1 /* $NetBSD: tables.c,v 1.17 2002/01/31 19:27:54 tv Exp $ */ 2 3 /*- 4 * Copyright (c) 1992 Keith Muller. 5 * Copyright (c) 1992, 1993 6 * The Regents of the University of California. All rights reserved. 7 * 8 * This code is derived from software contributed to Berkeley by 9 * Keith Muller of the University of California, San Diego. 10 * 11 * Redistribution and use in source and binary forms, with or without 12 * modification, are permitted provided that the following conditions 13 * are met: 14 * 1. Redistributions of source code must retain the above copyright 15 * notice, this list of conditions and the following disclaimer. 16 * 2. Redistributions in binary form must reproduce the above copyright 17 * notice, this list of conditions and the following disclaimer in the 18 * documentation and/or other materials provided with the distribution. 19 * 3. All advertising materials mentioning features or use of this software 20 * must display the following acknowledgement: 21 * This product includes software developed by the University of 22 * California, Berkeley and its contributors. 23 * 4. Neither the name of the University nor the names of its contributors 24 * may be used to endorse or promote products derived from this software 25 * without specific prior written permission. 26 * 27 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 28 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 29 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 30 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 31 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 32 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 33 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 34 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 35 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 36 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 37 * SUCH DAMAGE. 38 */ 39 40 #include <sys/cdefs.h> 41 #if defined(__RCSID) && !defined(lint) 42 #if 0 43 static char sccsid[] = "@(#)tables.c 8.1 (Berkeley) 5/31/93"; 44 #else 45 __RCSID("$NetBSD: tables.c,v 1.17 2002/01/31 19:27:54 tv Exp $"); 46 #endif 47 #endif /* not lint */ 48 49 #include <sys/types.h> 50 #include <sys/time.h> 51 #include <sys/stat.h> 52 #include <sys/param.h> 53 #include <stdio.h> 54 #include <ctype.h> 55 #include <fcntl.h> 56 #include <paths.h> 57 #include <string.h> 58 #include <unistd.h> 59 #include <errno.h> 60 #include <stdlib.h> 61 #include "pax.h" 62 #include "tables.h" 63 #include "extern.h" 64 65 /* 66 * Routines for controlling the contents of all the different databases pax 67 * keeps. Tables are dynamically created only when they are needed. The 68 * goal was speed and the ability to work with HUGE archives. The databases 69 * were kept simple, but do have complex rules for when the contents change. 70 * As of this writing, the posix library functions were more complex than 71 * needed for this application (pax databases have very short lifetimes and 72 * do not survive after pax is finished). Pax is required to handle very 73 * large archives. These database routines carefully combine memory usage and 74 * temporary file storage in ways which will not significantly impact runtime 75 * performance while allowing the largest possible archives to be handled. 76 * Trying to force the fit to the posix databases routines was not considered 77 * time well spent. 78 */ 79 80 static HRDLNK **ltab = NULL; /* hard link table for detecting hard links */ 81 static FTM **ftab = NULL; /* file time table for updating arch */ 82 static NAMT **ntab = NULL; /* interactive rename storage table */ 83 static DEVT **dtab = NULL; /* device/inode mapping tables */ 84 static ATDIR **atab = NULL; /* file tree directory time reset table */ 85 #ifdef DIRS_USE_FILE 86 static int dirfd = -1; /* storage for setting created dir time/mode */ 87 static u_long dircnt; /* entries in dir time/mode storage */ 88 #endif 89 static int ffd = -1; /* tmp file for file time table name storage */ 90 91 static DEVT *chk_dev(dev_t, int); 92 93 /* 94 * hard link table routines 95 * 96 * The hard link table tries to detect hard links to files using the device and 97 * inode values. We do this when writing an archive, so we can tell the format 98 * write routine that this file is a hard link to another file. The format 99 * write routine then can store this file in whatever way it wants (as a hard 100 * link if the format supports that like tar, or ignore this info like cpio). 101 * (Actually a field in the format driver table tells us if the format wants 102 * hard link info. if not, we do not waste time looking for them). We also use 103 * the same table when reading an archive. In that situation, this table is 104 * used by the format read routine to detect hard links from stored dev and 105 * inode numbers (like cpio). This will allow pax to create a link when one 106 * can be detected by the archive format. 107 */ 108 109 /* 110 * lnk_start 111 * Creates the hard link table. 