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 1.15 2004/08/28 21:32:28 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/proc.h> 185 #include <sys/namei.h> 186 #include <sys/malloc.h> 187 #include <sys/buf.h> 188 #include "null.h" 189 190 static int null_bug_bypass = 0; /* for debugging: enables bypass printf'ing */ 191 SYSCTL_INT(_debug, OID_AUTO, nullfs_bug_bypass, CTLFLAG_RW, 192 &null_bug_bypass, 0, ""); 193 194 static int null_access(struct vop_access_args *ap); 195 static int null_createvobject(struct vop_createvobject_args *ap); 196 static int null_destroyvobject(struct vop_destroyvobject_args *ap); 197 static int null_getattr(struct vop_getattr_args *ap); 198 static int null_getvobject(struct vop_getvobject_args *ap); 199 static int null_inactive(struct vop_inactive_args *ap); 200 static int null_islocked(struct vop_islocked_args *ap); 201 static int null_lock(struct vop_lock_args *ap); 202 static int null_lookup(struct vop_lookup_args *ap); 203 static int null_open(struct vop_open_args *ap); 204 static int null_print(struct vop_print_args *ap); 205 static int null_reclaim(struct vop_reclaim_args *ap); 206 static int null_rename(struct vop_rename_args *ap); 207 static int null_setattr(struct vop_setattr_args *ap); 208 static int null_unlock(struct vop_unlock_args *ap); 209 210 /* 211 * This is the 10-Apr-92 bypass routine. 212 * This version has been optimized for speed, throwing away some 213 * safety checks. It should still always work, but it's not as 214 * robust to programmer errors. 215 * 216 * In general, we map all vnodes going down and unmap them on the way back. 217 * As an exception to this, vnodes can be marked "unmapped" by setting 218 * the Nth bit in operation's vdesc_flags. 219 * 220 * Also, some BSD vnode operations have the side effect of vrele'ing 221 * their arguments. With stacking, the reference counts are held 222 * by the upper node, not the lower one, so we must handle these 223 * side-effects here. This is not of concern in Sun-derived systems 224 * since there are no such side-effects. 225 * 226 * This makes the following assumptions: 227 * - only one returned vpp 228 * - no INOUT vpp's (Sun's vop_open has one of these) 229 * - the vnode operation vector of the first vnode should be used 230 * to determine what implementation of the op should be invoked 231 * - all mapped vnodes are of our vnode-type (NEEDSWORK: 232 * problems on rmdir'ing mount points and renaming?) 233 * 234 * null_bypass(struct vnodeop_desc *a_desc, ...) 235 */ 236 int 237 null_bypass(struct vop_generic_args *ap) 238 { 239 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, j; 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 */ 262 reles = descp->vdesc_flags; 263 for (i = 0; i < VDESC_MAX_VPS; ++i) { 264 if (descp->vdesc_vp_offsets[i] == VDESC_NO_OFFSET) 265 break; /* bail out at end of list */ 266 vps_p[i] = this_vp_p = 267 VOPARG_OFFSETTO(struct vnode**,descp->vdesc_vp_offsets[i],ap); 268 /* 269 * We're not guaranteed that any but the first vnode 270 * are of our type. Check for and don't map any 271 * that aren't. (We must always map first vp or vclean fails.) 