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