union_vnops.c revision 1.73
1/* $NetBSD: union_vnops.c,v 1.73 2020/05/16 18:31:50 christos 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 1.73 2020/05/16 18:31:50 christos 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_lookup(void *); 97int union_create(void *); 98int union_whiteout(void *); 99int union_mknod(void *); 100int union_open(void *); 101int union_close(void *); 102int union_access(void *); 103int union_getattr(void *); 104int union_setattr(void *); 105int union_read(void *); 106int union_write(void *); 107int union_ioctl(void *); 108int union_poll(void *); 109int union_revoke(void *); 110int union_mmap(void *); 111int union_fsync(void *); 112int union_seek(void *); 113int union_remove(void *); 114int union_link(void *); 115int union_rename(void *); 116int union_mkdir(void *); 117int union_rmdir(void *); 118int union_symlink(void *); 119int union_readdir(void *); 120int union_readlink(void *); 121int union_abortop(void *); 122int union_inactive(void *); 123int union_reclaim(void *); 124int union_lock(void *); 125int union_unlock(void *); 126int union_bmap(void *); 127int union_print(void *); 128int union_islocked(void *); 129int union_pathconf(void *); 130int union_advlock(void *); 131int union_strategy(void *); 132int union_bwrite(void *); 133int union_getpages(void *); 134int union_putpages(void *); 135int union_kqfilter(void *); 136 137static int union_lookup1(struct vnode *, struct vnode **, 138 struct vnode **, struct componentname *); 139 140 141/* 142 * Global vfs data structures 143 */ 144int (**union_vnodeop_p)(void *); 145const struct vnodeopv_entry_desc union_vnodeop_entries[] = { 146 { &vop_default_desc, vn_default_error }, 147 { &vop_lookup_desc, union_lookup }, /* lookup */ 148 { &vop_create_desc, union_create }, /* create */ 149 { &vop_whiteout_desc, union_whiteout }, /* whiteout */ 150 { &vop_mknod_desc, union_mknod }, /* mknod */ 151 { &vop_open_desc, union_open }, /* open */ 152 { &vop_close_desc, union_close }, /* close */ 153 { &vop_access_desc, union_access }, /* access */ 154 { &vop_accessx_desc, genfs_accessx }, /* accessx */ 155 { &vop_getattr_desc, union_getattr }, /* getattr */ 156 { &vop_setattr_desc, union_setattr }, /* setattr */ 157 { &vop_read_desc, union_read }, /* read */ 158 { &vop_write_desc, union_write }, /* write */ 159 { &vop_fallocate_desc, genfs_eopnotsupp }, /* fallocate */ 160 { &vop_fdiscard_desc, genfs_eopnotsupp }, /* fdiscard */ 161 { &vop_ioctl_desc, union_ioctl }, /* ioctl */ 162 { &vop_poll_desc, union_poll }, /* select */ 163 { &vop_revoke_desc, union_revoke }, /* revoke */ 164 { &vop_mmap_desc, union_mmap }, /* mmap */ 165 { &vop_fsync_desc, union_fsync }, /* fsync */ 166 { &vop_seek_desc, union_seek }, /* seek */ 167 { &vop_remove_desc, union_remove }, /* remove */ 168 { &vop_link_desc, union_link }, /* link */ 169 { &vop_rename_desc, union_rename }, /* rename */ 170 { &vop_mkdir_desc, union_mkdir }, /* mkdir */ 171 { &vop_rmdir_desc, union_rmdir }, /* rmdir */ 172 { &vop_symlink_desc, union_symlink }, /* symlink */ 173 { &vop_readdir_desc, union_readdir }, /* readdir */ 174 { &vop_readlink_desc, union_readlink }, /* readlink */ 175 { &vop_abortop_desc, union_abortop }, /* abortop */ 176 { &vop_inactive_desc, union_inactive }, /* inactive */ 177 { &vop_reclaim_desc, union_reclaim }, /* reclaim */ 178 { &vop_lock_desc, union_lock }, /* lock */ 179 { &vop_unlock_desc, union_unlock }, /* unlock */ 180 { &vop_bmap_desc, union_bmap }, /* bmap */ 181 { &vop_strategy_desc, union_strategy }, /* strategy */ 182 { &vop_bwrite_desc, union_bwrite }, /* bwrite */ 183 { &vop_print_desc, union_print }, /* print */ 184 { &vop_islocked_desc, union_islocked }, /* islocked */ 185 { &vop_pathconf_desc, union_pathconf }, /* pathconf */ 186 { &vop_advlock_desc, union_advlock }, /* advlock */ 187 { &vop_getpages_desc, union_getpages }, /* getpages */ 188 { &vop_putpages_desc, union_putpages }, /* putpages */ 189 { &vop_kqfilter_desc, union_kqfilter }, /* kqfilter */ 190 { NULL, NULL } 191}; 192const struct vnodeopv_desc union_vnodeop_opv_desc = 193 { &union_vnodeop_p, union_vnodeop_entries }; 194 195#define NODE_IS_SPECIAL(vp) \ 196 ((vp)->v_type == VBLK || (vp)->v_type == VCHR || \ 197 (vp)->v_type == VSOCK || (vp)->v_type == VFIFO) 198 199static int 200union_lookup1(struct vnode *udvp, struct vnode **dvpp, struct vnode **vpp, 201 struct componentname *cnp) 202{ 203 int error; 204 struct vnode *tdvp; 205 struct vnode *dvp; 206 struct mount *mp; 207 208 dvp = *dvpp; 209 210 /* 211 * If stepping up the directory tree, check for going 212 * back across the mount point, in which case do what 213 * lookup would do by stepping back down the mount 214 * hierarchy. 215 */ 216 if (cnp->cn_flags & ISDOTDOT) { 217 while ((dvp != udvp) && (dvp->v_vflag & VV_ROOT)) { 218 /* 219 * Don't do the NOCROSSMOUNT check 220 * at this level. By definition, 221 * union fs deals with namespaces, not 222 * filesystems. 223 */ 224 tdvp = dvp; 225 *dvpp = dvp = dvp->v_mount->mnt_vnodecovered; 226 VOP_UNLOCK(tdvp); 227 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 228 } 229 } 230 231 error = VOP_LOOKUP(dvp, &tdvp, cnp); 232 if (error) 233 return (error); 234 if (dvp != tdvp) { 235 if (cnp->cn_flags & ISDOTDOT) 236 VOP_UNLOCK(dvp); 237 error = vn_lock(tdvp, LK_EXCLUSIVE); 238 if (cnp->cn_flags & ISDOTDOT) 239 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 240 if (error) { 241 vrele(tdvp); 242 return error; 243 } 244 dvp = tdvp; 245 } 246 247 /* 248 * Lastly check if the current node is a mount point in 249 * which case walk up the mount hierarchy making sure not to 250 * bump into the root of the mount tree (ie. dvp != udvp). 