xref: /freebsd/sys/fs/nullfs/null_vnops.c (revision a10cee30c94cf5944826d2a495e9cdf339dfbcc8)
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 
184 #include <fs/nullfs/null.h>
185 
186 #include <vm/vm.h>
187 #include <vm/vm_extern.h>
188 #include <vm/vm_object.h>
189 #include <vm/vnode_pager.h>
190 
191 static int null_bug_bypass = 0;   /* for debugging: enables bypass printf'ing */
192 SYSCTL_INT(_debug, OID_AUTO, nullfs_bug_bypass, CTLFLAG_RW,
193 	&null_bug_bypass, 0, "");
194 
195 /*
196  * This is the 10-Apr-92 bypass routine.
197  *    This version has been optimized for speed, throwing away some
198  * safety checks.  It should still always work, but it's not as
199  * robust to programmer errors.
200  *
201  * In general, we map all vnodes going down and unmap them on the way back.
202  * As an exception to this, vnodes can be marked "unmapped" by setting
203  * the Nth bit in operation's vdesc_flags.
204  *
205  * Also, some BSD vnode operations have the side effect of vrele'ing
206  * their arguments.  With stacking, the reference counts are held
207  * by the upper node, not the lower one, so we must handle these
208  * side-effects here.  This is not of concern in Sun-derived systems
209  * since there are no such side-effects.
210  *
211  * This makes the following assumptions:
212  * - only one returned vpp
213  * - no INOUT vpp's (Sun's vop_open has one of these)
214  * - the vnode operation vector of the first vnode should be used
215  *   to determine what implementation of the op should be invoked
216  * - all mapped vnodes are of our vnode-type (NEEDSWORK:
217  *   problems on rmdir'ing mount points and renaming?)
218  */
219 int
220 null_bypass(struct vop_generic_args *ap)
221 {
222 	struct vnode **this_vp_p;
223 	int error;
224 	struct vnode *old_vps[VDESC_MAX_VPS];
225 	struct vnode **vps_p[VDESC_MAX_VPS];
226 	struct vnode ***vppp;
227 	struct vnodeop_desc *descp = ap->a_desc;
228 	int reles, i;
229 
230 	if (null_bug_bypass)
231 		printf ("null_bypass: %s\n", descp->vdesc_name);
232 
233 #ifdef DIAGNOSTIC
234 	/*
235 	 * We require at least one vp.
236 	 */
237 	if (descp->vdesc_vp_offsets == NULL ||
238 	    descp->vdesc_vp_offsets[0] == VDESC_NO_OFFSET)
239 		panic ("null_bypass: no vp's in map");
240 #endif
241 
242 	/*
243 	 * Map the vnodes going in.
244 	 * Later, we'll invoke the operation based on
245 	 * the first mapped vnode's operation vector.
246 	 */
247 	reles = descp->vdesc_flags;
248 	for (i = 0; i < VDESC_MAX_VPS; reles >>= 1, i++) {
249 		if (descp->vdesc_vp_offsets[i] == VDESC_NO_OFFSET)
250 			break;   /* bail out at end of list */
251 		vps_p[i] = this_vp_p =
252 			VOPARG_OFFSETTO(struct vnode**,descp->vdesc_vp_offsets[i],ap);
253 		/*
254 		 * We're not guaranteed that any but the first vnode
255 		 * are of our type.  Check for and don't map any
256 		 * that aren't.  (We must always map first vp or vclean fails.)
257 		 */
258 		if (i && (*this_vp_p == NULLVP ||
259 		    (*this_vp_p)->v_op != &null_vnodeops)) {
260 			old_vps[i] = NULLVP;
261 		} else {
262 			old_vps[i] = *this_vp_p;
263 			*(vps_p[i]) = NULLVPTOLOWERVP(*this_vp_p);
264 			/*
265 			 * XXX - Several operations have the side effect
266 			 * of vrele'ing their vp's.  We must account for
267 			 * that.  (This should go away in the future.)
