1/* $NetBSD$ */ 2 3/* 4 * Copyright (c) 1999 National Aeronautics & Space Administration 5 * All rights reserved. 6 * 7 * This software was written by William Studenmund of the 8 * Numerical Aerospace Simulation Facility, NASA Ames Research Center. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. Neither the name of the National Aeronautics & Space Administration 19 * nor the names of its contributors may be used to endorse or promote 20 * products derived from this software without specific prior written 21 * permission. 22 * 23 * THIS SOFTWARE IS PROVIDED BY THE NATIONAL AERONAUTICS & SPACE ADMINISTRATION 24 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 25 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 26 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE ADMINISTRATION OR CONTRIB- 27 * UTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, 28 * OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 29 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 30 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 31 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 32 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 33 * POSSIBILITY OF SUCH DAMAGE. 34 */ 35 36/* 37 * Copyright (c) 1992, 1993 38 * The Regents of the University of California. All rights reserved. 39 * 40 * This code is derived from software contributed to Berkeley by 41 * John Heidemann of the UCLA Ficus project. 42 * 43 * Redistribution and use in source and binary forms, with or without 44 * modification, are permitted provided that the following conditions 45 * are met: 46 * 1. Redistributions of source code must retain the above copyright 47 * notice, this list of conditions and the following disclaimer. 48 * 2. Redistributions in binary form must reproduce the above copyright 49 * notice, this list of conditions and the following disclaimer in the 50 * documentation and/or other materials provided with the distribution. 51 * 3. Neither the name of the University nor the names of its contributors 52 * may be used to endorse or promote products derived from this software 53 * without specific prior written permission. 54 * 55 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 56 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 57 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 58 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 59 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 60 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 61 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 62 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 63 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 64 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 65 * SUCH DAMAGE. 66 * 67 * @(#)null_vnops.c 8.6 (Berkeley) 5/27/95 68 * 69 * Ancestors: 70 * @(#)lofs_vnops.c 1.2 (Berkeley) 6/18/92 71 * Id: lofs_vnops.c,v 1.11 1992/05/30 10:05:43 jsp Exp jsp 72 * ...and... 73 * @(#)null_vnodeops.c 1.20 92/07/07 UCLA Ficus project 74 */ 75 76/* 77 * Generic layer vnode operations. 78 * 79 * The layer.h, layer_extern.h, layer_vfs.c, and layer_vnops.c files provide 80 * the core implementation of stacked file-systems. 81 * 82 * The layerfs duplicates a portion of the file system name space under 83 * a new name. In this respect, it is similar to the loopback file system. 84 * It differs from the loopback fs in two respects: it is implemented using 85 * a stackable layers technique, and it is "layerfs-nodes" stack above all 86 * lower-layer vnodes, not just over directory vnodes. 