fs/xfs/xfs_vfsops.c

   1 /*
   2  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
   3  * All Rights Reserved.
   4  *
   5  * This program is free software; you can redistribute it and/or
   6  * modify it under the terms of the GNU General Public License as
   7  * published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it would be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write the Free Software Foundation,
  16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17  */
  18 #include "xfs.h"
  19 #include "xfs_fs.h"
  20 #include "xfs_types.h"
  21 #include "xfs_bit.h"
  22 #include "xfs_log.h"
  23 #include "xfs_inum.h"
  24 #include "xfs_trans.h"
  25 #include "xfs_sb.h"
  26 #include "xfs_ag.h"
  27 #include "xfs_dir.h"
  28 #include "xfs_dir2.h"
  29 #include "xfs_dmapi.h"
  30 #include "xfs_mount.h"
  31 #include "xfs_da_btree.h"
  32 #include "xfs_bmap_btree.h"
  33 #include "xfs_ialloc_btree.h"
  34 #include "xfs_alloc_btree.h"
  35 #include "xfs_dir_sf.h"
  36 #include "xfs_dir2_sf.h"
  37 #include "xfs_attr_sf.h"
  38 #include "xfs_dinode.h"
  39 #include "xfs_inode.h"
  40 #include "xfs_inode_item.h"
  41 #include "xfs_btree.h"
  42 #include "xfs_alloc.h"
  43 #include "xfs_ialloc.h"
  44 #include "xfs_quota.h"
  45 #include "xfs_error.h"
  46 #include "xfs_bmap.h"
  47 #include "xfs_rw.h"
  48 #include "xfs_refcache.h"
  49 #include "xfs_buf_item.h"
  50 #include "xfs_log_priv.h"
  51 #include "xfs_dir2_trace.h"
  52 #include "xfs_extfree_item.h"
  53 #include "xfs_acl.h"
  54 #include "xfs_attr.h"
  55 #include "xfs_clnt.h"
  56
  57 STATIC int xfs_sync(bhv_desc_t *, int, cred_t *);
  58
  59 int
  60 xfs_init(void)
  61 {
  62         extern kmem_zone_t      *xfs_bmap_free_item_zone;
  63         extern kmem_zone_t      *xfs_btree_cur_zone;
  64         extern kmem_zone_t      *xfs_trans_zone;
  65         extern kmem_zone_t      *xfs_buf_item_zone;
  66         extern kmem_zone_t      *xfs_dabuf_zone;
  67 #ifdef XFS_DABUF_DEBUG
  68         extern lock_t           xfs_dabuf_global_lock;
  69         spinlock_init(&xfs_dabuf_global_lock, "xfsda");
  70 #endif
  71
  72         /*
  73          * Initialize all of the zone allocators we use.
  74          */
  75         xfs_bmap_free_item_zone = kmem_zone_init(sizeof(xfs_bmap_free_item_t),
  76                                                  "xfs_bmap_free_item");
  77         xfs_btree_cur_zone = kmem_zone_init(sizeof(xfs_btree_cur_t),
  78                                             "xfs_btree_cur");
  79         xfs_inode_zone = kmem_zone_init(sizeof(xfs_inode_t), "xfs_inode");
  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
  85         /*
  86          * The size of the zone allocated buf log item is the maximum
  87          * size possible under XFS.  This wastes a little bit of memory,
  88          * but it is much faster.
  89          */
  90         xfs_buf_item_zone =
  91                 kmem_zone_init((sizeof(xfs_buf_log_item_t) +
  92                                 (((XFS_MAX_BLOCKSIZE / XFS_BLI_CHUNK) /
  93                                   NBWORD) * sizeof(int))),
  94                                "xfs_buf_item");
  95         xfs_efd_zone = kmem_zone_init((sizeof(xfs_efd_log_item_t) +
  96                                        ((XFS_EFD_MAX_FAST_EXTENTS - 1) * sizeof(xfs_extent_t))),
  97                                       "xfs_efd_item");
  98         xfs_efi_zone = kmem_zone_init((sizeof(xfs_efi_log_item_t) +
  99                                        ((XFS_EFI_MAX_FAST_EXTENTS - 1) * sizeof(xfs_extent_t))),
 100                                       "xfs_efi_item");
 101         xfs_ifork_zone = kmem_zone_init(sizeof(xfs_ifork_t), "xfs_ifork");
 102         xfs_ili_zone = kmem_zone_init(sizeof(xfs_inode_log_item_t), "xfs_ili");
 103         xfs_chashlist_zone = kmem_zone_init(sizeof(xfs_chashlist_t),
 104                                             "xfs_chashlist");
 105         xfs_acl_zone_init(xfs_acl_zone, "xfs_acl");
 106
 107         /*
 108          * Allocate global trace buffers.
 109          */
 110 #ifdef XFS_ALLOC_TRACE
 111         xfs_alloc_trace_buf = ktrace_alloc(XFS_ALLOC_TRACE_SIZE, KM_SLEEP);
 112 #endif
 113 #ifdef XFS_BMAP_TRACE
 114         xfs_bmap_trace_buf = ktrace_alloc(XFS_BMAP_TRACE_SIZE, KM_SLEEP);
 115 #endif
 116 #ifdef XFS_BMBT_TRACE
 117         xfs_bmbt_trace_buf = ktrace_alloc(XFS_BMBT_TRACE_SIZE, KM_SLEEP);
 118 #endif
 119 #ifdef XFS_DIR_TRACE
 120         xfs_dir_trace_buf = ktrace_alloc(XFS_DIR_TRACE_SIZE, KM_SLEEP);
 121 #endif
 122 #ifdef XFS_ATTR_TRACE
 123         xfs_attr_trace_buf = ktrace_alloc(XFS_ATTR_TRACE_SIZE, KM_SLEEP);
 124 #endif
 125 #ifdef XFS_DIR2_TRACE
 126         xfs_dir2_trace_buf = ktrace_alloc(XFS_DIR2_GTRACE_SIZE, KM_SLEEP);
 127 #endif
 128
 129         xfs_dir_startup();
 130
 131 #if (defined(DEBUG) || defined(INDUCE_IO_ERROR))
 132         xfs_error_test_init();
 133 #endif /* DEBUG || INDUCE_IO_ERROR */
 134
 135         xfs_init_procfs();
 136         xfs_sysctl_register();
 137         return 0;
 138 }
 139
 140 void
 141 xfs_cleanup(void)
 142 {
 143         extern kmem_zone_t      *xfs_bmap_free_item_zone;
 144         extern kmem_zone_t      *xfs_btree_cur_zone;
 145         extern kmem_zone_t      *xfs_inode_zone;
 146         extern kmem_zone_t      *xfs_trans_zone;
 147         extern kmem_zone_t      *xfs_da_state_zone;
 148         extern kmem_zone_t      *xfs_dabuf_zone;
 149         extern kmem_zone_t      *xfs_efd_zone;
 150         extern kmem_zone_t      *xfs_efi_zone;
 151         extern kmem_zone_t      *xfs_buf_item_zone;
 152         extern kmem_zone_t      *xfs_chashlist_zone;
 153
 154         xfs_cleanup_procfs();
 155         xfs_sysctl_unregister();
 156         xfs_refcache_destroy();
 157         xfs_acl_zone_destroy(xfs_acl_zone);
 158
 159 #ifdef XFS_DIR2_TRACE
 160         ktrace_free(xfs_dir2_trace_buf);
 161 #endif
 162 #ifdef XFS_ATTR_TRACE
 163         ktrace_free(xfs_attr_trace_buf);
 164 #endif
 165 #ifdef XFS_DIR_TRACE
 166         ktrace_free(xfs_dir_trace_buf);
 167 #endif
 168 #ifdef XFS_BMBT_TRACE
 169         ktrace_free(xfs_bmbt_trace_buf);
 170 #endif
 171 #ifdef XFS_BMAP_TRACE
 172         ktrace_free(xfs_bmap_trace_buf);
 173 #endif
 174 #ifdef XFS_ALLOC_TRACE
 175         ktrace_free(xfs_alloc_trace_buf);
 176 #endif
 177
 178         kmem_cache_destroy(xfs_bmap_free_item_zone);
 179         kmem_cache_destroy(xfs_btree_cur_zone);
 180         kmem_cache_destroy(xfs_inode_zone);
 181         kmem_cache_destroy(xfs_trans_zone);
 182         kmem_cache_destroy(xfs_da_state_zone);
 183         kmem_cache_destroy(xfs_dabuf_zone);
 184         kmem_cache_destroy(xfs_buf_item_zone);
 185         kmem_cache_destroy(xfs_efd_zone);
 186         kmem_cache_destroy(xfs_efi_zone);
 187         kmem_cache_destroy(xfs_ifork_zone);
 188         kmem_cache_destroy(xfs_ili_zone);
 189         kmem_cache_destroy(xfs_chashlist_zone);
 190 }
 191
 192 /*
 193  * xfs_start_flags
 194  *
 195  * This function fills in xfs_mount_t fields based on mount args.
 196  * Note: the superblock has _not_ yet been read in.
 197  */
 198 STATIC int
 199 xfs_start_flags(
 200         struct vfs              *vfs,
 201         struct xfs_mount_args   *ap,
 202         struct xfs_mount        *mp)
 203 {
 204         /* Values are in BBs */
 205         if ((ap->flags & XFSMNT_NOALIGN) != XFSMNT_NOALIGN) {
 206                 /*
 207                  * At this point the superblock has not been read
 208                  * in, therefore we do not know the block size.
 209                  * Before the mount call ends we will convert
 210                  * these to FSBs.
