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_version.h"
  50
  51 #include <linux/namei.h>
  52 #include <linux/init.h>
  53 #include <linux/mount.h>
  54 #include <linux/mempool.h>
  55 #include <linux/writeback.h>
  56 #include <linux/kthread.h>
  57 #include <linux/freezer.h>
  58
  59 static struct quotactl_ops xfs_quotactl_operations;
  60 static struct super_operations xfs_super_operations;
  61 static kmem_zone_t *xfs_vnode_zone;
  62 static kmem_zone_t *xfs_ioend_zone;
  63 mempool_t *xfs_ioend_pool;
  64
  65 STATIC struct xfs_mount_args *
  66 xfs_args_allocate(
  67         struct super_block      *sb,
  68         int                     silent)
  69 {
  70         struct xfs_mount_args   *args;
  71
  72         args = kmem_zalloc(sizeof(struct xfs_mount_args), KM_SLEEP);
  73         args->logbufs = args->logbufsize = -1;
  74         strncpy(args->fsname, sb->s_id, MAXNAMELEN);
  75
  76         /* Copy the already-parsed mount(2) flags we're interested in */
  77         if (sb->s_flags & MS_DIRSYNC)
  78                 args->flags |= XFSMNT_DIRSYNC;
  79         if (sb->s_flags & MS_SYNCHRONOUS)
  80                 args->flags |= XFSMNT_WSYNC;
  81         if (silent)
  82                 args->flags |= XFSMNT_QUIET;
  83         args->flags |= XFSMNT_32BITINODES;
  84
  85         return args;
  86 }
  87
  88 __uint64_t
  89 xfs_max_file_offset(
  90         unsigned int            blockshift)
  91 {
  92         unsigned int            pagefactor = 1;
  93         unsigned int            bitshift = BITS_PER_LONG - 1;
  94
  95         /* Figure out maximum filesize, on Linux this can depend on
  96          * the filesystem blocksize (on 32 bit platforms).
  97          * __block_prepare_write does this in an [unsigned] long...
  98          *      page->index << (PAGE_CACHE_SHIFT - bbits)
  99          * So, for page sized blocks (4K on 32 bit platforms),
 100          * this wraps at around 8Tb (hence MAX_LFS_FILESIZE which is
 101          *      (((u64)PAGE_CACHE_SIZE << (BITS_PER_LONG-1))-1)
 102          * but for smaller blocksizes it is less (bbits = log2 bsize).
 103          * Note1: get_block_t takes a long (implicit cast from above)
 104          * Note2: The Large Block Device (LBD and HAVE_SECTOR_T) patch
 105          * can optionally convert the [unsigned] long from above into
 106          * an [unsigned] long long.
 107          */
 108
 109 #if BITS_PER_LONG == 32
 110 # if defined(CONFIG_LBD)
 111         ASSERT(sizeof(sector_t) == 8);
 112         pagefactor = PAGE_CACHE_SIZE;
 113         bitshift = BITS_PER_LONG;
 114 # else
 115         pagefactor = PAGE_CACHE_SIZE >> (PAGE_CACHE_SHIFT - blockshift);
 116 # endif
 117 #endif
 118
 119         return (((__uint64_t)pagefactor) << bitshift) - 1;
 120 }
 121
 122 STATIC_INLINE void
 123 xfs_set_inodeops(
 124         struct inode            *inode)
 125 {
 126         switch (inode->i_mode & S_IFMT) {
 127         case S_IFREG:
 128                 inode->i_op = &xfs_inode_operations;
 129                 inode->i_fop = &xfs_file_operations;
 130                 inode->i_mapping->a_ops = &xfs_address_space_operations;
 131                 break;
 132         case S_IFDIR:
 133                 inode->i_op = &xfs_dir_inode_operations;
 134                 inode->i_fop = &xfs_dir_file_operations;
 135                 break;
 136         case S_IFLNK:
 137                 inode->i_op = &xfs_symlink_inode_operations;
 138                 if (inode->i_blocks)
 139                         inode->i_mapping->a_ops = &xfs_address_space_operations;
 140                 break;
 141         default:
 142                 inode->i_op = &xfs_inode_operations;
 143                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
 144                 break;
 145         }
 146 }
 147
 148 STATIC_INLINE void
 149 xfs_revalidate_inode(
 150         xfs_mount_t             *mp,
 151         bhv_vnode_t             *vp,
 152         xfs_inode_t             *ip)
 153 {
 154         struct inode            *inode = vn_to_inode(vp);