112 * Return: 113 * 0 if created, -1 if failure 114 */ 115 116 int 117 lnk_start(void) 118 { 119 if (ltab != NULL) 120 return(0); 121 if ((ltab = (HRDLNK **)calloc(L_TAB_SZ, sizeof(HRDLNK *))) == NULL) { 122 tty_warn(1, "Cannot allocate memory for hard link table"); 123 return(-1); 124 } 125 return(0); 126 } 127 128 /* 129 * chk_lnk() 130 * Looks up entry in hard link hash table. If found, it copies the name 131 * of the file it is linked to (we already saw that file) into ln_name. 132 * lnkcnt is decremented and if goes to 1 the node is deleted from the 133 * database. (We have seen all the links to this file). If not found, 134 * we add the file to the database if it has the potential for having 135 * hard links to other files we may process (it has a link count > 1) 136 * Return: 137 * if found returns 1; if not found returns 0; -1 on error 138 */ 139 140 int 141 chk_lnk(ARCHD *arcn) 142 { 143 HRDLNK *pt; 144 HRDLNK **ppt; 145 u_int indx; 146 147 if (ltab == NULL) 148 return(-1); 149 /* 150 * ignore those nodes that cannot have hard links 151 */ 152 if ((arcn->type == PAX_DIR) || (arcn->sb.st_nlink <= 1)) 153 return(0); 154 155 /* 156 * hash inode number and look for this file 157 */ 158 indx = ((unsigned)arcn->sb.st_ino) % L_TAB_SZ; 159 if ((pt = ltab[indx]) != NULL) { 160 /* 161 * it's hash chain in not empty, walk down looking for it 162 */ 163 ppt = &(ltab[indx]); 164 while (pt != NULL) { 165 if ((pt->ino == arcn->sb.st_ino) && 166 (pt->dev == arcn->sb.st_dev)) 167 break; 168 ppt = &(pt->fow); 169 pt = pt->fow; 170 } 171 172 if (pt != NULL) { 173 /* 174 * found a link. set the node type and copy in the 175 * name of the file it is to link to. we need to 176 * handle hardlinks to regular files differently than 177 * other links. 178 */ 179 arcn->ln_nlen = l_strncpy(arcn->ln_name, pt->name, 180 PAXPATHLEN+1); 181 if (arcn->type == PAX_REG) 182 arcn->type = PAX_HRG; 183 else 184 arcn->type = PAX_HLK; 185 186 /* 187 * if we have found all the links to this file, remove 188 * it from the database 189 */ 190 if (--pt->nlink <= 1) { 191 *ppt = pt->fow; 192 (void)free((char *)pt->name); 193 (void)free((char *)pt); 194 } 195 return(1); 196 } 197 } 198 199 /* 200 * we never saw this file before. It has links so we add it to the 201 * front of this hash chain 202 */ 203 if ((pt = (HRDLNK *)malloc(sizeof(HRDLNK))) != NULL) { 204 if ((pt->name = strdup(arcn->name)) != NULL) { 205 pt->dev = arcn->sb.st_dev; 206 pt->ino = arcn->sb.st_ino; 207 pt->nlink = arcn->sb.st_nlink; 208 pt->fow = ltab[indx]; 209 ltab[indx] = pt; 210 return(0); 211 } 212 (void)free((char *)pt); 213 } 214 215 tty_warn(1, "Hard link table out of memory"); 216 return(-1); 217 } 218 219 /* 220 * purg_lnk 221 * remove reference for a file that we may have added to the data base as 222 * a potential source for hard links. We ended up not using the file, so 223 * we do not want to accidently point another file at it later on. 224 */ 225 226 void 227 purg_lnk(ARCHD *arcn) 228 { 229 HRDLNK *pt; 230 HRDLNK **ppt; 231 u_int indx; 232 233 if (ltab == NULL) 234 return; 235 /* 236 * do not bother to look if it could not be in the database 237 */ 238 if ((arcn->sb.st_nlink <= 1) || (arcn->type == PAX_DIR) || 239 (arcn->type == PAX_HLK) || (arcn->type == PAX_HRG)) 240 return; 241 242 /* 243 * find the hash chain for this inode value, if empty return 244 */ 245 indx = ((unsigned)arcn->sb.st_ino) % L_TAB_SZ; 246 if ((pt = ltab[indx]) == NULL) 247 return; 248 249 /* 250 * walk down the list looking for the inode/dev pair, unlink and 251 * free if found 252 */ 253 ppt = &(ltab[indx]); 254 while (pt != NULL) { 255 if ((pt->ino == arcn->sb.st_ino) && 256 (pt->dev == arcn->sb.st_dev)) 257 break; 258 ppt = &(pt->fow); 259 pt = pt->fow; 260 } 261 if (pt == NULL) 262 return; 263 264 /* 265 * remove and free it 266 */ 267 *ppt = pt->fow; 268 (void)free((char *)pt->name); 269 (void)free((char *)pt); 270 } 271 272 /* 273 * lnk_end() 274 * pull apart a existing link table so we can reuse it. We do this between 275 * read and write phases of append with update. (The format may have 276 * used the link table, and we need to start with a fresh table for the 277 * write phase 278 */ 279 280 void 281 lnk_end(void) 282 { 283 int i; 284 HRDLNK *pt; 285 HRDLNK *ppt; 286 287 if (ltab == NULL) 288 return; 289 290 for (i = 0; i < L_TAB_SZ; ++i) { 291 if (ltab[i] == NULL) 292 continue; 293 pt = ltab[i]; 294 ltab[i] = NULL; 295 296 /* 297 * free up each entry on this chain 298 */ 299 while (pt != NULL) { 300 ppt = pt; 301 pt = ppt->fow; 302 (void)free((char *)ppt->name); 303 (void)free((char *)ppt); 304 } 305 } 306 return; 307 } 308 309 /* 310 * modification time table routines 311 * 312 * The modification time table keeps track of last modification times for all 313 * files stored in an archive during a write phase when -u is set. We only 314 * add a file to the archive if it is newer than a file with the same name 315 * already stored on the archive (if there is no other file with the same 316 * name on the archive it is added). This applies to writes and appends. 