1 /*- 2 * Copyright (c) 1992 Keith Muller. 3 * Copyright (c) 1992, 1993 4 * The Regents of the University of California. All rights reserved. 5 * 6 * This code is derived from software contributed to Berkeley by 7 * Keith Muller of the University of California, San Diego. 8 * 9 * Redistribution and use in source and binary forms, with or without 10 * modification, are permitted provided that the following conditions 11 * are met: 12 * 1. Redistributions of source code must retain the above copyright 13 * notice, this list of conditions and the following disclaimer. 14 * 2. Redistributions in binary form must reproduce the above copyright 15 * notice, this list of conditions and the following disclaimer in the 16 * documentation and/or other materials provided with the distribution. 17 * 3. All advertising materials mentioning features or use of this software 18 * must display the following acknowledgement: 19 * This product includes software developed by the University of 20 * California, Berkeley and its contributors. 21 * 4. Neither the name of the University nor the names of its contributors 22 * may be used to endorse or promote products derived from this software 23 * without specific prior written permission. 24 * 25 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 26 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 27 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 28 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 29 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 30 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 31 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 32 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 33 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 34 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 35 * SUCH DAMAGE. 36 */ 37 38 #ifndef lint 39 #if 0 40 static char sccsid[] = "@(#)tables.c 8.1 (Berkeley) 5/31/93"; 41 #endif 42 static const char rcsid[] = 43 "$FreeBSD$"; 44 #endif /* not lint */ 45 46 #include <sys/types.h> 47 #include <sys/time.h> 48 #include <sys/stat.h> 49 #include <sys/fcntl.h> 50 #include <errno.h> 51 #include <stdio.h> 52 #include <stdlib.h> 53 #include <string.h> 54 #include <unistd.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 databases routines was not considered 71 * time well spent. 72 */ 73 74 static HRDLNK **ltab = NULL; /* hard link table for detecting hard links */ 75 static FTM **ftab = NULL; /* file time table for updating arch */ 76 static NAMT **ntab = NULL; /* interactive rename storage table */ 77 static DEVT **dtab = NULL; /* device/inode mapping tables */ 78 static ATDIR **atab = NULL; /* file tree directory time reset table */ 79 static int dirfd = -1; /* storage for setting created dir time/mode */ 80 static u_long dircnt; /* entries in dir time/mode storage */ 81 static int ffd = -1; /* tmp file for file time table name storage */ 82 83 static DEVT *chk_dev __P((dev_t, int)); 84 85 /* 86 * hard link table routines 87 * 88 * The hard link table tries to detect hard links to files using the device and 89 * inode values. We do this when writing an archive, so we can tell the format 90 * write routine that this file is a hard link to another file. The format 91 * write routine then can store this file in whatever way it wants (as a hard 92 * link if the format supports that like tar, or ignore this info like cpio). 93 * (Actually a field in the format driver table tells us if the format wants 94 * hard link info. if not, we do not waste time looking for them). We also use 95 * the same table when reading an archive. In that situation, this table is 96 * used by the format read routine to detect hard links from stored dev and 97 * inode numbers (like cpio). This will allow pax to create a link when one 98 * can be detected by the archive format. 99 */ 100 101 /* 102 * lnk_start 103 * Creates the hard link table. 104 * Return: 105 * 0 if created, -1 if failure 106 */ 107 108 #ifdef __STDC__ 109 int 110 lnk_start(void) 111 #else 112 int 113 lnk_start() 114 #endif 115 { 116 if (ltab != NULL) 117 return(0); 118 if ((ltab = (HRDLNK **)calloc(L_TAB_SZ, sizeof(HRDLNK *))) == NULL) { 119 paxwarn(1, "Cannot allocate memory for hard link table"); 120 return(-1); 121 } 122 return(0); 123 } 124 125 /* 126 * chk_lnk() 127 * Looks up entry in hard link hash table. If found, it copies the name 128 * of the file it is linked to (we already saw that file) into ln_name. 129 * lnkcnt is decremented and if goes to 1 the node is deleted from the 130 * database. (We have seen all the links to this file). If not found, 131 * we add the file to the database if it has the potential for having 132 * hard links to other files we may process (it has a link count > 1) 133 * Return: 134 * if found returns 1; if not found returns 0; -1 on error 135 */ 136 137 #ifdef __STDC__ 138 int 139 chk_lnk(register ARCHD *arcn) 140 #else 141 int 142 chk_lnk(arcn) 143 register ARCHD *arcn; 144 #endif 145 { 146 register HRDLNK *pt; 147 register HRDLNK **ppt; 148 register u_int indx; 149 150 if (ltab == NULL) 151 return(-1); 152 /* 153 * ignore those nodes that cannot have hard links 154 */ 155 if ((arcn->type == PAX_DIR) || (arcn->sb.st_nlink <= 1)) 156 return(0); 157 158 /* 159 * hash inode number and look for this file 160 */ 161 indx = ((unsigned)arcn->sb.st_ino) % L_TAB_SZ; 162 if ((pt = ltab[indx]) != NULL) { 163 /* 164 * it's hash chain in not empty, walk down looking for it 165 */ 166 ppt = &(ltab[indx]); 167 while (pt != NULL) { 168 if ((pt->ino == arcn->sb.st_ino) && 169 (pt->dev == arcn->sb.st_dev)) 170 break; 171 ppt = &(pt->fow); 172 pt = pt->fow; 173 } 174 175 if (pt != NULL) { 176 /* 177 * found a link. set the node type and copy in the 178 * name of the file it is to link to. we need to 179 * handle hardlinks to regular files differently than 180 * other links. 