 211                  */
 212                 mp->m_dalign = ap->sunit;
 213                 mp->m_swidth = ap->swidth;
 214         }
 215
 216         if (ap->logbufs != -1 &&
 217             ap->logbufs != 0 &&
 218             (ap->logbufs < XLOG_MIN_ICLOGS ||
 219              ap->logbufs > XLOG_MAX_ICLOGS)) {
 220                 cmn_err(CE_WARN,
 221                         "XFS: invalid logbufs value: %d [not %d-%d]",
 222                         ap->logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
 223                 return XFS_ERROR(EINVAL);
 224         }
 225         mp->m_logbufs = ap->logbufs;
 226         if (ap->logbufsize != -1 &&
 227             ap->logbufsize !=  0 &&
 228             ap->logbufsize != 16 * 1024 &&
 229             ap->logbufsize != 32 * 1024 &&
 230             ap->logbufsize != 64 * 1024 &&
 231             ap->logbufsize != 128 * 1024 &&
 232             ap->logbufsize != 256 * 1024) {
 233                 cmn_err(CE_WARN,
 234         "XFS: invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
 235                         ap->logbufsize);
 236                 return XFS_ERROR(EINVAL);
 237         }
 238         mp->m_ihsize = ap->ihashsize;
 239         mp->m_logbsize = ap->logbufsize;
 240         mp->m_fsname_len = strlen(ap->fsname) + 1;
 241         mp->m_fsname = kmem_alloc(mp->m_fsname_len, KM_SLEEP);
 242         strcpy(mp->m_fsname, ap->fsname);
 243         if (ap->rtname[0]) {
 244                 mp->m_rtname = kmem_alloc(strlen(ap->rtname) + 1, KM_SLEEP);
 245                 strcpy(mp->m_rtname, ap->rtname);
 246         }
 247         if (ap->logname[0]) {
 248                 mp->m_logname = kmem_alloc(strlen(ap->logname) + 1, KM_SLEEP);
 249                 strcpy(mp->m_logname, ap->logname);
 250         }
 251
 252         if (ap->flags & XFSMNT_WSYNC)
 253                 mp->m_flags |= XFS_MOUNT_WSYNC;
 254 #if XFS_BIG_INUMS
 255         if (ap->flags & XFSMNT_INO64) {
 256                 mp->m_flags |= XFS_MOUNT_INO64;
 257                 mp->m_inoadd = XFS_INO64_OFFSET;
 258         }
 259 #endif
 260         if (ap->flags & XFSMNT_NOATIME)
 261                 mp->m_flags |= XFS_MOUNT_NOATIME;
 262         if (ap->flags & XFSMNT_RETERR)
 263                 mp->m_flags |= XFS_MOUNT_RETERR;
 264         if (ap->flags & XFSMNT_NOALIGN)
 265                 mp->m_flags |= XFS_MOUNT_NOALIGN;
 266         if (ap->flags & XFSMNT_SWALLOC)
 267                 mp->m_flags |= XFS_MOUNT_SWALLOC;
 268         if (ap->flags & XFSMNT_OSYNCISOSYNC)
 269                 mp->m_flags |= XFS_MOUNT_OSYNCISOSYNC;
 270         if (ap->flags & XFSMNT_32BITINODES)
 271                 mp->m_flags |= XFS_MOUNT_32BITINODES;
 272
 273         if (ap->flags & XFSMNT_IOSIZE) {
 274                 if (ap->iosizelog > XFS_MAX_IO_LOG ||
 275                     ap->iosizelog < XFS_MIN_IO_LOG) {
 276                         cmn_err(CE_WARN,
 277                 "XFS: invalid log iosize: %d [not %d-%d]",
 278                                 ap->iosizelog, XFS_MIN_IO_LOG,
 279                                 XFS_MAX_IO_LOG);
 280                         return XFS_ERROR(EINVAL);
 281                 }
 282
 283                 mp->m_flags |= XFS_MOUNT_DFLT_IOSIZE;
 284                 mp->m_readio_log = mp->m_writeio_log = ap->iosizelog;
 285         }
 286
 287         if (ap->flags & XFSMNT_IHASHSIZE)
 288                 mp->m_flags |= XFS_MOUNT_IHASHSIZE;
 289         if (ap->flags & XFSMNT_IDELETE)
 290                 mp->m_flags |= XFS_MOUNT_IDELETE;
 291         if (ap->flags & XFSMNT_DIRSYNC)
 292                 mp->m_flags |= XFS_MOUNT_DIRSYNC;
 293         if (ap->flags & XFSMNT_COMPAT_ATTR)
 294                 mp->m_flags |= XFS_MOUNT_COMPAT_ATTR;
 295
 296         if (ap->flags2 & XFSMNT2_COMPAT_IOSIZE)
 297                 mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE;
 298
 299         /*
 300          * no recovery flag requires a read-only mount
 301          */
 302         if (ap->flags & XFSMNT_NORECOVERY) {
 303                 if (!(vfs->vfs_flag & VFS_RDONLY)) {
 304                         cmn_err(CE_WARN,
 305         "XFS: tried to mount a FS read-write without recovery!");
 306                         return XFS_ERROR(EINVAL);
 307                 }
 308                 mp->m_flags |= XFS_MOUNT_NORECOVERY;
 309         }
 310
 311         if (ap->flags & XFSMNT_NOUUID)
 312                 mp->m_flags |= XFS_MOUNT_NOUUID;
 313         if (ap->flags & XFSMNT_BARRIER)
 314                 mp->m_flags |= XFS_MOUNT_BARRIER;
 315         else
 316                 mp->m_flags &= ~XFS_MOUNT_BARRIER;
 317
 318         return 0;
 319 }
 320
 321 /*
 322  * This function fills in xfs_mount_t fields based on mount args.
 323  * Note: the superblock _has_ now been read in.
 324  */
 325 STATIC int
 326 xfs_finish_flags(
 327         struct vfs              *vfs,
 328         struct xfs_mount_args   *ap,
 329         struct xfs_mount        *mp)
 330 {
 331         int                     ronly = (vfs->vfs_flag & VFS_RDONLY);
 332
 333         /* Fail a mount where the logbuf is smaller then the log stripe */
 334         if (XFS_SB_VERSION_HASLOGV2(&mp->m_sb)) {
 335                 if ((ap->logbufsize == -1) &&
 336                     (mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE)) {
 337                         mp->m_logbsize = mp->m_sb.sb_logsunit;
 338                 } else if (ap->logbufsize < mp->m_sb.sb_logsunit) {
 339                         cmn_err(CE_WARN,
 340         "XFS: logbuf size must be greater than or equal to log stripe size");
 341                         return XFS_ERROR(EINVAL);
 342                 }
 343         } else {
 344                 /* Fail a mount if the logbuf is larger than 32K */
 345                 if (ap->logbufsize > XLOG_BIG_RECORD_BSIZE) {
 346                         cmn_err(CE_WARN,
 347         "XFS: logbuf size for version 1 logs must be 16K or 32K");
 348                         return XFS_ERROR(EINVAL);
 349                 }
 350         }
 351
 352         /*
 353          * prohibit r/w mounts of read-only filesystems
 354          */
 355         if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !ronly) {
 356                 cmn_err(CE_WARN,
 357         "XFS: cannot mount a read-only filesystem as read-write");
 358                 return XFS_ERROR(EROFS);
 359         }
 360
 361         /*
 362          * check for shared mount.
 363          */
 364         if (ap->flags & XFSMNT_SHARED) {
 365                 if (!XFS_SB_VERSION_HASSHARED(&mp->m_sb))
 366                         return XFS_ERROR(EINVAL);
 367
 368                 /*
 369                  * For IRIX 6.5, shared mounts must have the shared
 370                  * version bit set, have the persistent readonly
 371                  * field set, must be version 0 and can only be mounted
 372                  * read-only.
 373                  */
 374                 if (!ronly || !(mp->m_sb.sb_flags & XFS_SBF_READONLY) ||
 375                      (mp->m_sb.sb_shared_vn != 0))
 376                         return XFS_ERROR(EINVAL);
 377
 378                 mp->m_flags |= XFS_MOUNT_SHARED;
 379
 380                 /*
 381                  * Shared XFS V0 can't deal with DMI.  Return EINVAL.
 382                  */
 383                 if (mp->m_sb.sb_shared_vn == 0 && (ap->flags & XFSMNT_DMAPI))
 384                         return XFS_ERROR(EINVAL);
 385         }
 386
 387         if (XFS_SB_VERSION_HASATTR2(&mp->m_sb)) {
 388                 mp->m_flags &= ~XFS_MOUNT_COMPAT_ATTR;
 389         }
 390
 391         return 0;
 392 }
 393
 394 /*
 395  * xfs_mount
 396  *
 397  * The file system configurations are:
 398  *      (1) device (partition) with data and internal log
 399  *      (2) logical volume with data and log subvolumes.
 400  *      (3) logical volume with data, log, and realtime subvolumes.
 401  *
 402  * We only have to handle opening the log and realtime volumes here if
 403  * they are present.  The data subvolume has already been opened by
 404  * get_sb_bdev() and is stored in vfsp->vfs_super->s_bdev.
 405  */
 406 STATIC int
 407 xfs_mount(
 408         struct bhv_desc         *bhvp,
 409         struct xfs_mount_args   *args,
 410         cred_t                  *credp)
 411 {
 412         struct vfs              *vfsp = bhvtovfs(bhvp);
 413         struct bhv_desc         *p;
 414         struct xfs_mount        *mp = XFS_BHVTOM(bhvp);
 415         struct block_device     *ddev, *logdev, *rtdev;
 416         int                     flags = 0, error;
 417
 418         ddev = vfsp->vfs_super->s_bdev;
 419         logdev = rtdev = NULL;
 420
 421         /*
 422          * Setup xfs_mount function vectors from available behaviors
 423          */
 424         p = vfs_bhv_lookup(vfsp, VFS_POSITION_DM);
 425         mp->m_dm_ops = p ? *(xfs_dmops_t *) vfs_bhv_custom(p) : xfs_dmcore_stub;
 426         p = vfs_bhv_lookup(vfsp, VFS_POSITION_QM);
 427         mp->m_qm_ops = p ? *(xfs_qmops_t *) vfs_bhv_custom(p) : xfs_qmcore_stub;
 428         p = vfs_bhv_lookup(vfsp, VFS_POSITION_IO);
 429         mp->m_io_ops = p ? *(xfs_ioops_t *) vfs_bhv_custom(p) : xfs_iocore_xfs;
 430
 431         /*
 432          * Open real time and log devices - order is important.
 433          */
 434         if (args->logname[0]) {
 435                 error = xfs_blkdev_get(mp, args->logname, &logdev);
 436                 if (error)
 437                         return error;
 438         }
 439         if (args->rtname[0]) {
 440                 error = xfs_blkdev_get(mp, args->rtname, &rtdev);
 441                 if (error) {
 442                         xfs_blkdev_put(logdev);
 443                         return error;
 444                 }
 445
 446                 if (rtdev == ddev || rtdev == logdev) {
 447                         cmn_err(CE_WARN,
 448         "XFS: Cannot mount filesystem with identical rtdev and ddev/logdev.");
 449                         xfs_blkdev_put(logdev);
 450                         xfs_blkdev_put(rtdev);
 451                         return EINVAL;
 452                 }
 453         }
 454
 455         /*
 456          * Setup xfs_mount buffer target pointers
 457          */
 458         error = ENOMEM;
 459         mp->m_ddev_targp = xfs_alloc_buftarg(ddev, 0);
 460         if (!mp->m_ddev_targp) {
 461                 xfs_blkdev_put(logdev);
 462                 xfs_blkdev_put(rtdev);
 463                 return error;
 464         }
 465         if (rtdev) {
 466                 mp->m_rtdev_targp = xfs_alloc_buftarg(rtdev, 1);
 467                 if (!mp->m_rtdev_targp)
 468                         goto error0;
 469         }
 470         mp->m_logdev_targp = (logdev && logdev != ddev) ?
