xref: /netbsd-src/sys/fs/union/union_vfsops.c (revision c2f76ff004a2cb67efe5b12d97bd3ef7fe89e18d)
1 /*	$NetBSD: union_vfsops.c,v 1.63 2010/07/05 21:27:08 pooka 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 1.63 2010/07/05 21:27:08 pooka 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 #include <sys/kauth.h>
95 #include <sys/module.h>
96 
97 #include <fs/union/union.h>
98 
99 MODULE(MODULE_CLASS_VFS, union, NULL);
100 
101 VFS_PROTOS(union);
102 
103 static struct sysctllog *union_sysctl_log;
104 
105 /*
106  * Mount union filesystem
107  */
108 int
109 union_mount(struct mount *mp, const char *path, void *data, size_t *data_len)
110 {
111 	struct lwp *l = curlwp;
112 	int error = 0;
113 	struct union_args *args = data;
114 	struct vnode *lowerrootvp = NULLVP;
115 	struct vnode *upperrootvp = NULLVP;
116 	struct union_mount *um = 0;
117 	const char *cp;
118 	char *xp;
119 	int len;
120 	size_t size;
121 
122 	if (*data_len < sizeof *args)
123 		return EINVAL;
124 
125 #ifdef UNION_DIAGNOSTIC
126 	printf("union_mount(mp = %p)\n", mp);
127 #endif
128 
129 	if (mp->mnt_flag & MNT_GETARGS) {
130 		um = MOUNTTOUNIONMOUNT(mp);
131 		if (um == NULL)
132 			return EIO;
133 		args->target = NULL;
134 		args->mntflags = um->um_op;
135 		*data_len = sizeof *args;
136 		return 0;
137 	}
138 	/*
139 	 * Update is a no-op
140 	 */
141 	if (mp->mnt_flag & MNT_UPDATE) {
142 		/*
143 		 * Need to provide.
144 		 * 1. a way to convert between rdonly and rdwr mounts.
145 		 * 2. support for nfs exports.
146 		 */
147 		error = EOPNOTSUPP;
148 		goto bad;
149 	}
150 
151 	printf("WARNING: the union file system is experimental\n"
152 	    "WARNING: it can cause crashes and file system corruption\n");
153 
154 	lowerrootvp = mp->mnt_vnodecovered;
155 	vref(lowerrootvp);
156 
157 	/*
158 	 * Find upper node.
159 	 */
160 	error = namei_simple_user(args->target,
161 				NSM_FOLLOW_NOEMULROOT, &upperrootvp);
162 	if (error != 0)
163 		goto bad;
164 
165 	if (upperrootvp->v_type != VDIR) {
166 		error = EINVAL;
167 		goto bad;
168 	}
169 
170 	um = (struct union_mount *) malloc(sizeof(struct union_mount),
171 				M_UFSMNT, M_WAITOK);	/* XXX */
172 
173 	/*
174 	 * Keep a held reference to the target vnodes.
175 	 * They are vrele'd in union_unmount.
176 	 *
177 	 * Depending on the _BELOW flag, the filesystems are
178 	 * viewed in a different order.  In effect, this is the
179 	 * same as providing a mount under option to the mount syscall.
180 	 */
181 
182 	um->um_op = args->mntflags & UNMNT_OPMASK;
183 	switch (um->um_op) {
184 	case UNMNT_ABOVE:
185 		um->um_lowervp = lowerrootvp;
186 		um->um_uppervp = upperrootvp;
187 		break;
188 
189 	case UNMNT_BELOW:
190 		um->um_lowervp = upperrootvp;
191 		um->um_uppervp = lowerrootvp;
192 		break;
193 
194 	case UNMNT_REPLACE:
195 		vrele(lowerrootvp);
196 		lowerrootvp = NULLVP;
197 		um->um_uppervp = upperrootvp;
198 		um->um_lowervp = lowerrootvp;
199 		break;
200 
201 	default:
202 		error = EINVAL;
203 		goto bad;
204 	}
205 
206 	/*
207 	 * Unless the mount is readonly, ensure that the top layer
208 	 * supports whiteout operations
209 	 */
210 	if ((mp->mnt_flag & MNT_RDONLY) == 0) {
211 		vn_lock(um->um_uppervp, LK_EXCLUSIVE | LK_RETRY);
212 		error = VOP_WHITEOUT(um->um_uppervp,
213 		    (struct componentname *) 0, LOOKUP);
214 		VOP_UNLOCK(um->um_uppervp);
215 		if (error)
216 			goto bad;
217 	}
218 
219 	um->um_cred = l->l_cred;
220 	kauth_cred_hold(um->um_cred);
221 	um->um_cmode = UN_DIRMODE &~ l->l_proc->p_cwdi->cwdi_cmask;
222 
223 	/*
224 	 * Depending on what you think the MNT_LOCAL flag might mean,
225 	 * you may want the && to be || on the conditional below.
