xref: /freebsd/sys/ufs/ffs/ffs_vfsops.c (revision 1b6c76a2fe091c74f08427e6c870851025a9cf67)
1 /*
2  * Copyright (c) 1989, 1991, 1993, 1994
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 3. All advertising materials mentioning features or use of this software
14  *    must display the following acknowledgement:
15  *	This product includes software developed by the University of
16  *	California, Berkeley and its contributors.
17  * 4. Neither the name of the University nor the names of its contributors
18  *    may be used to endorse or promote products derived from this software
19  *    without specific prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  *
33  *	@(#)ffs_vfsops.c	8.31 (Berkeley) 5/20/95
34  * $FreeBSD$
35  */
36 
37 #include "opt_quota.h"
38 #include "opt_ufs.h"
39 
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/namei.h>
43 #include <sys/proc.h>
44 #include <sys/kernel.h>
45 #include <sys/vnode.h>
46 #include <sys/mount.h>
47 #include <sys/bio.h>
48 #include <sys/buf.h>
49 #include <sys/conf.h>
50 #include <sys/fcntl.h>
51 #include <sys/disklabel.h>
52 #include <sys/malloc.h>
53 #include <sys/mutex.h>
54 
55 #include <ufs/ufs/extattr.h>
56 #include <ufs/ufs/quota.h>
57 #include <ufs/ufs/ufsmount.h>
58 #include <ufs/ufs/inode.h>
59 #include <ufs/ufs/ufs_extern.h>
60 
61 #include <ufs/ffs/fs.h>
62 #include <ufs/ffs/ffs_extern.h>
63 
64 #include <vm/vm.h>
65 #include <vm/vm_page.h>
66 
67 static MALLOC_DEFINE(M_FFSNODE, "FFS node", "FFS vnode private part");
68 
69 static int	ffs_sbupdate __P((struct ufsmount *, int));
70 int	ffs_reload __P((struct mount *,struct ucred *,struct proc *));
71 static int	ffs_oldfscompat __P((struct fs *));
72 static int	ffs_init __P((struct vfsconf *));
73 
74 static struct vfsops ufs_vfsops = {
75 	ffs_mount,
76 	ufs_start,
77 	ffs_unmount,
78 	ufs_root,
79 	ufs_quotactl,
80 	ffs_statfs,
81 	ffs_sync,
82 	ffs_vget,
83 	ffs_fhtovp,
84 	vfs_stdcheckexp,
85 	ffs_vptofh,
86 	ffs_init,
87 	vfs_stduninit,
88 #ifdef UFS_EXTATTR
89 	ufs_extattrctl,
90 #else
91 	vfs_stdextattrctl,
92 #endif
93 };
94 
95 VFS_SET(ufs_vfsops, ufs, 0);
96 
97 /*
98  * ffs_mount
99  *
100  * Called when mounting local physical media
101  *
102  * PARAMETERS:
103  *		mountroot
104  *			mp	mount point structure
105  *			path	NULL (flag for root mount!!!)
106  *			data	<unused>
107  *			ndp	<unused>
108  *			p	process (user credentials check [statfs])
109  *
110  *		mount
111  *			mp	mount point structure
112  *			path	path to mount point
113  *			data	pointer to argument struct in user space
114  *			ndp	mount point namei() return (used for
115  *				credentials on reload), reused to look
116  *				up block device.
117  *			p	process (user credentials check)
118  *
119  * RETURNS:	0	Success
120  *		!0	error number (errno.h)
121  *
122  * LOCK STATE:
123  *
124  *		ENTRY
125  *			mount point is locked
126  *		EXIT
127  *			mount point is locked
128  *
129  * NOTES:
130  *		A NULL path can be used for a flag since the mount
131  *		system call will fail with EFAULT in copyinstr in
132  *		namei() if it is a genuine NULL from the user.
133  */
134 int
135 ffs_mount(mp, path, data, ndp, p)
136         struct mount		*mp;	/* mount struct pointer*/
137         char			*path;	/* path to mount point*/
138         caddr_t			data;	/* arguments to FS specific mount*/
139         struct nameidata	*ndp;	/* mount point credentials*/
140         struct proc		*p;	/* process requesting mount*/
141 {
142 	size_t		size;
143 	struct vnode	*devvp;
144 	struct ufs_args args;
145 	struct ufsmount *ump = 0;
146 	register struct fs *fs;
147 	int error, flags;
148 	mode_t accessmode;
149 
150 	/*
151 	 * Use NULL path to indicate we are mounting the root file system.
152 	 */
153 	if (path == NULL) {
154 		if ((error = bdevvp(rootdev, &rootvp))) {
155 			printf("ffs_mountroot: can't find rootvp\n");
156 			return (error);
157 		}
158 
159 		if ((error = ffs_mountfs(rootvp, mp, p, M_FFSNODE)) != 0)
160 			return (error);
161 
162 		(void)VFS_STATFS(mp, &mp->mnt_stat, p);
163 		return (0);
164 	}
165 
166 	/*
167 	 * Mounting non-root file system or updating a file system
168 	 */
169 	if ((error = copyin(data, (caddr_t)&args, sizeof(struct ufs_args)))!= 0)
170 		return (error);
171 
172 	/*
173 	 * If updating, check whether changing from read-only to
174 	 * read/write; if there is no device name, that's all we do.
175 	 */
176 	if (mp->mnt_flag & MNT_UPDATE) {
177 		ump = VFSTOUFS(mp);
178 		fs = ump->um_fs;
179 		devvp = ump->um_devvp;
180 		if (fs->fs_ronly == 0 && (mp->mnt_flag & MNT_RDONLY)) {
181 			if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0)
182 				return (error);
183 			flags = WRITECLOSE;
184 			if (mp->mnt_flag & MNT_FORCE)
185 				flags |= FORCECLOSE;
186 			if (mp->mnt_flag & MNT_SOFTDEP) {
187 				error = softdep_flushfiles(mp, flags, p);
188 			} else {
189 				error = ffs_flushfiles(mp, flags, p);
190 			}
191 			if (error) {
192 				vn_finished_write(mp);
193 				return (error);
194 			}
195 			if (fs->fs_pendingblocks != 0 ||
196 			    fs->fs_pendinginodes != 0) {
197 				printf("%s: update error: blocks %d files %d\n",
198 				    fs->fs_fsmnt, fs->fs_pendingblocks,
199 				    fs->fs_pendinginodes);
200 				fs->fs_pendingblocks = 0;
201 				fs->fs_pendinginodes = 0;
202 			}
203 			fs->fs_ronly = 1;
204 			if ((fs->fs_flags & (FS_UNCLEAN | FS_NEEDSFSCK)) == 0)
205 				fs->fs_clean = 1;
206 			if ((error = ffs_sbupdate(ump, MNT_WAIT)) != 0) {
207 				fs->fs_ronly = 0;
208 				fs->fs_clean = 0;
209 				vn_finished_write(mp);
210 				return (error);
211 			}
212 			vn_finished_write(mp);
213 		}
214 		if ((mp->mnt_flag & MNT_RELOAD) &&
215 		    (error = ffs_reload(mp, ndp->ni_cnd.cn_cred, p)) != 0)
216 			return (error);
217 		if (fs->fs_ronly && (mp->mnt_kern_flag & MNTK_WANTRDWR)) {
218 			/*
219 			 * If upgrade to read-write by non-root, then verify
220 			 * that user has necessary permissions on the device.
