fs/xfs/linux-2.6/xfs_super.c

   1 /*
   2  * Copyright (c) 2000-2006 Silicon Graphics, Inc.
   3  * All Rights Reserved.
   4  *
   5  * This program is free software; you can redistribute it and/or
   6  * modify it under the terms of the GNU General Public License as
   7  * published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it would be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write the Free Software Foundation,
  16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17  */
  18 #include "xfs.h"
  19 #include "xfs_bit.h"
  20 #include "xfs_log.h"
  21 #include "xfs_clnt.h"
  22 #include "xfs_inum.h"
  23 #include "xfs_trans.h"
  24 #include "xfs_sb.h"
  25 #include "xfs_ag.h"
  26 #include "xfs_dir2.h"
  27 #include "xfs_alloc.h"
  28 #include "xfs_dmapi.h"
  29 #include "xfs_quota.h"
  30 #include "xfs_mount.h"
  31 #include "xfs_bmap_btree.h"
  32 #include "xfs_alloc_btree.h"
  33 #include "xfs_ialloc_btree.h"
  34 #include "xfs_dir2_sf.h"
  35 #include "xfs_attr_sf.h"
  36 #include "xfs_dinode.h"
  37 #include "xfs_inode.h"
  38 #include "xfs_btree.h"
  39 #include "xfs_ialloc.h"
  40 #include "xfs_bmap.h"
  41 #include "xfs_rtalloc.h"
  42 #include "xfs_error.h"
  43 #include "xfs_itable.h"
  44 #include "xfs_rw.h"
  45 #include "xfs_acl.h"
  46 #include "xfs_cap.h"
  47 #include "xfs_mac.h"
  48 #include "xfs_attr.h"
  49 #include "xfs_buf_item.h"
  50 #include "xfs_utils.h"
  51 #include "xfs_version.h"
  52
  53 #include <linux/namei.h>
  54 #include <linux/init.h>
  55 #include <linux/mount.h>
  56 #include <linux/mempool.h>
  57 #include <linux/writeback.h>
  58 #include <linux/kthread.h>
  59
  60 STATIC struct quotactl_ops xfs_quotactl_operations;
  61 STATIC struct super_operations xfs_super_operations;
  62 STATIC kmem_zone_t *xfs_vnode_zone;
  63 STATIC kmem_zone_t *xfs_ioend_zone;
  64 mempool_t *xfs_ioend_pool;
  65
  66 STATIC struct xfs_mount_args *
  67 xfs_args_allocate(
  68         struct super_block      *sb,
  69         int                     silent)
  70 {
  71         struct xfs_mount_args   *args;
  72
  73         args = kmem_zalloc(sizeof(struct xfs_mount_args), KM_SLEEP);
  74         args->logbufs = args->logbufsize = -1;
  75         strncpy(args->fsname, sb->s_id, MAXNAMELEN);
  76
  77         /* Copy the already-parsed mount(2) flags we're interested in */
  78         if (sb->s_flags & MS_DIRSYNC)
  79                 args->flags |= XFSMNT_DIRSYNC;
  80         if (sb->s_flags & MS_SYNCHRONOUS)
  81                 args->flags |= XFSMNT_WSYNC;
  82         if (silent)
  83                 args->flags |= XFSMNT_QUIET;
  84         args->flags |= XFSMNT_32BITINODES;
  85
  86         return args;
  87 }
  88
  89 __uint64_t
  90 xfs_max_file_offset(
  91         unsigned int            blockshift)
  92 {
  93         unsigned int            pagefactor = 1;
  94         unsigned int            bitshift = BITS_PER_LONG - 1;
  95
  96         /* Figure out maximum filesize, on Linux this can depend on
  97          * the filesystem blocksize (on 32 bit platforms).
  98          * __block_prepare_write does this in an [unsigned] long...
  99          *      page->index << (PAGE_CACHE_SHIFT - bbits)
 100          * So, for page sized blocks (4K on 32 bit platforms),
 101          * this wraps at around 8Tb (hence MAX_LFS_FILESIZE which is
 102          *      (((u64)PAGE_CACHE_SIZE << (BITS_PER_LONG-1))-1)
 103          * but for smaller blocksizes it is less (bbits = log2 bsize).
 104          * Note1: get_block_t takes a long (implicit cast from above)
 105          * Note2: The Large Block Device (LBD and HAVE_SECTOR_T) patch
 106          * can optionally convert the [unsigned] long from above into
 107          * an [unsigned] long long.
