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_mru_cache.h"
  55 #include "xfs_filestream.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         xfs_mru_cache_init();
  87         xfs_filestream_init();
  88
  89         /*
  90          * The size of the zone allocated buf log item is the maximum
  91          * size possible under XFS.  This wastes a little bit of memory,
  92          * but it is much faster.
  93          */
  94         xfs_buf_item_zone =
  95                 kmem_zone_init((sizeof(xfs_buf_log_item_t) +
  96                                 (((XFS_MAX_BLOCKSIZE / XFS_BLI_CHUNK) /
  97                                   NBWORD) * sizeof(int))),
  98                                "xfs_buf_item");
  99         xfs_efd_zone =
 100                 kmem_zone_init((sizeof(xfs_efd_log_item_t) +
 101                                ((XFS_EFD_MAX_FAST_EXTENTS - 1) *
 102                                  sizeof(xfs_extent_t))),
 103                                       "xfs_efd_item");
 104         xfs_efi_zone =
 105                 kmem_zone_init((sizeof(xfs_efi_log_item_t) +
 106                                ((XFS_EFI_MAX_FAST_EXTENTS - 1) *
 107                                  sizeof(xfs_extent_t))),
 108                                       "xfs_efi_item");
 109
 110         /*
 111          * These zones warrant special memory allocator hints
 112          */
 113         xfs_inode_zone =
 114                 kmem_zone_init_flags(sizeof(xfs_inode_t), "xfs_inode",
 115                                         KM_ZONE_HWALIGN | KM_ZONE_RECLAIM |
 116                                         KM_ZONE_SPREAD, NULL);
 117         xfs_ili_zone =
 118                 kmem_zone_init_flags(sizeof(xfs_inode_log_item_t), "xfs_ili",
 119                                         KM_ZONE_SPREAD, NULL);
 120         xfs_chashlist_zone =
 121                 kmem_zone_init_flags(sizeof(xfs_chashlist_t), "xfs_chashlist",
 122                                         KM_ZONE_SPREAD, NULL);
 123
 124         /*
 125          * Allocate global trace buffers.
 126          */
 127 #ifdef XFS_ALLOC_TRACE
 128         xfs_alloc_trace_buf = ktrace_alloc(XFS_ALLOC_TRACE_SIZE, KM_SLEEP);
 129 #endif
 130 #ifdef XFS_BMAP_TRACE
 131         xfs_bmap_trace_buf = ktrace_alloc(XFS_BMAP_TRACE_SIZE, KM_SLEEP);
 132 #endif
 133 #ifdef XFS_BMBT_TRACE
 134         xfs_bmbt_trace_buf = ktrace_alloc(XFS_BMBT_TRACE_SIZE, KM_SLEEP);
 135 #endif
 136 #ifdef XFS_ATTR_TRACE
 137         xfs_attr_trace_buf = ktrace_alloc(XFS_ATTR_TRACE_SIZE, KM_SLEEP);
 138 #endif
 139 #ifdef XFS_DIR2_TRACE
 140         xfs_dir2_trace_buf = ktrace_alloc(XFS_DIR2_GTRACE_SIZE, KM_SLEEP);
 141 #endif
 142
 143         xfs_dir_startup();
 144
 145 #if (defined(DEBUG) || defined(INDUCE_IO_ERROR))
 146         xfs_error_test_init();
 147 #endif /* DEBUG || INDUCE_IO_ERROR */
 148
 149         xfs_init_procfs();
 150         xfs_sysctl_register();
 151         return 0;
 152 }
 153
 154 void
 155 xfs_cleanup(void)
 156 {
 157         extern kmem_zone_t      *xfs_bmap_free_item_zone;
 158         extern kmem_zone_t      *xfs_btree_cur_zone;
 159         extern kmem_zone_t      *xfs_inode_zone;
 160         extern kmem_zone_t      *xfs_trans_zone;
 161         extern kmem_zone_t      *xfs_da_state_zone;
 162         extern kmem_zone_t      *xfs_dabuf_zone;
 163         extern kmem_zone_t      *xfs_efd_zone;
 164         extern kmem_zone_t      *xfs_efi_zone;
 165         extern kmem_zone_t      *xfs_buf_item_zone;
 166         extern kmem_zone_t      *xfs_chashlist_zone;
 167
 168         xfs_cleanup_procfs();
 169         xfs_sysctl_unregister();
 170         xfs_refcache_destroy();
 171         xfs_filestream_uninit();
 172         xfs_mru_cache_uninit();
 173         xfs_acl_zone_destroy(xfs_acl_zone);
 174
 175 #ifdef XFS_DIR2_TRACE
 176         ktrace_free(xfs_dir2_trace_buf);
 177 #endif
 178 #ifdef XFS_ATTR_TRACE
 179         ktrace_free(xfs_attr_trace_buf);
 180 #endif
 181 #ifdef XFS_BMBT_TRACE
 182         ktrace_free(xfs_bmbt_trace_buf);
 183 #endif
 184 #ifdef XFS_BMAP_TRACE
 185         ktrace_free(xfs_bmap_trace_buf);
 186 #endif
 187 #ifdef XFS_ALLOC_TRACE
 188         ktrace_free(xfs_alloc_trace_buf);
 189 #endif
 190
 191         kmem_zone_destroy(xfs_bmap_free_item_zone);
 192         kmem_zone_destroy(xfs_btree_cur_zone);
 193         kmem_zone_destroy(xfs_inode_zone);
 194         kmem_zone_destroy(xfs_trans_zone);
 195         kmem_zone_destroy(xfs_da_state_zone);
 196         kmem_zone_destroy(xfs_dabuf_zone);
 197         kmem_zone_destroy(xfs_buf_item_zone);
 198         kmem_zone_destroy(xfs_efd_zone);
 199         kmem_zone_destroy(xfs_efi_zone);
 200         kmem_zone_destroy(xfs_ifork_zone);
 201         kmem_zone_destroy(xfs_ili_zone);
 202         kmem_zone_destroy(xfs_chashlist_zone);
 203 }
 204
 205 /*
 206  * xfs_start_flags
 207  *
 208  * This function fills in xfs_mount_t fields based on mount args.
 209  * Note: the superblock has _not_ yet been read in.
 210  */
 211 STATIC int
 212 xfs_start_flags(
 213         struct bhv_vfs          *vfs,
 214         struct xfs_mount_args   *ap,
 215         struct xfs_mount        *mp)
 216 {
 217         /* Values are in BBs */
 218         if ((ap->flags & XFSMNT_NOALIGN) != XFSMNT_NOALIGN) {
 219                 /*
 220                  * At this point the superblock has not been read
 221                  * in, therefore we do not know the block size.
 222                  * Before the mount call ends we will convert
 223                  * these to FSBs.
 224                  */
 225                 mp->m_dalign = ap->sunit;
 226                 mp->m_swidth = ap->swidth;
 227         }
 228
 229         if (ap->logbufs != -1 &&
 230             ap->logbufs != 0 &&
 231             (ap->logbufs < XLOG_MIN_ICLOGS ||
 232              ap->logbufs > XLOG_MAX_ICLOGS)) {
 233                 cmn_err(CE_WARN,
 234                         "XFS: invalid logbufs value: %d [not %d-%d]",
 235                         ap->logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
 236                 return XFS_ERROR(EINVAL);
 237         }
 238         mp->m_logbufs = ap->logbufs;
 239         if (ap->logbufsize != -1 &&
 240             ap->logbufsize !=  0 &&
 241             ap->logbufsize != 16 * 1024 &&
 242             ap->logbufsize != 32 * 1024 &&
 243             ap->logbufsize != 64 * 1024 &&
 244             ap->logbufsize != 128 * 1024 &&
 245             ap->logbufsize != 256 * 1024) {
 246                 cmn_err(CE_WARN,
 247         "XFS: invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
 248                         ap->logbufsize);
 249                 return XFS_ERROR(EINVAL);
 250         }
 251         mp->m_ihsize = ap->ihashsize;
 252         mp->m_logbsize = ap->logbufsize;
 253         mp->m_fsname_len = strlen(ap->fsname) + 1;
 254         mp->m_fsname = kmem_alloc(mp->m_fsname_len, KM_SLEEP);
 255         strcpy(mp->m_fsname, ap->fsname);
 256         if (ap->rtname[0]) {
 257                 mp->m_rtname = kmem_alloc(strlen(ap->rtname) + 1, KM_SLEEP);
 258                 strcpy(mp->m_rtname, ap->rtname);
 259         }
 260         if (ap->logname[0]) {
 261                 mp->m_logname = kmem_alloc(strlen(ap->logname) + 1, KM_SLEEP);
 262                 strcpy(mp->m_logname, ap->logname);
 263         }
 264
 265         if (ap->flags & XFSMNT_WSYNC)
 266                 mp->m_flags |= XFS_MOUNT_WSYNC;
 267 #if XFS_BIG_INUMS
 268         if (ap->flags & XFSMNT_INO64) {
 269                 mp->m_flags |= XFS_MOUNT_INO64;
 270                 mp->m_inoadd = XFS_INO64_OFFSET;
 271         }
 272 #endif
 273         if (ap->flags & XFSMNT_RETERR)
 274                 mp->m_flags |= XFS_MOUNT_RETERR;
 275         if (ap->flags & XFSMNT_NOALIGN)
 276                 mp->m_flags |= XFS_MOUNT_NOALIGN;
 277         if (ap->flags & XFSMNT_SWALLOC)
 278                 mp->m_flags |= XFS_MOUNT_SWALLOC;
 279         if (ap->flags & XFSMNT_OSYNCISOSYNC)
 280                 mp->m_flags |= XFS_MOUNT_OSYNCISOSYNC;
 281         if (ap->flags & XFSMNT_32BITINODES)
 282                 mp->m_flags |= XFS_MOUNT_32BITINODES;
 283
 284         if (ap->flags & XFSMNT_IOSIZE) {
 285                 if (ap->iosizelog > XFS_MAX_IO_LOG ||
 286                     ap->iosizelog < XFS_MIN_IO_LOG) {
 287                         cmn_err(CE_WARN,
 288                 "XFS: invalid log iosize: %d [not %d-%d]",
 289                                 ap->iosizelog, XFS_MIN_IO_LOG,
 290                                 XFS_MAX_IO_LOG);
 291                         return XFS_ERROR(EINVAL);
 292                 }
 293
 294                 mp->m_flags |= XFS_MOUNT_DFLT_IOSIZE;
 295                 mp->m_readio_log = mp->m_writeio_log = ap->iosizelog;
 296         }
 297
 298         if (ap->flags & XFSMNT_IHASHSIZE)
 299                 mp->m_flags |= XFS_MOUNT_IHASHSIZE;
 300         if (ap->flags & XFSMNT_IDELETE)
 301                 mp->m_flags |= XFS_MOUNT_IDELETE;
 302         if (ap->flags & XFSMNT_DIRSYNC)
 303                 mp->m_flags |= XFS_MOUNT_DIRSYNC;
 304         if (ap->flags & XFSMNT_ATTR2)
 305                 mp->m_flags |= XFS_MOUNT_ATTR2;
 306
 307         if (ap->flags2 & XFSMNT2_COMPAT_IOSIZE)
 308                 mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE;
 309
 310         /*
 311          * no recovery flag requires a read-only mount
 312          */
 313         if (ap->flags & XFSMNT_NORECOVERY) {
 314                 if (!(vfs->vfs_flag & VFS_RDONLY)) {
 315                         cmn_err(CE_WARN,
 316         "XFS: tried to mount a FS read-write without recovery!");
 317                         return XFS_ERROR(EINVAL);
 318                 }
 319                 mp->m_flags |= XFS_MOUNT_NORECOVERY;
 320         }
 321
 322         if (ap->flags & XFSMNT_NOUUID)
 323                 mp->m_flags |= XFS_MOUNT_NOUUID;
 324         if (ap->flags & XFSMNT_BARRIER)
 325                 mp->m_flags |= XFS_MOUNT_BARRIER;
 326         else
 327                 mp->m_flags &= ~XFS_MOUNT_BARRIER;
 328
 329         if (ap->flags2 & XFSMNT2_FILESTREAMS)
 330                 mp->m_flags |= XFS_MOUNT_FILESTREAMS;
 331
 332         return 0;
 333 }
 334
 335 /*
 336  * This function fills in xfs_mount_t fields based on mount args.
