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"
63 #ifdef XFS_DABUF_DEBUG
64 extern spinlock_t xfs_dabuf_global_lock;
65 spin_lock_init(&xfs_dabuf_global_lock);
69 * Initialize all of the zone allocators we use.
71 xfs_bmap_free_item_zone = kmem_zone_init(sizeof(xfs_bmap_free_item_t),
72 "xfs_bmap_free_item");
73 xfs_btree_cur_zone = kmem_zone_init(sizeof(xfs_btree_cur_t),
75 xfs_trans_zone = kmem_zone_init(sizeof(xfs_trans_t), "xfs_trans");
77 kmem_zone_init(sizeof(xfs_da_state_t), "xfs_da_state");
78 xfs_dabuf_zone = kmem_zone_init(sizeof(xfs_dabuf_t), "xfs_dabuf");
79 xfs_ifork_zone = kmem_zone_init(sizeof(xfs_ifork_t), "xfs_ifork");
80 xfs_acl_zone_init(xfs_acl_zone, "xfs_acl");
82 xfs_filestream_init();
85 * The size of the zone allocated buf log item is the maximum
86 * size possible under XFS. This wastes a little bit of memory,
87 * but it is much faster.
90 kmem_zone_init((sizeof(xfs_buf_log_item_t) +
91 (((XFS_MAX_BLOCKSIZE / XFS_BLI_CHUNK) /
92 NBWORD) * sizeof(int))),
95 kmem_zone_init((sizeof(xfs_efd_log_item_t) +
96 ((XFS_EFD_MAX_FAST_EXTENTS - 1) *
97 sizeof(xfs_extent_t))),
100 kmem_zone_init((sizeof(xfs_efi_log_item_t) +
101 ((XFS_EFI_MAX_FAST_EXTENTS - 1) *
102 sizeof(xfs_extent_t))),
106 * These zones warrant special memory allocator hints
109 kmem_zone_init_flags(sizeof(xfs_inode_t), "xfs_inode",
110 KM_ZONE_HWALIGN | KM_ZONE_RECLAIM |
111 KM_ZONE_SPREAD, NULL);
113 kmem_zone_init_flags(sizeof(xfs_inode_log_item_t), "xfs_ili",
114 KM_ZONE_SPREAD, NULL);
116 kmem_zone_init_flags(sizeof(xfs_icluster_t), "xfs_icluster",
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;
155 extern kmem_zone_t *xfs_icluster_zone;
157 xfs_cleanup_procfs();
158 xfs_sysctl_unregister();
159 xfs_filestream_uninit();
160 xfs_mru_cache_uninit();
161 xfs_acl_zone_destroy(xfs_acl_zone);
163 #ifdef XFS_DIR2_TRACE
164 ktrace_free(xfs_dir2_trace_buf);
166 #ifdef XFS_ATTR_TRACE
167 ktrace_free(xfs_attr_trace_buf);
169 #ifdef XFS_BMBT_TRACE
170 ktrace_free(xfs_bmbt_trace_buf);
172 #ifdef XFS_BMAP_TRACE
173 ktrace_free(xfs_bmap_trace_buf);
175 #ifdef XFS_ALLOC_TRACE
176 ktrace_free(xfs_alloc_trace_buf);
179 kmem_zone_destroy(xfs_bmap_free_item_zone);
180 kmem_zone_destroy(xfs_btree_cur_zone);
181 kmem_zone_destroy(xfs_inode_zone);
182 kmem_zone_destroy(xfs_trans_zone);
183 kmem_zone_destroy(xfs_da_state_zone);
184 kmem_zone_destroy(xfs_dabuf_zone);
185 kmem_zone_destroy(xfs_buf_item_zone);
186 kmem_zone_destroy(xfs_efd_zone);
187 kmem_zone_destroy(xfs_efi_zone);
188 kmem_zone_destroy(xfs_ifork_zone);
189 kmem_zone_destroy(xfs_ili_zone);
190 kmem_zone_destroy(xfs_icluster_zone);
196 * This function fills in xfs_mount_t fields based on mount args.
197 * Note: the superblock has _not_ yet been read in.
201 struct xfs_mount_args *ap,
202 struct xfs_mount *mp)
204 /* Values are in BBs */
205 if ((ap->flags & XFSMNT_NOALIGN) != XFSMNT_NOALIGN) {
207 * At this point the superblock has not been read
208 * in, therefore we do not know the block size.
