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_refcache.h"
47 #include "xfs_buf_item.h"
48 #include "xfs_log_priv.h"
49 #include "xfs_dir2_trace.h"
50 #include "xfs_extfree_item.h"
54 #include "xfs_mru_cache.h"
55 #include "xfs_filestream.h"
56 #include "xfs_fsops.h"
57 #include "xfs_vnodeops.h"
58 #include "xfs_vfsops.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_bmap_free_item_zone = kmem_zone_init(sizeof(xfs_bmap_free_item_t),
73 "xfs_bmap_free_item");
74 xfs_btree_cur_zone = kmem_zone_init(sizeof(xfs_btree_cur_t),
76 xfs_trans_zone = kmem_zone_init(sizeof(xfs_trans_t), "xfs_trans");
78 kmem_zone_init(sizeof(xfs_da_state_t), "xfs_da_state");
79 xfs_dabuf_zone = kmem_zone_init(sizeof(xfs_dabuf_t), "xfs_dabuf");
80 xfs_ifork_zone = kmem_zone_init(sizeof(xfs_ifork_t), "xfs_ifork");
81 xfs_acl_zone_init(xfs_acl_zone, "xfs_acl");
83 xfs_filestream_init();
86 * The size of the zone allocated buf log item is the maximum
87 * size possible under XFS. This wastes a little bit of memory,
88 * but it is much faster.
91 kmem_zone_init((sizeof(xfs_buf_log_item_t) +
92 (((XFS_MAX_BLOCKSIZE / XFS_BLI_CHUNK) /
93 NBWORD) * sizeof(int))),
96 kmem_zone_init((sizeof(xfs_efd_log_item_t) +
97 ((XFS_EFD_MAX_FAST_EXTENTS - 1) *
98 sizeof(xfs_extent_t))),
101 kmem_zone_init((sizeof(xfs_efi_log_item_t) +
102 ((XFS_EFI_MAX_FAST_EXTENTS - 1) *
103 sizeof(xfs_extent_t))),
107 * These zones warrant special memory allocator hints
110 kmem_zone_init_flags(sizeof(xfs_inode_t), "xfs_inode",
111 KM_ZONE_HWALIGN | KM_ZONE_RECLAIM |
112 KM_ZONE_SPREAD, NULL);
114 kmem_zone_init_flags(sizeof(xfs_inode_log_item_t), "xfs_ili",
115 KM_ZONE_SPREAD, NULL);
117 kmem_zone_init_flags(sizeof(xfs_icluster_t), "xfs_icluster",
118 KM_ZONE_SPREAD, NULL);
121 * Allocate global trace buffers.
123 #ifdef XFS_ALLOC_TRACE
124 xfs_alloc_trace_buf = ktrace_alloc(XFS_ALLOC_TRACE_SIZE, KM_SLEEP);
126 #ifdef XFS_BMAP_TRACE
127 xfs_bmap_trace_buf = ktrace_alloc(XFS_BMAP_TRACE_SIZE, KM_SLEEP);
129 #ifdef XFS_BMBT_TRACE
130 xfs_bmbt_trace_buf = ktrace_alloc(XFS_BMBT_TRACE_SIZE, KM_SLEEP);
132 #ifdef XFS_ATTR_TRACE
133 xfs_attr_trace_buf = ktrace_alloc(XFS_ATTR_TRACE_SIZE, KM_SLEEP);
135 #ifdef XFS_DIR2_TRACE
136 xfs_dir2_trace_buf = ktrace_alloc(XFS_DIR2_GTRACE_SIZE, KM_SLEEP);
141 #if (defined(DEBUG) || defined(INDUCE_IO_ERROR))
142 xfs_error_test_init();
143 #endif /* DEBUG || INDUCE_IO_ERROR */
146 xfs_sysctl_register();
153 extern kmem_zone_t *xfs_inode_zone;
154 extern kmem_zone_t *xfs_efd_zone;
155 extern kmem_zone_t *xfs_efi_zone;
156 extern kmem_zone_t *xfs_icluster_zone;
158 xfs_cleanup_procfs();
159 xfs_sysctl_unregister();
160 xfs_refcache_destroy();
161 xfs_filestream_uninit();
162 xfs_mru_cache_uninit();
163 xfs_acl_zone_destroy(xfs_acl_zone);
165 #ifdef XFS_DIR2_TRACE
166 ktrace_free(xfs_dir2_trace_buf);
168 #ifdef XFS_ATTR_TRACE
169 ktrace_free(xfs_attr_trace_buf);
171 #ifdef XFS_BMBT_TRACE
172 ktrace_free(xfs_bmbt_trace_buf);
174 #ifdef XFS_BMAP_TRACE
175 ktrace_free(xfs_bmap_trace_buf);
177 #ifdef XFS_ALLOC_TRACE
178 ktrace_free(xfs_alloc_trace_buf);
181 kmem_zone_destroy(xfs_bmap_free_item_zone);
182 kmem_zone_destroy(xfs_btree_cur_zone);
183 kmem_zone_destroy(xfs_inode_zone);
184 kmem_zone_destroy(xfs_trans_zone);
185 kmem_zone_destroy(xfs_da_state_zone);
186 kmem_zone_destroy(xfs_dabuf_zone);
187 kmem_zone_destroy(xfs_buf_item_zone);
188 kmem_zone_destroy(xfs_efd_zone);
189 kmem_zone_destroy(xfs_efi_zone);
190 kmem_zone_destroy(xfs_ifork_zone);
191 kmem_zone_destroy(xfs_ili_zone);
192 kmem_zone_destroy(xfs_icluster_zone);
198 * This function fills in xfs_mount_t fields based on mount args.
