null_vnops.c revision 30636
119872Swollman/* 219872Swollman * Copyright (c) 1992, 1993 319872Swollman * The Regents of the University of California. All rights reserved. 419872Swollman * 519872Swollman * This code is derived from software contributed to Berkeley by 619872Swollman * John Heidemann of the UCLA Ficus project. 719872Swollman * 819872Swollman * Redistribution and use in source and binary forms, with or without 919872Swollman * modification, are permitted provided that the following conditions 1019872Swollman * are met: 1119872Swollman * 1. Redistributions of source code must retain the above copyright 1219872Swollman * notice, this list of conditions and the following disclaimer. 1319872Swollman * 2. Redistributions in binary form must reproduce the above copyright 1419872Swollman * notice, this list of conditions and the following disclaimer in the 15179530Sjkim * documentation and/or other materials provided with the distribution. 1619872Swollman * 3. All advertising materials mentioning features or use of this software 1719872Swollman * must display the following acknowledgement: 1819872Swollman * This product includes software developed by the University of 1919872Swollman * California, Berkeley and its contributors. 2019872Swollman * 4. Neither the name of the University nor the names of its contributors 2119872Swollman * may be used to endorse or promote products derived from this software 2219872Swollman * without specific prior written permission. 2319872Swollman * 2419872Swollman * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 2519872Swollman * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 2619872Swollman * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 2719872Swollman * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 2819872Swollman * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 2919872Swollman * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 3019872Swollman * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 3119872Swollman * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 3219872Swollman * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 3319872Swollman * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 3419872Swollman * SUCH DAMAGE. 35179530Sjkim * 36179530Sjkim * @(#)null_vnops.c 8.6 (Berkeley) 5/27/95 3730763Scharnier * 3819872Swollman * Ancestors: 3919872Swollman * @(#)lofs_vnops.c 1.2 (Berkeley) 6/18/92 4019872Swollman * $Id: null_vnops.c,v 1.24 1997/10/15 10:04:31 phk Exp $ 4119872Swollman * ...and... 4219872Swollman * @(#)null_vnodeops.c 1.20 92/07/07 UCLA Ficus project 4366907Swollman * 4419872Swollman * $Id: null_vnops.c,v 1.24 1997/10/15 10:04:31 phk Exp $ 4519872Swollman */ 4619872Swollman 47198267Sedwin/* 4819872Swollman * Null Layer 4919872Swollman * 50274394Sdteske * (See mount_null(8) for more information.) 5119872Swollman * 52227934Sfjoe * The null layer duplicates a portion of the file system 53227934Sfjoe * name space under a new name. In this respect, it is 54179530Sjkim * similar to the loopback file system. It differs from 55179530Sjkim * the loopback fs in two respects: it is implemented using 56179530Sjkim * a stackable layers techniques, and it's "null-node"s stack above 57179530Sjkim * all lower-layer vnodes, not just over directory vnodes. 58198267Sedwin * 59179530Sjkim * The null layer has two purposes. First, it serves as a demonstration 6019872Swollman * of layering by proving a layer which does nothing. (It actually 61230005Swollman * does everything the loopback file system does, which is slightly 62230005Swollman * more than nothing.) Second, the null layer can serve as a prototype 63230005Swollman * layer. Since it provides all necessary layer framework, 64230005Swollman * new file system layers can be created very easily be starting 65230005Swollman * with a null layer. 