fs/xfs/xfs_vfsops.c

   1 /*
   2  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
   3  * All Rights Reserved.
   4  *
   5  * This program is free software; you can redistribute it and/or
   6  * modify it under the terms of the GNU General Public License as
   7  * published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it would be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write the Free Software Foundation,
  16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17  */
  18 #include "xfs.h"
  19 #include "xfs_fs.h"
  20 #include "xfs_types.h"
  21 #include "xfs_bit.h"
  22 #include "xfs_log.h"
  23 #include "xfs_inum.h"
  24 #include "xfs_trans.h"
  25 #include "xfs_sb.h"
  26 #include "xfs_ag.h"
  27 #include "xfs_dir.h"
  28 #include "xfs_dir2.h"
  29 #include "xfs_dmapi.h"
  30 #include "xfs_mount.h"
  31 #include "xfs_da_btree.h"
  32 #include "xfs_bmap_btree.h"
  33 #include "xfs_ialloc_btree.h"
  34 #include "xfs_alloc_btree.h"
  35 #include "xfs_dir_sf.h"
  36 #include "xfs_dir2_sf.h"
  37 #include "xfs_attr_sf.h"
  38 #include "xfs_dinode.h"
  39 #include "xfs_inode.h"
  40 #include "xfs_inode_item.h"
  41 #include "xfs_btree.h"
  42 #include "xfs_alloc.h"
  43 #include "xfs_ialloc.h"
  44 #include "xfs_quota.h"
  45 #include "xfs_error.h"
  46 #include "xfs_bmap.h"
  47 #include "xfs_rw.h"
  48 #include "xfs_refcache.h"
  49 #include "xfs_buf_item.h"
  50 #include "xfs_log_priv.h"
  51 #include "xfs_dir2_trace.h"
  52 #include "xfs_extfree_item.h"
  53 #include "xfs_acl.h"
  54 #include "xfs_attr.h"
  55 #include "xfs_clnt.h"
  56 #include "xfs_fsops.h"
  57
  58 STATIC int      xfs_sync(bhv_desc_t *, int, cred_t *);
  59
  60 int
  61 xfs_init(void)
  62 {
  63         extern kmem_zone_t      *xfs_bmap_free_item_zone;
  64         extern kmem_zone_t      *xfs_btree_cur_zone;
  65         extern kmem_zone_t      *xfs_trans_zone;
  66         extern kmem_zone_t      *xfs_buf_item_zone;
  67         extern kmem_zone_t      *xfs_dabuf_zone;
  68 #ifdef XFS_DABUF_DEBUG
  69         extern lock_t           xfs_dabuf_global_lock;
  70         spinlock_init(&xfs_dabuf_global_lock, "xfsda");
  71 #endif
  72
  73         /*
  74          * Initialize all of the zone allocators we use.
  75          */
  76         xfs_bmap_free_item_zone = kmem_zone_init(sizeof(xfs_bmap_free_item_t),
  77                                                  "xfs_bmap_free_item");
  78         xfs_btree_cur_zone = kmem_zone_init(sizeof(xfs_btree_cur_t),
  79                                             "xfs_btree_cur");
  80         xfs_trans_zone = kmem_zone_init(sizeof(xfs_trans_t), "xfs_trans");
  81         xfs_da_state_zone =
  82                 kmem_zone_init(sizeof(xfs_da_state_t), "xfs_da_state");
  83         xfs_dabuf_zone = kmem_zone_init(sizeof(xfs_dabuf_t), "xfs_dabuf");
  84         xfs_ifork_zone = kmem_zone_init(sizeof(xfs_ifork_t), "xfs_ifork");
  85         xfs_acl_zone_init(xfs_acl_zone, "xfs_acl");
  86
  87         /*
  88          * The size of the zone allocated buf log item is the maximum
  89          * size possible under XFS.  This wastes a little bit of memory,
  90          * but it is much faster.
  91          */
  92         xfs_buf_item_zone =
  93                 kmem_zone_init((sizeof(xfs_buf_log_item_t) +
  94                                 (((XFS_MAX_BLOCKSIZE / XFS_BLI_CHUNK) /
  95                                   NBWORD) * sizeof(int))),
  96                                "xfs_buf_item");
  97         xfs_efd_zone =
  98                 kmem_zone_init((sizeof(xfs_efd_log_item_t) +
  99                                ((XFS_EFD_MAX_FAST_EXTENTS - 1) *
 100                                  sizeof(xfs_extent_t))),
 101                                       "xfs_efd_item");
 102         xfs_efi_zone =
 103                 kmem_zone_init((sizeof(xfs_efi_log_item_t) +
 104                                ((XFS_EFI_MAX_FAST_EXTENTS - 1) *
 105                                  sizeof(xfs_extent_t))),
 106                                       "xfs_efi_item");
 107
 108         /*
 109          * These zones warrant special memory allocator hints
 110          */
 111         xfs_inode_zone =
 112                 kmem_zone_init_flags(sizeof(xfs_inode_t), "xfs_inode",
 113                                         KM_ZONE_HWALIGN | KM_ZONE_RECLAIM |
 114                                         KM_ZONE_SPREAD, NULL);
 115         xfs_ili_zone =
 116                 kmem_zone_init_flags(sizeof(xfs_inode_log_item_t), "xfs_ili",
 117                                         KM_ZONE_SPREAD, NULL);
 118         xfs_chashlist_zone =
 119                 kmem_zone_init_flags(sizeof(xfs_chashlist_t), "xfs_chashlist",
 120                                         KM_ZONE_SPREAD, NULL);
 121
 122         /*
 123          * Allocate global trace buffers.
 124          */
 125 #ifdef XFS_ALLOC_TRACE
 126         xfs_alloc_trace_buf = ktrace_alloc(XFS_ALLOC_TRACE_SIZE, KM_SLEEP);
 127 #endif
 128 #ifdef XFS_BMAP_TRACE
 129         xfs_bmap_trace_buf = ktrace_alloc(XFS_BMAP_TRACE_SIZE, KM_SLEEP);
 130 #endif
 131 #ifdef XFS_BMBT_TRACE
 132         xfs_bmbt_trace_buf = ktrace_alloc(XFS_BMBT_TRACE_SIZE, KM_SLEEP);
 133 #endif
 134 #ifdef XFS_DIR_TRACE
 135         xfs_dir_trace_buf = ktrace_alloc(XFS_DIR_TRACE_SIZE, KM_SLEEP);
 136 #endif
 137 #ifdef XFS_ATTR_TRACE
 138         xfs_attr_trace_buf = ktrace_alloc(XFS_ATTR_TRACE_SIZE, KM_SLEEP);
 139 #endif
 140 #ifdef XFS_DIR2_TRACE
 141         xfs_dir2_trace_buf = ktrace_alloc(XFS_DIR2_GTRACE_SIZE, KM_SLEEP);
 142 #endif
 143
 144         xfs_dir_startup();
 145
 146 #if (defined(DEBUG) || defined(INDUCE_IO_ERROR))
 147         xfs_error_test_init();
 148 #endif /* DEBUG || INDUCE_IO_ERROR */
 149
 150         xfs_init_procfs();
 151         xfs_sysctl_register();
 152         return 0;
 153 }
 154
 155 void
 156 xfs_cleanup(void)
 157 {
 158         extern kmem_zone_t      *xfs_bmap_free_item_zone;
 159         extern kmem_zone_t      *xfs_btree_cur_zone;
 160         extern kmem_zone_t      *xfs_inode_zone;
 161         extern kmem_zone_t      *xfs_trans_zone;
 162         extern kmem_zone_t      *xfs_da_state_zone;
 163         extern kmem_zone_t      *xfs_dabuf_zone;
 164         extern kmem_zone_t      *xfs_efd_zone;
 165         extern kmem_zone_t      *xfs_efi_zone;
 166         extern kmem_zone_t      *xfs_buf_item_zone;
 167         extern kmem_zone_t      *xfs_chashlist_zone;
 168
 169         xfs_cleanup_procfs();
 170         xfs_sysctl_unregister();
 171         xfs_refcache_destroy();
 172         xfs_acl_zone_destroy(xfs_acl_zone);
 173
 174 #ifdef XFS_DIR2_TRACE
 175         ktrace_free(xfs_dir2_trace_buf);
 176 #endif
 177 #ifdef XFS_ATTR_TRACE
 178         ktrace_free(xfs_attr_trace_buf);
 179 #endif
 180 #ifdef XFS_DIR_TRACE
 181         ktrace_free(xfs_dir_trace_buf);
 182 #endif
 183 #ifdef XFS_BMBT_TRACE
 184         ktrace_free(xfs_bmbt_trace_buf);
 185 #endif
 186 #ifdef XFS_BMAP_TRACE
 187         ktrace_free(xfs_bmap_trace_buf);
 188 #endif
 189 #ifdef XFS_ALLOC_TRACE
 190         ktrace_free(xfs_alloc_trace_buf);
 191 #endif
 192
 193         kmem_zone_destroy(xfs_bmap_free_item_zone);
 194         kmem_zone_destroy(xfs_btree_cur_zone);
 195         kmem_zone_destroy(xfs_inode_zone);
 196         kmem_zone_destroy(xfs_trans_zone);
 197         kmem_zone_destroy(xfs_da_state_zone);
 198         kmem_zone_destroy(xfs_dabuf_zone);
 199         kmem_zone_destroy(xfs_buf_item_zone);
 200         kmem_zone_destroy(xfs_efd_zone);
 201         kmem_zone_destroy(xfs_efi_zone);
 202         kmem_zone_destroy(xfs_ifork_zone);
 203         kmem_zone_destroy(xfs_ili_zone);
 204         kmem_zone_destroy(xfs_chashlist_zone);
 205 }
 206
 207 /*
 208  * xfs_start_flags
 209  *
 210  * This function fills in xfs_mount_t fields based on mount args.
 211  * Note: the superblock has _not_ yet been read in.
 212  */
 213 STATIC int
 214 xfs_start_flags(
 215         struct vfs              *vfs,
 216         struct xfs_mount_args   *ap,
 217         struct xfs_mount        *mp)
 218 {
 219         /* Values are in BBs */
 220         if ((ap->flags & XFSMNT_NOALIGN) != XFSMNT_NOALIGN) {
 221                 /*
 222                  * At this point the superblock has not been read
 223                  * in, therefore we do not know the block size.
 224                  * Before the mount call ends we will convert
 225                  * these to FSBs.
