2 * XFS filesystem operations.
4 * Copyright (c) 2000-2005 Silicon Graphics, Inc. All Rights Reserved.
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms of version 2 of the GNU General Public License as
8 * published by the Free Software Foundation.
10 * This program is distributed in the hope that it would be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
14 * Further, this software is distributed without any warranty that it is
15 * free of the rightful claim of any third person regarding infringement
16 * or the like. Any license provided herein, whether implied or
17 * otherwise, applies only to this software file. Patent licenses, if
18 * any, provided herein do not apply to combinations of this program with
19 * other software, or any other product whatsoever.
21 * You should have received a copy of the GNU General Public License along
22 * with this program; if not, write the Free Software Foundation, Inc., 59
23 * Temple Place - Suite 330, Boston MA 02111-1307, USA.
25 * Contact information: Silicon Graphics, Inc., 1600 Amphitheatre Pkwy,
26 * Mountain View, CA 94043, or:
30 * For further information regarding this notice, see:
32 * http://oss.sgi.com/projects/GenInfo/SGIGPLNoticeExplan/
36 #include "xfs_macros.h"
37 #include "xfs_types.h"
40 #include "xfs_trans.h"
44 #include "xfs_dmapi.h"
45 #include "xfs_mount.h"
46 #include "xfs_bmap_btree.h"
47 #include "xfs_ialloc_btree.h"
48 #include "xfs_alloc_btree.h"
49 #include "xfs_btree.h"
50 #include "xfs_alloc.h"
51 #include "xfs_ialloc.h"
52 #include "xfs_attr_sf.h"
53 #include "xfs_dir_sf.h"
54 #include "xfs_dir2_sf.h"
55 #include "xfs_dinode.h"
56 #include "xfs_inode_item.h"
57 #include "xfs_inode.h"
59 #include "xfs_error.h"
61 #include "xfs_da_btree.h"
63 #include "xfs_refcache.h"
64 #include "xfs_buf_item.h"
65 #include "xfs_extfree_item.h"
66 #include "xfs_quota.h"
67 #include "xfs_dir2_trace.h"
71 #include "xfs_log_priv.h"
73 STATIC int xfs_sync(bhv_desc_t *, int, cred_t *);
78 extern kmem_zone_t *xfs_bmap_free_item_zone;
79 extern kmem_zone_t *xfs_btree_cur_zone;
80 extern kmem_zone_t *xfs_trans_zone;
81 extern kmem_zone_t *xfs_buf_item_zone;
82 extern kmem_zone_t *xfs_dabuf_zone;
83 #ifdef XFS_DABUF_DEBUG
84 extern lock_t xfs_dabuf_global_lock;
85 spinlock_init(&xfs_dabuf_global_lock, "xfsda");
89 * Initialize all of the zone allocators we use.
91 xfs_bmap_free_item_zone = kmem_zone_init(sizeof(xfs_bmap_free_item_t),
92 "xfs_bmap_free_item");
93 xfs_btree_cur_zone = kmem_zone_init(sizeof(xfs_btree_cur_t),
95 xfs_inode_zone = kmem_zone_init(sizeof(xfs_inode_t), "xfs_inode");
96 xfs_trans_zone = kmem_zone_init(sizeof(xfs_trans_t), "xfs_trans");
98 kmem_zone_init(sizeof(xfs_da_state_t), "xfs_da_state");
99 xfs_dabuf_zone = kmem_zone_init(sizeof(xfs_dabuf_t), "xfs_dabuf");
102 * The size of the zone allocated buf log item is the maximum
103 * size possible under XFS. This wastes a little bit of memory,
104 * but it is much faster.
107 kmem_zone_init((sizeof(xfs_buf_log_item_t) +
108 (((XFS_MAX_BLOCKSIZE / XFS_BLI_CHUNK) /
109 NBWORD) * sizeof(int))),
111 xfs_efd_zone = kmem_zone_init((sizeof(xfs_efd_log_item_t) +
112 ((XFS_EFD_MAX_FAST_EXTENTS - 1) * sizeof(xfs_extent_t))),
114 xfs_efi_zone = kmem_zone_init((sizeof(xfs_efi_log_item_t) +
115 ((XFS_EFI_MAX_FAST_EXTENTS - 1) * sizeof(xfs_extent_t))),
117 xfs_ifork_zone = kmem_zone_init(sizeof(xfs_ifork_t), "xfs_ifork");
118 xfs_ili_zone = kmem_zone_init(sizeof(xfs_inode_log_item_t), "xfs_ili");
119 xfs_chashlist_zone = kmem_zone_init(sizeof(xfs_chashlist_t),
121 xfs_acl_zone_init(xfs_acl_zone, "xfs_acl");
124 * Allocate global trace buffers.
126 #ifdef XFS_ALLOC_TRACE
127 xfs_alloc_trace_buf = ktrace_alloc(XFS_ALLOC_TRACE_SIZE, KM_SLEEP);
129 #ifdef XFS_BMAP_TRACE
130 xfs_bmap_trace_buf = ktrace_alloc(XFS_BMAP_TRACE_SIZE, KM_SLEEP);
132 #ifdef XFS_BMBT_TRACE
133 xfs_bmbt_trace_buf = ktrace_alloc(XFS_BMBT_TRACE_SIZE, KM_SLEEP);
136 xfs_dir_trace_buf = ktrace_alloc(XFS_DIR_TRACE_SIZE, KM_SLEEP);
138 #ifdef XFS_ATTR_TRACE
139 xfs_attr_trace_buf = ktrace_alloc(XFS_ATTR_TRACE_SIZE, KM_SLEEP);
141 #ifdef XFS_DIR2_TRACE
142 xfs_dir2_trace_buf = ktrace_alloc(XFS_DIR2_GTRACE_SIZE, KM_SLEEP);
147 #if (defined(DEBUG) || defined(INDUCE_IO_ERROR))
148 xfs_error_test_init();
149 #endif /* DEBUG || INDUCE_IO_ERROR */
152 xfs_sysctl_register();
159 extern kmem_zone_t *xfs_bmap_free_item_zone;
160 extern kmem_zone_t *xfs_btree_cur_zone;
161 extern kmem_zone_t *xfs_inode_zone;
162 extern kmem_zone_t *xfs_trans_zone;
163 extern kmem_zone_t *xfs_da_state_zone;
164 extern kmem_zone_t *xfs_dabuf_zone;
165 extern kmem_zone_t *xfs_efd_zone;
166 extern kmem_zone_t *xfs_efi_zone;
167 extern kmem_zone_t *xfs_buf_item_zone;
168 extern kmem_zone_t *xfs_chashlist_zone;
170 xfs_cleanup_procfs();
171 xfs_sysctl_unregister();
172 xfs_refcache_destroy();
173 xfs_acl_zone_destroy(xfs_acl_zone);
175 #ifdef XFS_DIR2_TRACE
176 ktrace_free(xfs_dir2_trace_buf);
178 #ifdef XFS_ATTR_TRACE
179 ktrace_free(xfs_attr_trace_buf);
182 ktrace_free(xfs_dir_trace_buf);
184 #ifdef XFS_BMBT_TRACE
185 ktrace_free(xfs_bmbt_trace_buf);
187 #ifdef XFS_BMAP_TRACE
188 ktrace_free(xfs_bmap_trace_buf);
190 #ifdef XFS_ALLOC_TRACE
191 ktrace_free(xfs_alloc_trace_buf);
194 kmem_cache_destroy(xfs_bmap_free_item_zone);
195 kmem_cache_destroy(xfs_btree_cur_zone);
196 kmem_cache_destroy(xfs_inode_zone);
197 kmem_cache_destroy(xfs_trans_zone);
198 kmem_cache_destroy(xfs_da_state_zone);
199 kmem_cache_destroy(xfs_dabuf_zone);
200 kmem_cache_destroy(xfs_buf_item_zone);
201 kmem_cache_destroy(xfs_efd_zone);
202 kmem_cache_destroy(xfs_efi_zone);
203 kmem_cache_destroy(xfs_ifork_zone);
204 kmem_cache_destroy(xfs_ili_zone);
205 kmem_cache_destroy(xfs_chashlist_zone);
211 * This function fills in xfs_mount_t fields based on mount args.
