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 * $DragonFly: src/sys/vfs/nullfs/null_vnops.c,v 1.2 2003/06/17 04:28:42 dillon Exp $ 42 * ...and... 43 * @(#)null_vnodeops.c 1.20 92/07/07 UCLA Ficus project 44 * 45 * $FreeBSD: src/sys/miscfs/nullfs/null_vnops.c,v 1.38.2.6 2002/07/31 00:32:28 semenu Exp $ 46 */ 47 48 /* 49 * Null Layer 50 * 51 * (See mount_null(8) for more information.) 52 * 53 * The null layer duplicates a portion of the file system 54 * name space under a new name. In this respect, it is 55 * similar to the loopback file system. It differs from 56 * the loopback fs in two respects: it is implemented using 57 * a stackable layers techniques, and its "null-node"s stack above 58 * all lower-layer vnodes, not just over directory vnodes. 59 * 60 * The null layer has two purposes. First, it serves as a demonstration 61 * of layering by proving a layer which does nothing. (It actually 62 * does everything the loopback file system does, which is slightly 63 * more than nothing.) Second, the null layer can serve as a prototype 64 * layer. Since it provides all necessary layer framework, 65 * new file system layers can be created very easily be starting 66 * with a null layer. 67 * 68 * The remainder of this man page examines the null layer as a basis 69 * for constructing new layers. 70 * 71 * 72 * INSTANTIATING NEW NULL LAYERS 73 * 74 * New null layers are created with mount_null(8). 75 * Mount_null(8) takes two arguments, the pathname 76 * of the lower vfs (target-pn) and the pathname where the null 77 * layer will appear in the namespace (alias-pn). After 78 * the null layer is put into place, the contents 79 * of target-pn subtree will be aliased under alias-pn. 80 * 81 * 82 * OPERATION OF A NULL LAYER 83 * 84 * The null layer is the minimum file system layer, 85 * simply bypassing all possible operations to the lower layer 86 * for processing there. The majority of its activity centers 87 * on the bypass routine, through which nearly all vnode operations 88 * pass. 89 * 90 * The bypass routine accepts arbitrary vnode operations for 91 * handling by the lower layer. It begins by examing vnode 92 * operation arguments and replacing any null-nodes by their 93 * lower-layer equivlants. It then invokes the operation 94 * on the lower layer. Finally, it replaces the null-nodes 95 * in the arguments and, if a vnode is return by the operation, 96 * stacks a null-node on top of the returned vnode. 97 * 98 * Although bypass handles most operations, vop_getattr, vop_lock, 99 * vop_unlock, vop_inactive, vop_reclaim, and vop_print are not 100 * bypassed. Vop_getattr must change the fsid being returned. 101 * Vop_lock and vop_unlock must handle any locking for the 102 * current vnode as well as pass the lock request down. 103 * Vop_inactive and vop_reclaim are not bypassed so that 104 * they can handle freeing null-layer specific data. Vop_print 105 * is not bypassed to avoid excessive debugging information. 106 * Also, certain vnode operations change the locking state within 107 * the operation (create, mknod, remove, link, rename, mkdir, rmdir, 108 * and symlink). Ideally these operations should not change the 109 * lock state, but should be changed to let the caller of the 110 * function unlock them. Otherwise all intermediate vnode layers 111 * (such as union, umapfs, etc) must catch these functions to do 112 * the necessary locking at their layer. 113 * 114 * 115 * INSTANTIATING VNODE STACKS 116 * 117 * Mounting associates the null layer with a lower layer, 118 * effect stacking two VFSes. Vnode stacks are instead 119 * created on demand as files are accessed. 