317 * An append with an -u must read the archive and store the modification time 318 * for every file on that archive before starting the write phase. It is clear 319 * that this is one HUGE database. To save memory space, the actual file names 320 * are stored in a scatch file and indexed by an in memory hash table. The 321 * hash table is indexed by hashing the file path. The nodes in the table store 322 * the length of the filename and the lseek offset within the scratch file 323 * where the actual name is stored. Since there are never any deletions to this 324 * table, fragmentation of the scratch file is never a issue. Lookups seem to 325 * not exhibit any locality at all (files in the database are rarely 326 * looked up more than once...). So caching is just a waste of memory. The 327 * only limitation is the amount of scatch file space available to store the 328 * path names. 329 */ 330 331 /* 332 * ftime_start() 333 * create the file time hash table and open for read/write the scratch 334 * file. (after created it is unlinked, so when we exit we leave 335 * no witnesses). 336 * Return: 337 * 0 if the table and file was created ok, -1 otherwise 338 */ 339 340 int 341 ftime_start(void) 342 { 343 const char *tmpdir; 344 char template[MAXPATHLEN]; 345 346 if (ftab != NULL) 347 return(0); 348 if ((ftab = (FTM **)calloc(F_TAB_SZ, sizeof(FTM *))) == NULL) { 349 tty_warn(1, "Cannot allocate memory for file time table"); 350 return(-1); 351 } 352 353 /* 354 * get random name and create temporary scratch file, unlink name 355 * so it will get removed on exit 356 */ 357 if ((tmpdir = getenv("TMPDIR")) == NULL) 358 tmpdir = _PATH_TMP; 359 (void)snprintf(template, sizeof(template), "%s/%s", tmpdir, TMPFILE); 360 if ((ffd = mkstemp(template)) == -1) { 361 syswarn(1, errno, "Unable to create temporary file: %s", 362 template); 363 return(-1); 364 } 365 366 (void)unlink(template); 367 return(0); 368 } 369 370 /* 371 * chk_ftime() 372 * looks up entry in file time hash table. If not found, the file is 373 * added to the hash table and the file named stored in the scratch file. 374 * If a file with the same name is found, the file times are compared and 375 * the most recent file time is retained. If the new file was younger (or 376 * was not in the database) the new file is selected for storage. 377 * Return: 378 * 0 if file should be added to the archive, 1 if it should be skipped, 379 * -1 on error 380 */ 381 382 int 383 chk_ftime(ARCHD *arcn) 384 { 385 FTM *pt; 386 int namelen; 387 u_int indx; 388 char ckname[PAXPATHLEN+1]; 389 390 /* 391 * no info, go ahead and add to archive 392 */ 393 if (ftab == NULL) 394 return(0); 395 396 /* 397 * hash the pathname and look up in table 398 */ 399 namelen = arcn->nlen; 400 indx = st_hash(arcn->name, namelen, F_TAB_SZ); 401 if ((pt = ftab[indx]) != NULL) { 402 /* 403 * the hash chain is not empty, walk down looking for match 404 * only read up the path names if the lengths match, speeds 405 * up the search a lot 406 */ 407 while (pt != NULL) { 408 if (pt->namelen == namelen) { 409 /* 410 * potential match, have to read the name 411 * from the scratch file. 412 */ 413 if (lseek(ffd,pt->seek,SEEK_SET) != pt->seek) { 414 syswarn(1, errno, 415 "Failed ftime table seek"); 416 return(-1); 417 } 418 if (xread(ffd, ckname, namelen) != namelen) { 419 syswarn(1, errno, 420 "Failed ftime table read"); 421 return(-1); 422 } 423 424 /* 425 * if the names match, we are done 426 */ 427 if (!strncmp(ckname, arcn->name, namelen)) 428 break; 429 } 430 431 /* 432 * try the next entry on the chain 433 */ 434 pt = pt->fow; 435 } 436 437 if (pt != NULL) { 438 /* 439 * found the file, compare the times, save the newer 440 */ 441 if (arcn->sb.st_mtime > pt->mtime) { 442 /* 443 * file is newer 444 */ 445 pt->mtime = arcn->sb.st_mtime; 446 return(0); 447 } 448 /* 449 * file is older 450 */ 451 return(1); 452 } 453 } 454 455 /* 456 * not in table, add it 457 */ 458 if ((pt = (FTM *)malloc(sizeof(FTM))) != NULL) { 459 /* 460 * add the name at the end of the scratch file, saving the 461 * offset. add the file to the head of the hash chain 462 */ 463 if ((pt->seek = lseek(ffd, (off_t)0, SEEK_END)) >= 0) { 464 if (xwrite(ffd, arcn->name, namelen) == namelen) { 465 pt->mtime = arcn->sb.st_mtime; 466 pt->namelen = namelen; 467 pt->fow = ftab[indx]; 468 ftab[indx] = pt; 469 return(0); 470 } 471 syswarn(1, errno, "Failed write to file time table"); 472 } else 473 syswarn(1, errno, "Failed seek on file time table"); 474 } else 475 tty_warn(1, "File time table ran out of memory"); 476 477 if (pt != NULL) 478 (void)free((char *)pt); 479 return(-1); 480 } 481 482 /* 483 * Interactive rename table routines 484 * 485 * The interactive rename table keeps track of the new names that the user 486 * assigns to files from tty input. Since this map is unique for each file 487 * we must store it in case there is a reference to the file later in archive 488 * (a link). Otherwise we will be unable to find the file we know was 489 * extracted. The remapping of these files is stored in a memory based hash 490 * table (it is assumed since input must come from /dev/tty, it is unlikely to 491 * be a very large table). 