null_vnops.c revision 72200
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 * ...and... 41 * @(#)null_vnodeops.c 1.20 92/07/07 UCLA Ficus project 42 * 43 * $FreeBSD: head/sys/fs/nullfs/null_vnops.c 72200 2001-02-09 06:11:45Z bmilekic $ 44 */ 45 46/* 47 * Null Layer 48 * 49 * (See mount_null(8) for more information.) 50 * 51 * The null layer duplicates a portion of the file system 52 * name space under a new name. In this respect, it is 53 * similar to the loopback file system. It differs from 54 * the loopback fs in two respects: it is implemented using 55 * a stackable layers techniques, and its "null-node"s stack above 56 * all lower-layer vnodes, not just over directory vnodes. 57 * 58 * The null layer has two purposes. First, it serves as a demonstration 59 * of layering by proving a layer which does nothing. (It actually 60 * does everything the loopback file system does, which is slightly 61 * more than nothing.) Second, the null layer can serve as a prototype 62 * layer. Since it provides all necessary layer framework, 63 * new file system layers can be created very easily be starting 64 * with a null layer. 65 * 66 * The remainder of this man page examines the null layer as a basis 67 * for constructing new layers. 68 * 69 * 70 * INSTANTIATING NEW NULL LAYERS 71 * 72 * New null layers are created with mount_null(8). 73 * Mount_null(8) takes two arguments, the pathname 74 * of the lower vfs (target-pn) and the pathname where the null 75 * layer will appear in the namespace (alias-pn). After 76 * the null layer is put into place, the contents 77 * of target-pn subtree will be aliased under alias-pn. 78 * 79 * 80 * OPERATION OF A NULL LAYER 81 * 82 * The null layer is the minimum file system layer, 83 * simply bypassing all possible operations to the lower layer 84 * for processing there. The majority of its activity centers 85 * on the bypass routine, through which nearly all vnode operations 86 * pass. 87 * 88 * The bypass routine accepts arbitrary vnode operations for 89 * handling by the lower layer. It begins by examing vnode 90 * operation arguments and replacing any null-nodes by their 91 * lower-layer equivlants. It then invokes the operation 92 * on the lower layer. Finally, it replaces the null-nodes 93 * in the arguments and, if a vnode is return by the operation, 94 * stacks a null-node on top of the returned vnode. 95 * 96 * Although bypass handles most operations, vop_getattr, vop_lock, 97 * vop_unlock, vop_inactive, vop_reclaim, and vop_print are not 98 * bypassed. Vop_getattr must change the fsid being returned. 99 * Vop_lock and vop_unlock must handle any locking for the 100 * current vnode as well as pass the lock request down. 101 * Vop_inactive and vop_reclaim are not bypassed so that 102 * they can handle freeing null-layer specific data. Vop_print 103 * is not bypassed to avoid excessive debugging information. 104 * Also, certain vnode operations change the locking state within 105 * the operation (create, mknod, remove, link, rename, mkdir, rmdir, 106 * and symlink). Ideally these operations should not change the 107 * lock state, but should be changed to let the caller of the 108 * function unlock them. Otherwise all intermediate vnode layers 109 * (such as union, umapfs, etc) must catch these functions to do 110 * the necessary locking at their layer. 111 * 112 * 113 * INSTANTIATING VNODE STACKS 114 * 115 * Mounting associates the null layer with a lower layer, 116 * effect stacking two VFSes. Vnode stacks are instead 117 * created on demand as files are accessed. 