xref: /freebsd/sys/fs/unionfs/union_vfsops.c (revision cb166ce422ac2bc81f42c2a2e2cd68625c11478d)
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/proc.h>
49 #include <sys/vnode.h>
50 #include <sys/mount.h>
51 #include <sys/namei.h>
52 #include <sys/malloc.h>
53 #include <sys/filedesc.h>
54 #include <miscfs/union/union.h>
55 #include <vm/vm_zone.h>
56 
57 static MALLOC_DEFINE(M_UNIONFSMNT, "UNION mount", "UNION mount structure");
58 
59 extern int	union_init __P((struct vfsconf *));
60 static int	union_mount __P((struct mount *mp, char *path, caddr_t data,
61 				 struct nameidata *ndp, struct proc *p));
62 static int	union_root __P((struct mount *mp, struct vnode **vpp));
63 static int	union_statfs __P((struct mount *mp, struct statfs *sbp,
64 				  struct proc *p));
65 static int	union_unmount __P((struct mount *mp, int mntflags,
66 				   struct proc *p));
67 
68 /*
69  * Mount union filesystem
70  */
71 static int
72 union_mount(mp, path, data, ndp, p)
73 	struct mount *mp;
74 	char *path;
75 	caddr_t data;
76 	struct nameidata *ndp;
77 	struct proc *p;
78 {
79 	int error = 0;
80 	struct union_args args;
81 	struct vnode *lowerrootvp = NULLVP;
82 	struct vnode *upperrootvp = NULLVP;
83 	struct union_mount *um = 0;
84 	struct ucred *cred = 0;
85 	char *cp = 0;
86 	int len;
87 	u_int size;
88 
89 	UDEBUG(("union_mount(mp = %p)\n", (void *)mp));
90 
91 	/*
92 	 * Disable clustered write, otherwise system becomes unstable.
93 	 */
94 	mp->mnt_flag |= MNT_NOCLUSTERW;
95 
96 	/*
97 	 * Update is a no-op
98 	 */
99 	if (mp->mnt_flag & MNT_UPDATE) {
100 		/*
101 		 * Need to provide.
102 		 * 1. a way to convert between rdonly and rdwr mounts.
103 		 * 2. support for nfs exports.
104 		 */
105 		error = EOPNOTSUPP;
106 		goto bad;
107 	}
108 
109 	/*
110 	 * Get argument
111 	 */
112 	error = copyin(data, (caddr_t)&args, sizeof(struct union_args));
113 	if (error)
114 		goto bad;
115 
116 	/*
117 	 * Obtain lower vnode.  Vnode is stored in mp->mnt_vnodecovered.
118 	 * We need to reference it but not lock it.
119 	 */
120 
121 	lowerrootvp = mp->mnt_vnodecovered;
122 	VREF(lowerrootvp);
123 
124 #if 0
125 	/*
126 	 * Unlock lower node to avoid deadlock.
127 	 */
128 	if (lowerrootvp->v_op == union_vnodeop_p)
129 		VOP_UNLOCK(lowerrootvp, 0, p);
130 #endif
131 
132 	/*
133 	 * Obtain upper vnode by calling namei() on the path.  The
134 	 * upperrootvp will be turned referenced but not locked.
135 	 */
136 	NDINIT(ndp, LOOKUP, FOLLOW|WANTPARENT,
137 	       UIO_USERSPACE, args.target, p);
138 
139 	error = namei(ndp);
140 
141 #if 0
142 	if (lowerrootvp->v_op == union_vnodeop_p)
143 		vn_lock(lowerrootvp, LK_EXCLUSIVE | LK_RETRY, p);
144 #endif
145 	if (error)
146 		goto bad;
147 
148 	NDFREE(ndp, NDF_ONLY_PNBUF);
149 	upperrootvp = ndp->ni_vp;
150 	vrele(ndp->ni_dvp);
151 	ndp->ni_dvp = NULL;
152 
153 	UDEBUG(("mount_root UPPERVP %p locked = %d\n", upperrootvp,
154 	    VOP_ISLOCKED(upperrootvp, NULL)));
155 
156 	/*
157 	 * Check multi union mount to avoid `lock myself again' panic.
158 	 * Also require that it be a directory.
