fs/xfs/xfs_vfsops.c

   1 /*
   2  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
   3  * All Rights Reserved.
   4  *
   5  * This program is free software; you can redistribute it and/or
   6  * modify it under the terms of the GNU General Public License as
   7  * published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it would be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write the Free Software Foundation,
  16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17  */
  18 #include "xfs.h"
  19 #include "xfs_fs.h"
  20 #include "xfs_types.h"
  21 #include "xfs_bit.h"
  22 #include "xfs_log.h"
  23 #include "xfs_inum.h"
  24 #include "xfs_trans.h"
  25 #include "xfs_sb.h"
  26 #include "xfs_ag.h"
  27 #include "xfs_dir2.h"
  28 #include "xfs_dmapi.h"
  29 #include "xfs_mount.h"
  30 #include "xfs_da_btree.h"
  31 #include "xfs_bmap_btree.h"
  32 #include "xfs_ialloc_btree.h"
  33 #include "xfs_alloc_btree.h"
  34 #include "xfs_dir2_sf.h"
  35 #include "xfs_attr_sf.h"
  36 #include "xfs_dinode.h"
  37 #include "xfs_inode.h"
  38 #include "xfs_inode_item.h"
  39 #include "xfs_btree.h"
  40 #include "xfs_alloc.h"
  41 #include "xfs_ialloc.h"
  42 #include "xfs_quota.h"
  43 #include "xfs_error.h"
  44 #include "xfs_bmap.h"
  45 #include "xfs_rw.h"
  46 #include "xfs_refcache.h"
  47 #include "xfs_buf_item.h"
  48 #include "xfs_log_priv.h"
  49 #include "xfs_dir2_trace.h"
  50 #include "xfs_extfree_item.h"
  51 #include "xfs_acl.h"
  52 #include "xfs_attr.h"
  53 #include "xfs_clnt.h"
  54 #include "xfs_mru_cache.h"
  55 #include "xfs_filestream.h"
  56 #include "xfs_fsops.h"
  57 #include "xfs_vnodeops.h"
  58
  59
  60 STATIC int      xfs_sync(bhv_desc_t *, int, cred_t *);
  61
  62 int
  63 xfs_init(void)
  64 {
  65         extern kmem_zone_t      *xfs_bmap_free_item_zone;
  66         extern kmem_zone_t      *xfs_btree_cur_zone;
  67         extern kmem_zone_t      *xfs_trans_zone;
  68         extern kmem_zone_t      *xfs_buf_item_zone;
  69         extern kmem_zone_t      *xfs_dabuf_zone;
  70 #ifdef XFS_DABUF_DEBUG
  71         extern lock_t           xfs_dabuf_global_lock;
  72         spinlock_init(&xfs_dabuf_global_lock, "xfsda");
  73 #endif
  74
  75         /*
  76          * Initialize all of the zone allocators we use.
  77          */
  78         xfs_bmap_free_item_zone = kmem_zone_init(sizeof(xfs_bmap_free_item_t),
  79                                                  "xfs_bmap_free_item");
  80         xfs_btree_cur_zone = kmem_zone_init(sizeof(xfs_btree_cur_t),
  81                                             "xfs_btree_cur");
  82         xfs_trans_zone = kmem_zone_init(sizeof(xfs_trans_t), "xfs_trans");
  83         xfs_da_state_zone =
  84                 kmem_zone_init(sizeof(xfs_da_state_t), "xfs_da_state");
  85         xfs_dabuf_zone = kmem_zone_init(sizeof(xfs_dabuf_t), "xfs_dabuf");
  86         xfs_ifork_zone = kmem_zone_init(sizeof(xfs_ifork_t), "xfs_ifork");
  87         xfs_acl_zone_init(xfs_acl_zone, "xfs_acl");
  88         xfs_mru_cache_init();
  89         xfs_filestream_init();
  90
  91         /*
  92          * The size of the zone allocated buf log item is the maximum
  93          * size possible under XFS.  This wastes a little bit of memory,
  94          * but it is much faster.
  95          */
  96         xfs_buf_item_zone =
  97                 kmem_zone_init((sizeof(xfs_buf_log_item_t) +
  98                                 (((XFS_MAX_BLOCKSIZE / XFS_BLI_CHUNK) /
  99                                   NBWORD) * sizeof(int))),
 100                                "xfs_buf_item");
 101         xfs_efd_zone =
 102                 kmem_zone_init((sizeof(xfs_efd_log_item_t) +
 103                                ((XFS_EFD_MAX_FAST_EXTENTS - 1) *
 104                                  sizeof(xfs_extent_t))),
 105                                       "xfs_efd_item");
 106         xfs_efi_zone =
 107                 kmem_zone_init((sizeof(xfs_efi_log_item_t) +
 108                                ((XFS_EFI_MAX_FAST_EXTENTS - 1) *
 109                                  sizeof(xfs_extent_t))),
 110                                       "xfs_efi_item");
 111
 112         /*
 113          * These zones warrant special memory allocator hints
 114          */
 115         xfs_inode_zone =
 116                 kmem_zone_init_flags(sizeof(xfs_inode_t), "xfs_inode",
 117                                         KM_ZONE_HWALIGN | KM_ZONE_RECLAIM |
 118                                         KM_ZONE_SPREAD, NULL);
 119         xfs_ili_zone =
 120                 kmem_zone_init_flags(sizeof(xfs_inode_log_item_t), "xfs_ili",
 121                                         KM_ZONE_SPREAD, NULL);
 122         xfs_icluster_zone =
 123                 kmem_zone_init_flags(sizeof(xfs_icluster_t), "xfs_icluster",
 124                                         KM_ZONE_SPREAD, NULL);
 125
 126         /*
 127          * Allocate global trace buffers.
 128          */
 129 #ifdef XFS_ALLOC_TRACE
 130         xfs_alloc_trace_buf = ktrace_alloc(XFS_ALLOC_TRACE_SIZE, KM_SLEEP);
 131 #endif
 132 #ifdef XFS_BMAP_TRACE
 133         xfs_bmap_trace_buf = ktrace_alloc(XFS_BMAP_TRACE_SIZE, KM_SLEEP);
 134 #endif
 135 #ifdef XFS_BMBT_TRACE
 136         xfs_bmbt_trace_buf = ktrace_alloc(XFS_BMBT_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_icluster_zone;
 169
 170         xfs_cleanup_procfs();
 171         xfs_sysctl_unregister();
 172         xfs_refcache_destroy();
 173         xfs_filestream_uninit();
 174         xfs_mru_cache_uninit();
 175         xfs_acl_zone_destroy(xfs_acl_zone);
 176
 177 #ifdef XFS_DIR2_TRACE
 178         ktrace_free(xfs_dir2_trace_buf);
 179 #endif
 180 #ifdef XFS_ATTR_TRACE
 181         ktrace_free(xfs_attr_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_icluster_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 < XLOG_MIN_RECORD_BSIZE ||
 244              ap->logbufsize > XLOG_MAX_RECORD_BSIZE ||
 245              !is_power_of_2(ap->logbufsize))) {
 246                 cmn_err(CE_WARN,
 247         "XFS: invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
 248                         ap->logbufsize);
 249                 return XFS_ERROR(EINVAL);
 250         }
 251         mp->m_logbsize = ap->logbufsize;
 252         mp->m_fsname_len = strlen(ap->fsname) + 1;
 253         mp->m_fsname = kmem_alloc(mp->m_fsname_len, KM_SLEEP);
 254         strcpy(mp->m_fsname, ap->fsname);
 255         if (ap->rtname[0]) {
 256                 mp->m_rtname = kmem_alloc(strlen(ap->rtname) + 1, KM_SLEEP);
 257                 strcpy(mp->m_rtname, ap->rtname);
 258         }
 259         if (ap->logname[0]) {
 260                 mp->m_logname = kmem_alloc(strlen(ap->logname) + 1, KM_SLEEP);
 261                 strcpy(mp->m_logname, ap->logname);
 262         }
 263
 264         if (ap->flags & XFSMNT_WSYNC)
 265                 mp->m_flags |= XFS_MOUNT_WSYNC;
 266 #if XFS_BIG_INUMS
 267         if (ap->flags & XFSMNT_INO64) {
 268                 mp->m_flags |= XFS_MOUNT_INO64;
 269                 mp->m_inoadd = XFS_INO64_OFFSET;
 270         }
 271 #endif
 272         if (ap->flags & XFSMNT_RETERR)
 273                 mp->m_flags |= XFS_MOUNT_RETERR;
 274         if (ap->flags & XFSMNT_NOALIGN)
 275                 mp->m_flags |= XFS_MOUNT_NOALIGN;
 276         if (ap->flags & XFSMNT_SWALLOC)
 277                 mp->m_flags |= XFS_MOUNT_SWALLOC;
 278         if (ap->flags & XFSMNT_OSYNCISOSYNC)
 279                 mp->m_flags |= XFS_MOUNT_OSYNCISOSYNC;
 280         if (ap->flags & XFSMNT_32BITINODES)
 281                 mp->m_flags |= XFS_MOUNT_32BITINODES;
 282
 283         if (ap->flags & XFSMNT_IOSIZE) {
 284                 if (ap->iosizelog > XFS_MAX_IO_LOG ||
 285                     ap->iosizelog < XFS_MIN_IO_LOG) {
 286                         cmn_err(CE_WARN,
 287                 "XFS: invalid log iosize: %d [not %d-%d]",
 288                                 ap->iosizelog, XFS_MIN_IO_LOG,
 289                                 XFS_MAX_IO_LOG);
 290                         return XFS_ERROR(EINVAL);
 291                 }
 292
 293                 mp->m_flags |= XFS_MOUNT_DFLT_IOSIZE;
 294                 mp->m_readio_log = mp->m_writeio_log = ap->iosizelog;
 295         }
 296
 297         if (ap->flags & XFSMNT_IDELETE)
 298                 mp->m_flags |= XFS_MOUNT_IDELETE;
 299         if (ap->flags & XFSMNT_DIRSYNC)
 300                 mp->m_flags |= XFS_MOUNT_DIRSYNC;
 301         if (ap->flags & XFSMNT_ATTR2)
 302                 mp->m_flags |= XFS_MOUNT_ATTR2;
 303
 304         if (ap->flags2 & XFSMNT2_COMPAT_IOSIZE)
 305                 mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE;
 306
 307         /*
 308          * no recovery flag requires a read-only mount
 309          */
 310         if (ap->flags & XFSMNT_NORECOVERY) {
 311                 if (!(vfs->vfs_flag & VFS_RDONLY)) {
 312                         cmn_err(CE_WARN,
 313         "XFS: tried to mount a FS read-write without recovery!");
 314                         return XFS_ERROR(EINVAL);
 315                 }
 316                 mp->m_flags |= XFS_MOUNT_NORECOVERY;
 317         }
 318
 319         if (ap->flags & XFSMNT_NOUUID)
 320                 mp->m_flags |= XFS_MOUNT_NOUUID;
 321         if (ap->flags & XFSMNT_BARRIER)
 322                 mp->m_flags |= XFS_MOUNT_BARRIER;
 323         else
 324                 mp->m_flags &= ~XFS_MOUNT_BARRIER;
 325
 326         if (ap->flags2 & XFSMNT2_FILESTREAMS)
 327                 mp->m_flags |= XFS_MOUNT_FILESTREAMS;
 328
 329         if (ap->flags & XFSMNT_DMAPI)
 330                 vfs->vfs_flag |= VFS_DMI;
 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         error = xfs_dmops_get(mp, args);
 436         if (error)
 437                 return error;
 438
 439         /*
 440          * Setup xfs_mount function vectors from available behaviors
 441          */
 442         p = vfs_bhv_lookup(vfsp, VFS_POSITION_QM);
 443         mp->m_qm_ops = p ? *(xfs_qmops_t *) vfs_bhv_custom(p) : xfs_qmcore_stub;
 444         p = vfs_bhv_lookup(vfsp, VFS_POSITION_IO);
 445         mp->m_io_ops = p ? *(xfs_ioops_t *) vfs_bhv_custom(p) : xfs_iocore_xfs;
 446
 447         if (args->flags & XFSMNT_QUIET)
 448                 flags |= XFS_MFSI_QUIET;
 449
 450         /*
 451          * Open real time and log devices - order is important.
