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 1.22 2005/02/15 08:32:18 joerg 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_nresolve(struct vop_nresolve_args *ap); 195 static int null_ncreate(struct vop_ncreate_args *ap); 196 static int null_nmkdir(struct vop_nmkdir_args *ap); 197 static int null_nremove(struct vop_nremove_args *ap); 198 static int null_nrmdir(struct vop_nrmdir_args *ap); 199 static int null_nrename(struct vop_nrename_args *ap); 200 201 static int null_revoke(struct vop_revoke_args *ap); 202 static int null_access(struct vop_access_args *ap); 203 static int null_createvobject(struct vop_createvobject_args *ap); 204 static int null_destroyvobject(struct vop_destroyvobject_args *ap); 205 static int null_getattr(struct vop_getattr_args *ap); 206 static int null_getvobject(struct vop_getvobject_args *ap); 207 static int null_inactive(struct vop_inactive_args *ap); 208 static int null_islocked(struct vop_islocked_args *ap); 209 static int null_lock(struct vop_lock_args *ap); 210 static int null_lookup(struct vop_lookup_args *ap); 211 static int null_open(struct vop_open_args *ap); 212 static int null_print(struct vop_print_args *ap); 213 static int null_reclaim(struct vop_reclaim_args *ap); 214 static int null_rename(struct vop_rename_args *ap); 215 static int null_setattr(struct vop_setattr_args *ap); 216 static int null_unlock(struct vop_unlock_args *ap); 217 218 /* 219 * This is the 10-Apr-92 bypass routine. 220 * This version has been optimized for speed, throwing away some 221 * safety checks. It should still always work, but it's not as 222 * robust to programmer errors. 223 * 224 * In general, we map all vnodes going down and unmap them on the way back. 225 * As an exception to this, vnodes can be marked "unmapped" by setting 226 * the Nth bit in operation's vdesc_flags. 227 * 228 * Also, some BSD vnode operations have the side effect of vrele'ing 229 * their arguments. With stacking, the reference counts are held 230 * by the upper node, not the lower one, so we must handle these 231 * side-effects here. This is not of concern in Sun-derived systems 232 * since there are no such side-effects. 233 * 234 * This makes the following assumptions: 235 * - only one returned vpp 236 * - no INOUT vpp's (Sun's vop_open has one of these) 237 * - the vnode operation vector of the first vnode should be used 238 * to determine what implementation of the op should be invoked 239 * - all mapped vnodes are of our vnode-type (NEEDSWORK: 240 * problems on rmdir'ing mount points and renaming?) 241 * 242 * null_bypass(struct vnodeop_desc *a_desc, ...) 243 */ 244 int 245 null_bypass(struct vop_generic_args *ap) 246 { 247 struct vnode **this_vp_p; 248 int error; 249 struct vnode *old_vps[VDESC_MAX_VPS]; 250 struct vnode **vps_p[VDESC_MAX_VPS]; 251 struct vnode ***vppp; 252 struct vnodeop_desc *descp = ap->a_desc; 253 int reles, i, j; 254 255 if (null_bug_bypass) 256 printf ("null_bypass: %s\n", descp->vdesc_name); 257 258 #ifdef DIAGNOSTIC 259 /* 260 * We require at least one vp. 261 */ 262 if (descp->vdesc_vp_offsets == NULL || 263 descp->vdesc_vp_offsets[0] == VDESC_NO_OFFSET) 264 panic ("null_bypass: no vp's in map"); 265 #endif 266 267 /* 268 * Map the vnodes going in. 269 */ 270 reles = descp->vdesc_flags; 271 for (i = 0; i < VDESC_MAX_VPS; ++i) { 272 if (descp->vdesc_vp_offsets[i] == VDESC_NO_OFFSET) 273 break; /* bail out at end of list */ 274 vps_p[i] = this_vp_p = 275 VOPARG_OFFSETTO(struct vnode**,descp->vdesc_vp_offsets[i],ap); 276 /* 277 * We're not guaranteed that any but the first vnode 278 * are of our type. Check for and don't map any 279 * that aren't. (We must always map first vp or vclean fails.) 