251 */ 252 while (dvp != udvp && (dvp->v_type == VDIR) && 253 (mp = dvp->v_mountedhere)) { 254 if (vfs_busy(mp)) 255 continue; 256 vput(dvp); 257 error = VFS_ROOT(mp, LK_EXCLUSIVE, &tdvp); 258 vfs_unbusy(mp); 259 if (error) { 260 return (error); 261 } 262 dvp = tdvp; 263 } 264 265 *vpp = dvp; 266 return (0); 267} 268 269int 270union_lookup(void *v) 271{ 272 struct vop_lookup_v2_args /* { 273 struct vnodeop_desc *a_desc; 274 struct vnode *a_dvp; 275 struct vnode **a_vpp; 276 struct componentname *a_cnp; 277 } */ *ap = v; 278 int error; 279 int uerror, lerror; 280 struct vnode *uppervp, *lowervp; 281 struct vnode *upperdvp, *lowerdvp; 282 struct vnode *dvp = ap->a_dvp; 283 struct union_node *dun = VTOUNION(dvp); 284 struct componentname *cnp = ap->a_cnp; 285 struct union_mount *um = MOUNTTOUNIONMOUNT(dvp->v_mount); 286 kauth_cred_t saved_cred = NULL; 287 int iswhiteout; 288 struct vattr va; 289 290#ifdef notyet 291 if (cnp->cn_namelen == 3 && 292 cnp->cn_nameptr[2] == '.' && 293 cnp->cn_nameptr[1] == '.' && 294 cnp->cn_nameptr[0] == '.') { 295 dvp = *ap->a_vpp = LOWERVP(ap->a_dvp); 296 if (dvp == NULLVP) 297 return (ENOENT); 298 vref(dvp); 299 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 300 return (0); 301 } 302#endif 303 304 if ((cnp->cn_flags & ISLASTCN) && 305 (dvp->v_mount->mnt_flag & MNT_RDONLY) && 306 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) 307 return (EROFS); 308 309start: 310 upperdvp = dun->un_uppervp; 311 lowerdvp = dun->un_lowervp; 312 uppervp = NULLVP; 313 lowervp = NULLVP; 314 iswhiteout = 0; 315 316 /* 317 * do the lookup in the upper level. 318 * if that level comsumes additional pathnames, 319 * then assume that something special is going 320 * on and just return that vnode. 321 */ 322 if (upperdvp != NULLVP) { 323 uerror = union_lookup1(um->um_uppervp, &upperdvp, 324 &uppervp, cnp); 325 if (cnp->cn_consume != 0) { 326 if (uppervp != upperdvp) 327 VOP_UNLOCK(uppervp); 328 *ap->a_vpp = uppervp; 329 return (uerror); 330 } 331 if (uerror == ENOENT || uerror == EJUSTRETURN) { 332 if (cnp->cn_flags & ISWHITEOUT) { 333 iswhiteout = 1; 334 } else if (lowerdvp != NULLVP) { 335 lerror = VOP_GETATTR(upperdvp, &va, 336 cnp->cn_cred); 337 if (lerror == 0 && (va.va_flags & OPAQUE)) 338 iswhiteout = 1; 339 } 340 } 341 } else { 342 uerror = ENOENT; 343 } 344 345 /* 346 * in a similar way to the upper layer, do the lookup 347 * in the lower layer. this time, if there is some 348 * component magic going on, then vput whatever we got 349 * back from the upper layer and return the lower vnode 350 * instead. 351 */ 352 if (lowerdvp != NULLVP && !iswhiteout) { 353 int nameiop; 354 355 vn_lock(lowerdvp, LK_EXCLUSIVE | LK_RETRY); 356 357 /* 358 * Only do a LOOKUP on the bottom node, since 359 * we won't be making changes to it anyway. 360 */ 361 nameiop = cnp->cn_nameiop; 362 cnp->cn_nameiop = LOOKUP; 363 if (um->um_op == UNMNT_BELOW) { 364 saved_cred = cnp->cn_cred; 365 cnp->cn_cred = um->um_cred; 366 } 367 368 /* 369 * we shouldn't have to worry about locking interactions 370 * between the lower layer and our union layer (w.r.t. 371 * `..' processing) because we don't futz with lowervp 372 * locks in the union-node instantiation code path. 373 */ 374 lerror = union_lookup1(um->um_lowervp, &lowerdvp, 375 &lowervp, cnp); 376 if (um->um_op == UNMNT_BELOW) 377 cnp->cn_cred = saved_cred; 378 cnp->cn_nameiop = nameiop; 379 380 if (lowervp != lowerdvp) 381 VOP_UNLOCK(lowerdvp); 382 383 if (cnp->cn_consume != 0) { 384 if (uppervp != NULLVP) { 385 if (uppervp == upperdvp) 386 vrele(uppervp); 387 else 388 vput(uppervp); 389 uppervp = NULLVP; 390 } 391 *ap->a_vpp = lowervp; 392 return (lerror); 393 } 394 } else { 395 lerror = ENOENT; 396 if ((cnp->cn_flags & ISDOTDOT) && dun->un_pvp != NULLVP) { 397 lowervp = LOWERVP(dun->un_pvp); 398 if (lowervp != NULLVP) { 399 vref(lowervp); 400 vn_lock(lowervp, LK_EXCLUSIVE | LK_RETRY); 401 lerror = 0; 402 } 403 } 404 } 405 406 /* 407 * EJUSTRETURN is used by underlying filesystems to indicate that 408 * a directory modification op was started successfully. 409 * This will only happen in the upper layer, since 410 * the lower layer only does LOOKUPs. 411 * If this union is mounted read-only, bounce it now. 412 */ 413 414 if ((uerror == EJUSTRETURN) && (cnp->cn_flags & ISLASTCN) && 415 (dvp->v_mount->mnt_flag & MNT_RDONLY) && 416 ((cnp->cn_nameiop == CREATE) || (cnp->cn_nameiop == RENAME))) 417 uerror = EROFS; 418 419 /* 420 * at this point, we have uerror and lerror indicating 421 * possible errors with the lookups in the upper and lower 422 * layers. additionally, uppervp and lowervp are (locked) 423 * references to existing vnodes in the upper and lower layers. 424 * 425 * there are now three cases to consider. 426 * 1. if both layers returned an error, then return whatever 427 * error the upper layer generated. 428 * 429 * 2. if the top layer failed and the bottom layer succeeded 430 * then two subcases occur. 431 * a. the bottom vnode is not a directory, in which 432 * case just return a new union vnode referencing 433 * an empty top layer and the existing bottom layer. 434 * b. the bottom vnode is a directory, in which case 435 * create a new directory in the top-level and 436 * continue as in case 3. 437 * 438 * 3. if the top layer succeeded then return a new union 439 * vnode referencing whatever the new top layer and 440 * whatever the bottom layer returned. 441 */ 442 443 *ap->a_vpp = NULLVP; 444 445 446 /* case 1. */ 447 if ((uerror != 0) && (lerror != 0)) { 448 return (uerror); 449 } 450 451 /* case 2. */ 452 if (uerror != 0 /* && (lerror == 0) */ ) { 453 if (lowervp->v_type == VDIR) { /* case 2b. */ 454 /* 455 * We may be racing another process to make the 456 * upper-level shadow directory. Be careful with 457 * locks/etc! 458 * If we have to create a shadow directory and want 459 * to commit the node we have to restart the lookup 460 * to get the componentname right. 