2 * XFS filesystem operations.
4 * Copyright (c) 2000-2005 Silicon Graphics, Inc. All Rights Reserved.
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms of version 2 of the GNU General Public License as
8 * published by the Free Software Foundation.
10 * This program is distributed in the hope that it would be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
14 * Further, this software is distributed without any warranty that it is
15 * free of the rightful claim of any third person regarding infringement
16 * or the like. Any license provided herein, whether implied or
17 * otherwise, applies only to this software file. Patent licenses, if
18 * any, provided herein do not apply to combinations of this program with
19 * other software, or any other product whatsoever.
21 * You should have received a copy of the GNU General Public License along
22 * with this program; if not, write the Free Software Foundation, Inc., 59
23 * Temple Place - Suite 330, Boston MA 02111-1307, USA.
25 * Contact information: Silicon Graphics, Inc., 1600 Amphitheatre Pkwy,
26 * Mountain View, CA 94043, or:
30 * For further information regarding this notice, see:
32 * http://oss.sgi.com/projects/GenInfo/SGIGPLNoticeExplan/
36 #include "xfs_macros.h"
37 #include "xfs_types.h"
40 #include "xfs_trans.h"
44 #include "xfs_dmapi.h"
45 #include "xfs_mount.h"
46 #include "xfs_bmap_btree.h"
47 #include "xfs_ialloc_btree.h"
48 #include "xfs_alloc_btree.h"
49 #include "xfs_btree.h"
50 #include "xfs_alloc.h"
51 #include "xfs_ialloc.h"
52 #include "xfs_attr_sf.h"
53 #include "xfs_dir_sf.h"
54 #include "xfs_dir2_sf.h"
55 #include "xfs_dinode.h"
56 #include "xfs_inode_item.h"
57 #include "xfs_inode.h"
59 #include "xfs_error.h"
61 #include "xfs_da_btree.h"
63 #include "xfs_refcache.h"
64 #include "xfs_buf_item.h"
65 #include "xfs_extfree_item.h"
66 #include "xfs_quota.h"
67 #include "xfs_dir2_trace.h"
71 #include "xfs_log_priv.h"
73 STATIC
int xfs_sync(bhv_desc_t
*, int, cred_t
*);
78 extern kmem_zone_t
*xfs_bmap_free_item_zone
;
79 extern kmem_zone_t
*xfs_btree_cur_zone
;
80 extern kmem_zone_t
*xfs_trans_zone
;
81 extern kmem_zone_t
*xfs_buf_item_zone
;
82 extern kmem_zone_t
*xfs_dabuf_zone
;
83 #ifdef XFS_DABUF_DEBUG
84 extern lock_t xfs_dabuf_global_lock
;
85 spinlock_init(&xfs_dabuf_global_lock
, "xfsda");
89 * Initialize all of the zone allocators we use.
91 xfs_bmap_free_item_zone
= kmem_zone_init(sizeof(xfs_bmap_free_item_t
),
92 "xfs_bmap_free_item");
93 xfs_btree_cur_zone
= kmem_zone_init(sizeof(xfs_btree_cur_t
),
95 xfs_inode_zone
= kmem_zone_init(sizeof(xfs_inode_t
), "xfs_inode");
96 xfs_trans_zone
= kmem_zone_init(sizeof(xfs_trans_t
), "xfs_trans");
98 kmem_zone_init(sizeof(xfs_da_state_t
), "xfs_da_state");
99 xfs_dabuf_zone
= kmem_zone_init(sizeof(xfs_dabuf_t
), "xfs_dabuf");
102 * The size of the zone allocated buf log item is the maximum
103 * size possible under XFS. This wastes a little bit of memory,
104 * but it is much faster.
107 kmem_zone_init((sizeof(xfs_buf_log_item_t
) +
108 (((XFS_MAX_BLOCKSIZE
/ XFS_BLI_CHUNK
) /
109 NBWORD
) * sizeof(int))),
111 xfs_efd_zone
= kmem_zone_init((sizeof(xfs_efd_log_item_t
) +
112 ((XFS_EFD_MAX_FAST_EXTENTS
- 1) * sizeof(xfs_extent_t
))),
114 xfs_efi_zone
= kmem_zone_init((sizeof(xfs_efi_log_item_t
) +
115 ((XFS_EFI_MAX_FAST_EXTENTS
- 1) * sizeof(xfs_extent_t
))),
117 xfs_ifork_zone
= kmem_zone_init(sizeof(xfs_ifork_t
), "xfs_ifork");
118 xfs_ili_zone
= kmem_zone_init(sizeof(xfs_inode_log_item_t
), "xfs_ili");
119 xfs_chashlist_zone
= kmem_zone_init(sizeof(xfs_chashlist_t
),
121 xfs_acl_zone_init(xfs_acl_zone
, "xfs_acl");
124 * Allocate global trace buffers.
126 #ifdef XFS_ALLOC_TRACE
127 xfs_alloc_trace_buf
= ktrace_alloc(XFS_ALLOC_TRACE_SIZE
, KM_SLEEP
);
129 #ifdef XFS_BMAP_TRACE
130 xfs_bmap_trace_buf
= ktrace_alloc(XFS_BMAP_TRACE_SIZE
, KM_SLEEP
);
132 #ifdef XFS_BMBT_TRACE
133 xfs_bmbt_trace_buf
= ktrace_alloc(XFS_BMBT_TRACE_SIZE
, KM_SLEEP
);
136 xfs_dir_trace_buf
= ktrace_alloc(XFS_DIR_TRACE_SIZE
, KM_SLEEP
);
138 #ifdef XFS_ATTR_TRACE
139 xfs_attr_trace_buf
= ktrace_alloc(XFS_ATTR_TRACE_SIZE
, KM_SLEEP
);
141 #ifdef XFS_DIR2_TRACE
142 xfs_dir2_trace_buf
= ktrace_alloc(XFS_DIR2_GTRACE_SIZE
, KM_SLEEP
);
147 #if (defined(DEBUG) || defined(INDUCE_IO_ERROR))
148 xfs_error_test_init();
149 #endif /* DEBUG || INDUCE_IO_ERROR */
152 xfs_sysctl_register();
159 extern kmem_zone_t
*xfs_bmap_free_item_zone
;
160 extern kmem_zone_t
*xfs_btree_cur_zone
;
161 extern kmem_zone_t
*xfs_inode_zone
;
162 extern kmem_zone_t
*xfs_trans_zone
;
163 extern kmem_zone_t
*xfs_da_state_zone
;
164 extern kmem_zone_t
*xfs_dabuf_zone
;
165 extern kmem_zone_t
*xfs_efd_zone
;
166 extern kmem_zone_t
*xfs_efi_zone
;
167 extern kmem_zone_t
*xfs_buf_item_zone
;
168 extern kmem_zone_t
*xfs_chashlist_zone
;
170 xfs_cleanup_procfs();
171 xfs_sysctl_unregister();
172 xfs_refcache_destroy();
173 xfs_acl_zone_destroy(xfs_acl_zone
);
175 #ifdef XFS_DIR2_TRACE
176 ktrace_free(xfs_dir2_trace_buf
);
178 #ifdef XFS_ATTR_TRACE
179 ktrace_free(xfs_attr_trace_buf
);
182 ktrace_free(xfs_dir_trace_buf
);
184 #ifdef XFS_BMBT_TRACE
185 ktrace_free(xfs_bmbt_trace_buf
);
187 #ifdef XFS_BMAP_TRACE
188 ktrace_free(xfs_bmap_trace_buf
);
190 #ifdef XFS_ALLOC_TRACE
191 ktrace_free(xfs_alloc_trace_buf
);
194 kmem_cache_destroy(xfs_bmap_free_item_zone
);
195 kmem_cache_destroy(xfs_btree_cur_zone
);
196 kmem_cache_destroy(xfs_inode_zone
);
197 kmem_cache_destroy(xfs_trans_zone
);
198 kmem_cache_destroy(xfs_da_state_zone
);
199 kmem_cache_destroy(xfs_dabuf_zone
);
200 kmem_cache_destroy(xfs_buf_item_zone
);
201 kmem_cache_destroy(xfs_efd_zone
);
202 kmem_cache_destroy(xfs_efi_zone
);
203 kmem_cache_destroy(xfs_ifork_zone
);
204 kmem_cache_destroy(xfs_ili_zone
);
205 kmem_cache_destroy(xfs_chashlist_zone
);
211 * This function fills in xfs_mount_t fields based on mount args.
