fs/xfs/xfs_vfsops.c

   1 /*
   2  * XFS filesystem operations.
   3  *
   4  * Copyright (c) 2000-2004 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         /*
 262          * Pull in the 'wsync' and 'ino64' mount options before we do the real
 263          * work of mounting and recovery.  The arg pointer will
 264          * be NULL when we are being called from the root mount code.
 265          */
 266         if (ap->flags & XFSMNT_WSYNC)
 267                 mp->m_flags |= XFS_MOUNT_WSYNC;
 268 #if XFS_BIG_INUMS
 269         if (ap->flags & XFSMNT_INO64) {
 270                 mp->m_flags |= XFS_MOUNT_INO64;
 271                 mp->m_inoadd = XFS_INO64_OFFSET;
 272         }
 273 #endif
 274         if (ap->flags & XFSMNT_NOATIME)
 275                 mp->m_flags |= XFS_MOUNT_NOATIME;
 276
 277         if (ap->flags & XFSMNT_RETERR)
 278                 mp->m_flags |= XFS_MOUNT_RETERR;
 279
 280         if (ap->flags & XFSMNT_NOALIGN)
 281                 mp->m_flags |= XFS_MOUNT_NOALIGN;
 282
 283         if (ap->flags & XFSMNT_SWALLOC)
 284                 mp->m_flags |= XFS_MOUNT_SWALLOC;
 285
 286         if (ap->flags & XFSMNT_OSYNCISOSYNC)
 287                 mp->m_flags |= XFS_MOUNT_OSYNCISOSYNC;
 288
 289         if (ap->flags & XFSMNT_32BITINODES)
 290                 mp->m_flags |= (XFS_MOUNT_32BITINODES | XFS_MOUNT_32BITINOOPT);
 291
 292         if (ap->flags & XFSMNT_IOSIZE) {
 293                 if (ap->iosizelog > XFS_MAX_IO_LOG ||
 294                     ap->iosizelog < XFS_MIN_IO_LOG) {
 295                         cmn_err(CE_WARN,
 296                 "XFS: invalid log iosize: %d [not %d-%d]",
 297                                 ap->iosizelog, XFS_MIN_IO_LOG,
 298                                 XFS_MAX_IO_LOG);
 299                         return XFS_ERROR(EINVAL);
 300                 }
 301
 302                 mp->m_flags |= XFS_MOUNT_DFLT_IOSIZE;
 303                 mp->m_readio_log = mp->m_writeio_log = ap->iosizelog;
 304         }
 305         if (ap->flags & XFSMNT_IDELETE)
 306                 mp->m_flags |= XFS_MOUNT_IDELETE;
 307
 308         /*
 309          * no recovery flag requires a read-only mount
 310          */
 311         if (ap->flags & XFSMNT_NORECOVERY) {
 312                 if (!(vfs->vfs_flag & VFS_RDONLY)) {
 313                         cmn_err(CE_WARN,
 314         "XFS: tried to mount a FS read-write without recovery!");
 315                         return XFS_ERROR(EINVAL);
 316                 }
 317                 mp->m_flags |= XFS_MOUNT_NORECOVERY;
 318         }
 319
 320         if (ap->flags & XFSMNT_NOUUID)
 321                 mp->m_flags |= XFS_MOUNT_NOUUID;
 322         if (ap->flags & XFSMNT_TAGXID)
 323                 mp->m_flags |= XFS_MOUNT_TAGXID;
 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          * disallow mount attempts with (IRIX) project quota enabled
 372          */
 373         if (XFS_SB_VERSION_HASQUOTA(&mp->m_sb) &&
 374             (mp->m_sb.sb_qflags & XFS_PQUOTA_ACCT)) {
 375                 cmn_err(CE_WARN,
 376         "XFS: cannot mount a filesystem with IRIX project quota enabled");
 377                 return XFS_ERROR(ENOSYS);
 378         }
 379
 380         /*
 381          * check for shared mount.
 382          */
 383         if (ap->flags & XFSMNT_SHARED) {
 384                 if (!XFS_SB_VERSION_HASSHARED(&mp->m_sb))
 385                         return XFS_ERROR(EINVAL);
 386
 387                 /*
 388                  * For IRIX 6.5, shared mounts must have the shared
 389                  * version bit set, have the persistent readonly
 390                  * field set, must be version 0 and can only be mounted
 391                  * read-only.
 392                  */
 393                 if (!ronly || !(mp->m_sb.sb_flags & XFS_SBF_READONLY) ||
 394                      (mp->m_sb.sb_shared_vn != 0))
 395                         return XFS_ERROR(EINVAL);
 396
 397                 mp->m_flags |= XFS_MOUNT_SHARED;
 398
 399                 /*
 400                  * Shared XFS V0 can't deal with DMI.  Return EINVAL.
 401                  */
 402                 if (mp->m_sb.sb_shared_vn == 0 && (ap->flags & XFSMNT_DMAPI))
 403                         return XFS_ERROR(EINVAL);
 404         }
 405
 406         if (ap->flags & XFSMNT_TAGXID)
 407                 vfs->vfs_super->s_flags |= MS_TAGXID;
 408         return 0;
 409 }
 410
 411 /*
 412  * xfs_mount
 413  *
 414  * The file system configurations are:
 415  *      (1) device (partition) with data and internal log
 416  *      (2) logical volume with data and log subvolumes.
 417  *      (3) logical volume with data, log, and realtime subvolumes.
 418  *
 419  * We only have to handle opening the log and realtime volumes here if
 420  * they are present.  The data subvolume has already been opened by
 421  * get_sb_bdev() and is stored in vfsp->vfs_super->s_bdev.
 422  */
 423 STATIC int
 424 xfs_mount(
 425         struct bhv_desc         *bhvp,
 426         struct xfs_mount_args   *args,
 427         cred_t                  *credp)
 428 {
 429         struct vfs              *vfsp = bhvtovfs(bhvp);
 430         struct bhv_desc         *p;
 431         struct xfs_mount        *mp = XFS_BHVTOM(bhvp);
 432         struct block_device     *ddev, *logdev, *rtdev;
 433         int                     flags = 0, error;
 434
 435         ddev = vfsp->vfs_super->s_bdev;
 436         logdev = rtdev = NULL;
 437
 438         /*
 439          * Setup xfs_mount function vectors from available behaviors
 440          */
 441         p = vfs_bhv_lookup(vfsp, VFS_POSITION_DM);
 442         mp->m_dm_ops = p ? *(xfs_dmops_t *) vfs_bhv_custom(p) : xfs_dmcore_stub;
 443         p = vfs_bhv_lookup(vfsp, VFS_POSITION_QM);
 444         mp->m_qm_ops = p ? *(xfs_qmops_t *) vfs_bhv_custom(p) : xfs_qmcore_stub;
 445         p = vfs_bhv_lookup(vfsp, VFS_POSITION_IO);
 446         mp->m_io_ops = p ? *(xfs_ioops_t *) vfs_bhv_custom(p) : xfs_iocore_xfs;
 447
 448         /*
 449          * Open real time and log devices - order is important.
 450          */
 451         if (args->logname[0]) {
 452                 error = xfs_blkdev_get(mp, args->logname, &logdev);
 453                 if (error)
 454                         return error;
 455         }
 456         if (args->rtname[0]) {
 457                 error = xfs_blkdev_get(mp, args->rtname, &rtdev);
 458                 if (error) {
 459                         xfs_blkdev_put(logdev);
 460                         return error;
 461                 }
 462
 463                 if (rtdev == ddev || rtdev == logdev) {
 464                         cmn_err(CE_WARN,
 465         "XFS: Cannot mount filesystem with identical rtdev and ddev/logdev.");
 466                         xfs_blkdev_put(logdev);
 467                         xfs_blkdev_put(rtdev);
 468                         return EINVAL;
 469                 }
 470         }
 471
 472         /*
 473          * Setup xfs_mount buffer target pointers
 474          */
 475         error = ENOMEM;
 476         mp->m_ddev_targp = xfs_alloc_buftarg(ddev, 0);
 477         if (!mp->m_ddev_targp) {
 478                 xfs_blkdev_put(logdev);
 479                 xfs_blkdev_put(rtdev);
 480                 return error;
 481         }
 482         if (rtdev) {
 483                 mp->m_rtdev_targp = xfs_alloc_buftarg(rtdev, 1);
 484                 if (!mp->m_rtdev_targp)
 485                         goto error0;
 486         }
 487         mp->m_logdev_targp = (logdev && logdev != ddev) ?
