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_vfsops.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 #include <linux/freezer.h>
  60
  61 static struct quotactl_ops xfs_quotactl_operations;
  62 static struct super_operations xfs_super_operations;
  63 static kmem_zone_t *xfs_vnode_zone;
  64 static kmem_zone_t *xfs_ioend_zone;
  65 mempool_t *xfs_ioend_pool;
  66
  67 STATIC struct xfs_mount_args *
  68 xfs_args_allocate(
  69         struct super_block      *sb,
  70         int                     silent)
  71 {
  72         struct xfs_mount_args   *args;
  73
  74         args = kmem_zalloc(sizeof(struct xfs_mount_args), KM_SLEEP);
  75         args->logbufs = args->logbufsize = -1;
  76         strncpy(args->fsname, sb->s_id, MAXNAMELEN);
  77
  78         /* Copy the already-parsed mount(2) flags we're interested in */
  79         if (sb->s_flags & MS_DIRSYNC)
  80                 args->flags |= XFSMNT_DIRSYNC;
  81         if (sb->s_flags & MS_SYNCHRONOUS)
  82                 args->flags |= XFSMNT_WSYNC;
  83         if (silent)
  84                 args->flags |= XFSMNT_QUIET;
  85         args->flags |= XFSMNT_32BITINODES;
  86
  87         return args;
  88 }
  89
  90 __uint64_t
  91 xfs_max_file_offset(
  92         unsigned int            blockshift)
  93 {
  94         unsigned int            pagefactor = 1;
  95         unsigned int            bitshift = BITS_PER_LONG - 1;
  96
  97         /* Figure out maximum filesize, on Linux this can depend on
  98          * the filesystem blocksize (on 32 bit platforms).
  99          * __block_prepare_write does this in an [unsigned] long...
 100          *      page->index << (PAGE_CACHE_SHIFT - bbits)
 101          * So, for page sized blocks (4K on 32 bit platforms),
 102          * this wraps at around 8Tb (hence MAX_LFS_FILESIZE which is
 103          *      (((u64)PAGE_CACHE_SIZE << (BITS_PER_LONG-1))-1)
 104          * but for smaller blocksizes it is less (bbits = log2 bsize).
 105          * Note1: get_block_t takes a long (implicit cast from above)
 106          * Note2: The Large Block Device (LBD and HAVE_SECTOR_T) patch
 107          * can optionally convert the [unsigned] long from above into
 108          * an [unsigned] long long.
 109          */
 110
 111 #if BITS_PER_LONG == 32
 112 # if defined(CONFIG_LBD)
 113         ASSERT(sizeof(sector_t) == 8);
 114         pagefactor = PAGE_CACHE_SIZE;
 115         bitshift = BITS_PER_LONG;
 116 # else
 117         pagefactor = PAGE_CACHE_SIZE >> (PAGE_CACHE_SHIFT - blockshift);
 118 # endif
 119 #endif
 120
 121         return (((__uint64_t)pagefactor) << bitshift) - 1;
 122 }
 123
 124 STATIC_INLINE void
 125 xfs_set_inodeops(
 126         struct inode            *inode)
 127 {
 128         switch (inode->i_mode & S_IFMT) {
 129         case S_IFREG:
 130                 inode->i_op = &xfs_inode_operations;
 131                 inode->i_fop = &xfs_file_operations;
 132                 inode->i_mapping->a_ops = &xfs_address_space_operations;
 133                 break;
 134         case S_IFDIR:
 135                 inode->i_op = &xfs_dir_inode_operations;
 136                 inode->i_fop = &xfs_dir_file_operations;
 137                 break;
 138         case S_IFLNK:
 139                 inode->i_op = &xfs_symlink_inode_operations;
 140                 if (inode->i_blocks)
 141                         inode->i_mapping->a_ops = &xfs_address_space_operations;
 142                 break;
 143         default:
 144                 inode->i_op = &xfs_inode_operations;
 145                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
 146                 break;
 147         }
 148 }
 149
 150 STATIC_INLINE void
 151 xfs_revalidate_inode(
 152         xfs_mount_t             *mp,
 153         bhv_vnode_t             *vp,
 154         xfs_inode_t             *ip)
 155 {
