xref: /freebsd/sys/fs/unionfs/union_vfsops.c (revision 5521ff5a4d1929056e7ffc982fac3341ca54df7c)
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 __P((struct vfsconf *));
61 static int	union_mount __P((struct mount *mp, char *path, caddr_t data,
62 				 struct nameidata *ndp, struct proc *p));
63 static int	union_root __P((struct mount *mp, struct vnode **vpp));
64 static int	union_statfs __P((struct mount *mp, struct statfs *sbp,
65 				  struct proc *p));
66 static int	union_unmount __P((struct mount *mp, int mntflags,
67 				   struct proc *p));
68 
69 /*
70  * Mount union filesystem
71  */
72 static int
73 union_mount(mp, path, data, ndp, p)
74 	struct mount *mp;
75 	char *path;
76 	caddr_t data;
77 	struct nameidata *ndp;
78 	struct proc *p;
79 {
80 	int error = 0;
81 	struct union_args args;
82 	struct vnode *lowerrootvp = NULLVP;
83 	struct vnode *upperrootvp = NULLVP;
84 	struct union_mount *um = 0;
85 	struct ucred *cred = 0;
86 	char *cp = 0;
87 	int len;
88 	u_int size;
89 
90 	UDEBUG(("union_mount(mp = %p)\n", (void *)mp));
91 
92 	/*
93 	 * Disable clustered write, otherwise system becomes unstable.
94 	 */
95 	mp->mnt_flag |= MNT_NOCLUSTERW;
96 
97 	/*
98 	 * Update is a no-op
99 	 */
100 	if (mp->mnt_flag & MNT_UPDATE) {
101 		/*
102 		 * Need to provide.
103 		 * 1. a way to convert between rdonly and rdwr mounts.
104 		 * 2. support for nfs exports.
105 		 */
106 		error = EOPNOTSUPP;
107 		goto bad;
108 	}
109 
110 	/*
111 	 * Get argument
112 	 */
113 	error = copyin(data, (caddr_t)&args, sizeof(struct union_args));
114 	if (error)
115 		goto bad;
116 
117 	/*
118 	 * Obtain lower vnode.  Vnode is stored in mp->mnt_vnodecovered.
119 	 * We need to reference it but not lock it.
120 	 */
121 
122 	lowerrootvp = mp->mnt_vnodecovered;
123 	VREF(lowerrootvp);
124 
125 #if 0
126 	/*
127 	 * Unlock lower node to avoid deadlock.
128 	 */
129 	if (lowerrootvp->v_op == union_vnodeop_p)
130 		VOP_UNLOCK(lowerrootvp, 0, p);
131 #endif
132 
133 	/*
134 	 * Obtain upper vnode by calling namei() on the path.  The
135 	 * upperrootvp will be turned referenced but not locked.
136 	 */
137 	NDINIT(ndp, LOOKUP, FOLLOW|WANTPARENT,
138 	       UIO_USERSPACE, args.target, p);
139 
140 	error = namei(ndp);
141 
142 #if 0
143 	if (lowerrootvp->v_op == union_vnodeop_p)
144 		vn_lock(lowerrootvp, LK_EXCLUSIVE | LK_RETRY, p);
145 #endif
146 	if (error)
147 		goto bad;
148 
149 	NDFREE(ndp, NDF_ONLY_PNBUF);
150 	upperrootvp = ndp->ni_vp;
151 	vrele(ndp->ni_dvp);
152 	ndp->ni_dvp = NULL;
153 
154 	UDEBUG(("mount_root UPPERVP %p locked = %d\n", upperrootvp,
155 	    VOP_ISLOCKED(upperrootvp, NULL)));
156 
157 	/*
158 	 * Check multi union mount to avoid `lock myself again' panic.
159 	 * Also require that it be a directory.
160 	 */
161 	if (upperrootvp == VTOUNION(lowerrootvp)->un_uppervp) {
162 #ifdef DIAGNOSTIC
163 		printf("union_mount: multi union mount?\n");
164 #endif
165 		error = EDEADLK;
166 		goto bad;
167 	}
168 
169 	if (upperrootvp->v_type != VDIR) {
170 		error = EINVAL;
171 		goto bad;
172 	}
173 
174 	/*
175 	 * Allocate our union_mount structure and populate the fields.
176 	 * The vnode references are stored in the union_mount as held,
177 	 * unlocked references.  Depending on the _BELOW flag, the
178 	 * filesystems are viewed in a different order.  In effect this
179 	 * is the same as providing a mount-under option to the mount
180 	 * syscall.