 108          */
 109
 110 #if BITS_PER_LONG == 32
 111 # if defined(CONFIG_LBD)
 112         ASSERT(sizeof(sector_t) == 8);
 113         pagefactor = PAGE_CACHE_SIZE;
 114         bitshift = BITS_PER_LONG;
 115 # else
 116         pagefactor = PAGE_CACHE_SIZE >> (PAGE_CACHE_SHIFT - blockshift);
 117 # endif
 118 #endif
 119
 120         return (((__uint64_t)pagefactor) << bitshift) - 1;
 121 }
 122
 123 STATIC __inline__ void
 124 xfs_set_inodeops(
 125         struct inode            *inode)
 126 {
 127         switch (inode->i_mode & S_IFMT) {
 128         case S_IFREG:
 129                 inode->i_op = &xfs_inode_operations;
 130                 inode->i_fop = &xfs_file_operations;
 131                 inode->i_mapping->a_ops = &xfs_address_space_operations;
 132                 break;
 133         case S_IFDIR:
 134                 inode->i_op = &xfs_dir_inode_operations;
 135                 inode->i_fop = &xfs_dir_file_operations;
 136                 break;
 137         case S_IFLNK:
 138                 inode->i_op = &xfs_symlink_inode_operations;
 139                 if (inode->i_blocks)
 140                         inode->i_mapping->a_ops = &xfs_address_space_operations;
 141                 break;
 142         default:
 143                 inode->i_op = &xfs_inode_operations;
 144                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
 145                 break;
 146         }
 147 }
 148
 149 STATIC __inline__ void
 150 xfs_revalidate_inode(
 151         xfs_mount_t             *mp,
 152         bhv_vnode_t             *vp,
 153         xfs_inode_t             *ip)
 154 {
 155         struct inode            *inode = vn_to_inode(vp);
 156
 157         inode->i_mode   = ip->i_d.di_mode;
 158         inode->i_nlink  = ip->i_d.di_nlink;
 159         inode->i_uid    = ip->i_d.di_uid;
 160         inode->i_gid    = ip->i_d.di_gid;
 161
 162         switch (inode->i_mode & S_IFMT) {
 163         case S_IFBLK:
 164         case S_IFCHR:
 165                 inode->i_rdev =
 166                         MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff,
 167                               sysv_minor(ip->i_df.if_u2.if_rdev));
 168                 break;
 169         default:
 170                 inode->i_rdev = 0;
 171                 break;
 172         }
 173
 174         inode->i_generation = ip->i_d.di_gen;
 175         i_size_write(inode, ip->i_d.di_size);
 176         inode->i_blocks =
 177                 XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
 178         inode->i_atime.tv_sec   = ip->i_d.di_atime.t_sec;
 179         inode->i_atime.tv_nsec  = ip->i_d.di_atime.t_nsec;
 180         inode->i_mtime.tv_sec   = ip->i_d.di_mtime.t_sec;
 181         inode->i_mtime.tv_nsec  = ip->i_d.di_mtime.t_nsec;
 182         inode->i_ctime.tv_sec   = ip->i_d.di_ctime.t_sec;
 183         inode->i_ctime.tv_nsec  = ip->i_d.di_ctime.t_nsec;
 184         if (ip->i_d.di_flags & XFS_DIFLAG_IMMUTABLE)
 185                 inode->i_flags |= S_IMMUTABLE;
 186         else
 187                 inode->i_flags &= ~S_IMMUTABLE;
 188         if (ip->i_d.di_flags & XFS_DIFLAG_APPEND)
 189                 inode->i_flags |= S_APPEND;
 190         else
 191                 inode->i_flags &= ~S_APPEND;
 192         if (ip->i_d.di_flags & XFS_DIFLAG_SYNC)
 193                 inode->i_flags |= S_SYNC;
 194         else
 195                 inode->i_flags &= ~S_SYNC;
 196         if (ip->i_d.di_flags & XFS_DIFLAG_NOATIME)
 197                 inode->i_flags |= S_NOATIME;
 198         else
 199                 inode->i_flags &= ~S_NOATIME;
 200         vp->v_flag &= ~VMODIFIED;
 201 }
 202
 203 void
 204 xfs_initialize_vnode(
 205         bhv_desc_t              *bdp,
 206         bhv_vnode_t             *vp,
 207         bhv_desc_t              *inode_bhv,
 208         int                     unlock)
 209 {
 210         xfs_inode_t             *ip = XFS_BHVTOI(inode_bhv);
 211         struct inode            *inode = vn_to_inode(vp);
 212
 213         if (!inode_bhv->bd_vobj) {
 214                 vp->v_vfsp = bhvtovfs(bdp);
 215                 bhv_desc_init(inode_bhv, ip, vp, &xfs_vnodeops);
 216                 bhv_insert(VN_BHV_HEAD(vp), inode_bhv);
 217         }
 218
 219         /*
 220          * We need to set the ops vectors, and unlock the inode, but if
 221          * we have been called during the new inode create process, it is
 222          * too early to fill in the Linux inode.  We will get called a
 223          * second time once the inode is properly set up, and then we can
 224          * finish our work.