 337  * Note: the superblock _has_ now been read in.
 338  */
 339 STATIC int
 340 xfs_finish_flags(
 341         struct bhv_vfs          *vfs,
 342         struct xfs_mount_args   *ap,
 343         struct xfs_mount        *mp)
 344 {
 345         int                     ronly = (vfs->vfs_flag & VFS_RDONLY);
 346
 347         /* Fail a mount where the logbuf is smaller then the log stripe */
 348         if (XFS_SB_VERSION_HASLOGV2(&mp->m_sb)) {
 349                 if ((ap->logbufsize <= 0) &&
 350                     (mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE)) {
 351                         mp->m_logbsize = mp->m_sb.sb_logsunit;
 352                 } else if (ap->logbufsize > 0 &&
 353                            ap->logbufsize < mp->m_sb.sb_logsunit) {
 354                         cmn_err(CE_WARN,
 355         "XFS: logbuf size must be greater than or equal to log stripe size");
 356                         return XFS_ERROR(EINVAL);
 357                 }
 358         } else {
 359                 /* Fail a mount if the logbuf is larger than 32K */
 360                 if (ap->logbufsize > XLOG_BIG_RECORD_BSIZE) {
 361                         cmn_err(CE_WARN,
 362         "XFS: logbuf size for version 1 logs must be 16K or 32K");
 363                         return XFS_ERROR(EINVAL);
 364                 }
 365         }
 366
 367         if (XFS_SB_VERSION_HASATTR2(&mp->m_sb)) {
 368                 mp->m_flags |= XFS_MOUNT_ATTR2;
 369         }
 370
 371         /*
 372          * prohibit r/w mounts of read-only filesystems
 373          */
 374         if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !ronly) {
 375                 cmn_err(CE_WARN,
 376         "XFS: cannot mount a read-only filesystem as read-write");
 377                 return XFS_ERROR(EROFS);
 378         }
 379
 380         /*
 381          * check for shared mount.
 382          */
 383         if (ap->flags & XFSMNT_SHARED) {
 384                 if (!XFS_SB_VERSION_HASSHARED(&mp->m_sb))
 385                         return XFS_ERROR(EINVAL);
 386
 387                 /*
 388                  * For IRIX 6.5, shared mounts must have the shared
 389                  * version bit set, have the persistent readonly
 390                  * field set, must be version 0 and can only be mounted
 391                  * read-only.
 392                  */
 393                 if (!ronly || !(mp->m_sb.sb_flags & XFS_SBF_READONLY) ||
 394                      (mp->m_sb.sb_shared_vn != 0))
 395                         return XFS_ERROR(EINVAL);
 396
 397                 mp->m_flags |= XFS_MOUNT_SHARED;
 398
 399                 /*
 400                  * Shared XFS V0 can't deal with DMI.  Return EINVAL.
 401                  */
 402                 if (mp->m_sb.sb_shared_vn == 0 && (ap->flags & XFSMNT_DMAPI))
 403                         return XFS_ERROR(EINVAL);
 404         }
 405
 406         return 0;
 407 }
 408
 409 /*
 410  * xfs_mount
 411  *
 412  * The file system configurations are:
 413  *      (1) device (partition) with data and internal log
 414  *      (2) logical volume with data and log subvolumes.
 415  *      (3) logical volume with data, log, and realtime subvolumes.
 416  *
 417  * We only have to handle opening the log and realtime volumes here if
 418  * they are present.  The data subvolume has already been opened by
 419  * get_sb_bdev() and is stored in vfsp->vfs_super->s_bdev.
 420  */
 421 STATIC int
 422 xfs_mount(
 423         struct bhv_desc         *bhvp,
 424         struct xfs_mount_args   *args,
 425         cred_t                  *credp)
 426 {
 427         struct bhv_vfs          *vfsp = bhvtovfs(bhvp);
 428         struct bhv_desc         *p;
 429         struct xfs_mount        *mp = XFS_BHVTOM(bhvp);
 430         struct block_device     *ddev, *logdev, *rtdev;
 431         int                     flags = 0, error;
 432
 433         ddev = vfsp->vfs_super->s_bdev;
 434         logdev = rtdev = NULL;
 435
 436         /*
 437          * Setup xfs_mount function vectors from available behaviors
 438          */
 439         p = vfs_bhv_lookup(vfsp, VFS_POSITION_DM);
 440         mp->m_dm_ops = p ? *(xfs_dmops_t *) vfs_bhv_custom(p) : xfs_dmcore_stub;
 441         p = vfs_bhv_lookup(vfsp, VFS_POSITION_QM);
 442         mp->m_qm_ops = p ? *(xfs_qmops_t *) vfs_bhv_custom(p) : xfs_qmcore_stub;
 443         p = vfs_bhv_lookup(vfsp, VFS_POSITION_IO);
 444         mp->m_io_ops = p ? *(xfs_ioops_t *) vfs_bhv_custom(p) : xfs_iocore_xfs;
 445
 446         if (args->flags & XFSMNT_QUIET)
 447                 flags |= XFS_MFSI_QUIET;
 448
 449         /*
 450          * Open real time and log devices - order is important.
 451          */
 452         if (args->logname[0]) {
 453                 error = xfs_blkdev_get(mp, args->logname, &logdev);
 454                 if (error)
 455                         return error;
 456         }
 457         if (args->rtname[0]) {
 458                 error = xfs_blkdev_get(mp, args->rtname, &rtdev);
 459                 if (error) {
 460                         xfs_blkdev_put(logdev);
 461                         return error;
 462                 }
 463
 464                 if (rtdev == ddev || rtdev == logdev) {
 465                         cmn_err(CE_WARN,
 466         "XFS: Cannot mount filesystem with identical rtdev and ddev/logdev.");
 467                         xfs_blkdev_put(logdev);
 468                         xfs_blkdev_put(rtdev);
 469                         return EINVAL;
 470                 }
 471         }
 472
 473         /*
 474          * Setup xfs_mount buffer target pointers
 475          */
 476         error = ENOMEM;
 477         mp->m_ddev_targp = xfs_alloc_buftarg(ddev, 0);
 478         if (!mp->m_ddev_targp) {
 479                 xfs_blkdev_put(logdev);
 480                 xfs_blkdev_put(rtdev);
 481                 return error;
 482         }
 483         if (rtdev) {
 484                 mp->m_rtdev_targp = xfs_alloc_buftarg(rtdev, 1);
 485                 if (!mp->m_rtdev_targp)
 486                         goto error0;
 487         }
 488         mp->m_logdev_targp = (logdev && logdev != ddev) ?
 489                                 xfs_alloc_buftarg(logdev, 1) : mp->m_ddev_targp;
 490         if (!mp->m_logdev_targp)
 491                 goto error0;
 492
 493         /*
 494          * Setup flags based on mount(2) options and then the superblock
 495          */
 496         error = xfs_start_flags(vfsp, args, mp);
 497         if (error)
 498                 goto error1;
 499         error = xfs_readsb(mp, flags);
 500         if (error)
 501                 goto error1;
 502         error = xfs_finish_flags(vfsp, args, mp);
 503         if (error)
 504                 goto error2;
 505
 506         /*
 507          * Setup xfs_mount buffer target pointers based on superblock
 508          */
 509         error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_blocksize,
 510                                     mp->m_sb.sb_sectsize);
 511         if (!error && logdev && logdev != ddev) {
 512                 unsigned int    log_sector_size = BBSIZE;
 513
 514                 if (XFS_SB_VERSION_HASSECTOR(&mp->m_sb))
 515                         log_sector_size = mp->m_sb.sb_logsectsize;
 516                 error = xfs_setsize_buftarg(mp->m_logdev_targp,
 517                                             mp->m_sb.sb_blocksize,
 518                                             log_sector_size);
 519         }
 520         if (!error && rtdev)
 521                 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
 522                                             mp->m_sb.sb_blocksize,
 523                                             mp->m_sb.sb_sectsize);
 524         if (error)
 525                 goto error2;
 526
 527         if (mp->m_flags & XFS_MOUNT_BARRIER)
 528                 xfs_mountfs_check_barriers(mp);
 529
 530         if ((error = xfs_filestream_mount(mp)))
 531                 goto error2;
 532
 533         error = XFS_IOINIT(vfsp, args, flags);
 534         if (error)
 535                 goto error2;
 536
 537         return 0;
 538
 539 error2:
 540         if (mp->m_sb_bp)
 541                 xfs_freesb(mp);
 542 error1:
 543         xfs_binval(mp->m_ddev_targp);
 544         if (logdev && logdev != ddev)
 545                 xfs_binval(mp->m_logdev_targp);
 546         if (rtdev)
 547                 xfs_binval(mp->m_rtdev_targp);
 548 error0:
 549         xfs_unmountfs_close(mp, credp);
 550         return error;
 551 }
 552
 553 STATIC int
 554 xfs_unmount(
 555         bhv_desc_t      *bdp,
 556         int             flags,
 557         cred_t          *credp)
 558 {
 559         bhv_vfs_t       *vfsp = bhvtovfs(bdp);
 560         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
 561         xfs_inode_t     *rip;
 562         bhv_vnode_t     *rvp;
 563         int             unmount_event_wanted = 0;
 564         int             unmount_event_flags = 0;
 565         int             xfs_unmountfs_needed = 0;
 566         int             error;
 567
 568         rip = mp->m_rootip;
 569         rvp = XFS_ITOV(rip);
 570
 571         if (vfsp->vfs_flag & VFS_DMI) {
 572                 error = XFS_SEND_PREUNMOUNT(mp, vfsp,
 573                                 rvp, DM_RIGHT_NULL, rvp, DM_RIGHT_NULL,
 574                                 NULL, NULL, 0, 0,
 575                                 (mp->m_dmevmask & (1<<DM_EVENT_PREUNMOUNT))?