120 * 121 * The initial mount creates a single vnode stack for the 122 * root of the new null layer. All other vnode stacks 123 * are created as a result of vnode operations on 124 * this or other null vnode stacks. 125 * 126 * New vnode stacks come into existance as a result of 127 * an operation which returns a vnode. 128 * The bypass routine stacks a null-node above the new 129 * vnode before returning it to the caller. 130 * 131 * For example, imagine mounting a null layer with 132 * "mount_null /usr/include /dev/layer/null". 133 * Changing directory to /dev/layer/null will assign 134 * the root null-node (which was created when the null layer was mounted). 135 * Now consider opening "sys". A vop_lookup would be 136 * done on the root null-node. This operation would bypass through 137 * to the lower layer which would return a vnode representing 138 * the UFS "sys". Null_bypass then builds a null-node 139 * aliasing the UFS "sys" and returns this to the caller. 140 * Later operations on the null-node "sys" will repeat this 141 * process when constructing other vnode stacks. 142 * 143 * 144 * CREATING OTHER FILE SYSTEM LAYERS 145 * 146 * One of the easiest ways to construct new file system layers is to make 147 * a copy of the null layer, rename all files and variables, and 148 * then begin modifing the copy. Sed can be used to easily rename 149 * all variables. 150 * 151 * The umap layer is an example of a layer descended from the 152 * null layer. 153 * 154 * 155 * INVOKING OPERATIONS ON LOWER LAYERS 156 * 157 * There are two techniques to invoke operations on a lower layer 158 * when the operation cannot be completely bypassed. Each method 159 * is appropriate in different situations. In both cases, 160 * it is the responsibility of the aliasing layer to make 161 * the operation arguments "correct" for the lower layer 162 * by mapping an vnode arguments to the lower layer. 163 * 164 * The first approach is to call the aliasing layer's bypass routine. 165 * This method is most suitable when you wish to invoke the operation 166 * currently being handled on the lower layer. It has the advantage 167 * that the bypass routine already must do argument mapping. 168 * An example of this is null_getattrs in the null layer. 169 * 170 * A second approach is to directly invoke vnode operations on 171 * the lower layer with the VOP_OPERATIONNAME interface. 172 * The advantage of this method is that it is easy to invoke 173 * arbitrary operations on the lower layer. The disadvantage 174 * is that vnode arguments must be manualy mapped. 175 * 176 */ 177 178 #include <sys/param.h> 179 #include <sys/systm.h> 180 #include <sys/kernel.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 <sys/buf.h> 187 #include <miscfs/nullfs/null.h> 188 189 static int null_bug_bypass = 0; /* for debugging: enables bypass printf'ing */ 190 SYSCTL_INT(_debug, OID_AUTO, nullfs_bug_bypass, CTLFLAG_RW, 191 &null_bug_bypass, 0, ""); 192 193 static int null_access(struct vop_access_args *ap); 194 static int null_createvobject(struct vop_createvobject_args *ap); 195 static int null_destroyvobject(struct vop_destroyvobject_args *ap); 196 static int null_getattr(struct vop_getattr_args *ap); 197 static int null_getvobject(struct vop_getvobject_args *ap); 198 static int null_inactive(struct vop_inactive_args *ap); 199 static int null_islocked(struct vop_islocked_args *ap); 200 static int null_lock(struct vop_lock_args *ap); 201 static int null_lookup(struct vop_lookup_args *ap); 202 static int null_open(struct vop_open_args *ap); 203 static int null_print(struct vop_print_args *ap); 204 static int null_reclaim(struct vop_reclaim_args *ap); 205 static int null_rename(struct vop_rename_args *ap); 206 static int null_setattr(struct vop_setattr_args *ap); 207 static int null_unlock(struct vop_unlock_args *ap); 208 209 /* 210 * This is the 10-Apr-92 bypass routine. 