]> pilppa.org Git - linux-2.6-omap-h63xx.git/blob - fs/xfs/linux-2.6/xfs_super.c
Merge with /pub/scm/linux/kernel/git/torvalds/linux-2.6.git
[linux-2.6-omap-h63xx.git] / fs / xfs / linux-2.6 / xfs_super.c
1 /*
2  * Copyright (c) 2000-2005 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_dir.h"
27 #include "xfs_dir2.h"
28 #include "xfs_alloc.h"
29 #include "xfs_dmapi.h"
30 #include "xfs_quota.h"
31 #include "xfs_mount.h"
32 #include "xfs_bmap_btree.h"
33 #include "xfs_alloc_btree.h"
34 #include "xfs_ialloc_btree.h"
35 #include "xfs_dir_sf.h"
36 #include "xfs_dir2_sf.h"
37 #include "xfs_attr_sf.h"
38 #include "xfs_dinode.h"
39 #include "xfs_inode.h"
40 #include "xfs_btree.h"
41 #include "xfs_ialloc.h"
42 #include "xfs_bmap.h"
43 #include "xfs_rtalloc.h"
44 #include "xfs_error.h"
45 #include "xfs_itable.h"
46 #include "xfs_rw.h"
47 #include "xfs_acl.h"
48 #include "xfs_cap.h"
49 #include "xfs_mac.h"
50 #include "xfs_attr.h"
51 #include "xfs_buf_item.h"
52 #include "xfs_utils.h"
53 #include "xfs_version.h"
54
55 #include <linux/namei.h>
56 #include <linux/init.h>
57 #include <linux/mount.h>
58 #include <linux/mempool.h>
59 #include <linux/writeback.h>
60 #include <linux/kthread.h>
61
62 STATIC struct quotactl_ops xfs_quotactl_operations;
63 STATIC struct super_operations xfs_super_operations;
64 STATIC kmem_zone_t *xfs_vnode_zone;
65 STATIC kmem_zone_t *xfs_ioend_zone;
66 mempool_t *xfs_ioend_pool;
67
68 STATIC struct xfs_mount_args *
69 xfs_args_allocate(
70         struct super_block      *sb)
71 {
72         struct xfs_mount_args   *args;
73
74         args = kmem_zalloc(sizeof(struct xfs_mount_args), KM_SLEEP);
75         args->logbufs = args->logbufsize = -1;
76         strncpy(args->fsname, sb->s_id, MAXNAMELEN);
77
78         /* Copy the already-parsed mount(2) flags we're interested in */
79         if (sb->s_flags & MS_DIRSYNC)
80                 args->flags |= XFSMNT_DIRSYNC;
81         if (sb->s_flags & MS_SYNCHRONOUS)
82                 args->flags |= XFSMNT_WSYNC;
83
84         /* Default to 32 bit inodes on Linux all the time */
85         args->flags |= XFSMNT_32BITINODES;
86
87         return args;
88 }
89
90 __uint64_t
91 xfs_max_file_offset(
92         unsigned int            blockshift)
93 {
94         unsigned int            pagefactor = 1;
95         unsigned int            bitshift = BITS_PER_LONG - 1;
96
97         /* Figure out maximum filesize, on Linux this can depend on
98          * the filesystem blocksize (on 32 bit platforms).
99          * __block_prepare_write does this in an [unsigned] long...
100          *      page->index << (PAGE_CACHE_SHIFT - bbits)
101          * So, for page sized blocks (4K on 32 bit platforms),
102          * this wraps at around 8Tb (hence MAX_LFS_FILESIZE which is
103          *      (((u64)PAGE_CACHE_SIZE << (BITS_PER_LONG-1))-1)
104          * but for smaller blocksizes it is less (bbits = log2 bsize).
105          * Note1: get_block_t takes a long (implicit cast from above)
106          * Note2: The Large Block Device (LBD and HAVE_SECTOR_T) patch
107          * can optionally convert the [unsigned] long from above into
108          * an [unsigned] long long.
