1/* $OpenBSD: tables.c,v 1.25 2007/09/02 15:19:08 deraadt Exp $ */ 2/* $NetBSD: tables.c,v 1.4 1995/03/21 09:07:45 cgd Exp $ */ 3 4/*- 5 * Copyright (c) 1992 Keith Muller. 6 * Copyright (c) 1992, 1993 7 * The Regents of the University of California. All rights reserved. 8 * 9 * This code is derived from software contributed to Berkeley by 10 * Keith Muller of the University of California, San Diego. 11 * 12 * Redistribution and use in source and binary forms, with or without 13 * modification, are permitted provided that the following conditions 14 * are met: 15 * 1. Redistributions of source code must retain the above copyright 16 * notice, this list of conditions and the following disclaimer. 17 * 2. Redistributions in binary form must reproduce the above copyright 18 * notice, this list of conditions and the following disclaimer in the 19 * documentation and/or other materials provided with the distribution. 20 * 3. Neither the name of the University nor the names of its contributors 21 * may be used to endorse or promote products derived from this software 22 * without specific prior written permission. 23 * 24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 34 * SUCH DAMAGE. 35 */ 36 37#ifndef lint 38#if 0 39static const char sccsid[] = "@(#)tables.c 8.1 (Berkeley) 5/31/93"; 40#else 41static const char rcsid[] = "$OpenBSD: tables.c,v 1.25 2007/09/02 15:19:08 deraadt Exp $"; 42#endif 43#endif /* not lint */ 44 45#include <sys/types.h> 46#include <sys/time.h> 47#include <sys/stat.h> 48#include <sys/param.h> 49#include <sys/fcntl.h> 50#include <stdio.h> 51#include <string.h> 52#include <unistd.h> 53#include <errno.h> 54#include <stdlib.h> 55#include "pax.h" 56#include "tables.h" 57#include "extern.h" 58 59/* 60 * Routines for controlling the contents of all the different databases pax 61 * keeps. Tables are dynamically created only when they are needed. The 62 * goal was speed and the ability to work with HUGE archives. The databases 63 * were kept simple, but do have complex rules for when the contents change. 64 * As of this writing, the posix library functions were more complex than 65 * needed for this application (pax databases have very short lifetimes and 66 * do not survive after pax is finished). Pax is required to handle very 67 * large archives. These database routines carefully combine memory usage and 68 * temporary file storage in ways which will not significantly impact runtime 69 * performance while allowing the largest possible archives to be handled. 70 * Trying to force the fit to the posix database routines was not considered 71 * time well spent. 72 */ 73 74static HRDLNK **ltab = NULL; /* hard link table for detecting hard links */ 75static FTM **ftab = NULL; /* file time table for updating arch */ 76static NAMT **ntab = NULL; /* interactive rename storage table */ 77static DEVT **dtab = NULL; /* device/inode mapping tables */ 78static ATDIR **atab = NULL; /* file tree directory time reset table */ 79static DIRDATA *dirp = NULL; /* storage for setting created dir time/mode */ 80static size_t dirsize; /* size of dirp table */ 81static long dircnt = 0; /* entries in dir time/mode storage */ 82static int ffd = -1; /* tmp file for file time table name storage */ 83 84static DEVT *chk_dev(dev_t, int); 85 86/* 87 * hard link table routines 88 * 89 * The hard link table tries to detect hard links to files using the device and 90 * inode values. We do this when writing an archive, so we can tell the format 91 * write routine that this file is a hard link to another file. The format 92 * write routine then can store this file in whatever way it wants (as a hard 93 * link if the format supports that like tar, or ignore this info like cpio). 94 * (Actually a field in the format driver table tells us if the format wants 95 * hard link info. if not, we do not waste time looking for them). We also use 96 * the same table when reading an archive. In that situation, this table is 97 * used by the format read routine to detect hard links from stored dev and 98 * inode numbers (like cpio). This will allow pax to create a link when one 99 * can be detected by the archive format. 100 */ 101 102/* 103 * lnk_start 104 * Creates the hard link table. 