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_REMOUNT|SYNC_ATTR|SYNC_WAIT, NULL);
644 pincount = xfs_flush_buftarg(mp->m_ddev_targp, 1);
658 struct xfs_mount_args *args)
660 bhv_vfs_t *vfsp = bhvtovfs(bdp);
661 xfs_mount_t *mp = XFS_BHVTOM(bdp);
663 if (!(*flags & MS_RDONLY)) { /* rw/ro -> rw */
664 if (vfsp->vfs_flag & VFS_RDONLY)
665 vfsp->vfs_flag &= ~VFS_RDONLY;
666 if (args->flags & XFSMNT_BARRIER) {
667 mp->m_flags |= XFS_MOUNT_BARRIER;
668 xfs_mountfs_check_barriers(mp);
670 mp->m_flags &= ~XFS_MOUNT_BARRIER;
672 } else if (!(vfsp->vfs_flag & VFS_RDONLY)) { /* rw -> ro */
673 bhv_vfs_sync(vfsp, SYNC_FSDATA|SYNC_BDFLUSH|SYNC_ATTR, NULL);
675 xfs_log_unmount_write(mp);
676 xfs_unmountfs_writesb(mp);
677 vfsp->vfs_flag |= VFS_RDONLY;
683 * xfs_unmount_flush implements a set of flush operation on special
684 * inodes, which are needed as a separate set of operations so that
685 * they can be called as part of relocation process.
689 xfs_mount_t *mp, /* Mount structure we are getting
691 int relocation) /* Called from vfs relocation. */
693 xfs_inode_t *rip = mp->m_rootip;
695 xfs_inode_t *rsumip = NULL;
696 bhv_vnode_t *rvp = XFS_ITOV(rip);
699 xfs_ilock(rip, XFS_ILOCK_EXCL);
703 * Flush out the real time inodes.
705 if ((rbmip = mp->m_rbmip) != NULL) {
706 xfs_ilock(rbmip, XFS_ILOCK_EXCL);
708 error = xfs_iflush(rbmip, XFS_IFLUSH_SYNC);
709 xfs_iunlock(rbmip, XFS_ILOCK_EXCL);
711 if (error == EFSCORRUPTED)
714 ASSERT(vn_count(XFS_ITOV(rbmip)) == 1);
716 rsumip = mp->m_rsumip;
717 xfs_ilock(rsumip, XFS_ILOCK_EXCL);
719 error = xfs_iflush(rsumip, XFS_IFLUSH_SYNC);
720 xfs_iunlock(rsumip, XFS_ILOCK_EXCL);
722 if (error == EFSCORRUPTED)
725 ASSERT(vn_count(XFS_ITOV(rsumip)) == 1);
729 * Synchronously flush root inode to disk
731 error = xfs_iflush(rip, XFS_IFLUSH_SYNC);
732 if (error == EFSCORRUPTED)
735 if (vn_count(rvp) != 1 && !relocation) {
736 xfs_iunlock(rip, XFS_ILOCK_EXCL);
737 return XFS_ERROR(EBUSY);
741 * Release dquot that rootinode, rbmino and rsumino might be holding,
742 * flush and purge the quota inodes.
744 error = XFS_QM_UNMOUNT(mp);
745 if (error == EFSCORRUPTED)
749 VN_RELE(XFS_ITOV(rbmip));
750 VN_RELE(XFS_ITOV(rsumip));
753 xfs_iunlock(rip, XFS_ILOCK_EXCL);
760 xfs_iunlock(rip, XFS_ILOCK_EXCL);
762 return XFS_ERROR(EFSCORRUPTED);
766 * xfs_root extracts the root vnode from a vfs.
768 * vfsp -- the vfs struct for the desired file system
769 * vpp -- address of the caller's vnode pointer which should be
770 * set to the desired fs root vnode
779 vp = XFS_ITOV((XFS_BHVTOM(bdp))->m_rootip);
788 * Fill in the statvfs structure for the given file system. We use
789 * the superblock lock in the mount structure to ensure a consistent
790 * snapshot of the counters returned.
795 bhv_statvfs_t *statp,
804 mp = XFS_BHVTOM(bdp);
807 statp->f_type = XFS_SB_MAGIC;
809 xfs_icsb_sync_counters_lazy(mp);
811 statp->f_bsize = sbp->sb_blocksize;
812 lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
813 statp->f_blocks = sbp->sb_dblocks - lsize;
814 statp->f_bfree = statp->f_bavail =
815 sbp->sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
816 fakeinos = statp->f_bfree << sbp->sb_inopblog;
818 fakeinos += mp->m_inoadd;
821 MIN(sbp->sb_icount + fakeinos, (__uint64_t)XFS_MAXINUMBER);
826 statp->f_files = min_t(typeof(statp->f_files),
829 statp->f_ffree = statp->f_files - (sbp->sb_icount - sbp->sb_ifree);
830 XFS_SB_UNLOCK(mp, s);
832 xfs_statvfs_fsid(statp, mp);
833 statp->f_namelen = MAXNAMELEN - 1;
840 * xfs_sync flushes any pending I/O to file system vfsp.
