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"
29 #include "xfs_dmapi.h"
30 #include "xfs_mount.h"
31 #include "xfs_da_btree.h"
32 #include "xfs_bmap_btree.h"
33 #include "xfs_ialloc_btree.h"
34 #include "xfs_alloc_btree.h"
35 #include "xfs_dir_sf.h"
36 #include "xfs_dir2_sf.h"
37 #include "xfs_attr_sf.h"
38 #include "xfs_dinode.h"
39 #include "xfs_inode.h"
40 #include "xfs_inode_item.h"
41 #include "xfs_btree.h"
42 #include "xfs_alloc.h"
43 #include "xfs_ialloc.h"
44 #include "xfs_quota.h"
45 #include "xfs_error.h"
48 #include "xfs_refcache.h"
49 #include "xfs_buf_item.h"
50 #include "xfs_log_priv.h"
51 #include "xfs_dir2_trace.h"
52 #include "xfs_extfree_item.h"
56 #include "xfs_fsops.h"
58 STATIC int xfs_sync(bhv_desc_t *, int, cred_t *);
63 extern kmem_zone_t *xfs_bmap_free_item_zone;
64 extern kmem_zone_t *xfs_btree_cur_zone;
65 extern kmem_zone_t *xfs_trans_zone;
66 extern kmem_zone_t *xfs_buf_item_zone;
67 extern kmem_zone_t *xfs_dabuf_zone;
68 #ifdef XFS_DABUF_DEBUG
69 extern lock_t xfs_dabuf_global_lock;
70 spinlock_init(&xfs_dabuf_global_lock, "xfsda");
74 * Initialize all of the zone allocators we use.
76 xfs_bmap_free_item_zone = kmem_zone_init(sizeof(xfs_bmap_free_item_t),
77 "xfs_bmap_free_item");
78 xfs_btree_cur_zone = kmem_zone_init(sizeof(xfs_btree_cur_t),
80 xfs_inode_zone = kmem_zone_init(sizeof(xfs_inode_t), "xfs_inode");
81 xfs_trans_zone = kmem_zone_init(sizeof(xfs_trans_t), "xfs_trans");
83 kmem_zone_init(sizeof(xfs_da_state_t), "xfs_da_state");
84 xfs_dabuf_zone = kmem_zone_init(sizeof(xfs_dabuf_t), "xfs_dabuf");
87 * The size of the zone allocated buf log item is the maximum
88 * size possible under XFS. This wastes a little bit of memory,
89 * but it is much faster.
92 kmem_zone_init((sizeof(xfs_buf_log_item_t) +
93 (((XFS_MAX_BLOCKSIZE / XFS_BLI_CHUNK) /
94 NBWORD) * sizeof(int))),
96 xfs_efd_zone = kmem_zone_init((sizeof(xfs_efd_log_item_t) +
97 ((XFS_EFD_MAX_FAST_EXTENTS - 1) * sizeof(xfs_extent_t))),
99 xfs_efi_zone = kmem_zone_init((sizeof(xfs_efi_log_item_t) +
100 ((XFS_EFI_MAX_FAST_EXTENTS - 1) * sizeof(xfs_extent_t))),
102 xfs_ifork_zone = kmem_zone_init(sizeof(xfs_ifork_t), "xfs_ifork");
103 xfs_ili_zone = kmem_zone_init(sizeof(xfs_inode_log_item_t), "xfs_ili");
104 xfs_chashlist_zone = kmem_zone_init(sizeof(xfs_chashlist_t),
106 xfs_acl_zone_init(xfs_acl_zone, "xfs_acl");
109 * Allocate global trace buffers.
111 #ifdef XFS_ALLOC_TRACE
112 xfs_alloc_trace_buf = ktrace_alloc(XFS_ALLOC_TRACE_SIZE, KM_SLEEP);
114 #ifdef XFS_BMAP_TRACE
115 xfs_bmap_trace_buf = ktrace_alloc(XFS_BMAP_TRACE_SIZE, KM_SLEEP);
117 #ifdef XFS_BMBT_TRACE
118 xfs_bmbt_trace_buf = ktrace_alloc(XFS_BMBT_TRACE_SIZE, KM_SLEEP);
121 xfs_dir_trace_buf = ktrace_alloc(XFS_DIR_TRACE_SIZE, KM_SLEEP);
123 #ifdef XFS_ATTR_TRACE
124 xfs_attr_trace_buf = ktrace_alloc(XFS_ATTR_TRACE_SIZE, KM_SLEEP);
126 #ifdef XFS_DIR2_TRACE
127 xfs_dir2_trace_buf = ktrace_alloc(XFS_DIR2_GTRACE_SIZE, KM_SLEEP);
132 #if (defined(DEBUG) || defined(INDUCE_IO_ERROR))
133 xfs_error_test_init();
134 #endif /* DEBUG || INDUCE_IO_ERROR */
137 xfs_sysctl_register();
144 extern kmem_zone_t *xfs_bmap_free_item_zone;
145 extern kmem_zone_t *xfs_btree_cur_zone;
146 extern kmem_zone_t *xfs_inode_zone;
147 extern kmem_zone_t *xfs_trans_zone;
148 extern kmem_zone_t *xfs_da_state_zone;
149 extern kmem_zone_t *xfs_dabuf_zone;
150 extern kmem_zone_t *xfs_efd_zone;
151 extern kmem_zone_t *xfs_efi_zone;
152 extern kmem_zone_t *xfs_buf_item_zone;
153 extern kmem_zone_t *xfs_chashlist_zone;
155 xfs_cleanup_procfs();
156 xfs_sysctl_unregister();
157 xfs_refcache_destroy();
158 xfs_acl_zone_destroy(xfs_acl_zone);
160 #ifdef XFS_DIR2_TRACE
161 ktrace_free(xfs_dir2_trace_buf);
163 #ifdef XFS_ATTR_TRACE
164 ktrace_free(xfs_attr_trace_buf);
167 ktrace_free(xfs_dir_trace_buf);
169 #ifdef XFS_BMBT_TRACE
170 ktrace_free(xfs_bmbt_trace_buf);
172 #ifdef XFS_BMAP_TRACE
173 ktrace_free(xfs_bmap_trace_buf);
175 #ifdef XFS_ALLOC_TRACE
176 ktrace_free(xfs_alloc_trace_buf);
179 kmem_cache_destroy(xfs_bmap_free_item_zone);
180 kmem_cache_destroy(xfs_btree_cur_zone);
181 kmem_cache_destroy(xfs_inode_zone);
182 kmem_cache_destroy(xfs_trans_zone);
183 kmem_cache_destroy(xfs_da_state_zone);
184 kmem_cache_destroy(xfs_dabuf_zone);
185 kmem_cache_destroy(xfs_buf_item_zone);
186 kmem_cache_destroy(xfs_efd_zone);
187 kmem_cache_destroy(xfs_efi_zone);
188 kmem_cache_destroy(xfs_ifork_zone);
189 kmem_cache_destroy(xfs_ili_zone);
190 kmem_cache_destroy(xfs_chashlist_zone);
196 * This function fills in xfs_mount_t fields based on mount args.
197 * Note: the superblock has _not_ yet been read in.
202 struct xfs_mount_args *ap,
203 struct xfs_mount *mp)
205 /* Values are in BBs */
206 if ((ap->flags & XFSMNT_NOALIGN) != XFSMNT_NOALIGN) {
208 * At this point the superblock has not been read
209 * in, therefore we do not know the block size.
