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 * @(#)tables.c 8.1 (Berkeley) 5/31/93 38 * $FreeBSD: src/bin/pax/tables.c,v 1.13.2.1 2001/08/01 05:03:12 obrien Exp $ 39 */ 40 41 #include <sys/types.h> 42 #include <sys/time.h> 43 #include <sys/stat.h> 44 #include <sys/fcntl.h> 45 #include <errno.h> 46 #include <stdio.h> 47 #include <stdlib.h> 48 #include <string.h> 49 #include <unistd.h> 50 #include "pax.h" 51 #include "tables.h" 52 #include "extern.h" 53 54 /* 55 * Routines for controlling the contents of all the different databases pax 56 * keeps. Tables are dynamically created only when they are needed. The 57 * goal was speed and the ability to work with HUGE archives. The databases 58 * were kept simple, but do have complex rules for when the contents change. 59 * As of this writing, the POSIX library functions were more complex than 60 * needed for this application (pax databases have very short lifetimes and 61 * do not survive after pax is finished). Pax is required to handle very 62 * large archives. These database routines carefully combine memory usage and 63 * temporary file storage in ways which will not significantly impact runtime 64 * performance while allowing the largest possible archives to be handled. 65 * Trying to force the fit to the POSIX database routines was not considered 66 * time well spent. 67 */ 68 69 static HRDLNK **ltab = NULL; /* hard link table for detecting hard links */ 70 static FTM **ftab = NULL; /* file time table for updating arch */ 71 static NAMT **ntab = NULL; /* interactive rename storage table */ 72 static DEVT **dtab = NULL; /* device/inode mapping tables */ 73 static ATDIR **atab = NULL; /* file tree directory time reset table */ 74 static int dirfd = -1; /* storage for setting created dir time/mode */ 75 static u_long dircnt; /* entries in dir time/mode storage */ 76 static int ffd = -1; /* tmp file for file time table name storage */ 77 78 static DEVT *chk_dev (dev_t, int); 79 80 /* 81 * hard link table routines 82 * 83 * The hard link table tries to detect hard links to files using the device and 84 * inode values. We do this when writing an archive, so we can tell the format 85 * write routine that this file is a hard link to another file. The format 86 * write routine then can store this file in whatever way it wants (as a hard 87 * link if the format supports that like tar, or ignore this info like cpio). 88 * (Actually a field in the format driver table tells us if the format wants 89 * hard link info. if not, we do not waste time looking for them). We also use 90 * the same table when reading an archive. In that situation, this table is 91 * used by the format read routine to detect hard links from stored dev and 92 * inode numbers (like cpio). This will allow pax to create a link when one 93 * can be detected by the archive format. 94 */ 95 96 /* 97 * lnk_start 98 * Creates the hard link table. 99 * Return: 100 * 0 if created, -1 if failure 101 */ 102 103 int 104 lnk_start(void) 105 { 106 if (ltab != NULL) 107 return(0); 108 if ((ltab = (HRDLNK **)calloc(L_TAB_SZ, sizeof(HRDLNK *))) == NULL) { 109 paxwarn(1, "Cannot allocate memory for hard link table"); 110 return(-1); 111 } 112 return(0); 113 } 114 115 /* 116 * chk_lnk() 117 * Looks up entry in hard link hash table. If found, it copies the name 118 * of the file it is linked to (we already saw that file) into ln_name. 119 * lnkcnt is decremented and if goes to 1 the node is deleted from the 120 * database. (We have seen all the links to this file). If not found, 121 * we add the file to the database if it has the potential for having 122 * hard links to other files we may process (it has a link count > 1) 123 * Return: 124 * if found returns 1; if not found returns 0; -1 on error 125 */ 126 127 int 128 chk_lnk(ARCHD *arcn) 129 { 130 HRDLNK *pt; 131 HRDLNK **ppt; 132 u_int indx; 133 134 if (ltab == NULL) 135 return(-1); 136 /* 137 * ignore those nodes that cannot have hard links 138 */ 139 if ((arcn->type == PAX_DIR) || (arcn->sb.st_nlink <= 1)) 140 return(0); 141 142 /* 143 * hash inode number and look for this file 144 */ 145 indx = ((unsigned)arcn->sb.st_ino) % L_TAB_SZ; 146 if ((pt = ltab[indx]) != NULL) { 147 /* 148 * it's hash chain in not empty, walk down looking for it 149 */ 150 ppt = &(ltab[indx]); 151 while (pt != NULL) { 152 if ((pt->ino == arcn->sb.st_ino) && 153 (pt->dev == arcn->sb.st_dev)) 154 break; 155 ppt = &(pt->fow); 156 pt = pt->fow; 157 } 158 159 if (pt != NULL) { 160 /* 161 * found a link. set the node type and copy in the 162 * name of the file it is to link to. we need to 163 * handle hardlinks to regular files differently than 164 * other links. 