xref: /original-bsd/sys/ufs/ffs/ffs_vfsops.c (revision b3c06cab)
1 /*
2  * Copyright (c) 1989, 1991, 1993, 1994
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * %sccs.include.redist.c%
6  *
7  *	@(#)ffs_vfsops.c	8.24 (Berkeley) 05/09/95
8  */
9 
10 #include <sys/param.h>
11 #include <sys/systm.h>
12 #include <sys/namei.h>
13 #include <sys/proc.h>
14 #include <sys/kernel.h>
15 #include <sys/vnode.h>
16 #include <sys/socket.h>
17 #include <sys/mount.h>
18 #include <sys/buf.h>
19 #include <sys/mbuf.h>
20 #include <sys/file.h>
21 #include <sys/disklabel.h>
22 #include <sys/ioctl.h>
23 #include <sys/errno.h>
24 #include <sys/malloc.h>
25 
26 #include <miscfs/specfs/specdev.h>
27 
28 #include <ufs/ufs/quota.h>
29 #include <ufs/ufs/ufsmount.h>
30 #include <ufs/ufs/inode.h>
31 #include <ufs/ufs/ufs_extern.h>
32 
33 #include <ufs/ffs/fs.h>
34 #include <ufs/ffs/ffs_extern.h>
35 
36 int ffs_sbupdate __P((struct ufsmount *, int));
37 
38 struct vfsops ufs_vfsops = {
39 	ffs_mount,
40 	ufs_start,
41 	ffs_unmount,
42 	ufs_root,
43 	ufs_quotactl,
44 	ffs_statfs,
45 	ffs_sync,
46 	ffs_vget,
47 	ffs_fhtovp,
48 	ffs_vptofh,
49 	ffs_init,
50 	ffs_sysctl,
51 };
52 
53 extern u_long nextgennumber;
54 
55 /*
56  * Called by main() when ufs is going to be mounted as root.
57  *
58  * Name is updated by mount(8) after booting.
59  */
60 #define ROOTNAME	"root_device"
61 
62 ffs_mountroot()
63 {
64 	extern struct vnode *rootvp;
65 	register struct fs *fs;
66 	register struct mount *mp;
67 	struct proc *p = curproc;	/* XXX */
68 	struct ufsmount *ump;
69 	struct vfsconf *vfsp;
70 	u_int size;
71 	int error;
72 
73 	/*
74 	 * Get vnodes for swapdev and rootdev.
75 	 */
76 	if (bdevvp(swapdev, &swapdev_vp) || bdevvp(rootdev, &rootvp))
77 		panic("ffs_mountroot: can't setup bdevvp's");
78 
79 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
80 		if (!strcmp(vfsp->vfc_name, "ufs"))
81 			break;
82 	if (vfsp == NULL)
83 		return (ENODEV);
84 	mp = malloc((u_long)sizeof(struct mount), M_MOUNT, M_WAITOK);
85 	bzero((char *)mp, (u_long)sizeof(struct mount));
86 	mp->mnt_vfc = vfsp;
87 	mp->mnt_op = vfsp->vfc_vfsops;
88 	mp->mnt_flag = MNT_RDONLY;
89 	if (error = ffs_mountfs(rootvp, mp, p)) {
90 		free(mp, M_MOUNT);
91 		return (error);
92 	}
93 	if (error = vfs_lock(mp)) {
94 		(void)ffs_unmount(mp, 0, p);
95 		free(mp, M_MOUNT);
96 		return (error);
97 	}
98 	CIRCLEQ_INSERT_TAIL(&mountlist, mp, mnt_list);
99 	mp->mnt_vnodecovered = NULLVP;
100 	vfsp->vfc_refcount++;
101 	mp->mnt_stat.f_type = vfsp->vfc_typenum;
102 	mp->mnt_flag |= (vfsp->vfc_flags & MNT_VISFLAGMASK) | MNT_ROOTFS;
103 	strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
104 	ump = VFSTOUFS(mp);
105 	fs = ump->um_fs;
106 	bzero(fs->fs_fsmnt, sizeof(fs->fs_fsmnt));
107 	fs->fs_fsmnt[0] = '/';
108 	bcopy((caddr_t)fs->fs_fsmnt, (caddr_t)mp->mnt_stat.f_mntonname,
109 	    MNAMELEN);
110 	(void) copystr(ROOTNAME, mp->mnt_stat.f_mntfromname, MNAMELEN - 1,
111 	    &size);
112 	bzero(mp->mnt_stat.f_mntfromname + size, MNAMELEN - size);
113 	(void)ffs_statfs(mp, &mp->mnt_stat, p);
114 	vfs_unlock(mp);
115 	inittodr(fs->fs_time);
116 	return (0);
117 }
118 
119 /*
120  * VFS Operations.