210 * Before the mount call ends we will convert
213 mp->m_dalign = ap->sunit;
214 mp->m_swidth = ap->swidth;
217 if (ap->logbufs != -1 &&
219 (ap->logbufs < XLOG_MIN_ICLOGS ||
220 ap->logbufs > XLOG_MAX_ICLOGS)) {
222 "XFS: invalid logbufs value: %d [not %d-%d]",
223 ap->logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
224 return XFS_ERROR(EINVAL);
226 mp->m_logbufs = ap->logbufs;
227 if (ap->logbufsize != -1 &&
228 ap->logbufsize != 0 &&
229 ap->logbufsize != 16 * 1024 &&
230 ap->logbufsize != 32 * 1024 &&
231 ap->logbufsize != 64 * 1024 &&
232 ap->logbufsize != 128 * 1024 &&
233 ap->logbufsize != 256 * 1024) {
235 "XFS: invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
237 return XFS_ERROR(EINVAL);
239 mp->m_ihsize = ap->ihashsize;
240 mp->m_logbsize = ap->logbufsize;
241 mp->m_fsname_len = strlen(ap->fsname) + 1;
242 mp->m_fsname = kmem_alloc(mp->m_fsname_len, KM_SLEEP);
243 strcpy(mp->m_fsname, ap->fsname);
245 mp->m_rtname = kmem_alloc(strlen(ap->rtname) + 1, KM_SLEEP);
246 strcpy(mp->m_rtname, ap->rtname);
248 if (ap->logname[0]) {
249 mp->m_logname = kmem_alloc(strlen(ap->logname) + 1, KM_SLEEP);
250 strcpy(mp->m_logname, ap->logname);
253 if (ap->flags & XFSMNT_WSYNC)
254 mp->m_flags |= XFS_MOUNT_WSYNC;
256 if (ap->flags & XFSMNT_INO64) {
257 mp->m_flags |= XFS_MOUNT_INO64;
258 mp->m_inoadd = XFS_INO64_OFFSET;
261 if (ap->flags & XFSMNT_NOATIME)
262 mp->m_flags |= XFS_MOUNT_NOATIME;
263 if (ap->flags & XFSMNT_RETERR)
264 mp->m_flags |= XFS_MOUNT_RETERR;
265 if (ap->flags & XFSMNT_NOALIGN)
266 mp->m_flags |= XFS_MOUNT_NOALIGN;
267 if (ap->flags & XFSMNT_SWALLOC)
268 mp->m_flags |= XFS_MOUNT_SWALLOC;
269 if (ap->flags & XFSMNT_OSYNCISOSYNC)
270 mp->m_flags |= XFS_MOUNT_OSYNCISOSYNC;
271 if (ap->flags & XFSMNT_32BITINODES)
272 mp->m_flags |= XFS_MOUNT_32BITINODES;
274 if (ap->flags & XFSMNT_IOSIZE) {
275 if (ap->iosizelog > XFS_MAX_IO_LOG ||
276 ap->iosizelog < XFS_MIN_IO_LOG) {
278 "XFS: invalid log iosize: %d [not %d-%d]",
279 ap->iosizelog, XFS_MIN_IO_LOG,
281 return XFS_ERROR(EINVAL);
284 mp->m_flags |= XFS_MOUNT_DFLT_IOSIZE;
285 mp->m_readio_log = mp->m_writeio_log = ap->iosizelog;
288 if (ap->flags & XFSMNT_IHASHSIZE)
289 mp->m_flags |= XFS_MOUNT_IHASHSIZE;
290 if (ap->flags & XFSMNT_IDELETE)
291 mp->m_flags |= XFS_MOUNT_IDELETE;
292 if (ap->flags & XFSMNT_DIRSYNC)
293 mp->m_flags |= XFS_MOUNT_DIRSYNC;
294 if (ap->flags & XFSMNT_ATTR2)
295 mp->m_flags |= XFS_MOUNT_ATTR2;
297 if (ap->flags2 & XFSMNT2_COMPAT_IOSIZE)
298 mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE;
301 * no recovery flag requires a read-only mount
303 if (ap->flags & XFSMNT_NORECOVERY) {
304 if (!(vfs->vfs_flag & VFS_RDONLY)) {
306 "XFS: tried to mount a FS read-write without recovery!");
307 return XFS_ERROR(EINVAL);
309 mp->m_flags |= XFS_MOUNT_NORECOVERY;
312 if (ap->flags & XFSMNT_NOUUID)
313 mp->m_flags |= XFS_MOUNT_NOUUID;
314 if (ap->flags & XFSMNT_BARRIER)
315 mp->m_flags |= XFS_MOUNT_BARRIER;
317 mp->m_flags &= ~XFS_MOUNT_BARRIER;
323 * This function fills in xfs_mount_t fields based on mount args.
324 * Note: the superblock _has_ now been read in.
329 struct xfs_mount_args *ap,
330 struct xfs_mount *mp)
332 int ronly = (vfs->vfs_flag & VFS_RDONLY);
334 /* Fail a mount where the logbuf is smaller then the log stripe */
335 if (XFS_SB_VERSION_HASLOGV2(&mp->m_sb)) {
336 if ((ap->logbufsize == -1) &&
337 (mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE)) {
338 mp->m_logbsize = mp->m_sb.sb_logsunit;
339 } else if (ap->logbufsize < mp->m_sb.sb_logsunit) {
341 "XFS: logbuf size must be greater than or equal to log stripe size");
342 return XFS_ERROR(EINVAL);
345 /* Fail a mount if the logbuf is larger than 32K */
346 if (ap->logbufsize > XLOG_BIG_RECORD_BSIZE) {
348 "XFS: logbuf size for version 1 logs must be 16K or 32K");
349 return XFS_ERROR(EINVAL);
353 if (XFS_SB_VERSION_HASATTR2(&mp->m_sb)) {
354 mp->m_flags |= XFS_MOUNT_ATTR2;
358 * prohibit r/w mounts of read-only filesystems
360 if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !ronly) {
362 "XFS: cannot mount a read-only filesystem as read-write");
363 return XFS_ERROR(EROFS);
367 * check for shared mount.
369 if (ap->flags & XFSMNT_SHARED) {
370 if (!XFS_SB_VERSION_HASSHARED(&mp->m_sb))
371 return XFS_ERROR(EINVAL);
374 * For IRIX 6.5, shared mounts must have the shared
375 * version bit set, have the persistent readonly
376 * field set, must be version 0 and can only be mounted
379 if (!ronly || !(mp->m_sb.sb_flags & XFS_SBF_READONLY) ||
380 (mp->m_sb.sb_shared_vn != 0))
381 return XFS_ERROR(EINVAL);
383 mp->m_flags |= XFS_MOUNT_SHARED;
386 * Shared XFS V0 can't deal with DMI. Return EINVAL.
388 if (mp->m_sb.sb_shared_vn == 0 && (ap->flags & XFSMNT_DMAPI))
389 return XFS_ERROR(EINVAL);
398 * The file system configurations are:
399 * (1) device (partition) with data and internal log
400 * (2) logical volume with data and log subvolumes.
401 * (3) logical volume with data, log, and realtime subvolumes.
403 * We only have to handle opening the log and realtime volumes here if
404 * they are present. The data subvolume has already been opened by
405 * get_sb_bdev() and is stored in vfsp->vfs_super->s_bdev.
409 struct bhv_desc *bhvp,
410 struct xfs_mount_args *args,
413 struct vfs *vfsp = bhvtovfs(bhvp);
415 struct xfs_mount *mp = XFS_BHVTOM(bhvp);
416 struct block_device *ddev, *logdev, *rtdev;
417 int flags = 0, error;
419 ddev = vfsp->vfs_super->s_bdev;
420 logdev = rtdev = NULL;
423 * Setup xfs_mount function vectors from available behaviors
425 p = vfs_bhv_lookup(vfsp, VFS_POSITION_DM);
426 mp->m_dm_ops = p ? *(xfs_dmops_t *) vfs_bhv_custom(p) : xfs_dmcore_stub;
427 p = vfs_bhv_lookup(vfsp, VFS_POSITION_QM);
428 mp->m_qm_ops = p ? *(xfs_qmops_t *) vfs_bhv_custom(p) : xfs_qmcore_stub;
429 p = vfs_bhv_lookup(vfsp, VFS_POSITION_IO);
430 mp->m_io_ops = p ? *(xfs_ioops_t *) vfs_bhv_custom(p) : xfs_iocore_xfs;
433 * Open real time and log devices - order is important.
