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 * $DragonFly: src/bin/pax/tables.c,v 1.7 2006/09/27 21:58:08 pavalos Exp $ 40 */ 41 42 #include <sys/types.h> 43 #include <sys/time.h> 44 #include <sys/stat.h> 45 #include <sys/fcntl.h> 46 #include <errno.h> 47 #include <stdio.h> 48 #include <stdlib.h> 49 #include <string.h> 50 #include <unistd.h> 51 #include "pax.h" 52 #include "tables.h" 53 #include "extern.h" 54 55 /* 56 * Routines for controlling the contents of all the different databases pax 57 * keeps. Tables are dynamically created only when they are needed. The 58 * goal was speed and the ability to work with HUGE archives. The databases 59 * were kept simple, but do have complex rules for when the contents change. 60 * As of this writing, the POSIX library functions were more complex than 61 * needed for this application (pax databases have very short lifetimes and 62 * do not survive after pax is finished). Pax is required to handle very 63 * large archives. These database routines carefully combine memory usage and 64 * temporary file storage in ways which will not significantly impact runtime 65 * performance while allowing the largest possible archives to be handled. 66 * Trying to force the fit to the POSIX database routines was not considered 67 * time well spent. 68 */ 69 70 static HRDLNK **ltab = NULL; /* hard link table for detecting hard links */ 71 static FTM **ftab = NULL; /* file time table for updating arch */ 72 static NAMT **ntab = NULL; /* interactive rename storage table */ 73 static DEVT **dtab = NULL; /* device/inode mapping tables */ 74 static ATDIR **atab = NULL; /* file tree directory time reset table */ 75 static int dirfd = -1; /* storage for setting created dir time/mode */ 76 static u_long dircnt; /* entries in dir time/mode storage */ 77 static int ffd = -1; /* tmp file for file time table name storage */ 78 79 static DEVT *chk_dev (dev_t, int); 80 81 /* 82 * hard link table routines 83 * 84 * The hard link table tries to detect hard links to files using the device and 85 * inode values. We do this when writing an archive, so we can tell the format 86 * write routine that this file is a hard link to another file. The format 87 * write routine then can store this file in whatever way it wants (as a hard 88 * link if the format supports that like tar, or ignore this info like cpio). 89 * (Actually a field in the format driver table tells us if the format wants 90 * hard link info. if not, we do not waste time looking for them). We also use 91 * the same table when reading an archive. In that situation, this table is 92 * used by the format read routine to detect hard links from stored dev and 93 * inode numbers (like cpio). This will allow pax to create a link when one 94 * can be detected by the archive format. 95 */ 96 97 /* 98 * lnk_start 99 * Creates the hard link table. 100 * Return: 101 * 0 if created, -1 if failure 102 */ 103 104 int 105 lnk_start(void) 106 { 107 if (ltab != NULL) 108 return(0); 109 if ((ltab = (HRDLNK **)calloc(L_TAB_SZ, sizeof(HRDLNK *))) == NULL) { 110 paxwarn(1, "Cannot allocate memory for hard link table"); 111 return(-1); 112 } 113 return(0); 114 } 115 116 /* 117 * chk_lnk() 118 * Looks up entry in hard link hash table. If found, it copies the name 119 * of the file it is linked to (we already saw that file) into ln_name. 120 * lnkcnt is decremented and if goes to 1 the node is deleted from the 121 * database. (We have seen all the links to this file). If not found, 122 * we add the file to the database if it has the potential for having 123 * hard links to other files we may process (it has a link count > 1) 124 * Return: 125 * if found returns 1; if not found returns 0; -1 on error 126 */ 127 128 int 129 chk_lnk(ARCHD *arcn) 130 { 131 HRDLNK *pt; 132 HRDLNK **ppt; 133 u_int indx; 134 135 if (ltab == NULL) 136 return(-1); 137 /* 138 * ignore those nodes that cannot have hard links 139 */ 140 if ((arcn->type == PAX_DIR) || (arcn->sb.st_nlink <= 1)) 141 return(0); 142 143 /* 144 * hash inode number and look for this file 145 */ 146 indx = ((unsigned)arcn->sb.st_ino) % L_TAB_SZ; 147 if ((pt = ltab[indx]) != NULL) { 148 /* 149 * it's hash chain in not empty, walk down looking for it 150 */ 151 ppt = &(ltab[indx]); 152 while (pt != NULL) { 153 if ((pt->ino == arcn->sb.st_ino) && 154 (pt->dev == arcn->sb.st_dev)) 155 break; 156 ppt = &(pt->fow); 157 pt = pt->fow; 158 } 159 160 if (pt != NULL) { 161 /* 162 * found a link. set the node type and copy in the 163 * name of the file it is to link to. we need to 164 * handle hardlinks to regular files differently than 165 * other links. 