null_vnops.c revision 140783
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 * 4. Neither the name of the University nor the names of its contributors 17 * may be used to endorse or promote products derived from this software 18 * without specific prior written permission. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 30 * SUCH DAMAGE. 31 * 32 * @(#)null_vnops.c 8.6 (Berkeley) 5/27/95 33 * 34 * Ancestors: 35 * @(#)lofs_vnops.c 1.2 (Berkeley) 6/18/92 36 * ...and... 37 * @(#)null_vnodeops.c 1.20 92/07/07 UCLA Ficus project 38 * 39 * $FreeBSD: head/sys/fs/nullfs/null_vnops.c 140783 2005-01-25 00:42:16Z phk $ 40 */ 41 42/* 43 * Null Layer 44 * 45 * (See mount_nullfs(8) for more information.) 46 * 47 * The null layer duplicates a portion of the filesystem 48 * name space under a new name. In this respect, it is 49 * similar to the loopback filesystem. It differs from 50 * the loopback fs in two respects: it is implemented using 51 * a stackable layers techniques, and its "null-node"s stack above 52 * all lower-layer vnodes, not just over directory vnodes. 53 * 54 * The null layer has two purposes. First, it serves as a demonstration 55 * of layering by proving a layer which does nothing. (It actually 56 * does everything the loopback filesystem does, which is slightly 57 * more than nothing.) Second, the null layer can serve as a prototype 58 * layer. Since it provides all necessary layer framework, 59 * new filesystem layers can be created very easily be starting 60 * with a null layer. 61 * 62 * The remainder of this man page examines the null layer as a basis 63 * for constructing new layers. 64 * 65 * 66 * INSTANTIATING NEW NULL LAYERS 67 * 68 * New null layers are created with mount_nullfs(8). 69 * Mount_nullfs(8) takes two arguments, the pathname 70 * of the lower vfs (target-pn) and the pathname where the null 71 * layer will appear in the namespace (alias-pn). After 72 * the null layer is put into place, the contents 73 * of target-pn subtree will be aliased under alias-pn. 74 * 75 * 76 * OPERATION OF A NULL LAYER 77 * 78 * The null layer is the minimum filesystem layer, 79 * simply bypassing all possible operations to the lower layer 80 * for processing there. The majority of its activity centers 81 * on the bypass routine, through which nearly all vnode operations 82 * pass. 83 * 84 * The bypass routine accepts arbitrary vnode operations for 85 * handling by the lower layer. It begins by examing vnode 86 * operation arguments and replacing any null-nodes by their 87 * lower-layer equivlants. It then invokes the operation 88 * on the lower layer. Finally, it replaces the null-nodes 89 * in the arguments and, if a vnode is return by the operation, 90 * stacks a null-node on top of the returned vnode. 91 * 92 * Although bypass handles most operations, vop_getattr, vop_lock, 93 * vop_unlock, vop_inactive, vop_reclaim, and vop_print are not 94 * bypassed. Vop_getattr must change the fsid being returned. 95 * Vop_lock and vop_unlock must handle any locking for the 96 * current vnode as well as pass the lock request down. 97 * Vop_inactive and vop_reclaim are not bypassed so that 98 * they can handle freeing null-layer specific data. Vop_print 99 * is not bypassed to avoid excessive debugging information. 100 * Also, certain vnode operations change the locking state within 101 * the operation (create, mknod, remove, link, rename, mkdir, rmdir, 102 * and symlink). Ideally these operations should not change the 103 * lock state, but should be changed to let the caller of the 104 * function unlock them. Otherwise all intermediate vnode layers 105 * (such as union, umapfs, etc) must catch these functions to do 106 * the necessary locking at their layer. 