165 */ 166 arcn->ln_nlen = l_strncpy(arcn->ln_name, pt->name, 167 sizeof(arcn->ln_name) - 1); 168 arcn->ln_name[arcn->ln_nlen] = '\0'; 169 if (arcn->type == PAX_REG) 170 arcn->type = PAX_HRG; 171 else 172 arcn->type = PAX_HLK; 173 174 /* 175 * if we have found all the links to this file, remove 176 * it from the database 177 */ 178 if (--pt->nlink <= 1) { 179 *ppt = pt->fow; 180 free((char *)pt->name); 181 free((char *)pt); 182 } 183 return(1); 184 } 185 } 186 187 /* 188 * we never saw this file before. It has links so we add it to the 189 * front of this hash chain 190 */ 191 if ((pt = (HRDLNK *)malloc(sizeof(HRDLNK))) != NULL) { 192 if ((pt->name = strdup(arcn->name)) != NULL) { 193 pt->dev = arcn->sb.st_dev; 194 pt->ino = arcn->sb.st_ino; 195 pt->nlink = arcn->sb.st_nlink; 196 pt->fow = ltab[indx]; 197 ltab[indx] = pt; 198 return(0); 199 } 200 free((char *)pt); 201 } 202 203 paxwarn(1, "Hard link table out of memory"); 204 return(-1); 205 } 206 207 /* 208 * purg_lnk 209 * remove reference for a file that we may have added to the data base as 210 * a potential source for hard links. We ended up not using the file, so 211 * we do not want to accidently point another file at it later on. 212 */ 213 214 void 215 purg_lnk(ARCHD *arcn) 216 { 217 HRDLNK *pt; 218 HRDLNK **ppt; 219 u_int indx; 220 221 if (ltab == NULL) 222 return; 223 /* 224 * do not bother to look if it could not be in the database 225 */ 226 if ((arcn->sb.st_nlink <= 1) || (arcn->type == PAX_DIR) || 227 (arcn->type == PAX_HLK) || (arcn->type == PAX_HRG)) 228 return; 229 230 /* 231 * find the hash chain for this inode value, if empty return 232 */ 233 indx = ((unsigned)arcn->sb.st_ino) % L_TAB_SZ; 234 if ((pt = ltab[indx]) == NULL) 235 return; 236 237 /* 238 * walk down the list looking for the inode/dev pair, unlink and 239 * free if found 240 */ 241 ppt = &(ltab[indx]); 242 while (pt != NULL) { 243 if ((pt->ino == arcn->sb.st_ino) && 244 (pt->dev == arcn->sb.st_dev)) 245 break; 246 ppt = &(pt->fow); 247 pt = pt->fow; 248 } 249 if (pt == NULL) 250 return; 251 252 /* 253 * remove and free it 254 */ 255 *ppt = pt->fow; 256 free((char *)pt->name); 257 free((char *)pt); 258 } 259 260 /* 261 * lnk_end() 262 * pull apart a existing link table so we can reuse it. We do this between 263 * read and write phases of append with update. (The format may have 264 * used the link table, and we need to start with a fresh table for the 265 * write phase 266 */ 267 268 void 269 lnk_end(void) 270 { 271 int i; 272 HRDLNK *pt; 273 HRDLNK *ppt; 274 275 if (ltab == NULL) 276 return; 277 278 for (i = 0; i < L_TAB_SZ; ++i) { 279 if (ltab[i] == NULL) 280 continue; 281 pt = ltab[i]; 282 ltab[i] = NULL; 283 284 /* 285 * free up each entry on this chain 286 */ 287 while (pt != NULL) { 288 ppt = pt; 289 pt = ppt->fow; 290 free((char *)ppt->name); 291 free((char *)ppt); 292 } 293 } 294 return; 295 } 296 297 /* 298 * modification time table routines 299 * 300 * The modification time table keeps track of last modification times for all 301 * files stored in an archive during a write phase when -u is set. We only 302 * add a file to the archive if it is newer than a file with the same name 303 * already stored on the archive (if there is no other file with the same 304 * name on the archive it is added). This applies to writes and appends. 305 * An append with an -u must read the archive and store the modification time 306 * for every file on that archive before starting the write phase. It is clear 307 * that this is one HUGE database. To save memory space, the actual file names 308 * are stored in a scratch file and indexed by an in memory hash table. The 309 * hash table is indexed by hashing the file path. The nodes in the table store 310 * the length of the filename and the lseek offset within the scratch file 311 * where the actual name is stored. Since there are never any deletions from 312 * this table, fragmentation of the scratch file is never a issue. Lookups 313 * seem to not exhibit any locality at all (files in the database are rarely 314 * looked up more than once...), so caching is just a waste of memory. The 315 * only limitation is the amount of scratch file space available to store the 316 * path names. 317 */ 318 319 /* 320 * ftime_start() 321 * create the file time hash table and open for read/write the scratch 322 * file. (after created it is unlinked, so when we exit we leave 323 * no witnesses). 