err.no Git - linux-2.6/blob - fs/xfs/xfs_vfsops.c

   1 /*
   2  * XFS filesystem operations.
   3  *
   4  * Copyright (c) 2000-2005 Silicon Graphics, Inc.  All Rights Reserved.
   5  *
   6  * This program is free software; you can redistribute it and/or modify it
   7  * under the terms of version 2 of the GNU General Public License as
   8  * published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it would be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
  13  *
  14  * Further, this software is distributed without any warranty that it is
  15  * free of the rightful claim of any third person regarding infringement
  16  * or the like.  Any license provided herein, whether implied or
  17  * otherwise, applies only to this software file.  Patent licenses, if
  18  * any, provided herein do not apply to combinations of this program with
  19  * other software, or any other product whatsoever.
  20  *
  21  * You should have received a copy of the GNU General Public License along
  22  * with this program; if not, write the Free Software Foundation, Inc., 59
  23  * Temple Place - Suite 330, Boston MA 02111-1307, USA.
  24  *
  25  * Contact information: Silicon Graphics, Inc., 1600 Amphitheatre Pkwy,
  26  * Mountain View, CA  94043, or:
  27  *
  28  * http://www.sgi.com
  29  *
  30  * For further information regarding this notice, see:
  31  *
  32  * http://oss.sgi.com/projects/GenInfo/SGIGPLNoticeExplan/
  33  */
  34
  35 #include "xfs.h"
  36 #include "xfs_macros.h"
  37 #include "xfs_types.h"
  38 #include "xfs_inum.h"
  39 #include "xfs_log.h"
  40 #include "xfs_trans.h"
  41 #include "xfs_sb.h"
  42 #include "xfs_dir.h"
  43 #include "xfs_dir2.h"
  44 #include "xfs_dmapi.h"
  45 #include "xfs_mount.h"
  46 #include "xfs_bmap_btree.h"
  47 #include "xfs_ialloc_btree.h"
  48 #include "xfs_alloc_btree.h"
  49 #include "xfs_btree.h"
  50 #include "xfs_alloc.h"
  51 #include "xfs_ialloc.h"
  52 #include "xfs_attr_sf.h"
  53 #include "xfs_dir_sf.h"
  54 #include "xfs_dir2_sf.h"
  55 #include "xfs_dinode.h"
  56 #include "xfs_inode_item.h"
  57 #include "xfs_inode.h"
  58 #include "xfs_ag.h"
  59 #include "xfs_error.h"
  60 #include "xfs_bmap.h"
  61 #include "xfs_da_btree.h"
  62 #include "xfs_rw.h"
  63 #include "xfs_refcache.h"
  64 #include "xfs_buf_item.h"
  65 #include "xfs_extfree_item.h"
  66 #include "xfs_quota.h"
  67 #include "xfs_dir2_trace.h"
  68 #include "xfs_acl.h"
  69 #include "xfs_attr.h"
  70 #include "xfs_clnt.h"
  71 #include "xfs_log_priv.h"
  72
  73 STATIC int xfs_sync(bhv_desc_t *, int, cred_t *);
  74
  75 int
  76 xfs_init(void)
  77 {
  78         extern kmem_zone_t      *xfs_bmap_free_item_zone;
  79         extern kmem_zone_t      *xfs_btree_cur_zone;
  80         extern kmem_zone_t      *xfs_trans_zone;
  81         extern kmem_zone_t      *xfs_buf_item_zone;
  82         extern kmem_zone_t      *xfs_dabuf_zone;
  83 #ifdef XFS_DABUF_DEBUG
  84         extern lock_t           xfs_dabuf_global_lock;
  85         spinlock_init(&xfs_dabuf_global_lock, "xfsda");
  86 #endif
  87
  88         /*
  89          * Initialize all of the zone allocators we use.
  90          */
  91         xfs_bmap_free_item_zone = kmem_zone_init(sizeof(xfs_bmap_free_item_t),
  92                                                  "xfs_bmap_free_item");
  93         xfs_btree_cur_zone = kmem_zone_init(sizeof(xfs_btree_cur_t),
  94                                             "xfs_btree_cur");
  95         xfs_inode_zone = kmem_zone_init(sizeof(xfs_inode_t), "xfs_inode");
  96         xfs_trans_zone = kmem_zone_init(sizeof(xfs_trans_t), "xfs_trans");
  97         xfs_da_state_zone =
  98                 kmem_zone_init(sizeof(xfs_da_state_t), "xfs_da_state");
  99         xfs_dabuf_zone = kmem_zone_init(sizeof(xfs_dabuf_t), "xfs_dabuf");
 100
 101         /*
 102          * The size of the zone allocated buf log item is the maximum
 103          * size possible under XFS.  This wastes a little bit of memory,
 104          * but it is much faster.
 105          */
 106         xfs_buf_item_zone =
 107                 kmem_zone_init((sizeof(xfs_buf_log_item_t) +
 108                                 (((XFS_MAX_BLOCKSIZE / XFS_BLI_CHUNK) /
 109                                   NBWORD) * sizeof(int))),
 110                                "xfs_buf_item");
 111         xfs_efd_zone = kmem_zone_init((sizeof(xfs_efd_log_item_t) +
 112                                        ((XFS_EFD_MAX_FAST_EXTENTS - 1) * sizeof(xfs_extent_t))),
 113                                       "xfs_efd_item");
 114         xfs_efi_zone = kmem_zone_init((sizeof(xfs_efi_log_item_t) +
 115                                        ((XFS_EFI_MAX_FAST_EXTENTS - 1) * sizeof(xfs_extent_t))),
 116                                       "xfs_efi_item");
 117         xfs_ifork_zone = kmem_zone_init(sizeof(xfs_ifork_t), "xfs_ifork");
 118         xfs_ili_zone = kmem_zone_init(sizeof(xfs_inode_log_item_t), "xfs_ili");
 119         xfs_chashlist_zone = kmem_zone_init(sizeof(xfs_chashlist_t),
 120                                             "xfs_chashlist");
 121         xfs_acl_zone_init(xfs_acl_zone, "xfs_acl");
 122
 123         /*
 124          * Allocate global trace buffers.
 125          */
 126 #ifdef XFS_ALLOC_TRACE
 127         xfs_alloc_trace_buf = ktrace_alloc(XFS_ALLOC_TRACE_SIZE, KM_SLEEP);
 128 #endif
 129 #ifdef XFS_BMAP_TRACE
 130         xfs_bmap_trace_buf = ktrace_alloc(XFS_BMAP_TRACE_SIZE, KM_SLEEP);
 131 #endif
 132 #ifdef XFS_BMBT_TRACE
 133         xfs_bmbt_trace_buf = ktrace_alloc(XFS_BMBT_TRACE_SIZE, KM_SLEEP);
 134 #endif
 135 #ifdef XFS_DIR_TRACE
 136         xfs_dir_trace_buf = ktrace_alloc(XFS_DIR_TRACE_SIZE, KM_SLEEP);
 137 #endif
 138 #ifdef XFS_ATTR_TRACE
 139         xfs_attr_trace_buf = ktrace_alloc(XFS_ATTR_TRACE_SIZE, KM_SLEEP);
 140 #endif
 141 #ifdef XFS_DIR2_TRACE
 142         xfs_dir2_trace_buf = ktrace_alloc(XFS_DIR2_GTRACE_SIZE, KM_SLEEP);
 143 #endif
 144
 145         xfs_dir_startup();
 146
 147 #if (defined(DEBUG) || defined(INDUCE_IO_ERROR))
 148         xfs_error_test_init();
 149 #endif /* DEBUG || INDUCE_IO_ERROR */
 150
 151         xfs_init_procfs();
 152         xfs_sysctl_register();
 153         return 0;
 154 }
 155
 156 void
 157 xfs_cleanup(void)
 158 {
 159         extern kmem_zone_t      *xfs_bmap_free_item_zone;
 160         extern kmem_zone_t      *xfs_btree_cur_zone;
 161         extern kmem_zone_t      *xfs_inode_zone;
 162         extern kmem_zone_t      *xfs_trans_zone;
 163         extern kmem_zone_t      *xfs_da_state_zone;
 164         extern kmem_zone_t      *xfs_dabuf_zone;
 165         extern kmem_zone_t      *xfs_efd_zone;
 166         extern kmem_zone_t      *xfs_efi_zone;
 167         extern kmem_zone_t      *xfs_buf_item_zone;
 168         extern kmem_zone_t      *xfs_chashlist_zone;
 169
 170         xfs_cleanup_procfs();
 171         xfs_sysctl_unregister();
 172         xfs_refcache_destroy();
 173         xfs_acl_zone_destroy(xfs_acl_zone);
 174
 175 #ifdef XFS_DIR2_TRACE
 176         ktrace_free(xfs_dir2_trace_buf);
 177 #endif
 178 #ifdef XFS_ATTR_TRACE
 179         ktrace_free(xfs_attr_trace_buf);
 180 #endif
 181 #ifdef XFS_DIR_TRACE
 182         ktrace_free(xfs_dir_trace_buf);
 183 #endif
 184 #ifdef XFS_BMBT_TRACE
 185         ktrace_free(xfs_bmbt_trace_buf);
 186 #endif
 187 #ifdef XFS_BMAP_TRACE
 188         ktrace_free(xfs_bmap_trace_buf);
 189 #endif
 190 #ifdef XFS_ALLOC_TRACE
 191         ktrace_free(xfs_alloc_trace_buf);
 192 #endif
 193
 194         kmem_cache_destroy(xfs_bmap_free_item_zone);
 195         kmem_cache_destroy(xfs_btree_cur_zone);
 196         kmem_cache_destroy(xfs_inode_zone);
 197         kmem_cache_destroy(xfs_trans_zone);
 198         kmem_cache_destroy(xfs_da_state_zone);
 199         kmem_cache_destroy(xfs_dabuf_zone);
 200         kmem_cache_destroy(xfs_buf_item_zone);
 201         kmem_cache_destroy(xfs_efd_zone);
 202         kmem_cache_destroy(xfs_efi_zone);
 203         kmem_cache_destroy(xfs_ifork_zone);
 204         kmem_cache_destroy(xfs_ili_zone);
 205         kmem_cache_destroy(xfs_chashlist_zone);
 206 }
 207
 208 /*
 209  * xfs_start_flags
 210  *
 211  * This function fills in xfs_mount_t fields based on mount args.
 212  * Note: the superblock has _not_ yet been read in.
