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 #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         inode->i_tag    = ip->i_d.di_tag;
 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_generation = ip->i_d.di_gen;
 177         i_size_write(inode, ip->i_d.di_size);
 178         inode->i_blocks =
 179                 XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
 180         inode->i_atime.tv_sec   = ip->i_d.di_atime.t_sec;
 181         inode->i_atime.tv_nsec  = ip->i_d.di_atime.t_nsec;
 182         inode->i_mtime.tv_sec   = ip->i_d.di_mtime.t_sec;
 183         inode->i_mtime.tv_nsec  = ip->i_d.di_mtime.t_nsec;
 184         inode->i_ctime.tv_sec   = ip->i_d.di_ctime.t_sec;
 185         inode->i_ctime.tv_nsec  = ip->i_d.di_ctime.t_nsec;
 186         if (ip->i_d.di_flags & XFS_DIFLAG_IMMUTABLE)
 187                 inode->i_flags |= S_IMMUTABLE;
 188         else
 189                 inode->i_flags &= ~S_IMMUTABLE;
 190         if (ip->i_d.di_flags & XFS_DIFLAG_IUNLINK)
 191                 inode->i_flags |= S_IUNLINK;
 192         else
 193                 inode->i_flags &= ~S_IUNLINK;
 194         if (ip->i_d.di_flags & XFS_DIFLAG_BARRIER)
 195                 inode->i_flags |= S_BARRIER;
 196         else
 197                 inode->i_flags &= ~S_BARRIER;
 198         if (ip->i_d.di_flags & XFS_DIFLAG_APPEND)
 199                 inode->i_flags |= S_APPEND;
 200         else
 201                 inode->i_flags &= ~S_APPEND;
 202         if (ip->i_d.di_flags & XFS_DIFLAG_SYNC)
 203                 inode->i_flags |= S_SYNC;
 204         else
 205                 inode->i_flags &= ~S_SYNC;
 206         if (ip->i_d.di_flags & XFS_DIFLAG_NOATIME)
 207                 inode->i_flags |= S_NOATIME;
 208         else
 209                 inode->i_flags &= ~S_NOATIME;
 210         vp->v_flag &= ~VMODIFIED;
 211 }
 212
 213 void
 214 xfs_initialize_vnode(
 215         bhv_desc_t              *bdp,
 216         bhv_vnode_t             *vp,
 217         bhv_desc_t              *inode_bhv,
 218         int                     unlock)
 219 {
 220         xfs_inode_t             *ip = XFS_BHVTOI(inode_bhv);
 221         struct inode            *inode = vn_to_inode(vp);
 222
 223         if (!inode_bhv->bd_vobj) {
 224                 vp->v_vfsp = bhvtovfs(bdp);
 225                 bhv_desc_init(inode_bhv, ip, vp, &xfs_vnodeops);
 226                 bhv_insert(VN_BHV_HEAD(vp), inode_bhv);
 227         }
 228
 229         /*
 230          * We need to set the ops vectors, and unlock the inode, but if
 231          * we have been called during the new inode create process, it is
 232          * too early to fill in the Linux inode.  We will get called a
 233          * second time once the inode is properly set up, and then we can
 234          * finish our work.
 235          */
 236         if (ip->i_d.di_mode != 0 && unlock && (inode->i_state & I_NEW)) {
 237                 xfs_revalidate_inode(XFS_BHVTOM(bdp), vp, ip);
 238                 xfs_set_inodeops(inode);
 239
 240                 xfs_iflags_clear(ip, XFS_INEW);
 241                 barrier();
 242
 243                 unlock_new_inode(inode);
 244         }
 245 }
 246
 247 int
 248 xfs_blkdev_get(
 249         xfs_mount_t             *mp,
 250         const char              *name,
 251         struct block_device     **bdevp)
 252 {
 253         int                     error = 0;
 254
 255         *bdevp = open_bdev_excl(name, 0, mp);
 256         if (IS_ERR(*bdevp)) {
 257                 error = PTR_ERR(*bdevp);
 258                 printk("XFS: Invalid device [%s], error=%d\n", name, error);
 259         }
 260
 261         return -error;
 262 }
 263
 264 void
 265 xfs_blkdev_put(
 266         struct block_device     *bdev)
 267 {
 268         if (bdev)
 269                 close_bdev_excl(bdev);
 270 }
 271
 272 /*
 273  * Try to write out the superblock using barriers.
