file_subs.c revision 1.37
1/* $NetBSD: file_subs.c,v 1.37 2004/02/12 22:19:18 matt 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. Neither the name of the University nor the names of its contributors 20 * may be used to endorse or promote products derived from this software 21 * without specific prior written permission. 22 * 23 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 24 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 25 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 26 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 27 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 28 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 29 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 30 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 31 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 32 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 33 * SUCH DAMAGE. 34 */ 35 36#if HAVE_NBTOOL_CONFIG_H 37#include "nbtool_config.h" 38#endif 39 40#include <sys/cdefs.h> 41#if !defined(lint) 42#if 0 43static char sccsid[] = "@(#)file_subs.c 8.1 (Berkeley) 5/31/93"; 44#else 45__RCSID("$NetBSD: file_subs.c,v 1.37 2004/02/12 22:19:18 matt 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 <unistd.h> 53#include <sys/param.h> 54#include <fcntl.h> 55#include <string.h> 56#include <stdio.h> 57#include <ctype.h> 58#include <errno.h> 59#include <sys/uio.h> 60#include <stdlib.h> 61#include "pax.h" 62#include "extern.h" 63#include "options.h" 64 65static int 66mk_link(char *,struct stat *,char *, int); 67 68/* 69 * routines that deal with file operations such as: creating, removing; 70 * and setting access modes, uid/gid and times of files 71 */ 72 73#define FILEBITS (S_ISVTX | S_IRWXU | S_IRWXG | S_IRWXO) 74#define SETBITS (S_ISUID | S_ISGID) 75#define ABITS (FILEBITS | SETBITS) 76 77/* 78 * file_creat() 79 * Create and open a file. 80 * Return: 81 * file descriptor or -1 for failure 82 */ 83 84int 85file_creat(ARCHD *arcn) 86{ 87 int fd = -1; 88 int oerrno; 89 90 /* 91 * Create a temporary file name so that the file doesn't have partial 92 * contents while restoring. 93 */ 94 arcn->tmp_name = malloc(arcn->nlen + 8); 95 if (arcn->tmp_name == NULL) { 96 syswarn(1, errno, "Cannot malloc %d bytes", arcn->nlen + 8); 97 return(-1); 98 } 99 (void)snprintf(arcn->tmp_name, arcn->nlen + 8, "%s.XXXXXX", arcn->name); 100 101 fd = mkstemp(arcn->tmp_name); 102 if (fd >= 0) 103 return(fd); 104 105 for (;;) { 106 /* 107 * try to create the temporary file we use to restore the 108 * contents info. if this fails, keep checking all the nodes 109 * in the path until chk_path() finds that it cannot fix 110 * anything further. if that happens we just give up. 111 */ 112 fd = mkstemp(arcn->tmp_name); 113 if (fd >= 0) 114 break; 115 oerrno = errno; 116 if (nodirs || chk_path(arcn->tmp_name,arcn->sb.st_uid,arcn->sb.st_gid) < 0) { 117 (void)fflush(listf); 118 syswarn(1, oerrno, "Cannot create %s", arcn->tmp_name); 119 free(arcn->tmp_name); 120 arcn->tmp_name = NULL; 121 return(-1); 122 } 123 } 124 return(fd); 125} 126 127/* 128 * file_close() 129 * Close file descriptor to a file just created by pax. Sets modes, 130 * ownership and times as required. 131 * Return: 132 * 0 for success, -1 for failure 133 */ 134 135void 136file_close(ARCHD *arcn, int fd) 137{ 138 int res = 0; 139 140 if (fd < 0) 141 return; 142 if (close(fd) < 0) 143 syswarn(0, errno, "Cannot close file descriptor on %s", 144 arcn->tmp_name); 145 146 /* 147 * set owner/groups first as this may strip off mode bits we want 148 * then set file permission modes. Then set file access and 149 * modification times. 