[XFS] Put back the QUEUE_ORDERED_NONE test in the barrier check.
[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_vfsops.h"
51 #include "xfs_version.h"
52
53 #include <linux/namei.h>
54 #include <linux/init.h>
55 #include <linux/mount.h>
56 #include <linux/mempool.h>
57 #include <linux/writeback.h>
58 #include <linux/kthread.h>
59 #include <linux/freezer.h>
60
61 static struct quotactl_ops xfs_quotactl_operations;
62 static struct super_operations xfs_super_operations;
63 static kmem_zone_t *xfs_vnode_zone;
64 static kmem_zone_t *xfs_ioend_zone;
65 mempool_t *xfs_ioend_pool;
66
67 STATIC struct xfs_mount_args *
68 xfs_args_allocate(
69         struct super_block      *sb,
70         int                     silent)
71 {
72         struct xfs_mount_args   *args;
73
74         args = kmem_zalloc(sizeof(struct xfs_mount_args), KM_SLEEP);
75         args->logbufs = args->logbufsize = -1;
76         strncpy(args->fsname, sb->s_id, MAXNAMELEN);
77
78         /* Copy the already-parsed mount(2) flags we're interested in */
79         if (sb->s_flags & MS_DIRSYNC)
80                 args->flags |= XFSMNT_DIRSYNC;
81         if (sb->s_flags & MS_SYNCHRONOUS)
82                 args->flags |= XFSMNT_WSYNC;
83         if (silent)
84                 args->flags |= XFSMNT_QUIET;
85         args->flags |= XFSMNT_32BITINODES;
86
87         return args;
88 }
89
90 __uint64_t
91 xfs_max_file_offset(
92         unsigned int            blockshift)
93 {
94         unsigned int            pagefactor = 1;
95         unsigned int            bitshift = BITS_PER_LONG - 1;
96
97         /* Figure out maximum filesize, on Linux this can depend on
98          * the filesystem blocksize (on 32 bit platforms).
99          * __block_prepare_write does this in an [unsigned] long...
100          *      page->index << (PAGE_CACHE_SHIFT - bbits)
101          * So, for page sized blocks (4K on 32 bit platforms),
102          * this wraps at around 8Tb (hence MAX_LFS_FILESIZE which is
103          *      (((u64)PAGE_CACHE_SIZE << (BITS_PER_LONG-1))-1)
104          * but for smaller blocksizes it is less (bbits = log2 bsize).
105          * Note1: get_block_t takes a long (implicit cast from above)
106          * Note2: The Large Block Device (LBD and HAVE_SECTOR_T) patch
107          * can optionally convert the [unsigned] long from above into
108          * an [unsigned] long long.
109          */
110
111 #if BITS_PER_LONG == 32
112 # if defined(CONFIG_LBD)
113         ASSERT(sizeof(sector_t) == 8);
114         pagefactor = PAGE_CACHE_SIZE;
115         bitshift = BITS_PER_LONG;
116 # else
117         pagefactor = PAGE_CACHE_SIZE >> (PAGE_CACHE_SHIFT - blockshift);
118 # endif
119 #endif
120
121         return (((__uint64_t)pagefactor) << bitshift) - 1;
122 }
123
124 STATIC_INLINE void
125 xfs_set_inodeops(
126         struct inode            *inode)
127 {
128         switch (inode->i_mode & S_IFMT) {
129         case S_IFREG:
130                 inode->i_op = &xfs_inode_operations;
131                 inode->i_fop = &xfs_file_operations;
132                 inode->i_mapping->a_ops = &xfs_address_space_operations;
133                 break;
134         case S_IFDIR:
135                 inode->i_op = &xfs_dir_inode_operations;
136                 inode->i_fop = &xfs_dir_file_operations;
137                 break;
138         case S_IFLNK:
139                 inode->i_op = &xfs_symlink_inode_operations;
140                 if (inode->i_blocks)
141                         inode->i_mapping->a_ops = &xfs_address_space_operations;
142                 break;
143         default:
144                 inode->i_op = &xfs_inode_operations;
145                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
146                 break;
147         }
148 }
149
150 STATIC_INLINE void
151 xfs_revalidate_inode(
152         xfs_mount_t             *mp,