159 	 */
160 	if (upperrootvp == VTOUNION(lowerrootvp)->un_uppervp) {
161 #ifdef DIAGNOSTIC
162 		printf("union_mount: multi union mount?\n");
163 #endif
164 		error = EDEADLK;
165 		goto bad;
166 	}
167 
168 	if (upperrootvp->v_type != VDIR) {
169 		error = EINVAL;
170 		goto bad;
171 	}
172 
173 	/*
174 	 * Allocate our union_mount structure and populate the fields.
175 	 * The vnode references are stored in the union_mount as held,
176 	 * unlocked references.  Depending on the _BELOW flag, the
177 	 * filesystems are viewed in a different order.  In effect this
178 	 * is the same as providing a mount-under option to the mount
179 	 * syscall.
180 	 */
181 
182 	um = (struct union_mount *) malloc(sizeof(struct union_mount),
183 				M_UNIONFSMNT, M_WAITOK);
184 
185 	bzero(um, sizeof(struct union_mount));
186 
187 	um->um_op = args.mntflags & UNMNT_OPMASK;
188 
189 	switch (um->um_op) {
190 	case UNMNT_ABOVE:
191 		um->um_lowervp = lowerrootvp;
192 		um->um_uppervp = upperrootvp;
193 		upperrootvp = NULL;
194 		lowerrootvp = NULL;
195 		break;
196 
197 	case UNMNT_BELOW:
198 		um->um_lowervp = upperrootvp;
199 		um->um_uppervp = lowerrootvp;
200 		upperrootvp = NULL;
201 		lowerrootvp = NULL;
202 		break;
203 
204 	case UNMNT_REPLACE:
205 		vrele(lowerrootvp);
206 		lowerrootvp = NULL;
207 		um->um_uppervp = upperrootvp;
208 		um->um_lowervp = lowerrootvp;
209 		upperrootvp = NULL;
210 		break;
211 
212 	default:
213 		error = EINVAL;
214 		goto bad;
215 	}
216 
217 	/*
218 	 * Unless the mount is readonly, ensure that the top layer
219 	 * supports whiteout operations
220 	 */
221 	if ((mp->mnt_flag & MNT_RDONLY) == 0) {
222 		error = VOP_WHITEOUT(um->um_uppervp, NULL, LOOKUP);
223 		if (error)
224 			goto bad;
225 	}
226 
227 	um->um_cred = p->p_ucred;
228 	crhold(um->um_cred);
229 	um->um_cmode = UN_DIRMODE &~ p->p_fd->fd_cmask;
230 
231 	/*
232 	 * Depending on what you think the MNT_LOCAL flag might mean,
233 	 * you may want the && to be || on the conditional below.
234 	 * At the moment it has been defined that the filesystem is
235 	 * only local if it is all local, ie the MNT_LOCAL flag implies
236 	 * that the entire namespace is local.  If you think the MNT_LOCAL
237 	 * flag implies that some of the files might be stored locally
238 	 * then you will want to change the conditional.
239 	 */
240 	if (um->um_op == UNMNT_ABOVE) {
241 		if (((um->um_lowervp == NULLVP) ||
242 		     (um->um_lowervp->v_mount->mnt_flag & MNT_LOCAL)) &&
243 		    (um->um_uppervp->v_mount->mnt_flag & MNT_LOCAL))
244 			mp->mnt_flag |= MNT_LOCAL;
245 	}
246 
247 	/*
248 	 * Copy in the upper layer's RDONLY flag.  This is for the benefit
249 	 * of lookup() which explicitly checks the flag, rather than asking
250 	 * the filesystem for its own opinion.  This means, that an update
251 	 * mount of the underlying filesystem to go from rdonly to rdwr
252 	 * will leave the unioned view as read-only.