 274  */
 275 STATIC int
 276 xfs_barrier_test(
 277         xfs_mount_t     *mp)
 278 {
 279         xfs_buf_t       *sbp = xfs_getsb(mp, 0);
 280         int             error;
 281
 282         XFS_BUF_UNDONE(sbp);
 283         XFS_BUF_UNREAD(sbp);
 284         XFS_BUF_UNDELAYWRITE(sbp);
 285         XFS_BUF_WRITE(sbp);
 286         XFS_BUF_UNASYNC(sbp);
 287         XFS_BUF_ORDERED(sbp);
 288
 289         xfsbdstrat(mp, sbp);
 290         error = xfs_iowait(sbp);
 291
 292         /*
 293          * Clear all the flags we set and possible error state in the
 294          * buffer.  We only did the write to try out whether barriers
 295          * worked and shouldn't leave any traces in the superblock
 296          * buffer.
 297          */
 298         XFS_BUF_DONE(sbp);
 299         XFS_BUF_ERROR(sbp, 0);
 300         XFS_BUF_UNORDERED(sbp);
 301
 302         xfs_buf_relse(sbp);
 303         return error;
 304 }
 305
 306 void
 307 xfs_mountfs_check_barriers(xfs_mount_t *mp)
 308 {
 309         int error;
 310
 311         if (mp->m_logdev_targp != mp->m_ddev_targp) {
 312                 xfs_fs_cmn_err(CE_NOTE, mp,
 313                   "Disabling barriers, not supported with external log device");
 314                 mp->m_flags &= ~XFS_MOUNT_BARRIER;
 315                 return;
 316         }
 317
 318         if (mp->m_ddev_targp->bt_bdev->bd_disk->queue->ordered ==
 319                                         QUEUE_ORDERED_NONE) {
 320                 xfs_fs_cmn_err(CE_NOTE, mp,
 321                   "Disabling barriers, not supported by the underlying device");
 322                 mp->m_flags &= ~XFS_MOUNT_BARRIER;
 323                 return;
 324         }
 325
 326         if (xfs_readonly_buftarg(mp->m_ddev_targp)) {
 327                 xfs_fs_cmn_err(CE_NOTE, mp,
 328                   "Disabling barriers, underlying device is readonly");
 329                 mp->m_flags &= ~XFS_MOUNT_BARRIER;
 330                 return;
 331         }
 332
 333         error = xfs_barrier_test(mp);
 334         if (error) {
 335                 xfs_fs_cmn_err(CE_NOTE, mp,
 336                   "Disabling barriers, trial barrier write failed");
 337                 mp->m_flags &= ~XFS_MOUNT_BARRIER;
 338                 return;
 339         }
 340 }
 341
 342 void
 343 xfs_blkdev_issue_flush(
 344         xfs_buftarg_t           *buftarg)
 345 {
 346         blkdev_issue_flush(buftarg->bt_bdev, NULL);
 347 }
 348
 349 STATIC struct inode *
 350 xfs_fs_alloc_inode(
 351         struct super_block      *sb)
 352 {
 353         bhv_vnode_t             *vp;
 354
 355         vp = kmem_zone_alloc(xfs_vnode_zone, KM_SLEEP);
 356         if (unlikely(!vp))
 357                 return NULL;
 358         return vn_to_inode(vp);
 359 }
 360
 361 STATIC void
 362 xfs_fs_destroy_inode(
 363         struct inode            *inode)
 364 {
 365         kmem_zone_free(xfs_vnode_zone, vn_from_inode(inode));
 366 }
 367
 368 STATIC void
 369 xfs_fs_inode_init_once(
 370         void                    *vnode,
 371         kmem_zone_t             *zonep,
 372         unsigned long           flags)
 373 {
 374         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
 375                       SLAB_CTOR_CONSTRUCTOR)
 376                 inode_init_once(vn_to_inode((bhv_vnode_t *)vnode));
 377 }
 378
