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_fsops.h"
56 STATIC int xfs_sync(bhv_desc_t *, int, cred_t *);
61 extern kmem_zone_t *xfs_bmap_free_item_zone;
62 extern kmem_zone_t *xfs_btree_cur_zone;
63 extern kmem_zone_t *xfs_trans_zone;
64 extern kmem_zone_t *xfs_buf_item_zone;
65 extern kmem_zone_t *xfs_dabuf_zone;
66 #ifdef XFS_DABUF_DEBUG
67 extern lock_t xfs_dabuf_global_lock;
68 spinlock_init(&xfs_dabuf_global_lock, "xfsda");
72 * Initialize all of the zone allocators we use.
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_trans_zone = kmem_zone_init(sizeof(xfs_trans_t), "xfs_trans");
80 kmem_zone_init(sizeof(xfs_da_state_t), "xfs_da_state");
81 xfs_dabuf_zone = kmem_zone_init(sizeof(xfs_dabuf_t), "xfs_dabuf");
82 xfs_ifork_zone = kmem_zone_init(sizeof(xfs_ifork_t), "xfs_ifork");
83 xfs_acl_zone_init(xfs_acl_zone, "xfs_acl");
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_chashlist_t), "xfs_chashlist",
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_bmap_free_item_zone;
154 extern kmem_zone_t *xfs_btree_cur_zone;
155 extern kmem_zone_t *xfs_inode_zone;
156 extern kmem_zone_t *xfs_trans_zone;
157 extern kmem_zone_t *xfs_da_state_zone;
158 extern kmem_zone_t *xfs_dabuf_zone;
159 extern kmem_zone_t *xfs_efd_zone;
160 extern kmem_zone_t *xfs_efi_zone;
161 extern kmem_zone_t *xfs_buf_item_zone;
162 extern kmem_zone_t *xfs_chashlist_zone;
164 xfs_cleanup_procfs();
165 xfs_sysctl_unregister();
166 xfs_refcache_destroy();
167 xfs_acl_zone_destroy(xfs_acl_zone);
169 #ifdef XFS_DIR2_TRACE
170 ktrace_free(xfs_dir2_trace_buf);
172 #ifdef XFS_ATTR_TRACE
173 ktrace_free(xfs_attr_trace_buf);
175 #ifdef XFS_BMBT_TRACE
176 ktrace_free(xfs_bmbt_trace_buf);
178 #ifdef XFS_BMAP_TRACE
179 ktrace_free(xfs_bmap_trace_buf);
181 #ifdef XFS_ALLOC_TRACE
182 ktrace_free(xfs_alloc_trace_buf);
185 kmem_zone_destroy(xfs_bmap_free_item_zone);
186 kmem_zone_destroy(xfs_btree_cur_zone);
187 kmem_zone_destroy(xfs_inode_zone);
188 kmem_zone_destroy(xfs_trans_zone);
189 kmem_zone_destroy(xfs_da_state_zone);
190 kmem_zone_destroy(xfs_dabuf_zone);
191 kmem_zone_destroy(xfs_buf_item_zone);
192 kmem_zone_destroy(xfs_efd_zone);
193 kmem_zone_destroy(xfs_efi_zone);
194 kmem_zone_destroy(xfs_ifork_zone);
195 kmem_zone_destroy(xfs_ili_zone);
196 kmem_zone_destroy(xfs_chashlist_zone);
202 * This function fills in xfs_mount_t fields based on mount args.
203 * Note: the superblock has _not_ yet been read in.
208 struct xfs_mount_args *ap,
209 struct xfs_mount *mp)
211 /* Values are in BBs */
212 if ((ap->flags & XFSMNT_NOALIGN) != XFSMNT_NOALIGN) {
214 * At this point the superblock has not been read
215 * in, therefore we do not know the block size.
216 * Before the mount call ends we will convert
219 mp->m_dalign = ap->sunit;
220 mp->m_swidth = ap->swidth;
223 if (ap->logbufs != -1 &&
225 (ap->logbufs < XLOG_MIN_ICLOGS ||
226 ap->logbufs > XLOG_MAX_ICLOGS)) {
228 "XFS: invalid logbufs value: %d [not %d-%d]",
229 ap->logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
230 return XFS_ERROR(EINVAL);
232 mp->m_logbufs = ap->logbufs;
233 if (ap->logbufsize != -1 &&
234 ap->logbufsize != 0 &&
235 ap->logbufsize != 16 * 1024 &&
236 ap->logbufsize != 32 * 1024 &&
237 ap->logbufsize != 64 * 1024 &&
238 ap->logbufsize != 128 * 1024 &&
239 ap->logbufsize != 256 * 1024) {
241 "XFS: invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
243 return XFS_ERROR(EINVAL);
245 mp->m_ihsize = ap->ihashsize;
246 mp->m_logbsize = ap->logbufsize;
247 mp->m_fsname_len = strlen(ap->fsname) + 1;
248 mp->m_fsname = kmem_alloc(mp->m_fsname_len, KM_SLEEP);
249 strcpy(mp->m_fsname, ap->fsname);
251 mp->m_rtname = kmem_alloc(strlen(ap->rtname) + 1, KM_SLEEP);
252 strcpy(mp->m_rtname, ap->rtname);
254 if (ap->logname[0]) {
255 mp->m_logname = kmem_alloc(strlen(ap->logname) + 1, KM_SLEEP);
256 strcpy(mp->m_logname, ap->logname);
259 if (ap->flags & XFSMNT_WSYNC)
260 mp->m_flags |= XFS_MOUNT_WSYNC;
262 if (ap->flags & XFSMNT_INO64) {
263 mp->m_flags |= XFS_MOUNT_INO64;
264 mp->m_inoadd = XFS_INO64_OFFSET;
267 if (ap->flags & XFSMNT_RETERR)
268 mp->m_flags |= XFS_MOUNT_RETERR;
269 if (ap->flags & XFSMNT_NOALIGN)
270 mp->m_flags |= XFS_MOUNT_NOALIGN;
271 if (ap->flags & XFSMNT_SWALLOC)
272 mp->m_flags |= XFS_MOUNT_SWALLOC;
273 if (ap->flags & XFSMNT_OSYNCISOSYNC)
274 mp->m_flags |= XFS_MOUNT_OSYNCISOSYNC;
275 if (ap->flags & XFSMNT_32BITINODES)
276 mp->m_flags |= XFS_MOUNT_32BITINODES;
278 if (ap->flags & XFSMNT_IOSIZE) {
279 if (ap->iosizelog > XFS_MAX_IO_LOG ||
280 ap->iosizelog < XFS_MIN_IO_LOG) {
282 "XFS: invalid log iosize: %d [not %d-%d]",
283 ap->iosizelog, XFS_MIN_IO_LOG,
285 return XFS_ERROR(EINVAL);
288 mp->m_flags |= XFS_MOUNT_DFLT_IOSIZE;
289 mp->m_readio_log = mp->m_writeio_log = ap->iosizelog;
292 if (ap->flags & XFSMNT_IHASHSIZE)
293 mp->m_flags |= XFS_MOUNT_IHASHSIZE;
294 if (ap->flags & XFSMNT_IDELETE)
295 mp->m_flags |= XFS_MOUNT_IDELETE;
296 if (ap->flags & XFSMNT_DIRSYNC)
297 mp->m_flags |= XFS_MOUNT_DIRSYNC;
298 if (ap->flags & XFSMNT_ATTR2)
299 mp->m_flags |= XFS_MOUNT_ATTR2;
301 if (ap->flags2 & XFSMNT2_COMPAT_IOSIZE)
302 mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE;
305 * no recovery flag requires a read-only mount
307 if (ap->flags & XFSMNT_NORECOVERY) {
308 if (!(vfs->vfs_flag & VFS_RDONLY)) {
310 "XFS: tried to mount a FS read-write without recovery!");
311 return XFS_ERROR(EINVAL);
313 mp->m_flags |= XFS_MOUNT_NORECOVERY;
316 if (ap->flags & XFSMNT_NOUUID)
317 mp->m_flags |= XFS_MOUNT_NOUUID;
318 if (ap->flags & XFSMNT_BARRIER)
319 mp->m_flags |= XFS_MOUNT_BARRIER;
321 mp->m_flags &= ~XFS_MOUNT_BARRIER;
327 * This function fills in xfs_mount_t fields based on mount args.