 488                                 xfs_alloc_buftarg(logdev, 1) : mp->m_ddev_targp;
 489         if (!mp->m_logdev_targp)
 490                 goto error0;
 491
 492         /*
 493          * Setup flags based on mount(2) options and then the superblock
 494          */
 495         error = xfs_start_flags(vfsp, args, mp);
 496         if (error)
 497                 goto error1;
 498         error = xfs_readsb(mp);
 499         if (error)
 500                 goto error1;
 501         error = xfs_finish_flags(vfsp, args, mp);
 502         if (error)
 503                 goto error2;
 504
 505         /*
 506          * Setup xfs_mount buffer target pointers based on superblock
 507          */
 508         error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_blocksize,
 509                                     mp->m_sb.sb_sectsize);
 510         if (!error && logdev && logdev != ddev) {
 511                 unsigned int    log_sector_size = BBSIZE;
 512
 513                 if (XFS_SB_VERSION_HASSECTOR(&mp->m_sb))
 514                         log_sector_size = mp->m_sb.sb_logsectsize;
 515                 error = xfs_setsize_buftarg(mp->m_logdev_targp,
 516                                             mp->m_sb.sb_blocksize,
 517                                             log_sector_size);
 518         }
 519         if (!error && rtdev)
 520                 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
 521                                             mp->m_sb.sb_blocksize,
 522                                             mp->m_sb.sb_sectsize);
 523         if (error)
 524                 goto error2;
 525
 526         error = XFS_IOINIT(vfsp, args, flags);
 527         if (!error)
 528                 return 0;
 529 error2:
 530         if (mp->m_sb_bp)
 531                 xfs_freesb(mp);
 532 error1:
 533         xfs_binval(mp->m_ddev_targp);
 534         if (logdev && logdev != ddev)
 535                 xfs_binval(mp->m_logdev_targp);
 536         if (rtdev)
 537                 xfs_binval(mp->m_rtdev_targp);
 538 error0:
 539         xfs_unmountfs_close(mp, credp);
 540         return error;
 541 }
 542
 543 STATIC int
 544 xfs_unmount(
 545         bhv_desc_t      *bdp,
 546         int             flags,
 547         cred_t          *credp)
 548 {
 549         struct vfs      *vfsp = bhvtovfs(bdp);
 550         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
 551         xfs_inode_t     *rip;
 552         vnode_t         *rvp;
 553         int             unmount_event_wanted = 0;
 554         int             unmount_event_flags = 0;
 555         int             xfs_unmountfs_needed = 0;
 556         int             error;
 557
 558         rip = mp->m_rootip;
 559         rvp = XFS_ITOV(rip);
 560
 561         if (vfsp->vfs_flag & VFS_DMI) {
 562                 error = XFS_SEND_PREUNMOUNT(mp, vfsp,
 563                                 rvp, DM_RIGHT_NULL, rvp, DM_RIGHT_NULL,
 564                                 NULL, NULL, 0, 0,
 565                                 (mp->m_dmevmask & (1<<DM_EVENT_PREUNMOUNT))?
 566                                         0:DM_FLAGS_UNWANTED);
 567                         if (error)
 568                                 return XFS_ERROR(error);
 569                 unmount_event_wanted = 1;
 570                 unmount_event_flags = (mp->m_dmevmask & (1<<DM_EVENT_UNMOUNT))?
 571                                         0 : DM_FLAGS_UNWANTED;
 572         }
 573
 574         /*
 575          * First blow any referenced inode from this file system
 576          * out of the reference cache, and delete the timer.
 577          */
 578         xfs_refcache_purge_mp(mp);
 579
 580         XFS_bflush(mp->m_ddev_targp);
 581         error = xfs_unmount_flush(mp, 0);
 582         if (error)
 583                 goto out;
 584
 585         ASSERT(vn_count(rvp) == 1);
 586
 587         /*
 588          * Drop the reference count
 589          */
 590         VN_RELE(rvp);
 591
 592         /*
 593          * If we're forcing a shutdown, typically because of a media error,
 594          * we want to make sure we invalidate dirty pages that belong to
 595          * referenced vnodes as well.
 596          */
 597         if (XFS_FORCED_SHUTDOWN(mp)) {
 598                 error = xfs_sync(&mp->m_bhv,
 599                          (SYNC_WAIT | SYNC_CLOSE), credp);
 600                 ASSERT(error != EFSCORRUPTED);
 601         }
 602         xfs_unmountfs_needed = 1;
 603
 604 out:
 605         /*      Send DMAPI event, if required.
 606          *      Then do xfs_unmountfs() if needed.
 607          *      Then return error (or zero).
 608          */
 609         if (unmount_event_wanted) {
 610                 /* Note: mp structure must still exist for
 611                  * XFS_SEND_UNMOUNT() call.
 612                  */
 613                 XFS_SEND_UNMOUNT(mp, vfsp, error == 0 ? rvp : NULL,
 614                         DM_RIGHT_NULL, 0, error, unmount_event_flags);
 615         }
 616         if (xfs_unmountfs_needed) {
 617                 /*
 618                  * Call common unmount function to flush to disk
 619                  * and free the super block buffer & mount structures.
 620                  */
 621                 xfs_unmountfs(mp, credp);
 622         }
 623
 624         return XFS_ERROR(error);
 625 }
 626
 627 #define REMOUNT_READONLY_FLAGS  (SYNC_REMOUNT|SYNC_ATTR|SYNC_WAIT)
 628
 629 STATIC int
 630 xfs_mntupdate(
 631         bhv_desc_t                      *bdp,
 632         int                             *flags,
 633         struct xfs_mount_args           *args)
 634 {
 635         struct vfs      *vfsp = bhvtovfs(bdp);
 636         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
 637         int             pincount, error;
 638         int             count = 0;
 639
 640         if (args->flags & XFSMNT_NOATIME)
 641                 mp->m_flags |= XFS_MOUNT_NOATIME;
 642         else
 643                 mp->m_flags &= ~XFS_MOUNT_NOATIME;
 644
 645         if (!(vfsp->vfs_flag & VFS_RDONLY)) {
 646                 VFS_SYNC(vfsp, SYNC_FSDATA|SYNC_BDFLUSH|SYNC_ATTR, NULL, error);
 647         }
 648
 649         if (*flags & MS_RDONLY) {
 650                 xfs_refcache_purge_mp(mp);
 651                 xfs_flush_buftarg(mp->m_ddev_targp, 0);
 652                 xfs_finish_reclaim_all(mp, 0);
 653
 654                 /* This loop must run at least twice.
 655                  * The first instance of the loop will flush
 656                  * most meta data but that will generate more
 657                  * meta data (typically directory updates).
 658                  * Which then must be flushed and logged before
 659                  * we can write the unmount record.
 660                  */
 661                 do {
 662                         VFS_SYNC(vfsp, REMOUNT_READONLY_FLAGS, NULL, error);
 663                         pincount = xfs_flush_buftarg(mp->m_ddev_targp, 1);
 664                         if (!pincount) {
 665                                 delay(50);
 666                                 count++;
 667                         }
 668                 } while (count < 2);
 669
 670                 /* Ok now write out an unmount record */
 671                 xfs_log_unmount_write(mp);
 672                 xfs_unmountfs_writesb(mp);
 673                 vfsp->vfs_flag |= VFS_RDONLY;
 674         } else {
 675                 vfsp->vfs_flag &= ~VFS_RDONLY;
 676         }
 677
 678         return 0;
 679 }
 680
 681 /*
 682  * xfs_unmount_flush implements a set of flush operation on special
 683  * inodes, which are needed as a separate set of operations so that
 684  * they can be called as part of relocation process.
 685  */
 686 int
 687 xfs_unmount_flush(
 688         xfs_mount_t     *mp,            /* Mount structure we are getting
 689                                            rid of. */
 690         int             relocation)     /* Called from vfs relocation. */
 691 {
 692         xfs_inode_t     *rip = mp->m_rootip;
 693         xfs_inode_t     *rbmip;
 694         xfs_inode_t     *rsumip = NULL;
 695         vnode_t         *rvp = XFS_ITOV(rip);
 696         int             error;
 697
 698         xfs_ilock(rip, XFS_ILOCK_EXCL);
 699         xfs_iflock(rip);
 700
 701         /*
 702          * Flush out the real time inodes.
