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);
189 kmem_zone_destroy(xfs_log_ticket_zone);
195 * This function fills in xfs_mount_t fields based on mount args.
196 * Note: the superblock has _not_ yet been read in.
200 struct xfs_mount_args *ap,
201 struct xfs_mount *mp)
203 /* Values are in BBs */
204 if ((ap->flags & XFSMNT_NOALIGN) != XFSMNT_NOALIGN) {
206 * At this point the superblock has not been read
207 * in, therefore we do not know the block size.
208 * Before the mount call ends we will convert
211 mp->m_dalign = ap->sunit;
212 mp->m_swidth = ap->swidth;
215 if (ap->logbufs != -1 &&
217 (ap->logbufs < XLOG_MIN_ICLOGS ||
218 ap->logbufs > XLOG_MAX_ICLOGS)) {
220 "XFS: invalid logbufs value: %d [not %d-%d]",
221 ap->logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
222 return XFS_ERROR(EINVAL);
224 mp->m_logbufs = ap->logbufs;
225 if (ap->logbufsize != -1 &&
226 ap->logbufsize != 0 &&
227 (ap->logbufsize < XLOG_MIN_RECORD_BSIZE ||
228 ap->logbufsize > XLOG_MAX_RECORD_BSIZE ||
229 !is_power_of_2(ap->logbufsize))) {
231 "XFS: invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
233 return XFS_ERROR(EINVAL);
235 mp->m_logbsize = ap->logbufsize;
236 mp->m_fsname_len = strlen(ap->fsname) + 1;
237 mp->m_fsname = kmem_alloc(mp->m_fsname_len, KM_SLEEP);
238 strcpy(mp->m_fsname, ap->fsname);
240 mp->m_rtname = kmem_alloc(strlen(ap->rtname) + 1, KM_SLEEP);
241 strcpy(mp->m_rtname, ap->rtname);
243 if (ap->logname[0]) {
244 mp->m_logname = kmem_alloc(strlen(ap->logname) + 1, KM_SLEEP);
245 strcpy(mp->m_logname, ap->logname);
248 if (ap->flags & XFSMNT_WSYNC)
249 mp->m_flags |= XFS_MOUNT_WSYNC;
251 if (ap->flags & XFSMNT_INO64) {
252 mp->m_flags |= XFS_MOUNT_INO64;
253 mp->m_inoadd = XFS_INO64_OFFSET;
256 if (ap->flags & XFSMNT_RETERR)
257 mp->m_flags |= XFS_MOUNT_RETERR;
258 if (ap->flags & XFSMNT_NOALIGN)
259 mp->m_flags |= XFS_MOUNT_NOALIGN;
260 if (ap->flags & XFSMNT_SWALLOC)
261 mp->m_flags |= XFS_MOUNT_SWALLOC;
262 if (ap->flags & XFSMNT_OSYNCISOSYNC)
263 mp->m_flags |= XFS_MOUNT_OSYNCISOSYNC;
264 if (ap->flags & XFSMNT_32BITINODES)
265 mp->m_flags |= XFS_MOUNT_32BITINODES;
267 if (ap->flags & XFSMNT_IOSIZE) {
268 if (ap->iosizelog > XFS_MAX_IO_LOG ||
269 ap->iosizelog < XFS_MIN_IO_LOG) {
271 "XFS: invalid log iosize: %d [not %d-%d]",
272 ap->iosizelog, XFS_MIN_IO_LOG,
274 return XFS_ERROR(EINVAL);
277 mp->m_flags |= XFS_MOUNT_DFLT_IOSIZE;
278 mp->m_readio_log = mp->m_writeio_log = ap->iosizelog;
281 if (ap->flags & XFSMNT_IKEEP)
282 mp->m_flags |= XFS_MOUNT_IKEEP;
283 if (ap->flags & XFSMNT_DIRSYNC)
284 mp->m_flags |= XFS_MOUNT_DIRSYNC;
285 if (ap->flags & XFSMNT_ATTR2)
286 mp->m_flags |= XFS_MOUNT_ATTR2;
288 if (ap->flags2 & XFSMNT2_COMPAT_IOSIZE)
289 mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE;
292 * no recovery flag requires a read-only mount
294 if (ap->flags & XFSMNT_NORECOVERY) {
295 if (!(mp->m_flags & XFS_MOUNT_RDONLY)) {
297 "XFS: tried to mount a FS read-write without recovery!");
298 return XFS_ERROR(EINVAL);
300 mp->m_flags |= XFS_MOUNT_NORECOVERY;
303 if (ap->flags & XFSMNT_NOUUID)
304 mp->m_flags |= XFS_MOUNT_NOUUID;
305 if (ap->flags & XFSMNT_BARRIER)
306 mp->m_flags |= XFS_MOUNT_BARRIER;
308 mp->m_flags &= ~XFS_MOUNT_BARRIER;
310 if (ap->flags2 & XFSMNT2_FILESTREAMS)
311 mp->m_flags |= XFS_MOUNT_FILESTREAMS;
313 if (ap->flags & XFSMNT_DMAPI)
314 mp->m_flags |= XFS_MOUNT_DMAPI;
319 * This function fills in xfs_mount_t fields based on mount args.