 452          */
 453         if (args->logname[0]) {
 454                 error = xfs_blkdev_get(mp, args->logname, &logdev);
 455                 if (error)
 456                         return error;
 457         }
 458         if (args->rtname[0]) {
 459                 error = xfs_blkdev_get(mp, args->rtname, &rtdev);
 460                 if (error) {
 461                         xfs_blkdev_put(logdev);
 462                         return error;
 463                 }
 464
 465                 if (rtdev == ddev || rtdev == logdev) {
 466                         cmn_err(CE_WARN,
 467         "XFS: Cannot mount filesystem with identical rtdev and ddev/logdev.");
 468                         xfs_blkdev_put(logdev);
 469                         xfs_blkdev_put(rtdev);
 470                         return EINVAL;
 471                 }
 472         }
 473
 474         /*
 475          * Setup xfs_mount buffer target pointers
 476          */
 477         error = ENOMEM;
 478         mp->m_ddev_targp = xfs_alloc_buftarg(ddev, 0);
 479         if (!mp->m_ddev_targp) {
 480                 xfs_blkdev_put(logdev);
 481                 xfs_blkdev_put(rtdev);
 482                 return error;
 483         }
 484         if (rtdev) {
 485                 mp->m_rtdev_targp = xfs_alloc_buftarg(rtdev, 1);
 486                 if (!mp->m_rtdev_targp) {
 487                         xfs_blkdev_put(logdev);
 488                         xfs_blkdev_put(rtdev);
 489                         goto error0;
 490                 }
 491         }
 492         mp->m_logdev_targp = (logdev && logdev != ddev) ?
 493                                 xfs_alloc_buftarg(logdev, 1) : mp->m_ddev_targp;
 494         if (!mp->m_logdev_targp) {
 495                 xfs_blkdev_put(logdev);
 496                 xfs_blkdev_put(rtdev);
 497                 goto error0;
 498         }
 499
 500         /*
 501          * Setup flags based on mount(2) options and then the superblock
 502          */
 503         error = xfs_start_flags(vfsp, args, mp);
 504         if (error)
 505                 goto error1;
 506         error = xfs_readsb(mp, flags);
 507         if (error)
 508                 goto error1;
 509         error = xfs_finish_flags(vfsp, args, mp);
 510         if (error)
 511                 goto error2;
 512
 513         /*
 514          * Setup xfs_mount buffer target pointers based on superblock
 515          */
 516         error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_blocksize,
 517                                     mp->m_sb.sb_sectsize);
 518         if (!error && logdev && logdev != ddev) {
 519                 unsigned int    log_sector_size = BBSIZE;
 520
 521                 if (XFS_SB_VERSION_HASSECTOR(&mp->m_sb))
 522                         log_sector_size = mp->m_sb.sb_logsectsize;
 523                 error = xfs_setsize_buftarg(mp->m_logdev_targp,
 524                                             mp->m_sb.sb_blocksize,
 525                                             log_sector_size);
 526         }
 527         if (!error && rtdev)
 528                 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
 529                                             mp->m_sb.sb_blocksize,
 530                                             mp->m_sb.sb_sectsize);
 531         if (error)
 532                 goto error2;
 533
 534         if (mp->m_flags & XFS_MOUNT_BARRIER)
 535                 xfs_mountfs_check_barriers(mp);
 536
 537         if ((error = xfs_filestream_mount(mp)))
 538                 goto error2;
 539
 540         error = XFS_IOINIT(vfsp, args, flags);
 541         if (error)
 542                 goto error2;
 543
 544         XFS_SEND_MOUNT(mp, DM_RIGHT_NULL, args->mtpt, args->fsname);
 545
 546         return 0;
 547
 548 error2:
 549         if (mp->m_sb_bp)
 550                 xfs_freesb(mp);
 551 error1:
 552         xfs_binval(mp->m_ddev_targp);
 553         if (logdev && logdev != ddev)
 554                 xfs_binval(mp->m_logdev_targp);
 555         if (rtdev)
 556                 xfs_binval(mp->m_rtdev_targp);
 557 error0:
 558         xfs_unmountfs_close(mp, credp);
 559         xfs_dmops_put(mp);
 560         return error;
 561 }
 562
 563 STATIC int
 564 xfs_unmount(
 565         bhv_desc_t      *bdp,
 566         int             flags,
 567         cred_t          *credp)
 568 {
 569         bhv_vfs_t       *vfsp = bhvtovfs(bdp);
 570         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
 571         xfs_inode_t     *rip;
 572         bhv_vnode_t     *rvp;
 573         int             unmount_event_wanted = 0;
 574         int             unmount_event_flags = 0;
 575         int             xfs_unmountfs_needed = 0;
 576         int             error;
 577
 578         rip = mp->m_rootip;
 579         rvp = XFS_ITOV(rip);
 580
 581 #ifdef HAVE_DMAPI
 582         if (vfsp->vfs_flag & VFS_DMI) {
 583                 error = XFS_SEND_PREUNMOUNT(mp, vfsp,
 584                                 rvp, DM_RIGHT_NULL, rvp, DM_RIGHT_NULL,
 585                                 NULL, NULL, 0, 0,
 586                                 (mp->m_dmevmask & (1<<DM_EVENT_PREUNMOUNT))?
 587                                         0:DM_FLAGS_UNWANTED);
 588                         if (error)
 589                                 return XFS_ERROR(error);
 590                 unmount_event_wanted = 1;
 591                 unmount_event_flags = (mp->m_dmevmask & (1<<DM_EVENT_UNMOUNT))?
 592                                         0 : DM_FLAGS_UNWANTED;
 593         }
 594 #endif
 595         /*
 596          * First blow any referenced inode from this file system
 597          * out of the reference cache, and delete the timer.
 598          */
 599         xfs_refcache_purge_mp(mp);
 600
 601         /*
 602          * Blow away any referenced inode in the filestreams cache.
 603          * This can and will cause log traffic as inodes go inactive
 604          * here.
 605          */
 606         xfs_filestream_unmount(mp);
 607
 608         XFS_bflush(mp->m_ddev_targp);
 609         error = xfs_unmount_flush(mp, 0);
 610         if (error)
 611                 goto out;
 612
 613         ASSERT(vn_count(rvp) == 1);
 614
 615         /*
 616          * Drop the reference count
 617          */
 618         VN_RELE(rvp);
 619
 620         /*
 621          * If we're forcing a shutdown, typically because of a media error,
 622          * we want to make sure we invalidate dirty pages that belong to
 623          * referenced vnodes as well.
 624          */
 625         if (XFS_FORCED_SHUTDOWN(mp)) {
 626                 error = xfs_sync(&mp->m_bhv,
 627                          (SYNC_WAIT | SYNC_CLOSE), credp);
 628                 ASSERT(error != EFSCORRUPTED);
 629         }
 630         xfs_unmountfs_needed = 1;
 631
 632 out:
 633         /*      Send DMAPI event, if required.
 634          *      Then do xfs_unmountfs() if needed.
 635          *      Then return error (or zero).
 636          */
 637         if (unmount_event_wanted) {
 638                 /* Note: mp structure must still exist for
 639                  * XFS_SEND_UNMOUNT() call.
 640                  */
 641                 XFS_SEND_UNMOUNT(mp, vfsp, error == 0 ? rvp : NULL,
 642                         DM_RIGHT_NULL, 0, error, unmount_event_flags);
 643         }
 644         if (xfs_unmountfs_needed) {
 645                 /*
 646                  * Call common unmount function to flush to disk
 647                  * and free the super block buffer & mount structures.
