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 * $FreeBSD: src/sys/miscfs/nullfs/null_vnops.c,v 1.38.2.6 2002/07/31 00:32:28 semenu Exp $ 41 * ...and... 42 * @(#)null_vnodeops.c 1.20 92/07/07 UCLA Ficus project 43 * 44 * $FreeBSD: src/sys/miscfs/nullfs/null_vnops.c,v 1.38.2.6 2002/07/31 00:32:28 semenu Exp $ 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 its "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, through 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 handled 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 invoke 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 vnode 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/vnode.h> 182 #include <sys/mount.h> 183 #include <sys/namei.h> 184 #include <sys/malloc.h> 185 #include <sys/buf.h> 186 #include <miscfs/nullfs/null.h> 187 188 static int null_bug_bypass = 0; /* for debugging: enables bypass printf'ing */ 189 SYSCTL_INT(_debug, OID_AUTO, nullfs_bug_bypass, CTLFLAG_RW, 190 &null_bug_bypass, 0, ""); 191 192 static int null_access(struct vop_access_args *ap); 193 static int null_createvobject(struct vop_createvobject_args *ap); 194 static int null_destroyvobject(struct vop_destroyvobject_args *ap); 195 static int null_getattr(struct vop_getattr_args *ap); 196 static int null_getvobject(struct vop_getvobject_args *ap); 197 static int null_inactive(struct vop_inactive_args *ap); 198 static int null_islocked(struct vop_islocked_args *ap); 199 static int null_lock(struct vop_lock_args *ap); 200 static int null_lookup(struct vop_lookup_args *ap); 201 static int null_open(struct vop_open_args *ap); 202 static int null_print(struct vop_print_args *ap); 203 static int null_reclaim(struct vop_reclaim_args *ap); 204 static int null_rename(struct vop_rename_args *ap); 205 static int null_setattr(struct vop_setattr_args *ap); 206 static int null_unlock(struct vop_unlock_args *ap); 207 208 /* 209 * This is the 10-Apr-92 bypass routine. 210 * This version has been optimized for speed, throwing away some 211 * safety checks. It should still always work, but it's not as 212 * robust to programmer errors. 213 * 214 * In general, we map all vnodes going down and unmap them on the way back. 215 * As an exception to this, vnodes can be marked "unmapped" by setting 216 * the Nth bit in operation's vdesc_flags. 217 * 218 * Also, some BSD vnode operations have the side effect of vrele'ing 219 * their arguments. With stacking, the reference counts are held 220 * by the upper node, not the lower one, so we must handle these 221 * side-effects here. This is not of concern in Sun-derived systems 222 * since there are no such side-effects. 223 * 224 * This makes the following assumptions: 225 * - only one returned vpp 226 * - no INOUT vpp's (Sun's vop_open has one of these) 227 * - the vnode operation vector of the first vnode should be used 228 * to determine what implementation of the op should be invoked 229 * - all mapped vnodes are of our vnode-type (NEEDSWORK: 230 * problems on rmdir'ing mount points and renaming?) 231 */ 232 int 233 null_bypass(ap) 234 struct vop_generic_args /* { 235 struct vnodeop_desc *a_desc; 236 <other random data follows, presumably> 237 } */ *ap; 238 { 239 register struct vnode **this_vp_p; 240 int error; 241 struct vnode *old_vps[VDESC_MAX_VPS]; 242 struct vnode **vps_p[VDESC_MAX_VPS]; 243 struct vnode ***vppp; 244 struct vnodeop_desc *descp = ap->a_desc; 245 int reles, i; 246 247 if (null_bug_bypass) 248 printf ("null_bypass: %s\n", descp->vdesc_name); 249 250 #ifdef DIAGNOSTIC 251 /* 252 * We require at least one vp. 253 */ 254 if (descp->vdesc_vp_offsets == NULL || 255 descp->vdesc_vp_offsets[0] == VDESC_NO_OFFSET) 256 panic ("null_bypass: no vp's in map"); 257 #endif 258 259 /* 260 * Map the vnodes going in. 261 * Later, we'll invoke the operation based on 262 * the first mapped vnode's operation vector. 