212 * Note: the superblock has _not_ yet been read in.
217 struct xfs_mount_args
*ap
,
218 struct xfs_mount
*mp
)
220 /* Values are in BBs */
221 if ((ap
->flags
& XFSMNT_NOALIGN
) != XFSMNT_NOALIGN
) {
223 * At this point the superblock has not been read
224 * in, therefore we do not know the block size.
225 * Before the mount call ends we will convert
228 mp
->m_dalign
= ap
->sunit
;
229 mp
->m_swidth
= ap
->swidth
;
232 if (ap
->logbufs
!= -1 &&
233 #if defined(DEBUG) || defined(XLOG_NOLOG)
236 (ap
->logbufs
< XLOG_MIN_ICLOGS
||
237 ap
->logbufs
> XLOG_MAX_ICLOGS
)) {
239 "XFS: invalid logbufs value: %d [not %d-%d]",
240 ap
->logbufs
, XLOG_MIN_ICLOGS
, XLOG_MAX_ICLOGS
);
241 return XFS_ERROR(EINVAL
);
243 mp
->m_logbufs
= ap
->logbufs
;
244 if (ap
->logbufsize
!= -1 &&
245 ap
->logbufsize
!= 16 * 1024 &&
246 ap
->logbufsize
!= 32 * 1024 &&
247 ap
->logbufsize
!= 64 * 1024 &&
248 ap
->logbufsize
!= 128 * 1024 &&
249 ap
->logbufsize
!= 256 * 1024) {
251 "XFS: invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
253 return XFS_ERROR(EINVAL
);
255 mp
->m_ihsize
= ap
->ihashsize
;
256 mp
->m_logbsize
= ap
->logbufsize
;
257 mp
->m_fsname_len
= strlen(ap
->fsname
) + 1;
258 mp
->m_fsname
= kmem_alloc(mp
->m_fsname_len
, KM_SLEEP
);
259 strcpy(mp
->m_fsname
, ap
->fsname
);
261 if (ap
->flags
& XFSMNT_WSYNC
)
262 mp
->m_flags
|= XFS_MOUNT_WSYNC
;
264 if (ap
->flags
& XFSMNT_INO64
) {
265 mp
->m_flags
|= XFS_MOUNT_INO64
;
266 mp
->m_inoadd
= XFS_INO64_OFFSET
;
269 if (ap
->flags
& XFSMNT_NOATIME
)
270 mp
->m_flags
|= XFS_MOUNT_NOATIME
;
272 if (ap
->flags
& XFSMNT_RETERR
)
273 mp
->m_flags
|= XFS_MOUNT_RETERR
;
275 if (ap
->flags
& XFSMNT_NOALIGN
)
276 mp
->m_flags
|= XFS_MOUNT_NOALIGN
;
278 if (ap
->flags
& XFSMNT_SWALLOC
)
279 mp
->m_flags
|= XFS_MOUNT_SWALLOC
;
281 if (ap
->flags
& XFSMNT_OSYNCISOSYNC
)
282 mp
->m_flags
|= XFS_MOUNT_OSYNCISOSYNC
;
284 if (ap
->flags
& XFSMNT_32BITINODES
)
285 mp
->m_flags
|= (XFS_MOUNT_32BITINODES
| XFS_MOUNT_32BITINOOPT
);
287 if (ap
->flags
& XFSMNT_IOSIZE
) {
288 if (ap
->iosizelog
> XFS_MAX_IO_LOG
||
289 ap
->iosizelog
< XFS_MIN_IO_LOG
) {
291 "XFS: invalid log iosize: %d [not %d-%d]",
292 ap
->iosizelog
, XFS_MIN_IO_LOG
,
294 return XFS_ERROR(EINVAL
);
297 mp
->m_flags
|= XFS_MOUNT_DFLT_IOSIZE
;
298 mp
->m_readio_log
= mp
->m_writeio_log
= ap
->iosizelog
;
301 if (ap
->flags
& XFSMNT_IHASHSIZE
)
302 mp
->m_flags
|= XFS_MOUNT_IHASHSIZE
;
304 if (ap
->flags
& XFSMNT_IDELETE
)
305 mp
->m_flags
|= XFS_MOUNT_IDELETE
;
307 if (ap
->flags
& XFSMNT_DIRSYNC
)
308 mp
->m_flags
|= XFS_MOUNT_DIRSYNC
;
311 * no recovery flag requires a read-only mount
313 if (ap
->flags
& XFSMNT_NORECOVERY
) {
314 if (!(vfs
->vfs_flag
& VFS_RDONLY
)) {
316 "XFS: tried to mount a FS read-write without recovery!");
317 return XFS_ERROR(EINVAL
);
319 mp
->m_flags
|= XFS_MOUNT_NORECOVERY
;
322 if (ap
->flags
& XFSMNT_NOUUID
)
323 mp
->m_flags
|= XFS_MOUNT_NOUUID
;
324 if (ap
->flags
& XFSMNT_NOLOGFLUSH
)
325 mp
->m_flags
|= XFS_MOUNT_NOLOGFLUSH
;
331 * This function fills in xfs_mount_t fields based on mount args.
332 * Note: the superblock _has_ now been read in.
337 struct xfs_mount_args
*ap
,
338 struct xfs_mount
*mp
)
340 int ronly
= (vfs
->vfs_flag
& VFS_RDONLY
);
342 /* Fail a mount where the logbuf is smaller then the log stripe */
343 if (XFS_SB_VERSION_HASLOGV2(&mp
->m_sb
)) {
344 if ((ap
->logbufsize
== -1) &&
345 (mp
->m_sb
.sb_logsunit
> XLOG_BIG_RECORD_BSIZE
)) {
346 mp
->m_logbsize
= mp
->m_sb
.sb_logsunit
;
347 } else if (ap
->logbufsize
< mp
->m_sb
.sb_logsunit
) {
349 "XFS: logbuf size must be greater than or equal to log stripe size");
350 return XFS_ERROR(EINVAL
);
353 /* Fail a mount if the logbuf is larger than 32K */
354 if (ap
->logbufsize
> XLOG_BIG_RECORD_BSIZE
) {
356 "XFS: logbuf size for version 1 logs must be 16K or 32K");
357 return XFS_ERROR(EINVAL
);
362 * prohibit r/w mounts of read-only filesystems
364 if ((mp
->m_sb
.sb_flags
& XFS_SBF_READONLY
) && !ronly
) {
366 "XFS: cannot mount a read-only filesystem as read-write");
367 return XFS_ERROR(EROFS
);
371 * check for shared mount.
