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#include <sys/cdefs.h> 38#ifndef lint 39#if 0 40static const char sccsid[] = "@(#)tables.c 8.1 (Berkeley) 5/31/93"; 41#else 42__used static const char rcsid[] = "$OpenBSD: tables.c,v 1.25 2007/09/02 15:19:08 deraadt Exp $"; 43#endif 44#endif /* not lint */ 45 46#include <sys/types.h> 47#include <sys/time.h> 48#include <sys/stat.h> 49#include <sys/param.h> 50#include <sys/fcntl.h> 51#include <stdio.h> 52#include <string.h> 53#include <unistd.h> 54#include <errno.h> 55#include <stdlib.h> 56#include "pax.h" 57#include "tables.h" 58#include "extern.h" 59 60/* 61 * Routines for controlling the contents of all the different databases pax 62 * keeps. Tables are dynamically created only when they are needed. The 63 * goal was speed and the ability to work with HUGE archives. The databases 64 * were kept simple, but do have complex rules for when the contents change. 65 * As of this writing, the posix library functions were more complex than 66 * needed for this application (pax databases have very short lifetimes and 67 * do not survive after pax is finished). Pax is required to handle very 68 * large archives. These database routines carefully combine memory usage and 69 * temporary file storage in ways which will not significantly impact runtime 70 * performance while allowing the largest possible archives to be handled. 71 * Trying to force the fit to the posix database routines was not considered 72 * time well spent. 73 */ 74 75static HRDLNK **ltab = NULL; /* hard link table for detecting hard links */ 76static FTM **ftab = NULL; /* file time table for updating arch */ 77static NAMT **ntab = NULL; /* interactive rename storage table */ 78static DEVT **dtab = NULL; /* device/inode mapping tables */ 79static ATDIR **atab = NULL; /* file tree directory time reset table */ 80static DIRDATA *dirp = NULL; /* storage for setting created dir time/mode */ 81static size_t dirsize; /* size of dirp table */ 82static long dircnt = 0; /* entries in dir time/mode storage */ 83static int ffd = -1; /* tmp file for file time table name storage */ 84 85static DEVT *chk_dev(dev_t, int); 86 87/* 88 * hard link table routines 89 * 90 * The hard link table tries to detect hard links to files using the device and 91 * inode values. We do this when writing an archive, so we can tell the format 92 * write routine that this file is a hard link to another file. The format 93 * write routine then can store this file in whatever way it wants (as a hard 94 * link if the format supports that like tar, or ignore this info like cpio). 95 * (Actually a field in the format driver table tells us if the format wants 96 * hard link info. if not, we do not waste time looking for them). We also use 97 * the same table when reading an archive. In that situation, this table is 98 * used by the format read routine to detect hard links from stored dev and 99 * inode numbers (like cpio). This will allow pax to create a link when one 100 * can be detected by the archive format. 101 */ 102 103/* 104 * lnk_start 105 * Creates the hard link table. 106 * Return: 107 * 0 if created, -1 if failure 108 */ 109 110int 111lnk_start(void) 112{ 113 if (ltab != NULL) 114 return(0); 115 if ((ltab = (HRDLNK **)calloc(L_TAB_SZ, sizeof(HRDLNK *))) == NULL) { 116 paxwarn(1, "Cannot allocate memory for hard link table"); 117 return(-1); 118 } 119 return(0); 120} 121 122/* 123 * chk_lnk() 124 * Looks up entry in hard link hash table. If found, it copies the name 125 * of the file it is linked to (we already saw that file) into ln_name. 126 * lnkcnt is decremented and if goes to 1 the node is deleted from the 127 * database. (We have seen all the links to this file). If not found, 128 * we add the file to the database if it has the potential for having 129 * hard links to other files we may process (it has a link count > 1) 130 * Return: 131 * if found returns 1; if not found returns 0; -1 on error 132 */ 133 134int 135chk_lnk(ARCHD *arcn) 136{ 137 HRDLNK *pt; 138 HRDLNK **ppt; 139 u_int indx; 140 141 if (ltab == NULL) 142 return(-1); 143 /* 144 * ignore those nodes that cannot have hard links 145 */ 146 if ((arcn->type == PAX_DIR) || (arcn->sb.st_nlink <= 1)) 147 return(0); 148 149 /* 150 * hash inode number and look for this file 151 */ 152 indx = ((unsigned)arcn->sb.st_ino) % L_TAB_SZ; 153 if ((pt = ltab[indx]) != NULL) { 154 /* 155 * its hash chain in not empty, walk down looking for it 156 */ 157 ppt = &(ltab[indx]); 158 while (pt != NULL) { 159 if ((pt->ino == arcn->sb.st_ino) && 160 (pt->dev == arcn->sb.st_dev)) 161 break; 162 ppt = &(pt->fow); 163 pt = pt->fow; 164 } 165 166 if (pt != NULL) { 167 /* 168 * found a link. set the node type and copy in the 169 * name of the file it is to link to. we need to 170 * handle hardlinks to regular files differently than 171 * other links. 