 471                                 xfs_alloc_buftarg(logdev, 1) : mp->m_ddev_targp;
 472         if (!mp->m_logdev_targp)
 473                 goto error0;
 474
 475         /*
 476          * Setup flags based on mount(2) options and then the superblock
 477          */
 478         error = xfs_start_flags(vfsp, args, mp);
 479         if (error)
 480                 goto error1;
 481         error = xfs_readsb(mp);
 482         if (error)
 483                 goto error1;
 484         error = xfs_finish_flags(vfsp, args, mp);
 485         if (error)
 486                 goto error2;
 487
 488         /*
 489          * Setup xfs_mount buffer target pointers based on superblock
 490          */
 491         error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_blocksize,
 492                                     mp->m_sb.sb_sectsize);
 493         if (!error && 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                 error = xfs_setsize_buftarg(mp->m_logdev_targp,
 499                                             mp->m_sb.sb_blocksize,
 500                                             log_sector_size);
 501         }
 502         if (!error && rtdev)
 503                 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
 504                                             mp->m_sb.sb_blocksize,
 505                                             mp->m_sb.sb_sectsize);
 506         if (error)
 507                 goto error2;
 508
 509         error = XFS_IOINIT(vfsp, args, flags);
 510         if (error)
 511                 goto error2;
 512
 513         if ((args->flags & XFSMNT_BARRIER) &&
 514             !(XFS_MTOVFS(mp)->vfs_flag & VFS_RDONLY))
 515                 xfs_mountfs_check_barriers(mp);
 516         return 0;
 517
 518 error2:
 519         if (mp->m_sb_bp)
 520                 xfs_freesb(mp);
 521 error1:
 522         xfs_binval(mp->m_ddev_targp);
 523         if (logdev && logdev != ddev)
 524                 xfs_binval(mp->m_logdev_targp);
 525         if (rtdev)
 526                 xfs_binval(mp->m_rtdev_targp);
 527 error0:
 528         xfs_unmountfs_close(mp, credp);
 529         return error;
 530 }
 531
 532 STATIC int
 533 xfs_unmount(
 534         bhv_desc_t      *bdp,
 535         int             flags,
 536         cred_t          *credp)
 537 {
 538         struct vfs      *vfsp = bhvtovfs(bdp);
 539         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
 540         xfs_inode_t     *rip;
 541         vnode_t         *rvp;
 542         int             unmount_event_wanted = 0;
 543         int             unmount_event_flags = 0;
 544         int             xfs_unmountfs_needed = 0;
 545         int             error;
 546
 547         rip = mp->m_rootip;
 548         rvp = XFS_ITOV(rip);
 549
 550         if (vfsp->vfs_flag & VFS_DMI) {
 551                 error = XFS_SEND_PREUNMOUNT(mp, vfsp,
 552                                 rvp, DM_RIGHT_NULL, rvp, DM_RIGHT_NULL,
 553                                 NULL, NULL, 0, 0,
 554                                 (mp->m_dmevmask & (1<<DM_EVENT_PREUNMOUNT))?
 555                                         0:DM_FLAGS_UNWANTED);
 556                         if (error)
 557                                 return XFS_ERROR(error);
 558                 unmount_event_wanted = 1;
 559                 unmount_event_flags = (mp->m_dmevmask & (1<<DM_EVENT_UNMOUNT))?
 560                                         0 : DM_FLAGS_UNWANTED;
 561         }
 562
 563         /*
 564          * First blow any referenced inode from this file system
 565          * out of the reference cache, and delete the timer.
 566          */
 567         xfs_refcache_purge_mp(mp);
 568
 569         XFS_bflush(mp->m_ddev_targp);
 570         error = xfs_unmount_flush(mp, 0);
 571         if (error)
 572                 goto out;
 573
 574         ASSERT(vn_count(rvp) == 1);
 575
 576         /*
 577          * Drop the reference count
 578          */
 579         VN_RELE(rvp);
 580
 581         /*
 582          * If we're forcing a shutdown, typically because of a media error,
 583          * we want to make sure we invalidate dirty pages that belong to
 584          * referenced vnodes as well.
 585          */
 586         if (XFS_FORCED_SHUTDOWN(mp)) {
 587                 error = xfs_sync(&mp->m_bhv,
 588                          (SYNC_WAIT | SYNC_CLOSE), credp);
 589                 ASSERT(error != EFSCORRUPTED);
 590         }
 591         xfs_unmountfs_needed = 1;
 592
 593 out:
 594         /*      Send DMAPI event, if required.
 595          *      Then do xfs_unmountfs() if needed.
 596          *      Then return error (or zero).
 597          */
 598         if (unmount_event_wanted) {
 599                 /* Note: mp structure must still exist for
 600                  * XFS_SEND_UNMOUNT() call.
 601                  */
 602                 XFS_SEND_UNMOUNT(mp, vfsp, error == 0 ? rvp : NULL,
 603                         DM_RIGHT_NULL, 0, error, unmount_event_flags);
 604         }
 605         if (xfs_unmountfs_needed) {
 606                 /*
 607                  * Call common unmount function to flush to disk
 608                  * and free the super block buffer & mount structures.
 609                  */
 610                 xfs_unmountfs(mp, credp);
 611         }
 612
 613         return XFS_ERROR(error);
 614 }
 615
 616 STATIC int
 617 xfs_quiesce_fs(
 618         xfs_mount_t             *mp)
 619 {
 620         int                     count = 0, pincount;
 621
 622         xfs_refcache_purge_mp(mp);
 623         xfs_flush_buftarg(mp->m_ddev_targp, 0);
 624         xfs_finish_reclaim_all(mp, 0);
 625
 626         /* This loop must run at least twice.
 627          * The first instance of the loop will flush
 628          * most meta data but that will generate more
 629          * meta data (typically directory updates).
 630          * Which then must be flushed and logged before
 631          * we can write the unmount record.
 632          */
 633         do {
 634                 xfs_syncsub(mp, SYNC_REMOUNT|SYNC_ATTR|SYNC_WAIT, 0, NULL);
 635                 pincount = xfs_flush_buftarg(mp->m_ddev_targp, 1);
 636                 if (!pincount) {
 637                         delay(50);
 638                         count++;
 639                 }
 640         } while (count < 2);
 641
 642         return 0;
 643 }
 644
 645 STATIC int
 646 xfs_mntupdate(
 647         bhv_desc_t                      *bdp,
 648         int                             *flags,
 649         struct xfs_mount_args           *args)
 650 {
 651         struct vfs      *vfsp = bhvtovfs(bdp);
 652         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
 653         int             error;
 654
 655         if (args->flags & XFSMNT_NOATIME)
 656                 mp->m_flags |= XFS_MOUNT_NOATIME;
 657         else
 658                 mp->m_flags &= ~XFS_MOUNT_NOATIME;
 659
 660         if (args->flags & XFSMNT_BARRIER)
 661                 mp->m_flags |= XFS_MOUNT_BARRIER;
 662         else
 663                 mp->m_flags &= ~XFS_MOUNT_BARRIER;
 664
 665         if ((vfsp->vfs_flag & VFS_RDONLY) &&
 666             !(*flags & MS_RDONLY)) {
 667                 vfsp->vfs_flag &= ~VFS_RDONLY;
 668
 669                 if (args->flags & XFSMNT_BARRIER)
 670                         xfs_mountfs_check_barriers(mp);
 671         }
 672
 673         if (!(vfsp->vfs_flag & VFS_RDONLY) &&
 674             (*flags & MS_RDONLY)) {
 675                 VFS_SYNC(vfsp, SYNC_FSDATA|SYNC_BDFLUSH|SYNC_ATTR, NULL, error);
 676
 677                 xfs_quiesce_fs(mp);
 678
 679                 /* Ok now write out an unmount record */
 680                 xfs_log_unmount_write(mp);
 681                 xfs_unmountfs_writesb(mp);
 682                 vfsp->vfs_flag |= VFS_RDONLY;
 683         }
 684
 685         return 0;
 686 }
 687
 688 /*
 689  * xfs_unmount_flush implements a set of flush operation on special
 690  * inodes, which are needed as a separate set of operations so that
 691  * they can be called as part of relocation process.
 692  */
 693 int
 694 xfs_unmount_flush(
 695         xfs_mount_t     *mp,            /* Mount structure we are getting
 696                                            rid of. */
 697         int             relocation)     /* Called from vfs relocation. */
 698 {
 699         xfs_inode_t     *rip = mp->m_rootip;
 700         xfs_inode_t     *rbmip;
 701         xfs_inode_t     *rsumip = NULL;
 702         vnode_t         *rvp = XFS_ITOV(rip);
 703         int             error;
 704
 705         xfs_ilock(rip, XFS_ILOCK_EXCL);
 706         xfs_iflock(rip);
 707
 708         /*
 709          * Flush out the real time inodes.
 710          */
 711         if ((rbmip = mp->m_rbmip) != NULL) {
 712                 xfs_ilock(rbmip, XFS_ILOCK_EXCL);
 713                 xfs_iflock(rbmip);
 714                 error = xfs_iflush(rbmip, XFS_IFLUSH_SYNC);
 715                 xfs_iunlock(rbmip, XFS_ILOCK_EXCL);
 716
 717                 if (error == EFSCORRUPTED)
 718                         goto fscorrupt_out;
 719
 720                 ASSERT(vn_count(XFS_ITOV(rbmip)) == 1);
 721
 722                 rsumip = mp->m_rsumip;
 723                 xfs_ilock(rsumip, XFS_ILOCK_EXCL);
 724                 xfs_iflock(rsumip);
 725                 error = xfs_iflush(rsumip, XFS_IFLUSH_SYNC);
 726                 xfs_iunlock(rsumip, XFS_ILOCK_EXCL);
 727
 728                 if (error == EFSCORRUPTED)
 729                         goto fscorrupt_out;
 730
 731                 ASSERT(vn_count(XFS_ITOV(rsumip)) == 1);
 732         }
 733
 734         /*
 735          * Synchronously flush root inode to disk
 736          */
 737         error = xfs_iflush(rip, XFS_IFLUSH_SYNC);
 738         if (error == EFSCORRUPTED)
 739                 goto fscorrupt_out2;
 740
 741         if (vn_count(rvp) != 1 && !relocation) {
 742                 xfs_iunlock(rip, XFS_ILOCK_EXCL);
 743                 return XFS_ERROR(EBUSY);
 744         }
 745
 746         /*
 747          * Release dquot that rootinode, rbmino and rsumino might be holding,
 748          * flush and purge the quota inodes.
