git.oblomov.eu Git - linux-2.6/blob - 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_inode_zone = kmem_zone_init(sizeof(xfs_inode_t), "xfs_inode");
  81         xfs_trans_zone = kmem_zone_init(sizeof(xfs_trans_t), "xfs_trans");
  82         xfs_da_state_zone =
  83                 kmem_zone_init(sizeof(xfs_da_state_t), "xfs_da_state");
  84         xfs_dabuf_zone = kmem_zone_init(sizeof(xfs_dabuf_t), "xfs_dabuf");
  85
  86         /*
  87          * The size of the zone allocated buf log item is the maximum
  88          * size possible under XFS.  This wastes a little bit of memory,
  89          * but it is much faster.
  90          */
  91         xfs_buf_item_zone =
  92                 kmem_zone_init((sizeof(xfs_buf_log_item_t) +
  93                                 (((XFS_MAX_BLOCKSIZE / XFS_BLI_CHUNK) /
  94                                   NBWORD) * sizeof(int))),
  95                                "xfs_buf_item");
  96         xfs_efd_zone = kmem_zone_init((sizeof(xfs_efd_log_item_t) +
  97                                        ((XFS_EFD_MAX_FAST_EXTENTS - 1) * sizeof(xfs_extent_t))),
  98                                       "xfs_efd_item");
  99         xfs_efi_zone = kmem_zone_init((sizeof(xfs_efi_log_item_t) +
 100                                        ((XFS_EFI_MAX_FAST_EXTENTS - 1) * sizeof(xfs_extent_t))),
 101                                       "xfs_efi_item");
 102         xfs_ifork_zone = kmem_zone_init(sizeof(xfs_ifork_t), "xfs_ifork");
 103         xfs_ili_zone = kmem_zone_init(sizeof(xfs_inode_log_item_t), "xfs_ili");
 104         xfs_chashlist_zone = kmem_zone_init(sizeof(xfs_chashlist_t),
 105                                             "xfs_chashlist");
 106         xfs_acl_zone_init(xfs_acl_zone, "xfs_acl");
 107
 108         /*
 109          * Allocate global trace buffers.
 110          */
 111 #ifdef XFS_ALLOC_TRACE
 112         xfs_alloc_trace_buf = ktrace_alloc(XFS_ALLOC_TRACE_SIZE, KM_SLEEP);
 113 #endif
 114 #ifdef XFS_BMAP_TRACE
 115         xfs_bmap_trace_buf = ktrace_alloc(XFS_BMAP_TRACE_SIZE, KM_SLEEP);
 116 #endif
 117 #ifdef XFS_BMBT_TRACE
 118         xfs_bmbt_trace_buf = ktrace_alloc(XFS_BMBT_TRACE_SIZE, KM_SLEEP);
 119 #endif
 120 #ifdef XFS_DIR_TRACE
 121         xfs_dir_trace_buf = ktrace_alloc(XFS_DIR_TRACE_SIZE, KM_SLEEP);
 122 #endif
 123 #ifdef XFS_ATTR_TRACE
 124         xfs_attr_trace_buf = ktrace_alloc(XFS_ATTR_TRACE_SIZE, KM_SLEEP);
 125 #endif
 126 #ifdef XFS_DIR2_TRACE
 127         xfs_dir2_trace_buf = ktrace_alloc(XFS_DIR2_GTRACE_SIZE, KM_SLEEP);
 128 #endif
 129
 130         xfs_dir_startup();
 131
 132 #if (defined(DEBUG) || defined(INDUCE_IO_ERROR))
 133         xfs_error_test_init();
 134 #endif /* DEBUG || INDUCE_IO_ERROR */
 135
 136         xfs_init_procfs();
 137         xfs_sysctl_register();
 138         return 0;
 139 }
 140
 141 void
 142 xfs_cleanup(void)
 143 {
 144         extern kmem_zone_t      *xfs_bmap_free_item_zone;
 145         extern kmem_zone_t      *xfs_btree_cur_zone;
 146         extern kmem_zone_t      *xfs_inode_zone;
 147         extern kmem_zone_t      *xfs_trans_zone;
 148         extern kmem_zone_t      *xfs_da_state_zone;
 149         extern kmem_zone_t      *xfs_dabuf_zone;
 150         extern kmem_zone_t      *xfs_efd_zone;
 151         extern kmem_zone_t      *xfs_efi_zone;
 152         extern kmem_zone_t      *xfs_buf_item_zone;
 153         extern kmem_zone_t      *xfs_chashlist_zone;
 154
 155         xfs_cleanup_procfs();
 156         xfs_sysctl_unregister();
 157         xfs_refcache_destroy();
 158         xfs_acl_zone_destroy(xfs_acl_zone);
 159
 160 #ifdef XFS_DIR2_TRACE
 161         ktrace_free(xfs_dir2_trace_buf);
 162 #endif
 163 #ifdef XFS_ATTR_TRACE
 164         ktrace_free(xfs_attr_trace_buf);
 165 #endif
 166 #ifdef XFS_DIR_TRACE
 167         ktrace_free(xfs_dir_trace_buf);
 168 #endif
 169 #ifdef XFS_BMBT_TRACE
 170         ktrace_free(xfs_bmbt_trace_buf);
 171 #endif
 172 #ifdef XFS_BMAP_TRACE
 173         ktrace_free(xfs_bmap_trace_buf);
 174 #endif
 175 #ifdef XFS_ALLOC_TRACE
 176         ktrace_free(xfs_alloc_trace_buf);
 177 #endif
 178
 179         kmem_cache_destroy(xfs_bmap_free_item_zone);
 180         kmem_cache_destroy(xfs_btree_cur_zone);
 181         kmem_cache_destroy(xfs_inode_zone);
 182         kmem_cache_destroy(xfs_trans_zone);
 183         kmem_cache_destroy(xfs_da_state_zone);
 184         kmem_cache_destroy(xfs_dabuf_zone);
 185         kmem_cache_destroy(xfs_buf_item_zone);
 186         kmem_cache_destroy(xfs_efd_zone);
 187         kmem_cache_destroy(xfs_efi_zone);
 188         kmem_cache_destroy(xfs_ifork_zone);
 189         kmem_cache_destroy(xfs_ili_zone);
 190         kmem_cache_destroy(xfs_chashlist_zone);
 191 }
 192
 193 /*
 194  * xfs_start_flags
 195  *
 196  * This function fills in xfs_mount_t fields based on mount args.
 197  * Note: the superblock has _not_ yet been read in.
 198  */
 199 STATIC int
 200 xfs_start_flags(
 201         struct vfs              *vfs,
 202         struct xfs_mount_args   *ap,
 203         struct xfs_mount        *mp)
 204 {
 205         /* Values are in BBs */
 206         if ((ap->flags & XFSMNT_NOALIGN) != XFSMNT_NOALIGN) {
 207                 /*
 208                  * At this point the superblock has not been read
 209                  * in, therefore we do not know the block size.
 210                  * Before the mount call ends we will convert
 211                  * these to FSBs.
