tables.c revision 76017
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 40static char sccsid[] = "@(#)tables.c 8.1 (Berkeley) 5/31/93"; 41#endif 42static const char rcsid[] = 43 "$FreeBSD: head/bin/pax/tables.c 76017 2001-04-26 08:37:00Z kris $"; 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 74static HRDLNK **ltab = NULL; /* hard link table for detecting hard links */ 75static FTM **ftab = NULL; /* file time table for updating arch */ 76static NAMT **ntab = NULL; /* interactive rename storage table */ 77static DEVT **dtab = NULL; /* device/inode mapping tables */ 78static ATDIR **atab = NULL; /* file tree directory time reset table */ 79static int dirfd = -1; /* storage for setting created dir time/mode */ 80static u_long dircnt; /* entries in dir time/mode storage */ 81static int ffd = -1; /* tmp file for file time table name storage */ 82 83static 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__ 109int 110lnk_start(void) 111#else 112int 113lnk_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__ 138int 139chk_lnk(register ARCHD *arcn) 140#else 141int 142chk_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 PAXPATHLEN+1); 184 arcn->ln_name[PAXPATHLEN] = '\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__ 231void 232purg_lnk(register ARCHD *arcn) 233#else 234void 235purg_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__ 291void 292lnk_end(void) 293#else 294void 295lnk_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__ 356int 357ftime_start(void) 358#else 359int 360ftime_start() 361#endif 362{ 363 if (ftab != NULL) 364 return(0); 365 if ((ftab = (FTM **)calloc(F_TAB_SZ, sizeof(FTM *))) == NULL) { 366 paxwarn(1, "Cannot allocate memory for file time table"); 367 return(-1); 368 } 369 370 /* 371 * get random name and create temporary scratch file, unlink name 372 * so it will get removed on exit 373 */ 374 memcpy(tempbase, _TFILE_BASE, sizeof(_TFILE_BASE)); 375 if ((ffd = mkstemp(tempfile)) < 0) { 376 syswarn(1, errno, "Unable to create temporary file: %s", 377 tempfile); 378 return(-1); 379 } 380 (void)unlink(tempfile); 381 382 return(0); 383} 384 385/* 386 * chk_ftime() 387 * looks up entry in file time hash table. If not found, the file is 388 * added to the hash table and the file named stored in the scratch file. 389 * If a file with the same name is found, the file times are compared and 390 * the most recent file time is retained. If the new file was younger (or 391 * was not in the database) the new file is selected for storage. 392 * Return: 393 * 0 if file should be added to the archive, 1 if it should be skipped, 394 * -1 on error 395 */ 396 397#ifdef __STDC__ 398int 399chk_ftime(register ARCHD *arcn) 400#else 401int 402chk_ftime(arcn) 403 register ARCHD *arcn; 404#endif 405{ 406 register FTM *pt; 407 register int namelen; 408 register u_int indx; 409 char ckname[PAXPATHLEN+1]; 410 411 /* 412 * no info, go ahead and add to archive 413 */ 414 if (ftab == NULL) 415 return(0); 416 417 /* 418 * hash the pathname and look up in table 419 */ 420 namelen = arcn->nlen; 421 indx = st_hash(arcn->name, namelen, F_TAB_SZ); 422 if ((pt = ftab[indx]) != NULL) { 423 /* 424 * the hash chain is not empty, walk down looking for match 425 * only read up the path names if the lengths match, speeds 426 * up the search a lot 427 */ 428 while (pt != NULL) { 429 if (pt->namelen == namelen) { 430 /* 431 * potential match, have to read the name 432 * from the scratch file. 433 */ 434 if (lseek(ffd,pt->seek,SEEK_SET) != pt->seek) { 435 syswarn(1, errno, 436 "Failed ftime table seek"); 437 return(-1); 438 } 439 if (read(ffd, ckname, namelen) != namelen) { 440 syswarn(1, errno, 441 "Failed ftime table read"); 442 return(-1); 443 } 444 445 /* 446 * if the names match, we are done 447 */ 448 if (!strncmp(ckname, arcn->name, namelen)) 449 break; 450 } 451 452 /* 453 * try the next entry on the chain 454 */ 455 pt = pt->fow; 456 } 457 458 if (pt != NULL) { 459 /* 460 * found the file, compare the times, save the newer 461 */ 462 if (arcn->sb.st_mtime > pt->mtime) { 463 /* 464 * file is newer 465 */ 466 pt->mtime = arcn->sb.st_mtime; 467 return(0); 468 } 469 /* 470 * file is older 471 */ 472 return(1); 473 } 474 } 475 476 /* 477 * not in table, add it 478 */ 479 if ((pt = (FTM *)malloc(sizeof(FTM))) != NULL) { 480 /* 481 * add the name at the end of the scratch file, saving the 482 * offset. add the file to the head of the hash chain 483 */ 484 if ((pt->seek = lseek(ffd, (off_t)0, SEEK_END)) >= 0) { 485 if (write(ffd, arcn->name, namelen) == namelen) { 486 pt->mtime = arcn->sb.st_mtime; 487 pt->namelen = namelen; 488 pt->fow = ftab[indx]; 489 ftab[indx] = pt; 490 return(0); 491 } 492 syswarn(1, errno, "Failed write to file time table"); 493 } else 494 syswarn(1, errno, "Failed seek on file time table"); 495 } else 496 paxwarn(1, "File time table ran out of memory"); 497 498 if (pt != NULL) 499 (void)free((char *)pt); 500 return(-1); 501} 502 503/* 504 * Interactive rename table routines 505 * 506 * The interactive rename table keeps track of the new names that the user 507 * assigns to files from tty input. Since this map is unique for each file 508 * we must store it in case there is a reference to the file later in archive 509 * (a link). Otherwise we will be unable to find the file we know was 510 * extracted. The remapping of these files is stored in a memory based hash 511 * table (it is assumed since input must come from /dev/tty, it is unlikely to 512 * be a very large table). 513 */ 514 515/* 516 * name_start() 517 * create the interactive rename table 518 * Return: 519 * 0 if successful, -1 otherwise 520 */ 521 522#ifdef __STDC__ 523int 524name_start(void) 525#else 526int 527name_start() 528#endif 529{ 530 if (ntab != NULL) 531 return(0); 532 if ((ntab = (NAMT **)calloc(N_TAB_SZ, sizeof(NAMT *))) == NULL) { 533 paxwarn(1, "Cannot allocate memory for interactive rename table"); 534 return(-1); 535 } 536 return(0); 537} 538 539/* 540 * add_name() 541 * add the new name to old name mapping just created by the user. 542 * If an old name mapping is found (there may be duplicate names on an 543 * archive) only the most recent is kept. 544 * Return: 545 * 0 if added, -1 otherwise 546 */ 547 548#ifdef __STDC__ 549int 550add_name(register char *oname, int onamelen, char *nname) 551#else 552int 553add_name(oname, onamelen, nname) 554 register char *oname; 555 int onamelen; 556 char *nname; 557#endif 558{ 559 register NAMT *pt; 560 register u_int indx; 561 562 if (ntab == NULL) { 563 /* 564 * should never happen 565 */ 566 paxwarn(0, "No interactive rename table, links may fail\n"); 567 return(0); 568 } 569 570 /* 571 * look to see if we have already mapped this file, if so we 572 * will update it 573 */ 574 indx = st_hash(oname, onamelen, N_TAB_SZ); 575 if ((pt = ntab[indx]) != NULL) { 576 /* 577 * look down the has chain for the file 578 */ 579 while ((pt != NULL) && (strcmp(oname, pt->oname) != 0)) 580 pt = pt->fow; 581 582 if (pt != NULL) { 583 /* 584 * found an old mapping, replace it with the new one 585 * the user just input (if it is different) 586 */ 587 if (strcmp(nname, pt->nname) == 0) 588 return(0); 589 590 (void)free((char *)pt->nname); 591 if ((pt->nname = strdup(nname)) == NULL) { 592 paxwarn(1, "Cannot update rename table"); 593 return(-1); 594 } 595 return(0); 596 } 597 } 598 599 /* 600 * this is a new mapping, add it to the table 601 */ 602 if ((pt = (NAMT *)malloc(sizeof(NAMT))) != NULL) { 603 if ((pt->oname = strdup(oname)) != NULL) { 604 if ((pt->nname = strdup(nname)) != NULL) { 605 pt->fow = ntab[indx]; 606 ntab[indx] = pt; 607 return(0); 608 } 609 (void)free((char *)pt->oname); 610 } 611 (void)free((char *)pt); 612 } 613 paxwarn(1, "Interactive rename table out of memory"); 614 return(-1); 615} 616 617/* 618 * sub_name() 619 * look up a link name to see if it points at a file that has been 620 * remapped by the