]> err.no Git - linux-2.6/blob - fs/xfs/linux-2.6/xfs_super.c
[XFS] kill v_vfsp member from struct bhv_vnode
[linux-2.6] / fs / xfs / linux-2.6 / xfs_super.c
1 /*
2  * Copyright (c) 2000-2006 Silicon Graphics, Inc.
3  * All Rights Reserved.
4  *
5  * This program is free software; you can redistribute it and/or
6  * modify it under the terms of the GNU General Public License as
7  * published by the Free Software Foundation.
8  *
9  * This program is distributed in the hope that it would be useful,
10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
12  * GNU General Public License for more details.
13  *
14  * You should have received a copy of the GNU General Public License
15  * along with this program; if not, write the Free Software Foundation,
16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
17  */
18 #include "xfs.h"
19 #include "xfs_bit.h"
20 #include "xfs_log.h"
21 #include "xfs_clnt.h"
22 #include "xfs_inum.h"
23 #include "xfs_trans.h"
24 #include "xfs_sb.h"
25 #include "xfs_ag.h"
26 #include "xfs_dir2.h"
27 #include "xfs_alloc.h"
28 #include "xfs_dmapi.h"
29 #include "xfs_quota.h"
30 #include "xfs_mount.h"
31 #include "xfs_bmap_btree.h"
32 #include "xfs_alloc_btree.h"
33 #include "xfs_ialloc_btree.h"
34 #include "xfs_dir2_sf.h"
35 #include "xfs_attr_sf.h"
36 #include "xfs_dinode.h"
37 #include "xfs_inode.h"
38 #include "xfs_btree.h"
39 #include "xfs_ialloc.h"
40 #include "xfs_bmap.h"
41 #include "xfs_rtalloc.h"
42 #include "xfs_error.h"
43 #include "xfs_itable.h"
44 #include "xfs_rw.h"
45 #include "xfs_acl.h"
46 #include "xfs_attr.h"
47 #include "xfs_buf_item.h"
48 #include "xfs_utils.h"
49 #include "xfs_vnodeops.h"
50 #include "xfs_version.h"
51
52 #include <linux/namei.h>
53 #include <linux/init.h>
54 #include <linux/mount.h>
55 #include <linux/mempool.h>
56 #include <linux/writeback.h>
57 #include <linux/kthread.h>
58 #include <linux/freezer.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         struct xfs_inode        *ip,
208         int                     unlock)
209 {
210         struct inode            *inode = vn_to_inode(vp);
211
212         if (!ip->i_vnode) {
213                 ip->i_vnode = vp;
214                 inode->i_private = ip;
215         }
216
217         /*
218          * We need to set the ops vectors, and unlock the inode, but if
219          * we have been called during the new inode create process, it is
220          * too early to fill in the Linux inode.  We will get called a
221          * second time once the inode is properly set up, and then we can
222          * finish our work.
223          */
224         if (ip->i_d.di_mode != 0 && unlock && (inode->i_state & I_NEW)) {
225                 xfs_revalidate_inode(XFS_BHVTOM(bdp), vp, ip);
226                 xfs_set_inodeops(inode);
227
228                 xfs_iflags_clear(ip, XFS_INEW);
229                 barrier();
230
231                 unlock_new_inode(inode);
232         }
233 }
234
235 int
236 xfs_blkdev_get(
237         xfs_mount_t             *mp,
238         const char              *name,
239         struct block_device     **bdevp)
240 {
241         int                     error = 0;
242
243         *bdevp = open_bdev_excl(name, 0, mp);
244         if (IS_ERR(*bdevp)) {
245                 error = PTR_ERR(*bdevp);
246                 printk("XFS: Invalid device [%s], error=%d\n", name, error);
247         }
248
249         return -error;
250 }
251
252 void
253 xfs_blkdev_put(
254         struct block_device     *bdev)
255 {
256         if (bdev)
257                 close_bdev_excl(bdev);
258 }
259
260 /*
261  * Try to write out the superblock using barriers.