272 */ 273 if (i && (*this_vp_p == NULLVP || 274 (*this_vp_p)->v_tag != VT_NULL)) { 275 old_vps[i] = NULLVP; 276 } else { 277 old_vps[i] = *this_vp_p; 278 *this_vp_p = NULLVPTOLOWERVP(*this_vp_p); 279 /* 280 * Several operations have the side effect of vrele'ing 281 * their vp's. We must account for that in the lower 282 * vp we pass down. 283 */ 284 if (reles & (VDESC_VP0_WILLRELE << i)) 285 vref(*this_vp_p); 286 } 287 288 } 289 290 /* 291 * Call the operation on the lower layer with the modified 292 * argument structure. We have to adjust a_fm to point to the 293 * lower vp's vop_ops structure. 294 */ 295 if (vps_p[0] && *vps_p[0]) { 296 ap->a_ops = (*(vps_p[0]))->v_ops; 297 error = vop_vnoperate_ap(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 by restoring vnodes in the 305 * argument structure to their original value. 306 */ 307 reles = descp->vdesc_flags; 308 for (i = 0; i < VDESC_MAX_VPS; ++i) { 309 if (descp->vdesc_vp_offsets[i] == VDESC_NO_OFFSET) 310 break; /* bail out at end of list */ 311 if (old_vps[i]) { 312 *(vps_p[i]) = old_vps[i]; 313 314 /* 315 * Since we operated on the lowervp's instead of the 316 * null node vp's, we have to adjust the null node 317 * vp's based on what the VOP did to the lower vp. 318 * 319 * Note: the unlock case only occurs with rename. 320 * tdvp and tvp are both locked on call and must be 321 * unlocked on return. 322 * 323 * Unlock semantics indicate that if two locked vp's 324 * are passed and they are the same vp, they are only 325 * actually locked once. 326 */ 327 if (reles & (VDESC_VP0_WILLUNLOCK << i)) { 328 VOP_UNLOCK(old_vps[i], NULL, 329 LK_THISLAYER, curthread); 330 for (j = i + 1; j < VDESC_MAX_VPS; ++j) { 331 if (descp->vdesc_vp_offsets[j] == VDESC_NO_OFFSET) 332 break; 333 if (old_vps[i] == old_vps[j]) { 334 reles &= ~(1 << (VDESC_VP0_WILLUNLOCK << j)); 335 } 336 } 337 } 338 339 if (reles & (VDESC_VP0_WILLRELE << i)) 340 vrele(old_vps[i]); 341 } 342 } 343 344 /* 345 * Map the possible out-going vpp 346 * (Assumes that the lower layer always returns 347 * a vref'ed vpp unless it gets an error.) 348 */ 349 if (descp->vdesc_vpp_offset != VDESC_NO_OFFSET && 350 !(descp->vdesc_flags & VDESC_NOMAP_VPP) && 351 !error) { 352 /* 353 * XXX - even though some ops have vpp returned vp's, 354 * several ops actually vrele this before returning. 355 * We must avoid these ops. 356 * (This should go away when these ops are regularized.) 357 */ 358 if (descp->vdesc_flags & VDESC_VPP_WILLRELE) 359 goto out; 360 vppp = VOPARG_OFFSETTO(struct vnode***, 361 descp->vdesc_vpp_offset,ap); 362 if (*vppp) 363 error = null_node_create(old_vps[0]->v_mount, **vppp, *vppp); 364 } 365 366 out: 367 return (error); 368 } 369 370 /* 371 * We have to carry on the locking protocol on the null layer vnodes 372 * as we progress through the tree. We also have to enforce read-only 373 * if this layer is mounted read-only. 374 * 375 * null_lookup(struct vnode *a_dvp, struct vnode **a_vpp, 376 * struct componentname *a_cnp) 377 */ 378 static int 379 null_lookup(struct vop_lookup_args *ap) 380 { 381 struct componentname *cnp = ap->a_cnp; 382 struct vnode *dvp = ap->a_dvp; 383 struct thread *td = cnp->cn_td; 384 int flags = cnp->cn_flags; 385 struct vnode *vp, *ldvp, *lvp; 386 int error; 387 388 if ((flags & CNP_ISLASTCN) && 389 (dvp->v_mount->mnt_flag & MNT_RDONLY) && 390 (cnp->cn_nameiop == NAMEI_DELETE || 391 cnp->cn_nameiop == NAMEI_RENAME)) { 392 return (EROFS); 393 } 394 ldvp = NULLVPTOLOWERVP(dvp); 395 396 /* 397 * If we are doing a ".." lookup we must release the lock on dvp 398 * now, before we run a lookup in the underlying fs, or we may 399 * deadlock. If we do this we must protect ldvp by ref'ing it. 400 */ 401 if (flags & CNP_ISDOTDOT) { 402 vref(ldvp); 403 VOP_UNLOCK(dvp, NULL, LK_THISLAYER, td); 404 } 405 406 /* 407 * Due to the non-deterministic nature of the handling of the 408 * parent directory lock by lookup, we cannot call null_bypass() 409 * here. We must make a direct call. It's faster to do a direct 410 * call, anyway. 