tables.c revision 26363
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 * $Id: tables.c,v 1.7 1997/02/22 14:04:43 peter Exp $ 38 */ 39 40#ifndef lint 41static char const sccsid[] = "@(#)tables.c 8.1 (Berkeley) 5/31/93"; 42#endif /* not lint */ 43 44#include <sys/types.h> 45#include <sys/time.h> 46#include <sys/stat.h> 47#include <sys/param.h> 48#include <sys/fcntl.h> 49#include <stdio.h> 50#include <string.h> 51#include <unistd.h> 52#include <errno.h> 53#include <stdlib.h> 54#include "pax.h" 55#include "tables.h" 56#include "extern.h" 57 58/* 59 * Routines for controlling the contents of all the different databases pax 60 * keeps. Tables are dynamically created only when they are needed. The 61 * goal was speed and the ability to work with HUGE archives. The databases 62 * were kept simple, but do have complex rules for when the contents change. 63 * As of this writing, the posix library functions were more complex than 64 * needed for this application (pax databases have very short lifetimes and 65 * do not survive after pax is finished). Pax is required to handle very 66 * large archives. These database routines carefully combine memory usage and 67 * temporary file storage in ways which will not significantly impact runtime 68 * performance while allowing the largest possible archives to be handled. 69 * Trying to force the fit to the posix databases routines was not considered 70 * time well spent. 71 */ 72 73static HRDLNK **ltab = NULL; /* hard link table for detecting hard links */ 74static FTM **ftab = NULL; /* file time table for updating arch */ 75static NAMT **ntab = NULL; /* interactive rename storage table */ 76static DEVT **dtab = NULL; /* device/inode mapping tables */ 77static ATDIR **atab = NULL; /* file tree directory time reset table */ 78static int dirfd = -1; /* storage for setting created dir time/mode */ 79static u_long dircnt; /* entries in dir time/mode storage */ 80static int ffd = -1; /* tmp file for file time table name storage */ 81 82static DEVT *chk_dev __P((dev_t, int)); 83 84/* 85 * hard link table routines 86 * 87 * The hard link table tries to detect hard links to files using the device and 88 * inode values. We do this when writing an archive, so we can tell the format 89 * write routine that this file is a hard link to another file. The format 90 * write routine then can store this file in whatever way it wants (as a hard 91 * link if the format supports that like tar, or ignore this info like cpio). 92 * (Actually a field in the format driver table tells us if the format wants 93 * hard link info. if not, we do not waste time looking for them). We also use 94 * the same table when reading an archive. In that situation, this table is 95 * used by the format read routine to detect hard links from stored dev and 96 * inode numbers (like cpio). This will allow pax to create a link when one 97 * can be detected by the archive format. 98 */ 99 100/* 101 * lnk_start 102 * Creates the hard link table. 103 * Return: 104 * 0 if created, -1 if failure 105 */ 106 107#if __STDC__ 108int 109lnk_start(void) 110#else 111int 112lnk_start() 113#endif 114{ 115 if (ltab != NULL) 116 return(0); 117 if ((ltab = (HRDLNK **)calloc(L_TAB_SZ, sizeof(HRDLNK *))) == NULL) { 118 warn(1, "Cannot allocate memory for hard link table"); 119 return(-1); 120 } 121 return(0); 122} 123 124/* 125 * chk_lnk() 126 * Looks up entry in hard link hash table. If found, it copies the name 127 * of the file it is linked to (we already saw that file) into ln_name. 128 * lnkcnt is decremented and if goes to 1 the node is deleted from the 129 * database. (We have seen all the links to this file). If not found, 130 * we add the file to the database if it has the potential for having 131 * hard links to other files we may process (it has a link count > 1) 132 * Return: 133 * if found returns 1; if not found returns 0; -1 on error 134 */ 135 136#if __STDC__ 137int 138chk_lnk(register ARCHD *arcn) 139#else 140int 141chk_lnk(arcn) 142 register ARCHD *arcn; 143#endif 144{ 145 register HRDLNK *pt; 146 register HRDLNK **ppt; 147 register u_int indx; 148 149 if (ltab == NULL) 150 return(-1); 151 /* 152 * ignore those nodes that cannot have hard links 153 */ 154 if ((arcn->type == PAX_DIR) || (arcn->sb.st_nlink <= 1)) 155 return(0); 156 157 /* 158 * hash inode number and look for this file 159 */ 160 indx = ((unsigned)arcn->sb.st_ino) % L_TAB_SZ; 161 if ((pt = ltab[indx]) != NULL) { 162 /* 163 * it's hash chain in not empty, walk down looking for it 164 */ 165 ppt = &(ltab[indx]); 166 while (pt != NULL) { 167 if ((pt->ino == arcn->sb.st_ino) && 168 (pt->dev == arcn->sb.st_dev)) 169 break; 170 ppt = &(pt->fow); 171 pt = pt->fow; 172 } 173 174 if (pt != NULL) { 175 /* 176 * found a link. set the node type and copy in the 177 * name of the file it is to link to. we need to 178 * handle hardlinks to regular files differently than 179 * other links. 