xref: /dflybsd-src/sys/vfs/nullfs/null_vnops.c (revision 57fed2afae86702adfd8bc0f2b73e76280fa6847)
1 /*
2  * Copyright (c) 1992, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * This code is derived from software contributed to Berkeley by
6  * John Heidemann of the UCLA Ficus project.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. All advertising materials mentioning features or use of this software
17  *    must display the following acknowledgement:
18  *	This product includes software developed by the University of
19  *	California, Berkeley and its contributors.
20  * 4. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)null_vnops.c	8.6 (Berkeley) 5/27/95
37  *
38  * Ancestors:
39  *	@(#)lofs_vnops.c	1.2 (Berkeley) 6/18/92
40  * $FreeBSD: src/sys/miscfs/nullfs/null_vnops.c,v 1.38.2.6 2002/07/31 00:32:28 semenu Exp $
41  * $DragonFly: src/sys/vfs/nullfs/null_vnops.c,v 1.19 2004/10/27 08:52:06 dillon Exp $
42  *	...and...
43  *	@(#)null_vnodeops.c 1.20 92/07/07 UCLA Ficus project
44  *
45  * $FreeBSD: src/sys/miscfs/nullfs/null_vnops.c,v 1.38.2.6 2002/07/31 00:32:28 semenu Exp $
46  */
47 
48 /*
49  * Null Layer
50  *
51  * (See mount_null(8) for more information.)
52  *
53  * The null layer duplicates a portion of the file system
54  * name space under a new name.  In this respect, it is
55  * similar to the loopback file system.  It differs from
56  * the loopback fs in two respects:  it is implemented using
57  * a stackable layers techniques, and its "null-node"s stack above
58  * all lower-layer vnodes, not just over directory vnodes.
59  *
60  * The null layer has two purposes.  First, it serves as a demonstration
61  * of layering by proving a layer which does nothing.  (It actually
62  * does everything the loopback file system does, which is slightly
63  * more than nothing.)  Second, the null layer can serve as a prototype
64  * layer.  Since it provides all necessary layer framework,
65  * new file system layers can be created very easily be starting
66  * with a null layer.
67  *
68  * The remainder of this man page examines the null layer as a basis
69  * for constructing new layers.
70  *
71  *
72  * INSTANTIATING NEW NULL LAYERS
73  *
74  * New null layers are created with mount_null(8).
75  * Mount_null(8) takes two arguments, the pathname
76  * of the lower vfs (target-pn) and the pathname where the null
77  * layer will appear in the namespace (alias-pn).  After
78  * the null layer is put into place, the contents
79  * of target-pn subtree will be aliased under alias-pn.
80  *
81  *
82  * OPERATION OF A NULL LAYER
83  *
84  * The null layer is the minimum file system layer,
85  * simply bypassing all possible operations to the lower layer
86  * for processing there.  The majority of its activity centers
87  * on the bypass routine, through which nearly all vnode operations
88  * pass.
89  *
90  * The bypass routine accepts arbitrary vnode operations for
91  * handling by the lower layer.  It begins by examing vnode
92  * operation arguments and replacing any null-nodes by their
93  * lower-layer equivlants.  It then invokes the operation
94  * on the lower layer.  Finally, it replaces the null-nodes
95  * in the arguments and, if a vnode is return by the operation,
96  * stacks a null-node on top of the returned vnode.
97  *
98  * Although bypass handles most operations, vop_getattr, vop_lock,
99  * vop_unlock, vop_inactive, vop_reclaim, and vop_print are not
100  * bypassed. Vop_getattr must change the fsid being returned.
101  * Vop_lock and vop_unlock must handle any locking for the
102  * current vnode as well as pass the lock request down.
103  * Vop_inactive and vop_reclaim are not bypassed so that
104  * they can handle freeing null-layer specific data. Vop_print
105  * is not bypassed to avoid excessive debugging information.
106  * Also, certain vnode operations change the locking state within
107  * the operation (create, mknod, remove, link, rename, mkdir, rmdir,
108  * and symlink). Ideally these operations should not change the
109  * lock state, but should be changed to let the caller of the
110  * function unlock them. Otherwise all intermediate vnode layers
111  * (such as union, umapfs, etc) must catch these functions to do
112  * the necessary locking at their layer.
113  *
114  *
115  * INSTANTIATING VNODE STACKS
116  *
117  * Mounting associates the null layer with a lower layer,
118  * effect stacking two VFSes.  Vnode stacks are instead
119  * created on demand as files are accessed.