105 * Return: 106 * 0 if created, -1 if failure 107 */ 108 109int 110lnk_start(void) 111{ 112 if (ltab != NULL) 113 return(0); 114 if ((ltab = (HRDLNK **)calloc(L_TAB_SZ, sizeof(HRDLNK *))) == NULL) { 115 paxwarn(1, "Cannot allocate memory for hard link table"); 116 return(-1); 117 } 118 return(0); 119} 120 121/* 122 * chk_lnk() 123 * Looks up entry in hard link hash table. If found, it copies the name 124 * of the file it is linked to (we already saw that file) into ln_name. 125 * lnkcnt is decremented and if goes to 1 the node is deleted from the 126 * database. (We have seen all the links to this file). If not found, 127 * we add the file to the database if it has the potential for having 128 * hard links to other files we may process (it has a link count > 1) 129 * Return: 130 * if found returns 1; if not found returns 0; -1 on error 131 */ 132 133int 134chk_lnk(ARCHD *arcn) 135{ 136 HRDLNK *pt; 137 HRDLNK **ppt; 138 u_int indx; 139 140 if (ltab == NULL) 141 return(-1); 142 /* 143 * ignore those nodes that cannot have hard links 144 */ 145 if ((arcn->type == PAX_DIR) || (arcn->sb.st_nlink <= 1)) 146 return(0); 147 148 /* 149 * hash inode number and look for this file 150 */ 151 indx = ((unsigned)arcn->sb.st_ino) % L_TAB_SZ; 152 if ((pt = ltab[indx]) != NULL) { 153 /* 154 * its hash chain in not empty, walk down looking for it 155 */ 156 ppt = &(ltab[indx]); 157 while (pt != NULL) { 158 if ((pt->ino == arcn->sb.st_ino) && 159 (pt->dev == arcn->sb.st_dev)) 160 break; 161 ppt = &(pt->fow); 162 pt = pt->fow; 163 } 164 165 if (pt != NULL) { 166 /* 167 * found a link. set the node type and copy in the 168 * name of the file it is to link to. we need to 169 * handle hardlinks to regular files differently than 170 * other links. 171 */ 172 arcn->ln_nlen = strlcpy(arcn->ln_name, pt->name, 173 sizeof(arcn->ln_name)); 174 /* XXX truncate? */ 175 if (arcn->nlen >= sizeof(arcn->name)) 176 arcn->nlen = sizeof(arcn->name) - 1; 177 if (arcn->type == PAX_REG) 178 arcn->type = PAX_HRG; 179 else 180 arcn->type = PAX_HLK; 181 182 /* 183 * if we have found all the links to this file, remove 184 * it from the database 185 */ 186 if (--pt->nlink <= 1) { 187 *ppt = pt->fow; 188 (void)free((char *)pt->name); 189 (void)free((char *)pt); 190 } 191 return(1); 192 } 193 } 194 195 /* 196 * we never saw this file before. It has links so we add it to the 197 * front of this hash chain 198 */ 199 if ((pt = (HRDLNK *)malloc(sizeof(HRDLNK))) != NULL) { 200 if ((pt->name = strdup(arcn->name)) != NULL) { 201 pt->dev = arcn->sb.st_dev; 202 pt->ino = arcn->sb.st_ino; 203 pt->nlink = arcn->sb.st_nlink; 204 pt->fow = ltab[indx]; 205 ltab[indx] = pt; 206 return(0); 207 } 208 (void)free((char *)pt); 209 } 210 211 paxwarn(1, "Hard link table out of memory"); 212 return(-1); 213} 214 215/* 216 * purg_lnk 217 * remove reference for a file that we may have added to the data base as 218 * a potential source for hard links. We ended up not using the file, so 219 * we do not want to accidently point another file at it later on. 220 */ 221 222void 223purg_lnk(ARCHD *arcn) 224{ 225 HRDLNK *pt; 226 HRDLNK **ppt; 227 u_int indx; 228 229 if (ltab == NULL) 230 return; 231 /* 232 * do not bother to look if it could not be in the database 233 */ 234 if ((arcn->sb.st_nlink <= 1) || (arcn->type == PAX_DIR) || 235 (arcn->type == PAX_HLK) || (arcn->type == PAX_HRG)) 236 return; 237 238 /* 239 * find the hash chain for this inode value, if empty return 240 */ 241 indx = ((unsigned)arcn->sb.st_ino) % L_TAB_SZ; 242 if ((pt = ltab[indx]) == NULL) 243 return; 244 245 /* 246 * walk down the list looking for the inode/dev pair, unlink and 247 * free if found 248 */ 249 ppt = &(ltab[indx]); 250 while (pt != NULL) { 251 if ((pt->ino == arcn->sb.st_ino) && 252 (pt->dev == arcn->sb.st_dev)) 253 break; 254 ppt = &(pt->fow); 255 pt = pt->fow; 256 } 257 if (pt == NULL) 258 return; 259 260 /* 261 * remove and free it 262 */ 263 *ppt = pt->fow; 264 (void)free((char *)pt->name); 265 (void)free((char *)pt); 266} 267 268/* 269 * lnk_end() 270 * pull apart a existing link table so we can reuse it. We do this between 271 * read and write phases of append with update. (The format may have 272 * used the link table, and we need to start with a fresh table for the 273 * write phase 274 */ 275 276void 277lnk_end(void) 278{ 279 int i; 280 HRDLNK *pt; 281 HRDLNK *ppt; 282 283 if (ltab == NULL) 284 return; 285 286 for (i = 0; i < L_TAB_SZ; ++i) { 287 if (ltab[i] == NULL) 288 continue; 289 pt = ltab[i]; 290 ltab[i] = NULL; 291 292 /* 293 * free up each entry on this chain 294 */ 295 while (pt != NULL) { 296 ppt = pt; 297 pt = ppt->fow; 298 (void)free((char *)ppt->name); 299 (void)free((char *)ppt); 300 } 301 } 302 return; 303} 304 305/* 306 * modification time table routines 307 * 308 * The modification time table keeps track of last modification times for all 309 * files stored in an archive during a write phase when -u is set. We only 310 * add a file to the archive if it is newer than a file with the same name 311 * already stored on the archive (if there is no other file with the same 312 * name on the archive it is added). This applies to writes and appends. 