280 */ 281 if (i && (*this_vp_p == NULLVP || 282 (*this_vp_p)->v_tag != VT_NULL)) { 283 old_vps[i] = NULLVP; 284 } else { 285 old_vps[i] = *this_vp_p; 286 *this_vp_p = NULLVPTOLOWERVP(*this_vp_p); 287 /* 288 * Several operations have the side effect of vrele'ing 289 * their vp's. We must account for that in the lower 290 * vp we pass down. 291 */ 292 if (reles & (VDESC_VP0_WILLRELE << i)) 293 vref(*this_vp_p); 294 } 295 296 } 297 298 /* 299 * Call the operation on the lower layer with the modified 300 * argument structure. We have to adjust a_fm to point to the 301 * lower vp's vop_ops structure. 302 */ 303 if (vps_p[0] && *vps_p[0]) { 304 ap->a_ops = *(*(vps_p[0]))->v_ops; 305 error = vop_vnoperate_ap(ap); 306 } else { 307 printf("null_bypass: no map for %s\n", descp->vdesc_name); 308 error = EINVAL; 309 } 310 311 /* 312 * Maintain the illusion of call-by-value by restoring vnodes in the 313 * argument structure to their original value. 314 */ 315 reles = descp->vdesc_flags; 316 for (i = 0; i < VDESC_MAX_VPS; ++i) { 317 if (descp->vdesc_vp_offsets[i] == VDESC_NO_OFFSET) 318 break; /* bail out at end of list */ 319 if (old_vps[i]) { 320 *(vps_p[i]) = old_vps[i]; 321 322 /* 323 * Since we operated on the lowervp's instead of the 324 * null node vp's, we have to adjust the null node 325 * vp's based on what the VOP did to the lower vp. 326 * 327 * Note: the unlock case only occurs with rename. 328 * tdvp and tvp are both locked on call and must be 329 * unlocked on return. 330 * 331 * Unlock semantics indicate that if two locked vp's 332 * are passed and they are the same vp, they are only 333 * actually locked once. 334 */ 335 if (reles & (VDESC_VP0_WILLUNLOCK << i)) { 336 VOP_UNLOCK(old_vps[i], LK_THISLAYER, curthread); 337 for (j = i + 1; j < VDESC_MAX_VPS; ++j) { 338 if (descp->vdesc_vp_offsets[j] == VDESC_NO_OFFSET) 339 break; 340 if (old_vps[i] == old_vps[j]) { 341 reles &= ~(1 << (VDESC_VP0_WILLUNLOCK << j)); 342 } 343 } 344 } 345 346 if (reles & (VDESC_VP0_WILLRELE << i)) 347 vrele(old_vps[i]); 348 } 349 } 350 351 /* 352 * Map the possible out-going vpp 353 * (Assumes that the lower layer always returns 354 * a vref'ed vpp unless it gets an error.) 355 */ 356 if (descp->vdesc_vpp_offset != VDESC_NO_OFFSET && 357 !(descp->vdesc_flags & VDESC_NOMAP_VPP) && 358 !error) { 359 /* 360 * XXX - even though some ops have vpp returned vp's, 361 * several ops actually vrele this before returning. 362 * We must avoid these ops. 363 * (This should go away when these ops are regularized.) 