err.no Git - linux-2.6/blob - fs/xfs/xfs_vfsops.c

   1 /*
   2  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
   3  * All Rights Reserved.
   4  *
   5  * This program is free software; you can redistribute it and/or
   6  * modify it under the terms of the GNU General Public License as
   7  * published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it would be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write the Free Software Foundation,
  16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17  */
  18 #include "xfs.h"
  19 #include "xfs_fs.h"
  20 #include "xfs_types.h"
  21 #include "xfs_bit.h"
  22 #include "xfs_log.h"
  23 #include "xfs_inum.h"
  24 #include "xfs_trans.h"
  25 #include "xfs_sb.h"
  26 #include "xfs_ag.h"
  27 #include "xfs_dir2.h"
  28 #include "xfs_dmapi.h"
  29 #include "xfs_mount.h"
  30 #include "xfs_da_btree.h"
  31 #include "xfs_bmap_btree.h"
  32 #include "xfs_ialloc_btree.h"
  33 #include "xfs_alloc_btree.h"
  34 #include "xfs_dir2_sf.h"
  35 #include "xfs_attr_sf.h"
  36 #include "xfs_dinode.h"
  37 #include "xfs_inode.h"
  38 #include "xfs_inode_item.h"
  39 #include "xfs_btree.h"
  40 #include "xfs_alloc.h"
  41 #include "xfs_ialloc.h"
  42 #include "xfs_quota.h"
  43 #include "xfs_error.h"
  44 #include "xfs_bmap.h"
  45 #include "xfs_rw.h"
  46 #include "xfs_refcache.h"
  47 #include "xfs_buf_item.h"
  48 #include "xfs_log_priv.h"
  49 #include "xfs_dir2_trace.h"
  50 #include "xfs_extfree_item.h"
  51 #include "xfs_acl.h"
  52 #include "xfs_attr.h"
  53 #include "xfs_clnt.h"
  54 #include "xfs_mru_cache.h"
  55 #include "xfs_filestream.h"
  56 #include "xfs_fsops.h"
  57 #include "xfs_vnodeops.h"
  58 #include "xfs_vfsops.h"
  59
  60
  61 int
  62 xfs_init(void)
  63 {
  64         extern kmem_zone_t      *xfs_bmap_free_item_zone;
  65         extern kmem_zone_t      *xfs_btree_cur_zone;
  66         extern kmem_zone_t      *xfs_trans_zone;
  67         extern kmem_zone_t      *xfs_buf_item_zone;
  68         extern kmem_zone_t      *xfs_dabuf_zone;
  69 #ifdef XFS_DABUF_DEBUG
  70         extern lock_t           xfs_dabuf_global_lock;
  71         spinlock_init(&xfs_dabuf_global_lock, "xfsda");
  72 #endif
  73
  74         /*
  75          * Initialize all of the zone allocators we use.
  76          */
  77         xfs_bmap_free_item_zone = kmem_zone_init(sizeof(xfs_bmap_free_item_t),
  78                                                  "xfs_bmap_free_item");
  79         xfs_btree_cur_zone = kmem_zone_init(sizeof(xfs_btree_cur_t),
  80                                             "xfs_btree_cur");
  81         xfs_trans_zone = kmem_zone_init(sizeof(xfs_trans_t), "xfs_trans");
  82         xfs_da_state_zone =
  83                 kmem_zone_init(sizeof(xfs_da_state_t), "xfs_da_state");
  84         xfs_dabuf_zone = kmem_zone_init(sizeof(xfs_dabuf_t), "xfs_dabuf");
  85         xfs_ifork_zone = kmem_zone_init(sizeof(xfs_ifork_t), "xfs_ifork");
  86         xfs_acl_zone_init(xfs_acl_zone, "xfs_acl");
  87         xfs_mru_cache_init();
  88         xfs_filestream_init();
  89
  90         /*
  91          * The size of the zone allocated buf log item is the maximum
  92          * size possible under XFS.  This wastes a little bit of memory,
  93          * but it is much faster.
  94          */
  95         xfs_buf_item_zone =
  96                 kmem_zone_init((sizeof(xfs_buf_log_item_t) +
  97                                 (((XFS_MAX_BLOCKSIZE / XFS_BLI_CHUNK) /
  98                                   NBWORD) * sizeof(int))),
  99                                "xfs_buf_item");
 100         xfs_efd_zone =
 101                 kmem_zone_init((sizeof(xfs_efd_log_item_t) +
 102                                ((XFS_EFD_MAX_FAST_EXTENTS - 1) *
 103                                  sizeof(xfs_extent_t))),
 104                                       "xfs_efd_item");
 105         xfs_efi_zone =
 106                 kmem_zone_init((sizeof(xfs_efi_log_item_t) +
 107                                ((XFS_EFI_MAX_FAST_EXTENTS - 1) *
 108                                  sizeof(xfs_extent_t))),
 109                                       "xfs_efi_item");
 110
 111         /*
 112          * These zones warrant special memory allocator hints
 113          */
 114         xfs_inode_zone =
 115                 kmem_zone_init_flags(sizeof(xfs_inode_t), "xfs_inode",
 116                                         KM_ZONE_HWALIGN | KM_ZONE_RECLAIM |
 117                                         KM_ZONE_SPREAD, NULL);
 118         xfs_ili_zone =
 119                 kmem_zone_init_flags(sizeof(xfs_inode_log_item_t), "xfs_ili",
 120                                         KM_ZONE_SPREAD, NULL);
 121         xfs_icluster_zone =
 122                 kmem_zone_init_flags(sizeof(xfs_icluster_t), "xfs_icluster",
 123                                         KM_ZONE_SPREAD, NULL);
 124
 125         /*
 126          * Allocate global trace buffers.
 127          */
 128 #ifdef XFS_ALLOC_TRACE
 129         xfs_alloc_trace_buf = ktrace_alloc(XFS_ALLOC_TRACE_SIZE, KM_SLEEP);
 130 #endif
 131 #ifdef XFS_BMAP_TRACE
 132         xfs_bmap_trace_buf = ktrace_alloc(XFS_BMAP_TRACE_SIZE, KM_SLEEP);
 133 #endif
 134 #ifdef XFS_BMBT_TRACE
 135         xfs_bmbt_trace_buf = ktrace_alloc(XFS_BMBT_TRACE_SIZE, KM_SLEEP);
 136 #endif
 137 #ifdef XFS_ATTR_TRACE
 138         xfs_attr_trace_buf = ktrace_alloc(XFS_ATTR_TRACE_SIZE, KM_SLEEP);
 139 #endif
 140 #ifdef XFS_DIR2_TRACE
 141         xfs_dir2_trace_buf = ktrace_alloc(XFS_DIR2_GTRACE_SIZE, KM_SLEEP);
 142 #endif
 143
 144         xfs_dir_startup();
 145
 146 #if (defined(DEBUG) || defined(INDUCE_IO_ERROR))
 147         xfs_error_test_init();
 148 #endif /* DEBUG || INDUCE_IO_ERROR */
 149
 150         xfs_init_procfs();
 151         xfs_sysctl_register();
 152         return 0;
 153 }
 154
 155 void
 156 xfs_cleanup(void)
 157 {
 158         extern kmem_zone_t      *xfs_bmap_free_item_zone;
 159         extern kmem_zone_t      *xfs_btree_cur_zone;
 160         extern kmem_zone_t      *xfs_inode_zone;
 161         extern kmem_zone_t      *xfs_trans_zone;
 162         extern kmem_zone_t      *xfs_da_state_zone;
 163         extern kmem_zone_t      *xfs_dabuf_zone;
 164         extern kmem_zone_t      *xfs_efd_zone;
 165         extern kmem_zone_t      *xfs_efi_zone;
 166         extern kmem_zone_t      *xfs_buf_item_zone;
 167         extern kmem_zone_t      *xfs_icluster_zone;
 168
 169         xfs_cleanup_procfs();
 170         xfs_sysctl_unregister();
 171         xfs_refcache_destroy();
 172         xfs_filestream_uninit();
 173         xfs_mru_cache_uninit();
 174         xfs_acl_zone_destroy(xfs_acl_zone);
 175
 176 #ifdef XFS_DIR2_TRACE
 177         ktrace_free(xfs_dir2_trace_buf);
 178 #endif
 179 #ifdef XFS_ATTR_TRACE
 180         ktrace_free(xfs_attr_trace_buf);
 181 #endif
 182 #ifdef XFS_BMBT_TRACE
 183         ktrace_free(xfs_bmbt_trace_buf);
 184 #endif
 185 #ifdef XFS_BMAP_TRACE
 186         ktrace_free(xfs_bmap_trace_buf);
 187 #endif
 188 #ifdef XFS_ALLOC_TRACE
 189         ktrace_free(xfs_alloc_trace_buf);
 190 #endif
 191
 192         kmem_zone_destroy(xfs_bmap_free_item_zone);
 193         kmem_zone_destroy(xfs_btree_cur_zone);
 194         kmem_zone_destroy(xfs_inode_zone);
 195         kmem_zone_destroy(xfs_trans_zone);
 196         kmem_zone_destroy(xfs_da_state_zone);
 197         kmem_zone_destroy(xfs_dabuf_zone);
 198         kmem_zone_destroy(xfs_buf_item_zone);
 199         kmem_zone_destroy(xfs_efd_zone);
 200         kmem_zone_destroy(xfs_efi_zone);
 201         kmem_zone_destroy(xfs_ifork_zone);
 202         kmem_zone_destroy(xfs_ili_zone);
 203         kmem_zone_destroy(xfs_icluster_zone);
 204 }
 205
 206 /*
 207  * xfs_start_flags
 208  *
 209  * This function fills in xfs_mount_t fields based on mount args.
 210  * Note: the superblock has _not_ yet been read in.
 211  */
 212 STATIC int
 213 xfs_start_flags(
 214         struct bhv_vfs          *vfs,
 215         struct xfs_mount_args   *ap,
 216         struct xfs_mount        *mp)
 217 {
 218         /* Values are in BBs */
 219         if ((ap->flags & XFSMNT_NOALIGN) != XFSMNT_NOALIGN) {
 220                 /*
 221                  * At this point the superblock has not been read
 222                  * in, therefore we do not know the block size.
 223                  * Before the mount call ends we will convert
 224                  * these to FSBs.
