fs/xfs/xfs_vfsops.c

   1 /*
   2  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
   3  * All Rights Reserved.
   4  *
   5  * This program is free software; you can redistribute it and/or
   6  * modify it under the terms of the GNU General Public License as
   7  * published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it would be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write the Free Software Foundation,
  16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17  */
  18 #include "xfs.h"
  19 #include "xfs_fs.h"
  20 #include "xfs_types.h"
  21 #include "xfs_bit.h"
  22 #include "xfs_log.h"
  23 #include "xfs_inum.h"
  24 #include "xfs_trans.h"
  25 #include "xfs_sb.h"
  26 #include "xfs_ag.h"
  27 #include "xfs_dir.h"
  28 #include "xfs_dir2.h"
  29 #include "xfs_dmapi.h"
  30 #include "xfs_mount.h"
  31 #include "xfs_da_btree.h"
  32 #include "xfs_bmap_btree.h"
  33 #include "xfs_ialloc_btree.h"
  34 #include "xfs_alloc_btree.h"
  35 #include "xfs_dir_sf.h"
  36 #include "xfs_dir2_sf.h"
  37 #include "xfs_attr_sf.h"
  38 #include "xfs_dinode.h"
  39 #include "xfs_inode.h"
  40 #include "xfs_inode_item.h"
  41 #include "xfs_btree.h"
  42 #include "xfs_alloc.h"
  43 #include "xfs_ialloc.h"
  44 #include "xfs_quota.h"
  45 #include "xfs_error.h"
  46 #include "xfs_bmap.h"
  47 #include "xfs_rw.h"
  48 #include "xfs_refcache.h"
  49 #include "xfs_buf_item.h"
  50 #include "xfs_log_priv.h"
  51 #include "xfs_dir2_trace.h"
  52 #include "xfs_extfree_item.h"
  53 #include "xfs_acl.h"
  54 #include "xfs_attr.h"
  55 #include "xfs_clnt.h"
  56 #include "xfs_fsops.h"
  57
  58 STATIC int      xfs_sync(bhv_desc_t *, int, cred_t *);
  59
  60 int
  61 xfs_init(void)
  62 {
  63         extern kmem_zone_t      *xfs_bmap_free_item_zone;
  64         extern kmem_zone_t      *xfs_btree_cur_zone;
  65         extern kmem_zone_t      *xfs_trans_zone;
  66         extern kmem_zone_t      *xfs_buf_item_zone;
  67         extern kmem_zone_t      *xfs_dabuf_zone;
  68 #ifdef XFS_DABUF_DEBUG
  69         extern lock_t           xfs_dabuf_global_lock;
  70         spinlock_init(&xfs_dabuf_global_lock, "xfsda");
  71 #endif
  72
  73         /*
  74          * Initialize all of the zone allocators we use.
  75          */
  76         xfs_bmap_free_item_zone = kmem_zone_init(sizeof(xfs_bmap_free_item_t),
  77                                                  "xfs_bmap_free_item");
  78         xfs_btree_cur_zone = kmem_zone_init(sizeof(xfs_btree_cur_t),
  79                                             "xfs_btree_cur");
  80         xfs_trans_zone = kmem_zone_init(sizeof(xfs_trans_t), "xfs_trans");
  81         xfs_da_state_zone =
  82                 kmem_zone_init(sizeof(xfs_da_state_t), "xfs_da_state");
  83         xfs_dabuf_zone = kmem_zone_init(sizeof(xfs_dabuf_t), "xfs_dabuf");
  84         xfs_ifork_zone = kmem_zone_init(sizeof(xfs_ifork_t), "xfs_ifork");
  85         xfs_acl_zone_init(xfs_acl_zone, "xfs_acl");
  86
  87         /*
  88          * The size of the zone allocated buf log item is the maximum
  89          * size possible under XFS.  This wastes a little bit of memory,
  90          * but it is much faster.
  91          */
  92         xfs_buf_item_zone =
  93                 kmem_zone_init((sizeof(xfs_buf_log_item_t) +
  94                                 (((XFS_MAX_BLOCKSIZE / XFS_BLI_CHUNK) /
  95                                   NBWORD) * sizeof(int))),
  96                                "xfs_buf_item");
  97         xfs_efd_zone =
  98                 kmem_zone_init((sizeof(xfs_efd_log_item_t) +
  99                                ((XFS_EFD_MAX_FAST_EXTENTS - 1) *
 100                                  sizeof(xfs_extent_t))),
 101                                       "xfs_efd_item");
 102         xfs_efi_zone =
 103                 kmem_zone_init((sizeof(xfs_efi_log_item_t) +
 104                                ((XFS_EFI_MAX_FAST_EXTENTS - 1) *
 105                                  sizeof(xfs_extent_t))),
 106                                       "xfs_efi_item");
 107
 108         /*
 109          * These zones warrant special memory allocator hints
 110          */
 111         xfs_inode_zone =
 112                 kmem_zone_init_flags(sizeof(xfs_inode_t), "xfs_inode",
 113                                         KM_ZONE_HWALIGN | KM_ZONE_RECLAIM |
 114                                         KM_ZONE_SPREAD, NULL);
 115         xfs_ili_zone =
 116                 kmem_zone_init_flags(sizeof(xfs_inode_log_item_t), "xfs_ili",
 117                                         KM_ZONE_SPREAD, NULL);
 118         xfs_chashlist_zone =
 119                 kmem_zone_init_flags(sizeof(xfs_chashlist_t), "xfs_chashlist",
 120                                         KM_ZONE_SPREAD, NULL);
 121
 122         /*
 123          * Allocate global trace buffers.
 124          */
 125 #ifdef XFS_ALLOC_TRACE
 126         xfs_alloc_trace_buf = ktrace_alloc(XFS_ALLOC_TRACE_SIZE, KM_SLEEP);
 127 #endif
 128 #ifdef XFS_BMAP_TRACE
 129         xfs_bmap_trace_buf = ktrace_alloc(XFS_BMAP_TRACE_SIZE, KM_SLEEP);
 130 #endif
 131 #ifdef XFS_BMBT_TRACE
 132         xfs_bmbt_trace_buf = ktrace_alloc(XFS_BMBT_TRACE_SIZE, KM_SLEEP);
 133 #endif
 134 #ifdef XFS_DIR_TRACE
 135         xfs_dir_trace_buf = ktrace_alloc(XFS_DIR_TRACE_SIZE, KM_SLEEP);
 136 #endif
 137 #ifdef XFS_ATTR_TRACE
 138         xfs_attr_trace_buf = ktrace_alloc(XFS_ATTR_TRACE_SIZE, KM_SLEEP);
 139 #endif
 140 #ifdef XFS_DIR2_TRACE
 141         xfs_dir2_trace_buf = ktrace_alloc(XFS_DIR2_GTRACE_SIZE, KM_SLEEP);
 142 #endif
 143
 144         xfs_dir_startup();
 145
 146 #if (defined(DEBUG) || defined(INDUCE_IO_ERROR))
 147         xfs_error_test_init();
 148 #endif /* DEBUG || INDUCE_IO_ERROR */
 149
 150         xfs_init_procfs();
 151         xfs_sysctl_register();
 152         return 0;
 153 }
 154
 155 void
 156 xfs_cleanup(void)
 157 {
 158         extern kmem_zone_t      *xfs_bmap_free_item_zone;
 159         extern kmem_zone_t      *xfs_btree_cur_zone;
 160         extern kmem_zone_t      *xfs_inode_zone;
 161         extern kmem_zone_t      *xfs_trans_zone;
 162         extern kmem_zone_t      *xfs_da_state_zone;
 163         extern kmem_zone_t      *xfs_dabuf_zone;
 164         extern kmem_zone_t      *xfs_efd_zone;
 165         extern kmem_zone_t      *xfs_efi_zone;
 166         extern kmem_zone_t      *xfs_buf_item_zone;
 167         extern kmem_zone_t      *xfs_chashlist_zone;
 168
 169         xfs_cleanup_procfs();
 170         xfs_sysctl_unregister();
 171         xfs_refcache_destroy();
 172         xfs_acl_zone_destroy(xfs_acl_zone);
 173
 174 #ifdef XFS_DIR2_TRACE
 175         ktrace_free(xfs_dir2_trace_buf);
 176 #endif
 177 #ifdef XFS_ATTR_TRACE
 178         ktrace_free(xfs_attr_trace_buf);
 179 #endif
 180 #ifdef XFS_DIR_TRACE
 181         ktrace_free(xfs_dir_trace_buf);
 182 #endif
 183 #ifdef XFS_BMBT_TRACE
 184         ktrace_free(xfs_bmbt_trace_buf);
 185 #endif
 186 #ifdef XFS_BMAP_TRACE
 187         ktrace_free(xfs_bmap_trace_buf);
 188 #endif
 189 #ifdef XFS_ALLOC_TRACE
 190         ktrace_free(xfs_alloc_trace_buf);
 191 #endif
 192
 193         kmem_zone_destroy(xfs_bmap_free_item_zone);
 194         kmem_zone_destroy(xfs_btree_cur_zone);
 195         kmem_zone_destroy(xfs_inode_zone);
 196         kmem_zone_destroy(xfs_trans_zone);
 197         kmem_zone_destroy(xfs_da_state_zone);
 198         kmem_zone_destroy(xfs_dabuf_zone);
 199         kmem_zone_destroy(xfs_buf_item_zone);
 200         kmem_zone_destroy(xfs_efd_zone);
 201         kmem_zone_destroy(xfs_efi_zone);
 202         kmem_zone_destroy(xfs_ifork_zone);
 203         kmem_zone_destroy(xfs_ili_zone);
 204         kmem_zone_destroy(xfs_chashlist_zone);
 205 }
 206
 207 /*
 208  * xfs_start_flags
 209  *
 210  * This function fills in xfs_mount_t fields based on mount args.
 211  * Note: the superblock has _not_ yet been read in.
 212  */
 213 STATIC int
 214 xfs_start_flags(
 215         struct bhv_vfs          *vfs,
 216         struct xfs_mount_args   *ap,
 217         struct xfs_mount        *mp)
 218 {
 219         /* Values are in BBs */
 220         if ((ap->flags & XFSMNT_NOALIGN) != XFSMNT_NOALIGN) {
 221                 /*
 222                  * At this point the superblock has not been read
 223                  * in, therefore we do not know the block size.
 224                  * Before the mount call ends we will convert
 225                  * these to FSBs.
