2 * Copyright (c) 1989, 1991, 1993, 1994
3 * The Regents of the University of California. All rights reserved.
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 3. Neither the name of the University nor the names of its contributors
14 * may be used to endorse or promote products derived from this software
15 * without specific prior written permission.
17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 * @(#)ffs_vfsops.c 8.31 (Berkeley) 5/20/95
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
35 #include "opt_quota.h"
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/namei.h>
45 #include <sys/taskqueue.h>
46 #include <sys/kernel.h>
47 #include <sys/vnode.h>
48 #include <sys/mount.h>
52 #include <sys/fcntl.h>
53 #include <sys/ioccom.h>
54 #include <sys/malloc.h>
55 #include <sys/mutex.h>
56 #include <sys/rwlock.h>
58 #include <security/mac/mac_framework.h>
60 #include <ufs/ufs/extattr.h>
61 #include <ufs/ufs/gjournal.h>
62 #include <ufs/ufs/quota.h>
63 #include <ufs/ufs/ufsmount.h>
64 #include <ufs/ufs/inode.h>
65 #include <ufs/ufs/ufs_extern.h>
67 #include <ufs/ffs/fs.h>
68 #include <ufs/ffs/ffs_extern.h>
72 #include <vm/vm_page.h>
74 #include <geom/geom.h>
75 #include <geom/geom_vfs.h>
79 static uma_zone_t uma_inode, uma_ufs1, uma_ufs2;
81 static int ffs_mountfs(struct vnode *, struct mount *, struct thread *);
82 static void ffs_oldfscompat_read(struct fs *, struct ufsmount *,
84 static void ffs_ifree(struct ufsmount *ump, struct inode *ip);
85 static int ffs_sync_lazy(struct mount *mp);
87 static vfs_init_t ffs_init;
88 static vfs_uninit_t ffs_uninit;
89 static vfs_extattrctl_t ffs_extattrctl;
90 static vfs_cmount_t ffs_cmount;
91 static vfs_unmount_t ffs_unmount;
92 static vfs_mount_t ffs_mount;
93 static vfs_statfs_t ffs_statfs;
94 static vfs_fhtovp_t ffs_fhtovp;
95 static vfs_sync_t ffs_sync;
97 static struct vfsops ufs_vfsops = {
98 .vfs_extattrctl = ffs_extattrctl,
99 .vfs_fhtovp = ffs_fhtovp,
100 .vfs_init = ffs_init,
101 .vfs_mount = ffs_mount,
102 .vfs_cmount = ffs_cmount,
103 .vfs_quotactl = ufs_quotactl,
104 .vfs_root = ufs_root,
105 .vfs_statfs = ffs_statfs,
106 .vfs_sync = ffs_sync,
107 .vfs_uninit = ffs_uninit,
108 .vfs_unmount = ffs_unmount,
109 .vfs_vget = ffs_vget,
110 .vfs_susp_clean = process_deferred_inactive,
113 VFS_SET(ufs_vfsops, ufs, 0);
114 MODULE_VERSION(ufs, 1);
116 static b_strategy_t ffs_geom_strategy;
117 static b_write_t ffs_bufwrite;
119 static struct buf_ops ffs_ops = {
121 .bop_write = ffs_bufwrite,
122 .bop_strategy = ffs_geom_strategy,
124 #ifdef NO_FFS_SNAPSHOT
125 .bop_bdflush = bufbdflush,
127 .bop_bdflush = ffs_bdflush,
132 * Note that userquota and groupquota options are not currently used
133 * by UFS/FFS code and generally mount(8) does not pass those options
134 * from userland, but they can be passed by loader(8) via
135 * vfs.root.mountfrom.options.
137 static const char *ffs_opts[] = { "acls", "async", "noatime", "noclusterr",
138 "noclusterw", "noexec", "export", "force", "from", "groupquota",
139 "multilabel", "nfsv4acls", "fsckpid", "snapshot", "nosuid", "suiddir",
140 "nosymfollow", "sync", "union", "userquota", NULL };
143 ffs_mount(struct mount *mp)
147 struct ufsmount *ump = NULL;
150 int error, error1, flags;
153 struct nameidata ndp;
157 if (vfs_filteropt(mp->mnt_optnew, ffs_opts))
159 if (uma_inode == NULL) {
160 uma_inode = uma_zcreate("FFS inode",
161 sizeof(struct inode), NULL, NULL, NULL, NULL,
163 uma_ufs1 = uma_zcreate("FFS1 dinode",
164 sizeof(struct ufs1_dinode), NULL, NULL, NULL, NULL,
166 uma_ufs2 = uma_zcreate("FFS2 dinode",
167 sizeof(struct ufs2_dinode), NULL, NULL, NULL, NULL,
171 vfs_deleteopt(mp->mnt_optnew, "groupquota");
172 vfs_deleteopt(mp->mnt_optnew, "userquota");
174 fspec = vfs_getopts(mp->mnt_optnew, "from", &error);
179 if (vfs_getopt(mp->mnt_optnew, "acls", NULL, NULL) == 0)
180 mntorflags |= MNT_ACLS;
182 if (vfs_getopt(mp->mnt_optnew, "snapshot", NULL, NULL) == 0) {
183 mntorflags |= MNT_SNAPSHOT;
185 * Once we have set the MNT_SNAPSHOT flag, do not
186 * persist "snapshot" in the options list.
188 vfs_deleteopt(mp->mnt_optnew, "snapshot");
189 vfs_deleteopt(mp->mnt_opt, "snapshot");
192 if (vfs_getopt(mp->mnt_optnew, "fsckpid", NULL, NULL) == 0 &&
193 vfs_scanopt(mp->mnt_optnew, "fsckpid", "%d", &fsckpid) == 1) {
195 * Once we have set the restricted PID, do not
196 * persist "fsckpid" in the options list.
198 vfs_deleteopt(mp->mnt_optnew, "fsckpid");
199 vfs_deleteopt(mp->mnt_opt, "fsckpid");
200 if (mp->mnt_flag & MNT_UPDATE) {
201 if (VFSTOUFS(mp)->um_fs->fs_ronly == 0 &&
202 vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0) == 0) {
204 "Checker enable: Must be read-only");
207 } else if (vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0) == 0) {
209 "Checker enable: Must be read-only");
212 /* Set to -1 if we are done */
217 if (vfs_getopt(mp->mnt_optnew, "nfsv4acls", NULL, NULL) == 0) {
218 if (mntorflags & MNT_ACLS) {
220 "\"acls\" and \"nfsv4acls\" options "
221 "are mutually exclusive");
224 mntorflags |= MNT_NFS4ACLS;
228 mp->mnt_flag |= mntorflags;
231 * If updating, check whether changing from read-only to
232 * read/write; if there is no device name, that's all we do.
234 if (mp->mnt_flag & MNT_UPDATE) {
237 devvp = ump->um_devvp;
238 if (fsckpid == -1 && ump->um_fsckpid > 0) {
239 if ((error = ffs_flushfiles(mp, WRITECLOSE, td)) != 0 ||
240 (error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0)
244 * Return to normal read-only mode.
246 error = g_access(ump->um_cp, 0, -1, 0);
250 if (fs->fs_ronly == 0 &&
251 vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0)) {
253 * Flush any dirty data and suspend filesystem.
255 if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0)
257 error = vfs_write_suspend_umnt(mp);
261 * Check for and optionally get rid of files open
265 if (mp->mnt_flag & MNT_FORCE)
267 if (MOUNTEDSOFTDEP(mp)) {
268 error = softdep_flushfiles(mp, flags, td);
270 error = ffs_flushfiles(mp, flags, td);
273 vfs_write_resume(mp, 0);
276 if (fs->fs_pendingblocks != 0 ||
277 fs->fs_pendinginodes != 0) {
278 printf("WARNING: %s Update error: blocks %jd "
279 "files %d\n", fs->fs_fsmnt,
280 (intmax_t)fs->fs_pendingblocks,
281 fs->fs_pendinginodes);
282 fs->fs_pendingblocks = 0;
283 fs->fs_pendinginodes = 0;
285 if ((fs->fs_flags & (FS_UNCLEAN | FS_NEEDSFSCK)) == 0)
287 if ((error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0) {
290 vfs_write_resume(mp, 0);
293 if (MOUNTEDSOFTDEP(mp))
297 * Drop our write and exclusive access.
