xref: /freebsd/sys/fs/unionfs/union_vfsops.c (revision f9218d3d4fd34f082473b3a021c6d4d109fb47cf)
1 /*
2  * Copyright (c) 1994, 1995 The Regents of the University of California.
3  * Copyright (c) 1994, 1995 Jan-Simon Pendry.
4  * All rights reserved.
5  *
6  * This code is derived from software donated to Berkeley by
7  * Jan-Simon Pendry.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  * 3. All advertising materials mentioning features or use of this software
18  *    must display the following acknowledgement:
19  *	This product includes software developed by the University of
20  *	California, Berkeley and its contributors.
21  * 4. Neither the name of the University nor the names of its contributors
22  *    may be used to endorse or promote products derived from this software
23  *    without specific prior written permission.
24  *
25  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
26  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
29  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35  * SUCH DAMAGE.
36  *
37  *	@(#)union_vfsops.c	8.20 (Berkeley) 5/20/95
38  * $FreeBSD$
39  */
40 
41 /*
42  * Union Layer
43  */
44 
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/kernel.h>
48 #include <sys/lock.h>
49 #include <sys/mutex.h>
50 #include <sys/proc.h>
51 #include <sys/vnode.h>
52 #include <sys/mount.h>
53 #include <sys/namei.h>
54 #include <sys/malloc.h>
55 #include <sys/filedesc.h>
56 #include <fs/unionfs/union.h>
57 
58 static MALLOC_DEFINE(M_UNIONFSMNT, "UNION mount", "UNION mount structure");
59 
60 extern int	union_init(struct vfsconf *);
61 static int	union_mount(struct mount *mp, struct nameidata *ndp,
62 				  struct thread *td);
63 static int	union_root(struct mount *mp, struct vnode **vpp);
64 static int	union_statfs(struct mount *mp, struct statfs *sbp,
65 				  struct thread *td);
66 static int	union_unmount(struct mount *mp, int mntflags,
67 				   struct thread *td);
68 
69 /*
70  * Mount union filesystem.
71  */
72 static int
73 union_mount(mp, ndp, td)
74 	struct mount *mp;
75 	struct nameidata *ndp;
76 	struct thread *td;
77 {
78 	int error = 0;
79 	struct vfsoptlist *opts;
80 	struct vnode *lowerrootvp = NULLVP;
81 	struct vnode *upperrootvp = NULLVP;
82 	struct union_mount *um = 0;
83 	struct ucred *cred = 0;
84 	char *cp = 0, *target;
85 	int op;
86 	int len;
87 	size_t size;
88 	struct componentname fakecn;
89 
90 	UDEBUG(("union_mount(mp = %p)\n", (void *)mp));
91 
92 	opts = mp->mnt_optnew;
93 	/*
94 	 * Disable clustered write, otherwise system becomes unstable.
95 	 */
96 	mp->mnt_flag |= MNT_NOCLUSTERW;
97 
98 	/*
99 	 * Update is a no-op
100 	 */
101 	if (mp->mnt_flag & MNT_UPDATE)
102 		/*
103 		 * Need to provide:
104 		 * 1. a way to convert between rdonly and rdwr mounts.
105 		 * 2. support for nfs exports.
106 		 */
107 		return (EOPNOTSUPP);
108 
109 	/*
110 	 * Get arguments.
111 	 */
112 	error = vfs_getopt(opts, "target", (void **)&target, &len);
113 	if (error || target[len - 1] != '\0')
114 		return (EINVAL);
115 
116 	op = 0;
117 	if (vfs_getopt(opts, "below", NULL, NULL) == 0)
118 		op = UNMNT_BELOW;
119 	if (vfs_getopt(opts, "replace", NULL, NULL) == 0) {
120 		/* These options are mutually exclusive. */
121 		if (op)
122 			return (EINVAL);
123 		op = UNMNT_REPLACE;
124 	}
125 	/*
126 	 * UNMNT_ABOVE is the default.
127 	 */
128 	if (op == 0)
129 		op = UNMNT_ABOVE;
130 
131 	/*
132 	 * Obtain lower vnode.  Vnode is stored in mp->mnt_vnodecovered.
133 	 * We need to reference it but not lock it.
134 	 */
135 
136 	lowerrootvp = mp->mnt_vnodecovered;
137 	VREF(lowerrootvp);
138 
139 #if 0
140 	/*
141 	 * Unlock lower node to avoid deadlock.
142 	 */
143 	if (lowerrootvp->v_op == union_vnodeop_p)
144 		VOP_UNLOCK(lowerrootvp, 0, td);
145 #endif
146 
147 	/*
148 	 * Obtain upper vnode by calling namei() on the path.  The
149 	 * upperrootvp will be turned referenced but not locked.
