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         inode->i_xid    = ip->i_d.di_xid;
 162
 163         switch (inode->i_mode & S_IFMT) {
 164         case S_IFBLK:
 165         case S_IFCHR:
 166                 inode->i_rdev =
 167                         MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff,
 168                               sysv_minor(ip->i_df.if_u2.if_rdev));
 169                 break;
 170         default:
 171                 inode->i_rdev = 0;
 172                 break;
 173         }
 174
 175         inode->i_generation = ip->i_d.di_gen;
 176         i_size_write(inode, ip->i_d.di_size);
 177         inode->i_blocks =
 178                 XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
 179         inode->i_atime.tv_sec   = ip->i_d.di_atime.t_sec;
 180         inode->i_atime.tv_nsec  = ip->i_d.di_atime.t_nsec;
 181         inode->i_mtime.tv_sec   = ip->i_d.di_mtime.t_sec;
 182         inode->i_mtime.tv_nsec  = ip->i_d.di_mtime.t_nsec;
 183         inode->i_ctime.tv_sec   = ip->i_d.di_ctime.t_sec;
 184         inode->i_ctime.tv_nsec  = ip->i_d.di_ctime.t_nsec;
 185         if (ip->i_d.di_flags & XFS_DIFLAG_IMMUTABLE)
 186                 inode->i_flags |= S_IMMUTABLE;
 187         else
 188                 inode->i_flags &= ~S_IMMUTABLE;
 189         if (ip->i_d.di_flags & XFS_DIFLAG_IUNLINK)
 190                 inode->i_flags |= S_IUNLINK;
 191         else
 192                 inode->i_flags &= ~S_IUNLINK;
 193         if (ip->i_d.di_flags & XFS_DIFLAG_BARRIER)
 194                 inode->i_flags |= S_BARRIER;
 195         else
 196                 inode->i_flags &= ~S_BARRIER;
 197         if (ip->i_d.di_flags & XFS_DIFLAG_APPEND)
 198                 inode->i_flags |= S_APPEND;
 199         else
 200                 inode->i_flags &= ~S_APPEND;
 201         if (ip->i_d.di_flags & XFS_DIFLAG_SYNC)
 202                 inode->i_flags |= S_SYNC;
 203         else
 204                 inode->i_flags &= ~S_SYNC;
 205         if (ip->i_d.di_flags & XFS_DIFLAG_NOATIME)
 206                 inode->i_flags |= S_NOATIME;
 207         else
 208                 inode->i_flags &= ~S_NOATIME;
 209         vp->v_flag &= ~VMODIFIED;
 210 }
 211
 212 void
 213 xfs_initialize_vnode(
 214         bhv_desc_t              *bdp,
 215         bhv_vnode_t             *vp,
 216         bhv_desc_t              *inode_bhv,
 217         int                     unlock)
 218 {
 219         xfs_inode_t             *ip = XFS_BHVTOI(inode_bhv);
 220         struct inode            *inode = vn_to_inode(vp);
 221
 222         if (!inode_bhv->bd_vobj) {
 223                 vp->v_vfsp = bhvtovfs(bdp);
 224                 bhv_desc_init(inode_bhv, ip, vp, &xfs_vnodeops);
 225                 bhv_insert(VN_BHV_HEAD(vp), inode_bhv);
 226         }
 227
 228         /*
 229          * We need to set the ops vectors, and unlock the inode, but if
 230          * we have been called during the new inode create process, it is
 231          * too early to fill in the Linux inode.  We will get called a
 232          * second time once the inode is properly set up, and then we can
 233          * finish our work.