107 * 108 * 109 * INSTANTIATING VNODE STACKS 110 * 111 * Mounting associates the null layer with a lower layer, 112 * effect stacking two VFSes. Vnode stacks are instead 113 * created on demand as files are accessed. 114 * 115 * The initial mount creates a single vnode stack for the 116 * root of the new null layer. All other vnode stacks 117 * are created as a result of vnode operations on 118 * this or other null vnode stacks. 119 * 120 * New vnode stacks come into existance as a result of 121 * an operation which returns a vnode. 122 * The bypass routine stacks a null-node above the new 123 * vnode before returning it to the caller. 124 * 125 * For example, imagine mounting a null layer with 126 * "mount_nullfs /usr/include /dev/layer/null". 127 * Changing directory to /dev/layer/null will assign 128 * the root null-node (which was created when the null layer was mounted). 129 * Now consider opening "sys". A vop_lookup would be 130 * done on the root null-node. This operation would bypass through 131 * to the lower layer which would return a vnode representing 132 * the UFS "sys". Null_bypass then builds a null-node 133 * aliasing the UFS "sys" and returns this to the caller. 134 * Later operations on the null-node "sys" will repeat this 135 * process when constructing other vnode stacks. 136 * 137 * 138 * CREATING OTHER FILE SYSTEM LAYERS 139 * 140 * One of the easiest ways to construct new filesystem layers is to make 141 * a copy of the null layer, rename all files and variables, and 142 * then begin modifing the copy. Sed can be used to easily rename 143 * all variables. 144 * 145 * The umap layer is an example of a layer descended from the 146 * null layer. 147 * 148 * 149 * INVOKING OPERATIONS ON LOWER LAYERS 150 * 151 * There are two techniques to invoke operations on a lower layer 152 * when the operation cannot be completely bypassed. Each method 153 * is appropriate in different situations. In both cases, 154 * it is the responsibility of the aliasing layer to make 155 * the operation arguments "correct" for the lower layer 156 * by mapping a vnode arguments to the lower layer. 157 * 158 * The first approach is to call the aliasing layer's bypass routine. 159 * This method is most suitable when you wish to invoke the operation 160 * currently being handled on the lower layer. It has the advantage 161 * that the bypass routine already must do argument mapping. 162 * An example of this is null_getattrs in the null layer. 163 * 164 * A second approach is to directly invoke vnode operations on 165 * the lower layer with the VOP_OPERATIONNAME interface. 166 * The advantage of this method is that it is easy to invoke 167 * arbitrary operations on the lower layer. The disadvantage 168 * is that vnode arguments must be manualy mapped. 169 * 170 */ 171 172#include <sys/param.h> 173#include <sys/systm.h> 174#include <sys/conf.h> 175#include <sys/kernel.h> 176#include <sys/lock.h> 177#include <sys/malloc.h> 178#include <sys/mount.h> 179#include <sys/mutex.h> 180#include <sys/namei.h> 181#include <sys/sysctl.h> 182#include <sys/vnode.h> 183 184#include <fs/nullfs/null.h> 185 186#include <vm/vm.h> 187#include <vm/vm_extern.h> 188#include <vm/vm_object.h> 189#include <vm/vnode_pager.h> 190 191static int null_bug_bypass = 0; /* for debugging: enables bypass printf'ing */ 192SYSCTL_INT(_debug, OID_AUTO, nullfs_bug_bypass, CTLFLAG_RW, 193 &null_bug_bypass, 0, ""); 194 195/* 196 * This is the 10-Apr-92 bypass routine. 197 * This version has been optimized for speed, throwing away some 198 * safety checks. It should still always work, but it's not as 199 * robust to programmer errors. 200 * 201 * In general, we map all vnodes going down and unmap them on the way back. 202 * As an exception to this, vnodes can be marked "unmapped" by setting 203 * the Nth bit in operation's vdesc_flags. 