364 */ 365 if (descp->vdesc_flags & VDESC_VPP_WILLRELE) 366 goto out; 367 vppp = VOPARG_OFFSETTO(struct vnode***, 368 descp->vdesc_vpp_offset,ap); 369 if (*vppp) 370 error = null_node_create(old_vps[0]->v_mount, **vppp, *vppp); 371 } 372 373 out: 374 return (error); 375 } 376 377 /* 378 * We have to carry on the locking protocol on the null layer vnodes 379 * as we progress through the tree. We also have to enforce read-only 380 * if this layer is mounted read-only. 381 * 382 * null_lookup(struct vnode *a_dvp, struct vnode **a_vpp, 383 * struct componentname *a_cnp) 384 */ 385 static int 386 null_lookup(struct vop_lookup_args *ap) 387 { 388 struct componentname *cnp = ap->a_cnp; 389 struct vnode *dvp = ap->a_dvp; 390 struct thread *td = cnp->cn_td; 391 int flags = cnp->cn_flags; 392 struct vnode *vp, *ldvp, *lvp; 393 int error; 394 395 if ((dvp->v_mount->mnt_flag & MNT_RDONLY) && 396 (cnp->cn_nameiop == NAMEI_DELETE || 397 cnp->cn_nameiop == NAMEI_RENAME)) { 398 return (EROFS); 399 } 400 ldvp = NULLVPTOLOWERVP(dvp); 401 402 /* 403 * If we are doing a ".." lookup we must release the lock on dvp 404 * now, before we run a lookup in the underlying fs, or we may 405 * deadlock. If we do this we must protect ldvp by ref'ing it. 406 */ 407 if (flags & CNP_ISDOTDOT) { 408 vref(ldvp); 409 VOP_UNLOCK(dvp, LK_THISLAYER, td); 410 } 411 412 /* 413 * Due to the non-deterministic nature of the handling of the 414 * parent directory lock by lookup, we cannot call null_bypass() 415 * here. We must make a direct call. It's faster to do a direct 416 * call, anyway. 417 */ 418 vp = lvp = NULL; 419 error = VOP_LOOKUP(ldvp, &lvp, cnp); 420 if (error == EJUSTRETURN && 421 (dvp->v_mount->mnt_flag & MNT_RDONLY) && 422 (cnp->cn_nameiop == NAMEI_CREATE || 423 cnp->cn_nameiop == NAMEI_RENAME)) { 424 error = EROFS; 425 } 426 427 if ((error == 0 || error == EJUSTRETURN) && lvp != NULL) { 428 if (ldvp == lvp) { 429 *ap->a_vpp = dvp; 430 vref(dvp); 431 vrele(lvp); 432 } else { 433 error = null_node_create(dvp->v_mount, lvp, &vp); 434 if (error == 0) 435 *ap->a_vpp = vp; 436 } 437 } 438 439 /* 440 * The underlying fs will set PDIRUNLOCK if it unlocked the parent 441 * directory, which means we have to follow suit in the nullfs layer. 442 * Note that the parent directory may have already been unlocked due 443 * to the ".." case. Note that use of cnp->cn_flags instead of flags. 444 */ 445 if (flags & CNP_ISDOTDOT) { 446 if ((cnp->cn_flags & CNP_PDIRUNLOCK) == 0) 447 VOP_LOCK(dvp, LK_THISLAYER | LK_EXCLUSIVE, td); 448 vrele(ldvp); 449 } else if (cnp->cn_flags & CNP_PDIRUNLOCK) { 450 VOP_UNLOCK(dvp, LK_THISLAYER, td); 451 } 452 return (error); 453 } 454 455 /* 456 * Setattr call. Disallow write attempts if the layer is mounted read-only. 457 * 458 * null_setattr(struct vnodeop_desc *a_desc, struct vnode *a_vp, 459 * struct vattr *a_vap, struct ucred *a_cred, 460 * struct thread *a_td) 461 */ 462 int 463 null_setattr(struct vop_setattr_args *ap) 464 { 465 struct vnode *vp = ap->a_vp; 466 struct vattr *vap = ap->a_vap; 467 468 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL || 469 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL || 470 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) && 471 (vp->v_mount->mnt_flag & MNT_RDONLY)) 472 return (EROFS); 473 if (vap->va_size != VNOVAL) { 474 switch (vp->v_type) { 475 case VDIR: 476 return (EISDIR); 477 case VCHR: 478 case VBLK: 479 case VSOCK: 480 case VFIFO: 481 if (vap->va_flags != VNOVAL) 482 return (EOPNOTSUPP); 483 return (0); 484 case VREG: 485 case VLNK: 486 default: 487 /* 488 * Disallow write attempts if the filesystem is 489 * mounted read-only. 490 */ 491 if (vp->v_mount->mnt_flag & MNT_RDONLY) 492 return (EROFS); 493 } 494 } 495 496 return (null_bypass(&ap->a_head)); 497 } 498 499 /* 500 * We handle getattr only to change the fsid. 501 * 502 * null_getattr(struct vnode *a_vp, struct vattr *a_vap, struct ucred *a_cred, 503 * struct thread *a_td) 504 */ 505 static int 506 null_getattr(struct vop_getattr_args *ap) 507 { 508 int error; 509 510 if ((error = null_bypass(&ap->a_head)) != 0) 511 return (error); 512 513 ap->a_vap->va_fsid = ap->a_vp->v_mount->mnt_stat.f_fsid.val[0]; 514 return (0); 515 } 516 517 /* 518 * Resolve a locked ncp at the nullfs layer. 519 */ 520 static int 521 null_nresolve(struct vop_nresolve_args *ap) 522 { 523 return(vop_compat_nresolve(ap)); 524 } 525 526 /* 527 * Create a file 528 */ 529 static int 530 null_ncreate(struct vop_ncreate_args *ap) 531 { 532 return(vop_compat_ncreate(ap)); 533 } 534 535 static int 536 null_nmkdir(struct vop_nmkdir_args *ap) 537 { 538 return(vop_compat_nmkdir(ap)); 539 } 540 541 static int 542 null_nremove(struct vop_nremove_args *ap) 543 { 544 return(vop_compat_nremove(ap)); 545 } 546 547 static int 548 null_nrmdir(struct vop_nrmdir_args *ap) 549 { 550 return(vop_compat_nrmdir(ap)); 551 } 552 553 static int 554 null_nrename(struct vop_nrename_args *ap) 555 { 556 return(vop_compat_nrename(ap)); 557 } 558 559 /* 560 * revoke is VX locked, we can't go through null_bypass 561 */ 562 static int 563 null_revoke(struct vop_revoke_args *ap) 564 { 565 struct null_node *np; 566 struct vnode *lvp; 567 568 np = VTONULL(ap->a_vp); 569 vx_unlock(ap->a_vp); 570 if ((lvp = np->null_lowervp) != NULL) { 571 vx_get(lvp); 572 VOP_REVOKE(lvp, ap->a_flags); 573 vx_put(lvp); 574 } 575 vx_lock(ap->a_vp); 576 vgone(ap->a_vp); 577 return(0); 578 } 579 580 /* 581 * Handle to disallow write access if mounted read-only. 582 * 583 * null_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred, 584 * struct thread *a_td) 585 */ 586 static int 587 null_access(struct vop_access_args *ap) 588 { 589 struct vnode *vp = ap->a_vp; 590 mode_t mode = ap->a_mode; 591 592 /* 593 * Disallow write attempts on read-only layers; 594 * unless the file is a socket, fifo, or a block or 595 * character device resident on the file system. 596 */ 597 if (mode & VWRITE) { 598 switch (vp->v_type) { 599 case VDIR: 600 case VLNK: 601 case VREG: 602 if (vp->v_mount->mnt_flag & MNT_RDONLY) 603 return (EROFS); 604 break; 605 default: 606 break; 607 } 608 } 609 return (null_bypass(&ap->a_head)); 610 } 611 612 /* 613 * We must handle open to be able to catch MNT_NODEV and friends. 614 * 615 * null_open(struct vnode *a_vp, int a_mode, struct ucred *a_cred, 616 * struct thread *a_td) 617 */ 618 static int 619 null_open(struct vop_open_args *ap) 620 { 621 struct vnode *vp = ap->a_vp; 622 struct vnode *lvp = NULLVPTOLOWERVP(ap->a_vp); 623 624 if ((vp->v_mount->mnt_flag & MNT_NODEV) && 625 (lvp->v_type == VBLK || lvp->v_type == VCHR)) 626 return ENXIO; 627 628 return (null_bypass(&ap->a_head)); 629 } 630 631 /* 632 * We handle this to eliminate null FS to lower FS 633 * file moving. Don't know why we don't allow this, 634 * possibly we should. 