 225                  */
 226                 mp->m_dalign = ap->sunit;
 227                 mp->m_swidth = ap->swidth;
 228         }
 229
 230         if (ap->logbufs != -1 &&
 231             ap->logbufs != 0 &&
 232             (ap->logbufs < XLOG_MIN_ICLOGS ||
 233              ap->logbufs > XLOG_MAX_ICLOGS)) {
 234                 cmn_err(CE_WARN,
 235                         "XFS: invalid logbufs value: %d [not %d-%d]",
 236                         ap->logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
 237                 return XFS_ERROR(EINVAL);
 238         }
 239         mp->m_logbufs = ap->logbufs;
 240         if (ap->logbufsize != -1 &&
 241             ap->logbufsize !=  0 &&
 242             (ap->logbufsize < XLOG_MIN_RECORD_BSIZE ||
 243              ap->logbufsize > XLOG_MAX_RECORD_BSIZE ||
 244              !is_power_of_2(ap->logbufsize))) {
 245                 cmn_err(CE_WARN,
 246         "XFS: invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
 247                         ap->logbufsize);
 248                 return XFS_ERROR(EINVAL);
 249         }
 250         mp->m_logbsize = ap->logbufsize;
 251         mp->m_fsname_len = strlen(ap->fsname) + 1;
 252         mp->m_fsname = kmem_alloc(mp->m_fsname_len, KM_SLEEP);
 253         strcpy(mp->m_fsname, ap->fsname);
 254         if (ap->rtname[0]) {
 255                 mp->m_rtname = kmem_alloc(strlen(ap->rtname) + 1, KM_SLEEP);
 256                 strcpy(mp->m_rtname, ap->rtname);
 257         }
 258         if (ap->logname[0]) {
 259                 mp->m_logname = kmem_alloc(strlen(ap->logname) + 1, KM_SLEEP);
 260                 strcpy(mp->m_logname, ap->logname);
 261         }
 262
 263         if (ap->flags & XFSMNT_WSYNC)
 264                 mp->m_flags |= XFS_MOUNT_WSYNC;
 265 #if XFS_BIG_INUMS
 266         if (ap->flags & XFSMNT_INO64) {
 267                 mp->m_flags |= XFS_MOUNT_INO64;
 268                 mp->m_inoadd = XFS_INO64_OFFSET;
 269         }
 270 #endif
 271         if (ap->flags & XFSMNT_RETERR)
 272                 mp->m_flags |= XFS_MOUNT_RETERR;
 273         if (ap->flags & XFSMNT_NOALIGN)
 274                 mp->m_flags |= XFS_MOUNT_NOALIGN;
 275         if (ap->flags & XFSMNT_SWALLOC)
 276                 mp->m_flags |= XFS_MOUNT_SWALLOC;
 277         if (ap->flags & XFSMNT_OSYNCISOSYNC)
 278                 mp->m_flags |= XFS_MOUNT_OSYNCISOSYNC;
 279         if (ap->flags & XFSMNT_32BITINODES)
 280                 mp->m_flags |= XFS_MOUNT_32BITINODES;
 281
 282         if (ap->flags & XFSMNT_IOSIZE) {
 283                 if (ap->iosizelog > XFS_MAX_IO_LOG ||
 284                     ap->iosizelog < XFS_MIN_IO_LOG) {
 285                         cmn_err(CE_WARN,
 286                 "XFS: invalid log iosize: %d [not %d-%d]",
 287                                 ap->iosizelog, XFS_MIN_IO_LOG,
 288                                 XFS_MAX_IO_LOG);
 289                         return XFS_ERROR(EINVAL);
 290                 }
 291
 292                 mp->m_flags |= XFS_MOUNT_DFLT_IOSIZE;
 293                 mp->m_readio_log = mp->m_writeio_log = ap->iosizelog;
 294         }
 295
 296         if (ap->flags & XFSMNT_IDELETE)
 297                 mp->m_flags |= XFS_MOUNT_IDELETE;
 298         if (ap->flags & XFSMNT_DIRSYNC)
 299                 mp->m_flags |= XFS_MOUNT_DIRSYNC;
 300         if (ap->flags & XFSMNT_ATTR2)
 301                 mp->m_flags |= XFS_MOUNT_ATTR2;
 302
 303         if (ap->flags2 & XFSMNT2_COMPAT_IOSIZE)
 304                 mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE;
 305
 306         /*
 307          * no recovery flag requires a read-only mount
 308          */
 309         if (ap->flags & XFSMNT_NORECOVERY) {
 310                 if (!(mp->m_flags & XFS_MOUNT_RDONLY)) {
 311                         cmn_err(CE_WARN,
 312         "XFS: tried to mount a FS read-write without recovery!");
 313                         return XFS_ERROR(EINVAL);
 314                 }
 315                 mp->m_flags |= XFS_MOUNT_NORECOVERY;
 316         }
 317
 318         if (ap->flags & XFSMNT_NOUUID)
 319                 mp->m_flags |= XFS_MOUNT_NOUUID;
 320         if (ap->flags & XFSMNT_BARRIER)
 321                 mp->m_flags |= XFS_MOUNT_BARRIER;
 322         else
 323                 mp->m_flags &= ~XFS_MOUNT_BARRIER;
 324
 325         if (ap->flags2 & XFSMNT2_FILESTREAMS)
 326                 mp->m_flags |= XFS_MOUNT_FILESTREAMS;
 327
 328         if (ap->flags & XFSMNT_DMAPI)
 329                 mp->m_flags |= XFS_MOUNT_DMAPI;
 330         return 0;
 331 }
 332
 333 /*
 334  * This function fills in xfs_mount_t fields based on mount args.
 335  * Note: the superblock _has_ now been read in.
 336  */
 337 STATIC int
 338 xfs_finish_flags(
 339         struct bhv_vfs          *vfs,
 340         struct xfs_mount_args   *ap,
 341         struct xfs_mount        *mp)
 342 {
 343         int                     ronly = (mp->m_flags & XFS_MOUNT_RDONLY);
 344
 345         /* Fail a mount where the logbuf is smaller then the log stripe */
 346         if (XFS_SB_VERSION_HASLOGV2(&mp->m_sb)) {
 347                 if ((ap->logbufsize <= 0) &&
 348                     (mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE)) {
 349                         mp->m_logbsize = mp->m_sb.sb_logsunit;
 350                 } else if (ap->logbufsize > 0 &&
 351                            ap->logbufsize < mp->m_sb.sb_logsunit) {
 352                         cmn_err(CE_WARN,
 353         "XFS: logbuf size must be greater than or equal to log stripe size");
 354                         return XFS_ERROR(EINVAL);
 355                 }
 356         } else {
 357                 /* Fail a mount if the logbuf is larger than 32K */
 358                 if (ap->logbufsize > XLOG_BIG_RECORD_BSIZE) {
 359                         cmn_err(CE_WARN,
 360         "XFS: logbuf size for version 1 logs must be 16K or 32K");
 361                         return XFS_ERROR(EINVAL);
 362                 }
 363         }
 364
 365         if (XFS_SB_VERSION_HASATTR2(&mp->m_sb)) {
 366                 mp->m_flags |= XFS_MOUNT_ATTR2;
 367         }
 368
 369         /*
 370          * prohibit r/w mounts of read-only filesystems
 371          */
 372         if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !ronly) {
 373                 cmn_err(CE_WARN,
 374         "XFS: cannot mount a read-only filesystem as read-write");
 375                 return XFS_ERROR(EROFS);
 376         }
 377
 378         /*
 379          * check for shared mount.
 380          */
 381         if (ap->flags & XFSMNT_SHARED) {
 382                 if (!XFS_SB_VERSION_HASSHARED(&mp->m_sb))
 383                         return XFS_ERROR(EINVAL);
 384
 385                 /*
 386                  * For IRIX 6.5, shared mounts must have the shared
 387                  * version bit set, have the persistent readonly
 388                  * field set, must be version 0 and can only be mounted
 389                  * read-only.
 390                  */
 391                 if (!ronly || !(mp->m_sb.sb_flags & XFS_SBF_READONLY) ||
 392                      (mp->m_sb.sb_shared_vn != 0))
 393                         return XFS_ERROR(EINVAL);
 394
 395                 mp->m_flags |= XFS_MOUNT_SHARED;
 396
 397                 /*
 398                  * Shared XFS V0 can't deal with DMI.  Return EINVAL.
 399                  */
 400                 if (mp->m_sb.sb_shared_vn == 0 && (ap->flags & XFSMNT_DMAPI))
 401                         return XFS_ERROR(EINVAL);
 402         }
 403
 404         if (ap->flags & XFSMNT_UQUOTA) {
 405                 mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE);
 406                 if (ap->flags & XFSMNT_UQUOTAENF)
 407                         mp->m_qflags |= XFS_UQUOTA_ENFD;
 408         }
 409
 410         if (ap->flags & XFSMNT_GQUOTA) {
 411                 mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE);
 412                 if (ap->flags & XFSMNT_GQUOTAENF)
 413                         mp->m_qflags |= XFS_OQUOTA_ENFD;
 414         } else if (ap->flags & XFSMNT_PQUOTA) {
 415                 mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE);
 416                 if (ap->flags & XFSMNT_PQUOTAENF)
 417                         mp->m_qflags |= XFS_OQUOTA_ENFD;
 418         }
 419
 420         return 0;
 421 }
 422
 423 /*
 424  * xfs_mount
 425  *
 426  * The file system configurations are:
 427  *      (1) device (partition) with data and internal log
 428  *      (2) logical volume with data and log subvolumes.
 429  *      (3) logical volume with data, log, and realtime subvolumes.
 430  *
 431  * We only have to handle opening the log and realtime volumes here if
 432  * they are present.  The data subvolume has already been opened by
 433  * get_sb_bdev() and is stored in vfsp->vfs_super->s_bdev.
 434  */
 435 int
 436 xfs_mount(
 437         struct xfs_mount        *mp,
 438         struct xfs_mount_args   *args,
 439         cred_t                  *credp)
 440 {
 441         struct bhv_vfs          *vfsp = XFS_MTOVFS(mp);
 442         struct block_device     *ddev, *logdev, *rtdev;
 443         int                     flags = 0, error;
 444
 445         ddev = vfsp->vfs_super->s_bdev;
 446         logdev = rtdev = NULL;
 447
 448         error = xfs_dmops_get(mp, args);
 449         if (error)
 450                 return error;
 451         error = xfs_qmops_get(mp, args);
 452         if (error)
 453                 return error;
 454
 455         mp->m_io_ops = xfs_iocore_xfs;
 456
 457         if (args->flags & XFSMNT_QUIET)
 458                 flags |= XFS_MFSI_QUIET;
 459
 460         /*
 461          * Open real time and log devices - order is important.
 462          */
 463         if (args->logname[0]) {
 464                 error = xfs_blkdev_get(mp, args->logname, &logdev);
 465                 if (error)
 466                         return error;
 467         }
 468         if (args->rtname[0]) {
 469                 error = xfs_blkdev_get(mp, args->rtname, &rtdev);
 470                 if (error) {
 471                         xfs_blkdev_put(logdev);
 472                         return error;
 473                 }
 474
 475                 if (rtdev == ddev || rtdev == logdev) {
 476                         cmn_err(CE_WARN,
 477         "XFS: Cannot mount filesystem with identical rtdev and ddev/logdev.");
 478                         xfs_blkdev_put(logdev);
 479                         xfs_blkdev_put(rtdev);
 480                         return EINVAL;
 481                 }
 482         }
 483
 484         /*
 485          * Setup xfs_mount buffer target pointers
 486          */
 487         error = ENOMEM;
 488         mp->m_ddev_targp = xfs_alloc_buftarg(ddev, 0);
 489         if (!mp->m_ddev_targp) {
 490                 xfs_blkdev_put(logdev);
 491                 xfs_blkdev_put(rtdev);
 492                 return error;
 493         }
 494         if (rtdev) {
 495                 mp->m_rtdev_targp = xfs_alloc_buftarg(rtdev, 1);
 496                 if (!mp->m_rtdev_targp) {
 497                         xfs_blkdev_put(logdev);
 498                         xfs_blkdev_put(rtdev);
 499                         goto error0;
 500                 }
 501         }
 502         mp->m_logdev_targp = (logdev && logdev != ddev) ?
