xref: /netbsd-src/sys/fs/union/union_vfsops.c (revision aaf4ece63a859a04e37cf3a7229b5fab0157cc06)
1 /*	$NetBSD: union_vfsops.c,v 1.32 2006/01/05 20:31:33 wrstuden Exp $	*/
2 
3 /*
4  * Copyright (c) 1994 The Regents of the University of California.
5  * All rights reserved.
6  *
7  * This code is derived from software donated to Berkeley by
8  * Jan-Simon Pendry.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)union_vfsops.c	8.20 (Berkeley) 5/20/95
35  */
36 
37 /*
38  * Copyright (c) 1994 Jan-Simon Pendry.
39  * All rights reserved.
40  *
41  * This code is derived from software donated to Berkeley by
42  * Jan-Simon Pendry.
43  *
44  * Redistribution and use in source and binary forms, with or without
45  * modification, are permitted provided that the following conditions
46  * are met:
47  * 1. Redistributions of source code must retain the above copyright
48  *    notice, this list of conditions and the following disclaimer.
49  * 2. Redistributions in binary form must reproduce the above copyright
50  *    notice, this list of conditions and the following disclaimer in the
51  *    documentation and/or other materials provided with the distribution.
52  * 3. All advertising materials mentioning features or use of this software
53  *    must display the following acknowledgement:
54  *	This product includes software developed by the University of
55  *	California, Berkeley and its contributors.
56  * 4. Neither the name of the University nor the names of its contributors
57  *    may be used to endorse or promote products derived from this software
58  *    without specific prior written permission.
59  *
60  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
61  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
62  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
63  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
64  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
65  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
66  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
67  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
68  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
69  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
70  * SUCH DAMAGE.
71  *
72  *	@(#)union_vfsops.c	8.20 (Berkeley) 5/20/95
73  */
74 
75 /*
76  * Union Layer
77  */
78 
79 #include <sys/cdefs.h>
80 __KERNEL_RCSID(0, "$NetBSD: union_vfsops.c,v 1.32 2006/01/05 20:31:33 wrstuden Exp $");
81 
82 #include <sys/param.h>
83 #include <sys/systm.h>
84 #include <sys/sysctl.h>
85 #include <sys/time.h>
86 #include <sys/proc.h>
87 #include <sys/vnode.h>
88 #include <sys/mount.h>
89 #include <sys/namei.h>
90 #include <sys/malloc.h>
91 #include <sys/filedesc.h>
92 #include <sys/queue.h>
93 #include <sys/stat.h>
94 
95 #include <fs/union/union.h>
96 
97 int union_mount(struct mount *, const char *, void *, struct nameidata *,
98 		     struct lwp *);
99 int union_start(struct mount *, int, struct lwp *);
100 int union_unmount(struct mount *, int, struct lwp *);
101 int union_root(struct mount *, struct vnode **);
102 int union_quotactl(struct mount *, int, uid_t, void *, struct lwp *);
103 int union_statvfs(struct mount *, struct statvfs *, struct lwp *);
104 int union_sync(struct mount *, int, struct ucred *, struct lwp *);
105 int union_vget(struct mount *, ino_t, struct vnode **);
106 
107 /*
108  * Mount union filesystem
109  */
110 int
111 union_mount(mp, path, data, ndp, l)
112 	struct mount *mp;
113 	const char *path;
114 	void *data;
115 	struct nameidata *ndp;
116 	struct lwp *l;
117 {
118 	int error = 0;
119 	struct union_args args;
120 	struct vnode *lowerrootvp = NULLVP;
121 	struct vnode *upperrootvp = NULLVP;
122 	struct union_mount *um = 0;
123 	struct ucred *cred = 0;
124 	const char *cp;
125 	char *xp;
126 	int len;
127 	size_t size;
128 
129 #ifdef UNION_DIAGNOSTIC
130 	printf("union_mount(mp = %p)\n", mp);
131 #endif
132 
133 	if (mp->mnt_flag & MNT_GETARGS) {
134 		um = MOUNTTOUNIONMOUNT(mp);
135 		if (um == NULL)
136 			return EIO;
137 		args.target = NULL;
138 		args.mntflags = um->um_op;
139 		return copyout(&args, data, sizeof(args));
140 	}
141 	/*
142 	 * Update is a no-op
143 	 */
144 	if (mp->mnt_flag & MNT_UPDATE) {
145 		/*
146 		 * Need to provide.
