err.no Git - linux-2.6/blob - fs/xfs/linux-2.6/xfs_super.c

   1 /*
   2  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
   3  * All Rights Reserved.
   4  *
   5  * This program is free software; you can redistribute it and/or
   6  * modify it under the terms of the GNU General Public License as
   7  * published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it would be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write the Free Software Foundation,
  16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17  */
  18 #include "xfs.h"
  19 #include "xfs_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_dir.h"
  27 #include "xfs_dir2.h"
  28 #include "xfs_alloc.h"
  29 #include "xfs_dmapi.h"
  30 #include "xfs_quota.h"
  31 #include "xfs_mount.h"
  32 #include "xfs_bmap_btree.h"
  33 #include "xfs_alloc_btree.h"
  34 #include "xfs_ialloc_btree.h"
  35 #include "xfs_dir_sf.h"
  36 #include "xfs_dir2_sf.h"
  37 #include "xfs_attr_sf.h"
  38 #include "xfs_dinode.h"
  39 #include "xfs_inode.h"
  40 #include "xfs_btree.h"
  41 #include "xfs_ialloc.h"
  42 #include "xfs_bmap.h"
  43 #include "xfs_rtalloc.h"
  44 #include "xfs_error.h"
  45 #include "xfs_itable.h"
  46 #include "xfs_rw.h"
  47 #include "xfs_acl.h"
  48 #include "xfs_cap.h"
  49 #include "xfs_mac.h"
  50 #include "xfs_attr.h"
  51 #include "xfs_buf_item.h"
  52 #include "xfs_utils.h"
  53 #include "xfs_version.h"
  54
  55 #include <linux/namei.h>
  56 #include <linux/init.h>
  57 #include <linux/mount.h>
  58 #include <linux/mempool.h>
  59 #include <linux/writeback.h>
  60 #include <linux/kthread.h>
  61
  62 STATIC struct quotactl_ops xfs_quotactl_operations;
  63 STATIC struct super_operations xfs_super_operations;
  64 STATIC kmem_zone_t *xfs_vnode_zone;
  65 STATIC kmem_zone_t *xfs_ioend_zone;
  66 mempool_t *xfs_ioend_pool;
  67
  68 STATIC struct xfs_mount_args *
  69 xfs_args_allocate(
  70         struct super_block      *sb,
  71         int                     silent)
  72 {
  73         struct xfs_mount_args   *args;
  74
  75         args = kmem_zalloc(sizeof(struct xfs_mount_args), KM_SLEEP);
  76         args->logbufs = args->logbufsize = -1;
  77         strncpy(args->fsname, sb->s_id, MAXNAMELEN);
  78
  79         /* Copy the already-parsed mount(2) flags we're interested in */
  80         if (sb->s_flags & MS_DIRSYNC)
  81                 args->flags |= XFSMNT_DIRSYNC;
  82         if (sb->s_flags & MS_SYNCHRONOUS)
  83                 args->flags |= XFSMNT_WSYNC;
  84         if (silent)
  85                 args->flags |= XFSMNT_QUIET;
  86         args->flags |= XFSMNT_32BITINODES;
  87
  88         return args;
  89 }
  90
  91 __uint64_t
  92 xfs_max_file_offset(
  93         unsigned int            blockshift)
  94 {
  95         unsigned int            pagefactor = 1;
  96         unsigned int            bitshift = BITS_PER_LONG - 1;
  97
  98         /* Figure out maximum filesize, on Linux this can depend on
  99          * the filesystem blocksize (on 32 bit platforms).
 100          * __block_prepare_write does this in an [unsigned] long...
 101          *      page->index << (PAGE_CACHE_SHIFT - bbits)
 102          * So, for page sized blocks (4K on 32 bit platforms),
 103          * this wraps at around 8Tb (hence MAX_LFS_FILESIZE which is
 104          *      (((u64)PAGE_CACHE_SIZE << (BITS_PER_LONG-1))-1)
 105          * but for smaller blocksizes it is less (bbits = log2 bsize).
 106          * Note1: get_block_t takes a long (implicit cast from above)
 107          * Note2: The Large Block Device (LBD and HAVE_SECTOR_T) patch
 108          * can optionally convert the [unsigned] long from above into
 109          * an [unsigned] long long.
