git.oblomov.eu Git - linux-2.6/blob - fs/xfs/xfs_vfsops.c

   1 /*
   2  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
   3  * All Rights Reserved.
   4  *
   5  * This program is free software; you can redistribute it and/or
   6  * modify it under the terms of the GNU General Public License as
   7  * published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it would be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write the Free Software Foundation,
  16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17  */
  18 #include "xfs.h"
  19 #include "xfs_fs.h"
  20 #include "xfs_types.h"
  21 #include "xfs_bit.h"
  22 #include "xfs_log.h"
  23 #include "xfs_inum.h"
  24 #include "xfs_trans.h"
  25 #include "xfs_sb.h"
  26 #include "xfs_ag.h"
  27 #include "xfs_dir2.h"
  28 #include "xfs_dmapi.h"
  29 #include "xfs_mount.h"
  30 #include "xfs_da_btree.h"
  31 #include "xfs_bmap_btree.h"
  32 #include "xfs_ialloc_btree.h"
  33 #include "xfs_alloc_btree.h"
  34 #include "xfs_dir2_sf.h"
  35 #include "xfs_attr_sf.h"
  36 #include "xfs_dinode.h"
  37 #include "xfs_inode.h"
  38 #include "xfs_inode_item.h"
  39 #include "xfs_btree.h"
  40 #include "xfs_alloc.h"
  41 #include "xfs_ialloc.h"
  42 #include "xfs_quota.h"
  43 #include "xfs_error.h"
  44 #include "xfs_bmap.h"
  45 #include "xfs_rw.h"
  46 #include "xfs_refcache.h"
  47 #include "xfs_buf_item.h"
  48 #include "xfs_log_priv.h"
  49 #include "xfs_dir2_trace.h"
  50 #include "xfs_extfree_item.h"
  51 #include "xfs_acl.h"
  52 #include "xfs_attr.h"
  53 #include "xfs_clnt.h"
  54 #include "xfs_fsops.h"
  55
  56 STATIC int      xfs_sync(bhv_desc_t *, int, cred_t *);
  57
  58 int
  59 xfs_init(void)
  60 {
  61         extern kmem_zone_t      *xfs_bmap_free_item_zone;
  62         extern kmem_zone_t      *xfs_btree_cur_zone;
  63         extern kmem_zone_t      *xfs_trans_zone;
  64         extern kmem_zone_t      *xfs_buf_item_zone;
  65         extern kmem_zone_t      *xfs_dabuf_zone;
  66 #ifdef XFS_DABUF_DEBUG
  67         extern lock_t           xfs_dabuf_global_lock;
  68         spinlock_init(&xfs_dabuf_global_lock, "xfsda");
  69 #endif
  70
  71         /*
  72          * Initialize all of the zone allocators we use.
  73          */
  74         xfs_bmap_free_item_zone = kmem_zone_init(sizeof(xfs_bmap_free_item_t),
  75                                                  "xfs_bmap_free_item");
  76         xfs_btree_cur_zone = kmem_zone_init(sizeof(xfs_btree_cur_t),
  77                                             "xfs_btree_cur");
  78         xfs_trans_zone = kmem_zone_init(sizeof(xfs_trans_t), "xfs_trans");
  79         xfs_da_state_zone =
  80                 kmem_zone_init(sizeof(xfs_da_state_t), "xfs_da_state");
  81         xfs_dabuf_zone = kmem_zone_init(sizeof(xfs_dabuf_t), "xfs_dabuf");
  82         xfs_ifork_zone = kmem_zone_init(sizeof(xfs_ifork_t), "xfs_ifork");
  83         xfs_acl_zone_init(xfs_acl_zone, "xfs_acl");
  84
  85         /*
  86          * The size of the zone allocated buf log item is the maximum
  87          * size possible under XFS.  This wastes a little bit of memory,
  88          * but it is much faster.
  89          */
  90         xfs_buf_item_zone =
  91                 kmem_zone_init((sizeof(xfs_buf_log_item_t) +
  92                                 (((XFS_MAX_BLOCKSIZE / XFS_BLI_CHUNK) /
  93                                   NBWORD) * sizeof(int))),
  94                                "xfs_buf_item");
  95         xfs_efd_zone =
  96                 kmem_zone_init((sizeof(xfs_efd_log_item_t) +
  97                                ((XFS_EFD_MAX_FAST_EXTENTS - 1) *
  98                                  sizeof(xfs_extent_t))),
  99                                       "xfs_efd_item");
 100         xfs_efi_zone =
 101                 kmem_zone_init((sizeof(xfs_efi_log_item_t) +
 102                                ((XFS_EFI_MAX_FAST_EXTENTS - 1) *
 103                                  sizeof(xfs_extent_t))),
 104                                       "xfs_efi_item");
 105
 106         /*
 107          * These zones warrant special memory allocator hints
 108          */
 109         xfs_inode_zone =
 110                 kmem_zone_init_flags(sizeof(xfs_inode_t), "xfs_inode",
 111                                         KM_ZONE_HWALIGN | KM_ZONE_RECLAIM |
 112                                         KM_ZONE_SPREAD, NULL);
 113         xfs_ili_zone =
 114                 kmem_zone_init_flags(sizeof(xfs_inode_log_item_t), "xfs_ili",
 115                                         KM_ZONE_SPREAD, NULL);
 116         xfs_chashlist_zone =
 117                 kmem_zone_init_flags(sizeof(xfs_chashlist_t), "xfs_chashlist",
 118                                         KM_ZONE_SPREAD, NULL);
 119
 120         /*
 121          * Allocate global trace buffers.
 122          */
 123 #ifdef XFS_ALLOC_TRACE
 124         xfs_alloc_trace_buf = ktrace_alloc(XFS_ALLOC_TRACE_SIZE, KM_SLEEP);
 125 #endif
 126 #ifdef XFS_BMAP_TRACE
 127         xfs_bmap_trace_buf = ktrace_alloc(XFS_BMAP_TRACE_SIZE, KM_SLEEP);
 128 #endif
 129 #ifdef XFS_BMBT_TRACE
 130         xfs_bmbt_trace_buf = ktrace_alloc(XFS_BMBT_TRACE_SIZE, KM_SLEEP);
 131 #endif
 132 #ifdef XFS_ATTR_TRACE
 133         xfs_attr_trace_buf = ktrace_alloc(XFS_ATTR_TRACE_SIZE, KM_SLEEP);
 134 #endif
 135 #ifdef XFS_DIR2_TRACE
 136         xfs_dir2_trace_buf = ktrace_alloc(XFS_DIR2_GTRACE_SIZE, KM_SLEEP);
 137 #endif
 138
 139         xfs_dir_startup();
 140
 141 #if (defined(DEBUG) || defined(INDUCE_IO_ERROR))
 142         xfs_error_test_init();
 143 #endif /* DEBUG || INDUCE_IO_ERROR */
 144
 145         xfs_init_procfs();
 146         xfs_sysctl_register();
 147         return 0;
 148 }
 149
 150 void
 151 xfs_cleanup(void)
 152 {
 153         extern kmem_zone_t      *xfs_bmap_free_item_zone;
 154         extern kmem_zone_t      *xfs_btree_cur_zone;
 155         extern kmem_zone_t      *xfs_inode_zone;
 156         extern kmem_zone_t      *xfs_trans_zone;
 157         extern kmem_zone_t      *xfs_da_state_zone;
 158         extern kmem_zone_t      *xfs_dabuf_zone;
 159         extern kmem_zone_t      *xfs_efd_zone;
 160         extern kmem_zone_t      *xfs_efi_zone;
 161         extern kmem_zone_t      *xfs_buf_item_zone;
 162         extern kmem_zone_t      *xfs_chashlist_zone;
 163
 164         xfs_cleanup_procfs();
 165         xfs_sysctl_unregister();
 166         xfs_refcache_destroy();
 167         xfs_acl_zone_destroy(xfs_acl_zone);
 168
 169 #ifdef XFS_DIR2_TRACE
 170         ktrace_free(xfs_dir2_trace_buf);
 171 #endif
 172 #ifdef XFS_ATTR_TRACE
 173         ktrace_free(xfs_attr_trace_buf);
 174 #endif
 175 #ifdef XFS_BMBT_TRACE
 176         ktrace_free(xfs_bmbt_trace_buf);
 177 #endif
 178 #ifdef XFS_BMAP_TRACE
 179         ktrace_free(xfs_bmap_trace_buf);
 180 #endif
 181 #ifdef XFS_ALLOC_TRACE
 182         ktrace_free(xfs_alloc_trace_buf);
 183 #endif
 184
 185         kmem_zone_destroy(xfs_bmap_free_item_zone);
 186         kmem_zone_destroy(xfs_btree_cur_zone);
 187         kmem_zone_destroy(xfs_inode_zone);
 188         kmem_zone_destroy(xfs_trans_zone);
 189         kmem_zone_destroy(xfs_da_state_zone);
 190         kmem_zone_destroy(xfs_dabuf_zone);
 191         kmem_zone_destroy(xfs_buf_item_zone);
 192         kmem_zone_destroy(xfs_efd_zone);
 193         kmem_zone_destroy(xfs_efi_zone);
 194         kmem_zone_destroy(xfs_ifork_zone);
 195         kmem_zone_destroy(xfs_ili_zone);
 196         kmem_zone_destroy(xfs_chashlist_zone);
 197 }
 198
 199 /*
 200  * xfs_start_flags
 201  *
 202  * This function fills in xfs_mount_t fields based on mount args.
 203  * Note: the superblock has _not_ yet been read in.
 204  */
 205 STATIC int
 206 xfs_start_flags(
 207         struct bhv_vfs          *vfs,
 208         struct xfs_mount_args   *ap,
 209         struct xfs_mount        *mp)
 210 {
 211         /* Values are in BBs */
 212         if ((ap->flags & XFSMNT_NOALIGN) != XFSMNT_NOALIGN) {
 213                 /*
 214                  * At this point the superblock has not been read
 215                  * in, therefore we do not know the block size.
 216                  * Before the mount call ends we will convert
 217                  * these to FSBs.
