xref: /freebsd/sys/ufs/ffs/ffs_vfsops.c (revision f9218d3d4fd34f082473b3a021c6d4d109fb47cf)
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_mac.h"
38 #include "opt_quota.h"
39 #include "opt_ufs.h"
40 
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/stdint.h>
44 #include <sys/namei.h>
45 #include <sys/proc.h>
46 #include <sys/kernel.h>
47 #include <sys/mac.h>
48 #include <sys/vnode.h>
49 #include <sys/mount.h>
50 #include <sys/bio.h>
51 #include <sys/buf.h>
52 #include <sys/conf.h>
53 #include <sys/fcntl.h>
54 #include <sys/disk.h>
55 #include <sys/malloc.h>
56 #include <sys/mutex.h>
57 
58 #include <ufs/ufs/extattr.h>
59 #include <ufs/ufs/quota.h>
60 #include <ufs/ufs/ufsmount.h>
61 #include <ufs/ufs/inode.h>
62 #include <ufs/ufs/ufs_extern.h>
63 
64 #include <ufs/ffs/fs.h>
65 #include <ufs/ffs/ffs_extern.h>
66 
67 #include <vm/vm.h>
68 #include <vm/uma.h>
69 #include <vm/vm_page.h>
70 
71 uma_zone_t uma_inode, uma_ufs1, uma_ufs2;
72 
73 static int	ffs_sbupdate(struct ufsmount *, int);
74        int	ffs_reload(struct mount *,struct ucred *,struct thread *);
75 static int	ffs_mountfs(struct vnode *, struct mount *, struct thread *);
76 static void	ffs_oldfscompat_read(struct fs *, struct ufsmount *,
77 		    ufs2_daddr_t);
78 static void	ffs_oldfscompat_write(struct fs *, struct ufsmount *);
79 static void	ffs_ifree(struct ufsmount *ump, struct inode *ip);
80 static vfs_init_t ffs_init;
81 static vfs_uninit_t ffs_uninit;
82 static vfs_extattrctl_t ffs_extattrctl;
83 
84 static struct vfsops ufs_vfsops = {
85 	ffs_mount,
86 	ufs_start,
87 	ffs_unmount,
88 	ufs_root,
89 	ufs_quotactl,
90 	ffs_statfs,
91 	ffs_sync,
92 	ffs_vget,
93 	ffs_fhtovp,
94 	vfs_stdcheckexp,
95 	ffs_vptofh,
96 	ffs_init,
97 	ffs_uninit,
98 	ffs_extattrctl,
99 };
100 
101 VFS_SET(ufs_vfsops, ufs, 0);
102 
103 /*
104  * ffs_mount
105  *
106  * Called when mounting local physical media
107  *
108  * PARAMETERS:
109  *		mountroot
110  *			mp	mount point structure
111  *			path	NULL (flag for root mount!!!)
112  *			data	<unused>
113  *			ndp	<unused>
114  *			p	process (user credentials check [statfs])
115  *
116  *		mount
117  *			mp	mount point structure
118  *			path	path to mount point
119  *			data	pointer to argument struct in user space
120  *			ndp	mount point namei() return (used for
121  *				credentials on reload), reused to look
122  *				up block device.
123  *			p	process (user credentials check)
124  *
125  * RETURNS:	0	Success
126  *		!0	error number (errno.h)
127  *
128  * LOCK STATE:
129  *
130  *		ENTRY
131  *			mount point is locked
132  *		EXIT
133  *			mount point is locked
134  *
135  * NOTES:
136  *		A NULL path can be used for a flag since the mount
137  *		system call will fail with EFAULT in copyinstr in
138  *		namei() if it is a genuine NULL from the user.
139  */
140 int
141 ffs_mount(mp, path, data, ndp, td)
142         struct mount		*mp;	/* mount struct pointer*/
143         char			*path;	/* path to mount point*/
144         caddr_t			data;	/* arguments to FS specific mount*/
145         struct nameidata	*ndp;	/* mount point credentials*/
146         struct thread		*td;	/* process requesting mount*/
147 {
148 	size_t size;
149 	struct vnode *devvp;
150 	struct ufs_args args;
151 	struct ufsmount *ump = 0;
152 	struct fs *fs;
153 	int error, flags;
154 	mode_t accessmode;
155 
156 	if (uma_inode == NULL) {
157 		uma_inode = uma_zcreate("FFS inode",
158 		    sizeof(struct inode), NULL, NULL, NULL, NULL,
159 		    UMA_ALIGN_PTR, 0);
160 		uma_ufs1 = uma_zcreate("FFS1 dinode",
161 		    sizeof(struct ufs1_dinode), NULL, NULL, NULL, NULL,
162 		    UMA_ALIGN_PTR, 0);
163 		uma_ufs2 = uma_zcreate("FFS2 dinode",
164 		    sizeof(struct ufs2_dinode), NULL, NULL, NULL, NULL,
165 		    UMA_ALIGN_PTR, 0);
166 	}
167 	/*
168 	 * Use NULL path to indicate we are mounting the root filesystem.
169 	 */
170 	if (path == NULL) {
171 		if ((error = bdevvp(rootdev, &rootvp))) {
172 			printf("ffs_mountroot: can't find rootvp\n");
173 			return (error);
174 		}
175 
176 		if ((error = ffs_mountfs(rootvp, mp, td)) != 0)
177 			return (error);
178 		(void)VFS_STATFS(mp, &mp->mnt_stat, td);
179 		return (0);
180 	}
181 
182 	/*
183 	 * Mounting non-root filesystem or updating a filesystem
184 	 */
185 	if ((error = copyin(data, (caddr_t)&args, sizeof(struct ufs_args)))!= 0)
186 		return (error);
187 
188 	/*
189 	 * If updating, check whether changing from read-only to
190 	 * read/write; if there is no device name, that's all we do.
191 	 */
192 	if (mp->mnt_flag & MNT_UPDATE) {
193 		ump = VFSTOUFS(mp);
194 		fs = ump->um_fs;
195 		devvp = ump->um_devvp;
196 		if (fs->fs_ronly == 0 && (mp->mnt_flag & MNT_RDONLY)) {
197 			if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0)
198 				return (error);
199 			/*
200 			 * Flush any dirty data.
201 			 */
202 			if ((error = VFS_SYNC(mp, MNT_WAIT,
203 			    td->td_proc->p_ucred, td)) != 0) {
204 				vn_finished_write(mp);
205 				return (error);
206 			}
207 			/*
208 			 * Check for and optionally get rid of files open
209 			 * for writing.
210 			 */
211 			flags = WRITECLOSE;
212 			if (mp->mnt_flag & MNT_FORCE)
213 				flags |= FORCECLOSE;
214 			if (mp->mnt_flag & MNT_SOFTDEP) {
215 				error = softdep_flushfiles(mp, flags, td);
216 			} else {
217 				error = ffs_flushfiles(mp, flags, td);
218 			}
219 			if (error) {
220 				vn_finished_write(mp);
221 				return (error);
222 			}
223 			if (fs->fs_pendingblocks != 0 ||
224 			    fs->fs_pendinginodes != 0) {
225 				printf("%s: %s: blocks %jd files %d\n",
226 				    fs->fs_fsmnt, "update error",
227 				    (intmax_t)fs->fs_pendingblocks,
228 				    fs->fs_pendinginodes);
229 				fs->fs_pendingblocks = 0;
230 				fs->fs_pendinginodes = 0;
231 			}
232 			fs->fs_ronly = 1;
233 			if ((fs->fs_flags & (FS_UNCLEAN | FS_NEEDSFSCK)) == 0)
234 				fs->fs_clean = 1;
235 			if ((error = ffs_sbupdate(ump, MNT_WAIT)) != 0) {
236 				fs->fs_ronly = 0;
237 				fs->fs_clean = 0;
238 				vn_finished_write(mp);
239 				return (error);
240 			}
241 			vn_finished_write(mp);
242 		}
243 		if ((mp->mnt_flag & MNT_RELOAD) &&
244 		    (error = ffs_reload(mp, ndp->ni_cnd.cn_cred, td)) != 0)
245 			return (error);
246 		if (fs->fs_ronly && (mp->mnt_kern_flag & MNTK_WANTRDWR)) {
247 			/*
248 			 * If upgrade to read-write by non-root, then verify
249 			 * that user has necessary permissions on the device.