 703          */
 704         if ((rbmip = mp->m_rbmip) != NULL) {
 705                 xfs_ilock(rbmip, XFS_ILOCK_EXCL);
 706                 xfs_iflock(rbmip);
 707                 error = xfs_iflush(rbmip, XFS_IFLUSH_SYNC);
 708                 xfs_iunlock(rbmip, XFS_ILOCK_EXCL);
 709
 710                 if (error == EFSCORRUPTED)
 711                         goto fscorrupt_out;
 712
 713                 ASSERT(vn_count(XFS_ITOV(rbmip)) == 1);
 714
 715                 rsumip = mp->m_rsumip;
 716                 xfs_ilock(rsumip, XFS_ILOCK_EXCL);
 717                 xfs_iflock(rsumip);
 718                 error = xfs_iflush(rsumip, XFS_IFLUSH_SYNC);
 719                 xfs_iunlock(rsumip, XFS_ILOCK_EXCL);
 720
 721                 if (error == EFSCORRUPTED)
 722                         goto fscorrupt_out;
 723
 724                 ASSERT(vn_count(XFS_ITOV(rsumip)) == 1);
 725         }
 726
 727         /*
 728          * Synchronously flush root inode to disk
 729          */
 730         error = xfs_iflush(rip, XFS_IFLUSH_SYNC);
 731         if (error == EFSCORRUPTED)
 732                 goto fscorrupt_out2;
 733
 734         if (vn_count(rvp) != 1 && !relocation) {
 735                 xfs_iunlock(rip, XFS_ILOCK_EXCL);
 736                 return XFS_ERROR(EBUSY);
 737         }
 738
 739         /*
 740          * Release dquot that rootinode, rbmino and rsumino might be holding,
 741          * flush and purge the quota inodes.
 742          */
 743         error = XFS_QM_UNMOUNT(mp);
 744         if (error == EFSCORRUPTED)
 745                 goto fscorrupt_out2;
 746
 747         if (rbmip) {
 748                 VN_RELE(XFS_ITOV(rbmip));
 749                 VN_RELE(XFS_ITOV(rsumip));
 750         }
 751
 752         xfs_iunlock(rip, XFS_ILOCK_EXCL);
 753         return 0;
 754
 755 fscorrupt_out:
 756         xfs_ifunlock(rip);
 757
 758 fscorrupt_out2:
 759         xfs_iunlock(rip, XFS_ILOCK_EXCL);
 760
 761         return XFS_ERROR(EFSCORRUPTED);
 762 }
 763
 764 /*
 765  * xfs_root extracts the root vnode from a vfs.
 766  *
 767  * vfsp -- the vfs struct for the desired file system
 768  * vpp  -- address of the caller's vnode pointer which should be
 769  *         set to the desired fs root vnode
 770  */
 771 STATIC int
 772 xfs_root(
 773         bhv_desc_t      *bdp,
 774         vnode_t         **vpp)
 775 {
 776         vnode_t         *vp;
 777
 778         vp = XFS_ITOV((XFS_BHVTOM(bdp))->m_rootip);
 779         VN_HOLD(vp);
 780         *vpp = vp;
 781         return 0;
 782 }
 783
 784 /*
 785  * xfs_statvfs
 786  *
 787  * Fill in the statvfs structure for the given file system.  We use
 788  * the superblock lock in the mount structure to ensure a consistent
 789  * snapshot of the counters returned.
 790  */
 791 STATIC int
 792 xfs_statvfs(
 793         bhv_desc_t      *bdp,
 794         xfs_statfs_t    *statp,
 795         vnode_t         *vp)
 796 {
 797         __uint64_t      fakeinos;
 798         xfs_extlen_t    lsize;
 799         xfs_mount_t     *mp;
 800         xfs_sb_t        *sbp;
 801         unsigned long   s;
 802         u64 id;
 803
 804         mp = XFS_BHVTOM(bdp);
 805         sbp = &(mp->m_sb);
 806
 807         statp->f_type = XFS_SB_MAGIC;
 808
 809         s = XFS_SB_LOCK(mp);
 810         statp->f_bsize = sbp->sb_blocksize;
 811         lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
 812         statp->f_blocks = sbp->sb_dblocks - lsize;
 813         statp->f_bfree = statp->f_bavail = sbp->sb_fdblocks;
 814         fakeinos = statp->f_bfree << sbp->sb_inopblog;
 815 #if XFS_BIG_INUMS
 816         fakeinos += mp->m_inoadd;
 817 #endif
 818         statp->f_files =
 819             MIN(sbp->sb_icount + fakeinos, (__uint64_t)XFS_MAXINUMBER);
 820         if (mp->m_maxicount)
 821 #if XFS_BIG_INUMS
 822                 if (!mp->m_inoadd)
 823 #endif
 824                         statp->f_files = min_t(typeof(statp->f_files),
 825                                                 statp->f_files,
 826                                                 mp->m_maxicount);
 827         statp->f_ffree = statp->f_files - (sbp->sb_icount - sbp->sb_ifree);
 828         XFS_SB_UNLOCK(mp, s);
 829
 830         id = huge_encode_dev(mp->m_dev);
 831         statp->f_fsid.val[0] = (u32)id;
 832         statp->f_fsid.val[1] = (u32)(id >> 32);
 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;
 885
 886         mp = XFS_BHVTOM(bdp);
 887         return (xfs_syncsub(mp, flags, 0, NULL));
 888 }
 889
 890 /*
 891  * xfs sync routine for internal use
 892  *
 893  * This routine supports all of the flags defined for the generic VFS_SYNC
 894  * interface as explained above under xfs_sync.  In the interests of not
 895  * changing interfaces within the 6.5 family, additional internallly-
 896  * required functions are specified within a separate xflags parameter,
 897  * only available by calling this routine.
 898  *
 899  */
 900 STATIC int
 901 xfs_sync_inodes(
 902         xfs_mount_t     *mp,
 903         int             flags,
 904         int             xflags,
 905         int             *bypassed)
 906 {
 907         xfs_inode_t     *ip = NULL;
 908         xfs_inode_t     *ip_next;
 909         xfs_buf_t       *bp;
 910         vnode_t         *vp = NULL;
 911         vmap_t          vmap;
 912         int             error;
 913         int             last_error;
 914         uint64_t        fflag;
 915         uint            lock_flags;
 916         uint            base_lock_flags;
 917         boolean_t       mount_locked;
 918         boolean_t       vnode_refed;
 919         int             preempt;
 920         xfs_dinode_t    *dip;
 921         xfs_iptr_t      *ipointer;
 922 #ifdef DEBUG
 923         boolean_t       ipointer_in = B_FALSE;
 924
 925 #define IPOINTER_SET    ipointer_in = B_TRUE
 926 #define IPOINTER_CLR    ipointer_in = B_FALSE
 927 #else
 928 #define IPOINTER_SET
 929 #define IPOINTER_CLR
 930 #endif
 931
 932
 933 /* Insert a marker record into the inode list after inode ip. The list
 934  * must be locked when this is called. After the call the list will no
 935  * longer be locked.
 936  */
 937 #define IPOINTER_INSERT(ip, mp) { \
 938                 ASSERT(ipointer_in == B_FALSE); \
 939                 ipointer->ip_mnext = ip->i_mnext; \
 940                 ipointer->ip_mprev = ip; \
 941                 ip->i_mnext = (xfs_inode_t *)ipointer; \
 942                 ipointer->ip_mnext->i_mprev = (xfs_inode_t *)ipointer; \
 943                 preempt = 0; \
 944                 XFS_MOUNT_IUNLOCK(mp); \
 945                 mount_locked = B_FALSE; \
 946                 IPOINTER_SET; \
 947         }
 948
 949 /* Remove the marker from the inode list. If the marker was the only item
 950  * in the list then there are no remaining inodes and we should zero out
 951  * the whole list. If we are the current head of the list then move the head
 952  * past us.
