SAFE public projects git trees. - safe/jmp/linux-2.6/blob - fs/xfs/xfs_vfsops.c

   1 /*
   2  * Copyright (c) 2000-2005 Silicon Graphics, Inc.  All Rights Reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or modify it
   5  * under the terms of version 2 of the GNU General Public License as
   6  * published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it would be useful, but
   9  * WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
  11  *
  12  * Further, this software is distributed without any warranty that it is
  13  * free of the rightful claim of any third person regarding infringement
  14  * or the like.  Any license provided herein, whether implied or
  15  * otherwise, applies only to this software file.  Patent licenses, if
  16  * any, provided herein do not apply to combinations of this program with
  17  * other software, or any other product whatsoever.
  18  *
  19  * You should have received a copy of the GNU General Public License along
  20  * with this program; if not, write the Free Software Foundation, Inc., 59
  21  * Temple Place - Suite 330, Boston MA 02111-1307, USA.
  22  *
  23  * Contact information: Silicon Graphics, Inc., 1600 Amphitheatre Pkwy,
  24  * Mountain View, CA  94043, or:
  25  *
  26  * http://www.sgi.com
  27  *
  28  * For further information regarding this notice, see:
  29  *
  30  * http://oss.sgi.com/projects/GenInfo/SGIGPLNoticeExplan/
  31  */
  32 #include "xfs.h"
  33 #include "xfs_fs.h"
  34 #include "xfs_types.h"
  35 #include "xfs_bit.h"
  36 #include "xfs_log.h"
  37 #include "xfs_inum.h"
  38 #include "xfs_trans.h"
  39 #include "xfs_sb.h"
  40 #include "xfs_ag.h"
  41 #include "xfs_dir.h"
  42 #include "xfs_dir2.h"
  43 #include "xfs_dmapi.h"
  44 #include "xfs_mount.h"
  45 #include "xfs_da_btree.h"
  46 #include "xfs_bmap_btree.h"
  47 #include "xfs_ialloc_btree.h"
  48 #include "xfs_alloc_btree.h"
  49 #include "xfs_dir_sf.h"
  50 #include "xfs_dir2_sf.h"
  51 #include "xfs_attr_sf.h"
  52 #include "xfs_dinode.h"
  53 #include "xfs_inode.h"
  54 #include "xfs_inode_item.h"
  55 #include "xfs_btree.h"
  56 #include "xfs_alloc.h"
  57 #include "xfs_ialloc.h"
  58 #include "xfs_quota.h"
  59 #include "xfs_error.h"
  60 #include "xfs_bmap.h"
  61 #include "xfs_rw.h"
  62 #include "xfs_refcache.h"
  63 #include "xfs_buf_item.h"
  64 #include "xfs_log_priv.h"
  65 #include "xfs_dir2_trace.h"
  66 #include "xfs_extfree_item.h"
  67 #include "xfs_acl.h"
  68 #include "xfs_attr.h"
  69 #include "xfs_clnt.h"
  70
  71 STATIC int xfs_sync(bhv_desc_t *, int, cred_t *);
  72
  73 int
  74 xfs_init(void)
  75 {
  76         extern kmem_zone_t      *xfs_bmap_free_item_zone;
  77         extern kmem_zone_t      *xfs_btree_cur_zone;
  78         extern kmem_zone_t      *xfs_trans_zone;
  79         extern kmem_zone_t      *xfs_buf_item_zone;
  80         extern kmem_zone_t      *xfs_dabuf_zone;
  81 #ifdef XFS_DABUF_DEBUG
  82         extern lock_t           xfs_dabuf_global_lock;
  83         spinlock_init(&xfs_dabuf_global_lock, "xfsda");
  84 #endif
  85
  86         /*
  87          * Initialize all of the zone allocators we use.
  88          */
  89         xfs_bmap_free_item_zone = kmem_zone_init(sizeof(xfs_bmap_free_item_t),
  90                                                  "xfs_bmap_free_item");
  91         xfs_btree_cur_zone = kmem_zone_init(sizeof(xfs_btree_cur_t),
  92                                             "xfs_btree_cur");
  93         xfs_inode_zone = kmem_zone_init(sizeof(xfs_inode_t), "xfs_inode");
  94         xfs_trans_zone = kmem_zone_init(sizeof(xfs_trans_t), "xfs_trans");
  95         xfs_da_state_zone =
  96                 kmem_zone_init(sizeof(xfs_da_state_t), "xfs_da_state");
  97         xfs_dabuf_zone = kmem_zone_init(sizeof(xfs_dabuf_t), "xfs_dabuf");
  98
  99         /*
 100          * The size of the zone allocated buf log item is the maximum
 101          * size possible under XFS.  This wastes a little bit of memory,
 102          * but it is much faster.
 103          */
 104         xfs_buf_item_zone =
 105                 kmem_zone_init((sizeof(xfs_buf_log_item_t) +
 106                                 (((XFS_MAX_BLOCKSIZE / XFS_BLI_CHUNK) /
 107                                   NBWORD) * sizeof(int))),
 108                                "xfs_buf_item");
 109         xfs_efd_zone = kmem_zone_init((sizeof(xfs_efd_log_item_t) +
 110                                        ((XFS_EFD_MAX_FAST_EXTENTS - 1) * sizeof(xfs_extent_t))),
 111                                       "xfs_efd_item");
 112         xfs_efi_zone = kmem_zone_init((sizeof(xfs_efi_log_item_t) +
 113                                        ((XFS_EFI_MAX_FAST_EXTENTS - 1) * sizeof(xfs_extent_t))),
 114                                       "xfs_efi_item");
 115         xfs_ifork_zone = kmem_zone_init(sizeof(xfs_ifork_t), "xfs_ifork");
 116         xfs_ili_zone = kmem_zone_init(sizeof(xfs_inode_log_item_t), "xfs_ili");
 117         xfs_chashlist_zone = kmem_zone_init(sizeof(xfs_chashlist_t),
 118                                             "xfs_chashlist");
 119         xfs_acl_zone_init(xfs_acl_zone, "xfs_acl");
 120
 121         /*
 122          * Allocate global trace buffers.
 123          */
 124 #ifdef XFS_ALLOC_TRACE
 125         xfs_alloc_trace_buf = ktrace_alloc(XFS_ALLOC_TRACE_SIZE, KM_SLEEP);
 126 #endif
 127 #ifdef XFS_BMAP_TRACE
 128         xfs_bmap_trace_buf = ktrace_alloc(XFS_BMAP_TRACE_SIZE, KM_SLEEP);
 129 #endif
 130 #ifdef XFS_BMBT_TRACE
 131         xfs_bmbt_trace_buf = ktrace_alloc(XFS_BMBT_TRACE_SIZE, KM_SLEEP);
 132 #endif
 133 #ifdef XFS_DIR_TRACE
 134         xfs_dir_trace_buf = ktrace_alloc(XFS_DIR_TRACE_SIZE, KM_SLEEP);
 135 #endif
 136 #ifdef XFS_ATTR_TRACE
 137         xfs_attr_trace_buf = ktrace_alloc(XFS_ATTR_TRACE_SIZE, KM_SLEEP);
 138 #endif
 139 #ifdef XFS_DIR2_TRACE
 140         xfs_dir2_trace_buf = ktrace_alloc(XFS_DIR2_GTRACE_SIZE, KM_SLEEP);
 141 #endif
 142
 143         xfs_dir_startup();
 144
 145 #if (defined(DEBUG) || defined(INDUCE_IO_ERROR))
 146         xfs_error_test_init();
 147 #endif /* DEBUG || INDUCE_IO_ERROR */
 148
 149         xfs_init_procfs();
 150         xfs_sysctl_register();
 151         return 0;
 152 }
 153
 154 void
 155 xfs_cleanup(void)
 156 {
 157         extern kmem_zone_t      *xfs_bmap_free_item_zone;
 158         extern kmem_zone_t      *xfs_btree_cur_zone;
 159         extern kmem_zone_t      *xfs_inode_zone;
 160         extern kmem_zone_t      *xfs_trans_zone;
 161         extern kmem_zone_t      *xfs_da_state_zone;
 162         extern kmem_zone_t      *xfs_dabuf_zone;
 163         extern kmem_zone_t      *xfs_efd_zone;
 164         extern kmem_zone_t      *xfs_efi_zone;
 165         extern kmem_zone_t      *xfs_buf_item_zone;
 166         extern kmem_zone_t      *xfs_chashlist_zone;
 167
 168         xfs_cleanup_procfs();
 169         xfs_sysctl_unregister();
 170         xfs_refcache_destroy();
 171         xfs_acl_zone_destroy(xfs_acl_zone);
 172
 173 #ifdef XFS_DIR2_TRACE
 174         ktrace_free(xfs_dir2_trace_buf);
 175 #endif
 176 #ifdef XFS_ATTR_TRACE
 177         ktrace_free(xfs_attr_trace_buf);
 178 #endif
 179 #ifdef XFS_DIR_TRACE
 180         ktrace_free(xfs_dir_trace_buf);
 181 #endif
 182 #ifdef XFS_BMBT_TRACE
 183         ktrace_free(xfs_bmbt_trace_buf);
 184 #endif
 185 #ifdef XFS_BMAP_TRACE
 186         ktrace_free(xfs_bmap_trace_buf);
 187 #endif
 188 #ifdef XFS_ALLOC_TRACE
 189         ktrace_free(xfs_alloc_trace_buf);
 190 #endif
 191
 192         kmem_cache_destroy(xfs_bmap_free_item_zone);
 193         kmem_cache_destroy(xfs_btree_cur_zone);
 194         kmem_cache_destroy(xfs_inode_zone);
 195         kmem_cache_destroy(xfs_trans_zone);
 196         kmem_cache_destroy(xfs_da_state_zone);
 197         kmem_cache_destroy(xfs_dabuf_zone);
 198         kmem_cache_destroy(xfs_buf_item_zone);
 199         kmem_cache_destroy(xfs_efd_zone);
 200         kmem_cache_destroy(xfs_efi_zone);
 201         kmem_cache_destroy(xfs_ifork_zone);
 202         kmem_cache_destroy(xfs_ili_zone);
 203         kmem_cache_destroy(xfs_chashlist_zone);
 204 }
 205
 206 /*
 207  * xfs_start_flags
 208  *
 209  * This function fills in xfs_mount_t fields based on mount args.
 210  * Note: the superblock has _not_ yet been read in.