250 			 */
251 			if (suser(td)) {
252 				vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY, td);
253 				if ((error = VOP_ACCESS(devvp, VREAD | VWRITE,
254 				    td->td_ucred, td)) != 0) {
255 					VOP_UNLOCK(devvp, 0, td);
256 					return (error);
257 				}
258 				VOP_UNLOCK(devvp, 0, td);
259 			}
260 			fs->fs_flags &= ~FS_UNCLEAN;
261 			if (fs->fs_clean == 0) {
262 				fs->fs_flags |= FS_UNCLEAN;
263 				if ((mp->mnt_flag & MNT_FORCE) ||
264 				    ((fs->fs_flags & FS_NEEDSFSCK) == 0 &&
265 				     (fs->fs_flags & FS_DOSOFTDEP))) {
266 					printf("WARNING: %s was not %s\n",
267 					   fs->fs_fsmnt, "properly dismounted");
268 				} else {
269 					printf(
270 "WARNING: R/W mount of %s denied.  Filesystem is not clean - run fsck\n",
271 					    fs->fs_fsmnt);
272 					return (EPERM);
273 				}
274 			}
275 			if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0)
276 				return (error);
277 			fs->fs_ronly = 0;
278 			fs->fs_clean = 0;
279 			if ((error = ffs_sbupdate(ump, MNT_WAIT)) != 0) {
280 				vn_finished_write(mp);
281 				return (error);
282 			}
283 			/* check to see if we need to start softdep */
284 			if ((fs->fs_flags & FS_DOSOFTDEP) &&
285 			    (error = softdep_mount(devvp, mp, fs, td->td_ucred))){
286 				vn_finished_write(mp);
287 				return (error);
288 			}
289 			if (fs->fs_snapinum[0] != 0)
290 				ffs_snapshot_mount(mp);
291 			vn_finished_write(mp);
292 		}
293 		/*
294 		 * Soft updates is incompatible with "async",
295 		 * so if we are doing softupdates stop the user
296 		 * from setting the async flag in an update.
297 		 * Softdep_mount() clears it in an initial mount
298 		 * or ro->rw remount.
299 		 */
300 		if (mp->mnt_flag & MNT_SOFTDEP)
301 			mp->mnt_flag &= ~MNT_ASYNC;
302 		/*
303 		 * If not updating name, process export requests.
304 		 */
305 		if (args.fspec == 0)
306 			return (vfs_export(mp, &args.export));
307 		/*
308 		 * If this is a snapshot request, take the snapshot.
309 		 */
310 		if (mp->mnt_flag & MNT_SNAPSHOT)
311 			return (ffs_snapshot(mp, args.fspec));
312 	}
313 
314 	/*
315 	 * Not an update, or updating the name: look up the name
316 	 * and verify that it refers to a sensible block device.
317 	 */
318 	NDINIT(ndp, LOOKUP, FOLLOW, UIO_USERSPACE, args.fspec, td);
319 	if ((error = namei(ndp)) != 0)
320 		return (error);
321 	NDFREE(ndp, NDF_ONLY_PNBUF);
322 	devvp = ndp->ni_vp;
323 	if (!vn_isdisk(devvp, &error)) {
324 		vrele(devvp);
325 		return (error);
326 	}
327 
328 	/*
329 	 * If mount by non-root, then verify that user has necessary
330 	 * permissions on the device.
331 	 */
332 	if (suser(td)) {
333 		accessmode = VREAD;
334 		if ((mp->mnt_flag & MNT_RDONLY) == 0)
335 			accessmode |= VWRITE;
336 		vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY, td);
337 		if ((error = VOP_ACCESS(devvp, accessmode, td->td_ucred, td))!= 0){
338 			vput(devvp);
339 			return (error);
340 		}
341 		VOP_UNLOCK(devvp, 0, td);
342 	}
343 
344 	if (mp->mnt_flag & MNT_UPDATE) {
345 		/*
346 		 * Update only
347 		 *
348 		 * If it's not the same vnode, or at least the same device
349 		 * then it's not correct.
350 		 */
351 
352 		if (devvp != ump->um_devvp &&
353 		    devvp->v_rdev != ump->um_devvp->v_rdev)
354 			error = EINVAL;	/* needs translation */
355 		vrele(devvp);
356 		if (error)
357 			return (error);
358 	} else {
359 		/*
360 		 * New mount
361 		 *
362 		 * We need the name for the mount point (also used for
363 		 * "last mounted on") copied in. If an error occurs,
364 		 * the mount point is discarded by the upper level code.
365 		 * Note that vfs_mount() populates f_mntonname for us.
366 		 */
367 		if ((error = ffs_mountfs(devvp, mp, td)) != 0) {
368 			vrele(devvp);
369 			return (error);
370 		}
371 	}
372 	/*
373 	 * Save "mounted from" device name info for mount point (NULL pad).
374 	 */
375 	copyinstr(args.fspec, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, &size);
376 	bzero( mp->mnt_stat.f_mntfromname + size, MNAMELEN - size);
377 	/*
378 	 * Initialize filesystem stat information in mount struct.
379 	 */
380 	(void)VFS_STATFS(mp, &mp->mnt_stat, td);
381 	return (0);
382 }
383 
384 /*
385  * Reload all incore data for a filesystem (used after running fsck on
386  * the root filesystem and finding things to fix). The filesystem must
387  * be mounted read-only.
388  *
389  * Things to do to update the mount:
390  *	1) invalidate all cached meta-data.
391  *	2) re-read superblock from disk.
392  *	3) re-read summary information from disk.
393  *	4) invalidate all inactive vnodes.
394  *	5) invalidate all cached file data.
395  *	6) re-read inode data for all active vnodes.
396  */
397 int
398 ffs_reload(mp, cred, td)
399 	struct mount *mp;
400 	struct ucred *cred;
401 	struct thread *td;
402 {
403 	struct vnode *vp, *nvp, *devvp;
404 	struct inode *ip;
405 	void *space;
406 	struct buf *bp;
407 	struct fs *fs, *newfs;
408 	dev_t dev;
409 	ufs2_daddr_t sblockloc;
410 	int i, blks, size, error;
411 	int32_t *lp;
412 
413 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
414 		return (EINVAL);
415 	/*
416 	 * Step 1: invalidate all cached meta-data.
417 	 */
418 	devvp = VFSTOUFS(mp)->um_devvp;
419 	vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY, td);
420 	error = vinvalbuf(devvp, 0, cred, td, 0, 0);
421 	VOP_UNLOCK(devvp, 0, td);
422 	if (error)
423 		panic("ffs_reload: dirty1");
424 
425 	dev = devvp->v_rdev;
426 
427 	/*
428 	 * Only VMIO the backing device if the backing device is a real
429 	 * block device.
430 	 */
431 	if (vn_isdisk(devvp, NULL)) {
432 		vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY, td);
433 		vfs_object_create(devvp, td, td->td_ucred);
434 		/* XXX Why lock only to release immediately?? */
435 		mtx_lock(&devvp->v_interlock);
436 		VOP_UNLOCK(devvp, LK_INTERLOCK, td);
437 	}
438 
439 	/*
440 	 * Step 2: re-read superblock from disk.