181 */ 182 arcn->ln_nlen = l_strncpy(arcn->ln_name, pt->name, 183 sizeof(arcn->ln_name) - 1); 184 arcn->ln_name[arcn->ln_nlen] = '\0'; 185 if (arcn->type == PAX_REG) 186 arcn->type = PAX_HRG; 187 else 188 arcn->type = PAX_HLK; 189 190 /* 191 * if we have found all the links to this file, remove 192 * it from the database 193 */ 194 if (--pt->nlink <= 1) { 195 *ppt = pt->fow; 196 (void)free((char *)pt->name); 197 (void)free((char *)pt); 198 } 199 return(1); 200 } 201 } 202 203 /* 204 * we never saw this file before. It has links so we add it to the 205 * front of this hash chain 206 */ 207 if ((pt = (HRDLNK *)malloc(sizeof(HRDLNK))) != NULL) { 208 if ((pt->name = strdup(arcn->name)) != NULL) { 209 pt->dev = arcn->sb.st_dev; 210 pt->ino = arcn->sb.st_ino; 211 pt->nlink = arcn->sb.st_nlink; 212 pt->fow = ltab[indx]; 213 ltab[indx] = pt; 214 return(0); 215 } 216 (void)free((char *)pt); 217 } 218 219 paxwarn(1, "Hard link table out of memory"); 220 return(-1); 221 } 222 223 /* 224 * purg_lnk 225 * remove reference for a file that we may have added to the data base as 226 * a potential source for hard links. We ended up not using the file, so 227 * we do not want to accidently point another file at it later on. 228 */ 229 230 #ifdef __STDC__ 231 void 232 purg_lnk(register ARCHD *arcn) 233 #else 234 void 235 purg_lnk(arcn) 236 register ARCHD *arcn; 237 #endif 238 { 239 register HRDLNK *pt; 240 register HRDLNK **ppt; 241 register u_int indx; 242 243 if (ltab == NULL) 244 return; 245 /* 246 * do not bother to look if it could not be in the database 247 */ 248 if ((arcn->sb.st_nlink <= 1) || (arcn->type == PAX_DIR) || 249 (arcn->type == PAX_HLK) || (arcn->type == PAX_HRG)) 250 return; 251 252 /* 253 * find the hash chain for this inode value, if empty return 254 */ 255 indx = ((unsigned)arcn->sb.st_ino) % L_TAB_SZ; 256 if ((pt = ltab[indx]) == NULL) 257 return; 258 259 /* 260 * walk down the list looking for the inode/dev pair, unlink and 261 * free if found 262 */ 263 ppt = &(ltab[indx]); 264 while (pt != NULL) { 265 if ((pt->ino == arcn->sb.st_ino) && 266 (pt->dev == arcn->sb.st_dev)) 267 break; 268 ppt = &(pt->fow); 269 pt = pt->fow; 270 } 271 if (pt == NULL) 272 return; 273 274 /* 275 * remove and free it 276 */ 277 *ppt = pt->fow; 278 (void)free((char *)pt->name); 279 (void)free((char *)pt); 280 } 281 282 /* 283 * lnk_end() 284 * pull apart a existing link table so we can reuse it. We do this between 285 * read and write phases of append with update. (The format may have 286 * used the link table, and we need to start with a fresh table for the 287 * write phase 288 */ 289 290 #ifdef __STDC__ 291 void 292 lnk_end(void) 293 #else 294 void 295 lnk_end() 296 #endif 297 { 298 register int i; 299 register HRDLNK *pt; 300 register HRDLNK *ppt; 301 302 if (ltab == NULL) 303 return; 304 305 for (i = 0; i < L_TAB_SZ; ++i) { 306 if (ltab[i] == NULL) 307 continue; 308 pt = ltab[i]; 309 ltab[i] = NULL; 310 311 /* 312 * free up each entry on this chain 313 */ 314 while (pt != NULL) { 315 ppt = pt; 316 pt = ppt->fow; 317 (void)free((char *)ppt->name); 318 (void)free((char *)ppt); 319 } 320 } 321 return; 322 } 323 324 /* 325 * modification time table routines 326 * 327 * The modification time table keeps track of last modification times for all 328 * files stored in an archive during a write phase when -u is set. We only 329 * add a file to the archive if it is newer than a file with the same name 330 * already stored on the archive (if there is no other file with the same 331 * name on the archive it is added). This applies to writes and appends. 332 * An append with an -u must read the archive and store the modification time 333 * for every file on that archive before starting the write phase. It is clear 334 * that this is one HUGE database. To save memory space, the actual file names 335 * are stored in a scatch file and indexed by an in memory hash table. The 336 * hash table is indexed by hashing the file path. The nodes in the table store 337 * the length of the filename and the lseek offset within the scratch file 338 * where the actual name is stored. Since there are never any deletions to this 339 * table, fragmentation of the scratch file is never a issue. Lookups seem to 340 * not exhibit any locality at all (files in the database are rarely 341 * looked up more than once...). So caching is just a waste of memory. The 342 * only limitation is the amount of scatch file space available to store the 343 * path names. 