1/*- 2 * Copyright (c) 1994 Jan-Simon Pendry 3 * Copyright (c) 1994 4 * The Regents of the University of California. All rights reserved. 5 * Copyright (c) 2005, 2006, 2012 Masanori Ozawa <ozawa@ongs.co.jp>, ONGS Inc. 6 * Copyright (c) 2006, 2012 Daichi Goto <daichi@freebsd.org> 7 * 8 * This code is derived from software contributed to Berkeley by 9 * Jan-Simon Pendry. 10 * 11 * Redistribution and use in source and binary forms, with or without 12 * modification, are permitted provided that the following conditions 13 * are met: 14 * 1. Redistributions of source code must retain the above copyright 15 * notice, this list of conditions and the following disclaimer. 16 * 2. Redistributions in binary form must reproduce the above copyright 17 * notice, this list of conditions and the following disclaimer in the 18 * documentation and/or other materials provided with the distribution. 19 * 4. Neither the name of the University nor the names of its contributors 20 * may be used to endorse or promote products derived from this software 21 * without specific prior written permission. 22 * 23 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 24 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 25 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 26 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 27 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 28 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 29 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 30 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 31 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 32 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 33 * SUCH DAMAGE. 34 * 35 * @(#)union_subr.c 8.20 (Berkeley) 5/20/95 36 * $FreeBSD$ 37 */ 38 39#include <sys/param.h> 40#include <sys/systm.h> 41#include <sys/kernel.h> 42#include <sys/lock.h> 43#include <sys/mutex.h> 44#include <sys/malloc.h> 45#include <sys/mount.h> 46#include <sys/namei.h> 47#include <sys/proc.h> 48#include <sys/vnode.h> 49#include <sys/dirent.h> 50#include <sys/fcntl.h> 51#include <sys/filedesc.h> 52#include <sys/stat.h> 53#include <sys/resourcevar.h> 54 55#include <security/mac/mac_framework.h> 56 57#include <vm/uma.h> 58 59#include <fs/unionfs/union.h> 60 61#define NUNIONFSNODECACHE 16 62 63static MALLOC_DEFINE(M_UNIONFSHASH, "UNIONFS hash", "UNIONFS hash table"); 64MALLOC_DEFINE(M_UNIONFSNODE, "UNIONFS node", "UNIONFS vnode private part"); 65MALLOC_DEFINE(M_UNIONFSPATH, "UNIONFS path", "UNIONFS path private part"); 66 67/* 68 * Initialize 69 */ 70int 71unionfs_init(struct vfsconf *vfsp) 72{ 73 UNIONFSDEBUG("unionfs_init\n"); /* printed during system boot */ 74 return (0); 75} 76 77/* 78 * Uninitialize 79 */ 80int 81unionfs_uninit(struct vfsconf *vfsp) 82{ 83 return (0); 84} 85 86static struct unionfs_node_hashhead * 87unionfs_get_hashhead(struct vnode *dvp, char *path) 88{ 89 int count; 90 char hash; 91 struct unionfs_node *unp; 92 93 hash = 0; 94 unp = VTOUNIONFS(dvp); 95 if (path != NULL) { 96 for (count = 0; path[count]; count++) 97 hash += path[count]; 98 } 99 100 return (&(unp->un_hashtbl[hash & (unp->un_hashmask)])); 101} 102 103/* 104 * Get the cached vnode. 105 */ 106static struct vnode * 107unionfs_get_cached_vnode(struct vnode *uvp, struct vnode *lvp, 108 struct vnode *dvp, char *path) 109{ 110 struct unionfs_node_hashhead *hd; 111 struct unionfs_node *unp; 112 struct vnode *vp; 113 114 KASSERT((uvp == NULLVP || uvp->v_type == VDIR), 115 ("unionfs_get_cached_vnode: v_type != VDIR")); 116 KASSERT((lvp == NULLVP || lvp->v_type == VDIR), 117 ("unionfs_get_cached_vnode: v_type != VDIR")); 118 119 VI_LOCK(dvp); 120 hd = unionfs_get_hashhead(dvp, path); 121 LIST_FOREACH(unp, hd, un_hash) { 122 if (!strcmp(unp->un_path, path)) { 123 vp = UNIONFSTOV(unp); 124 VI_LOCK_FLAGS(vp, MTX_DUPOK); 125 VI_UNLOCK(dvp); 126 vp->v_iflag &= ~VI_OWEINACT; 127 if ((vp->v_iflag & (VI_DOOMED | VI_DOINGINACT)) != 0) { 128 VI_UNLOCK(vp); 129 vp = NULLVP; 130 } else 131 VI_UNLOCK(vp); 132 return (vp); 133 } 134 } 135 VI_UNLOCK(dvp); 136 137 return (NULLVP); 138} 139 140/* 141 * Add the new vnode into cache. 