435 if (args->logname[0]) {
436 error = xfs_blkdev_get(mp, args->logname, &logdev);
440 if (args->rtname[0]) {
441 error = xfs_blkdev_get(mp, args->rtname, &rtdev);
443 xfs_blkdev_put(logdev);
447 if (rtdev == ddev || rtdev == logdev) {
449 "XFS: Cannot mount filesystem with identical rtdev and ddev/logdev.");
450 xfs_blkdev_put(logdev);
451 xfs_blkdev_put(rtdev);
457 * Setup xfs_mount buffer target pointers
460 mp->m_ddev_targp = xfs_alloc_buftarg(ddev, 0);
461 if (!mp->m_ddev_targp) {
462 xfs_blkdev_put(logdev);
463 xfs_blkdev_put(rtdev);
467 mp->m_rtdev_targp = xfs_alloc_buftarg(rtdev, 1);
468 if (!mp->m_rtdev_targp)
471 mp->m_logdev_targp = (logdev && logdev != ddev) ?
472 xfs_alloc_buftarg(logdev, 1) : mp->m_ddev_targp;
473 if (!mp->m_logdev_targp)
477 * Setup flags based on mount(2) options and then the superblock
479 error = xfs_start_flags(vfsp, args, mp);
482 error = xfs_readsb(mp);
485 error = xfs_finish_flags(vfsp, args, mp);
490 * Setup xfs_mount buffer target pointers based on superblock
492 error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_blocksize,
493 mp->m_sb.sb_sectsize);
494 if (!error && logdev && logdev != ddev) {
495 unsigned int log_sector_size = BBSIZE;
497 if (XFS_SB_VERSION_HASSECTOR(&mp->m_sb))
498 log_sector_size = mp->m_sb.sb_logsectsize;
499 error = xfs_setsize_buftarg(mp->m_logdev_targp,
500 mp->m_sb.sb_blocksize,
504 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
505 mp->m_sb.sb_blocksize,
506 mp->m_sb.sb_sectsize);
510 if ((mp->m_flags & XFS_MOUNT_BARRIER) && !(vfsp->vfs_flag & VFS_RDONLY))
511 xfs_mountfs_check_barriers(mp);
513 error = XFS_IOINIT(vfsp, args, flags);
523 xfs_binval(mp->m_ddev_targp);
524 if (logdev && logdev != ddev)
525 xfs_binval(mp->m_logdev_targp);
527 xfs_binval(mp->m_rtdev_targp);
529 xfs_unmountfs_close(mp, credp);
539 struct vfs *vfsp = bhvtovfs(bdp);
540 xfs_mount_t *mp = XFS_BHVTOM(bdp);
543 int unmount_event_wanted = 0;
544 int unmount_event_flags = 0;
545 int xfs_unmountfs_needed = 0;
551 if (vfsp->vfs_flag & VFS_DMI) {
552 error = XFS_SEND_PREUNMOUNT(mp, vfsp,
553 rvp, DM_RIGHT_NULL, rvp, DM_RIGHT_NULL,
555 (mp->m_dmevmask & (1<<DM_EVENT_PREUNMOUNT))?
556 0:DM_FLAGS_UNWANTED);
558 return XFS_ERROR(error);
559 unmount_event_wanted = 1;
560 unmount_event_flags = (mp->m_dmevmask & (1<<DM_EVENT_UNMOUNT))?
561 0 : DM_FLAGS_UNWANTED;
565 * First blow any referenced inode from this file system
566 * out of the reference cache, and delete the timer.
568 xfs_refcache_purge_mp(mp);
570 XFS_bflush(mp->m_ddev_targp);
571 error = xfs_unmount_flush(mp, 0);
575 ASSERT(vn_count(rvp) == 1);
578 * Drop the reference count
583 * If we're forcing a shutdown, typically because of a media error,
584 * we want to make sure we invalidate dirty pages that belong to
585 * referenced vnodes as well.
587 if (XFS_FORCED_SHUTDOWN(mp)) {
588 error = xfs_sync(&mp->m_bhv,
589 (SYNC_WAIT | SYNC_CLOSE), credp);
590 ASSERT(error != EFSCORRUPTED);
592 xfs_unmountfs_needed = 1;
595 /* Send DMAPI event, if required.
596 * Then do xfs_unmountfs() if needed.
597 * Then return error (or zero).
599 if (unmount_event_wanted) {
600 /* Note: mp structure must still exist for
601 * XFS_SEND_UNMOUNT() call.
603 XFS_SEND_UNMOUNT(mp, vfsp, error == 0 ? rvp : NULL,
604 DM_RIGHT_NULL, 0, error, unmount_event_flags);
606 if (xfs_unmountfs_needed) {
608 * Call common unmount function to flush to disk
609 * and free the super block buffer & mount structures.
611 xfs_unmountfs(mp, credp);
614 return XFS_ERROR(error);
621 int count = 0, pincount;
623 xfs_refcache_purge_mp(mp);
624 xfs_flush_buftarg(mp->m_ddev_targp, 0);
625 xfs_finish_reclaim_all(mp, 0);
627 /* This loop must run at least twice.
628 * The first instance of the loop will flush
629 * most meta data but that will generate more
630 * meta data (typically directory updates).
631 * Which then must be flushed and logged before
632 * we can write the unmount record.
635 xfs_syncsub(mp, SYNC_REMOUNT|SYNC_ATTR|SYNC_WAIT, 0, NULL);
636 pincount = xfs_flush_buftarg(mp->m_ddev_targp, 1);
650 struct xfs_mount_args *args)
652 struct vfs *vfsp = bhvtovfs(bdp);
653 xfs_mount_t *mp = XFS_BHVTOM(bdp);
656 if (args->flags & XFSMNT_NOATIME)
657 mp->m_flags |= XFS_MOUNT_NOATIME;
659 mp->m_flags &= ~XFS_MOUNT_NOATIME;
661 if (args->flags & XFSMNT_BARRIER)
662 mp->m_flags |= XFS_MOUNT_BARRIER;
664 mp->m_flags &= ~XFS_MOUNT_BARRIER;
666 if ((vfsp->vfs_flag & VFS_RDONLY) &&
667 !(*flags & MS_RDONLY)) {
668 vfsp->vfs_flag &= ~VFS_RDONLY;
670 if (args->flags & XFSMNT_BARRIER)
671 xfs_mountfs_check_barriers(mp);
674 if (!(vfsp->vfs_flag & VFS_RDONLY) &&
675 (*flags & MS_RDONLY)) {
676 VFS_SYNC(vfsp, SYNC_FSDATA|SYNC_BDFLUSH|SYNC_ATTR, NULL, error);
680 /* Ok now write out an unmount record */
681 xfs_log_unmount_write(mp);
682 xfs_unmountfs_writesb(mp);
683 vfsp->vfs_flag |= VFS_RDONLY;
690 * xfs_unmount_flush implements a set of flush operation on special
691 * inodes, which are needed as a separate set of operations so that
692 * they can be called as part of relocation process.
696 xfs_mount_t *mp, /* Mount structure we are getting
698 int relocation) /* Called from vfs relocation. */
700 xfs_inode_t *rip = mp->m_rootip;
702 xfs_inode_t *rsumip = NULL;
703 vnode_t *rvp = XFS_ITOV(rip);
706 xfs_ilock(rip, XFS_ILOCK_EXCL);
710 * Flush out the real time inodes.
