2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
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.
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.
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
20 #include "xfs_types.h"
24 #include "xfs_trans.h"
28 #include "xfs_dmapi.h"
29 #include "xfs_mount.h"
30 #include "xfs_da_btree.h"
31 #include "xfs_bmap_btree.h"
32 #include "xfs_ialloc_btree.h"
33 #include "xfs_alloc_btree.h"
34 #include "xfs_dir2_sf.h"
35 #include "xfs_attr_sf.h"
36 #include "xfs_dinode.h"
37 #include "xfs_inode.h"
38 #include "xfs_inode_item.h"
39 #include "xfs_btree.h"
40 #include "xfs_alloc.h"
41 #include "xfs_ialloc.h"
42 #include "xfs_quota.h"
43 #include "xfs_error.h"
46 #include "xfs_buf_item.h"
47 #include "xfs_log_priv.h"
48 #include "xfs_dir2_trace.h"
49 #include "xfs_extfree_item.h"
53 #include "xfs_mru_cache.h"
54 #include "xfs_filestream.h"
55 #include "xfs_fsops.h"
56 #include "xfs_vnodeops.h"
57 #include "xfs_vfsops.h"
58 #include "xfs_utils.h"
64 #ifdef XFS_DABUF_DEBUG
65 extern spinlock_t xfs_dabuf_global_lock
;
66 spin_lock_init(&xfs_dabuf_global_lock
);
70 * Initialize all of the zone allocators we use.
72 xfs_log_ticket_zone
= kmem_zone_init(sizeof(xlog_ticket_t
),
74 xfs_bmap_free_item_zone
= kmem_zone_init(sizeof(xfs_bmap_free_item_t
),
75 "xfs_bmap_free_item");
76 xfs_btree_cur_zone
= kmem_zone_init(sizeof(xfs_btree_cur_t
),
78 xfs_da_state_zone
= kmem_zone_init(sizeof(xfs_da_state_t
),
80 xfs_dabuf_zone
= kmem_zone_init(sizeof(xfs_dabuf_t
), "xfs_dabuf");
81 xfs_ifork_zone
= kmem_zone_init(sizeof(xfs_ifork_t
), "xfs_ifork");
82 xfs_trans_zone
= kmem_zone_init(sizeof(xfs_trans_t
), "xfs_trans");
83 xfs_acl_zone_init(xfs_acl_zone
, "xfs_acl");
85 xfs_filestream_init();
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.
93 kmem_zone_init((sizeof(xfs_buf_log_item_t
) +
94 (((XFS_MAX_BLOCKSIZE
/ XFS_BLI_CHUNK
) /
95 NBWORD
) * sizeof(int))),
98 kmem_zone_init((sizeof(xfs_efd_log_item_t
) +
99 ((XFS_EFD_MAX_FAST_EXTENTS
- 1) *
100 sizeof(xfs_extent_t
))),
103 kmem_zone_init((sizeof(xfs_efi_log_item_t
) +
104 ((XFS_EFI_MAX_FAST_EXTENTS
- 1) *
105 sizeof(xfs_extent_t
))),
109 * These zones warrant special memory allocator hints
112 kmem_zone_init_flags(sizeof(xfs_inode_t
), "xfs_inode",
113 KM_ZONE_HWALIGN
| KM_ZONE_RECLAIM
|
114 KM_ZONE_SPREAD
, NULL
);
116 kmem_zone_init_flags(sizeof(xfs_inode_log_item_t
), "xfs_ili",
117 KM_ZONE_SPREAD
, NULL
);
120 * Allocate global trace buffers.
122 #ifdef XFS_ALLOC_TRACE
123 xfs_alloc_trace_buf
= ktrace_alloc(XFS_ALLOC_TRACE_SIZE
, KM_SLEEP
);
125 #ifdef XFS_BMAP_TRACE
126 xfs_bmap_trace_buf
= ktrace_alloc(XFS_BMAP_TRACE_SIZE
, KM_SLEEP
);
128 #ifdef XFS_BMBT_TRACE
129 xfs_bmbt_trace_buf
= ktrace_alloc(XFS_BMBT_TRACE_SIZE
, KM_SLEEP
);
131 #ifdef XFS_ATTR_TRACE
132 xfs_attr_trace_buf
= ktrace_alloc(XFS_ATTR_TRACE_SIZE
, KM_SLEEP
);
134 #ifdef XFS_DIR2_TRACE
135 xfs_dir2_trace_buf
= ktrace_alloc(XFS_DIR2_GTRACE_SIZE
, KM_SLEEP
);
140 #if (defined(DEBUG) || defined(INDUCE_IO_ERROR))
141 xfs_error_test_init();
142 #endif /* DEBUG || INDUCE_IO_ERROR */
145 xfs_sysctl_register();
152 extern kmem_zone_t
*xfs_inode_zone
;
153 extern kmem_zone_t
*xfs_efd_zone
;
154 extern kmem_zone_t
*xfs_efi_zone
;
156 xfs_cleanup_procfs();
157 xfs_sysctl_unregister();
158 xfs_filestream_uninit();
159 xfs_mru_cache_uninit();
160 xfs_acl_zone_destroy(xfs_acl_zone
);
162 #ifdef XFS_DIR2_TRACE
163 ktrace_free(xfs_dir2_trace_buf
);
165 #ifdef XFS_ATTR_TRACE
166 ktrace_free(xfs_attr_trace_buf
);
168 #ifdef XFS_BMBT_TRACE
169 ktrace_free(xfs_bmbt_trace_buf
);
171 #ifdef XFS_BMAP_TRACE
172 ktrace_free(xfs_bmap_trace_buf
);
174 #ifdef XFS_ALLOC_TRACE
175 ktrace_free(xfs_alloc_trace_buf
);
178 kmem_zone_destroy(xfs_bmap_free_item_zone
);
179 kmem_zone_destroy(xfs_btree_cur_zone
);
180 kmem_zone_destroy(xfs_inode_zone
);
181 kmem_zone_destroy(xfs_trans_zone
);
182 kmem_zone_destroy(xfs_da_state_zone
);
183 kmem_zone_destroy(xfs_dabuf_zone
);
184 kmem_zone_destroy(xfs_buf_item_zone
);
185 kmem_zone_destroy(xfs_efd_zone
);
186 kmem_zone_destroy(xfs_efi_zone
);
187 kmem_zone_destroy(xfs_ifork_zone
);
188 kmem_zone_destroy(xfs_ili_zone
);
194 * This function fills in xfs_mount_t fields based on mount args.