211 * This version has been optimized for speed, throwing away some 212 * safety checks. It should still always work, but it's not as 213 * robust to programmer errors. 214 * 215 * In general, we map all vnodes going down and unmap them on the way back. 216 * As an exception to this, vnodes can be marked "unmapped" by setting 217 * the Nth bit in operation's vdesc_flags. 218 * 219 * Also, some BSD vnode operations have the side effect of vrele'ing 220 * their arguments. With stacking, the reference counts are held 221 * by the upper node, not the lower one, so we must handle these 222 * side-effects here. This is not of concern in Sun-derived systems 223 * since there are no such side-effects. 224 * 225 * This makes the following assumptions: 226 * - only one returned vpp 227 * - no INOUT vpp's (Sun's vop_open has one of these) 228 * - the vnode operation vector of the first vnode should be used 229 * to determine what implementation of the op should be invoked 230 * - all mapped vnodes are of our vnode-type (NEEDSWORK: 231 * problems on rmdir'ing mount points and renaming?) 232 */ 233 int 234 null_bypass(ap) 235 struct vop_generic_args /* { 236 struct vnodeop_desc *a_desc; 237 <other random data follows, presumably> 238 } */ *ap; 239 { 240 register struct vnode **this_vp_p; 241 int error; 242 struct vnode *old_vps[VDESC_MAX_VPS]; 243 struct vnode **vps_p[VDESC_MAX_VPS]; 244 struct vnode ***vppp; 245 struct vnodeop_desc *descp = ap->a_desc; 246 int reles, i; 247 248 if (null_bug_bypass) 249 printf ("null_bypass: %s\n", descp->vdesc_name); 250 251 #ifdef DIAGNOSTIC 252 /* 253 * We require at least one vp. 254 */ 255 if (descp->vdesc_vp_offsets == NULL || 256 descp->vdesc_vp_offsets[0] == VDESC_NO_OFFSET) 257 panic ("null_bypass: no vp's in map"); 258 #endif 259 260 /* 261 * Map the vnodes going in. 262 * Later, we'll invoke the operation based on 263 * the first mapped vnode's operation vector. 264 */ 265 reles = descp->vdesc_flags; 266 for (i = 0; i < VDESC_MAX_VPS; reles >>= 1, i++) { 267 if (descp->vdesc_vp_offsets[i] == VDESC_NO_OFFSET) 268 break; /* bail out at end of list */ 269 vps_p[i] = this_vp_p = 270 VOPARG_OFFSETTO(struct vnode**,descp->vdesc_vp_offsets[i],ap); 271 /* 272 * We're not guaranteed that any but the first vnode 273 * are of our type. Check for and don't map any 274 * that aren't. (We must always map first vp or vclean fails.) 275 */ 276 if (i && (*this_vp_p == NULLVP || 277 (*this_vp_p)->v_op != null_vnodeop_p)) { 278 old_vps[i] = NULLVP; 279 } else { 280 old_vps[i] = *this_vp_p; 281 *(vps_p[i]) = NULLVPTOLOWERVP(*this_vp_p); 282 /* 283 * XXX - Several operations have the side effect 284 * of vrele'ing their vp's. We must account for 285 * that. (This should go away in the future.) 286 */ 287 if (reles & VDESC_VP0_WILLRELE) 288 VREF(*this_vp_p); 289 } 290 291 } 292 293 /* 294 * Call the operation on the lower layer 295 * with the modified argument structure. 296 */ 297 if (vps_p[0] && *vps_p[0]) 298 error = VCALL(*(vps_p[0]), descp->vdesc_offset, ap); 299 else { 300 printf("null_bypass: no map for %s\n", descp->vdesc_name); 301 error = EINVAL; 302 } 303 304 /* 305 * Maintain the illusion of call-by-value 306 * by restoring vnodes in the argument structure 307 * to their original value. 