411 */ 412 vp = lvp = NULL; 413 error = VOP_LOOKUP(ldvp, NCPNULL, &lvp, NCPPNULL, cnp); 414 if (error == EJUSTRETURN && (flags & CNP_ISLASTCN) && 415 (dvp->v_mount->mnt_flag & MNT_RDONLY) && 416 (cnp->cn_nameiop == NAMEI_CREATE || 417 cnp->cn_nameiop == NAMEI_RENAME)) { 418 error = EROFS; 419 } 420 421 if ((error == 0 || error == EJUSTRETURN) && lvp != NULL) { 422 if (ldvp == lvp) { 423 *ap->a_vpp = dvp; 424 vref(dvp); 425 vrele(lvp); 426 } else { 427 error = null_node_create(dvp->v_mount, lvp, &vp); 428 if (error == 0) 429 *ap->a_vpp = vp; 430 } 431 } 432 433 /* 434 * The underlying fs will set PDIRUNLOCK if it unlocked the parent 435 * directory, which means we have to follow suit in the nullfs layer. 436 * Note that the parent directory may have already been unlocked due 437 * to the ".." case. Note that use of cnp->cn_flags instead of flags. 438 */ 439 if (flags & CNP_ISDOTDOT) { 440 if ((cnp->cn_flags & CNP_PDIRUNLOCK) == 0) 441 VOP_LOCK(dvp, NULL, LK_THISLAYER | LK_EXCLUSIVE, td); 442 vrele(ldvp); 443 } else if (cnp->cn_flags & CNP_PDIRUNLOCK) { 444 VOP_UNLOCK(dvp, NULL, LK_THISLAYER, td); 445 } 446 return (error); 447 } 448 449 /* 450 * Setattr call. Disallow write attempts if the layer is mounted read-only. 451 * 452 * null_setattr(struct vnodeop_desc *a_desc, struct vnode *a_vp, 453 * struct vattr *a_vap, struct ucred *a_cred, 454 * struct thread *a_td) 455 */ 456 int 457 null_setattr(struct vop_setattr_args *ap) 458 { 459 struct vnode *vp = ap->a_vp; 460 struct vattr *vap = ap->a_vap; 461 462 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL || 463 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL || 464 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) && 465 (vp->v_mount->mnt_flag & MNT_RDONLY)) 466 return (EROFS); 467 if (vap->va_size != VNOVAL) { 468 switch (vp->v_type) { 469 case VDIR: 470 return (EISDIR); 471 case VCHR: 472 case VBLK: 473 case VSOCK: 474 case VFIFO: 475 if (vap->va_flags != VNOVAL) 476 return (EOPNOTSUPP); 477 return (0); 478 case VREG: 479 case VLNK: 480 default: 481 /* 482 * Disallow write attempts if the filesystem is 483 * mounted read-only. 484 */ 485 if (vp->v_mount->mnt_flag & MNT_RDONLY) 486 return (EROFS); 487 } 488 } 489 490 return (null_bypass(&ap->a_head)); 491 } 492 493 /* 494 * We handle getattr only to change the fsid. 495 * 496 * null_getattr(struct vnode *a_vp, struct vattr *a_vap, struct ucred *a_cred, 497 * struct thread *a_td) 498 */ 499 static int 500 null_getattr(struct vop_getattr_args *ap) 501 { 502 int error; 503 504 if ((error = null_bypass(&ap->a_head)) != 0) 505 return (error); 506 507 ap->a_vap->va_fsid = ap->a_vp->v_mount->mnt_stat.f_fsid.val[0]; 508 return (0); 509 } 510 511 /* 512 * Handle to disallow write access if mounted read-only. 513 * 514 * null_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred, 515 * struct thread *a_td) 516 */ 517 static int 518 null_access(struct vop_access_args *ap) 519 { 520 struct vnode *vp = ap->a_vp; 521 mode_t mode = ap->a_mode; 522 523 /* 524 * Disallow write attempts on read-only layers; 525 * unless the file is a socket, fifo, or a block or 526 * character device resident on the file system. 