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;
88 static int ffs_mountfs(struct vnode *, struct mount *, struct thread *);
89 static void ffs_oldfscompat_read(struct fs *, struct ufsmount *,
91 static void ffs_ifree(struct ufsmount *ump, struct inode *ip);
92 static int ffs_sync_lazy(struct mount *mp);
93 static int ffs_use_bread(void *devfd, off_t loc, void **bufp, int size);
94 static int ffs_use_bwrite(void *devfd, off_t loc, void *buf, int size);
96 static vfs_init_t ffs_init;
97 static vfs_uninit_t ffs_uninit;
98 static vfs_extattrctl_t ffs_extattrctl;
99 static vfs_cmount_t ffs_cmount;
100 static vfs_unmount_t ffs_unmount;
101 static vfs_mount_t ffs_mount;
102 static vfs_statfs_t ffs_statfs;
103 static vfs_fhtovp_t ffs_fhtovp;
104 static vfs_sync_t ffs_sync;
106 static struct vfsops ufs_vfsops = {
107 .vfs_extattrctl = ffs_extattrctl,
108 .vfs_fhtovp = ffs_fhtovp,
109 .vfs_init = ffs_init,
110 .vfs_mount = ffs_mount,
111 .vfs_cmount = ffs_cmount,
112 .vfs_quotactl = ufs_quotactl,
113 .vfs_root = vfs_cache_root,
114 .vfs_cachedroot = ufs_root,
115 .vfs_statfs = ffs_statfs,
116 .vfs_sync = ffs_sync,
117 .vfs_uninit = ffs_uninit,
118 .vfs_unmount = ffs_unmount,
119 .vfs_vget = ffs_vget,
120 .vfs_susp_clean = process_deferred_inactive,
123 VFS_SET(ufs_vfsops, ufs, 0);
124 MODULE_VERSION(ufs, 1);
126 static b_strategy_t ffs_geom_strategy;
127 static b_write_t ffs_bufwrite;
129 static struct buf_ops ffs_ops = {
131 .bop_write = ffs_bufwrite,
132 .bop_strategy = ffs_geom_strategy,
134 #ifdef NO_FFS_SNAPSHOT
135 .bop_bdflush = bufbdflush,
137 .bop_bdflush = ffs_bdflush,
142 * Note that userquota and groupquota options are not currently used
143 * by UFS/FFS code and generally mount(8) does not pass those options
144 * from userland, but they can be passed by loader(8) via
145 * vfs.root.mountfrom.options.
147 static const char *ffs_opts[] = { "acls", "async", "noatime", "noclusterr",
148 "noclusterw", "noexec", "export", "force", "from", "groupquota",
149 "multilabel", "nfsv4acls", "fsckpid", "snapshot", "nosuid", "suiddir",
150 "nosymfollow", "sync", "union", "userquota", "untrusted", NULL };
152 static int ffs_enxio_enable = 1;
153 SYSCTL_DECL(_vfs_ffs);
154 SYSCTL_INT(_vfs_ffs, OID_AUTO, enxio_enable, CTLFLAG_RWTUN,
155 &ffs_enxio_enable, 0,
156 "enable mapping of other disk I/O errors to ENXIO");
159 * Return buffer with the contents of block "offset" from the beginning of
160 * directory "ip". If "res" is non-zero, fill it in with a pointer to the
161 * remaining space in the directory.
164 ffs_blkatoff(struct vnode *vp, off_t offset, char **res, struct buf **bpp)
174 lbn = lblkno(fs, offset);
175 bsize = blksize(fs, ip, lbn);
178 error = bread(vp, lbn, bsize, NOCRED, &bp);
183 *res = (char *)bp->b_data + blkoff(fs, offset);
189 * Load up the contents of an inode and copy the appropriate pieces
190 * to the incore copy.
193 ffs_load_inode(struct buf *bp, struct inode *ip, struct fs *fs, ino_t ino)
195 struct ufs1_dinode *dip1;
196 struct ufs2_dinode *dip2;
202 *((struct ufs1_dinode *)bp->b_data + ino_to_fsbo(fs, ino));
203 ip->i_mode = dip1->di_mode;
204 ip->i_nlink = dip1->di_nlink;
205 ip->i_effnlink = dip1->di_nlink;
206 ip->i_size = dip1->di_size;
207 ip->i_flags = dip1->di_flags;
208 ip->i_gen = dip1->di_gen;
209 ip->i_uid = dip1->di_uid;
210 ip->i_gid = dip1->di_gid;
213 dip2 = ((struct ufs2_dinode *)bp->b_data + ino_to_fsbo(fs, ino));
214 if ((error = ffs_verify_dinode_ckhash(fs, dip2)) != 0 &&
215 !ffs_fsfail_cleanup(ITOUMP(ip), error)) {
216 printf("%s: inode %jd: check-hash failed\n", fs->fs_fsmnt,
222 ip->i_mode = dip2->di_mode;
223 ip->i_nlink = dip2->di_nlink;
224 ip->i_effnlink = dip2->di_nlink;
225 ip->i_size = dip2->di_size;
226 ip->i_flags = dip2->di_flags;
227 ip->i_gen = dip2->di_gen;
228 ip->i_uid = dip2->di_uid;
229 ip->i_gid = dip2->di_gid;
234 * Verify that a filesystem block number is a valid data block.
235 * This routine is only called on untrusted filesystems.
238 ffs_check_blkno(struct mount *mp, ino_t inum, ufs2_daddr_t daddr, int blksize)
241 struct ufsmount *ump;
242 ufs2_daddr_t end_daddr;
245 KASSERT((mp->mnt_flag & MNT_UNTRUSTED) != 0,
246 ("ffs_check_blkno called on a trusted file system"));
249 cg = dtog(fs, daddr);
250 end_daddr = daddr + numfrags(fs, blksize);
252 * Verify that the block number is a valid data block. Also check
253 * that it does not point to an inode block or a superblock. Accept
254 * blocks that are unalloacted (0) or part of snapshot metadata
255 * (BLK_NOCOPY or BLK_SNAP).
257 * Thus, the block must be in a valid range for the filesystem and
258 * either in the space before a backup superblock (except the first
259 * cylinder group where that space is used by the bootstrap code) or
260 * after the inode blocks and before the end of the cylinder group.
262 if ((uint64_t)daddr <= BLK_SNAP ||
263 ((uint64_t)end_daddr <= fs->fs_size &&
264 ((cg > 0 && end_daddr <= cgsblock(fs, cg)) ||
265 (daddr >= cgdmin(fs, cg) &&
266 end_daddr <= cgbase(fs, cg) + fs->fs_fpg))))
268 if ((havemtx = mtx_owned(UFS_MTX(ump))) == 0)
270 if (ppsratecheck(&ump->um_last_integritymsg,
271 &ump->um_secs_integritymsg, 1)) {
273 uprintf("\n%s: inode %jd, out-of-range indirect block "
274 "number %jd\n", mp->mnt_stat.f_mntonname, inum, daddr);
283 * Initiate a forcible unmount.
284 * Used to unmount filesystems whose underlying media has gone away.
287 ffs_fsfail_unmount(void *v, int pending)
289 struct fsfail_task *etp;
295 * Find our mount and get a ref on it, then try to unmount.
297 mp = vfs_getvfs(&etp->fsid);
299 dounmount(mp, MNT_FORCE, curthread);
304 * On first ENXIO error, start a task that forcibly unmounts the filesystem.
306 * Return true if a cleanup is in progress.
309 ffs_fsfail_cleanup(struct ufsmount *ump, int error)
314 retval = ffs_fsfail_cleanup_locked(ump, error);
320 ffs_fsfail_cleanup_locked(struct ufsmount *ump, int error)
322 struct fsfail_task *etp;
325 mtx_assert(UFS_MTX(ump), MA_OWNED);
326 if (error == ENXIO && (ump->um_flags & UM_FSFAIL_CLEANUP) == 0) {
327 ump->um_flags |= UM_FSFAIL_CLEANUP;
329 * Queue an async forced unmount.
331 etp = ump->um_fsfail_task;
332 ump->um_fsfail_task = NULL;
335 TASK_INIT(tp, 0, ffs_fsfail_unmount, etp);
336 taskqueue_enqueue(taskqueue_thread, tp);
337 printf("UFS: forcibly unmounting %s from %s\n",
338 ump->um_mountp->mnt_stat.f_mntfromname,
339 ump->um_mountp->mnt_stat.f_mntonname);
342 return ((ump->um_flags & UM_FSFAIL_CLEANUP) != 0);
346 * Wrapper used during ENXIO cleanup to allocate empty buffers when
347 * the kernel is unable to read the real one. They are needed so that
348 * the soft updates code can use them to unwind its dependencies.