635 * 636 * null_rename(struct vnode *a_fdvp, struct vnode *a_fvp, 637 * struct componentname *a_fcnp, struct vnode *a_tdvp, 638 * struct vnode *a_tvp, struct componentname *a_tcnp) 639 */ 640 static int 641 null_rename(struct vop_rename_args *ap) 642 { 643 struct vnode *tdvp = ap->a_tdvp; 644 struct vnode *fvp = ap->a_fvp; 645 struct vnode *fdvp = ap->a_fdvp; 646 struct vnode *tvp = ap->a_tvp; 647 648 /* Check for cross-device rename. */ 649 if ((fvp->v_mount != tdvp->v_mount) || 650 (tvp && (fvp->v_mount != tvp->v_mount))) { 651 if (tdvp == tvp) 652 vrele(tdvp); 653 else 654 vput(tdvp); 655 if (tvp) 656 vput(tvp); 657 vrele(fdvp); 658 vrele(fvp); 659 return (EXDEV); 660 } 661 662 return (null_bypass(&ap->a_head)); 663 } 664 665 /* 666 * A special flag, LK_THISLAYER, causes the locking function to operate 667 * ONLY on the nullfs layer. Otherwise we are responsible for locking not 668 * only our layer, but the lower layer as well. 669 * 670 * null_lock(struct vnode *a_vp, int a_flags, struct thread *a_td) 671 */ 672 static int 673 null_lock(struct vop_lock_args *ap) 674 { 675 struct vnode *vp = ap->a_vp; 676 int flags = ap->a_flags; 677 struct null_node *np = VTONULL(vp); 678 struct vnode *lvp; 679 int error; 680 681 /* 682 * Lock the nullfs layer first, disposing of the interlock in the 683 * process. 684 */ 685 KKASSERT((flags & LK_INTERLOCK) == 0); 686 error = lockmgr(&vp->v_lock, flags & ~LK_THISLAYER, 687 NULL, ap->a_td); 688 689 /* 690 * If locking only the nullfs layer, or if there is no lower layer, 691 * or if an error occured while attempting to lock the nullfs layer, 692 * we are done. 693 * 694 * np can be NULL is the vnode is being recycled from a previous 695 * hash collision. 696 */ 697 if ((flags & LK_THISLAYER) || np == NULL || 698 np->null_lowervp == NULL || error) { 699 return (error); 700 } 701 702 /* 703 * Lock the underlying vnode. If we are draining we should not drain 704 * the underlying vnode, since it is not being destroyed, but we do 705 * lock it exclusively in that case. Note that any interlocks have 706 * already been disposed of above. 707 */ 708 lvp = np->null_lowervp; 709 if ((flags & LK_TYPE_MASK) == LK_DRAIN) { 710 NULLFSDEBUG("null_lock: avoiding LK_DRAIN\n"); 711 error = vn_lock(lvp, (flags & ~LK_TYPE_MASK) | LK_EXCLUSIVE, 712 ap->a_td); 713 } else { 714 error = vn_lock(lvp, flags, ap->a_td); 715 } 716 717 /* 718 * If an error occured we have to undo our nullfs lock, then return 719 * the original error. 720 */ 721 if (error) 722 lockmgr(&vp->v_lock, LK_RELEASE, NULL, ap->a_td); 723 return(error); 724 } 725 726 /* 727 * A special flag, LK_THISLAYER, causes the unlocking function to operate 728 * ONLY on the nullfs layer. Otherwise we are responsible for unlocking not 729 * only our layer, but the lower layer as well. 730 * 731 * null_unlock(struct vnode *a_vp, int a_flags, struct thread *a_td) 732 */ 733 static int 734 null_unlock(struct vop_unlock_args *ap) 735 { 736 struct vnode *vp = ap->a_vp; 737 int flags = ap->a_flags; 738 struct null_node *np = VTONULL(vp); 739 struct vnode *lvp; 740 int error; 741 742 KKASSERT((flags & LK_INTERLOCK) == 0); 743 /* 744 * nullfs layer only 745 */ 746 if (flags & LK_THISLAYER) { 747 error = lockmgr(&vp->v_lock, 748 (flags & ~LK_THISLAYER) | LK_RELEASE, 749 NULL, ap->a_td); 750 return (error); 751 } 752 753 /* 754 * If there is no underlying vnode the lock operation occurs at 755 * the nullfs layer. np can be NULL is the vnode is being recycled 756 * from a previous hash collision. 