fs/xfs/xfs_vfsops.c

   1 /*
   2  * XFS filesystem operations.
   3  *
   4  * Copyright (c) 2000-2004 Silicon Graphics, Inc.  All Rights Reserved.
   5  *
   6  * This program is free software; you can redistribute it and/or modify it
   7  * under the terms of version 2 of the GNU General Public License as
   8  * published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it would be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
  13  *
  14  * Further, this software is distributed without any warranty that it is
  15  * free of the rightful claim of any third person regarding infringement
  16  * or the like.  Any license provided herein, whether implied or
  17  * otherwise, applies only to this software file.  Patent licenses, if
  18  * any, provided herein do not apply to combinations of this program with
  19  * other software, or any other product whatsoever.
  20  *
  21  * You should have received a copy of the GNU General Public License along
  22  * with this program; if not, write the Free Software Foundation, Inc., 59
  23  * Temple Place - Suite 330, Boston MA 02111-1307, USA.
  24  *
  25  * Contact information: Silicon Graphics, Inc., 1600 Amphitheatre Pkwy,
  26  * Mountain View, CA  94043, or:
  27  *
  28  * http://www.sgi.com
  29  *
  30  * For further information regarding this notice, see:
  31  *
  32  * http://oss.sgi.com/projects/GenInfo/SGIGPLNoticeExplan/
  33  */
  34
  35 #include "xfs.h"
  36 #include "xfs_macros.h"
  37 #include "xfs_types.h"
  38 #include "xfs_inum.h"
  39 #include "xfs_log.h"
  40 #include "xfs_trans.h"
  41 #include "xfs_sb.h"
  42 #include "xfs_dir.h"
  43 #include "xfs_dir2.h"
  44 #include "xfs_dmapi.h"
  45 #include "xfs_mount.h"
  46 #include "xfs_bmap_btree.h"
  47 #include "xfs_ialloc_btree.h"
  48 #include "xfs_alloc_btree.h"
  49 #include "xfs_btree.h"
  50 #include "xfs_alloc.h"
  51 #include "xfs_ialloc.h"
  52 #include "xfs_attr_sf.h"
  53 #include "xfs_dir_sf.h"
  54 #include "xfs_dir2_sf.h"
  55 #include "xfs_dinode.h"
  56 #include "xfs_inode_item.h"
  57 #include "xfs_inode.h"
  58 #include "xfs_ag.h"
  59 #include "xfs_error.h"
  60 #include "xfs_bmap.h"
  61 #include "xfs_da_btree.h"
  62 #include "xfs_rw.h"
  63 #include "xfs_refcache.h"
  64 #include "xfs_buf_item.h"
  65 #include "xfs_extfree_item.h"
  66 #include "xfs_quota.h"
  67 #include "xfs_dir2_trace.h"
  68 #include "xfs_acl.h"
  69 #include "xfs_attr.h"
  70 #include "xfs_clnt.h"
  71 #include "xfs_log_priv.h"
  72
  73 STATIC int xfs_sync(bhv_desc_t *, int, cred_t *);
  74
  75 int
  76 xfs_init(void)
  77 {
  78         extern kmem_zone_t      *xfs_bmap_free_item_zone;
  79         extern kmem_zone_t      *xfs_btree_cur_zone;
  80         extern kmem_zone_t      *xfs_trans_zone;
  81         extern kmem_zone_t      *xfs_buf_item_zone;
  82         extern kmem_zone_t      *xfs_dabuf_zone;
  83 #ifdef XFS_DABUF_DEBUG
  84         extern lock_t           xfs_dabuf_global_lock;
  85         spinlock_init(&xfs_dabuf_global_lock, "xfsda");
  86 #endif
  87
  88         /*
  89          * Initialize all of the zone allocators we use.
  90          */
  91         xfs_bmap_free_item_zone = kmem_zone_init(sizeof(xfs_bmap_free_item_t),
  92                                                  "xfs_bmap_free_item");
  93         xfs_btree_cur_zone = kmem_zone_init(sizeof(xfs_btree_cur_t),
  94                                             "xfs_btree_cur");
  95         xfs_inode_zone = kmem_zone_init(sizeof(xfs_inode_t), "xfs_inode");
  96         xfs_trans_zone = kmem_zone_init(sizeof(xfs_trans_t), "xfs_trans");
  97         xfs_da_state_zone =
  98                 kmem_zone_init(sizeof(xfs_da_state_t), "xfs_da_state");
  99         xfs_dabuf_zone = kmem_zone_init(sizeof(xfs_dabuf_t), "xfs_dabuf");
 100
 101         /*
 102          * The size of the zone allocated buf log item is the maximum
 103          * size possible under XFS.  This wastes a little bit of memory,
 104          * but it is much faster.
 105          */
 106         xfs_buf_item_zone =
 107                 kmem_zone_init((sizeof(xfs_buf_log_item_t) +
 108                                 (((XFS_MAX_BLOCKSIZE / XFS_BLI_CHUNK) /
 109                                   NBWORD) * sizeof(int))),
 110                                "xfs_buf_item");
 111         xfs_efd_zone = kmem_zone_init((sizeof(xfs_efd_log_item_t) +
 112                                        ((XFS_EFD_MAX_FAST_EXTENTS - 1) * sizeof(xfs_extent_t))),
 113                                       "xfs_efd_item");
 114         xfs_efi_zone = kmem_zone_init((sizeof(xfs_efi_log_item_t) +
 115                                        ((XFS_EFI_MAX_FAST_EXTENTS - 1) * sizeof(xfs_extent_t))),
 116                                       "xfs_efi_item");
 117         xfs_ifork_zone = kmem_zone_init(sizeof(xfs_ifork_t), "xfs_ifork");
 118         xfs_ili_zone = kmem_zone_init(sizeof(xfs_inode_log_item_t), "xfs_ili");
 119         xfs_chashlist_zone = kmem_zone_init(sizeof(xfs_chashlist_t),
 120                                             "xfs_chashlist");
 121         _ACL_ZONE_INIT(xfs_acl_zone, "xfs_acl");
 122
 123         /*
 124          * Allocate global trace buffers.
 125          */
 126 #ifdef XFS_ALLOC_TRACE
 127         xfs_alloc_trace_buf = ktrace_alloc(XFS_ALLOC_TRACE_SIZE, KM_SLEEP);
 128 #endif
 129 #ifdef XFS_BMAP_TRACE
 130         xfs_bmap_trace_buf = ktrace_alloc(XFS_BMAP_TRACE_SIZE, KM_SLEEP);
 131 #endif
 132 #ifdef XFS_BMBT_TRACE
 133         xfs_bmbt_trace_buf = ktrace_alloc(XFS_BMBT_TRACE_SIZE, KM_SLEEP);
 134 #endif
 135 #ifdef XFS_DIR_TRACE
 136         xfs_dir_trace_buf = ktrace_alloc(XFS_DIR_TRACE_SIZE, KM_SLEEP);
 137 #endif
 138 #ifdef XFS_ATTR_TRACE
 139         xfs_attr_trace_buf = ktrace_alloc(XFS_ATTR_TRACE_SIZE, KM_SLEEP);
 140 #endif
 141 #ifdef XFS_DIR2_TRACE
 142         xfs_dir2_trace_buf = ktrace_alloc(XFS_DIR2_GTRACE_SIZE, KM_SLEEP);
 143 #endif
 144
 145         xfs_dir_startup();
 146
 147 #if (defined(DEBUG) || defined(INDUCE_IO_ERROR))
 148         xfs_error_test_init();
 149 #endif /* DEBUG || INDUCE_IO_ERROR */
 150
 151         xfs_init_procfs();
 152         xfs_sysctl_register();
 153         return 0;
 154 }
 155
 156 void
 157 xfs_cleanup(void)
 158 {
 159         extern kmem_zone_t      *xfs_bmap_free_item_zone;
 160         extern kmem_zone_t      *xfs_btree_cur_zone;
 161         extern kmem_zone_t      *xfs_inode_zone;
 162         extern kmem_zone_t      *xfs_trans_zone;
 163         extern kmem_zone_t      *xfs_da_state_zone;
 164         extern kmem_zone_t      *xfs_dabuf_zone;
 165         extern kmem_zone_t      *xfs_efd_zone;
 166         extern kmem_zone_t      *xfs_efi_zone;
 167         extern kmem_zone_t      *xfs_buf_item_zone;
 168         extern kmem_zone_t      *xfs_chashlist_zone;
 169
 170         xfs_cleanup_procfs();
 171         xfs_sysctl_unregister();
 172         xfs_refcache_destroy();
 173
 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_cache_destroy(xfs_bmap_free_item_zone);
 194         kmem_cache_destroy(xfs_btree_cur_zone);
 195         kmem_cache_destroy(xfs_inode_zone);
 196         kmem_cache_destroy(xfs_trans_zone);
 197         kmem_cache_destroy(xfs_da_state_zone);
 198         kmem_cache_destroy(xfs_dabuf_zone);
 199         kmem_cache_destroy(xfs_buf_item_zone);
 200         kmem_cache_destroy(xfs_efd_zone);
 201         kmem_cache_destroy(xfs_efi_zone);
 202         kmem_cache_destroy(xfs_ifork_zone);
 203         kmem_cache_destroy(xfs_ili_zone);
 204         kmem_cache_destroy(xfs_chashlist_zone);
 205         _ACL_ZONE_DESTROY(xfs_acl_zone);
 206 }
 207
 208 /*
 209  * xfs_start_flags
 210  *
 211  * This function fills in xfs_mount_t fields based on mount args.
 212  * Note: the superblock has _not_ yet been read in.
 213  */
 214 STATIC int
 215 xfs_start_flags(
 216         struct vfs              *vfs,
 217         struct xfs_mount_args   *ap,
 218         struct xfs_mount        *mp)
 219 {
 220         /* Values are in BBs */
 221         if ((ap->flags & XFSMNT_NOALIGN) != XFSMNT_NOALIGN) {
 222                 /*
 223                  * At this point the superblock has not been read
 224                  * in, therefore we do not know the block size.
