1 /* $NetBSD: union_vnops.c,v 1.84 2025/09/02 21:46:54 dholland Exp $ */ 2 3 /* 4 * Copyright (c) 1992, 1993, 1994, 1995 5 * The Regents of the University of California. All rights reserved. 6 * 7 * This code is derived from software contributed to Berkeley by 8 * Jan-Simon Pendry. 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 University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * @(#)union_vnops.c 8.33 (Berkeley) 7/31/95 35 */ 36 37 /* 38 * Copyright (c) 1992, 1993, 1994, 1995 Jan-Simon Pendry. 39 * 40 * This code is derived from software contributed to Berkeley by 41 * Jan-Simon Pendry. 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. All advertising materials mentioning features or use of this software 52 * must display the following acknowledgement: 53 * This product includes software developed by the University of 54 * California, Berkeley and its contributors. 55 * 4. Neither the name of the University nor the names of its contributors 56 * may be used to endorse or promote products derived from this software 57 * without specific prior written permission. 58 * 59 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 60 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 61 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 62 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 63 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 64 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 65 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 66 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 67 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 68 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 69 * SUCH DAMAGE. 70 * 71 * @(#)union_vnops.c 8.33 (Berkeley) 7/31/95 72 */ 73 74 #include <sys/cdefs.h> 75 __KERNEL_RCSID(0, "$NetBSD: union_vnops.c,v 1.84 2025/09/02 21:46:54 dholland Exp $"); 76 77 #include <sys/param.h> 78 #include <sys/systm.h> 79 #include <sys/proc.h> 80 #include <sys/file.h> 81 #include <sys/time.h> 82 #include <sys/stat.h> 83 #include <sys/vnode.h> 84 #include <sys/mount.h> 85 #include <sys/namei.h> 86 #include <sys/malloc.h> 87 #include <sys/buf.h> 88 #include <sys/queue.h> 89 #include <sys/lock.h> 90 #include <sys/kauth.h> 91 92 #include <fs/union/union.h> 93 #include <miscfs/genfs/genfs.h> 94 #include <miscfs/specfs/specdev.h> 95 96 int union_parsepath(void *); 97 int union_lookup(void *); 98 int union_create(void *); 99 int union_whiteout(void *); 100 int union_mknod(void *); 101 int union_open(void *); 102 int union_close(void *); 103 int union_access(void *); 104 int union_getattr(void *); 105 int union_setattr(void *); 106 int union_read(void *); 107 int union_write(void *); 108 int union_ioctl(void *); 109 int union_poll(void *); 110 int union_revoke(void *); 111 int union_mmap(void *); 112 int union_fsync(void *); 113 int union_seek(void *); 114 int union_remove(void *); 115 int union_link(void *); 116 int union_rename(void *); 117 int union_mkdir(void *); 118 int union_rmdir(void *); 119 int union_symlink(void *); 120 int union_readdir(void *); 121 int union_readlink(void *); 122 int union_abortop(void *); 123 int union_inactive(void *); 124 int union_reclaim(void *); 125 int union_lock(void *); 126 int union_unlock(void *); 127 int union_bmap(void *); 128 int union_print(void *); 129 int union_islocked(void *); 130 int union_pathconf(void *); 131 int union_advlock(void *); 132 int union_strategy(void *); 133 int union_bwrite(void *); 134 int union_getpages(void *); 135 int union_putpages(void *); 136 int union_kqfilter(void *); 137 138 static int union_lookup1(struct vnode *, struct vnode **, 139 struct vnode **, struct componentname *); 140 141 142 /* 143 * Global vfs data structures 144 */ 145 int (**union_vnodeop_p)(void *); 146 const struct vnodeopv_entry_desc union_vnodeop_entries[] = { 147 { &vop_default_desc, vn_default_error }, 148 { &vop_parsepath_desc, union_parsepath }, /* parsepath */ 149 { &vop_lookup_desc, union_lookup }, /* lookup */ 150 { &vop_create_desc, union_create }, /* create */ 151 { &vop_whiteout_desc, union_whiteout }, /* whiteout */ 152 { &vop_mknod_desc, union_mknod }, /* mknod */ 153 { &vop_open_desc, union_open }, /* open */ 154 { &vop_close_desc, union_close }, /* close */ 155 { &vop_access_desc, union_access }, /* access */ 156 { &vop_accessx_desc, genfs_accessx }, /* accessx */ 157 { &vop_getattr_desc, union_getattr }, /* getattr */ 158 { &vop_setattr_desc, union_setattr }, /* setattr */ 159 { &vop_read_desc, union_read }, /* read */ 160 { &vop_write_desc, union_write }, /* write */ 161 { &vop_fallocate_desc, genfs_eopnotsupp }, /* fallocate */ 162 { &vop_fdiscard_desc, genfs_eopnotsupp }, /* fdiscard */ 163 { &vop_ioctl_desc, union_ioctl }, /* ioctl */ 164 { &vop_poll_desc, union_poll }, /* select */ 165 { &vop_revoke_desc, union_revoke }, /* revoke */ 166 { &vop_mmap_desc, union_mmap }, /* mmap */ 167 { &vop_fsync_desc, union_fsync }, /* fsync */ 168 { &vop_seek_desc, union_seek }, /* seek */ 169 { &vop_remove_desc, union_remove }, /* remove */ 170 { &vop_link_desc, union_link }, /* link */ 171 { &vop_rename_desc, union_rename }, /* rename */ 172 { &vop_mkdir_desc, union_mkdir }, /* mkdir */ 173 { &vop_rmdir_desc, union_rmdir }, /* rmdir */ 174 { &vop_symlink_desc, union_symlink }, /* symlink */ 175 { &vop_readdir_desc, union_readdir }, /* readdir */ 176 { &vop_readlink_desc, union_readlink }, /* readlink */ 177 { &vop_abortop_desc, union_abortop }, /* abortop */ 178 { &vop_inactive_desc, union_inactive }, /* inactive */ 179 { &vop_reclaim_desc, union_reclaim }, /* reclaim */ 180 { &vop_lock_desc, union_lock }, /* lock */ 181 { &vop_unlock_desc, union_unlock }, /* unlock */ 182 { &vop_bmap_desc, union_bmap }, /* bmap */ 183 { &vop_strategy_desc, union_strategy }, /* strategy */ 184 { &vop_bwrite_desc, union_bwrite }, /* bwrite */ 185 { &vop_print_desc, union_print }, /* print */ 186 { &vop_islocked_desc, union_islocked }, /* islocked */ 187 { &vop_pathconf_desc, union_pathconf }, /* pathconf */ 188 { &vop_advlock_desc, union_advlock }, /* advlock */ 189 { &vop_getpages_desc, union_getpages }, /* getpages */ 190 { &vop_putpages_desc, union_putpages }, /* putpages */ 191 { &vop_kqfilter_desc, union_kqfilter }, /* kqfilter */ 192 { NULL, NULL } 193 }; 194 const struct vnodeopv_desc