union_subr.c revision 193511
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 Masanori Ozawa <ozawa@ongs.co.jp>, ONGS Inc. 6 * Copyright (c) 2006 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: head/sys/fs/unionfs/union_subr.c 193511 2009-06-05 14:55:22Z rwatson $ 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; 353 354 vp->v_vnlock = &(vp->v_lock); 355 vp->v_data = NULL; 356 lockmgr(vp->v_vnlock, LK_EXCLUSIVE | LK_INTERLOCK, VI_MTX(vp)); 357 if (lvp != NULLVP) 358 VOP_UNLOCK(lvp, 0); 359 if (uvp != NULLVP) 360 VOP_UNLOCK(uvp, 0); 361 vp->v_object = NULL; 362 363 if (dvp != NULLVP && unp->un_hash.le_prev != NULL) 364 unionfs_rem_cached_vnode(unp, dvp); 365 366 if (lvp != NULLVP) { 367 vfslocked = VFS_LOCK_GIANT(lvp->v_mount); 368 vrele(lvp); 369 VFS_UNLOCK_GIANT(vfslocked); 370 } 371 if (uvp != NULLVP) { 372 vfslocked = VFS_LOCK_GIANT(uvp->v_mount); 373 vrele(uvp); 374 VFS_UNLOCK_GIANT(vfslocked); 375 } 376 if (dvp != NULLVP) { 377 vfslocked = VFS_LOCK_GIANT(dvp->v_mount); 378 vrele(dvp); 379 VFS_UNLOCK_GIANT(vfslocked); 380 unp->un_dvp = NULLVP; 381 } 382 if (unp->un_path != NULL) { 383 free(unp->un_path, M_UNIONFSPATH); 384 unp->un_path = NULL; 385 } 386 387 if (unp->un_hashtbl != NULL) { 388 for (count = 0; count <= unp->un_hashmask; count++) { 389 hd = unp->un_hashtbl + count; 390 LIST_FOREACH_SAFE(unp_t1, hd, un_hash, unp_t2) { 391 LIST_REMOVE(unp_t1, un_hash); 392 unp_t1->un_hash.le_next = NULL; 393 unp_t1->un_hash.le_prev = NULL; 394 } 395 } 396 hashdestroy(unp->un_hashtbl, M_UNIONFSHASH, unp->un_hashmask); 397 } 398 399 LIST_FOREACH_SAFE(unsp, &(unp->un_unshead), uns_list, unsp_tmp) { 400 LIST_REMOVE(unsp, uns_list); 401 free(unsp, M_TEMP); 402 } 403 free(unp, M_UNIONFSNODE); 404} 405 406/* 407 * Get the unionfs node status. 408 * You need exclusive lock this vnode. 409 */ 410void 411unionfs_get_node_status(struct unionfs_node *unp, struct thread *td, 412 struct unionfs_node_status **unspp) 413{ 414 struct unionfs_node_status *unsp; 415 pid_t pid = td->td_proc->p_pid; 416 417 KASSERT(NULL != unspp, ("null pointer")); 418 ASSERT_VOP_ELOCKED(UNIONFSTOV(unp), "unionfs_get_node_status"); 419 420 LIST_FOREACH(unsp, &(unp->un_unshead), uns_list) { 421 if (unsp->uns_pid == pid) { 422 *unspp = unsp; 423 return; 424 } 425 } 426 427 /* create a new unionfs node status */ 428 unsp = malloc(sizeof(struct unionfs_node_status), 429 M_TEMP, M_WAITOK | M_ZERO); 430 431 unsp->uns_pid = pid; 432 LIST_INSERT_HEAD(&(unp->un_unshead), unsp, uns_list); 433 434 *unspp = unsp; 435} 436 437/* 438 * Remove the unionfs node status, if you can. 439 * You need exclusive lock this vnode. 