147 		 * 1. a way to convert between rdonly and rdwr mounts.
148 		 * 2. support for nfs exports.
149 		 */
150 		error = EOPNOTSUPP;
151 		goto bad;
152 	}
153 
154 	/*
155 	 * Get argument
156 	 */
157 	error = copyin(data, &args, sizeof(struct union_args));
158 	if (error)
159 		goto bad;
160 
161 	lowerrootvp = mp->mnt_vnodecovered;
162 	VREF(lowerrootvp);
163 
164 	/*
165 	 * Find upper node.
166 	 */
167 	NDINIT(ndp, LOOKUP, FOLLOW,
168 	       UIO_USERSPACE, args.target, l);
169 
170 	if ((error = namei(ndp)) != 0)
171 		goto bad;
172 
173 	upperrootvp = ndp->ni_vp;
174 
175 	if (upperrootvp->v_type != VDIR) {
176 		error = EINVAL;
177 		goto bad;
178 	}
179 
180 	um = (struct union_mount *) malloc(sizeof(struct union_mount),
181 				M_UFSMNT, M_WAITOK);	/* XXX */
182 
183 	/*
184 	 * Keep a held reference to the target vnodes.
185 	 * They are vrele'd in union_unmount.
186 	 *
187 	 * Depending on the _BELOW flag, the filesystems are
188 	 * viewed in a different order.  In effect, this is the
189 	 * same as providing a mount under option to the mount syscall.
190 	 */
191 
192 	um->um_op = args.mntflags & UNMNT_OPMASK;
193 	switch (um->um_op) {
194 	case UNMNT_ABOVE:
195 		um->um_lowervp = lowerrootvp;
196 		um->um_uppervp = upperrootvp;
197 		break;
198 
199 	case UNMNT_BELOW:
200 		um->um_lowervp = upperrootvp;
201 		um->um_uppervp = lowerrootvp;
202 		break;
203 
204 	case UNMNT_REPLACE:
205 		vrele(lowerrootvp);
206 		lowerrootvp = NULLVP;
207 		um->um_uppervp = upperrootvp;
208 		um->um_lowervp = lowerrootvp;
209 		break;
210 
211 	default:
212 		error = EINVAL;
213 		goto bad;
214 	}
215 
216 	/*
217 	 * Unless the mount is readonly, ensure that the top layer
218 	 * supports whiteout operations
219 	 */
220 	if ((mp->mnt_flag & MNT_RDONLY) == 0) {
221 		error = VOP_WHITEOUT(um->um_uppervp, (struct componentname *) 0, LOOKUP);
222 		if (error)
223 			goto bad;
224 	}
225 
226 	um->um_cred = l->l_proc->p_ucred;
227 	crhold(um->um_cred);
228 	um->um_cmode = UN_DIRMODE &~ l->l_proc->p_cwdi->cwdi_cmask;
229 
230 	/*
231 	 * Depending on what you think the MNT_LOCAL flag might mean,
232 	 * you may want the && to be || on the conditional below.
233 	 * At the moment it has been defined that the filesystem is
234 	 * only local if it is all local, ie the MNT_LOCAL flag implies
235 	 * that the entire namespace is local.  If you think the MNT_LOCAL
236 	 * flag implies that some of the files might be stored locally
237 	 * then you will want to change the conditional.