209 * Before the mount call ends we will convert
212 mp->m_dalign = ap->sunit;
213 mp->m_swidth = ap->swidth;
216 if (ap->logbufs != -1 &&
218 (ap->logbufs < XLOG_MIN_ICLOGS ||
219 ap->logbufs > XLOG_MAX_ICLOGS)) {
221 "XFS: invalid logbufs value: %d [not %d-%d]",
222 ap->logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
223 return XFS_ERROR(EINVAL);
225 mp->m_logbufs = ap->logbufs;
226 if (ap->logbufsize != -1 &&
227 ap->logbufsize != 0 &&
228 (ap->logbufsize < XLOG_MIN_RECORD_BSIZE ||
229 ap->logbufsize > XLOG_MAX_RECORD_BSIZE ||
230 !is_power_of_2(ap->logbufsize))) {
232 "XFS: invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
234 return XFS_ERROR(EINVAL);
236 mp->m_logbsize = ap->logbufsize;
237 mp->m_fsname_len = strlen(ap->fsname) + 1;
238 mp->m_fsname = kmem_alloc(mp->m_fsname_len, KM_SLEEP);
239 strcpy(mp->m_fsname, ap->fsname);
241 mp->m_rtname = kmem_alloc(strlen(ap->rtname) + 1, KM_SLEEP);
242 strcpy(mp->m_rtname, ap->rtname);
244 if (ap->logname[0]) {
245 mp->m_logname = kmem_alloc(strlen(ap->logname) + 1, KM_SLEEP);
246 strcpy(mp->m_logname, ap->logname);
249 if (ap->flags & XFSMNT_WSYNC)
250 mp->m_flags |= XFS_MOUNT_WSYNC;
252 if (ap->flags & XFSMNT_INO64) {
253 mp->m_flags |= XFS_MOUNT_INO64;
254 mp->m_inoadd = XFS_INO64_OFFSET;
257 if (ap->flags & XFSMNT_RETERR)
258 mp->m_flags |= XFS_MOUNT_RETERR;
259 if (ap->flags & XFSMNT_NOALIGN)
260 mp->m_flags |= XFS_MOUNT_NOALIGN;
261 if (ap->flags & XFSMNT_SWALLOC)
262 mp->m_flags |= XFS_MOUNT_SWALLOC;
263 if (ap->flags & XFSMNT_OSYNCISOSYNC)
264 mp->m_flags |= XFS_MOUNT_OSYNCISOSYNC;
265 if (ap->flags & XFSMNT_32BITINODES)
266 mp->m_flags |= XFS_MOUNT_32BITINODES;
268 if (ap->flags & XFSMNT_IOSIZE) {
269 if (ap->iosizelog > XFS_MAX_IO_LOG ||
270 ap->iosizelog < XFS_MIN_IO_LOG) {
272 "XFS: invalid log iosize: %d [not %d-%d]",
273 ap->iosizelog, XFS_MIN_IO_LOG,
275 return XFS_ERROR(EINVAL);
278 mp->m_flags |= XFS_MOUNT_DFLT_IOSIZE;
279 mp->m_readio_log = mp->m_writeio_log = ap->iosizelog;
282 if (ap->flags & XFSMNT_IKEEP)
283 mp->m_flags |= XFS_MOUNT_IKEEP;
284 if (ap->flags & XFSMNT_DIRSYNC)
285 mp->m_flags |= XFS_MOUNT_DIRSYNC;
286 if (ap->flags & XFSMNT_ATTR2)
287 mp->m_flags |= XFS_MOUNT_ATTR2;
289 if (ap->flags2 & XFSMNT2_COMPAT_IOSIZE)
290 mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE;
293 * no recovery flag requires a read-only mount
295 if (ap->flags & XFSMNT_NORECOVERY) {
296 if (!(mp->m_flags & XFS_MOUNT_RDONLY)) {
298 "XFS: tried to mount a FS read-write without recovery!");
299 return XFS_ERROR(EINVAL);
301 mp->m_flags |= XFS_MOUNT_NORECOVERY;
304 if (ap->flags & XFSMNT_NOUUID)
305 mp->m_flags |= XFS_MOUNT_NOUUID;
306 if (ap->flags & XFSMNT_BARRIER)
307 mp->m_flags |= XFS_MOUNT_BARRIER;
309 mp->m_flags &= ~XFS_MOUNT_BARRIER;
311 if (ap->flags2 & XFSMNT2_FILESTREAMS)
312 mp->m_flags |= XFS_MOUNT_FILESTREAMS;
314 if (ap->flags & XFSMNT_DMAPI)
315 mp->m_flags |= XFS_MOUNT_DMAPI;
320 * This function fills in xfs_mount_t fields based on mount args.
321 * Note: the superblock _has_ now been read in.