150 */ 151 if (pids) 152 res = set_ids(arcn->tmp_name, arcn->sb.st_uid, arcn->sb.st_gid); 153 154 /* 155 * IMPORTANT SECURITY NOTE: 156 * if not preserving mode or we cannot set uid/gid, then PROHIBIT 157 * set uid/gid bits but restore the file modes (since mkstemp doesn't). 158 */ 159 if (!pmode || res) 160 arcn->sb.st_mode &= ~(SETBITS); 161 if (pmode) 162 set_pmode(arcn->tmp_name, arcn->sb.st_mode); 163 else 164 set_pmode(arcn->tmp_name, arcn->sb.st_mode & FILEBITS); 165 if (patime || pmtime) 166 set_ftime(arcn->tmp_name, arcn->sb.st_mtime, arcn->sb.st_atime, 0); 167#if HAVE_STRUCT_STAT_ST_FLAGS 168 if (pfflags && arcn->type != PAX_SLK) 169 set_chflags(arcn->tmp_name, arcn->sb.st_flags); 170#endif 171 172 /* 173 * Finally, now the temp file is fully instantiated rename it to 174 * the desired file name. 175 */ 176 if (rename(arcn->tmp_name, arcn->name) < 0) { 177 syswarn(0, errno, "Cannot rename %s to %s", 178 arcn->tmp_name, arcn->name); 179 (void)unlink(arcn->tmp_name); 180 } 181 182 free(arcn->tmp_name); 183 arcn->tmp_name = NULL; 184} 185 186/* 187 * lnk_creat() 188 * Create a hard link to arcn->ln_name from arcn->name. arcn->ln_name 189 * must exist; 190 * Return: 191 * 0 if ok, -1 otherwise 192 */ 193 194int 195lnk_creat(ARCHD *arcn) 196{ 197 struct stat sb; 198 199 /* 200 * we may be running as root, so we have to be sure that link target 201 * is not a directory, so we lstat and check 202 */ 203 if (lstat(arcn->ln_name, &sb) < 0) { 204 syswarn(1, errno, "Cannot link to %s from %s", arcn->ln_name, 205 arcn->name); 206 return(-1); 207 } 208 209 if (S_ISDIR(sb.st_mode)) { 210 tty_warn(1, "A hard link to the directory %s is not allowed", 211 arcn->ln_name); 212 return(-1); 213 } 214 215 return(mk_link(arcn->ln_name, &sb, arcn->name, 0)); 216} 217 218/* 219 * cross_lnk() 220 * Create a hard link to arcn->org_name from arcn->name. Only used in copy 221 * with the -l flag. No warning or error if this does not succeed (we will 222 * then just create the file) 223 * Return: 224 * 1 if copy() should try to create this file node 225 * 0 if cross_lnk() ok, -1 for fatal flaw (like linking to self). 226 */ 227 228int 229cross_lnk(ARCHD *arcn) 230{ 231 /* 232 * try to make a link to original file (-l flag in copy mode). make 233 * sure we do not try to link to directories in case we are running as 234 * root (and it might succeed). 235 */ 236 if (arcn->type == PAX_DIR) 237 return(1); 238 return(mk_link(arcn->org_name, &(arcn->sb), arcn->name, 1)); 239} 240 241/* 242 * chk_same() 243 * In copy mode if we are not trying to make hard links between the src 244 * and destinations, make sure we are not going to overwrite ourselves by 245 * accident. This slows things down a little, but we have to protect all 246 * those people who make typing errors. 247 * Return: 248 * 1 the target does not exist, go ahead and copy 249 * 0 skip it file exists (-k) or may be the same as source file 250 */ 251 252int 253chk_same(ARCHD *arcn) 254{ 255 struct stat sb; 256 257 /* 258 * if file does not exist, return. if file exists and -k, skip it 259 * quietly 260 */ 261 if (lstat(arcn->name, &sb) < 0) 262 return(1); 263 if (kflag) 264 return(0); 265 266 /* 267 * better make sure the user does not have src == dest by mistake 268 */ 269 if ((arcn->sb.st_dev == sb.st_dev) && (arcn->sb.st_ino == sb.st_ino)) { 270 tty_warn(1, "Unable to copy %s, file would overwrite itself", 271 arcn->name); 272 return(0); 273 } 274 return(1); 275} 276 277/* 278 * mk_link() 279 * try to make a hard link between two files. if ign set, we do not 280 * complain. 