351 ffs_breadz(struct ufsmount *ump, struct vnode *vp, daddr_t lblkno,
352 daddr_t dblkno, int size, daddr_t *rablkno, int *rabsize, int cnt,
353 struct ucred *cred, int flags, void (*ckhashfunc)(struct buf *),
358 flags |= GB_CVTENXIO;
359 error = breadn_flags(vp, lblkno, dblkno, size, rablkno, rabsize, cnt,
360 cred, flags, ckhashfunc, bpp);
361 if (error != 0 && ffs_fsfail_cleanup(ump, error)) {
362 error = getblkx(vp, lblkno, dblkno, size, 0, 0, flags, bpp);
363 KASSERT(error == 0, ("getblkx failed"));
364 vfs_bio_bzero_buf(*bpp, 0, size);
370 ffs_mount(struct mount *mp)
372 struct vnode *devvp, *odevvp;
374 struct ufsmount *ump = NULL;
377 int error, error1, flags;
378 uint64_t mntorflags, saved_mnt_flag;
380 struct nameidata ndp;
384 if (vfs_filteropt(mp->mnt_optnew, ffs_opts))
386 if (uma_inode == NULL) {
387 uma_inode = uma_zcreate("FFS inode",
388 sizeof(struct inode), NULL, NULL, NULL, NULL,
390 uma_ufs1 = uma_zcreate("FFS1 dinode",
391 sizeof(struct ufs1_dinode), NULL, NULL, NULL, NULL,
393 uma_ufs2 = uma_zcreate("FFS2 dinode",
394 sizeof(struct ufs2_dinode), NULL, NULL, NULL, NULL,
398 vfs_deleteopt(mp->mnt_optnew, "groupquota");
399 vfs_deleteopt(mp->mnt_optnew, "userquota");
401 fspec = vfs_getopts(mp->mnt_optnew, "from", &error);
406 if (vfs_getopt(mp->mnt_optnew, "untrusted", NULL, NULL) == 0)
407 mntorflags |= MNT_UNTRUSTED;
409 if (vfs_getopt(mp->mnt_optnew, "acls", NULL, NULL) == 0)
410 mntorflags |= MNT_ACLS;
412 if (vfs_getopt(mp->mnt_optnew, "snapshot", NULL, NULL) == 0) {
413 mntorflags |= MNT_SNAPSHOT;
415 * Once we have set the MNT_SNAPSHOT flag, do not
416 * persist "snapshot" in the options list.
418 vfs_deleteopt(mp->mnt_optnew, "snapshot");
419 vfs_deleteopt(mp->mnt_opt, "snapshot");
422 if (vfs_getopt(mp->mnt_optnew, "fsckpid", NULL, NULL) == 0 &&
423 vfs_scanopt(mp->mnt_optnew, "fsckpid", "%d", &fsckpid) == 1) {
425 * Once we have set the restricted PID, do not
426 * persist "fsckpid" in the options list.
428 vfs_deleteopt(mp->mnt_optnew, "fsckpid");
429 vfs_deleteopt(mp->mnt_opt, "fsckpid");
430 if (mp->mnt_flag & MNT_UPDATE) {
431 if (VFSTOUFS(mp)->um_fs->fs_ronly == 0 &&
432 vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0) == 0) {
434 "Checker enable: Must be read-only");
437 } else if (vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0) == 0) {
439 "Checker enable: Must be read-only");
442 /* Set to -1 if we are done */
447 if (vfs_getopt(mp->mnt_optnew, "nfsv4acls", NULL, NULL) == 0) {
448 if (mntorflags & MNT_ACLS) {
450 "\"acls\" and \"nfsv4acls\" options "
451 "are mutually exclusive");
454 mntorflags |= MNT_NFS4ACLS;
458 mp->mnt_flag |= mntorflags;
461 * If updating, check whether changing from read-only to
462 * read/write; if there is no device name, that's all we do.
464 if (mp->mnt_flag & MNT_UPDATE) {
467 odevvp = ump->um_odevvp;
468 devvp = ump->um_devvp;
469 if (fsckpid == -1 && ump->um_fsckpid > 0) {
470 if ((error = ffs_flushfiles(mp, WRITECLOSE, td)) != 0 ||
471 (error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0)
475 * Return to normal read-only mode.
477 error = g_access(ump->um_cp, 0, -1, 0);
481 if (fs->fs_ronly == 0 &&
482 vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0)) {
484 * Flush any dirty data and suspend filesystem.
486 if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0)
488 error = vfs_write_suspend_umnt(mp);
492 * Check for and optionally get rid of files open
496 if (mp->mnt_flag & MNT_FORCE)
498 if (MOUNTEDSOFTDEP(mp)) {
499 error = softdep_flushfiles(mp, flags, td);
501 error = ffs_flushfiles(mp, flags, td);
504 vfs_write_resume(mp, 0);
507 if (fs->fs_pendingblocks != 0 ||
508 fs->fs_pendinginodes != 0) {
509 printf("WARNING: %s Update error: blocks %jd "
510 "files %d\n", fs->fs_fsmnt,
511 (intmax_t)fs->fs_pendingblocks,
512 fs->fs_pendinginodes);
513 fs->fs_pendingblocks = 0;
514 fs->fs_pendinginodes = 0;
516 if ((fs->fs_flags & (FS_UNCLEAN | FS_NEEDSFSCK)) == 0)
518 if ((error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0) {
521 vfs_write_resume(mp, 0);
524 if (MOUNTEDSOFTDEP(mp))
528 * Drop our write and exclusive access.
530 g_access(ump->um_cp, 0, -1, -1);
534 mp->mnt_flag |= MNT_RDONLY;
537 * Allow the writers to note that filesystem
540 vfs_write_resume(mp, 0);
542 if ((mp->mnt_flag & MNT_RELOAD) &&
543 (error = ffs_reload(mp, td, 0)) != 0)
546 !vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0)) {
548 * If we are running a checker, do not allow upgrade.
550 if (ump->um_fsckpid > 0) {
552 "Active checker, cannot upgrade to write");
556 * If upgrade to read-write by non-root, then verify
557 * that user has necessary permissions on the device.
559 vn_lock(odevvp, LK_EXCLUSIVE | LK_RETRY);
560 error = VOP_ACCESS(odevvp, VREAD | VWRITE,
563 error = priv_check(td, PRIV_VFS_MOUNT_PERM);
568 fs->fs_flags &= ~FS_UNCLEAN;
569 if (fs->fs_clean == 0) {
570 fs->fs_flags |= FS_UNCLEAN;
571 if ((mp->mnt_flag & MNT_FORCE) ||
573 (FS_SUJ | FS_NEEDSFSCK)) == 0 &&
574 (fs->fs_flags & FS_DOSOFTDEP))) {
575 printf("WARNING: %s was not properly "
576 "dismounted\n", fs->fs_fsmnt);
579 "R/W mount of %s denied. %s.%s",
581 "Filesystem is not clean - run fsck",
582 (fs->fs_flags & FS_SUJ) == 0 ? "" :
583 " Forced mount will invalidate"
584 " journal contents");
590 * Request exclusive write access.
592 error = g_access(ump->um_cp, 0, 1, 1);
596 if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0)
598 error = vfs_write_suspend_umnt(mp);
603 saved_mnt_flag = MNT_RDONLY;
604 if (MOUNTEDSOFTDEP(mp) && (mp->mnt_flag &
606 saved_mnt_flag |= MNT_ASYNC;
607 mp->mnt_flag &= ~saved_mnt_flag;
609 fs->fs_mtime = time_second;
610 /* check to see if we need to start softdep */
611 if ((fs->fs_flags & FS_DOSOFTDEP) &&
612 (error = softdep_mount(devvp, mp, fs, td->td_ucred))){
615 mp->mnt_flag |= saved_mnt_flag;
617 vfs_write_resume(mp, 0);
621 if ((error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 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(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);
798 vfs_mountedfrom(mp, fspec);
803 * Compatibility with old mount system call.
807 ffs_cmount(struct mntarg *ma, void *data, uint64_t flags)
809 struct ufs_args args;
814 error = copyin(data, &args, sizeof args);
818 ma = mount_argsu(ma, "from", args.fspec, MAXPATHLEN);
819 ma = mount_arg(ma, "export", &args.export, sizeof(args.export));
820 error = kernel_mount(ma, flags);
826 * Reload all incore data for a filesystem (used after running fsck on
827 * the root filesystem and finding things to fix). If the 'force' flag
828 * is 0, the filesystem must be mounted read-only.
830 * Things to do to update the mount:
831 * 1) invalidate all cached meta-data.
832 * 2) re-read superblock from disk.
833 * 3) re-read summary information from disk.
834 * 4) invalidate all inactive vnodes.
835 * 5) clear MNTK_SUSPEND2 and MNTK_SUSPENDED flags, allowing secondary
836 * writers, if requested.
837 * 6) invalidate all cached file data.
838 * 7) re-read inode data for all active vnodes.
841 ffs_reload(struct mount *mp, struct thread *td, int flags)
843 struct vnode *vp, *mvp, *devvp;
847 struct fs *fs, *newfs;
848 struct ufsmount *ump;
849 ufs2_daddr_t sblockloc;
857 if ((mp->mnt_flag & MNT_RDONLY) == 0 && (flags & FFSR_FORCE) == 0) {
864 * Step 1: invalidate all cached meta-data.
866 devvp = VFSTOUFS(mp)->um_devvp;
867 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
868 if (vinvalbuf(devvp, 0, 0, 0) != 0)
869 panic("ffs_reload: dirty1");
873 * Step 2: re-read superblock from disk.
