1 /* $NetBSD: tables.c,v 1.30 2008/01/10 04:24:51 tls Exp $ */ 2 3 /*- 4 * Copyright (c) 1992 Keith Muller. 5 * Copyright (c) 1992, 1993 6 * The Regents of the University of California. All rights reserved. 7 * 8 * This code is derived from software contributed to Berkeley by 9 * Keith Muller of the University of California, San Diego. 10 * 11 * Redistribution and use in source and binary forms, with or without 12 * modification, are permitted provided that the following conditions 13 * are met: 14 * 1. Redistributions of source code must retain the above copyright 15 * notice, this list of conditions and the following disclaimer. 16 * 2. Redistributions in binary form must reproduce the above copyright 17 * notice, this list of conditions and the following disclaimer in the 18 * documentation and/or other materials provided with the distribution. 19 * 3. Neither the name of the University nor the names of its contributors 20 * may be used to endorse or promote products derived from this software 21 * without specific prior written permission. 22 * 23 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 24 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 25 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 26 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 27 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 28 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 29 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 30 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 31 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 32 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 33 * SUCH DAMAGE. 34 */ 35 36 #if HAVE_NBTOOL_CONFIG_H 37 #include "nbtool_config.h" 38 #endif 39 40 #include <sys/cdefs.h> 41 #if !defined(lint) 42 #if 0 43 static char sccsid[] = "@(#)tables.c 8.1 (Berkeley) 5/31/93"; 44 #else 45 __RCSID("$NetBSD: tables.c,v 1.30 2008/01/10 04:24:51 tls Exp $"); 46 #endif 47 #endif /* not lint */ 48 49 #include <sys/types.h> 50 #include <sys/time.h> 51 #include <sys/stat.h> 52 #include <sys/param.h> 53 #include <stdio.h> 54 #include <ctype.h> 55 #include <fcntl.h> 56 #include <paths.h> 57 #include <string.h> 58 #include <unistd.h> 59 #include <errno.h> 60 #include <stdlib.h> 61 #include "pax.h" 62 #include "tables.h" 63 #include "extern.h" 64 65 /* 66 * Routines for controlling the contents of all the different databases pax 67 * keeps. Tables are dynamically created only when they are needed. The 68 * goal was speed and the ability to work with HUGE archives. The databases 69 * were kept simple, but do have complex rules for when the contents change. 70 * As of this writing, the POSIX library functions were more complex than 71 * needed for this application (pax databases have very short lifetimes and 72 * do not survive after pax is finished). Pax is required to handle very 73 * large archives. These database routines carefully combine memory usage and 74 * temporary file storage in ways which will not significantly impact runtime 75 * performance while allowing the largest possible archives to be handled. 76 * Trying to force the fit to the POSIX database routines was not considered 77 * time well spent. 78 */ 79 80 static HRDLNK **ltab = NULL; /* hard link table for detecting hard links */ 81 static FTM **ftab = NULL; /* file time table for updating arch */ 82 static NAMT **ntab = NULL; /* interactive rename storage table */ 83 static DEVT **dtab = NULL; /* device/inode mapping tables */ 84 static ATDIR **atab = NULL; /* file tree directory time reset table */ 85 #ifdef DIRS_USE_FILE 86 static int dirfd = -1; /* storage for setting created dir time/mode */ 87 static u_long dircnt; /* entries in dir time/mode storage */ 88 #endif 89 static int ffd = -1; /* tmp file for file time table name storage */ 90 91 static DEVT *chk_dev(dev_t, int); 92 93 /* 94 * hard link table routines 95 * 96 * The hard link table tries to detect hard links to files using the device and 97 * inode values. We do this when writing an archive, so we can tell the format 98 * write routine that this file is a hard link to another file. The format 99 * write routine then can store this file in whatever way it wants (as a hard 100 * link if the format supports that like tar, or ignore this info like cpio). 101 * (Actually a field in the format driver table tells us if the format wants 102 * hard link info. if not, we do not waste time looking for them). We also use 103 * the same table when reading an archive. In that situation, this table is 104 * used by the format read routine to detect hard links from stored dev and 105 * inode numbers (like cpio). This will allow pax to create a link when one 106 * can be detected by the archive format. 107 */ 108 109 /* 110 * lnk_start 111 * Creates the hard link table. 