union_vnodeop_opv_desc = 195 { &union_vnodeop_p, union_vnodeop_entries }; 196 197 #define NODE_IS_SPECIAL(vp) \ 198 ((vp)->v_type == VBLK || (vp)->v_type == VCHR || \ 199 (vp)->v_type == VSOCK || (vp)->v_type == VFIFO) 200 201 int 202 union_parsepath(void *v) 203 { 204 struct vop_parsepath_args /* { 205 struct vnode *a_dvp; 206 const char *a_name; 207 size_t *a_retval; 208 } */ *ap = v; 209 struct vnode *upperdvp, *lowerdvp; 210 size_t upper, lower; 211 int error; 212 213 upperdvp = UPPERVP(ap->a_dvp); 214 lowerdvp = LOWERVP(ap->a_dvp); 215 216 if (upperdvp != NULLVP) { 217 error = VOP_PARSEPATH(upperdvp, ap->a_name, &upper); 218 if (error) { 219 return error; 220 } 221 } else { 222 upper = 0; 223 } 224 225 if (lowerdvp != NULLVP) { 226 error = VOP_PARSEPATH(lowerdvp, ap->a_name, &lower); 227 if (error) { 228 return error; 229 } 230 } else { 231 lower = 0; 232 } 233 234 if (upper == 0 && lower == 0) { 235 panic("%s: missing both layers", __func__); 236 } 237 238 /* 239 * If they're different, use the larger one. This is not a 240 * comprehensive solution, but it's sufficient for the 241 * non-default cases of parsepath that currently exist. 242 */ 243 *ap->a_retval = MAX(upper, lower); 244 return 0; 245 } 246 247 static int 248 union_lookup1(struct vnode *udvp, struct vnode **dvpp, struct vnode **vpp, 249 struct componentname *cnp) 250 { 251 int error; 252 struct vnode *tdvp; 253 struct vnode *dvp; 254 struct mount *mp; 255 256 dvp = *dvpp; 257 258 /* 259 * If stepping up the directory tree, check for going 260 * back across the mount point, in which case do what 261 * lookup would do by stepping back down the mount 262 * hierarchy. 263 */ 264 if (cnp->cn_flags & ISDOTDOT) { 265 while ((dvp != udvp) && (dvp->v_vflag & VV_ROOT)) { 266 /* 267 * Don't do the NOCROSSMOUNT check 268 * at this level. By definition, 269 * union fs deals with namespaces, not 270 * filesystems. 271 */ 272 tdvp = dvp; 273 *dvpp = dvp = dvp->v_mount->mnt_vnodecovered; 274 VOP_UNLOCK(tdvp); 275 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 276 } 277 } 278 279 error = VOP_LOOKUP(dvp, &tdvp, cnp); 280 if (error) 281 return (error); 282 if (dvp != tdvp) { 283 if (cnp->cn_flags & ISDOTDOT) 284 VOP_UNLOCK(dvp); 285 error = vn_lock(tdvp, LK_EXCLUSIVE); 286 if (cnp->cn_flags & ISDOTDOT) 287 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 288 if (error) { 289 vrele(tdvp); 290 return error; 291 } 292 dvp = tdvp; 293 } 294 295 /* 296 * Lastly check if the current node is a mount point in 297 * which case walk up the mount hierarchy making sure not to 298 * bump into the root of the mount tree (ie. dvp != udvp). 299 */ 300 while (dvp != udvp && (dvp->v_type == VDIR) && 301 (mp = dvp->v_mountedhere)) { 302 if (vfs_busy(mp)) 303 continue; 304 vput(dvp); 305 error = VFS_ROOT(mp, LK_EXCLUSIVE, &tdvp); 306 vfs_unbusy(mp); 307 if (error) { 308 return (error); 309 } 310 dvp = tdvp; 311 } 312 313 *vpp = dvp; 314 return (0); 315 } 316 317 int 318 union_lookup(void *v) 319 { 320 struct vop_lookup_v2_args /* { 321 struct vnodeop_desc *a_desc; 322 struct vnode *a_dvp; 323 struct vnode **a_vpp; 324 struct componentname *a_cnp; 325 } */ *ap = v; 326 int error; 327 int uerror, lerror; 328 struct vnode *uppervp, *lowervp; 329 struct vnode *upperdvp, *lowerdvp; 330 struct vnode *dvp = ap->a_dvp; 331 struct union_node *dun = VTOUNION(dvp); 332 struct componentname *cnp = ap->a_cnp; 333 struct union_mount *um = MOUNTTOUNIONMOUNT(dvp->v_mount); 334 kauth_cred_t saved_cred = NULL; 335 int iswhiteout; 336 struct vattr va; 337 338 #ifdef notyet 339 if (cnp->cn_namelen == 3 && 340 cnp->cn_nameptr[2] == '.' && 341 cnp->cn_nameptr[1] == '.' && 342 cnp->cn_nameptr[0] == '.') { 343 dvp = *ap->a_vpp = LOWERVP(ap->a_dvp); 344 if (dvp == NULLVP) 345 return (ENOENT); 346 vref(dvp); 347 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 348 return (0); 349 } 350 #endif 351 352 if ((cnp->cn_flags & ISLASTCN) && 353 (dvp->v_mount->mnt_flag & MNT_RDONLY) && 354 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) 355 return (EROFS); 356 357 start: 358 upperdvp = dun->un_uppervp; 359 lowerdvp = dun->un_lowervp; 360 uppervp = NULLVP; 361 lowervp = NULLVP; 362 iswhiteout = 0; 363 364 /* 365 * do the lookup in the upper level. 366 * if that level comsumes additional pathnames, 367 * then assume that something special is going 368 * on and just return that vnode. 369 */ 370 if (upperdvp != NULLVP) { 371 uerror = union_lookup1(um->um_uppervp, &upperdvp, 372 &uppervp, cnp); 373 if (uerror == ENOENT || uerror == EJUSTRETURN) { 374 if (cnp->cn_flags & ISWHITEOUT) { 375 iswhiteout = 1; 376 } else if (lowerdvp != NULLVP) { 377 lerror = VOP_GETATTR(upperdvp, &va, 378 cnp->cn_cred); 379 if (lerror == 0 && (va.va_flags & OPAQUE)) 380 iswhiteout = 1; 381 } 382 } 383 } else { 384 uerror = ENOENT; 385 } 386 387 /* 388 * in a similar way to the upper layer, do the lookup 389 * in the lower layer. this time, if there is some 390 * component magic going on, then vput whatever we got 391 * back from the upper layer and return the lower vnode 392 * instead. 393 */ 394 if (lowerdvp != NULLVP && !iswhiteout) { 395 int nameiop; 396 397 vn_lock(lowerdvp, LK_EXCLUSIVE | LK_RETRY); 398 399 /* 400 * Only do a LOOKUP on the bottom node, since 401 * we won't be making changes to it anyway. 402 */ 403 nameiop = cnp->cn_nameiop; 404 cnp->cn_nameiop = LOOKUP; 405 if (um->um_op == UNMNT_BELOW) { 406 saved_cred = cnp->cn_cred; 407 cnp->cn_cred = um->um_cred; 408 } 409 410 /* 411 * we shouldn't have to worry about locking interactions 412 * between the lower layer and our union layer (w.r.t. 413 * `..' processing) because we don't futz with lowervp 414 * locks in the union-node instantiation code path. 415 */ 416 lerror = union_lookup1(um->um_lowervp, &lowerdvp, 417 &lowervp, cnp); 418 if (um->um_op == UNMNT_BELOW) 419 cnp->cn_cred = saved_cred; 420 cnp->cn_nameiop = nameiop; 421 422 if (lowervp != lowerdvp) 423 VOP_UNLOCK(lowerdvp); 424 } else { 425 lerror = ENOENT; 426 if ((cnp->cn_flags & ISDOTDOT) && dun->un_pvp != NULLVP) { 427 lowervp = LOWERVP(dun->un_pvp); 428 if (lowervp != NULLVP) { 429 vref(lowervp); 430 vn_lock(lowervp, LK_EXCLUSIVE | LK_RETRY); 431 lerror = 0; 432 } 433 } 434 } 435 436 /* 437 * EJUSTRETURN is used by underlying filesystems to indicate that 438 * a directory modification op was started successfully. 