875 fs = VFSTOUFS(mp)->um_fs;
876 if ((error = bread(devvp, btodb(fs->fs_sblockloc), fs->fs_sbsize,
879 newfs = (struct fs *)bp->b_data;
880 if ((newfs->fs_magic != FS_UFS1_MAGIC &&
881 newfs->fs_magic != FS_UFS2_MAGIC) ||
882 newfs->fs_bsize > MAXBSIZE ||
883 newfs->fs_bsize < sizeof(struct fs)) {
885 return (EIO); /* XXX needs translation */
888 * Preserve the summary information, read-only status, and
889 * superblock location by copying these fields into our new
890 * superblock before using it to update the existing superblock.
892 newfs->fs_si = fs->fs_si;
893 newfs->fs_ronly = fs->fs_ronly;
894 sblockloc = fs->fs_sblockloc;
895 bcopy(newfs, fs, (u_int)fs->fs_sbsize);
897 mp->mnt_maxsymlinklen = fs->fs_maxsymlinklen;
898 ffs_oldfscompat_read(fs, VFSTOUFS(mp), sblockloc);
900 if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
901 printf("WARNING: %s: reload pending error: blocks %jd "
902 "files %d\n", fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
903 fs->fs_pendinginodes);
904 fs->fs_pendingblocks = 0;
905 fs->fs_pendinginodes = 0;
910 * Step 3: re-read summary information from disk.
912 size = fs->fs_cssize;
913 blks = howmany(size, fs->fs_fsize);
914 if (fs->fs_contigsumsize > 0)
915 size += fs->fs_ncg * sizeof(int32_t);
916 size += fs->fs_ncg * sizeof(u_int8_t);
917 free(fs->fs_csp, M_UFSMNT);
918 space = malloc(size, M_UFSMNT, M_WAITOK);
920 for (i = 0; i < blks; i += fs->fs_frag) {
922 if (i + fs->fs_frag > blks)
923 size = (blks - i) * fs->fs_fsize;
924 error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), size,
928 bcopy(bp->b_data, space, (u_int)size);
929 space = (char *)space + size;
933 * We no longer know anything about clusters per cylinder group.
935 if (fs->fs_contigsumsize > 0) {
936 fs->fs_maxcluster = lp = space;
937 for (i = 0; i < fs->fs_ncg; i++)
938 *lp++ = fs->fs_contigsumsize;
941 size = fs->fs_ncg * sizeof(u_int8_t);
942 fs->fs_contigdirs = (u_int8_t *)space;
943 bzero(fs->fs_contigdirs, size);
944 if ((flags & FFSR_UNSUSPEND) != 0) {
946 mp->mnt_kern_flag &= ~(MNTK_SUSPENDED | MNTK_SUSPEND2);
947 wakeup(&mp->mnt_flag);
952 MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
956 if (vp->v_type == VNON) {
961 * Step 4: invalidate all cached file data.
963 if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, td)) {
964 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
967 if (vinvalbuf(vp, 0, 0, 0))
968 panic("ffs_reload: dirty2");
970 * Step 5: re-read inode data for all active vnodes.
974 bread(devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
975 (int)fs->fs_bsize, NOCRED, &bp);
978 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
981 if ((error = ffs_load_inode(bp, ip, fs, ip->i_number)) != 0) {
984 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
987 ip->i_effnlink = ip->i_nlink;
995 * Common code for mount and mountroot
998 ffs_mountfs(odevvp, mp, td)
999 struct vnode *odevvp;
1003 struct ufsmount *ump;
1006 int error, i, len, ronly;
1008 struct g_consumer *cp;
1010 struct vnode *devvp;
1011 struct fsfail_task *etp;
1012 int candelete, canspeedup;
1017 cred = td ? td->td_ucred : NOCRED;
1018 ronly = (mp->mnt_flag & MNT_RDONLY) != 0;
1020 devvp = mntfs_allocvp(mp, odevvp);
1022 KASSERT(devvp->v_type == VCHR, ("reclaimed devvp"));
1023 dev = devvp->v_rdev;
1024 if (atomic_cmpset_acq_ptr((uintptr_t *)&dev->si_mountpt, 0,
1025 (uintptr_t)mp) == 0) {
1026 mntfs_freevp(devvp);
1030 error = g_vfs_open(devvp, &cp, "ffs", ronly ? 0 : 1);
1031 g_topology_unlock();
1033 atomic_store_rel_ptr((uintptr_t *)&dev->si_mountpt, 0);
1034 mntfs_freevp(devvp);
1038 devvp->v_bufobj.bo_ops = &ffs_ops;
1039 BO_LOCK(&odevvp->v_bufobj);
1040 odevvp->v_bufobj.bo_flag |= BO_NOBUFS;
1041 BO_UNLOCK(&odevvp->v_bufobj);
1042 if (dev->si_iosize_max != 0)
1043 mp->mnt_iosize_max = dev->si_iosize_max;
1044 if (mp->mnt_iosize_max > MAXPHYS)
1045 mp->mnt_iosize_max = MAXPHYS;
1046 if ((SBLOCKSIZE % cp->provider->sectorsize) != 0) {
1049 "Invalid sectorsize %d for superblock size %d",
1050 cp->provider->sectorsize, SBLOCKSIZE);
1053 /* fetch the superblock and summary information */
1055 if ((mp->mnt_flag & MNT_ROOTFS) != 0)
1056 loc = STDSB_NOHASHFAIL;
1057 if ((error = ffs_sbget(devvp, &fs, loc, M_UFSMNT, ffs_use_bread)) != 0)
1059 /* none of these types of check-hashes are maintained by this kernel */
1060 fs->fs_metackhash &= ~(CK_INDIR | CK_DIR);
1061 /* no support for any undefined flags */
1062 fs->fs_flags &= FS_SUPPORTED;
1063 fs->fs_flags &= ~FS_UNCLEAN;
1064 if (fs->fs_clean == 0) {
1065 fs->fs_flags |= FS_UNCLEAN;
1066 if (ronly || (mp->mnt_flag & MNT_FORCE) ||
1067 ((fs->fs_flags & (FS_SUJ | FS_NEEDSFSCK)) == 0 &&
1068 (fs->fs_flags & FS_DOSOFTDEP))) {
1069 printf("WARNING: %s was not properly dismounted\n",
1072 vfs_mount_error(mp, "R/W mount of %s denied. %s%s",
1073 fs->fs_fsmnt, "Filesystem is not clean - run fsck.",
1074 (fs->fs_flags & FS_SUJ) == 0 ? "" :
1075 " Forced mount will invalidate journal contents");
1079 if ((fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) &&
1080 (mp->mnt_flag & MNT_FORCE)) {
1081 printf("WARNING: %s: lost blocks %jd files %d\n",
1082 fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
1083 fs->fs_pendinginodes);
1084 fs->fs_pendingblocks = 0;
1085 fs->fs_pendinginodes = 0;
1088 if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
1089 printf("WARNING: %s: mount pending error: blocks %jd "
1090 "files %d\n", fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
1091 fs->fs_pendinginodes);
1092 fs->fs_pendingblocks = 0;
1093 fs->fs_pendinginodes = 0;
1095 if ((fs->fs_flags & FS_GJOURNAL) != 0) {
1098 * Get journal provider name.
