1 /* $NetBSD: union_subr.c,v 1.18 2006/05/14 21:31:52 elad Exp $ */ 2 3 /* 4 * Copyright (c) 1994 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_subr.c 8.20 (Berkeley) 5/20/95 35 */ 36 37 /* 38 * Copyright (c) 1994 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_subr.c 8.20 (Berkeley) 5/20/95 72 */ 73 74 #include <sys/cdefs.h> 75 __KERNEL_RCSID(0, "$NetBSD: union_subr.c,v 1.18 2006/05/14 21:31:52 elad Exp $"); 76 77 #include <sys/param.h> 78 #include <sys/systm.h> 79 #include <sys/proc.h> 80 #include <sys/time.h> 81 #include <sys/kernel.h> 82 #include <sys/vnode.h> 83 #include <sys/namei.h> 84 #include <sys/malloc.h> 85 #include <sys/file.h> 86 #include <sys/filedesc.h> 87 #include <sys/queue.h> 88 #include <sys/mount.h> 89 #include <sys/stat.h> 90 #include <sys/kauth.h> 91 92 #include <uvm/uvm_extern.h> 93 94 #include <fs/union/union.h> 95 96 #ifdef DIAGNOSTIC 97 #include <sys/proc.h> 98 #endif 99 100 /* must be power of two, otherwise change UNION_HASH() */ 101 #define NHASH 32 102 103 /* unsigned int ... */ 104 #define UNION_HASH(u, l) \ 105 (((((unsigned long) (u)) + ((unsigned long) l)) >> 8) & (NHASH-1)) 106 107 static LIST_HEAD(unhead, union_node) unhead[NHASH]; 108 static int unvplock[NHASH]; 109 110 static int union_list_lock(int); 111 static void union_list_unlock(int); 112 void union_updatevp(struct union_node *, struct vnode *, struct vnode *); 113 static int union_relookup(struct union_mount *, struct vnode *, 114 struct vnode **, struct componentname *, 115 struct componentname *, const char *, int); 116 int union_vn_close(struct vnode *, int, kauth_cred_t, struct lwp *); 117 static void union_dircache_r(struct vnode *, struct vnode ***, int *); 118 struct vnode *union_dircache(struct vnode *, struct lwp *); 119 120 void 121 union_init() 122 { 123 int i; 124 125 for (i = 0; i < NHASH; i++) 126 LIST_INIT(&unhead[i]); 127 memset(unvplock, 0, sizeof(unvplock)); 128 } 129 130 /* 131 * Free global unionfs resources. 132 */ 133 void 134 union_done() 135 { 136 137 /* Make sure to unset the readdir hook. */ 138 vn_union_readdir_hook = NULL; 139 } 140 141 static int 142 union_list_lock(ix) 143 int ix; 144 { 145 146 if (unvplock[ix] & UN_LOCKED) { 147 unvplock[ix] |= UN_WANTED; 148 (void) tsleep(&unvplock[ix], PINOD, "unionlk", 0); 149 return (1); 150 } 151 152 unvplock[ix] |= UN_LOCKED; 153 154 return (0); 155 } 156 157 static void 158 union_list_unlock(ix) 159 int ix; 160 { 161 162 unvplock[ix] &= ~UN_LOCKED; 163 164 if (unvplock[ix] & UN_WANTED) { 165 unvplock[ix] &= ~UN_WANTED; 166 wakeup(&unvplock[ix]); 167 } 168 } 169 170 void 171 union_updatevp(un, uppervp, lowervp) 172 struct union_node *un; 173 struct vnode *uppervp; 174 struct vnode *lowervp; 175 { 176 int ohash = UNION_HASH(un->un_uppervp, un->un_lowervp); 177 int nhash = UNION_HASH(uppervp, lowervp); 178 int docache = (lowervp != NULLVP || uppervp != NULLVP); 179 int lhash, uhash; 180 181 /* 182 * Ensure locking is ordered from lower to higher 183 * to avoid deadlocks. 184 */ 185 if (nhash < ohash) { 186 lhash = nhash; 187 uhash = ohash; 188 } else { 189 lhash = ohash; 190 uhash = nhash; 191 } 192 193 if (lhash != uhash) 194 while (union_list_lock(lhash)) 195 continue; 196 197 while (union_list_lock(uhash)) 198 continue; 199 200 if (ohash != nhash || !docache) { 201 if (un->un_flags & UN_CACHED) { 202 un->un_flags &= ~UN_CACHED; 203 LIST_REMOVE(un, un_cache); 204 } 205 } 206 207 if (ohash != nhash) 208 union_list_unlock(ohash); 209 210 if (un->un_lowervp != lowervp) { 211 if (un->un_lowervp) { 212 vrele(un->un_lowervp); 213 if (un->un_path) { 214 free(un->un_path, M_TEMP); 215 un->un_path = 0; 216 } 217 if (un->un_dirvp) { 218 vrele(un->un_dirvp); 219 un->un_dirvp = NULLVP; 