 226                  */
 227                 mp->m_dalign = ap->sunit;
 228                 mp->m_swidth = ap->swidth;
 229         }
 230
 231         if (ap->logbufs != -1 &&
 232             ap->logbufs != 0 &&
 233             (ap->logbufs < XLOG_MIN_ICLOGS ||
 234              ap->logbufs > XLOG_MAX_ICLOGS)) {
 235                 cmn_err(CE_WARN,
 236                         "XFS: invalid logbufs value: %d [not %d-%d]",
 237                         ap->logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
 238                 return XFS_ERROR(EINVAL);
 239         }
 240         mp->m_logbufs = ap->logbufs;
 241         if (ap->logbufsize != -1 &&
 242             ap->logbufsize !=  0 &&
 243             ap->logbufsize != 16 * 1024 &&
 244             ap->logbufsize != 32 * 1024 &&
 245             ap->logbufsize != 64 * 1024 &&
 246             ap->logbufsize != 128 * 1024 &&
 247             ap->logbufsize != 256 * 1024) {
 248                 cmn_err(CE_WARN,
 249         "XFS: invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
 250                         ap->logbufsize);
 251                 return XFS_ERROR(EINVAL);
 252         }
 253         mp->m_ihsize = ap->ihashsize;
 254         mp->m_logbsize = ap->logbufsize;
 255         mp->m_fsname_len = strlen(ap->fsname) + 1;
 256         mp->m_fsname = kmem_alloc(mp->m_fsname_len, KM_SLEEP);
 257         strcpy(mp->m_fsname, ap->fsname);
 258         if (ap->rtname[0]) {
 259                 mp->m_rtname = kmem_alloc(strlen(ap->rtname) + 1, KM_SLEEP);
 260                 strcpy(mp->m_rtname, ap->rtname);
 261         }
 262         if (ap->logname[0]) {
 263                 mp->m_logname = kmem_alloc(strlen(ap->logname) + 1, KM_SLEEP);
 264                 strcpy(mp->m_logname, ap->logname);
 265         }
 266
 267         if (ap->flags & XFSMNT_WSYNC)
 268                 mp->m_flags |= XFS_MOUNT_WSYNC;
 269 #if XFS_BIG_INUMS
 270         if (ap->flags & XFSMNT_INO64) {
 271                 mp->m_flags |= XFS_MOUNT_INO64;
 272                 mp->m_inoadd = XFS_INO64_OFFSET;
 273         }
 274 #endif
 275         if (ap->flags & XFSMNT_RETERR)
 276                 mp->m_flags |= XFS_MOUNT_RETERR;
 277         if (ap->flags & XFSMNT_NOALIGN)
 278                 mp->m_flags |= XFS_MOUNT_NOALIGN;
 279         if (ap->flags & XFSMNT_SWALLOC)
 280                 mp->m_flags |= XFS_MOUNT_SWALLOC;
 281         if (ap->flags & XFSMNT_OSYNCISOSYNC)
 282                 mp->m_flags |= XFS_MOUNT_OSYNCISOSYNC;
 283         if (ap->flags & XFSMNT_32BITINODES)
 284                 mp->m_flags |= XFS_MOUNT_32BITINODES;
 285
 286         if (ap->flags & XFSMNT_IOSIZE) {
 287                 if (ap->iosizelog > XFS_MAX_IO_LOG ||
 288                     ap->iosizelog < XFS_MIN_IO_LOG) {
 289                         cmn_err(CE_WARN,
 290                 "XFS: invalid log iosize: %d [not %d-%d]",
 291                                 ap->iosizelog, XFS_MIN_IO_LOG,
 292                                 XFS_MAX_IO_LOG);
 293                         return XFS_ERROR(EINVAL);
 294                 }
 295
 296                 mp->m_flags |= XFS_MOUNT_DFLT_IOSIZE;
 297                 mp->m_readio_log = mp->m_writeio_log = ap->iosizelog;
 298         }
 299
 300         if (ap->flags & XFSMNT_IHASHSIZE)
 301                 mp->m_flags |= XFS_MOUNT_IHASHSIZE;
 302         if (ap->flags & XFSMNT_IDELETE)
 303                 mp->m_flags |= XFS_MOUNT_IDELETE;
 304         if (ap->flags & XFSMNT_DIRSYNC)
 305                 mp->m_flags |= XFS_MOUNT_DIRSYNC;
 306         if (ap->flags & XFSMNT_ATTR2)
 307                 mp->m_flags |= XFS_MOUNT_ATTR2;
 308
 309         if (ap->flags2 & XFSMNT2_COMPAT_IOSIZE)
 310                 mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE;
 311
 312         /*
 313          * no recovery flag requires a read-only mount
 314          */
 315         if (ap->flags & XFSMNT_NORECOVERY) {
 316                 if (!(vfs->vfs_flag & VFS_RDONLY)) {
 317                         cmn_err(CE_WARN,
 318         "XFS: tried to mount a FS read-write without recovery!");
 319                         return XFS_ERROR(EINVAL);
 320                 }
 321                 mp->m_flags |= XFS_MOUNT_NORECOVERY;
 322         }
 323
 324         if (ap->flags & XFSMNT_NOUUID)
 325                 mp->m_flags |= XFS_MOUNT_NOUUID;
 326         if (ap->flags & XFSMNT_BARRIER)
 327                 mp->m_flags |= XFS_MOUNT_BARRIER;
 328         else
 329                 mp->m_flags &= ~XFS_MOUNT_BARRIER;
 330
 331         return 0;
 332 }
 333
 334 /*
 335  * This function fills in xfs_mount_t fields based on mount args.
 336  * Note: the superblock _has_ now been read in.
 337  */
 338 STATIC int
 339 xfs_finish_flags(
 340         struct bhv_vfs          *vfs,
 341         struct xfs_mount_args   *ap,
 342         struct xfs_mount        *mp)
 343 {
 344         int                     ronly = (vfs->vfs_flag & VFS_RDONLY);
 345
 346         /* Fail a mount where the logbuf is smaller then the log stripe */
 347         if (XFS_SB_VERSION_HASLOGV2(&mp->m_sb)) {
 348                 if ((ap->logbufsize <= 0) &&
 349                     (mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE)) {
 350                         mp->m_logbsize = mp->m_sb.sb_logsunit;
 351                 } else if (ap->logbufsize > 0 &&
 352                            ap->logbufsize < mp->m_sb.sb_logsunit) {
 353                         cmn_err(CE_WARN,
 354         "XFS: logbuf size must be greater than or equal to log stripe size");
 355                         return XFS_ERROR(EINVAL);
 356                 }
 357         } else {
 358                 /* Fail a mount if the logbuf is larger than 32K */
 359                 if (ap->logbufsize > XLOG_BIG_RECORD_BSIZE) {
 360                         cmn_err(CE_WARN,
 361         "XFS: logbuf size for version 1 logs must be 16K or 32K");
 362                         return XFS_ERROR(EINVAL);
 363                 }
 364         }
 365
 366         if (XFS_SB_VERSION_HASATTR2(&mp->m_sb)) {
 367                 mp->m_flags |= XFS_MOUNT_ATTR2;
 368         }
 369
 370         /*
 371          * prohibit r/w mounts of read-only filesystems
 372          */
 373         if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !ronly) {
 374                 cmn_err(CE_WARN,
 375         "XFS: cannot mount a read-only filesystem as read-write");
 376                 return XFS_ERROR(EROFS);
 377         }
 378
 379         /*
 380          * check for shared mount.
 381          */
 382         if (ap->flags & XFSMNT_SHARED) {
 383                 if (!XFS_SB_VERSION_HASSHARED(&mp->m_sb))
 384                         return XFS_ERROR(EINVAL);
 385
 386                 /*
 387                  * For IRIX 6.5, shared mounts must have the shared
 388                  * version bit set, have the persistent readonly
 389                  * field set, must be version 0 and can only be mounted
 390                  * read-only.
 391                  */
 392                 if (!ronly || !(mp->m_sb.sb_flags & XFS_SBF_READONLY) ||
 393                      (mp->m_sb.sb_shared_vn != 0))
 394                         return XFS_ERROR(EINVAL);
 395
 396                 mp->m_flags |= XFS_MOUNT_SHARED;
 397
 398                 /*
 399                  * Shared XFS V0 can't deal with DMI.  Return EINVAL.
 400                  */
 401                 if (mp->m_sb.sb_shared_vn == 0 && (ap->flags & XFSMNT_DMAPI))
 402                         return XFS_ERROR(EINVAL);
 403         }
 404
 405         return 0;
 406 }
 407
 408 /*
 409  * xfs_mount
 410  *
 411  * The file system configurations are:
 412  *      (1) device (partition) with data and internal log
 413  *      (2) logical volume with data and log subvolumes.
 414  *      (3) logical volume with data, log, and realtime subvolumes.
 415  *
 416  * We only have to handle opening the log and realtime volumes here if
 417  * they are present.  The data subvolume has already been opened by
 418  * get_sb_bdev() and is stored in vfsp->vfs_super->s_bdev.
 419  */
 420 STATIC int
 421 xfs_mount(
 422         struct bhv_desc         *bhvp,
 423         struct xfs_mount_args   *args,
 424         cred_t                  *credp)
 425 {
 426         struct bhv_vfs          *vfsp = bhvtovfs(bhvp);
 427         struct bhv_desc         *p;
 428         struct xfs_mount        *mp = XFS_BHVTOM(bhvp);
 429         struct block_device     *ddev, *logdev, *rtdev;
 430         int                     flags = 0, error;
 431
 432         ddev = vfsp->vfs_super->s_bdev;
 433         logdev = rtdev = NULL;
 434
 435         /*
 436          * Setup xfs_mount function vectors from available behaviors
 437          */
 438         p = vfs_bhv_lookup(vfsp, VFS_POSITION_DM);
 439         mp->m_dm_ops = p ? *(xfs_dmops_t *) vfs_bhv_custom(p) : xfs_dmcore_stub;
 440         p = vfs_bhv_lookup(vfsp, VFS_POSITION_QM);
 441         mp->m_qm_ops = p ? *(xfs_qmops_t *) vfs_bhv_custom(p) : xfs_qmcore_stub;
 442         p = vfs_bhv_lookup(vfsp, VFS_POSITION_IO);
 443         mp->m_io_ops = p ? *(xfs_ioops_t *) vfs_bhv_custom(p) : xfs_iocore_xfs;
 444
 445         if (args->flags & XFSMNT_QUIET)
 446                 flags |= XFS_MFSI_QUIET;
 447
 448         /*
 449          * Open real time and log devices - order is important.
