[PATCH] cleanup in proc_check_chroot()
[linux-2.6] / fs / xfs / xfs_vfsops.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_fs.h"
20 #include "xfs_types.h"
21 #include "xfs_bit.h"
22 #include "xfs_log.h"
23 #include "xfs_inum.h"
24 #include "xfs_trans.h"
25 #include "xfs_sb.h"
26 #include "xfs_ag.h"
27 #include "xfs_dir.h"
28 #include "xfs_dir2.h"
29 #include "xfs_dmapi.h"
30 #include "xfs_mount.h"
31 #include "xfs_da_btree.h"
32 #include "xfs_bmap_btree.h"
33 #include "xfs_ialloc_btree.h"
34 #include "xfs_alloc_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_inode_item.h"
41 #include "xfs_btree.h"
42 #include "xfs_alloc.h"
43 #include "xfs_ialloc.h"
44 #include "xfs_quota.h"
45 #include "xfs_error.h"
46 #include "xfs_bmap.h"
47 #include "xfs_rw.h"
48 #include "xfs_refcache.h"
49 #include "xfs_buf_item.h"
50 #include "xfs_log_priv.h"
51 #include "xfs_dir2_trace.h"
52 #include "xfs_extfree_item.h"
53 #include "xfs_acl.h"
54 #include "xfs_attr.h"
55 #include "xfs_clnt.h"
56 #include "xfs_fsops.h"
57
58 STATIC int      xfs_sync(bhv_desc_t *, int, cred_t *);
59
60 int
61 xfs_init(void)
62 {
63         extern kmem_zone_t      *xfs_bmap_free_item_zone;
64         extern kmem_zone_t      *xfs_btree_cur_zone;
65         extern kmem_zone_t      *xfs_trans_zone;
66         extern kmem_zone_t      *xfs_buf_item_zone;
67         extern kmem_zone_t      *xfs_dabuf_zone;
68 #ifdef XFS_DABUF_DEBUG
69         extern lock_t           xfs_dabuf_global_lock;
70         spinlock_init(&xfs_dabuf_global_lock, "xfsda");
71 #endif
72
73         /*
74          * Initialize all of the zone allocators we use.
75          */
76         xfs_bmap_free_item_zone = kmem_zone_init(sizeof(xfs_bmap_free_item_t),
77                                                  "xfs_bmap_free_item");
78         xfs_btree_cur_zone = kmem_zone_init(sizeof(xfs_btree_cur_t),
79                                             "xfs_btree_cur");
80         xfs_trans_zone = kmem_zone_init(sizeof(xfs_trans_t), "xfs_trans");
81         xfs_da_state_zone =
82                 kmem_zone_init(sizeof(xfs_da_state_t), "xfs_da_state");
83         xfs_dabuf_zone = kmem_zone_init(sizeof(xfs_dabuf_t), "xfs_dabuf");
84         xfs_ifork_zone = kmem_zone_init(sizeof(xfs_ifork_t), "xfs_ifork");
85         xfs_acl_zone_init(xfs_acl_zone, "xfs_acl");
86
87         /*
88          * The size of the zone allocated buf log item is the maximum
89          * size possible under XFS.  This wastes a little bit of memory,
90          * but it is much faster.
91          */
92         xfs_buf_item_zone =
93                 kmem_zone_init((sizeof(xfs_buf_log_item_t) +
94                                 (((XFS_MAX_BLOCKSIZE / XFS_BLI_CHUNK) /
95                                   NBWORD) * sizeof(int))),
96                                "xfs_buf_item");
97         xfs_efd_zone =
98                 kmem_zone_init((sizeof(xfs_efd_log_item_t) +
99                                ((XFS_EFD_MAX_FAST_EXTENTS - 1) *
100                                  sizeof(xfs_extent_t))),
101                                       "xfs_efd_item");
102         xfs_efi_zone =
103                 kmem_zone_init((sizeof(xfs_efi_log_item_t) +
104                                ((XFS_EFI_MAX_FAST_EXTENTS - 1) *
105                                  sizeof(xfs_extent_t))),
106                                       "xfs_efi_item");
107
108         /*
109          * These zones warrant special memory allocator hints
110          */
111         xfs_inode_zone =
112                 kmem_zone_init_flags(sizeof(xfs_inode_t), "xfs_inode",
113                                         KM_ZONE_HWALIGN | KM_ZONE_RECLAIM |
114                                         KM_ZONE_SPREAD, NULL);
115         xfs_ili_zone =
116                 kmem_zone_init_flags(sizeof(xfs_inode_log_item_t), "xfs_ili",
117                                         KM_ZONE_SPREAD, NULL);
118         xfs_chashlist_zone =
119                 kmem_zone_init_flags(sizeof(xfs_chashlist_t), "xfs_chashlist",
120                                         KM_ZONE_SPREAD, NULL);
121
122         /*
123          * Allocate global trace buffers.
124          */
125 #ifdef XFS_ALLOC_TRACE
126         xfs_alloc_trace_buf = ktrace_alloc(XFS_ALLOC_TRACE_SIZE, KM_SLEEP);
127 #endif
128 #ifdef XFS_BMAP_TRACE
129         xfs_bmap_trace_buf = ktrace_alloc(XFS_BMAP_TRACE_SIZE, KM_SLEEP);
130 #endif
131 #ifdef XFS_BMBT_TRACE
132         xfs_bmbt_trace_buf = ktrace_alloc(XFS_BMBT_TRACE_SIZE, KM_SLEEP);
133 #endif
134 #ifdef XFS_DIR_TRACE
135         xfs_dir_trace_buf = ktrace_alloc(XFS_DIR_TRACE_SIZE, KM_SLEEP);
136 #endif
137 #ifdef XFS_ATTR_TRACE
138         xfs_attr_trace_buf = ktrace_alloc(XFS_ATTR_TRACE_SIZE, KM_SLEEP);
139 #endif
140 #ifdef XFS_DIR2_TRACE
141         xfs_dir2_trace_buf = ktrace_alloc(XFS_DIR2_GTRACE_SIZE, KM_SLEEP);
142 #endif
143
144         xfs_dir_startup();
145
146 #if (defined(DEBUG) || defined(INDUCE_IO_ERROR))
147         xfs_error_test_init();
148 #endif /* DEBUG || INDUCE_IO_ERROR */
149
150         xfs_init_procfs();
151         xfs_sysctl_register();
152         return 0;
153 }
154
155 void
156 xfs_cleanup(void)
157 {
158         extern kmem_zone_t      *xfs_bmap_free_item_zone;
159         extern kmem_zone_t      *xfs_btree_cur_zone;
160         extern kmem_zone_t      *xfs_inode_zone;
161         extern kmem_zone_t      *xfs_trans_zone;
162         extern kmem_zone_t      *xfs_da_state_zone;
163         extern kmem_zone_t      *xfs_dabuf_zone;
164         extern kmem_zone_t      *xfs_efd_zone;
165         extern kmem_zone_t      *xfs_efi_zone;
166         extern kmem_zone_t      *xfs_buf_item_zone;
167         extern kmem_zone_t      *xfs_chashlist_zone;
168
169         xfs_cleanup_procfs();
170         xfs_sysctl_unregister();
171         xfs_refcache_destroy();
172         xfs_acl_zone_destroy(xfs_acl_zone);
173
174 #ifdef XFS_DIR2_TRACE
175         ktrace_free(xfs_dir2_trace_buf);
176 #endif
177 #ifdef XFS_ATTR_TRACE
178         ktrace_free(xfs_attr_trace_buf);
179 #endif
180 #ifdef XFS_DIR_TRACE
181         ktrace_free(xfs_dir_trace_buf);
182 #endif
183 #ifdef XFS_BMBT_TRACE
184         ktrace_free(xfs_bmbt_trace_buf);
185 #endif
186 #ifdef XFS_BMAP_TRACE
187         ktrace_free(xfs_bmap_trace_buf);
188 #endif
189 #ifdef XFS_ALLOC_TRACE
190         ktrace_free(xfs_alloc_trace_buf);
191 #endif
192
193         kmem_zone_destroy(xfs_bmap_free_item_zone);
194         kmem_zone_destroy(xfs_btree_cur_zone);
195         kmem_zone_destroy(xfs_inode_zone);
196         kmem_zone_destroy(xfs_trans_zone);
197         kmem_zone_destroy(xfs_da_state_zone);
198         kmem_zone_destroy(xfs_dabuf_zone);
199         kmem_zone_destroy(xfs_buf_item_zone);
200         kmem_zone_destroy(xfs_efd_zone);
201         kmem_zone_destroy(xfs_efi_zone);
202         kmem_zone_destroy(xfs_ifork_zone);
203         kmem_zone_destroy(xfs_ili_zone);
204         kmem_zone_destroy(xfs_chashlist_zone);
205 }
206
207 /*
208  * xfs_start_flags
209  *
210  * This function fills in xfs_mount_t fields based on mount args.
211  * Note: the superblock has _not_ yet been read in.
212  */
213 STATIC int
214 xfs_start_flags(
215         struct vfs              *vfs,
216         struct xfs_mount_args   *ap,
217         struct xfs_mount        *mp)
218 {
219         /* Values are in BBs */
220         if ((ap->flags & XFSMNT_NOALIGN) != XFSMNT_NOALIGN) {
221                 /*
222                  * At this point the superblock has not been read
223                  * in, therefore we do not know the block size.
224                  * Before the mount call ends we will convert
225                  * these to FSBs.