153         bhv_vnode_t             *vp,
154         xfs_inode_t             *ip)
155 {
156         struct inode            *inode = vn_to_inode(vp);
157
158         inode->i_mode   = ip->i_d.di_mode;
159         inode->i_nlink  = ip->i_d.di_nlink;
160         inode->i_uid    = ip->i_d.di_uid;
161         inode->i_gid    = ip->i_d.di_gid;
162
163         switch (inode->i_mode & S_IFMT) {
164         case S_IFBLK:
165         case S_IFCHR:
166                 inode->i_rdev =
167                         MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff,
168                               sysv_minor(ip->i_df.if_u2.if_rdev));
169                 break;
170         default:
171                 inode->i_rdev = 0;
172                 break;
173         }
174
175         inode->i_generation = ip->i_d.di_gen;
176         i_size_write(inode, ip->i_d.di_size);
177         inode->i_blocks =
178                 XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
179         inode->i_atime.tv_sec   = ip->i_d.di_atime.t_sec;
180         inode->i_atime.tv_nsec  = ip->i_d.di_atime.t_nsec;
181         inode->i_mtime.tv_sec   = ip->i_d.di_mtime.t_sec;
182         inode->i_mtime.tv_nsec  = ip->i_d.di_mtime.t_nsec;
183         inode->i_ctime.tv_sec   = ip->i_d.di_ctime.t_sec;
184         inode->i_ctime.tv_nsec  = ip->i_d.di_ctime.t_nsec;
185         if (ip->i_d.di_flags & XFS_DIFLAG_IMMUTABLE)
186                 inode->i_flags |= S_IMMUTABLE;
187         else
188                 inode->i_flags &= ~S_IMMUTABLE;
189         if (ip->i_d.di_flags & XFS_DIFLAG_APPEND)
190                 inode->i_flags |= S_APPEND;
191         else
192                 inode->i_flags &= ~S_APPEND;
193         if (ip->i_d.di_flags & XFS_DIFLAG_SYNC)
194                 inode->i_flags |= S_SYNC;
195         else
196                 inode->i_flags &= ~S_SYNC;
197         if (ip->i_d.di_flags & XFS_DIFLAG_NOATIME)
198                 inode->i_flags |= S_NOATIME;
199         else
200                 inode->i_flags &= ~S_NOATIME;
201         xfs_iflags_clear(ip, XFS_IMODIFIED);
202 }
203
204 void
205 xfs_initialize_vnode(
206         struct xfs_mount        *mp,
207         bhv_vnode_t             *vp,
208         struct xfs_inode        *ip)
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 && (inode->i_state & I_NEW)) {
225                 xfs_revalidate_inode(mp, 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 (mp->m_ddev_targp->bt_bdev->bd_disk->queue->ordered ==
307                                         QUEUE_ORDERED_NONE) {
308                 xfs_fs_cmn_err(CE_NOTE, mp,
309                   "Disabling barriers, not supported by the underlying device");
310                 mp->m_flags &= ~XFS_MOUNT_BARRIER;
311                 return;
312         }
313
314         if (xfs_readonly_buftarg(mp->m_ddev_targp)) {
315                 xfs_fs_cmn_err(CE_NOTE, mp,
316                   "Disabling barriers, underlying device is readonly");
317                 mp->m_flags &= ~XFS_MOUNT_BARRIER;
318                 return;
319         }
320
321         error = xfs_barrier_test(mp);
322         if (error) {
323                 xfs_fs_cmn_err(CE_NOTE, mp,
324                   "Disabling barriers, trial barrier write failed");
325                 mp->m_flags &= ~XFS_MOUNT_BARRIER;
326                 return;
327         }
328 }
329
330 void
331 xfs_blkdev_issue_flush(
332         xfs_buftarg_t           *buftarg)
333 {
334         blkdev_issue_flush(buftarg->bt_bdev, NULL);
335 }
336
337 STATIC struct inode *
338 xfs_fs_alloc_inode(
339         struct super_block      *sb)
340 {
341         bhv_vnode_t             *vp;
342
343         vp = kmem_zone_alloc(xfs_vnode_zone, KM_SLEEP);
344         if (unlikely(!vp))
345                 return NULL;
346         return vn_to_inode(vp);
347 }
348
349 STATIC void
350 xfs_fs_destroy_inode(