user. If found, the link is adjusted to contain the 621 * new name (oname is the link to name) 622 */ 623 624#ifdef __STDC__ 625void 626sub_name(register char *oname, int *onamelen) 627#else 628void 629sub_name(oname, onamelen) 630 register char *oname; 631 int *onamelen; 632#endif 633{ 634 register NAMT *pt; 635 register u_int indx; 636 637 if (ntab == NULL) 638 return; 639 /* 640 * look the name up in the hash table 641 */ 642 indx = st_hash(oname, *onamelen, N_TAB_SZ); 643 if ((pt = ntab[indx]) == NULL) 644 return; 645 646 while (pt != NULL) { 647 /* 648 * walk down the hash chain looking for a match 649 */ 650 if (strcmp(oname, pt->oname) == 0) { 651 /* 652 * found it, replace it with the new name 653 * and return (we know that oname has enough space) 654 */ 655 *onamelen = l_strncpy(oname, pt->nname, PAXPATHLEN+1); 656 oname[PAXPATHLEN] = '\0'; 657 return; 658 } 659 pt = pt->fow; 660 } 661 662 /* 663 * no match, just return 664 */ 665 return; 666} 667 668/* 669 * device/inode mapping table routines 670 * (used with formats that store device and inodes fields) 671 * 672 * device/inode mapping tables remap the device field in a archive header. The 673 * device/inode fields are used to determine when files are hard links to each 674 * other. However these values have very little meaning outside of that. This 675 * database is used to solve one of two different problems. 676 * 677 * 1) when files are appended to an archive, while the new files may have hard 678 * links to each other, you cannot determine if they have hard links to any 679 * file already stored on the archive from a prior run of pax. We must assume 680 * that these inode/device pairs are unique only within a SINGLE run of pax 681 * (which adds a set of files to an archive). So we have to make sure the 682 * inode/dev pairs we add each time are always unique. We do this by observing 683 * while the inode field is very dense, the use of the dev field is fairly 684 * sparse. Within each run of pax, we remap any device number of a new archive 685 * member that has a device number used in a prior run and already stored in a 686 * file on the archive. During the read phase of the append, we store the 687 * device numbers used and mark them to not be used by any file during the 688 * write phase. If during write we go to use one of those old device numbers, 689 * we remap it to a new value. 690 * 691 * 2) Often the fields in the archive header used to store these values are 692 * too small to store the entire value. The result is an inode or device value 693 * which can be truncated. This really can foul up an archive. With truncation 694 * we end up creating links between files that are really not links (after 695 * truncation the inodes are the same value). We address that by detecting 696 * truncation and forcing a remap of the device field to split truncated 697 * inodes away from each other. Each truncation creates a pattern of bits that 698 * are removed. We use this pattern of truncated bits to partition the inodes 699 * on a single device to many different devices (each one represented by the 700 * truncated bit pattern). All inodes on the same device that have the same 701 * truncation pattern are mapped to the same new device. Two inodes that 702 * truncate to the same value clearly will always have different truncation 703 * bit patterns, so they will be split from away each other. When we spot 704 * device truncation we remap the device number to a non truncated value. 705 * (for more info see table.h for the data structures involved). 706 */ 707 708/* 709 * dev_start() 710 * create the device mapping table 711 * Return: 712 * 0 if successful, -1 otherwise 713 */ 714 715#ifdef __STDC__ 716int 717dev_start(void) 718#else 719int 720dev_start() 721#endif 722{ 723 if (dtab != NULL) 724 return(0); 725 if ((dtab = (DEVT **)calloc(D_TAB_SZ, sizeof(DEVT *))) == NULL) { 726 paxwarn(1, "Cannot allocate memory for device mapping table"); 727 return(-1); 728 } 729 return(0); 730} 731 732/* 733 * add_dev() 734 * add a device number to the table. this will force the device to be 735 * remapped to a new value if it be used during a write phase. This 736 * function is called during the read phase of an append to prohibit the 737 * use of any device number already in the archive. 