 155
 156         inode->i_mode   = ip->i_d.di_mode;
 157         inode->i_nlink  = ip->i_d.di_nlink;
 158         inode->i_uid    = ip->i_d.di_uid;
 159         inode->i_gid    = ip->i_d.di_gid;
 160
 161         switch (inode->i_mode & S_IFMT) {
 162         case S_IFBLK:
 163         case S_IFCHR:
 164                 inode->i_rdev =
 165                         MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff,
 166                               sysv_minor(ip->i_df.if_u2.if_rdev));
 167                 break;
 168         default:
 169                 inode->i_rdev = 0;
 170                 break;
 171         }
 172
 173         inode->i_generation = ip->i_d.di_gen;
 174         i_size_write(inode, ip->i_d.di_size);
 175         inode->i_blocks =
 176                 XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
 177         inode->i_atime.tv_sec   = ip->i_d.di_atime.t_sec;
 178         inode->i_atime.tv_nsec  = ip->i_d.di_atime.t_nsec;
 179         inode->i_mtime.tv_sec   = ip->i_d.di_mtime.t_sec;
 180         inode->i_mtime.tv_nsec  = ip->i_d.di_mtime.t_nsec;
 181         inode->i_ctime.tv_sec   = ip->i_d.di_ctime.t_sec;
 182         inode->i_ctime.tv_nsec  = ip->i_d.di_ctime.t_nsec;
 183         if (ip->i_d.di_flags & XFS_DIFLAG_IMMUTABLE)
 184                 inode->i_flags |= S_IMMUTABLE;
 185         else
 186                 inode->i_flags &= ~S_IMMUTABLE;
 187         if (ip->i_d.di_flags & XFS_DIFLAG_APPEND)
 188                 inode->i_flags |= S_APPEND;
 189         else
 190                 inode->i_flags &= ~S_APPEND;
 191         if (ip->i_d.di_flags & XFS_DIFLAG_SYNC)
 192                 inode->i_flags |= S_SYNC;
 193         else
 194                 inode->i_flags &= ~S_SYNC;
 195         if (ip->i_d.di_flags & XFS_DIFLAG_NOATIME)
 196                 inode->i_flags |= S_NOATIME;
 197         else
 198                 inode->i_flags &= ~S_NOATIME;
 199         vp->v_flag &= ~VMODIFIED;
 200 }
 201
 202 void
 203 xfs_initialize_vnode(
 204         bhv_desc_t              *bdp,
 205         bhv_vnode_t             *vp,
 206         bhv_desc_t              *inode_bhv,
 207         int                     unlock)
 208 {
 209         xfs_inode_t             *ip = XFS_BHVTOI(inode_bhv);
 210         struct inode            *inode = vn_to_inode(vp);
 211
 212         if (!inode_bhv->bd_vobj) {
 213                 vp->v_vfsp = bhvtovfs(bdp);
 214                 bhv_desc_init(inode_bhv, ip, vp, &xfs_vnodeops);
 215                 bhv_insert(VN_BHV_HEAD(vp), inode_bhv);
 216         }
 217
 218         /*
 219          * We need to set the ops vectors, and unlock the inode, but if
 220          * we have been called during the new inode create process, it is
 221          * too early to fill in the Linux inode.  We will get called a
 222          * second time once the inode is properly set up, and then we can
 223          * finish our work.
 224          */
 225         if (ip->i_d.di_mode != 0 && unlock && (inode->i_state & I_NEW)) {
 226                 xfs_revalidate_inode(XFS_BHVTOM(bdp), vp, ip);
 227                 xfs_set_inodeops(inode);
 228
 229                 xfs_iflags_clear(ip, XFS_INEW);
 230                 barrier();
 231
 232                 unlock_new_inode(inode);
 233         }
 234 }
 235
 236 int
 237 xfs_blkdev_get(
 238         xfs_mount_t             *mp,
 239         const char              *name,
 240         struct block_device     **bdevp)
 241 {
 242         int                     error = 0;
 243
 244         *bdevp = open_bdev_excl(name, 0, mp);
 245         if (IS_ERR(*bdevp)) {
 246                 error = PTR_ERR(*bdevp);
 247                 printk("XFS: Invalid device [%s], error=%d\n", name, error);
 248         }
 249
 250         return -error;
 251 }
 252
 253 void
 254 xfs_blkdev_put(
 255         struct block_device     *bdev)
 256 {
 257         if (bdev)
 258                 close_bdev_excl(bdev);
 259 }
 260
 261 /*
 262  * Try to write out the superblock using barriers.