351         struct inode            *inode)
352 {
353         kmem_zone_free(xfs_vnode_zone, vn_from_inode(inode));
354 }
355
356 STATIC void
357 xfs_fs_inode_init_once(
358         void                    *vnode,
359         kmem_zone_t             *zonep,
360         unsigned long           flags)
361 {
362         inode_init_once(vn_to_inode((bhv_vnode_t *)vnode));
363 }
364
365 STATIC int
366 xfs_init_zones(void)
367 {
368         xfs_vnode_zone = kmem_zone_init_flags(sizeof(bhv_vnode_t), "xfs_vnode",
369                                         KM_ZONE_HWALIGN | KM_ZONE_RECLAIM |
370                                         KM_ZONE_SPREAD,
371                                         xfs_fs_inode_init_once);
372         if (!xfs_vnode_zone)
373                 goto out;
374
375         xfs_ioend_zone = kmem_zone_init(sizeof(xfs_ioend_t), "xfs_ioend");
376         if (!xfs_ioend_zone)
377                 goto out_destroy_vnode_zone;
378
379         xfs_ioend_pool = mempool_create_slab_pool(4 * MAX_BUF_PER_PAGE,
380                                                   xfs_ioend_zone);
381         if (!xfs_ioend_pool)
382                 goto out_free_ioend_zone;
383         return 0;
384
385  out_free_ioend_zone:
386         kmem_zone_destroy(xfs_ioend_zone);
387  out_destroy_vnode_zone:
388         kmem_zone_destroy(xfs_vnode_zone);
389  out:
390         return -ENOMEM;
391 }
392
393 STATIC void
394 xfs_destroy_zones(void)
395 {
396         mempool_destroy(xfs_ioend_pool);
397         kmem_zone_destroy(xfs_vnode_zone);
398         kmem_zone_destroy(xfs_ioend_zone);
399 }
400
401 /*
402  * Attempt to flush the inode, this will actually fail
403  * if the inode is pinned, but we dirty the inode again
404  * at the point when it is unpinned after a log write,
405  * since this is when the inode itself becomes flushable.
406  */
407 STATIC int
408 xfs_fs_write_inode(
409         struct inode            *inode,
410         int                     sync)
411 {
412         int                     error = 0, flags = FLUSH_INODE;
413
414         vn_trace_entry(XFS_I(inode), __FUNCTION__,
415                         (inst_t *)__return_address);
416         if (sync) {
417                 filemap_fdatawait(inode->i_mapping);
418                 flags |= FLUSH_SYNC;
419         }
420         error = xfs_inode_flush(XFS_I(inode), flags);
421         /*
422          * if we failed to write out the inode then mark
423          * it dirty again so we'll try again later.
424          */
425         if (error)
426                 mark_inode_dirty_sync(inode);
427
428         return -error;
429 }
430
431 STATIC void
432 xfs_fs_clear_inode(
433         struct inode            *inode)
434 {
435         xfs_inode_t             *ip = XFS_I(inode);
436
437         /*
438          * ip can be null when xfs_iget_core calls xfs_idestroy if we
439          * find an inode with di_mode == 0 but without IGET_CREATE set.
440          */
441         if (ip) {
442                 vn_trace_entry(ip, __FUNCTION__, (inst_t *)__return_address);
443
444                 XFS_STATS_INC(vn_rele);
445                 XFS_STATS_INC(vn_remove);
446                 XFS_STATS_INC(vn_reclaim);
447                 XFS_STATS_DEC(vn_active);
448
449                 xfs_inactive(ip);
450                 xfs_iflags_clear(ip, XFS_IMODIFIED);
451                 if (xfs_reclaim(ip))
452                         panic("%s: cannot reclaim 0x%p\n", __FUNCTION__, inode);
453         }
454
455         ASSERT(XFS_I(inode) == NULL);
456 }
457
458 /*
459  * Enqueue a work item to be picked up by the vfs xfssyncd thread.
460  * Doing this has two advantages:
461  * - It saves on stack space, which is tight in certain situations
462  * - It can be used (with care) as a mechanism to avoid deadlocks.
463  * Flushing while allocating in a full filesystem requires both.