324 * Return: 325 * 0 if the table and file was created ok, -1 otherwise 326 */ 327 328 int 329 ftime_start(void) 330 { 331 332 if (ftab != NULL) 333 return(0); 334 if ((ftab = (FTM **)calloc(F_TAB_SZ, sizeof(FTM *))) == NULL) { 335 paxwarn(1, "Cannot allocate memory for file time table"); 336 return(-1); 337 } 338 339 /* 340 * get random name and create temporary scratch file, unlink name 341 * so it will get removed on exit 342 */ 343 memcpy(tempbase, _TFILE_BASE, sizeof(_TFILE_BASE)); 344 if ((ffd = mkstemp(tempfile)) < 0) { 345 syswarn(1, errno, "Unable to create temporary file: %s", 346 tempfile); 347 return(-1); 348 } 349 unlink(tempfile); 350 351 return(0); 352 } 353 354 /* 355 * chk_ftime() 356 * looks up entry in file time hash table. If not found, the file is 357 * added to the hash table and the file named stored in the scratch file. 358 * If a file with the same name is found, the file times are compared and 359 * the most recent file time is retained. If the new file was younger (or 360 * was not in the database) the new file is selected for storage. 361 * Return: 362 * 0 if file should be added to the archive, 1 if it should be skipped, 363 * -1 on error 364 */ 365 366 int 367 chk_ftime(ARCHD *arcn) 368 { 369 FTM *pt; 370 int namelen; 371 u_int indx; 372 char ckname[PAXPATHLEN+1]; 373 374 /* 375 * no info, go ahead and add to archive 376 */ 377 if (ftab == NULL) 378 return(0); 379 380 /* 381 * hash the pathname and look up in table 382 */ 383 namelen = arcn->nlen; 384 indx = st_hash(arcn->name, namelen, F_TAB_SZ); 385 if ((pt = ftab[indx]) != NULL) { 386 /* 387 * the hash chain is not empty, walk down looking for match 388 * only read up the path names if the lengths match, speeds 389 * up the search a lot 390 */ 391 while (pt != NULL) { 392 if (pt->namelen == namelen) { 393 /* 394 * potential match, have to read the name 395 * from the scratch file. 396 */ 397 if (lseek(ffd,pt->seek,SEEK_SET) != pt->seek) { 398 syswarn(1, errno, 399 "Failed ftime table seek"); 400 return(-1); 401 } 402 if (read(ffd, ckname, namelen) != namelen) { 403 syswarn(1, errno, 404 "Failed ftime table read"); 405 return(-1); 406 } 407 408 /* 409 * if the names match, we are done 410 */ 411 if (!strncmp(ckname, arcn->name, namelen)) 412 break; 413 } 414 415 /* 416 * try the next entry on the chain 417 */ 418 pt = pt->fow; 419 } 420 421 if (pt != NULL) { 422 /* 423 * found the file, compare the times, save the newer 424 */ 425 if (arcn->sb.st_mtime > pt->mtime) { 426 /* 427 * file is newer 428 */ 429 pt->mtime = arcn->sb.st_mtime; 430 return(0); 431 } 432 /* 433 * file is older 434 */ 435 return(1); 436 } 437 } 438 439 /* 440 * not in table, add it 441 */ 442 if ((pt = (FTM *)malloc(sizeof(FTM))) != NULL) { 443 /* 444 * add the name at the end of the scratch file, saving the 445 * offset. add the file to the head of the hash chain 446 */ 447 if ((pt->seek = lseek(ffd, (off_t)0, SEEK_END)) >= 0) { 448 if (write(ffd, arcn->name, namelen) == namelen) { 449 pt->mtime = arcn->sb.st_mtime; 450 pt->namelen = namelen; 451 pt->fow = ftab[indx]; 452 ftab[indx] = pt; 453 return(0); 454 } 455 syswarn(1, errno, "Failed write to file time table"); 456 } else 457 syswarn(1, errno, "Failed seek on file time table"); 458 } else 459 paxwarn(1, "File time table ran out of memory"); 460 461 if (pt != NULL) 462 free((char *)pt); 463 return(-1); 464 } 465 466 /* 467 * Interactive rename table routines 468 * 469 * The interactive rename table keeps track of the new names that the user 470 * assigns to files from tty input. Since this map is unique for each file 471 * we must store it in case there is a reference to the file later in archive 472 * (a link). Otherwise we will be unable to find the file we know was 473 * extracted. The remapping of these files is stored in a memory based hash 474 * table (it is assumed since input must come from /dev/tty, it is unlikely to 475 * be a very large table). 476 */ 477 478 /* 479 * name_start() 480 * create the interactive rename table 481 * Return: 482 * 0 if successful, -1 otherwise 483 */ 484 485 int 486 name_start(void) 487 { 488 if (ntab != NULL) 489 return(0); 490 if ((ntab = (NAMT **)calloc(N_TAB_SZ, sizeof(NAMT *))) == NULL) { 491 paxwarn(1, "Cannot allocate memory for interactive rename table"); 492 return(-1); 493 } 494 return(0); 495 } 496 497 /* 498 * add_name() 499 * add the new name to old name mapping just created by the user. 500 * If an old name mapping is found (there may be duplicate names on an 501 * archive) only the most recent is kept. 502 * Return: 503 * 0 if added, -1 otherwise 504 */ 505 506 int 507 add_name(char *oname, int onamelen, char *nname) 508 { 509 NAMT *pt; 510 u_int indx; 511 512 if (ntab == NULL) { 513 /* 514 * should never happen 515 */ 516 paxwarn(0, "No interactive rename table, links may fail\n"); 517 return(0); 518 } 519 520 /* 521 * look to see if we have already mapped this file, if so we 522 * will update it 523 */ 524 indx = st_hash(oname, onamelen, N_TAB_SZ); 525 if ((pt = ntab[indx]) != NULL) { 526 /* 527 * look down the has chain for the file 528 */ 529 while ((pt != NULL) && (strcmp(oname, pt->oname) != 0)) 530 pt = pt->fow; 531 532 if (pt != NULL) { 533 /* 534 * found an old mapping, replace it with the new one 535 * the user just input (if it is different) 536 */ 537 if (strcmp(nname, pt->nname) == 0) 538 return(0); 539 540 free((char *)pt->nname); 541 if ((pt->nname = strdup(nname)) == NULL) { 542 paxwarn(1, "Cannot update rename table"); 543 return(-1); 544 } 545 return(0); 546 } 547 } 548 549 /* 550 * this is a new mapping, add it to the table 551 */ 552 if ((pt = (NAMT *)malloc(sizeof(NAMT))) != NULL) { 553 if ((pt->oname = strdup(oname)) != NULL) { 554 if ((pt->nname = strdup(nname)) != NULL) { 555 pt->fow = ntab[indx]; 556 ntab[indx] = pt; 557 return(0); 558 } 559 free((char *)pt->oname); 560 } 561 free((char *)pt); 562 } 563 paxwarn(1, "Interactive rename table out of memory"); 564 return(-1); 565 } 566 567 /* 568 * sub_name() 569 * look up a link name to see if it points at a file that has been 570 * remapped by the user. If found, the link is adjusted to contain the 571 * new name (oname is the link to name) 572 */ 573 574 void 575 sub_name(char *oname, int *onamelen, size_t onamesize) 576 { 577 NAMT *pt; 578 u_int indx; 579 580 if (ntab == NULL) 581 return; 582 /* 583 * look the name up in the hash table 584 */ 585 indx = st_hash(oname, *onamelen, N_TAB_SZ); 586 if ((pt = ntab[indx]) == NULL) 587 return; 588 589 while (pt != NULL) { 590 /* 591 * walk down the hash chain looking for a match 592 */ 593 if (strcmp(oname, pt->oname) == 0) { 594 /* 595 * found it, replace it with the new name 596 * and return (we know that oname has enough space) 597 */ 598 *onamelen = l_strncpy(oname, pt->nname, onamesize - 1); 599 oname[*onamelen] = '\0'; 600 return; 601 } 602 pt = pt->fow; 603 } 604 605 /* 606 * no match, just return 607 */ 608 return; 609 } 610 611 /* 612 * device/inode mapping table routines 613 * (used with formats that store device and inodes fields) 614 * 615 * device/inode mapping tables remap the device field in a archive header. The 616 * device/inode fields are used to determine when files are hard links to each 617 * other. However these values have very little meaning outside of that. This 618 * database is used to solve one of two different problems. 619 * 620 * 1) when files are appended to an archive, while the new files may have hard 621 * links to each other, you cannot determine if they have hard links to any 622 * file already stored on the archive from a prior run of pax. We must assume 623 * that these inode/device pairs are unique only within a SINGLE run of pax 624 * (which adds a set of files to an archive). So we have to make sure the 625 * inode/dev pairs we add each time are always unique. We do this by observing 626 * while the inode field is very dense, the use of the dev field is fairly 627 * sparse. Within each run of pax, we remap any device number of a new archive 628 * member that has a device number used in a prior run and already stored in a 629 * file on the archive. During the read phase of the append, we store the 630 * device numbers used and mark them to not be used by any file during the 631 * write phase. If during write we go to use one of those old device numbers, 632 * we remap it to a new value. 633 * 634 * 2) Often the fields in the archive header used to store these values are 635 * too small to store the entire value. The result is an inode or device value 636 * which can be truncated. This really can foul up an archive. With truncation 637 * we end up creating links between files that are really not links (after 638 * truncation the inodes are the same value). We address that by detecting 639 * truncation and forcing a remap of the device field to split truncated 640 * inodes away from each other. Each truncation creates a pattern of bits that 641 * are removed. We use this pattern of truncated bits to partition the inodes 642 * on a single device to many different devices (each one represented by the 643 * truncated bit pattern). All inodes on the same device that have the same 644 * truncation pattern are mapped to the same new device. Two inodes that 645 * truncate to the same value clearly will always have different truncation 646 * bit patterns, so they will be split from away each other. When we spot 647 * device truncation we remap the device number to a non truncated value. 648 * (for more info see table.h for the data structures involved). 