226                  */
227                 mp->m_dalign = ap->sunit;
228                 mp->m_swidth = ap->swidth;
229         }
230
231         if (ap->logbufs != -1 &&
232             ap->logbufs != 0 &&
233             (ap->logbufs < XLOG_MIN_ICLOGS ||
234              ap->logbufs > XLOG_MAX_ICLOGS)) {
235                 cmn_err(CE_WARN,
236                         "XFS: invalid logbufs value: %d [not %d-%d]",
237                         ap->logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
238                 return XFS_ERROR(EINVAL);
239         }
240         mp->m_logbufs = ap->logbufs;
241         if (ap->logbufsize != -1 &&
242             ap->logbufsize !=  0 &&
243             ap->logbufsize != 16 * 1024 &&
244             ap->logbufsize != 32 * 1024 &&
245             ap->logbufsize != 64 * 1024 &&
246             ap->logbufsize != 128 * 1024 &&
247             ap->logbufsize != 256 * 1024) {
248                 cmn_err(CE_WARN,
249         "XFS: invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
250                         ap->logbufsize);
251                 return XFS_ERROR(EINVAL);
252         }
253         mp->m_ihsize = ap->ihashsize;
254         mp->m_logbsize = ap->logbufsize;
255         mp->m_fsname_len = strlen(ap->fsname) + 1;
256         mp->m_fsname = kmem_alloc(mp->m_fsname_len, KM_SLEEP);
257         strcpy(mp->m_fsname, ap->fsname);
258         if (ap->rtname[0]) {
259                 mp->m_rtname = kmem_alloc(strlen(ap->rtname) + 1, KM_SLEEP);
260                 strcpy(mp->m_rtname, ap->rtname);
261         }
262         if (ap->logname[0]) {
263                 mp->m_logname = kmem_alloc(strlen(ap->logname) + 1, KM_SLEEP);
264                 strcpy(mp->m_logname, ap->logname);
265         }
266
267         if (ap->flags & XFSMNT_WSYNC)
268                 mp->m_flags |= XFS_MOUNT_WSYNC;
269 #if XFS_BIG_INUMS
270         if (ap->flags & XFSMNT_INO64) {
271                 mp->m_flags |= XFS_MOUNT_INO64;
272                 mp->m_inoadd = XFS_INO64_OFFSET;
273         }
274 #endif
275         if (ap->flags & XFSMNT_RETERR)
276                 mp->m_flags |= XFS_MOUNT_RETERR;
277         if (ap->flags & XFSMNT_NOALIGN)
278                 mp->m_flags |= XFS_MOUNT_NOALIGN;
279         if (ap->flags & XFSMNT_SWALLOC)
280                 mp->m_flags |= XFS_MOUNT_SWALLOC;
281         if (ap->flags & XFSMNT_OSYNCISOSYNC)
282                 mp->m_flags |= XFS_MOUNT_OSYNCISOSYNC;
283         if (ap->flags & XFSMNT_32BITINODES)
284                 mp->m_flags |= XFS_MOUNT_32BITINODES;
285
286         if (ap->flags & XFSMNT_IOSIZE) {
287                 if (ap->iosizelog > XFS_MAX_IO_LOG ||
288                     ap->iosizelog < XFS_MIN_IO_LOG) {
289                         cmn_err(CE_WARN,
290                 "XFS: invalid log iosize: %d [not %d-%d]",
291                                 ap->iosizelog, XFS_MIN_IO_LOG,
292                                 XFS_MAX_IO_LOG);
293                         return XFS_ERROR(EINVAL);
294                 }
295
296                 mp->m_flags |= XFS_MOUNT_DFLT_IOSIZE;
297                 mp->m_readio_log = mp->m_writeio_log = ap->iosizelog;
298         }
299
300         if (ap->flags & XFSMNT_IHASHSIZE)
301                 mp->m_flags |= XFS_MOUNT_IHASHSIZE;
302         if (ap->flags & XFSMNT_IDELETE)
303                 mp->m_flags |= XFS_MOUNT_IDELETE;
304         if (ap->flags & XFSMNT_DIRSYNC)
305                 mp->m_flags |= XFS_MOUNT_DIRSYNC;
306         if (ap->flags & XFSMNT_ATTR2)
307                 mp->m_flags |= XFS_MOUNT_ATTR2;
308
309         if (ap->flags2 & XFSMNT2_COMPAT_IOSIZE)
310                 mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE;
311
312         /*
313          * no recovery flag requires a read-only mount
314          */
315         if (ap->flags & XFSMNT_NORECOVERY) {
316                 if (!(vfs->vfs_flag & VFS_RDONLY)) {
317                         cmn_err(CE_WARN,
318         "XFS: tried to mount a FS read-write without recovery!");
319                         return XFS_ERROR(EINVAL);
320                 }
321                 mp->m_flags |= XFS_MOUNT_NORECOVERY;
322         }
323
324         if (ap->flags & XFSMNT_NOUUID)
325                 mp->m_flags |= XFS_MOUNT_NOUUID;
326         if (ap->flags & XFSMNT_BARRIER)
327                 mp->m_flags |= XFS_MOUNT_BARRIER;
328         else
329                 mp->m_flags &= ~XFS_MOUNT_BARRIER;
330
331         return 0;
332 }
333
334 /*
335  * This function fills in xfs_mount_t fields based on mount args.
336  * Note: the superblock _has_ now been read in.
337  */
338 STATIC int
339 xfs_finish_flags(
340         struct vfs              *vfs,
341         struct xfs_mount_args   *ap,
342         struct xfs_mount        *mp)
343 {
344         int                     ronly = (vfs->vfs_flag & VFS_RDONLY);
345
346         /* Fail a mount where the logbuf is smaller then the log stripe */
347         if (XFS_SB_VERSION_HASLOGV2(&mp->m_sb)) {
348                 if ((ap->logbufsize <= 0) &&
349                     (mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE)) {
350                         mp->m_logbsize = mp->m_sb.sb_logsunit;
351                 } else if (ap->logbufsize > 0 &&
352                            ap->logbufsize < mp->m_sb.sb_logsunit) {
353                         cmn_err(CE_WARN,
354         "XFS: logbuf size must be greater than or equal to log stripe size");
355                         return XFS_ERROR(EINVAL);
356                 }
357         } else {
358                 /* Fail a mount if the logbuf is larger than 32K */
359                 if (ap->logbufsize > XLOG_BIG_RECORD_BSIZE) {
360                         cmn_err(CE_WARN,
361         "XFS: logbuf size for version 1 logs must be 16K or 32K");
362                         return XFS_ERROR(EINVAL);
363                 }
364         }
365
366         if (XFS_SB_VERSION_HASATTR2(&mp->m_sb)) {
367                 mp->m_flags |= XFS_MOUNT_ATTR2;
368         }
369
370         /*
371          * prohibit r/w mounts of read-only filesystems
372          */
373         if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !ronly) {
374                 cmn_err(CE_WARN,
375         "XFS: cannot mount a read-only filesystem as read-write");
376                 return XFS_ERROR(EROFS);
377         }
378
379         /*
380          * check for shared mount.
381          */
382         if (ap->flags & XFSMNT_SHARED) {
383                 if (!XFS_SB_VERSION_HASSHARED(&mp->m_sb))
384                         return XFS_ERROR(EINVAL);
385
386                 /*
387                  * For IRIX 6.5, shared mounts must have the shared
388                  * version bit set, have the persistent readonly
389                  * field set, must be version 0 and can only be mounted
390                  * read-only.
391                  */
392                 if (!ronly || !(mp->m_sb.sb_flags & XFS_SBF_READONLY) ||
393                      (mp->m_sb.sb_shared_vn != 0))
394                         return XFS_ERROR(EINVAL);
395
396                 mp->m_flags |= XFS_MOUNT_SHARED;
397
398                 /*
399                  * Shared XFS V0 can't deal with DMI.  Return EINVAL.
400                  */
401                 if (mp->m_sb.sb_shared_vn == 0 && (ap->flags & XFSMNT_DMAPI))
402                         return XFS_ERROR(EINVAL);
403         }
404
405         return 0;
406 }
407
408 /*
409  * xfs_mount
410  *
411  * The file system configurations are:
412  *      (1) device (partition) with data and internal log
413  *      (2) logical volume with data and log subvolumes.
414  *      (3) logical volume with data, log, and realtime subvolumes.
415  *
416  * We only have to handle opening the log and realtime volumes here if
417  * they are present.  The data subvolume has already been opened by
418  * get_sb_bdev() and is stored in vfsp->vfs_super->s_bdev.
419  */
420 STATIC int
421 xfs_mount(
422         struct bhv_desc         *bhvp,
423         struct xfs_mount_args   *args,
424         cred_t                  *credp)
425 {
426         struct vfs              *vfsp = bhvtovfs(bhvp);
427         struct bhv_desc         *p;
428         struct xfs_mount        *mp = XFS_BHVTOM(bhvp);
429         struct block_device     *ddev, *logdev, *rtdev;
430         int                     flags = 0, error;
431
432         ddev = vfsp->vfs_super->s_bdev;
433         logdev = rtdev = NULL;
434
435         /*
436          * Setup xfs_mount function vectors from available behaviors
437          */
438         p = vfs_bhv_lookup(vfsp, VFS_POSITION_DM);
439         mp->m_dm_ops = p ? *(xfs_dmops_t *) vfs_bhv_custom(p) : xfs_dmcore_stub;
440         p = vfs_bhv_lookup(vfsp, VFS_POSITION_QM);
441         mp->m_qm_ops = p ? *(xfs_qmops_t *) vfs_bhv_custom(p) : xfs_qmcore_stub;
442         p = vfs_bhv_lookup(vfsp, VFS_POSITION_IO);
443         mp->m_io_ops = p ? *(xfs_ioops_t *) vfs_bhv_custom(p) : xfs_iocore_xfs;
444
445         /*
446          * Open real time and log devices - order is important.
447          */
448         if (args->logname[0]) {
449                 error = xfs_blkdev_get(mp, args->logname, &logdev);
450                 if (error)
451                         return error;
452         }
453         if (args->rtname[0]) {
454                 error = xfs_blkdev_get(mp, args->rtname, &rtdev);
455                 if (error) {
456                         xfs_blkdev_put(logdev);
457                         return error;
458                 }
459
460                 if (rtdev == ddev || rtdev == logdev) {
461                         cmn_err(CE_WARN,
462         "XFS: Cannot mount filesystem with identical rtdev and ddev/logdev.");
463                         xfs_blkdev_put(logdev);
464                         xfs_blkdev_put(rtdev);
465                         return EINVAL;
466                 }
467         }
468
469         /*
470          * Setup xfs_mount buffer target pointers
471          */
472         error = ENOMEM;
473         mp->m_ddev_targp = xfs_alloc_buftarg(ddev, 0);
474         if (!mp->m_ddev_targp) {
475                 xfs_blkdev_put(logdev);
476                 xfs_blkdev_put(rtdev);
477                 return error;
478         }
479         if (rtdev) {
480                 mp->m_rtdev_targp = xfs_alloc_buftarg(rtdev, 1);
481                 if (!mp->m_rtdev_targp)
482                         goto error0;
483         }
484         mp->m_logdev_targp = (logdev && logdev != ddev) ?
