null_vnops.c revision 24987
150477Speter/* 22729Sdfr * Copyright (c) 1992, 1993 32729Sdfr * The Regents of the University of California. All rights reserved. 42729Sdfr * 52729Sdfr * This code is derived from software contributed to Berkeley by 62729Sdfr * John Heidemann of the UCLA Ficus project. 72729Sdfr * 82729Sdfr * Redistribution and use in source and binary forms, with or without 92729Sdfr * modification, are permitted provided that the following conditions 102729Sdfr * are met: 112729Sdfr * 1. Redistributions of source code must retain the above copyright 122729Sdfr * notice, this list of conditions and the following disclaimer. 132729Sdfr * 2. Redistributions in binary form must reproduce the above copyright 142729Sdfr * notice, this list of conditions and the following disclaimer in the 152729Sdfr * documentation and/or other materials provided with the distribution. 162729Sdfr * 3. All advertising materials mentioning features or use of this software 172729Sdfr * must display the following acknowledgement: 182729Sdfr * This product includes software developed by the University of 192729Sdfr * California, Berkeley and its contributors. 202729Sdfr * 4. Neither the name of the University nor the names of its contributors 212729Sdfr * may be used to endorse or promote products derived from this software 2259839Speter * without specific prior written permission. 2359839Speter * 242729Sdfr * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 252729Sdfr * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 2611626Sbde * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 272729Sdfr * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 282729Sdfr * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 2982607Sdillon * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 3082607Sdillon * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 312729Sdfr * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 3269449Salfred * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 3311626Sbde * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 3459839Speter * SUCH DAMAGE. 3559839Speter * 3668024Srwatson * @(#)null_vnops.c 8.6 (Berkeley) 5/27/95 372729Sdfr * 3859839Speter * Ancestors: 3959839Speter * @(#)lofs_vnops.c 1.2 (Berkeley) 6/18/92 4069449Salfred * $Id: null_vnops.c,v 1.16 1997/02/22 09:40:22 peter Exp $ 4169449Salfred * ...and... 4269449Salfred * @(#)null_vnodeops.c 1.20 92/07/07 UCLA Ficus project 4310358Sjulian * 442729Sdfr * $Id: null_vnops.c,v 1.16 1997/02/22 09:40:22 peter Exp $ 452729Sdfr */ 462729Sdfr 4711626Sbde/* 482729Sdfr * Null Layer 4911626Sbde * 5012819Sphk * (See mount_null(8) for more information.) 5111626Sbde * 5211626Sbde * The null layer duplicates a portion of the file system 5311626Sbde * name space under a new name. In this respect, it is 542729Sdfr * similar to the loopback file system. It differs from 5559839Speter * the loopback fs in two respects: it is implemented using 5659839Speter * a stackable layers techniques, and it's "null-node"s stack above 5759839Speter * all lower-layer vnodes, not just over directory vnodes. 5859839Speter * 5959839Speter * The null layer has two purposes. First, it serves as a demonstration 6059839Speter * of layering by proving a layer which does nothing. (It actually 6159839Speter * does everything the loopback file system does, which is slightly 6259839Speter * more than nothing.) Second, the null layer can serve as a prototype 6359839Speter * layer. Since it provides all necessary layer framework, 6459839Speter * new file system layers can be created very easily be starting 6559839Speter * with a null layer. 