440 */ 441void 442unionfs_tryrem_node_status(struct unionfs_node *unp, 443 struct unionfs_node_status *unsp) 444{ 445 KASSERT(NULL != unsp, ("null pointer")); 446 ASSERT_VOP_ELOCKED(UNIONFSTOV(unp), "unionfs_get_node_status"); 447 448 if (0 < unsp->uns_lower_opencnt || 0 < unsp->uns_upper_opencnt) 449 return; 450 451 LIST_REMOVE(unsp, uns_list); 452 free(unsp, M_TEMP); 453} 454 455/* 456 * Create upper node attr. 457 */ 458void 459unionfs_create_uppervattr_core(struct unionfs_mount *ump, 460 struct vattr *lva, 461 struct vattr *uva, 462 struct thread *td) 463{ 464 VATTR_NULL(uva); 465 uva->va_type = lva->va_type; 466 uva->va_atime = lva->va_atime; 467 uva->va_mtime = lva->va_mtime; 468 uva->va_ctime = lva->va_ctime; 469 470 switch (ump->um_copymode) { 471 case UNIONFS_TRANSPARENT: 472 uva->va_mode = lva->va_mode; 473 uva->va_uid = lva->va_uid; 474 uva->va_gid = lva->va_gid; 475 break; 476 case UNIONFS_MASQUERADE: 477 if (ump->um_uid == lva->va_uid) { 478 uva->va_mode = lva->va_mode & 077077; 479 uva->va_mode |= (lva->va_type == VDIR ? ump->um_udir : ump->um_ufile) & 0700; 480 uva->va_uid = lva->va_uid; 481 uva->va_gid = lva->va_gid; 482 } else { 483 uva->va_mode = (lva->va_type == VDIR ? ump->um_udir : ump->um_ufile); 484 uva->va_uid = ump->um_uid; 485 uva->va_gid = ump->um_gid; 486 } 487 break; 488 default: /* UNIONFS_TRADITIONAL */ 489 FILEDESC_SLOCK(td->td_proc->p_fd); 490 uva->va_mode = 0777 & ~td->td_proc->p_fd->fd_cmask; 491 FILEDESC_SUNLOCK(td->td_proc->p_fd); 492 uva->va_uid = ump->um_uid; 493 uva->va_gid = ump->um_gid; 494 break; 495 } 496} 497 498/* 499 * Create upper node attr. 500 */ 501int 502unionfs_create_uppervattr(struct unionfs_mount *ump, 503 struct vnode *lvp, 504 struct vattr *uva, 505 struct ucred *cred, 506 struct thread *td) 507{ 508 int error; 509 struct vattr lva; 510 511 if ((error = VOP_GETATTR(lvp, &lva, cred))) 512 return (error); 513 514 unionfs_create_uppervattr_core(ump, &lva, uva, td); 515 516 return (error); 517} 518 519/* 520 * relookup 521 * 522 * dvp should be locked on entry and will be locked on return. 523 * 524 * If an error is returned, *vpp will be invalid, otherwise it will hold a 525 * locked, referenced vnode. If *vpp == dvp then remember that only one 526 * LK_EXCLUSIVE lock is held. 527 */ 528int 529unionfs_relookup(struct vnode *dvp, struct vnode **vpp, 530 struct componentname *cnp, struct componentname *cn, 531 struct thread *td, char *path, int pathlen, u_long nameiop) 532{ 533 int error; 534 535 cn->cn_namelen = pathlen; 536 cn->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK); 537 bcopy(path, cn->cn_pnbuf, pathlen); 538 cn->cn_pnbuf[pathlen] = '\0'; 539 540 cn->cn_nameiop = nameiop; 541 cn->cn_flags = (LOCKPARENT | LOCKLEAF | HASBUF | SAVENAME | ISLASTCN); 542 cn->cn_lkflags = LK_EXCLUSIVE; 543 cn->cn_thread = td; 544 cn->cn_cred = cnp->cn_cred; 545 546 cn->cn_nameptr = cn->cn_pnbuf; 547 cn->cn_consume = cnp->cn_consume; 548 549 if (nameiop == DELETE) 550 cn->cn_flags |= (cnp->cn_flags & (DOWHITEOUT | SAVESTART)); 551 else if (RENAME == nameiop) 552 cn->cn_flags |= (cnp->cn_flags & SAVESTART); 553 554 vref(dvp); 555 VOP_UNLOCK(dvp, 0); 556 557 if ((error = relookup(dvp, vpp, cn))) { 558 uma_zfree(namei_zone, cn->cn_pnbuf); 559 cn->cn_flags &= ~HASBUF; 560 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 561 } else 562 vrele(dvp); 563 564 return (error); 565} 566 567/* 568 * relookup for CREATE namei operation. 