 212                  */
 213                 mp->m_dalign = ap->sunit;
 214                 mp->m_swidth = ap->swidth;
 215         }
 216
 217         if (ap->logbufs != -1 &&
 218             ap->logbufs != 0 &&
 219             (ap->logbufs < XLOG_MIN_ICLOGS ||
 220              ap->logbufs > XLOG_MAX_ICLOGS)) {
 221                 cmn_err(CE_WARN,
 222                         "XFS: invalid logbufs value: %d [not %d-%d]",
 223                         ap->logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
 224                 return XFS_ERROR(EINVAL);
 225         }
 226         mp->m_logbufs = ap->logbufs;
 227         if (ap->logbufsize != -1 &&
 228             ap->logbufsize !=  0 &&
 229             ap->logbufsize != 16 * 1024 &&
 230             ap->logbufsize != 32 * 1024 &&
 231             ap->logbufsize != 64 * 1024 &&
 232             ap->logbufsize != 128 * 1024 &&
 233             ap->logbufsize != 256 * 1024) {
 234                 cmn_err(CE_WARN,
 235         "XFS: invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
 236                         ap->logbufsize);
 237                 return XFS_ERROR(EINVAL);
 238         }
 239         mp->m_ihsize = ap->ihashsize;
 240         mp->m_logbsize = ap->logbufsize;
 241         mp->m_fsname_len = strlen(ap->fsname) + 1;
 242         mp->m_fsname = kmem_alloc(mp->m_fsname_len, KM_SLEEP);
 243         strcpy(mp->m_fsname, ap->fsname);
 244         if (ap->rtname[0]) {
 245                 mp->m_rtname = kmem_alloc(strlen(ap->rtname) + 1, KM_SLEEP);
 246                 strcpy(mp->m_rtname, ap->rtname);
 247         }
 248         if (ap->logname[0]) {
 249                 mp->m_logname = kmem_alloc(strlen(ap->logname) + 1, KM_SLEEP);
 250                 strcpy(mp->m_logname, ap->logname);
 251         }
 252
 253         if (ap->flags & XFSMNT_WSYNC)
 254                 mp->m_flags |= XFS_MOUNT_WSYNC;
 255 #if XFS_BIG_INUMS
 256         if (ap->flags & XFSMNT_INO64) {
 257                 mp->m_flags |= XFS_MOUNT_INO64;
 258                 mp->m_inoadd = XFS_INO64_OFFSET;
 259         }
 260 #endif
 261         if (ap->flags & XFSMNT_NOATIME)
 262                 mp->m_flags |= XFS_MOUNT_NOATIME;
 263         if (ap->flags & XFSMNT_RETERR)
 264                 mp->m_flags |= XFS_MOUNT_RETERR;
 265         if (ap->flags & XFSMNT_NOALIGN)
 266                 mp->m_flags |= XFS_MOUNT_NOALIGN;
 267         if (ap->flags & XFSMNT_SWALLOC)
 268                 mp->m_flags |= XFS_MOUNT_SWALLOC;
 269         if (ap->flags & XFSMNT_OSYNCISOSYNC)
 270                 mp->m_flags |= XFS_MOUNT_OSYNCISOSYNC;
 271         if (ap->flags & XFSMNT_32BITINODES)
 272                 mp->m_flags |= XFS_MOUNT_32BITINODES;
 273
 274         if (ap->flags & XFSMNT_IOSIZE) {
 275                 if (ap->iosizelog > XFS_MAX_IO_LOG ||
 276                     ap->iosizelog < XFS_MIN_IO_LOG) {
 277                         cmn_err(CE_WARN,
 278                 "XFS: invalid log iosize: %d [not %d-%d]",
 279                                 ap->iosizelog, XFS_MIN_IO_LOG,
 280                                 XFS_MAX_IO_LOG);
 281                         return XFS_ERROR(EINVAL);
 282                 }
 283
 284                 mp->m_flags |= XFS_MOUNT_DFLT_IOSIZE;
 285                 mp->m_readio_log = mp->m_writeio_log = ap->iosizelog;
 286         }
 287
 288         if (ap->flags & XFSMNT_IHASHSIZE)
 289                 mp->m_flags |= XFS_MOUNT_IHASHSIZE;
 290         if (ap->flags & XFSMNT_IDELETE)
 291                 mp->m_flags |= XFS_MOUNT_IDELETE;
 292         if (ap->flags & XFSMNT_DIRSYNC)
 293                 mp->m_flags |= XFS_MOUNT_DIRSYNC;
 294         if (ap->flags & XFSMNT_ATTR2)
 295                 mp->m_flags |= XFS_MOUNT_ATTR2;
 296
 297         if (ap->flags2 & XFSMNT2_COMPAT_IOSIZE)
 298                 mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE;
 299
 300         /*
 301          * no recovery flag requires a read-only mount
 302          */
 303         if (ap->flags & XFSMNT_NORECOVERY) {
 304                 if (!(vfs->vfs_flag & VFS_RDONLY)) {
 305                         cmn_err(CE_WARN,
 306         "XFS: tried to mount a FS read-write without recovery!");
 307                         return XFS_ERROR(EINVAL);
 308                 }
 309                 mp->m_flags |= XFS_MOUNT_NORECOVERY;
 310         }
 311
 312         if (ap->flags & XFSMNT_NOUUID)
 313                 mp->m_flags |= XFS_MOUNT_NOUUID;
 314         if (ap->flags & XFSMNT_BARRIER)
 315                 mp->m_flags |= XFS_MOUNT_BARRIER;
 316         else
 317                 mp->m_flags &= ~XFS_MOUNT_BARRIER;
 318
 319         return 0;
 320 }
 321
 322 /*
 323  * This function fills in xfs_mount_t fields based on mount args.
 324  * Note: the superblock _has_ now been read in.
 325  */
 326 STATIC int
 327 xfs_finish_flags(
 328         struct vfs              *vfs,
 329         struct xfs_mount_args   *ap,
 330         struct xfs_mount        *mp)
 331 {
 332         int                     ronly = (vfs->vfs_flag & VFS_RDONLY);
 333
 334         /* Fail a mount where the logbuf is smaller then the log stripe */
 335         if (XFS_SB_VERSION_HASLOGV2(&mp->m_sb)) {
 336                 if ((ap->logbufsize <= 0) &&
 337                     (mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE)) {
 338                         mp->m_logbsize = mp->m_sb.sb_logsunit;
 339                 } else if (ap->logbufsize > 0 &&
 340                            ap->logbufsize < mp->m_sb.sb_logsunit) {
 341                         cmn_err(CE_WARN,
 342         "XFS: logbuf size must be greater than or equal to log stripe size");
 343                         return XFS_ERROR(EINVAL);
 344                 }
 345         } else {
 346                 /* Fail a mount if the logbuf is larger than 32K */
 347                 if (ap->logbufsize > XLOG_BIG_RECORD_BSIZE) {
 348                         cmn_err(CE_WARN,
 349         "XFS: logbuf size for version 1 logs must be 16K or 32K");
 350                         return XFS_ERROR(EINVAL);
 351                 }
 352         }
 353
 354         if (XFS_SB_VERSION_HASATTR2(&mp->m_sb)) {
 355                 mp->m_flags |= XFS_MOUNT_ATTR2;
 356         }
 357
 358         /*
 359          * prohibit r/w mounts of read-only filesystems
 360          */
 361         if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !ronly) {
 362                 cmn_err(CE_WARN,
 363         "XFS: cannot mount a read-only filesystem as read-write");
 364                 return XFS_ERROR(EROFS);
 365         }
 366
 367         /*
 368          * check for shared mount.
 369          */
 370         if (ap->flags & XFSMNT_SHARED) {
 371                 if (!XFS_SB_VERSION_HASSHARED(&mp->m_sb))
 372                         return XFS_ERROR(EINVAL);
 373
 374                 /*
 375                  * For IRIX 6.5, shared mounts must have the shared
 376                  * version bit set, have the persistent readonly
 377                  * field set, must be version 0 and can only be mounted
 378                  * read-only.
 379                  */
 380                 if (!ronly || !(mp->m_sb.sb_flags & XFS_SBF_READONLY) ||
 381                      (mp->m_sb.sb_shared_vn != 0))
 382                         return XFS_ERROR(EINVAL);
 383
 384                 mp->m_flags |= XFS_MOUNT_SHARED;
 385
 386                 /*
 387                  * Shared XFS V0 can't deal with DMI.  Return EINVAL.
 388                  */
 389                 if (mp->m_sb.sb_shared_vn == 0 && (ap->flags & XFSMNT_DMAPI))
 390                         return XFS_ERROR(EINVAL);
 391         }
 392
 393         return 0;
 394 }
 395
 396 /*
 397  * xfs_mount
 398  *
 399  * The file system configurations are:
 400  *      (1) device (partition) with data and internal log
 401  *      (2) logical volume with data and log subvolumes.
 402  *      (3) logical volume with data, log, and realtime subvolumes.
 403  *
 404  * We only have to handle opening the log and realtime volumes here if
 405  * they are present.  The data subvolume has already been opened by
 406  * get_sb_bdev() and is stored in vfsp->vfs_super->s_bdev.
 407  */
 408 STATIC int
 409 xfs_mount(
 410         struct bhv_desc         *bhvp,
 411         struct xfs_mount_args   *args,
 412         cred_t                  *credp)
 413 {
 414         struct vfs              *vfsp = bhvtovfs(bhvp);
 415         struct bhv_desc         *p;
 416         struct xfs_mount        *mp = XFS_BHVTOM(bhvp);
 417         struct block_device     *ddev, *logdev, *rtdev;
 418         int                     flags = 0, error;
 419
 420         ddev = vfsp->vfs_super->s_bdev;
 421         logdev = rtdev = NULL;
 422
 423         /*
 424          * Setup xfs_mount function vectors from available behaviors
 425          */
 426         p = vfs_bhv_lookup(vfsp, VFS_POSITION_DM);
 427         mp->m_dm_ops = p ? *(xfs_dmops_t *) vfs_bhv_custom(p) : xfs_dmcore_stub;
 428         p = vfs_bhv_lookup(vfsp, VFS_POSITION_QM);
 429         mp->m_qm_ops = p ? *(xfs_qmops_t *) vfs_bhv_custom(p) : xfs_qmcore_stub;
 430         p = vfs_bhv_lookup(vfsp, VFS_POSITION_IO);
 431         mp->m_io_ops = p ? *(xfs_ioops_t *) vfs_bhv_custom(p) : xfs_iocore_xfs;
 432
 433         /*
 434          * Open real time and log devices - order is important.
 435          */
 436         if (args->logname[0]) {
 437                 error = xfs_blkdev_get(mp, args->logname, &logdev);
 438                 if (error)
 439                         return error;
 440         }
 441         if (args->rtname[0]) {
 442                 error = xfs_blkdev_get(mp, args->rtname, &rtdev);
 443                 if (error) {
 444                         xfs_blkdev_put(logdev);
 445                         return error;
 446                 }
 447
 448                 if (rtdev == ddev || rtdev == logdev) {
 449                         cmn_err(CE_WARN,
 450         "XFS: Cannot mount filesystem with identical rtdev and ddev/logdev.");
 451                         xfs_blkdev_put(logdev);
 452                         xfs_blkdev_put(rtdev);
 453                         return EINVAL;
 454                 }
 455         }
 456
 457         /*
 458          * Setup xfs_mount buffer target pointers
 459          */
 460         error = ENOMEM;
 461         mp->m_ddev_targp = xfs_alloc_buftarg(ddev, 0);
 462         if (!mp->m_ddev_targp) {
 463                 xfs_blkdev_put(logdev);
 464                 xfs_blkdev_put(rtdev);
 465                 return error;
 466         }
 467         if (rtdev) {
 468                 mp->m_rtdev_targp = xfs_alloc_buftarg(rtdev, 1);
 469                 if (!mp->m_rtdev_targp)
 470                         goto error0;
 471         }
 472         mp->m_logdev_targp = (logdev && logdev != ddev) ?
