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_ihsize = ap->ihashsize;
 256         mp->m_logbsize = ap->logbufsize;
 257         mp->m_fsname_len = strlen(ap->fsname) + 1;
 258         mp->m_fsname = kmem_alloc(mp->m_fsname_len, KM_SLEEP);
 259         strcpy(mp->m_fsname, ap->fsname);
 260
 261         /*
 262          * Pull in the 'wsync' and 'ino64' mount options before we do the real
 263          * work of mounting and recovery.  The arg pointer will
 264          * be NULL when we are being called from the root mount code.
 265          */
 266         if (ap->flags & XFSMNT_WSYNC)
 267                 mp->m_flags |= XFS_MOUNT_WSYNC;
 268 #if XFS_BIG_INUMS
 269         if (ap->flags & XFSMNT_INO64) {
 270                 mp->m_flags |= XFS_MOUNT_INO64;
 271                 mp->m_inoadd = XFS_INO64_OFFSET;
 272         }
 273 #endif
 274         if (ap->flags & XFSMNT_NOATIME)
 275                 mp->m_flags |= XFS_MOUNT_NOATIME;
 276
 277         if (ap->flags & XFSMNT_RETERR)
 278                 mp->m_flags |= XFS_MOUNT_RETERR;
 279
 280         if (ap->flags & XFSMNT_NOALIGN)
 281                 mp->m_flags |= XFS_MOUNT_NOALIGN;
 282
 283         if (ap->flags & XFSMNT_SWALLOC)
 284                 mp->m_flags |= XFS_MOUNT_SWALLOC;
 285
 286         if (ap->flags & XFSMNT_OSYNCISOSYNC)
 287                 mp->m_flags |= XFS_MOUNT_OSYNCISOSYNC;
 288
 289         if (ap->flags & XFSMNT_32BITINODES)
 290                 mp->m_flags |= (XFS_MOUNT_32BITINODES | XFS_MOUNT_32BITINOOPT);
 291
 292         if (ap->flags & XFSMNT_IOSIZE) {
 293                 if (ap->iosizelog > XFS_MAX_IO_LOG ||
 294                     ap->iosizelog < XFS_MIN_IO_LOG) {
 295                         cmn_err(CE_WARN,
 296                 "XFS: invalid log iosize: %d [not %d-%d]",
 297                                 ap->iosizelog, XFS_MIN_IO_LOG,
 298                                 XFS_MAX_IO_LOG);
 299                         return XFS_ERROR(EINVAL);
 300                 }
 301
 302                 mp->m_flags |= XFS_MOUNT_DFLT_IOSIZE;
 303                 mp->m_readio_log = mp->m_writeio_log = ap->iosizelog;
 304         }
 305         if (ap->flags & XFSMNT_IDELETE)
 306                 mp->m_flags |= XFS_MOUNT_IDELETE;
 307
 308         /*
 309          * no recovery flag requires a read-only mount
 310          */
 311         if (ap->flags & XFSMNT_NORECOVERY) {
 312                 if (!(vfs->vfs_flag & VFS_RDONLY)) {
 313                         cmn_err(CE_WARN,
 314         "XFS: tried to mount a FS read-write without recovery!");
 315                         return XFS_ERROR(EINVAL);
 316                 }
 317                 mp->m_flags |= XFS_MOUNT_NORECOVERY;
 318         }
 319
 320         if (ap->flags & XFSMNT_NOUUID)
 321                 mp->m_flags |= XFS_MOUNT_NOUUID;
 322         if (ap->flags & XFSMNT_NOLOGFLUSH)
 323                 mp->m_flags |= XFS_MOUNT_NOLOGFLUSH;
 324
 325         return 0;
 326 }
 327
 328 /*
 329  * This function fills in xfs_mount_t fields based on mount args.
 330  * Note: the superblock _has_ now been read in.
 331  */
 332 STATIC int
 333 xfs_finish_flags(
 334         struct vfs              *vfs,
 335         struct xfs_mount_args   *ap,
 336         struct xfs_mount        *mp)
 337 {
 338         int                     ronly = (vfs->vfs_flag & VFS_RDONLY);
 339
 340         /* Fail a mount where the logbuf is smaller then the log stripe */
 341         if (XFS_SB_VERSION_HASLOGV2(&mp->m_sb)) {
 342                 if ((ap->logbufsize == -1) &&
 343                     (mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE)) {
 344                         mp->m_logbsize = mp->m_sb.sb_logsunit;
 345                 } else if (ap->logbufsize < mp->m_sb.sb_logsunit) {
 346                         cmn_err(CE_WARN,
 347         "XFS: logbuf size must be greater than or equal to log stripe size");
 348                         return XFS_ERROR(EINVAL);
 349                 }
 350         } else {
 351                 /* Fail a mount if the logbuf is larger than 32K */
 352                 if (ap->logbufsize > XLOG_BIG_RECORD_BSIZE) {
 353                         cmn_err(CE_WARN,
 354         "XFS: logbuf size for version 1 logs must be 16K or 32K");
 355                         return XFS_ERROR(EINVAL);
 356                 }
 357         }
 358
 359         /*
 360          * prohibit r/w mounts of read-only filesystems
 361          */
 362         if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !ronly) {
 363                 cmn_err(CE_WARN,
 364         "XFS: cannot mount a read-only filesystem as read-write");
 365                 return XFS_ERROR(EROFS);
 366         }
 367
 368         /*
 369          * disallow mount attempts with (IRIX) project quota enabled
 370          */
 371         if (XFS_SB_VERSION_HASQUOTA(&mp->m_sb) &&
 372             (mp->m_sb.sb_qflags & XFS_PQUOTA_ACCT)) {
 373                 cmn_err(CE_WARN,
 374         "XFS: cannot mount a filesystem with IRIX project quota enabled");
 375                 return XFS_ERROR(ENOSYS);
 376         }
 377
 378         /*
 379          * check for shared mount.
 380          */
 381         if (ap->flags & XFSMNT_SHARED) {
 382                 if (!XFS_SB_VERSION_HASSHARED(&mp->m_sb))
 383                         return XFS_ERROR(EINVAL);
 384
 385                 /*
 386                  * For IRIX 6.5, shared mounts must have the shared
 387                  * version bit set, have the persistent readonly
 388                  * field set, must be version 0 and can only be mounted
 389                  * read-only.
 390                  */
 391                 if (!ronly || !(mp->m_sb.sb_flags & XFS_SBF_READONLY) ||
 392                      (mp->m_sb.sb_shared_vn != 0))
 393                         return XFS_ERROR(EINVAL);
 394
 395                 mp->m_flags |= XFS_MOUNT_SHARED;
 396
 397                 /*
 398                  * Shared XFS V0 can't deal with DMI.  Return EINVAL.
 399                  */
 400                 if (mp->m_sb.sb_shared_vn == 0 && (ap->flags & XFSMNT_DMAPI))
 401                         return XFS_ERROR(EINVAL);
 402         }
 403
 404         return 0;
 405 }
 406
 407 /*
 408  * xfs_mount
 409  *
 410  * The file system configurations are:
 411  *      (1) device (partition) with data and internal log
 412  *      (2) logical volume with data and log subvolumes.
 413  *      (3) logical volume with data, log, and realtime subvolumes.
 414  *
 415  * We only have to handle opening the log and realtime volumes here if
 416  * they are present.  The data subvolume has already been opened by
 417  * get_sb_bdev() and is stored in vfsp->vfs_super->s_bdev.
 418  */
 419 STATIC int
 420 xfs_mount(
 421         struct bhv_desc         *bhvp,
 422         struct xfs_mount_args   *args,
 423         cred_t                  *credp)
 424 {
 425         struct vfs              *vfsp = bhvtovfs(bhvp);
 426         struct bhv_desc         *p;
 427         struct xfs_mount        *mp = XFS_BHVTOM(bhvp);
 428         struct block_device     *ddev, *logdev, *rtdev;
 429         int                     flags = 0, error;
 430
 431         ddev = vfsp->vfs_super->s_bdev;
 432         logdev = rtdev = NULL;
 433
 434         /*
 435          * Setup xfs_mount function vectors from available behaviors
 436          */
 437         p = vfs_bhv_lookup(vfsp, VFS_POSITION_DM);
 438         mp->m_dm_ops = p ? *(xfs_dmops_t *) vfs_bhv_custom(p) : xfs_dmcore_stub;
 439         p = vfs_bhv_lookup(vfsp, VFS_POSITION_QM);
 440         mp->m_qm_ops = p ? *(xfs_qmops_t *) vfs_bhv_custom(p) : xfs_qmcore_stub;
 441         p = vfs_bhv_lookup(vfsp, VFS_POSITION_IO);
 442         mp->m_io_ops = p ? *(xfs_ioops_t *) vfs_bhv_custom(p) : xfs_iocore_xfs;
 443
 444         /*
 445          * Open real time and log devices - order is important.
