null_vnops.c revision 182943
1139776Simp/*-
21541Srgrimes * Copyright (c) 1992, 1993
31541Srgrimes *	The Regents of the University of California.  All rights reserved.
41541Srgrimes *
51541Srgrimes * This code is derived from software contributed to Berkeley by
61541Srgrimes * John Heidemann of the UCLA Ficus project.
71541Srgrimes *
81541Srgrimes * Redistribution and use in source and binary forms, with or without
91541Srgrimes * modification, are permitted provided that the following conditions
101541Srgrimes * are met:
111541Srgrimes * 1. Redistributions of source code must retain the above copyright
121541Srgrimes *    notice, this list of conditions and the following disclaimer.
131541Srgrimes * 2. Redistributions in binary form must reproduce the above copyright
141541Srgrimes *    notice, this list of conditions and the following disclaimer in the
151541Srgrimes *    documentation and/or other materials provided with the distribution.
161541Srgrimes * 4. Neither the name of the University nor the names of its contributors
171541Srgrimes *    may be used to endorse or promote products derived from this software
181541Srgrimes *    without specific prior written permission.
191541Srgrimes *
201541Srgrimes * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
211541Srgrimes * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
221541Srgrimes * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
231541Srgrimes * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
241541Srgrimes * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
251541Srgrimes * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
261541Srgrimes * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
271541Srgrimes * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
281541Srgrimes * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
291541Srgrimes * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
301541Srgrimes * SUCH DAMAGE.
311541Srgrimes *
3222521Sdyson *	@(#)null_vnops.c	8.6 (Berkeley) 5/27/95
331541Srgrimes *
3422521Sdyson * Ancestors:
3522521Sdyson *	@(#)lofs_vnops.c	1.2 (Berkeley) 6/18/92
3622521Sdyson *	...and...
3722521Sdyson *	@(#)null_vnodeops.c 1.20 92/07/07 UCLA Ficus project
3822521Sdyson *
3950477Speter * $FreeBSD: head/sys/fs/nullfs/null_vnops.c 182943 2008-09-11 20:15:34Z ed $
401541Srgrimes */
411541Srgrimes
421541Srgrimes/*
431541Srgrimes * Null Layer
441541Srgrimes *
4577130Sru * (See mount_nullfs(8) for more information.)
461541Srgrimes *
4796755Strhodes * The null layer duplicates a portion of the filesystem
481541Srgrimes * name space under a new name.  In this respect, it is
4996755Strhodes * similar to the loopback filesystem.  It differs from
501541Srgrimes * the loopback fs in two respects:  it is implemented using
5135256Sdes * a stackable layers techniques, and its "null-node"s stack above
521541Srgrimes * all lower-layer vnodes, not just over directory vnodes.
531541Srgrimes *
541541Srgrimes * The null layer has two purposes.  First, it serves as a demonstration
551541Srgrimes * of layering by proving a layer which does nothing.  (It actually
5696755Strhodes * does everything the loopback filesystem does, which is slightly
571541Srgrimes * more than nothing.)  Second, the null layer can serve as a prototype
581541Srgrimes * layer.  Since it provides all necessary layer framework,
5996755Strhodes * new filesystem layers can be created very easily be starting
601541Srgrimes * with a null layer.
611541Srgrimes *
621541Srgrimes * The remainder of this man page examines the null layer as a basis
631541Srgrimes * for constructing new layers.
641541Srgrimes *
651541Srgrimes *
661541Srgrimes * INSTANTIATING NEW NULL LAYERS
671541Srgrimes *
6877130Sru * New null layers are created with mount_nullfs(8).
6977130Sru * Mount_nullfs(8) takes two arguments, the pathname
701541Srgrimes * of the lower vfs (target-pn) and the pathname where the null
711541Srgrimes * layer will appear in the namespace (alias-pn).  After
721541Srgrimes * the null layer is put into place, the contents
731541Srgrimes * of target-pn subtree will be aliased under alias-pn.
741541Srgrimes *
751541Srgrimes *
761541Srgrimes * OPERATION OF A NULL LAYER
771541Srgrimes *
7896755Strhodes * The null layer is the minimum filesystem layer,
791541Srgrimes * simply bypassing all possible operations to the lower layer
801541Srgrimes * for processing there.  The majority of its activity centers
8126963Salex * on the bypass routine, through which nearly all vnode operations
821541Srgrimes * pass.
831541Srgrimes *
841541Srgrimes * The bypass routine accepts arbitrary vnode operations for
851541Srgrimes * handling by the lower layer.  It begins by examing vnode
861541Srgrimes * operation arguments and replacing any null-nodes by their
871541Srgrimes * lower-layer equivlants.  It then invokes the operation
881541Srgrimes * on the lower layer.  Finally, it replaces the null-nodes
891541Srgrimes * in the arguments and, if a vnode is return by the operation,
901541Srgrimes * stacks a null-node on top of the returned vnode.