 953  */
 954 #define IPOINTER_REMOVE(ip, mp) { \
 955                 ASSERT(ipointer_in == B_TRUE); \
 956                 if (ipointer->ip_mnext != (xfs_inode_t *)ipointer) { \
 957                         ip = ipointer->ip_mnext; \
 958                         ip->i_mprev = ipointer->ip_mprev; \
 959                         ipointer->ip_mprev->i_mnext = ip; \
 960                         if (mp->m_inodes == (xfs_inode_t *)ipointer) { \
 961                                 mp->m_inodes = ip; \
 962                         } \
 963                 } else { \
 964                         ASSERT(mp->m_inodes == (xfs_inode_t *)ipointer); \
 965                         mp->m_inodes = NULL; \
 966                         ip = NULL; \
 967                 } \
 968                 IPOINTER_CLR; \
 969         }
 970
 971 #define XFS_PREEMPT_MASK        0x7f
 972
 973         if (bypassed)
 974                 *bypassed = 0;
 975         if (XFS_MTOVFS(mp)->vfs_flag & VFS_RDONLY)
 976                 return 0;
 977         error = 0;
 978         last_error = 0;
 979         preempt = 0;
 980
 981         /* Allocate a reference marker */
 982         ipointer = (xfs_iptr_t *)kmem_zalloc(sizeof(xfs_iptr_t), KM_SLEEP);
 983
 984         fflag = XFS_B_ASYNC;            /* default is don't wait */
 985         if (flags & SYNC_BDFLUSH)
 986                 fflag = XFS_B_DELWRI;
 987         if (flags & SYNC_WAIT)
 988                 fflag = 0;              /* synchronous overrides all */
 989
 990         base_lock_flags = XFS_ILOCK_SHARED;
 991         if (flags & (SYNC_DELWRI | SYNC_CLOSE)) {
 992                 /*
 993                  * We need the I/O lock if we're going to call any of
 994                  * the flush/inval routines.
 995                  */
 996                 base_lock_flags |= XFS_IOLOCK_SHARED;
 997         }
 998
 999         XFS_MOUNT_ILOCK(mp);
1000
1001         ip = mp->m_inodes;
1002
1003         mount_locked = B_TRUE;
1004         vnode_refed  = B_FALSE;
1005
1006         IPOINTER_CLR;
1007
1008         do {
1009                 ASSERT(ipointer_in == B_FALSE);
1010                 ASSERT(vnode_refed == B_FALSE);
1011
1012                 lock_flags = base_lock_flags;
1013
1014                 /*
1015                  * There were no inodes in the list, just break out
1016                  * of the loop.
1017                  */
1018                 if (ip == NULL) {
1019                         break;
1020                 }
1021
1022                 /*
1023                  * We found another sync thread marker - skip it
1024                  */
1025                 if (ip->i_mount == NULL) {
1026                         ip = ip->i_mnext;
1027                         continue;
1028                 }
1029
1030                 vp = XFS_ITOV_NULL(ip);
1031
1032                 /*
1033                  * If the vnode is gone then this is being torn down,
1034                  * call reclaim if it is flushed, else let regular flush
1035                  * code deal with it later in the loop.
1036                  */
1037
1038                 if (vp == NULL) {
1039                         /* Skip ones already in reclaim */
1040                         if (ip->i_flags & XFS_IRECLAIM) {
1041                                 ip = ip->i_mnext;
1042                                 continue;
1043                         }
1044                         if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0) {
1045                                 ip = ip->i_mnext;
1046                         } else if ((xfs_ipincount(ip) == 0) &&
1047                                     xfs_iflock_nowait(ip)) {
1048                                 IPOINTER_INSERT(ip, mp);
1049
1050                                 xfs_finish_reclaim(ip, 1,
1051                                                 XFS_IFLUSH_DELWRI_ELSE_ASYNC);
1052
1053                                 XFS_MOUNT_ILOCK(mp);
1054                                 mount_locked = B_TRUE;
1055                                 IPOINTER_REMOVE(ip, mp);
1056                         } else {
1057                                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1058                                 ip = ip->i_mnext;
1059                         }
1060                         continue;
1061                 }
1062
1063                 if (VN_BAD(vp)) {
1064                         ip = ip->i_mnext;
1065                         continue;
1066                 }
1067
1068                 if (XFS_FORCED_SHUTDOWN(mp) && !(flags & SYNC_CLOSE)) {
1069                         XFS_MOUNT_IUNLOCK(mp);
1070                         kmem_free(ipointer, sizeof(xfs_iptr_t));
1071                         return 0;
1072                 }
1073
1074                 /*
1075                  * If this is just vfs_sync() or pflushd() calling
1076                  * then we can skip inodes for which it looks like
1077                  * there is nothing to do.  Since we don't have the
1078                  * inode locked this is racey, but these are periodic
1079                  * calls so it doesn't matter.  For the others we want
1080                  * to know for sure, so we at least try to lock them.
1081                  */
1082                 if (flags & SYNC_BDFLUSH) {
1083                         if (((ip->i_itemp == NULL) ||
1084                              !(ip->i_itemp->ili_format.ilf_fields &
1085                                XFS_ILOG_ALL)) &&
1086                             (ip->i_update_core == 0)) {
1087                                 ip = ip->i_mnext;
1088                                 continue;
1089                         }
1090                 }
1091
1092                 /*
1093                  * Try to lock without sleeping.  We're out of order with
1094                  * the inode list lock here, so if we fail we need to drop
1095                  * the mount lock and try again.  If we're called from
1096                  * bdflush() here, then don't bother.
1097                  *
1098                  * The inode lock here actually coordinates with the
1099                  * almost spurious inode lock in xfs_ireclaim() to prevent
1100                  * the vnode we handle here without a reference from
1101                  * being freed while we reference it.  If we lock the inode
1102                  * while it's on the mount list here, then the spurious inode
1103                  * lock in xfs_ireclaim() after the inode is pulled from
1104                  * the mount list will sleep until we release it here.
1105                  * This keeps the vnode from being freed while we reference
1106                  * it.  It is also cheaper and simpler than actually doing
1107                  * a vn_get() for every inode we touch here.
1108                  */
1109                 if (xfs_ilock_nowait(ip, lock_flags) == 0) {
1110
1111                         if ((flags & SYNC_BDFLUSH) || (vp == NULL)) {
1112                                 ip = ip->i_mnext;
1113                                 continue;
1114                         }
1115
1116                         /*
1117                          * We need to unlock the inode list lock in order
1118                          * to lock the inode. Insert a marker record into
1119                          * the inode list to remember our position, dropping
1120                          * the lock is now done inside the IPOINTER_INSERT
1121                          * macro.
1122                          *
1123                          * We also use the inode list lock to protect us
1124                          * in taking a snapshot of the vnode version number
1125                          * for use in calling vn_get().
1126                          */
1127                         VMAP(vp, vmap);
1128                         IPOINTER_INSERT(ip, mp);
1129
1130                         vp = vn_get(vp, &vmap);
1131                         if (vp == NULL) {
1132                                 /*
1133                                  * The vnode was reclaimed once we let go
1134                                  * of the inode list lock.  Skip to the
1135                                  * next list entry. Remove the marker.
1136                                  */
1137
1138                                 XFS_MOUNT_ILOCK(mp);
1139
1140                                 mount_locked = B_TRUE;
1141                                 vnode_refed  = B_FALSE;
1142
1143                                 IPOINTER_REMOVE(ip, mp);
1144
1145                                 continue;
1146                         }
1147
1148                         xfs_ilock(ip, lock_flags);
1149
1150                         ASSERT(vp == XFS_ITOV(ip));
1151                         ASSERT(ip->i_mount == mp);
1152
1153                         vnode_refed = B_TRUE;
1154                 }
1155
1156                 /* From here on in the loop we may have a marker record
1157                  * in the inode list.
1158                  */
1159
1160                 if ((flags & SYNC_CLOSE)  && (vp != NULL)) {
1161                         /*
1162                          * This is the shutdown case.  We just need to
1163                          * flush and invalidate all the pages associated
1164                          * with the inode.  Drop the inode lock since
1165                          * we can't hold it across calls to the buffer
1166                          * cache.
1167                          *
1168                          * We don't set the VREMAPPING bit in the vnode
1169                          * here, because we don't hold the vnode lock
1170                          * exclusively.  It doesn't really matter, though,
1171                          * because we only come here when we're shutting
1172                          * down anyway.
