union_vnops.c revision 1.77
1/* $NetBSD: union_vnops.c,v 1.77 2021/06/29 22:39:20 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 1.77 2021/06/29 22:39:20 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 96int union_parsepath(void *); 97int union_lookup(void *); 98int union_create(void *); 99int union_whiteout(void *); 100int union_mknod(void *); 101int union_open(void *); 102int union_close(void *); 103int union_access(void *); 104int union_getattr(void *); 105int union_setattr(void *); 106int union_read(void *); 107int union_write(void *); 108int union_ioctl(void *); 109int union_poll(void *); 110int union_revoke(void *); 111int union_mmap(void *); 112int union_fsync(void *); 113int union_seek(void *); 114int union_remove(void *); 115int union_link(void *); 116int union_rename(void *); 117int union_mkdir(void *); 118int union_rmdir(void *); 119int union_symlink(void *); 120int union_readdir(void *); 121int union_readlink(void *); 122int union_abortop(void *); 123int union_inactive(void *); 124int union_reclaim(void *); 125int union_lock(void *); 126int union_unlock(void *); 127int union_bmap(void *); 128int union_print(void *); 129int union_islocked(void *); 130int union_pathconf(void *); 131int union_advlock(void *); 132int union_strategy(void *); 133int union_bwrite(void *); 134int union_getpages(void *); 135int union_putpages(void *); 136int union_kqfilter(void *); 137 138static int union_lookup1(struct vnode *, struct vnode **, 139 struct vnode **, struct componentname *); 140 141 142/* 143 * Global vfs data structures 144 */ 145int (**union_vnodeop_p)(void *); 146const 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}; 194const 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 201int 202union_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 error = VOP_PARSEPATH(upperdvp, ap->a_name, &upper); 217 if (error) { 218 return error; 219 } 220 221 error = VOP_PARSEPATH(lowerdvp, ap->a_name, &lower); 222 if (error) { 223 return error; 224 } 225 226 /* 227 * If they're different, use the larger one. This is not a 228 * comprehensive solution, but it's sufficient for the 229 * non-default cases of parsepath that currently exist. 230 */ 231 *ap->a_retval = MAX(upper, lower); 232 return 0; 233} 234 235static int 236union_lookup1(struct vnode *udvp, struct vnode **dvpp, struct vnode **vpp, 237 struct componentname *cnp) 238{ 239 int error; 240 struct vnode *tdvp; 241 struct vnode *dvp; 242 struct mount *mp; 243 244 dvp = *dvpp; 245 246 /* 247 * If stepping up the directory tree, check for going 248 * back across the mount point, in which case do what 249 * lookup would do by stepping back down the mount 250 * hierarchy. 251 */ 252 if (cnp->cn_flags & ISDOTDOT) { 253 while ((dvp != udvp) && (dvp->v_vflag & VV_ROOT)) { 254 /* 255 * Don't do the NOCROSSMOUNT check 256 * at this level. By definition, 257 * union fs deals with namespaces, not 258 * filesystems. 259 */ 260 tdvp = dvp; 261 *dvpp = dvp = dvp->v_mount->mnt_vnodecovered; 262 VOP_UNLOCK(tdvp); 263 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 264 } 265 } 266 267 error = VOP_LOOKUP(dvp, &tdvp, cnp); 268 if (error) 269 return (error); 270 if (dvp != tdvp) { 271 if (cnp->cn_flags & ISDOTDOT) 272 VOP_UNLOCK(dvp); 273 error = vn_lock(tdvp, LK_EXCLUSIVE); 274 if (cnp->cn_flags & ISDOTDOT) 275 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 276 if (error) { 277 vrele(tdvp); 278 return error; 279 } 280 dvp = tdvp; 281 } 282 283 /* 284 * Lastly check if the current node is a mount point in 285 * which case walk up the mount hierarchy making sure not to 286 * bump into the root of the mount tree (ie. dvp != udvp). 287 */ 288 while (dvp != udvp && (dvp->v_type == VDIR) && 289 (mp = dvp->v_mountedhere)) { 290 if (vfs_busy(mp)) 291 continue; 292 vput(dvp); 293 error = VFS_ROOT(mp, LK_EXCLUSIVE, &tdvp); 294 vfs_unbusy(mp); 295 if (error) { 296 return (error); 297 } 298 dvp = tdvp; 299 } 300 301 *vpp = dvp; 302 return (0); 303} 304 305int 306union_lookup(void *v) 307{ 308 struct vop_lookup_v2_args /* { 309 struct vnodeop_desc *a_desc; 310 struct vnode *a_dvp; 311 struct vnode **a_vpp; 312 struct componentname *a_cnp; 313 } */ *ap = v; 314 int error; 315 int uerror, lerror; 316 struct vnode *uppervp, *lowervp; 317 struct vnode *upperdvp, *lowerdvp; 318 struct vnode *dvp = ap->a_dvp; 319 struct union_node *dun = VTOUNION(dvp); 320 struct componentname *cnp = ap->a_cnp; 321 struct union_mount *um = MOUNTTOUNIONMOUNT(dvp->v_mount); 322 kauth_cred_t saved_cred = NULL; 323 int iswhiteout; 324 struct vattr va; 325 326#ifdef notyet 327 if (cnp->cn_namelen == 3 && 328 cnp->cn_nameptr[2] == '.' && 329 cnp->cn_nameptr[1] == '.' && 330 cnp->cn_nameptr[0] == '.') { 331 dvp = *ap->a_vpp = LOWERVP(ap->a_dvp); 332 if (dvp == NULLVP) 333 return (ENOENT); 334 vref(dvp); 335 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 336 return (0); 337 } 338#endif 339 340 if ((cnp->cn_flags & ISLASTCN) && 341 (dvp->v_mount->mnt_flag & MNT_RDONLY) && 342 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) 343 return (EROFS); 344 345start: 346 upperdvp = dun->un_uppervp; 347 lowerdvp = dun->un_lowervp; 348 uppervp = NULLVP; 349 lowervp = NULLVP; 350 iswhiteout = 0; 351 352 /* 353 * do the lookup in the upper level. 354 * if that level comsumes additional pathnames, 355 * then assume that something special is going 356 * on and just return that vnode. 