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_TAGXID)
 325                 mp->m_flags |= XFS_MOUNT_TAGXID;
 326         if (ap->flags & XFSMNT_NOLOGFLUSH)
 327                 mp->m_flags |= XFS_MOUNT_NOLOGFLUSH;
 328
 329         return 0;
 330 }
 331
 332 /*
 333  * This function fills in xfs_mount_t fields based on mount args.
 334  * Note: the superblock _has_ now been read in.
 335  */
 336 STATIC int
 337 xfs_finish_flags(
 338         struct vfs              *vfs,
 339         struct xfs_mount_args   *ap,
 340         struct xfs_mount        *mp)
 341 {
 342         int                     ronly = (vfs->vfs_flag & VFS_RDONLY);
 343
 344         /* Fail a mount where the logbuf is smaller then the log stripe */
 345         if (XFS_SB_VERSION_HASLOGV2(&mp->m_sb)) {
 346                 if ((ap->logbufsize == -1) &&
 347                     (mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE)) {
 348                         mp->m_logbsize = mp->m_sb.sb_logsunit;
 349                 } else if (ap->logbufsize < mp->m_sb.sb_logsunit) {
 350                         cmn_err(CE_WARN,
 351         "XFS: logbuf size must be greater than or equal to log stripe size");
 352                         return XFS_ERROR(EINVAL);
 353                 }
 354         } else {
 355                 /* Fail a mount if the logbuf is larger than 32K */
 356                 if (ap->logbufsize > XLOG_BIG_RECORD_BSIZE) {
 357                         cmn_err(CE_WARN,
 358         "XFS: logbuf size for version 1 logs must be 16K or 32K");
 359                         return XFS_ERROR(EINVAL);
 360                 }
 361         }
 362
 363         /*
 364          * prohibit r/w mounts of read-only filesystems
 365          */
 366         if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !ronly) {
 367                 cmn_err(CE_WARN,
 368         "XFS: cannot mount a read-only filesystem as read-write");
 369                 return XFS_ERROR(EROFS);
 370         }
 371
 372         /*
 373          * disallow mount attempts with (IRIX) project quota enabled
 374          */
 375         if (XFS_SB_VERSION_HASQUOTA(&mp->m_sb) &&
 376             (mp->m_sb.sb_qflags & XFS_PQUOTA_ACCT)) {
 377                 cmn_err(CE_WARN,
 378         "XFS: cannot mount a filesystem with IRIX project quota enabled");
 379                 return XFS_ERROR(ENOSYS);
 380         }
 381
 382         /*
 383          * check for shared mount.
 384          */
 385         if (ap->flags & XFSMNT_SHARED) {
 386                 if (!XFS_SB_VERSION_HASSHARED(&mp->m_sb))
 387                         return XFS_ERROR(EINVAL);
 388
 389                 /*
 390                  * For IRIX 6.5, shared mounts must have the shared
 391                  * version bit set, have the persistent readonly
 392                  * field set, must be version 0 and can only be mounted
 393                  * read-only.
 394                  */
 395                 if (!ronly || !(mp->m_sb.sb_flags & XFS_SBF_READONLY) ||
 396                      (mp->m_sb.sb_shared_vn != 0))
 397                         return XFS_ERROR(EINVAL);
 398
 399                 mp->m_flags |= XFS_MOUNT_SHARED;
 400
 401                 /*
 402                  * Shared XFS V0 can't deal with DMI.  Return EINVAL.
 403                  */
 404                 if (mp->m_sb.sb_shared_vn == 0 && (ap->flags & XFSMNT_DMAPI))
 405                         return XFS_ERROR(EINVAL);
 406         }
 407
 408         if (ap->flags & XFSMNT_TAGXID)
 409                 vfs->vfs_super->s_flags |= MS_TAGXID;
 410         return 0;
 411 }
 412
 413 /*
 414  * xfs_mount
 415  *
 416  * The file system configurations are:
 417  *      (1) device (partition) with data and internal log
 418  *      (2) logical volume with data and log subvolumes.
 419  *      (3) logical volume with data, log, and realtime subvolumes.
 420  *
 421  * We only have to handle opening the log and realtime volumes here if
 422  * they are present.  The data subvolume has already been opened by
 423  * get_sb_bdev() and is stored in vfsp->vfs_super->s_bdev.
 424  */
 425 STATIC int
 426 xfs_mount(
 427         struct bhv_desc         *bhvp,
 428         struct xfs_mount_args   *args,
 429         cred_t                  *credp)
 430 {
 431         struct vfs              *vfsp = bhvtovfs(bhvp);
 432         struct bhv_desc         *p;
 433         struct xfs_mount        *mp = XFS_BHVTOM(bhvp);
 434         struct block_device     *ddev, *logdev, *rtdev;
 435         int                     flags = 0, error;
 436
 437         ddev = vfsp->vfs_super->s_bdev;
 438         logdev = rtdev = NULL;
 439
 440         /*
 441          * Setup xfs_mount function vectors from available behaviors
 442          */
 443         p = vfs_bhv_lookup(vfsp, VFS_POSITION_DM);
 444         mp->m_dm_ops = p ? *(xfs_dmops_t *) vfs_bhv_custom(p) : xfs_dmcore_stub;
 445         p = vfs_bhv_lookup(vfsp, VFS_POSITION_QM);
 446         mp->m_qm_ops = p ? *(xfs_qmops_t *) vfs_bhv_custom(p) : xfs_qmcore_stub;
 447         p = vfs_bhv_lookup(vfsp, VFS_POSITION_IO);
 448         mp->m_io_ops = p ? *(xfs_ioops_t *) vfs_bhv_custom(p) : xfs_iocore_xfs;
 449
 450         /*
 451          * Open real time and log devices - order is important.
 452          */
 453         if (args->logname[0]) {
 454                 error = xfs_blkdev_get(mp, args->logname, &logdev);
 455                 if (error)
 456                         return error;
 457         }
 458         if (args->rtname[0]) {
 459                 error = xfs_blkdev_get(mp, args->rtname, &rtdev);
 460                 if (error) {
 461                         xfs_blkdev_put(logdev);
 462                         return error;
 463                 }
 464
 465                 if (rtdev == ddev || rtdev == logdev) {
 466                         cmn_err(CE_WARN,
 467         "XFS: Cannot mount filesystem with identical rtdev and ddev/logdev.");
 468                         xfs_blkdev_put(logdev);
 469                         xfs_blkdev_put(rtdev);
 470                         return EINVAL;
 471                 }
 472         }
 473
 474         /*
 475          * Setup xfs_mount buffer target pointers
 476          */
 477         error = ENOMEM;
 478         mp->m_ddev_targp = xfs_alloc_buftarg(ddev, 0);
 479         if (!mp->m_ddev_targp) {
 480                 xfs_blkdev_put(logdev);
 481                 xfs_blkdev_put(rtdev);
 482                 return error;
 483         }
 484         if (rtdev) {
 485                 mp->m_rtdev_targp = xfs_alloc_buftarg(rtdev, 1);
 486                 if (!mp->m_rtdev_targp)
 487                         goto error0;
 488         }
 489         mp->m_logdev_targp = (logdev && logdev != ddev) ?