492 */ 493 494 /* 495 * name_start() 496 * create the interactive rename table 497 * Return: 498 * 0 if successful, -1 otherwise 499 */ 500 501 int 502 name_start(void) 503 { 504 if (ntab != NULL) 505 return(0); 506 if ((ntab = (NAMT **)calloc(N_TAB_SZ, sizeof(NAMT *))) == NULL) { 507 tty_warn(1, 508 "Cannot allocate memory for interactive rename table"); 509 return(-1); 510 } 511 return(0); 512 } 513 514 /* 515 * add_name() 516 * add the new name to old name mapping just created by the user. 517 * If an old name mapping is found (there may be duplicate names on an 518 * archive) only the most recent is kept. 519 * Return: 520 * 0 if added, -1 otherwise 521 */ 522 523 int 524 add_name(char *oname, int onamelen, char *nname) 525 { 526 NAMT *pt; 527 u_int indx; 528 529 if (ntab == NULL) { 530 /* 531 * should never happen 532 */ 533 tty_warn(0, "No interactive rename table, links may fail\n"); 534 return(0); 535 } 536 537 /* 538 * look to see if we have already mapped this file, if so we 539 * will update it 540 */ 541 indx = st_hash(oname, onamelen, N_TAB_SZ); 542 if ((pt = ntab[indx]) != NULL) { 543 /* 544 * look down the has chain for the file 545 */ 546 while ((pt != NULL) && (strcmp(oname, pt->oname) != 0)) 547 pt = pt->fow; 548 549 if (pt != NULL) { 550 /* 551 * found an old mapping, replace it with the new one 552 * the user just input (if it is different) 553 */ 554 if (strcmp(nname, pt->nname) == 0) 555 return(0); 556 557 (void)free((char *)pt->nname); 558 if ((pt->nname = strdup(nname)) == NULL) { 559 tty_warn(1, "Cannot update rename table"); 560 return(-1); 561 } 562 return(0); 563 } 564 } 565 566 /* 567 * this is a new mapping, add it to the table 568 */ 569 if ((pt = (NAMT *)malloc(sizeof(NAMT))) != NULL) { 570 if ((pt->oname = strdup(oname)) != NULL) { 571 if ((pt->nname = strdup(nname)) != NULL) { 572 pt->fow = ntab[indx]; 573 ntab[indx] = pt; 574 return(0); 575 } 576 (void)free((char *)pt->oname); 577 } 578 (void)free((char *)pt); 579 } 580 tty_warn(1, "Interactive rename table out of memory"); 581 return(-1); 582 } 583 584 /* 585 * sub_name() 586 * look up a link name to see if it points at a file that has been 587 * remapped by the user. If found, the link is adjusted to contain the 588 * new name (oname is the link to name) 589 */ 590 591 void 592 sub_name(char *oname, int *onamelen) 593 { 594 NAMT *pt; 595 u_int indx; 596 597 if (ntab == NULL) 598 return; 599 /* 600 * look the name up in the hash table 601 */ 602 indx = st_hash(oname, *onamelen, N_TAB_SZ); 603 if ((pt = ntab[indx]) == NULL) 604 return; 605 606 while (pt != NULL) { 607 /* 608 * walk down the hash chain looking for a match 609 */ 610 if (strcmp(oname, pt->oname) == 0) { 611 /* 612 * found it, replace it with the new name 613 * and return (we know that oname has enough space) 614 */ 615 *onamelen = l_strncpy(oname, pt->nname, PAXPATHLEN+1); 616 return; 617 } 618 pt = pt->fow; 619 } 620 621 /* 622 * no match, just return 623 */ 624 return; 625 } 626 627 /* 628 * device/inode mapping table routines 629 * (used with formats that store device and inodes fields) 630 * 631 * device/inode mapping tables remap the device field in a archive header. The 632 * device/inode fields are used to determine when files are hard links to each 633 * other. However these values have very little meaning outside of that. This 634 * database is used to solve one of two different problems. 635 * 636 * 1) when files are appended to an archive, while the new files may have hard 637 * links to each other, you cannot determine if they have hard links to any 638 * file already stored on the archive from a prior run of pax. We must assume 639 * that these inode/device pairs are unique only within a SINGLE run of pax 640 * (which adds a set of files to an archive). So we have to make sure the 641 * inode/dev pairs we add each time are always unique. We do this by observing 642 * while the inode field is very dense, the use of the dev field is fairly 643 * sparse. Within each run of pax, we remap any device number of a new archive 644 * member that has a device number used in a prior run and already stored in a 645 * file on the archive. During the read phase of the append, we store the 646 * device numbers used and mark them to not be used by any file during the 647 * write phase. If during write we go to use one of those old device numbers, 648 * we remap it to a new value. 649 * 650 * 2) Often the fields in the archive header used to store these values are 651 * too small to store the entire value. The result is an inode or device value 652 * which can be truncated. This really can foul up an archive. With truncation 653 * we end up creating links between files that are really not links (after 654 * truncation the inodes are the same value). We address that by detecting 655 * truncation and forcing a remap of the device field to split truncated 656 * inodes away from each other. Each truncation creates a pattern of bits that 657 * are removed. We use this pattern of truncated bits to partition the inodes 658 * on a single device to many different devices (each one represented by the 659 * truncated bit pattern). All inodes on the same device that have the same 660 * truncation pattern are mapped to the same new device. Two inodes that 661 * truncate to the same value clearly will always have different truncation 662 * bit patterns, so they will be split from away each other. When we spot 663 * device truncation we remap the device number to a non truncated value. 