6659839Speter * 6759839Speter * The remainder of this man page examines the null layer as a basis 6859839Speter * for constructing new layers. 6959839Speter * 7059839Speter * 7159839Speter * INSTANTIATING NEW NULL LAYERS 7259839Speter * 7359839Speter * New null layers are created with mount_null(8). 7459839Speter * Mount_null(8) takes two arguments, the pathname 7559839Speter * of the lower vfs (target-pn) and the pathname where the null 7659839Speter * layer will appear in the namespace (alias-pn). After 7759839Speter * the null layer is put into place, the contents 7859839Speter * of target-pn subtree will be aliased under alias-pn. 7959839Speter * 8059839Speter * 8159839Speter * OPERATION OF A NULL LAYER 8259839Speter * 8359839Speter * The null layer is the minimum file system layer, 8459839Speter * simply bypassing all possible operations to the lower layer 8559839Speter * for processing there. The majority of its activity centers 8659839Speter * on the bypass routine, though which nearly all vnode operations 8759839Speter * pass. 8859839Speter * 8959839Speter * The bypass routine accepts arbitrary vnode operations for 9059839Speter * handling by the lower layer. It begins by examing vnode 9159839Speter * operation arguments and replacing any null-nodes by their 9259839Speter * lower-layer equivlants. It then invokes the operation 9359839Speter * on the lower layer. Finally, it replaces the null-nodes 9459839Speter * in the arguments and, if a vnode is return by the operation, 9559839Speter * stacks a null-node on top of the returned vnode. 9659839Speter * 9759839Speter * Although bypass handles most operations, vop_getattr, vop_lock, 9859839Speter * vop_unlock, vop_inactive, vop_reclaim, and vop_print are not 9959839Speter * bypassed. Vop_getattr must change the fsid being returned. 10059839Speter * Vop_lock and vop_unlock must handle any locking for the 10159839Speter * current vnode as well as pass the lock request down. 10259839Speter * Vop_inactive and vop_reclaim are not bypassed so that 10359839Speter * they can handle freeing null-layer specific data. Vop_print 10459839Speter * is not bypassed to avoid excessive debugging information. 10559839Speter * Also, certain vnode operations change the locking state within 10659839Speter * the operation (create, mknod, remove, link, rename, mkdir, rmdir, 10759839Speter * and symlink). Ideally these operations should not change the 10859839Speter * lock state, but should be changed to let the caller of the 10959839Speter * function unlock them. Otherwise all intermediate vnode layers 11059839Speter * (such as union, umapfs, etc) must catch these functions to do 11159839Speter * the necessary locking at their layer. 11259839Speter * 11359839Speter * 11459839Speter * INSTANTIATING VNODE STACKS 11559839Speter * 11659839Speter * Mounting associates the null layer with a lower layer, 11759839Speter * effect stacking two VFSes. Vnode stacks are instead 11859839Speter * created on demand as files are accessed. 11959839Speter * 12059839Speter * The initial mount creates a single vnode stack for the 12159839Speter * root of the new null layer. All other vnode stacks 12212819Sphk * are created as a result of vnode operations on 12312819Sphk * this or other null vnode stacks. 12459839Speter * 12559839Speter * New vnode stacks come into existance as a result of 12659839Speter * an operation which returns a vnode. 