66230005Swollman * 67230005Swollman * The remainder of this man page examines the null layer as a basis 68227934Sfjoe * for constructing new layers. 69227934Sfjoe * 70227934Sfjoe * 71227934Sfjoe * INSTANTIATING NEW NULL LAYERS 72227934Sfjoe * 73227934Sfjoe * New null layers are created with mount_null(8). 74227934Sfjoe * Mount_null(8) takes two arguments, the pathname 75227934Sfjoe * of the lower vfs (target-pn) and the pathname where the null 76227934Sfjoe * layer will appear in the namespace (alias-pn). After 77227934Sfjoe * the null layer is put into place, the contents 78227934Sfjoe * of target-pn subtree will be aliased under alias-pn. 79227934Sfjoe * 80227934Sfjoe * 81227934Sfjoe * OPERATION OF A NULL LAYER 82227934Sfjoe * 83227934Sfjoe * The null layer is the minimum file system layer, 84228176Sfjoe * simply bypassing all possible operations to the lower layer 85228176Sfjoe * for processing there. The majority of its activity centers 86228176Sfjoe * on the bypass routine, through which nearly all vnode operations 87228176Sfjoe * pass. 88228176Sfjoe * 89228176Sfjoe * The bypass routine accepts arbitrary vnode operations for 90228176Sfjoe * handling by the lower layer. It begins by examing vnode 91228176Sfjoe * operation arguments and replacing any null-nodes by their 92228176Sfjoe * lower-layer equivlants. It then invokes the operation 93228176Sfjoe * on the lower layer. Finally, it replaces the null-nodes 94228176Sfjoe * in the arguments and, if a vnode is return by the operation, 95228176Sfjoe * stacks a null-node on top of the returned vnode. 96228176Sfjoe * 97228176Sfjoe * Although bypass handles most operations, vop_getattr, vop_lock, 98228176Sfjoe * vop_unlock, vop_inactive, vop_reclaim, and vop_print are not 99228176Sfjoe * bypassed. Vop_getattr must change the fsid being returned. 100228176Sfjoe * Vop_lock and vop_unlock must handle any locking for the 101228176Sfjoe * current vnode as well as pass the lock request down. 102228176Sfjoe * Vop_inactive and vop_reclaim are not bypassed so that 103228176Sfjoe * they can handle freeing null-layer specific data. Vop_print 104228176Sfjoe * is not bypassed to avoid excessive debugging information. 105228176Sfjoe * Also, certain vnode operations change the locking state within 106228176Sfjoe * the operation (create, mknod, remove, link, rename, mkdir, rmdir, 107228176Sfjoe * and symlink). Ideally these operations should not change the 108228176Sfjoe * lock state, but should be changed to let the caller of the 109228176Sfjoe * function unlock them. Otherwise all intermediate vnode layers 110228176Sfjoe * (such as union, umapfs, etc) must catch these functions to do 111228176Sfjoe * the necessary locking at their layer. 112228176Sfjoe * 113228176Sfjoe * 114228176Sfjoe * INSTANTIATING VNODE STACKS 115228176Sfjoe * 116227934Sfjoe * Mounting associates the null layer with a lower layer, 117227934Sfjoe * effect stacking two VFSes. Vnode stacks are instead 118227934Sfjoe * created on demand as files are accessed. 119227947Sfjoe * 120227934Sfjoe * The initial mount creates a single vnode stack for the 121227934Sfjoe * root of the new null layer. All other vnode stacks 122227934Sfjoe * are created as a result of vnode operations on 123227934Sfjoe * this or other null vnode stacks. 124227934Sfjoe * 125227934Sfjoe * New vnode stacks come into existance as a result of 126227947Sfjoe * an operation which returns a vnode. 127227934Sfjoe * The bypass routine stacks a null-node above the new 128227934Sfjoe * vnode before returning it to the caller. 129227934Sfjoe * 130227934Sfjoe * For example, imagine mounting a null layer with 131227934Sfjoe * "mount_null /usr/include /dev/layer/null". 