461 */ 462 if (upperdvp) { 463 VOP_UNLOCK(upperdvp); 464 uerror = union_mkshadow(um, upperdvp, cnp, 465 &uppervp); 466 vn_lock(upperdvp, LK_EXCLUSIVE | LK_RETRY); 467 if (uerror == 0 && cnp->cn_nameiop != LOOKUP) { 468 vrele(uppervp); 469 if (lowervp != NULLVP) 470 vput(lowervp); 471 goto start; 472 } 473 } 474 if (uerror) { 475 if (lowervp != NULLVP) { 476 vput(lowervp); 477 lowervp = NULLVP; 478 } 479 return (uerror); 480 } 481 } 482 } else { /* uerror == 0 */ 483 if (uppervp != upperdvp) 484 VOP_UNLOCK(uppervp); 485 } 486 487 if (lowervp != NULLVP) 488 VOP_UNLOCK(lowervp); 489 490 error = union_allocvp(ap->a_vpp, dvp->v_mount, dvp, upperdvp, cnp, 491 uppervp, lowervp, 1); 492 493 if (error) { 494 if (uppervp != NULLVP) 495 vrele(uppervp); 496 if (lowervp != NULLVP) 497 vrele(lowervp); 498 return error; 499 } 500 501 return 0; 502} 503 504int 505union_create(void *v) 506{ 507 struct vop_create_v3_args /* { 508 struct vnode *a_dvp; 509 struct vnode **a_vpp; 510 struct componentname *a_cnp; 511 struct vattr *a_vap; 512 } */ *ap = v; 513 struct union_node *un = VTOUNION(ap->a_dvp); 514 struct vnode *dvp = un->un_uppervp; 515 struct componentname *cnp = ap->a_cnp; 516 517 if (dvp != NULLVP) { 518 int error; 519 struct vnode *vp; 520 struct mount *mp; 521 522 mp = ap->a_dvp->v_mount; 523 524 vp = NULL; 525 error = VOP_CREATE(dvp, &vp, cnp, ap->a_vap); 526 if (error) 527 return (error); 528 529 error = union_allocvp(ap->a_vpp, mp, NULLVP, NULLVP, cnp, vp, 530 NULLVP, 1); 531 if (error) 532 vrele(vp); 533 return (error); 534 } 535 536 return (EROFS); 537} 538 539int 540union_whiteout(void *v) 541{ 542 struct vop_whiteout_args /* { 543 struct vnode *a_dvp; 544 struct componentname *a_cnp; 545 int a_flags; 546 } */ *ap = v; 547 struct union_node *un = VTOUNION(ap->a_dvp); 548 struct componentname *cnp = ap->a_cnp; 549 550 if (un->un_uppervp == NULLVP) 551 return (EOPNOTSUPP); 552 553 return (VOP_WHITEOUT(un->un_uppervp, cnp, ap->a_flags)); 554} 555 556int 557union_mknod(void *v) 558{ 559 struct vop_mknod_v3_args /* { 560 struct vnode *a_dvp; 561 struct vnode **a_vpp; 562 struct componentname *a_cnp; 563 struct vattr *a_vap; 564 } */ *ap = v; 565 struct union_node *un = VTOUNION(ap->a_dvp); 566 struct vnode *dvp = un->un_uppervp; 567 struct componentname *cnp = ap->a_cnp; 568 569 if (dvp != NULLVP) { 570 int error; 571 struct vnode *vp; 572 struct mount *mp; 573 574 mp = ap->a_dvp->v_mount; 575 error = VOP_MKNOD(dvp, &vp, cnp, ap->a_vap); 576 if (error) 577 return (error); 578 579 error = union_allocvp(ap->a_vpp, mp, NULLVP, NULLVP, 580 cnp, vp, NULLVP, 1); 581 if (error) 582 vrele(vp); 583 return (error); 584 } 585 586 return (EROFS); 587} 588 589int 590union_open(void *v) 591{ 592 struct vop_open_args /* { 593 struct vnodeop_desc *a_desc; 594 struct vnode *a_vp; 595 int a_mode; 596 kauth_cred_t a_cred; 597 } */ *ap = v; 598 struct union_node *un = VTOUNION(ap->a_vp); 599 struct vnode *tvp; 600 int mode = ap->a_mode; 601 kauth_cred_t cred = ap->a_cred; 602 struct lwp *l = curlwp; 603 int error; 604 605 /* 606 * If there is an existing upper vp then simply open that. 607 */ 608 tvp = un->un_uppervp; 609 if (tvp == NULLVP) { 610 /* 611 * If the lower vnode is being opened for writing, then 612 * copy the file contents to the upper vnode and open that, 613 * otherwise can simply open the lower vnode. 614 */ 615 tvp = un->un_lowervp; 616 if ((ap->a_mode & FWRITE) && (tvp->v_type == VREG)) { 617 error = union_copyup(un, (mode&O_TRUNC) == 0, cred, l); 618 if (error == 0) 619 error = VOP_OPEN(un->un_uppervp, mode, cred); 620 if (error == 0) { 621 mutex_enter(un->un_uppervp->v_interlock); 622 un->un_uppervp->v_writecount++; 623 mutex_exit(un->un_uppervp->v_interlock); 624 } 625 return (error); 626 } 627 628 /* 629 * Just open the lower vnode, but check for nodev mount flag 630 */ 631 if ((tvp->v_type == VBLK || tvp->v_type == VCHR) && 632 (ap->a_vp->v_mount->mnt_flag & MNT_NODEV)) 633 return ENXIO; 634 un->un_openl++; 635 vn_lock(tvp, LK_EXCLUSIVE | LK_RETRY); 636 error = VOP_OPEN(tvp, mode, cred); 637 VOP_UNLOCK(tvp); 638 639 return (error); 640 } 641 /* 642 * Just open the upper vnode, checking for nodev mount flag first 643 */ 644 if ((tvp->v_type == VBLK || tvp->v_type == VCHR) && 645 (ap->a_vp->v_mount->mnt_flag & MNT_NODEV)) 646 return ENXIO; 647 648 error = VOP_OPEN(tvp, mode, cred); 649 if (error == 0 && (ap->a_mode & FWRITE)) { 650 mutex_enter(tvp->v_interlock); 651 tvp->v_writecount++; 652 mutex_exit(tvp->v_interlock); 653 } 654 655 return (error); 656} 657 658int 659union_close(void *v) 660{ 661 struct vop_close_args /* { 662 struct vnode *a_vp; 663 int a_fflag; 664 kauth_cred_t a_cred; 665 } */ *ap = v; 666 struct union_node *un = VTOUNION(ap->a_vp); 667 struct vnode *vp; 668 int error; 669 bool do_lock; 670 671 vp = un->un_uppervp; 672 if (vp != NULLVP) { 673 do_lock = false; 674 } else { 675 KASSERT(un->un_openl > 0); 676 --un->un_openl; 677 vp = un->un_lowervp; 678 do_lock = true; 679 } 680 681 KASSERT(vp != NULLVP); 682 ap->a_vp = vp; 683 if ((ap->a_fflag & FWRITE)) { 684 KASSERT(vp == un->un_uppervp); 685 mutex_enter(vp->v_interlock); 686 vp->v_writecount--; 687 mutex_exit(vp->v_interlock); 688 } 689 if (do_lock) 690 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 691 error = VCALL(vp, VOFFSET(vop_close), ap); 692 if (do_lock) 693 VOP_UNLOCK(vp); 694 695 return error; 696} 697 698/* 699 * Check access permission on the union vnode. 