xref: /freebsd/sys/fs/nullfs/null_vnops.c (revision bfe691b2f75de2224c7ceb304ebcdef2b42d4179)
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  * 4. Neither the name of the University nor the names of its contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  *
32  *	@(#)null_vnops.c	8.6 (Berkeley) 5/27/95
33  *
34  * Ancestors:
35  *	@(#)lofs_vnops.c	1.2 (Berkeley) 6/18/92
36  *	...and...
37  *	@(#)null_vnodeops.c 1.20 92/07/07 UCLA Ficus project
38  *
39  * $FreeBSD$
40  */
41 
42 /*
43  * Null Layer
44  *
45  * (See mount_nullfs(8) for more information.)
46  *
47  * The null layer duplicates a portion of the filesystem
48  * name space under a new name.  In this respect, it is
49  * similar to the loopback filesystem.  It differs from
50  * the loopback fs in two respects:  it is implemented using
51  * a stackable layers techniques, and its "null-node"s stack above
52  * all lower-layer vnodes, not just over directory vnodes.
53  *
54  * The null layer has two purposes.  First, it serves as a demonstration
55  * of layering by proving a layer which does nothing.  (It actually
56  * does everything the loopback filesystem does, which is slightly
57  * more than nothing.)  Second, the null layer can serve as a prototype
58  * layer.  Since it provides all necessary layer framework,
59  * new filesystem layers can be created very easily be starting
60  * with a null layer.
61  *
62  * The remainder of this man page examines the null layer as a basis
63  * for constructing new layers.
64  *
65  *
66  * INSTANTIATING NEW NULL LAYERS
67  *
68  * New null layers are created with mount_nullfs(8).
69  * Mount_nullfs(8) takes two arguments, the pathname
70  * of the lower vfs (target-pn) and the pathname where the null
71  * layer will appear in the namespace (alias-pn).  After
72  * the null layer is put into place, the contents
73  * of target-pn subtree will be aliased under alias-pn.
74  *
75  *
76  * OPERATION OF A NULL LAYER
77  *
78  * The null layer is the minimum filesystem layer,
79  * simply bypassing all possible operations to the lower layer
80  * for processing there.  The majority of its activity centers
81  * on the bypass routine, through which nearly all vnode operations
82  * pass.
83  *
84  * The bypass routine accepts arbitrary vnode operations for
85  * handling by the lower layer.  It begins by examing vnode
86  * operation arguments and replacing any null-nodes by their
87  * lower-layer equivlants.  It then invokes the operation
88  * on the lower layer.  Finally, it replaces the null-nodes
89  * in the arguments and, if a vnode is return by the operation,
90  * stacks a null-node on top of the returned vnode.
91  *
92  * Although bypass handles most operations, vop_getattr, vop_lock,
93  * vop_unlock, vop_inactive, vop_reclaim, and vop_print are not
94  * bypassed. Vop_getattr must change the fsid being returned.
95  * Vop_lock and vop_unlock must handle any locking for the
96  * current vnode as well as pass the lock request down.
97  * Vop_inactive and vop_reclaim are not bypassed so that
98  * they can handle freeing null-layer specific data. Vop_print
99  * is not bypassed to avoid excessive debugging information.
100  * Also, certain vnode operations change the locking state within
101  * the operation (create, mknod, remove, link, rename, mkdir, rmdir,
102  * and symlink). Ideally these operations should not change the
103  * lock state, but should be changed to let the caller of the
104  * function unlock them. Otherwise all intermediate vnode layers
105  * (such as union, umapfs, etc) must catch these functions to do
106  * the necessary locking at their layer.
107  *
108  *
109  * INSTANTIATING VNODE STACKS
110  *
111  * Mounting associates the null layer with a lower layer,
112  * effect stacking two VFSes.  Vnode stacks are instead
113  * created on demand as files are accessed.
114  *
115  * The initial mount creates a single vnode stack for the
116  * root of the new null layer.  All other vnode stacks
117  * are created as a result of vnode operations on
118  * this or other null vnode stacks.
119  *
120  * New vnode stacks come into existance as a result of
121  * an operation which returns a vnode.
122  * The bypass routine stacks a null-node above the new
123  * vnode before returning it to the caller.
124  *
125  * For example, imagine mounting a null layer with
126  * "mount_nullfs /usr/include /dev/layer/null".
127  * Changing directory to /dev/layer/null will assign
128  * the root null-node (which was created when the null layer was mounted).
