xref: /original-bsd/sys/ufs/ffs/ffs_vfsops.c (revision e031425c)
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	8.17 (Berkeley) 01/02/95
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 };
51 
52 extern u_long nextgennumber;
53 
54 /*
55  * Called by main() when ufs is going to be mounted as root.
56  *
57  * Name is updated by mount(8) after booting.
58  */
59 #define ROOTNAME	"root_device"
60 
61 ffs_mountroot()
62 {
63 	extern struct vnode *rootvp;
64 	register struct fs *fs;
65 	register struct mount *mp;
66 	struct proc *p = curproc;	/* XXX */
67 	struct ufsmount *ump;
68 	u_int size;
69 	int error;
70 
71 	/*
72 	 * Get vnodes for swapdev and rootdev.
73 	 */
74 	if (bdevvp(swapdev, &swapdev_vp) || bdevvp(rootdev, &rootvp))
75 		panic("ffs_mountroot: can't setup bdevvp's");
76 
77 	mp = malloc((u_long)sizeof(struct mount), M_MOUNT, M_WAITOK);
78 	bzero((char *)mp, (u_long)sizeof(struct mount));
79 	mp->mnt_op = &ufs_vfsops;
80 	mp->mnt_flag = MNT_RDONLY;
81 	if (error = ffs_mountfs(rootvp, mp, p)) {
82 		free(mp, M_MOUNT);
83 		return (error);
84 	}
85 	if (error = vfs_lock(mp)) {
86 		(void)ffs_unmount(mp, 0, p);
87 		free(mp, M_MOUNT);
88 		return (error);
89 	}
90 	TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
91 	mp->mnt_flag |= MNT_ROOTFS;
92 	mp->mnt_vnodecovered = NULLVP;
93 	ump = VFSTOUFS(mp);
94 	fs = ump->um_fs;
95 	bzero(fs->fs_fsmnt, sizeof(fs->fs_fsmnt));
96 	fs->fs_fsmnt[0] = '/';
97 	bcopy((caddr_t)fs->fs_fsmnt, (caddr_t)mp->mnt_stat.f_mntonname,
98 	    MNAMELEN);
99 	(void) copystr(ROOTNAME, mp->mnt_stat.f_mntfromname, MNAMELEN - 1,
100 	    &size);
101 	bzero(mp->mnt_stat.f_mntfromname + size, MNAMELEN - size);
102 	(void)ffs_statfs(mp, &mp->mnt_stat, p);
103 	vfs_unlock(mp);
104 	inittodr(fs->fs_time);
105 	return (0);
106 }
107 
108 /*
109  * VFS Operations.
110  *
111  * mount system call
112  */
113 int
114 ffs_mount(mp, path, data, ndp, p)
115 	register struct mount *mp;
116 	char *path;
117 	caddr_t data;
118 	struct nameidata *ndp;
119 	struct proc *p;
120 {
121 	struct vnode *devvp;
122 	struct ufs_args args;
123 	struct ufsmount *ump;
124 	register struct fs *fs;
125 	u_int size;
126 	int error, flags;
127 	mode_t accessmode;
128 
129 	if (error = copyin(data, (caddr_t)&args, sizeof (struct ufs_args)))
130 		return (error);
131 	/*
132 	 * If updating, check whether changing from read-only to
133 	 * read/write; if there is no device name, that's all we do.
134 	 */
135 	if (mp->mnt_flag & MNT_UPDATE) {
136 		ump = VFSTOUFS(mp);
137 		fs = ump->um_fs;
138 		error = 0;
139 		if (fs->fs_ronly == 0 && (mp->mnt_flag & MNT_RDONLY)) {
140 			flags = WRITECLOSE;
141 			if (mp->mnt_flag & MNT_FORCE)
142 				flags |= FORCECLOSE;
143 			if (vfs_busy(mp))
144 				return (EBUSY);
145 			error = ffs_flushfiles(mp, flags, p);
146 			vfs_unbusy(mp);
147 		}
148 		if (!error && (mp->mnt_flag & MNT_RELOAD))
149 			error = ffs_reload(mp, ndp->ni_cnd.cn_cred, p);
150 		if (error)
151 			return (error);
152 		if (fs->fs_ronly && (mp->mnt_flag & MNT_WANTRDWR)) {
153 			/*
154 			 * If upgrade to read-write by non-root, then verify
155 			 * that user has necessary permissions on the device.