 749          */
 750         error = XFS_QM_UNMOUNT(mp);
 751         if (error == EFSCORRUPTED)
 752                 goto fscorrupt_out2;
 753
 754         if (rbmip) {
 755                 VN_RELE(XFS_ITOV(rbmip));
 756                 VN_RELE(XFS_ITOV(rsumip));
 757         }
 758
 759         xfs_iunlock(rip, XFS_ILOCK_EXCL);
 760         return 0;
 761
 762 fscorrupt_out:
 763         xfs_ifunlock(rip);
 764
 765 fscorrupt_out2:
 766         xfs_iunlock(rip, XFS_ILOCK_EXCL);
 767
 768         return XFS_ERROR(EFSCORRUPTED);
 769 }
 770
 771 /*
 772  * xfs_root extracts the root vnode from a vfs.
 773  *
 774  * vfsp -- the vfs struct for the desired file system
 775  * vpp  -- address of the caller's vnode pointer which should be
 776  *         set to the desired fs root vnode
 777  */
 778 STATIC int
 779 xfs_root(
 780         bhv_desc_t      *bdp,
 781         vnode_t         **vpp)
 782 {
 783         vnode_t         *vp;
 784
 785         vp = XFS_ITOV((XFS_BHVTOM(bdp))->m_rootip);
 786         VN_HOLD(vp);
 787         *vpp = vp;
 788         return 0;
 789 }
 790
 791 /*
 792  * xfs_statvfs
 793  *
 794  * Fill in the statvfs structure for the given file system.  We use
 795  * the superblock lock in the mount structure to ensure a consistent
 796  * snapshot of the counters returned.
 797  */
 798 STATIC int
 799 xfs_statvfs(
 800         bhv_desc_t      *bdp,
 801         xfs_statfs_t    *statp,
 802         vnode_t         *vp)
 803 {
 804         __uint64_t      fakeinos;
 805         xfs_extlen_t    lsize;
 806         xfs_mount_t     *mp;
 807         xfs_sb_t        *sbp;
 808         unsigned long   s;
 809
 810         mp = XFS_BHVTOM(bdp);
 811         sbp = &(mp->m_sb);
 812
 813         statp->f_type = XFS_SB_MAGIC;
 814
 815         s = XFS_SB_LOCK(mp);
 816         statp->f_bsize = sbp->sb_blocksize;
 817         lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
 818         statp->f_blocks = sbp->sb_dblocks - lsize;
 819         statp->f_bfree = statp->f_bavail = sbp->sb_fdblocks;
 820         fakeinos = statp->f_bfree << sbp->sb_inopblog;
 821 #if XFS_BIG_INUMS
 822         fakeinos += mp->m_inoadd;
 823 #endif
 824         statp->f_files =
 825             MIN(sbp->sb_icount + fakeinos, (__uint64_t)XFS_MAXINUMBER);
 826         if (mp->m_maxicount)
 827 #if XFS_BIG_INUMS
 828                 if (!mp->m_inoadd)
 829 #endif
 830                         statp->f_files = min_t(typeof(statp->f_files),
 831                                                 statp->f_files,
 832                                                 mp->m_maxicount);
 833         statp->f_ffree = statp->f_files - (sbp->sb_icount - sbp->sb_ifree);
 834         XFS_SB_UNLOCK(mp, s);
 835
 836         xfs_statvfs_fsid(statp, mp);
 837         statp->f_namelen = MAXNAMELEN - 1;
 838
 839         return 0;
 840 }
 841
 842
 843 /*
 844  * xfs_sync flushes any pending I/O to file system vfsp.
 845  *
 846  * This routine is called by vfs_sync() to make sure that things make it
 847  * out to disk eventually, on sync() system calls to flush out everything,
 848  * and when the file system is unmounted.  For the vfs_sync() case, all
 849  * we really need to do is sync out the log to make all of our meta-data
 850  * updates permanent (except for timestamps).  For calls from pflushd(),
 851  * dirty pages are kept moving by calling pdflush() on the inodes
 852  * containing them.  We also flush the inodes that we can lock without
 853  * sleeping and the superblock if we can lock it without sleeping from
 854  * vfs_sync() so that items at the tail of the log are always moving out.
 855  *
 856  * Flags:
 857  *      SYNC_BDFLUSH - We're being called from vfs_sync() so we don't want
 858  *                     to sleep if we can help it.  All we really need
 859  *                     to do is ensure that the log is synced at least
 860  *                     periodically.  We also push the inodes and
 861  *                     superblock if we can lock them without sleeping
 862  *                      and they are not pinned.
 863  *      SYNC_ATTR    - We need to flush the inodes.  If SYNC_BDFLUSH is not
 864  *                     set, then we really want to lock each inode and flush
 865  *                     it.
 866  *      SYNC_WAIT    - All the flushes that take place in this call should
 867  *                     be synchronous.
 868  *      SYNC_DELWRI  - This tells us to push dirty pages associated with
 869  *                     inodes.  SYNC_WAIT and SYNC_BDFLUSH are used to
 870  *                     determine if they should be flushed sync, async, or
 871  *                     delwri.
 872  *      SYNC_CLOSE   - This flag is passed when the system is being
 873  *                     unmounted.  We should sync and invalidate everthing.
 874  *      SYNC_FSDATA  - This indicates that the caller would like to make
 875  *                     sure the superblock is safe on disk.  We can ensure
 876  *                     this by simply makeing sure the log gets flushed
 877  *                     if SYNC_BDFLUSH is set, and by actually writing it
 878  *                     out otherwise.
 879  *
 880  */
 881 /*ARGSUSED*/
 882 STATIC int
 883 xfs_sync(
 884         bhv_desc_t      *bdp,
 885         int             flags,
 886         cred_t          *credp)
 887 {
 888         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
 889
 890         if (unlikely(flags == SYNC_QUIESCE))
 891                 return xfs_quiesce_fs(mp);
 892         else
 893                 return xfs_syncsub(mp, flags, 0, NULL);
 894 }
 895
 896 /*
 897  * xfs sync routine for internal use
 898  *
 899  * This routine supports all of the flags defined for the generic VFS_SYNC
 900  * interface as explained above under xfs_sync.  In the interests of not
 901  * changing interfaces within the 6.5 family, additional internallly-
 902  * required functions are specified within a separate xflags parameter,
 903  * only available by calling this routine.
 904  *
 905  */
 906 int
 907 xfs_sync_inodes(
 908         xfs_mount_t     *mp,
 909         int             flags,
 910         int             xflags,
 911         int             *bypassed)
 912 {
 913         xfs_inode_t     *ip = NULL;
 914         xfs_inode_t     *ip_next;
 915         xfs_buf_t       *bp;
 916         vnode_t         *vp = NULL;
 917         int             error;
 918         int             last_error;
 919         uint64_t        fflag;
 920         uint            lock_flags;
 921         uint            base_lock_flags;
 922         boolean_t       mount_locked;
 923         boolean_t       vnode_refed;
 924         int             preempt;
 925         xfs_dinode_t    *dip;
 926         xfs_iptr_t      *ipointer;
 927 #ifdef DEBUG
 928         boolean_t       ipointer_in = B_FALSE;
 929
 930 #define IPOINTER_SET    ipointer_in = B_TRUE
 931 #define IPOINTER_CLR    ipointer_in = B_FALSE
 932 #else
 933 #define IPOINTER_SET
 934 #define IPOINTER_CLR
 935 #endif
 936
 937
 938 /* Insert a marker record into the inode list after inode ip. The list
 939  * must be locked when this is called. After the call the list will no
 940  * longer be locked.
 941  */
 942 #define IPOINTER_INSERT(ip, mp) { \
 943                 ASSERT(ipointer_in == B_FALSE); \
 944                 ipointer->ip_mnext = ip->i_mnext; \
 945                 ipointer->ip_mprev = ip; \
 946                 ip->i_mnext = (xfs_inode_t *)ipointer; \
 947                 ipointer->ip_mnext->i_mprev = (xfs_inode_t *)ipointer; \
 948                 preempt = 0; \
 949                 XFS_MOUNT_IUNLOCK(mp); \
 950                 mount_locked = B_FALSE; \
 951                 IPOINTER_SET; \
 952         }
 953
 954 /* Remove the marker from the inode list. If the marker was the only item
 955  * in the list then there are no remaining inodes and we should zero out
 956  * the whole list. If we are the current head of the list then move the head
 957  * past us.
 958  */
 959 #define IPOINTER_REMOVE(ip, mp) { \
 960                 ASSERT(ipointer_in == B_TRUE); \
 961                 if (ipointer->ip_mnext != (xfs_inode_t *)ipointer) { \
 962                         ip = ipointer->ip_mnext; \
 963                         ip->i_mprev = ipointer->ip_mprev; \
 964                         ipointer->ip_mprev->i_mnext = ip; \
 965                         if (mp->m_inodes == (xfs_inode_t *)ipointer) { \
 966                                 mp->m_inodes = ip; \
 967                         } \
 968                 } else { \
 969                         ASSERT(mp->m_inodes == (xfs_inode_t *)ipointer); \
 970                         mp->m_inodes = NULL; \
 971                         ip = NULL; \
 972                 } \
 973                 IPOINTER_CLR; \
 974         }
 975
 976 #define XFS_PREEMPT_MASK        0x7f
 977
 978         if (bypassed)
 979                 *bypassed = 0;
 980         if (XFS_MTOVFS(mp)->vfs_flag & VFS_RDONLY)
 981                 return 0;
 982         error = 0;
 983         last_error = 0;
 984         preempt = 0;
 985
 986         /* Allocate a reference marker */
 987         ipointer = (xfs_iptr_t *)kmem_zalloc(sizeof(xfs_iptr_t), KM_SLEEP);
 988
 989         fflag = XFS_B_ASYNC;            /* default is don't wait */
 990         if (flags & (SYNC_BDFLUSH | SYNC_DELWRI))
 991                 fflag = XFS_B_DELWRI;
 992         if (flags & SYNC_WAIT)
 993                 fflag = 0;              /* synchronous overrides all */
 994
 995         base_lock_flags = XFS_ILOCK_SHARED;
 996         if (flags & (SYNC_DELWRI | SYNC_CLOSE)) {
 997                 /*
 998                  * We need the I/O lock if we're going to call any of
 999                  * the flush/inval routines.