199 * Note: the superblock has _not_ yet been read in.
203 struct xfs_mount_args *ap,
204 struct xfs_mount *mp)
206 /* Values are in BBs */
207 if ((ap->flags & XFSMNT_NOALIGN) != XFSMNT_NOALIGN) {
209 * At this point the superblock has not been read
210 * in, therefore we do not know the block size.
211 * Before the mount call ends we will convert
214 mp->m_dalign = ap->sunit;
215 mp->m_swidth = ap->swidth;
218 if (ap->logbufs != -1 &&
220 (ap->logbufs < XLOG_MIN_ICLOGS ||
221 ap->logbufs > XLOG_MAX_ICLOGS)) {
223 "XFS: invalid logbufs value: %d [not %d-%d]",
224 ap->logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
225 return XFS_ERROR(EINVAL);
227 mp->m_logbufs = ap->logbufs;
228 if (ap->logbufsize != -1 &&
229 ap->logbufsize != 0 &&
230 (ap->logbufsize < XLOG_MIN_RECORD_BSIZE ||
231 ap->logbufsize > XLOG_MAX_RECORD_BSIZE ||
232 !is_power_of_2(ap->logbufsize))) {
234 "XFS: invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
236 return XFS_ERROR(EINVAL);
238 mp->m_logbsize = ap->logbufsize;
239 mp->m_fsname_len = strlen(ap->fsname) + 1;
240 mp->m_fsname = kmem_alloc(mp->m_fsname_len, KM_SLEEP);
241 strcpy(mp->m_fsname, ap->fsname);
243 mp->m_rtname = kmem_alloc(strlen(ap->rtname) + 1, KM_SLEEP);
244 strcpy(mp->m_rtname, ap->rtname);
246 if (ap->logname[0]) {
247 mp->m_logname = kmem_alloc(strlen(ap->logname) + 1, KM_SLEEP);
248 strcpy(mp->m_logname, ap->logname);
251 if (ap->flags & XFSMNT_WSYNC)
252 mp->m_flags |= XFS_MOUNT_WSYNC;
254 if (ap->flags & XFSMNT_INO64) {
255 mp->m_flags |= XFS_MOUNT_INO64;
256 mp->m_inoadd = XFS_INO64_OFFSET;
259 if (ap->flags & XFSMNT_RETERR)
260 mp->m_flags |= XFS_MOUNT_RETERR;
261 if (ap->flags & XFSMNT_NOALIGN)
262 mp->m_flags |= XFS_MOUNT_NOALIGN;
263 if (ap->flags & XFSMNT_SWALLOC)
264 mp->m_flags |= XFS_MOUNT_SWALLOC;
265 if (ap->flags & XFSMNT_OSYNCISOSYNC)
266 mp->m_flags |= XFS_MOUNT_OSYNCISOSYNC;
267 if (ap->flags & XFSMNT_32BITINODES)
268 mp->m_flags |= XFS_MOUNT_32BITINODES;
270 if (ap->flags & XFSMNT_IOSIZE) {
271 if (ap->iosizelog > XFS_MAX_IO_LOG ||
272 ap->iosizelog < XFS_MIN_IO_LOG) {
274 "XFS: invalid log iosize: %d [not %d-%d]",
275 ap->iosizelog, XFS_MIN_IO_LOG,
277 return XFS_ERROR(EINVAL);
280 mp->m_flags |= XFS_MOUNT_DFLT_IOSIZE;
281 mp->m_readio_log = mp->m_writeio_log = ap->iosizelog;
284 if (ap->flags & XFSMNT_IKEEP)
285 mp->m_flags |= XFS_MOUNT_IKEEP;
286 if (ap->flags & XFSMNT_DIRSYNC)
287 mp->m_flags |= XFS_MOUNT_DIRSYNC;
288 if (ap->flags & XFSMNT_ATTR2)
289 mp->m_flags |= XFS_MOUNT_ATTR2;
291 if (ap->flags2 & XFSMNT2_COMPAT_IOSIZE)
292 mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE;
295 * no recovery flag requires a read-only mount
297 if (ap->flags & XFSMNT_NORECOVERY) {
298 if (!(mp->m_flags & XFS_MOUNT_RDONLY)) {
300 "XFS: tried to mount a FS read-write without recovery!");
301 return XFS_ERROR(EINVAL);
303 mp->m_flags |= XFS_MOUNT_NORECOVERY;
306 if (ap->flags & XFSMNT_NOUUID)
307 mp->m_flags |= XFS_MOUNT_NOUUID;
308 if (ap->flags & XFSMNT_BARRIER)
309 mp->m_flags |= XFS_MOUNT_BARRIER;
311 mp->m_flags &= ~XFS_MOUNT_BARRIER;
313 if (ap->flags2 & XFSMNT2_FILESTREAMS)
314 mp->m_flags |= XFS_MOUNT_FILESTREAMS;
316 if (ap->flags & XFSMNT_DMAPI)
317 mp->m_flags |= XFS_MOUNT_DMAPI;
322 * This function fills in xfs_mount_t fields based on mount args.