172 */ 173 arcn->ln_nlen = strlcpy(arcn->ln_name, pt->name, 174 sizeof(arcn->ln_name)); 175 /* XXX truncate? */ 176 if (arcn->nlen >= sizeof(arcn->name)) 177 arcn->nlen = sizeof(arcn->name) - 1; 178 if (arcn->type == PAX_REG) 179 arcn->type = PAX_HRG; 180 else 181 arcn->type = PAX_HLK; 182 183 /* 184 * if we have found all the links to this file, remove 185 * it from the database 186 */ 187 if (--pt->nlink <= 1) { 188 *ppt = pt->fow; 189 (void)free((char *)pt->name); 190 (void)free((char *)pt); 191 } 192 return(1); 193 } 194 } 195 196 /* 197 * we never saw this file before. It has links so we add it to the 198 * front of this hash chain 199 */ 200 if ((pt = (HRDLNK *)malloc(sizeof(HRDLNK))) != NULL) { 201 if ((pt->name = strdup(arcn->name)) != NULL) { 202 pt->dev = arcn->sb.st_dev; 203 pt->ino = arcn->sb.st_ino; 204 pt->nlink = arcn->sb.st_nlink; 205 pt->fow = ltab[indx]; 206 ltab[indx] = pt; 207 return(0); 208 } 209 (void)free((char *)pt); 210 } 211 212 paxwarn(1, "Hard link table out of memory"); 213 return(-1); 214} 215 216/* 217 * purg_lnk 218 * remove reference for a file that we may have added to the data base as 219 * a potential source for hard links. We ended up not using the file, so 220 * we do not want to accidently point another file at it later on. 221 */ 222 223void 224purg_lnk(ARCHD *arcn) 225{ 226 HRDLNK *pt; 227 HRDLNK **ppt; 228 u_int indx; 229 230 if (ltab == NULL) 231 return; 232 /* 233 * do not bother to look if it could not be in the database 234 */ 235 if ((arcn->sb.st_nlink <= 1) || (arcn->type == PAX_DIR) || 236 (arcn->type == PAX_HLK) || (arcn->type == PAX_HRG)) 237 return; 238 239 /* 240 * find the hash chain for this inode value, if empty return 241 */ 242 indx = ((unsigned)arcn->sb.st_ino) % L_TAB_SZ; 243 if ((pt = ltab[indx]) == NULL) 244 return; 245 246 /* 247 * walk down the list looking for the inode/dev pair, unlink and 248 * free if found 249 */ 250 ppt = &(ltab[indx]); 251 while (pt != NULL) { 252 if ((pt->ino == arcn->sb.st_ino) && 253 (pt->dev == arcn->sb.st_dev)) 254 break; 255 ppt = &(pt->fow); 256 pt = pt->fow; 257 } 258 if (pt == NULL) 259 return; 260 261 /* 262 * remove and free it 263 */ 264 *ppt = pt->fow; 265 (void)free((char *)pt->name); 266 (void)free((char *)pt); 267} 268 269/* 270 * lnk_end() 271 * pull apart a existing link table so we can reuse it. We do this between 272 * read and write phases of append with update. (The format may have 273 * used the link table, and we need to start with a fresh table for the 274 * write phase 275 */ 276 277void 278lnk_end(void) 279{ 280 int i; 281 HRDLNK *pt; 282 HRDLNK *ppt; 283 284 if (ltab == NULL) 285 return; 286 287 for (i = 0; i < L_TAB_SZ; ++i) { 288 if (ltab[i] == NULL) 289 continue; 290 pt = ltab[i]; 291 ltab[i] = NULL; 292 293 /* 294 * free up each entry on this chain 295 */ 296 while (pt != NULL) { 297 ppt = pt; 298 pt = ppt->fow; 299 (void)free((char *)ppt->name); 300 (void)free((char *)ppt); 301 } 302 } 303 return; 304} 305 306/* 307 * modification time table routines 308 * 309 * The modification time table keeps track of last modification times for all 310 * files stored in an archive during a write phase when -u is set. We only 311 * add a file to the archive if it is newer than a file with the same name 312 * already stored on the archive (if there is no other file with the same 313 * name on the archive it is added). This applies to writes and appends. 314 * An append with an -u must read the archive and store the modification time 315 * for every file on that archive before starting the write phase. It is clear 316 * that this is one HUGE database. To save memory space, the actual file names 317 * are stored in a scratch file and indexed by an in-memory hash table. The 318 * hash table is indexed by hashing the file path. The nodes in the table store 319 * the length of the filename and the lseek offset within the scratch file 320 * where the actual name is stored. Since there are never any deletions from 321 * this table, fragmentation of the scratch file is never a issue. Lookups 322 * seem to not exhibit any locality at all (files in the database are rarely 323 * looked up more than once...), so caching is just a waste of memory. The 324 * only limitation is the amount of scratch file space available to store the 325 * path names. 