262  */
263 STATIC int
264 xfs_barrier_test(
265         xfs_mount_t     *mp)
266 {
267         xfs_buf_t       *sbp = xfs_getsb(mp, 0);
268         int             error;
269
270         XFS_BUF_UNDONE(sbp);
271         XFS_BUF_UNREAD(sbp);
272         XFS_BUF_UNDELAYWRITE(sbp);
273         XFS_BUF_WRITE(sbp);
274         XFS_BUF_UNASYNC(sbp);
275         XFS_BUF_ORDERED(sbp);
276
277         xfsbdstrat(mp, sbp);
278         error = xfs_iowait(sbp);
279
280         /*
281          * Clear all the flags we set and possible error state in the
282          * buffer.  We only did the write to try out whether barriers
283          * worked and shouldn't leave any traces in the superblock
284          * buffer.
285          */
286         XFS_BUF_DONE(sbp);
287         XFS_BUF_ERROR(sbp, 0);
288         XFS_BUF_UNORDERED(sbp);
289
290         xfs_buf_relse(sbp);
291         return error;
292 }
293
294 void
295 xfs_mountfs_check_barriers(xfs_mount_t *mp)
296 {
297         int error;
298
299         if (mp->m_logdev_targp != mp->m_ddev_targp) {
300                 xfs_fs_cmn_err(CE_NOTE, mp,
301                   "Disabling barriers, not supported with external log device");
302                 mp->m_flags &= ~XFS_MOUNT_BARRIER;
303                 return;
304         }
305
306         if (xfs_readonly_buftarg(mp->m_ddev_targp)) {
307                 xfs_fs_cmn_err(CE_NOTE, mp,
308                   "Disabling barriers, underlying device is readonly");
309                 mp->m_flags &= ~XFS_MOUNT_BARRIER;
310                 return;
311         }
312
313         error = xfs_barrier_test(mp);
314         if (error) {
315                 xfs_fs_cmn_err(CE_NOTE, mp,
316                   "Disabling barriers, trial barrier write failed");
317                 mp->m_flags &= ~XFS_MOUNT_BARRIER;
318                 return;
319         }
320 }
321
322 void
323 xfs_blkdev_issue_flush(
324         xfs_buftarg_t           *buftarg)
325 {
326         blkdev_issue_flush(buftarg->bt_bdev, NULL);
327 }
328
329 STATIC struct inode *
330 xfs_fs_alloc_inode(
331         struct super_block      *sb)
332 {
333         bhv_vnode_t             *vp;
334
335         vp = kmem_zone_alloc(xfs_vnode_zone, KM_SLEEP);
336         if (unlikely(!vp))
337                 return NULL;
338         return vn_to_inode(vp);
339 }
340
341 STATIC void
342 xfs_fs_destroy_inode(
343         struct inode            *inode)
344 {
345         kmem_zone_free(xfs_vnode_zone, vn_from_inode(inode));
346 }
347
348 STATIC void
349 xfs_fs_inode_init_once(
350         void                    *vnode,
351         kmem_zone_t             *zonep,
352         unsigned long           flags)
353 {
354         inode_init_once(vn_to_inode((bhv_vnode_t *)vnode));
355 }
356
357 STATIC int
358 xfs_init_zones(void)
359 {
360         xfs_vnode_zone = kmem_zone_init_flags(sizeof(bhv_vnode_t), "xfs_vnode",
361                                         KM_ZONE_HWALIGN | KM_ZONE_RECLAIM |
362                                         KM_ZONE_SPREAD,
363                                         xfs_fs_inode_init_once);
364         if (!xfs_vnode_zone)
365                 goto out;
366
367         xfs_ioend_zone = kmem_zone_init(sizeof(xfs_ioend_t), "xfs_ioend");
368         if (!xfs_ioend_zone)
369                 goto out_destroy_vnode_zone;
370
371         xfs_ioend_pool = mempool_create_slab_pool(4 * MAX_BUF_PER_PAGE,
372                                                   xfs_ioend_zone);
373         if (!xfs_ioend_pool)
374                 goto out_free_ioend_zone;
375         return 0;
376
377  out_free_ioend_zone:
378         kmem_zone_destroy(xfs_ioend_zone);
379  out_destroy_vnode_zone:
380         kmem_zone_destroy(xfs_vnode_zone);
381  out:
382         return -ENOMEM;
383 }
384
385 STATIC void
386 xfs_destroy_zones(void)
387 {
388         mempool_destroy(xfs_ioend_pool);
389         kmem_zone_destroy(xfs_vnode_zone);
390         kmem_zone_destroy(xfs_ioend_zone);
391 }
392
393 /*
394  * Attempt to flush the inode, this will actually fail
395  * if the inode is pinned, but we dirty the inode again
396  * at the point when it is unpinned after a log write,
397  * since this is when the inode itself becomes flushable.
