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