 213  */
 214 STATIC int
 215 xfs_start_flags(
 216         struct vfs              *vfs,
 217         struct xfs_mount_args   *ap,
 218         struct xfs_mount        *mp)
 219 {
 220         /* Values are in BBs */
 221         if ((ap->flags & XFSMNT_NOALIGN) != XFSMNT_NOALIGN) {
 222                 /*
 223                  * At this point the superblock has not been read
 224                  * in, therefore we do not know the block size.
 225                  * Before the mount call ends we will convert
 226                  * these to FSBs.
 227                  */
 228                 mp->m_dalign = ap->sunit;
 229                 mp->m_swidth = ap->swidth;
 230         }
 231
 232         if (ap->logbufs != -1 &&
 233 #if defined(DEBUG) || defined(XLOG_NOLOG)
 234             ap->logbufs != 0 &&
 235 #endif
 236             (ap->logbufs < XLOG_MIN_ICLOGS ||
 237              ap->logbufs > XLOG_MAX_ICLOGS)) {
 238                 cmn_err(CE_WARN,
 239                         "XFS: invalid logbufs value: %d [not %d-%d]",
 240                         ap->logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
 241                 return XFS_ERROR(EINVAL);
 242         }
 243         mp->m_logbufs = ap->logbufs;
 244         if (ap->logbufsize != -1 &&
 245             ap->logbufsize != 16 * 1024 &&
 246             ap->logbufsize != 32 * 1024 &&
 247             ap->logbufsize != 64 * 1024 &&
 248             ap->logbufsize != 128 * 1024 &&
 249             ap->logbufsize != 256 * 1024) {
 250                 cmn_err(CE_WARN,
 251         "XFS: invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
 252                         ap->logbufsize);
 253                 return XFS_ERROR(EINVAL);
 254         }
 255         mp->m_ihsize = ap->ihashsize;
 256         mp->m_logbsize = ap->logbufsize;
 257         mp->m_fsname_len = strlen(ap->fsname) + 1;
 258         mp->m_fsname = kmem_alloc(mp->m_fsname_len, KM_SLEEP);
 259         strcpy(mp->m_fsname, ap->fsname);
 260         if (ap->rtname[0]) {
 261                 mp->m_rtname = kmem_alloc(strlen(ap->rtname) + 1, KM_SLEEP);
 262                 strcpy(mp->m_rtname, ap->rtname);
 263         }
 264         if (ap->logname[0]) {
 265                 mp->m_logname = kmem_alloc(strlen(ap->logname) + 1, KM_SLEEP);
 266                 strcpy(mp->m_logname, ap->logname);
 267         }
 268
 269         if (ap->flags & XFSMNT_WSYNC)
 270                 mp->m_flags |= XFS_MOUNT_WSYNC;
 271 #if XFS_BIG_INUMS
 272         if (ap->flags & XFSMNT_INO64) {
 273                 mp->m_flags |= XFS_MOUNT_INO64;
 274                 mp->m_inoadd = XFS_INO64_OFFSET;
 275         }
 276 #endif
 277         if (ap->flags & XFSMNT_NOATIME)
 278                 mp->m_flags |= XFS_MOUNT_NOATIME;
 279         if (ap->flags & XFSMNT_RETERR)
 280                 mp->m_flags |= XFS_MOUNT_RETERR;
 281         if (ap->flags & XFSMNT_NOALIGN)
 282                 mp->m_flags |= XFS_MOUNT_NOALIGN;
 283         if (ap->flags & XFSMNT_SWALLOC)
 284                 mp->m_flags |= XFS_MOUNT_SWALLOC;
 285         if (ap->flags & XFSMNT_OSYNCISOSYNC)
 286                 mp->m_flags |= XFS_MOUNT_OSYNCISOSYNC;
 287         if (ap->flags & XFSMNT_32BITINODES)
 288                 mp->m_flags |= (XFS_MOUNT_32BITINODES | XFS_MOUNT_32BITINOOPT);
 289
 290         if (ap->flags & XFSMNT_IOSIZE) {
 291                 if (ap->iosizelog > XFS_MAX_IO_LOG ||
 292                     ap->iosizelog < XFS_MIN_IO_LOG) {
 293                         cmn_err(CE_WARN,
 294                 "XFS: invalid log iosize: %d [not %d-%d]",
 295                                 ap->iosizelog, XFS_MIN_IO_LOG,
 296                                 XFS_MAX_IO_LOG);
 297                         return XFS_ERROR(EINVAL);
 298                 }
 299
 300                 mp->m_flags |= XFS_MOUNT_DFLT_IOSIZE;
 301                 mp->m_readio_log = mp->m_writeio_log = ap->iosizelog;
 302         }
 303
 304         if (ap->flags & XFSMNT_IHASHSIZE)
 305                 mp->m_flags |= XFS_MOUNT_IHASHSIZE;
 306         if (ap->flags & XFSMNT_IDELETE)
 307                 mp->m_flags |= XFS_MOUNT_IDELETE;
 308         if (ap->flags & XFSMNT_DIRSYNC)
 309                 mp->m_flags |= XFS_MOUNT_DIRSYNC;
 310         if (ap->flags & XFSMNT_COMPAT_IOSIZE)
 311                 mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE;
 312         if (ap->flags & XFSMNT_COMPAT_ATTR)
 313                 mp->m_flags |= XFS_MOUNT_COMPAT_ATTR;
 314
 315         /*
 316          * no recovery flag requires a read-only mount
 317          */
 318         if (ap->flags & XFSMNT_NORECOVERY) {
 319                 if (!(vfs->vfs_flag & VFS_RDONLY)) {
 320                         cmn_err(CE_WARN,
 321         "XFS: tried to mount a FS read-write without recovery!");
 322                         return XFS_ERROR(EINVAL);
 323                 }
 324                 mp->m_flags |= XFS_MOUNT_NORECOVERY;
 325         }
 326
 327         if (ap->flags & XFSMNT_NOUUID)
 328                 mp->m_flags |= XFS_MOUNT_NOUUID;
 329         if (ap->flags & XFSMNT_BARRIER)
 330                 mp->m_flags |= XFS_MOUNT_BARRIER;
 331
 332         return 0;
 333 }
 334
 335 /*
 336  * This function fills in xfs_mount_t fields based on mount args.
 337  * Note: the superblock _has_ now been read in.
 338  */
 339 STATIC int
 340 xfs_finish_flags(
 341         struct vfs              *vfs,
 342         struct xfs_mount_args   *ap,
 343         struct xfs_mount        *mp)
 344 {
 345         int                     ronly = (vfs->vfs_flag & VFS_RDONLY);
 346
 347         /* Fail a mount where the logbuf is smaller then the log stripe */
 348         if (XFS_SB_VERSION_HASLOGV2(&mp->m_sb)) {
 349                 if ((ap->logbufsize == -1) &&
 350                     (mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE)) {
 351                         mp->m_logbsize = mp->m_sb.sb_logsunit;
 352                 } else if (ap->logbufsize < mp->m_sb.sb_logsunit) {
 353                         cmn_err(CE_WARN,
 354         "XFS: logbuf size must be greater than or equal to log stripe size");
 355                         return XFS_ERROR(EINVAL);
 356                 }
 357         } else {
 358                 /* Fail a mount if the logbuf is larger than 32K */
 359                 if (ap->logbufsize > XLOG_BIG_RECORD_BSIZE) {
 360                         cmn_err(CE_WARN,
 361         "XFS: logbuf size for version 1 logs must be 16K or 32K");
 362                         return XFS_ERROR(EINVAL);
 363                 }
 364         }
 365
 366         /*
 367          * prohibit r/w mounts of read-only filesystems
 368          */
 369         if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !ronly) {
 370                 cmn_err(CE_WARN,
 371         "XFS: cannot mount a read-only filesystem as read-write");
 372                 return XFS_ERROR(EROFS);
 373         }
 374
 375         /*
 376          * check for shared mount.
 377          */
 378         if (ap->flags & XFSMNT_SHARED) {
 379                 if (!XFS_SB_VERSION_HASSHARED(&mp->m_sb))
 380                         return XFS_ERROR(EINVAL);
 381
 382                 /*
 383                  * For IRIX 6.5, shared mounts must have the shared
 384                  * version bit set, have the persistent readonly
 385                  * field set, must be version 0 and can only be mounted
 386                  * read-only.
 387                  */
 388                 if (!ronly || !(mp->m_sb.sb_flags & XFS_SBF_READONLY) ||
 389                      (mp->m_sb.sb_shared_vn != 0))
 390                         return XFS_ERROR(EINVAL);
 391
 392                 mp->m_flags |= XFS_MOUNT_SHARED;
 393
 394                 /*
 395                  * Shared XFS V0 can't deal with DMI.  Return EINVAL.
 396                  */
 397                 if (mp->m_sb.sb_shared_vn == 0 && (ap->flags & XFSMNT_DMAPI))
 398                         return XFS_ERROR(EINVAL);
 399         }
 400
 401         return 0;
 402 }
 403
 404 /*
 405  * xfs_mount
 406  *
 407  * The file system configurations are:
 408  *      (1) device (partition) with data and internal log
 409  *      (2) logical volume with data and log subvolumes.
 410  *      (3) logical volume with data, log, and realtime subvolumes.
 411  *
 412  * We only have to handle opening the log and realtime volumes here if
 413  * they are present.  The data subvolume has already been opened by
 414  * get_sb_bdev() and is stored in vfsp->vfs_super->s_bdev.
 415  */
 416 STATIC int
 417 xfs_mount(
 418         struct bhv_desc         *bhvp,
 419         struct xfs_mount_args   *args,
 420         cred_t                  *credp)
 421 {
 422         struct vfs              *vfsp = bhvtovfs(bhvp);
 423         struct bhv_desc         *p;
 424         struct xfs_mount        *mp = XFS_BHVTOM(bhvp);
 425         struct block_device     *ddev, *logdev, *rtdev;
 426         int                     flags = 0, error;
 427
 428         ddev = vfsp->vfs_super->s_bdev;
 429         logdev = rtdev = NULL;
 430
 431         /*
 432          * Setup xfs_mount function vectors from available behaviors
 433          */
 434         p = vfs_bhv_lookup(vfsp, VFS_POSITION_DM);
 435         mp->m_dm_ops = p ? *(xfs_dmops_t *) vfs_bhv_custom(p) : xfs_dmcore_stub;
 436         p = vfs_bhv_lookup(vfsp, VFS_POSITION_QM);
 437         mp->m_qm_ops = p ? *(xfs_qmops_t *) vfs_bhv_custom(p) : xfs_qmcore_stub;
 438         p = vfs_bhv_lookup(vfsp, VFS_POSITION_IO);
 439         mp->m_io_ops = p ? *(xfs_ioops_t *) vfs_bhv_custom(p) : xfs_iocore_xfs;
 440
 441         /*
 442          * Open real time and log devices - order is important.