 225          */
 226         if (ip->i_d.di_mode != 0 && unlock && (inode->i_state & I_NEW)) {
 227                 xfs_revalidate_inode(XFS_BHVTOM(bdp), vp, ip);
 228                 xfs_set_inodeops(inode);
 229
 230                 ip->i_flags &= ~XFS_INEW;
 231                 barrier();
 232
 233                 unlock_new_inode(inode);
 234         }
 235 }
 236
 237 int
 238 xfs_blkdev_get(
 239         xfs_mount_t             *mp,
 240         const char              *name,
 241         struct block_device     **bdevp)
 242 {
 243         int                     error = 0;
 244
 245         *bdevp = open_bdev_excl(name, 0, mp);
 246         if (IS_ERR(*bdevp)) {
 247                 error = PTR_ERR(*bdevp);
 248                 printk("XFS: Invalid device [%s], error=%d\n", name, error);
 249         }
 250
 251         return -error;
 252 }
 253
 254 void
 255 xfs_blkdev_put(
 256         struct block_device     *bdev)
 257 {
 258         if (bdev)
 259                 close_bdev_excl(bdev);
 260 }
 261
 262 /*
 263  * Try to write out the superblock using barriers.
 264  */
 265 STATIC int
 266 xfs_barrier_test(
 267         xfs_mount_t     *mp)
 268 {
 269         xfs_buf_t       *sbp = xfs_getsb(mp, 0);
 270         int             error;
 271
 272         XFS_BUF_UNDONE(sbp);
 273         XFS_BUF_UNREAD(sbp);
 274         XFS_BUF_UNDELAYWRITE(sbp);
 275         XFS_BUF_WRITE(sbp);
 276         XFS_BUF_UNASYNC(sbp);
 277         XFS_BUF_ORDERED(sbp);
 278
 279         xfsbdstrat(mp, sbp);
 280         error = xfs_iowait(sbp);
 281
 282         /*
 283          * Clear all the flags we set and possible error state in the
 284          * buffer.  We only did the write to try out whether barriers
 285          * worked and shouldn't leave any traces in the superblock
 286          * buffer.
 287          */
 288         XFS_BUF_DONE(sbp);
 289         XFS_BUF_ERROR(sbp, 0);
 290         XFS_BUF_UNORDERED(sbp);
 291
 292         xfs_buf_relse(sbp);
 293         return error;
 294 }
 295
 296 void
 297 xfs_mountfs_check_barriers(xfs_mount_t *mp)
 298 {
 299         int error;
 300
 301         if (mp->m_logdev_targp != mp->m_ddev_targp) {
 302                 xfs_fs_cmn_err(CE_NOTE, mp,
 303                   "Disabling barriers, not supported with external log device");
 304                 mp->m_flags &= ~XFS_MOUNT_BARRIER;
 305                 return;
 306         }
 307
 308         if (mp->m_ddev_targp->bt_bdev->bd_disk->queue->ordered ==
 309                                         QUEUE_ORDERED_NONE) {
 310                 xfs_fs_cmn_err(CE_NOTE, mp,
 311                   "Disabling barriers, not supported by the underlying device");
 312                 mp->m_flags &= ~XFS_MOUNT_BARRIER;
 313                 return;
 314         }
 315
 316         if (xfs_readonly_buftarg(mp->m_ddev_targp)) {
 317                 xfs_fs_cmn_err(CE_NOTE, mp,
 318                   "Disabling barriers, underlying device is readonly");
 319                 mp->m_flags &= ~XFS_MOUNT_BARRIER;
 320                 return;
 321         }
 322
 323         error = xfs_barrier_test(mp);
 324         if (error) {
 325                 xfs_fs_cmn_err(CE_NOTE, mp,
 326                   "Disabling barriers, trial barrier write failed");
 327                 mp->m_flags &= ~XFS_MOUNT_BARRIER;
 328                 return;
 329         }
 330 }
 331
 332 void
 333 xfs_blkdev_issue_flush(
 334         xfs_buftarg_t           *buftarg)
 335 {
 336         blkdev_issue_flush(buftarg->bt_bdev, NULL);
 337 }
 338
 339 STATIC struct inode *
 340 xfs_fs_alloc_inode(
 341         struct super_block      *sb)
 342 {
 343         bhv_vnode_t             *vp;
 344
 345         vp = kmem_zone_alloc(xfs_vnode_zone, KM_SLEEP);
 346         if (unlikely(!vp))
 347                 return NULL;
 348         return vn_to_inode(vp);
 349 }
 350
 351 STATIC void
 352 xfs_fs_destroy_inode(
 353         struct inode            *inode)
 354 {
 355         kmem_zone_free(xfs_vnode_zone, vn_from_inode(inode));