120  *
121  * The initial mount creates a single vnode stack for the
122  * root of the new null layer.  All other vnode stacks
123  * are created as a result of vnode operations on
124  * this or other null vnode stacks.
125  *
126  * New vnode stacks come into existance as a result of
127  * an operation which returns a vnode.
128  * The bypass routine stacks a null-node above the new
129  * vnode before returning it to the caller.
130  *
131  * For example, imagine mounting a null layer with
132  * "mount_null /usr/include /dev/layer/null".
133  * Changing directory to /dev/layer/null will assign
134  * the root null-node (which was created when the null layer was mounted).
135  * Now consider opening "sys".  A vop_lookup would be
136  * done on the root null-node.  This operation would bypass through
137  * to the lower layer which would return a vnode representing
138  * the UFS "sys".  Null_bypass then builds a null-node
139  * aliasing the UFS "sys" and returns this to the caller.
140  * Later operations on the null-node "sys" will repeat this
141  * process when constructing other vnode stacks.
142  *
143  *
144  * CREATING OTHER FILE SYSTEM LAYERS
145  *
146  * One of the easiest ways to construct new file system layers is to make
147  * a copy of the null layer, rename all files and variables, and
148  * then begin modifing the copy.  Sed can be used to easily rename
149  * all variables.
150  *
151  * The umap layer is an example of a layer descended from the
152  * null layer.
153  *
154  *
155  * INVOKING OPERATIONS ON LOWER LAYERS
156  *
157  * There are two techniques to invoke operations on a lower layer
158  * when the operation cannot be completely bypassed.  Each method
159  * is appropriate in different situations.  In both cases,
160  * it is the responsibility of the aliasing layer to make
161  * the operation arguments "correct" for the lower layer
162  * by mapping an vnode arguments to the lower layer.
163  *
164  * The first approach is to call the aliasing layer's bypass routine.
165  * This method is most suitable when you wish to invoke the operation
166  * currently being handled on the lower layer.  It has the advantage
167  * that the bypass routine already must do argument mapping.
168  * An example of this is null_getattrs in the null layer.
169  *
170  * A second approach is to directly invoke vnode operations on
171  * the lower layer with the VOP_OPERATIONNAME interface.
172  * The advantage of this method is that it is easy to invoke
173  * arbitrary operations on the lower layer.  The disadvantage
174  * is that vnode arguments must be manualy mapped.
175  *
176  */
177 
178 #include <sys/param.h>
179 #include <sys/systm.h>
180 #include <sys/kernel.h>
181 #include <sys/sysctl.h>
182 #include <sys/vnode.h>
183 #include <sys/mount.h>
184 #include <sys/proc.h>
185 #include <sys/namei.h>
186 #include <sys/malloc.h>
187 #include <sys/buf.h>
188 #include "null.h"
189 
190 static int null_bug_bypass = 0;   /* for debugging: enables bypass printf'ing */
191 SYSCTL_INT(_debug, OID_AUTO, nullfs_bug_bypass, CTLFLAG_RW,
192 	&null_bug_bypass, 0, "");
193 
194 static int	null_resolve(struct vop_resolve_args *ap);
195 static int	null_revoke(struct vop_revoke_args *ap);
196 static int	null_access(struct vop_access_args *ap);
197 static int	null_createvobject(struct vop_createvobject_args *ap);
198 static int	null_destroyvobject(struct vop_destroyvobject_args *ap);
199 static int	null_getattr(struct vop_getattr_args *ap);
200 static int	null_getvobject(struct vop_getvobject_args *ap);
201 static int	null_inactive(struct vop_inactive_args *ap);
202 static int	null_islocked(struct vop_islocked_args *ap);
203 static int	null_lock(struct vop_lock_args *ap);
204 static int	null_lookup(struct vop_lookup_args *ap);
205 static int	null_open(struct vop_open_args *ap);
206 static int	null_print(struct vop_print_args *ap);
207 static int	null_reclaim(struct vop_reclaim_args *ap);
208 static int	null_rename(struct vop_rename_args *ap);
209 static int	null_setattr(struct vop_setattr_args *ap);
210 static int	null_unlock(struct vop_unlock_args *ap);
211 
212 /*
213  * This is the 10-Apr-92 bypass routine.
214  *    This version has been optimized for speed, throwing away some
215  * safety checks.  It should still always work, but it's not as
216  * robust to programmer errors.