 211  */
 212 STATIC int
 213 xfs_start_flags(
 214         struct vfs              *vfs,
 215         struct xfs_mount_args   *ap,
 216         struct xfs_mount        *mp)
 217 {
 218         /* Values are in BBs */
 219         if ((ap->flags & XFSMNT_NOALIGN) != XFSMNT_NOALIGN) {
 220                 /*
 221                  * At this point the superblock has not been read
 222                  * in, therefore we do not know the block size.
 223                  * Before the mount call ends we will convert
 224                  * these to FSBs.
 225                  */
 226                 mp->m_dalign = ap->sunit;
 227                 mp->m_swidth = ap->swidth;
 228         }
 229
 230         if (ap->logbufs != -1 &&
 231 #if defined(DEBUG) || defined(XLOG_NOLOG)
 232             ap->logbufs != 0 &&
 233 #endif
 234             (ap->logbufs < XLOG_MIN_ICLOGS ||
 235              ap->logbufs > XLOG_MAX_ICLOGS)) {
 236                 cmn_err(CE_WARN,
 237                         "XFS: invalid logbufs value: %d [not %d-%d]",
 238                         ap->logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
 239                 return XFS_ERROR(EINVAL);
 240         }
 241         mp->m_logbufs = ap->logbufs;
 242         if (ap->logbufsize != -1 &&
 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_NOATIME)
 276                 mp->m_flags |= XFS_MOUNT_NOATIME;
 277         if (ap->flags & XFSMNT_RETERR)
 278                 mp->m_flags |= XFS_MOUNT_RETERR;
 279         if (ap->flags & XFSMNT_NOALIGN)
 280                 mp->m_flags |= XFS_MOUNT_NOALIGN;
 281         if (ap->flags & XFSMNT_SWALLOC)
 282                 mp->m_flags |= XFS_MOUNT_SWALLOC;
 283         if (ap->flags & XFSMNT_OSYNCISOSYNC)
 284                 mp->m_flags |= XFS_MOUNT_OSYNCISOSYNC;
 285         if (ap->flags & XFSMNT_32BITINODES)
 286                 mp->m_flags |= (XFS_MOUNT_32BITINODES | XFS_MOUNT_32BITINOOPT);
 287
 288         if (ap->flags & XFSMNT_IOSIZE) {
 289                 if (ap->iosizelog > XFS_MAX_IO_LOG ||
 290                     ap->iosizelog < XFS_MIN_IO_LOG) {
 291                         cmn_err(CE_WARN,
 292                 "XFS: invalid log iosize: %d [not %d-%d]",
 293                                 ap->iosizelog, XFS_MIN_IO_LOG,
 294                                 XFS_MAX_IO_LOG);
 295                         return XFS_ERROR(EINVAL);
 296                 }
 297
 298                 mp->m_flags |= XFS_MOUNT_DFLT_IOSIZE;
 299                 mp->m_readio_log = mp->m_writeio_log = ap->iosizelog;
 300         }
 301
 302         if (ap->flags & XFSMNT_IHASHSIZE)
 303                 mp->m_flags |= XFS_MOUNT_IHASHSIZE;
 304         if (ap->flags & XFSMNT_IDELETE)
 305                 mp->m_flags |= XFS_MOUNT_IDELETE;
 306         if (ap->flags & XFSMNT_DIRSYNC)
 307                 mp->m_flags |= XFS_MOUNT_DIRSYNC;
 308         if (ap->flags & XFSMNT_COMPAT_IOSIZE)
 309                 mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE;
 310         if (ap->flags & XFSMNT_COMPAT_ATTR)
 311                 mp->m_flags |= XFS_MOUNT_COMPAT_ATTR;
 312
 313         /*
 314          * no recovery flag requires a read-only mount
 315          */
 316         if (ap->flags & XFSMNT_NORECOVERY) {
 317                 if (!(vfs->vfs_flag & VFS_RDONLY)) {
 318                         cmn_err(CE_WARN,
 319         "XFS: tried to mount a FS read-write without recovery!");
 320                         return XFS_ERROR(EINVAL);
 321                 }
 322                 mp->m_flags |= XFS_MOUNT_NORECOVERY;
 323         }
 324
 325         if (ap->flags & XFSMNT_NOUUID)
 326                 mp->m_flags |= XFS_MOUNT_NOUUID;
 327         if (ap->flags & XFSMNT_BARRIER)
 328                 mp->m_flags |= XFS_MOUNT_BARRIER;
 329
 330         return 0;
 331 }
 332
 333 /*
 334  * This function fills in xfs_mount_t fields based on mount args.
 335  * Note: the superblock _has_ now been read in.
 336  */
 337 STATIC int
 338 xfs_finish_flags(
 339         struct vfs              *vfs,
 340         struct xfs_mount_args   *ap,
 341         struct xfs_mount        *mp)
 342 {
 343         int                     ronly = (vfs->vfs_flag & VFS_RDONLY);
 344
 345         /* Fail a mount where the logbuf is smaller then the log stripe */
 346         if (XFS_SB_VERSION_HASLOGV2(&mp->m_sb)) {
 347                 if ((ap->logbufsize == -1) &&
 348                     (mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE)) {
 349                         mp->m_logbsize = mp->m_sb.sb_logsunit;
 350                 } else if (ap->logbufsize < mp->m_sb.sb_logsunit) {
 351                         cmn_err(CE_WARN,
 352         "XFS: logbuf size must be greater than or equal to log stripe size");
 353                         return XFS_ERROR(EINVAL);
 354                 }
 355         } else {
 356                 /* Fail a mount if the logbuf is larger than 32K */
 357                 if (ap->logbufsize > XLOG_BIG_RECORD_BSIZE) {
 358                         cmn_err(CE_WARN,
 359         "XFS: logbuf size for version 1 logs must be 16K or 32K");
 360                         return XFS_ERROR(EINVAL);
 361                 }
 362         }
 363
 364         /*
 365          * prohibit r/w mounts of read-only filesystems
 366          */
 367         if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !ronly) {
 368                 cmn_err(CE_WARN,
 369         "XFS: cannot mount a read-only filesystem as read-write");
 370                 return XFS_ERROR(EROFS);
 371         }
 372
 373         /*
 374          * check for shared mount.
 375          */
 376         if (ap->flags & XFSMNT_SHARED) {
 377                 if (!XFS_SB_VERSION_HASSHARED(&mp->m_sb))
 378                         return XFS_ERROR(EINVAL);
 379
 380                 /*
 381                  * For IRIX 6.5, shared mounts must have the shared
 382                  * version bit set, have the persistent readonly
 383                  * field set, must be version 0 and can only be mounted
 384                  * read-only.
 385                  */
 386                 if (!ronly || !(mp->m_sb.sb_flags & XFS_SBF_READONLY) ||
 387                      (mp->m_sb.sb_shared_vn != 0))
 388                         return XFS_ERROR(EINVAL);
 389
 390                 mp->m_flags |= XFS_MOUNT_SHARED;
 391
 392                 /*
 393                  * Shared XFS V0 can't deal with DMI.  Return EINVAL.
 394                  */
 395                 if (mp->m_sb.sb_shared_vn == 0 && (ap->flags & XFSMNT_DMAPI))
 396                         return XFS_ERROR(EINVAL);
 397         }
 398
 399         return 0;
 400 }
 401
 402 /*
 403  * xfs_mount
 404  *
 405  * The file system configurations are:
 406  *      (1) device (partition) with data and internal log
 407  *      (2) logical volume with data and log subvolumes.
 408  *      (3) logical volume with data, log, and realtime subvolumes.
 409  *
 410  * We only have to handle opening the log and realtime volumes here if
 411  * they are present.  The data subvolume has already been opened by
 412  * get_sb_bdev() and is stored in vfsp->vfs_super->s_bdev.
 413  */
 414 STATIC int
 415 xfs_mount(
 416         struct bhv_desc         *bhvp,
 417         struct xfs_mount_args   *args,
 418         cred_t                  *credp)
 419 {
 420         struct vfs              *vfsp = bhvtovfs(bhvp);
 421         struct bhv_desc         *p;
 422         struct xfs_mount        *mp = XFS_BHVTOM(bhvp);
 423         struct block_device     *ddev, *logdev, *rtdev;
 424         int                     flags = 0, error;
 425
 426         ddev = vfsp->vfs_super->s_bdev;
 427         logdev = rtdev = NULL;
 428
 429         /*
 430          * Setup xfs_mount function vectors from available behaviors
 431          */
 432         p = vfs_bhv_lookup(vfsp, VFS_POSITION_DM);
 433         mp->m_dm_ops = p ? *(xfs_dmops_t *) vfs_bhv_custom(p) : xfs_dmcore_stub;
 434         p = vfs_bhv_lookup(vfsp, VFS_POSITION_QM);
 435         mp->m_qm_ops = p ? *(xfs_qmops_t *) vfs_bhv_custom(p) : xfs_qmcore_stub;
 436         p = vfs_bhv_lookup(vfsp, VFS_POSITION_IO);
 437         mp->m_io_ops = p ? *(xfs_ioops_t *) vfs_bhv_custom(p) : xfs_iocore_xfs;
 438
 439         /*
 440          * Open real time and log devices - order is important.
 441          */
 442         if (args->logname[0]) {
 443                 error = xfs_blkdev_get(mp, args->logname, &logdev);
 444                 if (error)
 445                         return error;
 446         }
 447         if (args->rtname[0]) {
 448                 error = xfs_blkdev_get(mp, args->rtname, &rtdev);
 449                 if (error) {
 450                         xfs_blkdev_put(logdev);
 451                         return error;
 452                 }
 453
 454                 if (rtdev == ddev || rtdev == logdev) {
 455                         cmn_err(CE_WARN,
 456         "XFS: Cannot mount filesystem with identical rtdev and ddev/logdev.");
 457                         xfs_blkdev_put(logdev);
 458                         xfs_blkdev_put(rtdev);
 459                         return EINVAL;
 460                 }
 461         }
 462
 463         /*
 464          * Setup xfs_mount buffer target pointers
 465          */
 466         error = ENOMEM;
 467         mp->m_ddev_targp = xfs_alloc_buftarg(ddev, 0);
 468         if (!mp->m_ddev_targp) {
 469                 xfs_blkdev_put(logdev);
 470                 xfs_blkdev_put(rtdev);
 471                 return error;
 472         }
 473         if (rtdev) {
 474                 mp->m_rtdev_targp = xfs_alloc_buftarg(rtdev, 1);
 475                 if (!mp->m_rtdev_targp)
 476                         goto error0;
 477         }
 478         mp->m_logdev_targp = (logdev && logdev != ddev) ?