212 * Note: the superblock has _not_ yet been read in.
217 struct xfs_mount_args *ap,
218 struct xfs_mount *mp)
220 /* Values are in BBs */
221 if ((ap->flags & XFSMNT_NOALIGN) != XFSMNT_NOALIGN) {
223 * At this point the superblock has not been read
224 * in, therefore we do not know the block size.
225 * Before the mount call ends we will convert
228 mp->m_dalign = ap->sunit;
229 mp->m_swidth = ap->swidth;
232 if (ap->logbufs != -1 &&
233 #if defined(DEBUG) || defined(XLOG_NOLOG)
236 (ap->logbufs < XLOG_MIN_ICLOGS ||
237 ap->logbufs > XLOG_MAX_ICLOGS)) {
239 "XFS: invalid logbufs value: %d [not %d-%d]",
240 ap->logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
241 return XFS_ERROR(EINVAL);
243 mp->m_logbufs = ap->logbufs;
244 if (ap->logbufsize != -1 &&
245 ap->logbufsize != 16 * 1024 &&
246 ap->logbufsize != 32 * 1024 &&
247 ap->logbufsize != 64 * 1024 &&
248 ap->logbufsize != 128 * 1024 &&
249 ap->logbufsize != 256 * 1024) {
251 "XFS: invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
253 return XFS_ERROR(EINVAL);
255 mp->m_ihsize = ap->ihashsize;
256 mp->m_logbsize = ap->logbufsize;
257 mp->m_fsname_len = strlen(ap->fsname) + 1;
258 mp->m_fsname = kmem_alloc(mp->m_fsname_len, KM_SLEEP);
259 strcpy(mp->m_fsname, ap->fsname);
261 if (ap->flags & XFSMNT_WSYNC)
262 mp->m_flags |= XFS_MOUNT_WSYNC;
264 if (ap->flags & XFSMNT_INO64) {
265 mp->m_flags |= XFS_MOUNT_INO64;
266 mp->m_inoadd = XFS_INO64_OFFSET;
269 if (ap->flags & XFSMNT_NOATIME)
270 mp->m_flags |= XFS_MOUNT_NOATIME;
271 if (ap->flags & XFSMNT_RETERR)
272 mp->m_flags |= XFS_MOUNT_RETERR;
273 if (ap->flags & XFSMNT_NOALIGN)
274 mp->m_flags |= XFS_MOUNT_NOALIGN;
275 if (ap->flags & XFSMNT_SWALLOC)
276 mp->m_flags |= XFS_MOUNT_SWALLOC;
277 if (ap->flags & XFSMNT_OSYNCISOSYNC)
278 mp->m_flags |= XFS_MOUNT_OSYNCISOSYNC;
279 if (ap->flags & XFSMNT_32BITINODES)
280 mp->m_flags |= (XFS_MOUNT_32BITINODES | XFS_MOUNT_32BITINOOPT);
282 if (ap->flags & XFSMNT_IOSIZE) {
283 if (ap->iosizelog > XFS_MAX_IO_LOG ||
284 ap->iosizelog < XFS_MIN_IO_LOG) {
286 "XFS: invalid log iosize: %d [not %d-%d]",
287 ap->iosizelog, XFS_MIN_IO_LOG,
289 return XFS_ERROR(EINVAL);
292 mp->m_flags |= XFS_MOUNT_DFLT_IOSIZE;
293 mp->m_readio_log = mp->m_writeio_log = ap->iosizelog;
296 if (ap->flags & XFSMNT_IHASHSIZE)
297 mp->m_flags |= XFS_MOUNT_IHASHSIZE;
298 if (ap->flags & XFSMNT_IDELETE)
299 mp->m_flags |= XFS_MOUNT_IDELETE;
300 if (ap->flags & XFSMNT_DIRSYNC)
301 mp->m_flags |= XFS_MOUNT_DIRSYNC;
302 if (ap->flags & XFSMNT_COMPAT_IOSIZE)
303 mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE;
304 if (ap->flags & XFSMNT_COMPAT_ATTR)
305 mp->m_flags |= XFS_MOUNT_COMPAT_ATTR;
308 * no recovery flag requires a read-only mount
310 if (ap->flags & XFSMNT_NORECOVERY) {
311 if (!(vfs->vfs_flag & VFS_RDONLY)) {
313 "XFS: tried to mount a FS read-write without recovery!");
314 return XFS_ERROR(EINVAL);
316 mp->m_flags |= XFS_MOUNT_NORECOVERY;
319 if (ap->flags & XFSMNT_NOUUID)
320 mp->m_flags |= XFS_MOUNT_NOUUID;
321 if (ap->flags & XFSMNT_BARRIER)
322 mp->m_flags |= XFS_MOUNT_BARRIER;
328 * This function fills in xfs_mount_t fields based on mount args.
329 * Note: the superblock _has_ now been read in.
334 struct xfs_mount_args *ap,
335 struct xfs_mount *mp)
337 int ronly = (vfs->vfs_flag & VFS_RDONLY);
339 /* Fail a mount where the logbuf is smaller then the log stripe */
340 if (XFS_SB_VERSION_HASLOGV2(&mp->m_sb)) {
341 if ((ap->logbufsize == -1) &&
342 (mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE)) {
343 mp->m_logbsize = mp->m_sb.sb_logsunit;
344 } else if (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);
359 * prohibit r/w mounts of read-only filesystems
361 if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !ronly) {
363 "XFS: cannot mount a read-only filesystem as read-write");
364 return XFS_ERROR(EROFS);
368 * check for shared mount.
370 if (ap->flags & XFSMNT_SHARED) {
371 if (!XFS_SB_VERSION_HASSHARED(&mp->m_sb))
372 return XFS_ERROR(EINVAL);
375 * For IRIX 6.5, shared mounts must have the shared
376 * version bit set, have the persistent readonly
377 * field set, must be version 0 and can only be mounted
380 if (!ronly || !(mp->m_sb.sb_flags & XFS_SBF_READONLY) ||
381 (mp->m_sb.sb_shared_vn != 0))
382 return XFS_ERROR(EINVAL);
384 mp->m_flags |= XFS_MOUNT_SHARED;
387 * Shared XFS V0 can't deal with DMI. Return EINVAL.
389 if (mp->m_sb.sb_shared_vn == 0 && (ap->flags & XFSMNT_DMAPI))
390 return XFS_ERROR(EINVAL);
399 * The file system configurations are:
400 * (1) device (partition) with data and internal log
401 * (2) logical volume with data and log subvolumes.
402 * (3) logical volume with data, log, and realtime subvolumes.
404 * We only have to handle opening the log and realtime volumes here if
405 * they are present. The data subvolume has already been opened by
406 * get_sb_bdev() and is stored in vfsp->vfs_super->s_bdev.
410 struct bhv_desc *bhvp,
411 struct xfs_mount_args *args,
414 struct vfs *vfsp = bhvtovfs(bhvp);
416 struct xfs_mount *mp = XFS_BHVTOM(bhvp);
417 struct block_device *ddev, *logdev, *rtdev;
418 int flags = 0, error;
420 ddev = vfsp->vfs_super->s_bdev;
421 logdev = rtdev = NULL;
424 * Setup xfs_mount function vectors from available behaviors
426 p = vfs_bhv_lookup(vfsp, VFS_POSITION_DM);
427 mp->m_dm_ops = p ? *(xfs_dmops_t *) vfs_bhv_custom(p) : xfs_dmcore_stub;
428 p = vfs_bhv_lookup(vfsp, VFS_POSITION_QM);
429 mp->m_qm_ops = p ? *(xfs_qmops_t *) vfs_bhv_custom(p) : xfs_qmcore_stub;
430 p = vfs_bhv_lookup(vfsp, VFS_POSITION_IO);
431 mp->m_io_ops = p ? *(xfs_ioops_t *) vfs_bhv_custom(p) : xfs_iocore_xfs;
434 * Open real time and log devices - order is important.