328 * Note: the superblock _has_ now been read in.
333 struct xfs_mount_args *ap,
334 struct xfs_mount *mp)
336 int ronly = (vfs->vfs_flag & VFS_RDONLY);
338 /* Fail a mount where the logbuf is smaller then the log stripe */
339 if (XFS_SB_VERSION_HASLOGV2(&mp->m_sb)) {
340 if ((ap->logbufsize <= 0) &&
341 (mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE)) {
342 mp->m_logbsize = mp->m_sb.sb_logsunit;
343 } else if (ap->logbufsize > 0 &&
344 ap->logbufsize < mp->m_sb.sb_logsunit) {
346 "XFS: logbuf size must be greater than or equal to log stripe size");
347 return XFS_ERROR(EINVAL);
350 /* Fail a mount if the logbuf is larger than 32K */
351 if (ap->logbufsize > XLOG_BIG_RECORD_BSIZE) {
353 "XFS: logbuf size for version 1 logs must be 16K or 32K");
354 return XFS_ERROR(EINVAL);
358 if (XFS_SB_VERSION_HASATTR2(&mp->m_sb)) {
359 mp->m_flags |= XFS_MOUNT_ATTR2;
363 * prohibit r/w mounts of read-only filesystems
365 if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !ronly) {
367 "XFS: cannot mount a read-only filesystem as read-write");
368 return XFS_ERROR(EROFS);
372 * check for shared mount.
374 if (ap->flags & XFSMNT_SHARED) {
375 if (!XFS_SB_VERSION_HASSHARED(&mp->m_sb))
376 return XFS_ERROR(EINVAL);
379 * For IRIX 6.5, shared mounts must have the shared
380 * version bit set, have the persistent readonly
381 * field set, must be version 0 and can only be mounted
384 if (!ronly || !(mp->m_sb.sb_flags & XFS_SBF_READONLY) ||
385 (mp->m_sb.sb_shared_vn != 0))
386 return XFS_ERROR(EINVAL);
388 mp->m_flags |= XFS_MOUNT_SHARED;
391 * Shared XFS V0 can't deal with DMI. Return EINVAL.
393 if (mp->m_sb.sb_shared_vn == 0 && (ap->flags & XFSMNT_DMAPI))
394 return XFS_ERROR(EINVAL);
403 * The file system configurations are:
404 * (1) device (partition) with data and internal log
405 * (2) logical volume with data and log subvolumes.
406 * (3) logical volume with data, log, and realtime subvolumes.
408 * We only have to handle opening the log and realtime volumes here if
409 * they are present. The data subvolume has already been opened by
410 * get_sb_bdev() and is stored in vfsp->vfs_super->s_bdev.
414 struct bhv_desc *bhvp,
415 struct xfs_mount_args *args,
418 struct bhv_vfs *vfsp = bhvtovfs(bhvp);
420 struct xfs_mount *mp = XFS_BHVTOM(bhvp);
421 struct block_device *ddev, *logdev, *rtdev;
422 int flags = 0, error;
424 ddev = vfsp->vfs_super->s_bdev;
425 logdev = rtdev = NULL;
428 * Setup xfs_mount function vectors from available behaviors
430 p = vfs_bhv_lookup(vfsp, VFS_POSITION_DM);
431 mp->m_dm_ops = p ? *(xfs_dmops_t *) vfs_bhv_custom(p) : xfs_dmcore_stub;
432 p = vfs_bhv_lookup(vfsp, VFS_POSITION_QM);
433 mp->m_qm_ops = p ? *(xfs_qmops_t *) vfs_bhv_custom(p) : xfs_qmcore_stub;
434 p = vfs_bhv_lookup(vfsp, VFS_POSITION_IO);
435 mp->m_io_ops = p ? *(xfs_ioops_t *) vfs_bhv_custom(p) : xfs_iocore_xfs;
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)
479 mp->m_logdev_targp = (logdev && logdev != ddev) ?
480 xfs_alloc_buftarg(logdev, 1) : mp->m_ddev_targp;
481 if (!mp->m_logdev_targp)
485 * Setup flags based on mount(2) options and then the superblock
487 error = xfs_start_flags(vfsp, args, mp);
490 error = xfs_readsb(mp, flags);
493 error = xfs_finish_flags(vfsp, args, mp);
498 * Setup xfs_mount buffer target pointers based on superblock
500 error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_blocksize,
501 mp->m_sb.sb_sectsize);
502 if (!error && logdev && logdev != ddev) {
503 unsigned int log_sector_size = BBSIZE;
505 if (XFS_SB_VERSION_HASSECTOR(&mp->m_sb))
506 log_sector_size = mp->m_sb.sb_logsectsize;
507 error = xfs_setsize_buftarg(mp->m_logdev_targp,
508 mp->m_sb.sb_blocksize,
512 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
513 mp->m_sb.sb_blocksize,
514 mp->m_sb.sb_sectsize);
518 if (mp->m_flags & XFS_MOUNT_BARRIER)
519 xfs_mountfs_check_barriers(mp);
521 error = XFS_IOINIT(vfsp, args, flags);
531 xfs_binval(mp->m_ddev_targp);
532 if (logdev && logdev != ddev)
533 xfs_binval(mp->m_logdev_targp);
535 xfs_binval(mp->m_rtdev_targp);
537 xfs_unmountfs_close(mp, credp);
547 bhv_vfs_t *vfsp = bhvtovfs(bdp);
548 xfs_mount_t *mp = XFS_BHVTOM(bdp);
551 int unmount_event_wanted = 0;
552 int unmount_event_flags = 0;
553 int xfs_unmountfs_needed = 0;
559 if (vfsp->vfs_flag & VFS_DMI) {
560 error = XFS_SEND_PREUNMOUNT(mp, vfsp,
561 rvp, DM_RIGHT_NULL, rvp, DM_RIGHT_NULL,
563 (mp->m_dmevmask & (1<<DM_EVENT_PREUNMOUNT))?
564 0:DM_FLAGS_UNWANTED);
566 return XFS_ERROR(error);
567 unmount_event_wanted = 1;
568 unmount_event_flags = (mp->m_dmevmask & (1<<DM_EVENT_UNMOUNT))?
569 0 : DM_FLAGS_UNWANTED;
573 * First blow any referenced inode from this file system
574 * out of the reference cache, and delete the timer.