118 * 119 * The initial mount creates a single vnode stack for the 120 * root of the new null layer. All other vnode stacks 121 * are created as a result of vnode operations on 122 * this or other null vnode stacks. 123 * 124 * New vnode stacks come into existance as a result of 125 * an operation which returns a vnode. 126 * The bypass routine stacks a null-node above the new 127 * vnode before returning it to the caller. 128 * 129 * For example, imagine mounting a null layer with 130 * "mount_null /usr/include /dev/layer/null". 131 * Changing directory to /dev/layer/null will assign 132 * the root null-node (which was created when the null layer was mounted). 133 * Now consider opening "sys". A vop_lookup would be 134 * done on the root null-node. This operation would bypass through 135 * to the lower layer which would return a vnode representing 136 * the UFS "sys". Null_bypass then builds a null-node 137 * aliasing the UFS "sys" and returns this to the caller. 138 * Later operations on the null-node "sys" will repeat this 139 * process when constructing other vnode stacks. 140 * 141 * 142 * CREATING OTHER FILE SYSTEM LAYERS 143 * 144 * One of the easiest ways to construct new file system layers is to make 145 * a copy of the null layer, rename all files and variables, and 146 * then begin modifing the copy. Sed can be used to easily rename 147 * all variables. 148 * 149 * The umap layer is an example of a layer descended from the 150 * null layer. 151 * 152 * 153 * INVOKING OPERATIONS ON LOWER LAYERS 154 * 155 * There are two techniques to invoke operations on a lower layer 156 * when the operation cannot be completely bypassed. Each method 157 * is appropriate in different situations. In both cases, 158 * it is the responsibility of the aliasing layer to make 159 * the operation arguments "correct" for the lower layer 160 * by mapping an vnode arguments to the lower layer. 161 * 162 * The first approach is to call the aliasing layer's bypass routine. 163 * This method is most suitable when you wish to invoke the operation 164 * currently being handled on the lower layer. It has the advantage 165 * that the bypass routine already must do argument mapping. 166 * An example of this is null_getattrs in the null layer. 167 * 168 * A second approach is to directly invoke vnode operations on 169 * the lower layer with the VOP_OPERATIONNAME interface. 170 * The advantage of this method is that it is easy to invoke 171 * arbitrary operations on the lower layer. The disadvantage 172 * is that vnode arguments must be manualy mapped. 173 * 174 */ 175 176#include <sys/param.h> 177#include <sys/systm.h> 178#include <sys/kernel.h> 179#include <sys/conf.h> 180#include <sys/sysctl.h> 181#include <sys/vnode.h> 182#include <sys/mount.h> 183#include <sys/namei.h> 184#include <sys/malloc.h> 185#include <miscfs/nullfs/null.h> 186 187#include <vm/vm.h> 188#include <vm/vm_extern.h> 189#include <vm/vm_object.h> 190#include <vm/vnode_pager.h> 191 192static int null_bug_bypass = 0; /* for debugging: enables bypass printf'ing */ 193SYSCTL_INT(_debug, OID_AUTO, nullfs_bug_bypass, CTLFLAG_RW, 194 &null_bug_bypass, 0, ""); 195 196static int null_access(struct vop_access_args *ap); 197static int null_createvobject(struct vop_createvobject_args *ap); 198static int null_destroyvobject(struct vop_destroyvobject_args *ap); 199static int null_getattr(struct vop_getattr_args *ap); 200static int null_getvobject(struct vop_getvobject_args *ap); 201static int null_inactive(struct vop_inactive_args *ap); 202static int null_islocked(struct vop_islocked_args *ap); 203static int null_lock(struct vop_lock_args *ap); 204static int null_lookup(struct vop_lookup_args *ap); 205static int null_open(struct vop_open_args *ap); 206static int null_print(struct vop_print_args *ap); 207static int null_reclaim(struct vop_reclaim_args *ap); 208static int null_rename(struct vop_rename_args *ap); 209static int null_setattr(struct vop_setattr_args *ap); 210static int null_unlock(struct vop_unlock_args *ap); 211 212/* 213 * This is the 10-Apr-92 bypass routine. 