87 * 88 * OPERATION OF LAYERFS 89 * 90 * The layerfs is the minimum file system layer, bypassing all possible 91 * operations to the lower layer for processing there. The majority of its 92 * activity centers on the bypass routine, through which nearly all vnode 93 * operations pass. 94 * 95 * The bypass routine accepts arbitrary vnode operations for handling by 96 * the lower layer. It begins by examining vnode operation arguments and 97 * replacing any layered nodes by their lower-layer equivalents. It then 98 * invokes an operation on the lower layer. Finally, it replaces the 99 * layered nodes in the arguments and, if a vnode is returned by the 100 * operation, stacks a layered node on top of the returned vnode. 101 * 102 * The bypass routine in this file, layer_bypass(), is suitable for use 103 * by many different layered filesystems. It can be used by multiple 104 * filesystems simultaneously. Alternatively, a layered fs may provide 105 * its own bypass routine, in which case layer_bypass() should be used as 106 * a model. For instance, the main functionality provided by umapfs, the user 107 * identity mapping file system, is handled by a custom bypass routine. 108 * 109 * Typically a layered fs registers its selected bypass routine as the 110 * default vnode operation in its vnodeopv_entry_desc table. Additionally 111 * the filesystem must store the bypass entry point in the layerm_bypass 112 * field of struct layer_mount. All other layer routines in this file will 113 * use the layerm_bypass() routine. 114 * 115 * Although the bypass routine handles most operations outright, a number 116 * of operations are special cased and handled by the layerfs. For instance, 117 * layer_getattr() must change the fsid being returned. While layer_lock() 118 * and layer_unlock() must handle any locking for the current vnode as well 119 * as pass the lock request down. layer_inactive() and layer_reclaim() are 120 * not bypassed so that they can handle freeing layerfs-specific data. Also, 121 * certain vnode operations (create, mknod, remove, link, rename, mkdir, 122 * rmdir, and symlink) change the locking state within the operation. Ideally 123 * these operations should not change the lock state, but should be changed 124 * to let the caller of the function unlock them. Otherwise, all intermediate 125 * vnode layers (such as union, umapfs, etc) must catch these functions to do 126 * the necessary locking at their layer. 127 * 128 * INSTANTIATING VNODE STACKS 129 * 130 * Mounting associates "layerfs-nodes" stack and lower layer, in effect 131 * stacking two VFSes. The initial mount creates a single vnode stack for 132 * the root of the new layerfs. All other vnode stacks are created as a 133 * result of vnode operations on this or other layerfs vnode stacks. 134 * 135 * New vnode stacks come into existence as a result of an operation which 136 * returns a vnode. The bypass routine stacks a layerfs-node above the new 137 * vnode before returning it to the caller. 138 * 139 * For example, imagine mounting a null layer with: 140 * 141 * "mount_null /usr/include /dev/layer/null" 142 * 143 * Changing directory to /dev/layer/null will assign the root layerfs-node, 144 * which was created when the null layer was mounted). Now consider opening 145 * "sys". A layer_lookup() would be performed on the root layerfs-node. 146 * This operation would bypass through to the lower layer which would return 147 * a vnode representing the UFS "sys". Then, layer_bypass() builds a 148 * layerfs-node aliasing the UFS "sys" and returns this to the caller. 