 490                                 xfs_alloc_buftarg(logdev, 1) : mp->m_ddev_targp;
 491         if (!mp->m_logdev_targp)
 492                 goto error0;
 493
 494         /*
 495          * Setup flags based on mount(2) options and then the superblock
 496          */
 497         error = xfs_start_flags(vfsp, args, mp);
 498         if (error)
 499                 goto error1;
 500         error = xfs_readsb(mp);
 501         if (error)
 502                 goto error1;
 503         error = xfs_finish_flags(vfsp, args, mp);
 504         if (error)
 505                 goto error2;
 506
 507         /*
 508          * Setup xfs_mount buffer target pointers based on superblock
 509          */
 510         error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_blocksize,
 511                                     mp->m_sb.sb_sectsize);
 512         if (!error && logdev && logdev != ddev) {
 513                 unsigned int    log_sector_size = BBSIZE;
 514
 515                 if (XFS_SB_VERSION_HASSECTOR(&mp->m_sb))
 516                         log_sector_size = mp->m_sb.sb_logsectsize;
 517                 error = xfs_setsize_buftarg(mp->m_logdev_targp,
 518                                             mp->m_sb.sb_blocksize,
 519                                             log_sector_size);
 520         }
 521         if (!error && rtdev)
 522                 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
 523                                             mp->m_sb.sb_blocksize,
 524                                             mp->m_sb.sb_sectsize);
 525         if (error)
 526                 goto error2;
 527
 528         error = XFS_IOINIT(vfsp, args, flags);
 529         if (!error)
 530                 return 0;
 531 error2:
 532         if (mp->m_sb_bp)
 533                 xfs_freesb(mp);
 534 error1:
 535         xfs_binval(mp->m_ddev_targp);
 536         if (logdev && logdev != ddev)
 537                 xfs_binval(mp->m_logdev_targp);
 538         if (rtdev)
 539                 xfs_binval(mp->m_rtdev_targp);
 540 error0:
 541         xfs_unmountfs_close(mp, credp);
 542         return error;
 543 }
 544
 545 STATIC int
 546 xfs_unmount(
 547         bhv_desc_t      *bdp,
 548         int             flags,
 549         cred_t          *credp)
 550 {
 551         struct vfs      *vfsp = bhvtovfs(bdp);
 552         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
 553         xfs_inode_t     *rip;
 554         vnode_t         *rvp;
 555         int             unmount_event_wanted = 0;
 556         int             unmount_event_flags = 0;
 557         int             xfs_unmountfs_needed = 0;
 558         int             error;
 559
 560         rip = mp->m_rootip;
 561         rvp = XFS_ITOV(rip);
 562
 563         if (vfsp->vfs_flag & VFS_DMI) {
 564                 error = XFS_SEND_PREUNMOUNT(mp, vfsp,
 565                                 rvp, DM_RIGHT_NULL, rvp, DM_RIGHT_NULL,
 566                                 NULL, NULL, 0, 0,
 567                                 (mp->m_dmevmask & (1<<DM_EVENT_PREUNMOUNT))?
 568                                         0:DM_FLAGS_UNWANTED);
 569                         if (error)
 570                                 return XFS_ERROR(error);
 571                 unmount_event_wanted = 1;
 572                 unmount_event_flags = (mp->m_dmevmask & (1<<DM_EVENT_UNMOUNT))?
 573                                         0 : DM_FLAGS_UNWANTED;
 574         }
 575
 576         /*
 577          * First blow any referenced inode from this file system
 578          * out of the reference cache, and delete the timer.
 579          */
 580         xfs_refcache_purge_mp(mp);
 581
 582         XFS_bflush(mp->m_ddev_targp);
 583         error = xfs_unmount_flush(mp, 0);
 584         if (error)
 585                 goto out;
 586
 587         ASSERT(vn_count(rvp) == 1);
 588
 589         /*
 590          * Drop the reference count
 591          */
 592         VN_RELE(rvp);
 593
 594         /*
 595          * If we're forcing a shutdown, typically because of a media error,
 596          * we want to make sure we invalidate dirty pages that belong to
 597          * referenced vnodes as well.
 598          */
 599         if (XFS_FORCED_SHUTDOWN(mp)) {
 600                 error = xfs_sync(&mp->m_bhv,
 601                          (SYNC_WAIT | SYNC_CLOSE), credp);
 602                 ASSERT(error != EFSCORRUPTED);
 603         }
 604         xfs_unmountfs_needed = 1;
 605
 606 out:
 607         /*      Send DMAPI event, if required.
 608          *      Then do xfs_unmountfs() if needed.
 609          *      Then return error (or zero).
 610          */
 611         if (unmount_event_wanted) {
 612                 /* Note: mp structure must still exist for
 613                  * XFS_SEND_UNMOUNT() call.
 614                  */
 615                 XFS_SEND_UNMOUNT(mp, vfsp, error == 0 ? rvp : NULL,
 616                         DM_RIGHT_NULL, 0, error, unmount_event_flags);
 617         }
 618         if (xfs_unmountfs_needed) {
 619                 /*
 620                  * Call common unmount function to flush to disk
 621                  * and free the super block buffer & mount structures.
 622                  */
 623                 xfs_unmountfs(mp, credp);
 624         }
 625
 626         return XFS_ERROR(error);
 627 }
 628
 629 #define REMOUNT_READONLY_FLAGS  (SYNC_REMOUNT|SYNC_ATTR|SYNC_WAIT)
 630
 631 STATIC int
 632 xfs_mntupdate(
 633         bhv_desc_t                      *bdp,
 634         int                             *flags,
 635         struct xfs_mount_args           *args)
 636 {
 637         struct vfs      *vfsp = bhvtovfs(bdp);
 638         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
 639         int             pincount, error;
 640         int             count = 0;
 641
 642         if (args->flags & XFSMNT_NOATIME)
 643                 mp->m_flags |= XFS_MOUNT_NOATIME;
 644         else
 645                 mp->m_flags &= ~XFS_MOUNT_NOATIME;
 646
 647         if (!(vfsp->vfs_flag & VFS_RDONLY)) {
 648                 VFS_SYNC(vfsp, SYNC_FSDATA|SYNC_BDFLUSH|SYNC_ATTR, NULL, error);
 649         }
 650
 651         if (*flags & MS_RDONLY) {
 652                 xfs_refcache_purge_mp(mp);
 653                 xfs_flush_buftarg(mp->m_ddev_targp, 0);
 654                 xfs_finish_reclaim_all(mp, 0);
 655
 656                 /* This loop must run at least twice.
 657                  * The first instance of the loop will flush
 658                  * most meta data but that will generate more
 659                  * meta data (typically directory updates).
 660                  * Which then must be flushed and logged before
 661                  * we can write the unmount record.
 662                  */
 663                 do {
 664                         VFS_SYNC(vfsp, REMOUNT_READONLY_FLAGS, NULL, error);
 665                         pincount = xfs_flush_buftarg(mp->m_ddev_targp, 1);
 666                         if (!pincount) {
 667                                 delay(50);
 668                                 count++;
 669                         }
 670                 } while (count < 2);
 671
 672                 /* Ok now write out an unmount record */
 673                 xfs_log_unmount_write(mp);
 674                 xfs_unmountfs_writesb(mp);
 675                 vfsp->vfs_flag |= VFS_RDONLY;
 676         } else {
 677                 vfsp->vfs_flag &= ~VFS_RDONLY;
 678         }
 679
 680         return 0;
 681 }
 682
 683 /*
 684  * xfs_unmount_flush implements a set of flush operation on special
 685  * inodes, which are needed as a separate set of operations so that
 686  * they can be called as part of relocation process.
 687  */
 688 int
 689 xfs_unmount_flush(
 690         xfs_mount_t     *mp,            /* Mount structure we are getting
 691                                            rid of. */
 692         int             relocation)     /* Called from vfs relocation. */
 693 {
 694         xfs_inode_t     *rip = mp->m_rootip;
 695         xfs_inode_t     *rbmip;
 696         xfs_inode_t     *rsumip = NULL;
 697         vnode_t         *rvp = XFS_ITOV(rip);
 698         int             error;
 699
 700         xfs_ilock(rip, XFS_ILOCK_EXCL);
 701         xfs_iflock(rip);
 702
 703         /*
 704          * Flush out the real time inodes.
 705          */
 706         if ((rbmip = mp->m_rbmip) != NULL) {
 707                 xfs_ilock(rbmip, XFS_ILOCK_EXCL);
 708                 xfs_iflock(rbmip);
 709                 error = xfs_iflush(rbmip, XFS_IFLUSH_SYNC);
 710                 xfs_iunlock(rbmip, XFS_ILOCK_EXCL);
 711
 712                 if (error == EFSCORRUPTED)
 713                         goto fscorrupt_out;
 714
 715                 ASSERT(vn_count(XFS_ITOV(rbmip)) == 1);
 716
 717                 rsumip = mp->m_rsumip;
 718                 xfs_ilock(rsumip, XFS_ILOCK_EXCL);
 719                 xfs_iflock(rsumip);
 720                 error = xfs_iflush(rsumip, XFS_IFLUSH_SYNC);
 721                 xfs_iunlock(rsumip, XFS_ILOCK_EXCL);
 722
 723                 if (error == EFSCORRUPTED)
 724                         goto fscorrupt_out;
 725
 726                 ASSERT(vn_count(XFS_ITOV(rsumip)) == 1);
 727         }
 728
 729         /*
 730          * Synchronously flush root inode to disk
 731          */
 732         error = xfs_iflush(rip, XFS_IFLUSH_SYNC);
 733         if (error == EFSCORRUPTED)
 734                 goto fscorrupt_out2;
 735
 736         if (vn_count(rvp) != 1 && !relocation) {
 737                 xfs_iunlock(rip, XFS_ILOCK_EXCL);
 738                 return XFS_ERROR(EBUSY);
 739         }
 740
 741         /*
 742          * Release dquot that rootinode, rbmino and rsumino might be holding,
 743          * flush and purge the quota inodes.