325 struct xfs_mount_args *ap,
326 struct xfs_mount *mp)
328 int ronly = (mp->m_flags & XFS_MOUNT_RDONLY);
330 /* Fail a mount where the logbuf is smaller then the log stripe */
331 if (xfs_sb_version_haslogv2(&mp->m_sb)) {
332 if ((ap->logbufsize <= 0) &&
333 (mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE)) {
334 mp->m_logbsize = mp->m_sb.sb_logsunit;
335 } else if (ap->logbufsize > 0 &&
336 ap->logbufsize < mp->m_sb.sb_logsunit) {
338 "XFS: logbuf size must be greater than or equal to log stripe size");
339 return XFS_ERROR(EINVAL);
342 /* Fail a mount if the logbuf is larger than 32K */
343 if (ap->logbufsize > XLOG_BIG_RECORD_BSIZE) {
345 "XFS: logbuf size for version 1 logs must be 16K or 32K");
346 return XFS_ERROR(EINVAL);
350 if (xfs_sb_version_hasattr2(&mp->m_sb))
351 mp->m_flags |= XFS_MOUNT_ATTR2;
354 * prohibit r/w mounts of read-only filesystems
356 if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !ronly) {
358 "XFS: cannot mount a read-only filesystem as read-write");
359 return XFS_ERROR(EROFS);
363 * check for shared mount.
365 if (ap->flags & XFSMNT_SHARED) {
366 if (!xfs_sb_version_hasshared(&mp->m_sb))
367 return XFS_ERROR(EINVAL);
370 * For IRIX 6.5, shared mounts must have the shared
371 * version bit set, have the persistent readonly
372 * field set, must be version 0 and can only be mounted
375 if (!ronly || !(mp->m_sb.sb_flags & XFS_SBF_READONLY) ||
376 (mp->m_sb.sb_shared_vn != 0))
377 return XFS_ERROR(EINVAL);
379 mp->m_flags |= XFS_MOUNT_SHARED;
382 * Shared XFS V0 can't deal with DMI. Return EINVAL.
384 if (mp->m_sb.sb_shared_vn == 0 && (ap->flags & XFSMNT_DMAPI))
385 return XFS_ERROR(EINVAL);
388 if (ap->flags & XFSMNT_UQUOTA) {
389 mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE);
390 if (ap->flags & XFSMNT_UQUOTAENF)
391 mp->m_qflags |= XFS_UQUOTA_ENFD;
394 if (ap->flags & XFSMNT_GQUOTA) {
395 mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE);
396 if (ap->flags & XFSMNT_GQUOTAENF)
397 mp->m_qflags |= XFS_OQUOTA_ENFD;
398 } else if (ap->flags & XFSMNT_PQUOTA) {
399 mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE);
400 if (ap->flags & XFSMNT_PQUOTAENF)
401 mp->m_qflags |= XFS_OQUOTA_ENFD;
410 * The file system configurations are:
411 * (1) device (partition) with data and internal log
412 * (2) logical volume with data and log subvolumes.
413 * (3) logical volume with data, log, and realtime subvolumes.
415 * We only have to handle opening the log and realtime volumes here if
416 * they are present. The data subvolume has already been opened by
417 * get_sb_bdev() and is stored in vfsp->vfs_super->s_bdev.
421 struct xfs_mount *mp,
422 struct xfs_mount_args *args,
425 struct block_device *ddev, *logdev, *rtdev;
426 int flags = 0, error;
428 ddev = mp->m_super->s_bdev;
429 logdev = rtdev = NULL;
431 error = xfs_dmops_get(mp, args);
434 error = xfs_qmops_get(mp, args);
438 if (args->flags & XFSMNT_QUIET)
439 flags |= XFS_MFSI_QUIET;
442 * Open real time and log devices - order is important.
444 if (args->logname[0]) {
445 error = xfs_blkdev_get(mp, args->logname, &logdev);
449 if (args->rtname[0]) {
450 error = xfs_blkdev_get(mp, args->rtname, &rtdev);
452 xfs_blkdev_put(logdev);
456 if (rtdev == ddev || rtdev == logdev) {
458 "XFS: Cannot mount filesystem with identical rtdev and ddev/logdev.");
459 xfs_blkdev_put(logdev);
460 xfs_blkdev_put(rtdev);
466 * Setup xfs_mount buffer target pointers
469 mp->m_ddev_targp = xfs_alloc_buftarg(ddev, 0);
470 if (!mp->m_ddev_targp) {
471 xfs_blkdev_put(logdev);
472 xfs_blkdev_put(rtdev);
476 mp->m_rtdev_targp = xfs_alloc_buftarg(rtdev, 1);
477 if (!mp->m_rtdev_targp) {
478 xfs_blkdev_put(logdev);
479 xfs_blkdev_put(rtdev);
483 mp->m_logdev_targp = (logdev && logdev != ddev) ?
