fs/xfs/linux-2.6/xfs_super.c

   1 /*
   2  * Copyright (c) 2000-2006 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_bit.h"
  20 #include "xfs_log.h"
  21 #include "xfs_clnt.h"
  22 #include "xfs_inum.h"
  23 #include "xfs_trans.h"
  24 #include "xfs_sb.h"
  25 #include "xfs_ag.h"
  26 #include "xfs_dir2.h"
  27 #include "xfs_alloc.h"
  28 #include "xfs_dmapi.h"
  29 #include "xfs_quota.h"
  30 #include "xfs_mount.h"
  31 #include "xfs_bmap_btree.h"
  32 #include "xfs_alloc_btree.h"
  33 #include "xfs_ialloc_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_btree.h"
  39 #include "xfs_ialloc.h"
  40 #include "xfs_bmap.h"
  41 #include "xfs_rtalloc.h"
  42 #include "xfs_error.h"
  43 #include "xfs_itable.h"
  44 #include "xfs_rw.h"
  45 #include "xfs_acl.h"
  46 #include "xfs_attr.h"
  47 #include "xfs_buf_item.h"
  48 #include "xfs_utils.h"
  49 #include "xfs_vnodeops.h"
  50 #include "xfs_version.h"
  51
  52 #include <linux/namei.h>
  53 #include <linux/init.h>
  54 #include <linux/mount.h>
  55 #include <linux/mempool.h>
  56 #include <linux/writeback.h>
  57 #include <linux/kthread.h>
  58 #include <linux/freezer.h>
  59
  60 static struct quotactl_ops xfs_quotactl_operations;
  61 static struct super_operations xfs_super_operations;
  62 static kmem_zone_t *xfs_vnode_zone;
  63 static kmem_zone_t *xfs_ioend_zone;
  64 mempool_t *xfs_ioend_pool;
  65
  66 STATIC struct xfs_mount_args *
  67 xfs_args_allocate(
  68         struct super_block      *sb,
  69         int                     silent)
  70 {
  71         struct xfs_mount_args   *args;
  72
  73         args = kmem_zalloc(sizeof(struct xfs_mount_args), KM_SLEEP);
  74         args->logbufs = args->logbufsize = -1;
  75         strncpy(args->fsname, sb->s_id, MAXNAMELEN);
  76
  77         /* Copy the already-parsed mount(2) flags we're interested in */
  78         if (sb->s_flags & MS_DIRSYNC)
  79                 args->flags |= XFSMNT_DIRSYNC;
  80         if (sb->s_flags & MS_SYNCHRONOUS)
  81                 args->flags |= XFSMNT_WSYNC;
  82         if (silent)
  83                 args->flags |= XFSMNT_QUIET;
  84         args->flags |= XFSMNT_32BITINODES;
  85
  86         return args;
  87 }
  88
  89 __uint64_t
  90 xfs_max_file_offset(
  91         unsigned int            blockshift)
  92 {
  93         unsigned int            pagefactor = 1;
  94         unsigned int            bitshift = BITS_PER_LONG - 1;
  95
  96         /* Figure out maximum filesize, on Linux this can depend on
  97          * the filesystem blocksize (on 32 bit platforms).
  98          * __block_prepare_write does this in an [unsigned] long...
  99          *      page->index << (PAGE_CACHE_SHIFT - bbits)
 100          * So, for page sized blocks (4K on 32 bit platforms),
 101          * this wraps at around 8Tb (hence MAX_LFS_FILESIZE which is
 102          *      (((u64)PAGE_CACHE_SIZE << (BITS_PER_LONG-1))-1)
 103          * but for smaller blocksizes it is less (bbits = log2 bsize).
 104          * Note1: get_block_t takes a long (implicit cast from above)
 105          * Note2: The Large Block Device (LBD and HAVE_SECTOR_T) patch
 106          * can optionally convert the [unsigned] long from above into
 107          * an [unsigned] long long.