357 */ 358 if (upperdvp != NULLVP) { 359 uerror = union_lookup1(um->um_uppervp, &upperdvp, 360 &uppervp, cnp); 361 if (uerror == ENOENT || uerror == EJUSTRETURN) { 362 if (cnp->cn_flags & ISWHITEOUT) { 363 iswhiteout = 1; 364 } else if (lowerdvp != NULLVP) { 365 lerror = VOP_GETATTR(upperdvp, &va, 366 cnp->cn_cred); 367 if (lerror == 0 && (va.va_flags & OPAQUE)) 368 iswhiteout = 1; 369 } 370 } 371 } else { 372 uerror = ENOENT; 373 } 374 375 /* 376 * in a similar way to the upper layer, do the lookup 377 * in the lower layer. this time, if there is some 378 * component magic going on, then vput whatever we got 379 * back from the upper layer and return the lower vnode 380 * instead. 381 */ 382 if (lowerdvp != NULLVP && !iswhiteout) { 383 int nameiop; 384 385 vn_lock(lowerdvp, LK_EXCLUSIVE | LK_RETRY); 386 387 /* 388 * Only do a LOOKUP on the bottom node, since 389 * we won't be making changes to it anyway. 390 */ 391 nameiop = cnp->cn_nameiop; 392 cnp->cn_nameiop = LOOKUP; 393 if (um->um_op == UNMNT_BELOW) { 394 saved_cred = cnp->cn_cred; 395 cnp->cn_cred = um->um_cred; 396 } 397 398 /* 399 * we shouldn't have to worry about locking interactions 400 * between the lower layer and our union layer (w.r.t. 401 * `..' processing) because we don't futz with lowervp 402 * locks in the union-node instantiation code path. 403 */ 404 lerror = union_lookup1(um->um_lowervp, &lowerdvp, 405 &lowervp, cnp); 406 if (um->um_op == UNMNT_BELOW) 407 cnp->cn_cred = saved_cred; 408 cnp->cn_nameiop = nameiop; 409 410 if (lowervp != lowerdvp) 411 VOP_UNLOCK(lowerdvp); 412 } else { 413 lerror = ENOENT; 414 if ((cnp->cn_flags & ISDOTDOT) && dun->un_pvp != NULLVP) { 415 lowervp = LOWERVP(dun->un_pvp); 416 if (lowervp != NULLVP) { 417 vref(lowervp); 418 vn_lock(lowervp, LK_EXCLUSIVE | LK_RETRY); 419 lerror = 0; 420 } 421 } 422 } 423 424 /* 425 * EJUSTRETURN is used by underlying filesystems to indicate that 426 * a directory modification op was started successfully. 427 * This will only happen in the upper layer, since 428 * the lower layer only does LOOKUPs. 429 * If this union is mounted read-only, bounce it now. 430 */ 431 432 if ((uerror == EJUSTRETURN) && (cnp->cn_flags & ISLASTCN) && 433 (dvp->v_mount->mnt_flag & MNT_RDONLY) && 434 ((cnp->cn_nameiop == CREATE) || (cnp->cn_nameiop == RENAME))) 435 uerror = EROFS; 436 437 /* 438 * at this point, we have uerror and lerror indicating 439 * possible errors with the lookups in the upper and lower 440 * layers. additionally, uppervp and lowervp are (locked) 441 * references to existing vnodes in the upper and lower layers. 442 * 443 * there are now three cases to consider. 444 * 1. if both layers returned an error, then return whatever 445 * error the upper layer generated. 446 * 447 * 2. if the top layer failed and the bottom layer succeeded 448 * then two subcases occur. 449 * a. the bottom vnode is not a directory, in which 450 * case just return a new union vnode referencing 451 * an empty top layer and the existing bottom layer. 452 * b. the bottom vnode is a directory, in which case 453 * create a new directory in the top-level and 454 * continue as in case 3. 455 * 456 * 3. if the top layer succeeded then return a new union 457 * vnode referencing whatever the new top layer and 458 * whatever the bottom layer returned. 459 */ 460 461 *ap->a_vpp = NULLVP; 462 463 464 /* case 1. */ 465 if ((uerror != 0) && (lerror != 0)) { 466 return (uerror); 467 } 468 469 /* case 2. */ 470 if (uerror != 0 /* && (lerror == 0) */ ) { 471 if (lowervp->v_type == VDIR) { /* case 2b. */ 472 /* 473 * We may be racing another process to make the 474 * upper-level shadow directory. Be careful with 475 * locks/etc! 476 * If we have to create a shadow directory and want 477 * to commit the node we have to restart the lookup 478 * to get the componentname right. 479 */ 480 if (upperdvp) { 481 VOP_UNLOCK(upperdvp); 482 uerror = union_mkshadow(um, upperdvp, cnp, 483 &uppervp); 484 vn_lock(upperdvp, LK_EXCLUSIVE | LK_RETRY); 485 if (uerror == 0 && cnp->cn_nameiop != LOOKUP) { 486 vrele(uppervp); 487 if (lowervp != NULLVP) 488 vput(lowervp); 489 goto start; 490 } 491 } 492 if (uerror) { 493 if (lowervp != NULLVP) { 494 vput(lowervp); 495 lowervp = NULLVP; 496 } 497 return (uerror); 498 } 499 } 500 } else { /* uerror == 0 */ 501 if (uppervp != upperdvp) 502 VOP_UNLOCK(uppervp); 503 } 504 505 if (lowervp != NULLVP) 506 VOP_UNLOCK(lowervp); 507 508 error = union_allocvp(ap->a_vpp, dvp->v_mount, dvp, upperdvp, cnp, 509 uppervp, lowervp, 1); 510 511 if (error) { 512 if (uppervp != NULLVP) 513 vrele(uppervp); 514 if (lowervp != NULLVP) 515 vrele(lowervp); 516 return error; 517 } 518 519 return 0; 520} 521 522int 523union_create(void *v) 524{ 525 struct vop_create_v3_args /* { 526 struct vnode *a_dvp; 527 struct vnode **a_vpp; 528 struct componentname *a_cnp; 529 struct vattr *a_vap; 530 } */ *ap = v; 531 struct union_node *un = VTOUNION(ap->a_dvp); 532 struct vnode *dvp = un->un_uppervp; 533 struct componentname *cnp = ap->a_cnp; 534 535 if (dvp != NULLVP) { 536 int error; 537 struct vnode *vp; 538 struct mount *mp; 539 540 mp = ap->a_dvp->v_mount; 541 542 vp = NULL; 543 error = VOP_CREATE(dvp, &vp, cnp, ap->a_vap); 544 if (error) 545 return (error); 546 547 error = union_allocvp(ap->a_vpp, mp, NULLVP, NULLVP, cnp, vp, 548 NULLVP, 1); 549 if (error) 550 vrele(vp); 551 return (error); 552 } 553 554 return (EROFS); 555} 556 557int 558union_whiteout(void *v) 559{ 560 struct vop_whiteout_args /* { 561 struct vnode *a_dvp; 562 struct componentname *a_cnp; 563 int a_flags; 564 } */ *ap = v; 565 struct union_node *un = VTOUNION(ap->a_dvp); 566 struct componentname *cnp = ap->a_cnp; 567 568 if (un->un_uppervp == NULLVP) 569 return (EOPNOTSUPP); 570 571 return (VOP_WHITEOUT(un->un_uppervp, cnp, ap->a_flags)); 572} 573 574int 575union_mknod(void *v) 576{ 577 struct vop_mknod_v3_args /* { 578 struct vnode *a_dvp; 579 struct vnode **a_vpp; 580 struct componentname *a_cnp; 581 struct vattr *a_vap; 582 } */ *ap = v; 583 struct union_node *un = VTOUNION(ap->a_dvp); 584 struct vnode *dvp = un->un_uppervp; 585 struct componentname *cnp = ap->a_cnp; 586 587 if (dvp != NULLVP) { 588 int error; 589 struct vnode *vp; 590 struct mount *mp; 591 592 mp = ap->a_dvp->v_mount; 593 error = VOP_MKNOD(dvp, &vp, cnp, ap->a_vap); 594 if (error) 595 return (error); 596 597 error = union_allocvp(ap->a_vpp, mp, NULLVP, NULLVP, 598 cnp, vp, NULLVP, 1); 599 if (error) 600 vrele(vp); 601 return (error); 602 } 603 604 return (EROFS); 605} 606 607int 608union_open(void *v) 609{ 610 struct vop_open_args /* { 611 struct vnodeop_desc *a_desc; 612 struct vnode *a_vp; 613 int a_mode; 614 kauth_cred_t a_cred; 615 } */ *ap = v; 616 struct union_node *un = VTOUNION(ap->a_vp); 617 struct vnode *tvp; 618 int mode = ap->a_mode; 619 kauth_cred_t cred = ap->a_cred; 620 struct lwp *l = curlwp; 621 int error; 622 623 /* 624 * If there is an existing upper vp then simply open that. 625 */ 626 tvp = un->un_uppervp; 627 if (tvp == NULLVP) { 628 /* 629 * If the lower vnode is being opened for writing, then 630 * copy the file contents to the upper vnode and open that, 631 * otherwise can simply open the lower vnode. 632 */ 633 tvp = un->un_lowervp; 634 if ((ap->a_mode & FWRITE) && (tvp->v_type == VREG)) { 635 error = union_copyup(un, (mode&O_TRUNC) == 0, cred, l); 636 if (error == 0) 637 error = VOP_OPEN(un->un_uppervp, mode, cred); 638 if (error == 0) { 639 mutex_enter(un->un_uppervp->v_interlock); 640 un->un_uppervp->v_writecount++; 641 mutex_exit(un->un_uppervp->v_interlock); 642 } 643 return (error); 644 } 645 646 /* 647 * Just open the lower vnode, but check for nodev mount flag 648 */ 649 if ((tvp->v_type == VBLK || tvp->v_type == VCHR) && 650 (ap->a_vp->v_mount->mnt_flag & MNT_NODEV)) 651 return ENXIO; 652 un->un_openl++; 653 vn_lock(tvp, LK_EXCLUSIVE | LK_RETRY); 654 error = VOP_OPEN(tvp, mode, cred); 655 VOP_UNLOCK(tvp); 656 657 return (error); 658 } 659 /* 660 * Just open the upper vnode, checking for nodev mount flag first 661 */ 662 if ((tvp->v_type == VBLK || tvp->v_type == VCHR) && 663 (ap->a_vp->v_mount->mnt_flag & MNT_NODEV)) 664 return ENXIO; 665 666 error = VOP_OPEN(tvp, mode, cred); 667 if (error == 0 && (ap->a_mode & FWRITE)) { 668 mutex_enter(tvp->v_interlock); 669 tvp->v_writecount++; 670 mutex_exit(tvp->v_interlock); 671 } 672 673 return (error); 674} 675 676int 677union_close(void *v) 678{ 679 struct vop_close_args /* { 680 struct vnode *a_vp; 681 int a_fflag; 682 kauth_cred_t a_cred; 683 } */ *ap = v; 684 struct union_node *un = VTOUNION(ap->a_vp); 685 struct vnode *vp; 686 int error; 687 bool do_lock; 688 689 vp = un->un_uppervp; 690 if (vp != NULLVP) { 691 do_lock = false; 692 } else { 693 KASSERT(un->un_openl > 0); 694 --un->un_openl; 695 vp = un->un_lowervp; 696 do_lock = true; 697 } 698 699 KASSERT(vp != NULLVP); 700 ap->a_vp = vp; 701 if ((ap->a_fflag & FWRITE)) { 702 KASSERT(vp == un->un_uppervp); 703 mutex_enter(vp->v_interlock); 704 vp->v_writecount--; 705 mutex_exit(vp->v_interlock); 706 } 707 if (do_lock) 708 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 709 error = VCALL(vp, VOFFSET(vop_close), ap); 710 if (do_lock) 711 VOP_UNLOCK(vp); 712 713 return error; 714} 715 716/* 717 * Check access permission on the union vnode. 718 * The access check being enforced is to check 719 * against both the underlying vnode, and any 720 * copied vnode. This ensures that no additional 721 * file permissions are given away simply because 722 * the user caused an implicit file copy. 723 */ 724int 725union_access(void *v) 726{ 727 struct vop_access_args /* { 728 struct vnodeop_desc *a_desc; 729 struct vnode *a_vp; 730 accmode_t a_accmode; 731 kauth_cred_t a_cred; 732 } */ *ap = v; 733 struct vnode *vp = ap->a_vp; 734 struct union_node *un = VTOUNION(vp); 735 int error = EACCES; 736 struct union_mount *um = MOUNTTOUNIONMOUNT(vp->v_mount); 737 738 /* 739 * Disallow write attempts on read-only file systems; 740 * unless the file is a socket, fifo, or a block or 741 * character device resident on the file system. 742 */ 743 if (ap->a_accmode & VWRITE) { 744 switch (vp->v_type) { 745 case VDIR: 746 case VLNK: 747 case VREG: 748 if (vp->v_mount->mnt_flag & MNT_RDONLY) 749 return (EROFS); 750 break; 751 case VBAD: 752 case VBLK: 753 case VCHR: 754 case VSOCK: 755 case VFIFO: 756 case VNON: 757 default: 758 break; 759 } 760 } 761 762 763 if ((vp = un->un_uppervp) != NULLVP) { 764 ap->a_vp = vp; 765 return (VCALL(vp, VOFFSET(vop_access), ap)); 766 } 767 768 if ((vp = un->un_lowervp) != NULLVP) { 769 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 770 ap->a_vp = vp; 771 error = VCALL(vp, VOFFSET(vop_access), ap); 772 if (error == 0) { 773 if (um->um_op == UNMNT_BELOW) { 774 ap->a_cred = um->um_cred; 775 error = VCALL(vp, VOFFSET(vop_access), ap); 776 } 777 } 778 VOP_UNLOCK(vp); 779 if (error) 780 return (error); 781 } 782 783 return (error); 784} 785 786/* 787 * We handle getattr only to change the fsid and 788 * track object sizes 789 */ 790int 791union_getattr(void *v) 792{ 793 struct vop_getattr_args /* { 794 struct vnode *a_vp; 795 struct vattr *a_vap; 796 kauth_cred_t a_cred; 797 } */ *ap = v; 798 int error; 799 struct union_node *un = VTOUNION(ap->a_vp); 800 struct vnode *vp = un->un_uppervp; 801 struct vattr *vap; 802 struct vattr va; 803 804 805 /* 806 * Some programs walk the filesystem hierarchy by counting 807 * links to directories to avoid stat'ing all the time. 808 * This means the link count on directories needs to be "correct". 