569 * 570 * dvp is unionfs vnode. dvp should be locked. 571 * 572 * If it called 'unionfs_copyfile' function by unionfs_link etc, 573 * VOP_LOOKUP information is broken. 574 * So it need relookup in order to create link etc. 575 */ 576int 577unionfs_relookup_for_create(struct vnode *dvp, struct componentname *cnp, 578 struct thread *td) 579{ 580 int error; 581 struct vnode *udvp; 582 struct vnode *vp; 583 struct componentname cn; 584 585 udvp = UNIONFSVPTOUPPERVP(dvp); 586 vp = NULLVP; 587 588 error = unionfs_relookup(udvp, &vp, cnp, &cn, td, cnp->cn_nameptr, 589 strlen(cnp->cn_nameptr), CREATE); 590 if (error) 591 return (error); 592 593 if (vp != NULLVP) { 594 if (udvp == vp) 595 vrele(vp); 596 else 597 vput(vp); 598 599 error = EEXIST; 600 } 601 602 if (cn.cn_flags & HASBUF) { 603 uma_zfree(namei_zone, cn.cn_pnbuf); 604 cn.cn_flags &= ~HASBUF; 605 } 606 607 if (!error) { 608 cn.cn_flags |= (cnp->cn_flags & HASBUF); 609 cnp->cn_flags = cn.cn_flags; 610 } 611 612 return (error); 613} 614 615/* 616 * relookup for DELETE namei operation. 617 * 618 * dvp is unionfs vnode. dvp should be locked. 619 */ 620int 621unionfs_relookup_for_delete(struct vnode *dvp, struct componentname *cnp, 622 struct thread *td) 623{ 624 int error; 625 struct vnode *udvp; 626 struct vnode *vp; 627 struct componentname cn; 628 629 udvp = UNIONFSVPTOUPPERVP(dvp); 630 vp = NULLVP; 631 632 error = unionfs_relookup(udvp, &vp, cnp, &cn, td, cnp->cn_nameptr, 633 strlen(cnp->cn_nameptr), DELETE); 634 if (error) 635 return (error); 636 637 if (vp == NULLVP) 638 error = ENOENT; 639 else { 640 if (udvp == vp) 641 vrele(vp); 642 else 643 vput(vp); 644 } 645 646 if (cn.cn_flags & HASBUF) { 647 uma_zfree(namei_zone, cn.cn_pnbuf); 648 cn.cn_flags &= ~HASBUF; 649 } 650 651 if (!error) { 652 cn.cn_flags |= (cnp->cn_flags & HASBUF); 653 cnp->cn_flags = cn.cn_flags; 654 } 655 656 return (error); 657} 658 659/* 660 * relookup for RENAME namei operation. 661 * 662 * dvp is unionfs vnode. dvp should be locked. 663 */ 664int 665unionfs_relookup_for_rename(struct vnode *dvp, struct componentname *cnp, 666 struct thread *td) 667{ 668 int error; 669 struct vnode *udvp; 670 struct vnode *vp; 671 struct componentname cn; 672 673 udvp = UNIONFSVPTOUPPERVP(dvp); 674 vp = NULLVP; 675 676 error = unionfs_relookup(udvp, &vp, cnp, &cn, td, cnp->cn_nameptr, 677 strlen(cnp->cn_nameptr), RENAME); 678 if (error) 679 return (error); 680 681 if (vp != NULLVP) { 682 if (udvp == vp) 683 vrele(vp); 684 else 685 vput(vp); 686 } 687 688 if (cn.cn_flags & HASBUF) { 689 uma_zfree(namei_zone, cn.cn_pnbuf); 690 cn.cn_flags &= ~HASBUF; 691 } 692 693 if (!error) { 694 cn.cn_flags |= (cnp->cn_flags & HASBUF); 695 cnp->cn_flags = cn.cn_flags; 696 } 697 698 return (error); 699 700} 701 702/* 703 * Update the unionfs_node. 704 * 705 * uvp is new locked upper vnode. unionfs vnode's lock will be exchanged to the 706 * uvp's lock and lower's lock will be unlocked. 