204 * 205 * Also, some BSD vnode operations have the side effect of vrele'ing 206 * their arguments. With stacking, the reference counts are held 207 * by the upper node, not the lower one, so we must handle these 208 * side-effects here. This is not of concern in Sun-derived systems 209 * since there are no such side-effects. 210 * 211 * This makes the following assumptions: 212 * - only one returned vpp 213 * - no INOUT vpp's (Sun's vop_open has one of these) 214 * - the vnode operation vector of the first vnode should be used 215 * to determine what implementation of the op should be invoked 216 * - all mapped vnodes are of our vnode-type (NEEDSWORK: 217 * problems on rmdir'ing mount points and renaming?) 218 */ 219int 220null_bypass(struct vop_generic_args *ap) 221{ 222 struct vnode **this_vp_p; 223 int error; 224 struct vnode *old_vps[VDESC_MAX_VPS]; 225 struct vnode **vps_p[VDESC_MAX_VPS]; 226 struct vnode ***vppp; 227 struct vnodeop_desc *descp = ap->a_desc; 228 int reles, i; 229 230 if (null_bug_bypass) 231 printf ("null_bypass: %s\n", descp->vdesc_name); 232 233#ifdef DIAGNOSTIC 234 /* 235 * We require at least one vp. 236 */ 237 if (descp->vdesc_vp_offsets == NULL || 238 descp->vdesc_vp_offsets[0] == VDESC_NO_OFFSET) 239 panic ("null_bypass: no vp's in map"); 240#endif 241 242 /* 243 * Map the vnodes going in. 244 * Later, we'll invoke the operation based on 245 * the first mapped vnode's operation vector. 246 */ 247 reles = descp->vdesc_flags; 248 for (i = 0; i < VDESC_MAX_VPS; reles >>= 1, i++) { 249 if (descp->vdesc_vp_offsets[i] == VDESC_NO_OFFSET) 250 break; /* bail out at end of list */ 251 vps_p[i] = this_vp_p = 252 VOPARG_OFFSETTO(struct vnode**,descp->vdesc_vp_offsets[i],ap); 253 /* 254 * We're not guaranteed that any but the first vnode 255 * are of our type. Check for and don't map any 256 * that aren't. (We must always map first vp or vclean fails.) 257 */ 258 if (i && (*this_vp_p == NULLVP || 259 (*this_vp_p)->v_op != &null_vnodeops)) { 260 old_vps[i] = NULLVP; 261 } else { 262 old_vps[i] = *this_vp_p; 263 *(vps_p[i]) = NULLVPTOLOWERVP(*this_vp_p); 264 /* 265 * XXX - Several operations have the side effect 266 * of vrele'ing their vp's. We must account for 267 * that. (This should go away in the future.) 268 */ 269 if (reles & VDESC_VP0_WILLRELE) 270 VREF(*this_vp_p); 271 } 272 273 } 274 275 /* 276 * Call the operation on the lower layer 277 * with the modified argument structure. 278 */ 279 if (vps_p[0] && *vps_p[0]) 280 error = VCALL(ap); 281 else { 282 printf("null_bypass: no map for %s\n", descp->vdesc_name); 283 error = EINVAL; 284 } 285 286 /* 287 * Maintain the illusion of call-by-value 288 * by restoring vnodes in the argument structure 289 * to their original value. 290 */ 291 reles = descp->vdesc_flags; 292 for (i = 0; i < VDESC_MAX_VPS; reles >>= 1, i++) { 293 if (descp->vdesc_vp_offsets[i] == VDESC_NO_OFFSET) 294 break; /* bail out at end of list */ 295 if (old_vps[i]) { 296 *(vps_p[i]) = old_vps[i]; 297#if 0 298 if (reles & VDESC_VP0_WILLUNLOCK) 299 VOP_UNLOCK(*(vps_p[i]), LK_THISLAYER, curthread); 300#endif 301 if (reles & VDESC_VP0_WILLRELE) 302 vrele(*(vps_p[i])); 303 } 304 } 305 306 /* 307 * Map the possible out-going vpp 308 * (Assumes that the lower layer always returns 309 * a VREF'ed vpp unless it gets an error.) 310 */ 311 if (descp->vdesc_vpp_offset != VDESC_NO_OFFSET && 312 !(descp->vdesc_flags & VDESC_NOMAP_VPP) && 313 !error) { 314 /* 315 * XXX - even though some ops have vpp returned vp's, 316 * several ops actually vrele this before returning. 317 * We must avoid these ops. 318 * (This should go away when these ops are regularized.) 