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