221 			 */
222 			if (p->p_ucred->cr_uid != 0) {
223 				vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY, p);
224 				if ((error = VOP_ACCESS(devvp, VREAD | VWRITE,
225 				    p->p_ucred, p)) != 0) {
226 					VOP_UNLOCK(devvp, 0, p);
227 					return (error);
228 				}
229 				VOP_UNLOCK(devvp, 0, p);
230 			}
231 			fs->fs_flags &= ~FS_UNCLEAN;
232 			if (fs->fs_clean == 0) {
233 				fs->fs_flags |= FS_UNCLEAN;
234 				if ((mp->mnt_flag & MNT_FORCE) ||
235 				    ((fs->fs_flags & FS_NEEDSFSCK) == 0 &&
236 				     (fs->fs_flags & FS_DOSOFTDEP))) {
237 					printf("WARNING: %s was not %s\n",
238 					   fs->fs_fsmnt, "properly dismounted");
239 				} else {
240 					printf(
241 "WARNING: R/W mount of %s denied.  Filesystem is not clean - run fsck\n",
242 					    fs->fs_fsmnt);
243 					return (EPERM);
244 				}
245 			}
246 			if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0)
247 				return (error);
248 			fs->fs_ronly = 0;
249 			fs->fs_clean = 0;
250 			if ((error = ffs_sbupdate(ump, MNT_WAIT)) != 0) {
251 				vn_finished_write(mp);
252 				return (error);
253 			}
254 			/* check to see if we need to start softdep */
255 			if ((fs->fs_flags & FS_DOSOFTDEP) &&
256 			    (error = softdep_mount(devvp, mp, fs, p->p_ucred))){
257 				vn_finished_write(mp);
258 				return (error);
259 			}
260 			if (fs->fs_snapinum[0] != 0)
261 				ffs_snapshot_mount(mp);
262 			vn_finished_write(mp);
263 		}
264 		/*
265 		 * Soft updates is incompatible with "async",
266 		 * so if we are doing softupdates stop the user
267 		 * from setting the async flag in an update.
268 		 * Softdep_mount() clears it in an initial mount
269 		 * or ro->rw remount.
270 		 */
271 		if (mp->mnt_flag & MNT_SOFTDEP)
272 			mp->mnt_flag &= ~MNT_ASYNC;
273 		/*
274 		 * If not updating name, process export requests.
275 		 */
276 		if (args.fspec == 0)
277 			return (vfs_export(mp, &args.export));
278 		/*
279 		 * If this is a snapshot request, take the snapshot.
280 		 */
281 		if (mp->mnt_flag & MNT_SNAPSHOT)
282 			return (ffs_snapshot(mp, args.fspec));
283 	}
284 
285 	/*
286 	 * Not an update, or updating the name: look up the name
287 	 * and verify that it refers to a sensible block device.
288 	 */
289 	NDINIT(ndp, LOOKUP, FOLLOW, UIO_USERSPACE, args.fspec, p);
290 	if ((error = namei(ndp)) != 0)
291 		return (error);
292 	NDFREE(ndp, NDF_ONLY_PNBUF);
293 	devvp = ndp->ni_vp;
294 	if (!vn_isdisk(devvp, &error)) {
295 		vrele(devvp);
296 		return (error);
297 	}
298 
299 	/*
300 	 * If mount by non-root, then verify that user has necessary
301 	 * permissions on the device.
302 	 */
303 	if (p->p_ucred->cr_uid != 0) {
304 		accessmode = VREAD;
305 		if ((mp->mnt_flag & MNT_RDONLY) == 0)
306 			accessmode |= VWRITE;
307 		vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY, p);
308 		if ((error = VOP_ACCESS(devvp, accessmode, p->p_ucred, p))!= 0){
309 			vput(devvp);
310 			return (error);
311 		}
312 		VOP_UNLOCK(devvp, 0, p);
313 	}
314 
315 	if (mp->mnt_flag & MNT_UPDATE) {
316 		/*
317 		 * Update only
318 		 *
319 		 * If it's not the same vnode, or at least the same device
320 		 * then it's not correct.
321 		 */
322 
323 		if (devvp != ump->um_devvp &&
324 		    devvp->v_rdev != ump->um_devvp->v_rdev)
325 			error = EINVAL;	/* needs translation */
326 		vrele(devvp);
327 		if (error)
328 			return (error);
329 	} else {
330 		/*
331 		 * New mount
332 		 *
333 		 * We need the name for the mount point (also used for
334 		 * "last mounted on") copied in. If an error occurs,
335 		 * the mount point is discarded by the upper level code.
336 		 * Note that vfs_mount() populates f_mntonname for us.
337 		 */
338 		if ((error = ffs_mountfs(devvp, mp, p, M_FFSNODE)) != 0) {
339 			vrele(devvp);
340 			return (error);
341 		}
342 	}
343 	/*
344 	 * Save "mounted from" device name info for mount point (NULL pad).
345 	 */
346 	copyinstr(args.fspec, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, &size);
347 	bzero( mp->mnt_stat.f_mntfromname + size, MNAMELEN - size);
348 	/*
349 	 * Initialize filesystem stat information in mount struct.
350 	 */
351 	(void)VFS_STATFS(mp, &mp->mnt_stat, p);
352 	return (0);
353 }
354 
355 /*
356  * Reload all incore data for a filesystem (used after running fsck on
357  * the root filesystem and finding things to fix). The filesystem must
358  * be mounted read-only.
359  *
360  * Things to do to update the mount:
361  *	1) invalidate all cached meta-data.
362  *	2) re-read superblock from disk.
363  *	3) re-read summary information from disk.
364  *	4) invalidate all inactive vnodes.
365  *	5) invalidate all cached file data.
366  *	6) re-read inode data for all active vnodes.
367  */
368 int
369 ffs_reload(mp, cred, p)
370 	register struct mount *mp;
371 	struct ucred *cred;
372 	struct proc *p;
373 {
374 	register struct vnode *vp, *nvp, *devvp;
375 	struct inode *ip;
376 	void *space;
377 	struct buf *bp;
378 	struct fs *fs, *newfs;
379 	struct partinfo dpart;
380 	dev_t dev;
381 	int i, blks, size, error;
382 	int32_t *lp;
383 
384 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
385 		return (EINVAL);
386 	/*
387 	 * Step 1: invalidate all cached meta-data.
388 	 */
389 	devvp = VFSTOUFS(mp)->um_devvp;
390 	vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY, p);
391 	error = vinvalbuf(devvp, 0, cred, p, 0, 0);
392 	VOP_UNLOCK(devvp, 0, p);
393 	if (error)
394 		panic("ffs_reload: dirty1");
395 
396 	dev = devvp->v_rdev;
397 
398 	/*
399 	 * Only VMIO the backing device if the backing device is a real
400 	 * block device.
401 	 */
402 	if (vn_isdisk(devvp, NULL)) {
403 		vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY, p);
404 		vfs_object_create(devvp, p, p->p_ucred);
405 		mtx_lock(&devvp->v_interlock);
406 		VOP_UNLOCK(devvp, LK_INTERLOCK, p);
407 	}
408 
409 	/*
410 	 * Step 2: re-read superblock from disk.