700 * The access check being enforced is to check 701 * against both the underlying vnode, and any 702 * copied vnode. This ensures that no additional 703 * file permissions are given away simply because 704 * the user caused an implicit file copy. 705 */ 706int 707union_access(void *v) 708{ 709 struct vop_access_args /* { 710 struct vnodeop_desc *a_desc; 711 struct vnode *a_vp; 712 accmode_t a_accmode; 713 kauth_cred_t a_cred; 714 } */ *ap = v; 715 struct vnode *vp = ap->a_vp; 716 struct union_node *un = VTOUNION(vp); 717 int error = EACCES; 718 struct union_mount *um = MOUNTTOUNIONMOUNT(vp->v_mount); 719 720 /* 721 * Disallow write attempts on read-only file systems; 722 * unless the file is a socket, fifo, or a block or 723 * character device resident on the file system. 724 */ 725 if (ap->a_accmode & VWRITE) { 726 switch (vp->v_type) { 727 case VDIR: 728 case VLNK: 729 case VREG: 730 if (vp->v_mount->mnt_flag & MNT_RDONLY) 731 return (EROFS); 732 break; 733 case VBAD: 734 case VBLK: 735 case VCHR: 736 case VSOCK: 737 case VFIFO: 738 case VNON: 739 default: 740 break; 741 } 742 } 743 744 745 if ((vp = un->un_uppervp) != NULLVP) { 746 ap->a_vp = vp; 747 return (VCALL(vp, VOFFSET(vop_access), ap)); 748 } 749 750 if ((vp = un->un_lowervp) != NULLVP) { 751 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 752 ap->a_vp = vp; 753 error = VCALL(vp, VOFFSET(vop_access), ap); 754 if (error == 0) { 755 if (um->um_op == UNMNT_BELOW) { 756 ap->a_cred = um->um_cred; 757 error = VCALL(vp, VOFFSET(vop_access), ap); 758 } 759 } 760 VOP_UNLOCK(vp); 761 if (error) 762 return (error); 763 } 764 765 return (error); 766} 767 768/* 769 * We handle getattr only to change the fsid and 770 * track object sizes 771 */ 772int 773union_getattr(void *v) 774{ 775 struct vop_getattr_args /* { 776 struct vnode *a_vp; 777 struct vattr *a_vap; 778 kauth_cred_t a_cred; 779 } */ *ap = v; 780 int error; 781 struct union_node *un = VTOUNION(ap->a_vp); 782 struct vnode *vp = un->un_uppervp; 783 struct vattr *vap; 784 struct vattr va; 785 786 787 /* 788 * Some programs walk the filesystem hierarchy by counting 789 * links to directories to avoid stat'ing all the time. 790 * This means the link count on directories needs to be "correct". 791 * The only way to do that is to call getattr on both layers 792 * and fix up the link count. The link count will not necessarily 793 * be accurate but will be large enough to defeat the tree walkers. 794 * 795 * To make life more interesting, some filesystems don't keep 796 * track of link counts in the expected way, and return a 797 * link count of `1' for those directories; if either of the 798 * component directories returns a link count of `1', we return a 1. 799 */ 800 801 vap = ap->a_vap; 802 803 vp = un->un_uppervp; 804 if (vp != NULLVP) { 805 error = VOP_GETATTR(vp, vap, ap->a_cred); 806 if (error) 807 return (error); 808 mutex_enter(&un->un_lock); 809 union_newsize(ap->a_vp, vap->va_size, VNOVAL); 810 } 811 812 if (vp == NULLVP) { 813 vp = un->un_lowervp; 814 } else if (vp->v_type == VDIR) { 815 vp = un->un_lowervp; 816 if (vp != NULLVP) 817 vap = &va; 818 } else { 819 vp = NULLVP; 820 } 821 822 if (vp != NULLVP) { 823 if (vp == un->un_lowervp) 824 vn_lock(vp, LK_SHARED | LK_RETRY); 825 error = VOP_GETATTR(vp, vap, ap->a_cred); 826 if (vp == un->un_lowervp) 827 VOP_UNLOCK(vp); 828 if (error) 829 return (error); 830 mutex_enter(&un->un_lock); 831 union_newsize(ap->a_vp, VNOVAL, vap->va_size); 832 } 833 834 if ((vap != ap->a_vap) && (vap->va_type == VDIR)) { 835 /* 836 * Link count manipulation: 837 * - If both return "2", return 2 (no subdirs) 838 * - If one or the other return "1", return "1" (ENOCLUE) 839 */ 840 if ((ap->a_vap->va_nlink == 2) && 841 (vap->va_nlink == 2)) 842 ; 843 else if (ap->a_vap->va_nlink != 1) { 844 if (vap->va_nlink == 1) 845 ap->a_vap->va_nlink = 1; 846 else 847 ap->a_vap->va_nlink += vap->va_nlink; 848 } 849 } 850 ap->a_vap->va_fsid = ap->a_vp->v_mount->mnt_stat.f_fsidx.__fsid_val[0]; 851 return (0); 852} 853 854int 855union_setattr(void *v) 856{ 857 struct vop_setattr_args /* { 858 struct vnode *a_vp; 859 struct vattr *a_vap; 860 kauth_cred_t a_cred; 861 } */ *ap = v; 862 struct vattr *vap = ap->a_vap; 863 struct vnode *vp = ap->a_vp; 864 struct union_node *un = VTOUNION(vp); 865 bool size_only; /* All but va_size are VNOVAL. */ 866 int error; 867 868 size_only = (vap->va_flags == VNOVAL && vap->va_uid == (uid_t)VNOVAL && 869 vap->va_gid == (gid_t)VNOVAL && vap->va_atime.tv_sec == VNOVAL && 870 vap->va_mtime.tv_sec == VNOVAL && vap->va_mode == (mode_t)VNOVAL); 871 872 if (!size_only && (vp->v_mount->mnt_flag & MNT_RDONLY)) 873 return (EROFS); 874 if (vap->va_size != VNOVAL) { 875 switch (vp->v_type) { 876 case VDIR: 877 return (EISDIR); 878 case VCHR: 879 case VBLK: 880 case VSOCK: 881 case VFIFO: 882 break; 883 case VREG: 884 case VLNK: 885 default: 886 /* 887 * Disallow write attempts if the filesystem is 888 * mounted read-only. 889 */ 890 if (vp->v_mount->mnt_flag & MNT_RDONLY) 891 return (EROFS); 892 } 893 } 894 895 /* 896 * Handle case of truncating lower object to zero size, 897 * by creating a zero length upper object. This is to 898 * handle the case of open with O_TRUNC and O_CREAT. 899 */ 900 if ((un->un_uppervp == NULLVP) && 901 /* assert(un->un_lowervp != NULLVP) */ 902 (un->un_lowervp->v_type == VREG)) { 903 error = union_copyup(un, (vap->va_size != 0), 904 ap->a_cred, curlwp); 905 if (error) 906 return (error); 907 } 908 909 /* 910 * Try to set attributes in upper layer, ignore size change to zero 911 * for devices to handle O_TRUNC and return read-only filesystem error 912 * otherwise. 