 379 STATIC int
 380 xfs_init_zones(void)
 381 {
 382         xfs_vnode_zone = kmem_zone_init_flags(sizeof(bhv_vnode_t), "xfs_vnode",
 383                                         KM_ZONE_HWALIGN | KM_ZONE_RECLAIM |
 384                                         KM_ZONE_SPREAD,
 385                                         xfs_fs_inode_init_once);
 386         if (!xfs_vnode_zone)
 387                 goto out;
 388
 389         xfs_ioend_zone = kmem_zone_init(sizeof(xfs_ioend_t), "xfs_ioend");
 390         if (!xfs_ioend_zone)
 391                 goto out_destroy_vnode_zone;
 392
 393         xfs_ioend_pool = mempool_create_slab_pool(4 * MAX_BUF_PER_PAGE,
 394                                                   xfs_ioend_zone);
 395         if (!xfs_ioend_pool)
 396                 goto out_free_ioend_zone;
 397         return 0;
 398
 399  out_free_ioend_zone:
 400         kmem_zone_destroy(xfs_ioend_zone);
 401  out_destroy_vnode_zone:
 402         kmem_zone_destroy(xfs_vnode_zone);
 403  out:
 404         return -ENOMEM;
 405 }
 406
 407 STATIC void
 408 xfs_destroy_zones(void)
 409 {
 410         mempool_destroy(xfs_ioend_pool);
 411         kmem_zone_destroy(xfs_vnode_zone);
 412         kmem_zone_destroy(xfs_ioend_zone);
 413 }
 414
 415 /*
 416  * Attempt to flush the inode, this will actually fail
 417  * if the inode is pinned, but we dirty the inode again
 418  * at the point when it is unpinned after a log write,
 419  * since this is when the inode itself becomes flushable.
 420  */
 421 STATIC int
 422 xfs_fs_write_inode(
 423         struct inode            *inode,
 424         int                     sync)
 425 {
 426         bhv_vnode_t             *vp = vn_from_inode(inode);
 427         int                     error = 0, flags = FLUSH_INODE;
 428
 429         if (vp) {
 430                 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
 431                 if (sync)
 432                         flags |= FLUSH_SYNC;
 433                 error = bhv_vop_iflush(vp, flags);
 434                 if (error == EAGAIN)
 435                         error = sync? bhv_vop_iflush(vp, flags | FLUSH_LOG) : 0;
 436         }
 437         return -error;
 438 }
 439
 440 STATIC void
 441 xfs_fs_clear_inode(
 442         struct inode            *inode)
 443 {
 444         bhv_vnode_t             *vp = vn_from_inode(inode);
 445
 446         vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
 447
 448         XFS_STATS_INC(vn_rele);
 449         XFS_STATS_INC(vn_remove);
 450         XFS_STATS_INC(vn_reclaim);
 451         XFS_STATS_DEC(vn_active);
 452
 453         /*
 454          * This can happen because xfs_iget_core calls xfs_idestroy if we
 455          * find an inode with di_mode == 0 but without IGET_CREATE set.
 456          */
 457         if (VNHEAD(vp))
 458                 bhv_vop_inactive(vp, NULL);
 459
 460         VN_LOCK(vp);
 461         vp->v_flag &= ~VMODIFIED;
 462         VN_UNLOCK(vp, 0);
 463
 464         if (VNHEAD(vp))
 465                 if (bhv_vop_reclaim(vp))
 466                         panic("%s: cannot reclaim 0x%p\n", __FUNCTION__, vp);
 467
 468         ASSERT(VNHEAD(vp) == NULL);
 469
 470 #ifdef XFS_VNODE_TRACE
 471         ktrace_free(vp->v_trace);
 472 #endif
 473 }
 474
 475 /*
 476  * Enqueue a work item to be picked up by the vfs xfssyncd thread.
