null_vnops.c revision 22521
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 * 3. All advertising materials mentioning features or use of this software 17 * must display the following acknowledgement: 18 * This product includes software developed by the University of 19 * California, Berkeley and its contributors. 20 * 4. Neither the name of the University nor the names of its contributors 21 * may be used to endorse or promote products derived from this software 22 * without specific prior written permission. 23 * 24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 34 * SUCH DAMAGE. 35 * 36 * @(#)null_vnops.c 8.6 (Berkeley) 5/27/95 37 * 38 * Ancestors: 39 * @(#)lofs_vnops.c 1.2 (Berkeley) 6/18/92 40 * $Id: null_vnops.c,v 1.11.2000.1 1996/09/17 14:32:31 peter Exp $ 41 * ...and... 42 * @(#)null_vnodeops.c 1.20 92/07/07 UCLA Ficus project 43 * 44 * $FreeBSD: head/sys/fs/nullfs/null_vnops.c 22521 1997-02-10 02:22:35Z dyson $ 45 */ 46 47/* 48 * Null Layer 49 * 50 * (See mount_null(8) for more information.) 51 * 52 * The null layer duplicates a portion of the file system 53 * name space under a new name. In this respect, it is 54 * similar to the loopback file system. It differs from 55 * the loopback fs in two respects: it is implemented using 56 * a stackable layers techniques, and it's "null-node"s stack above 57 * all lower-layer vnodes, not just over directory vnodes. 58 * 59 * The null layer has two purposes. First, it serves as a demonstration 60 * of layering by proving a layer which does nothing. (It actually 61 * does everything the loopback file system does, which is slightly 62 * more than nothing.) Second, the null layer can serve as a prototype 63 * layer. Since it provides all necessary layer framework, 64 * new file system layers can be created very easily be starting 65 * with a null layer. 66 * 67 * The remainder of this man page examines the null layer as a basis 68 * for constructing new layers. 69 * 70 * 71 * INSTANTIATING NEW NULL LAYERS 72 * 73 * New null layers are created with mount_null(8). 74 * Mount_null(8) takes two arguments, the pathname 75 * of the lower vfs (target-pn) and the pathname where the null 76 * layer will appear in the namespace (alias-pn). After 77 * the null layer is put into place, the contents 78 * of target-pn subtree will be aliased under alias-pn. 79 * 80 * 81 * OPERATION OF A NULL LAYER 82 * 83 * The null layer is the minimum file system layer, 84 * simply bypassing all possible operations to the lower layer 85 * for processing there. The majority of its activity centers 86 * on the bypass routine, though which nearly all vnode operations 87 * pass. 88 * 89 * The bypass routine accepts arbitrary vnode operations for 90 * handling by the lower layer. It begins by examing vnode 91 * operation arguments and replacing any null-nodes by their 92 * lower-layer equivlants. It then invokes the operation 93 * on the lower layer. Finally, it replaces the null-nodes 94 * in the arguments and, if a vnode is return by the operation, 95 * stacks a null-node on top of the returned vnode. 96 * 97 * Although bypass handles most operations, vop_getattr, vop_lock, 98 * vop_unlock, vop_inactive, vop_reclaim, and vop_print are not 99 * bypassed. Vop_getattr must change the fsid being returned. 