109          */
110
111 #if BITS_PER_LONG == 32
112 # if defined(CONFIG_LBD)
113         ASSERT(sizeof(sector_t) == 8);
114         pagefactor = PAGE_CACHE_SIZE;
115         bitshift = BITS_PER_LONG;
116 # else
117         pagefactor = PAGE_CACHE_SIZE >> (PAGE_CACHE_SHIFT - blockshift);
118 # endif
119 #endif
120
121         return (((__uint64_t)pagefactor) << bitshift) - 1;
122 }
123
124 STATIC __inline__ void
125 xfs_set_inodeops(
126         struct inode            *inode)
127 {
128         switch (inode->i_mode & S_IFMT) {
129         case S_IFREG:
130                 inode->i_op = &xfs_inode_operations;
131                 inode->i_fop = &xfs_file_operations;
132                 inode->i_mapping->a_ops = &xfs_address_space_operations;
133                 break;
134         case S_IFDIR:
135                 inode->i_op = &xfs_dir_inode_operations;
136                 inode->i_fop = &xfs_dir_file_operations;
137                 break;
138         case S_IFLNK:
139                 inode->i_op = &xfs_symlink_inode_operations;
140                 if (inode->i_blocks)
141                         inode->i_mapping->a_ops = &xfs_address_space_operations;
142                 break;
143         default:
144                 inode->i_op = &xfs_inode_operations;
145                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
146                 break;
147         }
148 }
149
150 STATIC __inline__ void
151 xfs_revalidate_inode(
152         xfs_mount_t             *mp,
153         vnode_t                 *vp,
154         xfs_inode_t             *ip)
155 {
156         struct inode            *inode = vn_to_inode(vp);
157
158         inode->i_mode   = ip->i_d.di_mode;
159         inode->i_nlink  = ip->i_d.di_nlink;
160         inode->i_uid    = ip->i_d.di_uid;
161         inode->i_gid    = ip->i_d.di_gid;
162
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_blksize = xfs_preferred_iosize(mp);
176         inode->i_generation = ip->i_d.di_gen;
177         i_size_write(inode, ip->i_d.di_size);
178         inode->i_blocks =
179                 XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
180         inode->i_atime.tv_sec   = ip->i_d.di_atime.t_sec;
181         inode->i_atime.tv_nsec  = ip->i_d.di_atime.t_nsec;
182         inode->i_mtime.tv_sec   = ip->i_d.di_mtime.t_sec;
183         inode->i_mtime.tv_nsec  = ip->i_d.di_mtime.t_nsec;
184         inode->i_ctime.tv_sec   = ip->i_d.di_ctime.t_sec;
185         inode->i_ctime.tv_nsec  = ip->i_d.di_ctime.t_nsec;
186         if (ip->i_d.di_flags & XFS_DIFLAG_IMMUTABLE)
187                 inode->i_flags |= S_IMMUTABLE;
188         else
189                 inode->i_flags &= ~S_IMMUTABLE;
190         if (ip->i_d.di_flags & XFS_DIFLAG_APPEND)
191                 inode->i_flags |= S_APPEND;
192         else
193                 inode->i_flags &= ~S_APPEND;
194         if (ip->i_d.di_flags & XFS_DIFLAG_SYNC)
195                 inode->i_flags |= S_SYNC;
196         else
197                 inode->i_flags &= ~S_SYNC;
198         if (ip->i_d.di_flags & XFS_DIFLAG_NOATIME)
199                 inode->i_flags |= S_NOATIME;
200         else
201                 inode->i_flags &= ~S_NOATIME;
202         vp->v_flag &= ~VMODIFIED;
203 }
204
205 void
206 xfs_initialize_vnode(
207         bhv_desc_t              *bdp,
208         vnode_t                 *vp,
209         bhv_desc_t              *inode_bhv,
210         int                     unlock)
211 {
212         xfs_inode_t             *ip = XFS_BHVTOI(inode_bhv);
213         struct inode            *inode = vn_to_inode(vp);
214
215         if (!inode_bhv->bd_vobj) {
216                 vp->v_vfsp = bhvtovfs(bdp);
217                 bhv_desc_init(inode_bhv, ip, vp, &xfs_vnodeops);
218                 bhv_insert(VN_BHV_HEAD(vp), inode_bhv);
219         }
220
221         /*
222          * We need to set the ops vectors, and unlock the inode, but if
223          * we have been called during the new inode create process, it is
224          * too early to fill in the Linux inode.  We will get called a
225          * second time once the inode is properly set up, and then we can
226          * finish our work.
