Home | History | Annotate | Line # | Download | only in ffs
ffs_vfsops.c revision 1.119
      1 /*	$NetBSD: ffs_vfsops.c,v 1.119 2003/08/07 16:34:31 agc Exp $	*/
      2 
      3 /*
      4  * Copyright (c) 1989, 1991, 1993, 1994
      5  *	The Regents of the University of California.  All rights reserved.
      6  *
      7  * Redistribution and use in source and binary forms, with or without
      8  * modification, are permitted provided that the following conditions
      9  * are met:
     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.
     18  *
     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
     29  * SUCH DAMAGE.
     30  *
     31  *	@(#)ffs_vfsops.c	8.31 (Berkeley) 5/20/95
     32  */
     33 
     34 #include <sys/cdefs.h>
     35 __KERNEL_RCSID(0, "$NetBSD: ffs_vfsops.c,v 1.119 2003/08/07 16:34:31 agc Exp $");
     36 
     37 #if defined(_KERNEL_OPT)
     38 #include "opt_ffs.h"
     39 #include "opt_quota.h"
     40 #include "opt_compat_netbsd.h"
     41 #include "opt_softdep.h"
     42 #endif
     43 
     44 #include <sys/param.h>
     45 #include <sys/systm.h>
     46 #include <sys/namei.h>
     47 #include <sys/proc.h>
     48 #include <sys/kernel.h>
     49 #include <sys/vnode.h>
     50 #include <sys/socket.h>
     51 #include <sys/mount.h>
     52 #include <sys/buf.h>
     53 #include <sys/device.h>
     54 #include <sys/mbuf.h>
     55 #include <sys/file.h>
     56 #include <sys/disklabel.h>
     57 #include <sys/ioctl.h>
     58 #include <sys/errno.h>
     59 #include <sys/malloc.h>
     60 #include <sys/pool.h>
     61 #include <sys/lock.h>
     62 #include <sys/sysctl.h>
     63 #include <sys/conf.h>
     64 
     65 #include <miscfs/specfs/specdev.h>
     66 
     67 #include <ufs/ufs/quota.h>
     68 #include <ufs/ufs/ufsmount.h>
     69 #include <ufs/ufs/inode.h>
     70 #include <ufs/ufs/dir.h>
     71 #include <ufs/ufs/ufs_extern.h>
     72 #include <ufs/ufs/ufs_bswap.h>
     73 
     74 #include <ufs/ffs/fs.h>
     75 #include <ufs/ffs/ffs_extern.h>
     76 
     77 /* how many times ffs_init() was called */
     78 int ffs_initcount = 0;
     79 
     80 extern struct lock ufs_hashlock;
     81 
     82 extern const struct vnodeopv_desc ffs_vnodeop_opv_desc;
     83 extern const struct vnodeopv_desc ffs_specop_opv_desc;
     84 extern const struct vnodeopv_desc ffs_fifoop_opv_desc;
     85 
     86 const struct vnodeopv_desc * const ffs_vnodeopv_descs[] = {
     87 	&ffs_vnodeop_opv_desc,
     88 	&ffs_specop_opv_desc,
     89 	&ffs_fifoop_opv_desc,
     90 	NULL,
     91 };
     92 
     93 struct vfsops ffs_vfsops = {
     94 	MOUNT_FFS,
     95 	ffs_mount,
     96 	ufs_start,
     97 	ffs_unmount,
     98 	ufs_root,
     99 	ufs_quotactl,
    100 	ffs_statfs,
    101 	ffs_sync,
    102 	ffs_vget,
    103 	ffs_fhtovp,
    104 	ffs_vptofh,
    105 	ffs_init,
    106 	ffs_reinit,
    107 	ffs_done,
    108 	ffs_sysctl,
    109 	ffs_mountroot,
    110 	ufs_check_export,
    111 	ffs_vnodeopv_descs,
    112 };
    113 
    114 struct genfs_ops ffs_genfsops = {
    115 	ffs_gop_size,
    116 	ufs_gop_alloc,
    117 	genfs_gop_write,
    118 };
    119 
    120 struct pool ffs_inode_pool;
    121 struct pool ffs_dinode1_pool;
    122 struct pool ffs_dinode2_pool;
    123 
    124 static void ffs_oldfscompat_read(struct fs *, struct ufsmount *,
    125 				   daddr_t);
    126 static void ffs_oldfscompat_write(struct fs *, struct ufsmount *);
    127 
    128 /*
    129  * Called by main() when ffs is going to be mounted as root.
    130  */
    131 
    132 int
    133 ffs_mountroot()
    134 {
    135 	struct fs *fs;
    136 	struct mount *mp;
    137 	struct proc *p = curproc;	/* XXX */
    138 	struct ufsmount *ump;
    139 	int error;
    140 
    141 	if (root_device->dv_class != DV_DISK)
    142 		return (ENODEV);
    143 
    144 	/*
    145 	 * Get vnodes for rootdev.
    146 	 */
    147 	if (bdevvp(rootdev, &rootvp))
    148 		panic("ffs_mountroot: can't setup bdevvp's");
    149 
    150 	if ((error = vfs_rootmountalloc(MOUNT_FFS, "root_device", &mp))) {
    151 		vrele(rootvp);
    152 		return (error);
    153 	}
    154 	if ((error = ffs_mountfs(rootvp, mp, p)) != 0) {
    155 		mp->mnt_op->vfs_refcount--;
    156 		vfs_unbusy(mp);
    157 		free(mp, M_MOUNT);
    158 		vrele(rootvp);
    159 		return (error);
    160 	}
    161 	simple_lock(&mountlist_slock);
    162 	CIRCLEQ_INSERT_TAIL(&mountlist, mp, mnt_list);
    163 	simple_unlock(&mountlist_slock);
    164 	ump = VFSTOUFS(mp);
    165 	fs = ump->um_fs;
    166 	memset(fs->fs_fsmnt, 0, sizeof(fs->fs_fsmnt));
    167 	(void)copystr(mp->mnt_stat.f_mntonname, fs->fs_fsmnt, MNAMELEN - 1, 0);
    168 	(void)ffs_statfs(mp, &mp->mnt_stat, p);
    169 	vfs_unbusy(mp);
    170 	inittodr(fs->fs_time);
    171 	return (0);
    172 }
    173 
    174 /*
    175  * VFS Operations.
    176  *
    177  * mount system call
    178  */
    179 int
    180 ffs_mount(mp, path, data, ndp, p)
    181 	struct mount *mp;
    182 	const char *path;
    183 	void *data;
    184 	struct nameidata *ndp;
    185 	struct proc *p;
    186 {
    187 	struct vnode *devvp = NULL;
    188 	struct ufs_args args;
    189 	struct ufsmount *ump = NULL;
    190 	struct fs *fs;
    191 	int error, flags, update;
    192 	mode_t accessmode;
    193 
    194 	if (mp->mnt_flag & MNT_GETARGS) {
    195 		ump = VFSTOUFS(mp);
    196 		if (ump == NULL)
    197 			return EIO;
    198 		args.fspec = NULL;
    199 		vfs_showexport(mp, &args.export, &ump->um_export);
    200 		return copyout(&args, data, sizeof(args));
    201 	}
    202 	error = copyin(data, &args, sizeof (struct ufs_args));
    203 	if (error)
    204 		return (error);
    205 
    206 #if !defined(SOFTDEP)
    207 	mp->mnt_flag &= ~MNT_SOFTDEP;
    208 #endif
    209 
    210 	update = mp->mnt_flag & MNT_UPDATE;
    211 
    212 	/* Check arguments */
    213 	if (update) {
    214 		/* Use the extant mount */
    215 		ump = VFSTOUFS(mp);
    216 		devvp = ump->um_devvp;
    217 		if (args.fspec == NULL)
    218 			vref(devvp);
    219 	} else {
    220 		/* New mounts must have a filename for the device */
    221 		if (args.fspec == NULL)
    222 			return (EINVAL);
    223 	}
    224 
    225 	if (args.fspec != NULL) {
    226 		/*
    227 		 * Look up the name and verify that it's sane.
    228 		 */
    229 		NDINIT(ndp, LOOKUP, FOLLOW, UIO_USERSPACE, args.fspec, p);
    230 		if ((error = namei(ndp)) != 0)
    231 			return (error);
    232 		devvp = ndp->ni_vp;
    233 
    234 		if (!update) {
    235 			/*
    236 			 * Be sure this is a valid block device
    237 			 */
    238 			if (devvp->v_type != VBLK)
    239 				error = ENOTBLK;
    240 			else if (bdevsw_lookup(devvp->v_rdev) == NULL)
    241 				error = ENXIO;
    242 		} else {
    243 			/*
    244 			 * Be sure we're still naming the same device
    245 			 * used for our initial mount
    246 			 */
    247 			if (devvp != ump->um_devvp)
    248 				error = EINVAL;
    249 		}
    250 	}
    251 
    252 	/*
    253 	 * If mount by non-root, then verify that user has necessary
    254 	 * permissions on the device.