 446          */
 447         if (args->logname[0]) {
 448                 error = xfs_blkdev_get(mp, args->logname, &logdev);
 449                 if (error)
 450                         return error;
 451         }
 452         if (args->rtname[0]) {
 453                 error = xfs_blkdev_get(mp, args->rtname, &rtdev);
 454                 if (error) {
 455                         xfs_blkdev_put(logdev);
 456                         return error;
 457                 }
 458
 459                 if (rtdev == ddev || rtdev == logdev) {
 460                         cmn_err(CE_WARN,
 461         "XFS: Cannot mount filesystem with identical rtdev and ddev/logdev.");
 462                         xfs_blkdev_put(logdev);
 463                         xfs_blkdev_put(rtdev);
 464                         return EINVAL;
 465                 }
 466         }
 467
 468         /*
 469          * Setup xfs_mount buffer target pointers
 470          */
 471         error = ENOMEM;
 472         mp->m_ddev_targp = xfs_alloc_buftarg(ddev, 0);
 473         if (!mp->m_ddev_targp) {
 474                 xfs_blkdev_put(logdev);
 475                 xfs_blkdev_put(rtdev);
 476                 return error;
 477         }
 478         if (rtdev) {
 479                 mp->m_rtdev_targp = xfs_alloc_buftarg(rtdev, 1);
 480                 if (!mp->m_rtdev_targp)
 481                         goto error0;
 482         }
 483         mp->m_logdev_targp = (logdev && logdev != ddev) ?
 484                                 xfs_alloc_buftarg(logdev, 1) : mp->m_ddev_targp;
 485         if (!mp->m_logdev_targp)
 486                 goto error0;
 487
 488         /*
 489          * Setup flags based on mount(2) options and then the superblock
 490          */
 491         error = xfs_start_flags(vfsp, args, mp);
 492         if (error)
 493                 goto error1;
 494         error = xfs_readsb(mp);
 495         if (error)
 496                 goto error1;
 497         error = xfs_finish_flags(vfsp, args, mp);
 498         if (error)
 499                 goto error2;
 500
 501         /*
 502          * Setup xfs_mount buffer target pointers based on superblock
 503          */
 504         error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_blocksize,
 505                                     mp->m_sb.sb_sectsize);
 506         if (!error && logdev && logdev != ddev) {
 507                 unsigned int    log_sector_size = BBSIZE;
 508
 509                 if (XFS_SB_VERSION_HASSECTOR(&mp->m_sb))
 510                         log_sector_size = mp->m_sb.sb_logsectsize;
 511                 error = xfs_setsize_buftarg(mp->m_logdev_targp,
 512                                             mp->m_sb.sb_blocksize,
 513                                             log_sector_size);
 514         }
 515         if (!error && rtdev)
 516                 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
 517                                             mp->m_sb.sb_blocksize,
 518                                             mp->m_sb.sb_sectsize);
 519         if (error)
 520                 goto error2;
 521
 522         error = XFS_IOINIT(vfsp, args, flags);
 523         if (!error)
 524                 return 0;
 525 error2:
 526         if (mp->m_sb_bp)
 527                 xfs_freesb(mp);
 528 error1:
 529         xfs_binval(mp->m_ddev_targp);
 530         if (logdev && logdev != ddev)
 531                 xfs_binval(mp->m_logdev_targp);
 532         if (rtdev)
 533                 xfs_binval(mp->m_rtdev_targp);
 534 error0:
 535         xfs_unmountfs_close(mp, credp);
 536         return error;
 537 }
 538
 539 STATIC int
 540 xfs_unmount(
 541         bhv_desc_t      *bdp,
 542         int             flags,
 543         cred_t          *credp)
 544 {
 545         struct vfs      *vfsp = bhvtovfs(bdp);
 546         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
 547         xfs_inode_t     *rip;
 548         vnode_t         *rvp;
 549         int             unmount_event_wanted = 0;
 550         int             unmount_event_flags = 0;
 551         int             xfs_unmountfs_needed = 0;
 552         int             error;
 553
 554         rip = mp->m_rootip;
 555         rvp = XFS_ITOV(rip);
 556
 557         if (vfsp->vfs_flag & VFS_DMI) {
 558                 error = XFS_SEND_PREUNMOUNT(mp, vfsp,
 559                                 rvp, DM_RIGHT_NULL, rvp, DM_RIGHT_NULL,
 560                                 NULL, NULL, 0, 0,
 561                                 (mp->m_dmevmask & (1<<DM_EVENT_PREUNMOUNT))?
 562                                         0:DM_FLAGS_UNWANTED);
 563                         if (error)
 564                                 return XFS_ERROR(error);
 565                 unmount_event_wanted = 1;
 566                 unmount_event_flags = (mp->m_dmevmask & (1<<DM_EVENT_UNMOUNT))?
 567                                         0 : DM_FLAGS_UNWANTED;
 568         }
 569
 570         /*
 571          * First blow any referenced inode from this file system
 572          * out of the reference cache, and delete the timer.
 573          */
 574         xfs_refcache_purge_mp(mp);
 575
 576         XFS_bflush(mp->m_ddev_targp);
 577         error = xfs_unmount_flush(mp, 0);
 578         if (error)
 579                 goto out;
 580
 581         ASSERT(vn_count(rvp) == 1);
 582
 583         /*
 584          * Drop the reference count
 585          */
 586         VN_RELE(rvp);
 587
 588         /*
 589          * If we're forcing a shutdown, typically because of a media error,
 590          * we want to make sure we invalidate dirty pages that belong to
 591          * referenced vnodes as well.
 592          */
 593         if (XFS_FORCED_SHUTDOWN(mp)) {
 594                 error = xfs_sync(&mp->m_bhv,
 595                          (SYNC_WAIT | SYNC_CLOSE), credp);
 596                 ASSERT(error != EFSCORRUPTED);
 597         }
 598         xfs_unmountfs_needed = 1;
 599
 600 out:
 601         /*      Send DMAPI event, if required.
 602          *      Then do xfs_unmountfs() if needed.
 603          *      Then return error (or zero).
 604          */
 605         if (unmount_event_wanted) {
 606                 /* Note: mp structure must still exist for
 607                  * XFS_SEND_UNMOUNT() call.
 608                  */
 609                 XFS_SEND_UNMOUNT(mp, vfsp, error == 0 ? rvp : NULL,
 610                         DM_RIGHT_NULL, 0, error, unmount_event_flags);
 611         }
 612         if (xfs_unmountfs_needed) {
 613                 /*
 614                  * Call common unmount function to flush to disk
 615                  * and free the super block buffer & mount structures.
 616                  */
 617                 xfs_unmountfs(mp, credp);
 618         }
 619
 620         return XFS_ERROR(error);
 621 }
 622
 623 #define REMOUNT_READONLY_FLAGS  (SYNC_REMOUNT|SYNC_ATTR|SYNC_WAIT)
 624
 625 STATIC int
 626 xfs_mntupdate(
 627         bhv_desc_t                      *bdp,
 628         int                             *flags,
 629         struct xfs_mount_args           *args)
 630 {
 631         struct vfs      *vfsp = bhvtovfs(bdp);
 632         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
 633         int             pincount, error;
 634         int             count = 0;
 635
 636         if (args->flags & XFSMNT_NOATIME)
 637                 mp->m_flags |= XFS_MOUNT_NOATIME;
 638         else
 639                 mp->m_flags &= ~XFS_MOUNT_NOATIME;
 640
 641         if (!(vfsp->vfs_flag & VFS_RDONLY)) {
 642                 VFS_SYNC(vfsp, SYNC_FSDATA|SYNC_BDFLUSH|SYNC_ATTR, NULL, error);
 643         }
 644
 645         if (*flags & MS_RDONLY) {
 646                 xfs_refcache_purge_mp(mp);
 647                 xfs_flush_buftarg(mp->m_ddev_targp, 0);
 648                 xfs_finish_reclaim_all(mp, 0);
 649
 650                 /* This loop must run at least twice.
 651                  * The first instance of the loop will flush
 652                  * most meta data but that will generate more
 653                  * meta data (typically directory updates).
 654                  * Which then must be flushed and logged before
 655                  * we can write the unmount record.
 656                  */
 657                 do {
 658                         VFS_SYNC(vfsp, REMOUNT_READONLY_FLAGS, NULL, error);
 659                         pincount = xfs_flush_buftarg(mp->m_ddev_targp, 1);
 660                         if (!pincount) {
 661                                 delay(50);
 662                                 count++;
 663                         }
 664                 } while (count < 2);
 665
 666                 /* Ok now write out an unmount record */
 667                 xfs_log_unmount_write(mp);
 668                 xfs_unmountfs_writesb(mp);
 669                 vfsp->vfs_flag |= VFS_RDONLY;
 670         } else {
 671                 vfsp->vfs_flag &= ~VFS_RDONLY;
 672         }
 673
 674         return 0;
 675 }
 676
 677 /*
 678  * xfs_unmount_flush implements a set of flush operation on special
 679  * inodes, which are needed as a separate set of operations so that
 680  * they can be called as part of relocation process.