911541Srgrimes *
9222521Sdyson * Although bypass handles most operations, vop_getattr, vop_lock,
9322521Sdyson * vop_unlock, vop_inactive, vop_reclaim, and vop_print are not
9422521Sdyson * bypassed. Vop_getattr must change the fsid being returned.
9522521Sdyson * Vop_lock and vop_unlock must handle any locking for the
9622521Sdyson * current vnode as well as pass the lock request down.
971541Srgrimes * Vop_inactive and vop_reclaim are not bypassed so that
9822521Sdyson * they can handle freeing null-layer specific data. Vop_print
9922521Sdyson * is not bypassed to avoid excessive debugging information.
10022521Sdyson * Also, certain vnode operations change the locking state within
10122521Sdyson * the operation (create, mknod, remove, link, rename, mkdir, rmdir,
10222521Sdyson * and symlink). Ideally these operations should not change the
10322521Sdyson * lock state, but should be changed to let the caller of the
10422521Sdyson * function unlock them. Otherwise all intermediate vnode layers
10522521Sdyson * (such as union, umapfs, etc) must catch these functions to do
10622521Sdyson * the necessary locking at their layer.
1071541Srgrimes *
1081541Srgrimes *
1091541Srgrimes * INSTANTIATING VNODE STACKS
1101541Srgrimes *
1111541Srgrimes * Mounting associates the null layer with a lower layer,
1121541Srgrimes * effect stacking two VFSes.  Vnode stacks are instead
1131541Srgrimes * created on demand as files are accessed.
1141541Srgrimes *
1151541Srgrimes * The initial mount creates a single vnode stack for the
1161541Srgrimes * root of the new null layer.  All other vnode stacks
1171541Srgrimes * are created as a result of vnode operations on
1181541Srgrimes * this or other null vnode stacks.
1191541Srgrimes *
1201541Srgrimes * New vnode stacks come into existance as a result of
1218876Srgrimes * an operation which returns a vnode.
1221541Srgrimes * The bypass routine stacks a null-node above the new
1231541Srgrimes * vnode before returning it to the caller.
1241541Srgrimes *
1251541Srgrimes * For example, imagine mounting a null layer with
12677130Sru * "mount_nullfs /usr/include /dev/layer/null".
1271541Srgrimes * Changing directory to /dev/layer/null will assign
1281541Srgrimes * the root null-node (which was created when the null layer was mounted).
1291541Srgrimes * Now consider opening "sys".  A vop_lookup would be
1301541Srgrimes * done on the root null-node.  This operation would bypass through
1318876Srgrimes * to the lower layer which would return a vnode representing
1321541Srgrimes * the UFS "sys".  Null_bypass then builds a null-node
1331541Srgrimes * aliasing the UFS "sys" and returns this to the caller.
1341541Srgrimes * Later operations on the null-node "sys" will repeat this
1351541Srgrimes * process when constructing other vnode stacks.
1361541Srgrimes *
1371541Srgrimes *
1381541Srgrimes * CREATING OTHER FILE SYSTEM LAYERS
1391541Srgrimes *
14096755Strhodes * One of the easiest ways to construct new filesystem layers is to make
1411541Srgrimes * a copy of the null layer, rename all files and variables, and
1421541Srgrimes * then begin modifing the copy.  Sed can be used to easily rename
1431541Srgrimes * all variables.
1441541Srgrimes *
1458876Srgrimes * The umap layer is an example of a layer descended from the
1461541Srgrimes * null layer.
1471541Srgrimes *
1481541Srgrimes *
1491541Srgrimes * INVOKING OPERATIONS ON LOWER LAYERS
1501541Srgrimes *
1518876Srgrimes * There are two techniques to invoke operations on a lower layer
1521541Srgrimes * when the operation cannot be completely bypassed.  Each method
1531541Srgrimes * is appropriate in different situations.  In both cases,
1541541Srgrimes * it is the responsibility of the aliasing layer to make
1551541Srgrimes * the operation arguments "correct" for the lower layer
156108470Sschweikh * by mapping a vnode arguments to the lower layer.
1571541Srgrimes *
1581541Srgrimes * The first approach is to call the aliasing layer's bypass routine.
1591541Srgrimes * This method is most suitable when you wish to invoke the operation
16026964Salex * currently being handled on the lower layer.  It has the advantage
1611541Srgrimes * that the bypass routine already must do argument mapping.
1621541Srgrimes * An example of this is null_getattrs in the null layer.
1631541Srgrimes *
16426964Salex * A second approach is to directly invoke vnode operations on
1651541Srgrimes * the lower layer with the VOP_OPERATIONNAME interface.
1661541Srgrimes * The advantage of this method is that it is easy to invoke
1671541Srgrimes * arbitrary operations on the lower layer.  The disadvantage
16826964Salex * is that vnode arguments must be manualy mapped.