576 xfs_refcache_purge_mp(mp);
578 XFS_bflush(mp->m_ddev_targp);
579 error = xfs_unmount_flush(mp, 0);
583 ASSERT(vn_count(rvp) == 1);
586 * Drop the reference count
591 * If we're forcing a shutdown, typically because of a media error,
592 * we want to make sure we invalidate dirty pages that belong to
593 * referenced vnodes as well.
595 if (XFS_FORCED_SHUTDOWN(mp)) {
596 error = xfs_sync(&mp->m_bhv,
597 (SYNC_WAIT | SYNC_CLOSE), credp);
598 ASSERT(error != EFSCORRUPTED);
600 xfs_unmountfs_needed = 1;
603 /* Send DMAPI event, if required.
604 * Then do xfs_unmountfs() if needed.
605 * Then return error (or zero).
607 if (unmount_event_wanted) {
608 /* Note: mp structure must still exist for
609 * XFS_SEND_UNMOUNT() call.
611 XFS_SEND_UNMOUNT(mp, vfsp, error == 0 ? rvp : NULL,
612 DM_RIGHT_NULL, 0, error, unmount_event_flags);
614 if (xfs_unmountfs_needed) {
616 * Call common unmount function to flush to disk
617 * and free the super block buffer & mount structures.
619 xfs_unmountfs(mp, credp);
622 return XFS_ERROR(error);
629 int count = 0, pincount;
631 xfs_refcache_purge_mp(mp);
632 xfs_flush_buftarg(mp->m_ddev_targp, 0);
633 xfs_finish_reclaim_all(mp, 0);
635 /* This loop must run at least twice.
636 * The first instance of the loop will flush
637 * most meta data but that will generate more
638 * meta data (typically directory updates).
639 * Which then must be flushed and logged before
640 * we can write the unmount record.
643 xfs_syncsub(mp, SYNC_INODE_QUIESCE, NULL);
644 pincount = xfs_flush_buftarg(mp->m_ddev_targp, 1);
655 * Second stage of a quiesce. The data is already synced, now we have to take
656 * care of the metadata. New transactions are already blocked, so we need to
657 * wait for any remaining transactions to drain out before proceding.
663 /* wait for all modifications to complete */
664 while (atomic_read(&mp->m_active_trans) > 0)
667 /* flush inodes and push all remaining buffers out to disk */
670 ASSERT_ALWAYS(atomic_read(&mp->m_active_trans) == 0);
672 /* Push the superblock and write an unmount record */
673 xfs_log_sbcount(mp, 1);
674 xfs_log_unmount_write(mp);
675 xfs_unmountfs_writesb(mp);
682 struct xfs_mount_args *args)
684 bhv_vfs_t *vfsp = bhvtovfs(bdp);
685 xfs_mount_t *mp = XFS_BHVTOM(bdp);
687 if (!(*flags & MS_RDONLY)) { /* rw/ro -> rw */
688 if (vfsp->vfs_flag & VFS_RDONLY)
689 vfsp->vfs_flag &= ~VFS_RDONLY;
690 if (args->flags & XFSMNT_BARRIER) {
691 mp->m_flags |= XFS_MOUNT_BARRIER;
692 xfs_mountfs_check_barriers(mp);
694 mp->m_flags &= ~XFS_MOUNT_BARRIER;
696 } else if (!(vfsp->vfs_flag & VFS_RDONLY)) { /* rw -> ro */
697 bhv_vfs_sync(vfsp, SYNC_DATA_QUIESCE, NULL);
698 xfs_attr_quiesce(mp);
699 vfsp->vfs_flag |= VFS_RDONLY;
705 * xfs_unmount_flush implements a set of flush operation on special
706 * inodes, which are needed as a separate set of operations so that
707 * they can be called as part of relocation process.
711 xfs_mount_t *mp, /* Mount structure we are getting
713 int relocation) /* Called from vfs relocation. */
715 xfs_inode_t *rip = mp->m_rootip;
717 xfs_inode_t *rsumip = NULL;
718 bhv_vnode_t *rvp = XFS_ITOV(rip);
721 xfs_ilock(rip, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
725 * Flush out the real time inodes.
727 if ((rbmip = mp->m_rbmip) != NULL) {
728 xfs_ilock(rbmip, XFS_ILOCK_EXCL);
730 error = xfs_iflush(rbmip, XFS_IFLUSH_SYNC);
731 xfs_iunlock(rbmip, XFS_ILOCK_EXCL);
733 if (error == EFSCORRUPTED)
736 ASSERT(vn_count(XFS_ITOV(rbmip)) == 1);
738 rsumip = mp->m_rsumip;
739 xfs_ilock(rsumip, XFS_ILOCK_EXCL);
741 error = xfs_iflush(rsumip, XFS_IFLUSH_SYNC);
742 xfs_iunlock(rsumip, XFS_ILOCK_EXCL);
744 if (error == EFSCORRUPTED)
747 ASSERT(vn_count(XFS_ITOV(rsumip)) == 1);
751 * Synchronously flush root inode to disk
753 error = xfs_iflush(rip, XFS_IFLUSH_SYNC);
754 if (error == EFSCORRUPTED)
757 if (vn_count(rvp) != 1 && !relocation) {
758 xfs_iunlock(rip, XFS_ILOCK_EXCL);
759 return XFS_ERROR(EBUSY);
763 * Release dquot that rootinode, rbmino and rsumino might be holding,
764 * flush and purge the quota inodes.
766 error = XFS_QM_UNMOUNT(mp);
767 if (error == EFSCORRUPTED)
771 VN_RELE(XFS_ITOV(rbmip));
772 VN_RELE(XFS_ITOV(rsumip));
775 xfs_iunlock(rip, XFS_ILOCK_EXCL);
782 xfs_iunlock(rip, XFS_ILOCK_EXCL);
784 return XFS_ERROR(EFSCORRUPTED);
788 * xfs_root extracts the root vnode from a vfs.
790 * vfsp -- the vfs struct for the desired file system
791 * vpp -- address of the caller's vnode pointer which should be
792 * set to the desired fs root vnode
801 vp = XFS_ITOV((XFS_BHVTOM(bdp))->m_rootip);
810 * Fill in the statvfs structure for the given file system. We use
811 * the superblock lock in the mount structure to ensure a consistent
812 * snapshot of the counters returned.
817 bhv_statvfs_t *statp,
826 mp = XFS_BHVTOM(bdp);
829 statp->f_type = XFS_SB_MAGIC;
831 xfs_icsb_sync_counters_flags(mp, XFS_ICSB_LAZY_COUNT);
833 statp->f_bsize = sbp->sb_blocksize;
834 lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
835 statp->f_blocks = sbp->sb_dblocks - lsize;
836 statp->f_bfree = statp->f_bavail =
837 sbp->sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
838 fakeinos = statp->f_bfree << sbp->sb_inopblog;
840 fakeinos += mp->m_inoadd;
843 MIN(sbp->sb_icount + fakeinos, (__uint64_t)XFS_MAXINUMBER);
848 statp->f_files = min_t(typeof(statp->f_files),
851 statp->f_ffree = statp->f_files - (sbp->sb_icount - sbp->sb_ifree);
852 XFS_SB_UNLOCK(mp, s);
854 xfs_statvfs_fsid(statp, mp);
855 statp->f_namelen = MAXNAMELEN - 1;
862 * xfs_sync flushes any pending I/O to file system vfsp.