 263  */
 264 STATIC int
 265 xfs_barrier_test(
 266         xfs_mount_t     *mp)
 267 {
 268         xfs_buf_t       *sbp = xfs_getsb(mp, 0);
 269         int             error;
 270
 271         XFS_BUF_UNDONE(sbp);
 272         XFS_BUF_UNREAD(sbp);
 273         XFS_BUF_UNDELAYWRITE(sbp);
 274         XFS_BUF_WRITE(sbp);
 275         XFS_BUF_UNASYNC(sbp);
 276         XFS_BUF_ORDERED(sbp);
 277
 278         xfsbdstrat(mp, sbp);
 279         error = xfs_iowait(sbp);
 280
 281         /*
 282          * Clear all the flags we set and possible error state in the
 283          * buffer.  We only did the write to try out whether barriers
 284          * worked and shouldn't leave any traces in the superblock
 285          * buffer.
 286          */
 287         XFS_BUF_DONE(sbp);
 288         XFS_BUF_ERROR(sbp, 0);
 289         XFS_BUF_UNORDERED(sbp);
 290
 291         xfs_buf_relse(sbp);
 292         return error;
 293 }
 294
 295 void
 296 xfs_mountfs_check_barriers(xfs_mount_t *mp)
 297 {
 298         int error;
 299
 300         if (mp->m_logdev_targp != mp->m_ddev_targp) {
 301                 xfs_fs_cmn_err(CE_NOTE, mp,
 302                   "Disabling barriers, not supported with external log device");
 303                 mp->m_flags &= ~XFS_MOUNT_BARRIER;
 304                 return;
 305         }
 306
 307         if (mp->m_ddev_targp->bt_bdev->bd_disk->queue->ordered ==
 308                                         QUEUE_ORDERED_NONE) {
 309                 xfs_fs_cmn_err(CE_NOTE, mp,
 310                   "Disabling barriers, not supported by the underlying device");
 311                 mp->m_flags &= ~XFS_MOUNT_BARRIER;
 312                 return;
 313         }
 314
 315         if (xfs_readonly_buftarg(mp->m_ddev_targp)) {
 316                 xfs_fs_cmn_err(CE_NOTE, mp,
 317                   "Disabling barriers, underlying device is readonly");
 318                 mp->m_flags &= ~XFS_MOUNT_BARRIER;
 319                 return;
 320         }
 321
 322         error = xfs_barrier_test(mp);
 323         if (error) {
 324                 xfs_fs_cmn_err(CE_NOTE, mp,
 325                   "Disabling barriers, trial barrier write failed");
 326                 mp->m_flags &= ~XFS_MOUNT_BARRIER;
 327                 return;
 328         }
 329 }
 330
 331 void
 332 xfs_blkdev_issue_flush(
 333         xfs_buftarg_t           *buftarg)
 334 {
 335         blkdev_issue_flush(buftarg->bt_bdev, NULL);
 336 }
 337
 338 STATIC struct inode *
 339 xfs_fs_alloc_inode(
 340         struct super_block      *sb)
 341 {
 342         bhv_vnode_t             *vp;
 343
 344         vp = kmem_zone_alloc(xfs_vnode_zone, KM_SLEEP);
 345         if (unlikely(!vp))
 346                 return NULL;
 347         return vn_to_inode(vp);
 348 }
 349
 350 STATIC void
 351 xfs_fs_destroy_inode(
 352         struct inode            *inode)
 353 {
 354         kmem_zone_free(xfs_vnode_zone, vn_from_inode(inode));
 355 }
 356
 357 STATIC void
 358 xfs_fs_inode_init_once(
 359         void                    *vnode,
 360         kmem_zone_t             *zonep,
 361         unsigned long           flags)
 362 {
 363         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
 364                       SLAB_CTOR_CONSTRUCTOR)
 365                 inode_init_once(vn_to_inode((bhv_vnode_t *)vnode));
 366 }
 367
 368 STATIC int
 369 xfs_init_zones(void)
 370 {
 371         xfs_vnode_zone = kmem_zone_init_flags(sizeof(bhv_vnode_t), "xfs_vnode",
 372                                         KM_ZONE_HWALIGN | KM_ZONE_RECLAIM |
 373                                         KM_ZONE_SPREAD,
 374                                         xfs_fs_inode_init_once);
 375         if (!xfs_vnode_zone)
 376                 goto out;
 377
 378         xfs_ioend_zone = kmem_zone_init(sizeof(xfs_ioend_t), "xfs_ioend");
 379         if (!xfs_ioend_zone)
 380                 goto out_destroy_vnode_zone;
 381
 382         xfs_ioend_pool = mempool_create_slab_pool(4 * MAX_BUF_PER_PAGE,
 383                                                   xfs_ioend_zone);
 384         if (!xfs_ioend_pool)
 385                 goto out_free_ioend_zone;
 386         return 0;
 387
 388  out_free_ioend_zone:
 389         kmem_zone_destroy(xfs_ioend_zone);
 390  out_destroy_vnode_zone:
 391         kmem_zone_destroy(xfs_vnode_zone);
 392  out:
 393         return -ENOMEM;
 394 }
 395
 396 STATIC void
 397 xfs_destroy_zones(void)
 398 {
 399         mempool_destroy(xfs_ioend_pool);
 400         kmem_zone_destroy(xfs_vnode_zone);
 401         kmem_zone_destroy(xfs_ioend_zone);
 402 }
 403
 404 /*
 405  * Attempt to flush the inode, this will actually fail
 406  * if the inode is pinned, but we dirty the inode again
 407  * at the point when it is unpinned after a log write,
 408  * since this is when the inode itself becomes flushable.