1000                  */
1001                 base_lock_flags |= XFS_IOLOCK_SHARED;
1002         }
1003
1004         XFS_MOUNT_ILOCK(mp);
1005
1006         ip = mp->m_inodes;
1007
1008         mount_locked = B_TRUE;
1009         vnode_refed  = B_FALSE;
1010
1011         IPOINTER_CLR;
1012
1013         do {
1014                 ASSERT(ipointer_in == B_FALSE);
1015                 ASSERT(vnode_refed == B_FALSE);
1016
1017                 lock_flags = base_lock_flags;
1018
1019                 /*
1020                  * There were no inodes in the list, just break out
1021                  * of the loop.
1022                  */
1023                 if (ip == NULL) {
1024                         break;
1025                 }
1026
1027                 /*
1028                  * We found another sync thread marker - skip it
1029                  */
1030                 if (ip->i_mount == NULL) {
1031                         ip = ip->i_mnext;
1032                         continue;
1033                 }
1034
1035                 vp = XFS_ITOV_NULL(ip);
1036
1037                 /*
1038                  * If the vnode is gone then this is being torn down,
1039                  * call reclaim if it is flushed, else let regular flush
1040                  * code deal with it later in the loop.
1041                  */
1042
1043                 if (vp == NULL) {
1044                         /* Skip ones already in reclaim */
1045                         if (ip->i_flags & XFS_IRECLAIM) {
1046                                 ip = ip->i_mnext;
1047                                 continue;
1048                         }
1049                         if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0) {
1050                                 ip = ip->i_mnext;
1051                         } else if ((xfs_ipincount(ip) == 0) &&
1052                                     xfs_iflock_nowait(ip)) {
1053                                 IPOINTER_INSERT(ip, mp);
1054
1055                                 xfs_finish_reclaim(ip, 1,
1056                                                 XFS_IFLUSH_DELWRI_ELSE_ASYNC);
1057
1058                                 XFS_MOUNT_ILOCK(mp);
1059                                 mount_locked = B_TRUE;
1060                                 IPOINTER_REMOVE(ip, mp);
1061                         } else {
1062                                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1063                                 ip = ip->i_mnext;
1064                         }
1065                         continue;
1066                 }
1067
1068                 if (VN_BAD(vp)) {
1069                         ip = ip->i_mnext;
1070                         continue;
1071                 }
1072
1073                 if (XFS_FORCED_SHUTDOWN(mp) && !(flags & SYNC_CLOSE)) {
1074                         XFS_MOUNT_IUNLOCK(mp);
1075                         kmem_free(ipointer, sizeof(xfs_iptr_t));
1076                         return 0;
1077                 }
1078
1079                 /*
1080                  * If this is just vfs_sync() or pflushd() calling
1081                  * then we can skip inodes for which it looks like
1082                  * there is nothing to do.  Since we don't have the
1083                  * inode locked this is racey, but these are periodic
1084                  * calls so it doesn't matter.  For the others we want
1085                  * to know for sure, so we at least try to lock them.
1086                  */
1087                 if (flags & SYNC_BDFLUSH) {
1088                         if (((ip->i_itemp == NULL) ||
1089                              !(ip->i_itemp->ili_format.ilf_fields &
1090                                XFS_ILOG_ALL)) &&
1091                             (ip->i_update_core == 0)) {
1092                                 ip = ip->i_mnext;
1093                                 continue;
1094                         }
1095                 }
1096
1097                 /*
1098                  * Try to lock without sleeping.  We're out of order with
1099                  * the inode list lock here, so if we fail we need to drop
1100                  * the mount lock and try again.  If we're called from
1101                  * bdflush() here, then don't bother.
1102                  *
1103                  * The inode lock here actually coordinates with the
1104                  * almost spurious inode lock in xfs_ireclaim() to prevent
1105                  * the vnode we handle here without a reference from
1106                  * being freed while we reference it.  If we lock the inode
1107                  * while it's on the mount list here, then the spurious inode
1108                  * lock in xfs_ireclaim() after the inode is pulled from
1109                  * the mount list will sleep until we release it here.
1110                  * This keeps the vnode from being freed while we reference
1111                  * it.
1112                  */
1113                 if (xfs_ilock_nowait(ip, lock_flags) == 0) {
1114                         if ((flags & SYNC_BDFLUSH) || (vp == NULL)) {
1115                                 ip = ip->i_mnext;
1116                                 continue;
1117                         }
1118
1119                         vp = vn_grab(vp);
1120                         if (vp == NULL) {
1121                                 ip = ip->i_mnext;
1122                                 continue;
1123                         }
1124
1125                         IPOINTER_INSERT(ip, mp);
1126                         xfs_ilock(ip, lock_flags);
1127
1128                         ASSERT(vp == XFS_ITOV(ip));
1129                         ASSERT(ip->i_mount == mp);
1130
1131                         vnode_refed = B_TRUE;
1132                 }
1133
1134                 /* From here on in the loop we may have a marker record
1135                  * in the inode list.
1136                  */
1137
1138                 if ((flags & SYNC_CLOSE)  && (vp != NULL)) {
1139                         /*
1140                          * This is the shutdown case.  We just need to
1141                          * flush and invalidate all the pages associated
1142                          * with the inode.  Drop the inode lock since
1143                          * we can't hold it across calls to the buffer
1144                          * cache.
1145                          *
1146                          * We don't set the VREMAPPING bit in the vnode
1147                          * here, because we don't hold the vnode lock
1148                          * exclusively.  It doesn't really matter, though,
1149                          * because we only come here when we're shutting
1150                          * down anyway.
1151                          */
1152                         xfs_iunlock(ip, XFS_ILOCK_SHARED);
1153
1154                         if (XFS_FORCED_SHUTDOWN(mp)) {
1155                                 VOP_TOSS_PAGES(vp, 0, -1, FI_REMAPF);
1156                         } else {
1157                                 VOP_FLUSHINVAL_PAGES(vp, 0, -1, FI_REMAPF);
1158                         }
1159
1160                         xfs_ilock(ip, XFS_ILOCK_SHARED);
1161
1162                 } else if ((flags & SYNC_DELWRI) && (vp != NULL)) {
1163                         if (VN_DIRTY(vp)) {
1164                                 /* We need to have dropped the lock here,
1165                                  * so insert a marker if we have not already
1166                                  * done so.
1167                                  */
1168                                 if (mount_locked) {
1169                                         IPOINTER_INSERT(ip, mp);
1170                                 }
1171
1172                                 /*
1173                                  * Drop the inode lock since we can't hold it
1174                                  * across calls to the buffer cache.
1175                                  */
1176                                 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1177                                 VOP_FLUSH_PAGES(vp, (xfs_off_t)0, -1,
1178                                                         fflag, FI_NONE, error);
1179                                 xfs_ilock(ip, XFS_ILOCK_SHARED);
1180                         }
1181
1182                 }
1183
1184                 if (flags & SYNC_BDFLUSH) {
1185                         if ((flags & SYNC_ATTR) &&
1186                             ((ip->i_update_core) ||
1187                              ((ip->i_itemp != NULL) &&
1188                               (ip->i_itemp->ili_format.ilf_fields != 0)))) {
1189
1190                                 /* Insert marker and drop lock if not already
1191                                  * done.
1192                                  */
1193                                 if (mount_locked) {
1194                                         IPOINTER_INSERT(ip, mp);
1195                                 }
1196
1197                                 /*
1198                                  * We don't want the periodic flushing of the
1199                                  * inodes by vfs_sync() to interfere with
1200                                  * I/O to the file, especially read I/O
1201                                  * where it is only the access time stamp
1202                                  * that is being flushed out.  To prevent
1203                                  * long periods where we have both inode
1204                                  * locks held shared here while reading the
1205                                  * inode's buffer in from disk, we drop the
1206                                  * inode lock while reading in the inode
1207                                  * buffer.  We have to release the buffer
1208                                  * and reacquire the inode lock so that they
1209                                  * are acquired in the proper order (inode
1210                                  * locks first).  The buffer will go at the
1211                                  * end of the lru chain, though, so we can
1212                                  * expect it to still be there when we go
1213                                  * for it again in xfs_iflush().
1214                                  */
1215                                 if ((xfs_ipincount(ip) == 0) &&
1216                                     xfs_iflock_nowait(ip)) {
1217
1218                                         xfs_ifunlock(ip);
1219                                         xfs_iunlock(ip, XFS_ILOCK_SHARED);
1220
1221                                         error = xfs_itobp(mp, NULL, ip,
1222                                                           &dip, &bp, 0);
1223                                         if (!error) {
1224                                                 xfs_buf_relse(bp);
1225                                         } else {
1226                                                 /* Bailing out, remove the
1227                                                  * marker and free it.
1228                                                  */
1229                                                 XFS_MOUNT_ILOCK(mp);
1230
1231                                                 IPOINTER_REMOVE(ip, mp);
1232
1233                                                 XFS_MOUNT_IUNLOCK(mp);
1234
1235                                                 ASSERT(!(lock_flags &
1236                                                         XFS_IOLOCK_SHARED));
1237
1238                                                 kmem_free(ipointer,
1239                                                         sizeof(xfs_iptr_t));
1240                                                 return (0);
1241                                         }
1242
1243                                         /*
1244                                          * Since we dropped the inode lock,
1245                                          * the inode may have been reclaimed.
1246                                          * Therefore, we reacquire the mount
1247                                          * lock and check to see if we were the
1248                                          * inode reclaimed. If this happened
1249                                          * then the ipointer marker will no
1250                                          * longer point back at us. In this
1251                                          * case, move ip along to the inode
1252                                          * after the marker, remove the marker
1253                                          * and continue.
1254                                          */
1255                                         XFS_MOUNT_ILOCK(mp);
1256                                         mount_locked = B_TRUE;
1257
1258                                         if (ip != ipointer->ip_mprev) {
1259                                                 IPOINTER_REMOVE(ip, mp);
1260
1261                                                 ASSERT(!vnode_refed);
1262                                                 ASSERT(!(lock_flags &
1263                                                         XFS_IOLOCK_SHARED));
1264                                                 continue;
1265                                         }
1266
1267                                         ASSERT(ip->i_mount == mp);
1268
1269                                         if (xfs_ilock_nowait(ip,
1270                                                     XFS_ILOCK_SHARED) == 0) {
1271                                                 ASSERT(ip->i_mount == mp);
1272                                                 /*
1273                                                  * We failed to reacquire
1274                                                  * the inode lock without
1275                                                  * sleeping, so just skip
1276                                                  * the inode for now.  We
1277                                                  * clear the ILOCK bit from
1278                                                  * the lock_flags so that we
1279                                                  * won't try to drop a lock
1280                                                  * we don't hold below.