132227934Sfjoe * Changing directory to /dev/layer/null will assign 133228176Sfjoe * the root null-node (which was created when the null layer was mounted). 134228176Sfjoe * Now consider opening "sys". A vop_lookup would be 135228176Sfjoe * done on the root null-node. This operation would bypass through 136228176Sfjoe * to the lower layer which would return a vnode representing 137228176Sfjoe * the UFS "sys". Null_bypass then builds a null-node 138228176Sfjoe * aliasing the UFS "sys" and returns this to the caller. 139227934Sfjoe * Later operations on the null-node "sys" will repeat this 140227934Sfjoe * process when constructing other vnode stacks. 141227934Sfjoe * 142227934Sfjoe * 143227934Sfjoe * CREATING OTHER FILE SYSTEM LAYERS 144227934Sfjoe * 145227934Sfjoe * One of the easiest ways to construct new file system layers is to make 146227934Sfjoe * a copy of the null layer, rename all files and variables, and 147227934Sfjoe * then begin modifing the copy. Sed can be used to easily rename 148227934Sfjoe * all variables. 149227934Sfjoe * 150227934Sfjoe * The umap layer is an example of a layer descended from the 151227934Sfjoe * null layer. 152228176Sfjoe * 153227934Sfjoe * 154227934Sfjoe * INVOKING OPERATIONS ON LOWER LAYERS 155227934Sfjoe * 156227934Sfjoe * There are two techniques to invoke operations on a lower layer 157227934Sfjoe * when the operation cannot be completely bypassed. Each method 158227934Sfjoe * is appropriate in different situations. In both cases, 159227934Sfjoe * it is the responsibility of the aliasing layer to make 160227947Sfjoe * the operation arguments "correct" for the lower layer 161227934Sfjoe * by mapping an vnode arguments to the lower layer. 162227934Sfjoe * 163227934Sfjoe * The first approach is to call the aliasing layer's bypass routine. 164227934Sfjoe * This method is most suitable when you wish to invoke the operation 165227934Sfjoe * currently being handled on the lower layer. It has the advantage 166227934Sfjoe * that the bypass routine already must do argument mapping. 167227934Sfjoe * An example of this is null_getattrs in the null layer. 168227934Sfjoe * 169227934Sfjoe * A second approach is to directly invoke vnode operations on 170227934Sfjoe * the lower layer with the VOP_OPERATIONNAME interface. 171227934Sfjoe * The advantage of this method is that it is easy to invoke 172227934Sfjoe * arbitrary operations on the lower layer. The disadvantage 173227934Sfjoe * is that vnode arguments must be manualy mapped. 174227934Sfjoe * 175227934Sfjoe */ 176227934Sfjoe 177227934Sfjoe#include <sys/param.h> 178227934Sfjoe#include <sys/systm.h> 179227934Sfjoe#include <sys/kernel.h> 180227934Sfjoe#include <sys/sysctl.h> 181227934Sfjoe#include <sys/vnode.h> 182227934Sfjoe#include <sys/mount.h> 183227934Sfjoe#include <sys/namei.h> 184227934Sfjoe#include <sys/malloc.h> 185227934Sfjoe#include <sys/buf.h> 186227934Sfjoe#include <miscfs/nullfs/null.h> 187227934Sfjoe 188227934Sfjoestatic int null_bug_bypass = 0; /* for debugging: enables bypass printf'ing */ 189227934SfjoeSYSCTL_INT(_debug, OID_AUTO, nullfs_bug_bypass, CTLFLAG_RW, 190198350Sedwin &null_bug_bypass, 0, ""); 191227934Sfjoe 192198350Sedwinstatic int null_access __P((struct vop_access_args *ap)); 193198350Sedwinstatic int null_bwrite __P((struct vop_bwrite_args *ap)); 19419872Swollmanstatic int null_getattr __P((struct vop_getattr_args *ap)); 195198267Sedwinstatic int null_inactive __P((struct vop_inactive_args *ap)); 196198350Sedwinstatic int null_lock __P((struct vop_lock_args *ap)); 197198350Sedwinstatic int null_lookup __P((struct vop_lookup_args *ap)); 19819872Swollmanstatic int null_print __P((struct vop_print_args *ap)); 199179530Sjkimstatic int null_reclaim __P((struct vop_reclaim_args *ap)); 200220172Sedwinstatic int null_setattr __P((struct vop_setattr_args *ap)); 201179530Sjkimstatic int null_strategy __P((struct vop_strategy_args *ap)); 202220172Sedwinstatic int null_unlock __P((struct vop_unlock_args *ap)); 203179530Sjkim 204179530Sjkim/* 205179530Sjkim * This is the 10-Apr-92 bypass routine. 