 156         struct inode            *inode = vn_to_inode(vp);
 157
 158         inode->i_mode   = ip->i_d.di_mode;
 159         inode->i_nlink  = ip->i_d.di_nlink;
 160         inode->i_uid    = ip->i_d.di_uid;
 161         inode->i_gid    = ip->i_d.di_gid;
 162
 163         switch (inode->i_mode & S_IFMT) {
 164         case S_IFBLK:
 165         case S_IFCHR:
 166                 inode->i_rdev =
 167                         MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff,
 168                               sysv_minor(ip->i_df.if_u2.if_rdev));
 169                 break;
 170         default:
 171                 inode->i_rdev = 0;
 172                 break;
 173         }
 174
 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         xfs_iflags_clear(ip, XFS_IMODIFIED);
 202 }
 203
 204 void
 205 xfs_initialize_vnode(
 206         struct xfs_mount        *mp,
 207         bhv_vnode_t             *vp,
 208         struct xfs_inode        *ip)
 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 && (inode->i_state & I_NEW)) {
 225                 xfs_revalidate_inode(mp, 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(XFS_I(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         xfs_inode_t             *ip = XFS_I(inode);
 429
 430         /*
 431          * ip can be null when xfs_iget_core calls xfs_idestroy if we
 432          * find an inode with di_mode == 0 but without IGET_CREATE set.
 433          */
 434         if (ip) {
 435                 vn_trace_entry(ip, __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                 xfs_inactive(ip);
 443                 xfs_iflags_clear(ip, XFS_IMODIFIED);
 444                 if (xfs_reclaim(ip))
 445                         panic("%s: cannot reclaim 0x%p\n", __FUNCTION__, inode);
 446         }
 447
 448         ASSERT(XFS_I(inode) == NULL);
 449 }
 450
 451 /*
 452  * Enqueue a work item to be picked up by the vfs xfssyncd thread.
 453  * Doing this has two advantages:
 454  * - It saves on stack space, which is tight in certain situations
 455  * - It can be used (with care) as a mechanism to avoid deadlocks.
 456  * Flushing while allocating in a full filesystem requires both.
 457  */
 458 STATIC void
 459 xfs_syncd_queue_work(
 460         struct bhv_vfs  *vfs,
 461         void            *data,
 462         void            (*syncer)(bhv_vfs_t *, void *))
 463 {
 464         struct bhv_vfs_sync_work *work;
 465
 466         work = kmem_alloc(sizeof(struct bhv_vfs_sync_work), KM_SLEEP);
 467         INIT_LIST_HEAD(&work->w_list);
 468         work->w_syncer = syncer;
 469         work->w_data = data;
 470         work->w_vfs = vfs;
 471         spin_lock(&vfs->vfs_sync_lock);
 472         list_add_tail(&work->w_list, &vfs->vfs_sync_list);
 473         spin_unlock(&vfs->vfs_sync_lock);
 474         wake_up_process(vfs->vfs_sync_task);
 475 }
 476
 477 /*
 478  * Flush delayed allocate data, attempting to free up reserved space
 479  * from existing allocations.  At this point a new allocation attempt
 480  * has failed with ENOSPC and we are in the process of scratching our
 481  * heads, looking about for more room...
 482  */
 483 STATIC void
 484 xfs_flush_inode_work(
 485         bhv_vfs_t       *vfs,
 486         void            *inode)
 487 {
 488         filemap_flush(((struct inode *)inode)->i_mapping);
 489         iput((struct inode *)inode);
 490 }
 491
 492 void
 493 xfs_flush_inode(
 494         xfs_inode_t     *ip)
 495 {
 496         struct inode    *inode = vn_to_inode(XFS_ITOV(ip));
 497         struct bhv_vfs  *vfs = XFS_MTOVFS(ip->i_mount);
 498
 499         igrab(inode);
 500         xfs_syncd_queue_work(vfs, inode, xfs_flush_inode_work);
 501         delay(msecs_to_jiffies(500));
 502 }
 503
 504 /*
 505  * This is the "bigger hammer" version of xfs_flush_inode_work...
 506  * (IOW, "If at first you don't succeed, use a Bigger Hammer").