112 * Return: 113 * 0 if created, -1 if failure 114 */ 115 116 int 117 lnk_start(void) 118 { 119 if (ltab != NULL) 120 return 0; 121 if ((ltab = (HRDLNK **)calloc(L_TAB_SZ, sizeof(HRDLNK *))) == NULL) { 122 tty_warn(1, "Cannot allocate memory for hard link table"); 123 return -1; 124 } 125 return 0; 126 } 127 128 /* 129 * chk_lnk() 130 * Looks up entry in hard link hash table. If found, it copies the name 131 * of the file it is linked to (we already saw that file) into ln_name. 132 * lnkcnt is decremented and if goes to 1 the node is deleted from the 133 * database. (We have seen all the links to this file). If not found, 134 * we add the file to the database if it has the potential for having 135 * hard links to other files we may process (it has a link count > 1) 136 * Return: 137 * if found returns 1; if not found returns 0; -1 on error 138 */ 139 140 int 141 chk_lnk(ARCHD *arcn) 142 { 143 HRDLNK *pt; 144 HRDLNK **ppt; 145 u_int indx; 146 147 if (ltab == NULL) 148 return -1; 149 /* 150 * ignore those nodes that cannot have hard links 151 */ 152 if ((arcn->type == PAX_DIR) || (arcn->sb.st_nlink <= 1)) 153 return 0; 154 155 /* 156 * hash inode number and look for this file 157 */ 158 indx = ((unsigned)arcn->sb.st_ino) % L_TAB_SZ; 159 if ((pt = ltab[indx]) != NULL) { 160 /* 161 * its hash chain is not empty, walk down looking for it 162 */ 163 ppt = &(ltab[indx]); 164 while (pt != NULL) { 165 if ((pt->ino == arcn->sb.st_ino) && 166 (pt->dev == arcn->sb.st_dev)) 167 break; 168 ppt = &(pt->fow); 169 pt = pt->fow; 170 } 171 172 if (pt != NULL) { 173 /* 174 * found a link. set the node type and copy in the 175 * name of the file it is to link to. we need to 176 * handle hardlinks to regular files differently than 177 * other links. 178 */ 179 arcn->ln_nlen = strlcpy(arcn->ln_name, pt->name, 180 sizeof(arcn->ln_name)); 181 if (arcn->type == PAX_REG) 182 arcn->type = PAX_HRG; 183 else 184 arcn->type = PAX_HLK; 185 186 /* 187 * if we have found all the links to this file, remove 188 * it from the database 189 */ 190 if (--pt->nlink <= 1) { 191 *ppt = pt->fow; 192 (void)free((char *)pt->name); 193 (void)free((char *)pt); 194 } 195 return 1; 196 } 197 } 198 199 /* 200 * we never saw this file before. It has links so we add it to the 201 * front of this hash chain 202 */ 203 if ((pt = (HRDLNK *)malloc(sizeof(HRDLNK))) != NULL) { 204 if ((pt->name = strdup(arcn->name)) != NULL) { 205 pt->dev = arcn->sb.st_dev; 206 pt->ino = arcn->sb.st_ino; 207 pt->nlink = arcn->sb.st_nlink; 208 pt->fow = ltab[indx]; 209 ltab[indx] = pt; 210 return 0; 211 } 212 (void)free((char *)pt); 213 } 214 215 tty_warn(1, "Hard link table out of memory"); 216 return -1; 217 } 218 219 /* 220 * purg_lnk 221 * remove reference for a file that we may have added to the data base as 222 * a potential source for hard links. We ended up not using the file, so 223 * we do not want to accidentally point another file at it later on. 224 */ 225 226 void 227 purg_lnk(ARCHD *arcn) 228 { 229 HRDLNK *pt; 230 HRDLNK **ppt; 231 u_int indx; 232 233 if (ltab == NULL) 234 return; 235 /* 236 * do not bother to look if it could not be in the database 237 */ 238 if ((arcn->sb.st_nlink <= 1) || (arcn->type == PAX_DIR) || 239 (arcn->type == PAX_HLK) || (arcn->type == PAX_HRG)) 240 return; 241 242 /* 243 * find the hash chain for this inode value, if empty return 244 */ 245 indx = ((unsigned)arcn->sb.st_ino) % L_TAB_SZ; 246 if ((pt = ltab[indx]) == NULL) 247 return; 248 249 /* 250 * walk down the list looking for the inode/dev pair, unlink and 251 * free if found 252 */ 253 ppt = &(ltab[indx]); 254 while (pt != NULL) { 255 if ((pt->ino == arcn->sb.st_ino) && 256 (pt->dev == arcn->sb.st_dev)) 257 break; 258 ppt = &(pt->fow); 259 pt = pt->fow; 260 } 261 if (pt == NULL) 262 return; 263 264 /* 265 * remove and free it 266 */ 267 *ppt = pt->fow; 268 (void)free((char *)pt->name); 269 (void)free((char *)pt); 270 } 271 272 /* 273 * lnk_end() 274 * pull apart a existing link table so we can reuse it. We do this between 275 * read and write phases of append with update. (The format may have 276 * used the link table, and we need to start with a fresh table for the 277 * write phase 278 */ 279 280 void 281 lnk_end(void) 282 { 283 int i; 284 HRDLNK *pt; 285 HRDLNK *ppt; 286 287 if (ltab == NULL) 288 return; 289 290 for (i = 0; i < L_TAB_SZ; ++i) { 291 if (ltab[i] == NULL) 292 continue; 293 pt = ltab[i]; 294 ltab[i] = NULL; 295 296 /* 297 * free up each entry on this chain 298 */ 299 while (pt != NULL) { 300 ppt = pt; 301 pt = ppt->fow; 302 (void)free((char *)ppt->name); 303 (void)free((char *)ppt); 304 } 305 } 306 return; 307 } 308 309 /* 310 * modification time table routines 311 * 312 * The modification time table keeps track of last modification times for all 313 * files stored in an archive during a write phase when -u is set. We only 314 * add a file to the archive if it is newer than a file with the same name 315 * already stored on the archive (if there is no other file with the same 316 * name on the archive it is added). This applies to writes and appends. 