1101 mp->mnt_gjprovider = malloc((u_long)len, M_UFSMNT, M_WAITOK);
1102 if (g_io_getattr("GJOURNAL::provider", cp, &len,
1103 mp->mnt_gjprovider) == 0) {
1104 mp->mnt_gjprovider = realloc(mp->mnt_gjprovider, len,
1105 M_UFSMNT, M_WAITOK);
1107 mp->mnt_flag |= MNT_GJOURNAL;
1110 printf("WARNING: %s: GJOURNAL flag on fs "
1111 "but no gjournal provider below\n",
1112 mp->mnt_stat.f_mntonname);
1113 free(mp->mnt_gjprovider, M_UFSMNT);
1114 mp->mnt_gjprovider = NULL;
1117 printf("WARNING: %s: GJOURNAL flag on fs but no "
1118 "UFS_GJOURNAL support\n", mp->mnt_stat.f_mntonname);
1121 mp->mnt_gjprovider = NULL;
1123 ump = malloc(sizeof *ump, M_UFSMNT, M_WAITOK | M_ZERO);
1125 ump->um_bo = &devvp->v_bufobj;
1127 if (fs->fs_magic == FS_UFS1_MAGIC) {
1128 ump->um_fstype = UFS1;
1129 ump->um_balloc = ffs_balloc_ufs1;
1131 ump->um_fstype = UFS2;
1132 ump->um_balloc = ffs_balloc_ufs2;
1134 ump->um_blkatoff = ffs_blkatoff;
1135 ump->um_truncate = ffs_truncate;
1136 ump->um_update = ffs_update;
1137 ump->um_valloc = ffs_valloc;
1138 ump->um_vfree = ffs_vfree;
1139 ump->um_ifree = ffs_ifree;
1140 ump->um_rdonly = ffs_rdonly;
1141 ump->um_snapgone = ffs_snapgone;
1142 if ((mp->mnt_flag & MNT_UNTRUSTED) != 0)
1143 ump->um_check_blkno = ffs_check_blkno;
1145 ump->um_check_blkno = NULL;
1146 mtx_init(UFS_MTX(ump), "FFS", "FFS Lock", MTX_DEF);
1147 ffs_oldfscompat_read(fs, ump, fs->fs_sblockloc);
1148 fs->fs_ronly = ronly;
1149 fs->fs_active = NULL;
1151 mp->mnt_stat.f_fsid.val[0] = fs->fs_id[0];
1152 mp->mnt_stat.f_fsid.val[1] = fs->fs_id[1];
1154 if (fs->fs_id[0] == 0 || fs->fs_id[1] == 0 ||
1155 (nmp = vfs_getvfs(&mp->mnt_stat.f_fsid))) {
1160 mp->mnt_maxsymlinklen = fs->fs_maxsymlinklen;
1162 mp->mnt_flag |= MNT_LOCAL;
1164 if ((fs->fs_flags & FS_MULTILABEL) != 0) {
1167 mp->mnt_flag |= MNT_MULTILABEL;
1170 printf("WARNING: %s: multilabel flag on fs but "
1171 "no MAC support\n", mp->mnt_stat.f_mntonname);
1174 if ((fs->fs_flags & FS_ACLS) != 0) {
1178 if (mp->mnt_flag & MNT_NFS4ACLS)
1179 printf("WARNING: %s: ACLs flag on fs conflicts with "
1180 "\"nfsv4acls\" mount option; option ignored\n",
1181 mp->mnt_stat.f_mntonname);
1182 mp->mnt_flag &= ~MNT_NFS4ACLS;
1183 mp->mnt_flag |= MNT_ACLS;
1187 printf("WARNING: %s: ACLs flag on fs but no ACLs support\n",
1188 mp->mnt_stat.f_mntonname);
1191 if ((fs->fs_flags & FS_NFS4ACLS) != 0) {
1195 if (mp->mnt_flag & MNT_ACLS)
1196 printf("WARNING: %s: NFSv4 ACLs flag on fs conflicts "
1197 "with \"acls\" mount option; option ignored\n",
1198 mp->mnt_stat.f_mntonname);
1199 mp->mnt_flag &= ~MNT_ACLS;
1200 mp->mnt_flag |= MNT_NFS4ACLS;
1204 printf("WARNING: %s: NFSv4 ACLs flag on fs but no "
1205 "ACLs support\n", mp->mnt_stat.f_mntonname);
1208 if ((fs->fs_flags & FS_TRIM) != 0) {
1210 if (g_io_getattr("GEOM::candelete", cp, &len,
1213 ump->um_flags |= UM_CANDELETE;
1215 printf("WARNING: %s: TRIM flag on fs but disk "
1216 "does not support TRIM\n",
1217 mp->mnt_stat.f_mntonname);
1219 printf("WARNING: %s: TRIM flag on fs but disk does "
1220 "not confirm that it supports TRIM\n",
1221 mp->mnt_stat.f_mntonname);
1223 if (((ump->um_flags) & UM_CANDELETE) != 0) {
1224 ump->um_trim_tq = taskqueue_create("trim", M_WAITOK,
1225 taskqueue_thread_enqueue, &ump->um_trim_tq);
1226 taskqueue_start_threads(&ump->um_trim_tq, 1, PVFS,
1227 "%s trim", mp->mnt_stat.f_mntonname);
1228 ump->um_trimhash = hashinit(MAXTRIMIO, M_TRIM,
1229 &ump->um_trimlisthashsize);
1234 if (g_io_getattr("GEOM::canspeedup", cp, &len, &canspeedup) == 0) {
1236 ump->um_flags |= UM_CANSPEEDUP;
1239 ump->um_mountp = mp;
1241 ump->um_devvp = devvp;
1242 ump->um_odevvp = odevvp;
1243 ump->um_nindir = fs->fs_nindir;
1244 ump->um_bptrtodb = fs->fs_fsbtodb;
1245 ump->um_seqinc = fs->fs_frag;
1246 for (i = 0; i < MAXQUOTAS; i++)
1247 ump->um_quotas[i] = NULLVP;
1249 ufs_extattr_uepm_init(&ump->um_extattr);
1252 * Set FS local "last mounted on" information (NULL pad)
1254 bzero(fs->fs_fsmnt, MAXMNTLEN);
1255 strlcpy(fs->fs_fsmnt, mp->mnt_stat.f_mntonname, MAXMNTLEN);
1256 mp->mnt_stat.f_iosize = fs->fs_bsize;
1258 if (mp->mnt_flag & MNT_ROOTFS) {
1260 * Root mount; update timestamp in mount structure.
1261 * this will be used by the common root mount code
1262 * to update the system clock.
1264 mp->mnt_time = fs->fs_time;
1268 fs->fs_mtime = time_second;
1269 if ((fs->fs_flags & FS_DOSOFTDEP) &&
1270 (error = softdep_mount(devvp, mp, fs, cred)) != 0) {
1271 ffs_flushfiles(mp, FORCECLOSE, td);
1274 if (fs->fs_snapinum[0] != 0)
1275 ffs_snapshot_mount(mp);
1278 (void) ffs_sbupdate(ump, MNT_WAIT, 0);
1281 * Initialize filesystem state information in mount struct.
1284 mp->mnt_kern_flag |= MNTK_LOOKUP_SHARED | MNTK_EXTENDED_SHARED |
1285 MNTK_NO_IOPF | MNTK_UNMAPPED_BUFS | MNTK_USES_BCACHE;
1288 #ifdef UFS_EXTATTR_AUTOSTART
1291 * Auto-starting does the following:
1292 * - check for /.attribute in the fs, and extattr_start if so
1293 * - for each file in .attribute, enable that file with
1294 * an attribute of the same name.
1295 * Not clear how to report errors -- probably eat them.
1296 * This would all happen while the filesystem was busy/not
1297 * available, so would effectively be "atomic".
1299 (void) ufs_extattr_autostart(mp, td);
1300 #endif /* !UFS_EXTATTR_AUTOSTART */
1301 #endif /* !UFS_EXTATTR */
1302 etp = malloc(sizeof *ump->um_fsfail_task, M_UFSMNT, M_WAITOK | M_ZERO);
1303 etp->fsid = mp->mnt_stat.f_fsid;
1304 ump->um_fsfail_task = etp;
1308 free(fs->fs_csp, M_UFSMNT);
1309 free(fs->fs_si, M_UFSMNT);
1315 g_topology_unlock();
1318 mtx_destroy(UFS_MTX(ump));
1319 if (mp->mnt_gjprovider != NULL) {
1320 free(mp->mnt_gjprovider, M_UFSMNT);
1321 mp->mnt_gjprovider = NULL;
1323 free(ump, M_UFSMNT);
1324 mp->mnt_data = NULL;
1326 BO_LOCK(&odevvp->v_bufobj);
1327 odevvp->v_bufobj.bo_flag &= ~BO_NOBUFS;
1328 BO_UNLOCK(&odevvp->v_bufobj);
1329 atomic_store_rel_ptr((uintptr_t *)&dev->si_mountpt, 0);
1330 mntfs_freevp(devvp);
1336 * A read function for use by filesystem-layer routines.
1339 ffs_use_bread(void *devfd, off_t loc, void **bufp, int size)
1344 KASSERT(*bufp == NULL, ("ffs_use_bread: non-NULL *bufp %p\n", *bufp));
1345 *bufp = malloc(size, M_UFSMNT, M_WAITOK);
1346 if ((error = bread((struct vnode *)devfd, btodb(loc), size, NOCRED,
1349 bcopy(bp->b_data, *bufp, size);
1350 bp->b_flags |= B_INVAL | B_NOCACHE;
1355 static int bigcgs = 0;
1356 SYSCTL_INT(_debug, OID_AUTO, bigcgs, CTLFLAG_RW, &bigcgs, 0, "");
1359 * Sanity checks for loading old filesystem superblocks.
1360 * See ffs_oldfscompat_write below for unwound actions.
1362 * XXX - Parts get retired eventually.
1363 * Unfortunately new bits get added.
1366 ffs_oldfscompat_read(fs, ump, sblockloc)
1368 struct ufsmount *ump;
1369 ufs2_daddr_t sblockloc;
1374 * If not yet done, update fs_flags location and value of fs_sblockloc.
1376 if ((fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) {
1377 fs->fs_flags = fs->fs_old_flags;
1378 fs->fs_old_flags |= FS_FLAGS_UPDATED;
1379 fs->fs_sblockloc = sblockloc;
1382 * If not yet done, update UFS1 superblock with new wider fields.
