2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1989, 1991, 1993, 1994
5 * The Regents of the University of California. All rights reserved.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 * 3. Neither the name of the University nor the names of its contributors
16 * may be used to endorse or promote products derived from this software
17 * without specific prior written permission.
19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 * @(#)ffs_vfsops.c 8.31 (Berkeley) 5/20/95
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
37 #include "opt_quota.h"
42 #include <sys/param.h>
43 #include <sys/gsb_crc32.h>
44 #include <sys/systm.h>
45 #include <sys/namei.h>
48 #include <sys/taskqueue.h>
49 #include <sys/kernel.h>
51 #include <sys/vnode.h>
52 #include <sys/mount.h>
56 #include <sys/fcntl.h>
57 #include <sys/ioccom.h>
58 #include <sys/malloc.h>
59 #include <sys/mutex.h>
60 #include <sys/rwlock.h>
61 #include <sys/sysctl.h>
62 #include <sys/vmmeter.h>
64 #include <security/mac/mac_framework.h>
66 #include <ufs/ufs/dir.h>
67 #include <ufs/ufs/extattr.h>
68 #include <ufs/ufs/gjournal.h>
69 #include <ufs/ufs/quota.h>
70 #include <ufs/ufs/ufsmount.h>
71 #include <ufs/ufs/inode.h>
72 #include <ufs/ufs/ufs_extern.h>
74 #include <ufs/ffs/fs.h>
75 #include <ufs/ffs/ffs_extern.h>
79 #include <vm/vm_page.h>
81 #include <geom/geom.h>
82 #include <geom/geom_vfs.h>
86 static uma_zone_t uma_inode, uma_ufs1, uma_ufs2;
89 static int ffs_mountfs(struct vnode *, struct mount *, struct thread *);
90 static void ffs_oldfscompat_read(struct fs *, struct ufsmount *,
92 static void ffs_ifree(struct ufsmount *ump, struct inode *ip);
93 static int ffs_sync_lazy(struct mount *mp);
94 static int ffs_use_bread(void *devfd, off_t loc, void **bufp, int size);
95 static int ffs_use_bwrite(void *devfd, off_t loc, void *buf, int size);
97 static vfs_init_t ffs_init;
98 static vfs_uninit_t ffs_uninit;
99 static vfs_extattrctl_t ffs_extattrctl;
100 static vfs_cmount_t ffs_cmount;
101 static vfs_unmount_t ffs_unmount;
102 static vfs_mount_t ffs_mount;
103 static vfs_statfs_t ffs_statfs;
104 static vfs_fhtovp_t ffs_fhtovp;
105 static vfs_sync_t ffs_sync;
107 static struct vfsops ufs_vfsops = {
108 .vfs_extattrctl = ffs_extattrctl,
109 .vfs_fhtovp = ffs_fhtovp,
110 .vfs_init = ffs_init,
111 .vfs_mount = ffs_mount,
112 .vfs_cmount = ffs_cmount,
113 .vfs_quotactl = ufs_quotactl,
114 .vfs_root = vfs_cache_root,
115 .vfs_cachedroot = ufs_root,
116 .vfs_statfs = ffs_statfs,
117 .vfs_sync = ffs_sync,
118 .vfs_uninit = ffs_uninit,
119 .vfs_unmount = ffs_unmount,
120 .vfs_vget = ffs_vget,
121 .vfs_susp_clean = process_deferred_inactive,
124 VFS_SET(ufs_vfsops, ufs, 0);
125 MODULE_VERSION(ufs, 1);
127 static b_strategy_t ffs_geom_strategy;
128 static b_write_t ffs_bufwrite;
130 static struct buf_ops ffs_ops = {
132 .bop_write = ffs_bufwrite,
133 .bop_strategy = ffs_geom_strategy,
135 #ifdef NO_FFS_SNAPSHOT
136 .bop_bdflush = bufbdflush,
138 .bop_bdflush = ffs_bdflush,
143 * Note that userquota and groupquota options are not currently used
144 * by UFS/FFS code and generally mount(8) does not pass those options
145 * from userland, but they can be passed by loader(8) via
146 * vfs.root.mountfrom.options.
148 static const char *ffs_opts[] = { "acls", "async", "noatime", "noclusterr",
149 "noclusterw", "noexec", "export", "force", "from", "groupquota",
150 "multilabel", "nfsv4acls", "fsckpid", "snapshot", "nosuid", "suiddir",
151 "nosymfollow", "sync", "union", "userquota", "untrusted", NULL };
153 static int ffs_enxio_enable = 1;
154 SYSCTL_DECL(_vfs_ffs);
155 SYSCTL_INT(_vfs_ffs, OID_AUTO, enxio_enable, CTLFLAG_RWTUN,
156 &ffs_enxio_enable, 0,
157 "enable mapping of other disk I/O errors to ENXIO");
160 * Return buffer with the contents of block "offset" from the beginning of
161 * directory "ip". If "res" is non-zero, fill it in with a pointer to the
162 * remaining space in the directory.
165 ffs_blkatoff(struct vnode *vp, off_t offset, char **res, struct buf **bpp)
175 lbn = lblkno(fs, offset);
176 bsize = blksize(fs, ip, lbn);
179 error = bread(vp, lbn, bsize, NOCRED, &bp);
184 *res = (char *)bp->b_data + blkoff(fs, offset);
190 * Load up the contents of an inode and copy the appropriate pieces
191 * to the incore copy.
194 ffs_load_inode(struct buf *bp, struct inode *ip, struct fs *fs, ino_t ino)
196 struct ufs1_dinode *dip1;
197 struct ufs2_dinode *dip2;
203 *((struct ufs1_dinode *)bp->b_data + ino_to_fsbo(fs, ino));
204 ip->i_mode = dip1->di_mode;
205 ip->i_nlink = dip1->di_nlink;
206 ip->i_effnlink = dip1->di_nlink;
207 ip->i_size = dip1->di_size;
208 ip->i_flags = dip1->di_flags;
209 ip->i_gen = dip1->di_gen;
210 ip->i_uid = dip1->di_uid;
211 ip->i_gid = dip1->di_gid;
214 dip2 = ((struct ufs2_dinode *)bp->b_data + ino_to_fsbo(fs, ino));
215 if ((error = ffs_verify_dinode_ckhash(fs, dip2)) != 0 &&
216 !ffs_fsfail_cleanup(ITOUMP(ip), error)) {
217 printf("%s: inode %jd: check-hash failed\n", fs->fs_fsmnt,
223 ip->i_mode = dip2->di_mode;
224 ip->i_nlink = dip2->di_nlink;
225 ip->i_effnlink = dip2->di_nlink;
226 ip->i_size = dip2->di_size;
227 ip->i_flags = dip2->di_flags;
228 ip->i_gen = dip2->di_gen;
229 ip->i_uid = dip2->di_uid;
230 ip->i_gid = dip2->di_gid;
235 * Verify that a filesystem block number is a valid data block.
236 * This routine is only called on untrusted filesystems.
239 ffs_check_blkno(struct mount *mp, ino_t inum, ufs2_daddr_t daddr, int blksize)
242 struct ufsmount *ump;
243 ufs2_daddr_t end_daddr;
246 KASSERT((mp->mnt_flag & MNT_UNTRUSTED) != 0,
247 ("ffs_check_blkno called on a trusted file system"));
250 cg = dtog(fs, daddr);
251 end_daddr = daddr + numfrags(fs, blksize);
253 * Verify that the block number is a valid data block. Also check
254 * that it does not point to an inode block or a superblock. Accept
255 * blocks that are unalloacted (0) or part of snapshot metadata
256 * (BLK_NOCOPY or BLK_SNAP).
258 * Thus, the block must be in a valid range for the filesystem and
259 * either in the space before a backup superblock (except the first
260 * cylinder group where that space is used by the bootstrap code) or
261 * after the inode blocks and before the end of the cylinder group.
263 if ((uint64_t)daddr <= BLK_SNAP ||
264 ((uint64_t)end_daddr <= fs->fs_size &&
265 ((cg > 0 && end_daddr <= cgsblock(fs, cg)) ||
266 (daddr >= cgdmin(fs, cg) &&
267 end_daddr <= cgbase(fs, cg) + fs->fs_fpg))))
269 if ((havemtx = mtx_owned(UFS_MTX(ump))) == 0)
271 if (ppsratecheck(&ump->um_last_integritymsg,
272 &ump->um_secs_integritymsg, 1)) {
274 uprintf("\n%s: inode %jd, out-of-range indirect block "
275 "number %jd\n", mp->mnt_stat.f_mntonname, inum, daddr);
284 * On first ENXIO error, initiate an asynchronous forcible unmount.
285 * Used to unmount filesystems whose underlying media has gone away.
287 * Return true if a cleanup is in progress.
290 ffs_fsfail_cleanup(struct ufsmount *ump, int error)
295 retval = ffs_fsfail_cleanup_locked(ump, error);
301 ffs_fsfail_cleanup_locked(struct ufsmount *ump, int error)
303 mtx_assert(UFS_MTX(ump), MA_OWNED);
304 if (error == ENXIO && (ump->um_flags & UM_FSFAIL_CLEANUP) == 0) {
305 ump->um_flags |= UM_FSFAIL_CLEANUP;
307 * Queue an async forced unmount.
309 vfs_ref(ump->um_mountp);
310 dounmount(ump->um_mountp,
311 MNT_FORCE | MNT_RECURSE | MNT_DEFERRED, curthread);
312 printf("UFS: forcibly unmounting %s from %s\n",
313 ump->um_mountp->mnt_stat.f_mntfromname,
314 ump->um_mountp->mnt_stat.f_mntonname);
316 return ((ump->um_flags & UM_FSFAIL_CLEANUP) != 0);
320 * Wrapper used during ENXIO cleanup to allocate empty buffers when
321 * the kernel is unable to read the real one. They are needed so that
322 * the soft updates code can use them to unwind its dependencies.