100 * Vop_lock and vop_unlock must handle any locking for the 101 * current vnode as well as pass the lock request down. 102 * Vop_inactive and vop_reclaim are not bypassed so that 103 * they can handle freeing null-layer specific data. Vop_print 104 * is not bypassed to avoid excessive debugging information. 105 * Also, certain vnode operations change the locking state within 106 * the operation (create, mknod, remove, link, rename, mkdir, rmdir, 107 * and symlink). Ideally these operations should not change the 108 * lock state, but should be changed to let the caller of the 109 * function unlock them. Otherwise all intermediate vnode layers 110 * (such as union, umapfs, etc) must catch these functions to do 111 * the necessary locking at their layer. 112 * 113 * 114 * INSTANTIATING VNODE STACKS 115 * 116 * Mounting associates the null layer with a lower layer, 117 * effect stacking two VFSes. Vnode stacks are instead 118 * created on demand as files are accessed. 119 * 120 * The initial mount creates a single vnode stack for the 121 * root of the new null layer. All other vnode stacks 122 * are created as a result of vnode operations on 123 * this or other null vnode stacks. 124 * 125 * New vnode stacks come into existance as a result of 126 * an operation which returns a vnode. 127 * The bypass routine stacks a null-node above the new 128 * vnode before returning it to the caller. 129 * 130 * For example, imagine mounting a null layer with 131 * "mount_null /usr/include /dev/layer/null". 132 * Changing directory to /dev/layer/null will assign 133 * the root null-node (which was created when the null layer was mounted). 134 * Now consider opening "sys". A vop_lookup would be 135 * done on the root null-node. This operation would bypass through 136 * to the lower layer which would return a vnode representing 137 * the UFS "sys". Null_bypass then builds a null-node 138 * aliasing the UFS "sys" and returns this to the caller. 139 * Later operations on the null-node "sys" will repeat this 140 * process when constructing other vnode stacks. 141 * 142 * 143 * CREATING OTHER FILE SYSTEM LAYERS 144 * 145 * One of the easiest ways to construct new file system layers is to make 146 * a copy of the null layer, rename all files and variables, and 147 * then begin modifing the copy. Sed can be used to easily rename 148 * all variables. 149 * 150 * The umap layer is an example of a layer descended from the 151 * null layer. 152 * 153 * 154 * INVOKING OPERATIONS ON LOWER LAYERS 155 * 156 * There are two techniques to invoke operations on a lower layer 157 * when the operation cannot be completely bypassed. Each method 158 * is appropriate in different situations. In both cases, 159 * it is the responsibility of the aliasing layer to make 160 * the operation arguments "correct" for the lower layer 161 * by mapping an vnode arguments to the lower layer. 162 * 163 * The first approach is to call the aliasing layer's bypass routine. 164 * This method is most suitable when you wish to invoke the operation 165 * currently being hanldled on the lower layer. It has the advantage 166 * that the bypass routine already must do argument mapping. 167 * An example of this is null_getattrs in the null layer. 168 * 169 * A second approach is to directly invoked vnode operations on 170 * the lower layer with the VOP_OPERATIONNAME interface. 