12759839Speter * The bypass routine stacks a null-node above the new 12859839Speter * vnode before returning it to the caller. 1292729Sdfr * 13059839Speter * For example, imagine mounting a null layer with 13169449Salfred * "mount_null /usr/include /dev/layer/null". 1322729Sdfr * Changing directory to /dev/layer/null will assign 1332729Sdfr * the root null-node (which was created when the null layer was mounted). 1342729Sdfr * Now consider opening "sys". A vop_lookup would be 13559839Speter * done on the root null-node. This operation would bypass through 13659839Speter * to the lower layer which would return a vnode representing 13759839Speter * the UFS "sys". Null_bypass then builds a null-node 13859839Speter * aliasing the UFS "sys" and returns this to the caller. 13959839Speter * Later operations on the null-node "sys" will repeat this 14059839Speter * process when constructing other vnode stacks. 14159839Speter * 14259839Speter * 14359839Speter * CREATING OTHER FILE SYSTEM LAYERS 14459839Speter * 14559839Speter * One of the easiest ways to construct new file system layers is to make 14659839Speter * a copy of the null layer, rename all files and variables, and 14759839Speter * then begin modifing the copy. Sed can be used to easily rename 1482729Sdfr * all variables. 1492729Sdfr * 1502729Sdfr * The umap layer is an example of a layer descended from the 1512729Sdfr * null layer. 1522729Sdfr * 1532729Sdfr * 1542729Sdfr * INVOKING OPERATIONS ON LOWER LAYERS 1552729Sdfr * 1562729Sdfr * There are two techniques to invoke operations on a lower layer 1572729Sdfr * when the operation cannot be completely bypassed. Each method 1582729Sdfr * is appropriate in different situations. In both cases, 1592729Sdfr * it is the responsibility of the aliasing layer to make 1602729Sdfr * the operation arguments "correct" for the lower layer 1612729Sdfr * by mapping an vnode arguments to the lower layer. 1622729Sdfr * 1632729Sdfr * The first approach is to call the aliasing layer's bypass routine. 1642729Sdfr * This method is most suitable when you wish to invoke the operation 1652729Sdfr * currently being hanldled on the lower layer. It has the advantage 1662729Sdfr * that the bypass routine already must do argument mapping. 1672729Sdfr * An example of this is null_getattrs in the null layer. 1682729Sdfr * 1692729Sdfr * A second approach is to directly invoked vnode operations on 1702729Sdfr * the lower layer with the VOP_OPERATIONNAME interface. 1712729Sdfr * The advantage of this method is that it is easy to invoke 1722729Sdfr * arbitrary operations on the lower layer. The disadvantage 1732729Sdfr * is that vnodes arguments must be manualy mapped. 1742729Sdfr * 1752729Sdfr */ 1762729Sdfr 1772729Sdfr#include <sys/param.h> 1782729Sdfr#include <sys/systm.h> 1792729Sdfr#include <sys/kernel.h> 1802729Sdfr#include <sys/sysctl.h> 1812729Sdfr#include <sys/proc.h> 1822729Sdfr#include <sys/time.h> 1832729Sdfr#include <sys/types.h> 1842729Sdfr#include <sys/vnode.h> 1852729Sdfr#include <sys/mount.h> 1862729Sdfr#include <sys/namei.h> 1872729Sdfr#include <sys/malloc.h> 1882729Sdfr#include <sys/buf.h> 1892729Sdfr#include <miscfs/nullfs/null.h> 1902729Sdfr 1912729Sdfrstatic int null_bug_bypass = 0; /* for debugging: enables bypass printf'ing */ 1922729SdfrSYSCTL_INT(_debug, OID_AUTO, nullfs_bug_bypass, CTLFLAG_RW, 1932729Sdfr &null_bug_bypass, 0, ""); 1942729Sdfr 1952729Sdfrstatic int null_access __P((struct vop_access_args *ap)); 19666085Speterint null_bypass __P((struct vop_generic_args *ap)); 1972729Sdfrstatic