 507  */
 508 STATIC void
 509 xfs_flush_device_work(
 510         bhv_vfs_t       *vfs,
 511         void            *inode)
 512 {
 513         sync_blockdev(vfs->vfs_super->s_bdev);
 514         iput((struct inode *)inode);
 515 }
 516
 517 void
 518 xfs_flush_device(
 519         xfs_inode_t     *ip)
 520 {
 521         struct inode    *inode = vn_to_inode(XFS_ITOV(ip));
 522         struct bhv_vfs  *vfs = XFS_MTOVFS(ip->i_mount);
 523
 524         igrab(inode);
 525         xfs_syncd_queue_work(vfs, inode, xfs_flush_device_work);
 526         delay(msecs_to_jiffies(500));
 527         xfs_log_force(ip->i_mount, (xfs_lsn_t)0, XFS_LOG_FORCE|XFS_LOG_SYNC);
 528 }
 529
 530 STATIC void
 531 vfs_sync_worker(
 532         bhv_vfs_t       *vfsp,
 533         void            *unused)
 534 {
 535         int             error;
 536
 537         if (!(vfsp->vfs_flag & VFS_RDONLY))
 538                 error = xfs_sync(XFS_VFSTOM(vfsp), SYNC_FSDATA | SYNC_BDFLUSH | \
 539                                         SYNC_ATTR | SYNC_REFCACHE | SYNC_SUPER);
 540         vfsp->vfs_sync_seq++;
 541         wake_up(&vfsp->vfs_wait_single_sync_task);
 542 }
 543
 544 STATIC int
 545 xfssyncd(
 546         void                    *arg)
 547 {
 548         long                    timeleft;
 549         bhv_vfs_t               *vfsp = (bhv_vfs_t *) arg;
 550         bhv_vfs_sync_work_t     *work, *n;
 551         LIST_HEAD               (tmp);
 552
 553         set_freezable();
 554         timeleft = xfs_syncd_centisecs * msecs_to_jiffies(10);
 555         for (;;) {
 556                 timeleft = schedule_timeout_interruptible(timeleft);
 557                 /* swsusp */
 558                 try_to_freeze();
 559                 if (kthread_should_stop() && list_empty(&vfsp->vfs_sync_list))
 560                         break;
 561
 562                 spin_lock(&vfsp->vfs_sync_lock);
 563                 /*
 564                  * We can get woken by laptop mode, to do a sync -
 565                  * that's the (only!) case where the list would be
 566                  * empty with time remaining.
 567                  */
 568                 if (!timeleft || list_empty(&vfsp->vfs_sync_list)) {
 569                         if (!timeleft)
 570                                 timeleft = xfs_syncd_centisecs *
 571                                                         msecs_to_jiffies(10);
 572                         INIT_LIST_HEAD(&vfsp->vfs_sync_work.w_list);
 573                         list_add_tail(&vfsp->vfs_sync_work.w_list,
 574                                         &vfsp->vfs_sync_list);
 575                 }
 576                 list_for_each_entry_safe(work, n, &vfsp->vfs_sync_list, w_list)
 577                         list_move(&work->w_list, &tmp);
 578                 spin_unlock(&vfsp->vfs_sync_lock);
 579
 580                 list_for_each_entry_safe(work, n, &tmp, w_list) {
 581                         (*work->w_syncer)(vfsp, work->w_data);
 582                         list_del(&work->w_list);
 583                         if (work == &vfsp->vfs_sync_work)
 584                                 continue;
 585                         kmem_free(work, sizeof(struct bhv_vfs_sync_work));
 586                 }
 587         }
 588
 589         return 0;
 590 }
 591
 592 STATIC int
 593 xfs_fs_start_syncd(
 594         bhv_vfs_t               *vfsp)
 595 {
 596         vfsp->vfs_sync_work.w_syncer = vfs_sync_worker;
 597         vfsp->vfs_sync_work.w_vfs = vfsp;
 598         vfsp->vfs_sync_task = kthread_run(xfssyncd, vfsp, "xfssyncd");
 599         if (IS_ERR(vfsp->vfs_sync_task))
 600                 return -PTR_ERR(vfsp->vfs_sync_task);
 601         return 0;
 602 }
 603
 604 STATIC void
 605 xfs_fs_stop_syncd(
 606         bhv_vfs_t               *vfsp)
 607 {
 608         kthread_stop(vfsp->vfs_sync_task);
 609 }
 610
 611 STATIC void
 612 xfs_fs_put_super(
