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 * 4. 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/kernel.h>
46 #include <sys/vnode.h>
47 #include <sys/mount.h>
51 #include <sys/fcntl.h>
52 #include <sys/ioccom.h>
53 #include <sys/malloc.h>
54 #include <sys/mutex.h>
56 #include <security/mac/mac_framework.h>
58 #include <ufs/ufs/extattr.h>
59 #include <ufs/ufs/gjournal.h>
60 #include <ufs/ufs/quota.h>
61 #include <ufs/ufs/ufsmount.h>
62 #include <ufs/ufs/inode.h>
63 #include <ufs/ufs/ufs_extern.h>
65 #include <ufs/ffs/fs.h>
66 #include <ufs/ffs/ffs_extern.h>
70 #include <vm/vm_page.h>
72 #include <geom/geom.h>
73 #include <geom/geom_vfs.h>
77 static uma_zone_t uma_inode, uma_ufs1, uma_ufs2;
79 static int ffs_mountfs(struct vnode *, struct mount *, struct thread *);
80 static void ffs_oldfscompat_read(struct fs *, struct ufsmount *,
82 static void ffs_ifree(struct ufsmount *ump, struct inode *ip);
83 static int ffs_sync_lazy(struct mount *mp);
85 static vfs_init_t ffs_init;
86 static vfs_uninit_t ffs_uninit;
87 static vfs_extattrctl_t ffs_extattrctl;
88 static vfs_cmount_t ffs_cmount;
89 static vfs_unmount_t ffs_unmount;
90 static vfs_mount_t ffs_mount;
91 static vfs_statfs_t ffs_statfs;
92 static vfs_fhtovp_t ffs_fhtovp;
93 static vfs_sync_t ffs_sync;
95 static struct vfsops ufs_vfsops = {
96 .vfs_extattrctl = ffs_extattrctl,
97 .vfs_fhtovp = ffs_fhtovp,
99 .vfs_mount = ffs_mount,
100 .vfs_cmount = ffs_cmount,
101 .vfs_quotactl = ufs_quotactl,
102 .vfs_root = ufs_root,
103 .vfs_statfs = ffs_statfs,
104 .vfs_sync = ffs_sync,
105 .vfs_uninit = ffs_uninit,
106 .vfs_unmount = ffs_unmount,
107 .vfs_vget = ffs_vget,
108 .vfs_susp_clean = process_deferred_inactive,
111 VFS_SET(ufs_vfsops, ufs, 0);
112 MODULE_VERSION(ufs, 1);
114 static b_strategy_t ffs_geom_strategy;
115 static b_write_t ffs_bufwrite;
117 static struct buf_ops ffs_ops = {
119 .bop_write = ffs_bufwrite,
120 .bop_strategy = ffs_geom_strategy,
122 #ifdef NO_FFS_SNAPSHOT
123 .bop_bdflush = bufbdflush,
125 .bop_bdflush = ffs_bdflush,
130 * Note that userquota and groupquota options are not currently used
131 * by UFS/FFS code and generally mount(8) does not pass those options
132 * from userland, but they can be passed by loader(8) via
133 * vfs.root.mountfrom.options.
135 static const char *ffs_opts[] = { "acls", "async", "noatime", "noclusterr",
136 "noclusterw", "noexec", "export", "force", "from", "groupquota",
137 "multilabel", "nfsv4acls", "fsckpid", "snapshot", "nosuid", "suiddir",
138 "nosymfollow", "sync", "union", "userquota", NULL };
141 ffs_mount(struct mount *mp)
145 struct ufsmount *ump = 0;
151 struct nameidata ndp;
155 if (vfs_filteropt(mp->mnt_optnew, ffs_opts))
157 if (uma_inode == NULL) {
158 uma_inode = uma_zcreate("FFS inode",
159 sizeof(struct inode), NULL, NULL, NULL, NULL,
161 uma_ufs1 = uma_zcreate("FFS1 dinode",
162 sizeof(struct ufs1_dinode), NULL, NULL, NULL, NULL,
164 uma_ufs2 = uma_zcreate("FFS2 dinode",
165 sizeof(struct ufs2_dinode), NULL, NULL, NULL, NULL,
169 vfs_deleteopt(mp->mnt_optnew, "groupquota");
170 vfs_deleteopt(mp->mnt_optnew, "userquota");
172 fspec = vfs_getopts(mp->mnt_optnew, "from", &error);
177 if (vfs_getopt(mp->mnt_optnew, "acls", NULL, NULL) == 0)
178 mntorflags |= MNT_ACLS;
180 if (vfs_getopt(mp->mnt_optnew, "snapshot", NULL, NULL) == 0) {
181 mntorflags |= MNT_SNAPSHOT;
183 * Once we have set the MNT_SNAPSHOT flag, do not
184 * persist "snapshot" in the options list.
186 vfs_deleteopt(mp->mnt_optnew, "snapshot");
187 vfs_deleteopt(mp->mnt_opt, "snapshot");
190 if (vfs_getopt(mp->mnt_optnew, "fsckpid", NULL, NULL) == 0 &&
191 vfs_scanopt(mp->mnt_optnew, "fsckpid", "%d", &fsckpid) == 1) {
193 * Once we have set the restricted PID, do not
194 * persist "fsckpid" in the options list.
196 vfs_deleteopt(mp->mnt_optnew, "fsckpid");
197 vfs_deleteopt(mp->mnt_opt, "fsckpid");
198 if (mp->mnt_flag & MNT_UPDATE) {
199 if (VFSTOUFS(mp)->um_fs->fs_ronly == 0 &&
200 vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0) == 0) {
202 "Checker enable: Must be read-only");
205 } else if (vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0) == 0) {
207 "Checker enable: Must be read-only");
210 /* Set to -1 if we are done */
215 if (vfs_getopt(mp->mnt_optnew, "nfsv4acls", NULL, NULL) == 0) {
216 if (mntorflags & MNT_ACLS) {
218 "\"acls\" and \"nfsv4acls\" options "
219 "are mutually exclusive");
222 mntorflags |= MNT_NFS4ACLS;
226 mp->mnt_flag |= mntorflags;
229 * If updating, check whether changing from read-only to
230 * read/write; if there is no device name, that's all we do.
232 if (mp->mnt_flag & MNT_UPDATE) {
235 devvp = ump->um_devvp;
236 if (fsckpid == -1 && ump->um_fsckpid > 0) {
237 if ((error = ffs_flushfiles(mp, WRITECLOSE, td)) != 0 ||
238 (error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0)
243 * Return to normal read-only mode.
245 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)
258 vn_finished_write(mp);
259 if ((error = vfs_write_suspend(mp)) != 0)
262 if (mp->mnt_kern_flag & MNTK_SUSPENDED) {
264 * Allow the secondary writes
267 mp->mnt_kern_flag &= ~(MNTK_SUSPENDED |
269 wakeup(&mp->mnt_flag);
272 * Allow the curthread to
273 * ignore the suspension to
274 * synchronize on-disk state.
276 td->td_pflags |= TDP_IGNSUSP;
280 vn_start_write(NULL, &mp, V_WAIT);
283 * Check for and optionally get rid of files open
287 if (mp->mnt_flag & MNT_FORCE)
289 if (MOUNTEDSOFTDEP(mp)) {
290 error = softdep_flushfiles(mp, flags, td);
292 error = ffs_flushfiles(mp, flags, td);
295 vfs_write_resume(mp);
298 if (fs->fs_pendingblocks != 0 ||
299 fs->fs_pendinginodes != 0) {
300 printf("WARNING: %s Update error: blocks %jd "
301 "files %d\n", fs->fs_fsmnt,
302 (intmax_t)fs->fs_pendingblocks,
303 fs->fs_pendinginodes);
304 fs->fs_pendingblocks = 0;
305 fs->fs_pendinginodes = 0;
307 if ((fs->fs_flags & (FS_UNCLEAN | FS_NEEDSFSCK)) == 0)
309 if ((error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0) {
312 vfs_write_resume(mp);
315 if (MOUNTEDSOFTDEP(mp))
320 * Drop our write and exclusive access.