281 * Return: 282 * 0 if successful (or we are done with this file but no error, such as 283 * finding the from file exists and the user has set -k). 284 * 1 when ign was set to indicates we could not make the link but we 285 * should try to copy/extract the file as that might work (and is an 286 * allowed option). -1 an error occurred. 287 */ 288 289static int 290mk_link(char *to, struct stat *to_sb, char *from, int ign) 291{ 292 struct stat sb; 293 int oerrno; 294 295 /* 296 * if from file exists, it has to be unlinked to make the link. If the 297 * file exists and -k is set, skip it quietly 298 */ 299 if (lstat(from, &sb) == 0) { 300 if (kflag) 301 return(0); 302 303 /* 304 * make sure it is not the same file, protect the user 305 */ 306 if ((to_sb->st_dev==sb.st_dev)&&(to_sb->st_ino == sb.st_ino)) { 307 tty_warn(1, "Cannot link file %s to itself", to); 308 return(-1); 309 } 310 311 /* 312 * try to get rid of the file, based on the type 313 */ 314 if (S_ISDIR(sb.st_mode)) { 315 if (rmdir(from) < 0) { 316 syswarn(1, errno, "Cannot remove %s", from); 317 return(-1); 318 } 319 } else if (unlink(from) < 0) { 320 if (!ign) { 321 syswarn(1, errno, "Cannot remove %s", from); 322 return(-1); 323 } 324 return(1); 325 } 326 } 327 328 /* 329 * from file is gone (or did not exist), try to make the hard link. 330 * if it fails, check the path and try it again (if chk_path() says to 331 * try again) 332 */ 333 for (;;) { 334 if (link(to, from) == 0) 335 break; 336 oerrno = errno; 337 if (chk_path(from, to_sb->st_uid, to_sb->st_gid) == 0) 338 continue; 339 if (!ign) { 340 syswarn(1, oerrno, "Cannot link to %s from %s", to, 341 from); 342 return(-1); 343 } 344 return(1); 345 } 346 347 /* 348 * all right the link was made 349 */ 350 return(0); 351} 352 353/* 354 * node_creat() 355 * create an entry in the file system (other than a file or hard link). 356 * If successful, sets uid/gid modes and times as required. 357 * Return: 358 * 0 if ok, -1 otherwise 359 */ 360 361int 362node_creat(ARCHD *arcn) 363{ 364 int res; 365 int ign = 0; 366 int oerrno; 367 int pass = 0; 368 mode_t file_mode; 369 struct stat sb; 370 char target[MAXPATHLEN]; 371 char *nm = arcn->name; 372 int len; 373 374 /* 375 * create node based on type, if that fails try to unlink the node and 376 * try again. finally check the path and try again. As noted in the 377 * file and link creation routines, this method seems to exhibit the 378 * best performance in general use workloads. 379 */ 380 file_mode = arcn->sb.st_mode & FILEBITS; 381 382 for (;;) { 383 switch(arcn->type) { 384 case PAX_DIR: 385 /* 386 * If -h (or -L) was given in tar-mode, follow the 387 * potential symlink chain before trying to create the 388 * directory. 389 */ 390 if (strcmp(NM_TAR, argv0) == 0 && Lflag) { 391 while (lstat(nm, &sb) == 0 && 392 S_ISLNK(sb.st_mode)) { 393 len = readlink(nm, target, 394 sizeof target - 1); 395 if (len == -1) { 396 syswarn(0, errno, 397 "cannot follow symlink %s in chain for %s", 398 nm, arcn->name); 399 res = -1; 400 goto badlink; 401 } 402 target[len] = '\0'; 403 nm = target; 404 } 405 } 406 res = mkdir(nm, file_mode); 407 408badlink: 409 if (ign) 410 res = 0; 411 break; 412 case PAX_CHR: 413 file_mode |= S_IFCHR; 414 res = mknod(nm, file_mode, arcn->sb.st_rdev); 415 break; 416 case PAX_BLK: 417 file_mode |= S_IFBLK; 418 res = mknod(nm, file_mode, arcn->sb.st_rdev); 419 break; 420 case PAX_FIF: 421 res = mkfifo(nm, file_mode); 422 break; 423 case PAX_SCK: 424 /* 425 * Skip sockets, operation has no meaning under BSD 426 */ 427 tty_warn(0, 428 "%s skipped. Sockets cannot be copied or extracted", 429 nm); 430 return(-1); 431 case PAX_SLK: 432 res = symlink(arcn->ln_name, nm); 433 break; 434 case PAX_CTG: 435 case PAX_HLK: 436 case PAX_HRG: 437 case PAX_REG: 438 default: 439 /* 440 * we should never get here 441 */ 442 tty_warn(0, "%s has an unknown file type, skipping", 443 nm); 444 return(-1); 445 } 446 447 /* 448 * if we were able to create the node break out of the loop, 449 * otherwise try to unlink the node and try again. if that 450 * fails check the full path and try a final time. 451 */ 452 if (res == 0) 453 break; 454 455 /* 456 * we failed to make the node 457 */ 458 oerrno = errno; 459 if ((ign = unlnk_exist(nm, arcn->type)) < 0) 460 return(-1); 461 462 if (++pass <= 1) 463 continue; 464 465 if (nodirs || chk_path(nm,arcn->sb.st_uid,arcn->sb.st_gid) < 0) { 466 syswarn(1, oerrno, "Cannot create %s", nm); 467 return(-1); 468 } 469 } 470 471 /* 472 * we were able to create the node. set uid/gid, modes and times 473 */ 474 if (pids) 475 res = set_ids(nm, arcn->sb.st_uid, arcn->sb.st_gid); 476 else 477 res = 0; 478 479 /* 480 * IMPORTANT SECURITY NOTE: 481 * if not preserving mode or we cannot set uid/gid, then PROHIBIT any 482 * set uid/gid bits 483 */ 484 if (!pmode || res) 485 arcn->sb.st_mode &= ~(SETBITS); 486 if (pmode) 487 set_pmode(arcn->name, arcn->sb.st_mode); 488 489 if (arcn->type == PAX_DIR && strcmp(NM_CPIO, argv0) != 0) { 490 /* 491 * Dirs must be processed again at end of extract to set times 492 * and modes to agree with those stored in the archive. However 493 * to allow extract to continue, we may have to also set owner 494 * rights. This allows nodes in the archive that are children 495 * of this directory to be extracted without failure. Both time 496 * and modes will be fixed after the entire archive is read and 497 * before pax exits. 498 */ 499 if (access(nm, R_OK | W_OK | X_OK) < 0) { 500 if (lstat(nm, &sb) < 0) { 501 syswarn(0, errno,"Cannot access %s (stat)", 502 arcn->name); 503 set_pmode(nm,file_mode | S_IRWXU); 504 } else { 505 /* 506 * We have to add rights to the dir, so we make 507 * sure to restore the mode. The mode must be 508 * restored AS CREATED and not as stored if 509 * pmode is not set. 510 */ 511 set_pmode(nm, 512 ((sb.st_mode & FILEBITS) | S_IRWXU)); 513 if (!pmode) 514 arcn->sb.st_mode = sb.st_mode; 515 } 516 517 /* 518 * we have to force the mode to what was set here, 519 * since we changed it from the default as created. 520 */ 521 add_dir(nm, arcn->nlen, &(arcn->sb), 1); 522 } else if (pmode || patime || pmtime) 523 add_dir(nm, arcn->nlen, &(arcn->sb), 0); 524 } 525 526#if HAVE_LUTIMES 527 if (patime || pmtime) 528#else 529 if ((patime || pmtime) && arcn->type != PAX_SLK) 530#endif 531 set_ftime(arcn->name, arcn->sb.st_mtime, arcn->sb.st_atime, 0); 532 533#if HAVE_STRUCT_STAT_ST_FLAGS 534 if (pfflags && arcn->type != PAX_SLK) 535 set_chflags(arcn->name, arcn->sb.st_flags); 536#endif 537 return(0); 538} 539 540/* 541 * unlnk_exist() 542 * Remove node from file system with the specified name. We pass the type 543 * of the node that is going to replace it. When we try to create a 544 * directory and find that it already exists, we allow processing to 545 * continue as proper modes etc will always be set for it later on. 546 * Return: 547 * 0 is ok to proceed, no file with the specified name exists 548 * -1 we were unable to remove the node, or we should not remove it (-k) 549 * 1 we found a directory and we were going to create a directory. 