441 	 */
442 	fs = VFSTOUFS(mp)->um_fs;
443 	if ((error = bread(devvp, btodb(fs->fs_sblockloc), fs->fs_sbsize,
444 	    NOCRED, &bp)) != 0)
445 		return (error);
446 	newfs = (struct fs *)bp->b_data;
447 	if ((newfs->fs_magic != FS_UFS1_MAGIC &&
448 	     newfs->fs_magic != FS_UFS2_MAGIC) ||
449 	    newfs->fs_bsize > MAXBSIZE ||
450 	    newfs->fs_bsize < sizeof(struct fs)) {
451 			brelse(bp);
452 			return (EIO);		/* XXX needs translation */
453 	}
454 	/*
455 	 * Copy pointer fields back into superblock before copying in	XXX
456 	 * new superblock. These should really be in the ufsmount.	XXX
457 	 * Note that important parameters (eg fs_ncg) are unchanged.
458 	 */
459 	newfs->fs_csp = fs->fs_csp;
460 	newfs->fs_maxcluster = fs->fs_maxcluster;
461 	newfs->fs_contigdirs = fs->fs_contigdirs;
462 	newfs->fs_active = fs->fs_active;
463 	sblockloc = fs->fs_sblockloc;
464 	bcopy(newfs, fs, (u_int)fs->fs_sbsize);
465 	brelse(bp);
466 	mp->mnt_maxsymlinklen = fs->fs_maxsymlinklen;
467 	ffs_oldfscompat_read(fs, VFSTOUFS(mp), sblockloc);
468 	if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
469 		printf("%s: reload pending error: blocks %jd files %d\n",
470 		    fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
471 		    fs->fs_pendinginodes);
472 		fs->fs_pendingblocks = 0;
473 		fs->fs_pendinginodes = 0;
474 	}
475 
476 	/*
477 	 * Step 3: re-read summary information from disk.
478 	 */
479 	blks = howmany(fs->fs_cssize, fs->fs_fsize);
480 	space = fs->fs_csp;
481 	for (i = 0; i < blks; i += fs->fs_frag) {
482 		size = fs->fs_bsize;
483 		if (i + fs->fs_frag > blks)
484 			size = (blks - i) * fs->fs_fsize;
485 		error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), size,
486 		    NOCRED, &bp);
487 		if (error)
488 			return (error);
489 		bcopy(bp->b_data, space, (u_int)size);
490 		space = (char *)space + size;
491 		brelse(bp);
492 	}
493 	/*
494 	 * We no longer know anything about clusters per cylinder group.
495 	 */
496 	if (fs->fs_contigsumsize > 0) {
497 		lp = fs->fs_maxcluster;
498 		for (i = 0; i < fs->fs_ncg; i++)
499 			*lp++ = fs->fs_contigsumsize;
500 	}
501 
502 loop:
503 	mtx_lock(&mntvnode_mtx);
504 	for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); vp != NULL; vp = nvp) {
505 		if (vp->v_mount != mp) {
506 			mtx_unlock(&mntvnode_mtx);
507 			goto loop;
508 		}
509 		nvp = TAILQ_NEXT(vp, v_nmntvnodes);
510 		mtx_unlock(&mntvnode_mtx);
511 		/*
512 		 * Step 4: invalidate all inactive vnodes.
513 		 */
514 		if (vrecycle(vp, NULL, td))
515 			goto loop;
516 		/*
517 		 * Step 5: invalidate all cached file data.
518 		 */
519 		/* XXX Why lock only to release immediately? */
520 		mtx_lock(&vp->v_interlock);
521 		if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, td)) {
522 			goto loop;
523 		}
524 		if (vinvalbuf(vp, 0, cred, td, 0, 0))
525 			panic("ffs_reload: dirty2");
526 		/*
527 		 * Step 6: re-read inode data for all active vnodes.
528 		 */
529 		ip = VTOI(vp);
530 		error =
531 		    bread(devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
532 		    (int)fs->fs_bsize, NOCRED, &bp);
533 		if (error) {
534 			vput(vp);
535 			return (error);
536 		}
537 		ffs_load_inode(bp, ip, fs, ip->i_number);
538 		ip->i_effnlink = ip->i_nlink;
539 		brelse(bp);
540 		vput(vp);
541 		mtx_lock(&mntvnode_mtx);
542 	}
543 	mtx_unlock(&mntvnode_mtx);
544 	return (0);
545 }
546 
547 /*
548  * Possible superblock locations ordered from most to least likely.
549  */
550 static int sblock_try[] = SBLOCKSEARCH;
551 
552 /*
553  * Common code for mount and mountroot
554  */
555 static int
556 ffs_mountfs(devvp, mp, td)
557 	struct vnode *devvp;
558 	struct mount *mp;
559 	struct thread *td;
560 {
561 	struct ufsmount *ump;
562 	struct buf *bp;
563 	struct fs *fs;
564 	dev_t dev;
565 	void *space;
566 	ufs2_daddr_t sblockloc;
567 	int error, i, blks, size, ronly;
568 	int32_t *lp;
569 	struct ucred *cred;
570 	size_t strsize;
571 	int ncount;
572 
573 	dev = devvp->v_rdev;
574 	cred = td ? td->td_ucred : NOCRED;
575 	/*
576 	 * Disallow multiple mounts of the same device.
577 	 * Disallow mounting of a device that is currently in use
578 	 * (except for root, which might share swap device for miniroot).
579 	 * Flush out any old buffers remaining from a previous use.
580 	 */
581 	error = vfs_mountedon(devvp);
582 	if (error)
583 		return (error);
584 	ncount = vcount(devvp);
585 
586 	if (ncount > 1 && devvp != rootvp)
587 		return (EBUSY);
588 	vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY, td);
589 	error = vinvalbuf(devvp, V_SAVE, cred, td, 0, 0);
590 	VOP_UNLOCK(devvp, 0, td);
591 	if (error)
592 		return (error);
593 
594 	/*
595 	 * Only VMIO the backing device if the backing device is a real
596 	 * block device.
597 	 * Note that it is optional that the backing device be VMIOed.  This
598 	 * increases the opportunity for metadata caching.
599 	 */
600 	if (vn_isdisk(devvp, NULL)) {
601 		vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY, td);
602 		vfs_object_create(devvp, td, cred);
603 		/* XXX Why lock only to release immediately?? */
604 		mtx_lock(&devvp->v_interlock);
605 		VOP_UNLOCK(devvp, LK_INTERLOCK, td);
606 	}
607 
608 	ronly = (mp->mnt_flag & MNT_RDONLY) != 0;
609 	vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY, td);
610 	/*
611 	 * XXX: We don't re-VOP_OPEN in FREAD|FWRITE mode if the filesystem
612 	 * XXX: is subsequently remounted, so open it FREAD|FWRITE from the
613 	 * XXX: start to avoid getting trashed later on.
614 	 */
615 #ifdef notyet
616 	error = VOP_OPEN(devvp, ronly ? FREAD : FREAD|FWRITE, FSCRED, td);
617 #else
618 	error = VOP_OPEN(devvp, FREAD|FWRITE, FSCRED, td);
619 #endif
620 	VOP_UNLOCK(devvp, 0, td);
621 	if (error)
622 		return (error);
623 	if (devvp->v_rdev->si_iosize_max != 0)
624 		mp->mnt_iosize_max = devvp->v_rdev->si_iosize_max;
625 	if (mp->mnt_iosize_max > MAXPHYS)
626 		mp->mnt_iosize_max = MAXPHYS;
627 
628 	bp = NULL;
629 	ump = NULL;
630 	fs = NULL;
631 	sblockloc = 0;
632 	/*
633 	 * Try reading the superblock in each of its possible locations.