 356 }
 357
 358 STATIC void
 359 xfs_fs_inode_init_once(
 360         void                    *vnode,
 361         kmem_zone_t             *zonep,
 362         unsigned long           flags)
 363 {
 364         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
 365                       SLAB_CTOR_CONSTRUCTOR)
 366                 inode_init_once(vn_to_inode((bhv_vnode_t *)vnode));
 367 }
 368
 369 STATIC int
 370 xfs_init_zones(void)
 371 {
 372         xfs_vnode_zone = kmem_zone_init_flags(sizeof(bhv_vnode_t), "xfs_vnode",
 373                                         KM_ZONE_HWALIGN | KM_ZONE_RECLAIM |
 374                                         KM_ZONE_SPREAD,
 375                                         xfs_fs_inode_init_once);
 376         if (!xfs_vnode_zone)
 377                 goto out;
 378
 379         xfs_ioend_zone = kmem_zone_init(sizeof(xfs_ioend_t), "xfs_ioend");
 380         if (!xfs_ioend_zone)
 381                 goto out_destroy_vnode_zone;
 382
 383         xfs_ioend_pool = mempool_create_slab_pool(4 * MAX_BUF_PER_PAGE,
 384                                                   xfs_ioend_zone);
 385         if (!xfs_ioend_pool)
 386                 goto out_free_ioend_zone;
 387         return 0;
 388
 389  out_free_ioend_zone:
 390         kmem_zone_destroy(xfs_ioend_zone);
 391  out_destroy_vnode_zone:
 392         kmem_zone_destroy(xfs_vnode_zone);
 393  out:
 394         return -ENOMEM;
 395 }
 396
 397 STATIC void
 398 xfs_destroy_zones(void)
 399 {
 400         mempool_destroy(xfs_ioend_pool);
 401         kmem_zone_destroy(xfs_vnode_zone);
 402         kmem_zone_destroy(xfs_ioend_zone);
 403 }
 404
 405 /*
 406  * Attempt to flush the inode, this will actually fail
 407  * if the inode is pinned, but we dirty the inode again
 408  * at the point when it is unpinned after a log write,
 409  * since this is when the inode itself becomes flushable.
 410  */
 411 STATIC int
 412 xfs_fs_write_inode(
 413         struct inode            *inode,
 414         int                     sync)
 415 {
 416         bhv_vnode_t             *vp = vn_from_inode(inode);
 417         int                     error = 0, flags = FLUSH_INODE;
 418
 419         if (vp) {
 420                 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
 421                 if (sync)
 422                         flags |= FLUSH_SYNC;
 423                 error = bhv_vop_iflush(vp, flags);
 424                 if (error == EAGAIN)
 425                         error = sync? bhv_vop_iflush(vp, flags | FLUSH_LOG) : 0;
 426         }
 427         return -error;
 428 }
 429
 430 STATIC void
 431 xfs_fs_clear_inode(
 432         struct inode            *inode)
 433 {
 434         bhv_vnode_t             *vp = vn_from_inode(inode);
 435
 436         vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
 437
 438         XFS_STATS_INC(vn_rele);
 439         XFS_STATS_INC(vn_remove);
 440         XFS_STATS_INC(vn_reclaim);
 441         XFS_STATS_DEC(vn_active);
 442
 443         /*
 444          * This can happen because xfs_iget_core calls xfs_idestroy if we
 445          * find an inode with di_mode == 0 but without IGET_CREATE set.
 446          */
 447         if (VNHEAD(vp))
 448                 bhv_vop_inactive(vp, NULL);
 449
 450         VN_LOCK(vp);
 451         vp->v_flag &= ~VMODIFIED;
 452         VN_UNLOCK(vp, 0);
 453
 454         if (VNHEAD(vp))
 455                 if (bhv_vop_reclaim(vp))
 456                         panic("%s: cannot reclaim 0x%p\n", __FUNCTION__, vp);
 457
 458         ASSERT(VNHEAD(vp) == NULL);
 459
 460 #ifdef XFS_VNODE_TRACE
 461         ktrace_free(vp->v_trace);
 462 #endif
 463 }
 464
 465 /*
 466  * Enqueue a work item to be picked up by the vfs xfssyncd thread.
 467  * Doing this has two advantages:
 468  * - It saves on stack space, which is tight in certain situations
 469  * - It can be used (with care) as a mechanism to avoid deadlocks.
 470  * Flushing while allocating in a full filesystem requires both.