 744          */
 745         error = XFS_QM_UNMOUNT(mp);
 746         if (error == EFSCORRUPTED)
 747                 goto fscorrupt_out2;
 748
 749         if (rbmip) {
 750                 VN_RELE(XFS_ITOV(rbmip));
 751                 VN_RELE(XFS_ITOV(rsumip));
 752         }
 753
 754         xfs_iunlock(rip, XFS_ILOCK_EXCL);
 755         return 0;
 756
 757 fscorrupt_out:
 758         xfs_ifunlock(rip);
 759
 760 fscorrupt_out2:
 761         xfs_iunlock(rip, XFS_ILOCK_EXCL);
 762
 763         return XFS_ERROR(EFSCORRUPTED);
 764 }
 765
 766 /*
 767  * xfs_root extracts the root vnode from a vfs.
 768  *
 769  * vfsp -- the vfs struct for the desired file system
 770  * vpp  -- address of the caller's vnode pointer which should be
 771  *         set to the desired fs root vnode
 772  */
 773 STATIC int
 774 xfs_root(
 775         bhv_desc_t      *bdp,
 776         vnode_t         **vpp)
 777 {
 778         vnode_t         *vp;
 779
 780         vp = XFS_ITOV((XFS_BHVTOM(bdp))->m_rootip);
 781         VN_HOLD(vp);
 782         *vpp = vp;
 783         return 0;
 784 }
 785
 786 /*
 787  * xfs_statvfs
 788  *
 789  * Fill in the statvfs structure for the given file system.  We use
 790  * the superblock lock in the mount structure to ensure a consistent
 791  * snapshot of the counters returned.
 792  */
 793 STATIC int
 794 xfs_statvfs(
 795         bhv_desc_t      *bdp,
 796         xfs_statfs_t    *statp,
 797         vnode_t         *vp)
 798 {
 799         __uint64_t      fakeinos;
 800         xfs_extlen_t    lsize;
 801         xfs_mount_t     *mp;
 802         xfs_sb_t        *sbp;
 803         unsigned long   s;
 804         u64 id;
 805
 806         mp = XFS_BHVTOM(bdp);
 807         sbp = &(mp->m_sb);
 808
 809         statp->f_type = XFS_SB_MAGIC;
 810
 811         s = XFS_SB_LOCK(mp);
 812         statp->f_bsize = sbp->sb_blocksize;
 813         lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
 814         statp->f_blocks = sbp->sb_dblocks - lsize;
 815         statp->f_bfree = statp->f_bavail = sbp->sb_fdblocks;
 816         fakeinos = statp->f_bfree << sbp->sb_inopblog;
 817 #if XFS_BIG_INUMS
 818         fakeinos += mp->m_inoadd;
 819 #endif
 820         statp->f_files =
 821             MIN(sbp->sb_icount + fakeinos, (__uint64_t)XFS_MAXINUMBER);
 822         if (mp->m_maxicount)
 823 #if XFS_BIG_INUMS
 824                 if (!mp->m_inoadd)
 825 #endif
 826                         statp->f_files = min_t(typeof(statp->f_files),
 827                                                 statp->f_files,
 828                                                 mp->m_maxicount);
 829         statp->f_ffree = statp->f_files - (sbp->sb_icount - sbp->sb_ifree);
 830         XFS_SB_UNLOCK(mp, s);
 831
 832         id = huge_encode_dev(mp->m_dev);
 833         statp->f_fsid.val[0] = (u32)id;
 834         statp->f_fsid.val[1] = (u32)(id >> 32);
 835         statp->f_namelen = MAXNAMELEN - 1;
 836
 837         return 0;
 838 }
 839
 840
 841 /*
 842  * xfs_sync flushes any pending I/O to file system vfsp.
 843  *
 844  * This routine is called by vfs_sync() to make sure that things make it
 845  * out to disk eventually, on sync() system calls to flush out everything,
 846  * and when the file system is unmounted.  For the vfs_sync() case, all
 847  * we really need to do is sync out the log to make all of our meta-data
 848  * updates permanent (except for timestamps).  For calls from pflushd(),
 849  * dirty pages are kept moving by calling pdflush() on the inodes
 850  * containing them.  We also flush the inodes that we can lock without
 851  * sleeping and the superblock if we can lock it without sleeping from
 852  * vfs_sync() so that items at the tail of the log are always moving out.
 853  *
 854  * Flags:
 855  *      SYNC_BDFLUSH - We're being called from vfs_sync() so we don't want
 856  *                     to sleep if we can help it.  All we really need
 857  *                     to do is ensure that the log is synced at least
 858  *                     periodically.  We also push the inodes and
 859  *                     superblock if we can lock them without sleeping
 860  *                      and they are not pinned.
 861  *      SYNC_ATTR    - We need to flush the inodes.  If SYNC_BDFLUSH is not
 862  *                     set, then we really want to lock each inode and flush
 863  *                     it.
 864  *      SYNC_WAIT    - All the flushes that take place in this call should
 865  *                     be synchronous.
 866  *      SYNC_DELWRI  - This tells us to push dirty pages associated with
 867  *                     inodes.  SYNC_WAIT and SYNC_BDFLUSH are used to
 868  *                     determine if they should be flushed sync, async, or
 869  *                     delwri.
 870  *      SYNC_CLOSE   - This flag is passed when the system is being
 871  *                     unmounted.  We should sync and invalidate everthing.
 872  *      SYNC_FSDATA  - This indicates that the caller would like to make
 873  *                     sure the superblock is safe on disk.  We can ensure
 874  *                     this by simply makeing sure the log gets flushed
 875  *                     if SYNC_BDFLUSH is set, and by actually writing it
 876  *                     out otherwise.
 877  *
 878  */
 879 /*ARGSUSED*/
 880 STATIC int
 881 xfs_sync(
 882         bhv_desc_t      *bdp,
 883         int             flags,
 884         cred_t          *credp)
 885 {
 886         xfs_mount_t     *mp;
 887
 888         mp = XFS_BHVTOM(bdp);
 889         return (xfs_syncsub(mp, flags, 0, NULL));
 890 }
 891
 892 /*
 893  * xfs sync routine for internal use
 894  *
 895  * This routine supports all of the flags defined for the generic VFS_SYNC
 896  * interface as explained above under xfs_sync.  In the interests of not
 897  * changing interfaces within the 6.5 family, additional internallly-
 898  * required functions are specified within a separate xflags parameter,
 899  * only available by calling this routine.
 900  *
 901  */
 902 STATIC int
 903 xfs_sync_inodes(
 904         xfs_mount_t     *mp,
 905         int             flags,
 906         int             xflags,
 907         int             *bypassed)
 908 {
 909         xfs_inode_t     *ip = NULL;
 910         xfs_inode_t     *ip_next;
 911         xfs_buf_t       *bp;
 912         vnode_t         *vp = NULL;
 913         vmap_t          vmap;
 914         int             error;
 915         int             last_error;
 916         uint64_t        fflag;
 917         uint            lock_flags;
 918         uint            base_lock_flags;
 919         boolean_t       mount_locked;
 920         boolean_t       vnode_refed;
 921         int             preempt;
 922         xfs_dinode_t    *dip;
 923         xfs_iptr_t      *ipointer;
 924 #ifdef DEBUG
 925         boolean_t       ipointer_in = B_FALSE;
 926
 927 #define IPOINTER_SET    ipointer_in = B_TRUE
 928 #define IPOINTER_CLR    ipointer_in = B_FALSE
 929 #else
 930 #define IPOINTER_SET
 931 #define IPOINTER_CLR
 932 #endif
 933
 934
 935 /* Insert a marker record into the inode list after inode ip. The list
 936  * must be locked when this is called. After the call the list will no
 937  * longer be locked.
 938  */
 939 #define IPOINTER_INSERT(ip, mp) { \
 940                 ASSERT(ipointer_in == B_FALSE); \
 941                 ipointer->ip_mnext = ip->i_mnext; \
 942                 ipointer->ip_mprev = ip; \
 943                 ip->i_mnext = (xfs_inode_t *)ipointer; \
 944                 ipointer->ip_mnext->i_mprev = (xfs_inode_t *)ipointer; \
 945                 preempt = 0; \
 946                 XFS_MOUNT_IUNLOCK(mp); \
 947                 mount_locked = B_FALSE; \
 948                 IPOINTER_SET; \
 949         }
 950
 951 /* Remove the marker from the inode list. If the marker was the only item
 952  * in the list then there are no remaining inodes and we should zero out
 953  * the whole list. If we are the current head of the list then move the head
 954  * past us.
