generic VFS_SYNC
1438  * interface as explained above under xfs_sync.  In the interests of not
1439  * changing interfaces within the 6.5 family, additional internallly-
1440  * required functions are specified within a separate xflags parameter,
1441  * only available by calling this routine.
1442  *
1443  */
1444 int
1445 xfs_syncsub(
1446         xfs_mount_t     *mp,
1447         int             flags,
1448         int             xflags,
1449         int             *bypassed)
1450 {
1451         int             error = 0;
1452         int             last_error = 0;
1453         uint            log_flags = XFS_LOG_FORCE;
1454         xfs_buf_t       *bp;
1455         xfs_buf_log_item_t      *bip;
1456
1457         /*
1458          * Sync out the log.  This ensures that the log is periodically
1459          * flushed even if there is not enough activity to fill it up.
1460          */
1461         if (flags & SYNC_WAIT)
1462                 log_flags |= XFS_LOG_SYNC;
1463
1464         xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1465
1466         if (flags & (SYNC_ATTR|SYNC_DELWRI)) {
1467                 if (flags & SYNC_BDFLUSH)
1468                         xfs_finish_reclaim_all(mp, 1);
1469                 else
1470                         error = xfs_sync_inodes(mp, flags, xflags, bypassed);
1471         }
1472
1473         /*
1474          * Flushing out dirty data above probably generated more
1475          * log activity, so if this isn't vfs_sync() then flush
1476          * the log again.
1477          */
1478         if (flags & SYNC_DELWRI) {
1479                 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1480         }
1481
1482         if (flags & SYNC_FSDATA) {
1483                 /*
1484                  * If this is vfs_sync() then only sync the superblock
1485                  * if we can lock it without sleeping and it is not pinned.
1486                  */
1487                 if (flags & SYNC_BDFLUSH) {
1488                         bp = xfs_getsb(mp, XFS_BUF_TRYLOCK);
1489                         if (bp != NULL) {
1490                                 bip = XFS_BUF_FSPRIVATE(bp,xfs_buf_log_item_t*);
1491                                 if ((bip != NULL) &&
1492                                     xfs_buf_item_dirty(bip)) {
1493                                         if (!(XFS_BUF_ISPINNED(bp))) {
1494                                                 XFS_BUF_ASYNC(bp);
1495                                                 error = xfs_bwrite(mp, bp);
1496                                         } else {
1497                                                 xfs_buf_relse(bp);
1498                                         }
1499                                 } else {
1500                                         xfs_buf_relse(bp);
1501                                 }
1502                         }
1503                 } else {
1504                         bp = xfs_getsb(mp, 0);
1505                         /*
1506                          * If the buffer is pinned then push on the log so
1507                          * we won't get stuck waiting in the write for
1508                          * someone, maybe ourselves, to flush the log.
1509                          * Even though we just pushed the log above, we
1510                          * did not have the superblock buffer locked at
1511                          * that point so it can become pinned in between
1512                          * there and here.
1513                          */
1514                         if (XFS_BUF_ISPINNED(bp))
1515                                 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
1516                         if (flags & SYNC_WAIT)
1517                                 XFS_BUF_UNASYNC(bp);
1518                         else
1519                                 XFS_BUF_ASYNC(bp);
1520                         error = xfs_bwrite(mp, bp);
1521                 }
1522                 if (error) {
1523                         last_error = error;
1524                 }
1525         }
1526
1527         /*
1528          * If this is the periodic sync, then kick some entries out of
1529          * the reference cache.  This ensures that idle entries are
1530          * eventually kicked out of the cache.
1531          */
1532         if (flags & SYNC_REFCACHE) {
1533                 xfs_refcache_purge_some(mp);
1534         }
1535
1536         /*
1537          * Now check to see if the log needs a "dummy" transaction.
1538          */
1539
1540         if (!(flags & SYNC_REMOUNT) && xfs_log_need_covered(mp)) {
1541                 xfs_trans_t *tp;
1542                 xfs_inode_t *ip;
1543
1544                 /*
1545                  * Put a dummy transaction in the log to tell
1546                  * recovery that all others are OK.