373 if (ap
->flags
& XFSMNT_SHARED
) {
374 if (!XFS_SB_VERSION_HASSHARED(&mp
->m_sb
))
375 return XFS_ERROR(EINVAL
);
378 * For IRIX 6.5, shared mounts must have the shared
379 * version bit set, have the persistent readonly
380 * field set, must be version 0 and can only be mounted
383 if (!ronly
|| !(mp
->m_sb
.sb_flags
& XFS_SBF_READONLY
) ||
384 (mp
->m_sb
.sb_shared_vn
!= 0))
385 return XFS_ERROR(EINVAL
);
387 mp
->m_flags
|= XFS_MOUNT_SHARED
;
390 * Shared XFS V0 can't deal with DMI. Return EINVAL.
392 if (mp
->m_sb
.sb_shared_vn
== 0 && (ap
->flags
& XFSMNT_DMAPI
))
393 return XFS_ERROR(EINVAL
);
402 * The file system configurations are:
403 * (1) device (partition) with data and internal log
404 * (2) logical volume with data and log subvolumes.
405 * (3) logical volume with data, log, and realtime subvolumes.
407 * We only have to handle opening the log and realtime volumes here if
408 * they are present. The data subvolume has already been opened by
409 * get_sb_bdev() and is stored in vfsp->vfs_super->s_bdev.
413 struct bhv_desc
*bhvp
,
414 struct xfs_mount_args
*args
,
417 struct vfs
*vfsp
= bhvtovfs(bhvp
);
419 struct xfs_mount
*mp
= XFS_BHVTOM(bhvp
);
420 struct block_device
*ddev
, *logdev
, *rtdev
;
421 int flags
= 0, error
;
423 ddev
= vfsp
->vfs_super
->s_bdev
;
424 logdev
= rtdev
= NULL
;
427 * Setup xfs_mount function vectors from available behaviors
429 p
= vfs_bhv_lookup(vfsp
, VFS_POSITION_DM
);
430 mp
->m_dm_ops
= p
? *(xfs_dmops_t
*) vfs_bhv_custom(p
) : xfs_dmcore_stub
;
431 p
= vfs_bhv_lookup(vfsp
, VFS_POSITION_QM
);
432 mp
->m_qm_ops
= p
? *(xfs_qmops_t
*) vfs_bhv_custom(p
) : xfs_qmcore_stub
;
433 p
= vfs_bhv_lookup(vfsp
, VFS_POSITION_IO
);
434 mp
->m_io_ops
= p
? *(xfs_ioops_t
*) vfs_bhv_custom(p
) : xfs_iocore_xfs
;
437 * Open real time and log devices - order is important.
439 if (args
->logname
[0]) {
440 error
= xfs_blkdev_get(mp
, args
->logname
, &logdev
);
444 if (args
->rtname
[0]) {
445 error
= xfs_blkdev_get(mp
, args
->rtname
, &rtdev
);
447 xfs_blkdev_put(logdev
);
451 if (rtdev
== ddev
|| rtdev
== logdev
) {
453 "XFS: Cannot mount filesystem with identical rtdev and ddev/logdev.");
454 xfs_blkdev_put(logdev
);
455 xfs_blkdev_put(rtdev
);
461 * Setup xfs_mount buffer target pointers
464 mp
->m_ddev_targp
= xfs_alloc_buftarg(ddev
, 0);
465 if (!mp
->m_ddev_targp
) {
466 xfs_blkdev_put(logdev
);
467 xfs_blkdev_put(rtdev
);
471 mp
->m_rtdev_targp
= xfs_alloc_buftarg(rtdev
, 1);
472 if (!mp
->m_rtdev_targp
)
475 mp
->m_logdev_targp
= (logdev
&& logdev
!= ddev
) ?
476 xfs_alloc_buftarg(logdev
, 1) : mp
->m_ddev_targp
;
477 if (!mp
->m_logdev_targp
)
481 * Setup flags based on mount(2) options and then the superblock
483 error
= xfs_start_flags(vfsp
, args
, mp
);
486 error
= xfs_readsb(mp
);
489 error
= xfs_finish_flags(vfsp
, args
, mp
);
494 * Setup xfs_mount buffer target pointers based on superblock
496 error
= xfs_setsize_buftarg(mp
->m_ddev_targp
, mp
->m_sb
.sb_blocksize
,
497 mp
->m_sb
.sb_sectsize
);
498 if (!error
&& logdev
&& logdev
!= ddev
) {
499 unsigned int log_sector_size
= BBSIZE
;
501 if (XFS_SB_VERSION_HASSECTOR(&mp
->m_sb
))
502 log_sector_size
= mp
->m_sb
.sb_logsectsize
;
503 error
= xfs_setsize_buftarg(mp
->m_logdev_targp
,
504 mp
->m_sb
.sb_blocksize
,
508 error
= xfs_setsize_buftarg(mp
->m_rtdev_targp
,
509 mp
->m_sb
.sb_blocksize
,
510 mp
->m_sb
.sb_sectsize
);
514 error
= XFS_IOINIT(vfsp
, args
, flags
);
521 xfs_binval(mp
->m_ddev_targp
);
522 if (logdev
&& logdev
!= ddev
)
523 xfs_binval(mp
->m_logdev_targp
);
525 xfs_binval(mp
->m_rtdev_targp
);
527 xfs_unmountfs_close(mp
, credp
);
537 struct vfs
*vfsp
= bhvtovfs(bdp
);
538 xfs_mount_t
*mp
= XFS_BHVTOM(bdp
);
541 int unmount_event_wanted
= 0;
542 int unmount_event_flags
= 0;
543 int xfs_unmountfs_needed
= 0;
549 if (vfsp
->vfs_flag
& VFS_DMI
) {
550 error
= XFS_SEND_PREUNMOUNT(mp
, vfsp
,
551 rvp
, DM_RIGHT_NULL
, rvp
, DM_RIGHT_NULL
,
553 (mp
->m_dmevmask
& (1<<DM_EVENT_PREUNMOUNT
))?
554 0:DM_FLAGS_UNWANTED
);
556 return XFS_ERROR(error
);
557 unmount_event_wanted
= 1;
558 unmount_event_flags
= (mp
->m_dmevmask
& (1<<DM_EVENT_UNMOUNT
))?
559 0 : DM_FLAGS_UNWANTED
;
563 * First blow any referenced inode from this file system
564 * out of the reference cache, and delete the timer.
566 xfs_refcache_purge_mp(mp
);
568 XFS_bflush(mp
->m_ddev_targp
);
569 error
= xfs_unmount_flush(mp
, 0);
573 ASSERT(vn_count(rvp
) == 1);
576 * Drop the reference count
581 * If we're forcing a shutdown, typically because of a media error,
582 * we want to make sure we invalidate dirty pages that belong to
583 * referenced vnodes as well.
585 if (XFS_FORCED_SHUTDOWN(mp
)) {
586 error
= xfs_sync(&mp
->m_bhv
,
587 (SYNC_WAIT
| SYNC_CLOSE
), credp
);
588 ASSERT(error
!= EFSCORRUPTED
);
590 xfs_unmountfs_needed
= 1;
593 /* Send DMAPI event, if required.
594 * Then do xfs_unmountfs() if needed.
595 * Then return error (or zero).
597 if (unmount_event_wanted
) {
598 /* Note: mp structure must still exist for
599 * XFS_SEND_UNMOUNT() call.
601 XFS_SEND_UNMOUNT(mp
, vfsp
, error
== 0 ? rvp
: NULL
,
602 DM_RIGHT_NULL
, 0, error
, unmount_event_flags
);
604 if (xfs_unmountfs_needed
) {
606 * Call common unmount function to flush to disk
607 * and free the super block buffer & mount structures.