913 */ 914 if (un->un_uppervp != NULLVP) { 915 error = VOP_SETATTR(un->un_uppervp, vap, ap->a_cred); 916 if ((error == 0) && (vap->va_size != VNOVAL)) { 917 mutex_enter(&un->un_lock); 918 union_newsize(ap->a_vp, vap->va_size, VNOVAL); 919 } 920 } else { 921 KASSERT(un->un_lowervp != NULLVP); 922 if (NODE_IS_SPECIAL(un->un_lowervp)) { 923 if (size_only && 924 (vap->va_size == 0 || vap->va_size == VNOVAL)) 925 error = 0; 926 else 927 error = EROFS; 928 } else { 929 error = EROFS; 930 } 931 } 932 933 return (error); 934} 935 936int 937union_read(void *v) 938{ 939 struct vop_read_args /* { 940 struct vnode *a_vp; 941 struct uio *a_uio; 942 int a_ioflag; 943 kauth_cred_t a_cred; 944 } */ *ap = v; 945 int error; 946 struct vnode *vp = OTHERVP(ap->a_vp); 947 int dolock = (vp == LOWERVP(ap->a_vp)); 948 949 if (dolock) 950 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 951 error = VOP_READ(vp, ap->a_uio, ap->a_ioflag, ap->a_cred); 952 if (dolock) 953 VOP_UNLOCK(vp); 954 955 /* 956 * XXX 957 * perhaps the size of the underlying object has changed under 958 * our feet. take advantage of the offset information present 959 * in the uio structure. 960 */ 961 if (error == 0) { 962 struct union_node *un = VTOUNION(ap->a_vp); 963 off_t cur = ap->a_uio->uio_offset; 964 off_t usz = VNOVAL, lsz = VNOVAL; 965 966 mutex_enter(&un->un_lock); 967 if (vp == un->un_uppervp) { 968 if (cur > un->un_uppersz) 969 usz = cur; 970 } else { 971 if (cur > un->un_lowersz) 972 lsz = cur; 973 } 974 975 if (usz != VNOVAL || lsz != VNOVAL) 976 union_newsize(ap->a_vp, usz, lsz); 977 else 978 mutex_exit(&un->un_lock); 979 } 980 981 return (error); 982} 983 984int 985union_write(void *v) 986{ 987 struct vop_read_args /* { 988 struct vnode *a_vp; 989 struct uio *a_uio; 990 int a_ioflag; 991 kauth_cred_t a_cred; 992 } */ *ap = v; 993 int error; 994 struct vnode *vp; 995 struct union_node *un = VTOUNION(ap->a_vp); 996 997 vp = UPPERVP(ap->a_vp); 998 if (vp == NULLVP) { 999 vp = LOWERVP(ap->a_vp); 1000 if (NODE_IS_SPECIAL(vp)) { 1001 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1002 error = VOP_WRITE(vp, ap->a_uio, ap->a_ioflag, 1003 ap->a_cred); 1004 VOP_UNLOCK(vp); 1005 return error; 1006 } 1007 panic("union: missing upper layer in write"); 1008 } 1009 1010 error = VOP_WRITE(vp, ap->a_uio, ap->a_ioflag, ap->a_cred); 1011 1012 /* 1013 * the size of the underlying object may be changed by the 1014 * write. 1015 */ 1016 if (error == 0) { 1017 off_t cur = ap->a_uio->uio_offset; 1018 1019 mutex_enter(&un->un_lock); 1020 if (cur > un->un_uppersz) 1021 union_newsize(ap->a_vp, cur, VNOVAL); 1022 else 1023 mutex_exit(&un->un_lock); 1024 } 1025 1026 return (error); 1027} 1028 1029int 1030union_ioctl(void *v) 1031{ 1032 struct vop_ioctl_args /* { 1033 struct vnode *a_vp; 1034 int a_command; 1035 void *a_data; 1036 int a_fflag; 1037 kauth_cred_t a_cred; 1038 } */ *ap = v; 1039 struct vnode *ovp = OTHERVP(ap->a_vp); 1040 1041 ap->a_vp = ovp; 1042 return (VCALL(ovp, VOFFSET(vop_ioctl), ap)); 1043} 1044 1045int 1046union_poll(void *v) 1047{ 1048 struct vop_poll_args /* { 1049 struct vnode *a_vp; 1050 int a_events; 1051 } */ *ap = v; 1052 struct vnode *ovp = OTHERVP(ap->a_vp); 1053 1054 ap->a_vp = ovp; 1055 return (VCALL(ovp, VOFFSET(vop_poll), ap)); 1056} 1057 1058int 1059union_revoke(void *v) 1060{ 1061 struct vop_revoke_args /* { 1062 struct vnode *a_vp; 1063 int a_flags; 1064 struct proc *a_p; 1065 } */ *ap = v; 1066 struct vnode *vp = ap->a_vp; 1067 1068 if (UPPERVP(vp)) 1069 VOP_REVOKE(UPPERVP(vp), ap->a_flags); 1070 if (LOWERVP(vp)) 1071 VOP_REVOKE(LOWERVP(vp), ap->a_flags); 1072 vgone(vp); /* XXXAD?? */ 1073 return (0); 1074} 1075 1076int 1077union_mmap(void *v) 1078{ 1079 struct vop_mmap_args /* { 1080 struct vnode *a_vp; 1081 vm_prot_t a_prot; 1082 kauth_cred_t a_cred; 1083 } */ *ap = v; 1084 struct vnode *ovp = OTHERVP(ap->a_vp); 1085 1086 ap->a_vp = ovp; 1087 return (VCALL(ovp, VOFFSET(vop_mmap), ap)); 1088} 1089 1090int 1091union_fsync(void *v) 1092{ 1093 struct vop_fsync_args /* { 1094 struct vnode *a_vp; 1095 kauth_cred_t a_cred; 1096 int a_flags; 1097 off_t offhi; 1098 off_t offlo; 1099 } */ *ap = v; 1100 int error = 0; 1101 struct vnode *targetvp; 1102 1103 /* 1104 * If vinvalbuf is calling us, it's a "shallow fsync" -- don't 1105 * bother syncing the underlying vnodes, since (a) they'll be 1106 * fsync'ed when reclaimed and (b) we could deadlock if 1107 * they're locked; otherwise, pass it through to the 1108 * underlying layer. 1109 */ 1110 if (ap->a_vp->v_type == VBLK || ap->a_vp->v_type == VCHR) { 1111 error = spec_fsync(v); 1112 if (error) 1113 return error; 1114 } 1115 1116 if (ap->a_flags & FSYNC_RECLAIM) 1117 return 0; 1118 1119 targetvp = OTHERVP(ap->a_vp); 1120 if (targetvp != NULLVP) { 1121 int dolock = (targetvp == LOWERVP(ap->a_vp)); 1122 1123 if (dolock) 1124 vn_lock(targetvp, LK_EXCLUSIVE | LK_RETRY); 1125 error = VOP_FSYNC(targetvp, ap->a_cred, ap->a_flags, 1126 ap->a_offlo, ap->a_offhi); 1127 if (dolock) 1128 VOP_UNLOCK(targetvp); 1129 } 1130 1131 return (error); 1132} 1133 1134int 1135union_seek(void *v) 1136{ 1137 struct vop_seek_args /* { 1138 struct vnode *a_vp; 1139 off_t a_oldoff; 1140 off_t a_newoff; 1141 kauth_cred_t a_cred; 1142 } */ *ap = v; 1143 struct vnode *ovp = OTHERVP(ap->a_vp); 1144 1145 ap->a_vp = ovp; 1146 return (VCALL(ovp, VOFFSET(vop_seek), ap)); 1147} 1148 1149int 1150union_remove(void *v) 1151{ 1152 struct vop_remove_v2_args /* { 1153 struct vnode *a_dvp; 1154 struct vnode *a_vp; 1155 struct componentname *a_cnp; 1156 } */ *ap = v; 1157 int error; 1158 struct union_node *dun = VTOUNION(ap->a_dvp); 1159 struct union_node *un = VTOUNION(ap->a_vp); 1160 struct componentname *cnp = ap->a_cnp; 1161 1162 if (dun->un_uppervp == NULLVP) 1163 panic("union remove: null upper vnode"); 1164 1165 if (un->un_uppervp != NULLVP) { 1166 struct vnode *dvp = dun->un_uppervp; 1167 struct vnode *vp = un->un_uppervp; 1168 1169 /* Account for VOP_REMOVE to vrele vp. */ 1170 vref(vp); 1171 if (union_dowhiteout(un, cnp->cn_cred)) 1172 cnp->cn_flags |= DOWHITEOUT; 1173 error = VOP_REMOVE(dvp, vp, cnp); 1174 if (!error) 1175 union_removed_upper(un); 1176 vrele(ap->a_vp); 1177 } else { 1178 error = union_mkwhiteout( 1179 MOUNTTOUNIONMOUNT(UNIONTOV(dun)->v_mount), 1180 