 226                  */
 227                 mp->m_dalign = ap->sunit;
 228                 mp->m_swidth = ap->swidth;
 229         }
 230
 231         if (ap->logbufs != -1 &&
 232             ap->logbufs != 0 &&
 233             (ap->logbufs < XLOG_MIN_ICLOGS ||
 234              ap->logbufs > XLOG_MAX_ICLOGS)) {
 235                 cmn_err(CE_WARN,
 236                         "XFS: invalid logbufs value: %d [not %d-%d]",
 237                         ap->logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
 238                 return XFS_ERROR(EINVAL);
 239         }
 240         mp->m_logbufs = ap->logbufs;
 241         if (ap->logbufsize != -1 &&
 242             ap->logbufsize !=  0 &&
 243             ap->logbufsize != 16 * 1024 &&
 244             ap->logbufsize != 32 * 1024 &&
 245             ap->logbufsize != 64 * 1024 &&
 246             ap->logbufsize != 128 * 1024 &&
 247             ap->logbufsize != 256 * 1024) {
 248                 cmn_err(CE_WARN,
 249         "XFS: invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
 250                         ap->logbufsize);
 251                 return XFS_ERROR(EINVAL);
 252         }
 253         mp->m_ihsize = ap->ihashsize;
 254         mp->m_logbsize = ap->logbufsize;
 255         mp->m_fsname_len = strlen(ap->fsname) + 1;
 256         mp->m_fsname = kmem_alloc(mp->m_fsname_len, KM_SLEEP);
 257         strcpy(mp->m_fsname, ap->fsname);
 258         if (ap->rtname[0]) {
 259                 mp->m_rtname = kmem_alloc(strlen(ap->rtname) + 1, KM_SLEEP);
 260                 strcpy(mp->m_rtname, ap->rtname);
 261         }
 262         if (ap->logname[0]) {
 263                 mp->m_logname = kmem_alloc(strlen(ap->logname) + 1, KM_SLEEP);
 264                 strcpy(mp->m_logname, ap->logname);
 265         }
 266
 267         if (ap->flags & XFSMNT_WSYNC)
 268                 mp->m_flags |= XFS_MOUNT_WSYNC;
 269 #if XFS_BIG_INUMS
 270         if (ap->flags & XFSMNT_INO64) {
 271                 mp->m_flags |= XFS_MOUNT_INO64;
 272                 mp->m_inoadd = XFS_INO64_OFFSET;
 273         }
 274 #endif
 275         if (ap->flags & XFSMNT_RETERR)
 276                 mp->m_flags |= XFS_MOUNT_RETERR;
 277         if (ap->flags & XFSMNT_NOALIGN)
 278                 mp->m_flags |= XFS_MOUNT_NOALIGN;
 279         if (ap->flags & XFSMNT_SWALLOC)
 280                 mp->m_flags |= XFS_MOUNT_SWALLOC;
 281         if (ap->flags & XFSMNT_OSYNCISOSYNC)
 282                 mp->m_flags |= XFS_MOUNT_OSYNCISOSYNC;
 283         if (ap->flags & XFSMNT_32BITINODES)
 284                 mp->m_flags |= XFS_MOUNT_32BITINODES;
 285
 286         if (ap->flags & XFSMNT_IOSIZE) {
 287                 if (ap->iosizelog > XFS_MAX_IO_LOG ||
 288                     ap->iosizelog < XFS_MIN_IO_LOG) {
 289                         cmn_err(CE_WARN,
 290                 "XFS: invalid log iosize: %d [not %d-%d]",
 291                                 ap->iosizelog, XFS_MIN_IO_LOG,
 292                                 XFS_MAX_IO_LOG);
 293                         return XFS_ERROR(EINVAL);
 294                 }
 295
 296                 mp->m_flags |= XFS_MOUNT_DFLT_IOSIZE;
 297                 mp->m_readio_log = mp->m_writeio_log = ap->iosizelog;
 298         }
 299
 300         if (ap->flags & XFSMNT_IHASHSIZE)
 301                 mp->m_flags |= XFS_MOUNT_IHASHSIZE;
 302         if (ap->flags & XFSMNT_IDELETE)
 303                 mp->m_flags |= XFS_MOUNT_IDELETE;
 304         if (ap->flags & XFSMNT_DIRSYNC)
 305                 mp->m_flags |= XFS_MOUNT_DIRSYNC;
 306         if (ap->flags & XFSMNT_ATTR2)
 307                 mp->m_flags |= XFS_MOUNT_ATTR2;
 308
 309         if (ap->flags2 & XFSMNT2_COMPAT_IOSIZE)
 310                 mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE;
 311         if (ap->flags2 & XFSMNT2_TAGXID)
 312                 mp->m_flags |= XFS_MOUNT_TAGXID;
 313
 314         /*
 315          * no recovery flag requires a read-only mount
 316          */
 317         if (ap->flags & XFSMNT_NORECOVERY) {
 318                 if (!(vfs->vfs_flag & VFS_RDONLY)) {
 319                         cmn_err(CE_WARN,
 320         "XFS: tried to mount a FS read-write without recovery!");
 321                         return XFS_ERROR(EINVAL);
 322                 }
 323                 mp->m_flags |= XFS_MOUNT_NORECOVERY;
 324         }
 325
 326         if (ap->flags & XFSMNT_NOUUID)
 327                 mp->m_flags |= XFS_MOUNT_NOUUID;
 328         if (ap->flags & XFSMNT_BARRIER)
 329                 mp->m_flags |= XFS_MOUNT_BARRIER;
 330         else
 331                 mp->m_flags &= ~XFS_MOUNT_BARRIER;
 332
 333         return 0;
 334 }
 335
 336 /*
 337  * This function fills in xfs_mount_t fields based on mount args.
 338  * Note: the superblock _has_ now been read in.
 339  */
 340 STATIC int
 341 xfs_finish_flags(
 342         struct vfs              *vfs,
 343         struct xfs_mount_args   *ap,
 344         struct xfs_mount        *mp)
 345 {
 346         int                     ronly = (vfs->vfs_flag & VFS_RDONLY);
 347
 348         /* Fail a mount where the logbuf is smaller then the log stripe */
 349         if (XFS_SB_VERSION_HASLOGV2(&mp->m_sb)) {
 350                 if ((ap->logbufsize <= 0) &&
 351                     (mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE)) {
 352                         mp->m_logbsize = mp->m_sb.sb_logsunit;
 353                 } else if (ap->logbufsize > 0 &&
 354                            ap->logbufsize < mp->m_sb.sb_logsunit) {
 355                         cmn_err(CE_WARN,
 356         "XFS: logbuf size must be greater than or equal to log stripe size");
 357                         return XFS_ERROR(EINVAL);
 358                 }
 359         } else {
 360                 /* Fail a mount if the logbuf is larger than 32K */
 361                 if (ap->logbufsize > XLOG_BIG_RECORD_BSIZE) {
 362                         cmn_err(CE_WARN,
 363         "XFS: logbuf size for version 1 logs must be 16K or 32K");
 364                         return XFS_ERROR(EINVAL);
 365                 }
 366         }
 367
 368         if (XFS_SB_VERSION_HASATTR2(&mp->m_sb)) {
 369                 mp->m_flags |= XFS_MOUNT_ATTR2;
 370         }
 371
 372         /*
 373          * prohibit r/w mounts of read-only filesystems
 374          */
 375         if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !ronly) {
 376                 cmn_err(CE_WARN,
 377         "XFS: cannot mount a read-only filesystem as read-write");
 378                 return XFS_ERROR(EROFS);
 379         }
 380
 381         /*
 382          * check for shared mount.
 383          */
 384         if (ap->flags & XFSMNT_SHARED) {
 385                 if (!XFS_SB_VERSION_HASSHARED(&mp->m_sb))
 386                         return XFS_ERROR(EINVAL);
 387
 388                 /*
 389                  * For IRIX 6.5, shared mounts must have the shared
 390                  * version bit set, have the persistent readonly
 391                  * field set, must be version 0 and can only be mounted
 392                  * read-only.
 393                  */
 394                 if (!ronly || !(mp->m_sb.sb_flags & XFS_SBF_READONLY) ||
 395                      (mp->m_sb.sb_shared_vn != 0))
 396                         return XFS_ERROR(EINVAL);
 397
 398                 mp->m_flags |= XFS_MOUNT_SHARED;
 399
 400                 /*
 401                  * Shared XFS V0 can't deal with DMI.  Return EINVAL.
 402                  */
 403                 if (mp->m_sb.sb_shared_vn == 0 && (ap->flags & XFSMNT_DMAPI))
 404                         return XFS_ERROR(EINVAL);
 405         }
 406
 407         if (ap->flags2 & XFSMNT2_TAGXID)
 408                 vfs->vfs_super->s_flags |= MS_TAGXID;
 409         return 0;
 410 }
 411
 412 /*
 413  * xfs_mount
 414  *
 415  * The file system configurations are:
 416  *      (1) device (partition) with data and internal log
 417  *      (2) logical volume with data and log subvolumes.
 418  *      (3) logical volume with data, log, and realtime subvolumes.
 419  *
 420  * We only have to handle opening the log and realtime volumes here if
 421  * they are present.  The data subvolume has already been opened by
 422  * get_sb_bdev() and is stored in vfsp->vfs_super->s_bdev.
 423  */
 424 STATIC int
 425 xfs_mount(
 426         struct bhv_desc         *bhvp,
 427         struct xfs_mount_args   *args,
 428         cred_t                  *credp)
 429 {
 430         struct vfs              *vfsp = bhvtovfs(bhvp);
 431         struct bhv_desc         *p;
 432         struct xfs_mount        *mp = XFS_BHVTOM(bhvp);
 433         struct block_device     *ddev, *logdev, *rtdev;
 434         int                     flags = 0, error;
 435
 436         ddev = vfsp->vfs_super->s_bdev;
 437         logdev = rtdev = NULL;
 438
 439         /*
 440          * Setup xfs_mount function vectors from available behaviors
 441          */
 442         p = vfs_bhv_lookup(vfsp, VFS_POSITION_DM);
 443         mp->m_dm_ops = p ? *(xfs_dmops_t *) vfs_bhv_custom(p) : xfs_dmcore_stub;
 444         p = vfs_bhv_lookup(vfsp, VFS_POSITION_QM);
 445         mp->m_qm_ops = p ? *(xfs_qmops_t *) vfs_bhv_custom(p) : xfs_qmcore_stub;
 446         p = vfs_bhv_lookup(vfsp, VFS_POSITION_IO);
 447         mp->m_io_ops = p ? *(xfs_ioops_t *) vfs_bhv_custom(p) : xfs_iocore_xfs;
 448
 449         if (args->flags & XFSMNT_QUIET)
 450                 flags |= XFS_MFSI_QUIET;
 451
 452         /*
 453          * Open real time and log devices - order is important.