    255 	 */
    256 	if (error == 0 && p->p_ucred->cr_uid != 0) {
    257 		accessmode = VREAD;
    258 		if (update ?
    259 		    (mp->mnt_flag & MNT_WANTRDWR) != 0 :
    260 		    (mp->mnt_flag & MNT_RDONLY) == 0)
    261 			accessmode |= VWRITE;
    262 		vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
    263 		error = VOP_ACCESS(devvp, accessmode, p->p_ucred, p);
    264 		VOP_UNLOCK(devvp, 0);
    265 	}
    266 
    267 	if (error) {
    268 		vrele(devvp);
    269 		return (error);
    270 	}
    271 
    272 	if (!update) {
    273 		error = ffs_mountfs(devvp, mp, p);
    274 		if (error) {
    275 			vrele(devvp);
    276 			return (error);
    277 		}
    278 
    279 		ump = VFSTOUFS(mp);
    280 		fs = ump->um_fs;
    281 		if ((mp->mnt_flag & (MNT_SOFTDEP | MNT_ASYNC)) ==
    282 		    (MNT_SOFTDEP | MNT_ASYNC)) {
    283 			printf("%s fs uses soft updates, "
    284 			    "ignoring async mode\n",
    285 			    fs->fs_fsmnt);
    286 			mp->mnt_flag &= ~MNT_ASYNC;
    287 		}
    288 	} else {
    289 		/*
    290 		 * Update the mount.
    291 		 */
    292 
    293 		/*
    294 		 * The initial mount got a reference on this
    295 		 * device, so drop the one obtained via
    296 		 * namei(), above.
    297 		 */
    298 		vrele(devvp);
    299 
    300 		fs = ump->um_fs;
    301 		if (fs->fs_ronly == 0 && (mp->mnt_flag & MNT_RDONLY)) {
    302 			/*
    303 			 * Changing from r/w to r/o
    304 			 */
    305 			flags = WRITECLOSE;
    306 			if (mp->mnt_flag & MNT_FORCE)
    307 				flags |= FORCECLOSE;
    308 			if (mp->mnt_flag & MNT_SOFTDEP)
    309 				error = softdep_flushfiles(mp, flags, p);
    310 			else
    311 				error = ffs_flushfiles(mp, flags, p);
    312 			if (fs->fs_pendingblocks != 0 ||
    313 			    fs->fs_pendinginodes != 0) {
    314 				printf("%s: update error: blocks %" PRId64
    315 				       " files %d\n",
    316 				    fs->fs_fsmnt, fs->fs_pendingblocks,
    317 				    fs->fs_pendinginodes);
    318 				fs->fs_pendingblocks = 0;
    319 				fs->fs_pendinginodes = 0;
    320 			}
    321 			if (error == 0 &&
    322 			    ffs_cgupdate(ump, MNT_WAIT) == 0 &&
    323 			    fs->fs_clean & FS_WASCLEAN) {
    324 				if (mp->mnt_flag & MNT_SOFTDEP)
    325 					fs->fs_flags &= ~FS_DOSOFTDEP;
    326 				fs->fs_clean = FS_ISCLEAN;
    327 				(void) ffs_sbupdate(ump, MNT_WAIT);
    328 			}
    329 			if (error)
    330 				return (error);
    331 			fs->fs_ronly = 1;
    332 			fs->fs_fmod = 0;
    333 		}
    334 
    335 		/*
    336 		 * Flush soft dependencies if disabling it via an update
    337 		 * mount. This may leave some items to be processed,
    338 		 * so don't do this yet XXX.
    339 		 */
    340 		if ((fs->fs_flags & FS_DOSOFTDEP) &&
    341 		    !(mp->mnt_flag & MNT_SOFTDEP) && fs->fs_ronly == 0) {
    342 #ifdef notyet
    343 			flags = WRITECLOSE;
    344 			if (mp->mnt_flag & MNT_FORCE)
    345 				flags |= FORCECLOSE;
    346 			error = softdep_flushfiles(mp, flags, p);
    347 			if (error == 0 && ffs_cgupdate(ump, MNT_WAIT) == 0)
    348 				fs->fs_flags &= ~FS_DOSOFTDEP;
    349 				(void) ffs_sbupdate(ump, MNT_WAIT);
    350 #elif defined(SOFTDEP)
    351 			mp->mnt_flag |= MNT_SOFTDEP;
    352 #endif
    353 		}
    354 
    355 		/*
    356 		 * When upgrading to a softdep mount, we must first flush
    357 		 * all vnodes. (not done yet -- see above)
    358 		 */
    359 		if (!(fs->fs_flags & FS_DOSOFTDEP) &&
    360 		    (mp->mnt_flag & MNT_SOFTDEP) && fs->fs_ronly == 0) {
    361 #ifdef notyet
    362 			flags = WRITECLOSE;
    363 			if (mp->mnt_flag & MNT_FORCE)
    364 				flags |= FORCECLOSE;
    365 			error = ffs_flushfiles(mp, flags, p);
    366 #else
    367 			mp->mnt_flag &= ~MNT_SOFTDEP;
    368 #endif
    369 		}
    370 
    371 		if (mp->mnt_flag & MNT_RELOAD) {
    372 			error = ffs_reload(mp, p->p_ucred, p);
    373 			if (error)
    374 				return (error);
    375 		}
    376 
    377 		if (fs->fs_ronly && (mp->mnt_flag & MNT_WANTRDWR)) {
    378 			/*
    379 			 * Changing from read-only to read/write
    380 			 */
    381 			fs->fs_ronly = 0;
    382 			fs->fs_clean <<= 1;
    383 			fs->fs_fmod = 1;
    384 			if ((fs->fs_flags & FS_DOSOFTDEP)) {
    385 				error = softdep_mount(devvp, mp, fs,
    386 				    p->p_ucred);
    387 				if (error)
    388 					return (error);
    389 			}
    390 		}
    391 		if (args.fspec == 0) {
    392 			/*
    393 			 * Process export requests.
    394 			 */
    395 			return (vfs_export(mp, &ump->um_export, &args.export));
    396 		}
    397 		if ((mp->mnt_flag & (MNT_SOFTDEP | MNT_ASYNC)) ==
    398 		    (MNT_SOFTDEP | MNT_ASYNC)) {
    399 			printf("%s fs uses soft updates, ignoring async mode\n",
    400 			    fs->fs_fsmnt);
    401 			mp->mnt_flag &= ~MNT_ASYNC;
    402 		}
    403 	}
    404 
    405 	error = set_statfs_info(path, UIO_USERSPACE, args.fspec,
    406 	    UIO_USERSPACE, mp, p);
    407 	if (error == 0)
    408 		(void)strncpy(fs->fs_fsmnt, mp->mnt_stat.f_mntonname,
    409 		    sizeof(fs->fs_fsmnt));
    410 	if (mp->mnt_flag & MNT_SOFTDEP)
    411 		fs->fs_flags |= FS_DOSOFTDEP;
    412 	else
    413 		fs->fs_flags &= ~FS_DOSOFTDEP;
    414 	if (fs->fs_fmod != 0) {	/* XXX */
    415 		fs->fs_fmod = 0;
    416 		if (fs->fs_clean & FS_WASCLEAN)
    417 			fs->fs_time = time.tv_sec;
    418 		else {
    419 			printf("%s: file system not clean (fs_clean=%x); please fsck(8)\n",
    420 			    mp->mnt_stat.f_mntfromname, fs->fs_clean);
    421 			printf("%s: lost blocks %" PRId64 " files %d\n",
    422 			    mp->mnt_stat.f_mntfromname, fs->fs_pendingblocks,
    423 			    fs->fs_pendinginodes);
    424 		}
    425 		(void) ffs_cgupdate(ump, MNT_WAIT);
    426 	}
    427 	return error;
    428 }
    429 
    430 /*
    431  * Reload all incore data for a filesystem (used after running fsck on
    432  * the root filesystem and finding things to fix). The filesystem must
    433  * be mounted read-only.
    434  *
    435  * Things to do to update the mount:
    436  *	1) invalidate all cached meta-data.
    437  *	2) re-read superblock from disk.
    438  *	3) re-read summary information from disk.
    439  *	4) invalidate all inactive vnodes.
    440  *	5) invalidate all cached file data.
    441  *	6) re-read inode data for all active vnodes.