129  * Now consider opening "sys".  A vop_lookup would be
130  * done on the root null-node.  This operation would bypass through
131  * to the lower layer which would return a vnode representing
132  * the UFS "sys".  Null_bypass then builds a null-node
133  * aliasing the UFS "sys" and returns this to the caller.
134  * Later operations on the null-node "sys" will repeat this
135  * process when constructing other vnode stacks.
136  *
137  *
138  * CREATING OTHER FILE SYSTEM LAYERS
139  *
140  * One of the easiest ways to construct new filesystem layers is to make
141  * a copy of the null layer, rename all files and variables, and
142  * then begin modifing the copy.  Sed can be used to easily rename
143  * all variables.
144  *
145  * The umap layer is an example of a layer descended from the
146  * null layer.
147  *
148  *
149  * INVOKING OPERATIONS ON LOWER LAYERS
150  *
151  * There are two techniques to invoke operations on a lower layer
152  * when the operation cannot be completely bypassed.  Each method
153  * is appropriate in different situations.  In both cases,
154  * it is the responsibility of the aliasing layer to make
155  * the operation arguments "correct" for the lower layer
156  * by mapping a vnode arguments to the lower layer.
157  *
158  * The first approach is to call the aliasing layer's bypass routine.
159  * This method is most suitable when you wish to invoke the operation
160  * currently being handled on the lower layer.  It has the advantage
161  * that the bypass routine already must do argument mapping.
162  * An example of this is null_getattrs in the null layer.
163  *
164  * A second approach is to directly invoke vnode operations on
165  * the lower layer with the VOP_OPERATIONNAME interface.
166  * The advantage of this method is that it is easy to invoke
167  * arbitrary operations on the lower layer.  The disadvantage
168  * is that vnode arguments must be manualy mapped.
169  *
170  */
171 
172 #include <sys/param.h>
173 #include <sys/systm.h>
174 #include <sys/conf.h>
175 #include <sys/kernel.h>
176 #include <sys/lock.h>
177 #include <sys/malloc.h>
178 #include <sys/mount.h>
179 #include <sys/mutex.h>
180 #include <sys/namei.h>
181 #include <sys/sysctl.h>
182 #include <sys/vnode.h>
183 #include <sys/kdb.h>
184 
185 #include <fs/nullfs/null.h>
186 
187 #include <vm/vm.h>
188 #include <vm/vm_extern.h>
189 #include <vm/vm_object.h>
190 #include <vm/vnode_pager.h>
191 
192 static int null_bug_bypass = 0;   /* for debugging: enables bypass printf'ing */
193 SYSCTL_INT(_debug, OID_AUTO, nullfs_bug_bypass, CTLFLAG_RW,
194 	&null_bug_bypass, 0, "");
195 
196 /*
197  * This is the 10-Apr-92 bypass routine.
198  *    This version has been optimized for speed, throwing away some
199  * safety checks.  It should still always work, but it's not as
200  * robust to programmer errors.
201  *
202  * In general, we map all vnodes going down and unmap them on the way back.
203  * As an exception to this, vnodes can be marked "unmapped" by setting
204  * the Nth bit in operation's vdesc_flags.
205  *
206  * Also, some BSD vnode operations have the side effect of vrele'ing
207  * their arguments.  With stacking, the reference counts are held
208  * by the upper node, not the lower one, so we must handle these
209  * side-effects here.  This is not of concern in Sun-derived systems
210  * since there are no such side-effects.
211  *
212  * This makes the following assumptions:
213  * - only one returned vpp
214  * - no INOUT vpp's (Sun's vop_open has one of these)
215  * - the vnode operation vector of the first vnode should be used
216  *   to determine what implementation of the op should be invoked
217  * - all mapped vnodes are of our vnode-type (NEEDSWORK:
218  *   problems on rmdir'ing mount points and renaming?)
219  */
220 int
221 null_bypass(struct vop_generic_args *ap)
222 {
223 	struct vnode **this_vp_p;
224 	int error;
225 	struct vnode *old_vps[VDESC_MAX_VPS];
226 	struct vnode **vps_p[VDESC_MAX_VPS];
227 	struct vnode ***vppp;
228 	struct vnodeop_desc *descp = ap->a_desc;
229 	int reles, i;
230 
231 	if (null_bug_bypass)
232 		printf ("null_bypass: %s\n", descp->vdesc_name);
233 
234 #ifdef DIAGNOSTIC
235 	/*
236 	 * We require at least one vp.