 234          */
 235         if (ip->i_d.di_mode != 0 && unlock && (inode->i_state & I_NEW)) {
 236                 xfs_revalidate_inode(XFS_BHVTOM(bdp), vp, ip);
 237                 xfs_set_inodeops(inode);
 238
 239                 ip->i_flags &= ~XFS_INEW;
 240                 barrier();
 241
 242                 unlock_new_inode(inode);
 243         }
 244 }
 245
 246 int
 247 xfs_blkdev_get(
 248         xfs_mount_t             *mp,
 249         const char              *name,
 250         struct block_device     **bdevp)
 251 {
 252         int                     error = 0;
 253
 254         *bdevp = open_bdev_excl(name, 0, mp);
 255         if (IS_ERR(*bdevp)) {
 256                 error = PTR_ERR(*bdevp);
 257                 printk("XFS: Invalid device [%s], error=%d\n", name, error);
 258         }
 259
 260         return -error;
 261 }
 262
 263 void
 264 xfs_blkdev_put(
 265         struct block_device     *bdev)
 266 {
 267         if (bdev)
 268                 close_bdev_excl(bdev);
 269 }
 270
 271 /*
 272  * Try to write out the superblock using barriers.
 273  */
 274 STATIC int
 275 xfs_barrier_test(
 276         xfs_mount_t     *mp)
 277 {
 278         xfs_buf_t       *sbp = xfs_getsb(mp, 0);
 279         int             error;
 280
 281         XFS_BUF_UNDONE(sbp);
 282         XFS_BUF_UNREAD(sbp);
 283         XFS_BUF_UNDELAYWRITE(sbp);
 284         XFS_BUF_WRITE(sbp);
 285         XFS_BUF_UNASYNC(sbp);
 286         XFS_BUF_ORDERED(sbp);
 287
 288         xfsbdstrat(mp, sbp);
 289         error = xfs_iowait(sbp);
 290
 291         /*
 292          * Clear all the flags we set and possible error state in the
 293          * buffer.  We only did the write to try out whether barriers
 294          * worked and shouldn't leave any traces in the superblock
 295          * buffer.
 296          */
 297         XFS_BUF_DONE(sbp);
 298         XFS_BUF_ERROR(sbp, 0);
 299         XFS_BUF_UNORDERED(sbp);
 300
 301         xfs_buf_relse(sbp);
 302         return error;
 303 }
 304
 305 void
 306 xfs_mountfs_check_barriers(xfs_mount_t *mp)
 307 {
 308         int error;
 309
 310         if (mp->m_logdev_targp != mp->m_ddev_targp) {
 311                 xfs_fs_cmn_err(CE_NOTE, mp,
 312                   "Disabling barriers, not supported with external log device");
 313                 mp->m_flags &= ~XFS_MOUNT_BARRIER;
 314                 return;
 315         }
 316
 317         if (mp->m_ddev_targp->bt_bdev->bd_disk->queue->ordered ==
 318                                         QUEUE_ORDERED_NONE) {
 319                 xfs_fs_cmn_err(CE_NOTE, mp,
 320                   "Disabling barriers, not supported by the underlying device");
 321                 mp->m_flags &= ~XFS_MOUNT_BARRIER;
 322                 return;
 323         }
 324
 325         if (xfs_readonly_buftarg(mp->m_ddev_targp)) {
 326                 xfs_fs_cmn_err(CE_NOTE, mp,
 327                   "Disabling barriers, underlying device is readonly");
 328                 mp->m_flags &= ~XFS_MOUNT_BARRIER;
 329                 return;
 330         }
 331
 332         error = xfs_barrier_test(mp);
 333         if (error) {
 334                 xfs_fs_cmn_err(CE_NOTE, mp,
 335                   "Disabling barriers, trial barrier write failed");
 336                 mp->m_flags &= ~XFS_MOUNT_BARRIER;
 337                 return;
 338         }
 339 }
 340
 341 void
 342 xfs_blkdev_issue_flush(
 343         xfs_buftarg_t           *buftarg)
 344 {
 345         blkdev_issue_flush(buftarg->bt_bdev, NULL);
 346 }
 347
 348 STATIC struct inode *
 349 xfs_fs_alloc_inode(