322 g_access(ump->um_cp, 0, -1, -1);
327 mp->mnt_flag |= MNT_RDONLY;
330 * Allow the writers to note that filesystem
333 vfs_write_resume(mp);
335 if ((mp->mnt_flag & MNT_RELOAD) &&
336 (error = ffs_reload(mp, td, 0)) != 0)
339 !vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0)) {
341 * If we are running a checker, do not allow upgrade.
343 if (ump->um_fsckpid > 0) {
345 "Active checker, cannot upgrade to write");
349 * If upgrade to read-write by non-root, then verify
350 * that user has necessary permissions on the device.
352 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
353 error = VOP_ACCESS(devvp, VREAD | VWRITE,
356 error = priv_check(td, PRIV_VFS_MOUNT_PERM);
358 VOP_UNLOCK(devvp, 0);
361 VOP_UNLOCK(devvp, 0);
362 fs->fs_flags &= ~FS_UNCLEAN;
363 if (fs->fs_clean == 0) {
364 fs->fs_flags |= FS_UNCLEAN;
365 if ((mp->mnt_flag & MNT_FORCE) ||
367 (FS_SUJ | FS_NEEDSFSCK)) == 0 &&
368 (fs->fs_flags & FS_DOSOFTDEP))) {
369 printf("WARNING: %s was not properly "
370 "dismounted\n", fs->fs_fsmnt);
373 "R/W mount of %s denied. %s.%s",
375 "Filesystem is not clean - run fsck",
376 (fs->fs_flags & FS_SUJ) == 0 ? "" :
377 " Forced mount will invalidate"
378 " journal contents");
385 * Request exclusive write access.
387 error = g_access(ump->um_cp, 0, 1, 1);
392 if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0)
396 mp->mnt_flag &= ~MNT_RDONLY;
398 fs->fs_mtime = time_second;
399 /* check to see if we need to start softdep */
400 if ((fs->fs_flags & FS_DOSOFTDEP) &&
401 (error = softdep_mount(devvp, mp, fs, td->td_ucred))){
402 vn_finished_write(mp);
406 if ((error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0) {
407 vn_finished_write(mp);
410 if (fs->fs_snapinum[0] != 0)
411 ffs_snapshot_mount(mp);
412 vn_finished_write(mp);
415 * Soft updates is incompatible with "async",
416 * so if we are doing softupdates stop the user
417 * from setting the async flag in an update.
418 * Softdep_mount() clears it in an initial mount
421 if (MOUNTEDSOFTDEP(mp)) {
422 /* XXX: Reset too late ? */
424 mp->mnt_flag &= ~MNT_ASYNC;
428 * Keep MNT_ACLS flag if it is stored in superblock.
430 if ((fs->fs_flags & FS_ACLS) != 0) {
431 /* XXX: Set too late ? */
433 mp->mnt_flag |= MNT_ACLS;
437 if ((fs->fs_flags & FS_NFS4ACLS) != 0) {
438 /* XXX: Set too late ? */
440 mp->mnt_flag |= MNT_NFS4ACLS;
444 * If this is a request from fsck to clean up the filesystem,
445 * then allow the specified pid to proceed.
448 if (ump->um_fsckpid != 0) {
450 "Active checker already running on %s",
454 KASSERT(MOUNTEDSOFTDEP(mp) == 0,
455 ("soft updates enabled on read-only file system"));
459 * Request write access.
461 error = g_access(ump->um_cp, 0, 1, 0);
466 "Checker activation failed on %s",
470 ump->um_fsckpid = fsckpid;
471 if (fs->fs_snapinum[0] != 0)
472 ffs_snapshot_mount(mp);
473 fs->fs_mtime = time_second;
476 (void) ffs_sbupdate(ump, MNT_WAIT, 0);
480 * If this is a snapshot request, take the snapshot.
482 if (mp->mnt_flag & MNT_SNAPSHOT)
483 return (ffs_snapshot(mp, fspec));
487 * Not an update, or updating the name: look up the name
488 * and verify that it refers to a sensible disk device.
490 NDINIT(&ndp, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, fspec, td);
491 if ((error = namei(&ndp)) != 0)
493 NDFREE(&ndp, NDF_ONLY_PNBUF);
495 if (!vn_isdisk(devvp, &error)) {
501 * If mount by non-root, then verify that user has necessary
502 * permissions on the device.
505 if ((mp->mnt_flag & MNT_RDONLY) == 0)
507 error = VOP_ACCESS(devvp, accmode, td->td_ucred, td);
509 error = priv_check(td, PRIV_VFS_MOUNT_PERM);
515 if (mp->mnt_flag & MNT_UPDATE) {
519 * If it's not the same vnode, or at least the same device
520 * then it's not correct.
523 if (devvp->v_rdev != ump->um_devvp->v_rdev)
524 error = EINVAL; /* needs translation */
532 * We need the name for the mount point (also used for
533 * "last mounted on") copied in. If an error occurs,
534 * the mount point is discarded by the upper level code.
535 * Note that vfs_mount() populates f_mntonname for us.
537 if ((error = ffs_mountfs(devvp, mp, td)) != 0) {
542 KASSERT(MOUNTEDSOFTDEP(mp) == 0,
543 ("soft updates enabled on read-only file system"));
549 * Request write access.
551 error = g_access(ump->um_cp, 0, 1, 0);
555 printf("WARNING: %s: Checker activation "
556 "failed\n", fs->fs_fsmnt);
558 ump->um_fsckpid = fsckpid;
559 if (fs->fs_snapinum[0] != 0)
560 ffs_snapshot_mount(mp);
561 fs->fs_mtime = time_second;
563 (void) ffs_sbupdate(ump, MNT_WAIT, 0);
567 vfs_mountedfrom(mp, fspec);
572 * Compatibility with old mount system call.
576 ffs_cmount(struct mntarg *ma, void *data, uint64_t flags)
578 struct ufs_args args;
579 struct export_args exp;
584 error = copyin(data, &args, sizeof args);
587 vfs_oexport_conv(&args.export, &exp);
589 ma = mount_argsu(ma, "from", args.fspec, MAXPATHLEN);
590 ma = mount_arg(ma, "export", &exp, sizeof(exp));
591 error = kernel_mount(ma, flags);
597 * Reload all incore data for a filesystem (used after running fsck on
598 * the root filesystem and finding things to fix). If the 'force' flag
599 * is 0, the filesystem must be mounted read-only.
601 * Things to do to update the mount:
602 * 1) invalidate all cached meta-data.
603 * 2) re-read superblock from disk.
604 * 3) re-read summary information from disk.
605 * 4) invalidate all inactive vnodes.
606 * 5) invalidate all cached file data.
607 * 6) re-read inode data for all active vnodes.
610 ffs_reload(struct mount *mp, struct thread *td, int force)
612 struct vnode *vp, *mvp, *devvp;
616 struct fs *fs, *newfs;
617 struct ufsmount *ump;
618 ufs2_daddr_t sblockloc;
619 int i, blks, size, error;
625 if ((mp->mnt_flag & MNT_RDONLY) == 0 && force == 0) {
632 * Step 1: invalidate all cached meta-data.
634 devvp = VFSTOUFS(mp)->um_devvp;
635 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
636 if (vinvalbuf(devvp, 0, 0, 0) != 0)
637 panic("ffs_reload: dirty1");
638 VOP_UNLOCK(devvp, 0);
641 * Step 2: re-read superblock from disk.
643 fs = VFSTOUFS(mp)->um_fs;
644 if ((error = bread(devvp, btodb(fs->fs_sblockloc), fs->fs_sbsize,
647 newfs = (struct fs *)bp->b_data;
648 if ((newfs->fs_magic != FS_UFS1_MAGIC &&
649 newfs->fs_magic != FS_UFS2_MAGIC) ||
650 newfs->fs_bsize > MAXBSIZE ||
651 newfs->fs_bsize < sizeof(struct fs)) {
653 return (EIO); /* XXX needs translation */
656 * Copy pointer fields back into superblock before copying in XXX
657 * new superblock. These should really be in the ufsmount. XXX
658 * Note that important parameters (eg fs_ncg) are unchanged.