237 	 */
238 	if (descp->vdesc_vp_offsets == NULL ||
239 	    descp->vdesc_vp_offsets[0] == VDESC_NO_OFFSET)
240 		panic ("null_bypass: no vp's in map");
241 #endif
242 
243 	/*
244 	 * Map the vnodes going in.
245 	 * Later, we'll invoke the operation based on
246 	 * the first mapped vnode's operation vector.
247 	 */
248 	reles = descp->vdesc_flags;
249 	for (i = 0; i < VDESC_MAX_VPS; reles >>= 1, i++) {
250 		if (descp->vdesc_vp_offsets[i] == VDESC_NO_OFFSET)
251 			break;   /* bail out at end of list */
252 		vps_p[i] = this_vp_p =
253 			VOPARG_OFFSETTO(struct vnode**,descp->vdesc_vp_offsets[i],ap);
254 		/*
255 		 * We're not guaranteed that any but the first vnode
256 		 * are of our type.  Check for and don't map any
257 		 * that aren't.  (We must always map first vp or vclean fails.)
258 		 */
259 		if (i && (*this_vp_p == NULLVP ||
260 		    (*this_vp_p)->v_op != &null_vnodeops)) {
261 			old_vps[i] = NULLVP;
262 		} else {
263 			old_vps[i] = *this_vp_p;
264 			*(vps_p[i]) = NULLVPTOLOWERVP(*this_vp_p);
265 			/*
266 			 * XXX - Several operations have the side effect
267 			 * of vrele'ing their vp's.  We must account for
268 			 * that.  (This should go away in the future.)
269 			 */
270 			if (reles & VDESC_VP0_WILLRELE)
271 				VREF(*this_vp_p);
272 		}
273 
274 	}
275 
276 	/*
277 	 * Call the operation on the lower layer
278 	 * with the modified argument structure.
279 	 */
280 	if (vps_p[0] && *vps_p[0])
281 		error = VCALL(ap);
282 	else {
283 		printf("null_bypass: no map for %s\n", descp->vdesc_name);
284 		error = EINVAL;
285 	}
286 
287 	/*
288 	 * Maintain the illusion of call-by-value
289 	 * by restoring vnodes in the argument structure
290 	 * to their original value.
291 	 */
292 	reles = descp->vdesc_flags;
293 	for (i = 0; i < VDESC_MAX_VPS; reles >>= 1, i++) {
294 		if (descp->vdesc_vp_offsets[i] == VDESC_NO_OFFSET)
295 			break;   /* bail out at end of list */
296 		if (old_vps[i]) {
297 			*(vps_p[i]) = old_vps[i];
298 #if 0
299 			if (reles & VDESC_VP0_WILLUNLOCK)
300 				VOP_UNLOCK(*(vps_p[i]), 0, curthread);
301 #endif
302 			if (reles & VDESC_VP0_WILLRELE)
303 				vrele(*(vps_p[i]));
304 		}
305 	}
306 
307 	/*
308 	 * Map the possible out-going vpp
309 	 * (Assumes that the lower layer always returns
310 	 * a VREF'ed vpp unless it gets an error.)
311 	 */
312 	if (descp->vdesc_vpp_offset != VDESC_NO_OFFSET &&
313 	    !(descp->vdesc_flags & VDESC_NOMAP_VPP) &&
314 	    !error) {
315 		/*
316 		 * XXX - even though some ops have vpp returned vp's,
317 		 * several ops actually vrele this before returning.
318 		 * We must avoid these ops.
319 		 * (This should go away when these ops are regularized.)
320 		 */
321 		if (descp->vdesc_flags & VDESC_VPP_WILLRELE)
322 			goto out;
323 		vppp = VOPARG_OFFSETTO(struct vnode***,
324 				 descp->vdesc_vpp_offset,ap);
325 		if (*vppp)
326 			error = null_nodeget(old_vps[0]->v_mount, **vppp, *vppp);
327 	}
328 
329  out:
330 	return (error);
331 }
332 
333 /*
334  * We have to carry on the locking protocol on the null layer vnodes
335  * as we progress through the tree. We also have to enforce read-only
336  * if this layer is mounted read-only.