550 */ 551 552int 553unlnk_exist(char *name, int type) 554{ 555 struct stat sb; 556 557 /* 558 * the file does not exist, or -k we are done 559 */ 560 if (lstat(name, &sb) < 0) 561 return(0); 562 if (kflag) 563 return(-1); 564 565 if (S_ISDIR(sb.st_mode)) { 566 /* 567 * try to remove a directory, if it fails and we were going to 568 * create a directory anyway, tell the caller (return a 1) 569 */ 570 if (rmdir(name) < 0) { 571 if (type == PAX_DIR) 572 return(1); 573 syswarn(1, errno, "Cannot remove directory %s", name); 574 return(-1); 575 } 576 return(0); 577 } 578 579 /* 580 * try to get rid of all non-directory type nodes 581 */ 582 if (unlink(name) < 0) { 583 (void)fflush(listf); 584 syswarn(1, errno, "Cannot unlink %s", name); 585 return(-1); 586 } 587 return(0); 588} 589 590/* 591 * chk_path() 592 * We were trying to create some kind of node in the file system and it 593 * failed. chk_path() makes sure the path up to the node exists and is 594 * writable. When we have to create a directory that is missing along the 595 * path somewhere, the directory we create will be set to the same 596 * uid/gid as the file has (when uid and gid are being preserved). 597 * NOTE: this routine is a real performance loss. It is only used as a 598 * last resort when trying to create entries in the file system. 599 * Return: 600 * -1 when it could find nothing it is allowed to fix. 601 * 0 otherwise 602 */ 603 604int 605chk_path( char *name, uid_t st_uid, gid_t st_gid) 606{ 607 char *spt = name; 608 struct stat sb; 609 int retval = -1; 610 611 /* 612 * watch out for paths with nodes stored directly in / (e.g. /bozo) 613 */ 614 if (*spt == '/') 615 ++spt; 616 617 for(;;) { 618 /* 619 * work forward from the first / and check each part of 620 * the path 621 */ 622 spt = strchr(spt, '/'); 623 if (spt == NULL) 624 break; 625 *spt = '\0'; 626 627 /* 628 * if it exists we assume it is a directory, it is not within 629 * the spec (at least it seems to read that way) to alter the 630 * file system for nodes NOT EXPLICITLY stored on the archive. 631 * If that assumption is changed, you would test the node here 632 * and figure out how to get rid of it (probably like some 633 * recursive unlink()) or fix up the directory permissions if 634 * required (do an access()). 635 */ 636 if (lstat(name, &sb) == 0) { 637 *(spt++) = '/'; 638 continue; 639 } 640 641 /* 642 * the path fails at this point, see if we can create the 643 * needed directory and continue on 644 */ 645 if (mkdir(name, S_IRWXU | S_IRWXG | S_IRWXO) < 0) { 646 *spt = '/'; 647 retval = -1; 648 break; 649 } 650 651 /* 652 * we were able to create the directory. We will tell the 653 * caller that we found something to fix, and it is ok to try 654 * and create the node again. 655 */ 656 retval = 0; 657 if (pids) 658 (void)set_ids(name, st_uid, st_gid); 659 660 /* 661 * make sure the user doesn't have some strange umask that 662 * causes this newly created directory to be unusable. We fix 663 * the modes and restore them back to the creation default at 664 * the end of pax 665 */ 666 if ((access(name, R_OK | W_OK | X_OK) < 0) && 667 (lstat(name, &sb) == 0)) { 668 set_pmode(name, ((sb.st_mode & FILEBITS) | S_IRWXU)); 669 add_dir(name, spt - name, &sb, 1); 670 } 671 *(spt++) = '/'; 672 continue; 673 } 674 return(retval); 675} 676 677/* 678 * set_ftime() 679 * Set the access time and modification time for a named file. If frc 680 * is non-zero we force these times to be set even if the user did not 681 * request access and/or modification time preservation (this is also 682 * used by -t to reset access times). 