181 	 */
182 
183 	um = (struct union_mount *) malloc(sizeof(struct union_mount),
184 				M_UNIONFSMNT, M_WAITOK | M_ZERO);
185 
186 	um->um_op = args.mntflags & UNMNT_OPMASK;
187 
188 	switch (um->um_op) {
189 	case UNMNT_ABOVE:
190 		um->um_lowervp = lowerrootvp;
191 		um->um_uppervp = upperrootvp;
192 		upperrootvp = NULL;
193 		lowerrootvp = NULL;
194 		break;
195 
196 	case UNMNT_BELOW:
197 		um->um_lowervp = upperrootvp;
198 		um->um_uppervp = lowerrootvp;
199 		upperrootvp = NULL;
200 		lowerrootvp = NULL;
201 		break;
202 
203 	case UNMNT_REPLACE:
204 		vrele(lowerrootvp);
205 		lowerrootvp = NULL;
206 		um->um_uppervp = upperrootvp;
207 		um->um_lowervp = lowerrootvp;
208 		upperrootvp = NULL;
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, NULL, LOOKUP);
222 		if (error)
223 			goto bad;
224 	}
225 
226 	um->um_cred = p->p_ucred;
227 	crhold(um->um_cred);
228 	um->um_cmode = UN_DIRMODE &~ p->p_fd->fd_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 its 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 = (qaddr_t) um;
256 	vfs_getnewfsid(mp);
257 
258 	switch (um->um_op) {
259 	case UNMNT_ABOVE:
260 		cp = "<above>:";
261 		break;
262 	case UNMNT_BELOW:
263 		cp = "<below>:";
264 		break;
265 	case UNMNT_REPLACE:
266 		cp = "";
267 		break;
268 	}
269 	len = strlen(cp);
270 	bcopy(cp, mp->mnt_stat.f_mntfromname, len);
271 
272 	cp = mp->mnt_stat.f_mntfromname + len;
273 	len = MNAMELEN - len;
274 
275 	(void) copyinstr(args.target, cp, len - 1, &size);
276 	bzero(cp + size, len - size);
277 
278 	(void)union_statfs(mp, &mp->mnt_stat, p);
279 
280 	UDEBUG(("union_mount: from %s, on %s\n",
281 		mp->mnt_stat.f_mntfromname, mp->mnt_stat.f_mntonname));
282 	return (0);
283 
284 bad:
285 	if (um) {
286 		if (um->um_uppervp)
287 			vrele(um->um_uppervp);
288 		if (um->um_lowervp)
289 			vrele(um->um_lowervp);
290 		/* XXX other fields */
291 		free(um, M_UNIONFSMNT);
292 	}
293 	if (cred)
294 		crfree(cred);
295 	if (upperrootvp)
296 		vrele(upperrootvp);
297 	if (lowerrootvp)
298 		vrele(lowerrootvp);
299 	return (error);
300 }
301 
302 /*
303  * Free reference to union layer
304  */
305 static int
306 union_unmount(mp, mntflags, p)
307 	struct mount *mp;
308 	int mntflags;
309 	struct proc *p;
310 {
311 	struct union_mount *um = MOUNTTOUNIONMOUNT(mp);
312 	int error;
313 	int freeing;
314 	int flags = 0;
315 
316 	UDEBUG(("union_unmount(mp = %p)\n", (void *)mp));
317 
318 	if (mntflags & MNT_FORCE)
319 		flags |= FORCECLOSE;
320 
321 	/*
322 	 * Keep flushing vnodes from the mount list.
323 	 * This is needed because of the un_pvp held
324 	 * reference to the parent vnode.
325 	 * If more vnodes have been freed on a given pass,
326 	 * the try again.  The loop will iterate at most
327 	 * (d) times, where (d) is the maximum tree depth
328 	 * in the filesystem.
329 	 */
330 	for (freeing = 0; (error = vflush(mp, 0, flags)) != 0;) {
331 		struct vnode *vp;
332 		int n;
333 
334 		/* count #vnodes held on mount list */
335 		mtx_lock(&mntvnode_mtx);
336 		n = 0;
337 		LIST_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes)
338 			n++;
339 		mtx_unlock(&mntvnode_mtx);
340 
341 		/* if this is unchanged then stop */
342 		if (n == freeing)
343 			break;
344 
345 		/* otherwise try once more time */
346 		freeing = n;
347 	}
348 
349 	/* If the most recent vflush failed, the filesystem is still busy. */
350 	if (error)
351 		return (error);
352 
353 	/*
354 	 * Discard references to upper and lower target vnodes.
355 	 */
356 	if (um->um_lowervp)
357 		vrele(um->um_lowervp);
358 	vrele(um->um_uppervp);
359 	crfree(um->um_cred);
360 	/*
361 	 * Finally, throw away the union_mount structure
362 	 */
363 	free(mp->mnt_data, M_UNIONFSMNT);	/* XXX */
364 	mp->mnt_data = 0;
365 	return (0);
366 }
367 
368 static int
369 union_root(mp, vpp)
370 	struct mount *mp;
371 	struct vnode **vpp;
372 {
373 	struct union_mount *um = MOUNTTOUNIONMOUNT(mp);
374 	int error;
375 
376 	/*
377 	 * Supply an unlocked reference to um_uppervp and to um_lowervp.  It
378 	 * is possible for um_uppervp to be locked without the associated
379 	 * root union_node being locked.  We let union_allocvp() deal with
380 	 * it.