220 } 221 } 222 un->un_lowervp = lowervp; 223 un->un_lowersz = VNOVAL; 224 } 225 226 if (un->un_uppervp != uppervp) { 227 if (un->un_uppervp) 228 vrele(un->un_uppervp); 229 230 un->un_uppervp = uppervp; 231 un->un_uppersz = VNOVAL; 232 } 233 234 if (docache && (ohash != nhash)) { 235 LIST_INSERT_HEAD(&unhead[nhash], un, un_cache); 236 un->un_flags |= UN_CACHED; 237 } 238 239 union_list_unlock(nhash); 240 } 241 242 void 243 union_newlower(un, lowervp) 244 struct union_node *un; 245 struct vnode *lowervp; 246 { 247 248 union_updatevp(un, un->un_uppervp, lowervp); 249 } 250 251 void 252 union_newupper(un, uppervp) 253 struct union_node *un; 254 struct vnode *uppervp; 255 { 256 257 union_updatevp(un, uppervp, un->un_lowervp); 258 } 259 260 /* 261 * Keep track of size changes in the underlying vnodes. 262 * If the size changes, then callback to the vm layer 263 * giving priority to the upper layer size. 264 */ 265 void 266 union_newsize(vp, uppersz, lowersz) 267 struct vnode *vp; 268 off_t uppersz, lowersz; 269 { 270 struct union_node *un; 271 off_t sz; 272 273 /* only interested in regular files */ 274 if (vp->v_type != VREG) 275 return; 276 277 un = VTOUNION(vp); 278 sz = VNOVAL; 279 280 if ((uppersz != VNOVAL) && (un->un_uppersz != uppersz)) { 281 un->un_uppersz = uppersz; 282 if (sz == VNOVAL) 283 sz = un->un_uppersz; 284 } 285 286 if ((lowersz != VNOVAL) && (un->un_lowersz != lowersz)) { 287 un->un_lowersz = lowersz; 288 if (sz == VNOVAL) 289 sz = un->un_lowersz; 290 } 291 292 if (sz != VNOVAL) { 293 #ifdef UNION_DIAGNOSTIC 294 printf("union: %s size now %qd\n", 295 uppersz != VNOVAL ? "upper" : "lower", sz); 296 #endif 297 uvm_vnp_setsize(vp, sz); 298 } 299 } 300 301 /* 302 * allocate a union_node/vnode pair. the vnode is 303 * referenced and locked. the new vnode is returned 304 * via (vpp). (mp) is the mountpoint of the union filesystem, 305 * (dvp) is the parent directory where the upper layer object 306 * should exist (but doesn't) and (cnp) is the componentname 307 * information which is partially copied to allow the upper 308 * layer object to be created at a later time. (uppervp) 309 * and (lowervp) reference the upper and lower layer objects 310 * being mapped. either, but not both, can be nil. 311 * if supplied, (uppervp) is locked. 312 * the reference is either maintained in the new union_node 313 * object which is allocated, or they are vrele'd. 314 * 315 * all union_nodes are maintained on a singly-linked 316 * list. new nodes are only allocated when they cannot 317 * be found on this list. entries on the list are 318 * removed when the vfs reclaim entry is called. 319 * 320 * a single lock is kept for the entire list. this is 321 * needed because the getnewvnode() function can block 322 * waiting for a vnode to become free, in which case there 323 * may be more than one process trying to get the same 324 * vnode. this lock is only taken if we are going to 325 * call getnewvnode, since the kernel itself is single-threaded. 326 * 327 * if an entry is found on the list, then call vget() to 328 * take a reference. this is done because there may be 329 * zero references to it and so it needs to removed from 330 * the vnode free list. 331 */ 332 int 333 union_allocvp(vpp, mp, undvp, dvp, cnp, uppervp, lowervp, docache) 334 struct vnode **vpp; 335 struct mount *mp; 336 struct vnode *undvp; /* parent union vnode */ 337 struct vnode *dvp; /* may be null */ 338 struct componentname *cnp; /* may be null */ 339 struct vnode *uppervp; /* may be null */ 340 struct vnode *lowervp; /* may be null */ 341 int docache; 342 { 343 int error; 344 struct union_node *un = NULL; 345 struct vnode *xlowervp = NULLVP; 346 struct union_mount *um = MOUNTTOUNIONMOUNT(mp); 