dun->un_uppervp, ap->a_cnp, un); 1181 vput(ap->a_vp); 1182 } 1183 1184 return (error); 1185} 1186 1187int 1188union_link(void *v) 1189{ 1190 struct vop_link_v2_args /* { 1191 struct vnode *a_dvp; 1192 struct vnode *a_vp; 1193 struct componentname *a_cnp; 1194 } */ *ap = v; 1195 int error = 0; 1196 struct componentname *cnp = ap->a_cnp; 1197 struct union_node *dun; 1198 struct vnode *vp; 1199 struct vnode *dvp; 1200 1201 dun = VTOUNION(ap->a_dvp); 1202 1203 KASSERT((ap->a_cnp->cn_flags & LOCKPARENT) != 0); 1204 1205 if (ap->a_dvp->v_op != ap->a_vp->v_op) { 1206 vp = ap->a_vp; 1207 } else { 1208 struct union_node *un = VTOUNION(ap->a_vp); 1209 if (un->un_uppervp == NULLVP) { 1210 const bool droplock = (dun->un_uppervp == un->un_dirvp); 1211 1212 /* 1213 * Needs to be copied before we can link it. 1214 */ 1215 vn_lock(ap->a_vp, LK_EXCLUSIVE | LK_RETRY); 1216 if (droplock) 1217 VOP_UNLOCK(dun->un_uppervp); 1218 error = union_copyup(un, 1, cnp->cn_cred, curlwp); 1219 if (droplock) { 1220 vn_lock(dun->un_uppervp, 1221 LK_EXCLUSIVE | LK_RETRY); 1222 /* 1223 * During copyup, we dropped the lock on the 1224 * dir and invalidated any saved namei lookup 1225 * state for the directory we'll be entering 1226 * the link in. We need to re-run the lookup 1227 * in that directory to reset any state needed 1228 * for VOP_LINK. 1229 * Call relookup on the union-layer to reset 1230 * the state. 1231 */ 1232 vp = NULLVP; 1233 if (dun->un_uppervp == NULLVP) 1234 panic("union: null upperdvp?"); 1235 error = relookup(ap->a_dvp, &vp, ap->a_cnp, 0); 1236 if (error) { 1237 VOP_UNLOCK(ap->a_vp); 1238 return EROFS; /* ? */ 1239 } 1240 if (vp != NULLVP) { 1241 /* 1242 * The name we want to create has 1243 * mysteriously appeared (a race?) 1244 */ 1245 error = EEXIST; 1246 VOP_UNLOCK(ap->a_vp); 1247 vput(vp); 1248 return (error); 1249 } 1250 } 1251 VOP_UNLOCK(ap->a_vp); 1252 } 1253 vp = un->un_uppervp; 1254 } 1255 1256 dvp = dun->un_uppervp; 1257 if (dvp == NULLVP) 1258 error = EROFS; 1259 1260 if (error) 1261 return (error); 1262 1263 return VOP_LINK(dvp, vp, cnp); 1264} 1265 1266int 1267union_rename(void *v) 1268{ 1269 struct vop_rename_args /* { 1270 struct vnode *a_fdvp; 1271 struct vnode *a_fvp; 1272 struct componentname *a_fcnp; 1273 struct vnode *a_tdvp; 1274 struct vnode *a_tvp; 1275 struct componentname *a_tcnp; 1276 } */ *ap = v; 1277 int error; 1278 1279 struct vnode *fdvp = ap->a_fdvp; 1280 struct vnode *fvp = ap->a_fvp; 1281 struct vnode *tdvp = ap->a_tdvp; 1282 struct vnode *tvp = ap->a_tvp; 1283 1284 /* 1285 * Account for VOP_RENAME to vrele all nodes. 1286 * Note: VOP_RENAME will unlock tdvp. 1287 */ 1288 1289 if (fdvp->v_op == union_vnodeop_p) { /* always true */ 1290 struct union_node *un = VTOUNION(fdvp); 1291 if (un->un_uppervp == NULLVP) { 1292 /* 1293 * this should never happen in normal 1294 * operation but might if there was 1295 * a problem creating the top-level shadow 1296 * directory. 1297 */ 1298 error = EXDEV; 1299 goto bad; 1300 } 1301 1302 fdvp = un->un_uppervp; 1303 vref(fdvp); 1304 } 1305 1306 if (fvp->v_op == union_vnodeop_p) { /* always true */ 1307 struct union_node *un = VTOUNION(fvp); 1308 if (un->un_uppervp == NULLVP) { 1309 /* XXX: should do a copyup */ 1310 error = EXDEV; 1311 goto bad; 1312 } 1313 1314 if (un->un_lowervp != NULLVP) 1315 ap->a_fcnp->cn_flags |= DOWHITEOUT; 1316 1317 fvp = un->un_uppervp; 1318 vref(fvp); 1319 } 1320 1321 if (tdvp->v_op == union_vnodeop_p) { 1322 struct union_node *un = VTOUNION(tdvp); 1323 if (un->un_uppervp == NULLVP) { 1324 /* 1325 * this should never happen in normal 1326 * operation but might if there was 1327 * a problem creating the top-level shadow 1328 * directory. 1329 */ 1330 error = EXDEV; 1331 goto bad; 1332 } 1333 1334 tdvp = un->un_uppervp; 1335 vref(tdvp); 1336 } 1337 1338 if (tvp != NULLVP && tvp->v_op == union_vnodeop_p) { 1339 struct union_node *un = VTOUNION(tvp); 1340 1341 tvp = un->un_uppervp; 1342 if (tvp != NULLVP) { 1343 vref(tvp); 1344 } 1345 } 1346 1347 error = VOP_RENAME(fdvp, fvp, ap->a_fcnp, tdvp, tvp, ap->a_tcnp); 1348 goto out; 1349 1350bad: 1351 vput(tdvp); 1352 if (tvp != NULLVP) 1353 vput(tvp); 1354 vrele(fdvp); 1355 vrele(fvp); 1356 1357out: 1358 if (fdvp != ap->a_fdvp) { 1359 vrele(ap->a_fdvp); 1360 } 1361 if (fvp != ap->a_fvp) { 1362 vrele(ap->a_fvp); 1363 } 1364 if (tdvp != ap->a_tdvp) { 1365 vrele(ap->a_tdvp); 1366 } 1367 if (tvp != ap->a_tvp) { 1368 vrele(ap->a_tvp); 1369 } 1370 return (error); 1371} 1372 1373int 1374union_mkdir(void *v) 1375{ 1376 struct vop_mkdir_v3_args /* { 1377 struct vnode *a_dvp; 1378 struct vnode **a_vpp; 1379 struct componentname *a_cnp; 1380 struct vattr *a_vap; 1381 } */ *ap = v; 1382 struct union_node *un = VTOUNION(ap->a_dvp); 1383 struct vnode *dvp = un->un_uppervp; 1384 struct componentname *cnp = ap->a_cnp; 1385 1386 if (dvp != NULLVP) { 1387 int error; 1388 struct vnode *vp; 1389 1390 vp = NULL; 1391 error = VOP_MKDIR(dvp, &vp, cnp, ap->a_vap); 1392 if (error) { 1393 vrele(ap->a_dvp); 1394 return (error); 1395 } 1396 1397 error = union_allocvp(ap->a_vpp, ap->a_dvp->v_mount, ap->a_dvp, 1398 NULLVP, cnp, vp, NULLVP, 1); 1399 if (error) 1400 vrele(vp); 1401 return (error); 1402 } 1403 1404 return (EROFS); 1405} 1406 1407int 1408union_rmdir(void *v) 1409{ 1410 struct vop_rmdir_v2_args /* { 1411 struct vnode *a_dvp; 1412 struct vnode *a_vp; 1413 struct componentname *a_cnp; 1414 } */ *ap = v; 1415 int error; 1416 struct union_node *dun = VTOUNION(ap->a_dvp); 1417 struct union_node *un = VTOUNION(ap->a_vp); 1418 struct componentname *cnp = ap->a_cnp; 1419 1420 if (dun->un_uppervp == NULLVP) 1421 panic("union rmdir: null upper vnode"); 1422 1423 error = union_check_rmdir(un, cnp->cn_cred); 1424 if (error) { 1425 vput(ap->a_vp); 1426 return error; 1427 } 1428 1429 if (un->un_uppervp != NULLVP) { 1430 struct vnode *dvp = dun->un_uppervp; 1431 struct vnode *vp = un->un_uppervp; 1432 1433 /* Account for VOP_RMDIR to vrele vp. */ 1434 vref(vp); 1435 if (union_dowhiteout(un, cnp->cn_cred)) 1436 cnp->cn_flags |= DOWHITEOUT; 1437 error = VOP_RMDIR(dvp, vp, ap->a_cnp); 1438 if (!error) 1439 union_removed_upper(un); 1440 vrele(ap->a_vp); 1441 } else { 1442 error = union_mkwhiteout( 1443 MOUNTTOUNIONMOUNT(UNIONTOV(dun)->v_mount), 1444 dun->un_uppervp, ap->a_cnp, un); 1445 vput(ap->a_vp); 1446 } 1447 1448 return (error); 1449} 1450 1451int 1452union_symlink(void *v) 1453{ 1454 struct vop_symlink_v3_args /* { 1455 struct vnode *a_dvp; 1456 struct vnode **a_vpp; 1457 struct componentname *a_cnp; 1458 struct vattr *a_vap; 1459 char *a_target; 1460 } */ *ap = v; 1461 struct union_node *un = VTOUNION(ap->a_dvp); 1462 struct vnode *dvp = un->un_uppervp; 1463 struct componentname *cnp = ap->a_cnp; 1464 1465 if (dvp != NULLVP) { 1466 int error; 1467 1468 error = VOP_SYMLINK(dvp, ap->a_vpp, cnp, ap->a_vap, 1469 ap->a_target); 1470 return (error); 1471 } 1472 1473 return (EROFS); 1474} 1475 1476/* 1477 * union_readdir works in concert with getdirentries and 1478 * readdir(3) to provide a list of entries in the unioned 1479 * directories. getdirentries is responsible for walking 1480 * down the union stack. readdir(3) is responsible for 1481 * eliminating duplicate names from the returned data stream. 1482 */ 1483int 1484union_readdir(void *v) 1485{ 1486 struct vop_readdir_args /* { 1487 struct vnodeop_desc *a_desc; 1488 struct vnode *a_vp; 1489 struct uio *a_uio; 1490 kauth_cred_t a_cred; 1491 int *a_eofflag; 1492 u_long *a_cookies; 1493 int a_ncookies; 1494 } */ *ap = v; 1495 struct union_node *un = VTOUNION(ap->a_vp); 1496 struct vnode *uvp = un->un_uppervp; 1497 1498 if (uvp == NULLVP) 1499 return (0); 1500 1501 ap->a_vp = uvp; 1502 return (VCALL(uvp, VOFFSET(vop_readdir), ap)); 1503} 1504 1505int 1506union_readlink(void *v) 1507{ 1508 struct vop_readlink_args /* { 1509 struct vnode *a_vp; 1510 struct uio *a_uio; 1511 kauth_cred_t a_cred; 1512 } */ *ap = v; 1513 int error; 1514 struct vnode *vp = OTHERVP(ap->a_vp); 1515 int dolock = (vp == LOWERVP(ap->a_vp)); 1516 1517 if (dolock) 1518 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1519 ap->a_vp = vp; 1520 error = VCALL(vp, VOFFSET(vop_readlink), ap); 1521 if (dolock) 1522 VOP_UNLOCK(vp); 1523 1524 return (error); 1525} 1526 1527int 1528union_abortop(void *v) 1529{ 1530 struct vop_abortop_args /* { 1531 struct vnode *a_dvp; 1532 struct componentname *a_cnp; 1533 } */ *ap = v; 1534 1535 KASSERT(UPPERVP(ap->a_dvp) != NULL); 1536 1537 ap->a_dvp = UPPERVP(ap->a_dvp); 1538 return VCALL(ap->a_dvp, VOFFSET(vop_abortop), ap); 1539} 1540 1541int 1542union_inactive(void *v) 1543{ 1544 struct vop_inactive_v2_args /* { 1545 const struct vnodeop_desc *a_desc; 1546 struct vnode *a_vp; 1547 bool *a_recycle; 1548 } */ *ap = v; 1549 struct vnode *vp = ap->a_vp; 1550 struct union_node *un = VTOUNION(vp); 1551 struct vnode **vpp; 1552 1553 /* 1554 * Do nothing (and _don't_ bypass). 1555 * Wait to vrele lowervp until reclaim, 1556 * so that until then our union_node is in the 1557 * cache and reusable. 1558 * 1559 * NEEDSWORK: Someday, consider inactive'ing 1560 * the lowervp and then trying to reactivate it 1561 * with capabilities (v_id) 1562 * like they do in the name lookup cache code. 1563 * That's too much work for now. 1564 */ 1565 1566 if (un->un_dircache != 0) { 1567 for (vpp = un->un_dircache; *vpp != NULLVP; vpp++) 1568 vrele(*vpp); 1569 free(un->un_dircache, M_TEMP); 1570 un->un_dircache = 0; 1571 } 1572 1573 *ap->a_recycle = ((un->un_cflags & UN_CACHED) == 0); 1574 1575 return (0); 1576} 1577 1578int 1579union_reclaim(void *v) 1580{ 1581 struct vop_reclaim_v2_args /* { 1582 struct vnode *a_vp; 1583 } */ *ap = v; 1584 struct vnode *vp = ap->a_vp; 1585 struct vnode *uvp = UPPERVP(vp); 1586 1587 VOP_UNLOCK(vp); 1588 1589 if (uvp != NULL) { 1590 mutex_enter(uvp->v_interlock); 1591 KASSERT(vp->v_interlock == uvp->v_interlock); 1592 uvp->v_writecount -= vp->v_writecount; 1593 mutex_exit(uvp->v_interlock); 1594 } 1595 1596 union_freevp(vp); 1597 1598 return (0); 1599} 1600 1601static int 1602union_lock1(struct vnode *vp, struct vnode *lockvp, int flags) 1603{ 1604 struct vop_lock_args ap; 1605 1606 ap.a_desc = VDESC(vop_lock); 1607 ap.a_vp = lockvp; 1608 ap.a_flags = flags; 1609 1610 if (lockvp == vp) 1611 return genfs_lock(&ap); 1612 else 1613 return VCALL(ap.a_vp, VOFFSET(vop_lock), &ap); 1614} 1615 1616static int 1617union_unlock1(struct vnode *vp, struct vnode *lockvp) 1618{ 1619 struct vop_unlock_args ap; 1620 1621 ap.a_desc = VDESC(vop_unlock); 1622 ap.a_vp = lockvp; 1623 1624 if (lockvp == vp) 1625 return genfs_unlock(&ap); 1626 else 1627 return VCALL(ap.a_vp, VOFFSET(vop_unlock), &ap); 1628} 1629 1630int 1631union_lock(void *v) 1632{ 1633 struct vop_lock_args /* { 1634 struct vnode *a_vp; 1635 int a_flags; 1636 } */ *ap = v; 1637 struct vnode *vp = ap->a_vp, *lockvp; 1638 struct union_node *un = VTOUNION(vp); 1639 int flags = ap->a_flags; 1640 int error; 1641 1642 if ((flags & LK_NOWAIT) != 0) { 1643 if (!mutex_tryenter(&un->un_lock)) 1644 return EBUSY; 1645 lockvp = LOCKVP(vp); 1646 error = union_lock1(vp, lockvp, flags); 1647 mutex_exit(&un->un_lock); 1648 if (error) 1649 return error; 1650 if (mutex_tryenter(vp->v_interlock)) { 1651 error = vdead_check(vp, VDEAD_NOWAIT); 1652 mutex_exit(vp->v_interlock); 1653 } else 1654 error = EBUSY; 1655 if (error) 1656 union_unlock1(vp, lockvp); 1657 return error; 1658 } 1659 1660 mutex_enter(&un->un_lock); 1661 for (;;) { 1662 lockvp = LOCKVP(vp); 1663 mutex_exit(&un->un_lock); 1664 error = union_lock1(vp, lockvp, flags); 1665 if (error != 0) 1666 return error; 1667 mutex_enter(&un->un_lock); 