 409  */
 410 STATIC int
 411 xfs_fs_write_inode(
 412         struct inode            *inode,
 413         int                     sync)
 414 {
 415         bhv_vnode_t             *vp = vn_from_inode(inode);
 416         int                     error = 0, flags = FLUSH_INODE;
 417
 418         if (vp) {
 419                 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
 420                 if (sync)
 421                         flags |= FLUSH_SYNC;
 422                 error = bhv_vop_iflush(vp, flags);
 423                 if (error == EAGAIN)
 424                         error = sync? bhv_vop_iflush(vp, flags | FLUSH_LOG) : 0;
 425         }
 426         return -error;
 427 }
 428
 429 STATIC void
 430 xfs_fs_clear_inode(
 431         struct inode            *inode)
 432 {
 433         bhv_vnode_t             *vp = vn_from_inode(inode);
 434
 435         vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
 436
 437         XFS_STATS_INC(vn_rele);
 438         XFS_STATS_INC(vn_remove);
 439         XFS_STATS_INC(vn_reclaim);
 440         XFS_STATS_DEC(vn_active);
 441
 442         /*
 443          * This can happen because xfs_iget_core calls xfs_idestroy if we
 444          * find an inode with di_mode == 0 but without IGET_CREATE set.
 445          */
 446         if (VNHEAD(vp))
 447                 bhv_vop_inactive(vp, NULL);
 448
 449         VN_LOCK(vp);
 450         vp->v_flag &= ~VMODIFIED;
 451         VN_UNLOCK(vp, 0);
 452
 453         if (VNHEAD(vp))
 454                 if (bhv_vop_reclaim(vp))
 455                         panic("%s: cannot reclaim 0x%p\n", __FUNCTION__, vp);
 456
 457         ASSERT(VNHEAD(vp) == NULL);
 458
 459 #ifdef XFS_VNODE_TRACE
 460         ktrace_free(vp->v_trace);
 461 #endif
 462 }
 463
 464 /*
 465  * Enqueue a work item to be picked up by the vfs xfssyncd thread.
 466  * Doing this has two advantages:
 467  * - It saves on stack space, which is tight in certain situations
 468  * - It can be used (with care) as a mechanism to avoid deadlocks.
 469  * Flushing while allocating in a full filesystem requires both.
 470  */
 471 STATIC void
 472 xfs_syncd_queue_work(
 473         struct bhv_vfs  *vfs,
 474         void            *data,
 475         void            (*syncer)(bhv_vfs_t *, void *))
 476 {
 477         struct bhv_vfs_sync_work *work;
 478
 479         work = kmem_alloc(sizeof(struct bhv_vfs_sync_work), KM_SLEEP);
 480         INIT_LIST_HEAD(&work->w_list);
 481         work->w_syncer = syncer;
 482         work->w_data = data;
 483         work->w_vfs = vfs;
 484         spin_lock(&vfs->vfs_sync_lock);
 485         list_add_tail(&work->w_list, &vfs->vfs_sync_list);
 486         spin_unlock(&vfs->vfs_sync_lock);
 487         wake_up_process(vfs->vfs_sync_task);
 488 }
 489
 490 /*
 491  * Flush delayed allocate data, attempting to free up reserved space
 492  * from existing allocations.  At this point a new allocation attempt
 493  * has failed with ENOSPC and we are in the process of scratching our
 494  * heads, looking about for more room...