150 	 */
151 	NDINIT(ndp, LOOKUP, FOLLOW|WANTPARENT, UIO_SYSSPACE, target, td);
152 
153 	error = namei(ndp);
154 
155 #if 0
156 	if (lowerrootvp->v_op == union_vnodeop_p)
157 		vn_lock(lowerrootvp, LK_EXCLUSIVE | LK_RETRY, td);
158 #endif
159 	if (error)
160 		goto bad;
161 
162 	NDFREE(ndp, NDF_ONLY_PNBUF);
163 	upperrootvp = ndp->ni_vp;
164 	vrele(ndp->ni_dvp);
165 	ndp->ni_dvp = NULL;
166 
167 	UDEBUG(("mount_root UPPERVP %p locked = %d\n", upperrootvp,
168 	    VOP_ISLOCKED(upperrootvp, NULL)));
169 
170 	/*
171 	 * Check multi union mount to avoid `lock myself again' panic.
172 	 * Also require that it be a directory.
173 	 */
174 	if (upperrootvp == VTOUNION(lowerrootvp)->un_uppervp) {
175 #ifdef DIAGNOSTIC
176 		printf("union_mount: multi union mount?\n");
177 #endif
178 		error = EDEADLK;
179 		goto bad;
180 	}
181 
182 	if (upperrootvp->v_type != VDIR) {
183 		error = EINVAL;
184 		goto bad;
185 	}
186 
187 	/*
188 	 * Allocate our union_mount structure and populate the fields.
189 	 * The vnode references are stored in the union_mount as held,
190 	 * unlocked references.  Depending on the _BELOW flag, the
191 	 * filesystems are viewed in a different order.  In effect this
192 	 * is the same as providing a mount-under option to the mount
193 	 * syscall.
194 	 */
195 
196 	um = (struct union_mount *) malloc(sizeof(struct union_mount),
197 				M_UNIONFSMNT, M_WAITOK | M_ZERO);
198 
199 	um->um_op = op;
200 
201 	switch (um->um_op) {
202 	case UNMNT_ABOVE:
203 		um->um_lowervp = lowerrootvp;
204 		um->um_uppervp = upperrootvp;
205 		upperrootvp = NULL;
206 		lowerrootvp = NULL;
207 		break;
208 
209 	case UNMNT_BELOW:
210 		um->um_lowervp = upperrootvp;
211 		um->um_uppervp = lowerrootvp;
212 		upperrootvp = NULL;
213 		lowerrootvp = NULL;
214 		break;
215 
216 	case UNMNT_REPLACE:
217 		vrele(lowerrootvp);
218 		lowerrootvp = NULL;
219 		um->um_uppervp = upperrootvp;
220 		um->um_lowervp = lowerrootvp;
221 		upperrootvp = NULL;
222 		break;
223 
224 	default:
225 		error = EINVAL;
226 		goto bad;
227 	}
228 
229 	/*
230 	 * Unless the mount is readonly, ensure that the top layer
231 	 * supports whiteout operations.
232 	 */
233 	if ((mp->mnt_flag & MNT_RDONLY) == 0) {
234 		/*
235 		 * XXX Fake up a struct componentname with only cn_nameiop
236 		 * and cn_thread valid; union_whiteout() needs to use the
237 		 * thread pointer to lock the vnode.
238 		 */
239 		bzero(&fakecn, sizeof(fakecn));
240 		fakecn.cn_nameiop = LOOKUP;
241 		fakecn.cn_thread = td;
242 		error = VOP_WHITEOUT(um->um_uppervp, &fakecn, LOOKUP);
243 		if (error)
244 			goto bad;
245 	}
246 
247 	um->um_cred = crhold(td->td_ucred);
248 	FILEDESC_LOCK(td->td_proc->p_fd);
249 	um->um_cmode = UN_DIRMODE &~ td->td_proc->p_fd->fd_cmask;
250 	FILEDESC_UNLOCK(td->td_proc->p_fd);
251 
252 	/*
253 	 * Depending on what you think the MNT_LOCAL flag might mean,
254 	 * you may want the && to be || on the conditional below.
255 	 * At the moment it has been defined that the filesystem is
256 	 * only local if it is all local, ie the MNT_LOCAL flag implies
257 	 * that the entire namespace is local.  If you think the MNT_LOCAL
258 	 * flag implies that some of the files might be stored locally
259 	 * then you will want to change the conditional.
