Home | History | Annotate | Line # | Download | only in ffs
ffs_vfsops.c revision 1.258.2.3
      1 /*	$NetBSD: ffs_vfsops.c,v 1.258.2.3 2011/03/05 20:56:30 rmind Exp $	*/
      2 
      3 /*-
      4  * Copyright (c) 2008, 2009 The NetBSD Foundation, Inc.
      5  * All rights reserved.
      6  *
      7  * This code is derived from software contributed to The NetBSD Foundation
      8  * by Wasabi Systems, Inc, and by Andrew Doran.
      9  *
     10  * Redistribution and use in source and binary forms, with or without
     11  * modification, are permitted provided that the following conditions
     12  * are met:
     13  * 1. Redistributions of source code must retain the above copyright
     14  *    notice, this list of conditions and the following disclaimer.
     15  * 2. Redistributions in binary form must reproduce the above copyright
     16  *    notice, this list of conditions and the following disclaimer in the
     17  *    documentation and/or other materials provided with the distribution.
     18  *
     19  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
     20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     21  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     22  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
     23  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     29  * POSSIBILITY OF SUCH DAMAGE.
     30  */
     31 
     32 /*
     33  * Copyright (c) 1989, 1991, 1993, 1994
     34  *	The Regents of the University of California.  All rights reserved.
     35  *
     36  * Redistribution and use in source and binary forms, with or without
     37  * modification, are permitted provided that the following conditions
     38  * are met:
     39  * 1. Redistributions of source code must retain the above copyright
     40  *    notice, this list of conditions and the following disclaimer.
     41  * 2. Redistributions in binary form must reproduce the above copyright
     42  *    notice, this list of conditions and the following disclaimer in the
     43  *    documentation and/or other materials provided with the distribution.
     44  * 3. Neither the name of the University nor the names of its contributors
     45  *    may be used to endorse or promote products derived from this software
     46  *    without specific prior written permission.
     47  *
     48  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     49  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     50  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     51  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     52  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     53  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     54  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     55  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     56  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     57  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     58  * SUCH DAMAGE.
     59  *
     60  *	@(#)ffs_vfsops.c	8.31 (Berkeley) 5/20/95
     61  */
     62 
     63 #include <sys/cdefs.h>
     64 __KERNEL_RCSID(0, "$NetBSD: ffs_vfsops.c,v 1.258.2.3 2011/03/05 20:56:30 rmind Exp $");
     65 
     66 #if defined(_KERNEL_OPT)
     67 #include "opt_ffs.h"
     68 #include "opt_quota.h"
     69 #include "opt_wapbl.h"
     70 #endif
     71 
     72 #include <sys/param.h>
     73 #include <sys/systm.h>
     74 #include <sys/namei.h>
     75 #include <sys/proc.h>
     76 #include <sys/kernel.h>
     77 #include <sys/vnode.h>
     78 #include <sys/socket.h>
     79 #include <sys/mount.h>
     80 #include <sys/buf.h>
     81 #include <sys/device.h>
     82 #include <sys/mbuf.h>
     83 #include <sys/file.h>
     84 #include <sys/disklabel.h>
     85 #include <sys/ioctl.h>
     86 #include <sys/errno.h>
     87 #include <sys/malloc.h>
     88 #include <sys/pool.h>
     89 #include <sys/lock.h>
     90 #include <sys/sysctl.h>
     91 #include <sys/conf.h>
     92 #include <sys/kauth.h>
     93 #include <sys/wapbl.h>
     94 #include <sys/fstrans.h>
     95 #include <sys/module.h>
     96 
     97 #include <miscfs/genfs/genfs.h>
     98 #include <miscfs/specfs/specdev.h>
     99 
    100 #include <ufs/ufs/quota.h>
    101 #include <ufs/ufs/ufsmount.h>
    102 #include <ufs/ufs/inode.h>
    103 #include <ufs/ufs/dir.h>
    104 #include <ufs/ufs/ufs_extern.h>
    105 #include <ufs/ufs/ufs_bswap.h>
    106 #include <ufs/ufs/ufs_wapbl.h>
    107 
    108 #include <ufs/ffs/fs.h>
    109 #include <ufs/ffs/ffs_extern.h>
    110 
    111 MODULE(MODULE_CLASS_VFS, ffs, NULL);
    112 
    113 static int	ffs_vfs_fsync(vnode_t *, int);
    114 
    115 static struct sysctllog *ffs_sysctl_log;
    116 
    117 /* how many times ffs_init() was called */
    118 int ffs_initcount = 0;
    119 
    120 extern const struct vnodeopv_desc ffs_vnodeop_opv_desc;
    121 extern const struct vnodeopv_desc ffs_specop_opv_desc;
    122 extern const struct vnodeopv_desc ffs_fifoop_opv_desc;
    123 
    124 const struct vnodeopv_desc * const ffs_vnodeopv_descs[] = {
    125 	&ffs_vnodeop_opv_desc,
    126 	&ffs_specop_opv_desc,
    127 	&ffs_fifoop_opv_desc,
    128 	NULL,
    129 };
    130 
    131 struct vfsops ffs_vfsops = {
    132 	MOUNT_FFS,
    133 	sizeof (struct ufs_args),
    134 	ffs_mount,
    135 	ufs_start,
    136 	ffs_unmount,
    137 	ufs_root,
    138 	ufs_quotactl,
    139 	ffs_statvfs,
    140 	ffs_sync,
    141 	ffs_vget,
    142 	ffs_fhtovp,
    143 	ffs_vptofh,
    144 	ffs_init,
    145 	ffs_reinit,
    146 	ffs_done,
    147 	ffs_mountroot,
    148 	ffs_snapshot,
    149 	ffs_extattrctl,
    150 	ffs_suspendctl,
    151 	genfs_renamelock_enter,
    152 	genfs_renamelock_exit,
    153 	ffs_vfs_fsync,
    154 	ffs_vnodeopv_descs,
    155 	0,
    156 	{ NULL, NULL },
    157 };
    158 
    159 static const struct genfs_ops ffs_genfsops = {
    160 	.gop_size = ffs_gop_size,
    161 	.gop_alloc = ufs_gop_alloc,
    162 	.gop_write = genfs_gop_write,
    163 	.gop_markupdate = ufs_gop_markupdate,
    164 };
    165 
    166 static const struct ufs_ops ffs_ufsops = {
    167 	.uo_itimes = ffs_itimes,
    168 	.uo_update = ffs_update,
    169 	.uo_truncate = ffs_truncate,
    170 	.uo_valloc = ffs_valloc,
    171 	.uo_vfree = ffs_vfree,
    172 	.uo_balloc = ffs_balloc,
    173 	.uo_unmark_vnode = (void (*)(vnode_t *))nullop,
    174 };
    175 
    176 static int
    177 ffs_modcmd(modcmd_t cmd, void *arg)
    178 {
    179 	int error;
    180 
    181 #if 0
    182 	extern int doasyncfree;
    183 #endif
    184 	extern int ffs_log_changeopt;
    185 
    186 	switch (cmd) {
    187 	case MODULE_CMD_INIT:
    188 		error = vfs_attach(&ffs_vfsops);
    189 		if (error != 0)
    190 			break;
    191 
    192 		sysctl_createv(&ffs_sysctl_log, 0, NULL, NULL,
    193 			       CTLFLAG_PERMANENT,
    194 			       CTLTYPE_NODE, "vfs", NULL,
    195 			       NULL, 0, NULL, 0,
    196 			       CTL_VFS, CTL_EOL);
    197 		sysctl_createv(&ffs_sysctl_log, 0, NULL, NULL,
    198 			       CTLFLAG_PERMANENT,
    199 			       CTLTYPE_NODE, "ffs",
    200 			       SYSCTL_DESCR("Berkeley Fast File System"),
    201 			       NULL, 0, NULL, 0,
    202 			       CTL_VFS, 1, CTL_EOL);
    203 
    204 		/*
    205 		 * @@@ should we even bother with these first three?
    206 		 */
    207 		sysctl_createv(&ffs_sysctl_log, 0, NULL, NULL,
    208 			       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
    209 			       CTLTYPE_INT, "doclusterread", NULL,
    210 			       sysctl_notavail, 0, NULL, 0,
    211 			       CTL_VFS, 1, FFS_CLUSTERREAD, CTL_EOL);
    212 		sysctl_createv(&ffs_sysctl_log, 0, NULL, NULL,
    213 			       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
    214 			       CTLTYPE_INT, "doclusterwrite", NULL,
    215 			       sysctl_notavail, 0, NULL, 0,
    216 			       CTL_VFS, 1, FFS_CLUSTERWRITE, CTL_EOL);
    217 		sysctl_createv(&ffs_sysctl_log, 0, NULL, NULL,
    218 			       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
    219 			       CTLTYPE_INT, "doreallocblks", NULL,
    220 			       sysctl_notavail, 0, NULL, 0,
    221 			       CTL_VFS, 1, FFS_REALLOCBLKS, CTL_EOL);
    222 #if 0
    223 		sysctl_createv(&ffs_sysctl_log, 0, NULL, NULL,
    224 			       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
    225 			       CTLTYPE_INT, "doasyncfree",
    226 			       SYSCTL_DESCR("Release dirty blocks asynchronously"),
    227 			       NULL, 0, &doasyncfree, 0,
    228 			       CTL_VFS, 1, FFS_ASYNCFREE, CTL_EOL);
    229 #endif
    230 		sysctl_createv(&ffs_sysctl_log, 0, NULL, NULL,
    231 			       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
    232 			       CTLTYPE_INT, "log_changeopt",
    233 			       SYSCTL_DESCR("Log changes in optimization strategy"),
    234 			       NULL, 0, &ffs_log_changeopt, 0,
    235 			       CTL_VFS, 1, FFS_LOG_CHANGEOPT, CTL_EOL);
    236 		break;
    237 	case MODULE_CMD_FINI:
    238 		error = vfs_detach(&ffs_vfsops);
    239 		if (error != 0)
    240 			break;
    241 		sysctl_teardown(&ffs_sysctl_log);
    242 		break;
    243 	default:
    244 		error = ENOTTY;
    245 		break;
    246 	}
    247 
    248 	return (error);
    249 }
    250 
    251 pool_cache_t ffs_inode_cache;
    252 pool_cache_t ffs_dinode1_cache;
    253 pool_cache_t ffs_dinode2_cache;
    254 
    255 static void ffs_oldfscompat_read(struct fs *, struct ufsmount *, daddr_t);
    256 static void ffs_oldfscompat_write(struct fs *, struct ufsmount *);
    257 
    258 /*
    259  * Called by main() when ffs is going to be mounted as root.
    260  */
    261 
    262 int
    263 ffs_mountroot(void)
    264 {
    265 	struct fs *fs;
    266 	struct mount *mp;
    267 	struct lwp *l = curlwp;			/* XXX */
    268 	struct ufsmount *ump;
    269 	int error;
    270 
    271 	if (device_class(root_device) != DV_DISK)
    272 		return (ENODEV);
    273 
    274 	if ((error = vfs_rootmountalloc(MOUNT_FFS, "root_device", &mp))) {
    275 		vrele(rootvp);
    276 		return (error);
    277 	}
    278 
    279 	/*
    280 	 * We always need to be able to mount the root file system.