484 xfs_alloc_buftarg(logdev, 1) : mp->m_ddev_targp;
485 if (!mp->m_logdev_targp) {
486 xfs_blkdev_put(logdev);
487 xfs_blkdev_put(rtdev);
492 * Setup flags based on mount(2) options and then the superblock
494 error = xfs_start_flags(args, mp);
497 error = xfs_readsb(mp, flags);
500 error = xfs_finish_flags(args, mp);
505 * Setup xfs_mount buffer target pointers based on superblock
507 error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_blocksize,
508 mp->m_sb.sb_sectsize);
509 if (!error && logdev && logdev != ddev) {
510 unsigned int log_sector_size = BBSIZE;
512 if (xfs_sb_version_hassector(&mp->m_sb))
513 log_sector_size = mp->m_sb.sb_logsectsize;
514 error = xfs_setsize_buftarg(mp->m_logdev_targp,
515 mp->m_sb.sb_blocksize,
519 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
520 mp->m_sb.sb_blocksize,
521 mp->m_sb.sb_sectsize);
525 if (mp->m_flags & XFS_MOUNT_BARRIER)
526 xfs_mountfs_check_barriers(mp);
528 if ((error = xfs_filestream_mount(mp)))
531 error = xfs_mountfs(mp, flags);
535 XFS_SEND_MOUNT(mp, DM_RIGHT_NULL, args->mtpt, args->fsname);
543 xfs_binval(mp->m_ddev_targp);
544 if (logdev && logdev != ddev)
545 xfs_binval(mp->m_logdev_targp);
547 xfs_binval(mp->m_rtdev_targp);
549 xfs_unmountfs_close(mp, credp);
563 int unmount_event_wanted = 0;
564 int unmount_event_flags = 0;
565 int xfs_unmountfs_needed = 0;
572 if (mp->m_flags & XFS_MOUNT_DMAPI) {
573 error = XFS_SEND_PREUNMOUNT(mp,
574 rvp, DM_RIGHT_NULL, rvp, DM_RIGHT_NULL,
576 (mp->m_dmevmask & (1<<DM_EVENT_PREUNMOUNT))?
577 0:DM_FLAGS_UNWANTED);
579 return XFS_ERROR(error);
580 unmount_event_wanted = 1;
581 unmount_event_flags = (mp->m_dmevmask & (1<<DM_EVENT_UNMOUNT))?
582 0 : DM_FLAGS_UNWANTED;
587 * Blow away any referenced inode in the filestreams cache.
588 * This can and will cause log traffic as inodes go inactive
591 xfs_filestream_unmount(mp);
593 XFS_bflush(mp->m_ddev_targp);
594 error = xfs_unmount_flush(mp, 0);
598 ASSERT(vn_count(rvp) == 1);
601 * Drop the reference count
606 * If we're forcing a shutdown, typically because of a media error,
607 * we want to make sure we invalidate dirty pages that belong to
608 * referenced vnodes as well.
610 if (XFS_FORCED_SHUTDOWN(mp)) {
611 error = xfs_sync(mp, SYNC_WAIT | SYNC_CLOSE);
612 ASSERT(error != EFSCORRUPTED);
614 xfs_unmountfs_needed = 1;
617 /* Send DMAPI event, if required.
618 * Then do xfs_unmountfs() if needed.
619 * Then return error (or zero).
621 if (unmount_event_wanted) {
622 /* Note: mp structure must still exist for
623 * XFS_SEND_UNMOUNT() call.
625 XFS_SEND_UNMOUNT(mp, error == 0 ? rvp : NULL,
626 DM_RIGHT_NULL, 0, error, unmount_event_flags);
628 if (xfs_unmountfs_needed) {
630 * Call common unmount function to flush to disk
631 * and free the super block buffer & mount structures.
633 xfs_unmountfs(mp, credp);
636 kmem_free(mp, sizeof(xfs_mount_t));
639 return XFS_ERROR(error);
646 int count = 0, pincount;
648 xfs_flush_buftarg(mp->m_ddev_targp, 0);
649 xfs_finish_reclaim_all(mp, 0);
651 /* This loop must run at least twice.
652 * The first instance of the loop will flush
653 * most meta data but that will generate more
654 * meta data (typically directory updates).
655 * Which then must be flushed and logged before
656 * we can write the unmount record.
659 xfs_syncsub(mp, SYNC_INODE_QUIESCE, NULL);
660 pincount = xfs_flush_buftarg(mp->m_ddev_targp, 1);
671 * Second stage of a quiesce. The data is already synced, now we have to take
672 * care of the metadata. New transactions are already blocked, so we need to
673 * wait for any remaining transactions to drain out before proceding.