 955  */
 956 #define IPOINTER_REMOVE(ip, mp) { \
 957                 ASSERT(ipointer_in == B_TRUE); \
 958                 if (ipointer->ip_mnext != (xfs_inode_t *)ipointer) { \
 959                         ip = ipointer->ip_mnext; \
 960                         ip->i_mprev = ipointer->ip_mprev; \
 961                         ipointer->ip_mprev->i_mnext = ip; \
 962                         if (mp->m_inodes == (xfs_inode_t *)ipointer) { \
 963                                 mp->m_inodes = ip; \
 964                         } \
 965                 } else { \
 966                         ASSERT(mp->m_inodes == (xfs_inode_t *)ipointer); \
 967                         mp->m_inodes = NULL; \
 968                         ip = NULL; \
 969                 } \
 970                 IPOINTER_CLR; \
 971         }
 972
 973 #define XFS_PREEMPT_MASK        0x7f
 974
 975         if (bypassed)
 976                 *bypassed = 0;
 977         if (XFS_MTOVFS(mp)->vfs_flag & VFS_RDONLY)
 978                 return 0;
 979         error = 0;
 980         last_error = 0;
 981         preempt = 0;
 982
 983         /* Allocate a reference marker */
 984         ipointer = (xfs_iptr_t *)kmem_zalloc(sizeof(xfs_iptr_t), KM_SLEEP);
 985
 986         fflag = XFS_B_ASYNC;            /* default is don't wait */
 987         if (flags & SYNC_BDFLUSH)
 988                 fflag = XFS_B_DELWRI;
 989         if (flags & SYNC_WAIT)
 990                 fflag = 0;              /* synchronous overrides all */
 991
 992         base_lock_flags = XFS_ILOCK_SHARED;
 993         if (flags & (SYNC_DELWRI | SYNC_CLOSE)) {
 994                 /*
 995                  * We need the I/O lock if we're going to call any of
 996                  * the flush/inval routines.
 997                  */
 998                 base_lock_flags |= XFS_IOLOCK_SHARED;
 999         }
1000
1001         XFS_MOUNT_ILOCK(mp);
1002
1003         ip = mp->m_inodes;
1004
1005         mount_locked = B_TRUE;
1006         vnode_refed  = B_FALSE;
1007
1008         IPOINTER_CLR;
1009
1010         do {
1011                 ASSERT(ipointer_in == B_FALSE);
1012                 ASSERT(vnode_refed == B_FALSE);
1013
1014                 lock_flags = base_lock_flags;
1015
1016                 /*
1017                  * There were no inodes in the list, just break out
1018                  * of the loop.
1019                  */
1020                 if (ip == NULL) {
1021                         break;
1022                 }
1023
1024                 /*
1025                  * We found another sync thread marker - skip it
1026                  */
1027                 if (ip->i_mount == NULL) {
1028                         ip = ip->i_mnext;
1029                         continue;
1030                 }
1031
1032                 vp = XFS_ITOV_NULL(ip);
1033
1034                 /*
1035                  * If the vnode is gone then this is being torn down,
1036                  * call reclaim if it is flushed, else let regular flush
1037                  * code deal with it later in the loop.
1038                  */
1039
1040                 if (vp == NULL) {
1041                         /* Skip ones already in reclaim */
1042                         if (ip->i_flags & XFS_IRECLAIM) {
1043                                 ip = ip->i_mnext;
1044                                 continue;
1045                         }
1046                         if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0) {
1047                                 ip = ip->i_mnext;
1048                         } else if ((xfs_ipincount(ip) == 0) &&
1049                                     xfs_iflock_nowait(ip)) {
1050                                 IPOINTER_INSERT(ip, mp);
1051
1052                                 xfs_finish_reclaim(ip, 1,
1053                                                 XFS_IFLUSH_DELWRI_ELSE_ASYNC);
1054
1055                                 XFS_MOUNT_ILOCK(mp);
1056                                 mount_locked = B_TRUE;
1057                                 IPOINTER_REMOVE(ip, mp);
1058                         } else {
1059                                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1060                                 ip = ip->i_mnext;
1061                         }
1062                         continue;
1063                 }
1064
1065                 if (VN_BAD(vp)) {
1066                         ip = ip->i_mnext;
1067                         continue;
1068                 }
1069
1070                 if (XFS_FORCED_SHUTDOWN(mp) && !(flags & SYNC_CLOSE)) {
1071                         XFS_MOUNT_IUNLOCK(mp);
1072                         kmem_free(ipointer, sizeof(xfs_iptr_t));
1073                         return 0;
1074                 }
1075
1076                 /*
1077                  * If this is just vfs_sync() or pflushd() calling
1078                  * then we can skip inodes for which it looks like
1079                  * there is nothing to do.  Since we don't have the
1080                  * inode locked this is racey, but these are periodic
1081                  * calls so it doesn't matter.  For the others we want
1082                  * to know for sure, so we at least try to lock them.
1083                  */
1084                 if (flags & SYNC_BDFLUSH) {
1085                         if (((ip->i_itemp == NULL) ||
1086                              !(ip->i_itemp->ili_format.ilf_fields &
1087                                XFS_ILOG_ALL)) &&
1088                             (ip->i_update_core == 0)) {
1089                                 ip = ip->i_mnext;
1090                                 continue;
1091                         }
1092                 }
1093
1094                 /*
1095                  * Try to lock without sleeping.  We're out of order with
1096                  * the inode list lock here, so if we fail we need to drop
1097                  * the mount lock and try again.  If we're called from
1098                  * bdflush() here, then don't bother.
1099                  *
1100                  * The inode lock here actually coordinates with the
1101                  * almost spurious inode lock in xfs_ireclaim() to prevent
1102                  * the vnode we handle here without a reference from
1103                  * being freed while we reference it.  If we lock the inode
1104                  * while it's on the mount list here, then the spurious inode
1105                  * lock in xfs_ireclaim() after the inode is pulled from
1106                  * the mount list will sleep until we release it here.
1107                  * This keeps the vnode from being freed while we reference
1108                  * it.  It is also cheaper and simpler than actually doing
1109                  * a vn_get() for every inode we touch here.
1110                  */
1111                 if (xfs_ilock_nowait(ip, lock_flags) == 0) {
1112
1113                         if ((flags & SYNC_BDFLUSH) || (vp == NULL)) {
1114                                 ip = ip->i_mnext;
1115                                 continue;
1116                         }
1117
1118                         /*
1119                          * We need to unlock the inode list lock in order
1120                          * to lock the inode. Insert a marker record into
1121                          * the inode list to remember our position, dropping
1122                          * the lock is now done inside the IPOINTER_INSERT
1123                          * macro.
1124                          *
1125                          * We also use the inode list lock to protect us
1126                          * in taking a snapshot of the vnode version number
1127                          * for use in calling vn_get().
1128                          */
1129                         VMAP(vp, vmap);
1130                         IPOINTER_INSERT(ip, mp);
1131
1132                         vp = vn_get(vp, &vmap);
1133                         if (vp == NULL) {
1134                                 /*
1135                                  * The vnode was reclaimed once we let go
1136                                  * of the inode list lock.  Skip to the
1137                                  * next list entry. Remove the marker.
1138                                  */
1139
1140                                 XFS_MOUNT_ILOCK(mp);
1141
1142                                 mount_locked = B_TRUE;
1143                                 vnode_refed  = B_FALSE;
1144
1145                                 IPOINTER_REMOVE(ip, mp);
1146
1147                                 continue;
1148                         }
1149
1150                         xfs_ilock(ip, lock_flags);
1151
1152                         ASSERT(vp == XFS_ITOV(ip));
1153                         ASSERT(ip->i_mount == mp);
1154
1155                         vnode_refed = B_TRUE;
1156                 }
1157
1158                 /* From here on in the loop we may have a marker record
1159                  * in the inode list.
1160                  */
1161
1162                 if ((flags & SYNC_CLOSE)  && (vp != NULL)) {
1163                         /*
1164                          * This is the shutdown case.  We just need to
1165                          * flush and invalidate all the pages associated
1166                          * with the inode.  Drop the inode lock since
1167                          * we can't hold it across calls to the buffer
1168                          * cache.
1169                          *
1170                          * We don't set the VREMAPPING bit in the vnode
1171                          * here, because we don't hold the vnode lock
1172                          * exclusively.  It doesn't really matter, though,
1173                          * because we only come here when we're shutting
1174                          * down anyway.