398  */
399 STATIC int
400 xfs_fs_write_inode(
401         struct inode            *inode,
402         int                     sync)
403 {
404         int                     error = 0, flags = FLUSH_INODE;
405
406         vn_trace_entry(vn_from_inode(inode), __FUNCTION__,
407                         (inst_t *)__return_address);
408         if (sync) {
409                 filemap_fdatawait(inode->i_mapping);
410                 flags |= FLUSH_SYNC;
411         }
412         error = xfs_inode_flush(XFS_I(inode), flags);
413         if (error == EAGAIN) {
414                 if (sync)
415                         error = xfs_inode_flush(XFS_I(inode),
416                                                        flags | FLUSH_LOG);
417                 else
418                         error = 0;
419         }
420
421         return -error;
422 }
423
424 STATIC void
425 xfs_fs_clear_inode(
426         struct inode            *inode)
427 {
428         bhv_vnode_t             *vp = vn_from_inode(inode);
429
430         vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
431
432         XFS_STATS_INC(vn_rele);
433         XFS_STATS_INC(vn_remove);
434         XFS_STATS_INC(vn_reclaim);
435         XFS_STATS_DEC(vn_active);
436
437         /*
438          * This can happen because xfs_iget_core calls xfs_idestroy if we
439          * find an inode with di_mode == 0 but without IGET_CREATE set.
440          */
441         if (XFS_I(inode))
442                 xfs_inactive(XFS_I(inode));
443
444         VN_LOCK(vp);
445         vp->v_flag &= ~VMODIFIED;
446         VN_UNLOCK(vp, 0);
447
448         if (XFS_I(inode))
449                 if (xfs_reclaim(XFS_I(inode)))
450                         panic("%s: cannot reclaim 0x%p\n", __FUNCTION__, vp);
451
452         ASSERT(XFS_I(inode) == NULL);
453
454 #ifdef XFS_VNODE_TRACE
455         ktrace_free(vp->v_trace);
456 #endif
457 }
458
459 /*
460  * Enqueue a work item to be picked up by the vfs xfssyncd thread.
461  * Doing this has two advantages:
462  * - It saves on stack space, which is tight in certain situations
463  * - It can be used (with care) as a mechanism to avoid deadlocks.
464  * Flushing while allocating in a full filesystem requires both.
465  */
466 STATIC void
467 xfs_syncd_queue_work(
468         struct bhv_vfs  *vfs,
469         void            *data,
470         void            (*syncer)(bhv_vfs_t *, void *))
471 {
472         struct bhv_vfs_sync_work *work;
473
474         work = kmem_alloc(sizeof(struct bhv_vfs_sync_work), KM_SLEEP);
475         INIT_LIST_HEAD(&work->w_list);
476         work->w_syncer = syncer;
477         work->w_data = data;
478         work->w_vfs = vfs;
479         spin_lock(&vfs->vfs_sync_lock);
480         list_add_tail(&work->w_list, &vfs->vfs_sync_list);
481         spin_unlock(&vfs->vfs_sync_lock);
482         wake_up_process(vfs->vfs_sync_task);
483 }
484
485 /*
486  * Flush delayed allocate data, attempting to free up reserved space
487  * from existing allocations.  At this point a new allocation attempt
488  * has failed with ENOSPC and we are in the process of scratching our
489  * heads, looking about for more room...