1173                          */
1174                         xfs_iunlock(ip, XFS_ILOCK_SHARED);
1175
1176                         if (XFS_FORCED_SHUTDOWN(mp)) {
1177                                 VOP_TOSS_PAGES(vp, 0, -1, FI_REMAPF);
1178                         } else {
1179                                 VOP_FLUSHINVAL_PAGES(vp, 0, -1, FI_REMAPF);
1180                         }
1181
1182                         xfs_ilock(ip, XFS_ILOCK_SHARED);
1183
1184                 } else if ((flags & SYNC_DELWRI) && (vp != NULL)) {
1185                         if (VN_DIRTY(vp)) {
1186                                 /* We need to have dropped the lock here,
1187                                  * so insert a marker if we have not already
1188                                  * done so.
1189                                  */
1190                                 if (mount_locked) {
1191                                         IPOINTER_INSERT(ip, mp);
1192                                 }
1193
1194                                 /*
1195                                  * Drop the inode lock since we can't hold it
1196                                  * across calls to the buffer cache.
1197                                  */
1198                                 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1199                                 VOP_FLUSH_PAGES(vp, (xfs_off_t)0, -1,
1200                                                         fflag, FI_NONE, error);
1201                                 xfs_ilock(ip, XFS_ILOCK_SHARED);
1202                         }
1203
1204                 }
1205
1206                 if (flags & SYNC_BDFLUSH) {
1207                         if ((flags & SYNC_ATTR) &&
1208                             ((ip->i_update_core) ||
1209                              ((ip->i_itemp != NULL) &&
1210                               (ip->i_itemp->ili_format.ilf_fields != 0)))) {
1211
1212                                 /* Insert marker and drop lock if not already
1213                                  * done.
1214                                  */
1215                                 if (mount_locked) {
1216                                         IPOINTER_INSERT(ip, mp);
1217                                 }
1218
1219                                 /*
1220                                  * We don't want the periodic flushing of the
1221                                  * inodes by vfs_sync() to interfere with
1222                                  * I/O to the file, especially read I/O
1223                                  * where it is only the access time stamp
1224                                  * that is being flushed out.  To prevent
1225                                  * long periods where we have both inode
1226                                  * locks held shared here while reading the
1227                                  * inode's buffer in from disk, we drop the
1228                                  * inode lock while reading in the inode
1229                                  * buffer.  We have to release the buffer
1230                                  * and reacquire the inode lock so that they
1231                                  * are acquired in the proper order (inode
1232                                  * locks first).  The buffer will go at the
1233                                  * end of the lru chain, though, so we can
1234                                  * expect it to still be there when we go
1235                                  * for it again in xfs_iflush().
1236                                  */
1237                                 if ((xfs_ipincount(ip) == 0) &&
1238                                     xfs_iflock_nowait(ip)) {
1239
1240                                         xfs_ifunlock(ip);
1241                                         xfs_iunlock(ip, XFS_ILOCK_SHARED);
1242
1243                                         error = xfs_itobp(mp, NULL, ip,
1244                                                           &dip, &bp, 0);
1245                                         if (!error) {
1246                                                 xfs_buf_relse(bp);
1247                                         } else {
1248                                                 /* Bailing out, remove the
1249                                                  * marker and free it.
1250                                                  */
1251                                                 XFS_MOUNT_ILOCK(mp);
1252
1253                                                 IPOINTER_REMOVE(ip, mp);
1254
1255                                                 XFS_MOUNT_IUNLOCK(mp);
1256
1257                                                 ASSERT(!(lock_flags &
1258                                                         XFS_IOLOCK_SHARED));
1259
1260                                                 kmem_free(ipointer,
1261                                                         sizeof(xfs_iptr_t));
1262                                                 return (0);
1263                                         }
1264
1265                                         /*
1266                                          * Since we dropped the inode lock,
1267                                          * the inode may have been reclaimed.
1268                                          * Therefore, we reacquire the mount
1269                                          * lock and check to see if we were the
1270                                          * inode reclaimed. If this happened
1271                                          * then the ipointer marker will no
1272                                          * longer point back at us. In this
1273                                          * case, move ip along to the inode
1274                                          * after the marker, remove the marker
1275                                          * and continue.
1276                                          */
1277                                         XFS_MOUNT_ILOCK(mp);
1278                                         mount_locked = B_TRUE;
1279
1280                                         if (ip != ipointer->ip_mprev) {
1281                                                 IPOINTER_REMOVE(ip, mp);
1282
1283                                                 ASSERT(!vnode_refed);
1284                                                 ASSERT(!(lock_flags &
1285                                                         XFS_IOLOCK_SHARED));
1286                                                 continue;
1287                                         }
1288
1289                                         ASSERT(ip->i_mount == mp);
1290
1291                                         if (xfs_ilock_nowait(ip,
1292                                                     XFS_ILOCK_SHARED) == 0) {
1293                                                 ASSERT(ip->i_mount == mp);
1294                                                 /*
1295                                                  * We failed to reacquire
1296                                                  * the inode lock without
1297                                                  * sleeping, so just skip
1298                                                  * the inode for now.  We
1299                                                  * clear the ILOCK bit from
1300                                                  * the lock_flags so that we
1301                                                  * won't try to drop a lock
1302                                                  * we don't hold below.
1303                                                  */
1304                                                 lock_flags &= ~XFS_ILOCK_SHARED;
1305                                                 IPOINTER_REMOVE(ip_next, mp);
1306                                         } else if ((xfs_ipincount(ip) == 0) &&
1307                                                    xfs_iflock_nowait(ip)) {
1308                                                 ASSERT(ip->i_mount == mp);
1309                                                 /*
1310                                                  * Since this is vfs_sync()
1311                                                  * calling we only flush the
1312                                                  * inode out if we can lock
1313                                                  * it without sleeping and
1314                                                  * it is not pinned.  Drop
1315                                                  * the mount lock here so
1316                                                  * that we don't hold it for
1317                                                  * too long. We already have
1318                                                  * a marker in the list here.
1319                                                  */
1320                                                 XFS_MOUNT_IUNLOCK(mp);
1321                                                 mount_locked = B_FALSE;
1322                                                 error = xfs_iflush(ip,
1323                                                            XFS_IFLUSH_DELWRI);
1324                                         } else {
1325                                                 ASSERT(ip->i_mount == mp);
1326                                                 IPOINTER_REMOVE(ip_next, mp);
1327                                         }
1328                                 }
1329
1330                         }
1331
1332                 } else {
1333                         if ((flags & SYNC_ATTR) &&
1334                             ((ip->i_update_core) ||
1335                              ((ip->i_itemp != NULL) &&
1336                               (ip->i_itemp->ili_format.ilf_fields != 0)))) {
1337                                 if (mount_locked) {
1338                                         IPOINTER_INSERT(ip, mp);
1339                                 }
1340
1341                                 if (flags & SYNC_WAIT) {
1342                                         xfs_iflock(ip);
1343                                         error = xfs_iflush(ip,
1344                                                            XFS_IFLUSH_SYNC);
1345                                 } else {
1346                                         /*
1347                                          * If we can't acquire the flush
1348                                          * lock, then the inode is already
1349                                          * being flushed so don't bother
1350                                          * waiting.  If we can lock it then
1351                                          * do a delwri flush so we can
1352                                          * combine multiple inode flushes
1353                                          * in each disk write.
1354                                          */
1355                                         if (xfs_iflock_nowait(ip)) {
1356                                                 error = xfs_iflush(ip,
1357                                                            XFS_IFLUSH_DELWRI);
1358                                         }
1359                                         else if (bypassed)
1360                                                 (*bypassed)++;
1361                                 }
1362                         }
1363                 }
1364
1365                 if (lock_flags != 0) {
1366                         xfs_iunlock(ip, lock_flags);
1367                 }
1368
1369                 if (vnode_refed) {
1370                         /*
1371                          * If we had to take a reference on the vnode
1372                          * above, then wait until after we've unlocked
1373                          * the inode to release the reference.  This is
1374                          * because we can be already holding the inode
1375                          * lock when VN_RELE() calls xfs_inactive().
1376                          *
1377                          * Make sure to drop the mount lock before calling
1378                          * VN_RELE() so that we don't trip over ourselves if
1379                          * we have to go for the mount lock again in the
1380                          * inactive code.