757 */ 758 if (np == NULL || (lvp = np->null_lowervp) == NULL) { 759 error = lockmgr(&vp->v_lock, flags | LK_RELEASE, 760 NULL, ap->a_td); 761 return(error); 762 } 763 764 /* 765 * Unlock the lower layer first, then our nullfs layer. 766 */ 767 VOP_UNLOCK(lvp, flags, ap->a_td); 768 error = lockmgr(&vp->v_lock, flags | LK_RELEASE, NULL, ap->a_td); 769 return (error); 770 } 771 772 /* 773 * null_islocked(struct vnode *a_vp, struct thread *a_td) 774 * 775 * If a lower layer exists return the lock status of the lower layer, 776 * otherwise return the lock status of our nullfs layer. 777 */ 778 static int 779 null_islocked(struct vop_islocked_args *ap) 780 { 781 struct vnode *vp = ap->a_vp; 782 struct vnode *lvp; 783 struct null_node *np = VTONULL(vp); 784 int error; 785 786 lvp = np->null_lowervp; 787 if (lvp == NULL) 788 error = lockstatus(&vp->v_lock, ap->a_td); 789 else 790 error = VOP_ISLOCKED(lvp, ap->a_td); 791 return (error); 792 } 793 794 795 /* 796 * The vnode is no longer active. However, the new VFS API may retain 797 * the node in the vfs cache. There is no way to tell that someone issued 798 * a remove/rmdir operation on the underlying filesystem (yet), but we can't 799 * remove the lowervp reference here. 800 * 801 * null_inactive(struct vnode *a_vp, struct thread *a_td) 802 */ 803 static int 804 null_inactive(struct vop_inactive_args *ap) 805 { 806 /*struct vnode *vp = ap->a_vp;*/ 807 /*struct null_node *np = VTONULL(vp);*/ 808 809 /* 810 * At the moment don't do anything here. All the rest of the code 811 * assumes that lowervp will remain inact, and the inactive nullvp 812 * may be reactivated at any time. XXX I'm not sure why the 4.x code 813 * even worked. 814 */ 815 816 /* 817 * Now it is safe to release our nullfs layer vnode. 818 */ 819 return (0); 820 } 821 822 /* 823 * We can free memory in null_inactive, but we do this 824 * here. (Possible to guard vp->v_data to point somewhere) 825 * 826 * null_reclaim(struct vnode *a_vp, struct thread *a_td) 827 */ 828 static int 829 null_reclaim(struct vop_reclaim_args *ap) 830 { 831 struct vnode *vp = ap->a_vp; 832 struct vnode *lowervp; 833 struct null_node *np; 834 835 np = VTONULL(vp); 836 vp->v_data = NULL; 837 /* 838 * null_lowervp reference to lowervp. The lower vnode's 839 * inactive routine may or may not be called when we do the 840 * final vrele(). 841 */ 842 if (np) { 843 null_node_rem(np); 844 lowervp = np->null_lowervp; 845 np->null_lowervp = NULLVP; 846 if (lowervp) 847 vrele(lowervp); 848 free(np, M_NULLFSNODE); 849 } 850 return (0); 851 } 852 853 /* 854 * null_print(struct vnode *a_vp) 855 */ 856 static int 857 null_print(struct vop_print_args *ap) 858 { 859 struct vnode *vp = ap->a_vp; 860 struct null_node *np = VTONULL(vp); 861 862 if (np == NULL) { 863 printf ("\ttag VT_NULLFS, vp=%p, NULL v_data!