325 ffs_breadz(struct ufsmount *ump, struct vnode *vp, daddr_t lblkno,
326 daddr_t dblkno, int size, daddr_t *rablkno, int *rabsize, int cnt,
327 struct ucred *cred, int flags, void (*ckhashfunc)(struct buf *),
332 flags |= GB_CVTENXIO;
333 error = breadn_flags(vp, lblkno, dblkno, size, rablkno, rabsize, cnt,
334 cred, flags, ckhashfunc, bpp);
335 if (error != 0 && ffs_fsfail_cleanup(ump, error)) {
336 error = getblkx(vp, lblkno, dblkno, size, 0, 0, flags, bpp);
337 KASSERT(error == 0, ("getblkx failed"));
338 vfs_bio_bzero_buf(*bpp, 0, size);
344 ffs_mount(struct mount *mp)
346 struct vnode *devvp, *odevvp;
348 struct ufsmount *ump = NULL;
351 int error, error1, flags;
352 uint64_t mntorflags, saved_mnt_flag;
354 struct nameidata ndp;
356 bool mounted_softdep;
359 if (vfs_filteropt(mp->mnt_optnew, ffs_opts))
361 if (uma_inode == NULL) {
362 uma_inode = uma_zcreate("FFS inode",
363 sizeof(struct inode), NULL, NULL, NULL, NULL,
365 uma_ufs1 = uma_zcreate("FFS1 dinode",
366 sizeof(struct ufs1_dinode), NULL, NULL, NULL, NULL,
368 uma_ufs2 = uma_zcreate("FFS2 dinode",
369 sizeof(struct ufs2_dinode), NULL, NULL, NULL, NULL,
371 VFS_SMR_ZONE_SET(uma_inode);
374 vfs_deleteopt(mp->mnt_optnew, "groupquota");
375 vfs_deleteopt(mp->mnt_optnew, "userquota");
377 fspec = vfs_getopts(mp->mnt_optnew, "from", &error);
382 if (vfs_getopt(mp->mnt_optnew, "untrusted", NULL, NULL) == 0)
383 mntorflags |= MNT_UNTRUSTED;
385 if (vfs_getopt(mp->mnt_optnew, "acls", NULL, NULL) == 0)
386 mntorflags |= MNT_ACLS;
388 if (vfs_getopt(mp->mnt_optnew, "snapshot", NULL, NULL) == 0) {
389 mntorflags |= MNT_SNAPSHOT;
391 * Once we have set the MNT_SNAPSHOT flag, do not
392 * persist "snapshot" in the options list.
394 vfs_deleteopt(mp->mnt_optnew, "snapshot");
395 vfs_deleteopt(mp->mnt_opt, "snapshot");
398 if (vfs_getopt(mp->mnt_optnew, "fsckpid", NULL, NULL) == 0 &&
399 vfs_scanopt(mp->mnt_optnew, "fsckpid", "%d", &fsckpid) == 1) {
401 * Once we have set the restricted PID, do not
402 * persist "fsckpid" in the options list.
404 vfs_deleteopt(mp->mnt_optnew, "fsckpid");
405 vfs_deleteopt(mp->mnt_opt, "fsckpid");
406 if (mp->mnt_flag & MNT_UPDATE) {
407 if (VFSTOUFS(mp)->um_fs->fs_ronly == 0 &&
408 vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0) == 0) {
410 "Checker enable: Must be read-only");
413 } else if (vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0) == 0) {
415 "Checker enable: Must be read-only");
418 /* Set to -1 if we are done */
423 if (vfs_getopt(mp->mnt_optnew, "nfsv4acls", NULL, NULL) == 0) {
424 if (mntorflags & MNT_ACLS) {
426 "\"acls\" and \"nfsv4acls\" options "
427 "are mutually exclusive");
430 mntorflags |= MNT_NFS4ACLS;
434 mp->mnt_kern_flag &= ~MNTK_FPLOOKUP;
435 mp->mnt_flag |= mntorflags;
438 * If updating, check whether changing from read-only to
439 * read/write; if there is no device name, that's all we do.
441 if (mp->mnt_flag & MNT_UPDATE) {
444 odevvp = ump->um_odevvp;
445 devvp = ump->um_devvp;
446 if (fsckpid == -1 && ump->um_fsckpid > 0) {
447 if ((error = ffs_flushfiles(mp, WRITECLOSE, td)) != 0 ||
448 (error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0)
452 * Return to normal read-only mode.
454 error = g_access(ump->um_cp, 0, -1, 0);
458 if (fs->fs_ronly == 0 &&
459 vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0)) {
461 * Flush any dirty data and suspend filesystem.
463 if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0)
465 error = vfs_write_suspend_umnt(mp);
470 if (MOUNTEDSOFTDEP(mp)) {
472 mp->mnt_flag &= ~MNT_SOFTDEP;
474 mounted_softdep = true;
476 mounted_softdep = false;
479 * Check for and optionally get rid of files open
483 if (mp->mnt_flag & MNT_FORCE)
485 if (mounted_softdep) {
486 error = softdep_flushfiles(mp, flags, td);
488 error = ffs_flushfiles(mp, flags, td);
492 if (mounted_softdep) {
494 mp->mnt_flag |= MNT_SOFTDEP;
497 vfs_write_resume(mp, 0);
501 if (fs->fs_pendingblocks != 0 ||
502 fs->fs_pendinginodes != 0) {
503 printf("WARNING: %s Update error: blocks %jd "
504 "files %d\n", fs->fs_fsmnt,
505 (intmax_t)fs->fs_pendingblocks,
506 fs->fs_pendinginodes);
507 fs->fs_pendingblocks = 0;
508 fs->fs_pendinginodes = 0;
510 if ((fs->fs_flags & (FS_UNCLEAN | FS_NEEDSFSCK)) == 0)
512 if ((error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0) {
515 if (mounted_softdep) {
517 mp->mnt_flag |= MNT_SOFTDEP;
520 vfs_write_resume(mp, 0);
527 * Drop our write and exclusive access.
529 g_access(ump->um_cp, 0, -1, -1);
532 mp->mnt_flag |= MNT_RDONLY;
535 * Allow the writers to note that filesystem
538 vfs_write_resume(mp, 0);
540 if ((mp->mnt_flag & MNT_RELOAD) &&
541 (error = ffs_reload(mp, 0)) != 0)
544 !vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0)) {
546 * If we are running a checker, do not allow upgrade.
548 if (ump->um_fsckpid > 0) {
550 "Active checker, cannot upgrade to write");
554 * If upgrade to read-write by non-root, then verify
555 * that user has necessary permissions on the device.
557 vn_lock(odevvp, LK_EXCLUSIVE | LK_RETRY);
558 error = VOP_ACCESS(odevvp, VREAD | VWRITE,
561 error = priv_check(td, PRIV_VFS_MOUNT_PERM);
566 fs->fs_flags &= ~FS_UNCLEAN;
567 if (fs->fs_clean == 0) {
568 fs->fs_flags |= FS_UNCLEAN;
569 if ((mp->mnt_flag & MNT_FORCE) ||
571 (FS_SUJ | FS_NEEDSFSCK)) == 0 &&
572 (fs->fs_flags & FS_DOSOFTDEP))) {
573 printf("WARNING: %s was not properly "
574 "dismounted\n", fs->fs_fsmnt);
577 "R/W mount of %s denied. %s.%s",
579 "Filesystem is not clean - run fsck",
580 (fs->fs_flags & FS_SUJ) == 0 ? "" :
581 " Forced mount will invalidate"
582 " journal contents");
588 * Request exclusive write access.
590 error = g_access(ump->um_cp, 0, 1, 1);
594 if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0)
596 error = vfs_write_suspend_umnt(mp);
601 saved_mnt_flag = MNT_RDONLY;
602 if (MOUNTEDSOFTDEP(mp) && (mp->mnt_flag &
604 saved_mnt_flag |= MNT_ASYNC;
605 mp->mnt_flag &= ~saved_mnt_flag;
607 fs->fs_mtime = time_second;
608 /* check to see if we need to start softdep */
609 if ((fs->fs_flags & FS_DOSOFTDEP) &&
610 (error = softdep_mount(devvp, mp, fs, td->td_ucred))){
613 mp->mnt_flag |= saved_mnt_flag;
615 vfs_write_resume(mp, 0);
619 if ((error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0) {
621 if ((fs->fs_flags & FS_DOSOFTDEP) != 0)
624 mp->mnt_flag |= saved_mnt_flag;
626 vfs_write_resume(mp, 0);
629 if (fs->fs_snapinum[0] != 0)
630 ffs_snapshot_mount(mp);
631 vfs_write_resume(mp, 0);
634 * Soft updates is incompatible with "async",
635 * so if we are doing softupdates stop the user
636 * from setting the async flag in an update.
637 * Softdep_mount() clears it in an initial mount
640 if (MOUNTEDSOFTDEP(mp)) {
641 /* XXX: Reset too late ? */
643 mp->mnt_flag &= ~MNT_ASYNC;
647 * Keep MNT_ACLS flag if it is stored in superblock.
649 if ((fs->fs_flags & FS_ACLS) != 0) {
650 /* XXX: Set too late ? */
652 mp->mnt_flag |= MNT_ACLS;
656 if ((fs->fs_flags & FS_NFS4ACLS) != 0) {
657 /* XXX: Set too late ? */
659 mp->mnt_flag |= MNT_NFS4ACLS;
663 * If this is a request from fsck to clean up the filesystem,
664 * then allow the specified pid to proceed.
667 if (ump->um_fsckpid != 0) {
669 "Active checker already running on %s",
673 KASSERT(MOUNTEDSOFTDEP(mp) == 0,
674 ("soft updates enabled on read-only file system"));
677 * Request write access.
679 error = g_access(ump->um_cp, 0, 1, 0);
683 "Checker activation failed on %s",
687 ump->um_fsckpid = fsckpid;
688 if (fs->fs_snapinum[0] != 0)
689 ffs_snapshot_mount(mp);
690 fs->fs_mtime = time_second;
693 (void) ffs_sbupdate(ump, MNT_WAIT, 0);
697 * If this is a snapshot request, take the snapshot.