 450          */
 451         if (args->logname[0]) {
 452                 error = xfs_blkdev_get(mp, args->logname, &logdev);
 453                 if (error)
 454                         return error;
 455         }
 456         if (args->rtname[0]) {
 457                 error = xfs_blkdev_get(mp, args->rtname, &rtdev);
 458                 if (error) {
 459                         xfs_blkdev_put(logdev);
 460                         return error;
 461                 }
 462
 463                 if (rtdev == ddev || rtdev == logdev) {
 464                         cmn_err(CE_WARN,
 465         "XFS: Cannot mount filesystem with identical rtdev and ddev/logdev.");
 466                         xfs_blkdev_put(logdev);
 467                         xfs_blkdev_put(rtdev);
 468                         return EINVAL;
 469                 }
 470         }
 471
 472         /*
 473          * Setup xfs_mount buffer target pointers
 474          */
 475         error = ENOMEM;
 476         mp->m_ddev_targp = xfs_alloc_buftarg(ddev, 0);
 477         if (!mp->m_ddev_targp) {
 478                 xfs_blkdev_put(logdev);
 479                 xfs_blkdev_put(rtdev);
 480                 return error;
 481         }
 482         if (rtdev) {
 483                 mp->m_rtdev_targp = xfs_alloc_buftarg(rtdev, 1);
 484                 if (!mp->m_rtdev_targp)
 485                         goto error0;
 486         }
 487         mp->m_logdev_targp = (logdev && logdev != ddev) ?
 488                                 xfs_alloc_buftarg(logdev, 1) : mp->m_ddev_targp;
 489         if (!mp->m_logdev_targp)
 490                 goto error0;
 491
 492         /*
 493          * Setup flags based on mount(2) options and then the superblock
 494          */
 495         error = xfs_start_flags(vfsp, args, mp);
 496         if (error)
 497                 goto error1;
 498         error = xfs_readsb(mp, flags);
 499         if (error)
 500                 goto error1;
 501         error = xfs_finish_flags(vfsp, args, mp);
 502         if (error)
 503                 goto error2;
 504
 505         /*
 506          * Setup xfs_mount buffer target pointers based on superblock
 507          */
 508         error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_blocksize,
 509                                     mp->m_sb.sb_sectsize);
 510         if (!error && logdev && logdev != ddev) {
 511                 unsigned int    log_sector_size = BBSIZE;
 512
 513                 if (XFS_SB_VERSION_HASSECTOR(&mp->m_sb))
 514                         log_sector_size = mp->m_sb.sb_logsectsize;
 515                 error = xfs_setsize_buftarg(mp->m_logdev_targp,
 516                                             mp->m_sb.sb_blocksize,
 517                                             log_sector_size);
 518         }
 519         if (!error && rtdev)
 520                 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
 521                                             mp->m_sb.sb_blocksize,
 522                                             mp->m_sb.sb_sectsize);
 523         if (error)
 524                 goto error2;
 525
 526         if ((mp->m_flags & XFS_MOUNT_BARRIER) && !(vfsp->vfs_flag & VFS_RDONLY))
 527                 xfs_mountfs_check_barriers(mp);
 528
 529         error = XFS_IOINIT(vfsp, args, flags);
 530         if (error)
 531                 goto error2;
 532
 533         return 0;
 534
 535 error2:
 536         if (mp->m_sb_bp)
 537                 xfs_freesb(mp);
 538 error1:
 539         xfs_binval(mp->m_ddev_targp);
 540         if (logdev && logdev != ddev)
 541                 xfs_binval(mp->m_logdev_targp);
 542         if (rtdev)
 543                 xfs_binval(mp->m_rtdev_targp);
 544 error0:
 545         xfs_unmountfs_close(mp, credp);
 546         return error;
 547 }
 548
 549 STATIC int
 550 xfs_unmount(
 551         bhv_desc_t      *bdp,
 552         int             flags,
 553         cred_t          *credp)
 554 {
 555         bhv_vfs_t       *vfsp = bhvtovfs(bdp);
 556         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
 557         xfs_inode_t     *rip;
 558         bhv_vnode_t     *rvp;
 559         int             unmount_event_wanted = 0;
 560         int             unmount_event_flags = 0;
 561         int             xfs_unmountfs_needed = 0;
 562         int             error;
 563
 564         rip = mp->m_rootip;
 565         rvp = XFS_ITOV(rip);
 566
 567         if (vfsp->vfs_flag & VFS_DMI) {
 568                 error = XFS_SEND_PREUNMOUNT(mp, vfsp,
 569                                 rvp, DM_RIGHT_NULL, rvp, DM_RIGHT_NULL,
 570                                 NULL, NULL, 0, 0,
 571                                 (mp->m_dmevmask & (1<<DM_EVENT_PREUNMOUNT))?
 572                                         0:DM_FLAGS_UNWANTED);
 573                         if (error)
 574                                 return XFS_ERROR(error);
 575                 unmount_event_wanted = 1;
 576                 unmount_event_flags = (mp->m_dmevmask & (1<<DM_EVENT_UNMOUNT))?
 577                                         0 : DM_FLAGS_UNWANTED;
 578         }
 579
 580         /*
 581          * First blow any referenced inode from this file system
 582          * out of the reference cache, and delete the timer.
 583          */
 584         xfs_refcache_purge_mp(mp);
 585
 586         XFS_bflush(mp->m_ddev_targp);
 587         error = xfs_unmount_flush(mp, 0);
 588         if (error)
 589                 goto out;
 590
 591         ASSERT(vn_count(rvp) == 1);
 592
 593         /*
 594          * Drop the reference count
 595          */
 596         VN_RELE(rvp);
 597
 598         /*
 599          * If we're forcing a shutdown, typically because of a media error,
 600          * we want to make sure we invalidate dirty pages that belong to
 601          * referenced vnodes as well.
 602          */
 603         if (XFS_FORCED_SHUTDOWN(mp)) {
 604                 error = xfs_sync(&mp->m_bhv,
 605                          (SYNC_WAIT | SYNC_CLOSE), credp);
 606                 ASSERT(error != EFSCORRUPTED);
 607         }
 608         xfs_unmountfs_needed = 1;
 609
 610 out:
 611         /*      Send DMAPI event, if required.
 612          *      Then do xfs_unmountfs() if needed.
 613          *      Then return error (or zero).
 614          */
 615         if (unmount_event_wanted) {
 616                 /* Note: mp structure must still exist for
 617                  * XFS_SEND_UNMOUNT() call.
 618                  */
 619                 XFS_SEND_UNMOUNT(mp, vfsp, error == 0 ? rvp : NULL,
 620                         DM_RIGHT_NULL, 0, error, unmount_event_flags);
 621         }
 622         if (xfs_unmountfs_needed) {
 623                 /*
 624                  * Call common unmount function to flush to disk
 625                  * and free the super block buffer & mount structures.
 626                  */
 627                 xfs_unmountfs(mp, credp);
 628         }
 629
 630         return XFS_ERROR(error);
 631 }
 632
 633 STATIC int
 634 xfs_quiesce_fs(
 635         xfs_mount_t             *mp)
 636 {
 637         int                     count = 0, pincount;
 638
 639         xfs_refcache_purge_mp(mp);
 640         xfs_flush_buftarg(mp->m_ddev_targp, 0);
 641         xfs_finish_reclaim_all(mp, 0);
 642
 643         /* This loop must run at least twice.
 644          * The first instance of the loop will flush
 645          * most meta data but that will generate more
 646          * meta data (typically directory updates).
 647          * Which then must be flushed and logged before
 648          * we can write the unmount record.
 649          */
 650         do {
 651                 xfs_syncsub(mp, SYNC_REMOUNT|SYNC_ATTR|SYNC_WAIT, 0, NULL);
 652                 pincount = xfs_flush_buftarg(mp->m_ddev_targp, 1);
 653                 if (!pincount) {
 654                         delay(50);
 655                         count++;
 656                 }
 657         } while (count < 2);
 658
 659         return 0;
 660 }
 661
 662 STATIC int
 663 xfs_mntupdate(
 664         bhv_desc_t                      *bdp,
 665         int                             *flags,
 666         struct xfs_mount_args           *args)
 667 {
 668         bhv_vfs_t       *vfsp = bhvtovfs(bdp);
 669         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
 670
 671         if (!(*flags & MS_RDONLY)) {                    /* rw/ro -> rw */
 672                 if (vfsp->vfs_flag & VFS_RDONLY)
 673                         vfsp->vfs_flag &= ~VFS_RDONLY;
 674                 if (args->flags & XFSMNT_BARRIER) {
 675                         mp->m_flags |= XFS_MOUNT_BARRIER;
 676                         xfs_mountfs_check_barriers(mp);
 677                 } else {
 678                         mp->m_flags &= ~XFS_MOUNT_BARRIER;
 679                 }
 680         } else if (!(vfsp->vfs_flag & VFS_RDONLY)) {    /* rw -> ro */
 681                 bhv_vfs_sync(vfsp, SYNC_FSDATA|SYNC_BDFLUSH|SYNC_ATTR, NULL);
 682                 xfs_quiesce_fs(mp);
 683                 xfs_log_unmount_write(mp);
 684                 xfs_unmountfs_writesb(mp);
 685                 vfsp->vfs_flag |= VFS_RDONLY;
 686         }
 687         return 0;
 688 }
 689
 690 /*
 691  * xfs_unmount_flush implements a set of flush operation on special
 692  * inodes, which are needed as a separate set of operations so that
 693  * they can be called as part of relocation process.
 694  */
 695 int
 696 xfs_unmount_flush(
 697         xfs_mount_t     *mp,            /* Mount structure we are getting
 698                                            rid of. */
 699         int             relocation)     /* Called from vfs relocation. */
 700 {
 701         xfs_inode_t     *rip = mp->m_rootip;
 702         xfs_inode_t     *rbmip;
 703         xfs_inode_t     *rsumip = NULL;
 704         bhv_vnode_t     *rvp = XFS_ITOV(rip);
 705         int             error;
 706
 707         xfs_ilock(rip, XFS_ILOCK_EXCL);
 708         xfs_iflock(rip);
 709
 710         /*
 711          * Flush out the real time inodes.
 712          */
 713         if ((rbmip = mp->m_rbmip) != NULL) {
 714                 xfs_ilock(rbmip, XFS_ILOCK_EXCL);
 715                 xfs_iflock(rbmip);
 716                 error = xfs_iflush(rbmip, XFS_IFLUSH_SYNC);
 717                 xfs_iunlock(rbmip, XFS_ILOCK_EXCL);
 718
 719                 if (error == EFSCORRUPTED)
 720                         goto fscorrupt_out;
 721
 722                 ASSERT(vn_count(XFS_ITOV(rbmip)) == 1);
 723
 724                 rsumip = mp->m_rsumip;
 725                 xfs_ilock(rsumip, XFS_ILOCK_EXCL);
 726                 xfs_iflock(rsumip);
 727                 error = xfs_iflush(rsumip, XFS_IFLUSH_SYNC);
 728                 xfs_iunlock(rsumip, XFS_ILOCK_EXCL);
 729
 730                 if (error == EFSCORRUPTED)
 731                         goto fscorrupt_out;
 732
 733                 ASSERT(vn_count(XFS_ITOV(rsumip)) == 1);
 734         }
 735
 736         /*
 737          * Synchronously flush root inode to disk
 738          */
 739         error = xfs_iflush(rip, XFS_IFLUSH_SYNC);
 740         if (error == EFSCORRUPTED)
 741                 goto fscorrupt_out2;
 742
 743         if (vn_count(rvp) != 1 && !relocation) {
 744                 xfs_iunlock(rip, XFS_ILOCK_EXCL);
 745                 return XFS_ERROR(EBUSY);
 746         }
 747
 748         /*
 749          * Release dquot that rootinode, rbmino and rsumino might be holding,
 750          * flush and purge the quota inodes.