660 newfs->fs_csp = fs->fs_csp;
661 newfs->fs_maxcluster = fs->fs_maxcluster;
662 newfs->fs_contigdirs = fs->fs_contigdirs;
663 newfs->fs_active = fs->fs_active;
664 newfs->fs_ronly = fs->fs_ronly;
665 sblockloc = fs->fs_sblockloc;
666 bcopy(newfs, fs, (u_int)fs->fs_sbsize);
668 mp->mnt_maxsymlinklen = fs->fs_maxsymlinklen;
669 ffs_oldfscompat_read(fs, VFSTOUFS(mp), sblockloc);
671 if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
672 printf("WARNING: %s: reload pending error: blocks %jd "
673 "files %d\n", fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
674 fs->fs_pendinginodes);
675 fs->fs_pendingblocks = 0;
676 fs->fs_pendinginodes = 0;
681 * Step 3: re-read summary information from disk.
683 size = fs->fs_cssize;
684 blks = howmany(size, fs->fs_fsize);
685 if (fs->fs_contigsumsize > 0)
686 size += fs->fs_ncg * sizeof(int32_t);
687 size += fs->fs_ncg * sizeof(u_int8_t);
688 free(fs->fs_csp, M_UFSMNT);
689 space = malloc((u_long)size, M_UFSMNT, M_WAITOK);
691 for (i = 0; i < blks; i += fs->fs_frag) {
693 if (i + fs->fs_frag > blks)
694 size = (blks - i) * fs->fs_fsize;
695 error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), size,
699 bcopy(bp->b_data, space, (u_int)size);
700 space = (char *)space + size;
704 * We no longer know anything about clusters per cylinder group.
706 if (fs->fs_contigsumsize > 0) {
707 fs->fs_maxcluster = lp = space;
708 for (i = 0; i < fs->fs_ncg; i++)
709 *lp++ = fs->fs_contigsumsize;
712 size = fs->fs_ncg * sizeof(u_int8_t);
713 fs->fs_contigdirs = (u_int8_t *)space;
714 bzero(fs->fs_contigdirs, size);
717 MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
721 if (vp->v_type == VNON) {
726 * Step 4: invalidate all cached file data.
728 if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, td)) {
729 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
732 if (vinvalbuf(vp, 0, 0, 0))
733 panic("ffs_reload: dirty2");
735 * Step 5: re-read inode data for all active vnodes.
739 bread(devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
740 (int)fs->fs_bsize, NOCRED, &bp);
744 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
747 ffs_load_inode(bp, ip, fs, ip->i_number);
748 ip->i_effnlink = ip->i_nlink;
757 * Possible superblock locations ordered from most to least likely.
759 static int sblock_try[] = SBLOCKSEARCH;
762 * Common code for mount and mountroot
765 ffs_mountfs(devvp, mp, td)
770 struct ufsmount *ump;
775 ufs2_daddr_t sblockloc;
776 int error, i, blks, size, ronly;
779 struct g_consumer *cp;
784 cred = td ? td->td_ucred : NOCRED;
785 ronly = (mp->mnt_flag & MNT_RDONLY) != 0;
791 error = g_vfs_open(devvp, &cp, "ffs", ronly ? 0 : 1);
794 VOP_UNLOCK(devvp, 0);
797 if (devvp->v_rdev->si_iosize_max != 0)
798 mp->mnt_iosize_max = devvp->v_rdev->si_iosize_max;
799 if (mp->mnt_iosize_max > MAXPHYS)
800 mp->mnt_iosize_max = MAXPHYS;
802 devvp->v_bufobj.bo_ops = &ffs_ops;
807 * Try reading the superblock in each of its possible locations.
809 for (i = 0; sblock_try[i] != -1; i++) {
810 if ((SBLOCKSIZE % cp->provider->sectorsize) != 0) {
813 "Invalid sectorsize %d for superblock size %d",
814 cp->provider->sectorsize, SBLOCKSIZE);
817 if ((error = bread(devvp, btodb(sblock_try[i]), SBLOCKSIZE,
820 fs = (struct fs *)bp->b_data;
821 sblockloc = sblock_try[i];
822 if ((fs->fs_magic == FS_UFS1_MAGIC ||
823 (fs->fs_magic == FS_UFS2_MAGIC &&
824 (fs->fs_sblockloc == sblockloc ||
825 (fs->fs_old_flags & FS_FLAGS_UPDATED) == 0))) &&
826 fs->fs_bsize <= MAXBSIZE &&
827 fs->fs_bsize >= sizeof(struct fs))
832 if (sblock_try[i] == -1) {
833 error = EINVAL; /* XXX needs translation */
837 fs->fs_flags &= ~FS_INDEXDIRS; /* no support for directory indicies */
838 fs->fs_flags &= ~FS_UNCLEAN;
839 if (fs->fs_clean == 0) {
840 fs->fs_flags |= FS_UNCLEAN;
841 if (ronly || (mp->mnt_flag & MNT_FORCE) ||
842 ((fs->fs_flags & (FS_SUJ | FS_NEEDSFSCK)) == 0 &&
843 (fs->fs_flags & FS_DOSOFTDEP))) {
844 printf("WARNING: %s was not properly dismounted\n",
847 vfs_mount_error(mp, "R/W mount of %s denied. %s%s",
848 fs->fs_fsmnt, "Filesystem is not clean - run fsck.",
849 (fs->fs_flags & FS_SUJ) == 0 ? "" :
850 " Forced mount will invalidate journal contents");
854 if ((fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) &&
855 (mp->mnt_flag & MNT_FORCE)) {
856 printf("WARNING: %s: lost blocks %jd files %d\n",
857 fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
858 fs->fs_pendinginodes);
859 fs->fs_pendingblocks = 0;
860 fs->fs_pendinginodes = 0;
863 if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
864 printf("WARNING: %s: mount pending error: blocks %jd "
865 "files %d\n", fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
866 fs->fs_pendinginodes);
867 fs->fs_pendingblocks = 0;
868 fs->fs_pendinginodes = 0;
870 if ((fs->fs_flags & FS_GJOURNAL) != 0) {
873 * Get journal provider name.