649 */ 650 651 /* 652 * dev_start() 653 * create the device mapping table 654 * Return: 655 * 0 if successful, -1 otherwise 656 */ 657 658 int 659 dev_start(void) 660 { 661 if (dtab != NULL) 662 return(0); 663 if ((dtab = (DEVT **)calloc(D_TAB_SZ, sizeof(DEVT *))) == NULL) { 664 paxwarn(1, "Cannot allocate memory for device mapping table"); 665 return(-1); 666 } 667 return(0); 668 } 669 670 /* 671 * add_dev() 672 * add a device number to the table. this will force the device to be 673 * remapped to a new value if it be used during a write phase. This 674 * function is called during the read phase of an append to prohibit the 675 * use of any device number already in the archive. 676 * Return: 677 * 0 if added ok, -1 otherwise 678 */ 679 680 int 681 add_dev(ARCHD *arcn) 682 { 683 if (chk_dev(arcn->sb.st_dev, 1) == NULL) 684 return(-1); 685 return(0); 686 } 687 688 /* 689 * chk_dev() 690 * check for a device value in the device table. If not found and the add 691 * flag is set, it is added. This does NOT assign any mapping values, just 692 * adds the device number as one that need to be remapped. If this device 693 * is already mapped, just return with a pointer to that entry. 694 * Return: 695 * pointer to the entry for this device in the device map table. Null 696 * if the add flag is not set and the device is not in the table (it is 697 * not been seen yet). If add is set and the device cannot be added, null 698 * is returned (indicates an error). 699 */ 700 701 static DEVT * 702 chk_dev(dev_t dev, int add) 703 { 704 DEVT *pt; 705 u_int indx; 706 707 if (dtab == NULL) 708 return(NULL); 709 /* 710 * look to see if this device is already in the table 711 */ 712 indx = ((unsigned)dev) % D_TAB_SZ; 713 if ((pt = dtab[indx]) != NULL) { 714 while ((pt != NULL) && (pt->dev != dev)) 715 pt = pt->fow; 716 717 /* 718 * found it, return a pointer to it 719 */ 720 if (pt != NULL) 721 return(pt); 722 } 723 724 /* 725 * not in table, we add it only if told to as this may just be a check 726 * to see if a device number is being used. 727 */ 728 if (add == 0) 729 return(NULL); 730 731 /* 732 * allocate a node for this device and add it to the front of the hash 733 * chain. Note we do not assign remaps values here, so the pt->list 734 * list must be NULL. 735 */ 736 if ((pt = (DEVT *)malloc(sizeof(DEVT))) == NULL) { 737 paxwarn(1, "Device map table out of memory"); 738 return(NULL); 739 } 740 pt->dev = dev; 741 pt->list = NULL; 742 pt->fow = dtab[indx]; 743 dtab[indx] = pt; 744 return(pt); 745 } 746 /* 747 * map_dev() 748 * given an inode and device storage mask (the mask has a 1 for each bit 749 * the archive format is able to store in a header), we check for inode 750 * and device truncation and remap the device as required. Device mapping 751 * can also occur when during the read phase of append a device number was 752 * seen (and was marked as do not use during the write phase). WE ASSUME 753 * that unsigned longs are the same size or bigger than the fields used 754 * for ino_t and dev_t. If not the types will have to be changed. 755 * Return: 756 * 0 if all ok, -1 otherwise. 757 */ 758 759 int 760 map_dev(ARCHD *arcn, u_long dev_mask, u_long ino_mask) 761 { 762 DEVT *pt; 763 DLIST *dpt; 764 static dev_t lastdev = 0; /* next device number to try */ 765 int trc_ino = 0; 766 int trc_dev = 0; 767 ino_t trunc_bits = 0; 768 ino_t nino; 769 770 if (dtab == NULL) 771 return(0); 772 /* 773 * check for device and inode truncation, and extract the truncated 774 * bit pattern. 775 */ 776 if ((arcn->sb.st_dev & (dev_t)dev_mask) != arcn->sb.st_dev) 777 ++trc_dev; 778 if ((nino = arcn->sb.st_ino & (ino_t)ino_mask) != arcn->sb.st_ino) { 779 ++trc_ino; 780 trunc_bits = arcn->sb.st_ino & (ino_t)(~ino_mask); 781 } 782 783 /* 784 * see if this device is already being mapped, look up the device 785 * then find the truncation bit pattern which applies 786 */ 787 if ((pt = chk_dev(arcn->sb.st_dev, 0)) != NULL) { 788 /* 789 * this device is already marked to be remapped 790 */ 791 for (dpt = pt->list; dpt != NULL; dpt = dpt->fow) 792 if (dpt->trunc_bits == trunc_bits) 793 break; 794 795 if (dpt != NULL) { 796 /* 797 * we are being remapped for this device and pattern 798 * change the device number to be stored and return 799 */ 800 arcn->sb.st_dev = dpt->dev; 801 arcn->sb.st_ino = nino; 802 return(0); 803 } 804 } else { 805 /* 806 * this device is not being remapped YET. if we do not have any 807 * form of truncation, we do not need a remap 808 */ 809 if (!trc_ino && !trc_dev) 810 return(0); 811 812 /* 813 * we have truncation, have to add this as a device to remap 814 */ 815 if ((pt = chk_dev(arcn->sb.st_dev, 1)) == NULL) 816 goto bad; 817 818 /* 819 * if we just have a truncated inode, we have to make sure that 820 * all future inodes that do not truncate (they have the 821 * truncation pattern of all 0's) continue to map to the same 822 * device number. We probably have already written inodes with 823 * this device number to the archive with the truncation 824 * pattern of all 0's. So we add the mapping for all 0's to the 825 * same device number. 