git.oblomov.eu Git - linux-2.6/blob - fs/xfs/xfs_vfsops.c

   1 /*
   2  * XFS filesystem operations.
   3  *
   4  * Copyright (c) 2000-2005 Silicon Graphics, Inc.  All Rights Reserved.
   5  *
   6  * This program is free software; you can redistribute it and/or modify it
   7  * under the terms of version 2 of the GNU General Public License as
   8  * published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it would be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
  13  *
  14  * Further, this software is distributed without any warranty that it is
  15  * free of the rightful claim of any third person regarding infringement
  16  * or the like.  Any license provided herein, whether implied or
  17  * otherwise, applies only to this software file.  Patent licenses, if
  18  * any, provided herein do not apply to combinations of this program with
  19  * other software, or any other product whatsoever.
  20  *
  21  * You should have received a copy of the GNU General Public License along
  22  * with this program; if not, write the Free Software Foundation, Inc., 59
  23  * Temple Place - Suite 330, Boston MA 02111-1307, USA.
  24  *
  25  * Contact information: Silicon Graphics, Inc., 1600 Amphitheatre Pkwy,
  26  * Mountain View, CA  94043, or:
  27  *
  28  * http://www.sgi.com
  29  *
  30  * For further information regarding this notice, see:
  31  *
  32  * http://oss.sgi.com/projects/GenInfo/SGIGPLNoticeExplan/
  33  */
  34
  35 #include "xfs.h"
  36 #include "xfs_macros.h"
  37 #include "xfs_types.h"
  38 #include "xfs_inum.h"
  39 #include "xfs_log.h"
  40 #include "xfs_trans.h"
  41 #include "xfs_sb.h"
  42 #include "xfs_dir.h"
  43 #include "xfs_dir2.h"
  44 #include "xfs_dmapi.h"
  45 #include "xfs_mount.h"
  46 #include "xfs_bmap_btree.h"
  47 #include "xfs_ialloc_btree.h"
  48 #include "xfs_alloc_btree.h"
  49 #include "xfs_btree.h"
  50 #include "xfs_alloc.h"
  51 #include "xfs_ialloc.h"
  52 #include "xfs_attr_sf.h"
  53 #include "xfs_dir_sf.h"
  54 #include "xfs_dir2_sf.h"
  55 #include "xfs_dinode.h"
  56 #include "xfs_inode_item.h"
  57 #include "xfs_inode.h"
  58 #include "xfs_ag.h"
  59 #include "xfs_error.h"
  60 #include "xfs_bmap.h"
  61 #include "xfs_da_btree.h"
  62 #include "xfs_rw.h"
  63 #include "xfs_refcache.h"
  64 #include "xfs_buf_item.h"
  65 #include "xfs_extfree_item.h"
  66 #include "xfs_quota.h"
  67 #include "xfs_dir2_trace.h"
  68 #include "xfs_acl.h"
  69 #include "xfs_attr.h"
  70 #include "xfs_clnt.h"
  71 #include "xfs_log_priv.h"
  72
  73 STATIC int xfs_sync(bhv_desc_t *, int, cred_t *);
  74
  75 int
  76 xfs_init(void)
  77 {
  78         extern kmem_zone_t      *xfs_bmap_free_item_zone;
  79         extern kmem_zone_t      *xfs_btree_cur_zone;
  80         extern kmem_zone_t      *xfs_trans_zone;
  81         extern kmem_zone_t      *xfs_buf_item_zone;
  82         extern kmem_zone_t      *xfs_dabuf_zone;
  83 #ifdef XFS_DABUF_DEBUG
  84         extern lock_t           xfs_dabuf_global_lock;
  85         spinlock_init(&xfs_dabuf_global_lock, "xfsda");
  86 #endif
  87
  88         /*
  89          * Initialize all of the zone allocators we use.
  90          */
  91         xfs_bmap_free_item_zone = kmem_zone_init(sizeof(xfs_bmap_free_item_t),
  92                                                  "xfs_bmap_free_item");
  93         xfs_btree_cur_zone = kmem_zone_init(sizeof(xfs_btree_cur_t),
  94                                             "xfs_btree_cur");
  95         xfs_inode_zone = kmem_zone_init(sizeof(xfs_inode_t), "xfs_inode");
  96         xfs_trans_zone = kmem_zone_init(sizeof(xfs_trans_t), "xfs_trans");
  97         xfs_da_state_zone =
  98                 kmem_zone_init(sizeof(xfs_da_state_t), "xfs_da_state");
  99         xfs_dabuf_zone = kmem_zone_init(sizeof(xfs_dabuf_t), "xfs_dabuf");
 100
 101         /*
 102          * The size of the zone allocated buf log item is the maximum
 103          * size possible under XFS.  This wastes a little bit of memory,
 104          * but it is much faster.
 105          */
 106         xfs_buf_item_zone =
 107                 kmem_zone_init((sizeof(xfs_buf_log_item_t) +
 108                                 (((XFS_MAX_BLOCKSIZE / XFS_BLI_CHUNK) /
 109                                   NBWORD) * sizeof(int))),
 110                                "xfs_buf_item");
 111         xfs_efd_zone = kmem_zone_init((sizeof(xfs_efd_log_item_t) +
 112                                        ((XFS_EFD_MAX_FAST_EXTENTS - 1) * sizeof(xfs_extent_t))),
 113                                       "xfs_efd_item");
 114         xfs_efi_zone = kmem_zone_init((sizeof(xfs_efi_log_item_t) +
 115                                        ((XFS_EFI_MAX_FAST_EXTENTS - 1) * sizeof(xfs_extent_t))),
 116                                       "xfs_efi_item");
 117         xfs_ifork_zone = kmem_zone_init(sizeof(xfs_ifork_t), "xfs_ifork");
 118         xfs_ili_zone = kmem_zone_init(sizeof(xfs_inode_log_item_t), "xfs_ili");
 119         xfs_chashlist_zone = kmem_zone_init(sizeof(xfs_chashlist_t),
 120                                             "xfs_chashlist");
 121         xfs_acl_zone_init(xfs_acl_zone, "xfs_acl");
 122
 123         /*
 124          * Allocate global trace buffers.
 125          */
 126 #ifdef XFS_ALLOC_TRACE
 127         xfs_alloc_trace_buf = ktrace_alloc(XFS_ALLOC_TRACE_SIZE, KM_SLEEP);
 128 #endif
 129 #ifdef XFS_BMAP_TRACE
 130         xfs_bmap_trace_buf = ktrace_alloc(XFS_BMAP_TRACE_SIZE, KM_SLEEP);
 131 #endif
 132 #ifdef XFS_BMBT_TRACE
 133         xfs_bmbt_trace_buf = ktrace_alloc(XFS_BMBT_TRACE_SIZE, KM_SLEEP);
 134 #endif
 135 #ifdef XFS_DIR_TRACE
 136         xfs_dir_trace_buf = ktrace_alloc(XFS_DIR_TRACE_SIZE, KM_SLEEP);
 137 #endif
 138 #ifdef XFS_ATTR_TRACE
 139         xfs_attr_trace_buf = ktrace_alloc(XFS_ATTR_TRACE_SIZE, KM_SLEEP);
 140 #endif
 141 #ifdef XFS_DIR2_TRACE
 142         xfs_dir2_trace_buf = ktrace_alloc(XFS_DIR2_GTRACE_SIZE, KM_SLEEP);
 143 #endif
 144
 145         xfs_dir_startup();
 146
 147 #if (defined(DEBUG) || defined(INDUCE_IO_ERROR))
 148         xfs_error_test_init();
 149 #endif /* DEBUG || INDUCE_IO_ERROR */
 150
 151         xfs_init_procfs();
 152         xfs_sysctl_register();
 153         return 0;
 154 }
 155
 156 void
 157 xfs_cleanup(void)
 158 {
 159         extern kmem_zone_t      *xfs_bmap_free_item_zone;
 160         extern kmem_zone_t      *xfs_btree_cur_zone;
 161         extern kmem_zone_t      *xfs_inode_zone;
 162         extern kmem_zone_t      *xfs_trans_zone;
 163         extern kmem_zone_t      *xfs_da_state_zone;
 164         extern kmem_zone_t      *xfs_dabuf_zone;
 165         extern kmem_zone_t      *xfs_efd_zone;
 166         extern kmem_zone_t      *xfs_efi_zone;
 167         extern kmem_zone_t      *xfs_buf_item_zone;
 168         extern kmem_zone_t      *xfs_chashlist_zone;
 169
 170         xfs_cleanup_procfs();
 171         xfs_sysctl_unregister();
 172         xfs_refcache_destroy();
 173         xfs_acl_zone_destroy(xfs_acl_zone);
 174
 175 #ifdef XFS_DIR2_TRACE
 176         ktrace_free(xfs_dir2_trace_buf);
 177 #endif
 178 #ifdef XFS_ATTR_TRACE
 179         ktrace_free(xfs_attr_trace_buf);
 180 #endif
 181 #ifdef XFS_DIR_TRACE
 182         ktrace_free(xfs_dir_trace_buf);
 183 #endif
 184 #ifdef XFS_BMBT_TRACE
 185         ktrace_free(xfs_bmbt_trace_buf);
 186 #endif
 187 #ifdef XFS_BMAP_TRACE
 188         ktrace_free(xfs_bmap_trace_buf);
 189 #endif
 190 #ifdef XFS_ALLOC_TRACE
 191         ktrace_free(xfs_alloc_trace_buf);
 192 #endif
 193
 194         kmem_cache_destroy(xfs_bmap_free_item_zone);
 195         kmem_cache_destroy(xfs_btree_cur_zone);
 196         kmem_cache_destroy(xfs_inode_zone);
 197         kmem_cache_destroy(xfs_trans_zone);
 198         kmem_cache_destroy(xfs_da_state_zone);
 199         kmem_cache_destroy(xfs_dabuf_zone);
 200         kmem_cache_destroy(xfs_buf_item_zone);
 201         kmem_cache_destroy(xfs_efd_zone);
 202         kmem_cache_destroy(xfs_efi_zone);
 203         kmem_cache_destroy(xfs_ifork_zone);
 204         kmem_cache_destroy(xfs_ili_zone);
 205         kmem_cache_destroy(xfs_chashlist_zone);
 206 }
 207
 208 /*
 209  * xfs_start_flags
 210  *
 211  * This function fills in xfs_mount_t fields based on mount args.