485                                 xfs_alloc_buftarg(logdev, 1) : mp->m_ddev_targp;
486         if (!mp->m_logdev_targp)
487                 goto error0;
488
489         /*
490          * Setup flags based on mount(2) options and then the superblock
491          */
492         error = xfs_start_flags(vfsp, args, mp);
493         if (error)
494                 goto error1;
495         error = xfs_readsb(mp);
496         if (error)
497                 goto error1;
498         error = xfs_finish_flags(vfsp, args, mp);
499         if (error)
500                 goto error2;
501
502         /*
503          * Setup xfs_mount buffer target pointers based on superblock
504          */
505         error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_blocksize,
506                                     mp->m_sb.sb_sectsize);
507         if (!error && logdev && logdev != ddev) {
508                 unsigned int    log_sector_size = BBSIZE;
509
510                 if (XFS_SB_VERSION_HASSECTOR(&mp->m_sb))
511                         log_sector_size = mp->m_sb.sb_logsectsize;
512                 error = xfs_setsize_buftarg(mp->m_logdev_targp,
513                                             mp->m_sb.sb_blocksize,
514                                             log_sector_size);
515         }
516         if (!error && rtdev)
517                 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
518                                             mp->m_sb.sb_blocksize,
519                                             mp->m_sb.sb_sectsize);
520         if (error)
521                 goto error2;
522
523         if ((mp->m_flags & XFS_MOUNT_BARRIER) && !(vfsp->vfs_flag & VFS_RDONLY))
524                 xfs_mountfs_check_barriers(mp);
525
526         error = XFS_IOINIT(vfsp, args, flags);
527         if (error)
528                 goto error2;
529
530         return 0;
531
532 error2:
533         if (mp->m_sb_bp)
534                 xfs_freesb(mp);
535 error1:
536         xfs_binval(mp->m_ddev_targp);
537         if (logdev && logdev != ddev)
538                 xfs_binval(mp->m_logdev_targp);
539         if (rtdev)
540                 xfs_binval(mp->m_rtdev_targp);
541 error0:
542         xfs_unmountfs_close(mp, credp);
543         return error;
544 }
545
546 STATIC int
547 xfs_unmount(
548         bhv_desc_t      *bdp,
549         int             flags,
550         cred_t          *credp)
551 {
552         struct vfs      *vfsp = bhvtovfs(bdp);
553         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
554         xfs_inode_t     *rip;
555         vnode_t         *rvp;
556         int             unmount_event_wanted = 0;
557         int             unmount_event_flags = 0;
558         int             xfs_unmountfs_needed = 0;
559         int             error;
560
561         rip = mp->m_rootip;
562         rvp = XFS_ITOV(rip);
563
564         if (vfsp->vfs_flag & VFS_DMI) {
565                 error = XFS_SEND_PREUNMOUNT(mp, vfsp,
566                                 rvp, DM_RIGHT_NULL, rvp, DM_RIGHT_NULL,
567                                 NULL, NULL, 0, 0,
568                                 (mp->m_dmevmask & (1<<DM_EVENT_PREUNMOUNT))?
569                                         0:DM_FLAGS_UNWANTED);
570                         if (error)
571                                 return XFS_ERROR(error);
572                 unmount_event_wanted = 1;
573                 unmount_event_flags = (mp->m_dmevmask & (1<<DM_EVENT_UNMOUNT))?
574                                         0 : DM_FLAGS_UNWANTED;
575         }
576
577         /*
578          * First blow any referenced inode from this file system
579          * out of the reference cache, and delete the timer.
580          */
581         xfs_refcache_purge_mp(mp);
582
583         XFS_bflush(mp->m_ddev_targp);
584         error = xfs_unmount_flush(mp, 0);
585         if (error)
586                 goto out;
587
588         ASSERT(vn_count(rvp) == 1);
589
590         /*
591          * Drop the reference count
592          */
593         VN_RELE(rvp);
594
595         /*
596          * If we're forcing a shutdown, typically because of a media error,
597          * we want to make sure we invalidate dirty pages that belong to
598          * referenced vnodes as well.
599          */
600         if (XFS_FORCED_SHUTDOWN(mp)) {
601                 error = xfs_sync(&mp->m_bhv,
602                          (SYNC_WAIT | SYNC_CLOSE), credp);
603                 ASSERT(error != EFSCORRUPTED);
604         }
605         xfs_unmountfs_needed = 1;
606
607 out:
608         /*      Send DMAPI event, if required.
609          *      Then do xfs_unmountfs() if needed.
610          *      Then return error (or zero).
611          */
612         if (unmount_event_wanted) {
613                 /* Note: mp structure must still exist for
614                  * XFS_SEND_UNMOUNT() call.
615                  */
616                 XFS_SEND_UNMOUNT(mp, vfsp, error == 0 ? rvp : NULL,
617                         DM_RIGHT_NULL, 0, error, unmount_event_flags);
618         }
619         if (xfs_unmountfs_needed) {
620                 /*
621                  * Call common unmount function to flush to disk
622                  * and free the super block buffer & mount structures.
623                  */
624                 xfs_unmountfs(mp, credp);
625         }
626
627         return XFS_ERROR(error);
628 }
629
630 STATIC int
631 xfs_quiesce_fs(
632         xfs_mount_t             *mp)
633 {
634         int                     count = 0, pincount;
635
636         xfs_refcache_purge_mp(mp);
637         xfs_flush_buftarg(mp->m_ddev_targp, 0);
638         xfs_finish_reclaim_all(mp, 0);
639
640         /* This loop must run at least twice.
641          * The first instance of the loop will flush
642          * most meta data but that will generate more
643          * meta data (typically directory updates).
644          * Which then must be flushed and logged before
645          * we can write the unmount record.
646          */
647         do {
648                 xfs_syncsub(mp, SYNC_REMOUNT|SYNC_ATTR|SYNC_WAIT, 0, NULL);
649                 pincount = xfs_flush_buftarg(mp->m_ddev_targp, 1);
650                 if (!pincount) {
651                         delay(50);
652                         count++;
653                 }
654         } while (count < 2);
655
656         return 0;
657 }
658
659 STATIC int
660 xfs_mntupdate(
661         bhv_desc_t                      *bdp,
662         int                             *flags,
663         struct xfs_mount_args           *args)
664 {
665         struct vfs      *vfsp = bhvtovfs(bdp);
666         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
667         int             error;
668
669         if (args->flags & XFSMNT_BARRIER)
670                 mp->m_flags |= XFS_MOUNT_BARRIER;
671         else
672                 mp->m_flags &= ~XFS_MOUNT_BARRIER;
673
674         if ((vfsp->vfs_flag & VFS_RDONLY) &&
675             !(*flags & MS_RDONLY)) {
676                 vfsp->vfs_flag &= ~VFS_RDONLY;
677
678                 if (args->flags & XFSMNT_BARRIER)
679                         xfs_mountfs_check_barriers(mp);
680         }
681
682         if (!(vfsp->vfs_flag & VFS_RDONLY) &&
683             (*flags & MS_RDONLY)) {
684                 VFS_SYNC(vfsp, SYNC_FSDATA|SYNC_BDFLUSH|SYNC_ATTR, NULL, error);
685
686                 xfs_quiesce_fs(mp);
687
688                 /* Ok now write out an unmount record */
689                 xfs_log_unmount_write(mp);
690                 xfs_unmountfs_writesb(mp);
691                 vfsp->vfs_flag |= VFS_RDONLY;
692         }
693
694         return 0;
695 }
696
697 /*
698  * xfs_unmount_flush implements a set of flush operation on special
699  * inodes, which are needed as a separate set of operations so that
700  * they can be called as part of relocation process.
701  */
702 int
703 xfs_unmount_flush(
704         xfs_mount_t     *mp,            /* Mount structure we are getting
705                                            rid of. */
706         int             relocation)     /* Called from vfs relocation. */
707 {
708         xfs_inode_t     *rip = mp->m_rootip;
709         xfs_inode_t     *rbmip;
710         xfs_inode_t     *rsumip = NULL;
711         vnode_t         *rvp = XFS_ITOV(rip);
712         int             error;
713
714         xfs_ilock(rip, XFS_ILOCK_EXCL);
715         xfs_iflock(rip);
716
717         /*
718          * Flush out the real time inodes.
719          */
720         if ((rbmip = mp->m_rbmip) != NULL) {
721                 xfs_ilock(rbmip, XFS_ILOCK_EXCL);
722                 xfs_iflock(rbmip);
723                 error = xfs_iflush(rbmip, XFS_IFLUSH_SYNC);
724                 xfs_iunlock(rbmip, XFS_ILOCK_EXCL);
725
726                 if (error == EFSCORRUPTED)
727                         goto fscorrupt_out;
728
729                 ASSERT(vn_count(XFS_ITOV(rbmip)) == 1);
730
731                 rsumip = mp->m_rsumip;
732                 xfs_ilock(rsumip, XFS_ILOCK_EXCL);
733                 xfs_iflock(rsumip);
734                 error = xfs_iflush(rsumip, XFS_IFLUSH_SYNC);
735                 xfs_iunlock(rsumip, XFS_ILOCK_EXCL);
736
737                 if (error == EFSCORRUPTED)
738                         goto fscorrupt_out;
739
740                 ASSERT(vn_count(XFS_ITOV(rsumip)) == 1);
741         }
742
743         /*
744          * Synchronously flush root inode to disk
745          */
746         error = xfs_iflush(rip, XFS_IFLUSH_SYNC);
747         if (error == EFSCORRUPTED)
748                 goto fscorrupt_out2;
749
750         if (vn_count(rvp) != 1 && !relocation) {
751                 xfs_iunlock(rip, XFS_ILOCK_EXCL);
752                 return XFS_ERROR(EBUSY);
753         }
754
755         /*
756          * Release dquot that rootinode, rbmino and rsumino might be holding,
757          * flush and purge the quota inodes.
758          */
759         error = XFS_QM_UNMOUNT(mp);
760         if (error == EFSCORRUPTED)
761                 goto fscorrupt_out2;
762
763         if (rbmip) {
764                 VN_RELE(XFS_ITOV(rbmip));
765                 VN_RELE(XFS_ITOV(rsumip));
766         }
767
768         xfs_iunlock(rip, XFS_ILOCK_EXCL);
769         return 0;
770
771 fscorrupt_out:
772         xfs_ifunlock(rip);
773
774 fscorrupt_out2:
775         xfs_iunlock(rip, XFS_ILOCK_EXCL);
776
777         return XFS_ERROR(EFSCORRUPTED);
778 }
779
780 /*
781  * xfs_root extracts the root vnode from a vfs.