1691541Srgrimes *
1701541Srgrimes */
1711541Srgrimes
1721541Srgrimes#include <sys/param.h>
1731541Srgrimes#include <sys/systm.h>
17476166Smarkm#include <sys/conf.h>
1752960Swollman#include <sys/kernel.h>
17676166Smarkm#include <sys/lock.h>
17776166Smarkm#include <sys/malloc.h>
17876166Smarkm#include <sys/mount.h>
17976166Smarkm#include <sys/mutex.h>
18076166Smarkm#include <sys/namei.h>
18112769Sphk#include <sys/sysctl.h>
1821541Srgrimes#include <sys/vnode.h>
18376166Smarkm
18477031Sru#include <fs/nullfs/null.h>
1851541Srgrimes
18666356Sbp#include <vm/vm.h>
18766356Sbp#include <vm/vm_extern.h>
18866356Sbp#include <vm/vm_object.h>
18966356Sbp#include <vm/vnode_pager.h>
19066356Sbp
19112769Sphkstatic int null_bug_bypass = 0;   /* for debugging: enables bypass printf'ing */
19212769SphkSYSCTL_INT(_debug, OID_AUTO, nullfs_bug_bypass, CTLFLAG_RW,
19312769Sphk	&null_bug_bypass, 0, "");
1941541Srgrimes
1951541Srgrimes/*
1961541Srgrimes * This is the 10-Apr-92 bypass routine.
1971541Srgrimes *    This version has been optimized for speed, throwing away some
1981541Srgrimes * safety checks.  It should still always work, but it's not as
1991541Srgrimes * robust to programmer errors.
2001541Srgrimes *
2011541Srgrimes * In general, we map all vnodes going down and unmap them on the way back.
2021541Srgrimes * As an exception to this, vnodes can be marked "unmapped" by setting
2031541Srgrimes * the Nth bit in operation's vdesc_flags.
2041541Srgrimes *
2051541Srgrimes * Also, some BSD vnode operations have the side effect of vrele'ing
2061541Srgrimes * their arguments.  With stacking, the reference counts are held
2071541Srgrimes * by the upper node, not the lower one, so we must handle these
2081541Srgrimes * side-effects here.  This is not of concern in Sun-derived systems
2091541Srgrimes * since there are no such side-effects.
2101541Srgrimes *
2111541Srgrimes * This makes the following assumptions:
2121541Srgrimes * - only one returned vpp
2131541Srgrimes * - no INOUT vpp's (Sun's vop_open has one of these)
2141541Srgrimes * - the vnode operation vector of the first vnode should be used
2151541Srgrimes *   to determine what implementation of the op should be invoked
2161541Srgrimes * - all mapped vnodes are of our vnode-type (NEEDSWORK:
2171541Srgrimes *   problems on rmdir'ing mount points and renaming?)
2188876Srgrimes */
21922521Sdysonint
220140728Sphknull_bypass(struct vop_generic_args *ap)
2211541Srgrimes{
222140732Sphk	struct vnode **this_vp_p;
2231541Srgrimes	int error;
2241541Srgrimes	struct vnode *old_vps[VDESC_MAX_VPS];
2251541Srgrimes	struct vnode **vps_p[VDESC_MAX_VPS];
2261541Srgrimes	struct vnode ***vppp;
2271541Srgrimes	struct vnodeop_desc *descp = ap->a_desc;
2281541Srgrimes	int reles, i;
2291541Srgrimes
2301541Srgrimes	if (null_bug_bypass)
2311541Srgrimes		printf ("null_bypass: %s\n", descp->vdesc_name);
2321541Srgrimes
23350616Sbde#ifdef DIAGNOSTIC
2341541Srgrimes	/*
2351541Srgrimes	 * We require at least one vp.
2361541Srgrimes	 */
2371541Srgrimes	if (descp->vdesc_vp_offsets == NULL ||
2381541Srgrimes	    descp->vdesc_vp_offsets[0] == VDESC_NO_OFFSET)
23950616Sbde		panic ("null_bypass: no vp's in map");
2401541Srgrimes#endif
2411541Srgrimes
2421541Srgrimes	/*
2431541Srgrimes	 * Map the vnodes going in.
2441541Srgrimes	 * Later, we'll invoke the operation based on
2451541Srgrimes	 * the first mapped vnode's operation vector.
2461541Srgrimes	 */
2471541Srgrimes	reles = descp->vdesc_flags;
2481541Srgrimes	for (i = 0; i < VDESC_MAX_VPS; reles >>= 1, i++) {
2491541Srgrimes		if (descp->vdesc_vp_offsets[i] == VDESC_NO_OFFSET)
2501541Srgrimes			break;   /* bail out at end of list */
2518876Srgrimes		vps_p[i] = this_vp_p =
2521541Srgrimes			VOPARG_OFFSETTO(struct vnode**,descp->vdesc_vp_offsets[i],ap);
2531541Srgrimes		/*
2541541Srgrimes		 * We're not guaranteed that any but the first vnode
2551541Srgrimes		 * are of our type.  Check for and don't map any
2561541Srgrimes		 * that aren't.  (We must always map first vp or vclean fails.)
