file_subs.c revision 1.38
1/* $NetBSD: file_subs.c,v 1.38 2004/02/13 00:07:55 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 1.38 2004/02/13 00:07:55 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 + 9); 95 if (arcn->tmp_name == NULL) { 96 syswarn(1, errno, "Cannot malloc %d bytes", arcn->nlen + 9); 97 return(-1); 98 } 99 (void)snprintf(arcn->tmp_name, arcn->nlen + 9, ".%s.XXXXXX", 100 arcn->name); 101 102 fd = mkstemp(arcn->tmp_name); 103 if (fd >= 0) 104 return(fd); 105 106 for (;;) { 107 /* 108 * try to create the temporary file we use to restore the 109 * contents info. if this fails, keep checking all the nodes 110 * in the path until chk_path() finds that it cannot fix 111 * anything further. if that happens we just give up. 112 */ 113 fd = mkstemp(arcn->tmp_name); 114 if (fd >= 0) 115 break; 116 oerrno = errno; 117 if (nodirs || chk_path(arcn->tmp_name,arcn->sb.st_uid,arcn->sb.st_gid) < 0) { 118 (void)fflush(listf); 119 syswarn(1, oerrno, "Cannot create %s", arcn->tmp_name); 120 free(arcn->tmp_name); 121 arcn->tmp_name = NULL; 122 return(-1); 123 } 124 } 125 return(fd); 126} 127 128/* 129 * file_close() 130 * Close file descriptor to a file just created by pax. Sets modes, 131 * ownership and times as required. 132 * Return: 133 * 0 for success, -1 for failure 134 */ 135 136void 137file_close(ARCHD *arcn, int fd) 138{ 139 int res = 0; 140 141 if (fd < 0) 142 return; 143 if (close(fd) < 0) 144 syswarn(0, errno, "Cannot close file descriptor on %s", 145 arcn->tmp_name); 146 147 /* 148 * set owner/groups first as this may strip off mode bits we want 149 * then set file permission modes. Then set file access and 150 * modification times. 151 */ 152 if (pids) 153 res = set_ids(arcn->tmp_name, arcn->sb.st_uid, arcn->sb.st_gid); 154 155 /* 156 * IMPORTANT SECURITY NOTE: 157 * if not preserving mode or we cannot set uid/gid, then PROHIBIT 158 * set uid/gid bits but restore the file modes (since mkstemp doesn't). 159 */ 160 if (!pmode || res) 161 arcn->sb.st_mode &= ~(SETBITS); 162 if (pmode) 163 set_pmode(arcn->tmp_name, arcn->sb.st_mode); 164 else 165 set_pmode(arcn->tmp_name, arcn->sb.st_mode & FILEBITS); 166 if (patime || pmtime) 167 set_ftime(arcn->tmp_name, arcn->sb.st_mtime, arcn->sb.st_atime, 0); 168#if HAVE_STRUCT_STAT_ST_FLAGS 169 if (pfflags && arcn->type != PAX_SLK) 170 set_chflags(arcn->tmp_name, arcn->sb.st_flags); 171#endif 172 173 /* 174 * Finally, now the temp file is fully instantiated rename it to 175 * the desired file name. 176 */ 177 if (rename(arcn->tmp_name, arcn->name) < 0) { 178 syswarn(0, errno, "Cannot rename %s to %s", 179 arcn->tmp_name, arcn->name); 180 (void)unlink(arcn->tmp_name); 181 } 182 183 free(arcn->tmp_name); 184 arcn->tmp_name = NULL; 185} 186 187/* 188 * lnk_creat() 189 * Create a hard link to arcn->ln_name from arcn->name. arcn->ln_name 190 * must exist; 191 * Return: 192 * 0 if ok, -1 otherwise 193 */ 194 195int 196lnk_creat(ARCHD *arcn) 197{ 198 struct stat sb; 199 200 /* 201 * we may be running as root, so we have to be sure that link target 202 * is not a directory, so we lstat and check 203 */ 204 if (lstat(arcn->ln_name, &sb) < 0) { 205 syswarn(1, errno, "Cannot link to %s from %s", arcn->ln_name, 206 arcn->name); 207 return(-1); 208 } 209 210 if (S_ISDIR(sb.st_mode)) { 211 tty_warn(1, "A hard link to the directory %s is not allowed", 212 arcn->ln_name); 213 return(-1); 214 } 215 216 return(mk_link(arcn->ln_name, &sb, arcn->name, 0)); 217} 218 219/* 220 * cross_lnk() 221 * Create a hard link to arcn->org_name from arcn->name. Only used in copy 222 * with the -l flag. No warning or error if this does not succeed (we will 223 * then just create the file) 224 * Return: 225 * 1 if copy() should try to create this file node 226 * 0 if cross_lnk() ok, -1 for fatal flaw (like linking to self). 