238 	 */
239 	if (um->um_op == UNMNT_ABOVE) {
240 		if (((um->um_lowervp == NULLVP) ||
241 		     (um->um_lowervp->v_mount->mnt_flag & MNT_LOCAL)) &&
242 		    (um->um_uppervp->v_mount->mnt_flag & MNT_LOCAL))
243 			mp->mnt_flag |= MNT_LOCAL;
244 	}
245 
246 	/*
247 	 * Copy in the upper layer's RDONLY flag.  This is for the benefit
248 	 * of lookup() which explicitly checks the flag, rather than asking
249 	 * the filesystem for it's own opinion.  This means, that an update
250 	 * mount of the underlying filesystem to go from rdonly to rdwr
251 	 * will leave the unioned view as read-only.
252 	 */
253 	mp->mnt_flag |= (um->um_uppervp->v_mount->mnt_flag & MNT_RDONLY);
254 
255 	mp->mnt_data = um;
256 	mp->mnt_leaf = um->um_uppervp->v_mount->mnt_leaf;
257 	vfs_getnewfsid(mp);
258 
259 	error = set_statvfs_info( path, UIO_USERSPACE, NULL, UIO_USERSPACE,
260 	    mp, l);
261 	if (error)
262 		goto bad;
263 
264 	switch (um->um_op) {
265 	case UNMNT_ABOVE:
266 		cp = "<above>:";
267 		break;
268 	case UNMNT_BELOW:
269 		cp = "<below>:";
270 		break;
271 	case UNMNT_REPLACE:
272 		cp = "";
273 		break;
274 	default:
275 		cp = "<invalid>:";
276 #ifdef DIAGNOSTIC
277 		panic("union_mount: bad um_op");
278 #endif
279 		break;
280 	}
281 	len = strlen(cp);
282 	memcpy(mp->mnt_stat.f_mntfromname, cp, len);
283 
284 	xp = mp->mnt_stat.f_mntfromname + len;
285 	len = MNAMELEN - len;
286 
287 	(void) copyinstr(args.target, xp, len - 1, &size);
288 	memset(xp + size, 0, len - size);
289 
290 #ifdef UNION_DIAGNOSTIC
291 	printf("union_mount: from %s, on %s\n",
292 	    mp->mnt_stat.f_mntfromname, mp->mnt_stat.f_mntonname);
293 #endif
294 
295 	/* Setup the readdir hook if it's not set already */
296 	if (!vn_union_readdir_hook)
297 		vn_union_readdir_hook = union_readdirhook;
298 
299 	return (0);
300 
301 bad:
302 	if (um)
303 		free(um, M_UFSMNT);
304 	if (cred)
305 		crfree(cred);
306 	if (upperrootvp)
307 		vrele(upperrootvp);
308 	if (lowerrootvp)
309 		vrele(lowerrootvp);
310 	return (error);
311 }
312 
313 /*
314  * VFS start.  Nothing needed here - the start routine
315  * on the underlying filesystem(s) will have been called
316  * when that filesystem was mounted.
317  */
318  /*ARGSUSED*/
319 int
320 union_start(mp, flags, l)
321 	struct mount *mp;
322 	int flags;
323 	struct lwp *l;
324 {
325 
326 	return (0);
327 }
328 
329 /*
330  * Free reference to union layer
331  */
332 int
333 union_unmount(mp, mntflags, l)
334 	struct mount *mp;
335 	int mntflags;
336 	struct lwp *l;
337 {
338 	struct union_mount *um = MOUNTTOUNIONMOUNT(mp);
339 	int freeing;
340 
341 #ifdef UNION_DIAGNOSTIC
342 	printf("union_unmount(mp = %p)\n", mp);
343 #endif
344 
345 	/*
346 	 * Keep flushing vnodes from the mount list.
347 	 * This is needed because of the un_pvp held
348 	 * reference to the parent vnode.
349 	 * If more vnodes have been freed on a given pass,
350 	 * the try again.  The loop will iterate at most
351 	 * (d) times, where (d) is the maximum tree depth
352 	 * in the filesystem.