 110          */
 111
 112 #if BITS_PER_LONG == 32
 113 # if defined(CONFIG_LBD)
 114         ASSERT(sizeof(sector_t) == 8);
 115         pagefactor = PAGE_CACHE_SIZE;
 116         bitshift = BITS_PER_LONG;
 117 # else
 118         pagefactor = PAGE_CACHE_SIZE >> (PAGE_CACHE_SHIFT - blockshift);
 119 # endif
 120 #endif
 121
 122         return (((__uint64_t)pagefactor) << bitshift) - 1;
 123 }
 124
 125 STATIC __inline__ void
 126 xfs_set_inodeops(
 127         struct inode            *inode)
 128 {
 129         switch (inode->i_mode & S_IFMT) {
 130         case S_IFREG:
 131                 inode->i_op = &xfs_inode_operations;
 132                 inode->i_fop = &xfs_file_operations;
 133                 inode->i_mapping->a_ops = &xfs_address_space_operations;
 134                 break;
 135         case S_IFDIR:
 136                 inode->i_op = &xfs_dir_inode_operations;
 137                 inode->i_fop = &xfs_dir_file_operations;
 138                 break;
 139         case S_IFLNK:
 140                 inode->i_op = &xfs_symlink_inode_operations;
 141                 if (inode->i_blocks)
 142                         inode->i_mapping->a_ops = &xfs_address_space_operations;
 143                 break;
 144         default:
 145                 inode->i_op = &xfs_inode_operations;
 146                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
 147                 break;
 148         }
 149 }
 150
 151 STATIC __inline__ void
 152 xfs_revalidate_inode(
 153         xfs_mount_t             *mp,
 154         vnode_t                 *vp,
 155         xfs_inode_t             *ip)
 156 {
 157         struct inode            *inode = vn_to_inode(vp);
 158
 159         inode->i_mode   = ip->i_d.di_mode;
 160         inode->i_nlink  = ip->i_d.di_nlink;
 161         inode->i_uid    = ip->i_d.di_uid;
 162         inode->i_gid    = ip->i_d.di_gid;
 163
 164         switch (inode->i_mode & S_IFMT) {
 165         case S_IFBLK:
 166         case S_IFCHR:
 167                 inode->i_rdev =
 168                         MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff,
 169                               sysv_minor(ip->i_df.if_u2.if_rdev));
 170                 break;
 171         default:
 172                 inode->i_rdev = 0;
 173                 break;
 174         }
 175
 176         inode->i_blksize = xfs_preferred_iosize(mp);
 177         inode->i_generation = ip->i_d.di_gen;
 178         i_size_write(inode, ip->i_d.di_size);
 179         inode->i_blocks =
 180                 XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
 181         inode->i_atime.tv_sec   = ip->i_d.di_atime.t_sec;
 182         inode->i_atime.tv_nsec  = ip->i_d.di_atime.t_nsec;
 183         inode->i_mtime.tv_sec   = ip->i_d.di_mtime.t_sec;
 184         inode->i_mtime.tv_nsec  = ip->i_d.di_mtime.t_nsec;
 185         inode->i_ctime.tv_sec   = ip->i_d.di_ctime.t_sec;
 186         inode->i_ctime.tv_nsec  = ip->i_d.di_ctime.t_nsec;
 187         if (ip->i_d.di_flags & XFS_DIFLAG_IMMUTABLE)
 188                 inode->i_flags |= S_IMMUTABLE;
 189         else
 190                 inode->i_flags &= ~S_IMMUTABLE;
 191         if (ip->i_d.di_flags & XFS_DIFLAG_APPEND)
 192                 inode->i_flags |= S_APPEND;
 193         else
 194                 inode->i_flags &= ~S_APPEND;
 195         if (ip->i_d.di_flags & XFS_DIFLAG_SYNC)
 196                 inode->i_flags |= S_SYNC;
 197         else
 198                 inode->i_flags &= ~S_SYNC;
 199         if (ip->i_d.di_flags & XFS_DIFLAG_NOATIME)
 200                 inode->i_flags |= S_NOATIME;
 201         else
 202                 inode->i_flags &= ~S_NOATIME;
 203         vp->v_flag &= ~VMODIFIED;
 204 }
 205
 206 void
 207 xfs_initialize_vnode(
 208         bhv_desc_t              *bdp,
 209         vnode_t                 *vp,
 210         bhv_desc_t              *inode_bhv,
 211         int                     unlock)
 212 {
 213         xfs_inode_t             *ip = XFS_BHVTOI(inode_bhv);
 214         struct inode            *inode = vn_to_inode(vp);
 215
 216         if (!inode_bhv->bd_vobj) {
 217                 vp->v_vfsp = bhvtovfs(bdp);
 218                 bhv_desc_init(inode_bhv, ip, vp, &xfs_vnodeops);
 219                 bhv_insert(VN_BHV_HEAD(vp), inode_bhv);
 220         }
 221
 222         /*
 223          * We need to set the ops vectors, and unlock the inode, but if
 224          * we have been called during the new inode create process, it is
 225          * too early to fill in the Linux inode.  We will get called a
 226          * second time once the inode is properly set up, and then we can
 227          * finish our work.