227 */ 228 229int 230cross_lnk(ARCHD *arcn) 231{ 232 /* 233 * try to make a link to original file (-l flag in copy mode). make 234 * sure we do not try to link to directories in case we are running as 235 * root (and it might succeed). 236 */ 237 if (arcn->type == PAX_DIR) 238 return(1); 239 return(mk_link(arcn->org_name, &(arcn->sb), arcn->name, 1)); 240} 241 242/* 243 * chk_same() 244 * In copy mode if we are not trying to make hard links between the src 245 * and destinations, make sure we are not going to overwrite ourselves by 246 * accident. This slows things down a little, but we have to protect all 247 * those people who make typing errors. 248 * Return: 249 * 1 the target does not exist, go ahead and copy 250 * 0 skip it file exists (-k) or may be the same as source file 251 */ 252 253int 254chk_same(ARCHD *arcn) 255{ 256 struct stat sb; 257 258 /* 259 * if file does not exist, return. if file exists and -k, skip it 260 * quietly 261 */ 262 if (lstat(arcn->name, &sb) < 0) 263 return(1); 264 if (kflag) 265 return(0); 266 267 /* 268 * better make sure the user does not have src == dest by mistake 269 */ 270 if ((arcn->sb.st_dev == sb.st_dev) && (arcn->sb.st_ino == sb.st_ino)) { 271 tty_warn(1, "Unable to copy %s, file would overwrite itself", 272 arcn->name); 273 return(0); 274 } 275 return(1); 276} 277 278/* 279 * mk_link() 280 * try to make a hard link between two files. if ign set, we do not 281 * complain. 282 * Return: 283 * 0 if successful (or we are done with this file but no error, such as 284 * finding the from file exists and the user has set -k). 285 * 1 when ign was set to indicates we could not make the link but we 286 * should try to copy/extract the file as that might work (and is an 287 * allowed option). -1 an error occurred. 288 */ 289 290static int 291mk_link(char *to, struct stat *to_sb, char *from, int ign) 292{ 293 struct stat sb; 294 int oerrno; 295 296 /* 297 * if from file exists, it has to be unlinked to make the link. If the 298 * file exists and -k is set, skip it quietly 299 */ 300 if (lstat(from, &sb) == 0) { 301 if (kflag) 302 return(0); 303 304 /* 305 * make sure it is not the same file, protect the user 306 */ 307 if ((to_sb->st_dev==sb.st_dev)&&(to_sb->st_ino == sb.st_ino)) { 308 tty_warn(1, "Cannot link file %s to itself", to); 309 return(-1); 310 } 311 312 /* 313 * try to get rid of the file, based on the type 314 */ 315 if (S_ISDIR(sb.st_mode)) { 316 if (rmdir(from) < 0) { 317 syswarn(1, errno, "Cannot remove %s", from); 318 return(-1); 319 } 320 } else if (unlink(from) < 0) { 321 if (!ign) { 322 syswarn(1, errno, "Cannot remove %s", from); 323 return(-1); 324 } 325 return(1); 326 } 327 } 328 329 /* 330 * from file is gone (or did not exist), try to make the hard link. 331 * if it fails, check the path and try it again (if chk_path() says to 332 * try again) 333 */ 334 for (;;) { 335 if (link(to, from) == 0) 336 break; 337 oerrno = errno; 338 if (chk_path(from, to_sb->st_uid, to_sb->st_gid) == 0) 339 continue; 340 if (!ign) { 341 syswarn(1, oerrno, "Cannot link to %s from %s", to, 342 from); 343 return(-1); 344 } 345 return(1); 346 } 347 348 /* 349 * all right the link was made 350 */ 351 return(0); 352} 353 354/* 355 * node_creat() 356 * create an entry in the file system (other than a file or hard link). 