 225                  * Before the mount call ends we will convert
 226                  * these to FSBs.
 227                  */
 228                 mp->m_dalign = ap->sunit;
 229                 mp->m_swidth = ap->swidth;
 230         }
 231
 232         if (ap->logbufs != -1 &&
 233 #if defined(DEBUG) || defined(XLOG_NOLOG)
 234             ap->logbufs != 0 &&
 235 #endif
 236             (ap->logbufs < XLOG_MIN_ICLOGS ||
 237              ap->logbufs > XLOG_MAX_ICLOGS)) {
 238                 cmn_err(CE_WARN,
 239                         "XFS: invalid logbufs value: %d [not %d-%d]",
 240                         ap->logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
 241                 return XFS_ERROR(EINVAL);
 242         }
 243         mp->m_logbufs = ap->logbufs;
 244         if (ap->logbufsize != -1 &&
 245             ap->logbufsize != 16 * 1024 &&
 246             ap->logbufsize != 32 * 1024 &&
 247             ap->logbufsize != 64 * 1024 &&
 248             ap->logbufsize != 128 * 1024 &&
 249             ap->logbufsize != 256 * 1024) {
 250                 cmn_err(CE_WARN,
 251         "XFS: invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
 252                         ap->logbufsize);
 253                 return XFS_ERROR(EINVAL);
 254         }
 255         mp->m_logbsize = ap->logbufsize;
 256         mp->m_fsname_len = strlen(ap->fsname) + 1;
 257         mp->m_fsname = kmem_alloc(mp->m_fsname_len, KM_SLEEP);
 258         strcpy(mp->m_fsname, ap->fsname);
 259
 260         /*
 261          * Pull in the 'wsync' and 'ino64' mount options before we do the real
 262          * work of mounting and recovery.  The arg pointer will
 263          * be NULL when we are being called from the root mount code.
 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_NOATIME)
 274                 mp->m_flags |= XFS_MOUNT_NOATIME;
 275
 276         if (ap->flags & XFSMNT_RETERR)
 277                 mp->m_flags |= XFS_MOUNT_RETERR;
 278
 279         if (ap->flags & XFSMNT_NOALIGN)
 280                 mp->m_flags |= XFS_MOUNT_NOALIGN;
 281
 282         if (ap->flags & XFSMNT_OSYNCISOSYNC)
 283                 mp->m_flags |= XFS_MOUNT_OSYNCISOSYNC;
 284
 285         if (ap->flags & XFSMNT_32BITINODES)
 286                 mp->m_flags |= (XFS_MOUNT_32BITINODES | XFS_MOUNT_32BITINOOPT);
 287
 288         if (ap->flags & XFSMNT_IOSIZE) {
 289                 if (ap->iosizelog > XFS_MAX_IO_LOG ||
 290                     ap->iosizelog < XFS_MIN_IO_LOG) {
 291                         cmn_err(CE_WARN,
 292                 "XFS: invalid log iosize: %d [not %d-%d]",
 293                                 ap->iosizelog, XFS_MIN_IO_LOG,
 294                                 XFS_MAX_IO_LOG);
 295                         return XFS_ERROR(EINVAL);
 296                 }
 297
 298                 mp->m_flags |= XFS_MOUNT_DFLT_IOSIZE;
 299                 mp->m_readio_log = mp->m_writeio_log = ap->iosizelog;
 300         }
 301         if (ap->flags & XFSMNT_IDELETE)
 302                 mp->m_flags |= XFS_MOUNT_IDELETE;
 303
 304         /*
 305          * no recovery flag requires a read-only mount
 306          */
 307         if (ap->flags & XFSMNT_NORECOVERY) {
 308                 if (!(vfs->vfs_flag & VFS_RDONLY)) {
 309                         cmn_err(CE_WARN,
 310         "XFS: tried to mount a FS read-write without recovery!");
 311                         return XFS_ERROR(EINVAL);
 312                 }
 313                 mp->m_flags |= XFS_MOUNT_NORECOVERY;
 314         }
 315
 316         if (ap->flags & XFSMNT_NOUUID)
 317                 mp->m_flags |= XFS_MOUNT_NOUUID;
 318         if (ap->flags & XFSMNT_NOLOGFLUSH)
 319                 mp->m_flags |= XFS_MOUNT_NOLOGFLUSH;
 320
 321         return 0;
 322 }
 323
 324 /*
 325  * This function fills in xfs_mount_t fields based on mount args.
 326  * Note: the superblock _has_ now been read in.
 327  */
 328 STATIC int
 329 xfs_finish_flags(
 330         struct vfs              *vfs,
 331         struct xfs_mount_args   *ap,
 332         struct xfs_mount        *mp)
 333 {
 334         int                     ronly = (vfs->vfs_flag & VFS_RDONLY);
 335
 336         /* Fail a mount where the logbuf is smaller then the log stripe */
 337         if (XFS_SB_VERSION_HASLOGV2(&mp->m_sb)) {
 338                 if ((ap->logbufsize == -1) &&
 339                     (mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE)) {
 340                         mp->m_logbsize = mp->m_sb.sb_logsunit;
 341                 } else if (ap->logbufsize < mp->m_sb.sb_logsunit) {
 342                         cmn_err(CE_WARN,
 343         "XFS: logbuf size must be greater than or equal to log stripe size");
 344                         return XFS_ERROR(EINVAL);
 345                 }
 346         } else {
 347                 /* Fail a mount if the logbuf is larger than 32K */
 348                 if (ap->logbufsize > XLOG_BIG_RECORD_BSIZE) {
 349                         cmn_err(CE_WARN,
 350         "XFS: logbuf size for version 1 logs must be 16K or 32K");
 351                         return XFS_ERROR(EINVAL);
 352                 }
 353         }
 354
 355         /*
 356          * prohibit r/w mounts of read-only filesystems
 357          */
 358         if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !ronly) {
 359                 cmn_err(CE_WARN,
 360         "XFS: cannot mount a read-only filesystem as read-write");
 361                 return XFS_ERROR(EROFS);
 362         }
 363
 364         /*
 365          * disallow mount attempts with (IRIX) project quota enabled
 366          */
 367         if (XFS_SB_VERSION_HASQUOTA(&mp->m_sb) &&
 368             (mp->m_sb.sb_qflags & XFS_PQUOTA_ACCT)) {
 369                 cmn_err(CE_WARN,
 370         "XFS: cannot mount a filesystem with IRIX project quota enabled");
 371                 return XFS_ERROR(ENOSYS);
 372         }
 373
 374         /*
 375          * check for shared mount.
 376          */
 377         if (ap->flags & XFSMNT_SHARED) {
 378                 if (!XFS_SB_VERSION_HASSHARED(&mp->m_sb))
 379                         return XFS_ERROR(EINVAL);
 380
 381                 /*
 382                  * For IRIX 6.5, shared mounts must have the shared
 383                  * version bit set, have the persistent readonly
 384                  * field set, must be version 0 and can only be mounted
 385                  * read-only.
 386                  */
 387                 if (!ronly || !(mp->m_sb.sb_flags & XFS_SBF_READONLY) ||
 388                      (mp->m_sb.sb_shared_vn != 0))
 389                         return XFS_ERROR(EINVAL);
 390
 391                 mp->m_flags |= XFS_MOUNT_SHARED;
 392
 393                 /*
 394                  * Shared XFS V0 can't deal with DMI.  Return EINVAL.
 395                  */
 396                 if (mp->m_sb.sb_shared_vn == 0 && (ap->flags & XFSMNT_DMAPI))
 397                         return XFS_ERROR(EINVAL);
 398         }
 399
 400         return 0;
 401 }
 402
 403 /*
 404  * xfs_mount
 405  *
 406  * The file system configurations are:
 407  *      (1) device (partition) with data and internal log
 408  *      (2) logical volume with data and log subvolumes.
 409  *      (3) logical volume with data, log, and realtime subvolumes.
 410  *
 411  * We only have to handle opening the log and realtime volumes here if
 412  * they are present.  The data subvolume has already been opened by
 413  * get_sb_bdev() and is stored in vfsp->vfs_super->s_bdev.
 414  */
 415 STATIC int
 416 xfs_mount(
 417         struct bhv_desc         *bhvp,
 418         struct xfs_mount_args   *args,
 419         cred_t                  *credp)
 420 {
 421         struct vfs              *vfsp = bhvtovfs(bhvp);
 422         struct bhv_desc         *p;
 423         struct xfs_mount        *mp = XFS_BHVTOM(bhvp);
 424         struct block_device     *ddev, *logdev, *rtdev;
 425         int                     flags = 0, error;
 426
 427         ddev = vfsp->vfs_super->s_bdev;
 428         logdev = rtdev = NULL;
 429
 430         /*
 431          * Open real time and log devices - order is important.
 432          */
 433         if (args->logname[0]) {
 434                 error = xfs_blkdev_get(mp, args->logname, &logdev);
 435                 if (error)
 436                         return error;
 437         }
 438         if (args->rtname[0]) {
 439                 error = xfs_blkdev_get(mp, args->rtname, &rtdev);
 440                 if (error) {
 441                         xfs_blkdev_put(logdev);
 442                         return error;
 443                 }
 444
 445                 if (rtdev == ddev || rtdev == logdev) {
 446                         cmn_err(CE_WARN,
 447         "XFS: Cannot mount filesystem with identical rtdev and ddev/logdev.");
 448                         xfs_blkdev_put(logdev);
 449                         xfs_blkdev_put(rtdev);
 450                         return EINVAL;
 451                 }
 452         }
 453
 454         /*
 455          * Setup xfs_mount function vectors from available behaviors
 456          */
 457         p = vfs_bhv_lookup(vfsp, VFS_POSITION_DM);
 458         mp->m_dm_ops = p ? *(xfs_dmops_t *) vfs_bhv_custom(p) : xfs_dmcore_stub;
 459         p = vfs_bhv_lookup(vfsp, VFS_POSITION_QM);
 460         mp->m_qm_ops = p ? *(xfs_qmops_t *) vfs_bhv_custom(p) : xfs_qmcore_stub;
 461         p = vfs_bhv_lookup(vfsp, VFS_POSITION_IO);
 462         mp->m_io_ops = p ? *(xfs_ioops_t *) vfs_bhv_custom(p) : xfs_iocore_xfs;
 463
 464         /*
 465          * Setup xfs_mount buffer target pointers
 466          */
 467         mp->m_ddev_targp = xfs_alloc_buftarg(ddev);
 468         if (rtdev)
 469                 mp->m_rtdev_targp = xfs_alloc_buftarg(rtdev);
 470         mp->m_logdev_targp = (logdev && logdev != ddev) ?