320 * Note: the superblock _has_ now been read in.
324 struct xfs_mount_args *ap,
325 struct xfs_mount *mp)
327 int ronly = (mp->m_flags & XFS_MOUNT_RDONLY);
329 /* Fail a mount where the logbuf is smaller then the log stripe */
330 if (xfs_sb_version_haslogv2(&mp->m_sb)) {
331 if ((ap->logbufsize <= 0) &&
332 (mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE)) {
333 mp->m_logbsize = mp->m_sb.sb_logsunit;
334 } else if (ap->logbufsize > 0 &&
335 ap->logbufsize < mp->m_sb.sb_logsunit) {
337 "XFS: logbuf size must be greater than or equal to log stripe size");
338 return XFS_ERROR(EINVAL);
341 /* Fail a mount if the logbuf is larger than 32K */
342 if (ap->logbufsize > XLOG_BIG_RECORD_BSIZE) {
344 "XFS: logbuf size for version 1 logs must be 16K or 32K");
345 return XFS_ERROR(EINVAL);
349 if (xfs_sb_version_hasattr2(&mp->m_sb))
350 mp->m_flags |= XFS_MOUNT_ATTR2;
353 * prohibit r/w mounts of read-only filesystems
355 if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !ronly) {
357 "XFS: cannot mount a read-only filesystem as read-write");
358 return XFS_ERROR(EROFS);
362 * check for shared mount.
364 if (ap->flags & XFSMNT_SHARED) {
365 if (!xfs_sb_version_hasshared(&mp->m_sb))
366 return XFS_ERROR(EINVAL);
369 * For IRIX 6.5, shared mounts must have the shared
370 * version bit set, have the persistent readonly
371 * field set, must be version 0 and can only be mounted
374 if (!ronly || !(mp->m_sb.sb_flags & XFS_SBF_READONLY) ||
375 (mp->m_sb.sb_shared_vn != 0))
376 return XFS_ERROR(EINVAL);
378 mp->m_flags |= XFS_MOUNT_SHARED;
381 * Shared XFS V0 can't deal with DMI. Return EINVAL.
383 if (mp->m_sb.sb_shared_vn == 0 && (ap->flags & XFSMNT_DMAPI))
384 return XFS_ERROR(EINVAL);
387 if (ap->flags & XFSMNT_UQUOTA) {
388 mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE);
389 if (ap->flags & XFSMNT_UQUOTAENF)
390 mp->m_qflags |= XFS_UQUOTA_ENFD;
393 if (ap->flags & XFSMNT_GQUOTA) {
394 mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE);
395 if (ap->flags & XFSMNT_GQUOTAENF)
396 mp->m_qflags |= XFS_OQUOTA_ENFD;
397 } else if (ap->flags & XFSMNT_PQUOTA) {
398 mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE);
399 if (ap->flags & XFSMNT_PQUOTAENF)
400 mp->m_qflags |= XFS_OQUOTA_ENFD;
409 * The file system configurations are:
410 * (1) device (partition) with data and internal log
411 * (2) logical volume with data and log subvolumes.
412 * (3) logical volume with data, log, and realtime subvolumes.
414 * We only have to handle opening the log and realtime volumes here if
415 * they are present. The data subvolume has already been opened by
416 * get_sb_bdev() and is stored in vfsp->vfs_super->s_bdev.