411 	 */
412 	if (VOP_IOCTL(devvp, DIOCGPART, (caddr_t)&dpart, FREAD, NOCRED, p) != 0)
413 		size = DEV_BSIZE;
414 	else
415 		size = dpart.disklab->d_secsize;
416 	if ((error = bread(devvp, (ufs_daddr_t)(SBOFF/size), SBSIZE, NOCRED,&bp)) != 0)
417 		return (error);
418 	newfs = (struct fs *)bp->b_data;
419 	if (newfs->fs_magic != FS_MAGIC || newfs->fs_bsize > MAXBSIZE ||
420 		newfs->fs_bsize < sizeof(struct fs)) {
421 			brelse(bp);
422 			return (EIO);		/* XXX needs translation */
423 	}
424 	fs = VFSTOUFS(mp)->um_fs;
425 	/*
426 	 * Copy pointer fields back into superblock before copying in	XXX
427 	 * new superblock. These should really be in the ufsmount.	XXX
428 	 * Note that important parameters (eg fs_ncg) are unchanged.
429 	 */
430 	newfs->fs_csp = fs->fs_csp;
431 	newfs->fs_maxcluster = fs->fs_maxcluster;
432 	newfs->fs_contigdirs = fs->fs_contigdirs;
433 	bcopy(newfs, fs, (u_int)fs->fs_sbsize);
434 	if (fs->fs_sbsize < SBSIZE)
435 		bp->b_flags |= B_INVAL | B_NOCACHE;
436 	brelse(bp);
437 	mp->mnt_maxsymlinklen = fs->fs_maxsymlinklen;
438 	ffs_oldfscompat(fs);
439 	/* An old fsck may have zeroed these fields, so recheck them. */
440 	if (fs->fs_avgfilesize <= 0)		/* XXX */
441 		fs->fs_avgfilesize = AVFILESIZ;	/* XXX */
442 	if (fs->fs_avgfpdir <= 0)		/* XXX */
443 		fs->fs_avgfpdir = AFPDIR;	/* XXX */
444 	if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
445 		printf("%s: reload pending error: blocks %d files %d\n",
446 		    fs->fs_fsmnt, fs->fs_pendingblocks, fs->fs_pendinginodes);
447 		fs->fs_pendingblocks = 0;
448 		fs->fs_pendinginodes = 0;
449 	}
450 
451 	/*
452 	 * Step 3: re-read summary information from disk.
453 	 */
454 	blks = howmany(fs->fs_cssize, fs->fs_fsize);
455 	space = fs->fs_csp;
456 	for (i = 0; i < blks; i += fs->fs_frag) {
457 		size = fs->fs_bsize;
458 		if (i + fs->fs_frag > blks)
459 			size = (blks - i) * fs->fs_fsize;
460 		error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), size,
461 		    NOCRED, &bp);
462 		if (error)
463 			return (error);
464 		bcopy(bp->b_data, space, (u_int)size);
465 		space = (char *)space + size;
466 		brelse(bp);
467 	}
468 	/*
469 	 * We no longer know anything about clusters per cylinder group.
470 	 */
471 	if (fs->fs_contigsumsize > 0) {
472 		lp = fs->fs_maxcluster;
473 		for (i = 0; i < fs->fs_ncg; i++)
474 			*lp++ = fs->fs_contigsumsize;
475 	}
476 
477 loop:
478 	mtx_lock(&mntvnode_mtx);
479 	for (vp = LIST_FIRST(&mp->mnt_vnodelist); vp != NULL; vp = nvp) {
480 		if (vp->v_mount != mp) {
481 			mtx_unlock(&mntvnode_mtx);
482 			goto loop;
483 		}
484 		nvp = LIST_NEXT(vp, v_mntvnodes);
485 		mtx_unlock(&mntvnode_mtx);
486 		/*
487 		 * Step 4: invalidate all inactive vnodes.
488 		 */
489 		if (vrecycle(vp, NULL, p))
490 			goto loop;
491 		/*
492 		 * Step 5: invalidate all cached file data.
493 		 */
494 		mtx_lock(&vp->v_interlock);
495 		if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, p)) {
496 			goto loop;
497 		}
498 		if (vinvalbuf(vp, 0, cred, p, 0, 0))
499 			panic("ffs_reload: dirty2");
500 		/*
501 		 * Step 6: re-read inode data for all active vnodes.
502 		 */
503 		ip = VTOI(vp);
504 		error =
505 		    bread(devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
506 		    (int)fs->fs_bsize, NOCRED, &bp);
507 		if (error) {
508 			vput(vp);
509 			return (error);
510 		}
511 		ip->i_din = *((struct dinode *)bp->b_data +
512 		    ino_to_fsbo(fs, ip->i_number));
513 		ip->i_effnlink = ip->i_nlink;
514 		brelse(bp);
515 		vput(vp);
516 		mtx_lock(&mntvnode_mtx);
517 	}
518 	mtx_unlock(&mntvnode_mtx);
519 	return (0);
520 }
521 
522 #include <sys/sysctl.h>
523 int bigcgs = 0;
524 SYSCTL_INT(_debug, OID_AUTO, bigcgs, CTLFLAG_RW, &bigcgs, 0, "");
525 
526 /*
527  * Common code for mount and mountroot
528  */
529 int
530 ffs_mountfs(devvp, mp, p, malloctype)
531 	register struct vnode *devvp;
532 	struct mount *mp;
533 	struct proc *p;
534 	struct malloc_type *malloctype;
535 {
536 	register struct ufsmount *ump;
537 	struct buf *bp;
538 	register struct fs *fs;
539 	dev_t dev;
540 	struct partinfo dpart;
541 	void *space;
542 	int error, i, blks, size, ronly;
543 	int32_t *lp;
544 	struct ucred *cred;
545 	u_int64_t maxfilesize;					/* XXX */
546 	size_t strsize;
547 	int ncount;
548 
549 	dev = devvp->v_rdev;
550 	cred = p ? p->p_ucred : NOCRED;
551 	/*
552 	 * Disallow multiple mounts of the same device.
553 	 * Disallow mounting of a device that is currently in use
554 	 * (except for root, which might share swap device for miniroot).
555 	 * Flush out any old buffers remaining from a previous use.
556 	 */
557 	error = vfs_mountedon(devvp);
558 	if (error)
559 		return (error);
560 	ncount = vcount(devvp);
561 
562 	if (ncount > 1 && devvp != rootvp)
563 		return (EBUSY);
564 	vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY, p);
565 	error = vinvalbuf(devvp, V_SAVE, cred, p, 0, 0);
566 	VOP_UNLOCK(devvp, 0, p);
567 	if (error)
568 		return (error);
569 
570 	/*
571 	 * Only VMIO the backing device if the backing device is a real
572 	 * block device.
573 	 * Note that it is optional that the backing device be VMIOed.  This
574 	 * increases the opportunity for metadata caching.