 350         struct super_block      *sb)
 351 {
 352         bhv_vnode_t             *vp;
 353
 354         vp = kmem_zone_alloc(xfs_vnode_zone, KM_SLEEP);
 355         if (unlikely(!vp))
 356                 return NULL;
 357         return vn_to_inode(vp);
 358 }
 359
 360 STATIC void
 361 xfs_fs_destroy_inode(
 362         struct inode            *inode)
 363 {
 364         kmem_zone_free(xfs_vnode_zone, vn_from_inode(inode));
 365 }
 366
 367 STATIC void
 368 xfs_fs_inode_init_once(
 369         void                    *vnode,
 370         kmem_zone_t             *zonep,
 371         unsigned long           flags)
 372 {
 373         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
 374                       SLAB_CTOR_CONSTRUCTOR)
 375                 inode_init_once(vn_to_inode((bhv_vnode_t *)vnode));
 376 }
 377
 378 STATIC int
 379 xfs_init_zones(void)
 380 {
 381         xfs_vnode_zone = kmem_zone_init_flags(sizeof(bhv_vnode_t), "xfs_vnode",
 382                                         KM_ZONE_HWALIGN | KM_ZONE_RECLAIM |
 383                                         KM_ZONE_SPREAD,
 384                                         xfs_fs_inode_init_once);
 385         if (!xfs_vnode_zone)
 386                 goto out;
 387
 388         xfs_ioend_zone = kmem_zone_init(sizeof(xfs_ioend_t), "xfs_ioend");
 389         if (!xfs_ioend_zone)
 390                 goto out_destroy_vnode_zone;
 391
 392         xfs_ioend_pool = mempool_create_slab_pool(4 * MAX_BUF_PER_PAGE,
 393                                                   xfs_ioend_zone);
 394         if (!xfs_ioend_pool)
 395                 goto out_free_ioend_zone;
 396         return 0;
 397
 398  out_free_ioend_zone:
 399         kmem_zone_destroy(xfs_ioend_zone);
 400  out_destroy_vnode_zone:
 401         kmem_zone_destroy(xfs_vnode_zone);
 402  out:
 403         return -ENOMEM;
 404 }
 405
 406 STATIC void
 407 xfs_destroy_zones(void)
 408 {
 409         mempool_destroy(xfs_ioend_pool);
 410         kmem_zone_destroy(xfs_vnode_zone);
 411         kmem_zone_destroy(xfs_ioend_zone);
 412 }
 413
 414 /*
 415  * Attempt to flush the inode, this will actually fail
 416  * if the inode is pinned, but we dirty the inode again
 417  * at the point when it is unpinned after a log write,
 418  * since this is when the inode itself becomes flushable.
 419  */
 420 STATIC int
 421 xfs_fs_write_inode(
 422         struct inode            *inode,
 423         int                     sync)
 424 {
 425         bhv_vnode_t             *vp = vn_from_inode(inode);
 426         int                     error = 0, flags = FLUSH_INODE;
 427
 428         if (vp) {
 429                 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
 430                 if (sync)
 431                         flags |= FLUSH_SYNC;
 432                 error = bhv_vop_iflush(vp, flags);
 433                 if (error == EAGAIN)
 434                         error = sync? bhv_vop_iflush(vp, flags | FLUSH_LOG) : 0;
 435         }
 436         return -error;
 437 }
 438
 439 STATIC void
 440 xfs_fs_clear_inode(
 441         struct inode            *inode)
 442 {
 443         bhv_vnode_t             *vp = vn_from_inode(inode);
 444
 445         vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
 446
 447         XFS_STATS_INC(vn_rele);
 448         XFS_STATS_INC(vn_remove);
 449         XFS_STATS_INC(vn_reclaim);
 450         XFS_STATS_DEC(vn_active);
 451
 452         /*
 453          * This can happen because xfs_iget_core calls xfs_idestroy if we
 454          * find an inode with di_mode == 0 but without IGET_CREATE set.
