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