171 * The advantage of this method is that it is easy to invoke 172 * arbitrary operations on the lower layer. The disadvantage 173 * is that vnodes arguments must be manualy mapped. 174 * 175 */ 176 177#include <sys/param.h> 178#include <sys/systm.h> 179#include <sys/kernel.h> 180#include <sys/sysctl.h> 181#include <sys/proc.h> 182#include <sys/time.h> 183#include <sys/types.h> 184#include <sys/vnode.h> 185#include <sys/mount.h> 186#include <sys/namei.h> 187#include <sys/malloc.h> 188#include <sys/buf.h> 189#include <miscfs/nullfs/null.h> 190 191static int null_bug_bypass = 0; /* for debugging: enables bypass printf'ing */ 192SYSCTL_INT(_debug, OID_AUTO, nullfs_bug_bypass, CTLFLAG_RW, 193 &null_bug_bypass, 0, ""); 194 195int null_bypass __P((struct vop_generic_args *ap)); 196static int null_bwrite __P((struct vop_bwrite_args *ap)); 197static int null_getattr __P((struct vop_getattr_args *ap)); 198static int null_inactive __P((struct vop_inactive_args *ap)); 199static int null_print __P((struct vop_print_args *ap)); 200static int null_reclaim __P((struct vop_reclaim_args *ap)); 201static int null_strategy __P((struct vop_strategy_args *ap)); 202 203/* 204 * This is the 10-Apr-92 bypass routine. 205 * This version has been optimized for speed, throwing away some 206 * safety checks. It should still always work, but it's not as 207 * robust to programmer errors. 208 * Define SAFETY to include some error checking code. 209 * 210 * In general, we map all vnodes going down and unmap them on the way back. 211 * As an exception to this, vnodes can be marked "unmapped" by setting 212 * the Nth bit in operation's vdesc_flags. 213 * 214 * Also, some BSD vnode operations have the side effect of vrele'ing 215 * their arguments. With stacking, the reference counts are held 216 * by the upper node, not the lower one, so we must handle these 217 * side-effects here. This is not of concern in Sun-derived systems 218 * since there are no such side-effects. 219 * 220 * This makes the following assumptions: 221 * - only one returned vpp 222 * - no INOUT vpp's (Sun's vop_open has one of these) 223 * - the vnode operation vector of the first vnode should be used 224 * to determine what implementation of the op should be invoked 225 * - all mapped vnodes are of our vnode-type (NEEDSWORK: 226 * problems on rmdir'ing mount points and renaming?) 227 */ 228int 229null_bypass(ap) 230 struct vop_generic_args /* { 231 struct vnodeop_desc *a_desc; 232 <other random data follows, presumably> 233 } */ *ap; 234{ 235 register struct vnode **this_vp_p; 236 int error; 237 struct vnode *old_vps[VDESC_MAX_VPS]; 238 struct vnode **vps_p[VDESC_MAX_VPS]; 239 struct vnode ***vppp; 240 struct vnodeop_desc *descp = ap->a_desc; 241 int reles, i; 242 243 if (null_bug_bypass) 244 printf ("null_bypass: %s\n", descp->vdesc_name); 245 246#ifdef SAFETY 247 /* 248 * We require at least one vp. 249 */ 250 if (descp->vdesc_vp_offsets == NULL || 251 descp->vdesc_vp_offsets[0] == VDESC_NO_OFFSET) 252 panic ("null_bypass: no vp's in map."); 253#endif 254 255 /* 256 * Map the vnodes going in. 257 * Later, we'll invoke the operation based on 258 * the first mapped vnode's operation vector. 259 */ 260 reles = descp->vdesc_flags; 261 for (i = 0; i < VDESC_MAX_VPS; reles >>= 1, i++) { 262 if (descp->vdesc_vp_offsets[i] == VDESC_NO_OFFSET) 263 break; /* bail out at end of list */ 264 vps_p[i] = this_vp_p = 265 VOPARG_OFFSETTO(struct vnode**,descp->vdesc_vp_offsets[i],ap); 266 /* 267 * We're not guaranteed that any but the first vnode 268 * are of our type. Check for and don't map any 269 * that aren't. (We must always map first vp or vclean fails.) 