214 * This version has been optimized for speed, throwing away some 215 * safety checks. It should still always work, but it's not as 216 * robust to programmer errors. 217 * 218 * In general, we map all vnodes going down and unmap them on the way back. 219 * As an exception to this, vnodes can be marked "unmapped" by setting 220 * the Nth bit in operation's vdesc_flags. 221 * 222 * Also, some BSD vnode operations have the side effect of vrele'ing 223 * their arguments. With stacking, the reference counts are held 224 * by the upper node, not the lower one, so we must handle these 225 * side-effects here. This is not of concern in Sun-derived systems 226 * since there are no such side-effects. 227 * 228 * This makes the following assumptions: 229 * - only one returned vpp 230 * - no INOUT vpp's (Sun's vop_open has one of these) 231 * - the vnode operation vector of the first vnode should be used 232 * to determine what implementation of the op should be invoked 233 * - all mapped vnodes are of our vnode-type (NEEDSWORK: 234 * problems on rmdir'ing mount points and renaming?) 235 */ 236int 237null_bypass(ap) 238 struct vop_generic_args /* { 239 struct vnodeop_desc *a_desc; 240 <other random data follows, presumably> 241 } */ *ap; 242{ 243 register struct vnode **this_vp_p; 244 int error; 245 struct vnode *old_vps[VDESC_MAX_VPS]; 246 struct vnode **vps_p[VDESC_MAX_VPS]; 247 struct vnode ***vppp; 248 struct vnodeop_desc *descp = ap->a_desc; 249 int reles, i; 250 251 if (null_bug_bypass) 252 printf ("null_bypass: %s\n", descp->vdesc_name); 253 254#ifdef DIAGNOSTIC 255 /* 256 * We require at least one vp. 257 */ 258 if (descp->vdesc_vp_offsets == NULL || 259 descp->vdesc_vp_offsets[0] == VDESC_NO_OFFSET) 260 panic ("null_bypass: no vp's in map"); 261#endif 262 263 /* 264 * Map the vnodes going in. 265 * Later, we'll invoke the operation based on 266 * the first mapped vnode's operation vector. 267 */ 268 reles = descp->vdesc_flags; 269 for (i = 0; i < VDESC_MAX_VPS; reles >>= 1, i++) { 270 if (descp->vdesc_vp_offsets[i] == VDESC_NO_OFFSET) 271 break; /* bail out at end of list */ 272 vps_p[i] = this_vp_p = 273 VOPARG_OFFSETTO(struct vnode**,descp->vdesc_vp_offsets[i],ap); 274 /* 275 * We're not guaranteed that any but the first vnode 276 * are of our type. Check for and don't map any 277 * that aren't. (We must always map first vp or vclean fails.) 278 */ 279 if (i && (*this_vp_p == NULLVP || 280 (*this_vp_p)->v_op != null_vnodeop_p)) { 281 old_vps[i] = NULLVP; 282 } else { 283 old_vps[i] = *this_vp_p; 284 *(vps_p[i]) = NULLVPTOLOWERVP(*this_vp_p); 285 /* 286 * XXX - Several operations have the side effect 287 * of vrele'ing their vp's. We must account for 288 * that. (This should go away in the future.) 289 */ 290 if (reles & VDESC_VP0_WILLRELE) 291 VREF(*this_vp_p); 292 } 293 294 } 295 296 /* 297 * Call the operation on the lower layer 298 * with the modified argument structure. 299 */ 300 if (vps_p[0] && *vps_p[0]) 301 error = VCALL(*(vps_p[0]), descp->vdesc_offset, ap); 302 else { 303 printf("null_bypass: no map for %s\n", descp->vdesc_name); 304 error = EINVAL; 305 } 306 307 /* 308 * Maintain the illusion of call-by-value 309 * by restoring vnodes in the argument structure 310 * to their original value. 