664 * (for more info see table.h for the data structures involved). 665 */ 666 667 /* 668 * dev_start() 669 * create the device mapping table 670 * Return: 671 * 0 if successful, -1 otherwise 672 */ 673 674 int 675 dev_start(void) 676 { 677 if (dtab != NULL) 678 return(0); 679 if ((dtab = (DEVT **)calloc(D_TAB_SZ, sizeof(DEVT *))) == NULL) { 680 tty_warn(1, "Cannot allocate memory for device mapping table"); 681 return(-1); 682 } 683 return(0); 684 } 685 686 /* 687 * add_dev() 688 * add a device number to the table. this will force the device to be 689 * remapped to a new value if it be used during a write phase. This 690 * function is called during the read phase of an append to prohibit the 691 * use of any device number already in the archive. 692 * Return: 693 * 0 if added ok, -1 otherwise 694 */ 695 696 int 697 add_dev(ARCHD *arcn) 698 { 699 if (chk_dev(arcn->sb.st_dev, 1) == NULL) 700 return(-1); 701 return(0); 702 } 703 704 /* 705 * chk_dev() 706 * check for a device value in the device table. If not found and the add 707 * flag is set, it is added. This does NOT assign any mapping values, just 708 * adds the device number as one that need to be remapped. If this device 709 * is already mapped, just return with a pointer to that entry. 710 * Return: 711 * pointer to the entry for this device in the device map table. Null 712 * if the add flag is not set and the device is not in the table (it is 713 * not been seen yet). If add is set and the device cannot be added, null 714 * is returned (indicates an error). 715 */ 716 717 static DEVT * 718 chk_dev(dev_t dev, int add) 719 { 720 DEVT *pt; 721 u_int indx; 722 723 if (dtab == NULL) 724 return(NULL); 725 /* 726 * look to see if this device is already in the table 727 */ 728 indx = ((unsigned)dev) % D_TAB_SZ; 729 if ((pt = dtab[indx]) != NULL) { 730 while ((pt != NULL) && (pt->dev != dev)) 731 pt = pt->fow; 732 733 /* 734 * found it, return a pointer to it 735 */ 736 if (pt != NULL) 737 return(pt); 738 } 739 740 /* 741 * not in table, we add it only if told to as this may just be a check 742 * to see if a device number is being used. 743 */ 744 if (add == 0) 745 return(NULL); 746 747 /* 748 * allocate a node for this device and add it to the front of the hash 749 * chain. Note we do not assign remaps values here, so the pt->list 750 * list must be NULL. 751 */ 752 if ((pt = (DEVT *)malloc(sizeof(DEVT))) == NULL) { 753 tty_warn(1, "Device map table out of memory"); 754 return(NULL); 755 } 756 pt->dev = dev; 757 pt->list = NULL; 758 pt->fow = dtab[indx]; 759 dtab[indx] = pt; 760 return(pt); 761 } 762 /* 763 * map_dev() 764 * given an inode and device storage mask (the mask has a 1 for each bit 765 * the archive format is able to store in a header), we check for inode 766 * and device truncation and remap the device as required. Device mapping 767 * can also occur when during the read phase of append a device number was 768 * seen (and was marked as do not use during the write phase). WE ASSUME 769 * that unsigned longs are the same size or bigger than the fields used 770 * for ino_t and dev_t. If not the types will have to be changed. 771 * Return: 772 * 0 if all ok, -1 otherwise. 773 */ 774 775 int 776 map_dev(ARCHD *arcn, u_long dev_mask, u_long ino_mask) 777 { 778 DEVT *pt; 779 DLIST *dpt; 780 static dev_t lastdev = 0; /* next device number to try */ 781 int trc_ino = 0; 782 int trc_dev = 0; 783 ino_t trunc_bits = 0; 784 ino_t nino; 785 786 if (dtab == NULL) 787 return(0); 788 /* 789 * check for device and inode truncation, and extract the truncated 790 * bit pattern. 791 */ 792 if ((arcn->sb.st_dev & (dev_t)dev_mask) != arcn->sb.st_dev) 793 ++trc_dev; 794 if ((nino = arcn->sb.st_ino & (ino_t)ino_mask) != arcn->sb.st_ino) { 795 ++trc_ino; 796 trunc_bits = arcn->sb.st_ino & (ino_t)(~ino_mask); 797 } 798 799 /* 800 * see if this device is already being mapped, look up the device 801 * then find the truncation bit pattern which applies 802 */ 803 if ((pt = chk_dev(arcn->sb.st_dev, 0)) != NULL) { 804 /* 805 * this device is already marked to be remapped 806 */ 807 for (dpt = pt->list; dpt != NULL; dpt = dpt->fow) 808 if (dpt->trunc_bits == trunc_bits) 809 break; 810 811 if (dpt != NULL) { 812 /* 813 * we are being remapped for this device and pattern 814 * change the device number to be stored and return 815 */ 816 arcn->sb.st_dev = dpt->dev; 817 arcn->sb.st_ino = nino; 818 return(0); 819 } 820 } else { 821 /* 822 * this device is not being remapped YET. if we do not have any 823 * form of truncation, we do not need a remap 824 */ 825 if (!trc_ino && !trc_dev) 826 return(0); 827 828 /* 829 * we have truncation, have to add this as a device to remap 830 */ 831 if ((pt = chk_dev(arcn->sb.st_dev, 1)) == NULL) 832 goto bad; 833 834 /* 835 * if we just have a truncated inode, we have to make sure that 836 * all future inodes that do not truncate (they have the 837 * truncation pattern of all 0's) continue to map to the same 838 * device number. We probably have already written inodes with 839 * this device number to the archive with the truncation 840 * pattern of all 0's. So we add the mapping for all 0's to the 841 * same device number. 