 108          */
 109
 110 #if BITS_PER_LONG == 32
 111 # if defined(CONFIG_LBD)
 112         ASSERT(sizeof(sector_t) == 8);
 113         pagefactor = PAGE_CACHE_SIZE;
 114         bitshift = BITS_PER_LONG;
 115 # else
 116         pagefactor = PAGE_CACHE_SIZE >> (PAGE_CACHE_SHIFT - blockshift);
 117 # endif
 118 #endif
 119
 120         return (((__uint64_t)pagefactor) << bitshift) - 1;
 121 }
 122
 123 STATIC_INLINE void
 124 xfs_set_inodeops(
 125         struct inode            *inode)
 126 {
 127         switch (inode->i_mode & S_IFMT) {
 128         case S_IFREG:
 129                 inode->i_op = &xfs_inode_operations;
 130                 inode->i_fop = &xfs_file_operations;
 131                 inode->i_mapping->a_ops = &xfs_address_space_operations;
 132                 break;
 133         case S_IFDIR:
 134                 inode->i_op = &xfs_dir_inode_operations;
 135                 inode->i_fop = &xfs_dir_file_operations;
 136                 break;
 137         case S_IFLNK:
 138                 inode->i_op = &xfs_symlink_inode_operations;
 139                 if (inode->i_blocks)
 140                         inode->i_mapping->a_ops = &xfs_address_space_operations;
 141                 break;
 142         default:
 143                 inode->i_op = &xfs_inode_operations;
 144                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
 145                 break;
 146         }
 147 }
 148
 149 STATIC_INLINE void
 150 xfs_revalidate_inode(
 151         xfs_mount_t             *mp,
 152         bhv_vnode_t             *vp,
 153         xfs_inode_t             *ip)
 154 {
 155         struct inode            *inode = vn_to_inode(vp);
 156
 157         inode->i_mode   = ip->i_d.di_mode;
 158         inode->i_nlink  = ip->i_d.di_nlink;
 159         inode->i_uid    = ip->i_d.di_uid;
 160         inode->i_gid    = ip->i_d.di_gid;
 161
 162         switch (inode->i_mode & S_IFMT) {
 163         case S_IFBLK:
 164         case S_IFCHR:
 165                 inode->i_rdev =
 166                         MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff,
 167                               sysv_minor(ip->i_df.if_u2.if_rdev));
 168                 break;
 169         default:
 170                 inode->i_rdev = 0;
 171                 break;
 172         }
 173
 174         inode->i_generation = ip->i_d.di_gen;
 175         i_size_write(inode, ip->i_d.di_size);
 176         inode->i_blocks =
 177                 XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
 178         inode->i_atime.tv_sec   = ip->i_d.di_atime.t_sec;
 179         inode->i_atime.tv_nsec  = ip->i_d.di_atime.t_nsec;
 180         inode->i_mtime.tv_sec   = ip->i_d.di_mtime.t_sec;
 181         inode->i_mtime.tv_nsec  = ip->i_d.di_mtime.t_nsec;
 182         inode->i_ctime.tv_sec   = ip->i_d.di_ctime.t_sec;
 183         inode->i_ctime.tv_nsec  = ip->i_d.di_ctime.t_nsec;
 184         if (ip->i_d.di_flags & XFS_DIFLAG_IMMUTABLE)
 185                 inode->i_flags |= S_IMMUTABLE;
 186         else
 187                 inode->i_flags &= ~S_IMMUTABLE;
 188         if (ip->i_d.di_flags & XFS_DIFLAG_APPEND)
 189                 inode->i_flags |= S_APPEND;
 190         else
 191                 inode->i_flags &= ~S_APPEND;
 192         if (ip->i_d.di_flags & XFS_DIFLAG_SYNC)
 193                 inode->i_flags |= S_SYNC;
 194         else
 195                 inode->i_flags &= ~S_SYNC;
 196         if (ip->i_d.di_flags & XFS_DIFLAG_NOATIME)
 197                 inode->i_flags |= S_NOATIME;
 198         else
 199                 inode->i_flags &= ~S_NOATIME;
 200         xfs_iflags_clear(ip, XFS_IMODIFIED);
 201 }
 202
 203 void
 204 xfs_initialize_vnode(
 205         bhv_desc_t              *bdp,
 206         bhv_vnode_t             *vp,
 207         struct xfs_inode        *ip,
 208         int                     unlock)
 209 {
 210         struct inode            *inode = vn_to_inode(vp);
 211
 212         if (!ip->i_vnode) {
 213                 ip->i_vnode = vp;
 214                 inode->i_private = ip;
 215         }
 216
 217         /*
 218          * We need to set the ops vectors, and unlock the inode, but if
 219          * we have been called during the new inode create process, it is
 220          * too early to fill in the Linux inode.  We will get called a
 221          * second time once the inode is properly set up, and then we can
 222          * finish our work.