436 if (args->logname[0]) {
437 error = xfs_blkdev_get(mp, args->logname, &logdev);
441 if (args->rtname[0]) {
442 error = xfs_blkdev_get(mp, args->rtname, &rtdev);
444 xfs_blkdev_put(logdev);
448 if (rtdev == ddev || rtdev == logdev) {
450 "XFS: Cannot mount filesystem with identical rtdev and ddev/logdev.");
451 xfs_blkdev_put(logdev);
452 xfs_blkdev_put(rtdev);
458 * Setup xfs_mount buffer target pointers
461 mp->m_ddev_targp = xfs_alloc_buftarg(ddev, 0);
462 if (!mp->m_ddev_targp) {
463 xfs_blkdev_put(logdev);
464 xfs_blkdev_put(rtdev);
468 mp->m_rtdev_targp = xfs_alloc_buftarg(rtdev, 1);
469 if (!mp->m_rtdev_targp)
472 mp->m_logdev_targp = (logdev && logdev != ddev) ?
473 xfs_alloc_buftarg(logdev, 1) : mp->m_ddev_targp;
474 if (!mp->m_logdev_targp)
478 * Setup flags based on mount(2) options and then the superblock
480 error = xfs_start_flags(vfsp, args, mp);
483 error = xfs_readsb(mp);
486 error = xfs_finish_flags(vfsp, args, mp);
491 * Setup xfs_mount buffer target pointers based on superblock
493 error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_blocksize,
494 mp->m_sb.sb_sectsize);
495 if (!error && logdev && logdev != ddev) {
496 unsigned int log_sector_size = BBSIZE;
498 if (XFS_SB_VERSION_HASSECTOR(&mp->m_sb))
499 log_sector_size = mp->m_sb.sb_logsectsize;
500 error = xfs_setsize_buftarg(mp->m_logdev_targp,
501 mp->m_sb.sb_blocksize,
505 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
506 mp->m_sb.sb_blocksize,
507 mp->m_sb.sb_sectsize);
511 error = XFS_IOINIT(vfsp, args, flags);
515 if ((args->flags & XFSMNT_BARRIER) &&
516 !(XFS_MTOVFS(mp)->vfs_flag & VFS_RDONLY))
517 xfs_mountfs_check_barriers(mp);
524 xfs_binval(mp->m_ddev_targp);
525 if (logdev && logdev != ddev)
526 xfs_binval(mp->m_logdev_targp);
528 xfs_binval(mp->m_rtdev_targp);
530 xfs_unmountfs_close(mp, credp);
540 struct vfs *vfsp = bhvtovfs(bdp);
541 xfs_mount_t *mp = XFS_BHVTOM(bdp);
544 int unmount_event_wanted = 0;
545 int unmount_event_flags = 0;
546 int xfs_unmountfs_needed = 0;
552 if (vfsp->vfs_flag & VFS_DMI) {
553 error = XFS_SEND_PREUNMOUNT(mp, vfsp,
554 rvp, DM_RIGHT_NULL, rvp, DM_RIGHT_NULL,
556 (mp->m_dmevmask & (1<<DM_EVENT_PREUNMOUNT))?
557 0:DM_FLAGS_UNWANTED);
559 return XFS_ERROR(error);
560 unmount_event_wanted = 1;
561 unmount_event_flags = (mp->m_dmevmask & (1<<DM_EVENT_UNMOUNT))?
562 0 : DM_FLAGS_UNWANTED;
566 * First blow any referenced inode from this file system
567 * out of the reference cache, and delete the timer.
569 xfs_refcache_purge_mp(mp);
571 XFS_bflush(mp->m_ddev_targp);
572 error = xfs_unmount_flush(mp, 0);
576 ASSERT(vn_count(rvp) == 1);
579 * Drop the reference count
584 * If we're forcing a shutdown, typically because of a media error,
585 * we want to make sure we invalidate dirty pages that belong to
586 * referenced vnodes as well.
588 if (XFS_FORCED_SHUTDOWN(mp)) {
589 error = xfs_sync(&mp->m_bhv,
590 (SYNC_WAIT | SYNC_CLOSE), credp);
591 ASSERT(error != EFSCORRUPTED);
593 xfs_unmountfs_needed = 1;
596 /* Send DMAPI event, if required.
597 * Then do xfs_unmountfs() if needed.
598 * Then return error (or zero).
600 if (unmount_event_wanted) {
601 /* Note: mp structure must still exist for
602 * XFS_SEND_UNMOUNT() call.
604 XFS_SEND_UNMOUNT(mp, vfsp, error == 0 ? rvp : NULL,
605 DM_RIGHT_NULL, 0, error, unmount_event_flags);
607 if (xfs_unmountfs_needed) {
609 * Call common unmount function to flush to disk
610 * and free the super block buffer & mount structures.
612 xfs_unmountfs(mp, credp);
615 return XFS_ERROR(error);
622 int count = 0, pincount;
624 xfs_refcache_purge_mp(mp);
625 xfs_flush_buftarg(mp->m_ddev_targp, 0);
626 xfs_finish_reclaim_all(mp, 0);
628 /* This loop must run at least twice.
629 * The first instance of the loop will flush
630 * most meta data but that will generate more
631 * meta data (typically directory updates).
632 * Which then must be flushed and logged before
633 * we can write the unmount record.
636 xfs_syncsub(mp, SYNC_REMOUNT|SYNC_ATTR|SYNC_WAIT, 0, NULL);
637 pincount = xfs_flush_buftarg(mp->m_ddev_targp, 1);
651 struct xfs_mount_args *args)
653 struct vfs *vfsp = bhvtovfs(bdp);
654 xfs_mount_t *mp = XFS_BHVTOM(bdp);
657 if (args->flags & XFSMNT_NOATIME)
658 mp->m_flags |= XFS_MOUNT_NOATIME;
660 mp->m_flags &= ~XFS_MOUNT_NOATIME;
662 if ((vfsp->vfs_flag & VFS_RDONLY) &&
663 !(*flags & MS_RDONLY)) {
664 vfsp->vfs_flag &= ~VFS_RDONLY;
666 if (args->flags & XFSMNT_BARRIER)
667 xfs_mountfs_check_barriers(mp);
670 if (!(vfsp->vfs_flag & VFS_RDONLY) &&
671 (*flags & MS_RDONLY)) {
672 VFS_SYNC(vfsp, SYNC_FSDATA|SYNC_BDFLUSH|SYNC_ATTR, NULL, error);
676 /* Ok now write out an unmount record */
677 xfs_log_unmount_write(mp);
678 xfs_unmountfs_writesb(mp);
679 vfsp->vfs_flag |= VFS_RDONLY;
686 * xfs_unmount_flush implements a set of flush operation on special
687 * inodes, which are needed as a separate set of operations so that
688 * they can be called as part of relocation process.
692 xfs_mount_t *mp, /* Mount structure we are getting
694 int relocation) /* Called from vfs relocation. */
696 xfs_inode_t *rip = mp->m_rootip;
698 xfs_inode_t *rsumip = NULL;
699 vnode_t *rvp = XFS_ITOV(rip);
702 xfs_ilock(rip, XFS_ILOCK_EXCL);
706 * Flush out the real time inodes.
