null_vnops.c revision 67145
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 67145 2000-10-15 06:25:42Z bp $ 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/proc.h> 181#include <sys/sysctl.h> 182#include <sys/vnode.h> 183#include <sys/mount.h> 184#include <sys/namei.h> 185#include <sys/malloc.h> 186#include <miscfs/nullfs/null.h> 187 188#include <vm/vm.h> 189#include <vm/vm_extern.h> 190#include <vm/vm_object.h> 191#include <vm/vnode_pager.h> 192 193static int null_bug_bypass = 0; /* for debugging: enables bypass printf'ing */ 194SYSCTL_INT(_debug, OID_AUTO, nullfs_bug_bypass, CTLFLAG_RW, 195 &null_bug_bypass, 0, ""); 196 197static int null_access(struct vop_access_args *ap); 198static int null_createvobject(struct vop_createvobject_args *ap); 199static int null_destroyvobject(struct vop_destroyvobject_args *ap); 200static int null_getattr(struct vop_getattr_args *ap); 201static int null_getvobject(struct vop_getvobject_args *ap); 202static int null_inactive(struct vop_inactive_args *ap); 203static int null_islocked(struct vop_islocked_args *ap); 204static int null_lock(struct vop_lock_args *ap); 205static int null_lookup(struct vop_lookup_args *ap); 206static int null_open(struct vop_open_args *ap); 207static int null_print(struct vop_print_args *ap); 208static int null_reclaim(struct vop_reclaim_args *ap); 209static int null_rename(struct vop_rename_args *ap); 210static int null_setattr(struct vop_setattr_args *ap); 211static int null_unlock(struct vop_unlock_args *ap); 212 213/* 214 * This is the 10-Apr-92 bypass routine. 215 * This version has been optimized for speed, throwing away some 216 * safety checks. It should still always work, but it's not as 217 * robust to programmer errors. 218 * 219 * In general, we map all vnodes going down and unmap them on the way back. 220 * As an exception to this, vnodes can be marked "unmapped" by setting 221 * the Nth bit in operation's vdesc_flags. 222 * 223 * Also, some BSD vnode operations have the side effect of vrele'ing 224 * their arguments. With stacking, the reference counts are held 225 * by the upper node, not the lower one, so we must handle these 226 * side-effects here. This is not of concern in Sun-derived systems 227 * since there are no such side-effects. 228 * 229 * This makes the following assumptions: 230 * - only one returned vpp 231 * - no INOUT vpp's (Sun's vop_open has one of these) 232 * - the vnode operation vector of the first vnode should be used 233 * to determine what implementation of the op should be invoked 234 * - all mapped vnodes are of our vnode-type (NEEDSWORK: 235 * problems on rmdir'ing mount points and renaming?) 236 */ 237int 238null_bypass(ap) 239 struct vop_generic_args /* { 240 struct vnodeop_desc *a_desc; 241 <other random data follows, presumably> 242 } */ *ap; 243{ 244 register struct vnode **this_vp_p; 245 int error; 246 struct vnode *old_vps[VDESC_MAX_VPS]; 247 struct vnode **vps_p[VDESC_MAX_VPS]; 248 struct vnode ***vppp; 249 struct vnodeop_desc *descp = ap->a_desc; 250 int reles, i; 251 252 if (null_bug_bypass) 253 printf ("null_bypass: %s\n", descp->vdesc_name); 254 255#ifdef DIAGNOSTIC 256 /* 257 * We require at least one vp. 258 */ 259 if (descp->vdesc_vp_offsets == NULL || 260 descp->vdesc_vp_offsets[0] == VDESC_NO_OFFSET) 261 panic ("null_bypass: no vp's in map"); 262#endif 263 264 /* 265 * Map the vnodes going in. 266 * Later, we'll invoke the operation based on 267 * the first mapped vnode's operation vector. 