 223          */
 224         if (ip->i_d.di_mode != 0 && unlock && (inode->i_state & I_NEW)) {
 225                 xfs_revalidate_inode(XFS_BHVTOM(bdp), vp, ip);
 226                 xfs_set_inodeops(inode);
 227
 228                 xfs_iflags_clear(ip, XFS_INEW);
 229                 barrier();
 230
 231                 unlock_new_inode(inode);
 232         }
 233 }
 234
 235 int
 236 xfs_blkdev_get(
 237         xfs_mount_t             *mp,
 238         const char              *name,
 239         struct block_device     **bdevp)
 240 {
 241         int                     error = 0;
 242
 243         *bdevp = open_bdev_excl(name, 0, mp);
 244         if (IS_ERR(*bdevp)) {
 245                 error = PTR_ERR(*bdevp);
 246                 printk("XFS: Invalid device [%s], error=%d\n", name, error);
 247         }
 248
 249         return -error;
 250 }
 251
 252 void
 253 xfs_blkdev_put(
 254         struct block_device     *bdev)
 255 {
 256         if (bdev)
 257                 close_bdev_excl(bdev);
 258 }
 259
 260 /*
 261  * Try to write out the superblock using barriers.
 262  */
 263 STATIC int
 264 xfs_barrier_test(
 265         xfs_mount_t     *mp)
 266 {
 267         xfs_buf_t       *sbp = xfs_getsb(mp, 0);
 268         int             error;
 269
 270         XFS_BUF_UNDONE(sbp);
 271         XFS_BUF_UNREAD(sbp);
 272         XFS_BUF_UNDELAYWRITE(sbp);
 273         XFS_BUF_WRITE(sbp);
 274         XFS_BUF_UNASYNC(sbp);
 275         XFS_BUF_ORDERED(sbp);
 276
 277         xfsbdstrat(mp, sbp);
 278         error = xfs_iowait(sbp);
 279
 280         /*
 281          * Clear all the flags we set and possible error state in the
 282          * buffer.  We only did the write to try out whether barriers
 283          * worked and shouldn't leave any traces in the superblock
 284          * buffer.
 285          */
 286         XFS_BUF_DONE(sbp);
 287         XFS_BUF_ERROR(sbp, 0);
 288         XFS_BUF_UNORDERED(sbp);
 289
 290         xfs_buf_relse(sbp);
 291         return error;
 292 }
 293
 294 void
 295 xfs_mountfs_check_barriers(xfs_mount_t *mp)
 296 {
 297         int error;
 298
 299         if (mp->m_logdev_targp != mp->m_ddev_targp) {
 300                 xfs_fs_cmn_err(CE_NOTE, mp,
 301                   "Disabling barriers, not supported with external log device");
 302                 mp->m_flags &= ~XFS_MOUNT_BARRIER;
 303                 return;
 304         }
 305
 306         if (xfs_readonly_buftarg(mp->m_ddev_targp)) {
 307                 xfs_fs_cmn_err(CE_NOTE, mp,
 308                   "Disabling barriers, underlying device is readonly");
 309                 mp->m_flags &= ~XFS_MOUNT_BARRIER;
 310                 return;
 311         }
 312
 313         error = xfs_barrier_test(mp);
 314         if (error) {
 315                 xfs_fs_cmn_err(CE_NOTE, mp,
 316                   "Disabling barriers, trial barrier write failed");
 317                 mp->m_flags &= ~XFS_MOUNT_BARRIER;
 318                 return;
 319         }
 320 }
 321
 322 void
 323 xfs_blkdev_issue_flush(
 324         xfs_buftarg_t           *buftarg)
 325 {
 326         blkdev_issue_flush(buftarg->bt_bdev, NULL);
 327 }
 328
 329 STATIC struct inode *
 330 xfs_fs_alloc_inode(
 331         struct super_block      *sb)
 332 {
 333         bhv_vnode_t             *vp;
 334
 335         vp = kmem_zone_alloc(xfs_vnode_zone, KM_SLEEP);
 336         if (unlikely(!vp))
 337                 return NULL;
 338         return vn_to_inode(vp);
 339 }
 340
 341 STATIC void
 342 xfs_fs_destroy_inode(
 343         struct inode            *inode)
 344 {
 345         kmem_zone_free(xfs_vnode_zone, vn_from_inode(inode));
 346 }
 347
 348 STATIC void
 349 xfs_fs_inode_init_once(
 350         void                    *vnode,
 351         kmem_zone_t             *zonep,
 352         unsigned long           flags)
 353 {
 354         inode_init_once(vn_to_inode((bhv_vnode_t *)vnode));
 355 }
 356
 357 STATIC int
 358 xfs_init_zones(void)
 359 {
 360         xfs_vnode_zone = kmem_zone_init_flags(sizeof(bhv_vnode_t), "xfs_vnode",
 361                                         KM_ZONE_HWALIGN | KM_ZONE_RECLAIM |
 362                                         KM_ZONE_SPREAD,
 363                                         xfs_fs_inode_init_once);
 364         if (!xfs_vnode_zone)
 365                 goto out;
 366
 367         xfs_ioend_zone = kmem_zone_init(sizeof(xfs_ioend_t), "xfs_ioend");
 368         if (!xfs_ioend_zone)
 369                 goto out_destroy_vnode_zone;
 370
 371         xfs_ioend_pool = mempool_create_slab_pool(4 * MAX_BUF_PER_PAGE,
 372                                                   xfs_ioend_zone);
 373         if (!xfs_ioend_pool)
 374                 goto out_free_ioend_zone;
 375         return 0;
 376
 377  out_free_ioend_zone:
 378         kmem_zone_destroy(xfs_ioend_zone);
 379  out_destroy_vnode_zone:
 380         kmem_zone_destroy(xfs_vnode_zone);
 381  out:
 382         return -ENOMEM;
 383 }
 384
 385 STATIC void
 386 xfs_destroy_zones(void)
 387 {
 388         mempool_destroy(xfs_ioend_pool);
 389         kmem_zone_destroy(xfs_vnode_zone);
 390         kmem_zone_destroy(xfs_ioend_zone);
 391 }
 392
 393 /*
 394  * Attempt to flush the inode, this will actually fail
 395  * if the inode is pinned, but we dirty the inode again
 396  * at the point when it is unpinned after a log write,
 397  * since this is when the inode itself becomes flushable.