782  *
783  * vfsp -- the vfs struct for the desired file system
784  * vpp  -- address of the caller's vnode pointer which should be
785  *         set to the desired fs root vnode
786  */
787 STATIC int
788 xfs_root(
789         bhv_desc_t      *bdp,
790         vnode_t         **vpp)
791 {
792         vnode_t         *vp;
793
794         vp = XFS_ITOV((XFS_BHVTOM(bdp))->m_rootip);
795         VN_HOLD(vp);
796         *vpp = vp;
797         return 0;
798 }
799
800 /*
801  * xfs_statvfs
802  *
803  * Fill in the statvfs structure for the given file system.  We use
804  * the superblock lock in the mount structure to ensure a consistent
805  * snapshot of the counters returned.
806  */
807 STATIC int
808 xfs_statvfs(
809         bhv_desc_t      *bdp,
810         xfs_statfs_t    *statp,
811         vnode_t         *vp)
812 {
813         __uint64_t      fakeinos;
814         xfs_extlen_t    lsize;
815         xfs_mount_t     *mp;
816         xfs_sb_t        *sbp;
817         unsigned long   s;
818
819         mp = XFS_BHVTOM(bdp);
820         sbp = &(mp->m_sb);
821
822         statp->f_type = XFS_SB_MAGIC;
823
824         xfs_icsb_sync_counters_lazy(mp);
825         s = XFS_SB_LOCK(mp);
826         statp->f_bsize = sbp->sb_blocksize;
827         lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
828         statp->f_blocks = sbp->sb_dblocks - lsize;
829         statp->f_bfree = statp->f_bavail = sbp->sb_fdblocks;
830         fakeinos = statp->f_bfree << sbp->sb_inopblog;
831 #if XFS_BIG_INUMS
832         fakeinos += mp->m_inoadd;
833 #endif
834         statp->f_files =
835             MIN(sbp->sb_icount + fakeinos, (__uint64_t)XFS_MAXINUMBER);
836         if (mp->m_maxicount)
837 #if XFS_BIG_INUMS
838                 if (!mp->m_inoadd)
839 #endif
840                         statp->f_files = min_t(typeof(statp->f_files),
841                                                 statp->f_files,
842                                                 mp->m_maxicount);
843         statp->f_ffree = statp->f_files - (sbp->sb_icount - sbp->sb_ifree);
844         XFS_SB_UNLOCK(mp, s);
845
846         xfs_statvfs_fsid(statp, mp);
847         statp->f_namelen = MAXNAMELEN - 1;
848
849         return 0;
850 }
851
852
853 /*
854  * xfs_sync flushes any pending I/O to file system vfsp.
855  *
856  * This routine is called by vfs_sync() to make sure that things make it
857  * out to disk eventually, on sync() system calls to flush out everything,
858  * and when the file system is unmounted.  For the vfs_sync() case, all
859  * we really need to do is sync out the log to make all of our meta-data
860  * updates permanent (except for timestamps).  For calls from pflushd(),
861  * dirty pages are kept moving by calling pdflush() on the inodes
862  * containing them.  We also flush the inodes that we can lock without
863  * sleeping and the superblock if we can lock it without sleeping from
864  * vfs_sync() so that items at the tail of the log are always moving out.
865  *
866  * Flags:
867  *      SYNC_BDFLUSH - We're being called from vfs_sync() so we don't want
868  *                     to sleep if we can help it.  All we really need
869  *                     to do is ensure that the log is synced at least
870  *                     periodically.  We also push the inodes and
871  *                     superblock if we can lock them without sleeping
872  *                      and they are not pinned.
873  *      SYNC_ATTR    - We need to flush the inodes.  If SYNC_BDFLUSH is not
874  *                     set, then we really want to lock each inode and flush
875  *                     it.
876  *      SYNC_WAIT    - All the flushes that take place in this call should
877  *                     be synchronous.
878  *      SYNC_DELWRI  - This tells us to push dirty pages associated with
879  *                     inodes.  SYNC_WAIT and SYNC_BDFLUSH are used to
880  *                     determine if they should be flushed sync, async, or
881  *                     delwri.
882  *      SYNC_CLOSE   - This flag is passed when the system is being
883  *                     unmounted.  We should sync and invalidate everything.
884  *      SYNC_FSDATA  - This indicates that the caller would like to make
885  *                     sure the superblock is safe on disk.  We can ensure
886  *                     this by simply making sure the log gets flushed
887  *                     if SYNC_BDFLUSH is set, and by actually writing it
888  *                     out otherwise.
889  *
890  */
891 /*ARGSUSED*/
892 STATIC int
893 xfs_sync(
894         bhv_desc_t      *bdp,
895         int             flags,
896         cred_t          *credp)
897 {
898         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
899
900         if (unlikely(flags == SYNC_QUIESCE))
901                 return xfs_quiesce_fs(mp);
902         else
903                 return xfs_syncsub(mp, flags, 0, NULL);
904 }
905
906 /*
907  * xfs sync routine for internal use
908  *
909  * This routine supports all of the flags defined for the generic VFS_SYNC
910  * interface as explained above under xfs_sync.  In the interests of not
911  * changing interfaces within the 6.5 family, additional internally-
912  * required functions are specified within a separate xflags parameter,
913  * only available by calling this routine.
914  *
915  */
916 int
917 xfs_sync_inodes(
918         xfs_mount_t     *mp,
919         int             flags,
920         int             xflags,
921         int             *bypassed)
922 {
923         xfs_inode_t     *ip = NULL;
924         xfs_inode_t     *ip_next;
925         xfs_buf_t       *bp;
926         vnode_t         *vp = NULL;
927         int             error;
928         int             last_error;
929         uint64_t        fflag;
930         uint            lock_flags;
931         uint            base_lock_flags;
932         boolean_t       mount_locked;
933         boolean_t       vnode_refed;
934         int             preempt;
935         xfs_dinode_t    *dip;
936         xfs_iptr_t      *ipointer;
937 #ifdef DEBUG
938         boolean_t       ipointer_in = B_FALSE;
939
940 #define IPOINTER_SET    ipointer_in = B_TRUE
941 #define IPOINTER_CLR    ipointer_in = B_FALSE
942 #else
943 #define IPOINTER_SET
944 #define IPOINTER_CLR
945 #endif
946
947
948 /* Insert a marker record into the inode list after inode ip. The list
949  * must be locked when this is called. After the call the list will no
950  * longer be locked.
951  */
952 #define IPOINTER_INSERT(ip, mp) { \
953                 ASSERT(ipointer_in == B_FALSE); \
954                 ipointer->ip_mnext = ip->i_mnext; \
955                 ipointer->ip_mprev = ip; \
956                 ip->i_mnext = (xfs_inode_t *)ipointer; \
957                 ipointer->ip_mnext->i_mprev = (xfs_inode_t *)ipointer; \
958                 preempt = 0; \
959                 XFS_MOUNT_IUNLOCK(mp); \
960                 mount_locked = B_FALSE; \
961                 IPOINTER_SET; \
962         }
963
964 /* Remove the marker from the inode list. If the marker was the only item
965  * in the list then there are no remaining inodes and we should zero out
966  * the whole list. If we are the current head of the list then move the head
967  * past us.
968  */
969 #define IPOINTER_REMOVE(ip, mp) { \
970                 ASSERT(ipointer_in == B_TRUE); \
971                 if (ipointer->ip_mnext != (xfs_inode_t *)ipointer) { \
972                         ip = ipointer->ip_mnext; \
973                         ip->i_mprev = ipointer->ip_mprev; \
974                         ipointer->ip_mprev->i_mnext = ip; \
975                         if (mp->m_inodes == (xfs_inode_t *)ipointer) { \
976                                 mp->m_inodes = ip; \
977                         } \
978                 } else { \
979                         ASSERT(mp->m_inodes == (xfs_inode_t *)ipointer); \
980                         mp->m_inodes = NULL; \
981                         ip = NULL; \
982                 } \
983                 IPOINTER_CLR; \
984         }
985
986 #define XFS_PREEMPT_MASK        0x7f
987
988         if (bypassed)
989                 *bypassed = 0;
990         if (XFS_MTOVFS(mp)->vfs_flag & VFS_RDONLY)
991                 return 0;
992         error = 0;
993         last_error = 0;
994         preempt = 0;
995
996         /* Allocate a reference marker */
997         ipointer = (xfs_iptr_t *)kmem_zalloc(sizeof(xfs_iptr_t), KM_SLEEP);
998
999         fflag = XFS_B_ASYNC;            /* default is don't wait */
1000         if (flags & (SYNC_BDFLUSH | SYNC_DELWRI))
1001                 fflag = XFS_B_DELWRI;
1002         if (flags & SYNC_WAIT)
1003                 fflag = 0;              /* synchronous overrides all */
1004
1005         base_lock_flags = XFS_ILOCK_SHARED;
1006         if (flags & (SYNC_DELWRI | SYNC_CLOSE)) {
1007                 /*
1008                  * We need the I/O lock if we're going to call any of
1009                  * the flush/inval routines.
1010                  */
1011                 base_lock_flags |= XFS_IOLOCK_SHARED;
1012         }
1013
1014         XFS_MOUNT_ILOCK(mp);
1015
1016         ip = mp->m_inodes;
1017
1018         mount_locked = B_TRUE;
1019         vnode_refed  = B_FALSE;
1020
1021         IPOINTER_CLR;
1022
1023         do {
1024                 ASSERT(ipointer_in == B_FALSE);
1025                 ASSERT(vnode_refed == B_FALSE);
1026
1027                 lock_flags = base_lock_flags;
1028
1029                 /*
1030                  * There were no inodes in the list, just break out
1031                  * of the loop.
1032                  */
1033                 if (ip == NULL) {
1034                         break;
1035                 }
1036
1037                 /*
1038                  * We found another sync thread marker - skip it
1039                  */
1040                 if (ip->i_mount == NULL) {
1041                         ip = ip->i_mnext;
1042                         continue;
1043                 }
1044
1045                 vp = XFS_ITOV_NULL(ip);
1046
1047                 /*
1048                  * If the vnode is gone then this is being torn down,
1049                  * call reclaim if it is flushed, else let regular flush
1050                  * code deal with it later in the loop.