 576                                         0:DM_FLAGS_UNWANTED);
 577                         if (error)
 578                                 return XFS_ERROR(error);
 579                 unmount_event_wanted = 1;
 580                 unmount_event_flags = (mp->m_dmevmask & (1<<DM_EVENT_UNMOUNT))?
 581                                         0 : DM_FLAGS_UNWANTED;
 582         }
 583
 584         /*
 585          * First blow any referenced inode from this file system
 586          * out of the reference cache, and delete the timer.
 587          */
 588         xfs_refcache_purge_mp(mp);
 589
 590         /*
 591          * Blow away any referenced inode in the filestreams cache.
 592          * This can and will cause log traffic as inodes go inactive
 593          * here.
 594          */
 595         xfs_filestream_unmount(mp);
 596
 597         XFS_bflush(mp->m_ddev_targp);
 598         error = xfs_unmount_flush(mp, 0);
 599         if (error)
 600                 goto out;
 601
 602         ASSERT(vn_count(rvp) == 1);
 603
 604         /*
 605          * Drop the reference count
 606          */
 607         VN_RELE(rvp);
 608
 609         /*
 610          * If we're forcing a shutdown, typically because of a media error,
 611          * we want to make sure we invalidate dirty pages that belong to
 612          * referenced vnodes as well.
 613          */
 614         if (XFS_FORCED_SHUTDOWN(mp)) {
 615                 error = xfs_sync(&mp->m_bhv,
 616                          (SYNC_WAIT | SYNC_CLOSE), credp);
 617                 ASSERT(error != EFSCORRUPTED);
 618         }
 619         xfs_unmountfs_needed = 1;
 620
 621 out:
 622         /*      Send DMAPI event, if required.
 623          *      Then do xfs_unmountfs() if needed.
 624          *      Then return error (or zero).
 625          */
 626         if (unmount_event_wanted) {
 627                 /* Note: mp structure must still exist for
 628                  * XFS_SEND_UNMOUNT() call.
 629                  */
 630                 XFS_SEND_UNMOUNT(mp, vfsp, error == 0 ? rvp : NULL,
 631                         DM_RIGHT_NULL, 0, error, unmount_event_flags);
 632         }
 633         if (xfs_unmountfs_needed) {
 634                 /*
 635                  * Call common unmount function to flush to disk
 636                  * and free the super block buffer & mount structures.
 637                  */
 638                 xfs_unmountfs(mp, credp);
 639         }
 640
 641         return XFS_ERROR(error);
 642 }
 643
 644 STATIC int
 645 xfs_quiesce_fs(
 646         xfs_mount_t             *mp)
 647 {
 648         int                     count = 0, pincount;
 649
 650         xfs_refcache_purge_mp(mp);
 651         xfs_flush_buftarg(mp->m_ddev_targp, 0);
 652         xfs_finish_reclaim_all(mp, 0);
 653
 654         /* This loop must run at least twice.
 655          * The first instance of the loop will flush
 656          * most meta data but that will generate more
 657          * meta data (typically directory updates).
 658          * Which then must be flushed and logged before
 659          * we can write the unmount record.
 660          */
 661         do {
 662                 xfs_syncsub(mp, SYNC_INODE_QUIESCE, NULL);
 663                 pincount = xfs_flush_buftarg(mp->m_ddev_targp, 1);
 664                 if (!pincount) {
 665                         delay(50);
 666                         count++;
 667                 }
 668         } while (count < 2);
 669
 670         return 0;
 671 }
 672
 673 /*
 674  * Second stage of a quiesce. The data is already synced, now we have to take
 675  * care of the metadata. New transactions are already blocked, so we need to
 676  * wait for any remaining transactions to drain out before proceding.
 677  */
 678 STATIC void
 679 xfs_attr_quiesce(
 680         xfs_mount_t     *mp)
 681 {
 682         /* wait for all modifications to complete */
 683         while (atomic_read(&mp->m_active_trans) > 0)
 684                 delay(100);
 685
 686         /* flush inodes and push all remaining buffers out to disk */
 687         xfs_quiesce_fs(mp);
 688
 689         ASSERT_ALWAYS(atomic_read(&mp->m_active_trans) == 0);
 690
 691         /* Push the superblock and write an unmount record */
 692         xfs_log_sbcount(mp, 1);
 693         xfs_log_unmount_write(mp);
 694         xfs_unmountfs_writesb(mp);
 695 }
 696
 697 STATIC int
 698 xfs_mntupdate(
 699         bhv_desc_t                      *bdp,
 700         int                             *flags,
 701         struct xfs_mount_args           *args)
 702 {
 703         bhv_vfs_t       *vfsp = bhvtovfs(bdp);
 704         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
 705
 706         if (!(*flags & MS_RDONLY)) {                    /* rw/ro -> rw */
 707                 if (vfsp->vfs_flag & VFS_RDONLY)
 708                         vfsp->vfs_flag &= ~VFS_RDONLY;
 709                 if (args->flags & XFSMNT_BARRIER) {
 710                         mp->m_flags |= XFS_MOUNT_BARRIER;
 711                         xfs_mountfs_check_barriers(mp);
 712                 } else {
 713                         mp->m_flags &= ~XFS_MOUNT_BARRIER;
 714                 }
 715         } else if (!(vfsp->vfs_flag & VFS_RDONLY)) {    /* rw -> ro */
 716                 xfs_filestream_flush(mp);
 717                 bhv_vfs_sync(vfsp, SYNC_DATA_QUIESCE, NULL);
 718                 xfs_attr_quiesce(mp);
 719                 vfsp->vfs_flag |= VFS_RDONLY;
 720         }
 721         return 0;
 722 }
 723
 724 /*
 725  * xfs_unmount_flush implements a set of flush operation on special
 726  * inodes, which are needed as a separate set of operations so that
 727  * they can be called as part of relocation process.
 728  */
 729 int
 730 xfs_unmount_flush(
 731         xfs_mount_t     *mp,            /* Mount structure we are getting
 732                                            rid of. */
 733         int             relocation)     /* Called from vfs relocation. */
 734 {
 735         xfs_inode_t     *rip = mp->m_rootip;
 736         xfs_inode_t     *rbmip;
 737         xfs_inode_t     *rsumip = NULL;
 738         bhv_vnode_t     *rvp = XFS_ITOV(rip);
 739         int             error;
 740
 741         xfs_ilock(rip, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
 742         xfs_iflock(rip);
 743
 744         /*
 745          * Flush out the real time inodes.
 746          */
 747         if ((rbmip = mp->m_rbmip) != NULL) {
 748                 xfs_ilock(rbmip, XFS_ILOCK_EXCL);
 749                 xfs_iflock(rbmip);
 750                 error = xfs_iflush(rbmip, XFS_IFLUSH_SYNC);
 751                 xfs_iunlock(rbmip, XFS_ILOCK_EXCL);
 752
 753                 if (error == EFSCORRUPTED)
 754                         goto fscorrupt_out;
 755
 756                 ASSERT(vn_count(XFS_ITOV(rbmip)) == 1);
 757
 758                 rsumip = mp->m_rsumip;
 759                 xfs_ilock(rsumip, XFS_ILOCK_EXCL);
 760                 xfs_iflock(rsumip);
 761                 error = xfs_iflush(rsumip, XFS_IFLUSH_SYNC);
 762                 xfs_iunlock(rsumip, XFS_ILOCK_EXCL);
 763
 764                 if (error == EFSCORRUPTED)
 765                         goto fscorrupt_out;
 766
 767                 ASSERT(vn_count(XFS_ITOV(rsumip)) == 1);
 768         }
 769
 770         /*
 771          * Synchronously flush root inode to disk
 772          */
 773         error = xfs_iflush(rip, XFS_IFLUSH_SYNC);
 774         if (error == EFSCORRUPTED)
 775                 goto fscorrupt_out2;
 776
 777         if (vn_count(rvp) != 1 && !relocation) {
 778                 xfs_iunlock(rip, XFS_ILOCK_EXCL);
 779                 return XFS_ERROR(EBUSY);
 780         }
 781
 782         /*
 783          * Release dquot that rootinode, rbmino and rsumino might be holding,
 784          * flush and purge the quota inodes.
 785          */
 786         error = XFS_QM_UNMOUNT(mp);
 787         if (error == EFSCORRUPTED)
 788                 goto fscorrupt_out2;
 789
 790         if (rbmip) {
 791                 VN_RELE(XFS_ITOV(rbmip));
 792                 VN_RELE(XFS_ITOV(rsumip));
 793         }
 794
 795         xfs_iunlock(rip, XFS_ILOCK_EXCL);
 796         return 0;
 797
 798 fscorrupt_out:
 799         xfs_ifunlock(rip);
 800
 801 fscorrupt_out2:
 802         xfs_iunlock(rip, XFS_ILOCK_EXCL);
 803
 804         return XFS_ERROR(EFSCORRUPTED);
 805 }
 806
 807 /*
 808  * xfs_root extracts the root vnode from a vfs.