738 * Return: 739 * 0 if added ok, -1 otherwise 740 */ 741 742#ifdef __STDC__ 743int 744add_dev(register ARCHD *arcn) 745#else 746int 747add_dev(arcn) 748 register ARCHD *arcn; 749#endif 750{ 751 if (chk_dev(arcn->sb.st_dev, 1) == NULL) 752 return(-1); 753 return(0); 754} 755 756/* 757 * chk_dev() 758 * check for a device value in the device table. If not found and the add 759 * flag is set, it is added. This does NOT assign any mapping values, just 760 * adds the device number as one that need to be remapped. If this device 761 * is already mapped, just return with a pointer to that entry. 762 * Return: 763 * pointer to the entry for this device in the device map table. Null 764 * if the add flag is not set and the device is not in the table (it is 765 * not been seen yet). If add is set and the device cannot be added, null 766 * is returned (indicates an error). 767 */ 768 769#ifdef __STDC__ 770static DEVT * 771chk_dev(dev_t dev, int add) 772#else 773static DEVT * 774chk_dev(dev, add) 775 dev_t dev; 776 int add; 777#endif 778{ 779 register DEVT *pt; 780 register u_int indx; 781 782 if (dtab == NULL) 783 return(NULL); 784 /* 785 * look to see if this device is already in the table 786 */ 787 indx = ((unsigned)dev) % D_TAB_SZ; 788 if ((pt = dtab[indx]) != NULL) { 789 while ((pt != NULL) && (pt->dev != dev)) 790 pt = pt->fow; 791 792 /* 793 * found it, return a pointer to it 794 */ 795 if (pt != NULL) 796 return(pt); 797 } 798 799 /* 800 * not in table, we add it only if told to as this may just be a check 801 * to see if a device number is being used. 802 */ 803 if (add == 0) 804 return(NULL); 805 806 /* 807 * allocate a node for this device and add it to the front of the hash 808 * chain. Note we do not assign remaps values here, so the pt->list 809 * list must be NULL. 810 */ 811 if ((pt = (DEVT *)malloc(sizeof(DEVT))) == NULL) { 812 paxwarn(1, "Device map table out of memory"); 813 return(NULL); 814 } 815 pt->dev = dev; 816 pt->list = NULL; 817 pt->fow = dtab[indx]; 818 dtab[indx] = pt; 819 return(pt); 820} 821/* 822 * map_dev() 823 * given an inode and device storage mask (the mask has a 1 for each bit 824 * the archive format is able to store in a header), we check for inode 825 * and device truncation and remap the device as required. Device mapping 826 * can also occur when during the read phase of append a device number was 827 * seen (and was marked as do not use during the write phase). WE ASSUME 828 * that unsigned longs are the same size or bigger than the fields used 829 * for ino_t and dev_t. If not the types will have to be changed. 830 * Return: 831 * 0 if all ok, -1 otherwise. 832 */ 833 834#ifdef __STDC__ 835int 836map_dev(register ARCHD *arcn, u_long dev_mask, u_long ino_mask) 837#else 838int 839map_dev(arcn, dev_mask, ino_mask) 840 register ARCHD *arcn; 841 u_long dev_mask; 842 u_long ino_mask; 843#endif 844{ 845 register DEVT *pt; 846 register DLIST *dpt; 847 static dev_t lastdev = 0; /* next device number to try */ 848 int trc_ino = 0; 849 int trc_dev = 0; 850 ino_t trunc_bits = 0; 851 ino_t nino; 852 853 if (dtab == NULL) 854 return(0); 855 /* 856 * check for device and inode truncation, and extract the truncated 857 * bit pattern. 