149 * Later operations on the layerfs-node "sys" will repeat this process when 150 * constructing other vnode stacks. 151 * 152 * INVOKING OPERATIONS ON LOWER LAYERS 153 * 154 * There are two techniques to invoke operations on a lower layer when the 155 * operation cannot be completely bypassed. Each method is appropriate in 156 * different situations. In both cases, it is the responsibility of the 157 * aliasing layer to make the operation arguments "correct" for the lower 158 * layer by mapping any vnode arguments to the lower layer. 159 * 160 * The first approach is to call the aliasing layer's bypass routine. This 161 * method is most suitable when you wish to invoke the operation currently 162 * being handled on the lower layer. It has the advantage that the bypass 163 * routine already must do argument mapping. An example of this is 164 * layer_getattr(). 165 * 166 * A second approach is to directly invoke vnode operations on the lower 167 * layer with the VOP_OPERATIONNAME interface. The advantage of this method 168 * is that it is easy to invoke arbitrary operations on the lower layer. 169 * The disadvantage is that vnode's arguments must be manually mapped. 170 */ 171 172#include <sys/cdefs.h> 173__KERNEL_RCSID(0, "$NetBSD$"); 174 175#include <sys/param.h> 176#include <sys/systm.h> 177#include <sys/proc.h> 178#include <sys/time.h> 179#include <sys/vnode.h> 180#include <sys/mount.h> 181#include <sys/namei.h> 182#include <sys/kmem.h> 183#include <sys/buf.h> 184#include <sys/kauth.h> 185 186#include <miscfs/genfs/layer.h> 187#include <miscfs/genfs/layer_extern.h> 188#include <miscfs/genfs/genfs.h> 189#include <miscfs/specfs/specdev.h> 190 191/* 192 * This is the 08-June-99 bypass routine, based on the 10-Apr-92 bypass 193 * routine by John Heidemann. 194 * The new element for this version is that the whole nullfs 195 * system gained the concept of locks on the lower node. 196 * The 10-Apr-92 version was optimized for speed, throwing away some 197 * safety checks. It should still always work, but it's not as 198 * robust to programmer errors. 199 * 200 * In general, we map all vnodes going down and unmap them on the way back. 201 * 202 * Also, some BSD vnode operations have the side effect of vrele'ing 203 * their arguments. With stacking, the reference counts are held 204 * by the upper node, not the lower one, so we must handle these 205 * side-effects here. This is not of concern in Sun-derived systems 206 * since there are no such side-effects. 207 * 208 * New for the 08-June-99 version: we also handle operations which unlock 209 * the passed-in node (typically they vput the node). 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 220layer_bypass(void *v) 221{ 222 struct vop_generic_args /* { 223 struct vnodeop_desc *a_desc; 224 <other random data follows, presumably> 225 } */ *ap = v; 226 int (**our_vnodeop_p)(void *); 227 struct vnode **this_vp_p; 228 int error; 229 struct vnode *old_vps[VDESC_MAX_VPS], *vp0; 230 struct vnode **vps_p[VDESC_MAX_VPS]; 231 struct vnode ***vppp; 232 struct mount *mp; 233 struct vnodeop_desc *descp = ap->a_desc; 234 int reles, i, flags; 235 236#ifdef DIAGNOSTIC 237 /* 238 * We require at least one vp. 239 */ 240 if (descp->vdesc_vp_offsets == NULL || 241 descp->vdesc_vp_offsets[0] == VDESC_NO_OFFSET) 242 panic("%s: no vp's in map.