180 */ 181 arcn->ln_nlen = l_strncpy(arcn->ln_name, pt->name, 182 PAXPATHLEN+1); 183 arcn->ln_name[PAXPATHLEN] = '\0'; 184 if (arcn->type == PAX_REG) 185 arcn->type = PAX_HRG; 186 else 187 arcn->type = PAX_HLK; 188 189 /* 190 * if we have found all the links to this file, remove 191 * it from the database 192 */ 193 if (--pt->nlink <= 1) { 194 *ppt = pt->fow; 195 (void)free((char *)pt->name); 196 (void)free((char *)pt); 197 } 198 return(1); 199 } 200 } 201 202 /* 203 * we never saw this file before. It has links so we add it to the 204 * front of this hash chain 205 */ 206 if ((pt = (HRDLNK *)malloc(sizeof(HRDLNK))) != NULL) { 207 if ((pt->name = strdup(arcn->name)) != NULL) { 208 pt->dev = arcn->sb.st_dev; 209 pt->ino = arcn->sb.st_ino; 210 pt->nlink = arcn->sb.st_nlink; 211 pt->fow = ltab[indx]; 212 ltab[indx] = pt; 213 return(0); 214 } 215 (void)free((char *)pt); 216 } 217 218 warn(1, "Hard link table out of memory"); 219 return(-1); 220} 221 222/* 223 * purg_lnk 224 * remove reference for a file that we may have added to the data base as 225 * a potential source for hard links. We ended up not using the file, so 226 * we do not want to accidently point another file at it later on. 227 */ 228 229#if __STDC__ 230void 231purg_lnk(register ARCHD *arcn) 232#else 233void 234purg_lnk(arcn) 235 register ARCHD *arcn; 236#endif 237{ 238 register HRDLNK *pt; 239 register HRDLNK **ppt; 240 register u_int indx; 241 242 if (ltab == NULL) 243 return; 244 /* 245 * do not bother to look if it could not be in the database 246 */ 247 if ((arcn->sb.st_nlink <= 1) || (arcn->type == PAX_DIR) || 248 (arcn->type == PAX_HLK) || (arcn->type == PAX_HRG)) 249 return; 250 251 /* 252 * find the hash chain for this inode value, if empty return 253 */ 254 indx = ((unsigned)arcn->sb.st_ino) % L_TAB_SZ; 255 if ((pt = ltab[indx]) == NULL) 256 return; 257 258 /* 259 * walk down the list looking for the inode/dev pair, unlink and 260 * free if found 261 */ 262 ppt = &(ltab[indx]); 263 while (pt != NULL) { 264 if ((pt->ino == arcn->sb.st_ino) && 265 (pt->dev == arcn->sb.st_dev)) 266 break; 267 ppt = &(pt->fow); 268 pt = pt->fow; 269 } 270 if (pt == NULL) 271 return; 272 273 /* 274 * remove and free it 275 */ 276 *ppt = pt->fow; 277 (void)free((char *)pt->name); 278 (void)free((char *)pt); 279} 280 281/* 282 * lnk_end() 283 * pull apart a existing link table so we can reuse it. We do this between 284 * read and write phases of append with update. (The format may have 285 * used the link table, and we need to start with a fresh table for the 286 * write phase 287 */ 288 289#if __STDC__ 290void 291lnk_end(void) 292#else 293void 294lnk_end() 295#endif 296{ 297 register int i; 298 register HRDLNK *pt; 299 register HRDLNK *ppt; 300 301 if (ltab == NULL) 302 return; 303 304 for (i = 0; i < L_TAB_SZ; ++i) { 305 if (ltab[i] == NULL) 306 continue; 307 pt = ltab[i]; 308 ltab[i] = NULL; 309 310 /* 311 * free up each entry on this chain 312 */ 313 while (pt != NULL) { 314 ppt = pt; 315 pt = ppt->fow; 316 (void)free((char *)ppt->name); 317 (void)free((char *)ppt); 318 } 319 } 320 return; 321} 322 323/* 324 * modification time table routines 325 * 326 * The modification time table keeps track of last modification times for all 327 * files stored in an archive during a write phase when -u is set. We only 328 * add a file to the archive if it is newer than a file with the same name 329 * already stored on the archive (if there is no other file with the same 330 * name on the archive it is added). This applies to writes and appends. 331 * An append with an -u must read the archive and store the modification time 332 * for every file on that archive before starting the write phase. It is clear 333 * that this is one HUGE database. To save memory space, the actual file names 334 * are stored in a scatch file and indexed by an in memory hash table. The 335 * hash table is indexed by hashing the file path. The nodes in the table store 336 * the length of the filename and the lseek offset within the scratch file 337 * where the actual name is stored. Since there are never any deletions to this 338 * table, fragmentation of the scratch file is never a issue. Lookups seem to 339 * not exhibit any locality at all (files in the database are rarely 340 * looked up more than once...). So caching is just a waste of memory. The 341 * only limitation is the amount of scatch file space available to store the 342 * path names. 343 */ 344 345/* 346 * ftime_start() 347 * create the file time hash table and open for read/write the scratch 348 * file. (after created it is unlinked, so when we exit we leave 349 * no witnesses). 