317 * An append with an -u must read the archive and store the modification time 318 * for every file on that archive before starting the write phase. It is clear 319 * that this is one HUGE database. To save memory space, the actual file names 320 * are stored in a scratch file and indexed by an in-memory hash table. The 321 * hash table is indexed by hashing the file path. The nodes in the table store 322 * the length of the filename and the lseek offset within the scratch file 323 * where the actual name is stored. Since there are never any deletions from this 324 * table, fragmentation of the scratch file is never a issue. Lookups seem to 325 * not exhibit any locality at all (files in the database are rarely 326 * looked up more than once...), so caching is just a waste of memory. The 327 * only limitation is the amount of scratch file space available to store the 328 * path names. 329 */ 330 331 /* 332 * ftime_start() 333 * create the file time hash table and open for read/write the scratch 334 * file. (after created it is unlinked, so when we exit we leave 335 * no witnesses). 336 * Return: 337 * 0 if the table and file was created ok, -1 otherwise 338 */ 339 340 int 341 ftime_start(void) 342 { 343 if (ftab != NULL) 344 return 0; 345 if ((ftab = (FTM **)calloc(F_TAB_SZ, sizeof(FTM *))) == NULL) { 346 tty_warn(1, "Cannot allocate memory for file time table"); 347 return -1; 348 } 349 350 /* 351 * get random name and create temporary scratch file, unlink name 352 * so it will get removed on exit 353 */ 354 memcpy(tempbase, _TFILE_BASE, sizeof(_TFILE_BASE)); 355 if ((ffd = mkstemp(tempfile)) == -1) { 356 syswarn(1, errno, "Unable to create temporary file: %s", 357 tempfile); 358 return -1; 359 } 360 361 (void)unlink(tempfile); 362 return 0; 363 } 364 365 /* 366 * chk_ftime() 367 * looks up entry in file time hash table. If not found, the file is 368 * added to the hash table and the file named stored in the scratch file. 369 * If a file with the same name is found, the file times are compared and 370 * the most recent file time is retained. If the new file was younger (or 371 * was not in the database) the new file is selected for storage. 372 * Return: 373 * 0 if file should be added to the archive, 1 if it should be skipped, 374 * -1 on error 375 */ 376 377 int 378 chk_ftime(ARCHD *arcn) 379 { 380 FTM *pt; 381 int namelen; 382 u_int indx; 383 char ckname[PAXPATHLEN+1]; 384 385 /* 386 * no info, go ahead and add to archive 387 */ 388 if (ftab == NULL) 389 return 0; 390 391 /* 392 * hash the pathname and look up in table 393 */ 394 namelen = arcn->nlen; 395 indx = st_hash(arcn->name, namelen, F_TAB_SZ); 396 if ((pt = ftab[indx]) != NULL) { 397 /* 398 * the hash chain is not empty, walk down looking for match 399 * only read up the path names if the lengths match, speeds 400 * up the search a lot 401 */ 402 while (pt != NULL) { 403 if (pt->namelen == namelen) { 404 /* 405 * potential match, have to read the name 406 * from the scratch file. 407 */ 408 if (lseek(ffd,pt->seek,SEEK_SET) != pt->seek) { 409 syswarn(1, errno, 410 "Failed ftime table seek"); 411 return -1; 412 } 413 if (xread(ffd, ckname, namelen) != namelen) { 414 syswarn(1, errno, 415 "Failed ftime table read"); 416 return -1; 417 } 418 419 /* 420 * if the names match, we are done 421 */ 422 if (!strncmp(ckname, arcn->name, namelen)) 423 break; 424 } 425 426 /* 427 * try the next entry on the chain 428 */ 429 pt = pt->fow; 430 } 431 432 if (pt != NULL) { 433 /* 434 * found the file, compare the times, save the newer 435 */ 436 if (arcn->sb.st_mtime > pt->mtime) { 437 /* 438 * file is newer 439 */ 440 pt->mtime = arcn->sb.st_mtime; 441 return 0; 442 } 443 /* 444 * file is older 445 */ 446 return 1; 447 } 448 } 449 450 /* 451 * not in table, add it 452 */ 453 if ((pt = (FTM *)malloc(sizeof(FTM))) != NULL) { 454 /* 455 * add the name at the end of the scratch file, saving the 456 * offset. add the file to the head of the hash chain 457 */ 458 if ((pt->seek = lseek(ffd, (off_t)0, SEEK_END)) >= 0) { 459 if (xwrite(ffd, arcn->name, namelen) == namelen) { 460 pt->mtime = arcn->sb.st_mtime; 461 pt->namelen = namelen; 462 pt->fow = ftab[indx]; 463 ftab[indx] = pt; 464 return 0; 465 } 466 syswarn(1, errno, "Failed write to file time table"); 467 } else 468 syswarn(1, errno, "Failed seek on file time table"); 469 } else 470 tty_warn(1, "File time table ran out of memory"); 471 472 if (pt != NULL) 473 (void)free((char *)pt); 474 return -1; 475 } 476 477 /* 478 * Interactive rename table routines 479 * 480 * The interactive rename table keeps track of the new names that the user 481 * assigns to files from tty input. Since this map is unique for each file 482 * we must store it in case there is a reference to the file later in archive 483 * (a link). Otherwise we will be unable to find the file we know was 484 * extracted. The remapping of these files is stored in a memory based hash 485 * table (it is assumed since input must come from /dev/tty, it is unlikely to 486 * be a very large table). 