 495  */
 496 STATIC void
 497 xfs_flush_inode_work(
 498         bhv_vfs_t       *vfs,
 499         void            *inode)
 500 {
 501         filemap_flush(((struct inode *)inode)->i_mapping);
 502         iput((struct inode *)inode);
 503 }
 504
 505 void
 506 xfs_flush_inode(
 507         xfs_inode_t     *ip)
 508 {
 509         struct inode    *inode = vn_to_inode(XFS_ITOV(ip));
 510         struct bhv_vfs  *vfs = XFS_MTOVFS(ip->i_mount);
 511
 512         igrab(inode);
 513         xfs_syncd_queue_work(vfs, inode, xfs_flush_inode_work);
 514         delay(msecs_to_jiffies(500));
 515 }
 516
 517 /*
 518  * This is the "bigger hammer" version of xfs_flush_inode_work...
 519  * (IOW, "If at first you don't succeed, use a Bigger Hammer").
 520  */
 521 STATIC void
 522 xfs_flush_device_work(
 523         bhv_vfs_t       *vfs,
 524         void            *inode)
 525 {
 526         sync_blockdev(vfs->vfs_super->s_bdev);
 527         iput((struct inode *)inode);
 528 }
 529
 530 void
 531 xfs_flush_device(
 532         xfs_inode_t     *ip)
 533 {
 534         struct inode    *inode = vn_to_inode(XFS_ITOV(ip));
 535         struct bhv_vfs  *vfs = XFS_MTOVFS(ip->i_mount);
 536
 537         igrab(inode);
 538         xfs_syncd_queue_work(vfs, inode, xfs_flush_device_work);
 539         delay(msecs_to_jiffies(500));
 540         xfs_log_force(ip->i_mount, (xfs_lsn_t)0, XFS_LOG_FORCE|XFS_LOG_SYNC);
 541 }
 542
 543 STATIC void
 544 vfs_sync_worker(
 545         bhv_vfs_t       *vfsp,
 546         void            *unused)
 547 {
 548         int             error;
 549
 550         if (!(vfsp->vfs_flag & VFS_RDONLY))
 551                 error = bhv_vfs_sync(vfsp, SYNC_FSDATA | SYNC_BDFLUSH | \
 552                                         SYNC_ATTR | SYNC_REFCACHE, NULL);
 553         vfsp->vfs_sync_seq++;
 554         wake_up(&vfsp->vfs_wait_single_sync_task);
 555 }
 556
 557 STATIC int
 558 xfssyncd(
 559         void                    *arg)
 560 {
 561         long                    timeleft;
 562         bhv_vfs_t               *vfsp = (bhv_vfs_t *) arg;
 563         bhv_vfs_sync_work_t     *work, *n;
 564         LIST_HEAD               (tmp);
 565
 566         timeleft = xfs_syncd_centisecs * msecs_to_jiffies(10);
 567         for (;;) {
 568                 timeleft = schedule_timeout_interruptible(timeleft);
 569                 /* swsusp */
 570                 try_to_freeze();
 571                 if (kthread_should_stop() && list_empty(&vfsp->vfs_sync_list))
 572                         break;
 573
 574                 spin_lock(&vfsp->vfs_sync_lock);
 575                 /*
 576                  * We can get woken by laptop mode, to do a sync -
 577                  * that's the (only!) case where the list would be
 578                  * empty with time remaining.