575 	 */
576 	if (vn_isdisk(devvp, NULL)) {
577 		vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY, p);
578 		vfs_object_create(devvp, p, cred);
579 		mtx_lock(&devvp->v_interlock);
580 		VOP_UNLOCK(devvp, LK_INTERLOCK, p);
581 	}
582 
583 	ronly = (mp->mnt_flag & MNT_RDONLY) != 0;
584 	vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY, p);
585 	error = VOP_OPEN(devvp, ronly ? FREAD : FREAD|FWRITE, FSCRED, p);
586 	VOP_UNLOCK(devvp, 0, p);
587 	if (error)
588 		return (error);
589 	if (devvp->v_rdev->si_iosize_max > mp->mnt_iosize_max)
590 		mp->mnt_iosize_max = devvp->v_rdev->si_iosize_max;
591 	if (mp->mnt_iosize_max > MAXPHYS)
592 		mp->mnt_iosize_max = MAXPHYS;
593 
594 	if (VOP_IOCTL(devvp, DIOCGPART, (caddr_t)&dpart, FREAD, cred, p) != 0)
595 		size = DEV_BSIZE;
596 	else
597 		size = dpart.disklab->d_secsize;
598 
599 	bp = NULL;
600 	ump = NULL;
601 	if ((error = bread(devvp, SBLOCK, SBSIZE, cred, &bp)) != 0)
602 		goto out;
603 	fs = (struct fs *)bp->b_data;
604 	if (fs->fs_magic != FS_MAGIC || fs->fs_bsize > MAXBSIZE ||
605 	    fs->fs_bsize < sizeof(struct fs)) {
606 		error = EINVAL;		/* XXX needs translation */
607 		goto out;
608 	}
609 	fs->fs_fmod = 0;
610 	fs->fs_flags &= ~FS_UNCLEAN;
611 	if (fs->fs_clean == 0) {
612 		fs->fs_flags |= FS_UNCLEAN;
613 		if (ronly || (mp->mnt_flag & MNT_FORCE) ||
614 		    ((fs->fs_flags & FS_NEEDSFSCK) == 0 &&
615 		     (fs->fs_flags & FS_DOSOFTDEP))) {
616 			printf(
617 "WARNING: %s was not properly dismounted\n",
618 			    fs->fs_fsmnt);
619 		} else {
620 			printf(
621 "WARNING: R/W mount of %s denied.  Filesystem is not clean - run fsck\n",
622 			    fs->fs_fsmnt);
623 			error = EPERM;
624 			goto out;
625 		}
626 		if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
627 			printf("%s: lost blocks %d files %d\n", fs->fs_fsmnt,
628 			    fs->fs_pendingblocks, fs->fs_pendinginodes);
629 			fs->fs_pendingblocks = 0;
630 			fs->fs_pendinginodes = 0;
631 		}
632 	}
633 	if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
634 		printf("%s: mount pending error: blocks %d files %d\n",
635 		    fs->fs_fsmnt, fs->fs_pendingblocks, fs->fs_pendinginodes);
636 		fs->fs_pendingblocks = 0;
637 		fs->fs_pendinginodes = 0;
638 	}
639 	/* XXX updating 4.2 FFS superblocks trashes rotational layout tables */
640 	if (fs->fs_postblformat == FS_42POSTBLFMT && !ronly) {
641 		error = EROFS;          /* needs translation */
642 		goto out;
643 	}
644 	ump = malloc(sizeof *ump, M_UFSMNT, M_WAITOK | M_ZERO);
645 	ump->um_malloctype = malloctype;
646 	ump->um_i_effnlink_valid = 1;
647 	ump->um_fs = malloc((u_long)fs->fs_sbsize, M_UFSMNT,
648 	    M_WAITOK);
649 	ump->um_blkatoff = ffs_blkatoff;
650 	ump->um_truncate = ffs_truncate;
651 	ump->um_update = ffs_update;
652 	ump->um_valloc = ffs_valloc;
653 	ump->um_vfree = ffs_vfree;
654 	ump->um_balloc = ffs_balloc;
655 	bcopy(bp->b_data, ump->um_fs, (u_int)fs->fs_sbsize);
656 	if (fs->fs_sbsize < SBSIZE)
657 		bp->b_flags |= B_INVAL | B_NOCACHE;
658 	brelse(bp);
659 	bp = NULL;
660 	fs = ump->um_fs;
661 	fs->fs_ronly = ronly;
662 	size = fs->fs_cssize;
663 	blks = howmany(size, fs->fs_fsize);
664 	if (fs->fs_contigsumsize > 0)
665 		size += fs->fs_ncg * sizeof(int32_t);
666 	size += fs->fs_ncg * sizeof(u_int8_t);
667 	space = malloc((u_long)size, M_UFSMNT, M_WAITOK);
668 	fs->fs_csp = space;
669 	for (i = 0; i < blks; i += fs->fs_frag) {
670 		size = fs->fs_bsize;
671 		if (i + fs->fs_frag > blks)
672 			size = (blks - i) * fs->fs_fsize;
673 		if ((error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), size,
674 		    cred, &bp)) != 0) {
675 			free(fs->fs_csp, M_UFSMNT);
676 			goto out;
677 		}
678 		bcopy(bp->b_data, space, (u_int)size);
679 		space = (char *)space + size;
680 		brelse(bp);
681 		bp = NULL;
682 	}
683 	if (fs->fs_contigsumsize > 0) {
684 		fs->fs_maxcluster = lp = space;
685 		for (i = 0; i < fs->fs_ncg; i++)
686 			*lp++ = fs->fs_contigsumsize;
687 	}
688 	size = fs->fs_ncg * sizeof(u_int8_t);
689 	fs->fs_contigdirs = (u_int8_t *)space;
690 	space = (u_int8_t *)space + size;
691 	bzero(fs->fs_contigdirs, size);
692 	/* Compatibility for old filesystems 	   XXX */
693 	if (fs->fs_avgfilesize <= 0)		/* XXX */
694 		fs->fs_avgfilesize = AVFILESIZ;	/* XXX */
695 	if (fs->fs_avgfpdir <= 0)		/* XXX */
696 		fs->fs_avgfpdir = AFPDIR;	/* XXX */
697 	mp->mnt_data = (qaddr_t)ump;
698 	mp->mnt_stat.f_fsid.val[0] = fs->fs_id[0];
699 	mp->mnt_stat.f_fsid.val[1] = fs->fs_id[1];
700 	if (fs->fs_id[0] == 0 || fs->fs_id[1] == 0 ||
701 	    vfs_getvfs(&mp->mnt_stat.f_fsid))
702 		vfs_getnewfsid(mp);
703 	mp->mnt_maxsymlinklen = fs->fs_maxsymlinklen;
704 	mp->mnt_flag |= MNT_LOCAL;
705 	ump->um_mountp = mp;
706 	ump->um_dev = dev;
707 	ump->um_devvp = devvp;
708 	ump->um_nindir = fs->fs_nindir;
709 	ump->um_bptrtodb = fs->fs_fsbtodb;
710 	ump->um_seqinc = fs->fs_frag;
711 	for (i = 0; i < MAXQUOTAS; i++)
712 		ump->um_quotas[i] = NULLVP;
713 #ifdef UFS_EXTATTR
714 	ufs_extattr_uepm_init(&ump->um_extattr);
715 #endif
716 	devvp->v_rdev->si_mountpoint = mp;
717 	ffs_oldfscompat(fs);
718 
719 	/*
720 	 * Set FS local "last mounted on" information (NULL pad)
721 	 */
722 	copystr(	mp->mnt_stat.f_mntonname,	/* mount point*/
723 			fs->fs_fsmnt,			/* copy area*/
724 			sizeof(fs->fs_fsmnt) - 1,	/* max size*/
725 			&strsize);			/* real size*/
726 	bzero( fs->fs_fsmnt + strsize, sizeof(fs->fs_fsmnt) - strsize);
727 
728 	if( mp->mnt_flag & MNT_ROOTFS) {
729 		/*
730 		 * Root mount; update timestamp in mount structure.