634 	 */
635 	for (i = 0; sblock_try[i] != -1; i++) {
636 		if ((error = bread(devvp, sblock_try[i] / DEV_BSIZE, SBLOCKSIZE,
637 		    cred, &bp)) != 0)
638 			goto out;
639 		fs = (struct fs *)bp->b_data;
640 		sblockloc = sblock_try[i];
641 		if ((fs->fs_magic == FS_UFS1_MAGIC ||
642 		     (fs->fs_magic == FS_UFS2_MAGIC &&
643 		      (fs->fs_sblockloc == sblockloc ||
644 		       (fs->fs_old_flags & FS_FLAGS_UPDATED) == 0))) &&
645 		    fs->fs_bsize <= MAXBSIZE &&
646 		    fs->fs_bsize >= sizeof(struct fs))
647 			break;
648 		brelse(bp);
649 		bp = NULL;
650 	}
651 	if (sblock_try[i] == -1) {
652 		error = EINVAL;		/* XXX needs translation */
653 		goto out;
654 	}
655 	fs->fs_fmod = 0;
656 	fs->fs_flags &= ~FS_INDEXDIRS;	/* no support for directory indicies */
657 	fs->fs_flags &= ~FS_UNCLEAN;
658 	if (fs->fs_clean == 0) {
659 		fs->fs_flags |= FS_UNCLEAN;
660 		if (ronly || (mp->mnt_flag & MNT_FORCE) ||
661 		    ((fs->fs_flags & FS_NEEDSFSCK) == 0 &&
662 		     (fs->fs_flags & FS_DOSOFTDEP))) {
663 			printf(
664 "WARNING: %s was not properly dismounted\n",
665 			    fs->fs_fsmnt);
666 		} else {
667 			printf(
668 "WARNING: R/W mount of %s denied.  Filesystem is not clean - run fsck\n",
669 			    fs->fs_fsmnt);
670 			error = EPERM;
671 			goto out;
672 		}
673 		if ((fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) &&
674 		    (mp->mnt_flag & MNT_FORCE)) {
675 			printf("%s: lost blocks %jd files %d\n", fs->fs_fsmnt,
676 			    (intmax_t)fs->fs_pendingblocks,
677 			    fs->fs_pendinginodes);
678 			fs->fs_pendingblocks = 0;
679 			fs->fs_pendinginodes = 0;
680 		}
681 	}
682 	if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
683 		printf("%s: mount pending error: blocks %jd files %d\n",
684 		    fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
685 		    fs->fs_pendinginodes);
686 		fs->fs_pendingblocks = 0;
687 		fs->fs_pendinginodes = 0;
688 	}
689 	ump = malloc(sizeof *ump, M_UFSMNT, M_WAITOK | M_ZERO);
690 	ump->um_fs = malloc((u_long)fs->fs_sbsize, M_UFSMNT,
691 	    M_WAITOK);
692 	if (fs->fs_magic == FS_UFS1_MAGIC) {
693 		ump->um_fstype = UFS1;
694 		ump->um_balloc = ffs_balloc_ufs1;
695 	} else {
696 		ump->um_fstype = UFS2;
697 		ump->um_balloc = ffs_balloc_ufs2;
698 	}
699 	ump->um_blkatoff = ffs_blkatoff;
700 	ump->um_truncate = ffs_truncate;
701 	ump->um_update = ffs_update;
702 	ump->um_valloc = ffs_valloc;
703 	ump->um_vfree = ffs_vfree;
704 	ump->um_ifree = ffs_ifree;
705 	bcopy(bp->b_data, ump->um_fs, (u_int)fs->fs_sbsize);
706 	if (fs->fs_sbsize < SBLOCKSIZE)
707 		bp->b_flags |= B_INVAL | B_NOCACHE;
708 	brelse(bp);
709 	bp = NULL;
710 	fs = ump->um_fs;
711 	ffs_oldfscompat_read(fs, ump, sblockloc);
712 	fs->fs_ronly = ronly;
713 	size = fs->fs_cssize;
714 	blks = howmany(size, fs->fs_fsize);
715 	if (fs->fs_contigsumsize > 0)
716 		size += fs->fs_ncg * sizeof(int32_t);
717 	size += fs->fs_ncg * sizeof(u_int8_t);
718 	space = malloc((u_long)size, M_UFSMNT, M_WAITOK);
719 	fs->fs_csp = space;
720 	for (i = 0; i < blks; i += fs->fs_frag) {
721 		size = fs->fs_bsize;
722 		if (i + fs->fs_frag > blks)
723 			size = (blks - i) * fs->fs_fsize;
724 		if ((error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), size,
725 		    cred, &bp)) != 0) {
726 			free(fs->fs_csp, M_UFSMNT);
727 			goto out;
728 		}
729 		bcopy(bp->b_data, space, (u_int)size);
730 		space = (char *)space + size;
731 		brelse(bp);
732 		bp = NULL;
733 	}
734 	if (fs->fs_contigsumsize > 0) {
735 		fs->fs_maxcluster = lp = space;
736 		for (i = 0; i < fs->fs_ncg; i++)
737 			*lp++ = fs->fs_contigsumsize;
738 		space = lp;
739 	}
740 	size = fs->fs_ncg * sizeof(u_int8_t);
741 	fs->fs_contigdirs = (u_int8_t *)space;
742 	bzero(fs->fs_contigdirs, size);
743 	fs->fs_active = NULL;
744 	mp->mnt_data = (qaddr_t)ump;
745 	mp->mnt_stat.f_fsid.val[0] = fs->fs_id[0];
746 	mp->mnt_stat.f_fsid.val[1] = fs->fs_id[1];
747 	if (fs->fs_id[0] == 0 || fs->fs_id[1] == 0 ||
748 	    vfs_getvfs(&mp->mnt_stat.f_fsid))
749 		vfs_getnewfsid(mp);
750 	mp->mnt_maxsymlinklen = fs->fs_maxsymlinklen;
751 	mp->mnt_flag |= MNT_LOCAL;
752 	if ((fs->fs_flags & FS_MULTILABEL) != 0)
753 #ifdef MAC
754 		mp->mnt_flag |= MNT_MULTILABEL;
755 #else
756 		printf(
757 "WARNING: %s: multilabel flag on fs but no MAC support\n",
758 		    fs->fs_fsmnt);
759 #endif
760 	if ((fs->fs_flags & FS_ACLS) != 0)
761 #ifdef UFS_ACL
762 		mp->mnt_flag |= MNT_ACLS;
763 #else
764 		printf(
765 "WARNING: %s: ACLs flag on fs but no ACLs support\n",
766 		    fs->fs_fsmnt);
767 #endif
768 	ump->um_mountp = mp;
769 	ump->um_dev = dev;
770 	ump->um_devvp = devvp;
771 	ump->um_nindir = fs->fs_nindir;
772 	ump->um_bptrtodb = fs->fs_fsbtodb;
773 	ump->um_seqinc = fs->fs_frag;
774 	for (i = 0; i < MAXQUOTAS; i++)
775 		ump->um_quotas[i] = NULLVP;
776 #ifdef UFS_EXTATTR
777 	ufs_extattr_uepm_init(&ump->um_extattr);
778 #endif
779 	devvp->v_rdev->si_mountpoint = mp;
780 
781 	/*
782 	 * Set FS local "last mounted on" information (NULL pad)
783 	 */
784 	copystr(	mp->mnt_stat.f_mntonname,	/* mount point*/
785 			fs->fs_fsmnt,			/* copy area*/
786 			sizeof(fs->fs_fsmnt) - 1,	/* max size*/
787 			&strsize);			/* real size*/
788 	bzero( fs->fs_fsmnt + strsize, sizeof(fs->fs_fsmnt) - strsize);
789 
790 	if( mp->mnt_flag & MNT_ROOTFS) {
791 		/*
792 		 * Root mount; update timestamp in mount structure.
793 		 * this will be used by the common root mount code
794 		 * to update the system clock.