227          */
228         if (ip->i_d.di_mode != 0 && unlock && (inode->i_state & I_NEW)) {
229                 xfs_revalidate_inode(XFS_BHVTOM(bdp), vp, ip);
230                 xfs_set_inodeops(inode);
231
232                 ip->i_flags &= ~XFS_INEW;
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         error = xfs_barrier_test(mp);
319         if (error) {
320                 xfs_fs_cmn_err(CE_NOTE, mp,
321                   "Disabling barriers, trial barrier write failed");
322                 mp->m_flags &= ~XFS_MOUNT_BARRIER;
323                 return;
324         }
325 }
326
327 void
328 xfs_blkdev_issue_flush(
329         xfs_buftarg_t           *buftarg)
330 {
331         blkdev_issue_flush(buftarg->bt_bdev, NULL);
332 }
333
334 STATIC struct inode *
335 xfs_fs_alloc_inode(
336         struct super_block      *sb)
337 {
338         vnode_t                 *vp;
339
340         vp = kmem_zone_alloc(xfs_vnode_zone, KM_SLEEP);
341         if (unlikely(!vp))
342                 return NULL;
343         return vn_to_inode(vp);
344 }
345
346 STATIC void
347 xfs_fs_destroy_inode(
348         struct inode            *inode)
349 {
350         kmem_zone_free(xfs_vnode_zone, vn_from_inode(inode));
351 }
352
353 STATIC void
354 xfs_fs_inode_init_once(
355         void                    *vnode,
356         kmem_zone_t             *zonep,
357         unsigned long           flags)
358 {
359         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
360                       SLAB_CTOR_CONSTRUCTOR)
361                 inode_init_once(vn_to_inode((vnode_t *)vnode));
362 }
363
364 STATIC int
365 xfs_init_zones(void)
366 {
367         xfs_vnode_zone = kmem_zone_init_flags(sizeof(vnode_t), "xfs_vnode_t",
368                                         KM_ZONE_HWALIGN | KM_ZONE_RECLAIM |
369                                         KM_ZONE_SPREAD,
370                                         xfs_fs_inode_init_once);
371         if (!xfs_vnode_zone)
372                 goto out;
373
374         xfs_ioend_zone = kmem_zone_init(sizeof(xfs_ioend_t), "xfs_ioend");
375         if (!xfs_ioend_zone)
376                 goto out_destroy_vnode_zone;
377
378         xfs_ioend_pool = mempool_create_slab_pool(4 * MAX_BUF_PER_PAGE,
379                                                   xfs_ioend_zone);
380         if (!xfs_ioend_pool)
381                 goto out_free_ioend_zone;
382         return 0;
383
384  out_free_ioend_zone:
385         kmem_zone_destroy(xfs_ioend_zone);
386  out_destroy_vnode_zone:
387         kmem_zone_destroy(xfs_vnode_zone);
388  out:
389         return -ENOMEM;
390 }
391
392 STATIC void
393 xfs_destroy_zones(void)
394 {
395         mempool_destroy(xfs_ioend_pool);
396         kmem_zone_destroy(xfs_vnode_zone);
397         kmem_zone_destroy(xfs_ioend_zone);
398 }
399
400 /*
401  * Attempt to flush the inode, this will actually fail
402  * if the inode is pinned, but we dirty the inode again
403  * at the point when it is unpinned after a log write,
404  * since this is when the inode itself becomes flushable.