858 */ 859 if ((arcn->sb.st_dev & (dev_t)dev_mask) != arcn->sb.st_dev) 860 ++trc_dev; 861 if ((nino = arcn->sb.st_ino & (ino_t)ino_mask) != arcn->sb.st_ino) { 862 ++trc_ino; 863 trunc_bits = arcn->sb.st_ino & (ino_t)(~ino_mask); 864 } 865 866 /* 867 * see if this device is already being mapped, look up the device 868 * then find the truncation bit pattern which applies 869 */ 870 if ((pt = chk_dev(arcn->sb.st_dev, 0)) != NULL) { 871 /* 872 * this device is already marked to be remapped 873 */ 874 for (dpt = pt->list; dpt != NULL; dpt = dpt->fow) 875 if (dpt->trunc_bits == trunc_bits) 876 break; 877 878 if (dpt != NULL) { 879 /* 880 * we are being remapped for this device and pattern 881 * change the device number to be stored and return 882 */ 883 arcn->sb.st_dev = dpt->dev; 884 arcn->sb.st_ino = nino; 885 return(0); 886 } 887 } else { 888 /* 889 * this device is not being remapped YET. if we do not have any 890 * form of truncation, we do not need a remap 891 */ 892 if (!trc_ino && !trc_dev) 893 return(0); 894 895 /* 896 * we have truncation, have to add this as a device to remap 897 */ 898 if ((pt = chk_dev(arcn->sb.st_dev, 1)) == NULL) 899 goto bad; 900 901 /* 902 * if we just have a truncated inode, we have to make sure that 903 * all future inodes that do not truncate (they have the 904 * truncation pattern of all 0's) continue to map to the same 905 * device number. We probably have already written inodes with 906 * this device number to the archive with the truncation 907 * pattern of all 0's. So we add the mapping for all 0's to the 908 * same device number. 909 */ 910 if (!trc_dev && (trunc_bits != 0)) { 911 if ((dpt = (DLIST *)malloc(sizeof(DLIST))) == NULL) 912 goto bad; 913 dpt->trunc_bits = 0; 914 dpt->dev = arcn->sb.st_dev; 915 dpt->fow = pt->list; 916 pt->list = dpt; 917 } 918 } 919 920 /* 921 * look for a device number not being used. We must watch for wrap 922 * around on lastdev (so we do not get stuck looking forever!) 923 */ 924 while (++lastdev > 0) { 925 if (chk_dev(lastdev, 0) != NULL) 926 continue; 927 /* 928 * found an unused value. If we have reached truncation point 929 * for this format we are hosed, so we give up. Otherwise we 930 * mark it as being used. 931 */ 932 if (((lastdev & ((dev_t)dev_mask)) != lastdev) || 933 (chk_dev(lastdev, 1) == NULL)) 934 goto bad; 935 break; 936 } 937 938 if ((lastdev <= 0) || ((dpt = (DLIST *)malloc(sizeof(DLIST))) == NULL)) 939 goto bad; 940 941 /* 942 * got a new device number, store it under this truncation pattern. 943 * change the device number this file is being stored with. 944 */ 945 dpt->trunc_bits = trunc_bits; 946 dpt->dev = lastdev; 947 dpt->fow = pt->list; 948 pt->list = dpt; 949 arcn->sb.st_dev = lastdev; 950 arcn->sb.st_ino = nino; 951 return(0); 952 953 bad: 954 paxwarn(1, "Unable to fix truncated inode/device field when storing %s", 955 arcn->name); 956 paxwarn(0, "Archive may create improper hard links when extracted"); 957 return(0); 958} 959 960/* 961 * directory access/mod time reset table routines (for directories READ by pax) 962 * 963 * The pax -t flag requires that access times of archive files to be the same 964 * before being read by pax. For regular files, access time is restored after 965 * the file has been copied. This database provides the same functionality for 966 * directories read during file tree traversal. Restoring directory access time 967 * is more complex than files since directories may be read several times until 968 * all the descendants in their subtree are visited by fts. Directory access 969 * and modification times are stored during the fts pre-order visit (done 970 * before any descendants in the subtree is visited) and restored after the 971 * fts post-order visit (after all the descendants have been visited). In the 972 * case of premature exit from a subtree (like from the effects of -n), any 973 * directory entries left in this database are reset during final cleanup 974 * operations of pax. Entries are hashed by inode number for fast lookup. 975 */ 976 977/* 978 * atdir_start() 979 * create the directory access time database for directories READ by pax. 980 * Return: 981 * 0 is created ok, -1 otherwise. 