 477  * Doing this has two advantages:
 478  * - It saves on stack space, which is tight in certain situations
 479  * - It can be used (with care) as a mechanism to avoid deadlocks.
 480  * Flushing while allocating in a full filesystem requires both.
 481  */
 482 STATIC void
 483 xfs_syncd_queue_work(
 484         struct bhv_vfs  *vfs,
 485         void            *data,
 486         void            (*syncer)(bhv_vfs_t *, void *))
 487 {
 488         struct bhv_vfs_sync_work *work;
 489
 490         work = kmem_alloc(sizeof(struct bhv_vfs_sync_work), KM_SLEEP);
 491         INIT_LIST_HEAD(&work->w_list);
 492         work->w_syncer = syncer;
 493         work->w_data = data;
 494         work->w_vfs = vfs;
 495         spin_lock(&vfs->vfs_sync_lock);
 496         list_add_tail(&work->w_list, &vfs->vfs_sync_list);
 497         spin_unlock(&vfs->vfs_sync_lock);
 498         wake_up_process(vfs->vfs_sync_task);
 499 }
 500
 501 /*
 502  * Flush delayed allocate data, attempting to free up reserved space
 503  * from existing allocations.  At this point a new allocation attempt
 504  * has failed with ENOSPC and we are in the process of scratching our
 505  * heads, looking about for more room...
 506  */
 507 STATIC void
 508 xfs_flush_inode_work(
 509         bhv_vfs_t       *vfs,
 510         void            *inode)
 511 {
 512         filemap_flush(((struct inode *)inode)->i_mapping);
 513         iput((struct inode *)inode);
 514 }
 515
 516 void
 517 xfs_flush_inode(
 518         xfs_inode_t     *ip)
 519 {
 520         struct inode    *inode = vn_to_inode(XFS_ITOV(ip));
 521         struct bhv_vfs  *vfs = XFS_MTOVFS(ip->i_mount);
 522
 523         igrab(inode);
 524         xfs_syncd_queue_work(vfs, inode, xfs_flush_inode_work);
 525         delay(msecs_to_jiffies(500));
 526 }
 527
 528 /*
 529  * This is the "bigger hammer" version of xfs_flush_inode_work...
 530  * (IOW, "If at first you don't succeed, use a Bigger Hammer").
 531  */
 532 STATIC void
 533 xfs_flush_device_work(
 534         bhv_vfs_t       *vfs,
 535         void            *inode)
 536 {
 537         sync_blockdev(vfs->vfs_super->s_bdev);
 538         iput((struct inode *)inode);
 539 }
 540
 541 void
 542 xfs_flush_device(
 543         xfs_inode_t     *ip)
 544 {
 545         struct inode    *inode = vn_to_inode(XFS_ITOV(ip));
 546         struct bhv_vfs  *vfs = XFS_MTOVFS(ip->i_mount);
 547
 548         igrab(inode);
 549         xfs_syncd_queue_work(vfs, inode, xfs_flush_device_work);
 550         delay(msecs_to_jiffies(500));
 551         xfs_log_force(ip->i_mount, (xfs_lsn_t)0, XFS_LOG_FORCE|XFS_LOG_SYNC);
 552 }
 553
 554 STATIC void
 555 vfs_sync_worker(
 556         bhv_vfs_t       *vfsp,
 557         void            *unused)
 558 {
 559         int             error;
 560
 561         if (!(vfsp->vfs_flag & VFS_RDONLY))
 562                 error = bhv_vfs_sync(vfsp, SYNC_FSDATA | SYNC_BDFLUSH | \
 563                                         SYNC_ATTR | SYNC_REFCACHE, NULL);
 564         vfsp->vfs_sync_seq++;
 565         wmb();
 566         wake_up(&vfsp->vfs_wait_single_sync_task);
 567 }
 568
 569 STATIC int
 570 xfssyncd(
 571         void                    *arg)
 572 {
 573         long                    timeleft;
 574         bhv_vfs_t               *vfsp = (bhv_vfs_t *) arg;
 575         bhv_vfs_sync_work_t     *work, *n;
 576         LIST_HEAD               (tmp);
 577
 578         timeleft = xfs_syncd_centisecs * msecs_to_jiffies(10);
 579         for (;;) {
 580                 timeleft = schedule_timeout_interruptible(timeleft);
 581                 /* swsusp */
 582                 try_to_freeze();
 583                 if (kthread_should_stop() && list_empty(&vfsp->vfs_sync_list))
 584                         break;
 585
 586                 spin_lock(&vfsp->vfs_sync_lock);
 587                 /*
 588                  * We can get woken by laptop mode, to do a sync -
 589                  * that's the (only!) case where the list would be
 590                  * empty with time remaining.