707 */ 708static void 709unionfs_node_update(struct unionfs_node *unp, struct vnode *uvp, 710 struct thread *td) 711{ 712 unsigned count, lockrec; 713 struct vnode *vp; 714 struct vnode *lvp; 715 struct vnode *dvp; 716 717 vp = UNIONFSTOV(unp); 718 lvp = unp->un_lowervp; 719 ASSERT_VOP_ELOCKED(lvp, "unionfs_node_update"); 720 dvp = unp->un_dvp; 721 722 /* 723 * lock update 724 */ 725 VI_LOCK(vp); 726 unp->un_uppervp = uvp; 727 vp->v_vnlock = uvp->v_vnlock; 728 VI_UNLOCK(vp); 729 lockrec = lvp->v_vnlock->lk_recurse; 730 for (count = 0; count < lockrec; count++) 731 vn_lock(uvp, LK_EXCLUSIVE | LK_CANRECURSE | LK_RETRY); 732 733 /* 734 * cache update 735 */ 736 if (unp->un_path != NULL && dvp != NULLVP && vp->v_type == VDIR) { 737 static struct unionfs_node_hashhead *hd; 738 739 VI_LOCK(dvp); 740 hd = unionfs_get_hashhead(dvp, unp->un_path); 741 LIST_REMOVE(unp, un_hash); 742 LIST_INSERT_HEAD(hd, unp, un_hash); 743 VI_UNLOCK(dvp); 744 } 745} 746 747/* 748 * Create a new shadow dir. 749 * 750 * udvp should be locked on entry and will be locked on return. 751 * 752 * If no error returned, unp will be updated. 753 */ 754int 755unionfs_mkshadowdir(struct unionfs_mount *ump, struct vnode *udvp, 756 struct unionfs_node *unp, struct componentname *cnp, 757 struct thread *td) 758{ 759 int error; 760 struct vnode *lvp; 761 struct vnode *uvp; 762 struct vattr va; 763 struct vattr lva; 764 struct componentname cn; 765 struct mount *mp; 766 struct ucred *cred; 767 struct ucred *credbk; 768 struct uidinfo *rootinfo; 769 770 if (unp->un_uppervp != NULLVP) 771 return (EEXIST); 772 773 lvp = unp->un_lowervp; 774 uvp = NULLVP; 775 credbk = cnp->cn_cred; 776 777 /* Authority change to root */ 778 rootinfo = uifind((uid_t)0); 779 cred = crdup(cnp->cn_cred); 780 chgproccnt(cred->cr_ruidinfo, 1, 0); 781 change_euid(cred, rootinfo); 782 change_ruid(cred, rootinfo); 783 change_svuid(cred, (uid_t)0); 784 uifree(rootinfo); 785 cnp->cn_cred = cred; 786 787 memset(&cn, 0, sizeof(cn)); 788 789 if ((error = VOP_GETATTR(lvp, &lva, cnp->cn_cred))) 790 goto unionfs_mkshadowdir_abort; 791 792 if ((error = unionfs_relookup(udvp, &uvp, cnp, &cn, td, cnp->cn_nameptr, cnp->cn_namelen, CREATE))) 793 goto unionfs_mkshadowdir_abort; 794 if (uvp != NULLVP) { 795 if (udvp == uvp) 796 vrele(uvp); 797 else 798 vput(uvp); 799 800 error = EEXIST; 801 goto unionfs_mkshadowdir_free_out; 802 } 803 804 if ((error = vn_start_write(udvp, &mp, V_WAIT | PCATCH))) 805 goto unionfs_mkshadowdir_free_out; 806 unionfs_create_uppervattr_core(ump, &lva, &va, td); 807 808 error = VOP_MKDIR(udvp, &uvp, &cn, &va); 809 810 if (!error) { 811 unionfs_node_update(unp, uvp, td); 812 813 /* 814 * XXX The bug which cannot set uid/gid was corrected. 