121  *
122  * mount system call
123  */
124 int
125 ffs_mount(mp, path, data, ndp, p)
126 	register struct mount *mp;
127 	char *path;
128 	caddr_t data;
129 	struct nameidata *ndp;
130 	struct proc *p;
131 {
132 	struct vnode *devvp;
133 	struct ufs_args args;
134 	struct ufsmount *ump;
135 	register struct fs *fs;
136 	u_int size;
137 	int error, flags;
138 	mode_t accessmode;
139 
140 	if (error = copyin(data, (caddr_t)&args, sizeof (struct ufs_args)))
141 		return (error);
142 	/*
143 	 * If updating, check whether changing from read-only to
144 	 * read/write; if there is no device name, that's all we do.
145 	 */
146 	if (mp->mnt_flag & MNT_UPDATE) {
147 		ump = VFSTOUFS(mp);
148 		fs = ump->um_fs;
149 		if (fs->fs_ronly == 0 && (mp->mnt_flag & MNT_RDONLY)) {
150 			flags = WRITECLOSE;
151 			if (mp->mnt_flag & MNT_FORCE)
152 				flags |= FORCECLOSE;
153 			if (vfs_busy(mp))
154 				return (EBUSY);
155 			if (error = ffs_flushfiles(mp, flags, p)) {
156 				vfs_unbusy(mp);
157 				return (error);
158 			}
159 			fs->fs_clean = 1;
160 			fs->fs_ronly = 1;
161 			if (error = ffs_sbupdate(ump, MNT_WAIT)) {
162 				fs->fs_clean = 0;
163 				fs->fs_ronly = 0;
164 				vfs_unbusy(mp);
165 				return (error);
166 			}
167 			vfs_unbusy(mp);
168 		}
169 		if ((mp->mnt_flag & MNT_RELOAD) &&
170 		    (error = ffs_reload(mp, ndp->ni_cnd.cn_cred, p)))
171 			return (error);
172 		if (fs->fs_ronly && (mp->mnt_flag & MNT_WANTRDWR)) {
173 			/*
174 			 * If upgrade to read-write by non-root, then verify
175 			 * that user has necessary permissions on the device.
176 			 */
177 			if (p->p_ucred->cr_uid != 0) {
178 				devvp = ump->um_devvp;
179 				VOP_LOCK(devvp);
180 				if (error = VOP_ACCESS(devvp, VREAD | VWRITE,
181 				    p->p_ucred, p)) {
182 					VOP_UNLOCK(devvp);
183 					return (error);
184 				}
185 				VOP_UNLOCK(devvp);
186 			}
187 			fs->fs_ronly = 0;
188 			fs->fs_clean = 0;
189 			(void) ffs_sbupdate(ump, MNT_WAIT);
190 		}
191 		if (args.fspec == 0) {
192 			/*
193 			 * Process export requests.
194 			 */
195 			return (vfs_export(mp, &ump->um_export, &args.export));
196 		}
197 	}
198 	/*
199 	 * Not an update, or updating the name: look up the name
200 	 * and verify that it refers to a sensible block device.
201 	 */
202 	NDINIT(ndp, LOOKUP, FOLLOW, UIO_USERSPACE, args.fspec, p);
203 	if (error = namei(ndp))
204 		return (error);
205 	devvp = ndp->ni_vp;
206 
207 	if (devvp->v_type != VBLK) {
208 		vrele(devvp);
209 		return (ENOTBLK);
210 	}
211 	if (major(devvp->v_rdev) >= nblkdev) {
212 		vrele(devvp);
213 		return (ENXIO);
214 	}
215 	/*
216 	 * If mount by non-root, then verify that user has necessary
217 	 * permissions on the device.
218 	 */
219 	if (p->p_ucred->cr_uid != 0) {
220 		accessmode = VREAD;
221 		if ((mp->mnt_flag & MNT_RDONLY) == 0)
222 			accessmode |= VWRITE;
223 		VOP_LOCK(devvp);
224 		if (error = VOP_ACCESS(devvp, accessmode, p->p_ucred, p)) {
225 			vput(devvp);
226 			return (error);
227 		}
228 		VOP_UNLOCK(devvp);
229 	}
230 	if ((mp->mnt_flag & MNT_UPDATE) == 0)
231 		error = ffs_mountfs(devvp, mp, p);
232 	else {
233 		if (devvp != ump->um_devvp)
234 			error = EINVAL;	/* needs translation */
235 		else
236 			vrele(devvp);
237 	}
238 	if (error) {
239 		vrele(devvp);
240 		return (error);
241 	}
242 	ump = VFSTOUFS(mp);
243 	fs = ump->um_fs;
244 	(void) copyinstr(path, fs->fs_fsmnt, sizeof(fs->fs_fsmnt) - 1, &size);
245 	bzero(fs->fs_fsmnt + size, sizeof(fs->fs_fsmnt) - size);
246 	bcopy((caddr_t)fs->fs_fsmnt, (caddr_t)mp->mnt_stat.f_mntonname,
247 	    MNAMELEN);
248 	(void) copyinstr(args.fspec, mp->mnt_stat.f_mntfromname, MNAMELEN - 1,
249 	    &size);
250 	bzero(mp->mnt_stat.f_mntfromname + size, MNAMELEN - size);
251 	(void)ffs_statfs(mp, &mp->mnt_stat, p);
252 	return (0);
253 }
254 
255 /*
256  * Reload all incore data for a filesystem (used after running fsck on
257  * the root filesystem and finding things to fix). The filesystem must
258  * be mounted read-only.