 579                  */
 580                 if (!timeleft || list_empty(&vfsp->vfs_sync_list)) {
 581                         if (!timeleft)
 582                                 timeleft = xfs_syncd_centisecs *
 583                                                         msecs_to_jiffies(10);
 584                         INIT_LIST_HEAD(&vfsp->vfs_sync_work.w_list);
 585                         list_add_tail(&vfsp->vfs_sync_work.w_list,
 586                                         &vfsp->vfs_sync_list);
 587                 }
 588                 list_for_each_entry_safe(work, n, &vfsp->vfs_sync_list, w_list)
 589                         list_move(&work->w_list, &tmp);
 590                 spin_unlock(&vfsp->vfs_sync_lock);
 591
 592                 list_for_each_entry_safe(work, n, &tmp, w_list) {
 593                         (*work->w_syncer)(vfsp, work->w_data);
 594                         list_del(&work->w_list);
 595                         if (work == &vfsp->vfs_sync_work)
 596                                 continue;
 597                         kmem_free(work, sizeof(struct bhv_vfs_sync_work));
 598                 }
 599         }
 600
 601         return 0;
 602 }
 603
 604 STATIC int
 605 xfs_fs_start_syncd(
 606         bhv_vfs_t               *vfsp)
 607 {
 608         vfsp->vfs_sync_work.w_syncer = vfs_sync_worker;
 609         vfsp->vfs_sync_work.w_vfs = vfsp;
 610         vfsp->vfs_sync_task = kthread_run(xfssyncd, vfsp, "xfssyncd");
 611         if (IS_ERR(vfsp->vfs_sync_task))
 612                 return -PTR_ERR(vfsp->vfs_sync_task);
 613         return 0;
 614 }
 615
 616 STATIC void
 617 xfs_fs_stop_syncd(
 618         bhv_vfs_t               *vfsp)
 619 {
 620         kthread_stop(vfsp->vfs_sync_task);
 621 }
 622
 623 STATIC void
 624 xfs_fs_put_super(
 625         struct super_block      *sb)
 626 {
 627         bhv_vfs_t               *vfsp = vfs_from_sb(sb);
 628         int                     error;
 629
 630         xfs_fs_stop_syncd(vfsp);
 631         bhv_vfs_sync(vfsp, SYNC_ATTR | SYNC_DELWRI, NULL);
 632         error = bhv_vfs_unmount(vfsp, 0, NULL);
 633         if (error) {
 634                 printk("XFS: unmount got error=%d\n", error);
 635                 printk("%s: vfs=0x%p left dangling!\n", __FUNCTION__, vfsp);
 636         } else {
 637                 vfs_deallocate(vfsp);
 638         }
 639 }
 640
 641 STATIC void
 642 xfs_fs_write_super(
 643         struct super_block      *sb)
 644 {
 645         if (!(sb->s_flags & MS_RDONLY))
 646                 bhv_vfs_sync(vfs_from_sb(sb), SYNC_FSDATA, NULL);
 647         sb->s_dirt = 0;
 648 }
 649
 650 STATIC int
 651 xfs_fs_sync_super(
 652         struct super_block      *sb,
 653         int                     wait)
 654 {
 655         bhv_vfs_t               *vfsp = vfs_from_sb(sb);
 656         int                     error;
 657         int                     flags;
 658
 659         if (unlikely(sb->s_frozen == SB_FREEZE_WRITE)) {
 660                 /*
 661                  * First stage of freeze - no more writers will make progress
 662                  * now we are here, so we flush delwri and delalloc buffers
 663                  * here, then wait for all I/O to complete.  Data is frozen at
 664                  * that point. Metadata is not frozen, transactions can still
 665                  * occur here so don't bother flushing the buftarg (i.e
 666                  * SYNC_QUIESCE) because it'll just get dirty again.
 667                  */
 668                 flags = SYNC_FSDATA | SYNC_DELWRI | SYNC_WAIT | SYNC_IOWAIT;
 669         } else
 670                 flags = SYNC_FSDATA | (wait ? SYNC_WAIT : 0);
 671
 672         error = bhv_vfs_sync(vfsp, flags, NULL);
 673         sb->s_dirt = 0;
 674
 675         if (unlikely(laptop_mode)) {
 676                 int     prev_sync_seq = vfsp->vfs_sync_seq;
 677
 678                 /*
 679                  * The disk must be active because we're syncing.
 680                  * We schedule xfssyncd now (now that the disk is
 681                  * active) instead of later (when it might not be).
 682                  */
 683                 wake_up_process(vfsp->vfs_sync_task);
 684                 /*
 685                  * We have to wait for the sync iteration to complete.
 686                  * If we don't, the disk activity caused by the sync
 687                  * will come after the sync is completed, and that
 688                  * triggers another sync from laptop mode.