326 */ 327 328/* 329 * ftime_start() 330 * create the file time hash table and open for read/write the scratch 331 * file. (after created it is unlinked, so when we exit we leave 332 * no witnesses). 333 * Return: 334 * 0 if the table and file was created ok, -1 otherwise 335 */ 336 337int 338ftime_start(void) 339{ 340 341 if (ftab != NULL) 342 return(0); 343 if ((ftab = (FTM **)calloc(F_TAB_SZ, sizeof(FTM *))) == NULL) { 344 paxwarn(1, "Cannot allocate memory for file time table"); 345 return(-1); 346 } 347 348 /* 349 * get random name and create temporary scratch file, unlink name 350 * so it will get removed on exit 351 */ 352 memcpy(tempbase, _TFILE_BASE, sizeof(_TFILE_BASE)); 353 if ((ffd = mkstemp(tempfile)) < 0) { 354 syswarn(1, errno, "Unable to create temporary file: %s", 355 tempfile); 356 return(-1); 357 } 358 (void)unlink(tempfile); 359 360 return(0); 361} 362 363/* 364 * chk_ftime() 365 * looks up entry in file time hash table. If not found, the file is 366 * added to the hash table and the file named stored in the scratch file. 367 * If a file with the same name is found, the file times are compared and 368 * the most recent file time is retained. If the new file was younger (or 369 * was not in the database) the new file is selected for storage. 370 * Return: 371 * 0 if file should be added to the archive, 1 if it should be skipped, 372 * -1 on error 373 */ 374 375int 376chk_ftime(ARCHD *arcn) 377{ 378 FTM *pt; 379 int namelen; 380 u_int indx; 381 char ckname[PAXPATHLEN+1]; 382 383 /* 384 * no info, go ahead and add to archive 385 */ 386 if (ftab == NULL) 387 return(0); 388 389 /* 390 * hash the pathname and look up in table 391 */ 392 namelen = arcn->nlen; 393 indx = st_hash(arcn->name, namelen, F_TAB_SZ); 394 if ((pt = ftab[indx]) != NULL) { 395 /* 396 * the hash chain is not empty, walk down looking for match 397 * only read up the path names if the lengths match, speeds 398 * up the search a lot 399 */ 400 while (pt != NULL) { 401 if (pt->namelen == namelen) { 402 /* 403 * potential match, have to read the name 404 * from the scratch file. 405 */ 406 if (lseek(ffd,pt->seek,SEEK_SET) != pt->seek) { 407 syswarn(1, errno, 408 "Failed ftime table seek"); 409 return(-1); 410 } 411 if (read(ffd, ckname, namelen) != namelen) { 412 syswarn(1, errno, 413 "Failed ftime table read"); 414 return(-1); 415 } 416 417 /* 418 * if the names match, we are done 419 */ 420 if (!strncmp(ckname, arcn->name, namelen)) 421 break; 422 } 423 424 /* 425 * try the next entry on the chain 426 */ 427 pt = pt->fow; 428 } 429 430 if (pt != NULL) { 431 /* 432 * found the file, compare the times, save the newer 433 */ 434 if (arcn->sb.st_mtime > pt->mtime) { 435 /* 436 * file is newer 437 */ 438 pt->mtime = arcn->sb.st_mtime; 439 return(0); 440 } 441 /* 442 * file is older 443 */ 444 return(1); 445 } 446 } 447 448 /* 449 * not in table, add it 450 */ 451 if ((pt = (FTM *)malloc(sizeof(FTM))) != NULL) { 452 /* 453 * add the name at the end of the scratch file, saving the 454 * offset. add the file to the head of the hash chain 455 */ 456 if ((pt->seek = lseek(ffd, (off_t)0, SEEK_END)) >= 0) { 457 if (write(ffd, arcn->name, namelen) == namelen) { 458 pt->mtime = arcn->sb.st_mtime; 459 pt->namelen = namelen; 460 pt->fow = ftab[indx]; 461 ftab[indx] = pt; 462 return(0); 463 } 464 syswarn(1, errno, "Failed write to file time table"); 465 } else 466 syswarn(1, errno, "Failed seek on file time table"); 467 } else 468 paxwarn(1, "File time table ran out of memory"); 469 470 if (pt != NULL) 471 (void)free((char *)pt); 472 return(-1); 473} 474 475/* 476 * Interactive rename table routines 477 * 478 * The interactive rename table keeps track of the new names that the user 479 * assigns to files from tty input. Since this map is unique for each file 480 * we must store it in case there is a reference to the file later in archive 481 * (a link). Otherwise we will be unable to find the file we know was 482 * extracted. The remapping of these files is stored in a memory based hash 483 * table (it is assumed since input must come from /dev/tty, it is unlikely to 484 * be a very large table). 