527 */ 528 if (mode & VWRITE) { 529 switch (vp->v_type) { 530 case VDIR: 531 case VLNK: 532 case VREG: 533 if (vp->v_mount->mnt_flag & MNT_RDONLY) 534 return (EROFS); 535 break; 536 default: 537 break; 538 } 539 } 540 return (null_bypass(&ap->a_head)); 541 } 542 543 /* 544 * We must handle open to be able to catch MNT_NODEV and friends. 545 * 546 * null_open(struct vnode *a_vp, int a_mode, struct ucred *a_cred, 547 * struct thread *a_td) 548 */ 549 static int 550 null_open(struct vop_open_args *ap) 551 { 552 struct vnode *vp = ap->a_vp; 553 struct vnode *lvp = NULLVPTOLOWERVP(ap->a_vp); 554 555 if ((vp->v_mount->mnt_flag & MNT_NODEV) && 556 (lvp->v_type == VBLK || lvp->v_type == VCHR)) 557 return ENXIO; 558 559 return (null_bypass(&ap->a_head)); 560 } 561 562 /* 563 * We handle this to eliminate null FS to lower FS 564 * file moving. Don't know why we don't allow this, 565 * possibly we should. 566 * 567 * null_rename(struct vnode *a_fdvp, struct vnode *a_fvp, 568 * struct componentname *a_fcnp, struct vnode *a_tdvp, 569 * struct vnode *a_tvp, struct componentname *a_tcnp) 570 */ 571 static int 572 null_rename(struct vop_rename_args *ap) 573 { 574 struct vnode *tdvp = ap->a_tdvp; 575 struct vnode *fvp = ap->a_fvp; 576 struct vnode *fdvp = ap->a_fdvp; 577 struct vnode *tvp = ap->a_tvp; 578 579 /* Check for cross-device rename. */ 580 if ((fvp->v_mount != tdvp->v_mount) || 581 (tvp && (fvp->v_mount != tvp->v_mount))) { 582 if (tdvp == tvp) 583 vrele(tdvp); 584 else 585 vput(tdvp); 586 if (tvp) 587 vput(tvp); 588 vrele(fdvp); 589 vrele(fvp); 590 return (EXDEV); 591 } 592 593 return (null_bypass(&ap->a_head)); 594 } 595 596 /* 597 * A special flag, LK_THISLAYER, causes the locking function to operate 598 * ONLY on the nullfs layer. Otherwise we are responsible for locking not 599 * only our layer, but the lower layer as well. 600 * 601 * null_lock(struct vnode *a_vp, lwkt_tokref_t a_vlock, int a_flags, 602 * struct thread *a_td) 603 */ 604 static int 605 null_lock(struct vop_lock_args *ap) 606 { 607 struct vnode *vp = ap->a_vp; 608 int flags = ap->a_flags; 609 struct null_node *np = VTONULL(vp); 610 struct vnode *lvp; 611 int error; 612 613 /* 614 * Lock the nullfs layer first, disposing of the interlock in the 615 * process. 616 */ 617 error = lockmgr(&vp->v_lock, flags & ~LK_THISLAYER, 618 ap->a_vlock, ap->a_td); 619 flags &= ~LK_INTERLOCK; 620 621 /* 622 * If locking only the nullfs layer, or if there is no lower layer, 623 * or if an error occured while attempting to lock the nullfs layer, 624 * we are done. 625 * 626 * np can be NULL is the vnode is being recycled from a previous 627 * hash collision. 628 */ 629 if ((flags & LK_THISLAYER) || np == NULL || 630 np->null_lowervp == NULL || error) { 631 return (error); 632 } 633 634 /* 635 * Lock the underlying vnode. If we are draining we should not drain 636 * the underlying vnode, since it is not being destroyed, but we do 637 * lock it exclusively in that case. Note that any interlocks have 638 * already been disposed of above. 