876 mp->mnt_gjprovider = malloc(size, M_UFSMNT, M_WAITOK);
877 if (g_io_getattr("GJOURNAL::provider", cp, &size,
878 mp->mnt_gjprovider) == 0) {
879 mp->mnt_gjprovider = realloc(mp->mnt_gjprovider, size,
882 mp->mnt_flag |= MNT_GJOURNAL;
885 printf("WARNING: %s: GJOURNAL flag on fs "
886 "but no gjournal provider below\n",
887 mp->mnt_stat.f_mntonname);
888 free(mp->mnt_gjprovider, M_UFSMNT);
889 mp->mnt_gjprovider = NULL;
892 printf("WARNING: %s: GJOURNAL flag on fs but no "
893 "UFS_GJOURNAL support\n", mp->mnt_stat.f_mntonname);
896 mp->mnt_gjprovider = NULL;
898 ump = malloc(sizeof *ump, M_UFSMNT, M_WAITOK | M_ZERO);
900 ump->um_bo = &devvp->v_bufobj;
901 ump->um_fs = malloc((u_long)fs->fs_sbsize, M_UFSMNT, M_WAITOK);
902 if (fs->fs_magic == FS_UFS1_MAGIC) {
903 ump->um_fstype = UFS1;
904 ump->um_balloc = ffs_balloc_ufs1;
906 ump->um_fstype = UFS2;
907 ump->um_balloc = ffs_balloc_ufs2;
909 ump->um_blkatoff = ffs_blkatoff;
910 ump->um_truncate = ffs_truncate;
911 ump->um_update = ffs_update;
912 ump->um_valloc = ffs_valloc;
913 ump->um_vfree = ffs_vfree;
914 ump->um_ifree = ffs_ifree;
915 ump->um_rdonly = ffs_rdonly;
916 ump->um_snapgone = ffs_snapgone;
917 mtx_init(UFS_MTX(ump), "FFS", "FFS Lock", MTX_DEF);
918 bcopy(bp->b_data, ump->um_fs, (u_int)fs->fs_sbsize);
919 if (fs->fs_sbsize < SBLOCKSIZE)
920 bp->b_flags |= B_INVAL | B_NOCACHE;
924 ffs_oldfscompat_read(fs, ump, sblockloc);
925 fs->fs_ronly = ronly;
926 size = fs->fs_cssize;
927 blks = howmany(size, fs->fs_fsize);
928 if (fs->fs_contigsumsize > 0)
929 size += fs->fs_ncg * sizeof(int32_t);
930 size += fs->fs_ncg * sizeof(u_int8_t);
931 space = malloc((u_long)size, M_UFSMNT, M_WAITOK);
933 for (i = 0; i < blks; i += fs->fs_frag) {
935 if (i + fs->fs_frag > blks)
936 size = (blks - i) * fs->fs_fsize;
937 if ((error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), size,
939 free(fs->fs_csp, M_UFSMNT);
942 bcopy(bp->b_data, space, (u_int)size);
943 space = (char *)space + size;
947 if (fs->fs_contigsumsize > 0) {
948 fs->fs_maxcluster = lp = space;
949 for (i = 0; i < fs->fs_ncg; i++)
950 *lp++ = fs->fs_contigsumsize;
953 size = fs->fs_ncg * sizeof(u_int8_t);
954 fs->fs_contigdirs = (u_int8_t *)space;
955 bzero(fs->fs_contigdirs, size);
956 fs->fs_active = NULL;
958 mp->mnt_stat.f_fsid.val[0] = fs->fs_id[0];
959 mp->mnt_stat.f_fsid.val[1] = fs->fs_id[1];
961 if (fs->fs_id[0] == 0 || fs->fs_id[1] == 0 ||
962 (nmp = vfs_getvfs(&mp->mnt_stat.f_fsid))) {
967 mp->mnt_maxsymlinklen = fs->fs_maxsymlinklen;
969 mp->mnt_flag |= MNT_LOCAL;
971 if ((fs->fs_flags & FS_MULTILABEL) != 0) {
974 mp->mnt_flag |= MNT_MULTILABEL;
977 printf("WARNING: %s: multilabel flag on fs but "
978 "no MAC support\n", mp->mnt_stat.f_mntonname);
981 if ((fs->fs_flags & FS_ACLS) != 0) {
985 if (mp->mnt_flag & MNT_NFS4ACLS)
986 printf("WARNING: %s: ACLs flag on fs conflicts with "
987 "\"nfsv4acls\" mount option; option ignored\n",
988 mp->mnt_stat.f_mntonname);
989 mp->mnt_flag &= ~MNT_NFS4ACLS;
990 mp->mnt_flag |= MNT_ACLS;
994 printf("WARNING: %s: ACLs flag on fs but no ACLs support\n",
995 mp->mnt_stat.f_mntonname);
998 if ((fs->fs_flags & FS_NFS4ACLS) != 0) {
1002 if (mp->mnt_flag & MNT_ACLS)
1003 printf("WARNING: %s: NFSv4 ACLs flag on fs conflicts "
1004 "with \"acls\" mount option; option ignored\n",
1005 mp->mnt_stat.f_mntonname);
1006 mp->mnt_flag &= ~MNT_ACLS;
1007 mp->mnt_flag |= MNT_NFS4ACLS;
1011 printf("WARNING: %s: NFSv4 ACLs flag on fs but no "
1012 "ACLs support\n", mp->mnt_stat.f_mntonname);
1015 if ((fs->fs_flags & FS_TRIM) != 0) {
1017 if (g_io_getattr("GEOM::candelete", cp, &size,
1018 &ump->um_candelete) == 0) {
1019 if (!ump->um_candelete)
1020 printf("WARNING: %s: TRIM flag on fs but disk "
1021 "does not support TRIM\n",
1022 mp->mnt_stat.f_mntonname);
1024 printf("WARNING: %s: TRIM flag on fs but disk does "
1025 "not confirm that it supports TRIM\n",
1026 mp->mnt_stat.f_mntonname);
1027 ump->um_candelete = 0;
1031 ump->um_mountp = mp;
1033 ump->um_devvp = devvp;
1034 ump->um_nindir = fs->fs_nindir;
1035 ump->um_bptrtodb = fs->fs_fsbtodb;
1036 ump->um_seqinc = fs->fs_frag;
1037 for (i = 0; i < MAXQUOTAS; i++)
1038 ump->um_quotas[i] = NULLVP;
1040 ufs_extattr_uepm_init(&ump->um_extattr);
1043 * Set FS local "last mounted on" information (NULL pad)
1045 bzero(fs->fs_fsmnt, MAXMNTLEN);
1046 strlcpy(fs->fs_fsmnt, mp->mnt_stat.f_mntonname, MAXMNTLEN);
1047 mp->mnt_stat.f_iosize = fs->fs_bsize;
1049 if (mp->mnt_flag & MNT_ROOTFS) {
1051 * Root mount; update timestamp in mount structure.
1052 * this will be used by the common root mount code
1053 * to update the system clock.
1055 mp->mnt_time = fs->fs_time;
1059 fs->fs_mtime = time_second;
1060 if ((fs->fs_flags & FS_DOSOFTDEP) &&
1061 (error = softdep_mount(devvp, mp, fs, cred)) != 0) {
1062 free(fs->fs_csp, M_UFSMNT);
1063 ffs_flushfiles(mp, FORCECLOSE, td);
1066 if (devvp->v_type == VCHR && devvp->v_rdev != NULL)
1067 devvp->v_rdev->si_mountpt = mp;
1068 if (fs->fs_snapinum[0] != 0)
1069 ffs_snapshot_mount(mp);
1072 (void) ffs_sbupdate(ump, MNT_WAIT, 0);
1075 * Initialize filesystem stat information in mount struct.
1078 mp->mnt_kern_flag |= MNTK_MPSAFE | MNTK_LOOKUP_SHARED |
1079 MNTK_EXTENDED_SHARED | MNTK_NO_IOPF | MNTK_UNMAPPED_BUFS;
1082 #ifdef UFS_EXTATTR_AUTOSTART
1085 * Auto-starting does the following:
1086 * - check for /.attribute in the fs, and extattr_start if so
1087 * - for each file in .attribute, enable that file with
1088 * an attribute of the same name.
1089 * Not clear how to report errors -- probably eat them.
1090 * This would all happen while the filesystem was busy/not
1091 * available, so would effectively be "atomic".
1093 (void) ufs_extattr_autostart(mp, td);
1094 #endif /* !UFS_EXTATTR_AUTOSTART */
1095 #endif /* !UFS_EXTATTR */
1104 g_topology_unlock();
1108 mtx_destroy(UFS_MTX(ump));
1109 if (mp->mnt_gjprovider != NULL) {
1110 free(mp->mnt_gjprovider, M_UFSMNT);
1111 mp->mnt_gjprovider = NULL;
1113 free(ump->um_fs, M_UFSMNT);
1114 free(ump, M_UFSMNT);
1115 mp->mnt_data = NULL;
1121 #include <sys/sysctl.h>
1122 static int bigcgs = 0;
1123 SYSCTL_INT(_debug, OID_AUTO, bigcgs, CTLFLAG_RW, &bigcgs, 0, "");
1126 * Sanity checks for loading old filesystem superblocks.
1127 * See ffs_oldfscompat_write below for unwound actions.
1129 * XXX - Parts get retired eventually.
1130 * Unfortunately new bits get added.