420 struct xfs_mount *mp,
421 struct xfs_mount_args *args,
424 struct block_device *ddev, *logdev, *rtdev;
425 int flags = 0, error;
427 ddev = mp->m_super->s_bdev;
428 logdev = rtdev = NULL;
430 error = xfs_dmops_get(mp, args);
433 error = xfs_qmops_get(mp, args);
437 if (args->flags & XFSMNT_QUIET)
438 flags |= XFS_MFSI_QUIET;
441 * Open real time and log devices - order is important.
443 if (args->logname[0]) {
444 error = xfs_blkdev_get(mp, args->logname, &logdev);
448 if (args->rtname[0]) {
449 error = xfs_blkdev_get(mp, args->rtname, &rtdev);
451 xfs_blkdev_put(logdev);
455 if (rtdev == ddev || rtdev == logdev) {
457 "XFS: Cannot mount filesystem with identical rtdev and ddev/logdev.");
458 xfs_blkdev_put(logdev);
459 xfs_blkdev_put(rtdev);
465 * Setup xfs_mount buffer target pointers
468 mp->m_ddev_targp = xfs_alloc_buftarg(ddev, 0);
469 if (!mp->m_ddev_targp) {
470 xfs_blkdev_put(logdev);
471 xfs_blkdev_put(rtdev);
475 mp->m_rtdev_targp = xfs_alloc_buftarg(rtdev, 1);
476 if (!mp->m_rtdev_targp) {
477 xfs_blkdev_put(logdev);
478 xfs_blkdev_put(rtdev);
482 mp->m_logdev_targp = (logdev && logdev != ddev) ?
483 xfs_alloc_buftarg(logdev, 1) : mp->m_ddev_targp;
484 if (!mp->m_logdev_targp) {
485 xfs_blkdev_put(logdev);
486 xfs_blkdev_put(rtdev);
491 * Setup flags based on mount(2) options and then the superblock
493 error = xfs_start_flags(args, mp);
496 error = xfs_readsb(mp, flags);
499 error = xfs_finish_flags(args, mp);
504 * Setup xfs_mount buffer target pointers based on superblock
506 error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_blocksize,
507 mp->m_sb.sb_sectsize);
508 if (!error && logdev && logdev != ddev) {
509 unsigned int log_sector_size = BBSIZE;
511 if (xfs_sb_version_hassector(&mp->m_sb))
512 log_sector_size = mp->m_sb.sb_logsectsize;
513 error = xfs_setsize_buftarg(mp->m_logdev_targp,
514 mp->m_sb.sb_blocksize,
518 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
519 mp->m_sb.sb_blocksize,
520 mp->m_sb.sb_sectsize);
524 if (mp->m_flags & XFS_MOUNT_BARRIER)
525 xfs_mountfs_check_barriers(mp);
527 if ((error = xfs_filestream_mount(mp)))
530 error = xfs_mountfs(mp, flags);
534 XFS_SEND_MOUNT(mp, DM_RIGHT_NULL, args->mtpt, args->fsname);
542 xfs_binval(mp->m_ddev_targp);
543 if (logdev && logdev != ddev)
544 xfs_binval(mp->m_logdev_targp);
546 xfs_binval(mp->m_rtdev_targp);
548 xfs_unmountfs_close(mp, credp);
562 int unmount_event_wanted = 0;
563 int unmount_event_flags = 0;
564 int xfs_unmountfs_needed = 0;
571 if (mp->m_flags & XFS_MOUNT_DMAPI) {
572 error = XFS_SEND_PREUNMOUNT(mp,
573 rip, DM_RIGHT_NULL, rip, DM_RIGHT_NULL,
575 (mp->m_dmevmask & (1<<DM_EVENT_PREUNMOUNT))?
576 0:DM_FLAGS_UNWANTED);
578 return XFS_ERROR(error);
579 unmount_event_wanted = 1;
580 unmount_event_flags = (mp->m_dmevmask & (1<<DM_EVENT_UNMOUNT))?
581 0 : DM_FLAGS_UNWANTED;
586 * Blow away any referenced inode in the filestreams cache.