464  */
465 STATIC void
466 xfs_syncd_queue_work(
467         struct xfs_mount *mp,
468         void            *data,
469         void            (*syncer)(struct xfs_mount *, void *))
470 {
471         struct bhv_vfs_sync_work *work;
472
473         work = kmem_alloc(sizeof(struct bhv_vfs_sync_work), KM_SLEEP);
474         INIT_LIST_HEAD(&work->w_list);
475         work->w_syncer = syncer;
476         work->w_data = data;
477         work->w_mount = mp;
478         spin_lock(&mp->m_sync_lock);
479         list_add_tail(&work->w_list, &mp->m_sync_list);
480         spin_unlock(&mp->m_sync_lock);
481         wake_up_process(mp->m_sync_task);
482 }
483
484 /*
485  * Flush delayed allocate data, attempting to free up reserved space
486  * from existing allocations.  At this point a new allocation attempt
487  * has failed with ENOSPC and we are in the process of scratching our
488  * heads, looking about for more room...
489  */
490 STATIC void
491 xfs_flush_inode_work(
492         struct xfs_mount *mp,
493         void            *arg)
494 {
495         struct inode    *inode = arg;
496         filemap_flush(inode->i_mapping);
497         iput(inode);
498 }
499
500 void
501 xfs_flush_inode(
502         xfs_inode_t     *ip)
503 {
504         struct inode    *inode = ip->i_vnode;
505
506         igrab(inode);
507         xfs_syncd_queue_work(ip->i_mount, inode, xfs_flush_inode_work);
508         delay(msecs_to_jiffies(500));
509 }
510
511 /*
512  * This is the "bigger hammer" version of xfs_flush_inode_work...
513  * (IOW, "If at first you don't succeed, use a Bigger Hammer").
514  */
515 STATIC void
516 xfs_flush_device_work(
517         struct xfs_mount *mp,
518         void            *arg)
519 {
520         struct inode    *inode = arg;
521         sync_blockdev(mp->m_super->s_bdev);
522         iput(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
531         igrab(inode);
532         xfs_syncd_queue_work(ip->i_mount, inode, xfs_flush_device_work);
533         delay(msecs_to_jiffies(500));
534         xfs_log_force(ip->i_mount, (xfs_lsn_t)0, XFS_LOG_FORCE|XFS_LOG_SYNC);
535 }
536
537 STATIC void
538 xfs_sync_worker(
539         struct xfs_mount *mp,
540         void            *unused)
541 {
542         int             error;
543
544         if (!(mp->m_flags & XFS_MOUNT_RDONLY))
545                 error = xfs_sync(mp, SYNC_FSDATA | SYNC_BDFLUSH | SYNC_ATTR |
546                                      SYNC_REFCACHE | SYNC_SUPER);
547         mp->m_sync_seq++;
548         wake_up(&mp->m_wait_single_sync_task);
549 }
550
551 STATIC int
552 xfssyncd(
553         void                    *arg)
554 {
555         struct xfs_mount        *mp = arg;
556         long                    timeleft;
557         bhv_vfs_sync_work_t     *work, *n;
558         LIST_HEAD               (tmp);
559
560         set_freezable();
561         timeleft = xfs_syncd_centisecs * msecs_to_jiffies(10);
562         for (;;) {
563                 timeleft = schedule_timeout_interruptible(timeleft);
564                 /* swsusp */
565                 try_to_freeze();
566                 if (kthread_should_stop() && list_empty(&mp->m_sync_list))
567                         break;
568
569                 spin_lock(&mp->m_sync_lock);
570                 /*
571                  * We can get woken by laptop mode, to do a sync -
572                  * that's the (only!) case where the list would be
573                  * empty with time remaining.