 218                  */
 219                 mp->m_dalign = ap->sunit;
 220                 mp->m_swidth = ap->swidth;
 221         }
 222
 223         if (ap->logbufs != -1 &&
 224             ap->logbufs != 0 &&
 225             (ap->logbufs < XLOG_MIN_ICLOGS ||
 226              ap->logbufs > XLOG_MAX_ICLOGS)) {
 227                 cmn_err(CE_WARN,
 228                         "XFS: invalid logbufs value: %d [not %d-%d]",
 229                         ap->logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
 230                 return XFS_ERROR(EINVAL);
 231         }
 232         mp->m_logbufs = ap->logbufs;
 233         if (ap->logbufsize != -1 &&
 234             ap->logbufsize !=  0 &&
 235             ap->logbufsize != 16 * 1024 &&
 236             ap->logbufsize != 32 * 1024 &&
 237             ap->logbufsize != 64 * 1024 &&
 238             ap->logbufsize != 128 * 1024 &&
 239             ap->logbufsize != 256 * 1024) {
 240                 cmn_err(CE_WARN,
 241         "XFS: invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
 242                         ap->logbufsize);
 243                 return XFS_ERROR(EINVAL);
 244         }
 245         mp->m_ihsize = ap->ihashsize;
 246         mp->m_logbsize = ap->logbufsize;
 247         mp->m_fsname_len = strlen(ap->fsname) + 1;
 248         mp->m_fsname = kmem_alloc(mp->m_fsname_len, KM_SLEEP);
 249         strcpy(mp->m_fsname, ap->fsname);
 250         if (ap->rtname[0]) {
 251                 mp->m_rtname = kmem_alloc(strlen(ap->rtname) + 1, KM_SLEEP);
 252                 strcpy(mp->m_rtname, ap->rtname);
 253         }
 254         if (ap->logname[0]) {
 255                 mp->m_logname = kmem_alloc(strlen(ap->logname) + 1, KM_SLEEP);
 256                 strcpy(mp->m_logname, ap->logname);
 257         }
 258
 259         if (ap->flags & XFSMNT_WSYNC)
 260                 mp->m_flags |= XFS_MOUNT_WSYNC;
 261 #if XFS_BIG_INUMS
 262         if (ap->flags & XFSMNT_INO64) {
 263                 mp->m_flags |= XFS_MOUNT_INO64;
 264                 mp->m_inoadd = XFS_INO64_OFFSET;
 265         }
 266 #endif
 267         if (ap->flags & XFSMNT_RETERR)
 268                 mp->m_flags |= XFS_MOUNT_RETERR;
 269         if (ap->flags & XFSMNT_NOALIGN)
 270                 mp->m_flags |= XFS_MOUNT_NOALIGN;
 271         if (ap->flags & XFSMNT_SWALLOC)
 272                 mp->m_flags |= XFS_MOUNT_SWALLOC;
 273         if (ap->flags & XFSMNT_OSYNCISOSYNC)
 274                 mp->m_flags |= XFS_MOUNT_OSYNCISOSYNC;
 275         if (ap->flags & XFSMNT_32BITINODES)
 276                 mp->m_flags |= XFS_MOUNT_32BITINODES;
 277
 278         if (ap->flags & XFSMNT_IOSIZE) {
 279                 if (ap->iosizelog > XFS_MAX_IO_LOG ||
 280                     ap->iosizelog < XFS_MIN_IO_LOG) {
 281                         cmn_err(CE_WARN,
 282                 "XFS: invalid log iosize: %d [not %d-%d]",
 283                                 ap->iosizelog, XFS_MIN_IO_LOG,
 284                                 XFS_MAX_IO_LOG);
 285                         return XFS_ERROR(EINVAL);
 286                 }
 287
 288                 mp->m_flags |= XFS_MOUNT_DFLT_IOSIZE;
 289                 mp->m_readio_log = mp->m_writeio_log = ap->iosizelog;
 290         }
 291
 292         if (ap->flags & XFSMNT_IHASHSIZE)
 293                 mp->m_flags |= XFS_MOUNT_IHASHSIZE;
 294         if (ap->flags & XFSMNT_IDELETE)
 295                 mp->m_flags |= XFS_MOUNT_IDELETE;
 296         if (ap->flags & XFSMNT_DIRSYNC)
 297                 mp->m_flags |= XFS_MOUNT_DIRSYNC;
 298         if (ap->flags & XFSMNT_ATTR2)
 299                 mp->m_flags |= XFS_MOUNT_ATTR2;
 300
 301         if (ap->flags2 & XFSMNT2_COMPAT_IOSIZE)
 302                 mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE;
 303
 304         /*
 305          * no recovery flag requires a read-only mount
 306          */
 307         if (ap->flags & XFSMNT_NORECOVERY) {
 308                 if (!(vfs->vfs_flag & VFS_RDONLY)) {
 309                         cmn_err(CE_WARN,
 310         "XFS: tried to mount a FS read-write without recovery!");
 311                         return XFS_ERROR(EINVAL);
 312                 }
 313                 mp->m_flags |= XFS_MOUNT_NORECOVERY;
 314         }
 315
 316         if (ap->flags & XFSMNT_NOUUID)
 317                 mp->m_flags |= XFS_MOUNT_NOUUID;
 318         if (ap->flags & XFSMNT_BARRIER)
 319                 mp->m_flags |= XFS_MOUNT_BARRIER;
 320         else
 321                 mp->m_flags &= ~XFS_MOUNT_BARRIER;
 322
 323         return 0;
 324 }
 325
 326 /*
 327  * This function fills in xfs_mount_t fields based on mount args.
 328  * Note: the superblock _has_ now been read in.
 329  */
 330 STATIC int
 331 xfs_finish_flags(
 332         struct bhv_vfs          *vfs,
 333         struct xfs_mount_args   *ap,
 334         struct xfs_mount        *mp)
 335 {
 336         int                     ronly = (vfs->vfs_flag & VFS_RDONLY);
 337
 338         /* Fail a mount where the logbuf is smaller then the log stripe */
 339         if (XFS_SB_VERSION_HASLOGV2(&mp->m_sb)) {
 340                 if ((ap->logbufsize <= 0) &&
 341                     (mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE)) {
 342                         mp->m_logbsize = mp->m_sb.sb_logsunit;
 343                 } else if (ap->logbufsize > 0 &&
 344                            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         if (XFS_SB_VERSION_HASATTR2(&mp->m_sb)) {
 359                 mp->m_flags |= XFS_MOUNT_ATTR2;
 360         }
 361
 362         /*
 363          * prohibit r/w mounts of read-only filesystems
 364          */
 365         if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !ronly) {
 366                 cmn_err(CE_WARN,
 367         "XFS: cannot mount a read-only filesystem as read-write");
 368                 return XFS_ERROR(EROFS);
 369         }
 370
 371         /*
 372          * check for shared mount.
 373          */
 374         if (ap->flags & XFSMNT_SHARED) {
 375                 if (!XFS_SB_VERSION_HASSHARED(&mp->m_sb))
 376                         return XFS_ERROR(EINVAL);
 377
 378                 /*
 379                  * For IRIX 6.5, shared mounts must have the shared
 380                  * version bit set, have the persistent readonly
 381                  * field set, must be version 0 and can only be mounted
 382                  * read-only.
 383                  */
 384                 if (!ronly || !(mp->m_sb.sb_flags & XFS_SBF_READONLY) ||
 385                      (mp->m_sb.sb_shared_vn != 0))
 386                         return XFS_ERROR(EINVAL);
 387
 388                 mp->m_flags |= XFS_MOUNT_SHARED;
 389
 390                 /*
 391                  * Shared XFS V0 can't deal with DMI.  Return EINVAL.
 392                  */
 393                 if (mp->m_sb.sb_shared_vn == 0 && (ap->flags & XFSMNT_DMAPI))
 394                         return XFS_ERROR(EINVAL);
 395         }
 396
 397         return 0;
 398 }
 399
 400 /*
 401  * xfs_mount
 402  *
 403  * The file system configurations are:
 404  *      (1) device (partition) with data and internal log
 405  *      (2) logical volume with data and log subvolumes.
 406  *      (3) logical volume with data, log, and realtime subvolumes.
 407  *
 408  * We only have to handle opening the log and realtime volumes here if
 409  * they are present.  The data subvolume has already been opened by
 410  * get_sb_bdev() and is stored in vfsp->vfs_super->s_bdev.
 411  */
 412 STATIC int
 413 xfs_mount(
 414         struct bhv_desc         *bhvp,
 415         struct xfs_mount_args   *args,
 416         cred_t                  *credp)
 417 {
 418         struct bhv_vfs          *vfsp = bhvtovfs(bhvp);
 419         struct bhv_desc         *p;
 420         struct xfs_mount        *mp = XFS_BHVTOM(bhvp);
 421         struct block_device     *ddev, *logdev, *rtdev;
 422         int                     flags = 0, error;
 423
 424         ddev = vfsp->vfs_super->s_bdev;
 425         logdev = rtdev = NULL;
 426
 427         /*
 428          * Setup xfs_mount function vectors from available behaviors
 429          */
 430         p = vfs_bhv_lookup(vfsp, VFS_POSITION_DM);
 431         mp->m_dm_ops = p ? *(xfs_dmops_t *) vfs_bhv_custom(p) : xfs_dmcore_stub;
 432         p = vfs_bhv_lookup(vfsp, VFS_POSITION_QM);
 433         mp->m_qm_ops = p ? *(xfs_qmops_t *) vfs_bhv_custom(p) : xfs_qmcore_stub;
 434         p = vfs_bhv_lookup(vfsp, VFS_POSITION_IO);
 435         mp->m_io_ops = p ? *(xfs_ioops_t *) vfs_bhv_custom(p) : xfs_iocore_xfs;
 436
 437         if (args->flags & XFSMNT_QUIET)
 438                 flags |= XFS_MFSI_QUIET;
 439
 440         /*
 441          * Open real time and log devices - order is important.
 442          */
 443         if (args->logname[0]) {
 444                 error = xfs_blkdev_get(mp, args->logname, &logdev);
 445                 if (error)
 446                         return error;
 447         }
 448         if (args->rtname[0]) {
 449                 error = xfs_blkdev_get(mp, args->rtname, &rtdev);
 450                 if (error) {
 451                         xfs_blkdev_put(logdev);
 452                         return error;
 453                 }
 454
 455                 if (rtdev == ddev || rtdev == logdev) {
 456                         cmn_err(CE_WARN,
 457         "XFS: Cannot mount filesystem with identical rtdev and ddev/logdev.");
 458                         xfs_blkdev_put(logdev);
 459                         xfs_blkdev_put(rtdev);
 460                         return EINVAL;
 461                 }
 462         }
 463
 464         /*
 465          * Setup xfs_mount buffer target pointers
 466          */
 467         error = ENOMEM;
 468         mp->m_ddev_targp = xfs_alloc_buftarg(ddev, 0);
 469         if (!mp->m_ddev_targp) {
 470                 xfs_blkdev_put(logdev);
 471                 xfs_blkdev_put(rtdev);
 472                 return error;
 473         }
 474         if (rtdev) {
 475                 mp->m_rtdev_targp = xfs_alloc_buftarg(rtdev, 1);
 476                 if (!mp->m_rtdev_targp)
 477                         goto error0;
 478         }
 479         mp->m_logdev_targp = (logdev && logdev != ddev) ?