 689                  */
 690                 wait_event(vfsp->vfs_wait_single_sync_task,
 691                                 vfsp->vfs_sync_seq != prev_sync_seq);
 692         }
 693
 694         return -error;
 695 }
 696
 697 STATIC int
 698 xfs_fs_statfs(
 699         struct dentry           *dentry,
 700         struct kstatfs          *statp)
 701 {
 702         return -bhv_vfs_statvfs(vfs_from_sb(dentry->d_sb), statp,
 703                                 vn_from_inode(dentry->d_inode));
 704 }
 705
 706 STATIC int
 707 xfs_fs_remount(
 708         struct super_block      *sb,
 709         int                     *flags,
 710         char                    *options)
 711 {
 712         bhv_vfs_t               *vfsp = vfs_from_sb(sb);
 713         struct xfs_mount_args   *args = xfs_args_allocate(sb, 0);
 714         int                     error;
 715
 716         error = bhv_vfs_parseargs(vfsp, options, args, 1);
 717         if (!error)
 718                 error = bhv_vfs_mntupdate(vfsp, flags, args);
 719         kmem_free(args, sizeof(*args));
 720         return -error;
 721 }
 722
 723 STATIC void
 724 xfs_fs_lockfs(
 725         struct super_block      *sb)
 726 {
 727         bhv_vfs_freeze(vfs_from_sb(sb));
 728 }
 729
 730 STATIC int
 731 xfs_fs_show_options(
 732         struct seq_file         *m,
 733         struct vfsmount         *mnt)
 734 {
 735         return -bhv_vfs_showargs(vfs_from_sb(mnt->mnt_sb), m);
 736 }
 737
 738 STATIC int
 739 xfs_fs_quotasync(
 740         struct super_block      *sb,
 741         int                     type)
 742 {
 743         return -bhv_vfs_quotactl(vfs_from_sb(sb), Q_XQUOTASYNC, 0, NULL);
 744 }
 745
 746 STATIC int
 747 xfs_fs_getxstate(
 748         struct super_block      *sb,
 749         struct fs_quota_stat    *fqs)
 750 {
 751         return -bhv_vfs_quotactl(vfs_from_sb(sb), Q_XGETQSTAT, 0, (caddr_t)fqs);
 752 }
 753
 754 STATIC int
 755 xfs_fs_setxstate(
 756         struct super_block      *sb,
 757         unsigned int            flags,
 758         int                     op)
 759 {
 760         return -bhv_vfs_quotactl(vfs_from_sb(sb), op, 0, (caddr_t)&flags);
 761 }
 762
 763 STATIC int
 764 xfs_fs_getxquota(
 765         struct super_block      *sb,
 766         int                     type,
 767         qid_t                   id,
 768         struct fs_disk_quota    *fdq)
 769 {
 770         return -bhv_vfs_quotactl(vfs_from_sb(sb),
 771                                  (type == USRQUOTA) ? Q_XGETQUOTA :
 772                                   ((type == GRPQUOTA) ? Q_XGETGQUOTA :
 773                                    Q_XGETPQUOTA), id, (caddr_t)fdq);
 774 }
 775
 776 STATIC int
 777 xfs_fs_setxquota(
 778         struct super_block      *sb,
 779         int                     type,
 780         qid_t                   id,
 781         struct fs_disk_quota    *fdq)
 782 {
 783         return -bhv_vfs_quotactl(vfs_from_sb(sb),
 784                                  (type == USRQUOTA) ? Q_XSETQLIM :
 785                                   ((type == GRPQUOTA) ? Q_XSETGQLIM :
 786                                    Q_XSETPQLIM), id, (caddr_t)fdq);
 787 }
 788
 789 STATIC int
 790 xfs_fs_fill_super(
 791         struct super_block      *sb,
 792         void                    *data,
 793         int                     silent)
 794 {
 795         struct bhv_vnode        *rootvp;
 796         struct bhv_vfs          *vfsp = vfs_allocate(sb);
 797         struct xfs_mount_args   *args = xfs_args_allocate(sb, silent);
 798         struct kstatfs          statvfs;
 799         int                     error;
 800
 801         bhv_insert_all_vfsops(vfsp);
 802
 803         error = bhv_vfs_parseargs(vfsp, (char *)data, args, 0);
 804         if (error) {
 805                 bhv_remove_all_vfsops(vfsp, 1);
 806                 goto fail_vfsop;
 807         }
 808
 809         sb_min_blocksize(sb, BBSIZE);
 810         sb->s_export_op = &xfs_export_operations;
 811         sb->s_qcop = &xfs_quotactl_operations;
 812         sb->s_op = &xfs_super_operations;
 813
 814         error = bhv_vfs_mount(vfsp, args, NULL);
 815         if (error) {
 816                 bhv_remove_all_vfsops(vfsp, 1);
 817                 goto fail_vfsop;
 818         }
 819
 820         error = bhv_vfs_statvfs(vfsp, &statvfs, NULL);
 821         if (error)
 822                 goto fail_unmount;
 823
 824         sb->s_dirt = 1;
 825         sb->s_magic = statvfs.f_type;
 826         sb->s_blocksize = statvfs.f_bsize;
 827         sb->s_blocksize_bits = ffs(statvfs.f_bsize) - 1;