485 */ 486 487/* 488 * name_start() 489 * create the interactive rename table 490 * Return: 491 * 0 if successful, -1 otherwise 492 */ 493 494int 495name_start(void) 496{ 497 if (ntab != NULL) 498 return(0); 499 if ((ntab = (NAMT **)calloc(N_TAB_SZ, sizeof(NAMT *))) == NULL) { 500 paxwarn(1, "Cannot allocate memory for interactive rename table"); 501 return(-1); 502 } 503 return(0); 504} 505 506/* 507 * add_name() 508 * add the new name to old name mapping just created by the user. 509 * If an old name mapping is found (there may be duplicate names on an 510 * archive) only the most recent is kept. 511 * Return: 512 * 0 if added, -1 otherwise 513 */ 514 515int 516add_name(char *oname, int onamelen, char *nname) 517{ 518 NAMT *pt; 519 u_int indx; 520 521 if (ntab == NULL) { 522 /* 523 * should never happen 524 */ 525 paxwarn(0, "No interactive rename table, links may fail"); 526 return(0); 527 } 528 529 /* 530 * look to see if we have already mapped this file, if so we 531 * will update it 532 */ 533 indx = st_hash(oname, onamelen, N_TAB_SZ); 534 if ((pt = ntab[indx]) != NULL) { 535 /* 536 * look down the has chain for the file 537 */ 538 while ((pt != NULL) && (strcmp(oname, pt->oname) != 0)) 539 pt = pt->fow; 540 541 if (pt != NULL) { 542 /* 543 * found an old mapping, replace it with the new one 544 * the user just input (if it is different) 545 */ 546 if (strcmp(nname, pt->nname) == 0) 547 return(0); 548 549 (void)free((char *)pt->nname); 550 if ((pt->nname = strdup(nname)) == NULL) { 551 paxwarn(1, "Cannot update rename table"); 552 return(-1); 553 } 554 return(0); 555 } 556 } 557 558 /* 559 * this is a new mapping, add it to the table 560 */ 561 if ((pt = (NAMT *)malloc(sizeof(NAMT))) != NULL) { 562 if ((pt->oname = strdup(oname)) != NULL) { 563 if ((pt->nname = strdup(nname)) != NULL) { 564 pt->fow = ntab[indx]; 565 ntab[indx] = pt; 566 return(0); 567 } 568 (void)free((char *)pt->oname); 569 } 570 (void)free((char *)pt); 571 } 572 paxwarn(1, "Interactive rename table out of memory"); 573 return(-1); 574} 575 576/* 577 * sub_name() 578 * look up a link name to see if it points at a file that has been 579 * remapped by the user. If found, the link is adjusted to contain the 580 * new name (oname is the link to name) 581 */ 582 583void 584sub_name(char *oname, int *onamelen, size_t onamesize) 585{ 586 NAMT *pt; 587 u_int indx; 588 589 if (ntab == NULL) 590 return; 591 /* 592 * look the name up in the hash table 593 */ 594 indx = st_hash(oname, *onamelen, N_TAB_SZ); 595 if ((pt = ntab[indx]) == NULL) 596 return; 597 598 while (pt != NULL) { 599 /* 600 * walk down the hash chain looking for a match 601 */ 602 if (strcmp(oname, pt->oname) == 0) { 603 /* 604 * found it, replace it with the new name 605 * and return (we know that oname has enough space) 606 */ 607 *onamelen = strlcpy(oname, pt->nname, onamesize); 608 if (*onamelen >= onamesize) 609 *onamelen = onamesize - 1; /* XXX truncate? */ 610 return; 611 } 612 pt = pt->fow; 613 } 614 615 /* 616 * no match, just return 617 */ 618 return; 619} 620 621/* 622 * device/inode mapping table routines 623 * (used with formats that store device and inodes fields) 624 * 625 * device/inode mapping tables remap the device field in a archive header. The 626 * device/inode fields are used to determine when files are hard links to each 627 * other. However these values have very little meaning outside of that. This 628 * database is used to solve one of two different problems. 629 * 630 * 1) when files are appended to an archive, while the new files may have hard 631 * links to each other, you cannot determine if they have hard links to any 632 * file already stored on the archive from a prior run of pax. We must assume 633 * that these inode/device pairs are unique only within a SINGLE run of pax 634 * (which adds a set of files to an archive). So we have to make sure the 635 * inode/dev pairs we add each time are always unique. We do this by observing 636 * while the inode field is very dense, the use of the dev field is fairly 637 * sparse. Within each run of pax, we remap any device number of a new archive 638 * member that has a device number used in a prior run and already stored in a 639 * file on the archive. During the read phase of the append, we store the 640 * device numbers used and mark them to not be used by any file during the 641 * write phase. If during write we go to use one of those old device numbers, 642 * we remap it to a new value. 643 * 644 * 2) Often the fields in the archive header used to store these values are 645 * too small to store the entire value. The result is an inode or device value 646 * which can be truncated. This really can foul up an archive. With truncation 647 * we end up creating links between files that are really not links (after 648 * truncation the inodes are the same value). We address that by detecting 649 * truncation and forcing a remap of the device field to split truncated 650 * inodes away from each other. Each truncation creates a pattern of bits that 651 * are removed. We use this pattern of truncated bits to partition the inodes 652 * on a single device to many different devices (each one represented by the 653 * truncated bit pattern). All inodes on the same device that have the same 654 * truncation pattern are mapped to the same new device. Two inodes that 655 * truncate to the same value clearly will always have different truncation 656 * bit patterns, so they will be split from away each other. When we spot 657 * device truncation we remap the device number to a non truncated value. 