 473                                 xfs_alloc_buftarg(logdev, 1) : mp->m_ddev_targp;
 474         if (!mp->m_logdev_targp)
 475                 goto error0;
 476
 477         /*
 478          * Setup flags based on mount(2) options and then the superblock
 479          */
 480         error = xfs_start_flags(vfsp, args, mp);
 481         if (error)
 482                 goto error1;
 483         error = xfs_readsb(mp);
 484         if (error)
 485                 goto error1;
 486         error = xfs_finish_flags(vfsp, args, mp);
 487         if (error)
 488                 goto error2;
 489
 490         /*
 491          * Setup xfs_mount buffer target pointers based on superblock
 492          */
 493         error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_blocksize,
 494                                     mp->m_sb.sb_sectsize);
 495         if (!error && logdev && logdev != ddev) {
 496                 unsigned int    log_sector_size = BBSIZE;
 497
 498                 if (XFS_SB_VERSION_HASSECTOR(&mp->m_sb))
 499                         log_sector_size = mp->m_sb.sb_logsectsize;
 500                 error = xfs_setsize_buftarg(mp->m_logdev_targp,
 501                                             mp->m_sb.sb_blocksize,
 502                                             log_sector_size);
 503         }
 504         if (!error && rtdev)
 505                 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
 506                                             mp->m_sb.sb_blocksize,
 507                                             mp->m_sb.sb_sectsize);
 508         if (error)
 509                 goto error2;
 510
 511         if ((mp->m_flags & XFS_MOUNT_BARRIER) && !(vfsp->vfs_flag & VFS_RDONLY))
 512                 xfs_mountfs_check_barriers(mp);
 513
 514         error = XFS_IOINIT(vfsp, args, flags);
 515         if (error)
 516                 goto error2;
 517
 518         return 0;
 519
 520 error2:
 521         if (mp->m_sb_bp)
 522                 xfs_freesb(mp);
 523 error1:
 524         xfs_binval(mp->m_ddev_targp);
 525         if (logdev && logdev != ddev)
 526                 xfs_binval(mp->m_logdev_targp);
 527         if (rtdev)
 528                 xfs_binval(mp->m_rtdev_targp);
 529 error0:
 530         xfs_unmountfs_close(mp, credp);
 531         return error;
 532 }
 533
 534 STATIC int
 535 xfs_unmount(
 536         bhv_desc_t      *bdp,
 537         int             flags,
 538         cred_t          *credp)
 539 {
 540         struct vfs      *vfsp = bhvtovfs(bdp);
 541         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
 542         xfs_inode_t     *rip;
 543         vnode_t         *rvp;
 544         int             unmount_event_wanted = 0;
 545         int             unmount_event_flags = 0;
 546         int             xfs_unmountfs_needed = 0;
 547         int             error;
 548
 549         rip = mp->m_rootip;
 550         rvp = XFS_ITOV(rip);
 551
 552         if (vfsp->vfs_flag & VFS_DMI) {
 553                 error = XFS_SEND_PREUNMOUNT(mp, vfsp,
 554                                 rvp, DM_RIGHT_NULL, rvp, DM_RIGHT_NULL,
 555                                 NULL, NULL, 0, 0,
 556                                 (mp->m_dmevmask & (1<<DM_EVENT_PREUNMOUNT))?
 557                                         0:DM_FLAGS_UNWANTED);
 558                         if (error)
 559                                 return XFS_ERROR(error);
 560                 unmount_event_wanted = 1;
 561                 unmount_event_flags = (mp->m_dmevmask & (1<<DM_EVENT_UNMOUNT))?
 562                                         0 : DM_FLAGS_UNWANTED;
 563         }
 564
 565         /*
 566          * First blow any referenced inode from this file system
 567          * out of the reference cache, and delete the timer.
 568          */
 569         xfs_refcache_purge_mp(mp);
 570
 571         XFS_bflush(mp->m_ddev_targp);
 572         error = xfs_unmount_flush(mp, 0);
 573         if (error)
 574                 goto out;
 575
 576         ASSERT(vn_count(rvp) == 1);
 577
 578         /*
 579          * Drop the reference count
 580          */
 581         VN_RELE(rvp);
 582
 583         /*
 584          * If we're forcing a shutdown, typically because of a media error,
 585          * we want to make sure we invalidate dirty pages that belong to
 586          * referenced vnodes as well.
 587          */
 588         if (XFS_FORCED_SHUTDOWN(mp)) {
 589                 error = xfs_sync(&mp->m_bhv,
 590                          (SYNC_WAIT | SYNC_CLOSE), credp);
 591                 ASSERT(error != EFSCORRUPTED);
 592         }
 593         xfs_unmountfs_needed = 1;
 594
 595 out:
 596         /*      Send DMAPI event, if required.
 597          *      Then do xfs_unmountfs() if needed.
 598          *      Then return error (or zero).
 599          */
 600         if (unmount_event_wanted) {
 601                 /* Note: mp structure must still exist for
 602                  * XFS_SEND_UNMOUNT() call.
 603                  */
 604                 XFS_SEND_UNMOUNT(mp, vfsp, error == 0 ? rvp : NULL,
 605                         DM_RIGHT_NULL, 0, error, unmount_event_flags);
 606         }
 607         if (xfs_unmountfs_needed) {
 608                 /*
 609                  * Call common unmount function to flush to disk
 610                  * and free the super block buffer & mount structures.
 611                  */
 612                 xfs_unmountfs(mp, credp);
 613         }
 614
 615         return XFS_ERROR(error);
 616 }
 617
 618 STATIC int
 619 xfs_quiesce_fs(
 620         xfs_mount_t             *mp)
 621 {
 622         int                     count = 0, pincount;
 623
 624         xfs_refcache_purge_mp(mp);
 625         xfs_flush_buftarg(mp->m_ddev_targp, 0);
 626         xfs_finish_reclaim_all(mp, 0);
 627
 628         /* This loop must run at least twice.
 629          * The first instance of the loop will flush
 630          * most meta data but that will generate more
 631          * meta data (typically directory updates).
 632          * Which then must be flushed and logged before
 633          * we can write the unmount record.
 634          */
 635         do {
 636                 xfs_syncsub(mp, SYNC_REMOUNT|SYNC_ATTR|SYNC_WAIT, 0, NULL);
 637                 pincount = xfs_flush_buftarg(mp->m_ddev_targp, 1);
 638                 if (!pincount) {
 639                         delay(50);
 640                         count++;
 641                 }
 642         } while (count < 2);
 643
 644         return 0;
 645 }
 646
 647 STATIC int
 648 xfs_mntupdate(
 649         bhv_desc_t                      *bdp,
 650         int                             *flags,
 651         struct xfs_mount_args           *args)
 652 {
 653         struct vfs      *vfsp = bhvtovfs(bdp);
 654         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
 655         int             error;
 656
 657         if (args->flags & XFSMNT_NOATIME)
 658                 mp->m_flags |= XFS_MOUNT_NOATIME;
 659         else
 660                 mp->m_flags &= ~XFS_MOUNT_NOATIME;
 661
 662         if (args->flags & XFSMNT_BARRIER)
 663                 mp->m_flags |= XFS_MOUNT_BARRIER;
 664         else
 665                 mp->m_flags &= ~XFS_MOUNT_BARRIER;
 666
 667         if ((vfsp->vfs_flag & VFS_RDONLY) &&
 668             !(*flags & MS_RDONLY)) {
 669                 vfsp->vfs_flag &= ~VFS_RDONLY;
 670
 671                 if (args->flags & XFSMNT_BARRIER)
 672                         xfs_mountfs_check_barriers(mp);
 673         }
 674
 675         if (!(vfsp->vfs_flag & VFS_RDONLY) &&
 676             (*flags & MS_RDONLY)) {
 677                 VFS_SYNC(vfsp, SYNC_FSDATA|SYNC_BDFLUSH|SYNC_ATTR, NULL, error);
 678
 679                 xfs_quiesce_fs(mp);
 680
 681                 /* Ok now write out an unmount record */
 682                 xfs_log_unmount_write(mp);
 683                 xfs_unmountfs_writesb(mp);
 684                 vfsp->vfs_flag |= VFS_RDONLY;
 685         }
 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         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         vnode_t         **vpp)
 784 {
 785         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         xfs_statfs_t    *statp,
 804         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         s = XFS_SB_LOCK(mp);
 818         statp->f_bsize = sbp->sb_blocksize;
 819         lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
 820         statp->f_blocks = sbp->sb_dblocks - lsize;
 821         statp->f_bfree = statp->f_bavail = sbp->sb_fdblocks;
 822         fakeinos = statp->f_bfree << sbp->sb_inopblog;
 823 #if XFS_BIG_INUMS
 824         fakeinos += mp->m_inoadd;
 825 #endif
 826         statp->f_files =
 827             MIN(sbp->sb_icount + fakeinos, (__uint64_t)XFS_MAXINUMBER);
 828         if (mp->m_maxicount)
 829 #if XFS_BIG_INUMS
 830                 if (!mp->m_inoadd)
 831 #endif
 832                         statp->f_files = min_t(typeof(statp->f_files),
 833                                                 statp->f_files,
 834                                                 mp->m_maxicount);
 835         statp->f_ffree = statp->f_files - (sbp->sb_icount - sbp->sb_ifree);
 836         XFS_SB_UNLOCK(mp, s);
 837
 838         xfs_statvfs_fsid(statp, mp);
 839         statp->f_namelen = MAXNAMELEN - 1;
 840
 841         return 0;
 842 }
 843
 844
 845 /*
 846  * xfs_sync flushes any pending I/O to file system vfsp.
 847  *
 848  * This routine is called by vfs_sync() to make sure that things make it
 849  * out to disk eventually, on sync() system calls to flush out everything,
 850  * and when the file system is unmounted.  For the vfs_sync() case, all
 851  * we really need to do is sync out the log to make all of our meta-data
 852  * updates permanent (except for timestamps).  For calls from pflushd(),
 853  * dirty pages are kept moving by calling pdflush() on the inodes
 854  * containing them.  We also flush the inodes that we can lock without
 855  * sleeping and the superblock if we can lock it without sleeping from
 856  * vfs_sync() so that items at the tail of the log are always moving out.