259  *
260  * Things to do to update the mount:
261  *	1) invalidate all cached meta-data.
262  *	2) re-read superblock from disk.
263  *	3) re-read summary information from disk.
264  *	4) invalidate all inactive vnodes.
265  *	5) invalidate all cached file data.
266  *	6) re-read inode data for all active vnodes.
267  */
268 ffs_reload(mountp, cred, p)
269 	register struct mount *mountp;
270 	struct ucred *cred;
271 	struct proc *p;
272 {
273 	register struct vnode *vp, *nvp, *devvp;
274 	struct inode *ip;
275 	struct csum *space;
276 	struct buf *bp;
277 	struct fs *fs, *newfs;
278 	struct partinfo dpart;
279 	int i, blks, size, error;
280 	int32_t *lp;
281 
282 	if ((mountp->mnt_flag & MNT_RDONLY) == 0)
283 		return (EINVAL);
284 	/*
285 	 * Step 1: invalidate all cached meta-data.
286 	 */
287 	devvp = VFSTOUFS(mountp)->um_devvp;
288 	if (vinvalbuf(devvp, 0, cred, p, 0, 0))
289 		panic("ffs_reload: dirty1");
290 	/*
291 	 * Step 2: re-read superblock from disk.
292 	 */
293 	if (VOP_IOCTL(devvp, DIOCGPART, (caddr_t)&dpart, FREAD, NOCRED, p) != 0)
294 		size = DEV_BSIZE;
295 	else
296 		size = dpart.disklab->d_secsize;
297 	if (error = bread(devvp, (ufs_daddr_t)(SBOFF/size), SBSIZE, NOCRED,&bp))
298 		return (error);
299 	newfs = (struct fs *)bp->b_data;
300 	if (newfs->fs_magic != FS_MAGIC || newfs->fs_bsize > MAXBSIZE ||
301 	    newfs->fs_bsize < sizeof(struct fs)) {
302 		brelse(bp);
303 		return (EIO);		/* XXX needs translation */
304 	}
305 	fs = VFSTOUFS(mountp)->um_fs;
306 	/*
307 	 * Copy pointer fields back into superblock before copying in	XXX
308 	 * new superblock. These should really be in the ufsmount.	XXX
309 	 * Note that important parameters (eg fs_ncg) are unchanged.
310 	 */
311 	bcopy(&fs->fs_csp[0], &newfs->fs_csp[0], sizeof(fs->fs_csp));
312 	newfs->fs_maxcluster = fs->fs_maxcluster;
313 	bcopy(newfs, fs, (u_int)fs->fs_sbsize);
314 	if (fs->fs_sbsize < SBSIZE)
315 		bp->b_flags |= B_INVAL;
316 	brelse(bp);
317 	mountp->mnt_maxsymlinklen = fs->fs_maxsymlinklen;
318 	ffs_oldfscompat(fs);
319 	/*
320 	 * Step 3: re-read summary information from disk.
321 	 */
322 	blks = howmany(fs->fs_cssize, fs->fs_fsize);
323 	space = fs->fs_csp[0];
324 	for (i = 0; i < blks; i += fs->fs_frag) {
325 		size = fs->fs_bsize;
326 		if (i + fs->fs_frag > blks)
327 			size = (blks - i) * fs->fs_fsize;
328 		if (error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), size,
329 		    NOCRED, &bp))
330 			return (error);
331 		bcopy(bp->b_data, fs->fs_csp[fragstoblks(fs, i)], (u_int)size);
332 		brelse(bp);
333 	}
334 	/*
335 	 * We no longer know anything about clusters per cylinder group.
336 	 */
337 	if (fs->fs_contigsumsize > 0) {
338 		lp = fs->fs_maxcluster;
339 		for (i = 0; i < fs->fs_ncg; i++)
340 			*lp++ = fs->fs_contigsumsize;
341 	}
342 loop:
343 	for (vp = mountp->mnt_vnodelist.lh_first; vp != NULL; vp = nvp) {
344 		nvp = vp->v_mntvnodes.le_next;
345 		/*
346 		 * Step 4: invalidate all inactive vnodes.
347 		 */
348 		if (vp->v_usecount == 0) {
349 			vgone(vp);
350 			continue;
351 		}
352 		/*
353 		 * Step 5: invalidate all cached file data.
354 		 */
355 		if (vget(vp, 1))
356 			goto loop;
357 		if (vinvalbuf(vp, 0, cred, p, 0, 0))
358 			panic("ffs_reload: dirty2");
359 		/*
360 		 * Step 6: re-read inode data for all active vnodes.