809 * The only way to do that is to call getattr on both layers 810 * and fix up the link count. The link count will not necessarily 811 * be accurate but will be large enough to defeat the tree walkers. 812 * 813 * To make life more interesting, some filesystems don't keep 814 * track of link counts in the expected way, and return a 815 * link count of `1' for those directories; if either of the 816 * component directories returns a link count of `1', we return a 1. 817 */ 818 819 vap = ap->a_vap; 820 821 vp = un->un_uppervp; 822 if (vp != NULLVP) { 823 error = VOP_GETATTR(vp, vap, ap->a_cred); 824 if (error) 825 return (error); 826 mutex_enter(&un->un_lock); 827 union_newsize(ap->a_vp, vap->va_size, VNOVAL); 828 } 829 830 if (vp == NULLVP) { 831 vp = un->un_lowervp; 832 } else if (vp->v_type == VDIR) { 833 vp = un->un_lowervp; 834 if (vp != NULLVP) 835 vap = &va; 836 } else { 837 vp = NULLVP; 838 } 839 840 if (vp != NULLVP) { 841 if (vp == un->un_lowervp) 842 vn_lock(vp, LK_SHARED | LK_RETRY); 843 error = VOP_GETATTR(vp, vap, ap->a_cred); 844 if (vp == un->un_lowervp) 845 VOP_UNLOCK(vp); 846 if (error) 847 return (error); 848 mutex_enter(&un->un_lock); 849 union_newsize(ap->a_vp, VNOVAL, vap->va_size); 850 } 851 852 if ((vap != ap->a_vap) && (vap->va_type == VDIR)) { 853 /* 854 * Link count manipulation: 855 * - If both return "2", return 2 (no subdirs) 856 * - If one or the other return "1", return "1" (ENOCLUE) 857 */ 858 if ((ap->a_vap->va_nlink == 2) && 859 (vap->va_nlink == 2)) 860 ; 861 else if (ap->a_vap->va_nlink != 1) { 862 if (vap->va_nlink == 1) 863 ap->a_vap->va_nlink = 1; 864 else 865 ap->a_vap->va_nlink += vap->va_nlink; 866 } 867 } 868 ap->a_vap->va_fsid = ap->a_vp->v_mount->mnt_stat.f_fsidx.__fsid_val[0]; 869 return (0); 870} 871 872int 873union_setattr(void *v) 874{ 875 struct vop_setattr_args /* { 876 struct vnode *a_vp; 877 struct vattr *a_vap; 878 kauth_cred_t a_cred; 879 } */ *ap = v; 880 struct vattr *vap = ap->a_vap; 881 struct vnode *vp = ap->a_vp; 882 struct union_node *un = VTOUNION(vp); 883 bool size_only; /* All but va_size are VNOVAL. */ 884 int error; 885 886 size_only = (vap->va_flags == VNOVAL && vap->va_uid == (uid_t)VNOVAL && 887 vap->va_gid == (gid_t)VNOVAL && vap->va_atime.tv_sec == VNOVAL && 888 vap->va_mtime.tv_sec == VNOVAL && vap->va_mode == (mode_t)VNOVAL); 889 890 if (!size_only && (vp->v_mount->mnt_flag & MNT_RDONLY)) 891 return (EROFS); 892 if (vap->va_size != VNOVAL) { 893 switch (vp->v_type) { 894 case VDIR: 895 return (EISDIR); 896 case VCHR: 897 case VBLK: 898 case VSOCK: 899 case VFIFO: 900 break; 901 case VREG: 902 case VLNK: 903 default: 904 /* 905 * Disallow write attempts if the filesystem is 906 * mounted read-only. 907 */ 908 if (vp->v_mount->mnt_flag & MNT_RDONLY) 909 return (EROFS); 910 } 911 } 912 913 /* 914 * Handle case of truncating lower object to zero size, 915 * by creating a zero length upper object. This is to 916 * handle the case of open with O_TRUNC and O_CREAT. 917 */ 918 if ((un->un_uppervp == NULLVP) && 919 /* assert(un->un_lowervp != NULLVP) */ 920 (un->un_lowervp->v_type == VREG)) { 921 error = union_copyup(un, (vap->va_size != 0), 922 ap->a_cred, curlwp); 923 if (error) 924 return (error); 925 } 926 927 /* 928 * Try to set attributes in upper layer, ignore size change to zero 929 * for devices to handle O_TRUNC and return read-only filesystem error 930 * otherwise. 931 */ 932 if (un->un_uppervp != NULLVP) { 933 error = VOP_SETATTR(un->un_uppervp, vap, ap->a_cred); 934 if ((error == 0) && (vap->va_size != VNOVAL)) { 935 mutex_enter(&un->un_lock); 936 union_newsize(ap->a_vp, vap->va_size, VNOVAL); 937 } 938 } else { 939 KASSERT(un->un_lowervp != NULLVP); 940 if (NODE_IS_SPECIAL(un->un_lowervp)) { 941 if (size_only && 942 (vap->va_size == 0 || vap->va_size == VNOVAL)) 943 error = 0; 944 else 945 error = EROFS; 946 } else { 947 error = EROFS; 948 } 949 } 950 951 return (error); 952} 953 954int 955union_read(void *v) 956{ 957 struct vop_read_args /* { 958 struct vnode *a_vp; 959 struct uio *a_uio; 960 int a_ioflag; 961 kauth_cred_t a_cred; 962 } */ *ap = v; 963 int error; 964 struct vnode *vp = OTHERVP(ap->a_vp); 965 int dolock = (vp == LOWERVP(ap->a_vp)); 966 967 if (dolock) 968 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 969 error = VOP_READ(vp, ap->a_uio, ap->a_ioflag, ap->a_cred); 970 if (dolock) 971 VOP_UNLOCK(vp); 972 973 /* 974 * XXX 975 * perhaps the size of the underlying object has changed under 976 * our feet. take advantage of the offset information present 977 * in the uio structure. 978 */ 979 if (error == 0) { 980 struct union_node *un = VTOUNION(ap->a_vp); 981 off_t cur = ap->a_uio->uio_offset; 982 off_t usz = VNOVAL, lsz = VNOVAL; 983 984 mutex_enter(&un->un_lock); 985 if (vp == un->un_uppervp) { 986 if (cur > un->un_uppersz) 987 usz = cur; 988 } else { 989 if (cur > un->un_lowersz) 990 lsz = cur; 991 } 992 993 if (usz != VNOVAL || lsz != VNOVAL) 994 union_newsize(ap->a_vp, usz, lsz); 995 else 996 mutex_exit(&un->un_lock); 997 } 998 999 return (error); 1000} 1001 1002int 1003union_write(void *v) 1004{ 1005 struct vop_read_args /* { 1006 struct vnode *a_vp; 1007 struct uio *a_uio; 1008 int a_ioflag; 1009 kauth_cred_t a_cred; 1010 } */ *ap = v; 1011 int error; 1012 struct vnode *vp; 1013 struct union_node *un = VTOUNION(ap->a_vp); 1014 1015 vp = UPPERVP(ap->a_vp); 1016 if (vp == NULLVP) { 1017 vp = LOWERVP(ap->a_vp); 1018 if (NODE_IS_SPECIAL(vp)) { 1019 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1020 error = VOP_WRITE(vp, ap->a_uio, ap->a_ioflag, 1021 ap->a_cred); 1022 VOP_UNLOCK(vp); 1023 return error; 1024 } 1025 panic("union: missing upper layer in write"); 1026 } 1027 1028 error = VOP_WRITE(vp, ap->a_uio, ap->a_ioflag, ap->a_cred); 1029 1030 /* 1031 * the size of the underlying object may be changed by the 1032 * write. 