 857  *
 858  * Flags:
 859  *      SYNC_BDFLUSH - We're being called from vfs_sync() so we don't want
 860  *                     to sleep if we can help it.  All we really need
 861  *                     to do is ensure that the log is synced at least
 862  *                     periodically.  We also push the inodes and
 863  *                     superblock if we can lock them without sleeping
 864  *                      and they are not pinned.
 865  *      SYNC_ATTR    - We need to flush the inodes.  If SYNC_BDFLUSH is not
 866  *                     set, then we really want to lock each inode and flush
 867  *                     it.
 868  *      SYNC_WAIT    - All the flushes that take place in this call should
 869  *                     be synchronous.
 870  *      SYNC_DELWRI  - This tells us to push dirty pages associated with
 871  *                     inodes.  SYNC_WAIT and SYNC_BDFLUSH are used to
 872  *                     determine if they should be flushed sync, async, or
 873  *                     delwri.
 874  *      SYNC_CLOSE   - This flag is passed when the system is being
 875  *                     unmounted.  We should sync and invalidate everthing.
 876  *      SYNC_FSDATA  - This indicates that the caller would like to make
 877  *                     sure the superblock is safe on disk.  We can ensure
 878  *                     this by simply makeing sure the log gets flushed
 879  *                     if SYNC_BDFLUSH is set, and by actually writing it
 880  *                     out otherwise.
 881  *
 882  */
 883 /*ARGSUSED*/
 884 STATIC int
 885 xfs_sync(
 886         bhv_desc_t      *bdp,
 887         int             flags,
 888         cred_t          *credp)
 889 {
 890         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
 891
 892         if (unlikely(flags == SYNC_QUIESCE))
 893                 return xfs_quiesce_fs(mp);
 894         else
 895                 return xfs_syncsub(mp, flags, 0, NULL);
 896 }
 897
 898 /*
 899  * xfs sync routine for internal use
 900  *
 901  * This routine supports all of the flags defined for the generic VFS_SYNC
 902  * interface as explained above under xfs_sync.  In the interests of not
 903  * changing interfaces within the 6.5 family, additional internallly-
 904  * required functions are specified within a separate xflags parameter,
 905  * only available by calling this routine.
 906  *
 907  */
 908 int
 909 xfs_sync_inodes(
 910         xfs_mount_t     *mp,
 911         int             flags,
 912         int             xflags,
 913         int             *bypassed)
 914 {
 915         xfs_inode_t     *ip = NULL;
 916         xfs_inode_t     *ip_next;
 917         xfs_buf_t       *bp;
 918         vnode_t         *vp = NULL;
 919         int             error;
 920         int             last_error;
 921         uint64_t        fflag;
 922         uint            lock_flags;
 923         uint            base_lock_flags;
 924         boolean_t       mount_locked;
 925         boolean_t       vnode_refed;
 926         int             preempt;
 927         xfs_dinode_t    *dip;
 928         xfs_iptr_t      *ipointer;
 929 #ifdef DEBUG
 930         boolean_t       ipointer_in = B_FALSE;
 931
 932 #define IPOINTER_SET    ipointer_in = B_TRUE
 933 #define IPOINTER_CLR    ipointer_in = B_FALSE
 934 #else
 935 #define IPOINTER_SET
 936 #define IPOINTER_CLR
 937 #endif
 938
 939
 940 /* Insert a marker record into the inode list after inode ip. The list
 941  * must be locked when this is called. After the call the list will no
 942  * longer be locked.
 943  */
 944 #define IPOINTER_INSERT(ip, mp) { \
 945                 ASSERT(ipointer_in == B_FALSE); \
 946                 ipointer->ip_mnext = ip->i_mnext; \
 947                 ipointer->ip_mprev = ip; \
 948                 ip->i_mnext = (xfs_inode_t *)ipointer; \
 949                 ipointer->ip_mnext->i_mprev = (xfs_inode_t *)ipointer; \
 950                 preempt = 0; \
 951                 XFS_MOUNT_IUNLOCK(mp); \
 952                 mount_locked = B_FALSE; \
 953                 IPOINTER_SET; \
 954         }
 955
 956 /* Remove the marker from the inode list. If the marker was the only item
 957  * in the list then there are no remaining inodes and we should zero out
 958  * the whole list. If we are the current head of the list then move the head
 959  * past us.
 960  */
 961 #define IPOINTER_REMOVE(ip, mp) { \
 962                 ASSERT(ipointer_in == B_TRUE); \
 963                 if (ipointer->ip_mnext != (xfs_inode_t *)ipointer) { \
 964                         ip = ipointer->ip_mnext; \
 965                         ip->i_mprev = ipointer->ip_mprev; \
 966                         ipointer->ip_mprev->i_mnext = ip; \
 967                         if (mp->m_inodes == (xfs_inode_t *)ipointer) { \
 968                                 mp->m_inodes = ip; \
 969                         } \
 970                 } else { \
 971                         ASSERT(mp->m_inodes == (xfs_inode_t *)ipointer); \
 972                         mp->m_inodes = NULL; \
 973                         ip = NULL; \
 974                 } \
 975                 IPOINTER_CLR; \
 976         }
 977
 978 #define XFS_PREEMPT_MASK        0x7f
 979
 980         if (bypassed)
 981                 *bypassed = 0;
 982         if (XFS_MTOVFS(mp)->vfs_flag & VFS_RDONLY)
 983                 return 0;
 984         error = 0;
 985         last_error = 0;
 986         preempt = 0;
 987
 988         /* Allocate a reference marker */
 989         ipointer = (xfs_iptr_t *)kmem_zalloc(sizeof(xfs_iptr_t), KM_SLEEP);
 990
 991         fflag = XFS_B_ASYNC;            /* default is don't wait */
 992         if (flags & (SYNC_BDFLUSH | SYNC_DELWRI))
 993                 fflag = XFS_B_DELWRI;
 994         if (flags & SYNC_WAIT)
 995                 fflag = 0;              /* synchronous overrides all */
 996
 997         base_lock_flags = XFS_ILOCK_SHARED;
 998         if (flags & (SYNC_DELWRI | SYNC_CLOSE)) {
 999                 /*
1000                  * We need the I/O lock if we're going to call any of
1001                  * the flush/inval routines.
1002                  */
1003                 base_lock_flags |= XFS_IOLOCK_SHARED;
1004         }
1005
1006         XFS_MOUNT_ILOCK(mp);
1007
1008         ip = mp->m_inodes;
1009
1010         mount_locked = B_TRUE;
1011         vnode_refed  = B_FALSE;
1012
1013         IPOINTER_CLR;
1014
1015         do {
1016                 ASSERT(ipointer_in == B_FALSE);
1017                 ASSERT(vnode_refed == B_FALSE);
1018
1019                 lock_flags = base_lock_flags;
1020
1021                 /*
1022                  * There were no inodes in the list, just break out
1023                  * of the loop.
1024                  */
1025                 if (ip == NULL) {
1026                         break;
1027                 }
1028
1029                 /*
1030                  * We found another sync thread marker - skip it
1031                  */
1032                 if (ip->i_mount == NULL) {
1033                         ip = ip->i_mnext;
1034                         continue;
1035                 }
1036
1037                 vp = XFS_ITOV_NULL(ip);
1038
1039                 /*
1040                  * If the vnode is gone then this is being torn down,
1041                  * call reclaim if it is flushed, else let regular flush
1042                  * code deal with it later in the loop.
1043                  */
1044
1045                 if (vp == NULL) {
1046                         /* Skip ones already in reclaim */
1047                         if (ip->i_flags & XFS_IRECLAIM) {
1048                                 ip = ip->i_mnext;
1049                                 continue;
1050                         }
1051                         if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0) {
1052                                 ip = ip->i_mnext;
1053                         } else if ((xfs_ipincount(ip) == 0) &&
1054                                     xfs_iflock_nowait(ip)) {
1055                                 IPOINTER_INSERT(ip, mp);
1056
1057                                 xfs_finish_reclaim(ip, 1,
1058                                                 XFS_IFLUSH_DELWRI_ELSE_ASYNC);
1059
1060                                 XFS_MOUNT_ILOCK(mp);
1061                                 mount_locked = B_TRUE;
1062                                 IPOINTER_REMOVE(ip, mp);
1063                         } else {
1064                                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1065                                 ip = ip->i_mnext;
1066                         }
1067                         continue;
1068                 }
1069
1070                 if (VN_BAD(vp)) {
1071                         ip = ip->i_mnext;
1072                         continue;
1073                 }
1074
1075                 if (XFS_FORCED_SHUTDOWN(mp) && !(flags & SYNC_CLOSE)) {
1076                         XFS_MOUNT_IUNLOCK(mp);
1077                         kmem_free(ipointer, sizeof(xfs_iptr_t));
1078                         return 0;
1079                 }
1080
1081                 /*
1082                  * If this is just vfs_sync() or pflushd() calling
1083                  * then we can skip inodes for which it looks like
1084                  * there is nothing to do.  Since we don't have the
1085                  * inode locked this is racey, but these are periodic
1086                  * calls so it doesn't matter.  For the others we want
1087                  * to know for sure, so we at least try to lock them.
1088                  */
1089                 if (flags & SYNC_BDFLUSH) {
1090                         if (((ip->i_itemp == NULL) ||
1091                              !(ip->i_itemp->ili_format.ilf_fields &
1092                                XFS_ILOG_ALL)) &&
1093                             (ip->i_update_core == 0)) {
1094                                 ip = ip->i_mnext;
1095                                 continue;
1096                         }
1097                 }
1098
1099                 /*
1100                  * Try to lock without sleeping.  We're out of order with
1101                  * the inode list lock here, so if we fail we need to drop
1102                  * the mount lock and try again.  If we're called from
1103                  * bdflush() here, then don't bother.