490  */
491 STATIC void
492 xfs_flush_inode_work(
493         bhv_vfs_t       *vfs,
494         void            *inode)
495 {
496         filemap_flush(((struct inode *)inode)->i_mapping);
497         iput((struct inode *)inode);
498 }
499
500 void
501 xfs_flush_inode(
502         xfs_inode_t     *ip)
503 {
504         struct inode    *inode = vn_to_inode(XFS_ITOV(ip));
505         struct bhv_vfs  *vfs = XFS_MTOVFS(ip->i_mount);
506
507         igrab(inode);
508         xfs_syncd_queue_work(vfs, inode, xfs_flush_inode_work);
509         delay(msecs_to_jiffies(500));
510 }
511
512 /*
513  * This is the "bigger hammer" version of xfs_flush_inode_work...
514  * (IOW, "If at first you don't succeed, use a Bigger Hammer").
515  */
516 STATIC void
517 xfs_flush_device_work(
518         bhv_vfs_t       *vfs,
519         void            *inode)
520 {
521         sync_blockdev(vfs->vfs_super->s_bdev);
522         iput((struct inode *)inode);
523 }
524
525 void
526 xfs_flush_device(
527         xfs_inode_t     *ip)
528 {
529         struct inode    *inode = vn_to_inode(XFS_ITOV(ip));
530         struct bhv_vfs  *vfs = XFS_MTOVFS(ip->i_mount);
531
532         igrab(inode);
533         xfs_syncd_queue_work(vfs, inode, xfs_flush_device_work);
534         delay(msecs_to_jiffies(500));
535         xfs_log_force(ip->i_mount, (xfs_lsn_t)0, XFS_LOG_FORCE|XFS_LOG_SYNC);
536 }
537
538 STATIC void
539 vfs_sync_worker(
540         bhv_vfs_t       *vfsp,
541         void            *unused)
542 {
543         int             error;
544
545         if (!(vfsp->vfs_flag & VFS_RDONLY))
546                 error = bhv_vfs_sync(vfsp, SYNC_FSDATA | SYNC_BDFLUSH | \
547                                         SYNC_ATTR | SYNC_REFCACHE | SYNC_SUPER,
548                                         NULL);
549         vfsp->vfs_sync_seq++;
550         wake_up(&vfsp->vfs_wait_single_sync_task);
551 }
552
553 STATIC int
554 xfssyncd(
555         void                    *arg)
556 {
557         long                    timeleft;
558         bhv_vfs_t               *vfsp = (bhv_vfs_t *) arg;
559         bhv_vfs_sync_work_t     *work, *n;
560         LIST_HEAD               (tmp);
561
562         set_freezable();
563         timeleft = xfs_syncd_centisecs * msecs_to_jiffies(10);
564         for (;;) {
565                 timeleft = schedule_timeout_interruptible(timeleft);
566                 /* swsusp */
567                 try_to_freeze();
568                 if (kthread_should_stop() && list_empty(&vfsp->vfs_sync_list))
569                         break;
570
571                 spin_lock(&vfsp->vfs_sync_lock);
572                 /*
573                  * We can get woken by laptop mode, to do a sync -
574                  * that's the (only!) case where the list would be
575                  * empty with time remaining.
576                  */
577                 if (!timeleft || list_empty(&vfsp->vfs_sync_list)) {
578                         if (!timeleft)
579                                 timeleft = xfs_syncd_centisecs *
580                                                         msecs_to_jiffies(10);
581                         INIT_LIST_HEAD(&vfsp->vfs_sync_work.w_list);
582                         list_add_tail(&vfsp->vfs_sync_work.w_list,
583                                         &vfsp->vfs_sync_list);
584                 }
585                 list_for_each_entry_safe(work, n, &vfsp->vfs_sync_list, w_list)
586                         list_move(&work->w_list, &tmp);
587                 spin_unlock(&vfsp->vfs_sync_lock);
588
589                 list_for_each_entry_safe(work, n, &tmp, w_list) {