 398  */
 399 STATIC int
 400 xfs_fs_write_inode(
 401         struct inode            *inode,
 402         int                     sync)
 403 {
 404         int                     error = 0, flags = FLUSH_INODE;
 405
 406         vn_trace_entry(vn_from_inode(inode), __FUNCTION__,
 407                         (inst_t *)__return_address);
 408         if (sync) {
 409                 filemap_fdatawait(inode->i_mapping);
 410                 flags |= FLUSH_SYNC;
 411         }
 412         error = xfs_inode_flush(XFS_I(inode), flags);
 413         if (error == EAGAIN) {
 414                 if (sync)
 415                         error = xfs_inode_flush(XFS_I(inode),
 416                                                        flags | FLUSH_LOG);
 417                 else
 418                         error = 0;
 419         }
 420
 421         return -error;
 422 }
 423
 424 STATIC void
 425 xfs_fs_clear_inode(
 426         struct inode            *inode)
 427 {
 428         bhv_vnode_t             *vp = vn_from_inode(inode);
 429
 430         vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
 431
 432         XFS_STATS_INC(vn_rele);
 433         XFS_STATS_INC(vn_remove);
 434         XFS_STATS_INC(vn_reclaim);
 435         XFS_STATS_DEC(vn_active);
 436
 437         /*
 438          * This can happen because xfs_iget_core calls xfs_idestroy if we
 439          * find an inode with di_mode == 0 but without IGET_CREATE set.
 440          */
 441         if (XFS_I(inode))
 442                 xfs_inactive(XFS_I(inode));
 443
 444
 445         if (XFS_I(inode)) {
 446                 xfs_iflags_clear(XFS_I(inode), XFS_IMODIFIED);
 447                 if (xfs_reclaim(XFS_I(inode)))
 448                         panic("%s: cannot reclaim 0x%p\n", __FUNCTION__, vp);
 449         }
 450
 451         ASSERT(XFS_I(inode) == NULL);
 452
 453 #ifdef XFS_VNODE_TRACE
 454         ktrace_free(vp->v_trace);
 455 #endif
 456 }
 457
 458 /*
 459  * Enqueue a work item to be picked up by the vfs xfssyncd thread.
 460  * Doing this has two advantages:
 461  * - It saves on stack space, which is tight in certain situations
 462  * - It can be used (with care) as a mechanism to avoid deadlocks.
 463  * Flushing while allocating in a full filesystem requires both.
 464  */
 465 STATIC void
 466 xfs_syncd_queue_work(
 467         struct bhv_vfs  *vfs,
 468         void            *data,
 469         void            (*syncer)(bhv_vfs_t *, void *))
 470 {
 471         struct bhv_vfs_sync_work *work;
 472
 473         work = kmem_alloc(sizeof(struct bhv_vfs_sync_work), KM_SLEEP);
 474         INIT_LIST_HEAD(&work->w_list);
 475         work->w_syncer = syncer;
 476         work->w_data = data;
 477         work->w_vfs = vfs;
 478         spin_lock(&vfs->vfs_sync_lock);
 479         list_add_tail(&work->w_list, &vfs->vfs_sync_list);
 480         spin_unlock(&vfs->vfs_sync_lock);
 481         wake_up_process(vfs->vfs_sync_task);
 482 }
 483
 484 /*
 485  * Flush delayed allocate data, attempting to free up reserved space
 486  * from existing allocations.  At this point a new allocation attempt
 487  * has failed with ENOSPC and we are in the process of scratching our
 488  * heads, looking about for more room...