574                  */
575                 if (!timeleft || list_empty(&mp->m_sync_list)) {
576                         if (!timeleft)
577                                 timeleft = xfs_syncd_centisecs *
578                                                         msecs_to_jiffies(10);
579                         INIT_LIST_HEAD(&mp->m_sync_work.w_list);
580                         list_add_tail(&mp->m_sync_work.w_list,
581                                         &mp->m_sync_list);
582                 }
583                 list_for_each_entry_safe(work, n, &mp->m_sync_list, w_list)
584                         list_move(&work->w_list, &tmp);
585                 spin_unlock(&mp->m_sync_lock);
586
587                 list_for_each_entry_safe(work, n, &tmp, w_list) {
588                         (*work->w_syncer)(mp, work->w_data);
589                         list_del(&work->w_list);
590                         if (work == &mp->m_sync_work)
591                                 continue;
592                         kmem_free(work, sizeof(struct bhv_vfs_sync_work));
593                 }
594         }
595
596         return 0;
597 }
598
599 STATIC void
600 xfs_fs_put_super(
601         struct super_block      *sb)
602 {
603         struct xfs_mount        *mp = XFS_M(sb);
604         int                     error;
605
606         kthread_stop(mp->m_sync_task);
607
608         xfs_sync(mp, SYNC_ATTR | SYNC_DELWRI);
609         error = xfs_unmount(mp, 0, NULL);
610         if (error)
611                 printk("XFS: unmount got error=%d\n", error);
612 }
613
614 STATIC void
615 xfs_fs_write_super(
616         struct super_block      *sb)
617 {
618         if (!(sb->s_flags & MS_RDONLY))
619                 xfs_sync(XFS_M(sb), SYNC_FSDATA);
620         sb->s_dirt = 0;
621 }
622
623 STATIC int
624 xfs_fs_sync_super(
625         struct super_block      *sb,
626         int                     wait)
627 {
628         struct xfs_mount        *mp = XFS_M(sb);
629         int                     error;
630         int                     flags;
631
632         /*
633          * Treat a sync operation like a freeze.  This is to work
634          * around a race in sync_inodes() which works in two phases
635          * - an asynchronous flush, which can write out an inode
636          * without waiting for file size updates to complete, and a
637          * synchronous flush, which wont do anything because the
638          * async flush removed the inode's dirty flag.  Also
639          * sync_inodes() will not see any files that just have
640          * outstanding transactions to be flushed because we don't
641          * dirty the Linux inode until after the transaction I/O
642          * completes.
643          */
644         if (wait || unlikely(sb->s_frozen == SB_FREEZE_WRITE)) {
645                 /*
646                  * First stage of freeze - no more writers will make progress
647                  * now we are here, so we flush delwri and delalloc buffers
648                  * here, then wait for all I/O to complete.  Data is frozen at
649                  * that point. Metadata is not frozen, transactions can still
650                  * occur here so don't bother flushing the buftarg (i.e
651                  * SYNC_QUIESCE) because it'll just get dirty again.
652                  */
653                 flags = SYNC_DATA_QUIESCE;
654         } else
655                 flags = SYNC_FSDATA;
656
657         error = xfs_sync(mp, flags);
658         sb->s_dirt = 0;
659
660         if (unlikely(laptop_mode)) {
661                 int     prev_sync_seq = mp->m_sync_seq;
662
663                 /*
664                  * The disk must be active because we're syncing.
665                  * We schedule xfssyncd now (now that the disk is
666                  * active) instead of later (when it might not be).
667                  */
668                 wake_up_process(mp->m_sync_task);
669                 /*
670                  * We have to wait for the sync iteration to complete.
671                  * If we don't, the disk activity caused by the sync
672                  * will come after the sync is completed, and that
673                  * triggers another sync from laptop mode.