 471  */
 472 STATIC void
 473 xfs_syncd_queue_work(
 474         struct bhv_vfs  *vfs,
 475         void            *data,
 476         void            (*syncer)(bhv_vfs_t *, void *))
 477 {
 478         struct bhv_vfs_sync_work *work;
 479
 480         work = kmem_alloc(sizeof(struct bhv_vfs_sync_work), KM_SLEEP);
 481         INIT_LIST_HEAD(&work->w_list);
 482         work->w_syncer = syncer;
 483         work->w_data = data;
 484         work->w_vfs = vfs;
 485         spin_lock(&vfs->vfs_sync_lock);
 486         list_add_tail(&work->w_list, &vfs->vfs_sync_list);
 487         spin_unlock(&vfs->vfs_sync_lock);
 488         wake_up_process(vfs->vfs_sync_task);
 489 }
 490
 491 /*
 492  * Flush delayed allocate data, attempting to free up reserved space
 493  * from existing allocations.  At this point a new allocation attempt
 494  * has failed with ENOSPC and we are in the process of scratching our
 495  * heads, looking about for more room...
 496  */
 497 STATIC void
 498 xfs_flush_inode_work(
 499         bhv_vfs_t       *vfs,
 500         void            *inode)
 501 {
 502         filemap_flush(((struct inode *)inode)->i_mapping);
 503         iput((struct inode *)inode);
 504 }
 505
 506 void
 507 xfs_flush_inode(
 508         xfs_inode_t     *ip)
 509 {
 510         struct inode    *inode = vn_to_inode(XFS_ITOV(ip));
 511         struct bhv_vfs  *vfs = XFS_MTOVFS(ip->i_mount);
 512
 513         igrab(inode);
 514         xfs_syncd_queue_work(vfs, inode, xfs_flush_inode_work);
 515         delay(msecs_to_jiffies(500));
 516 }
 517
 518 /*
 519  * This is the "bigger hammer" version of xfs_flush_inode_work...
 520  * (IOW, "If at first you don't succeed, use a Bigger Hammer").
 521  */
 522 STATIC void
 523 xfs_flush_device_work(
 524         bhv_vfs_t       *vfs,
 525         void            *inode)
 526 {
 527         sync_blockdev(vfs->vfs_super->s_bdev);
 528         iput((struct inode *)inode);
 529 }
 530
 531 void
 532 xfs_flush_device(
 533         xfs_inode_t     *ip)
 534 {
 535         struct inode    *inode = vn_to_inode(XFS_ITOV(ip));
 536         struct bhv_vfs  *vfs = XFS_MTOVFS(ip->i_mount);
 537
 538         igrab(inode);
 539         xfs_syncd_queue_work(vfs, inode, xfs_flush_device_work);
 540         delay(msecs_to_jiffies(500));
 541         xfs_log_force(ip->i_mount, (xfs_lsn_t)0, XFS_LOG_FORCE|XFS_LOG_SYNC);
 542 }
 543
 544 STATIC void
 545 vfs_sync_worker(
 546         bhv_vfs_t       *vfsp,
 547         void            *unused)
 548 {
 549         int             error;
 550
 551         if (!(vfsp->vfs_flag & VFS_RDONLY))
 552                 error = bhv_vfs_sync(vfsp, SYNC_FSDATA | SYNC_BDFLUSH | \
 553                                         SYNC_ATTR | SYNC_REFCACHE, NULL);
 554         vfsp->vfs_sync_seq++;
 555         wmb();
 556         wake_up(&vfsp->vfs_wait_single_sync_task);
 557 }
 558
 559 STATIC int
 560 xfssyncd(
 561         void                    *arg)
 562 {
 563         long                    timeleft;
 564         bhv_vfs_t               *vfsp = (bhv_vfs_t *) arg;
 565         bhv_vfs_sync_work_t     *work, *n;
 566         LIST_HEAD               (tmp);
 567
 568         timeleft = xfs_syncd_centisecs * msecs_to_jiffies(10);
 569         for (;;) {
 570                 timeleft = schedule_timeout_interruptible(timeleft);
 571                 /* swsusp */
 572                 try_to_freeze();
 573                 if (kthread_should_stop() && list_empty(&vfsp->vfs_sync_list))
 574                         break;
 575
 576                 spin_lock(&vfsp->vfs_sync_lock);
 577                 /*
 578                  * We can get woken by laptop mode, to do a sync -
 579                  * that's the (only!) case where the list would be
 580                  * empty with time remaining.