405  */
406 STATIC int
407 xfs_fs_write_inode(
408         struct inode            *inode,
409         int                     sync)
410 {
411         vnode_t                 *vp = vn_from_inode(inode);
412         int                     error = 0, flags = FLUSH_INODE;
413
414         if (vp) {
415                 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
416                 if (sync)
417                         flags |= FLUSH_SYNC;
418                 VOP_IFLUSH(vp, flags, error);
419                 if (error == EAGAIN) {
420                         if (sync)
421                                 VOP_IFLUSH(vp, flags | FLUSH_LOG, error);
422                         else
423                                 error = 0;
424                 }
425         }
426
427         return -error;
428 }
429
430 STATIC void
431 xfs_fs_clear_inode(
432         struct inode            *inode)
433 {
434         vnode_t                 *vp = vn_from_inode(inode);
435         int                     error, cache;
436
437         vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
438
439         XFS_STATS_INC(vn_rele);
440         XFS_STATS_INC(vn_remove);
441         XFS_STATS_INC(vn_reclaim);
442         XFS_STATS_DEC(vn_active);
443
444         /*
445          * This can happen because xfs_iget_core calls xfs_idestroy if we
446          * find an inode with di_mode == 0 but without IGET_CREATE set.
447          */
448         if (vp->v_fbhv)
449                 VOP_INACTIVE(vp, NULL, cache);
450
451         VN_LOCK(vp);
452         vp->v_flag &= ~VMODIFIED;
453         VN_UNLOCK(vp, 0);
454
455         if (vp->v_fbhv) {
456                 VOP_RECLAIM(vp, error);
457                 if (error)
458                         panic("vn_purge: cannot reclaim");
459         }
460
461         ASSERT(vp->v_fbhv == NULL);
462
463 #ifdef XFS_VNODE_TRACE
464         ktrace_free(vp->v_trace);
465 #endif
466 }
467
468 /*
469  * Enqueue a work item to be picked up by the vfs xfssyncd thread.
470  * Doing this has two advantages:
471  * - It saves on stack space, which is tight in certain situations
472  * - It can be used (with care) as a mechanism to avoid deadlocks.
473  * Flushing while allocating in a full filesystem requires both.
474  */
475 STATIC void
476 xfs_syncd_queue_work(
477         struct vfs      *vfs,
478         void            *data,
479         void            (*syncer)(vfs_t *, void *))
480 {
481         vfs_sync_work_t *work;
482
483         work = kmem_alloc(sizeof(struct vfs_sync_work), KM_SLEEP);
484         INIT_LIST_HEAD(&work->w_list);
485         work->w_syncer = syncer;
486         work->w_data = data;
487         work->w_vfs = vfs;
488         spin_lock(&vfs->vfs_sync_lock);
489         list_add_tail(&work->w_list, &vfs->vfs_sync_list);
490         spin_unlock(&vfs->vfs_sync_lock);
491         wake_up_process(vfs->vfs_sync_task);
492 }
493
494 /*
495  * Flush delayed allocate data, attempting to free up reserved space
496  * from existing allocations.  At this point a new allocation attempt
497  * has failed with ENOSPC and we are in the process of scratching our
498  * heads, looking about for more room...
499  */
500 STATIC void
501 xfs_flush_inode_work(
502         vfs_t           *vfs,
503         void            *inode)
504 {
505         filemap_flush(((struct inode *)inode)->i_mapping);
506         iput((struct inode *)inode);
507 }
508
509 void
510 xfs_flush_inode(
511         xfs_inode_t     *ip)
512 {
513         struct inode    *inode = vn_to_inode(XFS_ITOV(ip));
514         struct vfs      *vfs = XFS_MTOVFS(ip->i_mount);
515
516         igrab(inode);
517         xfs_syncd_queue_work(vfs, inode, xfs_flush_inode_work);
518         delay(msecs_to_jiffies(500));
519 }
520
521 /*
522  * This is the "bigger hammer" version of xfs_flush_inode_work...
523  * (IOW, "If at first you don't succeed, use a Bigger Hammer").