 751          */
 752         error = XFS_QM_UNMOUNT(mp);
 753         if (error == EFSCORRUPTED)
 754                 goto fscorrupt_out2;
 755
 756         if (rbmip) {
 757                 VN_RELE(XFS_ITOV(rbmip));
 758                 VN_RELE(XFS_ITOV(rsumip));
 759         }
 760
 761         xfs_iunlock(rip, XFS_ILOCK_EXCL);
 762         return 0;
 763
 764 fscorrupt_out:
 765         xfs_ifunlock(rip);
 766
 767 fscorrupt_out2:
 768         xfs_iunlock(rip, XFS_ILOCK_EXCL);
 769
 770         return XFS_ERROR(EFSCORRUPTED);
 771 }
 772
 773 /*
 774  * xfs_root extracts the root vnode from a vfs.
 775  *
 776  * vfsp -- the vfs struct for the desired file system
 777  * vpp  -- address of the caller's vnode pointer which should be
 778  *         set to the desired fs root vnode
 779  */
 780 STATIC int
 781 xfs_root(
 782         bhv_desc_t      *bdp,
 783         bhv_vnode_t     **vpp)
 784 {
 785         bhv_vnode_t     *vp;
 786
 787         vp = XFS_ITOV((XFS_BHVTOM(bdp))->m_rootip);
 788         VN_HOLD(vp);
 789         *vpp = vp;
 790         return 0;
 791 }
 792
 793 /*
 794  * xfs_statvfs
 795  *
 796  * Fill in the statvfs structure for the given file system.  We use
 797  * the superblock lock in the mount structure to ensure a consistent
 798  * snapshot of the counters returned.
 799  */
 800 STATIC int
 801 xfs_statvfs(
 802         bhv_desc_t      *bdp,
 803         bhv_statvfs_t   *statp,
 804         bhv_vnode_t     *vp)
 805 {
 806         __uint64_t      fakeinos;
 807         xfs_extlen_t    lsize;
 808         xfs_mount_t     *mp;
 809         xfs_sb_t        *sbp;
 810         unsigned long   s;
 811
 812         mp = XFS_BHVTOM(bdp);
 813         sbp = &(mp->m_sb);
 814
 815         statp->f_type = XFS_SB_MAGIC;
 816
 817         xfs_icsb_sync_counters_lazy(mp);
 818         s = XFS_SB_LOCK(mp);
 819         statp->f_bsize = sbp->sb_blocksize;
 820         lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
 821         statp->f_blocks = sbp->sb_dblocks - lsize;
 822         statp->f_bfree = statp->f_bavail = sbp->sb_fdblocks;
 823         fakeinos = statp->f_bfree << sbp->sb_inopblog;
 824 #if XFS_BIG_INUMS
 825         fakeinos += mp->m_inoadd;
 826 #endif
 827         statp->f_files =
 828             MIN(sbp->sb_icount + fakeinos, (__uint64_t)XFS_MAXINUMBER);
 829         if (mp->m_maxicount)
 830 #if XFS_BIG_INUMS
 831                 if (!mp->m_inoadd)
 832 #endif
 833                         statp->f_files = min_t(typeof(statp->f_files),
 834                                                 statp->f_files,
 835                                                 mp->m_maxicount);
 836         statp->f_ffree = statp->f_files - (sbp->sb_icount - sbp->sb_ifree);
 837         XFS_SB_UNLOCK(mp, s);
 838
 839         xfs_statvfs_fsid(statp, mp);
 840         statp->f_namelen = MAXNAMELEN - 1;
 841
 842         return 0;
 843 }
 844
 845
 846 /*
 847  * xfs_sync flushes any pending I/O to file system vfsp.
 848  *
 849  * This routine is called by vfs_sync() to make sure that things make it
 850  * out to disk eventually, on sync() system calls to flush out everything,
 851  * and when the file system is unmounted.  For the vfs_sync() case, all
 852  * we really need to do is sync out the log to make all of our meta-data
 853  * updates permanent (except for timestamps).  For calls from pflushd(),
 854  * dirty pages are kept moving by calling pdflush() on the inodes
 855  * containing them.  We also flush the inodes that we can lock without
 856  * sleeping and the superblock if we can lock it without sleeping from
 857  * vfs_sync() so that items at the tail of the log are always moving out.
 858  *
 859  * Flags:
 860  *      SYNC_BDFLUSH - We're being called from vfs_sync() so we don't want
 861  *                     to sleep if we can help it.  All we really need
 862  *                     to do is ensure that the log is synced at least
 863  *                     periodically.  We also push the inodes and
 864  *                     superblock if we can lock them without sleeping
 865  *                      and they are not pinned.
 866  *      SYNC_ATTR    - We need to flush the inodes.  If SYNC_BDFLUSH is not
 867  *                     set, then we really want to lock each inode and flush
 868  *                     it.
 869  *      SYNC_WAIT    - All the flushes that take place in this call should
 870  *                     be synchronous.
 871  *      SYNC_DELWRI  - This tells us to push dirty pages associated with
 872  *                     inodes.  SYNC_WAIT and SYNC_BDFLUSH are used to
 873  *                     determine if they should be flushed sync, async, or
 874  *                     delwri.
 875  *      SYNC_CLOSE   - This flag is passed when the system is being
 876  *                     unmounted.  We should sync and invalidate everything.
 877  *      SYNC_FSDATA  - This indicates that the caller would like to make
 878  *                     sure the superblock is safe on disk.  We can ensure
 879  *                     this by simply making sure the log gets flushed
 880  *                     if SYNC_BDFLUSH is set, and by actually writing it
 881  *                     out otherwise.
 882  *
 883  */
 884 /*ARGSUSED*/
 885 STATIC int
 886 xfs_sync(
 887         bhv_desc_t      *bdp,
 888         int             flags,
 889         cred_t          *credp)
 890 {
 891         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
 892
 893         if (unlikely(flags == SYNC_QUIESCE))
 894                 return xfs_quiesce_fs(mp);
 895         else
 896                 return xfs_syncsub(mp, flags, 0, NULL);
 897 }
 898
 899 /*
 900  * xfs sync routine for internal use
 901  *
 902  * This routine supports all of the flags defined for the generic vfs_sync
 903  * interface as explained above under xfs_sync.  In the interests of not
 904  * changing interfaces within the 6.5 family, additional internally-
 905  * required functions are specified within a separate xflags parameter,
 906  * only available by calling this routine.
 907  *
 908  */
 909 int
 910 xfs_sync_inodes(
 911         xfs_mount_t     *mp,
 912         int             flags,
 913         int             xflags,
 914         int             *bypassed)
 915 {
 916         xfs_inode_t     *ip = NULL;
 917         xfs_inode_t     *ip_next;
 918         xfs_buf_t       *bp;
 919         bhv_vnode_t     *vp = NULL;
 920         int             error;
 921         int             last_error;
 922         uint64_t        fflag;
 923         uint            lock_flags;
 924         uint            base_lock_flags;
 925         boolean_t       mount_locked;
 926         boolean_t       vnode_refed;
 927         int             preempt;
 928         xfs_dinode_t    *dip;
 929         xfs_iptr_t      *ipointer;
 930 #ifdef DEBUG
 931         boolean_t       ipointer_in = B_FALSE;
 932
 933 #define IPOINTER_SET    ipointer_in = B_TRUE
 934 #define IPOINTER_CLR    ipointer_in = B_FALSE
 935 #else
 936 #define IPOINTER_SET
 937 #define IPOINTER_CLR
 938 #endif
 939
 940
 941 /* Insert a marker record into the inode list after inode ip. The list
 942  * must be locked when this is called. After the call the list will no
 943  * longer be locked.
 944  */
 945 #define IPOINTER_INSERT(ip, mp) { \
 946                 ASSERT(ipointer_in == B_FALSE); \
 947                 ipointer->ip_mnext = ip->i_mnext; \
 948                 ipointer->ip_mprev = ip; \
 949                 ip->i_mnext = (xfs_inode_t *)ipointer; \
 950                 ipointer->ip_mnext->i_mprev = (xfs_inode_t *)ipointer; \
 951                 preempt = 0; \
 952                 XFS_MOUNT_IUNLOCK(mp); \
 953                 mount_locked = B_FALSE; \
 954                 IPOINTER_SET; \
 955         }
 956
 957 /* Remove the marker from the inode list. If the marker was the only item
 958  * in the list then there are no remaining inodes and we should zero out
 959  * the whole list. If we are the current head of the list then move the head
 960  * past us.
 961  */
 962 #define IPOINTER_REMOVE(ip, mp) { \
 963                 ASSERT(ipointer_in == B_TRUE); \
 964                 if (ipointer->ip_mnext != (xfs_inode_t *)ipointer) { \
 965                         ip = ipointer->ip_mnext; \
 966                         ip->i_mprev = ipointer->ip_mprev; \
 967                         ipointer->ip_mprev->i_mnext = ip; \
 968                         if (mp->m_inodes == (xfs_inode_t *)ipointer) { \
 969                                 mp->m_inodes = ip; \
 970                         } \
 971                 } else { \
 972                         ASSERT(mp->m_inodes == (xfs_inode_t *)ipointer); \
 973                         mp->m_inodes = NULL; \
 974                         ip = NULL; \
 975                 } \
 976                 IPOINTER_CLR; \
 977         }
 978
 979 #define XFS_PREEMPT_MASK        0x7f
 980
 981         if (bypassed)
 982                 *bypassed = 0;
 983         if (XFS_MTOVFS(mp)->vfs_flag & VFS_RDONLY)
 984                 return 0;
 985         error = 0;
 986         last_error = 0;
 987         preempt = 0;
 988
 989         /* Allocate a reference marker */
 990         ipointer = (xfs_iptr_t *)kmem_zalloc(sizeof(xfs_iptr_t), KM_SLEEP);
 991
 992         fflag = XFS_B_ASYNC;            /* default is don't wait */
 993         if (flags & (SYNC_BDFLUSH | SYNC_DELWRI))
 994                 fflag = XFS_B_DELWRI;
 995         if (flags & SYNC_WAIT)
 996                 fflag = 0;              /* synchronous overrides all */
 997
 998         base_lock_flags = XFS_ILOCK_SHARED;
 999         if (flags & (SYNC_DELWRI | SYNC_CLOSE)) {
1000                 /*
1001                  * We need the I/O lock if we're going to call any of
1002                  * the flush/inval routines.