712 if ((rbmip = mp->m_rbmip) != NULL) {
713 xfs_ilock(rbmip, XFS_ILOCK_EXCL);
715 error = xfs_iflush(rbmip, XFS_IFLUSH_SYNC);
716 xfs_iunlock(rbmip, XFS_ILOCK_EXCL);
718 if (error == EFSCORRUPTED)
721 ASSERT(vn_count(XFS_ITOV(rbmip)) == 1);
723 rsumip = mp->m_rsumip;
724 xfs_ilock(rsumip, XFS_ILOCK_EXCL);
726 error = xfs_iflush(rsumip, XFS_IFLUSH_SYNC);
727 xfs_iunlock(rsumip, XFS_ILOCK_EXCL);
729 if (error == EFSCORRUPTED)
732 ASSERT(vn_count(XFS_ITOV(rsumip)) == 1);
736 * Synchronously flush root inode to disk
738 error = xfs_iflush(rip, XFS_IFLUSH_SYNC);
739 if (error == EFSCORRUPTED)
742 if (vn_count(rvp) != 1 && !relocation) {
743 xfs_iunlock(rip, XFS_ILOCK_EXCL);
744 return XFS_ERROR(EBUSY);
748 * Release dquot that rootinode, rbmino and rsumino might be holding,
749 * flush and purge the quota inodes.
751 error = XFS_QM_UNMOUNT(mp);
752 if (error == EFSCORRUPTED)
756 VN_RELE(XFS_ITOV(rbmip));
757 VN_RELE(XFS_ITOV(rsumip));
760 xfs_iunlock(rip, XFS_ILOCK_EXCL);
767 xfs_iunlock(rip, XFS_ILOCK_EXCL);
769 return XFS_ERROR(EFSCORRUPTED);
773 * xfs_root extracts the root vnode from a vfs.
775 * vfsp -- the vfs struct for the desired file system
776 * vpp -- address of the caller's vnode pointer which should be
777 * set to the desired fs root vnode
786 vp = XFS_ITOV((XFS_BHVTOM(bdp))->m_rootip);
795 * Fill in the statvfs structure for the given file system. We use
796 * the superblock lock in the mount structure to ensure a consistent
797 * snapshot of the counters returned.
811 mp = XFS_BHVTOM(bdp);
814 statp->f_type = XFS_SB_MAGIC;
817 statp->f_bsize = sbp->sb_blocksize;
818 lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
819 statp->f_blocks = sbp->sb_dblocks - lsize;
820 statp->f_bfree = statp->f_bavail = sbp->sb_fdblocks;
821 fakeinos = statp->f_bfree << sbp->sb_inopblog;
823 fakeinos += mp->m_inoadd;
826 MIN(sbp->sb_icount + fakeinos, (__uint64_t)XFS_MAXINUMBER);
831 statp->f_files = min_t(typeof(statp->f_files),
834 statp->f_ffree = statp->f_files - (sbp->sb_icount - sbp->sb_ifree);
835 XFS_SB_UNLOCK(mp, s);
837 xfs_statvfs_fsid(statp, mp);
838 statp->f_namelen = MAXNAMELEN - 1;
845 * xfs_sync flushes any pending I/O to file system vfsp.
847 * This routine is called by vfs_sync() to make sure that things make it
848 * out to disk eventually, on sync() system calls to flush out everything,
849 * and when the file system is unmounted. For the vfs_sync() case, all
850 * we really need to do is sync out the log to make all of our meta-data
851 * updates permanent (except for timestamps). For calls from pflushd(),
852 * dirty pages are kept moving by calling pdflush() on the inodes
853 * containing them. We also flush the inodes that we can lock without
854 * sleeping and the superblock if we can lock it without sleeping from
855 * vfs_sync() so that items at the tail of the log are always moving out.
858 * SYNC_BDFLUSH - We're being called from vfs_sync() so we don't want
859 * to sleep if we can help it. All we really need
860 * to do is ensure that the log is synced at least
861 * periodically. We also push the inodes and
862 * superblock if we can lock them without sleeping
863 * and they are not pinned.
864 * SYNC_ATTR - We need to flush the inodes. If SYNC_BDFLUSH is not
865 * set, then we really want to lock each inode and flush
867 * SYNC_WAIT - All the flushes that take place in this call should
869 * SYNC_DELWRI - This tells us to push dirty pages associated with
870 * inodes. SYNC_WAIT and SYNC_BDFLUSH are used to
871 * determine if they should be flushed sync, async, or
873 * SYNC_CLOSE - This flag is passed when the system is being
874 * unmounted. We should sync and invalidate everthing.
875 * SYNC_FSDATA - This indicates that the caller would like to make
876 * sure the superblock is safe on disk. We can ensure
877 * this by simply makeing sure the log gets flushed
878 * if SYNC_BDFLUSH is set, and by actually writing it
889 xfs_mount_t *mp = XFS_BHVTOM(bdp);
891 if (unlikely(flags == SYNC_QUIESCE))
892 return xfs_quiesce_fs(mp);
894 return xfs_syncsub(mp, flags, 0, NULL);
898 * xfs sync routine for internal use
900 * This routine supports all of the flags defined for the generic VFS_SYNC
901 * interface as explained above under xfs_sync. In the interests of not
902 * changing interfaces within the 6.5 family, additional internallly-
903 * required functions are specified within a separate xflags parameter,
904 * only available by calling this routine.
914 xfs_inode_t *ip = NULL;
915 xfs_inode_t *ip_next;
922 uint base_lock_flags;
923 boolean_t mount_locked;
924 boolean_t vnode_refed;
927 xfs_iptr_t *ipointer;
929 boolean_t ipointer_in = B_FALSE;
931 #define IPOINTER_SET ipointer_in = B_TRUE
932 #define IPOINTER_CLR ipointer_in = B_FALSE
939 /* Insert a marker record into the inode list after inode ip. The list
940 * must be locked when this is called. After the call the list will no
943 #define IPOINTER_INSERT(ip, mp) { \
944 ASSERT(ipointer_in == B_FALSE); \
945 ipointer->ip_mnext = ip->i_mnext; \
946 ipointer->ip_mprev = ip; \
947 ip->i_mnext = (xfs_inode_t *)ipointer; \
948 ipointer->ip_mnext->i_mprev = (xfs_inode_t *)ipointer; \
950 XFS_MOUNT_IUNLOCK(mp); \
951 mount_locked = B_FALSE; \
955 /* Remove the marker from the inode list. If the marker was the only item
956 * in the list then there are no remaining inodes and we should zero out
957 * the whole list. If we are the current head of the list then move the head
960 #define IPOINTER_REMOVE(ip, mp) { \
961 ASSERT(ipointer_in == B_TRUE); \
962 if (ipointer->ip_mnext != (xfs_inode_t *)ipointer) { \
963 ip = ipointer->ip_mnext; \
964 ip->i_mprev = ipointer->ip_mprev; \
965 ipointer->ip_mprev->i_mnext = ip; \
966 if (mp->m_inodes == (xfs_inode_t *)ipointer) { \
970 ASSERT(mp->m_inodes == (xfs_inode_t *)ipointer); \
971 mp->m_inodes = NULL; \
977 #define XFS_PREEMPT_MASK 0x7f
981 if (XFS_MTOVFS(mp)->vfs_flag & VFS_RDONLY)
987 /* Allocate a reference marker */
988 ipointer = (xfs_iptr_t *)kmem_zalloc(sizeof(xfs_iptr_t), KM_SLEEP);
990 fflag = XFS_B_ASYNC; /* default is don't wait */
991 if (flags & (SYNC_BDFLUSH | SYNC_DELWRI))
992 fflag = XFS_B_DELWRI;
993 if (flags & SYNC_WAIT)
994 fflag = 0; /* synchronous overrides all */
996 base_lock_flags = XFS_ILOCK_SHARED;
997 if (flags & (SYNC_DELWRI | SYNC_CLOSE)) {
999 * We need the I/O lock if we're going to call any of
1000 * the flush/inval routines.