268 			 */
269 			if (reles & VDESC_VP0_WILLRELE)
270 				VREF(*this_vp_p);
271 		}
272 
273 	}
274 
275 	/*
276 	 * Call the operation on the lower layer
277 	 * with the modified argument structure.
278 	 */
279 	if (vps_p[0] && *vps_p[0])
280 		error = VCALL(ap);
281 	else {
282 		printf("null_bypass: no map for %s\n", descp->vdesc_name);
283 		error = EINVAL;
284 	}
285 
286 	/*
287 	 * Maintain the illusion of call-by-value
288 	 * by restoring vnodes in the argument structure
289 	 * to their original value.
290 	 */
291 	reles = descp->vdesc_flags;
292 	for (i = 0; i < VDESC_MAX_VPS; reles >>= 1, i++) {
293 		if (descp->vdesc_vp_offsets[i] == VDESC_NO_OFFSET)
294 			break;   /* bail out at end of list */
295 		if (old_vps[i]) {
296 			*(vps_p[i]) = old_vps[i];
297 #if 0
298 			if (reles & VDESC_VP0_WILLUNLOCK)
299 				VOP_UNLOCK(*(vps_p[i]), 0);
300 #endif
301 			if (reles & VDESC_VP0_WILLRELE)
302 				vrele(*(vps_p[i]));
303 		}
304 	}
305 
306 	/*
307 	 * Map the possible out-going vpp
308 	 * (Assumes that the lower layer always returns
309 	 * a VREF'ed vpp unless it gets an error.)
310 	 */
311 	if (descp->vdesc_vpp_offset != VDESC_NO_OFFSET &&
312 	    !(descp->vdesc_flags & VDESC_NOMAP_VPP) &&
313 	    !error) {
314 		/*
315 		 * XXX - even though some ops have vpp returned vp's,
316 		 * several ops actually vrele this before returning.
317 		 * We must avoid these ops.
318 		 * (This should go away when these ops are regularized.)
319 		 */
320 		if (descp->vdesc_flags & VDESC_VPP_WILLRELE)
321 			goto out;
322 		vppp = VOPARG_OFFSETTO(struct vnode***,
323 				 descp->vdesc_vpp_offset,ap);
324 		if (*vppp)
325 			error = null_nodeget(old_vps[0]->v_mount, **vppp, *vppp);
326 	}
327 
328  out:
329 	return (error);
330 }
331 
332 /*
333  * We have to carry on the locking protocol on the null layer vnodes
334  * as we progress through the tree. We also have to enforce read-only
335  * if this layer is mounted read-only.
336  */
337 static int
338 null_lookup(struct vop_lookup_args *ap)
339 {
340 	struct componentname *cnp = ap->a_cnp;
341 	struct vnode *dvp = ap->a_dvp;
342 	int flags = cnp->cn_flags;
343 	struct vnode *vp, *ldvp, *lvp;
344 	int error;
345 
346 	if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
347 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
348 		return (EROFS);
349 	/*
350 	 * Although it is possible to call null_bypass(), we'll do
351 	 * a direct call to reduce overhead
352 	 */
353 	ldvp = NULLVPTOLOWERVP(dvp);
354 	vp = lvp = NULL;
355 	error = VOP_LOOKUP(ldvp, &lvp, cnp);
356 	if (error == EJUSTRETURN && (flags & ISLASTCN) &&
357 	    (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
358 	    (cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME))
359 		error = EROFS;
360 
361 	if ((error == 0 || error == EJUSTRETURN) && lvp != NULL) {
362 		if (ldvp == lvp) {
363 			*ap->a_vpp = dvp;
364 			VREF(dvp);
365 			vrele(lvp);
366 		} else {
367 			error = null_nodeget(dvp->v_mount, lvp, &vp);
368 			if (error == 0)
369 				*ap->a_vpp = vp;
370 		}
371 	}
372 	return (error);
373 }
374 
375 static int
376 null_open(struct vop_open_args *ap)
377 {
378 	int retval;
379 	struct vnode *vp, *ldvp;
380 
381 	vp = ap->a_vp;
382 	ldvp = NULLVPTOLOWERVP(vp);
383 	retval = null_bypass(&ap->a_gen);
384 	if (retval == 0)
385 		vp->v_object = ldvp->v_object;
386 	return (retval);
387 }
388 
389 /*
390  * Setattr call. Disallow write attempts if the layer is mounted read-only.