1384 if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_maxbsize != fs->fs_bsize) {
1385 fs->fs_maxbsize = fs->fs_bsize;
1386 fs->fs_time = fs->fs_old_time;
1387 fs->fs_size = fs->fs_old_size;
1388 fs->fs_dsize = fs->fs_old_dsize;
1389 fs->fs_csaddr = fs->fs_old_csaddr;
1390 fs->fs_cstotal.cs_ndir = fs->fs_old_cstotal.cs_ndir;
1391 fs->fs_cstotal.cs_nbfree = fs->fs_old_cstotal.cs_nbfree;
1392 fs->fs_cstotal.cs_nifree = fs->fs_old_cstotal.cs_nifree;
1393 fs->fs_cstotal.cs_nffree = fs->fs_old_cstotal.cs_nffree;
1395 if (fs->fs_magic == FS_UFS1_MAGIC &&
1396 fs->fs_old_inodefmt < FS_44INODEFMT) {
1397 fs->fs_maxfilesize = ((uint64_t)1 << 31) - 1;
1398 fs->fs_qbmask = ~fs->fs_bmask;
1399 fs->fs_qfmask = ~fs->fs_fmask;
1401 if (fs->fs_magic == FS_UFS1_MAGIC) {
1402 ump->um_savedmaxfilesize = fs->fs_maxfilesize;
1403 maxfilesize = (uint64_t)0x80000000 * fs->fs_bsize - 1;
1404 if (fs->fs_maxfilesize > maxfilesize)
1405 fs->fs_maxfilesize = maxfilesize;
1407 /* Compatibility for old filesystems */
1408 if (fs->fs_avgfilesize <= 0)
1409 fs->fs_avgfilesize = AVFILESIZ;
1410 if (fs->fs_avgfpdir <= 0)
1411 fs->fs_avgfpdir = AFPDIR;
1413 fs->fs_save_cgsize = fs->fs_cgsize;
1414 fs->fs_cgsize = fs->fs_bsize;
1419 * Unwinding superblock updates for old filesystems.
1420 * See ffs_oldfscompat_read above for details.
1422 * XXX - Parts get retired eventually.
1423 * Unfortunately new bits get added.
1426 ffs_oldfscompat_write(fs, ump)
1428 struct ufsmount *ump;
1432 * Copy back UFS2 updated fields that UFS1 inspects.
1434 if (fs->fs_magic == FS_UFS1_MAGIC) {
1435 fs->fs_old_time = fs->fs_time;
1436 fs->fs_old_cstotal.cs_ndir = fs->fs_cstotal.cs_ndir;
1437 fs->fs_old_cstotal.cs_nbfree = fs->fs_cstotal.cs_nbfree;
1438 fs->fs_old_cstotal.cs_nifree = fs->fs_cstotal.cs_nifree;
1439 fs->fs_old_cstotal.cs_nffree = fs->fs_cstotal.cs_nffree;
1440 fs->fs_maxfilesize = ump->um_savedmaxfilesize;
1443 fs->fs_cgsize = fs->fs_save_cgsize;
1444 fs->fs_save_cgsize = 0;
1449 * unmount system call
1452 ffs_unmount(mp, mntflags)
1457 struct ufsmount *ump = VFSTOUFS(mp);
1459 int error, flags, susp;
1467 if (mntflags & MNT_FORCE)
1468 flags |= FORCECLOSE;
1469 susp = fs->fs_ronly == 0;
1471 if ((error = ufs_extattr_stop(mp, td))) {
1472 if (error != EOPNOTSUPP)
1473 printf("WARNING: unmount %s: ufs_extattr_stop "
1474 "returned errno %d\n", mp->mnt_stat.f_mntonname,
1478 ufs_extattr_uepm_destroy(&ump->um_extattr);
1483 error = vfs_write_suspend_umnt(mp);
1487 if (MOUNTEDSOFTDEP(mp))
1488 error = softdep_flushfiles(mp, flags, td);
1490 error = ffs_flushfiles(mp, flags, td);
1491 if (error != 0 && !ffs_fsfail_cleanup(ump, error))
1495 if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
1496 printf("WARNING: unmount %s: pending error: blocks %jd "
1497 "files %d\n", fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
1498 fs->fs_pendinginodes);
1499 fs->fs_pendingblocks = 0;
1500 fs->fs_pendinginodes = 0;
1503 if (MOUNTEDSOFTDEP(mp))
1504 softdep_unmount(mp);
1505 if (fs->fs_ronly == 0 || ump->um_fsckpid > 0) {
1506 fs->fs_clean = fs->fs_flags & (FS_UNCLEAN|FS_NEEDSFSCK) ? 0 : 1;
1507 error = ffs_sbupdate(ump, MNT_WAIT, 0);
1508 if (ffs_fsfail_cleanup(ump, error))
1510 if (error != 0 && !ffs_fsfail_cleanup(ump, error)) {
1516 vfs_write_resume(mp, VR_START_WRITE);
1517 if (ump->um_trim_tq != NULL) {
1518 while (ump->um_trim_inflight != 0)
1519 pause("ufsutr", hz);
1520 taskqueue_drain_all(ump->um_trim_tq);
1521 taskqueue_free(ump->um_trim_tq);
1522 free (ump->um_trimhash, M_TRIM);
1525 if (ump->um_fsckpid > 0) {
1527 * Return to normal read-only mode.
1529 error = g_access(ump->um_cp, 0, -1, 0);
1530 ump->um_fsckpid = 0;
1532 g_vfs_close(ump->um_cp);
1533 g_topology_unlock();
1534 BO_LOCK(&ump->um_odevvp->v_bufobj);
1535 ump->um_odevvp->v_bufobj.bo_flag &= ~BO_NOBUFS;
1536 BO_UNLOCK(&ump->um_odevvp->v_bufobj);
1537 atomic_store_rel_ptr((uintptr_t *)&ump->um_dev->si_mountpt, 0);
1538 mntfs_freevp(ump->um_devvp);
1539 vrele(ump->um_odevvp);
1540 dev_rel(ump->um_dev);
1541 mtx_destroy(UFS_MTX(ump));
1542 if (mp->mnt_gjprovider != NULL) {
1543 free(mp->mnt_gjprovider, M_UFSMNT);
1544 mp->mnt_gjprovider = NULL;
1546 free(fs->fs_csp, M_UFSMNT);
1547 free(fs->fs_si, M_UFSMNT);
1549 if (ump->um_fsfail_task != NULL)
1550 free(ump->um_fsfail_task, M_UFSMNT);
1551 free(ump, M_UFSMNT);
1552 mp->mnt_data = NULL;
1554 mp->mnt_flag &= ~MNT_LOCAL;
1556 if (td->td_su == mp) {
1564 vfs_write_resume(mp, VR_START_WRITE);
1568 ufs_extattr_uepm_init(&ump->um_extattr);
1569 #ifdef UFS_EXTATTR_AUTOSTART
1570 (void) ufs_extattr_autostart(mp, td);
1579 * Flush out all the files in a filesystem.
1582 ffs_flushfiles(mp, flags, td)
1587 struct ufsmount *ump;
1593 if (mp->mnt_flag & MNT_QUOTA) {
1595 error = vflush(mp, 0, SKIPSYSTEM|flags, td);
1598 for (i = 0; i < MAXQUOTAS; i++) {
1599 error = quotaoff(td, mp, i);
1601 if ((flags & EARLYFLUSH) == 0)
1609 * Here we fall through to vflush again to ensure that
1610 * we have gotten rid of all the system vnodes, unless
1611 * quotas must not be closed.
1615 ASSERT_VOP_LOCKED(ump->um_devvp, "ffs_flushfiles");
1616 if (ump->um_devvp->v_vflag & VV_COPYONWRITE) {
1617 if ((error = vflush(mp, 0, SKIPSYSTEM | flags, td)) != 0)
1619 ffs_snapshot_unmount(mp);
1620 flags |= FORCECLOSE;
1622 * Here we fall through to vflush again to ensure
1623 * that we have gotten rid of all the system vnodes.
1628 * Do not close system files if quotas were not closed, to be
1629 * able to sync the remaining dquots. The freeblks softupdate
1630 * workitems might hold a reference on a dquot, preventing
1631 * quotaoff() from completing. Next round of
1632 * softdep_flushworklist() iteration should process the
1633 * blockers, allowing the next run of quotaoff() to finally
1634 * flush held dquots.
1636 * Otherwise, flush all the files.
1638 if (qerror == 0 && (error = vflush(mp, 0, flags, td)) != 0)
1642 * Flush filesystem metadata.
1644 vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY);
1645 error = VOP_FSYNC(ump->um_devvp, MNT_WAIT, td);
1646 VOP_UNLOCK(ump->um_devvp);
1651 * Get filesystem statistics.
1658 struct ufsmount *ump;
1663 if (fs->fs_magic != FS_UFS1_MAGIC && fs->fs_magic != FS_UFS2_MAGIC)
1664 panic("ffs_statfs");
1665 sbp->f_version = STATFS_VERSION;
1666 sbp->f_bsize = fs->fs_fsize;
1667 sbp->f_iosize = fs->fs_bsize;
1668 sbp->f_blocks = fs->fs_dsize;
1670 sbp->f_bfree = fs->fs_cstotal.cs_nbfree * fs->fs_frag +
1671 fs->fs_cstotal.cs_nffree + dbtofsb(fs, fs->fs_pendingblocks);
1672 sbp->f_bavail = freespace(fs, fs->fs_minfree) +
1673 dbtofsb(fs, fs->fs_pendingblocks);
1674 sbp->f_files = fs->fs_ncg * fs->fs_ipg - UFS_ROOTINO;
1675 sbp->f_ffree = fs->fs_cstotal.cs_nifree + fs->fs_pendinginodes;
1677 sbp->f_namemax = UFS_MAXNAMLEN;
1682 sync_doupdate(struct inode *ip)
1685 return ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_MODIFIED |
1690 ffs_sync_lazy_filter(struct vnode *vp, void *arg __unused)
1695 * Flags are safe to access because ->v_data invalidation
1696 * is held off by listmtx.
