xref: /netbsd-src/sys/fs/union/union_vfsops.c (revision bf1e9b32e27832f0c493206710fb8b58a980838a)
1 /*	$NetBSD: union_vfsops.c,v 1.27 2005/05/29 21:00:29 christos 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 1.27 2005/05/29 21:00:29 christos 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 __P((struct mount *, const char *, void *, struct nameidata *,
98 		     struct proc *));
99 int union_start __P((struct mount *, int, struct proc *));
100 int union_unmount __P((struct mount *, int, struct proc *));
101 int union_root __P((struct mount *, struct vnode **));
102 int union_quotactl __P((struct mount *, int, uid_t, void *, struct proc *));
103 int union_statvfs __P((struct mount *, struct statvfs *, struct proc *));
104 int union_sync __P((struct mount *, int, struct ucred *, struct proc *));
105 int union_vget __P((struct mount *, ino_t, struct vnode **));
106 int union_fhtovp __P((struct mount *, struct fid *, struct vnode **));
107 int union_checkexp __P((struct mount *, struct mbuf *, int *,
108 		      struct ucred **));
109 int union_vptofh __P((struct vnode *, struct fid *));
110 
111 /*
112  * Mount union filesystem
113  */
114 int
115 union_mount(mp, path, data, ndp, p)
116 	struct mount *mp;
117 	const char *path;
118 	void *data;
119 	struct nameidata *ndp;
120 	struct proc *p;
121 {
122 	int error = 0;
123 	struct union_args args;
124 	struct vnode *lowerrootvp = NULLVP;
125 	struct vnode *upperrootvp = NULLVP;
126 	struct union_mount *um = 0;
127 	struct ucred *cred = 0;
128 	const char *cp;
129 	char *xp;
130 	int len;
131 	size_t size;
132 
133 #ifdef UNION_DIAGNOSTIC
134 	printf("union_mount(mp = %p)\n", mp);
135 #endif
136 
137 	if (mp->mnt_flag & MNT_GETARGS) {
138 		um = MOUNTTOUNIONMOUNT(mp);
139 		if (um == NULL)
140 			return EIO;
141 		args.target = NULL;
142 		args.mntflags = um->um_op;
143 		return copyout(&args, data, sizeof(args));
144 	}
145 	/*
146 	 * Update is a no-op
147 	 */
148 	if (mp->mnt_flag & MNT_UPDATE) {
149 		/*
150 		 * Need to provide.
151 		 * 1. a way to convert between rdonly and rdwr mounts.
152 		 * 2. support for nfs exports.
153 		 */
154 		error = EOPNOTSUPP;
155 		goto bad;
156 	}
157 
158 	/*
159 	 * Get argument
160 	 */
161 	error = copyin(data, &args, sizeof(struct union_args));
162 	if (error)
163 		goto bad;
164 
165 	lowerrootvp = mp->mnt_vnodecovered;
166 	VREF(lowerrootvp);
167 
168 	/*
169 	 * Find upper node.
170 	 */
171 	NDINIT(ndp, LOOKUP, FOLLOW,
172 	       UIO_USERSPACE, args.target, p);
173 
174 	if ((error = namei(ndp)) != 0)
175 		goto bad;
176 
177 	upperrootvp = ndp->ni_vp;
178 
179 	if (upperrootvp->v_type != VDIR) {
180 		error = EINVAL;
181 		goto bad;
182 	}
183 
184 	um = (struct union_mount *) malloc(sizeof(struct union_mount),
185 				M_UFSMNT, M_WAITOK);	/* XXX */
186 
187 	/*
188 	 * Keep a held reference to the target vnodes.
189 	 * They are vrele'd in union_unmount.
190 	 *
191 	 * Depending on the _BELOW flag, the filesystems are
192 	 * viewed in a different order.  In effect, this is the
193 	 * same as providing a mount under option to the mount syscall.