524  */
525 STATIC void
526 xfs_flush_device_work(
527         vfs_t           *vfs,
528         void            *inode)
529 {
530         sync_blockdev(vfs->vfs_super->s_bdev);
531         iput((struct inode *)inode);
532 }
533
534 void
535 xfs_flush_device(
536         xfs_inode_t     *ip)
537 {
538         struct inode    *inode = vn_to_inode(XFS_ITOV(ip));
539         struct vfs      *vfs = XFS_MTOVFS(ip->i_mount);
540
541         igrab(inode);
542         xfs_syncd_queue_work(vfs, inode, xfs_flush_device_work);
543         delay(msecs_to_jiffies(500));
544         xfs_log_force(ip->i_mount, (xfs_lsn_t)0, XFS_LOG_FORCE|XFS_LOG_SYNC);
545 }
546
547 #define SYNCD_FLAGS     (SYNC_FSDATA|SYNC_BDFLUSH|SYNC_ATTR|SYNC_REFCACHE)
548 STATIC void
549 vfs_sync_worker(
550         vfs_t           *vfsp,
551         void            *unused)
552 {
553         int             error;
554
555         if (!(vfsp->vfs_flag & VFS_RDONLY))
556                 VFS_SYNC(vfsp, SYNCD_FLAGS, NULL, error);
557         vfsp->vfs_sync_seq++;
558         wmb();
559         wake_up(&vfsp->vfs_wait_single_sync_task);
560 }
561
562 STATIC int
563 xfssyncd(
564         void                    *arg)
565 {
566         long                    timeleft;
567         vfs_t                   *vfsp = (vfs_t *) arg;
568         struct vfs_sync_work    *work, *n;
569         LIST_HEAD               (tmp);
570
571         timeleft = xfs_syncd_centisecs * msecs_to_jiffies(10);
572         for (;;) {
573                 timeleft = schedule_timeout_interruptible(timeleft);
574                 /* swsusp */
575                 try_to_freeze();
576                 if (kthread_should_stop() && list_empty(&vfsp->vfs_sync_list))
577                         break;
578
579                 spin_lock(&vfsp->vfs_sync_lock);
580                 /*
581                  * We can get woken by laptop mode, to do a sync -
582                  * that's the (only!) case where the list would be
583                  * empty with time remaining.
584                  */
585                 if (!timeleft || list_empty(&vfsp->vfs_sync_list)) {
586                         if (!timeleft)
587                                 timeleft = xfs_syncd_centisecs *
588                                                         msecs_to_jiffies(10);
589                         INIT_LIST_HEAD(&vfsp->vfs_sync_work.w_list);
590                         list_add_tail(&vfsp->vfs_sync_work.w_list,
591                                         &vfsp->vfs_sync_list);
592                 }
593                 list_for_each_entry_safe(work, n, &vfsp->vfs_sync_list, w_list)
594                         list_move(&work->w_list, &tmp);
595                 spin_unlock(&vfsp->vfs_sync_lock);
596
597                 list_for_each_entry_safe(work, n, &tmp, w_list) {
598                         (*work->w_syncer)(vfsp, work->w_data);
599                         list_del(&work->w_list);
600                         if (work == &vfsp->vfs_sync_work)
601                                 continue;
602                         kmem_free(work, sizeof(struct vfs_sync_work));
603                 }
604         }
605
606         return 0;
607 }
608
609 STATIC int
610 xfs_fs_start_syncd(
611         vfs_t                   *vfsp)
612 {
613         vfsp->vfs_sync_work.w_syncer = vfs_sync_worker;
614         vfsp->vfs_sync_work.w_vfs = vfsp;
615         vfsp->vfs_sync_task = kthread_run(xfssyncd, vfsp, "xfssyncd");
616         if (IS_ERR(vfsp->vfs_sync_task))
617                 return -PTR_ERR(vfsp->vfs_sync_task);
618         return 0;
619 }
620
621 STATIC void
622 xfs_fs_stop_syncd(
623         vfs_t                   *vfsp)
624 {
625         kthread_stop(vfsp->vfs_sync_task);
626 }
627
628 STATIC void
629 xfs_fs_put_super(
630         struct super_block      *sb)
631 {
632         vfs_t                   *vfsp = vfs_from_sb(sb);