1281                                                  */
1282                                                 lock_flags &= ~XFS_ILOCK_SHARED;
1283                                                 IPOINTER_REMOVE(ip_next, mp);
1284                                         } else if ((xfs_ipincount(ip) == 0) &&
1285                                                    xfs_iflock_nowait(ip)) {
1286                                                 ASSERT(ip->i_mount == mp);
1287                                                 /*
1288                                                  * Since this is vfs_sync()
1289                                                  * calling we only flush the
1290                                                  * inode out if we can lock
1291                                                  * it without sleeping and
1292                                                  * it is not pinned.  Drop
1293                                                  * the mount lock here so
1294                                                  * that we don't hold it for
1295                                                  * too long. We already have
1296                                                  * a marker in the list here.
1297                                                  */
1298                                                 XFS_MOUNT_IUNLOCK(mp);
1299                                                 mount_locked = B_FALSE;
1300                                                 error = xfs_iflush(ip,
1301                                                            XFS_IFLUSH_DELWRI);
1302                                         } else {
1303                                                 ASSERT(ip->i_mount == mp);
1304                                                 IPOINTER_REMOVE(ip_next, mp);
1305                                         }
1306                                 }
1307
1308                         }
1309
1310                 } else {
1311                         if ((flags & SYNC_ATTR) &&
1312                             ((ip->i_update_core) ||
1313                              ((ip->i_itemp != NULL) &&
1314                               (ip->i_itemp->ili_format.ilf_fields != 0)))) {
1315                                 if (mount_locked) {
1316                                         IPOINTER_INSERT(ip, mp);
1317                                 }
1318
1319                                 if (flags & SYNC_WAIT) {
1320                                         xfs_iflock(ip);
1321                                         error = xfs_iflush(ip,
1322                                                            XFS_IFLUSH_SYNC);
1323                                 } else {
1324                                         /*
1325                                          * If we can't acquire the flush
1326                                          * lock, then the inode is already
1327                                          * being flushed so don't bother
1328                                          * waiting.  If we can lock it then
1329                                          * do a delwri flush so we can
1330                                          * combine multiple inode flushes
1331                                          * in each disk write.
1332                                          */
1333                                         if (xfs_iflock_nowait(ip)) {
1334                                                 error = xfs_iflush(ip,
1335                                                            XFS_IFLUSH_DELWRI);
1336                                         }
1337                                         else if (bypassed)
1338                                                 (*bypassed)++;
1339                                 }
1340                         }
1341                 }
1342
1343                 if (lock_flags != 0) {
1344                         xfs_iunlock(ip, lock_flags);
1345                 }
1346
1347                 if (vnode_refed) {
1348                         /*
1349                          * If we had to take a reference on the vnode
1350                          * above, then wait until after we've unlocked
1351                          * the inode to release the reference.  This is
1352                          * because we can be already holding the inode
1353                          * lock when VN_RELE() calls xfs_inactive().
1354                          *
1355                          * Make sure to drop the mount lock before calling
1356                          * VN_RELE() so that we don't trip over ourselves if
1357                          * we have to go for the mount lock again in the
1358                          * inactive code.
1359                          */
1360                         if (mount_locked) {
1361                                 IPOINTER_INSERT(ip, mp);
1362                         }
1363
1364                         VN_RELE(vp);
1365
1366                         vnode_refed = B_FALSE;
1367                 }
1368
1369                 if (error) {
1370                         last_error = error;
1371                 }
1372
1373                 /*
1374                  * bail out if the filesystem is corrupted.
1375                  */
1376                 if (error == EFSCORRUPTED)  {
1377                         if (!mount_locked) {
1378                                 XFS_MOUNT_ILOCK(mp);
1379                                 IPOINTER_REMOVE(ip, mp);
1380                         }
1381                         XFS_MOUNT_IUNLOCK(mp);
1382                         ASSERT(ipointer_in == B_FALSE);
1383                         kmem_free(ipointer, sizeof(xfs_iptr_t));
1384                         return XFS_ERROR(error);
1385                 }
1386
1387                 /* Let other threads have a chance at the mount lock
1388                  * if we have looped many times without dropping the
1389                  * lock.
1390                  */
1391                 if ((++preempt & XFS_PREEMPT_MASK) == 0) {
1392                         if (mount_locked) {
1393                                 IPOINTER_INSERT(ip, mp);
1394                         }
1395                 }
1396
1397                 if (mount_locked == B_FALSE) {
1398                         XFS_MOUNT_ILOCK(mp);
1399                         mount_locked = B_TRUE;
1400                         IPOINTER_REMOVE(ip, mp);
1401                         continue;
1402                 }
1403
1404                 ASSERT(ipointer_in == B_FALSE);
1405                 ip = ip->i_mnext;
1406
1407         } while (ip != mp->m_inodes);
1408
1409         XFS_MOUNT_IUNLOCK(mp);
1410
1411         ASSERT(ipointer_in == B_FALSE);
1412
1413         kmem_free(ipointer, sizeof(xfs_iptr_t));
1414         return XFS_ERROR(last_error);
1415 }
1416
1417 /*
1418  * xfs sync routine for internal use
1419  *
1420  * This routine supports all of the flags defined for the generic VFS_SYNC
1421  * interface as explained above under xfs_sync.  In the interests of not
1422  * changing interfaces within the 6.5 family, additional internallly-
1423  * required functions are specified within a separate xflags parameter,
1424  * only available by calling this routine.
1425  *
1426  */
1427 int
1428 xfs_syncsub(
1429         xfs_mount_t     *mp,
1430         int             flags,
1431         int             xflags,
1432         int             *bypassed)
1433 {
1434         int             error = 0;
1435         int             last_error = 0;
1436         uint            log_flags = XFS_LOG_FORCE;
1437         xfs_buf_t       *bp;
1438         xfs_buf_log_item_t      *bip;
1439
1440         /*
1441          * Sync out the log.  This ensures that the log is periodically
1442          * flushed even if there is not enough activity to fill it up.
1443          */
1444         if (flags & SYNC_WAIT)
1445                 log_flags |= XFS_LOG_SYNC;
1446
1447         xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1448
1449         if (flags & (SYNC_ATTR|SYNC_DELWRI)) {
1450                 if (flags & SYNC_BDFLUSH)
1451                         xfs_finish_reclaim_all(mp, 1);
1452                 else
1453                         error = xfs_sync_inodes(mp, flags, xflags, bypassed);
1454         }
1455
1456         /*
1457          * Flushing out dirty data above probably generated more
1458          * log activity, so if this isn't vfs_sync() then flush
1459          * the log again.
1460          */
1461         if (flags & SYNC_DELWRI) {
1462                 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1463         }
1464
1465         if (flags & SYNC_FSDATA) {
1466                 /*
1467                  * If this is vfs_sync() then only sync the superblock
1468                  * if we can lock it without sleeping and it is not pinned.
1469                  */
1470                 if (flags & SYNC_BDFLUSH) {
1471                         bp = xfs_getsb(mp, XFS_BUF_TRYLOCK);
1472                         if (bp != NULL) {
1473                                 bip = XFS_BUF_FSPRIVATE(bp,xfs_buf_log_item_t*);
1474                                 if ((bip != NULL) &&
1475                                     xfs_buf_item_dirty(bip)) {
1476                                         if (!(XFS_BUF_ISPINNED(bp))) {
1477                                                 XFS_BUF_ASYNC(bp);
1478                                                 error = xfs_bwrite(mp, bp);
1479                                         } else {
1480                                                 xfs_buf_relse(bp);
1481                                         }
1482                                 } else {
1483                                         xfs_buf_relse(bp);
1484                                 }
1485                         }
1486                 } else {
1487                         bp = xfs_getsb(mp, 0);
1488                         /*
1489                          * If the buffer is pinned then push on the log so
1490                          * we won't get stuck waiting in the write for
1491                          * someone, maybe ourselves, to flush the log.
1492                          * Even though we just pushed the log above, we
1493                          * did not have the superblock buffer locked at
1494                          * that point so it can become pinned in between
1495                          * there and here.
1496                          */
1497                         if (XFS_BUF_ISPINNED(bp))
1498                                 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
1499                         if (flags & SYNC_WAIT)
1500                                 XFS_BUF_UNASYNC(bp);
1501                         else
1502                                 XFS_BUF_ASYNC(bp);
1503                         error = xfs_bwrite(mp, bp);
1504                 }
1505                 if (error) {
1506                         last_error = error;
1507                 }
1508         }
1509
1510         /*
1511          * If this is the periodic sync, then kick some entries out of
1512          * the reference cache.  This ensures that idle entries are
1513          * eventually kicked out of the cache.
1514          */
1515         if (flags & SYNC_REFCACHE) {
1516                 if (flags & SYNC_WAIT)
1517                         xfs_refcache_purge_mp(mp);
1518                 else
1519                         xfs_refcache_purge_some(mp);
1520         }
1521
1522         /*
1523          * Now check to see if the log needs a "dummy" transaction.
1524          */
1525
1526         if (!(flags & SYNC_REMOUNT) && xfs_log_need_covered(mp)) {
1527                 xfs_trans_t *tp;
1528                 xfs_inode_t *ip;
1529
1530                 /*
1531                  * Put a dummy transaction in the log to tell
1532                  * recovery that all others are OK.
1533                  */
1534                 tp = xfs_trans_alloc(mp, XFS_TRANS_DUMMY1);
1535                 if ((error = xfs_trans_reserve(tp, 0,
1536                                 XFS_ICHANGE_LOG_RES(mp),
1537                                 0, 0, 0)))  {
1538                         xfs_trans_cancel(tp, 0);
1539                         return error;
1540                 }
1541
1542                 ip = mp->m_rootip;
1543                 xfs_ilock(ip, XFS_ILOCK_EXCL);
1544
1545                 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1546                 xfs_trans_ihold(tp, ip);
1547                 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1548                 error = xfs_trans_commit(tp, 0, NULL);
1549                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1550                 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1551         }
1552
1553         /*
1554          * When shutting down, we need to insure that the AIL is pushed
1555          * to disk or the filesystem can appear corrupt from the PROM.
1556          */
1557         if ((flags & (SYNC_CLOSE|SYNC_WAIT)) == (SYNC_CLOSE|SYNC_WAIT)) {
1558                 XFS_bflush(mp->m_ddev_targp);
1559                 if (mp->m_rtdev_targp) {
1560                         XFS_bflush(mp->m_rtdev_targp);
1561                 }
1562         }
1563
1564         return XFS_ERROR(last_error);
1565 }
1566
1567 /*
1568  * xfs_vget - called by DMAPI and NFSD to get vnode from file handle
1569  */
1570 STATIC int
1571 xfs_vget(
1572         bhv_desc_t      *bdp,
1573         vnode_t         **vpp,
1574         fid_t           *fidp)
1575 {
1576         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
1577         xfs_fid_t       *xfid = (struct xfs_fid *)fidp;
1578         xfs_inode_t     *ip;
1579         int             error;
1580         xfs_ino_t       ino;
1581         unsigned int    igen;
1582
1583         /*
1584          * Invalid.  Since handles can be created in user space and passed in
1585          * via gethandle(), this is not cause for a panic.