 613         struct super_block      *sb)
 614 {
 615         bhv_vfs_t               *vfsp = vfs_from_sb(sb);
 616         struct xfs_mount        *mp = XFS_M(sb);
 617         int                     error;
 618
 619         xfs_fs_stop_syncd(vfsp);
 620         xfs_sync(mp, SYNC_ATTR | SYNC_DELWRI);
 621         error = xfs_unmount(mp, 0, NULL);
 622         if (error) {
 623                 printk("XFS: unmount got error=%d\n", error);
 624                 printk("%s: vfs=0x%p left dangling!\n", __FUNCTION__, vfsp);
 625         } else {
 626                 vfs_deallocate(vfsp);
 627         }
 628 }
 629
 630 STATIC void
 631 xfs_fs_write_super(
 632         struct super_block      *sb)
 633 {
 634         if (!(sb->s_flags & MS_RDONLY))
 635                 xfs_sync(XFS_M(sb), SYNC_FSDATA);
 636         sb->s_dirt = 0;
 637 }
 638
 639 STATIC int
 640 xfs_fs_sync_super(
 641         struct super_block      *sb,
 642         int                     wait)
 643 {
 644         bhv_vfs_t               *vfsp = vfs_from_sb(sb);
 645         struct xfs_mount        *mp = XFS_M(sb);
 646         int                     error;
 647         int                     flags;
 648
 649         if (unlikely(sb->s_frozen == SB_FREEZE_WRITE)) {
 650                 /*
 651                  * First stage of freeze - no more writers will make progress
 652                  * now we are here, so we flush delwri and delalloc buffers
 653                  * here, then wait for all I/O to complete.  Data is frozen at
 654                  * that point. Metadata is not frozen, transactions can still
 655                  * occur here so don't bother flushing the buftarg (i.e
 656                  * SYNC_QUIESCE) because it'll just get dirty again.
 657                  */
 658                 flags = SYNC_DATA_QUIESCE;
 659         } else
 660                 flags = SYNC_FSDATA | (wait ? SYNC_WAIT : 0);
 661
 662         error = xfs_sync(mp, flags);
 663         sb->s_dirt = 0;
 664
 665         if (unlikely(laptop_mode)) {
 666                 int     prev_sync_seq = vfsp->vfs_sync_seq;
 667
 668                 /*
 669                  * The disk must be active because we're syncing.
 670                  * We schedule xfssyncd now (now that the disk is
 671                  * active) instead of later (when it might not be).
 672                  */
 673                 wake_up_process(vfsp->vfs_sync_task);
 674                 /*
 675                  * We have to wait for the sync iteration to complete.
 676                  * If we don't, the disk activity caused by the sync
 677                  * will come after the sync is completed, and that
 678                  * triggers another sync from laptop mode.
 679                  */
 680                 wait_event(vfsp->vfs_wait_single_sync_task,
 681                                 vfsp->vfs_sync_seq != prev_sync_seq);
 682         }
 683
 684         return -error;
 685 }
 686
 687 STATIC int
 688 xfs_fs_statfs(
 689         struct dentry           *dentry,
 690         struct kstatfs          *statp)
 691 {
 692         return -xfs_statvfs(XFS_M(dentry->d_sb), statp,
 693                                 vn_from_inode(dentry->d_inode));
 694 }
 695
 696 STATIC int
 697 xfs_fs_remount(
 698         struct super_block      *sb,
 699         int                     *flags,
 700         char                    *options)
 701 {
 702         struct xfs_mount        *mp = XFS_M(sb);
 703         struct xfs_mount_args   *args = xfs_args_allocate(sb, 0);
 704         int                     error;
 705
 706         error = xfs_parseargs(mp, options, args, 1);
 707         if (!error)
 708                 error = xfs_mntupdate(mp, flags, args);
 709         kmem_free(args, sizeof(*args));
 710         return -error;
 711 }
 712
 713 STATIC void
 714 xfs_fs_lockfs(
 715         struct super_block      *sb)
 716 {