 503                                 xfs_alloc_buftarg(logdev, 1) : mp->m_ddev_targp;
 504         if (!mp->m_logdev_targp) {
 505                 xfs_blkdev_put(logdev);
 506                 xfs_blkdev_put(rtdev);
 507                 goto error0;
 508         }
 509
 510         /*
 511          * Setup flags based on mount(2) options and then the superblock
 512          */
 513         error = xfs_start_flags(vfsp, args, mp);
 514         if (error)
 515                 goto error1;
 516         error = xfs_readsb(mp, flags);
 517         if (error)
 518                 goto error1;
 519         error = xfs_finish_flags(vfsp, args, mp);
 520         if (error)
 521                 goto error2;
 522
 523         /*
 524          * Setup xfs_mount buffer target pointers based on superblock
 525          */
 526         error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_blocksize,
 527                                     mp->m_sb.sb_sectsize);
 528         if (!error && logdev && logdev != ddev) {
 529                 unsigned int    log_sector_size = BBSIZE;
 530
 531                 if (XFS_SB_VERSION_HASSECTOR(&mp->m_sb))
 532                         log_sector_size = mp->m_sb.sb_logsectsize;
 533                 error = xfs_setsize_buftarg(mp->m_logdev_targp,
 534                                             mp->m_sb.sb_blocksize,
 535                                             log_sector_size);
 536         }
 537         if (!error && rtdev)
 538                 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
 539                                             mp->m_sb.sb_blocksize,
 540                                             mp->m_sb.sb_sectsize);
 541         if (error)
 542                 goto error2;
 543
 544         if (mp->m_flags & XFS_MOUNT_BARRIER)
 545                 xfs_mountfs_check_barriers(mp);
 546
 547         if ((error = xfs_filestream_mount(mp)))
 548                 goto error2;
 549
 550         error = XFS_IOINIT(vfsp, args, flags);
 551         if (error)
 552                 goto error2;
 553
 554         XFS_SEND_MOUNT(mp, DM_RIGHT_NULL, args->mtpt, args->fsname);
 555
 556         return 0;
 557
 558 error2:
 559         if (mp->m_sb_bp)
 560                 xfs_freesb(mp);
 561 error1:
 562         xfs_binval(mp->m_ddev_targp);
 563         if (logdev && logdev != ddev)
 564                 xfs_binval(mp->m_logdev_targp);
 565         if (rtdev)
 566                 xfs_binval(mp->m_rtdev_targp);
 567 error0:
 568         xfs_unmountfs_close(mp, credp);
 569         xfs_qmops_put(mp);
 570         xfs_dmops_put(mp);
 571         return error;
 572 }
 573
 574 int
 575 xfs_unmount(
 576         xfs_mount_t     *mp,
 577         int             flags,
 578         cred_t          *credp)
 579 {
 580         bhv_vfs_t       *vfsp = XFS_MTOVFS(mp);
 581         xfs_inode_t     *rip;
 582         bhv_vnode_t     *rvp;
 583         int             unmount_event_wanted = 0;
 584         int             unmount_event_flags = 0;
 585         int             xfs_unmountfs_needed = 0;
 586         int             error;
 587
 588         rip = mp->m_rootip;
 589         rvp = XFS_ITOV(rip);
 590
 591 #ifdef HAVE_DMAPI
 592         if (mp->m_flags & XFS_MOUNT_DMAPI) {
 593                 error = XFS_SEND_PREUNMOUNT(mp, vfsp,
 594                                 rvp, DM_RIGHT_NULL, rvp, DM_RIGHT_NULL,
 595                                 NULL, NULL, 0, 0,
 596                                 (mp->m_dmevmask & (1<<DM_EVENT_PREUNMOUNT))?
 597                                         0:DM_FLAGS_UNWANTED);
 598                         if (error)
 599                                 return XFS_ERROR(error);
 600                 unmount_event_wanted = 1;
 601                 unmount_event_flags = (mp->m_dmevmask & (1<<DM_EVENT_UNMOUNT))?
 602                                         0 : DM_FLAGS_UNWANTED;
 603         }
 604 #endif
 605         /*
 606          * First blow any referenced inode from this file system
 607          * out of the reference cache, and delete the timer.
 608          */
 609         xfs_refcache_purge_mp(mp);
 610
 611         /*
 612          * Blow away any referenced inode in the filestreams cache.
 613          * This can and will cause log traffic as inodes go inactive
 614          * here.
 615          */
 616         xfs_filestream_unmount(mp);
 617
 618         XFS_bflush(mp->m_ddev_targp);
 619         error = xfs_unmount_flush(mp, 0);
 620         if (error)
 621                 goto out;
 622
 623         ASSERT(vn_count(rvp) == 1);
 624
 625         /*
 626          * Drop the reference count
 627          */
 628         VN_RELE(rvp);
 629
 630         /*
 631          * If we're forcing a shutdown, typically because of a media error,
 632          * we want to make sure we invalidate dirty pages that belong to
 633          * referenced vnodes as well.
 634          */
 635         if (XFS_FORCED_SHUTDOWN(mp)) {
 636                 error = xfs_sync(mp, SYNC_WAIT | SYNC_CLOSE);
 637                 ASSERT(error != EFSCORRUPTED);
 638         }
 639         xfs_unmountfs_needed = 1;
 640
 641 out:
 642         /*      Send DMAPI event, if required.
 643          *      Then do xfs_unmountfs() if needed.
 644          *      Then return error (or zero).
 645          */
 646         if (unmount_event_wanted) {
 647                 /* Note: mp structure must still exist for
 648                  * XFS_SEND_UNMOUNT() call.
 649                  */
 650                 XFS_SEND_UNMOUNT(mp, vfsp, error == 0 ? rvp : NULL,
 651                         DM_RIGHT_NULL, 0, error, unmount_event_flags);
 652         }
 653         if (xfs_unmountfs_needed) {
 654                 /*
 655                  * Call common unmount function to flush to disk
 656                  * and free the super block buffer & mount structures.
 657                  */
 658                 xfs_unmountfs(mp, credp);
 659                 xfs_qmops_put(mp);
 660                 xfs_dmops_put(mp);
 661                 kmem_free(mp, sizeof(xfs_mount_t));
 662         }
 663
 664         return XFS_ERROR(error);
 665 }
 666
 667 STATIC int
 668 xfs_quiesce_fs(
 669         xfs_mount_t             *mp)
 670 {
 671         int                     count = 0, pincount;
 672
 673         xfs_refcache_purge_mp(mp);
 674         xfs_flush_buftarg(mp->m_ddev_targp, 0);
 675         xfs_finish_reclaim_all(mp, 0);
 676
 677         /* This loop must run at least twice.
 678          * The first instance of the loop will flush
 679          * most meta data but that will generate more
 680          * meta data (typically directory updates).
 681          * Which then must be flushed and logged before
 682          * we can write the unmount record.
 683          */
 684         do {
 685                 xfs_syncsub(mp, SYNC_INODE_QUIESCE, NULL);
 686                 pincount = xfs_flush_buftarg(mp->m_ddev_targp, 1);
 687                 if (!pincount) {
 688                         delay(50);
 689                         count++;
 690                 }
 691         } while (count < 2);
 692
 693         return 0;
 694 }
 695
 696 /*
 697  * Second stage of a quiesce. The data is already synced, now we have to take
 698  * care of the metadata. New transactions are already blocked, so we need to
 699  * wait for any remaining transactions to drain out before proceding.
 700  */
 701 STATIC void
 702 xfs_attr_quiesce(
 703         xfs_mount_t     *mp)
 704 {
 705         /* wait for all modifications to complete */
 706         while (atomic_read(&mp->m_active_trans) > 0)
 707                 delay(100);
 708
 709         /* flush inodes and push all remaining buffers out to disk */
 710         xfs_quiesce_fs(mp);
 711
 712         ASSERT_ALWAYS(atomic_read(&mp->m_active_trans) == 0);
 713
 714         /* Push the superblock and write an unmount record */
 715         xfs_log_sbcount(mp, 1);
 716         xfs_log_unmount_write(mp);
 717         xfs_unmountfs_writesb(mp);
 718 }
 719
 720 int
 721 xfs_mntupdate(
 722         struct xfs_mount                *mp,
 723         int                             *flags,
 724         struct xfs_mount_args           *args)
 725 {
 726         if (!(*flags & MS_RDONLY)) {                    /* rw/ro -> rw */
 727                 if (mp->m_flags & XFS_MOUNT_RDONLY)
 728                         mp->m_flags &= ~XFS_MOUNT_RDONLY;
 729                 if (args->flags & XFSMNT_BARRIER) {
 730                         mp->m_flags |= XFS_MOUNT_BARRIER;
 731                         xfs_mountfs_check_barriers(mp);
 732                 } else {
 733                         mp->m_flags &= ~XFS_MOUNT_BARRIER;
 734                 }
 735         } else if (!(mp->m_flags & XFS_MOUNT_RDONLY)) { /* rw -> ro */
 736                 xfs_filestream_flush(mp);
 737                 xfs_sync(mp, SYNC_DATA_QUIESCE);
 738                 xfs_attr_quiesce(mp);
 739                 mp->m_flags |= XFS_MOUNT_RDONLY;
 740         }
 741         return 0;
 742 }
 743
 744 /*
 745  * xfs_unmount_flush implements a set of flush operation on special
 746  * inodes, which are needed as a separate set of operations so that
 747  * they can be called as part of relocation process.
 748  */
 749 int
 750 xfs_unmount_flush(
 751         xfs_mount_t     *mp,            /* Mount structure we are getting
 752                                            rid of. */
 753         int             relocation)     /* Called from vfs relocation. */
 754 {
 755         xfs_inode_t     *rip = mp->m_rootip;
 756         xfs_inode_t     *rbmip;
 757         xfs_inode_t     *rsumip = NULL;
 758         bhv_vnode_t     *rvp = XFS_ITOV(rip);
 759         int             error;
 760
 761         xfs_ilock(rip, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
 762         xfs_iflock(rip);
 763
 764         /*
 765          * Flush out the real time inodes.
 766          */
 767         if ((rbmip = mp->m_rbmip) != NULL) {
 768                 xfs_ilock(rbmip, XFS_ILOCK_EXCL);
 769                 xfs_iflock(rbmip);
 770                 error = xfs_iflush(rbmip, XFS_IFLUSH_SYNC);
 771                 xfs_iunlock(rbmip, XFS_ILOCK_EXCL);
 772
 773                 if (error == EFSCORRUPTED)
 774                         goto fscorrupt_out;
 775
 776                 ASSERT(vn_count(XFS_ITOV(rbmip)) == 1);
 777
 778                 rsumip = mp->m_rsumip;
 779                 xfs_ilock(rsumip, XFS_ILOCK_EXCL);
 780                 xfs_iflock(rsumip);
 781                 error = xfs_iflush(rsumip, XFS_IFLUSH_SYNC);
 782                 xfs_iunlock(rsumip, XFS_ILOCK_EXCL);
 783
 784                 if (error == EFSCORRUPTED)
 785                         goto fscorrupt_out;
 786
 787                 ASSERT(vn_count(XFS_ITOV(rsumip)) == 1);
 788         }
 789
 790         /*
 791          * Synchronously flush root inode to disk
 792          */
 793         error = xfs_iflush(rip, XFS_IFLUSH_SYNC);
 794         if (error == EFSCORRUPTED)
 795                 goto fscorrupt_out2;
 796
 797         if (vn_count(rvp) != 1 && !relocation) {
 798                 xfs_iunlock(rip, XFS_ILOCK_EXCL);
 799                 return XFS_ERROR(EBUSY);
 800         }
 801
 802         /*
 803          * Release dquot that rootinode, rbmino and rsumino might be holding,
 804          * flush and purge the quota inodes.