353 	 */
354 	for (freeing = 0; vflush(mp, NULL, 0) != 0;) {
355 		struct vnode *vp;
356 		int n;
357 
358 		/* count #vnodes held on mount list */
359 		for (n = 0, vp = mp->mnt_vnodelist.lh_first;
360 				vp != NULLVP;
361 				vp = vp->v_mntvnodes.le_next)
362 			n++;
363 
364 		/* if this is unchanged then stop */
365 		if (n == freeing)
366 			break;
367 
368 		/* otherwise try once more time */
369 		freeing = n;
370 	}
371 
372 	/*
373 	 * Ok, now that we've tried doing it gently, get out the hammer.
374 	 */
375 
376 	if (mntflags & MNT_FORCE)
377 		vflush(mp, NULL, FORCECLOSE);
378 
379 
380 	/*
381 	 * Discard references to upper and lower target vnodes.
382 	 */
383 	if (um->um_lowervp)
384 		vrele(um->um_lowervp);
385 	vrele(um->um_uppervp);
386 	crfree(um->um_cred);
387 	/*
388 	 * Finally, throw away the union_mount structure
389 	 */
390 	free(mp->mnt_data, M_UFSMNT);	/* XXX */
391 	mp->mnt_data = 0;
392 	return (0);
393 }
394 
395 int
396 union_root(mp, vpp)
397 	struct mount *mp;
398 	struct vnode **vpp;
399 {
400 	struct union_mount *um = MOUNTTOUNIONMOUNT(mp);
401 	int error;
402 	int loselock;
403 
404 	/*
405 	 * Return locked reference to root.
406 	 */
407 	VREF(um->um_uppervp);
408 	if ((um->um_op == UNMNT_BELOW) &&
409 	     VOP_ISLOCKED(um->um_uppervp)) {
410 		loselock = 1;
411 	} else {
412 		vn_lock(um->um_uppervp, LK_EXCLUSIVE | LK_RETRY);
413 		loselock = 0;
414 	}
415 	if (um->um_lowervp)
416 		VREF(um->um_lowervp);
417 	error = union_allocvp(vpp, mp,
418 			      (struct vnode *) 0,
419 			      (struct vnode *) 0,
420 			      (struct componentname *) 0,
421 			      um->um_uppervp,
422 			      um->um_lowervp,
423 			      1);
424 
425 	if (error) {
426 		if (!loselock)
427 			VOP_UNLOCK(um->um_uppervp, 0);
428 		vrele(um->um_uppervp);
429 		if (um->um_lowervp)
430 			vrele(um->um_lowervp);
431 	} else {
432 		if (loselock)
433 			VTOUNION(*vpp)->un_flags &= ~UN_ULOCK;
434 	}
435 
436 	return (error);
437 }
438 
439 /*ARGSUSED*/
440 int
441 union_quotactl(mp, cmd, uid, arg, l)
442 	struct mount *mp;
443 	int cmd;
444 	uid_t uid;
445 	void *arg;
446 	struct lwp *l;
447 {
448 
449 	return (EOPNOTSUPP);
450 }
451 
452 int
453 union_statvfs(mp, sbp, l)
454 	struct mount *mp;
455 	struct statvfs *sbp;
456 	struct lwp *l;
457 {
458 	int error;
459 	struct union_mount *um = MOUNTTOUNIONMOUNT(mp);
460 	struct statvfs *sbuf = malloc(sizeof(*sbuf), M_TEMP, M_WAITOK | M_ZERO);
461 	unsigned long lbsize;
462 
463 #ifdef UNION_DIAGNOSTIC
464 	printf("union_statvfs(mp = %p, lvp = %p, uvp = %p)\n", mp,
465 	    um->um_lowervp, um->um_uppervp);
466 #endif
467 
468 	if (um->um_lowervp) {
469 		error = VFS_STATVFS(um->um_lowervp->v_mount, sbuf, l);
470 		if (error)
471 			goto done;
472 	}
473 
474 	/* now copy across the "interesting" information and fake the rest */
475 	lbsize = sbuf->f_bsize;
476 	sbp->f_blocks = sbuf->f_blocks - sbuf->f_bfree;
477 	sbp->f_files = sbuf->f_files - sbuf->f_ffree;
478 
479 	error = VFS_STATVFS(um->um_uppervp->v_mount, sbuf, l);
480 	if (error)
481 		goto done;
482 
483 	sbp->f_flag = sbuf->f_flag;
484 	sbp->f_bsize = sbuf->f_bsize;
485 	sbp->f_frsize = sbuf->f_frsize;
486 	sbp->f_iosize = sbuf->f_iosize;
487 
488 	/*
489 	 * The "total" fields count total resources in all layers,
490 	 * the "free" fields count only those resources which are
491 	 * free in the upper layer (since only the upper layer
492 	 * is writable).