270 */ 271 if (i && (*this_vp_p == NULL || 272 (*this_vp_p)->v_op != null_vnodeop_p)) { 273 old_vps[i] = NULL; 274 } else { 275 old_vps[i] = *this_vp_p; 276 *(vps_p[i]) = NULLVPTOLOWERVP(*this_vp_p); 277 /* 278 * XXX - Several operations have the side effect 279 * of vrele'ing their vp's. We must account for 280 * that. (This should go away in the future.) 281 */ 282 if (reles & 1) 283 VREF(*this_vp_p); 284 } 285 286 } 287 288 /* 289 * Call the operation on the lower layer 290 * with the modified argument structure. 291 */ 292 error = VCALL(*(vps_p[0]), descp->vdesc_offset, ap); 293 294 /* 295 * Maintain the illusion of call-by-value 296 * by restoring vnodes in the argument structure 297 * to their original value. 298 */ 299 reles = descp->vdesc_flags; 300 for (i = 0; i < VDESC_MAX_VPS; reles >>= 1, i++) { 301 if (descp->vdesc_vp_offsets[i] == VDESC_NO_OFFSET) 302 break; /* bail out at end of list */ 303 if (old_vps[i]) { 304 *(vps_p[i]) = old_vps[i]; 305 if (reles & 1) 306 vrele(*(vps_p[i])); 307 } 308 } 309 310 /* 311 * Map the possible out-going vpp 312 * (Assumes that the lower layer always returns 313 * a VREF'ed vpp unless it gets an error.) 314 */ 315 if (descp->vdesc_vpp_offset != VDESC_NO_OFFSET && 316 !(descp->vdesc_flags & VDESC_NOMAP_VPP) && 317 !error) { 318 /* 319 * XXX - even though some ops have vpp returned vp's, 320 * several ops actually vrele this before returning. 321 * We must avoid these ops. 322 * (This should go away when these ops are regularized.) 323 */ 324 if (descp->vdesc_flags & VDESC_VPP_WILLRELE) 325 goto out; 326 vppp = VOPARG_OFFSETTO(struct vnode***, 327 descp->vdesc_vpp_offset,ap); 328 error = null_node_create(old_vps[0]->v_mount, **vppp, *vppp); 329 } 330 331 out: 332 return (error); 333} 334 335/* 336 * We have to carry on the locking protocol on the null layer vnodes 337 * as we progress through the tree. We also have to enforce read-only 338 * if this layer is mounted read-only. 339 */ 340static int 341null_lookup(ap) 342 struct vop_lookup_args /* { 343 struct vnode * a_dvp; 344 struct vnode ** a_vpp; 345 struct componentname * a_cnp; 346 } */ *ap; 347{ 348 struct componentname *cnp = ap->a_cnp; 349 struct proc *p = cnp->cn_proc; 350 int flags = cnp->cn_flags; 351 struct vop_lock_args lockargs; 352 struct vop_unlock_args unlockargs; 353 struct vnode *dvp, *vp; 354 int error; 355 356 if ((flags & ISLASTCN) && (ap->a_dvp->v_mount->mnt_flag & MNT_RDONLY) && 357 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) 358 return (EROFS); 359 error = null_bypass(ap); 360 if (error == EJUSTRETURN && (flags & ISLASTCN) && 361 (ap->a_dvp->v_mount->mnt_flag & MNT_RDONLY) && 362 (cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME)) 363 error = EROFS; 364 /* 365 * We must do the same locking and unlocking at this layer as 366 * is done in the layers below us. We could figure this out 367 * based on the error return and the LASTCN, LOCKPARENT, and 368 * LOCKLEAF flags. However, it is more expidient to just find 369 * out the state of the lower level vnodes and set ours to the 370 * same state. 