731 		 * this will be used by the common root mount code
732 		 * to update the system clock.
733 		 */
734 		mp->mnt_time = fs->fs_time;
735 	}
736 
737 	ump->um_savedmaxfilesize = fs->fs_maxfilesize;		/* XXX */
738 	maxfilesize = (u_int64_t)0x40000000 * fs->fs_bsize - 1;	/* XXX */
739 	if (fs->fs_maxfilesize > maxfilesize)			/* XXX */
740 		fs->fs_maxfilesize = maxfilesize;		/* XXX */
741 	if (bigcgs) {
742 		if (fs->fs_sparecon[0] <= 0)
743 			fs->fs_sparecon[0] = fs->fs_cgsize;
744 		fs->fs_cgsize = fs->fs_bsize;
745 	}
746 	if (ronly == 0) {
747 		if ((fs->fs_flags & FS_DOSOFTDEP) &&
748 		    (error = softdep_mount(devvp, mp, fs, cred)) != 0) {
749 			free(fs->fs_csp, M_UFSMNT);
750 			goto out;
751 		}
752 		if (fs->fs_snapinum[0] != 0)
753 			ffs_snapshot_mount(mp);
754 		fs->fs_fmod = 1;
755 		fs->fs_clean = 0;
756 		(void) ffs_sbupdate(ump, MNT_WAIT);
757 	}
758 #ifdef UFS_EXTATTR
759 #ifdef UFS_EXTATTR_AUTOSTART
760 	/*
761 	 *
762 	 * Auto-starting does the following:
763 	 *	- check for /.attribute in the fs, and extattr_start if so
764 	 *	- for each file in .attribute, enable that file with
765 	 * 	  an attribute of the same name.
766 	 * Not clear how to report errors -- probably eat them.
767 	 * This would all happen while the file system was busy/not
768 	 * available, so would effectively be "atomic".
769 	 */
770 	(void) ufs_extattr_autostart(mp, p);
771 #endif /* !UFS_EXTATTR_AUTOSTART */
772 #endif /* !UFS_EXTATTR */
773 	return (0);
774 out:
775 	devvp->v_rdev->si_mountpoint = NULL;
776 	if (bp)
777 		brelse(bp);
778 	(void)VOP_CLOSE(devvp, ronly ? FREAD : FREAD|FWRITE, cred, p);
779 	if (ump) {
780 		free(ump->um_fs, M_UFSMNT);
781 		free(ump, M_UFSMNT);
782 		mp->mnt_data = (qaddr_t)0;
783 	}
784 	return (error);
785 }
786 
787 /*
788  * Sanity checks for old file systems.
789  *
790  * XXX - goes away some day.
791  */
792 static int
793 ffs_oldfscompat(fs)
794 	struct fs *fs;
795 {
796 
797 	fs->fs_npsect = max(fs->fs_npsect, fs->fs_nsect);	/* XXX */
798 	fs->fs_interleave = max(fs->fs_interleave, 1);		/* XXX */
799 	if (fs->fs_postblformat == FS_42POSTBLFMT)		/* XXX */
800 		fs->fs_nrpos = 8;				/* XXX */
801 	if (fs->fs_inodefmt < FS_44INODEFMT) {			/* XXX */
802 #if 0
803 		int i;						/* XXX */
804 		u_int64_t sizepb = fs->fs_bsize;		/* XXX */
805 								/* XXX */
806 		fs->fs_maxfilesize = fs->fs_bsize * NDADDR - 1;	/* XXX */
807 		for (i = 0; i < NIADDR; i++) {			/* XXX */
808 			sizepb *= NINDIR(fs);			/* XXX */
809 			fs->fs_maxfilesize += sizepb;		/* XXX */
810 		}						/* XXX */
811 #endif
812 		fs->fs_maxfilesize = (u_quad_t) 1LL << 39;
813 		fs->fs_qbmask = ~fs->fs_bmask;			/* XXX */
814 		fs->fs_qfmask = ~fs->fs_fmask;			/* XXX */
815 	}							/* XXX */
816 	return (0);
817 }
818 
819 /*
820  * unmount system call
821  */
822 int
823 ffs_unmount(mp, mntflags, p)
824 	struct mount *mp;
825 	int mntflags;
826 	struct proc *p;
827 {
828 	register struct ufsmount *ump = VFSTOUFS(mp);
829 	register struct fs *fs;
830 	int error, flags;
831 
832 	flags = 0;
833 	if (mntflags & MNT_FORCE) {
834 		flags |= FORCECLOSE;
835 	}
836 #ifdef UFS_EXTATTR
837 	if ((error = ufs_extattr_stop(mp, p)))
838 		if (error != EOPNOTSUPP)
839 			printf("ffs_unmount: ufs_extattr_stop returned %d\n",
840 			    error);
841 	ufs_extattr_uepm_destroy(&ump->um_extattr);
842 #endif
843 	if (mp->mnt_flag & MNT_SOFTDEP) {
844 		if ((error = softdep_flushfiles(mp, flags, p)) != 0)
845 			return (error);
846 	} else {
847 		if ((error = ffs_flushfiles(mp, flags, p)) != 0)
848 			return (error);
849 	}
850 	fs = ump->um_fs;
851 	if (bigcgs) {
852 		fs->fs_cgsize = fs->fs_sparecon[0];
853 		fs->fs_sparecon[0] = 0;
854 	}
855 	if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
856 		printf("%s: unmount pending error: blocks %d files %d\n",
857 		    fs->fs_fsmnt, fs->fs_pendingblocks, fs->fs_pendinginodes);
858 		fs->fs_pendingblocks = 0;
859 		fs->fs_pendinginodes = 0;
860 	}
861 	if (fs->fs_ronly == 0) {
862 		fs->fs_clean = fs->fs_flags & (FS_UNCLEAN|FS_NEEDSFSCK) ? 0 : 1;
863 		error = ffs_sbupdate(ump, MNT_WAIT);
864 		if (error) {
865 			fs->fs_clean = 0;
866 			return (error);
867 		}
868 	}
869 	ump->um_devvp->v_rdev->si_mountpoint = NULL;
870 
871 	vinvalbuf(ump->um_devvp, V_SAVE, NOCRED, p, 0, 0);
872 	error = VOP_CLOSE(ump->um_devvp, fs->fs_ronly ? FREAD : FREAD|FWRITE,
873 		NOCRED, p);
874 
875 	vrele(ump->um_devvp);
876 
877 	free(fs->fs_csp, M_UFSMNT);
878 	free(fs, M_UFSMNT);
879 	free(ump, M_UFSMNT);
880 	mp->mnt_data = (qaddr_t)0;
881 	mp->mnt_flag &= ~MNT_LOCAL;
882 	return (error);
883 }
884 
885 /*
886  * Flush out all the files in a filesystem.