487 */ 488 489 /* 490 * name_start() 491 * create the interactive rename table 492 * Return: 493 * 0 if successful, -1 otherwise 494 */ 495 496 int 497 name_start(void) 498 { 499 if (ntab != NULL) 500 return 0; 501 if ((ntab = (NAMT **)calloc(N_TAB_SZ, sizeof(NAMT *))) == NULL) { 502 tty_warn(1, 503 "Cannot allocate memory for interactive rename table"); 504 return -1; 505 } 506 return 0; 507 } 508 509 /* 510 * add_name() 511 * add the new name to old name mapping just created by the user. 512 * If an old name mapping is found (there may be duplicate names on an 513 * archive) only the most recent is kept. 514 * Return: 515 * 0 if added, -1 otherwise 516 */ 517 518 int 519 add_name(char *oname, int onamelen, char *nname) 520 { 521 NAMT *pt; 522 u_int indx; 523 524 if (ntab == NULL) { 525 /* 526 * should never happen 527 */ 528 tty_warn(0, "No interactive rename table, links may fail\n"); 529 return 0; 530 } 531 532 /* 533 * look to see if we have already mapped this file, if so we 534 * will update it 535 */ 536 indx = st_hash(oname, onamelen, N_TAB_SZ); 537 if ((pt = ntab[indx]) != NULL) { 538 /* 539 * look down the has chain for the file 540 */ 541 while ((pt != NULL) && (strcmp(oname, pt->oname) != 0)) 542 pt = pt->fow; 543 544 if (pt != NULL) { 545 /* 546 * found an old mapping, replace it with the new one 547 * the user just input (if it is different) 548 */ 549 if (strcmp(nname, pt->nname) == 0) 550 return 0; 551 552 (void)free((char *)pt->nname); 553 if ((pt->nname = strdup(nname)) == NULL) { 554 tty_warn(1, "Cannot update rename table"); 555 return -1; 556 } 557 return 0; 558 } 559 } 560 561 /* 562 * this is a new mapping, add it to the table 563 */ 564 if ((pt = (NAMT *)malloc(sizeof(NAMT))) != NULL) { 565 if ((pt->oname = strdup(oname)) != NULL) { 566 if ((pt->nname = strdup(nname)) != NULL) { 567 pt->fow = ntab[indx]; 568 ntab[indx] = pt; 569 return 0; 570 } 571 (void)free((char *)pt->oname); 572 } 573 (void)free((char *)pt); 574 } 575 tty_warn(1, "Interactive rename table out of memory"); 576 return -1; 577 } 578 579 /* 580 * sub_name() 581 * look up a link name to see if it points at a file that has been 582 * remapped by the user. If found, the link is adjusted to contain the 583 * new name (oname is the link to name) 584 */ 585 586 void 587 sub_name(char *oname, int *onamelen, size_t onamesize) 588 { 589 NAMT *pt; 590 u_int indx; 591 592 if (ntab == NULL) 593 return; 594 /* 595 * look the name up in the hash table 596 */ 597 indx = st_hash(oname, *onamelen, N_TAB_SZ); 598 if ((pt = ntab[indx]) == NULL) 599 return; 600 601 while (pt != NULL) { 602 /* 603 * walk down the hash chain looking for a match 604 */ 605 if (strcmp(oname, pt->oname) == 0) { 606 /* 607 * found it, replace it with the new name 608 * and return (we know that oname has enough space) 609 */ 610 *onamelen = strlcpy(oname, pt->nname, onamesize); 611 return; 612 } 613 pt = pt->fow; 614 } 615 616 /* 617 * no match, just return 618 */ 619 return; 620 } 621 622 /* 623 * device/inode mapping table routines 624 * (used with formats that store device and inodes fields) 625 * 626 * device/inode mapping tables remap the device field in an archive header. The 627 * device/inode fields are used to determine when files are hard links to each 628 * other. However these values have very little meaning outside of that. This 629 * database is used to solve one of two different problems. 630 * 631 * 1) when files are appended to an archive, while the new files may have hard 632 * links to each other, you cannot determine if they have hard links to any 633 * file already stored on the archive from a prior run of pax. We must assume 634 * that these inode/device pairs are unique only within a SINGLE run of pax 635 * (which adds a set of files to an archive). So we have to make sure the 636 * inode/dev pairs we add each time are always unique. We do this by observing 637 * while the inode field is very dense, the use of the dev field is fairly 638 * sparse. Within each run of pax, we remap any device number of a new archive 639 * member that has a device number used in a prior run and already stored in a 640 * file on the archive. During the read phase of the append, we store the 641 * device numbers used and mark them to not be used by any file during the 642 * write phase. If during write we go to use one of those old device numbers, 643 * we remap it to a new value. 644 * 645 * 2) Often the fields in the archive header used to store these values are 646 * too small to store the entire value. The result is an inode or device value 647 * which can be truncated. This really can foul up an archive. With truncation 648 * we end up creating links between files that are really not links (after 649 * truncation the inodes are the same value). We address that by detecting 650 * truncation and forcing a remap of the device field to split truncated 651 * inodes away from each other. Each truncation creates a pattern of bits that 652 * are removed. We use this pattern of truncated bits to partition the inodes 653 * on a single device to many different devices (each one represented by the 654 * truncated bit pattern). All inodes on the same device that have the same 655 * truncation pattern are mapped to the same new device. Two inodes that 656 * truncate to the same value clearly will always have different truncation 657 * bit patterns, so they will be split from away each other. When we spot 658 * device truncation we remap the device number to a non truncated value. 659 * (for more info see table.h for the data structures involved). 