1586          */
1587         if (xfid->xfs_fid_len != sizeof(*xfid) - sizeof(xfid->xfs_fid_len))
1588                 return XFS_ERROR(EINVAL);
1589
1590         ino  = xfid->xfs_fid_ino;
1591         igen = xfid->xfs_fid_gen;
1592
1593         /*
1594          * NFS can sometimes send requests for ino 0.  Fail them gracefully.
1595          */
1596         if (ino == 0)
1597                 return XFS_ERROR(ESTALE);
1598
1599         error = xfs_iget(mp, NULL, ino, 0, XFS_ILOCK_SHARED, &ip, 0);
1600         if (error) {
1601                 *vpp = NULL;
1602                 return error;
1603         }
1604
1605         if (ip == NULL) {
1606                 *vpp = NULL;
1607                 return XFS_ERROR(EIO);
1608         }
1609
1610         if (ip->i_d.di_mode == 0 || ip->i_d.di_gen != igen) {
1611                 xfs_iput_new(ip, XFS_ILOCK_SHARED);
1612                 *vpp = NULL;
1613                 return XFS_ERROR(ENOENT);
1614         }
1615
1616         *vpp = XFS_ITOV(ip);
1617         xfs_iunlock(ip, XFS_ILOCK_SHARED);
1618         return 0;
1619 }
1620
1621
1622 #define MNTOPT_LOGBUFS  "logbufs"       /* number of XFS log buffers */
1623 #define MNTOPT_LOGBSIZE "logbsize"      /* size of XFS log buffers */
1624 #define MNTOPT_LOGDEV   "logdev"        /* log device */
1625 #define MNTOPT_RTDEV    "rtdev"         /* realtime I/O device */
1626 #define MNTOPT_BIOSIZE  "biosize"       /* log2 of preferred buffered io size */
1627 #define MNTOPT_WSYNC    "wsync"         /* safe-mode nfs compatible mount */
1628 #define MNTOPT_INO64    "ino64"         /* force inodes into 64-bit range */
1629 #define MNTOPT_NOALIGN  "noalign"       /* turn off stripe alignment */
1630 #define MNTOPT_SWALLOC  "swalloc"       /* turn on stripe width allocation */
1631 #define MNTOPT_SUNIT    "sunit"         /* data volume stripe unit */
1632 #define MNTOPT_SWIDTH   "swidth"        /* data volume stripe width */
1633 #define MNTOPT_NOUUID   "nouuid"        /* ignore filesystem UUID */
1634 #define MNTOPT_MTPT     "mtpt"          /* filesystem mount point */
1635 #define MNTOPT_GRPID    "grpid"         /* group-ID from parent directory */
1636 #define MNTOPT_NOGRPID  "nogrpid"       /* group-ID from current process */
1637 #define MNTOPT_BSDGROUPS    "bsdgroups"    /* group-ID from parent directory */
1638 #define MNTOPT_SYSVGROUPS   "sysvgroups"   /* group-ID from current process */
1639 #define MNTOPT_ALLOCSIZE    "allocsize"    /* preferred allocation size */
1640 #define MNTOPT_IHASHSIZE    "ihashsize"    /* size of inode hash table */
1641 #define MNTOPT_NORECOVERY   "norecovery"   /* don't run XFS recovery */
1642 #define MNTOPT_BARRIER  "barrier"       /* use writer barriers for log write and
1643                                          * unwritten extent conversion */
1644 #define MNTOPT_NOBARRIER "nobarrier"    /* .. disable */
1645 #define MNTOPT_OSYNCISOSYNC "osyncisosync" /* o_sync is REALLY o_sync */
1646 #define MNTOPT_64BITINODE   "inode64"   /* inodes can be allocated anywhere */
1647 #define MNTOPT_IKEEP    "ikeep"         /* do not free empty inode clusters */
1648 #define MNTOPT_NOIKEEP  "noikeep"       /* free empty inode clusters */
1649 #define MNTOPT_LARGEIO     "largeio"    /* report large I/O sizes in stat() */
1650 #define MNTOPT_NOLARGEIO   "nolargeio"  /* do not report large I/O sizes
1651                                          * in stat(). */
1652 #define MNTOPT_ATTR2    "attr2"         /* do use attr2 attribute format */
1653 #define MNTOPT_NOATTR2  "noattr2"       /* do not use attr2 attribute format */
1654
1655 STATIC unsigned long
1656 suffix_strtoul(const char *cp, char **endp, unsigned int base)
1657 {
1658         int     last, shift_left_factor = 0;
1659         char    *value = (char *)cp;
1660
1661         last = strlen(value) - 1;
1662         if (value[last] == 'K' || value[last] == 'k') {
1663                 shift_left_factor = 10;
1664                 value[last] = '\0';
1665         }
1666         if (value[last] == 'M' || value[last] == 'm') {
1667                 shift_left_factor = 20;
1668                 value[last] = '\0';
1669         }
1670         if (value[last] == 'G' || value[last] == 'g') {
1671                 shift_left_factor = 30;
1672                 value[last] = '\0';
1673         }
1674
1675         return simple_strtoul(cp, endp, base) << shift_left_factor;
1676 }
1677
1678 STATIC int
1679 xfs_parseargs(
1680         struct bhv_desc         *bhv,
1681         char                    *options,
1682         struct xfs_mount_args   *args,
1683         int                     update)
1684 {
1685         struct vfs              *vfsp = bhvtovfs(bhv);
1686         char                    *this_char, *value, *eov;
1687         int                     dsunit, dswidth, vol_dsunit, vol_dswidth;
1688         int                     iosize;
1689
1690         args->flags2 |= XFSMNT2_COMPAT_IOSIZE;
1691         args->flags |= XFSMNT_COMPAT_ATTR;
1692         args->flags |= XFSMNT_BARRIER;
1693
1694 #if 0   /* XXX: off by default, until some remaining issues ironed out */
1695         args->flags |= XFSMNT_IDELETE; /* default to on */
1696 #endif
1697
1698         if (!options)
1699                 goto done;
1700
1701         iosize = dsunit = dswidth = vol_dsunit = vol_dswidth = 0;
1702
1703         while ((this_char = strsep(&options, ",")) != NULL) {
1704                 if (!*this_char)
1705                         continue;
1706                 if ((value = strchr(this_char, '=')) != NULL)
1707                         *value++ = 0;
1708
1709                 if (!strcmp(this_char, MNTOPT_LOGBUFS)) {
1710                         if (!value || !*value) {
1711                                 printk("XFS: %s option requires an argument\n",
1712                                         this_char);
1713                                 return EINVAL;
1714                         }
1715                         args->logbufs = simple_strtoul(value, &eov, 10);
1716                 } else if (!strcmp(this_char, MNTOPT_LOGBSIZE)) {
1717                         if (!value || !*value) {
1718                                 printk("XFS: %s option requires an argument\n",
1719                                         this_char);
1720                                 return EINVAL;
1721                         }
1722                         args->logbufsize = suffix_strtoul(value, &eov, 10);
1723                 } else if (!strcmp(this_char, MNTOPT_LOGDEV)) {
1724                         if (!value || !*value) {
1725                                 printk("XFS: %s option requires an argument\n",
1726                                         this_char);
1727                                 return EINVAL;
1728                         }
1729                         strncpy(args->logname, value, MAXNAMELEN);
1730                 } else if (!strcmp(this_char, MNTOPT_MTPT)) {
1731                         if (!value || !*value) {
1732                                 printk("XFS: %s option requires an argument\n",
1733                                         this_char);
1734                                 return EINVAL;
1735                         }
1736                         strncpy(args->mtpt, value, MAXNAMELEN);
1737                 } else if (!strcmp(this_char, MNTOPT_RTDEV)) {
1738                         if (!value || !*value) {
1739                                 printk("XFS: %s option requires an argument\n",
1740                                         this_char);
1741                                 return EINVAL;
1742                         }
1743                         strncpy(args->rtname, value, MAXNAMELEN);
1744                 } else if (!strcmp(this_char, MNTOPT_BIOSIZE)) {
1745                         if (!value || !*value) {
1746                                 printk("XFS: %s option requires an argument\n",
1747                                         this_char);
1748                                 return EINVAL;
1749                         }
1750                         iosize = simple_strtoul(value, &eov, 10);
1751                         args->flags |= XFSMNT_IOSIZE;
1752                         args->iosizelog = (uint8_t) iosize;
1753                 } else if (!strcmp(this_char, MNTOPT_ALLOCSIZE)) {
1754                         if (!value || !*value) {
1755                                 printk("XFS: %s option requires an argument\n",
1756                                         this_char);
1757                                 return EINVAL;
1758                         }
1759                         iosize = suffix_strtoul(value, &eov, 10);
1760                         args->flags |= XFSMNT_IOSIZE;
1761                         args->iosizelog = ffs(iosize) - 1;
1762                 } else if (!strcmp(this_char, MNTOPT_IHASHSIZE)) {
1763                         if (!value || !*value) {
1764                                 printk("XFS: %s option requires an argument\n",
1765                                         this_char);
1766                                 return EINVAL;
1767                         }
1768                         args->flags |= XFSMNT_IHASHSIZE;
1769                         args->ihashsize = simple_strtoul(value, &eov, 10);
1770                 } else if (!strcmp(this_char, MNTOPT_GRPID) ||
1771                            !strcmp(this_char, MNTOPT_BSDGROUPS)) {
1772                         vfsp->vfs_flag |= VFS_GRPID;
1773                 } else if (!strcmp(this_char, MNTOPT_NOGRPID) ||
1774                            !strcmp(this_char, MNTOPT_SYSVGROUPS)) {
1775                         vfsp->vfs_flag &= ~VFS_GRPID;
1776                 } else if (!strcmp(this_char, MNTOPT_WSYNC)) {
1777                         args->flags |= XFSMNT_WSYNC;
1778                 } else if (!strcmp(this_char, MNTOPT_OSYNCISOSYNC)) {