142 */ 143static struct vnode * 144unionfs_ins_cached_vnode(struct unionfs_node *uncp, 145 struct vnode *dvp, char *path) 146{ 147 struct unionfs_node_hashhead *hd; 148 struct unionfs_node *unp; 149 struct vnode *vp; 150 151 KASSERT((uncp->un_uppervp==NULLVP || uncp->un_uppervp->v_type==VDIR), 152 ("unionfs_ins_cached_vnode: v_type != VDIR")); 153 KASSERT((uncp->un_lowervp==NULLVP || uncp->un_lowervp->v_type==VDIR), 154 ("unionfs_ins_cached_vnode: v_type != VDIR")); 155 156 VI_LOCK(dvp); 157 hd = unionfs_get_hashhead(dvp, path); 158 LIST_FOREACH(unp, hd, un_hash) { 159 if (!strcmp(unp->un_path, path)) { 160 vp = UNIONFSTOV(unp); 161 VI_LOCK_FLAGS(vp, MTX_DUPOK); 162 vp->v_iflag &= ~VI_OWEINACT; 163 if ((vp->v_iflag & (VI_DOOMED | VI_DOINGINACT)) != 0) { 164 LIST_INSERT_HEAD(hd, uncp, un_hash); 165 VI_UNLOCK(vp); 166 vp = NULLVP; 167 } else 168 VI_UNLOCK(vp); 169 VI_UNLOCK(dvp); 170 return (vp); 171 } 172 } 173 174 LIST_INSERT_HEAD(hd, uncp, un_hash); 175 VI_UNLOCK(dvp); 176 177 return (NULLVP); 178} 179 180/* 181 * Remove the vnode. 182 */ 183static void 184unionfs_rem_cached_vnode(struct unionfs_node *unp, struct vnode *dvp) 185{ 186 KASSERT((unp != NULL), ("unionfs_rem_cached_vnode: null node")); 187 KASSERT((dvp != NULLVP), 188 ("unionfs_rem_cached_vnode: null parent vnode")); 189 KASSERT((unp->un_hash.le_prev != NULL), 190 ("unionfs_rem_cached_vnode: null hash")); 191 192 VI_LOCK(dvp); 193 LIST_REMOVE(unp, un_hash); 194 unp->un_hash.le_next = NULL; 195 unp->un_hash.le_prev = NULL; 196 VI_UNLOCK(dvp); 197} 198 199/* 200 * Make a new or get existing unionfs node. 201 * 202 * uppervp and lowervp should be unlocked. Because if new unionfs vnode is 203 * locked, uppervp or lowervp is locked too. In order to prevent dead lock, 204 * you should not lock plurality simultaneously. 205 */ 206int 207unionfs_nodeget(struct mount *mp, struct vnode *uppervp, 208 struct vnode *lowervp, struct vnode *dvp, 209 struct vnode **vpp, struct componentname *cnp, 210 struct thread *td) 211{ 212 struct unionfs_mount *ump; 213 struct unionfs_node *unp; 214 struct vnode *vp; 215 int error; 216 int lkflags; 217 enum vtype vt; 218 char *path; 219 220 ump = MOUNTTOUNIONFSMOUNT(mp); 221 lkflags = (cnp ? cnp->cn_lkflags : 0); 222 path = (cnp ? cnp->cn_nameptr : NULL); 223 *vpp = NULLVP; 224 225 if (uppervp == NULLVP && lowervp == NULLVP) 226 panic("unionfs_nodeget: upper and lower is null"); 227 228 vt = (uppervp != NULLVP ? uppervp->v_type : lowervp->v_type); 229 230 /* If it has no ISLASTCN flag, path check is skipped. */ 231 if (cnp && !(cnp->cn_flags & ISLASTCN)) 232 path = NULL; 233 234 /* check the cache */ 235 if (path != NULL && dvp != NULLVP && vt == VDIR) { 236 vp = unionfs_get_cached_vnode(uppervp, lowervp, dvp, path); 237 if (vp != NULLVP) { 238 vref(vp); 239 *vpp = vp; 240 goto unionfs_nodeget_out; 241 } 242 } 243 244 if ((uppervp == NULLVP || ump->um_uppervp != uppervp) || 245 (lowervp == NULLVP || ump->um_lowervp != lowervp)) { 246 /* dvp will be NULLVP only in case of root vnode. */ 247 if (dvp == NULLVP) 248 return (EINVAL); 249 } 250 251 /* 252 * Do the MALLOC before the getnewvnode since doing so afterward 253 * might cause a bogus v_data pointer to get dereferenced elsewhere 254 * if MALLOC should block. 255 */ 256 unp = malloc(sizeof(struct unionfs_node), 257 M_UNIONFSNODE, M_WAITOK | M_ZERO); 258 259 error = getnewvnode("unionfs", mp, &unionfs_vnodeops, &vp); 260 if (error != 0) { 261 free(unp, M_UNIONFSNODE); 262 return (error); 263 } 264 error = insmntque(vp, mp); /* XXX: Too early for mpsafe fs */ 265 if (error != 0) { 266 free(unp, M_UNIONFSNODE); 267 return (error); 268 } 269 if (dvp != NULLVP) 270 vref(dvp); 271 if (uppervp != NULLVP) 272 vref(uppervp); 273 if (lowervp != NULLVP) 274 vref(lowervp); 275 276 if (vt == VDIR) 277 unp->un_hashtbl = hashinit(NUNIONFSNODECACHE, M_UNIONFSHASH, 278 &(unp->un_hashmask)); 279 280 unp->un_vnode = vp; 281 unp->un_uppervp = uppervp; 282 unp->un_lowervp = lowervp; 283 unp->un_dvp = dvp; 284 if (uppervp != NULLVP) 285 vp->v_vnlock = uppervp->v_vnlock; 286 else 287 vp->v_vnlock = lowervp->v_vnlock; 288 289 if (path != NULL) { 290 unp->un_path = (char *) 291 malloc(cnp->cn_namelen +1, M_UNIONFSPATH, M_WAITOK|M_ZERO); 292 bcopy(cnp->cn_nameptr, unp->un_path, cnp->cn_namelen); 293 unp->un_path[cnp->cn_namelen] = '\0'; 294 } 295 vp->v_type = vt; 296 vp->v_data = unp; 297 298 if ((uppervp != NULLVP && ump->um_uppervp == uppervp) && 299 (lowervp != NULLVP && ump->um_lowervp == lowervp)) 300 vp->v_vflag |= VV_ROOT; 301 302 if (path != NULL && dvp != NULLVP && vt == VDIR) 303 *vpp = unionfs_ins_cached_vnode(unp, dvp, path); 304 if ((*vpp) != NULLVP) { 305 if (dvp != NULLVP) 306 vrele(dvp); 307 if (uppervp != NULLVP) 308 vrele(uppervp); 309 if (lowervp != NULLVP) 310 vrele(lowervp); 311 312 unp->un_uppervp = NULLVP; 313 unp->un_lowervp = NULLVP; 314 unp->un_dvp = NULLVP; 315 vrele(vp); 316 vp = *vpp; 317 vref(vp); 318 } else 319 *vpp = vp; 320 321unionfs_nodeget_out: 322 if (lkflags & LK_TYPE_MASK) 323 vn_lock(vp, lkflags | LK_RETRY); 324 325 return (0); 326} 327 328/* 329 * Clean up the unionfs node. 330 */ 331void 332unionfs_noderem(struct vnode *vp, struct thread *td) 333{ 334 int vfslocked; 335 int count; 336 struct unionfs_node *unp, *unp_t1, *unp_t2; 337 struct unionfs_node_hashhead *hd; 338 struct unionfs_node_status *unsp, *unsp_tmp; 339 struct vnode *lvp; 340 struct vnode *uvp; 341 struct vnode *dvp; 342 343 /* 344 * Use the interlock to protect the clearing of v_data to 345 * prevent faults in unionfs_lock(). 346 */ 347 VI_LOCK(vp); 348 unp = VTOUNIONFS(vp); 349 lvp = unp->un_lowervp; 350 uvp = unp->un_uppervp; 351 dvp = unp->un_dvp; 352 unp->un_lowervp = unp->un_uppervp = NULLVP; 353 vp->v_vnlock = &(vp->v_lock); 354 vp->v_data = NULL; 355 vp->v_object = NULL; 356 VI_UNLOCK(vp); 357 358 if (lvp != NULLVP) 359 VOP_UNLOCK(lvp, LK_RELEASE); 360 if (uvp != NULLVP) 361 VOP_UNLOCK(uvp, LK_RELEASE); 362 363 if (dvp != NULLVP && unp->un_hash.le_prev != NULL) 364 unionfs_rem_cached_vnode(unp, dvp); 365 366 if (lockmgr(vp->v_vnlock, LK_EXCLUSIVE, VI_MTX(vp)) != 0) 367 panic("the lock for deletion is unacquirable."); 368 369 if (lvp != NULLVP) { 370 vfslocked = VFS_LOCK_GIANT(lvp->v_mount); 371 vrele(lvp); 372 VFS_UNLOCK_GIANT(vfslocked); 373 } 374 if (uvp != NULLVP) { 375 vfslocked = VFS_LOCK_GIANT(uvp->v_mount); 376 vrele(uvp); 377 VFS_UNLOCK_GIANT(vfslocked); 378 } 379 if (dvp != NULLVP) { 380 vfslocked = VFS_LOCK_GIANT(dvp->v_mount); 381 vrele(dvp); 382 VFS_UNLOCK_GIANT(vfslocked); 383 unp->un_dvp = NULLVP; 384 } 385 if (unp->un_path != NULL) { 386 free(unp->un_path, M_UNIONFSPATH); 387 unp->un_path = NULL; 388 } 389 390 if (unp->un_hashtbl != NULL) { 391 for (count = 0; count <= unp->un_hashmask; count++) { 392 hd = unp->un_hashtbl + count; 393 LIST_FOREACH_SAFE(unp_t1, hd, un_hash, unp_t2) { 394 LIST_REMOVE(unp_t1, un_hash); 395 unp_t1->un_hash.le_next = NULL; 396 unp_t1->un_hash.le_prev = NULL; 397 } 398 } 399 hashdestroy(unp->un_hashtbl, M_UNIONFSHASH, unp->un_hashmask); 400 } 401 402 LIST_FOREACH_SAFE(unsp, &(unp->un_unshead), uns_list, unsp_tmp) { 403 LIST_REMOVE(unsp, uns_list); 404 free(unsp, M_TEMP); 405 } 406 free(unp, M_UNIONFSNODE); 407} 408 409/* 410 * Get the unionfs node status. 411 * You need exclusive lock this vnode. 412 */ 413void 414unionfs_get_node_status(struct unionfs_node *unp, struct thread *td, 415 struct unionfs_node_status **unspp) 416{ 417 struct unionfs_node_status *unsp; 418 pid_t pid = td->td_proc->p_pid; 419 420 KASSERT(NULL != unspp, ("null pointer")); 421 ASSERT_VOP_ELOCKED(UNIONFSTOV(unp), "unionfs_get_node_status"); 422 423 LIST_FOREACH(unsp, &(unp->un_unshead), uns_list) { 424 if (unsp->uns_pid == pid) { 425 *unspp = unsp; 426 return; 427 } 428 } 429 430 /* create a new unionfs node status */ 431 unsp = malloc(sizeof(struct unionfs_node_status), 432 M_TEMP, M_WAITOK | M_ZERO); 433 434 unsp->uns_pid = pid; 435 LIST_INSERT_HEAD(&(unp->un_unshead), unsp, uns_list); 436 437 *unspp = unsp; 438} 439 440/* 441 * Remove the unionfs node status, if you can. 442 * You need exclusive lock this vnode. 