 581                  */
 582                 if (!timeleft || list_empty(&vfsp->vfs_sync_list)) {
 583                         if (!timeleft)
 584                                 timeleft = xfs_syncd_centisecs *
 585                                                         msecs_to_jiffies(10);
 586                         INIT_LIST_HEAD(&vfsp->vfs_sync_work.w_list);
 587                         list_add_tail(&vfsp->vfs_sync_work.w_list,
 588                                         &vfsp->vfs_sync_list);
 589                 }
 590                 list_for_each_entry_safe(work, n, &vfsp->vfs_sync_list, w_list)
 591                         list_move(&work->w_list, &tmp);
 592                 spin_unlock(&vfsp->vfs_sync_lock);
 593
 594                 list_for_each_entry_safe(work, n, &tmp, w_list) {
 595                         (*work->w_syncer)(vfsp, work->w_data);
 596                         list_del(&work->w_list);
 597                         if (work == &vfsp->vfs_sync_work)
 598                                 continue;
 599                         kmem_free(work, sizeof(struct bhv_vfs_sync_work));
 600                 }
 601         }
 602
 603         return 0;
 604 }
 605
 606 STATIC int
 607 xfs_fs_start_syncd(
 608         bhv_vfs_t               *vfsp)
 609 {
 610         vfsp->vfs_sync_work.w_syncer = vfs_sync_worker;
 611         vfsp->vfs_sync_work.w_vfs = vfsp;
 612         vfsp->vfs_sync_task = kthread_run(xfssyncd, vfsp, "xfssyncd");
 613         if (IS_ERR(vfsp->vfs_sync_task))
 614                 return -PTR_ERR(vfsp->vfs_sync_task);
 615         return 0;
 616 }
 617
 618 STATIC void
 619 xfs_fs_stop_syncd(
 620         bhv_vfs_t               *vfsp)
 621 {
 622         kthread_stop(vfsp->vfs_sync_task);
 623 }
 624
 625 STATIC void
 626 xfs_fs_put_super(
 627         struct super_block      *sb)
 628 {
 629         bhv_vfs_t               *vfsp = vfs_from_sb(sb);
 630         int                     error;
 631
 632         xfs_fs_stop_syncd(vfsp);
 633         bhv_vfs_sync(vfsp, SYNC_ATTR | SYNC_DELWRI, NULL);
 634         error = bhv_vfs_unmount(vfsp, 0, NULL);
 635         if (error) {
 636                 printk("XFS: unmount got error=%d\n", error);
 637                 printk("%s: vfs=0x%p left dangling!\n", __FUNCTION__, vfsp);
 638         } else {
 639                 vfs_deallocate(vfsp);
 640         }
 641 }
 642
 643 STATIC void
 644 xfs_fs_write_super(
 645         struct super_block      *sb)
 646 {
 647         if (!(sb->s_flags & MS_RDONLY))
 648                 bhv_vfs_sync(vfs_from_sb(sb), SYNC_FSDATA, NULL);
 649         sb->s_dirt = 0;
 650 }
 651
 652 STATIC int
 653 xfs_fs_sync_super(
 654         struct super_block      *sb,
 655         int                     wait)
 656 {
 657         bhv_vfs_t               *vfsp = vfs_from_sb(sb);
 658         int                     error;
 659         int                     flags;
 660
 661         if (unlikely(sb->s_frozen == SB_FREEZE_WRITE))
 662                 flags = SYNC_QUIESCE;
 663         else
 664                 flags = SYNC_FSDATA | (wait ? SYNC_WAIT : 0);
 665
 666         error = bhv_vfs_sync(vfsp, flags, NULL);
 667         sb->s_dirt = 0;
 668
 669         if (unlikely(laptop_mode)) {
 670                 int     prev_sync_seq = vfsp->vfs_sync_seq;
 671
 672                 /*
 673                  * The disk must be active because we're syncing.
 674                  * We schedule xfssyncd now (now that the disk is
 675                  * active) instead of later (when it might not be).
 676                  */
 677                 wake_up_process(vfsp->vfs_sync_task);
 678                 /*
 679                  * We have to wait for the sync iteration to complete.
 680                  * If we don't, the disk activity caused by the sync
 681                  * will come after the sync is completed, and that
 682                  * triggers another sync from laptop mode.