815 * Ignore errors. 816 */ 817 va.va_type = VNON; 818 VOP_SETATTR(uvp, &va, cn.cn_cred); 819 } 820 vn_finished_write(mp); 821 822unionfs_mkshadowdir_free_out: 823 if (cn.cn_flags & HASBUF) { 824 uma_zfree(namei_zone, cn.cn_pnbuf); 825 cn.cn_flags &= ~HASBUF; 826 } 827 828unionfs_mkshadowdir_abort: 829 cnp->cn_cred = credbk; 830 chgproccnt(cred->cr_ruidinfo, -1, 0); 831 crfree(cred); 832 833 return (error); 834} 835 836/* 837 * Create a new whiteout. 838 * 839 * dvp should be locked on entry and will be locked on return. 840 */ 841int 842unionfs_mkwhiteout(struct vnode *dvp, struct componentname *cnp, 843 struct thread *td, char *path) 844{ 845 int error; 846 struct vnode *wvp; 847 struct componentname cn; 848 struct mount *mp; 849 850 if (path == NULL) 851 path = cnp->cn_nameptr; 852 853 wvp = NULLVP; 854 if ((error = unionfs_relookup(dvp, &wvp, cnp, &cn, td, path, strlen(path), CREATE))) 855 return (error); 856 if (wvp != NULLVP) { 857 if (cn.cn_flags & HASBUF) { 858 uma_zfree(namei_zone, cn.cn_pnbuf); 859 cn.cn_flags &= ~HASBUF; 860 } 861 if (dvp == wvp) 862 vrele(wvp); 863 else 864 vput(wvp); 865 866 return (EEXIST); 867 } 868 869 if ((error = vn_start_write(dvp, &mp, V_WAIT | PCATCH))) 870 goto unionfs_mkwhiteout_free_out; 871 error = VOP_WHITEOUT(dvp, &cn, CREATE); 872 873 vn_finished_write(mp); 874 875unionfs_mkwhiteout_free_out: 876 if (cn.cn_flags & HASBUF) { 877 uma_zfree(namei_zone, cn.cn_pnbuf); 878 cn.cn_flags &= ~HASBUF; 879 } 880 881 return (error); 882} 883 884/* 885 * Create a new vnode for create a new shadow file. 886 * 887 * If an error is returned, *vpp will be invalid, otherwise it will hold a 888 * locked, referenced and opened vnode. 889 * 890 * unp is never updated. 891 */ 892static int 893unionfs_vn_create_on_upper(struct vnode **vpp, struct vnode *udvp, 894 struct unionfs_node *unp, struct vattr *uvap, 895 struct thread *td) 896{ 897 struct unionfs_mount *ump; 898 struct vnode *vp; 899 struct vnode *lvp; 900 struct ucred *cred; 901 struct vattr lva; 902 int fmode; 903 int error; 904 struct componentname cn; 905 906 ump = MOUNTTOUNIONFSMOUNT(UNIONFSTOV(unp)->v_mount); 907 vp = NULLVP; 908 lvp = unp->un_lowervp; 909 cred = td->td_ucred; 910 fmode = FFLAGS(O_WRONLY | O_CREAT | O_TRUNC | O_EXCL); 911 error = 0; 912 913 if ((error = VOP_GETATTR(lvp, &lva, cred)) != 0) 914 return (error); 915 unionfs_create_uppervattr_core(ump, &lva, uvap, td); 916 917 if (unp->un_path == NULL) 918 panic("unionfs: un_path is null"); 919 920 cn.cn_namelen = strlen(unp->un_path); 921 cn.cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK); 922 bcopy(unp->un_path, cn.cn_pnbuf, cn.cn_namelen + 1); 923 cn.cn_nameiop = CREATE; 924 cn.cn_flags = (LOCKPARENT | LOCKLEAF | HASBUF | SAVENAME | ISLASTCN); 925 cn.cn_lkflags = LK_EXCLUSIVE; 926 cn.cn_thread = td; 927 cn.cn_cred = cred; 928 cn.cn_nameptr = cn.cn_pnbuf; 929 cn.cn_consume = 0; 930 931 vref(udvp); 932 if ((error = relookup(udvp, &vp, &cn)) != 0) 933 goto unionfs_vn_create_on_upper_free_out2; 934 vrele(udvp); 935 936 if (vp != NULLVP) { 937 if (vp == udvp) 938 vrele(vp); 939 else 940 vput(vp); 941 error = EEXIST; 942 goto unionfs_vn_create_on_upper_free_out1; 943 } 944 945 if ((error = VOP_CREATE(udvp, &vp, &cn, uvap)) != 0) 946 goto unionfs_vn_create_on_upper_free_out1; 947 948 if ((error = VOP_OPEN(vp, fmode, cred, td, NULL)) != 0) { 949 vput(vp); 950 goto unionfs_vn_create_on_upper_free_out1; 951 } 952 vp->v_writecount++; 953 *vpp = vp; 954 955unionfs_vn_create_on_upper_free_out1: 956 VOP_UNLOCK(udvp, 0); 957 958unionfs_vn_create_on_upper_free_out2: 959 if (cn.cn_flags & HASBUF) { 960 uma_zfree(namei_zone, cn.cn_pnbuf); 961 cn.cn_flags &= ~HASBUF; 962 } 963 964 return (error); 965} 966 967/* 968 * Copy from lvp to uvp. 969 * 970 * lvp and uvp should be locked and opened on entry and will be locked and 971 * opened on return. 972 */ 973static int 974unionfs_copyfile_core(struct vnode *lvp, struct vnode *uvp, 975 struct ucred *cred, struct thread *td) 976{ 977 int error; 978 off_t offset; 979 int count; 980 int bufoffset; 981 char *buf; 982 struct uio uio; 983 struct iovec iov; 984 985 error = 0; 986 memset(&uio, 0, sizeof(uio)); 987 988 uio.uio_td = td; 989 uio.uio_segflg = UIO_SYSSPACE; 990 uio.uio_offset = 0; 991 992 buf = malloc(MAXBSIZE, M_TEMP, M_WAITOK); 993 994 while (error == 0) { 995 offset = uio.uio_offset; 996 997 uio.uio_iov = &iov; 998 uio.uio_iovcnt = 1; 999 iov.iov_base = buf; 1000 iov.iov_len = MAXBSIZE; 1001 uio.uio_resid = iov.iov_len; 1002 uio.uio_rw = UIO_READ; 1003 1004 if ((error = VOP_READ(lvp, &uio, 0, cred)) != 0) 1005 break; 1006 if ((count = MAXBSIZE - uio.uio_resid) == 0) 1007 break; 1008 1009 bufoffset = 0; 1010 while (bufoffset < count) { 1011 uio.uio_iov = &iov; 1012 uio.uio_iovcnt = 1; 1013 iov.iov_base = buf + bufoffset; 1014 iov.iov_len = count - bufoffset; 1015 uio.uio_offset = offset + bufoffset; 1016 uio.uio_resid = iov.iov_len; 1017 uio.uio_rw = UIO_WRITE; 1018 1019 if ((error = VOP_WRITE(uvp, &uio, 0, cred)) != 0) 1020 break; 1021 1022 bufoffset += (count - bufoffset) - uio.uio_resid; 1023 } 1024 1025 uio.uio_offset = offset + bufoffset; 1026 } 1027 1028 free(buf, M_TEMP); 1029 1030 return (error); 1031} 1032 1033/* 1034 * Copy file from lower to upper. 1035 * 1036 * If you need copy of the contents, set 1 to docopy. Otherwise, set 0 to 1037 * docopy. 1038 * 1039 * If no error returned, unp will be updated. 