    281 	 */
    282 	mp->mnt_flag |= MNT_FORCE;
    283 	if ((error = ffs_mountfs(rootvp, mp, l)) != 0) {
    284 		vfs_unbusy(mp, false, NULL);
    285 		vfs_destroy(mp);
    286 		return (error);
    287 	}
    288 	mp->mnt_flag &= ~MNT_FORCE;
    289 	mutex_enter(&mountlist_lock);
    290 	CIRCLEQ_INSERT_TAIL(&mountlist, mp, mnt_list);
    291 	mutex_exit(&mountlist_lock);
    292 	ump = VFSTOUFS(mp);
    293 	fs = ump->um_fs;
    294 	memset(fs->fs_fsmnt, 0, sizeof(fs->fs_fsmnt));
    295 	(void)copystr(mp->mnt_stat.f_mntonname, fs->fs_fsmnt, MNAMELEN - 1, 0);
    296 	(void)ffs_statvfs(mp, &mp->mnt_stat);
    297 	vfs_unbusy(mp, false, NULL);
    298 	setrootfstime((time_t)fs->fs_time);
    299 	return (0);
    300 }
    301 
    302 /*
    303  * VFS Operations.
    304  *
    305  * mount system call
    306  */
    307 int
    308 ffs_mount(struct mount *mp, const char *path, void *data, size_t *data_len)
    309 {
    310 	struct lwp *l = curlwp;
    311 	struct vnode *devvp = NULL;
    312 	struct ufs_args *args = data;
    313 	struct ufsmount *ump = NULL;
    314 	struct fs *fs;
    315 	int error = 0, flags, update;
    316 	mode_t accessmode;
    317 
    318 	if (*data_len < sizeof *args)
    319 		return EINVAL;
    320 
    321 	if (mp->mnt_flag & MNT_GETARGS) {
    322 		ump = VFSTOUFS(mp);
    323 		if (ump == NULL)
    324 			return EIO;
    325 		args->fspec = NULL;
    326 		*data_len = sizeof *args;
    327 		return 0;
    328 	}
    329 
    330 	update = mp->mnt_flag & MNT_UPDATE;
    331 
    332 	/* Check arguments */
    333 	if (args->fspec != NULL) {
    334 		/*
    335 		 * Look up the name and verify that it's sane.
    336 		 */
    337 		error = namei_simple_user(args->fspec,
    338 					NSM_FOLLOW_NOEMULROOT, &devvp);
    339 		if (error != 0)
    340 			return (error);
    341 
    342 		if (!update) {
    343 			/*
    344 			 * Be sure this is a valid block device
    345 			 */
    346 			if (devvp->v_type != VBLK)
    347 				error = ENOTBLK;
    348 			else if (bdevsw_lookup(devvp->v_rdev) == NULL)
    349 				error = ENXIO;
    350 		} else {
    351 			/*
    352 			 * Be sure we're still naming the same device
    353 			 * used for our initial mount
    354 			 */
    355 			ump = VFSTOUFS(mp);
    356 			if (devvp != ump->um_devvp) {
    357 				if (devvp->v_rdev != ump->um_devvp->v_rdev)
    358 					error = EINVAL;
    359 				else {
    360 					vrele(devvp);
    361 					devvp = ump->um_devvp;
    362 					vref(devvp);
    363 				}
    364 			}
    365 		}
    366 	} else {
    367 		if (!update) {
    368 			/* New mounts must have a filename for the device */
    369 			return (EINVAL);
    370 		} else {
    371 			/* Use the extant mount */
    372 			ump = VFSTOUFS(mp);
    373 			devvp = ump->um_devvp;
    374 			vref(devvp);
    375 		}
    376 	}
    377 
    378 	/*
    379 	 * If mount by non-root, then verify that user has necessary
    380 	 * permissions on the device.
    381 	 *
    382 	 * Permission to update a mount is checked higher, so here we presume
    383 	 * updating the mount is okay (for example, as far as securelevel goes)
    384 	 * which leaves us with the normal check.
    385 	 */
    386 	if (error == 0) {
    387 		accessmode = VREAD;
    388 		if (update ?
    389 		    (mp->mnt_iflag & IMNT_WANTRDWR) != 0 :
    390 		    (mp->mnt_flag & MNT_RDONLY) == 0)
    391 			accessmode |= VWRITE;
    392 		vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
    393 		error = genfs_can_mount(devvp, accessmode, l->l_cred);
    394 		VOP_UNLOCK(devvp);
    395 	}
    396 
    397 	if (error) {
    398 		vrele(devvp);
    399 		return (error);
    400 	}
    401 
    402 #ifdef WAPBL
    403 	/* WAPBL can only be enabled on a r/w mount. */
    404 	if ((mp->mnt_flag & MNT_RDONLY) && !(mp->mnt_iflag & IMNT_WANTRDWR)) {
    405 		mp->mnt_flag &= ~MNT_LOG;
    406 	}
    407 #else /* !WAPBL */
    408 	mp->mnt_flag &= ~MNT_LOG;
    409 #endif /* !WAPBL */
    410 
    411 	if (!update) {
    412 		int xflags;
    413 
    414 		if (mp->mnt_flag & MNT_RDONLY)
    415 			xflags = FREAD;
    416 		else
    417 			xflags = FREAD | FWRITE;
    418 		error = VOP_OPEN(devvp, xflags, FSCRED);
    419 		if (error)
    420 			goto fail;
    421 		error = ffs_mountfs(devvp, mp, l);
    422 		if (error) {
    423 			vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
    424 			(void)VOP_CLOSE(devvp, xflags, NOCRED);
    425 			VOP_UNLOCK(devvp);
    426 			goto fail;
    427 		}
    428 
    429 		ump = VFSTOUFS(mp);
    430 		fs = ump->um_fs;
    431 	} else {
    432 		/*
    433 		 * Update the mount.
    434 		 */
    435 
    436 		/*
    437 		 * The initial mount got a reference on this
    438 		 * device, so drop the one obtained via
    439 		 * namei(), above.
    440 		 */
    441 		vrele(devvp);
    442 
    443 		ump = VFSTOUFS(mp);
    444 		fs = ump->um_fs;
    445 		if (fs->fs_ronly == 0 && (mp->mnt_flag & MNT_RDONLY)) {
    446 			/*
    447 			 * Changing from r/w to r/o
    448 			 */
    449 			flags = WRITECLOSE;
    450 			if (mp->mnt_flag & MNT_FORCE)
    451 				flags |= FORCECLOSE;
    452 			error = ffs_flushfiles(mp, flags, l);
    453 			if (error == 0)
    454 				error = UFS_WAPBL_BEGIN(mp);
    455 			if (error == 0 &&
    456 			    ffs_cgupdate(ump, MNT_WAIT) == 0 &&
    457 			    fs->fs_clean & FS_WASCLEAN) {
    458 				if (mp->mnt_flag & MNT_SOFTDEP)
    459 					fs->fs_flags &= ~FS_DOSOFTDEP;
    460 				fs->fs_clean = FS_ISCLEAN;
    461 				(void) ffs_sbupdate(ump, MNT_WAIT);
    462 			}
    463 			if (error == 0)
    464 				UFS_WAPBL_END(mp);
    465 			if (error)
    466 				return (error);
    467 		}
    468 
    469 #ifdef WAPBL
    470 		if ((mp->mnt_flag & MNT_LOG) == 0) {
    471 			error = ffs_wapbl_stop(mp, mp->mnt_flag & MNT_FORCE);
    472 			if (error)
    473 				return error;
    474 		}
    475 #endif /* WAPBL */
    476 
    477 		if (fs->fs_ronly == 0 && (mp->mnt_flag & MNT_RDONLY)) {
    478 			/*
    479 			 * Finish change from r/w to r/o
    480 			 */
    481 			fs->fs_ronly = 1;
    482 			fs->fs_fmod = 0;
    483 		}
    484 
    485 		if (mp->mnt_flag & MNT_RELOAD) {
    486 			error = ffs_reload(mp, l->l_cred, l);
    487 			if (error)
    488 				return (error);
    489 		}
    490 
    491 		if (fs->fs_ronly && (mp->mnt_iflag & IMNT_WANTRDWR)) {
    492 			/*
    493 			 * Changing from read-only to read/write
    494 			 */
    495 			fs->fs_ronly = 0;
    496 			fs->fs_clean <<= 1;
    497 			fs->fs_fmod = 1;
    498 #ifdef WAPBL
    499 			if (fs->fs_flags & FS_DOWAPBL) {
    500 				printf("%s: replaying log to disk\n",
    501 				    fs->fs_fsmnt);
    502 				KDASSERT(mp->mnt_wapbl_replay);
    503 				error = wapbl_replay_write(mp->mnt_wapbl_replay,
    504 							   devvp);
    505 				if (error) {
    506 					return error;
    507 				}
    508 				wapbl_replay_stop(mp->mnt_wapbl_replay);
    509 				fs->fs_clean = FS_WASCLEAN;
    510 			}
    511 #endif /* WAPBL */
    512 			if (fs->fs_snapinum[0] != 0)
    513 				ffs_snapshot_mount(mp);
    514 		}
    515 
    516 #ifdef WAPBL
    517 		error = ffs_wapbl_start(mp);
    518 		if (error)
    519 			return error;
    520 #endif /* WAPBL */
    521 
    522 		if (args->fspec == NULL)
    523 			return 0;
    524 	}
    525 
    526 	error = set_statvfs_info(path, UIO_USERSPACE, args->fspec,
    527 	    UIO_USERSPACE, mp->mnt_op->vfs_name, mp, l);
    528 	if (error == 0)
    529 		(void)strncpy(fs->fs_fsmnt, mp->mnt_stat.f_mntonname,
    530 		    sizeof(fs->fs_fsmnt));
    531 	fs->fs_flags &= ~FS_DOSOFTDEP;
    532 	if (fs->fs_fmod != 0) {	/* XXX */
    533 		int err;
    534 
    535 		fs->fs_fmod = 0;
    536 		if (fs->fs_clean & FS_WASCLEAN)
    537 			fs->fs_time = time_second;
    538 		else {
    539 			printf("%s: file system not clean (fs_clean=%#x); "
    540 			    "please fsck(8)\n", mp->mnt_stat.f_mntfromname,
    541 			    fs->fs_clean);
    542 			printf("%s: lost blocks %" PRId64 " files %d\n",
    543 			    mp->mnt_stat.f_mntfromname, fs->fs_pendingblocks,
    544 			    fs->fs_pendinginodes);
    545 		}
    546 		err = UFS_WAPBL_BEGIN(mp);
    547 		if (err == 0) {
    548 			(void) ffs_cgupdate(ump, MNT_WAIT);
    549 			UFS_WAPBL_END(mp);
    550 		}
    551 	}
    552 	if ((mp->mnt_flag & MNT_SOFTDEP) != 0) {
    553 		printf("%s: `-o softdep' is no longer supported, "
    554 		    "consider `-o log'\n", mp->mnt_stat.f_mntfromname);
    555 		mp->mnt_flag &= ~MNT_SOFTDEP;
    556 	}
    557 
    558 	return (error);
    559 
    560 fail:
    561 	vrele(devvp);
    562 	return (error);
    563 }
    564 
    565 /*
    566  * Reload all incore data for a filesystem (used after running fsck on
    567  * the root filesystem and finding things to fix). The filesystem must
    568  * be mounted read-only.
    569  *
    570  * Things to do to update the mount:
    571  *	1) invalidate all cached meta-data.
    572  *	2) re-read superblock from disk.
    573  *	3) re-read summary information from disk.
    574  *	4) invalidate all inactive vnodes.
    575  *	5) invalidate all cached file data.
    576  *	6) re-read inode data for all active vnodes.
    577  */
    578 int
    579 ffs_reload(struct mount *mp, kauth_cred_t cred, struct lwp *l)
    580 {
    581 	struct vnode *vp, *mvp, *devvp;
    582 	struct inode *ip;
    583 	void *space;
    584 	struct buf *bp;
    585 	struct fs *fs, *newfs;
    586 	struct partinfo dpart;
    587 	int i, bsize, blks, error;
    588 	int32_t *lp;
    589 	struct ufsmount *ump;
    590 	daddr_t sblockloc;
    591 
    592 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
    593 		return (EINVAL);
    594 
    595 	ump = VFSTOUFS(mp);
    596 	/*
    597 	 * Step 1: invalidate all cached meta-data.