195 * Note: the superblock has _not_ yet been read in.
199 struct xfs_mount_args
*ap
,
200 struct xfs_mount
*mp
)
202 /* Values are in BBs */
203 if ((ap
->flags
& XFSMNT_NOALIGN
) != XFSMNT_NOALIGN
) {
205 * At this point the superblock has not been read
206 * in, therefore we do not know the block size.
207 * Before the mount call ends we will convert
210 mp
->m_dalign
= ap
->sunit
;
211 mp
->m_swidth
= ap
->swidth
;
214 if (ap
->logbufs
!= -1 &&
216 (ap
->logbufs
< XLOG_MIN_ICLOGS
||
217 ap
->logbufs
> XLOG_MAX_ICLOGS
)) {
219 "XFS: invalid logbufs value: %d [not %d-%d]",
220 ap
->logbufs
, XLOG_MIN_ICLOGS
, XLOG_MAX_ICLOGS
);
221 return XFS_ERROR(EINVAL
);
223 mp
->m_logbufs
= ap
->logbufs
;
224 if (ap
->logbufsize
!= -1 &&
225 ap
->logbufsize
!= 0 &&
226 (ap
->logbufsize
< XLOG_MIN_RECORD_BSIZE
||
227 ap
->logbufsize
> XLOG_MAX_RECORD_BSIZE
||
228 !is_power_of_2(ap
->logbufsize
))) {
230 "XFS: invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
232 return XFS_ERROR(EINVAL
);
234 mp
->m_logbsize
= ap
->logbufsize
;
235 mp
->m_fsname_len
= strlen(ap
->fsname
) + 1;
236 mp
->m_fsname
= kmem_alloc(mp
->m_fsname_len
, KM_SLEEP
);
237 strcpy(mp
->m_fsname
, ap
->fsname
);
239 mp
->m_rtname
= kmem_alloc(strlen(ap
->rtname
) + 1, KM_SLEEP
);
240 strcpy(mp
->m_rtname
, ap
->rtname
);
242 if (ap
->logname
[0]) {
243 mp
->m_logname
= kmem_alloc(strlen(ap
->logname
) + 1, KM_SLEEP
);
244 strcpy(mp
->m_logname
, ap
->logname
);
247 if (ap
->flags
& XFSMNT_WSYNC
)
248 mp
->m_flags
|= XFS_MOUNT_WSYNC
;
250 if (ap
->flags
& XFSMNT_INO64
) {
251 mp
->m_flags
|= XFS_MOUNT_INO64
;
252 mp
->m_inoadd
= XFS_INO64_OFFSET
;
255 if (ap
->flags
& XFSMNT_RETERR
)
256 mp
->m_flags
|= XFS_MOUNT_RETERR
;
257 if (ap
->flags
& XFSMNT_NOALIGN
)
258 mp
->m_flags
|= XFS_MOUNT_NOALIGN
;
259 if (ap
->flags
& XFSMNT_SWALLOC
)
260 mp
->m_flags
|= XFS_MOUNT_SWALLOC
;
261 if (ap
->flags
& XFSMNT_OSYNCISOSYNC
)
262 mp
->m_flags
|= XFS_MOUNT_OSYNCISOSYNC
;
263 if (ap
->flags
& XFSMNT_32BITINODES
)
264 mp
->m_flags
|= XFS_MOUNT_32BITINODES
;
266 if (ap
->flags
& XFSMNT_IOSIZE
) {
267 if (ap
->iosizelog
> XFS_MAX_IO_LOG
||
268 ap
->iosizelog
< XFS_MIN_IO_LOG
) {
270 "XFS: invalid log iosize: %d [not %d-%d]",
271 ap
->iosizelog
, XFS_MIN_IO_LOG
,
273 return XFS_ERROR(EINVAL
);
276 mp
->m_flags
|= XFS_MOUNT_DFLT_IOSIZE
;
277 mp
->m_readio_log
= mp
->m_writeio_log
= ap
->iosizelog
;
280 if (ap
->flags
& XFSMNT_IKEEP
)
281 mp
->m_flags
|= XFS_MOUNT_IKEEP
;
282 if (ap
->flags
& XFSMNT_DIRSYNC
)
283 mp
->m_flags
|= XFS_MOUNT_DIRSYNC
;
284 if (ap
->flags
& XFSMNT_ATTR2
)
285 mp
->m_flags
|= XFS_MOUNT_ATTR2
;
287 if (ap
->flags2
& XFSMNT2_COMPAT_IOSIZE
)
288 mp
->m_flags
|= XFS_MOUNT_COMPAT_IOSIZE
;
291 * no recovery flag requires a read-only mount
293 if (ap
->flags
& XFSMNT_NORECOVERY
) {
294 if (!(mp
->m_flags
& XFS_MOUNT_RDONLY
)) {
296 "XFS: tried to mount a FS read-write without recovery!");
297 return XFS_ERROR(EINVAL
);
299 mp
->m_flags
|= XFS_MOUNT_NORECOVERY
;
302 if (ap
->flags
& XFSMNT_NOUUID
)
303 mp
->m_flags
|= XFS_MOUNT_NOUUID
;
304 if (ap
->flags
& XFSMNT_BARRIER
)
305 mp
->m_flags
|= XFS_MOUNT_BARRIER
;
307 mp
->m_flags
&= ~XFS_MOUNT_BARRIER
;
309 if (ap
->flags2
& XFSMNT2_FILESTREAMS
)
310 mp
->m_flags
|= XFS_MOUNT_FILESTREAMS
;
312 if (ap
->flags
& XFSMNT_DMAPI
)
313 mp
->m_flags
|= XFS_MOUNT_DMAPI
;
318 * This function fills in xfs_mount_t fields based on mount args.