    442  */
    443 int
    444 ffs_reload(mountp, cred, p)
    445 	struct mount *mountp;
    446 	struct ucred *cred;
    447 	struct proc *p;
    448 {
    449 	struct vnode *vp, *nvp, *devvp;
    450 	struct inode *ip;
    451 	void *space;
    452 	struct buf *bp;
    453 	struct fs *fs, *newfs;
    454 	struct partinfo dpart;
    455 	daddr_t sblockloc;
    456 	int i, blks, size, error;
    457 	int32_t *lp;
    458 	struct ufsmount *ump;
    459 
    460 	if ((mountp->mnt_flag & MNT_RDONLY) == 0)
    461 		return (EINVAL);
    462 
    463 	ump = VFSTOUFS(mountp);
    464 	/*
    465 	 * Step 1: invalidate all cached meta-data.
    466 	 */
    467 	devvp = ump->um_devvp;
    468 	vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
    469 	error = vinvalbuf(devvp, 0, cred, p, 0, 0);
    470 	VOP_UNLOCK(devvp, 0);
    471 	if (error)
    472 		panic("ffs_reload: dirty1");
    473 	/*
    474 	 * Step 2: re-read superblock from disk.
    475 	 */
    476 	fs = ump->um_fs;
    477 	if (VOP_IOCTL(devvp, DIOCGPART, &dpart, FREAD, NOCRED, p) != 0)
    478 		size = DEV_BSIZE;
    479 	else
    480 		size = dpart.disklab->d_secsize;
    481 	error = bread(devvp, fs->fs_sblockloc / size, fs->fs_sbsize,
    482 		      NOCRED, &bp);
    483 	if (error) {
    484 		brelse(bp);
    485 		return (error);
    486 	}
    487 	newfs = malloc(fs->fs_sbsize, M_UFSMNT, M_WAITOK);
    488 	memcpy(newfs, bp->b_data, fs->fs_sbsize);
    489 #ifdef SUPPORT_42POSTBLFMT_WRITE
    490 	if (fs->fs_old_postblformat == FS_42POSTBLFMT)
    491 		memcpy(ump->um_opostsave, &newfs->fs_old_postbl_start, 256);
    492 #endif
    493 #ifdef FFS_EI
    494 	if (ump->um_flags & UFS_NEEDSWAP) {
    495 		ffs_sb_swap((struct fs*)bp->b_data, newfs);
    496 		fs->fs_flags |= FS_SWAPPED;
    497 	}
    498 #endif
    499 	if ((newfs->fs_magic != FS_UFS1_MAGIC &&
    500 	     newfs->fs_magic != FS_UFS2_MAGIC)||
    501 	     newfs->fs_bsize > MAXBSIZE ||
    502 	     newfs->fs_bsize < sizeof(struct fs)) {
    503 		brelse(bp);
    504 		free(newfs, M_UFSMNT);
    505 		return (EIO);		/* XXX needs translation */
    506 	}
    507 	/*
    508 	 * Copy pointer fields back into superblock before copying in	XXX
    509 	 * new superblock. These should really be in the ufsmount.	XXX
    510 	 * Note that important parameters (eg fs_ncg) are unchanged.
    511 	 */
    512 	newfs->fs_csp = fs->fs_csp;
    513 	newfs->fs_maxcluster = fs->fs_maxcluster;
    514 	newfs->fs_contigdirs = fs->fs_contigdirs;
    515 	newfs->fs_ronly = fs->fs_ronly;
    516 	newfs->fs_active = fs->fs_active;
    517 	sblockloc = fs->fs_sblockloc;
    518 	memcpy(fs, newfs, (u_int)fs->fs_sbsize);
    519 	brelse(bp);
    520 	free(newfs, M_UFSMNT);
    521 
    522 	/* Recheck for apple UFS filesystem */
    523 	VFSTOUFS(mountp)->um_flags &= ~UFS_ISAPPLEUFS;
    524 	/* First check to see if this is tagged as an Apple UFS filesystem
    525 	 * in the disklabel
    526 	 */
    527 	if ((VOP_IOCTL(devvp, DIOCGPART, &dpart, FREAD, cred, p) == 0) &&
    528 		(dpart.part->p_fstype == FS_APPLEUFS)) {
    529 		VFSTOUFS(mountp)->um_flags |= UFS_ISAPPLEUFS;
    530 	}
    531 #ifdef APPLE_UFS
    532 	else {
    533 		/* Manually look for an apple ufs label, and if a valid one
    534 		 * is found, then treat it like an Apple UFS filesystem anyway
    535 		 */
    536 		error = bread(devvp, (daddr_t)(APPLEUFS_LABEL_OFFSET / size),
    537 			APPLEUFS_LABEL_SIZE, cred, &bp);
    538 		if (error) {
    539 			brelse(bp);
    540 			return (error);
    541 		}
    542 		error = ffs_appleufs_validate(fs->fs_fsmnt,
    543 			(struct appleufslabel *)bp->b_data,NULL);
    544 		if (error == 0) {
    545 			VFSTOUFS(mountp)->um_flags |= UFS_ISAPPLEUFS;
    546 		}
    547 		brelse(bp);
    548 		bp = NULL;
    549 	}
    550 #else
    551 	if (VFSTOUFS(mountp)->um_flags & UFS_ISAPPLEUFS)
    552 		return (EIO);
    553 #endif
    554 
    555 	mountp->mnt_maxsymlinklen = fs->fs_maxsymlinklen;
    556 	if (UFS_MPISAPPLEUFS(mountp)) {
    557 		/* see comment about NeXT below */
    558 		mountp->mnt_maxsymlinklen = APPLEUFS_MAXSYMLINKLEN;
    559 	}
    560 	ffs_oldfscompat_read(fs, VFSTOUFS(mountp), fs->fs_sblockloc);
    561 	if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
    562 		fs->fs_pendingblocks = 0;
    563 		fs->fs_pendinginodes = 0;
    564 	}
    565 
    566 	ffs_statfs(mountp, &mountp->mnt_stat, p);
    567 	/*
    568 	 * Step 3: re-read summary information from disk.
    569 	 */
    570 	blks = howmany(fs->fs_cssize, fs->fs_fsize);
    571 	space = fs->fs_csp;
    572 	for (i = 0; i < blks; i += fs->fs_frag) {
    573 		size = fs->fs_bsize;
    574 		if (i + fs->fs_frag > blks)
    575 			size = (blks - i) * fs->fs_fsize;
    576 		error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), size,
    577 			      NOCRED, &bp);
    578 		if (error) {
    579 			brelse(bp);
    580 			return (error);
    581 		}
    582 #ifdef FFS_EI
    583 		if (UFS_FSNEEDSWAP(fs))
    584 			ffs_csum_swap((struct csum *)bp->b_data,
    585 			    (struct csum *)space, size);
    586 		else
    587 #endif
    588 			memcpy(space, bp->b_data, (size_t)size);
    589 		space = (char *)space + size;
    590 		brelse(bp);
    591 	}
    592 	if ((fs->fs_flags & FS_DOSOFTDEP))
    593 		softdep_mount(devvp, mountp, fs, cred);
    594 	/*
    595 	 * We no longer know anything about clusters per cylinder group.
    596 	 */
    597 	if (fs->fs_contigsumsize > 0) {
    598 		lp = fs->fs_maxcluster;
    599 		for (i = 0; i < fs->fs_ncg; i++)
    600 			*lp++ = fs->fs_contigsumsize;
    601 	}
    602 
    603 loop:
    604 	simple_lock(&mntvnode_slock);
    605 	for (vp = mountp->mnt_vnodelist.lh_first; vp != NULL; vp = nvp) {
    606 		if (vp->v_mount != mountp) {
    607 			simple_unlock(&mntvnode_slock);
    608 			goto loop;
    609 		}
    610 		nvp = vp->v_mntvnodes.le_next;
    611 		/*
    612 		 * Step 4: invalidate all inactive vnodes.
    613 		 */
    614 		if (vrecycle(vp, &mntvnode_slock, p))
    615 			goto loop;
    616 		/*
    617 		 * Step 5: invalidate all cached file data.
    618 		 */
    619 		simple_lock(&vp->v_interlock);
    620 		simple_unlock(&mntvnode_slock);
    621 		if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK))
    622 			goto loop;
    623 		if (vinvalbuf(vp, 0, cred, p, 0, 0))
    624 			panic("ffs_reload: dirty2");
    625 		/*
    626 		 * Step 6: re-read inode data for all active vnodes.