 479                                 xfs_alloc_buftarg(logdev, 1) : mp->m_ddev_targp;
 480         if (!mp->m_logdev_targp)
 481                 goto error0;
 482
 483         /*
 484          * Setup flags based on mount(2) options and then the superblock
 485          */
 486         error = xfs_start_flags(vfsp, args, mp);
 487         if (error)
 488                 goto error1;
 489         error = xfs_readsb(mp);
 490         if (error)
 491                 goto error1;
 492         error = xfs_finish_flags(vfsp, args, mp);
 493         if (error)
 494                 goto error2;
 495
 496         /*
 497          * Setup xfs_mount buffer target pointers based on superblock
 498          */
 499         error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_blocksize,
 500                                     mp->m_sb.sb_sectsize);
 501         if (!error && logdev && logdev != ddev) {
 502                 unsigned int    log_sector_size = BBSIZE;
 503
 504                 if (XFS_SB_VERSION_HASSECTOR(&mp->m_sb))
 505                         log_sector_size = mp->m_sb.sb_logsectsize;
 506                 error = xfs_setsize_buftarg(mp->m_logdev_targp,
 507                                             mp->m_sb.sb_blocksize,
 508                                             log_sector_size);
 509         }
 510         if (!error && rtdev)
 511                 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
 512                                             mp->m_sb.sb_blocksize,
 513                                             mp->m_sb.sb_sectsize);
 514         if (error)
 515                 goto error2;
 516
 517         error = XFS_IOINIT(vfsp, args, flags);
 518         if (error)
 519                 goto error2;
 520
 521         if ((args->flags & XFSMNT_BARRIER) &&
 522             !(XFS_MTOVFS(mp)->vfs_flag & VFS_RDONLY))
 523                 xfs_mountfs_check_barriers(mp);
 524         return 0;
 525
 526 error2:
 527         if (mp->m_sb_bp)
 528                 xfs_freesb(mp);
 529 error1:
 530         xfs_binval(mp->m_ddev_targp);
 531         if (logdev && logdev != ddev)
 532                 xfs_binval(mp->m_logdev_targp);
 533         if (rtdev)
 534                 xfs_binval(mp->m_rtdev_targp);
 535 error0:
 536         xfs_unmountfs_close(mp, credp);
 537         return error;
 538 }
 539
 540 STATIC int
 541 xfs_unmount(
 542         bhv_desc_t      *bdp,
 543         int             flags,
 544         cred_t          *credp)
 545 {
 546         struct vfs      *vfsp = bhvtovfs(bdp);
 547         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
 548         xfs_inode_t     *rip;
 549         vnode_t         *rvp;
 550         int             unmount_event_wanted = 0;
 551         int             unmount_event_flags = 0;
 552         int             xfs_unmountfs_needed = 0;
 553         int             error;
 554
 555         rip = mp->m_rootip;
 556         rvp = XFS_ITOV(rip);
 557
 558         if (vfsp->vfs_flag & VFS_DMI) {
 559                 error = XFS_SEND_PREUNMOUNT(mp, vfsp,
 560                                 rvp, DM_RIGHT_NULL, rvp, DM_RIGHT_NULL,
 561                                 NULL, NULL, 0, 0,
 562                                 (mp->m_dmevmask & (1<<DM_EVENT_PREUNMOUNT))?
 563                                         0:DM_FLAGS_UNWANTED);
 564                         if (error)
 565                                 return XFS_ERROR(error);
 566                 unmount_event_wanted = 1;
 567                 unmount_event_flags = (mp->m_dmevmask & (1<<DM_EVENT_UNMOUNT))?
 568                                         0 : DM_FLAGS_UNWANTED;
 569         }
 570
 571         /*
 572          * First blow any referenced inode from this file system
 573          * out of the reference cache, and delete the timer.
 574          */
 575         xfs_refcache_purge_mp(mp);
 576
 577         XFS_bflush(mp->m_ddev_targp);
 578         error = xfs_unmount_flush(mp, 0);
 579         if (error)
 580                 goto out;
 581
 582         ASSERT(vn_count(rvp) == 1);
 583
 584         /*
 585          * Drop the reference count
 586          */
 587         VN_RELE(rvp);
 588
 589         /*
 590          * If we're forcing a shutdown, typically because of a media error,
 591          * we want to make sure we invalidate dirty pages that belong to
 592          * referenced vnodes as well.
 593          */
 594         if (XFS_FORCED_SHUTDOWN(mp)) {
 595                 error = xfs_sync(&mp->m_bhv,
 596                          (SYNC_WAIT | SYNC_CLOSE), credp);
 597                 ASSERT(error != EFSCORRUPTED);
 598         }
 599         xfs_unmountfs_needed = 1;
 600
 601 out:
 602         /*      Send DMAPI event, if required.
 603          *      Then do xfs_unmountfs() if needed.
 604          *      Then return error (or zero).
 605          */
 606         if (unmount_event_wanted) {
 607                 /* Note: mp structure must still exist for
 608                  * XFS_SEND_UNMOUNT() call.
 609                  */
 610                 XFS_SEND_UNMOUNT(mp, vfsp, error == 0 ? rvp : NULL,
 611                         DM_RIGHT_NULL, 0, error, unmount_event_flags);
 612         }
 613         if (xfs_unmountfs_needed) {
 614                 /*
 615                  * Call common unmount function to flush to disk
 616                  * and free the super block buffer & mount structures.
 617                  */
 618                 xfs_unmountfs(mp, credp);
 619         }
 620
 621         return XFS_ERROR(error);
 622 }
 623
 624 STATIC int
 625 xfs_quiesce_fs(
 626         xfs_mount_t             *mp)
 627 {
 628         int                     count = 0, pincount;
 629
 630         xfs_refcache_purge_mp(mp);
 631         xfs_flush_buftarg(mp->m_ddev_targp, 0);
 632         xfs_finish_reclaim_all(mp, 0);
 633
 634         /* This loop must run at least twice.
 635          * The first instance of the loop will flush
 636          * most meta data but that will generate more
 637          * meta data (typically directory updates).
 638          * Which then must be flushed and logged before
 639          * we can write the unmount record.
 640          */
 641         do {
 642                 xfs_syncsub(mp, SYNC_REMOUNT|SYNC_ATTR|SYNC_WAIT, 0, NULL);
 643                 pincount = xfs_flush_buftarg(mp->m_ddev_targp, 1);
 644                 if (!pincount) {
 645                         delay(50);
 646                         count++;
 647                 }
 648         } while (count < 2);
 649
 650         return 0;
 651 }
 652
 653 STATIC int
 654 xfs_mntupdate(
 655         bhv_desc_t                      *bdp,
 656         int                             *flags,
 657         struct xfs_mount_args           *args)
 658 {
 659         struct vfs      *vfsp = bhvtovfs(bdp);
 660         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
 661         int             error;
 662
 663         if (args->flags & XFSMNT_NOATIME)
 664                 mp->m_flags |= XFS_MOUNT_NOATIME;
 665         else
 666                 mp->m_flags &= ~XFS_MOUNT_NOATIME;
 667
 668         if ((vfsp->vfs_flag & VFS_RDONLY) &&
 669             !(*flags & MS_RDONLY)) {
 670                 vfsp->vfs_flag &= ~VFS_RDONLY;
 671
 672                 if (args->flags & XFSMNT_BARRIER)
 673                         xfs_mountfs_check_barriers(mp);
 674         }
 675
 676         if (!(vfsp->vfs_flag & VFS_RDONLY) &&
 677             (*flags & MS_RDONLY)) {
 678                 VFS_SYNC(vfsp, SYNC_FSDATA|SYNC_BDFLUSH|SYNC_ATTR, NULL, error);
 679
 680                 xfs_quiesce_fs(mp);
 681
 682                 /* Ok now write out an unmount record */
 683                 xfs_log_unmount_write(mp);
 684                 xfs_unmountfs_writesb(mp);
 685                 vfsp->vfs_flag |= VFS_RDONLY;
 686         }
 687
 688         return 0;
 689 }
 690
 691 /*
 692  * xfs_unmount_flush implements a set of flush operation on special
 693  * inodes, which are needed as a separate set of operations so that
 694  * they can be called as part of relocation process.
 695  */
 696 int
 697 xfs_unmount_flush(
 698         xfs_mount_t     *mp,            /* Mount structure we are getting
 699                                            rid of. */
 700         int             relocation)     /* Called from vfs relocation. */
 701 {
 702         xfs_inode_t     *rip = mp->m_rootip;
 703         xfs_inode_t     *rbmip;
 704         xfs_inode_t     *rsumip = NULL;
 705         vnode_t         *rvp = XFS_ITOV(rip);
 706         int             error;
 707
 708         xfs_ilock(rip, XFS_ILOCK_EXCL);
 709         xfs_iflock(rip);
 710
 711         /*
 712          * Flush out the real time inodes.
 713          */
 714         if ((rbmip = mp->m_rbmip) != NULL) {
 715                 xfs_ilock(rbmip, XFS_ILOCK_EXCL);
 716                 xfs_iflock(rbmip);
 717                 error = xfs_iflush(rbmip, XFS_IFLUSH_SYNC);
 718                 xfs_iunlock(rbmip, XFS_ILOCK_EXCL);
 719
 720                 if (error == EFSCORRUPTED)
 721                         goto fscorrupt_out;
 722
 723                 ASSERT(vn_count(XFS_ITOV(rbmip)) == 1);
 724
 725                 rsumip = mp->m_rsumip;
 726                 xfs_ilock(rsumip, XFS_ILOCK_EXCL);
 727                 xfs_iflock(rsumip);
 728                 error = xfs_iflush(rsumip, XFS_IFLUSH_SYNC);
 729                 xfs_iunlock(rsumip, XFS_ILOCK_EXCL);
 730
 731                 if (error == EFSCORRUPTED)
 732                         goto fscorrupt_out;
 733
 734                 ASSERT(vn_count(XFS_ITOV(rsumip)) == 1);
 735         }
 736
 737         /*
 738          * Synchronously flush root inode to disk
 739          */
 740         error = xfs_iflush(rip, XFS_IFLUSH_SYNC);
 741         if (error == EFSCORRUPTED)
 742                 goto fscorrupt_out2;
 743
 744         if (vn_count(rvp) != 1 && !relocation) {
 745                 xfs_iunlock(rip, XFS_ILOCK_EXCL);
 746                 return XFS_ERROR(EBUSY);
 747         }
 748
 749         /*
 750          * Release dquot that rootinode, rbmino and rsumino might be holding,
 751          * flush and purge the quota inodes.