 443          */
 444         if (args->logname[0]) {
 445                 error = xfs_blkdev_get(mp, args->logname, &logdev);
 446                 if (error)
 447                         return error;
 448         }
 449         if (args->rtname[0]) {
 450                 error = xfs_blkdev_get(mp, args->rtname, &rtdev);
 451                 if (error) {
 452                         xfs_blkdev_put(logdev);
 453                         return error;
 454                 }
 455
 456                 if (rtdev == ddev || rtdev == logdev) {
 457                         cmn_err(CE_WARN,
 458         "XFS: Cannot mount filesystem with identical rtdev and ddev/logdev.");
 459                         xfs_blkdev_put(logdev);
 460                         xfs_blkdev_put(rtdev);
 461                         return EINVAL;
 462                 }
 463         }
 464
 465         /*
 466          * Setup xfs_mount buffer target pointers
 467          */
 468         error = ENOMEM;
 469         mp->m_ddev_targp = xfs_alloc_buftarg(ddev, 0);
 470         if (!mp->m_ddev_targp) {
 471                 xfs_blkdev_put(logdev);
 472                 xfs_blkdev_put(rtdev);
 473                 return error;
 474         }
 475         if (rtdev) {
 476                 mp->m_rtdev_targp = xfs_alloc_buftarg(rtdev, 1);
 477                 if (!mp->m_rtdev_targp)
 478                         goto error0;
 479         }
 480         mp->m_logdev_targp = (logdev && logdev != ddev) ?
 481                                 xfs_alloc_buftarg(logdev, 1) : mp->m_ddev_targp;
 482         if (!mp->m_logdev_targp)
 483                 goto error0;
 484
 485         /*
 486          * Setup flags based on mount(2) options and then the superblock
 487          */
 488         error = xfs_start_flags(vfsp, args, mp);
 489         if (error)
 490                 goto error1;
 491         error = xfs_readsb(mp);
 492         if (error)
 493                 goto error1;
 494         error = xfs_finish_flags(vfsp, args, mp);
 495         if (error)
 496                 goto error2;
 497
 498         /*
 499          * Setup xfs_mount buffer target pointers based on superblock
 500          */
 501         error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_blocksize,
 502                                     mp->m_sb.sb_sectsize);
 503         if (!error && logdev && logdev != ddev) {
 504                 unsigned int    log_sector_size = BBSIZE;
 505
 506                 if (XFS_SB_VERSION_HASSECTOR(&mp->m_sb))
 507                         log_sector_size = mp->m_sb.sb_logsectsize;
 508                 error = xfs_setsize_buftarg(mp->m_logdev_targp,
 509                                             mp->m_sb.sb_blocksize,
 510                                             log_sector_size);
 511         }
 512         if (!error && rtdev)
 513                 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
 514                                             mp->m_sb.sb_blocksize,
 515                                             mp->m_sb.sb_sectsize);
 516         if (error)
 517                 goto error2;
 518
 519         error = XFS_IOINIT(vfsp, args, flags);
 520         if (error)
 521                 goto error2;
 522
 523         if ((args->flags & XFSMNT_BARRIER) &&
 524             !(XFS_MTOVFS(mp)->vfs_flag & VFS_RDONLY))
 525                 xfs_mountfs_check_barriers(mp);
 526         return 0;
 527
 528 error2:
 529         if (mp->m_sb_bp)
 530                 xfs_freesb(mp);
 531 error1:
 532         xfs_binval(mp->m_ddev_targp);
 533         if (logdev && logdev != ddev)
 534                 xfs_binval(mp->m_logdev_targp);
 535         if (rtdev)
 536                 xfs_binval(mp->m_rtdev_targp);
 537 error0:
 538         xfs_unmountfs_close(mp, credp);
 539         return error;
 540 }
 541
 542 STATIC int
 543 xfs_unmount(
 544         bhv_desc_t      *bdp,
 545         int             flags,
 546         cred_t          *credp)
 547 {
 548         struct vfs      *vfsp = bhvtovfs(bdp);
 549         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
 550         xfs_inode_t     *rip;
 551         vnode_t         *rvp;
 552         int             unmount_event_wanted = 0;
 553         int             unmount_event_flags = 0;
 554         int             xfs_unmountfs_needed = 0;
 555         int             error;
 556
 557         rip = mp->m_rootip;
 558         rvp = XFS_ITOV(rip);
 559
 560         if (vfsp->vfs_flag & VFS_DMI) {
 561                 error = XFS_SEND_PREUNMOUNT(mp, vfsp,
 562                                 rvp, DM_RIGHT_NULL, rvp, DM_RIGHT_NULL,
 563                                 NULL, NULL, 0, 0,
 564                                 (mp->m_dmevmask & (1<<DM_EVENT_PREUNMOUNT))?
 565                                         0:DM_FLAGS_UNWANTED);
 566                         if (error)
 567                                 return XFS_ERROR(error);
 568                 unmount_event_wanted = 1;
 569                 unmount_event_flags = (mp->m_dmevmask & (1<<DM_EVENT_UNMOUNT))?
 570                                         0 : DM_FLAGS_UNWANTED;
 571         }
 572
 573         /*
 574          * First blow any referenced inode from this file system
 575          * out of the reference cache, and delete the timer.
 576          */
 577         xfs_refcache_purge_mp(mp);
 578
 579         XFS_bflush(mp->m_ddev_targp);
 580         error = xfs_unmount_flush(mp, 0);
 581         if (error)
 582                 goto out;
 583
 584         ASSERT(vn_count(rvp) == 1);
 585
 586         /*
 587          * Drop the reference count
 588          */
 589         VN_RELE(rvp);
 590
 591         /*
 592          * If we're forcing a shutdown, typically because of a media error,
 593          * we want to make sure we invalidate dirty pages that belong to
 594          * referenced vnodes as well.
 595          */
 596         if (XFS_FORCED_SHUTDOWN(mp)) {
 597                 error = xfs_sync(&mp->m_bhv,
 598                          (SYNC_WAIT | SYNC_CLOSE), credp);
 599                 ASSERT(error != EFSCORRUPTED);
 600         }
 601         xfs_unmountfs_needed = 1;
 602
 603 out:
 604         /*      Send DMAPI event, if required.
 605          *      Then do xfs_unmountfs() if needed.
 606          *      Then return error (or zero).
 607          */
 608         if (unmount_event_wanted) {
 609                 /* Note: mp structure must still exist for
 610                  * XFS_SEND_UNMOUNT() call.
 611                  */
 612                 XFS_SEND_UNMOUNT(mp, vfsp, error == 0 ? rvp : NULL,
 613                         DM_RIGHT_NULL, 0, error, unmount_event_flags);
 614         }
 615         if (xfs_unmountfs_needed) {
 616                 /*
 617                  * Call common unmount function to flush to disk
 618                  * and free the super block buffer & mount structures.
 619                  */
 620                 xfs_unmountfs(mp, credp);
 621         }
 622
 623         return XFS_ERROR(error);
 624 }
 625
 626 STATIC int
 627 xfs_quiesce_fs(
 628         xfs_mount_t             *mp)
 629 {
 630         int                     count = 0, pincount;
 631
 632         xfs_refcache_purge_mp(mp);
 633         xfs_flush_buftarg(mp->m_ddev_targp, 0);
 634         xfs_finish_reclaim_all(mp, 0);
 635
 636         /* This loop must run at least twice.
 637          * The first instance of the loop will flush
 638          * most meta data but that will generate more
 639          * meta data (typically directory updates).
 640          * Which then must be flushed and logged before
 641          * we can write the unmount record.
 642          */
 643         do {
 644                 xfs_syncsub(mp, SYNC_REMOUNT|SYNC_ATTR|SYNC_WAIT, 0, NULL);
 645                 pincount = xfs_flush_buftarg(mp->m_ddev_targp, 1);
 646                 if (!pincount) {
 647                         delay(50);
 648                         count++;
 649                 }
 650         } while (count < 2);
 651
 652         return 0;
 653 }
 654
 655 STATIC int
 656 xfs_mntupdate(
 657         bhv_desc_t                      *bdp,
 658         int                             *flags,
 659         struct xfs_mount_args           *args)
 660 {
 661         struct vfs      *vfsp = bhvtovfs(bdp);
 662         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
 663         int             error;
 664
 665         if (args->flags & XFSMNT_NOATIME)
 666                 mp->m_flags |= XFS_MOUNT_NOATIME;
 667         else
 668                 mp->m_flags &= ~XFS_MOUNT_NOATIME;
 669
 670         if ((vfsp->vfs_flag & VFS_RDONLY) &&
 671             !(*flags & MS_RDONLY)) {
 672                 vfsp->vfs_flag &= ~VFS_RDONLY;
 673
 674                 if (args->flags & XFSMNT_BARRIER)
 675                         xfs_mountfs_check_barriers(mp);
 676         }
 677
 678         if (!(vfsp->vfs_flag & VFS_RDONLY) &&
 679             (*flags & MS_RDONLY)) {
 680                 VFS_SYNC(vfsp, SYNC_FSDATA|SYNC_BDFLUSH|SYNC_ATTR, NULL, error);
 681
 682                 xfs_quiesce_fs(mp);
 683
 684                 /* Ok now write out an unmount record */
 685                 xfs_log_unmount_write(mp);
 686                 xfs_unmountfs_writesb(mp);
 687                 vfsp->vfs_flag |= VFS_RDONLY;
 688         }
 689
 690         return 0;
 691 }
 692
 693 /*
 694  * xfs_unmount_flush implements a set of flush operation on special
 695  * inodes, which are needed as a separate set of operations so that
 696  * they can be called as part of relocation process.
 697  */
 698 int
 699 xfs_unmount_flush(
 700         xfs_mount_t     *mp,            /* Mount structure we are getting
 701                                            rid of. */
 702         int             relocation)     /* Called from vfs relocation. */
 703 {
 704         xfs_inode_t     *rip = mp->m_rootip;
 705         xfs_inode_t     *rbmip;
 706         xfs_inode_t     *rsumip = NULL;
 707         vnode_t         *rvp = XFS_ITOV(rip);
 708         int             error;
 709
 710         xfs_ilock(rip, XFS_ILOCK_EXCL);
 711         xfs_iflock(rip);
 712
 713         /*
 714          * Flush out the real time inodes.