361 		 */
362 		ip = VTOI(vp);
363 		if (error =
364 		    bread(devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
365 		    (int)fs->fs_bsize, NOCRED, &bp)) {
366 			vput(vp);
367 			return (error);
368 		}
369 		ip->i_din = *((struct dinode *)bp->b_data +
370 		    ino_to_fsbo(fs, ip->i_number));
371 		brelse(bp);
372 		vput(vp);
373 		if (vp->v_mount != mountp)
374 			goto loop;
375 	}
376 	return (0);
377 }
378 
379 /*
380  * Common code for mount and mountroot
381  */
382 int
383 ffs_mountfs(devvp, mp, p)
384 	register struct vnode *devvp;
385 	struct mount *mp;
386 	struct proc *p;
387 {
388 	register struct ufsmount *ump;
389 	struct buf *bp;
390 	register struct fs *fs;
391 	dev_t dev;
392 	struct partinfo dpart;
393 	caddr_t base, space;
394 	int error, i, blks, size, ronly;
395 	int32_t *lp;
396 	struct ucred *cred;
397 	extern struct vnode *rootvp;
398 	u_int64_t maxfilesize;					/* XXX */
399 
400 	dev = devvp->v_rdev;
401 	cred = p ? p->p_ucred : NOCRED;
402 	/*
403 	 * Disallow multiple mounts of the same device.
404 	 * Disallow mounting of a device that is currently in use
405 	 * (except for root, which might share swap device for miniroot).
406 	 * Flush out any old buffers remaining from a previous use.
407 	 */
408 	if (error = vfs_mountedon(devvp))
409 		return (error);
410 	if (vcount(devvp) > 1 && devvp != rootvp)
411 		return (EBUSY);
412 	if (error = vinvalbuf(devvp, V_SAVE, cred, p, 0, 0))
413 		return (error);
414 
415 	ronly = (mp->mnt_flag & MNT_RDONLY) != 0;
416 	if (error = VOP_OPEN(devvp, ronly ? FREAD : FREAD|FWRITE, FSCRED, p))
417 		return (error);
418 	if (VOP_IOCTL(devvp, DIOCGPART, (caddr_t)&dpart, FREAD, cred, p) != 0)
419 		size = DEV_BSIZE;
420 	else
421 		size = dpart.disklab->d_secsize;
422 
423 	bp = NULL;
424 	ump = NULL;
425 	if (error = bread(devvp, (ufs_daddr_t)(SBOFF/size), SBSIZE, cred, &bp))
426 		goto out;
427 	fs = (struct fs *)bp->b_data;
428 	if (fs->fs_magic != FS_MAGIC || fs->fs_bsize > MAXBSIZE ||
429 	    fs->fs_bsize < sizeof(struct fs)) {
430 		error = EINVAL;		/* XXX needs translation */
431 		goto out;
432 	}
433 	/* XXX updating 4.2 FFS superblocks trashes rotational layout tables */
434 	if (fs->fs_postblformat == FS_42POSTBLFMT && !ronly) {
435 		error = EROFS;          /* needs translation */
436 		goto out;
437 	}
438 	ump = malloc(sizeof *ump, M_UFSMNT, M_WAITOK);
439 	bzero((caddr_t)ump, sizeof *ump);
440 	ump->um_fs = malloc((u_long)fs->fs_sbsize, M_UFSMNT,
441 	    M_WAITOK);
442 	bcopy(bp->b_data, ump->um_fs, (u_int)fs->fs_sbsize);
443 	if (fs->fs_sbsize < SBSIZE)
444 		bp->b_flags |= B_INVAL;
445 	brelse(bp);
446 	bp = NULL;
447 	fs = ump->um_fs;
448 	fs->fs_ronly = ronly;
449 	size = fs->fs_cssize;
450 	blks = howmany(size, fs->fs_fsize);
451 	if (fs->fs_contigsumsize > 0)
452 		size += fs->fs_ncg * sizeof(int32_t);
453 	base = space = malloc((u_long)size, M_UFSMNT, M_WAITOK);
454 	for (i = 0; i < blks; i += fs->fs_frag) {
455 		size = fs->fs_bsize;
456 		if (i + fs->fs_frag > blks)
457 			size = (blks - i) * fs->fs_fsize;
458 		if (error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), size,
459 		    cred, &bp)) {
460 			free(base, M_UFSMNT);
461 			goto out;
462 		}
463 		bcopy(bp->b_data, space, (u_int)size);
464 		fs->fs_csp[fragstoblks(fs, i)] = (struct csum *)space;
465 		space += size;
466 		brelse(bp);
467 		bp = NULL;
468 	}
469 	if (fs->fs_contigsumsize > 0) {
470 		fs->fs_maxcluster = lp = (int32_t *)space;
471 		for (i = 0; i < fs->fs_ncg; i++)
472 			*lp++ = fs->fs_contigsumsize;
473 	}
474 	mp->mnt_data = (qaddr_t)ump;
475 	mp->mnt_stat.f_fsid.val[0] = (long)dev;