    598 	 */
    599 	devvp = ump->um_devvp;
    600 	vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
    601 	error = vinvalbuf(devvp, 0, cred, l, 0, 0);
    602 	VOP_UNLOCK(devvp);
    603 	if (error)
    604 		panic("ffs_reload: dirty1");
    605 	/*
    606 	 * Step 2: re-read superblock from disk.
    607 	 */
    608 	fs = ump->um_fs;
    609 
    610 	/* XXX we don't handle possibility that superblock moved. */
    611 	error = bread(devvp, fs->fs_sblockloc / DEV_BSIZE, fs->fs_sbsize,
    612 		      NOCRED, 0, &bp);
    613 	if (error) {
    614 		brelse(bp, 0);
    615 		return (error);
    616 	}
    617 	newfs = malloc(fs->fs_sbsize, M_UFSMNT, M_WAITOK);
    618 	memcpy(newfs, bp->b_data, fs->fs_sbsize);
    619 #ifdef FFS_EI
    620 	if (ump->um_flags & UFS_NEEDSWAP) {
    621 		ffs_sb_swap((struct fs*)bp->b_data, newfs);
    622 		fs->fs_flags |= FS_SWAPPED;
    623 	} else
    624 #endif
    625 		fs->fs_flags &= ~FS_SWAPPED;
    626 	if ((newfs->fs_magic != FS_UFS1_MAGIC &&
    627 	     newfs->fs_magic != FS_UFS2_MAGIC)||
    628 	     newfs->fs_bsize > MAXBSIZE ||
    629 	     newfs->fs_bsize < sizeof(struct fs)) {
    630 		brelse(bp, 0);
    631 		free(newfs, M_UFSMNT);
    632 		return (EIO);		/* XXX needs translation */
    633 	}
    634 	/* Store off old fs_sblockloc for fs_oldfscompat_read. */
    635 	sblockloc = fs->fs_sblockloc;
    636 	/*
    637 	 * Copy pointer fields back into superblock before copying in	XXX
    638 	 * new superblock. These should really be in the ufsmount.	XXX
    639 	 * Note that important parameters (eg fs_ncg) are unchanged.
    640 	 */
    641 	newfs->fs_csp = fs->fs_csp;
    642 	newfs->fs_maxcluster = fs->fs_maxcluster;
    643 	newfs->fs_contigdirs = fs->fs_contigdirs;
    644 	newfs->fs_ronly = fs->fs_ronly;
    645 	newfs->fs_active = fs->fs_active;
    646 	memcpy(fs, newfs, (u_int)fs->fs_sbsize);
    647 	brelse(bp, 0);
    648 	free(newfs, M_UFSMNT);
    649 
    650 	/* Recheck for apple UFS filesystem */
    651 	ump->um_flags &= ~UFS_ISAPPLEUFS;
    652 	/* First check to see if this is tagged as an Apple UFS filesystem
    653 	 * in the disklabel
    654 	 */
    655 	if ((VOP_IOCTL(devvp, DIOCGPART, &dpart, FREAD, cred) == 0) &&
    656 		(dpart.part->p_fstype == FS_APPLEUFS)) {
    657 		ump->um_flags |= UFS_ISAPPLEUFS;
    658 	}
    659 #ifdef APPLE_UFS
    660 	else {
    661 		/* Manually look for an apple ufs label, and if a valid one
    662 		 * is found, then treat it like an Apple UFS filesystem anyway
    663 		 */
    664 		error = bread(devvp, (daddr_t)(APPLEUFS_LABEL_OFFSET / DEV_BSIZE),
    665 			APPLEUFS_LABEL_SIZE, cred, 0, &bp);
    666 		if (error) {
    667 			brelse(bp, 0);
    668 			return (error);
    669 		}
    670 		error = ffs_appleufs_validate(fs->fs_fsmnt,
    671 			(struct appleufslabel *)bp->b_data, NULL);
    672 		if (error == 0)
    673 			ump->um_flags |= UFS_ISAPPLEUFS;
    674 		brelse(bp, 0);
    675 		bp = NULL;
    676 	}
    677 #else
    678 	if (ump->um_flags & UFS_ISAPPLEUFS)
    679 		return (EIO);
    680 #endif
    681 
    682 	if (UFS_MPISAPPLEUFS(ump)) {
    683 		/* see comment about NeXT below */
    684 		ump->um_maxsymlinklen = APPLEUFS_MAXSYMLINKLEN;
    685 		ump->um_dirblksiz = APPLEUFS_DIRBLKSIZ;
    686 		mp->mnt_iflag |= IMNT_DTYPE;
    687 	} else {
    688 		ump->um_maxsymlinklen = fs->fs_maxsymlinklen;
    689 		ump->um_dirblksiz = DIRBLKSIZ;
    690 		if (ump->um_maxsymlinklen > 0)
    691 			mp->mnt_iflag |= IMNT_DTYPE;
    692 		else
    693 			mp->mnt_iflag &= ~IMNT_DTYPE;
    694 	}
    695 	ffs_oldfscompat_read(fs, ump, sblockloc);
    696 
    697 	mutex_enter(&ump->um_lock);
    698 	ump->um_maxfilesize = fs->fs_maxfilesize;
    699 	if (fs->fs_flags & ~(FS_KNOWN_FLAGS | FS_INTERNAL)) {
    700 		uprintf("%s: unknown ufs flags: 0x%08"PRIx32"%s\n",
    701 		    mp->mnt_stat.f_mntonname, fs->fs_flags,
    702 		    (mp->mnt_flag & MNT_FORCE) ? "" : ", not mounting");
    703 		if ((mp->mnt_flag & MNT_FORCE) == 0) {
    704 			mutex_exit(&ump->um_lock);
    705 			return (EINVAL);
    706 		}
    707 	}
    708 	if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
    709 		fs->fs_pendingblocks = 0;
    710 		fs->fs_pendinginodes = 0;
    711 	}
    712 	mutex_exit(&ump->um_lock);
    713 
    714 	ffs_statvfs(mp, &mp->mnt_stat);
    715 	/*
    716 	 * Step 3: re-read summary information from disk.
    717 	 */
    718 	blks = howmany(fs->fs_cssize, fs->fs_fsize);
    719 	space = fs->fs_csp;
    720 	for (i = 0; i < blks; i += fs->fs_frag) {
    721 		bsize = fs->fs_bsize;
    722 		if (i + fs->fs_frag > blks)
    723 			bsize = (blks - i) * fs->fs_fsize;
    724 		error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), bsize,
    725 			      NOCRED, 0, &bp);
    726 		if (error) {
    727 			brelse(bp, 0);
    728 			return (error);
    729 		}
    730 #ifdef FFS_EI
    731 		if (UFS_FSNEEDSWAP(fs))
    732 			ffs_csum_swap((struct csum *)bp->b_data,
    733 			    (struct csum *)space, bsize);
    734 		else
    735 #endif
    736 			memcpy(space, bp->b_data, (size_t)bsize);
    737 		space = (char *)space + bsize;
    738 		brelse(bp, 0);
    739 	}
    740 	if (fs->fs_snapinum[0] != 0)
    741 		ffs_snapshot_mount(mp);
    742 	/*
    743 	 * We no longer know anything about clusters per cylinder group.
    744 	 */
    745 	if (fs->fs_contigsumsize > 0) {
    746 		lp = fs->fs_maxcluster;
    747 		for (i = 0; i < fs->fs_ncg; i++)
    748 			*lp++ = fs->fs_contigsumsize;
    749 	}
    750 
    751 	/* Allocate a marker vnode. */
    752 	if ((mvp = vnalloc(mp)) == NULL)
    753 		return ENOMEM;
    754 	/*
    755 	 * NOTE: not using the TAILQ_FOREACH here since in this loop vgone()
    756 	 * and vclean() can be called indirectly
    757 	 */
    758 	mutex_enter(&mntvnode_lock);
    759  loop:
    760 	for (vp = TAILQ_FIRST(&mp->mnt_vnodelist); vp; vp = vunmark(mvp)) {
    761 		vmark(mvp, vp);
    762 		if (vp->v_mount != mp || vismarker(vp))
    763 			continue;
    764 		/*
    765 		 * Step 4: invalidate all inactive vnodes.
    766 		 */
    767 		if (vrecycle(vp, &mntvnode_lock, l)) {
    768 			mutex_enter(&mntvnode_lock);
    769 			(void)vunmark(mvp);
    770 			goto loop;
    771 		}
    772 		/*
    773 		 * Step 5: invalidate all cached file data.
    774 		 */
    775 		mutex_enter(vp->v_interlock);
    776 		mutex_exit(&mntvnode_lock);
    777 		if (vget(vp, LK_EXCLUSIVE)) {
    778 			(void)vunmark(mvp);
    779 			goto loop;
    780 		}
    781 		if (vinvalbuf(vp, 0, cred, l, 0, 0))
    782 			panic("ffs_reload: dirty2");
    783 		/*
    784 		 * Step 6: re-read inode data for all active vnodes.
    785 		 */
    786 		ip = VTOI(vp);
    787 		error = bread(devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
    788 			      (int)fs->fs_bsize, NOCRED, 0, &bp);
    789 		if (error) {
    790 			brelse(bp, 0);
    791 			vput(vp);
    792 			(void)vunmark(mvp);
    793 			break;
    794 		}
    795 		ffs_load_inode(bp, ip, fs, ip->i_number);
    796 		brelse(bp, 0);
    797 		vput(vp);
    798 		mutex_enter(&mntvnode_lock);
    799 	}
    800 	mutex_exit(&mntvnode_lock);
    801 	vnfree(mvp);
    802 	return (error);
    803 }
    804 
    805 /*
    806  * Possible superblock locations ordered from most to least likely.
    807  */
    808 static const int sblock_try[] = SBLOCKSEARCH;
    809 
    810 /*
    811  * Common code for mount and mountroot
    812  */
    813 int
    814 ffs_mountfs(struct vnode *devvp, struct mount *mp, struct lwp *l)
    815 {
    816 	struct ufsmount *ump;
    817 	struct buf *bp;
    818 	struct fs *fs;
    819 	dev_t dev;
    820 	struct partinfo dpart;
    821 	void *space;
    822 	daddr_t sblockloc, fsblockloc;
    823 	int blks, fstype;
    824 	int error, i, bsize, ronly, bset = 0;
    825 #ifdef FFS_EI
    826 	int needswap = 0;		/* keep gcc happy */
    827 #endif
    828 	int32_t *lp;
    829 	kauth_cred_t cred;
    830 	u_int32_t sbsize = 8192;	/* keep gcc happy*/
    831 	int32_t fsbsize;
    832 
    833 	dev = devvp->v_rdev;
    834 	cred = l ? l->l_cred : NOCRED;
    835 
    836 	/* Flush out any old buffers remaining from a previous use. */
    837 	vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
    838 	error = vinvalbuf(devvp, V_SAVE, cred, l, 0, 0);
    839 	VOP_UNLOCK(devvp);
    840 	if (error)
    841 		return (error);
    842 
    843 	ronly = (mp->mnt_flag & MNT_RDONLY) != 0;
    844 
    845 	bp = NULL;
    846 	ump = NULL;
    847 	fs = NULL;
    848 	sblockloc = 0;
    849 	fstype = 0;
    850 
    851 	error = fstrans_mount(mp);
    852 	if (error)
    853 		return error;
    854 
    855 	ump = malloc(sizeof *ump, M_UFSMNT, M_WAITOK);
    856 	memset(ump, 0, sizeof *ump);
    857 	mutex_init(&ump->um_lock, MUTEX_DEFAULT, IPL_NONE);
    858 	error = ffs_snapshot_init(ump);
    859 	if (error)
    860 		goto out;
    861 	ump->um_ops = &ffs_ufsops;
    862 
    863 #ifdef WAPBL
    864  sbagain:
    865 #endif
    866 	/*
    867 	 * Try reading the superblock in each of its possible locations.