699 if (mp->mnt_flag & MNT_SNAPSHOT)
700 return (ffs_snapshot(mp, fspec));
703 * Must not call namei() while owning busy ref.
709 * Not an update, or updating the name: look up the name
710 * and verify that it refers to a sensible disk device.
712 NDINIT(&ndp, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, fspec, td);
714 if ((mp->mnt_flag & MNT_UPDATE) != 0) {
716 * Unmount does not start if MNT_UPDATE is set. Mount
717 * update busies mp before setting MNT_UPDATE. We
718 * must be able to retain our busy ref succesfully,
721 error1 = vfs_busy(mp, MBF_NOWAIT);
726 NDFREE(&ndp, NDF_ONLY_PNBUF);
728 if (!vn_isdisk_error(devvp, &error)) {
734 * If mount by non-root, then verify that user has necessary
735 * permissions on the device.
738 if ((mp->mnt_flag & MNT_RDONLY) == 0)
740 error = VOP_ACCESS(devvp, accmode, td->td_ucred, td);
742 error = priv_check(td, PRIV_VFS_MOUNT_PERM);
748 if (mp->mnt_flag & MNT_UPDATE) {
752 * If it's not the same vnode, or at least the same device
753 * then it's not correct.
756 if (devvp->v_rdev != ump->um_devvp->v_rdev)
757 error = EINVAL; /* needs translation */
765 * We need the name for the mount point (also used for
766 * "last mounted on") copied in. If an error occurs,
767 * the mount point is discarded by the upper level code.
768 * Note that vfs_mount_alloc() populates f_mntonname for us.
770 if ((error = ffs_mountfs(devvp, mp, td)) != 0) {
775 KASSERT(MOUNTEDSOFTDEP(mp) == 0,
776 ("soft updates enabled on read-only file system"));
781 * Request write access.
783 error = g_access(ump->um_cp, 0, 1, 0);
786 printf("WARNING: %s: Checker activation "
787 "failed\n", fs->fs_fsmnt);
789 ump->um_fsckpid = fsckpid;
790 if (fs->fs_snapinum[0] != 0)
791 ffs_snapshot_mount(mp);
792 fs->fs_mtime = time_second;
794 (void) ffs_sbupdate(ump, MNT_WAIT, 0);
801 * This is racy versus lookup, see ufs_fplookup_vexec for details.
803 if ((mp->mnt_kern_flag & MNTK_FPLOOKUP) != 0)
804 panic("MNTK_FPLOOKUP set on mount %p when it should not be", mp);
805 if ((mp->mnt_flag & (MNT_ACLS | MNT_NFS4ACLS | MNT_UNION)) == 0)
806 mp->mnt_kern_flag |= MNTK_FPLOOKUP;
809 vfs_mountedfrom(mp, fspec);
814 * Compatibility with old mount system call.
818 ffs_cmount(struct mntarg *ma, void *data, uint64_t flags)
820 struct ufs_args args;
825 error = copyin(data, &args, sizeof args);
829 ma = mount_argsu(ma, "from", args.fspec, MAXPATHLEN);
830 ma = mount_arg(ma, "export", &args.export, sizeof(args.export));
831 error = kernel_mount(ma, flags);
837 * Reload all incore data for a filesystem (used after running fsck on
838 * the root filesystem and finding things to fix). If the 'force' flag
839 * is 0, the filesystem must be mounted read-only.
841 * Things to do to update the mount:
842 * 1) invalidate all cached meta-data.
843 * 2) re-read superblock from disk.
844 * 3) re-read summary information from disk.
845 * 4) invalidate all inactive vnodes.
846 * 5) clear MNTK_SUSPEND2 and MNTK_SUSPENDED flags, allowing secondary
847 * writers, if requested.
848 * 6) invalidate all cached file data.
849 * 7) re-read inode data for all active vnodes.
852 ffs_reload(struct mount *mp, int flags)
854 struct vnode *vp, *mvp, *devvp;
858 struct fs *fs, *newfs;
859 struct ufsmount *ump;
860 ufs2_daddr_t sblockloc;
868 if ((mp->mnt_flag & MNT_RDONLY) == 0 && (flags & FFSR_FORCE) == 0) {
875 * Step 1: invalidate all cached meta-data.
877 devvp = VFSTOUFS(mp)->um_devvp;
878 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
879 if (vinvalbuf(devvp, 0, 0, 0) != 0)
880 panic("ffs_reload: dirty1");
884 * Step 2: re-read superblock from disk.
886 fs = VFSTOUFS(mp)->um_fs;
887 if ((error = bread(devvp, btodb(fs->fs_sblockloc), fs->fs_sbsize,
890 newfs = (struct fs *)bp->b_data;
891 if ((newfs->fs_magic != FS_UFS1_MAGIC &&
892 newfs->fs_magic != FS_UFS2_MAGIC) ||
893 newfs->fs_bsize > MAXBSIZE ||
894 newfs->fs_bsize < sizeof(struct fs)) {
896 return (EIO); /* XXX needs translation */
899 * Preserve the summary information, read-only status, and
900 * superblock location by copying these fields into our new
901 * superblock before using it to update the existing superblock.
903 newfs->fs_si = fs->fs_si;
904 newfs->fs_ronly = fs->fs_ronly;
905 sblockloc = fs->fs_sblockloc;
906 bcopy(newfs, fs, (u_int)fs->fs_sbsize);
908 ump->um_maxsymlinklen = fs->fs_maxsymlinklen;
909 ffs_oldfscompat_read(fs, VFSTOUFS(mp), sblockloc);
911 if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
912 printf("WARNING: %s: reload pending error: blocks %jd "
913 "files %d\n", fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
914 fs->fs_pendinginodes);
915 fs->fs_pendingblocks = 0;
916 fs->fs_pendinginodes = 0;
921 * Step 3: re-read summary information from disk.
923 size = fs->fs_cssize;
924 blks = howmany(size, fs->fs_fsize);
925 if (fs->fs_contigsumsize > 0)
926 size += fs->fs_ncg * sizeof(int32_t);
927 size += fs->fs_ncg * sizeof(u_int8_t);
928 free(fs->fs_csp, M_UFSMNT);
929 space = malloc(size, M_UFSMNT, M_WAITOK);
931 for (i = 0; i < blks; i += fs->fs_frag) {
933 if (i + fs->fs_frag > blks)
934 size = (blks - i) * fs->fs_fsize;
935 error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), size,
939 bcopy(bp->b_data, space, (u_int)size);
940 space = (char *)space + size;
944 * We no longer know anything about clusters per cylinder group.
946 if (fs->fs_contigsumsize > 0) {
947 fs->fs_maxcluster = lp = space;
948 for (i = 0; i < fs->fs_ncg; i++)
949 *lp++ = fs->fs_contigsumsize;
952 size = fs->fs_ncg * sizeof(u_int8_t);
953 fs->fs_contigdirs = (u_int8_t *)space;
954 bzero(fs->fs_contigdirs, size);
955 if ((flags & FFSR_UNSUSPEND) != 0) {
957 mp->mnt_kern_flag &= ~(MNTK_SUSPENDED | MNTK_SUSPEND2);
958 wakeup(&mp->mnt_flag);
963 MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
967 if (vp->v_type == VNON) {
972 * Step 4: invalidate all cached file data.
974 if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK)) {
975 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
978 if (vinvalbuf(vp, 0, 0, 0))
979 panic("ffs_reload: dirty2");
981 * Step 5: re-read inode data for all active vnodes.
985 bread(devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
986 (int)fs->fs_bsize, NOCRED, &bp);
989 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
992 if ((error = ffs_load_inode(bp, ip, fs, ip->i_number)) != 0) {
995 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
998 ip->i_effnlink = ip->i_nlink;
1006 * Common code for mount and mountroot
1009 ffs_mountfs(odevvp, mp, td)
1010 struct vnode *odevvp;
1014 struct ufsmount *ump;
1017 int error, i, len, ronly;
1019 struct g_consumer *cp;
1021 struct vnode *devvp;
1022 int candelete, canspeedup;
1027 cred = td ? td->td_ucred : NOCRED;
1028 ronly = (mp->mnt_flag & MNT_RDONLY) != 0;
1030 devvp = mntfs_allocvp(mp, odevvp);
1032 KASSERT(devvp->v_type == VCHR, ("reclaimed devvp"));
1033 dev = devvp->v_rdev;
1034 KASSERT(dev->si_snapdata == NULL, ("non-NULL snapshot data"));
1035 if (atomic_cmpset_acq_ptr((uintptr_t *)&dev->si_mountpt, 0,
1036 (uintptr_t)mp) == 0) {
1037 mntfs_freevp(devvp);
1041 error = g_vfs_open(devvp, &cp, "ffs", ronly ? 0 : 1);
1042 g_topology_unlock();
1044 atomic_store_rel_ptr((uintptr_t *)&dev->si_mountpt, 0);
1045 mntfs_freevp(devvp);
1049 devvp->v_bufobj.bo_ops = &ffs_ops;
1050 BO_LOCK(&odevvp->v_bufobj);
1051 odevvp->v_bufobj.bo_flag |= BO_NOBUFS;
1052 BO_UNLOCK(&odevvp->v_bufobj);
1053 if (dev->si_iosize_max != 0)
1054 mp->mnt_iosize_max = dev->si_iosize_max;
1055 if (mp->mnt_iosize_max > maxphys)
1056 mp->mnt_iosize_max = maxphys;
1057 if ((SBLOCKSIZE % cp->provider->sectorsize) != 0) {
1060 "Invalid sectorsize %d for superblock size %d",
1061 cp->provider->sectorsize, SBLOCKSIZE);
1064 /* fetch the superblock and summary information */
1066 if ((mp->mnt_flag & MNT_ROOTFS) != 0)
1067 loc = STDSB_NOHASHFAIL;
1068 if ((error = ffs_sbget(devvp, &fs, loc, M_UFSMNT, ffs_use_bread)) != 0)
1070 fs->fs_flags &= ~FS_UNCLEAN;
1071 if (fs->fs_clean == 0) {
1072 fs->fs_flags |= FS_UNCLEAN;
1073 if (ronly || (mp->mnt_flag & MNT_FORCE) ||
1074 ((fs->fs_flags & (FS_SUJ | FS_NEEDSFSCK)) == 0 &&
1075 (fs->fs_flags & FS_DOSOFTDEP))) {
1076 printf("WARNING: %s was not properly dismounted\n",
1079 vfs_mount_error(mp, "R/W mount of %s denied. %s%s",
1080 fs->fs_fsmnt, "Filesystem is not clean - run fsck.",
1081 (fs->fs_flags & FS_SUJ) == 0 ? "" :
1082 " Forced mount will invalidate journal contents");
1086 if ((fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) &&
1087 (mp->mnt_flag & MNT_FORCE)) {
1088 printf("WARNING: %s: lost blocks %jd files %d\n",
1089 fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
1090 fs->fs_pendinginodes);
1091 fs->fs_pendingblocks = 0;
1092 fs->fs_pendinginodes = 0;
1095 if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
1096 printf("WARNING: %s: mount pending error: blocks %jd "
1097 "files %d\n", fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
1098 fs->fs_pendinginodes);
1099 fs->fs_pendingblocks = 0;
1100 fs->fs_pendinginodes = 0;
1102 if ((fs->fs_flags & FS_GJOURNAL) != 0) {
1105 * Get journal provider name.
