fs/xfs/xfs_vfsops.c

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