 715          */
 716         if ((rbmip = mp->m_rbmip) != NULL) {
 717                 xfs_ilock(rbmip, XFS_ILOCK_EXCL);
 718                 xfs_iflock(rbmip);
 719                 error = xfs_iflush(rbmip, XFS_IFLUSH_SYNC);
 720                 xfs_iunlock(rbmip, XFS_ILOCK_EXCL);
 721
 722                 if (error == EFSCORRUPTED)
 723                         goto fscorrupt_out;
 724
 725                 ASSERT(vn_count(XFS_ITOV(rbmip)) == 1);
 726
 727                 rsumip = mp->m_rsumip;
 728                 xfs_ilock(rsumip, XFS_ILOCK_EXCL);
 729                 xfs_iflock(rsumip);
 730                 error = xfs_iflush(rsumip, XFS_IFLUSH_SYNC);
 731                 xfs_iunlock(rsumip, XFS_ILOCK_EXCL);
 732
 733                 if (error == EFSCORRUPTED)
 734                         goto fscorrupt_out;
 735
 736                 ASSERT(vn_count(XFS_ITOV(rsumip)) == 1);
 737         }
 738
 739         /*
 740          * Synchronously flush root inode to disk
 741          */
 742         error = xfs_iflush(rip, XFS_IFLUSH_SYNC);
 743         if (error == EFSCORRUPTED)
 744                 goto fscorrupt_out2;
 745
 746         if (vn_count(rvp) != 1 && !relocation) {
 747                 xfs_iunlock(rip, XFS_ILOCK_EXCL);
 748                 return XFS_ERROR(EBUSY);
 749         }
 750
 751         /*
 752          * Release dquot that rootinode, rbmino and rsumino might be holding,
 753          * flush and purge the quota inodes.
 754          */
 755         error = XFS_QM_UNMOUNT(mp);
 756         if (error == EFSCORRUPTED)
 757                 goto fscorrupt_out2;
 758
 759         if (rbmip) {
 760                 VN_RELE(XFS_ITOV(rbmip));
 761                 VN_RELE(XFS_ITOV(rsumip));
 762         }
 763
 764         xfs_iunlock(rip, XFS_ILOCK_EXCL);
 765         return 0;
 766
 767 fscorrupt_out:
 768         xfs_ifunlock(rip);
 769
 770 fscorrupt_out2:
 771         xfs_iunlock(rip, XFS_ILOCK_EXCL);
 772
 773         return XFS_ERROR(EFSCORRUPTED);
 774 }
 775
 776 /*
 777  * xfs_root extracts the root vnode from a vfs.
 778  *
 779  * vfsp -- the vfs struct for the desired file system
 780  * vpp  -- address of the caller's vnode pointer which should be
 781  *         set to the desired fs root vnode
 782  */
 783 STATIC int
 784 xfs_root(
 785         bhv_desc_t      *bdp,
 786         vnode_t         **vpp)
 787 {
 788         vnode_t         *vp;
 789
 790         vp = XFS_ITOV((XFS_BHVTOM(bdp))->m_rootip);
 791         VN_HOLD(vp);
 792         *vpp = vp;
 793         return 0;
 794 }
 795
 796 /*
 797  * xfs_statvfs
 798  *
 799  * Fill in the statvfs structure for the given file system.  We use
 800  * the superblock lock in the mount structure to ensure a consistent
 801  * snapshot of the counters returned.
 802  */
 803 STATIC int
 804 xfs_statvfs(
 805         bhv_desc_t      *bdp,
 806         xfs_statfs_t    *statp,
 807         vnode_t         *vp)
 808 {
 809         __uint64_t      fakeinos;
 810         xfs_extlen_t    lsize;
 811         xfs_mount_t     *mp;
 812         xfs_sb_t        *sbp;
 813         unsigned long   s;
 814
 815         mp = XFS_BHVTOM(bdp);
 816         sbp = &(mp->m_sb);
 817
 818         statp->f_type = XFS_SB_MAGIC;
 819
 820         s = XFS_SB_LOCK(mp);
 821         statp->f_bsize = sbp->sb_blocksize;
 822         lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
 823         statp->f_blocks = sbp->sb_dblocks - lsize;
 824         statp->f_bfree = statp->f_bavail = sbp->sb_fdblocks;
 825         fakeinos = statp->f_bfree << sbp->sb_inopblog;
 826 #if XFS_BIG_INUMS
 827         fakeinos += mp->m_inoadd;
 828 #endif
 829         statp->f_files =
 830             MIN(sbp->sb_icount + fakeinos, (__uint64_t)XFS_MAXINUMBER);
 831         if (mp->m_maxicount)
 832 #if XFS_BIG_INUMS
 833                 if (!mp->m_inoadd)
 834 #endif
 835                         statp->f_files = min_t(typeof(statp->f_files),
 836                                                 statp->f_files,
 837                                                 mp->m_maxicount);
 838         statp->f_ffree = statp->f_files - (sbp->sb_icount - sbp->sb_ifree);
 839         XFS_SB_UNLOCK(mp, s);
 840
 841         xfs_statvfs_fsid(statp, mp);
 842         statp->f_namelen = MAXNAMELEN - 1;
 843
 844         return 0;
 845 }
 846
 847
 848 /*
 849  * xfs_sync flushes any pending I/O to file system vfsp.
 850  *
 851  * This routine is called by vfs_sync() to make sure that things make it
 852  * out to disk eventually, on sync() system calls to flush out everything,
 853  * and when the file system is unmounted.  For the vfs_sync() case, all
 854  * we really need to do is sync out the log to make all of our meta-data
 855  * updates permanent (except for timestamps).  For calls from pflushd(),
 856  * dirty pages are kept moving by calling pdflush() on the inodes
 857  * containing them.  We also flush the inodes that we can lock without
 858  * sleeping and the superblock if we can lock it without sleeping from
 859  * vfs_sync() so that items at the tail of the log are always moving out.
 860  *
 861  * Flags:
 862  *      SYNC_BDFLUSH - We're being called from vfs_sync() so we don't want
 863  *                     to sleep if we can help it.  All we really need
 864  *                     to do is ensure that the log is synced at least
 865  *                     periodically.  We also push the inodes and
 866  *                     superblock if we can lock them without sleeping
 867  *                      and they are not pinned.
 868  *      SYNC_ATTR    - We need to flush the inodes.  If SYNC_BDFLUSH is not
 869  *                     set, then we really want to lock each inode and flush
 870  *                     it.
 871  *      SYNC_WAIT    - All the flushes that take place in this call should
 872  *                     be synchronous.
 873  *      SYNC_DELWRI  - This tells us to push dirty pages associated with
 874  *                     inodes.  SYNC_WAIT and SYNC_BDFLUSH are used to
 875  *                     determine if they should be flushed sync, async, or
 876  *                     delwri.
 877  *      SYNC_CLOSE   - This flag is passed when the system is being
 878  *                     unmounted.  We should sync and invalidate everthing.
 879  *      SYNC_FSDATA  - This indicates that the caller would like to make
 880  *                     sure the superblock is safe on disk.  We can ensure
 881  *                     this by simply makeing sure the log gets flushed
 882  *                     if SYNC_BDFLUSH is set, and by actually writing it
 883  *                     out otherwise.
 884  *
 885  */
 886 /*ARGSUSED*/
 887 STATIC int
 888 xfs_sync(
 889         bhv_desc_t      *bdp,
 890         int             flags,
 891         cred_t          *credp)
 892 {
 893         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
 894
 895         if (unlikely(flags == SYNC_QUIESCE))
 896                 return xfs_quiesce_fs(mp);
 897         else
 898                 return xfs_syncsub(mp, flags, 0, NULL);
 899 }
 900
 901 /*
 902  * xfs sync routine for internal use
 903  *
 904  * This routine supports all of the flags defined for the generic VFS_SYNC
 905  * interface as explained above under xfs_sync.  In the interests of not
 906  * changing interfaces within the 6.5 family, additional internallly-
 907  * required functions are specified within a separate xflags parameter,
 908  * only available by calling this routine.
 909  *
 910  */
 911 int
 912 xfs_sync_inodes(
 913         xfs_mount_t     *mp,
 914         int             flags,
 915         int             xflags,
 916         int             *bypassed)
 917 {
 918         xfs_inode_t     *ip = NULL;
 919         xfs_inode_t     *ip_next;
 920         xfs_buf_t       *bp;
 921         vnode_t         *vp = NULL;
 922         int             error;
 923         int             last_error;
 924         uint64_t        fflag;
 925         uint            lock_flags;
 926         uint            base_lock_flags;
 927         boolean_t       mount_locked;
 928         boolean_t       vnode_refed;
 929         int             preempt;
 930         xfs_dinode_t    *dip;
 931         xfs_iptr_t      *ipointer;
 932 #ifdef DEBUG
 933         boolean_t       ipointer_in = B_FALSE;
 934
 935 #define IPOINTER_SET    ipointer_in = B_TRUE
 936 #define IPOINTER_CLR    ipointer_in = B_FALSE
 937 #else
 938 #define IPOINTER_SET
 939 #define IPOINTER_CLR
 940 #endif
 941
 942
 943 /* Insert a marker record into the inode list after inode ip. The list
 944  * must be locked when this is called. After the call the list will no
 945  * longer be locked.
 946  */
 947 #define IPOINTER_INSERT(ip, mp) { \
 948                 ASSERT(ipointer_in == B_FALSE); \
 949                 ipointer->ip_mnext = ip->i_mnext; \
 950                 ipointer->ip_mprev = ip; \
 951                 ip->i_mnext = (xfs_inode_t *)ipointer; \
 952                 ipointer->ip_mnext->i_mprev = (xfs_inode_t *)ipointer; \
 953                 preempt = 0; \
 954                 XFS_MOUNT_IUNLOCK(mp); \
 955                 mount_locked = B_FALSE; \
 956                 IPOINTER_SET; \
 957         }
 958
 959 /* Remove the marker from the inode list. If the marker was the only item
 960  * in the list then there are no remaining inodes and we should zero out
 961  * the whole list. If we are the current head of the list then move the head
 962  * past us.