 212  * Note: the superblock has _not_ yet been read in.
 213  */
 214 STATIC int
 215 xfs_start_flags(
 216         struct vfs              *vfs,
 217         struct xfs_mount_args   *ap,
 218         struct xfs_mount        *mp)
 219 {
 220         /* Values are in BBs */
 221         if ((ap->flags & XFSMNT_NOALIGN) != XFSMNT_NOALIGN) {
 222                 /*
 223                  * At this point the superblock has not been read
 224                  * in, therefore we do not know the block size.
 225                  * Before the mount call ends we will convert
 226                  * these to FSBs.
 227                  */
 228                 mp->m_dalign = ap->sunit;
 229                 mp->m_swidth = ap->swidth;
 230         }
 231
 232         if (ap->logbufs != -1 &&
 233 #if defined(DEBUG) || defined(XLOG_NOLOG)
 234             ap->logbufs != 0 &&
 235 #endif
 236             (ap->logbufs < XLOG_MIN_ICLOGS ||
 237              ap->logbufs > XLOG_MAX_ICLOGS)) {
 238                 cmn_err(CE_WARN,
 239                         "XFS: invalid logbufs value: %d [not %d-%d]",
 240                         ap->logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
 241                 return XFS_ERROR(EINVAL);
 242         }
 243         mp->m_logbufs = ap->logbufs;
 244         if (ap->logbufsize != -1 &&
 245             ap->logbufsize != 16 * 1024 &&
 246             ap->logbufsize != 32 * 1024 &&
 247             ap->logbufsize != 64 * 1024 &&
 248             ap->logbufsize != 128 * 1024 &&
 249             ap->logbufsize != 256 * 1024) {
 250                 cmn_err(CE_WARN,
 251         "XFS: invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
 252                         ap->logbufsize);
 253                 return XFS_ERROR(EINVAL);
 254         }
 255         mp->m_ihsize = ap->ihashsize;
 256         mp->m_logbsize = ap->logbufsize;
 257         mp->m_fsname_len = strlen(ap->fsname) + 1;
 258         mp->m_fsname = kmem_alloc(mp->m_fsname_len, KM_SLEEP);
 259         strcpy(mp->m_fsname, ap->fsname);
 260
 261         if (ap->flags & XFSMNT_WSYNC)
 262                 mp->m_flags |= XFS_MOUNT_WSYNC;
 263 #if XFS_BIG_INUMS
 264         if (ap->flags & XFSMNT_INO64) {
 265                 mp->m_flags |= XFS_MOUNT_INO64;
 266                 mp->m_inoadd = XFS_INO64_OFFSET;
 267         }
 268 #endif
 269         if (ap->flags & XFSMNT_NOATIME)
 270                 mp->m_flags |= XFS_MOUNT_NOATIME;
 271
 272         if (ap->flags & XFSMNT_RETERR)
 273                 mp->m_flags |= XFS_MOUNT_RETERR;
 274
 275         if (ap->flags & XFSMNT_NOALIGN)
 276                 mp->m_flags |= XFS_MOUNT_NOALIGN;
 277
 278         if (ap->flags & XFSMNT_SWALLOC)
 279                 mp->m_flags |= XFS_MOUNT_SWALLOC;
 280
 281         if (ap->flags & XFSMNT_OSYNCISOSYNC)
 282                 mp->m_flags |= XFS_MOUNT_OSYNCISOSYNC;
 283
 284         if (ap->flags & XFSMNT_32BITINODES)
 285                 mp->m_flags |= (XFS_MOUNT_32BITINODES | XFS_MOUNT_32BITINOOPT);
 286
 287         if (ap->flags & XFSMNT_IOSIZE) {
 288                 if (ap->iosizelog > XFS_MAX_IO_LOG ||
 289                     ap->iosizelog < XFS_MIN_IO_LOG) {
 290                         cmn_err(CE_WARN,
 291                 "XFS: invalid log iosize: %d [not %d-%d]",
 292                                 ap->iosizelog, XFS_MIN_IO_LOG,
 293                                 XFS_MAX_IO_LOG);
 294                         return XFS_ERROR(EINVAL);
 295                 }
 296
 297                 mp->m_flags |= XFS_MOUNT_DFLT_IOSIZE;
 298                 mp->m_readio_log = mp->m_writeio_log = ap->iosizelog;
 299         }
 300
 301         if (ap->flags & XFSMNT_IHASHSIZE)
 302                 mp->m_flags |= XFS_MOUNT_IHASHSIZE;
 303
 304         if (ap->flags & XFSMNT_IDELETE)
 305                 mp->m_flags |= XFS_MOUNT_IDELETE;
 306
 307         if (ap->flags & XFSMNT_DIRSYNC)
 308                 mp->m_flags |= XFS_MOUNT_DIRSYNC;
 309
 310         /*
 311          * no recovery flag requires a read-only mount
 312          */
 313         if (ap->flags & XFSMNT_NORECOVERY) {
 314                 if (!(vfs->vfs_flag & VFS_RDONLY)) {
 315                         cmn_err(CE_WARN,
 316         "XFS: tried to mount a FS read-write without recovery!");
 317                         return XFS_ERROR(EINVAL);
 318                 }
 319                 mp->m_flags |= XFS_MOUNT_NORECOVERY;
 320         }
 321
 322         if (ap->flags & XFSMNT_NOUUID)
 323                 mp->m_flags |= XFS_MOUNT_NOUUID;
 324         if (ap->flags & XFSMNT_NOLOGFLUSH)
 325                 mp->m_flags |= XFS_MOUNT_NOLOGFLUSH;
 326
 327         return 0;
 328 }
 329
 330 /*
 331  * This function fills in xfs_mount_t fields based on mount args.
 332  * Note: the superblock _has_ now been read in.
 333  */
 334 STATIC int
 335 xfs_finish_flags(
 336         struct vfs              *vfs,
 337         struct xfs_mount_args   *ap,
 338         struct xfs_mount        *mp)
 339 {
 340         int                     ronly = (vfs->vfs_flag & VFS_RDONLY);
 341
 342         /* Fail a mount where the logbuf is smaller then the log stripe */
 343         if (XFS_SB_VERSION_HASLOGV2(&mp->m_sb)) {
 344                 if ((ap->logbufsize == -1) &&
 345                     (mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE)) {
 346                         mp->m_logbsize = mp->m_sb.sb_logsunit;
 347                 } else if (ap->logbufsize < mp->m_sb.sb_logsunit) {
 348                         cmn_err(CE_WARN,
 349         "XFS: logbuf size must be greater than or equal to log stripe size");
 350                         return XFS_ERROR(EINVAL);
 351                 }
 352         } else {
 353                 /* Fail a mount if the logbuf is larger than 32K */
 354                 if (ap->logbufsize > XLOG_BIG_RECORD_BSIZE) {
 355                         cmn_err(CE_WARN,
 356         "XFS: logbuf size for version 1 logs must be 16K or 32K");
 357                         return XFS_ERROR(EINVAL);
 358                 }
 359         }
 360
 361         /*
 362          * prohibit r/w mounts of read-only filesystems
 363          */
 364         if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !ronly) {
 365                 cmn_err(CE_WARN,
 366         "XFS: cannot mount a read-only filesystem as read-write");
 367                 return XFS_ERROR(EROFS);
 368         }
 369
 370         /*
 371          * check for shared mount.
 372          */
 373         if (ap->flags & XFSMNT_SHARED) {
 374                 if (!XFS_SB_VERSION_HASSHARED(&mp->m_sb))
 375                         return XFS_ERROR(EINVAL);
 376
 377                 /*
 378                  * For IRIX 6.5, shared mounts must have the shared
 379                  * version bit set, have the persistent readonly
 380                  * field set, must be version 0 and can only be mounted
 381                  * read-only.
 382                  */
 383                 if (!ronly || !(mp->m_sb.sb_flags & XFS_SBF_READONLY) ||
 384                      (mp->m_sb.sb_shared_vn != 0))
 385                         return XFS_ERROR(EINVAL);
 386
 387                 mp->m_flags |= XFS_MOUNT_SHARED;
 388
 389                 /*
 390                  * Shared XFS V0 can't deal with DMI.  Return EINVAL.
 391                  */
 392                 if (mp->m_sb.sb_shared_vn == 0 && (ap->flags & XFSMNT_DMAPI))
 393                         return XFS_ERROR(EINVAL);
 394         }
 395
 396         return 0;
 397 }
 398
 399 /*
 400  * xfs_mount
 401  *
 402  * The file system configurations are:
 403  *      (1) device (partition) with data and internal log
 404  *      (2) logical volume with data and log subvolumes.
 405  *      (3) logical volume with data, log, and realtime subvolumes.
 406  *
 407  * We only have to handle opening the log and realtime volumes here if
 408  * they are present.  The data subvolume has already been opened by
 409  * get_sb_bdev() and is stored in vfsp->vfs_super->s_bdev.
 410  */
 411 STATIC int
 412 xfs_mount(
 413         struct bhv_desc         *bhvp,
 414         struct xfs_mount_args   *args,
 415         cred_t                  *credp)
 416 {
 417         struct vfs              *vfsp = bhvtovfs(bhvp);
 418         struct bhv_desc         *p;
 419         struct xfs_mount        *mp = XFS_BHVTOM(bhvp);
 420         struct block_device     *ddev, *logdev, *rtdev;
 421         int                     flags = 0, error;
 422
 423         ddev = vfsp->vfs_super->s_bdev;
 424         logdev = rtdev = NULL;
 425
 426         /*
 427          * Setup xfs_mount function vectors from available behaviors
 428          */
 429         p = vfs_bhv_lookup(vfsp, VFS_POSITION_DM);
 430         mp->m_dm_ops = p ? *(xfs_dmops_t *) vfs_bhv_custom(p) : xfs_dmcore_stub;
 431         p = vfs_bhv_lookup(vfsp, VFS_POSITION_QM);
 432         mp->m_qm_ops = p ? *(xfs_qmops_t *) vfs_bhv_custom(p) : xfs_qmcore_stub;
 433         p = vfs_bhv_lookup(vfsp, VFS_POSITION_IO);
 434         mp->m_io_ops = p ? *(xfs_ioops_t *) vfs_bhv_custom(p) : xfs_iocore_xfs;
 435
 436         /*
 437          * Open real time and log devices - order is important.