 681  */
 682 int
 683 xfs_unmount_flush(
 684         xfs_mount_t     *mp,            /* Mount structure we are getting
 685                                            rid of. */
 686         int             relocation)     /* Called from vfs relocation. */
 687 {
 688         xfs_inode_t     *rip = mp->m_rootip;
 689         xfs_inode_t     *rbmip;
 690         xfs_inode_t     *rsumip = NULL;
 691         vnode_t         *rvp = XFS_ITOV(rip);
 692         int             error;
 693
 694         xfs_ilock(rip, XFS_ILOCK_EXCL);
 695         xfs_iflock(rip);
 696
 697         /*
 698          * Flush out the real time inodes.
 699          */
 700         if ((rbmip = mp->m_rbmip) != NULL) {
 701                 xfs_ilock(rbmip, XFS_ILOCK_EXCL);
 702                 xfs_iflock(rbmip);
 703                 error = xfs_iflush(rbmip, XFS_IFLUSH_SYNC);
 704                 xfs_iunlock(rbmip, XFS_ILOCK_EXCL);
 705
 706                 if (error == EFSCORRUPTED)
 707                         goto fscorrupt_out;
 708
 709                 ASSERT(vn_count(XFS_ITOV(rbmip)) == 1);
 710
 711                 rsumip = mp->m_rsumip;
 712                 xfs_ilock(rsumip, XFS_ILOCK_EXCL);
 713                 xfs_iflock(rsumip);
 714                 error = xfs_iflush(rsumip, XFS_IFLUSH_SYNC);
 715                 xfs_iunlock(rsumip, XFS_ILOCK_EXCL);
 716
 717                 if (error == EFSCORRUPTED)
 718                         goto fscorrupt_out;
 719
 720                 ASSERT(vn_count(XFS_ITOV(rsumip)) == 1);
 721         }
 722
 723         /*
 724          * Synchronously flush root inode to disk
 725          */
 726         error = xfs_iflush(rip, XFS_IFLUSH_SYNC);
 727         if (error == EFSCORRUPTED)
 728                 goto fscorrupt_out2;
 729
 730         if (vn_count(rvp) != 1 && !relocation) {
 731                 xfs_iunlock(rip, XFS_ILOCK_EXCL);
 732                 return XFS_ERROR(EBUSY);
 733         }
 734
 735         /*
 736          * Release dquot that rootinode, rbmino and rsumino might be holding,
 737          * flush and purge the quota inodes.
 738          */
 739         error = XFS_QM_UNMOUNT(mp);
 740         if (error == EFSCORRUPTED)
 741                 goto fscorrupt_out2;
 742
 743         if (rbmip) {
 744                 VN_RELE(XFS_ITOV(rbmip));
 745                 VN_RELE(XFS_ITOV(rsumip));
 746         }
 747
 748         xfs_iunlock(rip, XFS_ILOCK_EXCL);
 749         return 0;
 750
 751 fscorrupt_out:
 752         xfs_ifunlock(rip);
 753
 754 fscorrupt_out2:
 755         xfs_iunlock(rip, XFS_ILOCK_EXCL);
 756
 757         return XFS_ERROR(EFSCORRUPTED);
 758 }
 759
 760 /*
 761  * xfs_root extracts the root vnode from a vfs.
 762  *
 763  * vfsp -- the vfs struct for the desired file system
 764  * vpp  -- address of the caller's vnode pointer which should be
 765  *         set to the desired fs root vnode
 766  */
 767 STATIC int
 768 xfs_root(
 769         bhv_desc_t      *bdp,
 770         vnode_t         **vpp)
 771 {
 772         vnode_t         *vp;
 773
 774         vp = XFS_ITOV((XFS_BHVTOM(bdp))->m_rootip);
 775         VN_HOLD(vp);
 776         *vpp = vp;
 777         return 0;
 778 }
 779
 780 /*
 781  * xfs_statvfs
 782  *
 783  * Fill in the statvfs structure for the given file system.  We use
 784  * the superblock lock in the mount structure to ensure a consistent
 785  * snapshot of the counters returned.
 786  */
 787 STATIC int
 788 xfs_statvfs(
 789         bhv_desc_t      *bdp,
 790         xfs_statfs_t    *statp,
 791         vnode_t         *vp)
 792 {
 793         __uint64_t      fakeinos;
 794         xfs_extlen_t    lsize;
 795         xfs_mount_t     *mp;
 796         xfs_sb_t        *sbp;
 797         unsigned long   s;
 798         u64 id;
 799
 800         mp = XFS_BHVTOM(bdp);
 801         sbp = &(mp->m_sb);
 802
 803         statp->f_type = XFS_SB_MAGIC;
 804
 805         s = XFS_SB_LOCK(mp);
 806         statp->f_bsize = sbp->sb_blocksize;
 807         lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
 808         statp->f_blocks = sbp->sb_dblocks - lsize;
 809         statp->f_bfree = statp->f_bavail = sbp->sb_fdblocks;
 810         fakeinos = statp->f_bfree << sbp->sb_inopblog;
 811 #if XFS_BIG_INUMS
 812         fakeinos += mp->m_inoadd;
 813 #endif
 814         statp->f_files =
 815             MIN(sbp->sb_icount + fakeinos, (__uint64_t)XFS_MAXINUMBER);
 816         if (mp->m_maxicount)
 817 #if XFS_BIG_INUMS
 818                 if (!mp->m_inoadd)
 819 #endif
 820                         statp->f_files = min_t(typeof(statp->f_files),
 821                                                 statp->f_files,
 822                                                 mp->m_maxicount);
 823         statp->f_ffree = statp->f_files - (sbp->sb_icount - sbp->sb_ifree);
 824         XFS_SB_UNLOCK(mp, s);
 825
 826         id = huge_encode_dev(mp->m_dev);
 827         statp->f_fsid.val[0] = (u32)id;
 828         statp->f_fsid.val[1] = (u32)(id >> 32);
 829         statp->f_namelen = MAXNAMELEN - 1;
 830
 831         return 0;
 832 }
 833
 834
 835 /*
 836  * xfs_sync flushes any pending I/O to file system vfsp.
 837  *
 838  * This routine is called by vfs_sync() to make sure that things make it
 839  * out to disk eventually, on sync() system calls to flush out everything,
 840  * and when the file system is unmounted.  For the vfs_sync() case, all
 841  * we really need to do is sync out the log to make all of our meta-data
 842  * updates permanent (except for timestamps).  For calls from pflushd(),
 843  * dirty pages are kept moving by calling pdflush() on the inodes
 844  * containing them.  We also flush the inodes that we can lock without
 845  * sleeping and the superblock if we can lock it without sleeping from
 846  * vfs_sync() so that items at the tail of the log are always moving out.
 847  *
 848  * Flags:
 849  *      SYNC_BDFLUSH - We're being called from vfs_sync() so we don't want
 850  *                     to sleep if we can help it.  All we really need
 851  *                     to do is ensure that the log is synced at least
 852  *                     periodically.  We also push the inodes and
 853  *                     superblock if we can lock them without sleeping
 854  *                      and they are not pinned.
 855  *      SYNC_ATTR    - We need to flush the inodes.  If SYNC_BDFLUSH is not
 856  *                     set, then we really want to lock each inode and flush
 857  *                     it.
 858  *      SYNC_WAIT    - All the flushes that take place in this call should
 859  *                     be synchronous.
 860  *      SYNC_DELWRI  - This tells us to push dirty pages associated with
 861  *                     inodes.  SYNC_WAIT and SYNC_BDFLUSH are used to
 862  *                     determine if they should be flushed sync, async, or
 863  *                     delwri.
 864  *      SYNC_CLOSE   - This flag is passed when the system is being
 865  *                     unmounted.  We should sync and invalidate everthing.
 866  *      SYNC_FSDATA  - This indicates that the caller would like to make
 867  *                     sure the superblock is safe on disk.  We can ensure
 868  *                     this by simply makeing sure the log gets flushed
 869  *                     if SYNC_BDFLUSH is set, and by actually writing it
 870  *                     out otherwise.