    627 		 */
    628 		ip = VTOI(vp);
    629 		error = bread(devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
    630 			      (int)fs->fs_bsize, NOCRED, &bp);
    631 		if (error) {
    632 			brelse(bp);
    633 			vput(vp);
    634 			return (error);
    635 		}
    636 		ffs_load_inode(bp, ip, fs, ip->i_number);
    637 		ip->i_ffs_effnlink = ip->i_nlink;
    638 		brelse(bp);
    639 		vput(vp);
    640 		simple_lock(&mntvnode_slock);
    641 	}
    642 	simple_unlock(&mntvnode_slock);
    643 	return (0);
    644 }
    645 
    646 /*
    647  * Possible superblock locations ordered from most to least likely.
    648  */
    649 static int sblock_try[] = SBLOCKSEARCH;
    650 
    651 /*
    652  * Common code for mount and mountroot
    653  */
    654 int
    655 ffs_mountfs(devvp, mp, p)
    656 	struct vnode *devvp;
    657 	struct mount *mp;
    658 	struct proc *p;
    659 {
    660 	struct ufsmount *ump;
    661 	struct buf *bp;
    662 	struct fs *fs;
    663 	dev_t dev;
    664 	struct partinfo dpart;
    665 	void *space;
    666 	daddr_t sblockloc, fsblockloc;
    667 	int blks, fstype;
    668 	int error, i, size, ronly;
    669 #ifdef FFS_EI
    670 	int needswap = 0;		/* keep gcc happy */
    671 #endif
    672 	int32_t *lp;
    673 	struct ucred *cred;
    674 	u_int32_t sbsize = 8192;	/* keep gcc happy*/
    675 
    676 	dev = devvp->v_rdev;
    677 	cred = p ? p->p_ucred : NOCRED;
    678 	/*
    679 	 * Disallow multiple mounts of the same device.
    680 	 * Disallow mounting of a device that is currently in use
    681 	 * (except for root, which might share swap device for miniroot).
    682 	 * Flush out any old buffers remaining from a previous use.
    683 	 */
    684 	if ((error = vfs_mountedon(devvp)) != 0)
    685 		return (error);
    686 	if (vcount(devvp) > 1 && devvp != rootvp)
    687 		return (EBUSY);
    688 	vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
    689 	error = vinvalbuf(devvp, V_SAVE, cred, p, 0, 0);
    690 	VOP_UNLOCK(devvp, 0);
    691 	if (error)
    692 		return (error);
    693 
    694 	ronly = (mp->mnt_flag & MNT_RDONLY) != 0;
    695 	error = VOP_OPEN(devvp, ronly ? FREAD : FREAD|FWRITE, FSCRED, p);
    696 	if (error)
    697 		return (error);
    698 	if (VOP_IOCTL(devvp, DIOCGPART, &dpart, FREAD, cred, p) != 0)
    699 		size = DEV_BSIZE;
    700 	else
    701 		size = dpart.disklab->d_secsize;
    702 
    703 	bp = NULL;
    704 	ump = NULL;
    705 	fs = NULL;
    706 	fsblockloc = sblockloc = 0;
    707 	fstype = 0;
    708 
    709 	/*
    710 	 * Try reading the superblock in each of its possible locations.		 */
    711 	for (i = 0; sblock_try[i] != -1; i++) {
    712 		error = bread(devvp, sblock_try[i] / size, SBLOCKSIZE, cred,
    713 			      &bp);
    714 		if (error)
    715 			goto out;
    716 		fs = (struct fs*)bp->b_data;
    717 		fsblockloc = sblockloc = sblock_try[i];
    718 		if (fs->fs_magic == FS_UFS1_MAGIC) {
    719 			sbsize = fs->fs_sbsize;
    720 			fstype = UFS1;
    721 #ifdef FFS_EI
    722 			needswap = 0;
    723 		} else if (fs->fs_magic == bswap32(FS_UFS1_MAGIC)) {
    724 			sbsize = bswap32(fs->fs_sbsize);
    725 			fstype = UFS1;
    726 			needswap = 1;
    727 #endif
    728 		} else if (fs->fs_magic == FS_UFS2_MAGIC) {
    729 			sbsize = fs->fs_sbsize;
    730 			fstype = UFS2;
    731 			fsblockloc = fs->fs_sblockloc;
    732 #ifdef FFS_EI
    733 			needswap = 0;
    734 		} else if (fs->fs_magic == bswap32(FS_UFS2_MAGIC)) {
    735 			sbsize = bswap32(fs->fs_sbsize);
    736 			fstype = UFS2;
    737 			fsblockloc = bswap64(fs->fs_sblockloc);
    738 			needswap = 1;
    739 #endif
    740 		} else
    741 			goto next_sblock;
    742 
    743 		if ((fsblockloc == sblockloc ||
    744 		     (fs->fs_old_flags & FS_FLAGS_UPDATED) == 0)
    745 		    && sbsize <= MAXBSIZE && sbsize >= sizeof(struct fs))
    746 			break;
    747 
    748 next_sblock:
    749 		bp->b_flags |= B_NOCACHE;
    750 		brelse(bp);
    751 		bp = NULL;
    752 	}
    753 
    754 	if (sblock_try[i] == -1) {
    755 		error = EINVAL;
    756 		goto out;
    757 	}
    758 
    759 	fs = malloc((u_long)sbsize, M_UFSMNT, M_WAITOK);
    760 	memcpy(fs, bp->b_data, sbsize);
    761 
    762 	ump = malloc(sizeof *ump, M_UFSMNT, M_WAITOK);
    763 	memset(ump, 0, sizeof *ump);
    764 	ump->um_fs = fs;
    765 
    766 	if (fs->fs_old_postblformat == FS_42POSTBLFMT) {
    767 #ifdef SUPPORT_FS_42POSTBLFMT_WRITE
    768 		ump->um_opostsave = malloc(256, M_UFSMNT, M_WAITOK);
    769 		memcpy(ump->um_opostsave, &fs->fs_old_postbl_start, 256);
    770 #else
    771 		if ((mp->mnt_flag & MNT_RDONLY) == 0) {
    772 			error = EROFS;
    773 			goto out2;
    774 		}
    775 #endif
    776 	}
    777 
    778 #ifdef FFS_EI
    779 	if (needswap) {
    780 		ffs_sb_swap((struct fs*)bp->b_data, fs);
    781 		fs->fs_flags |= FS_SWAPPED;
    782 	}
    783 #endif
    784 
    785 	if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
    786 		fs->fs_pendingblocks = 0;
    787 		fs->fs_pendinginodes = 0;
    788 	}
    789 
    790 	ump->um_fstype = fstype;
    791 	if (fs->fs_sbsize < SBLOCKSIZE)
    792 		bp->b_flags |= B_INVAL;
    793 	brelse(bp);
    794 	bp = NULL;
    795 
    796 	ffs_oldfscompat_read(fs, ump, sblockloc);
    797 
    798 	/* First check to see if this is tagged as an Apple UFS filesystem
    799 	 * in the disklabel
    800 	 */
    801 	if ((VOP_IOCTL(devvp, DIOCGPART, &dpart, FREAD, cred, p) == 0) &&
    802 		(dpart.part->p_fstype == FS_APPLEUFS)) {
    803 		ump->um_flags |= UFS_ISAPPLEUFS;
    804 	}
    805 #ifdef APPLE_UFS
    806 	else {
    807 		/* Manually look for an apple ufs label, and if a valid one
    808 		 * is found, then treat it like an Apple UFS filesystem anyway
    809 		 */
    810 		error = bread(devvp, (daddr_t)(APPLEUFS_LABEL_OFFSET / size),
    811 			APPLEUFS_LABEL_SIZE, cred, &bp);
    812 		if (error)
    813 			goto out;
    814 		error = ffs_appleufs_validate(fs->fs_fsmnt,
    815 			(struct appleufslabel *)bp->b_data,NULL);
    816 		if (error == 0) {
    817 			ump->um_flags |= UFS_ISAPPLEUFS;
    818 		}
    819 		brelse(bp);
    820 		bp = NULL;
    821 	}
    822 #else
    823 	if (ump->um_flags & UFS_ISAPPLEUFS) {
    824 		error = EINVAL;
    825 		goto out;
    826 	}
    827 #endif
    828 
    829 	/*
    830 	 * verify that we can access the last block in the fs
    831 	 * if we're mounting read/write.