319 */ 320 if (descp->vdesc_flags & VDESC_VPP_WILLRELE) 321 goto out; 322 vppp = VOPARG_OFFSETTO(struct vnode***, 323 descp->vdesc_vpp_offset,ap); 324 if (*vppp) 325 error = null_nodeget(old_vps[0]->v_mount, **vppp, *vppp); 326 } 327 328 out: 329 return (error); 330} 331 332static int 333null_close(struct vop_close_args *ap) 334{ 335 int retval; 336 struct vnode *vp; 337 338 vp = ap->a_vp; 339 retval = null_bypass(&ap->a_gen); 340 if (retval == 0) 341 vp->v_object = NULL; 342 return (retval); 343} 344 345/* 346 * We have to carry on the locking protocol on the null layer vnodes 347 * as we progress through the tree. We also have to enforce read-only 348 * if this layer is mounted read-only. 349 */ 350static int 351null_lookup(struct vop_lookup_args *ap) 352{ 353 struct componentname *cnp = ap->a_cnp; 354 struct vnode *dvp = ap->a_dvp; 355 struct thread *td = cnp->cn_thread; 356 int flags = cnp->cn_flags; 357 struct vnode *vp, *ldvp, *lvp; 358 int error; 359 360 if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) && 361 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) 362 return (EROFS); 363 /* 364 * Although it is possible to call null_bypass(), we'll do 365 * a direct call to reduce overhead 366 */ 367 ldvp = NULLVPTOLOWERVP(dvp); 368 vp = lvp = NULL; 369 error = VOP_LOOKUP(ldvp, &lvp, cnp); 370 if (error == EJUSTRETURN && (flags & ISLASTCN) && 371 (dvp->v_mount->mnt_flag & MNT_RDONLY) && 372 (cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME)) 373 error = EROFS; 374 375 /* 376 * Rely only on the PDIRUNLOCK flag which should be carefully 377 * tracked by underlying filesystem. 378 */ 379 if ((cnp->cn_flags & PDIRUNLOCK) && dvp->v_vnlock != ldvp->v_vnlock) 380 VOP_UNLOCK(dvp, LK_THISLAYER, td); 381 if ((error == 0 || error == EJUSTRETURN) && lvp != NULL) { 382 if (ldvp == lvp) { 383 *ap->a_vpp = dvp; 384 VREF(dvp); 385 vrele(lvp); 386 } else { 387 error = null_nodeget(dvp->v_mount, lvp, &vp); 388 if (error) { 389 /* XXX Cleanup needed... */ 390 panic("null_nodeget failed"); 391 } 392 *ap->a_vpp = vp; 393 } 394 } 395 return (error); 396} 397 398static int 399null_open(struct vop_open_args *ap) 400{ 401 int retval; 402 struct vnode *vp, *ldvp; 403 404 vp = ap->a_vp; 405 ldvp = NULLVPTOLOWERVP(vp); 406 retval = null_bypass(&ap->a_gen); 407 if (retval == 0) 408 vp->v_object = ldvp->v_object; 409 return (retval); 410} 411 412/* 413 * Setattr call. Disallow write attempts if the layer is mounted read-only. 414 */ 415static int 416null_setattr(struct vop_setattr_args *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 */ 455static int 456null_getattr(struct vop_getattr_args *ap) 457{ 458 int error; 459 460 if ((error = null_bypass((struct vop_generic_args *)ap)) != 0) 461 return (error); 462 463 ap->a_vap->va_fsid = ap->a_vp->v_mount->mnt_stat.f_fsid.val[0]; 464 return (0); 465} 466 467/* 468 * Handle to disallow write access if mounted read-only. 469 */ 470static int 471null_access(struct vop_access_args *ap) 472{ 473 struct vnode *vp = ap->a_vp; 474 mode_t mode = ap->a_mode; 475 476 /* 477 * Disallow write attempts on read-only layers; 478 * unless the file is a socket, fifo, or a block or 479 * character device resident on the filesystem. 480 */ 481 if (mode & VWRITE) { 482 switch (vp->v_type) { 483 case VDIR: 484 case VLNK: 485 case VREG: 486 if (vp->v_mount->mnt_flag & MNT_RDONLY) 487 return (EROFS); 488 break; 489 default: 490 break; 491 } 492 } 493 return (null_bypass((struct vop_generic_args *)ap)); 494} 495 496/* 497 * We handle this to eliminate null FS to lower FS 498 * file moving. Don't know why we don't allow this, 499 * possibly we should. 