 454          */
 455         if (args->logname[0]) {
 456                 error = xfs_blkdev_get(mp, args->logname, &logdev);
 457                 if (error)
 458                         return error;
 459         }
 460         if (args->rtname[0]) {
 461                 error = xfs_blkdev_get(mp, args->rtname, &rtdev);
 462                 if (error) {
 463                         xfs_blkdev_put(logdev);
 464                         return error;
 465                 }
 466
 467                 if (rtdev == ddev || rtdev == logdev) {
 468                         cmn_err(CE_WARN,
 469         "XFS: Cannot mount filesystem with identical rtdev and ddev/logdev.");
 470                         xfs_blkdev_put(logdev);
 471                         xfs_blkdev_put(rtdev);
 472                         return EINVAL;
 473                 }
 474         }
 475
 476         /*
 477          * Setup xfs_mount buffer target pointers
 478          */
 479         error = ENOMEM;
 480         mp->m_ddev_targp = xfs_alloc_buftarg(ddev, 0);
 481         if (!mp->m_ddev_targp) {
 482                 xfs_blkdev_put(logdev);
 483                 xfs_blkdev_put(rtdev);
 484                 return error;
 485         }
 486         if (rtdev) {
 487                 mp->m_rtdev_targp = xfs_alloc_buftarg(rtdev, 1);
 488                 if (!mp->m_rtdev_targp)
 489                         goto error0;
 490         }
 491         mp->m_logdev_targp = (logdev && logdev != ddev) ?
 492                                 xfs_alloc_buftarg(logdev, 1) : mp->m_ddev_targp;
 493         if (!mp->m_logdev_targp)
 494                 goto error0;
 495
 496         /*
 497          * Setup flags based on mount(2) options and then the superblock
 498          */
 499         error = xfs_start_flags(vfsp, args, mp);
 500         if (error)
 501                 goto error1;
 502         error = xfs_readsb(mp, flags);
 503         if (error)
 504                 goto error1;
 505         error = xfs_finish_flags(vfsp, args, mp);
 506         if (error)
 507                 goto error2;
 508
 509         /*
 510          * Setup xfs_mount buffer target pointers based on superblock
 511          */
 512         error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_blocksize,
 513                                     mp->m_sb.sb_sectsize);
 514         if (!error && logdev && logdev != ddev) {
 515                 unsigned int    log_sector_size = BBSIZE;
 516
 517                 if (XFS_SB_VERSION_HASSECTOR(&mp->m_sb))
 518                         log_sector_size = mp->m_sb.sb_logsectsize;
 519                 error = xfs_setsize_buftarg(mp->m_logdev_targp,
 520                                             mp->m_sb.sb_blocksize,
 521                                             log_sector_size);
 522         }
 523         if (!error && rtdev)
 524                 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
 525                                             mp->m_sb.sb_blocksize,
 526                                             mp->m_sb.sb_sectsize);
 527         if (error)
 528                 goto error2;
 529
 530         if ((mp->m_flags & XFS_MOUNT_BARRIER) && !(vfsp->vfs_flag & VFS_RDONLY))
 531                 xfs_mountfs_check_barriers(mp);
 532
 533         error = XFS_IOINIT(vfsp, args, flags);
 534         if (error)
 535                 goto error2;
 536
 537         return 0;
 538
 539 error2:
 540         if (mp->m_sb_bp)
 541                 xfs_freesb(mp);
 542 error1:
 543         xfs_binval(mp->m_ddev_targp);
 544         if (logdev && logdev != ddev)
 545                 xfs_binval(mp->m_logdev_targp);
 546         if (rtdev)
 547                 xfs_binval(mp->m_rtdev_targp);
 548 error0:
 549         xfs_unmountfs_close(mp, credp);
 550         return error;
 551 }
 552
 553 STATIC int
 554 xfs_unmount(
 555         bhv_desc_t      *bdp,
 556         int             flags,
 557         cred_t          *credp)
 558 {
 559         struct vfs      *vfsp = bhvtovfs(bdp);
 560         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
 561         xfs_inode_t     *rip;
 562         vnode_t         *rvp;
 563         int             unmount_event_wanted = 0;
 564         int             unmount_event_flags = 0;
 565         int             xfs_unmountfs_needed = 0;
 566         int             error;
 567
 568         rip = mp->m_rootip;
 569         rvp = XFS_ITOV(rip);
 570
 571         if (vfsp->vfs_flag & VFS_DMI) {
 572                 error = XFS_SEND_PREUNMOUNT(mp, vfsp,
 573                                 rvp, DM_RIGHT_NULL, rvp, DM_RIGHT_NULL,
 574                                 NULL, NULL, 0, 0,
 575                                 (mp->m_dmevmask & (1<<DM_EVENT_PREUNMOUNT))?
 576                                         0:DM_FLAGS_UNWANTED);
 577                         if (error)
 578                                 return XFS_ERROR(error);
 579                 unmount_event_wanted = 1;
 580                 unmount_event_flags = (mp->m_dmevmask & (1<<DM_EVENT_UNMOUNT))?
 581                                         0 : DM_FLAGS_UNWANTED;
 582         }
 583
 584         /*
 585          * First blow any referenced inode from this file system
 586          * out of the reference cache, and delete the timer.
 587          */
 588         xfs_refcache_purge_mp(mp);
 589
 590         XFS_bflush(mp->m_ddev_targp);
 591         error = xfs_unmount_flush(mp, 0);
 592         if (error)
 593                 goto out;
 594
 595         ASSERT(vn_count(rvp) == 1);
 596
 597         /*
 598          * Drop the reference count
 599          */
 600         VN_RELE(rvp);
 601
 602         /*
 603          * If we're forcing a shutdown, typically because of a media error,
 604          * we want to make sure we invalidate dirty pages that belong to
 605          * referenced vnodes as well.
 606          */
 607         if (XFS_FORCED_SHUTDOWN(mp)) {
 608                 error = xfs_sync(&mp->m_bhv,
 609                          (SYNC_WAIT | SYNC_CLOSE), credp);
 610                 ASSERT(error != EFSCORRUPTED);
 611         }
 612         xfs_unmountfs_needed = 1;
 613
 614 out:
 615         /*      Send DMAPI event, if required.
 616          *      Then do xfs_unmountfs() if needed.
 617          *      Then return error (or zero).
 618          */
 619         if (unmount_event_wanted) {
 620                 /* Note: mp structure must still exist for
 621                  * XFS_SEND_UNMOUNT() call.
 622                  */
 623                 XFS_SEND_UNMOUNT(mp, vfsp, error == 0 ? rvp : NULL,
 624                         DM_RIGHT_NULL, 0, error, unmount_event_flags);
 625         }
 626         if (xfs_unmountfs_needed) {
 627                 /*
 628                  * Call common unmount function to flush to disk
 629                  * and free the super block buffer & mount structures.
 630                  */
 631                 xfs_unmountfs(mp, credp);
 632         }
 633
 634         return XFS_ERROR(error);
 635 }
 636
 637 STATIC int
 638 xfs_quiesce_fs(
 639         xfs_mount_t             *mp)
 640 {
 641         int                     count = 0, pincount;
 642
 643         xfs_refcache_purge_mp(mp);
 644         xfs_flush_buftarg(mp->m_ddev_targp, 0);
 645         xfs_finish_reclaim_all(mp, 0);
 646
 647         /* This loop must run at least twice.
 648          * The first instance of the loop will flush
 649          * most meta data but that will generate more
 650          * meta data (typically directory updates).
 651          * Which then must be flushed and logged before
 652          * we can write the unmount record.
 653          */
 654         do {
 655                 xfs_syncsub(mp, SYNC_REMOUNT|SYNC_ATTR|SYNC_WAIT, 0, NULL);
 656                 pincount = xfs_flush_buftarg(mp->m_ddev_targp, 1);
 657                 if (!pincount) {
 658                         delay(50);
 659                         count++;
 660                 }
 661         } while (count < 2);
 662
 663         return 0;
 664 }
 665
 666 STATIC int
 667 xfs_mntupdate(
 668         bhv_desc_t                      *bdp,
 669         int                             *flags,
 670         struct xfs_mount_args           *args)
 671 {
 672         struct vfs      *vfsp = bhvtovfs(bdp);
 673         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
 674         int             error;
 675
 676         if (!(*flags & MS_RDONLY)) {                    /* rw/ro -> rw */
 677                 if (vfsp->vfs_flag & VFS_RDONLY)
 678                         vfsp->vfs_flag &= ~VFS_RDONLY;
 679                 if (args->flags & XFSMNT_BARRIER) {
 680                         mp->m_flags |= XFS_MOUNT_BARRIER;
 681                         xfs_mountfs_check_barriers(mp);
 682                 } else {
 683                         mp->m_flags &= ~XFS_MOUNT_BARRIER;
 684                 }
 685         } else if (!(vfsp->vfs_flag & VFS_RDONLY)) {    /* rw -> ro */
 686                 VFS_SYNC(vfsp, SYNC_FSDATA|SYNC_BDFLUSH|SYNC_ATTR, NULL, error);
 687                 xfs_quiesce_fs(mp);
 688                 xfs_log_unmount_write(mp);
 689                 xfs_unmountfs_writesb(mp);
 690                 vfsp->vfs_flag |= VFS_RDONLY;
 691         }
 692         return 0;
 693 }
 694
 695 /*
 696  * xfs_unmount_flush implements a set of flush operation on special
 697  * inodes, which are needed as a separate set of operations so that
 698  * they can be called as part of relocation process.
 699  */
 700 int
 701 xfs_unmount_flush(
 702         xfs_mount_t     *mp,            /* Mount structure we are getting
 703                                            rid of. */
 704         int             relocation)     /* Called from vfs relocation. */
 705 {
 706         xfs_inode_t     *rip = mp->m_rootip;
 707         xfs_inode_t     *rbmip;
 708         xfs_inode_t     *rsumip = NULL;
 709         vnode_t         *rvp = XFS_ITOV(rip);
 710         int             error;
 711
 712         xfs_ilock(rip, XFS_ILOCK_EXCL);
 713         xfs_iflock(rip);
 714
 715         /*
 716          * Flush out the real time inodes.