1033 */ 1034 if (error == 0) { 1035 off_t cur = ap->a_uio->uio_offset; 1036 1037 mutex_enter(&un->un_lock); 1038 if (cur > un->un_uppersz) 1039 union_newsize(ap->a_vp, cur, VNOVAL); 1040 else 1041 mutex_exit(&un->un_lock); 1042 } 1043 1044 return (error); 1045} 1046 1047int 1048union_ioctl(void *v) 1049{ 1050 struct vop_ioctl_args /* { 1051 struct vnode *a_vp; 1052 int a_command; 1053 void *a_data; 1054 int a_fflag; 1055 kauth_cred_t a_cred; 1056 } */ *ap = v; 1057 struct vnode *ovp = OTHERVP(ap->a_vp); 1058 1059 ap->a_vp = ovp; 1060 return (VCALL(ovp, VOFFSET(vop_ioctl), ap)); 1061} 1062 1063int 1064union_poll(void *v) 1065{ 1066 struct vop_poll_args /* { 1067 struct vnode *a_vp; 1068 int a_events; 1069 } */ *ap = v; 1070 struct vnode *ovp = OTHERVP(ap->a_vp); 1071 1072 ap->a_vp = ovp; 1073 return (VCALL(ovp, VOFFSET(vop_poll), ap)); 1074} 1075 1076int 1077union_revoke(void *v) 1078{ 1079 struct vop_revoke_args /* { 1080 struct vnode *a_vp; 1081 int a_flags; 1082 struct proc *a_p; 1083 } */ *ap = v; 1084 struct vnode *vp = ap->a_vp; 1085 1086 if (UPPERVP(vp)) 1087 VOP_REVOKE(UPPERVP(vp), ap->a_flags); 1088 if (LOWERVP(vp)) 1089 VOP_REVOKE(LOWERVP(vp), ap->a_flags); 1090 vgone(vp); /* XXXAD?? */ 1091 return (0); 1092} 1093 1094int 1095union_mmap(void *v) 1096{ 1097 struct vop_mmap_args /* { 1098 struct vnode *a_vp; 1099 vm_prot_t a_prot; 1100 kauth_cred_t a_cred; 1101 } */ *ap = v; 1102 struct vnode *ovp = OTHERVP(ap->a_vp); 1103 1104 ap->a_vp = ovp; 1105 return (VCALL(ovp, VOFFSET(vop_mmap), ap)); 1106} 1107 1108int 1109union_fsync(void *v) 1110{ 1111 struct vop_fsync_args /* { 1112 struct vnode *a_vp; 1113 kauth_cred_t a_cred; 1114 int a_flags; 1115 off_t offhi; 1116 off_t offlo; 1117 } */ *ap = v; 1118 int error = 0; 1119 struct vnode *targetvp; 1120 1121 /* 1122 * If vinvalbuf is calling us, it's a "shallow fsync" -- don't 1123 * bother syncing the underlying vnodes, since (a) they'll be 1124 * fsync'ed when reclaimed and (b) we could deadlock if 1125 * they're locked; otherwise, pass it through to the 1126 * underlying layer. 1127 */ 1128 if (ap->a_vp->v_type == VBLK || ap->a_vp->v_type == VCHR) { 1129 error = spec_fsync(v); 1130 if (error) 1131 return error; 1132 } 1133 1134 if (ap->a_flags & FSYNC_RECLAIM) 1135 return 0; 1136 1137 targetvp = OTHERVP(ap->a_vp); 1138 if (targetvp != NULLVP) { 1139 int dolock = (targetvp == LOWERVP(ap->a_vp)); 1140 1141 if (dolock) 1142 vn_lock(targetvp, LK_EXCLUSIVE | LK_RETRY); 1143 error = VOP_FSYNC(targetvp, ap->a_cred, ap->a_flags, 1144 ap->a_offlo, ap->a_offhi); 1145 if (dolock) 1146 VOP_UNLOCK(targetvp); 1147 } 1148 1149 return (error); 1150} 1151 1152int 1153union_seek(void *v) 1154{ 1155 struct vop_seek_args /* { 1156 struct vnode *a_vp; 1157 off_t a_oldoff; 1158 off_t a_newoff; 1159 kauth_cred_t a_cred; 1160 } */ *ap = v; 1161 struct vnode *ovp = OTHERVP(ap->a_vp); 1162 1163 ap->a_vp = ovp; 1164 return (VCALL(ovp, VOFFSET(vop_seek), ap)); 1165} 1166 1167int 1168union_remove(void *v) 1169{ 1170 struct vop_remove_v2_args /* { 1171 struct vnode *a_dvp; 1172 struct vnode *a_vp; 1173 struct componentname *a_cnp; 1174 } */ *ap = v; 1175 int error; 1176 struct union_node *dun = VTOUNION(ap->a_dvp); 1177 struct union_node *un = VTOUNION(ap->a_vp); 1178 struct componentname *cnp = ap->a_cnp; 1179 1180 if (dun->un_uppervp == NULLVP) 1181 panic("union remove: null upper vnode"); 1182 1183 if (un->un_uppervp != NULLVP) { 1184 struct vnode *dvp = dun->un_uppervp; 1185 struct vnode *vp = un->un_uppervp; 1186 1187 /* Account for VOP_REMOVE to vrele vp. */ 1188 vref(vp); 1189 if (union_dowhiteout(un, cnp->cn_cred)) 1190 cnp->cn_flags |= DOWHITEOUT; 1191 error = VOP_REMOVE(dvp, vp, cnp); 1192 if (!error) 1193 union_removed_upper(un); 1194 vrele(ap->a_vp); 1195 } else { 1196 error = union_mkwhiteout( 1197 MOUNTTOUNIONMOUNT(UNIONTOV(dun)->v_mount), 1198 dun->un_uppervp, ap->a_cnp, un); 1199 vput(ap->a_vp); 1200 } 1201 1202 return (error); 1203} 1204 1205int 1206union_link(void *v) 1207{ 1208 struct vop_link_v2_args /* { 1209 struct vnode *a_dvp; 1210 struct vnode *a_vp; 1211 struct componentname *a_cnp; 1212 } */ *ap = v; 1213 int error = 0; 1214 struct componentname *cnp = ap->a_cnp; 1215 struct union_node *dun; 1216 struct vnode *vp; 1217 struct vnode *dvp; 1218 1219 dun = VTOUNION(ap->a_dvp); 1220 1221 KASSERT((ap->a_cnp->cn_flags & LOCKPARENT) != 0); 1222 1223 if (ap->a_dvp->v_op != ap->a_vp->v_op) { 1224 vp = ap->a_vp; 1225 } else { 1226 struct union_node *un = VTOUNION(ap->a_vp); 1227 if (un->un_uppervp == NULLVP) { 1228 const bool droplock = (dun->un_uppervp == un->un_dirvp); 1229 1230 /* 1231 * Needs to be copied before we can link it. 1232 */ 1233 vn_lock(ap->a_vp, LK_EXCLUSIVE | LK_RETRY); 1234 if (droplock) 1235 VOP_UNLOCK(dun->un_uppervp); 1236 error = union_copyup(un, 1, cnp->cn_cred, curlwp); 1237 if (droplock) { 1238 vn_lock(dun->un_uppervp, 1239 LK_EXCLUSIVE | LK_RETRY); 1240 /* 1241 * During copyup, we dropped the lock on the 1242 * dir and invalidated any saved namei lookup 1243 * state for the directory we'll be entering 1244 * the link in. We need to re-run the lookup 1245 * in that directory to reset any state needed 1246 * for VOP_LINK. 1247 * Call relookup on the union-layer to reset 1248 * the state. 