206220172Sedwin * This version has been optimized for speed, throwing away some 20719872Swollman * safety checks. It should still always work, but it's not as 20819872Swollman * robust to programmer errors. 20919872Swollman * Define SAFETY to include some error checking code. 210179530Sjkim * 21119872Swollman * In general, we map all vnodes going down and unmap them on the way back. 21219872Swollman * As an exception to this, vnodes can be marked "unmapped" by setting 213179530Sjkim * the Nth bit in operation's vdesc_flags. 214220172Sedwin * 21519872Swollman * Also, some BSD vnode operations have the side effect of vrele'ing 21619872Swollman * their arguments. With stacking, the reference counts are held 217179530Sjkim * by the upper node, not the lower one, so we must handle these 21819872Swollman * side-effects here. This is not of concern in Sun-derived systems 21919872Swollman * since there are no such side-effects. 220179530Sjkim * 221179530Sjkim * This makes the following assumptions: 222179530Sjkim * - only one returned vpp 223179530Sjkim * - no INOUT vpp's (Sun's vop_open has one of these) 224179530Sjkim * - the vnode operation vector of the first vnode should be used 225179530Sjkim * to determine what implementation of the op should be invoked 226179530Sjkim * - all mapped vnodes are of our vnode-type (NEEDSWORK: 227179530Sjkim * problems on rmdir'ing mount points and renaming?) 228179530Sjkim */ 229220172Sedwinint 230227934Sfjoenull_bypass(ap) 23119872Swollman struct vop_generic_args /* { 23219872Swollman struct vnodeop_desc *a_desc; 233179530Sjkim <other random data follows, presumably> 234179530Sjkim } */ *ap; 235179530Sjkim{ 236179530Sjkim register struct vnode **this_vp_p; 237179530Sjkim int error; 238179530Sjkim struct vnode *old_vps[VDESC_MAX_VPS]; 239179530Sjkim struct vnode **vps_p[VDESC_MAX_VPS]; 240179530Sjkim struct vnode ***vppp; 241179530Sjkim struct vnodeop_desc *descp = ap->a_desc; 242179530Sjkim int reles, i; 243179530Sjkim 244179530Sjkim if (null_bug_bypass) 245179530Sjkim printf ("null_bypass: %s\n", descp->vdesc_name); 246220172Sedwin 247220172Sedwin#ifdef SAFETY 24819872Swollman /* 24919872Swollman * We require at least one vp. 250179530Sjkim */ 251179530Sjkim if (descp->vdesc_vp_offsets == NULL || 252179530Sjkim descp->vdesc_vp_offsets[0] == VDESC_NO_OFFSET) 253179530Sjkim panic ("null_bypass: no vp's in map."); 254179530Sjkim#endif 255179530Sjkim 25619872Swollman /* 25719872Swollman * Map the vnodes going in. 25819872Swollman * Later, we'll invoke the operation based on 259179530Sjkim * the first mapped vnode's operation vector. 26019872Swollman */ 261179530Sjkim reles = descp->vdesc_flags; 262179530Sjkim for (i = 0; i < VDESC_MAX_VPS; reles >>= 1, i++) { 263179530Sjkim if (descp->vdesc_vp_offsets[i] == VDESC_NO_OFFSET) 26419872Swollman break; /* bail out at end of list */ 265220172Sedwin vps_p[i] = this_vp_p = 266227934Sfjoe VOPARG_OFFSETTO(struct vnode**,descp->vdesc_vp_offsets[i],ap); 267220172Sedwin /* 26819872Swollman * We're not guaranteed that any but the first vnode 269179497Sjkim * are of our type. Check for and don't map any 270179497Sjkim * that aren't. (We must always map first vp or vclean fails.) 