 963  */
 964 #define IPOINTER_REMOVE(ip, mp) { \
 965                 ASSERT(ipointer_in == B_TRUE); \
 966                 if (ipointer->ip_mnext != (xfs_inode_t *)ipointer) { \
 967                         ip = ipointer->ip_mnext; \
 968                         ip->i_mprev = ipointer->ip_mprev; \
 969                         ipointer->ip_mprev->i_mnext = ip; \
 970                         if (mp->m_inodes == (xfs_inode_t *)ipointer) { \
 971                                 mp->m_inodes = ip; \
 972                         } \
 973                 } else { \
 974                         ASSERT(mp->m_inodes == (xfs_inode_t *)ipointer); \
 975                         mp->m_inodes = NULL; \
 976                         ip = NULL; \
 977                 } \
 978                 IPOINTER_CLR; \
 979         }
 980
 981 #define XFS_PREEMPT_MASK        0x7f
 982
 983         if (bypassed)
 984                 *bypassed = 0;
 985         if (XFS_MTOVFS(mp)->vfs_flag & VFS_RDONLY)
 986                 return 0;
 987         error = 0;
 988         last_error = 0;
 989         preempt = 0;
 990
 991         /* Allocate a reference marker */
 992         ipointer = (xfs_iptr_t *)kmem_zalloc(sizeof(xfs_iptr_t), KM_SLEEP);
 993
 994         fflag = XFS_B_ASYNC;            /* default is don't wait */
 995         if (flags & (SYNC_BDFLUSH | SYNC_DELWRI))
 996                 fflag = XFS_B_DELWRI;
 997         if (flags & SYNC_WAIT)
 998                 fflag = 0;              /* synchronous overrides all */
 999
1000         base_lock_flags = XFS_ILOCK_SHARED;
1001         if (flags & (SYNC_DELWRI | SYNC_CLOSE)) {
1002                 /*
1003                  * We need the I/O lock if we're going to call any of
1004                  * the flush/inval routines.
1005                  */
1006                 base_lock_flags |= XFS_IOLOCK_SHARED;
1007         }
1008
1009         XFS_MOUNT_ILOCK(mp);
1010
1011         ip = mp->m_inodes;
1012
1013         mount_locked = B_TRUE;
1014         vnode_refed  = B_FALSE;
1015
1016         IPOINTER_CLR;
1017
1018         do {
1019                 ASSERT(ipointer_in == B_FALSE);
1020                 ASSERT(vnode_refed == B_FALSE);
1021
1022                 lock_flags = base_lock_flags;
1023
1024                 /*
1025                  * There were no inodes in the list, just break out
1026                  * of the loop.
1027                  */
1028                 if (ip == NULL) {
1029                         break;
1030                 }
1031
1032                 /*
1033                  * We found another sync thread marker - skip it
1034                  */
1035                 if (ip->i_mount == NULL) {
1036                         ip = ip->i_mnext;
1037                         continue;
1038                 }
1039
1040                 vp = XFS_ITOV_NULL(ip);
1041
1042                 /*
1043                  * If the vnode is gone then this is being torn down,
1044                  * call reclaim if it is flushed, else let regular flush
1045                  * code deal with it later in the loop.
1046                  */
1047
1048                 if (vp == NULL) {
1049                         /* Skip ones already in reclaim */
1050                         if (ip->i_flags & XFS_IRECLAIM) {
1051                                 ip = ip->i_mnext;
1052                                 continue;
1053                         }
1054                         if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0) {
1055                                 ip = ip->i_mnext;
1056                         } else if ((xfs_ipincount(ip) == 0) &&
1057                                     xfs_iflock_nowait(ip)) {
1058                                 IPOINTER_INSERT(ip, mp);
1059
1060                                 xfs_finish_reclaim(ip, 1,
1061                                                 XFS_IFLUSH_DELWRI_ELSE_ASYNC);
1062
1063                                 XFS_MOUNT_ILOCK(mp);
1064                                 mount_locked = B_TRUE;
1065                                 IPOINTER_REMOVE(ip, mp);
1066                         } else {
1067                                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1068                                 ip = ip->i_mnext;
1069                         }
1070                         continue;
1071                 }
1072
1073                 if (VN_BAD(vp)) {
1074                         ip = ip->i_mnext;
1075                         continue;
1076                 }
1077
1078                 if (XFS_FORCED_SHUTDOWN(mp) && !(flags & SYNC_CLOSE)) {
1079                         XFS_MOUNT_IUNLOCK(mp);
1080                         kmem_free(ipointer, sizeof(xfs_iptr_t));
1081                         return 0;
1082                 }
1083
1084                 /*
1085                  * If this is just vfs_sync() or pflushd() calling
1086                  * then we can skip inodes for which it looks like
1087                  * there is nothing to do.  Since we don't have the
1088                  * inode locked this is racey, but these are periodic
1089                  * calls so it doesn't matter.  For the others we want
1090                  * to know for sure, so we at least try to lock them.
1091                  */
1092                 if (flags & SYNC_BDFLUSH) {
1093                         if (((ip->i_itemp == NULL) ||
1094                              !(ip->i_itemp->ili_format.ilf_fields &
1095                                XFS_ILOG_ALL)) &&
1096                             (ip->i_update_core == 0)) {
1097                                 ip = ip->i_mnext;
1098                                 continue;
1099                         }
1100                 }
1101
1102                 /*
1103                  * Try to lock without sleeping.  We're out of order with
1104                  * the inode list lock here, so if we fail we need to drop
1105                  * the mount lock and try again.  If we're called from
1106                  * bdflush() here, then don't bother.
1107                  *
1108                  * The inode lock here actually coordinates with the
1109                  * almost spurious inode lock in xfs_ireclaim() to prevent
1110                  * the vnode we handle here without a reference from
1111                  * being freed while we reference it.  If we lock the inode
1112                  * while it's on the mount list here, then the spurious inode
1113                  * lock in xfs_ireclaim() after the inode is pulled from
1114                  * the mount list will sleep until we release it here.
1115                  * This keeps the vnode from being freed while we reference
1116                  * it.
1117                  */
1118                 if (xfs_ilock_nowait(ip, lock_flags) == 0) {
1119                         if ((flags & SYNC_BDFLUSH) || (vp == NULL)) {
1120                                 ip = ip->i_mnext;
1121                                 continue;
1122                         }
1123
1124                         vp = vn_grab(vp);
1125                         if (vp == NULL) {
1126                                 ip = ip->i_mnext;
1127                                 continue;
1128                         }
1129
1130                         IPOINTER_INSERT(ip, mp);
1131                         xfs_ilock(ip, lock_flags);
1132
1133                         ASSERT(vp == XFS_ITOV(ip));
1134                         ASSERT(ip->i_mount == mp);
1135
1136                         vnode_refed = B_TRUE;
1137                 }
1138
1139                 /* From here on in the loop we may have a marker record
1140                  * in the inode list.
1141                  */
1142
1143                 if ((flags & SYNC_CLOSE)  && (vp != NULL)) {
1144                         /*
1145                          * This is the shutdown case.  We just need to
1146                          * flush and invalidate all the pages associated
1147                          * with the inode.  Drop the inode lock since
1148                          * we can't hold it across calls to the buffer
1149                          * cache.
1150                          *
1151                          * We don't set the VREMAPPING bit in the vnode
1152                          * here, because we don't hold the vnode lock
1153                          * exclusively.  It doesn't really matter, though,
1154                          * because we only come here when we're shutting
1155                          * down anyway.
1156                          */
1157                         xfs_iunlock(ip, XFS_ILOCK_SHARED);
1158
1159                         if (XFS_FORCED_SHUTDOWN(mp)) {
1160                                 VOP_TOSS_PAGES(vp, 0, -1, FI_REMAPF);
1161                         } else {
1162                                 VOP_FLUSHINVAL_PAGES(vp, 0, -1, FI_REMAPF);
1163                         }
1164
1165                         xfs_ilock(ip, XFS_ILOCK_SHARED);
1166
1167                 } else if ((flags & SYNC_DELWRI) && (vp != NULL)) {
1168                         if (VN_DIRTY(vp)) {
1169                                 /* We need to have dropped the lock here,
1170                                  * so insert a marker if we have not already
1171                                  * done so.
1172                                  */
1173                                 if (mount_locked) {
1174                                         IPOINTER_INSERT(ip, mp);
1175                                 }
1176
1177                                 /*
1178                                  * Drop the inode lock since we can't hold it
1179                                  * across calls to the buffer cache.
1180                                  */
1181                                 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1182                                 VOP_FLUSH_PAGES(vp, (xfs_off_t)0, -1,
1183                                                         fflag, FI_NONE, error);
1184                                 xfs_ilock(ip, XFS_ILOCK_SHARED);
1185                         }
1186
1187                 }
1188
1189                 if (flags & SYNC_BDFLUSH) {
1190                         if ((flags & SYNC_ATTR) &&
1191                             ((ip->i_update_core) ||
1192                              ((ip->i_itemp != NULL) &&
1193                               (ip->i_itemp->ili_format.ilf_fields != 0)))) {
1194
1195                                 /* Insert marker and drop lock if not already
1196                                  * done.
1197                                  */
1198                                 if (mount_locked) {
1199                                         IPOINTER_INSERT(ip, mp);
1200                                 }
1201
1202                                 /*
1203                                  * We don't want the periodic flushing of the
1204                                  * inodes by vfs_sync() to interfere with
1205                                  * I/O to the file, especially read I/O
1206                                  * where it is only the access time stamp
1207                                  * that is being flushed out.  To prevent
1208                                  * long periods where we have both inode
1209                                  * locks held shared here while reading the
1210                                  * inode's buffer in from disk, we drop the
1211                                  * inode lock while reading in the inode
1212                                  * buffer.  We have to release the buffer
1213                                  * and reacquire the inode lock so that they
1214                                  * are acquired in the proper order (inode
1215                                  * locks first).  The buffer will go at the
1216                                  * end of the lru chain, though, so we can
1217                                  * expect it to still be there when we go
1218                                  * for it again in xfs_iflush().
1219                                  */
1220                                 if ((xfs_ipincount(ip) == 0) &&
1221                                     xfs_iflock_nowait(ip)) {
1222
1223                                         xfs_ifunlock(ip);
1224                                         xfs_iunlock(ip, XFS_ILOCK_SHARED);
1225
1226                                         error = xfs_itobp(mp, NULL, ip,
1227                                                           &dip, &bp, 0);
1228                                         if (!error) {
1229                                                 xfs_buf_relse(bp);
1230                                         } else {
1231                                                 /* Bailing out, remove the
1232                                                  * marker and free it.
1233                                                  */
1234                                                 XFS_MOUNT_ILOCK(mp);
1235
1236                                                 IPOINTER_REMOVE(ip, mp);
1237
1238                                                 XFS_MOUNT_IUNLOCK(mp);
1239
1240                                                 ASSERT(!(lock_flags &
1241                                                         XFS_IOLOCK_SHARED));
1242
1243                                                 kmem_free(ipointer,
1244                                                         sizeof(xfs_iptr_t));
1245                                                 return (0);
1246                                         }
1247
1248                                         /*
1249                                          * Since we dropped the inode lock,
1250                                          * the inode may have been reclaimed.
1251                                          * Therefore, we reacquire the mount
1252                                          * lock and check to see if we were the
1253                                          * inode reclaimed. If this happened
1254                                          * then the ipointer marker will no
1255                                          * longer point back at us. In this
1256                                          * case, move ip along to the inode
1257                                          * after the marker, remove the marker
1258                                          * and continue.