439 * This will only happen in the upper layer, since 440 * the lower layer only does LOOKUPs. 441 * If this union is mounted read-only, bounce it now. 442 */ 443 444 if ((uerror == EJUSTRETURN) && (cnp->cn_flags & ISLASTCN) && 445 (dvp->v_mount->mnt_flag & MNT_RDONLY) && 446 ((cnp->cn_nameiop == CREATE) || (cnp->cn_nameiop == RENAME))) 447 uerror = EROFS; 448 449 /* 450 * at this point, we have uerror and lerror indicating 451 * possible errors with the lookups in the upper and lower 452 * layers. additionally, uppervp and lowervp are (locked) 453 * references to existing vnodes in the upper and lower layers. 454 * 455 * there are now three cases to consider. 456 * 1. if both layers returned an error, then return whatever 457 * error the upper layer generated. 458 * 459 * 2. if the top layer failed and the bottom layer succeeded 460 * then two subcases occur. 461 * a. the bottom vnode is not a directory, in which 462 * case just return a new union vnode referencing 463 * an empty top layer and the existing bottom layer. 464 * b. the bottom vnode is a directory, in which case 465 * create a new directory in the top-level and 466 * continue as in case 3. 467 * 468 * 3. if the top layer succeeded then return a new union 469 * vnode referencing whatever the new top layer and 470 * whatever the bottom layer returned. 471 */ 472 473 *ap->a_vpp = NULLVP; 474 475 476 /* case 1. */ 477 if ((uerror != 0) && (lerror != 0)) { 478 return (uerror); 479 } 480 481 /* case 2. */ 482 if (uerror != 0 /* && (lerror == 0) */ ) { 483 if (lowervp->v_type == VDIR) { /* case 2b. */ 484 /* 485 * We may be racing another process to make the 486 * upper-level shadow directory. Be careful with 487 * locks/etc! 488 * If we have to create a shadow directory and want 489 * to commit the node we have to restart the lookup 490 * to get the componentname right. 491 */ 492 if (upperdvp) { 493 VOP_UNLOCK(upperdvp); 494 uerror = union_mkshadow(um, upperdvp, cnp, 495 &uppervp); 496 vn_lock(upperdvp, LK_EXCLUSIVE | LK_RETRY); 497 if (uerror == 0 && cnp->cn_nameiop != LOOKUP) { 498 vrele(uppervp); 499 if (lowervp != NULLVP) 500 vput(lowervp); 501 goto start; 502 } 503 } 504 if (uerror) { 505 if (lowervp != NULLVP) { 506 vput(lowervp); 507 lowervp = NULLVP; 508 } 509 return (uerror); 510 } 511 } 512 } else { /* uerror == 0 */ 513 if (uppervp != upperdvp) 514 VOP_UNLOCK(uppervp); 515 } 516 517 if (lowervp != NULLVP) 518 VOP_UNLOCK(lowervp); 519 520 error = union_allocvp(ap->a_vpp, dvp->v_mount, dvp, upperdvp, cnp, 521 uppervp, lowervp, 1); 522 523 if (error) { 524 if (uppervp != NULLVP) 525 vrele(uppervp); 526 if (lowervp != NULLVP) 527 vrele(lowervp); 528 return error; 529 } 530 531 return 0; 532 } 533 534 int 535 union_create(void *v) 536 { 537 struct vop_create_v3_args /* { 538 struct vnode *a_dvp; 539 struct vnode **a_vpp; 540 struct componentname *a_cnp; 541 struct vattr *a_vap; 542 } */ *ap = v; 543 struct union_node *un = VTOUNION(ap->a_dvp); 544 struct vnode *dvp = un->un_uppervp; 545 struct componentname *cnp = ap->a_cnp; 546 547 if (dvp != NULLVP) { 548 int error; 549 struct vnode *vp; 550 struct mount *mp; 551 552 mp = ap->a_dvp->v_mount; 553 554 vp = NULL; 555 error = VOP_CREATE(dvp, &vp, cnp, ap->a_vap); 556 if (error) 557 return (error); 558 559 error = union_allocvp(ap->a_vpp, mp, NULLVP, NULLVP, cnp, vp, 560 NULLVP, 1); 561 if (error) 562 vrele(vp); 563 return (error); 564 } 565 566 return (EROFS); 567 } 568 569 int 570 union_whiteout(void *v) 571 { 572 struct vop_whiteout_args /* { 573 struct vnode *a_dvp; 574 struct componentname *a_cnp; 575 int a_flags; 576 } */ *ap = v; 577 struct union_node *un = VTOUNION(ap->a_dvp); 578 struct componentname *cnp = ap->a_cnp; 579 580 if (un->un_uppervp == NULLVP) 581 return (EOPNOTSUPP); 582 583 return (VOP_WHITEOUT(un->un_uppervp, cnp, ap->a_flags)); 584 } 585 586 int 587 union_mknod(void *v) 588 { 589 struct vop_mknod_v3_args /* { 590 struct vnode *a_dvp; 591 struct vnode **a_vpp; 592 struct componentname *a_cnp; 593 struct vattr *a_vap; 594 } */ *ap = v; 595 struct union_node *un = VTOUNION(ap->a_dvp); 596 struct vnode *dvp = un->un_uppervp; 597 struct componentname *cnp = ap->a_cnp; 598 599 if (dvp != NULLVP) { 600 int error; 601 struct vnode *vp; 602 struct mount *mp; 603 604 mp = ap->a_dvp->v_mount; 605 error = VOP_MKNOD(dvp, &vp, cnp, ap->a_vap); 606 if (error) 607 return (error); 608 609 error = union_allocvp(ap->a_vpp, mp, NULLVP, NULLVP, 610 cnp, vp, NULLVP, 1); 611 if (error) 612 vrele(vp); 613 return (error); 614 } 615 616 return (EROFS); 617 } 618 619 int 620 union_open(void *v) 621 { 622 struct vop_open_args /* { 623 struct vnodeop_desc *a_desc; 624 struct vnode *a_vp; 625 int a_mode; 626 kauth_cred_t a_cred; 627 } */ *ap = v; 628 struct union_node *un = VTOUNION(ap->a_vp); 629 struct vnode *tvp; 630 int mode = ap->a_mode; 631 kauth_cred_t cred = ap->a_cred; 632 struct lwp *l = curlwp; 633 int error; 634 635 /* 636 * If there is an existing upper vp then simply open that. 637 */ 638 tvp = un->un_uppervp; 639 if (tvp == NULLVP) { 640 /* 641 * If the lower vnode is being opened for writing, then 642 * copy the file contents to the upper vnode and open that, 643 * otherwise can simply open the lower vnode. 644 */ 645 tvp = un->un_lowervp; 646 if ((ap->a_mode & FWRITE) && (tvp->v_type == VREG)) { 647 error = union_copyup(un, (mode&O_TRUNC) == 0, cred, l); 648 if (error == 0) 649 error = VOP_OPEN(un->un_uppervp, mode, cred); 650 if (error == 0) { 651 mutex_enter(un->un_uppervp->v_interlock); 652 un->un_uppervp->v_writecount++; 653 mutex_exit(un->un_uppervp->v_interlock); 654 } 655 return (error); 656 } 657 658 /* 659 * Just open the lower vnode, but check for nodev mount flag 660 */ 661 if ((tvp->v_type == VBLK || tvp->v_type == VCHR) && 662 (ap->a_vp->v_mount->mnt_flag & MNT_NODEV)) 663 return ENXIO; 664 un->un_openl++; 665 vn_lock(tvp, LK_EXCLUSIVE | LK_RETRY); 666 error = VOP_OPEN(tvp, mode, cred); 667 VOP_UNLOCK(tvp); 668 669 return (error); 670 } 671 /* 672 * Just open the upper vnode, checking for nodev mount flag first 673 */ 674 if ((tvp->v_type == VBLK || tvp->v_type == VCHR) && 675 (ap->a_vp->v_mount->mnt_flag & MNT_NODEV)) 676 return ENXIO; 677 678 error = VOP_OPEN(tvp, mode, cred); 679 if (error == 0 && (ap->a_mode & FWRITE)) { 