658 * (for more info see table.h for the data structures involved). 659 */ 660 661/* 662 * dev_start() 663 * create the device mapping table 664 * Return: 665 * 0 if successful, -1 otherwise 666 */ 667 668int 669dev_start(void) 670{ 671 if (dtab != NULL) 672 return(0); 673 if ((dtab = (DEVT **)calloc(D_TAB_SZ, sizeof(DEVT *))) == NULL) { 674 paxwarn(1, "Cannot allocate memory for device mapping table"); 675 return(-1); 676 } 677 return(0); 678} 679 680/* 681 * add_dev() 682 * add a device number to the table. this will force the device to be 683 * remapped to a new value if it be used during a write phase. This 684 * function is called during the read phase of an append to prohibit the 685 * use of any device number already in the archive. 686 * Return: 687 * 0 if added ok, -1 otherwise 688 */ 689 690int 691add_dev(ARCHD *arcn) 692{ 693 if (chk_dev(arcn->sb.st_dev, 1) == NULL) 694 return(-1); 695 return(0); 696} 697 698/* 699 * chk_dev() 700 * check for a device value in the device table. If not found and the add 701 * flag is set, it is added. This does NOT assign any mapping values, just 702 * adds the device number as one that need to be remapped. If this device 703 * is already mapped, just return with a pointer to that entry. 704 * Return: 705 * pointer to the entry for this device in the device map table. Null 706 * if the add flag is not set and the device is not in the table (it is 707 * not been seen yet). If add is set and the device cannot be added, null 708 * is returned (indicates an error). 709 */ 710 711static DEVT * 712chk_dev(dev_t dev, int add) 713{ 714 DEVT *pt; 715 u_int indx; 716 717 if (dtab == NULL) 718 return(NULL); 719 /* 720 * look to see if this device is already in the table 721 */ 722 indx = ((unsigned)dev) % D_TAB_SZ; 723 if ((pt = dtab[indx]) != NULL) { 724 while ((pt != NULL) && (pt->dev != dev)) 725 pt = pt->fow; 726 727 /* 728 * found it, return a pointer to it 729 */ 730 if (pt != NULL) 731 return(pt); 732 } 733 734 /* 735 * not in table, we add it only if told to as this may just be a check 736 * to see if a device number is being used. 737 */ 738 if (add == 0) 739 return(NULL); 740 741 /* 742 * allocate a node for this device and add it to the front of the hash 743 * chain. Note we do not assign remaps values here, so the pt->list 744 * list must be NULL. 745 */ 746 if ((pt = (DEVT *)malloc(sizeof(DEVT))) == NULL) { 747 paxwarn(1, "Device map table out of memory"); 748 return(NULL); 749 } 750 pt->dev = dev; 751 pt->list = NULL; 752 pt->fow = dtab[indx]; 753 dtab[indx] = pt; 754 return(pt); 755} 756/* 757 * map_dev() 758 * given an inode and device storage mask (the mask has a 1 for each bit 759 * the archive format is able to store in a header), we check for inode 760 * and device truncation and remap the device as required. Device mapping 761 * can also occur when during the read phase of append a device number was 762 * seen (and was marked as do not use during the write phase). WE ASSUME 763 * that unsigned longs are the same size or bigger than the fields used 764 * for ino_t and dev_t. If not the types will have to be changed. 765 * Return: 766 * 0 if all ok, -1 otherwise. 767 */ 768 769int 770map_dev(ARCHD *arcn, u_long dev_mask, u_long ino_mask) 771{ 772 DEVT *pt; 773 DLIST *dpt; 774 static dev_t lastdev = 0; /* next device number to try */ 775 int trc_ino = 0; 776 int trc_dev = 0; 777 ino_t trunc_bits = 0; 778 ino_t nino; 779 780 if (dtab == NULL) 781 return(0); 782 /* 783 * check for device and inode truncation, and extract the truncated 784 * bit pattern. 