1133 ffs_oldfscompat_read(fs, ump, sblockloc)
1135 struct ufsmount *ump;
1136 ufs2_daddr_t sblockloc;
1141 * If not yet done, update fs_flags location and value of fs_sblockloc.
1143 if ((fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) {
1144 fs->fs_flags = fs->fs_old_flags;
1145 fs->fs_old_flags |= FS_FLAGS_UPDATED;
1146 fs->fs_sblockloc = sblockloc;
1149 * If not yet done, update UFS1 superblock with new wider fields.
1151 if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_maxbsize != fs->fs_bsize) {
1152 fs->fs_maxbsize = fs->fs_bsize;
1153 fs->fs_time = fs->fs_old_time;
1154 fs->fs_size = fs->fs_old_size;
1155 fs->fs_dsize = fs->fs_old_dsize;
1156 fs->fs_csaddr = fs->fs_old_csaddr;
1157 fs->fs_cstotal.cs_ndir = fs->fs_old_cstotal.cs_ndir;
1158 fs->fs_cstotal.cs_nbfree = fs->fs_old_cstotal.cs_nbfree;
1159 fs->fs_cstotal.cs_nifree = fs->fs_old_cstotal.cs_nifree;
1160 fs->fs_cstotal.cs_nffree = fs->fs_old_cstotal.cs_nffree;
1162 if (fs->fs_magic == FS_UFS1_MAGIC &&
1163 fs->fs_old_inodefmt < FS_44INODEFMT) {
1164 fs->fs_maxfilesize = ((uint64_t)1 << 31) - 1;
1165 fs->fs_qbmask = ~fs->fs_bmask;
1166 fs->fs_qfmask = ~fs->fs_fmask;
1168 if (fs->fs_magic == FS_UFS1_MAGIC) {
1169 ump->um_savedmaxfilesize = fs->fs_maxfilesize;
1170 maxfilesize = (uint64_t)0x80000000 * fs->fs_bsize - 1;
1171 if (fs->fs_maxfilesize > maxfilesize)
1172 fs->fs_maxfilesize = maxfilesize;
1174 /* Compatibility for old filesystems */
1175 if (fs->fs_avgfilesize <= 0)
1176 fs->fs_avgfilesize = AVFILESIZ;
1177 if (fs->fs_avgfpdir <= 0)
1178 fs->fs_avgfpdir = AFPDIR;
1180 fs->fs_save_cgsize = fs->fs_cgsize;
1181 fs->fs_cgsize = fs->fs_bsize;
1186 * Unwinding superblock updates for old filesystems.
1187 * See ffs_oldfscompat_read above for details.
1189 * XXX - Parts get retired eventually.
1190 * Unfortunately new bits get added.
1193 ffs_oldfscompat_write(fs, ump)
1195 struct ufsmount *ump;
1199 * Copy back UFS2 updated fields that UFS1 inspects.
1201 if (fs->fs_magic == FS_UFS1_MAGIC) {
1202 fs->fs_old_time = fs->fs_time;
1203 fs->fs_old_cstotal.cs_ndir = fs->fs_cstotal.cs_ndir;
1204 fs->fs_old_cstotal.cs_nbfree = fs->fs_cstotal.cs_nbfree;
1205 fs->fs_old_cstotal.cs_nifree = fs->fs_cstotal.cs_nifree;
1206 fs->fs_old_cstotal.cs_nffree = fs->fs_cstotal.cs_nffree;
1207 fs->fs_maxfilesize = ump->um_savedmaxfilesize;
1210 fs->fs_cgsize = fs->fs_save_cgsize;
1211 fs->fs_save_cgsize = 0;
1216 * unmount system call
1219 ffs_unmount(mp, mntflags)
1224 struct ufsmount *ump = VFSTOUFS(mp);
1226 int error, flags, susp;
1235 if (mntflags & MNT_FORCE) {
1236 flags |= FORCECLOSE;
1237 susp = fs->fs_ronly != 0;
1240 if ((error = ufs_extattr_stop(mp, td))) {
1241 if (error != EOPNOTSUPP)
1242 printf("WARNING: unmount %s: ufs_extattr_stop "
1243 "returned errno %d\n", mp->mnt_stat.f_mntonname,
1247 ufs_extattr_uepm_destroy(&ump->um_extattr);
1253 * dounmount already called vn_start_write().
1256 vn_finished_write(mp);
1257 if ((error = vfs_write_suspend(mp)) != 0)
1260 if (mp->mnt_kern_flag & MNTK_SUSPENDED) {
1261 mp->mnt_kern_flag &= ~(MNTK_SUSPENDED |
1263 wakeup(&mp->mnt_flag);
1265 td->td_pflags |= TDP_IGNSUSP;
1269 vn_start_write(NULL, &mp, V_WAIT);
1272 if (MOUNTEDSOFTDEP(mp))
1273 error = softdep_flushfiles(mp, flags, td);
1275 error = ffs_flushfiles(mp, flags, td);
1276 if (error != 0 && error != ENXIO)
1280 if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
1281 printf("WARNING: unmount %s: pending error: blocks %jd "
1282 "files %d\n", fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
1283 fs->fs_pendinginodes);
1284 fs->fs_pendingblocks = 0;
1285 fs->fs_pendinginodes = 0;
1288 softdep_unmount(mp);
1289 if (fs->fs_ronly == 0 || ump->um_fsckpid > 0) {
1290 fs->fs_clean = fs->fs_flags & (FS_UNCLEAN|FS_NEEDSFSCK) ? 0 : 1;
1291 error = ffs_sbupdate(ump, MNT_WAIT, 0);
1292 if (error && error != ENXIO) {
1298 vfs_write_resume(mp);
1299 vn_start_write(NULL, &mp, V_WAIT);
1303 if (ump->um_fsckpid > 0) {
1305 * Return to normal read-only mode.
1307 error = g_access(ump->um_cp, 0, -1, 0);
1308 ump->um_fsckpid = 0;
1310 g_vfs_close(ump->um_cp);
1311 g_topology_unlock();
1313 if (ump->um_devvp->v_type == VCHR && ump->um_devvp->v_rdev != NULL)
1314 ump->um_devvp->v_rdev->si_mountpt = NULL;
1315 vrele(ump->um_devvp);
1316 dev_rel(ump->um_dev);
1317 mtx_destroy(UFS_MTX(ump));
1318 if (mp->mnt_gjprovider != NULL) {
1319 free(mp->mnt_gjprovider, M_UFSMNT);
1320 mp->mnt_gjprovider = NULL;
1322 free(fs->fs_csp, M_UFSMNT);
1324 free(ump, M_UFSMNT);
1325 mp->mnt_data = NULL;
1327 mp->mnt_flag &= ~MNT_LOCAL;
1333 vfs_write_resume(mp);
1334 vn_start_write(NULL, &mp, V_WAIT);
1338 ufs_extattr_uepm_init(&ump->um_extattr);
1339 #ifdef UFS_EXTATTR_AUTOSTART
1340 (void) ufs_extattr_autostart(mp, td);
1349 * Flush out all the files in a filesystem.
1352 ffs_flushfiles(mp, flags, td)
1357 struct ufsmount *ump;
1363 if (mp->mnt_flag & MNT_QUOTA) {
1365 error = vflush(mp, 0, SKIPSYSTEM|flags, td);
1368 for (i = 0; i < MAXQUOTAS; i++) {
1369 error = quotaoff(td, mp, i);
1371 if ((flags & EARLYFLUSH) == 0)
1379 * Here we fall through to vflush again to ensure that
1380 * we have gotten rid of all the system vnodes, unless
1381 * quotas must not be closed.
1385 ASSERT_VOP_LOCKED(ump->um_devvp, "ffs_flushfiles");
1386 if (ump->um_devvp->v_vflag & VV_COPYONWRITE) {
1387 if ((error = vflush(mp, 0, SKIPSYSTEM | flags, td)) != 0)
1389 ffs_snapshot_unmount(mp);
1390 flags |= FORCECLOSE;
1392 * Here we fall through to vflush again to ensure
1393 * that we have gotten rid of all the system vnodes.