260 	 */
261 	if (um->um_op == UNMNT_ABOVE) {
262 		if (((um->um_lowervp == NULLVP) ||
263 		     (um->um_lowervp->v_mount->mnt_flag & MNT_LOCAL)) &&
264 		    (um->um_uppervp->v_mount->mnt_flag & MNT_LOCAL))
265 			mp->mnt_flag |= MNT_LOCAL;
266 	}
267 
268 	/*
269 	 * Copy in the upper layer's RDONLY flag.  This is for the benefit
270 	 * of lookup() which explicitly checks the flag, rather than asking
271 	 * the filesystem for its own opinion.  This means, that an update
272 	 * mount of the underlying filesystem to go from rdonly to rdwr
273 	 * will leave the unioned view as read-only.
274 	 */
275 	mp->mnt_flag |= (um->um_uppervp->v_mount->mnt_flag & MNT_RDONLY);
276 
277 	mp->mnt_data = (qaddr_t) um;
278 	vfs_getnewfsid(mp);
279 
280 	switch (um->um_op) {
281 	case UNMNT_ABOVE:
282 		cp = "<above>:";
283 		break;
284 	case UNMNT_BELOW:
285 		cp = "<below>:";
286 		break;
287 	case UNMNT_REPLACE:
288 		cp = "";
289 		break;
290 	}
291 	len = strlen(cp);
292 	bcopy(cp, mp->mnt_stat.f_mntfromname, len);
293 
294 	cp = mp->mnt_stat.f_mntfromname + len;
295 	len = MNAMELEN - len;
296 
297 	(void) copystr(target, cp, len - 1, &size);
298 	bzero(cp + size, len - size);
299 
300 	(void)union_statfs(mp, &mp->mnt_stat, td);
301 
302 	UDEBUG(("union_mount: from %s, on %s\n",
303 		mp->mnt_stat.f_mntfromname, mp->mnt_stat.f_mntonname));
304 	return (0);
305 
306 bad:
307 	if (um) {
308 		if (um->um_uppervp)
309 			vrele(um->um_uppervp);
310 		if (um->um_lowervp)
311 			vrele(um->um_lowervp);
312 		/* XXX other fields */
313 		free(um, M_UNIONFSMNT);
314 	}
315 	if (cred)
316 		crfree(cred);
317 	if (upperrootvp)
318 		vrele(upperrootvp);
319 	if (lowerrootvp)
320 		vrele(lowerrootvp);
321 	return (error);
322 }
323 
324 /*
325  * Free reference to union layer.
326  */
327 static int
328 union_unmount(mp, mntflags, td)
329 	struct mount *mp;
330 	int mntflags;
331 	struct thread *td;
332 {
333 	struct union_mount *um = MOUNTTOUNIONMOUNT(mp);
334 	int error;
335 	int freeing;
336 	int flags = 0;
337 
338 	UDEBUG(("union_unmount(mp = %p)\n", (void *)mp));
339 
340 	if (mntflags & MNT_FORCE)
341 		flags |= FORCECLOSE;
342 
343 	/*
344 	 * Keep flushing vnodes from the mount list.
345 	 * This is needed because of the un_pvp held
346 	 * reference to the parent vnode.
347 	 * If more vnodes have been freed on a given pass,
348 	 * the try again.  The loop will iterate at most
349 	 * (d) times, where (d) is the maximum tree depth
350 	 * in the filesystem.
351 	 */
352 	for (freeing = 0; (error = vflush(mp, 0, flags)) != 0;) {
353 		struct vnode *vp;
354 		int n;
355 
356 		/* count #vnodes held on mount list */
357 		mtx_lock(&mntvnode_mtx);
358 		n = 0;
359 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes)
360 			n++;
361 		mtx_unlock(&mntvnode_mtx);
362 
363 		/* if this is unchanged then stop */
364 		if (n == freeing)
365 			break;
366 
367 		/* otherwise try once more time */
368 		freeing = n;
369 	}
370 
371 	/*
372 	 * If the most recent vflush failed, the filesystem is still busy.
373 	 */
374 	if (error)
375 		return (error);
376 
377 	/*
378 	 * Discard references to upper and lower target vnodes.
379 	 */
380 	if (um->um_lowervp)
381 		vrele(um->um_lowervp);
382 	vrele(um->um_uppervp);
383 	crfree(um->um_cred);
384 	/*
385 	 * Finally, throw away the union_mount structure.
386 	 */
387 	free(mp->mnt_data, M_UNIONFSMNT);	/* XXX */
388 	mp->mnt_data = 0;
389 	return (0);
390 }
391 
392 static int
393 union_root(mp, vpp)
394 	struct mount *mp;
395 	struct vnode **vpp;
396 {
397 	struct union_mount *um = MOUNTTOUNIONMOUNT(mp);
398 	int error;
399 
400 	/*
401 	 * Supply an unlocked reference to um_uppervp and to um_lowervp.  It
402 	 * is possible for um_uppervp to be locked without the associated
403 	 * root union_node being locked.  We let union_allocvp() deal with
404 	 * it.