683 * When ign is zero, only those times the user has asked for are set, the 684 * other ones are left alone. We do not assume the un-documented feature 685 * of many utimes() implementations that consider a 0 time value as a do 686 * not set request. 687 */ 688 689void 690set_ftime(char *fnm, time_t mtime, time_t atime, int frc) 691{ 692 struct timeval tv[2]; 693 struct stat sb; 694 695 tv[0].tv_sec = (long)atime; 696 tv[0].tv_usec = 0; 697 tv[1].tv_sec = (long)mtime; 698 tv[1].tv_usec = 0; 699 if (!frc && (!patime || !pmtime)) { 700 /* 701 * if we are not forcing, only set those times the user wants 702 * set. We get the current values of the times if we need them. 703 */ 704 if (lstat(fnm, &sb) == 0) { 705#ifdef BSD4_4 706 if (!patime) 707 TIMESPEC_TO_TIMEVAL(&tv[0], &sb.st_atimespec); 708 if (!pmtime) 709 TIMESPEC_TO_TIMEVAL(&tv[1], &sb.st_mtimespec); 710#else 711 if (!patime) 712 tv[0].tv_sec = sb.st_atime; 713 if (!pmtime) 714 tv[1].tv_sec = sb.st_mtime; 715#endif 716 } else 717 syswarn(0, errno, "Cannot obtain file stats %s", fnm); 718 } 719 720 /* 721 * set the times 722 */ 723#if HAVE_LUTIMES 724 if (lutimes(fnm, tv)) 725#else 726 if (utimes(fnm, tv)) 727#endif 728 syswarn(1, errno, "Access/modification time set failed on: %s", 729 fnm); 730 return; 731} 732 733/* 734 * set_ids() 735 * set the uid and gid of a file system node 736 * Return: 737 * 0 when set, -1 on failure 738 */ 739 740int 741set_ids(char *fnm, uid_t uid, gid_t gid) 742{ 743 if (geteuid() == 0) 744 if (lchown(fnm, uid, gid)) { 745 (void)fflush(listf); 746 syswarn(1, errno, "Cannot set file uid/gid of %s", 747 fnm); 748 return(-1); 749 } 750 return(0); 751} 752 753/* 754 * set_pmode() 755 * Set file access mode 756 */ 757 758void 759set_pmode(char *fnm, mode_t mode) 760{ 761 mode &= ABITS; 762 if (lchmod(fnm, mode)) { 763 (void)fflush(listf); 764 syswarn(1, errno, "Cannot set permissions on %s", fnm); 765 } 766 return; 767} 768 769/* 770 * set_chflags() 771 * Set 4.4BSD file flags 772 */ 773void 774set_chflags(char *fnm, u_int32_t flags) 775{ 776 777#if 0 778 if (chflags(fnm, flags) < 0 && errno != EOPNOTSUPP) 779 syswarn(1, errno, "Cannot set file flags on %s", fnm); 780#endif 781 return; 782} 783 784/* 785 * file_write() 786 * Write/copy a file (during copy or archive extract). This routine knows 787 * how to copy files with lseek holes in it. (Which are read as file 788 * blocks containing all 0's but do not have any file blocks associated 789 * with the data). Typical examples of these are files created by dbm 790 * variants (.pag files). While the file size of these files are huge, the 791 * actual storage is quite small (the files are sparse). The problem is 792 * the holes read as all zeros so are probably stored on the archive that 793 * way (there is no way to determine if the file block is really a hole, 794 * we only know that a file block of all zero's can be a hole). 