xref: /netbsd-src/sys/fs/union/union_vfsops.c (revision bdc22b2e01993381dcefeff2bc9b56ca75a4235c)
1 /*	$NetBSD: union_vfsops.c,v 1.78 2017/04/01 19:35:56 riastradh 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 1.78 2017/04/01 19:35:56 riastradh 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 <miscfs/genfs/genfs.h>
98 #include <fs/union/union.h>
99 
100 MODULE(MODULE_CLASS_VFS, union, NULL);
101 
102 static struct sysctllog *union_sysctl_log;
103 
104 /*
105  * Mount union filesystem
106  */
107 int
108 union_mount(struct mount *mp, const char *path, void *data, size_t *data_len)
109 {
110 	struct lwp *l = curlwp;
111 	int error = 0;
112 	struct union_args *args = data;
113 	struct vnode *lowerrootvp = NULLVP;
114 	struct vnode *upperrootvp = NULLVP;
115 	struct union_mount *um = 0;
116 	const char *cp;
117 	char *xp;
118 	int len;
119 	size_t size;
120 
121 	if (args == NULL)
122 		return EINVAL;
123 	if (*data_len < sizeof *args)
124 		return EINVAL;
125 
126 #ifdef UNION_DIAGNOSTIC
127 	printf("union_mount(mp = %p)\n", mp);
128 #endif
129 
130 	if (mp->mnt_flag & MNT_GETARGS) {
131 		um = MOUNTTOUNIONMOUNT(mp);
132 		if (um == NULL)
133 			return EIO;
134 		args->target = NULL;
135 		args->mntflags = um->um_op;
136 		*data_len = sizeof *args;
137 		return 0;
138 	}
139 	/*
140 	 * Update is a no-op
141 	 */
142 	if (mp->mnt_flag & MNT_UPDATE) {
143 		/*
144 		 * Need to provide.
145 		 * 1. a way to convert between rdonly and rdwr mounts.
146 		 * 2. support for nfs exports.
147 		 */
148 		error = EOPNOTSUPP;
149 		goto bad;
150 	}
151 
152 	lowerrootvp = mp->mnt_vnodecovered;
153 	vref(lowerrootvp);
154 
155 	/*
156 	 * Find upper node.
157 	 */
158 	error = namei_simple_user(args->target,
159 				NSM_FOLLOW_NOEMULROOT, &upperrootvp);
160 	if (error != 0)
161 		goto bad;
162 
163 	if (upperrootvp->v_type != VDIR) {
164 		error = EINVAL;
165 		goto bad;
166 	}
167 
168 	um = kmem_zalloc(sizeof(struct union_mount), KM_SLEEP);
169 
170 	/*
171 	 * Keep a held reference to the target vnodes.
172 	 * They are vrele'd in union_unmount.
173 	 *
174 	 * Depending on the _BELOW flag, the filesystems are
175 	 * viewed in a different order.  In effect, this is the
176 	 * same as providing a mount under option to the mount syscall.
177 	 */
178 
179 	um->um_op = args->mntflags & UNMNT_OPMASK;
180 	switch (um->um_op) {
181 	case UNMNT_ABOVE:
182 		um->um_lowervp = lowerrootvp;
183 		um->um_uppervp = upperrootvp;
184 		break;
185 
186 	case UNMNT_BELOW:
187 		um->um_lowervp = upperrootvp;
188 		um->um_uppervp = lowerrootvp;
189 		break;
190 
191 	case UNMNT_REPLACE:
192 		vrele(lowerrootvp);
193 		lowerrootvp = NULLVP;
194 		um->um_uppervp = upperrootvp;
195 		um->um_lowervp = lowerrootvp;
196 		break;
197 
198 	default:
199 		error = EINVAL;
200 		goto bad;
201 	}
202 
203 	mp->mnt_iflag |= IMNT_MPSAFE;
204 
205 	/*
206 	 * Unless the mount is readonly, ensure that the top layer
207 	 * supports whiteout operations
208 	 */
209 	if ((mp->mnt_flag & MNT_RDONLY) == 0) {
210 		vn_lock(um->um_uppervp, LK_EXCLUSIVE | LK_RETRY);
211 		error = VOP_WHITEOUT(um->um_uppervp,
212 		    (struct componentname *) 0, LOOKUP);
213 		VOP_UNLOCK(um->um_uppervp);
214 		if (error)
215 			goto bad;
216 	}
217 
218 	um->um_cred = l->l_cred;
219 	kauth_cred_hold(um->um_cred);
220 	um->um_cmode = UN_DIRMODE &~ l->l_proc->p_cwdi->cwdi_cmask;
221 
222 	/*
223 	 * Depending on what you think the MNT_LOCAL flag might mean,
224 	 * you may want the && to be || on the conditional below.