 805          */
 806         error = XFS_QM_UNMOUNT(mp);
 807         if (error == EFSCORRUPTED)
 808                 goto fscorrupt_out2;
 809
 810         if (rbmip) {
 811                 VN_RELE(XFS_ITOV(rbmip));
 812                 VN_RELE(XFS_ITOV(rsumip));
 813         }
 814
 815         xfs_iunlock(rip, XFS_ILOCK_EXCL);
 816         return 0;
 817
 818 fscorrupt_out:
 819         xfs_ifunlock(rip);
 820
 821 fscorrupt_out2:
 822         xfs_iunlock(rip, XFS_ILOCK_EXCL);
 823
 824         return XFS_ERROR(EFSCORRUPTED);
 825 }
 826
 827 /*
 828  * xfs_root extracts the root vnode from a vfs.
 829  *
 830  * vfsp -- the vfs struct for the desired file system
 831  * vpp  -- address of the caller's vnode pointer which should be
 832  *         set to the desired fs root vnode
 833  */
 834 int
 835 xfs_root(
 836         xfs_mount_t     *mp,
 837         bhv_vnode_t     **vpp)
 838 {
 839         bhv_vnode_t     *vp;
 840
 841         vp = XFS_ITOV(mp->m_rootip);
 842         VN_HOLD(vp);
 843         *vpp = vp;
 844         return 0;
 845 }
 846
 847 /*
 848  * xfs_statvfs
 849  *
 850  * Fill in the statvfs structure for the given file system.  We use
 851  * the superblock lock in the mount structure to ensure a consistent
 852  * snapshot of the counters returned.
 853  */
 854 int
 855 xfs_statvfs(
 856         xfs_mount_t     *mp,
 857         bhv_statvfs_t   *statp,
 858         bhv_vnode_t     *vp)
 859 {
 860         __uint64_t      fakeinos;
 861         xfs_extlen_t    lsize;
 862         xfs_sb_t        *sbp;
 863         unsigned long   s;
 864
 865         sbp = &(mp->m_sb);
 866
 867         statp->f_type = XFS_SB_MAGIC;
 868
 869         xfs_icsb_sync_counters_flags(mp, XFS_ICSB_LAZY_COUNT);
 870         s = XFS_SB_LOCK(mp);
 871         statp->f_bsize = sbp->sb_blocksize;
 872         lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
 873         statp->f_blocks = sbp->sb_dblocks - lsize;
 874         statp->f_bfree = statp->f_bavail =
 875                                 sbp->sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
 876         fakeinos = statp->f_bfree << sbp->sb_inopblog;
 877 #if XFS_BIG_INUMS
 878         fakeinos += mp->m_inoadd;
 879 #endif
 880         statp->f_files =
 881             MIN(sbp->sb_icount + fakeinos, (__uint64_t)XFS_MAXINUMBER);
 882         if (mp->m_maxicount)
 883 #if XFS_BIG_INUMS
 884                 if (!mp->m_inoadd)
 885 #endif
 886                         statp->f_files = min_t(typeof(statp->f_files),
 887                                                 statp->f_files,
 888                                                 mp->m_maxicount);
 889         statp->f_ffree = statp->f_files - (sbp->sb_icount - sbp->sb_ifree);
 890         XFS_SB_UNLOCK(mp, s);
 891
 892         xfs_statvfs_fsid(statp, mp);
 893         statp->f_namelen = MAXNAMELEN - 1;
 894
 895         if (vp)
 896                 XFS_QM_DQSTATVFS(xfs_vtoi(vp), statp);
 897         return 0;
 898 }
 899
 900
 901 /*
 902  * xfs_sync flushes any pending I/O to file system vfsp.
 903  *
 904  * This routine is called by vfs_sync() to make sure that things make it
 905  * out to disk eventually, on sync() system calls to flush out everything,
 906  * and when the file system is unmounted.  For the vfs_sync() case, all
 907  * we really need to do is sync out the log to make all of our meta-data
 908  * updates permanent (except for timestamps).  For calls from pflushd(),
 909  * dirty pages are kept moving by calling pdflush() on the inodes
 910  * containing them.  We also flush the inodes that we can lock without
 911  * sleeping and the superblock if we can lock it without sleeping from
 912  * vfs_sync() so that items at the tail of the log are always moving out.
 913  *
 914  * Flags:
 915  *      SYNC_BDFLUSH - We're being called from vfs_sync() so we don't want
 916  *                     to sleep if we can help it.  All we really need
 917  *                     to do is ensure that the log is synced at least
 918  *                     periodically.  We also push the inodes and
 919  *                     superblock if we can lock them without sleeping
 920  *                      and they are not pinned.
 921  *      SYNC_ATTR    - We need to flush the inodes.  If SYNC_BDFLUSH is not
 922  *                     set, then we really want to lock each inode and flush
 923  *                     it.
 924  *      SYNC_WAIT    - All the flushes that take place in this call should
 925  *                     be synchronous.
 926  *      SYNC_DELWRI  - This tells us to push dirty pages associated with
 927  *                     inodes.  SYNC_WAIT and SYNC_BDFLUSH are used to
 928  *                     determine if they should be flushed sync, async, or
 929  *                     delwri.
 930  *      SYNC_CLOSE   - This flag is passed when the system is being
 931  *                     unmounted.  We should sync and invalidate everything.
 932  *      SYNC_FSDATA  - This indicates that the caller would like to make
 933  *                     sure the superblock is safe on disk.  We can ensure
 934  *                     this by simply making sure the log gets flushed
 935  *                     if SYNC_BDFLUSH is set, and by actually writing it
 936  *                     out otherwise.
 937  *      SYNC_IOWAIT  - The caller wants us to wait for all data I/O to complete
 938  *                     before we return (including direct I/O). Forms the drain
 939  *                     side of the write barrier needed to safely quiesce the
 940  *                     filesystem.
 941  *
 942  */
 943 int
 944 xfs_sync(
 945         xfs_mount_t     *mp,
 946         int             flags)
 947 {
 948         int             error;
 949
 950         /*
 951          * Get the Quota Manager to flush the dquots.
 952          *
 953          * If XFS quota support is not enabled or this filesystem
 954          * instance does not use quotas XFS_QM_DQSYNC will always
 955          * return zero.
 956          */
 957         error = XFS_QM_DQSYNC(mp, flags);
 958         if (error) {
 959                 /*
 960                  * If we got an IO error, we will be shutting down.
 961                  * So, there's nothing more for us to do here.
 962                  */
 963                 ASSERT(error != EIO || XFS_FORCED_SHUTDOWN(mp));
 964                 if (XFS_FORCED_SHUTDOWN(mp))
 965                         return XFS_ERROR(error);
 966         }
 967
 968         if (flags & SYNC_IOWAIT)
 969                 xfs_filestream_flush(mp);
 970
 971         return xfs_syncsub(mp, flags, NULL);
 972 }
 973
 974 /*
 975  * xfs sync routine for internal use
 976  *
 977  * This routine supports all of the flags defined for the generic vfs_sync
 978  * interface as explained above under xfs_sync.
 979  *
 980  */
 981 int
 982 xfs_sync_inodes(
 983         xfs_mount_t     *mp,
 984         int             flags,
 985         int             *bypassed)
 986 {
 987         xfs_inode_t     *ip = NULL;
 988         xfs_inode_t     *ip_next;
 989         xfs_buf_t       *bp;
 990         bhv_vnode_t     *vp = NULL;
 991         int             error;
 992         int             last_error;
 993         uint64_t        fflag;
 994         uint            lock_flags;
 995         uint            base_lock_flags;
 996         boolean_t       mount_locked;
 997         boolean_t       vnode_refed;
 998         int             preempt;
 999         xfs_dinode_t    *dip;
1000         xfs_iptr_t      *ipointer;
1001 #ifdef DEBUG
1002         boolean_t       ipointer_in = B_FALSE;
1003
1004 #define IPOINTER_SET    ipointer_in = B_TRUE
1005 #define IPOINTER_CLR    ipointer_in = B_FALSE
1006 #else
1007 #define IPOINTER_SET
1008 #define IPOINTER_CLR
1009 #endif
1010
1011
1012 /* Insert a marker record into the inode list after inode ip. The list
1013  * must be locked when this is called. After the call the list will no
1014  * longer be locked.
1015  */
1016 #define IPOINTER_INSERT(ip, mp) { \
1017                 ASSERT(ipointer_in == B_FALSE); \
1018                 ipointer->ip_mnext = ip->i_mnext; \
1019                 ipointer->ip_mprev = ip; \
1020                 ip->i_mnext = (xfs_inode_t *)ipointer; \
1021                 ipointer->ip_mnext->i_mprev = (xfs_inode_t *)ipointer; \
1022                 preempt = 0; \
1023                 XFS_MOUNT_IUNLOCK(mp); \
1024                 mount_locked = B_FALSE; \
1025                 IPOINTER_SET; \
1026         }
1027
1028 /* Remove the marker from the inode list. If the marker was the only item
1029  * in the list then there are no remaining inodes and we should zero out
1030  * the whole list. If we are the current head of the list then move the head
1031  * past us.
1032  */
1033 #define IPOINTER_REMOVE(ip, mp) { \
1034                 ASSERT(ipointer_in == B_TRUE); \
1035                 if (ipointer->ip_mnext != (xfs_inode_t *)ipointer) { \
1036                         ip = ipointer->ip_mnext; \
1037                         ip->i_mprev = ipointer->ip_mprev; \
1038                         ipointer->ip_mprev->i_mnext = ip; \
1039                         if (mp->m_inodes == (xfs_inode_t *)ipointer) { \
1040                                 mp->m_inodes = ip; \
1041                         } \
1042                 } else { \
1043                         ASSERT(mp->m_inodes == (xfs_inode_t *)ipointer); \
1044                         mp->m_inodes = NULL; \
1045                         ip = NULL; \
1046                 } \
1047                 IPOINTER_CLR; \
1048         }
1049
1050 #define XFS_PREEMPT_MASK        0x7f
1051
1052         if (bypassed)
1053                 *bypassed = 0;
1054         if (mp->m_flags & XFS_MOUNT_RDONLY)
1055                 return 0;
1056         error = 0;
1057         last_error = 0;
1058         preempt = 0;
1059
1060         /* Allocate a reference marker */
1061         ipointer = (xfs_iptr_t *)kmem_zalloc(sizeof(xfs_iptr_t), KM_SLEEP);
1062
1063         fflag = XFS_B_ASYNC;            /* default is don't wait */
1064         if (flags & (SYNC_BDFLUSH | SYNC_DELWRI))
1065                 fflag = XFS_B_DELWRI;
1066         if (flags & SYNC_WAIT)
1067                 fflag = 0;              /* synchronous overrides all */
1068
1069         base_lock_flags = XFS_ILOCK_SHARED;
1070         if (flags & (SYNC_DELWRI | SYNC_CLOSE)) {
1071                 /*
1072                  * We need the I/O lock if we're going to call any of
1073                  * the flush/inval routines.
1074                  */
1075                 base_lock_flags |= XFS_IOLOCK_SHARED;
1076         }
1077
1078         XFS_MOUNT_ILOCK(mp);
1079
1080         ip = mp->m_inodes;
1081
1082         mount_locked = B_TRUE;
1083         vnode_refed  = B_FALSE;
1084
1085         IPOINTER_CLR;
1086
1087         do {
1088                 ASSERT(ipointer_in == B_FALSE);
1089                 ASSERT(vnode_refed == B_FALSE);
1090
1091                 lock_flags = base_lock_flags;
1092
1093                 /*
1094                  * There were no inodes in the list, just break out
1095                  * of the loop.
1096                  */
1097                 if (ip == NULL) {
1098                         break;
1099                 }
1100
1101                 /*
1102                  * We found another sync thread marker - skip it
1103                  */
1104                 if (ip->i_mount == NULL) {
1105                         ip = ip->i_mnext;
1106                         continue;
1107                 }
1108
1109                 vp = XFS_ITOV_NULL(ip);
1110
1111                 /*
1112                  * If the vnode is gone then this is being torn down,
1113                  * call reclaim if it is flushed, else let regular flush
1114                  * code deal with it later in the loop.