443 */ 444void 445unionfs_tryrem_node_status(struct unionfs_node *unp, 446 struct unionfs_node_status *unsp) 447{ 448 KASSERT(NULL != unsp, ("null pointer")); 449 ASSERT_VOP_ELOCKED(UNIONFSTOV(unp), "unionfs_get_node_status"); 450 451 if (0 < unsp->uns_lower_opencnt || 0 < unsp->uns_upper_opencnt) 452 return; 453 454 LIST_REMOVE(unsp, uns_list); 455 free(unsp, M_TEMP); 456} 457 458/* 459 * Create upper node attr. 460 */ 461void 462unionfs_create_uppervattr_core(struct unionfs_mount *ump, 463 struct vattr *lva, 464 struct vattr *uva, 465 struct thread *td) 466{ 467 VATTR_NULL(uva); 468 uva->va_type = lva->va_type; 469 uva->va_atime = lva->va_atime; 470 uva->va_mtime = lva->va_mtime; 471 uva->va_ctime = lva->va_ctime; 472 473 switch (ump->um_copymode) { 474 case UNIONFS_TRANSPARENT: 475 uva->va_mode = lva->va_mode; 476 uva->va_uid = lva->va_uid; 477 uva->va_gid = lva->va_gid; 478 break; 479 case UNIONFS_MASQUERADE: 480 if (ump->um_uid == lva->va_uid) { 481 uva->va_mode = lva->va_mode & 077077; 482 uva->va_mode |= (lva->va_type == VDIR ? ump->um_udir : ump->um_ufile) & 0700; 483 uva->va_uid = lva->va_uid; 484 uva->va_gid = lva->va_gid; 485 } else { 486 uva->va_mode = (lva->va_type == VDIR ? ump->um_udir : ump->um_ufile); 487 uva->va_uid = ump->um_uid; 488 uva->va_gid = ump->um_gid; 489 } 490 break; 491 default: /* UNIONFS_TRADITIONAL */ 492 uva->va_mode = 0777 & ~td->td_proc->p_fd->fd_cmask; 493 uva->va_uid = ump->um_uid; 494 uva->va_gid = ump->um_gid; 495 break; 496 } 497} 498 499/* 500 * Create upper node attr. 501 */ 502int 503unionfs_create_uppervattr(struct unionfs_mount *ump, 504 struct vnode *lvp, 505 struct vattr *uva, 506 struct ucred *cred, 507 struct thread *td) 508{ 509 int error; 510 struct vattr lva; 511 512 if ((error = VOP_GETATTR(lvp, &lva, cred))) 513 return (error); 514 515 unionfs_create_uppervattr_core(ump, &lva, uva, td); 516 517 return (error); 518} 519 520/* 521 * relookup 522 * 523 * dvp should be locked on entry and will be locked on return. 524 * 525 * If an error is returned, *vpp will be invalid, otherwise it will hold a 526 * locked, referenced vnode. If *vpp == dvp then remember that only one 527 * LK_EXCLUSIVE lock is held. 528 */ 529int 530unionfs_relookup(struct vnode *dvp, struct vnode **vpp, 531 struct componentname *cnp, struct componentname *cn, 532 struct thread *td, char *path, int pathlen, u_long nameiop) 533{ 534 int error; 535 536 cn->cn_namelen = pathlen; 537 cn->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK); 538 bcopy(path, cn->cn_pnbuf, pathlen); 539 cn->cn_pnbuf[pathlen] = '\0'; 540 541 cn->cn_nameiop = nameiop; 542 cn->cn_flags = (LOCKPARENT | LOCKLEAF | HASBUF | SAVENAME | ISLASTCN); 543 cn->cn_lkflags = LK_EXCLUSIVE; 544 cn->cn_thread = td; 545 cn->cn_cred = cnp->cn_cred; 546 547 cn->cn_nameptr = cn->cn_pnbuf; 548 cn->cn_consume = cnp->cn_consume; 549 550 if (nameiop == DELETE) 551 cn->cn_flags |= (cnp->cn_flags & (DOWHITEOUT | SAVESTART)); 552 else if (RENAME == nameiop) 553 cn->cn_flags |= (cnp->cn_flags & SAVESTART); 554 555 vref(dvp); 556 VOP_UNLOCK(dvp, LK_RELEASE); 557 558 if ((error = relookup(dvp, vpp, cn))) { 559 uma_zfree(namei_zone, cn->cn_pnbuf); 560 cn->cn_flags &= ~HASBUF; 561 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 562 } else 563 vrele(dvp); 564 565 return (error); 566} 567 568/* 569 * relookup for CREATE namei operation. 570 * 571 * dvp is unionfs vnode. dvp should be locked. 572 * 573 * If it called 'unionfs_copyfile' function by unionfs_link etc, 574 * VOP_LOOKUP information is broken. 575 * So it need relookup in order to create link etc. 576 */ 577int 578unionfs_relookup_for_create(struct vnode *dvp, struct componentname *cnp, 579 struct thread *td) 580{ 581 int error; 582 struct vnode *udvp; 583 struct vnode *vp; 584 struct componentname cn; 585 586 udvp = UNIONFSVPTOUPPERVP(dvp); 587 vp = NULLVP; 588 589 error = unionfs_relookup(udvp, &vp, cnp, &cn, td, cnp->cn_nameptr, 590 strlen(cnp->cn_nameptr), CREATE); 591 if (error) 592 return (error); 593 594 if (vp != NULLVP) { 595 if (udvp == vp) 596 vrele(vp); 597 else 598 vput(vp); 599 600 error = EEXIST; 601 } 602 603 if (cn.cn_flags & HASBUF) { 604 uma_zfree(namei_zone, cn.cn_pnbuf); 605 cn.cn_flags &= ~HASBUF; 606 } 607 608 if (!error) { 609 cn.cn_flags |= (cnp->cn_flags & HASBUF); 610 cnp->cn_flags = cn.cn_flags; 611 } 612 613 return (error); 614} 615 616/* 617 * relookup for DELETE namei operation. 