 438          */
 439         if (args->logname[0]) {
 440                 error = xfs_blkdev_get(mp, args->logname, &logdev);
 441                 if (error)
 442                         return error;
 443         }
 444         if (args->rtname[0]) {
 445                 error = xfs_blkdev_get(mp, args->rtname, &rtdev);
 446                 if (error) {
 447                         xfs_blkdev_put(logdev);
 448                         return error;
 449                 }
 450
 451                 if (rtdev == ddev || rtdev == logdev) {
 452                         cmn_err(CE_WARN,
 453         "XFS: Cannot mount filesystem with identical rtdev and ddev/logdev.");
 454                         xfs_blkdev_put(logdev);
 455                         xfs_blkdev_put(rtdev);
 456                         return EINVAL;
 457                 }
 458         }
 459
 460         /*
 461          * Setup xfs_mount buffer target pointers
 462          */
 463         error = ENOMEM;
 464         mp->m_ddev_targp = xfs_alloc_buftarg(ddev, 0);
 465         if (!mp->m_ddev_targp) {
 466                 xfs_blkdev_put(logdev);
 467                 xfs_blkdev_put(rtdev);
 468                 return error;
 469         }
 470         if (rtdev) {
 471                 mp->m_rtdev_targp = xfs_alloc_buftarg(rtdev, 1);
 472                 if (!mp->m_rtdev_targp)
 473                         goto error0;
 474         }
 475         mp->m_logdev_targp = (logdev && logdev != ddev) ?
 476                                 xfs_alloc_buftarg(logdev, 1) : mp->m_ddev_targp;
 477         if (!mp->m_logdev_targp)
 478                 goto error0;
 479
 480         /*
 481          * Setup flags based on mount(2) options and then the superblock
 482          */
 483         error = xfs_start_flags(vfsp, args, mp);
 484         if (error)
 485                 goto error1;
 486         error = xfs_readsb(mp);
 487         if (error)
 488                 goto error1;
 489         error = xfs_finish_flags(vfsp, args, mp);
 490         if (error)
 491                 goto error2;
 492
 493         /*
 494          * Setup xfs_mount buffer target pointers based on superblock
 495          */
 496         error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_blocksize,
 497                                     mp->m_sb.sb_sectsize);
 498         if (!error && logdev && logdev != ddev) {
 499                 unsigned int    log_sector_size = BBSIZE;
 500
 501                 if (XFS_SB_VERSION_HASSECTOR(&mp->m_sb))
 502                         log_sector_size = mp->m_sb.sb_logsectsize;
 503                 error = xfs_setsize_buftarg(mp->m_logdev_targp,
 504                                             mp->m_sb.sb_blocksize,
 505                                             log_sector_size);
 506         }
 507         if (!error && rtdev)
 508                 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
 509                                             mp->m_sb.sb_blocksize,
 510                                             mp->m_sb.sb_sectsize);
 511         if (error)
 512                 goto error2;
 513
 514         error = XFS_IOINIT(vfsp, args, flags);
 515         if (!error)
 516                 return 0;
 517 error2:
 518         if (mp->m_sb_bp)
 519                 xfs_freesb(mp);
 520 error1:
 521         xfs_binval(mp->m_ddev_targp);
 522         if (logdev && logdev != ddev)
 523                 xfs_binval(mp->m_logdev_targp);
 524         if (rtdev)
 525                 xfs_binval(mp->m_rtdev_targp);
 526 error0:
 527         xfs_unmountfs_close(mp, credp);
 528         return error;
 529 }
 530
 531 STATIC int
 532 xfs_unmount(
 533         bhv_desc_t      *bdp,
 534         int             flags,
 535         cred_t          *credp)
 536 {
 537         struct vfs      *vfsp = bhvtovfs(bdp);
 538         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
 539         xfs_inode_t     *rip;
 540         vnode_t         *rvp;
 541         int             unmount_event_wanted = 0;
 542         int             unmount_event_flags = 0;
 543         int             xfs_unmountfs_needed = 0;
 544         int             error;
 545
 546         rip = mp->m_rootip;
 547         rvp = XFS_ITOV(rip);
 548
 549         if (vfsp->vfs_flag & VFS_DMI) {
 550                 error = XFS_SEND_PREUNMOUNT(mp, vfsp,
 551                                 rvp, DM_RIGHT_NULL, rvp, DM_RIGHT_NULL,
 552                                 NULL, NULL, 0, 0,
 553                                 (mp->m_dmevmask & (1<<DM_EVENT_PREUNMOUNT))?
 554                                         0:DM_FLAGS_UNWANTED);
 555                         if (error)
 556                                 return XFS_ERROR(error);
 557                 unmount_event_wanted = 1;
 558                 unmount_event_flags = (mp->m_dmevmask & (1<<DM_EVENT_UNMOUNT))?
 559                                         0 : DM_FLAGS_UNWANTED;
 560         }
 561
 562         /*
 563          * First blow any referenced inode from this file system
 564          * out of the reference cache, and delete the timer.
 565          */
 566         xfs_refcache_purge_mp(mp);
 567
 568         XFS_bflush(mp->m_ddev_targp);
 569         error = xfs_unmount_flush(mp, 0);
 570         if (error)
 571                 goto out;
 572
 573         ASSERT(vn_count(rvp) == 1);
 574
 575         /*
 576          * Drop the reference count
 577          */
 578         VN_RELE(rvp);
 579
 580         /*
 581          * If we're forcing a shutdown, typically because of a media error,
 582          * we want to make sure we invalidate dirty pages that belong to
 583          * referenced vnodes as well.
 584          */
 585         if (XFS_FORCED_SHUTDOWN(mp)) {
 586                 error = xfs_sync(&mp->m_bhv,
 587                          (SYNC_WAIT | SYNC_CLOSE), credp);
 588                 ASSERT(error != EFSCORRUPTED);
 589         }
 590         xfs_unmountfs_needed = 1;
 591
 592 out:
 593         /*      Send DMAPI event, if required.
 594          *      Then do xfs_unmountfs() if needed.
 595          *      Then return error (or zero).
 596          */
 597         if (unmount_event_wanted) {
 598                 /* Note: mp structure must still exist for
 599                  * XFS_SEND_UNMOUNT() call.
 600                  */
 601                 XFS_SEND_UNMOUNT(mp, vfsp, error == 0 ? rvp : NULL,
 602                         DM_RIGHT_NULL, 0, error, unmount_event_flags);
 603         }
 604         if (xfs_unmountfs_needed) {
 605                 /*
 606                  * Call common unmount function to flush to disk
 607                  * and free the super block buffer & mount structures.
 608                  */
 609                 xfs_unmountfs(mp, credp);
 610         }
 611
 612         return XFS_ERROR(error);
 613 }
 614
 615 STATIC int
 616 xfs_quiesce_fs(
 617         xfs_mount_t             *mp)
 618 {
 619         int                     count = 0, pincount;
 620
 621         xfs_refcache_purge_mp(mp);
 622         xfs_flush_buftarg(mp->m_ddev_targp, 0);
 623         xfs_finish_reclaim_all(mp, 0);
 624
 625         /* This loop must run at least twice.
 626          * The first instance of the loop will flush
 627          * most meta data but that will generate more
 628          * meta data (typically directory updates).
 629          * Which then must be flushed and logged before
 630          * we can write the unmount record.
 631          */
 632         do {
 633                 xfs_syncsub(mp, SYNC_REMOUNT|SYNC_ATTR|SYNC_WAIT, 0, NULL);
 634                 pincount = xfs_flush_buftarg(mp->m_ddev_targp, 1);
 635                 if (!pincount) {
 636                         delay(50);
 637                         count++;
 638                 }
 639         } while (count < 2);
 640
 641         return 0;
 642 }
 643
 644 STATIC int
 645 xfs_mntupdate(
 646         bhv_desc_t                      *bdp,
 647         int                             *flags,
 648         struct xfs_mount_args           *args)
 649 {
 650         struct vfs      *vfsp = bhvtovfs(bdp);
 651         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
 652         int             error;
 653
 654         if (args->flags & XFSMNT_NOATIME)
 655                 mp->m_flags |= XFS_MOUNT_NOATIME;
 656         else
 657                 mp->m_flags &= ~XFS_MOUNT_NOATIME;
 658
 659         if (!(vfsp->vfs_flag & VFS_RDONLY)) {
 660                 VFS_SYNC(vfsp, SYNC_FSDATA|SYNC_BDFLUSH|SYNC_ATTR, NULL, error);
 661         }
 662
 663         if (*flags & MS_RDONLY) {
 664                 xfs_quiesce_fs(mp);
 665
 666                 /* Ok now write out an unmount record */
 667                 xfs_log_unmount_write(mp);
 668                 xfs_unmountfs_writesb(mp);
 669                 vfsp->vfs_flag |= VFS_RDONLY;
 670         } else {
 671                 vfsp->vfs_flag &= ~VFS_RDONLY;
 672         }
 673
 674         return 0;
 675 }
 676
 677 /*
 678  * xfs_unmount_flush implements a set of flush operation on special
 679  * inodes, which are needed as a separate set of operations so that
 680  * they can be called as part of relocation process.
 681  */
 682 int
 683 xfs_unmount_flush(
 684         xfs_mount_t     *mp,            /* Mount structure we are getting
 685                                            rid of. */
 686         int             relocation)     /* Called from vfs relocation. */
 687 {
 688         xfs_inode_t     *rip = mp->m_rootip;
 689         xfs_inode_t     *rbmip;
 690         xfs_inode_t     *rsumip = NULL;
 691         vnode_t         *rvp = XFS_ITOV(rip);
 692         int             error;
 693
 694         xfs_ilock(rip, XFS_ILOCK_EXCL);
 695         xfs_iflock(rip);
 696
 697         /*
 698          * Flush out the real time inodes.