639 */ 640 lvp = np->null_lowervp; 641 if ((flags & LK_TYPE_MASK) == LK_DRAIN) { 642 NULLFSDEBUG("null_lock: avoiding LK_DRAIN\n"); 643 error = vn_lock(lvp, NULL, 644 (flags & ~LK_TYPE_MASK) | LK_EXCLUSIVE, 645 ap->a_td); 646 } else { 647 error = vn_lock(lvp, NULL, flags, ap->a_td); 648 } 649 650 /* 651 * If an error occured we have to undo our nullfs lock, then return 652 * the original error. 653 */ 654 if (error) 655 lockmgr(&vp->v_lock, LK_RELEASE, NULL, ap->a_td); 656 return(error); 657 } 658 659 /* 660 * A special flag, LK_THISLAYER, causes the unlocking function to operate 661 * ONLY on the nullfs layer. Otherwise we are responsible for unlocking not 662 * only our layer, but the lower layer as well. 663 * 664 * null_unlock(struct vnode *a_vp, lwkt_tokref_t a_vlock, int a_flags, 665 * struct thread *a_td) 666 */ 667 static int 668 null_unlock(struct vop_unlock_args *ap) 669 { 670 struct vnode *vp = ap->a_vp; 671 int flags = ap->a_flags; 672 struct null_node *np = VTONULL(vp); 673 struct vnode *lvp; 674 int error; 675 676 /* 677 * nullfs layer only 678 */ 679 if (flags & LK_THISLAYER) { 680 error = lockmgr(&vp->v_lock, 681 (flags & ~LK_THISLAYER) | LK_RELEASE, 682 ap->a_vlock, 683 ap->a_td); 684 return (error); 685 } 686 687 /* 688 * If there is no underlying vnode the lock operation occurs at 689 * the nullfs layer. np can be NULL is the vnode is being recycled 690 * from a previous hash collision. 691 */ 692 if (np == NULL || (lvp = np->null_lowervp) == NULL) { 693 error = lockmgr(&vp->v_lock, flags | LK_RELEASE, 694 ap->a_vlock, ap->a_td); 695 return(error); 696 } 697 698 /* 699 * Unlock the lower layer first, then our nullfs layer. 700 */ 701 VOP_UNLOCK(lvp, NULL, flags & ~LK_INTERLOCK, ap->a_td); 702 error = lockmgr(&vp->v_lock, flags | LK_RELEASE, 703 ap->a_vlock, ap->a_td); 704 return (error); 705 } 706 707 /* 708 * null_islocked(struct vnode *a_vp, struct thread *a_td) 709 * 710 * If a lower layer exists return the lock status of the lower layer, 711 * otherwise return the lock status of our nullfs layer. 712 */ 713 static int 714 null_islocked(struct vop_islocked_args *ap) 715 { 716 struct vnode *vp = ap->a_vp; 717 struct vnode *lvp; 718 struct null_node *np = VTONULL(vp); 719 int error; 720 721 lvp = np->null_lowervp; 722 if (lvp == NULL) 723 error = lockstatus(&vp->v_lock, ap->a_td); 724 else 725 error = VOP_ISLOCKED(lvp, ap->a_td); 726 return (error); 727 } 728 729 730 /* 731 * There is no way to tell that someone issued remove/rmdir operation 732 * on the underlying filesystem. For now we just have to release lowevrp 733 * as soon as possible. 734 * 735 * null_inactive(struct vnode *a_vp, struct thread *a_td) 736 */ 737 static int 738 null_inactive(struct vop_inactive_args *ap) 739 { 740 struct vnode *vp = ap->a_vp; 741 struct null_node *np = VTONULL(vp); 742 struct vnode *lowervp; 743 744 /* 745 * Clean out the lowervp. Due to the drain the lower vp has 746 * been exclusively locked. We undo that, then release our 747 * null_lowervp reference to lowervp. The lower vnode's 748 * inactive routine may or may not be called when we do the 749 * final vrele(). 750 */ 751 if (np) { 752 null_node_rem(np); 753 lowervp = np->null_lowervp; 754 np->null_lowervp = NULLVP; 755 if (lowervp) { 756 vput(lowervp); 757 vrele (lowervp); 758 } 759 } 760 761 /* 762 * Now it is safe to release our nullfs layer vnode. 