 471                                 xfs_alloc_buftarg(logdev) : mp->m_ddev_targp;
 472
 473         /*
 474          * Setup flags based on mount(2) options and then the superblock
 475          */
 476         error = xfs_start_flags(vfsp, args, mp);
 477         if (error)
 478                 goto error;
 479         error = xfs_readsb(mp);
 480         if (error)
 481                 goto error;
 482         error = xfs_finish_flags(vfsp, args, mp);
 483         if (error) {
 484                 xfs_freesb(mp);
 485                 goto error;
 486         }
 487
 488         /*
 489          * Setup xfs_mount buffer target pointers based on superblock
 490          */
 491         xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_blocksize,
 492                             mp->m_sb.sb_sectsize);
 493         if (logdev && logdev != ddev) {
 494                 unsigned int    log_sector_size = BBSIZE;
 495
 496                 if (XFS_SB_VERSION_HASSECTOR(&mp->m_sb))
 497                         log_sector_size = mp->m_sb.sb_logsectsize;
 498                 xfs_setsize_buftarg(mp->m_logdev_targp, mp->m_sb.sb_blocksize,
 499                                     log_sector_size);
 500         }
 501         if (rtdev)
 502                 xfs_setsize_buftarg(mp->m_rtdev_targp, mp->m_sb.sb_blocksize,
 503                                     mp->m_sb.sb_blocksize);
 504
 505         if (!(error = XFS_IOINIT(vfsp, args, flags)))
 506                 return 0;
 507
 508  error:
 509         xfs_binval(mp->m_ddev_targp);
 510         if (logdev != NULL && logdev != ddev) {
 511                 xfs_binval(mp->m_logdev_targp);
 512         }
 513         if (rtdev != NULL) {
 514                 xfs_binval(mp->m_rtdev_targp);
 515         }
 516         xfs_unmountfs_close(mp, NULL);
 517         return error;
 518 }
 519
 520 STATIC int
 521 xfs_unmount(
 522         bhv_desc_t      *bdp,
 523         int             flags,
 524         cred_t          *credp)
 525 {
 526         struct vfs      *vfsp = bhvtovfs(bdp);
 527         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
 528         xfs_inode_t     *rip;
 529         vnode_t         *rvp;
 530         int             unmount_event_wanted = 0;
 531         int             unmount_event_flags = 0;
 532         int             xfs_unmountfs_needed = 0;
 533         int             error;
 534
 535         rip = mp->m_rootip;
 536         rvp = XFS_ITOV(rip);
 537
 538         if (vfsp->vfs_flag & VFS_DMI) {
 539                 error = XFS_SEND_NAMESP(mp, DM_EVENT_PREUNMOUNT,
 540                                 rvp, DM_RIGHT_NULL, rvp, DM_RIGHT_NULL,
 541                                 NULL, NULL, 0, 0,
 542                                 (mp->m_dmevmask & (1<<DM_EVENT_PREUNMOUNT))?
 543                                         0:DM_FLAGS_UNWANTED);
 544                         if (error)
 545                                 return XFS_ERROR(error);
 546                 unmount_event_wanted = 1;
 547                 unmount_event_flags = (mp->m_dmevmask & (1<<DM_EVENT_UNMOUNT))?
 548                                         0 : DM_FLAGS_UNWANTED;
 549         }
 550
 551         /*
 552          * First blow any referenced inode from this file system
 553          * out of the reference cache, and delete the timer.
 554          */
 555         xfs_refcache_purge_mp(mp);
 556
 557         XFS_bflush(mp->m_ddev_targp);
 558         error = xfs_unmount_flush(mp, 0);
 559         if (error)
 560                 goto out;
 561
 562         ASSERT(vn_count(rvp) == 1);
 563
 564         /*
 565          * Drop the reference count
 566          */
 567         VN_RELE(rvp);
 568
 569         /*
 570          * If we're forcing a shutdown, typically because of a media error,
 571          * we want to make sure we invalidate dirty pages that belong to
 572          * referenced vnodes as well.
 573          */
 574         if (XFS_FORCED_SHUTDOWN(mp)) {
 575                 error = xfs_sync(&mp->m_bhv,
 576                          (SYNC_WAIT | SYNC_CLOSE), credp);
 577                 ASSERT(error != EFSCORRUPTED);
 578         }
 579         xfs_unmountfs_needed = 1;
 580
 581 out:
 582         /*      Send DMAPI event, if required.
 583          *      Then do xfs_unmountfs() if needed.
 584          *      Then return error (or zero).
 585          */
 586         if (unmount_event_wanted) {
 587                 /* Note: mp structure must still exist for
 588                  * XFS_SEND_UNMOUNT() call.
 589                  */
 590                 XFS_SEND_UNMOUNT(mp, vfsp, error == 0 ? rvp : NULL,
 591                         DM_RIGHT_NULL, 0, error, unmount_event_flags);
 592         }
 593         if (xfs_unmountfs_needed) {
 594                 /*
 595                  * Call common unmount function to flush to disk
 596                  * and free the super block buffer & mount structures.
 597                  */
 598                 xfs_unmountfs(mp, credp);
 599         }
 600
 601         return XFS_ERROR(error);
 602 }
 603
 604 #define REMOUNT_READONLY_FLAGS  (SYNC_REMOUNT|SYNC_ATTR|SYNC_WAIT)
 605
 606 STATIC int
 607 xfs_mntupdate(
 608         bhv_desc_t                      *bdp,
 609         int                             *flags,
 610         struct xfs_mount_args           *args)
 611 {
 612         struct vfs      *vfsp = bhvtovfs(bdp);
 613         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
 614         int             pincount, error;
 615         int             count = 0;
 616
 617         if (args->flags & XFSMNT_NOATIME)
 618                 mp->m_flags |= XFS_MOUNT_NOATIME;
 619         else
 620                 mp->m_flags &= ~XFS_MOUNT_NOATIME;
 621
 622         if (!(vfsp->vfs_flag & VFS_RDONLY)) {
 623                 VFS_SYNC(vfsp, SYNC_FSDATA|SYNC_BDFLUSH|SYNC_ATTR, NULL, error);
 624         }
 625
 626         if (*flags & MS_RDONLY) {
 627                 xfs_refcache_purge_mp(mp);
 628                 xfs_flush_buftarg(mp->m_ddev_targp, 0);
 629                 xfs_finish_reclaim_all(mp, 0);
 630
 631                 /* This loop must run at least twice.
 632                  * The first instance of the loop will flush
 633                  * most meta data but that will generate more
 634                  * meta data (typically directory updates).
 635                  * Which then must be flushed and logged before
 636                  * we can write the unmount record.
 637                  */
 638                 do {
 639                         VFS_SYNC(vfsp, REMOUNT_READONLY_FLAGS, NULL, error);
 640                         pincount = xfs_flush_buftarg(mp->m_ddev_targp, 1);
 641                         if (!pincount) {
 642                                 delay(50);
 643                                 count++;
 644                         }
 645                 } while (count < 2);
 646
 647                 /* Ok now write out an unmount record */
 648                 xfs_log_unmount_write(mp);
 649                 xfs_unmountfs_writesb(mp);
 650                 vfsp->vfs_flag |= VFS_RDONLY;
 651         } else {
 652                 vfsp->vfs_flag &= ~VFS_RDONLY;
 653         }
 654
 655         return 0;
 656 }
 657
 658 /*
 659  * xfs_unmount_flush implements a set of flush operation on special
 660  * inodes, which are needed as a separate set of operations so that
 661  * they can be called as part of relocation process.
 662  */
 663 int
 664 xfs_unmount_flush(
 665         xfs_mount_t     *mp,            /* Mount structure we are getting
 666                                            rid of. */
 667         int             relocation)     /* Called from vfs relocation. */
 668 {
 669         xfs_inode_t     *rip = mp->m_rootip;
 670         xfs_inode_t     *rbmip;
 671         xfs_inode_t     *rsumip = NULL;
 672         vnode_t         *rvp = XFS_ITOV(rip);
 673         int             error;
 674
 675         xfs_ilock(rip, XFS_ILOCK_EXCL);
 676         xfs_iflock(rip);
 677
 678         /*
 679          * Flush out the real time inodes.
 680          */
 681         if ((rbmip = mp->m_rbmip) != NULL) {
 682                 xfs_ilock(rbmip, XFS_ILOCK_EXCL);
 683                 xfs_iflock(rbmip);
 684                 error = xfs_iflush(rbmip, XFS_IFLUSH_SYNC);
 685                 xfs_iunlock(rbmip, XFS_ILOCK_EXCL);
 686
 687                 if (error == EFSCORRUPTED)
 688                         goto fscorrupt_out;
 689
 690                 ASSERT(vn_count(XFS_ITOV(rbmip)) == 1);
 691
 692                 rsumip = mp->m_rsumip;
 693                 xfs_ilock(rsumip, XFS_ILOCK_EXCL);
 694                 xfs_iflock(rsumip);
 695                 error = xfs_iflush(rsumip, XFS_IFLUSH_SYNC);
 696                 xfs_iunlock(rsumip, XFS_ILOCK_EXCL);
 697
 698                 if (error == EFSCORRUPTED)
 699                         goto fscorrupt_out;
 700
 701                 ASSERT(vn_count(XFS_ITOV(rsumip)) == 1);
 702         }
 703
 704         /*
 705          * Synchronously flush root inode to disk
 706          */
 707         error = xfs_iflush(rip, XFS_IFLUSH_SYNC);
 708         if (error == EFSCORRUPTED)
 709                 goto fscorrupt_out2;
 710
 711         if (vn_count(rvp) != 1 && !relocation) {
 712                 xfs_iunlock(rip, XFS_ILOCK_EXCL);
 713                 return XFS_ERROR(EBUSY);
 714         }
 715
 716         /*
 717          * Release dquot that rootinode, rbmino and rsumino might be holding,
 718          * flush and purge the quota inodes.