 480                                 xfs_alloc_buftarg(logdev, 1) : mp->m_ddev_targp;
 481         if (!mp->m_logdev_targp)
 482                 goto error0;
 483
 484         /*
 485          * Setup flags based on mount(2) options and then the superblock
 486          */
 487         error = xfs_start_flags(vfsp, args, mp);
 488         if (error)
 489                 goto error1;
 490         error = xfs_readsb(mp, flags);
 491         if (error)
 492                 goto error1;
 493         error = xfs_finish_flags(vfsp, args, mp);
 494         if (error)
 495                 goto error2;
 496
 497         /*
 498          * Setup xfs_mount buffer target pointers based on superblock
 499          */
 500         error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_blocksize,
 501                                     mp->m_sb.sb_sectsize);
 502         if (!error && logdev && logdev != ddev) {
 503                 unsigned int    log_sector_size = BBSIZE;
 504
 505                 if (XFS_SB_VERSION_HASSECTOR(&mp->m_sb))
 506                         log_sector_size = mp->m_sb.sb_logsectsize;
 507                 error = xfs_setsize_buftarg(mp->m_logdev_targp,
 508                                             mp->m_sb.sb_blocksize,
 509                                             log_sector_size);
 510         }
 511         if (!error && rtdev)
 512                 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
 513                                             mp->m_sb.sb_blocksize,
 514                                             mp->m_sb.sb_sectsize);
 515         if (error)
 516                 goto error2;
 517
 518         if (mp->m_flags & XFS_MOUNT_BARRIER)
 519                 xfs_mountfs_check_barriers(mp);
 520
 521         error = XFS_IOINIT(vfsp, args, flags);
 522         if (error)
 523                 goto error2;
 524
 525         return 0;
 526
 527 error2:
 528         if (mp->m_sb_bp)
 529                 xfs_freesb(mp);
 530 error1:
 531         xfs_binval(mp->m_ddev_targp);
 532         if (logdev && logdev != ddev)
 533                 xfs_binval(mp->m_logdev_targp);
 534         if (rtdev)
 535                 xfs_binval(mp->m_rtdev_targp);
 536 error0:
 537         xfs_unmountfs_close(mp, credp);
 538         return error;
 539 }
 540
 541 STATIC int
 542 xfs_unmount(
 543         bhv_desc_t      *bdp,
 544         int             flags,
 545         cred_t          *credp)
 546 {
 547         bhv_vfs_t       *vfsp = bhvtovfs(bdp);
 548         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
 549         xfs_inode_t     *rip;
 550         bhv_vnode_t     *rvp;
 551         int             unmount_event_wanted = 0;
 552         int             unmount_event_flags = 0;
 553         int             xfs_unmountfs_needed = 0;
 554         int             error;
 555
 556         rip = mp->m_rootip;
 557         rvp = XFS_ITOV(rip);
 558
 559         if (vfsp->vfs_flag & VFS_DMI) {
 560                 error = XFS_SEND_PREUNMOUNT(mp, vfsp,
 561                                 rvp, DM_RIGHT_NULL, rvp, DM_RIGHT_NULL,
 562                                 NULL, NULL, 0, 0,
 563                                 (mp->m_dmevmask & (1<<DM_EVENT_PREUNMOUNT))?
 564                                         0:DM_FLAGS_UNWANTED);
 565                         if (error)
 566                                 return XFS_ERROR(error);
 567                 unmount_event_wanted = 1;
 568                 unmount_event_flags = (mp->m_dmevmask & (1<<DM_EVENT_UNMOUNT))?
 569                                         0 : DM_FLAGS_UNWANTED;
 570         }
 571
 572         /*
 573          * First blow any referenced inode from this file system
 574          * out of the reference cache, and delete the timer.
 575          */
 576         xfs_refcache_purge_mp(mp);
 577
 578         XFS_bflush(mp->m_ddev_targp);
 579         error = xfs_unmount_flush(mp, 0);
 580         if (error)
 581                 goto out;
 582
 583         ASSERT(vn_count(rvp) == 1);
 584
 585         /*
 586          * Drop the reference count
 587          */
 588         VN_RELE(rvp);
 589
 590         /*
 591          * If we're forcing a shutdown, typically because of a media error,
 592          * we want to make sure we invalidate dirty pages that belong to
 593          * referenced vnodes as well.
 594          */
 595         if (XFS_FORCED_SHUTDOWN(mp)) {
 596                 error = xfs_sync(&mp->m_bhv,
 597                          (SYNC_WAIT | SYNC_CLOSE), credp);
 598                 ASSERT(error != EFSCORRUPTED);
 599         }
 600         xfs_unmountfs_needed = 1;
 601
 602 out:
 603         /*      Send DMAPI event, if required.
 604          *      Then do xfs_unmountfs() if needed.
 605          *      Then return error (or zero).
 606          */
 607         if (unmount_event_wanted) {
 608                 /* Note: mp structure must still exist for
 609                  * XFS_SEND_UNMOUNT() call.
 610                  */
 611                 XFS_SEND_UNMOUNT(mp, vfsp, error == 0 ? rvp : NULL,
 612                         DM_RIGHT_NULL, 0, error, unmount_event_flags);
 613         }
 614         if (xfs_unmountfs_needed) {
 615                 /*
 616                  * Call common unmount function to flush to disk
 617                  * and free the super block buffer & mount structures.
 618                  */
 619                 xfs_unmountfs(mp, credp);
 620         }
 621
 622         return XFS_ERROR(error);
 623 }
 624
 625 STATIC int
 626 xfs_quiesce_fs(
 627         xfs_mount_t             *mp)
 628 {
 629         int                     count = 0, pincount;
 630
 631         xfs_refcache_purge_mp(mp);
 632         xfs_flush_buftarg(mp->m_ddev_targp, 0);
 633         xfs_finish_reclaim_all(mp, 0);
 634
 635         /* This loop must run at least twice.
 636          * The first instance of the loop will flush
 637          * most meta data but that will generate more
 638          * meta data (typically directory updates).
 639          * Which then must be flushed and logged before
 640          * we can write the unmount record.
 641          */
 642         do {
 643                 xfs_syncsub(mp, SYNC_REMOUNT|SYNC_ATTR|SYNC_WAIT, NULL);
 644                 pincount = xfs_flush_buftarg(mp->m_ddev_targp, 1);
 645                 if (!pincount) {
 646                         delay(50);
 647                         count++;
 648                 }
 649         } while (count < 2);
 650
 651         return 0;
 652 }
 653
 654 STATIC int
 655 xfs_mntupdate(
 656         bhv_desc_t                      *bdp,
 657         int                             *flags,
 658         struct xfs_mount_args           *args)
 659 {
 660         bhv_vfs_t       *vfsp = bhvtovfs(bdp);
 661         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
 662
 663         if (!(*flags & MS_RDONLY)) {                    /* rw/ro -> rw */
 664                 if (vfsp->vfs_flag & VFS_RDONLY)
 665                         vfsp->vfs_flag &= ~VFS_RDONLY;
 666                 if (args->flags & XFSMNT_BARRIER) {
 667                         mp->m_flags |= XFS_MOUNT_BARRIER;
 668                         xfs_mountfs_check_barriers(mp);
 669                 } else {
 670                         mp->m_flags &= ~XFS_MOUNT_BARRIER;
 671                 }
 672         } else if (!(vfsp->vfs_flag & VFS_RDONLY)) {    /* rw -> ro */
 673                 bhv_vfs_sync(vfsp, SYNC_FSDATA|SYNC_BDFLUSH|SYNC_ATTR, NULL);
 674                 xfs_quiesce_fs(mp);
 675                 xfs_log_unmount_write(mp);
 676                 xfs_unmountfs_writesb(mp);
 677                 vfsp->vfs_flag |= VFS_RDONLY;
 678         }
 679         return 0;
 680 }
 681
 682 /*
 683  * xfs_unmount_flush implements a set of flush operation on special
 684  * inodes, which are needed as a separate set of operations so that
 685  * they can be called as part of relocation process.
 686  */
 687 int
 688 xfs_unmount_flush(
 689         xfs_mount_t     *mp,            /* Mount structure we are getting
 690                                            rid of. */
 691         int             relocation)     /* Called from vfs relocation. */
 692 {
 693         xfs_inode_t     *rip = mp->m_rootip;
 694         xfs_inode_t     *rbmip;
 695         xfs_inode_t     *rsumip = NULL;
 696         bhv_vnode_t     *rvp = XFS_ITOV(rip);
 697         int             error;
 698
 699         xfs_ilock(rip, XFS_ILOCK_EXCL);
 700         xfs_iflock(rip);
 701
 702         /*
 703          * Flush out the real time inodes.
 704          */
 705         if ((rbmip = mp->m_rbmip) != NULL) {
 706                 xfs_ilock(rbmip, XFS_ILOCK_EXCL);
 707                 xfs_iflock(rbmip);
 708                 error = xfs_iflush(rbmip, XFS_IFLUSH_SYNC);
 709                 xfs_iunlock(rbmip, XFS_ILOCK_EXCL);
 710
 711                 if (error == EFSCORRUPTED)
 712                         goto fscorrupt_out;
 713
 714                 ASSERT(vn_count(XFS_ITOV(rbmip)) == 1);
 715
 716                 rsumip = mp->m_rsumip;
 717                 xfs_ilock(rsumip, XFS_ILOCK_EXCL);
 718                 xfs_iflock(rsumip);
 719                 error = xfs_iflush(rsumip, XFS_IFLUSH_SYNC);
 720                 xfs_iunlock(rsumip, XFS_ILOCK_EXCL);
 721
 722                 if (error == EFSCORRUPTED)
 723                         goto fscorrupt_out;
 724
 725                 ASSERT(vn_count(XFS_ITOV(rsumip)) == 1);
 726         }
 727
 728         /*
 729          * Synchronously flush root inode to disk
 730          */
 731         error = xfs_iflush(rip, XFS_IFLUSH_SYNC);
 732         if (error == EFSCORRUPTED)
 733                 goto fscorrupt_out2;
 734
 735         if (vn_count(rvp) != 1 && !relocation) {
 736                 xfs_iunlock(rip, XFS_ILOCK_EXCL);
 737                 return XFS_ERROR(EBUSY);
 738         }
 739
 740         /*
 741          * Release dquot that rootinode, rbmino and rsumino might be holding,
 742          * flush and purge the quota inodes.