1249 */ 1250 vp = NULLVP; 1251 if (dun->un_uppervp == NULLVP) 1252 panic("union: null upperdvp?"); 1253 error = relookup(ap->a_dvp, &vp, ap->a_cnp, 0); 1254 if (error) { 1255 VOP_UNLOCK(ap->a_vp); 1256 return EROFS; /* ? */ 1257 } 1258 if (vp != NULLVP) { 1259 /* 1260 * The name we want to create has 1261 * mysteriously appeared (a race?) 1262 */ 1263 error = EEXIST; 1264 VOP_UNLOCK(ap->a_vp); 1265 vput(vp); 1266 return (error); 1267 } 1268 } 1269 VOP_UNLOCK(ap->a_vp); 1270 } 1271 vp = un->un_uppervp; 1272 } 1273 1274 dvp = dun->un_uppervp; 1275 if (dvp == NULLVP) 1276 error = EROFS; 1277 1278 if (error) 1279 return (error); 1280 1281 return VOP_LINK(dvp, vp, cnp); 1282} 1283 1284int 1285union_rename(void *v) 1286{ 1287 struct vop_rename_args /* { 1288 struct vnode *a_fdvp; 1289 struct vnode *a_fvp; 1290 struct componentname *a_fcnp; 1291 struct vnode *a_tdvp; 1292 struct vnode *a_tvp; 1293 struct componentname *a_tcnp; 1294 } */ *ap = v; 1295 int error; 1296 1297 struct vnode *fdvp = ap->a_fdvp; 1298 struct vnode *fvp = ap->a_fvp; 1299 struct vnode *tdvp = ap->a_tdvp; 1300 struct vnode *tvp = ap->a_tvp; 1301 1302 /* 1303 * Account for VOP_RENAME to vrele all nodes. 1304 * Note: VOP_RENAME will unlock tdvp. 1305 */ 1306 1307 if (fdvp->v_op == union_vnodeop_p) { /* always true */ 1308 struct union_node *un = VTOUNION(fdvp); 1309 if (un->un_uppervp == NULLVP) { 1310 /* 1311 * this should never happen in normal 1312 * operation but might if there was 1313 * a problem creating the top-level shadow 1314 * directory. 1315 */ 1316 error = EXDEV; 1317 goto bad; 1318 } 1319 1320 fdvp = un->un_uppervp; 1321 vref(fdvp); 1322 } 1323 1324 if (fvp->v_op == union_vnodeop_p) { /* always true */ 1325 struct union_node *un = VTOUNION(fvp); 1326 if (un->un_uppervp == NULLVP) { 1327 /* XXX: should do a copyup */ 1328 error = EXDEV; 1329 goto bad; 1330 } 1331 1332 if (un->un_lowervp != NULLVP) 1333 ap->a_fcnp->cn_flags |= DOWHITEOUT; 1334 1335 fvp = un->un_uppervp; 1336 vref(fvp); 1337 } 1338 1339 if (tdvp->v_op == union_vnodeop_p) { 1340 struct union_node *un = VTOUNION(tdvp); 1341 if (un->un_uppervp == NULLVP) { 1342 /* 1343 * this should never happen in normal 1344 * operation but might if there was 1345 * a problem creating the top-level shadow 1346 * directory. 1347 */ 1348 error = EXDEV; 1349 goto bad; 1350 } 1351 1352 tdvp = un->un_uppervp; 1353 vref(tdvp); 1354 } 1355 1356 if (tvp != NULLVP && tvp->v_op == union_vnodeop_p) { 1357 struct union_node *un = VTOUNION(tvp); 1358 1359 tvp = un->un_uppervp; 1360 if (tvp != NULLVP) { 1361 vref(tvp); 1362 } 1363 } 1364 1365 error = VOP_RENAME(fdvp, fvp, ap->a_fcnp, tdvp, tvp, ap->a_tcnp); 1366 goto out; 1367 1368bad: 1369 vput(tdvp); 1370 if (tvp != NULLVP) 1371 vput(tvp); 1372 vrele(fdvp); 1373 vrele(fvp); 1374 1375out: 1376 if (fdvp != ap->a_fdvp) { 1377 vrele(ap->a_fdvp); 1378 } 1379 if (fvp != ap->a_fvp) { 1380 vrele(ap->a_fvp); 1381 } 1382 if (tdvp != ap->a_tdvp) { 1383 vrele(ap->a_tdvp); 1384 } 1385 if (tvp != ap->a_tvp) { 1386 vrele(ap->a_tvp); 1387 } 1388 return (error); 1389} 1390 1391int 1392union_mkdir(void *v) 1393{ 1394 struct vop_mkdir_v3_args /* { 1395 struct vnode *a_dvp; 1396 struct vnode **a_vpp; 1397 struct componentname *a_cnp; 1398 struct vattr *a_vap; 1399 } */ *ap = v; 1400 struct union_node *un = VTOUNION(ap->a_dvp); 1401 struct vnode *dvp = un->un_uppervp; 1402 struct componentname *cnp = ap->a_cnp; 1403 1404 if (dvp != NULLVP) { 1405 int error; 1406 struct vnode *vp; 1407 1408 vp = NULL; 1409 error = VOP_MKDIR(dvp, &vp, cnp, ap->a_vap); 1410 if (error) { 1411 vrele(ap->a_dvp); 1412 return (error); 1413 } 1414 1415 error = union_allocvp(ap->a_vpp, ap->a_dvp->v_mount, ap->a_dvp, 1416 NULLVP, cnp, vp, NULLVP, 1); 1417 if (error) 1418 vrele(vp); 1419 return (error); 1420 } 1421 1422 return (EROFS); 1423} 1424 1425int 1426union_rmdir(void *v) 1427{ 1428 struct vop_rmdir_v2_args /* { 1429 struct vnode *a_dvp; 1430 struct vnode *a_vp; 1431 struct componentname *a_cnp; 1432 } */ *ap = v; 1433 int error; 1434 struct union_node *dun = VTOUNION(ap->a_dvp); 1435 struct union_node *un = VTOUNION(ap->a_vp); 1436 struct componentname *cnp = ap->a_cnp; 1437 1438 if (dun->un_uppervp == NULLVP) 1439 panic("union rmdir: null upper vnode"); 1440 1441 error = union_check_rmdir(un, cnp->cn_cred); 1442 if (error) { 1443 vput(ap->a_vp); 1444 return error; 1445 } 1446 1447 if (un->un_uppervp != NULLVP) { 1448 struct vnode *dvp = dun->un_uppervp; 1449 struct vnode *vp = un->un_uppervp; 1450 1451 /* Account for VOP_RMDIR to vrele vp. */ 1452 vref(vp); 1453 if (union_dowhiteout(un, cnp->cn_cred)) 1454 cnp->cn_flags |= DOWHITEOUT; 1455 error = VOP_RMDIR(dvp, vp, ap->a_cnp); 1456 if (!error) 1457 union_removed_upper(un); 1458 vrele(ap->a_vp); 1459 } else { 1460 error = union_mkwhiteout( 1461 MOUNTTOUNIONMOUNT(UNIONTOV(dun)->v_mount), 1462 dun->un_uppervp, ap->a_cnp, un); 1463 vput(ap->a_vp); 1464 } 1465 1466 return (error); 1467} 1468 1469int 1470union_symlink(void *v) 1471{ 1472 struct vop_symlink_v3_args /* { 1473 struct vnode *a_dvp; 1474 struct vnode **a_vpp; 1475 struct componentname *a_cnp; 1476 struct vattr *a_vap; 1477 char *a_target; 1478 } */ *ap = v; 1479 struct union_node *un = VTOUNION(ap->a_dvp); 1480 struct vnode *dvp = un->un_uppervp; 1481 struct componentname *cnp = ap->a_cnp; 1482 1483 if (dvp != NULLVP) { 1484 int error; 1485 1486 error = VOP_SYMLINK(dvp, ap->a_vpp, cnp, ap->a_vap, 1487 ap->a_target); 1488 return (error); 1489 } 1490 1491 return (EROFS); 1492} 1493 1494/* 1495 * union_readdir works in concert with getdirentries and 1496 * readdir(3) to provide a list of entries in the unioned 1497 * directories. getdirentries is responsible for walking 1498 * down the union stack. readdir(3) is responsible for 1499 * eliminating duplicate names from the returned data stream. 