194 	 */
195 
196 	um->um_op = args.mntflags & UNMNT_OPMASK;
197 	switch (um->um_op) {
198 	case UNMNT_ABOVE:
199 		um->um_lowervp = lowerrootvp;
200 		um->um_uppervp = upperrootvp;
201 		break;
202 
203 	case UNMNT_BELOW:
204 		um->um_lowervp = upperrootvp;
205 		um->um_uppervp = lowerrootvp;
206 		break;
207 
208 	case UNMNT_REPLACE:
209 		vrele(lowerrootvp);
210 		lowerrootvp = NULLVP;
211 		um->um_uppervp = upperrootvp;
212 		um->um_lowervp = lowerrootvp;
213 		break;
214 
215 	default:
216 		error = EINVAL;
217 		goto bad;
218 	}
219 
220 	/*
221 	 * Unless the mount is readonly, ensure that the top layer
222 	 * supports whiteout operations
223 	 */
224 	if ((mp->mnt_flag & MNT_RDONLY) == 0) {
225 		error = VOP_WHITEOUT(um->um_uppervp, (struct componentname *) 0, LOOKUP);
226 		if (error)
227 			goto bad;
228 	}
229 
230 	um->um_cred = p->p_ucred;
231 	crhold(um->um_cred);
232 	um->um_cmode = UN_DIRMODE &~ p->p_cwdi->cwdi_cmask;
233 
234 	/*
235 	 * Depending on what you think the MNT_LOCAL flag might mean,
236 	 * you may want the && to be || on the conditional below.
237 	 * At the moment it has been defined that the filesystem is
238 	 * only local if it is all local, ie the MNT_LOCAL flag implies
239 	 * that the entire namespace is local.  If you think the MNT_LOCAL
240 	 * flag implies that some of the files might be stored locally
241 	 * then you will want to change the conditional.
242 	 */
243 	if (um->um_op == UNMNT_ABOVE) {
244 		if (((um->um_lowervp == NULLVP) ||
245 		     (um->um_lowervp->v_mount->mnt_flag & MNT_LOCAL)) &&
246 		    (um->um_uppervp->v_mount->mnt_flag & MNT_LOCAL))
247 			mp->mnt_flag |= MNT_LOCAL;
248 	}
249 
250 	/*
251 	 * Copy in the upper layer's RDONLY flag.  This is for the benefit
252 	 * of lookup() which explicitly checks the flag, rather than asking
253 	 * the filesystem for it's own opinion.  This means, that an update
254 	 * mount of the underlying filesystem to go from rdonly to rdwr
255 	 * will leave the unioned view as read-only.
256 	 */
257 	mp->mnt_flag |= (um->um_uppervp->v_mount->mnt_flag & MNT_RDONLY);
258 
259 	mp->mnt_data = um;
260 	mp->mnt_leaf = um->um_uppervp->v_mount->mnt_leaf;
261 	vfs_getnewfsid(mp);
262 
263 	error = set_statvfs_info( path, UIO_USERSPACE, NULL, UIO_USERSPACE,
264 	    mp, p);
265 	if (error)
266 		goto bad;
267 
268 	switch (um->um_op) {
269 	case UNMNT_ABOVE:
270 		cp = "<above>:";
271 		break;
272 	case UNMNT_BELOW:
273 		cp = "<below>:";
274 		break;
275 	case UNMNT_REPLACE:
276 		cp = "";
277 		break;
278 	default:
279 		cp = "<invalid>:";
280 #ifdef DIAGNOSTIC
281 		panic("union_mount: bad um_op");
282 #endif
283 		break;
284 	}
285 	len = strlen(cp);
286 	memcpy(mp->mnt_stat.f_mntfromname, cp, len);
287 
288 	xp = mp->mnt_stat.f_mntfromname + len;
289 	len = MNAMELEN - len;
290 
291 	(void) copyinstr(args.target, xp, len - 1, &size);
292 	memset(xp + size, 0, len - size);
293 
294 #ifdef UNION_DIAGNOSTIC
295 	printf("union_mount: from %s, on %s\n",
296 	    mp->mnt_stat.f_mntfromname, mp->mnt_stat.f_mntonname);
297 #endif
298 
299 	/* Setup the readdir hook if it's not set already */
300 	if (!vn_union_readdir_hook)
301 		vn_union_readdir_hook = union_readdirhook;
302 
303 	return (0);
304 
305 bad:
306 	if (um)
307 		free(um, M_UFSMNT);
308 	if (cred)
309 		crfree(cred);
310 	if (upperrootvp)
311 		vrele(upperrootvp);
312 	if (lowerrootvp)
313 		vrele(lowerrootvp);
314 	return (error);
315 }
316 
317 /*
318  * VFS start.  Nothing needed here - the start routine
319  * on the underlying filesystem(s) will have been called
320  * when that filesystem was mounted.