795 		 */
796 		mp->mnt_time = fs->fs_time;
797 	}
798 
799 	if (ronly == 0) {
800 		if ((fs->fs_flags & FS_DOSOFTDEP) &&
801 		    (error = softdep_mount(devvp, mp, fs, cred)) != 0) {
802 			free(fs->fs_csp, M_UFSMNT);
803 			goto out;
804 		}
805 		if (fs->fs_snapinum[0] != 0)
806 			ffs_snapshot_mount(mp);
807 		fs->fs_fmod = 1;
808 		fs->fs_clean = 0;
809 		(void) ffs_sbupdate(ump, MNT_WAIT);
810 	}
811 #ifdef UFS_EXTATTR
812 #ifdef UFS_EXTATTR_AUTOSTART
813 	/*
814 	 *
815 	 * Auto-starting does the following:
816 	 *	- check for /.attribute in the fs, and extattr_start if so
817 	 *	- for each file in .attribute, enable that file with
818 	 * 	  an attribute of the same name.
819 	 * Not clear how to report errors -- probably eat them.
820 	 * This would all happen while the filesystem was busy/not
821 	 * available, so would effectively be "atomic".
822 	 */
823 	(void) ufs_extattr_autostart(mp, td);
824 #endif /* !UFS_EXTATTR_AUTOSTART */
825 #endif /* !UFS_EXTATTR */
826 	return (0);
827 out:
828 	devvp->v_rdev->si_mountpoint = NULL;
829 	if (bp)
830 		brelse(bp);
831 	/* XXX: see comment above VOP_OPEN */
832 #ifdef notyet
833 	(void)VOP_CLOSE(devvp, ronly ? FREAD : FREAD|FWRITE, cred, td);
834 #else
835 	(void)VOP_CLOSE(devvp, FREAD|FWRITE, cred, td);
836 #endif
837 	if (ump) {
838 		free(ump->um_fs, M_UFSMNT);
839 		free(ump, M_UFSMNT);
840 		mp->mnt_data = (qaddr_t)0;
841 	}
842 	return (error);
843 }
844 
845 #include <sys/sysctl.h>
846 int bigcgs = 0;
847 SYSCTL_INT(_debug, OID_AUTO, bigcgs, CTLFLAG_RW, &bigcgs, 0, "");
848 
849 /*
850  * Sanity checks for loading old filesystem superblocks.
851  * See ffs_oldfscompat_write below for unwound actions.
852  *
853  * XXX - Parts get retired eventually.
854  * Unfortunately new bits get added.
855  */
856 static void
857 ffs_oldfscompat_read(fs, ump, sblockloc)
858 	struct fs *fs;
859 	struct ufsmount *ump;
860 	ufs2_daddr_t sblockloc;
861 {
862 	off_t maxfilesize;
863 
864 	/*
865 	 * If not yet done, update fs_flags location and value of fs_sblockloc.
866 	 */
867 	if ((fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) {
868 		fs->fs_flags = fs->fs_old_flags;
869 		fs->fs_old_flags |= FS_FLAGS_UPDATED;
870 		fs->fs_sblockloc = sblockloc;
871 	}
872 	/*
873 	 * If not yet done, update UFS1 superblock with new wider fields.
874 	 */
875 	if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_maxbsize != fs->fs_bsize) {
876 		fs->fs_maxbsize = fs->fs_bsize;
877 		fs->fs_time = fs->fs_old_time;
878 		fs->fs_size = fs->fs_old_size;
879 		fs->fs_dsize = fs->fs_old_dsize;
880 		fs->fs_csaddr = fs->fs_old_csaddr;
881 		fs->fs_cstotal.cs_ndir = fs->fs_old_cstotal.cs_ndir;
882 		fs->fs_cstotal.cs_nbfree = fs->fs_old_cstotal.cs_nbfree;
883 		fs->fs_cstotal.cs_nifree = fs->fs_old_cstotal.cs_nifree;
884 		fs->fs_cstotal.cs_nffree = fs->fs_old_cstotal.cs_nffree;
885 	}
886 	if (fs->fs_magic == FS_UFS1_MAGIC &&
887 	    fs->fs_old_inodefmt < FS_44INODEFMT) {
888 		fs->fs_maxfilesize = (u_quad_t) 1LL << 39;
889 		fs->fs_qbmask = ~fs->fs_bmask;
890 		fs->fs_qfmask = ~fs->fs_fmask;
891 	}
892 	if (fs->fs_magic == FS_UFS1_MAGIC) {
893 		ump->um_savedmaxfilesize = fs->fs_maxfilesize;
894 		maxfilesize = (u_int64_t)0x40000000 * fs->fs_bsize - 1;
895 		if (fs->fs_maxfilesize > maxfilesize)
896 			fs->fs_maxfilesize = maxfilesize;
897 	}
898 	/* Compatibility for old filesystems */
899 	if (fs->fs_avgfilesize <= 0)
900 		fs->fs_avgfilesize = AVFILESIZ;
901 	if (fs->fs_avgfpdir <= 0)
902 		fs->fs_avgfpdir = AFPDIR;
903 	if (bigcgs) {
904 		fs->fs_save_cgsize = fs->fs_cgsize;
905 		fs->fs_cgsize = fs->fs_bsize;
906 	}
907 }
908 
909 /*
910  * Unwinding superblock updates for old filesystems.
911  * See ffs_oldfscompat_read above for details.
912  *
913  * XXX - Parts get retired eventually.
914  * Unfortunately new bits get added.
915  */
916 static void
917 ffs_oldfscompat_write(fs, ump)
918 	struct fs *fs;
919 	struct ufsmount *ump;
920 {
921 
922 	/*
923 	 * Copy back UFS2 updated fields that UFS1 inspects.
924 	 */
925 	if (fs->fs_magic == FS_UFS1_MAGIC) {
926 		fs->fs_old_time = fs->fs_time;
927 		fs->fs_old_cstotal.cs_ndir = fs->fs_cstotal.cs_ndir;
928 		fs->fs_old_cstotal.cs_nbfree = fs->fs_cstotal.cs_nbfree;
929 		fs->fs_old_cstotal.cs_nifree = fs->fs_cstotal.cs_nifree;
930 		fs->fs_old_cstotal.cs_nffree = fs->fs_cstotal.cs_nffree;
931 		fs->fs_maxfilesize = ump->um_savedmaxfilesize;
932 	}
933 	if (bigcgs) {
934 		fs->fs_cgsize = fs->fs_save_cgsize;
935 		fs->fs_save_cgsize = 0;
936 	}
937 }
938 
939 /*
940  * unmount system call
941  */
942 int
943 ffs_unmount(mp, mntflags, td)
944 	struct mount *mp;
945 	int mntflags;
946 	struct thread *td;
947 {
948 	struct ufsmount *ump = VFSTOUFS(mp);
949 	struct fs *fs;
950 	int error, flags;
951 
952 	flags = 0;
953 	if (mntflags & MNT_FORCE) {
954 		flags |= FORCECLOSE;
955 	}
956 #ifdef UFS_EXTATTR
957 	if ((error = ufs_extattr_stop(mp, td))) {
958 		if (error != EOPNOTSUPP)
959 			printf("ffs_unmount: ufs_extattr_stop returned %d\n",
960 			    error);
961 	} else {
962 		ufs_extattr_uepm_destroy(&ump->um_extattr);
963 	}
964 #endif
965 	if (mp->mnt_flag & MNT_SOFTDEP) {
966 		if ((error = softdep_flushfiles(mp, flags, td)) != 0)
967 			return (error);
968 	} else {
969 		if ((error = ffs_flushfiles(mp, flags, td)) != 0)
970 			return (error);
971 	}
972 	fs = ump->um_fs;
973 	if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
974 		printf("%s: unmount pending error: blocks %jd files %d\n",
975 		    fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
976 		    fs->fs_pendinginodes);
977 		fs->fs_pendingblocks = 0;
978 		fs->fs_pendinginodes = 0;
979 	}
980 	if (fs->fs_ronly == 0) {
981 		fs->fs_clean = fs->fs_flags & (FS_UNCLEAN|FS_NEEDSFSCK) ? 0 : 1;
982 		error = ffs_sbupdate(ump, MNT_WAIT);
983 		if (error) {
984 			fs->fs_clean = 0;
985 			return (error);
986 		}
987 	}
988 	ump->um_devvp->v_rdev->si_mountpoint = NULL;
989 
990 	vinvalbuf(ump->um_devvp, V_SAVE, NOCRED, td, 0, 0);
991 	/* XXX: see comment above VOP_OPEN */
992 #ifdef notyet
993 	error = VOP_CLOSE(ump->um_devvp, fs->fs_ronly ? FREAD : FREAD|FWRITE,
994 		NOCRED, td);
995 #else
996 	error = VOP_CLOSE(ump->um_devvp, FREAD|FWRITE, NOCRED, td);
997 #endif
998 
999 	vrele(ump->um_devvp);
1000 
1001 	free(fs->fs_csp, M_UFSMNT);
1002 	free(fs, M_UFSMNT);
1003 	free(ump, M_UFSMNT);
1004 	mp->mnt_data = (qaddr_t)0;
1005 	mp->mnt_flag &= ~MNT_LOCAL;
1006 	return (error);
1007 }
1008 
1009 /*
1010  * Flush out all the files in a filesystem.