 809  *
 810  * vfsp -- the vfs struct for the desired file system
 811  * vpp  -- address of the caller's vnode pointer which should be
 812  *         set to the desired fs root vnode
 813  */
 814 STATIC int
 815 xfs_root(
 816         bhv_desc_t      *bdp,
 817         bhv_vnode_t     **vpp)
 818 {
 819         bhv_vnode_t     *vp;
 820
 821         vp = XFS_ITOV((XFS_BHVTOM(bdp))->m_rootip);
 822         VN_HOLD(vp);
 823         *vpp = vp;
 824         return 0;
 825 }
 826
 827 /*
 828  * xfs_statvfs
 829  *
 830  * Fill in the statvfs structure for the given file system.  We use
 831  * the superblock lock in the mount structure to ensure a consistent
 832  * snapshot of the counters returned.
 833  */
 834 STATIC int
 835 xfs_statvfs(
 836         bhv_desc_t      *bdp,
 837         bhv_statvfs_t   *statp,
 838         bhv_vnode_t     *vp)
 839 {
 840         __uint64_t      fakeinos;
 841         xfs_extlen_t    lsize;
 842         xfs_mount_t     *mp;
 843         xfs_sb_t        *sbp;
 844         unsigned long   s;
 845
 846         mp = XFS_BHVTOM(bdp);
 847         sbp = &(mp->m_sb);
 848
 849         statp->f_type = XFS_SB_MAGIC;
 850
 851         xfs_icsb_sync_counters_flags(mp, XFS_ICSB_LAZY_COUNT);
 852         s = XFS_SB_LOCK(mp);
 853         statp->f_bsize = sbp->sb_blocksize;
 854         lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
 855         statp->f_blocks = sbp->sb_dblocks - lsize;
 856         statp->f_bfree = statp->f_bavail =
 857                                 sbp->sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
 858         fakeinos = statp->f_bfree << sbp->sb_inopblog;
 859 #if XFS_BIG_INUMS
 860         fakeinos += mp->m_inoadd;
 861 #endif
 862         statp->f_files =
 863             MIN(sbp->sb_icount + fakeinos, (__uint64_t)XFS_MAXINUMBER);
 864         if (mp->m_maxicount)
 865 #if XFS_BIG_INUMS
 866                 if (!mp->m_inoadd)
 867 #endif
 868                         statp->f_files = min_t(typeof(statp->f_files),
 869                                                 statp->f_files,
 870                                                 mp->m_maxicount);
 871         statp->f_ffree = statp->f_files - (sbp->sb_icount - sbp->sb_ifree);
 872         XFS_SB_UNLOCK(mp, s);
 873
 874         xfs_statvfs_fsid(statp, mp);
 875         statp->f_namelen = MAXNAMELEN - 1;
 876
 877         return 0;
 878 }
 879
 880
 881 /*
 882  * xfs_sync flushes any pending I/O to file system vfsp.
 883  *
 884  * This routine is called by vfs_sync() to make sure that things make it
 885  * out to disk eventually, on sync() system calls to flush out everything,
 886  * and when the file system is unmounted.  For the vfs_sync() case, all
 887  * we really need to do is sync out the log to make all of our meta-data
 888  * updates permanent (except for timestamps).  For calls from pflushd(),
 889  * dirty pages are kept moving by calling pdflush() on the inodes
 890  * containing them.  We also flush the inodes that we can lock without
 891  * sleeping and the superblock if we can lock it without sleeping from
 892  * vfs_sync() so that items at the tail of the log are always moving out.
 893  *
 894  * Flags:
 895  *      SYNC_BDFLUSH - We're being called from vfs_sync() so we don't want
 896  *                     to sleep if we can help it.  All we really need
 897  *                     to do is ensure that the log is synced at least
 898  *                     periodically.  We also push the inodes and
 899  *                     superblock if we can lock them without sleeping
 900  *                      and they are not pinned.
 901  *      SYNC_ATTR    - We need to flush the inodes.  If SYNC_BDFLUSH is not
 902  *                     set, then we really want to lock each inode and flush
 903  *                     it.
 904  *      SYNC_WAIT    - All the flushes that take place in this call should
 905  *                     be synchronous.
 906  *      SYNC_DELWRI  - This tells us to push dirty pages associated with
 907  *                     inodes.  SYNC_WAIT and SYNC_BDFLUSH are used to
 908  *                     determine if they should be flushed sync, async, or
 909  *                     delwri.
 910  *      SYNC_CLOSE   - This flag is passed when the system is being
 911  *                     unmounted.  We should sync and invalidate everything.
 912  *      SYNC_FSDATA  - This indicates that the caller would like to make
 913  *                     sure the superblock is safe on disk.  We can ensure
 914  *                     this by simply making sure the log gets flushed
 915  *                     if SYNC_BDFLUSH is set, and by actually writing it
 916  *                     out otherwise.
 917  *      SYNC_IOWAIT  - The caller wants us to wait for all data I/O to complete
 918  *                     before we return (including direct I/O). Forms the drain
 919  *                     side of the write barrier needed to safely quiesce the
 920  *                     filesystem.
 921  *
 922  */
 923 /*ARGSUSED*/
 924 STATIC int
 925 xfs_sync(
 926         bhv_desc_t      *bdp,
 927         int             flags,
 928         cred_t          *credp)
 929 {
 930         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
 931
 932         if (flags & SYNC_IOWAIT)
 933                 xfs_filestream_flush(mp);
 934
 935         return xfs_syncsub(mp, flags, NULL);
 936 }
 937
 938 /*
 939  * xfs sync routine for internal use
 940  *
 941  * This routine supports all of the flags defined for the generic vfs_sync
 942  * interface as explained above under xfs_sync.
 943  *
 944  */
 945 int
 946 xfs_sync_inodes(
 947         xfs_mount_t     *mp,
 948         int             flags,
 949         int             *bypassed)
 950 {
 951         xfs_inode_t     *ip = NULL;
 952         xfs_inode_t     *ip_next;
 953         xfs_buf_t       *bp;
 954         bhv_vnode_t     *vp = NULL;
 955         int             error;
 956         int             last_error;
 957         uint64_t        fflag;
 958         uint            lock_flags;
 959         uint            base_lock_flags;
 960         boolean_t       mount_locked;
 961         boolean_t       vnode_refed;
 962         int             preempt;
 963         xfs_dinode_t    *dip;
 964         xfs_iptr_t      *ipointer;
 965 #ifdef DEBUG
 966         boolean_t       ipointer_in = B_FALSE;
 967
 968 #define IPOINTER_SET    ipointer_in = B_TRUE
 969 #define IPOINTER_CLR    ipointer_in = B_FALSE
 970 #else
 971 #define IPOINTER_SET
 972 #define IPOINTER_CLR
 973 #endif
 974
 975
 976 /* Insert a marker record into the inode list after inode ip. The list
 977  * must be locked when this is called. After the call the list will no
 978  * longer be locked.
 979  */
 980 #define IPOINTER_INSERT(ip, mp) { \
 981                 ASSERT(ipointer_in == B_FALSE); \
 982                 ipointer->ip_mnext = ip->i_mnext; \
 983                 ipointer->ip_mprev = ip; \
 984                 ip->i_mnext = (xfs_inode_t *)ipointer; \
 985                 ipointer->ip_mnext->i_mprev = (xfs_inode_t *)ipointer; \
 986                 preempt = 0; \
 987                 XFS_MOUNT_IUNLOCK(mp); \
 988                 mount_locked = B_FALSE; \
 989                 IPOINTER_SET; \
 990         }
 991
 992 /* Remove the marker from the inode list. If the marker was the only item
 993  * in the list then there are no remaining inodes and we should zero out
 994  * the whole list. If we are the current head of the list then move the head
 995  * past us.
 996  */
 997 #define IPOINTER_REMOVE(ip, mp) { \
 998                 ASSERT(ipointer_in == B_TRUE); \
 999                 if (ipointer->ip_mnext != (xfs_inode_t *)ipointer) { \
1000                         ip = ipointer->ip_mnext; \
1001                         ip->i_mprev = ipointer->ip_mprev; \
1002                         ipointer->ip_mprev->i_mnext = ip; \
1003                         if (mp->m_inodes == (xfs_inode_t *)ipointer) { \
1004                                 mp->m_inodes = ip; \
1005                         } \
1006                 } else { \
1007                         ASSERT(mp->m_inodes == (xfs_inode_t *)ipointer); \
1008                         mp->m_inodes = NULL; \
1009                         ip = NULL; \
1010                 } \
1011                 IPOINTER_CLR; \
1012         }
1013
1014 #define XFS_PREEMPT_MASK        0x7f
1015
1016         if (bypassed)
1017                 *bypassed = 0;
1018         if (XFS_MTOVFS(mp)->vfs_flag & VFS_RDONLY)
1019                 return 0;
1020         error = 0;
1021         last_error = 0;
1022         preempt = 0;
1023
1024         /* Allocate a reference marker */
1025         ipointer = (xfs_iptr_t *)kmem_zalloc(sizeof(xfs_iptr_t), KM_SLEEP);
1026
1027         fflag = XFS_B_ASYNC;            /* default is don't wait */
1028         if (flags & (SYNC_BDFLUSH | SYNC_DELWRI))
1029                 fflag = XFS_B_DELWRI;
1030         if (flags & SYNC_WAIT)
1031                 fflag = 0;              /* synchronous overrides all */
1032
1033         base_lock_flags = XFS_ILOCK_SHARED;
1034         if (flags & (SYNC_DELWRI | SYNC_CLOSE)) {
1035                 /*
1036                  * We need the I/O lock if we're going to call any of
1037                  * the flush/inval routines.
1038                  */
1039                 base_lock_flags |= XFS_IOLOCK_SHARED;
1040         }
1041
1042         XFS_MOUNT_ILOCK(mp);
1043
1044         ip = mp->m_inodes;
1045
1046         mount_locked = B_TRUE;
1047         vnode_refed  = B_FALSE;
1048
1049         IPOINTER_CLR;
1050
1051         do {
1052                 ASSERT(ipointer_in == B_FALSE);
1053                 ASSERT(vnode_refed == B_FALSE);
1054
1055                 lock_flags = base_lock_flags;
1056
1057                 /*
1058                  * There were no inodes in the list, just break out
1059                  * of the loop.
1060                  */
1061                 if (ip == NULL) {
1062                         break;
1063                 }
1064
1065                 /*
1066                  * We found another sync thread marker - skip it
1067                  */
1068                 if (ip->i_mount == NULL) {
1069                         ip = ip->i_mnext;
1070                         continue;
1071                 }
1072
1073                 vp = XFS_ITOV_NULL(ip);
1074
1075                 /*
1076                  * If the vnode is gone then this is being torn down,
1077                  * call reclaim if it is flushed, else let regular flush
1078                  * code deal with it later in the loop.