int null_bwrite __P((struct vop_bwrite_args *ap)); 1982729Sdfrstatic int null_getattr __P((struct vop_getattr_args *ap)); 1992729Sdfrstatic int null_inactive __P((struct vop_inactive_args *ap)); 20069449Salfredstatic int null_lock __P((struct vop_lock_args *ap)); 20169449Salfredstatic int null_lookup __P((struct vop_lookup_args *ap)); 20269449Salfredstatic int null_print __P((struct vop_print_args *ap)); 20369449Salfredstatic int null_reclaim __P((struct vop_reclaim_args *ap)); 20469449Salfredstatic int null_setattr __P((struct vop_setattr_args *ap)); 20569449Salfredstatic int null_strategy __P((struct vop_strategy_args *ap)); 20669449Salfredstatic int null_unlock __P((struct vop_unlock_args *ap)); 20769449Salfred 20869449Salfred/* 20969449Salfred * This is the 10-Apr-92 bypass routine. 21069449Salfred * This version has been optimized for speed, throwing away some 21169449Salfred * safety checks. It should still always work, but it's not as 21269449Salfred * robust to programmer errors. 21369449Salfred * Define SAFETY to include some error checking code. 21469449Salfred * 21569449Salfred * In general, we map all vnodes going down and unmap them on the way back. 21669449Salfred * As an exception to this, vnodes can be marked "unmapped" by setting 21769449Salfred * the Nth bit in operation's vdesc_flags. 21869449Salfred * 21969449Salfred * Also, some BSD vnode operations have the side effect of vrele'ing 22069449Salfred * their arguments. With stacking, the reference counts are held 22169449Salfred * by the upper node, not the lower one, so we must handle these 22269449Salfred * side-effects here. This is not of concern in Sun-derived systems 22369449Salfred * since there are no such side-effects. 22469449Salfred * 22569449Salfred * This makes the following assumptions: 22669449Salfred * - only one returned vpp 22769449Salfred * - no INOUT vpp's (Sun's vop_open has one of these) 22869449Salfred * - the vnode operation vector of the first vnode should be used 22969449Salfred * to determine what implementation of the op should be invoked 23069449Salfred * - all mapped vnodes are of our vnode-type (NEEDSWORK: 23169449Salfred * problems on rmdir'ing mount points and renaming?) 23269449Salfred */ 23369449Salfredint 23469449Salfrednull_bypass(ap) 23569449Salfred struct vop_generic_args /* { 23669449Salfred struct vnodeop_desc *a_desc; 23769449Salfred <other random data follows, presumably> 23869449Salfred } */ *ap; 23969449Salfred{ 24069449Salfred register struct vnode **this_vp_p; 24169449Salfred int error; 24269449Salfred struct vnode *old_vps[VDESC_MAX_VPS]; 24369449Salfred struct vnode **vps_p[VDESC_MAX_VPS]; 24469449Salfred struct vnode ***vppp; 24569449Salfred struct vnodeop_desc *descp = ap->a_desc; 24669449Salfred int reles, i; 24769449Salfred 24869449Salfred if (null_bug_bypass) 24969449Salfred printf ("null_bypass: %s\n", descp->vdesc_name); 25071038Sdes 25171038Sdes#ifdef SAFETY 25269449Salfred /* 25369449Salfred * We require at least one vp. 25469449Salfred */ 25569449Salfred if (descp->vdesc_vp_offsets == NULL || 25669644Salfred descp->vdesc_vp_offsets[0] == VDESC_NO_OFFSET) 25769449Salfred panic ("null_bypass: no vp's in map."); 25869449Salfred#endif 25969449Salfred 26069449Salfred /* 26169449Salfred * Map the vnodes going in. 26271038Sdes * Later, we'll invoke the operation based on 26369449Salfred * the first mapped vnode's operation vector. 