344 */ 345 346 /* 347 * ftime_start() 348 * create the file time hash table and open for read/write the scratch 349 * file. (after created it is unlinked, so when we exit we leave 350 * no witnesses). 351 * Return: 352 * 0 if the table and file was created ok, -1 otherwise 353 */ 354 355 #ifdef __STDC__ 356 int 357 ftime_start(void) 358 #else 359 int 360 ftime_start() 361 #endif 362 { 363 364 if (ftab != NULL) 365 return(0); 366 if ((ftab = (FTM **)calloc(F_TAB_SZ, sizeof(FTM *))) == NULL) { 367 paxwarn(1, "Cannot allocate memory for file time table"); 368 return(-1); 369 } 370 371 /* 372 * get random name and create temporary scratch file, unlink name 373 * so it will get removed on exit 374 */ 375 memcpy(tempbase, _TFILE_BASE, sizeof(_TFILE_BASE)); 376 if ((ffd = mkstemp(tempfile)) < 0) { 377 syswarn(1, errno, "Unable to create temporary file: %s", 378 tempfile); 379 return(-1); 380 } 381 (void)unlink(tempfile); 382 383 return(0); 384 } 385 386 /* 387 * chk_ftime() 388 * looks up entry in file time hash table. If not found, the file is 389 * added to the hash table and the file named stored in the scratch file. 390 * If a file with the same name is found, the file times are compared and 391 * the most recent file time is retained. If the new file was younger (or 392 * was not in the database) the new file is selected for storage. 393 * Return: 394 * 0 if file should be added to the archive, 1 if it should be skipped, 395 * -1 on error 396 */ 397 398 #ifdef __STDC__ 399 int 400 chk_ftime(register ARCHD *arcn) 401 #else 402 int 403 chk_ftime(arcn) 404 register ARCHD *arcn; 405 #endif 406 { 407 register FTM *pt; 408 register int namelen; 409 register u_int indx; 410 char ckname[PAXPATHLEN+1]; 411 412 /* 413 * no info, go ahead and add to archive 414 */ 415 if (ftab == NULL) 416 return(0); 417 418 /* 419 * hash the pathname and look up in table 420 */ 421 namelen = arcn->nlen; 422 indx = st_hash(arcn->name, namelen, F_TAB_SZ); 423 if ((pt = ftab[indx]) != NULL) { 424 /* 425 * the hash chain is not empty, walk down looking for match 426 * only read up the path names if the lengths match, speeds 427 * up the search a lot 428 */ 429 while (pt != NULL) { 430 if (pt->namelen == namelen) { 431 /* 432 * potential match, have to read the name 433 * from the scratch file. 434 */ 435 if (lseek(ffd,pt->seek,SEEK_SET) != pt->seek) { 436 syswarn(1, errno, 437 "Failed ftime table seek"); 438 return(-1); 439 } 440 if (read(ffd, ckname, namelen) != namelen) { 441 syswarn(1, errno, 442 "Failed ftime table read"); 443 return(-1); 444 } 445 446 /* 447 * if the names match, we are done 448 */ 449 if (!strncmp(ckname, arcn->name, namelen)) 450 break; 451 } 452 453 /* 454 * try the next entry on the chain 455 */ 456 pt = pt->fow; 457 } 458 459 if (pt != NULL) { 460 /* 461 * found the file, compare the times, save the newer 462 */ 463 if (arcn->sb.st_mtime > pt->mtime) { 464 /* 465 * file is newer 466 */ 467 pt->mtime = arcn->sb.st_mtime; 468 return(0); 469 } 470 /* 471 * file is older 472 */ 473 return(1); 474 } 475 } 476 477 /* 478 * not in table, add it 479 */ 480 if ((pt = (FTM *)malloc(sizeof(FTM))) != NULL) { 481 /* 482 * add the name at the end of the scratch file, saving the 483 * offset. add the file to the head of the hash chain 484 */ 485 if ((pt->seek = lseek(ffd, (off_t)0, SEEK_END)) >= 0) { 486 if (write(ffd, arcn->name, namelen) == namelen) { 487 pt->mtime = arcn->sb.st_mtime; 488 pt->namelen = namelen; 489 pt->fow = ftab[indx]; 490 ftab[indx] = pt; 491 return(0); 492 } 493 syswarn(1, errno, "Failed write to file time table"); 494 } else 495 syswarn(1, errno, "Failed seek on file time table"); 496 } else 497 paxwarn(1, "File time table ran out of memory"); 498 499 if (pt != NULL) 500 (void)free((char *)pt); 501 return(-1); 502 } 503 504 /* 505 * Interactive rename table routines 506 * 507 * The interactive rename table keeps track of the new names that the user 508 * assigns to files from tty input. Since this map is unique for each file 509 * we must store it in case there is a reference to the file later in archive 510 * (a link). Otherwise we will be unable to find the file we know was 511 * extracted. The remapping of these files is stored in a memory based hash 512 * table (it is assumed since input must come from /dev/tty, it is unlikely to 513 * be a very large table). 