371 */ 372 dvp = ap->a_dvp; 373 vp = *ap->a_vpp; 374 if (dvp == vp) 375 return (error); 376 if (!VOP_ISLOCKED(dvp)) { 377 unlockargs.a_vp = dvp; 378 unlockargs.a_flags = 0; 379 unlockargs.a_p = p; 380 vop_nounlock(&unlockargs); 381 } 382 if (vp != NULL && VOP_ISLOCKED(vp)) { 383 lockargs.a_vp = vp; 384 lockargs.a_flags = LK_SHARED; 385 lockargs.a_p = p; 386 vop_nolock(&lockargs); 387 } 388 return (error); 389} 390 391/* 392 * Setattr call. Disallow write attempts if the layer is mounted read-only. 393 */ 394int 395null_setattr(ap) 396 struct vop_setattr_args /* { 397 struct vnodeop_desc *a_desc; 398 struct vnode *a_vp; 399 struct vattr *a_vap; 400 struct ucred *a_cred; 401 struct proc *a_p; 402 } */ *ap; 403{ 404 struct vnode *vp = ap->a_vp; 405 struct vattr *vap = ap->a_vap; 406 407 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL || 408 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.ts_sec != VNOVAL || 409 vap->va_mtime.ts_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) && 410 (vp->v_mount->mnt_flag & MNT_RDONLY)) 411 return (EROFS); 412 if (vap->va_size != VNOVAL) { 413 switch (vp->v_type) { 414 case VDIR: 415 return (EISDIR); 416 case VCHR: 417 case VBLK: 418 case VSOCK: 419 case VFIFO: 420 return (0); 421 case VREG: 422 case VLNK: 423 default: 424 /* 425 * Disallow write attempts if the filesystem is 426 * mounted read-only. 427 */ 428 if (vp->v_mount->mnt_flag & MNT_RDONLY) 429 return (EROFS); 430 } 431 } 432 return (null_bypass(ap)); 433} 434 435/* 436 * We handle getattr only to change the fsid. 437 */ 438static int 439null_getattr(ap) 440 struct vop_getattr_args /* { 441 struct vnode *a_vp; 442 struct vattr *a_vap; 443 struct ucred *a_cred; 444 struct proc *a_p; 445 } */ *ap; 446{ 447 int error; 448 449 if (error = null_bypass(ap)) 450 return (error); 451 /* Requires that arguments be restored. */ 452 ap->a_vap->va_fsid = ap->a_vp->v_mount->mnt_stat.f_fsid.val[0]; 453 return (0); 454} 455 456static int 457null_access(ap) 458 struct vop_access_args /* { 459 struct vnode *a_vp; 460 int a_mode; 461 struct ucred *a_cred; 462 struct proc *a_p; 463 } */ *ap; 464{ 465 struct vnode *vp = ap->a_vp; 466 mode_t mode = ap->a_mode; 467 468 /* 469 * Disallow write attempts on read-only layers; 470 * unless the file is a socket, fifo, or a block or 471 * character device resident on the file system. 472 */ 473 if (mode & VWRITE) { 474 switch (vp->v_type) { 475 case VDIR: 476 case VLNK: 477 case VREG: 478 if (vp->v_mount->mnt_flag & MNT_RDONLY) 479 return (EROFS); 480 break; 481 } 482 } 483 return (null_bypass(ap)); 484} 485 486/* 487 * We need to process our own vnode lock and then clear the 488 * interlock flag as it applies only to our vnode, not the 489 * vnodes below us on the stack. 490 */ 491int 492null_lock(ap) 493 struct vop_lock_args /* { 494 struct vnode *a_vp; 495 int a_flags; 496 struct proc *a_p; 497 } */ *ap; 498{ 499 500 vop_nolock(ap); 501 if ((ap->a_flags & LK_TYPE_MASK) == LK_DRAIN) 502 return (0); 503 ap->a_flags &= ~LK_INTERLOCK; 504 return (null_bypass(ap)); 505} 506 507/* 508 * We need to process our own vnode unlock and then clear the 509 * interlock flag as it applies only to our vnode, not the 510 * vnodes below us on the stack. 511 */ 512int 513null_unlock(ap) 514 struct vop_unlock_args /* { 515 struct vnode *a_vp; 516 int a_flags; 517 struct proc *a_p; 518 } */ *ap; 519{ 520 struct vnode *vp = ap->a_vp; 521 522 vop_nounlock(ap); 523 ap->a_flags &= ~LK_INTERLOCK; 524 return (null_bypass(ap)); 525} 526 527int 528null_inactive(ap) 529 struct vop_inactive_args /* { 530 struct vnode *a_vp; 531 struct proc *a_p; 532 } */ *ap; 533{ 534 /* 535 * Do nothing (and _don't_ bypass). 