319 * Note: the superblock _has_ now been read in.
323 struct xfs_mount_args
*ap
,
324 struct xfs_mount
*mp
)
326 int ronly
= (mp
->m_flags
& XFS_MOUNT_RDONLY
);
328 /* Fail a mount where the logbuf is smaller then the log stripe */
329 if (xfs_sb_version_haslogv2(&mp
->m_sb
)) {
330 if ((ap
->logbufsize
<= 0) &&
331 (mp
->m_sb
.sb_logsunit
> XLOG_BIG_RECORD_BSIZE
)) {
332 mp
->m_logbsize
= mp
->m_sb
.sb_logsunit
;
333 } else if (ap
->logbufsize
> 0 &&
334 ap
->logbufsize
< mp
->m_sb
.sb_logsunit
) {
336 "XFS: logbuf size must be greater than or equal to log stripe size");
337 return XFS_ERROR(EINVAL
);
340 /* Fail a mount if the logbuf is larger than 32K */
341 if (ap
->logbufsize
> XLOG_BIG_RECORD_BSIZE
) {
343 "XFS: logbuf size for version 1 logs must be 16K or 32K");
344 return XFS_ERROR(EINVAL
);
348 if (xfs_sb_version_hasattr2(&mp
->m_sb
))
349 mp
->m_flags
|= XFS_MOUNT_ATTR2
;
352 * prohibit r/w mounts of read-only filesystems
354 if ((mp
->m_sb
.sb_flags
& XFS_SBF_READONLY
) && !ronly
) {
356 "XFS: cannot mount a read-only filesystem as read-write");
357 return XFS_ERROR(EROFS
);
361 * check for shared mount.
363 if (ap
->flags
& XFSMNT_SHARED
) {
364 if (!xfs_sb_version_hasshared(&mp
->m_sb
))
365 return XFS_ERROR(EINVAL
);
368 * For IRIX 6.5, shared mounts must have the shared
369 * version bit set, have the persistent readonly
370 * field set, must be version 0 and can only be mounted
373 if (!ronly
|| !(mp
->m_sb
.sb_flags
& XFS_SBF_READONLY
) ||
374 (mp
->m_sb
.sb_shared_vn
!= 0))
375 return XFS_ERROR(EINVAL
);
377 mp
->m_flags
|= XFS_MOUNT_SHARED
;
380 * Shared XFS V0 can't deal with DMI. Return EINVAL.
382 if (mp
->m_sb
.sb_shared_vn
== 0 && (ap
->flags
& XFSMNT_DMAPI
))
383 return XFS_ERROR(EINVAL
);
386 if (ap
->flags
& XFSMNT_UQUOTA
) {
387 mp
->m_qflags
|= (XFS_UQUOTA_ACCT
| XFS_UQUOTA_ACTIVE
);
388 if (ap
->flags
& XFSMNT_UQUOTAENF
)
389 mp
->m_qflags
|= XFS_UQUOTA_ENFD
;
392 if (ap
->flags
& XFSMNT_GQUOTA
) {
393 mp
->m_qflags
|= (XFS_GQUOTA_ACCT
| XFS_GQUOTA_ACTIVE
);
394 if (ap
->flags
& XFSMNT_GQUOTAENF
)
395 mp
->m_qflags
|= XFS_OQUOTA_ENFD
;
396 } else if (ap
->flags
& XFSMNT_PQUOTA
) {
397 mp
->m_qflags
|= (XFS_PQUOTA_ACCT
| XFS_PQUOTA_ACTIVE
);
398 if (ap
->flags
& XFSMNT_PQUOTAENF
)
399 mp
->m_qflags
|= XFS_OQUOTA_ENFD
;
408 * The file system configurations are:
409 * (1) device (partition) with data and internal log
410 * (2) logical volume with data and log subvolumes.
411 * (3) logical volume with data, log, and realtime subvolumes.