1104                  *
1105                  * The inode lock here actually coordinates with the
1106                  * almost spurious inode lock in xfs_ireclaim() to prevent
1107                  * the vnode we handle here without a reference from
1108                  * being freed while we reference it.  If we lock the inode
1109                  * while it's on the mount list here, then the spurious inode
1110                  * lock in xfs_ireclaim() after the inode is pulled from
1111                  * the mount list will sleep until we release it here.
1112                  * This keeps the vnode from being freed while we reference
1113                  * it.
1114                  */
1115                 if (xfs_ilock_nowait(ip, lock_flags) == 0) {
1116                         if ((flags & SYNC_BDFLUSH) || (vp == NULL)) {
1117                                 ip = ip->i_mnext;
1118                                 continue;
1119                         }
1120
1121                         vp = vn_grab(vp);
1122                         if (vp == NULL) {
1123                                 ip = ip->i_mnext;
1124                                 continue;
1125                         }
1126
1127                         IPOINTER_INSERT(ip, mp);
1128                         xfs_ilock(ip, lock_flags);
1129
1130                         ASSERT(vp == XFS_ITOV(ip));
1131                         ASSERT(ip->i_mount == mp);
1132
1133                         vnode_refed = B_TRUE;
1134                 }
1135
1136                 /* From here on in the loop we may have a marker record
1137                  * in the inode list.
1138                  */
1139
1140                 if ((flags & SYNC_CLOSE)  && (vp != NULL)) {
1141                         /*
1142                          * This is the shutdown case.  We just need to
1143                          * flush and invalidate all the pages associated
1144                          * with the inode.  Drop the inode lock since
1145                          * we can't hold it across calls to the buffer
1146                          * cache.
1147                          *
1148                          * We don't set the VREMAPPING bit in the vnode
1149                          * here, because we don't hold the vnode lock
1150                          * exclusively.  It doesn't really matter, though,
1151                          * because we only come here when we're shutting
1152                          * down anyway.
1153                          */
1154                         xfs_iunlock(ip, XFS_ILOCK_SHARED);
1155
1156                         if (XFS_FORCED_SHUTDOWN(mp)) {
1157                                 VOP_TOSS_PAGES(vp, 0, -1, FI_REMAPF);
1158                         } else {
1159                                 VOP_FLUSHINVAL_PAGES(vp, 0, -1, FI_REMAPF);
1160                         }
1161
1162                         xfs_ilock(ip, XFS_ILOCK_SHARED);
1163
1164                 } else if ((flags & SYNC_DELWRI) && (vp != NULL)) {
1165                         if (VN_DIRTY(vp)) {
1166                                 /* We need to have dropped the lock here,
1167                                  * so insert a marker if we have not already
1168                                  * done so.
1169                                  */
1170                                 if (mount_locked) {
1171                                         IPOINTER_INSERT(ip, mp);
1172                                 }
1173
1174                                 /*
1175                                  * Drop the inode lock since we can't hold it
1176                                  * across calls to the buffer cache.
1177                                  */
1178                                 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1179                                 VOP_FLUSH_PAGES(vp, (xfs_off_t)0, -1,
1180                                                         fflag, FI_NONE, error);
1181                                 xfs_ilock(ip, XFS_ILOCK_SHARED);
1182                         }
1183
1184                 }
1185
1186                 if (flags & SYNC_BDFLUSH) {
1187                         if ((flags & SYNC_ATTR) &&
1188                             ((ip->i_update_core) ||
1189                              ((ip->i_itemp != NULL) &&
1190                               (ip->i_itemp->ili_format.ilf_fields != 0)))) {
1191
1192                                 /* Insert marker and drop lock if not already
1193                                  * done.
1194                                  */
1195                                 if (mount_locked) {
1196                                         IPOINTER_INSERT(ip, mp);
1197                                 }
1198
1199                                 /*
1200                                  * We don't want the periodic flushing of the
1201                                  * inodes by vfs_sync() to interfere with
1202                                  * I/O to the file, especially read I/O
1203                                  * where it is only the access time stamp
1204                                  * that is being flushed out.  To prevent
1205                                  * long periods where we have both inode
1206                                  * locks held shared here while reading the
1207                                  * inode's buffer in from disk, we drop the
1208                                  * inode lock while reading in the inode
1209                                  * buffer.  We have to release the buffer
1210                                  * and reacquire the inode lock so that they
1211                                  * are acquired in the proper order (inode
1212                                  * locks first).  The buffer will go at the
1213                                  * end of the lru chain, though, so we can
1214                                  * expect it to still be there when we go
1215                                  * for it again in xfs_iflush().
1216                                  */
1217                                 if ((xfs_ipincount(ip) == 0) &&
1218                                     xfs_iflock_nowait(ip)) {
1219
1220                                         xfs_ifunlock(ip);
1221                                         xfs_iunlock(ip, XFS_ILOCK_SHARED);
1222
1223                                         error = xfs_itobp(mp, NULL, ip,
1224                                                           &dip, &bp, 0);
1225                                         if (!error) {
1226                                                 xfs_buf_relse(bp);
1227                                         } else {
1228                                                 /* Bailing out, remove the
1229                                                  * marker and free it.
1230                                                  */
1231                                                 XFS_MOUNT_ILOCK(mp);
1232
1233                                                 IPOINTER_REMOVE(ip, mp);
1234
1235                                                 XFS_MOUNT_IUNLOCK(mp);
1236
1237                                                 ASSERT(!(lock_flags &
1238                                                         XFS_IOLOCK_SHARED));
1239
1240                                                 kmem_free(ipointer,
1241                                                         sizeof(xfs_iptr_t));
1242                                                 return (0);
1243                                         }
1244
1245                                         /*
1246                                          * Since we dropped the inode lock,
1247                                          * the inode may have been reclaimed.
1248                                          * Therefore, we reacquire the mount
1249                                          * lock and check to see if we were the
1250                                          * inode reclaimed. If this happened
1251                                          * then the ipointer marker will no
1252                                          * longer point back at us. In this
1253                                          * case, move ip along to the inode
1254                                          * after the marker, remove the marker
1255                                          * and continue.
1256                                          */
1257                                         XFS_MOUNT_ILOCK(mp);
1258                                         mount_locked = B_TRUE;
1259
1260                                         if (ip != ipointer->ip_mprev) {
1261                                                 IPOINTER_REMOVE(ip, mp);
1262
1263                                                 ASSERT(!vnode_refed);
1264                                                 ASSERT(!(lock_flags &
1265                                                         XFS_IOLOCK_SHARED));
1266                                                 continue;
1267                                         }
1268
1269                                         ASSERT(ip->i_mount == mp);
1270
1271                                         if (xfs_ilock_nowait(ip,
1272                                                     XFS_ILOCK_SHARED) == 0) {
1273                                                 ASSERT(ip->i_mount == mp);
1274                                                 /*
1275                                                  * We failed to reacquire
1276                                                  * the inode lock without
1277                                                  * sleeping, so just skip
1278                                                  * the inode for now.  We
1279                                                  * clear the ILOCK bit from
1280                                                  * the lock_flags so that we
1281                                                  * won't try to drop a lock
1282                                                  * we don't hold below.
1283                                                  */
1284                                                 lock_flags &= ~XFS_ILOCK_SHARED;
1285                                                 IPOINTER_REMOVE(ip_next, mp);
1286                                         } else if ((xfs_ipincount(ip) == 0) &&
1287                                                    xfs_iflock_nowait(ip)) {
1288                                                 ASSERT(ip->i_mount == mp);
1289                                                 /*
1290                                                  * Since this is vfs_sync()
1291                                                  * calling we only flush the
1292                                                  * inode out if we can lock
1293                                                  * it without sleeping and
1294                                                  * it is not pinned.  Drop
1295                                                  * the mount lock here so
1296                                                  * that we don't hold it for
1297                                                  * too long. We already have
1298                                                  * a marker in the list here.
1299                                                  */
1300                                                 XFS_MOUNT_IUNLOCK(mp);
1301                                                 mount_locked = B_FALSE;
1302                                                 error = xfs_iflush(ip,
1303                                                            XFS_IFLUSH_DELWRI);
1304                                         } else {
1305                                                 ASSERT(ip->i_mount == mp);
1306                                                 IPOINTER_REMOVE(ip_next, mp);
1307                                         }
1308                                 }
1309
1310                         }
1311
1312                 } else {
1313                         if ((flags & SYNC_ATTR) &&
1314                             ((ip->i_update_core) ||
1315                              ((ip->i_itemp != NULL) &&
1316                               (ip->i_itemp->ili_format.ilf_fields != 0)))) {
1317                                 if (mount_locked) {
1318                                         IPOINTER_INSERT(ip, mp);
1319                                 }
1320
1321                                 if (flags & SYNC_WAIT) {
1322                                         xfs_iflock(ip);
1323                                         error = xfs_iflush(ip,
1324                                                            XFS_IFLUSH_SYNC);
1325                                 } else {
1326                                         /*
1327                                          * If we can't acquire the flush
1328                                          * lock, then the inode is already
1329                                          * being flushed so don't bother
1330                                          * waiting.  If we can lock it then
1331                                          * do a delwri flush so we can
1332                                          * combine multiple inode flushes
1333                                          * in each disk write.