357 * If successful, sets uid/gid modes and times as required. 358 * Return: 359 * 0 if ok, -1 otherwise 360 */ 361 362int 363node_creat(ARCHD *arcn) 364{ 365 int res; 366 int ign = 0; 367 int oerrno; 368 int pass = 0; 369 mode_t file_mode; 370 struct stat sb; 371 char target[MAXPATHLEN]; 372 char *nm = arcn->name; 373 int len; 374 375 /* 376 * create node based on type, if that fails try to unlink the node and 377 * try again. finally check the path and try again. As noted in the 378 * file and link creation routines, this method seems to exhibit the 379 * best performance in general use workloads. 380 */ 381 file_mode = arcn->sb.st_mode & FILEBITS; 382 383 for (;;) { 384 switch(arcn->type) { 385 case PAX_DIR: 386 /* 387 * If -h (or -L) was given in tar-mode, follow the 388 * potential symlink chain before trying to create the 389 * directory. 390 */ 391 if (strcmp(NM_TAR, argv0) == 0 && Lflag) { 392 while (lstat(nm, &sb) == 0 && 393 S_ISLNK(sb.st_mode)) { 394 len = readlink(nm, target, 395 sizeof target - 1); 396 if (len == -1) { 397 syswarn(0, errno, 398 "cannot follow symlink %s in chain for %s", 399 nm, arcn->name); 400 res = -1; 401 goto badlink; 402 } 403 target[len] = '\0'; 404 nm = target; 405 } 406 } 407 res = mkdir(nm, file_mode); 408 409badlink: 410 if (ign) 411 res = 0; 412 break; 413 case PAX_CHR: 414 file_mode |= S_IFCHR; 415 res = mknod(nm, file_mode, arcn->sb.st_rdev); 416 break; 417 case PAX_BLK: 418 file_mode |= S_IFBLK; 419 res = mknod(nm, file_mode, arcn->sb.st_rdev); 420 break; 421 case PAX_FIF: 422 res = mkfifo(nm, file_mode); 423 break; 424 case PAX_SCK: 425 /* 426 * Skip sockets, operation has no meaning under BSD 427 */ 428 tty_warn(0, 429 "%s skipped. Sockets cannot be copied or extracted", 430 nm); 431 return(-1); 432 case PAX_SLK: 433 res = symlink(arcn->ln_name, nm); 434 break; 435 case PAX_CTG: 436 case PAX_HLK: 437 case PAX_HRG: 438 case PAX_REG: 439 default: 440 /* 441 * we should never get here 442 */ 443 tty_warn(0, "%s has an unknown file type, skipping", 444 nm); 445 return(-1); 446 } 447 448 /* 449 * if we were able to create the node break out of the loop, 450 * otherwise try to unlink the node and try again. if that 451 * fails check the full path and try a final time. 452 */ 453 if (res == 0) 454 break; 455 456 /* 457 * we failed to make the node 458 */ 459 oerrno = errno; 460 if ((ign = unlnk_exist(nm, arcn->type)) < 0) 461 return(-1); 462 463 if (++pass <= 1) 464 continue; 465 466 if (nodirs || chk_path(nm,arcn->sb.st_uid,arcn->sb.st_gid) < 0) { 467 syswarn(1, oerrno, "Cannot create %s", nm); 468 return(-1); 469 } 470 } 471 472 /* 473 * we were able to create the node. set uid/gid, modes and times 474 */ 475 if (pids) 476 res = set_ids(nm, arcn->sb.st_uid, arcn->sb.st_gid); 477 else 478 res = 0; 479 480 /* 481 * IMPORTANT SECURITY NOTE: 482 * if not preserving mode or we cannot set uid/gid, then PROHIBIT any 483 * set uid/gid bits 484 */ 485 if (!pmode || res) 486 arcn->sb.st_mode &= ~(SETBITS); 487 if (pmode) 488 set_pmode(arcn->name, arcn->sb.st_mode); 489 490 if (arcn->type == PAX_DIR && strcmp(NM_CPIO, argv0) != 0) { 491 /* 492 * Dirs must be processed again at end of extract to set times 493 * and modes to agree with those stored in the archive. However 494 * to allow extract to continue, we may have to also set owner 495 * rights. This allows nodes in the archive that are children 496 * of this directory to be extracted without failure. Both time 497 * and modes will be fixed after the entire archive is read and 498 * before pax exits. 