313 * An append with an -u must read the archive and store the modification time 314 * for every file on that archive before starting the write phase. It is clear 315 * that this is one HUGE database. To save memory space, the actual file names 316 * are stored in a scratch file and indexed by an in-memory hash table. The 317 * hash table is indexed by hashing the file path. The nodes in the table store 318 * the length of the filename and the lseek offset within the scratch file 319 * where the actual name is stored. Since there are never any deletions from 320 * this table, fragmentation of the scratch file is never a issue. Lookups 321 * seem to not exhibit any locality at all (files in the database are rarely 322 * looked up more than once...), so caching is just a waste of memory. The 323 * only limitation is the amount of scratch file space available to store the 324 * path names. 325 */ 326 327/* 328 * ftime_start() 329 * create the file time hash table and open for read/write the scratch 330 * file. (after created it is unlinked, so when we exit we leave 331 * no witnesses). 332 * Return: 333 * 0 if the table and file was created ok, -1 otherwise 334 */ 335 336int 337ftime_start(void) 338{ 339 340 if (ftab != NULL) 341 return(0); 342 if ((ftab = (FTM **)calloc(F_TAB_SZ, sizeof(FTM *))) == NULL) { 343 paxwarn(1, "Cannot allocate memory for file time table"); 344 return(-1); 345 } 346 347 /* 348 * get random name and create temporary scratch file, unlink name 349 * so it will get removed on exit 350 */ 351 memcpy(tempbase, _TFILE_BASE, sizeof(_TFILE_BASE)); 352 if ((ffd = mkstemp(tempfile)) < 0) { 353 syswarn(1, errno, "Unable to create temporary file: %s", 354 tempfile); 355 return(-1); 356 } 357 (void)unlink(tempfile); 358 359 return(0); 360} 361 362/* 363 * chk_ftime() 364 * looks up entry in file time hash table. If not found, the file is 365 * added to the hash table and the file named stored in the scratch file. 366 * If a file with the same name is found, the file times are compared and 367 * the most recent file time is retained. If the new file was younger (or 368 * was not in the database) the new file is selected for storage. 369 * Return: 370 * 0 if file should be added to the archive, 1 if it should be skipped, 371 * -1 on error 372 */ 373 374int 375chk_ftime(ARCHD *arcn) 376{ 377 FTM *pt; 378 int namelen; 379 u_int indx; 380 char ckname[PAXPATHLEN+1]; 381 382 /* 383 * no info, go ahead and add to archive 384 */ 385 if (ftab == NULL) 386 return(0); 387 388 /* 389 * hash the pathname and look up in table 390 */ 391 namelen = arcn->nlen; 392 indx = st_hash(arcn->name, namelen, F_TAB_SZ); 393 if ((pt = ftab[indx]) != NULL) { 394 /* 395 * the hash chain is not empty, walk down looking for match 396 * only read up the path names if the lengths match, speeds 397 * up the search a lot 398 */ 399 while (pt != NULL) { 400 if (pt->namelen == namelen) { 401 /* 402 * potential match, have to read the name 403 * from the scratch file. 404 */ 405 if (lseek(ffd,pt->seek,SEEK_SET) != pt->seek) { 406 syswarn(1, errno, 407 "Failed ftime table seek"); 408 return(-1); 409 } 410 if (read(ffd, ckname, namelen) != namelen) { 411 syswarn(1, errno, 412 "Failed ftime table read"); 413 return(-1); 414 } 415 416 /* 417 * if the names match, we are done 418 */ 419 if (!strncmp(ckname, arcn->name, namelen)) 420 break; 421 } 422 423 /* 424 * try the next entry on the chain 425 */ 426 pt = pt->fow; 427 } 428 429 if (pt != NULL) { 430 /* 431 * found the file, compare the times, save the newer 432 */ 433 if (arcn->sb.st_mtime > pt->mtime) { 434 /* 435 * file is newer 436 */ 437 pt->mtime = arcn->sb.st_mtime; 438 return(0); 439 } 440 /* 441 * file is older 442 */ 443 return(1); 444 } 445 } 446 447 /* 448 * not in table, add it 449 */ 450 if ((pt = (FTM *)malloc(sizeof(FTM))) != NULL) { 451 /* 452 * add the name at the end of the scratch file, saving the 453 * offset. add the file to the head of the hash chain 454 */ 455 if ((pt->seek = lseek(ffd, (off_t)0, SEEK_END)) >= 0) { 456 if (write(ffd, arcn->name, namelen) == namelen) { 457 pt->mtime = arcn->sb.st_mtime; 458 pt->namelen = namelen; 459 pt->fow = ftab[indx]; 460 ftab[indx] = pt; 461 return(0); 462 } 463 syswarn(1, errno, "Failed write to file time table"); 464 } else 465 syswarn(1, errno, "Failed seek on file time table"); 466 } else 467 paxwarn(1, "File time table ran out of memory"); 468 469 if (pt != NULL) 470 (void)free((char *)pt); 471 return(-1); 472} 473 474/* 475 * Interactive rename table routines 476 * 477 * The interactive rename table keeps track of the new names that the user 478 * assigns to files from tty input. Since this map is unique for each file 479 * we must store it in case there is a reference to the file later in archive 480 * (a link). Otherwise we will be unable to find the file we know was 481 * extracted. The remapping of these files is stored in a memory based hash 482 * table (it is assumed since input must come from /dev/tty, it is unlikely to 483 * be a very large table). 