166 */ 167 arcn->ln_nlen = l_strncpy(arcn->ln_name, pt->name, 168 sizeof(arcn->ln_name) - 1); 169 arcn->ln_name[arcn->ln_nlen] = '\0'; 170 if (arcn->type == PAX_REG) 171 arcn->type = PAX_HRG; 172 else 173 arcn->type = PAX_HLK; 174 175 /* 176 * if we have found all the links to this file, remove 177 * it from the database 178 */ 179 if (--pt->nlink <= 1) { 180 *ppt = pt->fow; 181 free((char *)pt->name); 182 free((char *)pt); 183 } 184 return(1); 185 } 186 } 187 188 /* 189 * we never saw this file before. It has links so we add it to the 190 * front of this hash chain 191 */ 192 if ((pt = (HRDLNK *)malloc(sizeof(HRDLNK))) != NULL) { 193 if ((pt->name = strdup(arcn->name)) != NULL) { 194 pt->dev = arcn->sb.st_dev; 195 pt->ino = arcn->sb.st_ino; 196 pt->nlink = arcn->sb.st_nlink; 197 pt->fow = ltab[indx]; 198 ltab[indx] = pt; 199 return(0); 200 } 201 free((char *)pt); 202 } 203 204 paxwarn(1, "Hard link table out of memory"); 205 return(-1); 206 } 207 208 /* 209 * purg_lnk 210 * remove reference for a file that we may have added to the data base as 211 * a potential source for hard links. We ended up not using the file, so 212 * we do not want to accidently point another file at it later on. 213 */ 214 215 void 216 purg_lnk(ARCHD *arcn) 217 { 218 HRDLNK *pt; 219 HRDLNK **ppt; 220 u_int indx; 221 222 if (ltab == NULL) 223 return; 224 /* 225 * do not bother to look if it could not be in the database 226 */ 227 if ((arcn->sb.st_nlink <= 1) || (arcn->type == PAX_DIR) || 228 (arcn->type == PAX_HLK) || (arcn->type == PAX_HRG)) 229 return; 230 231 /* 232 * find the hash chain for this inode value, if empty return 233 */ 234 indx = ((unsigned)arcn->sb.st_ino) % L_TAB_SZ; 235 if ((pt = ltab[indx]) == NULL) 236 return; 237 238 /* 239 * walk down the list looking for the inode/dev pair, unlink and 240 * free if found 241 */ 242 ppt = &(ltab[indx]); 243 while (pt != NULL) { 244 if ((pt->ino == arcn->sb.st_ino) && 245 (pt->dev == arcn->sb.st_dev)) 246 break; 247 ppt = &(pt->fow); 248 pt = pt->fow; 249 } 250 if (pt == NULL) 251 return; 252 253 /* 254 * remove and free it 255 */ 256 *ppt = pt->fow; 257 free((char *)pt->name); 258 free((char *)pt); 259 } 260 261 /* 262 * lnk_end() 263 * pull apart a existing link table so we can reuse it. We do this between 264 * read and write phases of append with update. (The format may have 265 * used the link table, and we need to start with a fresh table for the 266 * write phase 267 */ 268 269 void 270 lnk_end(void) 271 { 272 int i; 273 HRDLNK *pt; 274 HRDLNK *ppt; 275 276 if (ltab == NULL) 277 return; 278 279 for (i = 0; i < L_TAB_SZ; ++i) { 280 if (ltab[i] == NULL) 281 continue; 282 pt = ltab[i]; 283 ltab[i] = NULL; 284 285 /* 286 * free up each entry on this chain 287 */ 288 while (pt != NULL) { 289 ppt = pt; 290 pt = ppt->fow; 291 free((char *)ppt->name); 292 free((char *)ppt); 293 } 294 } 295 return; 296 } 297 298 /* 299 * modification time table routines 300 * 301 * The modification time table keeps track of last modification times for all 302 * files stored in an archive during a write phase when -u is set. We only 303 * add a file to the archive if it is newer than a file with the same name 304 * already stored on the archive (if there is no other file with the same 305 * name on the archive it is added). This applies to writes and appends. 306 * An append with an -u must read the archive and store the modification time 307 * for every file on that archive before starting the write phase. It is clear 308 * that this is one HUGE database. To save memory space, the actual file names 309 * are stored in a scratch file and indexed by an in memory hash table. The 310 * hash table is indexed by hashing the file path. The nodes in the table store 311 * the length of the filename and the lseek offset within the scratch file 312 * where the actual name is stored. Since there are never any deletions from 313 * this table, fragmentation of the scratch file is never a issue. Lookups 314 * seem to not exhibit any locality at all (files in the database are rarely 315 * looked up more than once...), so caching is just a waste of memory. The 316 * only limitation is the amount of scratch file space available to store the 317 * path names. 318 */ 319 320 /* 321 * ftime_start() 322 * create the file time hash table and open for read/write the scratch 323 * file. (after created it is unlinked, so when we exit we leave 324 * no witnesses). 