674                  */
675                 wait_event(mp->m_wait_single_sync_task,
676                                 mp->m_sync_seq != prev_sync_seq);
677         }
678
679         return -error;
680 }
681
682 STATIC int
683 xfs_fs_statfs(
684         struct dentry           *dentry,
685         struct kstatfs          *statp)
686 {
687         return -xfs_statvfs(XFS_M(dentry->d_sb), statp,
688                                 vn_from_inode(dentry->d_inode));
689 }
690
691 STATIC int
692 xfs_fs_remount(
693         struct super_block      *sb,
694         int                     *flags,
695         char                    *options)
696 {
697         struct xfs_mount        *mp = XFS_M(sb);
698         struct xfs_mount_args   *args = xfs_args_allocate(sb, 0);
699         int                     error;
700
701         error = xfs_parseargs(mp, options, args, 1);
702         if (!error)
703                 error = xfs_mntupdate(mp, flags, args);
704         kmem_free(args, sizeof(*args));
705         return -error;
706 }
707
708 STATIC void
709 xfs_fs_lockfs(
710         struct super_block      *sb)
711 {
712         xfs_freeze(XFS_M(sb));
713 }
714
715 STATIC int
716 xfs_fs_show_options(
717         struct seq_file         *m,
718         struct vfsmount         *mnt)
719 {
720         return -xfs_showargs(XFS_M(mnt->mnt_sb), m);
721 }
722
723 STATIC int
724 xfs_fs_quotasync(
725         struct super_block      *sb,
726         int                     type)
727 {
728         return -XFS_QM_QUOTACTL(XFS_M(sb), Q_XQUOTASYNC, 0, NULL);
729 }
730
731 STATIC int
732 xfs_fs_getxstate(
733         struct super_block      *sb,
734         struct fs_quota_stat    *fqs)
735 {
736         return -XFS_QM_QUOTACTL(XFS_M(sb), Q_XGETQSTAT, 0, (caddr_t)fqs);
737 }
738
739 STATIC int
740 xfs_fs_setxstate(
741         struct super_block      *sb,
742         unsigned int            flags,
743         int                     op)
744 {
745         return -XFS_QM_QUOTACTL(XFS_M(sb), op, 0, (caddr_t)&flags);
746 }
747
748 STATIC int
749 xfs_fs_getxquota(
750         struct super_block      *sb,
751         int                     type,
752         qid_t                   id,
753         struct fs_disk_quota    *fdq)
754 {
755         return -XFS_QM_QUOTACTL(XFS_M(sb),
756                                  (type == USRQUOTA) ? Q_XGETQUOTA :
757                                   ((type == GRPQUOTA) ? Q_XGETGQUOTA :
758                                    Q_XGETPQUOTA), id, (caddr_t)fdq);
759 }
760
761 STATIC int
762 xfs_fs_setxquota(
763         struct super_block      *sb,
764         int                     type,
765         qid_t                   id,
766         struct fs_disk_quota    *fdq)
767 {
768         return -XFS_QM_QUOTACTL(XFS_M(sb),
769                                  (type == USRQUOTA) ? Q_XSETQLIM :
770                                   ((type == GRPQUOTA) ? Q_XSETGQLIM :
771                                    Q_XSETPQLIM), id, (caddr_t)fdq);
772 }
773
774 STATIC int
775 xfs_fs_fill_super(
776         struct super_block      *sb,
777         void                    *data,
778         int                     silent)
779 {
780         struct inode            *rootvp;
781         struct xfs_mount        *mp = NULL;
782         struct xfs_mount_args   *args = xfs_args_allocate(sb, silent);
783         struct kstatfs          statvfs;
784         int                     error;
785
786         mp = xfs_mount_init();
787
788         INIT_LIST_HEAD(&mp->m_sync_list);
789         spin_lock_init(&mp->m_sync_lock);
790         init_waitqueue_head(&mp->m_wait_single_sync_task);
791
792         mp->m_super = sb;
793         sb->s_fs_info = mp;
794
795         if (sb->s_flags & MS_RDONLY)
796                 mp->m_flags |= XFS_MOUNT_RDONLY;
797
798         error = xfs_parseargs(mp, (char *)data, args, 0);
799         if (error)
800                 goto fail_vfsop;
801
802         sb_min_blocksize(sb, BBSIZE);
803         sb->s_export_op = &xfs_export_operations;
804         sb->s_qcop = &xfs_quotactl_operations;
805         sb->s_op = &xfs_super_operations;
806
807         error = xfs_mount(mp, args, NULL);
808         if (error)
809                 goto fail_vfsop;
810
811         error = xfs_statvfs(mp, &statvfs, NULL);
812         if (error)
813                 goto fail_unmount;
814
815         sb->s_dirt = 1;
816         sb->s_magic = statvfs.f_type;
817         sb->s_blocksize = statvfs.f_bsize;
818         sb->s_blocksize_bits = ffs(statvfs.f_bsize) - 1;
819         sb->s_maxbytes = xfs_max_file_offset(sb->s_blocksize_bits);