633         int                     error;
634
635         xfs_fs_stop_syncd(vfsp);
636         VFS_SYNC(vfsp, SYNC_ATTR|SYNC_DELWRI, NULL, error);
637         if (!error)
638                 VFS_UNMOUNT(vfsp, 0, NULL, error);
639         if (error) {
640                 printk("XFS unmount got error %d\n", error);
641                 printk("%s: vfsp/0x%p left dangling!\n", __FUNCTION__, vfsp);
642                 return;
643         }
644
645         vfs_deallocate(vfsp);
646 }
647
648 STATIC void
649 xfs_fs_write_super(
650         struct super_block      *sb)
651 {
652         vfs_t                   *vfsp = vfs_from_sb(sb);
653         int                     error;
654
655         if (sb->s_flags & MS_RDONLY) {
656                 sb->s_dirt = 0; /* paranoia */
657                 return;
658         }
659         /* Push the log and superblock a little */
660         VFS_SYNC(vfsp, SYNC_FSDATA, NULL, error);
661         sb->s_dirt = 0;
662 }
663
664 STATIC int
665 xfs_fs_sync_super(
666         struct super_block      *sb,
667         int                     wait)
668 {
669         vfs_t           *vfsp = vfs_from_sb(sb);
670         int             error;
671         int             flags = SYNC_FSDATA;
672
673         if (unlikely(sb->s_frozen == SB_FREEZE_WRITE))
674                 flags = SYNC_QUIESCE;
675         else
676                 flags = SYNC_FSDATA | (wait ? SYNC_WAIT : 0);
677
678         VFS_SYNC(vfsp, flags, NULL, error);
679         sb->s_dirt = 0;
680
681         if (unlikely(laptop_mode)) {
682                 int     prev_sync_seq = vfsp->vfs_sync_seq;
683
684                 /*
685                  * The disk must be active because we're syncing.
686                  * We schedule xfssyncd now (now that the disk is
687                  * active) instead of later (when it might not be).
688                  */
689                 wake_up_process(vfsp->vfs_sync_task);
690                 /*
691                  * We have to wait for the sync iteration to complete.
692                  * If we don't, the disk activity caused by the sync
693                  * will come after the sync is completed, and that
694                  * triggers another sync from laptop mode.
695                  */
696                 wait_event(vfsp->vfs_wait_single_sync_task,
697                                 vfsp->vfs_sync_seq != prev_sync_seq);
698         }
699
700         return -error;
701 }
702
703 STATIC int
704 xfs_fs_statfs(
705         struct super_block      *sb,
706         struct kstatfs          *statp)
707 {
708         vfs_t                   *vfsp = vfs_from_sb(sb);
709         int                     error;
710
711         VFS_STATVFS(vfsp, statp, NULL, error);
712         return -error;
713 }
714
715 STATIC int
716 xfs_fs_remount(
717         struct super_block      *sb,
718         int                     *flags,
719         char                    *options)
720 {
721         vfs_t                   *vfsp = vfs_from_sb(sb);
722         struct xfs_mount_args   *args = xfs_args_allocate(sb);
723         int                     error;
724
725         VFS_PARSEARGS(vfsp, options, args, 1, error);
726         if (!error)
727                 VFS_MNTUPDATE(vfsp, flags, args, error);
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         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         struct vfs              *vfsp = vfs_from_sb(mnt->mnt_sb);
745         int                     error;
746
747         VFS_SHOWARGS(vfsp, m, error);
748         return error;
749 }
750
751 STATIC int
752 xfs_fs_quotasync(
753         struct super_block      *sb,
754         int                     type)
755 {
756         struct vfs              *vfsp = vfs_from_sb(sb);
757         int                     error;
758
759         VFS_QUOTACTL(vfsp, Q_XQUOTASYNC, 0, (caddr_t)NULL, error);