    832 	 */
    833 
    834 	if (!ronly) {
    835 		error = bread(devvp, fsbtodb(fs, fs->fs_size - 1), fs->fs_fsize,
    836 		    cred, &bp);
    837 		if (bp->b_bcount != fs->fs_fsize)
    838 			error = EINVAL;
    839 		bp->b_flags |= B_INVAL;
    840 		if (error)
    841 			goto out;
    842 		brelse(bp);
    843 		bp = NULL;
    844 	}
    845 
    846 	fs->fs_ronly = ronly;
    847 	if (ronly == 0) {
    848 		fs->fs_clean <<= 1;
    849 		fs->fs_fmod = 1;
    850 	}
    851 	size = fs->fs_cssize;
    852 	blks = howmany(size, fs->fs_fsize);
    853 	if (fs->fs_contigsumsize > 0)
    854 		size += fs->fs_ncg * sizeof(int32_t);
    855 	size += fs->fs_ncg * sizeof(*fs->fs_contigdirs);
    856 	space = malloc((u_long)size, M_UFSMNT, M_WAITOK);
    857 	fs->fs_csp = space;
    858 	for (i = 0; i < blks; i += fs->fs_frag) {
    859 		size = fs->fs_bsize;
    860 		if (i + fs->fs_frag > blks)
    861 			size = (blks - i) * fs->fs_fsize;
    862 		error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), size,
    863 			      cred, &bp);
    864 		if (error) {
    865 			free(fs->fs_csp, M_UFSMNT);
    866 			goto out2;
    867 		}
    868 #ifdef FFS_EI
    869 		if (needswap)
    870 			ffs_csum_swap((struct csum *)bp->b_data,
    871 				(struct csum *)space, size);
    872 		else
    873 #endif
    874 			memcpy(space, bp->b_data, (u_int)size);
    875 
    876 		space = (char *)space + size;
    877 		brelse(bp);
    878 		bp = NULL;
    879 	}
    880 	if (fs->fs_contigsumsize > 0) {
    881 		fs->fs_maxcluster = lp = space;
    882 		for (i = 0; i < fs->fs_ncg; i++)
    883 			*lp++ = fs->fs_contigsumsize;
    884 		space = lp;
    885 	}
    886 	size = fs->fs_ncg * sizeof(*fs->fs_contigdirs);
    887 	fs->fs_contigdirs = space;
    888 	space = (char *)space + size;
    889 	memset(fs->fs_contigdirs, 0, size);
    890 		/* Compatibility for old filesystems - XXX */
    891 	if (fs->fs_avgfilesize <= 0)
    892 		fs->fs_avgfilesize = AVFILESIZ;
    893 	if (fs->fs_avgfpdir <= 0)
    894 		fs->fs_avgfpdir = AFPDIR;
    895 	mp->mnt_data = ump;
    896 	mp->mnt_stat.f_fsid.val[0] = (long)dev;
    897 	mp->mnt_stat.f_fsid.val[1] = makefstype(MOUNT_FFS);
    898 	mp->mnt_maxsymlinklen = fs->fs_maxsymlinklen;
    899 	if (UFS_MPISAPPLEUFS(mp)) {
    900 		/* NeXT used to keep short symlinks in the inode even
    901 		 * when using FS_42INODEFMT.  In that case fs->fs_maxsymlinklen
    902 		 * is probably -1, but we still need to be able to identify
    903 		 * short symlinks.
    904 		 */
    905 		mp->mnt_maxsymlinklen = APPLEUFS_MAXSYMLINKLEN;
    906 	}
    907 	mp->mnt_fs_bshift = fs->fs_bshift;
    908 	mp->mnt_dev_bshift = DEV_BSHIFT;	/* XXX */
    909 	mp->mnt_flag |= MNT_LOCAL;
    910 #ifdef FFS_EI
    911 	if (needswap)
    912 		ump->um_flags |= UFS_NEEDSWAP;
    913 #endif
    914 	ump->um_mountp = mp;
    915 	ump->um_dev = dev;
    916 	ump->um_devvp = devvp;
    917 	ump->um_nindir = fs->fs_nindir;
    918 	ump->um_lognindir = ffs(fs->fs_nindir) - 1;
    919 	ump->um_bptrtodb = fs->fs_fsbtodb;
    920 	ump->um_seqinc = fs->fs_frag;
    921 	for (i = 0; i < MAXQUOTAS; i++)
    922 		ump->um_quotas[i] = NULLVP;
    923 	devvp->v_specmountpoint = mp;
    924 	if (ronly == 0 && (fs->fs_flags & FS_DOSOFTDEP)) {
    925 		error = softdep_mount(devvp, mp, fs, cred);
    926 		if (error) {
    927 			free(fs->fs_csp, M_UFSMNT);
    928 			goto out;
    929 		}
    930 	}
    931 	return (0);
    932 out2:
    933 	free(fs, M_UFSMNT);
    934 out:
    935 	devvp->v_specmountpoint = NULL;
    936 	if (bp)
    937 		brelse(bp);
    938 	vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
    939 	(void)VOP_CLOSE(devvp, ronly ? FREAD : FREAD|FWRITE, cred, p);
    940 	VOP_UNLOCK(devvp, 0);
    941 	if (ump) {
    942 		free(ump, M_UFSMNT);
    943 		mp->mnt_data = NULL;
    944 	}
    945 	return (error);
    946 }
    947 
    948 /*
    949  * Sanity checks for loading old filesystem superblocks.
    950  * See ffs_oldfscompat_write below for unwound actions.
    951  *
    952  * XXX - Parts get retired eventually.
    953  * Unfortunately new bits get added.
    954  */
    955 static void
    956 ffs_oldfscompat_read(fs, ump, sblockloc)
    957 	struct fs *fs;
    958 	struct ufsmount *ump;
    959 	daddr_t sblockloc;
    960 {
    961 	off_t maxfilesize;
    962 
    963 	if (fs->fs_magic != FS_UFS1_MAGIC)
    964 		return;
    965 
    966 	/*
    967 	 * If not yet done, update fs_flags location and value of fs_sblockloc.
    968 	 */
    969 	if ((fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) {
    970 		fs->fs_flags = fs->fs_old_flags;
    971 		fs->fs_old_flags |= FS_FLAGS_UPDATED;
    972 		fs->fs_sblockloc = sblockloc;
    973 	}
    974 
    975 	/*
    976 	 * If the new fields haven't been set yet, or if the filesystem
    977 	 * was mounted and modified by an old kernel, use the old csum
    978 	 * totals.
    979 	 */
    980 	if (fs->fs_maxbsize != fs->fs_bsize || fs->fs_time < fs->fs_old_time) {
    981 		fs->fs_cstotal.cs_ndir = fs->fs_old_cstotal.cs_ndir;
    982 		fs->fs_cstotal.cs_nbfree = fs->fs_old_cstotal.cs_nbfree;
    983 		fs->fs_cstotal.cs_nifree = fs->fs_old_cstotal.cs_nifree;
    984 		fs->fs_cstotal.cs_nffree = fs->fs_old_cstotal.cs_nffree;
    985 	}
    986 
    987 	/*
    988 	 * If not yet done, update UFS1 superblock with new wider fields.
    989 	 */
    990 	if (fs->fs_maxbsize != fs->fs_bsize) {
    991 		fs->fs_maxbsize = fs->fs_bsize;
    992 		fs->fs_time = fs->fs_old_time;
    993 		fs->fs_size = fs->fs_old_size;
    994 		fs->fs_dsize = fs->fs_old_dsize;
    995 		fs->fs_csaddr = fs->fs_old_csaddr;
    996 	}
    997 
    998 	if (fs->fs_old_inodefmt < FS_44INODEFMT) {
    999 		fs->fs_maxfilesize = (u_quad_t) 1LL << 39;
   1000 		fs->fs_qbmask = ~fs->fs_bmask;
   1001 		fs->fs_qfmask = ~fs->fs_fmask;
   1002 	}
   1003 
   1004 	ump->um_savedmaxfilesize = fs->fs_maxfilesize;
   1005 	maxfilesize = (u_int64_t)0x80000000 * fs->fs_bsize - 1;
   1006 	if (fs->fs_maxfilesize > maxfilesize)
   1007 		fs->fs_maxfilesize = maxfilesize;
   1008 
   1009 	/* Compatibility for old filesystems */
   1010 	if (fs->fs_avgfilesize <= 0)
   1011 		fs->fs_avgfilesize = AVFILESIZ;
   1012 	if (fs->fs_avgfpdir <= 0)
   1013 		fs->fs_avgfpdir = AFPDIR;
   1014 #if 0
   1015 	if (bigcgs) {
   1016 		fs->fs_save_cgsize = fs->fs_cgsize;
   1017 		fs->fs_cgsize = fs->fs_bsize;
   1018 	}
   1019 #endif
   1020 }
   1021 
   1022 /*
   1023  * Unwinding superblock updates for old filesystems.
   1024  * See ffs_oldfscompat_read above for details.
   1025  *
   1026  * XXX - Parts get retired eventually.
   1027  * Unfortunately new bits get added.
   1028  */
   1029 static void
   1030 ffs_oldfscompat_write(fs, ump)
   1031 	struct fs *fs;
   1032 	struct ufsmount *ump;
   1033 {
   1034 	if (fs->fs_magic != FS_UFS1_MAGIC)
   1035 		return;
   1036 
   1037 	/*
   1038 	 * OS X somehow still seems to use this field and panic.