680 mutex_enter(tvp->v_interlock); 681 tvp->v_writecount++; 682 mutex_exit(tvp->v_interlock); 683 } 684 685 return (error); 686 } 687 688 int 689 union_close(void *v) 690 { 691 struct vop_close_args /* { 692 struct vnode *a_vp; 693 int a_fflag; 694 kauth_cred_t a_cred; 695 } */ *ap = v; 696 struct union_node *un = VTOUNION(ap->a_vp); 697 struct vnode *vp; 698 int error; 699 bool do_lock; 700 701 vp = un->un_uppervp; 702 if (vp != NULLVP) { 703 do_lock = false; 704 } else { 705 KASSERT(un->un_openl > 0); 706 --un->un_openl; 707 vp = un->un_lowervp; 708 do_lock = true; 709 } 710 711 KASSERT(vp != NULLVP); 712 ap->a_vp = vp; 713 if ((ap->a_fflag & FWRITE)) { 714 KASSERT(vp == un->un_uppervp); 715 mutex_enter(vp->v_interlock); 716 vp->v_writecount--; 717 mutex_exit(vp->v_interlock); 718 } 719 if (do_lock) 720 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 721 error = VCALL(vp, VOFFSET(vop_close), ap); 722 if (do_lock) 723 VOP_UNLOCK(vp); 724 725 return error; 726 } 727 728 /* 729 * Check access permission on the union vnode. 730 * The access check being enforced is to check 731 * against both the underlying vnode, and any 732 * copied vnode. This ensures that no additional 733 * file permissions are given away simply because 734 * the user caused an implicit file copy. 735 */ 736 int 737 union_access(void *v) 738 { 739 struct vop_access_args /* { 740 struct vnodeop_desc *a_desc; 741 struct vnode *a_vp; 742 accmode_t a_accmode; 743 kauth_cred_t a_cred; 744 } */ *ap = v; 745 struct vnode *vp = ap->a_vp; 746 struct union_node *un = VTOUNION(vp); 747 int error = EACCES; 748 struct union_mount *um = MOUNTTOUNIONMOUNT(vp->v_mount); 749 750 /* 751 * Disallow write attempts on read-only file systems; 752 * unless the file is a socket, fifo, or a block or 753 * character device resident on the file system. 754 */ 755 if (ap->a_accmode & VWRITE) { 756 switch (vp->v_type) { 757 case VDIR: 758 case VLNK: 759 case VREG: 760 if (vp->v_mount->mnt_flag & MNT_RDONLY) 761 return (EROFS); 762 break; 763 case VBAD: 764 case VBLK: 765 case VCHR: 766 case VSOCK: 767 case VFIFO: 768 case VNON: 769 default: 770 break; 771 } 772 } 773 774 /* 775 * Copy up to prevent checking (and failing) against 776 * underlying file system mounted read only. 777 * Check for read access first to prevent implicit 778 * copy of inaccessible underlying vnode. 779 */ 780 if (un->un_uppervp == NULLVP && 781 (un->un_lowervp->v_type == VREG) && 782 (ap->a_accmode & VWRITE)) { 783 vn_lock(un->un_lowervp, LK_EXCLUSIVE | LK_RETRY); 784 error = VOP_ACCESS(un->un_lowervp, VREAD, ap->a_cred); 785 VOP_UNLOCK(un->un_lowervp); 786 if (error == 0) 787 error = union_copyup(un, 1, ap->a_cred, curlwp); 788 if (error) 789 return error; 790 } 791 792 if ((vp = un->un_uppervp) != NULLVP) { 793 ap->a_vp = vp; 794 return (VCALL(vp, VOFFSET(vop_access), ap)); 795 } 796 797 if ((vp = un->un_lowervp) != NULLVP) { 798 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 799 ap->a_vp = vp; 800 error = VCALL(vp, VOFFSET(vop_access), ap); 801 if (error == 0) { 802 if (um->um_op == UNMNT_BELOW) { 803 ap->a_cred = um->um_cred; 804 error = VCALL(vp, VOFFSET(vop_access), ap); 805 } 806 } 807 VOP_UNLOCK(vp); 808 if (error) 809 return (error); 810 } 811 812 return (error); 813 } 814 815 /* 816 * We handle getattr only to change the fsid and 817 * track object sizes 818 */ 819 int 820 union_getattr(void *v) 821 { 822 struct vop_getattr_args /* { 823 struct vnode *a_vp; 824 struct vattr *a_vap; 825 kauth_cred_t a_cred; 826 } */ *ap = v; 827 int error; 828 struct union_node *un = VTOUNION(ap->a_vp); 829 struct vnode *vp = un->un_uppervp; 830 struct vattr *vap; 831 struct vattr va; 832 833 834 /* 835 * Some programs walk the filesystem hierarchy by counting 836 * links to directories to avoid stat'ing all the time. 837 * This means the link count on directories needs to be "correct". 838 * The only way to do that is to call getattr on both layers 839 * and fix up the link count. The link count will not necessarily 840 * be accurate but will be large enough to defeat the tree walkers. 841 * 842 * To make life more interesting, some filesystems don't keep 843 * track of link counts in the expected way, and return a 844 * link count of `1' for those directories; if either of the 845 * component directories returns a link count of `1', we return a 1. 846 */ 847 848 vap = ap->a_vap; 849 850 vp = un->un_uppervp; 851 if (vp != NULLVP) { 852 error = VOP_GETATTR(vp, vap, ap->a_cred); 853 if (error) 854 return (error); 855 mutex_enter(&un->un_lock); 856 union_newsize(ap->a_vp, vap->va_size, VNOVAL); 857 } 858 859 if (vp == NULLVP) { 860 vp = un->un_lowervp; 861 } else if (vp->v_type == VDIR) { 862 vp = un->un_lowervp; 863 if (vp != NULLVP) 864 vap = &va; 865 } else { 866 vp = NULLVP; 867 } 868 869 if (vp != NULLVP) { 870 if (vp == un->un_lowervp) 871 vn_lock(vp, LK_SHARED | LK_RETRY); 872 error = VOP_GETATTR(vp, vap, ap->a_cred); 873 if (vp == un->un_lowervp) 874 VOP_UNLOCK(vp); 875 if (error) 876 return (error); 877 mutex_enter(&un->un_lock); 878 union_newsize(ap->a_vp, VNOVAL, vap->va_size); 879 } 880 881 if ((vap != ap->a_vap) && (vap->va_type == VDIR)) { 882 /* 883 * Link count manipulation: 884 * - If both return "2", return 2 (no subdirs) 885 * - If one or the other return "1", return "1" (ENOCLUE) 886 */ 887 if ((ap->a_vap->va_nlink == 2) && 888 (vap->va_nlink == 2)) 889 ; 890 else if (ap->a_vap->va_nlink != 1) { 891 if (vap->va_nlink == 1) 892 ap->a_vap->va_nlink = 1; 893 else 894 ap->a_vap->va_nlink += vap->va_nlink; 895 } 896 } 897 ap->a_vap->va_fsid = ap->a_vp->v_mount->mnt_stat.f_fsidx.__fsid_val[0]; 898 return (0); 899 } 900 901 int 902 union_setattr(void *v) 903 { 904 struct vop_setattr_args /* { 905 struct vnode *a_vp; 906 struct vattr *a_vap; 907 kauth_cred_t a_cred; 908 } */ *ap = v; 909 struct vattr *vap = ap->a_vap; 910 struct vnode *vp = ap->a_vp; 911 struct union_node *un = VTOUNION(vp); 912 bool size_only; /* All but va_size are VNOVAL. */ 913 int error; 914 915 size_only = (vap->va_flags == VNOVAL && vap->va_uid == (uid_t)VNOVAL && 916 vap->va_gid == (gid_t)VNOVAL && vap->va_atime.tv_sec == VNOVAL && 917 vap->va_mtime.tv_sec == VNOVAL && vap->va_mode == (mode_t)VNOVAL); 918 919 if (!size_only && (vp->v_mount->mnt_flag & MNT_RDONLY)) 920 return (EROFS); 921 if (vap->va_size != VNOVAL) { 922 switch (vp->v_type) { 923 case VDIR: 924 return (EISDIR); 925 case VCHR: 926 case VBLK: 927 case VSOCK: 928 case VFIFO: 929 break; 930 case VREG: 931 case VLNK: 932 default: 933 /* 934 * Disallow write attempts if the filesystem is 935 * mounted read-only. 936 */ 937 if (vp->v_mount->mnt_flag & MNT_RDONLY) 938 return (EROFS); 939 } 940 } 941 942 /* 943 * Handle case of truncating lower object to zero size, 944 * by creating a zero length upper object. This is to 945 * handle the case of open with O_TRUNC and O_CREAT. 