708 if ((rbmip = mp->m_rbmip) != NULL) {
709 xfs_ilock(rbmip, XFS_ILOCK_EXCL);
711 error = xfs_iflush(rbmip, XFS_IFLUSH_SYNC);
712 xfs_iunlock(rbmip, XFS_ILOCK_EXCL);
714 if (error == EFSCORRUPTED)
717 ASSERT(vn_count(XFS_ITOV(rbmip)) == 1);
719 rsumip = mp->m_rsumip;
720 xfs_ilock(rsumip, XFS_ILOCK_EXCL);
722 error = xfs_iflush(rsumip, XFS_IFLUSH_SYNC);
723 xfs_iunlock(rsumip, XFS_ILOCK_EXCL);
725 if (error == EFSCORRUPTED)
728 ASSERT(vn_count(XFS_ITOV(rsumip)) == 1);
732 * Synchronously flush root inode to disk
734 error = xfs_iflush(rip, XFS_IFLUSH_SYNC);
735 if (error == EFSCORRUPTED)
738 if (vn_count(rvp) != 1 && !relocation) {
739 xfs_iunlock(rip, XFS_ILOCK_EXCL);
740 return XFS_ERROR(EBUSY);
744 * Release dquot that rootinode, rbmino and rsumino might be holding,
745 * flush and purge the quota inodes.
747 error = XFS_QM_UNMOUNT(mp);
748 if (error == EFSCORRUPTED)
752 VN_RELE(XFS_ITOV(rbmip));
753 VN_RELE(XFS_ITOV(rsumip));
756 xfs_iunlock(rip, XFS_ILOCK_EXCL);
763 xfs_iunlock(rip, XFS_ILOCK_EXCL);
765 return XFS_ERROR(EFSCORRUPTED);
769 * xfs_root extracts the root vnode from a vfs.
771 * vfsp -- the vfs struct for the desired file system
772 * vpp -- address of the caller's vnode pointer which should be
773 * set to the desired fs root vnode
782 vp = XFS_ITOV((XFS_BHVTOM(bdp))->m_rootip);
791 * Fill in the statvfs structure for the given file system. We use
792 * the superblock lock in the mount structure to ensure a consistent
793 * snapshot of the counters returned.
807 mp = XFS_BHVTOM(bdp);
810 statp->f_type = XFS_SB_MAGIC;
813 statp->f_bsize = sbp->sb_blocksize;
814 lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
815 statp->f_blocks = sbp->sb_dblocks - lsize;
816 statp->f_bfree = statp->f_bavail = sbp->sb_fdblocks;
817 fakeinos = statp->f_bfree << sbp->sb_inopblog;
819 fakeinos += mp->m_inoadd;
822 MIN(sbp->sb_icount + fakeinos, (__uint64_t)XFS_MAXINUMBER);
827 statp->f_files = min_t(typeof(statp->f_files),
830 statp->f_ffree = statp->f_files - (sbp->sb_icount - sbp->sb_ifree);
831 XFS_SB_UNLOCK(mp, s);
833 xfs_statvfs_fsid(statp, mp);
834 statp->f_namelen = MAXNAMELEN - 1;
841 * xfs_sync flushes any pending I/O to file system vfsp.
843 * This routine is called by vfs_sync() to make sure that things make it
844 * out to disk eventually, on sync() system calls to flush out everything,
845 * and when the file system is unmounted. For the vfs_sync() case, all
846 * we really need to do is sync out the log to make all of our meta-data
847 * updates permanent (except for timestamps). For calls from pflushd(),
848 * dirty pages are kept moving by calling pdflush() on the inodes
849 * containing them. We also flush the inodes that we can lock without
850 * sleeping and the superblock if we can lock it without sleeping from
851 * vfs_sync() so that items at the tail of the log are always moving out.
854 * SYNC_BDFLUSH - We're being called from vfs_sync() so we don't want
855 * to sleep if we can help it. All we really need
856 * to do is ensure that the log is synced at least
857 * periodically. We also push the inodes and
858 * superblock if we can lock them without sleeping
859 * and they are not pinned.
860 * SYNC_ATTR - We need to flush the inodes. If SYNC_BDFLUSH is not
861 * set, then we really want to lock each inode and flush
863 * SYNC_WAIT - All the flushes that take place in this call should
865 * SYNC_DELWRI - This tells us to push dirty pages associated with
866 * inodes. SYNC_WAIT and SYNC_BDFLUSH are used to
867 * determine if they should be flushed sync, async, or
869 * SYNC_CLOSE - This flag is passed when the system is being
870 * unmounted. We should sync and invalidate everthing.
871 * SYNC_FSDATA - This indicates that the caller would like to make
872 * sure the superblock is safe on disk. We can ensure
873 * this by simply makeing sure the log gets flushed
874 * if SYNC_BDFLUSH is set, and by actually writing it
885 xfs_mount_t *mp = XFS_BHVTOM(bdp);
887 if (unlikely(flags == SYNC_QUIESCE))
888 return xfs_quiesce_fs(mp);
890 return xfs_syncsub(mp, flags, 0, NULL);
894 * xfs sync routine for internal use
896 * This routine supports all of the flags defined for the generic VFS_SYNC
897 * interface as explained above under xfs_sync. In the interests of not
898 * changing interfaces within the 6.5 family, additional internallly-
899 * required functions are specified within a separate xflags parameter,
900 * only available by calling this routine.
910 xfs_inode_t *ip = NULL;
911 xfs_inode_t *ip_next;
918 uint base_lock_flags;
919 boolean_t mount_locked;
920 boolean_t vnode_refed;
923 xfs_iptr_t *ipointer;
925 boolean_t ipointer_in = B_FALSE;
927 #define IPOINTER_SET ipointer_in = B_TRUE
928 #define IPOINTER_CLR ipointer_in = B_FALSE
935 /* Insert a marker record into the inode list after inode ip. The list
936 * must be locked when this is called. After the call the list will no
939 #define IPOINTER_INSERT(ip, mp) { \
940 ASSERT(ipointer_in == B_FALSE); \
941 ipointer->ip_mnext = ip->i_mnext; \
942 ipointer->ip_mprev = ip; \
943 ip->i_mnext = (xfs_inode_t *)ipointer; \
944 ipointer->ip_mnext->i_mprev = (xfs_inode_t *)ipointer; \
946 XFS_MOUNT_IUNLOCK(mp); \
947 mount_locked = B_FALSE; \
951 /* Remove the marker from the inode list. If the marker was the only item
952 * in the list then there are no remaining inodes and we should zero out
953 * the whole list. If we are the current head of the list then move the head
956 #define IPOINTER_REMOVE(ip, mp) { \
957 ASSERT(ipointer_in == B_TRUE); \
958 if (ipointer->ip_mnext != (xfs_inode_t *)ipointer) { \
959 ip = ipointer->ip_mnext; \
960 ip->i_mprev = ipointer->ip_mprev; \
961 ipointer->ip_mprev->i_mnext = ip; \
962 if (mp->m_inodes == (xfs_inode_t *)ipointer) { \
966 ASSERT(mp->m_inodes == (xfs_inode_t *)ipointer); \
967 mp->m_inodes = NULL; \
973 #define XFS_PREEMPT_MASK 0x7f
977 if (XFS_MTOVFS(mp)->vfs_flag & VFS_RDONLY)
983 /* Allocate a reference marker */
984 ipointer = (xfs_iptr_t *)kmem_zalloc(sizeof(xfs_iptr_t), KM_SLEEP);
986 fflag = XFS_B_ASYNC; /* default is don't wait */
987 if (flags & (SYNC_BDFLUSH | SYNC_DELWRI))
988 fflag = XFS_B_DELWRI;
989 if (flags & SYNC_WAIT)
990 fflag = 0; /* synchronous overrides all */
992 base_lock_flags = XFS_ILOCK_SHARED;
993 if (flags & (SYNC_DELWRI | SYNC_CLOSE)) {
995 * We need the I/O lock if we're going to call any of
996 * the flush/inval routines.