381 	 */
382 	UDEBUG(("union_root UPPERVP %p locked = %d\n", um->um_uppervp,
383 	    VOP_ISLOCKED(um->um_uppervp, NULL)));
384 
385 	VREF(um->um_uppervp);
386 	if (um->um_lowervp)
387 		VREF(um->um_lowervp);
388 
389 	error = union_allocvp(vpp, mp, NULLVP, NULLVP, NULL,
390 		    um->um_uppervp, um->um_lowervp, 1);
391 	UDEBUG(("error %d\n", error));
392 	UDEBUG(("union_root2 UPPERVP %p locked = %d\n", um->um_uppervp,
393 	    VOP_ISLOCKED(um->um_uppervp, NULL)));
394 
395 	return (error);
396 }
397 
398 static int
399 union_statfs(mp, sbp, p)
400 	struct mount *mp;
401 	struct statfs *sbp;
402 	struct proc *p;
403 {
404 	int error;
405 	struct union_mount *um = MOUNTTOUNIONMOUNT(mp);
406 	struct statfs mstat;
407 	int lbsize;
408 
409 	UDEBUG(("union_statfs(mp = %p, lvp = %p, uvp = %p)\n",
410 	    (void *)mp, (void *)um->um_lowervp, (void *)um->um_uppervp));
411 
412 	bzero(&mstat, sizeof(mstat));
413 
414 	if (um->um_lowervp) {
415 		error = VFS_STATFS(um->um_lowervp->v_mount, &mstat, p);
416 		if (error)
417 			return (error);
418 	}
419 
420 	/* now copy across the "interesting" information and fake the rest */
421 #if 0
422 	sbp->f_type = mstat.f_type;
423 	sbp->f_flags = mstat.f_flags;
424 	sbp->f_bsize = mstat.f_bsize;
425 	sbp->f_iosize = mstat.f_iosize;
426 #endif
427 	lbsize = mstat.f_bsize;
428 	sbp->f_blocks = mstat.f_blocks;
429 	sbp->f_bfree = mstat.f_bfree;
430 	sbp->f_bavail = mstat.f_bavail;
431 	sbp->f_files = mstat.f_files;
432 	sbp->f_ffree = mstat.f_ffree;
433 
434 	error = VFS_STATFS(um->um_uppervp->v_mount, &mstat, p);
435 	if (error)
436 		return (error);
437 
438 	sbp->f_flags = mstat.f_flags;
439 	sbp->f_bsize = mstat.f_bsize;
440 	sbp->f_iosize = mstat.f_iosize;
441 
442 	/*
443 	 * if the lower and upper blocksizes differ, then frig the
444 	 * block counts so that the sizes reported by df make some
445 	 * kind of sense.  none of this makes sense though.
446 	 */
447 
448 	if (mstat.f_bsize != lbsize)
449 		sbp->f_blocks = ((off_t) sbp->f_blocks * lbsize) / mstat.f_bsize;
450 
451 	/*
452 	 * The "total" fields count total resources in all layers,
453 	 * the "free" fields count only those resources which are
454 	 * free in the upper layer (since only the upper layer
455 	 * is writeable).
456 	 */
457 	sbp->f_blocks += mstat.f_blocks;
458 	sbp->f_bfree = mstat.f_bfree;
459 	sbp->f_bavail = mstat.f_bavail;
460 	sbp->f_files += mstat.f_files;
461 	sbp->f_ffree = mstat.f_ffree;
462 
463 	if (sbp != &mp->mnt_stat) {
464 		sbp->f_type = mp->mnt_vfc->vfc_typenum;
465 		bcopy(&mp->mnt_stat.f_fsid, &sbp->f_fsid, sizeof(sbp->f_fsid));
466 		bcopy(mp->mnt_stat.f_mntonname, sbp->f_mntonname, MNAMELEN);
467 		bcopy(mp->mnt_stat.f_mntfromname, sbp->f_mntfromname, MNAMELEN);
468 	}
469 	return (0);
470 }
471 
472 static struct vfsops union_vfsops = {
473 	union_mount,
474 	vfs_stdstart,	/* underlying start already done */
475 	union_unmount,
476 	union_root,
477 	vfs_stdquotactl,
478 	union_statfs,
479 	vfs_stdsync,    /* XXX assumes no cached data on union level */
480 	vfs_stdvget,
481 	vfs_stdfhtovp,
482 	vfs_stdcheckexp,
483 	vfs_stdvptofh,
484 	union_init,
485 	vfs_stduninit,
486 	vfs_stdextattrctl,
487 };
488 
489 VFS_SET(union_vfsops, unionfs, VFCF_LOOPBACK);
490