946 */ 947 if ((un->un_uppervp == NULLVP) && 948 /* assert(un->un_lowervp != NULLVP) */ 949 (un->un_lowervp->v_type == VREG)) { 950 error = union_copyup(un, (vap->va_size != 0), 951 ap->a_cred, curlwp); 952 if (error) 953 return (error); 954 } 955 956 /* 957 * Try to set attributes in upper layer, ignore size change to zero 958 * for devices to handle O_TRUNC and return read-only filesystem error 959 * otherwise. 960 */ 961 if (un->un_uppervp != NULLVP) { 962 error = VOP_SETATTR(un->un_uppervp, vap, ap->a_cred); 963 if ((error == 0) && (vap->va_size != VNOVAL)) { 964 mutex_enter(&un->un_lock); 965 union_newsize(ap->a_vp, vap->va_size, VNOVAL); 966 } 967 } else { 968 KASSERT(un->un_lowervp != NULLVP); 969 if (NODE_IS_SPECIAL(un->un_lowervp)) { 970 if (size_only && 971 (vap->va_size == 0 || vap->va_size == VNOVAL)) 972 error = 0; 973 else 974 error = EROFS; 975 } else { 976 error = EROFS; 977 } 978 } 979 980 return (error); 981 } 982 983 int 984 union_read(void *v) 985 { 986 struct vop_read_args /* { 987 struct vnode *a_vp; 988 struct uio *a_uio; 989 int a_ioflag; 990 kauth_cred_t a_cred; 991 } */ *ap = v; 992 int error; 993 struct vnode *vp = OTHERVP(ap->a_vp); 994 int dolock = (vp == LOWERVP(ap->a_vp)); 995 996 if (dolock) 997 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 998 error = VOP_READ(vp, ap->a_uio, ap->a_ioflag, ap->a_cred); 999 if (dolock) 1000 VOP_UNLOCK(vp); 1001 1002 /* 1003 * XXX 1004 * perhaps the size of the underlying object has changed under 1005 * our feet. take advantage of the offset information present 1006 * in the uio structure. 1007 */ 1008 if (error == 0) { 1009 struct union_node *un = VTOUNION(ap->a_vp); 1010 off_t cur = ap->a_uio->uio_offset; 1011 off_t usz = VNOVAL, lsz = VNOVAL; 1012 1013 mutex_enter(&un->un_lock); 1014 if (vp == un->un_uppervp) { 1015 if (cur > un->un_uppersz) 1016 usz = cur; 1017 } else { 1018 if (cur > un->un_lowersz) 1019 lsz = cur; 1020 } 1021 1022 if (usz != VNOVAL || lsz != VNOVAL) 1023 union_newsize(ap->a_vp, usz, lsz); 1024 else 1025 mutex_exit(&un->un_lock); 1026 } 1027 1028 return (error); 1029 } 1030 1031 int 1032 union_write(void *v) 1033 { 1034 struct vop_read_args /* { 1035 struct vnode *a_vp; 1036 struct uio *a_uio; 1037 int a_ioflag; 1038 kauth_cred_t a_cred; 1039 } */ *ap = v; 1040 int error; 1041 struct vnode *vp; 1042 struct union_node *un = VTOUNION(ap->a_vp); 1043 1044 vp = UPPERVP(ap->a_vp); 1045 if (vp == NULLVP) { 1046 vp = LOWERVP(ap->a_vp); 1047 if (NODE_IS_SPECIAL(vp)) { 1048 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1049 error = VOP_WRITE(vp, ap->a_uio, ap->a_ioflag, 1050 ap->a_cred); 1051 VOP_UNLOCK(vp); 1052 return error; 1053 } 1054 panic("union: missing upper layer in write"); 1055 } 1056 1057 error = VOP_WRITE(vp, ap->a_uio, ap->a_ioflag, ap->a_cred); 1058 1059 /* 1060 * the size of the underlying object may be changed by the 1061 * write. 1062 */ 1063 if (error == 0) { 1064 off_t cur = ap->a_uio->uio_offset; 1065 1066 mutex_enter(&un->un_lock); 1067 if (cur > un->un_uppersz) 1068 union_newsize(ap->a_vp, cur, VNOVAL); 1069 else 1070 mutex_exit(&un->un_lock); 1071 } 1072 1073 return (error); 1074 } 1075 1076 int 1077 union_ioctl(void *v) 1078 { 1079 struct vop_ioctl_args /* { 1080 struct vnode *a_vp; 1081 int a_command; 1082 void *a_data; 1083 int a_fflag; 1084 kauth_cred_t a_cred; 1085 } */ *ap = v; 1086 struct vnode *ovp = OTHERVP(ap->a_vp); 1087 1088 ap->a_vp = ovp; 1089 return (VCALL(ovp, VOFFSET(vop_ioctl), ap)); 1090 } 1091 1092 int 1093 union_poll(void *v) 1094 { 1095 struct vop_poll_args /* { 1096 struct vnode *a_vp; 1097 int a_events; 1098 } */ *ap = v; 1099 struct vnode *ovp = OTHERVP(ap->a_vp); 1100 1101 ap->a_vp = ovp; 1102 return (VCALL(ovp, VOFFSET(vop_poll), ap)); 1103 } 1104 1105 int 1106 union_revoke(void *v) 1107 { 1108 struct vop_revoke_args /* { 1109 struct vnode *a_vp; 1110 int a_flags; 1111 struct proc *a_p; 1112 } */ *ap = v; 1113 struct vnode *vp = ap->a_vp; 1114 1115 if (UPPERVP(vp)) 1116 VOP_REVOKE(UPPERVP(vp), ap->a_flags); 1117 if (LOWERVP(vp)) 1118 VOP_REVOKE(LOWERVP(vp), ap->a_flags); 1119 vgone(vp); /* XXXAD?? */ 1120 return (0); 1121 } 1122 1123 int 1124 union_mmap(void *v) 1125 { 1126 struct vop_mmap_args /* { 1127 struct vnode *a_vp; 1128 vm_prot_t a_prot; 1129 kauth_cred_t a_cred; 1130 } */ *ap = v; 1131 struct vnode *ovp = OTHERVP(ap->a_vp); 1132 1133 ap->a_vp = ovp; 1134 return (VCALL(ovp, VOFFSET(vop_mmap), ap)); 1135 } 1136 1137 int 1138 union_fsync(void *v) 1139 { 1140 struct vop_fsync_args /* { 1141 struct vnode *a_vp; 1142 kauth_cred_t a_cred; 1143 int a_flags; 1144 off_t offhi; 1145 off_t offlo; 1146 } */ *ap = v; 1147 int error = 0; 1148 struct vnode *targetvp; 1149 1150 /* 1151 * If vinvalbuf is calling us, it's a "shallow fsync" -- don't 1152 * bother syncing the underlying vnodes, since (a) they'll be 1153 * fsync'ed when reclaimed and (b) we could deadlock if 1154 * they're locked; otherwise, pass it through to the 1155 * underlying layer. 