 455          */
 456         if (VNHEAD(vp))
 457                 bhv_vop_inactive(vp, NULL);
 458
 459         VN_LOCK(vp);
 460         vp->v_flag &= ~VMODIFIED;
 461         VN_UNLOCK(vp, 0);
 462
 463         if (VNHEAD(vp))
 464                 if (bhv_vop_reclaim(vp))
 465                         panic("%s: cannot reclaim 0x%p\n", __FUNCTION__, vp);
 466
 467         ASSERT(VNHEAD(vp) == NULL);
 468
 469 #ifdef XFS_VNODE_TRACE
 470         ktrace_free(vp->v_trace);
 471 #endif
 472 }
 473
 474 /*
 475  * Enqueue a work item to be picked up by the vfs xfssyncd thread.
 476  * Doing this has two advantages:
 477  * - It saves on stack space, which is tight in certain situations
 478  * - It can be used (with care) as a mechanism to avoid deadlocks.
 479  * Flushing while allocating in a full filesystem requires both.
 480  */
 481 STATIC void
 482 xfs_syncd_queue_work(
 483         struct bhv_vfs  *vfs,
 484         void            *data,
 485         void            (*syncer)(bhv_vfs_t *, void *))
 486 {
 487         struct bhv_vfs_sync_work *work;
 488
 489         work = kmem_alloc(sizeof(struct bhv_vfs_sync_work), KM_SLEEP);
 490         INIT_LIST_HEAD(&work->w_list);
 491         work->w_syncer = syncer;
 492         work->w_data = data;
 493         work->w_vfs = vfs;
 494         spin_lock(&vfs->vfs_sync_lock);
 495         list_add_tail(&work->w_list, &vfs->vfs_sync_list);
 496         spin_unlock(&vfs->vfs_sync_lock);
 497         wake_up_process(vfs->vfs_sync_task);
 498 }
 499
 500 /*
 501  * Flush delayed allocate data, attempting to free up reserved space
 502  * from existing allocations.  At this point a new allocation attempt
 503  * has failed with ENOSPC and we are in the process of scratching our
 504  * heads, looking about for more room...
 505  */
 506 STATIC void
 507 xfs_flush_inode_work(
 508         bhv_vfs_t       *vfs,
 509         void            *inode)
 510 {
 511         filemap_flush(((struct inode *)inode)->i_mapping);
 512         iput((struct inode *)inode);
 513 }
 514
 515 void
 516 xfs_flush_inode(
 517         xfs_inode_t     *ip)
 518 {
 519         struct inode    *inode = vn_to_inode(XFS_ITOV(ip));
 520         struct bhv_vfs  *vfs = XFS_MTOVFS(ip->i_mount);
 521
 522         igrab(inode);
 523         xfs_syncd_queue_work(vfs, inode, xfs_flush_inode_work);
 524         delay(msecs_to_jiffies(500));
 525 }
 526
 527 /*
 528  * This is the "bigger hammer" version of xfs_flush_inode_work...
 529  * (IOW, "If at first you don't succeed, use a Bigger Hammer").
 530  */
 531 STATIC void
 532 xfs_flush_device_work(
 533         bhv_vfs_t       *vfs,
 534         void            *inode)
 535 {
 536         sync_blockdev(vfs->vfs_super->s_bdev);
 537         iput((struct inode *)inode);
 538 }
 539
 540 void
 541 xfs_flush_device(
 542         xfs_inode_t     *ip)
 543 {
 544         struct inode    *inode = vn_to_inode(XFS_ITOV(ip));
 545         struct bhv_vfs  *vfs = XFS_MTOVFS(ip->i_mount);
 546
 547         igrab(inode);
 548         xfs_syncd_queue_work(vfs, inode, xfs_flush_device_work);
 549         delay(msecs_to_jiffies(500));
 550         xfs_log_force(ip->i_mount, (xfs_lsn_t)0, XFS_LOG_FORCE|XFS_LOG_SYNC);
 551 }
 552
 553 STATIC void
 554 vfs_sync_worker(
 555         bhv_vfs_t       *vfsp,
 556         void            *unused)
 557 {
 558         int             error;
 559
 560         if (!(vfsp->vfs_flag & VFS_RDONLY))
 561                 error = bhv_vfs_sync(vfsp, SYNC_FSDATA | SYNC_BDFLUSH | \
 562                                         SYNC_ATTR | SYNC_REFCACHE, NULL);
 563         vfsp->vfs_sync_seq++;
 564         wmb();
 565         wake_up(&vfsp->vfs_wait_single_sync_task);
 566 }
 567
 568 STATIC int
 569 xfssyncd(
 570         void                    *arg)
 571 {
 572         long                    timeleft;
 573         bhv_vfs_t               *vfsp = (bhv_vfs_t *) arg;
 574         bhv_vfs_sync_work_t     *work, *n;
 575         LIST_HEAD               (tmp);
 576