 743          */
 744         error = XFS_QM_UNMOUNT(mp);
 745         if (error == EFSCORRUPTED)
 746                 goto fscorrupt_out2;
 747
 748         if (rbmip) {
 749                 VN_RELE(XFS_ITOV(rbmip));
 750                 VN_RELE(XFS_ITOV(rsumip));
 751         }
 752
 753         xfs_iunlock(rip, XFS_ILOCK_EXCL);
 754         return 0;
 755
 756 fscorrupt_out:
 757         xfs_ifunlock(rip);
 758
 759 fscorrupt_out2:
 760         xfs_iunlock(rip, XFS_ILOCK_EXCL);
 761
 762         return XFS_ERROR(EFSCORRUPTED);
 763 }
 764
 765 /*
 766  * xfs_root extracts the root vnode from a vfs.
 767  *
 768  * vfsp -- the vfs struct for the desired file system
 769  * vpp  -- address of the caller's vnode pointer which should be
 770  *         set to the desired fs root vnode
 771  */
 772 STATIC int
 773 xfs_root(
 774         bhv_desc_t      *bdp,
 775         bhv_vnode_t     **vpp)
 776 {
 777         bhv_vnode_t     *vp;
 778
 779         vp = XFS_ITOV((XFS_BHVTOM(bdp))->m_rootip);
 780         VN_HOLD(vp);
 781         *vpp = vp;
 782         return 0;
 783 }
 784
 785 /*
 786  * xfs_statvfs
 787  *
 788  * Fill in the statvfs structure for the given file system.  We use
 789  * the superblock lock in the mount structure to ensure a consistent
 790  * snapshot of the counters returned.
 791  */
 792 STATIC int
 793 xfs_statvfs(
 794         bhv_desc_t      *bdp,
 795         bhv_statvfs_t   *statp,
 796         bhv_vnode_t     *vp)
 797 {
 798         __uint64_t      fakeinos;
 799         xfs_extlen_t    lsize;
 800         xfs_mount_t     *mp;
 801         xfs_sb_t        *sbp;
 802         unsigned long   s;
 803
 804         mp = XFS_BHVTOM(bdp);
 805         sbp = &(mp->m_sb);
 806
 807         statp->f_type = XFS_SB_MAGIC;
 808
 809         xfs_icsb_sync_counters_flags(mp, XFS_ICSB_LAZY_COUNT);
 810         s = XFS_SB_LOCK(mp);
 811         statp->f_bsize = sbp->sb_blocksize;
 812         lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
 813         statp->f_blocks = sbp->sb_dblocks - lsize;
 814         statp->f_bfree = statp->f_bavail =
 815                                 sbp->sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
 816         fakeinos = statp->f_bfree << sbp->sb_inopblog;
 817 #if XFS_BIG_INUMS
 818         fakeinos += mp->m_inoadd;
 819 #endif
 820         statp->f_files =
 821             MIN(sbp->sb_icount + fakeinos, (__uint64_t)XFS_MAXINUMBER);
 822         if (mp->m_maxicount)
 823 #if XFS_BIG_INUMS
 824                 if (!mp->m_inoadd)
 825 #endif
 826                         statp->f_files = min_t(typeof(statp->f_files),
 827                                                 statp->f_files,
 828                                                 mp->m_maxicount);
 829         statp->f_ffree = statp->f_files - (sbp->sb_icount - sbp->sb_ifree);
 830         XFS_SB_UNLOCK(mp, s);
 831
 832         xfs_statvfs_fsid(statp, mp);
 833         statp->f_namelen = MAXNAMELEN - 1;
 834
 835         return 0;
 836 }
 837
 838
 839 /*
 840  * xfs_sync flushes any pending I/O to file system vfsp.
 841  *
 842  * This routine is called by vfs_sync() to make sure that things make it
 843  * out to disk eventually, on sync() system calls to flush out everything,
 844  * and when the file system is unmounted.  For the vfs_sync() case, all
 845  * we really need to do is sync out the log to make all of our meta-data
 846  * updates permanent (except for timestamps).  For calls from pflushd(),
 847  * dirty pages are kept moving by calling pdflush() on the inodes
 848  * containing them.  We also flush the inodes that we can lock without
 849  * sleeping and the superblock if we can lock it without sleeping from
 850  * vfs_sync() so that items at the tail of the log are always moving out.
 851  *
 852  * Flags:
 853  *      SYNC_BDFLUSH - We're being called from vfs_sync() so we don't want
 854  *                     to sleep if we can help it.  All we really need
 855  *                     to do is ensure that the log is synced at least
 856  *                     periodically.  We also push the inodes and
 857  *                     superblock if we can lock them without sleeping
 858  *                      and they are not pinned.
 859  *      SYNC_ATTR    - We need to flush the inodes.  If SYNC_BDFLUSH is not
 860  *                     set, then we really want to lock each inode and flush
 861  *                     it.
 862  *      SYNC_WAIT    - All the flushes that take place in this call should
 863  *                     be synchronous.
 864  *      SYNC_DELWRI  - This tells us to push dirty pages associated with
 865  *                     inodes.  SYNC_WAIT and SYNC_BDFLUSH are used to
 866  *                     determine if they should be flushed sync, async, or
 867  *                     delwri.
 868  *      SYNC_CLOSE   - This flag is passed when the system is being
 869  *                     unmounted.  We should sync and invalidate everything.
 870  *      SYNC_FSDATA  - This indicates that the caller would like to make
 871  *                     sure the superblock is safe on disk.  We can ensure
 872  *                     this by simply making sure the log gets flushed
 873  *                     if SYNC_BDFLUSH is set, and by actually writing it
 874  *                     out otherwise.
 875  *      SYNC_IOWAIT  - The caller wants us to wait for all data I/O to complete
 876  *                     before we return (including direct I/O). Forms the drain
 877  *                     side of the write barrier needed to safely quiesce the
 878  *                     filesystem.
 879  *
 880  */
 881 /*ARGSUSED*/
 882 STATIC int
 883 xfs_sync(
 884         bhv_desc_t      *bdp,
 885         int             flags,
 886         cred_t          *credp)
 887 {
 888         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
 889
 890         return xfs_syncsub(mp, flags, 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.
 898  *
 899  */
 900 int
 901 xfs_sync_inodes(
 902         xfs_mount_t     *mp,
 903         int             flags,
 904         int             *bypassed)
 905 {
 906         xfs_inode_t     *ip = NULL;
 907         xfs_inode_t     *ip_next;
 908         xfs_buf_t       *bp;
 909         bhv_vnode_t     *vp = NULL;
 910         int             error;
 911         int             last_error;
 912         uint64_t        fflag;
 913         uint            lock_flags;
 914         uint            base_lock_flags;
 915         boolean_t       mount_locked;
 916         boolean_t       vnode_refed;
 917         int             preempt;
 918         xfs_dinode_t    *dip;
 919         xfs_iptr_t      *ipointer;
 920 #ifdef DEBUG
 921         boolean_t       ipointer_in = B_FALSE;
 922
 923 #define IPOINTER_SET    ipointer_in = B_TRUE
 924 #define IPOINTER_CLR    ipointer_in = B_FALSE
 925 #else
 926 #define IPOINTER_SET
 927 #define IPOINTER_CLR
 928 #endif
 929
 930
 931 /* Insert a marker record into the inode list after inode ip. The list
 932  * must be locked when this is called. After the call the list will no
 933  * longer be locked.
 934  */
 935 #define IPOINTER_INSERT(ip, mp) { \
 936                 ASSERT(ipointer_in == B_FALSE); \
 937                 ipointer->ip_mnext = ip->i_mnext; \
 938                 ipointer->ip_mprev = ip; \
 939                 ip->i_mnext = (xfs_inode_t *)ipointer; \
 940                 ipointer->ip_mnext->i_mprev = (xfs_inode_t *)ipointer; \
 941                 preempt = 0; \
 942                 XFS_MOUNT_IUNLOCK(mp); \
 943                 mount_locked = B_FALSE; \
 944                 IPOINTER_SET; \
 945         }
 946
 947 /* Remove the marker from the inode list. If the marker was the only item
 948  * in the list then there are no remaining inodes and we should zero out
 949  * the whole list. If we are the current head of the list then move the head
 950  * past us.
 951  */
 952 #define IPOINTER_REMOVE(ip, mp) { \
 953                 ASSERT(ipointer_in == B_TRUE); \
 954                 if (ipointer->ip_mnext != (xfs_inode_t *)ipointer) { \
 955                         ip = ipointer->ip_mnext; \
 956                         ip->i_mprev = ipointer->ip_mprev; \
 957                         ipointer->ip_mprev->i_mnext = ip; \
 958                         if (mp->m_inodes == (xfs_inode_t *)ipointer) { \
 959                                 mp->m_inodes = ip; \
 960                         } \
 961                 } else { \
 962                         ASSERT(mp->m_inodes == (xfs_inode_t *)ipointer); \
 963                         mp->m_inodes = NULL; \
 964                         ip = NULL; \
 965                 } \
 966                 IPOINTER_CLR; \
 967         }
 968
 969 #define XFS_PREEMPT_MASK        0x7f
 970
 971         if (bypassed)
 972                 *bypassed = 0;
 973         if (XFS_MTOVFS(mp)->vfs_flag & VFS_RDONLY)
 974                 return 0;
 975         error = 0;
 976         last_error = 0;
 977         preempt = 0;
 978
 979         /* Allocate a reference marker */
 980         ipointer = (xfs_iptr_t *)kmem_zalloc(sizeof(xfs_iptr_t), KM_SLEEP);
 981
 982         fflag = XFS_B_ASYNC;            /* default is don't wait */
 983         if (flags & (SYNC_BDFLUSH | SYNC_DELWRI))
 984                 fflag = XFS_B_DELWRI;
 985         if (flags & SYNC_WAIT)
 986                 fflag = 0;              /* synchronous overrides all */
 987
 988         base_lock_flags = XFS_ILOCK_SHARED;
 989         if (flags & (SYNC_DELWRI | SYNC_CLOSE)) {
 990                 /*
 991                  * We need the I/O lock if we're going to call any of
 992                  * the flush/inval routines.
 993                  */
 994                 base_lock_flags |= XFS_IOLOCK_SHARED;
 995         }
 996
 997         XFS_MOUNT_ILOCK(mp);
 998
 999         ip = mp->m_inodes;
1000
1001         mount_locked = B_TRUE;
1002         vnode_refed  = B_FALSE;
1003
1004         IPOINTER_CLR;
1005
1006         do {
1007                 ASSERT(ipointer_in == B_FALSE);
1008                 ASSERT(vnode_refed == B_FALSE);
1009
1010                 lock_flags = base_lock_flags;
1011
1012                 /*
1013                  * There were no inodes in the list, just break out
1014                  * of the loop.