514 */ 515 516 /* 517 * name_start() 518 * create the interactive rename table 519 * Return: 520 * 0 if successful, -1 otherwise 521 */ 522 523 #ifdef __STDC__ 524 int 525 name_start(void) 526 #else 527 int 528 name_start() 529 #endif 530 { 531 if (ntab != NULL) 532 return(0); 533 if ((ntab = (NAMT **)calloc(N_TAB_SZ, sizeof(NAMT *))) == NULL) { 534 paxwarn(1, "Cannot allocate memory for interactive rename table"); 535 return(-1); 536 } 537 return(0); 538 } 539 540 /* 541 * add_name() 542 * add the new name to old name mapping just created by the user. 543 * If an old name mapping is found (there may be duplicate names on an 544 * archive) only the most recent is kept. 545 * Return: 546 * 0 if added, -1 otherwise 547 */ 548 549 #ifdef __STDC__ 550 int 551 add_name(register char *oname, int onamelen, char *nname) 552 #else 553 int 554 add_name(oname, onamelen, nname) 555 register char *oname; 556 int onamelen; 557 char *nname; 558 #endif 559 { 560 register NAMT *pt; 561 register u_int indx; 562 563 if (ntab == NULL) { 564 /* 565 * should never happen 566 */ 567 paxwarn(0, "No interactive rename table, links may fail\n"); 568 return(0); 569 } 570 571 /* 572 * look to see if we have already mapped this file, if so we 573 * will update it 574 */ 575 indx = st_hash(oname, onamelen, N_TAB_SZ); 576 if ((pt = ntab[indx]) != NULL) { 577 /* 578 * look down the has chain for the file 579 */ 580 while ((pt != NULL) && (strcmp(oname, pt->oname) != 0)) 581 pt = pt->fow; 582 583 if (pt != NULL) { 584 /* 585 * found an old mapping, replace it with the new one 586 * the user just input (if it is different) 587 */ 588 if (strcmp(nname, pt->nname) == 0) 589 return(0); 590 591 (void)free((char *)pt->nname); 592 if ((pt->nname = strdup(nname)) == NULL) { 593 paxwarn(1, "Cannot update rename table"); 594 return(-1); 595 } 596 return(0); 597 } 598 } 599 600 /* 601 * this is a new mapping, add it to the table 602 */ 603 if ((pt = (NAMT *)malloc(sizeof(NAMT))) != NULL) { 604 if ((pt->oname = strdup(oname)) != NULL) { 605 if ((pt->nname = strdup(nname)) != NULL) { 606 pt->fow = ntab[indx]; 607 ntab[indx] = pt; 608 return(0); 609 } 610 (void)free((char *)pt->oname); 611 } 612 (void)free((char *)pt); 613 } 614 paxwarn(1, "Interactive rename table out of memory"); 615 return(-1); 616 } 617 618 /* 619 * sub_name() 620 * look up a link name to see if it points at a file that has been 621 * remapped by the user. If found, the link is adjusted to contain the 622 * new name (oname is the link to name) 623 */ 624 625 #ifdef __STDC__ 626 void 627 sub_name(register char *oname, int *onamelen, size_t onamesize) 628 #else 629 void 630 sub_name(oname, onamelen, onamesize) 631 register char *oname; 632 int *onamelen; 633 size_t onamesize; 634 #endif 635 { 636 register NAMT *pt; 637 register u_int indx; 638 639 if (ntab == NULL) 640 return; 641 /* 642 * look the name up in the hash table 643 */ 644 indx = st_hash(oname, *onamelen, N_TAB_SZ); 645 if ((pt = ntab[indx]) == NULL) 646 return; 647 648 while (pt != NULL) { 649 /* 650 * walk down the hash chain looking for a match 651 */ 652 if (strcmp(oname, pt->oname) == 0) { 653 /* 654 * found it, replace it with the new name 655 * and return (we know that oname has enough space) 656 */ 657 *onamelen = l_strncpy(oname, pt->nname, onamesize - 1); 658 oname[*onamelen] = '\0'; 659 return; 660 } 661 pt = pt->fow; 662 } 663 664 /* 665 * no match, just return 666 */ 667 return; 668 } 669 670 /* 671 * device/inode mapping table routines 672 * (used with formats that store device and inodes fields) 673 * 674 * device/inode mapping tables remap the device field in a archive header. The 675 * device/inode fields are used to determine when files are hard links to each 676 * other. However these values have very little meaning outside of that. This 677 * database is used to solve one of two different problems. 678 * 679 * 1) when files are appended to an archive, while the new files may have hard 680 * links to each other, you cannot determine if they have hard links to any 681 * file already stored on the archive from a prior run of pax. We must assume 682 * that these inode/device pairs are unique only within a SINGLE run of pax 683 * (which adds a set of files to an archive). So we have to make sure the 684 * inode/dev pairs we add each time are always unique. We do this by observing 685 * while the inode field is very dense, the use of the dev field is fairly 686 * sparse. Within each run of pax, we remap any device number of a new archive 687 * member that has a device number used in a prior run and already stored in a 688 * file on the archive. During the read phase of the append, we store the 689 * device numbers used and mark them to not be used by any file during the 690 * write phase. If during write we go to use one of those old device numbers, 691 * we remap it to a new value. 692 * 693 * 2) Often the fields in the archive header used to store these values are 694 * too small to store the entire value. The result is an inode or device value 695 * which can be truncated. This really can foul up an archive. With truncation 696 * we end up creating links between files that are really not links (after 697 * truncation the inodes are the same value). We address that by detecting 698 * truncation and forcing a remap of the device field to split truncated 699 * inodes away from each other. Each truncation creates a pattern of bits that 700 * are removed. We use this pattern of truncated bits to partition the inodes 701 * on a single device to many different devices (each one represented by the 702 * truncated bit pattern). All inodes on the same device that have the same 703 * truncation pattern are mapped to the same new device. Two inodes that 704 * truncate to the same value clearly will always have different truncation 705 * bit patterns, so they will be split from away each other. When we spot 706 * device truncation we remap the device number to a non truncated value. 