 577         timeleft = xfs_syncd_centisecs * msecs_to_jiffies(10);
 578         for (;;) {
 579                 timeleft = schedule_timeout_interruptible(timeleft);
 580                 /* swsusp */
 581                 try_to_freeze();
 582                 if (kthread_should_stop() && list_empty(&vfsp->vfs_sync_list))
 583                         break;
 584
 585                 spin_lock(&vfsp->vfs_sync_lock);
 586                 /*
 587                  * We can get woken by laptop mode, to do a sync -
 588                  * that's the (only!) case where the list would be
 589                  * empty with time remaining.
 590                  */
 591                 if (!timeleft || list_empty(&vfsp->vfs_sync_list)) {
 592                         if (!timeleft)
 593                                 timeleft = xfs_syncd_centisecs *
 594                                                         msecs_to_jiffies(10);
 595                         INIT_LIST_HEAD(&vfsp->vfs_sync_work.w_list);
 596                         list_add_tail(&vfsp->vfs_sync_work.w_list,
 597                                         &vfsp->vfs_sync_list);
 598                 }
 599                 list_for_each_entry_safe(work, n, &vfsp->vfs_sync_list, w_list)
 600                         list_move(&work->w_list, &tmp);
 601                 spin_unlock(&vfsp->vfs_sync_lock);
 602
 603                 list_for_each_entry_safe(work, n, &tmp, w_list) {
 604                         (*work->w_syncer)(vfsp, work->w_data);
 605                         list_del(&work->w_list);
 606                         if (work == &vfsp->vfs_sync_work)
 607                                 continue;
 608                         kmem_free(work, sizeof(struct bhv_vfs_sync_work));
 609                 }
 610         }
 611
 612         return 0;
 613 }
 614
 615 STATIC int
 616 xfs_fs_start_syncd(
 617         bhv_vfs_t               *vfsp)
 618 {
 619         vfsp->vfs_sync_work.w_syncer = vfs_sync_worker;
 620         vfsp->vfs_sync_work.w_vfs = vfsp;
 621         vfsp->vfs_sync_task = kthread_run(xfssyncd, vfsp, "xfssyncd");
 622         if (IS_ERR(vfsp->vfs_sync_task))
 623                 return -PTR_ERR(vfsp->vfs_sync_task);
 624         return 0;
 625 }
 626
 627 STATIC void
 628 xfs_fs_stop_syncd(
 629         bhv_vfs_t               *vfsp)
 630 {
 631         kthread_stop(vfsp->vfs_sync_task);
 632 }
 633
 634 STATIC void
 635 xfs_fs_put_super(
 636         struct super_block      *sb)
 637 {
 638         bhv_vfs_t               *vfsp = vfs_from_sb(sb);
 639         int                     error;
 640
 641         xfs_fs_stop_syncd(vfsp);
 642         bhv_vfs_sync(vfsp, SYNC_ATTR | SYNC_DELWRI, NULL);
 643         error = bhv_vfs_unmount(vfsp, 0, NULL);
 644         if (error) {
 645                 printk("XFS: unmount got error=%d\n", error);
 646                 printk("%s: vfs=0x%p left dangling!\n", __FUNCTION__, vfsp);
 647         } else {
 648                 vfs_deallocate(vfsp);
 649         }
 650 }
 651
 652 STATIC void
 653 xfs_fs_write_super(
 654         struct super_block      *sb)
 655 {
 656         if (!(sb->s_flags & MS_RDONLY))
 657                 bhv_vfs_sync(vfs_from_sb(sb), SYNC_FSDATA, NULL);
 658         sb->s_dirt = 0;
 659 }
 660
 661 STATIC int
 662 xfs_fs_sync_super(
 663         struct super_block      *sb,
 664         int                     wait)
 665 {
 666         bhv_vfs_t               *vfsp = vfs_from_sb(sb);
 667         int                     error;
 668         int                     flags;
 669
 670         if (unlikely(sb->s_frozen == SB_FREEZE_WRITE))
 671                 flags = SYNC_QUIESCE;
 672         else
 673                 flags = SYNC_FSDATA | (wait ? SYNC_WAIT : 0);
 674
 675         error = bhv_vfs_sync(vfsp, flags, NULL);
 676         sb->s_dirt = 0;
 677
 678         if (unlikely(laptop_mode)) {
 679                 int     prev_sync_seq = vfsp->vfs_sync_seq;
 680
 681                 /*
 682                  * The disk must be active because we're syncing.