1175                          */
1176                         xfs_iunlock(ip, XFS_ILOCK_SHARED);
1177
1178                         if (XFS_FORCED_SHUTDOWN(mp)) {
1179                                 VOP_TOSS_PAGES(vp, 0, -1, FI_REMAPF);
1180                         } else {
1181                                 VOP_FLUSHINVAL_PAGES(vp, 0, -1, FI_REMAPF);
1182                         }
1183
1184                         xfs_ilock(ip, XFS_ILOCK_SHARED);
1185
1186                 } else if ((flags & SYNC_DELWRI) && (vp != NULL)) {
1187                         if (VN_DIRTY(vp)) {
1188                                 /* We need to have dropped the lock here,
1189                                  * so insert a marker if we have not already
1190                                  * done so.
1191                                  */
1192                                 if (mount_locked) {
1193                                         IPOINTER_INSERT(ip, mp);
1194                                 }
1195
1196                                 /*
1197                                  * Drop the inode lock since we can't hold it
1198                                  * across calls to the buffer cache.
1199                                  */
1200                                 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1201                                 VOP_FLUSH_PAGES(vp, (xfs_off_t)0, -1,
1202                                                         fflag, FI_NONE, error);
1203                                 xfs_ilock(ip, XFS_ILOCK_SHARED);
1204                         }
1205
1206                 }
1207
1208                 if (flags & SYNC_BDFLUSH) {
1209                         if ((flags & SYNC_ATTR) &&
1210                             ((ip->i_update_core) ||
1211                              ((ip->i_itemp != NULL) &&
1212                               (ip->i_itemp->ili_format.ilf_fields != 0)))) {
1213
1214                                 /* Insert marker and drop lock if not already
1215                                  * done.
1216                                  */
1217                                 if (mount_locked) {
1218                                         IPOINTER_INSERT(ip, mp);
1219                                 }
1220
1221                                 /*
1222                                  * We don't want the periodic flushing of the
1223                                  * inodes by vfs_sync() to interfere with
1224                                  * I/O to the file, especially read I/O
1225                                  * where it is only the access time stamp
1226                                  * that is being flushed out.  To prevent
1227                                  * long periods where we have both inode
1228                                  * locks held shared here while reading the
1229                                  * inode's buffer in from disk, we drop the
1230                                  * inode lock while reading in the inode
1231                                  * buffer.  We have to release the buffer
1232                                  * and reacquire the inode lock so that they
1233                                  * are acquired in the proper order (inode
1234                                  * locks first).  The buffer will go at the
1235                                  * end of the lru chain, though, so we can
1236                                  * expect it to still be there when we go
1237                                  * for it again in xfs_iflush().
1238                                  */
1239                                 if ((xfs_ipincount(ip) == 0) &&
1240                                     xfs_iflock_nowait(ip)) {
1241
1242                                         xfs_ifunlock(ip);
1243                                         xfs_iunlock(ip, XFS_ILOCK_SHARED);
1244
1245                                         error = xfs_itobp(mp, NULL, ip,
1246                                                           &dip, &bp, 0);
1247                                         if (!error) {
1248                                                 xfs_buf_relse(bp);
1249                                         } else {
1250                                                 /* Bailing out, remove the
1251                                                  * marker and free it.
1252                                                  */
1253                                                 XFS_MOUNT_ILOCK(mp);
1254
1255                                                 IPOINTER_REMOVE(ip, mp);
1256
1257                                                 XFS_MOUNT_IUNLOCK(mp);
1258
1259                                                 ASSERT(!(lock_flags &
1260                                                         XFS_IOLOCK_SHARED));
1261
1262                                                 kmem_free(ipointer,
1263                                                         sizeof(xfs_iptr_t));
1264                                                 return (0);
1265                                         }
1266
1267                                         /*
1268                                          * Since we dropped the inode lock,
1269                                          * the inode may have been reclaimed.
1270                                          * Therefore, we reacquire the mount
1271                                          * lock and check to see if we were the
1272                                          * inode reclaimed. If this happened
1273                                          * then the ipointer marker will no
1274                                          * longer point back at us. In this
1275                                          * case, move ip along to the inode
1276                                          * after the marker, remove the marker
1277                                          * and continue.
1278                                          */
1279                                         XFS_MOUNT_ILOCK(mp);
1280                                         mount_locked = B_TRUE;
1281
1282                                         if (ip != ipointer->ip_mprev) {
1283                                                 IPOINTER_REMOVE(ip, mp);
1284
1285                                                 ASSERT(!vnode_refed);
1286                                                 ASSERT(!(lock_flags &
1287                                                         XFS_IOLOCK_SHARED));
1288                                                 continue;
1289                                         }
1290
1291                                         ASSERT(ip->i_mount == mp);
1292
1293                                         if (xfs_ilock_nowait(ip,
1294                                                     XFS_ILOCK_SHARED) == 0) {
1295                                                 ASSERT(ip->i_mount == mp);
1296                                                 /*
1297                                                  * We failed to reacquire
1298                                                  * the inode lock without
1299                                                  * sleeping, so just skip
1300                                                  * the inode for now.  We
1301                                                  * clear the ILOCK bit from
1302                                                  * the lock_flags so that we
1303                                                  * won't try to drop a lock
1304                                                  * we don't hold below.
1305                                                  */
1306                                                 lock_flags &= ~XFS_ILOCK_SHARED;
1307                                                 IPOINTER_REMOVE(ip_next, mp);
1308                                         } else if ((xfs_ipincount(ip) == 0) &&
1309                                                    xfs_iflock_nowait(ip)) {
1310                                                 ASSERT(ip->i_mount == mp);
1311                                                 /*
1312                                                  * Since this is vfs_sync()
1313                                                  * calling we only flush the
1314                                                  * inode out if we can lock
1315                                                  * it without sleeping and
1316                                                  * it is not pinned.  Drop
1317                                                  * the mount lock here so
1318                                                  * that we don't hold it for
1319                                                  * too long. We already have
1320                                                  * a marker in the list here.
1321                                                  */
1322                                                 XFS_MOUNT_IUNLOCK(mp);
1323                                                 mount_locked = B_FALSE;
1324                                                 error = xfs_iflush(ip,
1325                                                            XFS_IFLUSH_DELWRI);
1326                                         } else {
1327                                                 ASSERT(ip->i_mount == mp);
1328                                                 IPOINTER_REMOVE(ip_next, mp);
1329                                         }
1330                                 }
1331
1332                         }
1333
1334                 } else {
1335                         if ((flags & SYNC_ATTR) &&
1336                             ((ip->i_update_core) ||
1337                              ((ip->i_itemp != NULL) &&
1338                               (ip->i_itemp->ili_format.ilf_fields != 0)))) {
1339                                 if (mount_locked) {
1340                                         IPOINTER_INSERT(ip, mp);
1341                                 }
1342
1343                                 if (flags & SYNC_WAIT) {
1344                                         xfs_iflock(ip);
1345                                         error = xfs_iflush(ip,
1346                                                            XFS_IFLUSH_SYNC);
1347                                 } else {
1348                                         /*
1349                                          * If we can't acquire the flush
1350                                          * lock, then the inode is already
1351                                          * being flushed so don't bother
1352                                          * waiting.  If we can lock it then
1353                                          * do a delwri flush so we can
1354                                          * combine multiple inode flushes
1355                                          * in each disk write.
1356                                          */
1357                                         if (xfs_iflock_nowait(ip)) {
1358                                                 error = xfs_iflush(ip,
1359                                                            XFS_IFLUSH_DELWRI);
1360                                         }
1361                                         else if (bypassed)
1362                                                 (*bypassed)++;
1363                                 }
1364                         }
1365                 }
1366
1367                 if (lock_flags != 0) {
1368                         xfs_iunlock(ip, lock_flags);
1369                 }
1370
1371                 if (vnode_refed) {
1372                         /*
1373                          * If we had to take a reference on the vnode
1374                          * above, then wait until after we've unlocked
1375                          * the inode to release the reference.  This is
1376                          * because we can be already holding the inode
1377                          * lock when VN_RELE() calls xfs_inactive().
1378                          *
1379                          * Make sure to drop the mount lock before calling
1380                          * VN_RELE() so that we don't trip over ourselves if
1381                          * we have to go for the mount lock again in the
1382                          * inactive code.
1383                          */
1384                         if (mount_locked) {
1385                                 IPOINTER_INSERT(ip, mp);
1386                         }
1387
1388                         VN_RELE(vp);
1389
1390                         vnode_refed = B_FALSE;
1391                 }
1392
1393                 if (error) {
1394                         last_error = error;
1395                 }
1396
1397                 /*
1398                  * bail out if the filesystem is corrupted.