391  */
392 static int
393 null_setattr(struct vop_setattr_args *ap)
394 {
395 	struct vnode *vp = ap->a_vp;
396 	struct vattr *vap = ap->a_vap;
397 
398   	if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
399 	    vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
400 	    vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
401 	    (vp->v_mount->mnt_flag & MNT_RDONLY))
402 		return (EROFS);
403 	if (vap->va_size != VNOVAL) {
404  		switch (vp->v_type) {
405  		case VDIR:
406  			return (EISDIR);
407  		case VCHR:
408  		case VBLK:
409  		case VSOCK:
410  		case VFIFO:
411 			if (vap->va_flags != VNOVAL)
412 				return (EOPNOTSUPP);
413 			return (0);
414 		case VREG:
415 		case VLNK:
416  		default:
417 			/*
418 			 * Disallow write attempts if the filesystem is
419 			 * mounted read-only.
420 			 */
421 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
422 				return (EROFS);
423 		}
424 	}
425 
426 	return (null_bypass((struct vop_generic_args *)ap));
427 }
428 
429 /*
430  *  We handle getattr only to change the fsid.
431  */
432 static int
433 null_getattr(struct vop_getattr_args *ap)
434 {
435 	int error;
436 
437 	if ((error = null_bypass((struct vop_generic_args *)ap)) != 0)
438 		return (error);
439 
440 	ap->a_vap->va_fsid = ap->a_vp->v_mount->mnt_stat.f_fsid.val[0];
441 	return (0);
442 }
443 
444 /*
445  * Handle to disallow write access if mounted read-only.
446  */
447 static int
448 null_access(struct vop_access_args *ap)
449 {
450 	struct vnode *vp = ap->a_vp;
451 	accmode_t accmode = ap->a_accmode;
452 
453 	/*
454 	 * Disallow write attempts on read-only layers;
455 	 * unless the file is a socket, fifo, or a block or
456 	 * character device resident on the filesystem.
457 	 */
458 	if (accmode & VWRITE) {
459 		switch (vp->v_type) {
460 		case VDIR:
461 		case VLNK:
462 		case VREG:
463 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
464 				return (EROFS);
465 			break;
466 		default:
467 			break;
468 		}
469 	}
470 	return (null_bypass((struct vop_generic_args *)ap));
471 }
472 
473 static int
474 null_accessx(struct vop_accessx_args *ap)
475 {
476 	struct vnode *vp = ap->a_vp;
477 	accmode_t accmode = ap->a_accmode;
478 
479 	/*
480 	 * Disallow write attempts on read-only layers;
481 	 * unless the file is a socket, fifo, or a block or
482 	 * character device resident on the filesystem.
483 	 */
484 	if (accmode & VWRITE) {
485 		switch (vp->v_type) {
486 		case VDIR:
487 		case VLNK:
488 		case VREG:
489 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
490 				return (EROFS);
491 			break;
492 		default:
493 			break;
494 		}
495 	}
496 	return (null_bypass((struct vop_generic_args *)ap));
497 }
498 
499 /*
500  * Increasing refcount of lower vnode is needed at least for the case
501  * when lower FS is NFS to do sillyrename if the file is in use.
502  * Unfortunately v_usecount is incremented in many places in
503  * the kernel and, as such, there may be races that result in
504  * the NFS client doing an extraneous silly rename, but that seems
505  * preferable to not doing a silly rename when it is needed.