1015                  */
1016                 if (ip == NULL) {
1017                         break;
1018                 }
1019
1020                 /*
1021                  * We found another sync thread marker - skip it
1022                  */
1023                 if (ip->i_mount == NULL) {
1024                         ip = ip->i_mnext;
1025                         continue;
1026                 }
1027
1028                 vp = XFS_ITOV_NULL(ip);
1029
1030                 /*
1031                  * If the vnode is gone then this is being torn down,
1032                  * call reclaim if it is flushed, else let regular flush
1033                  * code deal with it later in the loop.
1034                  */
1035
1036                 if (vp == NULL) {
1037                         /* Skip ones already in reclaim */
1038                         if (ip->i_flags & XFS_IRECLAIM) {
1039                                 ip = ip->i_mnext;
1040                                 continue;
1041                         }
1042                         if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0) {
1043                                 ip = ip->i_mnext;
1044                         } else if ((xfs_ipincount(ip) == 0) &&
1045                                     xfs_iflock_nowait(ip)) {
1046                                 IPOINTER_INSERT(ip, mp);
1047
1048                                 xfs_finish_reclaim(ip, 1,
1049                                                 XFS_IFLUSH_DELWRI_ELSE_ASYNC);
1050
1051                                 XFS_MOUNT_ILOCK(mp);
1052                                 mount_locked = B_TRUE;
1053                                 IPOINTER_REMOVE(ip, mp);
1054                         } else {
1055                                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1056                                 ip = ip->i_mnext;
1057                         }
1058                         continue;
1059                 }
1060
1061                 if (VN_BAD(vp)) {
1062                         ip = ip->i_mnext;
1063                         continue;
1064                 }
1065
1066                 if (XFS_FORCED_SHUTDOWN(mp) && !(flags & SYNC_CLOSE)) {
1067                         XFS_MOUNT_IUNLOCK(mp);
1068                         kmem_free(ipointer, sizeof(xfs_iptr_t));
1069                         return 0;
1070                 }
1071
1072                 /*
1073                  * If this is just vfs_sync() or pflushd() calling
1074                  * then we can skip inodes for which it looks like
1075                  * there is nothing to do.  Since we don't have the
1076                  * inode locked this is racy, but these are periodic
1077                  * calls so it doesn't matter.  For the others we want
1078                  * to know for sure, so we at least try to lock them.
1079                  */
1080                 if (flags & SYNC_BDFLUSH) {
1081                         if (((ip->i_itemp == NULL) ||
1082                              !(ip->i_itemp->ili_format.ilf_fields &
1083                                XFS_ILOG_ALL)) &&
1084                             (ip->i_update_core == 0)) {
1085                                 ip = ip->i_mnext;
1086                                 continue;
1087                         }
1088                 }
1089
1090                 /*
1091                  * Try to lock without sleeping.  We're out of order with
1092                  * the inode list lock here, so if we fail we need to drop
1093                  * the mount lock and try again.  If we're called from
1094                  * bdflush() here, then don't bother.
1095                  *
1096                  * The inode lock here actually coordinates with the
1097                  * almost spurious inode lock in xfs_ireclaim() to prevent
1098                  * the vnode we handle here without a reference from
1099                  * being freed while we reference it.  If we lock the inode
1100                  * while it's on the mount list here, then the spurious inode
1101                  * lock in xfs_ireclaim() after the inode is pulled from
1102                  * the mount list will sleep until we release it here.
1103                  * This keeps the vnode from being freed while we reference
1104                  * it.
1105                  */
1106                 if (xfs_ilock_nowait(ip, lock_flags) == 0) {
1107                         if ((flags & SYNC_BDFLUSH) || (vp == NULL)) {
1108                                 ip = ip->i_mnext;
1109                                 continue;
1110                         }
1111
1112                         vp = vn_grab(vp);
1113                         if (vp == NULL) {
1114                                 ip = ip->i_mnext;
1115                                 continue;
1116                         }
1117
1118                         IPOINTER_INSERT(ip, mp);
1119                         xfs_ilock(ip, lock_flags);
1120
1121                         ASSERT(vp == XFS_ITOV(ip));
1122                         ASSERT(ip->i_mount == mp);
1123
1124                         vnode_refed = B_TRUE;
1125                 }
1126
1127                 /* From here on in the loop we may have a marker record
1128                  * in the inode list.
1129                  */
1130
1131                 if ((flags & SYNC_CLOSE)  && (vp != NULL)) {
1132                         /*
1133                          * This is the shutdown case.  We just need to
1134                          * flush and invalidate all the pages associated
1135                          * with the inode.  Drop the inode lock since
1136                          * we can't hold it across calls to the buffer
1137                          * cache.
1138                          *
1139                          * We don't set the VREMAPPING bit in the vnode
1140                          * here, because we don't hold the vnode lock
1141                          * exclusively.  It doesn't really matter, though,
1142                          * because we only come here when we're shutting
1143                          * down anyway.
1144                          */
1145                         xfs_iunlock(ip, XFS_ILOCK_SHARED);
1146
1147                         if (XFS_FORCED_SHUTDOWN(mp)) {
1148                                 bhv_vop_toss_pages(vp, 0, -1, FI_REMAPF);
1149                         } else {
1150                                 bhv_vop_flushinval_pages(vp, 0, -1, FI_REMAPF);
1151                         }
1152
1153                         xfs_ilock(ip, XFS_ILOCK_SHARED);
1154
1155                 } else if ((flags & SYNC_DELWRI) && (vp != NULL)) {
1156                         if (VN_DIRTY(vp)) {
1157                                 /* We need to have dropped the lock here,
1158                                  * so insert a marker if we have not already
1159                                  * done so.
1160                                  */
1161                                 if (mount_locked) {
1162                                         IPOINTER_INSERT(ip, mp);
1163                                 }
1164
1165                                 /*
1166                                  * Drop the inode lock since we can't hold it
1167                                  * across calls to the buffer cache.
1168                                  */
1169                                 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1170                                 error = bhv_vop_flush_pages(vp, (xfs_off_t)0,
1171                                                         -1, fflag, FI_NONE);
1172                                 xfs_ilock(ip, XFS_ILOCK_SHARED);
1173                         }
1174
1175                 }
1176                 /*
1177                  * When freezing, we need to wait ensure all I/O (including direct
1178                  * I/O) is complete to ensure no further data modification can take
1179                  * place after this point
1180                  */
1181                 if (flags & SYNC_IOWAIT)
1182                         vn_iowait(vp);
1183
1184                 if (flags & SYNC_BDFLUSH) {
1185                         if ((flags & SYNC_ATTR) &&
1186                             ((ip->i_update_core) ||
1187                              ((ip->i_itemp != NULL) &&
1188                               (ip->i_itemp->ili_format.ilf_fields != 0)))) {
1189
1190                                 /* Insert marker and drop lock if not already
1191                                  * done.
1192                                  */
1193                                 if (mount_locked) {
1194                                         IPOINTER_INSERT(ip, mp);
1195                                 }
1196
1197                                 /*
1198                                  * We don't want the periodic flushing of the
1199                                  * inodes by vfs_sync() to interfere with
1200                                  * I/O to the file, especially read I/O
1201                                  * where it is only the access time stamp
1202                                  * that is being flushed out.  To prevent
1203                                  * long periods where we have both inode
1204                                  * locks held shared here while reading the
1205                                  * inode's buffer in from disk, we drop the
1206                                  * inode lock while reading in the inode
1207                                  * buffer.  We have to release the buffer
1208                                  * and reacquire the inode lock so that they
1209                                  * are acquired in the proper order (inode
1210                                  * locks first).  The buffer will go at the
1211                                  * end of the lru chain, though, so we can
1212                                  * expect it to still be there when we go
1213                                  * for it again in xfs_iflush().
1214                                  */
1215                                 if ((xfs_ipincount(ip) == 0) &&
1216                                     xfs_iflock_nowait(ip)) {
1217
1218                                         xfs_ifunlock(ip);
1219                                         xfs_iunlock(ip, XFS_ILOCK_SHARED);
1220
1221                                         error = xfs_itobp(mp, NULL, ip,
1222                                                           &dip, &bp, 0, 0);
1223                                         if (!error) {
1224                                                 xfs_buf_relse(bp);
1225                                         } else {
1226                                                 /* Bailing out, remove the
1227                                                  * marker and free it.
1228                                                  */
1229                                                 XFS_MOUNT_ILOCK(mp);
1230                                                 IPOINTER_REMOVE(ip, mp);
1231                                                 XFS_MOUNT_IUNLOCK(mp);
1232
1233                                                 ASSERT(!(lock_flags &
1234                                                         XFS_IOLOCK_SHARED));
1235
1236                                                 kmem_free(ipointer,
1237                                                         sizeof(xfs_iptr_t));
1238                                                 return (0);
1239                                         }
1240
1241                                         /*
1242                                          * Since we dropped the inode lock,
1243                                          * the inode may have been reclaimed.
1244                                          * Therefore, we reacquire the mount
1245                                          * lock and check to see if we were the
1246                                          * inode reclaimed. If this happened
1247                                          * then the ipointer marker will no
1248                                          * longer point back at us. In this
1249                                          * case, move ip along to the inode
1250                                          * after the marker, remove the marker
1251                                          * and continue.
1252                                          */
1253                                         XFS_MOUNT_ILOCK(mp);
1254                                         mount_locked = B_TRUE;
1255
1256                                         if (ip != ipointer->ip_mprev) {
1257                                                 IPOINTER_REMOVE(ip, mp);
1258
1259                                                 ASSERT(!vnode_refed);
1260                                                 ASSERT(!(lock_flags &
1261                                                         XFS_IOLOCK_SHARED));
1262                                                 continue;
1263                                         }
1264
1265                                         ASSERT(ip->i_mount == mp);
1266
1267                                         if (xfs_ilock_nowait(ip,
1268                                                     XFS_ILOCK_SHARED) == 0) {
1269                                                 ASSERT(ip->i_mount == mp);
1270                                                 /*
1271                                                  * We failed to reacquire
1272                                                  * the inode lock without
1273                                                  * sleeping, so just skip
1274                                                  * the inode for now.  We
1275                                                  * clear the ILOCK bit from
1276                                                  * the lock_flags so that we
1277                                                  * won't try to drop a lock
1278                                                  * we don't hold below.