27119872Swollman */ 27219872Swollman if (i && (*this_vp_p == NULLVP || 273179530Sjkim (*this_vp_p)->v_op != null_vnodeop_p)) { 274179530Sjkim old_vps[i] = NULLVP; 275179530Sjkim } else { 276179530Sjkim old_vps[i] = *this_vp_p; 277179530Sjkim *(vps_p[i]) = NULLVPTOLOWERVP(*this_vp_p); 278179530Sjkim /* 279179530Sjkim * XXX - Several operations have the side effect 280179530Sjkim * of vrele'ing their vp's. We must account for 281179530Sjkim * that. (This should go away in the future.) 28219872Swollman */ 28319872Swollman if (reles & 1) 284227934Sfjoe VREF(*this_vp_p); 285227934Sfjoe } 28619872Swollman 287179530Sjkim } 288179530Sjkim 28919872Swollman /* 29019872Swollman * Call the operation on the lower layer 29119872Swollman * with the modified argument structure. 29219872Swollman */ 29319872Swollman error = VCALL(*(vps_p[0]), descp->vdesc_offset, ap); 294179530Sjkim 29519872Swollman /* 296179530Sjkim * Maintain the illusion of call-by-value 29719872Swollman * by restoring vnodes in the argument structure 298179530Sjkim * to their original value. 299179530Sjkim */ 30019872Swollman reles = descp->vdesc_flags; 30119872Swollman for (i = 0; i < VDESC_MAX_VPS; reles >>= 1, i++) { 30219872Swollman if (descp->vdesc_vp_offsets[i] == VDESC_NO_OFFSET) 303179530Sjkim break; /* bail out at end of list */ 304179530Sjkim if (old_vps[i]) { 305179530Sjkim *(vps_p[i]) = old_vps[i]; 306179530Sjkim if (reles & 1) 307179530Sjkim vrele(*(vps_p[i])); 308179530Sjkim } 309179530Sjkim } 31019872Swollman 31119872Swollman /* 31219872Swollman * Map the possible out-going vpp 31360938Sjake * (Assumes that the lower layer always returns 314179530Sjkim * a VREF'ed vpp unless it gets an error.) 315179530Sjkim */ 31619872Swollman if (descp->vdesc_vpp_offset != VDESC_NO_OFFSET && 31719872Swollman !(descp->vdesc_flags & VDESC_NOMAP_VPP) && 31819872Swollman !error) { 31919872Swollman /* 32019872Swollman * XXX - even though some ops have vpp returned vp's, 32119872Swollman * several ops actually vrele this before returning. 32219872Swollman * We must avoid these ops. 32319872Swollman * (This should go away when these ops are regularized.) 324179530Sjkim */ 32522181Sjhay if (descp->vdesc_flags & VDESC_VPP_WILLRELE) 32619872Swollman goto out; 327179530Sjkim vppp = VOPARG_OFFSETTO(struct vnode***, 328179530Sjkim descp->vdesc_vpp_offset,ap); 32919872Swollman if (*vppp) 33019872Swollman error = null_node_create(old_vps[0]->v_mount, **vppp, *vppp); 33119872Swollman } 33219872Swollman 33319872Swollman out: 33419872Swollman return (error); 33519872Swollman} 336179530Sjkim 337179530Sjkim/* 338179530Sjkim * We have to carry on the locking protocol on the null layer vnodes 339179530Sjkim * as we progress through the tree. We also have to enforce read-only 340179530Sjkim * if this layer is mounted read-only. 34119872Swollman */ 342198350Sedwinstatic int 34319872Swollmannull_lookup(ap) 344209190Semaste struct vop_lookup_args /* { 34519872Swollman struct vnode * a_dvp; 34619872Swollman struct vnode ** a_vpp; 347298033Saraujo struct componentname * a_cnp; 34819872Swollman } */ *ap; 34919872Swollman{ 350198350Sedwin struct componentname *cnp = ap->a_cnp; 35119872Swollman struct proc *p = cnp->cn_proc; 35230999Sjoerg int flags = cnp->cn_flags; 35319872Swollman struct vop_lock_args lockargs; 35419872Swollman struct vop_unlock_args unlockargs; 35519872Swollman struct vnode *dvp, *vp; 35619872Swollman int error; 357298033Saraujo 358198350Sedwin if ((flags & ISLASTCN) && (ap->a_dvp->v_mount->mnt_flag & MNT_RDONLY) && 35919872Swollman (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) 360198350Sedwin return (EROFS); 361179530Sjkim error = null_bypass((struct vop_generic_args *)ap); 36219872Swollman if (error == EJUSTRETURN && (flags & ISLASTCN) && 36319872Swollman (ap->a_dvp->v_mount->mnt_flag & MNT_RDONLY) && 364179530Sjkim (cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME)) 365298033Saraujo error = EROFS; 366198350Sedwin /* 367179530Sjkim * We must do the same locking and unlocking at this layer as 368298033Saraujo * is done in the layers below us. We could figure this out 369198350Sedwin * based on the error return and the LASTCN, LOCKPARENT, and 37019872Swollman * LOCKLEAF flags. However, it is more expidient to just find 37119872Swollman * out the state of the lower level vnodes and set ours to the 37219872Swollman * same state. 