820         sb->s_time_gran = 1;
821         set_posix_acl_flag(sb);
822
823         error = xfs_root(mp, &rootvp);
824         if (error)
825                 goto fail_unmount;
826
827         sb->s_root = d_alloc_root(vn_to_inode(rootvp));
828         if (!sb->s_root) {
829                 error = ENOMEM;
830                 goto fail_vnrele;
831         }
832         if (is_bad_inode(sb->s_root->d_inode)) {
833                 error = EINVAL;
834                 goto fail_vnrele;
835         }
836
837         mp->m_sync_work.w_syncer = xfs_sync_worker;
838         mp->m_sync_work.w_mount = mp;
839         mp->m_sync_task = kthread_run(xfssyncd, mp, "xfssyncd");
840         if (IS_ERR(mp->m_sync_task)) {
841                 error = -PTR_ERR(mp->m_sync_task);
842                 goto fail_vnrele;
843         }
844
845         vn_trace_exit(XFS_I(sb->s_root->d_inode), __FUNCTION__,
846                         (inst_t *)__return_address);
847
848         kmem_free(args, sizeof(*args));
849         return 0;
850
851 fail_vnrele:
852         if (sb->s_root) {
853                 dput(sb->s_root);
854                 sb->s_root = NULL;
855         } else {
856                 VN_RELE(rootvp);
857         }
858
859 fail_unmount:
860         xfs_unmount(mp, 0, NULL);
861
862 fail_vfsop:
863         kmem_free(args, sizeof(*args));
864         return -error;
865 }
866
867 STATIC int
868 xfs_fs_get_sb(
869         struct file_system_type *fs_type,
870         int                     flags,
871         const char              *dev_name,
872         void                    *data,
873         struct vfsmount         *mnt)
874 {
875         return get_sb_bdev(fs_type, flags, dev_name, data, xfs_fs_fill_super,
876                            mnt);
877 }
878
879 static struct super_operations xfs_super_operations = {
880         .alloc_inode            = xfs_fs_alloc_inode,
881         .destroy_inode          = xfs_fs_destroy_inode,
882         .write_inode            = xfs_fs_write_inode,
883         .clear_inode            = xfs_fs_clear_inode,
884         .put_super              = xfs_fs_put_super,
885         .write_super            = xfs_fs_write_super,
886         .sync_fs                = xfs_fs_sync_super,
887         .write_super_lockfs     = xfs_fs_lockfs,
888         .statfs                 = xfs_fs_statfs,
889         .remount_fs             = xfs_fs_remount,
890         .show_options           = xfs_fs_show_options,
891 };
892
893 static struct quotactl_ops xfs_quotactl_operations = {
894         .quota_sync             = xfs_fs_quotasync,
895         .get_xstate             = xfs_fs_getxstate,
896         .set_xstate             = xfs_fs_setxstate,
897         .get_xquota             = xfs_fs_getxquota,
898         .set_xquota             = xfs_fs_setxquota,
899 };
900
901 static struct file_system_type xfs_fs_type = {
902         .owner                  = THIS_MODULE,
903         .name                   = "xfs",
904         .get_sb                 = xfs_fs_get_sb,
905         .kill_sb                = kill_block_super,
906         .fs_flags               = FS_REQUIRES_DEV,
907 };
908
909
910 STATIC int __init
911 init_xfs_fs( void )
912 {
913         int                     error;
914         static char             message[] __initdata = KERN_INFO \
915                 XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled\n";
916
917         printk(message);
918
919         ktrace_init(64);
920
921         error = xfs_init_zones();
922         if (error < 0)
923                 goto undo_zones;
924
925         error = xfs_buf_init();
926         if (error < 0)
927                 goto undo_buffers;
928
929         vn_init();
930         xfs_init();
931         uuid_init();
932         vfs_initquota();
933
934         error = register_filesystem(&xfs_fs_type);
935         if (error)
936                 goto undo_register;
937         return 0;
938
939 undo_register:
940         xfs_buf_terminate();
941
942 undo_buffers:
943         xfs_destroy_zones();
944
945 undo_zones:
946         return error;
947 }
948
949 STATIC void __exit
950 exit_xfs_fs( void )
951 {
952         vfs_exitquota();
953         unregister_filesystem(&xfs_fs_type);
954         xfs_cleanup();
955         xfs_buf_terminate();
956         xfs_destroy_zones();
957         ktrace_uninit();
958 }
959
960 module_init(init_xfs_fs);
961 module_exit(exit_xfs_fs);
962
963 MODULE_AUTHOR("Silicon Graphics, Inc.");
964 MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled");
965 MODULE_LICENSE("GPL");