\n", __func__); 243#endif 244 245 vps_p[0] = 246 VOPARG_OFFSETTO(struct vnode**, descp->vdesc_vp_offsets[0], ap); 247 vp0 = *vps_p[0]; 248 mp = vp0->v_mount; 249 flags = MOUNTTOLAYERMOUNT(mp)->layerm_flags; 250 our_vnodeop_p = vp0->v_op; 251 252 if (flags & LAYERFS_MBYPASSDEBUG) 253 printf("%s: %s\n", __func__, descp->vdesc_name); 254 255 /* 256 * Map the vnodes going in. 257 * Later, we'll invoke the operation based on 258 * the first mapped vnode's operation vector. 259 */ 260 reles = descp->vdesc_flags; 261 for (i = 0; i < VDESC_MAX_VPS; reles >>= 1, i++) { 262 if (descp->vdesc_vp_offsets[i] == VDESC_NO_OFFSET) 263 break; /* bail out at end of list */ 264 vps_p[i] = this_vp_p = 265 VOPARG_OFFSETTO(struct vnode**, descp->vdesc_vp_offsets[i], 266 ap); 267 /* 268 * We're not guaranteed that any but the first vnode 269 * are of our type. Check for and don't map any 270 * that aren't. (We must always map first vp or vclean fails.) 271 */ 272 if (i && (*this_vp_p == NULL || 273 (*this_vp_p)->v_op != our_vnodeop_p)) { 274 old_vps[i] = NULL; 275 } else { 276 old_vps[i] = *this_vp_p; 277 *(vps_p[i]) = LAYERVPTOLOWERVP(*this_vp_p); 278 /* 279 * XXX - Several operations have the side effect 280 * of vrele'ing their vp's. We must account for 281 * that. (This should go away in the future.) 282 */ 283 if (reles & VDESC_VP0_WILLRELE) 284 vref(*this_vp_p); 285 } 286 } 287 288 /* 289 * Call the operation on the lower layer 290 * with the modified argument structure. 291 */ 292 error = VCALL(*vps_p[0], descp->vdesc_offset, ap); 293 294 /* 295 * Maintain the illusion of call-by-value 296 * by restoring vnodes in the argument structure 297 * to their original value. 298 */ 299 reles = descp->vdesc_flags; 300 for (i = 0; i < VDESC_MAX_VPS; reles >>= 1, i++) { 301 if (descp->vdesc_vp_offsets[i] == VDESC_NO_OFFSET) 302 break; /* bail out at end of list */ 303 if (old_vps[i]) { 304 *(vps_p[i]) = old_vps[i]; 305 if (reles & VDESC_VP0_WILLRELE) 306 vrele(*(vps_p[i])); 307 } 308 } 309 310 /* 311 * Map the possible out-going vpp 312 * (Assumes that the lower layer always returns 313 * a VREF'ed vpp unless it gets an error.) 314 */ 315 if (descp->vdesc_vpp_offset != VDESC_NO_OFFSET && !error) { 316 vppp = VOPARG_OFFSETTO(struct vnode***, 317 descp->vdesc_vpp_offset, ap); 318 /* 319 * Only vop_lookup, vop_create, vop_makedir, vop_bmap, 320 * vop_mknod, and vop_symlink return vpp's. vop_bmap 321 * doesn't call bypass as the lower vpp is fine (we're just 322 * going to do i/o on it). vop_lookup doesn't call bypass 323 * as a lookup on "." would generate a locking error. 324 * So all the calls which get us here have a locked vpp. :-) 325 */ 326 error = layer_node_create(mp, **vppp, *vppp); 327 if (error) { 328 vput(**vppp); 329 **vppp = NULL; 330 } 331 } 332 return error; 333} 334 335/* 336 * We have to carry on the locking protocol on the layer vnodes 337 * as we progress through the tree. We also have to enforce read-only 338 * if this layer is mounted read-only. 339 */ 340int 341layer_lookup(void *v) 342{ 343 struct vop_lookup_args /* { 344 struct vnodeop_desc *a_desc; 345 struct vnode * a_dvp; 346 struct vnode ** a_vpp; 347 struct componentname * a_cnp; 348 } */ *ap = v; 349 struct componentname *cnp = ap->a_cnp; 350 struct vnode *dvp, *lvp, *ldvp; 351 int error, flags = cnp->cn_flags; 352 353 dvp = ap->a_dvp; 354 355 