325 * Return: 326 * 0 if the table and file was created ok, -1 otherwise 327 */ 328 329 int 330 ftime_start(void) 331 { 332 333 if (ftab != NULL) 334 return(0); 335 if ((ftab = (FTM **)calloc(F_TAB_SZ, sizeof(FTM *))) == NULL) { 336 paxwarn(1, "Cannot allocate memory for file time table"); 337 return(-1); 338 } 339 340 /* 341 * get random name and create temporary scratch file, unlink name 342 * so it will get removed on exit 343 */ 344 memcpy(tempbase, _TFILE_BASE, sizeof(_TFILE_BASE)); 345 if ((ffd = mkstemp(tempfile)) < 0) { 346 syswarn(1, errno, "Unable to create temporary file: %s", 347 tempfile); 348 return(-1); 349 } 350 unlink(tempfile); 351 352 return(0); 353 } 354 355 /* 356 * chk_ftime() 357 * looks up entry in file time hash table. If not found, the file is 358 * added to the hash table and the file named stored in the scratch file. 359 * If a file with the same name is found, the file times are compared and 360 * the most recent file time is retained. If the new file was younger (or 361 * was not in the database) the new file is selected for storage. 362 * Return: 363 * 0 if file should be added to the archive, 1 if it should be skipped, 364 * -1 on error 365 */ 366 367 int 368 chk_ftime(ARCHD *arcn) 369 { 370 FTM *pt; 371 int namelen; 372 u_int indx; 373 char ckname[PAXPATHLEN+1]; 374 375 /* 376 * no info, go ahead and add to archive 377 */ 378 if (ftab == NULL) 379 return(0); 380 381 /* 382 * hash the pathname and look up in table 383 */ 384 namelen = arcn->nlen; 385 indx = st_hash(arcn->name, namelen, F_TAB_SZ); 386 if ((pt = ftab[indx]) != NULL) { 387 /* 388 * the hash chain is not empty, walk down looking for match 389 * only read up the path names if the lengths match, speeds 390 * up the search a lot 391 */ 392 while (pt != NULL) { 393 if (pt->namelen == namelen) { 394 /* 395 * potential match, have to read the name 396 * from the scratch file. 397 */ 398 if (lseek(ffd,pt->seek,SEEK_SET) != pt->seek) { 399 syswarn(1, errno, 400 "Failed ftime table seek"); 401 return(-1); 402 } 403 if (read(ffd, ckname, namelen) != namelen) { 404 syswarn(1, errno, 405 "Failed ftime table read"); 406 return(-1); 407 } 408 409 /* 410 * if the names match, we are done 411 */ 412 if (!strncmp(ckname, arcn->name, namelen)) 413 break; 414 } 415 416 /* 417 * try the next entry on the chain 418 */ 419 pt = pt->fow; 420 } 421 422 if (pt != NULL) { 423 /* 424 * found the file, compare the times, save the newer 425 */ 426 if (arcn->sb.st_mtime > pt->mtime) { 427 /* 428 * file is newer 429 */ 430 pt->mtime = arcn->sb.st_mtime; 431 return(0); 432 } 433 /* 434 * file is older 435 */ 436 return(1); 437 } 438 } 439 440 /* 441 * not in table, add it 442 */ 443 if ((pt = (FTM *)malloc(sizeof(FTM))) != NULL) { 444 /* 445 * add the name at the end of the scratch file, saving the 446 * offset. add the file to the head of the hash chain 447 */ 448 if ((pt->seek = lseek(ffd, (off_t)0, SEEK_END)) >= 0) { 449 if (write(ffd, arcn->name, namelen) == namelen) { 450 pt->mtime = arcn->sb.st_mtime; 451 pt->namelen = namelen; 452 pt->fow = ftab[indx]; 453 ftab[indx] = pt; 454 return(0); 455 } 456 syswarn(1, errno, "Failed write to file time table"); 457 } else 458 syswarn(1, errno, "Failed seek on file time table"); 459 } else 460 paxwarn(1, "File time table ran out of memory"); 461 462 if (pt != NULL) 463 free((char *)pt); 464 return(-1); 465 } 466 467 /* 468 * Interactive rename table routines 469 * 470 * The interactive rename table keeps track of the new names that the user 471 * assigns to files from tty input. Since this map is unique for each file 472 * we must store it in case there is a reference to the file later in archive 473 * (a link). Otherwise we will be unable to find the file we know was 474 * extracted. The remapping of these files is stored in a memory based hash 475 * table (it is assumed since input must come from /dev/tty, it is unlikely to 476 * be a very large table). 477 */ 478 479 /* 480 * name_start() 481 * create the interactive rename table 482 * Return: 483 * 0 if successful, -1 otherwise 484 */ 485 486 int 487 name_start(void) 488 { 489 if (ntab != NULL) 490 return(0); 491 if ((ntab = (NAMT **)calloc(N_TAB_SZ, sizeof(NAMT *))) == NULL) { 492 paxwarn(1, "Cannot allocate memory for interactive rename table"); 493 return(-1); 494 } 495 return(0); 496 } 497 498 /* 499 * add_name() 500 * add the new name to old name mapping just created by the user. 501 * If an old name mapping is found (there may be duplicate names on an 502 * archive) only the most recent is kept. 503 * Return: 504 * 0 if added, -1 otherwise 505 */ 506 507 int 508 add_name(char *oname, int onamelen, char *nname) 509 { 510 NAMT *pt; 511 u_int indx; 512 513 if (ntab == NULL) { 514 /* 515 * should never happen 516 */ 517 paxwarn(0, "No interactive rename table, links may fail\n"); 518 return(0); 519 } 520 521 /* 522 * look to see if we have