 648                  */
 649                 xfs_unmountfs(mp, credp);
 650                 xfs_dmops_put(mp);
 651                 kmem_free(mp, sizeof(xfs_mount_t));
 652         }
 653
 654         return XFS_ERROR(error);
 655 }
 656
 657 STATIC int
 658 xfs_quiesce_fs(
 659         xfs_mount_t             *mp)
 660 {
 661         int                     count = 0, pincount;
 662
 663         xfs_refcache_purge_mp(mp);
 664         xfs_flush_buftarg(mp->m_ddev_targp, 0);
 665         xfs_finish_reclaim_all(mp, 0);
 666
 667         /* This loop must run at least twice.
 668          * The first instance of the loop will flush
 669          * most meta data but that will generate more
 670          * meta data (typically directory updates).
 671          * Which then must be flushed and logged before
 672          * we can write the unmount record.
 673          */
 674         do {
 675                 xfs_syncsub(mp, SYNC_INODE_QUIESCE, NULL);
 676                 pincount = xfs_flush_buftarg(mp->m_ddev_targp, 1);
 677                 if (!pincount) {
 678                         delay(50);
 679                         count++;
 680                 }
 681         } while (count < 2);
 682
 683         return 0;
 684 }
 685
 686 /*
 687  * Second stage of a quiesce. The data is already synced, now we have to take
 688  * care of the metadata. New transactions are already blocked, so we need to
 689  * wait for any remaining transactions to drain out before proceding.
 690  */
 691 STATIC void
 692 xfs_attr_quiesce(
 693         xfs_mount_t     *mp)
 694 {
 695         /* wait for all modifications to complete */
 696         while (atomic_read(&mp->m_active_trans) > 0)
 697                 delay(100);
 698
 699         /* flush inodes and push all remaining buffers out to disk */
 700         xfs_quiesce_fs(mp);
 701
 702         ASSERT_ALWAYS(atomic_read(&mp->m_active_trans) == 0);
 703
 704         /* Push the superblock and write an unmount record */
 705         xfs_log_sbcount(mp, 1);
 706         xfs_log_unmount_write(mp);
 707         xfs_unmountfs_writesb(mp);
 708 }
 709
 710 STATIC int
 711 xfs_mntupdate(
 712         bhv_desc_t                      *bdp,
 713         int                             *flags,
 714         struct xfs_mount_args           *args)
 715 {
 716         bhv_vfs_t       *vfsp = bhvtovfs(bdp);
 717         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
 718
 719         if (!(*flags & MS_RDONLY)) {                    /* rw/ro -> rw */
 720                 if (vfsp->vfs_flag & VFS_RDONLY)
 721                         vfsp->vfs_flag &= ~VFS_RDONLY;
 722                 if (args->flags & XFSMNT_BARRIER) {
 723                         mp->m_flags |= XFS_MOUNT_BARRIER;
 724                         xfs_mountfs_check_barriers(mp);
 725                 } else {
 726                         mp->m_flags &= ~XFS_MOUNT_BARRIER;
 727                 }
 728         } else if (!(vfsp->vfs_flag & VFS_RDONLY)) {    /* rw -> ro */
 729                 xfs_filestream_flush(mp);
 730                 bhv_vfs_sync(vfsp, SYNC_DATA_QUIESCE, NULL);
 731                 xfs_attr_quiesce(mp);
 732                 vfsp->vfs_flag |= VFS_RDONLY;
 733         }
 734         return 0;
 735 }
 736
 737 /*
 738  * xfs_unmount_flush implements a set of flush operation on special
 739  * inodes, which are needed as a separate set of operations so that
 740  * they can be called as part of relocation process.
 741  */
 742 int
 743 xfs_unmount_flush(
 744         xfs_mount_t     *mp,            /* Mount structure we are getting
 745                                            rid of. */
 746         int             relocation)     /* Called from vfs relocation. */
 747 {
 748         xfs_inode_t     *rip = mp->m_rootip;
 749         xfs_inode_t     *rbmip;
 750         xfs_inode_t     *rsumip = NULL;
 751         bhv_vnode_t     *rvp = XFS_ITOV(rip);
 752         int             error;
 753
 754         xfs_ilock(rip, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
 755         xfs_iflock(rip);
 756
 757         /*
 758          * Flush out the real time inodes.
 759          */
 760         if ((rbmip = mp->m_rbmip) != NULL) {
 761                 xfs_ilock(rbmip, XFS_ILOCK_EXCL);
 762                 xfs_iflock(rbmip);
 763                 error = xfs_iflush(rbmip, XFS_IFLUSH_SYNC);
 764                 xfs_iunlock(rbmip, XFS_ILOCK_EXCL);
 765
 766                 if (error == EFSCORRUPTED)
 767                         goto fscorrupt_out;
 768
 769                 ASSERT(vn_count(XFS_ITOV(rbmip)) == 1);
 770
 771                 rsumip = mp->m_rsumip;
 772                 xfs_ilock(rsumip, XFS_ILOCK_EXCL);
 773                 xfs_iflock(rsumip);
 774                 error = xfs_iflush(rsumip, XFS_IFLUSH_SYNC);
 775                 xfs_iunlock(rsumip, XFS_ILOCK_EXCL);
 776
 777                 if (error == EFSCORRUPTED)
 778                         goto fscorrupt_out;
 779
 780                 ASSERT(vn_count(XFS_ITOV(rsumip)) == 1);
 781         }
 782
 783         /*
 784          * Synchronously flush root inode to disk
 785          */
 786         error = xfs_iflush(rip, XFS_IFLUSH_SYNC);
 787         if (error == EFSCORRUPTED)
 788                 goto fscorrupt_out2;
 789
 790         if (vn_count(rvp) != 1 && !relocation) {
 791                 xfs_iunlock(rip, XFS_ILOCK_EXCL);
 792                 return XFS_ERROR(EBUSY);
 793         }
 794
 795         /*
 796          * Release dquot that rootinode, rbmino and rsumino might be holding,
 797          * flush and purge the quota inodes.
 798          */
 799         error = XFS_QM_UNMOUNT(mp);
 800         if (error == EFSCORRUPTED)
 801                 goto fscorrupt_out2;
 802
 803         if (rbmip) {
 804                 VN_RELE(XFS_ITOV(rbmip));
 805                 VN_RELE(XFS_ITOV(rsumip));
 806         }
 807
 808         xfs_iunlock(rip, XFS_ILOCK_EXCL);
 809         return 0;
 810
 811 fscorrupt_out:
 812         xfs_ifunlock(rip);
 813
 814 fscorrupt_out2:
 815         xfs_iunlock(rip, XFS_ILOCK_EXCL);
 816
 817         return XFS_ERROR(EFSCORRUPTED);
 818 }
 819
 820 /*
 821  * xfs_root extracts the root vnode from a vfs.
 822  *
 823  * vfsp -- the vfs struct for the desired file system
 824  * vpp  -- address of the caller's vnode pointer which should be
 825  *         set to the desired fs root vnode
 826  */
 827 STATIC int
 828 xfs_root(
 829         bhv_desc_t      *bdp,
 830         bhv_vnode_t     **vpp)
 831 {
 832         bhv_vnode_t     *vp;
 833
 834         vp = XFS_ITOV((XFS_BHVTOM(bdp))->m_rootip);
 835         VN_HOLD(vp);
 836         *vpp = vp;
 837         return 0;
 838 }
 839
 840 /*
 841  * xfs_statvfs
 842  *
 843  * Fill in the statvfs structure for the given file system.  We use
 844  * the superblock lock in the mount structure to ensure a consistent
 845  * snapshot of the counters returned.
 846  */
 847 STATIC int
 848 xfs_statvfs(
 849         bhv_desc_t      *bdp,
 850         bhv_statvfs_t   *statp,
 851         bhv_vnode_t     *vp)
 852 {
 853         __uint64_t      fakeinos;
 854         xfs_extlen_t    lsize;
 855         xfs_mount_t     *mp;
 856         xfs_sb_t        *sbp;
 857         unsigned long   s;
 858
 859         mp = XFS_BHVTOM(bdp);
 860         sbp = &(mp->m_sb);
 861
 862         statp->f_type = XFS_SB_MAGIC;
 863
 864         xfs_icsb_sync_counters_flags(mp, XFS_ICSB_LAZY_COUNT);
 865         s = XFS_SB_LOCK(mp);
 866         statp->f_bsize = sbp->sb_blocksize;
 867         lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
 868         statp->f_blocks = sbp->sb_dblocks - lsize;
 869         statp->f_bfree = statp->f_bavail =
 870                                 sbp->sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
 871         fakeinos = statp->f_bfree << sbp->sb_inopblog;
 872 #if XFS_BIG_INUMS
 873         fakeinos += mp->m_inoadd;
 874 #endif
 875         statp->f_files =
 876             MIN(sbp->sb_icount + fakeinos, (__uint64_t)XFS_MAXINUMBER);
 877         if (mp->m_maxicount)
 878 #if XFS_BIG_INUMS
 879                 if (!mp->m_inoadd)
 880 #endif
 881                         statp->f_files = min_t(typeof(statp->f_files),
 882                                                 statp->f_files,
 883                                                 mp->m_maxicount);
 884         statp->f_ffree = statp->f_files - (sbp->sb_icount - sbp->sb_ifree);
 885         XFS_SB_UNLOCK(mp, s);
 886
 887         xfs_statvfs_fsid(statp, mp);
 888         statp->f_namelen = MAXNAMELEN - 1;
 889
 890         return 0;
 891 }
 892
 893
 894 /*
 895  * xfs_sync flushes any pending I/O to file system vfsp.
 896  *
 897  * This routine is called by vfs_sync() to make sure that things make it
 898  * out to disk eventually, on sync() system calls to flush out everything,
 899  * and when the file system is unmounted.  For the vfs_sync() case, all
 900  * we really need to do is sync out the log to make all of our meta-data
 901  * updates permanent (except for timestamps).  For calls from pflushd(),
 902  * dirty pages are kept moving by calling pdflush() on the inodes
 903  * containing them.  We also flush the inodes that we can lock without
 904  * sleeping and the superblock if we can lock it without sleeping from
 905  * vfs_sync() so that items at the tail of the log are always moving out.