1003                  */
1004                 base_lock_flags |= XFS_IOLOCK_SHARED;
1005         }
1006
1007         XFS_MOUNT_ILOCK(mp);
1008
1009         ip = mp->m_inodes;
1010
1011         mount_locked = B_TRUE;
1012         vnode_refed  = B_FALSE;
1013
1014         IPOINTER_CLR;
1015
1016         do {
1017                 ASSERT(ipointer_in == B_FALSE);
1018                 ASSERT(vnode_refed == B_FALSE);
1019
1020                 lock_flags = base_lock_flags;
1021
1022                 /*
1023                  * There were no inodes in the list, just break out
1024                  * of the loop.
1025                  */
1026                 if (ip == NULL) {
1027                         break;
1028                 }
1029
1030                 /*
1031                  * We found another sync thread marker - skip it
1032                  */
1033                 if (ip->i_mount == NULL) {
1034                         ip = ip->i_mnext;
1035                         continue;
1036                 }
1037
1038                 vp = XFS_ITOV_NULL(ip);
1039
1040                 /*
1041                  * If the vnode is gone then this is being torn down,
1042                  * call reclaim if it is flushed, else let regular flush
1043                  * code deal with it later in the loop.
1044                  */
1045
1046                 if (vp == NULL) {
1047                         /* Skip ones already in reclaim */
1048                         if (ip->i_flags & XFS_IRECLAIM) {
1049                                 ip = ip->i_mnext;
1050                                 continue;
1051                         }
1052                         if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0) {
1053                                 ip = ip->i_mnext;
1054                         } else if ((xfs_ipincount(ip) == 0) &&
1055                                     xfs_iflock_nowait(ip)) {
1056                                 IPOINTER_INSERT(ip, mp);
1057
1058                                 xfs_finish_reclaim(ip, 1,
1059                                                 XFS_IFLUSH_DELWRI_ELSE_ASYNC);
1060
1061                                 XFS_MOUNT_ILOCK(mp);
1062                                 mount_locked = B_TRUE;
1063                                 IPOINTER_REMOVE(ip, mp);
1064                         } else {
1065                                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1066                                 ip = ip->i_mnext;
1067                         }
1068                         continue;
1069                 }
1070
1071                 if (VN_BAD(vp)) {
1072                         ip = ip->i_mnext;
1073                         continue;
1074                 }
1075
1076                 if (XFS_FORCED_SHUTDOWN(mp) && !(flags & SYNC_CLOSE)) {
1077                         XFS_MOUNT_IUNLOCK(mp);
1078                         kmem_free(ipointer, sizeof(xfs_iptr_t));
1079                         return 0;
1080                 }
1081
1082                 /*
1083                  * If this is just vfs_sync() or pflushd() calling
1084                  * then we can skip inodes for which it looks like
1085                  * there is nothing to do.  Since we don't have the
1086                  * inode locked this is racy, but these are periodic
1087                  * calls so it doesn't matter.  For the others we want
1088                  * to know for sure, so we at least try to lock them.
1089                  */
1090                 if (flags & SYNC_BDFLUSH) {
1091                         if (((ip->i_itemp == NULL) ||
1092                              !(ip->i_itemp->ili_format.ilf_fields &
1093                                XFS_ILOG_ALL)) &&
1094                             (ip->i_update_core == 0)) {
1095                                 ip = ip->i_mnext;
1096                                 continue;
1097                         }
1098                 }
1099
1100                 /*
1101                  * Try to lock without sleeping.  We're out of order with
1102                  * the inode list lock here, so if we fail we need to drop
1103                  * the mount lock and try again.  If we're called from
1104                  * bdflush() here, then don't bother.
1105                  *
1106                  * The inode lock here actually coordinates with the
1107                  * almost spurious inode lock in xfs_ireclaim() to prevent
1108                  * the vnode we handle here without a reference from
1109                  * being freed while we reference it.  If we lock the inode
1110                  * while it's on the mount list here, then the spurious inode
1111                  * lock in xfs_ireclaim() after the inode is pulled from
1112                  * the mount list will sleep until we release it here.
1113                  * This keeps the vnode from being freed while we reference
1114                  * it.
1115                  */
1116                 if (xfs_ilock_nowait(ip, lock_flags) == 0) {
1117                         if ((flags & SYNC_BDFLUSH) || (vp == NULL)) {
1118                                 ip = ip->i_mnext;
1119                                 continue;
1120                         }
1121
1122                         vp = vn_grab(vp);
1123                         if (vp == NULL) {
1124                                 ip = ip->i_mnext;
1125                                 continue;
1126                         }
1127
1128                         IPOINTER_INSERT(ip, mp);
1129                         xfs_ilock(ip, lock_flags);
1130
1131                         ASSERT(vp == XFS_ITOV(ip));
1132                         ASSERT(ip->i_mount == mp);
1133
1134                         vnode_refed = B_TRUE;
1135                 }
1136
1137                 /* From here on in the loop we may have a marker record
1138                  * in the inode list.
1139                  */
1140
1141                 if ((flags & SYNC_CLOSE)  && (vp != NULL)) {
1142                         /*
1143                          * This is the shutdown case.  We just need to
1144                          * flush and invalidate all the pages associated
1145                          * with the inode.  Drop the inode lock since
1146                          * we can't hold it across calls to the buffer
1147                          * cache.
1148                          *
1149                          * We don't set the VREMAPPING bit in the vnode
1150                          * here, because we don't hold the vnode lock
1151                          * exclusively.  It doesn't really matter, though,
1152                          * because we only come here when we're shutting
1153                          * down anyway.
1154                          */
1155                         xfs_iunlock(ip, XFS_ILOCK_SHARED);
1156
1157                         if (XFS_FORCED_SHUTDOWN(mp)) {
1158                                 bhv_vop_toss_pages(vp, 0, -1, FI_REMAPF);
1159                         } else {
1160                                 bhv_vop_flushinval_pages(vp, 0, -1, FI_REMAPF);
1161                         }
1162
1163                         xfs_ilock(ip, XFS_ILOCK_SHARED);
1164
1165                 } else if ((flags & SYNC_DELWRI) && (vp != NULL)) {
1166                         if (VN_DIRTY(vp)) {
1167                                 /* We need to have dropped the lock here,
1168                                  * so insert a marker if we have not already
1169                                  * done so.
1170                                  */
1171                                 if (mount_locked) {
1172                                         IPOINTER_INSERT(ip, mp);
1173                                 }
1174
1175                                 /*
1176                                  * Drop the inode lock since we can't hold it
1177                                  * across calls to the buffer cache.
1178                                  */
1179                                 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1180                                 error = bhv_vop_flush_pages(vp, (xfs_off_t)0,
1181                                                         -1, fflag, FI_NONE);
1182                                 xfs_ilock(ip, XFS_ILOCK_SHARED);
1183                         }
1184
1185                 }
1186
1187                 if (flags & SYNC_BDFLUSH) {
1188                         if ((flags & SYNC_ATTR) &&
1189                             ((ip->i_update_core) ||
1190                              ((ip->i_itemp != NULL) &&
1191                               (ip->i_itemp->ili_format.ilf_fields != 0)))) {
1192
1193                                 /* Insert marker and drop lock if not already
1194                                  * done.
1195                                  */
1196                                 if (mount_locked) {
1197                                         IPOINTER_INSERT(ip, mp);
1198                                 }
1199
1200                                 /*
1201                                  * We don't want the periodic flushing of the
1202                                  * inodes by vfs_sync() to interfere with
1203                                  * I/O to the file, especially read I/O
1204                                  * where it is only the access time stamp
1205                                  * that is being flushed out.  To prevent
1206                                  * long periods where we have both inode
1207                                  * locks held shared here while reading the
1208                                  * inode's buffer in from disk, we drop the
1209                                  * inode lock while reading in the inode
1210                                  * buffer.  We have to release the buffer
1211                                  * and reacquire the inode lock so that they
1212                                  * are acquired in the proper order (inode
1213                                  * locks first).  The buffer will go at the
1214                                  * end of the lru chain, though, so we can
1215                                  * expect it to still be there when we go
1216                                  * for it again in xfs_iflush().
1217                                  */
1218                                 if ((xfs_ipincount(ip) == 0) &&
1219                                     xfs_iflock_nowait(ip)) {
1220
1221                                         xfs_ifunlock(ip);
1222                                         xfs_iunlock(ip, XFS_ILOCK_SHARED);
1223
1224                                         error = xfs_itobp(mp, NULL, ip,
1225                                                           &dip, &bp, 0, 0);
1226                                         if (!error) {
1227                                                 xfs_buf_relse(bp);
1228                                         } else {
1229                                                 /* Bailing out, remove the
1230                                                  * marker and free it.
1231                                                  */
1232                                                 XFS_MOUNT_ILOCK(mp);
1233                                                 IPOINTER_REMOVE(ip, mp);
1234                                                 XFS_MOUNT_IUNLOCK(mp);
1235
1236                                                 ASSERT(!(lock_flags &
1237                                                         XFS_IOLOCK_SHARED));
1238
1239                                                 kmem_free(ipointer,
1240                                                         sizeof(xfs_iptr_t));
1241                                                 return (0);
1242                                         }
1243
1244                                         /*
1245                                          * Since we dropped the inode lock,
1246                                          * the inode may have been reclaimed.
1247                                          * Therefore, we reacquire the mount
1248                                          * lock and check to see if we were the
1249                                          * inode reclaimed. If this happened
1250                                          * then the ipointer marker will no
1251                                          * longer point back at us. In this
1252                                          * case, move ip along to the inode
1253                                          * after the marker, remove the marker
1254                                          * and continue.
1255                                          */
1256                                         XFS_MOUNT_ILOCK(mp);
1257                                         mount_locked = B_TRUE;
1258
1259                                         if (ip != ipointer->ip_mprev) {
1260                                                 IPOINTER_REMOVE(ip, mp);
1261
1262                                                 ASSERT(!vnode_refed);
1263                                                 ASSERT(!(lock_flags &
1264                                                         XFS_IOLOCK_SHARED));
1265                                                 continue;
1266                                         }
1267
1268                                         ASSERT(ip->i_mount == mp);
1269
1270                                         if (xfs_ilock_nowait(ip,
1271                                                     XFS_ILOCK_SHARED) == 0) {
1272                                                 ASSERT(ip->i_mount == mp);
1273                                                 /*
1274                                                  * We failed to reacquire
1275                                                  * the inode lock without
1276                                                  * sleeping, so just skip
1277                                                  * the inode for now.  We
1278                                                  * clear the ILOCK bit from
1279                                                  * the lock_flags so that we
1280                                                  * won't try to drop a lock
1281                                                  * we don't hold below.