 228          */
 229         if (ip->i_d.di_mode != 0 && unlock && (inode->i_state & I_NEW)) {
 230                 xfs_revalidate_inode(XFS_BHVTOM(bdp), vp, ip);
 231                 xfs_set_inodeops(inode);
 232
 233                 ip->i_flags &= ~XFS_INEW;
 234                 barrier();
 235
 236                 unlock_new_inode(inode);
 237         }
 238 }
 239
 240 int
 241 xfs_blkdev_get(
 242         xfs_mount_t             *mp,
 243         const char              *name,
 244         struct block_device     **bdevp)
 245 {
 246         int                     error = 0;
 247
 248         *bdevp = open_bdev_excl(name, 0, mp);
 249         if (IS_ERR(*bdevp)) {
 250                 error = PTR_ERR(*bdevp);
 251                 printk("XFS: Invalid device [%s], error=%d\n", name, error);
 252         }
 253
 254         return -error;
 255 }
 256
 257 void
 258 xfs_blkdev_put(
 259         struct block_device     *bdev)
 260 {
 261         if (bdev)
 262                 close_bdev_excl(bdev);
 263 }
 264
 265 /*
 266  * Try to write out the superblock using barriers.
 267  */
 268 STATIC int
 269 xfs_barrier_test(
 270         xfs_mount_t     *mp)
 271 {
 272         xfs_buf_t       *sbp = xfs_getsb(mp, 0);
 273         int             error;
 274
 275         XFS_BUF_UNDONE(sbp);
 276         XFS_BUF_UNREAD(sbp);
 277         XFS_BUF_UNDELAYWRITE(sbp);
 278         XFS_BUF_WRITE(sbp);
 279         XFS_BUF_UNASYNC(sbp);
 280         XFS_BUF_ORDERED(sbp);
 281
 282         xfsbdstrat(mp, sbp);
 283         error = xfs_iowait(sbp);
 284
 285         /*
 286          * Clear all the flags we set and possible error state in the
 287          * buffer.  We only did the write to try out whether barriers
 288          * worked and shouldn't leave any traces in the superblock
 289          * buffer.
 290          */
 291         XFS_BUF_DONE(sbp);
 292         XFS_BUF_ERROR(sbp, 0);
 293         XFS_BUF_UNORDERED(sbp);
 294
 295         xfs_buf_relse(sbp);
 296         return error;
 297 }
 298
 299 void
 300 xfs_mountfs_check_barriers(xfs_mount_t *mp)
 301 {
 302         int error;
 303
 304         if (mp->m_logdev_targp != mp->m_ddev_targp) {
 305                 xfs_fs_cmn_err(CE_NOTE, mp,
 306                   "Disabling barriers, not supported with external log device");
 307                 mp->m_flags &= ~XFS_MOUNT_BARRIER;
 308                 return;
 309         }
 310
 311         if (mp->m_ddev_targp->bt_bdev->bd_disk->queue->ordered ==
 312                                         QUEUE_ORDERED_NONE) {
 313                 xfs_fs_cmn_err(CE_NOTE, mp,
 314                   "Disabling barriers, not supported by the underlying device");
 315                 mp->m_flags &= ~XFS_MOUNT_BARRIER;
 316                 return;
 317         }
 318
 319         error = xfs_barrier_test(mp);
 320         if (error) {
 321                 xfs_fs_cmn_err(CE_NOTE, mp,
 322                   "Disabling barriers, trial barrier write failed");
 323                 mp->m_flags &= ~XFS_MOUNT_BARRIER;
 324                 return;
 325         }
 326 }
 327
 328 void
 329 xfs_blkdev_issue_flush(
 330         xfs_buftarg_t           *buftarg)
 331 {
 332         blkdev_issue_flush(buftarg->bt_bdev, NULL);
 333 }
 334
 335 STATIC struct inode *
 336 xfs_fs_alloc_inode(
 337         struct super_block      *sb)
 338 {
 339         vnode_t                 *vp;
 340
 341         vp = kmem_zone_alloc(xfs_vnode_zone, KM_SLEEP);
 342         if (unlikely(!vp))
 343                 return NULL;
 344         return vn_to_inode(vp);
 345 }
 346
 347 STATIC void
 348 xfs_fs_destroy_inode(
 349         struct inode            *inode)
 350 {
 351         kmem_zone_free(xfs_vnode_zone, vn_from_inode(inode));
 352 }
 353
 354 STATIC void
 355 xfs_fs_inode_init_once(
 356         void                    *vnode,
 357         kmem_zone_t             *zonep,
 358         unsigned long           flags)
 359 {
 360         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