707 * (for more info see table.h for the data structures involved). 708 */ 709 710 /* 711 * dev_start() 712 * create the device mapping table 713 * Return: 714 * 0 if successful, -1 otherwise 715 */ 716 717 #ifdef __STDC__ 718 int 719 dev_start(void) 720 #else 721 int 722 dev_start() 723 #endif 724 { 725 if (dtab != NULL) 726 return(0); 727 if ((dtab = (DEVT **)calloc(D_TAB_SZ, sizeof(DEVT *))) == NULL) { 728 paxwarn(1, "Cannot allocate memory for device mapping table"); 729 return(-1); 730 } 731 return(0); 732 } 733 734 /* 735 * add_dev() 736 * add a device number to the table. this will force the device to be 737 * remapped to a new value if it be used during a write phase. This 738 * function is called during the read phase of an append to prohibit the 739 * use of any device number already in the archive. 740 * Return: 741 * 0 if added ok, -1 otherwise 742 */ 743 744 #ifdef __STDC__ 745 int 746 add_dev(register ARCHD *arcn) 747 #else 748 int 749 add_dev(arcn) 750 register ARCHD *arcn; 751 #endif 752 { 753 if (chk_dev(arcn->sb.st_dev, 1) == NULL) 754 return(-1); 755 return(0); 756 } 757 758 /* 759 * chk_dev() 760 * check for a device value in the device table. If not found and the add 761 * flag is set, it is added. This does NOT assign any mapping values, just 762 * adds the device number as one that need to be remapped. If this device 763 * is already mapped, just return with a pointer to that entry. 764 * Return: 765 * pointer to the entry for this device in the device map table. Null 766 * if the add flag is not set and the device is not in the table (it is 767 * not been seen yet). If add is set and the device cannot be added, null 768 * is returned (indicates an error). 769 */ 770 771 #ifdef __STDC__ 772 static DEVT * 773 chk_dev(dev_t dev, int add) 774 #else 775 static DEVT * 776 chk_dev(dev, add) 777 dev_t dev; 778 int add; 779 #endif 780 { 781 register DEVT *pt; 782 register u_int indx; 783 784 if (dtab == NULL) 785 return(NULL); 786 /* 787 * look to see if this device is already in the table 788 */ 789 indx = ((unsigned)dev) % D_TAB_SZ; 790 if ((pt = dtab[indx]) != NULL) { 791 while ((pt != NULL) && (pt->dev != dev)) 792 pt = pt->fow; 793 794 /* 795 * found it, return a pointer to it 796 */ 797 if (pt != NULL) 798 return(pt); 799 } 800 801 /* 802 * not in table, we add it only if told to as this may just be a check 803 * to see if a device number is being used. 804 */ 805 if (add == 0) 806 return(NULL); 807 808 /* 809 * allocate a node for this device and add it to the front of the hash 810 * chain. Note we do not assign remaps values here, so the pt->list 811 * list must be NULL. 812 */ 813 if ((pt = (DEVT *)malloc(sizeof(DEVT))) == NULL) { 814 paxwarn(1, "Device map table out of memory"); 815 return(NULL); 816 } 817 pt->dev = dev; 818 pt->list = NULL; 819 pt->fow = dtab[indx]; 820 dtab[indx] = pt; 821 return(pt); 822 } 823 /* 824 * map_dev() 825 * given an inode and device storage mask (the mask has a 1 for each bit 826 * the archive format is able to store in a header), we check for inode 827 * and device truncation and remap the device as required. Device mapping 828 * can also occur when during the read phase of append a device number was 829 * seen (and was marked as do not use during the write phase). WE ASSUME 830 * that unsigned longs are the same size or bigger than the fields used 831 * for ino_t and dev_t. If not the types will have to be changed. 832 * Return: 833 * 0 if all ok, -1 otherwise. 834 */ 835 836 #ifdef __STDC__ 837 int 838 map_dev(register ARCHD *arcn, u_long dev_mask, u_long ino_mask) 839 #else 840 int 841 map_dev(arcn, dev_mask, ino_mask) 842 register ARCHD *arcn; 843 u_long dev_mask; 844 u_long ino_mask; 845 #endif 846 { 847 register DEVT *pt; 848 register DLIST *dpt; 849 static dev_t lastdev = 0; /* next device number to try */ 850 int trc_ino = 0; 851 int trc_dev = 0; 852 ino_t trunc_bits = 0; 853 ino_t nino; 854 855 if (dtab == NULL) 856 return(0); 857 /* 858 * check for device and inode truncation, and extract the truncated 859 * bit pattern. 860 */ 861 if ((arcn->sb.st_dev & (dev_t)dev_mask) != arcn->sb.st_dev) 862 ++trc_dev; 863 if ((nino = arcn->sb.st_ino & (ino_t)ino_mask) != arcn->sb.st_ino) { 864 ++trc_ino; 865 trunc_bits = arcn->sb.st_ino & (ino_t)(~ino_mask); 866 } 867 868 /* 869 * see if this device is already being mapped, look up the device 870 * then find the truncation bit pattern which applies 871 */ 872 if ((pt = chk_dev(arcn->sb.st_dev, 0)) != NULL) { 873 /* 874 * this device is already marked to be remapped 875 */ 876 for (dpt = pt->list; dpt != NULL; dpt = dpt->fow) 877 if (dpt->trunc_bits == trunc_bits) 878 break; 879 880 if (dpt != NULL) { 881 /* 882 * we are being remapped for this device and pattern 883 * change the device number to be stored and return 884 */ 885 arcn->sb.st_dev = dpt->dev; 886 arcn->sb.st_ino = nino; 887 return(0); 888 } 889 } else { 890 /* 891 * this device is not being remapped YET. if we do not have any 892 * form of truncation, we do not need a remap 893 */ 894 if (!trc_ino && !trc_dev) 895 return(0); 896 897 /* 898 * we have truncation, have to add this as a device to remap 899 */ 900 if ((pt = chk_dev(arcn->sb.st_dev, 1)) == NULL) 901 goto bad; 902 903 /* 904 * if we just have a truncated inode, we have to make sure that 905 * all future inodes that do not truncate (they have the 906 * truncation pattern of all 0's) continue to map to the same 907 * device number. We probably have already written inodes with 908 * this device number to the archive with the truncation 909 * pattern of all 0's. So we add the mapping for all 0's to the 910 * same device number. 