 683                  */
 684                 wait_event(vfsp->vfs_wait_single_sync_task,
 685                                 vfsp->vfs_sync_seq != prev_sync_seq);
 686         }
 687
 688         return -error;
 689 }
 690
 691 STATIC int
 692 xfs_fs_statfs(
 693         struct dentry           *dentry,
 694         struct kstatfs          *statp)
 695 {
 696         return -bhv_vfs_statvfs(vfs_from_sb(dentry->d_sb), statp,
 697                                 vn_from_inode(dentry->d_inode));
 698 }
 699
 700 STATIC int
 701 xfs_fs_remount(
 702         struct super_block      *sb,
 703         int                     *flags,
 704         char                    *options)
 705 {
 706         bhv_vfs_t               *vfsp = vfs_from_sb(sb);
 707         struct xfs_mount_args   *args = xfs_args_allocate(sb, 0);
 708         int                     error;
 709
 710         error = bhv_vfs_parseargs(vfsp, options, args, 1);
 711         if (!error)
 712                 error = bhv_vfs_mntupdate(vfsp, flags, args);
 713         kmem_free(args, sizeof(*args));
 714         return -error;
 715 }
 716
 717 STATIC void
 718 xfs_fs_lockfs(
 719         struct super_block      *sb)
 720 {
 721         bhv_vfs_freeze(vfs_from_sb(sb));
 722 }
 723
 724 STATIC int
 725 xfs_fs_show_options(
 726         struct seq_file         *m,
 727         struct vfsmount         *mnt)
 728 {
 729         return -bhv_vfs_showargs(vfs_from_sb(mnt->mnt_sb), m);
 730 }
 731
 732 STATIC int
 733 xfs_fs_quotasync(
 734         struct super_block      *sb,
 735         int                     type)
 736 {
 737         return -bhv_vfs_quotactl(vfs_from_sb(sb), Q_XQUOTASYNC, 0, NULL);
 738 }
 739
 740 STATIC int
 741 xfs_fs_getxstate(
 742         struct super_block      *sb,
 743         struct fs_quota_stat    *fqs)
 744 {
 745         return -bhv_vfs_quotactl(vfs_from_sb(sb), Q_XGETQSTAT, 0, (caddr_t)fqs);
 746 }
 747
 748 STATIC int
 749 xfs_fs_setxstate(
 750         struct super_block      *sb,
 751         unsigned int            flags,
 752         int                     op)
 753 {
 754         return -bhv_vfs_quotactl(vfs_from_sb(sb), op, 0, (caddr_t)&flags);
 755 }
 756
 757 STATIC int
 758 xfs_fs_getxquota(
 759         struct super_block      *sb,
 760         int                     type,
 761         qid_t                   id,
 762         struct fs_disk_quota    *fdq)
 763 {
 764         return -bhv_vfs_quotactl(vfs_from_sb(sb),
 765                                  (type == USRQUOTA) ? Q_XGETQUOTA :
 766                                   ((type == GRPQUOTA) ? Q_XGETGQUOTA :
 767                                    Q_XGETPQUOTA), id, (caddr_t)fdq);
 768 }
 769
 770 STATIC int
 771 xfs_fs_setxquota(
 772         struct super_block      *sb,
 773         int                     type,
 774         qid_t                   id,
 775         struct fs_disk_quota    *fdq)
 776 {
 777         return -bhv_vfs_quotactl(vfs_from_sb(sb),
 778                                  (type == USRQUOTA) ? Q_XSETQLIM :
 779                                   ((type == GRPQUOTA) ? Q_XSETGQLIM :
 780                                    Q_XSETPQLIM), id, (caddr_t)fdq);
 781 }
 782
 783 STATIC int
 784 xfs_fs_fill_super(
 785         struct super_block      *sb,
 786         void                    *data,
 787         int                     silent)
 788 {
 789         struct bhv_vnode        *rootvp;
 790         struct bhv_vfs          *vfsp = vfs_allocate(sb);
 791         struct xfs_mount_args   *args = xfs_args_allocate(sb, silent);
 792         struct kstatfs          statvfs;
 793         int                     error;
 794
 795         bhv_insert_all_vfsops(vfsp);
 796
 797         error = bhv_vfs_parseargs(vfsp, (char *)data, args, 0);
 798         if (error) {
 799                 bhv_remove_all_vfsops(vfsp, 1);
 800                 goto fail_vfsop;
 801         }
 802
 803         sb_min_blocksize(sb, BBSIZE);
 804         sb->s_export_op = &xfs_export_operations;
 805         sb->s_qcop = &xfs_quotactl_operations;
 806         sb->s_op = &xfs_super_operations;
 807
 808         error = bhv_vfs_mount(vfsp, args, NULL);
 809         if (error) {
 810                 bhv_remove_all_vfsops(vfsp, 1);
 811                 goto fail_vfsop;
 812         }
 813
 814         error = bhv_vfs_statvfs(vfsp, &statvfs, NULL);
 815         if (error)
 816                 goto fail_unmount;
 817
 818         sb->s_dirt = 1;
 819         sb->s_magic = statvfs.f_type;
 820         sb->s_blocksize = statvfs.f_bsize;
 821         sb->s_blocksize_bits = ffs(statvfs.f_bsize) - 1;