887  */
888 int
889 ffs_flushfiles(mp, flags, p)
890 	register struct mount *mp;
891 	int flags;
892 	struct proc *p;
893 {
894 	register struct ufsmount *ump;
895 	int error;
896 
897 	ump = VFSTOUFS(mp);
898 #ifdef QUOTA
899 	if (mp->mnt_flag & MNT_QUOTA) {
900 		int i;
901 		error = vflush(mp, 0, SKIPSYSTEM|flags);
902 		if (error)
903 			return (error);
904 		for (i = 0; i < MAXQUOTAS; i++) {
905 			if (ump->um_quotas[i] == NULLVP)
906 				continue;
907 			quotaoff(p, mp, i);
908 		}
909 		/*
910 		 * Here we fall through to vflush again to ensure
911 		 * that we have gotten rid of all the system vnodes.
912 		 */
913 	}
914 #endif
915 	if (ump->um_devvp->v_flag & VCOPYONWRITE) {
916 		if ((error = vflush(mp, 0, SKIPSYSTEM | flags)) != 0)
917 			return (error);
918 		ffs_snapshot_unmount(mp);
919 		/*
920 		 * Here we fall through to vflush again to ensure
921 		 * that we have gotten rid of all the system vnodes.
922 		 */
923 	}
924         /*
925 	 * Flush all the files.
926 	 */
927 	if ((error = vflush(mp, 0, flags)) != 0)
928 		return (error);
929 	/*
930 	 * Flush filesystem metadata.
931 	 */
932 	vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY, p);
933 	error = VOP_FSYNC(ump->um_devvp, p->p_ucred, MNT_WAIT, p);
934 	VOP_UNLOCK(ump->um_devvp, 0, p);
935 	return (error);
936 }
937 
938 /*
939  * Get file system statistics.
940  */
941 int
942 ffs_statfs(mp, sbp, p)
943 	struct mount *mp;
944 	register struct statfs *sbp;
945 	struct proc *p;
946 {
947 	register struct ufsmount *ump;
948 	register struct fs *fs;
949 
950 	ump = VFSTOUFS(mp);
951 	fs = ump->um_fs;
952 	if (fs->fs_magic != FS_MAGIC)
953 		panic("ffs_statfs");
954 	sbp->f_bsize = fs->fs_fsize;
955 	sbp->f_iosize = fs->fs_bsize;
956 	sbp->f_blocks = fs->fs_dsize;
957 	sbp->f_bfree = fs->fs_cstotal.cs_nbfree * fs->fs_frag +
958 	    fs->fs_cstotal.cs_nffree + dbtofsb(fs, fs->fs_pendingblocks);
959 	sbp->f_bavail = freespace(fs, fs->fs_minfree) +
960 	    dbtofsb(fs, fs->fs_pendingblocks);
961 	sbp->f_files =  fs->fs_ncg * fs->fs_ipg - ROOTINO;
962 	sbp->f_ffree = fs->fs_cstotal.cs_nifree + fs->fs_pendinginodes;
963 	if (sbp != &mp->mnt_stat) {
964 		sbp->f_type = mp->mnt_vfc->vfc_typenum;
965 		bcopy((caddr_t)mp->mnt_stat.f_mntonname,
966 			(caddr_t)&sbp->f_mntonname[0], MNAMELEN);
967 		bcopy((caddr_t)mp->mnt_stat.f_mntfromname,
968 			(caddr_t)&sbp->f_mntfromname[0], MNAMELEN);
969 	}
970 	return (0);
971 }
972 
973 /*
974  * Go through the disk queues to initiate sandbagged IO;
975  * go through the inodes to write those that have been modified;
976  * initiate the writing of the super block if it has been modified.
977  *
978  * Note: we are always called with the filesystem marked `MPBUSY'.
979  */
980 int
981 ffs_sync(mp, waitfor, cred, p)
982 	struct mount *mp;
983 	int waitfor;
984 	struct ucred *cred;
985 	struct proc *p;
986 {
987 	struct vnode *nvp, *vp, *devvp;
988 	struct inode *ip;
989 	struct ufsmount *ump = VFSTOUFS(mp);
990 	struct fs *fs;
991 	int error, count, wait, lockreq, allerror = 0;
992 
993 	fs = ump->um_fs;
994 	if (fs->fs_fmod != 0 && fs->fs_ronly != 0) {		/* XXX */
995 		printf("fs = %s\n", fs->fs_fsmnt);
996 		panic("ffs_sync: rofs mod");
997 	}
998 	/*
999 	 * Write back each (modified) inode.
1000 	 */
1001 	wait = 0;
1002 	lockreq = LK_EXCLUSIVE | LK_NOWAIT | LK_INTERLOCK;
1003 	if (waitfor == MNT_WAIT) {
1004 		wait = 1;
1005 		lockreq = LK_EXCLUSIVE | LK_INTERLOCK;
1006 	}
1007 	mtx_lock(&mntvnode_mtx);
1008 loop:
1009 	for (vp = LIST_FIRST(&mp->mnt_vnodelist); vp != NULL; vp = nvp) {
1010 		/*
1011 		 * If the vnode that we are about to sync is no longer
1012 		 * associated with this mount point, start over.
1013 		 */
1014 		if (vp->v_mount != mp)
1015 			goto loop;
1016 		nvp = LIST_NEXT(vp, v_mntvnodes);
1017 
1018 		mtx_unlock(&mntvnode_mtx);
1019 		mtx_lock(&vp->v_interlock);
1020 		ip = VTOI(vp);
1021 		if (vp->v_type == VNON || ((ip->i_flag &
1022 		     (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) == 0 &&
1023 		     TAILQ_EMPTY(&vp->v_dirtyblkhd))) {
1024 			mtx_unlock(&vp->v_interlock);
1025 			mtx_lock(&mntvnode_mtx);
1026 			continue;
1027 		}
1028 		if (vp->v_type != VCHR) {
1029 			if ((error = vget(vp, lockreq, p)) != 0) {
1030 				mtx_lock(&mntvnode_mtx);
1031 				if (error == ENOENT)
1032 					goto loop;
1033 				continue;
1034 			}
1035 			if ((error = VOP_FSYNC(vp, cred, waitfor, p)) != 0)
1036 				allerror = error;
1037 			VOP_UNLOCK(vp, 0, p);
1038 			vrele(vp);
1039 		} else {
1040 			mtx_unlock(&vp->v_interlock);
1041 			UFS_UPDATE(vp, wait);
1042 		}
1043 		mtx_lock(&mntvnode_mtx);
1044 	}
1045 	mtx_unlock(&mntvnode_mtx);
1046 	/*
1047 	 * Force stale file system control information to be flushed.