    868 	 */
    869 	for (i = 0; ; i++) {
    870 		if (bp != NULL) {
    871 			brelse(bp, BC_NOCACHE);
    872 			bp = NULL;
    873 		}
    874 		if (sblock_try[i] == -1) {
    875 			error = EINVAL;
    876 			fs = NULL;
    877 			goto out;
    878 		}
    879 		error = bread(devvp, sblock_try[i] / DEV_BSIZE, SBLOCKSIZE, cred,
    880 			      0, &bp);
    881 		if (error) {
    882 			fs = NULL;
    883 			goto out;
    884 		}
    885 		fs = (struct fs*)bp->b_data;
    886 		fsblockloc = sblockloc = sblock_try[i];
    887 		if (fs->fs_magic == FS_UFS1_MAGIC) {
    888 			sbsize = fs->fs_sbsize;
    889 			fstype = UFS1;
    890 			fsbsize = fs->fs_bsize;
    891 #ifdef FFS_EI
    892 			needswap = 0;
    893 		} else if (fs->fs_magic == bswap32(FS_UFS1_MAGIC)) {
    894 			sbsize = bswap32(fs->fs_sbsize);
    895 			fstype = UFS1;
    896 			fsbsize = bswap32(fs->fs_bsize);
    897 			needswap = 1;
    898 #endif
    899 		} else if (fs->fs_magic == FS_UFS2_MAGIC) {
    900 			sbsize = fs->fs_sbsize;
    901 			fstype = UFS2;
    902 			fsbsize = fs->fs_bsize;
    903 #ifdef FFS_EI
    904 			needswap = 0;
    905 		} else if (fs->fs_magic == bswap32(FS_UFS2_MAGIC)) {
    906 			sbsize = bswap32(fs->fs_sbsize);
    907 			fstype = UFS2;
    908 			fsbsize = bswap32(fs->fs_bsize);
    909 			needswap = 1;
    910 #endif
    911 		} else
    912 			continue;
    913 
    914 
    915 		/* fs->fs_sblockloc isn't defined for old filesystems */
    916 		if (fstype == UFS1 && !(fs->fs_old_flags & FS_FLAGS_UPDATED)) {
    917 			if (sblockloc == SBLOCK_UFS2)
    918 				/*
    919 				 * This is likely to be the first alternate
    920 				 * in a filesystem with 64k blocks.
    921 				 * Don't use it.
    922 				 */
    923 				continue;
    924 			fsblockloc = sblockloc;
    925 		} else {
    926 			fsblockloc = fs->fs_sblockloc;
    927 #ifdef FFS_EI
    928 			if (needswap)
    929 				fsblockloc = bswap64(fsblockloc);
    930 #endif
    931 		}
    932 
    933 		/* Check we haven't found an alternate superblock */
    934 		if (fsblockloc != sblockloc)
    935 			continue;
    936 
    937 		/* Validate size of superblock */
    938 		if (sbsize > MAXBSIZE || sbsize < sizeof(struct fs))
    939 			continue;
    940 
    941 		/* Check that we can handle the file system blocksize */
    942 		if (fsbsize > MAXBSIZE) {
    943 			printf("ffs_mountfs: block size (%d) > MAXBSIZE (%d)\n",
    944 			    fsbsize, MAXBSIZE);
    945 			continue;
    946 		}
    947 
    948 		/* Ok seems to be a good superblock */
    949 		break;
    950 	}
    951 
    952 	fs = malloc((u_long)sbsize, M_UFSMNT, M_WAITOK);
    953 	memcpy(fs, bp->b_data, sbsize);
    954 	ump->um_fs = fs;
    955 
    956 #ifdef FFS_EI
    957 	if (needswap) {
    958 		ffs_sb_swap((struct fs*)bp->b_data, fs);
    959 		fs->fs_flags |= FS_SWAPPED;
    960 	} else
    961 #endif
    962 		fs->fs_flags &= ~FS_SWAPPED;
    963 
    964 #ifdef WAPBL
    965 	if ((mp->mnt_wapbl_replay == 0) && (fs->fs_flags & FS_DOWAPBL)) {
    966 		error = ffs_wapbl_replay_start(mp, fs, devvp);
    967 		if (error && (mp->mnt_flag & MNT_FORCE) == 0)
    968 			goto out;
    969 		if (!error) {
    970 			if (!ronly) {
    971 				/* XXX fsmnt may be stale. */
    972 				printf("%s: replaying log to disk\n",
    973 				    fs->fs_fsmnt);
    974 				error = wapbl_replay_write(mp->mnt_wapbl_replay,
    975 				    devvp);
    976 				if (error)
    977 					goto out;
    978 				wapbl_replay_stop(mp->mnt_wapbl_replay);
    979 				fs->fs_clean = FS_WASCLEAN;
    980 			} else {
    981 				/* XXX fsmnt may be stale */
    982 				printf("%s: replaying log to memory\n",
    983 				    fs->fs_fsmnt);
    984 			}
    985 
    986 			/* Force a re-read of the superblock */
    987 			brelse(bp, BC_INVAL);
    988 			bp = NULL;
    989 			free(fs, M_UFSMNT);
    990 			fs = NULL;
    991 			goto sbagain;
    992 		}
    993 	}
    994 #else /* !WAPBL */
    995 	if ((fs->fs_flags & FS_DOWAPBL) && (mp->mnt_flag & MNT_FORCE) == 0) {
    996 		error = EPERM;
    997 		goto out;
    998 	}
    999 #endif /* !WAPBL */
   1000 
   1001 	ffs_oldfscompat_read(fs, ump, sblockloc);
   1002 	ump->um_maxfilesize = fs->fs_maxfilesize;
   1003 
   1004 	if (fs->fs_flags & ~(FS_KNOWN_FLAGS | FS_INTERNAL)) {
   1005 		uprintf("%s: unknown ufs flags: 0x%08"PRIx32"%s\n",
   1006 		    mp->mnt_stat.f_mntonname, fs->fs_flags,
   1007 		    (mp->mnt_flag & MNT_FORCE) ? "" : ", not mounting");
   1008 		if ((mp->mnt_flag & MNT_FORCE) == 0) {
   1009 			error = EINVAL;
   1010 			goto out;
   1011 		}
   1012 	}
   1013 
   1014 	if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
   1015 		fs->fs_pendingblocks = 0;
   1016 		fs->fs_pendinginodes = 0;
   1017 	}
   1018 
   1019 	ump->um_fstype = fstype;
   1020 	if (fs->fs_sbsize < SBLOCKSIZE)
   1021 		brelse(bp, BC_INVAL);
   1022 	else
   1023 		brelse(bp, 0);
   1024 	bp = NULL;
   1025 
   1026 	/* First check to see if this is tagged as an Apple UFS filesystem
   1027 	 * in the disklabel
   1028 	 */
   1029 	if ((VOP_IOCTL(devvp, DIOCGPART, &dpart, FREAD, cred) == 0) &&
   1030 		(dpart.part->p_fstype == FS_APPLEUFS)) {
   1031 		ump->um_flags |= UFS_ISAPPLEUFS;
   1032 	}
   1033 #ifdef APPLE_UFS
   1034 	else {
   1035 		/* Manually look for an apple ufs label, and if a valid one
   1036 		 * is found, then treat it like an Apple UFS filesystem anyway
   1037 		 */
   1038 		error = bread(devvp, (daddr_t)(APPLEUFS_LABEL_OFFSET / DEV_BSIZE),
   1039 			APPLEUFS_LABEL_SIZE, cred, 0, &bp);
   1040 		if (error)
   1041 			goto out;
   1042 		error = ffs_appleufs_validate(fs->fs_fsmnt,
   1043 			(struct appleufslabel *)bp->b_data, NULL);
   1044 		if (error == 0) {
   1045 			ump->um_flags |= UFS_ISAPPLEUFS;
   1046 		}
   1047 		brelse(bp, 0);
   1048 		bp = NULL;
   1049 	}
   1050 #else
   1051 	if (ump->um_flags & UFS_ISAPPLEUFS) {
   1052 		error = EINVAL;
   1053 		goto out;
   1054 	}
   1055 #endif
   1056 
   1057 #if 0
   1058 /*
   1059  * XXX This code changes the behaviour of mounting dirty filesystems, to
   1060  * XXX require "mount -f ..." to mount them.  This doesn't match what
   1061  * XXX mount(8) describes and is disabled for now.
   1062  */
   1063 	/*
   1064 	 * If the file system is not clean, don't allow it to be mounted
   1065 	 * unless MNT_FORCE is specified.  (Note: MNT_FORCE is always set
   1066 	 * for the root file system.)
   1067 	 */
   1068 	if (fs->fs_flags & FS_DOWAPBL) {
   1069 		/*
   1070 		 * wapbl normally expects to be FS_WASCLEAN when the FS_DOWAPBL
   1071 		 * bit is set, although there's a window in unmount where it
   1072 		 * could be FS_ISCLEAN
   1073 		 */
   1074 		if ((mp->mnt_flag & MNT_FORCE) == 0 &&
   1075 		    (fs->fs_clean & (FS_WASCLEAN | FS_ISCLEAN)) == 0) {
   1076 			error = EPERM;
   1077 			goto out;
   1078 		}
   1079 	} else
   1080 		if ((fs->fs_clean & FS_ISCLEAN) == 0 &&
   1081 		    (mp->mnt_flag & MNT_FORCE) == 0) {
   1082 			error = EPERM;
   1083 			goto out;
   1084 		}
   1085 #endif
   1086 
   1087 	/*
   1088 	 * verify that we can access the last block in the fs
   1089 	 * if we're mounting read/write.