 717          */
 718         if ((rbmip = mp->m_rbmip) != NULL) {
 719                 xfs_ilock(rbmip, XFS_ILOCK_EXCL);
 720                 xfs_iflock(rbmip);
 721                 error = xfs_iflush(rbmip, XFS_IFLUSH_SYNC);
 722                 xfs_iunlock(rbmip, XFS_ILOCK_EXCL);
 723
 724                 if (error == EFSCORRUPTED)
 725                         goto fscorrupt_out;
 726
 727                 ASSERT(vn_count(XFS_ITOV(rbmip)) == 1);
 728
 729                 rsumip = mp->m_rsumip;
 730                 xfs_ilock(rsumip, XFS_ILOCK_EXCL);
 731                 xfs_iflock(rsumip);
 732                 error = xfs_iflush(rsumip, XFS_IFLUSH_SYNC);
 733                 xfs_iunlock(rsumip, XFS_ILOCK_EXCL);
 734
 735                 if (error == EFSCORRUPTED)
 736                         goto fscorrupt_out;
 737
 738                 ASSERT(vn_count(XFS_ITOV(rsumip)) == 1);
 739         }
 740
 741         /*
 742          * Synchronously flush root inode to disk
 743          */
 744         error = xfs_iflush(rip, XFS_IFLUSH_SYNC);
 745         if (error == EFSCORRUPTED)
 746                 goto fscorrupt_out2;
 747
 748         if (vn_count(rvp) != 1 && !relocation) {
 749                 xfs_iunlock(rip, XFS_ILOCK_EXCL);
 750                 return XFS_ERROR(EBUSY);
 751         }
 752
 753         /*
 754          * Release dquot that rootinode, rbmino and rsumino might be holding,
 755          * flush and purge the quota inodes.
 756          */
 757         error = XFS_QM_UNMOUNT(mp);
 758         if (error == EFSCORRUPTED)
 759                 goto fscorrupt_out2;
 760
 761         if (rbmip) {
 762                 VN_RELE(XFS_ITOV(rbmip));
 763                 VN_RELE(XFS_ITOV(rsumip));
 764         }
 765
 766         xfs_iunlock(rip, XFS_ILOCK_EXCL);
 767         return 0;
 768
 769 fscorrupt_out:
 770         xfs_ifunlock(rip);
 771
 772 fscorrupt_out2:
 773         xfs_iunlock(rip, XFS_ILOCK_EXCL);
 774
 775         return XFS_ERROR(EFSCORRUPTED);
 776 }
 777
 778 /*
 779  * xfs_root extracts the root vnode from a vfs.
 780  *
 781  * vfsp -- the vfs struct for the desired file system
 782  * vpp  -- address of the caller's vnode pointer which should be
 783  *         set to the desired fs root vnode
 784  */
 785 STATIC int
 786 xfs_root(
 787         bhv_desc_t      *bdp,
 788         vnode_t         **vpp)
 789 {
 790         vnode_t         *vp;
 791
 792         vp = XFS_ITOV((XFS_BHVTOM(bdp))->m_rootip);
 793         VN_HOLD(vp);
 794         *vpp = vp;
 795         return 0;
 796 }
 797
 798 /*
 799  * xfs_statvfs
 800  *
 801  * Fill in the statvfs structure for the given file system.  We use
 802  * the superblock lock in the mount structure to ensure a consistent
 803  * snapshot of the counters returned.
 804  */
 805 STATIC int
 806 xfs_statvfs(
 807         bhv_desc_t      *bdp,
 808         xfs_statfs_t    *statp,
 809         vnode_t         *vp)
 810 {
 811         __uint64_t      fakeinos;
 812         xfs_extlen_t    lsize;
 813         xfs_mount_t     *mp;
 814         xfs_sb_t        *sbp;
 815         unsigned long   s;
 816
 817         mp = XFS_BHVTOM(bdp);
 818         sbp = &(mp->m_sb);
 819
 820         statp->f_type = XFS_SB_MAGIC;
 821
 822         xfs_icsb_sync_counters_lazy(mp);
 823         s = XFS_SB_LOCK(mp);
 824         statp->f_bsize = sbp->sb_blocksize;
 825         lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
 826         statp->f_blocks = sbp->sb_dblocks - lsize;
 827         statp->f_bfree = statp->f_bavail = sbp->sb_fdblocks;
 828         fakeinos = statp->f_bfree << sbp->sb_inopblog;
 829 #if XFS_BIG_INUMS
 830         fakeinos += mp->m_inoadd;
 831 #endif
 832         statp->f_files =
 833             MIN(sbp->sb_icount + fakeinos, (__uint64_t)XFS_MAXINUMBER);
 834         if (mp->m_maxicount)
 835 #if XFS_BIG_INUMS
 836                 if (!mp->m_inoadd)
 837 #endif
 838                         statp->f_files = min_t(typeof(statp->f_files),
 839                                                 statp->f_files,
 840                                                 mp->m_maxicount);
 841         statp->f_ffree = statp->f_files - (sbp->sb_icount - sbp->sb_ifree);
 842         XFS_SB_UNLOCK(mp, s);
 843
 844         xfs_statvfs_fsid(statp, mp);
 845         statp->f_namelen = MAXNAMELEN - 1;
 846
 847         return 0;
 848 }
 849
 850
 851 /*
 852  * xfs_sync flushes any pending I/O to file system vfsp.
 853  *
 854  * This routine is called by vfs_sync() to make sure that things make it
 855  * out to disk eventually, on sync() system calls to flush out everything,
 856  * and when the file system is unmounted.  For the vfs_sync() case, all
 857  * we really need to do is sync out the log to make all of our meta-data
 858  * updates permanent (except for timestamps).  For calls from pflushd(),
 859  * dirty pages are kept moving by calling pdflush() on the inodes
 860  * containing them.  We also flush the inodes that we can lock without
 861  * sleeping and the superblock if we can lock it without sleeping from
 862  * vfs_sync() so that items at the tail of the log are always moving out.
 863  *
 864  * Flags:
 865  *      SYNC_BDFLUSH - We're being called from vfs_sync() so we don't want
 866  *                     to sleep if we can help it.  All we really need
 867  *                     to do is ensure that the log is synced at least
 868  *                     periodically.  We also push the inodes and
 869  *                     superblock if we can lock them without sleeping
 870  *                      and they are not pinned.
 871  *      SYNC_ATTR    - We need to flush the inodes.  If SYNC_BDFLUSH is not
 872  *                     set, then we really want to lock each inode and flush
 873  *                     it.
 874  *      SYNC_WAIT    - All the flushes that take place in this call should
 875  *                     be synchronous.
 876  *      SYNC_DELWRI  - This tells us to push dirty pages associated with
 877  *                     inodes.  SYNC_WAIT and SYNC_BDFLUSH are used to
 878  *                     determine if they should be flushed sync, async, or
 879  *                     delwri.
 880  *      SYNC_CLOSE   - This flag is passed when the system is being
 881  *                     unmounted.  We should sync and invalidate everything.
 882  *      SYNC_FSDATA  - This indicates that the caller would like to make
 883  *                     sure the superblock is safe on disk.  We can ensure
 884  *                     this by simply making sure the log gets flushed
 885  *                     if SYNC_BDFLUSH is set, and by actually writing it
 886  *                     out otherwise.
 887  *
 888  */
 889 /*ARGSUSED*/
 890 STATIC int
 891 xfs_sync(
 892         bhv_desc_t      *bdp,
 893         int             flags,
 894         cred_t          *credp)
 895 {
 896         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
 897
 898         if (unlikely(flags == SYNC_QUIESCE))
 899                 return xfs_quiesce_fs(mp);
 900         else
 901                 return xfs_syncsub(mp, flags, 0, NULL);
 902 }
 903
 904 /*
 905  * xfs sync routine for internal use
 906  *
 907  * This routine supports all of the flags defined for the generic VFS_SYNC
 908  * interface as explained above under xfs_sync.  In the interests of not
 909  * changing interfaces within the 6.5 family, additional internally-
 910  * required functions are specified within a separate xflags parameter,
 911  * only available by calling this routine.
 912  *
 913  */
 914 int
 915 xfs_sync_inodes(
 916         xfs_mount_t     *mp,
 917         int             flags,
 918         int             xflags,
 919         int             *bypassed)
 920 {
 921         xfs_inode_t     *ip = NULL;
 922         xfs_inode_t     *ip_next;
 923         xfs_buf_t       *bp;
 924         vnode_t         *vp = NULL;
 925         int             error;
 926         int             last_error;
 927         uint64_t        fflag;
 928         uint            lock_flags;
 929         uint            base_lock_flags;
 930         boolean_t       mount_locked;
 931         boolean_t       vnode_refed;
 932         int             preempt;
 933         xfs_dinode_t    *dip;
 934         xfs_iptr_t      *ipointer;
 935 #ifdef DEBUG
 936         boolean_t       ipointer_in = B_FALSE;
 937
 938 #define IPOINTER_SET    ipointer_in = B_TRUE
 939 #define IPOINTER_CLR    ipointer_in = B_FALSE
 940 #else
 941 #define IPOINTER_SET
 942 #define IPOINTER_CLR
 943 #endif
 944
 945
 946 /* Insert a marker record into the inode list after inode ip. The list
 947  * must be locked when this is called. After the call the list will no
 948  * longer be locked.
 949  */
 950 #define IPOINTER_INSERT(ip, mp) { \
 951                 ASSERT(ipointer_in == B_FALSE); \
 952                 ipointer->ip_mnext = ip->i_mnext; \
 953                 ipointer->ip_mprev = ip; \
 954                 ip->i_mnext = (xfs_inode_t *)ipointer; \
 955                 ipointer->ip_mnext->i_mprev = (xfs_inode_t *)ipointer; \
 956                 preempt = 0; \
 957                 XFS_MOUNT_IUNLOCK(mp); \
 958                 mount_locked = B_FALSE; \
 959                 IPOINTER_SET; \
 960         }
 961
 962 /* Remove the marker from the inode list. If the marker was the only item
 963  * in the list then there are no remaining inodes and we should zero out
 964  * the whole list. If we are the current head of the list then move the head
 965  * past us.
 966  */
 967 #define IPOINTER_REMOVE(ip, mp) { \
 968                 ASSERT(ipointer_in == B_TRUE); \
 969                 if (ipointer->ip_mnext != (xfs_inode_t *)ipointer) { \
 970                         ip = ipointer->ip_mnext; \
 971                         ip->i_mprev = ipointer->ip_mprev; \
 972                         ipointer->ip_mprev->i_mnext = ip; \
 973                         if (mp->m_inodes == (xfs_inode_t *)ipointer) { \
 974                                 mp->m_inodes = ip; \
 975                         } \
 976                 } else { \
 977                         ASSERT(mp->m_inodes == (xfs_inode_t *)ipointer); \
 978                         mp->m_inodes = NULL; \
 979                         ip = NULL; \
 980                 } \
 981                 IPOINTER_CLR; \
 982         }
 983
 984 #define XFS_PREEMPT_MASK        0x7f
 985
 986         if (bypassed)
 987                 *bypassed = 0;
 988         if (XFS_MTOVFS(mp)->vfs_flag & VFS_RDONLY)
 989                 return 0;
 990         error = 0;
 991         last_error = 0;
 992         preempt = 0;
 993
 994         /* Allocate a reference marker */
 995         ipointer = (xfs_iptr_t *)kmem_zalloc(sizeof(xfs_iptr_t), KM_SLEEP);
 996
 997         fflag = XFS_B_ASYNC;            /* default is don't wait */
 998         if (flags & (SYNC_BDFLUSH | SYNC_DELWRI))
 999                 fflag = XFS_B_DELWRI;
1000         if (flags & SYNC_WAIT)
1001                 fflag = 0;              /* synchronous overrides all */
1002
1003         base_lock_flags = XFS_ILOCK_SHARED;
1004         if (flags & (SYNC_DELWRI | SYNC_CLOSE)) {
1005                 /*
1006                  * We need the I/O lock if we're going to call any of
1007                  * the flush/inval routines.