1547                  */
1548                 tp = xfs_trans_alloc(mp, XFS_TRANS_DUMMY1);
1549                 if ((error = xfs_trans_reserve(tp, 0,
1550                                 XFS_ICHANGE_LOG_RES(mp),
1551                                 0, 0, 0)))  {
1552                         xfs_trans_cancel(tp, 0);
1553                         return error;
1554                 }
1555
1556                 ip = mp->m_rootip;
1557                 xfs_ilock(ip, XFS_ILOCK_EXCL);
1558
1559                 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1560                 xfs_trans_ihold(tp, ip);
1561                 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1562                 error = xfs_trans_commit(tp, 0, NULL);
1563                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1564                 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1565         }
1566
1567         /*
1568          * When shutting down, we need to insure that the AIL is pushed
1569          * to disk or the filesystem can appear corrupt from the PROM.
1570          */
1571         if ((flags & (SYNC_CLOSE|SYNC_WAIT)) == (SYNC_CLOSE|SYNC_WAIT)) {
1572                 XFS_bflush(mp->m_ddev_targp);
1573                 if (mp->m_rtdev_targp) {
1574                         XFS_bflush(mp->m_rtdev_targp);
1575                 }
1576         }
1577
1578         return XFS_ERROR(last_error);
1579 }
1580
1581 /*
1582  * xfs_vget - called by DMAPI to get vnode from file handle
1583  */
1584 STATIC int
1585 xfs_vget(
1586         bhv_desc_t      *bdp,
1587         vnode_t         **vpp,
1588         fid_t           *fidp)
1589 {
1590         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
1591         xfs_fid_t       *xfid = (struct xfs_fid *)fidp;
1592         xfs_inode_t     *ip;
1593         int             error;
1594         xfs_ino_t       ino;
1595         unsigned int    igen;
1596
1597         /*
1598          * Invalid.  Since handles can be created in user space and passed in
1599          * via gethandle(), this is not cause for a panic.
1600          */
1601         if (xfid->xfs_fid_len != sizeof(*xfid) - sizeof(xfid->xfs_fid_len))
1602                 return XFS_ERROR(EINVAL);
1603
1604         ino  = xfid->xfs_fid_ino;
1605         igen = xfid->xfs_fid_gen;
1606
1607         /*
1608          * NFS can sometimes send requests for ino 0.  Fail them gracefully.
1609          */
1610         if (ino == 0)
1611                 return XFS_ERROR(ESTALE);
1612
1613         error = xfs_iget(mp, NULL, ino, XFS_ILOCK_SHARED, &ip, 0);
1614         if (error) {
1615                 *vpp = NULL;
1616                 return error;
1617         }
1618
1619         if (ip == NULL) {
1620                 *vpp = NULL;
1621                 return XFS_ERROR(EIO);
1622         }
1623
1624         if (ip->i_d.di_mode == 0 || (igen && (ip->i_d.di_gen != igen))) {
1625                 xfs_iput_new(ip, XFS_ILOCK_SHARED);
1626                 *vpp = NULL;
1627                 return XFS_ERROR(ENOENT);
1628         }
1629
1630         *vpp = XFS_ITOV(ip);
1631         xfs_iunlock(ip, XFS_ILOCK_SHARED);
1632         return 0;
1633 }
1634
1635
1636 #define MNTOPT_LOGBUFS  "logbufs"       /* number of XFS log buffers */
1637 #define MNTOPT_LOGBSIZE "logbsize"      /* size of XFS log buffers */
1638 #define MNTOPT_LOGDEV   "logdev"        /* log device */
1639 #define MNTOPT_RTDEV    "rtdev"         /* realtime I/O device */
1640 #define MNTOPT_BIOSIZE  "biosize"       /* log2 of preferred buffered io size */
1641 #define MNTOPT_WSYNC    "wsync"         /* safe-mode nfs compatible mount */
1642 #define MNTOPT_INO64    "ino64"         /* force inodes into 64-bit range */
1643 #define MNTOPT_NOALIGN  "noalign"       /* turn off stripe alignment */
1644 #define MNTOPT_SWALLOC  "swalloc"       /* turn on stripe width allocation */
1645 #define MNTOPT_SUNIT    "sunit"         /* data volume stripe unit */
1646 #define MNTOPT_SWIDTH   "swidth"        /* data volume stripe width */
1647 #define MNTOPT_NOUUID   "nouuid"        /* ignore filesystem UUID */
1648 #define MNTOPT_MTPT     "mtpt"          /* filesystem mount point */
1649 #define MNTOPT_NORECOVERY   "norecovery"   /* don't run XFS recovery */
1650 #define MNTOPT_NOLOGFLUSH   "nologflush"   /* don't hard flush on log writes */
1651 #define MNTOPT_OSYNCISOSYNC "osyncisosync" /* o_sync is REALLY o_sync */