1698 if (vp->v_type == VNON)
1701 if (!sync_doupdate(ip) && (vp->v_iflag & VI_OWEINACT) == 0)
1707 * For a lazy sync, we only care about access times, quotas and the
1708 * superblock. Other filesystem changes are already converted to
1709 * cylinder group blocks or inode blocks updates and are written to
1716 struct vnode *mvp, *vp;
1719 int allerror, error;
1723 if ((mp->mnt_flag & MNT_NOATIME) != 0) {
1729 MNT_VNODE_FOREACH_LAZY(vp, mp, mvp, ffs_sync_lazy_filter, NULL) {
1730 if (vp->v_type == VNON) {
1737 * The IN_ACCESS flag is converted to IN_MODIFIED by
1738 * ufs_close() and ufs_getattr() by the calls to
1739 * ufs_itimes_locked(), without subsequent UFS_UPDATE().
1740 * Test also all the other timestamp flags too, to pick up
1741 * any other cases that could be missed.
1743 if (!sync_doupdate(ip) && (vp->v_iflag & VI_OWEINACT) == 0) {
1747 if ((error = vget(vp, LK_EXCLUSIVE | LK_NOWAIT | LK_INTERLOCK,
1753 if (sync_doupdate(ip))
1754 error = ffs_update(vp, 0);
1760 if (VFSTOUFS(mp)->um_fs->fs_fmod != 0 &&
1761 (error = ffs_sbupdate(VFSTOUFS(mp), MNT_LAZY, 0)) != 0)
1767 * Go through the disk queues to initiate sandbagged IO;
1768 * go through the inodes to write those that have been modified;
1769 * initiate the writing of the super block if it has been modified.
1771 * Note: we are always called with the filesystem marked busy using
1775 ffs_sync(mp, waitfor)
1779 struct vnode *mvp, *vp, *devvp;
1782 struct ufsmount *ump = VFSTOUFS(mp);
1784 int error, count, lockreq, allerror = 0;
1787 int secondary_writes;
1788 int secondary_accwrites;
1790 int softdep_accdeps;
1797 if (fs->fs_fmod != 0 && fs->fs_ronly != 0 && ump->um_fsckpid == 0)
1798 panic("%s: ffs_sync: modification on read-only filesystem",
1800 if (waitfor == MNT_LAZY) {
1802 return (ffs_sync_lazy(mp));
1803 waitfor = MNT_NOWAIT;
1807 * Write back each (modified) inode.
1809 lockreq = LK_EXCLUSIVE | LK_NOWAIT;
1810 if (waitfor == MNT_SUSPEND) {
1814 if (waitfor == MNT_WAIT)
1815 lockreq = LK_EXCLUSIVE;
1816 lockreq |= LK_INTERLOCK | LK_SLEEPFAIL;
1818 /* Grab snapshot of secondary write counts */
1820 secondary_writes = mp->mnt_secondary_writes;
1821 secondary_accwrites = mp->mnt_secondary_accwrites;
1824 /* Grab snapshot of softdep dependency counts */
1825 softdep_get_depcounts(mp, &softdep_deps, &softdep_accdeps);
1827 MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
1829 * Depend on the vnode interlock to keep things stable enough
1830 * for a quick test. Since there might be hundreds of
1831 * thousands of vnodes, we cannot afford even a subroutine
1832 * call unless there's a good chance that we have work to do.
1834 if (vp->v_type == VNON) {
1840 (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) == 0 &&
1841 vp->v_bufobj.bo_dirty.bv_cnt == 0) {
1845 if ((error = vget(vp, lockreq, td)) != 0) {
1846 if (error == ENOENT || error == ENOLCK) {
1847 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
1855 if ((error = ffs_syncvnode(vp, waitfor, 0)) != 0)
1860 * Force stale filesystem control information to be flushed.
1862 if (waitfor == MNT_WAIT || rebooting) {
1863 if ((error = softdep_flushworklist(ump->um_mountp, &count, td)))
1865 if (ffs_fsfail_cleanup(ump, allerror))
1867 /* Flushed work items may create new vnodes to clean */
1868 if (allerror == 0 && count)
1872 devvp = ump->um_devvp;
1873 bo = &devvp->v_bufobj;
1875 if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0) {
1877 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
1878 error = VOP_FSYNC(devvp, waitfor, td);
1880 if (MOUNTEDSOFTDEP(mp) && (error == 0 || error == EAGAIN))
1881 error = ffs_sbupdate(ump, waitfor, 0);
1884 if (ffs_fsfail_cleanup(ump, allerror))
1886 if (allerror == 0 && waitfor == MNT_WAIT)
1888 } else if (suspend != 0) {
1889 if (softdep_check_suspend(mp,
1894 secondary_accwrites) != 0) {
1896 goto loop; /* More work needed */
1898 mtx_assert(MNT_MTX(mp), MA_OWNED);
1899 mp->mnt_kern_flag |= MNTK_SUSPEND2 | MNTK_SUSPENDED;
1905 * Write back modified superblock.
1907 if (fs->fs_fmod != 0 &&
1908 (error = ffs_sbupdate(ump, waitfor, suspended)) != 0)
1910 if (ffs_fsfail_cleanup(ump, allerror))
1916 ffs_vget(mp, ino, flags, vpp)
1922 return (ffs_vgetf(mp, ino, flags, vpp, 0));
1926 ffs_vgetf(mp, ino, flags, vpp, ffs_flags)
1935 struct ufsmount *ump;
1941 MPASS((ffs_flags & FFSV_REPLACE) == 0 || (flags & LK_EXCLUSIVE) != 0);
1943 error = vfs_hash_get(mp, ino, flags, curthread, vpp, NULL, NULL);
1947 if ((ffs_flags & FFSV_REPLACE) == 0)
1954 * We must promote to an exclusive lock for vnode creation. This
1955 * can happen if lookup is passed LOCKSHARED.
1957 if ((flags & LK_TYPE_MASK) == LK_SHARED) {
1958 flags &= ~LK_TYPE_MASK;
1959 flags |= LK_EXCLUSIVE;
1963 * We do not lock vnode creation as it is believed to be too
1964 * expensive for such rare case as simultaneous creation of vnode
1965 * for same ino by different processes. We just allow them to race
1966 * and check later to decide who wins. Let the race begin!
1971 ip = uma_zalloc(uma_inode, M_WAITOK | M_ZERO);
1973 /* Allocate a new vnode/inode. */
1974 error = getnewvnode("ufs", mp, fs->fs_magic == FS_UFS1_MAGIC ?
1975 &ffs_vnodeops1 : &ffs_vnodeops2, &vp);
1978 uma_zfree(uma_inode, ip);
1982 * FFS supports recursive locking.
1984 lockmgr(vp->v_vnlock, LK_EXCLUSIVE, NULL);
1987 vp->v_bufobj.bo_bsize = fs->fs_bsize;
1992 ip->i_nextclustercg = -1;
1993 ip->i_flag = fs->fs_magic == FS_UFS1_MAGIC ? 0 : IN_UFS2;
1994 ip->i_mode = 0; /* ensure error cases below throw away vnode */
1998 for (i = 0; i < MAXQUOTAS; i++)
1999 ip->i_dquot[i] = NODQUOT;
2003 if (ffs_flags & FFSV_FORCEINSMQ)
2004 vp->v_vflag |= VV_FORCEINSMQ;
2005 error = insmntque(vp, mp);
2007 uma_zfree(uma_inode, ip);
2011 vp->v_vflag &= ~VV_FORCEINSMQ;
2012 error = vfs_hash_insert(vp, ino, flags, curthread, vpp, NULL, NULL);
2017 * Calls from ffs_valloc() (i.e. FFSV_REPLACE set)
2018 * operate on empty inode, which must not be found by
2019 * other threads until fully filled. Vnode for empty
2020 * inode must be not re-inserted on the hash by other
2021 * thread, after removal by us at the beginning.
2023 MPASS((ffs_flags & FFSV_REPLACE) == 0);
2027 /* Read in the disk contents for the inode, copy into the inode. */
2028 dbn = fsbtodb(fs, ino_to_fsba(fs, ino));
2029 error = ffs_breadz(ump, ump->um_devvp, dbn, dbn, (int)fs->fs_bsize,
2030 NULL, NULL, 0, NOCRED, 0, NULL, &bp);
2033 * The inode does not contain anything useful, so it would
2034 * be misleading to leave it on its hash chain. With mode
2035 * still zero, it will be unlinked and returned to the free
2044 ip->i_din1 = uma_zalloc(uma_ufs1, M_WAITOK);
2046 ip->i_din2 = uma_zalloc(uma_ufs2, M_WAITOK);
2047 if ((error = ffs_load_inode(bp, ip, fs, ino)) != 0) {
2054 if (DOINGSOFTDEP(vp))
2055 softdep_load_inodeblock(ip);
2057 ip->i_effnlink = ip->i_nlink;
2061 * Initialize the vnode from the inode, check for aliases.