499 */ 500 if (access(nm, R_OK | W_OK | X_OK) < 0) { 501 if (lstat(nm, &sb) < 0) { 502 syswarn(0, errno,"Cannot access %s (stat)", 503 arcn->name); 504 set_pmode(nm,file_mode | S_IRWXU); 505 } else { 506 /* 507 * We have to add rights to the dir, so we make 508 * sure to restore the mode. The mode must be 509 * restored AS CREATED and not as stored if 510 * pmode is not set. 511 */ 512 set_pmode(nm, 513 ((sb.st_mode & FILEBITS) | S_IRWXU)); 514 if (!pmode) 515 arcn->sb.st_mode = sb.st_mode; 516 } 517 518 /* 519 * we have to force the mode to what was set here, 520 * since we changed it from the default as created. 521 */ 522 add_dir(nm, arcn->nlen, &(arcn->sb), 1); 523 } else if (pmode || patime || pmtime) 524 add_dir(nm, arcn->nlen, &(arcn->sb), 0); 525 } 526 527#if HAVE_LUTIMES 528 if (patime || pmtime) 529#else 530 if ((patime || pmtime) && arcn->type != PAX_SLK) 531#endif 532 set_ftime(arcn->name, arcn->sb.st_mtime, arcn->sb.st_atime, 0); 533 534#if HAVE_STRUCT_STAT_ST_FLAGS 535 if (pfflags && arcn->type != PAX_SLK) 536 set_chflags(arcn->name, arcn->sb.st_flags); 537#endif 538 return(0); 539} 540 541/* 542 * unlnk_exist() 543 * Remove node from file system with the specified name. We pass the type 544 * of the node that is going to replace it. When we try to create a 545 * directory and find that it already exists, we allow processing to 546 * continue as proper modes etc will always be set for it later on. 547 * Return: 548 * 0 is ok to proceed, no file with the specified name exists 549 * -1 we were unable to remove the node, or we should not remove it (-k) 550 * 1 we found a directory and we were going to create a directory. 551 */ 552 553int 554unlnk_exist(char *name, int type) 555{ 556 struct stat sb; 557 558 /* 559 * the file does not exist, or -k we are done 560 */ 561 if (lstat(name, &sb) < 0) 562 return(0); 563 if (kflag) 564 return(-1); 565 566 if (S_ISDIR(sb.st_mode)) { 567 /* 568 * try to remove a directory, if it fails and we were going to 569 * create a directory anyway, tell the caller (return a 1) 570 */ 571 if (rmdir(name) < 0) { 572 if (type == PAX_DIR) 573 return(1); 574 syswarn(1, errno, "Cannot remove directory %s", name); 575 return(-1); 576 } 577 return(0); 578 } 579 580 /* 581 * try to get rid of all non-directory type nodes 582 */ 583 if (unlink(name) < 0) { 584 (void)fflush(listf); 585 syswarn(1, errno, "Cannot unlink %s", name); 586 return(-1); 587 } 588 return(0); 589} 590 591/* 592 * chk_path() 593 * We were trying to create some kind of node in the file system and it 594 * failed. chk_path() makes sure the path up to the node exists and is 595 * writable. When we have to create a directory that is missing along the 596 * path somewhere, the directory we create will be set to the same 597 * uid/gid as the file has (when uid and gid are being preserved). 598 * NOTE: this routine is a real performance loss. It is only used as a 599 * last resort when trying to create entries in the file system. 