1115                  */
1116
1117                 if (vp == NULL) {
1118                         /* Skip ones already in reclaim */
1119                         if (ip->i_flags & XFS_IRECLAIM) {
1120                                 ip = ip->i_mnext;
1121                                 continue;
1122                         }
1123                         if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0) {
1124                                 ip = ip->i_mnext;
1125                         } else if ((xfs_ipincount(ip) == 0) &&
1126                                     xfs_iflock_nowait(ip)) {
1127                                 IPOINTER_INSERT(ip, mp);
1128
1129                                 xfs_finish_reclaim(ip, 1,
1130                                                 XFS_IFLUSH_DELWRI_ELSE_ASYNC);
1131
1132                                 XFS_MOUNT_ILOCK(mp);
1133                                 mount_locked = B_TRUE;
1134                                 IPOINTER_REMOVE(ip, mp);
1135                         } else {
1136                                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1137                                 ip = ip->i_mnext;
1138                         }
1139                         continue;
1140                 }
1141
1142                 if (VN_BAD(vp)) {
1143                         ip = ip->i_mnext;
1144                         continue;
1145                 }
1146
1147                 if (XFS_FORCED_SHUTDOWN(mp) && !(flags & SYNC_CLOSE)) {
1148                         XFS_MOUNT_IUNLOCK(mp);
1149                         kmem_free(ipointer, sizeof(xfs_iptr_t));
1150                         return 0;
1151                 }
1152
1153                 /*
1154                  * If this is just vfs_sync() or pflushd() calling
1155                  * then we can skip inodes for which it looks like
1156                  * there is nothing to do.  Since we don't have the
1157                  * inode locked this is racy, but these are periodic
1158                  * calls so it doesn't matter.  For the others we want
1159                  * to know for sure, so we at least try to lock them.
1160                  */
1161                 if (flags & SYNC_BDFLUSH) {
1162                         if (((ip->i_itemp == NULL) ||
1163                              !(ip->i_itemp->ili_format.ilf_fields &
1164                                XFS_ILOG_ALL)) &&
1165                             (ip->i_update_core == 0)) {
1166                                 ip = ip->i_mnext;
1167                                 continue;
1168                         }
1169                 }
1170
1171                 /*
1172                  * Try to lock without sleeping.  We're out of order with
1173                  * the inode list lock here, so if we fail we need to drop
1174                  * the mount lock and try again.  If we're called from
1175                  * bdflush() here, then don't bother.
1176                  *
1177                  * The inode lock here actually coordinates with the
1178                  * almost spurious inode lock in xfs_ireclaim() to prevent
1179                  * the vnode we handle here without a reference from
1180                  * being freed while we reference it.  If we lock the inode
1181                  * while it's on the mount list here, then the spurious inode
1182                  * lock in xfs_ireclaim() after the inode is pulled from
1183                  * the mount list will sleep until we release it here.
1184                  * This keeps the vnode from being freed while we reference
1185                  * it.
1186                  */
1187                 if (xfs_ilock_nowait(ip, lock_flags) == 0) {
1188                         if ((flags & SYNC_BDFLUSH) || (vp == NULL)) {
1189                                 ip = ip->i_mnext;
1190                                 continue;
1191                         }
1192
1193                         vp = vn_grab(vp);
1194                         if (vp == NULL) {
1195                                 ip = ip->i_mnext;
1196                                 continue;
1197                         }
1198
1199                         IPOINTER_INSERT(ip, mp);
1200                         xfs_ilock(ip, lock_flags);
1201
1202                         ASSERT(vp == XFS_ITOV(ip));
1203                         ASSERT(ip->i_mount == mp);
1204
1205                         vnode_refed = B_TRUE;
1206                 }
1207
1208                 /* From here on in the loop we may have a marker record
1209                  * in the inode list.
1210                  */
1211
1212                 /*
1213                  * If we have to flush data or wait for I/O completion
1214                  * we need to drop the ilock that we currently hold.
1215                  * If we need to drop the lock, insert a marker if we
1216                  * have not already done so.
1217                  */
1218                 if ((flags & (SYNC_CLOSE|SYNC_IOWAIT)) ||
1219                     ((flags & SYNC_DELWRI) && VN_DIRTY(vp))) {
1220                         if (mount_locked) {
1221                                 IPOINTER_INSERT(ip, mp);
1222                         }
1223                         xfs_iunlock(ip, XFS_ILOCK_SHARED);
1224
1225                         if (flags & SYNC_CLOSE) {
1226                                 /* Shutdown case. Flush and invalidate. */
1227                                 if (XFS_FORCED_SHUTDOWN(mp))
1228                                         xfs_tosspages(ip, 0, -1,
1229                                                              FI_REMAPF);
1230                                 else
1231                                         error = xfs_flushinval_pages(ip,
1232                                                         0, -1, FI_REMAPF);
1233                         } else if ((flags & SYNC_DELWRI) && VN_DIRTY(vp)) {
1234                                 error = xfs_flush_pages(ip, 0,
1235                                                         -1, fflag, FI_NONE);
1236                         }
1237
1238                         /*
1239                          * When freezing, we need to wait ensure all I/O (including direct
1240                          * I/O) is complete to ensure no further data modification can take
1241                          * place after this point
1242                          */
1243                         if (flags & SYNC_IOWAIT)
1244                                 vn_iowait(ip);
1245
1246                         xfs_ilock(ip, XFS_ILOCK_SHARED);
1247                 }
1248
1249                 if (flags & SYNC_BDFLUSH) {
1250                         if ((flags & SYNC_ATTR) &&
1251                             ((ip->i_update_core) ||
1252                              ((ip->i_itemp != NULL) &&
1253                               (ip->i_itemp->ili_format.ilf_fields != 0)))) {
1254
1255                                 /* Insert marker and drop lock if not already
1256                                  * done.
1257                                  */
1258                                 if (mount_locked) {
1259                                         IPOINTER_INSERT(ip, mp);
1260                                 }
1261
1262                                 /*
1263                                  * We don't want the periodic flushing of the
1264                                  * inodes by vfs_sync() to interfere with
1265                                  * I/O to the file, especially read I/O
1266                                  * where it is only the access time stamp
1267                                  * that is being flushed out.  To prevent
1268                                  * long periods where we have both inode
1269                                  * locks held shared here while reading the
1270                                  * inode's buffer in from disk, we drop the
1271                                  * inode lock while reading in the inode
1272                                  * buffer.  We have to release the buffer
1273                                  * and reacquire the inode lock so that they
1274                                  * are acquired in the proper order (inode
1275                                  * locks first).  The buffer will go at the
1276                                  * end of the lru chain, though, so we can
1277                                  * expect it to still be there when we go
1278                                  * for it again in xfs_iflush().
1279                                  */
1280                                 if ((xfs_ipincount(ip) == 0) &&
1281                                     xfs_iflock_nowait(ip)) {
1282
1283                                         xfs_ifunlock(ip);
1284                                         xfs_iunlock(ip, XFS_ILOCK_SHARED);
1285
1286                                         error = xfs_itobp(mp, NULL, ip,
1287                                                           &dip, &bp, 0, 0);
1288                                         if (!error) {
1289                                                 xfs_buf_relse(bp);
1290                                         } else {
1291                                                 /* Bailing out, remove the
1292                                                  * marker and free it.
1293                                                  */
1294                                                 XFS_MOUNT_ILOCK(mp);
1295                                                 IPOINTER_REMOVE(ip, mp);
1296                                                 XFS_MOUNT_IUNLOCK(mp);
1297
1298                                                 ASSERT(!(lock_flags &
1299                                                         XFS_IOLOCK_SHARED));
1300
1301                                                 kmem_free(ipointer,
1302                                                         sizeof(xfs_iptr_t));
1303                                                 return (0);
1304                                         }
1305
1306                                         /*
1307                                          * Since we dropped the inode lock,
1308                                          * the inode may have been reclaimed.
1309                                          * Therefore, we reacquire the mount
1310                                          * lock and check to see if we were the
1311                                          * inode reclaimed. If this happened
1312                                          * then the ipointer marker will no
1313                                          * longer point back at us. In this
1314                                          * case, move ip along to the inode
1315                                          * after the marker, remove the marker
1316                                          * and continue.
1317                                          */
1318                                         XFS_MOUNT_ILOCK(mp);
1319                                         mount_locked = B_TRUE;
1320
1321                                         if (ip != ipointer->ip_mprev) {
1322                                                 IPOINTER_REMOVE(ip, mp);
1323
1324                                                 ASSERT(!vnode_refed);
1325                                                 ASSERT(!(lock_flags &
1326                                                         XFS_IOLOCK_SHARED));
1327                                                 continue;
1328                                         }
1329
1330                                         ASSERT(ip->i_mount == mp);
1331
1332                                         if (xfs_ilock_nowait(ip,
1333                                                     XFS_ILOCK_SHARED) == 0) {
1334                                                 ASSERT(ip->i_mount == mp);
1335                                                 /*
1336                                                  * We failed to reacquire
1337                                                  * the inode lock without
1338                                                  * sleeping, so just skip
1339                                                  * the inode for now.  We
1340                                                  * clear the ILOCK bit from
1341                                                  * the lock_flags so that we
1342                                                  * won't try to drop a lock
1343                                                  * we don't hold below.
1344                                                  */
1345                                                 lock_flags &= ~XFS_ILOCK_SHARED;
1346                                                 IPOINTER_REMOVE(ip_next, mp);
1347                                         } else if ((xfs_ipincount(ip) == 0) &&
1348                                                    xfs_iflock_nowait(ip)) {
1349                                                 ASSERT(ip->i_mount == mp);
1350                                                 /*
1351                                                  * Since this is vfs_sync()
1352                                                  * calling we only flush the
1353                                                  * inode out if we can lock
1354                                                  * it without sleeping and
1355                                                  * it is not pinned.  Drop
1356                                                  * the mount lock here so
1357                                                  * that we don't hold it for
1358                                                  * too long. We already have
1359                                                  * a marker in the list here.
1360                                                  */
1361                                                 XFS_MOUNT_IUNLOCK(mp);
1362                                                 mount_locked = B_FALSE;
1363                                                 error = xfs_iflush(ip,
1364                                                            XFS_IFLUSH_DELWRI);
1365                                         } else {
1366                                                 ASSERT(ip->i_mount == mp);
1367                                                 IPOINTER_REMOVE(ip_next, mp);
1368                                         }
1369                                 }
1370
1371                         }
1372
1373                 } else {
1374                         if ((flags & SYNC_ATTR) &&
1375                             ((ip->i_update_core) ||
1376                              ((ip->i_itemp != NULL) &&
1377                               (ip->i_itemp->ili_format.ilf_fields != 0)))) {
1378                                 if (mount_locked) {
1379                                         IPOINTER_INSERT(ip, mp);
1380                                 }
1381
1382                                 if (flags & SYNC_WAIT) {
1383                                         xfs_iflock(ip);
1384                                         error = xfs_iflush(ip,
1385                                                            XFS_IFLUSH_SYNC);
1386                                 } else {
1387                                         /*
1388                                          * If we can't acquire the flush
1389                                          * lock, then the inode is already
1390                                          * being flushed so don't bother
1391                                          * waiting.  If we can lock it then
1392                                          * do a delwri flush so we can
1393                                          * combine multiple inode flushes
1394                                          * in each disk write.