982 */ 983 984#ifdef __STDC__ 985int 986atdir_start(void) 987#else 988int 989atdir_start() 990#endif 991{ 992 if (atab != NULL) 993 return(0); 994 if ((atab = (ATDIR **)calloc(A_TAB_SZ, sizeof(ATDIR *))) == NULL) { 995 paxwarn(1,"Cannot allocate space for directory access time table"); 996 return(-1); 997 } 998 return(0); 999} 1000 1001 1002/* 1003 * atdir_end() 1004 * walk through the directory access time table and reset the access time 1005 * of any directory who still has an entry left in the database. These 1006 * entries are for directories READ by pax 1007 */ 1008 1009#ifdef __STDC__ 1010void 1011atdir_end(void) 1012#else 1013void 1014atdir_end() 1015#endif 1016{ 1017 register ATDIR *pt; 1018 register int i; 1019 1020 if (atab == NULL) 1021 return; 1022 /* 1023 * for each non-empty hash table entry reset all the directories 1024 * chained there. 1025 */ 1026 for (i = 0; i < A_TAB_SZ; ++i) { 1027 if ((pt = atab[i]) == NULL) 1028 continue; 1029 /* 1030 * remember to force the times, set_ftime() looks at pmtime 1031 * and patime, which only applies to things CREATED by pax, 1032 * not read by pax. Read time reset is controlled by -t. 1033 */ 1034 for (; pt != NULL; pt = pt->fow) 1035 set_ftime(pt->name, pt->mtime, pt->atime, 1); 1036 } 1037} 1038 1039/* 1040 * add_atdir() 1041 * add a directory to the directory access time table. Table is hashed 1042 * and chained by inode number. This is for directories READ by pax 1043 */ 1044 1045#ifdef __STDC__ 1046void 1047add_atdir(char *fname, dev_t dev, ino_t ino, time_t mtime, time_t atime) 1048#else 1049void 1050add_atdir(fname, dev, ino, mtime, atime) 1051 char *fname; 1052 dev_t dev; 1053 ino_t ino; 1054 time_t mtime; 1055 time_t atime; 1056#endif 1057{ 1058 register ATDIR *pt; 1059 register u_int indx; 1060 1061 if (atab == NULL) 1062 return; 1063 1064 /* 1065 * make sure this directory is not already in the table, if so just 1066 * return (the older entry always has the correct time). The only 1067 * way this will happen is when the same subtree can be traversed by 1068 * different args to pax and the -n option is aborting fts out of a 1069 * subtree before all the post-order visits have been made). 1070 */ 1071 indx = ((unsigned)ino) % A_TAB_SZ; 1072 if ((pt = atab[indx]) != NULL) { 1073 while (pt != NULL) { 1074 if ((pt->ino == ino) && (pt->dev == dev)) 1075 break; 1076 pt = pt->fow; 1077 } 1078 1079 /* 1080 * oops, already there. Leave it alone. 1081 */ 1082 if (pt != NULL) 1083 return; 1084 } 1085 1086 /* 1087 * add it to the front of the hash chain 1088 */ 1089 if ((pt = (ATDIR *)malloc(sizeof(ATDIR))) != NULL) { 1090 if ((pt->name = strdup(fname)) != NULL) { 1091 pt->dev = dev; 1092 pt->ino = ino; 1093 pt->mtime = mtime; 1094 pt->atime = atime; 1095 pt->fow = atab[indx]; 1096 atab[indx] = pt; 1097 return; 1098 } 1099 (void)free((char *)pt); 1100 } 1101 1102 paxwarn(1, "Directory access time reset table ran out of memory"); 1103 return; 1104} 1105 1106/* 1107 * get_atdir() 1108 * look up a directory by inode and device number to obtain the access 1109 * and modification time you want to set to. If found, the modification 1110 * and access time parameters are set and the entry is removed from the 1111 * table (as it is no longer needed). These are for directories READ by 1112 * pax 1113 * Return: 1114 * 0 if found, -1 if not found. 1115 */ 1116 1117#ifdef __STDC__ 1118int 1119get_atdir(dev_t dev, ino_t ino, time_t *mtime, time_t *atime) 1120#else 1121int 1122get_atdir(dev, ino, mtime, atime) 1123 dev_t dev; 1124 ino_t ino; 1125 time_t *mtime; 1126 time_t *atime; 1127#endif 1128{ 1129 register ATDIR *pt; 1130 register ATDIR **ppt; 1131 register u_int indx; 1132 1133 if (atab == NULL) 1134 return(-1); 1135 /* 1136 * hash by inode and search the chain for an inode and device match 1137 */ 1138 indx = ((unsigned)ino) % A_TAB_SZ; 1139 if ((pt = atab[indx]) == NULL) 1140 return(-1); 1141 1142 ppt = &(atab[indx]); 1143 while (pt != NULL) { 1144 if ((pt->ino == ino) && (pt->dev == dev)) 1145 break; 1146 /* 1147 * no match, go to next one 1148 */ 1149 ppt = &(pt->fow); 1150 pt = pt->fow; 1151 } 1152 1153 /* 1154 * return if we did not find it. 1155 */ 1156 if (pt == NULL) 1157 return(-1); 1158 1159 /* 1160 * found it. return the times and remove the entry from the table. 