321  */
322  /*ARGSUSED*/
323 int
324 union_start(mp, flags, p)
325 	struct mount *mp;
326 	int flags;
327 	struct proc *p;
328 {
329 
330 	return (0);
331 }
332 
333 /*
334  * Free reference to union layer
335  */
336 int
337 union_unmount(mp, mntflags, p)
338 	struct mount *mp;
339 	int mntflags;
340 	struct proc *p;
341 {
342 	struct union_mount *um = MOUNTTOUNIONMOUNT(mp);
343 	struct vnode *um_rootvp;
344 	int error;
345 	int freeing;
346 
347 #ifdef UNION_DIAGNOSTIC
348 	printf("union_unmount(mp = %p)\n", mp);
349 #endif
350 
351 	if ((error = union_root(mp, &um_rootvp)) != 0)
352 		return (error);
353 
354 	/*
355 	 * Keep flushing vnodes from the mount list.
356 	 * This is needed because of the un_pvp held
357 	 * reference to the parent vnode.
358 	 * If more vnodes have been freed on a given pass,
359 	 * the try again.  The loop will iterate at most
360 	 * (d) times, where (d) is the maximum tree depth
361 	 * in the filesystem.
362 	 */
363 	for (freeing = 0; vflush(mp, um_rootvp, 0) != 0;) {
364 		struct vnode *vp;
365 		int n;
366 
367 		/* count #vnodes held on mount list */
368 		for (n = 0, vp = mp->mnt_vnodelist.lh_first;
369 				vp != NULLVP;
370 				vp = vp->v_mntvnodes.le_next)
371 			n++;
372 
373 		/* if this is unchanged then stop */
374 		if (n == freeing)
375 			break;
376 
377 		/* otherwise try once more time */
378 		freeing = n;
379 	}
380 
381 	/*
382 	 * Ok, now that we've tried doing it gently, get out the hammer.
383 	 */
384 
385 	if (mntflags & MNT_FORCE)
386 		vflush(mp, um_rootvp, FORCECLOSE);
387 
388 
389 	/* At this point the root vnode should have a single reference */
390 	if (um_rootvp->v_usecount > 1) {
391 		vput(um_rootvp);
392 		return (EBUSY);
393 	}
394 
395 #ifdef UNION_DIAGNOSTIC
396 	vprint("union root", um_rootvp);
397 #endif
398 	/*
399 	 * Discard references to upper and lower target vnodes.
400 	 */
401 	if (um->um_lowervp)
402 		vrele(um->um_lowervp);
403 	vrele(um->um_uppervp);
404 	crfree(um->um_cred);
405 	/*
406 	 * Release reference on underlying root vnode
407 	 */
408 	vput(um_rootvp);
409 	/*
410 	 * And blow it away for future re-use
411 	 */
412 	vgone(um_rootvp);
413 	/*
414 	 * Finally, throw away the union_mount structure
415 	 */
416 	free(mp->mnt_data, M_UFSMNT);	/* XXX */
417 	mp->mnt_data = 0;
418 	return (0);
419 }
420 
421 int
422 union_root(mp, vpp)
423 	struct mount *mp;
424 	struct vnode **vpp;
425 {
426 	struct union_mount *um = MOUNTTOUNIONMOUNT(mp);
427 	int error;
428 	int loselock;
429 
430 	/*
431 	 * Return locked reference to root.