842 * This routine is called by vfs_sync() to make sure that things make it
843 * out to disk eventually, on sync() system calls to flush out everything,
844 * and when the file system is unmounted. For the vfs_sync() case, all
845 * we really need to do is sync out the log to make all of our meta-data
846 * updates permanent (except for timestamps). For calls from pflushd(),
847 * dirty pages are kept moving by calling pdflush() on the inodes
848 * containing them. We also flush the inodes that we can lock without
849 * sleeping and the superblock if we can lock it without sleeping from
850 * vfs_sync() so that items at the tail of the log are always moving out.
853 * SYNC_BDFLUSH - We're being called from vfs_sync() so we don't want
854 * to sleep if we can help it. All we really need
855 * to do is ensure that the log is synced at least
856 * periodically. We also push the inodes and
857 * superblock if we can lock them without sleeping
858 * and they are not pinned.
859 * SYNC_ATTR - We need to flush the inodes. If SYNC_BDFLUSH is not
860 * set, then we really want to lock each inode and flush
862 * SYNC_WAIT - All the flushes that take place in this call should
864 * SYNC_DELWRI - This tells us to push dirty pages associated with
865 * inodes. SYNC_WAIT and SYNC_BDFLUSH are used to
866 * determine if they should be flushed sync, async, or
868 * SYNC_CLOSE - This flag is passed when the system is being
869 * unmounted. We should sync and invalidate everything.
870 * SYNC_FSDATA - This indicates that the caller would like to make
871 * sure the superblock is safe on disk. We can ensure
872 * this by simply making sure the log gets flushed
873 * if SYNC_BDFLUSH is set, and by actually writing it
884 xfs_mount_t *mp = XFS_BHVTOM(bdp);
886 if (unlikely(flags == SYNC_QUIESCE))
887 return xfs_quiesce_fs(mp);
889 return xfs_syncsub(mp, flags, NULL);
893 * xfs sync routine for internal use
895 * This routine supports all of the flags defined for the generic vfs_sync
896 * interface as explained above under xfs_sync.
905 xfs_inode_t *ip = NULL;
906 xfs_inode_t *ip_next;
908 bhv_vnode_t *vp = NULL;
913 uint base_lock_flags;
914 boolean_t mount_locked;
915 boolean_t vnode_refed;
918 xfs_iptr_t *ipointer;
920 boolean_t ipointer_in = B_FALSE;
922 #define IPOINTER_SET ipointer_in = B_TRUE
923 #define IPOINTER_CLR ipointer_in = B_FALSE
930 /* Insert a marker record into the inode list after inode ip. The list
931 * must be locked when this is called. After the call the list will no
934 #define IPOINTER_INSERT(ip, mp) { \
935 ASSERT(ipointer_in == B_FALSE); \
936 ipointer->ip_mnext = ip->i_mnext; \
937 ipointer->ip_mprev = ip; \
938 ip->i_mnext = (xfs_inode_t *)ipointer; \
939 ipointer->ip_mnext->i_mprev = (xfs_inode_t *)ipointer; \
941 XFS_MOUNT_IUNLOCK(mp); \
942 mount_locked = B_FALSE; \
946 /* Remove the marker from the inode list. If the marker was the only item
947 * in the list then there are no remaining inodes and we should zero out
948 * the whole list. If we are the current head of the list then move the head
951 #define IPOINTER_REMOVE(ip, mp) { \
952 ASSERT(ipointer_in == B_TRUE); \
953 if (ipointer->ip_mnext != (xfs_inode_t *)ipointer) { \
954 ip = ipointer->ip_mnext; \
955 ip->i_mprev = ipointer->ip_mprev; \
956 ipointer->ip_mprev->i_mnext = ip; \
957 if (mp->m_inodes == (xfs_inode_t *)ipointer) { \
961 ASSERT(mp->m_inodes == (xfs_inode_t *)ipointer); \
962 mp->m_inodes = NULL; \
968 #define XFS_PREEMPT_MASK 0x7f
972 if (XFS_MTOVFS(mp)->vfs_flag & VFS_RDONLY)
978 /* Allocate a reference marker */
979 ipointer = (xfs_iptr_t *)kmem_zalloc(sizeof(xfs_iptr_t), KM_SLEEP);
981 fflag = XFS_B_ASYNC; /* default is don't wait */
982 if (flags & (SYNC_BDFLUSH | SYNC_DELWRI))
983 fflag = XFS_B_DELWRI;
984 if (flags & SYNC_WAIT)
985 fflag = 0; /* synchronous overrides all */
987 base_lock_flags = XFS_ILOCK_SHARED;
988 if (flags & (SYNC_DELWRI | SYNC_CLOSE)) {
990 * We need the I/O lock if we're going to call any of
991 * the flush/inval routines.
993 base_lock_flags |= XFS_IOLOCK_SHARED;
1000 mount_locked = B_TRUE;
1001 vnode_refed = B_FALSE;
1006 ASSERT(ipointer_in == B_FALSE);
1007 ASSERT(vnode_refed == B_FALSE);
1009 lock_flags = base_lock_flags;
1012 * There were no inodes in the list, just break out
1020 * We found another sync thread marker - skip it
1022 if (ip->i_mount == NULL) {
1027 vp = XFS_ITOV_NULL(ip);
1030 * If the vnode is gone then this is being torn down,
1031 * call reclaim if it is flushed, else let regular flush
1032 * code deal with it later in the loop.