826 */ 827 if (!trc_dev && (trunc_bits != 0)) { 828 if ((dpt = (DLIST *)malloc(sizeof(DLIST))) == NULL) 829 goto bad; 830 dpt->trunc_bits = 0; 831 dpt->dev = arcn->sb.st_dev; 832 dpt->fow = pt->list; 833 pt->list = dpt; 834 } 835 } 836 837 /* 838 * look for a device number not being used. We must watch for wrap 839 * around on lastdev (so we do not get stuck looking forever!) 840 */ 841 while (++lastdev > 0) { 842 if (chk_dev(lastdev, 0) != NULL) 843 continue; 844 /* 845 * found an unused value. If we have reached truncation point 846 * for this format we are hosed, so we give up. Otherwise we 847 * mark it as being used. 848 */ 849 if (((lastdev & ((dev_t)dev_mask)) != lastdev) || 850 (chk_dev(lastdev, 1) == NULL)) 851 goto bad; 852 break; 853 } 854 855 if ((lastdev <= 0) || ((dpt = (DLIST *)malloc(sizeof(DLIST))) == NULL)) 856 goto bad; 857 858 /* 859 * got a new device number, store it under this truncation pattern. 860 * change the device number this file is being stored with. 861 */ 862 dpt->trunc_bits = trunc_bits; 863 dpt->dev = lastdev; 864 dpt->fow = pt->list; 865 pt->list = dpt; 866 arcn->sb.st_dev = lastdev; 867 arcn->sb.st_ino = nino; 868 return(0); 869 870 bad: 871 paxwarn(1, "Unable to fix truncated inode/device field when storing %s", 872 arcn->name); 873 paxwarn(0, "Archive may create improper hard links when extracted"); 874 return(0); 875 } 876 877 /* 878 * directory access/mod time reset table routines (for directories READ by pax) 879 * 880 * The pax -t flag requires that access times of archive files be the same 881 * before being read by pax. For regular files, access time is restored after 882 * the file has been copied. This database provides the same functionality for 883 * directories read during file tree traversal. Restoring directory access time 884 * is more complex than files since directories may be read several times until 885 * all the descendants in their subtree are visited by fts. Directory access 886 * and modification times are stored during the fts pre-order visit (done 887 * before any descendants in the subtree are visited) and restored after the 888 * fts post-order visit (after all the descendants have been visited). In the 889 * case of premature exit from a subtree (like from the effects of -n), any 890 * directory entries left in this database are reset during final cleanup 891 * operations of pax. Entries are hashed by inode number for fast lookup. 892 */ 893 894 /* 895 * atdir_start() 896 * create the directory access time database for directories READ by pax. 897 * Return: 898 * 0 is created ok, -1 otherwise. 899 */ 900 901 int 902 atdir_start(void) 903 { 904 if (atab != NULL) 905 return(0); 906 if ((atab = (ATDIR **)calloc(A_TAB_SZ, sizeof(ATDIR *))) == NULL) { 907 paxwarn(1,"Cannot allocate space for directory access time table"); 908 return(-1); 909 } 910 return(0); 911 } 912 913 914 /* 915 * atdir_end() 916 * walk through the directory access time table and reset the access time 917 * of any directory who still has an entry left in the database. These 918 * entries are for directories READ by pax 919 */ 920 921 void 922 atdir_end(void) 923 { 924 ATDIR *pt; 925 int i; 926 927 if (atab == NULL) 928 return; 929 /* 930 * for each non-empty hash table entry reset all the directories 931 * chained there. 932 */ 933 for (i = 0; i < A_TAB_SZ; ++i) { 934 if ((pt = atab[i]) == NULL) 935 continue; 936 /* 937 * remember to force the times, set_ftime() looks at pmtime 938 * and patime, which only applies to things CREATED by pax, 939 * not read by pax. Read time reset is controlled by -t. 940 */ 941 for (; pt != NULL; pt = pt->fow) 942 set_ftime(pt->name, pt->mtime, pt->atime, 1); 943 } 944 } 945 946 /* 947 * add_atdir() 948 * add a directory to the directory access time table. Table is hashed 949 * and chained by inode number. This is for directories READ by pax 950 */ 951 952 void 953 add_atdir(char *fname, dev_t dev, ino_t ino, time_t mtime, time_t atime) 954 { 955 ATDIR *pt; 956 u_int indx; 957 958 if (atab == NULL) 959 return; 960 961 /* 962 * make sure this directory is not already in the table, if so just 963 * return (the older entry always has the correct time). The only 964 * way this will happen is when the same subtree can be traversed by 965 * different args to pax and the -n option is aborting fts out of a 966 * subtree before all the post-order visits have been made. 967 */ 968 indx = ((unsigned)ino) % A_TAB_SZ; 969 if ((pt = atab[indx]) != NULL) { 970 while (pt != NULL) { 971 if ((pt->ino == ino) && (pt->dev == dev)) 972 break; 973 pt = pt->fow; 974 } 975 976 /* 977 * oops, already there. Leave it alone. 