1079                  */
1080
1081                 if (vp == NULL) {
1082                         /* Skip ones already in reclaim */
1083                         if (ip->i_flags & XFS_IRECLAIM) {
1084                                 ip = ip->i_mnext;
1085                                 continue;
1086                         }
1087                         if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0) {
1088                                 ip = ip->i_mnext;
1089                         } else if ((xfs_ipincount(ip) == 0) &&
1090                                     xfs_iflock_nowait(ip)) {
1091                                 IPOINTER_INSERT(ip, mp);
1092
1093                                 xfs_finish_reclaim(ip, 1,
1094                                                 XFS_IFLUSH_DELWRI_ELSE_ASYNC);
1095
1096                                 XFS_MOUNT_ILOCK(mp);
1097                                 mount_locked = B_TRUE;
1098                                 IPOINTER_REMOVE(ip, mp);
1099                         } else {
1100                                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1101                                 ip = ip->i_mnext;
1102                         }
1103                         continue;
1104                 }
1105
1106                 if (VN_BAD(vp)) {
1107                         ip = ip->i_mnext;
1108                         continue;
1109                 }
1110
1111                 if (XFS_FORCED_SHUTDOWN(mp) && !(flags & SYNC_CLOSE)) {
1112                         XFS_MOUNT_IUNLOCK(mp);
1113                         kmem_free(ipointer, sizeof(xfs_iptr_t));
1114                         return 0;
1115                 }
1116
1117                 /*
1118                  * If this is just vfs_sync() or pflushd() calling
1119                  * then we can skip inodes for which it looks like
1120                  * there is nothing to do.  Since we don't have the
1121                  * inode locked this is racy, but these are periodic
1122                  * calls so it doesn't matter.  For the others we want
1123                  * to know for sure, so we at least try to lock them.
1124                  */
1125                 if (flags & SYNC_BDFLUSH) {
1126                         if (((ip->i_itemp == NULL) ||
1127                              !(ip->i_itemp->ili_format.ilf_fields &
1128                                XFS_ILOG_ALL)) &&
1129                             (ip->i_update_core == 0)) {
1130                                 ip = ip->i_mnext;
1131                                 continue;
1132                         }
1133                 }
1134
1135                 /*
1136                  * Try to lock without sleeping.  We're out of order with
1137                  * the inode list lock here, so if we fail we need to drop
1138                  * the mount lock and try again.  If we're called from
1139                  * bdflush() here, then don't bother.
1140                  *
1141                  * The inode lock here actually coordinates with the
1142                  * almost spurious inode lock in xfs_ireclaim() to prevent
1143                  * the vnode we handle here without a reference from
1144                  * being freed while we reference it.  If we lock the inode
1145                  * while it's on the mount list here, then the spurious inode
1146                  * lock in xfs_ireclaim() after the inode is pulled from
1147                  * the mount list will sleep until we release it here.
1148                  * This keeps the vnode from being freed while we reference
1149                  * it.
1150                  */
1151                 if (xfs_ilock_nowait(ip, lock_flags) == 0) {
1152                         if ((flags & SYNC_BDFLUSH) || (vp == NULL)) {
1153                                 ip = ip->i_mnext;
1154                                 continue;
1155                         }
1156
1157                         vp = vn_grab(vp);
1158                         if (vp == NULL) {
1159                                 ip = ip->i_mnext;
1160                                 continue;
1161                         }
1162
1163                         IPOINTER_INSERT(ip, mp);
1164                         xfs_ilock(ip, lock_flags);
1165
1166                         ASSERT(vp == XFS_ITOV(ip));
1167                         ASSERT(ip->i_mount == mp);
1168
1169                         vnode_refed = B_TRUE;
1170                 }
1171
1172                 /* From here on in the loop we may have a marker record
1173                  * in the inode list.
1174                  */
1175
1176                 /*
1177                  * If we have to flush data or wait for I/O completion
1178                  * we need to drop the ilock that we currently hold.
1179                  * If we need to drop the lock, insert a marker if we
1180                  * have not already done so.
1181                  */
1182                 if ((flags & (SYNC_CLOSE|SYNC_IOWAIT)) ||
1183                     ((flags & SYNC_DELWRI) && VN_DIRTY(vp))) {
1184                         if (mount_locked) {
1185                                 IPOINTER_INSERT(ip, mp);
1186                         }
1187                         xfs_iunlock(ip, XFS_ILOCK_SHARED);
1188
1189                         if (flags & SYNC_CLOSE) {
1190                                 /* Shutdown case. Flush and invalidate. */
1191                                 if (XFS_FORCED_SHUTDOWN(mp))
1192                                         bhv_vop_toss_pages(vp, 0, -1, FI_REMAPF);
1193                                 else
1194                                         error = bhv_vop_flushinval_pages(vp, 0,
1195                                                                 -1, FI_REMAPF);
1196                         } else if ((flags & SYNC_DELWRI) && VN_DIRTY(vp)) {
1197                                 error = bhv_vop_flush_pages(vp, (xfs_off_t)0,
1198                                                         -1, fflag, FI_NONE);
1199                         }
1200
1201                         /*
1202                          * When freezing, we need to wait ensure all I/O (including direct
1203                          * I/O) is complete to ensure no further data modification can take
1204                          * place after this point
1205                          */
1206                         if (flags & SYNC_IOWAIT)
1207                                 vn_iowait(vp);
1208
1209                         xfs_ilock(ip, XFS_ILOCK_SHARED);
1210                 }
1211
1212                 if (flags & SYNC_BDFLUSH) {
1213                         if ((flags & SYNC_ATTR) &&
1214                             ((ip->i_update_core) ||
1215                              ((ip->i_itemp != NULL) &&
1216                               (ip->i_itemp->ili_format.ilf_fields != 0)))) {
1217
1218                                 /* Insert marker and drop lock if not already
1219                                  * done.
1220                                  */
1221                                 if (mount_locked) {
1222                                         IPOINTER_INSERT(ip, mp);
1223                                 }
1224
1225                                 /*
1226                                  * We don't want the periodic flushing of the
1227                                  * inodes by vfs_sync() to interfere with
1228                                  * I/O to the file, especially read I/O
1229                                  * where it is only the access time stamp
1230                                  * that is being flushed out.  To prevent
1231                                  * long periods where we have both inode
1232                                  * locks held shared here while reading the
1233                                  * inode's buffer in from disk, we drop the
1234                                  * inode lock while reading in the inode
1235                                  * buffer.  We have to release the buffer
1236                                  * and reacquire the inode lock so that they
1237                                  * are acquired in the proper order (inode
1238                                  * locks first).  The buffer will go at the
1239                                  * end of the lru chain, though, so we can
1240                                  * expect it to still be there when we go
1241                                  * for it again in xfs_iflush().
1242                                  */
1243                                 if ((xfs_ipincount(ip) == 0) &&
1244                                     xfs_iflock_nowait(ip)) {
1245
1246                                         xfs_ifunlock(ip);
1247                                         xfs_iunlock(ip, XFS_ILOCK_SHARED);
1248
1249                                         error = xfs_itobp(mp, NULL, ip,
1250                                                           &dip, &bp, 0, 0);
1251                                         if (!error) {
1252                                                 xfs_buf_relse(bp);
1253                                         } else {
1254                                                 /* Bailing out, remove the
1255                                                  * marker and free it.
1256                                                  */
1257                                                 XFS_MOUNT_ILOCK(mp);
1258                                                 IPOINTER_REMOVE(ip, mp);
1259                                                 XFS_MOUNT_IUNLOCK(mp);
1260
1261                                                 ASSERT(!(lock_flags &
1262                                                         XFS_IOLOCK_SHARED));
1263
1264                                                 kmem_free(ipointer,
1265                                                         sizeof(xfs_iptr_t));
1266                                                 return (0);
1267                                         }
1268
1269                                         /*
1270                                          * Since we dropped the inode lock,
1271                                          * the inode may have been reclaimed.
1272                                          * Therefore, we reacquire the mount
1273                                          * lock and check to see if we were the
1274                                          * inode reclaimed. If this happened
1275                                          * then the ipointer marker will no
1276                                          * longer point back at us. In this
1277                                          * case, move ip along to the inode
1278                                          * after the marker, remove the marker
1279                                          * and continue.
1280                                          */
1281                                         XFS_MOUNT_ILOCK(mp);
1282                                         mount_locked = B_TRUE;
1283
1284                                         if (ip != ipointer->ip_mprev) {
1285                                                 IPOINTER_REMOVE(ip, mp);
1286
1287                                                 ASSERT(!vnode_refed);
1288                                                 ASSERT(!(lock_flags &
1289                                                         XFS_IOLOCK_SHARED));
1290                                                 continue;
1291                                         }
1292
1293                                         ASSERT(ip->i_mount == mp);
1294
1295                                         if (xfs_ilock_nowait(ip,
1296                                                     XFS_ILOCK_SHARED) == 0) {
1297                                                 ASSERT(ip->i_mount == mp);
1298                                                 /*
1299                                                  * We failed to reacquire
1300                                                  * the inode lock without
1301                                                  * sleeping, so just skip
1302                                                  * the inode for now.  We
1303                                                  * clear the ILOCK bit from
1304                                                  * the lock_flags so that we
1305                                                  * won't try to drop a lock
1306                                                  * we don't hold below.
1307                                                  */
1308                                                 lock_flags &= ~XFS_ILOCK_SHARED;
1309                                                 IPOINTER_REMOVE(ip_next, mp);
1310                                         } else if ((xfs_ipincount(ip) == 0) &&
1311                                                    xfs_iflock_nowait(ip)) {
1312                                                 ASSERT(ip->i_mount == mp);
1313                                                 /*
1314                                                  * Since this is vfs_sync()
1315                                                  * calling we only flush the
1316                                                  * inode out if we can lock
1317                                                  * it without sleeping and
1318                                                  * it is not pinned.  Drop
1319                                                  * the mount lock here so
1320                                                  * that we don't hold it for
1321                                                  * too long. We already have
1322                                                  * a marker in the list here.