484 */ 485 486/* 487 * name_start() 488 * create the interactive rename table 489 * Return: 490 * 0 if successful, -1 otherwise 491 */ 492 493int 494name_start(void) 495{ 496 if (ntab != NULL) 497 return(0); 498 if ((ntab = (NAMT **)calloc(N_TAB_SZ, sizeof(NAMT *))) == NULL) { 499 paxwarn(1, "Cannot allocate memory for interactive rename table"); 500 return(-1); 501 } 502 return(0); 503} 504 505/* 506 * add_name() 507 * add the new name to old name mapping just created by the user. 508 * If an old name mapping is found (there may be duplicate names on an 509 * archive) only the most recent is kept. 510 * Return: 511 * 0 if added, -1 otherwise 512 */ 513 514int 515add_name(char *oname, int onamelen, char *nname) 516{ 517 NAMT *pt; 518 u_int indx; 519 520 if (ntab == NULL) { 521 /* 522 * should never happen 523 */ 524 paxwarn(0, "No interactive rename table, links may fail"); 525 return(0); 526 } 527 528 /* 529 * look to see if we have already mapped this file, if so we 530 * will update it 531 */ 532 indx = st_hash(oname, onamelen, N_TAB_SZ); 533 if ((pt = ntab[indx]) != NULL) { 534 /* 535 * look down the has chain for the file 536 */ 537 while ((pt != NULL) && (strcmp(oname, pt->oname) != 0)) 538 pt = pt->fow; 539 540 if (pt != NULL) { 541 /* 542 * found an old mapping, replace it with the new one 543 * the user just input (if it is different) 544 */ 545 if (strcmp(nname, pt->nname) == 0) 546 return(0); 547 548 (void)free((char *)pt->nname); 549 if ((pt->nname = strdup(nname)) == NULL) { 550 paxwarn(1, "Cannot update rename table"); 551 return(-1); 552 } 553 return(0); 554 } 555 } 556 557 /* 558 * this is a new mapping, add it to the table 559 */ 560 if ((pt = (NAMT *)malloc(sizeof(NAMT))) != NULL) { 561 if ((pt->oname = strdup(oname)) != NULL) { 562 if ((pt->nname = strdup(nname)) != NULL) { 563 pt->fow = ntab[indx]; 564 ntab[indx] = pt; 565 return(0); 566 } 567 (void)free((char *)pt->oname); 568 } 569 (void)free((char *)pt); 570 } 571 paxwarn(1, "Interactive rename table out of memory"); 572 return(-1); 573} 574 575/* 576 * sub_name() 577 * look up a link name to see if it points at a file that has been 578 * remapped by the user. If found, the link is adjusted to contain the 579 * new name (oname is the link to name) 580 */ 581 582void 583sub_name(char *oname, int *onamelen, size_t onamesize) 584{ 585 NAMT *pt; 586 u_int indx; 587 588 if (ntab == NULL) 589 return; 590 /* 591 * look the name up in the hash table 592 */ 593 indx = st_hash(oname, *onamelen, N_TAB_SZ); 594 if ((pt = ntab[indx]) == NULL) 595 return; 596 597 while (pt != NULL) { 598 /* 599 * walk down the hash chain looking for a match 600 */ 601 if (strcmp(oname, pt->oname) == 0) { 602 /* 603 * found it, replace it with the new name 604 * and return (we know that oname has enough space) 605 */ 606 *onamelen = strlcpy(oname, pt->nname, onamesize); 607 if (*onamelen >= onamesize) 608 *onamelen = onamesize - 1; /* XXX truncate? */ 609 return; 610 } 611 pt = pt->fow; 612 } 613 614 /* 615 * no match, just return 616 */ 617 return; 618} 619 620/* 621 * device/inode mapping table routines 622 * (used with formats that store device and inodes fields) 623 * 624 * device/inode mapping tables remap the device field in a archive header. The 625 * device/inode fields are used to determine when files are hard links to each 626 * other. However these values have very little meaning outside of that. This 627 * database is used to solve one of two different problems. 628 * 629 * 1) when files are appended to an archive, while the new files may have hard 630 * links to each other, you cannot determine if they have hard links to any 631 * file already stored on the archive from a prior run of pax. We must assume 632 * that these inode/device pairs are unique only within a SINGLE run of pax 633 * (which adds a set of files to an archive). So we have to make sure the 634 * inode/dev pairs we add each time are always unique. We do this by observing 635 * while the inode field is very dense, the use of the dev field is fairly 636 * sparse. Within each run of pax, we remap any device number of a new archive 637 * member that has a device number used in a prior run and already stored in a 638 * file on the archive. During the read phase of the append, we store the 639 * device numbers used and mark them to not be used by any file during the 640 * write phase. If during write we go to use one of those old device numbers, 641 * we remap it to a new value. 