1008                  */
1009                 base_lock_flags |= XFS_IOLOCK_SHARED;
1010         }
1011
1012         XFS_MOUNT_ILOCK(mp);
1013
1014         ip = mp->m_inodes;
1015
1016         mount_locked = B_TRUE;
1017         vnode_refed  = B_FALSE;
1018
1019         IPOINTER_CLR;
1020
1021         do {
1022                 ASSERT(ipointer_in == B_FALSE);
1023                 ASSERT(vnode_refed == B_FALSE);
1024
1025                 lock_flags = base_lock_flags;
1026
1027                 /*
1028                  * There were no inodes in the list, just break out
1029                  * of the loop.
1030                  */
1031                 if (ip == NULL) {
1032                         break;
1033                 }
1034
1035                 /*
1036                  * We found another sync thread marker - skip it
1037                  */
1038                 if (ip->i_mount == NULL) {
1039                         ip = ip->i_mnext;
1040                         continue;
1041                 }
1042
1043                 vp = XFS_ITOV_NULL(ip);
1044
1045                 /*
1046                  * If the vnode is gone then this is being torn down,
1047                  * call reclaim if it is flushed, else let regular flush
1048                  * code deal with it later in the loop.
1049                  */
1050
1051                 if (vp == NULL) {
1052                         /* Skip ones already in reclaim */
1053                         if (ip->i_flags & XFS_IRECLAIM) {
1054                                 ip = ip->i_mnext;
1055                                 continue;
1056                         }
1057                         if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0) {
1058                                 ip = ip->i_mnext;
1059                         } else if ((xfs_ipincount(ip) == 0) &&
1060                                     xfs_iflock_nowait(ip)) {
1061                                 IPOINTER_INSERT(ip, mp);
1062
1063                                 xfs_finish_reclaim(ip, 1,
1064                                                 XFS_IFLUSH_DELWRI_ELSE_ASYNC);
1065
1066                                 XFS_MOUNT_ILOCK(mp);
1067                                 mount_locked = B_TRUE;
1068                                 IPOINTER_REMOVE(ip, mp);
1069                         } else {
1070                                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1071                                 ip = ip->i_mnext;
1072                         }
1073                         continue;
1074                 }
1075
1076                 if (VN_BAD(vp)) {
1077                         ip = ip->i_mnext;
1078                         continue;
1079                 }
1080
1081                 if (XFS_FORCED_SHUTDOWN(mp) && !(flags & SYNC_CLOSE)) {
1082                         XFS_MOUNT_IUNLOCK(mp);
1083                         kmem_free(ipointer, sizeof(xfs_iptr_t));
1084                         return 0;
1085                 }
1086
1087                 /*
1088                  * If this is just vfs_sync() or pflushd() calling
1089                  * then we can skip inodes for which it looks like
1090                  * there is nothing to do.  Since we don't have the
1091                  * inode locked this is racy, but these are periodic
1092                  * calls so it doesn't matter.  For the others we want
1093                  * to know for sure, so we at least try to lock them.
1094                  */
1095                 if (flags & SYNC_BDFLUSH) {
1096                         if (((ip->i_itemp == NULL) ||
1097                              !(ip->i_itemp->ili_format.ilf_fields &
1098                                XFS_ILOG_ALL)) &&
1099                             (ip->i_update_core == 0)) {
1100                                 ip = ip->i_mnext;
1101                                 continue;
1102                         }
1103                 }
1104
1105                 /*
1106                  * Try to lock without sleeping.  We're out of order with
1107                  * the inode list lock here, so if we fail we need to drop
1108                  * the mount lock and try again.  If we're called from
1109                  * bdflush() here, then don't bother.
1110                  *
1111                  * The inode lock here actually coordinates with the
1112                  * almost spurious inode lock in xfs_ireclaim() to prevent
1113                  * the vnode we handle here without a reference from
1114                  * being freed while we reference it.  If we lock the inode
1115                  * while it's on the mount list here, then the spurious inode
1116                  * lock in xfs_ireclaim() after the inode is pulled from
1117                  * the mount list will sleep until we release it here.
1118                  * This keeps the vnode from being freed while we reference
1119                  * it.
1120                  */
1121                 if (xfs_ilock_nowait(ip, lock_flags) == 0) {
1122                         if ((flags & SYNC_BDFLUSH) || (vp == NULL)) {
1123                                 ip = ip->i_mnext;
1124                                 continue;
1125                         }
1126
1127                         vp = vn_grab(vp);
1128                         if (vp == NULL) {
1129                                 ip = ip->i_mnext;
1130                                 continue;
1131                         }
1132
1133                         IPOINTER_INSERT(ip, mp);
1134                         xfs_ilock(ip, lock_flags);
1135
1136                         ASSERT(vp == XFS_ITOV(ip));
1137                         ASSERT(ip->i_mount == mp);
1138
1139                         vnode_refed = B_TRUE;
1140                 }
1141
1142                 /* From here on in the loop we may have a marker record
1143                  * in the inode list.
1144                  */
1145
1146                 if ((flags & SYNC_CLOSE)  && (vp != NULL)) {
1147                         /*
1148                          * This is the shutdown case.  We just need to
1149                          * flush and invalidate all the pages associated
1150                          * with the inode.  Drop the inode lock since
1151                          * we can't hold it across calls to the buffer
1152                          * cache.
1153                          *
1154                          * We don't set the VREMAPPING bit in the vnode
1155                          * here, because we don't hold the vnode lock
1156                          * exclusively.  It doesn't really matter, though,
1157                          * because we only come here when we're shutting
1158                          * down anyway.
1159                          */
1160                         xfs_iunlock(ip, XFS_ILOCK_SHARED);
1161
1162                         if (XFS_FORCED_SHUTDOWN(mp)) {
1163                                 VOP_TOSS_PAGES(vp, 0, -1, FI_REMAPF);
1164                         } else {
1165                                 VOP_FLUSHINVAL_PAGES(vp, 0, -1, FI_REMAPF);
1166                         }
1167
1168                         xfs_ilock(ip, XFS_ILOCK_SHARED);
1169
1170                 } else if ((flags & SYNC_DELWRI) && (vp != NULL)) {
1171                         if (VN_DIRTY(vp)) {
1172                                 /* We need to have dropped the lock here,
1173                                  * so insert a marker if we have not already
1174                                  * done so.
1175                                  */
1176                                 if (mount_locked) {
1177                                         IPOINTER_INSERT(ip, mp);
1178                                 }
1179
1180                                 /*
1181                                  * Drop the inode lock since we can't hold it
1182                                  * across calls to the buffer cache.
1183                                  */
1184                                 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1185                                 VOP_FLUSH_PAGES(vp, (xfs_off_t)0, -1,
1186                                                         fflag, FI_NONE, error);
1187                                 xfs_ilock(ip, XFS_ILOCK_SHARED);
1188                         }
1189
1190                 }
1191
1192                 if (flags & SYNC_BDFLUSH) {
1193                         if ((flags & SYNC_ATTR) &&
1194                             ((ip->i_update_core) ||
1195                              ((ip->i_itemp != NULL) &&
1196                               (ip->i_itemp->ili_format.ilf_fields != 0)))) {
1197
1198                                 /* Insert marker and drop lock if not already
1199                                  * done.
1200                                  */
1201                                 if (mount_locked) {
1202                                         IPOINTER_INSERT(ip, mp);
1203                                 }
1204
1205                                 /*
1206                                  * We don't want the periodic flushing of the
1207                                  * inodes by vfs_sync() to interfere with
1208                                  * I/O to the file, especially read I/O
1209                                  * where it is only the access time stamp
1210                                  * that is being flushed out.  To prevent
1211                                  * long periods where we have both inode
1212                                  * locks held shared here while reading the
1213                                  * inode's buffer in from disk, we drop the
1214                                  * inode lock while reading in the inode
1215                                  * buffer.  We have to release the buffer
1216                                  * and reacquire the inode lock so that they
1217                                  * are acquired in the proper order (inode
1218                                  * locks first).  The buffer will go at the
1219                                  * end of the lru chain, though, so we can
1220                                  * expect it to still be there when we go
1221                                  * for it again in xfs_iflush().
1222                                  */
1223                                 if ((xfs_ipincount(ip) == 0) &&
1224                                     xfs_iflock_nowait(ip)) {
1225
1226                                         xfs_ifunlock(ip);
1227                                         xfs_iunlock(ip, XFS_ILOCK_SHARED);
1228
1229                                         error = xfs_itobp(mp, NULL, ip,
1230                                                           &dip, &bp, 0, 0);
1231                                         if (!error) {
1232                                                 xfs_buf_relse(bp);
1233                                         } else {
1234                                                 /* Bailing out, remove the
1235                                                  * marker and free it.
1236                                                  */
1237                                                 XFS_MOUNT_ILOCK(mp);
1238
1239                                                 IPOINTER_REMOVE(ip, mp);
1240
1241                                                 XFS_MOUNT_IUNLOCK(mp);
1242
1243                                                 ASSERT(!(lock_flags &
1244                                                         XFS_IOLOCK_SHARED));
1245
1246                                                 kmem_free(ipointer,
1247                                                         sizeof(xfs_iptr_t));
1248                                                 return (0);
1249                                         }
1250
1251                                         /*
1252                                          * Since we dropped the inode lock,
1253                                          * the inode may have been reclaimed.
1254                                          * Therefore, we reacquire the mount
1255                                          * lock and check to see if we were the
1256                                          * inode reclaimed. If this happened
1257                                          * then the ipointer marker will no
1258                                          * longer point back at us. In this
1259                                          * case, move ip along to the inode
1260                                          * after the marker, remove the marker
1261                                          * and continue.
1262                                          */
1263                                         XFS_MOUNT_ILOCK(mp);
1264                                         mount_locked = B_TRUE;
1265
1266                                         if (ip != ipointer->ip_mprev) {
1267                                                 IPOINTER_REMOVE(ip, mp);
1268
1269                                                 ASSERT(!vnode_refed);
1270                                                 ASSERT(!(lock_flags &
1271                                                         XFS_IOLOCK_SHARED));
1272                                                 continue;
1273                                         }
1274
1275                                         ASSERT(ip->i_mount == mp);
1276
1277                                         if (xfs_ilock_nowait(ip,
1278                                                     XFS_ILOCK_SHARED) == 0) {
1279                                                 ASSERT(ip->i_mount == mp);
1280                                                 /*
1281                                                  * We failed to reacquire
1282                                                  * the inode lock without
1283                                                  * sleeping, so just skip
1284                                                  * the inode for now.  We
1285                                                  * clear the ILOCK bit from
1286                                                  * the lock_flags so that we
1287                                                  * won't try to drop a lock
1288                                                  * we don't hold below.