1398 * Do not close system files if quotas were not closed, to be
1399 * able to sync the remaining dquots. The freeblks softupdate
1400 * workitems might hold a reference on a dquot, preventing
1401 * quotaoff() from completing. Next round of
1402 * softdep_flushworklist() iteration should process the
1403 * blockers, allowing the next run of quotaoff() to finally
1404 * flush held dquots.
1406 * Otherwise, flush all the files.
1408 if (qerror == 0 && (error = vflush(mp, 0, flags, td)) != 0)
1412 * Flush filesystem metadata.
1414 vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY);
1415 error = VOP_FSYNC(ump->um_devvp, MNT_WAIT, td);
1416 VOP_UNLOCK(ump->um_devvp, 0);
1421 * Get filesystem statistics.
1428 struct ufsmount *ump;
1433 if (fs->fs_magic != FS_UFS1_MAGIC && fs->fs_magic != FS_UFS2_MAGIC)
1434 panic("ffs_statfs");
1435 sbp->f_version = STATFS_VERSION;
1436 sbp->f_bsize = fs->fs_fsize;
1437 sbp->f_iosize = fs->fs_bsize;
1438 sbp->f_blocks = fs->fs_dsize;
1440 sbp->f_bfree = fs->fs_cstotal.cs_nbfree * fs->fs_frag +
1441 fs->fs_cstotal.cs_nffree + dbtofsb(fs, fs->fs_pendingblocks);
1442 sbp->f_bavail = freespace(fs, fs->fs_minfree) +
1443 dbtofsb(fs, fs->fs_pendingblocks);
1444 sbp->f_files = fs->fs_ncg * fs->fs_ipg - ROOTINO;
1445 sbp->f_ffree = fs->fs_cstotal.cs_nifree + fs->fs_pendinginodes;
1447 sbp->f_namemax = NAME_MAX;
1452 * For a lazy sync, we only care about access times, quotas and the
1453 * superblock. Other filesystem changes are already converted to
1454 * cylinder group blocks or inode blocks updates and are written to
1461 struct vnode *mvp, *vp;
1464 int allerror, error;
1468 if ((mp->mnt_flag & MNT_NOATIME) != 0)
1470 MNT_VNODE_FOREACH_ACTIVE(vp, mp, mvp) {
1471 if (vp->v_type == VNON) {
1478 * The IN_ACCESS flag is converted to IN_MODIFIED by
1479 * ufs_close() and ufs_getattr() by the calls to
1480 * ufs_itimes_locked(), without subsequent UFS_UPDATE().
1481 * Test also all the other timestamp flags too, to pick up
1482 * any other cases that could be missed.
1484 if ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_MODIFIED |
1489 if ((error = vget(vp, LK_EXCLUSIVE | LK_NOWAIT | LK_INTERLOCK,
1492 error = ffs_update(vp, 0);
1503 if (VFSTOUFS(mp)->um_fs->fs_fmod != 0 &&
1504 (error = ffs_sbupdate(VFSTOUFS(mp), MNT_LAZY, 0)) != 0)
1510 * Go through the disk queues to initiate sandbagged IO;
1511 * go through the inodes to write those that have been modified;
1512 * initiate the writing of the super block if it has been modified.
1514 * Note: we are always called with the filesystem marked busy using
1518 ffs_sync(mp, waitfor)
1522 struct vnode *mvp, *vp, *devvp;
1525 struct ufsmount *ump = VFSTOUFS(mp);
1527 int error, count, wait, lockreq, allerror = 0;
1530 int secondary_writes;
1531 int secondary_accwrites;
1533 int softdep_accdeps;
1541 if (fs->fs_fmod != 0 && fs->fs_ronly != 0 && ump->um_fsckpid == 0)
1542 panic("%s: ffs_sync: modification on read-only filesystem",
1544 if (waitfor == MNT_LAZY)
1545 return (ffs_sync_lazy(mp));
1548 * Write back each (modified) inode.
1550 lockreq = LK_EXCLUSIVE | LK_NOWAIT;
1551 if (waitfor == MNT_SUSPEND) {
1555 if (waitfor == MNT_WAIT) {
1557 lockreq = LK_EXCLUSIVE;
1559 lockreq |= LK_INTERLOCK | LK_SLEEPFAIL;
1561 /* Grab snapshot of secondary write counts */
1563 secondary_writes = mp->mnt_secondary_writes;
1564 secondary_accwrites = mp->mnt_secondary_accwrites;
1567 /* Grab snapshot of softdep dependency counts */
1568 softdep_get_depcounts(mp, &softdep_deps, &softdep_accdeps);
1570 MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
1572 * Depend on the vnode interlock to keep things stable enough
1573 * for a quick test. Since there might be hundreds of
1574 * thousands of vnodes, we cannot afford even a subroutine
1575 * call unless there's a good chance that we have work to do.
1577 if (vp->v_type == VNON) {
1583 (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) == 0 &&
1584 vp->v_bufobj.bo_dirty.bv_cnt == 0) {
1588 if ((error = vget(vp, lockreq, td)) != 0) {
1589 if (error == ENOENT || error == ENOLCK) {
1590 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
1595 if ((error = ffs_syncvnode(vp, waitfor, 0)) != 0)
1600 * Force stale filesystem control information to be flushed.
1602 if (waitfor == MNT_WAIT) {
1603 if ((error = softdep_flushworklist(ump->um_mountp, &count, td)))
1605 /* Flushed work items may create new vnodes to clean */
1606 if (allerror == 0 && count)
1613 devvp = ump->um_devvp;
1614 bo = &devvp->v_bufobj;
1616 if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0) {
1618 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
1619 if ((error = VOP_FSYNC(devvp, waitfor, td)) != 0)
1621 VOP_UNLOCK(devvp, 0);
1622 if (allerror == 0 && waitfor == MNT_WAIT)
1624 } else if (suspend != 0) {
1625 if (softdep_check_suspend(mp,
1630 secondary_accwrites) != 0) {
1632 goto loop; /* More work needed */
1634 mtx_assert(MNT_MTX(mp), MA_OWNED);
1635 mp->mnt_kern_flag |= MNTK_SUSPEND2 | MNTK_SUSPENDED;
1641 * Write back modified superblock.
1643 if (fs->fs_fmod != 0 &&
1644 (error = ffs_sbupdate(ump, waitfor, suspended)) != 0)
1650 ffs_vget(mp, ino, flags, vpp)
1656 return (ffs_vgetf(mp, ino, flags, vpp, 0));
1660 ffs_vgetf(mp, ino, flags, vpp, ffs_flags)
1669 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!
1700 * If this malloc() is performed after the getnewvnode()
1701 * it might block, leaving a vnode with a NULL v_data to be
1702 * found by ffs_sync() if a sync happens to fire right then,
1703 * which will cause a panic because ffs_sync() blindly
1704 * dereferences vp->v_data (as well it should).
1706 ip = uma_zalloc(uma_inode, M_WAITOK | M_ZERO);
1708 /* Allocate a new vnode/inode. */
1709 if (fs->fs_magic == FS_UFS1_MAGIC)
1710 error = getnewvnode("ufs", mp, &ffs_vnodeops1, &vp);
1712 error = getnewvnode("ufs", mp, &ffs_vnodeops2, &vp);
1715 uma_zfree(uma_inode, ip);
1719 * FFS supports recursive locking.