600 * Return: 601 * -1 when it could find nothing it is allowed to fix. 602 * 0 otherwise 603 */ 604 605int 606chk_path( char *name, uid_t st_uid, gid_t st_gid) 607{ 608 char *spt = name; 609 struct stat sb; 610 int retval = -1; 611 612 /* 613 * watch out for paths with nodes stored directly in / (e.g. /bozo) 614 */ 615 if (*spt == '/') 616 ++spt; 617 618 for(;;) { 619 /* 620 * work forward from the first / and check each part of 621 * the path 622 */ 623 spt = strchr(spt, '/'); 624 if (spt == NULL) 625 break; 626 *spt = '\0'; 627 628 /* 629 * if it exists we assume it is a directory, it is not within 630 * the spec (at least it seems to read that way) to alter the 631 * file system for nodes NOT EXPLICITLY stored on the archive. 632 * If that assumption is changed, you would test the node here 633 * and figure out how to get rid of it (probably like some 634 * recursive unlink()) or fix up the directory permissions if 635 * required (do an access()). 636 */ 637 if (lstat(name, &sb) == 0) { 638 *(spt++) = '/'; 639 continue; 640 } 641 642 /* 643 * the path fails at this point, see if we can create the 644 * needed directory and continue on 645 */ 646 if (mkdir(name, S_IRWXU | S_IRWXG | S_IRWXO) < 0) { 647 *spt = '/'; 648 retval = -1; 649 break; 650 } 651 652 /* 653 * we were able to create the directory. We will tell the 654 * caller that we found something to fix, and it is ok to try 655 * and create the node again. 656 */ 657 retval = 0; 658 if (pids) 659 (void)set_ids(name, st_uid, st_gid); 660 661 /* 662 * make sure the user doesn't have some strange umask that 663 * causes this newly created directory to be unusable. We fix 664 * the modes and restore them back to the creation default at 665 * the end of pax 666 */ 667 if ((access(name, R_OK | W_OK | X_OK) < 0) && 668 (lstat(name, &sb) == 0)) { 669 set_pmode(name, ((sb.st_mode & FILEBITS) | S_IRWXU)); 670 add_dir(name, spt - name, &sb, 1); 671 } 672 *(spt++) = '/'; 673 continue; 674 } 675 return(retval); 676} 677 678/* 679 * set_ftime() 680 * Set the access time and modification time for a named file. If frc 681 * is non-zero we force these times to be set even if the user did not 682 * request access and/or modification time preservation (this is also 683 * used by -t to reset access times). 684 * When ign is zero, only those times the user has asked for are set, the 685 * other ones are left alone. We do not assume the un-documented feature 686 * of many utimes() implementations that consider a 0 time value as a do 687 * not set request. 688 */ 689 690void 691set_ftime(char *fnm, time_t mtime, time_t atime, int frc) 692{ 693 struct timeval tv[2]; 694 struct stat sb; 695 696 tv[0].tv_sec = (long)atime; 697 tv[0].tv_usec = 0; 698 tv[1].tv_sec = (long)mtime; 699 tv[1].tv_usec = 0; 700 if (!frc && (!patime || !pmtime)) { 701 /* 702 * if we are not forcing, only set those times the user wants 703 * set. We get the current values of the times if we need them. 704 */ 705 if (lstat(fnm, &sb) == 0) { 706#ifdef BSD4_4 707 if (!patime) 708 TIMESPEC_TO_TIMEVAL(&tv[0], &sb.st_atimespec); 709 if (!pmtime) 710 TIMESPEC_TO_TIMEVAL(&tv[1], &sb.st_mtimespec); 711#else 712 if (!patime) 713 tv[0].tv_sec = sb.st_atime; 714 if (!pmtime) 715 tv[1].tv_sec = sb.st_mtime; 716#endif 717 } else 718 syswarn(0, errno, "Cannot obtain file stats %s", fnm); 719 } 720 721 /* 722 * set the times 723 */ 724#if HAVE_LUTIMES 725 if (lutimes(fnm, tv)) 726#else 727 if (utimes(fnm, tv)) 728#endif 729 syswarn(1, errno, "Access/modification time set failed on: %s", 730 fnm); 731 return; 732} 733 734/* 735 * set_ids() 736 * set the uid and gid of a file system node 737 * Return: 738 * 0 when set, -1 on failure 739 */ 740 741int 742set_ids(char *fnm, uid_t uid, gid_t gid) 743{ 744 if (geteuid() == 0) 745 if (lchown(fnm, uid, gid)) { 746 (void)fflush(listf); 747 syswarn(1, errno, "Cannot