 361                       SLAB_CTOR_CONSTRUCTOR)
 362                 inode_init_once(vn_to_inode((vnode_t *)vnode));
 363 }
 364
 365 STATIC int
 366 xfs_init_zones(void)
 367 {
 368         xfs_vnode_zone = kmem_zone_init_flags(sizeof(vnode_t), "xfs_vnode_t",
 369                                         KM_ZONE_HWALIGN | KM_ZONE_RECLAIM |
 370                                         KM_ZONE_SPREAD,
 371                                         xfs_fs_inode_init_once);
 372         if (!xfs_vnode_zone)
 373                 goto out;
 374
 375         xfs_ioend_zone = kmem_zone_init(sizeof(xfs_ioend_t), "xfs_ioend");
 376         if (!xfs_ioend_zone)
 377                 goto out_destroy_vnode_zone;
 378
 379         xfs_ioend_pool = mempool_create_slab_pool(4 * MAX_BUF_PER_PAGE,
 380                                                   xfs_ioend_zone);
 381         if (!xfs_ioend_pool)
 382                 goto out_free_ioend_zone;
 383         return 0;
 384
 385  out_free_ioend_zone:
 386         kmem_zone_destroy(xfs_ioend_zone);
 387  out_destroy_vnode_zone:
 388         kmem_zone_destroy(xfs_vnode_zone);
 389  out:
 390         return -ENOMEM;
 391 }
 392
 393 STATIC void
 394 xfs_destroy_zones(void)
 395 {
 396         mempool_destroy(xfs_ioend_pool);
 397         kmem_zone_destroy(xfs_vnode_zone);
 398         kmem_zone_destroy(xfs_ioend_zone);
 399 }
 400
 401 /*
 402  * Attempt to flush the inode, this will actually fail
 403  * if the inode is pinned, but we dirty the inode again
 404  * at the point when it is unpinned after a log write,
 405  * since this is when the inode itself becomes flushable.
 406  */
 407 STATIC int
 408 xfs_fs_write_inode(
 409         struct inode            *inode,
 410         int                     sync)
 411 {
 412         vnode_t                 *vp = vn_from_inode(inode);
 413         int                     error = 0, flags = FLUSH_INODE;
 414
 415         if (vp) {
 416                 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
 417                 if (sync)
 418                         flags |= FLUSH_SYNC;
 419                 VOP_IFLUSH(vp, flags, error);
 420                 if (error == EAGAIN) {
 421                         if (sync)
 422                                 VOP_IFLUSH(vp, flags | FLUSH_LOG, error);
 423                         else
 424                                 error = 0;
 425                 }
 426         }
 427
 428         return -error;
 429 }
 430
 431 STATIC void
 432 xfs_fs_clear_inode(
 433         struct inode            *inode)
 434 {
 435         vnode_t                 *vp = vn_from_inode(inode);
 436         int                     error, cache;
 437
 438         vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
 439
 440         XFS_STATS_INC(vn_rele);
 441         XFS_STATS_INC(vn_remove);
 442         XFS_STATS_INC(vn_reclaim);
 443         XFS_STATS_DEC(vn_active);
 444
 445         /*
 446          * This can happen because xfs_iget_core calls xfs_idestroy if we
 447          * find an inode with di_mode == 0 but without IGET_CREATE set.
 448          */
 449         if (vp->v_fbhv)
 450                 VOP_INACTIVE(vp, NULL, cache);
 451
 452         VN_LOCK(vp);
 453         vp->v_flag &= ~VMODIFIED;
 454         VN_UNLOCK(vp, 0);
 455
 456         if (vp->v_fbhv) {
 457                 VOP_RECLAIM(vp, error);
 458                 if (error)
 459                         panic("vn_purge: cannot reclaim");
 460         }
 461
 462         ASSERT(vp->v_fbhv == NULL);
 463
 464 #ifdef XFS_VNODE_TRACE
 465         ktrace_free(vp->v_trace);
 466 #endif
 467 }
 468
 469 /*
 470  * Enqueue a work item to be picked up by the vfs xfssyncd thread.
 471  * Doing this has two advantages:
 472  * - It saves on stack space, which is tight in certain situations
 473  * - It can be used (with care) as a mechanism to avoid deadlocks.
 474  * Flushing while allocating in a full filesystem requires both.