\n", vp); 864 return(0); 865 } 866 printf ("\ttag VT_NULLFS, vp=%p, lowervp=%p\n", vp, np->null_lowervp); 867 if (np->null_lowervp != NULL) { 868 printf("\tlowervp_lock: "); 869 lockmgr_printinfo(&np->null_lowervp->v_lock); 870 } else { 871 printf("\tnull_lock: "); 872 lockmgr_printinfo(&vp->v_lock); 873 } 874 printf("\n"); 875 return (0); 876 } 877 878 /* 879 * Let an underlying filesystem do the work 880 * 881 * null_createvobject(struct vnode *vp, struct ucred *cred, struct proc *p) 882 */ 883 static int 884 null_createvobject(struct vop_createvobject_args *ap) 885 { 886 struct vnode *vp = ap->a_vp; 887 struct vnode *lowervp = VTONULL(vp) ? NULLVPTOLOWERVP(vp) : NULL; 888 int error; 889 890 if (vp->v_type == VNON || lowervp == NULL) 891 return 0; 892 error = VOP_CREATEVOBJECT(lowervp, ap->a_td); 893 if (error) 894 return (error); 895 vp->v_flag |= VOBJBUF; 896 return (0); 897 } 898 899 /* 900 * We have nothing to destroy and this operation shouldn't be bypassed. 901 * 902 * null_destroyvobject(struct vnode *vp) 903 */ 904 static int 905 null_destroyvobject(struct vop_destroyvobject_args *ap) 906 { 907 struct vnode *vp = ap->a_vp; 908 909 vp->v_flag &= ~VOBJBUF; 910 return (0); 911 } 912 913 /* 914 * null_getvobject(struct vnode *vp, struct vm_object **objpp) 915 * 916 * Note that this can be called when a vnode is being recycled, and 917 * v_data may be NULL in that case if nullfs had to recycle a vnode 918 * due to a null_node collision. 919 */ 920 static int 921 null_getvobject(struct vop_getvobject_args *ap) 922 { 923 struct vnode *lvp; 924 925 if (ap->a_vp->v_data == NULL) 926 return EINVAL; 927 928 lvp = NULLVPTOLOWERVP(ap->a_vp); 929 if (lvp == NULL) 930 return EINVAL; 931 return (VOP_GETVOBJECT(lvp, ap->a_objpp)); 932 } 933 934 /* 935 * Global vfs data structures 936 */ 937 struct vnodeopv_entry_desc null_vnodeop_entries[] = { 938 { &vop_default_desc, (vnodeopv_entry_t) null_bypass }, 939 { &vop_access_desc, (vnodeopv_entry_t) null_access }, 940 { &vop_createvobject_desc, (vnodeopv_entry_t) null_createvobject }, 941 { &vop_destroyvobject_desc, (vnodeopv_entry_t) null_destroyvobject }, 942 { &vop_getattr_desc, (vnodeopv_entry_t) null_getattr }, 943 { &vop_getvobject_desc, (vnodeopv_entry_t) null_getvobject }, 944 { &vop_inactive_desc, (vnodeopv_entry_t) null_inactive }, 945 { &vop_islocked_desc, (vnodeopv_entry_t) null_islocked }, 946 { &vop_lock_desc, (vnodeopv_entry_t) null_lock }, 947 { &vop_lookup_desc, (vnodeopv_entry_t) null_lookup }, 948 { &vop_open_desc, (vnodeopv_entry_t) null_open }, 949 { &vop_print_desc, (vnodeopv_entry_t) null_print }, 950 { &vop_reclaim_desc, (vnodeopv_entry_t) null_reclaim }, 951 { &vop_rename_desc, (vnodeopv_entry_t) null_rename }, 952 { &vop_setattr_desc, (vnodeopv_entry_t) null_setattr }, 953 { &vop_unlock_desc, (vnodeopv_entry_t) null_unlock }, 954 { &vop_revoke_desc, (vnodeopv_entry_t) null_revoke }, 955 956 { &vop_nresolve_desc, (vnodeopv_entry_t) null_nresolve }, 957 { &vop_ncreate_desc, (vnodeopv_entry_t) null_ncreate }, 958 { &vop_nmkdir_desc, (vnodeopv_entry_t) null_nmkdir }, 959 { &vop_nremove_desc, (vnodeopv_entry_t) null_nremove }, 960 { &vop_nrmdir_desc, (vnodeopv_entry_t) null_nrmdir }, 961 { &vop_nrename_desc, (vnodeopv_entry_t) null_nrename }, 962 { NULL, NULL } 963 }; 964 965