 752          */
 753         error = XFS_QM_UNMOUNT(mp);
 754         if (error == EFSCORRUPTED)
 755                 goto fscorrupt_out2;
 756
 757         if (rbmip) {
 758                 VN_RELE(XFS_ITOV(rbmip));
 759                 VN_RELE(XFS_ITOV(rsumip));
 760         }
 761
 762         xfs_iunlock(rip, XFS_ILOCK_EXCL);
 763         return 0;
 764
 765 fscorrupt_out:
 766         xfs_ifunlock(rip);
 767
 768 fscorrupt_out2:
 769         xfs_iunlock(rip, XFS_ILOCK_EXCL);
 770
 771         return XFS_ERROR(EFSCORRUPTED);
 772 }
 773
 774 /*
 775  * xfs_root extracts the root vnode from a vfs.
 776  *
 777  * vfsp -- the vfs struct for the desired file system
 778  * vpp  -- address of the caller's vnode pointer which should be
 779  *         set to the desired fs root vnode
 780  */
 781 STATIC int
 782 xfs_root(
 783         bhv_desc_t      *bdp,
 784         vnode_t         **vpp)
 785 {
 786         vnode_t         *vp;
 787
 788         vp = XFS_ITOV((XFS_BHVTOM(bdp))->m_rootip);
 789         VN_HOLD(vp);
 790         *vpp = vp;
 791         return 0;
 792 }
 793
 794 /*
 795  * xfs_statvfs
 796  *
 797  * Fill in the statvfs structure for the given file system.  We use
 798  * the superblock lock in the mount structure to ensure a consistent
 799  * snapshot of the counters returned.
 800  */
 801 STATIC int
 802 xfs_statvfs(
 803         bhv_desc_t      *bdp,
 804         xfs_statfs_t    *statp,
 805         vnode_t         *vp)
 806 {
 807         __uint64_t      fakeinos;
 808         xfs_extlen_t    lsize;
 809         xfs_mount_t     *mp;
 810         xfs_sb_t        *sbp;
 811         unsigned long   s;
 812
 813         mp = XFS_BHVTOM(bdp);
 814         sbp = &(mp->m_sb);
 815
 816         statp->f_type = XFS_SB_MAGIC;
 817
 818         s = XFS_SB_LOCK(mp);
 819         statp->f_bsize = sbp->sb_blocksize;
 820         lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
 821         statp->f_blocks = sbp->sb_dblocks - lsize;
 822         statp->f_bfree = statp->f_bavail = sbp->sb_fdblocks;
 823         fakeinos = statp->f_bfree << sbp->sb_inopblog;
 824 #if XFS_BIG_INUMS
 825         fakeinos += mp->m_inoadd;
 826 #endif
 827         statp->f_files =
 828             MIN(sbp->sb_icount + fakeinos, (__uint64_t)XFS_MAXINUMBER);
 829         if (mp->m_maxicount)
 830 #if XFS_BIG_INUMS
 831                 if (!mp->m_inoadd)
 832 #endif
 833                         statp->f_files = min_t(typeof(statp->f_files),
 834                                                 statp->f_files,
 835                                                 mp->m_maxicount);
 836         statp->f_ffree = statp->f_files - (sbp->sb_icount - sbp->sb_ifree);
 837         XFS_SB_UNLOCK(mp, s);
 838
 839         xfs_statvfs_fsid(statp, mp);
 840         statp->f_namelen = MAXNAMELEN - 1;
 841
 842         return 0;
 843 }
 844
 845
 846 /*
 847  * xfs_sync flushes any pending I/O to file system vfsp.
 848  *
 849  * This routine is called by vfs_sync() to make sure that things make it
 850  * out to disk eventually, on sync() system calls to flush out everything,
 851  * and when the file system is unmounted.  For the vfs_sync() case, all
 852  * we really need to do is sync out the log to make all of our meta-data
 853  * updates permanent (except for timestamps).  For calls from pflushd(),
 854  * dirty pages are kept moving by calling pdflush() on the inodes
 855  * containing them.  We also flush the inodes that we can lock without
 856  * sleeping and the superblock if we can lock it without sleeping from
 857  * vfs_sync() so that items at the tail of the log are always moving out.
 858  *
 859  * Flags:
 860  *      SYNC_BDFLUSH - We're being called from vfs_sync() so we don't want
 861  *                     to sleep if we can help it.  All we really need
 862  *                     to do is ensure that the log is synced at least
 863  *                     periodically.  We also push the inodes and
 864  *                     superblock if we can lock them without sleeping
 865  *                      and they are not pinned.
 866  *      SYNC_ATTR    - We need to flush the inodes.  If SYNC_BDFLUSH is not
 867  *                     set, then we really want to lock each inode and flush
 868  *                     it.
 869  *      SYNC_WAIT    - All the flushes that take place in this call should
 870  *                     be synchronous.
 871  *      SYNC_DELWRI  - This tells us to push dirty pages associated with
 872  *                     inodes.  SYNC_WAIT and SYNC_BDFLUSH are used to
 873  *                     determine if they should be flushed sync, async, or
 874  *                     delwri.
 875  *      SYNC_CLOSE   - This flag is passed when the system is being
 876  *                     unmounted.  We should sync and invalidate everthing.
 877  *      SYNC_FSDATA  - This indicates that the caller would like to make
 878  *                     sure the superblock is safe on disk.  We can ensure
 879  *                     this by simply makeing sure the log gets flushed
 880  *                     if SYNC_BDFLUSH is set, and by actually writing it
 881  *                     out otherwise.
 882  *
 883  */
 884 /*ARGSUSED*/
 885 STATIC int
 886 xfs_sync(
 887         bhv_desc_t      *bdp,
 888         int             flags,
 889         cred_t          *credp)
 890 {
 891         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
 892
 893         if (unlikely(flags == SYNC_QUIESCE))
 894                 return xfs_quiesce_fs(mp);
 895         else
 896                 return xfs_syncsub(mp, flags, 0, NULL);
 897 }
 898
 899 /*
 900  * xfs sync routine for internal use
 901  *
 902  * This routine supports all of the flags defined for the generic VFS_SYNC
 903  * interface as explained above under xfs_sync.  In the interests of not
 904  * changing interfaces within the 6.5 family, additional internallly-
 905  * required functions are specified within a separate xflags parameter,
 906  * only available by calling this routine.
 907  *
 908  */
 909 int
 910 xfs_sync_inodes(
 911         xfs_mount_t     *mp,
 912         int             flags,
 913         int             xflags,
 914         int             *bypassed)
 915 {
 916         xfs_inode_t     *ip = NULL;
 917         xfs_inode_t     *ip_next;
 918         xfs_buf_t       *bp;
 919         vnode_t         *vp = NULL;
 920         int             error;
 921         int             last_error;
 922         uint64_t        fflag;
 923         uint            lock_flags;
 924         uint            base_lock_flags;
 925         boolean_t       mount_locked;
 926         boolean_t       vnode_refed;
 927         int             preempt;
 928         xfs_dinode_t    *dip;
 929         xfs_iptr_t      *ipointer;
 930 #ifdef DEBUG
 931         boolean_t       ipointer_in = B_FALSE;
 932
 933 #define IPOINTER_SET    ipointer_in = B_TRUE
 934 #define IPOINTER_CLR    ipointer_in = B_FALSE
 935 #else
 936 #define IPOINTER_SET
 937 #define IPOINTER_CLR
 938 #endif
 939
 940
 941 /* Insert a marker record into the inode list after inode ip. The list
 942  * must be locked when this is called. After the call the list will no
 943  * longer be locked.
 944  */
 945 #define IPOINTER_INSERT(ip, mp) { \
 946                 ASSERT(ipointer_in == B_FALSE); \
 947                 ipointer->ip_mnext = ip->i_mnext; \
 948                 ipointer->ip_mprev = ip; \
 949                 ip->i_mnext = (xfs_inode_t *)ipointer; \
 950                 ipointer->ip_mnext->i_mprev = (xfs_inode_t *)ipointer; \
 951                 preempt = 0; \
 952                 XFS_MOUNT_IUNLOCK(mp); \
 953                 mount_locked = B_FALSE; \
 954                 IPOINTER_SET; \
 955         }
 956
 957 /* Remove the marker from the inode list. If the marker was the only item
 958  * in the list then there are no remaining inodes and we should zero out
 959  * the whole list. If we are the current head of the list then move the head
 960  * past us.
 961  */
 962 #define IPOINTER_REMOVE(ip, mp) { \
 963                 ASSERT(ipointer_in == B_TRUE); \
 964                 if (ipointer->ip_mnext != (xfs_inode_t *)ipointer) { \
 965                         ip = ipointer->ip_mnext; \
 966                         ip->i_mprev = ipointer->ip_mprev; \
 967                         ipointer->ip_mprev->i_mnext = ip; \
 968                         if (mp->m_inodes == (xfs_inode_t *)ipointer) { \
 969                                 mp->m_inodes = ip; \
 970                         } \
 971                 } else { \
 972                         ASSERT(mp->m_inodes == (xfs_inode_t *)ipointer); \
 973                         mp->m_inodes = NULL; \
 974                         ip = NULL; \
 975                 } \
 976                 IPOINTER_CLR; \
 977         }
 978
 979 #define XFS_PREEMPT_MASK        0x7f
 980
 981         if (bypassed)
 982                 *bypassed = 0;
 983         if (XFS_MTOVFS(mp)->vfs_flag & VFS_RDONLY)
 984                 return 0;
 985         error = 0;
 986         last_error = 0;
 987         preempt = 0;
 988
 989         /* Allocate a reference marker */
 990         ipointer = (xfs_iptr_t *)kmem_zalloc(sizeof(xfs_iptr_t), KM_SLEEP);
 991
 992         fflag = XFS_B_ASYNC;            /* default is don't wait */
 993         if (flags & (SYNC_BDFLUSH | SYNC_DELWRI))
 994                 fflag = XFS_B_DELWRI;
 995         if (flags & SYNC_WAIT)
 996                 fflag = 0;              /* synchronous overrides all */
 997
 998         base_lock_flags = XFS_ILOCK_SHARED;
 999         if (flags & (SYNC_DELWRI | SYNC_CLOSE)) {
1000                 /*
1001                  * We need the I/O lock if we're going to call any of
1002                  * the flush/inval routines.