156 			 */
157 			if (p->p_ucred->cr_uid != 0) {
158 				devvp = ump->um_devvp;
159 				VOP_LOCK(devvp);
160 				if (error = VOP_ACCESS(devvp, VREAD | VWRITE,
161 				    p->p_ucred, p)) {
162 					VOP_UNLOCK(devvp);
163 					return (error);
164 				}
165 				VOP_UNLOCK(devvp);
166 			}
167 			fs->fs_ronly = 0;
168 		}
169 		if (args.fspec == 0) {
170 			/*
171 			 * Process export requests.
172 			 */
173 			return (vfs_export(mp, &ump->um_export, &args.export));
174 		}
175 	}
176 	/*
177 	 * Not an update, or updating the name: look up the name
178 	 * and verify that it refers to a sensible block device.
179 	 */
180 	NDINIT(ndp, LOOKUP, FOLLOW, UIO_USERSPACE, args.fspec, p);
181 	if (error = namei(ndp))
182 		return (error);
183 	devvp = ndp->ni_vp;
184 
185 	if (devvp->v_type != VBLK) {
186 		vrele(devvp);
187 		return (ENOTBLK);
188 	}
189 	if (major(devvp->v_rdev) >= nblkdev) {
190 		vrele(devvp);
191 		return (ENXIO);
192 	}
193 	/*
194 	 * If mount by non-root, then verify that user has necessary
195 	 * permissions on the device.
196 	 */
197 	if (p->p_ucred->cr_uid != 0) {
198 		accessmode = VREAD;
199 		if ((mp->mnt_flag & MNT_RDONLY) == 0)
200 			accessmode |= VWRITE;
201 		VOP_LOCK(devvp);
202 		if (error = VOP_ACCESS(devvp, accessmode, p->p_ucred, p)) {
203 			vput(devvp);
204 			return (error);
205 		}
206 		VOP_UNLOCK(devvp);
207 	}
208 	if ((mp->mnt_flag & MNT_UPDATE) == 0)
209 		error = ffs_mountfs(devvp, mp, p);
210 	else {
211 		if (devvp != ump->um_devvp)
212 			error = EINVAL;	/* needs translation */
213 		else
214 			vrele(devvp);
215 	}
216 	if (error) {
217 		vrele(devvp);
218 		return (error);
219 	}
220 	ump = VFSTOUFS(mp);
221 	fs = ump->um_fs;
222 	(void) copyinstr(path, fs->fs_fsmnt, sizeof(fs->fs_fsmnt) - 1, &size);
223 	bzero(fs->fs_fsmnt + size, sizeof(fs->fs_fsmnt) - size);
224 	bcopy((caddr_t)fs->fs_fsmnt, (caddr_t)mp->mnt_stat.f_mntonname,
225 	    MNAMELEN);
226 	(void) copyinstr(args.fspec, mp->mnt_stat.f_mntfromname, MNAMELEN - 1,
227 	    &size);
228 	bzero(mp->mnt_stat.f_mntfromname + size, MNAMELEN - size);
229 	(void)ffs_statfs(mp, &mp->mnt_stat, p);
230 	return (0);
231 }
232 
233 /*
234  * Reload all incore data for a filesystem (used after running fsck on
235  * the root filesystem and finding things to fix). The filesystem must
236  * be mounted read-only.
237  *
238  * Things to do to update the mount:
239  *	1) invalidate all cached meta-data.
240  *	2) re-read superblock from disk.
241  *	3) re-read summary information from disk.
242  *	4) invalidate all inactive vnodes.
243  *	5) invalidate all cached file data.
244  *	6) re-read inode data for all active vnodes.
245  */
246 ffs_reload(mountp, cred, p)
247 	register struct mount *mountp;
248 	struct ucred *cred;
249 	struct proc *p;
250 {
251 	register struct vnode *vp, *nvp, *devvp;
252 	struct inode *ip;
253 	struct csum *space;
254 	struct buf *bp;
255 	struct fs *fs;
256 	struct partinfo dpart;
257 	int i, blks, size, error;
258 
259 	if ((mountp->mnt_flag & MNT_RDONLY) == 0)
260 		return (EINVAL);
261 	/*
262 	 * Step 1: invalidate all cached meta-data.