1108 mp->mnt_gjprovider = malloc((u_long)len, M_UFSMNT, M_WAITOK);
1109 if (g_io_getattr("GJOURNAL::provider", cp, &len,
1110 mp->mnt_gjprovider) == 0) {
1111 mp->mnt_gjprovider = realloc(mp->mnt_gjprovider, len,
1112 M_UFSMNT, M_WAITOK);
1114 mp->mnt_flag |= MNT_GJOURNAL;
1117 printf("WARNING: %s: GJOURNAL flag on fs "
1118 "but no gjournal provider below\n",
1119 mp->mnt_stat.f_mntonname);
1120 free(mp->mnt_gjprovider, M_UFSMNT);
1121 mp->mnt_gjprovider = NULL;
1124 printf("WARNING: %s: GJOURNAL flag on fs but no "
1125 "UFS_GJOURNAL support\n", mp->mnt_stat.f_mntonname);
1128 mp->mnt_gjprovider = NULL;
1130 ump = malloc(sizeof *ump, M_UFSMNT, M_WAITOK | M_ZERO);
1132 ump->um_bo = &devvp->v_bufobj;
1134 if (fs->fs_magic == FS_UFS1_MAGIC) {
1135 ump->um_fstype = UFS1;
1136 ump->um_balloc = ffs_balloc_ufs1;
1138 ump->um_fstype = UFS2;
1139 ump->um_balloc = ffs_balloc_ufs2;
1141 ump->um_blkatoff = ffs_blkatoff;
1142 ump->um_truncate = ffs_truncate;
1143 ump->um_update = ffs_update;
1144 ump->um_valloc = ffs_valloc;
1145 ump->um_vfree = ffs_vfree;
1146 ump->um_ifree = ffs_ifree;
1147 ump->um_rdonly = ffs_rdonly;
1148 ump->um_snapgone = ffs_snapgone;
1149 if ((mp->mnt_flag & MNT_UNTRUSTED) != 0)
1150 ump->um_check_blkno = ffs_check_blkno;
1152 ump->um_check_blkno = NULL;
1153 mtx_init(UFS_MTX(ump), "FFS", "FFS Lock", MTX_DEF);
1154 sx_init(&ump->um_checkpath_lock, "uchpth");
1155 ffs_oldfscompat_read(fs, ump, fs->fs_sblockloc);
1156 fs->fs_ronly = ronly;
1157 fs->fs_active = NULL;
1159 mp->mnt_stat.f_fsid.val[0] = fs->fs_id[0];
1160 mp->mnt_stat.f_fsid.val[1] = fs->fs_id[1];
1162 if (fs->fs_id[0] == 0 || fs->fs_id[1] == 0 ||
1163 (nmp = vfs_getvfs(&mp->mnt_stat.f_fsid))) {
1168 ump->um_maxsymlinklen = fs->fs_maxsymlinklen;
1170 mp->mnt_flag |= MNT_LOCAL;
1172 if ((fs->fs_flags & FS_MULTILABEL) != 0) {
1175 mp->mnt_flag |= MNT_MULTILABEL;
1178 printf("WARNING: %s: multilabel flag on fs but "
1179 "no MAC support\n", mp->mnt_stat.f_mntonname);
1182 if ((fs->fs_flags & FS_ACLS) != 0) {
1186 if (mp->mnt_flag & MNT_NFS4ACLS)
1187 printf("WARNING: %s: ACLs flag on fs conflicts with "
1188 "\"nfsv4acls\" mount option; option ignored\n",
1189 mp->mnt_stat.f_mntonname);
1190 mp->mnt_flag &= ~MNT_NFS4ACLS;
1191 mp->mnt_flag |= MNT_ACLS;
1195 printf("WARNING: %s: ACLs flag on fs but no ACLs support\n",
1196 mp->mnt_stat.f_mntonname);
1199 if ((fs->fs_flags & FS_NFS4ACLS) != 0) {
1203 if (mp->mnt_flag & MNT_ACLS)
1204 printf("WARNING: %s: NFSv4 ACLs flag on fs conflicts "
1205 "with \"acls\" mount option; option ignored\n",
1206 mp->mnt_stat.f_mntonname);
1207 mp->mnt_flag &= ~MNT_ACLS;
1208 mp->mnt_flag |= MNT_NFS4ACLS;
1212 printf("WARNING: %s: NFSv4 ACLs flag on fs but no "
1213 "ACLs support\n", mp->mnt_stat.f_mntonname);
1216 if ((fs->fs_flags & FS_TRIM) != 0) {
1218 if (g_io_getattr("GEOM::candelete", cp, &len,
1221 ump->um_flags |= UM_CANDELETE;
1223 printf("WARNING: %s: TRIM flag on fs but disk "
1224 "does not support TRIM\n",
1225 mp->mnt_stat.f_mntonname);
1227 printf("WARNING: %s: TRIM flag on fs but disk does "
1228 "not confirm that it supports TRIM\n",
1229 mp->mnt_stat.f_mntonname);
1231 if (((ump->um_flags) & UM_CANDELETE) != 0) {
1232 ump->um_trim_tq = taskqueue_create("trim", M_WAITOK,
1233 taskqueue_thread_enqueue, &ump->um_trim_tq);
1234 taskqueue_start_threads(&ump->um_trim_tq, 1, PVFS,
1235 "%s trim", mp->mnt_stat.f_mntonname);
1236 ump->um_trimhash = hashinit(MAXTRIMIO, M_TRIM,
1237 &ump->um_trimlisthashsize);
1242 if (g_io_getattr("GEOM::canspeedup", cp, &len, &canspeedup) == 0) {
1244 ump->um_flags |= UM_CANSPEEDUP;
1247 ump->um_mountp = mp;
1249 ump->um_devvp = devvp;
1250 ump->um_odevvp = odevvp;
1251 ump->um_nindir = fs->fs_nindir;
1252 ump->um_bptrtodb = fs->fs_fsbtodb;
1253 ump->um_seqinc = fs->fs_frag;
1254 for (i = 0; i < MAXQUOTAS; i++)
1255 ump->um_quotas[i] = NULLVP;
1257 ufs_extattr_uepm_init(&ump->um_extattr);
1260 * Set FS local "last mounted on" information (NULL pad)
1262 bzero(fs->fs_fsmnt, MAXMNTLEN);
1263 strlcpy(fs->fs_fsmnt, mp->mnt_stat.f_mntonname, MAXMNTLEN);
1264 mp->mnt_stat.f_iosize = fs->fs_bsize;
1266 if (mp->mnt_flag & MNT_ROOTFS) {
1268 * Root mount; update timestamp in mount structure.
1269 * this will be used by the common root mount code
1270 * to update the system clock.
1272 mp->mnt_time = fs->fs_time;
1276 fs->fs_mtime = time_second;
1277 if ((fs->fs_flags & FS_DOSOFTDEP) &&
1278 (error = softdep_mount(devvp, mp, fs, cred)) != 0) {
1279 ffs_flushfiles(mp, FORCECLOSE, td);
1282 if (fs->fs_snapinum[0] != 0)
1283 ffs_snapshot_mount(mp);
1286 (void) ffs_sbupdate(ump, MNT_WAIT, 0);
1289 * Initialize filesystem state information in mount struct.
1292 mp->mnt_kern_flag |= MNTK_LOOKUP_SHARED | MNTK_EXTENDED_SHARED |
1293 MNTK_NO_IOPF | MNTK_UNMAPPED_BUFS | MNTK_USES_BCACHE;
1296 #ifdef UFS_EXTATTR_AUTOSTART
1299 * Auto-starting does the following:
1300 * - check for /.attribute in the fs, and extattr_start if so
1301 * - for each file in .attribute, enable that file with
1302 * an attribute of the same name.
1303 * Not clear how to report errors -- probably eat them.
1304 * This would all happen while the filesystem was busy/not
1305 * available, so would effectively be "atomic".