 906  *
 907  * Flags:
 908  *      SYNC_BDFLUSH - We're being called from vfs_sync() so we don't want
 909  *                     to sleep if we can help it.  All we really need
 910  *                     to do is ensure that the log is synced at least
 911  *                     periodically.  We also push the inodes and
 912  *                     superblock if we can lock them without sleeping
 913  *                      and they are not pinned.
 914  *      SYNC_ATTR    - We need to flush the inodes.  If SYNC_BDFLUSH is not
 915  *                     set, then we really want to lock each inode and flush
 916  *                     it.
 917  *      SYNC_WAIT    - All the flushes that take place in this call should
 918  *                     be synchronous.
 919  *      SYNC_DELWRI  - This tells us to push dirty pages associated with
 920  *                     inodes.  SYNC_WAIT and SYNC_BDFLUSH are used to
 921  *                     determine if they should be flushed sync, async, or
 922  *                     delwri.
 923  *      SYNC_CLOSE   - This flag is passed when the system is being
 924  *                     unmounted.  We should sync and invalidate everything.
 925  *      SYNC_FSDATA  - This indicates that the caller would like to make
 926  *                     sure the superblock is safe on disk.  We can ensure
 927  *                     this by simply making sure the log gets flushed
 928  *                     if SYNC_BDFLUSH is set, and by actually writing it
 929  *                     out otherwise.
 930  *      SYNC_IOWAIT  - The caller wants us to wait for all data I/O to complete
 931  *                     before we return (including direct I/O). Forms the drain
 932  *                     side of the write barrier needed to safely quiesce the
 933  *                     filesystem.
 934  *
 935  */
 936 /*ARGSUSED*/
 937 STATIC int
 938 xfs_sync(
 939         bhv_desc_t      *bdp,
 940         int             flags,
 941         cred_t          *credp)
 942 {
 943         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
 944
 945         if (flags & SYNC_IOWAIT)
 946                 xfs_filestream_flush(mp);
 947
 948         return xfs_syncsub(mp, flags, NULL);
 949 }
 950
 951 /*
 952  * xfs sync routine for internal use
 953  *
 954  * This routine supports all of the flags defined for the generic vfs_sync
 955  * interface as explained above under xfs_sync.
 956  *
 957  */
 958 int
 959 xfs_sync_inodes(
 960         xfs_mount_t     *mp,
 961         int             flags,
 962         int             *bypassed)
 963 {
 964         xfs_inode_t     *ip = NULL;
 965         xfs_inode_t     *ip_next;
 966         xfs_buf_t       *bp;
 967         bhv_vnode_t     *vp = NULL;
 968         int             error;
 969         int             last_error;
 970         uint64_t        fflag;
 971         uint            lock_flags;
 972         uint            base_lock_flags;
 973         boolean_t       mount_locked;
 974         boolean_t       vnode_refed;
 975         int             preempt;
 976         xfs_dinode_t    *dip;
 977         xfs_iptr_t      *ipointer;
 978 #ifdef DEBUG
 979         boolean_t       ipointer_in = B_FALSE;
 980
 981 #define IPOINTER_SET    ipointer_in = B_TRUE
 982 #define IPOINTER_CLR    ipointer_in = B_FALSE
 983 #else
 984 #define IPOINTER_SET
 985 #define IPOINTER_CLR
 986 #endif
 987
 988
 989 /* Insert a marker record into the inode list after inode ip. The list
 990  * must be locked when this is called. After the call the list will no
 991  * longer be locked.
 992  */
 993 #define IPOINTER_INSERT(ip, mp) { \
 994                 ASSERT(ipointer_in == B_FALSE); \
 995                 ipointer->ip_mnext = ip->i_mnext; \
 996                 ipointer->ip_mprev = ip; \
 997                 ip->i_mnext = (xfs_inode_t *)ipointer; \
 998                 ipointer->ip_mnext->i_mprev = (xfs_inode_t *)ipointer; \
 999                 preempt = 0; \
1000                 XFS_MOUNT_IUNLOCK(mp); \
1001                 mount_locked = B_FALSE; \
1002                 IPOINTER_SET; \
1003         }
1004
1005 /* Remove the marker from the inode list. If the marker was the only item
1006  * in the list then there are no remaining inodes and we should zero out
1007  * the whole list. If we are the current head of the list then move the head
1008  * past us.
1009  */
1010 #define IPOINTER_REMOVE(ip, mp) { \
1011                 ASSERT(ipointer_in == B_TRUE); \
1012                 if (ipointer->ip_mnext != (xfs_inode_t *)ipointer) { \
1013                         ip = ipointer->ip_mnext; \
1014                         ip->i_mprev = ipointer->ip_mprev; \
1015                         ipointer->ip_mprev->i_mnext = ip; \
1016                         if (mp->m_inodes == (xfs_inode_t *)ipointer) { \
1017                                 mp->m_inodes = ip; \
1018                         } \
1019                 } else { \
1020                         ASSERT(mp->m_inodes == (xfs_inode_t *)ipointer); \
1021                         mp->m_inodes = NULL; \
1022                         ip = NULL; \
1023                 } \
1024                 IPOINTER_CLR; \
1025         }
1026
1027 #define XFS_PREEMPT_MASK        0x7f
1028
1029         if (bypassed)
1030                 *bypassed = 0;
1031         if (XFS_MTOVFS(mp)->vfs_flag & VFS_RDONLY)
1032                 return 0;
1033         error = 0;
1034         last_error = 0;
1035         preempt = 0;
1036
1037         /* Allocate a reference marker */
1038         ipointer = (xfs_iptr_t *)kmem_zalloc(sizeof(xfs_iptr_t), KM_SLEEP);
1039
1040         fflag = XFS_B_ASYNC;            /* default is don't wait */
1041         if (flags & (SYNC_BDFLUSH | SYNC_DELWRI))
1042                 fflag = XFS_B_DELWRI;
1043         if (flags & SYNC_WAIT)
1044                 fflag = 0;              /* synchronous overrides all */
1045
1046         base_lock_flags = XFS_ILOCK_SHARED;
1047         if (flags & (SYNC_DELWRI | SYNC_CLOSE)) {
1048                 /*
1049                  * We need the I/O lock if we're going to call any of
1050                  * the flush/inval routines.
1051                  */
1052                 base_lock_flags |= XFS_IOLOCK_SHARED;
1053         }
1054
1055         XFS_MOUNT_ILOCK(mp);
1056
1057         ip = mp->m_inodes;
1058
1059         mount_locked = B_TRUE;
1060         vnode_refed  = B_FALSE;
1061
1062         IPOINTER_CLR;
1063
1064         do {
1065                 ASSERT(ipointer_in == B_FALSE);
1066                 ASSERT(vnode_refed == B_FALSE);
1067
1068                 lock_flags = base_lock_flags;
1069
1070                 /*
1071                  * There were no inodes in the list, just break out
1072                  * of the loop.
1073                  */
1074                 if (ip == NULL) {
1075                         break;
1076                 }
1077
1078                 /*
1079                  * We found another sync thread marker - skip it
1080                  */
1081                 if (ip->i_mount == NULL) {
1082                         ip = ip->i_mnext;
1083                         continue;
1084                 }
1085
1086                 vp = XFS_ITOV_NULL(ip);
1087
1088                 /*
1089                  * If the vnode is gone then this is being torn down,
1090                  * call reclaim if it is flushed, else let regular flush
1091                  * code deal with it later in the loop.
1092                  */
1093
1094                 if (vp == NULL) {
1095                         /* Skip ones already in reclaim */
1096                         if (ip->i_flags & XFS_IRECLAIM) {
1097                                 ip = ip->i_mnext;
1098                                 continue;
1099                         }
1100                         if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0) {
1101                                 ip = ip->i_mnext;
1102                         } else if ((xfs_ipincount(ip) == 0) &&
1103                                     xfs_iflock_nowait(ip)) {
1104                                 IPOINTER_INSERT(ip, mp);
1105
1106                                 xfs_finish_reclaim(ip, 1,
1107                                                 XFS_IFLUSH_DELWRI_ELSE_ASYNC);
1108
1109                                 XFS_MOUNT_ILOCK(mp);
1110                                 mount_locked = B_TRUE;
1111                                 IPOINTER_REMOVE(ip, mp);
1112                         } else {
1113                                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1114                                 ip = ip->i_mnext;
1115                         }
1116                         continue;
1117                 }
1118
1119                 if (VN_BAD(vp)) {
1120                         ip = ip->i_mnext;
1121                         continue;
1122                 }
1123
1124                 if (XFS_FORCED_SHUTDOWN(mp) && !(flags & SYNC_CLOSE)) {
1125                         XFS_MOUNT_IUNLOCK(mp);
1126                         kmem_free(ipointer, sizeof(xfs_iptr_t));
1127                         return 0;
1128                 }
1129
1130                 /*
1131                  * If this is just vfs_sync() or pflushd() calling
1132                  * then we can skip inodes for which it looks like
1133                  * there is nothing to do.  Since we don't have the
1134                  * inode locked this is racy, but these are periodic
1135                  * calls so it doesn't matter.  For the others we want
1136                  * to know for sure, so we at least try to lock them.
1137                  */
1138                 if (flags & SYNC_BDFLUSH) {
1139                         if (((ip->i_itemp == NULL) ||
1140                              !(ip->i_itemp->ili_format.ilf_fields &
1141                                XFS_ILOG_ALL)) &&
1142                             (ip->i_update_core == 0)) {
1143                                 ip = ip->i_mnext;
1144                                 continue;
1145                         }
1146                 }
1147
1148                 /*
1149                  * Try to lock without sleeping.  We're out of order with
1150                  * the inode list lock here, so if we fail we need to drop
1151                  * the mount lock and try again.  If we're called from
1152                  * bdflush() here, then don't bother.
1153                  *
1154                  * The inode lock here actually coordinates with the
1155                  * almost spurious inode lock in xfs_ireclaim() to prevent
1156                  * the vnode we handle here without a reference from
1157                  * being freed while we reference it.  If we lock the inode
1158                  * while it's on the mount list here, then the spurious inode
1159                  * lock in xfs_ireclaim() after the inode is pulled from
1160                  * the mount list will sleep until we release it here.