263 	 */
264 	devvp = VFSTOUFS(mountp)->um_devvp;
265 	if (vinvalbuf(devvp, 0, cred, p, 0, 0))
266 		panic("ffs_reload: dirty1");
267 	/*
268 	 * Step 2: re-read superblock from disk.
269 	 */
270 	if (VOP_IOCTL(devvp, DIOCGPART, (caddr_t)&dpart, FREAD, NOCRED, p) != 0)
271 		size = DEV_BSIZE;
272 	else
273 		size = dpart.disklab->d_secsize;
274 	if (error = bread(devvp, (daddr_t)(SBOFF / size), SBSIZE, NOCRED, &bp))
275 		return (error);
276 	fs = (struct fs *)bp->b_data;
277 	if (fs->fs_magic != FS_MAGIC || fs->fs_bsize > MAXBSIZE ||
278 	    fs->fs_bsize < sizeof(struct fs)) {
279 		brelse(bp);
280 		return (EIO);		/* XXX needs translation */
281 	}
282 	fs = VFSTOUFS(mountp)->um_fs;
283 	bcopy(&fs->fs_csp[0], &((struct fs *)bp->b_data)->fs_csp[0],
284 	    sizeof(fs->fs_csp));
285 	bcopy(bp->b_data, fs, (u_int)fs->fs_sbsize);
286 	if (fs->fs_sbsize < SBSIZE)
287 		bp->b_flags |= B_INVAL;
288 	brelse(bp);
289 	mountp->mnt_maxsymlinklen = fs->fs_maxsymlinklen;
290 	ffs_oldfscompat(fs);
291 	/*
292 	 * Step 3: re-read summary information from disk.
293 	 */
294 	blks = howmany(fs->fs_cssize, fs->fs_fsize);
295 	space = fs->fs_csp[0];
296 	for (i = 0; i < blks; i += fs->fs_frag) {
297 		size = fs->fs_bsize;
298 		if (i + fs->fs_frag > blks)
299 			size = (blks - i) * fs->fs_fsize;
300 		if (error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), size,
301 		    NOCRED, &bp))
302 			return (error);
303 		bcopy(bp->b_data, fs->fs_csp[fragstoblks(fs, i)], (u_int)size);
304 		brelse(bp);
305 	}
306 loop:
307 	for (vp = mountp->mnt_vnodelist.lh_first; vp != NULL; vp = nvp) {
308 		nvp = vp->v_mntvnodes.le_next;
309 		/*
310 		 * Step 4: invalidate all inactive vnodes.
311 		 */
312 		if (vp->v_usecount == 0) {
313 			vgone(vp);
314 			continue;
315 		}
316 		/*
317 		 * Step 5: invalidate all cached file data.
318 		 */
319 		if (vget(vp, 1))
320 			goto loop;
321 		if (vinvalbuf(vp, 0, cred, p, 0, 0))
322 			panic("ffs_reload: dirty2");
323 		/*
324 		 * Step 6: re-read inode data for all active vnodes.
325 		 */
326 		ip = VTOI(vp);
327 		if (error =
328 		    bread(devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
329 		    (int)fs->fs_bsize, NOCRED, &bp)) {
330 			vput(vp);
331 			return (error);
332 		}
333 		ip->i_din = *((struct dinode *)bp->b_data +
334 		    ino_to_fsbo(fs, ip->i_number));
335 		brelse(bp);
336 		vput(vp);
337 		if (vp->v_mount != mountp)
338 			goto loop;
339 	}
340 	return (0);
341 }
342 
343 /*
344  * Common code for mount and mountroot
345  */
346 int
347 ffs_mountfs(devvp, mp, p)
348 	register struct vnode *devvp;
349 	struct mount *mp;
350 	struct proc *p;
351 {
352 	register struct ufsmount *ump;
353 	struct buf *bp;
354 	register struct fs *fs;
355 	dev_t dev;
356 	struct partinfo dpart;
357 	caddr_t base, space;
358 	int error, i, blks, size, ronly;
359 	int32_t *lp;
360 	struct ucred *cred;
361 	extern struct vnode *rootvp;
362 	u_int64_t maxfilesize;					/* XXX */
363 
364 	dev = devvp->v_rdev;
365 	cred = p ? p->p_ucred : NOCRED;
366 	/*
367 	 * Disallow multiple mounts of the same device.
368 	 * Disallow mounting of a device that is currently in use
369 	 * (except for root, which might share swap device for miniroot).