1002 base_lock_flags |= XFS_IOLOCK_SHARED;
1005 XFS_MOUNT_ILOCK(mp);
1009 mount_locked = B_TRUE;
1010 vnode_refed = B_FALSE;
1015 ASSERT(ipointer_in == B_FALSE);
1016 ASSERT(vnode_refed == B_FALSE);
1018 lock_flags = base_lock_flags;
1021 * There were no inodes in the list, just break out
1029 * We found another sync thread marker - skip it
1031 if (ip->i_mount == NULL) {
1036 vp = XFS_ITOV_NULL(ip);
1039 * If the vnode is gone then this is being torn down,
1040 * call reclaim if it is flushed, else let regular flush
1041 * code deal with it later in the loop.
1045 /* Skip ones already in reclaim */
1046 if (ip->i_flags & XFS_IRECLAIM) {
1050 if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0) {
1052 } else if ((xfs_ipincount(ip) == 0) &&
1053 xfs_iflock_nowait(ip)) {
1054 IPOINTER_INSERT(ip, mp);
1056 xfs_finish_reclaim(ip, 1,
1057 XFS_IFLUSH_DELWRI_ELSE_ASYNC);
1059 XFS_MOUNT_ILOCK(mp);
1060 mount_locked = B_TRUE;
1061 IPOINTER_REMOVE(ip, mp);
1063 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1074 if (XFS_FORCED_SHUTDOWN(mp) && !(flags & SYNC_CLOSE)) {
1075 XFS_MOUNT_IUNLOCK(mp);
1076 kmem_free(ipointer, sizeof(xfs_iptr_t));
1081 * If this is just vfs_sync() or pflushd() calling
1082 * then we can skip inodes for which it looks like
1083 * there is nothing to do. Since we don't have the
1084 * inode locked this is racey, but these are periodic
1085 * calls so it doesn't matter. For the others we want
1086 * to know for sure, so we at least try to lock them.
1088 if (flags & SYNC_BDFLUSH) {
1089 if (((ip->i_itemp == NULL) ||
1090 !(ip->i_itemp->ili_format.ilf_fields &
1092 (ip->i_update_core == 0)) {
1099 * Try to lock without sleeping. We're out of order with
1100 * the inode list lock here, so if we fail we need to drop
1101 * the mount lock and try again. If we're called from
1102 * bdflush() here, then don't bother.
1104 * The inode lock here actually coordinates with the
1105 * almost spurious inode lock in xfs_ireclaim() to prevent
1106 * the vnode we handle here without a reference from
1107 * being freed while we reference it. If we lock the inode
1108 * while it's on the mount list here, then the spurious inode
1109 * lock in xfs_ireclaim() after the inode is pulled from
1110 * the mount list will sleep until we release it here.
1111 * This keeps the vnode from being freed while we reference
1114 if (xfs_ilock_nowait(ip, lock_flags) == 0) {
1115 if ((flags & SYNC_BDFLUSH) || (vp == NULL)) {
1126 IPOINTER_INSERT(ip, mp);
1127 xfs_ilock(ip, lock_flags);
1129 ASSERT(vp == XFS_ITOV(ip));
1130 ASSERT(ip->i_mount == mp);
1132 vnode_refed = B_TRUE;
1135 /* From here on in the loop we may have a marker record
1136 * in the inode list.
1139 if ((flags & SYNC_CLOSE) && (vp != NULL)) {
1141 * This is the shutdown case. We just need to
1142 * flush and invalidate all the pages associated
1143 * with the inode. Drop the inode lock since
1144 * we can't hold it across calls to the buffer
1147 * We don't set the VREMAPPING bit in the vnode
1148 * here, because we don't hold the vnode lock
1149 * exclusively. It doesn't really matter, though,
1150 * because we only come here when we're shutting
1153 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1155 if (XFS_FORCED_SHUTDOWN(mp)) {
1156 VOP_TOSS_PAGES(vp, 0, -1, FI_REMAPF);
1158 VOP_FLUSHINVAL_PAGES(vp, 0, -1, FI_REMAPF);
1161 xfs_ilock(ip, XFS_ILOCK_SHARED);
1163 } else if ((flags & SYNC_DELWRI) && (vp != NULL)) {
1165 /* We need to have dropped the lock here,
1166 * so insert a marker if we have not already
1170 IPOINTER_INSERT(ip, mp);
1174 * Drop the inode lock since we can't hold it
1175 * across calls to the buffer cache.
1177 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1178 VOP_FLUSH_PAGES(vp, (xfs_off_t)0, -1,
1179 fflag, FI_NONE, error);
1180 xfs_ilock(ip, XFS_ILOCK_SHARED);
1185 if (flags & SYNC_BDFLUSH) {
1186 if ((flags & SYNC_ATTR) &&
1187 ((ip->i_update_core) ||
1188 ((ip->i_itemp != NULL) &&
1189 (ip->i_itemp->ili_format.ilf_fields != 0)))) {
1191 /* Insert marker and drop lock if not already
1195 IPOINTER_INSERT(ip, mp);
1199 * We don't want the periodic flushing of the
1200 * inodes by vfs_sync() to interfere with
1201 * I/O to the file, especially read I/O
1202 * where it is only the access time stamp
1203 * that is being flushed out. To prevent
1204 * long periods where we have both inode
1205 * locks held shared here while reading the
1206 * inode's buffer in from disk, we drop the
1207 * inode lock while reading in the inode
1208 * buffer. We have to release the buffer
1209 * and reacquire the inode lock so that they
1210 * are acquired in the proper order (inode
1211 * locks first). The buffer will go at the
1212 * end of the lru chain, though, so we can
1213 * expect it to still be there when we go
1214 * for it again in xfs_iflush().
1216 if ((xfs_ipincount(ip) == 0) &&
1217 xfs_iflock_nowait(ip)) {
1220 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1222 error = xfs_itobp(mp, NULL, ip,
1227 /* Bailing out, remove the
1228 * marker and free it.
1230 XFS_MOUNT_ILOCK(mp);
1232 IPOINTER_REMOVE(ip, mp);
1234 XFS_MOUNT_IUNLOCK(mp);
1236 ASSERT(!(lock_flags &
1237 XFS_IOLOCK_SHARED));
1240 sizeof(xfs_iptr_t));
1245 * Since we dropped the inode lock,
1246 * the inode may have been reclaimed.
1247 * Therefore, we reacquire the mount
1248 * lock and check to see if we were the
1249 * inode reclaimed. If this happened
1250 * then the ipointer marker will no
1251 * longer point back at us. In this
1252 * case, move ip along to the inode
1253 * after the marker, remove the marker
1256 XFS_MOUNT_ILOCK(mp);
1257 mount_locked = B_TRUE;
1259 if (ip != ipointer->ip_mprev) {
1260 IPOINTER_REMOVE(ip, mp);
1262 ASSERT(!vnode_refed);
1263 ASSERT(!(lock_flags &
1264 XFS_IOLOCK_SHARED));
1268 ASSERT(ip->i_mount == mp);
1270 if (xfs_ilock_nowait(ip,
1271 XFS_ILOCK_SHARED) == 0) {
1272 ASSERT(ip->i_mount == mp);
1274 * We failed to reacquire
1275 * the inode lock without
1276 * sleeping, so just skip
1277 * the inode for now. We
1278 * clear the ILOCK bit from
1279 * the lock_flags so that we
1280 * won't try to drop a lock
1281 * we don't hold below.