 719          */
 720         error = XFS_QM_UNMOUNT(mp);
 721         if (error == EFSCORRUPTED)
 722                 goto fscorrupt_out2;
 723
 724         if (rbmip) {
 725                 VN_RELE(XFS_ITOV(rbmip));
 726                 VN_RELE(XFS_ITOV(rsumip));
 727         }
 728
 729         xfs_iunlock(rip, XFS_ILOCK_EXCL);
 730         return 0;
 731
 732 fscorrupt_out:
 733         xfs_ifunlock(rip);
 734
 735 fscorrupt_out2:
 736         xfs_iunlock(rip, XFS_ILOCK_EXCL);
 737
 738         return XFS_ERROR(EFSCORRUPTED);
 739 }
 740
 741 /*
 742  * xfs_root extracts the root vnode from a vfs.
 743  *
 744  * vfsp -- the vfs struct for the desired file system
 745  * vpp  -- address of the caller's vnode pointer which should be
 746  *         set to the desired fs root vnode
 747  */
 748 STATIC int
 749 xfs_root(
 750         bhv_desc_t      *bdp,
 751         vnode_t         **vpp)
 752 {
 753         vnode_t         *vp;
 754
 755         vp = XFS_ITOV((XFS_BHVTOM(bdp))->m_rootip);
 756         VN_HOLD(vp);
 757         *vpp = vp;
 758         return 0;
 759 }
 760
 761 /*
 762  * xfs_statvfs
 763  *
 764  * Fill in the statvfs structure for the given file system.  We use
 765  * the superblock lock in the mount structure to ensure a consistent
 766  * snapshot of the counters returned.
 767  */
 768 STATIC int
 769 xfs_statvfs(
 770         bhv_desc_t      *bdp,
 771         xfs_statfs_t    *statp,
 772         vnode_t         *vp)
 773 {
 774         __uint64_t      fakeinos;
 775         xfs_extlen_t    lsize;
 776         xfs_mount_t     *mp;
 777         xfs_sb_t        *sbp;
 778         unsigned long   s;
 779         u64 id;
 780
 781         mp = XFS_BHVTOM(bdp);
 782         sbp = &(mp->m_sb);
 783
 784         statp->f_type = XFS_SB_MAGIC;
 785
 786         s = XFS_SB_LOCK(mp);
 787         statp->f_bsize = sbp->sb_blocksize;
 788         lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
 789         statp->f_blocks = sbp->sb_dblocks - lsize;
 790         statp->f_bfree = statp->f_bavail = sbp->sb_fdblocks;
 791         fakeinos = statp->f_bfree << sbp->sb_inopblog;
 792 #if XFS_BIG_INUMS
 793         fakeinos += mp->m_inoadd;
 794 #endif
 795         statp->f_files =
 796             MIN(sbp->sb_icount + fakeinos, (__uint64_t)XFS_MAXINUMBER);
 797         if (mp->m_maxicount)
 798 #if XFS_BIG_INUMS
 799                 if (!mp->m_inoadd)
 800 #endif
 801                         statp->f_files = min_t(typeof(statp->f_files),
 802                                                 statp->f_files,
 803                                                 mp->m_maxicount);
 804         statp->f_ffree = statp->f_files - (sbp->sb_icount - sbp->sb_ifree);
 805         XFS_SB_UNLOCK(mp, s);
 806
 807         id = huge_encode_dev(mp->m_dev);
 808         statp->f_fsid.val[0] = (u32)id;
 809         statp->f_fsid.val[1] = (u32)(id >> 32);
 810         statp->f_namelen = MAXNAMELEN - 1;
 811
 812         return 0;
 813 }
 814
 815
 816 /*
 817  * xfs_sync flushes any pending I/O to file system vfsp.
 818  *
 819  * This routine is called by vfs_sync() to make sure that things make it
 820  * out to disk eventually, on sync() system calls to flush out everything,
 821  * and when the file system is unmounted.  For the vfs_sync() case, all
 822  * we really need to do is sync out the log to make all of our meta-data
 823  * updates permanent (except for timestamps).  For calls from pflushd(),
 824  * dirty pages are kept moving by calling pdflush() on the inodes
 825  * containing them.  We also flush the inodes that we can lock without
 826  * sleeping and the superblock if we can lock it without sleeping from
 827  * vfs_sync() so that items at the tail of the log are always moving out.
 828  *
 829  * Flags:
 830  *      SYNC_BDFLUSH - We're being called from vfs_sync() so we don't want
 831  *                     to sleep if we can help it.  All we really need
 832  *                     to do is ensure that the log is synced at least
 833  *                     periodically.  We also push the inodes and
 834  *                     superblock if we can lock them without sleeping
 835  *                      and they are not pinned.
 836  *      SYNC_ATTR    - We need to flush the inodes.  If SYNC_BDFLUSH is not
 837  *                     set, then we really want to lock each inode and flush
 838  *                     it.
 839  *      SYNC_WAIT    - All the flushes that take place in this call should
 840  *                     be synchronous.
 841  *      SYNC_DELWRI  - This tells us to push dirty pages associated with
 842  *                     inodes.  SYNC_WAIT and SYNC_BDFLUSH are used to
 843  *                     determine if they should be flushed sync, async, or
 844  *                     delwri.
 845  *      SYNC_CLOSE   - This flag is passed when the system is being
 846  *                     unmounted.  We should sync and invalidate everthing.
 847  *      SYNC_FSDATA  - This indicates that the caller would like to make
 848  *                     sure the superblock is safe on disk.  We can ensure
 849  *                     this by simply makeing sure the log gets flushed
 850  *                     if SYNC_BDFLUSH is set, and by actually writing it
 851  *                     out otherwise.
 852  *
 853  */
 854 /*ARGSUSED*/
 855 STATIC int
 856 xfs_sync(
 857         bhv_desc_t      *bdp,
 858         int             flags,
 859         cred_t          *credp)
 860 {
 861         xfs_mount_t     *mp;
 862
 863         mp = XFS_BHVTOM(bdp);
 864         return (xfs_syncsub(mp, flags, 0, NULL));
 865 }
 866
 867 /*
 868  * xfs sync routine for internal use
 869  *
 870  * This routine supports all of the flags defined for the generic VFS_SYNC
 871  * interface as explained above under xfs_sync.  In the interests of not
 872  * changing interfaces within the 6.5 family, additional internallly-
 873  * required functions are specified within a separate xflags parameter,
 874  * only available by calling this routine.
 875  *
 876  */
 877 STATIC int
 878 xfs_sync_inodes(
 879         xfs_mount_t     *mp,
 880         int             flags,
 881         int             xflags,
 882         int             *bypassed)
 883 {
 884         xfs_inode_t     *ip = NULL;
 885         xfs_inode_t     *ip_next;
 886         xfs_buf_t       *bp;
 887         vnode_t         *vp = NULL;
 888         vmap_t          vmap;
 889         int             error;
 890         int             last_error;
 891         uint64_t        fflag;
 892         uint            lock_flags;
 893         uint            base_lock_flags;
 894         boolean_t       mount_locked;
 895         boolean_t       vnode_refed;
 896         int             preempt;
 897         xfs_dinode_t    *dip;
 898         xfs_iptr_t      *ipointer;
 899 #ifdef DEBUG
 900         boolean_t       ipointer_in = B_FALSE;
 901
 902 #define IPOINTER_SET    ipointer_in = B_TRUE
 903 #define IPOINTER_CLR    ipointer_in = B_FALSE
 904 #else
 905 #define IPOINTER_SET
 906 #define IPOINTER_CLR
 907 #endif
 908
 909
 910 /* Insert a marker record into the inode list after inode ip. The list
 911  * must be locked when this is called. After the call the list will no
 912  * longer be locked.
 913  */
 914 #define IPOINTER_INSERT(ip, mp) { \
 915                 ASSERT(ipointer_in == B_FALSE); \
 916                 ipointer->ip_mnext = ip->i_mnext; \
 917                 ipointer->ip_mprev = ip; \
 918                 ip->i_mnext = (xfs_inode_t *)ipointer; \
 919                 ipointer->ip_mnext->i_mprev = (xfs_inode_t *)ipointer; \
 920                 preempt = 0; \
 921                 XFS_MOUNT_IUNLOCK(mp); \
 922                 mount_locked = B_FALSE; \
 923                 IPOINTER_SET; \
 924         }
 925
 926 /* Remove the marker from the inode list. If the marker was the only item
 927  * in the list then there are no remaining inodes and we should zero out
 928  * the whole list. If we are the current head of the list then move the head
 929  * past us.