323 * Note: the superblock _has_ now been read in.
327 struct xfs_mount_args *ap,
328 struct xfs_mount *mp)
330 int ronly = (mp->m_flags & XFS_MOUNT_RDONLY);
332 /* Fail a mount where the logbuf is smaller then the log stripe */
333 if (xfs_sb_version_haslogv2(&mp->m_sb)) {
334 if ((ap->logbufsize <= 0) &&
335 (mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE)) {
336 mp->m_logbsize = mp->m_sb.sb_logsunit;
337 } else if (ap->logbufsize > 0 &&
338 ap->logbufsize < mp->m_sb.sb_logsunit) {
340 "XFS: logbuf size must be greater than or equal to log stripe size");
341 return XFS_ERROR(EINVAL);
344 /* Fail a mount if the logbuf is larger than 32K */
345 if (ap->logbufsize > XLOG_BIG_RECORD_BSIZE) {
347 "XFS: logbuf size for version 1 logs must be 16K or 32K");
348 return XFS_ERROR(EINVAL);
352 if (xfs_sb_version_hasattr2(&mp->m_sb))
353 mp->m_flags |= XFS_MOUNT_ATTR2;
356 * prohibit r/w mounts of read-only filesystems
358 if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !ronly) {
360 "XFS: cannot mount a read-only filesystem as read-write");
361 return XFS_ERROR(EROFS);
365 * check for shared mount.
367 if (ap->flags & XFSMNT_SHARED) {
368 if (!xfs_sb_version_hasshared(&mp->m_sb))
369 return XFS_ERROR(EINVAL);
372 * For IRIX 6.5, shared mounts must have the shared
373 * version bit set, have the persistent readonly
374 * field set, must be version 0 and can only be mounted
377 if (!ronly || !(mp->m_sb.sb_flags & XFS_SBF_READONLY) ||
378 (mp->m_sb.sb_shared_vn != 0))
379 return XFS_ERROR(EINVAL);
381 mp->m_flags |= XFS_MOUNT_SHARED;
384 * Shared XFS V0 can't deal with DMI. Return EINVAL.
386 if (mp->m_sb.sb_shared_vn == 0 && (ap->flags & XFSMNT_DMAPI))
387 return XFS_ERROR(EINVAL);
390 if (ap->flags & XFSMNT_UQUOTA) {
391 mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE);
392 if (ap->flags & XFSMNT_UQUOTAENF)
393 mp->m_qflags |= XFS_UQUOTA_ENFD;
396 if (ap->flags & XFSMNT_GQUOTA) {
397 mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE);
398 if (ap->flags & XFSMNT_GQUOTAENF)
399 mp->m_qflags |= XFS_OQUOTA_ENFD;
400 } else if (ap->flags & XFSMNT_PQUOTA) {
401 mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE);
402 if (ap->flags & XFSMNT_PQUOTAENF)
403 mp->m_qflags |= XFS_OQUOTA_ENFD;
412 * The file system configurations are:
413 * (1) device (partition) with data and internal log
414 * (2) logical volume with data and log subvolumes.
415 * (3) logical volume with data, log, and realtime subvolumes.
417 * We only have to handle opening the log and realtime volumes here if
418 * they are present. The data subvolume has already been opened by
419 * get_sb_bdev() and is stored in vfsp->vfs_super->s_bdev.
423 struct xfs_mount *mp,
424 struct xfs_mount_args *args,
427 struct block_device *ddev, *logdev, *rtdev;
428 int flags = 0, error;
430 ddev = mp->m_super->s_bdev;
431 logdev = rtdev = NULL;
433 error = xfs_dmops_get(mp, args);
436 error = xfs_qmops_get(mp, args);
440 if (args->flags & XFSMNT_QUIET)
441 flags |= XFS_MFSI_QUIET;
444 * Open real time and log devices - order is important.