 683                  * We schedule xfssyncd now (now that the disk is
 684                  * active) instead of later (when it might not be).
 685                  */
 686                 wake_up_process(vfsp->vfs_sync_task);
 687                 /*
 688                  * We have to wait for the sync iteration to complete.
 689                  * If we don't, the disk activity caused by the sync
 690                  * will come after the sync is completed, and that
 691                  * triggers another sync from laptop mode.
 692                  */
 693                 wait_event(vfsp->vfs_wait_single_sync_task,
 694                                 vfsp->vfs_sync_seq != prev_sync_seq);
 695         }
 696
 697         return -error;
 698 }
 699
 700 STATIC int
 701 xfs_fs_statfs(
 702         struct dentry           *dentry,
 703         struct kstatfs          *statp)
 704 {
 705         return -bhv_vfs_statvfs(vfs_from_sb(dentry->d_sb), statp,
 706                                 vn_from_inode(dentry->d_inode));
 707 }
 708
 709 STATIC int
 710 xfs_fs_remount(
 711         struct super_block      *sb,
 712         int                     *flags,
 713         char                    *options)
 714 {
 715         bhv_vfs_t               *vfsp = vfs_from_sb(sb);
 716         struct xfs_mount_args   *args = xfs_args_allocate(sb, 0);
 717         int                     error;
 718
 719         error = bhv_vfs_parseargs(vfsp, options, args, 1);
 720         if ((args->flags2 & XFSMNT2_TAGXID) &&
 721                 !(sb->s_flags & MS_TAGXID)) {
 722                 printk("XFS: %s: tagxid not permitted on remount.\n",
 723                         sb->s_id);
 724                 error = EINVAL;
 725         }
 726         if (!error)
 727                 error = bhv_vfs_mntupdate(vfsp, flags, args);
 728         kmem_free(args, sizeof(*args));
 729         return -error;
 730 }
 731
 732 STATIC void
 733 xfs_fs_lockfs(
 734         struct super_block      *sb)
 735 {
 736         bhv_vfs_freeze(vfs_from_sb(sb));
 737 }
 738
 739 STATIC int
 740 xfs_fs_show_options(
 741         struct seq_file         *m,
 742         struct vfsmount         *mnt)
 743 {
 744         return -bhv_vfs_showargs(vfs_from_sb(mnt->mnt_sb), m);
 745 }
 746
 747 STATIC int
 748 xfs_fs_quotasync(
 749         struct super_block      *sb,
 750         int                     type)
 751 {
 752         return -bhv_vfs_quotactl(vfs_from_sb(sb), Q_XQUOTASYNC, 0, NULL);
 753 }
 754
 755 STATIC int
 756 xfs_fs_getxstate(
 757         struct super_block      *sb,
 758         struct fs_quota_stat    *fqs)
 759 {
 760         return -bhv_vfs_quotactl(vfs_from_sb(sb), Q_XGETQSTAT, 0, (caddr_t)fqs);
 761 }
 762
 763 STATIC int
 764 xfs_fs_setxstate(
 765         struct super_block      *sb,
 766         unsigned int            flags,
 767         int                     op)
 768 {
 769         return -bhv_vfs_quotactl(vfs_from_sb(sb), op, 0, (caddr_t)&flags);
 770 }
 771
 772 STATIC int
 773 xfs_fs_getxquota(
 774         struct super_block      *sb,
 775         int                     type,
 776         qid_t                   id,
 777         struct fs_disk_quota    *fdq)
 778 {
 779         return -bhv_vfs_quotactl(vfs_from_sb(sb),
 780                                  (type == USRQUOTA) ? Q_XGETQUOTA :