217  *
218  * In general, we map all vnodes going down and unmap them on the way back.
219  * As an exception to this, vnodes can be marked "unmapped" by setting
220  * the Nth bit in operation's vdesc_flags.
221  *
222  * Also, some BSD vnode operations have the side effect of vrele'ing
223  * their arguments.  With stacking, the reference counts are held
224  * by the upper node, not the lower one, so we must handle these
225  * side-effects here.  This is not of concern in Sun-derived systems
226  * since there are no such side-effects.
227  *
228  * This makes the following assumptions:
229  * - only one returned vpp
230  * - no INOUT vpp's (Sun's vop_open has one of these)
231  * - the vnode operation vector of the first vnode should be used
232  *   to determine what implementation of the op should be invoked
233  * - all mapped vnodes are of our vnode-type (NEEDSWORK:
234  *   problems on rmdir'ing mount points and renaming?)
235  *
236  * null_bypass(struct vnodeop_desc *a_desc, ...)
237  */
238 int
239 null_bypass(struct vop_generic_args *ap)
240 {
241 	struct vnode **this_vp_p;
242 	int error;
243 	struct vnode *old_vps[VDESC_MAX_VPS];
244 	struct vnode **vps_p[VDESC_MAX_VPS];
245 	struct vnode ***vppp;
246 	struct vnodeop_desc *descp = ap->a_desc;
247 	int reles, i, j;
248 
249 	if (null_bug_bypass)
250 		printf ("null_bypass: %s\n", descp->vdesc_name);
251 
252 #ifdef DIAGNOSTIC
253 	/*
254 	 * We require at least one vp.
255 	 */
256 	if (descp->vdesc_vp_offsets == NULL ||
257 	    descp->vdesc_vp_offsets[0] == VDESC_NO_OFFSET)
258 		panic ("null_bypass: no vp's in map");
259 #endif
260 
261 	/*
262 	 * Map the vnodes going in.
263 	 */
264 	reles = descp->vdesc_flags;
265 	for (i = 0; i < VDESC_MAX_VPS; ++i) {
266 		if (descp->vdesc_vp_offsets[i] == VDESC_NO_OFFSET)
267 			break;   /* bail out at end of list */
268 		vps_p[i] = this_vp_p =
269 			VOPARG_OFFSETTO(struct vnode**,descp->vdesc_vp_offsets[i],ap);
270 		/*
271 		 * We're not guaranteed that any but the first vnode
272 		 * are of our type.  Check for and don't map any
273 		 * that aren't.  (We must always map first vp or vclean fails.)
274 		 */
275 		if (i && (*this_vp_p == NULLVP ||
276 		    (*this_vp_p)->v_tag != VT_NULL)) {
277 			old_vps[i] = NULLVP;
278 		} else {
279 			old_vps[i] = *this_vp_p;
280 			*this_vp_p = NULLVPTOLOWERVP(*this_vp_p);
281 			/*
282 			 * Several operations have the side effect of vrele'ing
283 			 * their vp's.  We must account for that in the lower
284 			 * vp we pass down.
285 			 */
286 			if (reles & (VDESC_VP0_WILLRELE << i))
287 				vref(*this_vp_p);
288 		}
289 
290 	}
291 
292 	/*
293 	 * Call the operation on the lower layer with the modified
294 	 * argument structure.  We have to adjust a_fm to point to the
295 	 * lower vp's vop_ops structure.
296 	 */
297 	if (vps_p[0] && *vps_p[0]) {
298 		ap->a_ops = (*(vps_p[0]))->v_ops;
299 		error = vop_vnoperate_ap(ap);
300 	} else {
301 		printf("null_bypass: no map for %s\n", descp->vdesc_name);
302 		error = EINVAL;
303 	}
304 
305 	/*
306 	 * Maintain the illusion of call-by-value by restoring vnodes in the
307 	 * argument structure to their original value.
308 	 */
309 	reles = descp->vdesc_flags;
310 	for (i = 0; i < VDESC_MAX_VPS; ++i) {
311 		if (descp->vdesc_vp_offsets[i] == VDESC_NO_OFFSET)
312 			break;   /* bail out at end of list */
313 		if (old_vps[i]) {
314 			*(vps_p[i]) = old_vps[i];
315 
316 			/*
317 			 * Since we operated on the lowervp's instead of the
318 			 * null node vp's, we have to adjust the null node
319 			 * vp's based on what the VOP did to the lower vp.
320 			 *
321 			 * Note: the unlock case only occurs with rename.