299 g_access(ump->um_cp, 0, -1, -1);
303 mp->mnt_flag |= MNT_RDONLY;
306 * Allow the writers to note that filesystem
309 vfs_write_resume(mp, 0);
311 if ((mp->mnt_flag & MNT_RELOAD) &&
312 (error = ffs_reload(mp, td, 0)) != 0)
315 !vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0)) {
317 * If we are running a checker, do not allow upgrade.
319 if (ump->um_fsckpid > 0) {
321 "Active checker, cannot upgrade to write");
325 * If upgrade to read-write by non-root, then verify
326 * that user has necessary permissions on the device.
328 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
329 error = VOP_ACCESS(devvp, VREAD | VWRITE,
332 error = priv_check(td, PRIV_VFS_MOUNT_PERM);
334 VOP_UNLOCK(devvp, 0);
337 VOP_UNLOCK(devvp, 0);
338 fs->fs_flags &= ~FS_UNCLEAN;
339 if (fs->fs_clean == 0) {
340 fs->fs_flags |= FS_UNCLEAN;
341 if ((mp->mnt_flag & MNT_FORCE) ||
343 (FS_SUJ | FS_NEEDSFSCK)) == 0 &&
344 (fs->fs_flags & FS_DOSOFTDEP))) {
345 printf("WARNING: %s was not properly "
346 "dismounted\n", fs->fs_fsmnt);
349 "R/W mount of %s denied. %s.%s",
351 "Filesystem is not clean - run fsck",
352 (fs->fs_flags & FS_SUJ) == 0 ? "" :
353 " Forced mount will invalidate"
354 " journal contents");
360 * Request exclusive write access.
362 error = g_access(ump->um_cp, 0, 1, 1);
366 if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0)
370 mp->mnt_flag &= ~MNT_RDONLY;
372 fs->fs_mtime = time_second;
373 /* check to see if we need to start softdep */
374 if ((fs->fs_flags & FS_DOSOFTDEP) &&
375 (error = softdep_mount(devvp, mp, fs, td->td_ucred))){
376 vn_finished_write(mp);
380 if ((error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0) {
381 vn_finished_write(mp);
384 if (fs->fs_snapinum[0] != 0)
385 ffs_snapshot_mount(mp);
386 vn_finished_write(mp);
389 * Soft updates is incompatible with "async",
390 * so if we are doing softupdates stop the user
391 * from setting the async flag in an update.
392 * Softdep_mount() clears it in an initial mount
395 if (MOUNTEDSOFTDEP(mp)) {
396 /* XXX: Reset too late ? */
398 mp->mnt_flag &= ~MNT_ASYNC;
402 * Keep MNT_ACLS flag if it is stored in superblock.
404 if ((fs->fs_flags & FS_ACLS) != 0) {
405 /* XXX: Set too late ? */
407 mp->mnt_flag |= MNT_ACLS;
411 if ((fs->fs_flags & FS_NFS4ACLS) != 0) {
412 /* XXX: Set too late ? */
414 mp->mnt_flag |= MNT_NFS4ACLS;
418 * If this is a request from fsck to clean up the filesystem,
419 * then allow the specified pid to proceed.
422 if (ump->um_fsckpid != 0) {
424 "Active checker already running on %s",
428 KASSERT(MOUNTEDSOFTDEP(mp) == 0,
429 ("soft updates enabled on read-only file system"));
432 * Request write access.
434 error = g_access(ump->um_cp, 0, 1, 0);
438 "Checker activation failed on %s",
442 ump->um_fsckpid = fsckpid;
443 if (fs->fs_snapinum[0] != 0)
444 ffs_snapshot_mount(mp);
445 fs->fs_mtime = time_second;
448 (void) ffs_sbupdate(ump, MNT_WAIT, 0);
452 * If this is a snapshot request, take the snapshot.
454 if (mp->mnt_flag & MNT_SNAPSHOT)
455 return (ffs_snapshot(mp, fspec));
458 * Must not call namei() while owning busy ref.
464 * Not an update, or updating the name: look up the name
465 * and verify that it refers to a sensible disk device.
467 NDINIT(&ndp, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, fspec, td);
469 if ((mp->mnt_flag & MNT_UPDATE) != 0) {
471 * Unmount does not start if MNT_UPDATE is set. Mount
472 * update busies mp before setting MNT_UPDATE. We
473 * must be able to retain our busy ref succesfully,
476 error1 = vfs_busy(mp, MBF_NOWAIT);
481 NDFREE(&ndp, NDF_ONLY_PNBUF);
483 if (!vn_isdisk(devvp, &error)) {
489 * If mount by non-root, then verify that user has necessary
490 * permissions on the device.
493 if ((mp->mnt_flag & MNT_RDONLY) == 0)
495 error = VOP_ACCESS(devvp, accmode, td->td_ucred, td);
497 error = priv_check(td, PRIV_VFS_MOUNT_PERM);
503 if (mp->mnt_flag & MNT_UPDATE) {
507 * If it's not the same vnode, or at least the same device
508 * then it's not correct.
511 if (devvp->v_rdev != ump->um_devvp->v_rdev)
512 error = EINVAL; /* needs translation */
520 * We need the name for the mount point (also used for
521 * "last mounted on") copied in. If an error occurs,
522 * the mount point is discarded by the upper level code.
523 * Note that vfs_mount_alloc() populates f_mntonname for us.
525 if ((error = ffs_mountfs(devvp, mp, td)) != 0) {
530 KASSERT(MOUNTEDSOFTDEP(mp) == 0,
531 ("soft updates enabled on read-only file system"));
536 * Request write access.
538 error = g_access(ump->um_cp, 0, 1, 0);
541 printf("WARNING: %s: Checker activation "
542 "failed\n", fs->fs_fsmnt);
544 ump->um_fsckpid = fsckpid;
545 if (fs->fs_snapinum[0] != 0)
546 ffs_snapshot_mount(mp);
547 fs->fs_mtime = time_second;
549 (void) ffs_sbupdate(ump, MNT_WAIT, 0);
553 vfs_mountedfrom(mp, fspec);
558 * Compatibility with old mount system call.
562 ffs_cmount(struct mntarg *ma, void *data, uint64_t flags)
564 struct ufs_args args;
565 struct export_args exp;
570 error = copyin(data, &args, sizeof args);
573 vfs_oexport_conv(&args.export, &exp);
575 ma = mount_argsu(ma, "from", args.fspec, MAXPATHLEN);
576 ma = mount_arg(ma, "export", &exp, sizeof(exp));
577 error = kernel_mount(ma, flags);
583 * Reload all incore data for a filesystem (used after running fsck on
584 * the root filesystem and finding things to fix). If the 'force' flag
585 * is 0, the filesystem must be mounted read-only.
587 * Things to do to update the mount:
588 * 1) invalidate all cached meta-data.
589 * 2) re-read superblock from disk.
590 * 3) re-read summary information from disk.
591 * 4) invalidate all inactive vnodes.
592 * 5) clear MNTK_SUSPEND2 and MNTK_SUSPENDED flags, allowing secondary
593 * writers, if requested.
594 * 6) invalidate all cached file data.
595 * 7) re-read inode data for all active vnodes.
598 ffs_reload(struct mount *mp, struct thread *td, int flags)
600 struct vnode *vp, *mvp, *devvp;
604 struct fs *fs, *newfs;
605 struct ufsmount *ump;
606 ufs2_daddr_t sblockloc;
614 if ((mp->mnt_flag & MNT_RDONLY) == 0 && (flags & FFSR_FORCE) == 0) {
621 * Step 1: invalidate all cached meta-data.
623 devvp = VFSTOUFS(mp)->um_devvp;
624 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
625 if (vinvalbuf(devvp, 0, 0, 0) != 0)
626 panic("ffs_reload: dirty1");
627 VOP_UNLOCK(devvp, 0);
630 * Step 2: re-read superblock from disk.