311 */ 312 reles = descp->vdesc_flags; 313 for (i = 0; i < VDESC_MAX_VPS; reles >>= 1, i++) { 314 if (descp->vdesc_vp_offsets[i] == VDESC_NO_OFFSET) 315 break; /* bail out at end of list */ 316 if (old_vps[i]) { 317 *(vps_p[i]) = old_vps[i]; 318#if 0 319 if (reles & VDESC_VP0_WILLUNLOCK) 320 VOP_UNLOCK(*(vps_p[i]), LK_THISLAYER, curproc); 321#endif 322 if (reles & VDESC_VP0_WILLRELE) 323 vrele(*(vps_p[i])); 324 } 325 } 326 327 /* 328 * Map the possible out-going vpp 329 * (Assumes that the lower layer always returns 330 * a VREF'ed vpp unless it gets an error.) 331 */ 332 if (descp->vdesc_vpp_offset != VDESC_NO_OFFSET && 333 !(descp->vdesc_flags & VDESC_NOMAP_VPP) && 334 !error) { 335 /* 336 * XXX - even though some ops have vpp returned vp's, 337 * several ops actually vrele this before returning. 338 * We must avoid these ops. 339 * (This should go away when these ops are regularized.) 340 */ 341 if (descp->vdesc_flags & VDESC_VPP_WILLRELE) 342 goto out; 343 vppp = VOPARG_OFFSETTO(struct vnode***, 344 descp->vdesc_vpp_offset,ap); 345 if (*vppp) 346 error = null_node_create(old_vps[0]->v_mount, **vppp, *vppp); 347 } 348 349 out: 350 return (error); 351} 352 353/* 354 * We have to carry on the locking protocol on the null layer vnodes 355 * as we progress through the tree. We also have to enforce read-only 356 * if this layer is mounted read-only. 357 */ 358static int 359null_lookup(ap) 360 struct vop_lookup_args /* { 361 struct vnode * a_dvp; 362 struct vnode ** a_vpp; 363 struct componentname * a_cnp; 364 } */ *ap; 365{ 366 struct componentname *cnp = ap->a_cnp; 367 struct vnode *dvp = ap->a_dvp; 368 struct proc *p = cnp->cn_proc; 369 int flags = cnp->cn_flags; 370 struct vnode *vp, *ldvp, *lvp; 371 int error; 372 373 if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) && 374 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) 375 return (EROFS); 376 /* 377 * Although it is possible to call null_bypass(), we'll do 378 * a direct call to reduce overhead 379 */ 380 ldvp = NULLVPTOLOWERVP(dvp); 381 vp = lvp = NULL; 382 error = VOP_LOOKUP(ldvp, &lvp, cnp); 383 if (error == EJUSTRETURN && (flags & ISLASTCN) && 384 (dvp->v_mount->mnt_flag & MNT_RDONLY) && 385 (cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME)) 386 error = EROFS; 387 388 /* 389 * Rely only on the PDIRUNLOCK flag which should be carefully 390 * tracked by underlying filesystem. 391 */ 392 if (cnp->cn_flags & PDIRUNLOCK) 393 VOP_UNLOCK(dvp, LK_THISLAYER, p); 394 if ((error == 0 || error == EJUSTRETURN) && lvp != NULL) { 395 if (ldvp == lvp) { 396 *ap->a_vpp = dvp; 397 VREF(dvp); 398 vrele(lvp); 399 } else { 400 error = null_node_create(dvp->v_mount, lvp, &vp); 401 if (error == 0) 402 *ap->a_vpp = vp; 403 } 404 } 405 return (error); 406} 407 408/* 409 * Setattr call. Disallow write attempts if the layer is mounted read-only. 410 */ 411int 412null_setattr(ap) 413 struct vop_setattr_args /* { 414 struct vnodeop_desc *a_desc; 415 struct vnode *a_vp; 416 struct vattr *a_vap; 417 struct ucred *a_cred; 418 struct proc *a_p; 419 } */ *ap; 420{ 421 struct vnode *vp = ap->a_vp; 422 struct vattr *vap = ap->a_vap; 423 424 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL || 425 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL || 426 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) && 427 (vp->v_mount->mnt_flag & MNT_RDONLY)) 428 return (EROFS); 429 if (vap->va_size != VNOVAL) { 430 switch (vp->v_type) { 431 case VDIR: 432 return (EISDIR); 433 case VCHR: 434 