587 * This can and will cause log traffic as inodes go inactive
590 xfs_filestream_unmount(mp);
592 XFS_bflush(mp->m_ddev_targp);
593 error = xfs_unmount_flush(mp, 0);
597 ASSERT(vn_count(rvp) == 1);
600 * Drop the reference count
605 * If we're forcing a shutdown, typically because of a media error,
606 * we want to make sure we invalidate dirty pages that belong to
607 * referenced vnodes as well.
609 if (XFS_FORCED_SHUTDOWN(mp)) {
610 error = xfs_sync(mp, SYNC_WAIT | SYNC_CLOSE);
611 ASSERT(error != EFSCORRUPTED);
613 xfs_unmountfs_needed = 1;
616 /* Send DMAPI event, if required.
617 * Then do xfs_unmountfs() if needed.
618 * Then return error (or zero).
620 if (unmount_event_wanted) {
621 /* Note: mp structure must still exist for
622 * XFS_SEND_UNMOUNT() call.
624 XFS_SEND_UNMOUNT(mp, error == 0 ? rip : NULL,
625 DM_RIGHT_NULL, 0, error, unmount_event_flags);
627 if (xfs_unmountfs_needed) {
629 * Call common unmount function to flush to disk
630 * and free the super block buffer & mount structures.
632 xfs_unmountfs(mp, credp);
635 kmem_free(mp, sizeof(xfs_mount_t));
638 return XFS_ERROR(error);
645 int count = 0, pincount;
647 xfs_flush_buftarg(mp->m_ddev_targp, 0);
648 xfs_finish_reclaim_all(mp, 0);
650 /* This loop must run at least twice.
651 * The first instance of the loop will flush
652 * most meta data but that will generate more
653 * meta data (typically directory updates).
654 * Which then must be flushed and logged before
655 * we can write the unmount record.
658 xfs_syncsub(mp, SYNC_INODE_QUIESCE, NULL);
659 pincount = xfs_flush_buftarg(mp->m_ddev_targp, 1);
668 * Second stage of a quiesce. The data is already synced, now we have to take
669 * care of the metadata. New transactions are already blocked, so we need to
670 * wait for any remaining transactions to drain out before proceding.
678 /* wait for all modifications to complete */
679 while (atomic_read(&mp->m_active_trans) > 0)
682 /* flush inodes and push all remaining buffers out to disk */
685 ASSERT_ALWAYS(atomic_read(&mp->m_active_trans) == 0);
687 /* Push the superblock and write an unmount record */
688 error = xfs_log_sbcount(mp, 1);
690 xfs_fs_cmn_err(CE_WARN, mp,
691 "xfs_attr_quiesce: failed to log sb changes. "
692 "Frozen image may not be consistent.");
693 xfs_log_unmount_write(mp);
694 xfs_unmountfs_writesb(mp);
699 struct xfs_mount *mp,
701 struct xfs_mount_args *args)
703 if (!(*flags & MS_RDONLY)) { /* rw/ro -> rw */
704 if (mp->m_flags & XFS_MOUNT_RDONLY)
705 mp->m_flags &= ~XFS_MOUNT_RDONLY;
706 if (args->flags & XFSMNT_BARRIER) {
707 mp->m_flags |= XFS_MOUNT_BARRIER;
708 xfs_mountfs_check_barriers(mp);
710 mp->m_flags &= ~XFS_MOUNT_BARRIER;
712 } else if (!(mp->m_flags & XFS_MOUNT_RDONLY)) { /* rw -> ro */
713 xfs_filestream_flush(mp);
714 xfs_sync(mp, SYNC_DATA_QUIESCE);
715 xfs_attr_quiesce(mp);
716 mp->m_flags |= XFS_MOUNT_RDONLY;
722 * xfs_unmount_flush implements a set of flush operation on special
723 * inodes, which are needed as a separate set of operations so that
724 * they can be called as part of relocation process.
728 xfs_mount_t *mp, /* Mount structure we are getting
730 int relocation) /* Called from vfs relocation. */
732 xfs_inode_t *rip = mp->m_rootip;
734 xfs_inode_t *rsumip = NULL;
735 bhv_vnode_t *rvp = XFS_ITOV(rip);
738 xfs_ilock(rip, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
742 * Flush out the real time inodes.