263 */ 264 reles = descp->vdesc_flags; 265 for (i = 0; i < VDESC_MAX_VPS; reles >>= 1, i++) { 266 if (descp->vdesc_vp_offsets[i] == VDESC_NO_OFFSET) 267 break; /* bail out at end of list */ 268 vps_p[i] = this_vp_p = 269 VOPARG_OFFSETTO(struct vnode**,descp->vdesc_vp_offsets[i],ap); 270 /* 271 * We're not guaranteed that any but the first vnode 272 * are of our type. Check for and don't map any 273 * that aren't. (We must always map first vp or vclean fails.) 274 */ 275 if (i && (*this_vp_p == NULLVP || 276 (*this_vp_p)->v_op != null_vnodeop_p)) { 277 old_vps[i] = NULLVP; 278 } else { 279 old_vps[i] = *this_vp_p; 280 *(vps_p[i]) = NULLVPTOLOWERVP(*this_vp_p); 281 /* 282 * XXX - Several operations have the side effect 283 * of vrele'ing their vp's. We must account for 284 * that. (This should go away in the future.) 285 */ 286 if (reles & VDESC_VP0_WILLRELE) 287 VREF(*this_vp_p); 288 } 289 290 } 291 292 /* 293 * Call the operation on the lower layer 294 * with the modified argument structure. 295 */ 296 if (vps_p[0] && *vps_p[0]) 297 error = VCALL(*(vps_p[0]), descp->vdesc_offset, ap); 298 else { 299 printf("null_bypass: no map for %s\n", descp->vdesc_name); 300 error = EINVAL; 301 } 302 303 /* 304 * Maintain the illusion of call-by-value 305 * by restoring vnodes in the argument structure 306 * to their original value. 307 */ 308 reles = descp->vdesc_flags; 309 for (i = 0; i < VDESC_MAX_VPS; reles >>= 1, i++) { 310 if (descp->vdesc_vp_offsets[i] == VDESC_NO_OFFSET) 311 break; /* bail out at end of list */ 312 if (old_vps[i]) { 313 *(vps_p[i]) = old_vps[i]; 314 #if 0 315 if (reles & VDESC_VP0_WILLUNLOCK) 316 VOP_UNLOCK(*(vps_p[i]), LK_THISLAYER, curproc); 317 #endif 318 if (reles & VDESC_VP0_WILLRELE) 319 vrele(*(vps_p[i])); 320 } 321 } 322 323 /* 324 * Map the possible out-going vpp 325 * (Assumes that the lower layer always returns 326 * a VREF'ed vpp unless it gets an error.) 327 */ 328 if (descp->vdesc_vpp_offset != VDESC_NO_OFFSET && 329 !(descp->vdesc_flags & VDESC_NOMAP_VPP) && 330 !error) { 331 /* 332 * XXX - even though some ops have vpp returned vp's, 333 * several ops actually vrele this before returning. 334 * We must avoid these ops. 335 * (This should go away when these ops are regularized.) 336 */ 337 if (descp->vdesc_flags & VDESC_VPP_WILLRELE) 338 goto out; 339 vppp = VOPARG_OFFSETTO(struct vnode***, 340 descp->vdesc_vpp_offset,ap); 341 if (*vppp) 342 error = null_node_create(old_vps[0]->v_mount, **vppp, *vppp); 343 } 344 345 out: 346 return (error); 347 } 348 349 /* 350 * We have to carry on the locking protocol on the null layer vnodes 351 * as we progress through the tree. We also have to enforce read-only 352 * if this layer is mounted read-only. 353 */ 354 static int 355 null_lookup(ap) 356 struct vop_lookup_args /* { 357 struct vnode * a_dvp; 358 struct vnode ** a_vpp; 359 struct componentname * a_cnp; 360 } */ *ap; 361 { 362 struct componentname *cnp = ap->a_cnp; 363 struct vnode *dvp = ap->a_dvp; 364 struct proc *p = cnp->cn_proc; 365 int flags = cnp->cn_flags; 366 struct vnode *vp, *ldvp, *lvp; 367 int error; 368 369 if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) && 370 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) 371 return (EROFS); 372 /* 373 * Although it is possible to call null_bypass(), we'll do 374 * a direct call to reduce overhead 375 */ 376 ldvp = NULLVPTOLOWERVP(dvp); 377 vp = lvp = NULL; 378 error = VOP_LOOKUP(ldvp, &lvp, cnp); 379 if (error == EJUSTRETURN && (flags & ISLASTCN) && 380 (dvp->v_mount->mnt_flag & MNT_RDONLY) && 381 (cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME)) 382 error = EROFS; 383 384 /* 385 * Rely only on the PDIRUNLOCK flag which should be carefully 386 * tracked by underlying filesystem. 