1048 	 */
1049 	if (waitfor == MNT_WAIT) {
1050 		if ((error = softdep_flushworklist(ump->um_mountp, &count, p)))
1051 			allerror = error;
1052 		/* Flushed work items may create new vnodes to clean */
1053 		if (count) {
1054 			mtx_lock(&mntvnode_mtx);
1055 			goto loop;
1056 		}
1057 	}
1058 #ifdef QUOTA
1059 	qsync(mp);
1060 #endif
1061 	devvp = ump->um_devvp;
1062 	mtx_lock(&devvp->v_interlock);
1063 	if (waitfor != MNT_LAZY &&
1064 	    (devvp->v_numoutput > 0 || TAILQ_FIRST(&devvp->v_dirtyblkhd))) {
1065 		mtx_unlock(&devvp->v_interlock);
1066 		vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY, p);
1067 		if ((error = VOP_FSYNC(devvp, cred, waitfor, p)) != 0)
1068 			allerror = error;
1069 		VOP_UNLOCK(devvp, 0, p);
1070 		if (waitfor == MNT_WAIT) {
1071 			mtx_lock(&mntvnode_mtx);
1072 			goto loop;
1073 		}
1074 	} else
1075 		mtx_unlock(&devvp->v_interlock);
1076 	/*
1077 	 * Write back modified superblock.
1078 	 */
1079 	if (fs->fs_fmod != 0 && (error = ffs_sbupdate(ump, waitfor)) != 0)
1080 		allerror = error;
1081 	return (allerror);
1082 }
1083 
1084 /*
1085  * Look up a FFS dinode number to find its incore vnode, otherwise read it
1086  * in from disk.  If it is in core, wait for the lock bit to clear, then
1087  * return the inode locked.  Detection and handling of mount points must be
1088  * done by the calling routine.
1089  */
1090 static int ffs_inode_hash_lock;
1091 /*
1092  * ffs_inode_hash_lock is a variable to manage mutual exclusion
1093  * of vnode allocation and intertion to the hash, especially to
1094  * avoid holding more than one vnodes for the same inode in the
1095  * hash table. ffs_inode_hash_lock must hence be tested-and-set
1096  * or cleared atomically, accomplished by ffs_inode_hash_mtx.
1097  *
1098  * As vnode allocation may block during MALLOC() and zone
1099  * allocation, we should also do msleep() to give away the CPU
1100  * if anyone else is allocating a vnode. lockmgr is not suitable
1101  * here because someone else may insert to the hash table the
1102  * vnode we are trying to allocate during our sleep, in which
1103  * case the hash table needs to be examined once again after
1104  * waking up.
1105  */
1106 static struct mtx ffs_inode_hash_mtx;
1107 
1108 int
1109 ffs_vget(mp, ino, vpp)
1110 	struct mount *mp;
1111 	ino_t ino;
1112 	struct vnode **vpp;
1113 {
1114 	struct fs *fs;
1115 	struct inode *ip;
1116 	struct ufsmount *ump;
1117 	struct buf *bp;
1118 	struct vnode *vp;
1119 	dev_t dev;
1120 	int error, want_wakeup;
1121 
1122 	ump = VFSTOUFS(mp);
1123 	dev = ump->um_dev;
1124 restart:
1125 	if ((*vpp = ufs_ihashget(dev, ino)) != NULL) {
1126 		return (0);
1127 	}
1128 
1129 	/*
1130 	 * Lock out the creation of new entries in the FFS hash table in
1131 	 * case getnewvnode() or MALLOC() blocks, otherwise a duplicate
1132 	 * may occur!
1133 	 */
1134 	mtx_lock(&ffs_inode_hash_mtx);
1135 	if (ffs_inode_hash_lock) {
1136 		while (ffs_inode_hash_lock) {
1137 			ffs_inode_hash_lock = -1;
1138 			msleep(&ffs_inode_hash_lock, &ffs_inode_hash_mtx, PVM, "ffsvgt", 0);
1139 		}
1140 		mtx_unlock(&ffs_inode_hash_mtx);
1141 		goto restart;
1142 	}
1143 	ffs_inode_hash_lock = 1;
1144 	mtx_unlock(&ffs_inode_hash_mtx);
1145 
1146 	/*
1147 	 * If this MALLOC() is performed after the getnewvnode()
1148 	 * it might block, leaving a vnode with a NULL v_data to be
1149 	 * found by ffs_sync() if a sync happens to fire right then,
1150 	 * which will cause a panic because ffs_sync() blindly
1151 	 * dereferences vp->v_data (as well it should).
1152 	 */
1153 	MALLOC(ip, struct inode *, sizeof(struct inode),
1154 	    ump->um_malloctype, M_WAITOK);
1155 
1156 	/* Allocate a new vnode/inode. */
1157 	error = getnewvnode(VT_UFS, mp, ffs_vnodeop_p, &vp);
1158 	if (error) {
1159 		/*
1160 		 * Do not wake up processes while holding the mutex,
1161 		 * otherwise the processes waken up immediately hit
1162 		 * themselves into the mutex.
1163 		 */
1164 		mtx_lock(&ffs_inode_hash_mtx);
1165 		want_wakeup = ffs_inode_hash_lock < 0;
1166 		ffs_inode_hash_lock = 0;
1167 		mtx_unlock(&ffs_inode_hash_mtx);
1168 		if (want_wakeup)
1169 			wakeup(&ffs_inode_hash_lock);
1170 		*vpp = NULL;
1171 		FREE(ip, ump->um_malloctype);
1172 		return (error);
1173 	}
1174 	bzero((caddr_t)ip, sizeof(struct inode));
1175 	/*
1176 	 * FFS supports lock sharing in the stack of vnodes
1177 	 */
1178 	vp->v_vnlock = &vp->v_lock;
1179 	lockinit(vp->v_vnlock, PINOD, "inode", 0, LK_CANRECURSE);
1180 	vp->v_data = ip;
1181 	ip->i_vnode = vp;
1182 	ip->i_fs = fs = ump->um_fs;
1183 	ip->i_dev = dev;
1184 	ip->i_number = ino;
1185 #ifdef QUOTA
1186 	{
1187 		int i;
1188 		for (i = 0; i < MAXQUOTAS; i++)
1189 			ip->i_dquot[i] = NODQUOT;
1190 	}
1191 #endif
1192 	/*
1193 	 * Put it onto its hash chain and lock it so that other requests for
1194 	 * this inode will block if they arrive while we are sleeping waiting
1195 	 * for old data structures to be purged or for the contents of the
1196 	 * disk portion of this inode to be read.
1197 	 */
1198 	ufs_ihashins(ip);
1199 
1200 	/*
1201 	 * Do not wake up processes while holding the mutex,
1202 	 * otherwise the processes waken up immediately hit
1203 	 * themselves into the mutex.
1204 	 */
1205 	mtx_lock(&ffs_inode_hash_mtx);
1206 	want_wakeup = ffs_inode_hash_lock < 0;
1207 	ffs_inode_hash_lock = 0;
1208 	mtx_unlock(&ffs_inode_hash_mtx);
1209 	if (want_wakeup)
1210 		wakeup(&ffs_inode_hash_lock);
1211 
1212 	/* Read in the disk contents for the inode, copy into the inode. */
1213 	error = bread(ump->um_devvp, fsbtodb(fs, ino_to_fsba(fs, ino)),
1214 	    (int)fs->fs_bsize, NOCRED, &bp);
1215 	if (error) {
1216 		/*
1217 		 * The inode does not contain anything useful, so it would
1218 		 * be misleading to leave it on its hash chain. With mode
1219 		 * still zero, it will be unlinked and returned to the free
1220 		 * list by vput().