744 if ((rbmip = mp->m_rbmip) != NULL) {
745 xfs_ilock(rbmip, XFS_ILOCK_EXCL);
747 error = xfs_iflush(rbmip, XFS_IFLUSH_SYNC);
748 xfs_iunlock(rbmip, XFS_ILOCK_EXCL);
750 if (error == EFSCORRUPTED)
753 ASSERT(vn_count(XFS_ITOV(rbmip)) == 1);
755 rsumip = mp->m_rsumip;
756 xfs_ilock(rsumip, XFS_ILOCK_EXCL);
758 error = xfs_iflush(rsumip, XFS_IFLUSH_SYNC);
759 xfs_iunlock(rsumip, XFS_ILOCK_EXCL);
761 if (error == EFSCORRUPTED)
764 ASSERT(vn_count(XFS_ITOV(rsumip)) == 1);
768 * Synchronously flush root inode to disk
770 error = xfs_iflush(rip, XFS_IFLUSH_SYNC);
771 if (error == EFSCORRUPTED)
774 if (vn_count(rvp) != 1 && !relocation) {
775 xfs_iunlock(rip, XFS_ILOCK_EXCL);
776 return XFS_ERROR(EBUSY);
780 * Release dquot that rootinode, rbmino and rsumino might be holding,
781 * flush and purge the quota inodes.
783 error = XFS_QM_UNMOUNT(mp);
784 if (error == EFSCORRUPTED)
792 xfs_iunlock(rip, XFS_ILOCK_EXCL);
799 xfs_iunlock(rip, XFS_ILOCK_EXCL);
801 return XFS_ERROR(EFSCORRUPTED);
805 * xfs_sync flushes any pending I/O to file system vfsp.
807 * This routine is called by vfs_sync() to make sure that things make it
808 * out to disk eventually, on sync() system calls to flush out everything,
809 * and when the file system is unmounted. For the vfs_sync() case, all
810 * we really need to do is sync out the log to make all of our meta-data
811 * updates permanent (except for timestamps). For calls from pflushd(),
812 * dirty pages are kept moving by calling pdflush() on the inodes
813 * containing them. We also flush the inodes that we can lock without
814 * sleeping and the superblock if we can lock it without sleeping from
815 * vfs_sync() so that items at the tail of the log are always moving out.
818 * SYNC_BDFLUSH - We're being called from vfs_sync() so we don't want
819 * to sleep if we can help it. All we really need
820 * to do is ensure that the log is synced at least
821 * periodically. We also push the inodes and
822 * superblock if we can lock them without sleeping
823 * and they are not pinned.
824 * SYNC_ATTR - We need to flush the inodes. If SYNC_BDFLUSH is not
825 * set, then we really want to lock each inode and flush
827 * SYNC_WAIT - All the flushes that take place in this call should
829 * SYNC_DELWRI - This tells us to push dirty pages associated with
830 * inodes. SYNC_WAIT and SYNC_BDFLUSH are used to
831 * determine if they should be flushed sync, async, or
833 * SYNC_CLOSE - This flag is passed when the system is being
834 * unmounted. We should sync and invalidate everything.
835 * SYNC_FSDATA - This indicates that the caller would like to make
836 * sure the superblock is safe on disk. We can ensure
837 * this by simply making sure the log gets flushed
838 * if SYNC_BDFLUSH is set, and by actually writing it
840 * SYNC_IOWAIT - The caller wants us to wait for all data I/O to complete
841 * before we return (including direct I/O). Forms the drain
842 * side of the write barrier needed to safely quiesce the
854 * Get the Quota Manager to flush the dquots.
856 * If XFS quota support is not enabled or this filesystem
857 * instance does not use quotas XFS_QM_DQSYNC will always
860 error = XFS_QM_DQSYNC(mp, flags);
863 * If we got an IO error, we will be shutting down.
864 * So, there's nothing more for us to do here.
866 ASSERT(error != EIO || XFS_FORCED_SHUTDOWN(mp));
867 if (XFS_FORCED_SHUTDOWN(mp))
868 return XFS_ERROR(error);
871 if (flags & SYNC_IOWAIT)
872 xfs_filestream_flush(mp);
874 return xfs_syncsub(mp, flags, NULL);
878 * xfs sync routine for internal use
880 * This routine supports all of the flags defined for the generic vfs_sync
881 * interface as explained above under xfs_sync.