347 int hash = 0; 348 int vflag; 349 int try; 350 351 if (uppervp == NULLVP && lowervp == NULLVP) 352 panic("union: unidentifiable allocation"); 353 354 if (uppervp && lowervp && (uppervp->v_type != lowervp->v_type)) { 355 xlowervp = lowervp; 356 lowervp = NULLVP; 357 } 358 359 /* detect the root vnode (and aliases) */ 360 vflag = VLAYER; 361 if ((uppervp == um->um_uppervp) && 362 ((lowervp == NULLVP) || lowervp == um->um_lowervp)) { 363 if (lowervp == NULLVP) { 364 lowervp = um->um_lowervp; 365 if (lowervp != NULLVP) 366 VREF(lowervp); 367 } 368 vflag = VROOT; 369 } 370 371 loop: 372 if (!docache) { 373 un = 0; 374 } else for (try = 0; try < 3; try++) { 375 switch (try) { 376 case 0: 377 if (lowervp == NULLVP) 378 continue; 379 hash = UNION_HASH(uppervp, lowervp); 380 break; 381 382 case 1: 383 if (uppervp == NULLVP) 384 continue; 385 hash = UNION_HASH(uppervp, NULLVP); 386 break; 387 388 case 2: 389 if (lowervp == NULLVP) 390 continue; 391 hash = UNION_HASH(NULLVP, lowervp); 392 break; 393 } 394 395 while (union_list_lock(hash)) 396 continue; 397 398 for (un = unhead[hash].lh_first; un != 0; 399 un = un->un_cache.le_next) { 400 if ((un->un_lowervp == lowervp || 401 un->un_lowervp == NULLVP) && 402 (un->un_uppervp == uppervp || 403 un->un_uppervp == NULLVP) && 404 (UNIONTOV(un)->v_mount == mp)) { 405 if (vget(UNIONTOV(un), 0)) { 406 union_list_unlock(hash); 407 goto loop; 408 } 409 break; 410 } 411 } 412 413 union_list_unlock(hash); 414 415 if (un) 416 break; 417 } 418 419 if (un) { 420 /* 421 * Obtain a lock on the union_node. 422 * uppervp is locked, though un->un_uppervp 423 * may not be. this doesn't break the locking 424 * hierarchy since in the case that un->un_uppervp 425 * is not yet locked it will be vrele'd and replaced 426 * with uppervp. 427 */ 428 429 if ((dvp != NULLVP) && (uppervp == dvp)) { 430 /* 431 * Access ``.'', so (un) will already 432 * be locked. Since this process has 433 * the lock on (uppervp) no other 434 * process can hold the lock on (un). 435 */ 436 #ifdef DIAGNOSTIC 437 if ((un->un_flags & UN_LOCKED) == 0) 438 panic("union: . not locked"); 439 else if (curproc && un->un_pid != curproc->p_pid && 440 un->un_pid > -1 && curproc->p_pid > -1) 441 panic("union: allocvp not lock owner"); 442 #endif 443 } else { 444 if (un->un_flags & UN_LOCKED) { 445 vrele(UNIONTOV(un)); 446 un->un_flags |= UN_WANTED; 447 (void) tsleep(&un->un_flags, PINOD, 448 "unionalloc", 0); 449 goto loop; 450 } 451 un->un_flags |= UN_LOCKED; 452 453 #ifdef DIAGNOSTIC 454 if (curproc) 455 un->un_pid = curproc->p_pid; 456 else 457 un->un_pid = -1; 458 #endif 459 } 460 461 /* 462 * At this point, the union_node is locked, 463 * un->un_uppervp may not be locked, and uppervp 464 * is locked or nil. 465 */ 466 467 /* 468 * Save information about the upper layer. 469 */ 470 if (uppervp != un->un_uppervp) { 471 union_newupper(un, uppervp); 472 } else if (uppervp) { 473 vrele(uppervp); 474 } 475 476 if (un->un_uppervp) { 477 un->un_flags |= UN_ULOCK; 478 un->un_flags &= ~UN_KLOCK; 479 } 480 481 /* 482 * Save information about the lower layer. 483 * This needs to keep track of pathname 484 * and directory information which union_vn_create 485 * might need. 486 */ 487 if (lowervp != un->un_lowervp) { 488 union_newlower(un, lowervp); 489 if (cnp && (lowervp != NULLVP)) { 490 un->un_hash = cnp->cn_hash; 491 un->un_path = malloc(cnp->cn_namelen+1, 492 M_TEMP, M_WAITOK); 493 memcpy(un->un_path, cnp->cn_nameptr, 494 cnp->cn_namelen); 495 un->un_path[cnp->cn_namelen] = '\0'; 496 VREF(dvp); 497 un->un_dirvp = dvp; 498 } 499 } else if (lowervp) { 500 vrele(lowervp); 501 } 502 *vpp = UNIONTOV(un); 503 return (0); 504 } 505 506 if (docache) { 507 /* 508 * otherwise lock the vp list while we call getnewvnode 509 * since that can block. 