370 	 * Flush out any old buffers remaining from a previous use.
371 	 */
372 	if (error = vfs_mountedon(devvp))
373 		return (error);
374 	if (vcount(devvp) > 1 && devvp != rootvp)
375 		return (EBUSY);
376 	if (error = vinvalbuf(devvp, V_SAVE, cred, p, 0, 0))
377 		return (error);
378 
379 	ronly = (mp->mnt_flag & MNT_RDONLY) != 0;
380 	if (error = VOP_OPEN(devvp, ronly ? FREAD : FREAD|FWRITE, FSCRED, p))
381 		return (error);
382 	if (VOP_IOCTL(devvp, DIOCGPART, (caddr_t)&dpart, FREAD, cred, p) != 0)
383 		size = DEV_BSIZE;
384 	else
385 		size = dpart.disklab->d_secsize;
386 
387 	bp = NULL;
388 	ump = NULL;
389 	if (error = bread(devvp, (daddr_t)(SBOFF / size), SBSIZE, cred, &bp))
390 		goto out;
391 	fs = (struct fs *)bp->b_data;
392 	if (fs->fs_magic != FS_MAGIC || fs->fs_bsize > MAXBSIZE ||
393 	    fs->fs_bsize < sizeof(struct fs)) {
394 		error = EINVAL;		/* XXX needs translation */
395 		goto out;
396 	}
397 	/* XXX updating 4.2 FFS superblocks trashes rotational layout tables */
398 	if (fs->fs_postblformat == FS_42POSTBLFMT && !ronly) {
399 		error = EROFS;          /* needs translation */
400 		goto out;
401 	}
402 	ump = malloc(sizeof *ump, M_UFSMNT, M_WAITOK);
403 	bzero((caddr_t)ump, sizeof *ump);
404 	ump->um_fs = malloc((u_long)fs->fs_sbsize, M_UFSMNT,
405 	    M_WAITOK);
406 	bcopy(bp->b_data, ump->um_fs, (u_int)fs->fs_sbsize);
407 	if (fs->fs_sbsize < SBSIZE)
408 		bp->b_flags |= B_INVAL;
409 	brelse(bp);
410 	bp = NULL;
411 	fs = ump->um_fs;
412 	fs->fs_ronly = ronly;
413 	if (ronly == 0)
414 		fs->fs_fmod = 1;
415 	size = fs->fs_cssize;
416 	blks = howmany(size, fs->fs_fsize);
417 	if (fs->fs_contigsumsize > 0)
418 		size += fs->fs_ncg * sizeof(int32_t);
419 	base = space = malloc((u_long)size, M_UFSMNT, M_WAITOK);
420 	for (i = 0; i < blks; i += fs->fs_frag) {
421 		size = fs->fs_bsize;
422 		if (i + fs->fs_frag > blks)
423 			size = (blks - i) * fs->fs_fsize;
424 		if (error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), size,
425 		    cred, &bp)) {
426 			free(base, M_UFSMNT);
427 			goto out;
428 		}
429 		bcopy(bp->b_data, space, (u_int)size);
430 		fs->fs_csp[fragstoblks(fs, i)] = (struct csum *)space;
431 		space += size;
432 		brelse(bp);
433 		bp = NULL;
434 	}
435 	if (fs->fs_contigsumsize > 0) {
436 		fs->fs_maxcluster = lp = (int32_t *)space;
437 		for (i = 0; i < fs->fs_ncg; i++)
438 			*lp++ = fs->fs_contigsumsize;
439 	}
440 	mp->mnt_data = (qaddr_t)ump;
441 	mp->mnt_stat.f_fsid.val[0] = (long)dev;
442 	mp->mnt_stat.f_fsid.val[1] = MOUNT_UFS;
443 	mp->mnt_maxsymlinklen = fs->fs_maxsymlinklen;
444 	mp->mnt_flag |= MNT_LOCAL;
445 	ump->um_mountp = mp;
446 	ump->um_dev = dev;
447 	ump->um_devvp = devvp;
448 	ump->um_nindir = fs->fs_nindir;
449 	ump->um_bptrtodb = fs->fs_fsbtodb;
450 	ump->um_seqinc = fs->fs_frag;
451 	for (i = 0; i < MAXQUOTAS; i++)
452 		ump->um_quotas[i] = NULLVP;
453 	devvp->v_specflags |= SI_MOUNTEDON;
454 	ffs_oldfscompat(fs);
455 	ump->um_savedmaxfilesize = fs->fs_maxfilesize;		/* XXX */
456 	maxfilesize = (u_int64_t)0x40000000 * fs->fs_bsize - 1;	/* XXX */
457 	if (fs->fs_maxfilesize > maxfilesize)			/* XXX */
458 		fs->fs_maxfilesize = maxfilesize;		/* XXX */
459 	return (0);
460 out:
461 	if (bp)
462 		brelse(bp);
463 	(void)VOP_CLOSE(devvp, ronly ? FREAD : FREAD|FWRITE, cred, p);
464 	if (ump) {
465 		free(ump->um_fs, M_UFSMNT);
466 		free(ump, M_UFSMNT);
467 		mp->mnt_data = (qaddr_t)0;
468 	}
469 	return (error);
470 }
471 
472 /*
473  * Sanity checks for old file systems.