   1090 	 */
   1091 
   1092 	if (!ronly) {
   1093 		error = bread(devvp, fsbtodb(fs, fs->fs_size - 1), fs->fs_fsize,
   1094 		    cred, 0, &bp);
   1095 		if (bp->b_bcount != fs->fs_fsize)
   1096 			error = EINVAL;
   1097 		if (error) {
   1098 			bset = BC_INVAL;
   1099 			goto out;
   1100 		}
   1101 		brelse(bp, BC_INVAL);
   1102 		bp = NULL;
   1103 	}
   1104 
   1105 	fs->fs_ronly = ronly;
   1106 	/* Don't bump fs_clean if we're replaying journal */
   1107 	if (!((fs->fs_flags & FS_DOWAPBL) && (fs->fs_clean & FS_WASCLEAN)))
   1108 		if (ronly == 0) {
   1109 			fs->fs_clean <<= 1;
   1110 			fs->fs_fmod = 1;
   1111 		}
   1112 	bsize = fs->fs_cssize;
   1113 	blks = howmany(bsize, fs->fs_fsize);
   1114 	if (fs->fs_contigsumsize > 0)
   1115 		bsize += fs->fs_ncg * sizeof(int32_t);
   1116 	bsize += fs->fs_ncg * sizeof(*fs->fs_contigdirs);
   1117 	space = malloc((u_long)bsize, M_UFSMNT, M_WAITOK);
   1118 	fs->fs_csp = space;
   1119 	for (i = 0; i < blks; i += fs->fs_frag) {
   1120 		bsize = fs->fs_bsize;
   1121 		if (i + fs->fs_frag > blks)
   1122 			bsize = (blks - i) * fs->fs_fsize;
   1123 		error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), bsize,
   1124 			      cred, 0, &bp);
   1125 		if (error) {
   1126 			free(fs->fs_csp, M_UFSMNT);
   1127 			goto out;
   1128 		}
   1129 #ifdef FFS_EI
   1130 		if (needswap)
   1131 			ffs_csum_swap((struct csum *)bp->b_data,
   1132 				(struct csum *)space, bsize);
   1133 		else
   1134 #endif
   1135 			memcpy(space, bp->b_data, (u_int)bsize);
   1136 
   1137 		space = (char *)space + bsize;
   1138 		brelse(bp, 0);
   1139 		bp = NULL;
   1140 	}
   1141 	if (fs->fs_contigsumsize > 0) {
   1142 		fs->fs_maxcluster = lp = space;
   1143 		for (i = 0; i < fs->fs_ncg; i++)
   1144 			*lp++ = fs->fs_contigsumsize;
   1145 		space = lp;
   1146 	}
   1147 	bsize = fs->fs_ncg * sizeof(*fs->fs_contigdirs);
   1148 	fs->fs_contigdirs = space;
   1149 	space = (char *)space + bsize;
   1150 	memset(fs->fs_contigdirs, 0, bsize);
   1151 		/* Compatibility for old filesystems - XXX */
   1152 	if (fs->fs_avgfilesize <= 0)
   1153 		fs->fs_avgfilesize = AVFILESIZ;
   1154 	if (fs->fs_avgfpdir <= 0)
   1155 		fs->fs_avgfpdir = AFPDIR;
   1156 	fs->fs_active = NULL;
   1157 	mp->mnt_data = ump;
   1158 	mp->mnt_stat.f_fsidx.__fsid_val[0] = (long)dev;
   1159 	mp->mnt_stat.f_fsidx.__fsid_val[1] = makefstype(MOUNT_FFS);
   1160 	mp->mnt_stat.f_fsid = mp->mnt_stat.f_fsidx.__fsid_val[0];
   1161 	mp->mnt_stat.f_namemax = FFS_MAXNAMLEN;
   1162 	if (UFS_MPISAPPLEUFS(ump)) {
   1163 		/* NeXT used to keep short symlinks in the inode even
   1164 		 * when using FS_42INODEFMT.  In that case fs->fs_maxsymlinklen
   1165 		 * is probably -1, but we still need to be able to identify
   1166 		 * short symlinks.
   1167 		 */
   1168 		ump->um_maxsymlinklen = APPLEUFS_MAXSYMLINKLEN;
   1169 		ump->um_dirblksiz = APPLEUFS_DIRBLKSIZ;
   1170 		mp->mnt_iflag |= IMNT_DTYPE;
   1171 	} else {
   1172 		ump->um_maxsymlinklen = fs->fs_maxsymlinklen;
   1173 		ump->um_dirblksiz = DIRBLKSIZ;
   1174 		if (ump->um_maxsymlinklen > 0)
   1175 			mp->mnt_iflag |= IMNT_DTYPE;
   1176 		else
   1177 			mp->mnt_iflag &= ~IMNT_DTYPE;
   1178 	}
   1179 	mp->mnt_fs_bshift = fs->fs_bshift;
   1180 	mp->mnt_dev_bshift = DEV_BSHIFT;	/* XXX */
   1181 	mp->mnt_flag |= MNT_LOCAL;
   1182 	mp->mnt_iflag |= IMNT_MPSAFE;
   1183 #ifdef FFS_EI
   1184 	if (needswap)
   1185 		ump->um_flags |= UFS_NEEDSWAP;
   1186 #endif
   1187 	ump->um_mountp = mp;
   1188 	ump->um_dev = dev;
   1189 	ump->um_devvp = devvp;
   1190 	ump->um_nindir = fs->fs_nindir;
   1191 	ump->um_lognindir = ffs(fs->fs_nindir) - 1;
   1192 	ump->um_bptrtodb = fs->fs_fshift - DEV_BSHIFT;
   1193 	ump->um_seqinc = fs->fs_frag;
   1194 	for (i = 0; i < MAXQUOTAS; i++)
   1195 		ump->um_quotas[i] = NULLVP;
   1196 	devvp->v_specmountpoint = mp;
   1197 	if (ronly == 0 && fs->fs_snapinum[0] != 0)
   1198 		ffs_snapshot_mount(mp);
   1199 
   1200 #ifdef WAPBL
   1201 	if (!ronly) {
   1202 		KDASSERT(fs->fs_ronly == 0);
   1203 		/*
   1204 		 * ffs_wapbl_start() needs mp->mnt_stat initialised if it
   1205 		 * needs to create a new log file in-filesystem.
   1206 		 */
   1207 		ffs_statvfs(mp, &mp->mnt_stat);
   1208 
   1209 		error = ffs_wapbl_start(mp);
   1210 		if (error) {
   1211 			free(fs->fs_csp, M_UFSMNT);
   1212 			goto out;
   1213 		}
   1214 	}
   1215 #endif /* WAPBL */
   1216 #ifdef UFS_EXTATTR
   1217 	/*
   1218 	 * Initialize file-backed extended attributes on UFS1 file
   1219 	 * systems.
   1220 	 */
   1221 	if (ump->um_fstype == UFS1) {
   1222 		ufs_extattr_uepm_init(&ump->um_extattr);
   1223 #ifdef UFS_EXTATTR_AUTOSTART
   1224 		/*
   1225 		 * XXX Just ignore errors.  Not clear that we should
   1226 		 * XXX fail the mount in this case.
   1227 		 */
   1228 		(void) ufs_extattr_autostart(mp, l);
   1229 #endif
   1230 	}
   1231 #endif /* UFS_EXTATTR */
   1232 	return (0);
   1233 out:
   1234 #ifdef WAPBL
   1235 	if (mp->mnt_wapbl_replay) {
   1236 		wapbl_replay_stop(mp->mnt_wapbl_replay);
   1237 		wapbl_replay_free(mp->mnt_wapbl_replay);
   1238 		mp->mnt_wapbl_replay = 0;
   1239 	}
   1240 #endif
   1241 
   1242 	fstrans_unmount(mp);
   1243 	if (fs)
   1244 		free(fs, M_UFSMNT);
   1245 	devvp->v_specmountpoint = NULL;
   1246 	if (bp)
   1247 		brelse(bp, bset);
   1248 	if (ump) {
   1249 		if (ump->um_oldfscompat)
   1250 			free(ump->um_oldfscompat, M_UFSMNT);
   1251 		mutex_destroy(&ump->um_lock);
   1252 		free(ump, M_UFSMNT);
   1253 		mp->mnt_data = NULL;
   1254 	}
   1255 	return (error);
   1256 }
   1257 
   1258 /*
   1259  * Sanity checks for loading old filesystem superblocks.
   1260  * See ffs_oldfscompat_write below for unwound actions.
   1261  *
   1262  * XXX - Parts get retired eventually.
   1263  * Unfortunately new bits get added.
   1264  */
   1265 static void
   1266 ffs_oldfscompat_read(struct fs *fs, struct ufsmount *ump, daddr_t sblockloc)
   1267 {
   1268 	off_t maxfilesize;
   1269 	int32_t *extrasave;
   1270 
   1271 	if ((fs->fs_magic != FS_UFS1_MAGIC) ||
   1272 	    (fs->fs_old_flags & FS_FLAGS_UPDATED))
   1273 		return;
   1274 
   1275 	if (!ump->um_oldfscompat)
   1276 		ump->um_oldfscompat = malloc(512 + 3*sizeof(int32_t),
   1277 		    M_UFSMNT, M_WAITOK);
   1278 
   1279 	memcpy(ump->um_oldfscompat, &fs->fs_old_postbl_start, 512);
   1280 	extrasave = ump->um_oldfscompat;
   1281 	extrasave += 512/sizeof(int32_t);
   1282 	extrasave[0] = fs->fs_old_npsect;
   1283 	extrasave[1] = fs->fs_old_interleave;
   1284 	extrasave[2] = fs->fs_old_trackskew;
   1285 
   1286 	/* These fields will be overwritten by their
   1287 	 * original values in fs_oldfscompat_write, so it is harmless
   1288 	 * to modify them here.
   1289 	 */
   1290 	fs->fs_cstotal.cs_ndir = fs->fs_old_cstotal.cs_ndir;
   1291 	fs->fs_cstotal.cs_nbfree = fs->fs_old_cstotal.cs_nbfree;
   1292 	fs->fs_cstotal.cs_nifree = fs->fs_old_cstotal.cs_nifree;
   1293 	fs->fs_cstotal.cs_nffree = fs->fs_old_cstotal.cs_nffree;
   1294 
   1295 	fs->fs_maxbsize = fs->fs_bsize;
   1296 	fs->fs_time = fs->fs_old_time;
   1297 	fs->fs_size = fs->fs_old_size;
   1298 	fs->fs_dsize = fs->fs_old_dsize;
   1299 	fs->fs_csaddr = fs->fs_old_csaddr;
   1300 	fs->fs_sblockloc = sblockloc;
   1301 
   1302 	fs->fs_flags = fs->fs_old_flags | (fs->fs_flags & FS_INTERNAL);
   1303 
   1304 	if (fs->fs_old_postblformat == FS_42POSTBLFMT) {
   1305 		fs->fs_old_nrpos = 8;
   1306 		fs->fs_old_npsect = fs->fs_old_nsect;
   1307 		fs->fs_old_interleave = 1;
   1308 		fs->fs_old_trackskew = 0;
   1309 	}
   1310 
   1311 	if (fs->fs_old_inodefmt < FS_44INODEFMT) {
   1312 		fs->fs_maxfilesize = (u_quad_t) 1LL << 39;
   1313 		fs->fs_qbmask = ~fs->fs_bmask;
   1314 		fs->fs_qfmask = ~fs->fs_fmask;
   1315 	}
   1316 
   1317 	maxfilesize = (u_int64_t)0x80000000 * fs->fs_bsize - 1;
   1318 	if (fs->fs_maxfilesize > maxfilesize)
   1319 		fs->fs_maxfilesize = maxfilesize;
   1320 
   1321 	/* Compatibility for old filesystems */
   1322 	if (fs->fs_avgfilesize <= 0)
   1323 		fs->fs_avgfilesize = AVFILESIZ;
   1324 	if (fs->fs_avgfpdir <= 0)
   1325 		fs->fs_avgfpdir = AFPDIR;
   1326 
   1327 #if 0
   1328 	if (bigcgs) {
   1329 		fs->fs_save_cgsize = fs->fs_cgsize;
   1330 		fs->fs_cgsize = fs->fs_bsize;
   1331 	}
   1332 #endif
   1333 }
   1334 
   1335 /*
   1336  * Unwinding superblock updates for old filesystems.
   1337  * See ffs_oldfscompat_read above for details.
   1338  *
   1339  * XXX - Parts get retired eventually.
   1340  * Unfortunately new bits get added.