1289                                                  */
1290                                                 lock_flags &= ~XFS_ILOCK_SHARED;
1291                                                 IPOINTER_REMOVE(ip_next, mp);
1292                                         } else if ((xfs_ipincount(ip) == 0) &&
1293                                                    xfs_iflock_nowait(ip)) {
1294                                                 ASSERT(ip->i_mount == mp);
1295                                                 /*
1296                                                  * Since this is vfs_sync()
1297                                                  * calling we only flush the
1298                                                  * inode out if we can lock
1299                                                  * it without sleeping and
1300                                                  * it is not pinned.  Drop
1301                                                  * the mount lock here so
1302                                                  * that we don't hold it for
1303                                                  * too long. We already have
1304                                                  * a marker in the list here.
1305                                                  */
1306                                                 XFS_MOUNT_IUNLOCK(mp);
1307                                                 mount_locked = B_FALSE;
1308                                                 error = xfs_iflush(ip,
1309                                                            XFS_IFLUSH_DELWRI);
1310                                         } else {
1311                                                 ASSERT(ip->i_mount == mp);
1312                                                 IPOINTER_REMOVE(ip_next, mp);
1313                                         }
1314                                 }
1315
1316                         }
1317
1318                 } else {
1319                         if ((flags & SYNC_ATTR) &&
1320                             ((ip->i_update_core) ||
1321                              ((ip->i_itemp != NULL) &&
1322                               (ip->i_itemp->ili_format.ilf_fields != 0)))) {
1323                                 if (mount_locked) {
1324                                         IPOINTER_INSERT(ip, mp);
1325                                 }
1326
1327                                 if (flags & SYNC_WAIT) {
1328                                         xfs_iflock(ip);
1329                                         error = xfs_iflush(ip,
1330                                                            XFS_IFLUSH_SYNC);
1331                                 } else {
1332                                         /*
1333                                          * If we can't acquire the flush
1334                                          * lock, then the inode is already
1335                                          * being flushed so don't bother
1336                                          * waiting.  If we can lock it then
1337                                          * do a delwri flush so we can
1338                                          * combine multiple inode flushes
1339                                          * in each disk write.
1340                                          */
1341                                         if (xfs_iflock_nowait(ip)) {
1342                                                 error = xfs_iflush(ip,
1343                                                            XFS_IFLUSH_DELWRI);
1344                                         }
1345                                         else if (bypassed)
1346                                                 (*bypassed)++;
1347                                 }
1348                         }
1349                 }
1350
1351                 if (lock_flags != 0) {
1352                         xfs_iunlock(ip, lock_flags);
1353                 }
1354
1355                 if (vnode_refed) {
1356                         /*
1357                          * If we had to take a reference on the vnode
1358                          * above, then wait until after we've unlocked
1359                          * the inode to release the reference.  This is
1360                          * because we can be already holding the inode
1361                          * lock when VN_RELE() calls xfs_inactive().
1362                          *
1363                          * Make sure to drop the mount lock before calling
1364                          * VN_RELE() so that we don't trip over ourselves if
1365                          * we have to go for the mount lock again in the
1366                          * inactive code.
1367                          */
1368                         if (mount_locked) {
1369                                 IPOINTER_INSERT(ip, mp);
1370                         }
1371
1372                         VN_RELE(vp);
1373
1374                         vnode_refed = B_FALSE;
1375                 }
1376
1377                 if (error) {
1378                         last_error = error;
1379                 }
1380
1381                 /*
1382                  * bail out if the filesystem is corrupted.
1383                  */
1384                 if (error == EFSCORRUPTED)  {
1385                         if (!mount_locked) {
1386                                 XFS_MOUNT_ILOCK(mp);
1387                                 IPOINTER_REMOVE(ip, mp);
1388                         }
1389                         XFS_MOUNT_IUNLOCK(mp);
1390                         ASSERT(ipointer_in == B_FALSE);
1391                         kmem_free(ipointer, sizeof(xfs_iptr_t));
1392                         return XFS_ERROR(error);
1393                 }
1394
1395                 /* Let other threads have a chance at the mount lock
1396                  * if we have looped many times without dropping the
1397                  * lock.
1398                  */
1399                 if ((++preempt & XFS_PREEMPT_MASK) == 0) {
1400                         if (mount_locked) {
1401                                 IPOINTER_INSERT(ip, mp);
1402                         }
1403                 }
1404
1405                 if (mount_locked == B_FALSE) {
1406                         XFS_MOUNT_ILOCK(mp);
1407                         mount_locked = B_TRUE;
1408                         IPOINTER_REMOVE(ip, mp);
1409                         continue;
1410                 }
1411
1412                 ASSERT(ipointer_in == B_FALSE);
1413                 ip = ip->i_mnext;
1414
1415         } while (ip != mp->m_inodes);
1416
1417         XFS_MOUNT_IUNLOCK(mp);
1418
1419         ASSERT(ipointer_in == B_FALSE);
1420
1421         kmem_free(ipointer, sizeof(xfs_iptr_t));
1422         return XFS_ERROR(last_error);
1423 }
1424
1425 /*
1426  * xfs sync routine for internal use
1427  *
1428  * This routine supports all of the flags defined for the generic VFS_SYNC
1429  * interface as explained above under xfs_sync.  In the interests of not
1430  * changing interfaces within the 6.5 family, additional internally-
1431  * required functions are specified within a separate xflags parameter,
1432  * only available by calling this routine.
1433  *
1434  */
1435 int
1436 xfs_syncsub(
1437         xfs_mount_t     *mp,
1438         int             flags,
1439         int             xflags,
1440         int             *bypassed)
1441 {
1442         int             error = 0;
1443         int             last_error = 0;
1444         uint            log_flags = XFS_LOG_FORCE;
1445         xfs_buf_t       *bp;
1446         xfs_buf_log_item_t      *bip;
1447
1448         /*
1449          * Sync out the log.  This ensures that the log is periodically
1450          * flushed even if there is not enough activity to fill it up.
1451          */
1452         if (flags & SYNC_WAIT)
1453                 log_flags |= XFS_LOG_SYNC;
1454
1455         xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1456
1457         if (flags & (SYNC_ATTR|SYNC_DELWRI)) {
1458                 if (flags & SYNC_BDFLUSH)
1459                         xfs_finish_reclaim_all(mp, 1);
1460                 else
1461                         error = xfs_sync_inodes(mp, flags, xflags, bypassed);
1462         }
1463
1464         /*
1465          * Flushing out dirty data above probably generated more
1466          * log activity, so if this isn't vfs_sync() then flush
1467          * the log again.
1468          */
1469         if (flags & SYNC_DELWRI) {
1470                 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1471         }
1472
1473         if (flags & SYNC_FSDATA) {
1474                 /*
1475                  * If this is vfs_sync() then only sync the superblock
1476                  * if we can lock it without sleeping and it is not pinned.
1477                  */
1478                 if (flags & SYNC_BDFLUSH) {
1479                         bp = xfs_getsb(mp, XFS_BUF_TRYLOCK);
1480                         if (bp != NULL) {
1481                                 bip = XFS_BUF_FSPRIVATE(bp,xfs_buf_log_item_t*);
1482                                 if ((bip != NULL) &&
1483                                     xfs_buf_item_dirty(bip)) {
1484                                         if (!(XFS_BUF_ISPINNED(bp))) {
1485                                                 XFS_BUF_ASYNC(bp);
1486                                                 error = xfs_bwrite(mp, bp);
1487                                         } else {
1488                                                 xfs_buf_relse(bp);
1489                                         }
1490                                 } else {
1491                                         xfs_buf_relse(bp);
1492                                 }
1493                         }
1494                 } else {
1495                         bp = xfs_getsb(mp, 0);
1496                         /*
1497                          * If the buffer is pinned then push on the log so
1498                          * we won't get stuck waiting in the write for
1499                          * someone, maybe ourselves, to flush the log.
1500                          * Even though we just pushed the log above, we
1501                          * did not have the superblock buffer locked at
1502                          * that point so it can become pinned in between
1503                          * there and here.
1504                          */
1505                         if (XFS_BUF_ISPINNED(bp))
1506                                 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
1507                         if (flags & SYNC_WAIT)
1508                                 XFS_BUF_UNASYNC(bp);
1509                         else
1510                                 XFS_BUF_ASYNC(bp);
1511                         error = xfs_bwrite(mp, bp);
1512                 }
1513                 if (error) {
1514                         last_error = error;
1515                 }
1516         }
1517
1518         /*
1519          * If this is the periodic sync, then kick some entries out of
1520          * the reference cache.  This ensures that idle entries are
1521          * eventually kicked out of the cache.
1522          */
1523         if (flags & SYNC_REFCACHE) {
1524                 if (flags & SYNC_WAIT)
1525                         xfs_refcache_purge_mp(mp);
1526                 else
1527                         xfs_refcache_purge_some(mp);
1528         }
1529
1530         /*
1531          * Now check to see if the log needs a "dummy" transaction.
1532          */
1533
1534         if (!(flags & SYNC_REMOUNT) && xfs_log_need_covered(mp)) {
1535                 xfs_trans_t *tp;
1536                 xfs_inode_t *ip;
1537
1538                 /*
1539                  * Put a dummy transaction in the log to tell
1540                  * recovery that all others are OK.
1541                  */
1542                 tp = xfs_trans_alloc(mp, XFS_TRANS_DUMMY1);
1543                 if ((error = xfs_trans_reserve(tp, 0,
1544                                 XFS_ICHANGE_LOG_RES(mp),
1545                                 0, 0, 0)))  {
1546                         xfs_trans_cancel(tp, 0);
1547                         return error;
1548                 }
1549
1550                 ip = mp->m_rootip;
1551                 xfs_ilock(ip, XFS_ILOCK_EXCL);
1552
1553                 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1554                 xfs_trans_ihold(tp, ip);
1555                 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1556                 error = xfs_trans_commit(tp, 0, NULL);
1557                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1558                 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1559         }
1560
1561         /*
1562          * When shutting down, we need to insure that the AIL is pushed
1563          * to disk or the filesystem can appear corrupt from the PROM.