37319872Swollman */ 37419872Swollman dvp = ap->a_dvp; 375198350Sedwin vp = *ap->a_vpp; 376198350Sedwin if (dvp == vp) 37719872Swollman return (error); 37856487Scharnier if (!VOP_ISLOCKED(dvp)) { 37956487Scharnier unlockargs.a_vp = dvp; 38019872Swollman unlockargs.a_flags = 0; 38156487Scharnier unlockargs.a_p = p; 38256487Scharnier vop_nounlock(&unlockargs); 38319872Swollman } 38419872Swollman if (vp != NULLVP && VOP_ISLOCKED(vp)) { 38519872Swollman lockargs.a_vp = vp; 38619872Swollman lockargs.a_flags = LK_SHARED; 38719872Swollman lockargs.a_p = p; 38819872Swollman vop_nolock(&lockargs); 38919872Swollman } 390179530Sjkim return (error); 39119872Swollman} 392179530Sjkim 393179530Sjkim/* 39419872Swollman * Setattr call. Disallow write attempts if the layer is mounted read-only. 39519872Swollman */ 396198350Sedwinint 397179530Sjkimnull_setattr(ap) 398227934Sfjoe struct vop_setattr_args /* { 39919872Swollman struct vnodeop_desc *a_desc; 40019872Swollman struct vnode *a_vp; 401198350Sedwin struct vattr *a_vap; 402179530Sjkim struct ucred *a_cred; 40319872Swollman struct proc *a_p; 40419872Swollman } */ *ap; 40519872Swollman{ 406198350Sedwin struct vnode *vp = ap->a_vp; 407198350Sedwin struct vattr *vap = ap->a_vap; 40819872Swollman 409179530Sjkim if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL || 410298033Saraujo vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL || 411179530Sjkim vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) && 412227934Sfjoe (vp->v_mount->mnt_flag & MNT_RDONLY)) 41319872Swollman return (EROFS); 41419872Swollman if (vap->va_size != VNOVAL) { 41519872Swollman switch (vp->v_type) { 41619872Swollman case VDIR: 41756487Scharnier return (EISDIR); 41856487Scharnier case VCHR: 41919872Swollman case VBLK: 42056487Scharnier case VSOCK: 42156487Scharnier case VFIFO: 42219872Swollman return (0); 42319872Swollman case VREG: 42419872Swollman case VLNK: 42519872Swollman default: 42619872Swollman /* 427198350Sedwin * Disallow write attempts if the filesystem is 428198350Sedwin * mounted read-only. 42919872Swollman */ 430198350Sedwin if (vp->v_mount->mnt_flag & MNT_RDONLY) 431198350Sedwin return (EROFS); 43219872Swollman } 43319872Swollman } 43456487Scharnier return (null_bypass((struct vop_generic_args *)ap)); 43556487Scharnier} 43619872Swollman 43719872Swollman/* 43819872Swollman * We handle getattr only to change the fsid. 43919872Swollman */ 44019872Swollmanstatic int 44119872Swollmannull_getattr(ap) 44219872Swollman struct vop_getattr_args /* { 44319872Swollman struct vnode *a_vp; 44419872Swollman struct vattr *a_vap; 44519872Swollman struct ucred *a_cred; 44619872Swollman struct proc *a_p; 44719872Swollman } */ *ap; 44819872Swollman{ 44919872Swollman int error; 45019872Swollman 451179530Sjkim if (error = null_bypass((struct vop_generic_args *)ap)) 45219872Swollman return (error); 453179530Sjkim /* Requires that arguments be restored. */ 45419872Swollman ap->a_vap->va_fsid = ap->a_vp->v_mount->mnt_stat.f_fsid.val[0]; 455179530Sjkim return (0); 45619872Swollman} 457179530Sjkim 45819872Swollmanstatic int 45919872Swollmannull_access(ap) 46019872Swollman struct vop_access_args /* { 46119872Swollman struct vnode *a_vp; 46219872Swollman int a_mode; 46319872Swollman struct ucred *a_cred; 46419872Swollman struct proc *a_p; 46519872Swollman } */ *ap; 46619872Swollman{ 467179530Sjkim struct vnode *vp = ap->a_vp; 468179530Sjkim mode_t mode = ap->a_mode; 46919872Swollman 47019872Swollman /* 47119872Swollman * Disallow write attempts on read-only layers; 47219872Swollman * unless the file is a socket, fifo, or a block or 47319872Swollman * character device resident on the file system. 