   1039 	 * Just set it to zero.
   1040 	 */
   1041 	if (ump->um_flags & UFS_ISAPPLEUFS)
   1042 		fs->fs_old_nrpos = 0;
   1043 	/*
   1044 	 * Copy back UFS2 updated fields that UFS1 inspects.
   1045 	 */
   1046 
   1047 	fs->fs_old_time = fs->fs_time;
   1048 	fs->fs_old_cstotal.cs_ndir = fs->fs_cstotal.cs_ndir;
   1049 	fs->fs_old_cstotal.cs_nbfree = fs->fs_cstotal.cs_nbfree;
   1050 	fs->fs_old_cstotal.cs_nifree = fs->fs_cstotal.cs_nifree;
   1051 	fs->fs_old_cstotal.cs_nffree = fs->fs_cstotal.cs_nffree;
   1052 	fs->fs_maxfilesize = ump->um_savedmaxfilesize;
   1053 
   1054 #if 0
   1055 	if (bigcgs) {
   1056 		fs->fs_cgsize = fs->fs_save_cgsize;
   1057 		fs->fs_save_cgsize = 0;
   1058 	}
   1059 #endif
   1060 }
   1061 
   1062 /*
   1063  * unmount system call
   1064  */
   1065 int
   1066 ffs_unmount(mp, mntflags, p)
   1067 	struct mount *mp;
   1068 	int mntflags;
   1069 	struct proc *p;
   1070 {
   1071 	struct ufsmount *ump;
   1072 	struct fs *fs;
   1073 	int error, flags, penderr;
   1074 
   1075 	penderr = 0;
   1076 	flags = 0;
   1077 	if (mntflags & MNT_FORCE)
   1078 		flags |= FORCECLOSE;
   1079 	if (mp->mnt_flag & MNT_SOFTDEP) {
   1080 		if ((error = softdep_flushfiles(mp, flags, p)) != 0)
   1081 			return (error);
   1082 	} else {
   1083 		if ((error = ffs_flushfiles(mp, flags, p)) != 0)
   1084 			return (error);
   1085 	}
   1086 	ump = VFSTOUFS(mp);
   1087 	fs = ump->um_fs;
   1088 	if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
   1089 		printf("%s: unmount pending error: blocks %" PRId64
   1090 		       " files %d\n",
   1091 		    fs->fs_fsmnt, fs->fs_pendingblocks, fs->fs_pendinginodes);
   1092 		fs->fs_pendingblocks = 0;
   1093 		fs->fs_pendinginodes = 0;
   1094 		penderr = 1;
   1095 	}
   1096 	if (fs->fs_ronly == 0 &&
   1097 	    ffs_cgupdate(ump, MNT_WAIT) == 0 &&
   1098 	    fs->fs_clean & FS_WASCLEAN) {
   1099 		/*
   1100 		 * XXXX don't mark fs clean in the case of softdep
   1101 		 * pending block errors, until they are fixed.
   1102 		 */
   1103 		if (penderr == 0) {
   1104 			if (mp->mnt_flag & MNT_SOFTDEP)
   1105 				fs->fs_flags &= ~FS_DOSOFTDEP;
   1106 			fs->fs_clean = FS_ISCLEAN;
   1107 		}
   1108 		fs->fs_fmod = 0;
   1109 		(void) ffs_sbupdate(ump, MNT_WAIT);
   1110 	}
   1111 	if (ump->um_devvp->v_type != VBAD)
   1112 		ump->um_devvp->v_specmountpoint = NULL;
   1113 	vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY);
   1114 	error = VOP_CLOSE(ump->um_devvp, fs->fs_ronly ? FREAD : FREAD|FWRITE,
   1115 		NOCRED, p);
   1116 	vput(ump->um_devvp);
   1117 	free(fs->fs_csp, M_UFSMNT);
   1118 	free(fs, M_UFSMNT);
   1119 #ifdef SUPPORT_FS_42POSTBLFMT_WRITE
   1120 	if (ump->um_opostsave != NULL)
   1121 		free(ump->um_opostsave, M_UFSMNT);
   1122 #endif
   1123 	free(ump, M_UFSMNT);
   1124 	mp->mnt_data = NULL;
   1125 	mp->mnt_flag &= ~MNT_LOCAL;
   1126 	return (error);
   1127 }
   1128 
   1129 /*
   1130  * Flush out all the files in a filesystem.
   1131  */
   1132 int
   1133 ffs_flushfiles(mp, flags, p)
   1134 	struct mount *mp;
   1135 	int flags;
   1136 	struct proc *p;
   1137 {
   1138 	extern int doforce;
   1139 	struct ufsmount *ump;
   1140 	int error;
   1141 
   1142 	if (!doforce)
   1143 		flags &= ~FORCECLOSE;
   1144 	ump = VFSTOUFS(mp);
   1145 #ifdef QUOTA
   1146 	if (mp->mnt_flag & MNT_QUOTA) {
   1147 		int i;
   1148 		if ((error = vflush(mp, NULLVP, SKIPSYSTEM|flags)) != 0)
   1149 			return (error);
   1150 		for (i = 0; i < MAXQUOTAS; i++) {
   1151 			if (ump->um_quotas[i] == NULLVP)
   1152 				continue;
   1153 			quotaoff(p, mp, i);
   1154 		}
   1155 		/*
   1156 		 * Here we fall through to vflush again to ensure
   1157 		 * that we have gotten rid of all the system vnodes.
   1158 		 */
   1159 	}
   1160 #endif
   1161 	/*
   1162 	 * Flush all the files.
   1163 	 */
   1164 	error = vflush(mp, NULLVP, flags);
   1165 	if (error)
   1166 		return (error);
   1167 	/*
   1168 	 * Flush filesystem metadata.
   1169 	 */
   1170 	vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY);
   1171 	error = VOP_FSYNC(ump->um_devvp, p->p_ucred, FSYNC_WAIT, 0, 0, p);
   1172 	VOP_UNLOCK(ump->um_devvp, 0);
   1173 	return (error);
   1174 }
   1175 
   1176 /*
   1177  * Get file system statistics.
   1178  */
   1179 int
   1180 ffs_statfs(mp, sbp, p)
   1181 	struct mount *mp;
   1182 	struct statfs *sbp;
   1183 	struct proc *p;
   1184 {
   1185 	struct ufsmount *ump;
   1186 	struct fs *fs;
   1187 
   1188 	ump = VFSTOUFS(mp);
   1189 	fs = ump->um_fs;
   1190 #ifdef COMPAT_09
   1191 	sbp->f_type = 1;
   1192 #else
   1193 	sbp->f_type = 0;
   1194 #endif
   1195 	sbp->f_bsize = fs->fs_fsize;
   1196 	sbp->f_iosize = fs->fs_bsize;
   1197 	sbp->f_blocks = fs->fs_dsize;
   1198 	sbp->f_bfree = blkstofrags(fs, fs->fs_cstotal.cs_nbfree) +
   1199 		fs->fs_cstotal.cs_nffree + dbtofsb(fs, fs->fs_pendingblocks);
   1200 	sbp->f_bavail = (long) (((u_int64_t) fs->fs_dsize * (u_int64_t)
   1201 	    (100 - fs->fs_minfree) / (u_int64_t) 100) -
   1202 	    (u_int64_t) (fs->fs_dsize - sbp->f_bfree));
   1203 	sbp->f_files =  fs->fs_ncg * fs->fs_ipg - ROOTINO;
   1204 	sbp->f_ffree = fs->fs_cstotal.cs_nifree + fs->fs_pendinginodes;
   1205 	copy_statfs_info(sbp, mp);
   1206 	return (0);
   1207 }
   1208 
   1209 /*
   1210  * Go through the disk queues to initiate sandbagged IO;
   1211  * go through the inodes to write those that have been modified;
   1212  * initiate the writing of the super block if it has been modified.
   1213  *
   1214  * Note: we are always called with the filesystem marked `MPBUSY'.
   1215  */
   1216 int
   1217 ffs_sync(mp, waitfor, cred, p)
   1218 	struct mount *mp;
   1219 	int waitfor;
   1220 	struct ucred *cred;
   1221 	struct proc *p;
   1222 {
   1223 	struct vnode *vp, *nvp;
   1224 	struct inode *ip;
   1225 	struct ufsmount *ump = VFSTOUFS(mp);
   1226 	struct fs *fs;
   1227 	int error, allerror = 0;
   1228 
   1229 	fs = ump->um_fs;
   1230 	if (fs->fs_fmod != 0 && fs->fs_ronly != 0) {		/* XXX */
   1231 		printf("fs = %s\n", fs->fs_fsmnt);
   1232 		panic("update: rofs mod");
   1233 	}
   1234 	/*
   1235 	 * Write back each (modified) inode.