500 */ 501static int 502null_rename(struct vop_rename_args *ap) 503{ 504 struct vnode *tdvp = ap->a_tdvp; 505 struct vnode *fvp = ap->a_fvp; 506 struct vnode *fdvp = ap->a_fdvp; 507 struct vnode *tvp = ap->a_tvp; 508 509 /* Check for cross-device rename. */ 510 if ((fvp->v_mount != tdvp->v_mount) || 511 (tvp && (fvp->v_mount != tvp->v_mount))) { 512 if (tdvp == tvp) 513 vrele(tdvp); 514 else 515 vput(tdvp); 516 if (tvp) 517 vput(tvp); 518 vrele(fdvp); 519 vrele(fvp); 520 return (EXDEV); 521 } 522 523 return (null_bypass((struct vop_generic_args *)ap)); 524} 525 526/* 527 * We need to process our own vnode lock and then clear the 528 * interlock flag as it applies only to our vnode, not the 529 * vnodes below us on the stack. 530 */ 531static int 532null_lock(struct vop_lock_args *ap) 533{ 534 struct vnode *vp = ap->a_vp; 535 int flags = ap->a_flags; 536 struct thread *td = ap->a_td; 537 struct vnode *lvp; 538 int error; 539 struct null_node *nn; 540 541 if (flags & LK_THISLAYER) { 542 if (vp->v_vnlock != NULL) { 543 /* lock is shared across layers */ 544 if (flags & LK_INTERLOCK) 545 mtx_unlock(&vp->v_interlock); 546 return 0; 547 } 548 error = lockmgr(&vp->v_lock, flags & ~LK_THISLAYER, 549 &vp->v_interlock, td); 550 return (error); 551 } 552 553 if (vp->v_vnlock != NULL) { 554 /* 555 * The lower level has exported a struct lock to us. Use 556 * it so that all vnodes in the stack lock and unlock 557 * simultaneously. Note: we don't DRAIN the lock as DRAIN 558 * decommissions the lock - just because our vnode is 559 * going away doesn't mean the struct lock below us is. 560 * LK_EXCLUSIVE is fine. 561 */ 562 if ((flags & LK_INTERLOCK) == 0) { 563 VI_LOCK(vp); 564 flags |= LK_INTERLOCK; 565 } 566 nn = VTONULL(vp); 567 if ((flags & LK_TYPE_MASK) == LK_DRAIN) { 568 NULLFSDEBUG("null_lock: avoiding LK_DRAIN\n"); 569 /* 570 * Emulate lock draining by waiting for all other 571 * pending locks to complete. Afterwards the 572 * lockmgr call might block, but no other threads 573 * will attempt to use this nullfs vnode due to the 574 * VI_XLOCK flag. 575 */ 576 while (nn->null_pending_locks > 0) { 577 nn->null_drain_wakeup = 1; 578 msleep(&nn->null_pending_locks, 579 VI_MTX(vp), 580 PVFS, 581 "nuldr", 0); 582 } 583 error = lockmgr(vp->v_vnlock, 584 (flags & ~LK_TYPE_MASK) | LK_EXCLUSIVE, 585 VI_MTX(vp), td); 586 return error; 587 } 588 nn->null_pending_locks++; 589 error = lockmgr(vp->v_vnlock, flags, &vp->v_interlock, td); 590 VI_LOCK(vp); 591 /* 592 * If we're called from vrele then v_usecount can have been 0 593 * and another process might have initiated a recycle 594 * operation. When that happens, just back out. 595 */ 596 if (error == 0 && (vp->v_iflag & VI_XLOCK) != 0 && 597 td != vp->v_vxthread) { 598 lockmgr(vp->v_vnlock, 599 (flags & ~LK_TYPE_MASK) | LK_RELEASE, 600 VI_MTX(vp), td); 601 VI_LOCK(vp); 602 error = ENOENT; 603 } 604 nn->null_pending_locks--; 605 /* 606 * Wakeup the process draining the vnode after all 607 * pending lock attempts has been failed. 608 */ 609 if (nn->null_pending_locks == 0 && 610 nn->null_drain_wakeup != 0) { 611 nn->null_drain_wakeup = 0; 612 wakeup(&nn->null_pending_locks); 613 } 614 if (error == ENOENT && (vp->v_iflag & VI_XLOCK) != 0 && 615 vp->v_vxthread != curthread) { 616 vp->v_iflag |= VI_XWANT; 617 msleep(vp, VI_MTX(vp), PINOD, "nulbo", 0); 618 } 619 VI_UNLOCK(vp); 620 return error; 621 } else { 622 /* 623 * To prevent race conditions involving doing a lookup 624 * on "..", we have to lock the lower node, then lock our 625 * node. Most of the time it won't matter that we lock our 626 * node (as any locking would need the lower one locked 627 * first). But we can LK_DRAIN the upper lock as a step 628 * towards decomissioning it. 629 */ 630 lvp = NULLVPTOLOWERVP(vp); 631 if (lvp == NULL) 632 return (lockmgr(&vp->v_lock, flags, &vp->v_interlock, td)); 633 if (flags & LK_INTERLOCK) { 634 mtx_unlock(&vp->v_interlock); 635 flags &= ~LK_INTERLOCK; 636 } 637 if ((flags & LK_TYPE_MASK) == LK_DRAIN) { 638 error = VOP_LOCK(lvp, 639 (flags & ~LK_TYPE_MASK) | LK_EXCLUSIVE, td); 640 } else 641 error = VOP_LOCK(lvp, flags, td); 642 if (error) 643 return (error); 644 error = lockmgr(&vp->v_lock, flags, &vp->v_interlock, td); 645 if (error) 646 VOP_UNLOCK(lvp, 0, td); 647 return (error); 648 } 649} 650 651/* 652 * We need to process our own vnode unlock and then clear the 653 * interlock flag as it applies only to our vnode, not the 654 * vnodes below us on the stack. 