 930  */
 931 #define IPOINTER_REMOVE(ip, mp) { \
 932                 ASSERT(ipointer_in == B_TRUE); \
 933                 if (ipointer->ip_mnext != (xfs_inode_t *)ipointer) { \
 934                         ip = ipointer->ip_mnext; \
 935                         ip->i_mprev = ipointer->ip_mprev; \
 936                         ipointer->ip_mprev->i_mnext = ip; \
 937                         if (mp->m_inodes == (xfs_inode_t *)ipointer) { \
 938                                 mp->m_inodes = ip; \
 939                         } \
 940                 } else { \
 941                         ASSERT(mp->m_inodes == (xfs_inode_t *)ipointer); \
 942                         mp->m_inodes = NULL; \
 943                         ip = NULL; \
 944                 } \
 945                 IPOINTER_CLR; \
 946         }
 947
 948 #define XFS_PREEMPT_MASK        0x7f
 949
 950         if (bypassed)
 951                 *bypassed = 0;
 952         if (XFS_MTOVFS(mp)->vfs_flag & VFS_RDONLY)
 953                 return 0;
 954         error = 0;
 955         last_error = 0;
 956         preempt = 0;
 957
 958         /* Allocate a reference marker */
 959         ipointer = (xfs_iptr_t *)kmem_zalloc(sizeof(xfs_iptr_t), KM_SLEEP);
 960
 961         fflag = XFS_B_ASYNC;            /* default is don't wait */
 962         if (flags & SYNC_BDFLUSH)
 963                 fflag = XFS_B_DELWRI;
 964         if (flags & SYNC_WAIT)
 965                 fflag = 0;              /* synchronous overrides all */
 966
 967         base_lock_flags = XFS_ILOCK_SHARED;
 968         if (flags & (SYNC_DELWRI | SYNC_CLOSE)) {
 969                 /*
 970                  * We need the I/O lock if we're going to call any of
 971                  * the flush/inval routines.
 972                  */
 973                 base_lock_flags |= XFS_IOLOCK_SHARED;
 974         }
 975
 976         XFS_MOUNT_ILOCK(mp);
 977
 978         ip = mp->m_inodes;
 979
 980         mount_locked = B_TRUE;
 981         vnode_refed  = B_FALSE;
 982
 983         IPOINTER_CLR;
 984
 985         do {
 986                 ASSERT(ipointer_in == B_FALSE);
 987                 ASSERT(vnode_refed == B_FALSE);
 988
 989                 lock_flags = base_lock_flags;
 990
 991                 /*
 992                  * There were no inodes in the list, just break out
 993                  * of the loop.
 994                  */
 995                 if (ip == NULL) {
 996                         break;
 997                 }
 998
 999                 /*
1000                  * We found another sync thread marker - skip it
1001                  */
1002                 if (ip->i_mount == NULL) {
1003                         ip = ip->i_mnext;
1004                         continue;
1005                 }
1006
1007                 vp = XFS_ITOV_NULL(ip);
1008
1009                 /*
1010                  * If the vnode is gone then this is being torn down,
1011                  * call reclaim if it is flushed, else let regular flush
1012                  * code deal with it later in the loop.
1013                  */
1014
1015                 if (vp == NULL) {
1016                         /* Skip ones already in reclaim */
1017                         if (ip->i_flags & XFS_IRECLAIM) {
1018                                 ip = ip->i_mnext;
1019                                 continue;
1020                         }
1021                         if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0) {
1022                                 ip = ip->i_mnext;
1023                         } else if ((xfs_ipincount(ip) == 0) &&
1024                                     xfs_iflock_nowait(ip)) {
1025                                 IPOINTER_INSERT(ip, mp);
1026
1027                                 xfs_finish_reclaim(ip, 1,
1028                                                 XFS_IFLUSH_DELWRI_ELSE_ASYNC);
1029
1030                                 XFS_MOUNT_ILOCK(mp);
1031                                 mount_locked = B_TRUE;
1032                                 IPOINTER_REMOVE(ip, mp);
1033                         } else {
1034                                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1035                                 ip = ip->i_mnext;
1036                         }
1037                         continue;
1038                 }
1039
1040                 if (XFS_FORCED_SHUTDOWN(mp) && !(flags & SYNC_CLOSE)) {
1041                         XFS_MOUNT_IUNLOCK(mp);
1042                         kmem_free(ipointer, sizeof(xfs_iptr_t));
1043                         return 0;
1044                 }
1045
1046                 /*
1047                  * If this is just vfs_sync() or pflushd() calling
1048                  * then we can skip inodes for which it looks like
1049                  * there is nothing to do.  Since we don't have the
1050                  * inode locked this is racey, but these are periodic
1051                  * calls so it doesn't matter.  For the others we want
1052                  * to know for sure, so we at least try to lock them.
1053                  */
1054                 if (flags & SYNC_BDFLUSH) {
1055                         if (((ip->i_itemp == NULL) ||
1056                              !(ip->i_itemp->ili_format.ilf_fields &
1057                                XFS_ILOG_ALL)) &&
1058                             (ip->i_update_core == 0)) {
1059                                 ip = ip->i_mnext;
1060                                 continue;
1061                         }
1062                 }
1063
1064                 /*
1065                  * Try to lock without sleeping.  We're out of order with
1066                  * the inode list lock here, so if we fail we need to drop
1067                  * the mount lock and try again.  If we're called from
1068                  * bdflush() here, then don't bother.
1069                  *
1070                  * The inode lock here actually coordinates with the
1071                  * almost spurious inode lock in xfs_ireclaim() to prevent
1072                  * the vnode we handle here without a reference from
1073                  * being freed while we reference it.  If we lock the inode
1074                  * while it's on the mount list here, then the spurious inode
1075                  * lock in xfs_ireclaim() after the inode is pulled from
1076                  * the mount list will sleep until we release it here.
1077                  * This keeps the vnode from being freed while we reference
1078                  * it.  It is also cheaper and simpler than actually doing
1079                  * a vn_get() for every inode we touch here.
1080                  */
1081                 if (xfs_ilock_nowait(ip, lock_flags) == 0) {
1082
1083                         if ((flags & SYNC_BDFLUSH) || (vp == NULL)) {
1084                                 ip = ip->i_mnext;
1085                                 continue;
1086                         }
1087
1088                         /*
1089                          * We need to unlock the inode list lock in order
1090                          * to lock the inode. Insert a marker record into
1091                          * the inode list to remember our position, dropping
1092                          * the lock is now done inside the IPOINTER_INSERT
1093                          * macro.
1094                          *
1095                          * We also use the inode list lock to protect us
1096                          * in taking a snapshot of the vnode version number
1097                          * for use in calling vn_get().
1098                          */
1099                         VMAP(vp, vmap);
1100                         IPOINTER_INSERT(ip, mp);
1101
1102                         vp = vn_get(vp, &vmap);
1103                         if (vp == NULL) {
1104                                 /*
1105                                  * The vnode was reclaimed once we let go
1106                                  * of the inode list lock.  Skip to the
1107                                  * next list entry. Remove the marker.
1108                                  */
1109
1110                                 XFS_MOUNT_ILOCK(mp);
1111
1112                                 mount_locked = B_TRUE;
1113                                 vnode_refed  = B_FALSE;
1114
1115                                 IPOINTER_REMOVE(ip, mp);
1116
1117                                 continue;
1118                         }
1119
1120                         xfs_ilock(ip, lock_flags);
1121
1122                         ASSERT(vp == XFS_ITOV(ip));
1123                         ASSERT(ip->i_mount == mp);
1124
1125                         vnode_refed = B_TRUE;
1126                 }
1127
1128                 /* From here on in the loop we may have a marker record
1129                  * in the inode list.
1130                  */
1131
1132                 if ((flags & SYNC_CLOSE)  && (vp != NULL)) {
1133                         /*
1134                          * This is the shutdown case.  We just need to
1135                          * flush and invalidate all the pages associated
1136                          * with the inode.  Drop the inode lock since
1137                          * we can't hold it across calls to the buffer
1138                          * cache.
1139                          *
1140                          * We don't set the VREMAPPING bit in the vnode
1141                          * here, because we don't hold the vnode lock
1142                          * exclusively.  It doesn't really matter, though,
1143                          * because we only come here when we're shutting
1144                          * down anyway.
1145                          */
1146                         xfs_iunlock(ip, XFS_ILOCK_SHARED);
1147
1148                         if (XFS_FORCED_SHUTDOWN(mp)) {
1149                                 VOP_TOSS_PAGES(vp, 0, -1, FI_REMAPF);
1150                         } else {
1151                                 VOP_FLUSHINVAL_PAGES(vp, 0, -1, FI_REMAPF);
1152                         }
1153
1154                         xfs_ilock(ip, XFS_ILOCK_SHARED);
1155
1156                 } else if ((flags & SYNC_DELWRI) && (vp != NULL)) {
1157                         if (VN_DIRTY(vp)) {
1158                                 /* We need to have dropped the lock here,
1159                                  * so insert a marker if we have not already
1160                                  * done so.
1161                                  */
1162                                 if (mount_locked) {
1163                                         IPOINTER_INSERT(ip, mp);
1164                                 }
1165
1166                                 /*
1167                                  * Drop the inode lock since we can't hold it
1168                                  * across calls to the buffer cache.
1169                                  */
1170                                 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1171                                 VOP_FLUSH_PAGES(vp, (xfs_off_t)0, -1,
1172                                                         fflag, FI_NONE, error);
1173                                 xfs_ilock(ip, XFS_ILOCK_SHARED);
1174                         }
1175
1176                 }
1177
1178                 if (flags & SYNC_BDFLUSH) {
1179                         if ((flags & SYNC_ATTR) &&
1180                             ((ip->i_update_core) ||
1181                              ((ip->i_itemp != NULL) &&
1182                               (ip->i_itemp->ili_format.ilf_fields != 0)))) {
1183
1184                                 /* Insert marker and drop lock if not already
1185                                  * done.
1186                                  */
1187                                 if (mount_locked) {
1188                                         IPOINTER_INSERT(ip, mp);
1189                                 }
1190
1191                                 /*
1192                                  * We don't want the periodic flushing of the
1193                                  * inodes by vfs_sync() to interfere with
1194                                  * I/O to the file, especially read I/O
1195                                  * where it is only the access time stamp
1196                                  * that is being flushed out.  To prevent
1197                                  * long periods where we have both inode
1198                                  * locks held shared here while reading the
1199                                  * inode's buffer in from disk, we drop the
1200                                  * inode lock while reading in the inode
1201                                  * buffer.  We have to release the buffer
1202                                  * and reacquire the inode lock so that they
1203                                  * are acquired in the proper order (inode
1204                                  * locks first).  The buffer will go at the
1205                                  * end of the lru chain, though, so we can
1206                                  * expect it to still be there when we go
1207                                  * for it again in xfs_iflush().