1161 */ 1162 *ppt = pt->fow; 1163 *mtime = pt->mtime; 1164 *atime = pt->atime; 1165 (void)free((char *)pt->name); 1166 (void)free((char *)pt); 1167 return(0); 1168} 1169 1170/* 1171 * directory access mode and time storage routines (for directories CREATED 1172 * by pax). 1173 * 1174 * Pax requires that extracted directories, by default, have their access/mod 1175 * times and permissions set to the values specified in the archive. During the 1176 * actions of extracting (and creating the destination subtree during -rw copy) 1177 * directories extracted may be modified after being created. Even worse is 1178 * that these directories may have been created with file permissions which 1179 * prohibits any descendants of these directories from being extracted. When 1180 * directories are created by pax, access rights may be added to permit the 1181 * creation of files in their subtree. Every time pax creates a directory, the 1182 * times and file permissions specified by the archive are stored. After all 1183 * files have been extracted (or copied), these directories have their times 1184 * and file modes reset to the stored values. The directory info is restored in 1185 * reverse order as entries were added to the data file from root to leaf. To 1186 * restore atime properly, we must go backwards. The data file consists of 1187 * records with two parts, the file name followed by a DIRDATA trailer. The 1188 * fixed sized trailer contains the size of the name plus the off_t location in 1189 * the file. To restore we work backwards through the file reading the trailer 1190 * then the file name. 1191 */ 1192 1193/* 1194 * dir_start() 1195 * set up the directory time and file mode storage for directories CREATED 1196 * by pax. 1197 * Return: 1198 * 0 if ok, -1 otherwise 1199 */ 1200 1201#ifdef __STDC__ 1202int 1203dir_start(void) 1204#else 1205int 1206dir_start() 1207#endif 1208{ 1209 if (dirfd != -1) 1210 return(0); 1211 1212 /* 1213 * unlink the file so it goes away at termination by itself 1214 */ 1215 memcpy(tempbase, _TFILE_BASE, sizeof(_TFILE_BASE)); 1216 if ((dirfd = mkstemp(tempfile)) >= 0) { 1217 (void)unlink(tempfile); 1218 return(0); 1219 } 1220 paxwarn(1, "Unable to create temporary file for directory times: %s", 1221 tempfile); 1222 return(-1); 1223} 1224 1225/* 1226 * add_dir() 1227 * add the mode and times for a newly CREATED directory 1228 * name is name of the directory, psb the stat buffer with the data in it, 1229 * frc_mode is a flag that says whether to force the setting of the mode 1230 * (ignoring the user set values for preserving file mode). Frc_mode is 1231 * for the case where we created a file and found that the resulting 1232 * directory was not writeable and the user asked for file modes to NOT 1233 * be preserved. (we have to preserve what was created by default, so we 1234 * have to force the setting at the end. this is stated explicitly in the 1235 * pax spec) 1236 */ 1237 1238#ifdef __STDC__ 1239void 1240add_dir(char *name, int nlen, struct stat *psb, int frc_mode) 1241#else 1242void 1243add_dir(name, nlen, psb, frc_mode) 1244 char *name; 1245 int nlen; 1246 struct stat *psb; 1247 int frc_mode; 1248#endif 1249{ 1250 DIRDATA dblk; 1251 1252 if (dirfd < 0) 1253 return; 1254 1255 /* 1256 * get current position (where file name will start) so we can store it 1257 * in the trailer 1258 */ 1259 if ((dblk.npos = lseek(dirfd, 0L, SEEK_CUR)) < 0) { 1260 paxwarn(1,"Unable to store mode and times for directory: %s",name); 1261 return; 1262 } 1263 1264 /* 1265 * write the file name followed by the trailer 1266 */ 1267 dblk.nlen = nlen + 1; 1268 dblk.mode = psb->st_mode & 0xffff; 1269 dblk.mtime = psb->st_mtime; 1270 dblk.atime = psb->st_atime; 1271 