 871  *
 872  */
 873 /*ARGSUSED*/
 874 STATIC int
 875 xfs_sync(
 876         bhv_desc_t      *bdp,
 877         int             flags,
 878         cred_t          *credp)
 879 {
 880         xfs_mount_t     *mp;
 881
 882         mp = XFS_BHVTOM(bdp);
 883         return (xfs_syncsub(mp, flags, 0, NULL));
 884 }
 885
 886 /*
 887  * xfs sync routine for internal use
 888  *
 889  * This routine supports all of the flags defined for the generic VFS_SYNC
 890  * interface as explained above under xfs_sync.  In the interests of not
 891  * changing interfaces within the 6.5 family, additional internallly-
 892  * required functions are specified within a separate xflags parameter,
 893  * only available by calling this routine.
 894  *
 895  */
 896 STATIC int
 897 xfs_sync_inodes(
 898         xfs_mount_t     *mp,
 899         int             flags,
 900         int             xflags,
 901         int             *bypassed)
 902 {
 903         xfs_inode_t     *ip = NULL;
 904         xfs_inode_t     *ip_next;
 905         xfs_buf_t       *bp;
 906         vnode_t         *vp = NULL;
 907         vmap_t          vmap;
 908         int             error;
 909         int             last_error;
 910         uint64_t        fflag;
 911         uint            lock_flags;
 912         uint            base_lock_flags;
 913         boolean_t       mount_locked;
 914         boolean_t       vnode_refed;
 915         int             preempt;
 916         xfs_dinode_t    *dip;
 917         xfs_iptr_t      *ipointer;
 918 #ifdef DEBUG
 919         boolean_t       ipointer_in = B_FALSE;
 920
 921 #define IPOINTER_SET    ipointer_in = B_TRUE
 922 #define IPOINTER_CLR    ipointer_in = B_FALSE
 923 #else
 924 #define IPOINTER_SET
 925 #define IPOINTER_CLR
 926 #endif
 927
 928
 929 /* Insert a marker record into the inode list after inode ip. The list
 930  * must be locked when this is called. After the call the list will no
 931  * longer be locked.
 932  */
 933 #define IPOINTER_INSERT(ip, mp) { \
 934                 ASSERT(ipointer_in == B_FALSE); \
 935                 ipointer->ip_mnext = ip->i_mnext; \
 936                 ipointer->ip_mprev = ip; \
 937                 ip->i_mnext = (xfs_inode_t *)ipointer; \
 938                 ipointer->ip_mnext->i_mprev = (xfs_inode_t *)ipointer; \
 939                 preempt = 0; \
 940                 XFS_MOUNT_IUNLOCK(mp); \
 941                 mount_locked = B_FALSE; \
 942                 IPOINTER_SET; \
 943         }
 944
 945 /* Remove the marker from the inode list. If the marker was the only item
 946  * in the list then there are no remaining inodes and we should zero out
 947  * the whole list. If we are the current head of the list then move the head
 948  * past us.
 949  */
 950 #define IPOINTER_REMOVE(ip, mp) { \
 951                 ASSERT(ipointer_in == B_TRUE); \
 952                 if (ipointer->ip_mnext != (xfs_inode_t *)ipointer) { \
 953                         ip = ipointer->ip_mnext; \
 954                         ip->i_mprev = ipointer->ip_mprev; \
 955                         ipointer->ip_mprev->i_mnext = ip; \
 956                         if (mp->m_inodes == (xfs_inode_t *)ipointer) { \
 957                                 mp->m_inodes = ip; \
 958                         } \
 959                 } else { \
 960                         ASSERT(mp->m_inodes == (xfs_inode_t *)ipointer); \
 961                         mp->m_inodes = NULL; \
 962                         ip = NULL; \
 963                 } \
 964                 IPOINTER_CLR; \
 965         }
 966
 967 #define XFS_PREEMPT_MASK        0x7f
 968
 969         if (bypassed)
 970                 *bypassed = 0;
 971         if (XFS_MTOVFS(mp)->vfs_flag & VFS_RDONLY)
 972                 return 0;
 973         error = 0;
 974         last_error = 0;
 975         preempt = 0;
 976
 977         /* Allocate a reference marker */
 978         ipointer = (xfs_iptr_t *)kmem_zalloc(sizeof(xfs_iptr_t), KM_SLEEP);
 979
 980         fflag = XFS_B_ASYNC;            /* default is don't wait */
 981         if (flags & SYNC_BDFLUSH)
 982                 fflag = XFS_B_DELWRI;
 983         if (flags & SYNC_WAIT)
 984                 fflag = 0;              /* synchronous overrides all */
 985
 986         base_lock_flags = XFS_ILOCK_SHARED;
 987         if (flags & (SYNC_DELWRI | SYNC_CLOSE)) {
 988                 /*
 989                  * We need the I/O lock if we're going to call any of
 990                  * the flush/inval routines.
 991                  */
 992                 base_lock_flags |= XFS_IOLOCK_SHARED;
 993         }
 994
 995         XFS_MOUNT_ILOCK(mp);
 996
 997         ip = mp->m_inodes;
 998
 999         mount_locked = B_TRUE;
1000         vnode_refed  = B_FALSE;
1001
1002         IPOINTER_CLR;
1003
1004         do {
1005                 ASSERT(ipointer_in == B_FALSE);
1006                 ASSERT(vnode_refed == B_FALSE);
1007
1008                 lock_flags = base_lock_flags;
1009
1010                 /*
1011                  * There were no inodes in the list, just break out
1012                  * of the loop.
1013                  */
1014                 if (ip == NULL) {
1015                         break;
1016                 }
1017
1018                 /*
1019                  * We found another sync thread marker - skip it
1020                  */
1021                 if (ip->i_mount == NULL) {
1022                         ip = ip->i_mnext;
1023                         continue;
1024                 }
1025
1026                 vp = XFS_ITOV_NULL(ip);
1027
1028                 /*
1029                  * If the vnode is gone then this is being torn down,
1030                  * call reclaim if it is flushed, else let regular flush
1031                  * code deal with it later in the loop.
1032                  */
1033
1034                 if (vp == NULL) {
1035                         /* Skip ones already in reclaim */
1036                         if (ip->i_flags & XFS_IRECLAIM) {
1037                                 ip = ip->i_mnext;
1038                                 continue;
1039                         }
1040                         if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0) {
1041                                 ip = ip->i_mnext;
1042                         } else if ((xfs_ipincount(ip) == 0) &&
1043                                     xfs_iflock_nowait(ip)) {
1044                                 IPOINTER_INSERT(ip, mp);
1045
1046                                 xfs_finish_reclaim(ip, 1,
1047                                                 XFS_IFLUSH_DELWRI_ELSE_ASYNC);
1048
1049                                 XFS_MOUNT_ILOCK(mp);
1050                                 mount_locked = B_TRUE;
1051                                 IPOINTER_REMOVE(ip, mp);
1052                         } else {
1053                                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1054                                 ip = ip->i_mnext;
1055                         }
1056                         continue;
1057                 }
1058
1059                 if (VN_BAD(vp)) {
1060                         ip = ip->i_mnext;
1061                         continue;
1062                 }
1063
1064                 if (XFS_FORCED_SHUTDOWN(mp) && !(flags & SYNC_CLOSE)) {
1065                         XFS_MOUNT_IUNLOCK(mp);
1066                         kmem_free(ipointer, sizeof(xfs_iptr_t));
1067                         return 0;
1068                 }
1069
1070                 /*
1071                  * If this is just vfs_sync() or pflushd() calling
1072                  * then we can skip inodes for which it looks like
1073                  * there is nothing to do.  Since we don't have the
1074                  * inode locked this is racey, but these are periodic
1075                  * calls so it doesn't matter.  For the others we want
1076                  * to know for sure, so we at least try to lock them.
1077                  */
1078                 if (flags & SYNC_BDFLUSH) {
1079                         if (((ip->i_itemp == NULL) ||
1080                              !(ip->i_itemp->ili_format.ilf_fields &
1081                                XFS_ILOG_ALL)) &&
1082                             (ip->i_update_core == 0)) {
1083                                 ip = ip->i_mnext;
1084                                 continue;
1085                         }
1086                 }
1087
1088                 /*
1089                  * Try to lock without sleeping.  We're out of order with
1090                  * the inode list lock here, so if we fail we need to drop
1091                  * the mount lock and try again.  If we're called from
1092                  * bdflush() here, then don't bother.
1093                  *
1094                  * The inode lock here actually coordinates with the
1095                  * almost spurious inode lock in xfs_ireclaim() to prevent
1096                  * the vnode we handle here without a reference from
1097                  * being freed while we reference it.  If we lock the inode
1098                  * while it's on the mount list here, then the spurious inode
1099                  * lock in xfs_ireclaim() after the inode is pulled from
1100                  * the mount list will sleep until we release it here.
1101                  * This keeps the vnode from being freed while we reference
1102                  * it.  It is also cheaper and simpler than actually doing
1103                  * a vn_get() for every inode we touch here.