760         return -error;
761 }
762
763 STATIC int
764 xfs_fs_getxstate(
765         struct super_block      *sb,
766         struct fs_quota_stat    *fqs)
767 {
768         struct vfs              *vfsp = vfs_from_sb(sb);
769         int                     error;
770
771         VFS_QUOTACTL(vfsp, Q_XGETQSTAT, 0, (caddr_t)fqs, error);
772         return -error;
773 }
774
775 STATIC int
776 xfs_fs_setxstate(
777         struct super_block      *sb,
778         unsigned int            flags,
779         int                     op)
780 {
781         struct vfs              *vfsp = vfs_from_sb(sb);
782         int                     error;
783
784         VFS_QUOTACTL(vfsp, op, 0, (caddr_t)&flags, error);
785         return -error;
786 }
787
788 STATIC int
789 xfs_fs_getxquota(
790         struct super_block      *sb,
791         int                     type,
792         qid_t                   id,
793         struct fs_disk_quota    *fdq)
794 {
795         struct vfs              *vfsp = vfs_from_sb(sb);
796         int                     error, getmode;
797
798         getmode = (type == USRQUOTA) ? Q_XGETQUOTA :
799                  ((type == GRPQUOTA) ? Q_XGETGQUOTA : Q_XGETPQUOTA);
800         VFS_QUOTACTL(vfsp, getmode, id, (caddr_t)fdq, error);
801         return -error;
802 }
803
804 STATIC int
805 xfs_fs_setxquota(
806         struct super_block      *sb,
807         int                     type,
808         qid_t                   id,
809         struct fs_disk_quota    *fdq)
810 {
811         struct vfs              *vfsp = vfs_from_sb(sb);
812         int                     error, setmode;
813
814         setmode = (type == USRQUOTA) ? Q_XSETQLIM :
815                  ((type == GRPQUOTA) ? Q_XSETGQLIM : Q_XSETPQLIM);
816         VFS_QUOTACTL(vfsp, setmode, id, (caddr_t)fdq, error);
817         return -error;
818 }
819
820 STATIC int
821 xfs_fs_fill_super(
822         struct super_block      *sb,
823         void                    *data,
824         int                     silent)
825 {
826         vnode_t                 *rootvp;
827         struct vfs              *vfsp = vfs_allocate(sb);
828         struct xfs_mount_args   *args = xfs_args_allocate(sb);
829         struct kstatfs          statvfs;
830         int                     error, error2;
831
832         bhv_insert_all_vfsops(vfsp);
833
834         VFS_PARSEARGS(vfsp, (char *)data, args, 0, error);
835         if (error) {
836                 bhv_remove_all_vfsops(vfsp, 1);
837                 goto fail_vfsop;
838         }
839
840         sb_min_blocksize(sb, BBSIZE);
841 #ifdef CONFIG_XFS_EXPORT
842         sb->s_export_op = &xfs_export_operations;
843 #endif
844         sb->s_qcop = &xfs_quotactl_operations;
845         sb->s_op = &xfs_super_operations;
846
847         VFS_MOUNT(vfsp, args, NULL, error);
848         if (error) {
849                 bhv_remove_all_vfsops(vfsp, 1);
850                 goto fail_vfsop;
851         }
852
853         VFS_STATVFS(vfsp, &statvfs, NULL, error);
854         if (error)
855                 goto fail_unmount;
856
857         sb->s_dirt = 1;
858         sb->s_magic = statvfs.f_type;
859         sb->s_blocksize = statvfs.f_bsize;
860         sb->s_blocksize_bits = ffs(statvfs.f_bsize) - 1;
861         sb->s_maxbytes = xfs_max_file_offset(sb->s_blocksize_bits);
862         sb->s_time_gran = 1;
863         set_posix_acl_flag(sb);
864
865         VFS_ROOT(vfsp, &rootvp, error);
866         if (error)
867                 goto fail_unmount;
868
869         sb->s_root = d_alloc_root(vn_to_inode(rootvp));
870         if (!sb->s_root) {
871                 error = ENOMEM;
872                 goto fail_vnrele;
873         }
874         if (is_bad_inode(sb->s_root->d_inode)) {
875                 error = EINVAL;
876                 goto fail_vnrele;
877         }