   1236 	 */
   1237 	simple_lock(&mntvnode_slock);
   1238 loop:
   1239 	for (vp = LIST_FIRST(&mp->mnt_vnodelist); vp != NULL; vp = nvp) {
   1240 		/*
   1241 		 * If the vnode that we are about to sync is no longer
   1242 		 * associated with this mount point, start over.
   1243 		 */
   1244 		if (vp->v_mount != mp)
   1245 			goto loop;
   1246 		simple_lock(&vp->v_interlock);
   1247 		nvp = LIST_NEXT(vp, v_mntvnodes);
   1248 		ip = VTOI(vp);
   1249 		if (vp->v_type == VNON ||
   1250 		    ((ip->i_flag &
   1251 		      (IN_ACCESS | IN_CHANGE | IN_UPDATE | IN_MODIFIED | IN_ACCESSED)) == 0 &&
   1252 		     LIST_EMPTY(&vp->v_dirtyblkhd) &&
   1253 		     vp->v_uobj.uo_npages == 0))
   1254 		{
   1255 			simple_unlock(&vp->v_interlock);
   1256 			continue;
   1257 		}
   1258 		simple_unlock(&mntvnode_slock);
   1259 		error = vget(vp, LK_EXCLUSIVE | LK_NOWAIT | LK_INTERLOCK);
   1260 		if (error) {
   1261 			simple_lock(&mntvnode_slock);
   1262 			if (error == ENOENT)
   1263 				goto loop;
   1264 			continue;
   1265 		}
   1266 		if ((error = VOP_FSYNC(vp, cred,
   1267 		    waitfor == MNT_WAIT ? FSYNC_WAIT : 0, 0, 0, p)) != 0)
   1268 			allerror = error;
   1269 		vput(vp);
   1270 		simple_lock(&mntvnode_slock);
   1271 	}
   1272 	simple_unlock(&mntvnode_slock);
   1273 	/*
   1274 	 * Force stale file system control information to be flushed.
   1275 	 */
   1276 	if (waitfor != MNT_LAZY) {
   1277 		if (ump->um_mountp->mnt_flag & MNT_SOFTDEP)
   1278 			waitfor = MNT_NOWAIT;
   1279 		vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY);
   1280 		if ((error = VOP_FSYNC(ump->um_devvp, cred,
   1281 		    waitfor == MNT_WAIT ? FSYNC_WAIT : 0, 0, 0, p)) != 0)
   1282 			allerror = error;
   1283 		VOP_UNLOCK(ump->um_devvp, 0);
   1284 	}
   1285 #ifdef QUOTA
   1286 	qsync(mp);
   1287 #endif
   1288 	/*
   1289 	 * Write back modified superblock.
   1290 	 */
   1291 	if (fs->fs_fmod != 0) {
   1292 		fs->fs_fmod = 0;
   1293 		fs->fs_time = time.tv_sec;
   1294 		if ((error = ffs_cgupdate(ump, waitfor)))
   1295 			allerror = error;
   1296 	}
   1297 	return (allerror);
   1298 }
   1299 
   1300 /*
   1301  * Look up a FFS dinode number to find its incore vnode, otherwise read it
   1302  * in from disk.  If it is in core, wait for the lock bit to clear, then
   1303  * return the inode locked.  Detection and handling of mount points must be
   1304  * done by the calling routine.
   1305  */
   1306 int
   1307 ffs_vget(mp, ino, vpp)
   1308 	struct mount *mp;
   1309 	ino_t ino;
   1310 	struct vnode **vpp;
   1311 {
   1312 	struct fs *fs;
   1313 	struct inode *ip;
   1314 	struct ufsmount *ump;
   1315 	struct buf *bp;
   1316 	struct vnode *vp;
   1317 	dev_t dev;
   1318 	int error;
   1319 
   1320 	ump = VFSTOUFS(mp);
   1321 	dev = ump->um_dev;
   1322 
   1323 	if ((*vpp = ufs_ihashget(dev, ino, LK_EXCLUSIVE)) != NULL)
   1324 		return (0);
   1325 
   1326 	/* Allocate a new vnode/inode. */
   1327 	if ((error = getnewvnode(VT_UFS, mp, ffs_vnodeop_p, &vp)) != 0) {
   1328 		*vpp = NULL;
   1329 		return (error);
   1330 	}
   1331 
   1332 	/*
   1333 	 * If someone beat us to it while sleeping in getnewvnode(),
   1334 	 * push back the freshly allocated vnode we don't need, and return.
   1335 	 */
   1336 
   1337 	do {
   1338 		if ((*vpp = ufs_ihashget(dev, ino, LK_EXCLUSIVE)) != NULL) {
   1339 			ungetnewvnode(vp);
   1340 			return (0);
   1341 		}
   1342 	} while (lockmgr(&ufs_hashlock, LK_EXCLUSIVE|LK_SLEEPFAIL, 0));
   1343 
   1344 	/*
   1345 	 * XXX MFS ends up here, too, to allocate an inode.  Should we
   1346 	 * XXX create another pool for MFS inodes?
   1347 	 */
   1348 
   1349 	ip = pool_get(&ffs_inode_pool, PR_WAITOK);
   1350 	memset(ip, 0, sizeof(struct inode));
   1351 	vp->v_data = ip;
   1352 	ip->i_vnode = vp;
   1353 	ip->i_ump = ump;
   1354 	ip->i_fs = fs = ump->um_fs;
   1355 	ip->i_dev = dev;
   1356 	ip->i_number = ino;
   1357 	LIST_INIT(&ip->i_pcbufhd);
   1358 #ifdef QUOTA
   1359 	{
   1360 		int i;
   1361 
   1362 		for (i = 0; i < MAXQUOTAS; i++)
   1363 			ip->i_dquot[i] = NODQUOT;
   1364 	}
   1365 #endif
   1366 
   1367 	/*
   1368 	 * Put it onto its hash chain and lock it so that other requests for
   1369 	 * this inode will block if they arrive while we are sleeping waiting
   1370 	 * for old data structures to be purged or for the contents of the
   1371 	 * disk portion of this inode to be read.
   1372 	 */
   1373 
   1374 	ufs_ihashins(ip);
   1375 	lockmgr(&ufs_hashlock, LK_RELEASE, 0);
   1376 
   1377 	/* Read in the disk contents for the inode, copy into the inode. */
   1378 	error = bread(ump->um_devvp, fsbtodb(fs, ino_to_fsba(fs, ino)),
   1379 		      (int)fs->fs_bsize, NOCRED, &bp);
   1380 	if (error) {
   1381 
   1382 		/*
   1383 		 * The inode does not contain anything useful, so it would
   1384 		 * be misleading to leave it on its hash chain. With mode
   1385 		 * still zero, it will be unlinked and returned to the free
   1386 		 * list by vput().
   1387 		 */
   1388 
   1389 		vput(vp);
   1390 		brelse(bp);
   1391 		*vpp = NULL;
   1392 		return (error);
   1393 	}
   1394 	if (ip->i_ump->um_fstype == UFS1)
   1395 		ip->i_din.ffs1_din = pool_get(&ffs_dinode1_pool, PR_WAITOK);
   1396 	else
   1397 		ip->i_din.ffs2_din = pool_get(&ffs_dinode2_pool, PR_WAITOK);
   1398 	ffs_load_inode(bp, ip, fs, ino);
   1399 	if (DOINGSOFTDEP(vp))
   1400 		softdep_load_inodeblock(ip);
   1401 	else
   1402 		ip->i_ffs_effnlink = ip->i_nlink;
   1403 	brelse(bp);
   1404 
   1405 	/*
   1406 	 * Initialize the vnode from the inode, check for aliases.
   1407 	 * Note that the underlying vnode may have changed.
   1408 	 */
   1409 
   1410 	ufs_vinit(mp, ffs_specop_p, ffs_fifoop_p, &vp);
   1411 
   1412 	/*
   1413 	 * Finish inode initialization now that aliasing has been resolved.
   1414 	 */
   1415 
   1416 	genfs_node_init(vp, &ffs_genfsops);
   1417 	ip->i_devvp = ump->um_devvp;
   1418 	VREF(ip->i_devvp);
   1419 
   1420 	/*
   1421 	 * Ensure that uid and gid are correct. This is a temporary
   1422 	 * fix until fsck has been changed to do the update.