1036 /* Skip ones already in reclaim */
1037 if (ip->i_flags & XFS_IRECLAIM) {
1041 if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0) {
1043 } else if ((xfs_ipincount(ip) == 0) &&
1044 xfs_iflock_nowait(ip)) {
1045 IPOINTER_INSERT(ip, mp);
1047 xfs_finish_reclaim(ip, 1,
1048 XFS_IFLUSH_DELWRI_ELSE_ASYNC);
1050 XFS_MOUNT_ILOCK(mp);
1051 mount_locked = B_TRUE;
1052 IPOINTER_REMOVE(ip, mp);
1054 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1065 if (XFS_FORCED_SHUTDOWN(mp) && !(flags & SYNC_CLOSE)) {
1066 XFS_MOUNT_IUNLOCK(mp);
1067 kmem_free(ipointer, sizeof(xfs_iptr_t));
1072 * If this is just vfs_sync() or pflushd() calling
1073 * then we can skip inodes for which it looks like
1074 * there is nothing to do. Since we don't have the
1075 * inode locked this is racy, but these are periodic
1076 * calls so it doesn't matter. For the others we want
1077 * to know for sure, so we at least try to lock them.
1079 if (flags & SYNC_BDFLUSH) {
1080 if (((ip->i_itemp == NULL) ||
1081 !(ip->i_itemp->ili_format.ilf_fields &
1083 (ip->i_update_core == 0)) {
1090 * Try to lock without sleeping. We're out of order with
1091 * the inode list lock here, so if we fail we need to drop
1092 * the mount lock and try again. If we're called from
1093 * bdflush() here, then don't bother.
1095 * The inode lock here actually coordinates with the
1096 * almost spurious inode lock in xfs_ireclaim() to prevent
1097 * the vnode we handle here without a reference from
1098 * being freed while we reference it. If we lock the inode
1099 * while it's on the mount list here, then the spurious inode
1100 * lock in xfs_ireclaim() after the inode is pulled from
1101 * the mount list will sleep until we release it here.
1102 * This keeps the vnode from being freed while we reference
1105 if (xfs_ilock_nowait(ip, lock_flags) == 0) {
1106 if ((flags & SYNC_BDFLUSH) || (vp == NULL)) {
1117 IPOINTER_INSERT(ip, mp);
1118 xfs_ilock(ip, lock_flags);
1120 ASSERT(vp == XFS_ITOV(ip));
1121 ASSERT(ip->i_mount == mp);
1123 vnode_refed = B_TRUE;
1126 /* From here on in the loop we may have a marker record
1127 * in the inode list.
1130 if ((flags & SYNC_CLOSE) && (vp != NULL)) {
1132 * This is the shutdown case. We just need to
1133 * flush and invalidate all the pages associated
1134 * with the inode. Drop the inode lock since
1135 * we can't hold it across calls to the buffer
1138 * We don't set the VREMAPPING bit in the vnode
1139 * here, because we don't hold the vnode lock
1140 * exclusively. It doesn't really matter, though,
1141 * because we only come here when we're shutting
1144 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1146 if (XFS_FORCED_SHUTDOWN(mp)) {
1147 bhv_vop_toss_pages(vp, 0, -1, FI_REMAPF);
1149 bhv_vop_flushinval_pages(vp, 0, -1, FI_REMAPF);
1152 xfs_ilock(ip, XFS_ILOCK_SHARED);
1154 } else if ((flags & SYNC_DELWRI) && (vp != NULL)) {
1156 /* We need to have dropped the lock here,
1157 * so insert a marker if we have not already
1161 IPOINTER_INSERT(ip, mp);
1165 * Drop the inode lock since we can't hold it
1166 * across calls to the buffer cache.
1168 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1169 error = bhv_vop_flush_pages(vp, (xfs_off_t)0,
1170 -1, fflag, FI_NONE);
1171 xfs_ilock(ip, XFS_ILOCK_SHARED);
1176 if (flags & SYNC_BDFLUSH) {
1177 if ((flags & SYNC_ATTR) &&
1178 ((ip->i_update_core) ||
1179 ((ip->i_itemp != NULL) &&
1180 (ip->i_itemp->ili_format.ilf_fields != 0)))) {
1182 /* Insert marker and drop lock if not already
1186 IPOINTER_INSERT(ip, mp);
1190 * We don't want the periodic flushing of the
1191 * inodes by vfs_sync() to interfere with
1192 * I/O to the file, especially read I/O
1193 * where it is only the access time stamp
1194 * that is being flushed out. To prevent
1195 * long periods where we have both inode
1196 * locks held shared here while reading the
1197 * inode's buffer in from disk, we drop the
1198 * inode lock while reading in the inode
1199 * buffer. We have to release the buffer
1200 * and reacquire the inode lock so that they
1201 * are acquired in the proper order (inode
1202 * locks first). The buffer will go at the
1203 * end of the lru chain, though, so we can
1204 * expect it to still be there when we go
1205 * for it again in xfs_iflush().