785 */ 786 if ((arcn->sb.st_dev & (dev_t)dev_mask) != arcn->sb.st_dev) 787 ++trc_dev; 788 if ((nino = arcn->sb.st_ino & (ino_t)ino_mask) != arcn->sb.st_ino) { 789 ++trc_ino; 790 trunc_bits = arcn->sb.st_ino & (ino_t)(~ino_mask); 791 } 792 793 /* 794 * see if this device is already being mapped, look up the device 795 * then find the truncation bit pattern which applies 796 */ 797 if ((pt = chk_dev(arcn->sb.st_dev, 0)) != NULL) { 798 /* 799 * this device is already marked to be remapped 800 */ 801 for (dpt = pt->list; dpt != NULL; dpt = dpt->fow) 802 if (dpt->trunc_bits == trunc_bits) 803 break; 804 805 if (dpt != NULL) { 806 /* 807 * we are being remapped for this device and pattern 808 * change the device number to be stored and return 809 */ 810 arcn->sb.st_dev = dpt->dev; 811 arcn->sb.st_ino = nino; 812 return(0); 813 } 814 } else { 815 /* 816 * this device is not being remapped YET. if we do not have any 817 * form of truncation, we do not need a remap 818 */ 819 if (!trc_ino && !trc_dev) 820 return(0); 821 822 /* 823 * we have truncation, have to add this as a device to remap 824 */ 825 if ((pt = chk_dev(arcn->sb.st_dev, 1)) == NULL) 826 goto bad; 827 828 /* 829 * if we just have a truncated inode, we have to make sure that 830 * all future inodes that do not truncate (they have the 831 * truncation pattern of all 0's) continue to map to the same 832 * device number. We probably have already written inodes with 833 * this device number to the archive with the truncation 834 * pattern of all 0's. So we add the mapping for all 0's to the 835 * same device number. 836 */ 837 if (!trc_dev && (trunc_bits != 0)) { 838 if ((dpt = (DLIST *)malloc(sizeof(DLIST))) == NULL) 839 goto bad; 840 dpt->trunc_bits = 0; 841 dpt->dev = arcn->sb.st_dev; 842 dpt->fow = pt->list; 843 pt->list = dpt; 844 } 845 } 846 847 /* 848 * look for a device number not being used. We must watch for wrap 849 * around on lastdev (so we do not get stuck looking forever!) 850 */ 851 while (++lastdev > 0) { 852 if (chk_dev(lastdev, 0) != NULL) 853 continue; 854 /* 855 * found an unused value. If we have reached truncation point 856 * for this format we are hosed, so we give up. Otherwise we 857 * mark it as being used. 858 */ 859 if (((lastdev & ((dev_t)dev_mask)) != lastdev) || 860 (chk_dev(lastdev, 1) == NULL)) 861 goto bad; 862 break; 863 } 864 865 if ((lastdev <= 0) || ((dpt = (DLIST *)malloc(sizeof(DLIST))) == NULL)) 866 goto bad; 867 868 /* 869 * got a new device number, store it under this truncation pattern. 870 * change the device number this file is being stored with. 871 */ 872 dpt->trunc_bits = trunc_bits; 873 dpt->dev = lastdev; 874 dpt->fow = pt->list; 875 pt->list = dpt; 876 arcn->sb.st_dev = lastdev; 877 arcn->sb.st_ino = nino; 878 return(0); 879 880 bad: 881 paxwarn(1, "Unable to fix truncated inode/device field when storing %s", 882 arcn->name); 883 paxwarn(0, "Archive may create improper hard links when extracted"); 884 return(0); 885} 886 887/* 888 * directory access/mod time reset table routines (for directories READ by pax) 889 * 890 * The pax -t flag requires that access times of archive files be the same 891 * before being read by pax. For regular files, access time is restored after 892 * the file has been copied. This database provides the same functionality for 893 * directories read during file tree traversal. Restoring directory access time 894 * is more complex than files since directories may be read several times until 895 * all the descendants in their subtree are visited by fts. Directory access 896 * and modification times are stored during the fts pre-order visit (done 897 * before any descendants in the subtree are visited) and restored after the 898 * fts post-order visit (after all the descendants have been visited). In the 899 * case of premature exit from a subtree (like from the effects of -n), any 900 * directory entries left in this database are reset during final cleanup 901 * operations of pax. Entries are hashed by inode number for fast lookup. 902 */ 903 904/* 905 * atdir_start() 906 * create the directory access time database for directories READ by pax. 907 * Return: 908 * 0 is created ok, -1 otherwise. 909 */ 910 911int 912atdir_start(void) 913{ 914 if (atab != NULL) 915 return(0); 916 if ((atab = (ATDIR **)calloc(A_TAB_SZ, sizeof(ATDIR *))) == NULL) { 917 paxwarn(1,"Cannot allocate space for directory access time table"); 918 return(-1); 919 } 920 return(0); 921} 922 923 924/* 925 * atdir_end() 926 * walk through the directory access time table and reset the access time 927 * of any directory who still has an entry left in the database. These 928 * entries are for directories READ by pax 929 */ 930 931void 932atdir_end(void) 933{ 934 ATDIR *pt; 935 int i; 936 937 if (atab == NULL) 938 return; 939 /* 940 * for each non-empty hash table entry reset all the directories 941 * chained there. 