413 * We only have to handle opening the log and realtime volumes here if
414 * they are present. The data subvolume has already been opened by
415 * get_sb_bdev() and is stored in vfsp->vfs_super->s_bdev.
419 struct xfs_mount
*mp
,
420 struct xfs_mount_args
*args
,
423 struct block_device
*ddev
, *logdev
, *rtdev
;
424 int flags
= 0, error
;
426 ddev
= mp
->m_super
->s_bdev
;
427 logdev
= rtdev
= NULL
;
429 error
= xfs_dmops_get(mp
, args
);
432 error
= xfs_qmops_get(mp
, args
);
436 if (args
->flags
& XFSMNT_QUIET
)
437 flags
|= XFS_MFSI_QUIET
;
440 * Open real time and log devices - order is important.
442 if (args
->logname
[0]) {
443 error
= xfs_blkdev_get(mp
, args
->logname
, &logdev
);
447 if (args
->rtname
[0]) {
448 error
= xfs_blkdev_get(mp
, args
->rtname
, &rtdev
);
450 xfs_blkdev_put(logdev
);
454 if (rtdev
== ddev
|| rtdev
== logdev
) {
456 "XFS: Cannot mount filesystem with identical rtdev and ddev/logdev.");
457 xfs_blkdev_put(logdev
);
458 xfs_blkdev_put(rtdev
);
464 * Setup xfs_mount buffer target pointers
467 mp
->m_ddev_targp
= xfs_alloc_buftarg(ddev
, 0);
468 if (!mp
->m_ddev_targp
) {
469 xfs_blkdev_put(logdev
);
470 xfs_blkdev_put(rtdev
);
474 mp
->m_rtdev_targp
= xfs_alloc_buftarg(rtdev
, 1);
475 if (!mp
->m_rtdev_targp
) {
476 xfs_blkdev_put(logdev
);
477 xfs_blkdev_put(rtdev
);
481 mp
->m_logdev_targp
= (logdev
&& logdev
!= ddev
) ?
482 xfs_alloc_buftarg(logdev
, 1) : mp
->m_ddev_targp
;
483 if (!mp
->m_logdev_targp
) {
484 xfs_blkdev_put(logdev
);
485 xfs_blkdev_put(rtdev
);
490 * Setup flags based on mount(2) options and then the superblock
492 error
= xfs_start_flags(args
, mp
);
495 error
= xfs_readsb(mp
, flags
);
498 error
= xfs_finish_flags(args
, mp
);
503 * Setup xfs_mount buffer target pointers based on superblock
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
;
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
,
517 error
= xfs_setsize_buftarg(mp
->m_rtdev_targp
,
518 mp
->m_sb
.sb_blocksize
,
519 mp
->m_sb
.sb_sectsize
);
523 if (mp
->m_flags
& XFS_MOUNT_BARRIER
)
524 xfs_mountfs_check_barriers(mp
);
526 if ((error
= xfs_filestream_mount(mp
)))
529 error
= xfs_mountfs(mp
, flags
);
533 XFS_SEND_MOUNT(mp
, DM_RIGHT_NULL
, args
->mtpt
, args
->fsname
);
541 xfs_binval(mp
->m_ddev_targp
);
542 if (logdev
&& logdev
!= ddev
)
543 xfs_binval(mp
->m_logdev_targp
);
545 xfs_binval(mp
->m_rtdev_targp
);
547 xfs_unmountfs_close(mp
, credp
);
561 int unmount_event_wanted
= 0;
562 int unmount_event_flags
= 0;
563 int xfs_unmountfs_needed
= 0;
570 if (mp
->m_flags
& XFS_MOUNT_DMAPI
) {
571 error
= XFS_SEND_PREUNMOUNT(mp
,
572 rip
, DM_RIGHT_NULL
, rip
, DM_RIGHT_NULL
,
574 (mp
->m_dmevmask
& (1<<DM_EVENT_PREUNMOUNT
))?
575 0:DM_FLAGS_UNWANTED
);
577 return XFS_ERROR(error
);
578 unmount_event_wanted
= 1;
579 unmount_event_flags
= (mp
->m_dmevmask
& (1<<DM_EVENT_UNMOUNT
))?
580 0 : DM_FLAGS_UNWANTED
;
585 * Blow away any referenced inode in the filestreams cache.
586 * This can and will cause log traffic as inodes go inactive
589 xfs_filestream_unmount(mp
);
591 XFS_bflush(mp
->m_ddev_targp
);
592 error
= xfs_unmount_flush(mp
, 0);
596 ASSERT(vn_count(rvp
) == 1);
599 * Drop the reference count
604 * If we're forcing a shutdown, typically because of a media error,
605 * we want to make sure we invalidate dirty pages that belong to
606 * referenced vnodes as well.
608 if (XFS_FORCED_SHUTDOWN(mp
)) {
609 error
= xfs_sync(mp
, SYNC_WAIT
| SYNC_CLOSE
);
610 ASSERT(error
!= EFSCORRUPTED
);
612 xfs_unmountfs_needed
= 1;
615 /* Send DMAPI event, if required.
616 * Then do xfs_unmountfs() if needed.
617 * Then return error (or zero).
619 if (unmount_event_wanted
) {
620 /* Note: mp structure must still exist for
621 * XFS_SEND_UNMOUNT() call.