1283 lock_flags &= ~XFS_ILOCK_SHARED;
1284 IPOINTER_REMOVE(ip_next, mp);
1285 } else if ((xfs_ipincount(ip) == 0) &&
1286 xfs_iflock_nowait(ip)) {
1287 ASSERT(ip->i_mount == mp);
1289 * Since this is vfs_sync()
1290 * calling we only flush the
1291 * inode out if we can lock
1292 * it without sleeping and
1293 * it is not pinned. Drop
1294 * the mount lock here so
1295 * that we don't hold it for
1296 * too long. We already have
1297 * a marker in the list here.
1299 XFS_MOUNT_IUNLOCK(mp);
1300 mount_locked = B_FALSE;
1301 error = xfs_iflush(ip,
1304 ASSERT(ip->i_mount == mp);
1305 IPOINTER_REMOVE(ip_next, mp);
1312 if ((flags & SYNC_ATTR) &&
1313 ((ip->i_update_core) ||
1314 ((ip->i_itemp != NULL) &&
1315 (ip->i_itemp->ili_format.ilf_fields != 0)))) {
1317 IPOINTER_INSERT(ip, mp);
1320 if (flags & SYNC_WAIT) {
1322 error = xfs_iflush(ip,
1326 * If we can't acquire the flush
1327 * lock, then the inode is already
1328 * being flushed so don't bother
1329 * waiting. If we can lock it then
1330 * do a delwri flush so we can
1331 * combine multiple inode flushes
1332 * in each disk write.
1334 if (xfs_iflock_nowait(ip)) {
1335 error = xfs_iflush(ip,
1344 if (lock_flags != 0) {
1345 xfs_iunlock(ip, lock_flags);
1350 * If we had to take a reference on the vnode
1351 * above, then wait until after we've unlocked
1352 * the inode to release the reference. This is
1353 * because we can be already holding the inode
1354 * lock when VN_RELE() calls xfs_inactive().
1356 * Make sure to drop the mount lock before calling
1357 * VN_RELE() so that we don't trip over ourselves if
1358 * we have to go for the mount lock again in the
1362 IPOINTER_INSERT(ip, mp);
1367 vnode_refed = B_FALSE;
1375 * bail out if the filesystem is corrupted.
1377 if (error == EFSCORRUPTED) {
1378 if (!mount_locked) {
1379 XFS_MOUNT_ILOCK(mp);
1380 IPOINTER_REMOVE(ip, mp);
1382 XFS_MOUNT_IUNLOCK(mp);
1383 ASSERT(ipointer_in == B_FALSE);
1384 kmem_free(ipointer, sizeof(xfs_iptr_t));
1385 return XFS_ERROR(error);
1388 /* Let other threads have a chance at the mount lock
1389 * if we have looped many times without dropping the
1392 if ((++preempt & XFS_PREEMPT_MASK) == 0) {
1394 IPOINTER_INSERT(ip, mp);
1398 if (mount_locked == B_FALSE) {
1399 XFS_MOUNT_ILOCK(mp);
1400 mount_locked = B_TRUE;
1401 IPOINTER_REMOVE(ip, mp);
1405 ASSERT(ipointer_in == B_FALSE);
1408 } while (ip != mp->m_inodes);
1410 XFS_MOUNT_IUNLOCK(mp);
1412 ASSERT(ipointer_in == B_FALSE);
1414 kmem_free(ipointer, sizeof(xfs_iptr_t));
1415 return XFS_ERROR(last_error);
1419 * xfs sync routine for internal use
1421 * This routine supports all of the flags defined for the generic VFS_SYNC
1422 * interface as explained above under xfs_sync. In the interests of not
1423 * changing interfaces within the 6.5 family, additional internallly-
1424 * required functions are specified within a separate xflags parameter,
1425 * only available by calling this routine.
1437 uint log_flags = XFS_LOG_FORCE;
1439 xfs_buf_log_item_t *bip;
1442 * Sync out the log. This ensures that the log is periodically
1443 * flushed even if there is not enough activity to fill it up.
1445 if (flags & SYNC_WAIT)
1446 log_flags |= XFS_LOG_SYNC;
1448 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1450 if (flags & (SYNC_ATTR|SYNC_DELWRI)) {
1451 if (flags & SYNC_BDFLUSH)
1452 xfs_finish_reclaim_all(mp, 1);
1454 error = xfs_sync_inodes(mp, flags, xflags, bypassed);
1458 * Flushing out dirty data above probably generated more
1459 * log activity, so if this isn't vfs_sync() then flush
1462 if (flags & SYNC_DELWRI) {
1463 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1466 if (flags & SYNC_FSDATA) {
1468 * If this is vfs_sync() then only sync the superblock
1469 * if we can lock it without sleeping and it is not pinned.
1471 if (flags & SYNC_BDFLUSH) {
1472 bp = xfs_getsb(mp, XFS_BUF_TRYLOCK);
1474 bip = XFS_BUF_FSPRIVATE(bp,xfs_buf_log_item_t*);
1475 if ((bip != NULL) &&
1476 xfs_buf_item_dirty(bip)) {
1477 if (!(XFS_BUF_ISPINNED(bp))) {
1479 error = xfs_bwrite(mp, bp);
1488 bp = xfs_getsb(mp, 0);
1490 * If the buffer is pinned then push on the log so
1491 * we won't get stuck waiting in the write for
1492 * someone, maybe ourselves, to flush the log.
1493 * Even though we just pushed the log above, we
1494 * did not have the superblock buffer locked at
1495 * that point so it can become pinned in between
1498 if (XFS_BUF_ISPINNED(bp))
1499 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
1500 if (flags & SYNC_WAIT)
1501 XFS_BUF_UNASYNC(bp);
1504 error = xfs_bwrite(mp, bp);
1512 * If this is the periodic sync, then kick some entries out of
1513 * the reference cache. This ensures that idle entries are
1514 * eventually kicked out of the cache.
1516 if (flags & SYNC_REFCACHE) {
1517 if (flags & SYNC_WAIT)
1518 xfs_refcache_purge_mp(mp);
1520 xfs_refcache_purge_some(mp);
1524 * Now check to see if the log needs a "dummy" transaction.
1527 if (!(flags & SYNC_REMOUNT) && xfs_log_need_covered(mp)) {
1532 * Put a dummy transaction in the log to tell
1533 * recovery that all others are OK.
1535 tp = xfs_trans_alloc(mp, XFS_TRANS_DUMMY1);
1536 if ((error = xfs_trans_reserve(tp, 0,
1537 XFS_ICHANGE_LOG_RES(mp),
1539 xfs_trans_cancel(tp, 0);
1544 xfs_ilock(ip, XFS_ILOCK_EXCL);
1546 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1547 xfs_trans_ihold(tp, ip);
1548 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1549 error = xfs_trans_commit(tp, 0, NULL);
1550 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1551 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1555 * When shutting down, we need to insure that the AIL is pushed
1556 * to disk or the filesystem can appear corrupt from the PROM.
1558 if ((flags & (SYNC_CLOSE|SYNC_WAIT)) == (SYNC_CLOSE|SYNC_WAIT)) {
1559 XFS_bflush(mp->m_ddev_targp);
1560 if (mp->m_rtdev_targp) {
1561 XFS_bflush(mp->m_rtdev_targp);
1565 return XFS_ERROR(last_error);
1569 * xfs_vget - called by DMAPI and NFSD to get vnode from file handle
1577 xfs_mount_t *mp = XFS_BHVTOM(bdp);
1578 xfs_fid_t *xfid = (struct xfs_fid *)fidp;
1585 * Invalid. Since handles can be created in user space and passed in
1586 * via gethandle(), this is not cause for a panic.