864 * This routine is called by vfs_sync() to make sure that things make it
865 * out to disk eventually, on sync() system calls to flush out everything,
866 * and when the file system is unmounted. For the vfs_sync() case, all
867 * we really need to do is sync out the log to make all of our meta-data
868 * updates permanent (except for timestamps). For calls from pflushd(),
869 * dirty pages are kept moving by calling pdflush() on the inodes
870 * containing them. We also flush the inodes that we can lock without
871 * sleeping and the superblock if we can lock it without sleeping from
872 * vfs_sync() so that items at the tail of the log are always moving out.
875 * SYNC_BDFLUSH - We're being called from vfs_sync() so we don't want
876 * to sleep if we can help it. All we really need
877 * to do is ensure that the log is synced at least
878 * periodically. We also push the inodes and
879 * superblock if we can lock them without sleeping
880 * and they are not pinned.
881 * SYNC_ATTR - We need to flush the inodes. If SYNC_BDFLUSH is not
882 * set, then we really want to lock each inode and flush
884 * SYNC_WAIT - All the flushes that take place in this call should
886 * SYNC_DELWRI - This tells us to push dirty pages associated with
887 * inodes. SYNC_WAIT and SYNC_BDFLUSH are used to
888 * determine if they should be flushed sync, async, or
890 * SYNC_CLOSE - This flag is passed when the system is being
891 * unmounted. We should sync and invalidate everything.
892 * SYNC_FSDATA - This indicates that the caller would like to make
893 * sure the superblock is safe on disk. We can ensure
894 * this by simply making sure the log gets flushed
895 * if SYNC_BDFLUSH is set, and by actually writing it
897 * SYNC_IOWAIT - The caller wants us to wait for all data I/O to complete
898 * before we return (including direct I/O). Forms the drain
899 * side of the write barrier needed to safely quiesce the
910 xfs_mount_t *mp = XFS_BHVTOM(bdp);
912 return xfs_syncsub(mp, flags, NULL);
916 * xfs sync routine for internal use
918 * This routine supports all of the flags defined for the generic vfs_sync
919 * interface as explained above under xfs_sync.
928 xfs_inode_t *ip = NULL;
929 xfs_inode_t *ip_next;
931 bhv_vnode_t *vp = NULL;
936 uint base_lock_flags;
937 boolean_t mount_locked;
938 boolean_t vnode_refed;
941 xfs_iptr_t *ipointer;
943 boolean_t ipointer_in = B_FALSE;
945 #define IPOINTER_SET ipointer_in = B_TRUE
946 #define IPOINTER_CLR ipointer_in = B_FALSE
953 /* Insert a marker record into the inode list after inode ip. The list
954 * must be locked when this is called. After the call the list will no
957 #define IPOINTER_INSERT(ip, mp) { \
958 ASSERT(ipointer_in == B_FALSE); \
959 ipointer->ip_mnext = ip->i_mnext; \
960 ipointer->ip_mprev = ip; \
961 ip->i_mnext = (xfs_inode_t *)ipointer; \
962 ipointer->ip_mnext->i_mprev = (xfs_inode_t *)ipointer; \
964 XFS_MOUNT_IUNLOCK(mp); \
965 mount_locked = B_FALSE; \
969 /* Remove the marker from the inode list. If the marker was the only item
970 * in the list then there are no remaining inodes and we should zero out
971 * the whole list. If we are the current head of the list then move the head
974 #define IPOINTER_REMOVE(ip, mp) { \
975 ASSERT(ipointer_in == B_TRUE); \
976 if (ipointer->ip_mnext != (xfs_inode_t *)ipointer) { \
977 ip = ipointer->ip_mnext; \
978 ip->i_mprev = ipointer->ip_mprev; \
979 ipointer->ip_mprev->i_mnext = ip; \
980 if (mp->m_inodes == (xfs_inode_t *)ipointer) { \
984 ASSERT(mp->m_inodes == (xfs_inode_t *)ipointer); \
985 mp->m_inodes = NULL; \
991 #define XFS_PREEMPT_MASK 0x7f
995 if (XFS_MTOVFS(mp)->vfs_flag & VFS_RDONLY)
1001 /* Allocate a reference marker */
1002 ipointer = (xfs_iptr_t *)kmem_zalloc(sizeof(xfs_iptr_t), KM_SLEEP);
1004 fflag = XFS_B_ASYNC; /* default is don't wait */
1005 if (flags & (SYNC_BDFLUSH | SYNC_DELWRI))
1006 fflag = XFS_B_DELWRI;
1007 if (flags & SYNC_WAIT)
1008 fflag = 0; /* synchronous overrides all */
1010 base_lock_flags = XFS_ILOCK_SHARED;
1011 if (flags & (SYNC_DELWRI | SYNC_CLOSE)) {
1013 * We need the I/O lock if we're going to call any of
1014 * the flush/inval routines.
1016 base_lock_flags |= XFS_IOLOCK_SHARED;
1019 XFS_MOUNT_ILOCK(mp);
1023 mount_locked = B_TRUE;
1024 vnode_refed = B_FALSE;
1029 ASSERT(ipointer_in == B_FALSE);
1030 ASSERT(vnode_refed == B_FALSE);
1032 lock_flags = base_lock_flags;
1035 * There were no inodes in the list, just break out
1043 * We found another sync thread marker - skip it
1045 if (ip->i_mount == NULL) {
1050 vp = XFS_ITOV_NULL(ip);
1053 * If the vnode is gone then this is being torn down,
1054 * call reclaim if it is flushed, else let regular flush
1055 * code deal with it later in the loop.
1059 /* Skip ones already in reclaim */
1060 if (ip->i_flags & XFS_IRECLAIM) {
1064 if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0) {
1066 } else if ((xfs_ipincount(ip) == 0) &&
1067 xfs_iflock_nowait(ip)) {
1068 IPOINTER_INSERT(ip, mp);
1070 xfs_finish_reclaim(ip, 1,
1071 XFS_IFLUSH_DELWRI_ELSE_ASYNC);
1073 XFS_MOUNT_ILOCK(mp);
1074 mount_locked = B_TRUE;
1075 IPOINTER_REMOVE(ip, mp);
1077 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1088 if (XFS_FORCED_SHUTDOWN(mp) && !(flags & SYNC_CLOSE)) {
1089 XFS_MOUNT_IUNLOCK(mp);
1090 kmem_free(ipointer, sizeof(xfs_iptr_t));
1095 * If this is just vfs_sync() or pflushd() calling
1096 * then we can skip inodes for which it looks like
1097 * there is nothing to do. Since we don't have the
1098 * inode locked this is racy, but these are periodic
1099 * calls so it doesn't matter. For the others we want
1100 * to know for sure, so we at least try to lock them.
1102 if (flags & SYNC_BDFLUSH) {
1103 if (((ip->i_itemp == NULL) ||
1104 !(ip->i_itemp->ili_format.ilf_fields &
1106 (ip->i_update_core == 0)) {
1113 * Try to lock without sleeping. We're out of order with
1114 * the inode list lock here, so if we fail we need to drop
1115 * the mount lock and try again. If we're called from
1116 * bdflush() here, then don't bother.