405 	 */
406 	UDEBUG(("union_root UPPERVP %p locked = %d\n", um->um_uppervp,
407 	    VOP_ISLOCKED(um->um_uppervp, NULL)));
408 
409 	VREF(um->um_uppervp);
410 	if (um->um_lowervp)
411 		VREF(um->um_lowervp);
412 
413 	error = union_allocvp(vpp, mp, NULLVP, NULLVP, NULL,
414 		    um->um_uppervp, um->um_lowervp, 1);
415 	UDEBUG(("error %d\n", error));
416 	UDEBUG(("union_root2 UPPERVP %p locked = %d\n", um->um_uppervp,
417 	    VOP_ISLOCKED(um->um_uppervp, NULL)));
418 
419 	return (error);
420 }
421 
422 static int
423 union_statfs(mp, sbp, td)
424 	struct mount *mp;
425 	struct statfs *sbp;
426 	struct thread *td;
427 {
428 	int error;
429 	struct union_mount *um = MOUNTTOUNIONMOUNT(mp);
430 	struct statfs mstat;
431 	int lbsize;
432 
433 	UDEBUG(("union_statfs(mp = %p, lvp = %p, uvp = %p)\n",
434 	    (void *)mp, (void *)um->um_lowervp, (void *)um->um_uppervp));
435 
436 	bzero(&mstat, sizeof(mstat));
437 
438 	if (um->um_lowervp) {
439 		error = VFS_STATFS(um->um_lowervp->v_mount, &mstat, td);
440 		if (error)
441 			return (error);
442 	}
443 
444 	/*
445 	 * Now copy across the "interesting" information and fake the rest.
446 	 */
447 #if 0
448 	sbp->f_type = mstat.f_type;
449 	sbp->f_flags = mstat.f_flags;
450 	sbp->f_bsize = mstat.f_bsize;
451 	sbp->f_iosize = mstat.f_iosize;
452 #endif
453 	lbsize = mstat.f_bsize;
454 	sbp->f_blocks = mstat.f_blocks;
455 	sbp->f_bfree = mstat.f_bfree;
456 	sbp->f_bavail = mstat.f_bavail;
457 	sbp->f_files = mstat.f_files;
458 	sbp->f_ffree = mstat.f_ffree;
459 
460 	error = VFS_STATFS(um->um_uppervp->v_mount, &mstat, td);
461 	if (error)
462 		return (error);
463 
464 	sbp->f_flags = mstat.f_flags;
465 	sbp->f_bsize = mstat.f_bsize;
466 	sbp->f_iosize = mstat.f_iosize;
467 
468 	/*
469 	 * If the lower and upper blocksizes differ, then frig the
470 	 * block counts so that the sizes reported by df make some
471 	 * kind of sense.  None of this makes sense though.
472 	 */
473 
474 	if (mstat.f_bsize != lbsize)
475 		sbp->f_blocks = ((off_t) sbp->f_blocks * lbsize) / mstat.f_bsize;
476 
477 	/*
478 	 * The "total" fields count total resources in all layers,
479 	 * the "free" fields count only those resources which are
480 	 * free in the upper layer (since only the upper layer
481 	 * is writeable).
482 	 */
483 	sbp->f_blocks += mstat.f_blocks;
484 	sbp->f_bfree = mstat.f_bfree;
485 	sbp->f_bavail = mstat.f_bavail;
486 	sbp->f_files += mstat.f_files;
487 	sbp->f_ffree = mstat.f_ffree;
488 
489 	if (sbp != &mp->mnt_stat) {
490 		sbp->f_type = mp->mnt_vfc->vfc_typenum;
491 		bcopy(&mp->mnt_stat.f_fsid, &sbp->f_fsid, sizeof(sbp->f_fsid));
492 		bcopy(mp->mnt_stat.f_mntonname, sbp->f_mntonname, MNAMELEN);
493 		bcopy(mp->mnt_stat.f_mntfromname, sbp->f_mntfromname, MNAMELEN);
494 	}
495 	return (0);
496 }
497 
498 static struct vfsops union_vfsops = {
499 	NULL,
500 	vfs_stdstart,	/* underlying start already done */
501 	union_unmount,
502 	union_root,
503 	vfs_stdquotactl,
504 	union_statfs,
505 	vfs_stdsync,    /* XXX assumes no cached data on union level */
506 	vfs_stdvget,
507 	vfs_stdfhtovp,
508 	vfs_stdcheckexp,
509 	vfs_stdvptofh,
510 	union_init,
511 	vfs_stduninit,
512 	vfs_stdextattrctl,
513 	union_mount,
514 };
515 
516 VFS_SET(union_vfsops, unionfs, VFCF_LOOPBACK);
517