446 if (args->logname[0]) {
447 error = xfs_blkdev_get(mp, args->logname, &logdev);
451 if (args->rtname[0]) {
452 error = xfs_blkdev_get(mp, args->rtname, &rtdev);
454 xfs_blkdev_put(logdev);
458 if (rtdev == ddev || rtdev == logdev) {
460 "XFS: Cannot mount filesystem with identical rtdev and ddev/logdev.");
461 xfs_blkdev_put(logdev);
462 xfs_blkdev_put(rtdev);
468 * Setup xfs_mount buffer target pointers
471 mp->m_ddev_targp = xfs_alloc_buftarg(ddev, 0);
472 if (!mp->m_ddev_targp) {
473 xfs_blkdev_put(logdev);
474 xfs_blkdev_put(rtdev);
478 mp->m_rtdev_targp = xfs_alloc_buftarg(rtdev, 1);
479 if (!mp->m_rtdev_targp) {
480 xfs_blkdev_put(logdev);
481 xfs_blkdev_put(rtdev);
485 mp->m_logdev_targp = (logdev && logdev != ddev) ?
486 xfs_alloc_buftarg(logdev, 1) : mp->m_ddev_targp;
487 if (!mp->m_logdev_targp) {
488 xfs_blkdev_put(logdev);
489 xfs_blkdev_put(rtdev);
494 * Setup flags based on mount(2) options and then the superblock
496 error = xfs_start_flags(args, mp);
499 error = xfs_readsb(mp, flags);
502 error = xfs_finish_flags(args, mp);
507 * Setup xfs_mount buffer target pointers based on superblock
509 error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_blocksize,
510 mp->m_sb.sb_sectsize);
511 if (!error && logdev && logdev != ddev) {
512 unsigned int log_sector_size = BBSIZE;
514 if (xfs_sb_version_hassector(&mp->m_sb))
515 log_sector_size = mp->m_sb.sb_logsectsize;
516 error = xfs_setsize_buftarg(mp->m_logdev_targp,
517 mp->m_sb.sb_blocksize,
521 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
522 mp->m_sb.sb_blocksize,
523 mp->m_sb.sb_sectsize);
527 if (mp->m_flags & XFS_MOUNT_BARRIER)
528 xfs_mountfs_check_barriers(mp);
530 if ((error = xfs_filestream_mount(mp)))
533 error = xfs_mountfs(mp, flags);
537 XFS_SEND_MOUNT(mp, DM_RIGHT_NULL, args->mtpt, args->fsname);
545 xfs_binval(mp->m_ddev_targp);
546 if (logdev && logdev != ddev)
547 xfs_binval(mp->m_logdev_targp);
549 xfs_binval(mp->m_rtdev_targp);
551 xfs_unmountfs_close(mp, credp);
565 int unmount_event_wanted = 0;
566 int unmount_event_flags = 0;
567 int xfs_unmountfs_needed = 0;
574 if (mp->m_flags & XFS_MOUNT_DMAPI) {
575 error = XFS_SEND_PREUNMOUNT(mp,
576 rvp, DM_RIGHT_NULL, rvp, DM_RIGHT_NULL,
578 (mp->m_dmevmask & (1<<DM_EVENT_PREUNMOUNT))?
579 0:DM_FLAGS_UNWANTED);
581 return XFS_ERROR(error);
582 unmount_event_wanted = 1;
583 unmount_event_flags = (mp->m_dmevmask & (1<<DM_EVENT_UNMOUNT))?
584 0 : DM_FLAGS_UNWANTED;
588 * First blow any referenced inode from this file system
589 * out of the reference cache, and delete the timer.
591 xfs_refcache_purge_mp(mp);
594 * Blow away any referenced inode in the filestreams cache.
595 * This can and will cause log traffic as inodes go inactive
598 xfs_filestream_unmount(mp);
600 XFS_bflush(mp->m_ddev_targp);
601 error = xfs_unmount_flush(mp, 0);
605 ASSERT(vn_count(rvp) == 1);
608 * Drop the reference count
613 * If we're forcing a shutdown, typically because of a media error,
614 * we want to make sure we invalidate dirty pages that belong to
615 * referenced vnodes as well.
617 if (XFS_FORCED_SHUTDOWN(mp)) {
618 error = xfs_sync(mp, SYNC_WAIT | SYNC_CLOSE);
619 ASSERT(error != EFSCORRUPTED);
621 xfs_unmountfs_needed = 1;
624 /* Send DMAPI event, if required.
625 * Then do xfs_unmountfs() if needed.
626 * Then return error (or zero).
628 if (unmount_event_wanted) {
629 /* Note: mp structure must still exist for
630 * XFS_SEND_UNMOUNT() call.
632 XFS_SEND_UNMOUNT(mp, error == 0 ? rvp : NULL,
633 DM_RIGHT_NULL, 0, error, unmount_event_flags);
635 if (xfs_unmountfs_needed) {
637 * Call common unmount function to flush to disk
638 * and free the super block buffer & mount structures.
640 xfs_unmountfs(mp, credp);
643 kmem_free(mp, sizeof(xfs_mount_t));
646 return XFS_ERROR(error);
653 int count = 0, pincount;
655 xfs_refcache_purge_mp(mp);
656 xfs_flush_buftarg(mp->m_ddev_targp, 0);
657 xfs_finish_reclaim_all(mp, 0);
659 /* This loop must run at least twice.
660 * The first instance of the loop will flush
661 * most meta data but that will generate more
662 * meta data (typically directory updates).