1156 */ 1157 if (ap->a_vp->v_type == VBLK || ap->a_vp->v_type == VCHR) { 1158 error = spec_fsync(v); 1159 if (error) 1160 return error; 1161 } 1162 1163 if (ap->a_flags & FSYNC_RECLAIM) 1164 return 0; 1165 1166 targetvp = OTHERVP(ap->a_vp); 1167 if (targetvp != NULLVP) { 1168 int dolock = (targetvp == LOWERVP(ap->a_vp)); 1169 1170 if (dolock) 1171 vn_lock(targetvp, LK_EXCLUSIVE | LK_RETRY); 1172 error = VOP_FSYNC(targetvp, ap->a_cred, ap->a_flags, 1173 ap->a_offlo, ap->a_offhi); 1174 if (dolock) 1175 VOP_UNLOCK(targetvp); 1176 } 1177 1178 return (error); 1179 } 1180 1181 int 1182 union_seek(void *v) 1183 { 1184 struct vop_seek_args /* { 1185 struct vnode *a_vp; 1186 off_t a_oldoff; 1187 off_t a_newoff; 1188 kauth_cred_t a_cred; 1189 } */ *ap = v; 1190 struct vnode *ovp = OTHERVP(ap->a_vp); 1191 1192 ap->a_vp = ovp; 1193 return (VCALL(ovp, VOFFSET(vop_seek), ap)); 1194 } 1195 1196 int 1197 union_remove(void *v) 1198 { 1199 struct vop_remove_v3_args /* { 1200 struct vnode *a_dvp; 1201 struct vnode *a_vp; 1202 struct componentname *a_cnp; 1203 nlink_t ctx_vp_new_nlink; 1204 } */ *ap = v; 1205 int error; 1206 struct union_node *dun = VTOUNION(ap->a_dvp); 1207 struct union_node *un = VTOUNION(ap->a_vp); 1208 struct componentname *cnp = ap->a_cnp; 1209 1210 if (dun->un_uppervp == NULLVP) 1211 panic("union remove: null upper vnode"); 1212 1213 if (un->un_uppervp != NULLVP) { 1214 struct vnode *dvp = dun->un_uppervp; 1215 struct vnode *vp = un->un_uppervp; 1216 1217 /* Account for VOP_REMOVE to vrele vp. */ 1218 vref(vp); 1219 if (union_dowhiteout(un, cnp->cn_cred)) 1220 cnp->cn_flags |= DOWHITEOUT; 1221 error = VOP_REMOVE(dvp, vp, cnp); 1222 if (!error) 1223 union_removed_upper(un); 1224 vrele(ap->a_vp); 1225 } else { 1226 error = union_mkwhiteout( 1227 MOUNTTOUNIONMOUNT(UNIONTOV(dun)->v_mount), 1228 dun->un_uppervp, ap->a_cnp, un); 1229 vput(ap->a_vp); 1230 } 1231 1232 return (error); 1233 } 1234 1235 int 1236 union_link(void *v) 1237 { 1238 struct vop_link_v2_args /* { 1239 struct vnode *a_dvp; 1240 struct vnode *a_vp; 1241 struct componentname *a_cnp; 1242 } */ *ap = v; 1243 int error = 0; 1244 struct componentname *cnp = ap->a_cnp; 1245 struct union_node *dun; 1246 struct vnode *vp; 1247 struct vnode *dvp; 1248 1249 dun = VTOUNION(ap->a_dvp); 1250 1251 KASSERT((ap->a_cnp->cn_flags & LOCKPARENT) != 0); 1252 1253 if (ap->a_dvp->v_op != ap->a_vp->v_op) { 1254 vp = ap->a_vp; 1255 } else { 1256 struct union_node *un = VTOUNION(ap->a_vp); 1257 if (un->un_uppervp == NULLVP) { 1258 const bool droplock = (dun->un_uppervp == un->un_dirvp); 1259 1260 /* 1261 * Needs to be copied before we can link it. 1262 */ 1263 vn_lock(ap->a_vp, LK_EXCLUSIVE | LK_RETRY); 1264 if (droplock) 1265 VOP_UNLOCK(dun->un_uppervp); 1266 error = union_copyup(un, 1, cnp->cn_cred, curlwp); 1267 /* 1268 * Unlock the target vnode again immediately 1269 * to avoid deadlocking while relocking or in 1270 * relookup. In particular it's common for a 1271 * link operation to be entirely in one dir, 1272 * where ap->a_dvp and/or dun->un_uppervp are 1273 * the parent of ap->a_vp; locking the former 1274 * while holding the latter is an order 1275 * reversal. It's also possible for the 1276 * relookup call to find and lock the same 1277 * vnode if another process just did the same 1278 * link operation. 1279 */ 1280 VOP_UNLOCK(ap->a_vp); 1281 if (droplock) { 1282 vn_lock(dun->un_uppervp, 1283 LK_EXCLUSIVE | LK_RETRY); 1284 /* 1285 * During copyup, we dropped the lock on the 1286 * dir and invalidated any saved namei lookup 1287 * state for the directory we'll be entering 1288 * the link in. We need to re-run the lookup 1289 * in that directory to reset any state needed 1290 * for VOP_LINK. 1291 * Call relookup on the union-layer to reset 1292 * the state. 1293 */ 1294 vp = NULLVP; 1295 if (dun->un_uppervp == NULLVP) 1296 panic("union: null upperdvp?"); 1297 error = relookup(ap->a_dvp, &vp, ap->a_cnp, 0); 1298 if (error) { 1299 return EROFS; /* ? */ 1300 } 1301 if (vp != NULLVP) { 1302 /* 1303 * The name we wanted to 1304 * create has been created by 1305 * another process. 1306 */ 1307 vput(vp); 1308 return EEXIST; 1309 } 1310 } 1311 } 1312 vp = un->un_uppervp; 1313 } 1314 1315 dvp = dun->un_uppervp; 1316 if (dvp == NULLVP) 1317 error = EROFS; 1318 1319 if (error) 1320 return (error); 1321 1322 return VOP_LINK(dvp, vp, cnp); 1323 } 1324 1325 int 1326 union_rename(void *v) 1327 { 1328 struct vop_rename_args /* { 1329 struct vnode *a_fdvp; 1330 struct vnode *a_fvp; 1331 struct componentname *a_fcnp; 1332 struct vnode *a_tdvp; 1333 struct vnode *a_tvp; 1334 struct componentname *a_tcnp; 1335 } */ *ap = v; 1336 int error; 1337 1338 struct vnode *fdvp = ap->a_fdvp; 1339 struct vnode *fvp = ap->a_fvp; 1340 struct vnode *tdvp = ap->a_tdvp; 1341 struct vnode *tvp = ap->a_tvp; 1342 1343 /* 1344 * Account for VOP_RENAME to vrele all nodes. 1345 * Note: VOP_RENAME will unlock tdvp. 1346 */ 1347 1348 if (fdvp->v_op == union_vnodeop_p) { /* always true */ 1349 struct union_node *un = VTOUNION(fdvp); 1350 if (un->un_uppervp == NULLVP) { 1351 /* 1352 * this should never happen in normal 1353 * operation but might if there was 1354 * a problem creating the top-level shadow 1355 * directory. 1356 */ 1357 error = EXDEV; 1358 goto bad; 1359 } 1360 1361 fdvp = un->un_uppervp; 1362 vref(fdvp); 1363 } 1364 1365 if (fvp->v_op == union_vnodeop_p) { /* always true */ 1366 struct union_node *un = VTOUNION(fvp); 1367 if (un->un_uppervp == NULLVP) { 1368 /* XXX: should do a copyup */ 1369 error = EXDEV; 1370 goto bad; 1371 } 1372 1373 if (un->un_lowervp != NULLVP) 1374 ap->a_fcnp->cn_flags |= DOWHITEOUT; 1375 1376 fvp = un->un_uppervp; 1377 vref(fvp); 1378 } 1379 1380 if (tdvp->v_op == union_vnodeop_p) { 1381 struct union_node *un = VTOUNION(tdvp); 1382 if (un->un_uppervp == NULLVP) { 1383 /* 1384 * this should never happen in normal 1385 * operation but might if there was 1386 * a problem creating the top-level shadow 1387 * directory. 1388 */ 1389 error = EXDEV; 1390 goto bad; 1391 } 1392 1393 tdvp = un->un_uppervp; 1394 vref(tdvp); 1395 } 1396 1397 if (tvp != NULLVP && tvp->v_op == union_vnodeop_p) { 1398 struct union_node *un = VTOUNION(tvp); 1399 1400 tvp = un->un_uppervp; 1401 if (tvp != NULLVP) { 1402 vref(tvp); 1403 } 1404 } 1405 1406 error = VOP_RENAME(fdvp, fvp, ap->a_fcnp, tdvp, tvp, ap->a_tcnp); 1407 goto out; 1408 1409 bad: 1410 vput(tdvp); 1411 if (tvp != NULLVP) 1412 vput(tvp); 1413 vrele(fdvp); 1414 vrele(fvp); 1415 1416 out: 1417 if (fdvp != ap->a_fdvp) { 1418 vrele(ap->a_fdvp); 1419 } 1420 if (fvp != ap->a_fvp) { 1421 vrele(ap->a_fvp); 1422 } 1423 if (tdvp != ap->a_tdvp) { 1424 vrele(ap->a_tdvp); 1425 } 1426 if (tvp != ap->a_tvp) { 1427 vrele(ap->a_tvp); 1428 } 1429 return (error); 1430 } 1431 1432 int 1433 union_mkdir(void *v) 1434 { 1435 struct vop_mkdir_v3_args /* { 1436 struct vnode *a_dvp; 1437 struct vnode **a_vpp; 1438 struct componentname *a_cnp; 1439 struct vattr *a_vap; 1440 } */ *ap = v; 1441 struct union_node *un = VTOUNION(ap->a_dvp); 1442 struct vnode *dvp = un->un_uppervp; 1443 struct componentname *cnp = ap->a_cnp; 