350 * Return: 351 * 0 if the table and file was created ok, -1 otherwise 352 */ 353 354#if __STDC__ 355int 356ftime_start(void) 357#else 358int 359ftime_start() 360#endif 361{ 362 char *pt; 363 364 if (ftab != NULL) 365 return(0); 366 if ((ftab = (FTM **)calloc(F_TAB_SZ, sizeof(FTM *))) == NULL) { 367 warn(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 if ((pt = tempnam((char *)NULL, (char *)NULL)) == NULL) 376 return(-1); 377 (void)unlink(pt); 378 379 if ((ffd = open(pt, O_RDWR | O_CREAT, S_IRWXU)) < 0) { 380 syswarn(1, errno, "Unable to open temporary file: %s", pt); 381 return(-1); 382 } 383 384 (void)unlink(pt); 385 return(0); 386} 387 388/* 389 * chk_ftime() 390 * looks up entry in file time hash table. If not found, the file is 391 * added to the hash table and the file named stored in the scratch file. 392 * If a file with the same name is found, the file times are compared and 393 * the most recent file time is retained. If the new file was younger (or 394 * was not in the database) the new file is selected for storage. 395 * Return: 396 * 0 if file should be added to the archive, 1 if it should be skipped, 397 * -1 on error 398 */ 399 400#if __STDC__ 401int 402chk_ftime(register ARCHD *arcn) 403#else 404int 405chk_ftime(arcn) 406 register ARCHD *arcn; 407#endif 408{ 409 register FTM *pt; 410 register int namelen; 411 register u_int indx; 412 char ckname[PAXPATHLEN+1]; 413 414 /* 415 * no info, go ahead and add to archive 416 */ 417 if (ftab == NULL) 418 return(0); 419 420 /* 421 * hash the pathname and look up in table 422 */ 423 namelen = arcn->nlen; 424 indx = st_hash(arcn->name, namelen, F_TAB_SZ); 425 if ((pt = ftab[indx]) != NULL) { 426 /* 427 * the hash chain is not empty, walk down looking for match 428 * only read up the path names if the lengths match, speeds 429 * up the search a lot 430 */ 431 while (pt != NULL) { 432 if (pt->namelen == namelen) { 433 /* 434 * potential match, have to read the name 435 * from the scratch file. 436 */ 437 if (lseek(ffd,pt->seek,SEEK_SET) != pt->seek) { 438 syswarn(1, errno, 439 "Failed ftime table seek"); 440 return(-1); 441 } 442 if (read(ffd, ckname, namelen) != namelen) { 443 syswarn(1, errno, 444 "Failed ftime table read"); 445 return(-1); 446 } 447 448 /* 449 * if the names match, we are done 450 */ 451 if (!strncmp(ckname, arcn->name, namelen)) 452 break; 453 } 454 455 /* 456 * try the next entry on the chain 457 */ 458 pt = pt->fow; 459 } 460 461 if (pt != NULL) { 462 /* 463 * found the file, compare the times, save the newer 464 */ 465 if (arcn->sb.st_mtime > pt->mtime) { 466 /* 467 * file is newer 468 */ 469 pt->mtime = arcn->sb.st_mtime; 470 return(0); 471 } 472 /* 473 * file is older 474 */ 475 return(1); 476 } 477 } 478 479 /* 480 * not in table, add it 481 */ 482 if ((pt = (FTM *)malloc(sizeof(FTM))) != NULL) { 483 /* 484 * add the name at the end of the scratch file, saving the 485 * offset. add the file to the head of the hash chain 486 */ 487 if ((pt->seek = lseek(ffd, (off_t)0, SEEK_END)) >= 0) { 488 if (write(ffd, arcn->name, namelen) == namelen) { 489 pt->mtime = arcn->sb.st_mtime; 490 pt->namelen = namelen; 491 pt->fow = ftab[indx]; 492 ftab[indx] = pt; 493 return(0); 494 } 495 syswarn(1, errno, "Failed write to file time table"); 496 } else 497 syswarn(1, errno, "Failed seek on file time table"); 498 } else 499 warn(1, "File time table ran out of memory"); 500 501 if (pt != NULL) 502 (void)free((char *)pt); 503 return(-1); 504} 505 506/* 507 * Interactive rename table routines 508 * 509 * The interactive rename table keeps track of the new names that the user 510 * assignes to files from tty input. Since this map is unique for each file 511 * we must store it in case there is a reference to the file later in archive 512 * (a link). Otherwise we will be unable to find the file we know was 513 * extracted. The remapping of these files is stored in a memory based hash 514 * table (it is assumed since input must come from /dev/tty, it is unlikely to 515 * be a very large table). 516 */ 517 518/* 519 * name_start() 520 * create the interactive rename table 521 * Return: 522 * 0 if successful, -1 otherwise 523 */ 524 525#if __STDC__ 526int 527name_start(void) 528#else 529int 530name_start() 531#endif 532{ 533 if (ntab != NULL) 534 return(0); 535 if ((ntab = (NAMT **)calloc(N_TAB_SZ, sizeof(NAMT *))) == NULL) { 536 warn(1, "Cannot allocate memory for interactive rename table"); 537 return(-1); 538 } 539 return(0); 540} 541 542/* 543 * add_name() 544 * add the new name to old name mapping just created by the user. 545 * If an old name mapping is found (there may be duplicate names on an 546 * archive) only the most recent is kept. 547 * Return: 548 * 0 if added, -1 otherwise 549 */ 550 551#if __STDC__ 552int 553add_name(register char *oname, int onamelen, char *nname) 554#else 555int 556add_name(oname, onamelen, nname) 557 register char *oname; 558 int onamelen; 559 char *nname; 560#endif 561{ 562 register NAMT *pt; 563 register u_int indx; 564 565 if (ntab == NULL) { 566 /* 567 * should never happen 568 */ 569 warn(0, "No interactive rename table, links may fail\n"); 570 return(0); 571 } 572 573 /* 574 * look to see if we have already mapped this file, if so we 575 * will update it 576 */ 577 indx = st_hash(oname, onamelen, N_TAB_SZ); 578 if ((pt = ntab[indx]) != NULL) { 579 /* 580 * look down the has chain for the file 581 */ 582 while ((pt != NULL) && (strcmp(oname, pt->oname) != 0)) 583 pt = pt->fow; 584 585 if (pt != NULL) { 586 /* 587 * found an old mapping, replace it with the new one 588 * the user just input (if it is different) 589 */ 590 if (strcmp(nname, pt->nname) == 0) 591 return(0); 592 593 (void)free((char *)pt->nname); 594 if ((pt->nname = strdup(nname)) == NULL) { 595 warn(1, "Cannot update rename table"); 596 return(-1); 597 } 598 return(0); 599 } 600 } 601 602 /* 603 * this is a new mapping, add it to the table 604 */ 605 if ((pt = (NAMT *)malloc(sizeof(NAMT))) != NULL) { 606 if ((pt->oname = strdup(oname)) != NULL) { 607 if ((pt->nname = strdup(nname)) != NULL) { 608 pt->fow = ntab[indx]; 609 ntab[indx] = pt; 610 return(0); 611 } 612 (void)free((char *)pt->oname); 613 } 614 (void)free((char *)pt); 615 } 616 warn(1, "Interactive rename table out of memory"); 617 return(-1); 618} 619 620/* 621 * sub_name() 622 * look up a link name to see if it points at a file that has been 623 * remapped by the user. If found, the link is adjusted to contain the 624 * new name (oname is the link to name) 625 */ 626 627#if __STDC__ 628void 629sub_name(register char *oname, int *onamelen) 630#else 631void 632sub_name(oname, onamelen) 633 register char *oname; 634 int *onamelen; 635#endif 636{ 637 register NAMT *pt; 638 register u_int indx; 639 640 if (ntab == NULL) 641 return; 642 /* 643 * look the name up in the hash table 644 */ 645 indx = st_hash(oname, *onamelen, N_TAB_SZ); 646 if ((pt = ntab[indx]) == NULL) 647 return; 648 649 while (pt != NULL) { 650 /* 651 * walk down the hash cahin looking for a match 652 */ 653 if (strcmp(oname, pt->oname) == 0) { 654 /* 655 * found it, replace it with the new name 656 * and return (we know that oname has enough space) 657 */ 658 *onamelen = l_strncpy(oname, pt->nname, PAXPATHLEN+1); 659 oname[PAXPATHLEN] = '\0'; 660 return; 661 } 662 pt = pt->fow; 663 } 664 665 /* 666 * no match, just return 667 */ 668 return; 669} 670 671/* 672 * device/inode mapping table routines 673 * (used with formats that store device and inodes fields) 674 * 675 * device/inode mapping tables remap the device field in a archive header. The 676 * device/inode fields are used to determine when files are hard links to each 677 * other. However these values have very little meaning outside of that. This 678 * database is used to solve one of two different problems. 679 * 680 * 1) when files are appended to an archive, while the new files may have hard 681 * links to each other, you cannot determine if they have hard links to any 682 * file already stored on the archive from a prior run of pax. We must assume 683 * that these inode/device pairs are unique only within a SINGLE run of pax 684 * (which adds a set of files to an archive). So we have to make sure the 685 * inode/dev pairs we add each time are always unique. We do this by observing 686 * while the inode field is very dense, the use of the dev field is fairly 687 * sparse. Within each run of pax, we remap any device number of a new archive 688 * member that has a device number used in a prior run and already stored in a 689 * file on the archive. During the read phase of the append, we store the 690 * device numbers used and mark them to not be used by any file during the 691 * write phase. If during write we go to use one of those old device numbers, 692 * we remap it to a new value. 693 * 694 * 2) Often the fields in the archive header used to store these values are 695 * too small to store the entire value. The result is an inode or device value 696 * which can be truncated. This really can foul up an archive. With truncation 697 * we end up creating links between files that are really not links (after 698 * truncation the inodes are the same value). We address that by detecting 699 * truncation and forcing a remap of the device field to split truncated 700 * inodes away from each other. Each truncation creates a pattern of bits that 701 * are removed. We use this pattern of truncated bits to partition the inodes 702 * on a single device to many different devices (each one represented by the 703 * truncated bit pattern). All inodes on the same device that have the same 704 * truncation pattern are mapped to the same new device. Two inodes that 705 * truncate to the same value clearly will always have different truncation 706 * bit patterns, so they will be split from away each other. When we spot 707 * device truncation we remap the device number to a non truncated value. 708 * (for more info see table.h for the data structures involved). 