1221 		 */
1222 		brelse(bp);
1223 		vput(vp);
1224 		*vpp = NULL;
1225 		return (error);
1226 	}
1227 	ip->i_din = *((struct dinode *)bp->b_data + ino_to_fsbo(fs, ino));
1228 	if (DOINGSOFTDEP(vp))
1229 		softdep_load_inodeblock(ip);
1230 	else
1231 		ip->i_effnlink = ip->i_nlink;
1232 	bqrelse(bp);
1233 
1234 	/*
1235 	 * Initialize the vnode from the inode, check for aliases.
1236 	 * Note that the underlying vnode may have changed.
1237 	 */
1238 	error = ufs_vinit(mp, ffs_specop_p, ffs_fifoop_p, &vp);
1239 	if (error) {
1240 		vput(vp);
1241 		*vpp = NULL;
1242 		return (error);
1243 	}
1244 	/*
1245 	 * Finish inode initialization now that aliasing has been resolved.
1246 	 */
1247 	ip->i_devvp = ump->um_devvp;
1248 	VREF(ip->i_devvp);
1249 	/*
1250 	 * Set up a generation number for this inode if it does not
1251 	 * already have one. This should only happen on old filesystems.
1252 	 */
1253 	if (ip->i_gen == 0) {
1254 		ip->i_gen = random() / 2 + 1;
1255 		if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0)
1256 			ip->i_flag |= IN_MODIFIED;
1257 	}
1258 	/*
1259 	 * Ensure that uid and gid are correct. This is a temporary
1260 	 * fix until fsck has been changed to do the update.
1261 	 */
1262 	if (fs->fs_inodefmt < FS_44INODEFMT) {		/* XXX */
1263 		ip->i_uid = ip->i_din.di_ouid;		/* XXX */
1264 		ip->i_gid = ip->i_din.di_ogid;		/* XXX */
1265 	}						/* XXX */
1266 
1267 	*vpp = vp;
1268 	return (0);
1269 }
1270 
1271 /*
1272  * File handle to vnode
1273  *
1274  * Have to be really careful about stale file handles:
1275  * - check that the inode number is valid
1276  * - call ffs_vget() to get the locked inode
1277  * - check for an unallocated inode (i_mode == 0)
1278  * - check that the given client host has export rights and return
1279  *   those rights via. exflagsp and credanonp
1280  */
1281 int
1282 ffs_fhtovp(mp, fhp, vpp)
1283 	register struct mount *mp;
1284 	struct fid *fhp;
1285 	struct vnode **vpp;
1286 {
1287 	register struct ufid *ufhp;
1288 	struct fs *fs;
1289 
1290 	ufhp = (struct ufid *)fhp;
1291 	fs = VFSTOUFS(mp)->um_fs;
1292 	if (ufhp->ufid_ino < ROOTINO ||
1293 	    ufhp->ufid_ino >= fs->fs_ncg * fs->fs_ipg)
1294 		return (ESTALE);
1295 	return (ufs_fhtovp(mp, ufhp, vpp));
1296 }
1297 
1298 /*
1299  * Vnode pointer to File handle
1300  */
1301 /* ARGSUSED */
1302 int
1303 ffs_vptofh(vp, fhp)
1304 	struct vnode *vp;
1305 	struct fid *fhp;
1306 {
1307 	register struct inode *ip;
1308 	register struct ufid *ufhp;
1309 
1310 	ip = VTOI(vp);
1311 	ufhp = (struct ufid *)fhp;
1312 	ufhp->ufid_len = sizeof(struct ufid);
1313 	ufhp->ufid_ino = ip->i_number;
1314 	ufhp->ufid_gen = ip->i_gen;
1315 	return (0);
1316 }
1317 
1318 /*
1319  * Initialize the filesystem; just use ufs_init.
1320  */
1321 static int
1322 ffs_init(vfsp)
1323 	struct vfsconf *vfsp;
1324 {
1325 
1326 	softdep_initialize();
1327 	mtx_init(&ffs_inode_hash_mtx, "ifsvgt", MTX_DEF);
1328 	return (ufs_init(vfsp));
1329 }
1330 
1331 /*
1332  * Write a superblock and associated information back to disk.
1333  */
1334 static int
1335 ffs_sbupdate(mp, waitfor)
1336 	struct ufsmount *mp;
1337 	int waitfor;
1338 {
1339 	register struct fs *dfs, *fs = mp->um_fs;
1340 	register struct buf *bp;
1341 	int blks;
1342 	void *space;
1343 	int i, size, error, allerror = 0;
1344 
1345 	/*
1346 	 * First write back the summary information.
1347 	 */
1348 	blks = howmany(fs->fs_cssize, fs->fs_fsize);
1349 	space = fs->fs_csp;
1350 	for (i = 0; i < blks; i += fs->fs_frag) {
1351 		size = fs->fs_bsize;
1352 		if (i + fs->fs_frag > blks)
1353 			size = (blks - i) * fs->fs_fsize;
1354 		bp = getblk(mp->um_devvp, fsbtodb(fs, fs->fs_csaddr + i),
1355 		    size, 0, 0);
1356 		bcopy(space, bp->b_data, (u_int)size);
1357 		space = (char *)space + size;
1358 		if (waitfor != MNT_WAIT)
1359 			bawrite(bp);
1360 		else if ((error = bwrite(bp)) != 0)
1361 			allerror = error;
1362 	}
1363 	/*
1364 	 * Now write back the superblock itself. If any errors occurred
1365 	 * up to this point, then fail so that the superblock avoids
1366 	 * being written out as clean.
1367 	 */
1368 	if (allerror)
1369 		return (allerror);
1370 	bp = getblk(mp->um_devvp, SBLOCK, (int)fs->fs_sbsize, 0, 0);
1371 	fs->fs_fmod = 0;
1372 	fs->fs_time = time_second;
1373 	bcopy((caddr_t)fs, bp->b_data, (u_int)fs->fs_sbsize);
1374 	/* Restore compatibility to old file systems.		   XXX */
1375 	dfs = (struct fs *)bp->b_data;				/* XXX */
1376 	if (fs->fs_postblformat == FS_42POSTBLFMT)		/* XXX */
1377 		dfs->fs_nrpos = -1;				/* XXX */
1378 	if (fs->fs_inodefmt < FS_44INODEFMT) {			/* XXX */
1379 		int32_t *lp, tmp;				/* XXX */
1380 								/* XXX */
1381 		lp = (int32_t *)&dfs->fs_qbmask;		/* XXX */
1382 		tmp = lp[4];					/* XXX */
1383 		for (i = 4; i > 0; i--)				/* XXX */
1384 			lp[i] = lp[i-1];			/* XXX */
1385 		lp[0] = tmp;					/* XXX */
1386 	}							/* XXX */
1387 	dfs->fs_maxfilesize = mp->um_savedmaxfilesize;		/* XXX */
1388 	if (waitfor != MNT_WAIT)
1389 		bawrite(bp);
1390 	else if ((error = bwrite(bp)) != 0)
1391 		allerror = error;
1392 	return (allerror);
1393 }
1394