618 * 619 * dvp is unionfs vnode. dvp should be locked. 620 */ 621int 622unionfs_relookup_for_delete(struct vnode *dvp, struct componentname *cnp, 623 struct thread *td) 624{ 625 int error; 626 struct vnode *udvp; 627 struct vnode *vp; 628 struct componentname cn; 629 630 udvp = UNIONFSVPTOUPPERVP(dvp); 631 vp = NULLVP; 632 633 error = unionfs_relookup(udvp, &vp, cnp, &cn, td, cnp->cn_nameptr, 634 strlen(cnp->cn_nameptr), DELETE); 635 if (error) 636 return (error); 637 638 if (vp == NULLVP) 639 error = ENOENT; 640 else { 641 if (udvp == vp) 642 vrele(vp); 643 else 644 vput(vp); 645 } 646 647 if (cn.cn_flags & HASBUF) { 648 uma_zfree(namei_zone, cn.cn_pnbuf); 649 cn.cn_flags &= ~HASBUF; 650 } 651 652 if (!error) { 653 cn.cn_flags |= (cnp->cn_flags & HASBUF); 654 cnp->cn_flags = cn.cn_flags; 655 } 656 657 return (error); 658} 659 660/* 661 * relookup for RENAME namei operation. 662 * 663 * dvp is unionfs vnode. dvp should be locked. 664 */ 665int 666unionfs_relookup_for_rename(struct vnode *dvp, struct componentname *cnp, 667 struct thread *td) 668{ 669 int error; 670 struct vnode *udvp; 671 struct vnode *vp; 672 struct componentname cn; 673 674 udvp = UNIONFSVPTOUPPERVP(dvp); 675 vp = NULLVP; 676 677 error = unionfs_relookup(udvp, &vp, cnp, &cn, td, cnp->cn_nameptr, 678 strlen(cnp->cn_nameptr), RENAME); 679 if (error) 680 return (error); 681 682 if (vp != NULLVP) { 683 if (udvp == vp) 684 vrele(vp); 685 else 686 vput(vp); 687 } 688 689 if (cn.cn_flags & HASBUF) { 690 uma_zfree(namei_zone, cn.cn_pnbuf); 691 cn.cn_flags &= ~HASBUF; 692 } 693 694 if (!error) { 695 cn.cn_flags |= (cnp->cn_flags & HASBUF); 696 cnp->cn_flags = cn.cn_flags; 697 } 698 699 return (error); 700 701} 702 703/* 704 * Update the unionfs_node. 705 * 706 * uvp is new locked upper vnode. unionfs vnode's lock will be exchanged to the 707 * uvp's lock and lower's lock will be unlocked. 708 */ 709static void 710unionfs_node_update(struct unionfs_node *unp, struct vnode *uvp, 711 struct thread *td) 712{ 713 unsigned count, lockrec; 714 struct vnode *vp; 715 struct vnode *lvp; 716 struct vnode *dvp; 717 718 vp = UNIONFSTOV(unp); 719 lvp = unp->un_lowervp; 720 ASSERT_VOP_ELOCKED(lvp, "unionfs_node_update"); 721 dvp = unp->un_dvp; 722 723 /* 724 * lock update 725 */ 726 VI_LOCK(vp); 727 unp->un_uppervp = uvp; 728 vp->v_vnlock = uvp->v_vnlock; 729 VI_UNLOCK(vp); 730 lockrec = lvp->v_vnlock->lk_recurse; 731 for (count = 0; count < lockrec; count++) 732 vn_lock(uvp, LK_EXCLUSIVE | LK_CANRECURSE | LK_RETRY); 733 734 /* 735 * cache update 736 */ 737 if (unp->un_path != NULL && dvp != NULLVP && vp->v_type == VDIR) { 738 static struct unionfs_node_hashhead *hd; 739 740 VI_LOCK(dvp); 741 hd = unionfs_get_hashhead(dvp, unp->un_path); 742 LIST_REMOVE(unp, un_hash); 743 LIST_INSERT_HEAD(hd, unp, un_hash); 744 VI_UNLOCK(dvp); 745 } 746} 747 748/* 749 * Create a new shadow dir. 750 * 751 * udvp should be locked on entry and will be locked on return. 752 * 753 * If no error returned, unp will be updated. 754 */ 755int 756unionfs_mkshadowdir(struct unionfs_mount *ump, struct vnode *udvp, 757 struct unionfs_node *unp, struct componentname *cnp, 758 struct thread *td) 759{ 760 int error; 761 struct vnode *lvp; 762 struct vnode *uvp; 763 struct vattr va; 764 struct vattr lva; 765 struct componentname cn; 766 struct mount *mp; 767 struct ucred *cred; 768 struct ucred *credbk; 769 struct uidinfo *rootinfo; 770 771 if (unp->un_uppervp != NULLVP) 772 return (EEXIST); 773 774 lvp = unp->un_lowervp; 775 uvp = NULLVP; 776 credbk = cnp->cn_cred; 777 778 /* Authority change to root */ 779 rootinfo = uifind((uid_t)0); 780 cred = crdup(cnp->cn_cred); 781 /* 782 * The calls to chgproccnt() are needed to compensate for change_ruid() 783 * calling chgproccnt(). 