268 */ 269 reles = descp->vdesc_flags; 270 for (i = 0; i < VDESC_MAX_VPS; reles >>= 1, i++) { 271 if (descp->vdesc_vp_offsets[i] == VDESC_NO_OFFSET) 272 break; /* bail out at end of list */ 273 vps_p[i] = this_vp_p = 274 VOPARG_OFFSETTO(struct vnode**,descp->vdesc_vp_offsets[i],ap); 275 /* 276 * We're not guaranteed that any but the first vnode 277 * are of our type. Check for and don't map any 278 * that aren't. (We must always map first vp or vclean fails.) 279 */ 280 if (i && (*this_vp_p == NULLVP || 281 (*this_vp_p)->v_op != null_vnodeop_p)) { 282 old_vps[i] = NULLVP; 283 } else { 284 old_vps[i] = *this_vp_p; 285 *(vps_p[i]) = NULLVPTOLOWERVP(*this_vp_p); 286 /* 287 * XXX - Several operations have the side effect 288 * of vrele'ing their vp's. We must account for 289 * that. (This should go away in the future.) 290 */ 291 if (reles & VDESC_VP0_WILLRELE) 292 VREF(*this_vp_p); 293 } 294 295 } 296 297 /* 298 * Call the operation on the lower layer 299 * with the modified argument structure. 300 */ 301 if (vps_p[0] && *vps_p[0]) 302 error = VCALL(*(vps_p[0]), descp->vdesc_offset, ap); 303 else { 304 printf("null_bypass: no map for %s\n", descp->vdesc_name); 305 error = EINVAL; 306 } 307 308 /* 309 * Maintain the illusion of call-by-value 310 * by restoring vnodes in the argument structure 311 * to their original value. 312 */ 313 reles = descp->vdesc_flags; 314 for (i = 0; i < VDESC_MAX_VPS; reles >>= 1, i++) { 315 if (descp->vdesc_vp_offsets[i] == VDESC_NO_OFFSET) 316 break; /* bail out at end of list */ 317 if (old_vps[i]) { 318 *(vps_p[i]) = old_vps[i]; 319#if 0 320 if (reles & VDESC_VP0_WILLUNLOCK) 321 VOP_UNLOCK(*(vps_p[i]), LK_THISLAYER, curproc); 322#endif 323 if (reles & VDESC_VP0_WILLRELE) 324 vrele(*(vps_p[i])); 325 } 326 } 327 328 /* 329 * Map the possible out-going vpp 330 * (Assumes that the lower layer always returns 331 * a VREF'ed vpp unless it gets an error.) 332 */ 333 if (descp->vdesc_vpp_offset != VDESC_NO_OFFSET && 334 !(descp->vdesc_flags & VDESC_NOMAP_VPP) && 335 !error) { 336 /* 337 * XXX - even though some ops have vpp returned vp's, 338 * several ops actually vrele this before returning. 339 * We must avoid these ops. 340 * (This should go away when these ops are regularized.) 341 */ 342 if (descp->vdesc_flags & VDESC_VPP_WILLRELE) 343 goto out; 344 vppp = VOPARG_OFFSETTO(struct vnode***, 345 descp->vdesc_vpp_offset,ap); 346 if (*vppp) 347 error = null_node_create(old_vps[0]->v_mount, **vppp, *vppp); 348 } 349 350 out: 351 return (error); 352} 353 354/* 355 * We have to carry on the locking protocol on the null layer vnodes 356 * as we progress through the tree. We also have to enforce read-only 357 * if this layer is mounted read-only. 358 */ 359static int 360null_lookup(ap) 361 struct vop_lookup_args /* { 362 struct vnode * a_dvp; 363 struct vnode ** a_vpp; 364 struct componentname * a_cnp; 365 } */ *ap; 366{ 367 struct