 781                                   ((type == GRPQUOTA) ? Q_XGETGQUOTA :
 782                                    Q_XGETPQUOTA), id, (caddr_t)fdq);
 783 }
 784
 785 STATIC int
 786 xfs_fs_setxquota(
 787         struct super_block      *sb,
 788         int                     type,
 789         qid_t                   id,
 790         struct fs_disk_quota    *fdq)
 791 {
 792         return -bhv_vfs_quotactl(vfs_from_sb(sb),
 793                                  (type == USRQUOTA) ? Q_XSETQLIM :
 794                                   ((type == GRPQUOTA) ? Q_XSETGQLIM :
 795                                    Q_XSETPQLIM), id, (caddr_t)fdq);
 796 }
 797
 798 STATIC int
 799 xfs_fs_fill_super(
 800         struct super_block      *sb,
 801         void                    *data,
 802         int                     silent)
 803 {
 804         struct bhv_vnode        *rootvp;
 805         struct bhv_vfs          *vfsp = vfs_allocate(sb);
 806         struct xfs_mount_args   *args = xfs_args_allocate(sb, silent);
 807         struct kstatfs          statvfs;
 808         int                     error;
 809
 810         bhv_insert_all_vfsops(vfsp);
 811
 812         error = bhv_vfs_parseargs(vfsp, (char *)data, args, 0);
 813         if (error) {
 814                 bhv_remove_all_vfsops(vfsp, 1);
 815                 goto fail_vfsop;
 816         }
 817
 818         sb_min_blocksize(sb, BBSIZE);
 819         sb->s_export_op = &xfs_export_operations;
 820         sb->s_qcop = &xfs_quotactl_operations;
 821         sb->s_op = &xfs_super_operations;
 822
 823         error = bhv_vfs_mount(vfsp, args, NULL);
 824         if (error) {
 825                 bhv_remove_all_vfsops(vfsp, 1);
 826                 goto fail_vfsop;
 827         }
 828
 829         error = bhv_vfs_statvfs(vfsp, &statvfs, NULL);
 830         if (error)
 831                 goto fail_unmount;
 832
 833         sb->s_dirt = 1;
 834         sb->s_magic = statvfs.f_type;
 835         sb->s_blocksize = statvfs.f_bsize;
 836         sb->s_blocksize_bits = ffs(statvfs.f_bsize) - 1;
 837         sb->s_maxbytes = xfs_max_file_offset(sb->s_blocksize_bits);
 838         sb->s_time_gran = 1;
 839         set_posix_acl_flag(sb);
 840
 841         error = bhv_vfs_root(vfsp, &rootvp);
 842         if (error)
 843                 goto fail_unmount;
 844
 845         sb->s_root = d_alloc_root(vn_to_inode(rootvp));
 846         if (!sb->s_root) {
 847                 error = ENOMEM;
 848                 goto fail_vnrele;
 849         }
 850         if (is_bad_inode(sb->s_root->d_inode)) {
 851                 error = EINVAL;
 852                 goto fail_vnrele;
 853         }
 854         if ((error = xfs_fs_start_syncd(vfsp)))
 855                 goto fail_vnrele;
 856         vn_trace_exit(rootvp, __FUNCTION__, (inst_t *)__return_address);
 857
 858         kmem_free(args, sizeof(*args));
 859         return 0;
 860
 861 fail_vnrele:
 862         if (sb->s_root) {
 863                 dput(sb->s_root);
 864                 sb->s_root = NULL;
 865         } else {
 866                 VN_RELE(rootvp);
 867         }
 868
 869 fail_unmount:
 870         bhv_vfs_unmount(vfsp, 0, NULL);
 871
 872 fail_vfsop:
 873         vfs_deallocate(vfsp);
 874         kmem_free(args, sizeof(*args));
 875         return -error;
 876 }
 877
 878 STATIC int
 879 xfs_fs_get_sb(