226 	 * At the moment it has been defined that the filesystem is
227 	 * only local if it is all local, ie the MNT_LOCAL flag implies
228 	 * that the entire namespace is local.  If you think the MNT_LOCAL
229 	 * flag implies that some of the files might be stored locally
230 	 * then you will want to change the conditional.
231 	 */
232 	if (um->um_op == UNMNT_ABOVE) {
233 		if (((um->um_lowervp == NULLVP) ||
234 		     (um->um_lowervp->v_mount->mnt_flag & MNT_LOCAL)) &&
235 		    (um->um_uppervp->v_mount->mnt_flag & MNT_LOCAL))
236 			mp->mnt_flag |= MNT_LOCAL;
237 	}
238 
239 	/*
240 	 * Copy in the upper layer's RDONLY flag.  This is for the benefit
241 	 * of lookup() which explicitly checks the flag, rather than asking
242 	 * the filesystem for it's own opinion.  This means, that an update
243 	 * mount of the underlying filesystem to go from rdonly to rdwr
244 	 * will leave the unioned view as read-only.
245 	 */
246 	mp->mnt_flag |= (um->um_uppervp->v_mount->mnt_flag & MNT_RDONLY);
247 
248 	mp->mnt_data = um;
249 	vfs_getnewfsid(mp);
250 
251 	error = set_statvfs_info( path, UIO_USERSPACE, NULL, UIO_USERSPACE,
252 	    mp->mnt_op->vfs_name, mp, l);
253 	if (error)
254 		goto bad;
255 
256 	switch (um->um_op) {
257 	case UNMNT_ABOVE:
258 		cp = "<above>:";
259 		break;
260 	case UNMNT_BELOW:
261 		cp = "<below>:";
262 		break;
263 	case UNMNT_REPLACE:
264 		cp = "";
265 		break;
266 	default:
267 		cp = "<invalid>:";
268 #ifdef DIAGNOSTIC
269 		panic("union_mount: bad um_op");
270 #endif
271 		break;
272 	}
273 	len = strlen(cp);
274 	memcpy(mp->mnt_stat.f_mntfromname, cp, len);
275 
276 	xp = mp->mnt_stat.f_mntfromname + len;
277 	len = MNAMELEN - len;
278 
279 	(void) copyinstr(args->target, xp, len - 1, &size);
280 	memset(xp + size, 0, len - size);
281 
282 #ifdef UNION_DIAGNOSTIC
283 	printf("union_mount: from %s, on %s\n",
284 	    mp->mnt_stat.f_mntfromname, mp->mnt_stat.f_mntonname);
285 #endif
286 
287 	/* Setup the readdir hook if it's not set already */
288 	if (!vn_union_readdir_hook)
289 		vn_union_readdir_hook = union_readdirhook;
290 
291 	return (0);
292 
293 bad:
294 	if (um)
295 		free(um, M_UFSMNT);
296 	if (upperrootvp)
297 		vrele(upperrootvp);
298 	if (lowerrootvp)
299 		vrele(lowerrootvp);
300 	return (error);
301 }
302 
303 /*
304  * VFS start.  Nothing needed here - the start routine
305  * on the underlying filesystem(s) will have been called
306  * when that filesystem was mounted.
307  */
308  /*ARGSUSED*/
309 int
310 union_start(struct mount *mp, int flags)
311 {
312 
313 	return (0);
314 }
315 
316 /*
317  * Free reference to union layer
318  */
319 int
320 union_unmount(struct mount *mp, int mntflags)
321 {
322 	struct union_mount *um = MOUNTTOUNIONMOUNT(mp);
323 	int freeing;
324 	int error;
325 
326 #ifdef UNION_DIAGNOSTIC
327 	printf("union_unmount(mp = %p)\n", mp);
328 #endif
329 
330 	/*
331 	 * Keep flushing vnodes from the mount list.