308 */ 309 reles = descp->vdesc_flags; 310 for (i = 0; i < VDESC_MAX_VPS; reles >>= 1, i++) { 311 if (descp->vdesc_vp_offsets[i] == VDESC_NO_OFFSET) 312 break; /* bail out at end of list */ 313 if (old_vps[i]) { 314 *(vps_p[i]) = old_vps[i]; 315 #if 0 316 if (reles & VDESC_VP0_WILLUNLOCK) 317 VOP_UNLOCK(*(vps_p[i]), LK_THISLAYER, curproc); 318 #endif 319 if (reles & VDESC_VP0_WILLRELE) 320 vrele(*(vps_p[i])); 321 } 322 } 323 324 /* 325 * Map the possible out-going vpp 326 * (Assumes that the lower layer always returns 327 * a VREF'ed vpp unless it gets an error.) 328 */ 329 if (descp->vdesc_vpp_offset != VDESC_NO_OFFSET && 330 !(descp->vdesc_flags & VDESC_NOMAP_VPP) && 331 !error) { 332 /* 333 * XXX - even though some ops have vpp returned vp's, 334 * several ops actually vrele this before returning. 335 * We must avoid these ops. 336 * (This should go away when these ops are regularized.) 337 */ 338 if (descp->vdesc_flags & VDESC_VPP_WILLRELE) 339 goto out; 340 vppp = VOPARG_OFFSETTO(struct vnode***, 341 descp->vdesc_vpp_offset,ap); 342 if (*vppp) 343 error = null_node_create(old_vps[0]->v_mount, **vppp, *vppp); 344 } 345 346 out: 347 return (error); 348 } 349 350 /* 351 * We have to carry on the locking protocol on the null layer vnodes 352 * as we progress through the tree. We also have to enforce read-only 353 * if this layer is mounted read-only. 354 */ 355 static int 356 null_lookup(ap) 357 struct vop_lookup_args /* { 358 struct vnode * a_dvp; 359 struct vnode ** a_vpp; 360 struct componentname * a_cnp; 361 } */ *ap; 362 { 363 struct componentname *cnp = ap->a_cnp; 364 struct vnode *dvp = ap->a_dvp; 365 struct proc *p = cnp->cn_proc; 366 int flags = cnp->cn_flags; 367 struct vnode *vp, *ldvp, *lvp; 368 int error; 369 370 if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) && 371 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) 372 return (EROFS); 373 /* 374 * Although it is possible to call null_bypass(), we'll do 375 * a direct call to reduce overhead 376 */ 377 ldvp = NULLVPTOLOWERVP(dvp); 378 vp = lvp = NULL; 379 error = VOP_LOOKUP(ldvp, &lvp, cnp); 380 if (error == EJUSTRETURN && (flags & ISLASTCN) && 381 (dvp->v_mount->mnt_flag & MNT_RDONLY) && 382 (cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME)) 383 error = EROFS; 384 385 /* 386 * Rely only on the PDIRUNLOCK flag which should be carefully 387 * tracked by underlying filesystem. 388 */ 389 if (cnp->cn_flags & PDIRUNLOCK) 390 VOP_UNLOCK(dvp, LK_THISLAYER, p); 391 if ((error == 0 || error == EJUSTRETURN) && lvp != NULL) { 392 if (ldvp == lvp) { 393 *ap->a_vpp = dvp; 394 VREF(dvp); 395 vrele(lvp); 396 } else { 397 error = null_node_create(dvp->v_mount, lvp, &vp); 398 if (error == 0) 399 *ap->a_vpp = vp; 400 } 401 } 402 return (error); 403 } 404 405 /* 406 * Setattr call. Disallow write attempts if the layer is mounted read-only. 407 */ 408 int 409 null_setattr(ap) 410 struct vop_setattr_args /* { 411 struct vnodeop_desc *a_desc; 412 struct vnode *a_vp; 413 struct vattr *a_vap; 414 struct ucred *a_cred; 415 struct proc *a_p; 416 } */ *ap; 417 { 418 struct vnode *vp = ap->a_vp; 419 struct vattr *vap = ap->a_vap; 420 421 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL || 422 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL || 423 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) && 424 (vp->v_mount->mnt_flag & MNT_RDONLY)) 425 return (EROFS); 426 if (vap->va_size != VNOVAL) { 427 switch (vp->v_type) { 428 case VDIR: 429 return (EISDIR); 430 case VCHR: 431 case VBLK: 432 case