2571541Srgrimes		 */
25824987Skato		if (i && (*this_vp_p == NULLVP ||
259138290Sphk		    (*this_vp_p)->v_op != &null_vnodeops)) {
26024987Skato			old_vps[i] = NULLVP;
2611541Srgrimes		} else {
2621541Srgrimes			old_vps[i] = *this_vp_p;
2631541Srgrimes			*(vps_p[i]) = NULLVPTOLOWERVP(*this_vp_p);
2641541Srgrimes			/*
2651541Srgrimes			 * XXX - Several operations have the side effect
2661541Srgrimes			 * of vrele'ing their vp's.  We must account for
2671541Srgrimes			 * that.  (This should go away in the future.)
2681541Srgrimes			 */
26966356Sbp			if (reles & VDESC_VP0_WILLRELE)
2701541Srgrimes				VREF(*this_vp_p);
2711541Srgrimes		}
2728876Srgrimes
2731541Srgrimes	}
2741541Srgrimes
2751541Srgrimes	/*
2761541Srgrimes	 * Call the operation on the lower layer
2771541Srgrimes	 * with the modified argument structure.
2781541Srgrimes	 */
27966356Sbp	if (vps_p[0] && *vps_p[0])
280140165Sphk		error = VCALL(ap);
28166356Sbp	else {
28266356Sbp		printf("null_bypass: no map for %s\n", descp->vdesc_name);
28366356Sbp		error = EINVAL;
28466356Sbp	}
2851541Srgrimes
2861541Srgrimes	/*
2871541Srgrimes	 * Maintain the illusion of call-by-value
2881541Srgrimes	 * by restoring vnodes in the argument structure
2891541Srgrimes	 * to their original value.
2901541Srgrimes	 */
2911541Srgrimes	reles = descp->vdesc_flags;
2921541Srgrimes	for (i = 0; i < VDESC_MAX_VPS; reles >>= 1, i++) {
2931541Srgrimes		if (descp->vdesc_vp_offsets[i] == VDESC_NO_OFFSET)
2941541Srgrimes			break;   /* bail out at end of list */
2951541Srgrimes		if (old_vps[i]) {
2961541Srgrimes			*(vps_p[i]) = old_vps[i];
29766356Sbp#if 0
29866356Sbp			if (reles & VDESC_VP0_WILLUNLOCK)
299175294Sattilio				VOP_UNLOCK(*(vps_p[i]), 0);
30066356Sbp#endif
30166356Sbp			if (reles & VDESC_VP0_WILLRELE)
3021541Srgrimes				vrele(*(vps_p[i]));
3031541Srgrimes		}
3041541Srgrimes	}
3051541Srgrimes
3061541Srgrimes	/*
3071541Srgrimes	 * Map the possible out-going vpp
3081541Srgrimes	 * (Assumes that the lower layer always returns
3091541Srgrimes	 * a VREF'ed vpp unless it gets an error.)
3101541Srgrimes	 */
3111541Srgrimes	if (descp->vdesc_vpp_offset != VDESC_NO_OFFSET &&
3121541Srgrimes	    !(descp->vdesc_flags & VDESC_NOMAP_VPP) &&
3131541Srgrimes	    !error) {
3141541Srgrimes		/*
3151541Srgrimes		 * XXX - even though some ops have vpp returned vp's,
3161541Srgrimes		 * several ops actually vrele this before returning.
3171541Srgrimes		 * We must avoid these ops.
3181541Srgrimes		 * (This should go away when these ops are regularized.)
3191541Srgrimes		 */
3201541Srgrimes		if (descp->vdesc_flags & VDESC_VPP_WILLRELE)
3211541Srgrimes			goto out;
3221541Srgrimes		vppp = VOPARG_OFFSETTO(struct vnode***,
3231541Srgrimes				 descp->vdesc_vpp_offset,ap);
32429584Sphk		if (*vppp)
32598183Ssemenu			error = null_nodeget(old_vps[0]->v_mount, **vppp, *vppp);
3261541Srgrimes	}
3271541Srgrimes
3281541Srgrimes out:
3291541Srgrimes	return (error);
3301541Srgrimes}
3311541Srgrimes
33222521Sdyson/*
33322521Sdyson * We have to carry on the locking protocol on the null layer vnodes
33422521Sdyson * as we progress through the tree. We also have to enforce read-only
33522521Sdyson * if this layer is mounted read-only.