 475  */
 476 STATIC void
 477 xfs_syncd_queue_work(
 478         struct bhv_vfs  *vfs,
 479         void            *data,
 480         void            (*syncer)(bhv_vfs_t *, void *))
 481 {
 482         struct bhv_vfs_sync_work *work;
 483
 484         work = kmem_alloc(sizeof(struct bhv_vfs_sync_work), KM_SLEEP);
 485         INIT_LIST_HEAD(&work->w_list);
 486         work->w_syncer = syncer;
 487         work->w_data = data;
 488         work->w_vfs = vfs;
 489         spin_lock(&vfs->vfs_sync_lock);
 490         list_add_tail(&work->w_list, &vfs->vfs_sync_list);
 491         spin_unlock(&vfs->vfs_sync_lock);
 492         wake_up_process(vfs->vfs_sync_task);
 493 }
 494
 495 /*
 496  * Flush delayed allocate data, attempting to free up reserved space
 497  * from existing allocations.  At this point a new allocation attempt
 498  * has failed with ENOSPC and we are in the process of scratching our
 499  * heads, looking about for more room...
 500  */
 501 STATIC void
 502 xfs_flush_inode_work(
 503         bhv_vfs_t       *vfs,
 504         void            *inode)
 505 {
 506         filemap_flush(((struct inode *)inode)->i_mapping);
 507         iput((struct inode *)inode);
 508 }
 509
 510 void
 511 xfs_flush_inode(
 512         xfs_inode_t     *ip)
 513 {
 514         struct inode    *inode = vn_to_inode(XFS_ITOV(ip));
 515         struct bhv_vfs  *vfs = XFS_MTOVFS(ip->i_mount);
 516
 517         igrab(inode);
 518         xfs_syncd_queue_work(vfs, inode, xfs_flush_inode_work);
 519         delay(msecs_to_jiffies(500));
 520 }
 521
 522 /*
 523  * This is the "bigger hammer" version of xfs_flush_inode_work...
 524  * (IOW, "If at first you don't succeed, use a Bigger Hammer").
 525  */
 526 STATIC void
 527 xfs_flush_device_work(
 528         bhv_vfs_t       *vfs,
 529         void            *inode)
 530 {
 531         sync_blockdev(vfs->vfs_super->s_bdev);
 532         iput((struct inode *)inode);
 533 }
 534
 535 void
 536 xfs_flush_device(
 537         xfs_inode_t     *ip)
 538 {
 539         struct inode    *inode = vn_to_inode(XFS_ITOV(ip));
 540         struct bhv_vfs  *vfs = XFS_MTOVFS(ip->i_mount);
 541
 542         igrab(inode);
 543         xfs_syncd_queue_work(vfs, inode, xfs_flush_device_work);
 544         delay(msecs_to_jiffies(500));
 545         xfs_log_force(ip->i_mount, (xfs_lsn_t)0, XFS_LOG_FORCE|XFS_LOG_SYNC);
 546 }
 547
 548 STATIC void
 549 vfs_sync_worker(
 550         bhv_vfs_t       *vfsp,
 551         void            *unused)
 552 {
 553         int             error;
 554
 555         if (!(vfsp->vfs_flag & VFS_RDONLY))
 556                 error = bhv_vfs_sync(vfsp, SYNC_FSDATA | SYNC_BDFLUSH | \
 557                                         SYNC_ATTR | SYNC_REFCACHE, NULL);
 558         vfsp->vfs_sync_seq++;
 559         wmb();
 560         wake_up(&vfsp->vfs_wait_single_sync_task);
 561 }
 562
 563 STATIC int
 564 xfssyncd(
 565         void                    *arg)
 566 {
 567         long                    timeleft;
 568         bhv_vfs_t               *vfsp = (bhv_vfs_t *) arg;
 569         bhv_vfs_sync_work_t     *work, *n;
 570         LIST_HEAD               (tmp);
 571
 572         timeleft = xfs_syncd_centisecs * msecs_to_jiffies(10);
 573         for (;;) {
 574                 timeleft = schedule_timeout_interruptible(timeleft);
 575                 /* swsusp */
 576                 try_to_freeze();
 577                 if (kthread_should_stop() && list_empty(&vfsp->vfs_sync_list))
 578                         break;
 579
 580                 spin_lock(&vfsp->vfs_sync_lock);
 581                 /*
 582                  * We can get woken by laptop mode, to do a sync -
 583                  * that's the (only!) case where the list would be
 584                  * empty with time remaining.