998 base_lock_flags |= XFS_IOLOCK_SHARED;
1001 XFS_MOUNT_ILOCK(mp);
1005 mount_locked = B_TRUE;
1006 vnode_refed = B_FALSE;
1011 ASSERT(ipointer_in == B_FALSE);
1012 ASSERT(vnode_refed == B_FALSE);
1014 lock_flags = base_lock_flags;
1017 * There were no inodes in the list, just break out
1025 * We found another sync thread marker - skip it
1027 if (ip->i_mount == NULL) {
1032 vp = XFS_ITOV_NULL(ip);
1035 * If the vnode is gone then this is being torn down,
1036 * call reclaim if it is flushed, else let regular flush
1037 * code deal with it later in the loop.
1041 /* Skip ones already in reclaim */
1042 if (ip->i_flags & XFS_IRECLAIM) {
1046 if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0) {
1048 } else if ((xfs_ipincount(ip) == 0) &&
1049 xfs_iflock_nowait(ip)) {
1050 IPOINTER_INSERT(ip, mp);
1052 xfs_finish_reclaim(ip, 1,
1053 XFS_IFLUSH_DELWRI_ELSE_ASYNC);
1055 XFS_MOUNT_ILOCK(mp);
1056 mount_locked = B_TRUE;
1057 IPOINTER_REMOVE(ip, mp);
1059 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1070 if (XFS_FORCED_SHUTDOWN(mp) && !(flags & SYNC_CLOSE)) {
1071 XFS_MOUNT_IUNLOCK(mp);
1072 kmem_free(ipointer, sizeof(xfs_iptr_t));
1077 * If this is just vfs_sync() or pflushd() calling
1078 * then we can skip inodes for which it looks like
1079 * there is nothing to do. Since we don't have the
1080 * inode locked this is racey, but these are periodic
1081 * calls so it doesn't matter. For the others we want
1082 * to know for sure, so we at least try to lock them.
1084 if (flags & SYNC_BDFLUSH) {
1085 if (((ip->i_itemp == NULL) ||
1086 !(ip->i_itemp->ili_format.ilf_fields &
1088 (ip->i_update_core == 0)) {
1095 * Try to lock without sleeping. We're out of order with
1096 * the inode list lock here, so if we fail we need to drop
1097 * the mount lock and try again. If we're called from
1098 * bdflush() here, then don't bother.
1100 * The inode lock here actually coordinates with the
1101 * almost spurious inode lock in xfs_ireclaim() to prevent
1102 * the vnode we handle here without a reference from
1103 * being freed while we reference it. If we lock the inode
1104 * while it's on the mount list here, then the spurious inode
1105 * lock in xfs_ireclaim() after the inode is pulled from
1106 * the mount list will sleep until we release it here.
1107 * This keeps the vnode from being freed while we reference
1110 if (xfs_ilock_nowait(ip, lock_flags) == 0) {
1111 if ((flags & SYNC_BDFLUSH) || (vp == NULL)) {
1122 IPOINTER_INSERT(ip, mp);
1123 xfs_ilock(ip, lock_flags);
1125 ASSERT(vp == XFS_ITOV(ip));
1126 ASSERT(ip->i_mount == mp);
1128 vnode_refed = B_TRUE;
1131 /* From here on in the loop we may have a marker record
1132 * in the inode list.
1135 if ((flags & SYNC_CLOSE) && (vp != NULL)) {
1137 * This is the shutdown case. We just need to
1138 * flush and invalidate all the pages associated
1139 * with the inode. Drop the inode lock since
1140 * we can't hold it across calls to the buffer
1143 * We don't set the VREMAPPING bit in the vnode
1144 * here, because we don't hold the vnode lock
1145 * exclusively. It doesn't really matter, though,
1146 * because we only come here when we're shutting
1149 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1151 if (XFS_FORCED_SHUTDOWN(mp)) {
1152 VOP_TOSS_PAGES(vp, 0, -1, FI_REMAPF);
1154 VOP_FLUSHINVAL_PAGES(vp, 0, -1, FI_REMAPF);
1157 xfs_ilock(ip, XFS_ILOCK_SHARED);
1159 } else if ((flags & SYNC_DELWRI) && (vp != NULL)) {
1161 /* We need to have dropped the lock here,
1162 * so insert a marker if we have not already
1166 IPOINTER_INSERT(ip, mp);
1170 * Drop the inode lock since we can't hold it
1171 * across calls to the buffer cache.
1173 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1174 VOP_FLUSH_PAGES(vp, (xfs_off_t)0, -1,
1175 fflag, FI_NONE, error);
1176 xfs_ilock(ip, XFS_ILOCK_SHARED);
1181 if (flags & SYNC_BDFLUSH) {
1182 if ((flags & SYNC_ATTR) &&
1183 ((ip->i_update_core) ||
1184 ((ip->i_itemp != NULL) &&
1185 (ip->i_itemp->ili_format.ilf_fields != 0)))) {
1187 /* Insert marker and drop lock if not already
1191 IPOINTER_INSERT(ip, mp);
1195 * We don't want the periodic flushing of the
1196 * inodes by vfs_sync() to interfere with
1197 * I/O to the file, especially read I/O
1198 * where it is only the access time stamp
1199 * that is being flushed out. To prevent
1200 * long periods where we have both inode
1201 * locks held shared here while reading the
1202 * inode's buffer in from disk, we drop the
1203 * inode lock while reading in the inode
1204 * buffer. We have to release the buffer
1205 * and reacquire the inode lock so that they
1206 * are acquired in the proper order (inode
1207 * locks first). The buffer will go at the
1208 * end of the lru chain, though, so we can
1209 * expect it to still be there when we go
1210 * for it again in xfs_iflush().
1212 if ((xfs_ipincount(ip) == 0) &&
1213 xfs_iflock_nowait(ip)) {
1216 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1218 error = xfs_itobp(mp, NULL, ip,
1223 /* Bailing out, remove the
1224 * marker and free it.
1226 XFS_MOUNT_ILOCK(mp);
1228 IPOINTER_REMOVE(ip, mp);
1230 XFS_MOUNT_IUNLOCK(mp);
1232 ASSERT(!(lock_flags &
1233 XFS_IOLOCK_SHARED));
1236 sizeof(xfs_iptr_t));
1241 * Since we dropped the inode lock,
1242 * the inode may have been reclaimed.
1243 * Therefore, we reacquire the mount
1244 * lock and check to see if we were the
1245 * inode reclaimed. If this happened
1246 * then the ipointer marker will no
1247 * longer point back at us. In this
1248 * case, move ip along to the inode
1249 * after the marker, remove the marker
1252 XFS_MOUNT_ILOCK(mp);
1253 mount_locked = B_TRUE;
1255 if (ip != ipointer->ip_mprev) {
1256 IPOINTER_REMOVE(ip, mp);
1258 ASSERT(!vnode_refed);
1259 ASSERT(!(lock_flags &
1260 XFS_IOLOCK_SHARED));
1264 ASSERT(ip->i_mount == mp);
1266 if (xfs_ilock_nowait(ip,
1267 XFS_ILOCK_SHARED) == 0) {
1268 ASSERT(ip->i_mount == mp);
1270 * We failed to reacquire
1271 * the inode lock without
1272 * sleeping, so just skip
1273 * the inode for now. We
1274 * clear the ILOCK bit from
1275 * the lock_flags so that we
1276 * won't try to drop a lock
1277 * we don't hold below.
1279 lock_flags &= ~XFS_ILOCK_SHARED;
1280 IPOINTER_REMOVE(ip_next, mp);
1281 } else if ((xfs_ipincount(ip) == 0) &&
1282 xfs_iflock_nowait(ip)) {
1283 ASSERT(ip->i_mount == mp);
1285 * Since this is vfs_sync()
1286 * calling we only flush the
1287 * inode out if we can lock
1288 * it without sleeping and
1289 * it is not pinned. Drop
1290 * the mount lock here so
1291 * that we don't hold it for
1292 * too long. We already have
1293 * a marker in the list here.