1323                                                  */
1324                                                 XFS_MOUNT_IUNLOCK(mp);
1325                                                 mount_locked = B_FALSE;
1326                                                 error = xfs_iflush(ip,
1327                                                            XFS_IFLUSH_DELWRI);
1328                                         } else {
1329                                                 ASSERT(ip->i_mount == mp);
1330                                                 IPOINTER_REMOVE(ip_next, mp);
1331                                         }
1332                                 }
1333
1334                         }
1335
1336                 } else {
1337                         if ((flags & SYNC_ATTR) &&
1338                             ((ip->i_update_core) ||
1339                              ((ip->i_itemp != NULL) &&
1340                               (ip->i_itemp->ili_format.ilf_fields != 0)))) {
1341                                 if (mount_locked) {
1342                                         IPOINTER_INSERT(ip, mp);
1343                                 }
1344
1345                                 if (flags & SYNC_WAIT) {
1346                                         xfs_iflock(ip);
1347                                         error = xfs_iflush(ip,
1348                                                            XFS_IFLUSH_SYNC);
1349                                 } else {
1350                                         /*
1351                                          * If we can't acquire the flush
1352                                          * lock, then the inode is already
1353                                          * being flushed so don't bother
1354                                          * waiting.  If we can lock it then
1355                                          * do a delwri flush so we can
1356                                          * combine multiple inode flushes
1357                                          * in each disk write.
1358                                          */
1359                                         if (xfs_iflock_nowait(ip)) {
1360                                                 error = xfs_iflush(ip,
1361                                                            XFS_IFLUSH_DELWRI);
1362                                         }
1363                                         else if (bypassed)
1364                                                 (*bypassed)++;
1365                                 }
1366                         }
1367                 }
1368
1369                 if (lock_flags != 0) {
1370                         xfs_iunlock(ip, lock_flags);
1371                 }
1372
1373                 if (vnode_refed) {
1374                         /*
1375                          * If we had to take a reference on the vnode
1376                          * above, then wait until after we've unlocked
1377                          * the inode to release the reference.  This is
1378                          * because we can be already holding the inode
1379                          * lock when VN_RELE() calls xfs_inactive().
1380                          *
1381                          * Make sure to drop the mount lock before calling
1382                          * VN_RELE() so that we don't trip over ourselves if
1383                          * we have to go for the mount lock again in the
1384                          * inactive code.
1385                          */
1386                         if (mount_locked) {
1387                                 IPOINTER_INSERT(ip, mp);
1388                         }
1389
1390                         VN_RELE(vp);
1391
1392                         vnode_refed = B_FALSE;
1393                 }
1394
1395                 if (error) {
1396                         last_error = error;
1397                 }
1398
1399                 /*
1400                  * bail out if the filesystem is corrupted.
1401                  */
1402                 if (error == EFSCORRUPTED)  {
1403                         if (!mount_locked) {
1404                                 XFS_MOUNT_ILOCK(mp);
1405                                 IPOINTER_REMOVE(ip, mp);
1406                         }
1407                         XFS_MOUNT_IUNLOCK(mp);
1408                         ASSERT(ipointer_in == B_FALSE);
1409                         kmem_free(ipointer, sizeof(xfs_iptr_t));
1410                         return XFS_ERROR(error);
1411                 }
1412
1413                 /* Let other threads have a chance at the mount lock
1414                  * if we have looped many times without dropping the
1415                  * lock.
1416                  */
1417                 if ((++preempt & XFS_PREEMPT_MASK) == 0) {
1418                         if (mount_locked) {
1419                                 IPOINTER_INSERT(ip, mp);
1420                         }
1421                 }
1422
1423                 if (mount_locked == B_FALSE) {
1424                         XFS_MOUNT_ILOCK(mp);
1425                         mount_locked = B_TRUE;
1426                         IPOINTER_REMOVE(ip, mp);
1427                         continue;
1428                 }
1429
1430                 ASSERT(ipointer_in == B_FALSE);
1431                 ip = ip->i_mnext;
1432
1433         } while (ip != mp->m_inodes);
1434
1435         XFS_MOUNT_IUNLOCK(mp);
1436
1437         ASSERT(ipointer_in == B_FALSE);
1438
1439         kmem_free(ipointer, sizeof(xfs_iptr_t));
1440         return XFS_ERROR(last_error);
1441 }
1442
1443 /*
1444  * xfs sync routine for internal use
1445  *
1446  * This routine supports all of the flags defined for the generic vfs_sync
1447  * interface as explained above under xfs_sync.
1448  *
1449  */
1450 int
1451 xfs_syncsub(
1452         xfs_mount_t     *mp,
1453         int             flags,
1454         int             *bypassed)
1455 {
1456         int             error = 0;
1457         int             last_error = 0;
1458         uint            log_flags = XFS_LOG_FORCE;
1459         xfs_buf_t       *bp;
1460         xfs_buf_log_item_t      *bip;
1461
1462         /*
1463          * Sync out the log.  This ensures that the log is periodically
1464          * flushed even if there is not enough activity to fill it up.
1465          */
1466         if (flags & SYNC_WAIT)
1467                 log_flags |= XFS_LOG_SYNC;
1468
1469         xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1470
1471         if (flags & (SYNC_ATTR|SYNC_DELWRI)) {
1472                 if (flags & SYNC_BDFLUSH)
1473                         xfs_finish_reclaim_all(mp, 1);
1474                 else
1475                         error = xfs_sync_inodes(mp, flags, bypassed);
1476         }
1477
1478         /*
1479          * Flushing out dirty data above probably generated more
1480          * log activity, so if this isn't vfs_sync() then flush
1481          * the log again.
1482          */
1483         if (flags & SYNC_DELWRI) {
1484                 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1485         }
1486
1487         if (flags & SYNC_FSDATA) {
1488                 /*
1489                  * If this is vfs_sync() then only sync the superblock
1490                  * if we can lock it without sleeping and it is not pinned.
1491                  */
1492                 if (flags & SYNC_BDFLUSH) {
1493                         bp = xfs_getsb(mp, XFS_BUF_TRYLOCK);
1494                         if (bp != NULL) {
1495                                 bip = XFS_BUF_FSPRIVATE(bp,xfs_buf_log_item_t*);
1496                                 if ((bip != NULL) &&
1497                                     xfs_buf_item_dirty(bip)) {
1498                                         if (!(XFS_BUF_ISPINNED(bp))) {
1499                                                 XFS_BUF_ASYNC(bp);
1500                                                 error = xfs_bwrite(mp, bp);
1501                                         } else {
1502                                                 xfs_buf_relse(bp);
1503                                         }
1504                                 } else {
1505                                         xfs_buf_relse(bp);
1506                                 }
1507                         }
1508                 } else {
1509                         bp = xfs_getsb(mp, 0);
1510                         /*
1511                          * If the buffer is pinned then push on the log so
1512                          * we won't get stuck waiting in the write for
1513                          * someone, maybe ourselves, to flush the log.
1514                          * Even though we just pushed the log above, we
1515                          * did not have the superblock buffer locked at
1516                          * that point so it can become pinned in between
1517                          * there and here.
1518                          */
1519                         if (XFS_BUF_ISPINNED(bp))
1520                                 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
1521                         if (flags & SYNC_WAIT)
1522                                 XFS_BUF_UNASYNC(bp);
1523                         else
1524                                 XFS_BUF_ASYNC(bp);
1525                         error = xfs_bwrite(mp, bp);
1526                 }
1527                 if (error) {
1528                         last_error = error;
1529                 }
1530         }
1531
1532         /*
1533          * If this is the periodic sync, then kick some entries out of
1534          * the reference cache.  This ensures that idle entries are
1535          * eventually kicked out of the cache.
1536          */
1537         if (flags & SYNC_REFCACHE) {
1538                 if (flags & SYNC_WAIT)
1539                         xfs_refcache_purge_mp(mp);
1540                 else
1541                         xfs_refcache_purge_some(mp);
1542         }
1543
1544         /*
1545          * If asked, update the disk superblock with incore counter values if we
1546          * are using non-persistent counters so that they don't get too far out
1547          * of sync if we crash or get a forced shutdown. We don't want to force
1548          * this to disk, just get a transaction into the iclogs....
1549          */
1550         if (flags & SYNC_SUPER)
1551                 xfs_log_sbcount(mp, 0);
1552
1553         /*
1554          * Now check to see if the log needs a "dummy" transaction.
1555          */
1556
1557         if (!(flags & SYNC_REMOUNT) && xfs_log_need_covered(mp)) {
1558                 xfs_trans_t *tp;
1559                 xfs_inode_t *ip;
1560
1561                 /*
1562                  * Put a dummy transaction in the log to tell
1563                  * recovery that all others are OK.
1564                  */
1565                 tp = xfs_trans_alloc(mp, XFS_TRANS_DUMMY1);
1566                 if ((error = xfs_trans_reserve(tp, 0,
1567                                 XFS_ICHANGE_LOG_RES(mp),
1568                                 0, 0, 0)))  {
1569                         xfs_trans_cancel(tp, 0);
1570                         return error;
1571                 }
1572
1573                 ip = mp->m_rootip;
1574                 xfs_ilock(ip, XFS_ILOCK_EXCL);
1575
1576                 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1577                 xfs_trans_ihold(tp, ip);
1578                 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1579                 error = xfs_trans_commit(tp, 0);
1580                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1581                 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1582         }
1583
1584         /*
1585          * When shutting down, we need to insure that the AIL is pushed
1586          * to disk or the filesystem can appear corrupt from the PROM.
1587          */
1588         if ((flags & (SYNC_CLOSE|SYNC_WAIT)) == (SYNC_CLOSE|SYNC_WAIT)) {
1589                 XFS_bflush(mp->m_ddev_targp);
1590                 if (mp->m_rtdev_targp) {
1591                         XFS_bflush(mp->m_rtdev_targp);
1592                 }
1593         }
1594
1595         return XFS_ERROR(last_error);
1596 }
1597
1598 /*
1599  * xfs_vget - called by DMAPI and NFSD to get vnode from file handle
1600  */
1601 STATIC int
1602 xfs_vget(
1603         bhv_desc_t      *bdp,
1604         bhv_vnode_t     **vpp,
1605         fid_t           *fidp)
1606 {
1607         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
1608         xfs_fid_t       *xfid = (struct xfs_fid *)fidp;
1609         xfs_inode_t     *ip;
1610         int             error;
1611         xfs_ino_t       ino;
1612         unsigned int    igen;
1613
1614         /*
1615          * Invalid.  Since handles can be created in user space and passed in
1616          * via gethandle(), this is not cause for a panic.
1617          */
1618         if (xfid->xfs_fid_len != sizeof(*xfid) - sizeof(xfid->xfs_fid_len))
1619                 return XFS_ERROR(EINVAL);
1620
1621         ino  = xfid->xfs_fid_ino;
1622         igen = xfid->xfs_fid_gen;
1623
1624         /*
1625          * NFS can sometimes send requests for ino 0.  Fail them gracefully.