590                         (*work->w_syncer)(vfsp, work->w_data);
591                         list_del(&work->w_list);
592                         if (work == &vfsp->vfs_sync_work)
593                                 continue;
594                         kmem_free(work, sizeof(struct bhv_vfs_sync_work));
595                 }
596         }
597
598         return 0;
599 }
600
601 STATIC int
602 xfs_fs_start_syncd(
603         bhv_vfs_t               *vfsp)
604 {
605         vfsp->vfs_sync_work.w_syncer = vfs_sync_worker;
606         vfsp->vfs_sync_work.w_vfs = vfsp;
607         vfsp->vfs_sync_task = kthread_run(xfssyncd, vfsp, "xfssyncd");
608         if (IS_ERR(vfsp->vfs_sync_task))
609                 return -PTR_ERR(vfsp->vfs_sync_task);
610         return 0;
611 }
612
613 STATIC void
614 xfs_fs_stop_syncd(
615         bhv_vfs_t               *vfsp)
616 {
617         kthread_stop(vfsp->vfs_sync_task);
618 }
619
620 STATIC void
621 xfs_fs_put_super(
622         struct super_block      *sb)
623 {
624         bhv_vfs_t               *vfsp = vfs_from_sb(sb);
625         int                     error;
626
627         xfs_fs_stop_syncd(vfsp);
628         bhv_vfs_sync(vfsp, SYNC_ATTR | SYNC_DELWRI, NULL);
629         error = bhv_vfs_unmount(vfsp, 0, NULL);
630         if (error) {
631                 printk("XFS: unmount got error=%d\n", error);
632                 printk("%s: vfs=0x%p left dangling!\n", __FUNCTION__, vfsp);
633         } else {
634                 vfs_deallocate(vfsp);
635         }
636 }
637
638 STATIC void
639 xfs_fs_write_super(
640         struct super_block      *sb)
641 {
642         if (!(sb->s_flags & MS_RDONLY))
643                 bhv_vfs_sync(vfs_from_sb(sb), SYNC_FSDATA, NULL);
644         sb->s_dirt = 0;
645 }
646
647 STATIC int
648 xfs_fs_sync_super(
649         struct super_block      *sb,
650         int                     wait)
651 {
652         bhv_vfs_t               *vfsp = vfs_from_sb(sb);
653         int                     error;
654         int                     flags;
655
656         if (unlikely(sb->s_frozen == SB_FREEZE_WRITE)) {
657                 /*
658                  * First stage of freeze - no more writers will make progress
659                  * now we are here, so we flush delwri and delalloc buffers
660                  * here, then wait for all I/O to complete.  Data is frozen at
661                  * that point. Metadata is not frozen, transactions can still
662                  * occur here so don't bother flushing the buftarg (i.e
663                  * SYNC_QUIESCE) because it'll just get dirty again.
664                  */
665                 flags = SYNC_DATA_QUIESCE;
666         } else
667                 flags = SYNC_FSDATA | (wait ? SYNC_WAIT : 0);
668
669         error = bhv_vfs_sync(vfsp, flags, NULL);
670         sb->s_dirt = 0;
671
672         if (unlikely(laptop_mode)) {
673                 int     prev_sync_seq = vfsp->vfs_sync_seq;
674
675                 /*
676                  * The disk must be active because we're syncing.
677                  * We schedule xfssyncd now (now that the disk is
678                  * active) instead of later (when it might not be).
679                  */
680                 wake_up_process(vfsp->vfs_sync_task);
681                 /*
682                  * We have to wait for the sync iteration to complete.
683                  * If we don't, the disk activity caused by the sync
684                  * will come after the sync is completed, and that
685                  * triggers another sync from laptop mode.