1295 XFS_MOUNT_IUNLOCK(mp);
1296 mount_locked = B_FALSE;
1297 error = xfs_iflush(ip,
1300 ASSERT(ip->i_mount == mp);
1301 IPOINTER_REMOVE(ip_next, mp);
1308 if ((flags & SYNC_ATTR) &&
1309 ((ip->i_update_core) ||
1310 ((ip->i_itemp != NULL) &&
1311 (ip->i_itemp->ili_format.ilf_fields != 0)))) {
1313 IPOINTER_INSERT(ip, mp);
1316 if (flags & SYNC_WAIT) {
1318 error = xfs_iflush(ip,
1322 * If we can't acquire the flush
1323 * lock, then the inode is already
1324 * being flushed so don't bother
1325 * waiting. If we can lock it then
1326 * do a delwri flush so we can
1327 * combine multiple inode flushes
1328 * in each disk write.
1330 if (xfs_iflock_nowait(ip)) {
1331 error = xfs_iflush(ip,
1340 if (lock_flags != 0) {
1341 xfs_iunlock(ip, lock_flags);
1346 * If we had to take a reference on the vnode
1347 * above, then wait until after we've unlocked
1348 * the inode to release the reference. This is
1349 * because we can be already holding the inode
1350 * lock when VN_RELE() calls xfs_inactive().
1352 * Make sure to drop the mount lock before calling
1353 * VN_RELE() so that we don't trip over ourselves if
1354 * we have to go for the mount lock again in the
1358 IPOINTER_INSERT(ip, mp);
1363 vnode_refed = B_FALSE;
1371 * bail out if the filesystem is corrupted.
1373 if (error == EFSCORRUPTED) {
1374 if (!mount_locked) {
1375 XFS_MOUNT_ILOCK(mp);
1376 IPOINTER_REMOVE(ip, mp);
1378 XFS_MOUNT_IUNLOCK(mp);
1379 ASSERT(ipointer_in == B_FALSE);
1380 kmem_free(ipointer, sizeof(xfs_iptr_t));
1381 return XFS_ERROR(error);
1384 /* Let other threads have a chance at the mount lock
1385 * if we have looped many times without dropping the
1388 if ((++preempt & XFS_PREEMPT_MASK) == 0) {
1390 IPOINTER_INSERT(ip, mp);
1394 if (mount_locked == B_FALSE) {
1395 XFS_MOUNT_ILOCK(mp);
1396 mount_locked = B_TRUE;
1397 IPOINTER_REMOVE(ip, mp);
1401 ASSERT(ipointer_in == B_FALSE);
1404 } while (ip != mp->m_inodes);
1406 XFS_MOUNT_IUNLOCK(mp);
1408 ASSERT(ipointer_in == B_FALSE);
1410 kmem_free(ipointer, sizeof(xfs_iptr_t));
1411 return XFS_ERROR(last_error);
1415 * xfs sync routine for internal use
1417 * This routine supports all of the flags defined for the generic VFS_SYNC
1418 * interface as explained above under xfs_sync. In the interests of not
1419 * changing interfaces within the 6.5 family, additional internallly-
1420 * required functions are specified within a separate xflags parameter,
1421 * only available by calling this routine.
1433 uint log_flags = XFS_LOG_FORCE;
1435 xfs_buf_log_item_t *bip;
1438 * Sync out the log. This ensures that the log is periodically
1439 * flushed even if there is not enough activity to fill it up.
1441 if (flags & SYNC_WAIT)
1442 log_flags |= XFS_LOG_SYNC;
1444 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1446 if (flags & (SYNC_ATTR|SYNC_DELWRI)) {
1447 if (flags & SYNC_BDFLUSH)
1448 xfs_finish_reclaim_all(mp, 1);
1450 error = xfs_sync_inodes(mp, flags, xflags, bypassed);
1454 * Flushing out dirty data above probably generated more
1455 * log activity, so if this isn't vfs_sync() then flush
1458 if (flags & SYNC_DELWRI) {
1459 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1462 if (flags & SYNC_FSDATA) {
1464 * If this is vfs_sync() then only sync the superblock
1465 * if we can lock it without sleeping and it is not pinned.
1467 if (flags & SYNC_BDFLUSH) {
1468 bp = xfs_getsb(mp, XFS_BUF_TRYLOCK);
1470 bip = XFS_BUF_FSPRIVATE(bp,xfs_buf_log_item_t*);
1471 if ((bip != NULL) &&
1472 xfs_buf_item_dirty(bip)) {
1473 if (!(XFS_BUF_ISPINNED(bp))) {
1475 error = xfs_bwrite(mp, bp);
1484 bp = xfs_getsb(mp, 0);
1486 * If the buffer is pinned then push on the log so
1487 * we won't get stuck waiting in the write for
1488 * someone, maybe ourselves, to flush the log.
1489 * Even though we just pushed the log above, we
1490 * did not have the superblock buffer locked at
1491 * that point so it can become pinned in between
1494 if (XFS_BUF_ISPINNED(bp))
1495 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
1496 if (flags & SYNC_WAIT)
1497 XFS_BUF_UNASYNC(bp);
1500 error = xfs_bwrite(mp, bp);
1508 * If this is the periodic sync, then kick some entries out of
1509 * the reference cache. This ensures that idle entries are
1510 * eventually kicked out of the cache.
1512 if (flags & SYNC_REFCACHE) {
1513 if (flags & SYNC_WAIT)
1514 xfs_refcache_purge_mp(mp);
1516 xfs_refcache_purge_some(mp);
1520 * Now check to see if the log needs a "dummy" transaction.
1523 if (!(flags & SYNC_REMOUNT) && xfs_log_need_covered(mp)) {
1528 * Put a dummy transaction in the log to tell
1529 * recovery that all others are OK.
1531 tp = xfs_trans_alloc(mp, XFS_TRANS_DUMMY1);
1532 if ((error = xfs_trans_reserve(tp, 0,
1533 XFS_ICHANGE_LOG_RES(mp),
1535 xfs_trans_cancel(tp, 0);
1540 xfs_ilock(ip, XFS_ILOCK_EXCL);
1542 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1543 xfs_trans_ihold(tp, ip);
1544 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1545 error = xfs_trans_commit(tp, 0, NULL);
1546 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1547 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1551 * When shutting down, we need to insure that the AIL is pushed
1552 * to disk or the filesystem can appear corrupt from the PROM.
1554 if ((flags & (SYNC_CLOSE|SYNC_WAIT)) == (SYNC_CLOSE|SYNC_WAIT)) {
1555 XFS_bflush(mp->m_ddev_targp);
1556 if (mp->m_rtdev_targp) {
1557 XFS_bflush(mp->m_rtdev_targp);
1561 return XFS_ERROR(last_error);
1565 * xfs_vget - called by DMAPI and NFSD to get vnode from file handle
1573 xfs_mount_t *mp = XFS_BHVTOM(bdp);
1574 xfs_fid_t *xfid = (struct xfs_fid *)fidp;
1581 * Invalid. Since handles can be created in user space and passed in
1582 * via gethandle(), this is not cause for a panic.
1584 if (xfid->xfs_fid_len != sizeof(*xfid) - sizeof(xfid->xfs_fid_len))
1585 return XFS_ERROR(EINVAL);
1587 ino = xfid->xfs_fid_ino;
1588 igen = xfid->xfs_fid_gen;
1591 * NFS can sometimes send requests for ino 0. Fail them gracefully.