 489  */
 490 STATIC void
 491 xfs_flush_inode_work(
 492         bhv_vfs_t       *vfs,
 493         void            *inode)
 494 {
 495         filemap_flush(((struct inode *)inode)->i_mapping);
 496         iput((struct inode *)inode);
 497 }
 498
 499 void
 500 xfs_flush_inode(
 501         xfs_inode_t     *ip)
 502 {
 503         struct inode    *inode = vn_to_inode(XFS_ITOV(ip));
 504         struct bhv_vfs  *vfs = XFS_MTOVFS(ip->i_mount);
 505
 506         igrab(inode);
 507         xfs_syncd_queue_work(vfs, inode, xfs_flush_inode_work);
 508         delay(msecs_to_jiffies(500));
 509 }
 510
 511 /*
 512  * This is the "bigger hammer" version of xfs_flush_inode_work...
 513  * (IOW, "If at first you don't succeed, use a Bigger Hammer").
 514  */
 515 STATIC void
 516 xfs_flush_device_work(
 517         bhv_vfs_t       *vfs,
 518         void            *inode)
 519 {
 520         sync_blockdev(vfs->vfs_super->s_bdev);
 521         iput((struct inode *)inode);
 522 }
 523
 524 void
 525 xfs_flush_device(
 526         xfs_inode_t     *ip)
 527 {
 528         struct inode    *inode = vn_to_inode(XFS_ITOV(ip));
 529         struct bhv_vfs  *vfs = XFS_MTOVFS(ip->i_mount);
 530
 531         igrab(inode);
 532         xfs_syncd_queue_work(vfs, inode, xfs_flush_device_work);
 533         delay(msecs_to_jiffies(500));
 534         xfs_log_force(ip->i_mount, (xfs_lsn_t)0, XFS_LOG_FORCE|XFS_LOG_SYNC);
 535 }
 536
 537 STATIC void
 538 vfs_sync_worker(
 539         bhv_vfs_t       *vfsp,
 540         void            *unused)
 541 {
 542         int             error;
 543
 544         if (!(vfsp->vfs_flag & VFS_RDONLY))
 545                 error = bhv_vfs_sync(vfsp, SYNC_FSDATA | SYNC_BDFLUSH | \
 546                                         SYNC_ATTR | SYNC_REFCACHE | SYNC_SUPER,
 547                                         NULL);
 548         vfsp->vfs_sync_seq++;
 549         wake_up(&vfsp->vfs_wait_single_sync_task);
 550 }
 551
 552 STATIC int
 553 xfssyncd(
 554         void                    *arg)
 555 {
 556         long                    timeleft;
 557         bhv_vfs_t               *vfsp = (bhv_vfs_t *) arg;
 558         bhv_vfs_sync_work_t     *work, *n;
 559         LIST_HEAD               (tmp);
 560
 561         set_freezable();
 562         timeleft = xfs_syncd_centisecs * msecs_to_jiffies(10);
 563         for (;;) {
 564                 timeleft = schedule_timeout_interruptible(timeleft);
 565                 /* swsusp */
 566                 try_to_freeze();
 567                 if (kthread_should_stop() && list_empty(&vfsp->vfs_sync_list))
 568                         break;
 569
 570                 spin_lock(&vfsp->vfs_sync_lock);
 571                 /*
 572                  * We can get woken by laptop mode, to do a sync -
 573                  * that's the (only!) case where the list would be
 574                  * empty with time remaining.