1588 if (xfid->xfs_fid_len != sizeof(*xfid) - sizeof(xfid->xfs_fid_len))
1589 return XFS_ERROR(EINVAL);
1591 ino = xfid->xfs_fid_ino;
1592 igen = xfid->xfs_fid_gen;
1595 * NFS can sometimes send requests for ino 0. Fail them gracefully.
1598 return XFS_ERROR(ESTALE);
1600 error = xfs_iget(mp, NULL, ino, 0, XFS_ILOCK_SHARED, &ip, 0);
1608 return XFS_ERROR(EIO);
1611 if (ip->i_d.di_mode == 0 || ip->i_d.di_gen != igen) {
1612 xfs_iput_new(ip, XFS_ILOCK_SHARED);
1614 return XFS_ERROR(ENOENT);
1617 *vpp = XFS_ITOV(ip);
1618 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1623 #define MNTOPT_LOGBUFS "logbufs" /* number of XFS log buffers */
1624 #define MNTOPT_LOGBSIZE "logbsize" /* size of XFS log buffers */
1625 #define MNTOPT_LOGDEV "logdev" /* log device */
1626 #define MNTOPT_RTDEV "rtdev" /* realtime I/O device */
1627 #define MNTOPT_BIOSIZE "biosize" /* log2 of preferred buffered io size */
1628 #define MNTOPT_WSYNC "wsync" /* safe-mode nfs compatible mount */
1629 #define MNTOPT_INO64 "ino64" /* force inodes into 64-bit range */
1630 #define MNTOPT_NOALIGN "noalign" /* turn off stripe alignment */
1631 #define MNTOPT_SWALLOC "swalloc" /* turn on stripe width allocation */
1632 #define MNTOPT_SUNIT "sunit" /* data volume stripe unit */
1633 #define MNTOPT_SWIDTH "swidth" /* data volume stripe width */
1634 #define MNTOPT_NOUUID "nouuid" /* ignore filesystem UUID */
1635 #define MNTOPT_MTPT "mtpt" /* filesystem mount point */
1636 #define MNTOPT_GRPID "grpid" /* group-ID from parent directory */
1637 #define MNTOPT_NOGRPID "nogrpid" /* group-ID from current process */
1638 #define MNTOPT_BSDGROUPS "bsdgroups" /* group-ID from parent directory */
1639 #define MNTOPT_SYSVGROUPS "sysvgroups" /* group-ID from current process */
1640 #define MNTOPT_ALLOCSIZE "allocsize" /* preferred allocation size */
1641 #define MNTOPT_IHASHSIZE "ihashsize" /* size of inode hash table */
1642 #define MNTOPT_NORECOVERY "norecovery" /* don't run XFS recovery */
1643 #define MNTOPT_BARRIER "barrier" /* use writer barriers for log write and
1644 * unwritten extent conversion */
1645 #define MNTOPT_NOBARRIER "nobarrier" /* .. disable */
1646 #define MNTOPT_OSYNCISOSYNC "osyncisosync" /* o_sync is REALLY o_sync */
1647 #define MNTOPT_64BITINODE "inode64" /* inodes can be allocated anywhere */
1648 #define MNTOPT_IKEEP "ikeep" /* do not free empty inode clusters */
1649 #define MNTOPT_NOIKEEP "noikeep" /* free empty inode clusters */
1650 #define MNTOPT_LARGEIO "largeio" /* report large I/O sizes in stat() */
1651 #define MNTOPT_NOLARGEIO "nolargeio" /* do not report large I/O sizes
1653 #define MNTOPT_ATTR2 "attr2" /* do use attr2 attribute format */
1654 #define MNTOPT_NOATTR2 "noattr2" /* do not use attr2 attribute format */
1656 STATIC unsigned long
1657 suffix_strtoul(const char *cp, char **endp, unsigned int base)
1659 int last, shift_left_factor = 0;
1660 char *value = (char *)cp;
1662 last = strlen(value) - 1;
1663 if (value[last] == 'K' || value[last] == 'k') {
1664 shift_left_factor = 10;
1667 if (value[last] == 'M' || value[last] == 'm') {
1668 shift_left_factor = 20;
1671 if (value[last] == 'G' || value[last] == 'g') {
1672 shift_left_factor = 30;
1676 return simple_strtoul(cp, endp, base) << shift_left_factor;
1681 struct bhv_desc *bhv,
1683 struct xfs_mount_args *args,
1686 struct vfs *vfsp = bhvtovfs(bhv);
1687 char *this_char, *value, *eov;
1688 int dsunit, dswidth, vol_dsunit, vol_dswidth;
1691 args->flags2 |= XFSMNT2_COMPAT_IOSIZE;
1693 #if 0 /* XXX: off by default, until some remaining issues ironed out */
1694 args->flags |= XFSMNT_IDELETE; /* default to on */
1700 iosize = dsunit = dswidth = vol_dsunit = vol_dswidth = 0;
1702 while ((this_char = strsep(&options, ",")) != NULL) {
1705 if ((value = strchr(this_char, '=')) != NULL)
1708 if (!strcmp(this_char, MNTOPT_LOGBUFS)) {
1709 if (!value || !*value) {
1710 printk("XFS: %s option requires an argument\n",
1714 args->logbufs = simple_strtoul(value, &eov, 10);
1715 } else if (!strcmp(this_char, MNTOPT_LOGBSIZE)) {
1716 if (!value || !*value) {
1717 printk("XFS: %s option requires an argument\n",
1721 args->logbufsize = suffix_strtoul(value, &eov, 10);
1722 } else if (!strcmp(this_char, MNTOPT_LOGDEV)) {
1723 if (!value || !*value) {
1724 printk("XFS: %s option requires an argument\n",
1728 strncpy(args->logname, value, MAXNAMELEN);
1729 } else if (!strcmp(this_char, MNTOPT_MTPT)) {
1730 if (!value || !*value) {
1731 printk("XFS: %s option requires an argument\n",
1735 strncpy(args->mtpt, value, MAXNAMELEN);
1736 } else if (!strcmp(this_char, MNTOPT_RTDEV)) {
1737 if (!value || !*value) {
1738 printk("XFS: %s option requires an argument\n",
1742 strncpy(args->rtname, value, MAXNAMELEN);
1743 } else if (!strcmp(this_char, MNTOPT_BIOSIZE)) {
1744 if (!value || !*value) {
1745 printk("XFS: %s option requires an argument\n",
1749 iosize = simple_strtoul(value, &eov, 10);
1750 args->flags |= XFSMNT_IOSIZE;
1751 args->iosizelog = (uint8_t) iosize;
1752 } else if (!strcmp(this_char, MNTOPT_ALLOCSIZE)) {
1753 if (!value || !*value) {
1754 printk("XFS: %s option requires an argument\n",
1758 iosize = suffix_strtoul(value, &eov, 10);
1759 args->flags |= XFSMNT_IOSIZE;
1760 args->iosizelog = ffs(iosize) - 1;
1761 } else if (!strcmp(this_char, MNTOPT_IHASHSIZE)) {
1762 if (!value || !*value) {
1763 printk("XFS: %s option requires an argument\n",
1767 args->flags |= XFSMNT_IHASHSIZE;
1768 args->ihashsize = simple_strtoul(value, &eov, 10);