1668 if (lockvp == LOCKVP(vp)) 1669 break; 1670 union_unlock1(vp, lockvp); 1671 } 1672 mutex_exit(&un->un_lock); 1673 1674 mutex_enter(vp->v_interlock); 1675 error = vdead_check(vp, VDEAD_NOWAIT); 1676 if (error) { 1677 union_unlock1(vp, lockvp); 1678 error = vdead_check(vp, 0); 1679 KASSERT(error == ENOENT); 1680 } 1681 mutex_exit(vp->v_interlock); 1682 return error; 1683} 1684 1685int 1686union_unlock(void *v) 1687{ 1688 struct vop_unlock_args /* { 1689 struct vnode *a_vp; 1690 int a_flags; 1691 } */ *ap = v; 1692 struct vnode *vp = ap->a_vp, *lockvp; 1693 1694 lockvp = LOCKVP(vp); 1695 union_unlock1(vp, lockvp); 1696 1697 return 0; 1698} 1699 1700int 1701union_bmap(void *v) 1702{ 1703 struct vop_bmap_args /* { 1704 struct vnode *a_vp; 1705 daddr_t a_bn; 1706 struct vnode **a_vpp; 1707 daddr_t *a_bnp; 1708 int *a_runp; 1709 } */ *ap = v; 1710 int error; 1711 struct vnode *vp = OTHERVP(ap->a_vp); 1712 int dolock = (vp == LOWERVP(ap->a_vp)); 1713 1714 if (dolock) 1715 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1716 ap->a_vp = vp; 1717 error = VCALL(vp, VOFFSET(vop_bmap), ap); 1718 if (dolock) 1719 VOP_UNLOCK(vp); 1720 1721 return (error); 1722} 1723 1724int 1725union_print(void *v) 1726{ 1727 struct vop_print_args /* { 1728 struct vnode *a_vp; 1729 } */ *ap = v; 1730 struct vnode *vp = ap->a_vp; 1731 1732 printf("\ttag VT_UNION, vp=%p, uppervp=%p, lowervp=%p\n", 1733 vp, UPPERVP(vp), LOWERVP(vp)); 1734 if (UPPERVP(vp) != NULLVP) 1735 vprint("union: upper", UPPERVP(vp)); 1736 if (LOWERVP(vp) != NULLVP) 1737 vprint("union: lower", LOWERVP(vp)); 1738 if (VTOUNION(vp)->un_dircache) { 1739 struct vnode **vpp; 1740 for (vpp = VTOUNION(vp)->un_dircache; *vpp != NULLVP; vpp++) 1741 vprint("dircache:", *vpp); 1742 } 1743 1744 return (0); 1745} 1746 1747int 1748union_islocked(void *v) 1749{ 1750 struct vop_islocked_args /* { 1751 struct vnode *a_vp; 1752 } */ *ap = v; 1753 struct vnode *vp; 1754 struct union_node *un; 1755 1756 un = VTOUNION(ap->a_vp); 1757 mutex_enter(&un->un_lock); 1758 vp = LOCKVP(ap->a_vp); 1759 mutex_exit(&un->un_lock); 1760 1761 if (vp == ap->a_vp) 1762 return genfs_islocked(ap); 1763 else 1764 return VOP_ISLOCKED(vp); 1765} 1766 1767int 1768union_pathconf(void *v) 1769{ 1770 struct vop_pathconf_args /* { 1771 struct vnode *a_vp; 1772 int a_name; 1773 int *a_retval; 1774 } */ *ap = v; 1775 int error; 1776 struct vnode *vp = OTHERVP(ap->a_vp); 1777 int dolock = (vp == LOWERVP(ap->a_vp)); 1778 1779 if (dolock) 1780 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1781 ap->a_vp = vp; 1782 error = VCALL(vp, VOFFSET(vop_pathconf), ap); 1783 if (dolock) 1784 VOP_UNLOCK(vp); 1785 1786 return (error); 1787} 1788 1789int 1790union_advlock(void *v) 1791{ 1792 struct vop_advlock_args /* { 1793 struct vnode *a_vp; 1794 void *a_id; 1795 int a_op; 1796 struct flock *a_fl; 1797 int a_flags; 1798 } */ *ap = v; 1799 struct vnode *ovp = OTHERVP(ap->a_vp); 1800 1801 ap->a_vp = ovp; 1802 return (VCALL(ovp, VOFFSET(vop_advlock), ap)); 1803} 1804 1805int 1806union_strategy(void *v) 1807{ 1808 struct vop_strategy_args /* { 1809 struct vnode *a_vp; 1810 struct buf *a_bp; 1811 } */ *ap = v; 1812 struct vnode *ovp = OTHERVP(ap->a_vp); 1813 struct buf *bp = ap->a_bp; 1814 1815 KASSERT(ovp != NULLVP); 1816 if (!NODE_IS_SPECIAL(ovp)) 1817 KASSERT((bp->b_flags & B_READ) || ovp != LOWERVP(bp->b_vp)); 1818 1819 return (VOP_STRATEGY(ovp, bp)); 1820} 1821 1822int 1823union_bwrite(void *v) 1824{ 1825 struct vop_bwrite_args /* { 1826 struct vnode *a_vp; 1827 struct buf *a_bp; 1828 } */ *ap = v; 1829 struct vnode *ovp = OTHERVP(ap->a_vp); 1830 struct buf *bp = ap->a_bp; 1831 1832 KASSERT(ovp != NULLVP); 1833 if (!NODE_IS_SPECIAL(ovp)) 1834 KASSERT((bp->b_flags & B_READ) || ovp != LOWERVP(bp->b_vp)); 1835 1836 return (VOP_BWRITE(ovp, bp)); 1837} 1838 1839int 1840union_getpages(void *v) 1841{ 1842 struct vop_getpages_args /* { 1843 struct vnode *a_vp; 1844 voff_t a_offset; 1845 struct vm_page **a_m; 1846 int *a_count; 1847 int a_centeridx; 1848 vm_prot_t a_access_type; 1849 int a_advice; 1850 int a_flags; 1851 } */ *ap = v; 1852 struct vnode *vp = ap->a_vp; 1853 1854 KASSERT(rw_lock_held(vp->v_uobj.vmobjlock)); 1855 1856 if (ap->a_flags & PGO_LOCKED) { 1857 return EBUSY; 1858 } 1859 ap->a_vp = OTHERVP(vp); 1860 KASSERT(vp->v_uobj.vmobjlock == ap->a_vp->v_uobj.vmobjlock); 1861 1862 /* Just pass the request on to the underlying layer. */ 1863 return VCALL(ap->a_vp, VOFFSET(vop_getpages), ap); 1864} 1865 1866int 1867union_putpages(void *v) 1868{ 1869 struct vop_putpages_args /* { 1870 struct vnode *a_vp; 1871 voff_t a_offlo; 1872 voff_t a_offhi; 1873 int a_flags; 1874 } */ *ap = v; 1875 struct vnode *vp = ap->a_vp; 1876 1877 KASSERT(rw_lock_held(vp->v_uobj.vmobjlock)); 1878 1879 ap->a_vp = OTHERVP(vp); 1880 KASSERT(vp->v_uobj.vmobjlock == ap->a_vp->v_uobj.vmobjlock); 1881 1882 if (ap->a_flags & PGO_RECLAIM) { 1883 rw_exit(vp->v_uobj.vmobjlock); 1884 return 0; 1885 } 1886 1887 /* Just pass the request on to the underlying layer. */ 1888 return VCALL(ap->a_vp, VOFFSET(vop_putpages), ap); 1889} 1890 1891int 1892union_kqfilter(void *v) 1893{ 1894 struct vop_kqfilter_args /* { 1895 struct vnode *a_vp; 1896 struct knote *a_kn; 1897 } */ *ap = v; 1898 int error; 1899 1900 /* 1901 * We watch either the upper layer file (if it already exists), 1902 * or the lower layer one. If there is lower layer file only 1903 * at this moment, we will keep watching that lower layer file 1904 * even if upper layer file would be created later on. 1905 */ 1906 if (UPPERVP(ap->a_vp)) 1907 error = VOP_KQFILTER(UPPERVP(ap->a_vp), ap->a_kn); 1908 else if (LOWERVP(ap->a_vp)) 1909 error = VOP_KQFILTER(LOWERVP(ap->a_vp), ap->a_kn); 1910 else { 1911 /* panic? */ 1912 error = EOPNOTSUPP; 1913 } 1914 1915 return (error); 1916} 1917