1011  */
1012 int
1013 ffs_flushfiles(mp, flags, td)
1014 	struct mount *mp;
1015 	int flags;
1016 	struct thread *td;
1017 {
1018 	struct ufsmount *ump;
1019 	int error;
1020 
1021 	ump = VFSTOUFS(mp);
1022 #ifdef QUOTA
1023 	if (mp->mnt_flag & MNT_QUOTA) {
1024 		int i;
1025 		error = vflush(mp, 0, SKIPSYSTEM|flags);
1026 		if (error)
1027 			return (error);
1028 		for (i = 0; i < MAXQUOTAS; i++) {
1029 			if (ump->um_quotas[i] == NULLVP)
1030 				continue;
1031 			quotaoff(td, mp, i);
1032 		}
1033 		/*
1034 		 * Here we fall through to vflush again to ensure
1035 		 * that we have gotten rid of all the system vnodes.
1036 		 */
1037 	}
1038 #endif
1039 	ASSERT_VOP_LOCKED(ump->um_devvp, "ffs_flushfiles");
1040 	if (ump->um_devvp->v_vflag & VV_COPYONWRITE) {
1041 		if ((error = vflush(mp, 0, SKIPSYSTEM | flags)) != 0)
1042 			return (error);
1043 		ffs_snapshot_unmount(mp);
1044 		/*
1045 		 * Here we fall through to vflush again to ensure
1046 		 * that we have gotten rid of all the system vnodes.
1047 		 */
1048 	}
1049         /*
1050 	 * Flush all the files.
1051 	 */
1052 	if ((error = vflush(mp, 0, flags)) != 0)
1053 		return (error);
1054 	/*
1055 	 * Flush filesystem metadata.
1056 	 */
1057 	vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY, td);
1058 	error = VOP_FSYNC(ump->um_devvp, td->td_ucred, MNT_WAIT, td);
1059 	VOP_UNLOCK(ump->um_devvp, 0, td);
1060 	return (error);
1061 }
1062 
1063 /*
1064  * Get filesystem statistics.
1065  */
1066 int
1067 ffs_statfs(mp, sbp, td)
1068 	struct mount *mp;
1069 	struct statfs *sbp;
1070 	struct thread *td;
1071 {
1072 	struct ufsmount *ump;
1073 	struct fs *fs;
1074 
1075 	ump = VFSTOUFS(mp);
1076 	fs = ump->um_fs;
1077 	if (fs->fs_magic != FS_UFS1_MAGIC && fs->fs_magic != FS_UFS2_MAGIC)
1078 		panic("ffs_statfs");
1079 	sbp->f_bsize = fs->fs_fsize;
1080 	sbp->f_iosize = fs->fs_bsize;
1081 	sbp->f_blocks = fs->fs_dsize;
1082 	sbp->f_bfree = fs->fs_cstotal.cs_nbfree * fs->fs_frag +
1083 	    fs->fs_cstotal.cs_nffree + dbtofsb(fs, fs->fs_pendingblocks);
1084 	sbp->f_bavail = freespace(fs, fs->fs_minfree) +
1085 	    dbtofsb(fs, fs->fs_pendingblocks);
1086 	sbp->f_files =  fs->fs_ncg * fs->fs_ipg - ROOTINO;
1087 	sbp->f_ffree = fs->fs_cstotal.cs_nifree + fs->fs_pendinginodes;
1088 	if (sbp != &mp->mnt_stat) {
1089 		sbp->f_type = mp->mnt_vfc->vfc_typenum;
1090 		bcopy((caddr_t)mp->mnt_stat.f_mntonname,
1091 			(caddr_t)&sbp->f_mntonname[0], MNAMELEN);
1092 		bcopy((caddr_t)mp->mnt_stat.f_mntfromname,
1093 			(caddr_t)&sbp->f_mntfromname[0], MNAMELEN);
1094 	}
1095 	return (0);
1096 }
1097 
1098 /*
1099  * Go through the disk queues to initiate sandbagged IO;
1100  * go through the inodes to write those that have been modified;
1101  * initiate the writing of the super block if it has been modified.
1102  *
1103  * Note: we are always called with the filesystem marked `MPBUSY'.
1104  */
1105 int
1106 ffs_sync(mp, waitfor, cred, td)
1107 	struct mount *mp;
1108 	int waitfor;
1109 	struct ucred *cred;
1110 	struct thread *td;
1111 {
1112 	struct vnode *nvp, *vp, *devvp;
1113 	struct inode *ip;
1114 	struct ufsmount *ump = VFSTOUFS(mp);
1115 	struct fs *fs;
1116 	int error, count, wait, lockreq, allerror = 0;
1117 
1118 	fs = ump->um_fs;
1119 	if (fs->fs_fmod != 0 && fs->fs_ronly != 0) {		/* XXX */
1120 		printf("fs = %s\n", fs->fs_fsmnt);
1121 		panic("ffs_sync: rofs mod");
1122 	}
1123 	/*
1124 	 * Write back each (modified) inode.
1125 	 */
1126 	wait = 0;
1127 	lockreq = LK_EXCLUSIVE | LK_NOWAIT;
1128 	if (waitfor == MNT_WAIT) {
1129 		wait = 1;
1130 		lockreq = LK_EXCLUSIVE;
1131 	}
1132 	mtx_lock(&mntvnode_mtx);
1133 loop:
1134 	for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); vp != NULL; vp = nvp) {
1135 		/*
1136 		 * If the vnode that we are about to sync is no longer
1137 		 * associated with this mount point, start over.
1138 		 */
1139 		if (vp->v_mount != mp)
1140 			goto loop;
1141 
1142 		/*
1143 		 * Depend on the mntvnode_slock to keep things stable enough
1144 		 * for a quick test.  Since there might be hundreds of
1145 		 * thousands of vnodes, we cannot afford even a subroutine
1146 		 * call unless there's a good chance that we have work to do.