 699          */
 700         if ((rbmip = mp->m_rbmip) != NULL) {
 701                 xfs_ilock(rbmip, XFS_ILOCK_EXCL);
 702                 xfs_iflock(rbmip);
 703                 error = xfs_iflush(rbmip, XFS_IFLUSH_SYNC);
 704                 xfs_iunlock(rbmip, XFS_ILOCK_EXCL);
 705
 706                 if (error == EFSCORRUPTED)
 707                         goto fscorrupt_out;
 708
 709                 ASSERT(vn_count(XFS_ITOV(rbmip)) == 1);
 710
 711                 rsumip = mp->m_rsumip;
 712                 xfs_ilock(rsumip, XFS_ILOCK_EXCL);
 713                 xfs_iflock(rsumip);
 714                 error = xfs_iflush(rsumip, XFS_IFLUSH_SYNC);
 715                 xfs_iunlock(rsumip, XFS_ILOCK_EXCL);
 716
 717                 if (error == EFSCORRUPTED)
 718                         goto fscorrupt_out;
 719
 720                 ASSERT(vn_count(XFS_ITOV(rsumip)) == 1);
 721         }
 722
 723         /*
 724          * Synchronously flush root inode to disk
 725          */
 726         error = xfs_iflush(rip, XFS_IFLUSH_SYNC);
 727         if (error == EFSCORRUPTED)
 728                 goto fscorrupt_out2;
 729
 730         if (vn_count(rvp) != 1 && !relocation) {
 731                 xfs_iunlock(rip, XFS_ILOCK_EXCL);
 732                 return XFS_ERROR(EBUSY);
 733         }
 734
 735         /*
 736          * Release dquot that rootinode, rbmino and rsumino might be holding,
 737          * flush and purge the quota inodes.
 738          */
 739         error = XFS_QM_UNMOUNT(mp);
 740         if (error == EFSCORRUPTED)
 741                 goto fscorrupt_out2;
 742
 743         if (rbmip) {
 744                 VN_RELE(XFS_ITOV(rbmip));
 745                 VN_RELE(XFS_ITOV(rsumip));
 746         }
 747
 748         xfs_iunlock(rip, XFS_ILOCK_EXCL);
 749         return 0;
 750
 751 fscorrupt_out:
 752         xfs_ifunlock(rip);
 753
 754 fscorrupt_out2:
 755         xfs_iunlock(rip, XFS_ILOCK_EXCL);
 756
 757         return XFS_ERROR(EFSCORRUPTED);
 758 }
 759
 760 /*
 761  * xfs_root extracts the root vnode from a vfs.
 762  *
 763  * vfsp -- the vfs struct for the desired file system
 764  * vpp  -- address of the caller's vnode pointer which should be
 765  *         set to the desired fs root vnode
 766  */
 767 STATIC int
 768 xfs_root(
 769         bhv_desc_t      *bdp,
 770         vnode_t         **vpp)
 771 {
 772         vnode_t         *vp;
 773
 774         vp = XFS_ITOV((XFS_BHVTOM(bdp))->m_rootip);
 775         VN_HOLD(vp);
 776         *vpp = vp;
 777         return 0;
 778 }
 779
 780 /*
 781  * xfs_statvfs
 782  *
 783  * Fill in the statvfs structure for the given file system.  We use
 784  * the superblock lock in the mount structure to ensure a consistent
 785  * snapshot of the counters returned.
 786  */
 787 STATIC int
 788 xfs_statvfs(
 789         bhv_desc_t      *bdp,
 790         xfs_statfs_t    *statp,
 791         vnode_t         *vp)
 792 {
 793         __uint64_t      fakeinos;
 794         xfs_extlen_t    lsize;
 795         xfs_mount_t     *mp;
 796         xfs_sb_t        *sbp;
 797         unsigned long   s;
 798         u64 id;
 799
 800         mp = XFS_BHVTOM(bdp);
 801         sbp = &(mp->m_sb);
 802
 803         statp->f_type = XFS_SB_MAGIC;
 804
 805         s = XFS_SB_LOCK(mp);
 806         statp->f_bsize = sbp->sb_blocksize;
 807         lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
 808         statp->f_blocks = sbp->sb_dblocks - lsize;
 809         statp->f_bfree = statp->f_bavail = sbp->sb_fdblocks;
 810         fakeinos = statp->f_bfree << sbp->sb_inopblog;
 811 #if XFS_BIG_INUMS
 812         fakeinos += mp->m_inoadd;
 813 #endif
 814         statp->f_files =
 815             MIN(sbp->sb_icount + fakeinos, (__uint64_t)XFS_MAXINUMBER);
 816         if (mp->m_maxicount)
 817 #if XFS_BIG_INUMS
 818                 if (!mp->m_inoadd)
 819 #endif
 820                         statp->f_files = min_t(typeof(statp->f_files),
 821                                                 statp->f_files,
 822                                                 mp->m_maxicount);
 823         statp->f_ffree = statp->f_files - (sbp->sb_icount - sbp->sb_ifree);
 824         XFS_SB_UNLOCK(mp, s);
 825
 826         id = huge_encode_dev(mp->m_dev);
 827         statp->f_fsid.val[0] = (u32)id;
 828         statp->f_fsid.val[1] = (u32)(id >> 32);
 829         statp->f_namelen = MAXNAMELEN - 1;
 830
 831         return 0;
 832 }
 833
 834
 835 /*
 836  * xfs_sync flushes any pending I/O to file system vfsp.
 837  *
 838  * This routine is called by vfs_sync() to make sure that things make it
 839  * out to disk eventually, on sync() system calls to flush out everything,
 840  * and when the file system is unmounted.  For the vfs_sync() case, all
 841  * we really need to do is sync out the log to make all of our meta-data
 842  * updates permanent (except for timestamps).  For calls from pflushd(),
 843  * dirty pages are kept moving by calling pdflush() on the inodes
 844  * containing them.  We also flush the inodes that we can lock without
 845  * sleeping and the superblock if we can lock it without sleeping from
 846  * vfs_sync() so that items at the tail of the log are always moving out.
 847  *
 848  * Flags:
 849  *      SYNC_BDFLUSH - We're being called from vfs_sync() so we don't want
 850  *                     to sleep if we can help it.  All we really need
 851  *                     to do is ensure that the log is synced at least
 852  *                     periodically.  We also push the inodes and
 853  *                     superblock if we can lock them without sleeping
 854  *                      and they are not pinned.
 855  *      SYNC_ATTR    - We need to flush the inodes.  If SYNC_BDFLUSH is not
 856  *                     set, then we really want to lock each inode and flush
 857  *                     it.
 858  *      SYNC_WAIT    - All the flushes that take place in this call should
 859  *                     be synchronous.
 860  *      SYNC_DELWRI  - This tells us to push dirty pages associated with
 861  *                     inodes.  SYNC_WAIT and SYNC_BDFLUSH are used to
 862  *                     determine if they should be flushed sync, async, or
 863  *                     delwri.
 864  *      SYNC_CLOSE   - This flag is passed when the system is being
 865  *                     unmounted.  We should sync and invalidate everthing.
 866  *      SYNC_FSDATA  - This indicates that the caller would like to make
 867  *                     sure the superblock is safe on disk.  We can ensure
 868  *                     this by simply makeing sure the log gets flushed
 869  *                     if SYNC_BDFLUSH is set, and by actually writing it
 870  *                     out otherwise.
 871  *
 872  */
 873 /*ARGSUSED*/
 874 STATIC int
 875 xfs_sync(
 876         bhv_desc_t      *bdp,
 877         int             flags,
 878         cred_t          *credp)
 879 {
 880         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
 881
 882         if (unlikely(flags == SYNC_QUIESCE))
 883                 return xfs_quiesce_fs(mp);
 884         else
 885                 return xfs_syncsub(mp, flags, 0, NULL);
 886 }
 887
 888 /*
 889  * xfs sync routine for internal use
 890  *
 891  * This routine supports all of the flags defined for the generic VFS_SYNC
 892  * interface as explained above under xfs_sync.  In the interests of not
 893  * changing interfaces within the 6.5 family, additional internallly-
 894  * required functions are specified within a separate xflags parameter,
 895  * only available by calling this routine.
 896  *
 897  */
 898 STATIC int
 899 xfs_sync_inodes(
 900         xfs_mount_t     *mp,
 901         int             flags,
 902         int             xflags,
 903         int             *bypassed)
 904 {
 905         xfs_inode_t     *ip = NULL;
 906         xfs_inode_t     *ip_next;
 907         xfs_buf_t       *bp;
 908         vnode_t         *vp = NULL;
 909         int             error;
 910         int             last_error;
 911         uint64_t        fflag;
 912         uint            lock_flags;
 913         uint            base_lock_flags;
 914         boolean_t       mount_locked;
 915         boolean_t       vnode_refed;
 916         int             preempt;
 917         xfs_dinode_t    *dip;
 918         xfs_iptr_t      *ipointer;
 919 #ifdef DEBUG
 920         boolean_t       ipointer_in = B_FALSE;
 921
 922 #define IPOINTER_SET    ipointer_in = B_TRUE
 923 #define IPOINTER_CLR    ipointer_in = B_FALSE
 924 #else
 925 #define IPOINTER_SET
 926 #define IPOINTER_CLR
 927 #endif
 928
 929
 930 /* Insert a marker record into the inode list after inode ip. The list
 931  * must be locked when this is called. After the call the list will no
 932  * longer be locked.
 933  */
 934 #define IPOINTER_INSERT(ip, mp) { \
 935                 ASSERT(ipointer_in == B_FALSE); \
 936                 ipointer->ip_mnext = ip->i_mnext; \
 937                 ipointer->ip_mprev = ip; \
 938                 ip->i_mnext = (xfs_inode_t *)ipointer; \
 939                 ipointer->ip_mnext->i_mprev = (xfs_inode_t *)ipointer; \
 940                 preempt = 0; \
 941                 XFS_MOUNT_IUNLOCK(mp); \
 942                 mount_locked = B_FALSE; \
 943                 IPOINTER_SET; \
 944         }
 945
 946 /* Remove the marker from the inode list. If the marker was the only item
 947  * in the list then there are no remaining inodes and we should zero out
 948  * the whole list. If we are the current head of the list then move the head
 949  * past us.