1051                  */
1052
1053                 if (vp == NULL) {
1054                         /* Skip ones already in reclaim */
1055                         if (ip->i_flags & XFS_IRECLAIM) {
1056                                 ip = ip->i_mnext;
1057                                 continue;
1058                         }
1059                         if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0) {
1060                                 ip = ip->i_mnext;
1061                         } else if ((xfs_ipincount(ip) == 0) &&
1062                                     xfs_iflock_nowait(ip)) {
1063                                 IPOINTER_INSERT(ip, mp);
1064
1065                                 xfs_finish_reclaim(ip, 1,
1066                                                 XFS_IFLUSH_DELWRI_ELSE_ASYNC);
1067
1068                                 XFS_MOUNT_ILOCK(mp);
1069                                 mount_locked = B_TRUE;
1070                                 IPOINTER_REMOVE(ip, mp);
1071                         } else {
1072                                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1073                                 ip = ip->i_mnext;
1074                         }
1075                         continue;
1076                 }
1077
1078                 if (VN_BAD(vp)) {
1079                         ip = ip->i_mnext;
1080                         continue;
1081                 }
1082
1083                 if (XFS_FORCED_SHUTDOWN(mp) && !(flags & SYNC_CLOSE)) {
1084                         XFS_MOUNT_IUNLOCK(mp);
1085                         kmem_free(ipointer, sizeof(xfs_iptr_t));
1086                         return 0;
1087                 }
1088
1089                 /*
1090                  * If this is just vfs_sync() or pflushd() calling
1091                  * then we can skip inodes for which it looks like
1092                  * there is nothing to do.  Since we don't have the
1093                  * inode locked this is racy, but these are periodic
1094                  * calls so it doesn't matter.  For the others we want
1095                  * to know for sure, so we at least try to lock them.
1096                  */
1097                 if (flags & SYNC_BDFLUSH) {
1098                         if (((ip->i_itemp == NULL) ||
1099                              !(ip->i_itemp->ili_format.ilf_fields &
1100                                XFS_ILOG_ALL)) &&
1101                             (ip->i_update_core == 0)) {
1102                                 ip = ip->i_mnext;
1103                                 continue;
1104                         }
1105                 }
1106
1107                 /*
1108                  * Try to lock without sleeping.  We're out of order with
1109                  * the inode list lock here, so if we fail we need to drop
1110                  * the mount lock and try again.  If we're called from
1111                  * bdflush() here, then don't bother.
1112                  *
1113                  * The inode lock here actually coordinates with the
1114                  * almost spurious inode lock in xfs_ireclaim() to prevent
1115                  * the vnode we handle here without a reference from
1116                  * being freed while we reference it.  If we lock the inode
1117                  * while it's on the mount list here, then the spurious inode
1118                  * lock in xfs_ireclaim() after the inode is pulled from
1119                  * the mount list will sleep until we release it here.
1120                  * This keeps the vnode from being freed while we reference
1121                  * it.
1122                  */
1123                 if (xfs_ilock_nowait(ip, lock_flags) == 0) {
1124                         if ((flags & SYNC_BDFLUSH) || (vp == NULL)) {
1125                                 ip = ip->i_mnext;
1126                                 continue;
1127                         }
1128
1129                         vp = vn_grab(vp);
1130                         if (vp == NULL) {
1131                                 ip = ip->i_mnext;
1132                                 continue;
1133                         }
1134
1135                         IPOINTER_INSERT(ip, mp);
1136                         xfs_ilock(ip, lock_flags);
1137
1138                         ASSERT(vp == XFS_ITOV(ip));
1139                         ASSERT(ip->i_mount == mp);
1140
1141                         vnode_refed = B_TRUE;
1142                 }
1143
1144                 /* From here on in the loop we may have a marker record
1145                  * in the inode list.
1146                  */
1147
1148                 if ((flags & SYNC_CLOSE)  && (vp != NULL)) {
1149                         /*
1150                          * This is the shutdown case.  We just need to
1151                          * flush and invalidate all the pages associated
1152                          * with the inode.  Drop the inode lock since
1153                          * we can't hold it across calls to the buffer
1154                          * cache.
1155                          *
1156                          * We don't set the VREMAPPING bit in the vnode
1157                          * here, because we don't hold the vnode lock
1158                          * exclusively.  It doesn't really matter, though,
1159                          * because we only come here when we're shutting
1160                          * down anyway.
1161                          */
1162                         xfs_iunlock(ip, XFS_ILOCK_SHARED);
1163
1164                         if (XFS_FORCED_SHUTDOWN(mp)) {
1165                                 VOP_TOSS_PAGES(vp, 0, -1, FI_REMAPF);
1166                         } else {
1167                                 VOP_FLUSHINVAL_PAGES(vp, 0, -1, FI_REMAPF);
1168                         }
1169
1170                         xfs_ilock(ip, XFS_ILOCK_SHARED);
1171
1172                 } else if ((flags & SYNC_DELWRI) && (vp != NULL)) {
1173                         if (VN_DIRTY(vp)) {
1174                                 /* We need to have dropped the lock here,
1175                                  * so insert a marker if we have not already
1176                                  * done so.
1177                                  */
1178                                 if (mount_locked) {
1179                                         IPOINTER_INSERT(ip, mp);
1180                                 }
1181
1182                                 /*
1183                                  * Drop the inode lock since we can't hold it
1184                                  * across calls to the buffer cache.
1185                                  */
1186                                 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1187                                 VOP_FLUSH_PAGES(vp, (xfs_off_t)0, -1,
1188                                                         fflag, FI_NONE, error);
1189                                 xfs_ilock(ip, XFS_ILOCK_SHARED);
1190                         }
1191
1192                 }
1193
1194                 if (flags & SYNC_BDFLUSH) {
1195                         if ((flags & SYNC_ATTR) &&
1196                             ((ip->i_update_core) ||
1197                              ((ip->i_itemp != NULL) &&
1198                               (ip->i_itemp->ili_format.ilf_fields != 0)))) {
1199
1200                                 /* Insert marker and drop lock if not already
1201                                  * done.
1202                                  */
1203                                 if (mount_locked) {
1204                                         IPOINTER_INSERT(ip, mp);
1205                                 }
1206
1207                                 /*
1208                                  * We don't want the periodic flushing of the
1209                                  * inodes by vfs_sync() to interfere with
1210                                  * I/O to the file, especially read I/O
1211                                  * where it is only the access time stamp
1212                                  * that is being flushed out.  To prevent
1213                                  * long periods where we have both inode
1214                                  * locks held shared here while reading the
1215                                  * inode's buffer in from disk, we drop the
1216                                  * inode lock while reading in the inode
1217                                  * buffer.  We have to release the buffer
1218                                  * and reacquire the inode lock so that they
1219                                  * are acquired in the proper order (inode
1220                                  * locks first).  The buffer will go at the
1221                                  * end of the lru chain, though, so we can
1222                                  * expect it to still be there when we go
1223                                  * for it again in xfs_iflush().
1224                                  */
1225                                 if ((xfs_ipincount(ip) == 0) &&
1226                                     xfs_iflock_nowait(ip)) {
1227
1228                                         xfs_ifunlock(ip);
1229                                         xfs_iunlock(ip, XFS_ILOCK_SHARED);
1230
1231                                         error = xfs_itobp(mp, NULL, ip,
1232                                                           &dip, &bp, 0, 0);
1233                                         if (!error) {
1234                                                 xfs_buf_relse(bp);
1235                                         } else {
1236                                                 /* Bailing out, remove the
1237                                                  * marker and free it.
1238                                                  */
1239                                                 XFS_MOUNT_ILOCK(mp);
1240
1241                                                 IPOINTER_REMOVE(ip, mp);
1242
1243                                                 XFS_MOUNT_IUNLOCK(mp);
1244
1245                                                 ASSERT(!(lock_flags &
1246                                                         XFS_IOLOCK_SHARED));
1247
1248                                                 kmem_free(ipointer,
1249                                                         sizeof(xfs_iptr_t));
1250                                                 return (0);
1251                                         }
1252
1253                                         /*
1254                                          * Since we dropped the inode lock,
1255                                          * the inode may have been reclaimed.
1256                                          * Therefore, we reacquire the mount
1257                                          * lock and check to see if we were the
1258                                          * inode reclaimed. If this happened
1259                                          * then the ipointer marker will no
1260                                          * longer point back at us. In this
1261                                          * case, move ip along to the inode
1262                                          * after the marker, remove the marker
1263                                          * and continue.
1264                                          */
1265                                         XFS_MOUNT_ILOCK(mp);
1266                                         mount_locked = B_TRUE;
1267
1268                                         if (ip != ipointer->ip_mprev) {
1269                                                 IPOINTER_REMOVE(ip, mp);
1270
1271                                                 ASSERT(!vnode_refed);
1272                                                 ASSERT(!(lock_flags &
1273                                                         XFS_IOLOCK_SHARED));
1274                                                 continue;
1275                                         }
1276
1277                                         ASSERT(ip->i_mount == mp);
1278
1279                                         if (xfs_ilock_nowait(ip,
1280                                                     XFS_ILOCK_SHARED) == 0) {
1281                                                 ASSERT(ip->i_mount == mp);
1282                                                 /*
1283                                                  * We failed to reacquire
1284                                                  * the inode lock without
1285                                                  * sleeping, so just skip
1286                                                  * the inode for now.  We
1287                                                  * clear the ILOCK bit from
1288                                                  * the lock_flags so that we
1289                                                  * won't try to drop a lock
1290                                                  * we don't hold below.
1291                                                  */
1292                                                 lock_flags &= ~XFS_ILOCK_SHARED;
1293                                                 IPOINTER_REMOVE(ip_next, mp);
1294                                         } else if ((xfs_ipincount(ip) == 0) &&
1295                                                    xfs_iflock_nowait(ip)) {
1296                                                 ASSERT(ip->i_mount == mp);
1297                                                 /*
1298                                                  * Since this is vfs_sync()
1299                                                  * calling we only flush the
1300                                                  * inode out if we can lock
1301                                                  * it without sleeping and
1302                                                  * it is not pinned.  Drop
1303                                                  * the mount lock here so
1304                                                  * that we don't hold it for
1305                                                  * too long. We already have
1306                                                  * a marker in the list here.