33622521Sdyson */
33722521Sdysonstatic int
338140728Sphknull_lookup(struct vop_lookup_args *ap)
33922521Sdyson{
34022521Sdyson	struct componentname *cnp = ap->a_cnp;
34166356Sbp	struct vnode *dvp = ap->a_dvp;
34222521Sdyson	int flags = cnp->cn_flags;
34366356Sbp	struct vnode *vp, *ldvp, *lvp;
34422521Sdyson	int error;
3451541Srgrimes
34666356Sbp	if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
34722521Sdyson	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
34822521Sdyson		return (EROFS);
34966356Sbp	/*
35066356Sbp	 * Although it is possible to call null_bypass(), we'll do
35166356Sbp	 * a direct call to reduce overhead
35266356Sbp	 */
35366356Sbp	ldvp = NULLVPTOLOWERVP(dvp);
35466356Sbp	vp = lvp = NULL;
35566356Sbp	error = VOP_LOOKUP(ldvp, &lvp, cnp);
35622521Sdyson	if (error == EJUSTRETURN && (flags & ISLASTCN) &&
35766356Sbp	    (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
35822521Sdyson	    (cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME))
35922521Sdyson		error = EROFS;
36066356Sbp
36166356Sbp	if ((error == 0 || error == EJUSTRETURN) && lvp != NULL) {
36266356Sbp		if (ldvp == lvp) {
36366356Sbp			*ap->a_vpp = dvp;
36466356Sbp			VREF(dvp);
36566356Sbp			vrele(lvp);
36666356Sbp		} else {
36798183Ssemenu			error = null_nodeget(dvp->v_mount, lvp, &vp);
36898183Ssemenu			if (error) {
36998183Ssemenu				/* XXX Cleanup needed... */
37098183Ssemenu				panic("null_nodeget failed");
37198183Ssemenu			}
37298183Ssemenu			*ap->a_vpp = vp;
37366356Sbp		}
37422521Sdyson	}
37522521Sdyson	return (error);
37622521Sdyson}
37722521Sdyson
378140776Sphkstatic int
379140776Sphknull_open(struct vop_open_args *ap)
380140776Sphk{
381140776Sphk	int retval;
382140776Sphk	struct vnode *vp, *ldvp;
383140776Sphk
384140776Sphk	vp = ap->a_vp;
385140776Sphk	ldvp = NULLVPTOLOWERVP(vp);
386140776Sphk	retval = null_bypass(&ap->a_gen);
387140776Sphk	if (retval == 0)
388140776Sphk		vp->v_object = ldvp->v_object;
389140776Sphk	return (retval);
390140776Sphk}
391140776Sphk
3921541Srgrimes/*
39322521Sdyson * Setattr call. Disallow write attempts if the layer is mounted read-only.
39422521Sdyson */
395105211Sphkstatic int
396140728Sphknull_setattr(struct vop_setattr_args *ap)
39722521Sdyson{
39822521Sdyson	struct vnode *vp = ap->a_vp;
39922521Sdyson	struct vattr *vap = ap->a_vap;
40022521Sdyson
40122521Sdyson  	if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
40222597Smpp	    vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
40322597Smpp	    vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
40422521Sdyson	    (vp->v_mount->mnt_flag & MNT_RDONLY))
40522521Sdyson		return (EROFS);
40622521Sdyson	if (vap->va_size != VNOVAL) {
40722521Sdyson 		switch (vp->v_type) {
40822521Sdyson 		case VDIR:
40922521Sdyson 			return (EISDIR);
41022521Sdyson 		case VCHR:
41122521Sdyson 		case VBLK:
41222521Sdyson 		case VSOCK:
41322521Sdyson 		case VFIFO:
41436840Speter			if (vap->va_flags != VNOVAL)
41536840Speter				return (EOPNOTSUPP);
41622521Sdyson			return (0);
41722521Sdyson		case VREG:
41822521Sdyson		case VLNK:
41922521Sdyson 		default:
42022521Sdyson			/*
42122521Sdyson			 * Disallow write attempts if the filesystem is
42222521Sdyson			 * mounted read-only.
42322521Sdyson			 */
42422521Sdyson			if (vp->v_mount->mnt_flag & MNT_RDONLY)
42522521Sdyson				return (EROFS);
42622521Sdyson		}
42722521Sdyson	}
42866356Sbp
42922607Smpp	return (null_bypass((struct vop_generic_args *)ap));
43022521Sdyson}
43122521Sdyson
43222521Sdyson/*
4331541Srgrimes *  We handle getattr only to change the fsid.
4341541Srgrimes */
43512769Sphkstatic int
436140728Sphknull_getattr(struct vop_getattr_args *ap)
4371541Srgrimes{
4381541Srgrimes	int error;
43922521Sdyson
44043311Sdillon	if ((error = null_bypass((struct vop_generic_args *)ap)) != 0)
4411541Srgrimes		return (error);
44265467Sbp
44365467Sbp	ap->a_vap->va_fsid = ap->a_vp->v_mount->mnt_stat.f_fsid.val[0];
4441541Srgrimes	return (0);
4451541Srgrimes}
4461541Srgrimes
44766356Sbp/*
44866356Sbp * Handle to disallow write access if mounted read-only.
44966356Sbp */
45022521Sdysonstatic int
451140728Sphknull_access(struct vop_access_args *ap)
45222521Sdyson{
45322521Sdyson	struct vnode *vp = ap->a_vp;
45422521Sdyson	mode_t mode = ap->a_mode;
4551541Srgrimes
45622521Sdyson	/*
45722521Sdyson	 * Disallow write attempts on read-only layers;
45822521Sdyson	 * unless the file is a socket, fifo, or a block or
45996755Strhodes	 * character device resident on the filesystem.