1282                                                  */
1283                                                 lock_flags &= ~XFS_ILOCK_SHARED;
1284                                                 IPOINTER_REMOVE(ip_next, mp);
1285                                         } else if ((xfs_ipincount(ip) == 0) &&
1286                                                    xfs_iflock_nowait(ip)) {
1287                                                 ASSERT(ip->i_mount == mp);
1288                                                 /*
1289                                                  * Since this is vfs_sync()
1290                                                  * calling we only flush the
1291                                                  * inode out if we can lock
1292                                                  * it without sleeping and
1293                                                  * it is not pinned.  Drop
1294                                                  * the mount lock here so
1295                                                  * that we don't hold it for
1296                                                  * too long. We already have
1297                                                  * a marker in the list here.
1298                                                  */
1299                                                 XFS_MOUNT_IUNLOCK(mp);
1300                                                 mount_locked = B_FALSE;
1301                                                 error = xfs_iflush(ip,
1302                                                            XFS_IFLUSH_DELWRI);
1303                                         } else {
1304                                                 ASSERT(ip->i_mount == mp);
1305                                                 IPOINTER_REMOVE(ip_next, mp);
1306                                         }
1307                                 }
1308
1309                         }
1310
1311                 } else {
1312                         if ((flags & SYNC_ATTR) &&
1313                             ((ip->i_update_core) ||
1314                              ((ip->i_itemp != NULL) &&
1315                               (ip->i_itemp->ili_format.ilf_fields != 0)))) {
1316                                 if (mount_locked) {
1317                                         IPOINTER_INSERT(ip, mp);
1318                                 }
1319
1320                                 if (flags & SYNC_WAIT) {
1321                                         xfs_iflock(ip);
1322                                         error = xfs_iflush(ip,
1323                                                            XFS_IFLUSH_SYNC);
1324                                 } else {
1325                                         /*
1326                                          * If we can't acquire the flush
1327                                          * lock, then the inode is already
1328                                          * being flushed so don't bother
1329                                          * waiting.  If we can lock it then
1330                                          * do a delwri flush so we can
1331                                          * combine multiple inode flushes
1332                                          * in each disk write.
1333                                          */
1334                                         if (xfs_iflock_nowait(ip)) {
1335                                                 error = xfs_iflush(ip,
1336                                                            XFS_IFLUSH_DELWRI);
1337                                         }
1338                                         else if (bypassed)
1339                                                 (*bypassed)++;
1340                                 }
1341                         }
1342                 }
1343
1344                 if (lock_flags != 0) {
1345                         xfs_iunlock(ip, lock_flags);
1346                 }
1347
1348                 if (vnode_refed) {
1349                         /*
1350                          * If we had to take a reference on the vnode
1351                          * above, then wait until after we've unlocked
1352                          * the inode to release the reference.  This is
1353                          * because we can be already holding the inode
1354                          * lock when VN_RELE() calls xfs_inactive().
1355                          *
1356                          * Make sure to drop the mount lock before calling
1357                          * VN_RELE() so that we don't trip over ourselves if
1358                          * we have to go for the mount lock again in the
1359                          * inactive code.
1360                          */
1361                         if (mount_locked) {
1362                                 IPOINTER_INSERT(ip, mp);
1363                         }
1364
1365                         VN_RELE(vp);
1366
1367                         vnode_refed = B_FALSE;
1368                 }
1369
1370                 if (error) {
1371                         last_error = error;
1372                 }
1373
1374                 /*
1375                  * bail out if the filesystem is corrupted.
1376                  */
1377                 if (error == EFSCORRUPTED)  {
1378                         if (!mount_locked) {
1379                                 XFS_MOUNT_ILOCK(mp);
1380                                 IPOINTER_REMOVE(ip, mp);
1381                         }
1382                         XFS_MOUNT_IUNLOCK(mp);
1383                         ASSERT(ipointer_in == B_FALSE);
1384                         kmem_free(ipointer, sizeof(xfs_iptr_t));
1385                         return XFS_ERROR(error);
1386                 }
1387
1388                 /* Let other threads have a chance at the mount lock
1389                  * if we have looped many times without dropping the
1390                  * lock.
1391                  */
1392                 if ((++preempt & XFS_PREEMPT_MASK) == 0) {
1393                         if (mount_locked) {
1394                                 IPOINTER_INSERT(ip, mp);
1395                         }
1396                 }
1397
1398                 if (mount_locked == B_FALSE) {
1399                         XFS_MOUNT_ILOCK(mp);
1400                         mount_locked = B_TRUE;
1401                         IPOINTER_REMOVE(ip, mp);
1402                         continue;
1403                 }
1404
1405                 ASSERT(ipointer_in == B_FALSE);
1406                 ip = ip->i_mnext;
1407
1408         } while (ip != mp->m_inodes);
1409
1410         XFS_MOUNT_IUNLOCK(mp);
1411
1412         ASSERT(ipointer_in == B_FALSE);
1413
1414         kmem_free(ipointer, sizeof(xfs_iptr_t));
1415         return XFS_ERROR(last_error);
1416 }
1417
1418 /*
1419  * xfs sync routine for internal use
1420  *
1421  * This routine supports all of the flags defined for the generic vfs_sync
1422  * interface as explained above under xfs_sync.  In the interests of not
1423  * changing interfaces within the 6.5 family, additional internally-
1424  * required functions are specified within a separate xflags parameter,
1425  * only available by calling this routine.
1426  *
1427  */
1428 int
1429 xfs_syncsub(
1430         xfs_mount_t     *mp,
1431         int             flags,
1432         int             xflags,
1433         int             *bypassed)
1434 {
1435         int             error = 0;
1436         int             last_error = 0;
1437         uint            log_flags = XFS_LOG_FORCE;
1438         xfs_buf_t       *bp;
1439         xfs_buf_log_item_t      *bip;
1440
1441         /*
1442          * Sync out the log.  This ensures that the log is periodically
1443          * flushed even if there is not enough activity to fill it up.
1444          */
1445         if (flags & SYNC_WAIT)
1446                 log_flags |= XFS_LOG_SYNC;
1447
1448         xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1449
1450         if (flags & (SYNC_ATTR|SYNC_DELWRI)) {
1451                 if (flags & SYNC_BDFLUSH)
1452                         xfs_finish_reclaim_all(mp, 1);
1453                 else
1454                         error = xfs_sync_inodes(mp, flags, xflags, bypassed);
1455         }
1456
1457         /*
1458          * Flushing out dirty data above probably generated more
1459          * log activity, so if this isn't vfs_sync() then flush
1460          * the log again.
1461          */
1462         if (flags & SYNC_DELWRI) {
1463                 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1464         }
1465
1466         if (flags & SYNC_FSDATA) {
1467                 /*
1468                  * If this is vfs_sync() then only sync the superblock
1469                  * if we can lock it without sleeping and it is not pinned.
1470                  */
1471                 if (flags & SYNC_BDFLUSH) {
1472                         bp = xfs_getsb(mp, XFS_BUF_TRYLOCK);
1473                         if (bp != NULL) {
1474                                 bip = XFS_BUF_FSPRIVATE(bp,xfs_buf_log_item_t*);
1475                                 if ((bip != NULL) &&
1476                                     xfs_buf_item_dirty(bip)) {
1477                                         if (!(XFS_BUF_ISPINNED(bp))) {
1478                                                 XFS_BUF_ASYNC(bp);
1479                                                 error = xfs_bwrite(mp, bp);
1480                                         } else {
1481                                                 xfs_buf_relse(bp);
1482                                         }
1483                                 } else {
1484                                         xfs_buf_relse(bp);
1485                                 }
1486                         }
1487                 } else {
1488                         bp = xfs_getsb(mp, 0);
1489                         /*
1490                          * If the buffer is pinned then push on the log so
1491                          * we won't get stuck waiting in the write for
1492                          * someone, maybe ourselves, to flush the log.
1493                          * Even though we just pushed the log above, we
1494                          * did not have the superblock buffer locked at
1495                          * that point so it can become pinned in between
1496                          * there and here.
1497                          */
1498                         if (XFS_BUF_ISPINNED(bp))
1499                                 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
1500                         if (flags & SYNC_WAIT)
1501                                 XFS_BUF_UNASYNC(bp);
1502                         else
1503                                 XFS_BUF_ASYNC(bp);
1504                         error = xfs_bwrite(mp, bp);
1505                 }
1506                 if (error) {
1507                         last_error = error;
1508                 }
1509         }
1510
1511         /*
1512          * If this is the periodic sync, then kick some entries out of
1513          * the reference cache.  This ensures that idle entries are
1514          * eventually kicked out of the cache.
1515          */
1516         if (flags & SYNC_REFCACHE) {
1517                 if (flags & SYNC_WAIT)
1518                         xfs_refcache_purge_mp(mp);
1519                 else
1520                         xfs_refcache_purge_some(mp);
1521         }
1522
1523         /*
1524          * Now check to see if the log needs a "dummy" transaction.
1525          */
1526
1527         if (!(flags & SYNC_REMOUNT) && xfs_log_need_covered(mp)) {
1528                 xfs_trans_t *tp;
1529                 xfs_inode_t *ip;
1530
1531                 /*
1532                  * Put a dummy transaction in the log to tell
1533                  * recovery that all others are OK.
1534                  */
1535                 tp = xfs_trans_alloc(mp, XFS_TRANS_DUMMY1);
1536                 if ((error = xfs_trans_reserve(tp, 0,
1537                                 XFS_ICHANGE_LOG_RES(mp),
1538                                 0, 0, 0)))  {
1539                         xfs_trans_cancel(tp, 0);
1540                         return error;
1541                 }
1542
1543                 ip = mp->m_rootip;
1544                 xfs_ilock(ip, XFS_ILOCK_EXCL);
1545
1546                 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1547                 xfs_trans_ihold(tp, ip);
1548                 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1549                 error = xfs_trans_commit(tp, 0, NULL);
1550                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1551                 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1552         }
1553
1554         /*
1555          * When shutting down, we need to insure that the AIL is pushed
1556          * to disk or the filesystem can appear corrupt from the PROM.