 575                  */
 576                 if (!timeleft || list_empty(&vfsp->vfs_sync_list)) {
 577                         if (!timeleft)
 578                                 timeleft = xfs_syncd_centisecs *
 579                                                         msecs_to_jiffies(10);
 580                         INIT_LIST_HEAD(&vfsp->vfs_sync_work.w_list);
 581                         list_add_tail(&vfsp->vfs_sync_work.w_list,
 582                                         &vfsp->vfs_sync_list);
 583                 }
 584                 list_for_each_entry_safe(work, n, &vfsp->vfs_sync_list, w_list)
 585                         list_move(&work->w_list, &tmp);
 586                 spin_unlock(&vfsp->vfs_sync_lock);
 587
 588                 list_for_each_entry_safe(work, n, &tmp, w_list) {
 589                         (*work->w_syncer)(vfsp, work->w_data);
 590                         list_del(&work->w_list);
 591                         if (work == &vfsp->vfs_sync_work)
 592                                 continue;
 593                         kmem_free(work, sizeof(struct bhv_vfs_sync_work));
 594                 }
 595         }
 596
 597         return 0;
 598 }
 599
 600 STATIC int
 601 xfs_fs_start_syncd(
 602         bhv_vfs_t               *vfsp)
 603 {
 604         vfsp->vfs_sync_work.w_syncer = vfs_sync_worker;
 605         vfsp->vfs_sync_work.w_vfs = vfsp;
 606         vfsp->vfs_sync_task = kthread_run(xfssyncd, vfsp, "xfssyncd");
 607         if (IS_ERR(vfsp->vfs_sync_task))
 608                 return -PTR_ERR(vfsp->vfs_sync_task);
 609         return 0;
 610 }
 611
 612 STATIC void
 613 xfs_fs_stop_syncd(
 614         bhv_vfs_t               *vfsp)
 615 {
 616         kthread_stop(vfsp->vfs_sync_task);
 617 }
 618
 619 STATIC void
 620 xfs_fs_put_super(
 621         struct super_block      *sb)
 622 {
 623         bhv_vfs_t               *vfsp = vfs_from_sb(sb);
 624         int                     error;
 625
 626         xfs_fs_stop_syncd(vfsp);
 627         bhv_vfs_sync(vfsp, SYNC_ATTR | SYNC_DELWRI, NULL);
 628         error = bhv_vfs_unmount(vfsp, 0, NULL);
 629         if (error) {
 630                 printk("XFS: unmount got error=%d\n", error);
 631                 printk("%s: vfs=0x%p left dangling!\n", __FUNCTION__, vfsp);
 632         } else {
 633                 vfs_deallocate(vfsp);
 634         }
 635 }
 636
 637 STATIC void
 638 xfs_fs_write_super(
 639         struct super_block      *sb)
 640 {
 641         if (!(sb->s_flags & MS_RDONLY))
 642                 bhv_vfs_sync(vfs_from_sb(sb), SYNC_FSDATA, NULL);
 643         sb->s_dirt = 0;
 644 }
 645
 646 STATIC int
 647 xfs_fs_sync_super(
 648         struct super_block      *sb,
 649         int                     wait)
 650 {
 651         bhv_vfs_t               *vfsp = vfs_from_sb(sb);
 652         int                     error;
 653         int                     flags;
 654
 655         if (unlikely(sb->s_frozen == SB_FREEZE_WRITE)) {
 656                 /*
 657                  * First stage of freeze - no more writers will make progress
 658                  * now we are here, so we flush delwri and delalloc buffers
 659                  * here, then wait for all I/O to complete.  Data is frozen at
 660                  * that point. Metadata is not frozen, transactions can still
 661                  * occur here so don't bother flushing the buftarg (i.e
 662                  * SYNC_QUIESCE) because it'll just get dirty again.
 663                  */
 664                 flags = SYNC_DATA_QUIESCE;
 665         } else
 666                 flags = SYNC_FSDATA | (wait ? SYNC_WAIT : 0);
 667
 668         error = bhv_vfs_sync(vfsp, flags, NULL);
 669         sb->s_dirt = 0;
 670
 671         if (unlikely(laptop_mode)) {
 672                 int     prev_sync_seq = vfsp->vfs_sync_seq;
 673
 674                 /*
 675                  * The disk must be active because we're syncing.
 676                  * We schedule xfssyncd now (now that the disk is
 677                  * active) instead of later (when it might not be).
 678                  */
 679                 wake_up_process(vfsp->vfs_sync_task);
 680                 /*
 681                  * We have to wait for the sync iteration to complete.
 682                  * If we don't, the disk activity caused by the sync
 683                  * will come after the sync is completed, and that
 684                  * triggers another sync from laptop mode.