VSOCK: 433 case VFIFO: 434 if (vap->va_flags != VNOVAL) 435 return (EOPNOTSUPP); 436 return (0); 437 case VREG: 438 case VLNK: 439 default: 440 /* 441 * Disallow write attempts if the filesystem is 442 * mounted read-only. 443 */ 444 if (vp->v_mount->mnt_flag & MNT_RDONLY) 445 return (EROFS); 446 } 447 } 448 449 return (null_bypass((struct vop_generic_args *)ap)); 450 } 451 452 /* 453 * We handle getattr only to change the fsid. 454 */ 455 static int 456 null_getattr(ap) 457 struct vop_getattr_args /* { 458 struct vnode *a_vp; 459 struct vattr *a_vap; 460 struct ucred *a_cred; 461 struct proc *a_p; 462 } */ *ap; 463 { 464 int error; 465 466 if ((error = null_bypass((struct vop_generic_args *)ap)) != 0) 467 return (error); 468 469 ap->a_vap->va_fsid = ap->a_vp->v_mount->mnt_stat.f_fsid.val[0]; 470 return (0); 471 } 472 473 /* 474 * Handle to disallow write access if mounted read-only. 475 */ 476 static int 477 null_access(ap) 478 struct vop_access_args /* { 479 struct vnode *a_vp; 480 int a_mode; 481 struct ucred *a_cred; 482 struct proc *a_p; 483 } */ *ap; 484 { 485 struct vnode *vp = ap->a_vp; 486 mode_t mode = ap->a_mode; 487 488 /* 489 * Disallow write attempts on read-only layers; 490 * unless the file is a socket, fifo, or a block or 491 * character device resident on the file system. 492 */ 493 if (mode & VWRITE) { 494 switch (vp->v_type) { 495 case VDIR: 496 case VLNK: 497 case VREG: 498 if (vp->v_mount->mnt_flag & MNT_RDONLY) 499 return (EROFS); 500 break; 501 default: 502 break; 503 } 504 } 505 return (null_bypass((struct vop_generic_args *)ap)); 506 } 507 508 /* 509 * We must handle open to be able to catch MNT_NODEV and friends. 510 */ 511 static int 512 null_open(ap) 513 struct vop_open_args /* { 514 struct vnode *a_vp; 515 int a_mode; 516 struct ucred *a_cred; 517 struct proc *a_p; 518 } */ *ap; 519 { 520 struct vnode *vp = ap->a_vp; 521 struct vnode *lvp = NULLVPTOLOWERVP(ap->a_vp); 522 523 if ((vp->v_mount->mnt_flag & MNT_NODEV) && 524 (lvp->v_type == VBLK || lvp->v_type == VCHR)) 525 return ENXIO; 526 527 return (null_bypass((struct vop_generic_args *)ap)); 528 } 529 530 /* 531 * We handle this to eliminate null FS to lower FS 532 * file moving. Don't know why we don't allow this, 533 * possibly we should. 534 */ 535 static int 536 null_rename(ap) 537 struct vop_rename_args /* { 538 struct vnode *a_fdvp; 539 struct vnode *a_fvp; 540 struct componentname *a_fcnp; 541 struct vnode *a_tdvp; 542 struct vnode *a_tvp; 543 struct componentname *a_tcnp; 544 } */ *ap; 545 { 546 struct vnode *tdvp = ap->a_tdvp; 547 struct vnode *fvp = ap->a_fvp; 548 struct vnode *fdvp = ap->a_fdvp; 549 struct vnode *tvp = ap->a_tvp; 550 551 /* Check for cross-device rename. */ 552 if ((fvp->v_mount != tdvp->v_mount) || 553 (tvp && (fvp->v_mount != tvp->v_mount))) { 554 if (tdvp == tvp) 555 vrele(tdvp); 556 else 557 vput(tdvp); 558 if (tvp) 559 vput(tvp); 560 vrele(fdvp); 561 vrele(fvp); 562 return (EXDEV); 563 } 564 565 return (null_bypass((struct vop_generic_args *)ap)); 566 } 567 568 /* 569 * We need to process our own vnode lock and then clear the 570 * interlock flag as it applies only to our vnode, not the 571 * vnodes below us on the stack. 