660 */ 661 662 /* 663 * dev_start() 664 * create the device mapping table 665 * Return: 666 * 0 if successful, -1 otherwise 667 */ 668 669 int 670 dev_start(void) 671 { 672 if (dtab != NULL) 673 return 0; 674 if ((dtab = (DEVT **)calloc(D_TAB_SZ, sizeof(DEVT *))) == NULL) { 675 tty_warn(1, "Cannot allocate memory for device mapping table"); 676 return -1; 677 } 678 return 0; 679 } 680 681 /* 682 * add_dev() 683 * add a device number to the table. this will force the device to be 684 * remapped to a new value if it be used during a write phase. This 685 * function is called during the read phase of an append to prohibit the 686 * use of any device number already in the archive. 687 * Return: 688 * 0 if added ok, -1 otherwise 689 */ 690 691 int 692 add_dev(ARCHD *arcn) 693 { 694 if (chk_dev(arcn->sb.st_dev, 1) == NULL) 695 return -1; 696 return 0; 697 } 698 699 /* 700 * chk_dev() 701 * check for a device value in the device table. If not found and the add 702 * flag is set, it is added. This does NOT assign any mapping values, just 703 * adds the device number as one that need to be remapped. If this device 704 * is already mapped, just return with a pointer to that entry. 705 * Return: 706 * pointer to the entry for this device in the device map table. Null 707 * if the add flag is not set and the device is not in the table (it is 708 * not been seen yet). If add is set and the device cannot be added, null 709 * is returned (indicates an error). 710 */ 711 712 static DEVT * 713 chk_dev(dev_t dev, int add) 714 { 715 DEVT *pt; 716 u_int indx; 717 718 if (dtab == NULL) 719 return NULL; 720 /* 721 * look to see if this device is already in the table 722 */ 723 indx = ((unsigned)dev) % D_TAB_SZ; 724 if ((pt = dtab[indx]) != NULL) { 725 while ((pt != NULL) && (pt->dev != dev)) 726 pt = pt->fow; 727 728 /* 729 * found it, return a pointer to it 730 */ 731 if (pt != NULL) 732 return pt; 733 } 734 735 /* 736 * not in table, we add it only if told to as this may just be a check 737 * to see if a device number is being used. 738 */ 739 if (add == 0) 740 return NULL; 741 742 /* 743 * allocate a node for this device and add it to the front of the hash 744 * chain. Note we do not assign remaps values here, so the pt->list 745 * list must be NULL. 746 */ 747 if ((pt = (DEVT *)malloc(sizeof(DEVT))) == NULL) { 748 tty_warn(1, "Device map table out of memory"); 749 return NULL; 750 } 751 pt->dev = dev; 752 pt->list = NULL; 753 pt->fow = dtab[indx]; 754 dtab[indx] = pt; 755 return pt; 756 } 757 /* 758 * map_dev() 759 * given an inode and device storage mask (the mask has a 1 for each bit 760 * the archive format is able to store in a header), we check for inode 761 * and device truncation and remap the device as required. Device mapping 762 * can also occur when during the read phase of append a device number was 763 * seen (and was marked as do not use during the write phase). WE ASSUME 764 * that unsigned longs are the same size or bigger than the fields used 765 * for ino_t and dev_t. If not the types will have to be changed. 766 * Return: 767 * 0 if all ok, -1 otherwise. 768 */ 769 770 int 771 map_dev(ARCHD *arcn, u_long dev_mask, u_long ino_mask) 772 { 773 DEVT *pt; 774 DLIST *dpt; 775 static dev_t lastdev = 0; /* next device number to try */ 776 int trc_ino = 0; 777 int trc_dev = 0; 778 ino_t trunc_bits = 0; 779 ino_t nino; 780 781 if (dtab == NULL) 782 return 0; 783 /* 784 * check for device and inode truncation, and extract the truncated 785 * bit pattern. 786 */ 787 if ((arcn->sb.st_dev & (dev_t)dev_mask) != arcn->sb.st_dev) 788 ++trc_dev; 789 if ((nino = arcn->sb.st_ino & (ino_t)ino_mask) != arcn->sb.st_ino) { 790 ++trc_ino; 791 trunc_bits = arcn->sb.st_ino & (ino_t)(~ino_mask); 792 } 793 794 /* 795 * see if this device is already being mapped, look up the device 796 * then find the truncation bit pattern which applies 797 */ 798 if ((pt = chk_dev(arcn->sb.st_dev, 0)) != NULL) { 799 /* 800 * this device is already marked to be remapped 801 */ 802 for (dpt = pt->list; dpt != NULL; dpt = dpt->fow) 803 if (dpt->trunc_bits == trunc_bits) 804 break; 805 806 if (dpt != NULL) { 807 /* 808 * we are being remapped for this device and pattern 809 * change the device number to be stored and return 810 */ 811 arcn->sb.st_dev = dpt->dev; 812 arcn->sb.st_ino = nino; 813 return 0; 814 } 815 } else { 816 /* 817 * this device is not being remapped YET. if we do not have any 818 * form of truncation, we do not need a remap 819 */ 820 if (!trc_ino && !trc_dev) 821 return 0; 822 823 /* 824 * we have truncation, have to add this as a device to remap 825 */ 826 if ((pt = chk_dev(arcn->sb.st_dev, 1)) == NULL) 827 goto bad; 828 829 /* 830 * if we just have a truncated inode, we have to make sure that 831 * all future inodes that do not truncate (they have the 832 * truncation pattern of all 0's) continue to map to the same 833 * device number. We probably have already written inodes with 834 * this device number to the archive with the truncation 835 * pattern of all 0's. So we add the mapping for all 0's to the 836 * same device number. 