26471038Sdes */ 26569449Salfred reles = descp->vdesc_flags; 2662729Sdfr for (i = 0; i < VDESC_MAX_VPS; reles >>= 1, i++) { 2672729Sdfr if (descp->vdesc_vp_offsets[i] == VDESC_NO_OFFSET) 26882607Sdillon break; /* bail out at end of list */ 26982607Sdillon vps_p[i] = this_vp_p = 2702729Sdfr VOPARG_OFFSETTO(struct vnode**,descp->vdesc_vp_offsets[i],ap); 2712729Sdfr /* 27283366Sjulian * We're not guaranteed that any but the first vnode 27383366Sjulian * are of our type. Check for and don't map any 27411626Sbde * that aren't. (We must always map first vp or vclean fails.) 27511626Sbde */ 27611626Sbde if (i && (*this_vp_p == NULLVP || 27711626Sbde (*this_vp_p)->v_op != null_vnodeop_p)) { 27811626Sbde old_vps[i] = NULLVP; 27911626Sbde } else { 28011626Sbde old_vps[i] = *this_vp_p; 28111626Sbde *(vps_p[i]) = NULLVPTOLOWERVP(*this_vp_p); 28211626Sbde /* 2832729Sdfr * XXX - Several operations have the side effect 28482607Sdillon * of vrele'ing their vp's. We must account for 2852729Sdfr * that. (This should go away in the future.) 28682607Sdillon */ 28783366Sjulian if (reles & 1) 28882607Sdillon VREF(*this_vp_p); 28982607Sdillon } 29082607Sdillon 29182607Sdillon } 29282607Sdillon 29382607Sdillon /* 29482607Sdillon * Call the operation on the lower layer 29583366Sjulian * with the modified argument structure. 29682607Sdillon */ 29782607Sdillon error = VCALL(*(vps_p[0]), descp->vdesc_offset, ap); 29882607Sdillon 2992729Sdfr /* 3002729Sdfr * Maintain the illusion of call-by-value 3012729Sdfr * by restoring vnodes in the argument structure 3022729Sdfr * to their original value. 3032729Sdfr */ 3042729Sdfr reles = descp->vdesc_flags; 3052729Sdfr for (i = 0; i < VDESC_MAX_VPS; reles >>= 1, i++) { 3062729Sdfr if (descp->vdesc_vp_offsets[i] == VDESC_NO_OFFSET) 3072729Sdfr break; /* bail out at end of list */ 3082729Sdfr if (old_vps[i]) { 3092729Sdfr *(vps_p[i]) = old_vps[i]; 3102729Sdfr if (reles & 1) 3112729Sdfr vrele(*(vps_p[i])); 3122729Sdfr } 3132729Sdfr } 3142729Sdfr 3152729Sdfr /* 3162729Sdfr * Map the possible out-going vpp 3172729Sdfr * (Assumes that the lower layer always returns 3182729Sdfr * a VREF'ed vpp unless it gets an error.) 3192729Sdfr */ 3202729Sdfr if (descp->vdesc_vpp_offset != VDESC_NO_OFFSET && 3212729Sdfr !(descp->vdesc_flags & VDESC_NOMAP_VPP) && 3222729Sdfr !error) { 3232729Sdfr /* 3242729Sdfr * XXX - even though some ops have vpp returned vp's, 32512866Speter * several ops actually vrele this before returning. 3262729Sdfr * We must avoid these ops. 3272729Sdfr * (This should go away when these ops are regularized.) 3282729Sdfr */ 32912866Speter if (descp->vdesc_flags & VDESC_VPP_WILLRELE) 3302729Sdfr goto out; 33112866Speter vppp = VOPARG_OFFSETTO(struct vnode***, 3322729Sdfr descp->vdesc_vpp_offset,ap); 33382607Sdillon error = null_node_create(old_vps[0]->v_mount, **vppp, *vppp); 33482607Sdillon } 33582607Sdillon 33612866Speter out: 33783366Sjulian return (error); 33883366Sjulian} 3392729Sdfr 3402729Sdfr/* 3412729Sdfr * We have to carry on the locking protocol on the null layer vnodes 3422729Sdfr * as we progress through the tree. We also have to enforce read-only 34312866Speter * if this layer is mounted read-only. 34482607Sdillon */ 3452729Sdfrstatic int 3462729Sdfrnull_lookup(ap) 3472729Sdfr struct vop_lookup_args /* { 3482729Sdfr struct vnode * a_dvp; 3492729Sdfr struct vnode ** a_vpp; 3502729Sdfr struct componentname * a_cnp; 35182607Sdillon } */ *ap; 35283366Sjulian{ 35382607Sdillon struct componentname *cnp = ap->a_cnp; 35482607Sdillon struct proc *p = cnp->cn_proc; 35582607Sdillon int flags = cnp->cn_flags; 35668024Srwatson struct vop_lock_args lockargs; 3572729Sdfr struct