623 XFS_SEND_UNMOUNT(mp
, error
== 0 ? rip
: NULL
,
624 DM_RIGHT_NULL
, 0, error
, unmount_event_flags
);
626 if (xfs_unmountfs_needed
) {
628 * Call common unmount function to flush to disk
629 * and free the super block buffer & mount structures.
631 xfs_unmountfs(mp
, credp
);
634 kmem_free(mp
, sizeof(xfs_mount_t
));
637 return XFS_ERROR(error
);
644 int count
= 0, pincount
;
646 xfs_flush_buftarg(mp
->m_ddev_targp
, 0);
647 xfs_finish_reclaim_all(mp
, 0);
649 /* This loop must run at least twice.
650 * The first instance of the loop will flush
651 * most meta data but that will generate more
652 * meta data (typically directory updates).
653 * Which then must be flushed and logged before
654 * we can write the unmount record.
657 xfs_syncsub(mp
, SYNC_INODE_QUIESCE
, NULL
);
658 pincount
= xfs_flush_buftarg(mp
->m_ddev_targp
, 1);
669 * Second stage of a quiesce. The data is already synced, now we have to take
670 * care of the metadata. New transactions are already blocked, so we need to
671 * wait for any remaining transactions to drain out before proceding.
677 /* wait for all modifications to complete */
678 while (atomic_read(&mp
->m_active_trans
) > 0)
681 /* flush inodes and push all remaining buffers out to disk */
684 ASSERT_ALWAYS(atomic_read(&mp
->m_active_trans
) == 0);
686 /* Push the superblock and write an unmount record */
687 xfs_log_sbcount(mp
, 1);
688 xfs_log_unmount_write(mp
);
689 xfs_unmountfs_writesb(mp
);
694 struct xfs_mount
*mp
,
696 struct xfs_mount_args
*args
)
698 if (!(*flags
& MS_RDONLY
)) { /* rw/ro -> rw */
699 if (mp
->m_flags
& XFS_MOUNT_RDONLY
)
700 mp
->m_flags
&= ~XFS_MOUNT_RDONLY
;
701 if (args
->flags
& XFSMNT_BARRIER
) {
702 mp
->m_flags
|= XFS_MOUNT_BARRIER
;
703 xfs_mountfs_check_barriers(mp
);
705 mp
->m_flags
&= ~XFS_MOUNT_BARRIER
;
707 } else if (!(mp
->m_flags
& XFS_MOUNT_RDONLY
)) { /* rw -> ro */
708 xfs_filestream_flush(mp
);
709 xfs_sync(mp
, SYNC_DATA_QUIESCE
);
710 xfs_attr_quiesce(mp
);
711 mp
->m_flags
|= XFS_MOUNT_RDONLY
;
717 * xfs_unmount_flush implements a set of flush operation on special
718 * inodes, which are needed as a separate set of operations so that
719 * they can be called as part of relocation process.
723 xfs_mount_t
*mp
, /* Mount structure we are getting
725 int relocation
) /* Called from vfs relocation. */
727 xfs_inode_t
*rip
= mp
->m_rootip
;
729 xfs_inode_t
*rsumip
= NULL
;
730 bhv_vnode_t
*rvp
= XFS_ITOV(rip
);
733 xfs_ilock(rip
, XFS_ILOCK_EXCL
| XFS_ILOCK_PARENT
);
737 * Flush out the real time inodes.
739 if ((rbmip
= mp
->m_rbmip
) != NULL
) {
740 xfs_ilock(rbmip
, XFS_ILOCK_EXCL
);
742 error
= xfs_iflush(rbmip
, XFS_IFLUSH_SYNC
);
743 xfs_iunlock(rbmip
, XFS_ILOCK_EXCL
);
745 if (error
== EFSCORRUPTED
)
748 ASSERT(vn_count(XFS_ITOV(rbmip
)) == 1);
750 rsumip
= mp
->m_rsumip
;
751 xfs_ilock(rsumip
, XFS_ILOCK_EXCL
);
753 error
= xfs_iflush(rsumip
, XFS_IFLUSH_SYNC
);
754 xfs_iunlock(rsumip
, XFS_ILOCK_EXCL
);
756 if (error
== EFSCORRUPTED
)
759 ASSERT(vn_count(XFS_ITOV(rsumip
)) == 1);
763 * Synchronously flush root inode to disk
765 error
= xfs_iflush(rip
, XFS_IFLUSH_SYNC
);
766 if (error
== EFSCORRUPTED
)
769 if (vn_count(rvp
) != 1 && !relocation
) {
770 xfs_iunlock(rip
, XFS_ILOCK_EXCL
);
771 return XFS_ERROR(EBUSY
);
775 * Release dquot that rootinode, rbmino and rsumino might be holding,
776 * flush and purge the quota inodes.
778 error
= XFS_QM_UNMOUNT(mp
);
779 if (error
== EFSCORRUPTED
)
787 xfs_iunlock(rip
, XFS_ILOCK_EXCL
);
794 xfs_iunlock(rip
, XFS_ILOCK_EXCL
);
796 return XFS_ERROR(EFSCORRUPTED
);
800 * xfs_sync flushes any pending I/O to file system vfsp.
802 * This routine is called by vfs_sync() to make sure that things make it
803 * out to disk eventually, on sync() system calls to flush out everything,
804 * and when the file system is unmounted. For the vfs_sync() case, all
805 * we really need to do is sync out the log to make all of our meta-data
806 * updates permanent (except for timestamps). For calls from pflushd(),
807 * dirty pages are kept moving by calling pdflush() on the inodes
808 * containing them. We also flush the inodes that we can lock without
809 * sleeping and the superblock if we can lock it without sleeping from
810 * vfs_sync() so that items at the tail of the log are always moving out.