663 * Which then must be flushed and logged before
664 * we can write the unmount record.
667 xfs_syncsub(mp, SYNC_INODE_QUIESCE, NULL);
668 pincount = xfs_flush_buftarg(mp->m_ddev_targp, 1);
679 * Second stage of a quiesce. The data is already synced, now we have to take
680 * care of the metadata. New transactions are already blocked, so we need to
681 * wait for any remaining transactions to drain out before proceding.
687 /* wait for all modifications to complete */
688 while (atomic_read(&mp->m_active_trans) > 0)
691 /* flush inodes and push all remaining buffers out to disk */
694 ASSERT_ALWAYS(atomic_read(&mp->m_active_trans) == 0);
696 /* Push the superblock and write an unmount record */
697 xfs_log_sbcount(mp, 1);
698 xfs_log_unmount_write(mp);
699 xfs_unmountfs_writesb(mp);
704 struct xfs_mount *mp,
706 struct xfs_mount_args *args)
708 if (!(*flags & MS_RDONLY)) { /* rw/ro -> rw */
709 if (mp->m_flags & XFS_MOUNT_RDONLY)
710 mp->m_flags &= ~XFS_MOUNT_RDONLY;
711 if (args->flags & XFSMNT_BARRIER) {
712 mp->m_flags |= XFS_MOUNT_BARRIER;
713 xfs_mountfs_check_barriers(mp);
715 mp->m_flags &= ~XFS_MOUNT_BARRIER;
717 } else if (!(mp->m_flags & XFS_MOUNT_RDONLY)) { /* rw -> ro */
718 xfs_filestream_flush(mp);
719 xfs_sync(mp, SYNC_DATA_QUIESCE);
720 xfs_attr_quiesce(mp);
721 mp->m_flags |= XFS_MOUNT_RDONLY;
727 * xfs_unmount_flush implements a set of flush operation on special
728 * inodes, which are needed as a separate set of operations so that
729 * they can be called as part of relocation process.
733 xfs_mount_t *mp, /* Mount structure we are getting
735 int relocation) /* Called from vfs relocation. */
737 xfs_inode_t *rip = mp->m_rootip;
739 xfs_inode_t *rsumip = NULL;
740 bhv_vnode_t *rvp = XFS_ITOV(rip);
743 xfs_ilock(rip, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
747 * Flush out the real time inodes.
749 if ((rbmip = mp->m_rbmip) != NULL) {
750 xfs_ilock(rbmip, XFS_ILOCK_EXCL);
752 error = xfs_iflush(rbmip, XFS_IFLUSH_SYNC);
753 xfs_iunlock(rbmip, XFS_ILOCK_EXCL);
755 if (error == EFSCORRUPTED)
758 ASSERT(vn_count(XFS_ITOV(rbmip)) == 1);
760 rsumip = mp->m_rsumip;
761 xfs_ilock(rsumip, XFS_ILOCK_EXCL);
763 error = xfs_iflush(rsumip, XFS_IFLUSH_SYNC);
764 xfs_iunlock(rsumip, XFS_ILOCK_EXCL);
766 if (error == EFSCORRUPTED)
769 ASSERT(vn_count(XFS_ITOV(rsumip)) == 1);
773 * Synchronously flush root inode to disk
775 error = xfs_iflush(rip, XFS_IFLUSH_SYNC);
776 if (error == EFSCORRUPTED)
779 if (vn_count(rvp) != 1 && !relocation) {
780 xfs_iunlock(rip, XFS_ILOCK_EXCL);
781 return XFS_ERROR(EBUSY);
785 * Release dquot that rootinode, rbmino and rsumino might be holding,
786 * flush and purge the quota inodes.
788 error = XFS_QM_UNMOUNT(mp);
789 if (error == EFSCORRUPTED)
793 VN_RELE(XFS_ITOV(rbmip));
794 VN_RELE(XFS_ITOV(rsumip));
797 xfs_iunlock(rip, XFS_ILOCK_EXCL);
804 xfs_iunlock(rip, XFS_ILOCK_EXCL);
806 return XFS_ERROR(EFSCORRUPTED);
810 * xfs_sync flushes any pending I/O to file system vfsp.
812 * This routine is called by vfs_sync() to make sure that things make it
813 * out to disk eventually, on sync() system calls to flush out everything,
814 * and when the file system is unmounted. For the vfs_sync() case, all
815 * we really need to do is sync out the log to make all of our meta-data
816 * updates permanent (except for timestamps). For calls from pflushd(),
817 * dirty pages are kept moving by calling pdflush() on the inodes
818 * containing them. We also flush the inodes that we can lock without
819 * sleeping and the superblock if we can lock it without sleeping from
820 * vfs_sync() so that items at the tail of the log are always moving out.
823 * SYNC_BDFLUSH - We're being called from vfs_sync() so we don't want
824 * to sleep if we can help it. All we really need
825 * to do is ensure that the log is synced at least
826 * periodically. We also push the inodes and
827 * superblock if we can lock them without sleeping
828 * and they are not pinned.