510 */ 511 hash = UNION_HASH(uppervp, lowervp); 512 513 if (union_list_lock(hash)) 514 goto loop; 515 } 516 517 error = getnewvnode(VT_UNION, mp, union_vnodeop_p, vpp); 518 if (error) { 519 if (uppervp) { 520 if (dvp == uppervp) 521 vrele(uppervp); 522 else 523 vput(uppervp); 524 } 525 if (lowervp) 526 vrele(lowervp); 527 528 goto out; 529 } 530 531 MALLOC((*vpp)->v_data, void *, sizeof(struct union_node), 532 M_TEMP, M_WAITOK); 533 534 (*vpp)->v_flag |= vflag; 535 (*vpp)->v_vnlock = NULL; /* Make upper layers call VOP_LOCK */ 536 if (uppervp) 537 (*vpp)->v_type = uppervp->v_type; 538 else 539 (*vpp)->v_type = lowervp->v_type; 540 un = VTOUNION(*vpp); 541 un->un_vnode = *vpp; 542 un->un_uppervp = uppervp; 543 un->un_uppersz = VNOVAL; 544 un->un_lowervp = lowervp; 545 un->un_lowersz = VNOVAL; 546 un->un_pvp = undvp; 547 if (undvp != NULLVP) 548 VREF(undvp); 549 un->un_dircache = 0; 550 un->un_openl = 0; 551 un->un_flags = UN_LOCKED; 552 if (un->un_uppervp) 553 un->un_flags |= UN_ULOCK; 554 #ifdef DIAGNOSTIC 555 if (curproc) 556 un->un_pid = curproc->p_pid; 557 else 558 un->un_pid = -1; 559 #endif 560 if (dvp && cnp && (lowervp != NULLVP)) { 561 un->un_hash = cnp->cn_hash; 562 un->un_path = malloc(cnp->cn_namelen+1, M_TEMP, M_WAITOK); 563 memcpy(un->un_path, cnp->cn_nameptr, cnp->cn_namelen); 564 un->un_path[cnp->cn_namelen] = '\0'; 565 VREF(dvp); 566 un->un_dirvp = dvp; 567 } else { 568 un->un_hash = 0; 569 un->un_path = 0; 570 un->un_dirvp = 0; 571 } 572 573 if (docache) { 574 LIST_INSERT_HEAD(&unhead[hash], un, un_cache); 575 un->un_flags |= UN_CACHED; 576 } 577 578 if (xlowervp) 579 vrele(xlowervp); 580 581 out: 582 if (docache) 583 union_list_unlock(hash); 584 585 return (error); 586 } 587 588 int 589 union_freevp(vp) 590 struct vnode *vp; 591 { 592 struct union_node *un = VTOUNION(vp); 593 594 if (un->un_flags & UN_CACHED) { 595 un->un_flags &= ~UN_CACHED; 596 LIST_REMOVE(un, un_cache); 597 } 598 599 if (un->un_pvp != NULLVP) 600 vrele(un->un_pvp); 601 if (un->un_uppervp != NULLVP) 602 vrele(un->un_uppervp); 603 if (un->un_lowervp != NULLVP) 604 vrele(un->un_lowervp); 605 if (un->un_dirvp != NULLVP) 606 vrele(un->un_dirvp); 607 if (un->un_path) 608 free(un->un_path, M_TEMP); 609 610 FREE(vp->v_data, M_TEMP); 611 vp->v_data = 0; 612 613 return (0); 614 } 615 616 /* 617 * copyfile. copy the vnode (fvp) to the vnode (tvp) 618 * using a sequence of reads and writes. both (fvp) 619 * and (tvp) are locked on entry and exit. 620 */ 621 int 622 union_copyfile(fvp, tvp, cred, l) 623 struct vnode *fvp; 624 struct vnode *tvp; 625 kauth_cred_t cred; 626 struct lwp *l; 627 { 628 char *tbuf; 629 struct uio uio; 630 struct iovec iov; 631 int error = 0; 632 633 /* 634 * strategy: 635 * allocate a buffer of size MAXBSIZE. 636 * loop doing reads and writes, keeping track 637 * of the current uio offset. 638 * give up at the first sign of trouble. 639 */ 640 641 uio.uio_offset = 0; 642 UIO_SETUP_SYSSPACE(&uio); 643 644 VOP_UNLOCK(fvp, 0); /* XXX */ 645 VOP_LEASE(fvp, l, cred, LEASE_READ); 646 vn_lock(fvp, LK_EXCLUSIVE | LK_RETRY); /* XXX */ 647 VOP_UNLOCK(tvp, 0); /* XXX */ 648 VOP_LEASE(tvp, l, cred, LEASE_WRITE); 649 vn_lock(tvp, LK_EXCLUSIVE | LK_RETRY); /* XXX */ 650 651 tbuf = malloc(MAXBSIZE, M_TEMP, M_WAITOK); 652 653 /* ugly loop follows... */ 654 do { 655 off_t offset = uio.uio_offset; 656 657 uio.uio_iov = &iov; 658 uio.uio_iovcnt = 1; 659 iov.iov_base = tbuf; 660 iov.iov_len = MAXBSIZE; 661 uio.uio_resid = iov.iov_len; 662 uio.uio_rw = UIO_READ; 663 error = VOP_READ(fvp, &uio, 0, cred); 664 665 if (error == 0) { 666 uio.uio_iov = &iov; 667 uio.uio_iovcnt = 1; 668 iov.iov_base = tbuf; 