1564          */
1565         if ((flags & (SYNC_CLOSE|SYNC_WAIT)) == (SYNC_CLOSE|SYNC_WAIT)) {
1566                 XFS_bflush(mp->m_ddev_targp);
1567                 if (mp->m_rtdev_targp) {
1568                         XFS_bflush(mp->m_rtdev_targp);
1569                 }
1570         }
1571
1572         return XFS_ERROR(last_error);
1573 }
1574
1575 /*
1576  * xfs_vget - called by DMAPI and NFSD to get vnode from file handle
1577  */
1578 STATIC int
1579 xfs_vget(
1580         bhv_desc_t      *bdp,
1581         vnode_t         **vpp,
1582         fid_t           *fidp)
1583 {
1584         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
1585         xfs_fid_t       *xfid = (struct xfs_fid *)fidp;
1586         xfs_inode_t     *ip;
1587         int             error;
1588         xfs_ino_t       ino;
1589         unsigned int    igen;
1590
1591         /*
1592          * Invalid.  Since handles can be created in user space and passed in
1593          * via gethandle(), this is not cause for a panic.
1594          */
1595         if (xfid->xfs_fid_len != sizeof(*xfid) - sizeof(xfid->xfs_fid_len))
1596                 return XFS_ERROR(EINVAL);
1597
1598         ino  = xfid->xfs_fid_ino;
1599         igen = xfid->xfs_fid_gen;
1600
1601         /*
1602          * NFS can sometimes send requests for ino 0.  Fail them gracefully.
1603          */
1604         if (ino == 0)
1605                 return XFS_ERROR(ESTALE);
1606
1607         error = xfs_iget(mp, NULL, ino, 0, XFS_ILOCK_SHARED, &ip, 0);
1608         if (error) {
1609                 *vpp = NULL;
1610                 return error;
1611         }
1612
1613         if (ip == NULL) {
1614                 *vpp = NULL;
1615                 return XFS_ERROR(EIO);
1616         }
1617
1618         if (ip->i_d.di_mode == 0 || ip->i_d.di_gen != igen) {
1619                 xfs_iput_new(ip, XFS_ILOCK_SHARED);
1620                 *vpp = NULL;
1621                 return XFS_ERROR(ENOENT);
1622         }
1623
1624         *vpp = XFS_ITOV(ip);
1625         xfs_iunlock(ip, XFS_ILOCK_SHARED);
1626         return 0;
1627 }
1628
1629
1630 #define MNTOPT_LOGBUFS  "logbufs"       /* number of XFS log buffers */
1631 #define MNTOPT_LOGBSIZE "logbsize"      /* size of XFS log buffers */
1632 #define MNTOPT_LOGDEV   "logdev"        /* log device */
1633 #define MNTOPT_RTDEV    "rtdev"         /* realtime I/O device */
1634 #define MNTOPT_BIOSIZE  "biosize"       /* log2 of preferred buffered io size */
1635 #define MNTOPT_WSYNC    "wsync"         /* safe-mode nfs compatible mount */
1636 #define MNTOPT_INO64    "ino64"         /* force inodes into 64-bit range */
1637 #define MNTOPT_NOALIGN  "noalign"       /* turn off stripe alignment */
1638 #define MNTOPT_SWALLOC  "swalloc"       /* turn on stripe width allocation */
1639 #define MNTOPT_SUNIT    "sunit"         /* data volume stripe unit */
1640 #define MNTOPT_SWIDTH   "swidth"        /* data volume stripe width */
1641 #define MNTOPT_NOUUID   "nouuid"        /* ignore filesystem UUID */
1642 #define MNTOPT_MTPT     "mtpt"          /* filesystem mount point */
1643 #define MNTOPT_GRPID    "grpid"         /* group-ID from parent directory */
1644 #define MNTOPT_NOGRPID  "nogrpid"       /* group-ID from current process */
1645 #define MNTOPT_BSDGROUPS    "bsdgroups"    /* group-ID from parent directory */
1646 #define MNTOPT_SYSVGROUPS   "sysvgroups"   /* group-ID from current process */
1647 #define MNTOPT_ALLOCSIZE    "allocsize"    /* preferred allocation size */
1648 #define MNTOPT_IHASHSIZE    "ihashsize"    /* size of inode hash table */
1649 #define MNTOPT_NORECOVERY   "norecovery"   /* don't run XFS recovery */
1650 #define MNTOPT_BARRIER  "barrier"       /* use writer barriers for log write and
1651                                          * unwritten extent conversion */
1652 #define MNTOPT_NOBARRIER "nobarrier"    /* .. disable */
1653 #define MNTOPT_OSYNCISOSYNC "osyncisosync" /* o_sync is REALLY o_sync */
1654 #define MNTOPT_64BITINODE   "inode64"   /* inodes can be allocated anywhere */
1655 #define MNTOPT_IKEEP    "ikeep"         /* do not free empty inode clusters */
1656 #define MNTOPT_NOIKEEP  "noikeep"       /* free empty inode clusters */
1657 #define MNTOPT_LARGEIO     "largeio"    /* report large I/O sizes in stat() */
1658 #define MNTOPT_NOLARGEIO   "nolargeio"  /* do not report large I/O sizes
1659                                          * in stat(). */
1660 #define MNTOPT_ATTR2    "attr2"         /* do use attr2 attribute format */
1661 #define MNTOPT_NOATTR2  "noattr2"       /* do not use attr2 attribute format */
1662 #define MNTOPT_TAGXID   "tagxid"        /* context xid tagging for inodes */
1663
1664 STATIC unsigned long
1665 suffix_strtoul(const char *cp, char **endp, unsigned int base)
1666 {
1667         int     last, shift_left_factor = 0;
1668         char    *value = (char *)cp;
1669
1670         last = strlen(value) - 1;
1671         if (value[last] == 'K' || value[last] == 'k') {
1672                 shift_left_factor = 10;
1673                 value[last] = '\0';
1674         }
1675         if (value[last] == 'M' || value[last] == 'm') {
1676                 shift_left_factor = 20;
1677                 value[last] = '\0';
1678         }
1679         if (value[last] == 'G' || value[last] == 'g') {
1680                 shift_left_factor = 30;
1681                 value[last] = '\0';
1682         }
1683
1684         return simple_strtoul(cp, endp, base) << shift_left_factor;
1685 }
1686
1687 STATIC int
1688 xfs_parseargs(
1689         struct bhv_desc         *bhv,
1690         char                    *options,
1691         struct xfs_mount_args   *args,
1692         int                     update)
1693 {
1694         struct vfs              *vfsp = bhvtovfs(bhv);
1695         char                    *this_char, *value, *eov;
1696         int                     dsunit, dswidth, vol_dsunit, vol_dswidth;
1697         int                     iosize;
1698
1699         args->flags |= XFSMNT_IDELETE;
1700         args->flags |= XFSMNT_BARRIER;
1701         args->flags2 |= XFSMNT2_COMPAT_IOSIZE;
1702
1703         if (!options)
1704                 goto done;
1705
1706         iosize = dsunit = dswidth = vol_dsunit = vol_dswidth = 0;
1707
1708         while ((this_char = strsep(&options, ",")) != NULL) {
1709                 if (!*this_char)
1710                         continue;
1711                 if ((value = strchr(this_char, '=')) != NULL)
1712                         *value++ = 0;
1713
1714                 if (!strcmp(this_char, MNTOPT_LOGBUFS)) {
1715                         if (!value || !*value) {
1716                                 printk("XFS: %s option requires an argument\n",
1717                                         this_char);
1718                                 return EINVAL;
1719                         }
1720                         args->logbufs = simple_strtoul(value, &eov, 10);
1721                 } else if (!strcmp(this_char, MNTOPT_LOGBSIZE)) {
1722                         if (!value || !*value) {
1723                                 printk("XFS: %s option requires an argument\n",
1724                                         this_char);
1725                                 return EINVAL;
1726                         }
1727                         args->logbufsize = suffix_strtoul(value, &eov, 10);
1728                 } else if (!strcmp(this_char, MNTOPT_LOGDEV)) {
1729                         if (!value || !*value) {
1730                                 printk("XFS: %s option requires an argument\n",
1731                                         this_char);
1732                                 return EINVAL;
1733                         }
1734                         strncpy(args->logname, value, MAXNAMELEN);
1735                 } else if (!strcmp(this_char, MNTOPT_MTPT)) {
1736                         if (!value || !*value) {
1737                                 printk("XFS: %s option requires an argument\n",
1738                                         this_char);
1739                                 return EINVAL;
1740                         }
1741                         strncpy(args->mtpt, value, MAXNAMELEN);
1742                 } else if (!strcmp(this_char, MNTOPT_RTDEV)) {
1743                         if (!value || !*value) {
1744                                 printk("XFS: %s option requires an argument\n",
1745                                         this_char);
1746                                 return EINVAL;
1747                         }
1748                         strncpy(args->rtname, value, MAXNAMELEN);
1749                 } else if (!strcmp(this_char, MNTOPT_BIOSIZE)) {
1750                         if (!value || !*value) {
1751                                 printk("XFS: %s option requires an argument\n",
1752                                         this_char);
1753                                 return EINVAL;
1754                         }
1755                         iosize = simple_strtoul(value, &eov, 10);
1756                         args->flags |= XFSMNT_IOSIZE;
1757                         args->iosizelog = (uint8_t) iosize;
1758                 } else if (!strcmp(this_char, MNTOPT_ALLOCSIZE)) {
1759                         if (!value || !*value) {
1760                                 printk("XFS: %s option requires an argument\n",
1761                                         this_char);
1762                                 return EINVAL;
1763                         }
1764                         iosize = suffix_strtoul(value, &eov, 10);
1765                         args->flags |= XFSMNT_IOSIZE;
1766                         args->iosizelog = ffs(iosize) - 1;
1767                 } else if (!strcmp(this_char, MNTOPT_IHASHSIZE)) {
1768                         if (!value || !*value) {
1769                                 printk("XFS: %s option requires an argument\n",
1770                                         this_char);
1771                                 return EINVAL;
1772                         }
1773                         args->flags |= XFSMNT_IHASHSIZE;
1774                         args->ihashsize = simple_strtoul(value, &eov, 10);
1775                 } else if (!strcmp(this_char, MNTOPT_GRPID) ||
1776                            !strcmp(this_char, MNTOPT_BSDGROUPS)) {