709 */ 710 711/* 712 * dev_start() 713 * create the device mapping table 714 * Return: 715 * 0 if successful, -1 otherwise 716 */ 717 718#if __STDC__ 719int 720dev_start(void) 721#else 722int 723dev_start() 724#endif 725{ 726 if (dtab != NULL) 727 return(0); 728 if ((dtab = (DEVT **)calloc(D_TAB_SZ, sizeof(DEVT *))) == NULL) { 729 warn(1, "Cannot allocate memory for device mapping table"); 730 return(-1); 731 } 732 return(0); 733} 734 735/* 736 * add_dev() 737 * add a device number to the table. this will force the device to be 738 * remapped to a new value if it be used during a write phase. This 739 * function is called during the read phase of an append to prohibit the 740 * use of any device number already in the archive. 741 * Return: 742 * 0 if added ok, -1 otherwise 743 */ 744 745#if __STDC__ 746int 747add_dev(register ARCHD *arcn) 748#else 749int 750add_dev(arcn) 751 register ARCHD *arcn; 752#endif 753{ 754 if (chk_dev(arcn->sb.st_dev, 1) == NULL) 755 return(-1); 756 return(0); 757} 758 759/* 760 * chk_dev() 761 * check for a device value in the device table. If not found and the add 762 * flag is set, it is added. This does NOT assign any mapping values, just 763 * adds the device number as one that need to be remapped. If this device 764 * is alread mapped, just return with a pointer to that entry. 765 * Return: 766 * pointer to the entry for this device in the device map table. Null 767 * if the add flag is not set and the device is not in the table (it is 768 * not been seen yet). If add is set and the device cannot be added, null 769 * is returned (indicates an error). 770 */ 771 772#if __STDC__ 773static DEVT * 774chk_dev(dev_t dev, int add) 775#else 776static DEVT * 777chk_dev(dev, add) 778 dev_t dev; 779 int add; 780#endif 781{ 782 register DEVT *pt; 783 register u_int indx; 784 785 if (dtab == NULL) 786 return(NULL); 787 /* 788 * look to see if this device is already in the table 789 */ 790 indx = ((unsigned)dev) % D_TAB_SZ; 791 if ((pt = dtab[indx]) != NULL) { 792 while ((pt != NULL) && (pt->dev != dev)) 793 pt = pt->fow; 794 795 /* 796 * found it, return a pointer to it 797 */ 798 if (pt != NULL) 799 return(pt); 800 } 801 802 /* 803 * not in table, we add it only if told to as this may just be a check 804 * to see if a device number is being used. 805 */ 806 if (add == 0) 807 return(NULL); 808 809 /* 810 * allocate a node for this device and add it to the front of the hash 811 * chain. Note we do not assign remaps values here, so the pt->list 812 * list must be NULL. 813 */ 814 if ((pt = (DEVT *)malloc(sizeof(DEVT))) == NULL) { 815 warn(1, "Device map table out of memory"); 816 return(NULL); 817 } 818 pt->dev = dev; 819 pt->list = NULL; 820 pt->fow = dtab[indx]; 821 dtab[indx] = pt; 822 return(pt); 823} 824/* 825 * map_dev() 826 * given an inode and device storage mask (the mask has a 1 for each bit 827 * the archive format is able to store in a header), we check for inode 828 * and device truncation and remap the device as required. Device mapping 829 * can also occur when during the read phase of append a device number was 830 * seen (and was marked as do not use during the write phase). WE ASSUME 831 * that unsigned longs are the same size or bigger than the fields used 832 * for ino_t and dev_t. If not the types will have to be changed. 833 * Return: 834 * 0 if all ok, -1 otherwise. 835 */ 836 837#if __STDC__ 838int 839map_dev(register ARCHD *arcn, u_long dev_mask, u_long ino_mask) 840#else 841int 842map_dev(arcn, dev_mask, ino_mask) 843 register ARCHD *arcn; 844 u_long dev_mask; 845 u_long ino_mask; 846#endif 847{ 848 register DEVT *pt; 849 register DLIST *dpt; 850 static dev_t lastdev = 0; /* next device number to try */ 851 int trc_ino = 0; 852 int trc_dev = 0; 853 ino_t trunc_bits = 0; 854 ino_t nino; 855 856 if (dtab == NULL) 857 return(0); 858 /* 859 * check for device and inode truncation, and extract the truncated 860 * bit pattern. 861 */ 862 if ((arcn->sb.st_dev & (dev_t)dev_mask) != arcn->sb.st_dev) 863 ++trc_dev; 864 if ((nino = arcn->sb.st_ino & (ino_t)ino_mask) != arcn->sb.st_ino) { 865 ++trc_ino; 866 trunc_bits = arcn->sb.st_ino & (ino_t)(~ino_mask); 867 } 868 869 /* 870 * see if this device is already being mapped, look up the device 871 * then find the truncation bit pattern which applies 872 */ 873 if ((pt = chk_dev(arcn->sb.st_dev, 0)) != NULL) { 874 /* 875 * this device is already marked to be remapped 876 */ 877 for (dpt = pt->list; dpt != NULL; dpt = dpt->fow) 878 if (dpt->trunc_bits == trunc_bits) 879 break; 880 881 if (dpt != NULL) { 882 /* 883 * we are being remapped for this device and pattern 884 * change the device number to be stored and return 885 */ 886 arcn->sb.st_dev = dpt->dev; 887 arcn->sb.st_ino = nino; 888 return(0); 889 } 890 } else { 891 /* 892 * this device is not being remapped YET. if we do not have any 893 * form of truncation, we do not need a remap 894 */ 895 if (!trc_ino && !trc_dev) 896 return(0); 897 898 /* 899 * we have truncation, have to add this as a device to remap 900 */ 901 if ((pt = chk_dev(arcn->sb.st_dev, 1)) == NULL) 902 goto bad; 903 904 /* 905 * if we just have a truncated inode, we have to make sure that 906 * all future inodes that do not truncate (they have the 907 * truncation pattern of all 0's) continue to map to the same 908 * device number. We probably have already written inodes with 909 * this device number to the archive with the truncation 910 * pattern of all 0's. So we add the mapping for all 0's to the 911 * same device number. 