669 iov.iov_len = MAXBSIZE - uio.uio_resid; 670 uio.uio_offset = offset; 671 uio.uio_rw = UIO_WRITE; 672 uio.uio_resid = iov.iov_len; 673 674 if (uio.uio_resid == 0) 675 break; 676 677 do { 678 error = VOP_WRITE(tvp, &uio, 0, cred); 679 } while ((uio.uio_resid > 0) && (error == 0)); 680 } 681 682 } while (error == 0); 683 684 free(tbuf, M_TEMP); 685 return (error); 686 } 687 688 /* 689 * (un) is assumed to be locked on entry and remains 690 * locked on exit. 691 */ 692 int 693 union_copyup(un, docopy, cred, l) 694 struct union_node *un; 695 int docopy; 696 kauth_cred_t cred; 697 struct lwp *l; 698 { 699 int error; 700 struct mount *mp; 701 struct vnode *lvp, *uvp; 702 struct vattr lvattr, uvattr; 703 704 if ((error = vn_start_write(un->un_dirvp, &mp, V_WAIT | V_PCATCH)) != 0) 705 return (error); 706 error = union_vn_create(&uvp, un, l); 707 if (error) { 708 vn_finished_write(mp, 0); 709 return (error); 710 } 711 712 /* at this point, uppervp is locked */ 713 union_newupper(un, uvp); 714 un->un_flags |= UN_ULOCK; 715 716 lvp = un->un_lowervp; 717 718 if (docopy) { 719 /* 720 * XX - should not ignore errors 721 * from VOP_CLOSE 722 */ 723 vn_lock(lvp, LK_EXCLUSIVE | LK_RETRY); 724 725 error = VOP_GETATTR(lvp, &lvattr, cred, l); 726 if (error == 0) 727 error = VOP_OPEN(lvp, FREAD, cred, l); 728 if (error == 0) { 729 error = union_copyfile(lvp, uvp, cred, l); 730 (void) VOP_CLOSE(lvp, FREAD, cred, l); 731 } 732 if (error == 0) { 733 /* Copy permissions up too */ 734 VATTR_NULL(&uvattr); 735 uvattr.va_mode = lvattr.va_mode; 736 uvattr.va_flags = lvattr.va_flags; 737 error = VOP_SETATTR(uvp, &uvattr, cred, l); 738 } 739 VOP_UNLOCK(lvp, 0); 740 #ifdef UNION_DIAGNOSTIC 741 if (error == 0) 742 uprintf("union: copied up %s\n", un->un_path); 743 #endif 744 745 } 746 vn_finished_write(mp, 0); 747 union_vn_close(uvp, FWRITE, cred, l); 748 749 /* 750 * Subsequent IOs will go to the top layer, so 751 * call close on the lower vnode and open on the 752 * upper vnode to ensure that the filesystem keeps 753 * its references counts right. This doesn't do 754 * the right thing with (cred) and (FREAD) though. 755 * Ignoring error returns is not right, either. 756 */ 757 if (error == 0) { 758 int i; 759 760 vn_lock(lvp, LK_EXCLUSIVE | LK_RETRY); 761 for (i = 0; i < un->un_openl; i++) { 762 (void) VOP_CLOSE(lvp, FREAD, cred, l); 763 (void) VOP_OPEN(uvp, FREAD, cred, l); 764 } 765 un->un_openl = 0; 766 VOP_UNLOCK(lvp, 0); 767 } 768 769 return (error); 770 771 } 772 773 static int 774 union_relookup(um, dvp, vpp, cnp, cn, path, pathlen) 775 struct union_mount *um; 776 struct vnode *dvp; 777 struct vnode **vpp; 778 struct componentname *cnp; 779 struct componentname *cn; 780 const char *path; 781 int pathlen; 782 { 783 int error; 784 785 /* 786 * A new componentname structure must be faked up because 787 * there is no way to know where the upper level cnp came 788 * from or what it is being used for. This must duplicate 789 * some of the work done by NDINIT, some of the work done 790 * by namei, some of the work done by lookup and some of 791 * the work done by VOP_LOOKUP when given a CREATE flag. 792 * Conclusion: Horrible. 793 * 794 * The pathname buffer will be PNBUF_PUT'd by VOP_MKDIR. 795 */ 796 cn->cn_namelen = pathlen; 797 if ((cn->cn_namelen + 1) > MAXPATHLEN) 798 return (ENAMETOOLONG); 799 cn->cn_pnbuf = PNBUF_GET(); 800 memcpy(cn->cn_pnbuf, path, cn->cn_namelen); 801 cn->cn_pnbuf[cn->cn_namelen] = '\0'; 802 803 cn->cn_nameiop = CREATE; 804 cn->cn_flags = (LOCKPARENT|HASBUF|SAVENAME|SAVESTART|ISLASTCN); 805 cn->cn_lwp = cnp->cn_lwp; 806 if (um->um_op == UNMNT_ABOVE) 807 cn->cn_cred = cnp->cn_cred; 808 else 809 cn->cn_cred = um->um_cred; 810 cn->cn_nameptr = cn->cn_pnbuf; 811 cn->cn_hash = cnp->cn_hash; 812 cn->cn_consume = cnp->cn_consume; 813 814 VREF(dvp); 815 error = relookup(dvp, vpp, cn); 816 if (!error) 817 vrele(dvp); 818 else { 819 PNBUF_PUT(cn->cn_pnbuf); 820 cn->cn_pnbuf = 0; 821 } 822 823 return (error); 824 } 825 826 /* 827 * Create a shadow directory in the upper layer. 