1594 return XFS_ERROR(ESTALE);
1596 error = xfs_iget(mp, NULL, ino, 0, XFS_ILOCK_SHARED, &ip, 0);
1604 return XFS_ERROR(EIO);
1607 if (ip->i_d.di_mode == 0 || ip->i_d.di_gen != igen) {
1608 xfs_iput_new(ip, XFS_ILOCK_SHARED);
1610 return XFS_ERROR(ENOENT);
1613 *vpp = XFS_ITOV(ip);
1614 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1619 #define MNTOPT_LOGBUFS "logbufs" /* number of XFS log buffers */
1620 #define MNTOPT_LOGBSIZE "logbsize" /* size of XFS log buffers */
1621 #define MNTOPT_LOGDEV "logdev" /* log device */
1622 #define MNTOPT_RTDEV "rtdev" /* realtime I/O device */
1623 #define MNTOPT_BIOSIZE "biosize" /* log2 of preferred buffered io size */
1624 #define MNTOPT_WSYNC "wsync" /* safe-mode nfs compatible mount */
1625 #define MNTOPT_INO64 "ino64" /* force inodes into 64-bit range */
1626 #define MNTOPT_NOALIGN "noalign" /* turn off stripe alignment */
1627 #define MNTOPT_SWALLOC "swalloc" /* turn on stripe width allocation */
1628 #define MNTOPT_SUNIT "sunit" /* data volume stripe unit */
1629 #define MNTOPT_SWIDTH "swidth" /* data volume stripe width */
1630 #define MNTOPT_NOUUID "nouuid" /* ignore filesystem UUID */
1631 #define MNTOPT_MTPT "mtpt" /* filesystem mount point */
1632 #define MNTOPT_GRPID "grpid" /* group-ID from parent directory */
1633 #define MNTOPT_NOGRPID "nogrpid" /* group-ID from current process */
1634 #define MNTOPT_BSDGROUPS "bsdgroups" /* group-ID from parent directory */
1635 #define MNTOPT_SYSVGROUPS "sysvgroups" /* group-ID from current process */
1636 #define MNTOPT_ALLOCSIZE "allocsize" /* preferred allocation size */
1637 #define MNTOPT_IHASHSIZE "ihashsize" /* size of inode hash table */
1638 #define MNTOPT_NORECOVERY "norecovery" /* don't run XFS recovery */
1639 #define MNTOPT_BARRIER "barrier" /* use writer barriers for log write and
1640 * unwritten extent conversion */
1641 #define MNTOPT_OSYNCISOSYNC "osyncisosync" /* o_sync is REALLY o_sync */
1642 #define MNTOPT_64BITINODE "inode64" /* inodes can be allocated anywhere */
1643 #define MNTOPT_IKEEP "ikeep" /* do not free empty inode clusters */
1644 #define MNTOPT_NOIKEEP "noikeep" /* free empty inode clusters */
1645 #define MNTOPT_LARGEIO "largeio" /* report large I/O sizes in stat() */
1646 #define MNTOPT_NOLARGEIO "nolargeio" /* do not report large I/O sizes
1648 #define MNTOPT_ATTR2 "attr2" /* do use attr2 attribute format */
1649 #define MNTOPT_NOATTR2 "noattr2" /* do not use attr2 attribute format */
1651 STATIC unsigned long
1652 suffix_strtoul(const char *cp, char **endp, unsigned int base)
1654 int last, shift_left_factor = 0;
1655 char *value = (char *)cp;
1657 last = strlen(value) - 1;
1658 if (value[last] == 'K' || value[last] == 'k') {
1659 shift_left_factor = 10;
1662 if (value[last] == 'M' || value[last] == 'm') {
1663 shift_left_factor = 20;
1666 if (value[last] == 'G' || value[last] == 'g') {
1667 shift_left_factor = 30;
1671 return simple_strtoul(cp, endp, base) << shift_left_factor;
1676 struct bhv_desc *bhv,
1678 struct xfs_mount_args *args,
1681 struct vfs *vfsp = bhvtovfs(bhv);
1682 char *this_char, *value, *eov;
1683 int dsunit, dswidth, vol_dsunit, vol_dswidth;
1686 args->flags |= XFSMNT_COMPAT_IOSIZE;
1687 #if 0 /* XXX: off by default, until some remaining issues ironed out */
1688 args->flags |= XFSMNT_IDELETE; /* default to on */
1694 iosize = dsunit = dswidth = vol_dsunit = vol_dswidth = 0;
1696 while ((this_char = strsep(&options, ",")) != NULL) {
1699 if ((value = strchr(this_char, '=')) != NULL)
1702 if (!strcmp(this_char, MNTOPT_LOGBUFS)) {
1703 if (!value || !*value) {
1704 printk("XFS: %s option requires an argument\n",
1708 args->logbufs = simple_strtoul(value, &eov, 10);
1709 } else if (!strcmp(this_char, MNTOPT_LOGBSIZE)) {
1710 if (!value || !*value) {
1711 printk("XFS: %s option requires an argument\n",
1715 args->logbufsize = suffix_strtoul(value, &eov, 10);
1716 } else if (!strcmp(this_char, MNTOPT_LOGDEV)) {
1717 if (!value || !*value) {
1718 printk("XFS: %s option requires an argument\n",
1722 strncpy(args->logname, value, MAXNAMELEN);
1723 } else if (!strcmp(this_char, MNTOPT_MTPT)) {
1724 if (!value || !*value) {
1725 printk("XFS: %s option requires an argument\n",
1729 strncpy(args->mtpt, value, MAXNAMELEN);
1730 } else if (!strcmp(this_char, MNTOPT_RTDEV)) {
1731 if (!value || !*value) {
1732 printk("XFS: %s option requires an argument\n",
1736 strncpy(args->rtname, value, MAXNAMELEN);
1737 } else if (!strcmp(this_char, MNTOPT_BIOSIZE)) {
1738 if (!value || !*value) {
1739 printk("XFS: %s option requires an argument\n",
1743 iosize = simple_strtoul(value, &eov, 10);
1744 args->flags |= XFSMNT_IOSIZE;
1745 args->iosizelog = (uint8_t) iosize;
1746 } else if (!strcmp(this_char, MNTOPT_ALLOCSIZE)) {
1747 if (!value || !*value) {
1748 printk("XFS: %s option requires an argument\n",
1752 iosize = suffix_strtoul(value, &eov, 10);
1753 args->flags |= XFSMNT_IOSIZE;
1754 args->iosizelog = ffs(iosize) - 1;
1755 } else if (!strcmp(this_char, MNTOPT_IHASHSIZE)) {
1756 if (!value || !*value) {
1757 printk("XFS: %s option requires an argument\n",
1761 args->flags |= XFSMNT_IHASHSIZE;
1762 args->ihashsize = simple_strtoul(value, &eov, 10);
1763 } else if (!strcmp(this_char, MNTOPT_GRPID) ||
1764 !strcmp(this_char, MNTOPT_BSDGROUPS)) {
1765 vfsp->vfs_flag |= VFS_GRPID;
1766 } else if (!strcmp(this_char, MNTOPT_NOGRPID) ||
1767 !strcmp(this_char, MNTOPT_SYSVGROUPS)) {
1768 vfsp->vfs_flag &= ~VFS_GRPID;
1769 } else if (!strcmp(this_char, MNTOPT_WSYNC)) {
1770 args->flags |= XFSMNT_WSYNC;
1771 } else if (!strcmp(this_char, MNTOPT_OSYNCISOSYNC)) {
1772 args->flags |= XFSMNT_OSYNCISOSYNC;