1557          */
1558         if ((flags & (SYNC_CLOSE|SYNC_WAIT)) == (SYNC_CLOSE|SYNC_WAIT)) {
1559                 XFS_bflush(mp->m_ddev_targp);
1560                 if (mp->m_rtdev_targp) {
1561                         XFS_bflush(mp->m_rtdev_targp);
1562                 }
1563         }
1564
1565         return XFS_ERROR(last_error);
1566 }
1567
1568 /*
1569  * xfs_vget - called by DMAPI and NFSD to get vnode from file handle
1570  */
1571 STATIC int
1572 xfs_vget(
1573         bhv_desc_t      *bdp,
1574         bhv_vnode_t     **vpp,
1575         fid_t           *fidp)
1576 {
1577         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
1578         xfs_fid_t       *xfid = (struct xfs_fid *)fidp;
1579         xfs_inode_t     *ip;
1580         int             error;
1581         xfs_ino_t       ino;
1582         unsigned int    igen;
1583
1584         /*
1585          * Invalid.  Since handles can be created in user space and passed in
1586          * via gethandle(), this is not cause for a panic.
1587          */
1588         if (xfid->xfs_fid_len != sizeof(*xfid) - sizeof(xfid->xfs_fid_len))
1589                 return XFS_ERROR(EINVAL);
1590
1591         ino  = xfid->xfs_fid_ino;
1592         igen = xfid->xfs_fid_gen;
1593
1594         /*
1595          * NFS can sometimes send requests for ino 0.  Fail them gracefully.
1596          */
1597         if (ino == 0)
1598                 return XFS_ERROR(ESTALE);
1599
1600         error = xfs_iget(mp, NULL, ino, 0, XFS_ILOCK_SHARED, &ip, 0);
1601         if (error) {
1602                 *vpp = NULL;
1603                 return error;
1604         }
1605
1606         if (ip == NULL) {
1607                 *vpp = NULL;
1608                 return XFS_ERROR(EIO);
1609         }
1610
1611         if (ip->i_d.di_mode == 0 || ip->i_d.di_gen != igen) {
1612                 xfs_iput_new(ip, XFS_ILOCK_SHARED);
1613                 *vpp = NULL;
1614                 return XFS_ERROR(ENOENT);
1615         }
1616
1617         *vpp = XFS_ITOV(ip);
1618         xfs_iunlock(ip, XFS_ILOCK_SHARED);
1619         return 0;
1620 }
1621
1622
1623 #define MNTOPT_LOGBUFS  "logbufs"       /* number of XFS log buffers */
1624 #define MNTOPT_LOGBSIZE "logbsize"      /* size of XFS log buffers */
1625 #define MNTOPT_LOGDEV   "logdev"        /* log device */
1626 #define MNTOPT_RTDEV    "rtdev"         /* realtime I/O device */
1627 #define MNTOPT_BIOSIZE  "biosize"       /* log2 of preferred buffered io size */
1628 #define MNTOPT_WSYNC    "wsync"         /* safe-mode nfs compatible mount */
1629 #define MNTOPT_INO64    "ino64"         /* force inodes into 64-bit range */
1630 #define MNTOPT_NOALIGN  "noalign"       /* turn off stripe alignment */
1631 #define MNTOPT_SWALLOC  "swalloc"       /* turn on stripe width allocation */
1632 #define MNTOPT_SUNIT    "sunit"         /* data volume stripe unit */
1633 #define MNTOPT_SWIDTH   "swidth"        /* data volume stripe width */
1634 #define MNTOPT_NOUUID   "nouuid"        /* ignore filesystem UUID */
1635 #define MNTOPT_MTPT     "mtpt"          /* filesystem mount point */
1636 #define MNTOPT_GRPID    "grpid"         /* group-ID from parent directory */
1637 #define MNTOPT_NOGRPID  "nogrpid"       /* group-ID from current process */
1638 #define MNTOPT_BSDGROUPS    "bsdgroups"    /* group-ID from parent directory */
1639 #define MNTOPT_SYSVGROUPS   "sysvgroups"   /* group-ID from current process */
1640 #define MNTOPT_ALLOCSIZE    "allocsize"    /* preferred allocation size */
1641 #define MNTOPT_IHASHSIZE    "ihashsize"    /* size of inode hash table */
1642 #define MNTOPT_NORECOVERY   "norecovery"   /* don't run XFS recovery */
1643 #define MNTOPT_BARRIER  "barrier"       /* use writer barriers for log write and
1644                                          * unwritten extent conversion */
1645 #define MNTOPT_NOBARRIER "nobarrier"    /* .. disable */
1646 #define MNTOPT_OSYNCISOSYNC "osyncisosync" /* o_sync is REALLY o_sync */
1647 #define MNTOPT_64BITINODE   "inode64"   /* inodes can be allocated anywhere */
1648 #define MNTOPT_IKEEP    "ikeep"         /* do not free empty inode clusters */
1649 #define MNTOPT_NOIKEEP  "noikeep"       /* free empty inode clusters */
1650 #define MNTOPT_LARGEIO     "largeio"    /* report large I/O sizes in stat() */
1651 #define MNTOPT_NOLARGEIO   "nolargeio"  /* do not report large I/O sizes
1652                                          * in stat(). */
1653 #define MNTOPT_ATTR2    "attr2"         /* do use attr2 attribute format */
1654 #define MNTOPT_NOATTR2  "noattr2"       /* do not use attr2 attribute format */
1655
1656 STATIC unsigned long
1657 suffix_strtoul(char *s, char **endp, unsigned int base)
1658 {
1659         int     last, shift_left_factor = 0;
1660         char    *value = s;
1661
1662         last = strlen(value) - 1;
1663         if (value[last] == 'K' || value[last] == 'k') {
1664                 shift_left_factor = 10;
1665                 value[last] = '\0';
1666         }
1667         if (value[last] == 'M' || value[last] == 'm') {
1668                 shift_left_factor = 20;
1669                 value[last] = '\0';
1670         }
1671         if (value[last] == 'G' || value[last] == 'g') {
1672                 shift_left_factor = 30;
1673                 value[last] = '\0';
1674         }
1675
1676         return simple_strtoul((const char *)s, endp, base) << shift_left_factor;
1677 }
1678
1679 STATIC int
1680 xfs_parseargs(
1681         struct bhv_desc         *bhv,
1682         char                    *options,
1683         struct xfs_mount_args   *args,
1684         int                     update)
1685 {
1686         bhv_vfs_t               *vfsp = bhvtovfs(bhv);
1687         char                    *this_char, *value, *eov;
1688         int                     dsunit, dswidth, vol_dsunit, vol_dswidth;
1689         int                     iosize;
1690
1691         args->flags |= XFSMNT_IDELETE;
1692         args->flags |= XFSMNT_BARRIER;
1693         args->flags2 |= XFSMNT2_COMPAT_IOSIZE;
1694
1695         if (!options)
1696                 goto done;
1697
1698         iosize = dsunit = dswidth = vol_dsunit = vol_dswidth = 0;
1699
1700         while ((this_char = strsep(&options, ",")) != NULL) {
1701                 if (!*this_char)
1702                         continue;
1703                 if ((value = strchr(this_char, '=')) != NULL)
1704                         *value++ = 0;
1705
1706                 if (!strcmp(this_char, MNTOPT_LOGBUFS)) {
1707                         if (!value || !*value) {
1708                                 cmn_err(CE_WARN,
1709                                         "XFS: %s option requires an argument",
1710                                         this_char);
1711                                 return EINVAL;
1712                         }
1713                         args->logbufs = simple_strtoul(value, &eov, 10);
1714                 } else if (!strcmp(this_char, MNTOPT_LOGBSIZE)) {
1715                         if (!value || !*value) {
1716                                 cmn_err(CE_WARN,
1717                                         "XFS: %s option requires an argument",
1718                                         this_char);
1719                                 return EINVAL;
1720                         }
1721                         args->logbufsize = suffix_strtoul(value, &eov, 10);
1722                 } else if (!strcmp(this_char, MNTOPT_LOGDEV)) {
1723                         if (!value || !*value) {
1724                                 cmn_err(CE_WARN,
1725                                         "XFS: %s option requires an argument",
1726                                         this_char);
1727                                 return EINVAL;
1728                         }
1729                         strncpy(args->logname, value, MAXNAMELEN);
1730                 } else if (!strcmp(this_char, MNTOPT_MTPT)) {
1731                         if (!value || !*value) {
1732                                 cmn_err(CE_WARN,
1733                                         "XFS: %s option requires an argument",
1734                                         this_char);
1735                                 return EINVAL;
1736                         }
1737                         strncpy(args->mtpt, value, MAXNAMELEN);
1738                 } else if (!strcmp(this_char, MNTOPT_RTDEV)) {
1739                         if (!value || !*value) {
1740                                 cmn_err(CE_WARN,
1741                                         "XFS: %s option requires an argument",
1742                                         this_char);
1743                                 return EINVAL;
1744                         }
1745                         strncpy(args->rtname, value, MAXNAMELEN);
1746                 } else if (!strcmp(this_char, MNTOPT_BIOSIZE)) {
1747                         if (!value || !*value) {
1748                                 cmn_err(CE_WARN,
1749                                         "XFS: %s option requires an argument",
1750                                         this_char);
1751                                 return EINVAL;
1752                         }
1753                         iosize = simple_strtoul(value, &eov, 10);
1754                         args->flags |= XFSMNT_IOSIZE;
1755                         args->iosizelog = (uint8_t) iosize;
1756                 } else if (!strcmp(this_char, MNTOPT_ALLOCSIZE)) {
1757                         if (!value || !*value) {
1758                                 cmn_err(CE_WARN,
1759                                         "XFS: %s option requires an argument",
1760                                         this_char);
1761                                 return EINVAL;
1762                         }
1763                         iosize = suffix_strtoul(value, &eov, 10);
1764                         args->flags |= XFSMNT_IOSIZE;
1765                         args->iosizelog = ffs(iosize) - 1;
1766                 } else if (!strcmp(this_char, MNTOPT_IHASHSIZE)) {
1767                         if (!value || !*value) {
1768                                 cmn_err(CE_WARN,
1769                                         "XFS: %s option requires an argument",
1770                                         this_char);
1771                                 return EINVAL;
1772                         }
1773                         args->flags |= XFSMNT_IHASHSIZE;
1774                         args->ihashsize = simple_strtoul(value, &eov, 10);
1775                 } else if (!strcmp(this_char, MNTOPT_GRPID) ||
1776                            !strcmp(this_char, MNTOPT_BSDGROUPS)) {