828 * The new vnode is returned locked. 829 * 830 * (um) points to the union mount structure for access to the 831 * the mounting process's credentials. 832 * (dvp) is the directory in which to create the shadow directory. 833 * it is unlocked on entry and exit. 834 * (cnp) is the componentname to be created. 835 * (vpp) is the returned newly created shadow directory, which 836 * is returned locked. 837 * 838 * N.B. We still attempt to create shadow directories even if the union 839 * is mounted read-only, which is a little nonintuitive. 840 */ 841 int 842 union_mkshadow(um, dvp, cnp, vpp) 843 struct union_mount *um; 844 struct vnode *dvp; 845 struct componentname *cnp; 846 struct vnode **vpp; 847 { 848 int error; 849 struct vattr va; 850 struct lwp *l = cnp->cn_lwp; 851 struct componentname cn; 852 struct mount *mp; 853 854 if ((error = vn_start_write(dvp, &mp, V_WAIT | V_PCATCH)) != 0) 855 return (error); 856 error = union_relookup(um, dvp, vpp, cnp, &cn, 857 cnp->cn_nameptr, cnp->cn_namelen); 858 if (error) { 859 vn_finished_write(mp, 0); 860 return (error); 861 } 862 863 if (*vpp) { 864 VOP_ABORTOP(dvp, &cn); 865 VOP_UNLOCK(dvp, 0); 866 vrele(*vpp); 867 vn_finished_write(mp, 0); 868 *vpp = NULLVP; 869 return (EEXIST); 870 } 871 872 /* 873 * policy: when creating the shadow directory in the 874 * upper layer, create it owned by the user who did 875 * the mount, group from parent directory, and mode 876 * 777 modified by umask (ie mostly identical to the 877 * mkdir syscall). (jsp, kb) 878 */ 879 880 VATTR_NULL(&va); 881 va.va_type = VDIR; 882 va.va_mode = um->um_cmode; 883 884 /* VOP_LEASE: dvp is locked */ 885 VOP_LEASE(dvp, l, cn.cn_cred, LEASE_WRITE); 886 887 error = VOP_MKDIR(dvp, vpp, &cn, &va); 888 vn_finished_write(mp, 0); 889 return (error); 890 } 891 892 /* 893 * Create a whiteout entry in the upper layer. 894 * 895 * (um) points to the union mount structure for access to the 896 * the mounting process's credentials. 897 * (dvp) is the directory in which to create the whiteout. 898 * it is locked on entry and exit. 899 * (cnp) is the componentname to be created. 900 */ 901 int 902 union_mkwhiteout(um, dvp, cnp, path) 903 struct union_mount *um; 904 struct vnode *dvp; 905 struct componentname *cnp; 906 char *path; 907 { 908 int error; 909 struct lwp *l = cnp->cn_lwp; 910 struct vnode *wvp; 911 struct componentname cn; 912 struct mount *mp; 913 914 VOP_UNLOCK(dvp, 0); 915 if ((error = vn_start_write(dvp, &mp, V_WAIT | V_PCATCH)) != 0) 916 return (error); 917 error = union_relookup(um, dvp, &wvp, cnp, &cn, path, strlen(path)); 918 if (error) { 919 vn_finished_write(mp, 0); 920 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 921 return (error); 922 } 923 924 if (wvp) { 925 VOP_ABORTOP(dvp, &cn); 926 vrele(dvp); 927 vrele(wvp); 928 vn_finished_write(mp, 0); 929 return (EEXIST); 930 } 931 932 /* VOP_LEASE: dvp is locked */ 933 VOP_LEASE(dvp, l, l->l_proc->p_cred, LEASE_WRITE); 934 935 error = VOP_WHITEOUT(dvp, &cn, CREATE); 936 if (error) 937 VOP_ABORTOP(dvp, &cn); 938 939 vrele(dvp); 940 vn_finished_write(mp, 0); 941 942 return (error); 943 } 944 945 /* 946 * union_vn_create: creates and opens a new shadow file 947 * on the upper union layer. this function is similar 948 * in spirit to calling vn_open but it avoids calling namei(). 949 * the problem with calling namei is that a) it locks too many 950 * things, and b) it doesn't start at the "right" directory, 951 * whereas relookup is told where to start. 