1395                                          */
1396                                         if (xfs_iflock_nowait(ip)) {
1397                                                 error = xfs_iflush(ip,
1398                                                            XFS_IFLUSH_DELWRI);
1399                                         }
1400                                         else if (bypassed)
1401                                                 (*bypassed)++;
1402                                 }
1403                         }
1404                 }
1405
1406                 if (lock_flags != 0) {
1407                         xfs_iunlock(ip, lock_flags);
1408                 }
1409
1410                 if (vnode_refed) {
1411                         /*
1412                          * If we had to take a reference on the vnode
1413                          * above, then wait until after we've unlocked
1414                          * the inode to release the reference.  This is
1415                          * because we can be already holding the inode
1416                          * lock when VN_RELE() calls xfs_inactive().
1417                          *
1418                          * Make sure to drop the mount lock before calling
1419                          * VN_RELE() so that we don't trip over ourselves if
1420                          * we have to go for the mount lock again in the
1421                          * inactive code.
1422                          */
1423                         if (mount_locked) {
1424                                 IPOINTER_INSERT(ip, mp);
1425                         }
1426
1427                         VN_RELE(vp);
1428
1429                         vnode_refed = B_FALSE;
1430                 }
1431
1432                 if (error) {
1433                         last_error = error;
1434                 }
1435
1436                 /*
1437                  * bail out if the filesystem is corrupted.
1438                  */
1439                 if (error == EFSCORRUPTED)  {
1440                         if (!mount_locked) {
1441                                 XFS_MOUNT_ILOCK(mp);
1442                                 IPOINTER_REMOVE(ip, mp);
1443                         }
1444                         XFS_MOUNT_IUNLOCK(mp);
1445                         ASSERT(ipointer_in == B_FALSE);
1446                         kmem_free(ipointer, sizeof(xfs_iptr_t));
1447                         return XFS_ERROR(error);
1448                 }
1449
1450                 /* Let other threads have a chance at the mount lock
1451                  * if we have looped many times without dropping the
1452                  * lock.
1453                  */
1454                 if ((++preempt & XFS_PREEMPT_MASK) == 0) {
1455                         if (mount_locked) {
1456                                 IPOINTER_INSERT(ip, mp);
1457                         }
1458                 }
1459
1460                 if (mount_locked == B_FALSE) {
1461                         XFS_MOUNT_ILOCK(mp);
1462                         mount_locked = B_TRUE;
1463                         IPOINTER_REMOVE(ip, mp);
1464                         continue;
1465                 }
1466
1467                 ASSERT(ipointer_in == B_FALSE);
1468                 ip = ip->i_mnext;
1469
1470         } while (ip != mp->m_inodes);
1471
1472         XFS_MOUNT_IUNLOCK(mp);
1473
1474         ASSERT(ipointer_in == B_FALSE);
1475
1476         kmem_free(ipointer, sizeof(xfs_iptr_t));
1477         return XFS_ERROR(last_error);
1478 }
1479
1480 /*
1481  * xfs sync routine for internal use
1482  *
1483  * This routine supports all of the flags defined for the generic vfs_sync
1484  * interface as explained above under xfs_sync.
1485  *
1486  */
1487 int
1488 xfs_syncsub(
1489         xfs_mount_t     *mp,
1490         int             flags,
1491         int             *bypassed)
1492 {
1493         int             error = 0;
1494         int             last_error = 0;
1495         uint            log_flags = XFS_LOG_FORCE;
1496         xfs_buf_t       *bp;
1497         xfs_buf_log_item_t      *bip;
1498
1499         /*
1500          * Sync out the log.  This ensures that the log is periodically
1501          * flushed even if there is not enough activity to fill it up.
1502          */
1503         if (flags & SYNC_WAIT)
1504                 log_flags |= XFS_LOG_SYNC;
1505
1506         xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1507
1508         if (flags & (SYNC_ATTR|SYNC_DELWRI)) {
1509                 if (flags & SYNC_BDFLUSH)
1510                         xfs_finish_reclaim_all(mp, 1);
1511                 else
1512                         error = xfs_sync_inodes(mp, flags, bypassed);
1513         }
1514
1515         /*
1516          * Flushing out dirty data above probably generated more
1517          * log activity, so if this isn't vfs_sync() then flush
1518          * the log again.
1519          */
1520         if (flags & SYNC_DELWRI) {
1521                 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1522         }
1523
1524         if (flags & SYNC_FSDATA) {
1525                 /*
1526                  * If this is vfs_sync() then only sync the superblock
1527                  * if we can lock it without sleeping and it is not pinned.
1528                  */
1529                 if (flags & SYNC_BDFLUSH) {
1530                         bp = xfs_getsb(mp, XFS_BUF_TRYLOCK);
1531                         if (bp != NULL) {
1532                                 bip = XFS_BUF_FSPRIVATE(bp,xfs_buf_log_item_t*);
1533                                 if ((bip != NULL) &&
1534                                     xfs_buf_item_dirty(bip)) {
1535                                         if (!(XFS_BUF_ISPINNED(bp))) {
1536                                                 XFS_BUF_ASYNC(bp);
1537                                                 error = xfs_bwrite(mp, bp);
1538                                         } else {
1539                                                 xfs_buf_relse(bp);
1540                                         }
1541                                 } else {
1542                                         xfs_buf_relse(bp);
1543                                 }
1544                         }
1545                 } else {
1546                         bp = xfs_getsb(mp, 0);
1547                         /*
1548                          * If the buffer is pinned then push on the log so
1549                          * we won't get stuck waiting in the write for
1550                          * someone, maybe ourselves, to flush the log.
1551                          * Even though we just pushed the log above, we
1552                          * did not have the superblock buffer locked at
1553                          * that point so it can become pinned in between
1554                          * there and here.
1555                          */
1556                         if (XFS_BUF_ISPINNED(bp))
1557                                 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
1558                         if (flags & SYNC_WAIT)
1559                                 XFS_BUF_UNASYNC(bp);
1560                         else
1561                                 XFS_BUF_ASYNC(bp);
1562                         error = xfs_bwrite(mp, bp);
1563                 }
1564                 if (error) {
1565                         last_error = error;
1566                 }
1567         }
1568
1569         /*
1570          * If this is the periodic sync, then kick some entries out of
1571          * the reference cache.  This ensures that idle entries are
1572          * eventually kicked out of the cache.
1573          */
1574         if (flags & SYNC_REFCACHE) {
1575                 if (flags & SYNC_WAIT)
1576                         xfs_refcache_purge_mp(mp);
1577                 else
1578                         xfs_refcache_purge_some(mp);
1579         }
1580
1581         /*
1582          * If asked, update the disk superblock with incore counter values if we
1583          * are using non-persistent counters so that they don't get too far out
1584          * of sync if we crash or get a forced shutdown. We don't want to force
1585          * this to disk, just get a transaction into the iclogs....
1586          */
1587         if (flags & SYNC_SUPER)
1588                 xfs_log_sbcount(mp, 0);
1589
1590         /*
1591          * Now check to see if the log needs a "dummy" transaction.
1592          */
1593
1594         if (!(flags & SYNC_REMOUNT) && xfs_log_need_covered(mp)) {
1595                 xfs_trans_t *tp;
1596                 xfs_inode_t *ip;
1597
1598                 /*
1599                  * Put a dummy transaction in the log to tell
1600                  * recovery that all others are OK.
1601                  */
1602                 tp = xfs_trans_alloc(mp, XFS_TRANS_DUMMY1);
1603                 if ((error = xfs_trans_reserve(tp, 0,
1604                                 XFS_ICHANGE_LOG_RES(mp),
1605                                 0, 0, 0)))  {
1606                         xfs_trans_cancel(tp, 0);
1607                         return error;
1608                 }
1609
1610                 ip = mp->m_rootip;
1611                 xfs_ilock(ip, XFS_ILOCK_EXCL);
1612
1613                 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1614                 xfs_trans_ihold(tp, ip);
1615                 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1616                 error = xfs_trans_commit(tp, 0);
1617                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1618                 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1619         }
1620
1621         /*
1622          * When shutting down, we need to insure that the AIL is pushed
1623          * to disk or the filesystem can appear corrupt from the PROM.
1624          */
1625         if ((flags & (SYNC_CLOSE|SYNC_WAIT)) == (SYNC_CLOSE|SYNC_WAIT)) {
1626                 XFS_bflush(mp->m_ddev_targp);
1627                 if (mp->m_rtdev_targp) {
1628                         XFS_bflush(mp->m_rtdev_targp);
1629                 }
1630         }
1631
1632         return XFS_ERROR(last_error);
1633 }
1634
1635 /*
1636  * xfs_vget - called by DMAPI and NFSD to get vnode from file handle
1637  */
1638 int
1639 xfs_vget(
1640         xfs_mount_t     *mp,
1641         bhv_vnode_t     **vpp,
1642         fid_t           *fidp)
1643 {
1644         xfs_fid_t       *xfid = (struct xfs_fid *)fidp;
1645         xfs_inode_t     *ip;
1646         int             error;
1647         xfs_ino_t       ino;
1648         unsigned int    igen;
1649
1650         /*
1651          * Invalid.  Since handles can be created in user space and passed in
1652          * via gethandle(), this is not cause for a panic.
1653          */
1654         if (xfid->xfs_fid_len != sizeof(*xfid) - sizeof(xfid->xfs_fid_len))
1655                 return XFS_ERROR(EINVAL);
1656
1657         ino  = xfid->xfs_fid_ino;
1658         igen = xfid->xfs_fid_gen;
1659
1660         /*
1661          * NFS can sometimes send requests for ino 0.  Fail them gracefully.