1104                  */
1105                 if (xfs_ilock_nowait(ip, lock_flags) == 0) {
1106
1107                         if ((flags & SYNC_BDFLUSH) || (vp == NULL)) {
1108                                 ip = ip->i_mnext;
1109                                 continue;
1110                         }
1111
1112                         /*
1113                          * We need to unlock the inode list lock in order
1114                          * to lock the inode. Insert a marker record into
1115                          * the inode list to remember our position, dropping
1116                          * the lock is now done inside the IPOINTER_INSERT
1117                          * macro.
1118                          *
1119                          * We also use the inode list lock to protect us
1120                          * in taking a snapshot of the vnode version number
1121                          * for use in calling vn_get().
1122                          */
1123                         VMAP(vp, vmap);
1124                         IPOINTER_INSERT(ip, mp);
1125
1126                         vp = vn_get(vp, &vmap);
1127                         if (vp == NULL) {
1128                                 /*
1129                                  * The vnode was reclaimed once we let go
1130                                  * of the inode list lock.  Skip to the
1131                                  * next list entry. Remove the marker.
1132                                  */
1133
1134                                 XFS_MOUNT_ILOCK(mp);
1135
1136                                 mount_locked = B_TRUE;
1137                                 vnode_refed  = B_FALSE;
1138
1139                                 IPOINTER_REMOVE(ip, mp);
1140
1141                                 continue;
1142                         }
1143
1144                         xfs_ilock(ip, lock_flags);
1145
1146                         ASSERT(vp == XFS_ITOV(ip));
1147                         ASSERT(ip->i_mount == mp);
1148
1149                         vnode_refed = B_TRUE;
1150                 }
1151
1152                 /* From here on in the loop we may have a marker record
1153                  * in the inode list.
1154                  */
1155
1156                 if ((flags & SYNC_CLOSE)  && (vp != NULL)) {
1157                         /*
1158                          * This is the shutdown case.  We just need to
1159                          * flush and invalidate all the pages associated
1160                          * with the inode.  Drop the inode lock since
1161                          * we can't hold it across calls to the buffer
1162                          * cache.
1163                          *
1164                          * We don't set the VREMAPPING bit in the vnode
1165                          * here, because we don't hold the vnode lock
1166                          * exclusively.  It doesn't really matter, though,
1167                          * because we only come here when we're shutting
1168                          * down anyway.
1169                          */
1170                         xfs_iunlock(ip, XFS_ILOCK_SHARED);
1171
1172                         if (XFS_FORCED_SHUTDOWN(mp)) {
1173                                 VOP_TOSS_PAGES(vp, 0, -1, FI_REMAPF);
1174                         } else {
1175                                 VOP_FLUSHINVAL_PAGES(vp, 0, -1, FI_REMAPF);
1176                         }
1177
1178                         xfs_ilock(ip, XFS_ILOCK_SHARED);
1179
1180                 } else if ((flags & SYNC_DELWRI) && (vp != NULL)) {
1181                         if (VN_DIRTY(vp)) {
1182                                 /* We need to have dropped the lock here,
1183                                  * so insert a marker if we have not already
1184                                  * done so.
1185                                  */
1186                                 if (mount_locked) {
1187                                         IPOINTER_INSERT(ip, mp);
1188                                 }
1189
1190                                 /*
1191                                  * Drop the inode lock since we can't hold it
1192                                  * across calls to the buffer cache.
1193                                  */
1194                                 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1195                                 VOP_FLUSH_PAGES(vp, (xfs_off_t)0, -1,
1196                                                         fflag, FI_NONE, error);
1197                                 xfs_ilock(ip, XFS_ILOCK_SHARED);
1198                         }
1199
1200                 }
1201
1202                 if (flags & SYNC_BDFLUSH) {
1203                         if ((flags & SYNC_ATTR) &&
1204                             ((ip->i_update_core) ||
1205                              ((ip->i_itemp != NULL) &&
1206                               (ip->i_itemp->ili_format.ilf_fields != 0)))) {
1207
1208                                 /* Insert marker and drop lock if not already
1209                                  * done.
1210                                  */
1211                                 if (mount_locked) {
1212                                         IPOINTER_INSERT(ip, mp);
1213                                 }
1214
1215                                 /*
1216                                  * We don't want the periodic flushing of the
1217                                  * inodes by vfs_sync() to interfere with
1218                                  * I/O to the file, especially read I/O
1219                                  * where it is only the access time stamp
1220                                  * that is being flushed out.  To prevent
1221                                  * long periods where we have both inode
1222                                  * locks held shared here while reading the
1223                                  * inode's buffer in from disk, we drop the
1224                                  * inode lock while reading in the inode
1225                                  * buffer.  We have to release the buffer
1226                                  * and reacquire the inode lock so that they
1227                                  * are acquired in the proper order (inode
1228                                  * locks first).  The buffer will go at the
1229                                  * end of the lru chain, though, so we can
1230                                  * expect it to still be there when we go
1231                                  * for it again in xfs_iflush().
1232                                  */
1233                                 if ((xfs_ipincount(ip) == 0) &&
1234                                     xfs_iflock_nowait(ip)) {
1235
1236                                         xfs_ifunlock(ip);
1237                                         xfs_iunlock(ip, XFS_ILOCK_SHARED);
1238
1239                                         error = xfs_itobp(mp, NULL, ip,
1240                                                           &dip, &bp, 0);
1241                                         if (!error) {
1242                                                 xfs_buf_relse(bp);
1243                                         } else {
1244                                                 /* Bailing out, remove the
1245                                                  * marker and free it.
1246                                                  */
1247                                                 XFS_MOUNT_ILOCK(mp);
1248
1249                                                 IPOINTER_REMOVE(ip, mp);
1250
1251                                                 XFS_MOUNT_IUNLOCK(mp);
1252
1253                                                 ASSERT(!(lock_flags &
1254                                                         XFS_IOLOCK_SHARED));
1255
1256                                                 kmem_free(ipointer,
1257                                                         sizeof(xfs_iptr_t));
1258                                                 return (0);
1259                                         }
1260
1261                                         /*
1262                                          * Since we dropped the inode lock,
1263                                          * the inode may have been reclaimed.
1264                                          * Therefore, we reacquire the mount
1265                                          * lock and check to see if we were the
1266                                          * inode reclaimed. If this happened
1267                                          * then the ipointer marker will no
1268                                          * longer point back at us. In this
1269                                          * case, move ip along to the inode
1270                                          * after the marker, remove the marker
1271                                          * and continue.
1272                                          */
1273                                         XFS_MOUNT_ILOCK(mp);
1274                                         mount_locked = B_TRUE;
1275
1276                                         if (ip != ipointer->ip_mprev) {
1277                                                 IPOINTER_REMOVE(ip, mp);
1278
1279                                                 ASSERT(!vnode_refed);
1280                                                 ASSERT(!(lock_flags &
1281                                                         XFS_IOLOCK_SHARED));
1282                                                 continue;
1283                                         }
1284
1285                                         ASSERT(ip->i_mount == mp);
1286
1287                                         if (xfs_ilock_nowait(ip,
1288                                                     XFS_ILOCK_SHARED) == 0) {
1289                                                 ASSERT(ip->i_mount == mp);
1290                                                 /*
1291                                                  * We failed to reacquire
1292                                                  * the inode lock without
1293                                                  * sleeping, so just skip
1294                                                  * the inode for now.  We
1295                                                  * clear the ILOCK bit from
1296                                                  * the lock_flags so that we
1297                                                  * won't try to drop a lock
1298                                                  * we don't hold below.
1299                                                  */
1300                                                 lock_flags &= ~XFS_ILOCK_SHARED;
1301                                                 IPOINTER_REMOVE(ip_next, mp);
1302                                         } else if ((xfs_ipincount(ip) == 0) &&
1303                                                    xfs_iflock_nowait(ip)) {
1304                                                 ASSERT(ip->i_mount == mp);
1305                                                 /*
1306                                                  * Since this is vfs_sync()
1307                                                  * calling we only flush the
1308                                                  * inode out if we can lock
1309                                                  * it without sleeping and
1310                                                  * it is not pinned.  Drop
1311                                                  * the mount lock here so
1312                                                  * that we don't hold it for
1313                                                  * too long. We already have
1314                                                  * a marker in the list here.