1118 * The inode lock here actually coordinates with the
1119 * almost spurious inode lock in xfs_ireclaim() to prevent
1120 * the vnode we handle here without a reference from
1121 * being freed while we reference it. If we lock the inode
1122 * while it's on the mount list here, then the spurious inode
1123 * lock in xfs_ireclaim() after the inode is pulled from
1124 * the mount list will sleep until we release it here.
1125 * This keeps the vnode from being freed while we reference
1128 if (xfs_ilock_nowait(ip, lock_flags) == 0) {
1129 if ((flags & SYNC_BDFLUSH) || (vp == NULL)) {
1140 IPOINTER_INSERT(ip, mp);
1141 xfs_ilock(ip, lock_flags);
1143 ASSERT(vp == XFS_ITOV(ip));
1144 ASSERT(ip->i_mount == mp);
1146 vnode_refed = B_TRUE;
1149 /* From here on in the loop we may have a marker record
1150 * in the inode list.
1154 * If we have to flush data or wait for I/O completion
1155 * we need to drop the ilock that we currently hold.
1156 * If we need to drop the lock, insert a marker if we
1157 * have not already done so.
1159 if ((flags & (SYNC_CLOSE|SYNC_IOWAIT)) ||
1160 ((flags & SYNC_DELWRI) && VN_DIRTY(vp))) {
1162 IPOINTER_INSERT(ip, mp);
1164 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1166 if (flags & SYNC_CLOSE) {
1167 /* Shutdown case. Flush and invalidate. */
1168 if (XFS_FORCED_SHUTDOWN(mp))
1169 bhv_vop_toss_pages(vp, 0, -1, FI_REMAPF);
1171 error = bhv_vop_flushinval_pages(vp, 0,
1173 } else if ((flags & SYNC_DELWRI) && VN_DIRTY(vp)) {
1174 error = bhv_vop_flush_pages(vp, (xfs_off_t)0,
1175 -1, fflag, FI_NONE);
1179 * When freezing, we need to wait ensure all I/O (including direct
1180 * I/O) is complete to ensure no further data modification can take
1181 * place after this point
1183 if (flags & SYNC_IOWAIT)
1186 xfs_ilock(ip, XFS_ILOCK_SHARED);
1189 if (flags & SYNC_BDFLUSH) {
1190 if ((flags & SYNC_ATTR) &&
1191 ((ip->i_update_core) ||
1192 ((ip->i_itemp != NULL) &&
1193 (ip->i_itemp->ili_format.ilf_fields != 0)))) {
1195 /* Insert marker and drop lock if not already
1199 IPOINTER_INSERT(ip, mp);
1203 * We don't want the periodic flushing of the
1204 * inodes by vfs_sync() to interfere with
1205 * I/O to the file, especially read I/O
1206 * where it is only the access time stamp
1207 * that is being flushed out. To prevent
1208 * long periods where we have both inode
1209 * locks held shared here while reading the
1210 * inode's buffer in from disk, we drop the
1211 * inode lock while reading in the inode
1212 * buffer. We have to release the buffer
1213 * and reacquire the inode lock so that they
1214 * are acquired in the proper order (inode
1215 * locks first). The buffer will go at the
1216 * end of the lru chain, though, so we can
1217 * expect it to still be there when we go
1218 * for it again in xfs_iflush().
1220 if ((xfs_ipincount(ip) == 0) &&
1221 xfs_iflock_nowait(ip)) {
1224 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1226 error = xfs_itobp(mp, NULL, ip,
1231 /* Bailing out, remove the
1232 * marker and free it.
1234 XFS_MOUNT_ILOCK(mp);
1235 IPOINTER_REMOVE(ip, mp);
1236 XFS_MOUNT_IUNLOCK(mp);
1238 ASSERT(!(lock_flags &
1239 XFS_IOLOCK_SHARED));
1242 sizeof(xfs_iptr_t));
1247 * Since we dropped the inode lock,
1248 * the inode may have been reclaimed.
1249 * Therefore, we reacquire the mount
1250 * lock and check to see if we were the
1251 * inode reclaimed. If this happened
1252 * then the ipointer marker will no
1253 * longer point back at us. In this
1254 * case, move ip along to the inode
1255 * after the marker, remove the marker
1258 XFS_MOUNT_ILOCK(mp);
1259 mount_locked = B_TRUE;
1261 if (ip != ipointer->ip_mprev) {
1262 IPOINTER_REMOVE(ip, mp);
1264 ASSERT(!vnode_refed);
1265 ASSERT(!(lock_flags &
1266 XFS_IOLOCK_SHARED));
1270 ASSERT(ip->i_mount == mp);
1272 if (xfs_ilock_nowait(ip,
1273 XFS_ILOCK_SHARED) == 0) {
1274 ASSERT(ip->i_mount == mp);
1276 * We failed to reacquire
1277 * the inode lock without
1278 * sleeping, so just skip
1279 * the inode for now. We
1280 * clear the ILOCK bit from
1281 * the lock_flags so that we
1282 * won't try to drop a lock
1283 * we don't hold below.
1285 lock_flags &= ~XFS_ILOCK_SHARED;
1286 IPOINTER_REMOVE(ip_next, mp);
1287 } else if ((xfs_ipincount(ip) == 0) &&
1288 xfs_iflock_nowait(ip)) {
1289 ASSERT(ip->i_mount == mp);
1291 * Since this is vfs_sync()
1292 * calling we only flush the
1293 * inode out if we can lock
1294 * it without sleeping and
1295 * it is not pinned. Drop
1296 * the mount lock here so
1297 * that we don't hold it for
1298 * too long. We already have
1299 * a marker in the list here.
1301 XFS_MOUNT_IUNLOCK(mp);
1302 mount_locked = B_FALSE;
1303 error = xfs_iflush(ip,
1306 ASSERT(ip->i_mount == mp);
1307 IPOINTER_REMOVE(ip_next, mp);
1314 if ((flags & SYNC_ATTR) &&
1315 ((ip->i_update_core) ||
1316 ((ip->i_itemp != NULL) &&
1317 (ip->i_itemp->ili_format.ilf_fields != 0)))) {
1319 IPOINTER_INSERT(ip, mp);
1322 if (flags & SYNC_WAIT) {
1324 error = xfs_iflush(ip,
1328 * If we can't acquire the flush
1329 * lock, then the inode is already
1330 * being flushed so don't bother
1331 * waiting. If we can lock it then
1332 * do a delwri flush so we can
1333 * combine multiple inode flushes
1334 * in each disk write.
1336 if (xfs_iflock_nowait(ip)) {
1337 error = xfs_iflush(ip,
1346 if (lock_flags != 0) {
1347 xfs_iunlock(ip, lock_flags);
1352 * If we had to take a reference on the vnode
1353 * above, then wait until after we've unlocked
1354 * the inode to release the reference. This is
1355 * because we can be already holding the inode
1356 * lock when VN_RELE() calls xfs_inactive().
1358 * Make sure to drop the mount lock before calling
1359 * VN_RELE() so that we don't trip over ourselves if
1360 * we have to go for the mount lock again in the
1364 IPOINTER_INSERT(ip, mp);
1369 vnode_refed = B_FALSE;
1377 * bail out if the filesystem is corrupted.