 717         xfs_freeze(XFS_M(sb));
 718 }
 719
 720 STATIC int
 721 xfs_fs_show_options(
 722         struct seq_file         *m,
 723         struct vfsmount         *mnt)
 724 {
 725         return -xfs_showargs(XFS_M(mnt->mnt_sb), m);
 726 }
 727
 728 STATIC int
 729 xfs_fs_quotasync(
 730         struct super_block      *sb,
 731         int                     type)
 732 {
 733         return -XFS_QM_QUOTACTL(XFS_M(sb), Q_XQUOTASYNC, 0, NULL);
 734 }
 735
 736 STATIC int
 737 xfs_fs_getxstate(
 738         struct super_block      *sb,
 739         struct fs_quota_stat    *fqs)
 740 {
 741         return -XFS_QM_QUOTACTL(XFS_M(sb), Q_XGETQSTAT, 0, (caddr_t)fqs);
 742 }
 743
 744 STATIC int
 745 xfs_fs_setxstate(
 746         struct super_block      *sb,
 747         unsigned int            flags,
 748         int                     op)
 749 {
 750         return -XFS_QM_QUOTACTL(XFS_M(sb), op, 0, (caddr_t)&flags);
 751 }
 752
 753 STATIC int
 754 xfs_fs_getxquota(
 755         struct super_block      *sb,
 756         int                     type,
 757         qid_t                   id,
 758         struct fs_disk_quota    *fdq)
 759 {
 760         return -XFS_QM_QUOTACTL(XFS_M(sb),
 761                                  (type == USRQUOTA) ? Q_XGETQUOTA :
 762                                   ((type == GRPQUOTA) ? Q_XGETGQUOTA :
 763                                    Q_XGETPQUOTA), id, (caddr_t)fdq);
 764 }
 765
 766 STATIC int
 767 xfs_fs_setxquota(
 768         struct super_block      *sb,
 769         int                     type,
 770         qid_t                   id,
 771         struct fs_disk_quota    *fdq)
 772 {
 773         return -XFS_QM_QUOTACTL(XFS_M(sb),
 774                                  (type == USRQUOTA) ? Q_XSETQLIM :
 775                                   ((type == GRPQUOTA) ? Q_XSETGQLIM :
 776                                    Q_XSETPQLIM), id, (caddr_t)fdq);
 777 }
 778
 779 STATIC int
 780 xfs_fs_fill_super(
 781         struct super_block      *sb,
 782         void                    *data,
 783         int                     silent)
 784 {
 785         struct inode            *rootvp;
 786         struct bhv_vfs          *vfsp = vfs_allocate(sb);
 787         struct xfs_mount        *mp = NULL;
 788         struct xfs_mount_args   *args = xfs_args_allocate(sb, silent);
 789         struct kstatfs          statvfs;
 790         int                     error;
 791
 792         mp = xfs_mount_init();
 793         mp->m_vfsp = vfsp;
 794         vfsp->vfs_mount = mp;
 795
 796         error = xfs_parseargs(mp, (char *)data, args, 0);
 797         if (error)
 798                 goto fail_vfsop;
 799
 800         sb_min_blocksize(sb, BBSIZE);
 801         sb->s_export_op = &xfs_export_operations;
 802         sb->s_qcop = &xfs_quotactl_operations;
 803         sb->s_op = &xfs_super_operations;
 804
 805         error = xfs_mount(mp, args, NULL);
 806         if (error)
 807                 goto fail_vfsop;
 808
 809         error = xfs_statvfs(mp, &statvfs, NULL);
 810         if (error)
 811                 goto fail_unmount;
 812
 813         sb->s_dirt = 1;
 814         sb->s_magic = statvfs.f_type;
 815         sb->s_blocksize = statvfs.f_bsize;
 816         sb->s_blocksize_bits = ffs(statvfs.f_bsize) - 1;
 817         sb->s_maxbytes = xfs_max_file_offset(sb->s_blocksize_bits);
 818         sb->s_time_gran = 1;
 819         set_posix_acl_flag(sb);
 820
 821         error = xfs_root(mp, &rootvp);
 822         if (error)
 823                 goto fail_unmount;
 824
 825         sb->s_root = d_alloc_root(vn_to_inode(rootvp));
 826         if (!sb->s_root) {
 827                 error = ENOMEM;
 828                 goto fail_vnrele;
 829         }
 830         if (is_bad_inode(sb->s_root->d_inode)) {
 831                 error = EINVAL;
 832                 goto fail_vnrele;
 833         }
 834         if ((error = xfs_fs_start_syncd(vfsp)))