already mapped this file, if so we 523 * will update it 524 */ 525 indx = st_hash(oname, onamelen, N_TAB_SZ); 526 if ((pt = ntab[indx]) != NULL) { 527 /* 528 * look down the has chain for the file 529 */ 530 while ((pt != NULL) && (strcmp(oname, pt->oname) != 0)) 531 pt = pt->fow; 532 533 if (pt != NULL) { 534 /* 535 * found an old mapping, replace it with the new one 536 * the user just input (if it is different) 537 */ 538 if (strcmp(nname, pt->nname) == 0) 539 return(0); 540 541 free((char *)pt->nname); 542 if ((pt->nname = strdup(nname)) == NULL) { 543 paxwarn(1, "Cannot update rename table"); 544 return(-1); 545 } 546 return(0); 547 } 548 } 549 550 /* 551 * this is a new mapping, add it to the table 552 */ 553 if ((pt = (NAMT *)malloc(sizeof(NAMT))) != NULL) { 554 if ((pt->oname = strdup(oname)) != NULL) { 555 if ((pt->nname = strdup(nname)) != NULL) { 556 pt->fow = ntab[indx]; 557 ntab[indx] = pt; 558 return(0); 559 } 560 free((char *)pt->oname); 561 } 562 free((char *)pt); 563 } 564 paxwarn(1, "Interactive rename table out of memory"); 565 return(-1); 566 } 567 568 /* 569 * sub_name() 570 * look up a link name to see if it points at a file that has been 571 * remapped by the user. If found, the link is adjusted to contain the 572 * new name (oname is the link to name) 573 */ 574 575 void 576 sub_name(char *oname, int *onamelen, size_t onamesize) 577 { 578 NAMT *pt; 579 u_int indx; 580 581 if (ntab == NULL) 582 return; 583 /* 584 * look the name up in the hash table 585 */ 586 indx = st_hash(oname, *onamelen, N_TAB_SZ); 587 if ((pt = ntab[indx]) == NULL) 588 return; 589 590 while (pt != NULL) { 591 /* 592 * walk down the hash chain looking for a match 593 */ 594 if (strcmp(oname, pt->oname) == 0) { 595 /* 596 * found it, replace it with the new name 597 * and return (we know that oname has enough space) 598 */ 599 *onamelen = l_strncpy(oname, pt->nname, onamesize - 1); 600 oname[*onamelen] = '\0'; 601 return; 602 } 603 pt = pt->fow; 604 } 605 606 /* 607 * no match, just return 608 */ 609 return; 610 } 611 612 /* 613 * device/inode mapping table routines 614 * (used with formats that store device and inodes fields) 615 * 616 * device/inode mapping tables remap the device field in a archive header. The 617 * device/inode fields are used to determine when files are hard links to each 618 * other. However these values have very little meaning outside of that. This 619 * database is used to solve one of two different problems. 620 * 621 * 1) when files are appended to an archive, while the new files may have hard 622 * links to each other, you cannot determine if they have hard links to any 623 * file already stored on the archive from a prior run of pax. We must assume 624 * that these inode/device pairs are unique only within a SINGLE run of pax 625 * (which adds a set of files to an archive). So we have to make sure the 626 * inode/dev pairs we add each time are always unique. We do this by observing 627 * while the inode field is very dense, the use of the dev field is fairly 628 * sparse. Within each run of pax, we remap any device number of a new archive 629 * member that has a device number used in a prior run and already stored in a 630 * file on the archive. During the read phase of the append, we store the 631 * device numbers used and mark them to not be used by any file during the 632 * write phase. If during write we go to use one of those old device numbers, 633 * we remap it to a new value. 634 * 635 * 2) Often the fields in the archive header used to store these values are 636 * too small to store the entire value. The result is an inode or device value 637 * which can be truncated. This really can foul up an archive. With truncation 638 * we end up creating links between files that are really not links (after 639 * truncation the inodes are the same value). We address that by detecting 640 * truncation and forcing a remap of the device field to split truncated 641 * inodes away from each other. Each truncation creates a pattern of bits that 642 * are removed. We use this pattern of truncated bits to partition the inodes 643 * on a single device to many different devices (each one represented by the 644 * truncated bit pattern). All inodes on the same device that have the same 645 * truncation pattern are mapped to the same new device. Two inodes that 646 * truncate to the same value clearly will always have different truncation 647 * bit patterns, so they will be split from away each other. When we spot 648 * device truncation we remap the device number to a non truncated value. 649 * (for more info see table.h for the data structures involved). 