978 */ 979 if (pt != NULL) 980 return; 981 } 982 983 /* 984 * add it to the front of the hash chain 985 */ 986 if ((pt = (ATDIR *)malloc(sizeof(ATDIR))) != NULL) { 987 if ((pt->name = strdup(fname)) != NULL) { 988 pt->dev = dev; 989 pt->ino = ino; 990 pt->mtime = mtime; 991 pt->atime = atime; 992 pt->fow = atab[indx]; 993 atab[indx] = pt; 994 return; 995 } 996 free((char *)pt); 997 } 998 999 paxwarn(1, "Directory access time reset table ran out of memory"); 1000 return; 1001 } 1002 1003 /* 1004 * get_atdir() 1005 * look up a directory by inode and device number to obtain the access 1006 * and modification time you want to set to. If found, the modification 1007 * and access time parameters are set and the entry is removed from the 1008 * table (as it is no longer needed). These are for directories READ by 1009 * pax 1010 * Return: 1011 * 0 if found, -1 if not found. 1012 */ 1013 1014 int 1015 get_atdir(dev_t dev, ino_t ino, time_t *mtime, time_t *atime) 1016 { 1017 ATDIR *pt; 1018 ATDIR **ppt; 1019 u_int indx; 1020 1021 if (atab == NULL) 1022 return(-1); 1023 /* 1024 * hash by inode and search the chain for an inode and device match 1025 */ 1026 indx = ((unsigned)ino) % A_TAB_SZ; 1027 if ((pt = atab[indx]) == NULL) 1028 return(-1); 1029 1030 ppt = &(atab[indx]); 1031 while (pt != NULL) { 1032 if ((pt->ino == ino) && (pt->dev == dev)) 1033 break; 1034 /* 1035 * no match, go to next one 1036 */ 1037 ppt = &(pt->fow); 1038 pt = pt->fow; 1039 } 1040 1041 /* 1042 * return if we did not find it. 1043 */ 1044 if (pt == NULL) 1045 return(-1); 1046 1047 /* 1048 * found it. return the times and remove the entry from the table. 1049 */ 1050 *ppt = pt->fow; 1051 *mtime = pt->mtime; 1052 *atime = pt->atime; 1053 free((char *)pt->name); 1054 free((char *)pt); 1055 return(0); 1056 } 1057 1058 /* 1059 * directory access mode and time storage routines (for directories CREATED 1060 * by pax). 1061 * 1062 * Pax requires that extracted directories, by default, have their access/mod 1063 * times and permissions set to the values specified in the archive. During the 1064 * actions of extracting (and creating the destination subtree during -rw copy) 1065 * directories extracted may be modified after being created. Even worse is 1066 * that these directories may have been created with file permissions which 1067 * prohibits any descendants of these directories from being extracted. When 1068 * directories are created by pax, access rights may be added to permit the 1069 * creation of files in their subtree. Every time pax creates a directory, the 1070 * times and file permissions specified by the archive are stored. After all 1071 * files have been extracted (or copied), these directories have their times 1072 * and file modes reset to the stored values. The directory info is restored in 1073 * reverse order as entries were added to the data file from root to leaf. To 1074 * restore atime properly, we must go backwards. The data file consists of 1075 * records with two parts, the file name followed by a DIRDATA trailer. The 1076 * fixed sized trailer contains the size of the name plus the off_t location in 1077 * the file. To restore we work backwards through the file reading the trailer 1078 * then the file name. 1079 */ 1080 1081 /* 1082 * dir_start() 1083 * set up the directory time and file mode storage for directories CREATED 1084 * by pax. 1085 * Return: 1086 * 0 if ok, -1 otherwise 1087 */ 1088 1089 int 1090 dir_start(void) 1091 { 1092 1093 if (dirfd != -1) 1094 return(0); 1095 1096 /* 1097 * unlink the file so it goes away at termination by itself 1098 */ 1099 memcpy(tempbase, _TFILE_BASE, sizeof(_TFILE_BASE)); 1100 if ((dirfd = mkstemp(tempfile)) >= 0) { 1101 unlink(tempfile); 1102 return(0); 1103 } 1104 paxwarn(1, "Unable to create temporary file for directory times: %s", 1105 tempfile); 1106 return(-1); 1107 } 1108 1109 /* 1110 * add_dir() 1111 * add the mode and times for a newly CREATED directory 1112 * name is name of the directory, psb the stat buffer with the data in it, 1113 * frc_mode is a flag that says whether to force the setting of the mode 1114 * (ignoring the user set values for preserving file mode). Frc_mode is 