476 	mp->mnt_stat.f_fsid.val[1] = mp->mnt_vfc->vfc_typenum;
477 	mp->mnt_maxsymlinklen = fs->fs_maxsymlinklen;
478 	ump->um_mountp = mp;
479 	ump->um_dev = dev;
480 	ump->um_devvp = devvp;
481 	ump->um_nindir = fs->fs_nindir;
482 	ump->um_bptrtodb = fs->fs_fsbtodb;
483 	ump->um_seqinc = fs->fs_frag;
484 	for (i = 0; i < MAXQUOTAS; i++)
485 		ump->um_quotas[i] = NULLVP;
486 	devvp->v_specflags |= SI_MOUNTEDON;
487 	ffs_oldfscompat(fs);
488 	ump->um_savedmaxfilesize = fs->fs_maxfilesize;		/* XXX */
489 	maxfilesize = (u_int64_t)0x40000000 * fs->fs_bsize - 1;	/* XXX */
490 	if (fs->fs_maxfilesize > maxfilesize)			/* XXX */
491 		fs->fs_maxfilesize = maxfilesize;		/* XXX */
492 	if (ronly == 0) {
493 		fs->fs_clean = 0;
494 		(void) ffs_sbupdate(ump, MNT_WAIT);
495 	}
496 	return (0);
497 out:
498 	if (bp)
499 		brelse(bp);
500 	(void)VOP_CLOSE(devvp, ronly ? FREAD : FREAD|FWRITE, cred, p);
501 	if (ump) {
502 		free(ump->um_fs, M_UFSMNT);
503 		free(ump, M_UFSMNT);
504 		mp->mnt_data = (qaddr_t)0;
505 	}
506 	return (error);
507 }
508 
509 /*
510  * Sanity checks for old file systems.
511  *
512  * XXX - goes away some day.
513  */
514 ffs_oldfscompat(fs)
515 	struct fs *fs;
516 {
517 	int i;
518 
519 	fs->fs_npsect = max(fs->fs_npsect, fs->fs_nsect);	/* XXX */
520 	fs->fs_interleave = max(fs->fs_interleave, 1);		/* XXX */
521 	if (fs->fs_postblformat == FS_42POSTBLFMT)		/* XXX */
522 		fs->fs_nrpos = 8;				/* XXX */
523 	if (fs->fs_inodefmt < FS_44INODEFMT) {			/* XXX */
524 		u_int64_t sizepb = fs->fs_bsize;		/* XXX */
525 								/* XXX */
526 		fs->fs_maxfilesize = fs->fs_bsize * NDADDR - 1;	/* XXX */
527 		for (i = 0; i < NIADDR; i++) {			/* XXX */
528 			sizepb *= NINDIR(fs);			/* XXX */
529 			fs->fs_maxfilesize += sizepb;		/* XXX */
530 		}						/* XXX */
531 		fs->fs_qbmask = ~fs->fs_bmask;			/* XXX */
532 		fs->fs_qfmask = ~fs->fs_fmask;			/* XXX */
533 	}							/* XXX */
534 	return (0);
535 }
536 
537 /*
538  * unmount system call
539  */
540 int
541 ffs_unmount(mp, mntflags, p)
542 	struct mount *mp;
543 	int mntflags;
544 	struct proc *p;
545 {
546 	register struct ufsmount *ump;
547 	register struct fs *fs;
548 	int error, flags;
549 
550 	flags = 0;
551 	if (mntflags & MNT_FORCE) {
552 		if (mp->mnt_flag & MNT_ROOTFS)
553 			return (EINVAL);
554 		flags |= FORCECLOSE;
555 	}
556 	if (error = ffs_flushfiles(mp, flags, p))
557 		return (error);
558 	ump = VFSTOUFS(mp);
559 	fs = ump->um_fs;
560 	if (fs->fs_ronly == 0) {
561 		fs->fs_clean = 1;
562 		if (error = ffs_sbupdate(ump, MNT_WAIT)) {
563 			fs->fs_clean = 0;
564 			return (error);
565 		}
566 	}
567 	ump->um_devvp->v_specflags &= ~SI_MOUNTEDON;
568 	error = VOP_CLOSE(ump->um_devvp, fs->fs_ronly ? FREAD : FREAD|FWRITE,
569 		NOCRED, p);
570 	vrele(ump->um_devvp);
571 	free(fs->fs_csp[0], M_UFSMNT);
572 	free(fs, M_UFSMNT);
573 	free(ump, M_UFSMNT);
574 	mp->mnt_data = (qaddr_t)0;
575 	return (error);
576 }
577 
578 /*
579  * Flush out all the files in a filesystem.
580  */
581 ffs_flushfiles(mp, flags, p)
582 	register struct mount *mp;
583 	int flags;
584 	struct proc *p;
585 {
586 	extern int doforce;
587 	register struct ufsmount *ump;
588 	int i, error;
589 
590 	if (!doforce)
591 		flags &= ~FORCECLOSE;
592 	ump = VFSTOUFS(mp);
593 #ifdef QUOTA
594 	if (mp->mnt_flag & MNT_QUOTA) {
595 		if (error = vflush(mp, NULLVP, SKIPSYSTEM|flags))
596 			return (error);
597 		for (i = 0; i < MAXQUOTAS; i++) {
598 			if (ump->um_quotas[i] == NULLVP)
599 				continue;
600 			quotaoff(p, mp, i);
601 		}
602 		/*
603 		 * Here we fall through to vflush again to ensure
604 		 * that we have gotten rid of all the system vnodes.