536 * Wait to vrele lowervp until reclaim, 537 * so that until then our null_node is in the 538 * cache and reusable. 539 * 540 * NEEDSWORK: Someday, consider inactive'ing 541 * the lowervp and then trying to reactivate it 542 * with capabilities (v_id) 543 * like they do in the name lookup cache code. 544 * That's too much work for now. 545 */ 546 VOP_UNLOCK(ap->a_vp, 0, ap->a_p); 547 return (0); 548} 549 550static int 551null_reclaim(ap) 552 struct vop_reclaim_args /* { 553 struct vnode *a_vp; 554 struct proc *a_p; 555 } */ *ap; 556{ 557 struct vnode *vp = ap->a_vp; 558 struct null_node *xp = VTONULL(vp); 559 struct vnode *lowervp = xp->null_lowervp; 560 561 /* 562 * Note: in vop_reclaim, vp->v_op == dead_vnodeop_p, 563 * so we can't call VOPs on ourself. 564 */ 565 /* After this assignment, this node will not be re-used. */ 566 xp->null_lowervp = NULL; 567 LIST_REMOVE(xp, null_hash); 568 FREE(vp->v_data, M_TEMP); 569 vp->v_data = NULL; 570 vrele (lowervp); 571 return (0); 572} 573 574static int 575null_print(ap) 576 struct vop_print_args /* { 577 struct vnode *a_vp; 578 } */ *ap; 579{ 580 register struct vnode *vp = ap->a_vp; 581 printf ("\ttag VT_NULLFS, vp=%p, lowervp=%p\n", vp, NULLVPTOLOWERVP(vp)); 582 return (0); 583} 584 585/* 586 * XXX - vop_strategy must be hand coded because it has no 587 * vnode in its arguments. 588 * This goes away with a merged VM/buffer cache. 589 */ 590static int 591null_strategy(ap) 592 struct vop_strategy_args /* { 593 struct buf *a_bp; 594 } */ *ap; 595{ 596 struct buf *bp = ap->a_bp; 597 int error; 598 struct vnode *savedvp; 599 600 savedvp = bp->b_vp; 601 bp->b_vp = NULLVPTOLOWERVP(bp->b_vp); 602 603 error = VOP_STRATEGY(bp); 604 605 bp->b_vp = savedvp; 606 607 return (error); 608} 609 610/* 611 * XXX - like vop_strategy, vop_bwrite must be hand coded because it has no 612 * vnode in its arguments. 613 * This goes away with a merged VM/buffer cache. 614 */ 615static int 616null_bwrite(ap) 617 struct vop_bwrite_args /* { 618 struct buf *a_bp; 619 } */ *ap; 620{ 621 struct buf *bp = ap->a_bp; 622 int error; 623 struct vnode *savedvp; 624 625 savedvp = bp->b_vp; 626 bp->b_vp = NULLVPTOLOWERVP(bp->b_vp); 627 628 error = VOP_BWRITE(bp); 629 630 bp->b_vp = savedvp; 631 632 return (error); 633} 634 635/* 636 * Global vfs data structures 637 */ 638vop_t **null_vnodeop_p; 639static struct vnodeopv_entry_desc null_vnodeop_entries[] = { 640 { &vop_default_desc, (vop_t *)null_bypass }, 641 642 { &vop_lookup_desc, (vop_t *)null_lookup }, 643 { &vop_setattr_desc, (vop_t *)null_setattr }, 644 { &vop_getattr_desc, (vop_t *)null_getattr }, 645 { &vop_access_desc, (vop_t *)null_access }, 646 { &vop_lock_desc, (vop_t *)null_lock }, 647 { &vop_unlock_desc, (vop_t *)null_unlock }, 648 { &vop_inactive_desc, (vop_t *)null_inactive }, 649 { &vop_reclaim_desc, (vop_t *)null_reclaim }, 650 { &vop_print_desc, (vop_t *)null_print }, 651 652 { &vop_strategy_desc, (vop_t *)null_strategy }, 653 { &vop_bwrite_desc, (vop_t *)null_bwrite }, 654 655 { NULL, NULL } 656}; 657static struct vnodeopv_desc null_vnodeop_opv_desc = 658 { &null_vnodeop_p, null_vnodeop_entries }; 659 660VNODEOP_SET(null_vnodeop_opv_desc); 661