952 */ 953 int 954 union_vn_create(vpp, un, l) 955 struct vnode **vpp; 956 struct union_node *un; 957 struct lwp *l; 958 { 959 struct vnode *vp; 960 kauth_cred_t cred = l->l_proc->p_cred; 961 struct vattr vat; 962 struct vattr *vap = &vat; 963 int fmode = FFLAGS(O_WRONLY|O_CREAT|O_TRUNC|O_EXCL); 964 int error; 965 int cmode = UN_FILEMODE & ~l->l_proc->p_cwdi->cwdi_cmask; 966 struct componentname cn; 967 968 *vpp = NULLVP; 969 970 /* 971 * Build a new componentname structure (for the same 972 * reasons outlines in union_mkshadow). 973 * The difference here is that the file is owned by 974 * the current user, rather than by the person who 975 * did the mount, since the current user needs to be 976 * able to write the file (that's why it is being 977 * copied in the first place). 978 */ 979 cn.cn_namelen = strlen(un->un_path); 980 if ((cn.cn_namelen + 1) > MAXPATHLEN) 981 return (ENAMETOOLONG); 982 cn.cn_pnbuf = PNBUF_GET(); 983 memcpy(cn.cn_pnbuf, un->un_path, cn.cn_namelen+1); 984 cn.cn_nameiop = CREATE; 985 cn.cn_flags = (LOCKPARENT|HASBUF|SAVENAME|SAVESTART|ISLASTCN); 986 cn.cn_lwp = l; 987 cn.cn_cred = l->l_proc->p_cred; 988 cn.cn_nameptr = cn.cn_pnbuf; 989 cn.cn_hash = un->un_hash; 990 cn.cn_consume = 0; 991 992 VREF(un->un_dirvp); 993 if ((error = relookup(un->un_dirvp, &vp, &cn)) != 0) 994 return (error); 995 vrele(un->un_dirvp); 996 997 if (vp) { 998 VOP_ABORTOP(un->un_dirvp, &cn); 999 if (un->un_dirvp == vp) 1000 vrele(un->un_dirvp); 1001 else 1002 vput(un->un_dirvp); 1003 vrele(vp); 1004 return (EEXIST); 1005 } 1006 1007 /* 1008 * Good - there was no race to create the file 1009 * so go ahead and create it. The permissions 1010 * on the file will be 0666 modified by the 1011 * current user's umask. Access to the file, while 1012 * it is unioned, will require access to the top *and* 1013 * bottom files. Access when not unioned will simply 1014 * require access to the top-level file. 1015 * TODO: confirm choice of access permissions. 1016 */ 1017 VATTR_NULL(vap); 1018 vap->va_type = VREG; 1019 vap->va_mode = cmode; 1020 VOP_LEASE(un->un_dirvp, l, cred, LEASE_WRITE); 1021 if ((error = VOP_CREATE(un->un_dirvp, &vp, &cn, vap)) != 0) 1022 return (error); 1023 1024 if ((error = VOP_OPEN(vp, fmode, cred, l)) != 0) { 1025 vput(vp); 1026 return (error); 1027 } 1028 1029 vp->v_writecount++; 1030 *vpp = vp; 1031 return (0); 1032 } 1033 1034 int 1035 union_vn_close(vp, fmode, cred, l) 1036 struct vnode *vp; 1037 int fmode; 1038 kauth_cred_t cred; 1039 struct lwp *l; 1040 { 1041 1042 if (fmode & FWRITE) 1043 --vp->v_writecount; 1044 return (VOP_CLOSE(vp, fmode, cred, l)); 1045 } 1046 1047 void 1048 union_removed_upper(un) 1049 struct union_node *un; 1050 { 1051 #if 1 1052 /* 1053 * We do not set the uppervp to NULLVP here, because lowervp 1054 * may also be NULLVP, so this routine would end up creating 1055 * a bogus union node with no upper or lower VP (that causes 1056 * pain in many places that assume at least one VP exists). 1057 * Since we've removed this node from the cache hash chains, 1058 * it won't be found again. When all current holders 1059 * release it, union_inactive() will vgone() it. 1060 */ 1061 union_diruncache(un); 1062 #else 1063 union_newupper(un, NULLVP); 1064 #endif 1065 1066 if (un->un_flags & UN_CACHED) { 1067 un->un_flags &= ~UN_CACHED; 1068 LIST_REMOVE(un, un_cache); 1069 } 1070 1071 if (un->un_flags & UN_ULOCK) { 1072 un->un_flags &= ~UN_ULOCK; 1073 VOP_UNLOCK(un->un_uppervp, 0); 1074 } 1075 } 1076 1077 #if 0 1078 struct vnode * 1079 union_lowervp(vp) 1080 struct vnode *vp; 1081 { 1082 struct union_node *un = VTOUNION(vp); 1083 1084 if ((un->un_lowervp != NULLVP) && 1085 (vp->v_type == un->un_lowervp->v_type)) { 1086 if (vget(un->un_lowervp, 0) == 0) 1087 return (un->un_lowervp); 1088 } 1089 1090 return (NULLVP); 1091 } 1092 #endif 1093 1094 /* 1095 * determine whether a whiteout is needed 1096 * during a remove/rmdir operation. 