case VBLK: 435 case VSOCK: 436 case VFIFO: 437 if (vap->va_flags != VNOVAL) 438 return (EOPNOTSUPP); 439 return (0); 440 case VREG: 441 case VLNK: 442 default: 443 /* 444 * Disallow write attempts if the filesystem is 445 * mounted read-only. 446 */ 447 if (vp->v_mount->mnt_flag & MNT_RDONLY) 448 return (EROFS); 449 } 450 } 451 452 return (null_bypass((struct vop_generic_args *)ap)); 453} 454 455/* 456 * We handle getattr only to change the fsid. 457 */ 458static int 459null_getattr(ap) 460 struct vop_getattr_args /* { 461 struct vnode *a_vp; 462 struct vattr *a_vap; 463 struct ucred *a_cred; 464 struct proc *a_p; 465 } */ *ap; 466{ 467 int error; 468 469 if ((error = null_bypass((struct vop_generic_args *)ap)) != 0) 470 return (error); 471 472 ap->a_vap->va_fsid = ap->a_vp->v_mount->mnt_stat.f_fsid.val[0]; 473 return (0); 474} 475 476/* 477 * Handle to disallow write access if mounted read-only. 478 */ 479static int 480null_access(ap) 481 struct vop_access_args /* { 482 struct vnode *a_vp; 483 int a_mode; 484 struct ucred *a_cred; 485 struct proc *a_p; 486 } */ *ap; 487{ 488 struct vnode *vp = ap->a_vp; 489 mode_t mode = ap->a_mode; 490 491 /* 492 * Disallow write attempts on read-only layers; 493 * unless the file is a socket, fifo, or a block or 494 * character device resident on the file system. 495 */ 496 if (mode & VWRITE) { 497 switch (vp->v_type) { 498 case VDIR: 499 case VLNK: 500 case VREG: 501 if (vp->v_mount->mnt_flag & MNT_RDONLY) 502 return (EROFS); 503 break; 504 default: 505 break; 506 } 507 } 508 return (null_bypass((struct vop_generic_args *)ap)); 509} 510 511/* 512 * We must handle open to be able to catch MNT_NODEV and friends. 513 */ 514static int 515null_open(ap) 516 struct vop_open_args /* { 517 struct vnode *a_vp; 518 int a_mode; 519 struct ucred *a_cred; 520 struct proc *a_p; 521 } */ *ap; 522{ 523 struct vnode *vp = ap->a_vp; 524 struct vnode *lvp = NULLVPTOLOWERVP(ap->a_vp); 525 526 if ((vp->v_mount->mnt_flag & MNT_NODEV) && 527 (lvp->v_type == VBLK || lvp->v_type == VCHR)) 528 return ENXIO; 529 530 return (null_bypass((struct vop_generic_args *)ap)); 531} 532 533/* 534 * We handle this to eliminate null FS to lower FS 535 * file moving. Don't know why we don't allow this, 536 * possibly we should. 537 */ 538static int 539null_rename(ap) 540 struct vop_rename_args /* { 541 struct vnode *a_fdvp; 542 struct vnode *a_fvp; 543 struct componentname *a_fcnp; 544 struct vnode *a_tdvp; 545 struct vnode *a_tvp; 546 struct componentname *a_tcnp; 547 } */ *ap; 548{ 549 struct vnode *tdvp = ap->a_tdvp; 550 struct vnode *fvp = ap->a_fvp; 551 struct vnode *fdvp = ap->a_fdvp; 552 struct vnode *tvp = ap->a_tvp; 553 554 /* Check for cross-device rename. */ 555 if ((fvp->v_mount != tdvp->v_mount) || 556 (tvp && (fvp->v_mount != tvp->v_mount))) { 557 if (tdvp == tvp) 558 vrele(tdvp); 559 else 560 vput(tdvp); 561 if (tvp) 562 vput(tvp); 563 vrele(fdvp); 564 vrele(fvp); 565 return (EXDEV); 566 } 567 568 return (null_bypass((struct vop_generic_args *)ap)); 569} 570 571/* 572 * We need to process our own vnode lock and then clear the 573 * interlock flag as it applies only to our vnode, not the 574 * vnodes below us on the stack. 