609 xfs_unmountfs(mp
, credp
);
612 return XFS_ERROR(error
);
615 #define REMOUNT_READONLY_FLAGS (SYNC_REMOUNT|SYNC_ATTR|SYNC_WAIT)
621 struct xfs_mount_args
*args
)
623 struct vfs
*vfsp
= bhvtovfs(bdp
);
624 xfs_mount_t
*mp
= XFS_BHVTOM(bdp
);
628 if (args
->flags
& XFSMNT_NOATIME
)
629 mp
->m_flags
|= XFS_MOUNT_NOATIME
;
631 mp
->m_flags
&= ~XFS_MOUNT_NOATIME
;
633 if (!(vfsp
->vfs_flag
& VFS_RDONLY
)) {
634 VFS_SYNC(vfsp
, SYNC_FSDATA
|SYNC_BDFLUSH
|SYNC_ATTR
, NULL
, error
);
637 if (*flags
& MS_RDONLY
) {
638 xfs_refcache_purge_mp(mp
);
639 xfs_flush_buftarg(mp
->m_ddev_targp
, 0);
640 xfs_finish_reclaim_all(mp
, 0);
642 /* This loop must run at least twice.
643 * The first instance of the loop will flush
644 * most meta data but that will generate more
645 * meta data (typically directory updates).
646 * Which then must be flushed and logged before
647 * we can write the unmount record.
650 VFS_SYNC(vfsp
, REMOUNT_READONLY_FLAGS
, NULL
, error
);
651 pincount
= xfs_flush_buftarg(mp
->m_ddev_targp
, 1);
658 /* Ok now write out an unmount record */
659 xfs_log_unmount_write(mp
);
660 xfs_unmountfs_writesb(mp
);
661 vfsp
->vfs_flag
|= VFS_RDONLY
;
663 vfsp
->vfs_flag
&= ~VFS_RDONLY
;
670 * xfs_unmount_flush implements a set of flush operation on special
671 * inodes, which are needed as a separate set of operations so that
672 * they can be called as part of relocation process.
676 xfs_mount_t
*mp
, /* Mount structure we are getting
678 int relocation
) /* Called from vfs relocation. */
680 xfs_inode_t
*rip
= mp
->m_rootip
;
682 xfs_inode_t
*rsumip
= NULL
;
683 vnode_t
*rvp
= XFS_ITOV(rip
);
686 xfs_ilock(rip
, XFS_ILOCK_EXCL
);
690 * Flush out the real time inodes.
692 if ((rbmip
= mp
->m_rbmip
) != NULL
) {
693 xfs_ilock(rbmip
, XFS_ILOCK_EXCL
);
695 error
= xfs_iflush(rbmip
, XFS_IFLUSH_SYNC
);
696 xfs_iunlock(rbmip
, XFS_ILOCK_EXCL
);
698 if (error
== EFSCORRUPTED
)
701 ASSERT(vn_count(XFS_ITOV(rbmip
)) == 1);
703 rsumip
= mp
->m_rsumip
;
704 xfs_ilock(rsumip
, XFS_ILOCK_EXCL
);
706 error
= xfs_iflush(rsumip
, XFS_IFLUSH_SYNC
);
707 xfs_iunlock(rsumip
, XFS_ILOCK_EXCL
);
709 if (error
== EFSCORRUPTED
)
712 ASSERT(vn_count(XFS_ITOV(rsumip
)) == 1);
716 * Synchronously flush root inode to disk
718 error
= xfs_iflush(rip
, XFS_IFLUSH_SYNC
);
719 if (error
== EFSCORRUPTED
)
722 if (vn_count(rvp
) != 1 && !relocation
) {
723 xfs_iunlock(rip
, XFS_ILOCK_EXCL
);
724 return XFS_ERROR(EBUSY
);
728 * Release dquot that rootinode, rbmino and rsumino might be holding,
729 * flush and purge the quota inodes.
731 error
= XFS_QM_UNMOUNT(mp
);
732 if (error
== EFSCORRUPTED
)
736 VN_RELE(XFS_ITOV(rbmip
));
737 VN_RELE(XFS_ITOV(rsumip
));
740 xfs_iunlock(rip
, XFS_ILOCK_EXCL
);
747 xfs_iunlock(rip
, XFS_ILOCK_EXCL
);
749 return XFS_ERROR(EFSCORRUPTED
);
753 * xfs_root extracts the root vnode from a vfs.
755 * vfsp -- the vfs struct for the desired file system
756 * vpp -- address of the caller's vnode pointer which should be
757 * set to the desired fs root vnode
766 vp
= XFS_ITOV((XFS_BHVTOM(bdp
))->m_rootip
);
775 * Fill in the statvfs structure for the given file system. We use
776 * the superblock lock in the mount structure to ensure a consistent
777 * snapshot of the counters returned.
792 mp
= XFS_BHVTOM(bdp
);
795 statp
->f_type
= XFS_SB_MAGIC
;
798 statp
->f_bsize
= sbp
->sb_blocksize
;
799 lsize
= sbp
->sb_logstart
? sbp
->sb_logblocks
: 0;
800 statp
->f_blocks
= sbp
->sb_dblocks
- lsize
;
801 statp
->f_bfree
= statp
->f_bavail
= sbp
->sb_fdblocks
;
802 fakeinos
= statp
->f_bfree
<< sbp
->sb_inopblog
;
804 fakeinos
+= mp
->m_inoadd
;
807 MIN(sbp
->sb_icount
+ fakeinos
, (__uint64_t
)XFS_MAXINUMBER
);
812 statp
->f_files
= min_t(typeof(statp
->f_files
),
815 statp
->f_ffree
= statp
->f_files
- (sbp
->sb_icount
- sbp
->sb_ifree
);
816 XFS_SB_UNLOCK(mp
, s
);
818 id
= huge_encode_dev(mp
->m_dev
);
819 statp
->f_fsid
.val
[0] = (u32
)id
;
820 statp
->f_fsid
.val
[1] = (u32
)(id
>> 32);
821 statp
->f_namelen
= MAXNAMELEN
- 1;
828 * xfs_sync flushes any pending I/O to file system vfsp.
830 * This routine is called by vfs_sync() to make sure that things make it
831 * out to disk eventually, on sync() system calls to flush out everything,
832 * and when the file system is unmounted. For the vfs_sync() case, all
833 * we really need to do is sync out the log to make all of our meta-data
834 * updates permanent (except for timestamps). For calls from pflushd(),
835 * dirty pages are kept moving by calling pdflush() on the inodes
836 * containing them. We also flush the inodes that we can lock without
837 * sleeping and the superblock if we can lock it without sleeping from
838 * vfs_sync() so that items at the tail of the log are always moving out.
841 * SYNC_BDFLUSH - We're being called from vfs_sync() so we don't want
842 * to sleep if we can help it. All we really need
843 * to do is ensure that the log is synced at least
844 * periodically. We also push the inodes and
845 * superblock if we can lock them without sleeping
846 * and they are not pinned.
847 * SYNC_ATTR - We need to flush the inodes. If SYNC_BDFLUSH is not
848 * set, then we really want to lock each inode and flush
850 * SYNC_WAIT - All the flushes that take place in this call should
852 * SYNC_DELWRI - This tells us to push dirty pages associated with
853 * inodes. SYNC_WAIT and SYNC_BDFLUSH are used to
854 * determine if they should be flushed sync, async, or
856 * SYNC_CLOSE - This flag is passed when the system is being
857 * unmounted. We should sync and invalidate everthing.
858 * SYNC_FSDATA - This indicates that the caller would like to make
859 * sure the superblock is safe on disk. We can ensure
860 * this by simply makeing sure the log gets flushed
861 * if SYNC_BDFLUSH is set, and by actually writing it
874 mp
= XFS_BHVTOM(bdp
);
875 return (xfs_syncsub(mp
, flags
, 0, NULL
));
879 * xfs sync routine for internal use
881 * This routine supports all of the flags defined for the generic VFS_SYNC
882 * interface as explained above under xfs_sync. In the interests of not
883 * changing interfaces within the 6.5 family, additional internallly-
884 * required functions are specified within a separate xflags parameter,
885 * only available by calling this routine.