componentname *cnp = ap->a_cnp; 368 struct vnode *dvp = ap->a_dvp; 369 struct proc *p = cnp->cn_proc; 370 int flags = cnp->cn_flags; 371 struct vnode *vp, *ldvp, *lvp; 372 int error; 373 374 if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) && 375 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) 376 return (EROFS); 377 /* 378 * Although it is possible to call null_bypass(), we'll do 379 * a direct call to reduce overhead 380 */ 381 ldvp = NULLVPTOLOWERVP(dvp); 382 vp = lvp = NULL; 383 error = VOP_LOOKUP(ldvp, &lvp, cnp); 384 if (error == EJUSTRETURN && (flags & ISLASTCN) && 385 (dvp->v_mount->mnt_flag & MNT_RDONLY) && 386 (cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME)) 387 error = EROFS; 388 389 /* 390 * Rely only on the PDIRUNLOCK flag which should be carefully 391 * tracked by underlying filesystem. 392 */ 393 if (cnp->cn_flags & PDIRUNLOCK) 394 VOP_UNLOCK(dvp, LK_THISLAYER, p); 395 if ((error == 0 || error == EJUSTRETURN) && lvp != NULL) { 396 if (ldvp == lvp) { 397 *ap->a_vpp = dvp; 398 VREF(dvp); 399 vrele(lvp); 400 } else { 401 error = null_node_create(dvp->v_mount, lvp, &vp); 402 if (error == 0) 403 *ap->a_vpp = vp; 404 } 405 } 406 return (error); 407} 408 409/* 410 * Setattr call. Disallow write attempts if the layer is mounted read-only. 411 */ 412int 413null_setattr(ap) 414 struct vop_setattr_args /* { 415 struct vnodeop_desc *a_desc; 416 struct vnode *a_vp; 417 struct vattr *a_vap; 418 struct ucred *a_cred; 419 struct proc *a_p; 420 } */ *ap; 421{ 422 struct vnode *vp = ap->a_vp; 423 struct vattr *vap = ap->a_vap; 424 425 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL || 426 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL || 427 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) && 428 (vp->v_mount->mnt_flag & MNT_RDONLY)) 429 return (EROFS); 430 if (vap->va_size != VNOVAL) { 431 switch (vp->v_type) { 432 case VDIR: 433 return (EISDIR); 434 case VCHR: 435 case VBLK: 436 case VSOCK: 437 case VFIFO: 438 if (vap->va_flags != VNOVAL) 439 return (EOPNOTSUPP); 440 return (0); 441 case VREG: 442 case VLNK: 443 default: 444 /* 445 * Disallow write attempts if the filesystem is 446 * mounted read-only. 447 */ 448 if (vp->v_mount->mnt_flag & MNT_RDONLY) 449 return (EROFS); 450 } 451 } 452 453 return (null_bypass((struct vop_generic_args *)ap)); 454} 455 456/* 457 * We handle getattr only to change the fsid. 458 */ 459static int 460null_getattr(ap) 461 struct vop_getattr_args /* { 462 struct vnode *a_vp; 463 struct vattr *a_vap; 464 struct ucred *a_cred; 465 struct proc *a_p; 466 } */ *ap; 467{ 468 int error; 469 470 if ((error = null_bypass((struct vop_generic_args *)ap)) != 0) 471 return (error); 472 473 ap->a_vap->va_fsid = ap->a_vp->v_mount->mnt_stat.f_fsid.val[0]; 474 return (0); 475} 476 477/* 478 * Handle to disallow write access if mounted read-only. 479 */ 480static int 481null_access(ap) 482 struct vop_access_args /* { 483 struct vnode *a_vp; 484 int a_mode; 485 struct ucred *a_cred; 486 struct proc *a_p; 487 } */ *ap; 488{ 489 struct vnode *vp = ap->a_vp; 490 mode_t mode = ap->a_mode; 491 492 /* 493 * Disallow write attempts on read-only layers; 494 * unless the file is a socket, fifo, or a block or 495 * character device resident on the file system. 