1115 * for the case where we created a file and found that the resulting 1116 * directory was not writeable and the user asked for file modes to NOT 1117 * be preserved. (we have to preserve what was created by default, so we 1118 * have to force the setting at the end. this is stated explicitly in the 1119 * pax spec) 1120 */ 1121 1122 void 1123 add_dir(char *name, int nlen, struct stat *psb, int frc_mode) 1124 { 1125 DIRDATA dblk; 1126 1127 if (dirfd < 0) 1128 return; 1129 1130 /* 1131 * get current position (where file name will start) so we can store it 1132 * in the trailer 1133 */ 1134 if ((dblk.npos = lseek(dirfd, 0L, SEEK_CUR)) < 0) { 1135 paxwarn(1,"Unable to store mode and times for directory: %s",name); 1136 return; 1137 } 1138 1139 /* 1140 * write the file name followed by the trailer 1141 */ 1142 dblk.nlen = nlen + 1; 1143 dblk.mode = psb->st_mode & 0xffff; 1144 dblk.mtime = psb->st_mtime; 1145 dblk.atime = psb->st_atime; 1146 dblk.frc_mode = frc_mode; 1147 if ((write(dirfd, name, dblk.nlen) == dblk.nlen) && 1148 (write(dirfd, (char *)&dblk, sizeof(dblk)) == sizeof(dblk))) { 1149 ++dircnt; 1150 return; 1151 } 1152 1153 paxwarn(1,"Unable to store mode and times for created directory: %s",name); 1154 return; 1155 } 1156 1157 /* 1158 * proc_dir() 1159 * process all file modes and times stored for directories CREATED 1160 * by pax 1161 */ 1162 1163 void 1164 proc_dir(void) 1165 { 1166 char name[PAXPATHLEN+1]; 1167 DIRDATA dblk; 1168 u_long cnt; 1169 1170 if (dirfd < 0) 1171 return; 1172 /* 1173 * read backwards through the file and process each directory 1174 */ 1175 for (cnt = 0; cnt < dircnt; ++cnt) { 1176 /* 1177 * read the trailer, then the file name, if this fails 1178 * just give up. 1179 */ 1180 if (lseek(dirfd, -((off_t)sizeof(dblk)), SEEK_CUR) < 0) 1181 break; 1182 if (read(dirfd,(char *)&dblk, sizeof(dblk)) != sizeof(dblk)) 1183 break; 1184 if (lseek(dirfd, dblk.npos, SEEK_SET) < 0) 1185 break; 1186 if (read(dirfd, name, dblk.nlen) != dblk.nlen) 1187 break; 1188 if (lseek(dirfd, dblk.npos, SEEK_SET) < 0) 1189 break; 1190 1191 /* 1192 * frc_mode set, make sure we set the file modes even if 1193 * the user didn't ask for it (see file_subs.c for more info) 1194 */ 1195 if (pmode || dblk.frc_mode) 1196 set_pmode(name, dblk.mode); 1197 if (patime || pmtime) 1198 set_ftime(name, dblk.mtime, dblk.atime, 0); 1199 } 1200 1201 close(dirfd); 1202 dirfd = -1; 1203 if (cnt != dircnt) 1204 paxwarn(1,"Unable to set mode and times for created directories"); 1205 return; 1206 } 1207 1208 /* 1209 * database independent routines 1210 */ 1211 1212 /* 1213 * st_hash() 1214 * hashes filenames to a u_int for hashing into a table. Looks at the tail 1215 * end of file, as this provides far better distribution than any other 1216 * part of the name. For performance reasons we only care about the last 1217 * MAXKEYLEN chars (should be at LEAST large enough to pick off the file 1218 * name). Was tested on 500,000 name file tree traversal from the root 1219 * and gave almost a perfectly uniform distribution of keys when used with 1220 * prime sized tables (MAXKEYLEN was 128 in test). Hashes (sizeof int) 1221 * chars at a time and pads with 0 for last addition. 1222 * Return: 1223 * the hash value of the string MOD (%) the table size. 1224 */ 1225 1226 u_int 1227 st_hash(char *name, int len, int tabsz) 1228 { 1229 char *pt; 1230 char *dest; 1231 char *end; 1232 int i; 1233 u_int key = 0; 1234 int steps; 1235 int res; 1236 u_int val; 1237 1238 /* 1239 * only look at the tail up to MAXKEYLEN, we do not need to waste 1240 * time here (remember these are pathnames, the tail is what will 1241 * spread out the keys) 1242 */ 1243 if (len > MAXKEYLEN) { 1244 pt = &(name[len - MAXKEYLEN]); 1245 len = MAXKEYLEN; 1246 } else 1247 pt = name; 1248 1249 /* 1250 * calculate the number of u_int size steps in the string and if 1251 * there is a runt to deal with 1252 */ 1253 steps = len/sizeof(u_int); 1254 res = len % sizeof(u_int); 1255 1256 /* 1257 * add up the value of the string in unsigned integer sized pieces 1258 * too bad we cannot have unsigned int aligned strings, then we 1259 * could avoid the expensive copy. 1260 */ 1261 for (i = 0; i < steps; ++i) { 1262 end = pt + sizeof(u_int); 1263 dest = (char *)&val; 1264 while (pt < end) 1265 *dest++ = *pt++; 1266 key += val; 1267 } 1268 1269 /* 1270 * add in the runt padded with zero to the right 1271 */ 1272 if (res) { 1273 val = 0; 1274 end = pt + res; 1275 dest = (char *)&val; 1276 while (pt < end) 1277 *dest++ = *pt++; 1278 key += val; 1279 } 1280 1281 /* 1282 * return the result mod the table size 1283 */ 1284 return(key % tabsz); 1285 } 1286