506  */
507 static int
508 null_remove(struct vop_remove_args *ap)
509 {
510 	int retval, vreleit;
511 	struct vnode *lvp;
512 
513 	if (vrefcnt(ap->a_vp) > 1) {
514 		lvp = NULLVPTOLOWERVP(ap->a_vp);
515 		VREF(lvp);
516 		vreleit = 1;
517 	} else
518 		vreleit = 0;
519 	retval = null_bypass(&ap->a_gen);
520 	if (vreleit != 0)
521 		vrele(lvp);
522 	return (retval);
523 }
524 
525 /*
526  * We handle this to eliminate null FS to lower FS
527  * file moving. Don't know why we don't allow this,
528  * possibly we should.
529  */
530 static int
531 null_rename(struct vop_rename_args *ap)
532 {
533 	struct vnode *tdvp = ap->a_tdvp;
534 	struct vnode *fvp = ap->a_fvp;
535 	struct vnode *fdvp = ap->a_fdvp;
536 	struct vnode *tvp = ap->a_tvp;
537 
538 	/* Check for cross-device rename. */
539 	if ((fvp->v_mount != tdvp->v_mount) ||
540 	    (tvp && (fvp->v_mount != tvp->v_mount))) {
541 		if (tdvp == tvp)
542 			vrele(tdvp);
543 		else
544 			vput(tdvp);
545 		if (tvp)
546 			vput(tvp);
547 		vrele(fdvp);
548 		vrele(fvp);
549 		return (EXDEV);
550 	}
551 
552 	return (null_bypass((struct vop_generic_args *)ap));
553 }
554 
555 /*
556  * We need to process our own vnode lock and then clear the
557  * interlock flag as it applies only to our vnode, not the
558  * vnodes below us on the stack.
559  */
560 static int
561 null_lock(struct vop_lock1_args *ap)
562 {
563 	struct vnode *vp = ap->a_vp;
564 	int flags = ap->a_flags;
565 	struct null_node *nn;
566 	struct vnode *lvp;
567 	int error;
568 
569 
570 	if ((flags & LK_INTERLOCK) == 0) {
571 		VI_LOCK(vp);
572 		ap->a_flags = flags |= LK_INTERLOCK;
573 	}
574 	nn = VTONULL(vp);
575 	/*
576 	 * If we're still active we must ask the lower layer to
577 	 * lock as ffs has special lock considerations in it's
578 	 * vop lock.
579 	 */
580 	if (nn != NULL && (lvp = NULLVPTOLOWERVP(vp)) != NULL) {
581 		VI_LOCK_FLAGS(lvp, MTX_DUPOK);
582 		VI_UNLOCK(vp);
583 		/*
584 		 * We have to hold the vnode here to solve a potential
585 		 * reclaim race.  If we're forcibly vgone'd while we
586 		 * still have refs, a thread could be sleeping inside
587 		 * the lowervp's vop_lock routine.  When we vgone we will
588 		 * drop our last ref to the lowervp, which would allow it
589 		 * to be reclaimed.  The lowervp could then be recycled,
590 		 * in which case it is not legal to be sleeping in it's VOP.
591 		 * We prevent it from being recycled by holding the vnode
592 		 * here.
593 		 */
594 		vholdl(lvp);
595 		error = VOP_LOCK(lvp, flags);
596 
597 		/*
598 		 * We might have slept to get the lock and someone might have
599 		 * clean our vnode already, switching vnode lock from one in
600 		 * lowervp to v_lock in our own vnode structure.  Handle this
601 		 * case by reacquiring correct lock in requested mode.