322 			 * tdvp and tvp are both locked on call and must be
323 			 * unlocked on return.
324 			 *
325 			 * Unlock semantics indicate that if two locked vp's
326 			 * are passed and they are the same vp, they are only
327 			 * actually locked once.
328 			 */
329 			if (reles & (VDESC_VP0_WILLUNLOCK << i)) {
330 				VOP_UNLOCK(old_vps[i], LK_THISLAYER, curthread);
331 				for (j = i + 1; j < VDESC_MAX_VPS; ++j) {
332 					if (descp->vdesc_vp_offsets[j] == VDESC_NO_OFFSET)
333 						break;
334 					if (old_vps[i] == old_vps[j]) {
335 						reles &= ~(1 << (VDESC_VP0_WILLUNLOCK << j));
336 					}
337 				}
338 			}
339 
340 			if (reles & (VDESC_VP0_WILLRELE << i))
341 				vrele(old_vps[i]);
342 		}
343 	}
344 
345 	/*
346 	 * Map the possible out-going vpp
347 	 * (Assumes that the lower layer always returns
348 	 * a vref'ed vpp unless it gets an error.)
349 	 */
350 	if (descp->vdesc_vpp_offset != VDESC_NO_OFFSET &&
351 	    !(descp->vdesc_flags & VDESC_NOMAP_VPP) &&
352 	    !error) {
353 		/*
354 		 * XXX - even though some ops have vpp returned vp's,
355 		 * several ops actually vrele this before returning.
356 		 * We must avoid these ops.
357 		 * (This should go away when these ops are regularized.)
358 		 */
359 		if (descp->vdesc_flags & VDESC_VPP_WILLRELE)
360 			goto out;
361 		vppp = VOPARG_OFFSETTO(struct vnode***,
362 				 descp->vdesc_vpp_offset,ap);
363 		if (*vppp)
364 			error = null_node_create(old_vps[0]->v_mount, **vppp, *vppp);
365 	}
366 
367  out:
368 	return (error);
369 }
370 
371 /*
372  * We have to carry on the locking protocol on the null layer vnodes
373  * as we progress through the tree. We also have to enforce read-only
374  * if this layer is mounted read-only.
375  *
376  * null_lookup(struct vnode *a_dvp, struct vnode **a_vpp,
377  *		struct componentname *a_cnp)
378  */
379 static int
380 null_lookup(struct vop_lookup_args *ap)
381 {
382 	struct componentname *cnp = ap->a_cnp;
383 	struct vnode *dvp = ap->a_dvp;
384 	struct thread *td = cnp->cn_td;
385 	int flags = cnp->cn_flags;
386 	struct vnode *vp, *ldvp, *lvp;
387 	int error;
388 
389 	if ((flags & CNP_ISLASTCN) &&
390 	    (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
391 	    (cnp->cn_nameiop == NAMEI_DELETE ||
392 	     cnp->cn_nameiop == NAMEI_RENAME)) {
393 		return (EROFS);
394 	}
395 	ldvp = NULLVPTOLOWERVP(dvp);
396 
397 	/*
398 	 * If we are doing a ".." lookup we must release the lock on dvp
399 	 * now, before we run a lookup in the underlying fs, or we may
400 	 * deadlock.  If we do this we must protect ldvp by ref'ing it.
401 	 */
402 	if (flags & CNP_ISDOTDOT) {
403 		vref(ldvp);
404 		VOP_UNLOCK(dvp, LK_THISLAYER, td);
405 	}
406 
407 	/*
408 	 * Due to the non-deterministic nature of the handling of the
409 	 * parent directory lock by lookup, we cannot call null_bypass()
410 	 * here.  We must make a direct call.  It's faster to do a direct
411 	 * call, anyway.
412 	 */
413 	vp = lvp = NULL;
414 	error = VOP_LOOKUP(ldvp, &lvp, cnp);
415 	if (error == EJUSTRETURN && (flags & CNP_ISLASTCN) &&
416 	    (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
417 	    (cnp->cn_nameiop == NAMEI_CREATE ||
418 	     cnp->cn_nameiop == NAMEI_RENAME)) {
419 		error = EROFS;
420 	}
421 
422 	if ((error == 0 || error == EJUSTRETURN) && lvp != NULL) {
423 		if (ldvp == lvp) {
424 			*ap->a_vpp = dvp;
425 			vref(dvp);
426 			vrele(lvp);
427 		} else {
428 			error = null_node_create(dvp->v_mount, lvp, &vp);
429 			if (error == 0)
430 				*ap->a_vpp = vp;
431 		}
432 	}
433 
434 	/*
435 	 * The underlying fs will set PDIRUNLOCK if it unlocked the parent
436 	 * directory, which means we have to follow suit in the nullfs layer.