829 * SYNC_ATTR - We need to flush the inodes. If SYNC_BDFLUSH is not
830 * set, then we really want to lock each inode and flush
832 * SYNC_WAIT - All the flushes that take place in this call should
834 * SYNC_DELWRI - This tells us to push dirty pages associated with
835 * inodes. SYNC_WAIT and SYNC_BDFLUSH are used to
836 * determine if they should be flushed sync, async, or
838 * SYNC_CLOSE - This flag is passed when the system is being
839 * unmounted. We should sync and invalidate everything.
840 * SYNC_FSDATA - This indicates that the caller would like to make
841 * sure the superblock is safe on disk. We can ensure
842 * this by simply making sure the log gets flushed
843 * if SYNC_BDFLUSH is set, and by actually writing it
845 * SYNC_IOWAIT - The caller wants us to wait for all data I/O to complete
846 * before we return (including direct I/O). Forms the drain
847 * side of the write barrier needed to safely quiesce the
859 * Get the Quota Manager to flush the dquots.
861 * If XFS quota support is not enabled or this filesystem
862 * instance does not use quotas XFS_QM_DQSYNC will always
865 error = XFS_QM_DQSYNC(mp, flags);
868 * If we got an IO error, we will be shutting down.
869 * So, there's nothing more for us to do here.
871 ASSERT(error != EIO || XFS_FORCED_SHUTDOWN(mp));
872 if (XFS_FORCED_SHUTDOWN(mp))
873 return XFS_ERROR(error);
876 if (flags & SYNC_IOWAIT)
877 xfs_filestream_flush(mp);
879 return xfs_syncsub(mp, flags, NULL);
883 * xfs sync routine for internal use
885 * This routine supports all of the flags defined for the generic vfs_sync
886 * interface as explained above under xfs_sync.
895 xfs_inode_t *ip = NULL;
896 bhv_vnode_t *vp = NULL;
901 uint base_lock_flags;
902 boolean_t mount_locked;
903 boolean_t vnode_refed;
905 xfs_iptr_t *ipointer;
907 boolean_t ipointer_in = B_FALSE;
909 #define IPOINTER_SET ipointer_in = B_TRUE
910 #define IPOINTER_CLR ipointer_in = B_FALSE
917 /* Insert a marker record into the inode list after inode ip. The list
918 * must be locked when this is called. After the call the list will no
921 #define IPOINTER_INSERT(ip, mp) { \
922 ASSERT(ipointer_in == B_FALSE); \
923 ipointer->ip_mnext = ip->i_mnext; \
924 ipointer->ip_mprev = ip; \
925 ip->i_mnext = (xfs_inode_t *)ipointer; \
926 ipointer->ip_mnext->i_mprev = (xfs_inode_t *)ipointer; \
928 XFS_MOUNT_IUNLOCK(mp); \
929 mount_locked = B_FALSE; \
933 /* Remove the marker from the inode list. If the marker was the only item
934 * in the list then there are no remaining inodes and we should zero out
935 * the whole list. If we are the current head of the list then move the head
938 #define IPOINTER_REMOVE(ip, mp) { \
939 ASSERT(ipointer_in == B_TRUE); \
940 if (ipointer->ip_mnext != (xfs_inode_t *)ipointer) { \
941 ip = ipointer->ip_mnext; \
942 ip->i_mprev = ipointer->ip_mprev; \
943 ipointer->ip_mprev->i_mnext = ip; \
944 if (mp->m_inodes == (xfs_inode_t *)ipointer) { \
948 ASSERT(mp->m_inodes == (xfs_inode_t *)ipointer); \
949 mp->m_inodes = NULL; \
955 #define XFS_PREEMPT_MASK 0x7f
957 ASSERT(!(flags & SYNC_BDFLUSH));
961 if (mp->m_flags & XFS_MOUNT_RDONLY)
967 /* Allocate a reference marker */
968 ipointer = (xfs_iptr_t *)kmem_zalloc(sizeof(xfs_iptr_t), KM_SLEEP);
970 fflag = XFS_B_ASYNC; /* default is don't wait */
971 if (flags & SYNC_DELWRI)
972 fflag = XFS_B_DELWRI;
973 if (flags & SYNC_WAIT)
974 fflag = 0; /* synchronous overrides all */
976 base_lock_flags = XFS_ILOCK_SHARED;
977 if (flags & (SYNC_DELWRI | SYNC_CLOSE)) {
979 * We need the I/O lock if we're going to call any of
980 * the flush/inval routines.
982 base_lock_flags |= XFS_IOLOCK_SHARED;
989 mount_locked = B_TRUE;
990 vnode_refed = B_FALSE;
995 ASSERT(ipointer_in == B_FALSE);
996 ASSERT(vnode_refed == B_FALSE);
998 lock_flags = base_lock_flags;
1001 * There were no inodes in the list, just break out
1009 * We found another sync thread marker - skip it
1011 if (ip->i_mount == NULL) {
1016 vp = XFS_ITOV_NULL(ip);
1019 * If the vnode is gone then this is being torn down,
1020 * call reclaim if it is flushed, else let regular flush
1021 * code deal with it later in the loop.