253 	 */
254 	mp->mnt_flag |= (um->um_uppervp->v_mount->mnt_flag & MNT_RDONLY);
255 
256 	mp->mnt_data = (qaddr_t) um;
257 	vfs_getnewfsid(mp);
258 
259 	(void) copyinstr(path, mp->mnt_stat.f_mntonname, MNAMELEN - 1, &size);
260 	bzero(mp->mnt_stat.f_mntonname + size, MNAMELEN - size);
261 
262 	switch (um->um_op) {
263 	case UNMNT_ABOVE:
264 		cp = "<above>:";
265 		break;
266 	case UNMNT_BELOW:
267 		cp = "<below>:";
268 		break;
269 	case UNMNT_REPLACE:
270 		cp = "";
271 		break;
272 	}
273 	len = strlen(cp);
274 	bcopy(cp, mp->mnt_stat.f_mntfromname, len);
275 
276 	cp = mp->mnt_stat.f_mntfromname + len;
277 	len = MNAMELEN - len;
278 
279 	(void) copyinstr(args.target, cp, len - 1, &size);
280 	bzero(cp + size, len - size);
281 
282 	(void)union_statfs(mp, &mp->mnt_stat, p);
283 
284 	UDEBUG(("union_mount: from %s, on %s\n",
285 		mp->mnt_stat.f_mntfromname, mp->mnt_stat.f_mntonname));
286 	return (0);
287 
288 bad:
289 	if (um) {
290 		if (um->um_uppervp)
291 			vrele(um->um_uppervp);
292 		if (um->um_lowervp)
293 			vrele(um->um_lowervp);
294 		/* XXX other fields */
295 		free(um, M_UNIONFSMNT);
296 	}
297 	if (cred)
298 		crfree(cred);
299 	if (upperrootvp)
300 		vrele(upperrootvp);
301 	if (lowerrootvp)
302 		vrele(lowerrootvp);
303 	return (error);
304 }
305 
306 /*
307  * Free reference to union layer
308  */
309 static int
310 union_unmount(mp, mntflags, p)
311 	struct mount *mp;
312 	int mntflags;
313 	struct proc *p;
314 {
315 	struct union_mount *um = MOUNTTOUNIONMOUNT(mp);
316 	struct vnode *um_rootvp;
317 	int error;
318 	int freeing;
319 	int flags = 0;
320 
321 	UDEBUG(("union_unmount(mp = %p)\n", (void *)mp));
322 
323 	if (mntflags & MNT_FORCE)
324 		flags |= FORCECLOSE;
325 
326 	if ((error = union_root(mp, &um_rootvp)) != 0)
327 		return (error);
328 
329 	/*
330 	 * Keep flushing vnodes from the mount list.
331 	 * This is needed because of the un_pvp held
332 	 * reference to the parent vnode.
333 	 * If more vnodes have been freed on a given pass,
334 	 * the try again.  The loop will iterate at most
335 	 * (d) times, where (d) is the maximum tree depth
336 	 * in the filesystem.
337 	 */
338 	for (freeing = 0; vflush(mp, um_rootvp, flags) != 0;) {
339 		struct vnode *vp;
340 		int n;
341 
342 		/* count #vnodes held on mount list */
343 		for (n = 0, vp = mp->mnt_vnodelist.lh_first;
344 				vp != NULLVP;
345 				vp = vp->v_mntvnodes.le_next)
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 	/* At this point the root vnode should have a single reference */
357 	if (um_rootvp->v_usecount > 1) {
358 		vput(um_rootvp);
359 		return (EBUSY);
360 	}
361 
362 #ifdef DEBUG
363 	vprint("union root", um_rootvp);
364 #endif
365 	/*
366 	 * Discard references to upper and lower target vnodes.
367 	 */
368 	if (um->um_lowervp)
369 		vrele(um->um_lowervp);
370 	vrele(um->um_uppervp);
371 	crfree(um->um_cred);
372 	/*
373 	 * Release reference on underlying root vnode
374 	 */
375 	vput(um_rootvp);
376 	/*
377 	 * And blow it away for future re-use
378 	 */
379 	vgone(um_rootvp);
380 	/*
381 	 * Finally, throw away the union_mount structure
382 	 */
383 	free(mp->mnt_data, M_UNIONFSMNT);	/* XXX */
384 	mp->mnt_data = 0;
385 	return (0);
386 }
387 
388 static int
389 union_root(mp, vpp)
390 	struct mount *mp;
391 	struct vnode **vpp;
392 {
393 	struct union_mount *um = MOUNTTOUNIONMOUNT(mp);
394 	int error;
395 
396 	/*
397 	 * Supply an unlocked reference to um_uppervp and to um_lowervp.  It
398 	 * is possible for um_uppervp to be locked without the associated
399 	 * root union_node being locked.  We let union_allocvp() deal with
400 	 * it.