474  *
475  * XXX - goes away some day.
476  */
477 ffs_oldfscompat(fs)
478 	struct fs *fs;
479 {
480 	int i;
481 
482 	fs->fs_npsect = max(fs->fs_npsect, fs->fs_nsect);	/* XXX */
483 	fs->fs_interleave = max(fs->fs_interleave, 1);		/* XXX */
484 	if (fs->fs_postblformat == FS_42POSTBLFMT)		/* XXX */
485 		fs->fs_nrpos = 8;				/* XXX */
486 	if (fs->fs_inodefmt < FS_44INODEFMT) {			/* XXX */
487 		u_int64_t sizepb = fs->fs_bsize;		/* XXX */
488 								/* XXX */
489 		fs->fs_maxfilesize = fs->fs_bsize * NDADDR - 1;	/* XXX */
490 		for (i = 0; i < NIADDR; i++) {			/* XXX */
491 			sizepb *= NINDIR(fs);			/* XXX */
492 			fs->fs_maxfilesize += sizepb;		/* XXX */
493 		}						/* XXX */
494 		fs->fs_qbmask = ~fs->fs_bmask;			/* XXX */
495 		fs->fs_qfmask = ~fs->fs_fmask;			/* XXX */
496 	}							/* XXX */
497 	return (0);
498 }
499 
500 /*
501  * unmount system call
502  */
503 int
504 ffs_unmount(mp, mntflags, p)
505 	struct mount *mp;
506 	int mntflags;
507 	struct proc *p;
508 {
509 	register struct ufsmount *ump;
510 	register struct fs *fs;
511 	int error, flags;
512 
513 	flags = 0;
514 	if (mntflags & MNT_FORCE) {
515 		if (mp->mnt_flag & MNT_ROOTFS)
516 			return (EINVAL);
517 		flags |= FORCECLOSE;
518 	}
519 	if (error = ffs_flushfiles(mp, flags, p))
520 		return (error);
521 	ump = VFSTOUFS(mp);
522 	fs = ump->um_fs;
523 	ump->um_devvp->v_specflags &= ~SI_MOUNTEDON;
524 	error = VOP_CLOSE(ump->um_devvp, fs->fs_ronly ? FREAD : FREAD|FWRITE,
525 		NOCRED, p);
526 	vrele(ump->um_devvp);
527 	free(fs->fs_csp[0], M_UFSMNT);
528 	free(fs, M_UFSMNT);
529 	free(ump, M_UFSMNT);
530 	mp->mnt_data = (qaddr_t)0;
531 	mp->mnt_flag &= ~MNT_LOCAL;
532 	return (error);
533 }
534 
535 /*
536  * Flush out all the files in a filesystem.
537  */
538 ffs_flushfiles(mp, flags, p)
539 	register struct mount *mp;
540 	int flags;
541 	struct proc *p;
542 {
543 	extern int doforce;
544 	register struct ufsmount *ump;
545 	int i, error;
546 
547 	if (!doforce)
548 		flags &= ~FORCECLOSE;
549 	ump = VFSTOUFS(mp);
550 #ifdef QUOTA
551 	if (mp->mnt_flag & MNT_QUOTA) {
552 		if (error = vflush(mp, NULLVP, SKIPSYSTEM|flags))
553 			return (error);
554 		for (i = 0; i < MAXQUOTAS; i++) {
555 			if (ump->um_quotas[i] == NULLVP)
556 				continue;
557 			quotaoff(p, mp, i);
558 		}
559 		/*
560 		 * Here we fall through to vflush again to ensure
561 		 * that we have gotten rid of all the system vnodes.