632 fs = VFSTOUFS(mp)->um_fs;
633 if ((error = bread(devvp, btodb(fs->fs_sblockloc), fs->fs_sbsize,
636 newfs = (struct fs *)bp->b_data;
637 if ((newfs->fs_magic != FS_UFS1_MAGIC &&
638 newfs->fs_magic != FS_UFS2_MAGIC) ||
639 newfs->fs_bsize > MAXBSIZE ||
640 newfs->fs_bsize < sizeof(struct fs)) {
642 return (EIO); /* XXX needs translation */
645 * Copy pointer fields back into superblock before copying in XXX
646 * new superblock. These should really be in the ufsmount. XXX
647 * Note that important parameters (eg fs_ncg) are unchanged.
649 newfs->fs_csp = fs->fs_csp;
650 newfs->fs_maxcluster = fs->fs_maxcluster;
651 newfs->fs_contigdirs = fs->fs_contigdirs;
652 newfs->fs_active = fs->fs_active;
653 newfs->fs_ronly = fs->fs_ronly;
654 sblockloc = fs->fs_sblockloc;
655 bcopy(newfs, fs, (u_int)fs->fs_sbsize);
657 mp->mnt_maxsymlinklen = fs->fs_maxsymlinklen;
658 ffs_oldfscompat_read(fs, VFSTOUFS(mp), sblockloc);
660 if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
661 printf("WARNING: %s: reload pending error: blocks %jd "
662 "files %d\n", fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
663 fs->fs_pendinginodes);
664 fs->fs_pendingblocks = 0;
665 fs->fs_pendinginodes = 0;
670 * Step 3: re-read summary information from disk.
672 size = fs->fs_cssize;
673 blks = howmany(size, fs->fs_fsize);
674 if (fs->fs_contigsumsize > 0)
675 size += fs->fs_ncg * sizeof(int32_t);
676 size += fs->fs_ncg * sizeof(u_int8_t);
677 free(fs->fs_csp, M_UFSMNT);
678 space = malloc(size, M_UFSMNT, M_WAITOK);
680 for (i = 0; i < blks; i += fs->fs_frag) {
682 if (i + fs->fs_frag > blks)
683 size = (blks - i) * fs->fs_fsize;
684 error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), size,
688 bcopy(bp->b_data, space, (u_int)size);
689 space = (char *)space + size;
693 * We no longer know anything about clusters per cylinder group.
695 if (fs->fs_contigsumsize > 0) {
696 fs->fs_maxcluster = lp = space;
697 for (i = 0; i < fs->fs_ncg; i++)
698 *lp++ = fs->fs_contigsumsize;
701 size = fs->fs_ncg * sizeof(u_int8_t);
702 fs->fs_contigdirs = (u_int8_t *)space;
703 bzero(fs->fs_contigdirs, size);
704 if ((flags & FFSR_UNSUSPEND) != 0) {
706 mp->mnt_kern_flag &= ~(MNTK_SUSPENDED | MNTK_SUSPEND2);
707 wakeup(&mp->mnt_flag);
712 MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
716 if (vp->v_type == VNON) {
721 * Step 4: invalidate all cached file data.
723 if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, td)) {
724 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
727 if (vinvalbuf(vp, 0, 0, 0))
728 panic("ffs_reload: dirty2");
730 * Step 5: re-read inode data for all active vnodes.
734 bread(devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
735 (int)fs->fs_bsize, NOCRED, &bp);
739 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
742 ffs_load_inode(bp, ip, fs, ip->i_number);
743 ip->i_effnlink = ip->i_nlink;
752 * Possible superblock locations ordered from most to least likely.
754 static int sblock_try[] = SBLOCKSEARCH;
757 * Common code for mount and mountroot
760 ffs_mountfs(devvp, mp, td)
765 struct ufsmount *ump;
770 ufs2_daddr_t sblockloc;
771 int error, i, blks, len, ronly;
775 struct g_consumer *cp;
780 cred = td ? td->td_ucred : NOCRED;
781 ronly = (mp->mnt_flag & MNT_RDONLY) != 0;
783 KASSERT(devvp->v_type == VCHR, ("reclaimed devvp"));
785 if (atomic_cmpset_acq_ptr((uintptr_t *)&dev->si_mountpt, 0,
786 (uintptr_t)mp) == 0) {
787 VOP_UNLOCK(devvp, 0);
791 error = g_vfs_open(devvp, &cp, "ffs", ronly ? 0 : 1);
794 atomic_store_rel_ptr((uintptr_t *)&dev->si_mountpt, 0);
795 VOP_UNLOCK(devvp, 0);
799 devvp->v_bufobj.bo_ops = &ffs_ops;
800 VOP_UNLOCK(devvp, 0);
801 if (dev->si_iosize_max != 0)
802 mp->mnt_iosize_max = dev->si_iosize_max;
803 if (mp->mnt_iosize_max > MAXPHYS)
804 mp->mnt_iosize_max = MAXPHYS;
809 * Try reading the superblock in each of its possible locations.
811 for (i = 0; sblock_try[i] != -1; i++) {
812 if ((SBLOCKSIZE % cp->provider->sectorsize) != 0) {
815 "Invalid sectorsize %d for superblock size %d",
816 cp->provider->sectorsize, SBLOCKSIZE);
819 if ((error = bread(devvp, btodb(sblock_try[i]), SBLOCKSIZE,
822 fs = (struct fs *)bp->b_data;
823 sblockloc = sblock_try[i];
824 if ((fs->fs_magic == FS_UFS1_MAGIC ||
825 (fs->fs_magic == FS_UFS2_MAGIC &&
826 (fs->fs_sblockloc == sblockloc ||
827 (fs->fs_old_flags & FS_FLAGS_UPDATED) == 0))) &&
828 fs->fs_bsize <= MAXBSIZE &&
829 fs->fs_bsize >= sizeof(struct fs))
834 if (sblock_try[i] == -1) {
835 error = EINVAL; /* XXX needs translation */
839 fs->fs_flags &= ~FS_INDEXDIRS; /* no support for directory indices */
840 fs->fs_flags &= ~FS_UNCLEAN;
841 if (fs->fs_clean == 0) {
842 fs->fs_flags |= FS_UNCLEAN;
843 if (ronly || (mp->mnt_flag & MNT_FORCE) ||
844 ((fs->fs_flags & (FS_SUJ | FS_NEEDSFSCK)) == 0 &&
845 (fs->fs_flags & FS_DOSOFTDEP))) {
846 printf("WARNING: %s was not properly dismounted\n",
849 vfs_mount_error(mp, "R/W mount of %s denied. %s%s",
850 fs->fs_fsmnt, "Filesystem is not clean - run fsck.",
851 (fs->fs_flags & FS_SUJ) == 0 ? "" :
852 " Forced mount will invalidate journal contents");
856 if ((fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) &&
857 (mp->mnt_flag & MNT_FORCE)) {
858 printf("WARNING: %s: lost blocks %jd files %d\n",
859 fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
860 fs->fs_pendinginodes);
861 fs->fs_pendingblocks = 0;
862 fs->fs_pendinginodes = 0;
865 if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
866 printf("WARNING: %s: mount pending error: blocks %jd "
867 "files %d\n", fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
868 fs->fs_pendinginodes);
869 fs->fs_pendingblocks = 0;
870 fs->fs_pendinginodes = 0;
872 if ((fs->fs_flags & FS_GJOURNAL) != 0) {
875 * Get journal provider name.