1259                                          */
1260                                         XFS_MOUNT_ILOCK(mp);
1261                                         mount_locked = B_TRUE;
1262
1263                                         if (ip != ipointer->ip_mprev) {
1264                                                 IPOINTER_REMOVE(ip, mp);
1265
1266                                                 ASSERT(!vnode_refed);
1267                                                 ASSERT(!(lock_flags &
1268                                                         XFS_IOLOCK_SHARED));
1269                                                 continue;
1270                                         }
1271
1272                                         ASSERT(ip->i_mount == mp);
1273
1274                                         if (xfs_ilock_nowait(ip,
1275                                                     XFS_ILOCK_SHARED) == 0) {
1276                                                 ASSERT(ip->i_mount == mp);
1277                                                 /*
1278                                                  * We failed to reacquire
1279                                                  * the inode lock without
1280                                                  * sleeping, so just skip
1281                                                  * the inode for now.  We
1282                                                  * clear the ILOCK bit from
1283                                                  * the lock_flags so that we
1284                                                  * won't try to drop a lock
1285                                                  * we don't hold below.
1286                                                  */
1287                                                 lock_flags &= ~XFS_ILOCK_SHARED;
1288                                                 IPOINTER_REMOVE(ip_next, mp);
1289                                         } else if ((xfs_ipincount(ip) == 0) &&
1290                                                    xfs_iflock_nowait(ip)) {
1291                                                 ASSERT(ip->i_mount == mp);
1292                                                 /*
1293                                                  * Since this is vfs_sync()
1294                                                  * calling we only flush the
1295                                                  * inode out if we can lock
1296                                                  * it without sleeping and
1297                                                  * it is not pinned.  Drop
1298                                                  * the mount lock here so
1299                                                  * that we don't hold it for
1300                                                  * too long. We already have
1301                                                  * a marker in the list here.
1302                                                  */
1303                                                 XFS_MOUNT_IUNLOCK(mp);
1304                                                 mount_locked = B_FALSE;
1305                                                 error = xfs_iflush(ip,
1306                                                            XFS_IFLUSH_DELWRI);
1307                                         } else {
1308                                                 ASSERT(ip->i_mount == mp);
1309                                                 IPOINTER_REMOVE(ip_next, mp);
1310                                         }
1311                                 }
1312
1313                         }
1314
1315                 } else {
1316                         if ((flags & SYNC_ATTR) &&
1317                             ((ip->i_update_core) ||
1318                              ((ip->i_itemp != NULL) &&
1319                               (ip->i_itemp->ili_format.ilf_fields != 0)))) {
1320                                 if (mount_locked) {
1321                                         IPOINTER_INSERT(ip, mp);
1322                                 }
1323
1324                                 if (flags & SYNC_WAIT) {
1325                                         xfs_iflock(ip);
1326                                         error = xfs_iflush(ip,
1327                                                            XFS_IFLUSH_SYNC);
1328                                 } else {
1329                                         /*
1330                                          * If we can't acquire the flush
1331                                          * lock, then the inode is already
1332                                          * being flushed so don't bother
1333                                          * waiting.  If we can lock it then
1334                                          * do a delwri flush so we can
1335                                          * combine multiple inode flushes
1336                                          * in each disk write.
1337                                          */
1338                                         if (xfs_iflock_nowait(ip)) {
1339                                                 error = xfs_iflush(ip,
1340                                                            XFS_IFLUSH_DELWRI);
1341                                         }
1342                                         else if (bypassed)
1343                                                 (*bypassed)++;
1344                                 }
1345                         }
1346                 }
1347
1348                 if (lock_flags != 0) {
1349                         xfs_iunlock(ip, lock_flags);
1350                 }
1351
1352                 if (vnode_refed) {
1353                         /*
1354                          * If we had to take a reference on the vnode
1355                          * above, then wait until after we've unlocked
1356                          * the inode to release the reference.  This is
1357                          * because we can be already holding the inode
1358                          * lock when VN_RELE() calls xfs_inactive().
1359                          *
1360                          * Make sure to drop the mount lock before calling
1361                          * VN_RELE() so that we don't trip over ourselves if
1362                          * we have to go for the mount lock again in the
1363                          * inactive code.
1364                          */
1365                         if (mount_locked) {
1366                                 IPOINTER_INSERT(ip, mp);
1367                         }
1368
1369                         VN_RELE(vp);
1370
1371                         vnode_refed = B_FALSE;
1372                 }
1373
1374                 if (error) {
1375                         last_error = error;
1376                 }
1377
1378                 /*
1379                  * bail out if the filesystem is corrupted.
1380                  */
1381                 if (error == EFSCORRUPTED)  {
1382                         if (!mount_locked) {
1383                                 XFS_MOUNT_ILOCK(mp);
1384                                 IPOINTER_REMOVE(ip, mp);
1385                         }
1386                         XFS_MOUNT_IUNLOCK(mp);
1387                         ASSERT(ipointer_in == B_FALSE);
1388                         kmem_free(ipointer, sizeof(xfs_iptr_t));
1389                         return XFS_ERROR(error);
1390                 }
1391
1392                 /* Let other threads have a chance at the mount lock
1393                  * if we have looped many times without dropping the
1394                  * lock.
1395                  */
1396                 if ((++preempt & XFS_PREEMPT_MASK) == 0) {
1397                         if (mount_locked) {
1398                                 IPOINTER_INSERT(ip, mp);
1399                         }
1400                 }
1401
1402                 if (mount_locked == B_FALSE) {
1403                         XFS_MOUNT_ILOCK(mp);
1404                         mount_locked = B_TRUE;
1405                         IPOINTER_REMOVE(ip, mp);
1406                         continue;
1407                 }
1408
1409                 ASSERT(ipointer_in == B_FALSE);
1410                 ip = ip->i_mnext;
1411
1412         } while (ip != mp->m_inodes);
1413
1414         XFS_MOUNT_IUNLOCK(mp);
1415
1416         ASSERT(ipointer_in == B_FALSE);
1417
1418         kmem_free(ipointer, sizeof(xfs_iptr_t));
1419         return XFS_ERROR(last_error);
1420 }
1421
1422 /*
1423  * xfs sync routine for internal use
1424  *
1425  * This routine supports all of the flags defined for the generic VFS_SYNC
1426  * interface as explained above under xfs_sync.  In the interests of not
1427  * changing interfaces within the 6.5 family, additional internallly-
1428  * required functions are specified within a separate xflags parameter,
1429  * only available by calling this routine.
1430  *
1431  */
1432 int
1433 xfs_syncsub(
1434         xfs_mount_t     *mp,
1435         int             flags,
1436         int             xflags,
1437         int             *bypassed)
1438 {
1439         int             error = 0;
1440         int             last_error = 0;
1441         uint            log_flags = XFS_LOG_FORCE;
1442         xfs_buf_t       *bp;
1443         xfs_buf_log_item_t      *bip;
1444
1445         /*
1446          * Sync out the log.  This ensures that the log is periodically
1447          * flushed even if there is not enough activity to fill it up.
1448          */
1449         if (flags & SYNC_WAIT)
1450                 log_flags |= XFS_LOG_SYNC;
1451
1452         xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1453
1454         if (flags & (SYNC_ATTR|SYNC_DELWRI)) {
1455                 if (flags & SYNC_BDFLUSH)
1456                         xfs_finish_reclaim_all(mp, 1);
1457                 else
1458                         error = xfs_sync_inodes(mp, flags, xflags, bypassed);
1459         }
1460
1461         /*
1462          * Flushing out dirty data above probably generated more
1463          * log activity, so if this isn't vfs_sync() then flush
1464          * the log again.
1465          */
1466         if (flags & SYNC_DELWRI) {
1467                 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1468         }
1469
1470         if (flags & SYNC_FSDATA) {
1471                 /*
1472                  * If this is vfs_sync() then only sync the superblock
1473                  * if we can lock it without sleeping and it is not pinned.
1474                  */
1475                 if (flags & SYNC_BDFLUSH) {
1476                         bp = xfs_getsb(mp, XFS_BUF_TRYLOCK);
1477                         if (bp != NULL) {
1478                                 bip = XFS_BUF_FSPRIVATE(bp,xfs_buf_log_item_t*);
1479                                 if ((bip != NULL) &&
1480                                     xfs_buf_item_dirty(bip)) {
1481                                         if (!(XFS_BUF_ISPINNED(bp))) {
1482                                                 XFS_BUF_ASYNC(bp);
1483                                                 error = xfs_bwrite(mp, bp);
1484                                         } else {
1485                                                 xfs_buf_relse(bp);
1486                                         }
1487                                 } else {
1488                                         xfs_buf_relse(bp);
1489                                 }
1490                         }
1491                 } else {
1492                         bp = xfs_getsb(mp, 0);
1493                         /*
1494                          * If the buffer is pinned then push on the log so
1495                          * we won't get stuck waiting in the write for
1496                          * someone, maybe ourselves, to flush the log.
1497                          * Even though we just pushed the log above, we
1498                          * did not have the superblock buffer locked at
1499                          * that point so it can become pinned in between
1500                          * there and here.
1501                          */
1502                         if (XFS_BUF_ISPINNED(bp))
1503                                 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
1504                         if (flags & SYNC_WAIT)
1505                                 XFS_BUF_UNASYNC(bp);
1506                         else
1507                                 XFS_BUF_ASYNC(bp);
1508                         error = xfs_bwrite(mp, bp);
1509                 }
1510                 if (error) {
1511                         last_error = error;
1512                 }
1513         }
1514
1515         /*
1516          * If this is the periodic sync, then kick some entries out of
1517          * the reference cache.  This ensures that idle entries are
1518          * eventually kicked out of the cache.
1519          */
1520         if (flags & SYNC_REFCACHE) {
1521                 if (flags & SYNC_WAIT)
1522                         xfs_refcache_purge_mp(mp);
1523                 else
1524                         xfs_refcache_purge_some(mp);
1525         }
1526
1527         /*
1528          * Now check to see if the log needs a "dummy" transaction.
1529          */
1530
1531         if (!(flags & SYNC_REMOUNT) && xfs_log_need_covered(mp)) {
1532                 xfs_trans_t *tp;
1533                 xfs_inode_t *ip;
1534
1535                 /*
1536                  * Put a dummy transaction in the log to tell
1537                  * recovery that all others are OK.