225 	 * At the moment it has been defined that the filesystem is
226 	 * only local if it is all local, ie the MNT_LOCAL flag implies
227 	 * that the entire namespace is local.  If you think the MNT_LOCAL
228 	 * flag implies that some of the files might be stored locally
229 	 * then you will want to change the conditional.
230 	 */
231 	if (um->um_op == UNMNT_ABOVE) {
232 		if (((um->um_lowervp == NULLVP) ||
233 		     (um->um_lowervp->v_mount->mnt_flag & MNT_LOCAL)) &&
234 		    (um->um_uppervp->v_mount->mnt_flag & MNT_LOCAL))
235 			mp->mnt_flag |= MNT_LOCAL;
236 	}
237 
238 	/*
239 	 * Copy in the upper layer's RDONLY flag.  This is for the benefit
240 	 * of lookup() which explicitly checks the flag, rather than asking
241 	 * the filesystem for its own opinion.  This means, that an update
242 	 * mount of the underlying filesystem to go from rdonly to rdwr
243 	 * will leave the unioned view as read-only.
244 	 */
245 	mp->mnt_flag |= (um->um_uppervp->v_mount->mnt_flag & MNT_RDONLY);
246 
247 	mp->mnt_data = um;
248 	vfs_getnewfsid(mp);
249 
250 	error = set_statvfs_info( path, UIO_USERSPACE, NULL, UIO_USERSPACE,
251 	    mp->mnt_op->vfs_name, mp, l);
252 	if (error)
253 		goto bad;
254 
255 	mp->mnt_lower = um->um_uppervp->v_mount;
256 
257 	switch (um->um_op) {
258 	case UNMNT_ABOVE:
259 		cp = "<above>:";
260 		break;
261 	case UNMNT_BELOW:
262 		cp = "<below>:";
263 		break;
264 	case UNMNT_REPLACE:
265 		cp = "";
266 		break;
267 	default:
268 		cp = "<invalid>:";
269 #ifdef DIAGNOSTIC
270 		panic("union_mount: bad um_op");
271 #endif
272 		break;
273 	}
274 	len = strlen(cp);
275 	memcpy(mp->mnt_stat.f_mntfromname, cp, len);
276 
277 	xp = mp->mnt_stat.f_mntfromname + len;
278 	len = MNAMELEN - len;
279 
280 	(void) copyinstr(args->target, xp, len - 1, &size);
281 	memset(xp + size, 0, len - size);
282 
283 #ifdef UNION_DIAGNOSTIC
284 	printf("union_mount: from %s, on %s\n",
285 	    mp->mnt_stat.f_mntfromname, mp->mnt_stat.f_mntonname);
286 #endif
287 
288 	/* Setup the readdir hook if it's not set already */
289 	if (!vn_union_readdir_hook)
290 		vn_union_readdir_hook = union_readdirhook;
291 
292 	return (0);
293 
294 bad:
295 	if (um)
296 		kmem_free(um, sizeof(struct union_mount));
297 	if (upperrootvp)
298 		vrele(upperrootvp);
299 	if (lowerrootvp)
300 		vrele(lowerrootvp);
301 	return (error);
302 }
303 
304 /*
305  * VFS start.  Nothing needed here - the start routine
306  * on the underlying filesystem(s) will have been called
307  * when that filesystem was mounted.
308  */
309  /*ARGSUSED*/
310 int
311 union_start(struct mount *mp, int flags)
312 {
313 
314 	return (0);
315 }
316 
317 /*
318  * Free reference to union layer
319  */
320 static bool
321 union_unmount_selector(void *cl, struct vnode *vp)
322 {
323 	int *count = cl;
324 
325 	KASSERT(mutex_owned(vp->v_interlock));
326 
327 	*count += 1;
328 	return false;
329 }
330 
331 int
332 union_unmount(struct mount *mp, int mntflags)
333 {
334 	struct union_mount *um = MOUNTTOUNIONMOUNT(mp);
335 	int freeing;
336 	int error;
337 
338 #ifdef UNION_DIAGNOSTIC
339 	printf("union_unmount(mp = %p)\n", mp);
340 #endif
341 
342 	/*
343 	 * Keep flushing vnodes from the mount list.