 828         sb->s_maxbytes = xfs_max_file_offset(sb->s_blocksize_bits);
 829         sb->s_time_gran = 1;
 830         set_posix_acl_flag(sb);
 831
 832         error = bhv_vfs_root(vfsp, &rootvp);
 833         if (error)
 834                 goto fail_unmount;
 835
 836         sb->s_root = d_alloc_root(vn_to_inode(rootvp));
 837         if (!sb->s_root) {
 838                 error = ENOMEM;
 839                 goto fail_vnrele;
 840         }
 841         if (is_bad_inode(sb->s_root->d_inode)) {
 842                 error = EINVAL;
 843                 goto fail_vnrele;
 844         }
 845         if ((error = xfs_fs_start_syncd(vfsp)))
 846                 goto fail_vnrele;
 847         vn_trace_exit(rootvp, __FUNCTION__, (inst_t *)__return_address);
 848
 849         kmem_free(args, sizeof(*args));
 850         return 0;
 851
 852 fail_vnrele:
 853         if (sb->s_root) {
 854                 dput(sb->s_root);
 855                 sb->s_root = NULL;
 856         } else {
 857                 VN_RELE(rootvp);
 858         }
 859
 860 fail_unmount:
 861         bhv_vfs_unmount(vfsp, 0, NULL);
 862
 863 fail_vfsop:
 864         vfs_deallocate(vfsp);
 865         kmem_free(args, sizeof(*args));
 866         return -error;
 867 }
 868
 869 STATIC int
 870 xfs_fs_get_sb(
 871         struct file_system_type *fs_type,
 872         int                     flags,
 873         const char              *dev_name,
 874         void                    *data,
 875         struct vfsmount         *mnt)
 876 {
 877         return get_sb_bdev(fs_type, flags, dev_name, data, xfs_fs_fill_super,
 878                            mnt);
 879 }
 880
 881 static struct super_operations xfs_super_operations = {
 882         .alloc_inode            = xfs_fs_alloc_inode,
 883         .destroy_inode          = xfs_fs_destroy_inode,
 884         .write_inode            = xfs_fs_write_inode,
 885         .clear_inode            = xfs_fs_clear_inode,
 886         .put_super              = xfs_fs_put_super,
 887         .write_super            = xfs_fs_write_super,
 888         .sync_fs                = xfs_fs_sync_super,
 889         .write_super_lockfs     = xfs_fs_lockfs,
 890         .statfs                 = xfs_fs_statfs,
 891         .remount_fs             = xfs_fs_remount,
 892         .show_options           = xfs_fs_show_options,
 893 };
 894
 895 static struct quotactl_ops xfs_quotactl_operations = {
 896         .quota_sync             = xfs_fs_quotasync,
 897         .get_xstate             = xfs_fs_getxstate,
 898         .set_xstate             = xfs_fs_setxstate,
 899         .get_xquota             = xfs_fs_getxquota,
 900         .set_xquota             = xfs_fs_setxquota,
 901 };
 902
 903 static struct file_system_type xfs_fs_type = {
 904         .owner                  = THIS_MODULE,
 905         .name                   = "xfs",
 906         .get_sb                 = xfs_fs_get_sb,
 907         .kill_sb                = kill_block_super,
 908         .fs_flags               = FS_REQUIRES_DEV,
 909 };
 910
 911
 912 STATIC int __init
 913 init_xfs_fs( void )
 914 {
 915         int                     error;
 916         struct sysinfo          si;
 917         static char             message[] __initdata = KERN_INFO \
 918                 XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled\n";
 919
 920         printk(message);
 921
 922         si_meminfo(&si);
 923         xfs_physmem = si.totalram;
 924
 925         ktrace_init(64);
 926
 927         error = xfs_init_zones();
 928         if (error < 0)
 929                 goto undo_zones;
 930
 931         error = xfs_buf_init();
 932         if (error < 0)
 933                 goto undo_buffers;
 934
 935         vn_init();
 936         xfs_init();
 937         uuid_init();
 938         vfs_initquota();
 939
 940         error = register_filesystem(&xfs_fs_type);
 941         if (error)
 942                 goto undo_register;
 943         return 0;
 944
 945 undo_register:
 946         xfs_buf_terminate();
 947
 948 undo_buffers:
 949         xfs_destroy_zones();
 950
 951 undo_zones:
 952         return error;
 953 }
 954
 955 STATIC void __exit
 956 exit_xfs_fs( void )
 957 {
 958         vfs_exitquota();
 959         unregister_filesystem(&xfs_fs_type);
 960         xfs_cleanup();
 961         xfs_buf_terminate();
 962         xfs_destroy_zones();
 963         ktrace_uninit();
 964 }
 965
 966 module_init(init_xfs_fs);
 967 module_exit(exit_xfs_fs);
 968
 969 MODULE_AUTHOR("Silicon Graphics, Inc.");
 970 MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled");
 971 MODULE_LICENSE("GPL");