650 */ 651 652 /* 653 * dev_start() 654 * create the device mapping table 655 * Return: 656 * 0 if successful, -1 otherwise 657 */ 658 659 int 660 dev_start(void) 661 { 662 if (dtab != NULL) 663 return(0); 664 if ((dtab = (DEVT **)calloc(D_TAB_SZ, sizeof(DEVT *))) == NULL) { 665 paxwarn(1, "Cannot allocate memory for device mapping table"); 666 return(-1); 667 } 668 return(0); 669 } 670 671 /* 672 * add_dev() 673 * add a device number to the table. this will force the device to be 674 * remapped to a new value if it be used during a write phase. This 675 * function is called during the read phase of an append to prohibit the 676 * use of any device number already in the archive. 677 * Return: 678 * 0 if added ok, -1 otherwise 679 */ 680 681 int 682 add_dev(ARCHD *arcn) 683 { 684 if (chk_dev(arcn->sb.st_dev, 1) == NULL) 685 return(-1); 686 return(0); 687 } 688 689 /* 690 * chk_dev() 691 * check for a device value in the device table. If not found and the add 692 * flag is set, it is added. This does NOT assign any mapping values, just 693 * adds the device number as one that need to be remapped. If this device 694 * is already mapped, just return with a pointer to that entry. 695 * Return: 696 * pointer to the entry for this device in the device map table. Null 697 * if the add flag is not set and the device is not in the table (it is 698 * not been seen yet). If add is set and the device cannot be added, null 699 * is returned (indicates an error). 700 */ 701 702 static DEVT * 703 chk_dev(dev_t dev, int add) 704 { 705 DEVT *pt; 706 u_int indx; 707 708 if (dtab == NULL) 709 return(NULL); 710 /* 711 * look to see if this device is already in the table 712 */ 713 indx = ((unsigned)dev) % D_TAB_SZ; 714 if ((pt = dtab[indx]) != NULL) { 715 while ((pt != NULL) && (pt->dev != dev)) 716 pt = pt->fow; 717 718 /* 719 * found it, return a pointer to it 720 */ 721 if (pt != NULL) 722 return(pt); 723 } 724 725 /* 726 * not in table, we add it only if told to as this may just be a check 727 * to see if a device number is being used. 728 */ 729 if (add == 0) 730 return(NULL); 731 732 /* 733 * allocate a node for this device and add it to the front of the hash 734 * chain. Note we do not assign remaps values here, so the pt->list 735 * list must be NULL. 736 */ 737 if ((pt = (DEVT *)malloc(sizeof(DEVT))) == NULL) { 738 paxwarn(1, "Device map table out of memory"); 739 return(NULL); 740 } 741 pt->dev = dev; 742 pt->list = NULL; 743 pt->fow = dtab[indx]; 744 dtab[indx] = pt; 745 return(pt); 746 } 747 /* 748 * map_dev() 749 * given an inode and device storage mask (the mask has a 1 for each bit 750 * the archive format is able to store in a header), we check for inode 751 * and device truncation and remap the device as required. Device mapping 752 * can also occur when during the read phase of append a device number was 753 * seen (and was marked as do not use during the write phase). WE ASSUME 754 * that unsigned longs are the same size or bigger than the fields used 755 * for ino_t and dev_t. If not the types will have to be changed. 756 * Return: 757 * 0 if all ok, -1 otherwise. 758 */ 759 760 int 761 map_dev(ARCHD *arcn, u_long dev_mask, u_long ino_mask) 762 { 763 DEVT *pt; 764 DLIST *dpt; 765 static dev_t lastdev = 0; /* next device number to try */ 766 int trc_ino = 0; 767 int trc_dev = 0; 768 ino_t trunc_bits = 0; 769 ino_t nino; 770 771 if (dtab == NULL) 772 return(0); 773 /* 774 * check for device and inode truncation, and extract the truncated 775 * bit pattern. 776 */ 777 if ((arcn->sb.st_dev & (dev_t)dev_mask) != arcn->sb.st_dev) 778 ++trc_dev; 779 if ((nino = arcn->sb.st_ino & (ino_t)ino_mask) != arcn->sb.st_ino) { 780 ++trc_ino; 781 trunc_bits = arcn->sb.st_ino & (ino_t)(~ino_mask); 782 } 783 784 /* 785 * see if this device is already being mapped, look up the device 786 * then find the truncation bit pattern which applies 787 */ 788 if ((pt = chk_dev(arcn->sb.st_dev, 0)) != NULL) { 789 /* 790 * this device is already marked to be remapped 791 */ 792 for (dpt = pt->list; dpt != NULL; dpt = dpt->fow) 793 if (dpt->trunc_bits == trunc_bits) 794 break; 795 796 if (dpt != NULL) { 797 /* 798 * we are being remapped for this device and pattern 799 * change the device number to be stored and return 800 */ 801 arcn->sb.st_dev = dpt->dev; 802 arcn->sb.st_ino = nino; 803 return(0); 804 } 805 } else { 806 /* 807 * this device is not being remapped YET. if we do not have any 808 * form of truncation, we do not need a remap 809 */ 810 if (!trc_ino && !trc_dev) 811 return(0); 812 813 /* 814 * we have truncation, have to add this as a device to remap 815 */ 816 if ((pt = chk_dev(arcn->sb.st_dev, 1)) == NULL) 817 goto bad; 818 819 /* 820 * if we just have a truncated inode, we have to make sure that 821 * all future inodes that do not truncate (they have the 822 * truncation pattern of all 0's) continue to map to the same 823 * device number. We probably have already written inodes with 824 * this device number to the archive with the truncation 825 * pattern of all 0's. So we add the mapping for all 0's to the 826 * same device number. 