337  */
338 static int
339 null_lookup(struct vop_lookup_args *ap)
340 {
341 	struct componentname *cnp = ap->a_cnp;
342 	struct vnode *dvp = ap->a_dvp;
343 	int flags = cnp->cn_flags;
344 	struct vnode *vp, *ldvp, *lvp;
345 	int error;
346 
347 	if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
348 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
349 		return (EROFS);
350 	/*
351 	 * Although it is possible to call null_bypass(), we'll do
352 	 * a direct call to reduce overhead
353 	 */
354 	ldvp = NULLVPTOLOWERVP(dvp);
355 	vp = lvp = NULL;
356 	error = VOP_LOOKUP(ldvp, &lvp, cnp);
357 	if (error == EJUSTRETURN && (flags & ISLASTCN) &&
358 	    (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
359 	    (cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME))
360 		error = EROFS;
361 
362 	if ((error == 0 || error == EJUSTRETURN) && lvp != NULL) {
363 		if (ldvp == lvp) {
364 			*ap->a_vpp = dvp;
365 			VREF(dvp);
366 			vrele(lvp);
367 		} else {
368 			error = null_nodeget(dvp->v_mount, lvp, &vp);
369 			if (error) {
370 				/* XXX Cleanup needed... */
371 				panic("null_nodeget failed");
372 			}
373 			*ap->a_vpp = vp;
374 		}
375 	}
376 	return (error);
377 }
378 
379 static int
380 null_open(struct vop_open_args *ap)
381 {
382 	int retval;
383 	struct vnode *vp, *ldvp;
384 
385 	vp = ap->a_vp;
386 	ldvp = NULLVPTOLOWERVP(vp);
387 	retval = null_bypass(&ap->a_gen);
388 	if (retval == 0)
389 		vp->v_object = ldvp->v_object;
390 	return (retval);
391 }
392 
393 /*
394  * Setattr call. Disallow write attempts if the layer is mounted read-only.
395  */
396 static int
397 null_setattr(struct vop_setattr_args *ap)
398 {
399 	struct vnode *vp = ap->a_vp;
400 	struct vattr *vap = ap->a_vap;
401 
402   	if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
403 	    vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
404 	    vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
405 	    (vp->v_mount->mnt_flag & MNT_RDONLY))
406 		return (EROFS);
407 	if (vap->va_size != VNOVAL) {
408  		switch (vp->v_type) {
409  		case VDIR:
410  			return (EISDIR);
411  		case VCHR:
412  		case VBLK:
413  		case VSOCK:
414  		case VFIFO:
415 			if (vap->va_flags != VNOVAL)
416 				return (EOPNOTSUPP);
417 			return (0);
418 		case VREG:
419 		case VLNK:
420  		default:
421 			/*
422 			 * Disallow write attempts if the filesystem is
423 			 * mounted read-only.
424 			 */
425 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
426 				return (EROFS);
427 		}
428 	}
429 
430 	return (null_bypass((struct vop_generic_args *)ap));
431 }
432 
433 /*
434  *  We handle getattr only to change the fsid.
435  */
436 static int
437 null_getattr(struct vop_getattr_args *ap)
438 {
439 	int error;
440 
441 	if ((error = null_bypass((struct vop_generic_args *)ap)) != 0)
442 		return (error);
443 
444 	ap->a_vap->va_fsid = ap->a_vp->v_mount->mnt_stat.f_fsid.val[0];
445 	return (0);
446 }
447 
448 /*
449  * Handle to disallow write access if mounted read-only.
450  */
451 static int
452 null_access(struct vop_access_args *ap)
453 {
454 	struct vnode *vp = ap->a_vp;
455 	mode_t mode = ap->a_mode;
456 
457 	/*
458 	 * Disallow write attempts on read-only layers;
459 	 * unless the file is a socket, fifo, or a block or
460 	 * character device resident on the filesystem.
461 	 */
462 	if (mode & VWRITE) {
463 		switch (vp->v_type) {
464 		case VDIR:
465 		case VLNK:
466 		case VREG:
467 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
468 				return (EROFS);
469 			break;
470 		default:
471 			break;
472 		}
473 	}
474 	return (null_bypass((struct vop_generic_args *)ap));
475 }
476 
477 /*
478  * We handle this to eliminate null FS to lower FS
479  * file moving. Don't know why we don't allow this,
480  * possibly we should.
481  */
482 static int
483 null_rename(struct vop_rename_args *ap)
484 {
485 	struct vnode *tdvp = ap->a_tdvp;
486 	struct vnode *fvp = ap->a_fvp;
487 	struct vnode *fdvp = ap->a_fdvp;
488 	struct vnode *tvp = ap->a_tvp;
489 
490 	/* Check for cross-device rename. */
491 	if ((fvp->v_mount != tdvp->v_mount) ||
492 	    (tvp && (fvp->v_mount != tvp->v_mount))) {
493 		if (tdvp == tvp)
494 			vrele(tdvp);
495 		else
496 			vput(tdvp);
497 		if (tvp)
498 			vput(tvp);
499 		vrele(fdvp);
500 		vrele(fvp);
501 		return (EXDEV);
502 	}
503 
504 	return (null_bypass((struct vop_generic_args *)ap));
505 }
506 
507 /*
508  * We need to process our own vnode lock and then clear the
509  * interlock flag as it applies only to our vnode, not the
510  * vnodes below us on the stack.