 822         sb->s_maxbytes = xfs_max_file_offset(sb->s_blocksize_bits);
 823         sb->s_time_gran = 1;
 824         set_posix_acl_flag(sb);
 825
 826         error = bhv_vfs_root(vfsp, &rootvp);
 827         if (error)
 828                 goto fail_unmount;
 829
 830         sb->s_root = d_alloc_root(vn_to_inode(rootvp));
 831         if (!sb->s_root) {
 832                 error = ENOMEM;
 833                 goto fail_vnrele;
 834         }
 835         if (is_bad_inode(sb->s_root->d_inode)) {
 836                 error = EINVAL;
 837                 goto fail_vnrele;
 838         }
 839         if ((error = xfs_fs_start_syncd(vfsp)))
 840                 goto fail_vnrele;
 841         vn_trace_exit(rootvp, __FUNCTION__, (inst_t *)__return_address);
 842
 843         kmem_free(args, sizeof(*args));
 844         return 0;
 845
 846 fail_vnrele:
 847         if (sb->s_root) {
 848                 dput(sb->s_root);
 849                 sb->s_root = NULL;
 850         } else {
 851                 VN_RELE(rootvp);
 852         }
 853
 854 fail_unmount:
 855         bhv_vfs_unmount(vfsp, 0, NULL);
 856
 857 fail_vfsop:
 858         vfs_deallocate(vfsp);
 859         kmem_free(args, sizeof(*args));
 860         return -error;
 861 }
 862
 863 STATIC int
 864 xfs_fs_get_sb(
 865         struct file_system_type *fs_type,
 866         int                     flags,
 867         const char              *dev_name,
 868         void                    *data,
 869         struct vfsmount         *mnt)
 870 {
 871         return get_sb_bdev(fs_type, flags, dev_name, data, xfs_fs_fill_super,
 872                            mnt);
 873 }
 874
 875 STATIC struct super_operations xfs_super_operations = {
 876         .alloc_inode            = xfs_fs_alloc_inode,
 877         .destroy_inode          = xfs_fs_destroy_inode,
 878         .write_inode            = xfs_fs_write_inode,
 879         .clear_inode            = xfs_fs_clear_inode,
 880         .put_super              = xfs_fs_put_super,
 881         .write_super            = xfs_fs_write_super,
 882         .sync_fs                = xfs_fs_sync_super,
 883         .write_super_lockfs     = xfs_fs_lockfs,
 884         .statfs                 = xfs_fs_statfs,
 885         .remount_fs             = xfs_fs_remount,
 886         .show_options           = xfs_fs_show_options,
 887 };
 888
 889 STATIC struct quotactl_ops xfs_quotactl_operations = {
 890         .quota_sync             = xfs_fs_quotasync,
 891         .get_xstate             = xfs_fs_getxstate,
 892         .set_xstate             = xfs_fs_setxstate,
 893         .get_xquota             = xfs_fs_getxquota,
 894         .set_xquota             = xfs_fs_setxquota,
 895 };
 896
 897 STATIC struct file_system_type xfs_fs_type = {
 898         .owner                  = THIS_MODULE,
 899         .name                   = "xfs",
 900         .get_sb                 = xfs_fs_get_sb,
 901         .kill_sb                = kill_block_super,
 902         .fs_flags               = FS_REQUIRES_DEV,
 903 };
 904
 905
 906 STATIC int __init
 907 init_xfs_fs( void )
 908 {
 909         int                     error;
 910         struct sysinfo          si;
 911         static char             message[] __initdata = KERN_INFO \
 912                 XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled\n";
 913
 914         printk(message);
 915
 916         si_meminfo(&si);
 917         xfs_physmem = si.totalram;
 918
 919         ktrace_init(64);
 920
 921         error = xfs_init_zones();
 922         if (error < 0)
 923                 goto undo_zones;
 924
 925         error = xfs_buf_init();
 926         if (error < 0)
 927                 goto undo_buffers;
 928
 929         vn_init();
 930         xfs_init();
 931         uuid_init();
 932         vfs_initquota();
 933
 934         error = register_filesystem(&xfs_fs_type);
 935         if (error)
 936                 goto undo_register;
 937         return 0;
 938
 939 undo_register:
 940         xfs_buf_terminate();
 941
 942 undo_buffers:
 943         xfs_destroy_zones();
 944
 945 undo_zones:
 946         return error;
 947 }
 948
 949 STATIC void __exit
 950 exit_xfs_fs( void )
 951 {
 952         vfs_exitquota();
 953         unregister_filesystem(&xfs_fs_type);
 954         xfs_cleanup();
 955         xfs_buf_terminate();
 956         xfs_destroy_zones();
 957         ktrace_uninit();
 958 }
 959
 960 module_init(init_xfs_fs);
 961 module_exit(exit_xfs_fs);
 962
 963 MODULE_AUTHOR("Silicon Graphics, Inc.");
 964 MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled");
 965 MODULE_LICENSE("GPL");