605 		 */
606 	}
607 #endif
608 	error = vflush(mp, NULLVP, flags);
609 	return (error);
610 }
611 
612 /*
613  * Get file system statistics.
614  */
615 int
616 ffs_statfs(mp, sbp, p)
617 	struct mount *mp;
618 	register struct statfs *sbp;
619 	struct proc *p;
620 {
621 	register struct ufsmount *ump;
622 	register struct fs *fs;
623 
624 	ump = VFSTOUFS(mp);
625 	fs = ump->um_fs;
626 	if (fs->fs_magic != FS_MAGIC)
627 		panic("ffs_statfs");
628 	sbp->f_bsize = fs->fs_fsize;
629 	sbp->f_iosize = fs->fs_bsize;
630 	sbp->f_blocks = fs->fs_dsize;
631 	sbp->f_bfree = fs->fs_cstotal.cs_nbfree * fs->fs_frag +
632 		fs->fs_cstotal.cs_nffree;
633 	sbp->f_bavail = (fs->fs_dsize * (100 - fs->fs_minfree) / 100) -
634 		(fs->fs_dsize - sbp->f_bfree);
635 	sbp->f_files =  fs->fs_ncg * fs->fs_ipg - ROOTINO;
636 	sbp->f_ffree = fs->fs_cstotal.cs_nifree;
637 	if (sbp != &mp->mnt_stat) {
638 		sbp->f_type = mp->mnt_vfc->vfc_typenum;
639 		bcopy((caddr_t)mp->mnt_stat.f_mntonname,
640 			(caddr_t)&sbp->f_mntonname[0], MNAMELEN);
641 		bcopy((caddr_t)mp->mnt_stat.f_mntfromname,
642 			(caddr_t)&sbp->f_mntfromname[0], MNAMELEN);
643 	}
644 	return (0);
645 }
646 
647 /*
648  * Go through the disk queues to initiate sandbagged IO;
649  * go through the inodes to write those that have been modified;
650  * initiate the writing of the super block if it has been modified.
651  *
652  * Note: we are always called with the filesystem marked `MPBUSY'.
653  */
654 int
655 ffs_sync(mp, waitfor, cred, p)
656 	struct mount *mp;
657 	int waitfor;
658 	struct ucred *cred;
659 	struct proc *p;
660 {
661 	register struct vnode *vp;
662 	register struct inode *ip;
663 	register struct ufsmount *ump = VFSTOUFS(mp);
664 	register struct fs *fs;
665 	int error, allerror = 0;
666 
667 	fs = ump->um_fs;
668 	if (fs->fs_ronly != 0) {		/* XXX */
669 		printf("fs = %s\n", fs->fs_fsmnt);
670 		panic("update: rofs mod");
671 	}
672 	/*
673 	 * Write back each (modified) inode.
674 	 */
675 loop:
676 	for (vp = mp->mnt_vnodelist.lh_first;
677 	     vp != NULL;
678 	     vp = vp->v_mntvnodes.le_next) {
679 		/*
680 		 * If the vnode that we are about to sync is no longer
681 		 * associated with this mount point, start over.
682 		 */
683 		if (vp->v_mount != mp)
684 			goto loop;
685 		if (VOP_ISLOCKED(vp))
686 			continue;
687 		ip = VTOI(vp);
688 		if ((ip->i_flag &
689 		    (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) == 0 &&
690 		    vp->v_dirtyblkhd.lh_first == NULL)
691 			continue;
692 		if (vget(vp, 1))
693 			goto loop;
694 		if (error = VOP_FSYNC(vp, cred, waitfor, p))
695 			allerror = error;
696 		vput(vp);
697 	}
698 	/*
699 	 * Force stale file system control information to be flushed.
700 	 */
701 	if (error = VOP_FSYNC(ump->um_devvp, cred, waitfor, p))
702 		allerror = error;
703 #ifdef QUOTA
704 	qsync(mp);
705 #endif
706 	/*
707 	 * Write back modified superblock.
708 	 */
709 	if (fs->fs_fmod != 0) {
710 		fs->fs_fmod = 0;
711 		fs->fs_time = time.tv_sec;
712 		if (error = ffs_sbupdate(ump, waitfor))
713 			allerror = error;
714 	}
715 	return (allerror);
716 }
717 
718 /*
719  * Look up a FFS dinode number to find its incore vnode, otherwise read it
720  * in from disk.  If it is in core, wait for the lock bit to clear, then
721  * return the inode locked.  Detection and handling of mount points must be
722  * done by the calling routine.