1500 */ 1501int 1502union_readdir(void *v) 1503{ 1504 struct vop_readdir_args /* { 1505 struct vnodeop_desc *a_desc; 1506 struct vnode *a_vp; 1507 struct uio *a_uio; 1508 kauth_cred_t a_cred; 1509 int *a_eofflag; 1510 u_long *a_cookies; 1511 int a_ncookies; 1512 } */ *ap = v; 1513 struct union_node *un = VTOUNION(ap->a_vp); 1514 struct vnode *vp; 1515 int dolock, error; 1516 1517 if (un->un_hooknode) { 1518 KASSERT(un->un_uppervp == NULLVP); 1519 KASSERT(un->un_lowervp != NULLVP); 1520 vp = un->un_lowervp; 1521 dolock = 1; 1522 } else { 1523 vp = un->un_uppervp; 1524 dolock = 0; 1525 } 1526 if (vp == NULLVP) 1527 return 0; 1528 1529 if (dolock) 1530 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1531 ap->a_vp = vp; 1532 error = VCALL(vp, VOFFSET(vop_readdir), ap); 1533 if (dolock) 1534 VOP_UNLOCK(vp); 1535 1536 return error; 1537} 1538 1539int 1540union_readlink(void *v) 1541{ 1542 struct vop_readlink_args /* { 1543 struct vnode *a_vp; 1544 struct uio *a_uio; 1545 kauth_cred_t a_cred; 1546 } */ *ap = v; 1547 int error; 1548 struct vnode *vp = OTHERVP(ap->a_vp); 1549 int dolock = (vp == LOWERVP(ap->a_vp)); 1550 1551 if (dolock) 1552 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1553 ap->a_vp = vp; 1554 error = VCALL(vp, VOFFSET(vop_readlink), ap); 1555 if (dolock) 1556 VOP_UNLOCK(vp); 1557 1558 return (error); 1559} 1560 1561int 1562union_abortop(void *v) 1563{ 1564 struct vop_abortop_args /* { 1565 struct vnode *a_dvp; 1566 struct componentname *a_cnp; 1567 } */ *ap = v; 1568 1569 KASSERT(UPPERVP(ap->a_dvp) != NULL); 1570 1571 ap->a_dvp = UPPERVP(ap->a_dvp); 1572 return VCALL(ap->a_dvp, VOFFSET(vop_abortop), ap); 1573} 1574 1575int 1576union_inactive(void *v) 1577{ 1578 struct vop_inactive_v2_args /* { 1579 const struct vnodeop_desc *a_desc; 1580 struct vnode *a_vp; 1581 bool *a_recycle; 1582 } */ *ap = v; 1583 struct vnode *vp = ap->a_vp; 1584 struct union_node *un = VTOUNION(vp); 1585 struct vnode **vpp; 1586 1587 /* 1588 * Do nothing (and _don't_ bypass). 1589 * Wait to vrele lowervp until reclaim, 1590 * so that until then our union_node is in the 1591 * cache and reusable. 1592 * 1593 * NEEDSWORK: Someday, consider inactive'ing 1594 * the lowervp and then trying to reactivate it 1595 * with capabilities (v_id) 1596 * like they do in the name lookup cache code. 1597 * That's too much work for now. 1598 */ 1599 1600 if (un->un_dircache != 0) { 1601 for (vpp = un->un_dircache; *vpp != NULLVP; vpp++) 1602 vrele(*vpp); 1603 free(un->un_dircache, M_TEMP); 1604 un->un_dircache = 0; 1605 } 1606 1607 *ap->a_recycle = ((un->un_cflags & UN_CACHED) == 0); 1608 1609 return (0); 1610} 1611 1612int 1613union_reclaim(void *v) 1614{ 1615 struct vop_reclaim_v2_args /* { 1616 struct vnode *a_vp; 1617 } */ *ap = v; 1618 struct vnode *vp = ap->a_vp; 1619 struct vnode *uvp = UPPERVP(vp); 1620 1621 VOP_UNLOCK(vp); 1622 1623 if (uvp != NULL) { 1624 mutex_enter(uvp->v_interlock); 1625 KASSERT(vp->v_interlock == uvp->v_interlock); 1626 uvp->v_writecount -= vp->v_writecount; 1627 mutex_exit(uvp->v_interlock); 1628 } 1629 1630 union_freevp(vp); 1631 1632 return (0); 1633} 1634 1635static int 1636union_lock1(struct vnode *vp, struct vnode *lockvp, int flags) 1637{ 1638 struct vop_lock_args ap; 1639 1640 ap.a_desc = VDESC(vop_lock); 1641 ap.a_vp = lockvp; 1642 ap.a_flags = flags; 1643 1644 if (lockvp == vp) 1645 return genfs_lock(&ap); 1646 else 1647 return VCALL(ap.a_vp, VOFFSET(vop_lock), &ap); 1648} 1649 1650static int 1651union_unlock1(struct vnode *vp, struct vnode *lockvp) 1652{ 1653 struct vop_unlock_args ap; 1654 1655 ap.a_desc = VDESC(vop_unlock); 1656 ap.a_vp = lockvp; 1657 1658 if (lockvp == vp) 1659 return genfs_unlock(&ap); 1660 else 1661 return VCALL(ap.a_vp, VOFFSET(vop_unlock), &ap); 1662} 1663 1664int 1665union_lock(void *v) 1666{ 1667 struct vop_lock_args /* { 1668 struct vnode *a_vp; 1669 int a_flags; 1670 } */ *ap = v; 1671 struct vnode *vp = ap->a_vp, *lockvp; 1672 struct union_node *un = VTOUNION(vp); 1673 int flags = ap->a_flags; 1674 int error; 1675 1676 if ((flags & LK_NOWAIT) != 0) { 1677 if (!mutex_tryenter(&un->un_lock)) 1678 return EBUSY; 1679 lockvp = LOCKVP(vp); 1680 error = union_lock1(vp, lockvp, flags); 1681 mutex_exit(&un->un_lock); 1682 if (error) 1683 return error; 1684 if (mutex_tryenter(vp->v_interlock)) { 1685 error = vdead_check(vp, VDEAD_NOWAIT); 1686 mutex_exit(vp->v_interlock); 1687 } else 1688 error = EBUSY; 1689 if (error) 1690 union_unlock1(vp, lockvp); 1691 return error; 1692 } 1693 1694 mutex_enter(&un->un_lock); 1695 for (;;) { 1696 lockvp = LOCKVP(vp); 1697 mutex_exit(&un->un_lock); 1698 error = union_lock1(vp, lockvp, flags); 1699 if (error != 0) 1700 return error; 1701 mutex_enter(&un->un_lock); 1702 if (lockvp == LOCKVP(vp)) 1703 break; 1704 union_unlock1(vp, lockvp); 1705 } 1706 mutex_exit(&un->un_lock); 1707 1708 mutex_enter(vp->v_interlock); 1709 error = vdead_check(vp, VDEAD_NOWAIT); 1710 if (error) { 1711 union_unlock1(vp, lockvp); 