 880         struct file_system_type *fs_type,
 881         int                     flags,
 882         const char              *dev_name,
 883         void                    *data,
 884         struct vfsmount         *mnt)
 885 {
 886         return get_sb_bdev(fs_type, flags, dev_name, data, xfs_fs_fill_super,
 887                            mnt);
 888 }
 889
 890 STATIC struct super_operations xfs_super_operations = {
 891         .alloc_inode            = xfs_fs_alloc_inode,
 892         .destroy_inode          = xfs_fs_destroy_inode,
 893         .write_inode            = xfs_fs_write_inode,
 894         .clear_inode            = xfs_fs_clear_inode,
 895         .put_super              = xfs_fs_put_super,
 896         .write_super            = xfs_fs_write_super,
 897         .sync_fs                = xfs_fs_sync_super,
 898         .write_super_lockfs     = xfs_fs_lockfs,
 899         .statfs                 = xfs_fs_statfs,
 900         .remount_fs             = xfs_fs_remount,
 901         .show_options           = xfs_fs_show_options,
 902 };
 903
 904 STATIC struct quotactl_ops xfs_quotactl_operations = {
 905         .quota_sync             = xfs_fs_quotasync,
 906         .get_xstate             = xfs_fs_getxstate,
 907         .set_xstate             = xfs_fs_setxstate,
 908         .get_xquota             = xfs_fs_getxquota,
 909         .set_xquota             = xfs_fs_setxquota,
 910 };
 911
 912 STATIC struct file_system_type xfs_fs_type = {
 913         .owner                  = THIS_MODULE,
 914         .name                   = "xfs",
 915         .get_sb                 = xfs_fs_get_sb,
 916         .kill_sb                = kill_block_super,
 917         .fs_flags               = FS_REQUIRES_DEV,
 918 };
 919
 920
 921 STATIC int __init
 922 init_xfs_fs( void )
 923 {
 924         int                     error;
 925         struct sysinfo          si;
 926         static char             message[] __initdata = KERN_INFO \
 927                 XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled\n";
 928
 929         printk(message);
 930
 931         si_meminfo(&si);
 932         xfs_physmem = si.totalram;
 933
 934         ktrace_init(64);
 935
 936         error = xfs_init_zones();
 937         if (error < 0)
 938                 goto undo_zones;
 939
 940         error = xfs_buf_init();
 941         if (error < 0)
 942                 goto undo_buffers;
 943
 944         vn_init();
 945         xfs_init();
 946         uuid_init();
 947         vfs_initquota();
 948
 949         error = register_filesystem(&xfs_fs_type);
 950         if (error)
 951                 goto undo_register;
 952         return 0;
 953
 954 undo_register:
 955         xfs_buf_terminate();
 956
 957 undo_buffers:
 958         xfs_destroy_zones();
 959
 960 undo_zones:
 961         return error;
 962 }
 963
 964 STATIC void __exit
 965 exit_xfs_fs( void )
 966 {
 967         vfs_exitquota();
 968         unregister_filesystem(&xfs_fs_type);
 969         xfs_cleanup();
 970         xfs_buf_terminate();
 971         xfs_destroy_zones();
 972         ktrace_uninit();
 973 }
 974
 975 module_init(init_xfs_fs);
 976 module_exit(exit_xfs_fs);
 977
 978 MODULE_AUTHOR("Silicon Graphics, Inc.");
 979 MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled");
 980 MODULE_LICENSE("GPL");