572 */ 573 static int 574 null_lock(ap) 575 struct vop_lock_args /* { 576 struct vnode *a_vp; 577 int a_flags; 578 struct proc *a_p; 579 } */ *ap; 580 { 581 struct vnode *vp = ap->a_vp; 582 int flags = ap->a_flags; 583 struct proc *p = ap->a_p; 584 struct null_node *np = VTONULL(vp); 585 struct vnode *lvp; 586 int error; 587 588 if (flags & LK_THISLAYER) { 589 if (vp->v_vnlock != NULL) { 590 /* lock is shared across layers */ 591 if (flags & LK_INTERLOCK) 592 simple_unlock(&vp->v_interlock); 593 return 0; 594 } 595 error = lockmgr(&np->null_lock, flags & ~LK_THISLAYER, 596 &vp->v_interlock, p); 597 return (error); 598 } 599 600 if (vp->v_vnlock != NULL) { 601 /* 602 * The lower level has exported a struct lock to us. Use 603 * it so that all vnodes in the stack lock and unlock 604 * simultaneously. Note: we don't DRAIN the lock as DRAIN 605 * decommissions the lock - just because our vnode is 606 * going away doesn't mean the struct lock below us is. 607 * LK_EXCLUSIVE is fine. 608 */ 609 if ((flags & LK_TYPE_MASK) == LK_DRAIN) { 610 NULLFSDEBUG("null_lock: avoiding LK_DRAIN\n"); 611 return(lockmgr(vp->v_vnlock, 612 (flags & ~LK_TYPE_MASK) | LK_EXCLUSIVE, 613 &vp->v_interlock, p)); 614 } 615 return(lockmgr(vp->v_vnlock, flags, &vp->v_interlock, p)); 616 } 617 /* 618 * To prevent race conditions involving doing a lookup 619 * on "..", we have to lock the lower node, then lock our 620 * node. Most of the time it won't matter that we lock our 621 * node (as any locking would need the lower one locked 622 * first). But we can LK_DRAIN the upper lock as a step 623 * towards decomissioning it. 624 */ 625 lvp = NULLVPTOLOWERVP(vp); 626 if (lvp == NULL) 627 return (lockmgr(&np->null_lock, flags, &vp->v_interlock, p)); 628 if (flags & LK_INTERLOCK) { 629 VI_UNLOCK(vp); 630 flags &= ~LK_INTERLOCK; 631 } 632 if ((flags & LK_TYPE_MASK) == LK_DRAIN) { 633 error = VOP_LOCK(lvp, 634 (flags & ~LK_TYPE_MASK) | LK_EXCLUSIVE, p); 635 } else 636 error = VOP_LOCK(lvp, flags, p); 637 if (error) 638 return (error); 639 error = lockmgr(&np->null_lock, flags, &vp->v_interlock, p); 640 if (error) 641 VOP_UNLOCK(lvp, 0, p); 642 return (error); 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 */ 650 static int 651 null_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 null_node *np = VTONULL(vp); 662 struct vnode *lvp; 663 664 if (vp->v_vnlock != NULL) { 665 if (flags & LK_THISLAYER) 666 return 0; /* the lock is shared across layers */ 667 flags &= ~LK_THISLAYER; 668 return (lockmgr(vp->v_vnlock, flags | LK_RELEASE, 669 &vp->v_interlock, p)); 670 } 671 lvp = NULLVPTOLOWERVP(vp); 672 if (lvp == NULL) 673 return (lockmgr(&np->null_lock, flags | LK_RELEASE, &vp->v_interlock, p)); 674 if ((flags & LK_THISLAYER) == 0) { 675 if (flags & LK_INTERLOCK) { 676 VI_UNLOCK(vp); 677 flags &= ~LK_INTERLOCK; 678 } 679 VOP_UNLOCK(lvp, flags, p); 680 } else 681 flags &= ~LK_THISLAYER; 682 ap->a_flags = flags; 683 return (lockmgr(&np->null_lock, flags | LK_RELEASE, &vp->v_interlock, p)); 684 } 685 686 static int 687 null_islocked(ap) 688 struct vop_islocked_args /* { 689 struct vnode *a_vp; 690 struct proc *a_p; 691 } */ *ap; 692 { 693 struct vnode *vp = ap->a_vp; 694 struct proc *p = ap->a_p; 695 696 if (vp->v_vnlock != NULL) 697 return (lockstatus(vp->v_vnlock, p)); 698 return (lockstatus(&VTONULL(vp)->null_lock, p)); 699 } 700 701 702 /* 703 * There is no way to tell that someone issued remove/rmdir operation 704 * on the underlying filesystem. For now we just have to release lowevrp 705 * as soon as possible. 706 */ 707 static int 708 null_inactive(ap) 709 struct vop_inactive_args /* { 710 struct vnode *a_vp; 711 struct proc *a_p; 712 } */ *ap; 713 { 714 struct vnode *vp = ap->a_vp; 715 struct proc *p = ap->a_p; 716 struct null_node *xp = VTONULL(vp); 717 struct vnode *lowervp = xp->null_lowervp; 718 719 lockmgr(&null_hashlock, LK_EXCLUSIVE, NULL, p); 720 LIST_REMOVE(xp, null_hash); 721 lockmgr(&null_hashlock, LK_RELEASE, NULL, p); 722 723 xp->null_lowervp = NULLVP; 724 if (vp->v_vnlock != NULL) { 725 vp->v_vnlock = &xp->null_lock; /* we no longer share the lock */ 726 } else 727 VOP_UNLOCK(vp, LK_THISLAYER, p); 728 729 vput(lowervp); 730 /* 731 * Now it is safe to drop references to the lower vnode. 