1444 1445 if (dvp != NULLVP) { 1446 int error; 1447 struct vnode *vp; 1448 1449 vp = NULL; 1450 error = VOP_MKDIR(dvp, &vp, cnp, ap->a_vap); 1451 if (error) { 1452 vrele(ap->a_dvp); 1453 return (error); 1454 } 1455 1456 error = union_allocvp(ap->a_vpp, ap->a_dvp->v_mount, ap->a_dvp, 1457 NULLVP, cnp, vp, NULLVP, 1); 1458 if (error) 1459 vrele(vp); 1460 return (error); 1461 } 1462 1463 return (EROFS); 1464 } 1465 1466 int 1467 union_rmdir(void *v) 1468 { 1469 struct vop_rmdir_v2_args /* { 1470 struct vnode *a_dvp; 1471 struct vnode *a_vp; 1472 struct componentname *a_cnp; 1473 } */ *ap = v; 1474 int error; 1475 struct union_node *dun = VTOUNION(ap->a_dvp); 1476 struct union_node *un = VTOUNION(ap->a_vp); 1477 struct componentname *cnp = ap->a_cnp; 1478 1479 if (dun->un_uppervp == NULLVP) 1480 panic("union rmdir: null upper vnode"); 1481 1482 error = union_check_rmdir(un, cnp->cn_cred); 1483 if (error) { 1484 vput(ap->a_vp); 1485 return error; 1486 } 1487 1488 if (un->un_uppervp != NULLVP) { 1489 struct vnode *dvp = dun->un_uppervp; 1490 struct vnode *vp = un->un_uppervp; 1491 1492 /* Account for VOP_RMDIR to vrele vp. */ 1493 vref(vp); 1494 if (union_dowhiteout(un, cnp->cn_cred)) 1495 cnp->cn_flags |= DOWHITEOUT; 1496 error = VOP_RMDIR(dvp, vp, ap->a_cnp); 1497 if (!error) 1498 union_removed_upper(un); 1499 vrele(ap->a_vp); 1500 } else { 1501 error = union_mkwhiteout( 1502 MOUNTTOUNIONMOUNT(UNIONTOV(dun)->v_mount), 1503 dun->un_uppervp, ap->a_cnp, un); 1504 vput(ap->a_vp); 1505 } 1506 1507 return (error); 1508 } 1509 1510 int 1511 union_symlink(void *v) 1512 { 1513 struct vop_symlink_v3_args /* { 1514 struct vnode *a_dvp; 1515 struct vnode **a_vpp; 1516 struct componentname *a_cnp; 1517 struct vattr *a_vap; 1518 char *a_target; 1519 } */ *ap = v; 1520 struct union_node *un = VTOUNION(ap->a_dvp); 1521 struct vnode *dvp = un->un_uppervp; 1522 struct componentname *cnp = ap->a_cnp; 1523 1524 if (dvp != NULLVP) { 1525 int error; 1526 1527 error = VOP_SYMLINK(dvp, ap->a_vpp, cnp, ap->a_vap, 1528 ap->a_target); 1529 return (error); 1530 } 1531 1532 return (EROFS); 1533 } 1534 1535 /* 1536 * union_readdir works in concert with getdirentries and 1537 * readdir(3) to provide a list of entries in the unioned 1538 * directories. getdirentries is responsible for walking 1539 * down the union stack. readdir(3) is responsible for 1540 * eliminating duplicate names from the returned data stream. 1541 */ 1542 int 1543 union_readdir(void *v) 1544 { 1545 struct vop_readdir_args /* { 1546 struct vnodeop_desc *a_desc; 1547 struct vnode *a_vp; 1548 struct uio *a_uio; 1549 kauth_cred_t a_cred; 1550 int *a_eofflag; 1551 u_long *a_cookies; 1552 int a_ncookies; 1553 } */ *ap = v; 1554 struct union_node *un = VTOUNION(ap->a_vp); 1555 struct vnode *vp; 1556 int dolock, error; 1557 1558 if (un->un_hooknode) { 1559 KASSERT(un->un_uppervp == NULLVP); 1560 KASSERT(un->un_lowervp != NULLVP); 1561 vp = un->un_lowervp; 1562 dolock = 1; 1563 } else { 1564 vp = un->un_uppervp; 1565 dolock = 0; 1566 } 1567 if (vp == NULLVP) 1568 return 0; 1569 1570 if (dolock) 1571 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1572 ap->a_vp = vp; 1573 error = VCALL(vp, VOFFSET(vop_readdir), ap); 1574 if (dolock) 1575 VOP_UNLOCK(vp); 1576 1577 return error; 1578 } 1579 1580 int 1581 union_readlink(void *v) 1582 { 1583 struct vop_readlink_args /* { 1584 struct vnode *a_vp; 1585 struct uio *a_uio; 1586 kauth_cred_t a_cred; 1587 } */ *ap = v; 1588 int error; 1589 struct vnode *vp = OTHERVP(ap->a_vp); 1590 int dolock = (vp == LOWERVP(ap->a_vp)); 1591 1592 if (dolock) 1593 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1594 ap->a_vp = vp; 1595 error = VCALL(vp, VOFFSET(vop_readlink), ap); 1596 if (dolock) 1597 VOP_UNLOCK(vp); 1598 1599 return (error); 1600 } 1601 1602 int 1603 union_abortop(void *v) 1604 { 1605 struct vop_abortop_args /* { 1606 struct vnode *a_dvp; 1607 struct componentname *a_cnp; 1608 } */ *ap = v; 1609 1610 KASSERT(UPPERVP(ap->a_dvp) != NULL); 1611 1612 ap->a_dvp = UPPERVP(ap->a_dvp); 1613 return VCALL(ap->a_dvp, VOFFSET(vop_abortop), ap); 1614 } 1615 1616 int 1617 union_inactive(void *v) 1618 { 1619 struct vop_inactive_v2_args /* { 1620 const struct vnodeop_desc *a_desc; 1621 struct vnode *a_vp; 1622 bool *a_recycle; 1623 } */ *ap = v; 1624 struct vnode *vp = ap->a_vp; 1625 struct union_node *un = VTOUNION(vp); 1626 struct vnode **vpp; 1627 1628 /* 1629 * Do nothing (and _don't_ bypass). 1630 * Wait to vrele lowervp until reclaim, 1631 * so that until then our union_node is in the 1632 * cache and reusable. 1633 * 1634 * NEEDSWORK: Someday, consider inactive'ing 1635 * the lowervp and then trying to reactivate it 1636 * with capabilities (v_id) 1637 * like they do in the name lookup cache code. 1638 * That's too much work for now. 1639 */ 1640 1641 if (un->un_dircache != 0) { 1642 for (vpp = un->un_dircache; *vpp != NULLVP; vpp++) 1643 vrele(*vpp); 1644 free(un->un_dircache, M_TEMP); 1645 un->un_dircache = 0; 1646 } 1647 1648 *ap->a_recycle = ((un->un_cflags & UN_CACHED) == 0); 1649 1650 return (0); 1651 } 1652 1653 int 1654 union_reclaim(void *v) 1655 { 1656 struct vop_reclaim_v2_args /* { 1657 struct vnode *a_vp; 1658 } */ *ap = v; 1659 struct vnode *vp = ap->a_vp; 1660 struct vnode *uvp = UPPERVP(vp); 1661 1662 VOP_UNLOCK(vp); 1663 1664 if (uvp != NULL) { 1665 mutex_enter(uvp->v_interlock); 1666 KASSERT(vp->v_interlock == uvp->v_interlock); 1667 uvp->v_writecount -= vp->v_writecount; 1668 mutex_exit(uvp->v_interlock); 1669 } 1670 1671 union_freevp(vp); 1672 1673 return (0); 1674 } 1675 1676 static int 1677 union_lock1(struct vnode *vp, struct vnode *lockvp, int flags) 1678 { 1679 struct vop_lock_args ap; 1680 1681 ap.a_desc = VDESC(vop_lock); 1682 ap.a_vp = lockvp; 1683 ap.a_flags = flags; 1684 1685 if (lockvp == vp) 1686 return genfs_lock(&ap); 1687 else 1688 return VCALL(ap.a_vp, VOFFSET(vop_lock), &ap); 1689 } 1690 1691 static int 1692 union_unlock1(struct vnode *vp, struct vnode *lockvp) 1693 { 1694 struct vop_unlock_args ap; 1695 1696 ap.a_desc = VDESC(vop_unlock); 1697 ap.a_vp = lockvp; 1698 1699 if (lockvp == vp) 1700 return genfs_unlock(&ap); 1701 else 1702 return VCALL(ap.a_vp, VOFFSET(vop_unlock), &ap); 1703 } 1704 1705 int 1706 union_lock(void *v) 1707 { 1708 struct vop_lock_args /* { 1709 struct vnode *a_vp; 