 685                  */
 686                 wait_event(vfsp->vfs_wait_single_sync_task,
 687                                 vfsp->vfs_sync_seq != prev_sync_seq);
 688         }
 689
 690         return -error;
 691 }
 692
 693 STATIC int
 694 xfs_fs_statfs(
 695         struct dentry           *dentry,
 696         struct kstatfs          *statp)
 697 {
 698         return -bhv_vfs_statvfs(vfs_from_sb(dentry->d_sb), statp,
 699                                 vn_from_inode(dentry->d_inode));
 700 }
 701
 702 STATIC int
 703 xfs_fs_remount(
 704         struct super_block      *sb,
 705         int                     *flags,
 706         char                    *options)
 707 {
 708         bhv_vfs_t               *vfsp = vfs_from_sb(sb);
 709         struct xfs_mount_args   *args = xfs_args_allocate(sb, 0);
 710         int                     error;
 711
 712         error = bhv_vfs_parseargs(vfsp, options, args, 1);
 713         if (!error)
 714                 error = bhv_vfs_mntupdate(vfsp, flags, args);
 715         kmem_free(args, sizeof(*args));
 716         return -error;
 717 }
 718
 719 STATIC void
 720 xfs_fs_lockfs(
 721         struct super_block      *sb)
 722 {
 723         bhv_vfs_freeze(vfs_from_sb(sb));
 724 }
 725
 726 STATIC int
 727 xfs_fs_show_options(
 728         struct seq_file         *m,
 729         struct vfsmount         *mnt)
 730 {
 731         return -bhv_vfs_showargs(vfs_from_sb(mnt->mnt_sb), m);
 732 }
 733
 734 STATIC int
 735 xfs_fs_quotasync(
 736         struct super_block      *sb,
 737         int                     type)
 738 {
 739         return -bhv_vfs_quotactl(vfs_from_sb(sb), Q_XQUOTASYNC, 0, NULL);
 740 }
 741
 742 STATIC int
 743 xfs_fs_getxstate(
 744         struct super_block      *sb,
 745         struct fs_quota_stat    *fqs)
 746 {
 747         return -bhv_vfs_quotactl(vfs_from_sb(sb), Q_XGETQSTAT, 0, (caddr_t)fqs);
 748 }
 749
 750 STATIC int
 751 xfs_fs_setxstate(
 752         struct super_block      *sb,
 753         unsigned int            flags,
 754         int                     op)
 755 {
 756         return -bhv_vfs_quotactl(vfs_from_sb(sb), op, 0, (caddr_t)&flags);
 757 }
 758
 759 STATIC int
 760 xfs_fs_getxquota(
 761         struct super_block      *sb,
 762         int                     type,
 763         qid_t                   id,
 764         struct fs_disk_quota    *fdq)
 765 {
 766         return -bhv_vfs_quotactl(vfs_from_sb(sb),
 767                                  (type == USRQUOTA) ? Q_XGETQUOTA :
 768                                   ((type == GRPQUOTA) ? Q_XGETGQUOTA :
 769                                    Q_XGETPQUOTA), id, (caddr_t)fdq);
 770 }
 771
 772 STATIC int
 773 xfs_fs_setxquota(
 774         struct super_block      *sb,
 775         int                     type,
 776         qid_t                   id,
 777         struct fs_disk_quota    *fdq)
 778 {
 779         return -bhv_vfs_quotactl(vfs_from_sb(sb),
 780                                  (type == USRQUOTA) ? Q_XSETQLIM :
 781                                   ((type == GRPQUOTA) ? Q_XSETGQLIM :
 782                                    Q_XSETPQLIM), id, (caddr_t)fdq);
 783 }
 784
 785 STATIC int
 786 xfs_fs_fill_super(
 787         struct super_block      *sb,
 788         void                    *data,
 789         int                     silent)
 790 {
 791         struct bhv_vnode        *rootvp;
 792         struct bhv_vfs          *vfsp = vfs_allocate(sb);
 793         struct xfs_mount_args   *args = xfs_args_allocate(sb, silent);
 794         struct kstatfs          statvfs;
 795         int                     error;
 796
 797         bhv_insert_all_vfsops(vfsp);
 798
 799         error = bhv_vfs_parseargs(vfsp, (char *)data, args, 0);
 800         if (error) {
 801                 bhv_remove_all_vfsops(vfsp, 1);
 802                 goto fail_vfsop;
 803         }
 804
 805         sb_min_blocksize(sb, BBSIZE);
 806         sb->s_export_op = &xfs_export_operations;
 807         sb->s_qcop = &xfs_quotactl_operations;
 808         sb->s_op = &xfs_super_operations;
 809
 810         error = bhv_vfs_mount(vfsp, args, NULL);
 811         if (error) {
 812                 bhv_remove_all_vfsops(vfsp, 1);
 813                 goto fail_vfsop;
 814         }
 815
 816         error = bhv_vfs_statvfs(vfsp, &statvfs, NULL);
 817         if (error)
 818                 goto fail_unmount;
 819
 820         sb->s_dirt = 1;
 821         sb->s_magic = statvfs.f_type;
 822         sb->s_blocksize = statvfs.f_bsize;