911 */ 912 if (!trc_dev && (trunc_bits != 0)) { 913 if ((dpt = (DLIST *)malloc(sizeof(DLIST))) == NULL) 914 goto bad; 915 dpt->trunc_bits = 0; 916 dpt->dev = arcn->sb.st_dev; 917 dpt->fow = pt->list; 918 pt->list = dpt; 919 } 920 } 921 922 /* 923 * look for a device number not being used. We must watch for wrap 924 * around on lastdev (so we do not get stuck looking forever!) 925 */ 926 while (++lastdev > 0) { 927 if (chk_dev(lastdev, 0) != NULL) 928 continue; 929 /* 930 * found an unused value. If we have reached truncation point 931 * for this format we are hosed, so we give up. Otherwise we 932 * mark it as being used. 933 */ 934 if (((lastdev & ((dev_t)dev_mask)) != lastdev) || 935 (chk_dev(lastdev, 1) == NULL)) 936 goto bad; 937 break; 938 } 939 940 if ((lastdev <= 0) || ((dpt = (DLIST *)malloc(sizeof(DLIST))) == NULL)) 941 goto bad; 942 943 /* 944 * got a new device number, store it under this truncation pattern. 945 * change the device number this file is being stored with. 946 */ 947 dpt->trunc_bits = trunc_bits; 948 dpt->dev = lastdev; 949 dpt->fow = pt->list; 950 pt->list = dpt; 951 arcn->sb.st_dev = lastdev; 952 arcn->sb.st_ino = nino; 953 return(0); 954 955 bad: 956 paxwarn(1, "Unable to fix truncated inode/device field when storing %s", 957 arcn->name); 958 paxwarn(0, "Archive may create improper hard links when extracted"); 959 return(0); 960 } 961 962 /* 963 * directory access/mod time reset table routines (for directories READ by pax) 964 * 965 * The pax -t flag requires that access times of archive files to be the same 966 * before being read by pax. For regular files, access time is restored after 967 * the file has been copied. This database provides the same functionality for 968 * directories read during file tree traversal. Restoring directory access time 969 * is more complex than files since directories may be read several times until 970 * all the descendants in their subtree are visited by fts. Directory access 971 * and modification times are stored during the fts pre-order visit (done 972 * before any descendants in the subtree is visited) and restored after the 973 * fts post-order visit (after all the descendants have been visited). In the 974 * case of premature exit from a subtree (like from the effects of -n), any 975 * directory entries left in this database are reset during final cleanup 976 * operations of pax. Entries are hashed by inode number for fast lookup. 977 */ 978 979 /* 980 * atdir_start() 981 * create the directory access time database for directories READ by pax. 982 * Return: 983 * 0 is created ok, -1 otherwise. 984 */ 985 986 #ifdef __STDC__ 987 int 988 atdir_start(void) 989 #else 990 int 991 atdir_start() 992 #endif 993 { 994 if (atab != NULL) 995 return(0); 996 if ((atab = (ATDIR **)calloc(A_TAB_SZ, sizeof(ATDIR *))) == NULL) { 997 paxwarn(1,"Cannot allocate space for directory access time table"); 998 return(-1); 999 } 1000 return(0); 1001 } 1002 1003 1004 /* 1005 * atdir_end() 1006 * walk through the directory access time table and reset the access time 1007 * of any directory who still has an entry left in the database. These 1008 * entries are for directories READ by pax 1009 */ 1010 1011 #ifdef __STDC__ 1012 void 1013 atdir_end(void) 1014 #else 1015 void 1016 atdir_end() 1017 #endif 1018 { 1019 register ATDIR *pt; 1020 register int i; 1021 1022 if (atab == NULL) 1023 return; 1024 /* 1025 * for each non-empty hash table entry reset all the directories 1026 * chained there. 1027 */ 1028 for (i = 0; i < A_TAB_SZ; ++i) { 1029 if ((pt = atab[i]) == NULL) 1030 continue; 1031 /* 1032 * remember to force the times, set_ftime() looks at pmtime 1033 * and patime, which only applies to things CREATED by pax, 1034 * not read by pax. Read time reset is controlled by -t. 1035 */ 1036 for (; pt != NULL; pt = pt->fow) 1037 set_ftime(pt->name, pt->mtime, pt->atime, 1); 1038 } 1039 } 1040 1041 /* 1042 * add_atdir() 1043 * add a directory to the directory access time table. Table is hashed 1044 * and chained by inode number. This is for directories READ by pax 1045 */ 1046 1047 #ifdef __STDC__ 1048 void 1049 add_atdir(char *fname, dev_t dev, ino_t ino, time_t mtime, time_t atime) 1050 #else 1051 void 1052 add_atdir(fname, dev, ino, mtime, atime) 1053 char *fname; 1054 dev_t dev; 1055 ino_t ino; 1056 time_t mtime; 1057 time_t atime; 1058 #endif 1059 { 1060 register ATDIR *pt; 1061 register u_int indx; 1062 1063 if (atab == NULL) 1064 return; 1065 1066 /* 1067 * make sure this directory is not already in the table, if so just 1068 * return (the older entry always has the correct time). The only 1069 * way this will happen is when the same subtree can be traversed by 1070 * different args to pax and the -n option is aborting fts out of a 1071 * subtree before all the post-order visits have been made). 