1097 */ 1098 int 1099 union_dowhiteout(un, cred, l) 1100 struct union_node *un; 1101 kauth_cred_t cred; 1102 struct lwp *l; 1103 { 1104 struct vattr va; 1105 1106 if (un->un_lowervp != NULLVP) 1107 return (1); 1108 1109 if (VOP_GETATTR(un->un_uppervp, &va, cred, l) == 0 && 1110 (va.va_flags & OPAQUE)) 1111 return (1); 1112 1113 return (0); 1114 } 1115 1116 static void 1117 union_dircache_r(vp, vppp, cntp) 1118 struct vnode *vp; 1119 struct vnode ***vppp; 1120 int *cntp; 1121 { 1122 struct union_node *un; 1123 1124 if (vp->v_op != union_vnodeop_p) { 1125 if (vppp) { 1126 VREF(vp); 1127 *(*vppp)++ = vp; 1128 if (--(*cntp) == 0) 1129 panic("union: dircache table too small"); 1130 } else { 1131 (*cntp)++; 1132 } 1133 1134 return; 1135 } 1136 1137 un = VTOUNION(vp); 1138 if (un->un_uppervp != NULLVP) 1139 union_dircache_r(un->un_uppervp, vppp, cntp); 1140 if (un->un_lowervp != NULLVP) 1141 union_dircache_r(un->un_lowervp, vppp, cntp); 1142 } 1143 1144 struct vnode * 1145 union_dircache(vp, l) 1146 struct vnode *vp; 1147 struct lwp *l; 1148 { 1149 int cnt; 1150 struct vnode *nvp = NULLVP; 1151 struct vnode **vpp; 1152 struct vnode **dircache; 1153 int error; 1154 1155 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1156 dircache = VTOUNION(vp)->un_dircache; 1157 1158 nvp = NULLVP; 1159 1160 if (dircache == 0) { 1161 cnt = 0; 1162 union_dircache_r(vp, 0, &cnt); 1163 cnt++; 1164 dircache = (struct vnode **) 1165 malloc(cnt * sizeof(struct vnode *), 1166 M_TEMP, M_WAITOK); 1167 vpp = dircache; 1168 union_dircache_r(vp, &vpp, &cnt); 1169 VTOUNION(vp)->un_dircache = dircache; 1170 *vpp = NULLVP; 1171 vpp = dircache + 1; 1172 } else { 1173 vpp = dircache; 1174 do { 1175 if (*vpp++ == VTOUNION(vp)->un_uppervp) 1176 break; 1177 } while (*vpp != NULLVP); 1178 } 1179 1180 if (*vpp == NULLVP) 1181 goto out; 1182 1183 vn_lock(*vpp, LK_EXCLUSIVE | LK_RETRY); 1184 VREF(*vpp); 1185 error = union_allocvp(&nvp, vp->v_mount, NULLVP, NULLVP, 0, *vpp, NULLVP, 0); 1186 if (!error) { 1187 VTOUNION(vp)->un_dircache = 0; 1188 VTOUNION(nvp)->un_dircache = dircache; 1189 } 1190 1191 out: 1192 VOP_UNLOCK(vp, 0); 1193 return (nvp); 1194 } 1195 1196 void 1197 union_diruncache(un) 1198 struct union_node *un; 1199 { 1200 struct vnode **vpp; 1201 1202 if (un->un_dircache != 0) { 1203 for (vpp = un->un_dircache; *vpp != NULLVP; vpp++) 1204 vrele(*vpp); 1205 free(un->un_dircache, M_TEMP); 1206 un->un_dircache = 0; 1207 } 1208 } 1209 1210 /* 1211 * This hook is called from vn_readdir() to switch to lower directory 1212 * entry after the upper directory is read. 1213 */ 1214 int 1215 union_readdirhook(struct vnode **vpp, struct file *fp, struct lwp *l) 1216 { 1217 struct vnode *vp = *vpp, *lvp; 1218 struct vattr va; 1219 int error; 1220 1221 if (vp->v_op != union_vnodeop_p) 1222 return (0); 1223 1224 if ((lvp = union_dircache(vp, l)) == NULLVP) 1225 return (0); 1226 1227 /* 1228 * If the directory is opaque, 1229 * then don't show lower entries 1230 */ 1231 error = VOP_GETATTR(vp, &va, fp->f_cred, l); 1232 if (error || (va.va_flags & OPAQUE)) { 1233 vput(lvp); 1234 return (error); 1235 } 1236 1237 error = VOP_OPEN(lvp, FREAD, fp->f_cred, l); 1238 if (error) { 1239 vput(lvp); 1240 return (error); 1241 } 1242 VOP_UNLOCK(lvp, 0); 1243 fp->f_data = lvp; 1244 fp->f_offset = 0; 1245 error = vn_close(vp, FREAD, fp->f_cred, l); 1246 if (error) 1247 return (error); 1248 *vpp = lvp; 1249 return (0); 1250 } 1251