562 		 */
563 	}
564 #endif
565 	error = vflush(mp, NULLVP, flags);
566 	return (error);
567 }
568 
569 /*
570  * Get file system statistics.
571  */
572 int
573 ffs_statfs(mp, sbp, p)
574 	struct mount *mp;
575 	register struct statfs *sbp;
576 	struct proc *p;
577 {
578 	register struct ufsmount *ump;
579 	register struct fs *fs;
580 
581 	ump = VFSTOUFS(mp);
582 	fs = ump->um_fs;
583 	if (fs->fs_magic != FS_MAGIC)
584 		panic("ffs_statfs");
585 	sbp->f_type = MOUNT_UFS;
586 	sbp->f_bsize = fs->fs_fsize;
587 	sbp->f_iosize = fs->fs_bsize;
588 	sbp->f_blocks = fs->fs_dsize;
589 	sbp->f_bfree = fs->fs_cstotal.cs_nbfree * fs->fs_frag +
590 		fs->fs_cstotal.cs_nffree;
591 	sbp->f_bavail = (fs->fs_dsize * (100 - fs->fs_minfree) / 100) -
592 		(fs->fs_dsize - sbp->f_bfree);
593 	sbp->f_files =  fs->fs_ncg * fs->fs_ipg - ROOTINO;
594 	sbp->f_ffree = fs->fs_cstotal.cs_nifree;
595 	if (sbp != &mp->mnt_stat) {
596 		bcopy((caddr_t)mp->mnt_stat.f_mntonname,
597 			(caddr_t)&sbp->f_mntonname[0], MNAMELEN);
598 		bcopy((caddr_t)mp->mnt_stat.f_mntfromname,
599 			(caddr_t)&sbp->f_mntfromname[0], MNAMELEN);
600 	}
601 	return (0);
602 }
603 
604 /*
605  * Go through the disk queues to initiate sandbagged IO;
606  * go through the inodes to write those that have been modified;
607  * initiate the writing of the super block if it has been modified.
608  *
609  * Note: we are always called with the filesystem marked `MPBUSY'.
610  */
611 int
612 ffs_sync(mp, waitfor, cred, p)
613 	struct mount *mp;
614 	int waitfor;
615 	struct ucred *cred;
616 	struct proc *p;
617 {
618 	register struct vnode *vp;
619 	register struct inode *ip;
620 	register struct ufsmount *ump = VFSTOUFS(mp);
621 	register struct fs *fs;
622 	int error, allerror = 0;
623 
624 	fs = ump->um_fs;
625 	/*
626 	 * Write back modified superblock.
627 	 * Consistency check that the superblock
628 	 * is still in the buffer cache.
629 	 */
630 	if (fs->fs_fmod != 0) {
631 		if (fs->fs_ronly != 0) {		/* XXX */
632 			printf("fs = %s\n", fs->fs_fsmnt);
633 			panic("update: rofs mod");
634 		}
635 		fs->fs_fmod = 0;
636 		fs->fs_time = time.tv_sec;
637 		allerror = ffs_sbupdate(ump, waitfor);
638 	}
639 	/*
640 	 * Write back each (modified) inode.
641 	 */
642 loop:
643 	for (vp = mp->mnt_vnodelist.lh_first;
644 	     vp != NULL;
645 	     vp = vp->v_mntvnodes.le_next) {
646 		/*
647 		 * If the vnode that we are about to sync is no longer
648 		 * associated with this mount point, start over.
649 		 */
650 		if (vp->v_mount != mp)
651 			goto loop;
652 		if (VOP_ISLOCKED(vp))
653 			continue;
654 		ip = VTOI(vp);
655 		if ((ip->i_flag &
656 		    (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) == 0 &&
657 		    vp->v_dirtyblkhd.lh_first == NULL)
658 			continue;
659 		if (vget(vp, 1))
660 			goto loop;
661 		if (error = VOP_FSYNC(vp, cred, waitfor, p))
662 			allerror = error;
663 		vput(vp);
664 	}
665 	/*
666 	 * Force stale file system control information to be flushed.