895 xfs_inode_t
*ip
= NULL
;
896 xfs_inode_t
*ip_next
;
904 uint base_lock_flags
;
905 boolean_t mount_locked
;
906 boolean_t vnode_refed
;
909 xfs_iptr_t
*ipointer
;
911 boolean_t ipointer_in
= B_FALSE
;
913 #define IPOINTER_SET ipointer_in = B_TRUE
914 #define IPOINTER_CLR ipointer_in = B_FALSE
921 /* Insert a marker record into the inode list after inode ip. The list
922 * must be locked when this is called. After the call the list will no
925 #define IPOINTER_INSERT(ip, mp) { \
926 ASSERT(ipointer_in == B_FALSE); \
927 ipointer->ip_mnext = ip->i_mnext; \
928 ipointer->ip_mprev = ip; \
929 ip->i_mnext = (xfs_inode_t *)ipointer; \
930 ipointer->ip_mnext->i_mprev = (xfs_inode_t *)ipointer; \
932 XFS_MOUNT_IUNLOCK(mp); \
933 mount_locked = B_FALSE; \
937 /* Remove the marker from the inode list. If the marker was the only item
938 * in the list then there are no remaining inodes and we should zero out
939 * the whole list. If we are the current head of the list then move the head
942 #define IPOINTER_REMOVE(ip, mp) { \
943 ASSERT(ipointer_in == B_TRUE); \
944 if (ipointer->ip_mnext != (xfs_inode_t *)ipointer) { \
945 ip = ipointer->ip_mnext; \
946 ip->i_mprev = ipointer->ip_mprev; \
947 ipointer->ip_mprev->i_mnext = ip; \
948 if (mp->m_inodes == (xfs_inode_t *)ipointer) { \
952 ASSERT(mp->m_inodes == (xfs_inode_t *)ipointer); \
953 mp->m_inodes = NULL; \
959 #define XFS_PREEMPT_MASK 0x7f
963 if (XFS_MTOVFS(mp
)->vfs_flag
& VFS_RDONLY
)
969 /* Allocate a reference marker */
970 ipointer
= (xfs_iptr_t
*)kmem_zalloc(sizeof(xfs_iptr_t
), KM_SLEEP
);
972 fflag
= XFS_B_ASYNC
; /* default is don't wait */
973 if (flags
& SYNC_BDFLUSH
)
974 fflag
= XFS_B_DELWRI
;
975 if (flags
& SYNC_WAIT
)
976 fflag
= 0; /* synchronous overrides all */
978 base_lock_flags
= XFS_ILOCK_SHARED
;
979 if (flags
& (SYNC_DELWRI
| SYNC_CLOSE
)) {
981 * We need the I/O lock if we're going to call any of
982 * the flush/inval routines.
984 base_lock_flags
|= XFS_IOLOCK_SHARED
;
991 mount_locked
= B_TRUE
;
992 vnode_refed
= B_FALSE
;
997 ASSERT(ipointer_in
== B_FALSE
);
998 ASSERT(vnode_refed
== B_FALSE
);
1000 lock_flags
= base_lock_flags
;
1003 * There were no inodes in the list, just break out
1011 * We found another sync thread marker - skip it
1013 if (ip
->i_mount
== NULL
) {
1018 vp
= XFS_ITOV_NULL(ip
);
1021 * If the vnode is gone then this is being torn down,
1022 * call reclaim if it is flushed, else let regular flush
1023 * code deal with it later in the loop.
1027 /* Skip ones already in reclaim */
1028 if (ip
->i_flags
& XFS_IRECLAIM
) {
1032 if (xfs_ilock_nowait(ip
, XFS_ILOCK_EXCL
) == 0) {
1034 } else if ((xfs_ipincount(ip
) == 0) &&
1035 xfs_iflock_nowait(ip
)) {
1036 IPOINTER_INSERT(ip
, mp
);
1038 xfs_finish_reclaim(ip
, 1,
1039 XFS_IFLUSH_DELWRI_ELSE_ASYNC
);
1041 XFS_MOUNT_ILOCK(mp
);
1042 mount_locked
= B_TRUE
;
1043 IPOINTER_REMOVE(ip
, mp
);
1045 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1056 if (XFS_FORCED_SHUTDOWN(mp
) && !(flags
& SYNC_CLOSE
)) {
1057 XFS_MOUNT_IUNLOCK(mp
);
1058 kmem_free(ipointer
, sizeof(xfs_iptr_t
));
1063 * If this is just vfs_sync() or pflushd() calling
1064 * then we can skip inodes for which it looks like
1065 * there is nothing to do. Since we don't have the
1066 * inode locked this is racey, but these are periodic
1067 * calls so it doesn't matter. For the others we want
1068 * to know for sure, so we at least try to lock them.
1070 if (flags
& SYNC_BDFLUSH
) {
1071 if (((ip
->i_itemp
== NULL
) ||
1072 !(ip
->i_itemp
->ili_format
.ilf_fields
&
1074 (ip
->i_update_core
== 0)) {
1081 * Try to lock without sleeping. We're out of order with
1082 * the inode list lock here, so if we fail we need to drop
1083 * the mount lock and try again. If we're called from
1084 * bdflush() here, then don't bother.
1086 * The inode lock here actually coordinates with the
1087 * almost spurious inode lock in xfs_ireclaim() to prevent
1088 * the vnode we handle here without a reference from
1089 * being freed while we reference it. If we lock the inode
1090 * while it's on the mount list here, then the spurious inode
1091 * lock in xfs_ireclaim() after the inode is pulled from
1092 * the mount list will sleep until we release it here.
1093 * This keeps the vnode from being freed while we reference
1094 * it. It is also cheaper and simpler than actually doing
1095 * a vn_get() for every inode we touch here.
1097 if (xfs_ilock_nowait(ip
, lock_flags
) == 0) {
1099 if ((flags
& SYNC_BDFLUSH
) || (vp
== NULL
)) {
1105 * We need to unlock the inode list lock in order
1106 * to lock the inode. Insert a marker record into
1107 * the inode list to remember our position, dropping
1108 * the lock is now done inside the IPOINTER_INSERT
1111 * We also use the inode list lock to protect us
1112 * in taking a snapshot of the vnode version number
1113 * for use in calling vn_get().
1116 IPOINTER_INSERT(ip
, mp
);
1118 vp
= vn_get(vp
, &vmap
);
1121 * The vnode was reclaimed once we let go
1122 * of the inode list lock. Skip to the
1123 * next list entry. Remove the marker.
1126 XFS_MOUNT_ILOCK(mp
);
1128 mount_locked
= B_TRUE
;
1129 vnode_refed
= B_FALSE
;
1131 IPOINTER_REMOVE(ip
, mp
);
1136 xfs_ilock(ip
, lock_flags
);
1138 ASSERT(vp
== XFS_ITOV(ip
));
1139 ASSERT(ip
->i_mount
== mp
);
1141 vnode_refed
= B_TRUE
;
1144 /* From here on in the loop we may have a marker record
1145 * in the inode list.
1148 if ((flags
& SYNC_CLOSE
) && (vp
!= NULL
)) {
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
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
1162 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
1164 if (XFS_FORCED_SHUTDOWN(mp
)) {
1165 VOP_TOSS_PAGES(vp
, 0, -1, FI_REMAPF
);
1167 VOP_FLUSHINVAL_PAGES(vp
, 0, -1, FI_REMAPF
);
1170 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
1172 } else if ((flags
& SYNC_DELWRI
) && (vp
!= NULL
)) {
1174 /* We need to have dropped the lock here,
1175 * so insert a marker if we have not already
1179 IPOINTER_INSERT(ip
, mp
);
1183 * Drop the inode lock since we can't hold it
1184 * across calls to the buffer cache.