if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) && 356 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) { 357 *ap->a_vpp = NULL; 358 return EROFS; 359 } 360 361 ldvp = LAYERVPTOLOWERVP(dvp); 362 ap->a_dvp = ldvp; 363 error = VCALL(ldvp, ap->a_desc->vdesc_offset, ap); 364 lvp = *ap->a_vpp; 365 *ap->a_vpp = NULL; 366 367 if (error == EJUSTRETURN && (flags & ISLASTCN) && 368 (dvp->v_mount->mnt_flag & MNT_RDONLY) && 369 (cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME)) 370 error = EROFS; 371 372 /* 373 * We must do the same locking and unlocking at this layer as 374 * is done in the layers below us. 375 */ 376 if (ldvp == lvp) { 377 /* 378 * Got the same object back, because we looked up ".", 379 * or ".." in the root node of a mount point. 380 * So we make another reference to dvp and return it. 381 */ 382 vref(dvp); 383 *ap->a_vpp = dvp; 384 vrele(lvp); 385 } else if (lvp != NULL) { 386 /* Note: dvp, ldvp and lvp are all locked. */ 387 error = layer_node_create(dvp->v_mount, lvp, ap->a_vpp); 388 if (error) { 389 vput(lvp); 390 } 391 } 392 return error; 393} 394 395/* 396 * Setattr call. Disallow write attempts if the layer is mounted read-only. 397 */ 398int 399layer_setattr(void *v) 400{ 401 struct vop_setattr_args /* { 402 struct vnodeop_desc *a_desc; 403 struct vnode *a_vp; 404 struct vattr *a_vap; 405 kauth_cred_t a_cred; 406 struct lwp *a_l; 407 } */ *ap = v; 408 struct vnode *vp = ap->a_vp; 409 struct vattr *vap = ap->a_vap; 410 411 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL || 412 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL || 413 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) && 414 (vp->v_mount->mnt_flag & MNT_RDONLY)) 415 return EROFS; 416 if (vap->va_size != VNOVAL) { 417 switch (vp->v_type) { 418 case VDIR: 419 return EISDIR; 420 case VCHR: 421 case VBLK: 422 case VSOCK: 423 case VFIFO: 424 return 0; 425 case VREG: 426 case VLNK: 427 default: 428 /* 429 * Disallow write attempts if the filesystem is 430 * mounted read-only. 431 */ 432 if (vp->v_mount->mnt_flag & MNT_RDONLY) 433 return EROFS; 434 } 435 } 436 return LAYERFS_DO_BYPASS(vp, ap); 437} 438 439/* 440 * We handle getattr only to change the fsid. 441 */ 442int 443layer_getattr(void *v) 444{ 445 struct vop_getattr_args /* { 446 struct vnode *a_vp; 447 struct vattr *a_vap; 448 kauth_cred_t a_cred; 449 struct lwp *a_l; 450 } */ *ap = v; 451 struct vnode *vp = ap->a_vp; 452 int error; 453 454 error = LAYERFS_DO_BYPASS(vp, ap); 455 if (error) { 456 return error; 457 } 458 /* Requires that arguments be restored. */ 459 ap->a_vap->va_fsid = vp->v_mount->mnt_stat.f_fsidx.__fsid_val[0]; 460 return 0; 461} 462 463int 464layer_access(void *v) 465{ 466 struct vop_access_args /* { 467 struct vnode *a_vp; 468 int a_mode; 469 kauth_cred_t a_cred; 470 struct lwp *a_l; 471 } */ *ap = v; 472 struct vnode *vp = ap->a_vp; 473 mode_t mode = ap->a_mode; 474 475 /* 476 * Disallow write attempts on read-only layers; 477 * unless the file is a socket, fifo, or a block or 478 * character device resident on the file system. 479 */ 480 if (mode & VWRITE) { 481 switch (vp->v_type) { 482 case VDIR: 483 case VLNK: 484 case VREG: 485 if (vp->v_mount->mnt_flag & MNT_RDONLY) 486 return EROFS; 487 break; 488 default: 489 break; 490 } 491 } 492 return LAYERFS_DO_BYPASS(vp, ap); 493} 494 495/* 496 * We must handle open to be able to catch MNT_NODEV and friends. 