dblk.frc_mode = frc_mode; 1272 if ((write(dirfd, name, dblk.nlen) == dblk.nlen) && 1273 (write(dirfd, (char *)&dblk, sizeof(dblk)) == sizeof(dblk))) { 1274 ++dircnt; 1275 return; 1276 } 1277 1278 paxwarn(1,"Unable to store mode and times for created directory: %s",name); 1279 return; 1280} 1281 1282/* 1283 * proc_dir() 1284 * process all file modes and times stored for directories CREATED 1285 * by pax 1286 */ 1287 1288#ifdef __STDC__ 1289void 1290proc_dir(void) 1291#else 1292void 1293proc_dir() 1294#endif 1295{ 1296 char name[PAXPATHLEN+1]; 1297 DIRDATA dblk; 1298 u_long cnt; 1299 1300 if (dirfd < 0) 1301 return; 1302 /* 1303 * read backwards through the file and process each directory 1304 */ 1305 for (cnt = 0; cnt < dircnt; ++cnt) { 1306 /* 1307 * read the trailer, then the file name, if this fails 1308 * just give up. 1309 */ 1310 if (lseek(dirfd, -((off_t)sizeof(dblk)), SEEK_CUR) < 0) 1311 break; 1312 if (read(dirfd,(char *)&dblk, sizeof(dblk)) != sizeof(dblk)) 1313 break; 1314 if (lseek(dirfd, dblk.npos, SEEK_SET) < 0) 1315 break; 1316 if (read(dirfd, name, dblk.nlen) != dblk.nlen) 1317 break; 1318 if (lseek(dirfd, dblk.npos, SEEK_SET) < 0) 1319 break; 1320 1321 /* 1322 * frc_mode set, make sure we set the file modes even if 1323 * the user didn't ask for it (see file_subs.c for more info) 1324 */ 1325 if (pmode || dblk.frc_mode) 1326 set_pmode(name, dblk.mode); 1327 if (patime || pmtime) 1328 set_ftime(name, dblk.mtime, dblk.atime, 0); 1329 } 1330 1331 (void)close(dirfd); 1332 dirfd = -1; 1333 if (cnt != dircnt) 1334 paxwarn(1,"Unable to set mode and times for created directories"); 1335 return; 1336} 1337 1338/* 1339 * database independent routines 1340 */ 1341 1342/* 1343 * st_hash() 1344 * hashes filenames to a u_int for hashing into a table. Looks at the tail 1345 * end of file, as this provides far better distribution than any other 1346 * part of the name. For performance reasons we only care about the last 1347 * MAXKEYLEN chars (should be at LEAST large enough to pick off the file 1348 * name). Was tested on 500,000 name file tree traversal from the root 1349 * and gave almost a perfectly uniform distribution of keys when used with 1350 * prime sized tables (MAXKEYLEN was 128 in test). Hashes (sizeof int) 1351 * chars at a time and pads with 0 for last addition. 1352 * Return: 1353 * the hash value of the string MOD (%) the table size. 1354 */ 1355 1356#ifdef __STDC__ 1357u_int 1358st_hash(char *name, int len, int tabsz) 1359#else 1360u_int 1361st_hash(name, len, tabsz) 1362 char *name; 1363 int len; 1364 int tabsz; 1365#endif 1366{ 1367 register char *pt; 1368 register char *dest; 1369 register char *end; 1370 register int i; 1371 register u_int key = 0; 1372 register int steps; 1373 register int res; 1374 u_int val; 1375 1376 /* 1377 * only look at the tail up to MAXKEYLEN, we do not need to waste 1378 * time here (remember these are pathnames, the tail is what will 1379 * spread out the keys) 1380 */ 1381 if (len > MAXKEYLEN) { 1382 pt = &(name[len - MAXKEYLEN]); 1383 len = MAXKEYLEN; 1384 } else 1385 pt = name; 1386 1387 /* 1388 * calculate the number of u_int size steps in the string and if 1389 * there is a runt to deal with 1390 */ 1391 steps = len/sizeof(u_int); 1392 res = len % sizeof(u_int); 1393 1394 /* 1395 * add up the value of the string in unsigned integer sized pieces 1396 * too bad we cannot have unsigned int aligned strings, then we 1397 * could avoid the expensive copy. 1398 */ 1399 for (i = 0; i < steps; ++i) { 1400 end = pt + sizeof(u_int); 1401 dest = (char *)&val; 1402 while (pt < end) 1403 *dest++ = *pt++; 1404 key += val; 1405 } 1406 1407 /* 1408 * add in the runt padded with zero to the right 1409 */ 1410 if (res) { 1411 val = 0; 1412 end = pt + res; 1413 dest = (char *)&val; 1414 while (pt < end) 1415 *dest++ = *pt++; 1416 key += val; 1417 } 1418 1419 /* 1420 * return the result mod the table size 1421 */ 1422 return(key % tabsz); 1423} 1424