1379 if (error == EFSCORRUPTED) {
1380 if (!mount_locked) {
1381 XFS_MOUNT_ILOCK(mp);
1382 IPOINTER_REMOVE(ip, mp);
1384 XFS_MOUNT_IUNLOCK(mp);
1385 ASSERT(ipointer_in == B_FALSE);
1386 kmem_free(ipointer, sizeof(xfs_iptr_t));
1387 return XFS_ERROR(error);
1390 /* Let other threads have a chance at the mount lock
1391 * if we have looped many times without dropping the
1394 if ((++preempt & XFS_PREEMPT_MASK) == 0) {
1396 IPOINTER_INSERT(ip, mp);
1400 if (mount_locked == B_FALSE) {
1401 XFS_MOUNT_ILOCK(mp);
1402 mount_locked = B_TRUE;
1403 IPOINTER_REMOVE(ip, mp);
1407 ASSERT(ipointer_in == B_FALSE);
1410 } while (ip != mp->m_inodes);
1412 XFS_MOUNT_IUNLOCK(mp);
1414 ASSERT(ipointer_in == B_FALSE);
1416 kmem_free(ipointer, sizeof(xfs_iptr_t));
1417 return XFS_ERROR(last_error);
1421 * xfs sync routine for internal use
1423 * This routine supports all of the flags defined for the generic vfs_sync
1424 * interface as explained above under xfs_sync.
1435 uint log_flags = XFS_LOG_FORCE;
1437 xfs_buf_log_item_t *bip;
1440 * Sync out the log. This ensures that the log is periodically
1441 * flushed even if there is not enough activity to fill it up.
1443 if (flags & SYNC_WAIT)
1444 log_flags |= XFS_LOG_SYNC;
1446 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1448 if (flags & (SYNC_ATTR|SYNC_DELWRI)) {
1449 if (flags & SYNC_BDFLUSH)
1450 xfs_finish_reclaim_all(mp, 1);
1452 error = xfs_sync_inodes(mp, flags, bypassed);
1456 * Flushing out dirty data above probably generated more
1457 * log activity, so if this isn't vfs_sync() then flush
1460 if (flags & SYNC_DELWRI) {
1461 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1464 if (flags & SYNC_FSDATA) {
1466 * If this is vfs_sync() then only sync the superblock
1467 * if we can lock it without sleeping and it is not pinned.
1469 if (flags & SYNC_BDFLUSH) {
1470 bp = xfs_getsb(mp, XFS_BUF_TRYLOCK);
1472 bip = XFS_BUF_FSPRIVATE(bp,xfs_buf_log_item_t*);
1473 if ((bip != NULL) &&
1474 xfs_buf_item_dirty(bip)) {
1475 if (!(XFS_BUF_ISPINNED(bp))) {
1477 error = xfs_bwrite(mp, bp);
1486 bp = xfs_getsb(mp, 0);
1488 * If the buffer is pinned then push on the log so
1489 * we won't get stuck waiting in the write for
1490 * someone, maybe ourselves, to flush the log.
1491 * Even though we just pushed the log above, we
1492 * did not have the superblock buffer locked at
1493 * that point so it can become pinned in between
1496 if (XFS_BUF_ISPINNED(bp))
1497 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
1498 if (flags & SYNC_WAIT)
1499 XFS_BUF_UNASYNC(bp);
1502 error = xfs_bwrite(mp, bp);
1510 * If this is the periodic sync, then kick some entries out of
1511 * the reference cache. This ensures that idle entries are
1512 * eventually kicked out of the cache.
1514 if (flags & SYNC_REFCACHE) {
1515 if (flags & SYNC_WAIT)
1516 xfs_refcache_purge_mp(mp);
1518 xfs_refcache_purge_some(mp);
1522 * If asked, update the disk superblock with incore counter values if we
1523 * are using non-persistent counters so that they don't get too far out
1524 * of sync if we crash or get a forced shutdown. We don't want to force
1525 * this to disk, just get a transaction into the iclogs....
1527 if (flags & SYNC_SUPER)
1528 xfs_log_sbcount(mp, 0);
1531 * Now check to see if the log needs a "dummy" transaction.
1534 if (!(flags & SYNC_REMOUNT) && xfs_log_need_covered(mp)) {
1539 * Put a dummy transaction in the log to tell
1540 * recovery that all others are OK.
1542 tp = xfs_trans_alloc(mp, XFS_TRANS_DUMMY1);
1543 if ((error = xfs_trans_reserve(tp, 0,
1544 XFS_ICHANGE_LOG_RES(mp),
1546 xfs_trans_cancel(tp, 0);
1551 xfs_ilock(ip, XFS_ILOCK_EXCL);
1553 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1554 xfs_trans_ihold(tp, ip);
1555 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1556 error = xfs_trans_commit(tp, 0);
1557 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1558 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1562 * When shutting down, we need to insure that the AIL is pushed
1563 * to disk or the filesystem can appear corrupt from the PROM.
1565 if ((flags & (SYNC_CLOSE|SYNC_WAIT)) == (SYNC_CLOSE|SYNC_WAIT)) {
1566 XFS_bflush(mp->m_ddev_targp);
1567 if (mp->m_rtdev_targp) {
1568 XFS_bflush(mp->m_rtdev_targp);
1572 return XFS_ERROR(last_error);
1576 * xfs_vget - called by DMAPI and NFSD to get vnode from file handle
1584 xfs_mount_t *mp = XFS_BHVTOM(bdp);
1585 xfs_fid_t *xfid = (struct xfs_fid *)fidp;
1592 * Invalid. Since handles can be created in user space and passed in
1593 * via gethandle(), this is not cause for a panic.
1595 if (xfid->xfs_fid_len != sizeof(*xfid) - sizeof(xfid->xfs_fid_len))
1596 return XFS_ERROR(EINVAL);
1598 ino = xfid->xfs_fid_ino;
1599 igen = xfid->xfs_fid_gen;
1602 * NFS can sometimes send requests for ino 0. Fail them gracefully.