2062 * Note that the underlying vnode may have changed.
2064 error = ufs_vinit(mp, I_IS_UFS1(ip) ? &ffs_fifoops1 : &ffs_fifoops2,
2074 * Finish inode initialization.
2076 if (vp->v_type != VFIFO) {
2077 /* FFS supports shared locking for all files except fifos. */
2082 * Set up a generation number for this inode if it does not
2083 * already have one. This should only happen on old filesystems.
2085 if (ip->i_gen == 0) {
2086 while (ip->i_gen == 0)
2087 ip->i_gen = arc4random();
2088 if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
2089 UFS_INODE_SET_FLAG(ip, IN_MODIFIED);
2090 DIP_SET(ip, i_gen, ip->i_gen);
2094 if ((mp->mnt_flag & MNT_MULTILABEL) && ip->i_mode) {
2096 * If this vnode is already allocated, and we're running
2097 * multi-label, attempt to perform a label association
2098 * from the extended attributes on the inode.
2100 error = mac_vnode_associate_extattr(mp, vp);
2102 /* ufs_inactive will release ip->i_devvp ref. */
2116 * File handle to vnode
2118 * Have to be really careful about stale file handles:
2119 * - check that the inode number is valid
2120 * - for UFS2 check that the inode number is initialized
2121 * - call ffs_vget() to get the locked inode
2122 * - check for an unallocated inode (i_mode == 0)
2123 * - check that the given client host has export rights and return
2124 * those rights via. exflagsp and credanonp
2127 ffs_fhtovp(mp, fhp, flags, vpp)
2134 struct ufsmount *ump;
2142 ufhp = (struct ufid *)fhp;
2143 ino = ufhp->ufid_ino;
2146 if (ino < UFS_ROOTINO || ino >= fs->fs_ncg * fs->fs_ipg)
2149 * Need to check if inode is initialized because UFS2 does lazy
2150 * initialization and nfs_fhtovp can offer arbitrary inode numbers.
2152 if (fs->fs_magic != FS_UFS2_MAGIC)
2153 return (ufs_fhtovp(mp, ufhp, flags, vpp));
2154 cg = ino_to_cg(fs, ino);
2155 if ((error = ffs_getcg(fs, ump->um_devvp, cg, 0, &bp, &cgp)) != 0)
2157 if (ino >= cg * fs->fs_ipg + cgp->cg_initediblk) {
2162 return (ufs_fhtovp(mp, ufhp, flags, vpp));
2166 * Initialize the filesystem.
2170 struct vfsconf *vfsp;
2173 ffs_susp_initialize();
2174 softdep_initialize();
2175 return (ufs_init(vfsp));
2179 * Undo the work of ffs_init().
2183 struct vfsconf *vfsp;
2187 ret = ufs_uninit(vfsp);
2188 softdep_uninitialize();
2189 ffs_susp_uninitialize();
2190 taskqueue_drain_all(taskqueue_thread);
2195 * Structure used to pass information from ffs_sbupdate to its
2196 * helper routine ffs_use_bwrite.
2199 struct ufsmount *ump;
2207 * Write a superblock and associated information back to disk.
2210 ffs_sbupdate(ump, waitfor, suspended)
2211 struct ufsmount *ump;
2220 if (fs->fs_ronly == 1 &&
2221 (ump->um_mountp->mnt_flag & (MNT_RDONLY | MNT_UPDATE)) !=
2222 (MNT_RDONLY | MNT_UPDATE) && ump->um_fsckpid == 0)
2223 panic("ffs_sbupdate: write read-only filesystem");
2225 * We use the superblock's buf to serialize calls to ffs_sbupdate().
2227 sbbp = getblk(ump->um_devvp, btodb(fs->fs_sblockloc),
2228 (int)fs->fs_sbsize, 0, 0, 0);
2230 * Initialize info needed for write function.
2234 devfd.waitfor = waitfor;
2235 devfd.suspended = suspended;
2237 return (ffs_sbput(&devfd, fs, fs->fs_sblockloc, ffs_use_bwrite));
2241 * Write function for use by filesystem-layer routines.
2244 ffs_use_bwrite(void *devfd, off_t loc, void *buf, int size)
2246 struct devfd *devfdp;
2247 struct ufsmount *ump;
2256 * Writing the superblock summary information.
2258 if (loc != fs->fs_sblockloc) {
2259 bp = getblk(ump->um_devvp, btodb(loc), size, 0, 0, 0);
2260 bcopy(buf, bp->b_data, (u_int)size);
2261 if (devfdp->suspended)
2262 bp->b_flags |= B_VALIDSUSPWRT;
2263 if (devfdp->waitfor != MNT_WAIT)
2265 else if ((error = bwrite(bp)) != 0)
2266 devfdp->error = error;
2270 * Writing the superblock itself. We need to do special checks for it.
2273 if (ffs_fsfail_cleanup(ump, devfdp->error))
2275 if (devfdp->error != 0) {
2277 return (devfdp->error);
2279 if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_sblockloc != SBLOCK_UFS1 &&
2280 (fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) {
2281 printf("WARNING: %s: correcting fs_sblockloc from %jd to %d\n",
2282 fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS1);
2283 fs->fs_sblockloc = SBLOCK_UFS1;
2285 if (fs->fs_magic == FS_UFS2_MAGIC && fs->fs_sblockloc != SBLOCK_UFS2 &&
2286 (fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) {
2287 printf("WARNING: %s: correcting fs_sblockloc from %jd to %d\n",
2288 fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS2);
2289 fs->fs_sblockloc = SBLOCK_UFS2;
2291 if (MOUNTEDSOFTDEP(ump->um_mountp))
2292 softdep_setup_sbupdate(ump, (struct fs *)bp->b_data, bp);
2293 bcopy((caddr_t)fs, bp->b_data, (u_int)fs->fs_sbsize);
2294 fs = (struct fs *)bp->b_data;
2295 ffs_oldfscompat_write(fs, ump);
2296 /* Recalculate the superblock hash */
2297 fs->fs_ckhash = ffs_calc_sbhash(fs);
2298 if (devfdp->suspended)
2299 bp->b_flags |= B_VALIDSUSPWRT;
2300 if (devfdp->waitfor != MNT_WAIT)
2302 else if ((error = bwrite(bp)) != 0)
2303 devfdp->error = error;
2304 return (devfdp->error);
2308 ffs_extattrctl(struct mount *mp, int cmd, struct vnode *filename_vp,
2309 int attrnamespace, const char *attrname)
2313 return (ufs_extattrctl(mp, cmd, filename_vp, attrnamespace,
2316 return (vfs_stdextattrctl(mp, cmd, filename_vp, attrnamespace,
2322 ffs_ifree(struct ufsmount *ump, struct inode *ip)
2325 if (ump->um_fstype == UFS1 && ip->i_din1 != NULL)
2326 uma_zfree(uma_ufs1, ip->i_din1);
2327 else if (ip->i_din2 != NULL)
2328 uma_zfree(uma_ufs2, ip->i_din2);
2329 uma_zfree(uma_inode, ip);
2332 static int dobkgrdwrite = 1;
2333 SYSCTL_INT(_debug, OID_AUTO, dobkgrdwrite, CTLFLAG_RW, &dobkgrdwrite, 0,
2334 "Do background writes (honoring the BV_BKGRDWRITE flag)?");
2337 * Complete a background write started from bwrite.
2340 ffs_backgroundwritedone(struct buf *bp)
2342 struct bufobj *bufobj;
2346 if (!LIST_EMPTY(&bp->b_dep) && (bp->b_ioflags & BIO_ERROR) != 0)
2347 softdep_handle_error(bp);
2351 * Find the original buffer that we are writing.
2353 bufobj = bp->b_bufobj;
2355 if ((origbp = gbincore(bp->b_bufobj, bp->b_lblkno)) == NULL)
2356 panic("backgroundwritedone: lost buffer");
2359 * We should mark the cylinder group buffer origbp as
2360 * dirty, to not lose the failed write.
2362 if ((bp->b_ioflags & BIO_ERROR) != 0)
2363 origbp->b_vflags |= BV_BKGRDERR;
2366 * Process dependencies then return any unfinished ones.
2368 if (!LIST_EMPTY(&bp->b_dep) && (bp->b_ioflags & BIO_ERROR) == 0)
2371 if (!LIST_EMPTY(&bp->b_dep))
2372 softdep_move_dependencies(bp, origbp);
2375 * This buffer is marked B_NOCACHE so when it is released
2376 * by biodone it will be tossed.