496 */ 497 if (mode & VWRITE) { 498 switch (vp->v_type) { 499 case VDIR: 500 case VLNK: 501 case VREG: 502 if (vp->v_mount->mnt_flag & MNT_RDONLY) 503 return (EROFS); 504 break; 505 default: 506 break; 507 } 508 } 509 return (null_bypass((struct vop_generic_args *)ap)); 510} 511 512/* 513 * We must handle open to be able to catch MNT_NODEV and friends. 514 */ 515static int 516null_open(ap) 517 struct vop_open_args /* { 518 struct vnode *a_vp; 519 int a_mode; 520 struct ucred *a_cred; 521 struct proc *a_p; 522 } */ *ap; 523{ 524 struct vnode *vp = ap->a_vp; 525 struct vnode *lvp = NULLVPTOLOWERVP(ap->a_vp); 526 527 if ((vp->v_mount->mnt_flag & MNT_NODEV) && 528 (lvp->v_type == VBLK || lvp->v_type == VCHR)) 529 return ENXIO; 530 531 return (null_bypass((struct vop_generic_args *)ap)); 532} 533 534/* 535 * We handle this to eliminate null FS to lower FS 536 * file moving. Don't know why we don't allow this, 537 * possibly we should. 538 */ 539static int 540null_rename(ap) 541 struct vop_rename_args /* { 542 struct vnode *a_fdvp; 543 struct vnode *a_fvp; 544 struct componentname *a_fcnp; 545 struct vnode *a_tdvp; 546 struct vnode *a_tvp; 547 struct componentname *a_tcnp; 548 } */ *ap; 549{ 550 struct vnode *tdvp = ap->a_tdvp; 551 struct vnode *fvp = ap->a_fvp; 552 struct vnode *fdvp = ap->a_fdvp; 553 struct vnode *tvp = ap->a_tvp; 554 555 /* Check for cross-device rename. */ 556 if ((fvp->v_mount != tdvp->v_mount) || 557 (tvp && (fvp->v_mount != tvp->v_mount))) { 558 if (tdvp == tvp) 559 vrele(tdvp); 560 else 561 vput(tdvp); 562 if (tvp) 563 vput(tvp); 564 vrele(fdvp); 565 vrele(fvp); 566 return (EXDEV); 567 } 568 569 return (null_bypass((struct vop_generic_args *)ap)); 570} 571 572/* 573 * We need to process our own vnode lock and then clear the 574 * interlock flag as it applies only to our vnode, not the 575 * vnodes below us on the stack. 576 */ 577static int 578null_lock(ap) 579 struct vop_lock_args /* { 580 struct vnode *a_vp; 581 int a_flags; 582 struct proc *a_p; 583 } */ *ap; 584{ 585 struct vnode *vp = ap->a_vp; 586 int flags = ap->a_flags; 587 struct proc *p = ap->a_p; 588 struct vnode *lvp; 589 int error; 590 591 if (flags & LK_THISLAYER) { 592 if (vp->v_vnlock != NULL) 593 return 0; /* lock is shared across layers */ 594 error = lockmgr(&vp->v_lock, flags & ~LK_THISLAYER, 595 &vp->v_interlock, p); 596 return (error); 597 } 598 599 if (vp->v_vnlock != NULL) { 600 /* 601 * The lower level has exported a struct lock to us. Use 602 * it so that all vnodes in the stack lock and unlock 603 * simultaneously. Note: we don't DRAIN the lock as DRAIN 604 * decommissions the lock - just because our vnode is 605 * going away doesn't mean the struct lock below us is. 606 * LK_EXCLUSIVE is fine. 607 */ 608 if ((flags & LK_TYPE_MASK) == LK_DRAIN) { 609 NULLFSDEBUG("null_lock: avoiding LK_DRAIN\n"); 610 return(lockmgr(vp->v_vnlock, 611 (flags & ~LK_TYPE_MASK) | LK_EXCLUSIVE, 612 &vp->v_interlock, p)); 613 } 614 return(lockmgr(vp->v_vnlock, flags, &vp->v_interlock, p)); 615 } else { 616 /* 617 * To prevent race conditions involving doing a lookup 618 * on "..", we have to