474179530Sjkim */ 475179530Sjkim if (mode & VWRITE) { 47619872Swollman switch (vp->v_type) { 477179530Sjkim case VDIR: 478281733Seadler case VLNK: 479179530Sjkim case VREG: 48019872Swollman if (vp->v_mount->mnt_flag & MNT_RDONLY) 481198350Sedwin return (EROFS); 48219872Swollman break; 483209190Semaste } 48419872Swollman } 48519872Swollman return (null_bypass((struct vop_generic_args *)ap)); 486298033Saraujo} 48719872Swollman 48819872Swollman/* 489198350Sedwin * We need to process our own vnode lock and then clear the 49019872Swollman * interlock flag as it applies only to our vnode, not the 49119872Swollman * vnodes below us on the stack. 49219872Swollman */ 49319872Swollmanstatic int 49419872Swollmannull_lock(ap) 49519872Swollman struct vop_lock_args /* { 496198350Sedwin struct vnode *a_vp; 497198350Sedwin int a_flags; 498281733Seadler struct proc *a_p; 49919872Swollman } */ *ap; 50019872Swollman{ 501298033Saraujo 502198350Sedwin vop_nolock(ap); 503179530Sjkim if ((ap->a_flags & LK_TYPE_MASK) == LK_DRAIN) 50419872Swollman return (0); 50519872Swollman ap->a_flags &= ~LK_INTERLOCK; 50619872Swollman return (null_bypass((struct vop_generic_args *)ap)); 50719872Swollman} 508198350Sedwin 509179530Sjkim/* 51019872Swollman * We need to process our own vnode unlock and then clear the 511179530Sjkim * interlock flag as it applies only to our vnode, not the 51219872Swollman * vnodes below us on the stack. 51319872Swollman */ 51419872Swollmanstatic int 51519872Swollmannull_unlock(ap) 51619872Swollman struct vop_unlock_args /* { 51719872Swollman struct vnode *a_vp; 51819872Swollman int a_flags; 51919872Swollman struct proc *a_p; 52019872Swollman } */ *ap; 521179530Sjkim{ 522179530Sjkim struct vnode *vp = ap->a_vp; 52319872Swollman 524179530Sjkim vop_nounlock(ap); 525179530Sjkim ap->a_flags &= ~LK_INTERLOCK; 52619872Swollman return (null_bypass((struct vop_generic_args *)ap)); 52719872Swollman} 52819872Swollman 52919872Swollmanstatic int 53019872Swollmannull_inactive(ap) 53119872Swollman struct vop_inactive_args /* { 53219872Swollman struct vnode *a_vp; 53319872Swollman struct proc *a_p; 53419872Swollman } */ *ap; 53519872Swollman{ 53619872Swollman struct vnode *vp = ap->a_vp; 53719872Swollman struct null_node *xp = VTONULL(vp); 53819872Swollman struct vnode *lowervp = xp->null_lowervp; 53970486Sben /* 54019872Swollman * Do nothing (and _don't_ bypass). 54170486Sben * Wait to vrele lowervp until reclaim, 542179530Sjkim * so that until then our null_node is in the 543179530Sjkim * cache and reusable. 54419872Swollman * We still have to tell the lower layer the vnode 54519872Swollman * is now inactive though. 54619872Swollman * 54719872Swollman * NEEDSWORK: Someday, consider inactive'ing 54819872Swollman * the lowervp and then trying to reactivate it 54919872Swollman * with capabilities (v_id) 55019872Swollman * like they do in the name lookup cache code. 55119872Swollman * That's too much work for now. 552179530Sjkim */ 553179530Sjkim VOP_INACTIVE(lowervp, ap->a_p); 554179530Sjkim VOP_UNLOCK(ap->a_vp, 0, ap->a_p); 55519872Swollman return (0); 556179530Sjkim} 55719872Swollman 55819872Swollmanstatic int 559179530Sjkimnull_reclaim(ap) 560179530Sjkim struct vop_reclaim_args /* { 561298033Saraujo struct vnode *a_vp; 56256487Scharnier struct proc *a_p; 56319872Swollman } */ *ap; 564179530Sjkim{ 565179530Sjkim struct vnode *vp = ap->a_vp; 566179530Sjkim struct null_node *xp = VTONULL(vp); 567179530Sjkim struct vnode *lowervp = xp->null_lowervp; 56819872Swollman 56919872Swollman /* 57019872Swollman * Note: in vop_reclaim, vp->v_op == dead_vnodeop_p, 57119872Swollman * so we can't call VOPs on ourself. 