942 */ 943 for (i = 0; i < A_TAB_SZ; ++i) { 944 if ((pt = atab[i]) == NULL) 945 continue; 946 /* 947 * remember to force the times, set_ftime() looks at pmtime 948 * and patime, which only applies to things CREATED by pax, 949 * not read by pax. Read time reset is controlled by -t. 950 */ 951 for (; pt != NULL; pt = pt->fow) 952 set_ftime(pt->name, pt->mtime, pt->atime, 1); 953 } 954} 955 956/* 957 * add_atdir() 958 * add a directory to the directory access time table. Table is hashed 959 * and chained by inode number. This is for directories READ by pax 960 */ 961 962void 963add_atdir(char *fname, dev_t dev, ino_t ino, time_t mtime, time_t atime) 964{ 965 ATDIR *pt; 966 u_int indx; 967 968 if (atab == NULL) 969 return; 970 971 /* 972 * make sure this directory is not already in the table, if so just 973 * return (the older entry always has the correct time). The only 974 * way this will happen is when the same subtree can be traversed by 975 * different args to pax and the -n option is aborting fts out of a 976 * subtree before all the post-order visits have been made. 977 */ 978 indx = ((unsigned)ino) % A_TAB_SZ; 979 if ((pt = atab[indx]) != NULL) { 980 while (pt != NULL) { 981 if ((pt->ino == ino) && (pt->dev == dev)) 982 break; 983 pt = pt->fow; 984 } 985 986 /* 987 * oops, already there. Leave it alone. 988 */ 989 if (pt != NULL) 990 return; 991 } 992 993 /* 994 * add it to the front of the hash chain 995 */ 996 if ((pt = (ATDIR *)malloc(sizeof(ATDIR))) != NULL) { 997 if ((pt->name = strdup(fname)) != NULL) { 998 pt->dev = dev; 999 pt->ino = ino; 1000 pt->mtime = mtime; 1001 pt->atime = atime; 1002 pt->fow = atab[indx]; 1003 atab[indx] = pt; 1004 return; 1005 } 1006 (void)free((char *)pt); 1007 } 1008 1009 paxwarn(1, "Directory access time reset table ran out of memory"); 1010 return; 1011} 1012 1013/* 1014 * get_atdir() 1015 * look up a directory by inode and device number to obtain the access 1016 * and modification time you want to set to. If found, the modification 1017 * and access time parameters are set and the entry is removed from the 1018 * table (as it is no longer needed). These are for directories READ by 1019 * pax 1020 * Return: 1021 * 0 if found, -1 if not found. 1022 */ 1023 1024int 1025get_atdir(dev_t dev, ino_t ino, time_t *mtime, time_t *atime) 1026{ 1027 ATDIR *pt; 1028 ATDIR **ppt; 1029 u_int indx; 1030 1031 if (atab == NULL) 1032 return(-1); 1033 /* 1034 * hash by inode and search the chain for an inode and device match 1035 */ 1036 indx = ((unsigned)ino) % A_TAB_SZ; 1037 if ((pt = atab[indx]) == NULL) 1038 return(-1); 1039 1040 ppt = &(atab[indx]); 1041 while (pt != NULL) { 1042 if ((pt->ino == ino) && (pt->dev == dev)) 1043 break; 1044 /* 1045 * no match, go to next one 1046 */ 1047 ppt = &(pt->fow); 1048 pt = pt->fow; 1049 } 1050 1051 /* 1052 * return if we did not find it. 1053 */ 1054 if (pt == NULL) 1055 return(-1); 1056 1057 /* 1058 * found it. return the times and remove the entry from the table. 1059 */ 1060 *ppt = pt->fow; 1061 *mtime = pt->mtime; 1062 *atime = pt->atime; 1063 (void)free((char *)pt->name); 1064 (void)free((char *)pt); 1065 return(0); 1066} 1067 1068/* 1069 * directory access mode and time storage routines (for directories CREATED 1070 * by pax). 1071 * 1072 * Pax requires that extracted directories, by default, have their access/mod 1073 * times and permissions set to the values specified in the archive. During the 1074 * actions of extracting (and creating the destination subtree during -rw copy) 1075 * directories extracted may be modified after being created. Even worse is 1076 * that these directories may have been created with file permissions which 1077 * prohibits any descendants of these directories from being extracted. When 1078 * directories are created by pax, access rights may be added to permit the 1079 * creation of files in their subtree. Every time pax creates a directory, the 1080 * times and file permissions specified by the archive are stored. After all 1081 * files have been extracted (or copied), these directories have their times 1082 * and file modes reset to the stored values. The directory info is restored in 1083 * reverse order as entries were added to the data file from root to leaf. To 1084 * restore atime properly, we must go backwards. The data file consists of 1085 * records with two parts, the file name followed by a DIRDATA trailer. The 1086 * fixed sized trailer contains the size of the name plus the off_t location in 1087 * the file. To restore we work backwards through the file reading the trailer 1088 * then the file name. 1089 */ 1090 1091/* 1092 * dir_start() 1093 * set up the directory time and file mode storage for directories CREATED 1094 * by pax. 1095 * Return: 1096 * 0 if ok, -1 otherwise 1097 */ 1098 1099int 1100dir_start(void) 1101{ 1102 if (dirp != NULL) 1103 return(0); 1104 1105 dirsize = DIRP_SIZE; 1106 if ((dirp = calloc(dirsize, sizeof(DIRDATA))) == NULL) { 1107 paxwarn(1, "Unable to allocate memory