795 * At this writing, no major archive format knows how to archive files 796 * with holes. However, on extraction (or during copy, -rw) we have to 797 * deal with these files. Without detecting the holes, the files can 798 * consume a lot of file space if just written to disk. This replacement 799 * for write when passed the basic allocation size of a file system block, 800 * uses lseek whenever it detects the input data is all 0 within that 801 * file block. In more detail, the strategy is as follows: 802 * While the input is all zero keep doing an lseek. Keep track of when we 803 * pass over file block boundaries. Only write when we hit a non zero 804 * input. once we have written a file block, we continue to write it to 805 * the end (we stop looking at the input). When we reach the start of the 806 * next file block, start checking for zero blocks again. Working on file 807 * block boundaries significantly reduces the overhead when copying files 808 * that are NOT very sparse. This overhead (when compared to a write) is 809 * almost below the measurement resolution on many systems. Without it, 810 * files with holes cannot be safely copied. It does has a side effect as 811 * it can put holes into files that did not have them before, but that is 812 * not a problem since the file contents are unchanged (in fact it saves 813 * file space). (Except on paging files for diskless clients. But since we 814 * cannot determine one of those file from here, we ignore them). If this 815 * ever ends up on a system where CTG files are supported and the holes 816 * are not desired, just do a conditional test in those routines that 817 * call file_write() and have it call write() instead. BEFORE CLOSING THE 818 * FILE, make sure to call file_flush() when the last write finishes with 819 * an empty block. A lot of file systems will not create an lseek hole at 820 * the end. In this case we drop a single 0 at the end to force the 821 * trailing 0's in the file. 822 * ---Parameters--- 823 * rem: how many bytes left in this file system block 824 * isempt: have we written to the file block yet (is it empty) 825 * sz: basic file block allocation size 826 * cnt: number of bytes on this write 827 * str: buffer to write 828 * Return: 829 * number of bytes written, -1 on write (or lseek) error. 830 */ 831 832int 833file_write(int fd, char *str, int cnt, int *rem, int *isempt, int sz, 834 char *name) 835{ 836 char *pt; 837 char *end; 838 int wcnt; 839 char *st = str; 840 char **strp; 841 842 /* 843 * while we have data to process 844 */ 845 while (cnt) { 846 if (!*rem) { 847 /* 848 * We are now at the start of file system block again 849 * (or what we think one is...). start looking for 850 * empty blocks again 851 */ 852 *isempt = 1; 853 *rem = sz; 854 } 855 856 /* 857 * only examine up to the end of the current file block or 858 * remaining characters to write, whatever is smaller 859 */ 860 wcnt = MIN(cnt, *rem); 861 cnt -= wcnt; 862 *rem -= wcnt; 863 if (*isempt) { 864 /* 865 * have not written to this block yet, so we keep 866 * looking for zero's 867 */ 868 pt = st; 869 end = st + wcnt; 870 871 /* 872 * look for a zero filled buffer 873 */ 874 while ((pt < end) && (*pt == '\0')) 875 ++pt; 876 877 if (pt == end) { 878 /* 879 * skip, buf is empty so far 880 */ 881 if (fd > -1 && 882 lseek(fd, (off_t)wcnt, SEEK_CUR) < 0) { 883 syswarn(1, errno, "File seek on %s", 884 name); 885 return(-1); 886 } 887 st = pt; 888 continue; 889 } 890 /* 891 * drat, the buf is not zero filled 892 */ 893 *isempt = 0; 894 } 895 896 /* 897 * have non-zero data in this file system block, have to write 898 */ 899 switch (fd) { 900 case -1: 901 strp = &gnu_name_string; 902 break; 903 case -2: 904 strp = &gnu_link_string; 905 break; 906 default: 907 strp = NULL; 908 