837 */ 838 if (!trc_dev && (trunc_bits != 0)) { 839 if ((dpt = (DLIST *)malloc(sizeof(DLIST))) == NULL) 840 goto bad; 841 dpt->trunc_bits = 0; 842 dpt->dev = arcn->sb.st_dev; 843 dpt->fow = pt->list; 844 pt->list = dpt; 845 } 846 } 847 848 /* 849 * look for a device number not being used. We must watch for wrap 850 * around on lastdev (so we do not get stuck looking forever!) 851 */ 852 while (++lastdev > 0) { 853 if (chk_dev(lastdev, 0) != NULL) 854 continue; 855 /* 856 * found an unused value. If we have reached truncation point 857 * for this format we are hosed, so we give up. Otherwise we 858 * mark it as being used. 859 */ 860 if (((lastdev & ((dev_t)dev_mask)) != lastdev) || 861 (chk_dev(lastdev, 1) == NULL)) 862 goto bad; 863 break; 864 } 865 866 if ((lastdev <= 0) || ((dpt = (DLIST *)malloc(sizeof(DLIST))) == NULL)) 867 goto bad; 868 869 /* 870 * got a new device number, store it under this truncation pattern. 871 * change the device number this file is being stored with. 872 */ 873 dpt->trunc_bits = trunc_bits; 874 dpt->dev = lastdev; 875 dpt->fow = pt->list; 876 pt->list = dpt; 877 arcn->sb.st_dev = lastdev; 878 arcn->sb.st_ino = nino; 879 return 0; 880 881 bad: 882 tty_warn(1, 883 "Unable to fix truncated inode/device field when storing %s", 884 arcn->name); 885 tty_warn(0, "Archive may create improper hard links when extracted"); 886 return 0; 887 } 888 889 /* 890 * directory access/mod time reset table routines (for directories READ by pax) 891 * 892 * The pax -t flag requires that access times of archive files to be the same 893 * as before being read by pax. For regular files, access time is restored after 894 * the file has been copied. This database provides the same functionality for 895 * directories read during file tree traversal. Restoring directory access time 896 * is more complex than files since directories may be read several times until 897 * all the descendants in their subtree are visited by fts. Directory access 898 * and modification times are stored during the fts pre-order visit (done 899 * before any descendants in the subtree is visited) and restored after the 900 * fts post-order visit (after all the descendants have been visited). In the 901 * case of premature exit from a subtree (like from the effects of -n), any 902 * directory entries left in this database are reset during final cleanup 903 * operations of pax. Entries are hashed by inode number for fast lookup. 904 */ 905 906 /* 907 * atdir_start() 908 * create the directory access time database for directories READ by pax. 909 * Return: 910 * 0 is created ok, -1 otherwise. 911 */ 912 913 int 914 atdir_start(void) 915 { 916 if (atab != NULL) 917 return 0; 918 if ((atab = (ATDIR **)calloc(A_TAB_SZ, sizeof(ATDIR *))) == NULL) { 919 tty_warn(1, 920 "Cannot allocate space for directory access time table"); 921 return -1; 922 } 923 return 0; 924 } 925 926 927 /* 928 * atdir_end() 929 * walk through the directory access time table and reset the access time 930 * of any directory who still has an entry left in the database. These 931 * entries are for directories READ by pax 932 */ 933 934 void 935 atdir_end(void) 936 { 937 ATDIR *pt; 938 int i; 939 940 if (atab == NULL) 941 return; 942 /* 943 * for each non-empty hash table entry reset all the directories 944 * chained there. 945 */ 946 for (i = 0; i < A_TAB_SZ; ++i) { 947 if ((pt = atab[i]) == NULL) 948 continue; 949 /* 950 * remember to force the times, set_ftime() looks at pmtime 951 * and patime, which only applies to things CREATED by pax, 952 * not read by pax. Read time reset is controlled by -t. 953 */ 954 for (; pt != NULL; pt = pt->fow) 955 set_ftime(pt->name, pt->mtime, pt->atime, 1, 0); 956 } 957 } 958 959 /* 960 * add_atdir() 961 * add a directory to the directory access time table. Table is hashed 962 * and chained by inode number. This is for directories READ by pax 963 */ 964 965 void 966 add_atdir(char *fname, dev_t dev, ino_t ino, time_t mtime, time_t atime) 967 { 968 ATDIR *pt; 969 u_int indx; 970 971 if (atab == NULL) 972 return; 973 974 /* 975 * make sure this directory is not already in the table, if so just 976 * return (the older entry always has the correct time). The only 977 * way this will happen is when the same subtree can be traversed by 978 * different args to pax and the -n option is aborting fts out of a 979 * subtree before all the post-order visits have been made. 980 */ 981 indx = ((unsigned)ino) % A_TAB_SZ; 982 if ((pt = atab[indx]) != NULL) { 983 while (pt != NULL) { 984 if ((pt->ino == ino) && (pt->dev == dev)) 985 break; 986 pt = pt->fow; 987 } 988 989 /* 990 * oops, already there. Leave it alone. 991 */ 992 if (pt != NULL) 993 return; 994 } 995 996 /* 997 * add it to the front of the hash chain 998 */ 999 if ((pt = (ATDIR *)malloc(sizeof(ATDIR))) != NULL) { 1000 if ((pt->name = strdup(fname)) != NULL) { 1001 pt->dev = dev; 1002 pt->ino = ino; 1003 pt->mtime = mtime; 1004 pt->atime = atime; 1005 pt->fow = atab[indx]; 1006 atab[indx] = pt; 1007 return; 1008 } 1009 (void)free((char *)pt); 1010 } 1011 1012 tty_warn(1, "Directory access time reset table ran out of memory"); 1013 return; 1014 } 1015 1016 /* 1017 * get_atdir() 1018 * look up a directory by inode and device number to obtain the access 1019 * and modification time you want to set to. If found, the modification 1020 * and access time parameters are set and the entry is removed from the 1021 * table (as it is no longer needed). These are for directories READ by 1022 * pax 1023 * Return: 1024 * 0 if found, -1 if not found. 