1003                  */
1004                 base_lock_flags |= XFS_IOLOCK_SHARED;
1005         }
1006
1007         XFS_MOUNT_ILOCK(mp);
1008
1009         ip = mp->m_inodes;
1010
1011         mount_locked = B_TRUE;
1012         vnode_refed  = B_FALSE;
1013
1014         IPOINTER_CLR;
1015
1016         do {
1017                 ASSERT(ipointer_in == B_FALSE);
1018                 ASSERT(vnode_refed == B_FALSE);
1019
1020                 lock_flags = base_lock_flags;
1021
1022                 /*
1023                  * There were no inodes in the list, just break out
1024                  * of the loop.
1025                  */
1026                 if (ip == NULL) {
1027                         break;
1028                 }
1029
1030                 /*
1031                  * We found another sync thread marker - skip it
1032                  */
1033                 if (ip->i_mount == NULL) {
1034                         ip = ip->i_mnext;
1035                         continue;
1036                 }
1037
1038                 vp = XFS_ITOV_NULL(ip);
1039
1040                 /*
1041                  * If the vnode is gone then this is being torn down,
1042                  * call reclaim if it is flushed, else let regular flush
1043                  * code deal with it later in the loop.
1044                  */
1045
1046                 if (vp == NULL) {
1047                         /* Skip ones already in reclaim */
1048                         if (ip->i_flags & XFS_IRECLAIM) {
1049                                 ip = ip->i_mnext;
1050                                 continue;
1051                         }
1052                         if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0) {
1053                                 ip = ip->i_mnext;
1054                         } else if ((xfs_ipincount(ip) == 0) &&
1055                                     xfs_iflock_nowait(ip)) {
1056                                 IPOINTER_INSERT(ip, mp);
1057
1058                                 xfs_finish_reclaim(ip, 1,
1059                                                 XFS_IFLUSH_DELWRI_ELSE_ASYNC);
1060
1061                                 XFS_MOUNT_ILOCK(mp);
1062                                 mount_locked = B_TRUE;
1063                                 IPOINTER_REMOVE(ip, mp);
1064                         } else {
1065                                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1066                                 ip = ip->i_mnext;
1067                         }
1068                         continue;
1069                 }
1070
1071                 if (VN_BAD(vp)) {
1072                         ip = ip->i_mnext;
1073                         continue;
1074                 }
1075
1076                 if (XFS_FORCED_SHUTDOWN(mp) && !(flags & SYNC_CLOSE)) {
1077                         XFS_MOUNT_IUNLOCK(mp);
1078                         kmem_free(ipointer, sizeof(xfs_iptr_t));
1079                         return 0;
1080                 }
1081
1082                 /*
1083                  * If this is just vfs_sync() or pflushd() calling
1084                  * then we can skip inodes for which it looks like
1085                  * there is nothing to do.  Since we don't have the
1086                  * inode locked this is racey, but these are periodic
1087                  * calls so it doesn't matter.  For the others we want
1088                  * to know for sure, so we at least try to lock them.
1089                  */
1090                 if (flags & SYNC_BDFLUSH) {
1091                         if (((ip->i_itemp == NULL) ||
1092                              !(ip->i_itemp->ili_format.ilf_fields &
1093                                XFS_ILOG_ALL)) &&
1094                             (ip->i_update_core == 0)) {
1095                                 ip = ip->i_mnext;
1096                                 continue;
1097                         }
1098                 }
1099
1100                 /*
1101                  * Try to lock without sleeping.  We're out of order with
1102                  * the inode list lock here, so if we fail we need to drop
1103                  * the mount lock and try again.  If we're called from
1104                  * bdflush() here, then don't bother.
1105                  *
1106                  * The inode lock here actually coordinates with the
1107                  * almost spurious inode lock in xfs_ireclaim() to prevent
1108                  * the vnode we handle here without a reference from
1109                  * being freed while we reference it.  If we lock the inode
1110                  * while it's on the mount list here, then the spurious inode
1111                  * lock in xfs_ireclaim() after the inode is pulled from
1112                  * the mount list will sleep until we release it here.
1113                  * This keeps the vnode from being freed while we reference
1114                  * it.
1115                  */
1116                 if (xfs_ilock_nowait(ip, lock_flags) == 0) {
1117                         if ((flags & SYNC_BDFLUSH) || (vp == NULL)) {
1118                                 ip = ip->i_mnext;
1119                                 continue;
1120                         }
1121
1122                         vp = vn_grab(vp);
1123                         if (vp == NULL) {
1124                                 ip = ip->i_mnext;
1125                                 continue;
1126                         }
1127
1128                         IPOINTER_INSERT(ip, mp);
1129                         xfs_ilock(ip, lock_flags);
1130
1131                         ASSERT(vp == XFS_ITOV(ip));
1132                         ASSERT(ip->i_mount == mp);
1133
1134                         vnode_refed = B_TRUE;
1135                 }
1136
1137                 /* From here on in the loop we may have a marker record
1138                  * in the inode list.
1139                  */
1140
1141                 if ((flags & SYNC_CLOSE)  && (vp != NULL)) {
1142                         /*
1143                          * This is the shutdown case.  We just need to
1144                          * flush and invalidate all the pages associated
1145                          * with the inode.  Drop the inode lock since
1146                          * we can't hold it across calls to the buffer
1147                          * cache.
1148                          *
1149                          * We don't set the VREMAPPING bit in the vnode
1150                          * here, because we don't hold the vnode lock
1151                          * exclusively.  It doesn't really matter, though,
1152                          * because we only come here when we're shutting
1153                          * down anyway.
1154                          */
1155                         xfs_iunlock(ip, XFS_ILOCK_SHARED);
1156
1157                         if (XFS_FORCED_SHUTDOWN(mp)) {
1158                                 VOP_TOSS_PAGES(vp, 0, -1, FI_REMAPF);
1159                         } else {
1160                                 VOP_FLUSHINVAL_PAGES(vp, 0, -1, FI_REMAPF);
1161                         }
1162
1163                         xfs_ilock(ip, XFS_ILOCK_SHARED);
1164
1165                 } else if ((flags & SYNC_DELWRI) && (vp != NULL)) {
1166                         if (VN_DIRTY(vp)) {
1167                                 /* We need to have dropped the lock here,
1168                                  * so insert a marker if we have not already
1169                                  * done so.
1170                                  */
1171                                 if (mount_locked) {
1172                                         IPOINTER_INSERT(ip, mp);
1173                                 }
1174
1175                                 /*
1176                                  * Drop the inode lock since we can't hold it
1177                                  * across calls to the buffer cache.
1178                                  */
1179                                 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1180                                 VOP_FLUSH_PAGES(vp, (xfs_off_t)0, -1,
1181                                                         fflag, FI_NONE, error);
1182                                 xfs_ilock(ip, XFS_ILOCK_SHARED);
1183                         }
1184
1185                 }
1186
1187                 if (flags & SYNC_BDFLUSH) {
1188                         if ((flags & SYNC_ATTR) &&
1189                             ((ip->i_update_core) ||
1190                              ((ip->i_itemp != NULL) &&
1191                               (ip->i_itemp->ili_format.ilf_fields != 0)))) {
1192
1193                                 /* Insert marker and drop lock if not already
1194                                  * done.
1195                                  */
1196                                 if (mount_locked) {
1197                                         IPOINTER_INSERT(ip, mp);
1198                                 }
1199
1200                                 /*
1201                                  * We don't want the periodic flushing of the
1202                                  * inodes by vfs_sync() to interfere with
1203                                  * I/O to the file, especially read I/O
1204                                  * where it is only the access time stamp
1205                                  * that is being flushed out.  To prevent
1206                                  * long periods where we have both inode
1207                                  * locks held shared here while reading the
1208                                  * inode's buffer in from disk, we drop the
1209                                  * inode lock while reading in the inode
1210                                  * buffer.  We have to release the buffer
1211                                  * and reacquire the inode lock so that they
1212                                  * are acquired in the proper order (inode
1213                                  * locks first).  The buffer will go at the
1214                                  * end of the lru chain, though, so we can
1215                                  * expect it to still be there when we go
1216                                  * for it again in xfs_iflush().
1217                                  */
1218                                 if ((xfs_ipincount(ip) == 0) &&
1219                                     xfs_iflock_nowait(ip)) {
1220
1221                                         xfs_ifunlock(ip);
1222                                         xfs_iunlock(ip, XFS_ILOCK_SHARED);
1223
1224                                         error = xfs_itobp(mp, NULL, ip,
1225                                                           &dip, &bp, 0);
1226                                         if (!error) {
1227                                                 xfs_buf_relse(bp);
1228                                         } else {
1229                                                 /* Bailing out, remove the
1230                                                  * marker and free it.
1231                                                  */
1232                                                 XFS_MOUNT_ILOCK(mp);
1233
1234                                                 IPOINTER_REMOVE(ip, mp);
1235
1236                                                 XFS_MOUNT_IUNLOCK(mp);
1237
1238                                                 ASSERT(!(lock_flags &
1239                                                         XFS_IOLOCK_SHARED));
1240
1241                                                 kmem_free(ipointer,
1242                                                         sizeof(xfs_iptr_t));
1243                                                 return (0);
1244                                         }
1245
1246                                         /*
1247                                          * Since we dropped the inode lock,
1248                                          * the inode may have been reclaimed.
1249                                          * Therefore, we reacquire the mount
1250                                          * lock and check to see if we were the
1251                                          * inode reclaimed. If this happened
1252                                          * then the ipointer marker will no
1253                                          * longer point back at us. In this
1254                                          * case, move ip along to the inode
1255                                          * after the marker, remove the marker
1256                                          * and continue.
1257                                          */
1258                                         XFS_MOUNT_ILOCK(mp);
1259                                         mount_locked = B_TRUE;
1260
1261                                         if (ip != ipointer->ip_mprev) {
1262                                                 IPOINTER_REMOVE(ip, mp);
1263
1264                                                 ASSERT(!vnode_refed);
1265                                                 ASSERT(!(lock_flags &
1266                                                         XFS_IOLOCK_SHARED));
1267                                                 continue;
1268                                         }
1269
1270                                         ASSERT(ip->i_mount == mp);
1271
1272                                         if (xfs_ilock_nowait(ip,
1273                                                     XFS_ILOCK_SHARED) == 0) {
1274                                                 ASSERT(ip->i_mount == mp);
1275                                                 /*
1276                                                  * We failed to reacquire
1277                                                  * the inode lock without
1278                                                  * sleeping, so just skip
1279                                                  * the inode for now.  We
1280                                                  * clear the ILOCK bit from
1281                                                  * the lock_flags so that we
1282                                                  * won't try to drop a lock
1283                                                  * we don't hold below.