1652 #define MNTOPT_64BITINODE   "inode64"   /* inodes can be allocated anywhere */
1653 #define MNTOPT_IKEEP    "ikeep"         /* do not free empty inode clusters */
1654 #define MNTOPT_NOIKEEP  "noikeep"       /* free empty inode clusters */
1655
1656
1657 int
1658 xfs_parseargs(
1659         struct bhv_desc         *bhv,
1660         char                    *options,
1661         struct xfs_mount_args   *args,
1662         int                     update)
1663 {
1664         struct vfs              *vfsp = bhvtovfs(bhv);
1665         char                    *this_char, *value, *eov;
1666         int                     dsunit, dswidth, vol_dsunit, vol_dswidth;
1667         int                     iosize;
1668
1669 #if 0   /* XXX: off by default, until some remaining issues ironed out */
1670         args->flags |= XFSMNT_IDELETE; /* default to on */
1671 #endif
1672
1673         if (!options)
1674                 return 0;
1675
1676         iosize = dsunit = dswidth = vol_dsunit = vol_dswidth = 0;
1677
1678         while ((this_char = strsep(&options, ",")) != NULL) {
1679                 if (!*this_char)
1680                         continue;
1681                 if ((value = strchr(this_char, '=')) != NULL)
1682                         *value++ = 0;
1683
1684                 if (!strcmp(this_char, MNTOPT_LOGBUFS)) {
1685                         if (!value || !*value) {
1686                                 printk("XFS: %s option requires an argument\n",
1687                                         MNTOPT_LOGBUFS);
1688                                 return EINVAL;
1689                         }
1690                         args->logbufs = simple_strtoul(value, &eov, 10);
1691                 } else if (!strcmp(this_char, MNTOPT_LOGBSIZE)) {
1692                         int     last, in_kilobytes = 0;
1693
1694                         if (!value || !*value) {
1695                                 printk("XFS: %s option requires an argument\n",
1696                                         MNTOPT_LOGBSIZE);
1697                                 return EINVAL;
1698                         }
1699                         last = strlen(value) - 1;
1700                         if (value[last] == 'K' || value[last] == 'k') {
1701                                 in_kilobytes = 1;
1702                                 value[last] = '\0';
1703                         }
1704                         args->logbufsize = simple_strtoul(value, &eov, 10);
1705                         if (in_kilobytes)
1706                                 args->logbufsize <<= 10;
1707                 } else if (!strcmp(this_char, MNTOPT_LOGDEV)) {
1708                         if (!value || !*value) {
1709                                 printk("XFS: %s option requires an argument\n",
1710                                         MNTOPT_LOGDEV);
1711                                 return EINVAL;
1712                         }
1713                         strncpy(args->logname, value, MAXNAMELEN);
1714                 } else if (!strcmp(this_char, MNTOPT_MTPT)) {
1715                         if (!value || !*value) {
1716                                 printk("XFS: %s option requires an argument\n",
1717                                         MNTOPT_MTPT);
1718                                 return EINVAL;
1719                         }
1720                         strncpy(args->mtpt, value, MAXNAMELEN);
1721                 } else if (!strcmp(this_char, MNTOPT_RTDEV)) {
1722                         if (!value || !*value) {
1723                                 printk("XFS: %s option requires an argument\n",
1724                                         MNTOPT_RTDEV);
1725                                 return EINVAL;
1726                         }
1727                         strncpy(args->rtname, value, MAXNAMELEN);
1728                 } else if (!strcmp(this_char, MNTOPT_BIOSIZE)) {
1729                         if (!value || !*value) {
1730                                 printk("XFS: %s option requires an argument\n",
1731                                         MNTOPT_BIOSIZE);
1732                                 return EINVAL;
1733                         }
1734                         iosize = simple_strtoul(value, &eov, 10);
1735                         args->flags |= XFSMNT_IOSIZE;
1736                         args->iosizelog = (uint8_t) iosize;