1777                         vfsp->vfs_flag |= VFS_GRPID;
1778                 } else if (!strcmp(this_char, MNTOPT_NOGRPID) ||
1779                            !strcmp(this_char, MNTOPT_SYSVGROUPS)) {
1780                         vfsp->vfs_flag &= ~VFS_GRPID;
1781                 } else if (!strcmp(this_char, MNTOPT_WSYNC)) {
1782                         args->flags |= XFSMNT_WSYNC;
1783                 } else if (!strcmp(this_char, MNTOPT_OSYNCISOSYNC)) {
1784                         args->flags |= XFSMNT_OSYNCISOSYNC;
1785                 } else if (!strcmp(this_char, MNTOPT_NORECOVERY)) {
1786                         args->flags |= XFSMNT_NORECOVERY;
1787                 } else if (!strcmp(this_char, MNTOPT_INO64)) {
1788                         args->flags |= XFSMNT_INO64;
1789 #if !XFS_BIG_INUMS
1790                         cmn_err(CE_WARN,
1791                                 "XFS: %s option not allowed on this system",
1792                                 this_char);
1793                         return EINVAL;
1794 #endif
1795                 } else if (!strcmp(this_char, MNTOPT_NOALIGN)) {
1796                         args->flags |= XFSMNT_NOALIGN;
1797                 } else if (!strcmp(this_char, MNTOPT_SWALLOC)) {
1798                         args->flags |= XFSMNT_SWALLOC;
1799                 } else if (!strcmp(this_char, MNTOPT_SUNIT)) {
1800                         if (!value || !*value) {
1801                                 cmn_err(CE_WARN,
1802                                         "XFS: %s option requires an argument",
1803                                         this_char);
1804                                 return EINVAL;
1805                         }
1806                         dsunit = simple_strtoul(value, &eov, 10);
1807                 } else if (!strcmp(this_char, MNTOPT_SWIDTH)) {
1808                         if (!value || !*value) {
1809                                 cmn_err(CE_WARN,
1810                                         "XFS: %s option requires an argument",
1811                                         this_char);
1812                                 return EINVAL;
1813                         }
1814                         dswidth = simple_strtoul(value, &eov, 10);
1815                 } else if (!strcmp(this_char, MNTOPT_64BITINODE)) {
1816                         args->flags &= ~XFSMNT_32BITINODES;
1817 #if !XFS_BIG_INUMS
1818                         cmn_err(CE_WARN,
1819                                 "XFS: %s option not allowed on this system",
1820                                 this_char);
1821                         return EINVAL;
1822 #endif
1823                 } else if (!strcmp(this_char, MNTOPT_NOUUID)) {
1824                         args->flags |= XFSMNT_NOUUID;
1825                 } else if (!strcmp(this_char, MNTOPT_BARRIER)) {
1826                         args->flags |= XFSMNT_BARRIER;
1827                 } else if (!strcmp(this_char, MNTOPT_NOBARRIER)) {
1828                         args->flags &= ~XFSMNT_BARRIER;
1829                 } else if (!strcmp(this_char, MNTOPT_IKEEP)) {
1830                         args->flags &= ~XFSMNT_IDELETE;
1831                 } else if (!strcmp(this_char, MNTOPT_NOIKEEP)) {
1832                         args->flags |= XFSMNT_IDELETE;
1833                 } else if (!strcmp(this_char, MNTOPT_LARGEIO)) {
1834                         args->flags2 &= ~XFSMNT2_COMPAT_IOSIZE;
1835                 } else if (!strcmp(this_char, MNTOPT_NOLARGEIO)) {
1836                         args->flags2 |= XFSMNT2_COMPAT_IOSIZE;
1837                 } else if (!strcmp(this_char, MNTOPT_ATTR2)) {
1838                         args->flags |= XFSMNT_ATTR2;
1839                 } else if (!strcmp(this_char, MNTOPT_NOATTR2)) {
1840                         args->flags &= ~XFSMNT_ATTR2;
1841                 } else if (!strcmp(this_char, "osyncisdsync")) {
1842                         /* no-op, this is now the default */
1843                         cmn_err(CE_WARN,
1844         "XFS: osyncisdsync is now the default, option is deprecated.");
1845                 } else if (!strcmp(this_char, "irixsgid")) {
1846                         cmn_err(CE_WARN,
1847         "XFS: irixsgid is now a sysctl(2) variable, option is deprecated.");
1848                 } else {
1849                         cmn_err(CE_WARN,
1850                                 "XFS: unknown mount option [%s].", this_char);
1851                         return EINVAL;
1852                 }
1853         }
1854
1855         if (args->flags & XFSMNT_NORECOVERY) {
1856                 if ((vfsp->vfs_flag & VFS_RDONLY) == 0) {
1857                         cmn_err(CE_WARN,
1858                                 "XFS: no-recovery mounts must be read-only.");
1859                         return EINVAL;
1860                 }
1861         }
1862
1863         if ((args->flags & XFSMNT_NOALIGN) && (dsunit || dswidth)) {
1864                 cmn_err(CE_WARN,
1865         "XFS: sunit and swidth options incompatible with the noalign option");
1866                 return EINVAL;
1867         }
1868
1869         if ((dsunit && !dswidth) || (!dsunit && dswidth)) {
1870                 cmn_err(CE_WARN,
1871                         "XFS: sunit and swidth must be specified together");
1872                 return EINVAL;
1873         }
1874
1875         if (dsunit && (dswidth % dsunit != 0)) {
1876                 cmn_err(CE_WARN,
1877         "XFS: stripe width (%d) must be a multiple of the stripe unit (%d)",
1878                         dswidth, dsunit);
1879                 return EINVAL;
1880         }
1881
1882         if ((args->flags & XFSMNT_NOALIGN) != XFSMNT_NOALIGN) {
1883                 if (dsunit) {
1884                         args->sunit = dsunit;
1885                         args->flags |= XFSMNT_RETERR;
1886                 } else {
1887                         args->sunit = vol_dsunit;
1888                 }
1889                 dswidth ? (args->swidth = dswidth) :
1890                           (args->swidth = vol_dswidth);
1891         } else {
1892                 args->sunit = args->swidth = 0;
1893         }
1894
1895 done:
1896         if (args->flags & XFSMNT_32BITINODES)
1897                 vfsp->vfs_flag |= VFS_32BITINODES;
1898         if (args->flags2)
1899                 args->flags |= XFSMNT_FLAGS2;
1900         return 0;
1901 }
1902
1903 STATIC int
1904 xfs_showargs(
1905         struct bhv_desc         *bhv,
1906         struct seq_file         *m)
1907 {
1908         static struct proc_xfs_info {
1909                 int     flag;
1910                 char    *str;
1911         } xfs_info[] = {
1912                 /* the few simple ones we can get from the mount struct */
1913                 { XFS_MOUNT_WSYNC,              "," MNTOPT_WSYNC },
1914                 { XFS_MOUNT_INO64,              "," MNTOPT_INO64 },
1915                 { XFS_MOUNT_NOALIGN,            "," MNTOPT_NOALIGN },
1916                 { XFS_MOUNT_SWALLOC,            "," MNTOPT_SWALLOC },
1917                 { XFS_MOUNT_NOUUID,             "," MNTOPT_NOUUID },
1918                 { XFS_MOUNT_NORECOVERY,         "," MNTOPT_NORECOVERY },
1919                 { XFS_MOUNT_OSYNCISOSYNC,       "," MNTOPT_OSYNCISOSYNC },
1920                 { 0, NULL }
1921         };
1922         struct proc_xfs_info    *xfs_infop;
1923         struct xfs_mount        *mp = XFS_BHVTOM(bhv);
1924         struct bhv_vfs          *vfsp = XFS_MTOVFS(mp);
1925
1926         for (xfs_infop = xfs_info; xfs_infop->flag; xfs_infop++) {
1927                 if (mp->m_flags & xfs_infop->flag)
1928                         seq_puts(m, xfs_infop->str);
1929         }
1930
1931         if (mp->m_flags & XFS_MOUNT_IHASHSIZE)
1932                 seq_printf(m, "," MNTOPT_IHASHSIZE "=%d", mp->m_ihsize);
1933
1934         if (mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)
1935                 seq_printf(m, "," MNTOPT_ALLOCSIZE "=%dk",
1936                                 (int)(1 << mp->m_writeio_log) >> 10);
1937
1938         if (mp->m_logbufs > 0)
1939                 seq_printf(m, "," MNTOPT_LOGBUFS "=%d", mp->m_logbufs);
1940         if (mp->m_logbsize > 0)
1941                 seq_printf(m, "," MNTOPT_LOGBSIZE "=%dk", mp->m_logbsize >> 10);
1942
1943         if (mp->m_logname)
1944                 seq_printf(m, "," MNTOPT_LOGDEV "=%s", mp->m_logname);
1945         if (mp->m_rtname)
1946                 seq_printf(m, "," MNTOPT_RTDEV "=%s", mp->m_rtname);
1947
1948         if (mp->m_dalign > 0)
1949                 seq_printf(m, "," MNTOPT_SUNIT "=%d",
1950                                 (int)XFS_FSB_TO_BB(mp, mp->m_dalign));
1951         if (mp->m_swidth > 0)
1952                 seq_printf(m, "," MNTOPT_SWIDTH "=%d",
1953                                 (int)XFS_FSB_TO_BB(mp, mp->m_swidth));
1954
1955         if (!(mp->m_flags & XFS_MOUNT_IDELETE))
1956                 seq_printf(m, "," MNTOPT_IKEEP);
1957         if (!(mp->m_flags & XFS_MOUNT_COMPAT_IOSIZE))
1958                 seq_printf(m, "," MNTOPT_LARGEIO);
1959
1960         if (!(vfsp->vfs_flag & VFS_32BITINODES))
1961                 seq_printf(m, "," MNTOPT_64BITINODE);
1962         if (vfsp->vfs_flag & VFS_GRPID)
1963                 seq_printf(m, "," MNTOPT_GRPID);
1964
1965         return 0;
1966 }
1967
1968 STATIC void
1969 xfs_freeze(
1970         bhv_desc_t      *bdp)
1971 {
1972         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
1973
1974         while (atomic_read(&mp->m_active_trans) > 0)
1975                 delay(100);
1976
1977         /* Push the superblock and write an unmount record */
1978         xfs_log_unmount_write(mp);
1979         xfs_unmountfs_writesb(mp);
1980         xfs_fs_log_dummy(mp);
1981 }
1982
1983
1984 bhv_vfsops_t xfs_vfsops = {
1985         BHV_IDENTITY_INIT(VFS_BHV_XFS,VFS_POSITION_XFS),
1986         .vfs_parseargs          = xfs_parseargs,
1987         .vfs_showargs           = xfs_showargs,
1988         .vfs_mount              = xfs_mount,
1989         .vfs_unmount            = xfs_unmount,
1990         .vfs_mntupdate          = xfs_mntupdate,
1991         .vfs_root               = xfs_root,
1992         .vfs_statvfs            = xfs_statvfs,
1993         .vfs_sync               = xfs_sync,
1994         .vfs_vget               = xfs_vget,
1995         .vfs_dmapiops           = (vfs_dmapiops_t)fs_nosys,
1996         .vfs_quotactl           = (vfs_quotactl_t)fs_nosys,
1997         .vfs_init_vnode         = xfs_initialize_vnode,
1998         .vfs_force_shutdown     = xfs_do_force_shutdown,
1999         .vfs_freeze             = xfs_freeze,
2000 };