1315                                                  */
1316                                                 XFS_MOUNT_IUNLOCK(mp);
1317                                                 mount_locked = B_FALSE;
1318                                                 error = xfs_iflush(ip,
1319                                                            XFS_IFLUSH_DELWRI);
1320                                         } else {
1321                                                 ASSERT(ip->i_mount == mp);
1322                                                 IPOINTER_REMOVE(ip_next, mp);
1323                                         }
1324                                 }
1325
1326                         }
1327
1328                 } else {
1329                         if ((flags & SYNC_ATTR) &&
1330                             ((ip->i_update_core) ||
1331                              ((ip->i_itemp != NULL) &&
1332                               (ip->i_itemp->ili_format.ilf_fields != 0)))) {
1333                                 if (mount_locked) {
1334                                         IPOINTER_INSERT(ip, mp);
1335                                 }
1336
1337                                 if (flags & SYNC_WAIT) {
1338                                         xfs_iflock(ip);
1339                                         error = xfs_iflush(ip,
1340                                                            XFS_IFLUSH_SYNC);
1341                                 } else {
1342                                         /*
1343                                          * If we can't acquire the flush
1344                                          * lock, then the inode is already
1345                                          * being flushed so don't bother
1346                                          * waiting.  If we can lock it then
1347                                          * do a delwri flush so we can
1348                                          * combine multiple inode flushes
1349                                          * in each disk write.
1350                                          */
1351                                         if (xfs_iflock_nowait(ip)) {
1352                                                 error = xfs_iflush(ip,
1353                                                            XFS_IFLUSH_DELWRI);
1354                                         }
1355                                         else if (bypassed)
1356                                                 (*bypassed)++;
1357                                 }
1358                         }
1359                 }
1360
1361                 if (lock_flags != 0) {
1362                         xfs_iunlock(ip, lock_flags);
1363                 }
1364
1365                 if (vnode_refed) {
1366                         /*
1367                          * If we had to take a reference on the vnode
1368                          * above, then wait until after we've unlocked
1369                          * the inode to release the reference.  This is
1370                          * because we can be already holding the inode
1371                          * lock when VN_RELE() calls xfs_inactive().
1372                          *
1373                          * Make sure to drop the mount lock before calling
1374                          * VN_RELE() so that we don't trip over ourselves if
1375                          * we have to go for the mount lock again in the
1376                          * inactive code.
1377                          */
1378                         if (mount_locked) {
1379                                 IPOINTER_INSERT(ip, mp);
1380                         }
1381
1382                         VN_RELE(vp);
1383
1384                         vnode_refed = B_FALSE;
1385                 }
1386
1387                 if (error) {
1388                         last_error = error;
1389                 }
1390
1391                 /*
1392                  * bail out if the filesystem is corrupted.
1393                  */
1394                 if (error == EFSCORRUPTED)  {
1395                         if (!mount_locked) {
1396                                 XFS_MOUNT_ILOCK(mp);
1397                                 IPOINTER_REMOVE(ip, mp);
1398                         }
1399                         XFS_MOUNT_IUNLOCK(mp);
1400                         ASSERT(ipointer_in == B_FALSE);
1401                         kmem_free(ipointer, sizeof(xfs_iptr_t));
1402                         return XFS_ERROR(error);
1403                 }
1404
1405                 /* Let other threads have a chance at the mount lock
1406                  * if we have looped many times without dropping the
1407                  * lock.
1408                  */
1409                 if ((++preempt & XFS_PREEMPT_MASK) == 0) {
1410                         if (mount_locked) {
1411                                 IPOINTER_INSERT(ip, mp);
1412                         }
1413                 }
1414
1415                 if (mount_locked == B_FALSE) {
1416                         XFS_MOUNT_ILOCK(mp);
1417                         mount_locked = B_TRUE;
1418                         IPOINTER_REMOVE(ip, mp);
1419                         continue;
1420                 }
1421
1422                 ASSERT(ipointer_in == B_FALSE);
1423                 ip = ip->i_mnext;
1424
1425         } while (ip != mp->m_inodes);
1426
1427         XFS_MOUNT_IUNLOCK(mp);
1428
1429         ASSERT(ipointer_in == B_FALSE);
1430
1431         kmem_free(ipointer, sizeof(xfs_iptr_t));
1432         return XFS_ERROR(last_error);
1433 }
1434
1435 /*
1436  * xfs sync routine for internal use
1437  *
1438  * This routine supports all of the flags defined for the generic VFS_SYNC
1439  * interface as explained above under xfs_sync.  In the interests of not
1440  * changing interfaces within the 6.5 family, additional internallly-
1441  * required functions are specified within a separate xflags parameter,
1442  * only available by calling this routine.
1443  *
1444  */
1445 int
1446 xfs_syncsub(
1447         xfs_mount_t     *mp,
1448         int             flags,
1449         int             xflags,
1450         int             *bypassed)
1451 {
1452         int             error = 0;
1453         int             last_error = 0;
1454         uint            log_flags = XFS_LOG_FORCE;
1455         xfs_buf_t       *bp;
1456         xfs_buf_log_item_t      *bip;
1457
1458         /*
1459          * Sync out the log.  This ensures that the log is periodically
1460          * flushed even if there is not enough activity to fill it up.
1461          */
1462         if (flags & SYNC_WAIT)
1463                 log_flags |= XFS_LOG_SYNC;
1464
1465         xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1466
1467         if (flags & (SYNC_ATTR|SYNC_DELWRI)) {
1468                 if (flags & SYNC_BDFLUSH)
1469                         xfs_finish_reclaim_all(mp, 1);
1470                 else
1471                         error = xfs_sync_inodes(mp, flags, xflags, bypassed);
1472         }
1473
1474         /*
1475          * Flushing out dirty data above probably generated more
1476          * log activity, so if this isn't vfs_sync() then flush
1477          * the log again.
1478          */
1479         if (flags & SYNC_DELWRI) {
1480                 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1481         }
1482
1483         if (flags & SYNC_FSDATA) {
1484                 /*
1485                  * If this is vfs_sync() then only sync the superblock
1486                  * if we can lock it without sleeping and it is not pinned.
1487                  */
1488                 if (flags & SYNC_BDFLUSH) {
1489                         bp = xfs_getsb(mp, XFS_BUF_TRYLOCK);
1490                         if (bp != NULL) {
1491                                 bip = XFS_BUF_FSPRIVATE(bp,xfs_buf_log_item_t*);
1492                                 if ((bip != NULL) &&
1493                                     xfs_buf_item_dirty(bip)) {
1494                                         if (!(XFS_BUF_ISPINNED(bp))) {
1495                                                 XFS_BUF_ASYNC(bp);
1496                                                 error = xfs_bwrite(mp, bp);
1497                                         } else {
1498                                                 xfs_buf_relse(bp);
1499                                         }
1500                                 } else {
1501                                         xfs_buf_relse(bp);
1502                                 }
1503                         }
1504                 } else {
1505                         bp = xfs_getsb(mp, 0);
1506                         /*
1507                          * If the buffer is pinned then push on the log so
1508                          * we won't get stuck waiting in the write for
1509                          * someone, maybe ourselves, to flush the log.
1510                          * Even though we just pushed the log above, we
1511                          * did not have the superblock buffer locked at
1512                          * that point so it can become pinned in between
1513                          * there and here.
1514                          */
1515                         if (XFS_BUF_ISPINNED(bp))
1516                                 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
1517                         if (flags & SYNC_WAIT)
1518                                 XFS_BUF_UNASYNC(bp);
1519                         else
1520                                 XFS_BUF_ASYNC(bp);
1521                         error = xfs_bwrite(mp, bp);
1522                 }
1523                 if (error) {
1524                         last_error = error;
1525                 }
1526         }
1527
1528         /*
1529          * If this is the periodic sync, then kick some entries out of
1530          * the reference cache.  This ensures that idle entries are
1531          * eventually kicked out of the cache.
1532          */
1533         if (flags & SYNC_REFCACHE) {
1534                 xfs_refcache_purge_some(mp);
1535         }
1536
1537         /*
1538          * Now check to see if the log needs a "dummy" transaction.
1539          */
1540
1541         if (!(flags & SYNC_REMOUNT) && xfs_log_need_covered(mp)) {
1542                 xfs_trans_t *tp;
1543                 xfs_inode_t *ip;
1544
1545                 /*
1546                  * Put a dummy transaction in the log to tell
1547                  * recovery that all others are OK.
1548                  */
1549                 tp = xfs_trans_alloc(mp, XFS_TRANS_DUMMY1);
1550                 if ((error = xfs_trans_reserve(tp, 0,
1551                                 XFS_ICHANGE_LOG_RES(mp),
1552                                 0, 0, 0)))  {
1553                         xfs_trans_cancel(tp, 0);
1554                         return error;
1555                 }
1556
1557                 ip = mp->m_rootip;
1558                 xfs_ilock(ip, XFS_ILOCK_EXCL);
1559
1560                 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1561                 xfs_trans_ihold(tp, ip);
1562                 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1563                 error = xfs_trans_commit(tp, 0, NULL);
1564                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1565                 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1566         }
1567
1568         /*
1569          * When shutting down, we need to insure that the AIL is pushed
1570          * to disk or the filesystem can appear corrupt from the PROM.