1710 int a_flags; 1711 } */ *ap = v; 1712 struct vnode *vp = ap->a_vp, *lockvp; 1713 struct union_node *un = VTOUNION(vp); 1714 int flags = ap->a_flags; 1715 int error; 1716 1717 if ((flags & LK_NOWAIT) != 0) { 1718 if (!mutex_tryenter(&un->un_lock)) 1719 return EBUSY; 1720 lockvp = LOCKVP(vp); 1721 error = union_lock1(vp, lockvp, flags); 1722 mutex_exit(&un->un_lock); 1723 return error; 1724 } 1725 1726 mutex_enter(&un->un_lock); 1727 for (;;) { 1728 lockvp = LOCKVP(vp); 1729 mutex_exit(&un->un_lock); 1730 error = union_lock1(vp, lockvp, flags); 1731 if (error != 0 || (flags & (LK_DOWNGRADE | LK_UPGRADE)) != 0) 1732 return error; 1733 mutex_enter(&un->un_lock); 1734 if (lockvp == LOCKVP(vp)) 1735 break; 1736 union_unlock1(vp, lockvp); 1737 } 1738 mutex_exit(&un->un_lock); 1739 1740 return error; 1741 } 1742 1743 int 1744 union_unlock(void *v) 1745 { 1746 struct vop_unlock_args /* { 1747 struct vnode *a_vp; 1748 int a_flags; 1749 } */ *ap = v; 1750 struct vnode *vp = ap->a_vp, *lockvp; 1751 1752 lockvp = LOCKVP(vp); 1753 union_unlock1(vp, lockvp); 1754 1755 return 0; 1756 } 1757 1758 int 1759 union_bmap(void *v) 1760 { 1761 struct vop_bmap_args /* { 1762 struct vnode *a_vp; 1763 daddr_t a_bn; 1764 struct vnode **a_vpp; 1765 daddr_t *a_bnp; 1766 int *a_runp; 1767 } */ *ap = v; 1768 int error; 1769 struct vnode *vp = OTHERVP(ap->a_vp); 1770 int dolock = (vp == LOWERVP(ap->a_vp)); 1771 1772 if (dolock) 1773 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1774 ap->a_vp = vp; 1775 error = VCALL(vp, VOFFSET(vop_bmap), ap); 1776 if (dolock) 1777 VOP_UNLOCK(vp); 1778 1779 return (error); 1780 } 1781 1782 int 1783 union_print(void *v) 1784 { 1785 struct vop_print_args /* { 1786 struct vnode *a_vp; 1787 } */ *ap = v; 1788 struct vnode *vp = ap->a_vp; 1789 1790 printf("\ttag VT_UNION, vp=%p, uppervp=%p, lowervp=%p\n", 1791 vp, UPPERVP(vp), LOWERVP(vp)); 1792 if (UPPERVP(vp) != NULLVP) 1793 vprint("union: upper", UPPERVP(vp)); 1794 if (LOWERVP(vp) != NULLVP) 1795 vprint("union: lower", LOWERVP(vp)); 1796 if (VTOUNION(vp)->un_dircache) { 1797 struct vnode **vpp; 1798 for (vpp = VTOUNION(vp)->un_dircache; *vpp != NULLVP; vpp++) 1799 vprint("dircache:", *vpp); 1800 } 1801 1802 return (0); 1803 } 1804 1805 int 1806 union_islocked(void *v) 1807 { 1808 struct vop_islocked_args /* { 1809 struct vnode *a_vp; 1810 } */ *ap = v; 1811 struct vnode *vp; 1812 struct union_node *un; 1813 1814 un = VTOUNION(ap->a_vp); 1815 mutex_enter(&un->un_lock); 1816 vp = LOCKVP(ap->a_vp); 1817 mutex_exit(&un->un_lock); 1818 1819 if (vp == ap->a_vp) 1820 return genfs_islocked(ap); 1821 else 1822 return VOP_ISLOCKED(vp); 1823 } 1824 1825 int 1826 union_pathconf(void *v) 1827 { 1828 struct vop_pathconf_args /* { 1829 struct vnode *a_vp; 1830 int a_name; 1831 int *a_retval; 1832 } */ *ap = v; 1833 int error; 1834 struct vnode *vp = OTHERVP(ap->a_vp); 1835 int dolock = (vp == LOWERVP(ap->a_vp)); 1836 1837 if (dolock) 1838 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1839 ap->a_vp = vp; 1840 error = VCALL(vp, VOFFSET(vop_pathconf), ap); 1841 if (dolock) 1842 VOP_UNLOCK(vp); 1843 1844 return (error); 1845 } 1846 1847 int 1848 union_advlock(void *v) 1849 { 1850 struct vop_advlock_args /* { 1851 struct vnode *a_vp; 1852 void *a_id; 1853 int a_op; 1854 struct flock *a_fl; 1855 int a_flags; 1856 } */ *ap = v; 1857 struct vnode *ovp = OTHERVP(ap->a_vp); 1858 1859 ap->a_vp = ovp; 1860 return (VCALL(ovp, VOFFSET(vop_advlock), ap)); 1861 } 1862 1863 int 1864 union_strategy(void *v) 1865 { 1866 struct vop_strategy_args /* { 1867 struct vnode *a_vp; 1868 struct buf *a_bp; 1869 } */ *ap = v; 1870 struct vnode *ovp = OTHERVP(ap->a_vp); 1871 struct buf *bp = ap->a_bp; 1872 1873 KASSERT(ovp != NULLVP); 1874 if (!NODE_IS_SPECIAL(ovp)) 1875 KASSERT((bp->b_flags & B_READ) || ovp != LOWERVP(bp->b_vp)); 1876 1877 return (VOP_STRATEGY(ovp, bp)); 1878 } 1879 1880 int 1881 union_bwrite(void *v) 1882 { 1883 struct vop_bwrite_args /* { 1884 struct vnode *a_vp; 1885 struct buf *a_bp; 1886 } */ *ap = v; 1887 struct vnode *ovp = OTHERVP(ap->a_vp); 1888 struct buf *bp = ap->a_bp; 1889 1890 KASSERT(ovp != NULLVP); 1891 if (!NODE_IS_SPECIAL(ovp)) 1892 KASSERT((bp->b_flags & B_READ) || ovp != LOWERVP(bp->b_vp)); 1893 1894 return (VOP_BWRITE(ovp, bp)); 1895 } 1896 1897 int 1898 union_getpages(void *v) 1899 { 1900 struct vop_getpages_args /* { 1901 struct vnode *a_vp; 1902 voff_t a_offset; 1903 struct vm_page **a_m; 1904 int *a_count; 1905 int a_centeridx; 1906 vm_prot_t a_access_type; 1907 int a_advice; 1908 int a_flags; 1909 } */ *ap = v; 1910 struct vnode *vp = ap->a_vp; 1911 1912 KASSERT(rw_lock_held(vp->v_uobj.vmobjlock)); 1913 1914 if (ap->a_flags & PGO_LOCKED) { 1915 return EBUSY; 1916 } 1917 ap->a_vp = OTHERVP(vp); 1918 KASSERT(vp->v_uobj.vmobjlock == ap->a_vp->v_uobj.vmobjlock); 1919 1920 /* Just pass the request on to the underlying layer. */ 1921 return VCALL(ap->a_vp, VOFFSET(vop_getpages), ap); 1922 } 1923 1924 int 1925 union_putpages(void *v) 1926 { 1927 struct vop_putpages_args /* { 1928 struct vnode *a_vp; 1929 voff_t a_offlo; 1930 voff_t a_offhi; 1931 int a_flags; 1932 } */ *ap = v; 1933 struct vnode *vp = ap->a_vp; 1934 1935 KASSERT(rw_lock_held(vp->v_uobj.vmobjlock)); 1936 1937 ap->a_vp = OTHERVP(vp); 1938 KASSERT(vp->v_uobj.vmobjlock == ap->a_vp->v_uobj.vmobjlock); 1939 1940 if (ap->a_flags & PGO_RECLAIM) { 1941 rw_exit(vp->v_uobj.vmobjlock); 1942 return 0; 1943 } 1944 1945 /* Just pass the request on to the underlying layer. */ 1946 return VCALL(ap->a_vp, VOFFSET(vop_putpages), ap); 1947 } 1948 1949 int 1950 union_kqfilter(void *v) 1951 { 1952 struct vop_kqfilter_args /* { 1953 struct vnode *a_vp; 1954 struct knote *a_kn; 1955 } */ *ap = v; 1956 int error; 1957 1958 /* 1959 * We watch either the upper layer file (if it already exists), 1960 * or the lower layer one. If there is lower layer file only 1961 * at this moment, we will keep watching that lower layer file 1962 * even if upper layer file would be created later on. 1963 */ 1964 if (UPPERVP(ap->a_vp)) 1965 error = VOP_KQFILTER(UPPERVP(ap->a_vp), ap->a_kn); 1966 else if (LOWERVP(ap->a_vp)) 1967 error = VOP_KQFILTER(LOWERVP(ap->a_vp), ap->a_kn); 1968 else { 1969 /* panic? */ 1970 error = EOPNOTSUPP; 1971 } 1972 1973 return (error); 1974 } 1975
Indexes created Sat Sep 20 22:09:52 GMT 2025