493 	 */
494 
495 	if (sbuf->f_bsize != lbsize)
496 		sbp->f_blocks = sbp->f_blocks * lbsize / sbuf->f_bsize;
497 	sbp->f_blocks += sbuf->f_blocks;
498 	sbp->f_bfree = sbuf->f_bfree;
499 	sbp->f_bavail = sbuf->f_bavail;
500 	sbp->f_bresvd = sbuf->f_bresvd;
501 	sbp->f_files += sbuf->f_files;
502 	sbp->f_ffree = sbuf->f_ffree;
503 	sbp->f_favail = sbuf->f_favail;
504 	sbp->f_fresvd = sbuf->f_fresvd;
505 
506 	copy_statvfs_info(sbp, mp);
507 done:
508 	free(sbuf, M_TEMP);
509 	return error;
510 }
511 
512 /*ARGSUSED*/
513 int
514 union_sync(mp, waitfor, cred, l)
515 	struct mount *mp;
516 	int waitfor;
517 	struct ucred *cred;
518 	struct lwp *l;
519 {
520 
521 	/*
522 	 * XXX - Assumes no data cached at union layer.
523 	 */
524 	return (0);
525 }
526 
527 /*ARGSUSED*/
528 int
529 union_vget(mp, ino, vpp)
530 	struct mount *mp;
531 	ino_t ino;
532 	struct vnode **vpp;
533 {
534 
535 	return (EOPNOTSUPP);
536 }
537 
538 SYSCTL_SETUP(sysctl_vfs_union_setup, "sysctl vfs.union subtree setup")
539 {
540 
541 	sysctl_createv(clog, 0, NULL, NULL,
542 		       CTLFLAG_PERMANENT,
543 		       CTLTYPE_NODE, "vfs", NULL,
544 		       NULL, 0, NULL, 0,
545 		       CTL_VFS, CTL_EOL);
546 	sysctl_createv(clog, 0, NULL, NULL,
547 		       CTLFLAG_PERMANENT,
548 		       CTLTYPE_NODE, "union",
549 		       SYSCTL_DESCR("Union file system"),
550 		       NULL, 0, NULL, 0,
551 		       CTL_VFS, 15, CTL_EOL);
552 	/*
553 	 * XXX the "15" above could be dynamic, thereby eliminating
554 	 * one more instance of the "number to vfs" mapping problem,
555 	 * but "15" is the order as taken from sys/mount.h
556 	 */
557 }
558 
559 extern const struct vnodeopv_desc union_vnodeop_opv_desc;
560 
561 const struct vnodeopv_desc * const union_vnodeopv_descs[] = {
562 	&union_vnodeop_opv_desc,
563 	NULL,
564 };
565 
566 struct vfsops union_vfsops = {
567 	MOUNT_UNION,
568 	union_mount,
569 	union_start,
570 	union_unmount,
571 	union_root,
572 	union_quotactl,
573 	union_statvfs,
574 	union_sync,
575 	union_vget,
576 	NULL,				/* vfs_fhtovp */
577 	NULL,				/* vfs_vptofh */
578 	union_init,
579 	NULL,				/* vfs_reinit */
580 	union_done,
581 	NULL,				/* vfs_mountroot */
582 	(int (*)(struct mount *, struct vnode *, struct timespec *)) eopnotsupp,
583 	vfs_stdextattrctl,
584 	union_vnodeopv_descs,
585 };
586 VFS_ATTACH(union_vfsops);
587