842 */ 843 if (!trc_dev && (trunc_bits != 0)) { 844 if ((dpt = (DLIST *)malloc(sizeof(DLIST))) == NULL) 845 goto bad; 846 dpt->trunc_bits = 0; 847 dpt->dev = arcn->sb.st_dev; 848 dpt->fow = pt->list; 849 pt->list = dpt; 850 } 851 } 852 853 /* 854 * look for a device number not being used. We must watch for wrap 855 * around on lastdev (so we do not get stuck looking forever!) 856 */ 857 while (++lastdev > 0) { 858 if (chk_dev(lastdev, 0) != NULL) 859 continue; 860 /* 861 * found an unused value. If we have reached truncation point 862 * for this format we are hosed, so we give up. Otherwise we 863 * mark it as being used. 864 */ 865 if (((lastdev & ((dev_t)dev_mask)) != lastdev) || 866 (chk_dev(lastdev, 1) == NULL)) 867 goto bad; 868 break; 869 } 870 871 if ((lastdev <= 0) || ((dpt = (DLIST *)malloc(sizeof(DLIST))) == NULL)) 872 goto bad; 873 874 /* 875 * got a new device number, store it under this truncation pattern. 876 * change the device number this file is being stored with. 877 */ 878 dpt->trunc_bits = trunc_bits; 879 dpt->dev = lastdev; 880 dpt->fow = pt->list; 881 pt->list = dpt; 882 arcn->sb.st_dev = lastdev; 883 arcn->sb.st_ino = nino; 884 return(0); 885 886 bad: 887 tty_warn(1, 888 "Unable to fix truncated inode/device field when storing %s", 889 arcn->name); 890 tty_warn(0, "Archive may create improper hard links when extracted"); 891 return(0); 892 } 893 894 /* 895 * directory access/mod time reset table routines (for directories READ by pax) 896 * 897 * The pax -t flag requires that access times of archive files to be the same 898 * before being read by pax. For regular files, access time is restored after 899 * the file has been copied. This database provides the same functionality for 900 * directories read during file tree traversal. Restoring directory access time 901 * is more complex than files since directories may be read several times until 902 * all the descendants in their subtree are visited by fts. Directory access 903 * and modification times are stored during the fts pre-order visit (done 904 * before any descendants in the subtree is visited) and restored after the 905 * fts post-order visit (after all the descendants have been visited). In the 906 * case of premature exit from a subtree (like from the effects of -n), any 907 * directory entries left in this database are reset during final cleanup 908 * operations of pax. Entries are hashed by inode number for fast lookup. 909 */ 910 911 /* 912 * atdir_start() 913 * create the directory access time database for directories READ by pax. 914 * Return: 915 * 0 is created ok, -1 otherwise. 916 */ 917 918 int 919 atdir_start(void) 920 { 921 if (atab != NULL) 922 return(0); 923 if ((atab = (ATDIR **)calloc(A_TAB_SZ, sizeof(ATDIR *))) == NULL) { 924 tty_warn(1, 925 "Cannot allocate space for directory access time table"); 926 return(-1); 927 } 928 return(0); 929 } 930 931 932 /* 933 * atdir_end() 934 * walk through the directory access time table and reset the access time 935 * of any directory who still has an entry left in the database. These 936 * entries are for directories READ by pax 937 */ 938 939 void 940 atdir_end(void) 941 { 942 ATDIR *pt; 943 int i; 944 945 if (atab == NULL) 946 return; 947 /* 948 * for each non-empty hash table entry reset all the directories 949 * chained there. 950 */ 951 for (i = 0; i < A_TAB_SZ; ++i) { 952 if ((pt = atab[i]) == NULL) 953 continue; 954 /* 955 * remember to force the times, set_ftime() looks at pmtime 956 * and patime, which only applies to things CREATED by pax, 957 * not read by pax. Read time reset is controlled by -t. 958 */ 959 for (; pt != NULL; pt = pt->fow) 960 set_ftime(pt->name, pt->mtime, pt->atime, 1); 961 } 962 } 963 964 /* 965 * add_atdir() 966 * add a directory to the directory access time table. Table is hashed 967 * and chained by inode number. This is for directories READ by pax 968 */ 969 970 void 971 add_atdir(char *fname, dev_t dev, ino_t ino, time_t mtime, time_t atime) 972 { 973 ATDIR *pt; 974 u_int indx; 975 976 if (atab == NULL) 977 return; 978 979 /* 980 * make sure this directory is not already in the table, if so just 981 * return (the older entry always has the correct time). The only 982 * way this will happen is when the same subtree can be traversed by 983 * different args to pax and the -n option is aborting fts out of a 984 * subtree before all the post-order visits have been made). 985 */ 986 indx = ((unsigned)ino) % A_TAB_SZ; 987 if ((pt = atab[indx]) != NULL) { 988 while (pt != NULL) { 989 if ((pt->ino == ino) && (pt->dev == dev)) 990 break; 991 pt = pt->fow; 992 } 993 994 /* 995 * oops, already there. Leave it alone. 