655 */ 656static int 657null_unlock(struct vop_unlock_args *ap) 658{ 659 struct vnode *vp = ap->a_vp; 660 int flags = ap->a_flags; 661 struct thread *td = ap->a_td; 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, td)); 670 } 671 lvp = NULLVPTOLOWERVP(vp); 672 if (lvp == NULL) 673 return (lockmgr(&vp->v_lock, flags | LK_RELEASE, &vp->v_interlock, td)); 674 if ((flags & LK_THISLAYER) == 0) { 675 if (flags & LK_INTERLOCK) { 676 mtx_unlock(&vp->v_interlock); 677 flags &= ~LK_INTERLOCK; 678 } 679 VOP_UNLOCK(lvp, flags & ~LK_INTERLOCK, td); 680 } else 681 flags &= ~LK_THISLAYER; 682 return (lockmgr(&vp->v_lock, flags | LK_RELEASE, &vp->v_interlock, td)); 683} 684 685static int 686null_islocked(struct vop_islocked_args *ap) 687{ 688 struct vnode *vp = ap->a_vp; 689 struct thread *td = ap->a_td; 690 691 if (vp->v_vnlock != NULL) 692 return (lockstatus(vp->v_vnlock, td)); 693 return (lockstatus(&vp->v_lock, td)); 694} 695 696/* 697 * There is no way to tell that someone issued remove/rmdir operation 698 * on the underlying filesystem. For now we just have to release lowevrp 699 * as soon as possible. 700 * 701 * Note, we can't release any resources nor remove vnode from hash before 702 * appropriate VXLOCK stuff is is done because other process can find this 703 * vnode in hash during inactivation and may be sitting in vget() and waiting 704 * for null_inactive to unlock vnode. Thus we will do all those in VOP_RECLAIM. 705 */ 706static int 707null_inactive(struct vop_inactive_args *ap) 708{ 709 struct vnode *vp = ap->a_vp; 710 struct thread *td = ap->a_td; 711 712 VOP_UNLOCK(vp, 0, td); 713 714 /* 715 * If this is the last reference, then free up the vnode 716 * so as not to tie up the lower vnodes. 717 */ 718 vrecycle(vp, NULL, td); 719 720 return (0); 721} 722 723/* 724 * Now, the VXLOCK is in force and we're free to destroy the null vnode. 725 */ 726static int 727null_reclaim(struct vop_reclaim_args *ap) 728{ 729 struct vnode *vp = ap->a_vp; 730 struct null_node *xp = VTONULL(vp); 731 struct vnode *lowervp = xp->null_lowervp; 732 733 if (lowervp) { 734 null_hashrem(xp); 735 736 vrele(lowervp); 737 vrele(lowervp); 738 } 739 740 vp->v_data = NULL; 741 vp->v_vnlock = &vp->v_lock; 742 FREE(xp, M_NULLFSNODE); 743 744 return (0); 745} 746 747static int 748null_print(struct vop_print_args *ap) 749{ 750 struct vnode *vp = ap->a_vp; 751 printf("\tvp=%p, lowervp=%p\n", vp, NULLVPTOLOWERVP(vp)); 752 return (0); 753} 754 755/* 756 * We have nothing to destroy and this operation shouldn't be bypassed. 757 */ 758static int 759null_destroyvobject(struct vop_destroyvobject_args *ap) 760{ 761 struct vnode *vp = ap->a_vp; 762 763 vp->v_object = NULL; 764 return (0); 765} 766 767/* 768 * Global vfs data structures 769 */ 770struct vop_vector null_vnodeops = { 771 .vop_bypass = null_bypass, 772 773 .vop_access = null_access, 774 .vop_bmap = VOP_EOPNOTSUPP, 775 .vop_close = null_close, 776 .vop_destroyvobject = null_destroyvobject, 777 .vop_getattr = null_getattr, 778 .vop_getwritemount = vop_stdgetwritemount, 779 .vop_inactive = null_inactive, 780 .vop_islocked = null_islocked, 781 .vop_lock = null_lock, 782 .vop_lookup = null_lookup, 783 .vop_open = null_open, 784 .vop_print = null_print, 785 .vop_reclaim = null_reclaim, 786 .vop_rename = null_rename, 787 .vop_setattr = null_setattr, 788 .vop_strategy = VOP_EOPNOTSUPP, 789 .vop_unlock = null_unlock, 790}; 791