679 /* wait for all modifications to complete */
680 while (atomic_read(&mp->m_active_trans) > 0)
683 /* flush inodes and push all remaining buffers out to disk */
686 ASSERT_ALWAYS(atomic_read(&mp->m_active_trans) == 0);
688 /* Push the superblock and write an unmount record */
689 xfs_log_sbcount(mp, 1);
690 xfs_log_unmount_write(mp);
691 xfs_unmountfs_writesb(mp);
696 struct xfs_mount *mp,
698 struct xfs_mount_args *args)
700 if (!(*flags & MS_RDONLY)) { /* rw/ro -> rw */
701 if (mp->m_flags & XFS_MOUNT_RDONLY)
702 mp->m_flags &= ~XFS_MOUNT_RDONLY;
703 if (args->flags & XFSMNT_BARRIER) {
704 mp->m_flags |= XFS_MOUNT_BARRIER;
705 xfs_mountfs_check_barriers(mp);
707 mp->m_flags &= ~XFS_MOUNT_BARRIER;
709 } else if (!(mp->m_flags & XFS_MOUNT_RDONLY)) { /* rw -> ro */
710 xfs_filestream_flush(mp);
711 xfs_sync(mp, SYNC_DATA_QUIESCE);
712 xfs_attr_quiesce(mp);
713 mp->m_flags |= XFS_MOUNT_RDONLY;
719 * xfs_unmount_flush implements a set of flush operation on special
720 * inodes, which are needed as a separate set of operations so that
721 * they can be called as part of relocation process.
725 xfs_mount_t *mp, /* Mount structure we are getting
727 int relocation) /* Called from vfs relocation. */
729 xfs_inode_t *rip = mp->m_rootip;
731 xfs_inode_t *rsumip = NULL;
732 bhv_vnode_t *rvp = XFS_ITOV(rip);
735 xfs_ilock(rip, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
739 * Flush out the real time inodes.
741 if ((rbmip = mp->m_rbmip) != NULL) {
742 xfs_ilock(rbmip, XFS_ILOCK_EXCL);
744 error = xfs_iflush(rbmip, XFS_IFLUSH_SYNC);
745 xfs_iunlock(rbmip, XFS_ILOCK_EXCL);
747 if (error == EFSCORRUPTED)
750 ASSERT(vn_count(XFS_ITOV(rbmip)) == 1);
752 rsumip = mp->m_rsumip;
753 xfs_ilock(rsumip, XFS_ILOCK_EXCL);
755 error = xfs_iflush(rsumip, XFS_IFLUSH_SYNC);
756 xfs_iunlock(rsumip, XFS_ILOCK_EXCL);
758 if (error == EFSCORRUPTED)
761 ASSERT(vn_count(XFS_ITOV(rsumip)) == 1);
765 * Synchronously flush root inode to disk
767 error = xfs_iflush(rip, XFS_IFLUSH_SYNC);
768 if (error == EFSCORRUPTED)
771 if (vn_count(rvp) != 1 && !relocation) {
772 xfs_iunlock(rip, XFS_ILOCK_EXCL);
773 return XFS_ERROR(EBUSY);
777 * Release dquot that rootinode, rbmino and rsumino might be holding,
778 * flush and purge the quota inodes.
780 error = XFS_QM_UNMOUNT(mp);
781 if (error == EFSCORRUPTED)
785 VN_RELE(XFS_ITOV(rbmip));
786 VN_RELE(XFS_ITOV(rsumip));
789 xfs_iunlock(rip, XFS_ILOCK_EXCL);
796 xfs_iunlock(rip, XFS_ILOCK_EXCL);
798 return XFS_ERROR(EFSCORRUPTED);
802 * xfs_sync flushes any pending I/O to file system vfsp.
804 * This routine is called by vfs_sync() to make sure that things make it
805 * out to disk eventually, on sync() system calls to flush out everything,
806 * and when the file system is unmounted. For the vfs_sync() case, all
807 * we really need to do is sync out the log to make all of our meta-data
808 * updates permanent (except for timestamps). For calls from pflushd(),
809 * dirty pages are kept moving by calling pdflush() on the inodes
810 * containing them. We also flush the inodes that we can lock without
811 * sleeping and the superblock if we can lock it without sleeping from
812 * vfs_sync() so that items at the tail of the log are always moving out.
815 * SYNC_BDFLUSH - We're being called from vfs_sync() so we don't want
816 * to sleep if we can help it. All we really need
817 * to do is ensure that the log is synced at least
818 * periodically. We also push the inodes and
819 * superblock if we can lock them without sleeping
820 * and they are not pinned.
821 * SYNC_ATTR - We need to flush the inodes. If SYNC_BDFLUSH is not
822 * set, then we really want to lock each inode and flush
824 * SYNC_WAIT - All the flushes that take place in this call should
826 * SYNC_DELWRI - This tells us to push dirty pages associated with
827 * inodes. SYNC_WAIT and SYNC_BDFLUSH are used to
828 * determine if they should be flushed sync, async, or
830 * SYNC_CLOSE - This flag is passed when the system is being
831 * unmounted. We should sync and invalidate everything.