1307 (void) ufs_extattr_autostart(mp, td);
1308 #endif /* !UFS_EXTATTR_AUTOSTART */
1309 #endif /* !UFS_EXTATTR */
1313 free(fs->fs_csp, M_UFSMNT);
1314 free(fs->fs_si, M_UFSMNT);
1320 g_topology_unlock();
1323 mtx_destroy(UFS_MTX(ump));
1324 sx_destroy(&ump->um_checkpath_lock);
1325 if (mp->mnt_gjprovider != NULL) {
1326 free(mp->mnt_gjprovider, M_UFSMNT);
1327 mp->mnt_gjprovider = NULL;
1329 MPASS(ump->um_softdep == NULL);
1330 free(ump, M_UFSMNT);
1331 mp->mnt_data = NULL;
1333 BO_LOCK(&odevvp->v_bufobj);
1334 odevvp->v_bufobj.bo_flag &= ~BO_NOBUFS;
1335 BO_UNLOCK(&odevvp->v_bufobj);
1336 atomic_store_rel_ptr((uintptr_t *)&dev->si_mountpt, 0);
1337 mntfs_freevp(devvp);
1343 * A read function for use by filesystem-layer routines.
1346 ffs_use_bread(void *devfd, off_t loc, void **bufp, int size)
1351 KASSERT(*bufp == NULL, ("ffs_use_bread: non-NULL *bufp %p\n", *bufp));
1352 *bufp = malloc(size, M_UFSMNT, M_WAITOK);
1353 if ((error = bread((struct vnode *)devfd, btodb(loc), size, NOCRED,
1356 bcopy(bp->b_data, *bufp, size);
1357 bp->b_flags |= B_INVAL | B_NOCACHE;
1362 static int bigcgs = 0;
1363 SYSCTL_INT(_debug, OID_AUTO, bigcgs, CTLFLAG_RW, &bigcgs, 0, "");
1366 * Sanity checks for loading old filesystem superblocks.
1367 * See ffs_oldfscompat_write below for unwound actions.
1369 * XXX - Parts get retired eventually.
1370 * Unfortunately new bits get added.
1373 ffs_oldfscompat_read(fs, ump, sblockloc)
1375 struct ufsmount *ump;
1376 ufs2_daddr_t sblockloc;
1381 * If not yet done, update fs_flags location and value of fs_sblockloc.
1383 if ((fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) {
1384 fs->fs_flags = fs->fs_old_flags;
1385 fs->fs_old_flags |= FS_FLAGS_UPDATED;
1386 fs->fs_sblockloc = sblockloc;
1389 * If not yet done, update UFS1 superblock with new wider fields.
1391 if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_maxbsize != fs->fs_bsize) {
1392 fs->fs_maxbsize = fs->fs_bsize;
1393 fs->fs_time = fs->fs_old_time;
1394 fs->fs_size = fs->fs_old_size;
1395 fs->fs_dsize = fs->fs_old_dsize;
1396 fs->fs_csaddr = fs->fs_old_csaddr;
1397 fs->fs_cstotal.cs_ndir = fs->fs_old_cstotal.cs_ndir;
1398 fs->fs_cstotal.cs_nbfree = fs->fs_old_cstotal.cs_nbfree;
1399 fs->fs_cstotal.cs_nifree = fs->fs_old_cstotal.cs_nifree;
1400 fs->fs_cstotal.cs_nffree = fs->fs_old_cstotal.cs_nffree;
1402 if (fs->fs_magic == FS_UFS1_MAGIC &&
1403 fs->fs_old_inodefmt < FS_44INODEFMT) {
1404 fs->fs_maxfilesize = ((uint64_t)1 << 31) - 1;
1405 fs->fs_qbmask = ~fs->fs_bmask;
1406 fs->fs_qfmask = ~fs->fs_fmask;
1408 if (fs->fs_magic == FS_UFS1_MAGIC) {
1409 ump->um_savedmaxfilesize = fs->fs_maxfilesize;
1410 maxfilesize = (uint64_t)0x80000000 * fs->fs_bsize - 1;
1411 if (fs->fs_maxfilesize > maxfilesize)
1412 fs->fs_maxfilesize = maxfilesize;
1414 /* Compatibility for old filesystems */
1415 if (fs->fs_avgfilesize <= 0)
1416 fs->fs_avgfilesize = AVFILESIZ;
1417 if (fs->fs_avgfpdir <= 0)
1418 fs->fs_avgfpdir = AFPDIR;
1420 fs->fs_save_cgsize = fs->fs_cgsize;
1421 fs->fs_cgsize = fs->fs_bsize;
1426 * Unwinding superblock updates for old filesystems.
1427 * See ffs_oldfscompat_read above for details.
1429 * XXX - Parts get retired eventually.
1430 * Unfortunately new bits get added.
1433 ffs_oldfscompat_write(fs, ump)
1435 struct ufsmount *ump;
1439 * Copy back UFS2 updated fields that UFS1 inspects.
1441 if (fs->fs_magic == FS_UFS1_MAGIC) {
1442 fs->fs_old_time = fs->fs_time;
1443 fs->fs_old_cstotal.cs_ndir = fs->fs_cstotal.cs_ndir;
1444 fs->fs_old_cstotal.cs_nbfree = fs->fs_cstotal.cs_nbfree;
1445 fs->fs_old_cstotal.cs_nifree = fs->fs_cstotal.cs_nifree;
1446 fs->fs_old_cstotal.cs_nffree = fs->fs_cstotal.cs_nffree;
1447 fs->fs_maxfilesize = ump->um_savedmaxfilesize;
1450 fs->fs_cgsize = fs->fs_save_cgsize;
1451 fs->fs_save_cgsize = 0;
1456 * unmount system call
1459 ffs_unmount(mp, mntflags)
1464 struct ufsmount *ump = VFSTOUFS(mp);
1466 int error, flags, susp;
1474 if (mntflags & MNT_FORCE)
1475 flags |= FORCECLOSE;
1476 susp = fs->fs_ronly == 0;
1478 if ((error = ufs_extattr_stop(mp, td))) {
1479 if (error != EOPNOTSUPP)
1480 printf("WARNING: unmount %s: ufs_extattr_stop "
1481 "returned errno %d\n", mp->mnt_stat.f_mntonname,
1485 ufs_extattr_uepm_destroy(&ump->um_extattr);
1490 error = vfs_write_suspend_umnt(mp);
1494 if (MOUNTEDSOFTDEP(mp))
1495 error = softdep_flushfiles(mp, flags, td);
1497 error = ffs_flushfiles(mp, flags, td);
1498 if (error != 0 && !ffs_fsfail_cleanup(ump, error))
1502 if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
1503 printf("WARNING: unmount %s: pending error: blocks %jd "
1504 "files %d\n", fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
1505 fs->fs_pendinginodes);
1506 fs->fs_pendingblocks = 0;
1507 fs->fs_pendinginodes = 0;
1510 if (MOUNTEDSOFTDEP(mp))
1511 softdep_unmount(mp);
1512 MPASS(ump->um_softdep == NULL);
1513 if (fs->fs_ronly == 0 || ump->um_fsckpid > 0) {
1514 fs->fs_clean = fs->fs_flags & (FS_UNCLEAN|FS_NEEDSFSCK) ? 0 : 1;
1515 error = ffs_sbupdate(ump, MNT_WAIT, 0);
1516 if (ffs_fsfail_cleanup(ump, error))
1518 if (error != 0 && !ffs_fsfail_cleanup(ump, error)) {
1524 vfs_write_resume(mp, VR_START_WRITE);
1525 if (ump->um_trim_tq != NULL) {
1526 while (ump->um_trim_inflight != 0)
1527 pause("ufsutr", hz);
1528 taskqueue_drain_all(ump->um_trim_tq);
1529 taskqueue_free(ump->um_trim_tq);
1530 free (ump->um_trimhash, M_TRIM);
1533 if (ump->um_fsckpid > 0) {
1535 * Return to normal read-only mode.
1537 error = g_access(ump->um_cp, 0, -1, 0);
1538 ump->um_fsckpid = 0;
1540 g_vfs_close(ump->um_cp);
1541 g_topology_unlock();
1542 BO_LOCK(&ump->um_odevvp->v_bufobj);
1543 ump->um_odevvp->v_bufobj.bo_flag &= ~BO_NOBUFS;
1544 BO_UNLOCK(&ump->um_odevvp->v_bufobj);
1545 atomic_store_rel_ptr((uintptr_t *)&ump->um_dev->si_mountpt, 0);
1546 mntfs_freevp(ump->um_devvp);
1547 vrele(ump->um_odevvp);
1548 dev_rel(ump->um_dev);
1549 mtx_destroy(UFS_MTX(ump));
1550 sx_destroy(&ump->um_checkpath_lock);
1551 if (mp->mnt_gjprovider != NULL) {
1552 free(mp->mnt_gjprovider, M_UFSMNT);
1553 mp->mnt_gjprovider = NULL;
1555 free(fs->fs_csp, M_UFSMNT);
1556 free(fs->fs_si, M_UFSMNT);
1558 free(ump, M_UFSMNT);
1559 mp->mnt_data = NULL;
1561 mp->mnt_flag &= ~MNT_LOCAL;
1563 if (td->td_su == mp) {
1571 vfs_write_resume(mp, VR_START_WRITE);
1575 ufs_extattr_uepm_init(&ump->um_extattr);
1576 #ifdef UFS_EXTATTR_AUTOSTART
1577 (void) ufs_extattr_autostart(mp, td);
1586 * Flush out all the files in a filesystem.
1589 ffs_flushfiles(mp, flags, td)
1594 struct ufsmount *ump;
1600 if (mp->mnt_flag & MNT_QUOTA) {
1602 error = vflush(mp, 0, SKIPSYSTEM|flags, td);
1605 for (i = 0; i < MAXQUOTAS; i++) {
1606 error = quotaoff(td, mp, i);
1608 if ((flags & EARLYFLUSH) == 0)
1616 * Here we fall through to vflush again to ensure that
1617 * we have gotten rid of all the system vnodes, unless
1618 * quotas must not be closed.
1622 ASSERT_VOP_LOCKED(ump->um_devvp, "ffs_flushfiles");
1623 if (ump->um_devvp->v_vflag & VV_COPYONWRITE) {
1624 if ((error = vflush(mp, 0, SKIPSYSTEM | flags, td)) != 0)
1626 ffs_snapshot_unmount(mp);
1627 flags |= FORCECLOSE;
1629 * Here we fall through to vflush again to ensure
1630 * that we have gotten rid of all the system vnodes.