432 	 */
433 	VREF(um->um_uppervp);
434 	if ((um->um_op == UNMNT_BELOW) &&
435 	     VOP_ISLOCKED(um->um_uppervp)) {
436 		loselock = 1;
437 	} else {
438 		vn_lock(um->um_uppervp, LK_EXCLUSIVE | LK_RETRY);
439 		loselock = 0;
440 	}
441 	if (um->um_lowervp)
442 		VREF(um->um_lowervp);
443 	error = union_allocvp(vpp, mp,
444 			      (struct vnode *) 0,
445 			      (struct vnode *) 0,
446 			      (struct componentname *) 0,
447 			      um->um_uppervp,
448 			      um->um_lowervp,
449 			      1);
450 
451 	if (error) {
452 		if (!loselock)
453 			VOP_UNLOCK(um->um_uppervp, 0);
454 		vrele(um->um_uppervp);
455 		if (um->um_lowervp)
456 			vrele(um->um_lowervp);
457 	} else {
458 		if (loselock)
459 			VTOUNION(*vpp)->un_flags &= ~UN_ULOCK;
460 	}
461 
462 	return (error);
463 }
464 
465 /*ARGSUSED*/
466 int
467 union_quotactl(mp, cmd, uid, arg, p)
468 	struct mount *mp;
469 	int cmd;
470 	uid_t uid;
471 	void *arg;
472 	struct proc *p;
473 {
474 
475 	return (EOPNOTSUPP);
476 }
477 
478 int
479 union_statvfs(mp, sbp, p)
480 	struct mount *mp;
481 	struct statvfs *sbp;
482 	struct proc *p;
483 {
484 	int error;
485 	struct union_mount *um = MOUNTTOUNIONMOUNT(mp);
486 	struct statvfs *sbuf = malloc(sizeof(*sbuf), M_TEMP, M_WAITOK);
487 	unsigned long lbsize;
488 
489 #ifdef UNION_DIAGNOSTIC
490 	printf("union_statvfs(mp = %p, lvp = %p, uvp = %p)\n", mp,
491 	    um->um_lowervp, um->um_uppervp);
492 #endif
493 
494 	if (um->um_lowervp) {
495 		error = VFS_STATVFS(um->um_lowervp->v_mount, sbuf, p);
496 		if (error)
497 			goto done;
498 	}
499 
500 	/* now copy across the "interesting" information and fake the rest */
501 	lbsize = sbuf->f_bsize;
502 	sbp->f_blocks = sbuf->f_blocks - sbuf->f_bfree;
503 	sbp->f_files = sbuf->f_files - sbuf->f_ffree;
504 
505 	error = VFS_STATVFS(um->um_uppervp->v_mount, sbuf, p);
506 	if (error)
507 		goto done;
508 
509 	sbp->f_flag = sbuf->f_flag;
510 	sbp->f_bsize = sbuf->f_bsize;
511 	sbp->f_frsize = sbuf->f_frsize;
512 	sbp->f_iosize = sbuf->f_iosize;
513 
514 	/*
515 	 * The "total" fields count total resources in all layers,
516 	 * the "free" fields count only those resources which are
517 	 * free in the upper layer (since only the upper layer
518 	 * is writable).