575 */ 576static int 577null_lock(ap) 578 struct vop_lock_args /* { 579 struct vnode *a_vp; 580 int a_flags; 581 struct proc *a_p; 582 } */ *ap; 583{ 584 struct vnode *vp = ap->a_vp; 585 int flags = ap->a_flags; 586 struct proc *p = ap->a_p; 587 struct vnode *lvp; 588 int error; 589 590 if (flags & LK_THISLAYER) { 591 if (vp->v_vnlock != NULL) 592 return 0; /* lock is shared across layers */ 593 error = lockmgr(&vp->v_lock, flags & ~LK_THISLAYER, 594 &vp->v_interlock, p); 595 return (error); 596 } 597 598 if (vp->v_vnlock != NULL) { 599 /* 600 * The lower level has exported a struct lock to us. Use 601 * it so that all vnodes in the stack lock and unlock 602 * simultaneously. Note: we don't DRAIN the lock as DRAIN 603 * decommissions the lock - just because our vnode is 604 * going away doesn't mean the struct lock below us is. 605 * LK_EXCLUSIVE is fine. 606 */ 607 if ((flags & LK_TYPE_MASK) == LK_DRAIN) { 608 NULLFSDEBUG("null_lock: avoiding LK_DRAIN\n"); 609 return(lockmgr(vp->v_vnlock, 610 (flags & ~LK_TYPE_MASK) | LK_EXCLUSIVE, 611 &vp->v_interlock, p)); 612 } 613 return(lockmgr(vp->v_vnlock, flags, &vp->v_interlock, p)); 614 } else { 615 /* 616 * To prevent race conditions involving doing a lookup 617 * on "..", we have to lock the lower node, then lock our 618 * node. Most of the time it won't matter that we lock our 619 * node (as any locking would need the lower one locked 620 * first). But we can LK_DRAIN the upper lock as a step 621 * towards decomissioning it. 622 */ 623 lvp = NULLVPTOLOWERVP(vp); 624 if (lvp == NULL) 625 return (lockmgr(&vp->v_lock, flags, &vp->v_interlock, p)); 626 if (flags & LK_INTERLOCK) { 627 mtx_unlock(&vp->v_interlock); 628 flags &= ~LK_INTERLOCK; 629 } 630 if ((flags & LK_TYPE_MASK) == LK_DRAIN) { 631 error = VOP_LOCK(lvp, 632 (flags & ~LK_TYPE_MASK) | LK_EXCLUSIVE, p); 633 } else 634 error = VOP_LOCK(lvp, flags, p); 635 if (error) 636 return (error); 637 error = lockmgr(&vp->v_lock, flags, &vp->v_interlock, p); 638 if (error) 639 VOP_UNLOCK(lvp, 0, p); 640 return (error); 641 } 642} 643 644/* 645 * We need to process our own vnode unlock and then clear the 646 * interlock flag as it applies only to our vnode, not the 647 * vnodes below us on the stack. 648 */ 649static int 650null_unlock(ap) 651 struct vop_unlock_args /* { 652 struct vnode *a_vp; 653 int a_flags; 654 struct proc *a_p; 655 } */ *ap; 656{ 657 struct vnode *vp = ap->a_vp; 658 int flags = ap->a_flags; 659 struct proc *p = ap->a_p; 660 struct vnode *lvp; 661 662 if (vp->v_vnlock != NULL) { 663 if (flags & LK_THISLAYER) 664 return 0; /* the lock is shared across layers */ 665 flags &= ~LK_THISLAYER; 666 return (lockmgr(vp->v_vnlock, flags | LK_RELEASE, 667 &vp->v_interlock, p)); 668 } 669 lvp = NULLVPTOLOWERVP(vp); 670 if (lvp == NULL) 671 return (lockmgr(&vp->v_lock, flags | LK_RELEASE, &vp->v_interlock, p)); 672 if ((flags & LK_THISLAYER) == 0) { 673 if (flags & LK_INTERLOCK) { 674 mtx_unlock(&vp->v_interlock); 675 flags &= ~LK_INTERLOCK; 676 } 677 VOP_UNLOCK(lvp, flags & ~LK_INTERLOCK, p); 678 } else 679 flags &= ~LK_THISLAYER; 680 return (lockmgr(&vp->v_lock, flags | LK_RELEASE, &vp->v_interlock, p)); 681} 682 683static int 684null_islocked(ap) 685 struct vop_islocked_args /* { 686 struct vnode *a_vp; 687 struct proc *a_p; 688 } */ *ap; 689{ 690 struct vnode *vp = ap->a_vp; 691 struct proc *p = ap->a_p; 692 693 if (vp->v_vnlock != NULL) 694 return (lockstatus(vp->v_vnlock, p)); 695 return (lockstatus(&vp->v_lock, p)); 696} 697 698/* 699 * There is no way to tell that someone issued remove/rmdir operation 700 * on the underlying filesystem. For now we just have to release lowevrp 701 * as soon as possible. 