890 xfs_inode_t *ip = NULL;
891 bhv_vnode_t *vp = NULL;
896 uint base_lock_flags;
897 boolean_t mount_locked;
898 boolean_t vnode_refed;
900 xfs_iptr_t *ipointer;
902 boolean_t ipointer_in = B_FALSE;
904 #define IPOINTER_SET ipointer_in = B_TRUE
905 #define IPOINTER_CLR ipointer_in = B_FALSE
912 /* Insert a marker record into the inode list after inode ip. The list
913 * must be locked when this is called. After the call the list will no
916 #define IPOINTER_INSERT(ip, mp) { \
917 ASSERT(ipointer_in == B_FALSE); \
918 ipointer->ip_mnext = ip->i_mnext; \
919 ipointer->ip_mprev = ip; \
920 ip->i_mnext = (xfs_inode_t *)ipointer; \
921 ipointer->ip_mnext->i_mprev = (xfs_inode_t *)ipointer; \
923 XFS_MOUNT_IUNLOCK(mp); \
924 mount_locked = B_FALSE; \
928 /* Remove the marker from the inode list. If the marker was the only item
929 * in the list then there are no remaining inodes and we should zero out
930 * the whole list. If we are the current head of the list then move the head
933 #define IPOINTER_REMOVE(ip, mp) { \
934 ASSERT(ipointer_in == B_TRUE); \
935 if (ipointer->ip_mnext != (xfs_inode_t *)ipointer) { \
936 ip = ipointer->ip_mnext; \
937 ip->i_mprev = ipointer->ip_mprev; \
938 ipointer->ip_mprev->i_mnext = ip; \
939 if (mp->m_inodes == (xfs_inode_t *)ipointer) { \
943 ASSERT(mp->m_inodes == (xfs_inode_t *)ipointer); \
944 mp->m_inodes = NULL; \
950 #define XFS_PREEMPT_MASK 0x7f
952 ASSERT(!(flags & SYNC_BDFLUSH));
956 if (mp->m_flags & XFS_MOUNT_RDONLY)
962 /* Allocate a reference marker */
963 ipointer = (xfs_iptr_t *)kmem_zalloc(sizeof(xfs_iptr_t), KM_SLEEP);
965 fflag = XFS_B_ASYNC; /* default is don't wait */
966 if (flags & SYNC_DELWRI)
967 fflag = XFS_B_DELWRI;
968 if (flags & SYNC_WAIT)
969 fflag = 0; /* synchronous overrides all */
971 base_lock_flags = XFS_ILOCK_SHARED;
972 if (flags & (SYNC_DELWRI | SYNC_CLOSE)) {
974 * We need the I/O lock if we're going to call any of
975 * the flush/inval routines.
977 base_lock_flags |= XFS_IOLOCK_SHARED;
984 mount_locked = B_TRUE;
985 vnode_refed = B_FALSE;
990 ASSERT(ipointer_in == B_FALSE);
991 ASSERT(vnode_refed == B_FALSE);
993 lock_flags = base_lock_flags;
996 * There were no inodes in the list, just break out
1004 * We found another sync thread marker - skip it
1006 if (ip->i_mount == NULL) {
1011 vp = XFS_ITOV_NULL(ip);
1014 * If the vnode is gone then this is being torn down,
1015 * call reclaim if it is flushed, else let regular flush
1016 * code deal with it later in the loop.
1020 /* Skip ones already in reclaim */
1021 if (ip->i_flags & XFS_IRECLAIM) {
1025 if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0) {
1027 } else if ((xfs_ipincount(ip) == 0) &&
1028 xfs_iflock_nowait(ip)) {
1029 IPOINTER_INSERT(ip, mp);
1031 xfs_finish_reclaim(ip, 1,
1032 XFS_IFLUSH_DELWRI_ELSE_ASYNC);
1034 XFS_MOUNT_ILOCK(mp);
1035 mount_locked = B_TRUE;
1036 IPOINTER_REMOVE(ip, mp);
1038 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1049 if (XFS_FORCED_SHUTDOWN(mp) && !(flags & SYNC_CLOSE)) {
1050 XFS_MOUNT_IUNLOCK(mp);
1051 kmem_free(ipointer, sizeof(xfs_iptr_t));
1056 * Try to lock without sleeping. We're out of order with
1057 * the inode list lock here, so if we fail we need to drop
1058 * the mount lock and try again. If we're called from
1059 * bdflush() here, then don't bother.