 950  */
 951 #define IPOINTER_REMOVE(ip, mp) { \
 952                 ASSERT(ipointer_in == B_TRUE); \
 953                 if (ipointer->ip_mnext != (xfs_inode_t *)ipointer) { \
 954                         ip = ipointer->ip_mnext; \
 955                         ip->i_mprev = ipointer->ip_mprev; \
 956                         ipointer->ip_mprev->i_mnext = ip; \
 957                         if (mp->m_inodes == (xfs_inode_t *)ipointer) { \
 958                                 mp->m_inodes = ip; \
 959                         } \
 960                 } else { \
 961                         ASSERT(mp->m_inodes == (xfs_inode_t *)ipointer); \
 962                         mp->m_inodes = NULL; \
 963                         ip = NULL; \
 964                 } \
 965                 IPOINTER_CLR; \
 966         }
 967
 968 #define XFS_PREEMPT_MASK        0x7f
 969
 970         if (bypassed)
 971                 *bypassed = 0;
 972         if (XFS_MTOVFS(mp)->vfs_flag & VFS_RDONLY)
 973                 return 0;
 974         error = 0;
 975         last_error = 0;
 976         preempt = 0;
 977
 978         /* Allocate a reference marker */
 979         ipointer = (xfs_iptr_t *)kmem_zalloc(sizeof(xfs_iptr_t), KM_SLEEP);
 980
 981         fflag = XFS_B_ASYNC;            /* default is don't wait */
 982         if (flags & SYNC_BDFLUSH)
 983                 fflag = XFS_B_DELWRI;
 984         if (flags & SYNC_WAIT)
 985                 fflag = 0;              /* synchronous overrides all */
 986
 987         base_lock_flags = XFS_ILOCK_SHARED;
 988         if (flags & (SYNC_DELWRI | SYNC_CLOSE)) {
 989                 /*
 990                  * We need the I/O lock if we're going to call any of
 991                  * the flush/inval routines.
 992                  */
 993                 base_lock_flags |= XFS_IOLOCK_SHARED;
 994         }
 995
 996         XFS_MOUNT_ILOCK(mp);
 997
 998         ip = mp->m_inodes;
 999
1000         mount_locked = B_TRUE;
1001         vnode_refed  = B_FALSE;
1002
1003         IPOINTER_CLR;
1004
1005         do {
1006                 ASSERT(ipointer_in == B_FALSE);
1007                 ASSERT(vnode_refed == B_FALSE);
1008
1009                 lock_flags = base_lock_flags;
1010
1011                 /*
1012                  * There were no inodes in the list, just break out
1013                  * of the loop.
1014                  */
1015                 if (ip == NULL) {
1016                         break;
1017                 }
1018
1019                 /*
1020                  * We found another sync thread marker - skip it
1021                  */
1022                 if (ip->i_mount == NULL) {
1023                         ip = ip->i_mnext;
1024                         continue;
1025                 }
1026
1027                 vp = XFS_ITOV_NULL(ip);
1028
1029                 /*
1030                  * If the vnode is gone then this is being torn down,
1031                  * call reclaim if it is flushed, else let regular flush
1032                  * code deal with it later in the loop.
1033                  */
1034
1035                 if (vp == NULL) {
1036                         /* Skip ones already in reclaim */
1037                         if (ip->i_flags & XFS_IRECLAIM) {
1038                                 ip = ip->i_mnext;
1039                                 continue;
1040                         }
1041                         if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0) {
1042                                 ip = ip->i_mnext;
1043                         } else if ((xfs_ipincount(ip) == 0) &&
1044                                     xfs_iflock_nowait(ip)) {
1045                                 IPOINTER_INSERT(ip, mp);
1046
1047                                 xfs_finish_reclaim(ip, 1,
1048                                                 XFS_IFLUSH_DELWRI_ELSE_ASYNC);
1049
1050                                 XFS_MOUNT_ILOCK(mp);
1051                                 mount_locked = B_TRUE;
1052                                 IPOINTER_REMOVE(ip, mp);
1053                         } else {
1054                                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1055                                 ip = ip->i_mnext;
1056                         }
1057                         continue;
1058                 }
1059
1060                 if (VN_BAD(vp)) {
1061                         ip = ip->i_mnext;
1062                         continue;
1063                 }
1064
1065                 if (XFS_FORCED_SHUTDOWN(mp) && !(flags & SYNC_CLOSE)) {
1066                         XFS_MOUNT_IUNLOCK(mp);
1067                         kmem_free(ipointer, sizeof(xfs_iptr_t));
1068                         return 0;
1069                 }
1070
1071                 /*
1072                  * If this is just vfs_sync() or pflushd() calling
1073                  * then we can skip inodes for which it looks like
1074                  * there is nothing to do.  Since we don't have the
1075                  * inode locked this is racey, but these are periodic
1076                  * calls so it doesn't matter.  For the others we want
1077                  * to know for sure, so we at least try to lock them.
1078                  */
1079                 if (flags & SYNC_BDFLUSH) {
1080                         if (((ip->i_itemp == NULL) ||
1081                              !(ip->i_itemp->ili_format.ilf_fields &
1082                                XFS_ILOG_ALL)) &&
1083                             (ip->i_update_core == 0)) {
1084                                 ip = ip->i_mnext;
1085                                 continue;
1086                         }
1087                 }
1088
1089                 /*
1090                  * Try to lock without sleeping.  We're out of order with
1091                  * the inode list lock here, so if we fail we need to drop
1092                  * the mount lock and try again.  If we're called from
1093                  * bdflush() here, then don't bother.
1094                  *
1095                  * The inode lock here actually coordinates with the
1096                  * almost spurious inode lock in xfs_ireclaim() to prevent
1097                  * the vnode we handle here without a reference from
1098                  * being freed while we reference it.  If we lock the inode
1099                  * while it's on the mount list here, then the spurious inode
1100                  * lock in xfs_ireclaim() after the inode is pulled from
1101                  * the mount list will sleep until we release it here.
1102                  * This keeps the vnode from being freed while we reference
1103                  * it.
1104                  */
1105                 if (xfs_ilock_nowait(ip, lock_flags) == 0) {
1106                         if ((flags & SYNC_BDFLUSH) || (vp == NULL)) {
1107                                 ip = ip->i_mnext;
1108                                 continue;
1109                         }
1110
1111                         vp = vn_grab(vp);
1112                         if (vp == NULL) {
1113                                 ip = ip->i_mnext;
1114                                 continue;
1115                         }
1116
1117                         IPOINTER_INSERT(ip, mp);
1118                         xfs_ilock(ip, lock_flags);
1119
1120                         ASSERT(vp == XFS_ITOV(ip));
1121                         ASSERT(ip->i_mount == mp);
1122
1123                         vnode_refed = B_TRUE;
1124                 }
1125
1126                 /* From here on in the loop we may have a marker record
1127                  * in the inode list.
1128                  */
1129
1130                 if ((flags & SYNC_CLOSE)  && (vp != NULL)) {
1131                         /*
1132                          * This is the shutdown case.  We just need to
1133                          * flush and invalidate all the pages associated
1134                          * with the inode.  Drop the inode lock since
1135                          * we can't hold it across calls to the buffer
1136                          * cache.
1137                          *
1138                          * We don't set the VREMAPPING bit in the vnode
1139                          * here, because we don't hold the vnode lock
1140                          * exclusively.  It doesn't really matter, though,
1141                          * because we only come here when we're shutting
1142                          * down anyway.
1143                          */
1144                         xfs_iunlock(ip, XFS_ILOCK_SHARED);
1145
1146                         if (XFS_FORCED_SHUTDOWN(mp)) {
1147                                 VOP_TOSS_PAGES(vp, 0, -1, FI_REMAPF);
1148                         } else {
1149                                 VOP_FLUSHINVAL_PAGES(vp, 0, -1, FI_REMAPF);
1150                         }
1151
1152                         xfs_ilock(ip, XFS_ILOCK_SHARED);
1153
1154                 } else if ((flags & SYNC_DELWRI) && (vp != NULL)) {
1155                         if (VN_DIRTY(vp)) {
1156                                 /* We need to have dropped the lock here,
1157                                  * so insert a marker if we have not already
1158                                  * done so.
1159                                  */
1160                                 if (mount_locked) {
1161                                         IPOINTER_INSERT(ip, mp);
1162                                 }
1163
1164                                 /*
1165                                  * Drop the inode lock since we can't hold it
1166                                  * across calls to the buffer cache.
1167                                  */
1168                                 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1169                                 VOP_FLUSH_PAGES(vp, (xfs_off_t)0, -1,
1170                                                         fflag, FI_NONE, error);
1171                                 xfs_ilock(ip, XFS_ILOCK_SHARED);
1172                         }
1173
1174                 }
1175
1176                 if (flags & SYNC_BDFLUSH) {
1177                         if ((flags & SYNC_ATTR) &&
1178                             ((ip->i_update_core) ||
1179                              ((ip->i_itemp != NULL) &&
1180                               (ip->i_itemp->ili_format.ilf_fields != 0)))) {
1181
1182                                 /* Insert marker and drop lock if not already
1183                                  * done.
1184                                  */
1185                                 if (mount_locked) {
1186                                         IPOINTER_INSERT(ip, mp);
1187                                 }
1188
1189                                 /*
1190                                  * We don't want the periodic flushing of the
1191                                  * inodes by vfs_sync() to interfere with
1192                                  * I/O to the file, especially read I/O
1193                                  * where it is only the access time stamp
1194                                  * that is being flushed out.  To prevent
1195                                  * long periods where we have both inode
1196                                  * locks held shared here while reading the
1197                                  * inode's buffer in from disk, we drop the
1198                                  * inode lock while reading in the inode
1199                                  * buffer.  We have to release the buffer
1200                                  * and reacquire the inode lock so that they
1201                                  * are acquired in the proper order (inode
1202                                  * locks first).  The buffer will go at the
1203                                  * end of the lru chain, though, so we can
1204                                  * expect it to still be there when we go
1205                                  * for it again in xfs_iflush().
1206                                  */
1207                                 if ((xfs_ipincount(ip) == 0) &&
1208                                     xfs_iflock_nowait(ip)) {
1209
1210                                         xfs_ifunlock(ip);
1211                                         xfs_iunlock(ip, XFS_ILOCK_SHARED);
1212
1213                                         error = xfs_itobp(mp, NULL, ip,
1214                                                           &dip, &bp, 0);
1215                                         if (!error) {
1216                                                 xfs_buf_relse(bp);
1217                                         } else {
1218                                                 /* Bailing out, remove the
1219                                                  * marker and free it.
1220                                                  */
1221                                                 XFS_MOUNT_ILOCK(mp);
1222
1223                                                 IPOINTER_REMOVE(ip, mp);
1224
1225                                                 XFS_MOUNT_IUNLOCK(mp);
1226
1227                                                 ASSERT(!(lock_flags &
1228                                                         XFS_IOLOCK_SHARED));
1229
1230                                                 kmem_free(ipointer,
1231                                                         sizeof(xfs_iptr_t));
1232                                                 return (0);
1233                                         }
1234
1235                                         /*
1236                                          * Since we dropped the inode lock,
1237                                          * the inode may have been reclaimed.