1399                  */
1400                 if (error == EFSCORRUPTED)  {
1401                         if (!mount_locked) {
1402                                 XFS_MOUNT_ILOCK(mp);
1403                                 IPOINTER_REMOVE(ip, mp);
1404                         }
1405                         XFS_MOUNT_IUNLOCK(mp);
1406                         ASSERT(ipointer_in == B_FALSE);
1407                         kmem_free(ipointer, sizeof(xfs_iptr_t));
1408                         return XFS_ERROR(error);
1409                 }
1410
1411                 /* Let other threads have a chance at the mount lock
1412                  * if we have looped many times without dropping the
1413                  * lock.
1414                  */
1415                 if ((++preempt & XFS_PREEMPT_MASK) == 0) {
1416                         if (mount_locked) {
1417                                 IPOINTER_INSERT(ip, mp);
1418                         }
1419                 }
1420
1421                 if (mount_locked == B_FALSE) {
1422                         XFS_MOUNT_ILOCK(mp);
1423                         mount_locked = B_TRUE;
1424                         IPOINTER_REMOVE(ip, mp);
1425                         continue;
1426                 }
1427
1428                 ASSERT(ipointer_in == B_FALSE);
1429                 ip = ip->i_mnext;
1430
1431         } while (ip != mp->m_inodes);
1432
1433         XFS_MOUNT_IUNLOCK(mp);
1434
1435         ASSERT(ipointer_in == B_FALSE);
1436
1437         kmem_free(ipointer, sizeof(xfs_iptr_t));
1438         return XFS_ERROR(last_error);
1439 }
1440
1441 /*
1442  * xfs sync routine for internal use
1443  *
1444  * This routine supports all of the flags defined for the generic VFS_SYNC
1445  * interface as explained above under xfs_sync.  In the interests of not
1446  * changing interfaces within the 6.5 family, additional internallly-
1447  * required functions are specified within a separate xflags parameter,
1448  * only available by calling this routine.
1449  *
1450  */
1451 int
1452 xfs_syncsub(
1453         xfs_mount_t     *mp,
1454         int             flags,
1455         int             xflags,
1456         int             *bypassed)
1457 {
1458         int             error = 0;
1459         int             last_error = 0;
1460         uint            log_flags = XFS_LOG_FORCE;
1461         xfs_buf_t       *bp;
1462         xfs_buf_log_item_t      *bip;
1463
1464         /*
1465          * Sync out the log.  This ensures that the log is periodically
1466          * flushed even if there is not enough activity to fill it up.
1467          */
1468         if (flags & SYNC_WAIT)
1469                 log_flags |= XFS_LOG_SYNC;
1470
1471         xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1472
1473         if (flags & (SYNC_ATTR|SYNC_DELWRI)) {
1474                 if (flags & SYNC_BDFLUSH)
1475                         xfs_finish_reclaim_all(mp, 1);
1476                 else
1477                         error = xfs_sync_inodes(mp, flags, xflags, bypassed);
1478         }
1479
1480         /*
1481          * Flushing out dirty data above probably generated more
1482          * log activity, so if this isn't vfs_sync() then flush
1483          * the log again.
1484          */
1485         if (flags & SYNC_DELWRI) {
1486                 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1487         }
1488
1489         if (flags & SYNC_FSDATA) {
1490                 /*
1491                  * If this is vfs_sync() then only sync the superblock
1492                  * if we can lock it without sleeping and it is not pinned.
1493                  */
1494                 if (flags & SYNC_BDFLUSH) {
1495                         bp = xfs_getsb(mp, XFS_BUF_TRYLOCK);
1496                         if (bp != NULL) {
1497                                 bip = XFS_BUF_FSPRIVATE(bp,xfs_buf_log_item_t*);
1498                                 if ((bip != NULL) &&
1499                                     xfs_buf_item_dirty(bip)) {
1500                                         if (!(XFS_BUF_ISPINNED(bp))) {
1501                                                 XFS_BUF_ASYNC(bp);
1502                                                 error = xfs_bwrite(mp, bp);
1503                                         } else {
1504                                                 xfs_buf_relse(bp);
1505                                         }
1506                                 } else {
1507                                         xfs_buf_relse(bp);
1508                                 }
1509                         }
1510                 } else {
1511                         bp = xfs_getsb(mp, 0);
1512                         /*
1513                          * If the buffer is pinned then push on the log so
1514                          * we won't get stuck waiting in the write for
1515                          * someone, maybe ourselves, to flush the log.
1516                          * Even though we just pushed the log above, we
1517                          * did not have the superblock buffer locked at
1518                          * that point so it can become pinned in between
1519                          * there and here.
1520                          */
1521                         if (XFS_BUF_ISPINNED(bp))
1522                                 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
1523                         if (flags & SYNC_WAIT)
1524                                 XFS_BUF_UNASYNC(bp);
1525                         else
1526                                 XFS_BUF_ASYNC(bp);
1527                         error = xfs_bwrite(mp, bp);
1528                 }
1529                 if (error) {
1530                         last_error = error;
1531                 }
1532         }
1533
1534         /*
1535          * If this is the periodic sync, then kick some entries out of
1536          * the reference cache.  This ensures that idle entries are
1537          * eventually kicked out of the cache.
1538          */
1539         if (flags & SYNC_REFCACHE) {
1540                 xfs_refcache_purge_some(mp);
1541         }
1542
1543         /*
1544          * Now check to see if the log needs a "dummy" transaction.
1545          */
1546
1547         if (!(flags & SYNC_REMOUNT) && xfs_log_need_covered(mp)) {
1548                 xfs_trans_t *tp;
1549                 xfs_inode_t *ip;
1550
1551                 /*
1552                  * Put a dummy transaction in the log to tell
1553                  * recovery that all others are OK.
1554                  */
1555                 tp = xfs_trans_alloc(mp, XFS_TRANS_DUMMY1);
1556                 if ((error = xfs_trans_reserve(tp, 0,
1557                                 XFS_ICHANGE_LOG_RES(mp),
1558                                 0, 0, 0)))  {
1559                         xfs_trans_cancel(tp, 0);
1560                         return error;
1561                 }
1562
1563                 ip = mp->m_rootip;
1564                 xfs_ilock(ip, XFS_ILOCK_EXCL);
1565
1566                 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1567                 xfs_trans_ihold(tp, ip);
1568                 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1569                 error = xfs_trans_commit(tp, 0, NULL);
1570                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1571                 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1572         }
1573
1574         /*
1575          * When shutting down, we need to insure that the AIL is pushed
1576          * to disk or the filesystem can appear corrupt from the PROM.
1577          */
1578         if ((flags & (SYNC_CLOSE|SYNC_WAIT)) == (SYNC_CLOSE|SYNC_WAIT)) {
1579                 XFS_bflush(mp->m_ddev_targp);
1580                 if (mp->m_rtdev_targp) {
1581                         XFS_bflush(mp->m_rtdev_targp);
1582                 }
1583         }
1584
1585         return XFS_ERROR(last_error);
1586 }
1587
1588 /*
1589  * xfs_vget - called by DMAPI and NFSD to get vnode from file handle
1590  */
1591 STATIC int
1592 xfs_vget(
1593         bhv_desc_t      *bdp,
1594         vnode_t         **vpp,
1595         fid_t           *fidp)
1596 {
1597         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
1598         xfs_fid_t       *xfid = (struct xfs_fid *)fidp;
1599         xfs_inode_t     *ip;
1600         int             error;
1601         xfs_ino_t       ino;
1602         unsigned int    igen;
1603
1604         /*
1605          * Invalid.  Since handles can be created in user space and passed in
1606          * via gethandle(), this is not cause for a panic.
1607          */
1608         if (xfid->xfs_fid_len != sizeof(*xfid) - sizeof(xfid->xfs_fid_len))
1609                 return XFS_ERROR(EINVAL);
1610
1611         ino  = xfid->xfs_fid_ino;
1612         igen = xfid->xfs_fid_gen;
1613
1614         /*
1615          * NFS can sometimes send requests for ino 0.  Fail them gracefully.