1334                                          */
1335                                         if (xfs_iflock_nowait(ip)) {
1336                                                 error = xfs_iflush(ip,
1337                                                            XFS_IFLUSH_DELWRI);
1338                                         }
1339                                         else if (bypassed)
1340                                                 (*bypassed)++;
1341                                 }
1342                         }
1343                 }
1344
1345                 if (lock_flags != 0) {
1346                         xfs_iunlock(ip, lock_flags);
1347                 }
1348
1349                 if (vnode_refed) {
1350                         /*
1351                          * If we had to take a reference on the vnode
1352                          * above, then wait until after we've unlocked
1353                          * the inode to release the reference.  This is
1354                          * because we can be already holding the inode
1355                          * lock when VN_RELE() calls xfs_inactive().
1356                          *
1357                          * Make sure to drop the mount lock before calling
1358                          * VN_RELE() so that we don't trip over ourselves if
1359                          * we have to go for the mount lock again in the
1360                          * inactive code.
1361                          */
1362                         if (mount_locked) {
1363                                 IPOINTER_INSERT(ip, mp);
1364                         }
1365
1366                         VN_RELE(vp);
1367
1368                         vnode_refed = B_FALSE;
1369                 }
1370
1371                 if (error) {
1372                         last_error = error;
1373                 }
1374
1375                 /*
1376                  * bail out if the filesystem is corrupted.
1377                  */
1378                 if (error == EFSCORRUPTED)  {
1379                         if (!mount_locked) {
1380                                 XFS_MOUNT_ILOCK(mp);
1381                                 IPOINTER_REMOVE(ip, mp);
1382                         }
1383                         XFS_MOUNT_IUNLOCK(mp);
1384                         ASSERT(ipointer_in == B_FALSE);
1385                         kmem_free(ipointer, sizeof(xfs_iptr_t));
1386                         return XFS_ERROR(error);
1387                 }
1388
1389                 /* Let other threads have a chance at the mount lock
1390                  * if we have looped many times without dropping the
1391                  * lock.
1392                  */
1393                 if ((++preempt & XFS_PREEMPT_MASK) == 0) {
1394                         if (mount_locked) {
1395                                 IPOINTER_INSERT(ip, mp);
1396                         }
1397                 }
1398
1399                 if (mount_locked == B_FALSE) {
1400                         XFS_MOUNT_ILOCK(mp);
1401                         mount_locked = B_TRUE;
1402                         IPOINTER_REMOVE(ip, mp);
1403                         continue;
1404                 }
1405
1406                 ASSERT(ipointer_in == B_FALSE);
1407                 ip = ip->i_mnext;
1408
1409         } while (ip != mp->m_inodes);
1410
1411         XFS_MOUNT_IUNLOCK(mp);
1412
1413         ASSERT(ipointer_in == B_FALSE);
1414
1415         kmem_free(ipointer, sizeof(xfs_iptr_t));
1416         return XFS_ERROR(last_error);
1417 }
1418
1419 /*
1420  * xfs sync routine for internal use
1421  *
1422  * This routine supports all of the flags defined for the generic VFS_SYNC
1423  * interface as explained above under xfs_sync.  In the interests of not
1424  * changing interfaces within the 6.5 family, additional internallly-
1425  * required functions are specified within a separate xflags parameter,
1426  * only available by calling this routine.
1427  *
1428  */
1429 int
1430 xfs_syncsub(
1431         xfs_mount_t     *mp,
1432         int             flags,
1433         int             xflags,
1434         int             *bypassed)
1435 {
1436         int             error = 0;
1437         int             last_error = 0;
1438         uint            log_flags = XFS_LOG_FORCE;
1439         xfs_buf_t       *bp;
1440         xfs_buf_log_item_t      *bip;
1441
1442         /*
1443          * Sync out the log.  This ensures that the log is periodically
1444          * flushed even if there is not enough activity to fill it up.
1445          */
1446         if (flags & SYNC_WAIT)
1447                 log_flags |= XFS_LOG_SYNC;
1448
1449         xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1450
1451         if (flags & (SYNC_ATTR|SYNC_DELWRI)) {
1452                 if (flags & SYNC_BDFLUSH)
1453                         xfs_finish_reclaim_all(mp, 1);
1454                 else
1455                         error = xfs_sync_inodes(mp, flags, xflags, bypassed);
1456         }
1457
1458         /*
1459          * Flushing out dirty data above probably generated more
1460          * log activity, so if this isn't vfs_sync() then flush
1461          * the log again.
1462          */
1463         if (flags & SYNC_DELWRI) {
1464                 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1465         }
1466
1467         if (flags & SYNC_FSDATA) {
1468                 /*
1469                  * If this is vfs_sync() then only sync the superblock
1470                  * if we can lock it without sleeping and it is not pinned.
1471                  */
1472                 if (flags & SYNC_BDFLUSH) {
1473                         bp = xfs_getsb(mp, XFS_BUF_TRYLOCK);
1474                         if (bp != NULL) {
1475                                 bip = XFS_BUF_FSPRIVATE(bp,xfs_buf_log_item_t*);
1476                                 if ((bip != NULL) &&
1477                                     xfs_buf_item_dirty(bip)) {
1478                                         if (!(XFS_BUF_ISPINNED(bp))) {
1479                                                 XFS_BUF_ASYNC(bp);
1480                                                 error = xfs_bwrite(mp, bp);
1481                                         } else {
1482                                                 xfs_buf_relse(bp);
1483                                         }
1484                                 } else {
1485                                         xfs_buf_relse(bp);
1486                                 }
1487                         }
1488                 } else {
1489                         bp = xfs_getsb(mp, 0);
1490                         /*
1491                          * If the buffer is pinned then push on the log so
1492                          * we won't get stuck waiting in the write for
1493                          * someone, maybe ourselves, to flush the log.
1494                          * Even though we just pushed the log above, we
1495                          * did not have the superblock buffer locked at
1496                          * that point so it can become pinned in between
1497                          * there and here.
1498                          */
1499                         if (XFS_BUF_ISPINNED(bp))
1500                                 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
1501                         if (flags & SYNC_WAIT)
1502                                 XFS_BUF_UNASYNC(bp);
1503                         else
1504                                 XFS_BUF_ASYNC(bp);
1505                         error = xfs_bwrite(mp, bp);
1506                 }
1507                 if (error) {
1508                         last_error = error;
1509                 }
1510         }
1511
1512         /*
1513          * If this is the periodic sync, then kick some entries out of
1514          * the reference cache.  This ensures that idle entries are
1515          * eventually kicked out of the cache.
1516          */
1517         if (flags & SYNC_REFCACHE) {
1518                 if (flags & SYNC_WAIT)
1519                         xfs_refcache_purge_mp(mp);
1520                 else
1521                         xfs_refcache_purge_some(mp);
1522         }
1523
1524         /*
1525          * Now check to see if the log needs a "dummy" transaction.
1526          */
1527
1528         if (!(flags & SYNC_REMOUNT) && xfs_log_need_covered(mp)) {
1529                 xfs_trans_t *tp;
1530                 xfs_inode_t *ip;
1531
1532                 /*
1533                  * Put a dummy transaction in the log to tell
1534                  * recovery that all others are OK.
1535                  */
1536                 tp = xfs_trans_alloc(mp, XFS_TRANS_DUMMY1);
1537                 if ((error = xfs_trans_reserve(tp, 0,
1538                                 XFS_ICHANGE_LOG_RES(mp),
1539                                 0, 0, 0)))  {
1540                         xfs_trans_cancel(tp, 0);
1541                         return error;
1542                 }
1543
1544                 ip = mp->m_rootip;
1545                 xfs_ilock(ip, XFS_ILOCK_EXCL);
1546
1547                 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1548                 xfs_trans_ihold(tp, ip);
1549                 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1550                 error = xfs_trans_commit(tp, 0, NULL);
1551                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1552                 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1553         }
1554
1555         /*
1556          * When shutting down, we need to insure that the AIL is pushed
1557          * to disk or the filesystem can appear corrupt from the PROM.
1558          */
1559         if ((flags & (SYNC_CLOSE|SYNC_WAIT)) == (SYNC_CLOSE|SYNC_WAIT)) {
1560                 XFS_bflush(mp->m_ddev_targp);
1561                 if (mp->m_rtdev_targp) {
1562                         XFS_bflush(mp->m_rtdev_targp);
1563                 }
1564         }
1565
1566         return XFS_ERROR(last_error);
1567 }
1568
1569 /*
1570  * xfs_vget - called by DMAPI and NFSD to get vnode from file handle
1571  */
1572 STATIC int
1573 xfs_vget(
1574         bhv_desc_t      *bdp,
1575         vnode_t         **vpp,
1576         fid_t           *fidp)
1577 {
1578         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
1579         xfs_fid_t       *xfid = (struct xfs_fid *)fidp;
1580         xfs_inode_t     *ip;
1581         int             error;
1582         xfs_ino_t       ino;
1583         unsigned int    igen;
1584
1585         /*
1586          * Invalid.  Since handles can be created in user space and passed in
1587          * via gethandle(), this is not cause for a panic.
1588          */
1589         if (xfid->xfs_fid_len != sizeof(*xfid) - sizeof(xfid->xfs_fid_len))
1590                 return XFS_ERROR(EINVAL);
1591
1592         ino  = xfid->xfs_fid_ino;
1593         igen = xfid->xfs_fid_gen;
1594
1595         /*
1596          * NFS can sometimes send requests for ino 0.  Fail them gracefully.