1279                                                  */
1280                                                 lock_flags &= ~XFS_ILOCK_SHARED;
1281                                                 IPOINTER_REMOVE(ip_next, mp);
1282                                         } else if ((xfs_ipincount(ip) == 0) &&
1283                                                    xfs_iflock_nowait(ip)) {
1284                                                 ASSERT(ip->i_mount == mp);
1285                                                 /*
1286                                                  * Since this is vfs_sync()
1287                                                  * calling we only flush the
1288                                                  * inode out if we can lock
1289                                                  * it without sleeping and
1290                                                  * it is not pinned.  Drop
1291                                                  * the mount lock here so
1292                                                  * that we don't hold it for
1293                                                  * too long. We already have
1294                                                  * a marker in the list here.
1295                                                  */
1296                                                 XFS_MOUNT_IUNLOCK(mp);
1297                                                 mount_locked = B_FALSE;
1298                                                 error = xfs_iflush(ip,
1299                                                            XFS_IFLUSH_DELWRI);
1300                                         } else {
1301                                                 ASSERT(ip->i_mount == mp);
1302                                                 IPOINTER_REMOVE(ip_next, mp);
1303                                         }
1304                                 }
1305
1306                         }
1307
1308                 } else {
1309                         if ((flags & SYNC_ATTR) &&
1310                             ((ip->i_update_core) ||
1311                              ((ip->i_itemp != NULL) &&
1312                               (ip->i_itemp->ili_format.ilf_fields != 0)))) {
1313                                 if (mount_locked) {
1314                                         IPOINTER_INSERT(ip, mp);
1315                                 }
1316
1317                                 if (flags & SYNC_WAIT) {
1318                                         xfs_iflock(ip);
1319                                         error = xfs_iflush(ip,
1320                                                            XFS_IFLUSH_SYNC);
1321                                 } else {
1322                                         /*
1323                                          * If we can't acquire the flush
1324                                          * lock, then the inode is already
1325                                          * being flushed so don't bother
1326                                          * waiting.  If we can lock it then
1327                                          * do a delwri flush so we can
1328                                          * combine multiple inode flushes
1329                                          * in each disk write.
1330                                          */
1331                                         if (xfs_iflock_nowait(ip)) {
1332                                                 error = xfs_iflush(ip,
1333                                                            XFS_IFLUSH_DELWRI);
1334                                         }
1335                                         else if (bypassed)
1336                                                 (*bypassed)++;
1337                                 }
1338                         }
1339                 }
1340
1341                 if (lock_flags != 0) {
1342                         xfs_iunlock(ip, lock_flags);
1343                 }
1344
1345                 if (vnode_refed) {
1346                         /*
1347                          * If we had to take a reference on the vnode
1348                          * above, then wait until after we've unlocked
1349                          * the inode to release the reference.  This is
1350                          * because we can be already holding the inode
1351                          * lock when VN_RELE() calls xfs_inactive().
1352                          *
1353                          * Make sure to drop the mount lock before calling
1354                          * VN_RELE() so that we don't trip over ourselves if
1355                          * we have to go for the mount lock again in the
1356                          * inactive code.
1357                          */
1358                         if (mount_locked) {
1359                                 IPOINTER_INSERT(ip, mp);
1360                         }
1361
1362                         VN_RELE(vp);
1363
1364                         vnode_refed = B_FALSE;
1365                 }
1366
1367                 if (error) {
1368                         last_error = error;
1369                 }
1370
1371                 /*
1372                  * bail out if the filesystem is corrupted.
1373                  */
1374                 if (error == EFSCORRUPTED)  {
1375                         if (!mount_locked) {
1376                                 XFS_MOUNT_ILOCK(mp);
1377                                 IPOINTER_REMOVE(ip, mp);
1378                         }
1379                         XFS_MOUNT_IUNLOCK(mp);
1380                         ASSERT(ipointer_in == B_FALSE);
1381                         kmem_free(ipointer, sizeof(xfs_iptr_t));
1382                         return XFS_ERROR(error);
1383                 }
1384
1385                 /* Let other threads have a chance at the mount lock
1386                  * if we have looped many times without dropping the
1387                  * lock.
1388                  */
1389                 if ((++preempt & XFS_PREEMPT_MASK) == 0) {
1390                         if (mount_locked) {
1391                                 IPOINTER_INSERT(ip, mp);
1392                         }
1393                 }
1394
1395                 if (mount_locked == B_FALSE) {
1396                         XFS_MOUNT_ILOCK(mp);
1397                         mount_locked = B_TRUE;
1398                         IPOINTER_REMOVE(ip, mp);
1399                         continue;
1400                 }
1401
1402                 ASSERT(ipointer_in == B_FALSE);
1403                 ip = ip->i_mnext;
1404
1405         } while (ip != mp->m_inodes);
1406
1407         XFS_MOUNT_IUNLOCK(mp);
1408
1409         ASSERT(ipointer_in == B_FALSE);
1410
1411         kmem_free(ipointer, sizeof(xfs_iptr_t));
1412         return XFS_ERROR(last_error);
1413 }
1414
1415 /*
1416  * xfs sync routine for internal use
1417  *
1418  * This routine supports all of the flags defined for the generic vfs_sync
1419  * interface as explained above under xfs_sync.
1420  *
1421  */
1422 int
1423 xfs_syncsub(
1424         xfs_mount_t     *mp,
1425         int             flags,
1426         int             *bypassed)
1427 {
1428         int             error = 0;
1429         int             last_error = 0;
1430         uint            log_flags = XFS_LOG_FORCE;
1431         xfs_buf_t       *bp;
1432         xfs_buf_log_item_t      *bip;
1433
1434         /*
1435          * Sync out the log.  This ensures that the log is periodically
1436          * flushed even if there is not enough activity to fill it up.
1437          */
1438         if (flags & SYNC_WAIT)
1439                 log_flags |= XFS_LOG_SYNC;
1440
1441         xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1442
1443         if (flags & (SYNC_ATTR|SYNC_DELWRI)) {
1444                 if (flags & SYNC_BDFLUSH)
1445                         xfs_finish_reclaim_all(mp, 1);
1446                 else
1447                         error = xfs_sync_inodes(mp, flags, bypassed);
1448         }
1449
1450         /*
1451          * Flushing out dirty data above probably generated more
1452          * log activity, so if this isn't vfs_sync() then flush
1453          * the log again.
1454          */
1455         if (flags & SYNC_DELWRI) {
1456                 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1457         }
1458
1459         if (flags & SYNC_FSDATA) {
1460                 /*
1461                  * If this is vfs_sync() then only sync the superblock
1462                  * if we can lock it without sleeping and it is not pinned.
1463                  */
1464                 if (flags & SYNC_BDFLUSH) {
1465                         bp = xfs_getsb(mp, XFS_BUF_TRYLOCK);
1466                         if (bp != NULL) {
1467                                 bip = XFS_BUF_FSPRIVATE(bp,xfs_buf_log_item_t*);
1468                                 if ((bip != NULL) &&
1469                                     xfs_buf_item_dirty(bip)) {
1470                                         if (!(XFS_BUF_ISPINNED(bp))) {
1471                                                 XFS_BUF_ASYNC(bp);
1472                                                 error = xfs_bwrite(mp, bp);
1473                                         } else {
1474                                                 xfs_buf_relse(bp);
1475                                         }
1476                                 } else {
1477                                         xfs_buf_relse(bp);
1478                                 }
1479                         }
1480                 } else {
1481                         bp = xfs_getsb(mp, 0);
1482                         /*
1483                          * If the buffer is pinned then push on the log so
1484                          * we won't get stuck waiting in the write for
1485                          * someone, maybe ourselves, to flush the log.
1486                          * Even though we just pushed the log above, we
1487                          * did not have the superblock buffer locked at
1488                          * that point so it can become pinned in between
1489                          * there and here.
1490                          */
1491                         if (XFS_BUF_ISPINNED(bp))
1492                                 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
1493                         if (flags & SYNC_WAIT)
1494                                 XFS_BUF_UNASYNC(bp);
1495                         else
1496                                 XFS_BUF_ASYNC(bp);
1497                         error = xfs_bwrite(mp, bp);
1498                 }
1499                 if (error) {
1500                         last_error = error;
1501                 }
1502         }
1503
1504         /*
1505          * If this is the periodic sync, then kick some entries out of
1506          * the reference cache.  This ensures that idle entries are
1507          * eventually kicked out of the cache.
1508          */
1509         if (flags & SYNC_REFCACHE) {
1510                 if (flags & SYNC_WAIT)
1511                         xfs_refcache_purge_mp(mp);
1512                 else
1513                         xfs_refcache_purge_some(mp);
1514         }
1515
1516         /*
1517          * Now check to see if the log needs a "dummy" transaction.
1518          */
1519
1520         if (!(flags & SYNC_REMOUNT) && xfs_log_need_covered(mp)) {
1521                 xfs_trans_t *tp;
1522                 xfs_inode_t *ip;
1523
1524                 /*
1525                  * Put a dummy transaction in the log to tell
1526                  * recovery that all others are OK.
1527                  */
1528                 tp = xfs_trans_alloc(mp, XFS_TRANS_DUMMY1);
1529                 if ((error = xfs_trans_reserve(tp, 0,
1530                                 XFS_ICHANGE_LOG_RES(mp),
1531                                 0, 0, 0)))  {
1532                         xfs_trans_cancel(tp, 0);
1533                         return error;
1534                 }
1535
1536                 ip = mp->m_rootip;
1537                 xfs_ilock(ip, XFS_ILOCK_EXCL);
1538
1539                 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1540                 xfs_trans_ihold(tp, ip);
1541                 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1542                 error = xfs_trans_commit(tp, 0, NULL);
1543                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1544                 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1545         }
1546
1547         /*
1548          * When shutting down, we need to insure that the AIL is pushed
1549          * to disk or the filesystem can appear corrupt from the PROM.
1550          */
1551         if ((flags & (SYNC_CLOSE|SYNC_WAIT)) == (SYNC_CLOSE|SYNC_WAIT)) {
1552                 XFS_bflush(mp->m_ddev_targp);
1553                 if (mp->m_rtdev_targp) {
1554                         XFS_bflush(mp->m_rtdev_targp);
1555                 }
1556         }
1557
1558         return XFS_ERROR(last_error);
1559 }
1560
1561 /*
1562  * xfs_vget - called by DMAPI and NFSD to get vnode from file handle
1563  */
1564 STATIC int
1565 xfs_vget(
1566         bhv_desc_t      *bdp,
1567         bhv_vnode_t     **vpp,
1568         fid_t           *fidp)
1569 {
1570         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
1571         xfs_fid_t       *xfid = (struct xfs_fid *)fidp;
1572         xfs_inode_t     *ip;
1573         int             error;
1574         xfs_ino_t       ino;
1575         unsigned int    igen;
1576
1577         /*
1578          * Invalid.  Since handles can be created in user space and passed in
1579          * via gethandle(), this is not cause for a panic.