1773 } else if (!strcmp(this_char, MNTOPT_NORECOVERY)) {
1774 args->flags |= XFSMNT_NORECOVERY;
1775 } else if (!strcmp(this_char, MNTOPT_INO64)) {
1776 args->flags |= XFSMNT_INO64;
1778 printk("XFS: %s option not allowed on this system\n",
1782 } else if (!strcmp(this_char, MNTOPT_NOALIGN)) {
1783 args->flags |= XFSMNT_NOALIGN;
1784 } else if (!strcmp(this_char, MNTOPT_SWALLOC)) {
1785 args->flags |= XFSMNT_SWALLOC;
1786 } else if (!strcmp(this_char, MNTOPT_SUNIT)) {
1787 if (!value || !*value) {
1788 printk("XFS: %s option requires an argument\n",
1792 dsunit = simple_strtoul(value, &eov, 10);
1793 } else if (!strcmp(this_char, MNTOPT_SWIDTH)) {
1794 if (!value || !*value) {
1795 printk("XFS: %s option requires an argument\n",
1799 dswidth = simple_strtoul(value, &eov, 10);
1800 } else if (!strcmp(this_char, MNTOPT_64BITINODE)) {
1801 args->flags &= ~XFSMNT_32BITINODES;
1803 printk("XFS: %s option not allowed on this system\n",
1807 } else if (!strcmp(this_char, MNTOPT_NOUUID)) {
1808 args->flags |= XFSMNT_NOUUID;
1809 } else if (!strcmp(this_char, MNTOPT_BARRIER)) {
1810 args->flags |= XFSMNT_BARRIER;
1811 } else if (!strcmp(this_char, MNTOPT_IKEEP)) {
1812 args->flags &= ~XFSMNT_IDELETE;
1813 } else if (!strcmp(this_char, MNTOPT_NOIKEEP)) {
1814 args->flags |= XFSMNT_IDELETE;
1815 } else if (!strcmp(this_char, MNTOPT_LARGEIO)) {
1816 args->flags &= ~XFSMNT_COMPAT_IOSIZE;
1817 } else if (!strcmp(this_char, MNTOPT_NOLARGEIO)) {
1818 args->flags |= XFSMNT_COMPAT_IOSIZE;
1819 } else if (!strcmp(this_char, MNTOPT_ATTR2)) {
1820 args->flags &= ~XFSMNT_COMPAT_ATTR;
1821 } else if (!strcmp(this_char, MNTOPT_NOATTR2)) {
1822 args->flags |= XFSMNT_COMPAT_ATTR;
1823 } else if (!strcmp(this_char, "osyncisdsync")) {
1824 /* no-op, this is now the default */
1825 printk("XFS: osyncisdsync is now the default, option is deprecated.\n");
1826 } else if (!strcmp(this_char, "irixsgid")) {
1827 printk("XFS: irixsgid is now a sysctl(2) variable, option is deprecated.\n");
1829 printk("XFS: unknown mount option [%s].\n", this_char);
1834 if (args->flags & XFSMNT_NORECOVERY) {
1835 if ((vfsp->vfs_flag & VFS_RDONLY) == 0) {
1836 printk("XFS: no-recovery mounts must be read-only.\n");
1841 if ((args->flags & XFSMNT_NOALIGN) && (dsunit || dswidth)) {
1843 "XFS: sunit and swidth options incompatible with the noalign option\n");
1847 if ((dsunit && !dswidth) || (!dsunit && dswidth)) {
1848 printk("XFS: sunit and swidth must be specified together\n");
1852 if (dsunit && (dswidth % dsunit != 0)) {
1854 "XFS: stripe width (%d) must be a multiple of the stripe unit (%d)\n",
1859 if ((args->flags & XFSMNT_NOALIGN) != XFSMNT_NOALIGN) {
1861 args->sunit = dsunit;
1862 args->flags |= XFSMNT_RETERR;
1864 args->sunit = vol_dsunit;
1866 dswidth ? (args->swidth = dswidth) :
1867 (args->swidth = vol_dswidth);
1869 args->sunit = args->swidth = 0;
1877 struct bhv_desc *bhv,
1880 static struct proc_xfs_info {
1884 /* the few simple ones we can get from the mount struct */
1885 { XFS_MOUNT_WSYNC, "," MNTOPT_WSYNC },
1886 { XFS_MOUNT_INO64, "," MNTOPT_INO64 },
1887 { XFS_MOUNT_NOALIGN, "," MNTOPT_NOALIGN },
1888 { XFS_MOUNT_SWALLOC, "," MNTOPT_SWALLOC },
1889 { XFS_MOUNT_NOUUID, "," MNTOPT_NOUUID },
1890 { XFS_MOUNT_NORECOVERY, "," MNTOPT_NORECOVERY },
1891 { XFS_MOUNT_OSYNCISOSYNC, "," MNTOPT_OSYNCISOSYNC },
1892 { XFS_MOUNT_BARRIER, "," MNTOPT_BARRIER },
1893 { XFS_MOUNT_IDELETE, "," MNTOPT_NOIKEEP },
1896 struct proc_xfs_info *xfs_infop;
1897 struct xfs_mount *mp = XFS_BHVTOM(bhv);
1898 struct vfs *vfsp = XFS_MTOVFS(mp);
1900 for (xfs_infop = xfs_info; xfs_infop->flag; xfs_infop++) {
1901 if (mp->m_flags & xfs_infop->flag)
1902 seq_puts(m, xfs_infop->str);
1905 if (mp->m_flags & XFS_MOUNT_IHASHSIZE)
1906 seq_printf(m, "," MNTOPT_IHASHSIZE "=%d", mp->m_ihsize);
1908 if (mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)
1909 seq_printf(m, "," MNTOPT_ALLOCSIZE "=%d", 1<<mp->m_writeio_log);
1911 if (mp->m_logbufs > 0)
1912 seq_printf(m, "," MNTOPT_LOGBUFS "=%d", mp->m_logbufs);
1914 if (mp->m_logbsize > 0)
1915 seq_printf(m, "," MNTOPT_LOGBSIZE "=%d", mp->m_logbsize);
1917 if (mp->m_ddev_targp != mp->m_logdev_targp)
1918 seq_printf(m, "," MNTOPT_LOGDEV "=%s",
1919 XFS_BUFTARG_NAME(mp->m_logdev_targp));
1921 if (mp->m_rtdev_targp && mp->m_ddev_targp != mp->m_rtdev_targp)
1922 seq_printf(m, "," MNTOPT_RTDEV "=%s",
1923 XFS_BUFTARG_NAME(mp->m_rtdev_targp));
1925 if (mp->m_dalign > 0)
1926 seq_printf(m, "," MNTOPT_SUNIT "=%d",
1927 (int)XFS_FSB_TO_BB(mp, mp->m_dalign));
1929 if (mp->m_swidth > 0)
1930 seq_printf(m, "," MNTOPT_SWIDTH "=%d",
1931 (int)XFS_FSB_TO_BB(mp, mp->m_swidth));
1933 if (!(mp->m_flags & XFS_MOUNT_32BITINOOPT))
1934 seq_printf(m, "," MNTOPT_64BITINODE);
1936 if (vfsp->vfs_flag & VFS_GRPID)
1937 seq_printf(m, "," MNTOPT_GRPID);
1946 xfs_mount_t *mp = XFS_BHVTOM(bdp);
1948 while (atomic_read(&mp->m_active_trans) > 0)
1951 /* Push the superblock and write an unmount record */
1952 xfs_log_unmount_write(mp);
1953 xfs_unmountfs_writesb(mp);
1957 vfsops_t xfs_vfsops = {
1958 BHV_IDENTITY_INIT(VFS_BHV_XFS,VFS_POSITION_XFS),
1959 .vfs_parseargs = xfs_parseargs,
1960 .vfs_showargs = xfs_showargs,
1961 .vfs_mount = xfs_mount,
1962 .vfs_unmount = xfs_unmount,
1963 .vfs_mntupdate = xfs_mntupdate,
1964 .vfs_root = xfs_root,
1965 .vfs_statvfs = xfs_statvfs,
1966 .vfs_sync = xfs_sync,
1967 .vfs_vget = xfs_vget,
1968 .vfs_dmapiops = (vfs_dmapiops_t)fs_nosys,
1969 .vfs_quotactl = (vfs_quotactl_t)fs_nosys,
1970 .vfs_init_vnode = xfs_initialize_vnode,
1971 .vfs_force_shutdown = xfs_do_force_shutdown,
1972 .vfs_freeze = xfs_freeze,