602 		 */
603 		if (VTONULL(vp) == NULL && error == 0) {
604 			ap->a_flags &= ~(LK_TYPE_MASK | LK_INTERLOCK);
605 			switch (flags & LK_TYPE_MASK) {
606 			case LK_SHARED:
607 				ap->a_flags |= LK_SHARED;
608 				break;
609 			case LK_UPGRADE:
610 			case LK_EXCLUSIVE:
611 				ap->a_flags |= LK_EXCLUSIVE;
612 				break;
613 			default:
614 				panic("Unsupported lock request %d\n",
615 				    ap->a_flags);
616 			}
617 			VOP_UNLOCK(lvp, 0);
618 			error = vop_stdlock(ap);
619 		}
620 		vdrop(lvp);
621 	} else
622 		error = vop_stdlock(ap);
623 
624 	return (error);
625 }
626 
627 /*
628  * We need to process our own vnode unlock and then clear the
629  * interlock flag as it applies only to our vnode, not the
630  * vnodes below us on the stack.
631  */
632 static int
633 null_unlock(struct vop_unlock_args *ap)
634 {
635 	struct vnode *vp = ap->a_vp;
636 	int flags = ap->a_flags;
637 	int mtxlkflag = 0;
638 	struct null_node *nn;
639 	struct vnode *lvp;
640 	int error;
641 
642 	if ((flags & LK_INTERLOCK) != 0)
643 		mtxlkflag = 1;
644 	else if (mtx_owned(VI_MTX(vp)) == 0) {
645 		VI_LOCK(vp);
646 		mtxlkflag = 2;
647 	}
648 	nn = VTONULL(vp);
649 	if (nn != NULL && (lvp = NULLVPTOLOWERVP(vp)) != NULL) {
650 		VI_LOCK_FLAGS(lvp, MTX_DUPOK);
651 		flags |= LK_INTERLOCK;
652 		vholdl(lvp);
653 		VI_UNLOCK(vp);
654 		error = VOP_UNLOCK(lvp, flags);
655 		vdrop(lvp);
656 		if (mtxlkflag == 0)
657 			VI_LOCK(vp);
658 	} else {
659 		if (mtxlkflag == 2)
660 			VI_UNLOCK(vp);
661 		error = vop_stdunlock(ap);
662 	}
663 
664 	return (error);
665 }
666 
667 /*
668  * XXXKIB
669  */
670 static int
671 null_inactive(struct vop_inactive_args *ap __unused)
672 {
673 
674 	return (0);
675 }
676 
677 /*
678  * Now, the nullfs vnode and, due to the sharing lock, the lower
679  * vnode, are exclusively locked, and we shall destroy the null vnode.
680  */
681 static int
682 null_reclaim(struct vop_reclaim_args *ap)
683 {
684 	struct vnode *vp;
685 	struct null_node *xp;
686 	struct vnode *lowervp;
687 
688 	vp = ap->a_vp;
689 	xp = VTONULL(vp);
690 	lowervp = xp->null_lowervp;
691 
692 	KASSERT(lowervp != NULL && vp->v_vnlock != &vp->v_lock,
693 	    ("Reclaiming inclomplete null vnode %p", vp));
694 
695 	null_hashrem(xp);
696 	/*
697 	 * Use the interlock to protect the clearing of v_data to
698 	 * prevent faults in null_lock().