   1423 	 */
   1424 
   1425 	if (fs->fs_old_inodefmt < FS_44INODEFMT) {		/* XXX */
   1426 		ip->i_uid = ip->i_ffs1_ouid;			/* XXX */
   1427 		ip->i_gid = ip->i_ffs1_ogid;			/* XXX */
   1428 	}							/* XXX */
   1429 	uvm_vnp_setsize(vp, ip->i_size);
   1430 	*vpp = vp;
   1431 	return (0);
   1432 }
   1433 
   1434 /*
   1435  * File handle to vnode
   1436  *
   1437  * Have to be really careful about stale file handles:
   1438  * - check that the inode number is valid
   1439  * - call ffs_vget() to get the locked inode
   1440  * - check for an unallocated inode (i_mode == 0)
   1441  * - check that the given client host has export rights and return
   1442  *   those rights via. exflagsp and credanonp
   1443  */
   1444 int
   1445 ffs_fhtovp(mp, fhp, vpp)
   1446 	struct mount *mp;
   1447 	struct fid *fhp;
   1448 	struct vnode **vpp;
   1449 {
   1450 	struct ufid *ufhp;
   1451 	struct fs *fs;
   1452 
   1453 	ufhp = (struct ufid *)fhp;
   1454 	fs = VFSTOUFS(mp)->um_fs;
   1455 	if (ufhp->ufid_ino < ROOTINO ||
   1456 	    ufhp->ufid_ino >= fs->fs_ncg * fs->fs_ipg)
   1457 		return (ESTALE);
   1458 	return (ufs_fhtovp(mp, ufhp, vpp));
   1459 }
   1460 
   1461 /*
   1462  * Vnode pointer to File handle
   1463  */
   1464 /* ARGSUSED */
   1465 int
   1466 ffs_vptofh(vp, fhp)
   1467 	struct vnode *vp;
   1468 	struct fid *fhp;
   1469 {
   1470 	struct inode *ip;
   1471 	struct ufid *ufhp;
   1472 
   1473 	ip = VTOI(vp);
   1474 	ufhp = (struct ufid *)fhp;
   1475 	ufhp->ufid_len = sizeof(struct ufid);
   1476 	ufhp->ufid_ino = ip->i_number;
   1477 	ufhp->ufid_gen = ip->i_gen;
   1478 	return (0);
   1479 }
   1480 
   1481 void
   1482 ffs_init()
   1483 {
   1484 	if (ffs_initcount++ > 0)
   1485 		return;
   1486 
   1487 	softdep_initialize();
   1488 	ufs_init();
   1489 
   1490 	pool_init(&ffs_inode_pool, sizeof(struct inode), 0, 0, 0, "ffsinopl",
   1491 	    &pool_allocator_nointr);
   1492 	pool_init(&ffs_dinode1_pool, sizeof(struct ufs1_dinode), 0, 0, 0,
   1493 	    "dino1pl", &pool_allocator_nointr);
   1494 	pool_init(&ffs_dinode2_pool, sizeof(struct ufs2_dinode), 0, 0, 0,
   1495 	    "dino2pl", &pool_allocator_nointr);
   1496 }
   1497 
   1498 void
   1499 ffs_reinit()
   1500 {
   1501 	softdep_reinitialize();
   1502 	ufs_reinit();
   1503 }
   1504 
   1505 void
   1506 ffs_done()
   1507 {
   1508 	if (--ffs_initcount > 0)
   1509 		return;
   1510 
   1511 	/* XXX softdep cleanup ? */
   1512 	ufs_done();
   1513 	pool_destroy(&ffs_inode_pool);
   1514 }
   1515 
   1516 int
   1517 ffs_sysctl(name, namelen, oldp, oldlenp, newp, newlen, p)
   1518 	int *name;
   1519 	u_int namelen;
   1520 	void *oldp;
   1521 	size_t *oldlenp;
   1522 	void *newp;
   1523 	size_t newlen;
   1524 	struct proc *p;
   1525 {
   1526 	extern int doasyncfree;
   1527 	extern int ffs_log_changeopt;
   1528 
   1529 	/* all sysctl names at this level are terminal */
   1530 	if (namelen != 1)
   1531 		return (ENOTDIR);		/* overloaded */
   1532 
   1533 	switch (name[0]) {
   1534 	case FFS_ASYNCFREE:
   1535 		return (sysctl_int(oldp, oldlenp, newp, newlen, &doasyncfree));
   1536 	case FFS_LOG_CHANGEOPT:
   1537 		return (sysctl_int(oldp, oldlenp, newp, newlen,
   1538 			&ffs_log_changeopt));
   1539 	default:
   1540 		return (EOPNOTSUPP);
   1541 	}
   1542 	/* NOTREACHED */
   1543 }
   1544 
   1545 /*
   1546  * Write a superblock and associated information back to disk.
   1547  */
   1548 int
   1549 ffs_sbupdate(mp, waitfor)
   1550 	struct ufsmount *mp;
   1551 	int waitfor;
   1552 {
   1553 	struct fs *fs = mp->um_fs;
   1554 	struct buf *bp;
   1555 	int error = 0;
   1556 	u_int32_t saveflag;
   1557 
   1558 	bp = getblk(mp->um_devvp,
   1559 	    fs->fs_sblockloc >> (fs->fs_fshift - fs->fs_fsbtodb),
   1560 	    (int)fs->fs_sbsize, 0, 0);
   1561 	saveflag = fs->fs_flags & FS_INTERNAL;
   1562 	fs->fs_flags &= ~FS_INTERNAL;
   1563 
   1564 	if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_sblockloc != SBLOCK_UFS1 &&
   1565 	    (fs->fs_flags & FS_FLAGS_UPDATED) == 0) {
   1566 		printf("%s: correcting fs_sblockloc from %" PRId64 " to %d\n",
   1567 		    fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS1);
   1568 		fs->fs_sblockloc = SBLOCK_UFS1;
   1569 	}
   1570 
   1571 	if (fs->fs_magic == FS_UFS2_MAGIC && fs->fs_sblockloc != SBLOCK_UFS2 &&
   1572 	    (fs->fs_flags & FS_FLAGS_UPDATED) == 0) {
   1573 		printf("%s: correcting fs_sblockloc from %" PRId64 " to %d\n",
   1574 		    fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS2);
   1575 		fs->fs_sblockloc = SBLOCK_UFS2;
   1576 	}
   1577 
   1578 	memcpy(bp->b_data, fs, fs->fs_sbsize);
   1579 
   1580 	ffs_oldfscompat_write((struct fs *)bp->b_data, mp);
   1581 #ifdef FFS_EI
   1582 	if (mp->um_flags & UFS_NEEDSWAP)
   1583 		ffs_sb_swap(fs, (struct fs *)bp->b_data);
   1584 #endif
   1585 #ifdef SUPPORT_FS_42POSTBLFMT_WRITE
   1586 	if (fs->fs_old_postblformat == FS_42POSTBLFMT)
   1587 		memcpy(&((struct fs *)bp->b_data)->fs_old_postbl_start,
   1588 		    mp->um_opostsave, 256);
   1589 #endif
   1590 	fs->fs_flags |= saveflag;
   1591 
   1592 	if (waitfor == MNT_WAIT)
   1593 		error = bwrite(bp);
   1594 	else
   1595 		bawrite(bp);
   1596 	return (error);
   1597 }
   1598 
   1599 int
   1600 ffs_cgupdate(mp, waitfor)
   1601 	struct ufsmount *mp;
   1602 	int waitfor;
   1603 {
   1604 	struct fs *fs = mp->um_fs;
   1605 	struct buf *bp;
   1606 	int blks;
   1607 	void *space;
   1608 	int i, size, error = 0, allerror = 0;
   1609 
   1610 	allerror = ffs_sbupdate(mp, waitfor);
   1611 	blks = howmany(fs->fs_cssize, fs->fs_fsize);
   1612 	space = fs->fs_csp;
   1613 	for (i = 0; i < blks; i += fs->fs_frag) {
   1614 		size = fs->fs_bsize;
   1615 		if (i + fs->fs_frag > blks)
   1616 			size = (blks - i) * fs->fs_fsize;
   1617 		bp = getblk(mp->um_devvp, fsbtodb(fs, fs->fs_csaddr + i),
   1618 		    size, 0, 0);
   1619 #ifdef FFS_EI
   1620 		if (mp->um_flags & UFS_NEEDSWAP)
   1621 			ffs_csum_swap((struct csum*)space,
   1622 			    (struct csum*)bp->b_data, size);
   1623 		else
   1624 #endif
   1625 			memcpy(bp->b_data, space, (u_int)size);
   1626 		space = (char *)space + size;
   1627 		if (waitfor == MNT_WAIT)
   1628 			error = bwrite(bp);
   1629 		else
   1630 			bawrite(bp);
   1631 	}
   1632 	if (!allerror && error)
   1633 		allerror = error;
   1634 	return (allerror);
   1635 }
   1636