878 mp->mnt_gjprovider = malloc((u_long)len, M_UFSMNT, M_WAITOK);
879 if (g_io_getattr("GJOURNAL::provider", cp, &len,
880 mp->mnt_gjprovider) == 0) {
881 mp->mnt_gjprovider = realloc(mp->mnt_gjprovider, len,
884 mp->mnt_flag |= MNT_GJOURNAL;
887 printf("WARNING: %s: GJOURNAL flag on fs "
888 "but no gjournal provider below\n",
889 mp->mnt_stat.f_mntonname);
890 free(mp->mnt_gjprovider, M_UFSMNT);
891 mp->mnt_gjprovider = NULL;
894 printf("WARNING: %s: GJOURNAL flag on fs but no "
895 "UFS_GJOURNAL support\n", mp->mnt_stat.f_mntonname);
898 mp->mnt_gjprovider = NULL;
900 ump = malloc(sizeof *ump, M_UFSMNT, M_WAITOK | M_ZERO);
902 ump->um_bo = &devvp->v_bufobj;
903 ump->um_fs = malloc((u_long)fs->fs_sbsize, M_UFSMNT, M_WAITOK);
904 if (fs->fs_magic == FS_UFS1_MAGIC) {
905 ump->um_fstype = UFS1;
906 ump->um_balloc = ffs_balloc_ufs1;
908 ump->um_fstype = UFS2;
909 ump->um_balloc = ffs_balloc_ufs2;
911 ump->um_blkatoff = ffs_blkatoff;
912 ump->um_truncate = ffs_truncate;
913 ump->um_update = ffs_update;
914 ump->um_valloc = ffs_valloc;
915 ump->um_vfree = ffs_vfree;
916 ump->um_ifree = ffs_ifree;
917 ump->um_rdonly = ffs_rdonly;
918 ump->um_snapgone = ffs_snapgone;
919 mtx_init(UFS_MTX(ump), "FFS", "FFS Lock", MTX_DEF);
920 bcopy(bp->b_data, ump->um_fs, (u_int)fs->fs_sbsize);
921 if (fs->fs_sbsize < SBLOCKSIZE)
922 bp->b_flags |= B_INVAL | B_NOCACHE;
926 ffs_oldfscompat_read(fs, ump, sblockloc);
927 fs->fs_ronly = ronly;
928 size = fs->fs_cssize;
929 blks = howmany(size, fs->fs_fsize);
930 if (fs->fs_contigsumsize > 0)
931 size += fs->fs_ncg * sizeof(int32_t);
932 size += fs->fs_ncg * sizeof(u_int8_t);
933 space = malloc(size, M_UFSMNT, M_WAITOK);
935 for (i = 0; i < blks; i += fs->fs_frag) {
937 if (i + fs->fs_frag > blks)
938 size = (blks - i) * fs->fs_fsize;
939 if ((error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), size,
941 free(fs->fs_csp, M_UFSMNT);
944 bcopy(bp->b_data, space, (u_int)size);
945 space = (char *)space + size;
949 if (fs->fs_contigsumsize > 0) {
950 fs->fs_maxcluster = lp = space;
951 for (i = 0; i < fs->fs_ncg; i++)
952 *lp++ = fs->fs_contigsumsize;
955 size = fs->fs_ncg * sizeof(u_int8_t);
956 fs->fs_contigdirs = (u_int8_t *)space;
957 bzero(fs->fs_contigdirs, size);
958 fs->fs_active = NULL;
960 mp->mnt_stat.f_fsid.val[0] = fs->fs_id[0];
961 mp->mnt_stat.f_fsid.val[1] = fs->fs_id[1];
963 if (fs->fs_id[0] == 0 || fs->fs_id[1] == 0 ||
964 (nmp = vfs_getvfs(&mp->mnt_stat.f_fsid))) {
969 mp->mnt_maxsymlinklen = fs->fs_maxsymlinklen;
971 mp->mnt_flag |= MNT_LOCAL;
973 if ((fs->fs_flags & FS_MULTILABEL) != 0) {
976 mp->mnt_flag |= MNT_MULTILABEL;
979 printf("WARNING: %s: multilabel flag on fs but "
980 "no MAC support\n", mp->mnt_stat.f_mntonname);
983 if ((fs->fs_flags & FS_ACLS) != 0) {
987 if (mp->mnt_flag & MNT_NFS4ACLS)
988 printf("WARNING: %s: ACLs flag on fs conflicts with "
989 "\"nfsv4acls\" mount option; option ignored\n",
990 mp->mnt_stat.f_mntonname);
991 mp->mnt_flag &= ~MNT_NFS4ACLS;
992 mp->mnt_flag |= MNT_ACLS;
996 printf("WARNING: %s: ACLs flag on fs but no ACLs support\n",
997 mp->mnt_stat.f_mntonname);
1000 if ((fs->fs_flags & FS_NFS4ACLS) != 0) {
1004 if (mp->mnt_flag & MNT_ACLS)
1005 printf("WARNING: %s: NFSv4 ACLs flag on fs conflicts "
1006 "with \"acls\" mount option; option ignored\n",
1007 mp->mnt_stat.f_mntonname);
1008 mp->mnt_flag &= ~MNT_ACLS;
1009 mp->mnt_flag |= MNT_NFS4ACLS;
1013 printf("WARNING: %s: NFSv4 ACLs flag on fs but no "
1014 "ACLs support\n", mp->mnt_stat.f_mntonname);
1017 if ((fs->fs_flags & FS_TRIM) != 0) {
1019 if (g_io_getattr("GEOM::candelete", cp, &len,
1020 &ump->um_candelete) == 0) {
1021 if (!ump->um_candelete)
1022 printf("WARNING: %s: TRIM flag on fs but disk "
1023 "does not support TRIM\n",
1024 mp->mnt_stat.f_mntonname);
1026 printf("WARNING: %s: TRIM flag on fs but disk does "
1027 "not confirm that it supports TRIM\n",
1028 mp->mnt_stat.f_mntonname);
1029 ump->um_candelete = 0;
1031 if (ump->um_candelete) {
1032 ump->um_trim_tq = taskqueue_create("trim", M_WAITOK,
1033 taskqueue_thread_enqueue, &ump->um_trim_tq);
1034 taskqueue_start_threads(&ump->um_trim_tq, 1, PVFS,
1035 "%s trim", mp->mnt_stat.f_mntonname);
1039 ump->um_mountp = mp;
1041 ump->um_devvp = devvp;
1042 ump->um_nindir = fs->fs_nindir;
1043 ump->um_bptrtodb = fs->fs_fsbtodb;
1044 ump->um_seqinc = fs->fs_frag;
1045 for (i = 0; i < MAXQUOTAS; i++)
1046 ump->um_quotas[i] = NULLVP;
1048 ufs_extattr_uepm_init(&ump->um_extattr);
1051 * Set FS local "last mounted on" information (NULL pad)
1053 bzero(fs->fs_fsmnt, MAXMNTLEN);
1054 strlcpy(fs->fs_fsmnt, mp->mnt_stat.f_mntonname, MAXMNTLEN);
1055 mp->mnt_stat.f_iosize = fs->fs_bsize;
1057 if (mp->mnt_flag & MNT_ROOTFS) {
1059 * Root mount; update timestamp in mount structure.
1060 * this will be used by the common root mount code
1061 * to update the system clock.
1063 mp->mnt_time = fs->fs_time;
1067 fs->fs_mtime = time_second;
1068 if ((fs->fs_flags & FS_DOSOFTDEP) &&
1069 (error = softdep_mount(devvp, mp, fs, cred)) != 0) {
1070 free(fs->fs_csp, M_UFSMNT);
1071 ffs_flushfiles(mp, FORCECLOSE, td);
1074 if (fs->fs_snapinum[0] != 0)
1075 ffs_snapshot_mount(mp);
1078 (void) ffs_sbupdate(ump, MNT_WAIT, 0);
1081 * Initialize filesystem state information in mount struct.
1084 mp->mnt_kern_flag |= MNTK_LOOKUP_SHARED | MNTK_EXTENDED_SHARED |
1085 MNTK_NO_IOPF | MNTK_UNMAPPED_BUFS | MNTK_USES_BCACHE;
1088 #ifdef UFS_EXTATTR_AUTOSTART
1091 * Auto-starting does the following:
1092 * - check for /.attribute in the fs, and extattr_start if so
1093 * - for each file in .attribute, enable that file with
1094 * an attribute of the same name.
1095 * Not clear how to report errors -- probably eat them.
1096 * This would all happen while the filesystem was busy/not
1097 * available, so would effectively be "atomic".
1099 (void) ufs_extattr_autostart(mp, td);
1100 #endif /* !UFS_EXTATTR_AUTOSTART */
1101 #endif /* !UFS_EXTATTR */
1109 g_topology_unlock();
1112 mtx_destroy(UFS_MTX(ump));
1113 if (mp->mnt_gjprovider != NULL) {
1114 free(mp->mnt_gjprovider, M_UFSMNT);
1115 mp->mnt_gjprovider = NULL;
1117 free(ump->um_fs, M_UFSMNT);
1118 free(ump, M_UFSMNT);
1119 mp->mnt_data = NULL;
1121 atomic_store_rel_ptr((uintptr_t *)&dev->si_mountpt, 0);
1126 #include <sys/sysctl.h>
1127 static int bigcgs = 0;
1128 SYSCTL_INT(_debug, OID_AUTO, bigcgs, CTLFLAG_RW, &bigcgs, 0, "");
1131 * Sanity checks for loading old filesystem superblocks.