1207 if ((xfs_ipincount(ip) == 0) &&
1208 xfs_iflock_nowait(ip)) {
1211 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1213 error = xfs_itobp(mp, NULL, ip,
1218 /* Bailing out, remove the
1219 * marker and free it.
1221 XFS_MOUNT_ILOCK(mp);
1222 IPOINTER_REMOVE(ip, mp);
1223 XFS_MOUNT_IUNLOCK(mp);
1225 ASSERT(!(lock_flags &
1226 XFS_IOLOCK_SHARED));
1229 sizeof(xfs_iptr_t));
1234 * Since we dropped the inode lock,
1235 * the inode may have been reclaimed.
1236 * Therefore, we reacquire the mount
1237 * lock and check to see if we were the
1238 * inode reclaimed. If this happened
1239 * then the ipointer marker will no
1240 * longer point back at us. In this
1241 * case, move ip along to the inode
1242 * after the marker, remove the marker
1245 XFS_MOUNT_ILOCK(mp);
1246 mount_locked = B_TRUE;
1248 if (ip != ipointer->ip_mprev) {
1249 IPOINTER_REMOVE(ip, mp);
1251 ASSERT(!vnode_refed);
1252 ASSERT(!(lock_flags &
1253 XFS_IOLOCK_SHARED));
1257 ASSERT(ip->i_mount == mp);
1259 if (xfs_ilock_nowait(ip,
1260 XFS_ILOCK_SHARED) == 0) {
1261 ASSERT(ip->i_mount == mp);
1263 * We failed to reacquire
1264 * the inode lock without
1265 * sleeping, so just skip
1266 * the inode for now. We
1267 * clear the ILOCK bit from
1268 * the lock_flags so that we
1269 * won't try to drop a lock
1270 * we don't hold below.
1272 lock_flags &= ~XFS_ILOCK_SHARED;
1273 IPOINTER_REMOVE(ip_next, mp);
1274 } else if ((xfs_ipincount(ip) == 0) &&
1275 xfs_iflock_nowait(ip)) {
1276 ASSERT(ip->i_mount == mp);
1278 * Since this is vfs_sync()
1279 * calling we only flush the
1280 * inode out if we can lock
1281 * it without sleeping and
1282 * it is not pinned. Drop
1283 * the mount lock here so
1284 * that we don't hold it for
1285 * too long. We already have
1286 * a marker in the list here.
1288 XFS_MOUNT_IUNLOCK(mp);
1289 mount_locked = B_FALSE;
1290 error = xfs_iflush(ip,
1293 ASSERT(ip->i_mount == mp);
1294 IPOINTER_REMOVE(ip_next, mp);
1301 if ((flags & SYNC_ATTR) &&
1302 ((ip->i_update_core) ||
1303 ((ip->i_itemp != NULL) &&
1304 (ip->i_itemp->ili_format.ilf_fields != 0)))) {
1306 IPOINTER_INSERT(ip, mp);
1309 if (flags & SYNC_WAIT) {
1311 error = xfs_iflush(ip,
1315 * If we can't acquire the flush
1316 * lock, then the inode is already
1317 * being flushed so don't bother
1318 * waiting. If we can lock it then
1319 * do a delwri flush so we can
1320 * combine multiple inode flushes
1321 * in each disk write.
1323 if (xfs_iflock_nowait(ip)) {
1324 error = xfs_iflush(ip,
1333 if (lock_flags != 0) {
1334 xfs_iunlock(ip, lock_flags);
1339 * If we had to take a reference on the vnode
1340 * above, then wait until after we've unlocked
1341 * the inode to release the reference. This is
1342 * because we can be already holding the inode
1343 * lock when VN_RELE() calls xfs_inactive().
1345 * Make sure to drop the mount lock before calling
1346 * VN_RELE() so that we don't trip over ourselves if
1347 * we have to go for the mount lock again in the
1351 IPOINTER_INSERT(ip, mp);
1356 vnode_refed = B_FALSE;
1364 * bail out if the filesystem is corrupted.
1366 if (error == EFSCORRUPTED) {
1367 if (!mount_locked) {
1368 XFS_MOUNT_ILOCK(mp);
1369 IPOINTER_REMOVE(ip, mp);
1371 XFS_MOUNT_IUNLOCK(mp);
1372 ASSERT(ipointer_in == B_FALSE);
1373 kmem_free(ipointer, sizeof(xfs_iptr_t));
1374 return XFS_ERROR(error);
1377 /* Let other threads have a chance at the mount lock
1378 * if we have looped many times without dropping the
1381 if ((++preempt & XFS_PREEMPT_MASK) == 0) {
1383 IPOINTER_INSERT(ip, mp);
1387 if (mount_locked == B_FALSE) {
1388 XFS_MOUNT_ILOCK(mp);
1389 mount_locked = B_TRUE;
1390 IPOINTER_REMOVE(ip, mp);
1394 ASSERT(ipointer_in == B_FALSE);
1397 } while (ip != mp->m_inodes);
1399 XFS_MOUNT_IUNLOCK(mp);
1401 ASSERT(ipointer_in == B_FALSE);
1403 kmem_free(ipointer, sizeof(xfs_iptr_t));
1404 return XFS_ERROR(last_error);
1408 * xfs sync routine for internal use
1410 * This routine supports all of the flags defined for the generic vfs_sync
1411 * interface as explained above under xfs_sync.