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
);
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)))) {
1200 /* Insert marker and drop lock if not already
1204 IPOINTER_INSERT(ip
, mp
);
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().
1225 if ((xfs_ipincount(ip
) == 0) &&
1226 xfs_iflock_nowait(ip
)) {
1229 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
1231 error
= xfs_itobp(mp
, NULL
, ip
,
1236 /* Bailing out, remove the
1237 * marker and free it.
1239 XFS_MOUNT_ILOCK(mp
);
1241 IPOINTER_REMOVE(ip
, mp
);
1243 XFS_MOUNT_IUNLOCK(mp
);
1245 ASSERT(!(lock_flags
&
1246 XFS_IOLOCK_SHARED
));
1249 sizeof(xfs_iptr_t
));
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
1265 XFS_MOUNT_ILOCK(mp
);
1266 mount_locked
= B_TRUE
;
1268 if (ip
!= ipointer
->ip_mprev
) {
1269 IPOINTER_REMOVE(ip
, mp
);
1271 ASSERT(!vnode_refed
);
1272 ASSERT(!(lock_flags
&
1273 XFS_IOLOCK_SHARED
));
1277 ASSERT(ip
->i_mount
== mp
);
1279 if (xfs_ilock_nowait(ip
,
1280 XFS_ILOCK_SHARED
) == 0) {
1281 ASSERT(ip
->i_mount
== mp
);
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.
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
);
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.
1308 XFS_MOUNT_IUNLOCK(mp
);
1309 mount_locked
= B_FALSE
;
1310 error
= xfs_iflush(ip
,
1313 ASSERT(ip
->i_mount
== mp
);
1314 IPOINTER_REMOVE(ip_next
, mp
);
1321 if ((flags
& SYNC_ATTR
) &&
1322 ((ip
->i_update_core
) ||
1323 ((ip
->i_itemp
!= NULL
) &&
1324 (ip
->i_itemp
->ili_format
.ilf_fields
!= 0)))) {
1326 IPOINTER_INSERT(ip
, mp
);
1329 if (flags
& SYNC_WAIT
) {
1331 error
= xfs_iflush(ip
,
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.
1343 if (xfs_iflock_nowait(ip
)) {
1344 error
= xfs_iflush(ip
,
1353 if (lock_flags
!= 0) {
1354 xfs_iunlock(ip
, lock_flags
);
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().
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
1371 IPOINTER_INSERT(ip
, mp
);
1376 vnode_refed
= B_FALSE
;
1384 * bail out if the filesystem is corrupted.
1386 if (error
== EFSCORRUPTED
) {
1387 if (!mount_locked
) {
1388 XFS_MOUNT_ILOCK(mp
);
1389 IPOINTER_REMOVE(ip
, mp
);
1391 XFS_MOUNT_IUNLOCK(mp
);
1392 ASSERT(ipointer_in
== B_FALSE
);
1393 kmem_free(ipointer
, sizeof(xfs_iptr_t
));
1394 return XFS_ERROR(error
);
1397 /* Let other threads have a chance at the mount lock
1398 * if we have looped many times without dropping the
1401 if ((++preempt
& XFS_PREEMPT_MASK
) == 0) {
1403 IPOINTER_INSERT(ip
, mp
);
1407 if (mount_locked
== B_FALSE
) {
1408 XFS_MOUNT_ILOCK(mp
);
1409 mount_locked
= B_TRUE
;
1410 IPOINTER_REMOVE(ip
, mp
);
1414 ASSERT(ipointer_in
== B_FALSE
);
1417 } while (ip
!= mp
->m_inodes
);
1419 XFS_MOUNT_IUNLOCK(mp
);
1421 ASSERT(ipointer_in
== B_FALSE
);
1423 kmem_free(ipointer
, sizeof(xfs_iptr_t
));
1424 return XFS_ERROR(last_error
);
1428 * xfs sync routine for internal use
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 internallly-
1433 * required functions are specified within a separate xflags parameter,
1434 * only available by calling this routine.
1446 uint log_flags
= XFS_LOG_FORCE
;
1448 xfs_buf_log_item_t
*bip
;
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.
1454 if (flags
& SYNC_WAIT
)
1455 log_flags
|= XFS_LOG_SYNC
;
1457 xfs_log_force(mp
, (xfs_lsn_t
)0, log_flags
);
1459 if (flags
& (SYNC_ATTR
|SYNC_DELWRI
)) {
1460 if (flags
& SYNC_BDFLUSH
)
1461 xfs_finish_reclaim_all(mp
, 1);
1463 error
= xfs_sync_inodes(mp
, flags
, xflags
, bypassed
);
1467 * Flushing out dirty data above probably generated more
1468 * log activity, so if this isn't vfs_sync() then flush
1471 if (flags
& SYNC_DELWRI
) {
1472 xfs_log_force(mp
, (xfs_lsn_t
)0, log_flags
);
1475 if (flags
& SYNC_FSDATA
) {
1477 * If this is vfs_sync() then only sync the superblock
1478 * if we can lock it without sleeping and it is not pinned.
1480 if (flags
& SYNC_BDFLUSH
) {
1481 bp
= xfs_getsb(mp
, XFS_BUF_TRYLOCK
);
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
))) {
1488 error
= xfs_bwrite(mp
, bp
);
1497 bp
= xfs_getsb(mp
, 0);
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
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
);
1513 error
= xfs_bwrite(mp
, bp
);
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.
1525 if (flags
& SYNC_REFCACHE
) {
1526 xfs_refcache_purge_some(mp
);
1530 * Now check to see if the log needs a "dummy" transaction.
1533 if (!(flags
& SYNC_REMOUNT
) && xfs_log_need_covered(mp
)) {
1538 * Put a dummy transaction in the log to tell
1539 * recovery that all others are OK.
1541 tp
= xfs_trans_alloc(mp
, XFS_TRANS_DUMMY1
);
1542 if ((error
= xfs_trans_reserve(tp
, 0,
1543 XFS_ICHANGE_LOG_RES(mp
),
1545 xfs_trans_cancel(tp
, 0);
1550 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1552 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
1553 xfs_trans_ihold(tp
, ip
);
1554 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1555 error
= xfs_trans_commit(tp
, 0, NULL
);
1556 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1557 xfs_log_force(mp
, (xfs_lsn_t
)0, log_flags
);
1561 * When shutting down, we need to insure that the AIL is pushed
1562 * to disk or the filesystem can appear corrupt from the PROM.
1564 if ((flags
& (SYNC_CLOSE
|SYNC_WAIT
)) == (SYNC_CLOSE
|SYNC_WAIT
)) {
1565 XFS_bflush(mp
->m_ddev_targp
);
1566 if (mp
->m_rtdev_targp
) {
1567 XFS_bflush(mp
->m_rtdev_targp
);
1571 return XFS_ERROR(last_error
);
1575 * xfs_vget - called by DMAPI and NFSD to get vnode from file handle
1583 xfs_mount_t
*mp
= XFS_BHVTOM(bdp
);
1584 xfs_fid_t
*xfid
= (struct xfs_fid
*)fidp
;
1591 * Invalid. Since handles can be created in user space and passed in
1592 * via gethandle(), this is not cause for a panic.