723  */
724 int
725 ffs_vget(mp, ino, vpp)
726 	struct mount *mp;
727 	ino_t ino;
728 	struct vnode **vpp;
729 {
730 	register struct fs *fs;
731 	register struct inode *ip;
732 	struct ufsmount *ump;
733 	struct buf *bp;
734 	struct vnode *vp;
735 	dev_t dev;
736 	int i, type, error;
737 
738 	ump = VFSTOUFS(mp);
739 	dev = ump->um_dev;
740 	if ((*vpp = ufs_ihashget(dev, ino)) != NULL)
741 		return (0);
742 
743 	/* Allocate a new vnode/inode. */
744 	if (error = getnewvnode(VT_UFS, mp, ffs_vnodeop_p, &vp)) {
745 		*vpp = NULL;
746 		return (error);
747 	}
748 	type = ump->um_devvp->v_tag == VT_MFS ? M_MFSNODE : M_FFSNODE; /* XXX */
749 	MALLOC(ip, struct inode *, sizeof(struct inode), type, M_WAITOK);
750 	bzero((caddr_t)ip, sizeof(struct inode));
751 	vp->v_data = ip;
752 	ip->i_vnode = vp;
753 	ip->i_fs = fs = ump->um_fs;
754 	ip->i_dev = dev;
755 	ip->i_number = ino;
756 #ifdef QUOTA
757 	for (i = 0; i < MAXQUOTAS; i++)
758 		ip->i_dquot[i] = NODQUOT;
759 #endif
760 	/*
761 	 * Put it onto its hash chain and lock it so that other requests for
762 	 * this inode will block if they arrive while we are sleeping waiting
763 	 * for old data structures to be purged or for the contents of the
764 	 * disk portion of this inode to be read.
765 	 */
766 	ufs_ihashins(ip);
767 
768 	/* Read in the disk contents for the inode, copy into the inode. */
769 	if (error = bread(ump->um_devvp, fsbtodb(fs, ino_to_fsba(fs, ino)),
770 	    (int)fs->fs_bsize, NOCRED, &bp)) {
771 		/*
772 		 * The inode does not contain anything useful, so it would
773 		 * be misleading to leave it on its hash chain. With mode
774 		 * still zero, it will be unlinked and returned to the free
775 		 * list by vput().
776 		 */
777 		vput(vp);
778 		brelse(bp);
779 		*vpp = NULL;
780 		return (error);
781 	}
782 	ip->i_din = *((struct dinode *)bp->b_data + ino_to_fsbo(fs, ino));
783 	brelse(bp);
784 
785 	/*
786 	 * Initialize the vnode from the inode, check for aliases.
787 	 * Note that the underlying vnode may have changed.
788 	 */
789 	if (error = ufs_vinit(mp, ffs_specop_p, FFS_FIFOOPS, &vp)) {
790 		vput(vp);
791 		*vpp = NULL;
792 		return (error);
793 	}
794 	/*
795 	 * Finish inode initialization now that aliasing has been resolved.
796 	 */
797 	ip->i_devvp = ump->um_devvp;
798 	VREF(ip->i_devvp);
799 	/*
800 	 * Set up a generation number for this inode if it does not
801 	 * already have one. This should only happen on old filesystems.
802 	 */
803 	if (ip->i_gen == 0) {
804 		if (++nextgennumber < (u_long)time.tv_sec)
805 			nextgennumber = time.tv_sec;
806 		ip->i_gen = nextgennumber;
807 		if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0)
808 			ip->i_flag |= IN_MODIFIED;
809 	}
810 	/*
811 	 * Ensure that uid and gid are correct. This is a temporary
812 	 * fix until fsck has been changed to do the update.
813 	 */
814 	if (fs->fs_inodefmt < FS_44INODEFMT) {		/* XXX */
815 		ip->i_uid = ip->i_din.di_ouid;		/* XXX */
816 		ip->i_gid = ip->i_din.di_ogid;		/* XXX */
817 	}						/* XXX */
818 
819 	*vpp = vp;
820 	return (0);
821 }
822 
823 /*
824  * File handle to vnode
825  *
826  * Have to be really careful about stale file handles:
827  * - check that the inode number is valid
828  * - call ffs_vget() to get the locked inode
829  * - check for an unallocated inode (i_mode == 0)
830  * - check that the given client host has export rights and return
831  *   those rights via. exflagsp and credanonp
832  */
833 int
834 ffs_fhtovp(mp, fhp, nam, vpp, exflagsp, credanonp)
835 	register struct mount *mp;
836 	struct fid *fhp;
837 	struct mbuf *nam;
838 	struct vnode **vpp;
839 	int *exflagsp;
840 	struct ucred **credanonp;
841 {
842 	register struct ufid *ufhp;
843 	struct fs *fs;
844 
845 	ufhp = (struct ufid *)fhp;
846 	fs = VFSTOUFS(mp)->um_fs;
847 	if (ufhp->ufid_ino < ROOTINO ||
848 	    ufhp->ufid_ino >= fs->fs_ncg * fs->fs_ipg)
849 		return (ESTALE);
850 	return (ufs_check_export(mp, ufhp, nam, vpp, exflagsp, credanonp));
851 }
852 
853 /*
854  * Vnode pointer to File handle
855  */
856 /* ARGSUSED */
857 ffs_vptofh(vp, fhp)
858 	struct vnode *vp;
859 	struct fid *fhp;
860 {
861 	register struct inode *ip;
862 	register struct ufid *ufhp;
863 
864 	ip = VTOI(vp);
865 	ufhp = (struct ufid *)fhp;
866 	ufhp->ufid_len = sizeof(struct ufid);
867 	ufhp->ufid_ino = ip->i_number;
868 	ufhp->ufid_gen = ip->i_gen;
869 	return (0);
870 }
871 
872 /*
873  * Initialize the filesystem; just use ufs_init.