lock the lower node, then lock our 619 * node. Most of the time it won't matter that we lock our 620 * node (as any locking would need the lower one locked 621 * first). But we can LK_DRAIN the upper lock as a step 622 * towards decomissioning it. 623 */ 624 lvp = NULLVPTOLOWERVP(vp); 625 if (lvp == NULL) 626 return (lockmgr(&vp->v_lock, flags, &vp->v_interlock, p)); 627 if (flags & LK_INTERLOCK) { 628 mtx_exit(&vp->v_interlock, MTX_DEF); 629 flags &= ~LK_INTERLOCK; 630 } 631 if ((flags & LK_TYPE_MASK) == LK_DRAIN) { 632 error = VOP_LOCK(lvp, 633 (flags & ~LK_TYPE_MASK) | LK_EXCLUSIVE, p); 634 } else 635 error = VOP_LOCK(lvp, flags, p); 636 if (error) 637 return (error); 638 error = lockmgr(&vp->v_lock, flags, &vp->v_interlock, p); 639 if (error) 640 VOP_UNLOCK(lvp, 0, p); 641 return (error); 642 } 643} 644 645/* 646 * We need to process our own vnode unlock and then clear the 647 * interlock flag as it applies only to our vnode, not the 648 * vnodes below us on the stack. 649 */ 650static int 651null_unlock(ap) 652 struct vop_unlock_args /* { 653 struct vnode *a_vp; 654 int a_flags; 655 struct proc *a_p; 656 } */ *ap; 657{ 658 struct vnode *vp = ap->a_vp; 659 int flags = ap->a_flags; 660 struct proc *p = ap->a_p; 661 struct vnode *lvp; 662 663 if (vp->v_vnlock != NULL) { 664 if (flags & LK_THISLAYER) 665 return 0; /* the lock is shared across layers */ 666 flags &= ~LK_THISLAYER; 667 return (lockmgr(vp->v_vnlock, flags | LK_RELEASE, 668 &vp->v_interlock, p)); 669 } 670 lvp = NULLVPTOLOWERVP(vp); 671 if (lvp == NULL) 672 return (lockmgr(&vp->v_lock, flags | LK_RELEASE, &vp->v_interlock, p)); 673 if ((flags & LK_THISLAYER) == 0) { 674 if (flags & LK_INTERLOCK) { 675 mtx_exit(&vp->v_interlock, MTX_DEF); 676 flags &= ~LK_INTERLOCK; 677 } 678 VOP_UNLOCK(lvp, flags & ~LK_INTERLOCK, p); 679 } else 680 flags &= ~LK_THISLAYER; 681 return (lockmgr(&vp->v_lock, flags | LK_RELEASE, &vp->v_interlock, p)); 682} 683 684static int 685null_islocked(ap) 686 struct vop_islocked_args /* { 687 struct vnode *a_vp; 688 struct proc *a_p; 689 } */ *ap; 690{ 691 struct vnode *vp = ap->a_vp; 692 struct proc *p = ap->a_p; 693 694 if (vp->v_vnlock != NULL) 695 return (lockstatus(vp->v_vnlock, p)); 696 return (lockstatus(&vp->v_lock, p)); 697} 698 699/* 700 * There is no way to tell that someone issued remove/rmdir operation 701 * on the underlying filesystem. For now we just have to release lowevrp 702 * as soon as possible. 703 */ 704static int 705null_inactive(ap) 706 struct vop_inactive_args /* { 707 struct vnode *a_vp; 708 struct proc *a_p; 709 } */ *ap; 710{ 711 struct vnode *vp = ap->a_vp; 712 struct proc *p = ap->a_p; 713 struct null_node *xp = VTONULL(vp); 714 struct vnode *lowervp = xp->null_lowervp; 715 716 lockmgr(&null_hashlock, LK_EXCLUSIVE, NULL, p); 717 LIST_REMOVE(xp, null_hash); 718 lockmgr(&null_hashlock, LK_RELEASE, NULL, p); 719 720 xp->null_lowervp = NULLVP; 721 if (vp->v_vnlock != NULL) { 722 vp->v_vnlock = &vp->v_lock; /* we no longer share the lock */ 723 } else 724 VOP_UNLOCK(vp, LK_THISLAYER, p); 725 726 vput(lowervp); 727 /* 728 * Now it is safe to drop references to the lower vnode. 729 * VOP_INACTIVE() will be called by vrele() if necessary. 