1208                                  */
1209                                 if ((xfs_ipincount(ip) == 0) &&
1210                                     xfs_iflock_nowait(ip)) {
1211
1212                                         xfs_ifunlock(ip);
1213                                         xfs_iunlock(ip, XFS_ILOCK_SHARED);
1214
1215                                         error = xfs_itobp(mp, NULL, ip,
1216                                                           &dip, &bp, 0);
1217                                         if (!error) {
1218                                                 xfs_buf_relse(bp);
1219                                         } else {
1220                                                 /* Bailing out, remove the
1221                                                  * marker and free it.
1222                                                  */
1223                                                 XFS_MOUNT_ILOCK(mp);
1224
1225                                                 IPOINTER_REMOVE(ip, mp);
1226
1227                                                 XFS_MOUNT_IUNLOCK(mp);
1228
1229                                                 ASSERT(!(lock_flags &
1230                                                         XFS_IOLOCK_SHARED));
1231
1232                                                 kmem_free(ipointer,
1233                                                         sizeof(xfs_iptr_t));
1234                                                 return (0);
1235                                         }
1236
1237                                         /*
1238                                          * Since we dropped the inode lock,
1239                                          * the inode may have been reclaimed.
1240                                          * Therefore, we reacquire the mount
1241                                          * lock and check to see if we were the
1242                                          * inode reclaimed. If this happened
1243                                          * then the ipointer marker will no
1244                                          * longer point back at us. In this
1245                                          * case, move ip along to the inode
1246                                          * after the marker, remove the marker
1247                                          * and continue.
1248                                          */
1249                                         XFS_MOUNT_ILOCK(mp);
1250                                         mount_locked = B_TRUE;
1251
1252                                         if (ip != ipointer->ip_mprev) {
1253                                                 IPOINTER_REMOVE(ip, mp);
1254
1255                                                 ASSERT(!vnode_refed);
1256                                                 ASSERT(!(lock_flags &
1257                                                         XFS_IOLOCK_SHARED));
1258                                                 continue;
1259                                         }
1260
1261                                         ASSERT(ip->i_mount == mp);
1262
1263                                         if (xfs_ilock_nowait(ip,
1264                                                     XFS_ILOCK_SHARED) == 0) {
1265                                                 ASSERT(ip->i_mount == mp);
1266                                                 /*
1267                                                  * We failed to reacquire
1268                                                  * the inode lock without
1269                                                  * sleeping, so just skip
1270                                                  * the inode for now.  We
1271                                                  * clear the ILOCK bit from
1272                                                  * the lock_flags so that we
1273                                                  * won't try to drop a lock
1274                                                  * we don't hold below.
1275                                                  */
1276                                                 lock_flags &= ~XFS_ILOCK_SHARED;
1277                                                 IPOINTER_REMOVE(ip_next, mp);
1278                                         } else if ((xfs_ipincount(ip) == 0) &&
1279                                                    xfs_iflock_nowait(ip)) {
1280                                                 ASSERT(ip->i_mount == mp);
1281                                                 /*
1282                                                  * Since this is vfs_sync()
1283                                                  * calling we only flush the
1284                                                  * inode out if we can lock
1285                                                  * it without sleeping and
1286                                                  * it is not pinned.  Drop
1287                                                  * the mount lock here so
1288                                                  * that we don't hold it for
1289                                                  * too long. We already have
1290                                                  * a marker in the list here.
1291                                                  */
1292                                                 XFS_MOUNT_IUNLOCK(mp);
1293                                                 mount_locked = B_FALSE;
1294                                                 error = xfs_iflush(ip,
1295                                                            XFS_IFLUSH_DELWRI);
1296                                         } else {
1297                                                 ASSERT(ip->i_mount == mp);
1298                                                 IPOINTER_REMOVE(ip_next, mp);
1299                                         }
1300                                 }
1301
1302                         }
1303
1304                 } else {
1305                         if ((flags & SYNC_ATTR) &&
1306                             ((ip->i_update_core) ||
1307                              ((ip->i_itemp != NULL) &&
1308                               (ip->i_itemp->ili_format.ilf_fields != 0)))) {
1309                                 if (mount_locked) {
1310                                         IPOINTER_INSERT(ip, mp);
1311                                 }
1312
1313                                 if (flags & SYNC_WAIT) {
1314                                         xfs_iflock(ip);
1315                                         error = xfs_iflush(ip,
1316                                                            XFS_IFLUSH_SYNC);
1317                                 } else {
1318                                         /*
1319                                          * If we can't acquire the flush
1320                                          * lock, then the inode is already
1321                                          * being flushed so don't bother
1322                                          * waiting.  If we can lock it then
1323                                          * do a delwri flush so we can
1324                                          * combine multiple inode flushes
1325                                          * in each disk write.
1326                                          */
1327                                         if (xfs_iflock_nowait(ip)) {
1328                                                 error = xfs_iflush(ip,
1329                                                            XFS_IFLUSH_DELWRI);
1330                                         }
1331                                         else if (bypassed)
1332                                                 (*bypassed)++;
1333                                 }
1334                         }
1335                 }
1336
1337                 if (lock_flags != 0) {
1338                         xfs_iunlock(ip, lock_flags);
1339                 }
1340
1341                 if (vnode_refed) {
1342                         /*
1343                          * If we had to take a reference on the vnode
1344                          * above, then wait until after we've unlocked
1345                          * the inode to release the reference.  This is
1346                          * because we can be already holding the inode
1347                          * lock when VN_RELE() calls xfs_inactive().
1348                          *
1349                          * Make sure to drop the mount lock before calling
1350                          * VN_RELE() so that we don't trip over ourselves if
1351                          * we have to go for the mount lock again in the
1352                          * inactive code.
1353                          */
1354                         if (mount_locked) {
1355                                 IPOINTER_INSERT(ip, mp);
1356                         }
1357
1358                         VN_RELE(vp);
1359
1360                         vnode_refed = B_FALSE;
1361                 }
1362
1363                 if (error) {
1364                         last_error = error;
1365                 }
1366
1367                 /*
1368                  * bail out if the filesystem is corrupted.
1369                  */
1370                 if (error == EFSCORRUPTED)  {
1371                         if (!mount_locked) {
1372                                 XFS_MOUNT_ILOCK(mp);
1373                                 IPOINTER_REMOVE(ip, mp);
1374                         }
1375                         XFS_MOUNT_IUNLOCK(mp);
1376                         ASSERT(ipointer_in == B_FALSE);
1377                         kmem_free(ipointer, sizeof(xfs_iptr_t));
1378                         return XFS_ERROR(error);
1379                 }
1380
1381                 /* Let other threads have a chance at the mount lock
1382                  * if we have looped many times without dropping the
1383                  * lock.
1384                  */
1385                 if ((++preempt & XFS_PREEMPT_MASK) == 0) {
1386                         if (mount_locked) {
1387                                 IPOINTER_INSERT(ip, mp);
1388                         }
1389                 }
1390
1391                 if (mount_locked == B_FALSE) {
1392                         XFS_MOUNT_ILOCK(mp);
1393                         mount_locked = B_TRUE;
1394                         IPOINTER_REMOVE(ip, mp);
1395                         continue;
1396                 }
1397
1398                 ASSERT(ipointer_in == B_FALSE);
1399                 ip = ip->i_mnext;
1400
1401         } while (ip != mp->m_inodes);
1402
1403         XFS_MOUNT_IUNLOCK(mp);
1404
1405         ASSERT(ipointer_in == B_FALSE);
1406
1407         kmem_free(ipointer, sizeof(xfs_iptr_t));
1408         return XFS_ERROR(last_error);
1409 }
1410
1411 /*
1412  * xfs sync routine for internal use
1413  *
1414  * This routine supports all of the flags defined for the generic VFS_SYNC
1415  * interface as explained above under xfs_sync.  In the interests of not
1416  * changing interfaces within the 6.5 family, additional internallly-
1417  * required functions are specified within a separate xflags parameter,
1418  * only available by calling this routine.
1419  *
1420  */
1421 int
1422 xfs_syncsub(
1423         xfs_mount_t     *mp,
1424         int             flags,
1425         int             xflags,
1426         int             *bypassed)
1427 {
1428         int             error = 0;
1429         int             last_error = 0;
1430         uint            log_flags = XFS_LOG_FORCE;
1431         xfs_buf_t       *bp;
1432         xfs_buf_log_item_t      *bip;
1433
1434         /*
1435          * Sync out the log.  This ensures that the log is periodically
1436          * flushed even if there is not enough activity to fill it up.
1437          */
1438         if (flags & SYNC_WAIT)
1439                 log_flags |= XFS_LOG_SYNC;
1440
1441         xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1442
1443         if (flags & (SYNC_ATTR|SYNC_DELWRI)) {
1444                 if (flags & SYNC_BDFLUSH)
1445                         xfs_finish_reclaim_all(mp, 1);
1446                 else
1447                         error = xfs_sync_inodes(mp, flags, xflags, bypassed);
1448         }
1449
1450         /*
1451          * Flushing out dirty data above probably generated more
1452          * log activity, so if this isn't vfs_sync() then flush
1453          * the log again.
1454          */
1455         if (flags & SYNC_DELWRI) {
1456                 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1457         }
1458
1459         if (flags & SYNC_FSDATA) {
1460                 /*
1461                  * If this is vfs_sync() then only sync the superblock
1462                  * if we can lock it without sleeping and it is not pinned.