1635 * Do not close system files if quotas were not closed, to be
1636 * able to sync the remaining dquots. The freeblks softupdate
1637 * workitems might hold a reference on a dquot, preventing
1638 * quotaoff() from completing. Next round of
1639 * softdep_flushworklist() iteration should process the
1640 * blockers, allowing the next run of quotaoff() to finally
1641 * flush held dquots.
1643 * Otherwise, flush all the files.
1645 if (qerror == 0 && (error = vflush(mp, 0, flags, td)) != 0)
1649 * Flush filesystem metadata.
1651 vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY);
1652 error = VOP_FSYNC(ump->um_devvp, MNT_WAIT, td);
1653 VOP_UNLOCK(ump->um_devvp);
1658 * Get filesystem statistics.
1665 struct ufsmount *ump;
1670 if (fs->fs_magic != FS_UFS1_MAGIC && fs->fs_magic != FS_UFS2_MAGIC)
1671 panic("ffs_statfs");
1672 sbp->f_version = STATFS_VERSION;
1673 sbp->f_bsize = fs->fs_fsize;
1674 sbp->f_iosize = fs->fs_bsize;
1675 sbp->f_blocks = fs->fs_dsize;
1677 sbp->f_bfree = fs->fs_cstotal.cs_nbfree * fs->fs_frag +
1678 fs->fs_cstotal.cs_nffree + dbtofsb(fs, fs->fs_pendingblocks);
1679 sbp->f_bavail = freespace(fs, fs->fs_minfree) +
1680 dbtofsb(fs, fs->fs_pendingblocks);
1681 sbp->f_files = fs->fs_ncg * fs->fs_ipg - UFS_ROOTINO;
1682 sbp->f_ffree = fs->fs_cstotal.cs_nifree + fs->fs_pendinginodes;
1684 sbp->f_namemax = UFS_MAXNAMLEN;
1689 sync_doupdate(struct inode *ip)
1692 return ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_MODIFIED |
1697 ffs_sync_lazy_filter(struct vnode *vp, void *arg __unused)
1702 * Flags are safe to access because ->v_data invalidation
1703 * is held off by listmtx.
1705 if (vp->v_type == VNON)
1708 if (!sync_doupdate(ip) && (vp->v_iflag & VI_OWEINACT) == 0)
1714 * For a lazy sync, we only care about access times, quotas and the
1715 * superblock. Other filesystem changes are already converted to
1716 * cylinder group blocks or inode blocks updates and are written to
1723 struct vnode *mvp, *vp;
1726 int allerror, error;
1730 if ((mp->mnt_flag & MNT_NOATIME) != 0) {
1736 MNT_VNODE_FOREACH_LAZY(vp, mp, mvp, ffs_sync_lazy_filter, NULL) {
1737 if (vp->v_type == VNON) {
1744 * The IN_ACCESS flag is converted to IN_MODIFIED by
1745 * ufs_close() and ufs_getattr() by the calls to
1746 * ufs_itimes_locked(), without subsequent UFS_UPDATE().
1747 * Test also all the other timestamp flags too, to pick up
1748 * any other cases that could be missed.
1750 if (!sync_doupdate(ip) && (vp->v_iflag & VI_OWEINACT) == 0) {
1754 if ((error = vget(vp, LK_EXCLUSIVE | LK_NOWAIT | LK_INTERLOCK)) != 0)
1759 if (sync_doupdate(ip))
1760 error = ffs_update(vp, 0);
1766 if (VFSTOUFS(mp)->um_fs->fs_fmod != 0 &&
1767 (error = ffs_sbupdate(VFSTOUFS(mp), MNT_LAZY, 0)) != 0)
1773 * Go through the disk queues to initiate sandbagged IO;
1774 * go through the inodes to write those that have been modified;
1775 * initiate the writing of the super block if it has been modified.
1777 * Note: we are always called with the filesystem marked busy using
1781 ffs_sync(mp, waitfor)
1785 struct vnode *mvp, *vp, *devvp;
1788 struct ufsmount *ump = VFSTOUFS(mp);
1790 int error, count, lockreq, allerror = 0;
1793 int secondary_writes;
1794 int secondary_accwrites;
1796 int softdep_accdeps;
1803 if (fs->fs_fmod != 0 && fs->fs_ronly != 0 && ump->um_fsckpid == 0)
1804 panic("%s: ffs_sync: modification on read-only filesystem",
1806 if (waitfor == MNT_LAZY) {
1808 return (ffs_sync_lazy(mp));
1809 waitfor = MNT_NOWAIT;
1813 * Write back each (modified) inode.
1815 lockreq = LK_EXCLUSIVE | LK_NOWAIT;
1816 if (waitfor == MNT_SUSPEND) {
1820 if (waitfor == MNT_WAIT)
1821 lockreq = LK_EXCLUSIVE;
1822 lockreq |= LK_INTERLOCK | LK_SLEEPFAIL;
1824 /* Grab snapshot of secondary write counts */
1826 secondary_writes = mp->mnt_secondary_writes;
1827 secondary_accwrites = mp->mnt_secondary_accwrites;
1830 /* Grab snapshot of softdep dependency counts */
1831 softdep_get_depcounts(mp, &softdep_deps, &softdep_accdeps);
1833 MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
1835 * Depend on the vnode interlock to keep things stable enough
1836 * for a quick test. Since there might be hundreds of
1837 * thousands of vnodes, we cannot afford even a subroutine
1838 * call unless there's a good chance that we have work to do.
1840 if (vp->v_type == VNON) {
1846 (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) == 0 &&
1847 vp->v_bufobj.bo_dirty.bv_cnt == 0) {
1851 if ((error = vget(vp, lockreq)) != 0) {
1852 if (error == ENOENT || error == ENOLCK) {
1853 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
1862 error = ffs_syncvnode(vp, waitfor, 0);
1863 if (error == ERELOOKUP)
1872 * Force stale filesystem control information to be flushed.
1874 if (waitfor == MNT_WAIT || rebooting) {
1875 if ((error = softdep_flushworklist(ump->um_mountp, &count, td)))
1877 if (ffs_fsfail_cleanup(ump, allerror))
1879 /* Flushed work items may create new vnodes to clean */
1880 if (allerror == 0 && count)
1884 devvp = ump->um_devvp;
1885 bo = &devvp->v_bufobj;
1887 if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0) {
1889 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
1890 error = VOP_FSYNC(devvp, waitfor, td);
1892 if (MOUNTEDSOFTDEP(mp) && (error == 0 || error == EAGAIN))
1893 error = ffs_sbupdate(ump, waitfor, 0);
1896 if (ffs_fsfail_cleanup(ump, allerror))
1898 if (allerror == 0 && waitfor == MNT_WAIT)
1900 } else if (suspend != 0) {
1901 if (softdep_check_suspend(mp,
1906 secondary_accwrites) != 0) {
1908 goto loop; /* More work needed */
1910 mtx_assert(MNT_MTX(mp), MA_OWNED);
1911 mp->mnt_kern_flag |= MNTK_SUSPEND2 | MNTK_SUSPENDED;
1917 * Write back modified superblock.
1919 if (fs->fs_fmod != 0 &&
1920 (error = ffs_sbupdate(ump, waitfor, suspended)) != 0)
1922 if (ffs_fsfail_cleanup(ump, allerror))
1928 ffs_vget(mp, ino, flags, vpp)
1934 return (ffs_vgetf(mp, ino, flags, vpp, 0));
1938 ffs_vgetf(mp, ino, flags, vpp, ffs_flags)
1947 struct ufsmount *ump;
1953 MPASS((ffs_flags & (FFSV_REPLACE | FFSV_REPLACE_DOOMED)) == 0 ||
1954 (flags & LK_EXCLUSIVE) != 0);
1956 error = vfs_hash_get(mp, ino, flags, curthread, vpp, NULL, NULL);
1960 if ((ffs_flags & FFSV_REPLACE) == 0 ||
1961 ((ffs_flags & FFSV_REPLACE_DOOMED) == 0 ||
1962 !VN_IS_DOOMED(*vpp)))
1969 * We must promote to an exclusive lock for vnode creation. This
1970 * can happen if lookup is passed LOCKSHARED.
1972 if ((flags & LK_TYPE_MASK) == LK_SHARED) {
1973 flags &= ~LK_TYPE_MASK;
1974 flags |= LK_EXCLUSIVE;
1978 * We do not lock vnode creation as it is believed to be too
1979 * expensive for such rare case as simultaneous creation of vnode
1980 * for same ino by different processes. We just allow them to race
1981 * and check later to decide who wins. Let the race begin!
1986 ip = uma_zalloc_smr(uma_inode, M_WAITOK | M_ZERO);
1988 /* Allocate a new vnode/inode. */
1989 error = getnewvnode("ufs", mp, fs->fs_magic == FS_UFS1_MAGIC ?
1990 &ffs_vnodeops1 : &ffs_vnodeops2, &vp);
1993 uma_zfree_smr(uma_inode, ip);
1997 * FFS supports recursive locking.
1999 lockmgr(vp->v_vnlock, LK_EXCLUSIVE, NULL);
2002 vp->v_bufobj.bo_bsize = fs->fs_bsize;
2007 ip->i_nextclustercg = -1;
2008 ip->i_flag = fs->fs_magic == FS_UFS1_MAGIC ? 0 : IN_UFS2;
2009 ip->i_mode = 0; /* ensure error cases below throw away vnode */
2010 cluster_init_vn(&ip->i_clusterw);
2012 ufs_init_trackers(ip);
2017 for (i = 0; i < MAXQUOTAS; i++)
2018 ip->i_dquot[i] = NODQUOT;
2022 if (ffs_flags & FFSV_FORCEINSMQ)
2023 vp->v_vflag |= VV_FORCEINSMQ;
2024 error = insmntque(vp, mp);
2026 uma_zfree_smr(uma_inode, ip);
2030 vp->v_vflag &= ~VV_FORCEINSMQ;
2031 error = vfs_hash_insert(vp, ino, flags, curthread, vpp, NULL, NULL);
2036 * Calls from ffs_valloc() (i.e. FFSV_REPLACE set)
2037 * operate on empty inode, which must not be found by
2038 * other threads until fully filled. Vnode for empty
2039 * inode must be not re-inserted on the hash by other
2040 * thread, after removal by us at the beginning.