 585                  */
 586                 if (!timeleft || list_empty(&vfsp->vfs_sync_list)) {
 587                         if (!timeleft)
 588                                 timeleft = xfs_syncd_centisecs *
 589                                                         msecs_to_jiffies(10);
 590                         INIT_LIST_HEAD(&vfsp->vfs_sync_work.w_list);
 591                         list_add_tail(&vfsp->vfs_sync_work.w_list,
 592                                         &vfsp->vfs_sync_list);
 593                 }
 594                 list_for_each_entry_safe(work, n, &vfsp->vfs_sync_list, w_list)
 595                         list_move(&work->w_list, &tmp);
 596                 spin_unlock(&vfsp->vfs_sync_lock);
 597
 598                 list_for_each_entry_safe(work, n, &tmp, w_list) {
 599                         (*work->w_syncer)(vfsp, work->w_data);
 600                         list_del(&work->w_list);
 601                         if (work == &vfsp->vfs_sync_work)
 602                                 continue;
 603                         kmem_free(work, sizeof(struct bhv_vfs_sync_work));
 604                 }
 605         }
 606
 607         return 0;
 608 }
 609
 610 STATIC int
 611 xfs_fs_start_syncd(
 612         bhv_vfs_t               *vfsp)
 613 {
 614         vfsp->vfs_sync_work.w_syncer = vfs_sync_worker;
 615         vfsp->vfs_sync_work.w_vfs = vfsp;
 616         vfsp->vfs_sync_task = kthread_run(xfssyncd, vfsp, "xfssyncd");
 617         if (IS_ERR(vfsp->vfs_sync_task))
 618                 return -PTR_ERR(vfsp->vfs_sync_task);
 619         return 0;
 620 }
 621
 622 STATIC void
 623 xfs_fs_stop_syncd(
 624         bhv_vfs_t               *vfsp)
 625 {
 626         kthread_stop(vfsp->vfs_sync_task);
 627 }
 628
 629 STATIC void
 630 xfs_fs_put_super(
 631         struct super_block      *sb)
 632 {
 633         bhv_vfs_t               *vfsp = vfs_from_sb(sb);
 634         int                     error;
 635
 636         xfs_fs_stop_syncd(vfsp);
 637         bhv_vfs_sync(vfsp, SYNC_ATTR | SYNC_DELWRI, NULL);
 638         error = bhv_vfs_unmount(vfsp, 0, NULL);
 639         if (error) {
 640                 printk("XFS: unmount got error=%d\n", error);
 641                 printk("%s: vfs=0x%p left dangling!\n", __FUNCTION__, vfsp);
 642         } else {
 643                 vfs_deallocate(vfsp);
 644         }
 645 }
 646
 647 STATIC void
 648 xfs_fs_write_super(
 649         struct super_block      *sb)
 650 {
 651         if (!(sb->s_flags & MS_RDONLY))
 652                 bhv_vfs_sync(vfs_from_sb(sb), SYNC_FSDATA, NULL);
 653         sb->s_dirt = 0;
 654 }
 655
 656 STATIC int
 657 xfs_fs_sync_super(
 658         struct super_block      *sb,
 659         int                     wait)
 660 {
 661         bhv_vfs_t               *vfsp = vfs_from_sb(sb);
 662         int                     error;
 663         int                     flags;
 664
 665         if (unlikely(sb->s_frozen == SB_FREEZE_WRITE))
 666                 flags = SYNC_QUIESCE;
 667         else
 668                 flags = SYNC_FSDATA | (wait ? SYNC_WAIT : 0);
 669
 670         error = bhv_vfs_sync(vfsp, flags, NULL);
 671         sb->s_dirt = 0;
 672
 673         if (unlikely(laptop_mode)) {
 674                 int     prev_sync_seq = vfsp->vfs_sync_seq;
 675
 676                 /*
 677                  * The disk must be active because we're syncing.
 678                  * We schedule xfssyncd now (now that the disk is
 679                  * active) instead of later (when it might not be).
 680                  */
 681                 wake_up_process(vfsp->vfs_sync_task);
 682                 /*
 683                  * We have to wait for the sync iteration to complete.
 684                  * If we don't, the disk activity caused by the sync
 685                  * will come after the sync is completed, and that
 686                  * triggers another sync from laptop mode.