break; 909 } 910 if (strp) { 911 if (*strp) 912 err(1, "WARNING! Major Internal Error! GNU hack Failing!"); 913 *strp = malloc(wcnt + 1); 914 if (*strp == NULL) { 915 tty_warn(1, "Out of memory"); 916 return(-1); 917 } 918 strlcpy(*strp, st, wcnt); 919 break; 920 } else if (xwrite(fd, st, wcnt) != wcnt) { 921 syswarn(1, errno, "Failed write to file %s", name); 922 return(-1); 923 } 924 st += wcnt; 925 } 926 return(st - str); 927} 928 929/* 930 * file_flush() 931 * when the last file block in a file is zero, many file systems will not 932 * let us create a hole at the end. To get the last block with zeros, we 933 * write the last BYTE with a zero (back up one byte and write a zero). 934 */ 935 936void 937file_flush(int fd, char *fname, int isempt) 938{ 939 static char blnk[] = "\0"; 940 941 /* 942 * silly test, but make sure we are only called when the last block is 943 * filled with all zeros. 944 */ 945 if (!isempt) 946 return; 947 948 /* 949 * move back one byte and write a zero 950 */ 951 if (lseek(fd, (off_t)-1, SEEK_CUR) < 0) { 952 syswarn(1, errno, "Failed seek on file %s", fname); 953 return; 954 } 955 956 if (write_with_restart(fd, blnk, 1) < 0) 957 syswarn(1, errno, "Failed write to file %s", fname); 958 return; 959} 960 961/* 962 * rdfile_close() 963 * close a file we have been reading (to copy or archive). If we have to 964 * reset access time (tflag) do so (the times are stored in arcn). 965 */ 966 967void 968rdfile_close(ARCHD *arcn, int *fd) 969{ 970 /* 971 * make sure the file is open 972 */ 973 if (*fd < 0) 974 return; 975 976 (void)close(*fd); 977 *fd = -1; 978 if (!tflag) 979 return; 980 981 /* 982 * user wants last access time reset 983 */ 984 set_ftime(arcn->org_name, arcn->sb.st_mtime, arcn->sb.st_atime, 1); 985 return; 986} 987 988/* 989 * set_crc() 990 * read a file to calculate its crc. This is a real drag. Archive formats 991 * that have this, end up reading the file twice (we have to write the 992 * header WITH the crc before writing the file contents. Oh well... 993 * Return: 994 * 0 if was able to calculate the crc, -1 otherwise 995 */ 996 997int 998set_crc(ARCHD *arcn, int fd) 999{ 1000 int i; 1001 int res; 1002 off_t cpcnt = 0L; 1003 u_long size; 1004 unsigned long crc = 0L; 1005 char tbuf[FILEBLK]; 1006 struct stat sb; 1007 1008 if (fd < 0) { 1009 /* 1010 * hmm, no fd, should never happen. well no crc then. 1011 */ 1012 arcn->crc = 0L; 1013 return(0); 1014 } 1015 1016 if ((size = (u_long)arcn->sb.st_blksize) > (u_long)sizeof(tbuf)) 1017 size = (u_long)sizeof(tbuf); 1018 1019 /* 1020 * read all the bytes we think that there are in the file. If the user 1021 * is trying to archive an active file, forget this file. 1022 */ 1023 for(;;) { 1024 if ((res = read(fd, tbuf, size)) <= 0) 1025 break; 1026 cpcnt += res; 1027 for (i = 0; i < res; ++i) 1028 crc += (tbuf[i] & 0xff); 1029 } 1030 1031 /* 1032 * safety check. we want to avoid archiving files that are active as 1033 * they can create inconsistant archive copies. 1034 */ 1035 if (cpcnt != arcn->sb.st_size) 1036 tty_warn(1, "File changed size %s", arcn->org_name); 1037 else if (fstat(fd, &sb) < 0) 1038 syswarn(1, errno, "Failed stat on %s", arcn->org_name); 1039 else if (arcn->sb.st_mtime != sb.st_mtime) 1040 tty_warn(1, "File %s was modified during read", arcn->org_name); 1041 else if (lseek(fd, (off_t)0L, SEEK_SET) < 0) 1042 syswarn(1, errno, "File rewind failed on: %s", arcn->org_name); 1043 else { 1044 arcn->crc = crc; 1045 return(0); 1046 } 1047 return(-1); 1048} 1049