set file uid/gid of %s", 748 fnm); 749 return(-1); 750 } 751 return(0); 752} 753 754/* 755 * set_pmode() 756 * Set file access mode 757 */ 758 759void 760set_pmode(char *fnm, mode_t mode) 761{ 762 mode &= ABITS; 763 if (lchmod(fnm, mode)) { 764 (void)fflush(listf); 765 syswarn(1, errno, "Cannot set permissions on %s", fnm); 766 } 767 return; 768} 769 770/* 771 * set_chflags() 772 * Set 4.4BSD file flags 773 */ 774void 775set_chflags(char *fnm, u_int32_t flags) 776{ 777 778#if 0 779 if (chflags(fnm, flags) < 0 && errno != EOPNOTSUPP) 780 syswarn(1, errno, "Cannot set file flags on %s", fnm); 781#endif 782 return; 783} 784 785/* 786 * file_write() 787 * Write/copy a file (during copy or archive extract). This routine knows 788 * how to copy files with lseek holes in it. (Which are read as file 789 * blocks containing all 0's but do not have any file blocks associated 790 * with the data). Typical examples of these are files created by dbm 791 * variants (.pag files). While the file size of these files are huge, the 792 * actual storage is quite small (the files are sparse). The problem is 793 * the holes read as all zeros so are probably stored on the archive that 794 * way (there is no way to determine if the file block is really a hole, 795 * we only know that a file block of all zero's can be a hole). 796 * At this writing, no major archive format knows how to archive files 797 * with holes. However, on extraction (or during copy, -rw) we have to 798 * deal with these files. Without detecting the holes, the files can 799 * consume a lot of file space if just written to disk. This replacement 800 * for write when passed the basic allocation size of a file system block, 801 * uses lseek whenever it detects the input data is all 0 within that 802 * file block. In more detail, the strategy is as follows: 803 * While the input is all zero keep doing an lseek. Keep track of when we 804 * pass over file block boundaries. Only write when we hit a non zero 805 * input. once we have written a file block, we continue to write it to 806 * the end (we stop looking at the input). When we reach the start of the 807 * next file block, start checking for zero blocks again. Working on file 808 * block boundaries significantly reduces the overhead when copying files 809 * that are NOT very sparse. This overhead (when compared to a write) is 810 * almost below the measurement resolution on many systems. Without it, 811 * files with holes cannot be safely copied. It does has a side effect as 812 * it can put holes into files that did not have them before, but that is 813 * not a problem since the file contents are unchanged (in fact it saves 814 * file space). (Except on paging files for diskless clients. But since we 815 * cannot determine one of those file from here, we ignore them). If this 816 * ever ends up on a system where CTG files are supported and the holes 817 * are not desired, just do a conditional test in those routines that 818 * call file_write() and have it call write() instead. BEFORE CLOSING THE 819 * FILE, make sure to call file_flush() when the last write finishes with 820 * an empty block. A lot of file systems will not create an lseek hole at 821 * the end. In this case we drop a single 0 at the end to force the 822 * trailing 0's in the file. 823 * ---Parameters--- 824 * rem: how many bytes left in this file system block 825 * isempt: have we written to the file block yet (is it empty) 826 * sz: basic file block allocation size 827 * cnt: number of bytes on this write 828 * str: buffer to write 829 * Return: 830 * number of bytes written, -1 on write (or lseek) error. 