1161                  * This keeps the vnode from being freed while we reference
1162                  * it.
1163                  */
1164                 if (xfs_ilock_nowait(ip, lock_flags) == 0) {
1165                         if ((flags & SYNC_BDFLUSH) || (vp == NULL)) {
1166                                 ip = ip->i_mnext;
1167                                 continue;
1168                         }
1169
1170                         vp = vn_grab(vp);
1171                         if (vp == NULL) {
1172                                 ip = ip->i_mnext;
1173                                 continue;
1174                         }
1175
1176                         IPOINTER_INSERT(ip, mp);
1177                         xfs_ilock(ip, lock_flags);
1178
1179                         ASSERT(vp == XFS_ITOV(ip));
1180                         ASSERT(ip->i_mount == mp);
1181
1182                         vnode_refed = B_TRUE;
1183                 }
1184
1185                 /* From here on in the loop we may have a marker record
1186                  * in the inode list.
1187                  */
1188
1189                 /*
1190                  * If we have to flush data or wait for I/O completion
1191                  * we need to drop the ilock that we currently hold.
1192                  * If we need to drop the lock, insert a marker if we
1193                  * have not already done so.
1194                  */
1195                 if ((flags & (SYNC_CLOSE|SYNC_IOWAIT)) ||
1196                     ((flags & SYNC_DELWRI) && VN_DIRTY(vp))) {
1197                         if (mount_locked) {
1198                                 IPOINTER_INSERT(ip, mp);
1199                         }
1200                         xfs_iunlock(ip, XFS_ILOCK_SHARED);
1201
1202                         if (flags & SYNC_CLOSE) {
1203                                 /* Shutdown case. Flush and invalidate. */
1204                                 if (XFS_FORCED_SHUTDOWN(mp))
1205                                         xfs_tosspages(ip, 0, -1,
1206                                                              FI_REMAPF);
1207                                 else
1208                                         error = xfs_flushinval_pages(ip,
1209                                                         0, -1, FI_REMAPF);
1210                         } else if ((flags & SYNC_DELWRI) && VN_DIRTY(vp)) {
1211                                 error = xfs_flush_pages(ip, 0,
1212                                                         -1, fflag, FI_NONE);
1213                         }
1214
1215                         /*
1216                          * When freezing, we need to wait ensure all I/O (including direct
1217                          * I/O) is complete to ensure no further data modification can take
1218                          * place after this point
1219                          */
1220                         if (flags & SYNC_IOWAIT)
1221                                 vn_iowait(ip);
1222
1223                         xfs_ilock(ip, XFS_ILOCK_SHARED);
1224                 }
1225
1226                 if (flags & SYNC_BDFLUSH) {
1227                         if ((flags & SYNC_ATTR) &&
1228                             ((ip->i_update_core) ||
1229                              ((ip->i_itemp != NULL) &&
1230                               (ip->i_itemp->ili_format.ilf_fields != 0)))) {
1231
1232                                 /* Insert marker and drop lock if not already
1233                                  * done.
1234                                  */
1235                                 if (mount_locked) {
1236                                         IPOINTER_INSERT(ip, mp);
1237                                 }
1238
1239                                 /*
1240                                  * We don't want the periodic flushing of the
1241                                  * inodes by vfs_sync() to interfere with
1242                                  * I/O to the file, especially read I/O
1243                                  * where it is only the access time stamp
1244                                  * that is being flushed out.  To prevent
1245                                  * long periods where we have both inode
1246                                  * locks held shared here while reading the
1247                                  * inode's buffer in from disk, we drop the
1248                                  * inode lock while reading in the inode
1249                                  * buffer.  We have to release the buffer
1250                                  * and reacquire the inode lock so that they
1251                                  * are acquired in the proper order (inode
1252                                  * locks first).  The buffer will go at the
1253                                  * end of the lru chain, though, so we can
1254                                  * expect it to still be there when we go
1255                                  * for it again in xfs_iflush().
1256                                  */
1257                                 if ((xfs_ipincount(ip) == 0) &&
1258                                     xfs_iflock_nowait(ip)) {
1259
1260                                         xfs_ifunlock(ip);
1261                                         xfs_iunlock(ip, XFS_ILOCK_SHARED);
1262
1263                                         error = xfs_itobp(mp, NULL, ip,
1264                                                           &dip, &bp, 0, 0);
1265                                         if (!error) {
1266                                                 xfs_buf_relse(bp);
1267                                         } else {
1268                                                 /* Bailing out, remove the
1269                                                  * marker and free it.
1270                                                  */
1271                                                 XFS_MOUNT_ILOCK(mp);
1272                                                 IPOINTER_REMOVE(ip, mp);
1273                                                 XFS_MOUNT_IUNLOCK(mp);
1274
1275                                                 ASSERT(!(lock_flags &
1276                                                         XFS_IOLOCK_SHARED));
1277
1278                                                 kmem_free(ipointer,
1279                                                         sizeof(xfs_iptr_t));
1280                                                 return (0);
1281                                         }
1282
1283                                         /*
1284                                          * Since we dropped the inode lock,
1285                                          * the inode may have been reclaimed.
1286                                          * Therefore, we reacquire the mount
1287                                          * lock and check to see if we were the
1288                                          * inode reclaimed. If this happened
1289                                          * then the ipointer marker will no
1290                                          * longer point back at us. In this
1291                                          * case, move ip along to the inode
1292                                          * after the marker, remove the marker
1293                                          * and continue.
1294                                          */
1295                                         XFS_MOUNT_ILOCK(mp);
1296                                         mount_locked = B_TRUE;
1297
1298                                         if (ip != ipointer->ip_mprev) {
1299                                                 IPOINTER_REMOVE(ip, mp);
1300
1301                                                 ASSERT(!vnode_refed);
1302                                                 ASSERT(!(lock_flags &
1303                                                         XFS_IOLOCK_SHARED));
1304                                                 continue;
1305                                         }
1306
1307                                         ASSERT(ip->i_mount == mp);
1308
1309                                         if (xfs_ilock_nowait(ip,
1310                                                     XFS_ILOCK_SHARED) == 0) {
1311                                                 ASSERT(ip->i_mount == mp);
1312                                                 /*
1313                                                  * We failed to reacquire
1314                                                  * the inode lock without
1315                                                  * sleeping, so just skip
1316                                                  * the inode for now.  We
1317                                                  * clear the ILOCK bit from
1318                                                  * the lock_flags so that we
1319                                                  * won't try to drop a lock
1320                                                  * we don't hold below.
1321                                                  */
1322                                                 lock_flags &= ~XFS_ILOCK_SHARED;
1323                                                 IPOINTER_REMOVE(ip_next, mp);
1324                                         } else if ((xfs_ipincount(ip) == 0) &&
1325                                                    xfs_iflock_nowait(ip)) {
1326                                                 ASSERT(ip->i_mount == mp);
1327                                                 /*
1328                                                  * Since this is vfs_sync()
1329                                                  * calling we only flush the
1330                                                  * inode out if we can lock
1331                                                  * it without sleeping and
1332                                                  * it is not pinned.  Drop
1333                                                  * the mount lock here so
1334                                                  * that we don't hold it for
1335                                                  * too long. We already have
1336                                                  * a marker in the list here.
1337                                                  */
1338                                                 XFS_MOUNT_IUNLOCK(mp);
1339                                                 mount_locked = B_FALSE;
1340                                                 error = xfs_iflush(ip,
1341                                                            XFS_IFLUSH_DELWRI);
1342                                         } else {
1343                                                 ASSERT(ip->i_mount == mp);
1344                                                 IPOINTER_REMOVE(ip_next, mp);
1345                                         }
1346                                 }
1347
1348                         }
1349
1350                 } else {
1351                         if ((flags & SYNC_ATTR) &&
1352                             ((ip->i_update_core) ||
1353                              ((ip->i_itemp != NULL) &&
1354                               (ip->i_itemp->ili_format.ilf_fields != 0)))) {
1355                                 if (mount_locked) {
1356                                         IPOINTER_INSERT(ip, mp);
1357                                 }
1358
1359                                 if (flags & SYNC_WAIT) {
1360                                         xfs_iflock(ip);
1361                                         error = xfs_iflush(ip,
1362                                                            XFS_IFLUSH_SYNC);
1363                                 } else {
1364                                         /*
1365                                          * If we can't acquire the flush
1366                                          * lock, then the inode is already
1367                                          * being flushed so don't bother
1368                                          * waiting.  If we can lock it then
1369                                          * do a delwri flush so we can
1370                                          * combine multiple inode flushes
1371                                          * in each disk write.
1372                                          */
1373                                         if (xfs_iflock_nowait(ip)) {
1374                                                 error = xfs_iflush(ip,
1375                                                            XFS_IFLUSH_DELWRI);
1376                                         }
1377                                         else if (bypassed)
1378                                                 (*bypassed)++;
1379                                 }
1380                         }
1381                 }
1382
1383                 if (lock_flags != 0) {
1384                         xfs_iunlock(ip, lock_flags);
1385                 }
1386
1387                 if (vnode_refed) {
1388                         /*
1389                          * If we had to take a reference on the vnode
1390                          * above, then wait until after we've unlocked
1391                          * the inode to release the reference.  This is
1392                          * because we can be already holding the inode
1393                          * lock when VN_RELE() calls xfs_inactive().
1394                          *
1395                          * Make sure to drop the mount lock before calling
1396                          * VN_RELE() so that we don't trip over ourselves if
1397                          * we have to go for the mount lock again in the
1398                          * inactive code.
1399                          */
1400                         if (mount_locked) {
1401                                 IPOINTER_INSERT(ip, mp);
1402                         }
1403
1404                         VN_RELE(vp);
1405
1406                         vnode_refed = B_FALSE;
1407                 }
1408
1409                 if (error) {
1410                         last_error = error;
1411                 }
1412
1413                 /*
1414                  * bail out if the filesystem is corrupted.