996 */ 997 if (pt != NULL) 998 return; 999 } 1000 1001 /* 1002 * add it to the front of the hash chain 1003 */ 1004 if ((pt = (ATDIR *)malloc(sizeof(ATDIR))) != NULL) { 1005 if ((pt->name = strdup(fname)) != NULL) { 1006 pt->dev = dev; 1007 pt->ino = ino; 1008 pt->mtime = mtime; 1009 pt->atime = atime; 1010 pt->fow = atab[indx]; 1011 atab[indx] = pt; 1012 return; 1013 } 1014 (void)free((char *)pt); 1015 } 1016 1017 tty_warn(1, "Directory access time reset table ran out of memory"); 1018 return; 1019 } 1020 1021 /* 1022 * get_atdir() 1023 * look up a directory by inode and device number to obtain the access 1024 * and modification time you want to set to. If found, the modification 1025 * and access time parameters are set and the entry is removed from the 1026 * table (as it is no longer needed). These are for directories READ by 1027 * pax 1028 * Return: 1029 * 0 if found, -1 if not found. 1030 */ 1031 1032 int 1033 get_atdir(dev_t dev, ino_t ino, time_t *mtime, time_t *atime) 1034 { 1035 ATDIR *pt; 1036 ATDIR **ppt; 1037 u_int indx; 1038 1039 if (atab == NULL) 1040 return(-1); 1041 /* 1042 * hash by inode and search the chain for an inode and device match 1043 */ 1044 indx = ((unsigned)ino) % A_TAB_SZ; 1045 if ((pt = atab[indx]) == NULL) 1046 return(-1); 1047 1048 ppt = &(atab[indx]); 1049 while (pt != NULL) { 1050 if ((pt->ino == ino) && (pt->dev == dev)) 1051 break; 1052 /* 1053 * no match, go to next one 1054 */ 1055 ppt = &(pt->fow); 1056 pt = pt->fow; 1057 } 1058 1059 /* 1060 * return if we did not find it. 1061 */ 1062 if (pt == NULL) 1063 return(-1); 1064 1065 /* 1066 * found it. return the times and remove the entry from the table. 1067 */ 1068 *ppt = pt->fow; 1069 *mtime = pt->mtime; 1070 *atime = pt->atime; 1071 (void)free((char *)pt->name); 1072 (void)free((char *)pt); 1073 return(0); 1074 } 1075 1076 /* 1077 * directory access mode and time storage routines (for directories CREATED 1078 * by pax). 1079 * 1080 * Pax requires that extracted directories, by default, have their access/mod 1081 * times and permissions set to the values specified in the archive. During the 1082 * actions of extracting (and creating the destination subtree during -rw copy) 1083 * directories extracted may be modified after being created. Even worse is 1084 * that these directories may have been created with file permissions which 1085 * prohibits any descendants of these directories from being extracted. When 1086 * directories are created by pax, access rights may be added to permit the 1087 * creation of files in their subtree. Every time pax creates a directory, the 1088 * times and file permissions specified by the archive are stored. After all 1089 * files have been extracted (or copied), these directories have their times 1090 * and file modes reset to the stored values. The directory info is restored in 1091 * reverse order as entries were added to the data file from root to leaf. To 1092 * restore atime properly, we must go backwards. The data file consists of 1093 * records with two parts, the file name followed by a DIRDATA trailer. The 1094 * fixed sized trailer contains the size of the name plus the off_t location in 1095 * the file. To restore we work backwards through the file reading the trailer 1096 * then the file name. 1097 */ 1098 1099 #ifndef DIRS_USE_FILE 1100 static DIRDATA *dirdata_head; 1101 #endif 1102 1103 /* 1104 * dir_start() 1105 * set up the directory time and file mode storage for directories CREATED 1106 * by pax. 1107 * Return: 1108 * 0 if ok, -1 otherwise 1109 */ 1110 1111 int 1112 dir_start(void) 1113 { 1114 #ifdef DIRS_USE_FILE 1115 const char *tmpdir; 1116 char template[MAXPATHLEN]; 1117 1118 if (dirfd != -1) 1119 return(0); 1120 1121 /* 1122 * unlink the file so it goes away at termination by itself 1123 */ 1124 if ((tmpdir = getenv("TMPDIR")) == NULL) 1125 tmpdir = _PATH_TMP; 1126 (void)snprintf(template, sizeof(template), "%s/%s", tmpdir, TMPFILE); 1127 if ((dirfd = mkstemp(template)) >= 0) { 1128 (void)unlink(template); 1129 return(0); 1130 } 1131 tty_warn(1, "Unable to create temporary file for directory times: %s", 1132 template); 1133 return(-1); 1134 #else 1135 return (0); 1136 #endif /* DIRS_USE_FILE */ 1137 } 1138 1139 /* 1140 * add_dir() 1141 * add the mode and times for a newly CREATED directory 1142 * name is name of the directory, psb the stat buffer with the data in it, 1143 * frc_mode is a flag that says whether to force the setting of the mode 1144 * (ignoring the user set values for preserving file mode). Frc_mode is 1145 * for the case where we created a file and found that the resulting 1146 * directory was not writeable and the user asked for file modes to NOT 1147 * be preserved. (we have to preserve what was created by default, so we 1148 * have to force the setting at the end. this is stated explicitly in the 1149 * pax spec) 1150 */ 1151 1152 void 1153 add_dir(char *name, int nlen, struct stat *psb, int frc_mode) 1154 { 1155 #ifdef DIRS_USE_FILE 1156 DIRDATA dblk; 1157 1158 if (dirfd < 0) 1159 return; 1160 1161 /* 1162 * get current position (where file name will start) so we can store it 1163 * in the trailer 1164 */ 1165 if ((dblk.npos = lseek(dirfd, 0L, SEEK_CUR)) < 0) { 1166 tty_warn(1, 1167 "Unable to store mode and times for directory: %s",name); 1168 return; 1169 } 1170 1171 /* 1172 * write the file name followed by the trailer 1173 */ 1174 dblk.nlen = nlen + 1; 1175 dblk.mode = psb->st_mode & 0xffff; 1176 dblk.mtime = psb->st_mtime; 1177 dblk.atime = psb->st_atime; 