57219872Swollman */ 57319872Swollman /* After this assignment, this node will not be re-used. */ 57419872Swollman xp->null_lowervp = NULLVP; 57519872Swollman LIST_REMOVE(xp, null_hash); 57619872Swollman FREE(vp->v_data, M_TEMP); 57719872Swollman vp->v_data = NULL; 57819872Swollman vrele (lowervp); 57919872Swollman return (0); 580179530Sjkim} 581179530Sjkim 58219872Swollmanstatic int 58319872Swollmannull_print(ap) 58419872Swollman struct vop_print_args /* { 58519872Swollman struct vnode *a_vp; 586179530Sjkim } */ *ap; 58719872Swollman{ 58856487Scharnier register struct vnode *vp = ap->a_vp; 58919872Swollman printf ("\ttag VT_NULLFS, vp=%p, lowervp=%p\n", vp, NULLVPTOLOWERVP(vp)); 59070486Sben return (0); 59119872Swollman} 59219872Swollman 593179530Sjkim/* 594179530Sjkim * XXX - vop_strategy must be hand coded because it has no 59519872Swollman * vnode in its arguments. 59619872Swollman * This goes away with a merged VM/buffer cache. 59719872Swollman */ 59819872Swollmanstatic int 59970486Sbennull_strategy(ap) 600179530Sjkim struct vop_strategy_args /* { 601179530Sjkim struct buf *a_bp; 60219872Swollman } */ *ap; 60319872Swollman{ 60419872Swollman struct buf *bp = ap->a_bp; 60519872Swollman int error; 60619872Swollman struct vnode *savedvp; 607179530Sjkim 60819872Swollman savedvp = bp->b_vp; 60919872Swollman bp->b_vp = NULLVPTOLOWERVP(bp->b_vp); 61019872Swollman 61119872Swollman error = VOP_STRATEGY(bp); 61219872Swollman 61319872Swollman bp->b_vp = savedvp; 61419872Swollman 61519872Swollman return (error); 61619872Swollman} 61719872Swollman 61819872Swollman/* 61919872Swollman * XXX - like vop_strategy, vop_bwrite must be hand coded because it has no 620179530Sjkim * vnode in its arguments. 621179530Sjkim * This goes away with a merged VM/buffer cache. 622179530Sjkim */ 623179530Sjkimstatic int 62419872Swollmannull_bwrite(ap) 625179530Sjkim struct vop_bwrite_args /* { 626179530Sjkim struct buf *a_bp; 627179530Sjkim } */ *ap; 62819872Swollman{ 629227934Sfjoe struct buf *bp = ap->a_bp; 630227934Sfjoe int error; 63119872Swollman struct vnode *savedvp; 632179530Sjkim 633179530Sjkim savedvp = bp->b_vp; 63419872Swollman bp->b_vp = NULLVPTOLOWERVP(bp->b_vp); 63519872Swollman 636220172Sedwin error = VOP_BWRITE(bp); 637220172Sedwin 638220172Sedwin bp->b_vp = savedvp; 639220172Sedwin 640220172Sedwin return (error); 641227934Sfjoe} 642220172Sedwin 643220172Sedwin/* 644220172Sedwin * Global vfs data structures 64519872Swollman */ 646198350Sedwinvop_t **null_vnodeop_p; 64719872Swollmanstatic struct vnodeopv_entry_desc null_vnodeop_entries[] = { 648179530Sjkim { &vop_default_desc, (vop_t *) null_bypass }, 649230005Swollman { &vop_access_desc, (vop_t *) null_access }, 650179530Sjkim { &vop_bwrite_desc, (vop_t *) null_bwrite }, 651179530Sjkim { &vop_getattr_desc, (vop_t *) null_getattr }, 652179530Sjkim { &vop_inactive_desc, (vop_t *) null_inactive }, 65319872Swollman { &vop_lock_desc, (vop_t *) null_lock }, 654198350Sedwin { &vop_lookup_desc, (vop_t *) null_lookup }, 65519872Swollman { &vop_print_desc, (vop_t *) null_print }, 65619872Swollman { &vop_reclaim_desc, (vop_t *) null_reclaim }, 657179530Sjkim { &vop_setattr_desc, (vop_t *) null_setattr }, 65819872Swollman { &vop_strategy_desc, (vop_t *) null_strategy }, 65919872Swollman { &vop_unlock_desc, (vop_t *) null_unlock }, 66019872Swollman { NULL, NULL } 66119872Swollman}; 66221915Sjkhstatic struct vnodeopv_desc null_vnodeop_opv_desc = 663230299Semaste { &null_vnodeop_p, null_vnodeop_entries }; 664230299Semaste 665179530SjkimVNODEOP_SET(null_vnodeop_opv_desc); 666230299Semaste