1721 lockmgr(vp->v_vnlock, LK_EXCLUSIVE, NULL);
1724 vp->v_bufobj.bo_bsize = fs->fs_bsize;
1734 for (i = 0; i < MAXQUOTAS; i++)
1735 ip->i_dquot[i] = NODQUOT;
1739 if (ffs_flags & FFSV_FORCEINSMQ)
1740 vp->v_vflag |= VV_FORCEINSMQ;
1741 error = insmntque(vp, mp);
1743 uma_zfree(uma_inode, ip);
1747 vp->v_vflag &= ~VV_FORCEINSMQ;
1748 error = vfs_hash_insert(vp, ino, flags, curthread, vpp, NULL, NULL);
1749 if (error || *vpp != NULL)
1752 /* Read in the disk contents for the inode, copy into the inode. */
1753 error = bread(ump->um_devvp, fsbtodb(fs, ino_to_fsba(fs, ino)),
1754 (int)fs->fs_bsize, NOCRED, &bp);
1757 * The inode does not contain anything useful, so it would
1758 * be misleading to leave it on its hash chain. With mode
1759 * still zero, it will be unlinked and returned to the free
1767 if (ip->i_ump->um_fstype == UFS1)
1768 ip->i_din1 = uma_zalloc(uma_ufs1, M_WAITOK);
1770 ip->i_din2 = uma_zalloc(uma_ufs2, M_WAITOK);
1771 ffs_load_inode(bp, ip, fs, ino);
1772 if (DOINGSOFTDEP(vp))
1773 softdep_load_inodeblock(ip);
1775 ip->i_effnlink = ip->i_nlink;
1779 * Initialize the vnode from the inode, check for aliases.
1780 * Note that the underlying vnode may have changed.
1782 if (ip->i_ump->um_fstype == UFS1)
1783 error = ufs_vinit(mp, &ffs_fifoops1, &vp);
1785 error = ufs_vinit(mp, &ffs_fifoops2, &vp);
1793 * Finish inode initialization.
1795 if (vp->v_type != VFIFO) {
1796 /* FFS supports shared locking for all files except fifos. */
1801 * Set up a generation number for this inode if it does not
1802 * already have one. This should only happen on old filesystems.
1804 if (ip->i_gen == 0) {
1805 ip->i_gen = arc4random() / 2 + 1;
1806 if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
1807 ip->i_flag |= IN_MODIFIED;
1808 DIP_SET(ip, i_gen, ip->i_gen);
1812 if ((mp->mnt_flag & MNT_MULTILABEL) && ip->i_mode) {
1814 * If this vnode is already allocated, and we're running
1815 * multi-label, attempt to perform a label association
1816 * from the extended attributes on the inode.
1818 error = mac_vnode_associate_extattr(mp, vp);
1820 /* ufs_inactive will release ip->i_devvp ref. */
1833 * File handle to vnode
1835 * Have to be really careful about stale file handles:
1836 * - check that the inode number is valid
1837 * - call ffs_vget() to get the locked inode
1838 * - check for an unallocated inode (i_mode == 0)
1839 * - check that the given client host has export rights and return
1840 * those rights via. exflagsp and credanonp
1843 ffs_fhtovp(mp, fhp, flags, vpp)
1852 ufhp = (struct ufid *)fhp;
1853 fs = VFSTOUFS(mp)->um_fs;
1854 if (ufhp->ufid_ino < ROOTINO ||
1855 ufhp->ufid_ino >= fs->fs_ncg * fs->fs_ipg)
1857 return (ufs_fhtovp(mp, ufhp, flags, vpp));
1861 * Initialize the filesystem.
1865 struct vfsconf *vfsp;
1868 ffs_susp_initialize();
1869 softdep_initialize();
1870 return (ufs_init(vfsp));
1874 * Undo the work of ffs_init().
1878 struct vfsconf *vfsp;
1882 ret = ufs_uninit(vfsp);
1883 softdep_uninitialize();
1884 ffs_susp_uninitialize();
1889 * Write a superblock and associated information back to disk.
1892 ffs_sbupdate(ump, waitfor, suspended)
1893 struct ufsmount *ump;
1897 struct fs *fs = ump->um_fs;
1902 int i, size, error, allerror = 0;
1904 if (fs->fs_ronly == 1 &&
1905 (ump->um_mountp->mnt_flag & (MNT_RDONLY | MNT_UPDATE)) !=
1906 (MNT_RDONLY | MNT_UPDATE) && ump->um_fsckpid == 0)
1907 panic("ffs_sbupdate: write read-only filesystem");
1909 * We use the superblock's buf to serialize calls to ffs_sbupdate().
1911 sbbp = getblk(ump->um_devvp, btodb(fs->fs_sblockloc),
1912 (int)fs->fs_sbsize, 0, 0, 0);
1914 * First write back the summary information.
1916 blks = howmany(fs->fs_cssize, fs->fs_fsize);
1918 for (i = 0; i < blks; i += fs->fs_frag) {
1919 size = fs->fs_bsize;
1920 if (i + fs->fs_frag > blks)
1921 size = (blks - i) * fs->fs_fsize;
1922 bp = getblk(ump->um_devvp, fsbtodb(fs, fs->fs_csaddr + i),
1924 bcopy(space, bp->b_data, (u_int)size);
1925 space = (char *)space + size;
1927 bp->b_flags |= B_VALIDSUSPWRT;
1928 if (waitfor != MNT_WAIT)
1930 else if ((error = bwrite(bp)) != 0)
1934 * Now write back the superblock itself. If any errors occurred
1935 * up to this point, then fail so that the superblock avoids
1936 * being written out as clean.
1943 if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_sblockloc != SBLOCK_UFS1 &&
1944 (fs->fs_flags & FS_FLAGS_UPDATED) == 0) {
1945 printf("WARNING: %s: correcting fs_sblockloc from %jd to %d\n",
1946 fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS1);
1947 fs->fs_sblockloc = SBLOCK_UFS1;
1949 if (fs->fs_magic == FS_UFS2_MAGIC && fs->fs_sblockloc != SBLOCK_UFS2 &&
1950 (fs->fs_flags & FS_FLAGS_UPDATED) == 0) {
1951 printf("WARNING: %s: correcting fs_sblockloc from %jd to %d\n",
1952 fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS2);
1953 fs->fs_sblockloc = SBLOCK_UFS2;
1956 fs->fs_time = time_second;
1957 if (fs->fs_flags & FS_DOSOFTDEP)
1958 softdep_setup_sbupdate(ump, (struct fs *)bp->b_data, bp);
1959 bcopy((caddr_t)fs, bp->b_data, (u_int)fs->fs_sbsize);
1960 ffs_oldfscompat_write((struct fs *)bp->b_data, ump);
1962 bp->b_flags |= B_VALIDSUSPWRT;
1963 if (waitfor != MNT_WAIT)
1965 else if ((error = bwrite(bp)) != 0)
1971 ffs_extattrctl(struct mount *mp, int cmd, struct vnode *filename_vp,
1972 int attrnamespace, const char *attrname)
1976 return (ufs_extattrctl(mp, cmd, filename_vp, attrnamespace,
1979 return (vfs_stdextattrctl(mp, cmd, filename_vp, attrnamespace,
1985 ffs_ifree(struct ufsmount *ump, struct inode *ip)
1988 if (ump->um_fstype == UFS1 && ip->i_din1 != NULL)
1989 uma_zfree(uma_ufs1, ip->i_din1);
1990 else if (ip->i_din2 != NULL)
1991 uma_zfree(uma_ufs2, ip->i_din2);
1992 uma_zfree(uma_inode, ip);
1995 static int dobkgrdwrite = 1;
1996 SYSCTL_INT(_debug, OID_AUTO, dobkgrdwrite, CTLFLAG_RW, &dobkgrdwrite, 0,
1997 "Do background writes (honoring the BV_BKGRDWRITE flag)?");
2000 * Complete a background write started from bwrite.
2003 ffs_backgroundwritedone(struct buf *bp)
2005 struct bufobj *bufobj;
2009 * Find the original buffer that we are writing.
2011 bufobj = bp->b_bufobj;
2013 if ((origbp = gbincore(bp->b_bufobj, bp->b_lblkno)) == NULL)
2014 panic("backgroundwritedone: lost buffer");
2015 /* Grab an extra reference to be dropped by the bufdone() below. */
2016 bufobj_wrefl(bufobj);
2019 * Process dependencies then return any unfinished ones.