1178 #if HAVE_STRUCT_STAT_ST_FLAGS 1179 dblk.fflags = psb->st_flags; 1180 #else 1181 dblk.fflags = 0; 1182 #endif 1183 dblk.frc_mode = frc_mode; 1184 if ((xwrite(dirfd, name, dblk.nlen) == dblk.nlen) && 1185 (xwrite(dirfd, (char *)&dblk, sizeof(dblk)) == sizeof(dblk))) { 1186 ++dircnt; 1187 return; 1188 } 1189 1190 tty_warn(1, 1191 "Unable to store mode and times for created directory: %s",name); 1192 return; 1193 #else 1194 DIRDATA *dblk; 1195 1196 if ((dblk = malloc(sizeof(*dblk))) == NULL || 1197 (dblk->name = strdup(name)) == NULL) { 1198 tty_warn(1, 1199 "Unable to store mode and times for directory: %s",name); 1200 if (dblk != NULL) 1201 free(dblk); 1202 return; 1203 } 1204 1205 dblk->mode = psb->st_mode & 0xffff; 1206 dblk->mtime = psb->st_mtime; 1207 dblk->atime = psb->st_atime; 1208 #if HAVE_STRUCT_STAT_ST_FLAGS 1209 dblk->fflags = psb->st_flags; 1210 #else 1211 dblk->fflags = 0; 1212 #endif 1213 dblk->frc_mode = frc_mode; 1214 1215 dblk->next = dirdata_head; 1216 dirdata_head = dblk; 1217 return; 1218 #endif /* DIRS_USE_FILE */ 1219 } 1220 1221 /* 1222 * proc_dir() 1223 * process all file modes and times stored for directories CREATED 1224 * by pax 1225 */ 1226 1227 void 1228 proc_dir(void) 1229 { 1230 #ifdef DIRS_USE_FILE 1231 char name[PAXPATHLEN+1]; 1232 DIRDATA dblk; 1233 u_long cnt; 1234 1235 if (dirfd < 0) 1236 return; 1237 /* 1238 * read backwards through the file and process each directory 1239 */ 1240 for (cnt = 0; cnt < dircnt; ++cnt) { 1241 /* 1242 * read the trailer, then the file name, if this fails 1243 * just give up. 1244 */ 1245 if (lseek(dirfd, -((off_t)sizeof(dblk)), SEEK_CUR) < 0) 1246 break; 1247 if (xread(dirfd,(char *)&dblk, sizeof(dblk)) != sizeof(dblk)) 1248 break; 1249 if (lseek(dirfd, dblk.npos, SEEK_SET) < 0) 1250 break; 1251 if (xread(dirfd, name, dblk.nlen) != dblk.nlen) 1252 break; 1253 if (lseek(dirfd, dblk.npos, SEEK_SET) < 0) 1254 break; 1255 1256 /* 1257 * frc_mode set, make sure we set the file modes even if 1258 * the user didn't ask for it (see file_subs.c for more info) 1259 */ 1260 if (pmode || dblk.frc_mode) 1261 set_pmode(name, dblk.mode); 1262 if (patime || pmtime) 1263 set_ftime(name, dblk.mtime, dblk.atime, 0); 1264 if (pfflags) 1265 set_chflags(name, dblk.fflags); 1266 } 1267 1268 (void)close(dirfd); 1269 dirfd = -1; 1270 if (cnt != dircnt) 1271 tty_warn(1, 1272 "Unable to set mode and times for created directories"); 1273 return; 1274 #else 1275 DIRDATA *dblk; 1276 1277 for (dblk = dirdata_head; dblk != NULL; dblk = dirdata_head) { 1278 dirdata_head = dblk->next; 1279 1280 /* 1281 * frc_mode set, make sure we set the file modes even if 1282 * the user didn't ask for it (see file_subs.c for more info) 1283 */ 1284 if (pmode || dblk->frc_mode) 1285 set_pmode(dblk->name, dblk->mode); 1286 if (patime || pmtime) 1287 set_ftime(dblk->name, dblk->mtime, dblk->atime, 0); 1288 if (pfflags) 1289 set_chflags(dblk->name, dblk->fflags); 1290 1291 free(dblk->name); 1292 free(dblk); 1293 } 1294 #endif /* DIRS_USE_FILE */ 1295 } 1296 1297 /* 1298 * database independent routines 1299 */ 1300 1301 /* 1302 * st_hash() 1303 * hashes filenames to a u_int for hashing into a table. Looks at the tail 1304 * end of file, as this provides far better distribution than any other 1305 * part of the name. For performance reasons we only care about the last 1306 * MAXKEYLEN chars (should be at LEAST large enough to pick off the file 1307 * name). Was tested on 500,000 name file tree traversal from the root 1308 * and gave almost a perfectly uniform distribution of keys when used with 1309 * prime sized tables (MAXKEYLEN was 128 in test). Hashes (sizeof int) 1310 * chars at a time and pads with 0 for last addition. 1311 * Return: 1312 * the hash value of the string MOD (%) the table size. 1313 */ 1314 1315 u_int 1316 st_hash(char *name, int len, int tabsz) 1317 { 1318 char *pt; 1319 char *dest; 1320 char *end; 1321 int i; 1322 u_int key = 0; 1323 int steps; 1324 int res; 1325 u_int val; 1326 1327 /* 1328 * only look at the tail up to MAXKEYLEN, we do not need to waste 1329 * time here (remember these are pathnames, the tail is what will 1330 * spread out the keys) 1331 */ 1332 if (len > MAXKEYLEN) { 1333 pt = &(name[len - MAXKEYLEN]); 1334 len = MAXKEYLEN; 1335 } else 1336 pt = name; 1337 1338 /* 1339 * calculate the number of u_int size steps in the string and if 1340 * there is a runt to deal with 1341 */ 1342 steps = len/sizeof(u_int); 1343 res = len % sizeof(u_int); 1344 1345 /* 1346 * add up the value of the string in unsigned integer sized pieces 1347 * too bad we cannot have unsigned int aligned strings, then we 1348 * could avoid the expensive copy. 1349 */ 1350 for (i = 0; i < steps; ++i) { 1351 end = pt + sizeof(u_int); 1352 dest = (char *)&val; 1353 while (pt < end) 1354 *dest++ = *pt++; 1355 key += val; 1356 } 1357 1358 /* 1359 * add in the runt padded with zero to the right 1360 */ 1361 if (res) { 1362 val = 0; 1363 end = pt + res; 1364 dest = (char *)&val; 1365 while (pt < end) 1366 *dest++ = *pt++; 1367 key += val; 1368 } 1369 1370 /* 1371 * return the result mod the table size 1372 */ 1373 return(key % tabsz); 1374 } 1375