1132 * See ffs_oldfscompat_write below for unwound actions.
1134 * XXX - Parts get retired eventually.
1135 * Unfortunately new bits get added.
1138 ffs_oldfscompat_read(fs, ump, sblockloc)
1140 struct ufsmount *ump;
1141 ufs2_daddr_t sblockloc;
1146 * If not yet done, update fs_flags location and value of fs_sblockloc.
1148 if ((fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) {
1149 fs->fs_flags = fs->fs_old_flags;
1150 fs->fs_old_flags |= FS_FLAGS_UPDATED;
1151 fs->fs_sblockloc = sblockloc;
1154 * If not yet done, update UFS1 superblock with new wider fields.
1156 if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_maxbsize != fs->fs_bsize) {
1157 fs->fs_maxbsize = fs->fs_bsize;
1158 fs->fs_time = fs->fs_old_time;
1159 fs->fs_size = fs->fs_old_size;
1160 fs->fs_dsize = fs->fs_old_dsize;
1161 fs->fs_csaddr = fs->fs_old_csaddr;
1162 fs->fs_cstotal.cs_ndir = fs->fs_old_cstotal.cs_ndir;
1163 fs->fs_cstotal.cs_nbfree = fs->fs_old_cstotal.cs_nbfree;
1164 fs->fs_cstotal.cs_nifree = fs->fs_old_cstotal.cs_nifree;
1165 fs->fs_cstotal.cs_nffree = fs->fs_old_cstotal.cs_nffree;
1167 if (fs->fs_magic == FS_UFS1_MAGIC &&
1168 fs->fs_old_inodefmt < FS_44INODEFMT) {
1169 fs->fs_maxfilesize = ((uint64_t)1 << 31) - 1;
1170 fs->fs_qbmask = ~fs->fs_bmask;
1171 fs->fs_qfmask = ~fs->fs_fmask;
1173 if (fs->fs_magic == FS_UFS1_MAGIC) {
1174 ump->um_savedmaxfilesize = fs->fs_maxfilesize;
1175 maxfilesize = (uint64_t)0x80000000 * fs->fs_bsize - 1;
1176 if (fs->fs_maxfilesize > maxfilesize)
1177 fs->fs_maxfilesize = maxfilesize;
1179 /* Compatibility for old filesystems */
1180 if (fs->fs_avgfilesize <= 0)
1181 fs->fs_avgfilesize = AVFILESIZ;
1182 if (fs->fs_avgfpdir <= 0)
1183 fs->fs_avgfpdir = AFPDIR;
1185 fs->fs_save_cgsize = fs->fs_cgsize;
1186 fs->fs_cgsize = fs->fs_bsize;
1191 * Unwinding superblock updates for old filesystems.
1192 * See ffs_oldfscompat_read above for details.
1194 * XXX - Parts get retired eventually.
1195 * Unfortunately new bits get added.
1198 ffs_oldfscompat_write(fs, ump)
1200 struct ufsmount *ump;
1204 * Copy back UFS2 updated fields that UFS1 inspects.
1206 if (fs->fs_magic == FS_UFS1_MAGIC) {
1207 fs->fs_old_time = fs->fs_time;
1208 fs->fs_old_cstotal.cs_ndir = fs->fs_cstotal.cs_ndir;
1209 fs->fs_old_cstotal.cs_nbfree = fs->fs_cstotal.cs_nbfree;
1210 fs->fs_old_cstotal.cs_nifree = fs->fs_cstotal.cs_nifree;
1211 fs->fs_old_cstotal.cs_nffree = fs->fs_cstotal.cs_nffree;
1212 fs->fs_maxfilesize = ump->um_savedmaxfilesize;
1215 fs->fs_cgsize = fs->fs_save_cgsize;
1216 fs->fs_save_cgsize = 0;
1221 * unmount system call
1224 ffs_unmount(mp, mntflags)
1229 struct ufsmount *ump = VFSTOUFS(mp);
1231 int error, flags, susp;
1240 if (mntflags & MNT_FORCE) {
1241 flags |= FORCECLOSE;
1242 susp = fs->fs_ronly == 0;
1245 if ((error = ufs_extattr_stop(mp, td))) {
1246 if (error != EOPNOTSUPP)
1247 printf("WARNING: unmount %s: ufs_extattr_stop "
1248 "returned errno %d\n", mp->mnt_stat.f_mntonname,
1252 ufs_extattr_uepm_destroy(&ump->um_extattr);
1257 error = vfs_write_suspend_umnt(mp);
1261 if (MOUNTEDSOFTDEP(mp))
1262 error = softdep_flushfiles(mp, flags, td);
1264 error = ffs_flushfiles(mp, flags, td);
1265 if (error != 0 && error != ENXIO)
1269 if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
1270 printf("WARNING: unmount %s: pending error: blocks %jd "
1271 "files %d\n", fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
1272 fs->fs_pendinginodes);
1273 fs->fs_pendingblocks = 0;
1274 fs->fs_pendinginodes = 0;
1277 if (MOUNTEDSOFTDEP(mp))
1278 softdep_unmount(mp);
1279 if (fs->fs_ronly == 0 || ump->um_fsckpid > 0) {
1280 fs->fs_clean = fs->fs_flags & (FS_UNCLEAN|FS_NEEDSFSCK) ? 0 : 1;
1281 error = ffs_sbupdate(ump, MNT_WAIT, 0);
1282 if (error && error != ENXIO) {
1288 vfs_write_resume(mp, VR_START_WRITE);
1289 if (ump->um_trim_tq != NULL) {
1290 while (ump->um_trim_inflight != 0)
1291 pause("ufsutr", hz);
1292 taskqueue_drain_all(ump->um_trim_tq);
1293 taskqueue_free(ump->um_trim_tq);
1296 if (ump->um_fsckpid > 0) {
1298 * Return to normal read-only mode.
1300 error = g_access(ump->um_cp, 0, -1, 0);
1301 ump->um_fsckpid = 0;
1303 g_vfs_close(ump->um_cp);
1304 g_topology_unlock();
1305 atomic_store_rel_ptr((uintptr_t *)&ump->um_dev->si_mountpt, 0);
1306 vrele(ump->um_devvp);
1307 dev_rel(ump->um_dev);
1308 mtx_destroy(UFS_MTX(ump));
1309 if (mp->mnt_gjprovider != NULL) {
1310 free(mp->mnt_gjprovider, M_UFSMNT);
1311 mp->mnt_gjprovider = NULL;
1313 free(fs->fs_csp, M_UFSMNT);
1315 free(ump, M_UFSMNT);
1316 mp->mnt_data = NULL;
1318 mp->mnt_flag &= ~MNT_LOCAL;
1324 vfs_write_resume(mp, VR_START_WRITE);
1328 ufs_extattr_uepm_init(&ump->um_extattr);
1329 #ifdef UFS_EXTATTR_AUTOSTART
1330 (void) ufs_extattr_autostart(mp, td);
1339 * Flush out all the files in a filesystem.
1342 ffs_flushfiles(mp, flags, td)
1347 struct ufsmount *ump;
1353 if (mp->mnt_flag & MNT_QUOTA) {
1355 error = vflush(mp, 0, SKIPSYSTEM|flags, td);
1358 for (i = 0; i < MAXQUOTAS; i++) {
1359 error = quotaoff(td, mp, i);
1361 if ((flags & EARLYFLUSH) == 0)
1369 * Here we fall through to vflush again to ensure that
1370 * we have gotten rid of all the system vnodes, unless
1371 * quotas must not be closed.
1375 ASSERT_VOP_LOCKED(ump->um_devvp, "ffs_flushfiles");
1376 if (ump->um_devvp->v_vflag & VV_COPYONWRITE) {
1377 if ((error = vflush(mp, 0, SKIPSYSTEM | flags, td)) != 0)
1379 ffs_snapshot_unmount(mp);
1380 flags |= FORCECLOSE;
1382 * Here we fall through to vflush again to ensure
1383 * that we have gotten rid of all the system vnodes.