1712 error = vdead_check(vp, 0); 1713 KASSERT(error == ENOENT); 1714 } 1715 mutex_exit(vp->v_interlock); 1716 return error; 1717} 1718 1719int 1720union_unlock(void *v) 1721{ 1722 struct vop_unlock_args /* { 1723 struct vnode *a_vp; 1724 int a_flags; 1725 } */ *ap = v; 1726 struct vnode *vp = ap->a_vp, *lockvp; 1727 1728 lockvp = LOCKVP(vp); 1729 union_unlock1(vp, lockvp); 1730 1731 return 0; 1732} 1733 1734int 1735union_bmap(void *v) 1736{ 1737 struct vop_bmap_args /* { 1738 struct vnode *a_vp; 1739 daddr_t a_bn; 1740 struct vnode **a_vpp; 1741 daddr_t *a_bnp; 1742 int *a_runp; 1743 } */ *ap = v; 1744 int error; 1745 struct vnode *vp = OTHERVP(ap->a_vp); 1746 int dolock = (vp == LOWERVP(ap->a_vp)); 1747 1748 if (dolock) 1749 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1750 ap->a_vp = vp; 1751 error = VCALL(vp, VOFFSET(vop_bmap), ap); 1752 if (dolock) 1753 VOP_UNLOCK(vp); 1754 1755 return (error); 1756} 1757 1758int 1759union_print(void *v) 1760{ 1761 struct vop_print_args /* { 1762 struct vnode *a_vp; 1763 } */ *ap = v; 1764 struct vnode *vp = ap->a_vp; 1765 1766 printf("\ttag VT_UNION, vp=%p, uppervp=%p, lowervp=%p\n", 1767 vp, UPPERVP(vp), LOWERVP(vp)); 1768 if (UPPERVP(vp) != NULLVP) 1769 vprint("union: upper", UPPERVP(vp)); 1770 if (LOWERVP(vp) != NULLVP) 1771 vprint("union: lower", LOWERVP(vp)); 1772 if (VTOUNION(vp)->un_dircache) { 1773 struct vnode **vpp; 1774 for (vpp = VTOUNION(vp)->un_dircache; *vpp != NULLVP; vpp++) 1775 vprint("dircache:", *vpp); 1776 } 1777 1778 return (0); 1779} 1780 1781int 1782union_islocked(void *v) 1783{ 1784 struct vop_islocked_args /* { 1785 struct vnode *a_vp; 1786 } */ *ap = v; 1787 struct vnode *vp; 1788 struct union_node *un; 1789 1790 un = VTOUNION(ap->a_vp); 1791 mutex_enter(&un->un_lock); 1792 vp = LOCKVP(ap->a_vp); 1793 mutex_exit(&un->un_lock); 1794 1795 if (vp == ap->a_vp) 1796 return genfs_islocked(ap); 1797 else 1798 return VOP_ISLOCKED(vp); 1799} 1800 1801int 1802union_pathconf(void *v) 1803{ 1804 struct vop_pathconf_args /* { 1805 struct vnode *a_vp; 1806 int a_name; 1807 int *a_retval; 1808 } */ *ap = v; 1809 int error; 1810 struct vnode *vp = OTHERVP(ap->a_vp); 1811 int dolock = (vp == LOWERVP(ap->a_vp)); 1812 1813 if (dolock) 1814 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1815 ap->a_vp = vp; 1816 error = VCALL(vp, VOFFSET(vop_pathconf), ap); 1817 if (dolock) 1818 VOP_UNLOCK(vp); 1819 1820 return (error); 1821} 1822 1823int 1824union_advlock(void *v) 1825{ 1826 struct vop_advlock_args /* { 1827 struct vnode *a_vp; 1828 void *a_id; 1829 int a_op; 1830 struct flock *a_fl; 1831 int a_flags; 1832 } */ *ap = v; 1833 struct vnode *ovp = OTHERVP(ap->a_vp); 1834 1835 ap->a_vp = ovp; 1836 return (VCALL(ovp, VOFFSET(vop_advlock), ap)); 1837} 1838 1839int 1840union_strategy(void *v) 1841{ 1842 struct vop_strategy_args /* { 1843 struct vnode *a_vp; 1844 struct buf *a_bp; 1845 } */ *ap = v; 1846 struct vnode *ovp = OTHERVP(ap->a_vp); 1847 struct buf *bp = ap->a_bp; 1848 1849 KASSERT(ovp != NULLVP); 1850 if (!NODE_IS_SPECIAL(ovp)) 1851 KASSERT((bp->b_flags & B_READ) || ovp != LOWERVP(bp->b_vp)); 1852 1853 return (VOP_STRATEGY(ovp, bp)); 1854} 1855 1856int 1857union_bwrite(void *v) 1858{ 1859 struct vop_bwrite_args /* { 1860 struct vnode *a_vp; 1861 struct buf *a_bp; 1862 } */ *ap = v; 1863 struct vnode *ovp = OTHERVP(ap->a_vp); 1864 struct buf *bp = ap->a_bp; 1865 1866 KASSERT(ovp != NULLVP); 1867 if (!NODE_IS_SPECIAL(ovp)) 1868 KASSERT((bp->b_flags & B_READ) || ovp != LOWERVP(bp->b_vp)); 1869 1870 return (VOP_BWRITE(ovp, bp)); 1871} 1872 1873int 1874union_getpages(void *v) 1875{ 1876 struct vop_getpages_args /* { 1877 struct vnode *a_vp; 1878 voff_t a_offset; 1879 struct vm_page **a_m; 1880 int *a_count; 1881 int a_centeridx; 1882 vm_prot_t a_access_type; 1883 int a_advice; 1884 int a_flags; 1885 } */ *ap = v; 1886 struct vnode *vp = ap->a_vp; 1887 1888 KASSERT(rw_lock_held(vp->v_uobj.vmobjlock)); 1889 1890 if (ap->a_flags & PGO_LOCKED) { 1891 return EBUSY; 1892 } 1893 ap->a_vp = OTHERVP(vp); 1894 KASSERT(vp->v_uobj.vmobjlock == ap->a_vp->v_uobj.vmobjlock); 1895 1896 /* Just pass the request on to the underlying layer. */ 1897 return VCALL(ap->a_vp, VOFFSET(vop_getpages), ap); 1898} 1899 1900int 1901union_putpages(void *v) 1902{ 1903 struct vop_putpages_args /* { 1904 struct vnode *a_vp; 1905 voff_t a_offlo; 1906 voff_t a_offhi; 1907 int a_flags; 1908 } */ *ap = v; 1909 struct vnode *vp = ap->a_vp; 1910 1911 KASSERT(rw_lock_held(vp->v_uobj.vmobjlock)); 1912 1913 ap->a_vp = OTHERVP(vp); 1914 KASSERT(vp->v_uobj.vmobjlock == ap->a_vp->v_uobj.vmobjlock); 1915 1916 if (ap->a_flags & PGO_RECLAIM) { 1917 rw_exit(vp->v_uobj.vmobjlock); 1918 return 0; 1919 } 1920 1921 /* Just pass the request on to the underlying layer. */ 1922 return VCALL(ap->a_vp, VOFFSET(vop_putpages), ap); 1923} 1924 1925int 1926union_kqfilter(void *v) 1927{ 1928 struct vop_kqfilter_args /* { 1929 struct vnode *a_vp; 1930 struct knote *a_kn; 1931 } */ *ap = v; 1932 int error; 1933 1934 /* 1935 * We watch either the upper layer file (if it already exists), 1936 * or the lower layer one. If there is lower layer file only 1937 * at this moment, we will keep watching that lower layer file 1938 * even if upper layer file would be created later on. 1939 */ 1940 if (UPPERVP(ap->a_vp)) 1941 error = VOP_KQFILTER(UPPERVP(ap->a_vp), ap->a_kn); 1942 else if (LOWERVP(ap->a_vp)) 1943 error = VOP_KQFILTER(LOWERVP(ap->a_vp), ap->a_kn); 1944 else { 1945 /* panic? */ 1946 error = EOPNOTSUPP; 1947 } 1948 1949 return (error); 1950} 1951