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