vop_unlock_args unlockargs; 3582729Sdfr struct vnode *dvp, *vp; 3592729Sdfr int error; 3602729Sdfr 3612729Sdfr if ((flags & ISLASTCN) && (ap->a_dvp->v_mount->mnt_flag & MNT_RDONLY) && 3622729Sdfr (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) 3632729Sdfr return (EROFS); 36482607Sdillon error = null_bypass((struct vop_generic_args *)ap); 36582607Sdillon if (error == EJUSTRETURN && (flags & ISLASTCN) && 3662729Sdfr (ap->a_dvp->v_mount->mnt_flag & MNT_RDONLY) && 3672729Sdfr (cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME)) 3682729Sdfr error = EROFS; 3692729Sdfr /* 3702729Sdfr * We must do the same locking and unlocking at this layer as 3712729Sdfr * is done in the layers below us. We could figure this out 3722729Sdfr * based on the error return and the LASTCN, LOCKPARENT, and 3732729Sdfr * LOCKLEAF flags. However, it is more expidient to just find 37482607Sdillon * out the state of the lower level vnodes and set ours to the 37582607Sdillon * same state. 3762729Sdfr */ 3772729Sdfr dvp = ap->a_dvp; 3782729Sdfr vp = *ap->a_vpp; 3792729Sdfr if (dvp == vp) 3802729Sdfr return (error); 38182607Sdillon if (!VOP_ISLOCKED(dvp)) { 38282607Sdillon unlockargs.a_vp = dvp; 3832729Sdfr unlockargs.a_flags = 0; 3842729Sdfr unlockargs.a_p = p; 38582607Sdillon vop_nounlock(&unlockargs); 3862729Sdfr } 3872729Sdfr if (vp != NULLVP && VOP_ISLOCKED(vp)) { 3882729Sdfr lockargs.a_vp = vp; 3892729Sdfr lockargs.a_flags = LK_SHARED; 3902729Sdfr lockargs.a_p = p; 3912729Sdfr vop_nolock(&lockargs); 3922729Sdfr } 39383366Sjulian return (error); 39482607Sdillon} 3952729Sdfr 3962729Sdfr/* 3972729Sdfr * Setattr call. Disallow write attempts if the layer is mounted read-only. 3982729Sdfr */ 3992729Sdfrint 4002729Sdfrnull_setattr(ap) 4012729Sdfr struct vop_setattr_args /* { 4022729Sdfr struct vnodeop_desc *a_desc; 4032729Sdfr struct vnode *a_vp; 4042729Sdfr struct vattr *a_vap; 4052729Sdfr struct ucred *a_cred; 4062729Sdfr struct proc *a_p; 4072729Sdfr } */ *ap; 4082729Sdfr{ 4092729Sdfr struct vnode *vp = ap->a_vp; 4102729Sdfr struct vattr *vap = ap->a_vap; 4112729Sdfr 4122729Sdfr if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL || 4132729Sdfr vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL || 4142729Sdfr vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) && 4152729Sdfr (vp->v_mount->mnt_flag & MNT_RDONLY)) 4162729Sdfr return (EROFS); 4172729Sdfr if (vap->va_size != VNOVAL) { 4182729Sdfr switch (vp->v_type) { 4192729Sdfr case VDIR: 4202729Sdfr return (EISDIR); 42183366Sjulian case VCHR: 42282607Sdillon case VBLK: 42382607Sdillon case VSOCK: 42482607Sdillon case VFIFO: 42543426Sphk return (0); 42683366Sjulian case VREG: 42782607Sdillon case VLNK: 42882607Sdillon default: 42943426Sphk /* 4302729Sdfr * Disallow write attempts if the filesystem is 4312729Sdfr * mounted read-only. 4322729Sdfr */ 4332729Sdfr if (vp->v_mount->mnt_flag & MNT_RDONLY) 4342729Sdfr return (EROFS); 4352729Sdfr } 4362729Sdfr } 4372729Sdfr return (null_bypass((struct vop_generic_args *)ap)); 4382729Sdfr} 4392729Sdfr 4402729Sdfr/* 44182607Sdillon * We handle getattr only to change the fsid. 44282607Sdillon */ 4432729Sdfrstatic int 4442729Sdfrnull_getattr(ap) 4452729Sdfr struct vop_getattr_args /* { 4462729Sdfr struct vnode *a_vp; 4472729Sdfr struct vattr *a_vap; 4482729Sdfr struct ucred *a_cred; 44934961Sphk struct proc *a_p; 4502729Sdfr } */ *ap; 4512729Sdfr{ 4522729Sdfr int error; 45383366Sjulian 4542729Sdfr if (error = null_bypass((struct vop_generic_args *)ap)) 4552729Sdfr return (error); 4562729Sdfr /* Requires that arguments be restored. */ 45782607Sdillon ap->a_vap->va_fsid = ap->a_vp->v_mount->mnt_stat.f_fsid.val[0]; 4582729Sdfr return (0); 45982607Sdillon} 4602729Sdfr 4612729Sdfrstatic int 4622729Sdfrnull_access(ap) 4632729Sdfr struct vop_access_args /* { 4642729Sdfr struct vnode *a_vp; 4652729Sdfr int a_mode; 4662729Sdfr struct ucred *a_cred; 46782607Sdillon struct proc *a_p; 46882607Sdillon } */ *ap; 4692729Sdfr{ 4702729Sdfr struct vnode *vp = ap->a_vp; 47182607Sdillon mode_t mode = ap->a_mode; 47283366Sjulian 47382607Sdillon /* 47482607Sdillon * Disallow write attempts on read-only layers; 47582607Sdillon * unless the file is a socket, fifo, or a block or 4762729Sdfr * character device resident on the file system. 4772729Sdfr */ 47812866Speter if (mode & VWRITE) { 4792729Sdfr switch (vp->v_type) { 4802729Sdfr case VDIR: 4812729Sdfr case VLNK: 4822729Sdfr case VREG: 48312866Speter if (vp->v_mount->mnt_flag & MNT_RDONLY) 4842729Sdfr return (EROFS); 48582607Sdillon break; 48682607Sdillon } 48782607Sdillon } 48812866Speter return (null_bypass((struct vop_generic_args *)ap)); 48983366Sjulian} 49083366Sjulian 4912729Sdfr/* 4922729Sdfr * We need to process our own vnode lock and then clear the 49382607Sdillon * interlock flag as it applies only to our vnode, not the 4942729Sdfr * vnodes below us on the stack. 4952729Sdfr */ 49683366Sjulianstatic int 4972836Sdgnull_lock(ap) 4982729Sdfr struct vop_lock_args /* { 4992729Sdfr struct vnode *a_vp; 5002729Sdfr int a_flags; 5012729Sdfr struct proc *a_p; 5022729Sdfr } */ *ap; 50382607Sdillon{ 50483366Sjulian 50582607Sdillon vop_nolock(ap); 50682607Sdillon if ((ap->a_flags & LK_TYPE_MASK) == LK_DRAIN) 50782607Sdillon return (0); 50882607Sdillon ap->a_flags &= ~LK_INTERLOCK; 5092729Sdfr return (null_bypass((struct vop_generic_args *)ap)); 5102729Sdfr} 5112729Sdfr 5122729Sdfr/* 5132729Sdfr * We need to process our own vnode unlock and then clear the 5142729Sdfr * interlock flag as it applies only to our vnode, not the 5152729Sdfr * vnodes below us on the stack. 5162729Sdfr */ 5172729Sdfrstatic int 5182729Sdfrnull_unlock(ap) 5192729Sdfr struct vop_unlock_args /* { 5202729Sdfr struct vnode *a_vp; 5212729Sdfr int a_flags; 5222729Sdfr struct proc *a_p; 5232729Sdfr } */ *ap; 52482607Sdillon{ 52582607Sdillon struct vnode *vp = ap->a_vp; 5262729Sdfr 52783366Sjulian vop_nounlock(ap); 5282729Sdfr ap->a_flags &= ~LK_INTERLOCK; 5292729Sdfr return (null_bypass((struct vop_generic_args *)ap)); 5302729Sdfr} 5312729Sdfr 53282607Sdillonstatic int 5332729Sdfrnull_inactive(ap) 5342729Sdfr struct vop_inactive_args /* { 5352729Sdfr struct vnode *a_vp; 5362729Sdfr struct proc *a_p; 5372729Sdfr } */ *ap; 5382729Sdfr{ 5392729Sdfr /* 5402729Sdfr * Do nothing (and _don't_ bypass). 5412729Sdfr * Wait to vrele lowervp until reclaim, 5422729Sdfr * so that until then our null_node is in the 5432729Sdfr * cache and reusable. 5442729Sdfr * 5452729Sdfr * NEEDSWORK: Someday, consider inactive'ing 5462729Sdfr * the lowervp and then trying to reactivate it 5472729Sdfr * with capabilities (v_id) 5482729Sdfr * like they do in the name lookup cache code. 5492729Sdfr * That's too much work for now. 5502729Sdfr */ 5512729Sdfr VOP_UNLOCK(ap->a_vp, 0, ap->a_p); 5522729Sdfr return (0); 5532729Sdfr} 5542729Sdfr 5552729Sdfrstatic int 5562729Sdfrnull_reclaim(ap) 5572729Sdfr struct vop_reclaim_args /* { 55882607Sdillon struct vnode *a_vp; 55982607Sdillon struct proc *a_p; 5602729Sdfr } */ *ap; 5612729Sdfr{ 5622729Sdfr struct vnode *vp = ap->a_vp; 5632729Sdfr struct null_node *xp = VTONULL(vp); 5642729Sdfr struct vnode *lowervp = xp->null_lowervp; 5652729Sdfr 5662729Sdfr /* 5672729Sdfr * Note: in vop_reclaim, vp->v_op == dead_vnodeop_p, 5682729Sdfr * so we can't call VOPs on ourself. 