1616          */
1617         if (ino == 0)
1618                 return XFS_ERROR(ESTALE);
1619
1620         error = xfs_iget(mp, NULL, ino, 0, XFS_ILOCK_SHARED, &ip, 0);
1621         if (error) {
1622                 *vpp = NULL;
1623                 return error;
1624         }
1625
1626         if (ip == NULL) {
1627                 *vpp = NULL;
1628                 return XFS_ERROR(EIO);
1629         }
1630
1631         if (ip->i_d.di_mode == 0 || ip->i_d.di_gen != igen) {
1632                 xfs_iput_new(ip, XFS_ILOCK_SHARED);
1633                 *vpp = NULL;
1634                 return XFS_ERROR(ENOENT);
1635         }
1636
1637         *vpp = XFS_ITOV(ip);
1638         xfs_iunlock(ip, XFS_ILOCK_SHARED);
1639         return 0;
1640 }
1641
1642
1643 #define MNTOPT_LOGBUFS  "logbufs"       /* number of XFS log buffers */
1644 #define MNTOPT_LOGBSIZE "logbsize"      /* size of XFS log buffers */
1645 #define MNTOPT_LOGDEV   "logdev"        /* log device */
1646 #define MNTOPT_RTDEV    "rtdev"         /* realtime I/O device */
1647 #define MNTOPT_BIOSIZE  "biosize"       /* log2 of preferred buffered io size */
1648 #define MNTOPT_WSYNC    "wsync"         /* safe-mode nfs compatible mount */
1649 #define MNTOPT_INO64    "ino64"         /* force inodes into 64-bit range */
1650 #define MNTOPT_NOALIGN  "noalign"       /* turn off stripe alignment */
1651 #define MNTOPT_SWALLOC  "swalloc"       /* turn on stripe width allocation */
1652 #define MNTOPT_SUNIT    "sunit"         /* data volume stripe unit */
1653 #define MNTOPT_SWIDTH   "swidth"        /* data volume stripe width */
1654 #define MNTOPT_NOUUID   "nouuid"        /* ignore filesystem UUID */
1655 #define MNTOPT_MTPT     "mtpt"          /* filesystem mount point */
1656 #define MNTOPT_ALLOCSIZE    "allocsize"    /* preferred allocation size */
1657 #define MNTOPT_IHASHSIZE    "ihashsize"    /* size of inode hash table */
1658 #define MNTOPT_NORECOVERY   "norecovery"   /* don't run XFS recovery */
1659 #define MNTOPT_NOLOGFLUSH   "nologflush"   /* don't hard flush on log writes */
1660 #define MNTOPT_OSYNCISOSYNC "osyncisosync" /* o_sync is REALLY o_sync */
1661 #define MNTOPT_64BITINODE   "inode64"   /* inodes can be allocated anywhere */
1662 #define MNTOPT_IKEEP    "ikeep"         /* do not free empty inode clusters */
1663 #define MNTOPT_NOIKEEP  "noikeep"       /* free empty inode clusters */
1664 #define MNTOPT_TAGXID   "tagxid"        /* context xid tagging for inodes */
1665
1666 STATIC unsigned long
1667 suffix_strtoul(const char *cp, char **endp, unsigned int base)
1668 {
1669         int     last, shift_left_factor = 0;
1670         char    *value = (char *)cp;
1671
1672         last = strlen(value) - 1;
1673         if (value[last] == 'K' || value[last] == 'k') {
1674                 shift_left_factor = 10;
1675                 value[last] = '\0';
1676         }
1677         if (value[last] == 'M' || value[last] == 'm') {
1678                 shift_left_factor = 20;
1679                 value[last] = '\0';
1680         }
1681         if (value[last] == 'G' || value[last] == 'g') {
1682                 shift_left_factor = 30;
1683                 value[last] = '\0';
1684         }
1685
1686         return simple_strtoul(cp, endp, base) << shift_left_factor;
1687 }
1688
1689 int
1690 xfs_parseargs(
1691         struct bhv_desc         *bhv,
1692         char                    *options,
1693         struct xfs_mount_args   *args,
1694         int                     update)
1695 {
1696         struct vfs              *vfsp = bhvtovfs(bhv);
1697         char                    *this_char, *value, *eov;
1698         int                     dsunit, dswidth, vol_dsunit, vol_dswidth;
1699         int                     iosize;
1700
1701 #if 0   /* XXX: off by default, until some remaining issues ironed out */
1702         args->flags |= XFSMNT_IDELETE; /* default to on */
1703 #endif
1704
1705         if (!options)
1706                 return 0;
1707
1708         iosize = dsunit = dswidth = vol_dsunit = vol_dswidth = 0;
1709
1710         while ((this_char = strsep(&options, ",")) != NULL) {
1711                 if (!*this_char)
1712                         continue;
1713                 if ((value = strchr(this_char, '=')) != NULL)
1714                         *value++ = 0;
1715
1716                 if (!strcmp(this_char, MNTOPT_LOGBUFS)) {
1717                         if (!value || !*value) {
1718                                 printk("XFS: %s option requires an argument\n",
1719                                         this_char);
1720                                 return EINVAL;
1721                         }
1722                         args->logbufs = simple_strtoul(value, &eov, 10);
1723                 } else if (!strcmp(this_char, MNTOPT_LOGBSIZE)) {
1724                         if (!value || !*value) {
1725                                 printk("XFS: %s option requires an argument\n",
1726                                         this_char);
1727                                 return EINVAL;
1728                         }
1729                         args->logbufsize = suffix_strtoul(value, &eov, 10);
1730                 } else if (!strcmp(this_char, MNTOPT_LOGDEV)) {
1731                         if (!value || !*value) {
1732                                 printk("XFS: %s option requires an argument\n",
1733                                         this_char);
1734                                 return EINVAL;
1735                         }
1736                         strncpy(args->logname, value, MAXNAMELEN);
1737                 } else if (!strcmp(this_char, MNTOPT_MTPT)) {
1738                         if (!value || !*value) {
1739                                 printk("XFS: %s option requires an argument\n",
1740                                         this_char);
1741                                 return EINVAL;
1742                         }
1743                         strncpy(args->mtpt, value, MAXNAMELEN);
1744                 } else if (!strcmp(this_char, MNTOPT_RTDEV)) {
1745                         if (!value || !*value) {
1746                                 printk("XFS: %s option requires an argument\n",
1747                                         this_char);
1748                                 return EINVAL;
1749                         }
1750                         strncpy(args->rtname, value, MAXNAMELEN);
1751                 } else if (!strcmp(this_char, MNTOPT_BIOSIZE)) {
1752                         if (!value || !*value) {
1753                                 printk("XFS: %s option requires an argument\n",
1754                                         this_char);
1755                                 return EINVAL;
1756                         }
1757                         iosize = simple_strtoul(value, &eov, 10);
1758                         args->flags |= XFSMNT_IOSIZE;
1759                         args->iosizelog = (uint8_t) iosize;
1760                 } else if (!strcmp(this_char, MNTOPT_ALLOCSIZE)) {
1761                         if (!value || !*value) {
1762                                 printk("XFS: %s option requires an argument\n",
1763                                         this_char);
1764                                 return EINVAL;
1765                         }
1766                         iosize = suffix_strtoul(value, &eov, 10);
1767                         args->flags |= XFSMNT_IOSIZE;
1768                         args->iosizelog = ffs(iosize) - 1;
1769                 } else if (!strcmp(this_char, MNTOPT_IHASHSIZE)) {
1770                         if (!value || !*value) {
1771                                 printk("XFS: %s option requires an argument\n",
1772                                         this_char);
1773                                 return EINVAL;
1774                         }
1775                         args->flags |= XFSMNT_IHASHSIZE;
1776                         args->ihashsize = simple_strtoul(value, &eov, 10);
1777                 } else if (!strcmp(this_char, MNTOPT_WSYNC)) {
1778                         args->flags |= XFSMNT_WSYNC;
1779                 } else if (!strcmp(this_char, MNTOPT_OSYNCISOSYNC)) {
1780                         args->flags |= XFSMNT_OSYNCISOSYNC;
1781                 } else if (!strcmp(this_char, MNTOPT_NORECOVERY)) {
1782                         args->flags |= XFSMNT_NORECOVERY;
1783                 } else if (!strcmp(this_char, MNTOPT_INO64)) {
1784                         args->flags |= XFSMNT_INO64;
1785 #if !XFS_BIG_INUMS