   1341  */
   1342 static void
   1343 ffs_oldfscompat_write(struct fs *fs, struct ufsmount *ump)
   1344 {
   1345 	int32_t *extrasave;
   1346 
   1347 	if ((fs->fs_magic != FS_UFS1_MAGIC) ||
   1348 	    (fs->fs_old_flags & FS_FLAGS_UPDATED))
   1349 		return;
   1350 
   1351 	fs->fs_old_time = fs->fs_time;
   1352 	fs->fs_old_cstotal.cs_ndir = fs->fs_cstotal.cs_ndir;
   1353 	fs->fs_old_cstotal.cs_nbfree = fs->fs_cstotal.cs_nbfree;
   1354 	fs->fs_old_cstotal.cs_nifree = fs->fs_cstotal.cs_nifree;
   1355 	fs->fs_old_cstotal.cs_nffree = fs->fs_cstotal.cs_nffree;
   1356 	fs->fs_old_flags = fs->fs_flags;
   1357 
   1358 #if 0
   1359 	if (bigcgs) {
   1360 		fs->fs_cgsize = fs->fs_save_cgsize;
   1361 	}
   1362 #endif
   1363 
   1364 	memcpy(&fs->fs_old_postbl_start, ump->um_oldfscompat, 512);
   1365 	extrasave = ump->um_oldfscompat;
   1366 	extrasave += 512/sizeof(int32_t);
   1367 	fs->fs_old_npsect = extrasave[0];
   1368 	fs->fs_old_interleave = extrasave[1];
   1369 	fs->fs_old_trackskew = extrasave[2];
   1370 
   1371 }
   1372 
   1373 /*
   1374  * unmount vfs operation
   1375  */
   1376 int
   1377 ffs_unmount(struct mount *mp, int mntflags)
   1378 {
   1379 	struct lwp *l = curlwp;
   1380 	struct ufsmount *ump = VFSTOUFS(mp);
   1381 	struct fs *fs = ump->um_fs;
   1382 	int error, flags;
   1383 #ifdef WAPBL
   1384 	extern int doforce;
   1385 #endif
   1386 
   1387 	flags = 0;
   1388 	if (mntflags & MNT_FORCE)
   1389 		flags |= FORCECLOSE;
   1390 	if ((error = ffs_flushfiles(mp, flags, l)) != 0)
   1391 		return (error);
   1392 	error = UFS_WAPBL_BEGIN(mp);
   1393 	if (error == 0)
   1394 		if (fs->fs_ronly == 0 &&
   1395 		    ffs_cgupdate(ump, MNT_WAIT) == 0 &&
   1396 		    fs->fs_clean & FS_WASCLEAN) {
   1397 			fs->fs_clean = FS_ISCLEAN;
   1398 			fs->fs_fmod = 0;
   1399 			(void) ffs_sbupdate(ump, MNT_WAIT);
   1400 		}
   1401 	if (error == 0)
   1402 		UFS_WAPBL_END(mp);
   1403 #ifdef WAPBL
   1404 	KASSERT(!(mp->mnt_wapbl_replay && mp->mnt_wapbl));
   1405 	if (mp->mnt_wapbl_replay) {
   1406 		KDASSERT(fs->fs_ronly);
   1407 		wapbl_replay_stop(mp->mnt_wapbl_replay);
   1408 		wapbl_replay_free(mp->mnt_wapbl_replay);
   1409 		mp->mnt_wapbl_replay = 0;
   1410 	}
   1411 	error = ffs_wapbl_stop(mp, doforce && (mntflags & MNT_FORCE));
   1412 	if (error) {
   1413 		return error;
   1414 	}
   1415 #endif /* WAPBL */
   1416 #ifdef UFS_EXTATTR
   1417 	if (ump->um_fstype == UFS1) {
   1418 		ufs_extattr_stop(mp, l);
   1419 		ufs_extattr_uepm_destroy(&ump->um_extattr);
   1420 	}
   1421 #endif /* UFS_EXTATTR */
   1422 
   1423 	if (ump->um_devvp->v_type != VBAD)
   1424 		ump->um_devvp->v_specmountpoint = NULL;
   1425 	vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY);
   1426 	(void)VOP_CLOSE(ump->um_devvp, fs->fs_ronly ? FREAD : FREAD | FWRITE,
   1427 		NOCRED);
   1428 	vput(ump->um_devvp);
   1429 	free(fs->fs_csp, M_UFSMNT);
   1430 	free(fs, M_UFSMNT);
   1431 	if (ump->um_oldfscompat != NULL)
   1432 		free(ump->um_oldfscompat, M_UFSMNT);
   1433 	mutex_destroy(&ump->um_lock);
   1434 	ffs_snapshot_fini(ump);
   1435 	free(ump, M_UFSMNT);
   1436 	mp->mnt_data = NULL;
   1437 	mp->mnt_flag &= ~MNT_LOCAL;
   1438 	fstrans_unmount(mp);
   1439 	return (0);
   1440 }
   1441 
   1442 /*
   1443  * Flush out all the files in a filesystem.
   1444  */
   1445 int
   1446 ffs_flushfiles(struct mount *mp, int flags, struct lwp *l)
   1447 {
   1448 	extern int doforce;
   1449 	struct ufsmount *ump;
   1450 	int error;
   1451 
   1452 	if (!doforce)
   1453 		flags &= ~FORCECLOSE;
   1454 	ump = VFSTOUFS(mp);
   1455 #ifdef QUOTA
   1456 	if (mp->mnt_flag & MNT_QUOTA) {
   1457 		int i;
   1458 		if ((error = vflush(mp, NULLVP, SKIPSYSTEM | flags)) != 0)
   1459 			return (error);
   1460 		for (i = 0; i < MAXQUOTAS; i++) {
   1461 			if (ump->um_quotas[i] == NULLVP)
   1462 				continue;
   1463 			quotaoff(l, mp, i);
   1464 		}
   1465 		/*
   1466 		 * Here we fall through to vflush again to ensure
   1467 		 * that we have gotten rid of all the system vnodes.
   1468 		 */
   1469 	}
   1470 #endif
   1471 	if ((error = vflush(mp, 0, SKIPSYSTEM | flags)) != 0)
   1472 		return (error);
   1473 	ffs_snapshot_unmount(mp);
   1474 	/*
   1475 	 * Flush all the files.
   1476 	 */
   1477 	error = vflush(mp, NULLVP, flags);
   1478 	if (error)
   1479 		return (error);
   1480 	/*
   1481 	 * Flush filesystem metadata.
   1482 	 */
   1483 	vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY);
   1484 	error = VOP_FSYNC(ump->um_devvp, l->l_cred, FSYNC_WAIT, 0, 0);
   1485 	VOP_UNLOCK(ump->um_devvp);
   1486 	if (flags & FORCECLOSE) /* XXXDBJ */
   1487 		error = 0;
   1488 
   1489 #ifdef WAPBL
   1490 	if (error)
   1491 		return error;
   1492 	if (mp->mnt_wapbl) {
   1493 		error = wapbl_flush(mp->mnt_wapbl, 1);
   1494 		if (flags & FORCECLOSE)
   1495 			error = 0;
   1496 	}
   1497 #endif
   1498 
   1499 	return (error);
   1500 }
   1501 
   1502 /*
   1503  * Get file system statistics.
   1504  */
   1505 int
   1506 ffs_statvfs(struct mount *mp, struct statvfs *sbp)
   1507 {
   1508 	struct ufsmount *ump;
   1509 	struct fs *fs;
   1510 
   1511 	ump = VFSTOUFS(mp);
   1512 	fs = ump->um_fs;
   1513 	mutex_enter(&ump->um_lock);
   1514 	sbp->f_bsize = fs->fs_bsize;
   1515 	sbp->f_frsize = fs->fs_fsize;
   1516 	sbp->f_iosize = fs->fs_bsize;
   1517 	sbp->f_blocks = fs->fs_dsize;
   1518 	sbp->f_bfree = blkstofrags(fs, fs->fs_cstotal.cs_nbfree) +
   1519 	    fs->fs_cstotal.cs_nffree + dbtofsb(fs, fs->fs_pendingblocks);
   1520 	sbp->f_bresvd = ((u_int64_t) fs->fs_dsize * (u_int64_t)
   1521 	    fs->fs_minfree) / (u_int64_t) 100;
   1522 	if (sbp->f_bfree > sbp->f_bresvd)
   1523 		sbp->f_bavail = sbp->f_bfree - sbp->f_bresvd;
   1524 	else
   1525 		sbp->f_bavail = 0;
   1526 	sbp->f_files =  fs->fs_ncg * fs->fs_ipg - ROOTINO;
   1527 	sbp->f_ffree = fs->fs_cstotal.cs_nifree + fs->fs_pendinginodes;
   1528 	sbp->f_favail = sbp->f_ffree;
   1529 	sbp->f_fresvd = 0;
   1530 	mutex_exit(&ump->um_lock);
   1531 	copy_statvfs_info(sbp, mp);
   1532 
   1533 	return (0);
   1534 }
   1535 
   1536 /*
   1537  * Go through the disk queues to initiate sandbagged IO;
   1538  * go through the inodes to write those that have been modified;
   1539  * initiate the writing of the super block if it has been modified.
   1540  *
   1541  * Note: we are always called with the filesystem marked `MPBUSY'.
   1542  */
   1543 int
   1544 ffs_sync(struct mount *mp, int waitfor, kauth_cred_t cred)
   1545 {
   1546 	struct vnode *vp, *mvp, *nvp;
   1547 	struct inode *ip;
   1548 	struct ufsmount *ump = VFSTOUFS(mp);
   1549 	struct fs *fs;
   1550 	int error, allerror = 0;
   1551 	bool is_suspending;
   1552 
   1553 	fs = ump->um_fs;
   1554 	if (fs->fs_fmod != 0 && fs->fs_ronly != 0) {		/* XXX */
   1555 		printf("fs = %s\n", fs->fs_fsmnt);
   1556 		panic("update: rofs mod");
   1557 	}
   1558 
   1559 	/* Allocate a marker vnode. */
   1560 	if ((mvp = vnalloc(mp)) == NULL)
   1561 		return (ENOMEM);
   1562 
   1563 	fstrans_start(mp, FSTRANS_SHARED);
   1564 	is_suspending = (fstrans_getstate(mp) == FSTRANS_SUSPENDING);
   1565 	/*
   1566 	 * Write back each (modified) inode.
   1567 	 */
   1568 	mutex_enter(&mntvnode_lock);
   1569 loop:
   1570 	/*
   1571 	 * NOTE: not using the TAILQ_FOREACH here since in this loop vgone()
   1572 	 * and vclean() can be called indirectly
   1573 	 */
   1574 	for (vp = TAILQ_FIRST(&mp->mnt_vnodelist); vp; vp = nvp) {
   1575 		nvp = TAILQ_NEXT(vp, v_mntvnodes);
   1576 		/*
   1577 		 * If the vnode that we are about to sync is no longer
   1578 		 * associated with this mount point, start over.
   1579 		 */
   1580 		if (vp->v_mount != mp)
   1581 			goto loop;
   1582 		/*
   1583 		 * Don't interfere with concurrent scans of this FS.
   1584 		 */
   1585 		if (vismarker(vp))
   1586 			continue;
   1587 		mutex_enter(vp->v_interlock);
   1588 		ip = VTOI(vp);
   1589 
   1590 		/*
   1591 		 * Skip the vnode/inode if inaccessible.
   1592 		 */
   1593 		if (ip == NULL || (vp->v_iflag & (VI_XLOCK | VI_CLEAN)) != 0 ||
   1594 		    vp->v_type == VNON) {
   1595 			mutex_exit(vp->v_interlock);
   1596 			continue;
   1597 		}
   1598 
   1599 		/*
   1600 		 * We deliberately update inode times here.  This will
   1601 		 * prevent a massive queue of updates accumulating, only
   1602 		 * to be handled by a call to unmount.
   1603 		 *
   1604 		 * XXX It would be better to have the syncer trickle these
   1605 		 * out.  Adjustment needed to allow registering vnodes for
   1606 		 * sync when the vnode is clean, but the inode dirty.  Or
   1607 		 * have ufs itself trickle out inode updates.
   1608 		 *
   1609 		 * If doing a lazy sync, we don't care about metadata or
   1610 		 * data updates, because they are handled by each vnode's
   1611 		 * synclist entry.  In this case we are only interested in
   1612 		 * writing back modified inodes.