763 */ 764 VOP_UNLOCK(vp, NULL, LK_THISLAYER, ap->a_td); 765 return (0); 766 } 767 768 /* 769 * We can free memory in null_inactive, but we do this 770 * here. (Possible to guard vp->v_data to point somewhere) 771 * 772 * null_reclaim(struct vnode *a_vp, struct thread *a_td) 773 */ 774 static int 775 null_reclaim(struct vop_reclaim_args *ap) 776 { 777 struct vnode *vp = ap->a_vp; 778 struct null_node *np; 779 780 np = VTONULL(vp); 781 vp->v_data = NULL; 782 if (np) 783 free(np, M_NULLFSNODE); 784 return (0); 785 } 786 787 /* 788 * null_print(struct vnode *a_vp) 789 */ 790 static int 791 null_print(struct vop_print_args *ap) 792 { 793 struct vnode *vp = ap->a_vp; 794 struct null_node *np = VTONULL(vp); 795 796 if (np == NULL) { 797 printf ("\ttag VT_NULLFS, vp=%p, NULL v_data!\n", vp); 798 return(0); 799 } 800 printf ("\ttag VT_NULLFS, vp=%p, lowervp=%p\n", vp, np->null_lowervp); 801 if (np->null_lowervp != NULL) { 802 printf("\tlowervp_lock: "); 803 lockmgr_printinfo(&np->null_lowervp->v_lock); 804 } else { 805 printf("\tnull_lock: "); 806 lockmgr_printinfo(&vp->v_lock); 807 } 808 printf("\n"); 809 return (0); 810 } 811 812 /* 813 * Let an underlying filesystem do the work 814 * 815 * null_createvobject(struct vnode *vp, struct ucred *cred, struct proc *p) 816 */ 817 static int 818 null_createvobject(struct vop_createvobject_args *ap) 819 { 820 struct vnode *vp = ap->a_vp; 821 struct vnode *lowervp = VTONULL(vp) ? NULLVPTOLOWERVP(vp) : NULL; 822 int error; 823 824 if (vp->v_type == VNON || lowervp == NULL) 825 return 0; 826 error = VOP_CREATEVOBJECT(lowervp, ap->a_td); 827 if (error) 828 return (error); 829 vp->v_flag |= VOBJBUF; 830 return (0); 831 } 832 833 /* 834 * We have nothing to destroy and this operation shouldn't be bypassed. 835 * 836 * null_destroyvobject(struct vnode *vp) 837 */ 838 static int 839 null_destroyvobject(struct vop_destroyvobject_args *ap) 840 { 841 struct vnode *vp = ap->a_vp; 842 843 vp->v_flag &= ~VOBJBUF; 844 return (0); 845 } 846 847 /* 848 * null_getvobject(struct vnode *vp, struct vm_object **objpp) 849 * 850 * Note that this can be called when a vnode is being recycled, and 851 * v_data may be NULL in that case if nullfs had to recycle a vnode 852 * due to a null_node collision. 853 */ 854 static int 855 null_getvobject(struct vop_getvobject_args *ap) 856 { 857 struct vnode *lvp; 858 859 if (ap->a_vp->v_data == NULL) 860 return EINVAL; 861 862 lvp = NULLVPTOLOWERVP(ap->a_vp); 863 if (lvp == NULL) 864 return EINVAL; 865 return (VOP_GETVOBJECT(lvp, ap->a_objpp)); 866 } 867 868 /* 869 * Global vfs data structures 870 */ 871 struct vnodeopv_entry_desc null_vnodeop_entries[] = { 872 { &vop_default_desc, (void *) null_bypass }, 873 { &vop_access_desc, (void *) null_access }, 874 { &vop_createvobject_desc, (void *) null_createvobject }, 875 { &vop_destroyvobject_desc, (void *) null_destroyvobject }, 876 { &vop_getattr_desc, (void *) null_getattr }, 877 { &vop_getvobject_desc, (void *) null_getvobject }, 878 { &vop_inactive_desc, (void *) null_inactive }, 879 { &vop_islocked_desc, (void *) null_islocked }, 880 { &vop_lock_desc, (void *) null_lock }, 881 { &vop_lookup_desc, (void *) null_lookup }, 882 { &vop_open_desc, (void *) null_open }, 883 { &vop_print_desc, (void *) null_print }, 884 { &vop_reclaim_desc, (void *) null_reclaim }, 885 { &vop_rename_desc, (void *) null_rename }, 886 { &vop_setattr_desc, (void *) null_setattr }, 887 { &vop_unlock_desc, (void *) null_unlock }, 888 { NULL, NULL } 889 }; 890 891