1538                  */
1539                 tp = xfs_trans_alloc(mp, XFS_TRANS_DUMMY1);
1540                 if ((error = xfs_trans_reserve(tp, 0,
1541                                 XFS_ICHANGE_LOG_RES(mp),
1542                                 0, 0, 0)))  {
1543                         xfs_trans_cancel(tp, 0);
1544                         return error;
1545                 }
1546
1547                 ip = mp->m_rootip;
1548                 xfs_ilock(ip, XFS_ILOCK_EXCL);
1549
1550                 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1551                 xfs_trans_ihold(tp, ip);
1552                 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1553                 error = xfs_trans_commit(tp, 0, NULL);
1554                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1555                 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1556         }
1557
1558         /*
1559          * When shutting down, we need to insure that the AIL is pushed
1560          * to disk or the filesystem can appear corrupt from the PROM.
1561          */
1562         if ((flags & (SYNC_CLOSE|SYNC_WAIT)) == (SYNC_CLOSE|SYNC_WAIT)) {
1563                 XFS_bflush(mp->m_ddev_targp);
1564                 if (mp->m_rtdev_targp) {
1565                         XFS_bflush(mp->m_rtdev_targp);
1566                 }
1567         }
1568
1569         return XFS_ERROR(last_error);
1570 }
1571
1572 /*
1573  * xfs_vget - called by DMAPI and NFSD to get vnode from file handle
1574  */
1575 STATIC int
1576 xfs_vget(
1577         bhv_desc_t      *bdp,
1578         vnode_t         **vpp,
1579         fid_t           *fidp)
1580 {
1581         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
1582         xfs_fid_t       *xfid = (struct xfs_fid *)fidp;
1583         xfs_inode_t     *ip;
1584         int             error;
1585         xfs_ino_t       ino;
1586         unsigned int    igen;
1587
1588         /*
1589          * Invalid.  Since handles can be created in user space and passed in
1590          * via gethandle(), this is not cause for a panic.
1591          */
1592         if (xfid->xfs_fid_len != sizeof(*xfid) - sizeof(xfid->xfs_fid_len))
1593                 return XFS_ERROR(EINVAL);
1594
1595         ino  = xfid->xfs_fid_ino;
1596         igen = xfid->xfs_fid_gen;
1597
1598         /*
1599          * NFS can sometimes send requests for ino 0.  Fail them gracefully.
1600          */
1601         if (ino == 0)
1602                 return XFS_ERROR(ESTALE);
1603
1604         error = xfs_iget(mp, NULL, ino, 0, XFS_ILOCK_SHARED, &ip, 0);
1605         if (error) {
1606                 *vpp = NULL;
1607                 return error;
1608         }
1609
1610         if (ip == NULL) {
1611                 *vpp = NULL;
1612                 return XFS_ERROR(EIO);
1613         }
1614
1615         if (ip->i_d.di_mode == 0 || ip->i_d.di_gen != igen) {
1616                 xfs_iput_new(ip, XFS_ILOCK_SHARED);
1617                 *vpp = NULL;
1618                 return XFS_ERROR(ENOENT);
1619         }
1620
1621         *vpp = XFS_ITOV(ip);
1622         xfs_iunlock(ip, XFS_ILOCK_SHARED);
1623         return 0;
1624 }
1625
1626
1627 #define MNTOPT_LOGBUFS  "logbufs"       /* number of XFS log buffers */
1628 #define MNTOPT_LOGBSIZE "logbsize"      /* size of XFS log buffers */
1629 #define MNTOPT_LOGDEV   "logdev"        /* log device */
1630 #define MNTOPT_RTDEV    "rtdev"         /* realtime I/O device */
1631 #define MNTOPT_BIOSIZE  "biosize"       /* log2 of preferred buffered io size */
1632 #define MNTOPT_WSYNC    "wsync"         /* safe-mode nfs compatible mount */
1633 #define MNTOPT_INO64    "ino64"         /* force inodes into 64-bit range */
1634 #define MNTOPT_NOALIGN  "noalign"       /* turn off stripe alignment */
1635 #define MNTOPT_SWALLOC  "swalloc"       /* turn on stripe width allocation */
1636 #define MNTOPT_SUNIT    "sunit"         /* data volume stripe unit */
1637 #define MNTOPT_SWIDTH   "swidth"        /* data volume stripe width */
1638 #define MNTOPT_NOUUID   "nouuid"        /* ignore filesystem UUID */
1639 #define MNTOPT_MTPT     "mtpt"          /* filesystem mount point */
1640 #define MNTOPT_GRPID    "grpid"         /* group-ID from parent directory */
1641 #define MNTOPT_NOGRPID  "nogrpid"       /* group-ID from current process */
1642 #define MNTOPT_BSDGROUPS    "bsdgroups"    /* group-ID from parent directory */
1643 #define MNTOPT_SYSVGROUPS   "sysvgroups"   /* group-ID from current process */
1644 #define MNTOPT_ALLOCSIZE    "allocsize"    /* preferred allocation size */
1645 #define MNTOPT_IHASHSIZE    "ihashsize"    /* size of inode hash table */
1646 #define MNTOPT_NORECOVERY   "norecovery"   /* don't run XFS recovery */
1647 #define MNTOPT_BARRIER  "barrier"       /* use writer barriers for log write and
1648                                          * unwritten extent conversion */
1649 #define MNTOPT_OSYNCISOSYNC "osyncisosync" /* o_sync is REALLY o_sync */
1650 #define MNTOPT_64BITINODE   "inode64"   /* inodes can be allocated anywhere */
1651 #define MNTOPT_IKEEP    "ikeep"         /* do not free empty inode clusters */
1652 #define MNTOPT_NOIKEEP  "noikeep"       /* free empty inode clusters */
1653 #define MNTOPT_LARGEIO     "largeio"    /* report large I/O sizes in stat() */
1654 #define MNTOPT_NOLARGEIO   "nolargeio"  /* do not report large I/O sizes
1655                                          * in stat(). */
1656 #define MNTOPT_ATTR2    "attr2"         /* do use attr2 attribute format */
1657 #define MNTOPT_NOATTR2  "noattr2"       /* do not use attr2 attribute format */
1658
1659 STATIC unsigned long
1660 suffix_strtoul(const char *cp, char **endp, unsigned int base)
1661 {
1662         int     last, shift_left_factor = 0;
1663         char    *value = (char *)cp;
1664
1665         last = strlen(value) - 1;
1666         if (value[last] == 'K' || value[last] == 'k') {
1667                 shift_left_factor = 10;
1668                 value[last] = '\0';
1669         }
1670         if (value[last] == 'M' || value[last] == 'm') {
1671                 shift_left_factor = 20;
1672                 value[last] = '\0';
1673         }
1674         if (value[last] == 'G' || value[last] == 'g') {
1675                 shift_left_factor = 30;
1676                 value[last] = '\0';
1677         }
1678
1679         return simple_strtoul(cp, endp, base) << shift_left_factor;
1680 }
1681
1682 STATIC int
1683 xfs_parseargs(
1684         struct bhv_desc         *bhv,
1685         char                    *options,
1686         struct xfs_mount_args   *args,
1687         int                     update)
1688 {
1689         struct vfs              *vfsp = bhvtovfs(bhv);
1690         char                    *this_char, *value, *eov;
1691         int                     dsunit, dswidth, vol_dsunit, vol_dswidth;
1692         int                     iosize;
1693
1694         args->flags |= XFSMNT_COMPAT_IOSIZE;
1695 #if 0   /* XXX: off by default, until some remaining issues ironed out */
1696         args->flags |= XFSMNT_IDELETE; /* default to on */
1697 #endif
1698
1699         if (!options)
1700                 return 0;
1701
1702         iosize = dsunit = dswidth = vol_dsunit = vol_dswidth = 0;
1703
1704         while ((this_char = strsep(&options, ",")) != NULL) {
1705                 if (!*this_char)
1706                         continue;
1707                 if ((value = strchr(this_char, '=')) != NULL)
1708                         *value++ = 0;
1709
1710                 if (!strcmp(this_char, MNTOPT_LOGBUFS)) {
1711                         if (!value || !*value) {
1712                                 printk("XFS: %s option requires an argument\n",
1713                                         this_char);
1714                                 return EINVAL;
1715                         }
1716                         args->logbufs = simple_strtoul(value, &eov, 10);
1717                 } else if (!strcmp(this_char, MNTOPT_LOGBSIZE)) {
1718                         if (!value || !*value) {
1719                                 printk("XFS: %s option requires an argument\n",
1720                                         this_char);
1721                                 return EINVAL;
1722                         }
1723                         args->logbufsize = suffix_strtoul(value, &eov, 10);
1724                 } else if (!strcmp(this_char, MNTOPT_LOGDEV)) {
1725                         if (!value || !*value) {
1726                                 printk("XFS: %s option requires an argument\n",
1727                                         this_char);
1728                                 return EINVAL;
1729                         }
1730                         strncpy(args->logname, value, MAXNAMELEN);
1731                 } else if (!strcmp(this_char, MNTOPT_MTPT)) {
1732                         if (!value || !*value) {
1733                                 printk("XFS: %s option requires an argument\n",
1734                                         this_char);
1735                                 return EINVAL;
1736                         }
1737                         strncpy(args->mtpt, value, MAXNAMELEN);
1738                 } else if (!strcmp(this_char, MNTOPT_RTDEV)) {
1739                         if (!value || !*value) {
1740                                 printk("XFS: %s option requires an argument\n",
1741                                         this_char);
1742                                 return EINVAL;
1743                         }
1744                         strncpy(args->rtname, value, MAXNAMELEN);
1745                 } else if (!strcmp(this_char, MNTOPT_BIOSIZE)) {
1746                         if (!value || !*value) {
1747                                 printk("XFS: %s option requires an argument\n",
1748                                         this_char);
1749                                 return EINVAL;
1750                         }
1751                         iosize = simple_strtoul(value, &eov, 10);
1752                         args->flags |= XFSMNT_IOSIZE;
1753                         args->iosizelog = (uint8_t) iosize;
1754                 } else if (!strcmp(this_char, MNTOPT_ALLOCSIZE)) {
1755                         if (!value || !*value) {
1756                                 printk("XFS: %s option requires an argument\n",
1757                                         this_char);
1758                                 return EINVAL;
1759                         }