for directory times"); 1108 return(-1); 1109 } 1110 return(0); 1111} 1112 1113/* 1114 * add_dir() 1115 * add the mode and times for a newly CREATED directory 1116 * name is name of the directory, psb the stat buffer with the data in it, 1117 * frc_mode is a flag that says whether to force the setting of the mode 1118 * (ignoring the user set values for preserving file mode). Frc_mode is 1119 * for the case where we created a file and found that the resulting 1120 * directory was not writeable and the user asked for file modes to NOT 1121 * be preserved. (we have to preserve what was created by default, so we 1122 * have to force the setting at the end. this is stated explicitly in the 1123 * pax spec) 1124 */ 1125 1126void 1127add_dir(char *name, size_t nlen, struct stat *psb, int frc_mode) 1128{ 1129 DIRDATA *dblk; 1130 char realname[MAXPATHLEN], *rp; 1131 1132 if (dirp == NULL) 1133 return; 1134 1135 if (havechd && *name != '/') { 1136 if ((rp = realpath(name, realname)) == NULL) { 1137 paxwarn(1, "Cannot canonicalize %s", name); 1138 return; 1139 } 1140 name = rp; 1141 } 1142 if (dircnt == dirsize) { 1143 dblk = realloc(dirp, 2 * dirsize * sizeof(DIRDATA)); 1144 if (dblk == NULL) { 1145 paxwarn(1, "Unable to store mode and times for created" 1146 " directory: %s", name); 1147 return; 1148 } 1149 dirp = dblk; 1150 dirsize *= 2; 1151 } 1152 dblk = &dirp[dircnt]; 1153 if ((dblk->name = strdup(name)) == NULL) { 1154 paxwarn(1, "Unable to store mode and times for created" 1155 " directory: %s", name); 1156 return; 1157 } 1158 dblk->mode = psb->st_mode & 0xffff; 1159 dblk->mtime = psb->st_mtime; 1160 dblk->atime = psb->st_atime; 1161 dblk->frc_mode = frc_mode; 1162 ++dircnt; 1163} 1164 1165/* 1166 * proc_dir() 1167 * process all file modes and times stored for directories CREATED 1168 * by pax 1169 */ 1170 1171void 1172proc_dir(void) 1173{ 1174 DIRDATA *dblk; 1175 long cnt; 1176 1177 if (dirp == NULL) 1178 return; 1179 /* 1180 * read backwards through the file and process each directory 1181 */ 1182 cnt = dircnt; 1183 while (--cnt >= 0) { 1184 /* 1185 * frc_mode set, make sure we set the file modes even if 1186 * the user didn't ask for it (see file_subs.c for more info) 1187 */ 1188 dblk = &dirp[cnt]; 1189 if (pmode || dblk->frc_mode) 1190 set_pmode(dblk->name, dblk->mode); 1191 if (patime || pmtime) 1192 set_ftime(dblk->name, dblk->mtime, dblk->atime, 0); 1193 free(dblk->name); 1194 } 1195 1196 free(dirp); 1197 dirp = NULL; 1198 dircnt = 0; 1199} 1200 1201/* 1202 * database independent routines 1203 */ 1204 1205/* 1206 * st_hash() 1207 * hashes filenames to a u_int for hashing into a table. Looks at the tail 1208 * end of file, as this provides far better distribution than any other 1209 * part of the name. For performance reasons we only care about the last 1210 * MAXKEYLEN chars (should be at LEAST large enough to pick off the file 1211 * name). Was tested on 500,000 name file tree traversal from the root 1212 * and gave almost a perfectly uniform distribution of keys when used with 1213 * prime sized tables (MAXKEYLEN was 128 in test). Hashes (sizeof int) 1214 * chars at a time and pads with 0 for last addition. 1215 * Return: 1216 * the hash value of the string MOD (%) the table size. 1217 */ 1218 1219u_int 1220st_hash(char *name, int len, int tabsz) 1221{ 1222 char *pt; 1223 char *dest; 1224 char *end; 1225 int i; 1226 u_int key = 0; 1227 int steps; 1228 int res; 1229 u_int val = 0; 1230 1231 /* 1232 * only look at the tail up to MAXKEYLEN, we do not need to waste 1233 * time here (remember these are pathnames, the tail is what will 1234 * spread out the keys) 1235 */ 1236 if (len > MAXKEYLEN) { 1237 pt = &(name[len - MAXKEYLEN]); 1238 len = MAXKEYLEN; 1239 } else 1240 pt = name; 1241 1242 /* 1243 * calculate the number of u_int size steps in the string and if 1244 * there is a runt to deal with 1245 */ 1246 steps = len/sizeof(u_int); 1247 res = len % sizeof(u_int); 1248 1249 /* 1250 * add up the value of the string in unsigned integer sized pieces 1251 * too bad we cannot have unsigned int aligned strings, then we 1252 * could avoid the expensive copy. 1253 */ 1254 for (i = 0; i < steps; ++i) { 1255 end = pt + sizeof(u_int); 1256 dest = (char *)&val; 1257 while (pt < end) 1258 *dest++ = *pt++; 1259 key += val; 1260 } 1261 1262 /* 1263 * add in the runt padded with zero to the right 1264 */ 1265 if (res) { 1266 val = 0; 1267 end = pt + res; 1268 dest = (char *)&val; 1269 while (pt < end) 1270 *dest++ = *pt++; 1271 key += val; 1272 } 1273 1274 /* 1275 * return the result mod the table size 1276 */ 1277 return(key % tabsz); 1278} 1279