2378 bp->b_flags |= B_NOCACHE;
2379 bp->b_flags &= ~B_CACHE;
2383 * Prevent brelse() from trying to keep and re-dirtying bp on
2384 * errors. It causes b_bufobj dereference in
2385 * bdirty()/reassignbuf(), and b_bufobj was cleared in
2388 if ((bp->b_ioflags & BIO_ERROR) != 0)
2389 bp->b_flags |= B_INVAL;
2393 * Clear the BV_BKGRDINPROG flag in the original buffer
2394 * and awaken it if it is waiting for the write to complete.
2395 * If BV_BKGRDINPROG is not set in the original buffer it must
2396 * have been released and re-instantiated - which is not legal.
2398 KASSERT((origbp->b_vflags & BV_BKGRDINPROG),
2399 ("backgroundwritedone: lost buffer2"));
2400 origbp->b_vflags &= ~BV_BKGRDINPROG;
2401 if (origbp->b_vflags & BV_BKGRDWAIT) {
2402 origbp->b_vflags &= ~BV_BKGRDWAIT;
2403 wakeup(&origbp->b_xflags);
2410 * Write, release buffer on completion. (Done by iodone
2411 * if async). Do not bother writing anything if the buffer
2414 * Note that we set B_CACHE here, indicating that buffer is
2415 * fully valid and thus cacheable. This is true even of NFS
2416 * now so we set it generally. This could be set either here
2417 * or in biodone() since the I/O is synchronous. We put it
2421 ffs_bufwrite(struct buf *bp)
2426 CTR3(KTR_BUF, "bufwrite(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
2427 if (bp->b_flags & B_INVAL) {
2432 if (!BUF_ISLOCKED(bp))
2433 panic("bufwrite: buffer is not busy???");
2435 * If a background write is already in progress, delay
2436 * writing this block if it is asynchronous. Otherwise
2437 * wait for the background write to complete.
2439 BO_LOCK(bp->b_bufobj);
2440 if (bp->b_vflags & BV_BKGRDINPROG) {
2441 if (bp->b_flags & B_ASYNC) {
2442 BO_UNLOCK(bp->b_bufobj);
2446 bp->b_vflags |= BV_BKGRDWAIT;
2447 msleep(&bp->b_xflags, BO_LOCKPTR(bp->b_bufobj), PRIBIO,
2449 if (bp->b_vflags & BV_BKGRDINPROG)
2450 panic("bufwrite: still writing");
2452 bp->b_vflags &= ~BV_BKGRDERR;
2453 BO_UNLOCK(bp->b_bufobj);
2456 * If this buffer is marked for background writing and we
2457 * do not have to wait for it, make a copy and write the
2458 * copy so as to leave this buffer ready for further use.
2460 * This optimization eats a lot of memory. If we have a page
2461 * or buffer shortfall we can't do it.
2463 if (dobkgrdwrite && (bp->b_xflags & BX_BKGRDWRITE) &&
2464 (bp->b_flags & B_ASYNC) &&
2465 !vm_page_count_severe() &&
2466 !buf_dirty_count_severe()) {
2467 KASSERT(bp->b_iodone == NULL,
2468 ("bufwrite: needs chained iodone (%p)", bp->b_iodone));
2470 /* get a new block */
2471 newbp = geteblk(bp->b_bufsize, GB_NOWAIT_BD);
2475 KASSERT(buf_mapped(bp), ("Unmapped cg"));
2476 memcpy(newbp->b_data, bp->b_data, bp->b_bufsize);
2477 BO_LOCK(bp->b_bufobj);
2478 bp->b_vflags |= BV_BKGRDINPROG;
2479 BO_UNLOCK(bp->b_bufobj);
2481 (bp->b_xflags & BX_FSPRIV) | BX_BKGRDMARKER;
2482 newbp->b_lblkno = bp->b_lblkno;
2483 newbp->b_blkno = bp->b_blkno;
2484 newbp->b_offset = bp->b_offset;
2485 newbp->b_iodone = ffs_backgroundwritedone;
2486 newbp->b_flags |= B_ASYNC;
2487 newbp->b_flags &= ~B_INVAL;
2488 pbgetvp(bp->b_vp, newbp);
2492 * Move over the dependencies. If there are rollbacks,
2493 * leave the parent buffer dirtied as it will need to
2496 if (LIST_EMPTY(&bp->b_dep) ||
2497 softdep_move_dependencies(bp, newbp) == 0)
2504 * Initiate write on the copy, release the original. The
2505 * BKGRDINPROG flag prevents it from going away until
2506 * the background write completes. We have to recalculate
2507 * its check hash in case the buffer gets freed and then
2508 * reconstituted from the buffer cache during a later read.
2510 if ((bp->b_xflags & BX_CYLGRP) != 0) {
2511 cgp = (struct cg *)bp->b_data;
2514 calculate_crc32c(~0L, bp->b_data, bp->b_bcount);
2519 /* Mark the buffer clean */
2523 /* Let the normal bufwrite do the rest for us */
2526 * If we are writing a cylinder group, update its time.
2528 if ((bp->b_xflags & BX_CYLGRP) != 0) {
2529 cgp = (struct cg *)bp->b_data;
2530 cgp->cg_old_time = cgp->cg_time = time_second;
2532 return (bufwrite(bp));
2537 ffs_geom_strategy(struct bufobj *bo, struct buf *bp)
2544 * This is the bufobj strategy for the private VCHR vnodes
2545 * used by FFS to access the underlying storage device.
2546 * We override the default bufobj strategy and thus bypass
2547 * VOP_STRATEGY() for these vnodes.
2550 KASSERT(bp->b_vp == NULL || bp->b_vp->v_type != VCHR ||
2551 bp->b_vp->v_rdev == NULL ||
2552 bp->b_vp->v_rdev->si_mountpt == NULL ||
2553 VFSTOUFS(bp->b_vp->v_rdev->si_mountpt) == NULL ||
2554 vp == VFSTOUFS(bp->b_vp->v_rdev->si_mountpt)->um_devvp,
2555 ("ffs_geom_strategy() with wrong vp"));
2556 if (bp->b_iocmd == BIO_WRITE) {
2557 if ((bp->b_flags & B_VALIDSUSPWRT) == 0 &&
2558 bp->b_vp != NULL && bp->b_vp->v_mount != NULL &&
2559 (bp->b_vp->v_mount->mnt_kern_flag & MNTK_SUSPENDED) != 0)
2560 panic("ffs_geom_strategy: bad I/O");
2561 nocopy = bp->b_flags & B_NOCOPY;
2562 bp->b_flags &= ~(B_VALIDSUSPWRT | B_NOCOPY);
2563 if ((vp->v_vflag & VV_COPYONWRITE) && nocopy == 0 &&
2564 vp->v_rdev->si_snapdata != NULL) {
2565 if ((bp->b_flags & B_CLUSTER) != 0) {
2566 runningbufwakeup(bp);
2567 TAILQ_FOREACH(tbp, &bp->b_cluster.cluster_head,
2568 b_cluster.cluster_entry) {
2569 error = ffs_copyonwrite(vp, tbp);
2571 error != EOPNOTSUPP) {
2572 bp->b_error = error;
2573 bp->b_ioflags |= BIO_ERROR;
2578 bp->b_runningbufspace = bp->b_bufsize;
2579 atomic_add_long(&runningbufspace,
2580 bp->b_runningbufspace);
2582 error = ffs_copyonwrite(vp, bp);
2583 if (error != 0 && error != EOPNOTSUPP) {
2584 bp->b_error = error;
2585 bp->b_ioflags |= BIO_ERROR;
2592 if ((bp->b_flags & B_CLUSTER) != 0) {
2593 TAILQ_FOREACH(tbp, &bp->b_cluster.cluster_head,
2594 b_cluster.cluster_entry) {
2595 if (!LIST_EMPTY(&tbp->b_dep))
2599 if (!LIST_EMPTY(&bp->b_dep))
2605 * Check for metadata that needs check-hashes and update them.
2607 switch (bp->b_xflags & BX_FSPRIV) {
2609 ((struct cg *)bp->b_data)->cg_ckhash = 0;
2610 ((struct cg *)bp->b_data)->cg_ckhash =
2611 calculate_crc32c(~0L, bp->b_data, bp->b_bcount);
2618 printf("Check-hash write is unimplemented!!!\n");
2625 printf("multiple buffer types 0x%b\n",
2626 (u_int)(bp->b_xflags & BX_FSPRIV),
2627 PRINT_UFS_BUF_XFLAGS);
2631 if (bp->b_iocmd != BIO_READ && ffs_enxio_enable)
2632 bp->b_xflags |= BX_CVTENXIO;
2633 g_vfs_strategy(bo, bp);
2637 ffs_own_mount(const struct mount *mp)
2640 if (mp->mnt_op == &ufs_vfsops)
2648 /* defined in ffs_softdep.c */
2649 extern void db_print_ffs(struct ufsmount *ump);
2651 DB_SHOW_COMMAND(ffs, db_show_ffs)
2654 struct ufsmount *ump;
2657 ump = VFSTOUFS((struct mount *)addr);
2662 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2663 if (!strcmp(mp->mnt_stat.f_fstypename, ufs_vfsconf.vfc_name))
2664 db_print_ffs(VFSTOUFS(mp));
2668 #endif /* SOFTUPDATES */