1147 		 */
1148 		nvp = TAILQ_NEXT(vp, v_nmntvnodes);
1149 		ip = VTOI(vp);
1150 		if (vp->v_type == VNON || ((ip->i_flag &
1151 		    (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) == 0 &&
1152 		    TAILQ_EMPTY(&vp->v_dirtyblkhd))) {
1153 			continue;
1154 		}
1155 		if (vp->v_type != VCHR) {
1156 			mtx_unlock(&mntvnode_mtx);
1157 			if ((error = vget(vp, lockreq, td)) != 0) {
1158 				mtx_lock(&mntvnode_mtx);
1159 				if (error == ENOENT)
1160 					goto loop;
1161 			} else {
1162 				if ((error = VOP_FSYNC(vp, cred, waitfor, td)) != 0)
1163 					allerror = error;
1164 				VOP_UNLOCK(vp, 0, td);
1165 				vrele(vp);
1166 				mtx_lock(&mntvnode_mtx);
1167 			}
1168 		} else {
1169 			mtx_unlock(&mntvnode_mtx);
1170 			UFS_UPDATE(vp, wait);
1171 			mtx_lock(&mntvnode_mtx);
1172 		}
1173 		if (TAILQ_NEXT(vp, v_nmntvnodes) != nvp)
1174 			goto loop;
1175 	}
1176 	mtx_unlock(&mntvnode_mtx);
1177 	/*
1178 	 * Force stale filesystem control information to be flushed.
1179 	 */
1180 	if (waitfor == MNT_WAIT) {
1181 		if ((error = softdep_flushworklist(ump->um_mountp, &count, td)))
1182 			allerror = error;
1183 		/* Flushed work items may create new vnodes to clean */
1184 		if (allerror == 0 && count) {
1185 			mtx_lock(&mntvnode_mtx);
1186 			goto loop;
1187 		}
1188 	}
1189 #ifdef QUOTA
1190 	qsync(mp);
1191 #endif
1192 	devvp = ump->um_devvp;
1193 	VI_LOCK(devvp);
1194 	if (waitfor != MNT_LAZY &&
1195 	    (devvp->v_numoutput > 0 || TAILQ_FIRST(&devvp->v_dirtyblkhd))) {
1196 		vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK, td);
1197 		if ((error = VOP_FSYNC(devvp, cred, waitfor, td)) != 0)
1198 			allerror = error;
1199 		VOP_UNLOCK(devvp, 0, td);
1200 		if (allerror == 0 && waitfor == MNT_WAIT) {
1201 			mtx_lock(&mntvnode_mtx);
1202 			goto loop;
1203 		}
1204 	} else
1205 		VI_UNLOCK(devvp);
1206 	/*
1207 	 * Write back modified superblock.
1208 	 */
1209 	if (fs->fs_fmod != 0 && (error = ffs_sbupdate(ump, waitfor)) != 0)
1210 		allerror = error;
1211 	return (allerror);
1212 }
1213 
1214 int
1215 ffs_vget(mp, ino, flags, vpp)
1216 	struct mount *mp;
1217 	ino_t ino;
1218 	int flags;
1219 	struct vnode **vpp;
1220 {
1221 	struct thread *td = curthread; 		/* XXX */
1222 	struct fs *fs;
1223 	struct inode *ip;
1224 	struct ufsmount *ump;
1225 	struct buf *bp;
1226 	struct vnode *vp;
1227 	dev_t dev;
1228 	int error;
1229 
1230 	ump = VFSTOUFS(mp);
1231 	dev = ump->um_dev;
1232 
1233 	/*
1234 	 * We do not lock vnode creation as it is believed to be too
1235 	 * expensive for such rare case as simultaneous creation of vnode
1236 	 * for same ino by different processes. We just allow them to race
1237 	 * and check later to decide who wins. Let the race begin!
1238 	 */
1239 	if ((error = ufs_ihashget(dev, ino, flags, vpp)) != 0)
1240 		return (error);
1241 	if (*vpp != NULL)
1242 		return (0);
1243 
1244 	/*
1245 	 * If this MALLOC() is performed after the getnewvnode()
1246 	 * it might block, leaving a vnode with a NULL v_data to be
1247 	 * found by ffs_sync() if a sync happens to fire right then,
1248 	 * which will cause a panic because ffs_sync() blindly
1249 	 * dereferences vp->v_data (as well it should).
1250 	 */
1251 	ip = uma_zalloc(uma_inode, M_WAITOK);
1252 
1253 	/* Allocate a new vnode/inode. */
1254 	error = getnewvnode("ufs", mp, ffs_vnodeop_p, &vp);
1255 	if (error) {
1256 		*vpp = NULL;
1257 		uma_zfree(uma_inode, ip);
1258 		return (error);
1259 	}
1260 	bzero((caddr_t)ip, sizeof(struct inode));
1261 	/*
1262 	 * FFS supports recursive locking.
1263 	 */
1264 	vp->v_vnlock->lk_flags |= LK_CANRECURSE;
1265 	vp->v_data = ip;
1266 	ip->i_vnode = vp;
1267 	ip->i_ump = ump;
1268 	ip->i_fs = fs = ump->um_fs;
1269 	ip->i_dev = dev;
1270 	ip->i_number = ino;
1271 #ifdef QUOTA
1272 	{
1273 		int i;
1274 		for (i = 0; i < MAXQUOTAS; i++)
1275 			ip->i_dquot[i] = NODQUOT;
1276 	}
1277 #endif
1278 	/*
1279 	 * Exclusively lock the vnode before adding to hash. Note, that we
1280 	 * must not release nor downgrade the lock (despite flags argument
1281 	 * says) till it is fully initialized.
1282 	 */
1283 	lockmgr(vp->v_vnlock, LK_EXCLUSIVE, (struct mtx *)0, td);
1284 
1285 	/*
1286 	 * Atomicaly (in terms of ufs_hash operations) check the hash for
1287 	 * duplicate of vnode being created and add it to the hash. If a
1288 	 * duplicate vnode was found, it will be vget()ed from hash for us.
1289 	 */
1290 	if ((error = ufs_ihashins(ip, flags, vpp)) != 0) {
1291 		vput(vp);
1292 		*vpp = NULL;
1293 		return (error);
1294 	}
1295 
1296 	/* We lost the race, then throw away our vnode and return existing */
1297 	if (*vpp != NULL) {
1298 		vput(vp);
1299 		return (0);
1300 	}
1301 
1302 	/* Read in the disk contents for the inode, copy into the inode. */
1303 	error = bread(ump->um_devvp, fsbtodb(fs, ino_to_fsba(fs, ino)),
1304 	    (int)fs->fs_bsize, NOCRED, &bp);
1305 	if (error) {
1306 		/*
1307 		 * The inode does not contain anything useful, so it would
1308 		 * be misleading to leave it on its hash chain. With mode
1309 		 * still zero, it will be unlinked and returned to the free
1310 		 * list by vput().
1311 		 */
1312 		brelse(bp);
1313 		vput(vp);
1314 		*vpp = NULL;
1315 		return (error);
1316 	}
1317 	if (ip->i_ump->um_fstype == UFS1)
1318 		ip->i_din1 = uma_zalloc(uma_ufs1, M_WAITOK);
1319 	else
1320 		ip->i_din2 = uma_zalloc(uma_ufs2, M_WAITOK);
1321 	ffs_load_inode(bp, ip, fs, ino);
1322 	if (DOINGSOFTDEP(vp))
1323 		softdep_load_inodeblock(ip);
1324 	else
1325 		ip->i_effnlink = ip->i_nlink;
1326 	bqrelse(bp);
1327 
1328 	/*
1329 	 * Initialize the vnode from the inode, check for aliases.
1330 	 * Note that the underlying vnode may have changed.
1331 	 */
1332 	error = ufs_vinit(mp, ffs_specop_p, ffs_fifoop_p, &vp);
1333 	if (error) {
1334 		vput(vp);
1335 		*vpp = NULL;
1336 		return (error);
1337 	}
1338 	/*
1339 	 * Finish inode initialization now that aliasing has been resolved.
1340 	 */
1341 	ip->i_devvp = ump->um_devvp;
1342 	VREF(ip->i_devvp);
1343 	/*
1344 	 * Set up a generation number for this inode if it does not
1345 	 * already have one. This should only happen on old filesystems.
1346 	 */
1347 	if (ip->i_gen == 0) {
1348 		ip->i_gen = arc4random() / 2 + 1;
1349 		if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
1350 			ip->i_flag |= IN_MODIFIED;
1351 			DIP(ip, i_gen) = ip->i_gen;
1352 		}
1353 	}
1354 	/*
1355 	 * Ensure that uid and gid are correct. This is a temporary
1356 	 * fix until fsck has been changed to do the update.