1238                                          * Therefore, we reacquire the mount
1239                                          * lock and check to see if we were the
1240                                          * inode reclaimed. If this happened
1241                                          * then the ipointer marker will no
1242                                          * longer point back at us. In this
1243                                          * case, move ip along to the inode
1244                                          * after the marker, remove the marker
1245                                          * and continue.
1246                                          */
1247                                         XFS_MOUNT_ILOCK(mp);
1248                                         mount_locked = B_TRUE;
1249
1250                                         if (ip != ipointer->ip_mprev) {
1251                                                 IPOINTER_REMOVE(ip, mp);
1252
1253                                                 ASSERT(!vnode_refed);
1254                                                 ASSERT(!(lock_flags &
1255                                                         XFS_IOLOCK_SHARED));
1256                                                 continue;
1257                                         }
1258
1259                                         ASSERT(ip->i_mount == mp);
1260
1261                                         if (xfs_ilock_nowait(ip,
1262                                                     XFS_ILOCK_SHARED) == 0) {
1263                                                 ASSERT(ip->i_mount == mp);
1264                                                 /*
1265                                                  * We failed to reacquire
1266                                                  * the inode lock without
1267                                                  * sleeping, so just skip
1268                                                  * the inode for now.  We
1269                                                  * clear the ILOCK bit from
1270                                                  * the lock_flags so that we
1271                                                  * won't try to drop a lock
1272                                                  * we don't hold below.
1273                                                  */
1274                                                 lock_flags &= ~XFS_ILOCK_SHARED;
1275                                                 IPOINTER_REMOVE(ip_next, mp);
1276                                         } else if ((xfs_ipincount(ip) == 0) &&
1277                                                    xfs_iflock_nowait(ip)) {
1278                                                 ASSERT(ip->i_mount == mp);
1279                                                 /*
1280                                                  * Since this is vfs_sync()
1281                                                  * calling we only flush the
1282                                                  * inode out if we can lock
1283                                                  * it without sleeping and
1284                                                  * it is not pinned.  Drop
1285                                                  * the mount lock here so
1286                                                  * that we don't hold it for
1287                                                  * too long. We already have
1288                                                  * a marker in the list here.
1289                                                  */
1290                                                 XFS_MOUNT_IUNLOCK(mp);
1291                                                 mount_locked = B_FALSE;
1292                                                 error = xfs_iflush(ip,
1293                                                            XFS_IFLUSH_DELWRI);
1294                                         } else {
1295                                                 ASSERT(ip->i_mount == mp);
1296                                                 IPOINTER_REMOVE(ip_next, mp);
1297                                         }
1298                                 }
1299
1300                         }
1301
1302                 } else {
1303                         if ((flags & SYNC_ATTR) &&
1304                             ((ip->i_update_core) ||
1305                              ((ip->i_itemp != NULL) &&
1306                               (ip->i_itemp->ili_format.ilf_fields != 0)))) {
1307                                 if (mount_locked) {
1308                                         IPOINTER_INSERT(ip, mp);
1309                                 }
1310
1311                                 if (flags & SYNC_WAIT) {
1312                                         xfs_iflock(ip);
1313                                         error = xfs_iflush(ip,
1314                                                            XFS_IFLUSH_SYNC);
1315                                 } else {
1316                                         /*
1317                                          * If we can't acquire the flush
1318                                          * lock, then the inode is already
1319                                          * being flushed so don't bother
1320                                          * waiting.  If we can lock it then
1321                                          * do a delwri flush so we can
1322                                          * combine multiple inode flushes
1323                                          * in each disk write.
1324                                          */
1325                                         if (xfs_iflock_nowait(ip)) {
1326                                                 error = xfs_iflush(ip,
1327                                                            XFS_IFLUSH_DELWRI);
1328                                         }
1329                                         else if (bypassed)
1330                                                 (*bypassed)++;
1331                                 }
1332                         }
1333                 }
1334
1335                 if (lock_flags != 0) {
1336                         xfs_iunlock(ip, lock_flags);
1337                 }
1338
1339                 if (vnode_refed) {
1340                         /*
1341                          * If we had to take a reference on the vnode
1342                          * above, then wait until after we've unlocked
1343                          * the inode to release the reference.  This is
1344                          * because we can be already holding the inode
1345                          * lock when VN_RELE() calls xfs_inactive().
1346                          *
1347                          * Make sure to drop the mount lock before calling
1348                          * VN_RELE() so that we don't trip over ourselves if
1349                          * we have to go for the mount lock again in the
1350                          * inactive code.
1351                          */
1352                         if (mount_locked) {
1353                                 IPOINTER_INSERT(ip, mp);
1354                         }
1355
1356                         VN_RELE(vp);
1357
1358                         vnode_refed = B_FALSE;
1359                 }
1360
1361                 if (error) {
1362                         last_error = error;
1363                 }
1364
1365                 /*
1366                  * bail out if the filesystem is corrupted.
1367                  */
1368                 if (error == EFSCORRUPTED)  {
1369                         if (!mount_locked) {
1370                                 XFS_MOUNT_ILOCK(mp);
1371                                 IPOINTER_REMOVE(ip, mp);
1372                         }
1373                         XFS_MOUNT_IUNLOCK(mp);
1374                         ASSERT(ipointer_in == B_FALSE);
1375                         kmem_free(ipointer, sizeof(xfs_iptr_t));
1376                         return XFS_ERROR(error);
1377                 }
1378
1379                 /* Let other threads have a chance at the mount lock
1380                  * if we have looped many times without dropping the
1381                  * lock.
1382                  */
1383                 if ((++preempt & XFS_PREEMPT_MASK) == 0) {
1384                         if (mount_locked) {
1385                                 IPOINTER_INSERT(ip, mp);
1386                         }
1387                 }
1388
1389                 if (mount_locked == B_FALSE) {
1390                         XFS_MOUNT_ILOCK(mp);
1391                         mount_locked = B_TRUE;
1392                         IPOINTER_REMOVE(ip, mp);
1393                         continue;
1394                 }
1395
1396                 ASSERT(ipointer_in == B_FALSE);
1397                 ip = ip->i_mnext;
1398
1399         } while (ip != mp->m_inodes);
1400
1401         XFS_MOUNT_IUNLOCK(mp);
1402
1403         ASSERT(ipointer_in == B_FALSE);
1404
1405         kmem_free(ipointer, sizeof(xfs_iptr_t));
1406         return XFS_ERROR(last_error);
1407 }
1408
1409 /*
1410  * xfs sync routine for internal use
1411  *
1412  * This routine supports all of the flags defined for the generic VFS_SYNC
1413  * interface as explained above under xfs_sync.  In the interests of not
1414  * changing interfaces within the 6.5 family, additional internallly-
1415  * required functions are specified within a separate xflags parameter,
1416  * only available by calling this routine.
1417  *
1418  */
1419 int
1420 xfs_syncsub(
1421         xfs_mount_t     *mp,
1422         int             flags,
1423         int             xflags,
1424         int             *bypassed)
1425 {
1426         int             error = 0;
1427         int             last_error = 0;
1428         uint            log_flags = XFS_LOG_FORCE;
1429         xfs_buf_t       *bp;
1430         xfs_buf_log_item_t      *bip;
1431
1432         /*
1433          * Sync out the log.  This ensures that the log is periodically
1434          * flushed even if there is not enough activity to fill it up.
1435          */
1436         if (flags & SYNC_WAIT)
1437                 log_flags |= XFS_LOG_SYNC;
1438
1439         xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1440
1441         if (flags & (SYNC_ATTR|SYNC_DELWRI)) {
1442                 if (flags & SYNC_BDFLUSH)
1443                         xfs_finish_reclaim_all(mp, 1);
1444                 else
1445                         error = xfs_sync_inodes(mp, flags, xflags, bypassed);
1446         }
1447
1448         /*
1449          * Flushing out dirty data above probably generated more
1450          * log activity, so if this isn't vfs_sync() then flush
1451          * the log again.
1452          */
1453         if (flags & SYNC_DELWRI) {
1454                 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1455         }
1456
1457         if (flags & SYNC_FSDATA) {
1458                 /*
1459                  * If this is vfs_sync() then only sync the superblock
1460                  * if we can lock it without sleeping and it is not pinned.
1461                  */
1462                 if (flags & SYNC_BDFLUSH) {
1463                         bp = xfs_getsb(mp, XFS_BUF_TRYLOCK);
1464                         if (bp != NULL) {
1465                                 bip = XFS_BUF_FSPRIVATE(bp,xfs_buf_log_item_t*);
1466                                 if ((bip != NULL) &&
1467                                     xfs_buf_item_dirty(bip)) {
1468                                         if (!(XFS_BUF_ISPINNED(bp))) {
1469                                                 XFS_BUF_ASYNC(bp);
1470                                                 error = xfs_bwrite(mp, bp);
1471                                         } else {
1472                                                 xfs_buf_relse(bp);
1473                                         }
1474                                 } else {
1475                                         xfs_buf_relse(bp);
1476                                 }
1477                         }
1478                 } else {
1479                         bp = xfs_getsb(mp, 0);
1480                         /*
1481                          * If the buffer is pinned then push on the log so
1482                          * we won't get stuck waiting in the write for
1483                          * someone, maybe ourselves, to flush the log.
1484                          * Even though we just pushed the log above, we
1485                          * did not have the superblock buffer locked at
1486                          * that point so it can become pinned in between
1487                          * there and here.
1488                          */
1489                         if (XFS_BUF_ISPINNED(bp))
1490                                 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
1491                         if (flags & SYNC_WAIT)
1492                                 XFS_BUF_UNASYNC(bp);
1493                         else
1494                                 XFS_BUF_ASYNC(bp);
1495                         error = xfs_bwrite(mp, bp);
1496                 }
1497                 if (error) {
1498                         last_error = error;
1499                 }
1500         }
1501
1502         /*
1503          * If this is the periodic sync, then kick some entries out of
1504          * the reference cache.  This ensures that idle entries are
1505          * eventually kicked out of the cache.
1506          */
1507         if (flags & SYNC_REFCACHE) {
1508                 xfs_refcache_purge_some(mp);
1509         }
1510
1511         /*
1512          * Now check to see if the log needs a "dummy" transaction.
1513          */
1514
1515         if (!(flags & SYNC_REMOUNT) && xfs_log_need_covered(mp)) {
1516                 xfs_trans_t *tp;
1517                 xfs_inode_t *ip;
1518
1519                 /*
1520                  * Put a dummy transaction in the log to tell
1521                  * recovery that all others are OK.