511  */
512 static int
513 null_lock(struct _vop_lock_args *ap)
514 {
515 	struct vnode *vp = ap->a_vp;
516 	int flags = ap->a_flags;
517 	struct thread *td = ap->a_td;
518 	struct null_node *nn;
519 	struct vnode *lvp;
520 	int error;
521 
522 
523 	if ((flags & LK_INTERLOCK) == 0) {
524 		VI_LOCK(vp);
525 		ap->a_flags = flags |= LK_INTERLOCK;
526 	}
527 	nn = VTONULL(vp);
528 	/*
529 	 * If we're still active we must ask the lower layer to
530 	 * lock as ffs has special lock considerations in it's
531 	 * vop lock.
532 	 */
533 	if (nn != NULL && (lvp = NULLVPTOLOWERVP(vp)) != NULL) {
534 		VI_LOCK_FLAGS(lvp, MTX_DUPOK);
535 		VI_UNLOCK(vp);
536 		/*
537 		 * We have to hold the vnode here to solve a potential
538 		 * reclaim race.  If we're forcibly vgone'd while we
539 		 * still have refs, a thread could be sleeping inside
540 		 * the lowervp's vop_lock routine.  When we vgone we will
541 		 * drop our last ref to the lowervp, which would allow it
542 		 * to be reclaimed.  The lowervp could then be recycled,
543 		 * in which case it is not legal to be sleeping in it's VOP.
544 		 * We prevent it from being recycled by holding the vnode
545 		 * here.
546 		 */
547 		vholdl(lvp);
548 		error = VOP_LOCK(lvp, flags, td);
549 
550 		/*
551 		 * We might have slept to get the lock and someone might have
552 		 * clean our vnode already, switching vnode lock from one in
553 		 * lowervp to v_lock in our own vnode structure.  Handle this
554 		 * case by reacquiring correct lock in requested mode.
555 		 */
556 		if (VTONULL(vp) == NULL && error == 0) {
557 			ap->a_flags &= ~(LK_TYPE_MASK | LK_INTERLOCK);
558 			switch (flags & LK_TYPE_MASK) {
559 			case LK_SHARED:
560 				ap->a_flags |= LK_SHARED;
561 				break;
562 			case LK_UPGRADE:
563 			case LK_EXCLUSIVE:
564 				ap->a_flags |= LK_EXCLUSIVE;
565 				break;
566 			default:
567 				panic("Unsupported lock request %d\n",
568 				    ap->a_flags);
569 			}
570 			VOP_UNLOCK(lvp, 0, td);
571 			error = vop_stdlock(ap);
572 		}
573 		vdrop(lvp);
574 	} else
575 		error = vop_stdlock(ap);
576 
577 	return (error);
578 }
579 
580 /*
581  * We need to process our own vnode unlock and then clear the
582  * interlock flag as it applies only to our vnode, not the
583  * vnodes below us on the stack.
584  */
585 static int
586 null_unlock(struct vop_unlock_args *ap)
587 {
588 	struct vnode *vp = ap->a_vp;
589 	int flags = ap->a_flags;
590 	struct thread *td = ap->a_td;
591 	struct null_node *nn;
592 	struct vnode *lvp;
593 	int error;
594 
595 	if ((flags & LK_INTERLOCK) != 0) {
596 		VI_UNLOCK(vp);
597 		ap->a_flags = flags &= ~LK_INTERLOCK;
598 	}
599 	nn = VTONULL(vp);
600 	if (nn != NULL && (lvp = NULLVPTOLOWERVP(vp)) != NULL)
601 		error = VOP_UNLOCK(lvp, flags, td);
602 	else
603 		error = vop_stdunlock(ap);
604 
605 	return (error);
606 }
607 
608 static int
609 null_islocked(struct vop_islocked_args *ap)
610 {
611 	struct vnode *vp = ap->a_vp;
612 	struct thread *td = ap->a_td;
613 
614 	return (lockstatus(vp->v_vnlock, td));
615 }
616 
617 /*
618  * There is no way to tell that someone issued remove/rmdir operation
619  * on the underlying filesystem. For now we just have to release lowevrp
620  * as soon as possible.