 687                  */
 688                 wait_event(vfsp->vfs_wait_single_sync_task,
 689                                 vfsp->vfs_sync_seq != prev_sync_seq);
 690         }
 691
 692         return -error;
 693 }
 694
 695 STATIC int
 696 xfs_fs_statfs(
 697         struct super_block      *sb,
 698         struct kstatfs          *statp)
 699 {
 700         return -bhv_vfs_statvfs(vfs_from_sb(sb), statp, NULL);
 701 }
 702
 703 STATIC int
 704 xfs_fs_remount(
 705         struct super_block      *sb,
 706         int                     *flags,
 707         char                    *options)
 708 {
 709         bhv_vfs_t               *vfsp = vfs_from_sb(sb);
 710         struct xfs_mount_args   *args = xfs_args_allocate(sb, 0);
 711         int                     error;
 712
 713         error = bhv_vfs_parseargs(vfsp, options, args, 1);
 714         if (!error)
 715                 error = bhv_vfs_mntupdate(vfsp, flags, args);
 716         kmem_free(args, sizeof(*args));
 717         return -error;
 718 }
 719
 720 STATIC void
 721 xfs_fs_lockfs(
 722         struct super_block      *sb)
 723 {
 724         bhv_vfs_freeze(vfs_from_sb(sb));
 725 }
 726
 727 STATIC int
 728 xfs_fs_show_options(
 729         struct seq_file         *m,
 730         struct vfsmount         *mnt)
 731 {
 732         return -bhv_vfs_showargs(vfs_from_sb(mnt->mnt_sb), m);
 733 }
 734
 735 STATIC int
 736 xfs_fs_quotasync(
 737         struct super_block      *sb,
 738         int                     type)
 739 {
 740         return -bhv_vfs_quotactl(vfs_from_sb(sb), Q_XQUOTASYNC, 0, NULL);
 741 }
 742
 743 STATIC int
 744 xfs_fs_getxstate(
 745         struct super_block      *sb,
 746         struct fs_quota_stat    *fqs)
 747 {
 748         return -bhv_vfs_quotactl(vfs_from_sb(sb), Q_XGETQSTAT, 0, (caddr_t)fqs);
 749 }
 750
 751 STATIC int
 752 xfs_fs_setxstate(
 753         struct super_block      *sb,
 754         unsigned int            flags,
 755         int                     op)
 756 {
 757         return -bhv_vfs_quotactl(vfs_from_sb(sb), op, 0, (caddr_t)&flags);
 758 }
 759
 760 STATIC int
 761 xfs_fs_getxquota(
 762         struct super_block      *sb,
 763         int                     type,
 764         qid_t                   id,
 765         struct fs_disk_quota    *fdq)
 766 {
 767         return -bhv_vfs_quotactl(vfs_from_sb(sb),
 768                                  (type == USRQUOTA) ? Q_XGETQUOTA :
 769                                   ((type == GRPQUOTA) ? Q_XGETGQUOTA :
 770                                    Q_XGETPQUOTA), id, (caddr_t)fdq);
 771 }
 772
 773 STATIC int
 774 xfs_fs_setxquota(
 775         struct super_block      *sb,
 776         int                     type,
 777         qid_t                   id,
 778         struct fs_disk_quota    *fdq)
 779 {
 780         return -bhv_vfs_quotactl(vfs_from_sb(sb),
 781                                  (type == USRQUOTA) ? Q_XSETQLIM :
 782                                   ((type == GRPQUOTA) ? Q_XSETGQLIM :
 783                                    Q_XSETPQLIM), id, (caddr_t)fdq);
 784 }
 785
 786 STATIC int
 787 xfs_fs_fill_super(
 788         struct super_block      *sb,
 789         void                    *data,
 790         int                     silent)
 791 {
 792         vnode_t                 *rootvp;
 793         struct bhv_vfs          *vfsp = vfs_allocate(sb);
 794         struct xfs_mount_args   *args = xfs_args_allocate(sb, silent);
 795         struct kstatfs          statvfs;
 796         int                     error;
 797
 798         bhv_insert_all_vfsops(vfsp);
 799
 800         error = bhv_vfs_parseargs(vfsp, (char *)data, args, 0);
 801         if (error) {
 802                 bhv_remove_all_vfsops(vfsp, 1);
 803                 goto fail_vfsop;
 804         }
 805
 806         sb_min_blocksize(sb, BBSIZE);
 807 #ifdef CONFIG_XFS_EXPORT
 808         sb->s_export_op = &xfs_export_operations;
 809 #endif
 810         sb->s_qcop = &xfs_quotactl_operations;
 811         sb->s_op = &xfs_super_operations;
 812
 813         error = bhv_vfs_mount(vfsp, args, NULL);
 814         if (error) {
 815                 bhv_remove_all_vfsops(vfsp, 1);
 816                 goto fail_vfsop;
 817         }
 818
 819         error = bhv_vfs_statvfs(vfsp, &statvfs, NULL);
 820         if (error)
 821                 goto fail_unmount;
 822
 823         sb->s_dirt = 1;
 824         sb->s_magic = statvfs.f_type;
 825         sb->s_blocksize = statvfs.f_bsize;
 826         sb->s_blocksize_bits = ffs(statvfs.f_bsize) - 1;