686                  */
687                 wait_event(vfsp->vfs_wait_single_sync_task,
688                                 vfsp->vfs_sync_seq != prev_sync_seq);
689         }
690
691         return -error;
692 }
693
694 STATIC int
695 xfs_fs_statfs(
696         struct dentry           *dentry,
697         struct kstatfs          *statp)
698 {
699         return -bhv_vfs_statvfs(vfs_from_sb(dentry->d_sb), statp,
700                                 vn_from_inode(dentry->d_inode));
701 }
702
703 STATIC int
704 xfs_fs_remount(
705         struct super_block      *sb,
706         int                     *flags,
707         char                    *options)
708 {
709         bhv_vfs_t               *vfsp = vfs_from_sb(sb);
710         struct xfs_mount_args   *args = xfs_args_allocate(sb, 0);
711         int                     error;
712
713         error = bhv_vfs_parseargs(vfsp, options, args, 1);
714         if (!error)
715                 error = bhv_vfs_mntupdate(vfsp, flags, args);
716         kmem_free(args, sizeof(*args));
717         return -error;
718 }
719
720 STATIC void
721 xfs_fs_lockfs(
722         struct super_block      *sb)
723 {
724         bhv_vfs_freeze(vfs_from_sb(sb));
725 }
726
727 STATIC int
728 xfs_fs_show_options(
729         struct seq_file         *m,
730         struct vfsmount         *mnt)
731 {
732         return -bhv_vfs_showargs(vfs_from_sb(mnt->mnt_sb), m);
733 }
734
735 STATIC int
736 xfs_fs_quotasync(
737         struct super_block      *sb,
738         int                     type)
739 {
740         return -bhv_vfs_quotactl(vfs_from_sb(sb), Q_XQUOTASYNC, 0, NULL);
741 }
742
743 STATIC int
744 xfs_fs_getxstate(
745         struct super_block      *sb,
746         struct fs_quota_stat    *fqs)
747 {
748         return -bhv_vfs_quotactl(vfs_from_sb(sb), Q_XGETQSTAT, 0, (caddr_t)fqs);
749 }
750
751 STATIC int
752 xfs_fs_setxstate(
753         struct super_block      *sb,
754         unsigned int            flags,
755         int                     op)
756 {
757         return -bhv_vfs_quotactl(vfs_from_sb(sb), op, 0, (caddr_t)&flags);
758 }
759
760 STATIC int
761 xfs_fs_getxquota(
762         struct super_block      *sb,
763         int                     type,
764         qid_t                   id,
765         struct fs_disk_quota    *fdq)
766 {
767         return -bhv_vfs_quotactl(vfs_from_sb(sb),
768                                  (type == USRQUOTA) ? Q_XGETQUOTA :
769                                   ((type == GRPQUOTA) ? Q_XGETGQUOTA :
770                                    Q_XGETPQUOTA), id, (caddr_t)fdq);
771 }
772
773 STATIC int
774 xfs_fs_setxquota(
775         struct super_block      *sb,
776         int                     type,
777         qid_t                   id,
778         struct fs_disk_quota    *fdq)
779 {
780         return -bhv_vfs_quotactl(vfs_from_sb(sb),
781                                  (type == USRQUOTA) ? Q_XSETQLIM :
782                                   ((type == GRPQUOTA) ? Q_XSETGQLIM :
783                                    Q_XSETPQLIM), id, (caddr_t)fdq);
784 }
785
786 STATIC int
787 xfs_fs_fill_super(
788         struct super_block      *sb,
789         void                    *data,
790         int                     silent)
791 {
792         struct bhv_vnode        *rootvp;
793         struct bhv_vfs          *vfsp = vfs_allocate(sb);
794         struct xfs_mount_args   *args = xfs_args_allocate(sb, silent);
795         struct kstatfs          statvfs;
796         int                     error;
797
798         bhv_insert_all_vfsops(vfsp);
799
800         error = bhv_vfs_parseargs(vfsp, (char *)data, args, 0);
801         if (error) {
802                 bhv_remove_all_vfsops(vfsp, 1);
803                 goto fail_vfsop;
804         }
805
806         sb_min_blocksize(sb, BBSIZE);
807         sb->s_export_op = &xfs_export_operations;
808         sb->s_qcop = &xfs_quotactl_operations;
809         sb->s_op = &xfs_super_operations;
810
811         error = bhv_vfs_mount(vfsp, args, NULL);
812         if (error) {
813                 bhv_remove_all_vfsops(vfsp, 1);
814                 goto fail_vfsop;
815         }
816
817         error = bhv_vfs_statvfs(vfsp, &statvfs, NULL);
818         if (error)
819                 goto fail_unmount;
820
821         sb->s_dirt = 1;
822         sb->s_magic = statvfs.f_type;
823         sb->s_blocksize = statvfs.f_bsize;