732 * VOP_INACTIVE() will be called by vrele() if necessary. 733 */ 734 vrele (lowervp); 735 736 return (0); 737 } 738 739 /* 740 * We can free memory in null_inactive, but we do this 741 * here. (Possible to guard vp->v_data to point somewhere) 742 */ 743 static int 744 null_reclaim(ap) 745 struct vop_reclaim_args /* { 746 struct vnode *a_vp; 747 struct proc *a_p; 748 } */ *ap; 749 { 750 struct vnode *vp = ap->a_vp; 751 void *vdata = vp->v_data; 752 753 vp->v_data = NULL; 754 FREE(vdata, M_NULLFSNODE); 755 756 return (0); 757 } 758 759 static int 760 null_print(ap) 761 struct vop_print_args /* { 762 struct vnode *a_vp; 763 } */ *ap; 764 { 765 struct vnode *vp = ap->a_vp; 766 767 printf ("\ttag VT_NULLFS, vp=%p, lowervp=%p\n", vp, NULLVPTOLOWERVP(vp)); 768 if (vp->v_vnlock != NULL) { 769 printf("\tvnlock: "); 770 lockmgr_printinfo(vp->v_vnlock); 771 } else { 772 printf("\tnull_lock: "); 773 lockmgr_printinfo(&VTONULL(vp)->null_lock); 774 } 775 printf("\n"); 776 return (0); 777 } 778 779 /* 780 * Let an underlying filesystem do the work 781 */ 782 static int 783 null_createvobject(ap) 784 struct vop_createvobject_args /* { 785 struct vnode *vp; 786 struct ucred *cred; 787 struct proc *p; 788 } */ *ap; 789 { 790 struct vnode *vp = ap->a_vp; 791 struct vnode *lowervp = VTONULL(vp) ? NULLVPTOLOWERVP(vp) : NULL; 792 int error; 793 794 if (vp->v_type == VNON || lowervp == NULL) 795 return 0; 796 error = VOP_CREATEVOBJECT(lowervp, ap->a_cred, ap->a_p); 797 if (error) 798 return (error); 799 vp->v_flag |= VOBJBUF; 800 return (0); 801 } 802 803 /* 804 * We have nothing to destroy and this operation shouldn't be bypassed. 805 */ 806 static int 807 null_destroyvobject(ap) 808 struct vop_destroyvobject_args /* { 809 struct vnode *vp; 810 } */ *ap; 811 { 812 struct vnode *vp = ap->a_vp; 813 814 vp->v_flag &= ~VOBJBUF; 815 return (0); 816 } 817 818 static int 819 null_getvobject(ap) 820 struct vop_getvobject_args /* { 821 struct vnode *vp; 822 struct vm_object **objpp; 823 } */ *ap; 824 { 825 struct vnode *lvp = NULLVPTOLOWERVP(ap->a_vp); 826 827 if (lvp == NULL) 828 return EINVAL; 829 return (VOP_GETVOBJECT(lvp, ap->a_objpp)); 830 } 831 832 /* 833 * Global vfs data structures 834 */ 835 vop_t **null_vnodeop_p; 836 static struct vnodeopv_entry_desc null_vnodeop_entries[] = { 837 { &vop_default_desc, (vop_t *) null_bypass }, 838 { &vop_access_desc, (vop_t *) null_access }, 839 { &vop_createvobject_desc, (vop_t *) null_createvobject }, 840 { &vop_destroyvobject_desc, (vop_t *) null_destroyvobject }, 841 { &vop_getattr_desc, (vop_t *) null_getattr }, 842 { &vop_getvobject_desc, (vop_t *) null_getvobject }, 843 { &vop_inactive_desc, (vop_t *) null_inactive }, 844 { &vop_islocked_desc, (vop_t *) null_islocked }, 845 { &vop_lock_desc, (vop_t *) null_lock }, 846 { &vop_lookup_desc, (vop_t *) null_lookup }, 847 { &vop_open_desc, (vop_t *) null_open }, 848 { &vop_print_desc, (vop_t *) null_print }, 849 { &vop_reclaim_desc, (vop_t *) null_reclaim }, 850 { &vop_rename_desc, (vop_t *) null_rename }, 851 { &vop_setattr_desc, (vop_t *) null_setattr }, 852 { &vop_unlock_desc, (vop_t *) null_unlock }, 853 { NULL, NULL } 854 }; 855 static struct vnodeopv_desc null_vnodeop_opv_desc = 856 { &null_vnodeop_p, null_vnodeop_entries }; 857 858 VNODEOP_SET(null_vnodeop_opv_desc); 859