1463                  */
1464                 if (flags & SYNC_BDFLUSH) {
1465                         bp = xfs_getsb(mp, XFS_BUF_TRYLOCK);
1466                         if (bp != NULL) {
1467                                 bip = XFS_BUF_FSPRIVATE(bp,xfs_buf_log_item_t*);
1468                                 if ((bip != NULL) &&
1469                                     xfs_buf_item_dirty(bip)) {
1470                                         if (!(XFS_BUF_ISPINNED(bp))) {
1471                                                 XFS_BUF_ASYNC(bp);
1472                                                 error = xfs_bwrite(mp, bp);
1473                                         } else {
1474                                                 xfs_buf_relse(bp);
1475                                         }
1476                                 } else {
1477                                         xfs_buf_relse(bp);
1478                                 }
1479                         }
1480                 } else {
1481                         bp = xfs_getsb(mp, 0);
1482                         /*
1483                          * If the buffer is pinned then push on the log so
1484                          * we won't get stuck waiting in the write for
1485                          * someone, maybe ourselves, to flush the log.
1486                          * Even though we just pushed the log above, we
1487                          * did not have the superblock buffer locked at
1488                          * that point so it can become pinned in between
1489                          * there and here.
1490                          */
1491                         if (XFS_BUF_ISPINNED(bp))
1492                                 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
1493                         if (flags & SYNC_WAIT)
1494                                 XFS_BUF_UNASYNC(bp);
1495                         else
1496                                 XFS_BUF_ASYNC(bp);
1497                         error = xfs_bwrite(mp, bp);
1498                 }
1499                 if (error) {
1500                         last_error = error;
1501                 }
1502         }
1503
1504         /*
1505          * If this is the periodic sync, then kick some entries out of
1506          * the reference cache.  This ensures that idle entries are
1507          * eventually kicked out of the cache.
1508          */
1509         if (flags & SYNC_REFCACHE) {
1510                 xfs_refcache_purge_some(mp);
1511         }
1512
1513         /*
1514          * Now check to see if the log needs a "dummy" transaction.
1515          */
1516
1517         if (!(flags & SYNC_REMOUNT) && xfs_log_need_covered(mp)) {
1518                 xfs_trans_t *tp;
1519                 xfs_inode_t *ip;
1520
1521                 /*
1522                  * Put a dummy transaction in the log to tell
1523                  * recovery that all others are OK.
1524                  */
1525                 tp = xfs_trans_alloc(mp, XFS_TRANS_DUMMY1);
1526                 if ((error = xfs_trans_reserve(tp, 0,
1527                                 XFS_ICHANGE_LOG_RES(mp),
1528                                 0, 0, 0)))  {
1529                         xfs_trans_cancel(tp, 0);
1530                         return error;
1531                 }
1532
1533                 ip = mp->m_rootip;
1534                 xfs_ilock(ip, XFS_ILOCK_EXCL);
1535
1536                 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1537                 xfs_trans_ihold(tp, ip);
1538                 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1539                 error = xfs_trans_commit(tp, 0, NULL);
1540                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1541                 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1542         }
1543
1544         /*
1545          * When shutting down, we need to insure that the AIL is pushed
1546          * to disk or the filesystem can appear corrupt from the PROM.
1547          */
1548         if ((flags & (SYNC_CLOSE|SYNC_WAIT)) == (SYNC_CLOSE|SYNC_WAIT)) {
1549                 XFS_bflush(mp->m_ddev_targp);
1550                 if (mp->m_rtdev_targp) {
1551                         XFS_bflush(mp->m_rtdev_targp);
1552                 }
1553         }
1554
1555         return XFS_ERROR(last_error);
1556 }
1557
1558 /*
1559  * xfs_vget - called by DMAPI to get vnode from file handle
1560  */
1561 STATIC int
1562 xfs_vget(
1563         bhv_desc_t      *bdp,
1564         vnode_t         **vpp,
1565         fid_t           *fidp)
1566 {
1567         xfs_fid_t       *xfid;
1568         xfs_inode_t     *ip;
1569         int             error;
1570         xfs_ino_t       ino;
1571         unsigned int    igen;
1572         xfs_mount_t     *mp;
1573
1574         xfid  = (struct xfs_fid *)fidp;
1575         if (xfid->xfs_fid_len == sizeof(*xfid) - sizeof(xfid->xfs_fid_len)) {
1576                 ino  = xfid->xfs_fid_ino;
1577                 igen = xfid->xfs_fid_gen;
1578         } else {
1579                 /*
1580                  * Invalid.  Since handles can be created in user space
1581                  * and passed in via gethandle(), this is not cause for
1582                  * a panic.
1583                  */
1584                 return XFS_ERROR(EINVAL);
1585         }
1586         mp = XFS_BHVTOM(bdp);
1587         error = xfs_iget(mp, NULL, ino, XFS_ILOCK_SHARED, &ip, 0);
1588         if (error) {
1589                 *vpp = NULL;
1590                 return error;
1591         }
1592         if (ip == NULL) {
1593                 *vpp = NULL;
1594                 return XFS_ERROR(EIO);
1595         }
1596
1597         if (ip->i_d.di_mode == 0 || (igen && (ip->i_d.di_gen != igen))) {
1598                 xfs_iput_new(ip, XFS_ILOCK_SHARED);
1599                 *vpp = NULL;
1600                 return XFS_ERROR(ENOENT);
1601         }
1602
1603         *vpp = XFS_ITOV(ip);
1604         xfs_iunlock(ip, XFS_ILOCK_SHARED);
1605         return 0;
1606 }
1607
1608
1609 #define MNTOPT_LOGBUFS  "logbufs"       /* number of XFS log buffers */
1610 #define MNTOPT_LOGBSIZE "logbsize"      /* size of XFS log buffers */
1611 #define MNTOPT_LOGDEV   "logdev"        /* log device */
1612 #define MNTOPT_RTDEV    "rtdev"         /* realtime I/O device */
1613 #define MNTOPT_BIOSIZE  "biosize"       /* log2 of preferred buffered io size */
1614 #define MNTOPT_WSYNC    "wsync"         /* safe-mode nfs compatible mount */
1615 #define MNTOPT_INO64    "ino64"         /* force inodes into 64-bit range */
1616 #define MNTOPT_NOALIGN  "noalign"       /* turn off stripe alignment */
1617 #define MNTOPT_SUNIT    "sunit"         /* data volume stripe unit */
1618 #define MNTOPT_SWIDTH   "swidth"        /* data volume stripe width */
1619 #define MNTOPT_NOUUID   "nouuid"        /* ignore filesystem UUID */
1620 #define MNTOPT_MTPT     "mtpt"          /* filesystem mount point */
1621 #define MNTOPT_NORECOVERY   "norecovery"   /* don't run XFS recovery */
1622 #define MNTOPT_NOLOGFLUSH   "nologflush"   /* don't hard flush on log writes */
1623 #define MNTOPT_OSYNCISOSYNC "osyncisosync" /* o_sync is REALLY o_sync */
1624 #define MNTOPT_64BITINODE   "inode64"   /* inodes can be allocated anywhere */
1625 #define MNTOPT_IKEEP    "ikeep"         /* do not free empty inode clusters */
1626 #define MNTOPT_NOIKEEP  "noikeep"       /* free empty inode clusters */
1627
1628
1629 int
1630 xfs_parseargs(
1631         struct bhv_desc         *bhv,
1632         char                    *options,
1633         struct xfs_mount_args   *args,
1634         int                     update)
1635 {
1636         struct vfs              *vfsp = bhvtovfs(bhv);
1637         char                    *this_char, *value, *eov;
1638         int                     dsunit, dswidth, vol_dsunit, vol_dswidth;
1639         int                     iosize;
1640
1641 #if 0   /* XXX: off by default, until some remaining issues ironed out */
1642         args->flags |= XFSMNT_IDELETE; /* default to on */
1643 #endif
1644
1645         if (!options)
1646                 return 0;
1647
1648         iosize = dsunit = dswidth = vol_dsunit = vol_dswidth = 0;
1649
1650         while ((this_char = strsep(&options, ",")) != NULL) {
1651                 if (!*this_char)
1652                         continue;
1653                 if ((value = strchr(this_char, '=')) != NULL)
1654                         *value++ = 0;
1655
1656                 if (!strcmp(this_char, MNTOPT_LOGBUFS)) {
1657                         if (!value || !*value) {
1658                                 printk("XFS: %s option requires an argument\n",
1659                                         MNTOPT_LOGBUFS);
1660                                 return EINVAL;
1661                         }
1662                         args->logbufs = simple_strtoul(value, &eov, 10);
1663                 } else if (!strcmp(this_char, MNTOPT_LOGBSIZE)) {
1664                         int     last, in_kilobytes = 0;
1665
1666                         if (!value || !*value) {
1667                                 printk("XFS: %s option requires an argument\n",
1668                                         MNTOPT_LOGBSIZE);
1669                                 return EINVAL;
1670                         }
1671                         last = strlen(value) - 1;
1672                         if (value[last] == 'K' || value[last] == 'k') {
1673                                 in_kilobytes = 1;
1674                                 value[last] = '\0';
1675                         }
1676                         args->logbufsize = simple_strtoul(value, &eov, 10);
1677                         if (in_kilobytes)
1678                                 args->logbufsize <<= 10;
1679                 } else if (!strcmp(this_char, MNTOPT_LOGDEV)) {
1680                         if (!value || !*value) {
1681                                 printk("XFS: %s option requires an argument\n",
1682                                         MNTOPT_LOGDEV);
1683                                 return EINVAL;