874  */
875 int
876 ffs_init(vfsp)
877 	struct vfsconf *vfsp;
878 {
879 
880 	return (ufs_init(vfsp));
881 }
882 
883 /*
884  * fast filesystem related variables.
885  */
886 ffs_sysctl(name, namelen, oldp, oldlenp, newp, newlen, p)
887 	int *name;
888 	u_int namelen;
889 	void *oldp;
890 	size_t *oldlenp;
891 	void *newp;
892 	size_t newlen;
893 	struct proc *p;
894 {
895 	extern int doclusterread, doclusterwrite, doreallocblks, doasyncfree;
896 
897 	/* all sysctl names at this level are terminal */
898 	if (namelen != 1)
899 		return (ENOTDIR);		/* overloaded */
900 
901 	switch (name[0]) {
902 	case FFS_CLUSTERREAD:
903 		return (sysctl_int(oldp, oldlenp, newp, newlen,
904 		    &doclusterread));
905 	case FFS_CLUSTERWRITE:
906 		return (sysctl_int(oldp, oldlenp, newp, newlen,
907 		    &doclusterwrite));
908 	case FFS_REALLOCBLKS:
909 		return (sysctl_int(oldp, oldlenp, newp, newlen,
910 		    &doreallocblks));
911 	case FFS_ASYNCFREE:
912 		return (sysctl_int(oldp, oldlenp, newp, newlen, &doasyncfree));
913 	default:
914 		return (EOPNOTSUPP);
915 	}
916 	/* NOTREACHED */
917 }
918 
919 /*
920  * Write a superblock and associated information back to disk.
921  */
922 int
923 ffs_sbupdate(mp, waitfor)
924 	struct ufsmount *mp;
925 	int waitfor;
926 {
927 	register struct fs *dfs, *fs = mp->um_fs;
928 	register struct buf *bp;
929 	int blks;
930 	caddr_t space;
931 	int i, size, error, allerror = 0;
932 
933 	/*
934 	 * First write back the summary information.
935 	 */
936 	blks = howmany(fs->fs_cssize, fs->fs_fsize);
937 	space = (caddr_t)fs->fs_csp[0];
938 	for (i = 0; i < blks; i += fs->fs_frag) {
939 		size = fs->fs_bsize;
940 		if (i + fs->fs_frag > blks)
941 			size = (blks - i) * fs->fs_fsize;
942 		bp = getblk(mp->um_devvp, fsbtodb(fs, fs->fs_csaddr + i),
943 		    size, 0, 0);
944 		bcopy(space, bp->b_data, (u_int)size);
945 		space += size;
946 		if (waitfor != MNT_WAIT)
947 			bawrite(bp);
948 		else if (error = bwrite(bp))
949 			allerror = error;
950 	}
951 	/*
952 	 * Now write back the superblock itself. If any errors occurred
953 	 * up to this point, then fail so that the superblock avoids
954 	 * being written out as clean.
955 	 */
956 	if (allerror)
957 		return (allerror);
958 	bp = getblk(mp->um_devvp, SBLOCK, (int)fs->fs_sbsize, 0, 0);
959 	bcopy((caddr_t)fs, bp->b_data, (u_int)fs->fs_sbsize);
960 	/* Restore compatibility to old file systems.		   XXX */
961 	dfs = (struct fs *)bp->b_data;				/* XXX */
962 	if (fs->fs_postblformat == FS_42POSTBLFMT)		/* XXX */
963 		dfs->fs_nrpos = -1;				/* XXX */
964 	if (fs->fs_inodefmt < FS_44INODEFMT) {			/* XXX */
965 		int32_t *lp, tmp;				/* XXX */
966 								/* XXX */
967 		lp = (int32_t *)&dfs->fs_qbmask;		/* XXX */
968 		tmp = lp[4];					/* XXX */
969 		for (i = 4; i > 0; i--)				/* XXX */
970 			lp[i] = lp[i-1];			/* XXX */
971 		lp[0] = tmp;					/* XXX */
972 	}							/* XXX */
973 	dfs->fs_maxfilesize = mp->um_savedmaxfilesize;		/* XXX */
974 	if (waitfor != MNT_WAIT)
975 		bawrite(bp);
976 	else if (error = bwrite(bp))
977 		allerror = error;
978 	return (allerror);
979 }
980