1388 * Do not close system files if quotas were not closed, to be
1389 * able to sync the remaining dquots. The freeblks softupdate
1390 * workitems might hold a reference on a dquot, preventing
1391 * quotaoff() from completing. Next round of
1392 * softdep_flushworklist() iteration should process the
1393 * blockers, allowing the next run of quotaoff() to finally
1394 * flush held dquots.
1396 * Otherwise, flush all the files.
1398 if (qerror == 0 && (error = vflush(mp, 0, flags, td)) != 0)
1402 * Flush filesystem metadata.
1404 vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY);
1405 error = VOP_FSYNC(ump->um_devvp, MNT_WAIT, td);
1406 VOP_UNLOCK(ump->um_devvp, 0);
1411 * Get filesystem statistics.
1418 struct ufsmount *ump;
1423 if (fs->fs_magic != FS_UFS1_MAGIC && fs->fs_magic != FS_UFS2_MAGIC)
1424 panic("ffs_statfs");
1425 sbp->f_version = STATFS_VERSION;
1426 sbp->f_bsize = fs->fs_fsize;
1427 sbp->f_iosize = fs->fs_bsize;
1428 sbp->f_blocks = fs->fs_dsize;
1430 sbp->f_bfree = fs->fs_cstotal.cs_nbfree * fs->fs_frag +
1431 fs->fs_cstotal.cs_nffree + dbtofsb(fs, fs->fs_pendingblocks);
1432 sbp->f_bavail = freespace(fs, fs->fs_minfree) +
1433 dbtofsb(fs, fs->fs_pendingblocks);
1434 sbp->f_files = fs->fs_ncg * fs->fs_ipg - UFS_ROOTINO;
1435 sbp->f_ffree = fs->fs_cstotal.cs_nifree + fs->fs_pendinginodes;
1437 sbp->f_namemax = NAME_MAX;
1442 sync_doupdate(struct inode *ip)
1445 return ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_MODIFIED |
1450 * For a lazy sync, we only care about access times, quotas and the
1451 * superblock. Other filesystem changes are already converted to
1452 * cylinder group blocks or inode blocks updates and are written to
1459 struct vnode *mvp, *vp;
1462 int allerror, error;
1466 if ((mp->mnt_flag & MNT_NOATIME) != 0)
1468 MNT_VNODE_FOREACH_ACTIVE(vp, mp, mvp) {
1469 if (vp->v_type == VNON) {
1476 * The IN_ACCESS flag is converted to IN_MODIFIED by
1477 * ufs_close() and ufs_getattr() by the calls to
1478 * ufs_itimes_locked(), without subsequent UFS_UPDATE().
1479 * Test also all the other timestamp flags too, to pick up
1480 * any other cases that could be missed.
1482 if (!sync_doupdate(ip) && (vp->v_iflag & VI_OWEINACT) == 0) {
1486 if ((error = vget(vp, LK_EXCLUSIVE | LK_NOWAIT | LK_INTERLOCK,
1489 if (sync_doupdate(ip))
1490 error = ffs_update(vp, 0);
1501 if (VFSTOUFS(mp)->um_fs->fs_fmod != 0 &&
1502 (error = ffs_sbupdate(VFSTOUFS(mp), MNT_LAZY, 0)) != 0)
1508 * Go through the disk queues to initiate sandbagged IO;
1509 * go through the inodes to write those that have been modified;
1510 * initiate the writing of the super block if it has been modified.
1512 * Note: we are always called with the filesystem marked busy using
1516 ffs_sync(mp, waitfor)
1520 struct vnode *mvp, *vp, *devvp;
1523 struct ufsmount *ump = VFSTOUFS(mp);
1525 int error, count, lockreq, allerror = 0;
1528 int secondary_writes;
1529 int secondary_accwrites;
1531 int softdep_accdeps;
1538 if (fs->fs_fmod != 0 && fs->fs_ronly != 0 && ump->um_fsckpid == 0)
1539 panic("%s: ffs_sync: modification on read-only filesystem",
1541 if (waitfor == MNT_LAZY) {
1543 return (ffs_sync_lazy(mp));
1544 waitfor = MNT_NOWAIT;
1548 * Write back each (modified) inode.
1550 lockreq = LK_EXCLUSIVE | LK_NOWAIT;
1551 if (waitfor == MNT_SUSPEND) {
1555 if (waitfor == MNT_WAIT)
1556 lockreq = LK_EXCLUSIVE;
1557 lockreq |= LK_INTERLOCK | LK_SLEEPFAIL;
1559 /* Grab snapshot of secondary write counts */
1561 secondary_writes = mp->mnt_secondary_writes;
1562 secondary_accwrites = mp->mnt_secondary_accwrites;
1565 /* Grab snapshot of softdep dependency counts */
1566 softdep_get_depcounts(mp, &softdep_deps, &softdep_accdeps);
1568 MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
1570 * Depend on the vnode interlock to keep things stable enough
1571 * for a quick test. Since there might be hundreds of
1572 * thousands of vnodes, we cannot afford even a subroutine
1573 * call unless there's a good chance that we have work to do.
1575 if (vp->v_type == VNON) {
1581 (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) == 0 &&
1582 vp->v_bufobj.bo_dirty.bv_cnt == 0) {
1586 if ((error = vget(vp, lockreq, td)) != 0) {
1587 if (error == ENOENT || error == ENOLCK) {
1588 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
1593 if ((error = ffs_syncvnode(vp, waitfor, 0)) != 0)
1598 * Force stale filesystem control information to be flushed.
1600 if (waitfor == MNT_WAIT || rebooting) {
1601 if ((error = softdep_flushworklist(ump->um_mountp, &count, td)))
1603 /* Flushed work items may create new vnodes to clean */
1604 if (allerror == 0 && count)
1611 devvp = ump->um_devvp;
1612 bo = &devvp->v_bufobj;
1614 if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0) {
1616 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
1617 error = VOP_FSYNC(devvp, waitfor, td);
1618 VOP_UNLOCK(devvp, 0);
1619 if (MOUNTEDSOFTDEP(mp) && (error == 0 || error == EAGAIN))
1620 error = ffs_sbupdate(ump, waitfor, 0);
1623 if (allerror == 0 && waitfor == MNT_WAIT)
1625 } else if (suspend != 0) {
1626 if (softdep_check_suspend(mp,
1631 secondary_accwrites) != 0) {
1633 goto loop; /* More work needed */
1635 mtx_assert(MNT_MTX(mp), MA_OWNED);
1636 mp->mnt_kern_flag |= MNTK_SUSPEND2 | MNTK_SUSPENDED;
1642 * Write back modified superblock.
1644 if (fs->fs_fmod != 0 &&
1645 (error = ffs_sbupdate(ump, waitfor, suspended)) != 0)
1651 ffs_vget(mp, ino, flags, vpp)
1657 return (ffs_vgetf(mp, ino, flags, vpp, 0));
1661 ffs_vgetf(mp, ino, flags, vpp, ffs_flags)
1670 struct ufsmount *ump;
1675 error = vfs_hash_get(mp, ino, flags, curthread, vpp, NULL, NULL);
1676 if (error || *vpp != NULL)
1680 * We must promote to an exclusive lock for vnode creation. This
1681 * can happen if lookup is passed LOCKSHARED.
1683 if ((flags & LK_TYPE_MASK) == LK_SHARED) {
1684 flags &= ~LK_TYPE_MASK;
1685 flags |= LK_EXCLUSIVE;
1689 * We do not lock vnode creation as it is believed to be too
1690 * expensive for such rare case as simultaneous creation of vnode
1691 * for same ino by different processes. We just allow them to race
1692 * and check later to decide who wins. Let the race begin!
1697 ip = uma_zalloc(uma_inode, M_WAITOK | M_ZERO);
1699 /* Allocate a new vnode/inode. */
1700 error = getnewvnode("ufs", mp, fs->fs_magic == FS_UFS1_MAGIC ?
1701 &ffs_vnodeops1 : &ffs_vnodeops2, &vp);
1704 uma_zfree(uma_inode, ip);
1708 * FFS supports recursive locking.