1307                                                  */
1308                                                 XFS_MOUNT_IUNLOCK(mp);
1309                                                 mount_locked = B_FALSE;
1310                                                 error = xfs_iflush(ip,
1311                                                            XFS_IFLUSH_DELWRI);
1312                                         } else {
1313                                                 ASSERT(ip->i_mount == mp);
1314                                                 IPOINTER_REMOVE(ip_next, mp);
1315                                         }
1316                                 }
1317
1318                         }
1319
1320                 } else {
1321                         if ((flags & SYNC_ATTR) &&
1322                             ((ip->i_update_core) ||
1323                              ((ip->i_itemp != NULL) &&
1324                               (ip->i_itemp->ili_format.ilf_fields != 0)))) {
1325                                 if (mount_locked) {
1326                                         IPOINTER_INSERT(ip, mp);
1327                                 }
1328
1329                                 if (flags & SYNC_WAIT) {
1330                                         xfs_iflock(ip);
1331                                         error = xfs_iflush(ip,
1332                                                            XFS_IFLUSH_SYNC);
1333                                 } else {
1334                                         /*
1335                                          * If we can't acquire the flush
1336                                          * lock, then the inode is already
1337                                          * being flushed so don't bother
1338                                          * waiting.  If we can lock it then
1339                                          * do a delwri flush so we can
1340                                          * combine multiple inode flushes
1341                                          * in each disk write.
1342                                          */
1343                                         if (xfs_iflock_nowait(ip)) {
1344                                                 error = xfs_iflush(ip,
1345                                                            XFS_IFLUSH_DELWRI);
1346                                         }
1347                                         else if (bypassed)
1348                                                 (*bypassed)++;
1349                                 }
1350                         }
1351                 }
1352
1353                 if (lock_flags != 0) {
1354                         xfs_iunlock(ip, lock_flags);
1355                 }
1356
1357                 if (vnode_refed) {
1358                         /*
1359                          * If we had to take a reference on the vnode
1360                          * above, then wait until after we've unlocked
1361                          * the inode to release the reference.  This is
1362                          * because we can be already holding the inode
1363                          * lock when VN_RELE() calls xfs_inactive().
1364                          *
1365                          * Make sure to drop the mount lock before calling
1366                          * VN_RELE() so that we don't trip over ourselves if
1367                          * we have to go for the mount lock again in the
1368                          * inactive code.
1369                          */
1370                         if (mount_locked) {
1371                                 IPOINTER_INSERT(ip, mp);
1372                         }
1373
1374                         VN_RELE(vp);
1375
1376                         vnode_refed = B_FALSE;
1377                 }
1378
1379                 if (error) {
1380                         last_error = error;
1381                 }
1382
1383                 /*
1384                  * bail out if the filesystem is corrupted.
1385                  */
1386                 if (error == EFSCORRUPTED)  {
1387                         if (!mount_locked) {
1388                                 XFS_MOUNT_ILOCK(mp);
1389                                 IPOINTER_REMOVE(ip, mp);
1390                         }
1391                         XFS_MOUNT_IUNLOCK(mp);
1392                         ASSERT(ipointer_in == B_FALSE);
1393                         kmem_free(ipointer, sizeof(xfs_iptr_t));
1394                         return XFS_ERROR(error);
1395                 }
1396
1397                 /* Let other threads have a chance at the mount lock
1398                  * if we have looped many times without dropping the
1399                  * lock.
1400                  */
1401                 if ((++preempt & XFS_PREEMPT_MASK) == 0) {
1402                         if (mount_locked) {
1403                                 IPOINTER_INSERT(ip, mp);
1404                         }
1405                 }
1406
1407                 if (mount_locked == B_FALSE) {
1408                         XFS_MOUNT_ILOCK(mp);
1409                         mount_locked = B_TRUE;
1410                         IPOINTER_REMOVE(ip, mp);
1411                         continue;
1412                 }
1413
1414                 ASSERT(ipointer_in == B_FALSE);
1415                 ip = ip->i_mnext;
1416
1417         } while (ip != mp->m_inodes);
1418
1419         XFS_MOUNT_IUNLOCK(mp);
1420
1421         ASSERT(ipointer_in == B_FALSE);
1422
1423         kmem_free(ipointer, sizeof(xfs_iptr_t));
1424         return XFS_ERROR(last_error);
1425 }
1426
1427 /*
1428  * xfs sync routine for internal use
1429  *
1430  * This routine supports all of the flags defined for the generic VFS_SYNC
1431  * interface as explained above under xfs_sync.  In the interests of not
1432  * changing interfaces within the 6.5 family, additional internally-
1433  * required functions are specified within a separate xflags parameter,
1434  * only available by calling this routine.
1435  *
1436  */
1437 int
1438 xfs_syncsub(
1439         xfs_mount_t     *mp,
1440         int             flags,
1441         int             xflags,
1442         int             *bypassed)
1443 {
1444         int             error = 0;
1445         int             last_error = 0;
1446         uint            log_flags = XFS_LOG_FORCE;
1447         xfs_buf_t       *bp;
1448         xfs_buf_log_item_t      *bip;
1449
1450         /*
1451          * Sync out the log.  This ensures that the log is periodically
1452          * flushed even if there is not enough activity to fill it up.
1453          */
1454         if (flags & SYNC_WAIT)
1455                 log_flags |= XFS_LOG_SYNC;
1456
1457         xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1458
1459         if (flags & (SYNC_ATTR|SYNC_DELWRI)) {
1460                 if (flags & SYNC_BDFLUSH)
1461                         xfs_finish_reclaim_all(mp, 1);
1462                 else
1463                         error = xfs_sync_inodes(mp, flags, xflags, bypassed);
1464         }
1465
1466         /*
1467          * Flushing out dirty data above probably generated more
1468          * log activity, so if this isn't vfs_sync() then flush
1469          * the log again.
1470          */
1471         if (flags & SYNC_DELWRI) {
1472                 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1473         }
1474
1475         if (flags & SYNC_FSDATA) {
1476                 /*
1477                  * If this is vfs_sync() then only sync the superblock
1478                  * if we can lock it without sleeping and it is not pinned.
1479                  */
1480                 if (flags & SYNC_BDFLUSH) {
1481                         bp = xfs_getsb(mp, XFS_BUF_TRYLOCK);
1482                         if (bp != NULL) {
1483                                 bip = XFS_BUF_FSPRIVATE(bp,xfs_buf_log_item_t*);
1484                                 if ((bip != NULL) &&
1485                                     xfs_buf_item_dirty(bip)) {
1486                                         if (!(XFS_BUF_ISPINNED(bp))) {
1487                                                 XFS_BUF_ASYNC(bp);
1488                                                 error = xfs_bwrite(mp, bp);
1489                                         } else {
1490                                                 xfs_buf_relse(bp);
1491                                         }
1492                                 } else {
1493                                         xfs_buf_relse(bp);
1494                                 }
1495                         }
1496                 } else {
1497                         bp = xfs_getsb(mp, 0);
1498                         /*
1499                          * If the buffer is pinned then push on the log so
1500                          * we won't get stuck waiting in the write for
1501                          * someone, maybe ourselves, to flush the log.
1502                          * Even though we just pushed the log above, we
1503                          * did not have the superblock buffer locked at
1504                          * that point so it can become pinned in between
1505                          * there and here.
1506                          */
1507                         if (XFS_BUF_ISPINNED(bp))
1508                                 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
1509                         if (flags & SYNC_WAIT)
1510                                 XFS_BUF_UNASYNC(bp);
1511                         else
1512                                 XFS_BUF_ASYNC(bp);
1513                         error = xfs_bwrite(mp, bp);
1514                 }
1515                 if (error) {
1516                         last_error = error;
1517                 }
1518         }
1519
1520         /*
1521          * If this is the periodic sync, then kick some entries out of
1522          * the reference cache.  This ensures that idle entries are
1523          * eventually kicked out of the cache.
1524          */
1525         if (flags & SYNC_REFCACHE) {
1526                 if (flags & SYNC_WAIT)
1527                         xfs_refcache_purge_mp(mp);
1528                 else
1529                         xfs_refcache_purge_some(mp);
1530         }
1531
1532         /*
1533          * Now check to see if the log needs a "dummy" transaction.
1534          */
1535
1536         if (!(flags & SYNC_REMOUNT) && xfs_log_need_covered(mp)) {
1537                 xfs_trans_t *tp;
1538                 xfs_inode_t *ip;
1539
1540                 /*
1541                  * Put a dummy transaction in the log to tell
1542                  * recovery that all others are OK.
1543                  */
1544                 tp = xfs_trans_alloc(mp, XFS_TRANS_DUMMY1);
1545                 if ((error = xfs_trans_reserve(tp, 0,
1546                                 XFS_ICHANGE_LOG_RES(mp),
1547                                 0, 0, 0)))  {
1548                         xfs_trans_cancel(tp, 0);
1549                         return error;
1550                 }
1551
1552                 ip = mp->m_rootip;
1553                 xfs_ilock(ip, XFS_ILOCK_EXCL);
1554
1555                 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1556                 xfs_trans_ihold(tp, ip);
1557                 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1558                 error = xfs_trans_commit(tp, 0, NULL);
1559                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1560                 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1561         }
1562
1563         /*
1564          * When shutting down, we need to insure that the AIL is pushed
1565          * to disk or the filesystem can appear corrupt from the PROM.
1566          */
1567         if ((flags & (SYNC_CLOSE|SYNC_WAIT)) == (SYNC_CLOSE|SYNC_WAIT)) {
1568                 XFS_bflush(mp->m_ddev_targp);
1569                 if (mp->m_rtdev_targp) {
1570                         XFS_bflush(mp->m_rtdev_targp);
1571                 }
1572         }
1573
1574         return XFS_ERROR(last_error);
1575 }
1576
1577 /*
1578  * xfs_vget - called by DMAPI and NFSD to get vnode from file handle
1579  */
1580 STATIC int
1581 xfs_vget(
1582         bhv_desc_t      *bdp,
1583         vnode_t         **vpp,
1584         fid_t           *fidp)
1585 {
1586         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
1587         xfs_fid_t       *xfid = (struct xfs_fid *)fidp;
1588         xfs_inode_t     *ip;
1589         int             error;
1590         xfs_ino_t       ino;
1591         unsigned int    igen;
1592
1593         /*
1594          * Invalid.  Since handles can be created in user space and passed in
1595          * via gethandle(), this is not cause for a panic.