912 */ 913 if (!trc_dev && (trunc_bits != 0)) { 914 if ((dpt = (DLIST *)malloc(sizeof(DLIST))) == NULL) 915 goto bad; 916 dpt->trunc_bits = 0; 917 dpt->dev = arcn->sb.st_dev; 918 dpt->fow = pt->list; 919 pt->list = dpt; 920 } 921 } 922 923 /* 924 * look for a device number not being used. We must watch for wrap 925 * around on lastdev (so we do not get stuck looking forever!) 926 */ 927 while (++lastdev > 0) { 928 if (chk_dev(lastdev, 0) != NULL) 929 continue; 930 /* 931 * found an unused value. If we have reached truncation point 932 * for this format we are hosed, so we give up. Otherwise we 933 * mark it as being used. 934 */ 935 if (((lastdev & ((dev_t)dev_mask)) != lastdev) || 936 (chk_dev(lastdev, 1) == NULL)) 937 goto bad; 938 break; 939 } 940 941 if ((lastdev <= 0) || ((dpt = (DLIST *)malloc(sizeof(DLIST))) == NULL)) 942 goto bad; 943 944 /* 945 * got a new device number, store it under this truncation pattern. 946 * change the device number this file is being stored with. 947 */ 948 dpt->trunc_bits = trunc_bits; 949 dpt->dev = lastdev; 950 dpt->fow = pt->list; 951 pt->list = dpt; 952 arcn->sb.st_dev = lastdev; 953 arcn->sb.st_ino = nino; 954 return(0); 955 956 bad: 957 warn(1, "Unable to fix truncated inode/device field when storing %s", 958 arcn->name); 959 warn(0, "Archive may create improper hard links when extracted"); 960 return(0); 961} 962 963/* 964 * directory access/mod time reset table routines (for directories READ by pax) 965 * 966 * The pax -t flag requires that access times of archive files to be the same 967 * before being read by pax. For regular files, access time is restored after 968 * the file has been copied. This database provides the same functionality for 969 * directories read during file tree traversal. Restoring directory access time 970 * is more complex than files since directories may be read several times until 971 * all the descendants in their subtree are visited by fts. Directory access 972 * and modification times are stored during the fts pre-order visit (done 973 * before any descendants in the subtree is visited) and restored after the 974 * fts post-order visit (after all the descendants have been visited). In the 975 * case of premature exit from a subtree (like from the effects of -n), any 976 * directory entries left in this database are reset during final cleanup 977 * operations of pax. Entries are hashed by inode number for fast lookup. 978 */ 979 980/* 981 * atdir_start() 982 * create the directory access time database for directories READ by pax. 983 * Return: 984 * 0 is created ok, -1 otherwise. 985 */ 986 987#if __STDC__ 988int 989atdir_start(void) 990#else 991int 992atdir_start() 993#endif 994{ 995 if (atab != NULL) 996 return(0); 997 if ((atab = (ATDIR **)calloc(A_TAB_SZ, sizeof(ATDIR *))) == NULL) { 998 warn(1,"Cannot allocate space for directory access time table"); 999 return(-1); 1000 } 1001 return(0); 1002} 1003 1004 1005/* 1006 * atdir_end() 1007 * walk through the directory access time table and reset the access time 1008 * of any directory who still has an entry left in the database. These 1009 * entries are for directories READ by pax 1010 */ 1011 1012#if __STDC__ 1013void 1014atdir_end(void) 1015#else 1016void 1017atdir_end() 1018#endif 1019{ 1020 register ATDIR *pt; 1021 register int i; 1022 1023 if (atab == NULL) 1024 return; 1025 /* 1026 * for each non-empty hash table entry reset all the directories 1027 * chained there. 1028 */ 1029 for (i = 0; i < A_TAB_SZ; ++i) { 1030 if ((pt = atab[i]) == NULL) 1031 continue; 1032 /* 1033 * remember to force the times, set_ftime() looks at pmtime 1034 * and patime, which only applies to things CREATED by pax, 1035 * not read by pax. Read time reset is controlled by -t. 1036 */ 1037 for (; pt != NULL; pt = pt->fow) 1038 set_ftime(pt->name, pt->mtime, pt->atime, 1); 1039 } 1040} 1041 1042/* 1043 * add_atdir() 1044 * add a directory to the directory access time table. Table is hashed 1045 * and chained by inode number. This is for directories READ by pax 1046 */ 1047 1048#if __STDC__ 1049void 1050add_atdir(char *fname, dev_t dev, ino_t ino, time_t mtime, time_t atime) 1051#else 1052void 1053add_atdir(fname, dev, ino, mtime, atime) 1054 char *fname; 1055 dev_t dev; 1056 ino_t ino; 1057 time_t mtime; 1058 time_t atime; 1059#endif 1060{ 1061 register ATDIR *pt; 1062 register u_int indx; 1063 1064 if (atab == NULL) 1065 return; 1066 1067 /* 1068 * make sure this directory is not already in the table, if so just 1069 * return (the older entry always has the correct time). The only 1070 * way this will happen is when the same subtree can be traversed by 1071 * different args to pax and the -n option is aborting fts out of a 1072 * subtree before all the post-order visits have been made). 