1357 	 */
1358 	if (fs->fs_magic == FS_UFS1_MAGIC &&		/* XXX */
1359 	    fs->fs_old_inodefmt < FS_44INODEFMT) {	/* XXX */
1360 		ip->i_uid = ip->i_din1->di_ouid;	/* XXX */
1361 		ip->i_gid = ip->i_din1->di_ogid;	/* XXX */
1362 	}						/* XXX */
1363 
1364 #ifdef MAC
1365 	if ((mp->mnt_flag & MNT_MULTILABEL) && ip->i_mode) {
1366 		/*
1367 		 * If this vnode is already allocated, and we're running
1368 		 * multi-label, attempt to perform a label association
1369 		 * from the extended attributes on the inode.
1370 		 */
1371 		error = mac_associate_vnode_extattr(mp, vp);
1372 		if (error) {
1373 			/* ufs_inactive will release ip->i_devvp ref. */
1374 			vput(vp);
1375 			*vpp = NULL;
1376 			return (error);
1377 		}
1378 	}
1379 #endif
1380 
1381 	*vpp = vp;
1382 	return (0);
1383 }
1384 
1385 /*
1386  * File handle to vnode
1387  *
1388  * Have to be really careful about stale file handles:
1389  * - check that the inode number is valid
1390  * - call ffs_vget() to get the locked inode
1391  * - check for an unallocated inode (i_mode == 0)
1392  * - check that the given client host has export rights and return
1393  *   those rights via. exflagsp and credanonp
1394  */
1395 int
1396 ffs_fhtovp(mp, fhp, vpp)
1397 	struct mount *mp;
1398 	struct fid *fhp;
1399 	struct vnode **vpp;
1400 {
1401 	struct ufid *ufhp;
1402 	struct fs *fs;
1403 
1404 	ufhp = (struct ufid *)fhp;
1405 	fs = VFSTOUFS(mp)->um_fs;
1406 	if (ufhp->ufid_ino < ROOTINO ||
1407 	    ufhp->ufid_ino >= fs->fs_ncg * fs->fs_ipg)
1408 		return (ESTALE);
1409 	return (ufs_fhtovp(mp, ufhp, vpp));
1410 }
1411 
1412 /*
1413  * Vnode pointer to File handle
1414  */
1415 /* ARGSUSED */
1416 int
1417 ffs_vptofh(vp, fhp)
1418 	struct vnode *vp;
1419 	struct fid *fhp;
1420 {
1421 	struct inode *ip;
1422 	struct ufid *ufhp;
1423 
1424 	ip = VTOI(vp);
1425 	ufhp = (struct ufid *)fhp;
1426 	ufhp->ufid_len = sizeof(struct ufid);
1427 	ufhp->ufid_ino = ip->i_number;
1428 	ufhp->ufid_gen = ip->i_gen;
1429 	return (0);
1430 }
1431 
1432 /*
1433  * Initialize the filesystem.
1434  */
1435 static int
1436 ffs_init(vfsp)
1437 	struct vfsconf *vfsp;
1438 {
1439 
1440 	softdep_initialize();
1441 	return (ufs_init(vfsp));
1442 }
1443 
1444 /*
1445  * Undo the work of ffs_init().
1446  */
1447 static int
1448 ffs_uninit(vfsp)
1449 	struct vfsconf *vfsp;
1450 {
1451 	int ret;
1452 
1453 	ret = ufs_uninit(vfsp);
1454 	softdep_uninitialize();
1455 	return (ret);
1456 }
1457 
1458 /*
1459  * Write a superblock and associated information back to disk.
1460  */
1461 static int
1462 ffs_sbupdate(mp, waitfor)
1463 	struct ufsmount *mp;
1464 	int waitfor;
1465 {
1466 	struct fs *fs = mp->um_fs;
1467 	struct buf *bp;
1468 	int blks;
1469 	void *space;
1470 	int i, size, error, allerror = 0;
1471 
1472 	if (fs->fs_ronly == 1 &&
1473 	    (mp->um_mountp->mnt_flag & (MNT_RDONLY | MNT_UPDATE)) !=
1474 	    (MNT_RDONLY | MNT_UPDATE))
1475 		panic("ffs_sbupdate: write read-only filesystem");
1476 	/*
1477 	 * First write back the summary information.
1478 	 */
1479 	blks = howmany(fs->fs_cssize, fs->fs_fsize);
1480 	space = fs->fs_csp;
1481 	for (i = 0; i < blks; i += fs->fs_frag) {
1482 		size = fs->fs_bsize;
1483 		if (i + fs->fs_frag > blks)
1484 			size = (blks - i) * fs->fs_fsize;
1485 		bp = getblk(mp->um_devvp, fsbtodb(fs, fs->fs_csaddr + i),
1486 		    size, 0, 0);
1487 		bcopy(space, bp->b_data, (u_int)size);
1488 		space = (char *)space + size;
1489 		if (waitfor != MNT_WAIT)
1490 			bawrite(bp);
1491 		else if ((error = bwrite(bp)) != 0)
1492 			allerror = error;
1493 	}
1494 	/*
1495 	 * Now write back the superblock itself. If any errors occurred
1496 	 * up to this point, then fail so that the superblock avoids
1497 	 * being written out as clean.
1498 	 */
1499 	if (allerror)
1500 		return (allerror);
1501 	if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_sblockloc != SBLOCK_UFS1 &&
1502 	    (fs->fs_flags & FS_FLAGS_UPDATED) == 0) {
1503 		printf("%s: correcting fs_sblockloc from %jd to %d\n",
1504 		    fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS1);
1505 		fs->fs_sblockloc = SBLOCK_UFS1;
1506 	}
1507 	if (fs->fs_magic == FS_UFS2_MAGIC && fs->fs_sblockloc != SBLOCK_UFS2 &&
1508 	    (fs->fs_flags & FS_FLAGS_UPDATED) == 0) {
1509 		printf("%s: correcting fs_sblockloc from %jd to %d\n",
1510 		    fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS2);
1511 		fs->fs_sblockloc = SBLOCK_UFS2;
1512 	}
1513 	bp = getblk(mp->um_devvp, btodb(fs->fs_sblockloc), (int)fs->fs_sbsize,
1514 	    0, 0);
1515 	fs->fs_fmod = 0;
1516 	fs->fs_time = time_second;
1517 	bcopy((caddr_t)fs, bp->b_data, (u_int)fs->fs_sbsize);
1518 	ffs_oldfscompat_write((struct fs *)bp->b_data, mp);
1519 	if (waitfor != MNT_WAIT)
1520 		bawrite(bp);
1521 	else if ((error = bwrite(bp)) != 0)
1522 		allerror = error;
1523 	return (allerror);
1524 }
1525 
1526 static int
1527 ffs_extattrctl(struct mount *mp, int cmd, struct vnode *filename_vp,
1528 	int attrnamespace, const char *attrname, struct thread *td)
1529 {
1530 
1531 #ifdef UFS_EXTATTR
1532 	return (ufs_extattrctl(mp, cmd, filename_vp, attrnamespace,
1533 	    attrname, td));
1534 #else
1535 	return (vfs_stdextattrctl(mp, cmd, filename_vp, attrnamespace,
1536 	    attrname, td));
1537 #endif
1538 }
1539 
1540 static void
1541 ffs_ifree(struct ufsmount *ump, struct inode *ip)
1542 {
1543 
1544 	if (ump->um_fstype == UFS1)
1545 		uma_zfree(uma_ufs1, ip->i_din1);
1546 	else
1547 		uma_zfree(uma_ufs2, ip->i_din1);
1548 	uma_zfree(uma_inode, ip);
1549 }
1550