832 * SYNC_FSDATA - This indicates that the caller would like to make
833 * sure the superblock is safe on disk. We can ensure
834 * this by simply making sure the log gets flushed
835 * if SYNC_BDFLUSH is set, and by actually writing it
837 * SYNC_IOWAIT - The caller wants us to wait for all data I/O to complete
838 * before we return (including direct I/O). Forms the drain
839 * side of the write barrier needed to safely quiesce the
851 * Get the Quota Manager to flush the dquots.
853 * If XFS quota support is not enabled or this filesystem
854 * instance does not use quotas XFS_QM_DQSYNC will always
857 error = XFS_QM_DQSYNC(mp, flags);
860 * If we got an IO error, we will be shutting down.
861 * So, there's nothing more for us to do here.
863 ASSERT(error != EIO || XFS_FORCED_SHUTDOWN(mp));
864 if (XFS_FORCED_SHUTDOWN(mp))
865 return XFS_ERROR(error);
868 if (flags & SYNC_IOWAIT)
869 xfs_filestream_flush(mp);
871 return xfs_syncsub(mp, flags, NULL);
875 * xfs sync routine for internal use
877 * This routine supports all of the flags defined for the generic vfs_sync
878 * interface as explained above under xfs_sync.
887 xfs_inode_t *ip = NULL;
888 bhv_vnode_t *vp = NULL;
893 uint base_lock_flags;
894 boolean_t mount_locked;
895 boolean_t vnode_refed;
897 xfs_iptr_t *ipointer;
899 boolean_t ipointer_in = B_FALSE;
901 #define IPOINTER_SET ipointer_in = B_TRUE
902 #define IPOINTER_CLR ipointer_in = B_FALSE
909 /* Insert a marker record into the inode list after inode ip. The list
910 * must be locked when this is called. After the call the list will no
913 #define IPOINTER_INSERT(ip, mp) { \
914 ASSERT(ipointer_in == B_FALSE); \
915 ipointer->ip_mnext = ip->i_mnext; \
916 ipointer->ip_mprev = ip; \
917 ip->i_mnext = (xfs_inode_t *)ipointer; \
918 ipointer->ip_mnext->i_mprev = (xfs_inode_t *)ipointer; \
920 XFS_MOUNT_IUNLOCK(mp); \
921 mount_locked = B_FALSE; \
925 /* Remove the marker from the inode list. If the marker was the only item
926 * in the list then there are no remaining inodes and we should zero out
927 * the whole list. If we are the current head of the list then move the head
930 #define IPOINTER_REMOVE(ip, mp) { \
931 ASSERT(ipointer_in == B_TRUE); \
932 if (ipointer->ip_mnext != (xfs_inode_t *)ipointer) { \
933 ip = ipointer->ip_mnext; \
934 ip->i_mprev = ipointer->ip_mprev; \
935 ipointer->ip_mprev->i_mnext = ip; \
936 if (mp->m_inodes == (xfs_inode_t *)ipointer) { \
940 ASSERT(mp->m_inodes == (xfs_inode_t *)ipointer); \
941 mp->m_inodes = NULL; \
947 #define XFS_PREEMPT_MASK 0x7f
949 ASSERT(!(flags & SYNC_BDFLUSH));
953 if (mp->m_flags & XFS_MOUNT_RDONLY)
959 /* Allocate a reference marker */
960 ipointer = (xfs_iptr_t *)kmem_zalloc(sizeof(xfs_iptr_t), KM_SLEEP);
962 fflag = XFS_B_ASYNC; /* default is don't wait */
963 if (flags & SYNC_DELWRI)
964 fflag = XFS_B_DELWRI;
965 if (flags & SYNC_WAIT)
966 fflag = 0; /* synchronous overrides all */
968 base_lock_flags = XFS_ILOCK_SHARED;
969 if (flags & (SYNC_DELWRI | SYNC_CLOSE)) {
971 * We need the I/O lock if we're going to call any of
972 * the flush/inval routines.
974 base_lock_flags |= XFS_IOLOCK_SHARED;
981 mount_locked = B_TRUE;
982 vnode_refed = B_FALSE;
987 ASSERT(ipointer_in == B_FALSE);
988 ASSERT(vnode_refed == B_FALSE);
990 lock_flags = base_lock_flags;
993 * There were no inodes in the list, just break out
1001 * We found another sync thread marker - skip it
1003 if (ip->i_mount == NULL) {
1008 vp = XFS_ITOV_NULL(ip);
1011 * If the vnode is gone then this is being torn down,
1012 * call reclaim if it is flushed, else let regular flush
1013 * code deal with it later in the loop.