   1613 		 */
   1614 		if ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_UPDATE |
   1615 		    IN_MODIFY | IN_MODIFIED | IN_ACCESSED)) == 0 &&
   1616 		    (waitfor == MNT_LAZY || (LIST_EMPTY(&vp->v_dirtyblkhd) &&
   1617 		    UVM_OBJ_IS_CLEAN(&vp->v_uobj)))) {
   1618 			mutex_exit(vp->v_interlock);
   1619 			continue;
   1620 		}
   1621 		if (vp->v_type == VBLK && is_suspending) {
   1622 			mutex_exit(vp->v_interlock);
   1623 			continue;
   1624 		}
   1625 		vmark(mvp, vp);
   1626 		mutex_exit(&mntvnode_lock);
   1627 		error = vget(vp, LK_EXCLUSIVE | LK_NOWAIT);
   1628 		if (error) {
   1629 			mutex_enter(&mntvnode_lock);
   1630 			nvp = vunmark(mvp);
   1631 			if (error == ENOENT) {
   1632 				goto loop;
   1633 			}
   1634 			continue;
   1635 		}
   1636 		if (waitfor == MNT_LAZY) {
   1637 			error = UFS_WAPBL_BEGIN(vp->v_mount);
   1638 			if (!error) {
   1639 				error = ffs_update(vp, NULL, NULL,
   1640 				    UPDATE_CLOSE);
   1641 				UFS_WAPBL_END(vp->v_mount);
   1642 			}
   1643 		} else {
   1644 			error = VOP_FSYNC(vp, cred, FSYNC_NOLOG |
   1645 			    (waitfor == MNT_WAIT ? FSYNC_WAIT : 0), 0, 0);
   1646 		}
   1647 		if (error)
   1648 			allerror = error;
   1649 		vput(vp);
   1650 		mutex_enter(&mntvnode_lock);
   1651 		nvp = vunmark(mvp);
   1652 	}
   1653 	mutex_exit(&mntvnode_lock);
   1654 	/*
   1655 	 * Force stale file system control information to be flushed.
   1656 	 */
   1657 	if (waitfor != MNT_LAZY && (ump->um_devvp->v_numoutput > 0 ||
   1658 	    !LIST_EMPTY(&ump->um_devvp->v_dirtyblkhd))) {
   1659 		vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY);
   1660 		if ((error = VOP_FSYNC(ump->um_devvp, cred,
   1661 		    (waitfor == MNT_WAIT ? FSYNC_WAIT : 0) | FSYNC_NOLOG,
   1662 		    0, 0)) != 0)
   1663 			allerror = error;
   1664 		VOP_UNLOCK(ump->um_devvp);
   1665 		if (allerror == 0 && waitfor == MNT_WAIT && !mp->mnt_wapbl) {
   1666 			mutex_enter(&mntvnode_lock);
   1667 			goto loop;
   1668 		}
   1669 	}
   1670 #ifdef QUOTA
   1671 	qsync(mp);
   1672 #endif
   1673 	/*
   1674 	 * Write back modified superblock.
   1675 	 */
   1676 	if (fs->fs_fmod != 0) {
   1677 		fs->fs_fmod = 0;
   1678 		fs->fs_time = time_second;
   1679 		error = UFS_WAPBL_BEGIN(mp);
   1680 		if (error)
   1681 			allerror = error;
   1682 		else {
   1683 			if ((error = ffs_cgupdate(ump, waitfor)))
   1684 				allerror = error;
   1685 			UFS_WAPBL_END(mp);
   1686 		}
   1687 	}
   1688 
   1689 #ifdef WAPBL
   1690 	if (mp->mnt_wapbl) {
   1691 		error = wapbl_flush(mp->mnt_wapbl, 0);
   1692 		if (error)
   1693 			allerror = error;
   1694 	}
   1695 #endif
   1696 
   1697 	fstrans_done(mp);
   1698 	vnfree(mvp);
   1699 	return (allerror);
   1700 }
   1701 
   1702 /*
   1703  * Look up a FFS dinode number to find its incore vnode, otherwise read it
   1704  * in from disk.  If it is in core, wait for the lock bit to clear, then
   1705  * return the inode locked.  Detection and handling of mount points must be
   1706  * done by the calling routine.
   1707  */
   1708 int
   1709 ffs_vget(struct mount *mp, ino_t ino, struct vnode **vpp)
   1710 {
   1711 	struct fs *fs;
   1712 	struct inode *ip;
   1713 	struct ufsmount *ump;
   1714 	struct buf *bp;
   1715 	struct vnode *vp;
   1716 	dev_t dev;
   1717 	int error;
   1718 
   1719 	ump = VFSTOUFS(mp);
   1720 	dev = ump->um_dev;
   1721 
   1722  retry:
   1723 	if ((*vpp = ufs_ihashget(dev, ino, LK_EXCLUSIVE)) != NULL)
   1724 		return (0);
   1725 
   1726 	/* Allocate a new vnode/inode. */
   1727 	if ((error = getnewvnode(VT_UFS, mp, ffs_vnodeop_p, &vp)) != 0) {
   1728 		*vpp = NULL;
   1729 		return (error);
   1730 	}
   1731 	ip = pool_cache_get(ffs_inode_cache, PR_WAITOK);
   1732 
   1733 	/*
   1734 	 * If someone beat us to it, put back the freshly allocated
   1735 	 * vnode/inode pair and retry.
   1736 	 */
   1737 	mutex_enter(&ufs_hashlock);
   1738 	if (ufs_ihashget(dev, ino, 0) != NULL) {
   1739 		mutex_exit(&ufs_hashlock);
   1740 		ungetnewvnode(vp);
   1741 		pool_cache_put(ffs_inode_cache, ip);
   1742 		goto retry;
   1743 	}
   1744 
   1745 	vp->v_vflag |= VV_LOCKSWORK;
   1746 
   1747 	/*
   1748 	 * XXX MFS ends up here, too, to allocate an inode.  Should we
   1749 	 * XXX create another pool for MFS inodes?
   1750 	 */
   1751 
   1752 	memset(ip, 0, sizeof(struct inode));
   1753 	vp->v_data = ip;
   1754 	ip->i_vnode = vp;
   1755 	ip->i_ump = ump;
   1756 	ip->i_fs = fs = ump->um_fs;
   1757 	ip->i_dev = dev;
   1758 	ip->i_number = ino;
   1759 #ifdef QUOTA
   1760 	ufsquota_init(ip);
   1761 #endif
   1762 
   1763 	/*
   1764 	 * Initialize genfs node, we might proceed to destroy it in
   1765 	 * error branches.
   1766 	 */
   1767 	genfs_node_init(vp, &ffs_genfsops);
   1768 
   1769 	/*
   1770 	 * Put it onto its hash chain and lock it so that other requests for
   1771 	 * this inode will block if they arrive while we are sleeping waiting
   1772 	 * for old data structures to be purged or for the contents of the
   1773 	 * disk portion of this inode to be read.
   1774 	 */
   1775 
   1776 	ufs_ihashins(ip);
   1777 	mutex_exit(&ufs_hashlock);
   1778 
   1779 	/* Read in the disk contents for the inode, copy into the inode. */
   1780 	error = bread(ump->um_devvp, fsbtodb(fs, ino_to_fsba(fs, ino)),
   1781 		      (int)fs->fs_bsize, NOCRED, 0, &bp);
   1782 	if (error) {
   1783 
   1784 		/*
   1785 		 * The inode does not contain anything useful, so it would
   1786 		 * be misleading to leave it on its hash chain. With mode
   1787 		 * still zero, it will be unlinked and returned to the free
   1788 		 * list by vput().
   1789 		 */
   1790 
   1791 		vput(vp);
   1792 		brelse(bp, 0);
   1793 		*vpp = NULL;
   1794 		return (error);
   1795 	}
   1796 	if (ip->i_ump->um_fstype == UFS1)
   1797 		ip->i_din.ffs1_din = pool_cache_get(ffs_dinode1_cache,
   1798 		    PR_WAITOK);
   1799 	else
   1800 		ip->i_din.ffs2_din = pool_cache_get(ffs_dinode2_cache,
   1801 		    PR_WAITOK);
   1802 	ffs_load_inode(bp, ip, fs, ino);
   1803 	brelse(bp, 0);
   1804 
   1805 	/*
   1806 	 * Initialize the vnode from the inode, check for aliases.
   1807 	 * Note that the underlying vnode may have changed.
   1808 	 */
   1809 
   1810 	ufs_vinit(mp, ffs_specop_p, ffs_fifoop_p, &vp);
   1811 
   1812 	/*
   1813 	 * Finish inode initialization now that aliasing has been resolved.
   1814 	 */
   1815 
   1816 	ip->i_devvp = ump->um_devvp;
   1817 	vref(ip->i_devvp);
   1818 
   1819 	/*
   1820 	 * Ensure that uid and gid are correct. This is a temporary
   1821 	 * fix until fsck has been changed to do the update.
   1822 	 */
   1823 
   1824 	if (fs->fs_old_inodefmt < FS_44INODEFMT) {		/* XXX */
   1825 		ip->i_uid = ip->i_ffs1_ouid;			/* XXX */
   1826 		ip->i_gid = ip->i_ffs1_ogid;			/* XXX */
   1827 	}							/* XXX */
   1828 	uvm_vnp_setsize(vp, ip->i_size);
   1829 	*vpp = vp;
   1830 	return (0);
   1831 }
   1832 
   1833 /*
   1834  * File handle to vnode
   1835  *
   1836  * Have to be really careful about stale file handles:
   1837  * - check that the inode number is valid
   1838  * - call ffs_vget() to get the locked inode
   1839  * - check for an unallocated inode (i_mode == 0)
   1840  * - check that the given client host has export rights and return
   1841  *   those rights via. exflagsp and credanonp
   1842  */
   1843 int
   1844 ffs_fhtovp(struct mount *mp, struct fid *fhp, struct vnode **vpp)
   1845 {
   1846 	struct ufid ufh;
   1847 	struct fs *fs;
   1848 
   1849 	if (fhp->fid_len != sizeof(struct ufid))
   1850 		return EINVAL;
   1851 
   1852 	memcpy(&ufh, fhp, sizeof(ufh));
   1853 	fs = VFSTOUFS(mp)->um_fs;
   1854 	if (ufh.ufid_ino < ROOTINO ||
   1855 	    ufh.ufid_ino >= fs->fs_ncg * fs->fs_ipg)
   1856 		return (ESTALE);
   1857 	return (ufs_fhtovp(mp, &ufh, vpp));
   1858 }
   1859 
   1860 /*
   1861  * Vnode pointer to File handle
   1862  */
   1863 /* ARGSUSED */
   1864 int
   1865 ffs_vptofh(struct vnode *vp, struct fid *fhp, size_t *fh_size)
   1866 {
   1867 	struct inode *ip;
   1868 	struct ufid ufh;
   1869 
   1870 	if (*fh_size < sizeof(struct ufid)) {
   1871 		*fh_size = sizeof(struct ufid);
   1872 		return E2BIG;
   1873 	}
   1874 	ip = VTOI(vp);
   1875 	*fh_size = sizeof(struct ufid);
   1876 	memset(&ufh, 0, sizeof(ufh));
   1877 	ufh.ufid_len = sizeof(struct ufid);
   1878 	ufh.ufid_ino = ip->i_number;
   1879 	ufh.ufid_gen = ip->i_gen;
   1880 	memcpy(fhp, &ufh, sizeof(ufh));
   1881 	return (0);
   1882 }
   1883 
   1884 void
   1885 ffs_init(void)
   1886 {
   1887 	if (ffs_initcount++ > 0)
   1888 		return;
   1889 
   1890 	ffs_inode_cache = pool_cache_init(sizeof(struct inode), 0, 0, 0,
   1891 	    "ffsino", NULL, IPL_NONE, NULL, NULL, NULL);
   1892 	ffs_dinode1_cache = pool_cache_init(sizeof(struct ufs1_dinode), 0, 0, 0,
   1893 	    "ffsdino1", NULL, IPL_NONE, NULL, NULL, NULL);
   1894 	ffs_dinode2_cache = pool_cache_init(sizeof(struct ufs2_dinode), 0, 0, 0,
   1895 	    "ffsdino2", NULL, IPL_NONE, NULL, NULL, NULL);
   1896 	ufs_init();
   1897 }
   1898 
   1899 void
   1900 ffs_reinit(void)
   1901 {
   1902 
   1903 	ufs_reinit();
   1904 }
   1905 
   1906 void
   1907 ffs_done(void)
   1908 {
   1909 	if (--ffs_initcount > 0)
   1910 		return;
   1911 
   1912 	ufs_done();
   1913 	pool_cache_destroy(ffs_dinode2_cache);
   1914 	pool_cache_destroy(ffs_dinode1_cache);
   1915 	pool_cache_destroy(ffs_inode_cache);
   1916 }
   1917 
   1918 /*
   1919  * Write a superblock and associated information back to disk.