 827         sb->s_maxbytes = xfs_max_file_offset(sb->s_blocksize_bits);
 828         sb->s_time_gran = 1;
 829         set_posix_acl_flag(sb);
 830
 831         error = bhv_vfs_root(vfsp, &rootvp);
 832         if (error)
 833                 goto fail_unmount;
 834
 835         sb->s_root = d_alloc_root(vn_to_inode(rootvp));
 836         if (!sb->s_root) {
 837                 error = ENOMEM;
 838                 goto fail_vnrele;
 839         }
 840         if (is_bad_inode(sb->s_root->d_inode)) {
 841                 error = EINVAL;
 842                 goto fail_vnrele;
 843         }
 844         if ((error = xfs_fs_start_syncd(vfsp)))
 845                 goto fail_vnrele;
 846         vn_trace_exit(rootvp, __FUNCTION__, (inst_t *)__return_address);
 847
 848         kmem_free(args, sizeof(*args));
 849         return 0;
 850
 851 fail_vnrele:
 852         if (sb->s_root) {
 853                 dput(sb->s_root);
 854                 sb->s_root = NULL;
 855         } else {
 856                 VN_RELE(rootvp);
 857         }
 858
 859 fail_unmount:
 860         bhv_vfs_unmount(vfsp, 0, NULL);
 861
 862 fail_vfsop:
 863         vfs_deallocate(vfsp);
 864         kmem_free(args, sizeof(*args));
 865         return -error;
 866 }
 867
 868 STATIC struct super_block *
 869 xfs_fs_get_sb(
 870         struct file_system_type *fs_type,
 871         int                     flags,
 872         const char              *dev_name,
 873         void                    *data)
 874 {
 875         return get_sb_bdev(fs_type, flags, dev_name, data, xfs_fs_fill_super);
 876 }
 877
 878 STATIC struct super_operations xfs_super_operations = {
 879         .alloc_inode            = xfs_fs_alloc_inode,
 880         .destroy_inode          = xfs_fs_destroy_inode,
 881         .write_inode            = xfs_fs_write_inode,
 882         .clear_inode            = xfs_fs_clear_inode,
 883         .put_super              = xfs_fs_put_super,
 884         .write_super            = xfs_fs_write_super,
 885         .sync_fs                = xfs_fs_sync_super,
 886         .write_super_lockfs     = xfs_fs_lockfs,
 887         .statfs                 = xfs_fs_statfs,
 888         .remount_fs             = xfs_fs_remount,
 889         .show_options           = xfs_fs_show_options,
 890 };
 891
 892 STATIC struct quotactl_ops xfs_quotactl_operations = {
 893         .quota_sync             = xfs_fs_quotasync,
 894         .get_xstate             = xfs_fs_getxstate,
 895         .set_xstate             = xfs_fs_setxstate,
 896         .get_xquota             = xfs_fs_getxquota,
 897         .set_xquota             = xfs_fs_setxquota,
 898 };
 899
 900 STATIC struct file_system_type xfs_fs_type = {
 901         .owner                  = THIS_MODULE,
 902         .name                   = "xfs",
 903         .get_sb                 = xfs_fs_get_sb,
 904         .kill_sb                = kill_block_super,
 905         .fs_flags               = FS_REQUIRES_DEV,
 906 };
 907
 908
 909 STATIC int __init
 910 init_xfs_fs( void )
 911 {
 912         int                     error;
 913         struct sysinfo          si;
 914         static char             message[] __initdata = KERN_INFO \
 915                 XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled\n";
 916
 917         printk(message);
 918
 919         si_meminfo(&si);
 920         xfs_physmem = si.totalram;
 921
 922         ktrace_init(64);
 923
 924         error = xfs_init_zones();
 925         if (error < 0)
 926                 goto undo_zones;
 927
 928         error = xfs_buf_init();
 929         if (error < 0)
 930                 goto undo_buffers;
 931
 932         vn_init();
 933         xfs_init();
 934         uuid_init();
 935         vfs_initquota();
 936
 937         error = register_filesystem(&xfs_fs_type);
 938         if (error)
 939                 goto undo_register;
 940         return 0;
 941
 942 undo_register:
 943         xfs_buf_terminate();
 944
 945 undo_buffers:
 946         xfs_destroy_zones();
 947
 948 undo_zones:
 949         return error;
 950 }
 951
 952 STATIC void __exit
 953 exit_xfs_fs( void )
 954 {
 955         vfs_exitquota();
 956         unregister_filesystem(&xfs_fs_type);
 957         xfs_cleanup();
 958         xfs_buf_terminate();
 959         xfs_destroy_zones();
 960         ktrace_uninit();
 961 }
 962
 963 module_init(init_xfs_fs);
 964 module_exit(exit_xfs_fs);
 965
 966 MODULE_AUTHOR("Silicon Graphics, Inc.");
 967 MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled");
 968 MODULE_LICENSE("GPL");