 591                  */
 592                 if (!timeleft || list_empty(&vfsp->vfs_sync_list)) {
 593                         if (!timeleft)
 594                                 timeleft = xfs_syncd_centisecs *
 595                                                         msecs_to_jiffies(10);
 596                         INIT_LIST_HEAD(&vfsp->vfs_sync_work.w_list);
 597                         list_add_tail(&vfsp->vfs_sync_work.w_list,
 598                                         &vfsp->vfs_sync_list);
 599                 }
 600                 list_for_each_entry_safe(work, n, &vfsp->vfs_sync_list, w_list)
 601                         list_move(&work->w_list, &tmp);
 602                 spin_unlock(&vfsp->vfs_sync_lock);
 603
 604                 list_for_each_entry_safe(work, n, &tmp, w_list) {
 605                         (*work->w_syncer)(vfsp, work->w_data);
 606                         list_del(&work->w_list);
 607                         if (work == &vfsp->vfs_sync_work)
 608                                 continue;
 609                         kmem_free(work, sizeof(struct bhv_vfs_sync_work));
 610                 }
 611         }
 612
 613         return 0;
 614 }
 615
 616 STATIC int
 617 xfs_fs_start_syncd(
 618         bhv_vfs_t               *vfsp)
 619 {
 620         vfsp->vfs_sync_work.w_syncer = vfs_sync_worker;
 621         vfsp->vfs_sync_work.w_vfs = vfsp;
 622         vfsp->vfs_sync_task = kthread_run(xfssyncd, vfsp, "xfssyncd");
 623         if (IS_ERR(vfsp->vfs_sync_task))
 624                 return -PTR_ERR(vfsp->vfs_sync_task);
 625         return 0;
 626 }
 627
 628 STATIC void
 629 xfs_fs_stop_syncd(
 630         bhv_vfs_t               *vfsp)
 631 {
 632         kthread_stop(vfsp->vfs_sync_task);
 633 }
 634
 635 STATIC void
 636 xfs_fs_put_super(
 637         struct super_block      *sb)
 638 {
 639         bhv_vfs_t               *vfsp = vfs_from_sb(sb);
 640         int                     error;
 641
 642         xfs_fs_stop_syncd(vfsp);
 643         bhv_vfs_sync(vfsp, SYNC_ATTR | SYNC_DELWRI, NULL);
 644         error = bhv_vfs_unmount(vfsp, 0, NULL);
 645         if (error) {
 646                 printk("XFS: unmount got error=%d\n", error);
 647                 printk("%s: vfs=0x%p left dangling!\n", __FUNCTION__, vfsp);
 648         } else {
 649                 vfs_deallocate(vfsp);
 650         }
 651 }
 652
 653 STATIC void
 654 xfs_fs_write_super(
 655         struct super_block      *sb)
 656 {
 657         if (!(sb->s_flags & MS_RDONLY))
 658                 bhv_vfs_sync(vfs_from_sb(sb), SYNC_FSDATA, NULL);
 659         sb->s_dirt = 0;
 660 }
 661
 662 STATIC int
 663 xfs_fs_sync_super(
 664         struct super_block      *sb,
 665         int                     wait)
 666 {
 667         bhv_vfs_t               *vfsp = vfs_from_sb(sb);
 668         int                     error;
 669         int                     flags;
 670
 671         if (unlikely(sb->s_frozen == SB_FREEZE_WRITE))
 672                 flags = SYNC_QUIESCE;
 673         else
 674                 flags = SYNC_FSDATA | (wait ? SYNC_WAIT : 0);
 675
 676         error = bhv_vfs_sync(vfsp, flags, NULL);
 677         sb->s_dirt = 0;
 678
 679         if (unlikely(laptop_mode)) {
 680                 int     prev_sync_seq = vfsp->vfs_sync_seq;
 681
 682                 /*
 683                  * The disk must be active because we're syncing.