1040 */ 1041int 1042unionfs_copyfile(struct unionfs_node *unp, int docopy, struct ucred *cred, 1043 struct thread *td) 1044{ 1045 int error; 1046 struct mount *mp; 1047 struct vnode *udvp; 1048 struct vnode *lvp; 1049 struct vnode *uvp; 1050 struct vattr uva; 1051 1052 lvp = unp->un_lowervp; 1053 uvp = NULLVP; 1054 1055 if ((UNIONFSTOV(unp)->v_mount->mnt_flag & MNT_RDONLY)) 1056 return (EROFS); 1057 if (unp->un_dvp == NULLVP) 1058 return (EINVAL); 1059 if (unp->un_uppervp != NULLVP) 1060 return (EEXIST); 1061 udvp = VTOUNIONFS(unp->un_dvp)->un_uppervp; 1062 if (udvp == NULLVP) 1063 return (EROFS); 1064 if ((udvp->v_mount->mnt_flag & MNT_RDONLY)) 1065 return (EROFS); 1066 1067 error = VOP_ACCESS(lvp, VREAD, cred, td); 1068 if (error != 0) 1069 return (error); 1070 1071 if ((error = vn_start_write(udvp, &mp, V_WAIT | PCATCH)) != 0) 1072 return (error); 1073 error = unionfs_vn_create_on_upper(&uvp, udvp, unp, &uva, td); 1074 if (error != 0) { 1075 vn_finished_write(mp); 1076 return (error); 1077 } 1078 1079 if (docopy != 0) { 1080 error = VOP_OPEN(lvp, FREAD, cred, td, NULL); 1081 if (error == 0) { 1082 error = unionfs_copyfile_core(lvp, uvp, cred, td); 1083 VOP_CLOSE(lvp, FREAD, cred, td); 1084 } 1085 } 1086 VOP_CLOSE(uvp, FWRITE, cred, td); 1087 uvp->v_writecount--; 1088 1089 vn_finished_write(mp); 1090 1091 if (error == 0) { 1092 /* Reset the attributes. Ignore errors. */ 1093 uva.va_type = VNON; 1094 VOP_SETATTR(uvp, &uva, cred); 1095 } 1096 1097 unionfs_node_update(unp, uvp, td); 1098 1099 return (error); 1100} 1101 1102/* 1103 * It checks whether vp can rmdir. (check empty) 1104 * 1105 * vp is unionfs vnode. 1106 * vp should be locked. 1107 */ 1108int 1109unionfs_check_rmdir(struct vnode *vp, struct ucred *cred, struct thread *td) 1110{ 1111 int error; 1112 int eofflag; 1113 int lookuperr; 1114 struct vnode *uvp; 1115 struct vnode *lvp; 1116 struct vnode *tvp; 1117 struct vattr va; 1118 struct componentname cn; 1119 /* 1120 * The size of buf needs to be larger than DIRBLKSIZ. 1121 */ 1122 char buf[256 * 6]; 1123 struct dirent *dp; 1124 struct dirent *edp; 1125 struct uio uio; 1126 struct iovec iov; 1127 1128 ASSERT_VOP_ELOCKED(vp, "unionfs_check_rmdir"); 1129 1130 eofflag = 0; 1131 uvp = UNIONFSVPTOUPPERVP(vp); 1132 lvp = UNIONFSVPTOLOWERVP(vp); 1133 1134 /* check opaque */ 1135 if ((error = VOP_GETATTR(uvp, &va, cred)) != 0) 1136 return (error); 1137 if (va.va_flags & OPAQUE) 1138 return (0); 1139 1140 /* open vnode */ 1141#ifdef MAC 1142 if ((error = mac_vnode_check_open(cred, vp, VEXEC|VREAD)) != 0) 1143 return (error); 1144#endif 1145 if ((error = VOP_ACCESS(vp, VEXEC|VREAD, cred, td)) != 0) 1146 return (error); 1147 if ((error = VOP_OPEN(vp, FREAD, cred, td, NULL)) != 0) 1148 return (error); 1149 1150 uio.uio_rw = UIO_READ; 1151 uio.uio_segflg = UIO_SYSSPACE; 1152 uio.uio_td = td; 1153 uio.uio_offset = 0; 1154 1155#ifdef MAC 1156 error = mac_vnode_check_readdir(td->td_ucred, lvp); 1157#endif 1158 while (!error && !eofflag) { 