1415                  */
1416                 if (error == EFSCORRUPTED)  {
1417                         if (!mount_locked) {
1418                                 XFS_MOUNT_ILOCK(mp);
1419                                 IPOINTER_REMOVE(ip, mp);
1420                         }
1421                         XFS_MOUNT_IUNLOCK(mp);
1422                         ASSERT(ipointer_in == B_FALSE);
1423                         kmem_free(ipointer, sizeof(xfs_iptr_t));
1424                         return XFS_ERROR(error);
1425                 }
1426
1427                 /* Let other threads have a chance at the mount lock
1428                  * if we have looped many times without dropping the
1429                  * lock.
1430                  */
1431                 if ((++preempt & XFS_PREEMPT_MASK) == 0) {
1432                         if (mount_locked) {
1433                                 IPOINTER_INSERT(ip, mp);
1434                         }
1435                 }
1436
1437                 if (mount_locked == B_FALSE) {
1438                         XFS_MOUNT_ILOCK(mp);
1439                         mount_locked = B_TRUE;
1440                         IPOINTER_REMOVE(ip, mp);
1441                         continue;
1442                 }
1443
1444                 ASSERT(ipointer_in == B_FALSE);
1445                 ip = ip->i_mnext;
1446
1447         } while (ip != mp->m_inodes);
1448
1449         XFS_MOUNT_IUNLOCK(mp);
1450
1451         ASSERT(ipointer_in == B_FALSE);
1452
1453         kmem_free(ipointer, sizeof(xfs_iptr_t));
1454         return XFS_ERROR(last_error);
1455 }
1456
1457 /*
1458  * xfs sync routine for internal use
1459  *
1460  * This routine supports all of the flags defined for the generic vfs_sync
1461  * interface as explained above under xfs_sync.
1462  *
1463  */
1464 int
1465 xfs_syncsub(
1466         xfs_mount_t     *mp,
1467         int             flags,
1468         int             *bypassed)
1469 {
1470         int             error = 0;
1471         int             last_error = 0;
1472         uint            log_flags = XFS_LOG_FORCE;
1473         xfs_buf_t       *bp;
1474         xfs_buf_log_item_t      *bip;
1475
1476         /*
1477          * Sync out the log.  This ensures that the log is periodically
1478          * flushed even if there is not enough activity to fill it up.
1479          */
1480         if (flags & SYNC_WAIT)
1481                 log_flags |= XFS_LOG_SYNC;
1482
1483         xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1484
1485         if (flags & (SYNC_ATTR|SYNC_DELWRI)) {
1486                 if (flags & SYNC_BDFLUSH)
1487                         xfs_finish_reclaim_all(mp, 1);
1488                 else
1489                         error = xfs_sync_inodes(mp, flags, bypassed);
1490         }
1491
1492         /*
1493          * Flushing out dirty data above probably generated more
1494          * log activity, so if this isn't vfs_sync() then flush
1495          * the log again.
1496          */
1497         if (flags & SYNC_DELWRI) {
1498                 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1499         }
1500
1501         if (flags & SYNC_FSDATA) {
1502                 /*
1503                  * If this is vfs_sync() then only sync the superblock
1504                  * if we can lock it without sleeping and it is not pinned.
1505                  */
1506                 if (flags & SYNC_BDFLUSH) {
1507                         bp = xfs_getsb(mp, XFS_BUF_TRYLOCK);
1508                         if (bp != NULL) {
1509                                 bip = XFS_BUF_FSPRIVATE(bp,xfs_buf_log_item_t*);
1510                                 if ((bip != NULL) &&
1511                                     xfs_buf_item_dirty(bip)) {
1512                                         if (!(XFS_BUF_ISPINNED(bp))) {
1513                                                 XFS_BUF_ASYNC(bp);
1514                                                 error = xfs_bwrite(mp, bp);
1515                                         } else {
1516                                                 xfs_buf_relse(bp);
1517                                         }
1518                                 } else {
1519                                         xfs_buf_relse(bp);
1520                                 }
1521                         }
1522                 } else {
1523                         bp = xfs_getsb(mp, 0);
1524                         /*
1525                          * If the buffer is pinned then push on the log so
1526                          * we won't get stuck waiting in the write for
1527                          * someone, maybe ourselves, to flush the log.
1528                          * Even though we just pushed the log above, we
1529                          * did not have the superblock buffer locked at
1530                          * that point so it can become pinned in between
1531                          * there and here.
1532                          */
1533                         if (XFS_BUF_ISPINNED(bp))
1534                                 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
1535                         if (flags & SYNC_WAIT)
1536                                 XFS_BUF_UNASYNC(bp);
1537                         else
1538                                 XFS_BUF_ASYNC(bp);
1539                         error = xfs_bwrite(mp, bp);
1540                 }
1541                 if (error) {
1542                         last_error = error;
1543                 }
1544         }
1545
1546         /*
1547          * If this is the periodic sync, then kick some entries out of
1548          * the reference cache.  This ensures that idle entries are
1549          * eventually kicked out of the cache.
1550          */
1551         if (flags & SYNC_REFCACHE) {
1552                 if (flags & SYNC_WAIT)
1553                         xfs_refcache_purge_mp(mp);
1554                 else
1555                         xfs_refcache_purge_some(mp);
1556         }
1557
1558         /*
1559          * If asked, update the disk superblock with incore counter values if we
1560          * are using non-persistent counters so that they don't get too far out
1561          * of sync if we crash or get a forced shutdown. We don't want to force
1562          * this to disk, just get a transaction into the iclogs....
1563          */
1564         if (flags & SYNC_SUPER)
1565                 xfs_log_sbcount(mp, 0);
1566
1567         /*
1568          * Now check to see if the log needs a "dummy" transaction.
1569          */
1570
1571         if (!(flags & SYNC_REMOUNT) && xfs_log_need_covered(mp)) {
1572                 xfs_trans_t *tp;
1573                 xfs_inode_t *ip;
1574
1575                 /*
1576                  * Put a dummy transaction in the log to tell
1577                  * recovery that all others are OK.
1578                  */
1579                 tp = xfs_trans_alloc(mp, XFS_TRANS_DUMMY1);
1580                 if ((error = xfs_trans_reserve(tp, 0,
1581                                 XFS_ICHANGE_LOG_RES(mp),
1582                                 0, 0, 0)))  {
1583                         xfs_trans_cancel(tp, 0);
1584                         return error;
1585                 }
1586
1587                 ip = mp->m_rootip;
1588                 xfs_ilock(ip, XFS_ILOCK_EXCL);
1589
1590                 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1591                 xfs_trans_ihold(tp, ip);
1592                 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1593                 error = xfs_trans_commit(tp, 0);
1594                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1595                 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1596         }
1597
1598         /*
1599          * When shutting down, we need to insure that the AIL is pushed
1600          * to disk or the filesystem can appear corrupt from the PROM.
1601          */
1602         if ((flags & (SYNC_CLOSE|SYNC_WAIT)) == (SYNC_CLOSE|SYNC_WAIT)) {
1603                 XFS_bflush(mp->m_ddev_targp);
1604                 if (mp->m_rtdev_targp) {
1605                         XFS_bflush(mp->m_rtdev_targp);
1606                 }
1607         }
1608
1609         return XFS_ERROR(last_error);
1610 }
1611
1612 /*
1613  * xfs_vget - called by DMAPI and NFSD to get vnode from file handle
1614  */
1615 STATIC int
1616 xfs_vget(
1617         bhv_desc_t      *bdp,
1618         bhv_vnode_t     **vpp,
1619         fid_t           *fidp)
1620 {
1621         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
1622         xfs_fid_t       *xfid = (struct xfs_fid *)fidp;
1623         xfs_inode_t     *ip;
1624         int             error;
1625         xfs_ino_t       ino;
1626         unsigned int    igen;
1627
1628         /*
1629          * Invalid.  Since handles can be created in user space and passed in
1630          * via gethandle(), this is not cause for a panic.
1631          */
1632         if (xfid->xfs_fid_len != sizeof(*xfid) - sizeof(xfid->xfs_fid_len))
1633                 return XFS_ERROR(EINVAL);
1634
1635         ino  = xfid->xfs_fid_ino;
1636         igen = xfid->xfs_fid_gen;
1637
1638         /*
1639          * NFS can sometimes send requests for ino 0.  Fail them gracefully.