702 */ 703static int 704null_inactive(ap) 705 struct vop_inactive_args /* { 706 struct vnode *a_vp; 707 struct proc *a_p; 708 } */ *ap; 709{ 710 struct vnode *vp = ap->a_vp; 711 struct proc *p = ap->a_p; 712 struct null_node *xp = VTONULL(vp); 713 struct vnode *lowervp = xp->null_lowervp; 714 715 lockmgr(&null_hashlock, LK_EXCLUSIVE, NULL, p); 716 LIST_REMOVE(xp, null_hash); 717 lockmgr(&null_hashlock, LK_RELEASE, NULL, p); 718 719 xp->null_lowervp = NULLVP; 720 if (vp->v_vnlock != NULL) { 721 vp->v_vnlock = &vp->v_lock; /* we no longer share the lock */ 722 } else 723 VOP_UNLOCK(vp, LK_THISLAYER, p); 724 725 vput(lowervp); 726 /* 727 * Now it is safe to drop references to the lower vnode. 728 * VOP_INACTIVE() will be called by vrele() if necessary. 729 */ 730 vrele (lowervp); 731 732 return (0); 733} 734 735/* 736 * We can free memory in null_inactive, but we do this 737 * here. (Possible to guard vp->v_data to point somewhere) 738 */ 739static int 740null_reclaim(ap) 741 struct vop_reclaim_args /* { 742 struct vnode *a_vp; 743 struct proc *a_p; 744 } */ *ap; 745{ 746 struct vnode *vp = ap->a_vp; 747 void *vdata = vp->v_data; 748 749 vp->v_data = NULL; 750 FREE(vdata, M_NULLFSNODE); 751 752 return (0); 753} 754 755static int 756null_print(ap) 757 struct vop_print_args /* { 758 struct vnode *a_vp; 759 } */ *ap; 760{ 761 register struct vnode *vp = ap->a_vp; 762 printf ("\ttag VT_NULLFS, vp=%p, lowervp=%p\n", vp, NULLVPTOLOWERVP(vp)); 763 return (0); 764} 765 766/* 767 * Let an underlying filesystem do the work 768 */ 769static int 770null_createvobject(ap) 771 struct vop_createvobject_args /* { 772 struct vnode *vp; 773 struct ucred *cred; 774 struct proc *p; 775 } */ *ap; 776{ 777 struct vnode *vp = ap->a_vp; 778 struct vnode *lowervp = VTONULL(vp) ? NULLVPTOLOWERVP(vp) : NULL; 779 int error; 780 781 if (vp->v_type == VNON || lowervp == NULL) 782 return 0; 783 error = VOP_CREATEVOBJECT(lowervp, ap->a_cred, ap->a_p); 784 if (error) 785 return (error); 786 vp->v_flag |= VOBJBUF; 787 return (0); 788} 789 790/* 791 * We have nothing to destroy and this operation shouldn't be bypassed. 792 */ 793static int 794null_destroyvobject(ap) 795 struct vop_destroyvobject_args /* { 796 struct vnode *vp; 797 } */ *ap; 798{ 799 struct vnode *vp = ap->a_vp; 800 801 vp->v_flag &= ~VOBJBUF; 802 return (0); 803} 804 805static int 806null_getvobject(ap) 807 struct vop_getvobject_args /* { 808 struct vnode *vp; 809 struct vm_object **objpp; 810 } */ *ap; 811{ 812 struct vnode *lvp = NULLVPTOLOWERVP(ap->a_vp); 813 814 if (lvp == NULL) 815 return EINVAL; 816 return (VOP_GETVOBJECT(lvp, ap->a_objpp)); 817} 818 819/* 820 * Global vfs data structures 821 */ 822vop_t **null_vnodeop_p; 823static struct vnodeopv_entry_desc null_vnodeop_entries[] = { 824 { &vop_default_desc, (vop_t *) null_bypass }, 825 826 { &vop_access_desc, (vop_t *) null_access }, 827 { &vop_bmap_desc, (vop_t *) vop_eopnotsupp }, 828 { &vop_createvobject_desc, (vop_t *) null_createvobject }, 829 { &vop_destroyvobject_desc, (vop_t *) null_destroyvobject }, 830 { &vop_getattr_desc, (vop_t *) null_getattr }, 831 { &vop_getvobject_desc, (vop_t *) null_getvobject }, 832 { &vop_getwritemount_desc, (vop_t *) vop_stdgetwritemount}, 833 { &vop_inactive_desc, (vop_t *) null_inactive }, 834 { &vop_islocked_desc, (vop_t *) null_islocked }, 835 { &vop_lock_desc, (vop_t *) null_lock }, 836 { &vop_lookup_desc, (vop_t *) null_lookup }, 837 { &vop_open_desc, (vop_t *) null_open }, 838 { &vop_print_desc, (vop_t *) null_print }, 839 { &vop_reclaim_desc, (vop_t *) null_reclaim }, 840 { &vop_rename_desc, (vop_t *) null_rename }, 841 { &vop_setattr_desc, (vop_t *) null_setattr }, 842 { &vop_strategy_desc, (vop_t *) vop_eopnotsupp }, 843 { &vop_unlock_desc, (vop_t *) null_unlock }, 844 { NULL, NULL } 845}; 846static struct vnodeopv_desc null_vnodeop_opv_desc = 847 { &null_vnodeop_p, null_vnodeop_entries }; 848 849VNODEOP_SET(null_vnodeop_opv_desc); 850