1597          */
1598         if (ino == 0)
1599                 return XFS_ERROR(ESTALE);
1600
1601         error = xfs_iget(mp, NULL, ino, 0, XFS_ILOCK_SHARED, &ip, 0);
1602         if (error) {
1603                 *vpp = NULL;
1604                 return error;
1605         }
1606
1607         if (ip == NULL) {
1608                 *vpp = NULL;
1609                 return XFS_ERROR(EIO);
1610         }
1611
1612         if (ip->i_d.di_mode == 0 || ip->i_d.di_gen != igen) {
1613                 xfs_iput_new(ip, XFS_ILOCK_SHARED);
1614                 *vpp = NULL;
1615                 return XFS_ERROR(ENOENT);
1616         }
1617
1618         *vpp = XFS_ITOV(ip);
1619         xfs_iunlock(ip, XFS_ILOCK_SHARED);
1620         return 0;
1621 }
1622
1623
1624 #define MNTOPT_LOGBUFS  "logbufs"       /* number of XFS log buffers */
1625 #define MNTOPT_LOGBSIZE "logbsize"      /* size of XFS log buffers */
1626 #define MNTOPT_LOGDEV   "logdev"        /* log device */
1627 #define MNTOPT_RTDEV    "rtdev"         /* realtime I/O device */
1628 #define MNTOPT_BIOSIZE  "biosize"       /* log2 of preferred buffered io size */
1629 #define MNTOPT_WSYNC    "wsync"         /* safe-mode nfs compatible mount */
1630 #define MNTOPT_INO64    "ino64"         /* force inodes into 64-bit range */
1631 #define MNTOPT_NOALIGN  "noalign"       /* turn off stripe alignment */
1632 #define MNTOPT_SWALLOC  "swalloc"       /* turn on stripe width allocation */
1633 #define MNTOPT_SUNIT    "sunit"         /* data volume stripe unit */
1634 #define MNTOPT_SWIDTH   "swidth"        /* data volume stripe width */
1635 #define MNTOPT_NOUUID   "nouuid"        /* ignore filesystem UUID */
1636 #define MNTOPT_MTPT     "mtpt"          /* filesystem mount point */
1637 #define MNTOPT_GRPID    "grpid"         /* group-ID from parent directory */
1638 #define MNTOPT_NOGRPID  "nogrpid"       /* group-ID from current process */
1639 #define MNTOPT_BSDGROUPS    "bsdgroups"    /* group-ID from parent directory */
1640 #define MNTOPT_SYSVGROUPS   "sysvgroups"   /* group-ID from current process */
1641 #define MNTOPT_ALLOCSIZE    "allocsize"    /* preferred allocation size */
1642 #define MNTOPT_IHASHSIZE    "ihashsize"    /* size of inode hash table */
1643 #define MNTOPT_NORECOVERY   "norecovery"   /* don't run XFS recovery */
1644 #define MNTOPT_BARRIER  "barrier"       /* use writer barriers for log write and
1645                                          * unwritten extent conversion */
1646 #define MNTOPT_NOBARRIER "nobarrier"    /* .. disable */
1647 #define MNTOPT_OSYNCISOSYNC "osyncisosync" /* o_sync is REALLY o_sync */
1648 #define MNTOPT_64BITINODE   "inode64"   /* inodes can be allocated anywhere */
1649 #define MNTOPT_IKEEP    "ikeep"         /* do not free empty inode clusters */
1650 #define MNTOPT_NOIKEEP  "noikeep"       /* free empty inode clusters */
1651 #define MNTOPT_LARGEIO     "largeio"    /* report large I/O sizes in stat() */
1652 #define MNTOPT_NOLARGEIO   "nolargeio"  /* do not report large I/O sizes
1653                                          * in stat(). */
1654 #define MNTOPT_ATTR2    "attr2"         /* do use attr2 attribute format */
1655 #define MNTOPT_NOATTR2  "noattr2"       /* do not use attr2 attribute format */
1656
1657 STATIC unsigned long
1658 suffix_strtoul(const char *cp, char **endp, unsigned int base)
1659 {
1660         int     last, shift_left_factor = 0;
1661         char    *value = (char *)cp;
1662
1663         last = strlen(value) - 1;
1664         if (value[last] == 'K' || value[last] == 'k') {
1665                 shift_left_factor = 10;
1666                 value[last] = '\0';
1667         }
1668         if (value[last] == 'M' || value[last] == 'm') {
1669                 shift_left_factor = 20;
1670                 value[last] = '\0';
1671         }
1672         if (value[last] == 'G' || value[last] == 'g') {
1673                 shift_left_factor = 30;
1674                 value[last] = '\0';
1675         }
1676
1677         return simple_strtoul(cp, endp, base) << shift_left_factor;
1678 }
1679
1680 STATIC int
1681 xfs_parseargs(
1682         struct bhv_desc         *bhv,
1683         char                    *options,
1684         struct xfs_mount_args   *args,
1685         int                     update)
1686 {
1687         struct vfs              *vfsp = bhvtovfs(bhv);
1688         char                    *this_char, *value, *eov;
1689         int                     dsunit, dswidth, vol_dsunit, vol_dswidth;
1690         int                     iosize;
1691
1692         args->flags2 |= XFSMNT2_COMPAT_IOSIZE;
1693
1694 #if 0   /* XXX: off by default, until some remaining issues ironed out */
1695         args->flags |= XFSMNT_IDELETE; /* default to on */
1696 #endif
1697
1698         if (!options)
1699                 goto done;
1700
1701         iosize = dsunit = dswidth = vol_dsunit = vol_dswidth = 0;
1702
1703         while ((this_char = strsep(&options, ",")) != NULL) {
1704                 if (!*this_char)
1705                         continue;
1706                 if ((value = strchr(this_char, '=')) != NULL)
1707                         *value++ = 0;
1708
1709                 if (!strcmp(this_char, MNTOPT_LOGBUFS)) {
1710                         if (!value || !*value) {
1711                                 printk("XFS: %s option requires an argument\n",
1712                                         this_char);
1713                                 return EINVAL;
1714                         }
1715                         args->logbufs = simple_strtoul(value, &eov, 10);
1716                 } else if (!strcmp(this_char, MNTOPT_LOGBSIZE)) {
1717                         if (!value || !*value) {
1718                                 printk("XFS: %s option requires an argument\n",
1719                                         this_char);
1720                                 return EINVAL;
1721                         }
1722                         args->logbufsize = suffix_strtoul(value, &eov, 10);
1723                 } else if (!strcmp(this_char, MNTOPT_LOGDEV)) {
1724                         if (!value || !*value) {
1725                                 printk("XFS: %s option requires an argument\n",
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                                 printk("XFS: %s option requires an argument\n",
1733                                         this_char);
1734                                 return EINVAL;
1735                         }
1736                         strncpy(args->mtpt, value, MAXNAMELEN);
1737                 } else if (!strcmp(this_char, MNTOPT_RTDEV)) {
1738                         if (!value || !*value) {
1739                                 printk("XFS: %s option requires an argument\n",
1740                                         this_char);
1741                                 return EINVAL;
1742                         }
1743                         strncpy(args->rtname, value, MAXNAMELEN);
1744                 } else if (!strcmp(this_char, MNTOPT_BIOSIZE)) {
1745                         if (!value || !*value) {
1746                                 printk("XFS: %s option requires an argument\n",
1747                                         this_char);
1748                                 return EINVAL;
1749                         }
1750                         iosize = simple_strtoul(value, &eov, 10);
1751                         args->flags |= XFSMNT_IOSIZE;
1752                         args->iosizelog = (uint8_t) iosize;
1753                 } else if (!strcmp(this_char, MNTOPT_ALLOCSIZE)) {
1754                         if (!value || !*value) {
1755                                 printk("XFS: %s option requires an argument\n",
1756                                         this_char);
1757                                 return EINVAL;
1758                         }
1759                         iosize = suffix_strtoul(value, &eov, 10);
1760                         args->flags |= XFSMNT_IOSIZE;
1761                         args->iosizelog = ffs(iosize) - 1;
1762                 } else if (!strcmp(this_char, MNTOPT_IHASHSIZE)) {
1763                         if (!value || !*value) {
1764                                 printk("XFS: %s option requires an argument\n",
1765                                         this_char);
1766                                 return EINVAL;
1767                         }
1768                         args->flags |= XFSMNT_IHASHSIZE;
1769                         args->ihashsize = simple_strtoul(value, &eov, 10);
1770                 } else if (!strcmp(this_char, MNTOPT_GRPID) ||
1771                            !strcmp(this_char, MNTOPT_BSDGROUPS)) {
1772                         vfsp->vfs_flag |= VFS_GRPID;
1773                 } else if (!strcmp(this_char, MNTOPT_NOGRPID) ||
1774                            !strcmp(this_char, MNTOPT_SYSVGROUPS)) {
1775                         vfsp->vfs_flag &= ~VFS_GRPID;
1776                 } else if (!strcmp(this_char, MNTOPT_WSYNC)) {
1777                         args->flags |= XFSMNT_WSYNC;
1778                 } else if (!strcmp(this_char, MNTOPT_OSYNCISOSYNC)) {
1779                         args->flags |= XFSMNT_OSYNCISOSYNC;
1780                 } else if (!strcmp(this_char, MNTOPT_NORECOVERY)) {
1781                         args->flags |= XFSMNT_NORECOVERY;
1782                 } else if (!strcmp(this_char, MNTOPT_INO64)) {
1783                         args->flags |= XFSMNT_INO64;