824         sb->s_blocksize_bits = ffs(statvfs.f_bsize) - 1;
825         sb->s_maxbytes = xfs_max_file_offset(sb->s_blocksize_bits);
826         sb->s_time_gran = 1;
827         set_posix_acl_flag(sb);
828
829         error = bhv_vfs_root(vfsp, &rootvp);
830         if (error)
831                 goto fail_unmount;
832
833         sb->s_root = d_alloc_root(vn_to_inode(rootvp));
834         if (!sb->s_root) {
835                 error = ENOMEM;
836                 goto fail_vnrele;
837         }
838         if (is_bad_inode(sb->s_root->d_inode)) {
839                 error = EINVAL;
840                 goto fail_vnrele;
841         }
842         if ((error = xfs_fs_start_syncd(vfsp)))
843                 goto fail_vnrele;
844         vn_trace_exit(rootvp, __FUNCTION__, (inst_t *)__return_address);
845
846         kmem_free(args, sizeof(*args));
847         return 0;
848
849 fail_vnrele:
850         if (sb->s_root) {
851                 dput(sb->s_root);
852                 sb->s_root = NULL;
853         } else {
854                 VN_RELE(rootvp);
855         }
856
857 fail_unmount:
858         bhv_vfs_unmount(vfsp, 0, NULL);
859
860 fail_vfsop:
861         vfs_deallocate(vfsp);
862         kmem_free(args, sizeof(*args));
863         return -error;
864 }
865
866 STATIC int
867 xfs_fs_get_sb(
868         struct file_system_type *fs_type,
869         int                     flags,
870         const char              *dev_name,
871         void                    *data,
872         struct vfsmount         *mnt)
873 {
874         return get_sb_bdev(fs_type, flags, dev_name, data, xfs_fs_fill_super,
875                            mnt);
876 }
877
878 static struct super_operations xfs_super_operations = {
879         .alloc_inode            = xfs_fs_alloc_inode,
880         .destroy_inode          = xfs_fs_destroy_inode,
881         .write_inode            = xfs_fs_write_inode,
882         .clear_inode            = xfs_fs_clear_inode,
883         .put_super              = xfs_fs_put_super,
884         .write_super            = xfs_fs_write_super,
885         .sync_fs                = xfs_fs_sync_super,
886         .write_super_lockfs     = xfs_fs_lockfs,
887         .statfs                 = xfs_fs_statfs,
888         .remount_fs             = xfs_fs_remount,
889         .show_options           = xfs_fs_show_options,
890 };
891
892 static struct quotactl_ops xfs_quotactl_operations = {
893         .quota_sync             = xfs_fs_quotasync,
894         .get_xstate             = xfs_fs_getxstate,
895         .set_xstate             = xfs_fs_setxstate,
896         .get_xquota             = xfs_fs_getxquota,
897         .set_xquota             = xfs_fs_setxquota,
898 };
899
900 static struct file_system_type xfs_fs_type = {
901         .owner                  = THIS_MODULE,
902         .name                   = "xfs",
903         .get_sb                 = xfs_fs_get_sb,
904         .kill_sb                = kill_block_super,
905         .fs_flags               = FS_REQUIRES_DEV,
906 };
907
908
909 STATIC int __init
910 init_xfs_fs( void )
911 {
912         int                     error;
913         static char             message[] __initdata = KERN_INFO \
914                 XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled\n";
915
916         printk(message);
917
918         ktrace_init(64);
919
920         error = xfs_init_zones();
921         if (error < 0)
922                 goto undo_zones;
923
924         error = xfs_buf_init();
925         if (error < 0)
926                 goto undo_buffers;
927
928         vn_init();
929         xfs_init();
930         uuid_init();
931         vfs_initquota();
932
933         error = register_filesystem(&xfs_fs_type);
934         if (error)
935                 goto undo_register;
936         return 0;
937
938 undo_register:
939         xfs_buf_terminate();
940
941 undo_buffers:
942         xfs_destroy_zones();
943
944 undo_zones:
945         return error;
946 }
947
948 STATIC void __exit
949 exit_xfs_fs( void )
950 {
951         vfs_exitquota();
952         unregister_filesystem(&xfs_fs_type);
953         xfs_cleanup();
954         xfs_buf_terminate();
955         xfs_destroy_zones();
956         ktrace_uninit();
957 }
958
959 module_init(init_xfs_fs);
960 module_exit(exit_xfs_fs);
961
962 MODULE_AUTHOR("Silicon Graphics, Inc.");
963 MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled");
964 MODULE_LICENSE("GPL");