2042 MPASS((ffs_flags & FFSV_REPLACE) == 0);
2046 /* Read in the disk contents for the inode, copy into the inode. */
2047 dbn = fsbtodb(fs, ino_to_fsba(fs, ino));
2048 error = ffs_breadz(ump, ump->um_devvp, dbn, dbn, (int)fs->fs_bsize,
2049 NULL, NULL, 0, NOCRED, 0, NULL, &bp);
2052 * The inode does not contain anything useful, so it would
2053 * be misleading to leave it on its hash chain. With mode
2054 * still zero, it will be unlinked and returned to the free
2063 ip->i_din1 = uma_zalloc(uma_ufs1, M_WAITOK);
2065 ip->i_din2 = uma_zalloc(uma_ufs2, M_WAITOK);
2066 if ((error = ffs_load_inode(bp, ip, fs, ino)) != 0) {
2073 if (DOINGSOFTDEP(vp) && (!fs->fs_ronly ||
2074 (ffs_flags & FFSV_FORCEINODEDEP) != 0))
2075 softdep_load_inodeblock(ip);
2077 ip->i_effnlink = ip->i_nlink;
2081 * Initialize the vnode from the inode, check for aliases.
2082 * Note that the underlying vnode may have changed.
2084 error = ufs_vinit(mp, I_IS_UFS1(ip) ? &ffs_fifoops1 : &ffs_fifoops2,
2094 * Finish inode initialization.
2096 if (vp->v_type != VFIFO) {
2097 /* FFS supports shared locking for all files except fifos. */
2102 * Set up a generation number for this inode if it does not
2103 * already have one. This should only happen on old filesystems.
2105 if (ip->i_gen == 0) {
2106 while (ip->i_gen == 0)
2107 ip->i_gen = arc4random();
2108 if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
2109 UFS_INODE_SET_FLAG(ip, IN_MODIFIED);
2110 DIP_SET(ip, i_gen, ip->i_gen);
2114 if ((mp->mnt_flag & MNT_MULTILABEL) && ip->i_mode) {
2116 * If this vnode is already allocated, and we're running
2117 * multi-label, attempt to perform a label association
2118 * from the extended attributes on the inode.
2120 error = mac_vnode_associate_extattr(mp, vp);
2122 /* ufs_inactive will release ip->i_devvp ref. */
2136 * File handle to vnode
2138 * Have to be really careful about stale file handles:
2139 * - check that the inode number is valid
2140 * - for UFS2 check that the inode number is initialized
2141 * - call ffs_vget() to get the locked inode
2142 * - check for an unallocated inode (i_mode == 0)
2143 * - check that the given client host has export rights and return
2144 * those rights via. exflagsp and credanonp
2147 ffs_fhtovp(mp, fhp, flags, vpp)
2155 ufhp = (struct ufid *)fhp;
2156 return (ffs_inotovp(mp, ufhp->ufid_ino, ufhp->ufid_gen, flags,
2161 ffs_inotovp(mp, ino, gen, lflags, vpp, ffs_flags)
2169 struct ufsmount *ump;
2182 if (ino < UFS_ROOTINO || ino >= fs->fs_ncg * fs->fs_ipg)
2186 * Need to check if inode is initialized because UFS2 does lazy
2187 * initialization and nfs_fhtovp can offer arbitrary inode numbers.
2189 if (fs->fs_magic == FS_UFS2_MAGIC) {
2190 cg = ino_to_cg(fs, ino);
2191 error = ffs_getcg(fs, ump->um_devvp, cg, 0, &bp, &cgp);
2194 if (ino >= cg * fs->fs_ipg + cgp->cg_initediblk) {
2201 error = ffs_vgetf(mp, ino, lflags, &nvp, ffs_flags);
2206 if (ip->i_mode == 0 || ip->i_gen != gen || ip->i_effnlink <= 0) {
2207 if (ip->i_mode == 0)
2213 vnode_create_vobject(nvp, DIP(ip, i_size), curthread);
2219 * Initialize the filesystem.
2223 struct vfsconf *vfsp;
2226 ffs_susp_initialize();
2227 softdep_initialize();
2228 return (ufs_init(vfsp));
2232 * Undo the work of ffs_init().
2236 struct vfsconf *vfsp;
2240 ret = ufs_uninit(vfsp);
2241 softdep_uninitialize();
2242 ffs_susp_uninitialize();
2243 taskqueue_drain_all(taskqueue_thread);
2248 * Structure used to pass information from ffs_sbupdate to its
2249 * helper routine ffs_use_bwrite.
2252 struct ufsmount *ump;
2260 * Write a superblock and associated information back to disk.
2263 ffs_sbupdate(ump, waitfor, suspended)
2264 struct ufsmount *ump;
2273 if (fs->fs_ronly == 1 &&
2274 (ump->um_mountp->mnt_flag & (MNT_RDONLY | MNT_UPDATE)) !=
2275 (MNT_RDONLY | MNT_UPDATE) && ump->um_fsckpid == 0)
2276 panic("ffs_sbupdate: write read-only filesystem");
2278 * We use the superblock's buf to serialize calls to ffs_sbupdate().
2280 sbbp = getblk(ump->um_devvp, btodb(fs->fs_sblockloc),
2281 (int)fs->fs_sbsize, 0, 0, 0);
2283 * Initialize info needed for write function.
2287 devfd.waitfor = waitfor;
2288 devfd.suspended = suspended;
2290 return (ffs_sbput(&devfd, fs, fs->fs_sblockloc, ffs_use_bwrite));
2294 * Write function for use by filesystem-layer routines.
2297 ffs_use_bwrite(void *devfd, off_t loc, void *buf, int size)
2299 struct devfd *devfdp;
2300 struct ufsmount *ump;
2309 * Writing the superblock summary information.
2311 if (loc != fs->fs_sblockloc) {
2312 bp = getblk(ump->um_devvp, btodb(loc), size, 0, 0, 0);
2313 bcopy(buf, bp->b_data, (u_int)size);
2314 if (devfdp->suspended)
2315 bp->b_flags |= B_VALIDSUSPWRT;
2316 if (devfdp->waitfor != MNT_WAIT)
2318 else if ((error = bwrite(bp)) != 0)
2319 devfdp->error = error;
2323 * Writing the superblock itself. We need to do special checks for it.
2326 if (ffs_fsfail_cleanup(ump, devfdp->error))
2328 if (devfdp->error != 0) {
2330 return (devfdp->error);
2332 if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_sblockloc != SBLOCK_UFS1 &&
2333 (fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) {
2334 printf("WARNING: %s: correcting fs_sblockloc from %jd to %d\n",
2335 fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS1);
2336 fs->fs_sblockloc = SBLOCK_UFS1;
2338 if (fs->fs_magic == FS_UFS2_MAGIC && fs->fs_sblockloc != SBLOCK_UFS2 &&
2339 (fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) {
2340 printf("WARNING: %s: correcting fs_sblockloc from %jd to %d\n",
2341 fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS2);
2342 fs->fs_sblockloc = SBLOCK_UFS2;
2344 if (MOUNTEDSOFTDEP(ump->um_mountp))
2345 softdep_setup_sbupdate(ump, (struct fs *)bp->b_data, bp);
2346 bcopy((caddr_t)fs, bp->b_data, (u_int)fs->fs_sbsize);
2347 fs = (struct fs *)bp->b_data;
2348 ffs_oldfscompat_write(fs, ump);
2350 /* Recalculate the superblock hash */
2351 fs->fs_ckhash = ffs_calc_sbhash(fs);
2352 if (devfdp->suspended)
2353 bp->b_flags |= B_VALIDSUSPWRT;
2354 if (devfdp->waitfor != MNT_WAIT)
2356 else if ((error = bwrite(bp)) != 0)
2357 devfdp->error = error;
2358 return (devfdp->error);
2362 ffs_extattrctl(struct mount *mp, int cmd, struct vnode *filename_vp,
2363 int attrnamespace, const char *attrname)
2367 return (ufs_extattrctl(mp, cmd, filename_vp, attrnamespace,
2370 return (vfs_stdextattrctl(mp, cmd, filename_vp, attrnamespace,
2376 ffs_ifree(struct ufsmount *ump, struct inode *ip)
2379 if (ump->um_fstype == UFS1 && ip->i_din1 != NULL)
2380 uma_zfree(uma_ufs1, ip->i_din1);
2381 else if (ip->i_din2 != NULL)
2382 uma_zfree(uma_ufs2, ip->i_din2);
2383 uma_zfree_smr(uma_inode, ip);
2386 static int dobkgrdwrite = 1;
2387 SYSCTL_INT(_debug, OID_AUTO, dobkgrdwrite, CTLFLAG_RW, &dobkgrdwrite, 0,
2388 "Do background writes (honoring the BV_BKGRDWRITE flag)?");
2391 * Complete a background write started from bwrite.
2394 ffs_backgroundwritedone(struct buf *bp)
2396 struct bufobj *bufobj;
2400 if (!LIST_EMPTY(&bp->b_dep) && (bp->b_ioflags & BIO_ERROR) != 0)
2401 softdep_handle_error(bp);
2405 * Find the original buffer that we are writing.
2407 bufobj = bp->b_bufobj;
2409 if ((origbp = gbincore(bp->b_bufobj, bp->b_lblkno)) == NULL)
2410 panic("backgroundwritedone: lost buffer");
2413 * We should mark the cylinder group buffer origbp as
2414 * dirty, to not lose the failed write.