827 */ 828 if (!trc_dev && (trunc_bits != 0)) { 829 if ((dpt = (DLIST *)malloc(sizeof(DLIST))) == NULL) 830 goto bad; 831 dpt->trunc_bits = 0; 832 dpt->dev = arcn->sb.st_dev; 833 dpt->fow = pt->list; 834 pt->list = dpt; 835 } 836 } 837 838 /* 839 * look for a device number not being used. We must watch for wrap 840 * around on lastdev (so we do not get stuck looking forever!) 841 */ 842 while (++lastdev > 0) { 843 if (chk_dev(lastdev, 0) != NULL) 844 continue; 845 /* 846 * found an unused value. If we have reached truncation point 847 * for this format we are hosed, so we give up. Otherwise we 848 * mark it as being used. 849 */ 850 if (((lastdev & ((dev_t)dev_mask)) != lastdev) || 851 (chk_dev(lastdev, 1) == NULL)) 852 goto bad; 853 break; 854 } 855 856 if ((lastdev <= 0) || ((dpt = (DLIST *)malloc(sizeof(DLIST))) == NULL)) 857 goto bad; 858 859 /* 860 * got a new device number, store it under this truncation pattern. 861 * change the device number this file is being stored with. 862 */ 863 dpt->trunc_bits = trunc_bits; 864 dpt->dev = lastdev; 865 dpt->fow = pt->list; 866 pt->list = dpt; 867 arcn->sb.st_dev = lastdev; 868 arcn->sb.st_ino = nino; 869 return(0); 870 871 bad: 872 paxwarn(1, "Unable to fix truncated inode/device field when storing %s", 873 arcn->name); 874 paxwarn(0, "Archive may create improper hard links when extracted"); 875 return(0); 876 } 877 878 /* 879 * directory access/mod time reset table routines (for directories READ by pax) 880 * 881 * The pax -t flag requires that access times of archive files be the same 882 * before being read by pax. For regular files, access time is restored after 883 * the file has been copied. This database provides the same functionality for 884 * directories read during file tree traversal. Restoring directory access time 885 * is more complex than files since directories may be read several times until 886 * all the descendants in their subtree are visited by fts. Directory access 887 * and modification times are stored during the fts pre-order visit (done 888 * before any descendants in the subtree are visited) and restored after the 889 * fts post-order visit (after all the descendants have been visited). In the 890 * case of premature exit from a subtree (like from the effects of -n), any 891 * directory entries left in this database are reset during final cleanup 892 * operations of pax. Entries are hashed by inode number for fast lookup. 893 */ 894 895 /* 896 * atdir_start() 897 * create the directory access time database for directories READ by pax. 898 * Return: 899 * 0 is created ok, -1 otherwise. 900 */ 901 902 int 903 atdir_start(void) 904 { 905 if (atab != NULL) 906 return(0); 907 if ((atab = (ATDIR **)calloc(A_TAB_SZ, sizeof(ATDIR *))) == NULL) { 908 paxwarn(1,"Cannot allocate space for directory access time table"); 909 return(-1); 910 } 911 return(0); 912 } 913 914 915 /* 916 * atdir_end() 917 * walk through the directory access time table and reset the access time 918 * of any directory who still has an entry left in the database. These 919 * entries are for directories READ by pax 920 */ 921 922 void 923 atdir_end(void) 924 { 925 ATDIR *pt; 926 int i; 927 928 if (atab == NULL) 929 return; 930 /* 931 * for each non-empty hash table entry reset all the directories 932 * chained there. 933 */ 934 for (i = 0; i < A_TAB_SZ; ++i) { 935 if ((pt = atab[i]) == NULL) 936 continue; 937 /* 938 * remember to force the times, set_ftime() looks at pmtime 939 * and patime, which only applies to things CREATED by pax, 940 * not read by pax. Read time reset is controlled by -t. 941 */ 942 for (; pt != NULL; pt = pt->fow) 943 set_ftime(pt->name, pt->mtime, pt->atime, 1); 944 } 945 } 946 947 /* 948 * add_atdir() 949 * add a directory to the directory access time table. Table is hashed 950 * and chained by inode number. This is for directories READ by pax 951 */ 952 953 void 954 add_atdir(char *fname, dev_t dev, ino_t ino, time_t mtime, time_t atime) 955 { 956 ATDIR *pt; 957 u_int indx; 958 959 if (atab == NULL) 960 return; 961 962 /* 963 * make sure this directory is not already in the table, if so just 964 * return (the older entry always has the correct time). The only 965 * way this will happen is when the same subtree can be traversed by 966 * different args to pax and the -n option is aborting fts out of a 967 * subtree before all the post-order visits have been made. 968 */ 969 indx = ((unsigned)ino) % A_TAB_SZ; 970 if ((pt = atab[indx]) != NULL) { 971 while (pt != NULL) { 972 if ((pt->ino == ino) && (pt->dev == dev)) 973 break; 974 pt = pt->fow; 975 } 976 977 /* 978 * oops, already there. Leave it alone. 