1596          */
1597         if (xfid->xfs_fid_len != sizeof(*xfid) - sizeof(xfid->xfs_fid_len))
1598                 return XFS_ERROR(EINVAL);
1599
1600         ino  = xfid->xfs_fid_ino;
1601         igen = xfid->xfs_fid_gen;
1602
1603         /*
1604          * NFS can sometimes send requests for ino 0.  Fail them gracefully.
1605          */
1606         if (ino == 0)
1607                 return XFS_ERROR(ESTALE);
1608
1609         error = xfs_iget(mp, NULL, ino, 0, XFS_ILOCK_SHARED, &ip, 0);
1610         if (error) {
1611                 *vpp = NULL;
1612                 return error;
1613         }
1614
1615         if (ip == NULL) {
1616                 *vpp = NULL;
1617                 return XFS_ERROR(EIO);
1618         }
1619
1620         if (ip->i_d.di_mode == 0 || ip->i_d.di_gen != igen) {
1621                 xfs_iput_new(ip, XFS_ILOCK_SHARED);
1622                 *vpp = NULL;
1623                 return XFS_ERROR(ENOENT);
1624         }
1625
1626         *vpp = XFS_ITOV(ip);
1627         xfs_iunlock(ip, XFS_ILOCK_SHARED);
1628         return 0;
1629 }
1630
1631
1632 #define MNTOPT_LOGBUFS  "logbufs"       /* number of XFS log buffers */
1633 #define MNTOPT_LOGBSIZE "logbsize"      /* size of XFS log buffers */
1634 #define MNTOPT_LOGDEV   "logdev"        /* log device */
1635 #define MNTOPT_RTDEV    "rtdev"         /* realtime I/O device */
1636 #define MNTOPT_BIOSIZE  "biosize"       /* log2 of preferred buffered io size */
1637 #define MNTOPT_WSYNC    "wsync"         /* safe-mode nfs compatible mount */
1638 #define MNTOPT_INO64    "ino64"         /* force inodes into 64-bit range */
1639 #define MNTOPT_NOALIGN  "noalign"       /* turn off stripe alignment */
1640 #define MNTOPT_SWALLOC  "swalloc"       /* turn on stripe width allocation */
1641 #define MNTOPT_SUNIT    "sunit"         /* data volume stripe unit */
1642 #define MNTOPT_SWIDTH   "swidth"        /* data volume stripe width */
1643 #define MNTOPT_NOUUID   "nouuid"        /* ignore filesystem UUID */
1644 #define MNTOPT_MTPT     "mtpt"          /* filesystem mount point */
1645 #define MNTOPT_GRPID    "grpid"         /* group-ID from parent directory */
1646 #define MNTOPT_NOGRPID  "nogrpid"       /* group-ID from current process */
1647 #define MNTOPT_BSDGROUPS    "bsdgroups"    /* group-ID from parent directory */
1648 #define MNTOPT_SYSVGROUPS   "sysvgroups"   /* group-ID from current process */
1649 #define MNTOPT_ALLOCSIZE    "allocsize"    /* preferred allocation size */
1650 #define MNTOPT_IHASHSIZE    "ihashsize"    /* size of inode hash table */
1651 #define MNTOPT_NORECOVERY   "norecovery"   /* don't run XFS recovery */
1652 #define MNTOPT_BARRIER  "barrier"       /* use writer barriers for log write and
1653                                          * unwritten extent conversion */
1654 #define MNTOPT_NOBARRIER "nobarrier"    /* .. disable */
1655 #define MNTOPT_OSYNCISOSYNC "osyncisosync" /* o_sync is REALLY o_sync */
1656 #define MNTOPT_64BITINODE   "inode64"   /* inodes can be allocated anywhere */
1657 #define MNTOPT_IKEEP    "ikeep"         /* do not free empty inode clusters */
1658 #define MNTOPT_NOIKEEP  "noikeep"       /* free empty inode clusters */
1659 #define MNTOPT_LARGEIO     "largeio"    /* report large I/O sizes in stat() */
1660 #define MNTOPT_NOLARGEIO   "nolargeio"  /* do not report large I/O sizes
1661                                          * in stat(). */
1662 #define MNTOPT_ATTR2    "attr2"         /* do use attr2 attribute format */
1663 #define MNTOPT_NOATTR2  "noattr2"       /* do not use attr2 attribute format */
1664
1665 STATIC unsigned long
1666 suffix_strtoul(const char *cp, char **endp, unsigned int base)
1667 {
1668         int     last, shift_left_factor = 0;
1669         char    *value = (char *)cp;
1670
1671         last = strlen(value) - 1;
1672         if (value[last] == 'K' || value[last] == 'k') {
1673                 shift_left_factor = 10;
1674                 value[last] = '\0';
1675         }
1676         if (value[last] == 'M' || value[last] == 'm') {
1677                 shift_left_factor = 20;
1678                 value[last] = '\0';
1679         }
1680         if (value[last] == 'G' || value[last] == 'g') {
1681                 shift_left_factor = 30;
1682                 value[last] = '\0';
1683         }
1684
1685         return simple_strtoul(cp, endp, base) << shift_left_factor;
1686 }
1687
1688 STATIC int
1689 xfs_parseargs(
1690         struct bhv_desc         *bhv,
1691         char                    *options,
1692         struct xfs_mount_args   *args,
1693         int                     update)
1694 {
1695         struct vfs              *vfsp = bhvtovfs(bhv);
1696         char                    *this_char, *value, *eov;
1697         int                     dsunit, dswidth, vol_dsunit, vol_dswidth;
1698         int                     iosize;
1699
1700         args->flags2 |= XFSMNT2_COMPAT_IOSIZE;
1701         args->flags |= XFSMNT_IDELETE;
1702
1703         if (!options)
1704                 goto done;
1705
1706         iosize = dsunit = dswidth = vol_dsunit = vol_dswidth = 0;
1707
1708         while ((this_char = strsep(&options, ",")) != NULL) {
1709                 if (!*this_char)
1710                         continue;
1711                 if ((value = strchr(this_char, '=')) != NULL)
1712                         *value++ = 0;
1713
1714                 if (!strcmp(this_char, MNTOPT_LOGBUFS)) {
1715                         if (!value || !*value) {
1716                                 printk("XFS: %s option requires an argument\n",
1717                                         this_char);
1718                                 return EINVAL;
1719                         }
1720                         args->logbufs = simple_strtoul(value, &eov, 10);
1721                 } else if (!strcmp(this_char, MNTOPT_LOGBSIZE)) {
1722                         if (!value || !*value) {
1723                                 printk("XFS: %s option requires an argument\n",
1724                                         this_char);
1725                                 return EINVAL;
1726                         }
1727                         args->logbufsize = suffix_strtoul(value, &eov, 10);
1728                 } else if (!strcmp(this_char, MNTOPT_LOGDEV)) {
1729                         if (!value || !*value) {
1730                                 printk("XFS: %s option requires an argument\n",
1731                                         this_char);
1732                                 return EINVAL;
1733                         }
1734                         strncpy(args->logname, value, MAXNAMELEN);
1735                 } else if (!strcmp(this_char, MNTOPT_MTPT)) {
1736                         if (!value || !*value) {
1737                                 printk("XFS: %s option requires an argument\n",
1738                                         this_char);
1739                                 return EINVAL;
1740                         }
1741                         strncpy(args->mtpt, value, MAXNAMELEN);
1742                 } else if (!strcmp(this_char, MNTOPT_RTDEV)) {
1743                         if (!value || !*value) {
1744                                 printk("XFS: %s option requires an argument\n",
1745                                         this_char);
1746                                 return EINVAL;
1747                         }
1748                         strncpy(args->rtname, value, MAXNAMELEN);
1749                 } else if (!strcmp(this_char, MNTOPT_BIOSIZE)) {
1750                         if (!value || !*value) {
1751                                 printk("XFS: %s option requires an argument\n",
1752                                         this_char);
1753                                 return EINVAL;
1754                         }
1755                         iosize = simple_strtoul(value, &eov, 10);
1756                         args->flags |= XFSMNT_IOSIZE;
1757                         args->iosizelog = (uint8_t) iosize;
1758                 } else if (!strcmp(this_char, MNTOPT_ALLOCSIZE)) {
1759                         if (!value || !*value) {
1760                                 printk("XFS: %s option requires an argument\n",
1761                                         this_char);
1762                                 return EINVAL;
1763                         }
1764                         iosize = suffix_strtoul(value, &eov, 10);
1765                         args->flags |= XFSMNT_IOSIZE;
1766                         args->iosizelog = ffs(iosize) - 1;
1767                 } else if (!strcmp(this_char, MNTOPT_IHASHSIZE)) {
1768                         if (!value || !*value) {
1769                                 printk("XFS: %s option requires an argument\n",
1770                                         this_char);
1771                                 return EINVAL;
1772                         }
1773                         args->flags |= XFSMNT_IHASHSIZE;
1774                         args->ihashsize = simple_strtoul(value, &eov, 10);
1775                 } else if (!strcmp(this_char, MNTOPT_GRPID) ||
1776                            !strcmp(this_char, MNTOPT_BSDGROUPS)) {
1777                         vfsp->vfs_flag |= VFS_GRPID;
1778                 } else if (!strcmp(this_char, MNTOPT_NOGRPID) ||
1779                            !strcmp(this_char, MNTOPT_SYSVGROUPS)) {
1780                         vfsp->vfs_flag &= ~VFS_GRPID;
1781                 } else if (!strcmp(this_char, MNTOPT_WSYNC)) {
1782                         args->flags |= XFSMNT_WSYNC;
1783                 } else if (!strcmp(this_char, MNTOPT_OSYNCISOSYNC)) {
1784                         args->flags |= XFSMNT_OSYNCISOSYNC;