1571          */
1572         if ((flags & (SYNC_CLOSE|SYNC_WAIT)) == (SYNC_CLOSE|SYNC_WAIT)) {
1573                 XFS_bflush(mp->m_ddev_targp);
1574                 if (mp->m_rtdev_targp) {
1575                         XFS_bflush(mp->m_rtdev_targp);
1576                 }
1577         }
1578
1579         return XFS_ERROR(last_error);
1580 }
1581
1582 /*
1583  * xfs_vget - called by DMAPI and NFSD to get vnode from file handle
1584  */
1585 STATIC int
1586 xfs_vget(
1587         bhv_desc_t      *bdp,
1588         vnode_t         **vpp,
1589         fid_t           *fidp)
1590 {
1591         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
1592         xfs_fid_t       *xfid = (struct xfs_fid *)fidp;
1593         xfs_inode_t     *ip;
1594         int             error;
1595         xfs_ino_t       ino;
1596         unsigned int    igen;
1597
1598         /*
1599          * Invalid.  Since handles can be created in user space and passed in
1600          * via gethandle(), this is not cause for a panic.
1601          */
1602         if (xfid->xfs_fid_len != sizeof(*xfid) - sizeof(xfid->xfs_fid_len))
1603                 return XFS_ERROR(EINVAL);
1604
1605         ino  = xfid->xfs_fid_ino;
1606         igen = xfid->xfs_fid_gen;
1607
1608         /*
1609          * NFS can sometimes send requests for ino 0.  Fail them gracefully.
1610          */
1611         if (ino == 0)
1612                 return XFS_ERROR(ESTALE);
1613
1614         error = xfs_iget(mp, NULL, ino, 0, XFS_ILOCK_SHARED, &ip, 0);
1615         if (error) {
1616                 *vpp = NULL;
1617                 return error;
1618         }
1619
1620         if (ip == NULL) {
1621                 *vpp = NULL;
1622                 return XFS_ERROR(EIO);
1623         }
1624
1625         if (ip->i_d.di_mode == 0 || ip->i_d.di_gen != igen) {
1626                 xfs_iput_new(ip, XFS_ILOCK_SHARED);
1627                 *vpp = NULL;
1628                 return XFS_ERROR(ENOENT);
1629         }
1630
1631         *vpp = XFS_ITOV(ip);
1632         xfs_iunlock(ip, XFS_ILOCK_SHARED);
1633         return 0;
1634 }
1635
1636
1637 #define MNTOPT_LOGBUFS  "logbufs"       /* number of XFS log buffers */
1638 #define MNTOPT_LOGBSIZE "logbsize"      /* size of XFS log buffers */
1639 #define MNTOPT_LOGDEV   "logdev"        /* log device */
1640 #define MNTOPT_RTDEV    "rtdev"         /* realtime I/O device */
1641 #define MNTOPT_BIOSIZE  "biosize"       /* log2 of preferred buffered io size */
1642 #define MNTOPT_WSYNC    "wsync"         /* safe-mode nfs compatible mount */
1643 #define MNTOPT_INO64    "ino64"         /* force inodes into 64-bit range */
1644 #define MNTOPT_NOALIGN  "noalign"       /* turn off stripe alignment */
1645 #define MNTOPT_SWALLOC  "swalloc"       /* turn on stripe width allocation */
1646 #define MNTOPT_SUNIT    "sunit"         /* data volume stripe unit */
1647 #define MNTOPT_SWIDTH   "swidth"        /* data volume stripe width */
1648 #define MNTOPT_NOUUID   "nouuid"        /* ignore filesystem UUID */
1649 #define MNTOPT_MTPT     "mtpt"          /* filesystem mount point */
1650 #define MNTOPT_IHASHSIZE    "ihashsize"    /* size of inode hash table */
1651 #define MNTOPT_NORECOVERY   "norecovery"   /* don't run XFS recovery */
1652 #define MNTOPT_NOLOGFLUSH   "nologflush"   /* don't hard flush on log writes */
1653 #define MNTOPT_OSYNCISOSYNC "osyncisosync" /* o_sync is REALLY o_sync */
1654 #define MNTOPT_64BITINODE   "inode64"   /* inodes can be allocated anywhere */
1655 #define MNTOPT_IKEEP    "ikeep"         /* do not free empty inode clusters */
1656 #define MNTOPT_NOIKEEP  "noikeep"       /* free empty inode clusters */
1657
1658
1659 int
1660 xfs_parseargs(
1661         struct bhv_desc         *bhv,
1662         char                    *options,
1663         struct xfs_mount_args   *args,
1664         int                     update)
1665 {
1666         struct vfs              *vfsp = bhvtovfs(bhv);
1667         char                    *this_char, *value, *eov;
1668         int                     dsunit, dswidth, vol_dsunit, vol_dswidth;
1669         int                     iosize;
1670
1671 #if 0   /* XXX: off by default, until some remaining issues ironed out */
1672         args->flags |= XFSMNT_IDELETE; /* default to on */
1673 #endif
1674
1675         if (!options)
1676                 return 0;
1677
1678         iosize = dsunit = dswidth = vol_dsunit = vol_dswidth = 0;
1679
1680         while ((this_char = strsep(&options, ",")) != NULL) {
1681                 if (!*this_char)
1682                         continue;
1683                 if ((value = strchr(this_char, '=')) != NULL)
1684                         *value++ = 0;
1685
1686                 if (!strcmp(this_char, MNTOPT_LOGBUFS)) {
1687                         if (!value || !*value) {
1688                                 printk("XFS: %s option requires an argument\n",
1689                                         MNTOPT_LOGBUFS);
1690                                 return EINVAL;
1691                         }
1692                         args->logbufs = simple_strtoul(value, &eov, 10);
1693                 } else if (!strcmp(this_char, MNTOPT_LOGBSIZE)) {
1694                         int     last, in_kilobytes = 0;
1695
1696                         if (!value || !*value) {
1697                                 printk("XFS: %s option requires an argument\n",
1698                                         MNTOPT_LOGBSIZE);
1699                                 return EINVAL;
1700                         }
1701                         last = strlen(value) - 1;
1702                         if (value[last] == 'K' || value[last] == 'k') {
1703                                 in_kilobytes = 1;
1704                                 value[last] = '\0';
1705                         }
1706                         args->logbufsize = simple_strtoul(value, &eov, 10);
1707                         if (in_kilobytes)
1708                                 args->logbufsize <<= 10;
1709                 } else if (!strcmp(this_char, MNTOPT_LOGDEV)) {
1710                         if (!value || !*value) {
1711                                 printk("XFS: %s option requires an argument\n",
1712                                         MNTOPT_LOGDEV);
1713                                 return EINVAL;
1714                         }
1715                         strncpy(args->logname, value, MAXNAMELEN);
1716                 } else if (!strcmp(this_char, MNTOPT_MTPT)) {
1717                         if (!value || !*value) {
1718                                 printk("XFS: %s option requires an argument\n",
1719                                         MNTOPT_MTPT);
1720                                 return EINVAL;
1721                         }
1722                         strncpy(args->mtpt, value, MAXNAMELEN);
1723                 } else if (!strcmp(this_char, MNTOPT_RTDEV)) {
1724                         if (!value || !*value) {
1725                                 printk("XFS: %s option requires an argument\n",
1726                                         MNTOPT_RTDEV);
1727                                 return EINVAL;
1728                         }
1729                         strncpy(args->rtname, value, MAXNAMELEN);
1730                 } else if (!strcmp(this_char, MNTOPT_BIOSIZE)) {
1731                         if (!value || !*value) {
1732                                 printk("XFS: %s option requires an argument\n",
1733                                         MNTOPT_BIOSIZE);
1734                                 return EINVAL;
1735                         }
1736                         iosize = simple_strtoul(value, &eov, 10);
1737                         args->flags |= XFSMNT_IOSIZE;
1738                         args->iosizelog = (uint8_t) iosize;
1739                 } else if (!strcmp(this_char, MNTOPT_IHASHSIZE)) {
1740                         if (!value || !*value) {
1741                                 printk("XFS: %s option requires an argument\n",
1742                                         this_char);
1743                                 return EINVAL;
1744                         }
1745                         args->ihashsize = simple_strtoul(value, &eov, 10);
1746                 } else if (!strcmp(this_char, MNTOPT_WSYNC)) {
1747                         args->flags |= XFSMNT_WSYNC;
1748                 } else if (!strcmp(this_char, MNTOPT_OSYNCISOSYNC)) {
1749                         args->flags |= XFSMNT_OSYNCISOSYNC;