1594 if (xfid
->xfs_fid_len
!= sizeof(*xfid
) - sizeof(xfid
->xfs_fid_len
))
1595 return XFS_ERROR(EINVAL
);
1597 ino
= xfid
->xfs_fid_ino
;
1598 igen
= xfid
->xfs_fid_gen
;
1601 * NFS can sometimes send requests for ino 0. Fail them gracefully.
1604 return XFS_ERROR(ESTALE
);
1606 error
= xfs_iget(mp
, NULL
, ino
, 0, XFS_ILOCK_SHARED
, &ip
, 0);
1614 return XFS_ERROR(EIO
);
1617 if (ip
->i_d
.di_mode
== 0 || ip
->i_d
.di_gen
!= igen
) {
1618 xfs_iput_new(ip
, XFS_ILOCK_SHARED
);
1620 return XFS_ERROR(ENOENT
);
1623 *vpp
= XFS_ITOV(ip
);
1624 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
1629 #define MNTOPT_LOGBUFS "logbufs" /* number of XFS log buffers */
1630 #define MNTOPT_LOGBSIZE "logbsize" /* size of XFS log buffers */
1631 #define MNTOPT_LOGDEV "logdev" /* log device */
1632 #define MNTOPT_RTDEV "rtdev" /* realtime I/O device */
1633 #define MNTOPT_BIOSIZE "biosize" /* log2 of preferred buffered io size */
1634 #define MNTOPT_WSYNC "wsync" /* safe-mode nfs compatible mount */
1635 #define MNTOPT_INO64 "ino64" /* force inodes into 64-bit range */
1636 #define MNTOPT_NOALIGN "noalign" /* turn off stripe alignment */
1637 #define MNTOPT_SWALLOC "swalloc" /* turn on stripe width allocation */
1638 #define MNTOPT_SUNIT "sunit" /* data volume stripe unit */
1639 #define MNTOPT_SWIDTH "swidth" /* data volume stripe width */
1640 #define MNTOPT_NOUUID "nouuid" /* ignore filesystem UUID */
1641 #define MNTOPT_MTPT "mtpt" /* filesystem mount point */
1642 #define MNTOPT_ALLOCSIZE "allocsize" /* preferred allocation size */
1643 #define MNTOPT_IHASHSIZE "ihashsize" /* size of inode hash table */
1644 #define MNTOPT_NORECOVERY "norecovery" /* don't run XFS recovery */
1645 #define MNTOPT_NOLOGFLUSH "nologflush" /* don't hard flush on log writes */
1646 #define MNTOPT_OSYNCISOSYNC "osyncisosync" /* o_sync is REALLY o_sync */
1647 #define MNTOPT_64BITINODE "inode64" /* inodes can be allocated anywhere */
1648 #define MNTOPT_IKEEP "ikeep" /* do not free empty inode clusters */
1649 #define MNTOPT_NOIKEEP "noikeep" /* free empty inode clusters */
1651 STATIC
unsigned long
1652 suffix_strtoul(const char *cp
, char **endp
, unsigned int base
)
1654 int last
, shift_left_factor
= 0;
1655 char *value
= (char *)cp
;
1657 last
= strlen(value
) - 1;
1658 if (value
[last
] == 'K' || value
[last
] == 'k') {
1659 shift_left_factor
= 10;
1662 if (value
[last
] == 'M' || value
[last
] == 'm') {
1663 shift_left_factor
= 20;
1666 if (value
[last
] == 'G' || value
[last
] == 'g') {
1667 shift_left_factor
= 30;
1671 return simple_strtoul(cp
, endp
, base
) << shift_left_factor
;
1676 struct bhv_desc
*bhv
,
1678 struct xfs_mount_args
*args
,
1681 struct vfs
*vfsp
= bhvtovfs(bhv
);
1682 char *this_char
, *value
, *eov
;
1683 int dsunit
, dswidth
, vol_dsunit
, vol_dswidth
;
1686 #if 0 /* XXX: off by default, until some remaining issues ironed out */
1687 args
->flags
|= XFSMNT_IDELETE
; /* default to on */
1693 iosize
= dsunit
= dswidth
= vol_dsunit
= vol_dswidth
= 0;
1695 while ((this_char
= strsep(&options
, ",")) != NULL
) {
1698 if ((value
= strchr(this_char
, '=')) != NULL
)
1701 if (!strcmp(this_char
, MNTOPT_LOGBUFS
)) {
1702 if (!value
|| !*value
) {
1703 printk("XFS: %s option requires an argument\n",
1707 args
->logbufs
= simple_strtoul(value
, &eov
, 10);
1708 } else if (!strcmp(this_char
, MNTOPT_LOGBSIZE
)) {
1709 if (!value
|| !*value
) {
1710 printk("XFS: %s option requires an argument\n",
1714 args
->logbufsize
= suffix_strtoul(value
, &eov
, 10);
1715 } else if (!strcmp(this_char
, MNTOPT_LOGDEV
)) {
1716 if (!value
|| !*value
) {
1717 printk("XFS: %s option requires an argument\n",
1721 strncpy(args
->logname
, value
, MAXNAMELEN
);
1722 } else if (!strcmp(this_char
, MNTOPT_MTPT
)) {
1723 if (!value
|| !*value
) {
1724 printk("XFS: %s option requires an argument\n",
1728 strncpy(args
->mtpt
, value
, MAXNAMELEN
);
1729 } else if (!strcmp(this_char
, MNTOPT_RTDEV
)) {
1730 if (!value
|| !*value
) {
1731 printk("XFS: %s option requires an argument\n",
1735 strncpy(args
->rtname
, value
, MAXNAMELEN
);
1736 } else if (!strcmp(this_char
, MNTOPT_BIOSIZE
)) {
1737 if (!value
|| !*value
) {
1738 printk("XFS: %s option requires an argument\n",
1742 iosize
= simple_strtoul(value
, &eov
, 10);
1743 args
->flags
|= XFSMNT_IOSIZE
;
1744 args
->iosizelog
= (uint8_t) iosize
;
1745 } else if (!strcmp(this_char
, MNTOPT_ALLOCSIZE
)) {
1746 if (!value
|| !*value
) {
1747 printk("XFS: %s option requires an argument\n",
1751 iosize
= suffix_strtoul(value
, &eov
, 10);
1752 args
->flags
|= XFSMNT_IOSIZE
;
1753 args
->iosizelog
= ffs(iosize
) - 1;
1754 } else if (!strcmp(this_char
, MNTOPT_IHASHSIZE
)) {
1755 if (!value
|| !*value
) {
1756 printk("XFS: %s option requires an argument\n",
1760 args
->flags
|= XFSMNT_IHASHSIZE
;
1761 args
->ihashsize
= simple_strtoul(value
, &eov
, 10);
1762 } else if (!strcmp(this_char
, MNTOPT_WSYNC
)) {
1763 args
->flags
|= XFSMNT_WSYNC
;
1764 } else if (!strcmp(this_char
, MNTOPT_OSYNCISOSYNC
)) {
1765 args
->flags
|= XFSMNT_OSYNCISOSYNC
;
1766 } else if (!strcmp(this_char
, MNTOPT_NORECOVERY
)) {
1767 args
->flags
|= XFSMNT_NORECOVERY
;