1779                         args->flags |= XFSMNT_OSYNCISOSYNC;
1780                 } else if (!strcmp(this_char, MNTOPT_NORECOVERY)) {
1781                         args->flags |= XFSMNT_NORECOVERY;
1782                 } else if (!strcmp(this_char, MNTOPT_INO64)) {
1783                         args->flags |= XFSMNT_INO64;
1784 #if !XFS_BIG_INUMS
1785                         printk("XFS: %s option not allowed on this system\n",
1786                                 this_char);
1787                         return EINVAL;
1788 #endif
1789                 } else if (!strcmp(this_char, MNTOPT_NOALIGN)) {
1790                         args->flags |= XFSMNT_NOALIGN;
1791                 } else if (!strcmp(this_char, MNTOPT_SWALLOC)) {
1792                         args->flags |= XFSMNT_SWALLOC;
1793                 } else if (!strcmp(this_char, MNTOPT_SUNIT)) {
1794                         if (!value || !*value) {
1795                                 printk("XFS: %s option requires an argument\n",
1796                                         this_char);
1797                                 return EINVAL;
1798                         }
1799                         dsunit = simple_strtoul(value, &eov, 10);
1800                 } else if (!strcmp(this_char, MNTOPT_SWIDTH)) {
1801                         if (!value || !*value) {
1802                                 printk("XFS: %s option requires an argument\n",
1803                                         this_char);
1804                                 return EINVAL;
1805                         }
1806                         dswidth = simple_strtoul(value, &eov, 10);
1807                 } else if (!strcmp(this_char, MNTOPT_64BITINODE)) {
1808                         args->flags &= ~XFSMNT_32BITINODES;
1809 #if !XFS_BIG_INUMS
1810                         printk("XFS: %s option not allowed on this system\n",
1811                                 this_char);
1812                         return EINVAL;
1813 #endif
1814                 } else if (!strcmp(this_char, MNTOPT_NOUUID)) {
1815                         args->flags |= XFSMNT_NOUUID;
1816                 } else if (!strcmp(this_char, MNTOPT_BARRIER)) {
1817                         args->flags |= XFSMNT_BARRIER;
1818                 } else if (!strcmp(this_char, MNTOPT_NOBARRIER)) {
1819                         args->flags &= ~XFSMNT_BARRIER;
1820                 } else if (!strcmp(this_char, MNTOPT_IKEEP)) {
1821                         args->flags &= ~XFSMNT_IDELETE;
1822                 } else if (!strcmp(this_char, MNTOPT_NOIKEEP)) {
1823                         args->flags |= XFSMNT_IDELETE;
1824                 } else if (!strcmp(this_char, MNTOPT_LARGEIO)) {
1825                         args->flags2 &= ~XFSMNT2_COMPAT_IOSIZE;
1826                 } else if (!strcmp(this_char, MNTOPT_NOLARGEIO)) {
1827                         args->flags2 |= XFSMNT2_COMPAT_IOSIZE;
1828                 } else if (!strcmp(this_char, MNTOPT_ATTR2)) {
1829                         args->flags &= ~XFSMNT_COMPAT_ATTR;
1830                 } else if (!strcmp(this_char, MNTOPT_NOATTR2)) {
1831                         args->flags |= XFSMNT_COMPAT_ATTR;
1832                 } else if (!strcmp(this_char, "osyncisdsync")) {
1833                         /* no-op, this is now the default */
1834 printk("XFS: osyncisdsync is now the default, option is deprecated.\n");
1835                 } else if (!strcmp(this_char, "irixsgid")) {
1836 printk("XFS: irixsgid is now a sysctl(2) variable, option is deprecated.\n");
1837                 } else {
1838                         printk("XFS: unknown mount option [%s].\n", this_char);
1839                         return EINVAL;
1840                 }
1841         }
1842
1843         if (args->flags & XFSMNT_NORECOVERY) {
1844                 if ((vfsp->vfs_flag & VFS_RDONLY) == 0) {
1845                         printk("XFS: no-recovery mounts must be read-only.\n");
1846                         return EINVAL;
1847                 }
1848         }
1849
1850         if ((args->flags & XFSMNT_NOALIGN) && (dsunit || dswidth)) {
1851                 printk(
1852         "XFS: sunit and swidth options incompatible with the noalign option\n");
1853                 return EINVAL;
1854         }
1855
1856         if ((dsunit && !dswidth) || (!dsunit && dswidth)) {
1857                 printk("XFS: sunit and swidth must be specified together\n");
1858                 return EINVAL;
1859         }
1860
1861         if (dsunit && (dswidth % dsunit != 0)) {
1862                 printk(
1863         "XFS: stripe width (%d) must be a multiple of the stripe unit (%d)\n",
1864                         dswidth, dsunit);
1865                 return EINVAL;
1866         }
1867
1868         if ((args->flags & XFSMNT_NOALIGN) != XFSMNT_NOALIGN) {
1869                 if (dsunit) {
1870                         args->sunit = dsunit;
1871                         args->flags |= XFSMNT_RETERR;
1872                 } else {
1873                         args->sunit = vol_dsunit;
1874                 }
1875                 dswidth ? (args->swidth = dswidth) :
1876                           (args->swidth = vol_dswidth);
1877         } else {
1878                 args->sunit = args->swidth = 0;
1879         }
1880
1881 done:
1882         if (args->flags & XFSMNT_32BITINODES)
1883                 vfsp->vfs_flag |= VFS_32BITINODES;
1884         if (args->flags2)
1885                 args->flags |= XFSMNT_FLAGS2;
1886         return 0;
1887 }
1888
1889 STATIC int
1890 xfs_showargs(
1891         struct bhv_desc         *bhv,
1892         struct seq_file         *m)
1893 {
1894         static struct proc_xfs_info {
1895                 int     flag;
1896                 char    *str;
1897         } xfs_info[] = {
1898                 /* the few simple ones we can get from the mount struct */
1899                 { XFS_MOUNT_WSYNC,              "," MNTOPT_WSYNC },
1900                 { XFS_MOUNT_INO64,              "," MNTOPT_INO64 },
1901                 { XFS_MOUNT_NOALIGN,            "," MNTOPT_NOALIGN },
1902                 { XFS_MOUNT_SWALLOC,            "," MNTOPT_SWALLOC },
1903                 { XFS_MOUNT_NOUUID,             "," MNTOPT_NOUUID },
1904                 { XFS_MOUNT_NORECOVERY,         "," MNTOPT_NORECOVERY },
1905                 { XFS_MOUNT_OSYNCISOSYNC,       "," MNTOPT_OSYNCISOSYNC },
1906                 { XFS_MOUNT_IDELETE,            "," MNTOPT_NOIKEEP },
1907                 { 0, NULL }
1908         };
1909         struct proc_xfs_info    *xfs_infop;
1910         struct xfs_mount        *mp = XFS_BHVTOM(bhv);
1911         struct vfs              *vfsp = XFS_MTOVFS(mp);
1912
1913         for (xfs_infop = xfs_info; xfs_infop->flag; xfs_infop++) {
1914                 if (mp->m_flags & xfs_infop->flag)
1915                         seq_puts(m, xfs_infop->str);
1916         }
1917
1918         if (mp->m_flags & XFS_MOUNT_IHASHSIZE)
1919                 seq_printf(m, "," MNTOPT_IHASHSIZE "=%d", mp->m_ihsize);
1920
1921         if (mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)
1922                 seq_printf(m, "," MNTOPT_ALLOCSIZE "=%dk",
1923                                 (int)(1 << mp->m_writeio_log) >> 10);
1924
1925         if (mp->m_logbufs > 0)
1926                 seq_printf(m, "," MNTOPT_LOGBUFS "=%d", mp->m_logbufs);
1927         if (mp->m_logbsize > 0)
1928                 seq_printf(m, "," MNTOPT_LOGBSIZE "=%dk", mp->m_logbsize >> 10);
1929
1930         if (mp->m_logname)
1931                 seq_printf(m, "," MNTOPT_LOGDEV "=%s", mp->m_logname);
1932         if (mp->m_rtname)
1933                 seq_printf(m, "," MNTOPT_RTDEV "=%s", mp->m_rtname);
1934
1935         if (mp->m_dalign > 0)
1936                 seq_printf(m, "," MNTOPT_SUNIT "=%d",
1937                                 (int)XFS_FSB_TO_BB(mp, mp->m_dalign));
1938         if (mp->m_swidth > 0)
1939                 seq_printf(m, "," MNTOPT_SWIDTH "=%d",
1940                                 (int)XFS_FSB_TO_BB(mp, mp->m_swidth));
1941
1942         if (!(mp->m_flags & XFS_MOUNT_COMPAT_ATTR))
1943                 seq_printf(m, "," MNTOPT_ATTR2);
1944         if (!(mp->m_flags & XFS_MOUNT_COMPAT_IOSIZE))
1945                 seq_printf(m, "," MNTOPT_LARGEIO);
1946         if (!(mp->m_flags & XFS_MOUNT_BARRIER))
1947                 seq_printf(m, "," MNTOPT_NOBARRIER);
1948
1949         if (!(vfsp->vfs_flag & VFS_32BITINODES))
1950                 seq_printf(m, "," MNTOPT_64BITINODE);
1951         if (vfsp->vfs_flag & VFS_GRPID)
1952                 seq_printf(m, "," MNTOPT_GRPID);
1953
1954         return 0;
1955 }
1956
1957 STATIC void
1958 xfs_freeze(
1959         bhv_desc_t      *bdp)
1960 {
1961         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
1962
1963         while (atomic_read(&mp->m_active_trans) > 0)
1964                 delay(100);
1965
1966         /* Push the superblock and write an unmount record */
1967         xfs_log_unmount_write(mp);
1968         xfs_unmountfs_writesb(mp);
1969 }
1970
1971
1972 vfsops_t xfs_vfsops = {
1973         BHV_IDENTITY_INIT(VFS_BHV_XFS,VFS_POSITION_XFS),
1974         .vfs_parseargs          = xfs_parseargs,
1975         .vfs_showargs           = xfs_showargs,
1976         .vfs_mount              = xfs_mount,
1977         .vfs_unmount            = xfs_unmount,
1978         .vfs_mntupdate          = xfs_mntupdate,
1979         .vfs_root               = xfs_root,
1980         .vfs_statvfs            = xfs_statvfs,
1981         .vfs_sync               = xfs_sync,
1982         .vfs_vget               = xfs_vget,
1983         .vfs_dmapiops           = (vfs_dmapiops_t)fs_nosys,
1984         .vfs_quotactl           = (vfs_quotactl_t)fs_nosys,
1985         .vfs_init_vnode         = xfs_initialize_vnode,
1986         .vfs_force_shutdown     = xfs_do_force_shutdown,
1987         .vfs_freeze             = xfs_freeze,
1988 };