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

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