878         if ((error = xfs_fs_start_syncd(vfsp)))
879                 goto fail_vnrele;
880         vn_trace_exit(rootvp, __FUNCTION__, (inst_t *)__return_address);
881
882         kmem_free(args, sizeof(*args));
883         return 0;
884
885 fail_vnrele:
886         if (sb->s_root) {
887                 dput(sb->s_root);
888                 sb->s_root = NULL;
889         } else {
890                 VN_RELE(rootvp);
891         }
892
893 fail_unmount:
894         VFS_UNMOUNT(vfsp, 0, NULL, error2);
895
896 fail_vfsop:
897         vfs_deallocate(vfsp);
898         kmem_free(args, sizeof(*args));
899         return -error;
900 }
901
902 STATIC struct super_block *
903 xfs_fs_get_sb(
904         struct file_system_type *fs_type,
905         int                     flags,
906         const char              *dev_name,
907         void                    *data)
908 {
909         return get_sb_bdev(fs_type, flags, dev_name, data, xfs_fs_fill_super);
910 }
911
912 STATIC struct super_operations xfs_super_operations = {
913         .alloc_inode            = xfs_fs_alloc_inode,
914         .destroy_inode          = xfs_fs_destroy_inode,
915         .write_inode            = xfs_fs_write_inode,
916         .clear_inode            = xfs_fs_clear_inode,
917         .put_super              = xfs_fs_put_super,
918         .write_super            = xfs_fs_write_super,
919         .sync_fs                = xfs_fs_sync_super,
920         .write_super_lockfs     = xfs_fs_lockfs,
921         .statfs                 = xfs_fs_statfs,
922         .remount_fs             = xfs_fs_remount,
923         .show_options           = xfs_fs_show_options,
924 };
925
926 STATIC struct quotactl_ops xfs_quotactl_operations = {
927         .quota_sync             = xfs_fs_quotasync,
928         .get_xstate             = xfs_fs_getxstate,
929         .set_xstate             = xfs_fs_setxstate,
930         .get_xquota             = xfs_fs_getxquota,
931         .set_xquota             = xfs_fs_setxquota,
932 };
933
934 STATIC struct file_system_type xfs_fs_type = {
935         .owner                  = THIS_MODULE,
936         .name                   = "xfs",
937         .get_sb                 = xfs_fs_get_sb,
938         .kill_sb                = kill_block_super,
939         .fs_flags               = FS_REQUIRES_DEV,
940 };
941
942
943 STATIC int __init
944 init_xfs_fs( void )
945 {
946         int                     error;
947         struct sysinfo          si;
948         static char             message[] __initdata = KERN_INFO \
949                 XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled\n";
950
951         printk(message);
952
953         si_meminfo(&si);
954         xfs_physmem = si.totalram;
955
956         ktrace_init(64);
957
958         error = xfs_init_zones();
959         if (error < 0)
960                 goto undo_zones;
961
962         error = xfs_buf_init();
963         if (error < 0)
964                 goto undo_buffers;
965
966         vn_init();
967         xfs_init();
968         uuid_init();
969         vfs_initquota();
970
971         error = register_filesystem(&xfs_fs_type);
972         if (error)
973                 goto undo_register;
974         return 0;
975
976 undo_register:
977         xfs_buf_terminate();
978
979 undo_buffers:
980         xfs_destroy_zones();
981
982 undo_zones:
983         return error;
984 }
985
986 STATIC void __exit
987 exit_xfs_fs( void )
988 {
989         vfs_exitquota();
990         unregister_filesystem(&xfs_fs_type);
991         xfs_cleanup();
992         xfs_buf_terminate();
993         xfs_destroy_zones();
994         ktrace_uninit();
995 }
996
997 module_init(init_xfs_fs);
998 module_exit(exit_xfs_fs);
999
1000 MODULE_AUTHOR("Silicon Graphics, Inc.");
1001 MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled");
1002 MODULE_LICENSE("GPL");