1284                                                  */
1285                                                 lock_flags &= ~XFS_ILOCK_SHARED;
1286                                                 IPOINTER_REMOVE(ip_next, mp);
1287                                         } else if ((xfs_ipincount(ip) == 0) &&
1288                                                    xfs_iflock_nowait(ip)) {
1289                                                 ASSERT(ip->i_mount == mp);
1290                                                 /*
1291                                                  * Since this is vfs_sync()
1292                                                  * calling we only flush the
1293                                                  * inode out if we can lock
1294                                                  * it without sleeping and
1295                                                  * it is not pinned.  Drop
1296                                                  * the mount lock here so
1297                                                  * that we don't hold it for
1298                                                  * too long. We already have
1299                                                  * a marker in the list here.
1300                                                  */
1301                                                 XFS_MOUNT_IUNLOCK(mp);
1302                                                 mount_locked = B_FALSE;
1303                                                 error = xfs_iflush(ip,
1304                                                            XFS_IFLUSH_DELWRI);
1305                                         } else {
1306                                                 ASSERT(ip->i_mount == mp);
1307                                                 IPOINTER_REMOVE(ip_next, mp);
1308                                         }
1309                                 }
1310
1311                         }
1312
1313                 } else {
1314                         if ((flags & SYNC_ATTR) &&
1315                             ((ip->i_update_core) ||
1316                              ((ip->i_itemp != NULL) &&
1317                               (ip->i_itemp->ili_format.ilf_fields != 0)))) {
1318                                 if (mount_locked) {
1319                                         IPOINTER_INSERT(ip, mp);
1320                                 }
1321
1322                                 if (flags & SYNC_WAIT) {
1323                                         xfs_iflock(ip);
1324                                         error = xfs_iflush(ip,
1325                                                            XFS_IFLUSH_SYNC);
1326                                 } else {
1327                                         /*
1328                                          * If we can't acquire the flush
1329                                          * lock, then the inode is already
1330                                          * being flushed so don't bother
1331                                          * waiting.  If we can lock it then
1332                                          * do a delwri flush so we can
1333                                          * combine multiple inode flushes
1334                                          * in each disk write.
1335                                          */
1336                                         if (xfs_iflock_nowait(ip)) {
1337                                                 error = xfs_iflush(ip,
1338                                                            XFS_IFLUSH_DELWRI);
1339                                         }
1340                                         else if (bypassed)
1341                                                 (*bypassed)++;
1342                                 }
1343                         }
1344                 }
1345
1346                 if (lock_flags != 0) {
1347                         xfs_iunlock(ip, lock_flags);
1348                 }
1349
1350                 if (vnode_refed) {
1351                         /*
1352                          * If we had to take a reference on the vnode
1353                          * above, then wait until after we've unlocked
1354                          * the inode to release the reference.  This is
1355                          * because we can be already holding the inode
1356                          * lock when VN_RELE() calls xfs_inactive().
1357                          *
1358                          * Make sure to drop the mount lock before calling
1359                          * VN_RELE() so that we don't trip over ourselves if
1360                          * we have to go for the mount lock again in the
1361                          * inactive code.
1362                          */
1363                         if (mount_locked) {
1364                                 IPOINTER_INSERT(ip, mp);
1365                         }
1366
1367                         VN_RELE(vp);
1368
1369                         vnode_refed = B_FALSE;
1370                 }
1371
1372                 if (error) {
1373                         last_error = error;
1374                 }
1375
1376                 /*
1377                  * bail out if the filesystem is corrupted.
1378                  */
1379                 if (error == EFSCORRUPTED)  {
1380                         if (!mount_locked) {
1381                                 XFS_MOUNT_ILOCK(mp);
1382                                 IPOINTER_REMOVE(ip, mp);
1383                         }
1384                         XFS_MOUNT_IUNLOCK(mp);
1385                         ASSERT(ipointer_in == B_FALSE);
1386                         kmem_free(ipointer, sizeof(xfs_iptr_t));
1387                         return XFS_ERROR(error);
1388                 }
1389
1390                 /* Let other threads have a chance at the mount lock
1391                  * if we have looped many times without dropping the
1392                  * lock.
1393                  */
1394                 if ((++preempt & XFS_PREEMPT_MASK) == 0) {
1395                         if (mount_locked) {
1396                                 IPOINTER_INSERT(ip, mp);
1397                         }
1398                 }
1399
1400                 if (mount_locked == B_FALSE) {
1401                         XFS_MOUNT_ILOCK(mp);
1402                         mount_locked = B_TRUE;
1403                         IPOINTER_REMOVE(ip, mp);
1404                         continue;
1405                 }
1406
1407                 ASSERT(ipointer_in == B_FALSE);
1408                 ip = ip->i_mnext;
1409
1410         } while (ip != mp->m_inodes);
1411
1412         XFS_MOUNT_IUNLOCK(mp);
1413
1414         ASSERT(ipointer_in == B_FALSE);
1415
1416         kmem_free(ipointer, sizeof(xfs_iptr_t));
1417         return XFS_ERROR(last_error);
1418 }
1419
1420 /*
1421  * xfs sync routine for internal use
1422  *
1423  * This routine supports all of the flags defined for the generic VFS_SYNC
1424  * interface as explained above under xfs_sync.  In the interests of not
1425  * changing interfaces within the 6.5 family, additional internallly-
1426  * required functions are specified within a separate xflags parameter,
1427  * only available by calling this routine.
1428  *
1429  */
1430 int
1431 xfs_syncsub(
1432         xfs_mount_t     *mp,
1433         int             flags,
1434         int             xflags,
1435         int             *bypassed)
1436 {
1437         int             error = 0;
1438         int             last_error = 0;
1439         uint            log_flags = XFS_LOG_FORCE;
1440         xfs_buf_t       *bp;
1441         xfs_buf_log_item_t      *bip;
1442
1443         /*
1444          * Sync out the log.  This ensures that the log is periodically
1445          * flushed even if there is not enough activity to fill it up.
1446          */
1447         if (flags & SYNC_WAIT)
1448                 log_flags |= XFS_LOG_SYNC;
1449
1450         xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1451
1452         if (flags & (SYNC_ATTR|SYNC_DELWRI)) {
1453                 if (flags & SYNC_BDFLUSH)
1454                         xfs_finish_reclaim_all(mp, 1);
1455                 else
1456                         error = xfs_sync_inodes(mp, flags, xflags, bypassed);
1457         }
1458
1459         /*
1460          * Flushing out dirty data above probably generated more
1461          * log activity, so if this isn't vfs_sync() then flush
1462          * the log again.
1463          */
1464         if (flags & SYNC_DELWRI) {
1465                 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1466         }
1467
1468         if (flags & SYNC_FSDATA) {
1469                 /*
1470                  * If this is vfs_sync() then only sync the superblock
1471                  * if we can lock it without sleeping and it is not pinned.
1472                  */
1473                 if (flags & SYNC_BDFLUSH) {
1474                         bp = xfs_getsb(mp, XFS_BUF_TRYLOCK);
1475                         if (bp != NULL) {
1476                                 bip = XFS_BUF_FSPRIVATE(bp,xfs_buf_log_item_t*);
1477                                 if ((bip != NULL) &&
1478                                     xfs_buf_item_dirty(bip)) {
1479                                         if (!(XFS_BUF_ISPINNED(bp))) {
1480                                                 XFS_BUF_ASYNC(bp);
1481                                                 error = xfs_bwrite(mp, bp);
1482                                         } else {
1483                                                 xfs_buf_relse(bp);
1484                                         }
1485                                 } else {
1486                                         xfs_buf_relse(bp);
1487                                 }
1488                         }
1489                 } else {
1490                         bp = xfs_getsb(mp, 0);
1491                         /*
1492                          * If the buffer is pinned then push on the log so
1493                          * we won't get stuck waiting in the write for
1494                          * someone, maybe ourselves, to flush the log.
1495                          * Even though we just pushed the log above, we
1496                          * did not have the superblock buffer locked at
1497                          * that point so it can become pinned in between
1498                          * there and here.
1499                          */
1500                         if (XFS_BUF_ISPINNED(bp))
1501                                 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
1502                         if (flags & SYNC_WAIT)
1503                                 XFS_BUF_UNASYNC(bp);
1504                         else
1505                                 XFS_BUF_ASYNC(bp);
1506                         error = xfs_bwrite(mp, bp);
1507                 }
1508                 if (error) {
1509                         last_error = error;
1510                 }
1511         }
1512
1513         /*
1514          * If this is the periodic sync, then kick some entries out of
1515          * the reference cache.  This ensures that idle entries are
1516          * eventually kicked out of the cache.
1517          */
1518         if (flags & SYNC_REFCACHE) {
1519                 if (flags & SYNC_WAIT)
1520                         xfs_refcache_purge_mp(mp);
1521                 else
1522                         xfs_refcache_purge_some(mp);
1523         }
1524
1525         /*
1526          * Now check to see if the log needs a "dummy" transaction.
1527          */
1528
1529         if (!(flags & SYNC_REMOUNT) && xfs_log_need_covered(mp)) {
1530                 xfs_trans_t *tp;
1531                 xfs_inode_t *ip;
1532
1533                 /*
1534                  * Put a dummy transaction in the log to tell
1535                  * recovery that all others are OK.
1536                  */
1537                 tp = xfs_trans_alloc(mp, XFS_TRANS_DUMMY1);
1538                 if ((error = xfs_trans_reserve(tp, 0,
1539                                 XFS_ICHANGE_LOG_RES(mp),
1540                                 0, 0, 0)))  {
1541                         xfs_trans_cancel(tp, 0);
1542                         return error;
1543                 }
1544
1545                 ip = mp->m_rootip;
1546                 xfs_ilock(ip, XFS_ILOCK_EXCL);
1547
1548                 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1549                 xfs_trans_ihold(tp, ip);
1550                 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1551                 error = xfs_trans_commit(tp, 0, NULL);
1552                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1553                 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1554         }
1555
1556         /*
1557          * When shutting down, we need to insure that the AIL is pushed
1558          * to disk or the filesystem can appear corrupt from the PROM.