1786                         printk("XFS: %s option not allowed on this system\n",
1787                                 this_char);
1788                         return EINVAL;
1789 #endif
1790                 } else if (!strcmp(this_char, MNTOPT_NOALIGN)) {
1791                         args->flags |= XFSMNT_NOALIGN;
1792                 } else if (!strcmp(this_char, MNTOPT_SWALLOC)) {
1793                         args->flags |= XFSMNT_SWALLOC;
1794                 } else if (!strcmp(this_char, MNTOPT_SUNIT)) {
1795                         if (!value || !*value) {
1796                                 printk("XFS: %s option requires an argument\n",
1797                                         this_char);
1798                                 return EINVAL;
1799                         }
1800                         dsunit = simple_strtoul(value, &eov, 10);
1801                 } else if (!strcmp(this_char, MNTOPT_SWIDTH)) {
1802                         if (!value || !*value) {
1803                                 printk("XFS: %s option requires an argument\n",
1804                                         this_char);
1805                                 return EINVAL;
1806                         }
1807                         dswidth = simple_strtoul(value, &eov, 10);
1808                 } else if (!strcmp(this_char, MNTOPT_64BITINODE)) {
1809                         args->flags &= ~XFSMNT_32BITINODES;
1810 #if !XFS_BIG_INUMS
1811                         printk("XFS: %s option not allowed on this system\n",
1812                                 this_char);
1813                         return EINVAL;
1814 #endif
1815                 } else if (!strcmp(this_char, MNTOPT_NOUUID)) {
1816                         args->flags |= XFSMNT_NOUUID;
1817                 } else if (!strcmp(this_char, MNTOPT_NOLOGFLUSH)) {
1818                         args->flags |= XFSMNT_NOLOGFLUSH;
1819                 } else if (!strcmp(this_char, MNTOPT_IKEEP)) {
1820                         args->flags &= ~XFSMNT_IDELETE;
1821                 } else if (!strcmp(this_char, MNTOPT_NOIKEEP)) {
1822                         args->flags |= XFSMNT_IDELETE;
1823                 } else if (!strcmp(this_char, MNTOPT_TAGXID)) {
1824                         args->flags |= XFSMNT_TAGXID;
1825                 } else if (!strcmp(this_char, "osyncisdsync")) {
1826                         /* no-op, this is now the default */
1827 printk("XFS: osyncisdsync is now the default, option is deprecated.\n");
1828                 } else if (!strcmp(this_char, "irixsgid")) {
1829 printk("XFS: irixsgid is now a sysctl(2) variable, option is deprecated.\n");
1830                 } else {
1831                         printk("XFS: unknown mount option [%s].\n", this_char);
1832                         return EINVAL;
1833                 }
1834         }
1835
1836         if (args->flags & XFSMNT_NORECOVERY) {
1837                 if ((vfsp->vfs_flag & VFS_RDONLY) == 0) {
1838                         printk("XFS: no-recovery mounts must be read-only.\n");
1839                         return EINVAL;
1840                 }
1841         }
1842
1843         if ((args->flags & XFSMNT_NOALIGN) && (dsunit || dswidth)) {
1844                 printk(
1845         "XFS: sunit and swidth options incompatible with the noalign option\n");
1846                 return EINVAL;
1847         }
1848
1849         if ((dsunit && !dswidth) || (!dsunit && dswidth)) {
1850                 printk("XFS: sunit and swidth must be specified together\n");
1851                 return EINVAL;
1852         }
1853
1854         if (dsunit && (dswidth % dsunit != 0)) {
1855                 printk(
1856         "XFS: stripe width (%d) must be a multiple of the stripe unit (%d)\n",
1857                         dswidth, dsunit);
1858                 return EINVAL;
1859         }
1860
1861         if ((args->flags & XFSMNT_NOALIGN) != XFSMNT_NOALIGN) {
1862                 if (dsunit) {
1863                         args->sunit = dsunit;
1864                         args->flags |= XFSMNT_RETERR;
1865                 } else {
1866                         args->sunit = vol_dsunit;
1867                 }
1868                 dswidth ? (args->swidth = dswidth) :
1869                           (args->swidth = vol_dswidth);
1870         } else {
1871                 args->sunit = args->swidth = 0;
1872         }
1873
1874         return 0;
1875 }
1876
1877 int
1878 xfs_showargs(
1879         struct bhv_desc         *bhv,
1880         struct seq_file         *m)
1881 {
1882         static struct proc_xfs_info {
1883                 int     flag;
1884                 char    *str;
1885         } xfs_info[] = {
1886                 /* the few simple ones we can get from the mount struct */
1887                 { XFS_MOUNT_WSYNC,              "," MNTOPT_WSYNC },
1888                 { XFS_MOUNT_INO64,              "," MNTOPT_INO64 },
1889                 { XFS_MOUNT_NOALIGN,            "," MNTOPT_NOALIGN },
1890                 { XFS_MOUNT_SWALLOC,            "," MNTOPT_SWALLOC },
1891                 { XFS_MOUNT_NOUUID,             "," MNTOPT_NOUUID },
1892                 { XFS_MOUNT_NORECOVERY,         "," MNTOPT_NORECOVERY },
1893                 { XFS_MOUNT_OSYNCISOSYNC,       "," MNTOPT_OSYNCISOSYNC },
1894                 { XFS_MOUNT_NOLOGFLUSH,         "," MNTOPT_NOLOGFLUSH },
1895                 { XFS_MOUNT_IDELETE,            "," MNTOPT_NOIKEEP },
1896                 { 0, NULL }
1897         };
1898         struct proc_xfs_info    *xfs_infop;
1899         struct xfs_mount        *mp = XFS_BHVTOM(bhv);
1900
1901         for (xfs_infop = xfs_info; xfs_infop->flag; xfs_infop++) {
1902                 if (mp->m_flags & xfs_infop->flag)
1903                         seq_puts(m, xfs_infop->str);
1904         }
1905
1906         if (mp->m_flags & XFS_MOUNT_IHASHSIZE)
1907                 seq_printf(m, "," MNTOPT_IHASHSIZE "=%d", mp->m_ihsize);
1908
1909         if (mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)
1910                 seq_printf(m, "," MNTOPT_ALLOCSIZE "=%d", 1<<mp->m_writeio_log);
1911
1912         if (mp->m_logbufs > 0)
1913                 seq_printf(m, "," MNTOPT_LOGBUFS "=%d", mp->m_logbufs);
1914
1915         if (mp->m_logbsize > 0)
1916                 seq_printf(m, "," MNTOPT_LOGBSIZE "=%d", mp->m_logbsize);
1917
1918         if (mp->m_ddev_targp != mp->m_logdev_targp)
1919                 seq_printf(m, "," MNTOPT_LOGDEV "=%s",
1920                                 XFS_BUFTARG_NAME(mp->m_logdev_targp));
1921
1922         if (mp->m_rtdev_targp && mp->m_ddev_targp != mp->m_rtdev_targp)
1923                 seq_printf(m, "," MNTOPT_RTDEV "=%s",
1924                                 XFS_BUFTARG_NAME(mp->m_rtdev_targp));
1925
1926         if (mp->m_dalign > 0)
1927                 seq_printf(m, "," MNTOPT_SUNIT "=%d",
1928                                 (int)XFS_FSB_TO_BB(mp, mp->m_dalign));
1929
1930         if (mp->m_swidth > 0)
1931                 seq_printf(m, "," MNTOPT_SWIDTH "=%d",
1932                                 (int)XFS_FSB_TO_BB(mp, mp->m_swidth));
1933
1934         if (!(mp->m_flags & XFS_MOUNT_32BITINOOPT))
1935                 seq_printf(m, "," MNTOPT_64BITINODE);
1936
1937         return 0;
1938 }
1939
1940 STATIC void
1941 xfs_freeze(
1942         bhv_desc_t      *bdp)
1943 {
1944         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
1945
1946         while (atomic_read(&mp->m_active_trans) > 0)
1947                 delay(100);
1948
1949         /* Push the superblock and write an unmount record */
1950         xfs_log_unmount_write(mp);
1951         xfs_unmountfs_writesb(mp);
1952 }
1953
1954
1955 vfsops_t xfs_vfsops = {
1956         BHV_IDENTITY_INIT(VFS_BHV_XFS,VFS_POSITION_XFS),
1957         .vfs_parseargs          = xfs_parseargs,
1958         .vfs_showargs           = xfs_showargs,
1959         .vfs_mount              = xfs_mount,
1960         .vfs_unmount            = xfs_unmount,
1961         .vfs_mntupdate          = xfs_mntupdate,
1962         .vfs_root               = xfs_root,
1963         .vfs_statvfs            = xfs_statvfs,
1964         .vfs_sync               = xfs_sync,
1965         .vfs_vget               = xfs_vget,
1966         .vfs_dmapiops           = (vfs_dmapiops_t)fs_nosys,
1967         .vfs_quotactl           = (vfs_quotactl_t)fs_nosys,
1968         .vfs_init_vnode         = xfs_initialize_vnode,
1969         .vfs_force_shutdown     = xfs_do_force_shutdown,
1970         .vfs_freeze             = xfs_freeze,
1971 };