813 * SYNC_BDFLUSH - We're being called from vfs_sync() so we don't want
814 * to sleep if we can help it. All we really need
815 * to do is ensure that the log is synced at least
816 * periodically. We also push the inodes and
817 * superblock if we can lock them without sleeping
818 * and they are not pinned.
819 * SYNC_ATTR - We need to flush the inodes. If SYNC_BDFLUSH is not
820 * set, then we really want to lock each inode and flush
822 * SYNC_WAIT - All the flushes that take place in this call should
824 * SYNC_DELWRI - This tells us to push dirty pages associated with
825 * inodes. SYNC_WAIT and SYNC_BDFLUSH are used to
826 * determine if they should be flushed sync, async, or
828 * SYNC_CLOSE - This flag is passed when the system is being
829 * unmounted. We should sync and invalidate everything.
830 * SYNC_FSDATA - This indicates that the caller would like to make
831 * sure the superblock is safe on disk. We can ensure
832 * this by simply making sure the log gets flushed
833 * if SYNC_BDFLUSH is set, and by actually writing it
835 * SYNC_IOWAIT - The caller wants us to wait for all data I/O to complete
836 * before we return (including direct I/O). Forms the drain
837 * side of the write barrier needed to safely quiesce the
849 * Get the Quota Manager to flush the dquots.
851 * If XFS quota support is not enabled or this filesystem
852 * instance does not use quotas XFS_QM_DQSYNC will always
855 error
= XFS_QM_DQSYNC(mp
, flags
);
858 * If we got an IO error, we will be shutting down.
859 * So, there's nothing more for us to do here.
861 ASSERT(error
!= EIO
|| XFS_FORCED_SHUTDOWN(mp
));
862 if (XFS_FORCED_SHUTDOWN(mp
))
863 return XFS_ERROR(error
);
866 if (flags
& SYNC_IOWAIT
)
867 xfs_filestream_flush(mp
);
869 return xfs_syncsub(mp
, flags
, NULL
);
873 * xfs sync routine for internal use
875 * This routine supports all of the flags defined for the generic vfs_sync
876 * interface as explained above under xfs_sync.
885 xfs_inode_t
*ip
= NULL
;
886 bhv_vnode_t
*vp
= NULL
;
891 uint base_lock_flags
;
892 boolean_t mount_locked
;
893 boolean_t vnode_refed
;
895 xfs_iptr_t
*ipointer
;
897 boolean_t ipointer_in
= B_FALSE
;
899 #define IPOINTER_SET ipointer_in = B_TRUE
900 #define IPOINTER_CLR ipointer_in = B_FALSE
907 /* Insert a marker record into the inode list after inode ip. The list
908 * must be locked when this is called. After the call the list will no
911 #define IPOINTER_INSERT(ip, mp) { \
912 ASSERT(ipointer_in == B_FALSE); \
913 ipointer->ip_mnext = ip->i_mnext; \
914 ipointer->ip_mprev = ip; \
915 ip->i_mnext = (xfs_inode_t *)ipointer; \
916 ipointer->ip_mnext->i_mprev = (xfs_inode_t *)ipointer; \
918 XFS_MOUNT_IUNLOCK(mp); \
919 mount_locked = B_FALSE; \
923 /* Remove the marker from the inode list. If the marker was the only item
924 * in the list then there are no remaining inodes and we should zero out
925 * the whole list. If we are the current head of the list then move the head
928 #define IPOINTER_REMOVE(ip, mp) { \
929 ASSERT(ipointer_in == B_TRUE); \
930 if (ipointer->ip_mnext != (xfs_inode_t *)ipointer) { \
931 ip = ipointer->ip_mnext; \
932 ip->i_mprev = ipointer->ip_mprev; \
933 ipointer->ip_mprev->i_mnext = ip; \
934 if (mp->m_inodes == (xfs_inode_t *)ipointer) { \
938 ASSERT(mp->m_inodes == (xfs_inode_t *)ipointer); \
939 mp->m_inodes = NULL; \
945 #define XFS_PREEMPT_MASK 0x7f
947 ASSERT(!(flags
& SYNC_BDFLUSH
));
951 if (mp
->m_flags
& XFS_MOUNT_RDONLY
)
957 /* Allocate a reference marker */
958 ipointer
= (xfs_iptr_t
*)kmem_zalloc(sizeof(xfs_iptr_t
), KM_SLEEP
);
960 fflag
= XFS_B_ASYNC
; /* default is don't wait */
961 if (flags
& SYNC_DELWRI
)
962 fflag
= XFS_B_DELWRI
;
963 if (flags
& SYNC_WAIT
)
964 fflag
= 0; /* synchronous overrides all */
966 base_lock_flags
= XFS_ILOCK_SHARED
;
967 if (flags
& (SYNC_DELWRI
| SYNC_CLOSE
)) {
969 * We need the I/O lock if we're going to call any of
970 * the flush/inval routines.