1710 lockmgr(vp->v_vnlock, LK_EXCLUSIVE, NULL);
1713 vp->v_bufobj.bo_bsize = fs->fs_bsize;
1718 ip->i_nextclustercg = -1;
1719 ip->i_flag = fs->fs_magic == FS_UFS1_MAGIC ? 0 : IN_UFS2;
1723 for (i = 0; i < MAXQUOTAS; i++)
1724 ip->i_dquot[i] = NODQUOT;
1728 if (ffs_flags & FFSV_FORCEINSMQ)
1729 vp->v_vflag |= VV_FORCEINSMQ;
1730 error = insmntque(vp, mp);
1732 uma_zfree(uma_inode, ip);
1736 vp->v_vflag &= ~VV_FORCEINSMQ;
1737 error = vfs_hash_insert(vp, ino, flags, curthread, vpp, NULL, NULL);
1738 if (error || *vpp != NULL)
1741 /* Read in the disk contents for the inode, copy into the inode. */
1742 error = bread(ump->um_devvp, fsbtodb(fs, ino_to_fsba(fs, ino)),
1743 (int)fs->fs_bsize, NOCRED, &bp);
1746 * The inode does not contain anything useful, so it would
1747 * be misleading to leave it on its hash chain. With mode
1748 * still zero, it will be unlinked and returned to the free
1757 ip->i_din1 = uma_zalloc(uma_ufs1, M_WAITOK);
1759 ip->i_din2 = uma_zalloc(uma_ufs2, M_WAITOK);
1760 ffs_load_inode(bp, ip, fs, ino);
1761 if (DOINGSOFTDEP(vp))
1762 softdep_load_inodeblock(ip);
1764 ip->i_effnlink = ip->i_nlink;
1768 * Initialize the vnode from the inode, check for aliases.
1769 * Note that the underlying vnode may have changed.
1771 error = ufs_vinit(mp, I_IS_UFS1(ip) ? &ffs_fifoops1 : &ffs_fifoops2,
1780 * Finish inode initialization.
1782 if (vp->v_type != VFIFO) {
1783 /* FFS supports shared locking for all files except fifos. */
1788 * Set up a generation number for this inode if it does not
1789 * already have one. This should only happen on old filesystems.
1791 if (ip->i_gen == 0) {
1792 while (ip->i_gen == 0)
1793 ip->i_gen = arc4random();
1794 if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
1795 ip->i_flag |= IN_MODIFIED;
1796 DIP_SET(ip, i_gen, ip->i_gen);
1800 if ((mp->mnt_flag & MNT_MULTILABEL) && ip->i_mode) {
1802 * If this vnode is already allocated, and we're running
1803 * multi-label, attempt to perform a label association
1804 * from the extended attributes on the inode.
1806 error = mac_vnode_associate_extattr(mp, vp);
1808 /* ufs_inactive will release ip->i_devvp ref. */
1821 * File handle to vnode
1823 * Have to be really careful about stale file handles:
1824 * - check that the inode number is valid
1825 * - for UFS2 check that the inode number is initialized
1826 * - call ffs_vget() to get the locked inode
1827 * - check for an unallocated inode (i_mode == 0)
1828 * - check that the given client host has export rights and return
1829 * those rights via. exflagsp and credanonp
1832 ffs_fhtovp(mp, fhp, flags, vpp)
1839 struct ufsmount *ump;
1847 ufhp = (struct ufid *)fhp;
1848 ino = ufhp->ufid_ino;
1851 if (ino < UFS_ROOTINO || ino >= fs->fs_ncg * fs->fs_ipg)
1854 * Need to check if inode is initialized because UFS2 does lazy
1855 * initialization and nfs_fhtovp can offer arbitrary inode numbers.
1857 if (fs->fs_magic != FS_UFS2_MAGIC)
1858 return (ufs_fhtovp(mp, ufhp, flags, vpp));
1859 cg = ino_to_cg(fs, ino);
1860 error = bread(ump->um_devvp, fsbtodb(fs, cgtod(fs, cg)),
1861 (int)fs->fs_cgsize, NOCRED, &bp);
1864 cgp = (struct cg *)bp->b_data;
1865 if (!cg_chkmagic(cgp) || ino >= cg * fs->fs_ipg + cgp->cg_initediblk) {
1870 return (ufs_fhtovp(mp, ufhp, flags, vpp));
1874 * Initialize the filesystem.
1878 struct vfsconf *vfsp;
1881 ffs_susp_initialize();
1882 softdep_initialize();
1883 return (ufs_init(vfsp));
1887 * Undo the work of ffs_init().
1891 struct vfsconf *vfsp;
1895 ret = ufs_uninit(vfsp);
1896 softdep_uninitialize();
1897 ffs_susp_uninitialize();
1902 * Write a superblock and associated information back to disk.
1905 ffs_sbupdate(ump, waitfor, suspended)
1906 struct ufsmount *ump;
1910 struct fs *fs = ump->um_fs;
1915 int i, size, error, allerror = 0;
1917 if (fs->fs_ronly == 1 &&
1918 (ump->um_mountp->mnt_flag & (MNT_RDONLY | MNT_UPDATE)) !=
1919 (MNT_RDONLY | MNT_UPDATE) && ump->um_fsckpid == 0)
1920 panic("ffs_sbupdate: write read-only filesystem");
1922 * We use the superblock's buf to serialize calls to ffs_sbupdate().
1924 sbbp = getblk(ump->um_devvp, btodb(fs->fs_sblockloc),
1925 (int)fs->fs_sbsize, 0, 0, 0);
1927 * First write back the summary information.
1929 blks = howmany(fs->fs_cssize, fs->fs_fsize);
1931 for (i = 0; i < blks; i += fs->fs_frag) {
1932 size = fs->fs_bsize;
1933 if (i + fs->fs_frag > blks)
1934 size = (blks - i) * fs->fs_fsize;
1935 bp = getblk(ump->um_devvp, fsbtodb(fs, fs->fs_csaddr + i),
1937 bcopy(space, bp->b_data, (u_int)size);
1938 space = (char *)space + size;
1940 bp->b_flags |= B_VALIDSUSPWRT;
1941 if (waitfor != MNT_WAIT)
1943 else if ((error = bwrite(bp)) != 0)
1947 * Now write back the superblock itself. If any errors occurred
1948 * up to this point, then fail so that the superblock avoids
1949 * being written out as clean.
1956 if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_sblockloc != SBLOCK_UFS1 &&
1957 (fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) {
1958 printf("WARNING: %s: correcting fs_sblockloc from %jd to %d\n",
1959 fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS1);
1960 fs->fs_sblockloc = SBLOCK_UFS1;
1962 if (fs->fs_magic == FS_UFS2_MAGIC && fs->fs_sblockloc != SBLOCK_UFS2 &&
1963 (fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) {
1964 printf("WARNING: %s: correcting fs_sblockloc from %jd to %d\n",
1965 fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS2);
1966 fs->fs_sblockloc = SBLOCK_UFS2;
1969 fs->fs_time = time_second;
1970 if (MOUNTEDSOFTDEP(ump->um_mountp))
1971 softdep_setup_sbupdate(ump, (struct fs *)bp->b_data, bp);
1972 bcopy((caddr_t)fs, bp->b_data, (u_int)fs->fs_sbsize);
1973 ffs_oldfscompat_write((struct fs *)bp->b_data, ump);
1975 bp->b_flags |= B_VALIDSUSPWRT;
1976 if (waitfor != MNT_WAIT)
1978 else if ((error = bwrite(bp)) != 0)
1984 ffs_extattrctl(struct mount *mp, int cmd, struct vnode *filename_vp,
1985 int attrnamespace, const char *attrname)
1989 return (ufs_extattrctl(mp, cmd, filename_vp, attrnamespace,
1992 return (vfs_stdextattrctl(mp, cmd, filename_vp, attrnamespace,
1998 ffs_ifree(struct ufsmount *ump, struct inode *ip)
2001 if (ump->um_fstype == UFS1 && ip->i_din1 != NULL)
2002 uma_zfree(uma_ufs1, ip->i_din1);
2003 else if (ip->i_din2 != NULL)
2004 uma_zfree(uma_ufs2, ip->i_din2);
2005 uma_zfree(uma_inode, ip);
2008 static int dobkgrdwrite = 1;
2009 SYSCTL_INT(_debug, OID_AUTO, dobkgrdwrite, CTLFLAG_RW, &dobkgrdwrite, 0,
2010 "Do background writes (honoring the BV_BKGRDWRITE flag)?");
2013 * Complete a background write started from bwrite.