1061 * The inode lock here actually coordinates with the
1062 * almost spurious inode lock in xfs_ireclaim() to prevent
1063 * the vnode we handle here without a reference from
1064 * being freed while we reference it. If we lock the inode
1065 * while it's on the mount list here, then the spurious inode
1066 * lock in xfs_ireclaim() after the inode is pulled from
1067 * the mount list will sleep until we release it here.
1068 * This keeps the vnode from being freed while we reference
1071 if (xfs_ilock_nowait(ip, lock_flags) == 0) {
1083 IPOINTER_INSERT(ip, mp);
1084 xfs_ilock(ip, lock_flags);
1086 ASSERT(vp == XFS_ITOV(ip));
1087 ASSERT(ip->i_mount == mp);
1089 vnode_refed = B_TRUE;
1092 /* From here on in the loop we may have a marker record
1093 * in the inode list.
1097 * If we have to flush data or wait for I/O completion
1098 * we need to drop the ilock that we currently hold.
1099 * If we need to drop the lock, insert a marker if we
1100 * have not already done so.
1102 if ((flags & (SYNC_CLOSE|SYNC_IOWAIT)) ||
1103 ((flags & SYNC_DELWRI) && VN_DIRTY(vp))) {
1105 IPOINTER_INSERT(ip, mp);
1107 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1109 if (flags & SYNC_CLOSE) {
1110 /* Shutdown case. Flush and invalidate. */
1111 if (XFS_FORCED_SHUTDOWN(mp))
1112 xfs_tosspages(ip, 0, -1,
1115 error = xfs_flushinval_pages(ip,
1117 } else if ((flags & SYNC_DELWRI) && VN_DIRTY(vp)) {
1118 error = xfs_flush_pages(ip, 0,
1119 -1, fflag, FI_NONE);
1123 * When freezing, we need to wait ensure all I/O (including direct
1124 * I/O) is complete to ensure no further data modification can take
1125 * place after this point
1127 if (flags & SYNC_IOWAIT)
1130 xfs_ilock(ip, XFS_ILOCK_SHARED);
1133 if ((flags & SYNC_ATTR) &&
1134 (ip->i_update_core ||
1135 (ip->i_itemp && ip->i_itemp->ili_format.ilf_fields))) {
1137 IPOINTER_INSERT(ip, mp);
1139 if (flags & SYNC_WAIT) {
1141 error = xfs_iflush(ip, XFS_IFLUSH_SYNC);
1144 * If we can't acquire the flush lock, then the inode
1145 * is already being flushed so don't bother waiting.
1147 * If we can lock it then do a delwri flush so we can
1148 * combine multiple inode flushes in each disk write.
1150 } else if (xfs_iflock_nowait(ip)) {
1151 error = xfs_iflush(ip, XFS_IFLUSH_DELWRI);
1152 } else if (bypassed) {
1157 if (lock_flags != 0) {
1158 xfs_iunlock(ip, lock_flags);
1163 * If we had to take a reference on the vnode
1164 * above, then wait until after we've unlocked
1165 * the inode to release the reference. This is
1166 * because we can be already holding the inode
1167 * lock when IRELE() calls xfs_inactive().
1169 * Make sure to drop the mount lock before calling
1170 * IRELE() so that we don't trip over ourselves if
1171 * we have to go for the mount lock again in the
1175 IPOINTER_INSERT(ip, mp);
1180 vnode_refed = B_FALSE;
1188 * bail out if the filesystem is corrupted.