972 base_lock_flags
|= XFS_IOLOCK_SHARED
;
979 mount_locked
= B_TRUE
;
980 vnode_refed
= B_FALSE
;
985 ASSERT(ipointer_in
== B_FALSE
);
986 ASSERT(vnode_refed
== B_FALSE
);
988 lock_flags
= base_lock_flags
;
991 * There were no inodes in the list, just break out
999 * We found another sync thread marker - skip it
1001 if (ip
->i_mount
== NULL
) {
1006 vp
= XFS_ITOV_NULL(ip
);
1009 * If the vnode is gone then this is being torn down,
1010 * call reclaim if it is flushed, else let regular flush
1011 * code deal with it later in the loop.
1015 /* Skip ones already in reclaim */
1016 if (ip
->i_flags
& XFS_IRECLAIM
) {
1020 if (xfs_ilock_nowait(ip
, XFS_ILOCK_EXCL
) == 0) {
1022 } else if ((xfs_ipincount(ip
) == 0) &&
1023 xfs_iflock_nowait(ip
)) {
1024 IPOINTER_INSERT(ip
, mp
);
1026 xfs_finish_reclaim(ip
, 1,
1027 XFS_IFLUSH_DELWRI_ELSE_ASYNC
);
1029 XFS_MOUNT_ILOCK(mp
);
1030 mount_locked
= B_TRUE
;
1031 IPOINTER_REMOVE(ip
, mp
);
1033 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1044 if (XFS_FORCED_SHUTDOWN(mp
) && !(flags
& SYNC_CLOSE
)) {
1045 XFS_MOUNT_IUNLOCK(mp
);
1046 kmem_free(ipointer
, sizeof(xfs_iptr_t
));
1051 * Try to lock without sleeping. We're out of order with
1052 * the inode list lock here, so if we fail we need to drop
1053 * the mount lock and try again. If we're called from
1054 * bdflush() here, then don't bother.
1056 * The inode lock here actually coordinates with the
1057 * almost spurious inode lock in xfs_ireclaim() to prevent
1058 * the vnode we handle here without a reference from
1059 * being freed while we reference it. If we lock the inode
1060 * while it's on the mount list here, then the spurious inode
1061 * lock in xfs_ireclaim() after the inode is pulled from
1062 * the mount list will sleep until we release it here.
1063 * This keeps the vnode from being freed while we reference
1066 if (xfs_ilock_nowait(ip
, lock_flags
) == 0) {
1078 IPOINTER_INSERT(ip
, mp
);
1079 xfs_ilock(ip
, lock_flags
);
1081 ASSERT(vp
== XFS_ITOV(ip
));
1082 ASSERT(ip
->i_mount
== mp
);
1084 vnode_refed
= B_TRUE
;
1087 /* From here on in the loop we may have a marker record
1088 * in the inode list.
1092 * If we have to flush data or wait for I/O completion
1093 * we need to drop the ilock that we currently hold.
1094 * If we need to drop the lock, insert a marker if we
1095 * have not already done so.
1097 if ((flags
& (SYNC_CLOSE
|SYNC_IOWAIT
)) ||
1098 ((flags
& SYNC_DELWRI
) && VN_DIRTY(vp
))) {
1100 IPOINTER_INSERT(ip
, mp
);
1102 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
1104 if (flags
& SYNC_CLOSE
) {
1105 /* Shutdown case. Flush and invalidate. */
1106 if (XFS_FORCED_SHUTDOWN(mp
))
1107 xfs_tosspages(ip
, 0, -1,
1110 error
= xfs_flushinval_pages(ip
,
1112 } else if ((flags
& SYNC_DELWRI
) && VN_DIRTY(vp
)) {
1113 error
= xfs_flush_pages(ip
, 0,
1114 -1, fflag
, FI_NONE
);
1118 * When freezing, we need to wait ensure all I/O (including direct
1119 * I/O) is complete to ensure no further data modification can take
1120 * place after this point
1122 if (flags
& SYNC_IOWAIT
)
1125 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
1128 if ((flags
& SYNC_ATTR
) &&
1129 (ip
->i_update_core
||
1130 (ip
->i_itemp
&& ip
->i_itemp
->ili_format
.ilf_fields
))) {
1132 IPOINTER_INSERT(ip
, mp
);
1134 if (flags
& SYNC_WAIT
) {
1136 error
= xfs_iflush(ip
, XFS_IFLUSH_SYNC
);
1139 * If we can't acquire the flush lock, then the inode
1140 * is already being flushed so don't bother waiting.
1142 * If we can lock it then do a delwri flush so we can
1143 * combine multiple inode flushes in each disk write.
1145 } else if (xfs_iflock_nowait(ip
)) {
1146 error
= xfs_iflush(ip
, XFS_IFLUSH_DELWRI
);
1147 } else if (bypassed
) {
1152 if (lock_flags
!= 0) {
1153 xfs_iunlock(ip
, lock_flags
);
1158 * If we had to take a reference on the vnode
1159 * above, then wait until after we've unlocked
1160 * the inode to release the reference. This is
1161 * because we can be already holding the inode
1162 * lock when IRELE() calls xfs_inactive().
1164 * Make sure to drop the mount lock before calling
1165 * IRELE() so that we don't trip over ourselves if
1166 * we have to go for the mount lock again in the
1170 IPOINTER_INSERT(ip
, mp
);
1175 vnode_refed
= B_FALSE
;
1183 * bail out if the filesystem is corrupted.