730 */ 731 vrele (lowervp); 732 733 return (0); 734} 735 736/* 737 * We can free memory in null_inactive, but we do this 738 * here. (Possible to guard vp->v_data to point somewhere) 739 */ 740static int 741null_reclaim(ap) 742 struct vop_reclaim_args /* { 743 struct vnode *a_vp; 744 struct proc *a_p; 745 } */ *ap; 746{ 747 struct vnode *vp = ap->a_vp; 748 void *vdata = vp->v_data; 749 750 vp->v_data = NULL; 751 FREE(vdata, M_NULLFSNODE); 752 753 return (0); 754} 755 756static int 757null_print(ap) 758 struct vop_print_args /* { 759 struct vnode *a_vp; 760 } */ *ap; 761{ 762 register struct vnode *vp = ap->a_vp; 763 printf ("\ttag VT_NULLFS, vp=%p, lowervp=%p\n", vp, NULLVPTOLOWERVP(vp)); 764 return (0); 765} 766 767/* 768 * Let an underlying filesystem do the work 769 */ 770static int 771null_createvobject(ap) 772 struct vop_createvobject_args /* { 773 struct vnode *vp; 774 struct ucred *cred; 775 struct proc *p; 776 } */ *ap; 777{ 778 struct vnode *vp = ap->a_vp; 779 struct vnode *lowervp = VTONULL(vp) ? NULLVPTOLOWERVP(vp) : NULL; 780 int error; 781 782 if (vp->v_type == VNON || lowervp == NULL) 783 return 0; 784 error = VOP_CREATEVOBJECT(lowervp, ap->a_cred, ap->a_p); 785 if (error) 786 return (error); 787 vp->v_flag |= VOBJBUF; 788 return (0); 789} 790 791/* 792 * We have nothing to destroy and this operation shouldn't be bypassed. 793 */ 794static int 795null_destroyvobject(ap) 796 struct vop_destroyvobject_args /* { 797 struct vnode *vp; 798 } */ *ap; 799{ 800 struct vnode *vp = ap->a_vp; 801 802 vp->v_flag &= ~VOBJBUF; 803 return (0); 804} 805 806static int 807null_getvobject(ap) 808 struct vop_getvobject_args /* { 809 struct vnode *vp; 810 struct vm_object **objpp; 811 } */ *ap; 812{ 813 struct vnode *lvp = NULLVPTOLOWERVP(ap->a_vp); 814 815 if (lvp == NULL) 816 return EINVAL; 817 return (VOP_GETVOBJECT(lvp, ap->a_objpp)); 818} 819 820/* 821 * Global vfs data structures 822 */ 823vop_t **null_vnodeop_p; 824static struct vnodeopv_entry_desc null_vnodeop_entries[] = { 825 { &vop_default_desc, (vop_t *) null_bypass }, 826 827 { &vop_access_desc, (vop_t *) null_access }, 828 { &vop_bmap_desc, (vop_t *) vop_eopnotsupp }, 829 { &vop_createvobject_desc, (vop_t *) null_createvobject }, 830 { &vop_destroyvobject_desc, (vop_t *) null_destroyvobject }, 831 { &vop_getattr_desc, (vop_t *) null_getattr }, 832 { &vop_getvobject_desc, (vop_t *) null_getvobject }, 833 { &vop_getwritemount_desc, (vop_t *) vop_stdgetwritemount}, 834 { &vop_inactive_desc, (vop_t *) null_inactive }, 835 { &vop_islocked_desc, (vop_t *) null_islocked }, 836 { &vop_lock_desc, (vop_t *) null_lock }, 837 { &vop_lookup_desc, (vop_t *) null_lookup }, 838 { &vop_open_desc, (vop_t *) null_open }, 839 { &vop_print_desc, (vop_t *) null_print }, 840 { &vop_reclaim_desc, (vop_t *) null_reclaim }, 841 { &vop_rename_desc, (vop_t *) null_rename }, 842 { &vop_setattr_desc, (vop_t *) null_setattr }, 843 { &vop_strategy_desc, (vop_t *) vop_eopnotsupp }, 844 { &vop_unlock_desc, (vop_t *) null_unlock }, 845 { NULL, NULL } 846}; 847static struct vnodeopv_desc null_vnodeop_opv_desc = 848 { &null_vnodeop_p, null_vnodeop_entries }; 849 850VNODEOP_SET(null_vnodeop_opv_desc); 851