46022521Sdyson	 */
46122521Sdyson	if (mode & VWRITE) {
46222521Sdyson		switch (vp->v_type) {
46322521Sdyson		case VDIR:
46422521Sdyson		case VLNK:
46522521Sdyson		case VREG:
46622521Sdyson			if (vp->v_mount->mnt_flag & MNT_RDONLY)
46722521Sdyson				return (EROFS);
46822521Sdyson			break;
46943305Sdillon		default:
47043305Sdillon			break;
47122521Sdyson		}
47222521Sdyson	}
47322607Smpp	return (null_bypass((struct vop_generic_args *)ap));
47422521Sdyson}
47522521Sdyson
47622521Sdyson/*
47765467Sbp * We handle this to eliminate null FS to lower FS
47865467Sbp * file moving. Don't know why we don't allow this,
47965467Sbp * possibly we should.
48065467Sbp */
48165467Sbpstatic int
482140728Sphknull_rename(struct vop_rename_args *ap)
48365467Sbp{
48465467Sbp	struct vnode *tdvp = ap->a_tdvp;
48565467Sbp	struct vnode *fvp = ap->a_fvp;
48665467Sbp	struct vnode *fdvp = ap->a_fdvp;
48765467Sbp	struct vnode *tvp = ap->a_tvp;
48865467Sbp
48965467Sbp	/* Check for cross-device rename. */
49065467Sbp	if ((fvp->v_mount != tdvp->v_mount) ||
49165467Sbp	    (tvp && (fvp->v_mount != tvp->v_mount))) {
49265467Sbp		if (tdvp == tvp)
49365467Sbp			vrele(tdvp);
49465467Sbp		else
49565467Sbp			vput(tdvp);
49665467Sbp		if (tvp)
49765467Sbp			vput(tvp);
49865467Sbp		vrele(fdvp);
49965467Sbp		vrele(fvp);
50065467Sbp		return (EXDEV);
50165467Sbp	}
50265467Sbp
50365467Sbp	return (null_bypass((struct vop_generic_args *)ap));
50465467Sbp}
50565467Sbp
50665467Sbp/*
50722521Sdyson * We need to process our own vnode lock and then clear the
50822521Sdyson * interlock flag as it applies only to our vnode, not the
50922521Sdyson * vnodes below us on the stack.
51022521Sdyson */
51122597Smppstatic int
512169671Skibnull_lock(struct vop_lock1_args *ap)
51322521Sdyson{
51466356Sbp	struct vnode *vp = ap->a_vp;
51566356Sbp	int flags = ap->a_flags;
516143642Sjeff	struct null_node *nn;
51766356Sbp	struct vnode *lvp;
51866356Sbp	int error;
51922521Sdyson
52066356Sbp
521143513Sjeff	if ((flags & LK_INTERLOCK) == 0) {
522143513Sjeff		VI_LOCK(vp);
523143642Sjeff		ap->a_flags = flags |= LK_INTERLOCK;
524143513Sjeff	}
525143642Sjeff	nn = VTONULL(vp);
526143642Sjeff	/*
527143642Sjeff	 * If we're still active we must ask the lower layer to
528143642Sjeff	 * lock as ffs has special lock considerations in it's
529143642Sjeff	 * vop lock.
530143642Sjeff	 */
531143642Sjeff	if (nn != NULL && (lvp = NULLVPTOLOWERVP(vp)) != NULL) {
532145424Sjeff		VI_LOCK_FLAGS(lvp, MTX_DUPOK);
533116469Stjr		VI_UNLOCK(vp);
53466356Sbp		/*
535143642Sjeff		 * We have to hold the vnode here to solve a potential
536143642Sjeff		 * reclaim race.  If we're forcibly vgone'd while we
537143642Sjeff		 * still have refs, a thread could be sleeping inside
538143642Sjeff		 * the lowervp's vop_lock routine.  When we vgone we will
539143642Sjeff		 * drop our last ref to the lowervp, which would allow it
540143642Sjeff		 * to be reclaimed.  The lowervp could then be recycled,
541143642Sjeff		 * in which case it is not legal to be sleeping in it's VOP.
542143642Sjeff		 * We prevent it from being recycled by holding the vnode
543143642Sjeff		 * here.
54466356Sbp		 */
545143642Sjeff		vholdl(lvp);
546175294Sattilio		error = VOP_LOCK(lvp, flags);
547150181Skan
548150181Skan		/*
549150181Skan		 * We might have slept to get the lock and someone might have
550150181Skan		 * clean our vnode already, switching vnode lock from one in
551150181Skan		 * lowervp to v_lock in our own vnode structure.  Handle this
552150181Skan		 * case by reacquiring correct lock in requested mode.