1662          */
1663         if (ino == 0)
1664                 return XFS_ERROR(ESTALE);
1665
1666         error = xfs_iget(mp, NULL, ino, 0, XFS_ILOCK_SHARED, &ip, 0);
1667         if (error) {
1668                 *vpp = NULL;
1669                 return error;
1670         }
1671
1672         if (ip == NULL) {
1673                 *vpp = NULL;
1674                 return XFS_ERROR(EIO);
1675         }
1676
1677         if (ip->i_d.di_mode == 0 || ip->i_d.di_gen != igen) {
1678                 xfs_iput_new(ip, XFS_ILOCK_SHARED);
1679                 *vpp = NULL;
1680                 return XFS_ERROR(ENOENT);
1681         }
1682
1683         *vpp = XFS_ITOV(ip);
1684         xfs_iunlock(ip, XFS_ILOCK_SHARED);
1685         return 0;
1686 }
1687
1688
1689 #define MNTOPT_LOGBUFS  "logbufs"       /* number of XFS log buffers */
1690 #define MNTOPT_LOGBSIZE "logbsize"      /* size of XFS log buffers */
1691 #define MNTOPT_LOGDEV   "logdev"        /* log device */
1692 #define MNTOPT_RTDEV    "rtdev"         /* realtime I/O device */
1693 #define MNTOPT_BIOSIZE  "biosize"       /* log2 of preferred buffered io size */
1694 #define MNTOPT_WSYNC    "wsync"         /* safe-mode nfs compatible mount */
1695 #define MNTOPT_INO64    "ino64"         /* force inodes into 64-bit range */
1696 #define MNTOPT_NOALIGN  "noalign"       /* turn off stripe alignment */
1697 #define MNTOPT_SWALLOC  "swalloc"       /* turn on stripe width allocation */
1698 #define MNTOPT_SUNIT    "sunit"         /* data volume stripe unit */
1699 #define MNTOPT_SWIDTH   "swidth"        /* data volume stripe width */
1700 #define MNTOPT_NOUUID   "nouuid"        /* ignore filesystem UUID */
1701 #define MNTOPT_MTPT     "mtpt"          /* filesystem mount point */
1702 #define MNTOPT_GRPID    "grpid"         /* group-ID from parent directory */
1703 #define MNTOPT_NOGRPID  "nogrpid"       /* group-ID from current process */
1704 #define MNTOPT_BSDGROUPS    "bsdgroups"    /* group-ID from parent directory */
1705 #define MNTOPT_SYSVGROUPS   "sysvgroups"   /* group-ID from current process */
1706 #define MNTOPT_ALLOCSIZE    "allocsize"    /* preferred allocation size */
1707 #define MNTOPT_NORECOVERY   "norecovery"   /* don't run XFS recovery */
1708 #define MNTOPT_BARRIER  "barrier"       /* use writer barriers for log write and
1709                                          * unwritten extent conversion */
1710 #define MNTOPT_NOBARRIER "nobarrier"    /* .. disable */
1711 #define MNTOPT_OSYNCISOSYNC "osyncisosync" /* o_sync is REALLY o_sync */
1712 #define MNTOPT_64BITINODE   "inode64"   /* inodes can be allocated anywhere */
1713 #define MNTOPT_IKEEP    "ikeep"         /* do not free empty inode clusters */
1714 #define MNTOPT_NOIKEEP  "noikeep"       /* free empty inode clusters */
1715 #define MNTOPT_LARGEIO     "largeio"    /* report large I/O sizes in stat() */
1716 #define MNTOPT_NOLARGEIO   "nolargeio"  /* do not report large I/O sizes
1717                                          * in stat(). */
1718 #define MNTOPT_ATTR2    "attr2"         /* do use attr2 attribute format */
1719 #define MNTOPT_NOATTR2  "noattr2"       /* do not use attr2 attribute format */
1720 #define MNTOPT_FILESTREAM  "filestreams" /* use filestreams allocator */
1721 #define MNTOPT_QUOTA    "quota"         /* disk quotas (user) */
1722 #define MNTOPT_NOQUOTA  "noquota"       /* no quotas */
1723 #define MNTOPT_USRQUOTA "usrquota"      /* user quota enabled */
1724 #define MNTOPT_GRPQUOTA "grpquota"      /* group quota enabled */
1725 #define MNTOPT_PRJQUOTA "prjquota"      /* project quota enabled */
1726 #define MNTOPT_UQUOTA   "uquota"        /* user quota (IRIX variant) */
1727 #define MNTOPT_GQUOTA   "gquota"        /* group quota (IRIX variant) */
1728 #define MNTOPT_PQUOTA   "pquota"        /* project quota (IRIX variant) */
1729 #define MNTOPT_UQUOTANOENF "uqnoenforce"/* user quota limit enforcement */
1730 #define MNTOPT_GQUOTANOENF "gqnoenforce"/* group quota limit enforcement */
1731 #define MNTOPT_PQUOTANOENF "pqnoenforce"/* project quota limit enforcement */
1732 #define MNTOPT_QUOTANOENF  "qnoenforce" /* same as uqnoenforce */
1733 #define MNTOPT_DMAPI    "dmapi"         /* DMI enabled (DMAPI / XDSM) */
1734 #define MNTOPT_XDSM     "xdsm"          /* DMI enabled (DMAPI / XDSM) */
1735 #define MNTOPT_DMI      "dmi"           /* DMI enabled (DMAPI / XDSM) */
1736
1737 STATIC unsigned long
1738 suffix_strtoul(char *s, char **endp, unsigned int base)
1739 {
1740         int     last, shift_left_factor = 0;
1741         char    *value = s;
1742
1743         last = strlen(value) - 1;
1744         if (value[last] == 'K' || value[last] == 'k') {
1745                 shift_left_factor = 10;
1746                 value[last] = '\0';
1747         }
1748         if (value[last] == 'M' || value[last] == 'm') {
1749                 shift_left_factor = 20;
1750                 value[last] = '\0';
1751         }
1752         if (value[last] == 'G' || value[last] == 'g') {
1753                 shift_left_factor = 30;
1754                 value[last] = '\0';
1755         }
1756
1757         return simple_strtoul((const char *)s, endp, base) << shift_left_factor;
1758 }
1759
1760 int
1761 xfs_parseargs(
1762         struct xfs_mount        *mp,
1763         char                    *options,
1764         struct xfs_mount_args   *args,
1765         int                     update)
1766 {
1767         char                    *this_char, *value, *eov;
1768         int                     dsunit, dswidth, vol_dsunit, vol_dswidth;
1769         int                     iosize;
1770
1771         /*
1772          * Applications using DMI filesystems often expect the
1773          * inode generation number to be monotonically increasing.
1774          * If we delete inode chunks we break this assumption, so
1775          * keep unused inode chunks on disk for DMI filesystems
1776          * until we come up with a better solution.
1777          * Note that if "ikeep" or "noikeep" mount options are
1778          * supplied, then they are honored.
1779          */
1780         if (!(args->flags & XFSMNT_DMAPI))
1781                 args->flags |= XFSMNT_IDELETE;
1782
1783         args->flags |= XFSMNT_BARRIER;
1784         args->flags2 |= XFSMNT2_COMPAT_IOSIZE;
1785
1786         if (!options)
1787                 goto done;
1788
1789         iosize = dsunit = dswidth = vol_dsunit = vol_dswidth = 0;
1790
1791         while ((this_char = strsep(&options, ",")) != NULL) {
1792                 if (!*this_char)
1793                         continue;
1794                 if ((value = strchr(this_char, '=')) != NULL)
1795                         *value++ = 0;
1796
1797                 if (!strcmp(this_char, MNTOPT_LOGBUFS)) {
1798                         if (!value || !*value) {
1799                                 cmn_err(CE_WARN,
1800                                         "XFS: %s option requires an argument",
1801                                         this_char);
1802                                 return EINVAL;
1803                         }
1804                         args->logbufs = simple_strtoul(value, &eov, 10);
1805                 } else if (!strcmp(this_char, MNTOPT_LOGBSIZE)) {
1806                         if (!value || !*value) {
1807                                 cmn_err(CE_WARN,
1808                                         "XFS: %s option requires an argument",
1809                                         this_char);
1810                                 return EINVAL;
1811                         }
1812                         args->logbufsize = suffix_strtoul(value, &eov, 10);
1813                 } else if (!strcmp(this_char, MNTOPT_LOGDEV)) {
1814                         if (!value || !*value) {
1815                                 cmn_err(CE_WARN,
1816                                         "XFS: %s option requires an argument",
1817                                         this_char);
1818                                 return EINVAL;
1819                         }
1820                         strncpy(args->logname, value, MAXNAMELEN);
1821                 } else if (!strcmp(this_char, MNTOPT_MTPT)) {
1822                         if (!value || !*value) {
1823                                 cmn_err(CE_WARN,
1824                                         "XFS: %s option requires an argument",
1825                                         this_char);
1826                                 return EINVAL;
1827                         }
1828                         strncpy(args->mtpt, value, MAXNAMELEN);
1829                 } else if (!strcmp(this_char, MNTOPT_RTDEV)) {
1830                         if (!value || !*value) {
1831                                 cmn_err(CE_WARN,
1832                                         "XFS: %s option requires an argument",
1833                                         this_char);
1834                                 return EINVAL;
1835                         }
1836                         strncpy(args->rtname, value, MAXNAMELEN);
1837                 } else if (!strcmp(this_char, MNTOPT_BIOSIZE)) {
1838                         if (!value || !*value) {
1839                                 cmn_err(CE_WARN,
1840                                         "XFS: %s option requires an argument",
1841                                         this_char);
1842                                 return EINVAL;
1843                         }
1844                         iosize = simple_strtoul(value, &eov, 10);
1845                         args->flags |= XFSMNT_IOSIZE;
1846                         args->iosizelog = (uint8_t) iosize;
1847                 } else if (!strcmp(this_char, MNTOPT_ALLOCSIZE)) {
1848                         if (!value || !*value) {
1849                                 cmn_err(CE_WARN,
1850                                         "XFS: %s option requires an argument",
1851                                         this_char);
1852                                 return EINVAL;
1853                         }
1854                         iosize = suffix_strtoul(value, &eov, 10);
1855                         args->flags |= XFSMNT_IOSIZE;
1856                         args->iosizelog = ffs(iosize) - 1;
1857                 } else if (!strcmp(this_char, MNTOPT_GRPID) ||
1858                            !strcmp(this_char, MNTOPT_BSDGROUPS)) {
1859                         mp->m_flags |= XFS_MOUNT_GRPID;
1860                 } else if (!strcmp(this_char, MNTOPT_NOGRPID) ||
1861                            !strcmp(this_char, MNTOPT_SYSVGROUPS)) {
1862                         mp->m_flags &= ~XFS_MOUNT_GRPID;
1863                 } else if (!strcmp(this_char, MNTOPT_WSYNC)) {
1864                         args->flags |= XFSMNT_WSYNC;
1865                 } else if (!strcmp(this_char, MNTOPT_OSYNCISOSYNC)) {
1866                         args->flags |= XFSMNT_OSYNCISOSYNC;
1867                 } else if (!strcmp(this_char, MNTOPT_NORECOVERY)) {
1868                         args->flags |= XFSMNT_NORECOVERY;
1869                 } else if (!strcmp(this_char, MNTOPT_INO64)) {
1870                         args->flags |= XFSMNT_INO64;
1871 #if !XFS_BIG_INUMS
1872                         cmn_err(CE_WARN,
1873                                 "XFS: %s option not allowed on this system",
1874                                 this_char);
1875                         return EINVAL;
1876 #endif
1877                 } else if (!strcmp(this_char, MNTOPT_NOALIGN)) {
1878                         args->flags |= XFSMNT_NOALIGN;
1879                 } else if (!strcmp(this_char, MNTOPT_SWALLOC)) {
1880                         args->flags |= XFSMNT_SWALLOC;
1881                 } else if (!strcmp(this_char, MNTOPT_SUNIT)) {
1882                         if (!value || !*value) {
1883                                 cmn_err(CE_WARN,