784 */ 785 chgproccnt(cred->cr_ruidinfo, 1, 0); 786 change_euid(cred, rootinfo); 787 change_ruid(cred, rootinfo); 788 change_svuid(cred, (uid_t)0); 789 uifree(rootinfo); 790 cnp->cn_cred = cred; 791 792 memset(&cn, 0, sizeof(cn)); 793 794 if ((error = VOP_GETATTR(lvp, &lva, cnp->cn_cred))) 795 goto unionfs_mkshadowdir_abort; 796 797 if ((error = unionfs_relookup(udvp, &uvp, cnp, &cn, td, cnp->cn_nameptr, cnp->cn_namelen, CREATE))) 798 goto unionfs_mkshadowdir_abort; 799 if (uvp != NULLVP) { 800 if (udvp == uvp) 801 vrele(uvp); 802 else 803 vput(uvp); 804 805 error = EEXIST; 806 goto unionfs_mkshadowdir_free_out; 807 } 808 809 if ((error = vn_start_write(udvp, &mp, V_WAIT | PCATCH))) 810 goto unionfs_mkshadowdir_free_out; 811 unionfs_create_uppervattr_core(ump, &lva, &va, td); 812 813 error = VOP_MKDIR(udvp, &uvp, &cn, &va); 814 815 if (!error) { 816 unionfs_node_update(unp, uvp, td); 817 818 /* 819 * XXX The bug which cannot set uid/gid was corrected. 820 * Ignore errors. 821 */ 822 va.va_type = VNON; 823 VOP_SETATTR(uvp, &va, cn.cn_cred); 824 } 825 vn_finished_write(mp); 826 827unionfs_mkshadowdir_free_out: 828 if (cn.cn_flags & HASBUF) { 829 uma_zfree(namei_zone, cn.cn_pnbuf); 830 cn.cn_flags &= ~HASBUF; 831 } 832 833unionfs_mkshadowdir_abort: 834 cnp->cn_cred = credbk; 835 chgproccnt(cred->cr_ruidinfo, -1, 0); 836 crfree(cred); 837 838 return (error); 839} 840 841/* 842 * Create a new whiteout. 843 * 844 * dvp should be locked on entry and will be locked on return. 845 */ 846int 847unionfs_mkwhiteout(struct vnode *dvp, struct componentname *cnp, 848 struct thread *td, char *path) 849{ 850 int error; 851 struct vnode *wvp; 852 struct componentname cn; 853 struct mount *mp; 854 855 if (path == NULL) 856 path = cnp->cn_nameptr; 857 858 wvp = NULLVP; 859 if ((error = unionfs_relookup(dvp, &wvp, cnp, &cn, td, path, strlen(path), CREATE))) 860 return (error); 861 if (wvp != NULLVP) { 862 if (cn.cn_flags & HASBUF) { 863 uma_zfree(namei_zone, cn.cn_pnbuf); 864 cn.cn_flags &= ~HASBUF; 865 } 866 if (dvp == wvp) 867 vrele(wvp); 868 else 869 vput(wvp); 870 871 return (EEXIST); 872 } 873 874 if ((error = vn_start_write(dvp, &mp, V_WAIT | PCATCH))) 875 goto unionfs_mkwhiteout_free_out; 876 error = VOP_WHITEOUT(dvp, &cn, CREATE); 877 878 vn_finished_write(mp); 879 880unionfs_mkwhiteout_free_out: 881 if (cn.cn_flags & HASBUF) { 882 uma_zfree(namei_zone, cn.cn_pnbuf); 883 cn.cn_flags &= ~HASBUF; 884 } 885 886 return (error); 887} 888 889/* 890 * Create a new vnode for create a new shadow file. 891 * 892 * If an error is returned, *vpp will be invalid, otherwise it will hold a 893 * locked, referenced and opened vnode. 894 * 895 * unp is never updated. 896 */ 897static int 898unionfs_vn_create_on_upper(struct vnode **vpp, struct vnode *udvp, 899 struct unionfs_node *unp, struct vattr *uvap, 900 struct thread *td) 901{ 902 struct unionfs_mount *ump; 903 struct vnode *vp; 904 struct vnode *lvp; 905 struct ucred *cred; 906 struct vattr lva; 907 int fmode; 908 int error; 909 struct componentname cn; 910 911 ump = MOUNTTOUNIONFSMOUNT(UNIONFSTOV(unp)->v_mount); 912 vp = NULLVP; 913 lvp = unp->un_lowervp; 914 cred = td->td_ucred; 915 fmode = FFLAGS(O_WRONLY | O_CREAT | O_TRUNC | O_EXCL); 916 error = 0; 917 918 if ((error = VOP_GETATTR(lvp, &lva, cred)) != 0) 919 return (error); 920 unionfs_create_uppervattr_core(ump, &lva, uvap, td); 921 922 if (unp->un_path == NULL) 923 panic("unionfs: un_path is null"); 924 925 cn.cn_namelen = strlen(unp->un_path); 926 cn.cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK); 927 bcopy(unp->un_path, cn.cn_pnbuf, cn.cn_namelen + 1); 928 cn.cn_nameiop = CREATE; 929 cn.cn_flags = (LOCKPARENT | LOCKLEAF | HASBUF | SAVENAME | ISLASTCN); 930 cn.cn_lkflags = LK_EXCLUSIVE; 931 cn.cn_thread = td; 932 cn.cn_cred = cred; 933 cn.cn_nameptr = cn.cn_pnbuf; 934 cn.cn_consume = 0; 935 936 vref(udvp); 937 if ((error = relookup(udvp, &vp, &cn)) != 0) 938 goto unionfs_vn_create_on_upper_free_out2; 939 vrele(udvp); 940 941 if (vp != NULLVP) { 942 if (vp == udvp) 943 vrele(vp); 944 else 945 vput(vp); 946 error = EEXIST; 