1640          */
1641         if (ino == 0)
1642                 return XFS_ERROR(ESTALE);
1643
1644         error = xfs_iget(mp, NULL, ino, 0, XFS_ILOCK_SHARED, &ip, 0);
1645         if (error) {
1646                 *vpp = NULL;
1647                 return error;
1648         }
1649
1650         if (ip == NULL) {
1651                 *vpp = NULL;
1652                 return XFS_ERROR(EIO);
1653         }
1654
1655         if (ip->i_d.di_mode == 0 || ip->i_d.di_gen != igen) {
1656                 xfs_iput_new(ip, XFS_ILOCK_SHARED);
1657                 *vpp = NULL;
1658                 return XFS_ERROR(ENOENT);
1659         }
1660
1661         *vpp = XFS_ITOV(ip);
1662         xfs_iunlock(ip, XFS_ILOCK_SHARED);
1663         return 0;
1664 }
1665
1666
1667 #define MNTOPT_LOGBUFS  "logbufs"       /* number of XFS log buffers */
1668 #define MNTOPT_LOGBSIZE "logbsize"      /* size of XFS log buffers */
1669 #define MNTOPT_LOGDEV   "logdev"        /* log device */
1670 #define MNTOPT_RTDEV    "rtdev"         /* realtime I/O device */
1671 #define MNTOPT_BIOSIZE  "biosize"       /* log2 of preferred buffered io size */
1672 #define MNTOPT_WSYNC    "wsync"         /* safe-mode nfs compatible mount */
1673 #define MNTOPT_INO64    "ino64"         /* force inodes into 64-bit range */
1674 #define MNTOPT_NOALIGN  "noalign"       /* turn off stripe alignment */
1675 #define MNTOPT_SWALLOC  "swalloc"       /* turn on stripe width allocation */
1676 #define MNTOPT_SUNIT    "sunit"         /* data volume stripe unit */
1677 #define MNTOPT_SWIDTH   "swidth"        /* data volume stripe width */
1678 #define MNTOPT_NOUUID   "nouuid"        /* ignore filesystem UUID */
1679 #define MNTOPT_MTPT     "mtpt"          /* filesystem mount point */
1680 #define MNTOPT_GRPID    "grpid"         /* group-ID from parent directory */
1681 #define MNTOPT_NOGRPID  "nogrpid"       /* group-ID from current process */
1682 #define MNTOPT_BSDGROUPS    "bsdgroups"    /* group-ID from parent directory */
1683 #define MNTOPT_SYSVGROUPS   "sysvgroups"   /* group-ID from current process */
1684 #define MNTOPT_ALLOCSIZE    "allocsize"    /* preferred allocation size */
1685 #define MNTOPT_NORECOVERY   "norecovery"   /* don't run XFS recovery */
1686 #define MNTOPT_BARRIER  "barrier"       /* use writer barriers for log write and
1687                                          * unwritten extent conversion */
1688 #define MNTOPT_NOBARRIER "nobarrier"    /* .. disable */
1689 #define MNTOPT_OSYNCISOSYNC "osyncisosync" /* o_sync is REALLY o_sync */
1690 #define MNTOPT_64BITINODE   "inode64"   /* inodes can be allocated anywhere */
1691 #define MNTOPT_IKEEP    "ikeep"         /* do not free empty inode clusters */
1692 #define MNTOPT_NOIKEEP  "noikeep"       /* free empty inode clusters */
1693 #define MNTOPT_LARGEIO     "largeio"    /* report large I/O sizes in stat() */
1694 #define MNTOPT_NOLARGEIO   "nolargeio"  /* do not report large I/O sizes
1695                                          * in stat(). */
1696 #define MNTOPT_ATTR2    "attr2"         /* do use attr2 attribute format */
1697 #define MNTOPT_NOATTR2  "noattr2"       /* do not use attr2 attribute format */
1698 #define MNTOPT_FILESTREAM  "filestreams" /* use filestreams allocator */
1699 #define MNTOPT_DMAPI    "dmapi"         /* DMI enabled (DMAPI / XDSM) */
1700 #define MNTOPT_XDSM     "xdsm"          /* DMI enabled (DMAPI / XDSM) */
1701 #define MNTOPT_DMI      "dmi"           /* DMI enabled (DMAPI / XDSM) */
1702
1703 STATIC unsigned long
1704 suffix_strtoul(char *s, char **endp, unsigned int base)
1705 {
1706         int     last, shift_left_factor = 0;
1707         char    *value = s;
1708
1709         last = strlen(value) - 1;
1710         if (value[last] == 'K' || value[last] == 'k') {
1711                 shift_left_factor = 10;
1712                 value[last] = '\0';
1713         }
1714         if (value[last] == 'M' || value[last] == 'm') {
1715                 shift_left_factor = 20;
1716                 value[last] = '\0';
1717         }
1718         if (value[last] == 'G' || value[last] == 'g') {
1719                 shift_left_factor = 30;
1720                 value[last] = '\0';
1721         }
1722
1723         return simple_strtoul((const char *)s, endp, base) << shift_left_factor;
1724 }
1725
1726 STATIC int
1727 xfs_parseargs(
1728         struct bhv_desc         *bhv,
1729         char                    *options,
1730         struct xfs_mount_args   *args,
1731         int                     update)
1732 {
1733         bhv_vfs_t               *vfsp = bhvtovfs(bhv);
1734         char                    *this_char, *value, *eov;
1735         int                     dsunit, dswidth, vol_dsunit, vol_dswidth;
1736         int                     iosize;
1737
1738         /*
1739          * Applications using DMI filesystems often expect the
1740          * inode generation number to be monotonically increasing.
1741          * If we delete inode chunks we break this assumption, so
1742          * keep unused inode chunks on disk for DMI filesystems
1743          * until we come up with a better solution.
1744          * Note that if "ikeep" or "noikeep" mount options are
1745          * supplied, then they are honored.
1746          */
1747         if (!(args->flags & XFSMNT_DMAPI))
1748                 args->flags |= XFSMNT_IDELETE;
1749
1750         args->flags |= XFSMNT_BARRIER;
1751         args->flags2 |= XFSMNT2_COMPAT_IOSIZE;
1752
1753         if (!options)
1754                 goto done;
1755
1756         iosize = dsunit = dswidth = vol_dsunit = vol_dswidth = 0;
1757
1758         while ((this_char = strsep(&options, ",")) != NULL) {
1759                 if (!*this_char)
1760                         continue;
1761                 if ((value = strchr(this_char, '=')) != NULL)
1762                         *value++ = 0;
1763
1764                 if (!strcmp(this_char, MNTOPT_LOGBUFS)) {
1765                         if (!value || !*value) {
1766                                 cmn_err(CE_WARN,
1767                                         "XFS: %s option requires an argument",
1768                                         this_char);
1769                                 return EINVAL;
1770                         }
1771                         args->logbufs = simple_strtoul(value, &eov, 10);
1772                 } else if (!strcmp(this_char, MNTOPT_LOGBSIZE)) {
1773                         if (!value || !*value) {
1774                                 cmn_err(CE_WARN,
1775                                         "XFS: %s option requires an argument",
1776                                         this_char);
1777                                 return EINVAL;
1778                         }
1779                         args->logbufsize = suffix_strtoul(value, &eov, 10);
1780                 } else if (!strcmp(this_char, MNTOPT_LOGDEV)) {
1781                         if (!value || !*value) {
1782                                 cmn_err(CE_WARN,
1783                                         "XFS: %s option requires an argument",
1784                                         this_char);
1785                                 return EINVAL;
1786                         }
1787                         strncpy(args->logname, value, MAXNAMELEN);
1788                 } else if (!strcmp(this_char, MNTOPT_MTPT)) {
1789                         if (!value || !*value) {
1790                                 cmn_err(CE_WARN,
1791                                         "XFS: %s option requires an argument",
1792                                         this_char);
1793                                 return EINVAL;
1794                         }
1795                         strncpy(args->mtpt, value, MAXNAMELEN);
1796                 } else if (!strcmp(this_char, MNTOPT_RTDEV)) {
1797                         if (!value || !*value) {
1798                                 cmn_err(CE_WARN,
1799                                         "XFS: %s option requires an argument",
1800                                         this_char);
1801                                 return EINVAL;
1802                         }
1803                         strncpy(args->rtname, value, MAXNAMELEN);
1804                 } else if (!strcmp(this_char, MNTOPT_BIOSIZE)) {
1805                         if (!value || !*value) {
1806                                 cmn_err(CE_WARN,
1807                                         "XFS: %s option requires an argument",
1808                                         this_char);
1809                                 return EINVAL;
1810                         }
1811                         iosize = simple_strtoul(value, &eov, 10);
1812                         args->flags |= XFSMNT_IOSIZE;
1813                         args->iosizelog = (uint8_t) iosize;
1814                 } else if (!strcmp(this_char, MNTOPT_ALLOCSIZE)) {
1815                         if (!value || !*value) {
1816                                 cmn_err(CE_WARN,
1817                                         "XFS: %s option requires an argument",
1818                                         this_char);
1819                                 return EINVAL;
1820                         }
1821                         iosize = suffix_strtoul(value, &eov, 10);
1822                         args->flags |= XFSMNT_IOSIZE;
1823                         args->iosizelog = ffs(iosize) - 1;
1824                 } else if (!strcmp(this_char, MNTOPT_GRPID) ||
1825                            !strcmp(this_char, MNTOPT_BSDGROUPS)) {
1826                         vfsp->vfs_flag |= VFS_GRPID;
1827                 } else if (!strcmp(this_char, MNTOPT_NOGRPID) ||
1828                            !strcmp(this_char, MNTOPT_SYSVGROUPS)) {
1829                         vfsp->vfs_flag &= ~VFS_GRPID;
1830                 } else if (!strcmp(this_char, MNTOPT_WSYNC)) {
1831                         args->flags |= XFSMNT_WSYNC;
1832                 } else if (!strcmp(this_char, MNTOPT_OSYNCISOSYNC)) {
1833                         args->flags |= XFSMNT_OSYNCISOSYNC;
1834                 } else if (!strcmp(this_char, MNTOPT_NORECOVERY)) {
1835                         args->flags |= XFSMNT_NORECOVERY;
1836                 } else if (!strcmp(this_char, MNTOPT_INO64)) {
1837                         args->flags |= XFSMNT_INO64;
1838 #if !XFS_BIG_INUMS
1839                         cmn_err(CE_WARN,
1840                                 "XFS: %s option not allowed on this system",
1841                                 this_char);