1159 iov.iov_base = buf; 1160 iov.iov_len = sizeof(buf); 1161 uio.uio_iov = &iov; 1162 uio.uio_iovcnt = 1; 1163 uio.uio_resid = iov.iov_len; 1164 1165 error = VOP_READDIR(lvp, &uio, cred, &eofflag, NULL, NULL); 1166 if (error != 0) 1167 break; 1168 if (eofflag == 0 && uio.uio_resid == sizeof(buf)) { 1169#ifdef DIAGNOSTIC 1170 panic("bad readdir response from lower FS."); 1171#endif 1172 break; 1173 } 1174 1175 edp = (struct dirent*)&buf[sizeof(buf) - uio.uio_resid]; 1176 for (dp = (struct dirent*)buf; !error && dp < edp; 1177 dp = (struct dirent*)((caddr_t)dp + dp->d_reclen)) { 1178 if (dp->d_type == DT_WHT || 1179 (dp->d_namlen == 1 && dp->d_name[0] == '.') || 1180 (dp->d_namlen == 2 && !bcmp(dp->d_name, "..", 2))) 1181 continue; 1182 1183 cn.cn_namelen = dp->d_namlen; 1184 cn.cn_pnbuf = NULL; 1185 cn.cn_nameptr = dp->d_name; 1186 cn.cn_nameiop = LOOKUP; 1187 cn.cn_flags = (LOCKPARENT | LOCKLEAF | SAVENAME | RDONLY | ISLASTCN); 1188 cn.cn_lkflags = LK_EXCLUSIVE; 1189 cn.cn_thread = td; 1190 cn.cn_cred = cred; 1191 cn.cn_consume = 0; 1192 1193 /* 1194 * check entry in lower. 1195 * Sometimes, readdir function returns 1196 * wrong entry. 1197 */ 1198 lookuperr = VOP_LOOKUP(lvp, &tvp, &cn); 1199 1200 if (!lookuperr) 1201 vput(tvp); 1202 else 1203 continue; /* skip entry */ 1204 1205 /* 1206 * check entry 1207 * If it has no exist/whiteout entry in upper, 1208 * directory is not empty. 1209 */ 1210 cn.cn_flags = (LOCKPARENT | LOCKLEAF | SAVENAME | RDONLY | ISLASTCN); 1211 lookuperr = VOP_LOOKUP(uvp, &tvp, &cn); 1212 1213 if (!lookuperr) 1214 vput(tvp); 1215 1216 /* ignore exist or whiteout entry */ 1217 if (!lookuperr || 1218 (lookuperr == ENOENT && (cn.cn_flags & ISWHITEOUT))) 1219 continue; 1220 1221 error = ENOTEMPTY; 1222 } 1223 } 1224 1225 /* close vnode */ 1226 VOP_CLOSE(vp, FREAD, cred, td); 1227 1228 return (error); 1229} 1230 1231#ifdef DIAGNOSTIC 1232 1233struct vnode * 1234unionfs_checkuppervp(struct vnode *vp, char *fil, int lno) 1235{ 1236 struct unionfs_node *unp; 1237 1238 unp = VTOUNIONFS(vp); 1239 1240#ifdef notyet 1241 if (vp->v_op != unionfs_vnodeop_p) { 1242 printf("unionfs_checkuppervp: on non-unionfs-node.\n"); 1243#ifdef KDB 1244 kdb_enter(KDB_WHY_UNIONFS, 1245 "unionfs_checkuppervp: on non-unionfs-node.\n"); 1246#endif 1247 panic("unionfs_checkuppervp"); 1248 }; 1249#endif 1250 return (unp->un_uppervp); 1251} 1252 1253struct vnode * 1254unionfs_checklowervp(struct vnode *vp, char *fil, int lno) 1255{ 1256 struct unionfs_node *unp; 1257 1258 unp = VTOUNIONFS(vp); 1259 1260#ifdef notyet 1261 if (vp->v_op != unionfs_vnodeop_p) { 1262 printf("unionfs_checklowervp: on non-unionfs-node.\n"); 1263#ifdef KDB 1264 kdb_enter(KDB_WHY_UNIONFS, 1265 "unionfs_checklowervp: on non-unionfs-node.\n"); 1266#endif 1267 panic("unionfs_checklowervp"); 1268 }; 1269#endif 1270 return (unp->un_lowervp); 1271} 1272#endif 1273