1626          */
1627         if (ino == 0)
1628                 return XFS_ERROR(ESTALE);
1629
1630         error = xfs_iget(mp, NULL, ino, 0, XFS_ILOCK_SHARED, &ip, 0);
1631         if (error) {
1632                 *vpp = NULL;
1633                 return error;
1634         }
1635
1636         if (ip == NULL) {
1637                 *vpp = NULL;
1638                 return XFS_ERROR(EIO);
1639         }
1640
1641         if (ip->i_d.di_mode == 0 || ip->i_d.di_gen != igen) {
1642                 xfs_iput_new(ip, XFS_ILOCK_SHARED);
1643                 *vpp = NULL;
1644                 return XFS_ERROR(ENOENT);
1645         }
1646
1647         *vpp = XFS_ITOV(ip);
1648         xfs_iunlock(ip, XFS_ILOCK_SHARED);
1649         return 0;
1650 }
1651
1652
1653 #define MNTOPT_LOGBUFS  "logbufs"       /* number of XFS log buffers */
1654 #define MNTOPT_LOGBSIZE "logbsize"      /* size of XFS log buffers */
1655 #define MNTOPT_LOGDEV   "logdev"        /* log device */
1656 #define MNTOPT_RTDEV    "rtdev"         /* realtime I/O device */
1657 #define MNTOPT_BIOSIZE  "biosize"       /* log2 of preferred buffered io size */
1658 #define MNTOPT_WSYNC    "wsync"         /* safe-mode nfs compatible mount */
1659 #define MNTOPT_INO64    "ino64"         /* force inodes into 64-bit range */
1660 #define MNTOPT_NOALIGN  "noalign"       /* turn off stripe alignment */
1661 #define MNTOPT_SWALLOC  "swalloc"       /* turn on stripe width allocation */
1662 #define MNTOPT_SUNIT    "sunit"         /* data volume stripe unit */
1663 #define MNTOPT_SWIDTH   "swidth"        /* data volume stripe width */
1664 #define MNTOPT_NOUUID   "nouuid"        /* ignore filesystem UUID */
1665 #define MNTOPT_MTPT     "mtpt"          /* filesystem mount point */
1666 #define MNTOPT_GRPID    "grpid"         /* group-ID from parent directory */
1667 #define MNTOPT_NOGRPID  "nogrpid"       /* group-ID from current process */
1668 #define MNTOPT_BSDGROUPS    "bsdgroups"    /* group-ID from parent directory */
1669 #define MNTOPT_SYSVGROUPS   "sysvgroups"   /* group-ID from current process */
1670 #define MNTOPT_ALLOCSIZE    "allocsize"    /* preferred allocation size */
1671 #define MNTOPT_IHASHSIZE    "ihashsize"    /* size of inode hash table */
1672 #define MNTOPT_NORECOVERY   "norecovery"   /* don't run XFS recovery */
1673 #define MNTOPT_BARRIER  "barrier"       /* use writer barriers for log write and
1674                                          * unwritten extent conversion */
1675 #define MNTOPT_NOBARRIER "nobarrier"    /* .. disable */
1676 #define MNTOPT_OSYNCISOSYNC "osyncisosync" /* o_sync is REALLY o_sync */
1677 #define MNTOPT_64BITINODE   "inode64"   /* inodes can be allocated anywhere */
1678 #define MNTOPT_IKEEP    "ikeep"         /* do not free empty inode clusters */
1679 #define MNTOPT_NOIKEEP  "noikeep"       /* free empty inode clusters */
1680 #define MNTOPT_LARGEIO     "largeio"    /* report large I/O sizes in stat() */
1681 #define MNTOPT_NOLARGEIO   "nolargeio"  /* do not report large I/O sizes
1682                                          * in stat(). */
1683 #define MNTOPT_ATTR2    "attr2"         /* do use attr2 attribute format */
1684 #define MNTOPT_NOATTR2  "noattr2"       /* do not use attr2 attribute format */
1685 #define MNTOPT_FILESTREAM  "filestreams" /* use filestreams allocator */
1686
1687 STATIC unsigned long
1688 suffix_strtoul(char *s, char **endp, unsigned int base)
1689 {
1690         int     last, shift_left_factor = 0;
1691         char    *value = s;
1692
1693         last = strlen(value) - 1;
1694         if (value[last] == 'K' || value[last] == 'k') {
1695                 shift_left_factor = 10;
1696                 value[last] = '\0';
1697         }
1698         if (value[last] == 'M' || value[last] == 'm') {
1699                 shift_left_factor = 20;
1700                 value[last] = '\0';
1701         }
1702         if (value[last] == 'G' || value[last] == 'g') {
1703                 shift_left_factor = 30;
1704                 value[last] = '\0';
1705         }
1706
1707         return simple_strtoul((const char *)s, endp, base) << shift_left_factor;
1708 }
1709
1710 STATIC int
1711 xfs_parseargs(
1712         struct bhv_desc         *bhv,
1713         char                    *options,
1714         struct xfs_mount_args   *args,
1715         int                     update)
1716 {
1717         bhv_vfs_t               *vfsp = bhvtovfs(bhv);
1718         char                    *this_char, *value, *eov;
1719         int                     dsunit, dswidth, vol_dsunit, vol_dswidth;
1720         int                     iosize;
1721
1722         args->flags |= XFSMNT_IDELETE;
1723         args->flags |= XFSMNT_BARRIER;
1724         args->flags2 |= XFSMNT2_COMPAT_IOSIZE;
1725
1726         if (!options)
1727                 goto done;
1728
1729         iosize = dsunit = dswidth = vol_dsunit = vol_dswidth = 0;
1730
1731         while ((this_char = strsep(&options, ",")) != NULL) {
1732                 if (!*this_char)
1733                         continue;
1734                 if ((value = strchr(this_char, '=')) != NULL)
1735                         *value++ = 0;
1736
1737                 if (!strcmp(this_char, MNTOPT_LOGBUFS)) {
1738                         if (!value || !*value) {
1739                                 cmn_err(CE_WARN,
1740                                         "XFS: %s option requires an argument",
1741                                         this_char);
1742                                 return EINVAL;
1743                         }
1744                         args->logbufs = simple_strtoul(value, &eov, 10);
1745                 } else if (!strcmp(this_char, MNTOPT_LOGBSIZE)) {
1746                         if (!value || !*value) {
1747                                 cmn_err(CE_WARN,
1748                                         "XFS: %s option requires an argument",
1749                                         this_char);
1750                                 return EINVAL;
1751                         }
1752                         args->logbufsize = suffix_strtoul(value, &eov, 10);
1753                 } else if (!strcmp(this_char, MNTOPT_LOGDEV)) {
1754                         if (!value || !*value) {
1755                                 cmn_err(CE_WARN,
1756                                         "XFS: %s option requires an argument",
1757                                         this_char);
1758                                 return EINVAL;
1759                         }
1760                         strncpy(args->logname, value, MAXNAMELEN);
1761                 } else if (!strcmp(this_char, MNTOPT_MTPT)) {
1762                         if (!value || !*value) {
1763                                 cmn_err(CE_WARN,
1764                                         "XFS: %s option requires an argument",
1765                                         this_char);
1766                                 return EINVAL;
1767                         }
1768                         strncpy(args->mtpt, value, MAXNAMELEN);
1769                 } else if (!strcmp(this_char, MNTOPT_RTDEV)) {
1770                         if (!value || !*value) {
1771                                 cmn_err(CE_WARN,
1772                                         "XFS: %s option requires an argument",
1773                                         this_char);
1774                                 return EINVAL;
1775                         }
1776                         strncpy(args->rtname, value, MAXNAMELEN);
1777                 } else if (!strcmp(this_char, MNTOPT_BIOSIZE)) {
1778                         if (!value || !*value) {
1779                                 cmn_err(CE_WARN,
1780                                         "XFS: %s option requires an argument",
1781                                         this_char);
1782                                 return EINVAL;
1783                         }
1784                         iosize = simple_strtoul(value, &eov, 10);
1785                         args->flags |= XFSMNT_IOSIZE;
1786                         args->iosizelog = (uint8_t) iosize;
1787                 } else if (!strcmp(this_char, MNTOPT_ALLOCSIZE)) {
1788                         if (!value || !*value) {
1789                                 cmn_err(CE_WARN,
1790                                         "XFS: %s option requires an argument",
1791                                         this_char);
1792                                 return EINVAL;
1793                         }
1794                         iosize = suffix_strtoul(value, &eov, 10);
1795                         args->flags |= XFSMNT_IOSIZE;
1796                         args->iosizelog = ffs(iosize) - 1;
1797                 } else if (!strcmp(this_char, MNTOPT_IHASHSIZE)) {
1798                         if (!value || !*value) {
1799                                 cmn_err(CE_WARN,
1800                                         "XFS: %s option requires an argument",
1801                                         this_char);
1802                                 return EINVAL;
1803                         }
1804                         args->flags |= XFSMNT_IHASHSIZE;
1805                         args->ihashsize = simple_strtoul(value, &eov, 10);
1806                 } else if (!strcmp(this_char, MNTOPT_GRPID) ||
1807                            !strcmp(this_char, MNTOPT_BSDGROUPS)) {
1808                         vfsp->vfs_flag |= VFS_GRPID;
1809                 } else if (!strcmp(this_char, MNTOPT_NOGRPID) ||
1810                            !strcmp(this_char, MNTOPT_SYSVGROUPS)) {
1811                         vfsp->vfs_flag &= ~VFS_GRPID;
1812                 } else if (!strcmp(this_char, MNTOPT_WSYNC)) {
1813                         args->flags |= XFSMNT_WSYNC;
1814                 } else if (!strcmp(this_char, MNTOPT_OSYNCISOSYNC)) {
1815                         args->flags |= XFSMNT_OSYNCISOSYNC;
1816                 } else if (!strcmp(this_char, MNTOPT_NORECOVERY)) {
1817                         args->flags |= XFSMNT_NORECOVERY;
1818                 } else if (!strcmp(this_char, MNTOPT_INO64)) {