1381                          */
1382                         if (mount_locked) {
1383                                 IPOINTER_INSERT(ip, mp);
1384                         }
1385
1386                         VN_RELE(vp);
1387
1388                         vnode_refed = B_FALSE;
1389                 }
1390
1391                 if (error) {
1392                         last_error = error;
1393                 }
1394
1395                 /*
1396                  * bail out if the filesystem is corrupted.
1397                  */
1398                 if (error == EFSCORRUPTED)  {
1399                         if (!mount_locked) {
1400                                 XFS_MOUNT_ILOCK(mp);
1401                                 IPOINTER_REMOVE(ip, mp);
1402                         }
1403                         XFS_MOUNT_IUNLOCK(mp);
1404                         ASSERT(ipointer_in == B_FALSE);
1405                         kmem_free(ipointer, sizeof(xfs_iptr_t));
1406                         return XFS_ERROR(error);
1407                 }
1408
1409                 /* Let other threads have a chance at the mount lock
1410                  * if we have looped many times without dropping the
1411                  * lock.
1412                  */
1413                 if ((++preempt & XFS_PREEMPT_MASK) == 0) {
1414                         if (mount_locked) {
1415                                 IPOINTER_INSERT(ip, mp);
1416                         }
1417                 }
1418
1419                 if (mount_locked == B_FALSE) {
1420                         XFS_MOUNT_ILOCK(mp);
1421                         mount_locked = B_TRUE;
1422                         IPOINTER_REMOVE(ip, mp);
1423                         continue;
1424                 }
1425
1426                 ASSERT(ipointer_in == B_FALSE);
1427                 ip = ip->i_mnext;
1428
1429         } while (ip != mp->m_inodes);
1430
1431         XFS_MOUNT_IUNLOCK(mp);
1432
1433         ASSERT(ipointer_in == B_FALSE);
1434
1435         kmem_free(ipointer, sizeof(xfs_iptr_t));
1436         return XFS_ERROR(last_error);
1437 }
1438
1439 /*
1440  * xfs sync routine for internal use
1441  *
1442  * This routine supports all of the flags defined for the generic VFS_SYNC
1443  * interface as explained above under xfs_sync.  In the interests of not
1444  * changing interfaces within the 6.5 family, additional internallly-
1445  * required functions are specified within a separate xflags parameter,
1446  * only available by calling this routine.
1447  *
1448  */
1449 int
1450 xfs_syncsub(
1451         xfs_mount_t     *mp,
1452         int             flags,
1453         int             xflags,
1454         int             *bypassed)
1455 {
1456         int             error = 0;
1457         int             last_error = 0;
1458         uint            log_flags = XFS_LOG_FORCE;
1459         xfs_buf_t       *bp;
1460         xfs_buf_log_item_t      *bip;
1461
1462         /*
1463          * Sync out the log.  This ensures that the log is periodically
1464          * flushed even if there is not enough activity to fill it up.
1465          */
1466         if (flags & SYNC_WAIT)
1467                 log_flags |= XFS_LOG_SYNC;
1468
1469         xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1470
1471         if (flags & (SYNC_ATTR|SYNC_DELWRI)) {
1472                 if (flags & SYNC_BDFLUSH)
1473                         xfs_finish_reclaim_all(mp, 1);
1474                 else
1475                         error = xfs_sync_inodes(mp, flags, xflags, bypassed);
1476         }
1477
1478         /*
1479          * Flushing out dirty data above probably generated more
1480          * log activity, so if this isn't vfs_sync() then flush
1481          * the log again.
1482          */
1483         if (flags & SYNC_DELWRI) {
1484                 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1485         }
1486
1487         if (flags & SYNC_FSDATA) {
1488                 /*
1489                  * If this is vfs_sync() then only sync the superblock
1490                  * if we can lock it without sleeping and it is not pinned.
1491                  */
1492                 if (flags & SYNC_BDFLUSH) {
1493                         bp = xfs_getsb(mp, XFS_BUF_TRYLOCK);
1494                         if (bp != NULL) {
1495                                 bip = XFS_BUF_FSPRIVATE(bp,xfs_buf_log_item_t*);
1496                                 if ((bip != NULL) &&
1497                                     xfs_buf_item_dirty(bip)) {
1498                                         if (!(XFS_BUF_ISPINNED(bp))) {
1499                                                 XFS_BUF_ASYNC(bp);
1500                                                 error = xfs_bwrite(mp, bp);
1501                                         } else {
1502                                                 xfs_buf_relse(bp);
1503                                         }
1504                                 } else {
1505                                         xfs_buf_relse(bp);
1506                                 }
1507                         }
1508                 } else {
1509                         bp = xfs_getsb(mp, 0);
1510                         /*
1511                          * If the buffer is pinned then push on the log so
1512                          * we won't get stuck waiting in the write for
1513                          * someone, maybe ourselves, to flush the log.
1514                          * Even though we just pushed the log above, we
1515                          * did not have the superblock buffer locked at
1516                          * that point so it can become pinned in between
1517                          * there and here.
1518                          */
1519                         if (XFS_BUF_ISPINNED(bp))
1520                                 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
1521                         if (flags & SYNC_WAIT)
1522                                 XFS_BUF_UNASYNC(bp);
1523                         else
1524                                 XFS_BUF_ASYNC(bp);
1525                         error = xfs_bwrite(mp, bp);
1526                 }
1527                 if (error) {
1528                         last_error = error;
1529                 }
1530         }
1531
1532         /*
1533          * If this is the periodic sync, then kick some entries out of
1534          * the reference cache.  This ensures that idle entries are
1535          * eventually kicked out of the cache.
1536          */
1537         if (flags & SYNC_REFCACHE) {
1538                 xfs_refcache_purge_some(mp);
1539         }
1540
1541         /*
1542          * Now check to see if the log needs a "dummy" transaction.
1543          */
1544
1545         if (!(flags & SYNC_REMOUNT) && xfs_log_need_covered(mp)) {
1546                 xfs_trans_t *tp;
1547                 xfs_inode_t *ip;
1548
1549                 /*
1550                  * Put a dummy transaction in the log to tell
1551                  * recovery that all others are OK.
1552                  */
1553                 tp = xfs_trans_alloc(mp, XFS_TRANS_DUMMY1);
1554                 if ((error = xfs_trans_reserve(tp, 0,
1555                                 XFS_ICHANGE_LOG_RES(mp),
1556                                 0, 0, 0)))  {
1557                         xfs_trans_cancel(tp, 0);
1558                         return error;
1559                 }
1560
1561                 ip = mp->m_rootip;
1562                 xfs_ilock(ip, XFS_ILOCK_EXCL);
1563
1564                 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1565                 xfs_trans_ihold(tp, ip);
1566                 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1567                 error = xfs_trans_commit(tp, 0, NULL);
1568                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1569                 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1570         }
1571
1572         /*
1573          * When shutting down, we need to insure that the AIL is pushed
1574          * to disk or the filesystem can appear corrupt from the PROM.
1575          */
1576         if ((flags & (SYNC_CLOSE|SYNC_WAIT)) == (SYNC_CLOSE|SYNC_WAIT)) {
1577                 XFS_bflush(mp->m_ddev_targp);
1578                 if (mp->m_rtdev_targp) {
1579                         XFS_bflush(mp->m_rtdev_targp);
1580                 }
1581         }
1582
1583         return XFS_ERROR(last_error);
1584 }
1585
1586 /*
1587  * xfs_vget - called by DMAPI and NFSD to get vnode from file handle
1588  */
1589 STATIC int
1590 xfs_vget(
1591         bhv_desc_t      *bdp,
1592         vnode_t         **vpp,
1593         fid_t           *fidp)
1594 {
1595         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
1596         xfs_fid_t       *xfid = (struct xfs_fid *)fidp;
1597         xfs_inode_t     *ip;
1598         int             error;
1599         xfs_ino_t       ino;
1600         unsigned int    igen;
1601
1602         /*
1603          * Invalid.  Since handles can be created in user space and passed in
1604          * via gethandle(), this is not cause for a panic.
1605          */
1606         if (xfid->xfs_fid_len != sizeof(*xfid) - sizeof(xfid->xfs_fid_len))
1607                 return XFS_ERROR(EINVAL);
1608
1609         ino  = xfid->xfs_fid_ino;
1610         igen = xfid->xfs_fid_gen;
1611
1612         /*
1613          * NFS can sometimes send requests for ino 0.  Fail them gracefully.