 684                  * We schedule xfssyncd now (now that the disk is
 685                  * active) instead of later (when it might not be).
 686                  */
 687                 wake_up_process(vfsp->vfs_sync_task);
 688                 /*
 689                  * We have to wait for the sync iteration to complete.
 690                  * If we don't, the disk activity caused by the sync
 691                  * will come after the sync is completed, and that
 692                  * triggers another sync from laptop mode.
 693                  */
 694                 wait_event(vfsp->vfs_wait_single_sync_task,
 695                                 vfsp->vfs_sync_seq != prev_sync_seq);
 696         }
 697
 698         return -error;
 699 }
 700
 701 STATIC int
 702 xfs_fs_statfs(
 703         struct dentry           *dentry,
 704         struct kstatfs          *statp)
 705 {
 706         return -bhv_vfs_statvfs(vfs_from_sb(dentry->d_sb), statp,
 707                                 vn_from_inode(dentry->d_inode));
 708 }
 709
 710 STATIC int
 711 xfs_fs_remount(
 712         struct super_block      *sb,
 713         int                     *flags,
 714         char                    *options)
 715 {
 716         bhv_vfs_t               *vfsp = vfs_from_sb(sb);
 717         struct xfs_mount_args   *args = xfs_args_allocate(sb, 0);
 718         int                     error;
 719
 720         error = bhv_vfs_parseargs(vfsp, options, args, 1);
 721         if ((args->flags2 & XFSMNT2_TAGGED) &&
 722                 !(sb->s_flags & MS_TAGGED)) {
 723                 printk("XFS: %s: tagging not permitted on remount.\n",
 724                         sb->s_id);
 725                 error = EINVAL;
 726         }
 727         if (!error)
 728                 error = bhv_vfs_mntupdate(vfsp, flags, args);
 729         kmem_free(args, sizeof(*args));
 730         return -error;
 731 }
 732
 733 STATIC void
 734 xfs_fs_lockfs(
 735         struct super_block      *sb)
 736 {
 737         bhv_vfs_freeze(vfs_from_sb(sb));
 738 }
 739
 740 STATIC int
 741 xfs_fs_show_options(
 742         struct seq_file         *m,
 743         struct vfsmount         *mnt)
 744 {
 745         return -bhv_vfs_showargs(vfs_from_sb(mnt->mnt_sb), m);
 746 }
 747
 748 STATIC int
 749 xfs_fs_quotasync(
 750         struct super_block      *sb,
 751         int                     type)
 752 {
 753         return -bhv_vfs_quotactl(vfs_from_sb(sb), Q_XQUOTASYNC, 0, NULL);
 754 }
 755
 756 STATIC int
 757 xfs_fs_getxstate(
 758         struct super_block      *sb,
 759         struct fs_quota_stat    *fqs)
 760 {
 761         return -bhv_vfs_quotactl(vfs_from_sb(sb), Q_XGETQSTAT, 0, (caddr_t)fqs);
 762 }
 763
 764 STATIC int
 765 xfs_fs_setxstate(
 766         struct super_block      *sb,
 767         unsigned int            flags,
 768         int                     op)
 769 {
 770         return -bhv_vfs_quotactl(vfs_from_sb(sb), op, 0, (caddr_t)&flags);
 771 }
 772
 773 STATIC int
 774 xfs_fs_getxquota(
 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_XGETQUOTA :
 782                                   ((type == GRPQUOTA) ? Q_XGETGQUOTA :
 783                                    Q_XGETPQUOTA), id, (caddr_t)fdq);
 784 }
 785
 786 STATIC int