2016 ffs_backgroundwritedone(struct buf *bp)
2018 struct bufobj *bufobj;
2022 * Find the original buffer that we are writing.
2024 bufobj = bp->b_bufobj;
2026 if ((origbp = gbincore(bp->b_bufobj, bp->b_lblkno)) == NULL)
2027 panic("backgroundwritedone: lost buffer");
2030 * We should mark the cylinder group buffer origbp as
2031 * dirty, to not loose the failed write.
2033 if ((bp->b_ioflags & BIO_ERROR) != 0)
2034 origbp->b_vflags |= BV_BKGRDERR;
2037 * Process dependencies then return any unfinished ones.
2040 if (!LIST_EMPTY(&bp->b_dep) && (bp->b_ioflags & BIO_ERROR) == 0)
2043 if (!LIST_EMPTY(&bp->b_dep))
2044 softdep_move_dependencies(bp, origbp);
2047 * This buffer is marked B_NOCACHE so when it is released
2048 * by biodone it will be tossed.
2050 bp->b_flags |= B_NOCACHE;
2051 bp->b_flags &= ~B_CACHE;
2054 * Prevent brelse() from trying to keep and re-dirtying bp on
2055 * errors. It causes b_bufobj dereference in
2056 * bdirty()/reassignbuf(), and b_bufobj was cleared in
2059 if ((bp->b_ioflags & BIO_ERROR) != 0)
2060 bp->b_flags |= B_INVAL;
2064 * Clear the BV_BKGRDINPROG flag in the original buffer
2065 * and awaken it if it is waiting for the write to complete.
2066 * If BV_BKGRDINPROG is not set in the original buffer it must
2067 * have been released and re-instantiated - which is not legal.
2069 KASSERT((origbp->b_vflags & BV_BKGRDINPROG),
2070 ("backgroundwritedone: lost buffer2"));
2071 origbp->b_vflags &= ~BV_BKGRDINPROG;
2072 if (origbp->b_vflags & BV_BKGRDWAIT) {
2073 origbp->b_vflags &= ~BV_BKGRDWAIT;
2074 wakeup(&origbp->b_xflags);
2081 * Write, release buffer on completion. (Done by iodone
2082 * if async). Do not bother writing anything if the buffer
2085 * Note that we set B_CACHE here, indicating that buffer is
2086 * fully valid and thus cacheable. This is true even of NFS
2087 * now so we set it generally. This could be set either here
2088 * or in biodone() since the I/O is synchronous. We put it
2092 ffs_bufwrite(struct buf *bp)
2096 CTR3(KTR_BUF, "bufwrite(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
2097 if (bp->b_flags & B_INVAL) {
2102 if (!BUF_ISLOCKED(bp))
2103 panic("bufwrite: buffer is not busy???");
2105 * If a background write is already in progress, delay
2106 * writing this block if it is asynchronous. Otherwise
2107 * wait for the background write to complete.
2109 BO_LOCK(bp->b_bufobj);
2110 if (bp->b_vflags & BV_BKGRDINPROG) {
2111 if (bp->b_flags & B_ASYNC) {
2112 BO_UNLOCK(bp->b_bufobj);
2116 bp->b_vflags |= BV_BKGRDWAIT;
2117 msleep(&bp->b_xflags, BO_LOCKPTR(bp->b_bufobj), PRIBIO,
2119 if (bp->b_vflags & BV_BKGRDINPROG)
2120 panic("bufwrite: still writing");
2122 bp->b_vflags &= ~BV_BKGRDERR;
2123 BO_UNLOCK(bp->b_bufobj);
2126 * If this buffer is marked for background writing and we
2127 * do not have to wait for it, make a copy and write the
2128 * copy so as to leave this buffer ready for further use.
2130 * This optimization eats a lot of memory. If we have a page
2131 * or buffer shortfall we can't do it.
2133 if (dobkgrdwrite && (bp->b_xflags & BX_BKGRDWRITE) &&
2134 (bp->b_flags & B_ASYNC) &&
2135 !vm_page_count_severe() &&
2136 !buf_dirty_count_severe()) {
2137 KASSERT(bp->b_iodone == NULL,
2138 ("bufwrite: needs chained iodone (%p)", bp->b_iodone));
2140 /* get a new block */
2141 newbp = geteblk(bp->b_bufsize, GB_NOWAIT_BD);
2145 KASSERT(buf_mapped(bp), ("Unmapped cg"));
2146 memcpy(newbp->b_data, bp->b_data, bp->b_bufsize);
2147 BO_LOCK(bp->b_bufobj);
2148 bp->b_vflags |= BV_BKGRDINPROG;
2149 BO_UNLOCK(bp->b_bufobj);
2150 newbp->b_xflags |= BX_BKGRDMARKER;
2151 newbp->b_lblkno = bp->b_lblkno;
2152 newbp->b_blkno = bp->b_blkno;
2153 newbp->b_offset = bp->b_offset;
2154 newbp->b_iodone = ffs_backgroundwritedone;
2155 newbp->b_flags |= B_ASYNC;
2156 newbp->b_flags &= ~B_INVAL;
2157 pbgetvp(bp->b_vp, newbp);
2161 * Move over the dependencies. If there are rollbacks,
2162 * leave the parent buffer dirtied as it will need to
2165 if (LIST_EMPTY(&bp->b_dep) ||
2166 softdep_move_dependencies(bp, newbp) == 0)
2173 * Initiate write on the copy, release the original. The
2174 * BKGRDINPROG flag prevents it from going away until
2175 * the background write completes.
2180 /* Mark the buffer clean */
2184 /* Let the normal bufwrite do the rest for us */
2186 return (bufwrite(bp));
2191 ffs_geom_strategy(struct bufobj *bo, struct buf *bp)
2199 if (bp->b_iocmd == BIO_WRITE) {
2200 if ((bp->b_flags & B_VALIDSUSPWRT) == 0 &&
2201 bp->b_vp != NULL && bp->b_vp->v_mount != NULL &&
2202 (bp->b_vp->v_mount->mnt_kern_flag & MNTK_SUSPENDED) != 0)
2203 panic("ffs_geom_strategy: bad I/O");
2204 nocopy = bp->b_flags & B_NOCOPY;
2205 bp->b_flags &= ~(B_VALIDSUSPWRT | B_NOCOPY);
2206 if ((vp->v_vflag & VV_COPYONWRITE) && nocopy == 0 &&
2207 vp->v_rdev->si_snapdata != NULL) {
2208 if ((bp->b_flags & B_CLUSTER) != 0) {
2209 runningbufwakeup(bp);
2210 TAILQ_FOREACH(tbp, &bp->b_cluster.cluster_head,
2211 b_cluster.cluster_entry) {
2212 error = ffs_copyonwrite(vp, tbp);
2214 error != EOPNOTSUPP) {
2215 bp->b_error = error;
2216 bp->b_ioflags |= BIO_ERROR;
2221 bp->b_runningbufspace = bp->b_bufsize;
2222 atomic_add_long(&runningbufspace,
2223 bp->b_runningbufspace);
2225 error = ffs_copyonwrite(vp, bp);
2226 if (error != 0 && error != EOPNOTSUPP) {
2227 bp->b_error = error;
2228 bp->b_ioflags |= BIO_ERROR;
2235 if ((bp->b_flags & B_CLUSTER) != 0) {
2236 TAILQ_FOREACH(tbp, &bp->b_cluster.cluster_head,
2237 b_cluster.cluster_entry) {
2238 if (!LIST_EMPTY(&tbp->b_dep))
2242 if (!LIST_EMPTY(&bp->b_dep))
2248 g_vfs_strategy(bo, bp);
2252 ffs_own_mount(const struct mount *mp)
2255 if (mp->mnt_op == &ufs_vfsops)
2263 /* defined in ffs_softdep.c */
2264 extern void db_print_ffs(struct ufsmount *ump);
2266 DB_SHOW_COMMAND(ffs, db_show_ffs)
2269 struct ufsmount *ump;
2272 ump = VFSTOUFS((struct mount *)addr);
2277 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2278 if (!strcmp(mp->mnt_stat.f_fstypename, ufs_vfsconf.vfc_name))
2279 db_print_ffs(VFSTOUFS(mp));
2283 #endif /* SOFTUPDATES */