519 	 */
520 
521 	if (sbuf->f_bsize != lbsize)
522 		sbp->f_blocks = sbp->f_blocks * lbsize / sbuf->f_bsize;
523 	sbp->f_blocks += sbuf->f_blocks;
524 	sbp->f_bfree = sbuf->f_bfree;
525 	sbp->f_bavail = sbuf->f_bavail;
526 	sbp->f_bresvd = sbuf->f_bresvd;
527 	sbp->f_files += sbuf->f_files;
528 	sbp->f_ffree = sbuf->f_ffree;
529 	sbp->f_favail = sbuf->f_favail;
530 	sbp->f_fresvd = sbuf->f_fresvd;
531 
532 	copy_statvfs_info(sbp, mp);
533 done:
534 	free(sbuf, M_TEMP);
535 	return error;
536 }
537 
538 /*ARGSUSED*/
539 int
540 union_sync(mp, waitfor, cred, p)
541 	struct mount *mp;
542 	int waitfor;
543 	struct ucred *cred;
544 	struct proc *p;
545 {
546 
547 	/*
548 	 * XXX - Assumes no data cached at union layer.
549 	 */
550 	return (0);
551 }
552 
553 /*ARGSUSED*/
554 int
555 union_vget(mp, ino, vpp)
556 	struct mount *mp;
557 	ino_t ino;
558 	struct vnode **vpp;
559 {
560 
561 	return (EOPNOTSUPP);
562 }
563 
564 /*ARGSUSED*/
565 int
566 union_fhtovp(mp, fidp, vpp)
567 	struct mount *mp;
568 	struct fid *fidp;
569 	struct vnode **vpp;
570 {
571 
572 	return (EOPNOTSUPP);
573 }
574 
575 /*ARGSUSED*/
576 int
577 union_checkexp(mp, nam, exflagsp, credanonp)
578 	struct mount *mp;
579 	struct mbuf *nam;
580 	int *exflagsp;
581 	struct ucred **credanonp;
582 {
583 
584 	return (EOPNOTSUPP);
585 }
586 
587 /*ARGSUSED*/
588 int
589 union_vptofh(vp, fhp)
590 	struct vnode *vp;
591 	struct fid *fhp;
592 {
593 
594 	return (EOPNOTSUPP);
595 }
596 
597 SYSCTL_SETUP(sysctl_vfs_union_setup, "sysctl vfs.union subtree setup")
598 {
599 
600 	sysctl_createv(clog, 0, NULL, NULL,
601 		       CTLFLAG_PERMANENT,
602 		       CTLTYPE_NODE, "vfs", NULL,
603 		       NULL, 0, NULL, 0,
604 		       CTL_VFS, CTL_EOL);
605 	sysctl_createv(clog, 0, NULL, NULL,
606 		       CTLFLAG_PERMANENT,
607 		       CTLTYPE_NODE, "union",
608 		       SYSCTL_DESCR("Union file system"),
609 		       NULL, 0, NULL, 0,
610 		       CTL_VFS, 15, CTL_EOL);
611 	/*
612 	 * XXX the "15" above could be dynamic, thereby eliminating
613 	 * one more instance of the "number to vfs" mapping problem,
614 	 * but "15" is the order as taken from sys/mount.h
615 	 */
616 }
617 
618 extern const struct vnodeopv_desc union_vnodeop_opv_desc;
619 
620 const struct vnodeopv_desc * const union_vnodeopv_descs[] = {
621 	&union_vnodeop_opv_desc,
622 	NULL,
623 };
624 
625 struct vfsops union_vfsops = {
626 	MOUNT_UNION,
627 	union_mount,
628 	union_start,
629 	union_unmount,
630 	union_root,
631 	union_quotactl,
632 	union_statvfs,
633 	union_sync,
634 	union_vget,
635 	union_fhtovp,
636 	union_vptofh,
637 	union_init,
638 	NULL,				/* vfs_reinit */
639 	union_done,
640 	NULL,
641 	NULL,				/* vfs_mountroot */
642 	union_checkexp,
643 	(int (*)(struct mount *, struct vnode *, struct timespec *)) eopnotsupp,
644 	vfs_stdextattrctl,
645 	union_vnodeopv_descs,
646 };
647 VFS_ATTACH(union_vfsops);
648