332 	 * This is needed because of the un_pvp held
333 	 * reference to the parent vnode.
334 	 * If more vnodes have been freed on a given pass,
335 	 * the try again.  The loop will iterate at most
336 	 * (d) times, where (d) is the maximum tree depth
337 	 * in the filesystem.
338 	 */
339 	for (freeing = 0; (error = vflush(mp, NULL, 0)) != 0;) {
340 		struct vnode *vp;
341 		int n;
342 
343 		/* count #vnodes held on mount list */
344 		n = 0;
345 		TAILQ_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes)
346 			n++;
347 
348 		/* if this is unchanged then stop */
349 		if (n == freeing)
350 			break;
351 
352 		/* otherwise try once more time */
353 		freeing = n;
354 	}
355 
356 	/*
357 	 * Ok, now that we've tried doing it gently, get out the hammer.
358 	 */
359 
360 	if (mntflags & MNT_FORCE)
361 		error = vflush(mp, NULL, FORCECLOSE);
362 
363 	if (error)
364 		return error;
365 
366 	/*
367 	 * Discard references to upper and lower target vnodes.
368 	 */
369 	if (um->um_lowervp)
370 		vrele(um->um_lowervp);
371 	vrele(um->um_uppervp);
372 	kauth_cred_free(um->um_cred);
373 	/*
374 	 * Finally, throw away the union_mount structure
375 	 */
376 	free(mp->mnt_data, M_UFSMNT);	/* XXX */
377 	mp->mnt_data = NULL;
378 	return (0);
379 }
380 
381 int
382 union_root(struct mount *mp, struct vnode **vpp)
383 {
384 	struct union_mount *um = MOUNTTOUNIONMOUNT(mp);
385 	int error;
386 
387 	/*
388 	 * Return locked reference to root.
389 	 */
390 	vref(um->um_uppervp);
391 	vn_lock(um->um_uppervp, LK_EXCLUSIVE | LK_RETRY);
392 	if (um->um_lowervp)
393 		vref(um->um_lowervp);
394 	error = union_allocvp(vpp, mp, NULL, NULL, NULL,
395 			      um->um_uppervp, um->um_lowervp, 1);
396 
397 	if (error) {
398 		vput(um->um_uppervp);
399 		if (um->um_lowervp)
400 			vrele(um->um_lowervp);
401 	}
402 
403 	return (error);
404 }
405 
406 int
407 union_statvfs(struct mount *mp, struct statvfs *sbp)
408 {
409 	int error;
410 	struct union_mount *um = MOUNTTOUNIONMOUNT(mp);
411 	struct statvfs *sbuf = malloc(sizeof(*sbuf), M_TEMP, M_WAITOK | M_ZERO);
412 	unsigned long lbsize;
413 
414 #ifdef UNION_DIAGNOSTIC
415 	printf("union_statvfs(mp = %p, lvp = %p, uvp = %p)\n", mp,
416 	    um->um_lowervp, um->um_uppervp);
417 #endif
418 
419 	if (um->um_lowervp) {
420 		error = VFS_STATVFS(um->um_lowervp->v_mount, sbuf);
421 		if (error)
422 			goto done;
423 	}
424 
425 	/* now copy across the "interesting" information and fake the rest */
426 	lbsize = sbuf->f_bsize;
427 	sbp->f_blocks = sbuf->f_blocks - sbuf->f_bfree;
428 	sbp->f_files = sbuf->f_files - sbuf->f_ffree;
429 
430 	error = VFS_STATVFS(um->um_uppervp->v_mount, sbuf);
431 	if (error)
432 		goto done;
433 
434 	sbp->f_flag = sbuf->f_flag;
435 	sbp->f_bsize = sbuf->f_bsize;
436 	sbp->f_frsize = sbuf->f_frsize;
437 	sbp->f_iosize = sbuf->f_iosize;
438 
439 	/*
440 	 * The "total" fields count total resources in all layers,
441 	 * the "free" fields count only those resources which are
442 	 * free in the upper layer (since only the upper layer
443 	 * is writable).