1185 if (error
== EFSCORRUPTED
) {
1186 if (!mount_locked
) {
1187 XFS_MOUNT_ILOCK(mp
);
1188 IPOINTER_REMOVE(ip
, mp
);
1190 XFS_MOUNT_IUNLOCK(mp
);
1191 ASSERT(ipointer_in
== B_FALSE
);
1192 kmem_free(ipointer
, sizeof(xfs_iptr_t
));
1193 return XFS_ERROR(error
);
1196 /* Let other threads have a chance at the mount lock
1197 * if we have looped many times without dropping the
1200 if ((++preempt
& XFS_PREEMPT_MASK
) == 0) {
1202 IPOINTER_INSERT(ip
, mp
);
1206 if (mount_locked
== B_FALSE
) {
1207 XFS_MOUNT_ILOCK(mp
);
1208 mount_locked
= B_TRUE
;
1209 IPOINTER_REMOVE(ip
, mp
);
1213 ASSERT(ipointer_in
== B_FALSE
);
1216 } while (ip
!= mp
->m_inodes
);
1218 XFS_MOUNT_IUNLOCK(mp
);
1220 ASSERT(ipointer_in
== B_FALSE
);
1222 kmem_free(ipointer
, sizeof(xfs_iptr_t
));
1223 return XFS_ERROR(last_error
);
1227 * xfs sync routine for internal use
1229 * This routine supports all of the flags defined for the generic vfs_sync
1230 * interface as explained above under xfs_sync.
1241 uint log_flags
= XFS_LOG_FORCE
;
1243 xfs_buf_log_item_t
*bip
;
1246 * Sync out the log. This ensures that the log is periodically
1247 * flushed even if there is not enough activity to fill it up.
1249 if (flags
& SYNC_WAIT
)
1250 log_flags
|= XFS_LOG_SYNC
;
1252 xfs_log_force(mp
, (xfs_lsn_t
)0, log_flags
);
1254 if (flags
& (SYNC_ATTR
|SYNC_DELWRI
)) {
1255 if (flags
& SYNC_BDFLUSH
)
1256 xfs_finish_reclaim_all(mp
, 1);
1258 error
= xfs_sync_inodes(mp
, flags
, bypassed
);
1262 * Flushing out dirty data above probably generated more
1263 * log activity, so if this isn't vfs_sync() then flush
1266 if (flags
& SYNC_DELWRI
) {
1267 xfs_log_force(mp
, (xfs_lsn_t
)0, log_flags
);
1270 if (flags
& SYNC_FSDATA
) {
1272 * If this is vfs_sync() then only sync the superblock
1273 * if we can lock it without sleeping and it is not pinned.
1275 if (flags
& SYNC_BDFLUSH
) {
1276 bp
= xfs_getsb(mp
, XFS_BUF_TRYLOCK
);
1278 bip
= XFS_BUF_FSPRIVATE(bp
,xfs_buf_log_item_t
*);
1279 if ((bip
!= NULL
) &&
1280 xfs_buf_item_dirty(bip
)) {
1281 if (!(XFS_BUF_ISPINNED(bp
))) {
1283 error
= xfs_bwrite(mp
, bp
);
1292 bp
= xfs_getsb(mp
, 0);
1294 * If the buffer is pinned then push on the log so
1295 * we won't get stuck waiting in the write for
1296 * someone, maybe ourselves, to flush the log.
1297 * Even though we just pushed the log above, we
1298 * did not have the superblock buffer locked at
1299 * that point so it can become pinned in between
1302 if (XFS_BUF_ISPINNED(bp
))
1303 xfs_log_force(mp
, (xfs_lsn_t
)0, XFS_LOG_FORCE
);
1304 if (flags
& SYNC_WAIT
)
1305 XFS_BUF_UNASYNC(bp
);
1308 error
= xfs_bwrite(mp
, bp
);
1316 * If asked, update the disk superblock with incore counter values if we
1317 * are using non-persistent counters so that they don't get too far out
1318 * of sync if we crash or get a forced shutdown. We don't want to force
1319 * this to disk, just get a transaction into the iclogs....
1321 if (flags
& SYNC_SUPER
)
1322 xfs_log_sbcount(mp
, 0);
1325 * Now check to see if the log needs a "dummy" transaction.
1328 if (!(flags
& SYNC_REMOUNT
) && xfs_log_need_covered(mp
)) {
1333 * Put a dummy transaction in the log to tell
1334 * recovery that all others are OK.
1336 tp
= xfs_trans_alloc(mp
, XFS_TRANS_DUMMY1
);
1337 if ((error
= xfs_trans_reserve(tp
, 0,
1338 XFS_ICHANGE_LOG_RES(mp
),
1340 xfs_trans_cancel(tp
, 0);
1345 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1347 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
1348 xfs_trans_ihold(tp
, ip
);
1349 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1350 error
= xfs_trans_commit(tp
, 0);
1351 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1352 xfs_log_force(mp
, (xfs_lsn_t
)0, log_flags
);
1356 * When shutting down, we need to insure that the AIL is pushed
1357 * to disk or the filesystem can appear corrupt from the PROM.
1359 if ((flags
& (SYNC_CLOSE
|SYNC_WAIT
)) == (SYNC_CLOSE
|SYNC_WAIT
)) {
1360 XFS_bflush(mp
->m_ddev_targp
);
1361 if (mp
->m_rtdev_targp
) {
1362 XFS_bflush(mp
->m_rtdev_targp
);
1366 return XFS_ERROR(last_error
);