 787 xfs_fs_setxquota(
 788         struct super_block      *sb,
 789         int                     type,
 790         qid_t                   id,
 791         struct fs_disk_quota    *fdq)
 792 {
 793         return -bhv_vfs_quotactl(vfs_from_sb(sb),
 794                                  (type == USRQUOTA) ? Q_XSETQLIM :
 795                                   ((type == GRPQUOTA) ? Q_XSETGQLIM :
 796                                    Q_XSETPQLIM), id, (caddr_t)fdq);
 797 }
 798
 799 STATIC int
 800 xfs_fs_fill_super(
 801         struct super_block      *sb,
 802         void                    *data,
 803         int                     silent)
 804 {
 805         struct bhv_vnode        *rootvp;
 806         struct bhv_vfs          *vfsp = vfs_allocate(sb);
 807         struct xfs_mount_args   *args = xfs_args_allocate(sb, silent);
 808         struct kstatfs          statvfs;
 809         int                     error;
 810
 811         bhv_insert_all_vfsops(vfsp);
 812
 813         error = bhv_vfs_parseargs(vfsp, (char *)data, args, 0);
 814         if (error) {
 815                 bhv_remove_all_vfsops(vfsp, 1);
 816                 goto fail_vfsop;
 817         }
 818
 819         sb_min_blocksize(sb, BBSIZE);
 820         sb->s_export_op = &xfs_export_operations;
 821         sb->s_qcop = &xfs_quotactl_operations;
 822         sb->s_op = &xfs_super_operations;
 823
 824         error = bhv_vfs_mount(vfsp, args, NULL);
 825         if (error) {
 826                 bhv_remove_all_vfsops(vfsp, 1);
 827                 goto fail_vfsop;
 828         }
 829
 830         error = bhv_vfs_statvfs(vfsp, &statvfs, NULL);
 831         if (error)
 832                 goto fail_unmount;
 833
 834         sb->s_dirt = 1;
 835         sb->s_magic = statvfs.f_type;
 836         sb->s_blocksize = statvfs.f_bsize;
 837         sb->s_blocksize_bits = ffs(statvfs.f_bsize) - 1;
 838         sb->s_maxbytes = xfs_max_file_offset(sb->s_blocksize_bits);
 839         sb->s_time_gran = 1;
 840         set_posix_acl_flag(sb);
 841
 842         error = bhv_vfs_root(vfsp, &rootvp);
 843         if (error)
 844                 goto fail_unmount;
 845
 846         sb->s_root = d_alloc_root(vn_to_inode(rootvp));
 847         if (!sb->s_root) {
 848                 error = ENOMEM;
 849                 goto fail_vnrele;
 850         }
 851         if (is_bad_inode(sb->s_root->d_inode)) {
 852                 error = EINVAL;
 853                 goto fail_vnrele;
 854         }
 855         if ((error = xfs_fs_start_syncd(vfsp)))
 856                 goto fail_vnrele;
 857         vn_trace_exit(rootvp, __FUNCTION__, (inst_t *)__return_address);
 858
 859         kmem_free(args, sizeof(*args));
 860         return 0;
 861
 862 fail_vnrele:
 863         if (sb->s_root) {
 864                 dput(sb->s_root);
 865                 sb->s_root = NULL;
 866         } else {
 867                 VN_RELE(rootvp);
 868         }
 869
 870 fail_unmount:
 871         bhv_vfs_unmount(vfsp, 0, NULL);
 872
 873 fail_vfsop:
 874         vfs_deallocate(vfsp);
 875         kmem_free(args, sizeof(*args));
 876         return -error;
 877 }
 878
 879 STATIC int
 880 xfs_fs_get_sb(
 881         struct file_system_type *fs_type,
 882         int                     flags,