553150181Skan		 */
554150181Skan		if (VTONULL(vp) == NULL && error == 0) {
555150181Skan			ap->a_flags &= ~(LK_TYPE_MASK | LK_INTERLOCK);
556150181Skan			switch (flags & LK_TYPE_MASK) {
557150181Skan			case LK_SHARED:
558150181Skan				ap->a_flags |= LK_SHARED;
559150181Skan				break;
560150181Skan			case LK_UPGRADE:
561150181Skan			case LK_EXCLUSIVE:
562150181Skan				ap->a_flags |= LK_EXCLUSIVE;
563150181Skan				break;
564150181Skan			default:
565150181Skan				panic("Unsupported lock request %d\n",
566150181Skan				    ap->a_flags);
567150181Skan			}
568175294Sattilio			VOP_UNLOCK(lvp, 0);
569150181Skan			error = vop_stdlock(ap);
570150181Skan		}
571143642Sjeff		vdrop(lvp);
572143642Sjeff	} else
573143642Sjeff		error = vop_stdlock(ap);
574143642Sjeff
575143642Sjeff	return (error);
57622521Sdyson}
57722521Sdyson
57822521Sdyson/*
57922521Sdyson * We need to process our own vnode unlock and then clear the
58022521Sdyson * interlock flag as it applies only to our vnode, not the
58122521Sdyson * vnodes below us on the stack.
58222521Sdyson */
58322597Smppstatic int
584140728Sphknull_unlock(struct vop_unlock_args *ap)
58522521Sdyson{
58666356Sbp	struct vnode *vp = ap->a_vp;
58766356Sbp	int flags = ap->a_flags;
588172644Sdaichi	int mtxlkflag = 0;
589143642Sjeff	struct null_node *nn;
59066570Sbp	struct vnode *lvp;
591143642Sjeff	int error;
59266356Sbp
593172644Sdaichi	if ((flags & LK_INTERLOCK) != 0)
594172644Sdaichi		mtxlkflag = 1;
595172644Sdaichi	else if (mtx_owned(VI_MTX(vp)) == 0) {
596172644Sdaichi		VI_LOCK(vp);
597172644Sdaichi		mtxlkflag = 2;
59866356Sbp	}
599143642Sjeff	nn = VTONULL(vp);
600172644Sdaichi	if (nn != NULL && (lvp = NULLVPTOLOWERVP(vp)) != NULL) {
601172644Sdaichi		VI_LOCK_FLAGS(lvp, MTX_DUPOK);
602172644Sdaichi		flags |= LK_INTERLOCK;
603172644Sdaichi		vholdl(lvp);
604172644Sdaichi		VI_UNLOCK(vp);
605175294Sattilio		error = VOP_UNLOCK(lvp, flags);
606172644Sdaichi		vdrop(lvp);
607172644Sdaichi		if (mtxlkflag == 0)
608172644Sdaichi			VI_LOCK(vp);
609172644Sdaichi	} else {
610172644Sdaichi		if (mtxlkflag == 2)
611172644Sdaichi			VI_UNLOCK(vp);
612143642Sjeff		error = vop_stdunlock(ap);
613172644Sdaichi	}
614143642Sjeff
615143642Sjeff	return (error);
61622521Sdyson}
61722521Sdyson
61822597Smppstatic int
619140728Sphknull_islocked(struct vop_islocked_args *ap)
62066356Sbp{
62166356Sbp	struct vnode *vp = ap->a_vp;
62266356Sbp
623176559Sattilio	return (lockstatus(vp->v_vnlock));
62466356Sbp}
62566356Sbp
62666356Sbp/*
62766356Sbp * There is no way to tell that someone issued remove/rmdir operation
628182943Sed * on the underlying filesystem. For now we just have to release lowervp
62966356Sbp * as soon as possible.
63098183Ssemenu *
63198183Ssemenu * Note, we can't release any resources nor remove vnode from hash before
63298183Ssemenu * appropriate VXLOCK stuff is is done because other process can find this
63398183Ssemenu * vnode in hash during inactivation and may be sitting in vget() and waiting
63498183Ssemenu * for null_inactive to unlock vnode. Thus we will do all those in VOP_RECLAIM.
63566356Sbp */
63666356Sbpstatic int
637140728Sphknull_inactive(struct vop_inactive_args *ap)
6381541Srgrimes{
63930636Sroberto	struct vnode *vp = ap->a_vp;
64098175Ssemenu	struct thread *td = ap->a_td;
64192540Smckusick
642141447Sphk	vp->v_object = NULL;
64398175Ssemenu
64492540Smckusick	/*
64592540Smckusick	 * If this is the last reference, then free up the vnode
64692540Smckusick	 * so as not to tie up the lower vnodes.
64792540Smckusick	 */
648140936Sphk	vrecycle(vp, td);
64998175Ssemenu
65092540Smckusick	return (0);
65192540Smckusick}
65292540Smckusick
65392540Smckusick/*
65498183Ssemenu * Now, the VXLOCK is in force and we're free to destroy the null vnode.