979 */ 980 if (pt != NULL) 981 return; 982 } 983 984 /* 985 * add it to the front of the hash chain 986 */ 987 if ((pt = (ATDIR *)malloc(sizeof(ATDIR))) != NULL) { 988 if ((pt->name = strdup(fname)) != NULL) { 989 pt->dev = dev; 990 pt->ino = ino; 991 pt->mtime = mtime; 992 pt->atime = atime; 993 pt->fow = atab[indx]; 994 atab[indx] = pt; 995 return; 996 } 997 free((char *)pt); 998 } 999 1000 paxwarn(1, "Directory access time reset table ran out of memory"); 1001 return; 1002 } 1003 1004 /* 1005 * get_atdir() 1006 * look up a directory by inode and device number to obtain the access 1007 * and modification time you want to set to. If found, the modification 1008 * and access time parameters are set and the entry is removed from the 1009 * table (as it is no longer needed). These are for directories READ by 1010 * pax 1011 * Return: 1012 * 0 if found, -1 if not found. 1013 */ 1014 1015 int 1016 get_atdir(dev_t dev, ino_t ino, time_t *mtime, time_t *atime) 1017 { 1018 ATDIR *pt; 1019 ATDIR **ppt; 1020 u_int indx; 1021 1022 if (atab == NULL) 1023 return(-1); 1024 /* 1025 * hash by inode and search the chain for an inode and device match 1026 */ 1027 indx = ((unsigned)ino) % A_TAB_SZ; 1028 if ((pt = atab[indx]) == NULL) 1029 return(-1); 1030 1031 ppt = &(atab[indx]); 1032 while (pt != NULL) { 1033 if ((pt->ino == ino) && (pt->dev == dev)) 1034 break; 1035 /* 1036 * no match, go to next one 1037 */ 1038 ppt = &(pt->fow); 1039 pt = pt->fow; 1040 } 1041 1042 /* 1043 * return if we did not find it. 1044 */ 1045 if (pt == NULL) 1046 return(-1); 1047 1048 /* 1049 * found it. return the times and remove the entry from the table. 1050 */ 1051 *ppt = pt->fow; 1052 *mtime = pt->mtime; 1053 *atime = pt->atime; 1054 free((char *)pt->name); 1055 free((char *)pt); 1056 return(0); 1057 } 1058 1059 /* 1060 * directory access mode and time storage routines (for directories CREATED 1061 * by pax). 1062 * 1063 * Pax requires that extracted directories, by default, have their access/mod 1064 * times and permissions set to the values specified in the archive. During the 1065 * actions of extracting (and creating the destination subtree during -rw copy) 1066 * directories extracted may be modified after being created. Even worse is 1067 * that these directories may have been created with file permissions which 1068 * prohibits any descendants of these directories from being extracted. When 1069 * directories are created by pax, access rights may be added to permit the 1070 * creation of files in their subtree. Every time pax creates a directory, the 1071 * times and file permissions specified by the archive are stored. After all 1072 * files have been extracted (or copied), these directories have their times 1073 * and file modes reset to the stored values. The directory info is restored in 1074 * reverse order as entries were added to the data file from root to leaf. To 1075 * restore atime properly, we must go backwards. The data file consists of 1076 * records with two parts, the file name followed by a DIRDATA trailer. The 1077 * fixed sized trailer contains the size of the name plus the off_t location in 1078 * the file. To restore we work backwards through the file reading the trailer 1079 * then the file name. 1080 */ 1081 1082 /* 1083 * dir_start() 1084 * set up the directory time and file mode storage for directories CREATED 1085 * by pax. 1086 * Return: 1087 * 0 if ok, -1 otherwise 1088 */ 1089 1090 int 1091 dir_start(void) 1092 { 1093 1094 if (dirfd != -1) 1095 return(0); 1096 1097 /* 1098 * unlink the file so it goes away at termination by itself 1099 */ 1100 memcpy(tempbase, _TFILE_BASE, sizeof(_TFILE_BASE)); 1101 if ((dirfd = mkstemp(tempfile)) >= 0) { 1102 unlink(tempfile); 1103 return(0); 1104 } 1105 paxwarn(1, "Unable to create temporary file for directory times: %s", 1106 tempfile); 1107 return(-1); 1108 } 1109 1110 /* 1111 * add_dir() 1112 * add the mode and times for a newly CREATED directory 1113 * name is name of the directory, psb the stat buffer with the data in it, 1114 * frc_mode is a flag that says whether to force the setting of the mode 1115 * (ignoring the user set values for preserving file mode). Frc_mode is 1116 * for the case where we created a file and found that the resulting 1117 * directory was not writeable and the user asked for file modes to NOT 1118 * be preserved. (we have to preserve what was created by default, so we 1119 * have to force the setting at the end. this is stated explicitly in the 1120 * pax spec) 1121 */ 1122 1123 void 1124 add_dir(char *name, int nlen, struct stat *psb, int frc_mode) 1125 { 1126 DIRDATA dblk; 1127 1128 if (dirfd < 0) 1129 return; 1130 1131 /* 1132 * get current position (where file name will start) so we can store it 1133 * in the trailer 1134 */ 1135 if ((dblk.npos = lseek(dirfd, 0L, SEEK_CUR)) < 0) { 1136 paxwarn(1,"Unable to store mode and times for directory: %s",name); 1137 return; 1138 } 1139 1140 /* 1141 * write the file name followed by the trailer 1142 */ 1143 dblk.nlen = nlen + 1; 1144 dblk.mode = psb->st_mode & 0xffff; 1145 dblk.mtime = psb->st_mtime; 1146 dblk.atime = psb->st_atime; 1147 dblk.frc_mode = frc_mode; 1148 if ((write(dirfd, name, dblk.nlen) == dblk.nlen) && 1149 (write(dirfd, (char *)&dblk, sizeof(dblk)) == sizeof(dblk))) { 1150 ++dircnt; 1151 return; 1152 } 1153 1154 paxwarn(1,"Unable to store mode and times for created directory: %s",name); 1155 return; 1156 } 1157 1158 /* 1159 * proc_dir() 1160 * process all file modes and times stored for directories CREATED 1161 * by pax 1162 */ 1163 1164 void 1165 proc_dir(void) 1166 { 1167 char name[PAXPATHLEN+1]; 1168 DIRDATA dblk; 1169 u_long cnt; 1170 1171 if (dirfd < 0) 1172 return; 1173 /* 1174 * read backwards through the file and process each directory 1175 */ 1176 for (cnt = 0; cnt < dircnt; ++cnt) { 1177 /* 1178 * read the trailer, then the file name, if this fails 1179 * just give up. 1180 */ 1181 if (lseek(dirfd, -((off_t)sizeof(dblk)), SEEK_CUR) < 0) 1182 break; 1183 if (read(dirfd,(char *)&dblk, sizeof(dblk)) != sizeof(dblk)) 1184 break; 1185 if (lseek(dirfd, dblk.npos, SEEK_SET) < 0) 1186 break; 1187 if (read(dirfd, name, dblk.nlen) != dblk.nlen) 1188 break; 1189 if (lseek(dirfd, dblk.npos, SEEK_SET) < 0) 1190 break; 1191 1192 /* 1193 * frc_mode set, make sure we set the file modes even if 1194 * the user didn't ask for it (see file_subs.c for more info) 1195 */ 1196 if (pmode || dblk.frc_mode) 1197 set_pmode(name, dblk.mode); 1198 if (patime || pmtime) 1199 set_ftime(name, dblk.mtime, dblk.atime, 0); 1200 } 1201 1202 close(dirfd); 1203 dirfd = -1; 1204 if (cnt != dircnt) 1205 paxwarn(1,"Unable to set mode and times for created directories"); 1206 return; 1207 } 1208 1209 /* 1210 * database independent routines 1211 */ 1212 1213 /* 1214 * st_hash() 1215 * hashes filenames to a u_int for hashing into a table. Looks at the tail 1216 * end of file, as this provides far better distribution than any other 1217 * part of the name. For performance reasons we only care about the last 1218 * MAXKEYLEN chars (should be at LEAST large enough to pick off the file 1219 * name). Was tested on 500,000 name file tree traversal from the root 1220 * and gave almost a perfectly uniform distribution of keys when used with 1221 * prime sized tables (MAXKEYLEN was 128 in test). Hashes (sizeof int) 1222 * chars at a time and pads with 0 for last addition. 1223 * Return: 1224 * the hash value of the string MOD (%) the table size. 1225 */ 1226 1227 u_int 1228 st_hash(char *name, int len, int tabsz) 1229 { 1230 char *pt; 1231 char *dest; 1232 char *end; 1233 int i; 1234 u_int key = 0; 1235 int steps; 1236 int res; 1237 u_int val; 1238 1239 /* 1240 * only look at the tail up to MAXKEYLEN, we do not need to waste 1241 * time here (remember these are pathnames, the tail is what will 1242 * spread out the keys) 1243 */ 1244 if (len > MAXKEYLEN) { 1245 pt = &(name[len - MAXKEYLEN]); 1246 len = MAXKEYLEN; 1247 } else 1248 pt = name; 1249 1250 /* 1251 * calculate the number of u_int size steps in the string and if 1252 * there is a runt to deal with 1253 */ 1254 steps = len/sizeof(u_int); 1255 res = len % sizeof(u_int); 1256 1257 /* 1258 * add up the value of the string in unsigned integer sized pieces 1259 * too bad we cannot have unsigned int aligned strings, then we 1260 * could avoid the expensive copy. 1261 */ 1262 for (i = 0; i < steps; ++i) { 1263 end = pt + sizeof(u_int); 1264 dest = (char *)&val; 1265 while (pt < end) 1266 *dest++ = *pt++; 1267 key += val; 1268 } 1269 1270 /* 1271 * add in the runt padded with zero to the right 1272 */ 1273 if (res) { 1274 val = 0; 1275 end = pt + res; 1276 dest = (char *)&val; 1277 while (pt < end) 1278 *dest++ = *pt++; 1279 key += val; 1280 } 1281 1282 /* 1283 * return the result mod the table size 1284 */ 1285 return(key % tabsz); 1286 } 1287