1777                         vfsp->vfs_flag |= VFS_GRPID;
1778                 } else if (!strcmp(this_char, MNTOPT_NOGRPID) ||
1779                            !strcmp(this_char, MNTOPT_SYSVGROUPS)) {
1780                         vfsp->vfs_flag &= ~VFS_GRPID;
1781                 } else if (!strcmp(this_char, MNTOPT_WSYNC)) {
1782                         args->flags |= XFSMNT_WSYNC;
1783                 } else if (!strcmp(this_char, MNTOPT_OSYNCISOSYNC)) {
1784                         args->flags |= XFSMNT_OSYNCISOSYNC;
1785                 } else if (!strcmp(this_char, MNTOPT_NORECOVERY)) {
1786                         args->flags |= XFSMNT_NORECOVERY;
1787                 } else if (!strcmp(this_char, MNTOPT_INO64)) {
1788                         args->flags |= XFSMNT_INO64;
1789 #if !XFS_BIG_INUMS
1790                         printk("XFS: %s option not allowed on this system\n",
1791                                 this_char);
1792                         return EINVAL;
1793 #endif
1794                 } else if (!strcmp(this_char, MNTOPT_NOALIGN)) {
1795                         args->flags |= XFSMNT_NOALIGN;
1796                 } else if (!strcmp(this_char, MNTOPT_SWALLOC)) {
1797                         args->flags |= XFSMNT_SWALLOC;
1798                 } else if (!strcmp(this_char, MNTOPT_SUNIT)) {
1799                         if (!value || !*value) {
1800                                 printk("XFS: %s option requires an argument\n",
1801                                         this_char);
1802                                 return EINVAL;
1803                         }
1804                         dsunit = simple_strtoul(value, &eov, 10);
1805                 } else if (!strcmp(this_char, MNTOPT_SWIDTH)) {
1806                         if (!value || !*value) {
1807                                 printk("XFS: %s option requires an argument\n",
1808                                         this_char);
1809                                 return EINVAL;
1810                         }
1811                         dswidth = simple_strtoul(value, &eov, 10);
1812                 } else if (!strcmp(this_char, MNTOPT_64BITINODE)) {
1813                         args->flags &= ~XFSMNT_32BITINODES;
1814 #if !XFS_BIG_INUMS
1815                         printk("XFS: %s option not allowed on this system\n",
1816                                 this_char);
1817                         return EINVAL;
1818 #endif
1819                 } else if (!strcmp(this_char, MNTOPT_NOUUID)) {
1820                         args->flags |= XFSMNT_NOUUID;
1821                 } else if (!strcmp(this_char, MNTOPT_BARRIER)) {
1822                         args->flags |= XFSMNT_BARRIER;
1823                 } else if (!strcmp(this_char, MNTOPT_NOBARRIER)) {
1824                         args->flags &= ~XFSMNT_BARRIER;
1825                 } else if (!strcmp(this_char, MNTOPT_IKEEP)) {
1826                         args->flags &= ~XFSMNT_IDELETE;
1827                 } else if (!strcmp(this_char, MNTOPT_NOIKEEP)) {
1828                         args->flags |= XFSMNT_IDELETE;
1829                 } else if (!strcmp(this_char, MNTOPT_LARGEIO)) {
1830                         args->flags2 &= ~XFSMNT2_COMPAT_IOSIZE;
1831                 } else if (!strcmp(this_char, MNTOPT_NOLARGEIO)) {
1832                         args->flags2 |= XFSMNT2_COMPAT_IOSIZE;
1833                 } else if (!strcmp(this_char, MNTOPT_ATTR2)) {
1834                         args->flags |= XFSMNT_ATTR2;
1835                 } else if (!strcmp(this_char, MNTOPT_NOATTR2)) {
1836                         args->flags &= ~XFSMNT_ATTR2;
1837 #ifndef CONFIG_INOXID_NONE
1838                 } else if (!strcmp(this_char, MNTOPT_TAGXID)) {
1839                         args->flags2 |= XFSMNT2_TAGXID;
1840 #endif
1841                 } else if (!strcmp(this_char, "osyncisdsync")) {
1842                         /* no-op, this is now the default */
1843 printk("XFS: osyncisdsync is now the default, option is deprecated.\n");
1844                 } else if (!strcmp(this_char, "irixsgid")) {
1845 printk("XFS: irixsgid is now a sysctl(2) variable, option is deprecated.\n");
1846                 } else {
1847                         printk("XFS: unknown mount option [%s].\n", this_char);
1848                         return EINVAL;
1849                 }
1850         }
1851
1852         if (args->flags & XFSMNT_NORECOVERY) {
1853                 if ((vfsp->vfs_flag & VFS_RDONLY) == 0) {
1854                         printk("XFS: no-recovery mounts must be read-only.\n");
1855                         return EINVAL;
1856                 }
1857         }
1858
1859         if ((args->flags & XFSMNT_NOALIGN) && (dsunit || dswidth)) {
1860                 printk(
1861         "XFS: sunit and swidth options incompatible with the noalign option\n");
1862                 return EINVAL;
1863         }
1864
1865         if ((dsunit && !dswidth) || (!dsunit && dswidth)) {
1866                 printk("XFS: sunit and swidth must be specified together\n");
1867                 return EINVAL;
1868         }
1869
1870         if (dsunit && (dswidth % dsunit != 0)) {
1871                 printk(
1872         "XFS: stripe width (%d) must be a multiple of the stripe unit (%d)\n",
1873                         dswidth, dsunit);
1874                 return EINVAL;
1875         }
1876
1877         if ((args->flags & XFSMNT_NOALIGN) != XFSMNT_NOALIGN) {
1878                 if (dsunit) {
1879                         args->sunit = dsunit;
1880                         args->flags |= XFSMNT_RETERR;
1881                 } else {
1882                         args->sunit = vol_dsunit;
1883                 }
1884                 dswidth ? (args->swidth = dswidth) :
1885                           (args->swidth = vol_dswidth);
1886         } else {
1887                 args->sunit = args->swidth = 0;
1888         }
1889
1890 done:
1891         if (args->flags & XFSMNT_32BITINODES)
1892                 vfsp->vfs_flag |= VFS_32BITINODES;
1893         if (args->flags2)
1894                 args->flags |= XFSMNT_FLAGS2;
1895         return 0;
1896 }
1897
1898 STATIC int
1899 xfs_showargs(
1900         struct bhv_desc         *bhv,
1901         struct seq_file         *m)
1902 {
1903         static struct proc_xfs_info {
1904                 int     flag;
1905                 char    *str;
1906         } xfs_info[] = {
1907                 /* the few simple ones we can get from the mount struct */
1908                 { XFS_MOUNT_WSYNC,              "," MNTOPT_WSYNC },
1909                 { XFS_MOUNT_INO64,              "," MNTOPT_INO64 },
1910                 { XFS_MOUNT_NOALIGN,            "," MNTOPT_NOALIGN },
1911                 { XFS_MOUNT_SWALLOC,            "," MNTOPT_SWALLOC },
1912                 { XFS_MOUNT_NOUUID,             "," MNTOPT_NOUUID },
1913                 { XFS_MOUNT_NORECOVERY,         "," MNTOPT_NORECOVERY },
1914                 { XFS_MOUNT_OSYNCISOSYNC,       "," MNTOPT_OSYNCISOSYNC },
1915                 { 0, NULL }
1916         };
1917         struct proc_xfs_info    *xfs_infop;
1918         struct xfs_mount        *mp = XFS_BHVTOM(bhv);
1919         struct vfs              *vfsp = XFS_MTOVFS(mp);
1920
1921         for (xfs_infop = xfs_info; xfs_infop->flag; xfs_infop++) {
1922                 if (mp->m_flags & xfs_infop->flag)
1923                         seq_puts(m, xfs_infop->str);
1924         }
1925
1926         if (mp->m_flags & XFS_MOUNT_IHASHSIZE)
1927                 seq_printf(m, "," MNTOPT_IHASHSIZE "=%d", mp->m_ihsize);
1928
1929         if (mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)
1930                 seq_printf(m, "," MNTOPT_ALLOCSIZE "=%dk",
1931                                 (int)(1 << mp->m_writeio_log) >> 10);
1932
1933         if (mp->m_logbufs > 0)
1934                 seq_printf(m, "," MNTOPT_LOGBUFS "=%d", mp->m_logbufs);
1935         if (mp->m_logbsize > 0)
1936                 seq_printf(m, "," MNTOPT_LOGBSIZE "=%dk", mp->m_logbsize >> 10);
1937
1938         if (mp->m_logname)
1939                 seq_printf(m, "," MNTOPT_LOGDEV "=%s", mp->m_logname);
1940         if (mp->m_rtname)
1941                 seq_printf(m, "," MNTOPT_RTDEV "=%s", mp->m_rtname);
1942
1943         if (mp->m_dalign > 0)
1944                 seq_printf(m, "," MNTOPT_SUNIT "=%d",
1945                                 (int)XFS_FSB_TO_BB(mp, mp->m_dalign));
1946         if (mp->m_swidth > 0)
1947                 seq_printf(m, "," MNTOPT_SWIDTH "=%d",
1948                                 (int)XFS_FSB_TO_BB(mp, mp->m_swidth));
1949
1950         if (!(mp->m_flags & XFS_MOUNT_IDELETE))
1951                 seq_printf(m, "," MNTOPT_IKEEP);
1952         if (!(mp->m_flags & XFS_MOUNT_COMPAT_IOSIZE))
1953                 seq_printf(m, "," MNTOPT_LARGEIO);
1954
1955         if (!(vfsp->vfs_flag & VFS_32BITINODES))
1956                 seq_printf(m, "," MNTOPT_64BITINODE);
1957         if (vfsp->vfs_flag & VFS_GRPID)
1958                 seq_printf(m, "," MNTOPT_GRPID);
1959
1960         return 0;
1961 }
1962
1963 STATIC void
1964 xfs_freeze(
1965         bhv_desc_t      *bdp)
1966 {
1967         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
1968
1969         while (atomic_read(&mp->m_active_trans) > 0)
1970                 delay(100);
1971
1972         /* Push the superblock and write an unmount record */
1973         xfs_log_unmount_write(mp);
1974         xfs_unmountfs_writesb(mp);
1975         xfs_fs_log_dummy(mp);
1976 }
1977
1978
1979 vfsops_t xfs_vfsops = {
1980         BHV_IDENTITY_INIT(VFS_BHV_XFS,VFS_POSITION_XFS),
1981         .vfs_parseargs          = xfs_parseargs,
1982         .vfs_showargs           = xfs_showargs,
1983         .vfs_mount              = xfs_mount,
1984         .vfs_unmount            = xfs_unmount,
1985         .vfs_mntupdate          = xfs_mntupdate,
1986         .vfs_root               = xfs_root,
1987         .vfs_statvfs            = xfs_statvfs,
1988         .vfs_sync               = xfs_sync,
1989         .vfs_vget               = xfs_vget,
1990         .vfs_dmapiops           = (vfs_dmapiops_t)fs_nosys,
1991         .vfs_quotactl           = (vfs_quotactl_t)fs_nosys,
1992         .vfs_init_vnode         = xfs_initialize_vnode,
1993         .vfs_force_shutdown     = xfs_do_force_shutdown,
1994         .vfs_freeze             = xfs_freeze,
1995 };