642 * 643 * 2) Often the fields in the archive header used to store these values are 644 * too small to store the entire value. The result is an inode or device value 645 * which can be truncated. This really can foul up an archive. With truncation 646 * we end up creating links between files that are really not links (after 647 * truncation the inodes are the same value). We address that by detecting 648 * truncation and forcing a remap of the device field to split truncated 649 * inodes away from each other. Each truncation creates a pattern of bits that 650 * are removed. We use this pattern of truncated bits to partition the inodes 651 * on a single device to many different devices (each one represented by the 652 * truncated bit pattern). All inodes on the same device that have the same 653 * truncation pattern are mapped to the same new device. Two inodes that 654 * truncate to the same value clearly will always have different truncation 655 * bit patterns, so they will be split from away each other. When we spot 656 * device truncation we remap the device number to a non truncated value. 657 * (for more info see table.h for the data structures involved). 658 */ 659 660/* 661 * dev_start() 662 * create the device mapping table 663 * Return: 664 * 0 if successful, -1 otherwise 665 */ 666 667int 668dev_start(void) 669{ 670 if (dtab != NULL) 671 return(0); 672 if ((dtab = (DEVT **)calloc(D_TAB_SZ, sizeof(DEVT *))) == NULL) { 673 paxwarn(1, "Cannot allocate memory for device mapping table"); 674 return(-1); 675 } 676 return(0); 677} 678 679/* 680 * add_dev() 681 * add a device number to the table. this will force the device to be 682 * remapped to a new value if it be used during a write phase. This 683 * function is called during the read phase of an append to prohibit the 684 * use of any device number already in the archive. 685 * Return: 686 * 0 if added ok, -1 otherwise 687 */ 688 689int 690add_dev(ARCHD *arcn) 691{ 692 if (chk_dev(arcn->sb.st_dev, 1) == NULL) 693 return(-1); 694 return(0); 695} 696 697/* 698 * chk_dev() 699 * check for a device value in the device table. If not found and the add 700 * flag is set, it is added. This does NOT assign any mapping values, just 701 * adds the device number as one that need to be remapped. If this device 702 * is already mapped, just return with a pointer to that entry. 703 * Return: 704 * pointer to the entry for this device in the device map table. Null 705 * if the add flag is not set and the device is not in the table (it is 706 * not been seen yet). If add is set and the device cannot be added, null 707 * is returned (indicates an error). 708 */ 709 710static DEVT * 711chk_dev(dev_t dev, int add) 712{ 713 DEVT *pt; 714 u_int indx; 715 716 if (dtab == NULL) 717 return(NULL); 718 /* 719 * look to see if this device is already in the table 720 */ 721 indx = ((unsigned)dev) % D_TAB_SZ; 722 if ((pt = dtab[indx]) != NULL) { 723 while ((pt != NULL) && (pt->dev != dev)) 724 pt = pt->fow; 725 726 /* 727 * found it, return a pointer to it 728 */ 729 if (pt != NULL) 730 return(pt); 731 } 732 733 /* 734 * not in table, we add it only if told to as this may just be a check 735 * to see if a device number is being used. 736 */ 737 if (add == 0) 738 return(NULL); 739 740 /* 741 * allocate a node for this device and add it to the front of the hash 742 * chain. Note we do not assign remaps values here, so the pt->list 743 * list must be NULL. 744 */ 745 if ((pt = (DEVT *)malloc(sizeof(DEVT))) == NULL) { 746 paxwarn(1, "Device map table out of memory"); 747 return(NULL); 748 } 749 pt->dev = dev; 750 pt->list = NULL; 751 pt->fow = dtab[indx]; 752 dtab[indx] = pt; 753 return(pt); 754} 755/* 756 * map_dev() 757 * given an inode and device storage mask (the mask has a 1 for each bit 758 * the archive format is able to store in a header), we check for inode 759 * and device truncation and remap the device as required. Device mapping 760 * can also occur when during the read phase of append a device number was 761 * seen (and was marked as do not use during the write phase). WE ASSUME 762 * that unsigned longs are the same size or bigger than the fields used 763 * for ino_t and dev_t. If not the types will have to be changed. 764 * Return: 765 * 0 if all ok, -1 otherwise. 766 */ 767 768int 769map_dev(ARCHD *arcn, u_long dev_mask, u_long ino_mask) 770{ 771 DEVT *pt; 772 DLIST *dpt; 773 static dev_t lastdev = 0; /* next device number to try */ 774 int trc_ino = 0; 775 int trc_dev = 0; 776 ino_t trunc_bits = 0; 777 ino_t nino; 778 779 if (dtab == NULL) 780 return(0); 781 /* 782 * check for device and inode truncation, and extract the truncated 783 * bit pattern. 