1017 /* Skip ones already in reclaim */
1018 if (ip->i_flags & XFS_IRECLAIM) {
1022 if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0) {
1024 } else if ((xfs_ipincount(ip) == 0) &&
1025 xfs_iflock_nowait(ip)) {
1026 IPOINTER_INSERT(ip, mp);
1028 xfs_finish_reclaim(ip, 1,
1029 XFS_IFLUSH_DELWRI_ELSE_ASYNC);
1031 XFS_MOUNT_ILOCK(mp);
1032 mount_locked = B_TRUE;
1033 IPOINTER_REMOVE(ip, mp);
1035 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1046 if (XFS_FORCED_SHUTDOWN(mp) && !(flags & SYNC_CLOSE)) {
1047 XFS_MOUNT_IUNLOCK(mp);
1048 kmem_free(ipointer, sizeof(xfs_iptr_t));
1053 * Try to lock without sleeping. We're out of order with
1054 * the inode list lock here, so if we fail we need to drop
1055 * the mount lock and try again. If we're called from
1056 * bdflush() here, then don't bother.
1058 * The inode lock here actually coordinates with the
1059 * almost spurious inode lock in xfs_ireclaim() to prevent
1060 * the vnode we handle here without a reference from
1061 * being freed while we reference it. If we lock the inode
1062 * while it's on the mount list here, then the spurious inode
1063 * lock in xfs_ireclaim() after the inode is pulled from
1064 * the mount list will sleep until we release it here.
1065 * This keeps the vnode from being freed while we reference
1068 if (xfs_ilock_nowait(ip, lock_flags) == 0) {
1080 IPOINTER_INSERT(ip, mp);
1081 xfs_ilock(ip, lock_flags);
1083 ASSERT(vp == XFS_ITOV(ip));
1084 ASSERT(ip->i_mount == mp);
1086 vnode_refed = B_TRUE;
1089 /* From here on in the loop we may have a marker record
1090 * in the inode list.
1094 * If we have to flush data or wait for I/O completion
1095 * we need to drop the ilock that we currently hold.
1096 * If we need to drop the lock, insert a marker if we
1097 * have not already done so.
1099 if ((flags & (SYNC_CLOSE|SYNC_IOWAIT)) ||
1100 ((flags & SYNC_DELWRI) && VN_DIRTY(vp))) {
1102 IPOINTER_INSERT(ip, mp);
1104 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1106 if (flags & SYNC_CLOSE) {
1107 /* Shutdown case. Flush and invalidate. */
1108 if (XFS_FORCED_SHUTDOWN(mp))
1109 xfs_tosspages(ip, 0, -1,
1112 error = xfs_flushinval_pages(ip,
1114 } else if ((flags & SYNC_DELWRI) && VN_DIRTY(vp)) {
1115 error = xfs_flush_pages(ip, 0,
1116 -1, fflag, FI_NONE);
1120 * When freezing, we need to wait ensure all I/O (including direct
1121 * I/O) is complete to ensure no further data modification can take
1122 * place after this point
1124 if (flags & SYNC_IOWAIT)
1127 xfs_ilock(ip, XFS_ILOCK_SHARED);
1130 if ((flags & SYNC_ATTR) &&
1131 (ip->i_update_core ||
1132 (ip->i_itemp && ip->i_itemp->ili_format.ilf_fields))) {
1134 IPOINTER_INSERT(ip, mp);
1136 if (flags & SYNC_WAIT) {
1138 error = xfs_iflush(ip, XFS_IFLUSH_SYNC);
1141 * If we can't acquire the flush lock, then the inode
1142 * is already being flushed so don't bother waiting.
1144 * If we can lock it then do a delwri flush so we can
1145 * combine multiple inode flushes in each disk write.
1147 } else if (xfs_iflock_nowait(ip)) {
1148 error = xfs_iflush(ip, XFS_IFLUSH_DELWRI);
1149 } else if (bypassed) {
1154 if (lock_flags != 0) {
1155 xfs_iunlock(ip, lock_flags);
1160 * If we had to take a reference on the vnode
1161 * above, then wait until after we've unlocked
1162 * the inode to release the reference. This is
1163 * because we can be already holding the inode
1164 * lock when VN_RELE() calls xfs_inactive().
1166 * Make sure to drop the mount lock before calling
1167 * VN_RELE() so that we don't trip over ourselves if
1168 * we have to go for the mount lock again in the
1172 IPOINTER_INSERT(ip, mp);
1177 vnode_refed = B_FALSE;
1185 * bail out if the filesystem is corrupted.