1422 uint log_flags = XFS_LOG_FORCE;
1424 xfs_buf_log_item_t *bip;
1427 * Sync out the log. This ensures that the log is periodically
1428 * flushed even if there is not enough activity to fill it up.
1430 if (flags & SYNC_WAIT)
1431 log_flags |= XFS_LOG_SYNC;
1433 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1435 if (flags & (SYNC_ATTR|SYNC_DELWRI)) {
1436 if (flags & SYNC_BDFLUSH)
1437 xfs_finish_reclaim_all(mp, 1);
1439 error = xfs_sync_inodes(mp, flags, bypassed);
1443 * Flushing out dirty data above probably generated more
1444 * log activity, so if this isn't vfs_sync() then flush
1447 if (flags & SYNC_DELWRI) {
1448 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1451 if (flags & SYNC_FSDATA) {
1453 * If this is vfs_sync() then only sync the superblock
1454 * if we can lock it without sleeping and it is not pinned.
1456 if (flags & SYNC_BDFLUSH) {
1457 bp = xfs_getsb(mp, XFS_BUF_TRYLOCK);
1459 bip = XFS_BUF_FSPRIVATE(bp,xfs_buf_log_item_t*);
1460 if ((bip != NULL) &&
1461 xfs_buf_item_dirty(bip)) {
1462 if (!(XFS_BUF_ISPINNED(bp))) {
1464 error = xfs_bwrite(mp, bp);
1473 bp = xfs_getsb(mp, 0);
1475 * If the buffer is pinned then push on the log so
1476 * we won't get stuck waiting in the write for
1477 * someone, maybe ourselves, to flush the log.
1478 * Even though we just pushed the log above, we
1479 * did not have the superblock buffer locked at
1480 * that point so it can become pinned in between
1483 if (XFS_BUF_ISPINNED(bp))
1484 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
1485 if (flags & SYNC_WAIT)
1486 XFS_BUF_UNASYNC(bp);
1489 error = xfs_bwrite(mp, bp);
1497 * If this is the periodic sync, then kick some entries out of
1498 * the reference cache. This ensures that idle entries are
1499 * eventually kicked out of the cache.
1501 if (flags & SYNC_REFCACHE) {
1502 if (flags & SYNC_WAIT)
1503 xfs_refcache_purge_mp(mp);
1505 xfs_refcache_purge_some(mp);
1509 * Now check to see if the log needs a "dummy" transaction.
1512 if (!(flags & SYNC_REMOUNT) && xfs_log_need_covered(mp)) {
1517 * Put a dummy transaction in the log to tell
1518 * recovery that all others are OK.
1520 tp = xfs_trans_alloc(mp, XFS_TRANS_DUMMY1);
1521 if ((error = xfs_trans_reserve(tp, 0,
1522 XFS_ICHANGE_LOG_RES(mp),
1524 xfs_trans_cancel(tp, 0);
1529 xfs_ilock(ip, XFS_ILOCK_EXCL);
1531 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1532 xfs_trans_ihold(tp, ip);
1533 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1534 error = xfs_trans_commit(tp, 0, NULL);
1535 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1536 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1540 * When shutting down, we need to insure that the AIL is pushed
1541 * to disk or the filesystem can appear corrupt from the PROM.
1543 if ((flags & (SYNC_CLOSE|SYNC_WAIT)) == (SYNC_CLOSE|SYNC_WAIT)) {
1544 XFS_bflush(mp->m_ddev_targp);
1545 if (mp->m_rtdev_targp) {
1546 XFS_bflush(mp->m_rtdev_targp);
1550 return XFS_ERROR(last_error);
1554 * xfs_vget - called by DMAPI and NFSD to get vnode from file handle
1562 xfs_mount_t *mp = XFS_BHVTOM(bdp);
1563 xfs_fid_t *xfid = (struct xfs_fid *)fidp;
1570 * Invalid. Since handles can be created in user space and passed in
1571 * via gethandle(), this is not cause for a panic.
1573 if (xfid->xfs_fid_len != sizeof(*xfid) - sizeof(xfid->xfs_fid_len))
1574 return XFS_ERROR(EINVAL);
1576 ino = xfid->xfs_fid_ino;
1577 igen = xfid->xfs_fid_gen;
1580 * NFS can sometimes send requests for ino 0. Fail them gracefully.