2021 if (!LIST_EMPTY(&bp->b_dep))
2024 if (!LIST_EMPTY(&bp->b_dep))
2025 softdep_move_dependencies(bp, origbp);
2028 * This buffer is marked B_NOCACHE so when it is released
2029 * by biodone it will be tossed.
2031 bp->b_flags |= B_NOCACHE;
2032 bp->b_flags &= ~B_CACHE;
2036 * Clear the BV_BKGRDINPROG flag in the original buffer
2037 * and awaken it if it is waiting for the write to complete.
2038 * If BV_BKGRDINPROG is not set in the original buffer it must
2039 * have been released and re-instantiated - which is not legal.
2041 KASSERT((origbp->b_vflags & BV_BKGRDINPROG),
2042 ("backgroundwritedone: lost buffer2"));
2043 origbp->b_vflags &= ~BV_BKGRDINPROG;
2044 if (origbp->b_vflags & BV_BKGRDWAIT) {
2045 origbp->b_vflags &= ~BV_BKGRDWAIT;
2046 wakeup(&origbp->b_xflags);
2053 * Write, release buffer on completion. (Done by iodone
2054 * if async). Do not bother writing anything if the buffer
2057 * Note that we set B_CACHE here, indicating that buffer is
2058 * fully valid and thus cacheable. This is true even of NFS
2059 * now so we set it generally. This could be set either here
2060 * or in biodone() since the I/O is synchronous. We put it
2064 ffs_bufwrite(struct buf *bp)
2069 CTR3(KTR_BUF, "bufwrite(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
2070 if (bp->b_flags & B_INVAL) {
2075 oldflags = bp->b_flags;
2077 if (!BUF_ISLOCKED(bp))
2078 panic("bufwrite: buffer is not busy???");
2081 * If a background write is already in progress, delay
2082 * writing this block if it is asynchronous. Otherwise
2083 * wait for the background write to complete.
2085 BO_LOCK(bp->b_bufobj);
2086 if (bp->b_vflags & BV_BKGRDINPROG) {
2087 if (bp->b_flags & B_ASYNC) {
2088 BO_UNLOCK(bp->b_bufobj);
2093 bp->b_vflags |= BV_BKGRDWAIT;
2094 msleep(&bp->b_xflags, BO_MTX(bp->b_bufobj), PRIBIO, "bwrbg", 0);
2095 if (bp->b_vflags & BV_BKGRDINPROG)
2096 panic("bufwrite: still writing");
2098 BO_UNLOCK(bp->b_bufobj);
2101 * If this buffer is marked for background writing and we
2102 * do not have to wait for it, make a copy and write the
2103 * copy so as to leave this buffer ready for further use.
2105 * This optimization eats a lot of memory. If we have a page
2106 * or buffer shortfall we can't do it.
2108 if (dobkgrdwrite && (bp->b_xflags & BX_BKGRDWRITE) &&
2109 (bp->b_flags & B_ASYNC) &&
2110 !vm_page_count_severe() &&
2111 !buf_dirty_count_severe()) {
2112 KASSERT(bp->b_iodone == NULL,
2113 ("bufwrite: needs chained iodone (%p)", bp->b_iodone));
2115 /* get a new block */
2116 newbp = geteblk(bp->b_bufsize, GB_NOWAIT_BD);
2121 * set it to be identical to the old block. We have to
2122 * set b_lblkno and BKGRDMARKER before calling bgetvp()
2123 * to avoid confusing the splay tree and gbincore().
2125 KASSERT((bp->b_flags & B_UNMAPPED) == 0, ("Unmapped cg"));
2126 memcpy(newbp->b_data, bp->b_data, bp->b_bufsize);
2127 newbp->b_lblkno = bp->b_lblkno;
2128 newbp->b_xflags |= BX_BKGRDMARKER;
2129 BO_LOCK(bp->b_bufobj);
2130 bp->b_vflags |= BV_BKGRDINPROG;
2131 bgetvp(bp->b_vp, newbp);
2132 BO_UNLOCK(bp->b_bufobj);
2133 newbp->b_bufobj = &bp->b_vp->v_bufobj;
2134 newbp->b_blkno = bp->b_blkno;
2135 newbp->b_offset = bp->b_offset;
2136 newbp->b_iodone = ffs_backgroundwritedone;
2137 newbp->b_flags |= B_ASYNC;
2138 newbp->b_flags &= ~B_INVAL;
2142 * Move over the dependencies. If there are rollbacks,
2143 * leave the parent buffer dirtied as it will need to
2146 if (LIST_EMPTY(&bp->b_dep) ||
2147 softdep_move_dependencies(bp, newbp) == 0)
2154 * Initiate write on the copy, release the original to
2155 * the B_LOCKED queue so that it cannot go away until
2156 * the background write completes. If not locked it could go
2157 * away and then be reconstituted while it was being written.
2158 * If the reconstituted buffer were written, we could end up
2159 * with two background copies being written at the same time.
2164 /* Mark the buffer clean */
2168 /* Let the normal bufwrite do the rest for us */
2170 return (bufwrite(bp));
2175 ffs_geom_strategy(struct bufobj *bo, struct buf *bp)
2182 vp = bo->__bo_vnode;
2183 if (bp->b_iocmd == BIO_WRITE) {
2184 if ((bp->b_flags & B_VALIDSUSPWRT) == 0 &&
2185 bp->b_vp != NULL && bp->b_vp->v_mount != NULL &&
2186 (bp->b_vp->v_mount->mnt_kern_flag & MNTK_SUSPENDED) != 0)
2187 panic("ffs_geom_strategy: bad I/O");
2188 nocopy = bp->b_flags & B_NOCOPY;
2189 bp->b_flags &= ~(B_VALIDSUSPWRT | B_NOCOPY);
2190 if ((vp->v_vflag & VV_COPYONWRITE) && nocopy == 0 &&
2191 vp->v_rdev->si_snapdata != NULL) {
2192 if ((bp->b_flags & B_CLUSTER) != 0) {
2193 runningbufwakeup(bp);
2194 TAILQ_FOREACH(tbp, &bp->b_cluster.cluster_head,
2195 b_cluster.cluster_entry) {
2196 error = ffs_copyonwrite(vp, tbp);
2198 error != EOPNOTSUPP) {
2199 bp->b_error = error;
2200 bp->b_ioflags |= BIO_ERROR;
2205 bp->b_runningbufspace = bp->b_bufsize;
2206 atomic_add_long(&runningbufspace,
2207 bp->b_runningbufspace);
2209 error = ffs_copyonwrite(vp, bp);
2210 if (error != 0 && error != EOPNOTSUPP) {
2211 bp->b_error = error;
2212 bp->b_ioflags |= BIO_ERROR;
2219 if ((bp->b_flags & B_CLUSTER) != 0) {
2220 TAILQ_FOREACH(tbp, &bp->b_cluster.cluster_head,
2221 b_cluster.cluster_entry) {
2222 if (!LIST_EMPTY(&tbp->b_dep))
2226 if (!LIST_EMPTY(&bp->b_dep))
2232 g_vfs_strategy(bo, bp);
2236 ffs_own_mount(const struct mount *mp)
2239 if (mp->mnt_op == &ufs_vfsops)
2247 db_print_ffs(struct ufsmount *ump)
2249 db_printf("mp %p %s devvp %p fs %p su_wl %d su_deps %d su_req %d\n",
2250 ump->um_mountp, ump->um_mountp->mnt_stat.f_mntonname,
2251 ump->um_devvp, ump->um_fs, ump->softdep_on_worklist,
2252 ump->softdep_deps, ump->softdep_req);
2255 DB_SHOW_COMMAND(ffs, db_show_ffs)
2258 struct ufsmount *ump;
2261 ump = VFSTOUFS((struct mount *)addr);
2266 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2267 if (!strcmp(mp->mnt_stat.f_fstypename, ufs_vfsconf.vfc_name))
2268 db_print_ffs(VFSTOUFS(mp));