1737                 } else if (!strcmp(this_char, MNTOPT_WSYNC)) {
1738                         args->flags |= XFSMNT_WSYNC;
1739                 } else if (!strcmp(this_char, MNTOPT_OSYNCISOSYNC)) {
1740                         args->flags |= XFSMNT_OSYNCISOSYNC;
1741                 } else if (!strcmp(this_char, MNTOPT_NORECOVERY)) {
1742                         args->flags |= XFSMNT_NORECOVERY;
1743                 } else if (!strcmp(this_char, MNTOPT_INO64)) {
1744                         args->flags |= XFSMNT_INO64;
1745 #if !XFS_BIG_INUMS
1746                         printk("XFS: %s option not allowed on this system\n",
1747                                 MNTOPT_INO64);
1748                         return EINVAL;
1749 #endif
1750                 } else if (!strcmp(this_char, MNTOPT_NOALIGN)) {
1751                         args->flags |= XFSMNT_NOALIGN;
1752                 } else if (!strcmp(this_char, MNTOPT_SWALLOC)) {
1753                         args->flags |= XFSMNT_SWALLOC;
1754                 } else if (!strcmp(this_char, MNTOPT_SUNIT)) {
1755                         if (!value || !*value) {
1756                                 printk("XFS: %s option requires an argument\n",
1757                                         MNTOPT_SUNIT);
1758                                 return EINVAL;
1759                         }
1760                         dsunit = simple_strtoul(value, &eov, 10);
1761                 } else if (!strcmp(this_char, MNTOPT_SWIDTH)) {
1762                         if (!value || !*value) {
1763                                 printk("XFS: %s option requires an argument\n",
1764                                         MNTOPT_SWIDTH);
1765                                 return EINVAL;
1766                         }
1767                         dswidth = simple_strtoul(value, &eov, 10);
1768                 } else if (!strcmp(this_char, MNTOPT_64BITINODE)) {
1769                         args->flags &= ~XFSMNT_32BITINODES;
1770 #if !XFS_BIG_INUMS
1771                         printk("XFS: %s option not allowed on this system\n",
1772                                 MNTOPT_64BITINODE);
1773                         return EINVAL;
1774 #endif
1775                 } else if (!strcmp(this_char, MNTOPT_NOUUID)) {
1776                         args->flags |= XFSMNT_NOUUID;
1777                 } else if (!strcmp(this_char, MNTOPT_NOLOGFLUSH)) {
1778                         args->flags |= XFSMNT_NOLOGFLUSH;
1779                 } else if (!strcmp(this_char, MNTOPT_IKEEP)) {
1780                         args->flags &= ~XFSMNT_IDELETE;
1781                 } else if (!strcmp(this_char, MNTOPT_NOIKEEP)) {
1782                         args->flags |= XFSMNT_IDELETE;
1783                 } else if (!strcmp(this_char, "osyncisdsync")) {
1784                         /* no-op, this is now the default */
1785 printk("XFS: osyncisdsync is now the default, option is deprecated.\n");
1786                 } else if (!strcmp(this_char, "irixsgid")) {
1787 printk("XFS: irixsgid is now a sysctl(2) variable, option is deprecated.\n");
1788                 } else {
1789                         printk("XFS: unknown mount option [%s].\n", this_char);
1790                         return EINVAL;
1791                 }
1792         }
1793
1794         if (args->flags & XFSMNT_NORECOVERY) {
1795                 if ((vfsp->vfs_flag & VFS_RDONLY) == 0) {
1796                         printk("XFS: no-recovery mounts must be read-only.\n");
1797                         return EINVAL;
1798                 }
1799         }
1800
1801         if ((args->flags & XFSMNT_NOALIGN) && (dsunit || dswidth)) {
1802                 printk(
1803         "XFS: sunit and swidth options incompatible with the noalign option\n");
1804                 return EINVAL;
1805         }
1806
1807         if ((dsunit && !dswidth) || (!dsunit && dswidth)) {
1808                 printk("XFS: sunit and swidth must be specified together\n");
1809                 return EINVAL;
1810         }
1811
1812         if (dsunit && (dswidth % dsunit != 0)) {
1813                 printk(
1814         "XFS: stripe width (%d) must be a multiple of the stripe unit (%d)\n",
1815                         dswidth, dsunit);
1816                 return EINVAL;
1817         }
1818
1819         if ((args->flags & XFSMNT_NOALIGN) != XFSMNT_NOALIGN) {
1820                 if (dsunit) {
1821                         args->sunit = dsunit;