1559          */
1560         if ((flags & (SYNC_CLOSE|SYNC_WAIT)) == (SYNC_CLOSE|SYNC_WAIT)) {
1561                 XFS_bflush(mp->m_ddev_targp);
1562                 if (mp->m_rtdev_targp) {
1563                         XFS_bflush(mp->m_rtdev_targp);
1564                 }
1565         }
1566
1567         return XFS_ERROR(last_error);
1568 }
1569
1570 /*
1571  * xfs_vget - called by DMAPI and NFSD to get vnode from file handle
1572  */
1573 STATIC int
1574 xfs_vget(
1575         bhv_desc_t      *bdp,
1576         vnode_t         **vpp,
1577         fid_t           *fidp)
1578 {
1579         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
1580         xfs_fid_t       *xfid = (struct xfs_fid *)fidp;
1581         xfs_inode_t     *ip;
1582         int             error;
1583         xfs_ino_t       ino;
1584         unsigned int    igen;
1585
1586         /*
1587          * Invalid.  Since handles can be created in user space and passed in
1588          * via gethandle(), this is not cause for a panic.
1589          */
1590         if (xfid->xfs_fid_len != sizeof(*xfid) - sizeof(xfid->xfs_fid_len))
1591                 return XFS_ERROR(EINVAL);
1592
1593         ino  = xfid->xfs_fid_ino;
1594         igen = xfid->xfs_fid_gen;
1595
1596         /*
1597          * NFS can sometimes send requests for ino 0.  Fail them gracefully.
1598          */
1599         if (ino == 0)
1600                 return XFS_ERROR(ESTALE);
1601
1602         error = xfs_iget(mp, NULL, ino, 0, XFS_ILOCK_SHARED, &ip, 0);
1603         if (error) {
1604                 *vpp = NULL;
1605                 return error;
1606         }
1607
1608         if (ip == NULL) {
1609                 *vpp = NULL;
1610                 return XFS_ERROR(EIO);
1611         }
1612
1613         if (ip->i_d.di_mode == 0 || ip->i_d.di_gen != igen) {
1614                 xfs_iput_new(ip, XFS_ILOCK_SHARED);
1615                 *vpp = NULL;
1616                 return XFS_ERROR(ENOENT);
1617         }
1618
1619         *vpp = XFS_ITOV(ip);
1620         xfs_iunlock(ip, XFS_ILOCK_SHARED);
1621         return 0;
1622 }
1623
1624
1625 #define MNTOPT_LOGBUFS  "logbufs"       /* number of XFS log buffers */
1626 #define MNTOPT_LOGBSIZE "logbsize"      /* size of XFS log buffers */
1627 #define MNTOPT_LOGDEV   "logdev"        /* log device */
1628 #define MNTOPT_RTDEV    "rtdev"         /* realtime I/O device */
1629 #define MNTOPT_BIOSIZE  "biosize"       /* log2 of preferred buffered io size */
1630 #define MNTOPT_WSYNC    "wsync"         /* safe-mode nfs compatible mount */
1631 #define MNTOPT_INO64    "ino64"         /* force inodes into 64-bit range */
1632 #define MNTOPT_NOALIGN  "noalign"       /* turn off stripe alignment */
1633 #define MNTOPT_SWALLOC  "swalloc"       /* turn on stripe width allocation */
1634 #define MNTOPT_SUNIT    "sunit"         /* data volume stripe unit */
1635 #define MNTOPT_SWIDTH   "swidth"        /* data volume stripe width */
1636 #define MNTOPT_NOUUID   "nouuid"        /* ignore filesystem UUID */
1637 #define MNTOPT_MTPT     "mtpt"          /* filesystem mount point */
1638 #define MNTOPT_GRPID    "grpid"         /* group-ID from parent directory */
1639 #define MNTOPT_NOGRPID  "nogrpid"       /* group-ID from current process */
1640 #define MNTOPT_BSDGROUPS    "bsdgroups"    /* group-ID from parent directory */
1641 #define MNTOPT_SYSVGROUPS   "sysvgroups"   /* group-ID from current process */
1642 #define MNTOPT_ALLOCSIZE    "allocsize"    /* preferred allocation size */
1643 #define MNTOPT_IHASHSIZE    "ihashsize"    /* size of inode hash table */
1644 #define MNTOPT_NORECOVERY   "norecovery"   /* don't run XFS recovery */
1645 #define MNTOPT_BARRIER  "barrier"       /* use writer barriers for log write and
1646                                          * unwritten extent conversion */
1647 #define MNTOPT_OSYNCISOSYNC "osyncisosync" /* o_sync is REALLY o_sync */
1648 #define MNTOPT_64BITINODE   "inode64"   /* inodes can be allocated anywhere */
1649 #define MNTOPT_IKEEP    "ikeep"         /* do not free empty inode clusters */
1650 #define MNTOPT_NOIKEEP  "noikeep"       /* free empty inode clusters */
1651 #define MNTOPT_LARGEIO     "largeio"    /* report large I/O sizes in stat() */
1652 #define MNTOPT_NOLARGEIO   "nolargeio"  /* do not report large I/O sizes
1653                                          * in stat(). */
1654 #define MNTOPT_ATTR2    "attr2"         /* do use attr2 attribute format */
1655 #define MNTOPT_NOATTR2  "noattr2"       /* do not use attr2 attribute format */
1656
1657 STATIC unsigned long
1658 suffix_strtoul(const char *cp, char **endp, unsigned int base)
1659 {
1660         int     last, shift_left_factor = 0;
1661         char    *value = (char *)cp;
1662
1663         last = strlen(value) - 1;
1664         if (value[last] == 'K' || value[last] == 'k') {
1665                 shift_left_factor = 10;
1666                 value[last] = '\0';
1667         }
1668         if (value[last] == 'M' || value[last] == 'm') {
1669                 shift_left_factor = 20;
1670                 value[last] = '\0';
1671         }
1672         if (value[last] == 'G' || value[last] == 'g') {
1673                 shift_left_factor = 30;
1674                 value[last] = '\0';
1675         }
1676
1677         return simple_strtoul(cp, endp, base) << shift_left_factor;
1678 }
1679
1680 STATIC int
1681 xfs_parseargs(
1682         struct bhv_desc         *bhv,
1683         char                    *options,
1684         struct xfs_mount_args   *args,
1685         int                     update)
1686 {
1687         struct vfs              *vfsp = bhvtovfs(bhv);
1688         char                    *this_char, *value, *eov;
1689         int                     dsunit, dswidth, vol_dsunit, vol_dswidth;
1690         int                     iosize;
1691
1692         args->flags |= XFSMNT_COMPAT_IOSIZE;
1693 #if 0   /* XXX: off by default, until some remaining issues ironed out */
1694         args->flags |= XFSMNT_IDELETE; /* default to on */
1695 #endif
1696
1697         if (!options)
1698                 return 0;
1699
1700         iosize = dsunit = dswidth = vol_dsunit = vol_dswidth = 0;
1701
1702         while ((this_char = strsep(&options, ",")) != NULL) {
1703                 if (!*this_char)
1704                         continue;
1705                 if ((value = strchr(this_char, '=')) != NULL)
1706                         *value++ = 0;
1707
1708                 if (!strcmp(this_char, MNTOPT_LOGBUFS)) {
1709                         if (!value || !*value) {
1710                                 printk("XFS: %s option requires an argument\n",
1711                                         this_char);
1712                                 return EINVAL;
1713                         }
1714                         args->logbufs = simple_strtoul(value, &eov, 10);
1715                 } else if (!strcmp(this_char, MNTOPT_LOGBSIZE)) {
1716                         if (!value || !*value) {
1717                                 printk("XFS: %s option requires an argument\n",
1718                                         this_char);
1719                                 return EINVAL;
1720                         }
1721                         args->logbufsize = suffix_strtoul(value, &eov, 10);
1722                 } else if (!strcmp(this_char, MNTOPT_LOGDEV)) {
1723                         if (!value || !*value) {
1724                                 printk("XFS: %s option requires an argument\n",
1725                                         this_char);
1726                                 return EINVAL;
1727                         }
1728                         strncpy(args->logname, value, MAXNAMELEN);
1729                 } else if (!strcmp(this_char, MNTOPT_MTPT)) {
1730                         if (!value || !*value) {
1731                                 printk("XFS: %s option requires an argument\n",
1732                                         this_char);
1733                                 return EINVAL;
1734                         }
1735                         strncpy(args->mtpt, value, MAXNAMELEN);
1736                 } else if (!strcmp(this_char, MNTOPT_RTDEV)) {
1737                         if (!value || !*value) {
1738                                 printk("XFS: %s option requires an argument\n",
1739                                         this_char);
1740                                 return EINVAL;
1741                         }
1742                         strncpy(args->rtname, value, MAXNAMELEN);
1743                 } else if (!strcmp(this_char, MNTOPT_BIOSIZE)) {
1744                         if (!value || !*value) {
1745                                 printk("XFS: %s option requires an argument\n",
1746                                         this_char);
1747                                 return EINVAL;
1748                         }
1749                         iosize = simple_strtoul(value, &eov, 10);
1750                         args->flags |= XFSMNT_IOSIZE;
1751                         args->iosizelog = (uint8_t) iosize;
1752                 } else if (!strcmp(this_char, MNTOPT_ALLOCSIZE)) {
1753                         if (!value || !*value) {
1754                                 printk("XFS: %s option requires an argument\n",
1755                                         this_char);
1756                                 return EINVAL;
1757                         }
1758                         iosize = suffix_strtoul(value, &eov, 10);
1759                         args->flags |= XFSMNT_IOSIZE;
1760                         args->iosizelog = ffs(iosize) - 1;
1761                 } else if (!strcmp(this_char, MNTOPT_IHASHSIZE)) {
1762                         if (!value || !*value) {
1763                                 printk("XFS: %s option requires an argument\n",
1764                                         this_char);
1765                                 return EINVAL;
1766                         }