1605 return XFS_ERROR(ESTALE);
1607 error = xfs_iget(mp, NULL, ino, 0, XFS_ILOCK_SHARED, &ip, 0);
1615 return XFS_ERROR(EIO);
1618 if (ip->i_d.di_mode == 0 || ip->i_d.di_gen != igen) {
1619 xfs_iput_new(ip, XFS_ILOCK_SHARED);
1621 return XFS_ERROR(ENOENT);
1624 *vpp = XFS_ITOV(ip);
1625 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1630 #define MNTOPT_LOGBUFS "logbufs" /* number of XFS log buffers */
1631 #define MNTOPT_LOGBSIZE "logbsize" /* size of XFS log buffers */
1632 #define MNTOPT_LOGDEV "logdev" /* log device */
1633 #define MNTOPT_RTDEV "rtdev" /* realtime I/O device */
1634 #define MNTOPT_BIOSIZE "biosize" /* log2 of preferred buffered io size */
1635 #define MNTOPT_WSYNC "wsync" /* safe-mode nfs compatible mount */
1636 #define MNTOPT_INO64 "ino64" /* force inodes into 64-bit range */
1637 #define MNTOPT_NOALIGN "noalign" /* turn off stripe alignment */
1638 #define MNTOPT_SWALLOC "swalloc" /* turn on stripe width allocation */
1639 #define MNTOPT_SUNIT "sunit" /* data volume stripe unit */
1640 #define MNTOPT_SWIDTH "swidth" /* data volume stripe width */
1641 #define MNTOPT_NOUUID "nouuid" /* ignore filesystem UUID */
1642 #define MNTOPT_MTPT "mtpt" /* filesystem mount point */
1643 #define MNTOPT_GRPID "grpid" /* group-ID from parent directory */
1644 #define MNTOPT_NOGRPID "nogrpid" /* group-ID from current process */
1645 #define MNTOPT_BSDGROUPS "bsdgroups" /* group-ID from parent directory */
1646 #define MNTOPT_SYSVGROUPS "sysvgroups" /* group-ID from current process */
1647 #define MNTOPT_ALLOCSIZE "allocsize" /* preferred allocation size */
1648 #define MNTOPT_IHASHSIZE "ihashsize" /* size of inode hash table */
1649 #define MNTOPT_NORECOVERY "norecovery" /* don't run XFS recovery */
1650 #define MNTOPT_BARRIER "barrier" /* use writer barriers for log write and
1651 * unwritten extent conversion */
1652 #define MNTOPT_NOBARRIER "nobarrier" /* .. disable */
1653 #define MNTOPT_OSYNCISOSYNC "osyncisosync" /* o_sync is REALLY o_sync */
1654 #define MNTOPT_64BITINODE "inode64" /* inodes can be allocated anywhere */
1655 #define MNTOPT_IKEEP "ikeep" /* do not free empty inode clusters */
1656 #define MNTOPT_NOIKEEP "noikeep" /* free empty inode clusters */
1657 #define MNTOPT_LARGEIO "largeio" /* report large I/O sizes in stat() */
1658 #define MNTOPT_NOLARGEIO "nolargeio" /* do not report large I/O sizes
1660 #define MNTOPT_ATTR2 "attr2" /* do use attr2 attribute format */
1661 #define MNTOPT_NOATTR2 "noattr2" /* do not use attr2 attribute format */
1663 STATIC unsigned long
1664 suffix_strtoul(char *s, char **endp, unsigned int base)
1666 int last, shift_left_factor = 0;
1669 last = strlen(value) - 1;
1670 if (value[last] == 'K' || value[last] == 'k') {
1671 shift_left_factor = 10;
1674 if (value[last] == 'M' || value[last] == 'm') {
1675 shift_left_factor = 20;
1678 if (value[last] == 'G' || value[last] == 'g') {
1679 shift_left_factor = 30;
1683 return simple_strtoul((const char *)s, endp, base) << shift_left_factor;
1688 struct bhv_desc *bhv,
1690 struct xfs_mount_args *args,
1693 bhv_vfs_t *vfsp = bhvtovfs(bhv);
1694 char *this_char, *value, *eov;
1695 int dsunit, dswidth, vol_dsunit, vol_dswidth;
1698 args->flags |= XFSMNT_IDELETE;
1699 args->flags |= XFSMNT_BARRIER;
1700 args->flags2 |= XFSMNT2_COMPAT_IOSIZE;
1705 iosize = dsunit = dswidth = vol_dsunit = vol_dswidth = 0;
1707 while ((this_char = strsep(&options, ",")) != NULL) {
1710 if ((value = strchr(this_char, '=')) != NULL)
1713 if (!strcmp(this_char, MNTOPT_LOGBUFS)) {
1714 if (!value || !*value) {
1716 "XFS: %s option requires an argument",
1720 args->logbufs = simple_strtoul(value, &eov, 10);
1721 } else if (!strcmp(this_char, MNTOPT_LOGBSIZE)) {
1722 if (!value || !*value) {
1724 "XFS: %s option requires an argument",
1728 args->logbufsize = suffix_strtoul(value, &eov, 10);
1729 } else if (!strcmp(this_char, MNTOPT_LOGDEV)) {
1730 if (!value || !*value) {
1732 "XFS: %s option requires an argument",
1736 strncpy(args->logname, value, MAXNAMELEN);
1737 } else if (!strcmp(this_char, MNTOPT_MTPT)) {
1738 if (!value || !*value) {
1740 "XFS: %s option requires an argument",
1744 strncpy(args->mtpt, value, MAXNAMELEN);
1745 } else if (!strcmp(this_char, MNTOPT_RTDEV)) {
1746 if (!value || !*value) {
1748 "XFS: %s option requires an argument",
1752 strncpy(args->rtname, value, MAXNAMELEN);
1753 } else if (!strcmp(this_char, MNTOPT_BIOSIZE)) {
1754 if (!value || !*value) {
1756 "XFS: %s option requires an argument",
1760 iosize = simple_strtoul(value, &eov, 10);
1761 args->flags |= XFSMNT_IOSIZE;
1762 args->iosizelog = (uint8_t) iosize;
1763 } else if (!strcmp(this_char, MNTOPT_ALLOCSIZE)) {
1764 if (!value || !*value) {
1766 "XFS: %s option requires an argument",
1770 iosize = suffix_strtoul(value, &eov, 10);
1771 args->flags |= XFSMNT_IOSIZE;
1772 args->iosizelog = ffs(iosize) - 1;
1773 } else if (!strcmp(this_char, MNTOPT_IHASHSIZE)) {
1774 if (!value || !*value) {
1776 "XFS: %s option requires an argument",
1780 args->flags |= XFSMNT_IHASHSIZE;
1781 args->ihashsize = simple_strtoul(value, &eov, 10);
1782 } else if (!strcmp(this_char, MNTOPT_GRPID) ||
1783 !strcmp(this_char, MNTOPT_BSDGROUPS)) {
1784 vfsp->vfs_flag |= VFS_GRPID;
1785 } else if (!strcmp(this_char, MNTOPT_NOGRPID) ||
1786 !strcmp(this_char, MNTOPT_SYSVGROUPS)) {
1787 vfsp->vfs_flag &= ~VFS_GRPID;
1788 } else if (!strcmp(this_char, MNTOPT_WSYNC)) {
1789 args->flags |= XFSMNT_WSYNC;
1790 } else if (!strcmp(this_char, MNTOPT_OSYNCISOSYNC)) {
1791 args->flags |= XFSMNT_OSYNCISOSYNC;
1792 } else if (!strcmp(this_char, MNTOPT_NORECOVERY)) {
1793 args->flags |= XFSMNT_NORECOVERY;