1073 */ 1074 indx = ((unsigned)ino) % A_TAB_SZ; 1075 if ((pt = atab[indx]) != NULL) { 1076 while (pt != NULL) { 1077 if ((pt->ino == ino) && (pt->dev == dev)) 1078 break; 1079 pt = pt->fow; 1080 } 1081 1082 /* 1083 * oops, already there. Leave it alone. 1084 */ 1085 if (pt != NULL) 1086 return; 1087 } 1088 1089 /* 1090 * add it to the front of the hash chain 1091 */ 1092 if ((pt = (ATDIR *)malloc(sizeof(ATDIR))) != NULL) { 1093 if ((pt->name = strdup(fname)) != NULL) { 1094 pt->dev = dev; 1095 pt->ino = ino; 1096 pt->mtime = mtime; 1097 pt->atime = atime; 1098 pt->fow = atab[indx]; 1099 atab[indx] = pt; 1100 return; 1101 } 1102 (void)free((char *)pt); 1103 } 1104 1105 warn(1, "Directory access time reset table ran out of memory"); 1106 return; 1107} 1108 1109/* 1110 * get_atdir() 1111 * look up a directory by inode and device number to obtain the access 1112 * and modification time you want to set to. If found, the modification 1113 * and access time parameters are set and the entry is removed from the 1114 * table (as it is no longer needed). These are for directories READ by 1115 * pax 1116 * Return: 1117 * 0 if found, -1 if not found. 1118 */ 1119 1120#if __STDC__ 1121int 1122get_atdir(dev_t dev, ino_t ino, time_t *mtime, time_t *atime) 1123#else 1124int 1125get_atdir(dev, ino, mtime, atime) 1126 dev_t dev; 1127 ino_t ino; 1128 time_t *mtime; 1129 time_t *atime; 1130#endif 1131{ 1132 register ATDIR *pt; 1133 register ATDIR **ppt; 1134 register u_int indx; 1135 1136 if (atab == NULL) 1137 return(-1); 1138 /* 1139 * hash by inode and search the chain for an inode and device match 1140 */ 1141 indx = ((unsigned)ino) % A_TAB_SZ; 1142 if ((pt = atab[indx]) == NULL) 1143 return(-1); 1144 1145 ppt = &(atab[indx]); 1146 while (pt != NULL) { 1147 if ((pt->ino == ino) && (pt->dev == dev)) 1148 break; 1149 /* 1150 * no match, go to next one 1151 */ 1152 ppt = &(pt->fow); 1153 pt = pt->fow; 1154 } 1155 1156 /* 1157 * return if we did not find it. 1158 */ 1159 if (pt == NULL) 1160 return(-1); 1161 1162 /* 1163 * found it. return the times and remove the entry from the table. 1164 */ 1165 *ppt = pt->fow; 1166 *mtime = pt->mtime; 1167 *atime = pt->atime; 1168 (void)free((char *)pt->name); 1169 (void)free((char *)pt); 1170 return(0); 1171} 1172 1173/* 1174 * directory access mode and time storage routines (for directories CREATED 1175 * by pax). 1176 * 1177 * Pax requires that extracted directories, by default, have their access/mod 1178 * times and permissions set to the values specified in the archive. During the 1179 * actions of extracting (and creating the destination subtree during -rw copy) 1180 * directories extracted may be modified after being created. Even worse is 1181 * that these directories may have been created with file permissions which 1182 * prohibits any descendants of these directories from being extracted. When 1183 * directories are created by pax, access rights may be added to permit the 1184 * creation of files in their subtree. Every time pax creates a directory, the 1185 * times and file permissions specified by the archive are stored. After all 1186 * files have been extracted (or copied), these directories have their times 1187 * and file modes reset to the stored values. The directory info is restored in 1188 * reverse order as entries were added to the data file from root to leaf. To 1189 * restore atime properly, we must go backwards. The data file consists of 1190 * records with two parts, the file name followed by a DIRDATA trailer. The 1191 * fixed sized trailer contains the size of the name plus the off_t location in 1192 * the file. To restore we work backwards through the file reading the trailer 1193 * then the file name. 1194 */ 1195 1196/* 1197 * dir_start() 1198 * set up the directory time and file mode storage for directories CREATED 1199 * by pax. 1200 * Return: 1201 * 0 if ok, -1 otherwise 1202 */ 1203 1204#if __STDC__ 1205int 1206dir_start(void) 1207#else 1208int 1209dir_start() 1210#endif 1211{ 1212 char *pt; 1213 1214 if (dirfd != -1) 1215 return(0); 1216 if ((pt = tempnam((char *)NULL, (char *)NULL)) == NULL) 1217 return(-1); 1218 1219 /* 1220 * unlink the file so it goes away at termination by itself 1221 */ 1222 (void)unlink(pt); 1223 if ((dirfd = open(pt, O_RDWR|O_CREAT, 0600)) >= 0) { 1224 (void)unlink(pt); 1225 return(0); 1226 } 1227 warn(1, "Unable to create temporary file for directory times: %s", pt); 1228 return(-1); 1229} 1230 1231/* 1232 * add_dir() 1233 * add the mode and times for a newly CREATED directory 1234 * name is name of the directory, psb the stat buffer with the data in it, 1235 * frc_mode is a flag that says whether to force the setting of the mode 1236 * (ignoring the user set values for preserving file mode). Frc_mode is 1237 * for the case where we created a file and found that the resulting 1238 * directory