497 */ 498int 499layer_open(void *v) 500{ 501 struct vop_open_args /* { 502 const struct vnodeop_desc *a_desc; 503 struct vnode *a_vp; 504 int a_mode; 505 kauth_cred_t a_cred; 506 } */ *ap = v; 507 struct vnode *vp = ap->a_vp; 508 enum vtype lower_type = LAYERVPTOLOWERVP(vp)->v_type; 509 510 if (((lower_type == VBLK) || (lower_type == VCHR)) && 511 (vp->v_mount->mnt_flag & MNT_NODEV)) 512 return ENXIO; 513 514 return LAYERFS_DO_BYPASS(vp, ap); 515} 516 517/* 518 * If vinvalbuf is calling us, it's a "shallow fsync" -- don't bother 519 * syncing the underlying vnodes, since they'll be fsync'ed when 520 * reclaimed; otherwise, pass it through to the underlying layer. 521 * 522 * XXX Do we still need to worry about shallow fsync? 523 */ 524int 525layer_fsync(void *v) 526{ 527 struct vop_fsync_args /* { 528 struct vnode *a_vp; 529 kauth_cred_t a_cred; 530 int a_flags; 531 off_t offlo; 532 off_t offhi; 533 struct lwp *a_l; 534 } */ *ap = v; 535 int error; 536 537 if (ap->a_flags & FSYNC_RECLAIM) { 538 return 0; 539 } 540 if (ap->a_vp->v_type == VBLK || ap->a_vp->v_type == VCHR) { 541 error = spec_fsync(v); 542 if (error) 543 return error; 544 } 545 return LAYERFS_DO_BYPASS(ap->a_vp, ap); 546} 547 548int 549layer_inactive(void *v) 550{ 551 struct vop_inactive_args /* { 552 struct vnode *a_vp; 553 bool *a_recycle; 554 } */ *ap = v; 555 struct vnode *vp = ap->a_vp; 556 557 /* 558 * If we did a remove, don't cache the node. 559 */ 560 *ap->a_recycle = ((VTOLAYER(vp)->layer_flags & LAYERFS_REMOVED) != 0); 561 562 /* 563 * Do nothing (and _don't_ bypass). 564 * Wait to vrele lowervp until reclaim, 565 * so that until then our layer_node is in the 566 * cache and reusable. 567 * 568 * NEEDSWORK: Someday, consider inactive'ing 569 * the lowervp and then trying to reactivate it 570 * with capabilities (v_id) 571 * like they do in the name lookup cache code. 572 * That's too much work for now. 573 */ 574 VOP_UNLOCK(vp); 575 return 0; 576} 577 578int 579layer_remove(void *v) 580{ 581 struct vop_remove_args /* { 582 struct vonde *a_dvp; 583 struct vnode *a_vp; 584 struct componentname *a_cnp; 585 } */ *ap = v; 586 struct vnode *vp = ap->a_vp; 587 int error; 588 589 vref(vp); 590 error = LAYERFS_DO_BYPASS(vp, ap); 591 if (error == 0) { 592 VTOLAYER(vp)->layer_flags |= LAYERFS_REMOVED; 593 } 594 vrele(vp); 595 596 return error; 597} 598 599int 600layer_rename(void *v) 601{ 602 struct vop_rename_args /* { 603 struct vnode *a_fdvp; 604 struct vnode *a_fvp; 605 struct componentname *a_fcnp; 606 struct vnode *a_tdvp; 607 struct vnode *a_tvp; 608 struct componentname *a_tcnp; 609 } */ *ap = v; 610 struct vnode *fdvp = ap->a_fdvp, *tvp; 611 int error; 612 613 tvp = ap->a_tvp; 614 if (tvp) { 615 if (tvp->v_mount != fdvp->v_mount) 616 tvp = NULL; 617 else 618 vref(tvp); 619 } 620 error = LAYERFS_DO_BYPASS(fdvp, ap); 621 if (tvp) { 622 if (error == 0) 623 VTOLAYER(tvp)->layer_flags |= LAYERFS_REMOVED; 624 vrele(tvp); 625 } 626 return error; 627} 628 629int 630layer_rmdir(void *v) 631{ 632 struct vop_rmdir_args /* { 633 struct vnode *a_dvp; 634 struct vnode *a_vp; 635 struct componentname *a_cnp; 636 } */ *ap = v; 637 int error; 638 struct vnode *vp = ap->a_vp; 639 640 vref(vp); 641 error = LAYERFS_DO_BYPASS(vp, ap); 642 if (error == 0) { 643 VTOLAYER(vp)->layer_flags |= LAYERFS_REMOVED; 644 } 645 vrele(vp); 646 647 return error; 648} 649 650int 651layer_revoke(void *v) 652{ 653 struct vop_revoke_args /* { 654 struct vnode *a_vp; 655 int a_flags; 656 } */ *ap = v; 657 struct vnode *vp = ap->a_vp; 658 struct vnode *lvp = LAYERVPTOLOWERVP(vp); 659 int error; 660 661 /* 662 * We will most likely end up in vclean which uses the v_usecount 663 * to determine if a vnode is active. Take an extra reference on 664 * the lower vnode so it will always close and inactivate. 