1072 */ 1073 indx = ((unsigned)ino) % A_TAB_SZ; 1074 if ((pt = atab[indx]) != NULL) { 1075 while (pt != NULL) { 1076 if ((pt->ino == ino) && (pt->dev == dev)) 1077 break; 1078 pt = pt->fow; 1079 } 1080 1081 /* 1082 * oops, already there. Leave it alone. 1083 */ 1084 if (pt != NULL) 1085 return; 1086 } 1087 1088 /* 1089 * add it to the front of the hash chain 1090 */ 1091 if ((pt = (ATDIR *)malloc(sizeof(ATDIR))) != NULL) { 1092 if ((pt->name = strdup(fname)) != NULL) { 1093 pt->dev = dev; 1094 pt->ino = ino; 1095 pt->mtime = mtime; 1096 pt->atime = atime; 1097 pt->fow = atab[indx]; 1098 atab[indx] = pt; 1099 return; 1100 } 1101 (void)free((char *)pt); 1102 } 1103 1104 paxwarn(1, "Directory access time reset table ran out of memory"); 1105 return; 1106 } 1107 1108 /* 1109 * get_atdir() 1110 * look up a directory by inode and device number to obtain the access 1111 * and modification time you want to set to. If found, the modification 1112 * and access time parameters are set and the entry is removed from the 1113 * table (as it is no longer needed). These are for directories READ by 1114 * pax 1115 * Return: 1116 * 0 if found, -1 if not found. 1117 */ 1118 1119 #ifdef __STDC__ 1120 int 1121 get_atdir(dev_t dev, ino_t ino, time_t *mtime, time_t *atime) 1122 #else 1123 int 1124 get_atdir(dev, ino, mtime, atime) 1125 dev_t dev; 1126 ino_t ino; 1127 time_t *mtime; 1128 time_t *atime; 1129 #endif 1130 { 1131 register ATDIR *pt; 1132 register ATDIR **ppt; 1133 register u_int indx; 1134 1135 if (atab == NULL) 1136 return(-1); 1137 /* 1138 * hash by inode and search the chain for an inode and device match 1139 */ 1140 indx = ((unsigned)ino) % A_TAB_SZ; 1141 if ((pt = atab[indx]) == NULL) 1142 return(-1); 1143 1144 ppt = &(atab[indx]); 1145 while (pt != NULL) { 1146 if ((pt->ino == ino) && (pt->dev == dev)) 1147 break; 1148 /* 1149 * no match, go to next one 1150 */ 1151 ppt = &(pt->fow); 1152 pt = pt->fow; 1153 } 1154 1155 /* 1156 * return if we did not find it. 1157 */ 1158 if (pt == NULL) 1159 return(-1); 1160 1161 /* 1162 * found it. return the times and remove the entry from the table. 1163 */ 1164 *ppt = pt->fow; 1165 *mtime = pt->mtime; 1166 *atime = pt->atime; 1167 (void)free((char *)pt->name); 1168 (void)free((char *)pt); 1169 return(0); 1170 } 1171 1172 /* 1173 * directory access mode and time storage routines (for directories CREATED 1174 * by pax). 1175 * 1176 * Pax requires that extracted directories, by default, have their access/mod 1177 * times and permissions set to the values specified in the archive. During the 1178 * actions of extracting (and creating the destination subtree during -rw copy) 1179 * directories extracted may be modified after being created. Even worse is 1180 * that these directories may have been created with file permissions which 1181 * prohibits any descendants of these directories from being extracted. When 1182 * directories are created by pax, access rights may be added to permit the 1183 * creation of files in their subtree. Every time pax creates a directory, the 1184 * times and file permissions specified by the archive are stored. After all 1185 * files have been extracted (or copied), these directories have their times 1186 * and file modes reset to the stored values. The directory info is restored in 1187 * reverse order as entries were added to the data file from root to leaf. To 1188 * restore atime properly, we must go backwards. The data file consists of 1189 * records with two parts, the file name followed by a DIRDATA trailer. The 1190 * fixed sized trailer contains the size of the name plus the off_t location in 1191 * the file. To restore we work backwards through the file reading the trailer 1192 * then the file name. 1193 */ 1194 1195 /* 1196 * dir_start() 1197 * set up the directory time and file mode storage for directories CREATED 1198 * by pax. 1199 * Return: 1200 * 0 if ok, -1 otherwise 1201 */ 1202 1203 #ifdef __STDC__ 1204 int 1205 dir_start(void) 1206 #else 1207 int 1208 dir_start() 1209 #endif 1210 { 1211 1212 if (dirfd != -1) 1213 return(0); 1214 1215 /* 1216 * unlink the file so it goes away at termination by itself 1217 */ 1218 memcpy(tempbase, _TFILE_BASE, sizeof(_TFILE_BASE)); 1219 if ((dirfd = mkstemp(tempfile)) >= 0) { 1220 (void)unlink(tempfile); 1221 return(0); 1222 } 1223 paxwarn(1, "Unable to create temporary file for directory times: %s", 1224 tempfile); 1225 return(-1); 1226 } 1227 1228 /* 1229 * add_dir() 1230 * add the mode and times for a newly CREATED directory 1231 * name is name of the directory, psb the stat buffer with the data in it, 1232 * frc_mode is a flag that says whether to force the setting of the mode 1233 * (ignoring the user set values for preserving