1583 return XFS_ERROR(ESTALE);
1585 error = xfs_iget(mp, NULL, ino, 0, XFS_ILOCK_SHARED, &ip, 0);
1593 return XFS_ERROR(EIO);
1596 if (ip->i_d.di_mode == 0 || ip->i_d.di_gen != igen) {
1597 xfs_iput_new(ip, XFS_ILOCK_SHARED);
1599 return XFS_ERROR(ENOENT);
1602 *vpp = XFS_ITOV(ip);
1603 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1608 #define MNTOPT_LOGBUFS "logbufs" /* number of XFS log buffers */
1609 #define MNTOPT_LOGBSIZE "logbsize" /* size of XFS log buffers */
1610 #define MNTOPT_LOGDEV "logdev" /* log device */
1611 #define MNTOPT_RTDEV "rtdev" /* realtime I/O device */
1612 #define MNTOPT_BIOSIZE "biosize" /* log2 of preferred buffered io size */
1613 #define MNTOPT_WSYNC "wsync" /* safe-mode nfs compatible mount */
1614 #define MNTOPT_INO64 "ino64" /* force inodes into 64-bit range */
1615 #define MNTOPT_NOALIGN "noalign" /* turn off stripe alignment */
1616 #define MNTOPT_SWALLOC "swalloc" /* turn on stripe width allocation */
1617 #define MNTOPT_SUNIT "sunit" /* data volume stripe unit */
1618 #define MNTOPT_SWIDTH "swidth" /* data volume stripe width */
1619 #define MNTOPT_NOUUID "nouuid" /* ignore filesystem UUID */
1620 #define MNTOPT_MTPT "mtpt" /* filesystem mount point */
1621 #define MNTOPT_GRPID "grpid" /* group-ID from parent directory */
1622 #define MNTOPT_NOGRPID "nogrpid" /* group-ID from current process */
1623 #define MNTOPT_BSDGROUPS "bsdgroups" /* group-ID from parent directory */
1624 #define MNTOPT_SYSVGROUPS "sysvgroups" /* group-ID from current process */
1625 #define MNTOPT_ALLOCSIZE "allocsize" /* preferred allocation size */
1626 #define MNTOPT_IHASHSIZE "ihashsize" /* size of inode hash table */
1627 #define MNTOPT_NORECOVERY "norecovery" /* don't run XFS recovery */
1628 #define MNTOPT_BARRIER "barrier" /* use writer barriers for log write and
1629 * unwritten extent conversion */
1630 #define MNTOPT_NOBARRIER "nobarrier" /* .. disable */
1631 #define MNTOPT_OSYNCISOSYNC "osyncisosync" /* o_sync is REALLY o_sync */
1632 #define MNTOPT_64BITINODE "inode64" /* inodes can be allocated anywhere */
1633 #define MNTOPT_IKEEP "ikeep" /* do not free empty inode clusters */
1634 #define MNTOPT_NOIKEEP "noikeep" /* free empty inode clusters */
1635 #define MNTOPT_LARGEIO "largeio" /* report large I/O sizes in stat() */
1636 #define MNTOPT_NOLARGEIO "nolargeio" /* do not report large I/O sizes
1638 #define MNTOPT_ATTR2 "attr2" /* do use attr2 attribute format */
1639 #define MNTOPT_NOATTR2 "noattr2" /* do not use attr2 attribute format */
1641 STATIC unsigned long
1642 suffix_strtoul(char *s, char **endp, unsigned int base)
1644 int last, shift_left_factor = 0;
1647 last = strlen(value) - 1;
1648 if (value[last] == 'K' || value[last] == 'k') {
1649 shift_left_factor = 10;
1652 if (value[last] == 'M' || value[last] == 'm') {
1653 shift_left_factor = 20;
1656 if (value[last] == 'G' || value[last] == 'g') {
1657 shift_left_factor = 30;
1661 return simple_strtoul((const char *)s, endp, base) << shift_left_factor;
1666 struct bhv_desc *bhv,
1668 struct xfs_mount_args *args,
1671 bhv_vfs_t *vfsp = bhvtovfs(bhv);
1672 char *this_char, *value, *eov;
1673 int dsunit, dswidth, vol_dsunit, vol_dswidth;
1676 args->flags |= XFSMNT_IDELETE;
1677 args->flags |= XFSMNT_BARRIER;
1678 args->flags2 |= XFSMNT2_COMPAT_IOSIZE;
1683 iosize = dsunit = dswidth = vol_dsunit = vol_dswidth = 0;
1685 while ((this_char = strsep(&options, ",")) != NULL) {
1688 if ((value = strchr(this_char, '=')) != NULL)
1691 if (!strcmp(this_char, MNTOPT_LOGBUFS)) {
1692 if (!value || !*value) {
1694 "XFS: %s option requires an argument",
1698 args->logbufs = simple_strtoul(value, &eov, 10);
1699 } else if (!strcmp(this_char, MNTOPT_LOGBSIZE)) {
1700 if (!value || !*value) {