1025 /* Skip ones already in reclaim */
1026 if (ip->i_flags & XFS_IRECLAIM) {
1030 if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0) {
1032 } else if ((xfs_ipincount(ip) == 0) &&
1033 xfs_iflock_nowait(ip)) {
1034 IPOINTER_INSERT(ip, mp);
1036 xfs_finish_reclaim(ip, 1,
1037 XFS_IFLUSH_DELWRI_ELSE_ASYNC);
1039 XFS_MOUNT_ILOCK(mp);
1040 mount_locked = B_TRUE;
1041 IPOINTER_REMOVE(ip, mp);
1043 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1054 if (XFS_FORCED_SHUTDOWN(mp) && !(flags & SYNC_CLOSE)) {
1055 XFS_MOUNT_IUNLOCK(mp);
1056 kmem_free(ipointer, sizeof(xfs_iptr_t));
1061 * Try to lock without sleeping. We're out of order with
1062 * the inode list lock here, so if we fail we need to drop
1063 * the mount lock and try again. If we're called from
1064 * bdflush() here, then don't bother.
1066 * The inode lock here actually coordinates with the
1067 * almost spurious inode lock in xfs_ireclaim() to prevent
1068 * the vnode we handle here without a reference from
1069 * being freed while we reference it. If we lock the inode
1070 * while it's on the mount list here, then the spurious inode
1071 * lock in xfs_ireclaim() after the inode is pulled from
1072 * the mount list will sleep until we release it here.
1073 * This keeps the vnode from being freed while we reference
1076 if (xfs_ilock_nowait(ip, lock_flags) == 0) {
1088 IPOINTER_INSERT(ip, mp);
1089 xfs_ilock(ip, lock_flags);
1091 ASSERT(vp == XFS_ITOV(ip));
1092 ASSERT(ip->i_mount == mp);
1094 vnode_refed = B_TRUE;
1097 /* From here on in the loop we may have a marker record
1098 * in the inode list.
1102 * If we have to flush data or wait for I/O completion
1103 * we need to drop the ilock that we currently hold.
1104 * If we need to drop the lock, insert a marker if we
1105 * have not already done so.
1107 if ((flags & (SYNC_CLOSE|SYNC_IOWAIT)) ||
1108 ((flags & SYNC_DELWRI) && VN_DIRTY(vp))) {
1110 IPOINTER_INSERT(ip, mp);
1112 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1114 if (flags & SYNC_CLOSE) {
1115 /* Shutdown case. Flush and invalidate. */
1116 if (XFS_FORCED_SHUTDOWN(mp))
1117 xfs_tosspages(ip, 0, -1,
1120 error = xfs_flushinval_pages(ip,
1122 } else if ((flags & SYNC_DELWRI) && VN_DIRTY(vp)) {
1123 error = xfs_flush_pages(ip, 0,
1124 -1, fflag, FI_NONE);
1128 * When freezing, we need to wait ensure all I/O (including direct
1129 * I/O) is complete to ensure no further data modification can take
1130 * place after this point
1132 if (flags & SYNC_IOWAIT)
1135 xfs_ilock(ip, XFS_ILOCK_SHARED);
1138 if ((flags & SYNC_ATTR) &&
1139 (ip->i_update_core ||
1140 (ip->i_itemp && ip->i_itemp->ili_format.ilf_fields))) {
1142 IPOINTER_INSERT(ip, mp);
1144 if (flags & SYNC_WAIT) {
1146 error = xfs_iflush(ip, XFS_IFLUSH_SYNC);
1149 * If we can't acquire the flush lock, then the inode
1150 * is already being flushed so don't bother waiting.
1152 * If we can lock it then do a delwri flush so we can
1153 * combine multiple inode flushes in each disk write.
1155 } else if (xfs_iflock_nowait(ip)) {
1156 error = xfs_iflush(ip, XFS_IFLUSH_DELWRI);
1157 } else if (bypassed) {
1162 if (lock_flags != 0) {
1163 xfs_iunlock(ip, lock_flags);
1168 * If we had to take a reference on the vnode
1169 * above, then wait until after we've unlocked
1170 * the inode to release the reference. This is
1171 * because we can be already holding the inode
1172 * lock when VN_RELE() calls xfs_inactive().
1174 * Make sure to drop the mount lock before calling
1175 * VN_RELE() so that we don't trip over ourselves if
1176 * we have to go for the mount lock again in the
1180 IPOINTER_INSERT(ip, mp);
1185 vnode_refed = B_FALSE;
1193 * bail out if the filesystem is corrupted.