1767                         args->flags |= XFSMNT_IHASHSIZE;
1768                         args->ihashsize = simple_strtoul(value, &eov, 10);
1769                 } else if (!strcmp(this_char, MNTOPT_GRPID) ||
1770                            !strcmp(this_char, MNTOPT_BSDGROUPS)) {
1771                         vfsp->vfs_flag |= VFS_GRPID;
1772                 } else if (!strcmp(this_char, MNTOPT_NOGRPID) ||
1773                            !strcmp(this_char, MNTOPT_SYSVGROUPS)) {
1774                         vfsp->vfs_flag &= ~VFS_GRPID;
1775                 } else if (!strcmp(this_char, MNTOPT_WSYNC)) {
1776                         args->flags |= XFSMNT_WSYNC;
1777                 } else if (!strcmp(this_char, MNTOPT_OSYNCISOSYNC)) {
1778                         args->flags |= XFSMNT_OSYNCISOSYNC;
1779                 } else if (!strcmp(this_char, MNTOPT_NORECOVERY)) {
1780                         args->flags |= XFSMNT_NORECOVERY;
1781                 } else if (!strcmp(this_char, MNTOPT_INO64)) {
1782                         args->flags |= XFSMNT_INO64;
1783 #if !XFS_BIG_INUMS
1784                         printk("XFS: %s option not allowed on this system\n",
1785                                 this_char);
1786                         return EINVAL;
1787 #endif
1788                 } else if (!strcmp(this_char, MNTOPT_NOALIGN)) {
1789                         args->flags |= XFSMNT_NOALIGN;
1790                 } else if (!strcmp(this_char, MNTOPT_SWALLOC)) {
1791                         args->flags |= XFSMNT_SWALLOC;
1792                 } else if (!strcmp(this_char, MNTOPT_SUNIT)) {
1793                         if (!value || !*value) {
1794                                 printk("XFS: %s option requires an argument\n",
1795                                         this_char);
1796                                 return EINVAL;
1797                         }
1798                         dsunit = simple_strtoul(value, &eov, 10);
1799                 } else if (!strcmp(this_char, MNTOPT_SWIDTH)) {
1800                         if (!value || !*value) {
1801                                 printk("XFS: %s option requires an argument\n",
1802                                         this_char);
1803                                 return EINVAL;
1804                         }
1805                         dswidth = simple_strtoul(value, &eov, 10);
1806                 } else if (!strcmp(this_char, MNTOPT_64BITINODE)) {
1807                         args->flags &= ~XFSMNT_32BITINODES;
1808 #if !XFS_BIG_INUMS
1809                         printk("XFS: %s option not allowed on this system\n",
1810                                 this_char);
1811                         return EINVAL;
1812 #endif
1813                 } else if (!strcmp(this_char, MNTOPT_NOUUID)) {
1814                         args->flags |= XFSMNT_NOUUID;
1815                 } else if (!strcmp(this_char, MNTOPT_BARRIER)) {
1816                         args->flags |= XFSMNT_BARRIER;
1817                 } else if (!strcmp(this_char, MNTOPT_IKEEP)) {
1818                         args->flags &= ~XFSMNT_IDELETE;
1819                 } else if (!strcmp(this_char, MNTOPT_NOIKEEP)) {
1820                         args->flags |= XFSMNT_IDELETE;
1821                 } else if (!strcmp(this_char, MNTOPT_LARGEIO)) {
1822                         args->flags &= ~XFSMNT_COMPAT_IOSIZE;
1823                 } else if (!strcmp(this_char, MNTOPT_NOLARGEIO)) {
1824                         args->flags |= XFSMNT_COMPAT_IOSIZE;
1825                 } else if (!strcmp(this_char, MNTOPT_ATTR2)) {
1826                         args->flags &= ~XFSMNT_COMPAT_ATTR;
1827                 } else if (!strcmp(this_char, MNTOPT_NOATTR2)) {
1828                         args->flags |= XFSMNT_COMPAT_ATTR;
1829                 } else if (!strcmp(this_char, "osyncisdsync")) {
1830                         /* no-op, this is now the default */
1831 printk("XFS: osyncisdsync is now the default, option is deprecated.\n");
1832                 } else if (!strcmp(this_char, "irixsgid")) {
1833 printk("XFS: irixsgid is now a sysctl(2) variable, option is deprecated.\n");
1834                 } else {
1835                         printk("XFS: unknown mount option [%s].\n", this_char);
1836                         return EINVAL;
1837                 }
1838         }
1839
1840         if (args->flags & XFSMNT_NORECOVERY) {
1841                 if ((vfsp->vfs_flag & VFS_RDONLY) == 0) {
1842                         printk("XFS: no-recovery mounts must be read-only.\n");
1843                         return EINVAL;
1844                 }
1845         }
1846
1847         if ((args->flags & XFSMNT_NOALIGN) && (dsunit || dswidth)) {
1848                 printk(
1849         "XFS: sunit and swidth options incompatible with the noalign option\n");
1850                 return EINVAL;
1851         }
1852
1853         if ((dsunit && !dswidth) || (!dsunit && dswidth)) {
1854                 printk("XFS: sunit and swidth must be specified together\n");
1855                 return EINVAL;
1856         }
1857
1858         if (dsunit && (dswidth % dsunit != 0)) {
1859                 printk(
1860         "XFS: stripe width (%d) must be a multiple of the stripe unit (%d)\n",
1861                         dswidth, dsunit);
1862                 return EINVAL;
1863         }
1864
1865         if ((args->flags & XFSMNT_NOALIGN) != XFSMNT_NOALIGN) {
1866                 if (dsunit) {
1867                         args->sunit = dsunit;
1868                         args->flags |= XFSMNT_RETERR;
1869                 } else {
1870                         args->sunit = vol_dsunit;
1871                 }
1872                 dswidth ? (args->swidth = dswidth) :
1873                           (args->swidth = vol_dswidth);
1874         } else {
1875                 args->sunit = args->swidth = 0;
1876         }
1877
1878         return 0;
1879 }
1880
1881 STATIC int
1882 xfs_showargs(
1883         struct bhv_desc         *bhv,
1884         struct seq_file         *m)
1885 {
1886         static struct proc_xfs_info {
1887                 int     flag;
1888                 char    *str;
1889         } xfs_info[] = {
1890                 /* the few simple ones we can get from the mount struct */
1891                 { XFS_MOUNT_WSYNC,              "," MNTOPT_WSYNC },
1892                 { XFS_MOUNT_INO64,              "," MNTOPT_INO64 },
1893                 { XFS_MOUNT_NOALIGN,            "," MNTOPT_NOALIGN },
1894                 { XFS_MOUNT_SWALLOC,            "," MNTOPT_SWALLOC },
1895                 { XFS_MOUNT_NOUUID,             "," MNTOPT_NOUUID },
1896                 { XFS_MOUNT_NORECOVERY,         "," MNTOPT_NORECOVERY },
1897                 { XFS_MOUNT_OSYNCISOSYNC,       "," MNTOPT_OSYNCISOSYNC },
1898                 { XFS_MOUNT_BARRIER,            "," MNTOPT_BARRIER },
1899                 { XFS_MOUNT_IDELETE,            "," MNTOPT_NOIKEEP },
1900                 { 0, NULL }
1901         };
1902         struct proc_xfs_info    *xfs_infop;
1903         struct xfs_mount        *mp = XFS_BHVTOM(bhv);
1904         struct vfs              *vfsp = XFS_MTOVFS(mp);
1905
1906         for (xfs_infop = xfs_info; xfs_infop->flag; xfs_infop++) {
1907                 if (mp->m_flags & xfs_infop->flag)
1908                         seq_puts(m, xfs_infop->str);
1909         }
1910
1911         if (mp->m_flags & XFS_MOUNT_IHASHSIZE)
1912                 seq_printf(m, "," MNTOPT_IHASHSIZE "=%d", mp->m_ihsize);
1913
1914         if (mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)
1915                 seq_printf(m, "," MNTOPT_ALLOCSIZE "=%d", 1<<mp->m_writeio_log);
1916
1917         if (mp->m_logbufs > 0)
1918                 seq_printf(m, "," MNTOPT_LOGBUFS "=%d", mp->m_logbufs);
1919
1920         if (mp->m_logbsize > 0)
1921                 seq_printf(m, "," MNTOPT_LOGBSIZE "=%d", mp->m_logbsize);
1922
1923         if (mp->m_logname)
1924                 seq_printf(m, "," MNTOPT_LOGDEV "=%s", mp->m_logname);
1925
1926         if (mp->m_rtname)
1927                 seq_printf(m, "," MNTOPT_RTDEV "=%s", mp->m_rtname);
1928
1929         if (mp->m_dalign > 0)
1930                 seq_printf(m, "," MNTOPT_SUNIT "=%d",
1931                                 (int)XFS_FSB_TO_BB(mp, mp->m_dalign));
1932
1933         if (mp->m_swidth > 0)
1934                 seq_printf(m, "," MNTOPT_SWIDTH "=%d",
1935                                 (int)XFS_FSB_TO_BB(mp, mp->m_swidth));
1936
1937         if (!(mp->m_flags & XFS_MOUNT_32BITINOOPT))
1938                 seq_printf(m, "," MNTOPT_64BITINODE);
1939
1940         if (vfsp->vfs_flag & VFS_GRPID)
1941                 seq_printf(m, "," MNTOPT_GRPID);
1942
1943         return 0;
1944 }
1945
1946 STATIC void
1947 xfs_freeze(
1948         bhv_desc_t      *bdp)
1949 {
1950         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
1951
1952         while (atomic_read(&mp->m_active_trans) > 0)
1953                 delay(100);
1954
1955         /* Push the superblock and write an unmount record */
1956         xfs_log_unmount_write(mp);
1957         xfs_unmountfs_writesb(mp);
1958 }
1959
1960
1961 vfsops_t xfs_vfsops = {
1962         BHV_IDENTITY_INIT(VFS_BHV_XFS,VFS_POSITION_XFS),
1963         .vfs_parseargs          = xfs_parseargs,
1964         .vfs_showargs           = xfs_showargs,
1965         .vfs_mount              = xfs_mount,
1966         .vfs_unmount            = xfs_unmount,
1967         .vfs_mntupdate          = xfs_mntupdate,
1968         .vfs_root               = xfs_root,
1969         .vfs_statvfs            = xfs_statvfs,
1970         .vfs_sync               = xfs_sync,
1971         .vfs_vget               = xfs_vget,
1972         .vfs_dmapiops           = (vfs_dmapiops_t)fs_nosys,
1973         .vfs_quotactl           = (vfs_quotactl_t)fs_nosys,
1974         .vfs_init_vnode         = xfs_initialize_vnode,
1975         .vfs_force_shutdown     = xfs_do_force_shutdown,
1976         .vfs_freeze             = xfs_freeze,
1977 };