665 */ 666 vref(lvp); 667 error = LAYERFS_DO_BYPASS(vp, ap); 668 vrele(lvp); 669 670 return error; 671} 672 673int 674layer_reclaim(void *v) 675{ 676 struct vop_reclaim_args /* { 677 struct vnode *a_vp; 678 struct lwp *a_l; 679 } */ *ap = v; 680 struct vnode *vp = ap->a_vp; 681 struct layer_mount *lmp = MOUNTTOLAYERMOUNT(vp->v_mount); 682 struct layer_node *xp = VTOLAYER(vp); 683 struct vnode *lowervp = xp->layer_lowervp; 684 685 /* 686 * Note: in vop_reclaim, the node's struct lock has been 687 * decomissioned, so we have to be careful about calling 688 * VOP's on ourself. We must be careful as VXLOCK is set. 689 */ 690 if (vp == lmp->layerm_rootvp) { 691 /* 692 * Oops! We no longer have a root node. Most likely reason is 693 * that someone forcably unmunted the underlying fs. 694 * 695 * Now getting the root vnode will fail. We're dead. :-( 696 */ 697 lmp->layerm_rootvp = NULL; 698 } 699 /* After this assignment, this node will not be re-used. */ 700 xp->layer_lowervp = NULL; 701 mutex_enter(&lmp->layerm_hashlock); 702 LIST_REMOVE(xp, layer_hash); 703 mutex_exit(&lmp->layerm_hashlock); 704 kmem_free(vp->v_data, lmp->layerm_size); 705 vp->v_data = NULL; 706 vrele(lowervp); 707 708 return 0; 709} 710 711/* 712 * We just feed the returned vnode up to the caller - there's no need 713 * to build a layer node on top of the node on which we're going to do 714 * i/o. :-) 715 */ 716int 717layer_bmap(void *v) 718{ 719 struct vop_bmap_args /* { 720 struct vnode *a_vp; 721 daddr_t a_bn; 722 struct vnode **a_vpp; 723 daddr_t *a_bnp; 724 int *a_runp; 725 } */ *ap = v; 726 struct vnode *vp; 727 728 vp = LAYERVPTOLOWERVP(ap->a_vp); 729 ap->a_vp = vp; 730 731 return VCALL(vp, ap->a_desc->vdesc_offset, ap); 732} 733 734int 735layer_print(void *v) 736{ 737 struct vop_print_args /* { 738 struct vnode *a_vp; 739 } */ *ap = v; 740 struct vnode *vp = ap->a_vp; 741 printf ("\ttag VT_LAYERFS, vp=%p, lowervp=%p\n", vp, LAYERVPTOLOWERVP(vp)); 742 return 0; 743} 744 745int 746layer_getpages(void *v) 747{ 748 struct vop_getpages_args /* { 749 struct vnode *a_vp; 750 voff_t a_offset; 751 struct vm_page **a_m; 752 int *a_count; 753 int a_centeridx; 754 vm_prot_t a_access_type; 755 int a_advice; 756 int a_flags; 757 } */ *ap = v; 758 struct vnode *vp = ap->a_vp; 759 760 KASSERT(mutex_owned(vp->v_interlock)); 761 762 if (ap->a_flags & PGO_LOCKED) { 763 return EBUSY; 764 } 765 ap->a_vp = LAYERVPTOLOWERVP(vp); 766 KASSERT(vp->v_interlock == ap->a_vp->v_interlock); 767 768 /* Just pass the request on to the underlying layer. */ 769 return VCALL(ap->a_vp, VOFFSET(vop_getpages), ap); 770} 771 772int 773layer_putpages(void *v) 774{ 775 struct vop_putpages_args /* { 776 struct vnode *a_vp; 777 voff_t a_offlo; 778 voff_t a_offhi; 779 int a_flags; 780 } */ *ap = v; 781 struct vnode *vp = ap->a_vp; 782 783 KASSERT(mutex_owned(vp->v_interlock)); 784 785 ap->a_vp = LAYERVPTOLOWERVP(vp); 786 KASSERT(vp->v_interlock == ap->a_vp->v_interlock); 787 788 if (ap->a_flags & PGO_RECLAIM) { 789 mutex_exit(vp->v_interlock); 790 return 0; 791 } 792 793 /* Just pass the request on to the underlying layer. */ 794 return VCALL(ap->a_vp, VOFFSET(vop_putpages), ap); 795} 796