1785                 } else if (!strcmp(this_char, MNTOPT_NORECOVERY)) {
1786                         args->flags |= XFSMNT_NORECOVERY;
1787                 } else if (!strcmp(this_char, MNTOPT_INO64)) {
1788                         args->flags |= XFSMNT_INO64;
1789 #if !XFS_BIG_INUMS
1790                         printk("XFS: %s option not allowed on this system\n",
1791                                 this_char);
1792                         return EINVAL;
1793 #endif
1794                 } else if (!strcmp(this_char, MNTOPT_NOALIGN)) {
1795                         args->flags |= XFSMNT_NOALIGN;
1796                 } else if (!strcmp(this_char, MNTOPT_SWALLOC)) {
1797                         args->flags |= XFSMNT_SWALLOC;
1798                 } else if (!strcmp(this_char, MNTOPT_SUNIT)) {
1799                         if (!value || !*value) {
1800                                 printk("XFS: %s option requires an argument\n",
1801                                         this_char);
1802                                 return EINVAL;
1803                         }
1804                         dsunit = simple_strtoul(value, &eov, 10);
1805                 } else if (!strcmp(this_char, MNTOPT_SWIDTH)) {
1806                         if (!value || !*value) {
1807                                 printk("XFS: %s option requires an argument\n",
1808                                         this_char);
1809                                 return EINVAL;
1810                         }
1811                         dswidth = simple_strtoul(value, &eov, 10);
1812                 } else if (!strcmp(this_char, MNTOPT_64BITINODE)) {
1813                         args->flags &= ~XFSMNT_32BITINODES;
1814 #if !XFS_BIG_INUMS
1815                         printk("XFS: %s option not allowed on this system\n",
1816                                 this_char);
1817                         return EINVAL;
1818 #endif
1819                 } else if (!strcmp(this_char, MNTOPT_NOUUID)) {
1820                         args->flags |= XFSMNT_NOUUID;
1821                 } else if (!strcmp(this_char, MNTOPT_BARRIER)) {
1822                         args->flags |= XFSMNT_BARRIER;
1823                 } else if (!strcmp(this_char, MNTOPT_NOBARRIER)) {
1824                         args->flags &= ~XFSMNT_BARRIER;
1825                 } else if (!strcmp(this_char, MNTOPT_IKEEP)) {
1826                         args->flags &= ~XFSMNT_IDELETE;
1827                 } else if (!strcmp(this_char, MNTOPT_NOIKEEP)) {
1828                         args->flags |= XFSMNT_IDELETE;
1829                 } else if (!strcmp(this_char, MNTOPT_LARGEIO)) {
1830                         args->flags2 &= ~XFSMNT2_COMPAT_IOSIZE;
1831                 } else if (!strcmp(this_char, MNTOPT_NOLARGEIO)) {
1832                         args->flags2 |= XFSMNT2_COMPAT_IOSIZE;
1833                 } else if (!strcmp(this_char, MNTOPT_ATTR2)) {
1834                         args->flags |= XFSMNT_ATTR2;
1835                 } else if (!strcmp(this_char, MNTOPT_NOATTR2)) {
1836                         args->flags &= ~XFSMNT_ATTR2;
1837                 } else if (!strcmp(this_char, "osyncisdsync")) {
1838                         /* no-op, this is now the default */
1839 printk("XFS: osyncisdsync is now the default, option is deprecated.\n");
1840                 } else if (!strcmp(this_char, "irixsgid")) {
1841 printk("XFS: irixsgid is now a sysctl(2) variable, option is deprecated.\n");
1842                 } else {
1843                         printk("XFS: unknown mount option [%s].\n", this_char);
1844                         return EINVAL;
1845                 }
1846         }
1847
1848         if (args->flags & XFSMNT_NORECOVERY) {
1849                 if ((vfsp->vfs_flag & VFS_RDONLY) == 0) {
1850                         printk("XFS: no-recovery mounts must be read-only.\n");
1851                         return EINVAL;
1852                 }
1853         }
1854
1855         if ((args->flags & XFSMNT_NOALIGN) && (dsunit || dswidth)) {
1856                 printk(
1857         "XFS: sunit and swidth options incompatible with the noalign option\n");
1858                 return EINVAL;
1859         }
1860
1861         if ((dsunit && !dswidth) || (!dsunit && dswidth)) {
1862                 printk("XFS: sunit and swidth must be specified together\n");
1863                 return EINVAL;
1864         }
1865
1866         if (dsunit && (dswidth % dsunit != 0)) {
1867                 printk(
1868         "XFS: stripe width (%d) must be a multiple of the stripe unit (%d)\n",
1869                         dswidth, dsunit);
1870                 return EINVAL;
1871         }
1872
1873         if ((args->flags & XFSMNT_NOALIGN) != XFSMNT_NOALIGN) {
1874                 if (dsunit) {
1875                         args->sunit = dsunit;
1876                         args->flags |= XFSMNT_RETERR;
1877                 } else {
1878                         args->sunit = vol_dsunit;
1879                 }
1880                 dswidth ? (args->swidth = dswidth) :
1881                           (args->swidth = vol_dswidth);
1882         } else {
1883                 args->sunit = args->swidth = 0;
1884         }
1885
1886 done:
1887         if (args->flags & XFSMNT_32BITINODES)
1888                 vfsp->vfs_flag |= VFS_32BITINODES;
1889         if (args->flags2)
1890                 args->flags |= XFSMNT_FLAGS2;
1891         return 0;
1892 }
1893
1894 STATIC int
1895 xfs_showargs(
1896         struct bhv_desc         *bhv,
1897         struct seq_file         *m)
1898 {
1899         static struct proc_xfs_info {
1900                 int     flag;
1901                 char    *str;
1902         } xfs_info[] = {
1903                 /* the few simple ones we can get from the mount struct */
1904                 { XFS_MOUNT_WSYNC,              "," MNTOPT_WSYNC },
1905                 { XFS_MOUNT_INO64,              "," MNTOPT_INO64 },
1906                 { XFS_MOUNT_NOALIGN,            "," MNTOPT_NOALIGN },
1907                 { XFS_MOUNT_SWALLOC,            "," MNTOPT_SWALLOC },
1908                 { XFS_MOUNT_NOUUID,             "," MNTOPT_NOUUID },
1909                 { XFS_MOUNT_NORECOVERY,         "," MNTOPT_NORECOVERY },
1910                 { XFS_MOUNT_OSYNCISOSYNC,       "," MNTOPT_OSYNCISOSYNC },
1911                 { 0, NULL }
1912         };
1913         struct proc_xfs_info    *xfs_infop;
1914         struct xfs_mount        *mp = XFS_BHVTOM(bhv);
1915         struct vfs              *vfsp = XFS_MTOVFS(mp);
1916
1917         for (xfs_infop = xfs_info; xfs_infop->flag; xfs_infop++) {
1918                 if (mp->m_flags & xfs_infop->flag)
1919                         seq_puts(m, xfs_infop->str);
1920         }
1921
1922         if (mp->m_flags & XFS_MOUNT_IHASHSIZE)
1923                 seq_printf(m, "," MNTOPT_IHASHSIZE "=%d", mp->m_ihsize);
1924
1925         if (mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)
1926                 seq_printf(m, "," MNTOPT_ALLOCSIZE "=%dk",
1927                                 (int)(1 << mp->m_writeio_log) >> 10);
1928
1929         if (mp->m_logbufs > 0)
1930                 seq_printf(m, "," MNTOPT_LOGBUFS "=%d", mp->m_logbufs);
1931         if (mp->m_logbsize > 0)
1932                 seq_printf(m, "," MNTOPT_LOGBSIZE "=%dk", mp->m_logbsize >> 10);
1933
1934         if (mp->m_logname)
1935                 seq_printf(m, "," MNTOPT_LOGDEV "=%s", mp->m_logname);
1936         if (mp->m_rtname)
1937                 seq_printf(m, "," MNTOPT_RTDEV "=%s", mp->m_rtname);
1938
1939         if (mp->m_dalign > 0)
1940                 seq_printf(m, "," MNTOPT_SUNIT "=%d",
1941                                 (int)XFS_FSB_TO_BB(mp, mp->m_dalign));
1942         if (mp->m_swidth > 0)
1943                 seq_printf(m, "," MNTOPT_SWIDTH "=%d",
1944                                 (int)XFS_FSB_TO_BB(mp, mp->m_swidth));
1945
1946         if (!(mp->m_flags & XFS_MOUNT_IDELETE))
1947                 seq_printf(m, "," MNTOPT_IKEEP);
1948         if (!(mp->m_flags & XFS_MOUNT_COMPAT_IOSIZE))
1949                 seq_printf(m, "," MNTOPT_LARGEIO);
1950         if (mp->m_flags & XFS_MOUNT_BARRIER)
1951                 seq_printf(m, "," MNTOPT_BARRIER);
1952
1953         if (!(vfsp->vfs_flag & VFS_32BITINODES))
1954                 seq_printf(m, "," MNTOPT_64BITINODE);
1955         if (vfsp->vfs_flag & VFS_GRPID)
1956                 seq_printf(m, "," MNTOPT_GRPID);
1957
1958         return 0;
1959 }
1960
1961 STATIC void
1962 xfs_freeze(
1963         bhv_desc_t      *bdp)
1964 {
1965         xfs_mount_t     *mp = XFS_BHVTOM(bdp);
1966
1967         while (atomic_read(&mp->m_active_trans) > 0)
1968                 delay(100);
1969
1970         /* Push the superblock and write an unmount record */
1971         xfs_log_unmount_write(mp);
1972         xfs_unmountfs_writesb(mp);
1973         xfs_fs_log_dummy(mp);
1974 }
1975
1976
1977 vfsops_t xfs_vfsops = {
1978         BHV_IDENTITY_INIT(VFS_BHV_XFS,VFS_POSITION_XFS),
1979         .vfs_parseargs          = xfs_parseargs,
1980         .vfs_showargs           = xfs_showargs,
1981         .vfs_mount              = xfs_mount,
1982         .vfs_unmount            = xfs_unmount,
1983         .vfs_mntupdate          = xfs_mntupdate,
1984         .vfs_root               = xfs_root,
1985         .vfs_statvfs            = xfs_statvfs,
1986         .vfs_sync               = xfs_sync,
1987         .vfs_vget               = xfs_vget,
1988         .vfs_dmapiops           = (vfs_dmapiops_t)fs_nosys,
1989         .vfs_quotactl           = (vfs_quotactl_t)fs_nosys,
1990         .vfs_init_vnode         = xfs_initialize_vnode,
1991         .vfs_force_shutdown     = xfs_do_force_shutdown,
1992         .vfs_freeze             = xfs_freeze,
1993 };