387 */ 388 if (cnp->cn_flags & PDIRUNLOCK) 389 VOP_UNLOCK(dvp, LK_THISLAYER, p); 390 if ((error == 0 || error == EJUSTRETURN) && lvp != NULL) { 391 if (ldvp == lvp) { 392 *ap->a_vpp = dvp; 393 VREF(dvp); 394 vrele(lvp); 395 } else { 396 error = null_node_create(dvp->v_mount, lvp, &vp); 397 if (error == 0) 398 *ap->a_vpp = vp; 399 } 400 } 401 return (error); 402 } 403 404 /* 405 * Setattr call. Disallow write attempts if the layer is mounted read-only. 406 */ 407 int 408 null_setattr(ap) 409 struct vop_setattr_args /* { 410 struct vnodeop_desc *a_desc; 411 struct vnode *a_vp; 412 struct vattr *a_vap; 413 struct ucred *a_cred; 414 struct proc *a_p; 415 } */ *ap; 416 { 417 struct vnode *vp = ap->a_vp; 418 struct vattr *vap = ap->a_vap; 419 420 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL || 421 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL || 422 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) && 423 (vp->v_mount->mnt_flag & MNT_RDONLY)) 424 return (EROFS); 425 if (vap->va_size != VNOVAL) { 426 switch (vp->v_type) { 427 case VDIR: 428 return (EISDIR); 429 case VCHR: 430 case VBLK: 431 case VSOCK: 432 case VFIFO: 433 if (vap->va_flags != VNOVAL) 434 return (EOPNOTSUPP); 435 return (0); 436 case VREG: 437 case VLNK: 438 default: 439 /* 440 * Disallow write attempts if the filesystem is 441 * mounted read-only. 442 */ 443 if (vp->v_mount->mnt_flag & MNT_RDONLY) 444 return (EROFS); 445 } 446 } 447 448 return (null_bypass((struct vop_generic_args *)ap)); 449 } 450 451 /* 452 * We handle getattr only to change the fsid. 453 */ 454 static int 455 null_getattr(ap) 456 struct vop_getattr_args /* { 457 struct vnode *a_vp; 458 struct vattr *a_vap; 459 struct ucred *a_cred; 460 struct proc *a_p; 461 } */ *ap; 462 { 463 int error; 464 465 if ((error = null_bypass((struct vop_generic_args *)ap)) != 0) 466 return (error); 467 468 ap->a_vap->va_fsid = ap->a_vp->v_mount->mnt_stat.f_fsid.val[0]; 469 return (0); 470 } 471 472 /* 473 * Handle to disallow write access if mounted read-only. 474 */ 475 static int 476 null_access(ap) 477 struct vop_access_args /* { 478 struct vnode *a_vp; 479 int a_mode; 480 struct ucred *a_cred; 481 struct proc *a_p; 482 } */ *ap; 483 { 484 struct vnode *vp = ap->a_vp; 485 mode_t mode = ap->a_mode; 486 487 /* 488 * Disallow write attempts on read-only layers; 489 * unless the file is a socket, fifo, or a block or 490 * character device resident on the file system. 491 */ 492 if (mode & VWRITE) { 493 switch (vp->v_type) { 494 case VDIR: 495 case VLNK: 496 case VREG: 497 if (vp->v_mount->mnt_flag & MNT_RDONLY) 498 return (EROFS); 499 break; 500 default: 501 break; 502 } 503 } 504 return (null_bypass((struct vop_generic_args *)ap)); 505 } 506 507 /* 508 * We must handle open to be able to catch MNT_NODEV and friends. 509 */ 510 static int 511 null_open(ap) 512 struct vop_open_args /* { 513 struct vnode *a_vp; 514 int a_mode; 515 struct ucred *a_cred; 516 struct proc *a_p; 517 } */ *ap; 518 { 519 struct vnode *vp = ap->a_vp; 520 struct vnode *lvp = NULLVPTOLOWERVP(ap->a_vp); 521 522 if ((vp->v_mount->mnt_flag & MNT_NODEV) && 523 (lvp->v_type == VBLK || lvp->v_type == VCHR)) 524 return ENXIO; 525 526 return (null_bypass((struct vop_generic_args *)ap)); 527 } 528 529 /* 530 * We handle this to eliminate null FS to lower FS 531 * file moving. Don't know why we don't allow this, 532 * possibly we should. 