947 goto unionfs_vn_create_on_upper_free_out1; 948 } 949 950 if ((error = VOP_CREATE(udvp, &vp, &cn, uvap)) != 0) 951 goto unionfs_vn_create_on_upper_free_out1; 952 953 if ((error = VOP_OPEN(vp, fmode, cred, td, NULL)) != 0) { 954 vput(vp); 955 goto unionfs_vn_create_on_upper_free_out1; 956 } 957 VOP_ADD_WRITECOUNT(vp, 1); 958 *vpp = vp; 959 960unionfs_vn_create_on_upper_free_out1: 961 VOP_UNLOCK(udvp, LK_RELEASE); 962 963unionfs_vn_create_on_upper_free_out2: 964 if (cn.cn_flags & HASBUF) { 965 uma_zfree(namei_zone, cn.cn_pnbuf); 966 cn.cn_flags &= ~HASBUF; 967 } 968 969 return (error); 970} 971 972/* 973 * Copy from lvp to uvp. 974 * 975 * lvp and uvp should be locked and opened on entry and will be locked and 976 * opened on return. 977 */ 978static int 979unionfs_copyfile_core(struct vnode *lvp, struct vnode *uvp, 980 struct ucred *cred, struct thread *td) 981{ 982 int error; 983 off_t offset; 984 int count; 985 int bufoffset; 986 char *buf; 987 struct uio uio; 988 struct iovec iov; 989 990 error = 0; 991 memset(&uio, 0, sizeof(uio)); 992 993 uio.uio_td = td; 994 uio.uio_segflg = UIO_SYSSPACE; 995 uio.uio_offset = 0; 996 997 buf = malloc(MAXBSIZE, M_TEMP, M_WAITOK); 998 999 while (error == 0) { 1000 offset = uio.uio_offset; 1001 1002 uio.uio_iov = &iov; 1003 uio.uio_iovcnt = 1; 1004 iov.iov_base = buf; 1005 iov.iov_len = MAXBSIZE; 1006 uio.uio_resid = iov.iov_len; 1007 uio.uio_rw = UIO_READ; 1008 1009 if ((error = VOP_READ(lvp, &uio, 0, cred)) != 0) 1010 break; 1011 if ((count = MAXBSIZE - uio.uio_resid) == 0) 1012 break; 1013 1014 bufoffset = 0; 1015 while (bufoffset < count) { 1016 uio.uio_iov = &iov; 1017 uio.uio_iovcnt = 1; 1018 iov.iov_base = buf + bufoffset; 1019 iov.iov_len = count - bufoffset; 1020 uio.uio_offset = offset + bufoffset; 1021 uio.uio_resid = iov.iov_len; 1022 uio.uio_rw = UIO_WRITE; 1023 1024 if ((error = VOP_WRITE(uvp, &uio, 0, cred)) != 0) 1025 break; 1026 1027 bufoffset += (count - bufoffset) - uio.uio_resid; 1028 } 1029 1030 uio.uio_offset = offset + bufoffset; 1031 } 1032 1033 free(buf, M_TEMP); 1034 1035 return (error); 1036} 1037 1038/* 1039 * Copy file from lower to upper. 1040 * 1041 * If you need copy of the contents, set 1 to docopy. Otherwise, set 0 to 1042 * docopy. 1043 * 1044 * If no error returned, unp will be updated. 1045 */ 1046int 1047unionfs_copyfile(struct unionfs_node *unp, int docopy, struct ucred *cred, 1048 struct thread *td) 1049{ 1050 int error; 1051 struct mount *mp; 1052 struct vnode *udvp; 1053 struct vnode *lvp; 1054 struct vnode *uvp; 1055 struct vattr uva; 1056 1057 lvp = unp->un_lowervp; 1058 uvp = NULLVP; 1059 1060 if ((UNIONFSTOV(unp)->v_mount->mnt_flag & MNT_RDONLY)) 1061 return (EROFS); 1062 if (unp->un_dvp == NULLVP) 1063 return (EINVAL); 1064 if (unp->un_uppervp != NULLVP) 1065 return (EEXIST); 1066 udvp = VTOUNIONFS(unp->un_dvp)->un_uppervp; 1067 if (udvp == NULLVP) 1068 return (EROFS); 1069 if ((udvp->v_mount->mnt_flag & MNT_RDONLY)) 1070 return (EROFS); 1071 1072 error = VOP_ACCESS(lvp, VREAD, cred, td); 1073 if (error != 0) 1074 return (error); 1075 1076 if ((error = vn_start_write(udvp, &mp, V_WAIT | PCATCH)) != 0) 1077 return (error); 1078 error = unionfs_vn_create_on_upper(&uvp, udvp, unp, &uva, td); 1079 if (error != 0) { 1080 vn_finished_write(mp); 1081 return (error); 1082 } 1083 1084 if (docopy != 0) { 1085 error = VOP_OPEN(lvp, FREAD, cred, td, NULL); 1086 if (error == 0) { 1087 error = unionfs_copyfile_core(lvp, uvp, cred, td); 1088 VOP_CLOSE(lvp, FREAD, cred, td); 1089 } 1090 } 1091 VOP_CLOSE(uvp, FWRITE, cred, td); 1092 VOP_ADD_WRITECOUNT(uvp, -1); 1093 1094 vn_finished_write(mp); 1095 1096 if (error == 0) { 1097 /* Reset the attributes. Ignore errors. */ 1098 uva.va_type = VNON; 1099 VOP_SETATTR(uvp, &uva, cred); 1100 } 1101 1102 unionfs_node_update(unp, uvp, td); 1103 1104 return (error); 1105} 1106 1107/* 1108 * It checks whether vp can rmdir. (check empty) 1109 * 1110 * vp is unionfs vnode. 1111 * vp should be locked. 