784 */ 785 if ((arcn->sb.st_dev & (dev_t)dev_mask) != arcn->sb.st_dev) 786 ++trc_dev; 787 if ((nino = arcn->sb.st_ino & (ino_t)ino_mask) != arcn->sb.st_ino) { 788 ++trc_ino; 789 trunc_bits = arcn->sb.st_ino & (ino_t)(~ino_mask); 790 } 791 792 /* 793 * see if this device is already being mapped, look up the device 794 * then find the truncation bit pattern which applies 795 */ 796 if ((pt = chk_dev(arcn->sb.st_dev, 0)) != NULL) { 797 /* 798 * this device is already marked to be remapped 799 */ 800 for (dpt = pt->list; dpt != NULL; dpt = dpt->fow) 801 if (dpt->trunc_bits == trunc_bits) 802 break; 803 804 if (dpt != NULL) { 805 /* 806 * we are being remapped for this device and pattern 807 * change the device number to be stored and return 808 */ 809 arcn->sb.st_dev = dpt->dev; 810 arcn->sb.st_ino = nino; 811 return(0); 812 } 813 } else { 814 /* 815 * this device is not being remapped YET. if we do not have any 816 * form of truncation, we do not need a remap 817 */ 818 if (!trc_ino && !trc_dev) 819 return(0); 820 821 /* 822 * we have truncation, have to add this as a device to remap 823 */ 824 if ((pt = chk_dev(arcn->sb.st_dev, 1)) == NULL) 825 goto bad; 826 827 /* 828 * if we just have a truncated inode, we have to make sure that 829 * all future inodes that do not truncate (they have the 830 * truncation pattern of all 0's) continue to map to the same 831 * device number. We probably have already written inodes with 832 * this device number to the archive with the truncation 833 * pattern of all 0's. So we add the mapping for all 0's to the 834 * same device number. 835 */ 836 if (!trc_dev && (trunc_bits != 0)) { 837 if ((dpt = (DLIST *)malloc(sizeof(DLIST))) == NULL) 838 goto bad; 839 dpt->trunc_bits = 0; 840 dpt->dev = arcn->sb.st_dev; 841 dpt->fow = pt->list; 842 pt->list = dpt; 843 } 844 } 845 846 /* 847 * look for a device number not being used. We must watch for wrap 848 * around on lastdev (so we do not get stuck looking forever!) 849 */ 850 while (++lastdev > 0) { 851 if (chk_dev(lastdev, 0) != NULL) 852 continue; 853 /* 854 * found an unused value. If we have reached truncation point 855 * for this format we are hosed, so we give up. Otherwise we 856 * mark it as being used. 857 */ 858 if (((lastdev & ((dev_t)dev_mask)) != lastdev) || 859 (chk_dev(lastdev, 1) == NULL)) 860 goto bad; 861 break; 862 } 863 864 if ((lastdev <= 0) || ((dpt = (DLIST *)malloc(sizeof(DLIST))) == NULL)) 865 goto bad; 866 867 /* 868 * got a new device number, store it under this truncation pattern. 869 * change the device number this file is being stored with. 870 */ 871 dpt->trunc_bits = trunc_bits; 872 dpt->dev = lastdev; 873 dpt->fow = pt->list; 874 pt->list = dpt; 875 arcn->sb.st_dev = lastdev; 876 arcn->sb.st_ino = nino; 877 return(0); 878 879 bad: 880 paxwarn(1, "Unable to fix truncated inode/device field when storing %s", 881 arcn->name); 882 paxwarn(0, "Archive may create improper hard links when extracted"); 883 return(0); 884} 885 886/* 887 * directory access/mod time reset table routines (for directories READ by pax) 888 * 889 * The pax -t flag requires that access times of archive files be the same 890 * before being read by pax. For regular files, access time is restored after 891 * the file has been copied. This database provides the same functionality for 892 * directories read during file tree traversal. Restoring directory access time 893 * is more complex than files since directories may be read several times until 894 * all the descendants in their subtree are visited by fts. Directory access 895 * and modification times are stored during the fts pre-order visit (done 896 * before any descendants in the subtree are visited) and restored after the 897 * fts post-order visit (after all the descendants have been visited). In the 898 * case of premature exit from a subtree (like from the effects of -n), any 899 * directory entries left in this database are reset during final cleanup 900 * operations of pax. Entries are hashed by inode number for fast lookup. 901 */ 902 903/* 904 * atdir_start() 905 * create the directory access time database for directories READ by pax. 906 * Return: 907 * 0 is created ok, -1 otherwise. 908 */ 909 910int 911atdir_start(void) 912{ 913 if (atab != NULL) 914 return(0); 915 if ((atab = (ATDIR **)calloc(A_TAB_SZ, sizeof(ATDIR *))) == NULL) { 916 paxwarn(1,"Cannot allocate space for directory access time table"); 917 return(-1); 918 } 919 return(0); 920} 921 922 923/* 924 * atdir_end() 925 * walk through the directory access time table and reset the access time 926 * of any directory who still has an entry left in the database. These 927 * entries are for directories READ by pax 928 */ 929 930void 931atdir_end(void) 932{ 933 ATDIR *pt; 934 int i; 935 936 if (atab == NULL) 937 return; 938 /* 939 * for each non-empty hash table entry reset all the directories 940 * chained there. 