1884                                         "XFS: %s option requires an argument",
1885                                         this_char);
1886                                 return EINVAL;
1887                         }
1888                         dsunit = simple_strtoul(value, &eov, 10);
1889                 } else if (!strcmp(this_char, MNTOPT_SWIDTH)) {
1890                         if (!value || !*value) {
1891                                 cmn_err(CE_WARN,
1892                                         "XFS: %s option requires an argument",
1893                                         this_char);
1894                                 return EINVAL;
1895                         }
1896                         dswidth = simple_strtoul(value, &eov, 10);
1897                 } else if (!strcmp(this_char, MNTOPT_64BITINODE)) {
1898                         args->flags &= ~XFSMNT_32BITINODES;
1899 #if !XFS_BIG_INUMS
1900                         cmn_err(CE_WARN,
1901                                 "XFS: %s option not allowed on this system",
1902                                 this_char);
1903                         return EINVAL;
1904 #endif
1905                 } else if (!strcmp(this_char, MNTOPT_NOUUID)) {
1906                         args->flags |= XFSMNT_NOUUID;
1907                 } else if (!strcmp(this_char, MNTOPT_BARRIER)) {
1908                         args->flags |= XFSMNT_BARRIER;
1909                 } else if (!strcmp(this_char, MNTOPT_NOBARRIER)) {
1910                         args->flags &= ~XFSMNT_BARRIER;
1911                 } else if (!strcmp(this_char, MNTOPT_IKEEP)) {
1912                         args->flags &= ~XFSMNT_IDELETE;
1913                 } else if (!strcmp(this_char, MNTOPT_NOIKEEP)) {
1914                         args->flags |= XFSMNT_IDELETE;
1915                 } else if (!strcmp(this_char, MNTOPT_LARGEIO)) {
1916                         args->flags2 &= ~XFSMNT2_COMPAT_IOSIZE;
1917                 } else if (!strcmp(this_char, MNTOPT_NOLARGEIO)) {
1918                         args->flags2 |= XFSMNT2_COMPAT_IOSIZE;
1919                 } else if (!strcmp(this_char, MNTOPT_ATTR2)) {
1920                         args->flags |= XFSMNT_ATTR2;
1921                 } else if (!strcmp(this_char, MNTOPT_NOATTR2)) {
1922                         args->flags &= ~XFSMNT_ATTR2;
1923                 } else if (!strcmp(this_char, MNTOPT_FILESTREAM)) {
1924                         args->flags2 |= XFSMNT2_FILESTREAMS;
1925                 } else if (!strcmp(this_char, MNTOPT_NOQUOTA)) {
1926                         args->flags &= ~(XFSMNT_UQUOTAENF|XFSMNT_UQUOTA);
1927                         args->flags &= ~(XFSMNT_GQUOTAENF|XFSMNT_GQUOTA);
1928                 } else if (!strcmp(this_char, MNTOPT_QUOTA) ||
1929                            !strcmp(this_char, MNTOPT_UQUOTA) ||
1930                            !strcmp(this_char, MNTOPT_USRQUOTA)) {
1931                         args->flags |= XFSMNT_UQUOTA | XFSMNT_UQUOTAENF;
1932                 } else if (!strcmp(this_char, MNTOPT_QUOTANOENF) ||
1933                            !strcmp(this_char, MNTOPT_UQUOTANOENF)) {
1934                         args->flags |= XFSMNT_UQUOTA;
1935                         args->flags &= ~XFSMNT_UQUOTAENF;
1936                 } else if (!strcmp(this_char, MNTOPT_PQUOTA) ||
1937                            !strcmp(this_char, MNTOPT_PRJQUOTA)) {
1938                         args->flags |= XFSMNT_PQUOTA | XFSMNT_PQUOTAENF;
1939                 } else if (!strcmp(this_char, MNTOPT_PQUOTANOENF)) {
1940                         args->flags |= XFSMNT_PQUOTA;
1941                         args->flags &= ~XFSMNT_PQUOTAENF;
1942                 } else if (!strcmp(this_char, MNTOPT_GQUOTA) ||
1943                            !strcmp(this_char, MNTOPT_GRPQUOTA)) {
1944                         args->flags |= XFSMNT_GQUOTA | XFSMNT_GQUOTAENF;
1945                 } else if (!strcmp(this_char, MNTOPT_GQUOTANOENF)) {
1946                         args->flags |= XFSMNT_GQUOTA;
1947                         args->flags &= ~XFSMNT_GQUOTAENF;
1948                 } else if (!strcmp(this_char, MNTOPT_DMAPI)) {
1949                         args->flags |= XFSMNT_DMAPI;
1950                 } else if (!strcmp(this_char, MNTOPT_XDSM)) {
1951                         args->flags |= XFSMNT_DMAPI;
1952                 } else if (!strcmp(this_char, MNTOPT_DMI)) {
1953                         args->flags |= XFSMNT_DMAPI;
1954                 } else if (!strcmp(this_char, "ihashsize")) {
1955                         cmn_err(CE_WARN,
1956         "XFS: ihashsize no longer used, option is deprecated.");
1957                 } else if (!strcmp(this_char, "osyncisdsync")) {
1958                         /* no-op, this is now the default */
1959                         cmn_err(CE_WARN,
1960         "XFS: osyncisdsync is now the default, option is deprecated.");
1961                 } else if (!strcmp(this_char, "irixsgid")) {
1962                         cmn_err(CE_WARN,
1963         "XFS: irixsgid is now a sysctl(2) variable, option is deprecated.");
1964                 } else {
1965                         cmn_err(CE_WARN,
1966                                 "XFS: unknown mount option [%s].", this_char);
1967                         return EINVAL;
1968                 }
1969         }
1970
1971         if (args->flags & XFSMNT_NORECOVERY) {
1972                 if ((mp->m_flags & XFS_MOUNT_RDONLY) == 0) {
1973                         cmn_err(CE_WARN,
1974                                 "XFS: no-recovery mounts must be read-only.");
1975                         return EINVAL;
1976                 }
1977         }
1978
1979         if ((args->flags & XFSMNT_NOALIGN) && (dsunit || dswidth)) {
1980                 cmn_err(CE_WARN,
1981         "XFS: sunit and swidth options incompatible with the noalign option");
1982                 return EINVAL;
1983         }
1984
1985         if ((args->flags & XFSMNT_GQUOTA) && (args->flags & XFSMNT_PQUOTA)) {
1986                 cmn_err(CE_WARN,
1987                         "XFS: cannot mount with both project and group quota");
1988                 return EINVAL;
1989         }
1990
1991         if ((args->flags & XFSMNT_DMAPI) && *args->mtpt == '\0') {
1992                 printk("XFS: %s option needs the mount point option as well\n",
1993                         MNTOPT_DMAPI);
1994                 return EINVAL;
1995         }
1996
1997         if ((dsunit && !dswidth) || (!dsunit && dswidth)) {
1998                 cmn_err(CE_WARN,
1999                         "XFS: sunit and swidth must be specified together");
2000                 return EINVAL;
2001         }
2002
2003         if (dsunit && (dswidth % dsunit != 0)) {
2004                 cmn_err(CE_WARN,
2005         "XFS: stripe width (%d) must be a multiple of the stripe unit (%d)",
2006                         dswidth, dsunit);
2007                 return EINVAL;
2008         }
2009
2010         if ((args->flags & XFSMNT_NOALIGN) != XFSMNT_NOALIGN) {
2011                 if (dsunit) {
2012                         args->sunit = dsunit;
2013                         args->flags |= XFSMNT_RETERR;
2014                 } else {
2015                         args->sunit = vol_dsunit;
2016                 }
2017                 dswidth ? (args->swidth = dswidth) :
2018                           (args->swidth = vol_dswidth);
2019         } else {
2020                 args->sunit = args->swidth = 0;
2021         }
2022
2023 done:
2024         if (args->flags & XFSMNT_32BITINODES)
2025                 mp->m_flags |= XFS_MOUNT_SMALL_INUMS;
2026         if (args->flags2)
2027                 args->flags |= XFSMNT_FLAGS2;
2028         return 0;
2029 }
2030
2031 int
2032 xfs_showargs(
2033         struct xfs_mount        *mp,
2034         struct seq_file         *m)
2035 {
2036         static struct proc_xfs_info {
2037                 int     flag;
2038                 char    *str;
2039         } xfs_info[] = {
2040                 /* the few simple ones we can get from the mount struct */
2041                 { XFS_MOUNT_WSYNC,              "," MNTOPT_WSYNC },
2042                 { XFS_MOUNT_INO64,              "," MNTOPT_INO64 },
2043                 { XFS_MOUNT_NOALIGN,            "," MNTOPT_NOALIGN },
2044                 { XFS_MOUNT_SWALLOC,            "," MNTOPT_SWALLOC },
2045                 { XFS_MOUNT_NOUUID,             "," MNTOPT_NOUUID },
2046                 { XFS_MOUNT_NORECOVERY,         "," MNTOPT_NORECOVERY },
2047                 { XFS_MOUNT_OSYNCISOSYNC,       "," MNTOPT_OSYNCISOSYNC },
2048                 { 0, NULL }
2049         };
2050         struct proc_xfs_info    *xfs_infop;
2051
2052         for (xfs_infop = xfs_info; xfs_infop->flag; xfs_infop++) {
2053                 if (mp->m_flags & xfs_infop->flag)
2054                         seq_puts(m, xfs_infop->str);
2055         }
2056
2057         if (mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)
2058                 seq_printf(m, "," MNTOPT_ALLOCSIZE "=%dk",
2059                                 (int)(1 << mp->m_writeio_log) >> 10);
2060
2061         if (mp->m_logbufs > 0)
2062                 seq_printf(m, "," MNTOPT_LOGBUFS "=%d", mp->m_logbufs);
2063         if (mp->m_logbsize > 0)
2064                 seq_printf(m, "," MNTOPT_LOGBSIZE "=%dk", mp->m_logbsize >> 10);
2065
2066         if (mp->m_logname)
2067                 seq_printf(m, "," MNTOPT_LOGDEV "=%s", mp->m_logname);
2068         if (mp->m_rtname)
2069                 seq_printf(m, "," MNTOPT_RTDEV "=%s", mp->m_rtname);
2070
2071         if (mp->m_dalign > 0)
2072                 seq_printf(m, "," MNTOPT_SUNIT "=%d",
2073                                 (int)XFS_FSB_TO_BB(mp, mp->m_dalign));
2074         if (mp->m_swidth > 0)
2075                 seq_printf(m, "," MNTOPT_SWIDTH "=%d",
2076                                 (int)XFS_FSB_TO_BB(mp, mp->m_swidth));
2077
2078         if (!(mp->m_flags & XFS_MOUNT_IDELETE))
2079                 seq_printf(m, "," MNTOPT_IKEEP);
2080         if (!(mp->m_flags & XFS_MOUNT_COMPAT_IOSIZE))
2081                 seq_printf(m, "," MNTOPT_LARGEIO);
2082
2083         if (!(mp->m_flags & XFS_MOUNT_SMALL_INUMS))
2084                 seq_printf(m, "," MNTOPT_64BITINODE);
2085         if (mp->m_flags & XFS_MOUNT_GRPID)
2086                 seq_printf(m, "," MNTOPT_GRPID);
2087
2088         if (mp->m_qflags & XFS_UQUOTA_ACCT) {
2089                 if (mp->m_qflags & XFS_UQUOTA_ENFD)
2090                         seq_puts(m, "," MNTOPT_USRQUOTA);
2091                 else
2092                         seq_puts(m, "," MNTOPT_UQUOTANOENF);
2093         }
2094
2095         if (mp->m_qflags & XFS_PQUOTA_ACCT) {
2096                 if (mp->m_qflags & XFS_OQUOTA_ENFD)
2097                         seq_puts(m, "," MNTOPT_PRJQUOTA);
2098                 else
2099                         seq_puts(m, "," MNTOPT_PQUOTANOENF);
2100         }
2101
2102         if (mp->m_qflags & XFS_GQUOTA_ACCT) {
2103                 if (mp->m_qflags & XFS_OQUOTA_ENFD)
2104                         seq_puts(m, "," MNTOPT_GRPQUOTA);
2105                 else
2106                         seq_puts(m, "," MNTOPT_GQUOTANOENF);
2107         }
2108
2109         if (!(mp->m_qflags & XFS_ALL_QUOTA_ACCT))
2110                 seq_puts(m, "," MNTOPT_NOQUOTA);
2111
2112         if (mp->m_flags & XFS_MOUNT_DMAPI)
2113                 seq_puts(m, "," MNTOPT_DMAPI);
2114         return 0;
2115 }
2116
2117 /*
2118  * Second stage of a freeze. The data is already frozen so we only
2119  * need to take care of themetadata. Once that's done write a dummy
2120  * record to dirty the log in case of a crash while frozen.
2121  */
2122 STATIC void
2123 xfs_freeze(
2124         xfs_mount_t     *mp)
2125 {
2126         xfs_attr_quiesce(mp);
2127         xfs_fs_log_dummy(mp);
2128 }