1794 } else if (!strcmp(this_char, MNTOPT_INO64)) {
1795 args->flags |= XFSMNT_INO64;
1798 "XFS: %s option not allowed on this system",
1802 } else if (!strcmp(this_char, MNTOPT_NOALIGN)) {
1803 args->flags |= XFSMNT_NOALIGN;
1804 } else if (!strcmp(this_char, MNTOPT_SWALLOC)) {
1805 args->flags |= XFSMNT_SWALLOC;
1806 } else if (!strcmp(this_char, MNTOPT_SUNIT)) {
1807 if (!value || !*value) {
1809 "XFS: %s option requires an argument",
1813 dsunit = simple_strtoul(value, &eov, 10);
1814 } else if (!strcmp(this_char, MNTOPT_SWIDTH)) {
1815 if (!value || !*value) {
1817 "XFS: %s option requires an argument",
1821 dswidth = simple_strtoul(value, &eov, 10);
1822 } else if (!strcmp(this_char, MNTOPT_64BITINODE)) {
1823 args->flags &= ~XFSMNT_32BITINODES;
1826 "XFS: %s option not allowed on this system",
1830 } else if (!strcmp(this_char, MNTOPT_NOUUID)) {
1831 args->flags |= XFSMNT_NOUUID;
1832 } else if (!strcmp(this_char, MNTOPT_BARRIER)) {
1833 args->flags |= XFSMNT_BARRIER;
1834 } else if (!strcmp(this_char, MNTOPT_NOBARRIER)) {
1835 args->flags &= ~XFSMNT_BARRIER;
1836 } else if (!strcmp(this_char, MNTOPT_IKEEP)) {
1837 args->flags &= ~XFSMNT_IDELETE;
1838 } else if (!strcmp(this_char, MNTOPT_NOIKEEP)) {
1839 args->flags |= XFSMNT_IDELETE;
1840 } else if (!strcmp(this_char, MNTOPT_LARGEIO)) {
1841 args->flags2 &= ~XFSMNT2_COMPAT_IOSIZE;
1842 } else if (!strcmp(this_char, MNTOPT_NOLARGEIO)) {
1843 args->flags2 |= XFSMNT2_COMPAT_IOSIZE;
1844 } else if (!strcmp(this_char, MNTOPT_ATTR2)) {
1845 args->flags |= XFSMNT_ATTR2;
1846 } else if (!strcmp(this_char, MNTOPT_NOATTR2)) {
1847 args->flags &= ~XFSMNT_ATTR2;
1848 } else if (!strcmp(this_char, "osyncisdsync")) {
1849 /* no-op, this is now the default */
1851 "XFS: osyncisdsync is now the default, option is deprecated.");
1852 } else if (!strcmp(this_char, "irixsgid")) {
1854 "XFS: irixsgid is now a sysctl(2) variable, option is deprecated.");
1857 "XFS: unknown mount option [%s].", this_char);
1862 if (args->flags & XFSMNT_NORECOVERY) {
1863 if ((vfsp->vfs_flag & VFS_RDONLY) == 0) {
1865 "XFS: no-recovery mounts must be read-only.");
1870 if ((args->flags & XFSMNT_NOALIGN) && (dsunit || dswidth)) {
1872 "XFS: sunit and swidth options incompatible with the noalign option");
1876 if ((dsunit && !dswidth) || (!dsunit && dswidth)) {
1878 "XFS: sunit and swidth must be specified together");
1882 if (dsunit && (dswidth % dsunit != 0)) {
1884 "XFS: stripe width (%d) must be a multiple of the stripe unit (%d)",
1889 if ((args->flags & XFSMNT_NOALIGN) != XFSMNT_NOALIGN) {
1891 args->sunit = dsunit;
1892 args->flags |= XFSMNT_RETERR;
1894 args->sunit = vol_dsunit;
1896 dswidth ? (args->swidth = dswidth) :
1897 (args->swidth = vol_dswidth);
1899 args->sunit = args->swidth = 0;
1903 if (args->flags & XFSMNT_32BITINODES)
1904 vfsp->vfs_flag |= VFS_32BITINODES;
1906 args->flags |= XFSMNT_FLAGS2;
1912 struct bhv_desc *bhv,
1915 static struct proc_xfs_info {
1919 /* the few simple ones we can get from the mount struct */
1920 { XFS_MOUNT_WSYNC, "," MNTOPT_WSYNC },
1921 { XFS_MOUNT_INO64, "," MNTOPT_INO64 },
1922 { XFS_MOUNT_NOALIGN, "," MNTOPT_NOALIGN },
1923 { XFS_MOUNT_SWALLOC, "," MNTOPT_SWALLOC },
1924 { XFS_MOUNT_NOUUID, "," MNTOPT_NOUUID },
1925 { XFS_MOUNT_NORECOVERY, "," MNTOPT_NORECOVERY },
1926 { XFS_MOUNT_OSYNCISOSYNC, "," MNTOPT_OSYNCISOSYNC },
1929 struct proc_xfs_info *xfs_infop;
1930 struct xfs_mount *mp = XFS_BHVTOM(bhv);
1931 struct bhv_vfs *vfsp = XFS_MTOVFS(mp);
1933 for (xfs_infop = xfs_info; xfs_infop->flag; xfs_infop++) {
1934 if (mp->m_flags & xfs_infop->flag)
1935 seq_puts(m, xfs_infop->str);
1938 if (mp->m_flags & XFS_MOUNT_IHASHSIZE)
1939 seq_printf(m, "," MNTOPT_IHASHSIZE "=%d", (int)mp->m_ihsize);
1941 if (mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)
1942 seq_printf(m, "," MNTOPT_ALLOCSIZE "=%dk",
1943 (int)(1 << mp->m_writeio_log) >> 10);
1945 if (mp->m_logbufs > 0)
1946 seq_printf(m, "," MNTOPT_LOGBUFS "=%d", mp->m_logbufs);
1947 if (mp->m_logbsize > 0)
1948 seq_printf(m, "," MNTOPT_LOGBSIZE "=%dk", mp->m_logbsize >> 10);
1951 seq_printf(m, "," MNTOPT_LOGDEV "=%s", mp->m_logname);
1953 seq_printf(m, "," MNTOPT_RTDEV "=%s", mp->m_rtname);
1955 if (mp->m_dalign > 0)
1956 seq_printf(m, "," MNTOPT_SUNIT "=%d",
1957 (int)XFS_FSB_TO_BB(mp, mp->m_dalign));
1958 if (mp->m_swidth > 0)
1959 seq_printf(m, "," MNTOPT_SWIDTH "=%d",
1960 (int)XFS_FSB_TO_BB(mp, mp->m_swidth));
1962 if (!(mp->m_flags & XFS_MOUNT_IDELETE))
1963 seq_printf(m, "," MNTOPT_IKEEP);
1964 if (!(mp->m_flags & XFS_MOUNT_COMPAT_IOSIZE))
1965 seq_printf(m, "," MNTOPT_LARGEIO);
1967 if (!(vfsp->vfs_flag & VFS_32BITINODES))
1968 seq_printf(m, "," MNTOPT_64BITINODE);
1969 if (vfsp->vfs_flag & VFS_GRPID)
1970 seq_printf(m, "," MNTOPT_GRPID);
1976 * Second stage of a freeze. The data is already frozen so we only
1977 * need to take care of themetadata. Once that's done write a dummy
1978 * record to dirty the log in case of a crash while frozen.
1984 xfs_mount_t *mp = XFS_BHVTOM(bdp);
1986 xfs_attr_quiesce(mp);
1987 xfs_fs_log_dummy(mp);
1991 bhv_vfsops_t xfs_vfsops = {
1992 BHV_IDENTITY_INIT(VFS_BHV_XFS,VFS_POSITION_XFS),
1993 .vfs_parseargs = xfs_parseargs,
1994 .vfs_showargs = xfs_showargs,
1995 .vfs_mount = xfs_mount,
1996 .vfs_unmount = xfs_unmount,
1997 .vfs_mntupdate = xfs_mntupdate,
1998 .vfs_root = xfs_root,
1999 .vfs_statvfs = xfs_statvfs,
2000 .vfs_sync = xfs_sync,
2001 .vfs_vget = xfs_vget,
2002 .vfs_dmapiops = (vfs_dmapiops_t)fs_nosys,
2003 .vfs_quotactl = (vfs_quotactl_t)fs_nosys,
2004 .vfs_init_vnode = xfs_initialize_vnode,
2005 .vfs_force_shutdown = xfs_do_force_shutdown,
2006 .vfs_freeze = xfs_freeze,