65592540Smckusick */
65692540Smckusickstatic int
657140728Sphknull_reclaim(struct vop_reclaim_args *ap)
65892540Smckusick{
65992540Smckusick	struct vnode *vp = ap->a_vp;
66030636Sroberto	struct null_node *xp = VTONULL(vp);
66130636Sroberto	struct vnode *lowervp = xp->null_lowervp;
662143630Sjeff	struct lock *vnlock;
66366356Sbp
664155899Sjeff	if (lowervp)
665155899Sjeff		null_hashrem(xp);
666143744Sjeff	/*
667143744Sjeff	 * Use the interlock to protect the clearing of v_data to
668143744Sjeff	 * prevent faults in null_lock().
669143744Sjeff	 */
670143744Sjeff	VI_LOCK(vp);
671143744Sjeff	vp->v_data = NULL;
672155899Sjeff	vp->v_object = NULL;
673150181Skan	vnlock = vp->v_vnlock;
674150181Skan	vp->v_vnlock = &vp->v_lock;
675149722Sssouhlal	if (lowervp) {
676175635Sattilio		lockmgr(vp->v_vnlock, LK_EXCLUSIVE | LK_INTERLOCK, VI_MTX(vp));
677149722Sssouhlal		vput(lowervp);
678149722Sssouhlal	} else
679182943Sed		panic("null_reclaim: reclaiming a node with no lowervp");
68098176Ssemenu	FREE(xp, M_NULLFSNODE);
68166356Sbp
6821541Srgrimes	return (0);
6831541Srgrimes}
6841541Srgrimes
68512769Sphkstatic int
686140728Sphknull_print(struct vop_print_args *ap)
6871541Srgrimes{
688140732Sphk	struct vnode *vp = ap->a_vp;
689155899Sjeff
690111841Snjl	printf("\tvp=%p, lowervp=%p\n", vp, NULLVPTOLOWERVP(vp));
6911541Srgrimes	return (0);
6921541Srgrimes}
6931541Srgrimes
694156585Sjeff/* ARGSUSED */
695156585Sjeffstatic int
696156585Sjeffnull_getwritemount(struct vop_getwritemount_args *ap)
697156585Sjeff{
698156585Sjeff	struct null_node *xp;
699156585Sjeff	struct vnode *lowervp;
700156585Sjeff	struct vnode *vp;
701156585Sjeff
702156585Sjeff	vp = ap->a_vp;
703156585Sjeff	VI_LOCK(vp);
704156585Sjeff	xp = VTONULL(vp);
705156585Sjeff	if (xp && (lowervp = xp->null_lowervp)) {
706156585Sjeff		VI_LOCK_FLAGS(lowervp, MTX_DUPOK);
707156585Sjeff		VI_UNLOCK(vp);
708156585Sjeff		vholdl(lowervp);
709156585Sjeff		VI_UNLOCK(lowervp);
710156585Sjeff		VOP_GETWRITEMOUNT(lowervp, ap->a_mpp);
711156585Sjeff		vdrop(lowervp);
712156585Sjeff	} else {
713156585Sjeff		VI_UNLOCK(vp);
714156585Sjeff		*(ap->a_mpp) = NULL;
715156585Sjeff	}
716156585Sjeff	return (0);
717156585Sjeff}
718156585Sjeff
719166774Spjdstatic int
720166774Spjdnull_vptofh(struct vop_vptofh_args *ap)
721166774Spjd{
722166774Spjd	struct vnode *lvp;
723166774Spjd
724166774Spjd	lvp = NULLVPTOLOWERVP(ap->a_vp);
725166774Spjd	return VOP_VPTOFH(lvp, ap->a_fhp);
726166774Spjd}
727166774Spjd
7281541Srgrimes/*
7291541Srgrimes * Global vfs data structures
7301541Srgrimes */
731138290Sphkstruct vop_vector null_vnodeops = {
732138290Sphk	.vop_bypass =		null_bypass,
733138290Sphk	.vop_access =		null_access,
734138290Sphk	.vop_bmap =		VOP_EOPNOTSUPP,
735138290Sphk	.vop_getattr =		null_getattr,
736156585Sjeff	.vop_getwritemount =	null_getwritemount,
737138290Sphk	.vop_inactive =		null_inactive,
738138290Sphk	.vop_islocked =		null_islocked,
739169671Skib	.vop_lock1 =		null_lock,
740138290Sphk	.vop_lookup =		null_lookup,
741140776Sphk	.vop_open =		null_open,
742138290Sphk	.vop_print =		null_print,
743138290Sphk	.vop_reclaim =		null_reclaim,
744138290Sphk	.vop_rename =		null_rename,
745138290Sphk	.vop_setattr =		null_setattr,
746138290Sphk	.vop_strategy =		VOP_EOPNOTSUPP,
747138290Sphk	.vop_unlock =		null_unlock,
748166774Spjd	.vop_vptofh =		null_vptofh,
7491541Srgrimes};
750