941 */ 942 for (i = 0; i < A_TAB_SZ; ++i) { 943 if ((pt = atab[i]) == NULL) 944 continue; 945 /* 946 * remember to force the times, set_ftime() looks at pmtime 947 * and patime, which only applies to things CREATED by pax, 948 * not read by pax. Read time reset is controlled by -t. 949 */ 950 for (; pt != NULL; pt = pt->fow) 951 set_ftime(pt->name, pt->mtime, pt->atime, 1); 952 } 953} 954 955/* 956 * add_atdir() 957 * add a directory to the directory access time table. Table is hashed 958 * and chained by inode number. This is for directories READ by pax 959 */ 960 961void 962add_atdir(char *fname, dev_t dev, ino_t ino, time_t mtime, time_t atime) 963{ 964 ATDIR *pt; 965 u_int indx; 966 967 if (atab == NULL) 968 return; 969 970 /* 971 * make sure this directory is not already in the table, if so just 972 * return (the older entry always has the correct time). The only 973 * way this will happen is when the same subtree can be traversed by 974 * different args to pax and the -n option is aborting fts out of a 975 * subtree before all the post-order visits have been made. 976 */ 977 indx = ((unsigned)ino) % A_TAB_SZ; 978 if ((pt = atab[indx]) != NULL) { 979 while (pt != NULL) { 980 if ((pt->ino == ino) && (pt->dev == dev)) 981 break; 982 pt = pt->fow; 983 } 984 985 /* 986 * oops, already there. Leave it alone. 987 */ 988 if (pt != NULL) 989 return; 990 } 991 992 /* 993 * add it to the front of the hash chain 994 */ 995 if ((pt = (ATDIR *)malloc(sizeof(ATDIR))) != NULL) { 996 if ((pt->name = strdup(fname)) != NULL) { 997 pt->dev = dev; 998 pt->ino = ino; 999 pt->mtime = mtime; 1000 pt->atime = atime; 1001 pt->fow = atab[indx]; 1002 atab[indx] = pt; 1003 return; 1004 } 1005 (void)free((char *)pt); 1006 } 1007 1008 paxwarn(1, "Directory access time reset table ran out of memory"); 1009 return; 1010} 1011 1012/* 1013 * get_atdir() 1014 * look up a directory by inode and device number to obtain the access 1015 * and modification time you want to set to. If found, the modification 1016 * and access time parameters are set and the entry is removed from the 1017 * table (as it is no longer needed). These are for directories READ by 1018 * pax 1019 * Return: 1020 * 0 if found, -1 if not found. 1021 */ 1022 1023int 1024get_atdir(dev_t dev, ino_t ino, time_t *mtime, time_t *atime) 1025{ 1026 ATDIR *pt; 1027 ATDIR **ppt; 1028 u_int indx; 1029 1030 if (atab == NULL) 1031 return(-1); 1032 /* 1033 * hash by inode and search the chain for an inode and device match 1034 */ 1035 indx = ((unsigned)ino) % A_TAB_SZ; 1036 if ((pt = atab[indx]) == NULL) 1037 return(-1); 1038 1039 ppt = &(atab[indx]); 1040 while (pt != NULL) { 1041 if ((pt->ino == ino) && (pt->dev == dev)) 1042 break; 1043 /* 1044 * no match, go to next one 1045 */ 1046 ppt = &(pt->fow); 1047 pt = pt->fow; 1048 } 1049 1050 /* 1051 * return if we did not find it. 1052 */ 1053 if (pt == NULL) 1054 return(-1); 1055 1056 /* 1057 * found it. return the times and remove the entry from the table. 1058 */ 1059 *ppt = pt->fow; 1060 *mtime = pt->mtime; 1061 *atime = pt->atime; 1062 (void)free((char *)pt->name); 1063 (void)free((char *)pt); 1064 return(0); 1065} 1066 1067/* 1068 * directory access mode and time storage routines (for directories CREATED 1069 * by pax). 1070 * 1071 * Pax requires that extracted directories, by default, have their access/mod 1072 * times and permissions set to the values specified in the archive. During the 1073 * actions of extracting (and creating the destination subtree during -rw copy) 1074 * directories extracted may be modified after being created. Even worse is 1075 * that these directories may have been created with file permissions which 1076 * prohibits any descendants of these directories from being extracted. When 1077 * directories are created by pax, access rights may be added to permit the 1078 * creation of files in their subtree. Every time pax creates a directory, the 1079 * times and file permissions specified by the archive are stored. After all 1080 * files have been extracted (or copied), these directories have their times 1081 * and file modes reset to the stored values. The directory info is restored in 1082 * reverse order as entries were added to the data file from root to leaf. To 1083 * restore atime properly, we must go backwards. The data file consists of 1084 * records with two parts, the file name followed by a DIRDATA trailer. The 1085 * fixed sized trailer contains the size of the name plus the off_t location in 1086 * the file. To restore we work backwards through the file reading the trailer 1087 * then the file name. 1088 */ 1089 1090/* 1091 * dir_start() 1092 * set up the directory time and file mode storage for directories CREATED 1093 * by pax. 1094 * Return: 1095 * 0 if ok, -1 otherwise 1096 */ 1097 1098int 1099dir_start(void) 1100{ 1101 if (dirp != NULL) 1102 