1614          */
1615         if (ino == 0)
1616                 return XFS_ERROR(ESTALE);
1617
1618         error = xfs_iget(mp, NULL, ino, 0, XFS_ILOCK_SHARED, &ip, 0);
1619         if (error) {
1620                 *vpp = NULL;
1621                 return error;
1622         }
1623
1624         if (ip == NULL) {
1625                 *vpp = NULL;
1626                 return XFS_ERROR(EIO);
1627         }
1628
1629         if (ip->i_d.di_mode == 0 || ip->i_d.di_gen != igen) {
1630                 xfs_iput_new(ip, XFS_ILOCK_SHARED);
1631                 *vpp = NULL;
1632                 return XFS_ERROR(ENOENT);
1633         }
1634
1635         *vpp = XFS_ITOV(ip);
1636         xfs_iunlock(ip, XFS_ILOCK_SHARED);
1637         return 0;
1638 }
1639
1640
1641 #define MNTOPT_LOGBUFS  "logbufs"       /* number of XFS log buffers */
1642 #define MNTOPT_LOGBSIZE "logbsize"      /* size of XFS log buffers */
1643 #define MNTOPT_LOGDEV   "logdev"        /* log device */
1644 #define MNTOPT_RTDEV    "rtdev"         /* realtime I/O device */
1645 #define MNTOPT_BIOSIZE  "biosize"       /* log2 of preferred buffered io size */
1646 #define MNTOPT_WSYNC    "wsync"         /* safe-mode nfs compatible mount */
1647 #define MNTOPT_INO64    "ino64"         /* force inodes into 64-bit range */
1648 #define MNTOPT_NOALIGN  "noalign"       /* turn off stripe alignment */
1649 #define MNTOPT_SWALLOC  "swalloc"       /* turn on stripe width allocation */
1650 #define MNTOPT_SUNIT    "sunit"         /* data volume stripe unit */
1651 #define MNTOPT_SWIDTH   "swidth"        /* data volume stripe width */
1652 #define MNTOPT_NOUUID   "nouuid"        /* ignore filesystem UUID */
1653 #define MNTOPT_MTPT     "mtpt"          /* filesystem mount point */
1654 #define MNTOPT_IHASHSIZE    "ihashsize"    /* size of inode hash table */
1655 #define MNTOPT_NORECOVERY   "norecovery"   /* don't run XFS recovery */
1656 #define MNTOPT_NOLOGFLUSH   "nologflush"   /* don't hard flush on log writes */
1657 #define MNTOPT_OSYNCISOSYNC "osyncisosync" /* o_sync is REALLY o_sync */
1658 #define MNTOPT_64BITINODE   "inode64"   /* inodes can be allocated anywhere */
1659 #define MNTOPT_IKEEP    "ikeep"         /* do not free empty inode clusters */
1660 #define MNTOPT_NOIKEEP  "noikeep"       /* free empty inode clusters */
1661 #define MNTOPT_TAGXID   "tagxid"        /* context xid tagging for inodes */
1662
1663
1664 int
1665 xfs_parseargs(
1666         struct bhv_desc         *bhv,
1667         char                    *options,
1668         struct xfs_mount_args   *args,
1669         int                     update)
1670 {
1671         struct vfs              *vfsp = bhvtovfs(bhv);
1672         char                    *this_char, *value, *eov;
1673         int                     dsunit, dswidth, vol_dsunit, vol_dswidth;
1674         int                     iosize;
1675
1676 #if 0   /* XXX: off by default, until some remaining issues ironed out */
1677         args->flags |= XFSMNT_IDELETE; /* default to on */
1678 #endif
1679
1680         if (!options)
1681                 return 0;
1682
1683         iosize = dsunit = dswidth = vol_dsunit = vol_dswidth = 0;
1684
1685         while ((this_char = strsep(&options, ",")) != NULL) {
1686                 if (!*this_char)
1687                         continue;
1688                 if ((value = strchr(this_char, '=')) != NULL)
1689                         *value++ = 0;
1690
1691                 if (!strcmp(this_char, MNTOPT_LOGBUFS)) {
1692                         if (!value || !*value) {
1693                                 printk("XFS: %s option requires an argument\n",
1694                                         MNTOPT_LOGBUFS);
1695                                 return EINVAL;
1696                         }
1697                         args->logbufs = simple_strtoul(value, &eov, 10);
1698                 } else if (!strcmp(this_char, MNTOPT_LOGBSIZE)) {
1699                         int     last, in_kilobytes = 0;
1700
1701                         if (!value || !*value) {
1702                                 printk("XFS: %s option requires an argument\n",
1703                                         MNTOPT_LOGBSIZE);
1704                                 return EINVAL;
1705                         }
1706                         last = strlen(value) - 1;
1707                         if (value[last] == 'K' || value[last] == 'k') {
1708                                 in_kilobytes = 1;
1709                                 value[last] = '\0';
1710                         }
1711                         args->logbufsize = simple_strtoul(value, &eov, 10);
1712                         if (in_kilobytes)
1713                                 args->logbufsize <<= 10;
1714                 } else if (!strcmp(this_char, MNTOPT_LOGDEV)) {
1715                         if (!value || !*value) {
1716                                 printk("XFS: %s option requires an argument\n",
1717                                         MNTOPT_LOGDEV);
1718                                 return EINVAL;
1719                         }
1720                         strncpy(args->logname, value, MAXNAMELEN);
1721                 } else if (!strcmp(this_char, MNTOPT_MTPT)) {
1722                         if (!value || !*value) {
1723                                 printk("XFS: %s option requires an argument\n",
1724                                         MNTOPT_MTPT);
1725                                 return EINVAL;
1726                         }
1727                         strncpy(args->mtpt, value, MAXNAMELEN);
1728                 } else if (!strcmp(this_char, MNTOPT_RTDEV)) {
1729                         if (!value || !*value) {
1730                                 printk("XFS: %s option requires an argument\n",
1731                                         MNTOPT_RTDEV);
1732                                 return EINVAL;
1733                         }
1734                         strncpy(args->rtname, value, MAXNAMELEN);
1735                 } else if (!strcmp(this_char, MNTOPT_BIOSIZE)) {
1736                         if (!value || !*value) {
1737                                 printk("XFS: %s option requires an argument\n",
1738                                         MNTOPT_BIOSIZE);
1739                                 return EINVAL;
1740                         }
1741                         iosize = simple_strtoul(value, &eov, 10);
1742                         args->flags |= XFSMNT_IOSIZE;
1743                         args->iosizelog = (uint8_t) iosize;
1744                 } else if (!strcmp(this_char, MNTOPT_IHASHSIZE)) {
1745                         if (!value || !*value) {
1746                                 printk("XFS: %s option requires an argument\n",
1747                                         this_char);
1748                                 return EINVAL;
1749                         }
1750                         args->ihashsize = simple_strtoul(value, &eov, 10);
1751                 } else if (!strcmp(this_char, MNTOPT_WSYNC)) {
1752                         args->flags |= XFSMNT_WSYNC;
1753                 } else if (!strcmp(this_char, MNTOPT_OSYNCISOSYNC)) {
1754                         args->flags |= XFSMNT_OSYNCISOSYNC;
1755                 } else if (!strcmp(this_char, MNTOPT_NORECOVERY)) {
1756                         args->flags |= XFSMNT_NORECOVERY;
1757                 } else if (!strcmp(this_char, MNTOPT_INO64)) {
1758                         args->flags |= XFSMNT_INO64;
1759 #if !XFS_BIG_INUMS
1760                         printk("XFS: %s option not allowed on this system\n",
1761                                 MNTOPT_INO64);
1762                         return EINVAL;
1763 #endif
1764                 } else if (!strcmp(this_char, MNTOPT_NOALIGN)) {
1765                         args->flags |= XFSMNT_NOALIGN;
1766                 } else if (!strcmp(this_char, MNTOPT_SWALLOC)) {
1767                         args->flags |= XFSMNT_SWALLOC;