1760                         iosize = suffix_strtoul(value, &eov, 10);
1761                         args->flags |= XFSMNT_IOSIZE;
1762                         args->iosizelog = ffs(iosize) - 1;
1763                 } else if (!strcmp(this_char, MNTOPT_IHASHSIZE)) {
1764                         if (!value || !*value) {
1765                                 printk("XFS: %s option requires an argument\n",
1766                                         this_char);
1767                                 return EINVAL;
1768                         }
1769                         args->flags |= XFSMNT_IHASHSIZE;
1770                         args->ihashsize = simple_strtoul(value, &eov, 10);
1771                 } else if (!strcmp(this_char, MNTOPT_GRPID) ||
1772                            !strcmp(this_char, MNTOPT_BSDGROUPS)) {
1773                         vfsp->vfs_flag |= VFS_GRPID;
1774                 } else if (!strcmp(this_char, MNTOPT_NOGRPID) ||
1775                            !strcmp(this_char, MNTOPT_SYSVGROUPS)) {
1776                         vfsp->vfs_flag &= ~VFS_GRPID;
1777                 } else if (!strcmp(this_char, MNTOPT_WSYNC)) {
1778                         args->flags |= XFSMNT_WSYNC;
1779                 } else if (!strcmp(this_char, MNTOPT_OSYNCISOSYNC)) {
1780                         args->flags |= XFSMNT_OSYNCISOSYNC;
1781                 } else if (!strcmp(this_char, MNTOPT_NORECOVERY)) {
1782                         args->flags |= XFSMNT_NORECOVERY;
1783                 } else if (!strcmp(this_char, MNTOPT_INO64)) {
1784                         args->flags |= XFSMNT_INO64;
1785 #if !XFS_BIG_INUMS
1786                         printk("XFS: %s option not allowed on this system\n",
1787                                 this_char);
1788                         return EINVAL;
1789 #endif
1790                 } else if (!strcmp(this_char, MNTOPT_NOALIGN)) {
1791                         args->flags |= XFSMNT_NOALIGN;
1792                 } else if (!strcmp(this_char, MNTOPT_SWALLOC)) {
1793                         args->flags |= XFSMNT_SWALLOC;
1794                 } else if (!strcmp(this_char, MNTOPT_SUNIT)) {
1795                         if (!value || !*value) {
1796                                 printk("XFS: %s option requires an argument\n",
1797                                         this_char);
1798                                 return EINVAL;
1799                         }
1800                         dsunit = simple_strtoul(value, &eov, 10);
1801                 } else if (!strcmp(this_char, MNTOPT_SWIDTH)) {
1802                         if (!value || !*value) {
1803                                 printk("XFS: %s option requires an argument\n",
1804                                         this_char);
1805                                 return EINVAL;
1806                         }
1807                         dswidth = simple_strtoul(value, &eov, 10);
1808                 } else if (!strcmp(this_char, MNTOPT_64BITINODE)) {
1809                         args->flags &= ~XFSMNT_32BITINODES;
1810 #if !XFS_BIG_INUMS
1811                         printk("XFS: %s option not allowed on this system\n",
1812                                 this_char);
1813                         return EINVAL;
1814 #endif
1815                 } else if (!strcmp(this_char, MNTOPT_NOUUID)) {
1816                         args->flags |= XFSMNT_NOUUID;
1817                 } else if (!strcmp(this_char, MNTOPT_BARRIER)) {
1818                         args->flags |= XFSMNT_BARRIER;
1819                 } else if (!strcmp(this_char, MNTOPT_IKEEP)) {
1820                         args->flags &= ~XFSMNT_IDELETE;
1821                 } else if (!strcmp(this_char, MNTOPT_NOIKEEP)) {
1822                         args->flags |= XFSMNT_IDELETE;
1823                 } else if (!strcmp(this_char, MNTOPT_LARGEIO)) {
1824                         args->flags &= ~XFSMNT_COMPAT_IOSIZE;
1825                 } else if (!strcmp(this_char, MNTOPT_NOLARGEIO)) {
1826                         args->flags |= XFSMNT_COMPAT_IOSIZE;
1827                 } else if (!strcmp(this_char, MNTOPT_ATTR2)) {
1828                         args->flags &= ~XFSMNT_COMPAT_ATTR;
1829                 } else if (!strcmp(this_char, MNTOPT_NOATTR2)) {
1830                         args->flags |= XFSMNT_COMPAT_ATTR;
1831                 } else if (!strcmp(this_char, "osyncisdsync")) {
1832                         /* no-op, this is now the default */
1833 printk("XFS: osyncisdsync is now the default, option is deprecated.\n");
1834                 } else if (!strcmp(this_char, "irixsgid")) {
1835 printk("XFS: irixsgid is now a sysctl(2) variable, option is deprecated.\n");
1836                 } else {
1837                         printk("XFS: unknown mount option [%s].\n", this_char);
1838                         return EINVAL;
1839                 }
1840         }
1841
1842         if (args->flags & XFSMNT_NORECOVERY) {
1843                 if ((vfsp->vfs_flag & VFS_RDONLY) == 0) {
1844                         printk("XFS: no-recovery mounts must be read-only.\n");
1845                         return EINVAL;
1846                 }
1847         }
1848
1849         if ((args->flags & XFSMNT_NOALIGN) && (dsunit || dswidth)) {
1850                 printk(
1851         "XFS: sunit and swidth options incompatible with the noalign option\n");
1852                 return EINVAL;
1853         }
1854
1855         if ((dsunit && !dswidth) || (!dsunit && dswidth)) {
1856                 printk("XFS: sunit and swidth must be specified together\n");
1857                 return EINVAL;
1858         }
1859
1860         if (dsunit && (dswidth % dsunit != 0)) {
1861                 printk(
1862         "XFS: stripe width (%d) must be a multiple of the stripe unit (%d)\n",
1863                         dswidth, dsunit);
1864                 return EINVAL;
1865         }
1866
1867         if ((args->flags & XFSMNT_NOALIGN) != XFSMNT_NOALIGN) {
1868                 if (dsunit) {
1869                         args->sunit = dsunit;
1870                         args->flags |= XFSMNT_RETERR;
1871                 } else {
1872                         args->sunit = vol_dsunit;
1873                 }
1874                 dswidth ? (args->swidth = dswidth) :
1875                           (args->swidth = vol_dswidth);
1876         } else {
1877                 args->sunit = args->swidth = 0;
1878         }
1879
1880         return 0;
1881 }
1882
1883 STATIC int
1884 xfs_showargs(
1885         struct bhv_desc         *bhv,
1886         struct seq_file         *m)
1887 {
1888         static struct proc_xfs_info {
1889                 int     flag;
1890                 char    *str;
1891         } xfs_info[] = {
1892                 /* the few simple ones we can get from the mount struct */
1893                 { XFS_MOUNT_WSYNC,              "," MNTOPT_WSYNC },
1894                 { XFS_MOUNT_INO64,              "," MNTOPT_INO64 },
1895                 { XFS_MOUNT_NOALIGN,            "," MNTOPT_NOALIGN },
1896                 { XFS_MOUNT_SWALLOC,            "," MNTOPT_SWALLOC },
1897                 { XFS_MOUNT_NOUUID,             "," MNTOPT_NOUUID },
1898                 { XFS_MOUNT_NORECOVERY,         "," MNTOPT_NORECOVERY },
1899                 { XFS_MOUNT_OSYNCISOSYNC,       "," MNTOPT_OSYNCISOSYNC },
1900                 { XFS_MOUNT_BARRIER,            "," MNTOPT_BARRIER },
1901                 { XFS_MOUNT_IDELETE,            "," MNTOPT_NOIKEEP },
1902                 { 0, NULL }
1903         };
1904         struct proc_xfs_info    *xfs_infop;
1905         struct xfs_mount        *mp = XFS_BHVTOM(bhv);
1906         struct vfs              *vfsp = XFS_MTOVFS(mp);
1907
1908         for (xfs_infop = xfs_info; xfs_infop->flag; xfs_infop++) {
1909                 if (mp->m_flags & xfs_infop->flag)
1910                         seq_puts(m, xfs_infop->str);
1911         }
1912
1913         if (mp->m_flags & XFS_MOUNT_IHASHSIZE)
1914                 seq_printf(m, "," MNTOPT_IHASHSIZE "=%d", mp->m_ihsize);
1915
1916         if (mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)
1917                 seq_printf(m, "," MNTOPT_ALLOCSIZE "=%d", 1<<mp->m_writeio_log);
1918
1919         if (mp->m_logbufs > 0)
1920                 seq_printf(m, "," MNTOPT_LOGBUFS "=%d", mp->m_logbufs);
1921
1922         if (mp->m_logbsize > 0)
1923                 seq_printf(m, "," MNTOPT_LOGBSIZE "=%d", mp->m_logbsize);
1924
1925         if (mp->m_logname)
1926                 seq_printf(m, "," MNTOPT_LOGDEV "=%s", mp->m_logname);
1927
1928         if (mp->m_rtname)
1929                 seq_printf(m, "," MNTOPT_RTDEV "=%s", mp->m_rtname);
1930
1931         if (mp->m_dalign > 0)
1932                 seq_printf(m, "," MNTOPT_SUNIT "=%d",
1933                                 (int)XFS_FSB_TO_BB(mp, mp->m_dalign));
1934
1935         if (mp->m_swidth > 0)
1936                 seq_printf(m, "," MNTOPT_SWIDTH "=%d",
1937                                 (int)XFS_FSB_TO_BB(mp, mp->m_swidth));
1938
1939         if (!(mp->m_flags & XFS_MOUNT_32BITINOOPT))
1940                 seq_printf(m, "," MNTOPT_64BITINODE);
1941
1942         if (vfsp->vfs_flag & VFS_GRPID)
1943                 seq_printf(m, "," MNTOPT_GRPID);
1944
1945         return 0;
1946 }
1947
1948 STATIC void
1949 xfs_freeze(
1950         bhv_desc_t      *bdp)
1951 {
1952         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
1953
1954         while (atomic_read(&mp->m_active_trans) > 0)
1955                 delay(100);
1956
1957         /* Push the superblock and write an unmount record */
1958         xfs_log_unmount_write(mp);
1959         xfs_unmountfs_writesb(mp);
1960 }
1961
1962
1963 vfsops_t xfs_vfsops = {
1964         BHV_IDENTITY_INIT(VFS_BHV_XFS,VFS_POSITION_XFS),
1965         .vfs_parseargs          = xfs_parseargs,
1966         .vfs_showargs           = xfs_showargs,
1967         .vfs_mount              = xfs_mount,
1968         .vfs_unmount            = xfs_unmount,
1969         .vfs_mntupdate          = xfs_mntupdate,
1970         .vfs_root               = xfs_root,
1971         .vfs_statvfs            = xfs_statvfs,
1972         .vfs_sync               = xfs_sync,
1973         .vfs_vget               = xfs_vget,
1974         .vfs_dmapiops           = (vfs_dmapiops_t)fs_nosys,
1975         .vfs_quotactl           = (vfs_quotactl_t)fs_nosys,
1976         .vfs_init_vnode         = xfs_initialize_vnode,
1977         .vfs_force_shutdown     = xfs_do_force_shutdown,
1978         .vfs_freeze             = xfs_freeze,
1979 };