file mode). Frc_mode is 1234 * for the case where we created a file and found that the resulting 1235 * directory was not writeable and the user asked for file modes to NOT 1236 * be preserved. (we have to preserve what was created by default, so we 1237 * have to force the setting at the end. this is stated explicitly in the 1238 * pax spec) 1239 */ 1240 1241 #ifdef __STDC__ 1242 void 1243 add_dir(char *name, int nlen, struct stat *psb, int frc_mode) 1244 #else 1245 void 1246 add_dir(name, nlen, psb, frc_mode) 1247 char *name; 1248 int nlen; 1249 struct stat *psb; 1250 int frc_mode; 1251 #endif 1252 { 1253 DIRDATA dblk; 1254 1255 if (dirfd < 0) 1256 return; 1257 1258 /* 1259 * get current position (where file name will start) so we can store it 1260 * in the trailer 1261 */ 1262 if ((dblk.npos = lseek(dirfd, 0L, SEEK_CUR)) < 0) { 1263 paxwarn(1,"Unable to store mode and times for directory: %s",name); 1264 return; 1265 } 1266 1267 /* 1268 * write the file name followed by the trailer 1269 */ 1270 dblk.nlen = nlen + 1; 1271 dblk.mode = psb->st_mode & 0xffff; 1272 dblk.mtime = psb->st_mtime; 1273 dblk.atime = psb->st_atime; 1274 dblk.frc_mode = frc_mode; 1275 if ((write(dirfd, name, dblk.nlen) == dblk.nlen) && 1276 (write(dirfd, (char *)&dblk, sizeof(dblk)) == sizeof(dblk))) { 1277 ++dircnt; 1278 return; 1279 } 1280 1281 paxwarn(1,"Unable to store mode and times for created directory: %s",name); 1282 return; 1283 } 1284 1285 /* 1286 * proc_dir() 1287 * process all file modes and times stored for directories CREATED 1288 * by pax 1289 */ 1290 1291 #ifdef __STDC__ 1292 void 1293 proc_dir(void) 1294 #else 1295 void 1296 proc_dir() 1297 #endif 1298 { 1299 char name[PAXPATHLEN+1]; 1300 DIRDATA dblk; 1301 u_long cnt; 1302 1303 if (dirfd < 0) 1304 return; 1305 /* 1306 * read backwards through the file and process each directory 1307 */ 1308 for (cnt = 0; cnt < dircnt; ++cnt) { 1309 /* 1310 * read the trailer, then the file name, if this fails 1311 * just give up. 1312 */ 1313 if (lseek(dirfd, -((off_t)sizeof(dblk)), SEEK_CUR) < 0) 1314 break; 1315 if (read(dirfd,(char *)&dblk, sizeof(dblk)) != sizeof(dblk)) 1316 break; 1317 if (lseek(dirfd, dblk.npos, SEEK_SET) < 0) 1318 break; 1319 if (read(dirfd, name, dblk.nlen) != dblk.nlen) 1320 break; 1321 if (lseek(dirfd, dblk.npos, SEEK_SET) < 0) 1322 break; 1323 1324 /* 1325 * frc_mode set, make sure we set the file modes even if 1326 * the user didn't ask for it (see file_subs.c for more info) 1327 */ 1328 if (pmode || dblk.frc_mode) 1329 set_pmode(name, dblk.mode); 1330 if (patime || pmtime) 1331 set_ftime(name, dblk.mtime, dblk.atime, 0); 1332 } 1333 1334 (void)close(dirfd); 1335 dirfd = -1; 1336 if (cnt != dircnt) 1337 paxwarn(1,"Unable to set mode and times for created directories"); 1338 return; 1339 } 1340 1341 /* 1342 * database independent routines 1343 */ 1344 1345 /* 1346 * st_hash() 1347 * hashes filenames to a u_int for hashing into a table. Looks at the tail 1348 * end of file, as this provides far better distribution than any other 1349 * part of the name. For performance reasons we only care about the last 1350 * MAXKEYLEN chars (should be at LEAST large enough to pick off the file 1351 * name). Was tested on 500,000 name file tree traversal from the root 1352 * and gave almost a perfectly uniform distribution of keys when used with 1353 * prime sized tables (MAXKEYLEN was 128 in test). Hashes (sizeof int) 1354 * chars at a time and pads with 0 for last addition. 1355 * Return: 1356 * the hash value of the string MOD (%) the table size. 1357 */ 1358 1359 #ifdef __STDC__ 1360 u_int 1361 st_hash(char *name, int len, int tabsz) 1362 #else 1363 u_int 1364 st_hash(name, len, tabsz) 1365 char *name; 1366 int len; 1367 int tabsz; 1368 #endif 1369 { 1370 register char *pt; 1371 register char *dest; 1372 register char *end; 1373 register int i; 1374 register u_int key = 0; 1375 register int steps; 1376 register int res; 1377 u_int val; 1378 1379 /* 1380 * only look at the tail up to MAXKEYLEN, we do not need to waste 1381 * time here (remember these are pathnames, the tail is what will 1382 * spread out the keys) 1383 */ 1384 if (len > MAXKEYLEN) { 1385 pt = &(name[len - MAXKEYLEN]); 1386 len = MAXKEYLEN; 1387 } else 1388 pt = name; 1389 1390 /* 1391 * calculate the number of u_int size steps in the string and if 1392 * there is a runt to deal with 1393 */ 1394 steps = len/sizeof(u_int); 1395 res = len % sizeof(u_int); 1396 1397 /* 1398 * add up the value of the string in unsigned integer sized pieces 1399 * too bad we cannot have unsigned int aligned strings, then we 1400 * could avoid the expensive copy. 1401 */ 1402 for (i = 0; i < steps; ++i) { 1403 end = pt + sizeof(u_int); 1404 dest = (char *)&val; 1405 while (pt < end) 1406 *dest++ = *pt++; 1407 key += val; 1408 } 1409 1410 /* 1411 * add in the runt padded with zero to the right 1412 */ 1413 if (res) { 1414 val = 0; 1415 end = pt + res; 1416 dest = (char *)&val; 1417 while (pt < end) 1418 *dest++ = *pt++; 1419 key += val; 1420 } 1421 1422 /* 1423 * return the result mod the table size 1424 */ 1425 return(key % tabsz); 1426 } 1427