1195 if (error == EFSCORRUPTED) {
1196 if (!mount_locked) {
1197 XFS_MOUNT_ILOCK(mp);
1198 IPOINTER_REMOVE(ip, mp);
1200 XFS_MOUNT_IUNLOCK(mp);
1201 ASSERT(ipointer_in == B_FALSE);
1202 kmem_free(ipointer, sizeof(xfs_iptr_t));
1203 return XFS_ERROR(error);
1206 /* Let other threads have a chance at the mount lock
1207 * if we have looped many times without dropping the
1210 if ((++preempt & XFS_PREEMPT_MASK) == 0) {
1212 IPOINTER_INSERT(ip, mp);
1216 if (mount_locked == B_FALSE) {
1217 XFS_MOUNT_ILOCK(mp);
1218 mount_locked = B_TRUE;
1219 IPOINTER_REMOVE(ip, mp);
1223 ASSERT(ipointer_in == B_FALSE);
1226 } while (ip != mp->m_inodes);
1228 XFS_MOUNT_IUNLOCK(mp);
1230 ASSERT(ipointer_in == B_FALSE);
1232 kmem_free(ipointer, sizeof(xfs_iptr_t));
1233 return XFS_ERROR(last_error);
1237 * xfs sync routine for internal use
1239 * This routine supports all of the flags defined for the generic vfs_sync
1240 * interface as explained above under xfs_sync.
1251 uint log_flags = XFS_LOG_FORCE;
1253 xfs_buf_log_item_t *bip;
1256 * Sync out the log. This ensures that the log is periodically
1257 * flushed even if there is not enough activity to fill it up.
1259 if (flags & SYNC_WAIT)
1260 log_flags |= XFS_LOG_SYNC;
1262 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1264 if (flags & (SYNC_ATTR|SYNC_DELWRI)) {
1265 if (flags & SYNC_BDFLUSH)
1266 xfs_finish_reclaim_all(mp, 1);
1268 error = xfs_sync_inodes(mp, flags, bypassed);
1272 * Flushing out dirty data above probably generated more
1273 * log activity, so if this isn't vfs_sync() then flush
1276 if (flags & SYNC_DELWRI) {
1277 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1280 if (flags & SYNC_FSDATA) {
1282 * If this is vfs_sync() then only sync the superblock
1283 * if we can lock it without sleeping and it is not pinned.
1285 if (flags & SYNC_BDFLUSH) {
1286 bp = xfs_getsb(mp, XFS_BUF_TRYLOCK);
1288 bip = XFS_BUF_FSPRIVATE(bp,xfs_buf_log_item_t*);
1289 if ((bip != NULL) &&
1290 xfs_buf_item_dirty(bip)) {
1291 if (!(XFS_BUF_ISPINNED(bp))) {
1293 error = xfs_bwrite(mp, bp);
1302 bp = xfs_getsb(mp, 0);
1304 * If the buffer is pinned then push on the log so
1305 * we won't get stuck waiting in the write for
1306 * someone, maybe ourselves, to flush the log.
1307 * Even though we just pushed the log above, we
1308 * did not have the superblock buffer locked at
1309 * that point so it can become pinned in between
1312 if (XFS_BUF_ISPINNED(bp))
1313 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
1314 if (flags & SYNC_WAIT)
1315 XFS_BUF_UNASYNC(bp);
1318 error = xfs_bwrite(mp, bp);
1326 * If this is the periodic sync, then kick some entries out of
1327 * the reference cache. This ensures that idle entries are
1328 * eventually kicked out of the cache.
1330 if (flags & SYNC_REFCACHE) {
1331 if (flags & SYNC_WAIT)
1332 xfs_refcache_purge_mp(mp);
1334 xfs_refcache_purge_some(mp);
1338 * If asked, update the disk superblock with incore counter values if we
1339 * are using non-persistent counters so that they don't get too far out
1340 * of sync if we crash or get a forced shutdown. We don't want to force
1341 * this to disk, just get a transaction into the iclogs....
1343 if (flags & SYNC_SUPER)
1344 xfs_log_sbcount(mp, 0);
1347 * Now check to see if the log needs a "dummy" transaction.
1350 if (!(flags & SYNC_REMOUNT) && xfs_log_need_covered(mp)) {
1355 * Put a dummy transaction in the log to tell
1356 * recovery that all others are OK.
1358 tp = xfs_trans_alloc(mp, XFS_TRANS_DUMMY1);
1359 if ((error = xfs_trans_reserve(tp, 0,
1360 XFS_ICHANGE_LOG_RES(mp),
1362 xfs_trans_cancel(tp, 0);
1367 xfs_ilock(ip, XFS_ILOCK_EXCL);
1369 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1370 xfs_trans_ihold(tp, ip);
1371 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1372 error = xfs_trans_commit(tp, 0);
1373 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1374 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1378 * When shutting down, we need to insure that the AIL is pushed
1379 * to disk or the filesystem can appear corrupt from the PROM.
1381 if ((flags & (SYNC_CLOSE|SYNC_WAIT)) == (SYNC_CLOSE|SYNC_WAIT)) {
1382 XFS_bflush(mp->m_ddev_targp);
1383 if (mp->m_rtdev_targp) {
1384 XFS_bflush(mp->m_rtdev_targp);
1388 return XFS_ERROR(last_error);