667 	 */
668 	if (error = VOP_FSYNC(ump->um_devvp, cred, waitfor, p))
669 		allerror = error;
670 #ifdef QUOTA
671 	qsync(mp);
672 #endif
673 	return (allerror);
674 }
675 
676 /*
677  * Look up a FFS dinode number to find its incore vnode, otherwise read it
678  * in from disk.  If it is in core, wait for the lock bit to clear, then
679  * return the inode locked.  Detection and handling of mount points must be
680  * done by the calling routine.
681  */
682 int
683 ffs_vget(mp, ino, vpp)
684 	struct mount *mp;
685 	ino_t ino;
686 	struct vnode **vpp;
687 {
688 	register struct fs *fs;
689 	register struct inode *ip;
690 	struct ufsmount *ump;
691 	struct buf *bp;
692 	struct vnode *vp;
693 	dev_t dev;
694 	int i, type, error;
695 
696 	ump = VFSTOUFS(mp);
697 	dev = ump->um_dev;
698 	if ((*vpp = ufs_ihashget(dev, ino)) != NULL)
699 		return (0);
700 
701 	/* Allocate a new vnode/inode. */
702 	if (error = getnewvnode(VT_UFS, mp, ffs_vnodeop_p, &vp)) {
703 		*vpp = NULL;
704 		return (error);
705 	}
706 	type = ump->um_devvp->v_tag == VT_MFS ? M_MFSNODE : M_FFSNODE; /* XXX */
707 	MALLOC(ip, struct inode *, sizeof(struct inode), type, M_WAITOK);
708 	bzero((caddr_t)ip, sizeof(struct inode));
709 	vp->v_data = ip;
710 	ip->i_vnode = vp;
711 	ip->i_fs = fs = ump->um_fs;
712 	ip->i_dev = dev;
713 	ip->i_number = ino;
714 #ifdef QUOTA
715 	for (i = 0; i < MAXQUOTAS; i++)
716 		ip->i_dquot[i] = NODQUOT;
717 #endif
718 	/*
719 	 * Put it onto its hash chain and lock it so that other requests for
720 	 * this inode will block if they arrive while we are sleeping waiting
721 	 * for old data structures to be purged or for the contents of the
722 	 * disk portion of this inode to be read.
723 	 */
724 	ufs_ihashins(ip);
725 
726 	/* Read in the disk contents for the inode, copy into the inode. */
727 	if (error = bread(ump->um_devvp, fsbtodb(fs, ino_to_fsba(fs, ino)),
728 	    (int)fs->fs_bsize, NOCRED, &bp)) {
729 		/*
730 		 * The inode does not contain anything useful, so it would
731 		 * be misleading to leave it on its hash chain. With mode
732 		 * still zero, it will be unlinked and returned to the free
733 		 * list by vput().
734 		 */
735 		vput(vp);
736 		brelse(bp);
737 		*vpp = NULL;
738 		return (error);
739 	}
740 	ip->i_din = *((struct dinode *)bp->b_data + ino_to_fsbo(fs, ino));
741 	brelse(bp);
742 
743 	/*
744 	 * Initialize the vnode from the inode, check for aliases.
745 	 * Note that the underlying vnode may have changed.
746 	 */
747 	if (error = ufs_vinit(mp, ffs_specop_p, FFS_FIFOOPS, &vp)) {
748 		vput(vp);
749 		*vpp = NULL;
750 		return (error);
751 	}
752 	/*
753 	 * Finish inode initialization now that aliasing has been resolved.
754 	 */
755 	ip->i_devvp = ump->um_devvp;
756 	VREF(ip->i_devvp);
757 	/*
758 	 * Set up a generation number for this inode if it does not
759 	 * already have one. This should only happen on old filesystems.
760 	 */
761 	if (ip->i_gen == 0) {
762 		if (++nextgennumber < (u_long)time.tv_sec)
763 			nextgennumber = time.tv_sec;
764 		ip->i_gen = nextgennumber;
765 		if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0)
766 			ip->i_flag |= IN_MODIFIED;
767 	}
768 	/*
769 	 * Ensure that uid and gid are correct. This is a temporary
770 	 * fix until fsck has been changed to do the update.