1580          */
1581         if (xfid->xfs_fid_len != sizeof(*xfid) - sizeof(xfid->xfs_fid_len))
1582                 return XFS_ERROR(EINVAL);
1583
1584         ino  = xfid->xfs_fid_ino;
1585         igen = xfid->xfs_fid_gen;
1586
1587         /*
1588          * NFS can sometimes send requests for ino 0.  Fail them gracefully.
1589          */
1590         if (ino == 0)
1591                 return XFS_ERROR(ESTALE);
1592
1593         error = xfs_iget(mp, NULL, ino, 0, XFS_ILOCK_SHARED, &ip, 0);
1594         if (error) {
1595                 *vpp = NULL;
1596                 return error;
1597         }
1598
1599         if (ip == NULL) {
1600                 *vpp = NULL;
1601                 return XFS_ERROR(EIO);
1602         }
1603
1604         if (ip->i_d.di_mode == 0 || ip->i_d.di_gen != igen) {
1605                 xfs_iput_new(ip, XFS_ILOCK_SHARED);
1606                 *vpp = NULL;
1607                 return XFS_ERROR(ENOENT);
1608         }
1609
1610         *vpp = XFS_ITOV(ip);
1611         xfs_iunlock(ip, XFS_ILOCK_SHARED);
1612         return 0;
1613 }
1614
1615
1616 #define MNTOPT_LOGBUFS  "logbufs"       /* number of XFS log buffers */
1617 #define MNTOPT_LOGBSIZE "logbsize"      /* size of XFS log buffers */
1618 #define MNTOPT_LOGDEV   "logdev"        /* log device */
1619 #define MNTOPT_RTDEV    "rtdev"         /* realtime I/O device */
1620 #define MNTOPT_BIOSIZE  "biosize"       /* log2 of preferred buffered io size */
1621 #define MNTOPT_WSYNC    "wsync"         /* safe-mode nfs compatible mount */
1622 #define MNTOPT_INO64    "ino64"         /* force inodes into 64-bit range */
1623 #define MNTOPT_NOALIGN  "noalign"       /* turn off stripe alignment */
1624 #define MNTOPT_SWALLOC  "swalloc"       /* turn on stripe width allocation */
1625 #define MNTOPT_SUNIT    "sunit"         /* data volume stripe unit */
1626 #define MNTOPT_SWIDTH   "swidth"        /* data volume stripe width */
1627 #define MNTOPT_NOUUID   "nouuid"        /* ignore filesystem UUID */
1628 #define MNTOPT_MTPT     "mtpt"          /* filesystem mount point */
1629 #define MNTOPT_GRPID    "grpid"         /* group-ID from parent directory */
1630 #define MNTOPT_NOGRPID  "nogrpid"       /* group-ID from current process */
1631 #define MNTOPT_BSDGROUPS    "bsdgroups"    /* group-ID from parent directory */
1632 #define MNTOPT_SYSVGROUPS   "sysvgroups"   /* group-ID from current process */
1633 #define MNTOPT_ALLOCSIZE    "allocsize"    /* preferred allocation size */
1634 #define MNTOPT_IHASHSIZE    "ihashsize"    /* size of inode hash table */
1635 #define MNTOPT_NORECOVERY   "norecovery"   /* don't run XFS recovery */
1636 #define MNTOPT_BARRIER  "barrier"       /* use writer barriers for log write and
1637                                          * unwritten extent conversion */
1638 #define MNTOPT_NOBARRIER "nobarrier"    /* .. disable */
1639 #define MNTOPT_OSYNCISOSYNC "osyncisosync" /* o_sync is REALLY o_sync */
1640 #define MNTOPT_64BITINODE   "inode64"   /* inodes can be allocated anywhere */
1641 #define MNTOPT_IKEEP    "ikeep"         /* do not free empty inode clusters */
1642 #define MNTOPT_NOIKEEP  "noikeep"       /* free empty inode clusters */
1643 #define MNTOPT_LARGEIO     "largeio"    /* report large I/O sizes in stat() */
1644 #define MNTOPT_NOLARGEIO   "nolargeio"  /* do not report large I/O sizes
1645                                          * in stat(). */
1646 #define MNTOPT_ATTR2    "attr2"         /* do use attr2 attribute format */
1647 #define MNTOPT_NOATTR2  "noattr2"       /* do not use attr2 attribute format */
1648
1649 STATIC unsigned long
1650 suffix_strtoul(char *s, char **endp, unsigned int base)
1651 {
1652         int     last, shift_left_factor = 0;
1653         char    *value = s;
1654
1655         last = strlen(value) - 1;
1656         if (value[last] == 'K' || value[last] == 'k') {
1657                 shift_left_factor = 10;
1658                 value[last] = '\0';
1659         }
1660         if (value[last] == 'M' || value[last] == 'm') {
1661                 shift_left_factor = 20;
1662                 value[last] = '\0';
1663         }
1664         if (value[last] == 'G' || value[last] == 'g') {
1665                 shift_left_factor = 30;
1666                 value[last] = '\0';
1667         }
1668
1669         return simple_strtoul((const char *)s, endp, base) << shift_left_factor;
1670 }
1671
1672 STATIC int
1673 xfs_parseargs(
1674         struct bhv_desc         *bhv,
1675         char                    *options,
1676         struct xfs_mount_args   *args,
1677         int                     update)
1678 {
1679         bhv_vfs_t               *vfsp = bhvtovfs(bhv);
1680         char                    *this_char, *value, *eov;
1681         int                     dsunit, dswidth, vol_dsunit, vol_dswidth;
1682         int                     iosize;
1683
1684         args->flags |= XFSMNT_IDELETE;
1685         args->flags |= XFSMNT_BARRIER;
1686         args->flags2 |= XFSMNT2_COMPAT_IOSIZE;
1687
1688         if (!options)
1689                 goto done;
1690
1691         iosize = dsunit = dswidth = vol_dsunit = vol_dswidth = 0;
1692
1693         while ((this_char = strsep(&options, ",")) != NULL) {
1694                 if (!*this_char)
1695                         continue;
1696                 if ((value = strchr(this_char, '=')) != NULL)
1697                         *value++ = 0;
1698
1699                 if (!strcmp(this_char, MNTOPT_LOGBUFS)) {
1700                         if (!value || !*value) {
1701                                 cmn_err(CE_WARN,
1702                                         "XFS: %s option requires an argument",
1703                                         this_char);
1704                                 return EINVAL;
1705                         }
1706                         args->logbufs = simple_strtoul(value, &eov, 10);
1707                 } else if (!strcmp(this_char, MNTOPT_LOGBSIZE)) {
1708                         if (!value || !*value) {
1709                                 cmn_err(CE_WARN,
1710                                         "XFS: %s option requires an argument",
1711                                         this_char);
1712                                 return EINVAL;
1713                         }
1714                         args->logbufsize = suffix_strtoul(value, &eov, 10);
1715                 } else if (!strcmp(this_char, MNTOPT_LOGDEV)) {
1716                         if (!value || !*value) {
1717                                 cmn_err(CE_WARN,
1718                                         "XFS: %s option requires an argument",
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                                 cmn_err(CE_WARN,
1726                                         "XFS: %s option requires an argument",
1727                                         this_char);
1728                                 return EINVAL;
1729                         }
1730                         strncpy(args->mtpt, value, MAXNAMELEN);
1731                 } else if (!strcmp(this_char, MNTOPT_RTDEV)) {
1732                         if (!value || !*value) {
1733                                 cmn_err(CE_WARN,
1734                                         "XFS: %s option requires an argument",
1735                                         this_char);
1736                                 return EINVAL;
1737                         }
1738                         strncpy(args->rtname, value, MAXNAMELEN);
1739                 } else if (!strcmp(this_char, MNTOPT_BIOSIZE)) {
1740                         if (!value || !*value) {
1741                                 cmn_err(CE_WARN,
1742                                         "XFS: %s option requires an argument",
1743                                         this_char);
1744                                 return EINVAL;
1745                         }
1746                         iosize = simple_strtoul(value, &eov, 10);
1747                         args->flags |= XFSMNT_IOSIZE;
1748                         args->iosizelog = (uint8_t) iosize;
1749                 } else if (!strcmp(this_char, MNTOPT_ALLOCSIZE)) {
1750                         if (!value || !*value) {
1751                                 cmn_err(CE_WARN,
1752                                         "XFS: %s option requires an argument",
1753                                         this_char);
1754                                 return EINVAL;
1755                         }
1756                         iosize = suffix_strtoul(value, &eov, 10);
1757                         args->flags |= XFSMNT_IOSIZE;
1758                         args->iosizelog = ffs(iosize) - 1;
1759                 } else if (!strcmp(this_char, MNTOPT_IHASHSIZE)) {
1760                         if (!value || !*value) {
1761                                 cmn_err(CE_WARN,
1762                                         "XFS: %s option requires an argument",
1763                                         this_char);
1764                                 return EINVAL;
1765                         }
1766                         args->flags |= XFSMNT_IHASHSIZE;
1767                         args->ihashsize = simple_strtoul(value, &eov, 10);
1768                 } else if (!strcmp(this_char, MNTOPT_GRPID) ||
1769                            !strcmp(this_char, MNTOPT_BSDGROUPS)) {
1770                         vfsp->vfs_flag |= VFS_GRPID;
1771                 } else if (!strcmp(this_char, MNTOPT_NOGRPID) ||
1772                            !strcmp(this_char, MNTOPT_SYSVGROUPS)) {
1773                         vfsp->vfs_flag &= ~VFS_GRPID;
1774                 } else if (!strcmp(this_char, MNTOPT_WSYNC)) {
1775                         args->flags |= XFSMNT_WSYNC;
1776                 } else if (!strcmp(this_char, MNTOPT_OSYNCISOSYNC)) {
1777                         args->flags |= XFSMNT_OSYNCISOSYNC;
1778                 } else if (!strcmp(this_char, MNTOPT_NORECOVERY)) {
1779                         args->flags |= XFSMNT_NORECOVERY;
1780                 } else if (!strcmp(this_char, MNTOPT_INO64)) {