401 	 */
402 	UDEBUG(("union_root UPPERVP %p locked = %d\n", um->um_uppervp,
403 	    VOP_ISLOCKED(um->um_uppervp, NULL)));
404 
405 	VREF(um->um_uppervp);
406 	if (um->um_lowervp)
407 		VREF(um->um_lowervp);
408 
409 	error = union_allocvp(vpp, mp, NULLVP, NULLVP, NULL,
410 		    um->um_uppervp, um->um_lowervp, 1);
411 	UDEBUG(("error %d\n", error));
412 	UDEBUG(("union_root2 UPPERVP %p locked = %d\n", um->um_uppervp,
413 	    VOP_ISLOCKED(um->um_uppervp, NULL)));
414 
415 	return (error);
416 }
417 
418 static int
419 union_statfs(mp, sbp, p)
420 	struct mount *mp;
421 	struct statfs *sbp;
422 	struct proc *p;
423 {
424 	int error;
425 	struct union_mount *um = MOUNTTOUNIONMOUNT(mp);
426 	struct statfs mstat;
427 	int lbsize;
428 
429 	UDEBUG(("union_statfs(mp = %p, lvp = %p, uvp = %p)\n",
430 	    (void *)mp, (void *)um->um_lowervp, (void *)um->um_uppervp));
431 
432 	bzero(&mstat, sizeof(mstat));
433 
434 	if (um->um_lowervp) {
435 		error = VFS_STATFS(um->um_lowervp->v_mount, &mstat, p);
436 		if (error)
437 			return (error);
438 	}
439 
440 	/* now copy across the "interesting" information and fake the rest */
441 #if 0
442 	sbp->f_type = mstat.f_type;
443 	sbp->f_flags = mstat.f_flags;
444 	sbp->f_bsize = mstat.f_bsize;
445 	sbp->f_iosize = mstat.f_iosize;
446 #endif
447 	lbsize = mstat.f_bsize;
448 	sbp->f_blocks = mstat.f_blocks;
449 	sbp->f_bfree = mstat.f_bfree;
450 	sbp->f_bavail = mstat.f_bavail;
451 	sbp->f_files = mstat.f_files;
452 	sbp->f_ffree = mstat.f_ffree;
453 
454 	error = VFS_STATFS(um->um_uppervp->v_mount, &mstat, p);
455 	if (error)
456 		return (error);
457 
458 	sbp->f_flags = mstat.f_flags;
459 	sbp->f_bsize = mstat.f_bsize;
460 	sbp->f_iosize = mstat.f_iosize;
461 
462 	/*
463 	 * if the lower and upper blocksizes differ, then frig the
464 	 * block counts so that the sizes reported by df make some
465 	 * kind of sense.  none of this makes sense though.
466 	 */
467 
468 	if (mstat.f_bsize != lbsize)
469 		sbp->f_blocks = ((off_t) sbp->f_blocks * lbsize) / mstat.f_bsize;
470 
471 	/*
472 	 * The "total" fields count total resources in all layers,
473 	 * the "free" fields count only those resources which are
474 	 * free in the upper layer (since only the upper layer
475 	 * is writeable).
476 	 */
477 	sbp->f_blocks += mstat.f_blocks;
478 	sbp->f_bfree = mstat.f_bfree;
479 	sbp->f_bavail = mstat.f_bavail;
480 	sbp->f_files += mstat.f_files;
481 	sbp->f_ffree = mstat.f_ffree;
482 
483 	if (sbp != &mp->mnt_stat) {
484 		sbp->f_type = mp->mnt_vfc->vfc_typenum;
485 		bcopy(&mp->mnt_stat.f_fsid, &sbp->f_fsid, sizeof(sbp->f_fsid));
486 		bcopy(mp->mnt_stat.f_mntonname, sbp->f_mntonname, MNAMELEN);
487 		bcopy(mp->mnt_stat.f_mntfromname, sbp->f_mntfromname, MNAMELEN);
488 	}
489 	return (0);
490 }
491 
492 static struct vfsops union_vfsops = {
493 	union_mount,
494 	vfs_stdstart,	/* underlying start already done */
495 	union_unmount,
496 	union_root,
497 	vfs_stdquotactl,
498 	union_statfs,
499 	vfs_stdsync,    /* XXX assumes no cached data on union level */
500 	vfs_stdvget,
501 	vfs_stdfhtovp,
502 	vfs_stdcheckexp,
503 	vfs_stdvptofh,
504 	union_init,
505 	vfs_stduninit,
506 	vfs_stdextattrctl,
507 };
508 
509 VFS_SET(union_vfsops, union, VFCF_LOOPBACK);
510