1190 if (error == EFSCORRUPTED) {
1191 if (!mount_locked) {
1192 XFS_MOUNT_ILOCK(mp);
1193 IPOINTER_REMOVE(ip, mp);
1195 XFS_MOUNT_IUNLOCK(mp);
1196 ASSERT(ipointer_in == B_FALSE);
1197 kmem_free(ipointer, sizeof(xfs_iptr_t));
1198 return XFS_ERROR(error);
1201 /* Let other threads have a chance at the mount lock
1202 * if we have looped many times without dropping the
1205 if ((++preempt & XFS_PREEMPT_MASK) == 0) {
1207 IPOINTER_INSERT(ip, mp);
1211 if (mount_locked == B_FALSE) {
1212 XFS_MOUNT_ILOCK(mp);
1213 mount_locked = B_TRUE;
1214 IPOINTER_REMOVE(ip, mp);
1218 ASSERT(ipointer_in == B_FALSE);
1221 } while (ip != mp->m_inodes);
1223 XFS_MOUNT_IUNLOCK(mp);
1225 ASSERT(ipointer_in == B_FALSE);
1227 kmem_free(ipointer, sizeof(xfs_iptr_t));
1228 return XFS_ERROR(last_error);
1232 * xfs sync routine for internal use
1234 * This routine supports all of the flags defined for the generic vfs_sync
1235 * interface as explained above under xfs_sync.
1246 uint log_flags = XFS_LOG_FORCE;
1248 xfs_buf_log_item_t *bip;
1251 * Sync out the log. This ensures that the log is periodically
1252 * flushed even if there is not enough activity to fill it up.
1254 if (flags & SYNC_WAIT)
1255 log_flags |= XFS_LOG_SYNC;
1257 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1259 if (flags & (SYNC_ATTR|SYNC_DELWRI)) {
1260 if (flags & SYNC_BDFLUSH)
1261 xfs_finish_reclaim_all(mp, 1);
1263 error = xfs_sync_inodes(mp, flags, bypassed);
1267 * Flushing out dirty data above probably generated more
1268 * log activity, so if this isn't vfs_sync() then flush
1271 if (flags & SYNC_DELWRI) {
1272 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1275 if (flags & SYNC_FSDATA) {
1277 * If this is vfs_sync() then only sync the superblock
1278 * if we can lock it without sleeping and it is not pinned.
1280 if (flags & SYNC_BDFLUSH) {
1281 bp = xfs_getsb(mp, XFS_BUF_TRYLOCK);
1283 bip = XFS_BUF_FSPRIVATE(bp,xfs_buf_log_item_t*);
1284 if ((bip != NULL) &&
1285 xfs_buf_item_dirty(bip)) {
1286 if (!(XFS_BUF_ISPINNED(bp))) {
1288 error = xfs_bwrite(mp, bp);
1297 bp = xfs_getsb(mp, 0);
1299 * If the buffer is pinned then push on the log so
1300 * we won't get stuck waiting in the write for
1301 * someone, maybe ourselves, to flush the log.
1302 * Even though we just pushed the log above, we
1303 * did not have the superblock buffer locked at
1304 * that point so it can become pinned in between
1307 if (XFS_BUF_ISPINNED(bp))
1308 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
1309 if (flags & SYNC_WAIT)
1310 XFS_BUF_UNASYNC(bp);
1313 error = xfs_bwrite(mp, bp);
1321 * Now check to see if the log needs a "dummy" transaction.
1323 if (!(flags & SYNC_REMOUNT) && xfs_log_need_covered(mp)) {
1328 * Put a dummy transaction in the log to tell
1329 * recovery that all others are OK.
1331 tp = xfs_trans_alloc(mp, XFS_TRANS_DUMMY1);
1332 if ((error = xfs_trans_reserve(tp, 0,
1333 XFS_ICHANGE_LOG_RES(mp),
1335 xfs_trans_cancel(tp, 0);
1340 xfs_ilock(ip, XFS_ILOCK_EXCL);
1342 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1343 xfs_trans_ihold(tp, ip);
1344 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1345 error = xfs_trans_commit(tp, 0);
1346 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1347 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1351 * When shutting down, we need to insure that the AIL is pushed
1352 * to disk or the filesystem can appear corrupt from the PROM.
1354 if ((flags & (SYNC_CLOSE|SYNC_WAIT)) == (SYNC_CLOSE|SYNC_WAIT)) {
1355 XFS_bflush(mp->m_ddev_targp);
1356 if (mp->m_rtdev_targp) {
1357 XFS_bflush(mp->m_rtdev_targp);
1361 return XFS_ERROR(last_error);