 835                 goto fail_vnrele;
 836         vn_trace_exit(XFS_I(sb->s_root->d_inode), __FUNCTION__,
 837                         (inst_t *)__return_address);
 838
 839         kmem_free(args, sizeof(*args));
 840         return 0;
 841
 842 fail_vnrele:
 843         if (sb->s_root) {
 844                 dput(sb->s_root);
 845                 sb->s_root = NULL;
 846         } else {
 847                 VN_RELE(rootvp);
 848         }
 849
 850 fail_unmount:
 851         xfs_unmount(mp, 0, NULL);
 852
 853 fail_vfsop:
 854         vfs_deallocate(vfsp);
 855         kmem_free(args, sizeof(*args));
 856         return -error;
 857 }
 858
 859 STATIC int
 860 xfs_fs_get_sb(
 861         struct file_system_type *fs_type,
 862         int                     flags,
 863         const char              *dev_name,
 864         void                    *data,
 865         struct vfsmount         *mnt)
 866 {
 867         return get_sb_bdev(fs_type, flags, dev_name, data, xfs_fs_fill_super,
 868                            mnt);
 869 }
 870
 871 static struct super_operations xfs_super_operations = {
 872         .alloc_inode            = xfs_fs_alloc_inode,
 873         .destroy_inode          = xfs_fs_destroy_inode,
 874         .write_inode            = xfs_fs_write_inode,
 875         .clear_inode            = xfs_fs_clear_inode,
 876         .put_super              = xfs_fs_put_super,
 877         .write_super            = xfs_fs_write_super,
 878         .sync_fs                = xfs_fs_sync_super,
 879         .write_super_lockfs     = xfs_fs_lockfs,
 880         .statfs                 = xfs_fs_statfs,
 881         .remount_fs             = xfs_fs_remount,
 882         .show_options           = xfs_fs_show_options,
 883 };
 884
 885 static struct quotactl_ops xfs_quotactl_operations = {
 886         .quota_sync             = xfs_fs_quotasync,
 887         .get_xstate             = xfs_fs_getxstate,
 888         .set_xstate             = xfs_fs_setxstate,
 889         .get_xquota             = xfs_fs_getxquota,
 890         .set_xquota             = xfs_fs_setxquota,
 891 };
 892
 893 static struct file_system_type xfs_fs_type = {
 894         .owner                  = THIS_MODULE,
 895         .name                   = "xfs",
 896         .get_sb                 = xfs_fs_get_sb,
 897         .kill_sb                = kill_block_super,
 898         .fs_flags               = FS_REQUIRES_DEV,
 899 };
 900
 901
 902 STATIC int __init
 903 init_xfs_fs( void )
 904 {
 905         int                     error;
 906         static char             message[] __initdata = KERN_INFO \
 907                 XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled\n";
 908
 909         printk(message);
 910
 911         ktrace_init(64);
 912
 913         error = xfs_init_zones();
 914         if (error < 0)
 915                 goto undo_zones;
 916
 917         error = xfs_buf_init();
 918         if (error < 0)
 919                 goto undo_buffers;
 920
 921         vn_init();
 922         xfs_init();
 923         uuid_init();
 924         vfs_initquota();
 925
 926         error = register_filesystem(&xfs_fs_type);
 927         if (error)
 928                 goto undo_register;
 929         return 0;
 930
 931 undo_register:
 932         xfs_buf_terminate();
 933
 934 undo_buffers:
 935         xfs_destroy_zones();
 936
 937 undo_zones:
 938         return error;
 939 }
 940
 941 STATIC void __exit
 942 exit_xfs_fs( void )
 943 {
 944         vfs_exitquota();
 945         unregister_filesystem(&xfs_fs_type);
 946         xfs_cleanup();
 947         xfs_buf_terminate();
 948         xfs_destroy_zones();
 949         ktrace_uninit();
 950 }
 951
 952 module_init(init_xfs_fs);
 953 module_exit(exit_xfs_fs);
 954
 955 MODULE_AUTHOR("Silicon Graphics, Inc.");
 956 MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled");
 957 MODULE_LICENSE("GPL");