1819                         args->flags |= XFSMNT_INO64;
1820 #if !XFS_BIG_INUMS
1821                         cmn_err(CE_WARN,
1822                                 "XFS: %s option not allowed on this system",
1823                                 this_char);
1824                         return EINVAL;
1825 #endif
1826                 } else if (!strcmp(this_char, MNTOPT_NOALIGN)) {
1827                         args->flags |= XFSMNT_NOALIGN;
1828                 } else if (!strcmp(this_char, MNTOPT_SWALLOC)) {
1829                         args->flags |= XFSMNT_SWALLOC;
1830                 } else if (!strcmp(this_char, MNTOPT_SUNIT)) {
1831                         if (!value || !*value) {
1832                                 cmn_err(CE_WARN,
1833                                         "XFS: %s option requires an argument",
1834                                         this_char);
1835                                 return EINVAL;
1836                         }
1837                         dsunit = simple_strtoul(value, &eov, 10);
1838                 } else if (!strcmp(this_char, MNTOPT_SWIDTH)) {
1839                         if (!value || !*value) {
1840                                 cmn_err(CE_WARN,
1841                                         "XFS: %s option requires an argument",
1842                                         this_char);
1843                                 return EINVAL;
1844                         }
1845                         dswidth = simple_strtoul(value, &eov, 10);
1846                 } else if (!strcmp(this_char, MNTOPT_64BITINODE)) {
1847                         args->flags &= ~XFSMNT_32BITINODES;
1848 #if !XFS_BIG_INUMS
1849                         cmn_err(CE_WARN,
1850                                 "XFS: %s option not allowed on this system",
1851                                 this_char);
1852                         return EINVAL;
1853 #endif
1854                 } else if (!strcmp(this_char, MNTOPT_NOUUID)) {
1855                         args->flags |= XFSMNT_NOUUID;
1856                 } else if (!strcmp(this_char, MNTOPT_BARRIER)) {
1857                         args->flags |= XFSMNT_BARRIER;
1858                 } else if (!strcmp(this_char, MNTOPT_NOBARRIER)) {
1859                         args->flags &= ~XFSMNT_BARRIER;
1860                 } else if (!strcmp(this_char, MNTOPT_IKEEP)) {
1861                         args->flags &= ~XFSMNT_IDELETE;
1862                 } else if (!strcmp(this_char, MNTOPT_NOIKEEP)) {
1863                         args->flags |= XFSMNT_IDELETE;
1864                 } else if (!strcmp(this_char, MNTOPT_LARGEIO)) {
1865                         args->flags2 &= ~XFSMNT2_COMPAT_IOSIZE;
1866                 } else if (!strcmp(this_char, MNTOPT_NOLARGEIO)) {
1867                         args->flags2 |= XFSMNT2_COMPAT_IOSIZE;
1868                 } else if (!strcmp(this_char, MNTOPT_ATTR2)) {
1869                         args->flags |= XFSMNT_ATTR2;
1870                 } else if (!strcmp(this_char, MNTOPT_NOATTR2)) {
1871                         args->flags &= ~XFSMNT_ATTR2;
1872                 } else if (!strcmp(this_char, MNTOPT_FILESTREAM)) {
1873                         args->flags2 |= XFSMNT2_FILESTREAMS;
1874                 } else if (!strcmp(this_char, "osyncisdsync")) {
1875                         /* no-op, this is now the default */
1876                         cmn_err(CE_WARN,
1877         "XFS: osyncisdsync is now the default, option is deprecated.");
1878                 } else if (!strcmp(this_char, "irixsgid")) {
1879                         cmn_err(CE_WARN,
1880         "XFS: irixsgid is now a sysctl(2) variable, option is deprecated.");
1881                 } else {
1882                         cmn_err(CE_WARN,
1883                                 "XFS: unknown mount option [%s].", this_char);
1884                         return EINVAL;
1885                 }
1886         }
1887
1888         if (args->flags & XFSMNT_NORECOVERY) {
1889                 if ((vfsp->vfs_flag & VFS_RDONLY) == 0) {
1890                         cmn_err(CE_WARN,
1891                                 "XFS: no-recovery mounts must be read-only.");
1892                         return EINVAL;
1893                 }
1894         }
1895
1896         if ((args->flags & XFSMNT_NOALIGN) && (dsunit || dswidth)) {
1897                 cmn_err(CE_WARN,
1898         "XFS: sunit and swidth options incompatible with the noalign option");
1899                 return EINVAL;
1900         }
1901
1902         if ((dsunit && !dswidth) || (!dsunit && dswidth)) {
1903                 cmn_err(CE_WARN,
1904                         "XFS: sunit and swidth must be specified together");
1905                 return EINVAL;
1906         }
1907
1908         if (dsunit && (dswidth % dsunit != 0)) {
1909                 cmn_err(CE_WARN,
1910         "XFS: stripe width (%d) must be a multiple of the stripe unit (%d)",
1911                         dswidth, dsunit);
1912                 return EINVAL;
1913         }
1914
1915         if ((args->flags & XFSMNT_NOALIGN) != XFSMNT_NOALIGN) {
1916                 if (dsunit) {
1917                         args->sunit = dsunit;
1918                         args->flags |= XFSMNT_RETERR;
1919                 } else {
1920                         args->sunit = vol_dsunit;
1921                 }
1922                 dswidth ? (args->swidth = dswidth) :
1923                           (args->swidth = vol_dswidth);
1924         } else {
1925                 args->sunit = args->swidth = 0;
1926         }
1927
1928 done:
1929         if (args->flags & XFSMNT_32BITINODES)
1930                 vfsp->vfs_flag |= VFS_32BITINODES;
1931         if (args->flags2)
1932                 args->flags |= XFSMNT_FLAGS2;
1933         return 0;
1934 }
1935
1936 STATIC int
1937 xfs_showargs(
1938         struct bhv_desc         *bhv,
1939         struct seq_file         *m)
1940 {
1941         static struct proc_xfs_info {
1942                 int     flag;
1943                 char    *str;
1944         } xfs_info[] = {
1945                 /* the few simple ones we can get from the mount struct */
1946                 { XFS_MOUNT_WSYNC,              "," MNTOPT_WSYNC },
1947                 { XFS_MOUNT_INO64,              "," MNTOPT_INO64 },
1948                 { XFS_MOUNT_NOALIGN,            "," MNTOPT_NOALIGN },
1949                 { XFS_MOUNT_SWALLOC,            "," MNTOPT_SWALLOC },
1950                 { XFS_MOUNT_NOUUID,             "," MNTOPT_NOUUID },
1951                 { XFS_MOUNT_NORECOVERY,         "," MNTOPT_NORECOVERY },
1952                 { XFS_MOUNT_OSYNCISOSYNC,       "," MNTOPT_OSYNCISOSYNC },
1953                 { 0, NULL }
1954         };
1955         struct proc_xfs_info    *xfs_infop;
1956         struct xfs_mount        *mp = XFS_BHVTOM(bhv);
1957         struct bhv_vfs          *vfsp = XFS_MTOVFS(mp);
1958
1959         for (xfs_infop = xfs_info; xfs_infop->flag; xfs_infop++) {
1960                 if (mp->m_flags & xfs_infop->flag)
1961                         seq_puts(m, xfs_infop->str);
1962         }
1963
1964         if (mp->m_flags & XFS_MOUNT_IHASHSIZE)
1965                 seq_printf(m, "," MNTOPT_IHASHSIZE "=%d", (int)mp->m_ihsize);
1966
1967         if (mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)
1968                 seq_printf(m, "," MNTOPT_ALLOCSIZE "=%dk",
1969                                 (int)(1 << mp->m_writeio_log) >> 10);
1970
1971         if (mp->m_logbufs > 0)
1972                 seq_printf(m, "," MNTOPT_LOGBUFS "=%d", mp->m_logbufs);
1973         if (mp->m_logbsize > 0)
1974                 seq_printf(m, "," MNTOPT_LOGBSIZE "=%dk", mp->m_logbsize >> 10);
1975
1976         if (mp->m_logname)
1977                 seq_printf(m, "," MNTOPT_LOGDEV "=%s", mp->m_logname);
1978         if (mp->m_rtname)
1979                 seq_printf(m, "," MNTOPT_RTDEV "=%s", mp->m_rtname);
1980
1981         if (mp->m_dalign > 0)
1982                 seq_printf(m, "," MNTOPT_SUNIT "=%d",
1983                                 (int)XFS_FSB_TO_BB(mp, mp->m_dalign));
1984         if (mp->m_swidth > 0)
1985                 seq_printf(m, "," MNTOPT_SWIDTH "=%d",
1986                                 (int)XFS_FSB_TO_BB(mp, mp->m_swidth));
1987
1988         if (!(mp->m_flags & XFS_MOUNT_IDELETE))
1989                 seq_printf(m, "," MNTOPT_IKEEP);
1990         if (!(mp->m_flags & XFS_MOUNT_COMPAT_IOSIZE))
1991                 seq_printf(m, "," MNTOPT_LARGEIO);
1992
1993         if (!(vfsp->vfs_flag & VFS_32BITINODES))
1994                 seq_printf(m, "," MNTOPT_64BITINODE);
1995         if (vfsp->vfs_flag & VFS_GRPID)
1996                 seq_printf(m, "," MNTOPT_GRPID);
1997
1998         return 0;
1999 }
2000
2001 /*
2002  * Second stage of a freeze. The data is already frozen so we only
2003  * need to take care of themetadata. Once that's done write a dummy
2004  * record to dirty the log in case of a crash while frozen.
2005  */
2006 STATIC void
2007 xfs_freeze(
2008         bhv_desc_t      *bdp)
2009 {
2010         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
2011
2012         xfs_attr_quiesce(mp);
2013         xfs_fs_log_dummy(mp);
2014 }
2015
2016
2017 bhv_vfsops_t xfs_vfsops = {
2018         BHV_IDENTITY_INIT(VFS_BHV_XFS,VFS_POSITION_XFS),
2019         .vfs_parseargs          = xfs_parseargs,
2020         .vfs_showargs           = xfs_showargs,
2021         .vfs_mount              = xfs_mount,
2022         .vfs_unmount            = xfs_unmount,
2023         .vfs_mntupdate          = xfs_mntupdate,
2024         .vfs_root               = xfs_root,
2025         .vfs_statvfs            = xfs_statvfs,
2026         .vfs_sync               = xfs_sync,
2027         .vfs_vget               = xfs_vget,
2028         .vfs_dmapiops           = (vfs_dmapiops_t)fs_nosys,
2029         .vfs_quotactl           = (vfs_quotactl_t)fs_nosys,
2030         .vfs_init_vnode         = xfs_initialize_vnode,
2031         .vfs_force_shutdown     = xfs_do_force_shutdown,
2032         .vfs_freeze             = xfs_freeze,
2033 };