1842                         return EINVAL;
1843 #endif
1844                 } else if (!strcmp(this_char, MNTOPT_NOALIGN)) {
1845                         args->flags |= XFSMNT_NOALIGN;
1846                 } else if (!strcmp(this_char, MNTOPT_SWALLOC)) {
1847                         args->flags |= XFSMNT_SWALLOC;
1848                 } else if (!strcmp(this_char, MNTOPT_SUNIT)) {
1849                         if (!value || !*value) {
1850                                 cmn_err(CE_WARN,
1851                                         "XFS: %s option requires an argument",
1852                                         this_char);
1853                                 return EINVAL;
1854                         }
1855                         dsunit = simple_strtoul(value, &eov, 10);
1856                 } else if (!strcmp(this_char, MNTOPT_SWIDTH)) {
1857                         if (!value || !*value) {
1858                                 cmn_err(CE_WARN,
1859                                         "XFS: %s option requires an argument",
1860                                         this_char);
1861                                 return EINVAL;
1862                         }
1863                         dswidth = simple_strtoul(value, &eov, 10);
1864                 } else if (!strcmp(this_char, MNTOPT_64BITINODE)) {
1865                         args->flags &= ~XFSMNT_32BITINODES;
1866 #if !XFS_BIG_INUMS
1867                         cmn_err(CE_WARN,
1868                                 "XFS: %s option not allowed on this system",
1869                                 this_char);
1870                         return EINVAL;
1871 #endif
1872                 } else if (!strcmp(this_char, MNTOPT_NOUUID)) {
1873                         args->flags |= XFSMNT_NOUUID;
1874                 } else if (!strcmp(this_char, MNTOPT_BARRIER)) {
1875                         args->flags |= XFSMNT_BARRIER;
1876                 } else if (!strcmp(this_char, MNTOPT_NOBARRIER)) {
1877                         args->flags &= ~XFSMNT_BARRIER;
1878                 } else if (!strcmp(this_char, MNTOPT_IKEEP)) {
1879                         args->flags &= ~XFSMNT_IDELETE;
1880                 } else if (!strcmp(this_char, MNTOPT_NOIKEEP)) {
1881                         args->flags |= XFSMNT_IDELETE;
1882                 } else if (!strcmp(this_char, MNTOPT_LARGEIO)) {
1883                         args->flags2 &= ~XFSMNT2_COMPAT_IOSIZE;
1884                 } else if (!strcmp(this_char, MNTOPT_NOLARGEIO)) {
1885                         args->flags2 |= XFSMNT2_COMPAT_IOSIZE;
1886                 } else if (!strcmp(this_char, MNTOPT_ATTR2)) {
1887                         args->flags |= XFSMNT_ATTR2;
1888                 } else if (!strcmp(this_char, MNTOPT_NOATTR2)) {
1889                         args->flags &= ~XFSMNT_ATTR2;
1890                 } else if (!strcmp(this_char, MNTOPT_FILESTREAM)) {
1891                         args->flags2 |= XFSMNT2_FILESTREAMS;
1892                 } else if (!strcmp(this_char, MNTOPT_DMAPI)) {
1893                         args->flags |= XFSMNT_DMAPI;
1894                 } else if (!strcmp(this_char, MNTOPT_XDSM)) {
1895                         args->flags |= XFSMNT_DMAPI;
1896                 } else if (!strcmp(this_char, MNTOPT_DMI)) {
1897                         args->flags |= XFSMNT_DMAPI;
1898                 } else if (!strcmp(this_char, "ihashsize")) {
1899                         cmn_err(CE_WARN,
1900         "XFS: ihashsize no longer used, option is deprecated.");
1901                 } else if (!strcmp(this_char, "osyncisdsync")) {
1902                         /* no-op, this is now the default */
1903                         cmn_err(CE_WARN,
1904         "XFS: osyncisdsync is now the default, option is deprecated.");
1905                 } else if (!strcmp(this_char, "irixsgid")) {
1906                         cmn_err(CE_WARN,
1907         "XFS: irixsgid is now a sysctl(2) variable, option is deprecated.");
1908                 } else {
1909                         cmn_err(CE_WARN,
1910                                 "XFS: unknown mount option [%s].", this_char);
1911                         return EINVAL;
1912                 }
1913         }
1914
1915         if (args->flags & XFSMNT_NORECOVERY) {
1916                 if ((vfsp->vfs_flag & VFS_RDONLY) == 0) {
1917                         cmn_err(CE_WARN,
1918                                 "XFS: no-recovery mounts must be read-only.");
1919                         return EINVAL;
1920                 }
1921         }
1922
1923         if ((args->flags & XFSMNT_NOALIGN) && (dsunit || dswidth)) {
1924                 cmn_err(CE_WARN,
1925         "XFS: sunit and swidth options incompatible with the noalign option");
1926                 return EINVAL;
1927         }
1928
1929         if ((args->flags & XFSMNT_DMAPI) && *args->mtpt == '\0') {
1930                 printk("XFS: %s option needs the mount point option as well\n",
1931                         MNTOPT_DMAPI);
1932                 return EINVAL;
1933         }
1934
1935         if ((dsunit && !dswidth) || (!dsunit && dswidth)) {
1936                 cmn_err(CE_WARN,
1937                         "XFS: sunit and swidth must be specified together");
1938                 return EINVAL;
1939         }
1940
1941         if (dsunit && (dswidth % dsunit != 0)) {
1942                 cmn_err(CE_WARN,
1943         "XFS: stripe width (%d) must be a multiple of the stripe unit (%d)",
1944                         dswidth, dsunit);
1945                 return EINVAL;
1946         }
1947
1948         if ((args->flags & XFSMNT_NOALIGN) != XFSMNT_NOALIGN) {
1949                 if (dsunit) {
1950                         args->sunit = dsunit;
1951                         args->flags |= XFSMNT_RETERR;
1952                 } else {
1953                         args->sunit = vol_dsunit;
1954                 }
1955                 dswidth ? (args->swidth = dswidth) :
1956                           (args->swidth = vol_dswidth);
1957         } else {
1958                 args->sunit = args->swidth = 0;
1959         }
1960
1961 done:
1962         if (args->flags & XFSMNT_32BITINODES)
1963                 vfsp->vfs_flag |= VFS_32BITINODES;
1964         if (args->flags2)
1965                 args->flags |= XFSMNT_FLAGS2;
1966         return 0;
1967 }
1968
1969 STATIC int
1970 xfs_showargs(
1971         struct bhv_desc         *bhv,
1972         struct seq_file         *m)
1973 {
1974         static struct proc_xfs_info {
1975                 int     flag;
1976                 char    *str;
1977         } xfs_info[] = {
1978                 /* the few simple ones we can get from the mount struct */
1979                 { XFS_MOUNT_WSYNC,              "," MNTOPT_WSYNC },
1980                 { XFS_MOUNT_INO64,              "," MNTOPT_INO64 },
1981                 { XFS_MOUNT_NOALIGN,            "," MNTOPT_NOALIGN },
1982                 { XFS_MOUNT_SWALLOC,            "," MNTOPT_SWALLOC },
1983                 { XFS_MOUNT_NOUUID,             "," MNTOPT_NOUUID },
1984                 { XFS_MOUNT_NORECOVERY,         "," MNTOPT_NORECOVERY },
1985                 { XFS_MOUNT_OSYNCISOSYNC,       "," MNTOPT_OSYNCISOSYNC },
1986                 { 0, NULL }
1987         };
1988         struct proc_xfs_info    *xfs_infop;
1989         struct xfs_mount        *mp = XFS_BHVTOM(bhv);
1990         struct bhv_vfs          *vfsp = XFS_MTOVFS(mp);
1991
1992         for (xfs_infop = xfs_info; xfs_infop->flag; xfs_infop++) {
1993                 if (mp->m_flags & xfs_infop->flag)
1994                         seq_puts(m, xfs_infop->str);
1995         }
1996
1997         if (mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)
1998                 seq_printf(m, "," MNTOPT_ALLOCSIZE "=%dk",
1999                                 (int)(1 << mp->m_writeio_log) >> 10);
2000
2001         if (mp->m_logbufs > 0)
2002                 seq_printf(m, "," MNTOPT_LOGBUFS "=%d", mp->m_logbufs);
2003         if (mp->m_logbsize > 0)
2004                 seq_printf(m, "," MNTOPT_LOGBSIZE "=%dk", mp->m_logbsize >> 10);
2005
2006         if (mp->m_logname)
2007                 seq_printf(m, "," MNTOPT_LOGDEV "=%s", mp->m_logname);
2008         if (mp->m_rtname)
2009                 seq_printf(m, "," MNTOPT_RTDEV "=%s", mp->m_rtname);
2010
2011         if (mp->m_dalign > 0)
2012                 seq_printf(m, "," MNTOPT_SUNIT "=%d",
2013                                 (int)XFS_FSB_TO_BB(mp, mp->m_dalign));
2014         if (mp->m_swidth > 0)
2015                 seq_printf(m, "," MNTOPT_SWIDTH "=%d",
2016                                 (int)XFS_FSB_TO_BB(mp, mp->m_swidth));
2017
2018         if (!(mp->m_flags & XFS_MOUNT_IDELETE))
2019                 seq_printf(m, "," MNTOPT_IKEEP);
2020         if (!(mp->m_flags & XFS_MOUNT_COMPAT_IOSIZE))
2021                 seq_printf(m, "," MNTOPT_LARGEIO);
2022
2023         if (!(vfsp->vfs_flag & VFS_32BITINODES))
2024                 seq_printf(m, "," MNTOPT_64BITINODE);
2025         if (vfsp->vfs_flag & VFS_GRPID)
2026                 seq_printf(m, "," MNTOPT_GRPID);
2027
2028         if (vfsp->vfs_flag & VFS_DMI)
2029                 seq_puts(m, "," MNTOPT_DMAPI);
2030
2031         return 0;
2032 }
2033
2034 /*
2035  * Second stage of a freeze. The data is already frozen so we only
2036  * need to take care of themetadata. Once that's done write a dummy
2037  * record to dirty the log in case of a crash while frozen.
2038  */
2039 STATIC void
2040 xfs_freeze(
2041         bhv_desc_t      *bdp)
2042 {
2043         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
2044
2045         xfs_attr_quiesce(mp);
2046         xfs_fs_log_dummy(mp);
2047 }
2048
2049
2050 bhv_vfsops_t xfs_vfsops = {
2051         BHV_IDENTITY_INIT(VFS_BHV_XFS,VFS_POSITION_XFS),
2052         .vfs_parseargs          = xfs_parseargs,
2053         .vfs_showargs           = xfs_showargs,
2054         .vfs_mount              = xfs_mount,
2055         .vfs_unmount            = xfs_unmount,
2056         .vfs_mntupdate          = xfs_mntupdate,
2057         .vfs_root               = xfs_root,
2058         .vfs_statvfs            = xfs_statvfs,
2059         .vfs_sync               = xfs_sync,
2060         .vfs_vget               = xfs_vget,
2061         .vfs_quotactl           = (vfs_quotactl_t)fs_nosys,
2062         .vfs_init_vnode         = xfs_initialize_vnode,
2063         .vfs_force_shutdown     = xfs_do_force_shutdown,
2064         .vfs_freeze             = xfs_freeze,
2065 };