was not writeable and the user asked for file modes to NOT 1239 * be preserved. (we have to preserve what was created by default, so we 1240 * have to force the setting at the end. this is stated explicitly in the 1241 * pax spec) 1242 */ 1243 1244#if __STDC__ 1245void 1246add_dir(char *name, int nlen, struct stat *psb, int frc_mode) 1247#else 1248void 1249add_dir(name, nlen, psb, frc_mode) 1250 char *name; 1251 int nlen; 1252 struct stat *psb; 1253 int frc_mode; 1254#endif 1255{ 1256 DIRDATA dblk; 1257 1258 if (dirfd < 0) 1259 return; 1260 1261 /* 1262 * get current position (where file name will start) so we can store it 1263 * in the trailer 1264 */ 1265 if ((dblk.npos = lseek(dirfd, 0L, SEEK_CUR)) < 0) { 1266 warn(1,"Unable to store mode and times for directory: %s",name); 1267 return; 1268 } 1269 1270 /* 1271 * write the file name followed by the trailer 1272 */ 1273 dblk.nlen = nlen + 1; 1274 dblk.mode = psb->st_mode & 0xffff; 1275 dblk.mtime = psb->st_mtime; 1276 dblk.atime = psb->st_atime; 1277 dblk.frc_mode = frc_mode; 1278 if ((write(dirfd, name, dblk.nlen) == dblk.nlen) && 1279 (write(dirfd, (char *)&dblk, sizeof(dblk)) == sizeof(dblk))) { 1280 ++dircnt; 1281 return; 1282 } 1283 1284 warn(1,"Unable to store mode and times for created directory: %s",name); 1285 return; 1286} 1287 1288/* 1289 * proc_dir() 1290 * process all file modes and times stored for directories CREATED 1291 * by pax 1292 */ 1293 1294#if __STDC__ 1295void 1296proc_dir(void) 1297#else 1298void 1299proc_dir() 1300#endif 1301{ 1302 char name[PAXPATHLEN+1]; 1303 DIRDATA dblk; 1304 u_long cnt; 1305 1306 if (dirfd < 0) 1307 return; 1308 /* 1309 * read backwards through the file and process each directory 1310 */ 1311 for (cnt = 0; cnt < dircnt; ++cnt) { 1312 /* 1313 * read the trailer, then the file name, if this fails 1314 * just give up. 1315 */ 1316 if (lseek(dirfd, -((off_t)sizeof(dblk)), SEEK_CUR) < 0) 1317 break; 1318 if (read(dirfd,(char *)&dblk, sizeof(dblk)) != sizeof(dblk)) 1319 break; 1320 if (lseek(dirfd, dblk.npos, SEEK_SET) < 0) 1321 break; 1322 if (read(dirfd, name, dblk.nlen) != dblk.nlen) 1323 break; 1324 if (lseek(dirfd, dblk.npos, SEEK_SET) < 0) 1325 break; 1326 1327 /* 1328 * frc_mode set, make sure we set the file modes even if 1329 * the user didn't ask for it (see file_subs.c for more info) 1330 */ 1331 if (pmode || dblk.frc_mode) 1332 set_pmode(name, dblk.mode); 1333 if (patime || pmtime) 1334 set_ftime(name, dblk.mtime, dblk.atime, 0); 1335 } 1336 1337 (void)close(dirfd); 1338 dirfd = -1; 1339 if (cnt != dircnt) 1340 warn(1,"Unable to set mode and times for created directories"); 1341 return; 1342} 1343 1344/* 1345 * database independent routines 1346 */ 1347 1348/* 1349 * st_hash() 1350 * hashes filenames to a u_int for hashing into a table. Looks at the tail 1351 * end of file, as this provides far better distribution than any other 1352 * part of the name. For performance reasons we only care about the last 1353 * MAXKEYLEN chars (should be at LEAST large enough to pick off the file 1354 * name). Was tested on 500,000 name file tree traversal from the root 1355 * and gave almost a perfectly uniform distribution of keys when used with 1356 * prime sized tables (MAXKEYLEN was 128 in test). Hashes (sizeof int) 1357 * chars at a time and pads with 0 for last addition. 1358 * Return: 1359 * the hash value of the string MOD (%) the table size. 1360 */ 1361 1362#if __STDC__ 1363u_int 1364st_hash(char *name, int len, int tabsz) 1365#else 1366u_int 1367st_hash(name, len, tabsz) 1368 char *name; 1369 int len; 1370 int tabsz; 1371#endif 1372{ 1373 register char *pt; 1374 register char *dest; 1375 register char *end; 1376 register int i; 1377 register u_int key = 0; 1378 register int steps; 1379 register int res; 1380 u_int val; 1381 1382 /* 1383 * only look at the tail up to MAXKEYLEN, we do not need to waste 1384 * time here (remember these are pathnames, the tail is what will 1385 * spread out the keys) 1386 */ 1387 if (len > MAXKEYLEN) { 1388 pt = &(name[len - MAXKEYLEN]); 1389 len = MAXKEYLEN; 1390 } else 1391 pt = name; 1392 1393 /* 1394 * calculate the number of u_int size steps in the string and if 1395 * there is a runt to deal with 1396 */ 1397 steps = len/sizeof(u_int); 1398 res = len % sizeof(u_int); 1399 1400 /* 1401 * add up the value of the string in unsigned integer sized pieces 1402 * too bad we cannot have unsigned int aligned strings, then we 1403 * could avoid the expensive copy. 1404 */ 1405 for (i = 0; i < steps; ++i) { 1406 end = pt + sizeof(u_int); 1407 dest = (char *)&val; 1408 while (pt < end) 1409 *dest++ = *pt++; 1410 key += val; 1411 } 1412 1413 /* 1414 * add in the runt padded with zero to the right 1415 */ 1416 if (res) { 1417 val = 0; 1418 end = pt + res; 1419 dest = (char *)&val; 1420 while (pt < end) 1421 *dest++ = *pt++; 1422 key += val; 1423 } 1424 1425 /* 1426 * return the result mod the table size 1427 */ 1428 return(key % tabsz); 1429} 1430