344 	 * This is needed because of the un_pvp held
345 	 * reference to the parent vnode.
346 	 * If more vnodes have been freed on a given pass,
347 	 * the try again.  The loop will iterate at most
348 	 * (d) times, where (d) is the maximum tree depth
349 	 * in the filesystem.
350 	 */
351 	for (freeing = 0; (error = vflush(mp, NULL, 0)) != 0;) {
352 		struct vnode_iterator *marker;
353 		int n;
354 
355 		/* count #vnodes held on mount list */
356 		n = 0;
357 		vfs_vnode_iterator_init(mp, &marker);
358 		vfs_vnode_iterator_next(marker, union_unmount_selector, &n);
359 		vfs_vnode_iterator_destroy(marker);
360 
361 		/* if this is unchanged then stop */
362 		if (n == freeing)
363 			break;
364 
365 		/* otherwise try once more time */
366 		freeing = n;
367 	}
368 
369 	/*
370 	 * Ok, now that we've tried doing it gently, get out the hammer.
371 	 */
372 
373 	if (mntflags & MNT_FORCE)
374 		error = vflush(mp, NULL, FORCECLOSE);
375 
376 	if (error)
377 		return error;
378 
379 	/*
380 	 * Discard references to upper and lower target vnodes.
381 	 */
382 	if (um->um_lowervp)
383 		vrele(um->um_lowervp);
384 	vrele(um->um_uppervp);
385 	kauth_cred_free(um->um_cred);
386 	/*
387 	 * Finally, throw away the union_mount structure
388 	 */
389 	kmem_free(um, sizeof(struct union_mount));
390 	mp->mnt_data = NULL;
391 	return 0;
392 }
393 
394 int
395 union_root(struct mount *mp, struct vnode **vpp)
396 {
397 	struct union_mount *um = MOUNTTOUNIONMOUNT(mp);
398 	int error;
399 
400 	/*
401 	 * Return locked reference to root.
402 	 */
403 	vref(um->um_uppervp);
404 	if (um->um_lowervp)
405 		vref(um->um_lowervp);
406 	error = union_allocvp(vpp, mp, NULL, NULL, NULL,
407 			      um->um_uppervp, um->um_lowervp, 1);
408 
409 	if (error) {
410 		vrele(um->um_uppervp);
411 		if (um->um_lowervp)
412 			vrele(um->um_lowervp);
413 		return error;
414 	}
415 
416 	vn_lock(*vpp, LK_EXCLUSIVE | LK_RETRY);
417 
418 	return 0;
419 }
420 
421 int
422 union_statvfs(struct mount *mp, struct statvfs *sbp)
423 {
424 	int error;
425 	struct union_mount *um = MOUNTTOUNIONMOUNT(mp);
426 	struct statvfs *sbuf = malloc(sizeof(*sbuf), M_TEMP, M_WAITOK | M_ZERO);
427 	unsigned long lbsize;
428 
429 #ifdef UNION_DIAGNOSTIC
430 	printf("union_statvfs(mp = %p, lvp = %p, uvp = %p)\n", mp,
431 	    um->um_lowervp, um->um_uppervp);
432 #endif
433 
434 	if (um->um_lowervp) {
435 		error = VFS_STATVFS(um->um_lowervp->v_mount, sbuf);
436 		if (error)
437 			goto done;
438 	}
439 
440 	/* now copy across the "interesting" information and fake the rest */
441 	lbsize = sbuf->f_bsize;
442 	sbp->f_blocks = sbuf->f_blocks - sbuf->f_bfree;
443 	sbp->f_files = sbuf->f_files - sbuf->f_ffree;
444 
445 	error = VFS_STATVFS(um->um_uppervp->v_mount, sbuf);
446 	if (error)
447 		goto done;
448 
449 	sbp->f_flag = sbuf->f_flag;
450 	sbp->f_bsize = sbuf->f_bsize;
451 	sbp->f_frsize = sbuf->f_frsize;
452 	sbp->f_iosize = sbuf->f_iosize;
453 
454 	/*
455 	 * The "total" fields count total resources in all layers,
456 	 * the "free" fields count only those resources which are
457 	 * free in the upper layer (since only the upper layer
458 	 * is writable).