1684                         }
1685                         strncpy(args->logname, value, MAXNAMELEN);
1686                 } else if (!strcmp(this_char, MNTOPT_MTPT)) {
1687                         if (!value || !*value) {
1688                                 printk("XFS: %s option requires an argument\n",
1689                                         MNTOPT_MTPT);
1690                                 return EINVAL;
1691                         }
1692                         strncpy(args->mtpt, value, MAXNAMELEN);
1693                 } else if (!strcmp(this_char, MNTOPT_RTDEV)) {
1694                         if (!value || !*value) {
1695                                 printk("XFS: %s option requires an argument\n",
1696                                         MNTOPT_RTDEV);
1697                                 return EINVAL;
1698                         }
1699                         strncpy(args->rtname, value, MAXNAMELEN);
1700                 } else if (!strcmp(this_char, MNTOPT_BIOSIZE)) {
1701                         if (!value || !*value) {
1702                                 printk("XFS: %s option requires an argument\n",
1703                                         MNTOPT_BIOSIZE);
1704                                 return EINVAL;
1705                         }
1706                         iosize = simple_strtoul(value, &eov, 10);
1707                         args->flags |= XFSMNT_IOSIZE;
1708                         args->iosizelog = (uint8_t) iosize;
1709                 } else if (!strcmp(this_char, MNTOPT_WSYNC)) {
1710                         args->flags |= XFSMNT_WSYNC;
1711                 } else if (!strcmp(this_char, MNTOPT_OSYNCISOSYNC)) {
1712                         args->flags |= XFSMNT_OSYNCISOSYNC;
1713                 } else if (!strcmp(this_char, MNTOPT_NORECOVERY)) {
1714                         args->flags |= XFSMNT_NORECOVERY;
1715                 } else if (!strcmp(this_char, MNTOPT_INO64)) {
1716                         args->flags |= XFSMNT_INO64;
1717 #if !XFS_BIG_INUMS
1718                         printk("XFS: %s option not allowed on this system\n",
1719                                 MNTOPT_INO64);
1720                         return EINVAL;
1721 #endif
1722                 } else if (!strcmp(this_char, MNTOPT_NOALIGN)) {
1723                         args->flags |= XFSMNT_NOALIGN;
1724                 } else if (!strcmp(this_char, MNTOPT_SUNIT)) {
1725                         if (!value || !*value) {
1726                                 printk("XFS: %s option requires an argument\n",
1727                                         MNTOPT_SUNIT);
1728                                 return EINVAL;
1729                         }
1730                         dsunit = simple_strtoul(value, &eov, 10);
1731                 } else if (!strcmp(this_char, MNTOPT_SWIDTH)) {
1732                         if (!value || !*value) {
1733                                 printk("XFS: %s option requires an argument\n",
1734                                         MNTOPT_SWIDTH);
1735                                 return EINVAL;
1736                         }
1737                         dswidth = simple_strtoul(value, &eov, 10);
1738                 } else if (!strcmp(this_char, MNTOPT_64BITINODE)) {
1739                         args->flags &= ~XFSMNT_32BITINODES;
1740 #if !XFS_BIG_INUMS
1741                         printk("XFS: %s option not allowed on this system\n",
1742                                 MNTOPT_64BITINODE);
1743                         return EINVAL;
1744 #endif
1745                 } else if (!strcmp(this_char, MNTOPT_NOUUID)) {
1746                         args->flags |= XFSMNT_NOUUID;
1747                 } else if (!strcmp(this_char, MNTOPT_NOLOGFLUSH)) {
1748                         args->flags |= XFSMNT_NOLOGFLUSH;
1749                 } else if (!strcmp(this_char, MNTOPT_IKEEP)) {
1750                         args->flags &= ~XFSMNT_IDELETE;
1751                 } else if (!strcmp(this_char, MNTOPT_NOIKEEP)) {
1752                         args->flags |= XFSMNT_IDELETE;
1753                 } else if (!strcmp(this_char, "osyncisdsync")) {
1754                         /* no-op, this is now the default */
1755 printk("XFS: osyncisdsync is now the default, option is deprecated.\n");
1756                 } else if (!strcmp(this_char, "irixsgid")) {
1757 printk("XFS: irixsgid is now a sysctl(2) variable, option is deprecated.\n");
1758                 } else {
1759                         printk("XFS: unknown mount option [%s].\n", this_char);
1760                         return EINVAL;
1761                 }
1762         }
1763
1764         if (args->flags & XFSMNT_NORECOVERY) {
1765                 if ((vfsp->vfs_flag & VFS_RDONLY) == 0) {
1766                         printk("XFS: no-recovery mounts must be read-only.\n");
1767                         return EINVAL;
1768                 }
1769         }
1770
1771         if ((args->flags & XFSMNT_NOALIGN) && (dsunit || dswidth)) {
1772                 printk(
1773         "XFS: sunit and swidth options incompatible with the noalign option\n");
1774                 return EINVAL;
1775         }
1776
1777         if ((dsunit && !dswidth) || (!dsunit && dswidth)) {
1778                 printk("XFS: sunit and swidth must be specified together\n");
1779                 return EINVAL;
1780         }
1781
1782         if (dsunit && (dswidth % dsunit != 0)) {
1783                 printk(
1784         "XFS: stripe width (%d) must be a multiple of the stripe unit (%d)\n",
1785                         dswidth, dsunit);
1786                 return EINVAL;
1787         }
1788
1789         if ((args->flags & XFSMNT_NOALIGN) != XFSMNT_NOALIGN) {
1790                 if (dsunit) {
1791                         args->sunit = dsunit;
1792                         args->flags |= XFSMNT_RETERR;
1793                 } else {
1794                         args->sunit = vol_dsunit;
1795                 }
1796                 dswidth ? (args->swidth = dswidth) :
1797                           (args->swidth = vol_dswidth);
1798         } else {
1799                 args->sunit = args->swidth = 0;
1800         }
1801
1802         return 0;
1803 }
1804
1805 int
1806 xfs_showargs(
1807         struct bhv_desc         *bhv,
1808         struct seq_file         *m)
1809 {
1810         static struct proc_xfs_info {
1811                 int     flag;
1812                 char    *str;
1813         } xfs_info[] = {
1814                 /* the few simple ones we can get from the mount struct */
1815                 { XFS_MOUNT_WSYNC,              "," MNTOPT_WSYNC },
1816                 { XFS_MOUNT_INO64,              "," MNTOPT_INO64 },
1817                 { XFS_MOUNT_NOALIGN,            "," MNTOPT_NOALIGN },
1818                 { XFS_MOUNT_NOUUID,             "," MNTOPT_NOUUID },
1819                 { XFS_MOUNT_NORECOVERY,         "," MNTOPT_NORECOVERY },
1820                 { XFS_MOUNT_OSYNCISOSYNC,       "," MNTOPT_OSYNCISOSYNC },
1821                 { XFS_MOUNT_NOLOGFLUSH,         "," MNTOPT_NOLOGFLUSH },
1822                 { XFS_MOUNT_IDELETE,            "," MNTOPT_NOIKEEP },
1823                 { 0, NULL }
1824         };
1825         struct proc_xfs_info    *xfs_infop;
1826         struct xfs_mount        *mp = XFS_BHVTOM(bhv);
1827
1828         for (xfs_infop = xfs_info; xfs_infop->flag; xfs_infop++) {
1829                 if (mp->m_flags & xfs_infop->flag)
1830                         seq_puts(m, xfs_infop->str);
1831         }
1832
1833         if (mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)
1834                 seq_printf(m, "," MNTOPT_BIOSIZE "=%d", mp->m_writeio_log);
1835
1836         if (mp->m_logbufs > 0)
1837                 seq_printf(m, "," MNTOPT_LOGBUFS "=%d", mp->m_logbufs);
1838
1839         if (mp->m_logbsize > 0)
1840                 seq_printf(m, "," MNTOPT_LOGBSIZE "=%d", mp->m_logbsize);
1841
1842         if (mp->m_ddev_targp != mp->m_logdev_targp)
1843                 seq_printf(m, "," MNTOPT_LOGDEV "=%s",
1844                                 XFS_BUFTARG_NAME(mp->m_logdev_targp));
1845
1846         if (mp->m_rtdev_targp && mp->m_ddev_targp != mp->m_rtdev_targp)
1847                 seq_printf(m, "," MNTOPT_RTDEV "=%s",
1848                                 XFS_BUFTARG_NAME(mp->m_rtdev_targp));
1849
1850         if (mp->m_dalign > 0)
1851                 seq_printf(m, "," MNTOPT_SUNIT "=%d",
1852                                 (int)XFS_FSB_TO_BB(mp, mp->m_dalign));
1853
1854         if (mp->m_swidth > 0)
1855                 seq_printf(m, "," MNTOPT_SWIDTH "=%d",
1856                                 (int)XFS_FSB_TO_BB(mp, mp->m_swidth));
1857
1858         if (!(mp->m_flags & XFS_MOUNT_32BITINOOPT))
1859                 seq_printf(m, "," MNTOPT_64BITINODE);
1860
1861         return 0;
1862 }
1863
1864 STATIC void
1865 xfs_freeze(
1866         bhv_desc_t      *bdp)
1867 {
1868         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
1869
1870         while (atomic_read(&mp->m_active_trans) > 0)
1871                 delay(100);
1872
1873         /* Push the superblock and write an unmount record */
1874         xfs_log_unmount_write(mp);
1875         xfs_unmountfs_writesb(mp);
1876 }
1877
1878
1879 vfsops_t xfs_vfsops = {
1880         BHV_IDENTITY_INIT(VFS_BHV_XFS,VFS_POSITION_XFS),
1881         .vfs_parseargs          = xfs_parseargs,
1882         .vfs_showargs           = xfs_showargs,
1883         .vfs_mount              = xfs_mount,
1884         .vfs_unmount            = xfs_unmount,
1885         .vfs_mntupdate          = xfs_mntupdate,
1886         .vfs_root               = xfs_root,
1887         .vfs_statvfs            = xfs_statvfs,
1888         .vfs_sync               = xfs_sync,
1889         .vfs_vget               = xfs_vget,
1890         .vfs_dmapiops           = (vfs_dmapiops_t)fs_nosys,
1891         .vfs_quotactl           = (vfs_quotactl_t)fs_nosys,
1892         .vfs_init_vnode         = xfs_initialize_vnode,
1893         .vfs_force_shutdown     = xfs_do_force_shutdown,
1894         .vfs_freeze             = xfs_freeze,
1895 };