1768 } else if (!strcmp(this_char
, MNTOPT_INO64
)) {
1769 args
->flags
|= XFSMNT_INO64
;
1771 printk("XFS: %s option not allowed on this system\n",
1775 } else if (!strcmp(this_char
, MNTOPT_NOALIGN
)) {
1776 args
->flags
|= XFSMNT_NOALIGN
;
1777 } else if (!strcmp(this_char
, MNTOPT_SWALLOC
)) {
1778 args
->flags
|= XFSMNT_SWALLOC
;
1779 } else if (!strcmp(this_char
, MNTOPT_SUNIT
)) {
1780 if (!value
|| !*value
) {
1781 printk("XFS: %s option requires an argument\n",
1785 dsunit
= simple_strtoul(value
, &eov
, 10);
1786 } else if (!strcmp(this_char
, MNTOPT_SWIDTH
)) {
1787 if (!value
|| !*value
) {
1788 printk("XFS: %s option requires an argument\n",
1792 dswidth
= simple_strtoul(value
, &eov
, 10);
1793 } else if (!strcmp(this_char
, MNTOPT_64BITINODE
)) {
1794 args
->flags
&= ~XFSMNT_32BITINODES
;
1796 printk("XFS: %s option not allowed on this system\n",
1800 } else if (!strcmp(this_char
, MNTOPT_NOUUID
)) {
1801 args
->flags
|= XFSMNT_NOUUID
;
1802 } else if (!strcmp(this_char
, MNTOPT_NOLOGFLUSH
)) {
1803 args
->flags
|= XFSMNT_NOLOGFLUSH
;
1804 } else if (!strcmp(this_char
, MNTOPT_IKEEP
)) {
1805 args
->flags
&= ~XFSMNT_IDELETE
;
1806 } else if (!strcmp(this_char
, MNTOPT_NOIKEEP
)) {
1807 args
->flags
|= XFSMNT_IDELETE
;
1808 } else if (!strcmp(this_char
, "osyncisdsync")) {
1809 /* no-op, this is now the default */
1810 printk("XFS: osyncisdsync is now the default, option is deprecated.\n");
1811 } else if (!strcmp(this_char
, "irixsgid")) {
1812 printk("XFS: irixsgid is now a sysctl(2) variable, option is deprecated.\n");
1814 printk("XFS: unknown mount option [%s].\n", this_char
);
1819 if (args
->flags
& XFSMNT_NORECOVERY
) {
1820 if ((vfsp
->vfs_flag
& VFS_RDONLY
) == 0) {
1821 printk("XFS: no-recovery mounts must be read-only.\n");
1826 if ((args
->flags
& XFSMNT_NOALIGN
) && (dsunit
|| dswidth
)) {
1828 "XFS: sunit and swidth options incompatible with the noalign option\n");
1832 if ((dsunit
&& !dswidth
) || (!dsunit
&& dswidth
)) {
1833 printk("XFS: sunit and swidth must be specified together\n");
1837 if (dsunit
&& (dswidth
% dsunit
!= 0)) {
1839 "XFS: stripe width (%d) must be a multiple of the stripe unit (%d)\n",
1844 if ((args
->flags
& XFSMNT_NOALIGN
) != XFSMNT_NOALIGN
) {
1846 args
->sunit
= dsunit
;
1847 args
->flags
|= XFSMNT_RETERR
;
1849 args
->sunit
= vol_dsunit
;
1851 dswidth
? (args
->swidth
= dswidth
) :
1852 (args
->swidth
= vol_dswidth
);
1854 args
->sunit
= args
->swidth
= 0;
1862 struct bhv_desc
*bhv
,
1865 static struct proc_xfs_info
{
1869 /* the few simple ones we can get from the mount struct */
1870 { XFS_MOUNT_WSYNC
, "," MNTOPT_WSYNC
},
1871 { XFS_MOUNT_INO64
, "," MNTOPT_INO64
},
1872 { XFS_MOUNT_NOALIGN
, "," MNTOPT_NOALIGN
},
1873 { XFS_MOUNT_SWALLOC
, "," MNTOPT_SWALLOC
},
1874 { XFS_MOUNT_NOUUID
, "," MNTOPT_NOUUID
},
1875 { XFS_MOUNT_NORECOVERY
, "," MNTOPT_NORECOVERY
},
1876 { XFS_MOUNT_OSYNCISOSYNC
, "," MNTOPT_OSYNCISOSYNC
},
1877 { XFS_MOUNT_NOLOGFLUSH
, "," MNTOPT_NOLOGFLUSH
},
1878 { XFS_MOUNT_IDELETE
, "," MNTOPT_NOIKEEP
},
1881 struct proc_xfs_info
*xfs_infop
;
1882 struct xfs_mount
*mp
= XFS_BHVTOM(bhv
);
1884 for (xfs_infop
= xfs_info
; xfs_infop
->flag
; xfs_infop
++) {
1885 if (mp
->m_flags
& xfs_infop
->flag
)
1886 seq_puts(m
, xfs_infop
->str
);
1889 if (mp
->m_flags
& XFS_MOUNT_IHASHSIZE
)
1890 seq_printf(m
, "," MNTOPT_IHASHSIZE
"=%d", mp
->m_ihsize
);
1892 if (mp
->m_flags
& XFS_MOUNT_DFLT_IOSIZE
)
1893 seq_printf(m
, "," MNTOPT_ALLOCSIZE
"=%d", 1<<mp
->m_writeio_log
);
1895 if (mp
->m_logbufs
> 0)
1896 seq_printf(m
, "," MNTOPT_LOGBUFS
"=%d", mp
->m_logbufs
);
1898 if (mp
->m_logbsize
> 0)
1899 seq_printf(m
, "," MNTOPT_LOGBSIZE
"=%d", mp
->m_logbsize
);
1901 if (mp
->m_ddev_targp
!= mp
->m_logdev_targp
)
1902 seq_printf(m
, "," MNTOPT_LOGDEV
"=%s",
1903 XFS_BUFTARG_NAME(mp
->m_logdev_targp
));
1905 if (mp
->m_rtdev_targp
&& mp
->m_ddev_targp
!= mp
->m_rtdev_targp
)
1906 seq_printf(m
, "," MNTOPT_RTDEV
"=%s",
1907 XFS_BUFTARG_NAME(mp
->m_rtdev_targp
));
1909 if (mp
->m_dalign
> 0)
1910 seq_printf(m
, "," MNTOPT_SUNIT
"=%d",
1911 (int)XFS_FSB_TO_BB(mp
, mp
->m_dalign
));
1913 if (mp
->m_swidth
> 0)
1914 seq_printf(m
, "," MNTOPT_SWIDTH
"=%d",
1915 (int)XFS_FSB_TO_BB(mp
, mp
->m_swidth
));
1917 if (!(mp
->m_flags
& XFS_MOUNT_32BITINOOPT
))
1918 seq_printf(m
, "," MNTOPT_64BITINODE
);
1927 xfs_mount_t
*mp
= XFS_BHVTOM(bdp
);
1929 while (atomic_read(&mp
->m_active_trans
) > 0)
1932 /* Push the superblock and write an unmount record */
1933 xfs_log_unmount_write(mp
);
1934 xfs_unmountfs_writesb(mp
);
1938 vfsops_t xfs_vfsops
= {
1939 BHV_IDENTITY_INIT(VFS_BHV_XFS
,VFS_POSITION_XFS
),
1940 .vfs_parseargs
= xfs_parseargs
,
1941 .vfs_showargs
= xfs_showargs
,
1942 .vfs_mount
= xfs_mount
,
1943 .vfs_unmount
= xfs_unmount
,
1944 .vfs_mntupdate
= xfs_mntupdate
,
1945 .vfs_root
= xfs_root
,
1946 .vfs_statvfs
= xfs_statvfs
,
1947 .vfs_sync
= xfs_sync
,
1948 .vfs_vget
= xfs_vget
,
1949 .vfs_dmapiops
= (vfs_dmapiops_t
)fs_nosys
,
1950 .vfs_quotactl
= (vfs_quotactl_t
)fs_nosys
,
1951 .vfs_init_vnode
= xfs_initialize_vnode
,
1952 .vfs_force_shutdown
= xfs_do_force_shutdown
,
1953 .vfs_freeze
= xfs_freeze
,