1522                  */
1523                 tp = xfs_trans_alloc(mp, XFS_TRANS_DUMMY1);
1524                 if ((error = xfs_trans_reserve(tp, 0,
1525                                 XFS_ICHANGE_LOG_RES(mp),
1526                                 0, 0, 0)))  {
1527                         xfs_trans_cancel(tp, 0);
1528                         return error;
1529                 }
1530
1531                 ip = mp->m_rootip;
1532                 xfs_ilock(ip, XFS_ILOCK_EXCL);
1533
1534                 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1535                 xfs_trans_ihold(tp, ip);
1536                 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1537                 error = xfs_trans_commit(tp, 0, NULL);
1538                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1539                 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1540         }
1541
1542         /*
1543          * When shutting down, we need to insure that the AIL is pushed
1544          * to disk or the filesystem can appear corrupt from the PROM.
1545          */
1546         if ((flags & (SYNC_CLOSE|SYNC_WAIT)) == (SYNC_CLOSE|SYNC_WAIT)) {
1547                 XFS_bflush(mp->m_ddev_targp);
1548                 if (mp->m_rtdev_targp) {
1549                         XFS_bflush(mp->m_rtdev_targp);
1550                 }
1551         }
1552
1553         return XFS_ERROR(last_error);
1554 }
1555
1556 /*
1557  * xfs_vget - called by DMAPI and NFSD to get vnode from file handle
1558  */
1559 STATIC int
1560 xfs_vget(
1561         bhv_desc_t      *bdp,
1562         vnode_t         **vpp,
1563         fid_t           *fidp)
1564 {
1565         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
1566         xfs_fid_t       *xfid = (struct xfs_fid *)fidp;
1567         xfs_inode_t     *ip;
1568         int             error;
1569         xfs_ino_t       ino;
1570         unsigned int    igen;
1571
1572         /*
1573          * Invalid.  Since handles can be created in user space and passed in
1574          * via gethandle(), this is not cause for a panic.
1575          */
1576         if (xfid->xfs_fid_len != sizeof(*xfid) - sizeof(xfid->xfs_fid_len))
1577                 return XFS_ERROR(EINVAL);
1578
1579         ino  = xfid->xfs_fid_ino;
1580         igen = xfid->xfs_fid_gen;
1581
1582         /*
1583          * NFS can sometimes send requests for ino 0.  Fail them gracefully.
1584          */
1585         if (ino == 0)
1586                 return XFS_ERROR(ESTALE);
1587
1588         error = xfs_iget(mp, NULL, ino, 0, XFS_ILOCK_SHARED, &ip, 0);
1589         if (error) {
1590                 *vpp = NULL;
1591                 return error;
1592         }
1593
1594         if (ip == NULL) {
1595                 *vpp = NULL;
1596                 return XFS_ERROR(EIO);
1597         }
1598
1599         if (ip->i_d.di_mode == 0 || ip->i_d.di_gen != igen) {
1600                 xfs_iput_new(ip, XFS_ILOCK_SHARED);
1601                 *vpp = NULL;
1602                 return XFS_ERROR(ENOENT);
1603         }
1604
1605         *vpp = XFS_ITOV(ip);
1606         xfs_iunlock(ip, XFS_ILOCK_SHARED);
1607         return 0;
1608 }
1609
1610
1611 #define MNTOPT_LOGBUFS  "logbufs"       /* number of XFS log buffers */
1612 #define MNTOPT_LOGBSIZE "logbsize"      /* size of XFS log buffers */
1613 #define MNTOPT_LOGDEV   "logdev"        /* log device */
1614 #define MNTOPT_RTDEV    "rtdev"         /* realtime I/O device */
1615 #define MNTOPT_BIOSIZE  "biosize"       /* log2 of preferred buffered io size */
1616 #define MNTOPT_WSYNC    "wsync"         /* safe-mode nfs compatible mount */
1617 #define MNTOPT_INO64    "ino64"         /* force inodes into 64-bit range */
1618 #define MNTOPT_NOALIGN  "noalign"       /* turn off stripe alignment */
1619 #define MNTOPT_SWALLOC  "swalloc"       /* turn on stripe width allocation */
1620 #define MNTOPT_SUNIT    "sunit"         /* data volume stripe unit */
1621 #define MNTOPT_SWIDTH   "swidth"        /* data volume stripe width */
1622 #define MNTOPT_NOUUID   "nouuid"        /* ignore filesystem UUID */
1623 #define MNTOPT_MTPT     "mtpt"          /* filesystem mount point */
1624 #define MNTOPT_GRPID    "grpid"         /* group-ID from parent directory */
1625 #define MNTOPT_NOGRPID  "nogrpid"       /* group-ID from current process */
1626 #define MNTOPT_BSDGROUPS    "bsdgroups"    /* group-ID from parent directory */
1627 #define MNTOPT_SYSVGROUPS   "sysvgroups"   /* group-ID from current process */
1628 #define MNTOPT_ALLOCSIZE    "allocsize"    /* preferred allocation size */
1629 #define MNTOPT_IHASHSIZE    "ihashsize"    /* size of inode hash table */
1630 #define MNTOPT_NORECOVERY   "norecovery"   /* don't run XFS recovery */
1631 #define MNTOPT_NOLOGFLUSH   "nologflush"   /* don't hard flush on log writes */
1632 #define MNTOPT_OSYNCISOSYNC "osyncisosync" /* o_sync is REALLY o_sync */
1633 #define MNTOPT_64BITINODE   "inode64"   /* inodes can be allocated anywhere */
1634 #define MNTOPT_IKEEP    "ikeep"         /* do not free empty inode clusters */
1635 #define MNTOPT_NOIKEEP  "noikeep"       /* free empty inode clusters */
1636
1637 STATIC unsigned long
1638 suffix_strtoul(const char *cp, char **endp, unsigned int base)
1639 {
1640         int     last, shift_left_factor = 0;
1641         char    *value = (char *)cp;
1642
1643         last = strlen(value) - 1;
1644         if (value[last] == 'K' || value[last] == 'k') {
1645                 shift_left_factor = 10;
1646                 value[last] = '\0';
1647         }
1648         if (value[last] == 'M' || value[last] == 'm') {
1649                 shift_left_factor = 20;
1650                 value[last] = '\0';
1651         }
1652         if (value[last] == 'G' || value[last] == 'g') {
1653                 shift_left_factor = 30;
1654                 value[last] = '\0';
1655         }
1656
1657         return simple_strtoul(cp, endp, base) << shift_left_factor;
1658 }
1659
1660 STATIC int
1661 xfs_parseargs(
1662         struct bhv_desc         *bhv,
1663         char                    *options,
1664         struct xfs_mount_args   *args,
1665         int                     update)
1666 {
1667         struct vfs              *vfsp = bhvtovfs(bhv);
1668         char                    *this_char, *value, *eov;
1669         int                     dsunit, dswidth, vol_dsunit, vol_dswidth;
1670         int                     iosize;
1671
1672 #if 0   /* XXX: off by default, until some remaining issues ironed out */
1673         args->flags |= XFSMNT_IDELETE; /* default to on */
1674 #endif
1675
1676         if (!options)
1677                 return 0;
1678
1679         iosize = dsunit = dswidth = vol_dsunit = vol_dswidth = 0;
1680
1681         while ((this_char = strsep(&options, ",")) != NULL) {
1682                 if (!*this_char)
1683                         continue;
1684                 if ((value = strchr(this_char, '=')) != NULL)
1685                         *value++ = 0;
1686
1687                 if (!strcmp(this_char, MNTOPT_LOGBUFS)) {
1688                         if (!value || !*value) {
1689                                 printk("XFS: %s option requires an argument\n",
1690                                         this_char);
1691                                 return EINVAL;
1692                         }
1693                         args->logbufs = simple_strtoul(value, &eov, 10);
1694                 } else if (!strcmp(this_char, MNTOPT_LOGBSIZE)) {
1695                         if (!value || !*value) {
1696                                 printk("XFS: %s option requires an argument\n",
1697                                         this_char);
1698                                 return EINVAL;
1699                         }
1700                         args->logbufsize = suffix_strtoul(value, &eov, 10);
1701                 } else if (!strcmp(this_char, MNTOPT_LOGDEV)) {
1702                         if (!value || !*value) {
1703                                 printk("XFS: %s option requires an argument\n",
1704                                         this_char);
1705                                 return EINVAL;
1706                         }
1707                         strncpy(args->logname, value, MAXNAMELEN);
1708                 } else if (!strcmp(this_char, MNTOPT_MTPT)) {
1709                         if (!value || !*value) {
1710                                 printk("XFS: %s option requires an argument\n",
1711                                         this_char);
1712                                 return EINVAL;
1713                         }
1714                         strncpy(args->mtpt, value, MAXNAMELEN);
1715                 } else if (!strcmp(this_char, MNTOPT_RTDEV)) {
1716                         if (!value || !*value) {
1717                                 printk("XFS: %s option requires an argument\n",
1718                                         this_char);
1719                                 return EINVAL;
1720                         }
1721                         strncpy(args->rtname, value, MAXNAMELEN);
1722                 } else if (!strcmp(this_char, MNTOPT_BIOSIZE)) {
1723                         if (!value || !*value) {
1724                                 printk("XFS: %s option requires an argument\n",
1725                                         this_char);
1726                                 return EINVAL;
1727                         }
1728                         iosize = simple_strtoul(value, &eov, 10);
1729                         args->flags |= XFSMNT_IOSIZE;
1730                         args->iosizelog = (uint8_t) iosize;
1731                 } else if (!strcmp(this_char, MNTOPT_ALLOCSIZE)) {
1732                         if (!value || !*value) {
1733                                 printk("XFS: %s option requires an argument\n",
1734                                         this_char);
1735                                 return EINVAL;
1736                         }
1737                         iosize = suffix_strtoul(value, &eov, 10);
1738                         args->flags |= XFSMNT_IOSIZE;