437 	 * Note that the parent directory may have already been unlocked due
438 	 * to the ".." case.  Note that use of cnp->cn_flags instead of flags.
439 	 */
440 	if (flags & CNP_ISDOTDOT) {
441 		if ((cnp->cn_flags & CNP_PDIRUNLOCK) == 0)
442 			VOP_LOCK(dvp, LK_THISLAYER | LK_EXCLUSIVE, td);
443 		vrele(ldvp);
444 	} else if (cnp->cn_flags & CNP_PDIRUNLOCK) {
445 		VOP_UNLOCK(dvp, LK_THISLAYER, td);
446 	}
447 	return (error);
448 }
449 
450 /*
451  * Setattr call. Disallow write attempts if the layer is mounted read-only.
452  *
453  * null_setattr(struct vnodeop_desc *a_desc, struct vnode *a_vp,
454  *		struct vattr *a_vap, struct ucred *a_cred,
455  *		struct thread *a_td)
456  */
457 int
458 null_setattr(struct vop_setattr_args *ap)
459 {
460 	struct vnode *vp = ap->a_vp;
461 	struct vattr *vap = ap->a_vap;
462 
463   	if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
464 	    vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
465 	    vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
466 	    (vp->v_mount->mnt_flag & MNT_RDONLY))
467 		return (EROFS);
468 	if (vap->va_size != VNOVAL) {
469  		switch (vp->v_type) {
470  		case VDIR:
471  			return (EISDIR);
472  		case VCHR:
473  		case VBLK:
474  		case VSOCK:
475  		case VFIFO:
476 			if (vap->va_flags != VNOVAL)
477 				return (EOPNOTSUPP);
478 			return (0);
479 		case VREG:
480 		case VLNK:
481  		default:
482 			/*
483 			 * Disallow write attempts if the filesystem is
484 			 * mounted read-only.
485 			 */
486 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
487 				return (EROFS);
488 		}
489 	}
490 
491 	return (null_bypass(&ap->a_head));
492 }
493 
494 /*
495  *  We handle getattr only to change the fsid.
496  *
497  * null_getattr(struct vnode *a_vp, struct vattr *a_vap, struct ucred *a_cred,
498  *		struct thread *a_td)
499  */
500 static int
501 null_getattr(struct vop_getattr_args *ap)
502 {
503 	int error;
504 
505 	if ((error = null_bypass(&ap->a_head)) != 0)
506 		return (error);
507 
508 	ap->a_vap->va_fsid = ap->a_vp->v_mount->mnt_stat.f_fsid.val[0];
509 	return (0);
510 }
511 
512 /*
513  * Resolve a locked ncp at the nullfs layer.
514  */
515 static int
516 null_resolve(struct vop_resolve_args *ap)
517 {
518 	return(vop_noresolve(ap));
519 }
520 
521 /*
522  * revoke is VX locked, we can't go through null_bypass
523  */
524 static int
525 null_revoke(struct vop_revoke_args *ap)
526 {
527 	struct null_node *np;
528 	struct vnode *lvp;
529 
530 	np = VTONULL(ap->a_vp);
531 	vx_unlock(ap->a_vp);
532 	if ((lvp = np->null_lowervp) != NULL) {
533 		vx_get(lvp);
534 		VOP_REVOKE(lvp, ap->a_flags);
535 		vx_put(lvp);
536 	}
537 	vx_lock(ap->a_vp);
538 	vgone(ap->a_vp);
539 	return(0);
540 }
541 
542 /*
543  * Handle to disallow write access if mounted read-only.
544  *
545  * null_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred,
546  *		struct thread *a_td)
547  */
548 static int
549 null_access(struct vop_access_args *ap)
550 {
551 	struct vnode *vp = ap->a_vp;
552 	mode_t mode = ap->a_mode;
553 
554 	/*
555 	 * Disallow write attempts on read-only layers;
556 	 * unless the file is a socket, fifo, or a block or
557 	 * character device resident on the file system.