1769 } else if (!strcmp(this_char, MNTOPT_GRPID) ||
1770 !strcmp(this_char, MNTOPT_BSDGROUPS)) {
1771 vfsp->vfs_flag |= VFS_GRPID;
1772 } else if (!strcmp(this_char, MNTOPT_NOGRPID) ||
1773 !strcmp(this_char, MNTOPT_SYSVGROUPS)) {
1774 vfsp->vfs_flag &= ~VFS_GRPID;
1775 } else if (!strcmp(this_char, MNTOPT_WSYNC)) {
1776 args->flags |= XFSMNT_WSYNC;
1777 } else if (!strcmp(this_char, MNTOPT_OSYNCISOSYNC)) {
1778 args->flags |= XFSMNT_OSYNCISOSYNC;
1779 } else if (!strcmp(this_char, MNTOPT_NORECOVERY)) {
1780 args->flags |= XFSMNT_NORECOVERY;
1781 } else if (!strcmp(this_char, MNTOPT_INO64)) {
1782 args->flags |= XFSMNT_INO64;
1784 printk("XFS: %s option not allowed on this system\n",
1788 } else if (!strcmp(this_char, MNTOPT_NOALIGN)) {
1789 args->flags |= XFSMNT_NOALIGN;
1790 } else if (!strcmp(this_char, MNTOPT_SWALLOC)) {
1791 args->flags |= XFSMNT_SWALLOC;
1792 } else if (!strcmp(this_char, MNTOPT_SUNIT)) {
1793 if (!value || !*value) {
1794 printk("XFS: %s option requires an argument\n",
1798 dsunit = simple_strtoul(value, &eov, 10);
1799 } else if (!strcmp(this_char, MNTOPT_SWIDTH)) {
1800 if (!value || !*value) {
1801 printk("XFS: %s option requires an argument\n",
1805 dswidth = simple_strtoul(value, &eov, 10);
1806 } else if (!strcmp(this_char, MNTOPT_64BITINODE)) {
1807 args->flags &= ~XFSMNT_32BITINODES;
1809 printk("XFS: %s option not allowed on this system\n",
1813 } else if (!strcmp(this_char, MNTOPT_NOUUID)) {
1814 args->flags |= XFSMNT_NOUUID;
1815 } else if (!strcmp(this_char, MNTOPT_BARRIER)) {
1816 args->flags |= XFSMNT_BARRIER;
1817 } else if (!strcmp(this_char, MNTOPT_NOBARRIER)) {
1818 args->flags &= ~XFSMNT_BARRIER;
1819 } else if (!strcmp(this_char, MNTOPT_IKEEP)) {
1820 args->flags &= ~XFSMNT_IDELETE;
1821 } else if (!strcmp(this_char, MNTOPT_NOIKEEP)) {
1822 args->flags |= XFSMNT_IDELETE;
1823 } else if (!strcmp(this_char, MNTOPT_LARGEIO)) {
1824 args->flags2 &= ~XFSMNT2_COMPAT_IOSIZE;
1825 } else if (!strcmp(this_char, MNTOPT_NOLARGEIO)) {
1826 args->flags2 |= XFSMNT2_COMPAT_IOSIZE;
1827 } else if (!strcmp(this_char, MNTOPT_ATTR2)) {
1828 args->flags |= XFSMNT_ATTR2;
1829 } else if (!strcmp(this_char, MNTOPT_NOATTR2)) {
1830 args->flags &= ~XFSMNT_ATTR2;
1831 } else if (!strcmp(this_char, "osyncisdsync")) {
1832 /* no-op, this is now the default */
1833 printk("XFS: osyncisdsync is now the default, option is deprecated.\n");
1834 } else if (!strcmp(this_char, "irixsgid")) {
1835 printk("XFS: irixsgid is now a sysctl(2) variable, option is deprecated.\n");
1837 printk("XFS: unknown mount option [%s].\n", this_char);
1842 if (args->flags & XFSMNT_NORECOVERY) {
1843 if ((vfsp->vfs_flag & VFS_RDONLY) == 0) {
1844 printk("XFS: no-recovery mounts must be read-only.\n");
1849 if ((args->flags & XFSMNT_NOALIGN) && (dsunit || dswidth)) {
1851 "XFS: sunit and swidth options incompatible with the noalign option\n");
1855 if ((dsunit && !dswidth) || (!dsunit && dswidth)) {
1856 printk("XFS: sunit and swidth must be specified together\n");
1860 if (dsunit && (dswidth % dsunit != 0)) {
1862 "XFS: stripe width (%d) must be a multiple of the stripe unit (%d)\n",
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 },
1905 { XFS_MOUNT_IDELETE, "," MNTOPT_NOIKEEP },
1908 struct proc_xfs_info *xfs_infop;
1909 struct xfs_mount *mp = XFS_BHVTOM(bhv);
1910 struct vfs *vfsp = XFS_MTOVFS(mp);
1912 for (xfs_infop = xfs_info; xfs_infop->flag; xfs_infop++) {
1913 if (mp->m_flags & xfs_infop->flag)
1914 seq_puts(m, xfs_infop->str);
1917 if (mp->m_flags & XFS_MOUNT_IHASHSIZE)
1918 seq_printf(m, "," MNTOPT_IHASHSIZE "=%d", mp->m_ihsize);
1920 if (mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)
1921 seq_printf(m, "," MNTOPT_ALLOCSIZE "=%dk",
1922 (int)(1 << mp->m_writeio_log) >> 10);
1924 if (mp->m_logbufs > 0)
1925 seq_printf(m, "," MNTOPT_LOGBUFS "=%d", mp->m_logbufs);
1926 if (mp->m_logbsize > 0)
1927 seq_printf(m, "," MNTOPT_LOGBSIZE "=%dk", mp->m_logbsize >> 10);
1930 seq_printf(m, "," MNTOPT_LOGDEV "=%s", mp->m_logname);
1932 seq_printf(m, "," MNTOPT_RTDEV "=%s", mp->m_rtname);
1934 if (mp->m_dalign > 0)
1935 seq_printf(m, "," MNTOPT_SUNIT "=%d",
1936 (int)XFS_FSB_TO_BB(mp, mp->m_dalign));
1937 if (mp->m_swidth > 0)
1938 seq_printf(m, "," MNTOPT_SWIDTH "=%d",
1939 (int)XFS_FSB_TO_BB(mp, mp->m_swidth));
1941 if (!(mp->m_flags & XFS_MOUNT_COMPAT_IOSIZE))
1942 seq_printf(m, "," MNTOPT_LARGEIO);
1943 if (mp->m_flags & XFS_MOUNT_BARRIER)
1944 seq_printf(m, "," MNTOPT_BARRIER);
1946 if (!(vfsp->vfs_flag & VFS_32BITINODES))
1947 seq_printf(m, "," MNTOPT_64BITINODE);
1948 if (vfsp->vfs_flag & VFS_GRPID)
1949 seq_printf(m, "," MNTOPT_GRPID);
1958 xfs_mount_t *mp = XFS_BHVTOM(bdp);
1960 while (atomic_read(&mp->m_active_trans) > 0)
1963 /* Push the superblock and write an unmount record */
1964 xfs_log_unmount_write(mp);
1965 xfs_unmountfs_writesb(mp);
1966 xfs_fs_log_dummy(mp);
1970 vfsops_t xfs_vfsops = {
1971 BHV_IDENTITY_INIT(VFS_BHV_XFS,VFS_POSITION_XFS),
1972 .vfs_parseargs = xfs_parseargs,
1973 .vfs_showargs = xfs_showargs,
1974 .vfs_mount = xfs_mount,
1975 .vfs_unmount = xfs_unmount,
1976 .vfs_mntupdate = xfs_mntupdate,
1977 .vfs_root = xfs_root,
1978 .vfs_statvfs = xfs_statvfs,
1979 .vfs_sync = xfs_sync,
1980 .vfs_vget = xfs_vget,
1981 .vfs_dmapiops = (vfs_dmapiops_t)fs_nosys,
1982 .vfs_quotactl = (vfs_quotactl_t)fs_nosys,
1983 .vfs_init_vnode = xfs_initialize_vnode,
1984 .vfs_force_shutdown = xfs_do_force_shutdown,
1985 .vfs_freeze = xfs_freeze,