1702 "XFS: %s option requires an argument",
1706 args->logbufsize = suffix_strtoul(value, &eov, 10);
1707 } else if (!strcmp(this_char, MNTOPT_LOGDEV)) {
1708 if (!value || !*value) {
1710 "XFS: %s option requires an argument",
1714 strncpy(args->logname, value, MAXNAMELEN);
1715 } else if (!strcmp(this_char, MNTOPT_MTPT)) {
1716 if (!value || !*value) {
1718 "XFS: %s option requires an argument",
1722 strncpy(args->mtpt, value, MAXNAMELEN);
1723 } else if (!strcmp(this_char, MNTOPT_RTDEV)) {
1724 if (!value || !*value) {
1726 "XFS: %s option requires an argument",
1730 strncpy(args->rtname, value, MAXNAMELEN);
1731 } else if (!strcmp(this_char, MNTOPT_BIOSIZE)) {
1732 if (!value || !*value) {
1734 "XFS: %s option requires an argument",
1738 iosize = simple_strtoul(value, &eov, 10);
1739 args->flags |= XFSMNT_IOSIZE;
1740 args->iosizelog = (uint8_t) iosize;
1741 } else if (!strcmp(this_char, MNTOPT_ALLOCSIZE)) {
1742 if (!value || !*value) {
1744 "XFS: %s option requires an argument",
1748 iosize = suffix_strtoul(value, &eov, 10);
1749 args->flags |= XFSMNT_IOSIZE;
1750 args->iosizelog = ffs(iosize) - 1;
1751 } else if (!strcmp(this_char, MNTOPT_IHASHSIZE)) {
1752 if (!value || !*value) {
1754 "XFS: %s option requires an argument",
1758 args->flags |= XFSMNT_IHASHSIZE;
1759 args->ihashsize = simple_strtoul(value, &eov, 10);
1760 } else if (!strcmp(this_char, MNTOPT_GRPID) ||
1761 !strcmp(this_char, MNTOPT_BSDGROUPS)) {
1762 vfsp->vfs_flag |= VFS_GRPID;
1763 } else if (!strcmp(this_char, MNTOPT_NOGRPID) ||
1764 !strcmp(this_char, MNTOPT_SYSVGROUPS)) {
1765 vfsp->vfs_flag &= ~VFS_GRPID;
1766 } else if (!strcmp(this_char, MNTOPT_WSYNC)) {
1767 args->flags |= XFSMNT_WSYNC;
1768 } else if (!strcmp(this_char, MNTOPT_OSYNCISOSYNC)) {
1769 args->flags |= XFSMNT_OSYNCISOSYNC;
1770 } else if (!strcmp(this_char, MNTOPT_NORECOVERY)) {
1771 args->flags |= XFSMNT_NORECOVERY;
1772 } else if (!strcmp(this_char, MNTOPT_INO64)) {
1773 args->flags |= XFSMNT_INO64;
1776 "XFS: %s option not allowed on this system",
1780 } else if (!strcmp(this_char, MNTOPT_NOALIGN)) {
1781 args->flags |= XFSMNT_NOALIGN;
1782 } else if (!strcmp(this_char, MNTOPT_SWALLOC)) {
1783 args->flags |= XFSMNT_SWALLOC;
1784 } else if (!strcmp(this_char, MNTOPT_SUNIT)) {
1785 if (!value || !*value) {
1787 "XFS: %s option requires an argument",
1791 dsunit = simple_strtoul(value, &eov, 10);
1792 } else if (!strcmp(this_char, MNTOPT_SWIDTH)) {
1793 if (!value || !*value) {
1795 "XFS: %s option requires an argument",
1799 dswidth = simple_strtoul(value, &eov, 10);
1800 } else if (!strcmp(this_char, MNTOPT_64BITINODE)) {
1801 args->flags &= ~XFSMNT_32BITINODES;
1804 "XFS: %s option not allowed on this system",
1808 } else if (!strcmp(this_char, MNTOPT_NOUUID)) {
1809 args->flags |= XFSMNT_NOUUID;
1810 } else if (!strcmp(this_char, MNTOPT_BARRIER)) {
1811 args->flags |= XFSMNT_BARRIER;
1812 } else if (!strcmp(this_char, MNTOPT_NOBARRIER)) {
1813 args->flags &= ~XFSMNT_BARRIER;
1814 } else if (!strcmp(this_char, MNTOPT_IKEEP)) {
1815 args->flags &= ~XFSMNT_IDELETE;
1816 } else if (!strcmp(this_char, MNTOPT_NOIKEEP)) {
1817 args->flags |= XFSMNT_IDELETE;
1818 } else if (!strcmp(this_char, MNTOPT_LARGEIO)) {
1819 args->flags2 &= ~XFSMNT2_COMPAT_IOSIZE;
1820 } else if (!strcmp(this_char, MNTOPT_NOLARGEIO)) {
1821 args->flags2 |= XFSMNT2_COMPAT_IOSIZE;
1822 } else if (!strcmp(this_char, MNTOPT_ATTR2)) {
1823 args->flags |= XFSMNT_ATTR2;
1824 } else if (!strcmp(this_char, MNTOPT_NOATTR2)) {
1825 args->flags &= ~XFSMNT_ATTR2;
1826 } else if (!strcmp(this_char, "osyncisdsync")) {
1827 /* no-op, this is now the default */
1829 "XFS: osyncisdsync is now the default, option is deprecated.");
1830 } else if (!strcmp(this_char, "irixsgid")) {