558 	 */
559 	if (mode & VWRITE) {
560 		switch (vp->v_type) {
561 		case VDIR:
562 		case VLNK:
563 		case VREG:
564 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
565 				return (EROFS);
566 			break;
567 		default:
568 			break;
569 		}
570 	}
571 	return (null_bypass(&ap->a_head));
572 }
573 
574 /*
575  * We must handle open to be able to catch MNT_NODEV and friends.
576  *
577  * null_open(struct vnode *a_vp, int a_mode, struct ucred *a_cred,
578  *	     struct thread *a_td)
579  */
580 static int
581 null_open(struct vop_open_args *ap)
582 {
583 	struct vnode *vp = ap->a_vp;
584 	struct vnode *lvp = NULLVPTOLOWERVP(ap->a_vp);
585 
586 	if ((vp->v_mount->mnt_flag & MNT_NODEV) &&
587 	    (lvp->v_type == VBLK || lvp->v_type == VCHR))
588 		return ENXIO;
589 
590 	return (null_bypass(&ap->a_head));
591 }
592 
593 /*
594  * We handle this to eliminate null FS to lower FS
595  * file moving. Don't know why we don't allow this,
596  * possibly we should.
597  *
598  * null_rename(struct vnode *a_fdvp, struct vnode *a_fvp,
599  *		struct componentname *a_fcnp, struct vnode *a_tdvp,
600  *		struct vnode *a_tvp, struct componentname *a_tcnp)
601  */
602 static int
603 null_rename(struct vop_rename_args *ap)
604 {
605 	struct vnode *tdvp = ap->a_tdvp;
606 	struct vnode *fvp = ap->a_fvp;
607 	struct vnode *fdvp = ap->a_fdvp;
608 	struct vnode *tvp = ap->a_tvp;
609 
610 	/* Check for cross-device rename. */
611 	if ((fvp->v_mount != tdvp->v_mount) ||
612 	    (tvp && (fvp->v_mount != tvp->v_mount))) {
613 		if (tdvp == tvp)
614 			vrele(tdvp);
615 		else
616 			vput(tdvp);
617 		if (tvp)
618 			vput(tvp);
619 		vrele(fdvp);
620 		vrele(fvp);
621 		return (EXDEV);
622 	}
623 
624 	return (null_bypass(&ap->a_head));
625 }
626 
627 /*
628  * A special flag, LK_THISLAYER, causes the locking function to operate
629  * ONLY on the nullfs layer.  Otherwise we are responsible for locking not
630  * only our layer, but the lower layer as well.
631  *
632  * null_lock(struct vnode *a_vp, int a_flags, struct thread *a_td)
633  */
634 static int
635 null_lock(struct vop_lock_args *ap)
636 {
637 	struct vnode *vp = ap->a_vp;
638 	int flags = ap->a_flags;
639 	struct null_node *np = VTONULL(vp);
640 	struct vnode *lvp;
641 	int error;
642 
643 	/*
644 	 * Lock the nullfs layer first, disposing of the interlock in the
645 	 * process.
646 	 */
647 	KKASSERT((flags & LK_INTERLOCK) == 0);
648 	error = lockmgr(&vp->v_lock, flags & ~LK_THISLAYER,
649 			NULL, ap->a_td);
650 
651 	/*
652 	 * If locking only the nullfs layer, or if there is no lower layer,
653 	 * or if an error occured while attempting to lock the nullfs layer,
654 	 * we are done.
655 	 *
656 	 * np can be NULL is the vnode is being recycled from a previous
657 	 * hash collision.
658 	 */
659 	if ((flags & LK_THISLAYER) || np == NULL ||
660 	    np->null_lowervp == NULL || error) {
661 		return (error);
662 	}
663 
664 	/*
665 	 * Lock the underlying vnode.  If we are draining we should not drain
666 	 * the underlying vnode, since it is not being destroyed, but we do
667 	 * lock it exclusively in that case.  Note that any interlocks have
668 	 * already been disposed of above.
669 	 */
670 	lvp = np->null_lowervp;
671 	if ((flags & LK_TYPE_MASK) == LK_DRAIN) {
672 		NULLFSDEBUG("null_lock: avoiding LK_DRAIN\n");
673 		error = vn_lock(lvp, (flags & ~LK_TYPE_MASK) | LK_EXCLUSIVE,
674 				ap->a_td);
675 	} else {
676 		error = vn_lock(lvp, flags, ap->a_td);
677 	}
678 
679 	/*
680 	 * If an error occured we have to undo our nullfs lock, then return
681 	 * the original error.