 883         const char              *dev_name,
 884         void                    *data,
 885         struct vfsmount         *mnt)
 886 {
 887         return get_sb_bdev(fs_type, flags, dev_name, data, xfs_fs_fill_super,
 888                            mnt);
 889 }
 890
 891 STATIC struct super_operations xfs_super_operations = {
 892         .alloc_inode            = xfs_fs_alloc_inode,
 893         .destroy_inode          = xfs_fs_destroy_inode,
 894         .write_inode            = xfs_fs_write_inode,
 895         .clear_inode            = xfs_fs_clear_inode,
 896         .put_super              = xfs_fs_put_super,
 897         .write_super            = xfs_fs_write_super,
 898         .sync_fs                = xfs_fs_sync_super,
 899         .write_super_lockfs     = xfs_fs_lockfs,
 900         .statfs                 = xfs_fs_statfs,
 901         .remount_fs             = xfs_fs_remount,
 902         .show_options           = xfs_fs_show_options,
 903 };
 904
 905 STATIC struct quotactl_ops xfs_quotactl_operations = {
 906         .quota_sync             = xfs_fs_quotasync,
 907         .get_xstate             = xfs_fs_getxstate,
 908         .set_xstate             = xfs_fs_setxstate,
 909         .get_xquota             = xfs_fs_getxquota,
 910         .set_xquota             = xfs_fs_setxquota,
 911 };
 912
 913 STATIC struct file_system_type xfs_fs_type = {
 914         .owner                  = THIS_MODULE,
 915         .name                   = "xfs",
 916         .get_sb                 = xfs_fs_get_sb,
 917         .kill_sb                = kill_block_super,
 918         .fs_flags               = FS_REQUIRES_DEV,
 919 };
 920
 921
 922 STATIC int __init
 923 init_xfs_fs( void )
 924 {
 925         int                     error;
 926         struct sysinfo          si;
 927         static char             message[] __initdata = KERN_INFO \
 928                 XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled\n";
 929
 930         printk(message);
 931
 932         si_meminfo(&si);
 933         xfs_physmem = si.totalram;
 934
 935         ktrace_init(64);
 936
 937         error = xfs_init_zones();
 938         if (error < 0)
 939                 goto undo_zones;
 940
 941         error = xfs_buf_init();
 942         if (error < 0)
 943                 goto undo_buffers;
 944
 945         vn_init();
 946         xfs_init();
 947         uuid_init();
 948         vfs_initquota();
 949
 950         error = register_filesystem(&xfs_fs_type);
 951         if (error)
 952                 goto undo_register;
 953         return 0;
 954
 955 undo_register:
 956         xfs_buf_terminate();
 957
 958 undo_buffers:
 959         xfs_destroy_zones();
 960
 961 undo_zones:
 962         return error;
 963 }
 964
 965 STATIC void __exit
 966 exit_xfs_fs( void )
 967 {
 968         vfs_exitquota();
 969         unregister_filesystem(&xfs_fs_type);
 970         xfs_cleanup();
 971         xfs_buf_terminate();
 972         xfs_destroy_zones();
 973         ktrace_uninit();
 974 }
 975
 976 module_init(init_xfs_fs);
 977 module_exit(exit_xfs_fs);
 978
 979 MODULE_AUTHOR("Silicon Graphics, Inc.");
 980 MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled");
 981 MODULE_LICENSE("GPL");