1768                 } else if (!strcmp(this_char, MNTOPT_SUNIT)) {
1769                         if (!value || !*value) {
1770                                 printk("XFS: %s option requires an argument\n",
1771                                         MNTOPT_SUNIT);
1772                                 return EINVAL;
1773                         }
1774                         dsunit = simple_strtoul(value, &eov, 10);
1775                 } else if (!strcmp(this_char, MNTOPT_SWIDTH)) {
1776                         if (!value || !*value) {
1777                                 printk("XFS: %s option requires an argument\n",
1778                                         MNTOPT_SWIDTH);
1779                                 return EINVAL;
1780                         }
1781                         dswidth = simple_strtoul(value, &eov, 10);
1782                 } else if (!strcmp(this_char, MNTOPT_64BITINODE)) {
1783                         args->flags &= ~XFSMNT_32BITINODES;
1784 #if !XFS_BIG_INUMS
1785                         printk("XFS: %s option not allowed on this system\n",
1786                                 MNTOPT_64BITINODE);
1787                         return EINVAL;
1788 #endif
1789                 } else if (!strcmp(this_char, MNTOPT_NOUUID)) {
1790                         args->flags |= XFSMNT_NOUUID;
1791                 } else if (!strcmp(this_char, MNTOPT_NOLOGFLUSH)) {
1792                         args->flags |= XFSMNT_NOLOGFLUSH;
1793                 } else if (!strcmp(this_char, MNTOPT_IKEEP)) {
1794                         args->flags &= ~XFSMNT_IDELETE;
1795                 } else if (!strcmp(this_char, MNTOPT_NOIKEEP)) {
1796                         args->flags |= XFSMNT_IDELETE;
1797                 } else if (!strcmp(this_char, MNTOPT_TAGXID)) {
1798                         args->flags |= XFSMNT_TAGXID;
1799                 } else if (!strcmp(this_char, "osyncisdsync")) {
1800                         /* no-op, this is now the default */
1801 printk("XFS: osyncisdsync is now the default, option is deprecated.\n");
1802                 } else if (!strcmp(this_char, "irixsgid")) {
1803 printk("XFS: irixsgid is now a sysctl(2) variable, option is deprecated.\n");
1804                 } else {
1805                         printk("XFS: unknown mount option [%s].\n", this_char);
1806                         return EINVAL;
1807                 }
1808         }
1809
1810         if (args->flags & XFSMNT_NORECOVERY) {
1811                 if ((vfsp->vfs_flag & VFS_RDONLY) == 0) {
1812                         printk("XFS: no-recovery mounts must be read-only.\n");
1813                         return EINVAL;
1814                 }
1815         }
1816
1817         if ((args->flags & XFSMNT_NOALIGN) && (dsunit || dswidth)) {
1818                 printk(
1819         "XFS: sunit and swidth options incompatible with the noalign option\n");
1820                 return EINVAL;
1821         }
1822
1823         if ((dsunit && !dswidth) || (!dsunit && dswidth)) {
1824                 printk("XFS: sunit and swidth must be specified together\n");
1825                 return EINVAL;
1826         }
1827
1828         if (dsunit && (dswidth % dsunit != 0)) {
1829                 printk(
1830         "XFS: stripe width (%d) must be a multiple of the stripe unit (%d)\n",
1831                         dswidth, dsunit);
1832                 return EINVAL;
1833         }
1834
1835         if ((args->flags & XFSMNT_NOALIGN) != XFSMNT_NOALIGN) {
1836                 if (dsunit) {
1837                         args->sunit = dsunit;
1838                         args->flags |= XFSMNT_RETERR;
1839                 } else {
1840                         args->sunit = vol_dsunit;
1841                 }
1842                 dswidth ? (args->swidth = dswidth) :
1843                           (args->swidth = vol_dswidth);
1844         } else {
1845                 args->sunit = args->swidth = 0;
1846         }
1847
1848         return 0;
1849 }
1850
1851 int
1852 xfs_showargs(
1853         struct bhv_desc         *bhv,
1854         struct seq_file         *m)
1855 {
1856         static struct proc_xfs_info {
1857                 int     flag;
1858                 char    *str;
1859         } xfs_info[] = {
1860                 /* the few simple ones we can get from the mount struct */
1861                 { XFS_MOUNT_WSYNC,              "," MNTOPT_WSYNC },
1862                 { XFS_MOUNT_INO64,              "," MNTOPT_INO64 },
1863                 { XFS_MOUNT_NOALIGN,            "," MNTOPT_NOALIGN },
1864                 { XFS_MOUNT_SWALLOC,            "," MNTOPT_SWALLOC },
1865                 { XFS_MOUNT_NOUUID,             "," MNTOPT_NOUUID },
1866                 { XFS_MOUNT_NORECOVERY,         "," MNTOPT_NORECOVERY },
1867                 { XFS_MOUNT_OSYNCISOSYNC,       "," MNTOPT_OSYNCISOSYNC },
1868                 { XFS_MOUNT_NOLOGFLUSH,         "," MNTOPT_NOLOGFLUSH },
1869                 { XFS_MOUNT_IDELETE,            "," MNTOPT_NOIKEEP },
1870                 { 0, NULL }
1871         };
1872         struct proc_xfs_info    *xfs_infop;
1873         struct xfs_mount        *mp = XFS_BHVTOM(bhv);
1874
1875         for (xfs_infop = xfs_info; xfs_infop->flag; xfs_infop++) {
1876                 if (mp->m_flags & xfs_infop->flag)
1877                         seq_puts(m, xfs_infop->str);
1878         }
1879
1880         if (mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)
1881                 seq_printf(m, "," MNTOPT_BIOSIZE "=%d", mp->m_writeio_log);
1882
1883         if (mp->m_logbufs > 0)
1884                 seq_printf(m, "," MNTOPT_LOGBUFS "=%d", mp->m_logbufs);
1885
1886         if (mp->m_logbsize > 0)
1887                 seq_printf(m, "," MNTOPT_LOGBSIZE "=%d", mp->m_logbsize);
1888
1889         if (mp->m_ddev_targp != mp->m_logdev_targp)
1890                 seq_printf(m, "," MNTOPT_LOGDEV "=%s",
1891                                 XFS_BUFTARG_NAME(mp->m_logdev_targp));
1892
1893         if (mp->m_rtdev_targp && mp->m_ddev_targp != mp->m_rtdev_targp)
1894                 seq_printf(m, "," MNTOPT_RTDEV "=%s",
1895                                 XFS_BUFTARG_NAME(mp->m_rtdev_targp));
1896
1897         if (mp->m_dalign > 0)
1898                 seq_printf(m, "," MNTOPT_SUNIT "=%d",
1899                                 (int)XFS_FSB_TO_BB(mp, mp->m_dalign));
1900
1901         if (mp->m_swidth > 0)
1902                 seq_printf(m, "," MNTOPT_SWIDTH "=%d",
1903                                 (int)XFS_FSB_TO_BB(mp, mp->m_swidth));
1904
1905         if (!(mp->m_flags & XFS_MOUNT_32BITINOOPT))
1906                 seq_printf(m, "," MNTOPT_64BITINODE);
1907
1908         return 0;
1909 }
1910
1911 STATIC void
1912 xfs_freeze(
1913         bhv_desc_t      *bdp)
1914 {
1915         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
1916
1917         while (atomic_read(&mp->m_active_trans) > 0)
1918                 delay(100);
1919
1920         /* Push the superblock and write an unmount record */
1921         xfs_log_unmount_write(mp);
1922         xfs_unmountfs_writesb(mp);
1923 }
1924
1925
1926 vfsops_t xfs_vfsops = {
1927         BHV_IDENTITY_INIT(VFS_BHV_XFS,VFS_POSITION_XFS),
1928         .vfs_parseargs          = xfs_parseargs,
1929         .vfs_showargs           = xfs_showargs,
1930         .vfs_mount              = xfs_mount,
1931         .vfs_unmount            = xfs_unmount,
1932         .vfs_mntupdate          = xfs_mntupdate,
1933         .vfs_root               = xfs_root,
1934         .vfs_statvfs            = xfs_statvfs,
1935         .vfs_sync               = xfs_sync,
1936         .vfs_vget               = xfs_vget,
1937         .vfs_dmapiops           = (vfs_dmapiops_t)fs_nosys,
1938         .vfs_quotactl           = (vfs_quotactl_t)fs_nosys,
1939         .vfs_init_vnode         = xfs_initialize_vnode,
1940         .vfs_force_shutdown     = xfs_do_force_shutdown,
1941         .vfs_freeze             = xfs_freeze,
1942 };