   1920  */
   1921 int
   1922 ffs_sbupdate(struct ufsmount *mp, int waitfor)
   1923 {
   1924 	struct fs *fs = mp->um_fs;
   1925 	struct buf *bp;
   1926 	int error = 0;
   1927 	u_int32_t saveflag;
   1928 
   1929 	error = ffs_getblk(mp->um_devvp,
   1930 	    fs->fs_sblockloc / DEV_BSIZE, FFS_NOBLK,
   1931 	    fs->fs_sbsize, false, &bp);
   1932 	if (error)
   1933 		return error;
   1934 	saveflag = fs->fs_flags & FS_INTERNAL;
   1935 	fs->fs_flags &= ~FS_INTERNAL;
   1936 
   1937 	memcpy(bp->b_data, fs, fs->fs_sbsize);
   1938 
   1939 	ffs_oldfscompat_write((struct fs *)bp->b_data, mp);
   1940 #ifdef FFS_EI
   1941 	if (mp->um_flags & UFS_NEEDSWAP)
   1942 		ffs_sb_swap((struct fs *)bp->b_data, (struct fs *)bp->b_data);
   1943 #endif
   1944 	fs->fs_flags |= saveflag;
   1945 
   1946 	if (waitfor == MNT_WAIT)
   1947 		error = bwrite(bp);
   1948 	else
   1949 		bawrite(bp);
   1950 	return (error);
   1951 }
   1952 
   1953 int
   1954 ffs_cgupdate(struct ufsmount *mp, int waitfor)
   1955 {
   1956 	struct fs *fs = mp->um_fs;
   1957 	struct buf *bp;
   1958 	int blks;
   1959 	void *space;
   1960 	int i, size, error = 0, allerror = 0;
   1961 
   1962 	allerror = ffs_sbupdate(mp, waitfor);
   1963 	blks = howmany(fs->fs_cssize, fs->fs_fsize);
   1964 	space = fs->fs_csp;
   1965 	for (i = 0; i < blks; i += fs->fs_frag) {
   1966 		size = fs->fs_bsize;
   1967 		if (i + fs->fs_frag > blks)
   1968 			size = (blks - i) * fs->fs_fsize;
   1969 		error = ffs_getblk(mp->um_devvp, fsbtodb(fs, fs->fs_csaddr + i),
   1970 		    FFS_NOBLK, size, false, &bp);
   1971 		if (error)
   1972 			break;
   1973 #ifdef FFS_EI
   1974 		if (mp->um_flags & UFS_NEEDSWAP)
   1975 			ffs_csum_swap((struct csum*)space,
   1976 			    (struct csum*)bp->b_data, size);
   1977 		else
   1978 #endif
   1979 			memcpy(bp->b_data, space, (u_int)size);
   1980 		space = (char *)space + size;
   1981 		if (waitfor == MNT_WAIT)
   1982 			error = bwrite(bp);
   1983 		else
   1984 			bawrite(bp);
   1985 	}
   1986 	if (!allerror && error)
   1987 		allerror = error;
   1988 	return (allerror);
   1989 }
   1990 
   1991 int
   1992 ffs_extattrctl(struct mount *mp, int cmd, struct vnode *vp,
   1993     int attrnamespace, const char *attrname)
   1994 {
   1995 #ifdef UFS_EXTATTR
   1996 	/*
   1997 	 * File-backed extended attributes are only supported on UFS1.
   1998 	 * UFS2 has native extended attributes.
   1999 	 */
   2000 	if (VFSTOUFS(mp)->um_fstype == UFS1)
   2001 		return (ufs_extattrctl(mp, cmd, vp, attrnamespace, attrname));
   2002 #endif
   2003 	return (vfs_stdextattrctl(mp, cmd, vp, attrnamespace, attrname));
   2004 }
   2005 
   2006 int
   2007 ffs_suspendctl(struct mount *mp, int cmd)
   2008 {
   2009 	int error;
   2010 	struct lwp *l = curlwp;
   2011 
   2012 	switch (cmd) {
   2013 	case SUSPEND_SUSPEND:
   2014 		if ((error = fstrans_setstate(mp, FSTRANS_SUSPENDING)) != 0)
   2015 			return error;
   2016 		error = ffs_sync(mp, MNT_WAIT, l->l_proc->p_cred);
   2017 		if (error == 0)
   2018 			error = fstrans_setstate(mp, FSTRANS_SUSPENDED);
   2019 #ifdef WAPBL
   2020 		if (error == 0 && mp->mnt_wapbl)
   2021 			error = wapbl_flush(mp->mnt_wapbl, 1);
   2022 #endif
   2023 		if (error != 0) {
   2024 			(void) fstrans_setstate(mp, FSTRANS_NORMAL);
   2025 			return error;
   2026 		}
   2027 		return 0;
   2028 
   2029 	case SUSPEND_RESUME:
   2030 		return fstrans_setstate(mp, FSTRANS_NORMAL);
   2031 
   2032 	default:
   2033 		return EINVAL;
   2034 	}
   2035 }
   2036 
   2037 /*
   2038  * Synch vnode for a mounted file system.  This is called for foreign
   2039  * vnodes, i.e. non-ffs.
   2040  */
   2041 static int
   2042 ffs_vfs_fsync(vnode_t *vp, int flags)
   2043 {
   2044 	int error, passes, skipmeta, i, pflags;
   2045 	buf_t *bp, *nbp;
   2046 #ifdef WAPBL
   2047 	struct mount *mp;
   2048 #endif
   2049 
   2050 	KASSERT(vp->v_type == VBLK);
   2051 	KASSERT(vp->v_specmountpoint != NULL);
   2052 
   2053 	/*
   2054 	 * Flush all dirty data associated with the vnode.
   2055 	 */
   2056 	pflags = PGO_ALLPAGES | PGO_CLEANIT;
   2057 	if ((flags & FSYNC_WAIT) != 0)
   2058 		pflags |= PGO_SYNCIO;
   2059 	mutex_enter(vp->v_interlock);
   2060 	error = VOP_PUTPAGES(vp, 0, 0, pflags);
   2061 	if (error)
   2062 		return error;
   2063 
   2064 #ifdef WAPBL
   2065 	mp = vp->v_specmountpoint;
   2066 	if (mp && mp->mnt_wapbl) {
   2067 		/*
   2068 		 * Don't bother writing out metadata if the syncer is
   2069 		 * making the request.  We will let the sync vnode
   2070 		 * write it out in a single burst through a call to
   2071 		 * VFS_SYNC().
   2072 		 */
   2073 		if ((flags & (FSYNC_DATAONLY | FSYNC_LAZY | FSYNC_NOLOG)) != 0)
   2074 			return 0;
   2075 
   2076 		/*
   2077 		 * Don't flush the log if the vnode being flushed
   2078 		 * contains no dirty buffers that could be in the log.
   2079 		 */
   2080 		if (!LIST_EMPTY(&vp->v_dirtyblkhd)) {
   2081 			error = wapbl_flush(mp->mnt_wapbl, 0);
   2082 			if (error)
   2083 				return error;
   2084 		}
   2085 
   2086 		if ((flags & FSYNC_WAIT) != 0) {
   2087 			mutex_enter(vp->v_interlock);
   2088 			while (vp->v_numoutput)
   2089 				cv_wait(&vp->v_cv, vp->v_interlock);
   2090 			mutex_exit(vp->v_interlock);
   2091 		}
   2092 
   2093 		return 0;
   2094 	}
   2095 #endif /* WAPBL */
   2096 
   2097 	/*
   2098 	 * Write out metadata for non-logging file systems. XXX This block
   2099 	 * should be simplified now that softdep is gone.
   2100 	 */
   2101 	passes = NIADDR + 1;
   2102 	skipmeta = 0;
   2103 	if (flags & FSYNC_WAIT)
   2104 		skipmeta = 1;
   2105 
   2106 loop:
   2107 	mutex_enter(&bufcache_lock);
   2108 	LIST_FOREACH(bp, &vp->v_dirtyblkhd, b_vnbufs) {
   2109 		bp->b_cflags &= ~BC_SCANNED;
   2110 	}
   2111 	for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
   2112 		nbp = LIST_NEXT(bp, b_vnbufs);
   2113 		if (bp->b_cflags & (BC_BUSY | BC_SCANNED))
   2114 			continue;
   2115 		if ((bp->b_oflags & BO_DELWRI) == 0)
   2116 			panic("ffs_fsync: not dirty");
   2117 		if (skipmeta && bp->b_lblkno < 0)
   2118 			continue;
   2119 		bp->b_cflags |= BC_BUSY | BC_VFLUSH | BC_SCANNED;
   2120 		mutex_exit(&bufcache_lock);
   2121 		/*
   2122 		 * On our final pass through, do all I/O synchronously
   2123 		 * so that we can find out if our flush is failing
   2124 		 * because of write errors.
   2125 		 */
   2126 		if (passes > 0 || !(flags & FSYNC_WAIT))
   2127 			(void) bawrite(bp);
   2128 		else if ((error = bwrite(bp)) != 0)
   2129 			return (error);
   2130 		/*
   2131 		 * Since we unlocked during the I/O, we need
   2132 		 * to start from a known point.
   2133 		 */
   2134 		mutex_enter(&bufcache_lock);
   2135 		nbp = LIST_FIRST(&vp->v_dirtyblkhd);
   2136 	}
   2137 	mutex_exit(&bufcache_lock);
   2138 	if (skipmeta) {
   2139 		skipmeta = 0;
   2140 		goto loop;
   2141 	}
   2142 
   2143 	if ((flags & FSYNC_WAIT) != 0) {
   2144 		mutex_enter(vp->v_interlock);
   2145 		while (vp->v_numoutput) {
   2146 			cv_wait(&vp->v_cv, vp->v_interlock);
   2147 		}
   2148 		mutex_exit(vp->v_interlock);
   2149 
   2150 		if (!LIST_EMPTY(&vp->v_dirtyblkhd)) {
   2151 			/*
   2152 			* Block devices associated with filesystems may
   2153 			* have new I/O requests posted for them even if
   2154 			* the vnode is locked, so no amount of trying will
   2155 			* get them clean. Thus we give block devices a
   2156 			* good effort, then just give up. For all other file
   2157 			* types, go around and try again until it is clean.
   2158 			*/
   2159 			if (passes > 0) {
   2160 				passes--;
   2161 				goto loop;
   2162 			}
   2163 #ifdef DIAGNOSTIC
   2164 			if (vp->v_type != VBLK)
   2165 				vprint("ffs_fsync: dirty", vp);
   2166 #endif
   2167 		}
   2168 	}
   2169 
   2170 	if (error == 0 && (flags & FSYNC_CACHE) != 0) {
   2171 		(void)VOP_IOCTL(vp, DIOCCACHESYNC, &i, FWRITE,
   2172 		    kauth_cred_get());
   2173 	}
   2174 
   2175 	return error;
   2176 }
   2177