771 	 */
772 	if (fs->fs_inodefmt < FS_44INODEFMT) {		/* XXX */
773 		ip->i_uid = ip->i_din.di_ouid;		/* XXX */
774 		ip->i_gid = ip->i_din.di_ogid;		/* XXX */
775 	}						/* XXX */
776 
777 	*vpp = vp;
778 	return (0);
779 }
780 
781 /*
782  * File handle to vnode
783  *
784  * Have to be really careful about stale file handles:
785  * - check that the inode number is valid
786  * - call ffs_vget() to get the locked inode
787  * - check for an unallocated inode (i_mode == 0)
788  * - check that the given client host has export rights and return
789  *   those rights via. exflagsp and credanonp
790  */
791 int
792 ffs_fhtovp(mp, fhp, nam, vpp, exflagsp, credanonp)
793 	register struct mount *mp;
794 	struct fid *fhp;
795 	struct mbuf *nam;
796 	struct vnode **vpp;
797 	int *exflagsp;
798 	struct ucred **credanonp;
799 {
800 	register struct ufid *ufhp;
801 	struct fs *fs;
802 
803 	ufhp = (struct ufid *)fhp;
804 	fs = VFSTOUFS(mp)->um_fs;
805 	if (ufhp->ufid_ino < ROOTINO ||
806 	    ufhp->ufid_ino >= fs->fs_ncg * fs->fs_ipg)
807 		return (ESTALE);
808 	return (ufs_check_export(mp, ufhp, nam, vpp, exflagsp, credanonp));
809 }
810 
811 /*
812  * Vnode pointer to File handle
813  */
814 /* ARGSUSED */
815 ffs_vptofh(vp, fhp)
816 	struct vnode *vp;
817 	struct fid *fhp;
818 {
819 	register struct inode *ip;
820 	register struct ufid *ufhp;
821 
822 	ip = VTOI(vp);
823 	ufhp = (struct ufid *)fhp;
824 	ufhp->ufid_len = sizeof(struct ufid);
825 	ufhp->ufid_ino = ip->i_number;
826 	ufhp->ufid_gen = ip->i_gen;
827 	return (0);
828 }
829 
830 /*
831  * Write a superblock and associated information back to disk.
832  */
833 int
834 ffs_sbupdate(mp, waitfor)
835 	struct ufsmount *mp;
836 	int waitfor;
837 {
838 	register struct fs *dfs, *fs = mp->um_fs;
839 	register struct buf *bp;
840 	int blks;
841 	caddr_t space;
842 	int i, size, error = 0;
843 
844 	bp = getblk(mp->um_devvp, SBLOCK, (int)fs->fs_sbsize, 0, 0);
845 	bcopy((caddr_t)fs, bp->b_data, (u_int)fs->fs_sbsize);
846 	/* Restore compatibility to old file systems.		   XXX */
847 	dfs = (struct fs *)bp->b_data;				/* XXX */
848 	if (fs->fs_postblformat == FS_42POSTBLFMT)		/* XXX */
849 		dfs->fs_nrpos = -1;				/* XXX */
850 	if (fs->fs_inodefmt < FS_44INODEFMT) {			/* XXX */
851 		int32_t *lp, tmp;				/* XXX */
852 								/* XXX */
853 		lp = (int32_t *)&dfs->fs_qbmask;		/* XXX */
854 		tmp = lp[4];					/* XXX */
855 		for (i = 4; i > 0; i--)				/* XXX */
856 			lp[i] = lp[i-1];			/* XXX */
857 		lp[0] = tmp;					/* XXX */
858 	}							/* XXX */
859 	dfs->fs_maxfilesize = mp->um_savedmaxfilesize;		/* XXX */
860 	if (waitfor == MNT_WAIT)
861 		error = bwrite(bp);
862 	else
863 		bawrite(bp);
864 	blks = howmany(fs->fs_cssize, fs->fs_fsize);
865 	space = (caddr_t)fs->fs_csp[0];
866 	for (i = 0; i < blks; i += fs->fs_frag) {
867 		size = fs->fs_bsize;
868 		if (i + fs->fs_frag > blks)
869 			size = (blks - i) * fs->fs_fsize;
870 		bp = getblk(mp->um_devvp, fsbtodb(fs, fs->fs_csaddr + i),
871 		    size, 0, 0);
872 		bcopy(space, bp->b_data, (u_int)size);
873 		space += size;
874 		if (waitfor == MNT_WAIT)
875 			error = bwrite(bp);
876 		else
877 			bawrite(bp);
878 	}
879 	return (error);
880 }
881