1750                 } else if (!strcmp(this_char, MNTOPT_NORECOVERY)) {
1751                         args->flags |= XFSMNT_NORECOVERY;
1752                 } else if (!strcmp(this_char, MNTOPT_INO64)) {
1753                         args->flags |= XFSMNT_INO64;
1754 #if !XFS_BIG_INUMS
1755                         printk("XFS: %s option not allowed on this system\n",
1756                                 MNTOPT_INO64);
1757                         return EINVAL;
1758 #endif
1759                 } else if (!strcmp(this_char, MNTOPT_NOALIGN)) {
1760                         args->flags |= XFSMNT_NOALIGN;
1761                 } else if (!strcmp(this_char, MNTOPT_SWALLOC)) {
1762                         args->flags |= XFSMNT_SWALLOC;
1763                 } else if (!strcmp(this_char, MNTOPT_SUNIT)) {
1764                         if (!value || !*value) {
1765                                 printk("XFS: %s option requires an argument\n",
1766                                         MNTOPT_SUNIT);
1767                                 return EINVAL;
1768                         }
1769                         dsunit = simple_strtoul(value, &eov, 10);
1770                 } else if (!strcmp(this_char, MNTOPT_SWIDTH)) {
1771                         if (!value || !*value) {
1772                                 printk("XFS: %s option requires an argument\n",
1773                                         MNTOPT_SWIDTH);
1774                                 return EINVAL;
1775                         }
1776                         dswidth = simple_strtoul(value, &eov, 10);
1777                 } else if (!strcmp(this_char, MNTOPT_64BITINODE)) {
1778                         args->flags &= ~XFSMNT_32BITINODES;
1779 #if !XFS_BIG_INUMS
1780                         printk("XFS: %s option not allowed on this system\n",
1781                                 MNTOPT_64BITINODE);
1782                         return EINVAL;
1783 #endif
1784                 } else if (!strcmp(this_char, MNTOPT_NOUUID)) {
1785                         args->flags |= XFSMNT_NOUUID;
1786                 } else if (!strcmp(this_char, MNTOPT_NOLOGFLUSH)) {
1787                         args->flags |= XFSMNT_NOLOGFLUSH;
1788                 } else if (!strcmp(this_char, MNTOPT_IKEEP)) {
1789                         args->flags &= ~XFSMNT_IDELETE;
1790                 } else if (!strcmp(this_char, MNTOPT_NOIKEEP)) {
1791                         args->flags |= XFSMNT_IDELETE;
1792                 } else if (!strcmp(this_char, "osyncisdsync")) {
1793                         /* no-op, this is now the default */
1794 printk("XFS: osyncisdsync is now the default, option is deprecated.\n");
1795                 } else if (!strcmp(this_char, "irixsgid")) {
1796 printk("XFS: irixsgid is now a sysctl(2) variable, option is deprecated.\n");
1797                 } else {
1798                         printk("XFS: unknown mount option [%s].\n", this_char);
1799                         return EINVAL;
1800                 }
1801         }
1802
1803         if (args->flags & XFSMNT_NORECOVERY) {
1804                 if ((vfsp->vfs_flag & VFS_RDONLY) == 0) {
1805                         printk("XFS: no-recovery mounts must be read-only.\n");
1806                         return EINVAL;
1807                 }
1808         }
1809
1810         if ((args->flags & XFSMNT_NOALIGN) && (dsunit || dswidth)) {
1811                 printk(
1812         "XFS: sunit and swidth options incompatible with the noalign option\n");
1813                 return EINVAL;
1814         }
1815
1816         if ((dsunit && !dswidth) || (!dsunit && dswidth)) {
1817                 printk("XFS: sunit and swidth must be specified together\n");
1818                 return EINVAL;
1819         }
1820
1821         if (dsunit && (dswidth % dsunit != 0)) {
1822                 printk(
1823         "XFS: stripe width (%d) must be a multiple of the stripe unit (%d)\n",
1824                         dswidth, dsunit);
1825                 return EINVAL;
1826         }
1827
1828         if ((args->flags & XFSMNT_NOALIGN) != XFSMNT_NOALIGN) {
1829                 if (dsunit) {
1830                         args->sunit = dsunit;
1831                         args->flags |= XFSMNT_RETERR;
1832                 } else {
1833                         args->sunit = vol_dsunit;
1834                 }
1835                 dswidth ? (args->swidth = dswidth) :
1836                           (args->swidth = vol_dswidth);
1837         } else {
1838                 args->sunit = args->swidth = 0;
1839         }
1840
1841         return 0;
1842 }
1843
1844 int
1845 xfs_showargs(
1846         struct bhv_desc         *bhv,
1847         struct seq_file         *m)
1848 {
1849         static struct proc_xfs_info {
1850                 int     flag;
1851                 char    *str;
1852         } xfs_info[] = {
1853                 /* the few simple ones we can get from the mount struct */
1854                 { XFS_MOUNT_WSYNC,              "," MNTOPT_WSYNC },
1855                 { XFS_MOUNT_INO64,              "," MNTOPT_INO64 },
1856                 { XFS_MOUNT_NOALIGN,            "," MNTOPT_NOALIGN },
1857                 { XFS_MOUNT_SWALLOC,            "," MNTOPT_SWALLOC },
1858                 { XFS_MOUNT_NOUUID,             "," MNTOPT_NOUUID },
1859                 { XFS_MOUNT_NORECOVERY,         "," MNTOPT_NORECOVERY },
1860                 { XFS_MOUNT_OSYNCISOSYNC,       "," MNTOPT_OSYNCISOSYNC },
1861                 { XFS_MOUNT_NOLOGFLUSH,         "," MNTOPT_NOLOGFLUSH },
1862                 { XFS_MOUNT_IDELETE,            "," MNTOPT_NOIKEEP },
1863                 { 0, NULL }
1864         };
1865         struct proc_xfs_info    *xfs_infop;
1866         struct xfs_mount        *mp = XFS_BHVTOM(bhv);
1867
1868         for (xfs_infop = xfs_info; xfs_infop->flag; xfs_infop++) {
1869                 if (mp->m_flags & xfs_infop->flag)
1870                         seq_puts(m, xfs_infop->str);
1871         }
1872
1873         if (mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)
1874                 seq_printf(m, "," MNTOPT_BIOSIZE "=%d", mp->m_writeio_log);
1875
1876         if (mp->m_logbufs > 0)
1877                 seq_printf(m, "," MNTOPT_LOGBUFS "=%d", mp->m_logbufs);
1878
1879         if (mp->m_logbsize > 0)
1880                 seq_printf(m, "," MNTOPT_LOGBSIZE "=%d", mp->m_logbsize);
1881
1882         if (mp->m_ddev_targp != mp->m_logdev_targp)
1883                 seq_printf(m, "," MNTOPT_LOGDEV "=%s",
1884                                 XFS_BUFTARG_NAME(mp->m_logdev_targp));
1885
1886         if (mp->m_rtdev_targp && mp->m_ddev_targp != mp->m_rtdev_targp)
1887                 seq_printf(m, "," MNTOPT_RTDEV "=%s",
1888                                 XFS_BUFTARG_NAME(mp->m_rtdev_targp));
1889
1890         if (mp->m_dalign > 0)
1891                 seq_printf(m, "," MNTOPT_SUNIT "=%d",
1892                                 (int)XFS_FSB_TO_BB(mp, mp->m_dalign));
1893
1894         if (mp->m_swidth > 0)
1895                 seq_printf(m, "," MNTOPT_SWIDTH "=%d",
1896                                 (int)XFS_FSB_TO_BB(mp, mp->m_swidth));
1897
1898         if (!(mp->m_flags & XFS_MOUNT_32BITINOOPT))
1899                 seq_printf(m, "," MNTOPT_64BITINODE);
1900
1901         return 0;
1902 }
1903
1904 STATIC void
1905 xfs_freeze(
1906         bhv_desc_t      *bdp)
1907 {
1908         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
1909
1910         while (atomic_read(&mp->m_active_trans) > 0)
1911                 delay(100);
1912
1913         /* Push the superblock and write an unmount record */
1914         xfs_log_unmount_write(mp);
1915         xfs_unmountfs_writesb(mp);
1916 }
1917
1918
1919 vfsops_t xfs_vfsops = {
1920         BHV_IDENTITY_INIT(VFS_BHV_XFS,VFS_POSITION_XFS),
1921         .vfs_parseargs          = xfs_parseargs,
1922         .vfs_showargs           = xfs_showargs,
1923         .vfs_mount              = xfs_mount,
1924         .vfs_unmount            = xfs_unmount,
1925         .vfs_mntupdate          = xfs_mntupdate,
1926         .vfs_root               = xfs_root,
1927         .vfs_statvfs            = xfs_statvfs,
1928         .vfs_sync               = xfs_sync,
1929         .vfs_vget               = xfs_vget,
1930         .vfs_dmapiops           = (vfs_dmapiops_t)fs_nosys,
1931         .vfs_quotactl           = (vfs_quotactl_t)fs_nosys,
1932         .vfs_init_vnode         = xfs_initialize_vnode,
1933         .vfs_force_shutdown     = xfs_do_force_shutdown,
1934         .vfs_freeze             = xfs_freeze,
1935 };