533 */ 534 static int 535 null_rename(ap) 536 struct vop_rename_args /* { 537 struct vnode *a_fdvp; 538 struct vnode *a_fvp; 539 struct componentname *a_fcnp; 540 struct vnode *a_tdvp; 541 struct vnode *a_tvp; 542 struct componentname *a_tcnp; 543 } */ *ap; 544 { 545 struct vnode *tdvp = ap->a_tdvp; 546 struct vnode *fvp = ap->a_fvp; 547 struct vnode *fdvp = ap->a_fdvp; 548 struct vnode *tvp = ap->a_tvp; 549 550 /* Check for cross-device rename. */ 551 if ((fvp->v_mount != tdvp->v_mount) || 552 (tvp && (fvp->v_mount != tvp->v_mount))) { 553 if (tdvp == tvp) 554 vrele(tdvp); 555 else 556 vput(tdvp); 557 if (tvp) 558 vput(tvp); 559 vrele(fdvp); 560 vrele(fvp); 561 return (EXDEV); 562 } 563 564 return (null_bypass((struct vop_generic_args *)ap)); 565 } 566 567 /* 568 * We need to process our own vnode lock and then clear the 569 * interlock flag as it applies only to our vnode, not the 570 * vnodes below us on the stack. 571 */ 572 static int 573 null_lock(ap) 574 struct vop_lock_args /* { 575 struct vnode *a_vp; 576 int a_flags; 577 struct proc *a_p; 578 } */ *ap; 579 { 580 struct vnode *vp = ap->a_vp; 581 int flags = ap->a_flags; 582 struct proc *p = ap->a_p; 583 struct null_node *np = VTONULL(vp); 584 struct vnode *lvp; 585 int error; 586 587 if (flags & LK_THISLAYER) { 588 if (vp->v_vnlock != NULL) { 589 /* lock is shared across layers */ 590 if (flags & LK_INTERLOCK) 591 simple_unlock(&vp->v_interlock); 592 return 0; 593 } 594 error = lockmgr(&np->null_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 } 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(&np->null_lock, flags, &vp->v_interlock, p)); 627 if (flags & LK_INTERLOCK) { 628 VI_UNLOCK(vp); 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(&np->null_lock, flags, &vp->v_interlock, p); 639 if (error) 640 VOP_UNLOCK(lvp, 0, p); 641 return (error); 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 */ 649 static int 650 null_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 null_node *np = VTONULL(vp); 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(&np->null_lock, flags | LK_RELEASE, &vp->v_interlock, p)); 673 if ((flags & LK_THISLAYER) == 0) { 674 if (flags & LK_INTERLOCK) { 675 VI_UNLOCK(vp); 676 flags &= ~LK_INTERLOCK; 677 } 678 VOP_UNLOCK(lvp, flags, p); 679 } else 680 flags &= ~LK_THISLAYER; 681 ap->a_flags = flags; 682 return (lockmgr(&np->null_lock, flags | LK_RELEASE, &vp->v_interlock, p)); 683 } 684 685 static int 686 null_islocked(ap) 687 struct vop_islocked_args /* { 688 struct vnode *a_vp; 689 struct proc *a_p; 690 } */ *ap; 691 { 692 struct vnode *vp = ap->a_vp; 693 struct proc *p = ap->a_p; 694 695 if (vp->v_vnlock != NULL) 696 return (lockstatus(vp->v_vnlock, p)); 697 return (lockstatus(&VTONULL(vp)->null_lock, p)); 698 } 699 700 701 /* 702 * There is no way to tell that someone issued remove/rmdir operation 703 * on the underlying filesystem. For now we just have to release lowevrp 704 * as soon as possible. 705 */ 706 static int 707 null_inactive(ap) 708 struct vop_inactive_args /* { 709 struct vnode *a_vp; 710 struct proc *a_p; 711 } */ *ap; 712 { 713 struct vnode *vp = ap->a_vp; 714 struct proc *p = ap->a_p; 715 struct null_node *xp = VTONULL(vp); 716 struct vnode *lowervp = xp->null_lowervp; 717 718 lockmgr(&null_hashlock, LK_EXCLUSIVE, NULL, p); 719 LIST_REMOVE(xp, null_hash); 720 lockmgr(&null_hashlock, LK_RELEASE, NULL, p); 721 722 xp->null_lowervp = NULLVP; 723 if (vp->v_vnlock != NULL) { 724 vp->v_vnlock = &xp->null_lock; /* we no longer share the lock */ 725 } else 726 VOP_UNLOCK(vp, LK_THISLAYER, p); 727 728 vput(lowervp); 729 /* 730 * Now it is safe to drop references to the lower vnode. 