1025 */ 1026 1027 int 1028 get_atdir(dev_t dev, ino_t ino, time_t *mtime, time_t *atime) 1029 { 1030 ATDIR *pt; 1031 ATDIR **ppt; 1032 u_int indx; 1033 1034 if (atab == NULL) 1035 return -1; 1036 /* 1037 * hash by inode and search the chain for an inode and device match 1038 */ 1039 indx = ((unsigned)ino) % A_TAB_SZ; 1040 if ((pt = atab[indx]) == NULL) 1041 return -1; 1042 1043 ppt = &(atab[indx]); 1044 while (pt != NULL) { 1045 if ((pt->ino == ino) && (pt->dev == dev)) 1046 break; 1047 /* 1048 * no match, go to next one 1049 */ 1050 ppt = &(pt->fow); 1051 pt = pt->fow; 1052 } 1053 1054 /* 1055 * return if we did not find it. 1056 */ 1057 if (pt == NULL) 1058 return -1; 1059 1060 /* 1061 * found it. return the times and remove the entry from the table. 1062 */ 1063 *ppt = pt->fow; 1064 *mtime = pt->mtime; 1065 *atime = pt->atime; 1066 (void)free((char *)pt->name); 1067 (void)free((char *)pt); 1068 return 0; 1069 } 1070 1071 /* 1072 * directory access mode and time storage routines (for directories CREATED 1073 * by pax). 1074 * 1075 * Pax requires that extracted directories, by default, have their access/mod 1076 * times and permissions set to the values specified in the archive. During the 1077 * actions of extracting (and creating the destination subtree during -rw copy) 1078 * directories extracted may be modified after being created. Even worse is 1079 * that these directories may have been created with file permissions which 1080 * prohibits any descendants of these directories from being extracted. When 1081 * directories are created by pax, access rights may be added to permit the 1082 * creation of files in their subtree. Every time pax creates a directory, the 1083 * times and file permissions specified by the archive are stored. After all 1084 * files have been extracted (or copied), these directories have their times 1085 * and file modes reset to the stored values. The directory info is restored in 1086 * reverse order as entries were added to the data file from root to leaf. To 1087 * restore atime properly, we must go backwards. The data file consists of 1088 * records with two parts, the file name followed by a DIRDATA trailer. The 1089 * fixed sized trailer contains the size of the name plus the off_t location in 1090 * the file. To restore we work backwards through the file reading the trailer 1091 * then the file name. 1092 */ 1093 1094 #ifndef DIRS_USE_FILE 1095 static DIRDATA *dirdata_head; 1096 #endif 1097 1098 /* 1099 * dir_start() 1100 * set up the directory time and file mode storage for directories CREATED 1101 * by pax. 1102 * Return: 1103 * 0 if ok, -1 otherwise 1104 */ 1105 1106 int 1107 dir_start(void) 1108 { 1109 #ifdef DIRS_USE_FILE 1110 if (dirfd != -1) 1111 return 0; 1112 1113 /* 1114 * unlink the file so it goes away at termination by itself 1115 */ 1116 memcpy(tempbase, _TFILE_BASE, sizeof(_TFILE_BASE)); 1117 if ((dirfd = mkstemp(tempfile)) >= 0) { 1118 (void)unlink(tempfile); 1119 return 0; 1120 } 1121 tty_warn(1, "Unable to create temporary file for directory times: %s", 1122 tempfile); 1123 return -1; 1124 #else 1125 return (0); 1126 #endif /* DIRS_USE_FILE */ 1127 } 1128 1129 /* 1130 * add_dir() 1131 * add the mode and times for a newly CREATED directory 1132 * name is name of the directory, psb the stat buffer with the data in it, 1133 * frc_mode is a flag that says whether to force the setting of the mode 1134 * (ignoring the user set values for preserving file mode). Frc_mode is 1135 * for the case where we created a file and found that the resulting 1136 * directory was not writable and the user asked for file modes to NOT 1137 * be preserved. (we have to preserve what was created by default, so we 1138 * have to force the setting at the end. this is stated explicitly in the 1139 * pax spec) 1140 */ 1141 1142 void 1143 add_dir(char *name, int nlen, struct stat *psb, int frc_mode) 1144 { 1145 #ifdef DIRS_USE_FILE 1146 DIRDATA dblk; 1147 #else 1148 DIRDATA *dblk; 1149 #endif 1150 char realname[MAXPATHLEN], *rp; 1151 1152 if (havechd && *name != '/') { 1153 if ((rp = realpath(name, realname)) == NULL) { 1154 tty_warn(1, "Cannot canonicalize %s", name); 1155 return; 1156 } 1157 name = rp; 1158 nlen = strlen(name); 1159 } 1160 1161 #ifdef DIRS_USE_FILE 1162 if (dirfd < 0) 1163 return; 1164 1165 /* 1166 * get current position (where file name will start) so we can store it 1167 * in the trailer 1168 */ 1169 if ((dblk.npos = lseek(dirfd, 0L, SEEK_CUR)) < 0) { 1170 tty_warn(1, 1171 "Unable to store mode and times for directory: %s",name); 1172 return; 1173 } 1174 1175 /* 1176 * write the file name followed by the trailer 1177 */ 1178 dblk.nlen = nlen + 1; 1179 dblk.mode = psb->st_mode & 0xffff; 1180 dblk.mtime = psb->st_mtime; 1181 dblk.atime = psb->st_atime; 1182 #if HAVE_STRUCT_STAT_ST_FLAGS 1183 dblk.fflags = psb->st_flags; 1184 #else 1185 dblk.fflags = 0; 1186 #endif 1187 dblk.frc_mode = frc_mode; 1188 if ((xwrite(dirfd, name, dblk.nlen) == dblk.nlen) && 1189 (xwrite(dirfd, (char *)&dblk, sizeof(dblk)) == sizeof(dblk))) { 1190 ++dircnt; 1191 return; 1192 } 1193 1194 tty_warn(1, 1195 "Unable to store mode and times for created directory: %s",name); 1196 return; 1197 #else 1198 1199 if ((dblk = malloc(sizeof(*dblk))) == NULL || 1200 (dblk->name = strdup(name)) == NULL) { 1201 tty_warn(1, 1202 "Unable to store mode and times for directory: %s",name); 1203 if (dblk != NULL) 1204 free(dblk); 1205 return; 1206 } 1207 1208 dblk->mode = psb->st_mode & 0xffff; 1209 dblk->mtime = psb->st_mtime; 1210 dblk->atime = psb->st_atime; 1211 #if HAVE_STRUCT_STAT_ST_FLAGS 1212 dblk->fflags = psb->st_flags; 1213 #else 1214 dblk->fflags = 0; 1215 #endif 1216 dblk->frc_mode = frc_mode; 1217 1218 dblk->next = dirdata_head; 1219 dirdata_head = dblk; 1220 return; 1221 #endif /* DIRS_USE_FILE */ 1222 } 1223 1224 /* 1225 * proc_dir() 1226 * process all file modes and times stored for directories CREATED 1227 * by pax 1228 */ 1229 1230 void 1231 proc_dir(void) 1232 { 1233 #ifdef DIRS_USE_FILE 1234 char name[PAXPATHLEN+1]; 1235 DIRDATA dblk; 1236 u_long cnt; 1237 1238 if (dirfd < 0) 1239 return; 1240 /* 1241 * read backwards through the file and process each directory 1242 */ 1243 for (cnt = 0; cnt < dircnt; ++cnt) { 1244 /* 1245 * read the trailer, then the file name, if this fails 1246 * just give up. 1247 */ 1248 if (lseek(dirfd, -((off_t)sizeof(dblk)), SEEK_CUR) < 0) 1249 break; 1250 if (xread(dirfd,(char *)&dblk, sizeof(dblk)) != sizeof(dblk)) 1251 break; 1252 if (lseek(dirfd, dblk.npos, SEEK_SET) < 0) 1253 break; 1254 if (xread(dirfd, name, dblk.nlen) != dblk.nlen) 1255 break; 1256 if (lseek(dirfd, dblk.npos, SEEK_SET) < 0) 1257 break; 1258 1259 /* 1260 * frc_mode set, make sure we set the file modes even if 1261 * the user didn't ask for it (see file_subs.c for more info) 1262 */ 1263 if (pmode || dblk.frc_mode) 1264 set_pmode(name, dblk.mode); 1265 if (patime || pmtime) 1266 set_ftime(name, dblk.mtime, dblk.atime, 0, 0); 1267 if (pfflags) 1268 set_chflags(name, dblk.fflags); 1269 } 1270 1271 (void)close(dirfd); 1272 dirfd = -1; 1273 if (cnt != dircnt) 1274 tty_warn(1, 1275 "Unable to set mode and times for created directories"); 1276 return; 1277 #else 1278 DIRDATA *dblk; 1279 1280 for (dblk = dirdata_head; dblk != NULL; dblk = dirdata_head) { 1281 dirdata_head = dblk->next; 1282 1283 /* 1284 * frc_mode set, make sure we set the file modes even if 1285 * the user didn't ask for it (see file_subs.c for more info) 1286 */ 1287 if (pmode || dblk->frc_mode) 1288 set_pmode(dblk->name, dblk->mode); 1289 if (patime || pmtime) 1290 set_ftime(dblk->name, dblk->mtime, dblk->atime, 0, 0); 1291 if (pfflags) 1292 set_chflags(dblk->name, dblk->fflags); 1293 1294 free(dblk->name); 1295 free(dblk); 1296 } 1297 #endif /* DIRS_USE_FILE */ 1298 } 1299 1300 /* 1301 * database independent routines 1302 */ 1303 1304 /* 1305 * st_hash() 1306 * hashes filenames to a u_int for hashing into a table. Looks at the tail 1307 * end of file, as this provides far better distribution than any other 1308 * part of the name. For performance reasons we only care about the last 1309 * MAXKEYLEN chars (should be at LEAST large enough to pick off the file 1310 * name). Was tested on 500,000 name file tree traversal from the root 1311 * and gave almost a perfectly uniform distribution of keys when used with 1312 * prime sized tables (MAXKEYLEN was 128 in test). Hashes (sizeof int) 1313 * chars at a time and pads with 0 for last addition. 1314 * Return: 1315 * the hash value of the string MOD (%) the table size. 1316 */ 1317 1318 u_int 1319 st_hash(char *name, int len, int tabsz) 1320 { 1321 char *pt; 1322 char *dest; 1323 char *end; 1324 int i; 1325 u_int key = 0; 1326 int steps; 1327 int res; 1328 u_int val; 1329 1330 /* 1331 * only look at the tail up to MAXKEYLEN, we do not need to waste 1332 * time here (remember these are pathnames, the tail is what will 1333 * spread out the keys) 1334 */ 1335 if (len > MAXKEYLEN) { 1336 pt = &(name[len - MAXKEYLEN]); 1337 len = MAXKEYLEN; 1338 } else 1339 pt = name; 1340 1341 /* 1342 * calculate the number of u_int size steps in the string and if 1343 * there is a runt to deal with 1344 */ 1345 steps = len/sizeof(u_int); 1346 res = len % sizeof(u_int); 1347 1348 /* 1349 * add up the value of the string in unsigned integer sized pieces 1350 * too bad we cannot have unsigned int aligned strings, then we 1351 * could avoid the expensive copy. 1352 */ 1353 for (i = 0; i < steps; ++i) { 1354 end = pt + sizeof(u_int); 1355 dest = (char *)&val; 1356 while (pt < end) 1357 *dest++ = *pt++; 1358 key += val; 1359 } 1360 1361 /* 1362 * add in the runt padded with zero to the right 1363 */ 1364 if (res) { 1365 val = 0; 1366 end = pt + res; 1367 dest = (char *)&val; 1368 while (pt < end) 1369 *dest++ = *pt++; 1370 key += val; 1371 } 1372 1373 /* 1374 * return the result mod the table size 1375 */ 1376 return key % tabsz; 1377 } 1378