444 	 */
445 
446 	if (sbuf->f_bsize != lbsize)
447 		sbp->f_blocks = sbp->f_blocks * lbsize / sbuf->f_bsize;
448 	sbp->f_blocks += sbuf->f_blocks;
449 	sbp->f_bfree = sbuf->f_bfree;
450 	sbp->f_bavail = sbuf->f_bavail;
451 	sbp->f_bresvd = sbuf->f_bresvd;
452 	sbp->f_files += sbuf->f_files;
453 	sbp->f_ffree = sbuf->f_ffree;
454 	sbp->f_favail = sbuf->f_favail;
455 	sbp->f_fresvd = sbuf->f_fresvd;
456 
457 	copy_statvfs_info(sbp, mp);
458 done:
459 	free(sbuf, M_TEMP);
460 	return error;
461 }
462 
463 /*ARGSUSED*/
464 int
465 union_sync(struct mount *mp, int waitfor,
466     kauth_cred_t cred)
467 {
468 
469 	/*
470 	 * XXX - Assumes no data cached at union layer.
471 	 */
472 	return (0);
473 }
474 
475 /*ARGSUSED*/
476 int
477 union_vget(struct mount *mp, ino_t ino,
478     struct vnode **vpp)
479 {
480 
481 	return (EOPNOTSUPP);
482 }
483 
484 static int
485 union_renamelock_enter(struct mount *mp)
486 {
487 	struct union_mount *um = MOUNTTOUNIONMOUNT(mp);
488 
489 	/* Lock just the upper fs, where the action happens. */
490 	return VFS_RENAMELOCK_ENTER(um->um_uppervp->v_mount);
491 }
492 
493 static void
494 union_renamelock_exit(struct mount *mp)
495 {
496 	struct union_mount *um = MOUNTTOUNIONMOUNT(mp);
497 
498 	VFS_RENAMELOCK_EXIT(um->um_uppervp->v_mount);
499 }
500 
501 extern const struct vnodeopv_desc union_vnodeop_opv_desc;
502 
503 const struct vnodeopv_desc * const union_vnodeopv_descs[] = {
504 	&union_vnodeop_opv_desc,
505 	NULL,
506 };
507 
508 struct vfsops union_vfsops = {
509 	MOUNT_UNION,
510 	sizeof (struct union_args),
511 	union_mount,
512 	union_start,
513 	union_unmount,
514 	union_root,
515 	(void *)eopnotsupp,		/* vfs_quotactl */
516 	union_statvfs,
517 	union_sync,
518 	union_vget,
519 	(void *)eopnotsupp,		/* vfs_fhtovp */
520 	(void *)eopnotsupp,		/* vfs_vptofh */
521 	union_init,
522 	NULL,				/* vfs_reinit */
523 	union_done,
524 	NULL,				/* vfs_mountroot */
525 	(int (*)(struct mount *, struct vnode *, struct timespec *)) eopnotsupp,
526 	vfs_stdextattrctl,
527 	(void *)eopnotsupp,		/* vfs_suspendctl */
528 	union_renamelock_enter,
529 	union_renamelock_exit,
530 	(void *)eopnotsupp,
531 	union_vnodeopv_descs,
532 	0,				/* vfs_refcount */
533 	{ NULL, NULL },
534 };
535 
536 static int
537 union_modcmd(modcmd_t cmd, void *arg)
538 {
539 	int error;
540 
541 	switch (cmd) {
542 	case MODULE_CMD_INIT:
543 		error = vfs_attach(&union_vfsops);
544 		if (error != 0)
545 			break;
546 		sysctl_createv(&union_sysctl_log, 0, NULL, NULL,
547 			       CTLFLAG_PERMANENT,
548 			       CTLTYPE_NODE, "vfs", NULL,
549 			       NULL, 0, NULL, 0,
550 			       CTL_VFS, CTL_EOL);
551 		sysctl_createv(&union_sysctl_log, 0, NULL, NULL,
552 			       CTLFLAG_PERMANENT,
553 			       CTLTYPE_NODE, "union",
554 			       SYSCTL_DESCR("Union file system"),
555 			       NULL, 0, NULL, 0,
556 			       CTL_VFS, 15, CTL_EOL);
557 		/*
558 		 * XXX the "15" above could be dynamic, thereby eliminating
559 		 * one more instance of the "number to vfs" mapping problem,
560 		 * but "15" is the order as taken from sys/mount.h
561 		 */
562 		break;
563 	case MODULE_CMD_FINI:
564 		error = vfs_detach(&union_vfsops);
565 		if (error != 0)
566 			break;
567 		sysctl_teardown(&union_sysctl_log);
568 		break;
569 	default:
570 		error = ENOTTY;
571 		break;
572 	}
573 
574 	return (error);
575 }
576