1784 #if !XFS_BIG_INUMS
1785                         printk("XFS: %s option not allowed on this system\n",
1786                                 this_char);
1787                         return EINVAL;
1788 #endif
1789                 } else if (!strcmp(this_char, MNTOPT_NOALIGN)) {
1790                         args->flags |= XFSMNT_NOALIGN;
1791                 } else if (!strcmp(this_char, MNTOPT_SWALLOC)) {
1792                         args->flags |= XFSMNT_SWALLOC;
1793                 } else if (!strcmp(this_char, MNTOPT_SUNIT)) {
1794                         if (!value || !*value) {
1795                                 printk("XFS: %s option requires an argument\n",
1796                                         this_char);
1797                                 return EINVAL;
1798                         }
1799                         dsunit = simple_strtoul(value, &eov, 10);
1800                 } else if (!strcmp(this_char, MNTOPT_SWIDTH)) {
1801                         if (!value || !*value) {
1802                                 printk("XFS: %s option requires an argument\n",
1803                                         this_char);
1804                                 return EINVAL;
1805                         }
1806                         dswidth = simple_strtoul(value, &eov, 10);
1807                 } else if (!strcmp(this_char, MNTOPT_64BITINODE)) {
1808                         args->flags &= ~XFSMNT_32BITINODES;
1809 #if !XFS_BIG_INUMS
1810                         printk("XFS: %s option not allowed on this system\n",
1811                                 this_char);
1812                         return EINVAL;
1813 #endif
1814                 } else if (!strcmp(this_char, MNTOPT_NOUUID)) {
1815                         args->flags |= XFSMNT_NOUUID;
1816                 } else if (!strcmp(this_char, MNTOPT_BARRIER)) {
1817                         args->flags |= XFSMNT_BARRIER;
1818                 } else if (!strcmp(this_char, MNTOPT_NOBARRIER)) {
1819                         args->flags &= ~XFSMNT_BARRIER;
1820                 } else if (!strcmp(this_char, MNTOPT_IKEEP)) {
1821                         args->flags &= ~XFSMNT_IDELETE;
1822                 } else if (!strcmp(this_char, MNTOPT_NOIKEEP)) {
1823                         args->flags |= XFSMNT_IDELETE;
1824                 } else if (!strcmp(this_char, MNTOPT_LARGEIO)) {
1825                         args->flags2 &= ~XFSMNT2_COMPAT_IOSIZE;
1826                 } else if (!strcmp(this_char, MNTOPT_NOLARGEIO)) {
1827                         args->flags2 |= XFSMNT2_COMPAT_IOSIZE;
1828                 } else if (!strcmp(this_char, MNTOPT_ATTR2)) {
1829                         args->flags |= XFSMNT_ATTR2;
1830                 } else if (!strcmp(this_char, MNTOPT_NOATTR2)) {
1831                         args->flags &= ~XFSMNT_ATTR2;
1832                 } else if (!strcmp(this_char, "osyncisdsync")) {
1833                         /* no-op, this is now the default */
1834 printk("XFS: osyncisdsync is now the default, option is deprecated.\n");
1835                 } else if (!strcmp(this_char, "irixsgid")) {
1836 printk("XFS: irixsgid is now a sysctl(2) variable, option is deprecated.\n");
1837                 } else {
1838                         printk("XFS: unknown mount option [%s].\n", this_char);
1839                         return EINVAL;
1840                 }
1841         }
1842
1843         if (args->flags & XFSMNT_NORECOVERY) {
1844                 if ((vfsp->vfs_flag & VFS_RDONLY) == 0) {
1845                         printk("XFS: no-recovery mounts must be read-only.\n");
1846                         return EINVAL;
1847                 }
1848         }
1849
1850         if ((args->flags & XFSMNT_NOALIGN) && (dsunit || dswidth)) {
1851                 printk(
1852         "XFS: sunit and swidth options incompatible with the noalign option\n");
1853                 return EINVAL;
1854         }
1855
1856         if ((dsunit && !dswidth) || (!dsunit && dswidth)) {
1857                 printk("XFS: sunit and swidth must be specified together\n");
1858                 return EINVAL;
1859         }
1860
1861         if (dsunit && (dswidth % dsunit != 0)) {
1862                 printk(
1863         "XFS: stripe width (%d) must be a multiple of the stripe unit (%d)\n",
1864                         dswidth, dsunit);
1865                 return EINVAL;
1866         }
1867
1868         if ((args->flags & XFSMNT_NOALIGN) != XFSMNT_NOALIGN) {
1869                 if (dsunit) {
1870                         args->sunit = dsunit;
1871                         args->flags |= XFSMNT_RETERR;
1872                 } else {
1873                         args->sunit = vol_dsunit;
1874                 }
1875                 dswidth ? (args->swidth = dswidth) :
1876                           (args->swidth = vol_dswidth);
1877         } else {
1878                 args->sunit = args->swidth = 0;
1879         }
1880
1881 done:
1882         if (args->flags & XFSMNT_32BITINODES)
1883                 vfsp->vfs_flag |= VFS_32BITINODES;
1884         if (args->flags2)
1885                 args->flags |= XFSMNT_FLAGS2;
1886         return 0;
1887 }
1888
1889 STATIC int
1890 xfs_showargs(
1891         struct bhv_desc         *bhv,
1892         struct seq_file         *m)
1893 {
1894         static struct proc_xfs_info {
1895                 int     flag;
1896                 char    *str;
1897         } xfs_info[] = {
1898                 /* the few simple ones we can get from the mount struct */
1899                 { XFS_MOUNT_WSYNC,              "," MNTOPT_WSYNC },
1900                 { XFS_MOUNT_INO64,              "," MNTOPT_INO64 },
1901                 { XFS_MOUNT_NOALIGN,            "," MNTOPT_NOALIGN },
1902                 { XFS_MOUNT_SWALLOC,            "," MNTOPT_SWALLOC },
1903                 { XFS_MOUNT_NOUUID,             "," MNTOPT_NOUUID },
1904                 { XFS_MOUNT_NORECOVERY,         "," MNTOPT_NORECOVERY },
1905                 { XFS_MOUNT_OSYNCISOSYNC,       "," MNTOPT_OSYNCISOSYNC },
1906                 { XFS_MOUNT_IDELETE,            "," MNTOPT_NOIKEEP },
1907                 { 0, NULL }
1908         };
1909         struct proc_xfs_info    *xfs_infop;
1910         struct xfs_mount        *mp = XFS_BHVTOM(bhv);
1911         struct vfs              *vfsp = XFS_MTOVFS(mp);
1912
1913         for (xfs_infop = xfs_info; xfs_infop->flag; xfs_infop++) {
1914                 if (mp->m_flags & xfs_infop->flag)
1915                         seq_puts(m, xfs_infop->str);
1916         }
1917
1918         if (mp->m_flags & XFS_MOUNT_IHASHSIZE)
1919                 seq_printf(m, "," MNTOPT_IHASHSIZE "=%d", mp->m_ihsize);
1920
1921         if (mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)
1922                 seq_printf(m, "," MNTOPT_ALLOCSIZE "=%dk",
1923                                 (int)(1 << mp->m_writeio_log) >> 10);
1924
1925         if (mp->m_logbufs > 0)
1926                 seq_printf(m, "," MNTOPT_LOGBUFS "=%d", mp->m_logbufs);
1927         if (mp->m_logbsize > 0)
1928                 seq_printf(m, "," MNTOPT_LOGBSIZE "=%dk", mp->m_logbsize >> 10);
1929
1930         if (mp->m_logname)
1931                 seq_printf(m, "," MNTOPT_LOGDEV "=%s", mp->m_logname);
1932         if (mp->m_rtname)
1933                 seq_printf(m, "," MNTOPT_RTDEV "=%s", mp->m_rtname);
1934
1935         if (mp->m_dalign > 0)
1936                 seq_printf(m, "," MNTOPT_SUNIT "=%d",
1937                                 (int)XFS_FSB_TO_BB(mp, mp->m_dalign));
1938         if (mp->m_swidth > 0)
1939                 seq_printf(m, "," MNTOPT_SWIDTH "=%d",
1940                                 (int)XFS_FSB_TO_BB(mp, mp->m_swidth));
1941
1942         if (!(mp->m_flags & XFS_MOUNT_COMPAT_IOSIZE))
1943                 seq_printf(m, "," MNTOPT_LARGEIO);
1944         if (mp->m_flags & XFS_MOUNT_BARRIER)
1945                 seq_printf(m, "," MNTOPT_BARRIER);
1946
1947         if (!(vfsp->vfs_flag & VFS_32BITINODES))
1948                 seq_printf(m, "," MNTOPT_64BITINODE);
1949         if (vfsp->vfs_flag & VFS_GRPID)
1950                 seq_printf(m, "," MNTOPT_GRPID);
1951
1952         return 0;
1953 }
1954
1955 STATIC void
1956 xfs_freeze(
1957         bhv_desc_t      *bdp)
1958 {
1959         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
1960
1961         while (atomic_read(&mp->m_active_trans) > 0)
1962                 delay(100);
1963
1964         /* Push the superblock and write an unmount record */
1965         xfs_log_unmount_write(mp);
1966         xfs_unmountfs_writesb(mp);
1967         xfs_fs_log_dummy(mp);
1968 }
1969
1970
1971 vfsops_t xfs_vfsops = {
1972         BHV_IDENTITY_INIT(VFS_BHV_XFS,VFS_POSITION_XFS),
1973         .vfs_parseargs          = xfs_parseargs,
1974         .vfs_showargs           = xfs_showargs,
1975         .vfs_mount              = xfs_mount,
1976         .vfs_unmount            = xfs_unmount,
1977         .vfs_mntupdate          = xfs_mntupdate,
1978         .vfs_root               = xfs_root,
1979         .vfs_statvfs            = xfs_statvfs,
1980         .vfs_sync               = xfs_sync,
1981         .vfs_vget               = xfs_vget,
1982         .vfs_dmapiops           = (vfs_dmapiops_t)fs_nosys,
1983         .vfs_quotactl           = (vfs_quotactl_t)fs_nosys,
1984         .vfs_init_vnode         = xfs_initialize_vnode,
1985         .vfs_force_shutdown     = xfs_do_force_shutdown,
1986         .vfs_freeze             = xfs_freeze,
1987 };