return(0); 1103 1104 dirsize = DIRP_SIZE; 1105 if ((dirp = calloc(dirsize, sizeof(DIRDATA))) == NULL) { 1106 paxwarn(1, "Unable to allocate memory for directory times"); 1107 return(-1); 1108 } 1109 return(0); 1110} 1111 1112/* 1113 * add_dir() 1114 * add the mode and times for a newly CREATED directory 1115 * name is name of the directory, psb the stat buffer with the data in it, 1116 * frc_mode is a flag that says whether to force the setting of the mode 1117 * (ignoring the user set values for preserving file mode). Frc_mode is 1118 * for the case where we created a file and found that the resulting 1119 * directory was not writeable and the user asked for file modes to NOT 1120 * be preserved. (we have to preserve what was created by default, so we 1121 * have to force the setting at the end. this is stated explicitly in the 1122 * pax spec) 1123 */ 1124 1125void 1126add_dir(char *name, size_t nlen, struct stat *psb, int frc_mode) 1127{ 1128 DIRDATA *dblk; 1129 char realname[MAXPATHLEN], *rp; 1130 1131 if (dirp == NULL) 1132 return; 1133 1134 if (havechd && *name != '/') { 1135 if ((rp = realpath(name, realname)) == NULL) { 1136 paxwarn(1, "Cannot canonicalize %s", name); 1137 return; 1138 } 1139 name = rp; 1140 } 1141 if (dircnt == dirsize) { 1142 dblk = realloc(dirp, 2 * dirsize * sizeof(DIRDATA)); 1143 if (dblk == NULL) { 1144 paxwarn(1, "Unable to store mode and times for created" 1145 " directory: %s", name); 1146 return; 1147 } 1148 dirp = dblk; 1149 dirsize *= 2; 1150 } 1151 dblk = &dirp[dircnt]; 1152 if ((dblk->name = strdup(name)) == NULL) { 1153 paxwarn(1, "Unable to store mode and times for created" 1154 " directory: %s", name); 1155 return; 1156 } 1157 dblk->mode = psb->st_mode & 0xffff; 1158 dblk->mtime = psb->st_mtime; 1159 dblk->atime = psb->st_atime; 1160 dblk->frc_mode = frc_mode; 1161 ++dircnt; 1162} 1163 1164/* 1165 * proc_dir() 1166 * process all file modes and times stored for directories CREATED 1167 * by pax 1168 */ 1169 1170void 1171proc_dir(void) 1172{ 1173 DIRDATA *dblk; 1174 long cnt; 1175 1176 if (dirp == NULL) 1177 return; 1178 /* 1179 * read backwards through the file and process each directory 1180 */ 1181 cnt = dircnt; 1182 while (--cnt >= 0) { 1183 /* 1184 * frc_mode set, make sure we set the file modes even if 1185 * the user didn't ask for it (see file_subs.c for more info) 1186 */ 1187 dblk = &dirp[cnt]; 1188 if (pmode || dblk->frc_mode) 1189 set_pmode(dblk->name, dblk->mode); 1190 if (patime || pmtime) 1191 set_ftime(dblk->name, dblk->mtime, dblk->atime, 0); 1192 free(dblk->name); 1193 } 1194 1195 free(dirp); 1196 dirp = NULL; 1197 dircnt = 0; 1198} 1199 1200/* 1201 * database independent routines 1202 */ 1203 1204/* 1205 * st_hash() 1206 * hashes filenames to a u_int for hashing into a table. Looks at the tail 1207 * end of file, as this provides far better distribution than any other 1208 * part of the name. For performance reasons we only care about the last 1209 * MAXKEYLEN chars (should be at LEAST large enough to pick off the file 1210 * name). Was tested on 500,000 name file tree traversal from the root 1211 * and gave almost a perfectly uniform distribution of keys when used with 1212 * prime sized tables (MAXKEYLEN was 128 in test). Hashes (sizeof int) 1213 * chars at a time and pads with 0 for last addition. 1214 * Return: 1215 * the hash value of the string MOD (%) the table size. 1216 */ 1217 1218u_int 1219st_hash(char *name, int len, int tabsz) 1220{ 1221 char *pt; 1222 char *dest; 1223 char *end; 1224 int i; 1225 u_int key = 0; 1226 int steps; 1227 int res; 1228 u_int val = 0; 1229 1230 /* 1231 * only look at the tail up to MAXKEYLEN, we do not need to waste 1232 * time here (remember these are pathnames, the tail is what will 1233 * spread out the keys) 1234 */ 1235 if (len > MAXKEYLEN) { 1236 pt = &(name[len - MAXKEYLEN]); 1237 len = MAXKEYLEN; 1238 } else 1239 pt = name; 1240 1241 /* 1242 * calculate the number of u_int size steps in the string and if 1243 * there is a runt to deal with 1244 */ 1245 steps = len/sizeof(u_int); 1246 res = len % sizeof(u_int); 1247 1248 /* 1249 * add up the value of the string in unsigned integer sized pieces 1250 * too bad we cannot have unsigned int aligned strings, then we 1251 * could avoid the expensive copy. 1252 */ 1253 for (i = 0; i < steps; ++i) { 1254 end = pt + sizeof(u_int); 1255 dest = (char *)&val; 1256 while (pt < end) 1257 *dest++ = *pt++; 1258 key += val; 1259 } 1260 1261 /* 1262 * add in the runt padded with zero to the right 1263 */ 1264 if (res) { 1265 val = 0; 1266 end = pt + res; 1267 dest = (char *)&val; 1268 while (pt < end) 1269 *dest++ = *pt++; 1270 key += val; 1271 } 1272 1273 /* 1274 * return the result mod the table size 1275 */ 1276 return(key % tabsz); 1277} 1278