fs/xfs/xfs_vfsops.c

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