682 	 */
683 	if (error)
684 		lockmgr(&vp->v_lock, LK_RELEASE, NULL, ap->a_td);
685 	return(error);
686 }
687 
688 /*
689  * A special flag, LK_THISLAYER, causes the unlocking function to operate
690  * ONLY on the nullfs layer.  Otherwise we are responsible for unlocking not
691  * only our layer, but the lower layer as well.
692  *
693  * null_unlock(struct vnode *a_vp, int a_flags, struct thread *a_td)
694  */
695 static int
696 null_unlock(struct vop_unlock_args *ap)
697 {
698 	struct vnode *vp = ap->a_vp;
699 	int flags = ap->a_flags;
700 	struct null_node *np = VTONULL(vp);
701 	struct vnode *lvp;
702 	int error;
703 
704 	KKASSERT((flags & LK_INTERLOCK) == 0);
705 	/*
706 	 * nullfs layer only
707 	 */
708 	if (flags & LK_THISLAYER) {
709 		error = lockmgr(&vp->v_lock,
710 				(flags & ~LK_THISLAYER) | LK_RELEASE,
711 				NULL, ap->a_td);
712 		return (error);
713 	}
714 
715 	/*
716 	 * If there is no underlying vnode the lock operation occurs at
717 	 * the nullfs layer.  np can be NULL is the vnode is being recycled
718 	 * from a previous hash collision.
719 	 */
720 	if (np == NULL || (lvp = np->null_lowervp) == NULL) {
721 		error = lockmgr(&vp->v_lock, flags | LK_RELEASE,
722 				NULL, ap->a_td);
723 		return(error);
724 	}
725 
726 	/*
727 	 * Unlock the lower layer first, then our nullfs layer.
728 	 */
729 	VOP_UNLOCK(lvp, flags, ap->a_td);
730 	error = lockmgr(&vp->v_lock, flags | LK_RELEASE, NULL, ap->a_td);
731 	return (error);
732 }
733 
734 /*
735  * null_islocked(struct vnode *a_vp, struct thread *a_td)
736  *
737  * If a lower layer exists return the lock status of the lower layer,
738  * otherwise return the lock status of our nullfs layer.
739  */
740 static int
741 null_islocked(struct vop_islocked_args *ap)
742 {
743 	struct vnode *vp = ap->a_vp;
744 	struct vnode *lvp;
745 	struct null_node *np = VTONULL(vp);
746 	int error;
747 
748 	lvp = np->null_lowervp;
749 	if (lvp == NULL)
750 		error = lockstatus(&vp->v_lock, ap->a_td);
751 	else
752 		error = VOP_ISLOCKED(lvp, ap->a_td);
753 	return (error);
754 }
755 
756 
757 /*
758  * The vnode is no longer active.  However, the new VFS API may retain
759  * the node in the vfs cache.  There is no way to tell that someone issued
760  * a remove/rmdir operation on the underlying filesystem (yet), but we can't
761  * remove the lowervp reference here.
762  *
763  * null_inactive(struct vnode *a_vp, struct thread *a_td)
764  */
765 static int
766 null_inactive(struct vop_inactive_args *ap)
767 {
768 	/*struct vnode *vp = ap->a_vp;*/
769 	/*struct null_node *np = VTONULL(vp);*/
770 
771 	/*
772 	 * At the moment don't do anything here.  All the rest of the code
773 	 * assumes that lowervp will remain inact, and the inactive nullvp
774 	 * may be reactivated at any time.  XXX I'm not sure why the 4.x code
775 	 * even worked.
776 	 */
777 
778 	/*
779 	 * Now it is safe to release our nullfs layer vnode.
780 	 */
781 	return (0);
782 }
783 
784 /*
785  * We can free memory in null_inactive, but we do this
786  * here. (Possible to guard vp->v_data to point somewhere)
787  *
788  * null_reclaim(struct vnode *a_vp, struct thread *a_td)
789  */
790 static int
791 null_reclaim(struct vop_reclaim_args *ap)
792 {
793 	struct vnode *vp = ap->a_vp;
794 	struct vnode *lowervp;
795 	struct null_node *np;
796 
797 	np = VTONULL(vp);
798 	vp->v_data = NULL;
799 	/*
800 	 * null_lowervp reference to lowervp.  The lower vnode's
801 	 * inactive routine may or may not be called when we do the
802 	 * final vrele().