2416 if ((bp->b_ioflags & BIO_ERROR) != 0)
2417 origbp->b_vflags |= BV_BKGRDERR;
2420 * Process dependencies then return any unfinished ones.
2422 if (!LIST_EMPTY(&bp->b_dep) && (bp->b_ioflags & BIO_ERROR) == 0)
2425 if (!LIST_EMPTY(&bp->b_dep))
2426 softdep_move_dependencies(bp, origbp);
2429 * This buffer is marked B_NOCACHE so when it is released
2430 * by biodone it will be tossed. Clear B_IOSTARTED in case of error.
2432 bp->b_flags |= B_NOCACHE;
2433 bp->b_flags &= ~(B_CACHE | B_IOSTARTED);
2437 * Prevent brelse() from trying to keep and re-dirtying bp on
2438 * errors. It causes b_bufobj dereference in
2439 * bdirty()/reassignbuf(), and b_bufobj was cleared in
2442 if ((bp->b_ioflags & BIO_ERROR) != 0)
2443 bp->b_flags |= B_INVAL;
2447 * Clear the BV_BKGRDINPROG flag in the original buffer
2448 * and awaken it if it is waiting for the write to complete.
2449 * If BV_BKGRDINPROG is not set in the original buffer it must
2450 * have been released and re-instantiated - which is not legal.
2452 KASSERT((origbp->b_vflags & BV_BKGRDINPROG),
2453 ("backgroundwritedone: lost buffer2"));
2454 origbp->b_vflags &= ~BV_BKGRDINPROG;
2455 if (origbp->b_vflags & BV_BKGRDWAIT) {
2456 origbp->b_vflags &= ~BV_BKGRDWAIT;
2457 wakeup(&origbp->b_xflags);
2463 * Write, release buffer on completion. (Done by iodone
2464 * if async). Do not bother writing anything if the buffer
2467 * Note that we set B_CACHE here, indicating that buffer is
2468 * fully valid and thus cacheable. This is true even of NFS
2469 * now so we set it generally. This could be set either here
2470 * or in biodone() since the I/O is synchronous. We put it
2474 ffs_bufwrite(struct buf *bp)
2479 CTR3(KTR_BUF, "bufwrite(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
2480 if (bp->b_flags & B_INVAL) {
2485 if (!BUF_ISLOCKED(bp))
2486 panic("bufwrite: buffer is not busy???");
2488 * If a background write is already in progress, delay
2489 * writing this block if it is asynchronous. Otherwise
2490 * wait for the background write to complete.
2492 BO_LOCK(bp->b_bufobj);
2493 if (bp->b_vflags & BV_BKGRDINPROG) {
2494 if (bp->b_flags & B_ASYNC) {
2495 BO_UNLOCK(bp->b_bufobj);
2499 bp->b_vflags |= BV_BKGRDWAIT;
2500 msleep(&bp->b_xflags, BO_LOCKPTR(bp->b_bufobj), PRIBIO,
2502 if (bp->b_vflags & BV_BKGRDINPROG)
2503 panic("bufwrite: still writing");
2505 bp->b_vflags &= ~BV_BKGRDERR;
2506 BO_UNLOCK(bp->b_bufobj);
2509 * If this buffer is marked for background writing and we
2510 * do not have to wait for it, make a copy and write the
2511 * copy so as to leave this buffer ready for further use.
2513 * This optimization eats a lot of memory. If we have a page
2514 * or buffer shortfall we can't do it.
2516 if (dobkgrdwrite && (bp->b_xflags & BX_BKGRDWRITE) &&
2517 (bp->b_flags & B_ASYNC) &&
2518 !vm_page_count_severe() &&
2519 !buf_dirty_count_severe()) {
2520 KASSERT(bp->b_iodone == NULL,
2521 ("bufwrite: needs chained iodone (%p)", bp->b_iodone));
2523 /* get a new block */
2524 newbp = geteblk(bp->b_bufsize, GB_NOWAIT_BD);
2528 KASSERT(buf_mapped(bp), ("Unmapped cg"));
2529 memcpy(newbp->b_data, bp->b_data, bp->b_bufsize);
2530 BO_LOCK(bp->b_bufobj);
2531 bp->b_vflags |= BV_BKGRDINPROG;
2532 BO_UNLOCK(bp->b_bufobj);
2534 (bp->b_xflags & BX_FSPRIV) | BX_BKGRDMARKER;
2535 newbp->b_lblkno = bp->b_lblkno;
2536 newbp->b_blkno = bp->b_blkno;
2537 newbp->b_offset = bp->b_offset;
2538 newbp->b_iodone = ffs_backgroundwritedone;
2539 newbp->b_flags |= B_ASYNC;
2540 newbp->b_flags &= ~B_INVAL;
2541 pbgetvp(bp->b_vp, newbp);
2545 * Move over the dependencies. If there are rollbacks,
2546 * leave the parent buffer dirtied as it will need to
2549 if (LIST_EMPTY(&bp->b_dep) ||
2550 softdep_move_dependencies(bp, newbp) == 0)
2557 * Initiate write on the copy, release the original. The
2558 * BKGRDINPROG flag prevents it from going away until
2559 * the background write completes. We have to recalculate
2560 * its check hash in case the buffer gets freed and then
2561 * reconstituted from the buffer cache during a later read.
2563 if ((bp->b_xflags & BX_CYLGRP) != 0) {
2564 cgp = (struct cg *)bp->b_data;
2567 calculate_crc32c(~0L, bp->b_data, bp->b_bcount);
2572 /* Mark the buffer clean */
2575 /* Let the normal bufwrite do the rest for us */
2578 * If we are writing a cylinder group, update its time.
2580 if ((bp->b_xflags & BX_CYLGRP) != 0) {
2581 cgp = (struct cg *)bp->b_data;
2582 cgp->cg_old_time = cgp->cg_time = time_second;
2584 return (bufwrite(bp));
2588 ffs_geom_strategy(struct bufobj *bo, struct buf *bp)
2595 * This is the bufobj strategy for the private VCHR vnodes
2596 * used by FFS to access the underlying storage device.
2597 * We override the default bufobj strategy and thus bypass
2598 * VOP_STRATEGY() for these vnodes.
2601 KASSERT(bp->b_vp == NULL || bp->b_vp->v_type != VCHR ||
2602 bp->b_vp->v_rdev == NULL ||
2603 bp->b_vp->v_rdev->si_mountpt == NULL ||
2604 VFSTOUFS(bp->b_vp->v_rdev->si_mountpt) == NULL ||
2605 vp == VFSTOUFS(bp->b_vp->v_rdev->si_mountpt)->um_devvp,
2606 ("ffs_geom_strategy() with wrong vp"));
2607 if (bp->b_iocmd == BIO_WRITE) {
2608 if ((bp->b_flags & B_VALIDSUSPWRT) == 0 &&
2609 bp->b_vp != NULL && bp->b_vp->v_mount != NULL &&
2610 (bp->b_vp->v_mount->mnt_kern_flag & MNTK_SUSPENDED) != 0)
2611 panic("ffs_geom_strategy: bad I/O");
2612 nocopy = bp->b_flags & B_NOCOPY;
2613 bp->b_flags &= ~(B_VALIDSUSPWRT | B_NOCOPY);
2614 if ((vp->v_vflag & VV_COPYONWRITE) && nocopy == 0 &&
2615 vp->v_rdev->si_snapdata != NULL) {
2616 if ((bp->b_flags & B_CLUSTER) != 0) {
2617 runningbufwakeup(bp);
2618 TAILQ_FOREACH(tbp, &bp->b_cluster.cluster_head,
2619 b_cluster.cluster_entry) {
2620 error = ffs_copyonwrite(vp, tbp);
2622 error != EOPNOTSUPP) {
2623 bp->b_error = error;
2624 bp->b_ioflags |= BIO_ERROR;
2625 bp->b_flags &= ~B_BARRIER;
2630 bp->b_runningbufspace = bp->b_bufsize;
2631 atomic_add_long(&runningbufspace,
2632 bp->b_runningbufspace);
2634 error = ffs_copyonwrite(vp, bp);
2635 if (error != 0 && error != EOPNOTSUPP) {
2636 bp->b_error = error;
2637 bp->b_ioflags |= BIO_ERROR;
2638 bp->b_flags &= ~B_BARRIER;
2645 if ((bp->b_flags & B_CLUSTER) != 0) {
2646 TAILQ_FOREACH(tbp, &bp->b_cluster.cluster_head,
2647 b_cluster.cluster_entry) {
2648 if (!LIST_EMPTY(&tbp->b_dep))
2652 if (!LIST_EMPTY(&bp->b_dep))
2658 * Check for metadata that needs check-hashes and update them.
2660 switch (bp->b_xflags & BX_FSPRIV) {
2662 ((struct cg *)bp->b_data)->cg_ckhash = 0;
2663 ((struct cg *)bp->b_data)->cg_ckhash =
2664 calculate_crc32c(~0L, bp->b_data, bp->b_bcount);
2671 printf("Check-hash write is unimplemented!!!\n");
2678 printf("multiple buffer types 0x%b\n",
2679 (u_int)(bp->b_xflags & BX_FSPRIV),
2680 PRINT_UFS_BUF_XFLAGS);
2684 if (bp->b_iocmd != BIO_READ && ffs_enxio_enable)
2685 bp->b_xflags |= BX_CVTENXIO;
2686 g_vfs_strategy(bo, bp);
2690 ffs_own_mount(const struct mount *mp)
2693 if (mp->mnt_op == &ufs_vfsops)
2701 /* defined in ffs_softdep.c */
2702 extern void db_print_ffs(struct ufsmount *ump);
2704 DB_SHOW_COMMAND(ffs, db_show_ffs)
2707 struct ufsmount *ump;
2710 ump = VFSTOUFS((struct mount *)addr);
2715 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2716 if (!strcmp(mp->mnt_stat.f_fstypename, ufs_vfsconf.vfc_name))
2717 db_print_ffs(VFSTOUFS(mp));
2721 #endif /* SOFTUPDATES */