699 	 */
700 	lockmgr(&vp->v_lock, LK_EXCLUSIVE, NULL);
701 	VI_LOCK(vp);
702 	vp->v_data = NULL;
703 	vp->v_object = NULL;
704 	vp->v_vnlock = &vp->v_lock;
705 	VI_UNLOCK(vp);
706 	vput(lowervp);
707 	free(xp, M_NULLFSNODE);
708 
709 	return (0);
710 }
711 
712 static int
713 null_print(struct vop_print_args *ap)
714 {
715 	struct vnode *vp = ap->a_vp;
716 
717 	printf("\tvp=%p, lowervp=%p\n", vp, VTONULL(vp)->null_lowervp);
718 	return (0);
719 }
720 
721 /* ARGSUSED */
722 static int
723 null_getwritemount(struct vop_getwritemount_args *ap)
724 {
725 	struct null_node *xp;
726 	struct vnode *lowervp;
727 	struct vnode *vp;
728 
729 	vp = ap->a_vp;
730 	VI_LOCK(vp);
731 	xp = VTONULL(vp);
732 	if (xp && (lowervp = xp->null_lowervp)) {
733 		VI_LOCK_FLAGS(lowervp, MTX_DUPOK);
734 		VI_UNLOCK(vp);
735 		vholdl(lowervp);
736 		VI_UNLOCK(lowervp);
737 		VOP_GETWRITEMOUNT(lowervp, ap->a_mpp);
738 		vdrop(lowervp);
739 	} else {
740 		VI_UNLOCK(vp);
741 		*(ap->a_mpp) = NULL;
742 	}
743 	return (0);
744 }
745 
746 static int
747 null_vptofh(struct vop_vptofh_args *ap)
748 {
749 	struct vnode *lvp;
750 
751 	lvp = NULLVPTOLOWERVP(ap->a_vp);
752 	return VOP_VPTOFH(lvp, ap->a_fhp);
753 }
754 
755 static int
756 null_vptocnp(struct vop_vptocnp_args *ap)
757 {
758 	struct vnode *vp = ap->a_vp;
759 	struct vnode **dvp = ap->a_vpp;
760 	struct vnode *lvp, *ldvp;
761 	struct ucred *cred = ap->a_cred;
762 	int error, locked;
763 
764 	if (vp->v_type == VDIR)
765 		return (vop_stdvptocnp(ap));
766 
767 	locked = VOP_ISLOCKED(vp);
768 	lvp = NULLVPTOLOWERVP(vp);
769 	vhold(lvp);
770 	VOP_UNLOCK(vp, 0); /* vp is held by vn_vptocnp_locked that called us */
771 	ldvp = lvp;
772 	vref(lvp);
773 	error = vn_vptocnp(&ldvp, cred, ap->a_buf, ap->a_buflen);
774 	vdrop(lvp);
775 	if (error != 0) {
776 		vn_lock(vp, locked | LK_RETRY);
777 		return (ENOENT);
778 	}
779 
780 	/*
781 	 * Exclusive lock is required by insmntque1 call in
782 	 * null_nodeget()
783 	 */
784 	error = vn_lock(ldvp, LK_EXCLUSIVE);
785 	if (error != 0) {
786 		vrele(ldvp);
787 		vn_lock(vp, locked | LK_RETRY);
788 		return (ENOENT);
789 	}
790 	vref(ldvp);
791 	error = null_nodeget(vp->v_mount, ldvp, dvp);
792 	if (error == 0) {
793 #ifdef DIAGNOSTIC
794 		NULLVPTOLOWERVP(*dvp);
795 #endif
796 		VOP_UNLOCK(*dvp, 0); /* keep reference on *dvp */
797 	}
798 	vn_lock(vp, locked | LK_RETRY);
799 	return (error);
800 }
801 
802 /*
803  * Global vfs data structures
804  */
805 struct vop_vector null_vnodeops = {
806 	.vop_bypass =		null_bypass,
807 	.vop_access =		null_access,
808 	.vop_accessx =		null_accessx,
809 	.vop_advlockpurge =	vop_stdadvlockpurge,
810 	.vop_bmap =		VOP_EOPNOTSUPP,
811 	.vop_getattr =		null_getattr,
812 	.vop_getwritemount =	null_getwritemount,
813 	.vop_inactive =		null_inactive,
814 	.vop_islocked =		vop_stdislocked,
815 	.vop_lock1 =		null_lock,
816 	.vop_lookup =		null_lookup,
817 	.vop_open =		null_open,
818 	.vop_print =		null_print,
819 	.vop_reclaim =		null_reclaim,
820 	.vop_remove =		null_remove,
821 	.vop_rename =		null_rename,
822 	.vop_setattr =		null_setattr,
823 	.vop_strategy =		VOP_EOPNOTSUPP,
824 	.vop_unlock =		null_unlock,
825 	.vop_vptocnp =		null_vptocnp,
826 	.vop_vptofh =		null_vptofh,
827 };
828