1781                         args->flags |= XFSMNT_INO64;
1782 #if !XFS_BIG_INUMS
1783                         cmn_err(CE_WARN,
1784                                 "XFS: %s option not allowed on this system",
1785                                 this_char);
1786                         return EINVAL;
1787 #endif
1788                 } else if (!strcmp(this_char, MNTOPT_NOALIGN)) {
1789                         args->flags |= XFSMNT_NOALIGN;
1790                 } else if (!strcmp(this_char, MNTOPT_SWALLOC)) {
1791                         args->flags |= XFSMNT_SWALLOC;
1792                 } else if (!strcmp(this_char, MNTOPT_SUNIT)) {
1793                         if (!value || !*value) {
1794                                 cmn_err(CE_WARN,
1795                                         "XFS: %s option requires an argument",
1796                                         this_char);
1797                                 return EINVAL;
1798                         }
1799                         dsunit = simple_strtoul(value, &eov, 10);
1800                 } else if (!strcmp(this_char, MNTOPT_SWIDTH)) {
1801                         if (!value || !*value) {
1802                                 cmn_err(CE_WARN,
1803                                         "XFS: %s option requires an argument",
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                         cmn_err(CE_WARN,
1812                                 "XFS: %s option not allowed on this system",
1813                                 this_char);
1814                         return EINVAL;
1815 #endif
1816                 } else if (!strcmp(this_char, MNTOPT_NOUUID)) {
1817                         args->flags |= XFSMNT_NOUUID;
1818                 } else if (!strcmp(this_char, MNTOPT_BARRIER)) {
1819                         args->flags |= XFSMNT_BARRIER;
1820                 } else if (!strcmp(this_char, MNTOPT_NOBARRIER)) {
1821                         args->flags &= ~XFSMNT_BARRIER;
1822                 } else if (!strcmp(this_char, MNTOPT_IKEEP)) {
1823                         args->flags &= ~XFSMNT_IDELETE;
1824                 } else if (!strcmp(this_char, MNTOPT_NOIKEEP)) {
1825                         args->flags |= XFSMNT_IDELETE;
1826                 } else if (!strcmp(this_char, MNTOPT_LARGEIO)) {
1827                         args->flags2 &= ~XFSMNT2_COMPAT_IOSIZE;
1828                 } else if (!strcmp(this_char, MNTOPT_NOLARGEIO)) {
1829                         args->flags2 |= XFSMNT2_COMPAT_IOSIZE;
1830                 } else if (!strcmp(this_char, MNTOPT_ATTR2)) {
1831                         args->flags |= XFSMNT_ATTR2;
1832                 } else if (!strcmp(this_char, MNTOPT_NOATTR2)) {
1833                         args->flags &= ~XFSMNT_ATTR2;
1834                 } else if (!strcmp(this_char, "osyncisdsync")) {
1835                         /* no-op, this is now the default */
1836                         cmn_err(CE_WARN,
1837         "XFS: osyncisdsync is now the default, option is deprecated.");
1838                 } else if (!strcmp(this_char, "irixsgid")) {
1839                         cmn_err(CE_WARN,
1840         "XFS: irixsgid is now a sysctl(2) variable, option is deprecated.");
1841                 } else {
1842                         cmn_err(CE_WARN,
1843                                 "XFS: unknown mount option [%s].", this_char);
1844                         return EINVAL;
1845                 }
1846         }
1847
1848         if (args->flags & XFSMNT_NORECOVERY) {
1849                 if ((vfsp->vfs_flag & VFS_RDONLY) == 0) {
1850                         cmn_err(CE_WARN,
1851                                 "XFS: no-recovery mounts must be read-only.");
1852                         return EINVAL;
1853                 }
1854         }
1855
1856         if ((args->flags & XFSMNT_NOALIGN) && (dsunit || dswidth)) {
1857                 cmn_err(CE_WARN,
1858         "XFS: sunit and swidth options incompatible with the noalign option");
1859                 return EINVAL;
1860         }
1861
1862         if ((dsunit && !dswidth) || (!dsunit && dswidth)) {
1863                 cmn_err(CE_WARN,
1864                         "XFS: sunit and swidth must be specified together");
1865                 return EINVAL;
1866         }
1867
1868         if (dsunit && (dswidth % dsunit != 0)) {
1869                 cmn_err(CE_WARN,
1870         "XFS: stripe width (%d) must be a multiple of the stripe unit (%d)",
1871                         dswidth, dsunit);
1872                 return EINVAL;
1873         }
1874
1875         if ((args->flags & XFSMNT_NOALIGN) != XFSMNT_NOALIGN) {
1876                 if (dsunit) {
1877                         args->sunit = dsunit;
1878                         args->flags |= XFSMNT_RETERR;
1879                 } else {
1880                         args->sunit = vol_dsunit;
1881                 }
1882                 dswidth ? (args->swidth = dswidth) :
1883                           (args->swidth = vol_dswidth);
1884         } else {
1885                 args->sunit = args->swidth = 0;
1886         }
1887
1888 done:
1889         if (args->flags & XFSMNT_32BITINODES)
1890                 vfsp->vfs_flag |= VFS_32BITINODES;
1891         if (args->flags2)
1892                 args->flags |= XFSMNT_FLAGS2;
1893         return 0;
1894 }
1895
1896 STATIC int
1897 xfs_showargs(
1898         struct bhv_desc         *bhv,
1899         struct seq_file         *m)
1900 {
1901         static struct proc_xfs_info {
1902                 int     flag;
1903                 char    *str;
1904         } xfs_info[] = {
1905                 /* the few simple ones we can get from the mount struct */
1906                 { XFS_MOUNT_WSYNC,              "," MNTOPT_WSYNC },
1907                 { XFS_MOUNT_INO64,              "," MNTOPT_INO64 },
1908                 { XFS_MOUNT_NOALIGN,            "," MNTOPT_NOALIGN },
1909                 { XFS_MOUNT_SWALLOC,            "," MNTOPT_SWALLOC },
1910                 { XFS_MOUNT_NOUUID,             "," MNTOPT_NOUUID },
1911                 { XFS_MOUNT_NORECOVERY,         "," MNTOPT_NORECOVERY },
1912                 { XFS_MOUNT_OSYNCISOSYNC,       "," MNTOPT_OSYNCISOSYNC },
1913                 { 0, NULL }
1914         };
1915         struct proc_xfs_info    *xfs_infop;
1916         struct xfs_mount        *mp = XFS_BHVTOM(bhv);
1917         struct bhv_vfs          *vfsp = XFS_MTOVFS(mp);
1918
1919         for (xfs_infop = xfs_info; xfs_infop->flag; xfs_infop++) {
1920                 if (mp->m_flags & xfs_infop->flag)
1921                         seq_puts(m, xfs_infop->str);
1922         }
1923
1924         if (mp->m_flags & XFS_MOUNT_IHASHSIZE)
1925                 seq_printf(m, "," MNTOPT_IHASHSIZE "=%d", (int)mp->m_ihsize);
1926
1927         if (mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)
1928                 seq_printf(m, "," MNTOPT_ALLOCSIZE "=%dk",
1929                                 (int)(1 << mp->m_writeio_log) >> 10);
1930
1931         if (mp->m_logbufs > 0)
1932                 seq_printf(m, "," MNTOPT_LOGBUFS "=%d", mp->m_logbufs);
1933         if (mp->m_logbsize > 0)
1934                 seq_printf(m, "," MNTOPT_LOGBSIZE "=%dk", mp->m_logbsize >> 10);
1935
1936         if (mp->m_logname)
1937                 seq_printf(m, "," MNTOPT_LOGDEV "=%s", mp->m_logname);
1938         if (mp->m_rtname)
1939                 seq_printf(m, "," MNTOPT_RTDEV "=%s", mp->m_rtname);
1940
1941         if (mp->m_dalign > 0)
1942                 seq_printf(m, "," MNTOPT_SUNIT "=%d",
1943                                 (int)XFS_FSB_TO_BB(mp, mp->m_dalign));
1944         if (mp->m_swidth > 0)
1945                 seq_printf(m, "," MNTOPT_SWIDTH "=%d",
1946                                 (int)XFS_FSB_TO_BB(mp, mp->m_swidth));
1947
1948         if (!(mp->m_flags & XFS_MOUNT_IDELETE))
1949                 seq_printf(m, "," MNTOPT_IKEEP);
1950         if (!(mp->m_flags & XFS_MOUNT_COMPAT_IOSIZE))
1951                 seq_printf(m, "," MNTOPT_LARGEIO);
1952
1953         if (!(vfsp->vfs_flag & VFS_32BITINODES))
1954                 seq_printf(m, "," MNTOPT_64BITINODE);
1955         if (vfsp->vfs_flag & VFS_GRPID)
1956                 seq_printf(m, "," MNTOPT_GRPID);
1957
1958         return 0;
1959 }
1960
1961 /*
1962  * Second stage of a freeze. The data is already frozen, now we have to take
1963  * care of the metadata. New transactions are already blocked, so we need to
1964  * wait for any remaining transactions to drain out before proceding.
1965  */
1966 STATIC void
1967 xfs_freeze(
1968         bhv_desc_t      *bdp)
1969 {
1970         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
1971
1972         /* wait for all modifications to complete */
1973         while (atomic_read(&mp->m_active_trans) > 0)
1974                 delay(100);
1975
1976         /* flush inodes and push all remaining buffers out to disk */
1977         xfs_quiesce_fs(mp);
1978
1979         ASSERT_ALWAYS(atomic_read(&mp->m_active_trans) == 0);
1980
1981         /* Push the superblock and write an unmount record */
1982         xfs_log_unmount_write(mp);
1983         xfs_unmountfs_writesb(mp);
1984         xfs_fs_log_dummy(mp);
1985 }
1986
1987
1988 bhv_vfsops_t xfs_vfsops = {
1989         BHV_IDENTITY_INIT(VFS_BHV_XFS,VFS_POSITION_XFS),
1990         .vfs_parseargs          = xfs_parseargs,
1991         .vfs_showargs           = xfs_showargs,
1992         .vfs_mount              = xfs_mount,
1993         .vfs_unmount            = xfs_unmount,
1994         .vfs_mntupdate          = xfs_mntupdate,
1995         .vfs_root               = xfs_root,
1996         .vfs_statvfs            = xfs_statvfs,
1997         .vfs_sync               = xfs_sync,
1998         .vfs_vget               = xfs_vget,
1999         .vfs_dmapiops           = (vfs_dmapiops_t)fs_nosys,
2000         .vfs_quotactl           = (vfs_quotactl_t)fs_nosys,
2001         .vfs_init_vnode         = xfs_initialize_vnode,
2002         .vfs_force_shutdown     = xfs_do_force_shutdown,
2003         .vfs_freeze             = xfs_freeze,
2004 };