5692729Sdfr */ 5702729Sdfr /* After this assignment, this node will not be re-used. */ 5712729Sdfr xp->null_lowervp = NULLVP; 5722729Sdfr LIST_REMOVE(xp, null_hash); 5732729Sdfr FREE(vp->v_data, M_TEMP); 5742729Sdfr vp->v_data = NULL; 5752729Sdfr vrele (lowervp); 5762729Sdfr return (0); 5772729Sdfr} 5782729Sdfr 5792729Sdfrstatic int 5802729Sdfrnull_print(ap) 58134961Sphk struct vop_print_args /* { 5822729Sdfr struct vnode *a_vp; 5832729Sdfr } */ *ap; 5842729Sdfr{ 5852729Sdfr register struct vnode *vp = ap->a_vp; 58682607Sdillon printf ("\ttag VT_NULLFS, vp=%p, lowervp=%p\n", vp, NULLVPTOLOWERVP(vp)); 58782607Sdillon return (0); 5882729Sdfr} 5892729Sdfr 5902729Sdfr/* 5912729Sdfr * XXX - vop_strategy must be hand coded because it has no 59283366Sjulian * vnode in its arguments. 59382607Sdillon * This goes away with a merged VM/buffer cache. 59482607Sdillon */ 59582607Sdillonstatic int 5962729Sdfrnull_strategy(ap) 5972729Sdfr struct vop_strategy_args /* { 59812866Speter struct buf *a_bp; 5992729Sdfr } */ *ap; 6002729Sdfr{ 60112866Speter struct buf *bp = ap->a_bp; 6022729Sdfr int error; 6032729Sdfr struct vnode *savedvp; 6042729Sdfr 60512866Speter savedvp = bp->b_vp; 6062729Sdfr bp->b_vp = NULLVPTOLOWERVP(bp->b_vp); 60782607Sdillon 60882607Sdillon error = VOP_STRATEGY(bp); 60982607Sdillon 61012866Speter bp->b_vp = savedvp; 61183366Sjulian 61283366Sjulian return (error); 6132729Sdfr} 6142729Sdfr 6152729Sdfr/* 61612866Speter * XXX - like vop_strategy, vop_bwrite must be hand coded because it has no 6172729Sdfr * vnode in its arguments. 6182729Sdfr * This goes away with a merged VM/buffer cache. 61982607Sdillon */ 6202729Sdfrstatic int 6212729Sdfrnull_bwrite(ap) 6222729Sdfr struct vop_bwrite_args /* { 6232729Sdfr struct buf *a_bp; 6242729Sdfr } */ *ap; 6252729Sdfr{ 6262729Sdfr struct buf *bp = ap->a_bp; 6272729Sdfr int error; 62882607Sdillon struct vnode *savedvp; 62983366Sjulian 63082607Sdillon savedvp = bp->b_vp; 63182607Sdillon bp->b_vp = NULLVPTOLOWERVP(bp->b_vp); 63282607Sdillon 63368024Srwatson error = VOP_BWRITE(bp); 6342729Sdfr 6352729Sdfr bp->b_vp = savedvp; 6362729Sdfr 6372729Sdfr return (error); 6382729Sdfr} 6392729Sdfr 6402729Sdfr/* 64182607Sdillon * Global vfs data structures 64282607Sdillon */ 6432729Sdfrvop_t **null_vnodeop_p; 6442729Sdfrstatic struct vnodeopv_entry_desc null_vnodeop_entries[] = { 6452729Sdfr { &vop_default_desc, (vop_t *)null_bypass }, 6462729Sdfr 6472729Sdfr { &vop_lookup_desc, (vop_t *)null_lookup }, 6482729Sdfr { &vop_setattr_desc, (vop_t *)null_setattr }, 6492729Sdfr { &vop_getattr_desc, (vop_t *)null_getattr }, 65082607Sdillon { &vop_access_desc, (vop_t *)null_access }, 65182607Sdillon { &vop_lock_desc, (vop_t *)null_lock }, 6522729Sdfr { &vop_unlock_desc, (vop_t *)null_unlock }, 6532729Sdfr { &vop_inactive_desc, (vop_t *)null_inactive }, 6542729Sdfr { &vop_reclaim_desc, (vop_t *)null_reclaim }, 6552729Sdfr { &vop_print_desc, (vop_t *)null_print }, 6562729Sdfr 65782607Sdillon { &vop_strategy_desc, (vop_t *)null_strategy }, 65882607Sdillon { &vop_bwrite_desc, (vop_t *)null_bwrite }, 6592729Sdfr 6602729Sdfr { NULL, NULL } 66183366Sjulian}; 6622729Sdfrstatic struct vnodeopv_desc null_vnodeop_opv_desc = 6632729Sdfr { &null_vnodeop_p, null_vnodeop_entries }; 6642729Sdfr 66582607SdillonVNODEOP_SET(null_vnodeop_opv_desc); 6662729Sdfr