1822                         args->flags |= XFSMNT_RETERR;
1823                 } else {
1824                         args->sunit = vol_dsunit;
1825                 }
1826                 dswidth ? (args->swidth = dswidth) :
1827                           (args->swidth = vol_dswidth);
1828         } else {
1829                 args->sunit = args->swidth = 0;
1830         }
1831
1832         return 0;
1833 }
1834
1835 int
1836 xfs_showargs(
1837         struct bhv_desc         *bhv,
1838         struct seq_file         *m)
1839 {
1840         static struct proc_xfs_info {
1841                 int     flag;
1842                 char    *str;
1843         } xfs_info[] = {
1844                 /* the few simple ones we can get from the mount struct */
1845                 { XFS_MOUNT_WSYNC,              "," MNTOPT_WSYNC },
1846                 { XFS_MOUNT_INO64,              "," MNTOPT_INO64 },
1847                 { XFS_MOUNT_NOALIGN,            "," MNTOPT_NOALIGN },
1848                 { XFS_MOUNT_SWALLOC,            "," MNTOPT_SWALLOC },
1849                 { XFS_MOUNT_NOUUID,             "," MNTOPT_NOUUID },
1850                 { XFS_MOUNT_NORECOVERY,         "," MNTOPT_NORECOVERY },
1851                 { XFS_MOUNT_OSYNCISOSYNC,       "," MNTOPT_OSYNCISOSYNC },
1852                 { XFS_MOUNT_NOLOGFLUSH,         "," MNTOPT_NOLOGFLUSH },
1853                 { XFS_MOUNT_IDELETE,            "," MNTOPT_NOIKEEP },
1854                 { 0, NULL }
1855         };
1856         struct proc_xfs_info    *xfs_infop;
1857         struct xfs_mount        *mp = XFS_BHVTOM(bhv);
1858
1859         for (xfs_infop = xfs_info; xfs_infop->flag; xfs_infop++) {
1860                 if (mp->m_flags & xfs_infop->flag)
1861                         seq_puts(m, xfs_infop->str);
1862         }
1863
1864         if (mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)
1865                 seq_printf(m, "," MNTOPT_BIOSIZE "=%d", mp->m_writeio_log);
1866
1867         if (mp->m_logbufs > 0)
1868                 seq_printf(m, "," MNTOPT_LOGBUFS "=%d", mp->m_logbufs);
1869
1870         if (mp->m_logbsize > 0)
1871                 seq_printf(m, "," MNTOPT_LOGBSIZE "=%d", mp->m_logbsize);
1872
1873         if (mp->m_ddev_targp != mp->m_logdev_targp)
1874                 seq_printf(m, "," MNTOPT_LOGDEV "=%s",
1875                                 XFS_BUFTARG_NAME(mp->m_logdev_targp));
1876
1877         if (mp->m_rtdev_targp && mp->m_ddev_targp != mp->m_rtdev_targp)
1878                 seq_printf(m, "," MNTOPT_RTDEV "=%s",
1879                                 XFS_BUFTARG_NAME(mp->m_rtdev_targp));
1880
1881         if (mp->m_dalign > 0)
1882                 seq_printf(m, "," MNTOPT_SUNIT "=%d",
1883                                 (int)XFS_FSB_TO_BB(mp, mp->m_dalign));
1884
1885         if (mp->m_swidth > 0)
1886                 seq_printf(m, "," MNTOPT_SWIDTH "=%d",
1887                                 (int)XFS_FSB_TO_BB(mp, mp->m_swidth));
1888
1889         if (!(mp->m_flags & XFS_MOUNT_32BITINOOPT))
1890                 seq_printf(m, "," MNTOPT_64BITINODE);
1891
1892         return 0;
1893 }
1894
1895 STATIC void
1896 xfs_freeze(
1897         bhv_desc_t      *bdp)
1898 {
1899         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
1900
1901         while (atomic_read(&mp->m_active_trans) > 0)
1902                 delay(100);
1903
1904         /* Push the superblock and write an unmount record */
1905         xfs_log_unmount_write(mp);
1906         xfs_unmountfs_writesb(mp);
1907 }
1908
1909
1910 vfsops_t xfs_vfsops = {
1911         BHV_IDENTITY_INIT(VFS_BHV_XFS,VFS_POSITION_XFS),
1912         .vfs_parseargs          = xfs_parseargs,
1913         .vfs_showargs           = xfs_showargs,
1914         .vfs_mount              = xfs_mount,
1915         .vfs_unmount            = xfs_unmount,
1916         .vfs_mntupdate          = xfs_mntupdate,
1917         .vfs_root               = xfs_root,
1918         .vfs_statvfs            = xfs_statvfs,
1919         .vfs_sync               = xfs_sync,
1920         .vfs_vget               = xfs_vget,
1921         .vfs_dmapiops           = (vfs_dmapiops_t)fs_nosys,
1922         .vfs_quotactl           = (vfs_quotactl_t)fs_nosys,
1923         .vfs_init_vnode         = xfs_initialize_vnode,
1924         .vfs_force_shutdown     = xfs_do_force_shutdown,
1925         .vfs_freeze             = xfs_freeze,
1926 };