1832 "XFS: irixsgid is now a sysctl(2) variable, option is deprecated.");
1835 "XFS: unknown mount option [%s].", this_char);
1840 if (args->flags & XFSMNT_NORECOVERY) {
1841 if ((vfsp->vfs_flag & VFS_RDONLY) == 0) {
1843 "XFS: no-recovery mounts must be read-only.");
1848 if ((args->flags & XFSMNT_NOALIGN) && (dsunit || dswidth)) {
1850 "XFS: sunit and swidth options incompatible with the noalign option");
1854 if ((dsunit && !dswidth) || (!dsunit && dswidth)) {
1856 "XFS: sunit and swidth must be specified together");
1860 if (dsunit && (dswidth % dsunit != 0)) {
1862 "XFS: stripe width (%d) must be a multiple of the stripe unit (%d)",
1867 if ((args->flags & XFSMNT_NOALIGN) != XFSMNT_NOALIGN) {
1869 args->sunit = dsunit;
1870 args->flags |= XFSMNT_RETERR;
1872 args->sunit = vol_dsunit;
1874 dswidth ? (args->swidth = dswidth) :
1875 (args->swidth = vol_dswidth);
1877 args->sunit = args->swidth = 0;
1881 if (args->flags & XFSMNT_32BITINODES)
1882 vfsp->vfs_flag |= VFS_32BITINODES;
1884 args->flags |= XFSMNT_FLAGS2;
1890 struct bhv_desc *bhv,
1893 static struct proc_xfs_info {
1897 /* the few simple ones we can get from the mount struct */
1898 { XFS_MOUNT_WSYNC, "," MNTOPT_WSYNC },
1899 { XFS_MOUNT_INO64, "," MNTOPT_INO64 },
1900 { XFS_MOUNT_NOALIGN, "," MNTOPT_NOALIGN },
1901 { XFS_MOUNT_SWALLOC, "," MNTOPT_SWALLOC },
1902 { XFS_MOUNT_NOUUID, "," MNTOPT_NOUUID },
1903 { XFS_MOUNT_NORECOVERY, "," MNTOPT_NORECOVERY },
1904 { XFS_MOUNT_OSYNCISOSYNC, "," MNTOPT_OSYNCISOSYNC },
1907 struct proc_xfs_info *xfs_infop;
1908 struct xfs_mount *mp = XFS_BHVTOM(bhv);
1909 struct bhv_vfs *vfsp = XFS_MTOVFS(mp);
1911 for (xfs_infop = xfs_info; xfs_infop->flag; xfs_infop++) {
1912 if (mp->m_flags & xfs_infop->flag)
1913 seq_puts(m, xfs_infop->str);
1916 if (mp->m_flags & XFS_MOUNT_IHASHSIZE)
1917 seq_printf(m, "," MNTOPT_IHASHSIZE "=%d", (int)mp->m_ihsize);
1919 if (mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)
1920 seq_printf(m, "," MNTOPT_ALLOCSIZE "=%dk",
1921 (int)(1 << mp->m_writeio_log) >> 10);
1923 if (mp->m_logbufs > 0)
1924 seq_printf(m, "," MNTOPT_LOGBUFS "=%d", mp->m_logbufs);
1925 if (mp->m_logbsize > 0)
1926 seq_printf(m, "," MNTOPT_LOGBSIZE "=%dk", mp->m_logbsize >> 10);
1929 seq_printf(m, "," MNTOPT_LOGDEV "=%s", mp->m_logname);
1931 seq_printf(m, "," MNTOPT_RTDEV "=%s", mp->m_rtname);
1933 if (mp->m_dalign > 0)
1934 seq_printf(m, "," MNTOPT_SUNIT "=%d",
1935 (int)XFS_FSB_TO_BB(mp, mp->m_dalign));
1936 if (mp->m_swidth > 0)
1937 seq_printf(m, "," MNTOPT_SWIDTH "=%d",
1938 (int)XFS_FSB_TO_BB(mp, mp->m_swidth));
1940 if (!(mp->m_flags & XFS_MOUNT_IDELETE))
1941 seq_printf(m, "," MNTOPT_IKEEP);
1942 if (!(mp->m_flags & XFS_MOUNT_COMPAT_IOSIZE))
1943 seq_printf(m, "," MNTOPT_LARGEIO);
1945 if (!(vfsp->vfs_flag & VFS_32BITINODES))
1946 seq_printf(m, "," MNTOPT_64BITINODE);
1947 if (vfsp->vfs_flag & VFS_GRPID)
1948 seq_printf(m, "," MNTOPT_GRPID);
1957 xfs_mount_t *mp = XFS_BHVTOM(bdp);
1959 while (atomic_read(&mp->m_active_trans) > 0)
1962 /* Push the superblock and write an unmount record */
1963 xfs_log_unmount_write(mp);
1964 xfs_unmountfs_writesb(mp);
1965 xfs_fs_log_dummy(mp);
1969 bhv_vfsops_t xfs_vfsops = {
1970 BHV_IDENTITY_INIT(VFS_BHV_XFS,VFS_POSITION_XFS),
1971 .vfs_parseargs = xfs_parseargs,
1972 .vfs_showargs = xfs_showargs,
1973 .vfs_mount = xfs_mount,
1974 .vfs_unmount = xfs_unmount,
1975 .vfs_mntupdate = xfs_mntupdate,
1976 .vfs_root = xfs_root,
1977 .vfs_statvfs = xfs_statvfs,
1978 .vfs_sync = xfs_sync,
1979 .vfs_vget = xfs_vget,
1980 .vfs_dmapiops = (vfs_dmapiops_t)fs_nosys,
1981 .vfs_quotactl = (vfs_quotactl_t)fs_nosys,
1982 .vfs_init_vnode = xfs_initialize_vnode,
1983 .vfs_force_shutdown = xfs_do_force_shutdown,
1984 .vfs_freeze = xfs_freeze,