1739                         args->iosizelog = ffs(iosize) - 1;
1740                 } else if (!strcmp(this_char, MNTOPT_IHASHSIZE)) {
1741                         if (!value || !*value) {
1742                                 printk("XFS: %s option requires an argument\n",
1743                                         this_char);
1744                                 return EINVAL;
1745                         }
1746                         args->flags |= XFSMNT_IHASHSIZE;
1747                         args->ihashsize = simple_strtoul(value, &eov, 10);
1748                 } else if (!strcmp(this_char, MNTOPT_GRPID) ||
1749                            !strcmp(this_char, MNTOPT_BSDGROUPS)) {
1750                         vfsp->vfs_flag |= VFS_GRPID;
1751                 } else if (!strcmp(this_char, MNTOPT_NOGRPID) ||
1752                            !strcmp(this_char, MNTOPT_SYSVGROUPS)) {
1753                         vfsp->vfs_flag &= ~VFS_GRPID;
1754                 } else if (!strcmp(this_char, MNTOPT_WSYNC)) {
1755                         args->flags |= XFSMNT_WSYNC;
1756                 } else if (!strcmp(this_char, MNTOPT_OSYNCISOSYNC)) {
1757                         args->flags |= XFSMNT_OSYNCISOSYNC;
1758                 } else if (!strcmp(this_char, MNTOPT_NORECOVERY)) {
1759                         args->flags |= XFSMNT_NORECOVERY;
1760                 } else if (!strcmp(this_char, MNTOPT_INO64)) {
1761                         args->flags |= XFSMNT_INO64;
1762 #if !XFS_BIG_INUMS
1763                         printk("XFS: %s option not allowed on this system\n",
1764                                 this_char);
1765                         return EINVAL;
1766 #endif
1767                 } else if (!strcmp(this_char, MNTOPT_NOALIGN)) {
1768                         args->flags |= XFSMNT_NOALIGN;
1769                 } else if (!strcmp(this_char, MNTOPT_SWALLOC)) {
1770                         args->flags |= XFSMNT_SWALLOC;
1771                 } else if (!strcmp(this_char, MNTOPT_SUNIT)) {
1772                         if (!value || !*value) {
1773                                 printk("XFS: %s option requires an argument\n",
1774                                         this_char);
1775                                 return EINVAL;
1776                         }
1777                         dsunit = simple_strtoul(value, &eov, 10);
1778                 } else if (!strcmp(this_char, MNTOPT_SWIDTH)) {
1779                         if (!value || !*value) {
1780                                 printk("XFS: %s option requires an argument\n",
1781                                         this_char);
1782                                 return EINVAL;
1783                         }
1784                         dswidth = simple_strtoul(value, &eov, 10);
1785                 } else if (!strcmp(this_char, MNTOPT_64BITINODE)) {
1786                         args->flags &= ~XFSMNT_32BITINODES;
1787 #if !XFS_BIG_INUMS
1788                         printk("XFS: %s option not allowed on this system\n",
1789                                 this_char);
1790                         return EINVAL;
1791 #endif
1792                 } else if (!strcmp(this_char, MNTOPT_NOUUID)) {
1793                         args->flags |= XFSMNT_NOUUID;
1794                 } else if (!strcmp(this_char, MNTOPT_NOLOGFLUSH)) {
1795                         args->flags |= XFSMNT_NOLOGFLUSH;
1796                 } else if (!strcmp(this_char, MNTOPT_IKEEP)) {
1797                         args->flags &= ~XFSMNT_IDELETE;
1798                 } else if (!strcmp(this_char, MNTOPT_NOIKEEP)) {
1799                         args->flags |= XFSMNT_IDELETE;
1800                 } else if (!strcmp(this_char, "osyncisdsync")) {
1801                         /* no-op, this is now the default */
1802 printk("XFS: osyncisdsync is now the default, option is deprecated.\n");
1803                 } else if (!strcmp(this_char, "irixsgid")) {
1804 printk("XFS: irixsgid is now a sysctl(2) variable, option is deprecated.\n");
1805                 } else {
1806                         printk("XFS: unknown mount option [%s].\n", this_char);
1807                         return EINVAL;
1808                 }
1809         }
1810
1811         if (args->flags & XFSMNT_NORECOVERY) {
1812                 if ((vfsp->vfs_flag & VFS_RDONLY) == 0) {
1813                         printk("XFS: no-recovery mounts must be read-only.\n");
1814                         return EINVAL;
1815                 }
1816         }
1817
1818         if ((args->flags & XFSMNT_NOALIGN) && (dsunit || dswidth)) {
1819                 printk(
1820         "XFS: sunit and swidth options incompatible with the noalign option\n");
1821                 return EINVAL;
1822         }
1823
1824         if ((dsunit && !dswidth) || (!dsunit && dswidth)) {
1825                 printk("XFS: sunit and swidth must be specified together\n");
1826                 return EINVAL;
1827         }
1828
1829         if (dsunit && (dswidth % dsunit != 0)) {
1830                 printk(
1831         "XFS: stripe width (%d) must be a multiple of the stripe unit (%d)\n",
1832                         dswidth, dsunit);
1833                 return EINVAL;
1834         }
1835
1836         if ((args->flags & XFSMNT_NOALIGN) != XFSMNT_NOALIGN) {
1837                 if (dsunit) {
1838                         args->sunit = dsunit;
1839                         args->flags |= XFSMNT_RETERR;
1840                 } else {
1841                         args->sunit = vol_dsunit;
1842                 }
1843                 dswidth ? (args->swidth = dswidth) :
1844                           (args->swidth = vol_dswidth);
1845         } else {
1846                 args->sunit = args->swidth = 0;
1847         }
1848
1849         return 0;
1850 }
1851
1852 STATIC int
1853 xfs_showargs(
1854         struct bhv_desc         *bhv,
1855         struct seq_file         *m)
1856 {
1857         static struct proc_xfs_info {
1858                 int     flag;
1859                 char    *str;
1860         } xfs_info[] = {
1861                 /* the few simple ones we can get from the mount struct */
1862                 { XFS_MOUNT_WSYNC,              "," MNTOPT_WSYNC },
1863                 { XFS_MOUNT_INO64,              "," MNTOPT_INO64 },
1864                 { XFS_MOUNT_NOALIGN,            "," MNTOPT_NOALIGN },
1865                 { XFS_MOUNT_SWALLOC,            "," MNTOPT_SWALLOC },
1866                 { XFS_MOUNT_NOUUID,             "," MNTOPT_NOUUID },
1867                 { XFS_MOUNT_NORECOVERY,         "," MNTOPT_NORECOVERY },
1868                 { XFS_MOUNT_OSYNCISOSYNC,       "," MNTOPT_OSYNCISOSYNC },
1869                 { XFS_MOUNT_NOLOGFLUSH,         "," MNTOPT_NOLOGFLUSH },
1870                 { XFS_MOUNT_IDELETE,            "," MNTOPT_NOIKEEP },
1871                 { 0, NULL }
1872         };
1873         struct proc_xfs_info    *xfs_infop;
1874         struct xfs_mount        *mp = XFS_BHVTOM(bhv);
1875         struct vfs              *vfsp = XFS_MTOVFS(mp);
1876
1877         for (xfs_infop = xfs_info; xfs_infop->flag; xfs_infop++) {
1878                 if (mp->m_flags & xfs_infop->flag)
1879                         seq_puts(m, xfs_infop->str);
1880         }
1881
1882         if (mp->m_flags & XFS_MOUNT_IHASHSIZE)
1883                 seq_printf(m, "," MNTOPT_IHASHSIZE "=%d", mp->m_ihsize);
1884
1885         if (mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)
1886                 seq_printf(m, "," MNTOPT_ALLOCSIZE "=%d", 1<<mp->m_writeio_log);
1887
1888         if (mp->m_logbufs > 0)
1889                 seq_printf(m, "," MNTOPT_LOGBUFS "=%d", mp->m_logbufs);
1890
1891         if (mp->m_logbsize > 0)
1892                 seq_printf(m, "," MNTOPT_LOGBSIZE "=%d", mp->m_logbsize);
1893
1894         if (mp->m_ddev_targp != mp->m_logdev_targp)
1895                 seq_printf(m, "," MNTOPT_LOGDEV "=%s",
1896                                 XFS_BUFTARG_NAME(mp->m_logdev_targp));
1897
1898         if (mp->m_rtdev_targp && mp->m_ddev_targp != mp->m_rtdev_targp)
1899                 seq_printf(m, "," MNTOPT_RTDEV "=%s",
1900                                 XFS_BUFTARG_NAME(mp->m_rtdev_targp));
1901
1902         if (mp->m_dalign > 0)
1903                 seq_printf(m, "," MNTOPT_SUNIT "=%d",
1904                                 (int)XFS_FSB_TO_BB(mp, mp->m_dalign));
1905
1906         if (mp->m_swidth > 0)
1907                 seq_printf(m, "," MNTOPT_SWIDTH "=%d",
1908                                 (int)XFS_FSB_TO_BB(mp, mp->m_swidth));
1909
1910         if (!(mp->m_flags & XFS_MOUNT_32BITINOOPT))
1911                 seq_printf(m, "," MNTOPT_64BITINODE);
1912
1913         if (vfsp->vfs_flag & VFS_GRPID)
1914                 seq_printf(m, "," MNTOPT_GRPID);
1915
1916         return 0;
1917 }
1918
1919 STATIC void
1920 xfs_freeze(
1921         bhv_desc_t      *bdp)
1922 {
1923         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
1924
1925         while (atomic_read(&mp->m_active_trans) > 0)
1926                 delay(100);
1927
1928         /* Push the superblock and write an unmount record */
1929         xfs_log_unmount_write(mp);
1930         xfs_unmountfs_writesb(mp);
1931 }
1932
1933
1934 vfsops_t xfs_vfsops = {
1935         BHV_IDENTITY_INIT(VFS_BHV_XFS,VFS_POSITION_XFS),
1936         .vfs_parseargs          = xfs_parseargs,
1937         .vfs_showargs           = xfs_showargs,
1938         .vfs_mount              = xfs_mount,
1939         .vfs_unmount            = xfs_unmount,
1940         .vfs_mntupdate          = xfs_mntupdate,
1941         .vfs_root               = xfs_root,
1942         .vfs_statvfs            = xfs_statvfs,
1943         .vfs_sync               = xfs_sync,
1944         .vfs_vget               = xfs_vget,
1945         .vfs_dmapiops           = (vfs_dmapiops_t)fs_nosys,
1946         .vfs_quotactl           = (vfs_quotactl_t)fs_nosys,
1947         .vfs_init_vnode         = xfs_initialize_vnode,
1948         .vfs_force_shutdown     = xfs_do_force_shutdown,
1949         .vfs_freeze             = xfs_freeze,
1950 };