731 * VOP_INACTIVE() will be called by vrele() if necessary. 732 */ 733 vrele (lowervp); 734 735 return (0); 736 } 737 738 /* 739 * We can free memory in null_inactive, but we do this 740 * here. (Possible to guard vp->v_data to point somewhere) 741 */ 742 static int 743 null_reclaim(ap) 744 struct vop_reclaim_args /* { 745 struct vnode *a_vp; 746 struct proc *a_p; 747 } */ *ap; 748 { 749 struct vnode *vp = ap->a_vp; 750 void *vdata = vp->v_data; 751 752 vp->v_data = NULL; 753 FREE(vdata, M_NULLFSNODE); 754 755 return (0); 756 } 757 758 static int 759 null_print(ap) 760 struct vop_print_args /* { 761 struct vnode *a_vp; 762 } */ *ap; 763 { 764 struct vnode *vp = ap->a_vp; 765 766 printf ("\ttag VT_NULLFS, vp=%p, lowervp=%p\n", vp, NULLVPTOLOWERVP(vp)); 767 if (vp->v_vnlock != NULL) { 768 printf("\tvnlock: "); 769 lockmgr_printinfo(vp->v_vnlock); 770 } else { 771 printf("\tnull_lock: "); 772 lockmgr_printinfo(&VTONULL(vp)->null_lock); 773 } 774 printf("\n"); 775 return (0); 776 } 777 778 /* 779 * Let an underlying filesystem do the work 780 */ 781 static int 782 null_createvobject(ap) 783 struct vop_createvobject_args /* { 784 struct vnode *vp; 785 struct ucred *cred; 786 struct proc *p; 787 } */ *ap; 788 { 789 struct vnode *vp = ap->a_vp; 790 struct vnode *lowervp = VTONULL(vp) ? NULLVPTOLOWERVP(vp) : NULL; 791 int error; 792 793 if (vp->v_type == VNON || lowervp == NULL) 794 return 0; 795 error = VOP_CREATEVOBJECT(lowervp, ap->a_cred, ap->a_p); 796 if (error) 797 return (error); 798 vp->v_flag |= VOBJBUF; 799 return (0); 800 } 801 802 /* 803 * We have nothing to destroy and this operation shouldn't be bypassed. 804 */ 805 static int 806 null_destroyvobject(ap) 807 struct vop_destroyvobject_args /* { 808 struct vnode *vp; 809 } */ *ap; 810 { 811 struct vnode *vp = ap->a_vp; 812 813 vp->v_flag &= ~VOBJBUF; 814 return (0); 815 } 816 817 static int 818 null_getvobject(ap) 819 struct vop_getvobject_args /* { 820 struct vnode *vp; 821 struct vm_object **objpp; 822 } */ *ap; 823 { 824 struct vnode *lvp = NULLVPTOLOWERVP(ap->a_vp); 825 826 if (lvp == NULL) 827 return EINVAL; 828 return (VOP_GETVOBJECT(lvp, ap->a_objpp)); 829 } 830 831 /* 832 * Global vfs data structures 833 */ 834 vop_t **null_vnodeop_p; 835 static struct vnodeopv_entry_desc null_vnodeop_entries[] = { 836 { &vop_default_desc, (vop_t *) null_bypass }, 837 { &vop_access_desc, (vop_t *) null_access }, 838 { &vop_createvobject_desc, (vop_t *) null_createvobject }, 839 { &vop_destroyvobject_desc, (vop_t *) null_destroyvobject }, 840 { &vop_getattr_desc, (vop_t *) null_getattr }, 841 { &vop_getvobject_desc, (vop_t *) null_getvobject }, 842 { &vop_inactive_desc, (vop_t *) null_inactive }, 843 { &vop_islocked_desc, (vop_t *) null_islocked }, 844 { &vop_lock_desc, (vop_t *) null_lock }, 845 { &vop_lookup_desc, (vop_t *) null_lookup }, 846 { &vop_open_desc, (vop_t *) null_open }, 847 { &vop_print_desc, (vop_t *) null_print }, 848 { &vop_reclaim_desc, (vop_t *) null_reclaim }, 849 { &vop_rename_desc, (vop_t *) null_rename }, 850 { &vop_setattr_desc, (vop_t *) null_setattr }, 851 { &vop_unlock_desc, (vop_t *) null_unlock }, 852 { NULL, NULL } 853 }; 854 static struct vnodeopv_desc null_vnodeop_opv_desc = 855 { &null_vnodeop_p, null_vnodeop_entries }; 856 857 VNODEOP_SET(null_vnodeop_opv_desc); 858