 823         sb->s_blocksize_bits = ffs(statvfs.f_bsize) - 1;
 824         sb->s_maxbytes = xfs_max_file_offset(sb->s_blocksize_bits);
 825         sb->s_time_gran = 1;
 826         set_posix_acl_flag(sb);
 827
 828         error = bhv_vfs_root(vfsp, &rootvp);
 829         if (error)
 830                 goto fail_unmount;
 831
 832         sb->s_root = d_alloc_root(vn_to_inode(rootvp));
 833         if (!sb->s_root) {
 834                 error = ENOMEM;
 835                 goto fail_vnrele;
 836         }
 837         if (is_bad_inode(sb->s_root->d_inode)) {
 838                 error = EINVAL;
 839                 goto fail_vnrele;
 840         }
 841         if ((error = xfs_fs_start_syncd(vfsp)))
 842                 goto fail_vnrele;
 843         vn_trace_exit(rootvp, __FUNCTION__, (inst_t *)__return_address);
 844
 845         kmem_free(args, sizeof(*args));
 846         return 0;
 847
 848 fail_vnrele:
 849         if (sb->s_root) {
 850                 dput(sb->s_root);
 851                 sb->s_root = NULL;
 852         } else {
 853                 VN_RELE(rootvp);
 854         }
 855
 856 fail_unmount:
 857         bhv_vfs_unmount(vfsp, 0, NULL);
 858
 859 fail_vfsop:
 860         vfs_deallocate(vfsp);
 861         kmem_free(args, sizeof(*args));
 862         return -error;
 863 }
 864
 865 STATIC int
 866 xfs_fs_get_sb(
 867         struct file_system_type *fs_type,
 868         int                     flags,
 869         const char              *dev_name,
 870         void                    *data,
 871         struct vfsmount         *mnt)
 872 {
 873         return get_sb_bdev(fs_type, flags, dev_name, data, xfs_fs_fill_super,
 874                            mnt);
 875 }
 876
 877 static struct super_operations xfs_super_operations = {
 878         .alloc_inode            = xfs_fs_alloc_inode,
 879         .destroy_inode          = xfs_fs_destroy_inode,
 880         .write_inode            = xfs_fs_write_inode,
 881         .clear_inode            = xfs_fs_clear_inode,
 882         .put_super              = xfs_fs_put_super,
 883         .write_super            = xfs_fs_write_super,
 884         .sync_fs                = xfs_fs_sync_super,
 885         .write_super_lockfs     = xfs_fs_lockfs,
 886         .statfs                 = xfs_fs_statfs,
 887         .remount_fs             = xfs_fs_remount,
 888         .show_options           = xfs_fs_show_options,
 889 };
 890
 891 static struct quotactl_ops xfs_quotactl_operations = {
 892         .quota_sync             = xfs_fs_quotasync,
 893         .get_xstate             = xfs_fs_getxstate,
 894         .set_xstate             = xfs_fs_setxstate,
 895         .get_xquota             = xfs_fs_getxquota,
 896         .set_xquota             = xfs_fs_setxquota,
 897 };
 898
 899 static struct file_system_type xfs_fs_type = {
 900         .owner                  = THIS_MODULE,
 901         .name                   = "xfs",
 902         .get_sb                 = xfs_fs_get_sb,
 903         .kill_sb                = kill_block_super,
 904         .fs_flags               = FS_REQUIRES_DEV,
 905 };
 906
 907
 908 STATIC int __init
 909 init_xfs_fs( void )
 910 {
 911         int                     error;
 912         static char             message[] __initdata = KERN_INFO \
 913                 XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled\n";
 914
 915         printk(message);
 916
 917         ktrace_init(64);
 918
 919         error = xfs_init_zones();
 920         if (error < 0)
 921                 goto undo_zones;
 922
 923         error = xfs_buf_init();
 924         if (error < 0)
 925                 goto undo_buffers;
 926
 927         vn_init();
 928         xfs_init();
 929         uuid_init();
 930         vfs_initquota();
 931
 932         error = register_filesystem(&xfs_fs_type);
 933         if (error)
 934                 goto undo_register;
 935         return 0;
 936
 937 undo_register:
 938         xfs_buf_terminate();
 939
 940 undo_buffers:
 941         xfs_destroy_zones();
 942
 943 undo_zones:
 944         return error;
 945 }
 946
 947 STATIC void __exit
 948 exit_xfs_fs( void )
 949 {
 950         vfs_exitquota();
 951         unregister_filesystem(&xfs_fs_type);
 952         xfs_cleanup();
 953         xfs_buf_terminate();
 954         xfs_destroy_zones();
 955         ktrace_uninit();
 956 }
 957
 958 module_init(init_xfs_fs);
 959 module_exit(exit_xfs_fs);
 960
 961 MODULE_AUTHOR("Silicon Graphics, Inc.");
 962 MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled");
 963 MODULE_LICENSE("GPL");