1187 if (error == EFSCORRUPTED) {
1188 if (!mount_locked) {
1189 XFS_MOUNT_ILOCK(mp);
1190 IPOINTER_REMOVE(ip, mp);
1192 XFS_MOUNT_IUNLOCK(mp);
1193 ASSERT(ipointer_in == B_FALSE);
1194 kmem_free(ipointer, sizeof(xfs_iptr_t));
1195 return XFS_ERROR(error);
1198 /* Let other threads have a chance at the mount lock
1199 * if we have looped many times without dropping the
1202 if ((++preempt & XFS_PREEMPT_MASK) == 0) {
1204 IPOINTER_INSERT(ip, mp);
1208 if (mount_locked == B_FALSE) {
1209 XFS_MOUNT_ILOCK(mp);
1210 mount_locked = B_TRUE;
1211 IPOINTER_REMOVE(ip, mp);
1215 ASSERT(ipointer_in == B_FALSE);
1218 } while (ip != mp->m_inodes);
1220 XFS_MOUNT_IUNLOCK(mp);
1222 ASSERT(ipointer_in == B_FALSE);
1224 kmem_free(ipointer, sizeof(xfs_iptr_t));
1225 return XFS_ERROR(last_error);
1229 * xfs sync routine for internal use
1231 * This routine supports all of the flags defined for the generic vfs_sync
1232 * interface as explained above under xfs_sync.
1243 uint log_flags = XFS_LOG_FORCE;
1245 xfs_buf_log_item_t *bip;
1248 * Sync out the log. This ensures that the log is periodically
1249 * flushed even if there is not enough activity to fill it up.
1251 if (flags & SYNC_WAIT)
1252 log_flags |= XFS_LOG_SYNC;
1254 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1256 if (flags & (SYNC_ATTR|SYNC_DELWRI)) {
1257 if (flags & SYNC_BDFLUSH)
1258 xfs_finish_reclaim_all(mp, 1);
1260 error = xfs_sync_inodes(mp, flags, bypassed);
1264 * Flushing out dirty data above probably generated more
1265 * log activity, so if this isn't vfs_sync() then flush
1268 if (flags & SYNC_DELWRI) {
1269 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1272 if (flags & SYNC_FSDATA) {
1274 * If this is vfs_sync() then only sync the superblock
1275 * if we can lock it without sleeping and it is not pinned.
1277 if (flags & SYNC_BDFLUSH) {
1278 bp = xfs_getsb(mp, XFS_BUF_TRYLOCK);
1280 bip = XFS_BUF_FSPRIVATE(bp,xfs_buf_log_item_t*);
1281 if ((bip != NULL) &&
1282 xfs_buf_item_dirty(bip)) {
1283 if (!(XFS_BUF_ISPINNED(bp))) {
1285 error = xfs_bwrite(mp, bp);
1294 bp = xfs_getsb(mp, 0);
1296 * If the buffer is pinned then push on the log so
1297 * we won't get stuck waiting in the write for
1298 * someone, maybe ourselves, to flush the log.
1299 * Even though we just pushed the log above, we
1300 * did not have the superblock buffer locked at
1301 * that point so it can become pinned in between
1304 if (XFS_BUF_ISPINNED(bp))
1305 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
1306 if (flags & SYNC_WAIT)
1307 XFS_BUF_UNASYNC(bp);
1310 error = xfs_bwrite(mp, bp);
1318 * If asked, update the disk superblock with incore counter values if we
1319 * are using non-persistent counters so that they don't get too far out
1320 * of sync if we crash or get a forced shutdown. We don't want to force
1321 * this to disk, just get a transaction into the iclogs....
1323 if (flags & SYNC_SUPER)
1324 xfs_log_sbcount(mp, 0);
1327 * Now check to see if the log needs a "dummy" transaction.
1330 if (!(flags & SYNC_REMOUNT) && xfs_log_need_covered(mp)) {
1335 * Put a dummy transaction in the log to tell
1336 * recovery that all others are OK.
1338 tp = xfs_trans_alloc(mp, XFS_TRANS_DUMMY1);
1339 if ((error = xfs_trans_reserve(tp, 0,
1340 XFS_ICHANGE_LOG_RES(mp),
1342 xfs_trans_cancel(tp, 0);
1347 xfs_ilock(ip, XFS_ILOCK_EXCL);
1349 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1350 xfs_trans_ihold(tp, ip);
1351 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1352 error = xfs_trans_commit(tp, 0);
1353 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1354 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1358 * When shutting down, we need to insure that the AIL is pushed
1359 * to disk or the filesystem can appear corrupt from the PROM.
1361 if ((flags & (SYNC_CLOSE|SYNC_WAIT)) == (SYNC_CLOSE|SYNC_WAIT)) {
1362 XFS_bflush(mp->m_ddev_targp);
1363 if (mp->m_rtdev_targp) {
1364 XFS_bflush(mp->m_rtdev_targp);
1368 return XFS_ERROR(last_error);