1112 */ 1113int 1114unionfs_check_rmdir(struct vnode *vp, struct ucred *cred, struct thread *td) 1115{ 1116 int error; 1117 int eofflag; 1118 int lookuperr; 1119 struct vnode *uvp; 1120 struct vnode *lvp; 1121 struct vnode *tvp; 1122 struct vattr va; 1123 struct componentname cn; 1124 /* 1125 * The size of buf needs to be larger than DIRBLKSIZ. 1126 */ 1127 char buf[256 * 6]; 1128 struct dirent *dp; 1129 struct dirent *edp; 1130 struct uio uio; 1131 struct iovec iov; 1132 1133 ASSERT_VOP_ELOCKED(vp, "unionfs_check_rmdir"); 1134 1135 eofflag = 0; 1136 uvp = UNIONFSVPTOUPPERVP(vp); 1137 lvp = UNIONFSVPTOLOWERVP(vp); 1138 1139 /* check opaque */ 1140 if ((error = VOP_GETATTR(uvp, &va, cred)) != 0) 1141 return (error); 1142 if (va.va_flags & OPAQUE) 1143 return (0); 1144 1145 /* open vnode */ 1146#ifdef MAC 1147 if ((error = mac_vnode_check_open(cred, vp, VEXEC|VREAD)) != 0) 1148 return (error); 1149#endif 1150 if ((error = VOP_ACCESS(vp, VEXEC|VREAD, cred, td)) != 0) 1151 return (error); 1152 if ((error = VOP_OPEN(vp, FREAD, cred, td, NULL)) != 0) 1153 return (error); 1154 1155 uio.uio_rw = UIO_READ; 1156 uio.uio_segflg = UIO_SYSSPACE; 1157 uio.uio_td = td; 1158 uio.uio_offset = 0; 1159 1160#ifdef MAC 1161 error = mac_vnode_check_readdir(td->td_ucred, lvp); 1162#endif 1163 while (!error && !eofflag) { 1164 iov.iov_base = buf; 1165 iov.iov_len = sizeof(buf); 1166 uio.uio_iov = &iov; 1167 uio.uio_iovcnt = 1; 1168 uio.uio_resid = iov.iov_len; 1169 1170 error = VOP_READDIR(lvp, &uio, cred, &eofflag, NULL, NULL); 1171 if (error != 0) 1172 break; 1173 if (eofflag == 0 && uio.uio_resid == sizeof(buf)) { 1174#ifdef DIAGNOSTIC 1175 panic("bad readdir response from lower FS."); 1176#endif 1177 break; 1178 } 1179 1180 edp = (struct dirent*)&buf[sizeof(buf) - uio.uio_resid]; 1181 for (dp = (struct dirent*)buf; !error && dp < edp; 1182 dp = (struct dirent*)((caddr_t)dp + dp->d_reclen)) { 1183 if (dp->d_type == DT_WHT || 1184 (dp->d_namlen == 1 && dp->d_name[0] == '.') || 1185 (dp->d_namlen == 2 && !bcmp(dp->d_name, "..", 2))) 1186 continue; 1187 1188 cn.cn_namelen = dp->d_namlen; 1189 cn.cn_pnbuf = NULL; 1190 cn.cn_nameptr = dp->d_name; 1191 cn.cn_nameiop = LOOKUP; 1192 cn.cn_flags = (LOCKPARENT | LOCKLEAF | SAVENAME | RDONLY | ISLASTCN); 1193 cn.cn_lkflags = LK_EXCLUSIVE; 1194 cn.cn_thread = td; 1195 cn.cn_cred = cred; 1196 cn.cn_consume = 0; 1197 1198 /* 1199 * check entry in lower. 1200 * Sometimes, readdir function returns 1201 * wrong entry. 1202 */ 1203 lookuperr = VOP_LOOKUP(lvp, &tvp, &cn); 1204 1205 if (!lookuperr) 1206 vput(tvp); 1207 else 1208 continue; /* skip entry */ 1209 1210 /* 1211 * check entry 1212 * If it has no exist/whiteout entry in upper, 1213 * directory is not empty. 1214 */ 1215 cn.cn_flags = (LOCKPARENT | LOCKLEAF | SAVENAME | RDONLY | ISLASTCN); 1216 lookuperr = VOP_LOOKUP(uvp, &tvp, &cn); 1217 1218 if (!lookuperr) 1219 vput(tvp); 1220 1221 /* ignore exist or whiteout entry */ 1222 if (!lookuperr || 1223 (lookuperr == ENOENT && (cn.cn_flags & ISWHITEOUT))) 1224 continue; 1225 1226 error = ENOTEMPTY; 1227 } 1228 } 1229 1230 /* close vnode */ 1231 VOP_CLOSE(vp, FREAD, cred, td); 1232 1233 return (error); 1234} 1235 1236#ifdef DIAGNOSTIC 1237 1238struct vnode * 1239unionfs_checkuppervp(struct vnode *vp, char *fil, int lno) 1240{ 1241 struct unionfs_node *unp; 1242 1243 unp = VTOUNIONFS(vp); 1244 1245#ifdef notyet 1246 if (vp->v_op != unionfs_vnodeop_p) { 1247 printf("unionfs_checkuppervp: on non-unionfs-node.\n"); 1248#ifdef KDB 1249 kdb_enter(KDB_WHY_UNIONFS, 1250 "unionfs_checkuppervp: on non-unionfs-node.\n"); 1251#endif 1252 panic("unionfs_checkuppervp"); 1253 }; 1254#endif 1255 return (unp->un_uppervp); 1256} 1257 1258struct vnode * 1259unionfs_checklowervp(struct vnode *vp, char *fil, int lno) 1260{ 1261 struct unionfs_node *unp; 1262 1263 unp = VTOUNIONFS(vp); 1264 1265#ifdef notyet 1266 if (vp->v_op != unionfs_vnodeop_p) { 1267 printf("unionfs_checklowervp: on non-unionfs-node.\n"); 1268#ifdef KDB 1269 kdb_enter(KDB_WHY_UNIONFS, 1270 "unionfs_checklowervp: on non-unionfs-node.\n"); 1271#endif 1272 panic("unionfs_checklowervp"); 1273 }; 1274#endif 1275 return (unp->un_lowervp); 1276} 1277#endif 1278