459 	 */
460 
461 	if (sbuf->f_bsize != lbsize)
462 		sbp->f_blocks = sbp->f_blocks * lbsize / sbuf->f_bsize;
463 	sbp->f_blocks += sbuf->f_blocks;
464 	sbp->f_bfree = sbuf->f_bfree;
465 	sbp->f_bavail = sbuf->f_bavail;
466 	sbp->f_bresvd = sbuf->f_bresvd;
467 	sbp->f_files += sbuf->f_files;
468 	sbp->f_ffree = sbuf->f_ffree;
469 	sbp->f_favail = sbuf->f_favail;
470 	sbp->f_fresvd = sbuf->f_fresvd;
471 
472 	copy_statvfs_info(sbp, mp);
473 done:
474 	free(sbuf, M_TEMP);
475 	return error;
476 }
477 
478 /*ARGSUSED*/
479 int
480 union_sync(struct mount *mp, int waitfor,
481     kauth_cred_t cred)
482 {
483 
484 	/*
485 	 * XXX - Assumes no data cached at union layer.
486 	 */
487 	return (0);
488 }
489 
490 /*ARGSUSED*/
491 int
492 union_vget(struct mount *mp, ino_t ino,
493     struct vnode **vpp)
494 {
495 
496 	return (EOPNOTSUPP);
497 }
498 
499 static int
500 union_renamelock_enter(struct mount *mp)
501 {
502 	struct union_mount *um = MOUNTTOUNIONMOUNT(mp);
503 
504 	/* Lock just the upper fs, where the action happens. */
505 	return VFS_RENAMELOCK_ENTER(um->um_uppervp->v_mount);
506 }
507 
508 static void
509 union_renamelock_exit(struct mount *mp)
510 {
511 	struct union_mount *um = MOUNTTOUNIONMOUNT(mp);
512 
513 	VFS_RENAMELOCK_EXIT(um->um_uppervp->v_mount);
514 }
515 
516 extern const struct vnodeopv_desc union_vnodeop_opv_desc;
517 
518 const struct vnodeopv_desc * const union_vnodeopv_descs[] = {
519 	&union_vnodeop_opv_desc,
520 	NULL,
521 };
522 
523 struct vfsops union_vfsops = {
524 	.vfs_name = MOUNT_UNION,
525 	.vfs_min_mount_data = sizeof (struct union_args),
526 	.vfs_mount = union_mount,
527 	.vfs_start = union_start,
528 	.vfs_unmount = union_unmount,
529 	.vfs_root = union_root,
530 	.vfs_quotactl = (void *)eopnotsupp,
531 	.vfs_statvfs = union_statvfs,
532 	.vfs_sync = union_sync,
533 	.vfs_vget = union_vget,
534 	.vfs_loadvnode = union_loadvnode,
535 	.vfs_fhtovp = (void *)eopnotsupp,
536 	.vfs_vptofh = (void *)eopnotsupp,
537 	.vfs_init = union_init,
538 	.vfs_reinit = union_reinit,
539 	.vfs_done = union_done,
540 	.vfs_snapshot = (void *)eopnotsupp,
541 	.vfs_extattrctl = vfs_stdextattrctl,
542 	.vfs_suspendctl = genfs_suspendctl,
543 	.vfs_renamelock_enter = union_renamelock_enter,
544 	.vfs_renamelock_exit = union_renamelock_exit,
545 	.vfs_fsync = (void *)eopnotsupp,
546 	.vfs_opv_descs = union_vnodeopv_descs
547 };
548 
549 static int
550 union_modcmd(modcmd_t cmd, void *arg)
551 {
552 	int error;
553 
554 	switch (cmd) {
555 	case MODULE_CMD_INIT:
556 		error = vfs_attach(&union_vfsops);
557 		if (error != 0)
558 			break;
559 		sysctl_createv(&union_sysctl_log, 0, NULL, NULL,
560 			       CTLFLAG_PERMANENT,
561 			       CTLTYPE_NODE, "union",
562 			       SYSCTL_DESCR("Union file system"),
563 			       NULL, 0, NULL, 0,
564 			       CTL_VFS, 15, CTL_EOL);
565 		/*
566 		 * XXX the "15" above could be dynamic, thereby eliminating
567 		 * one more instance of the "number to vfs" mapping problem,
568 		 * but "15" is the order as taken from sys/mount.h
569 		 */
570 		break;
571 	case MODULE_CMD_FINI:
572 		error = vfs_detach(&union_vfsops);
573 		if (error != 0)
574 			break;
575 		sysctl_teardown(&union_sysctl_log);
576 		break;
577 	default:
578 		error = ENOTTY;
579 		break;
580 	}
581 
582 	return (error);
583 }
584