803 	 */
804 	if (np) {
805 		null_node_rem(np);
806 		lowervp = np->null_lowervp;
807 		np->null_lowervp = NULLVP;
808 		if (lowervp)
809 			vrele(lowervp);
810 		free(np, M_NULLFSNODE);
811 	}
812 	return (0);
813 }
814 
815 /*
816  * null_print(struct vnode *a_vp)
817  */
818 static int
819 null_print(struct vop_print_args *ap)
820 {
821 	struct vnode *vp = ap->a_vp;
822 	struct null_node *np = VTONULL(vp);
823 
824 	if (np == NULL) {
825 		printf ("\ttag VT_NULLFS, vp=%p, NULL v_data!\n", vp);
826 		return(0);
827 	}
828 	printf ("\ttag VT_NULLFS, vp=%p, lowervp=%p\n", vp, np->null_lowervp);
829 	if (np->null_lowervp != NULL) {
830 		printf("\tlowervp_lock: ");
831 		lockmgr_printinfo(&np->null_lowervp->v_lock);
832 	} else {
833 		printf("\tnull_lock: ");
834 		lockmgr_printinfo(&vp->v_lock);
835 	}
836 	printf("\n");
837 	return (0);
838 }
839 
840 /*
841  * Let an underlying filesystem do the work
842  *
843  * null_createvobject(struct vnode *vp, struct ucred *cred, struct proc *p)
844  */
845 static int
846 null_createvobject(struct vop_createvobject_args *ap)
847 {
848 	struct vnode *vp = ap->a_vp;
849 	struct vnode *lowervp = VTONULL(vp) ? NULLVPTOLOWERVP(vp) : NULL;
850 	int error;
851 
852 	if (vp->v_type == VNON || lowervp == NULL)
853 		return 0;
854 	error = VOP_CREATEVOBJECT(lowervp, ap->a_td);
855 	if (error)
856 		return (error);
857 	vp->v_flag |= VOBJBUF;
858 	return (0);
859 }
860 
861 /*
862  * We have nothing to destroy and this operation shouldn't be bypassed.
863  *
864  * null_destroyvobject(struct vnode *vp)
865  */
866 static int
867 null_destroyvobject(struct vop_destroyvobject_args *ap)
868 {
869 	struct vnode *vp = ap->a_vp;
870 
871 	vp->v_flag &= ~VOBJBUF;
872 	return (0);
873 }
874 
875 /*
876  * null_getvobject(struct vnode *vp, struct vm_object **objpp)
877  *
878  * Note that this can be called when a vnode is being recycled, and
879  * v_data may be NULL in that case if nullfs had to recycle a vnode
880  * due to a null_node collision.
881  */
882 static int
883 null_getvobject(struct vop_getvobject_args *ap)
884 {
885 	struct vnode *lvp;
886 
887 	if (ap->a_vp->v_data == NULL)
888 		return EINVAL;
889 
890 	lvp = NULLVPTOLOWERVP(ap->a_vp);
891 	if (lvp == NULL)
892 		return EINVAL;
893 	return (VOP_GETVOBJECT(lvp, ap->a_objpp));
894 }
895 
896 /*
897  * Global vfs data structures
898  */
899 struct vnodeopv_entry_desc null_vnodeop_entries[] = {
900 	{ &vop_default_desc,		(void *) null_bypass },
901 	{ &vop_resolve_desc,		(void *) null_resolve },
902 	{ &vop_access_desc,		(void *) null_access },
903 	{ &vop_createvobject_desc,	(void *) null_createvobject },
904 	{ &vop_destroyvobject_desc,	(void *) null_destroyvobject },
905 	{ &vop_getattr_desc,		(void *) null_getattr },
906 	{ &vop_getvobject_desc,		(void *) null_getvobject },
907 	{ &vop_inactive_desc,		(void *) null_inactive },
908 	{ &vop_islocked_desc,		(void *) null_islocked },
909 	{ &vop_lock_desc,		(void *) null_lock },
910 	{ &vop_lookup_desc,		(void *) null_lookup },
911 	{ &vop_open_desc,		(void *) null_open },
912 	{ &vop_print_desc,		(void *) null_print },
913 	{ &vop_reclaim_desc,		(void *) null_reclaim },
914 	{ &vop_rename_desc,		(void *) null_rename },
915 	{ &vop_setattr_desc,		(void *) null_setattr },
916 	{ &vop_unlock_desc,		(void *) null_unlock },
917 	{ &vop_revoke_desc,		(void *) null_revoke },
918 	{ NULL, NULL }
919 };
920 
921