621  *
622  * Note, we can't release any resources nor remove vnode from hash before
623  * appropriate VXLOCK stuff is is done because other process can find this
624  * vnode in hash during inactivation and may be sitting in vget() and waiting
625  * for null_inactive to unlock vnode. Thus we will do all those in VOP_RECLAIM.
626  */
627 static int
628 null_inactive(struct vop_inactive_args *ap)
629 {
630 	struct vnode *vp = ap->a_vp;
631 	struct thread *td = ap->a_td;
632 
633 	vp->v_object = NULL;
634 
635 	/*
636 	 * If this is the last reference, then free up the vnode
637 	 * so as not to tie up the lower vnodes.
638 	 */
639 	vrecycle(vp, td);
640 
641 	return (0);
642 }
643 
644 /*
645  * Now, the VXLOCK is in force and we're free to destroy the null vnode.
646  */
647 static int
648 null_reclaim(struct vop_reclaim_args *ap)
649 {
650 	struct vnode *vp = ap->a_vp;
651 	struct null_node *xp = VTONULL(vp);
652 	struct vnode *lowervp = xp->null_lowervp;
653 	struct lock *vnlock;
654 
655 	if (lowervp)
656 		null_hashrem(xp);
657 	/*
658 	 * Use the interlock to protect the clearing of v_data to
659 	 * prevent faults in null_lock().
660 	 */
661 	VI_LOCK(vp);
662 	vp->v_data = NULL;
663 	vp->v_object = NULL;
664 	vnlock = vp->v_vnlock;
665 	vp->v_vnlock = &vp->v_lock;
666 	if (lowervp) {
667 		lockmgr(vp->v_vnlock,
668 		    LK_EXCLUSIVE|LK_INTERLOCK, VI_MTX(vp), curthread);
669 		vput(lowervp);
670 	} else
671 		panic("null_reclaim: reclaiming an node with now lowervp");
672 	FREE(xp, M_NULLFSNODE);
673 
674 	return (0);
675 }
676 
677 static int
678 null_print(struct vop_print_args *ap)
679 {
680 	struct vnode *vp = ap->a_vp;
681 
682 	printf("\tvp=%p, lowervp=%p\n", vp, NULLVPTOLOWERVP(vp));
683 	return (0);
684 }
685 
686 /* ARGSUSED */
687 static int
688 null_getwritemount(struct vop_getwritemount_args *ap)
689 {
690 	struct null_node *xp;
691 	struct vnode *lowervp;
692 	struct vnode *vp;
693 
694 	vp = ap->a_vp;
695 	VI_LOCK(vp);
696 	xp = VTONULL(vp);
697 	if (xp && (lowervp = xp->null_lowervp)) {
698 		VI_LOCK_FLAGS(lowervp, MTX_DUPOK);
699 		VI_UNLOCK(vp);
700 		vholdl(lowervp);
701 		VI_UNLOCK(lowervp);
702 		VOP_GETWRITEMOUNT(lowervp, ap->a_mpp);
703 		vdrop(lowervp);
704 	} else {
705 		VI_UNLOCK(vp);
706 		*(ap->a_mpp) = NULL;
707 	}
708 	return (0);
709 }
710 
711 static int
712 null_vptofh(struct vop_vptofh_args *ap)
713 {
714 	struct vnode *lvp;
715 
716 	lvp = NULLVPTOLOWERVP(ap->a_vp);
717 	return VOP_VPTOFH(lvp, ap->a_fhp);
718 }
719 
720 /*
721  * Global vfs data structures
722  */
723 struct vop_vector null_vnodeops = {
724 	.vop_bypass =		null_bypass,
725 	.vop_access =		null_access,
726 	.vop_bmap =		VOP_EOPNOTSUPP,
727 	.vop_getattr =		null_getattr,
728 	.vop_getwritemount =	null_getwritemount,
729 	.vop_inactive =		null_inactive,
730 	.vop_islocked =		null_islocked,
731 	._vop_lock =		null_lock,
732 	.vop_lookup =		null_lookup,
733 	.vop_open =		null_open,
734 	.vop_print =		null_print,
735 	.vop_reclaim =		null_reclaim,
736 	.vop_rename =		null_rename,
737 	.vop_setattr =		null_setattr,
738 	.vop_strategy =		VOP_EOPNOTSUPP,
739 	.vop_unlock =		null_unlock,
740 	.vop_vptofh =		null_vptofh,
741 };
742