831 */ 832 833int 834file_write(int fd, char *str, int cnt, int *rem, int *isempt, int sz, 835 char *name) 836{ 837 char *pt; 838 char *end; 839 int wcnt; 840 char *st = str; 841 char **strp; 842 843 /* 844 * while we have data to process 845 */ 846 while (cnt) { 847 if (!*rem) { 848 /* 849 * We are now at the start of file system block again 850 * (or what we think one is...). start looking for 851 * empty blocks again 852 */ 853 *isempt = 1; 854 *rem = sz; 855 } 856 857 /* 858 * only examine up to the end of the current file block or 859 * remaining characters to write, whatever is smaller 860 */ 861 wcnt = MIN(cnt, *rem); 862 cnt -= wcnt; 863 *rem -= wcnt; 864 if (*isempt) { 865 /* 866 * have not written to this block yet, so we keep 867 * looking for zero's 868 */ 869 pt = st; 870 end = st + wcnt; 871 872 /* 873 * look for a zero filled buffer 874 */ 875 while ((pt < end) && (*pt == '\0')) 876 ++pt; 877 878 if (pt == end) { 879 /* 880 * skip, buf is empty so far 881 */ 882 if (fd > -1 && 883 lseek(fd, (off_t)wcnt, SEEK_CUR) < 0) { 884 syswarn(1, errno, "File seek on %s", 885 name); 886 return(-1); 887 } 888 st = pt; 889 continue; 890 } 891 /* 892 * drat, the buf is not zero filled 893 */ 894 *isempt = 0; 895 } 896 897 /* 898 * have non-zero data in this file system block, have to write 899 */ 900 switch (fd) { 901 case -1: 902 strp = &gnu_name_string; 903 break; 904 case -2: 905 strp = &gnu_link_string; 906 break; 907 default: 908 strp = NULL; 909 break; 910 } 911 if (strp) { 912 if (*strp) 913 err(1, "WARNING! Major Internal Error! GNU hack Failing!"); 914 *strp = malloc(wcnt + 1); 915 if (*strp == NULL) { 916 tty_warn(1, "Out of memory"); 917 return(-1); 918 } 919 strlcpy(*strp, st, wcnt); 920 break; 921 } else if (xwrite(fd, st, wcnt) != wcnt) { 922 syswarn(1, errno, "Failed write to file %s", name); 923 return(-1); 924 } 925 st += wcnt; 926 } 927 return(st - str); 928} 929 930/* 931 * file_flush() 932 * when the last file block in a file is zero, many file systems will not 933 * let us create a hole at the end. To get the last block with zeros, we 934 * write the last BYTE with a zero (back up one byte and write a zero). 935 */ 936 937void 938file_flush(int fd, char *fname, int isempt) 939{ 940 static char blnk[] = "\0"; 941 942 /* 943 * silly test, but make sure we are only called when the last block is 944 * filled with all zeros. 945 */ 946 if (!isempt) 947 return; 948 949 /* 950 * move back one byte and write a zero 951 */ 952 if (lseek(fd, (off_t)-1, SEEK_CUR) < 0) { 953 syswarn(1, errno, "Failed seek on file %s", fname); 954 return; 955 } 956 957 if (write_with_restart(fd, blnk, 1) < 0) 958 syswarn(1, errno, "Failed write to file %s", fname); 959 return; 960} 961 962/* 963 * rdfile_close() 964 * close a file we have been reading (to copy or archive). If we have to 965 * reset access time (tflag) do so (the times are stored in arcn). 966 */ 967 968void 969rdfile_close(ARCHD *arcn, int *fd) 970{ 971 /* 972 * make sure the file is open 973 */ 974 if (*fd < 0) 975 return; 976 977 (void)close(*fd); 978 *fd = -1; 979 if (!tflag) 980 return; 981 982 /* 983 * user wants last access time reset 984 */ 985 set_ftime(arcn->org_name, arcn->sb.st_mtime, arcn->sb.st_atime, 1); 986 return; 987} 988 989/* 990 * set_crc() 991 * read a file to calculate its crc. This is a real drag. Archive formats 992 * that have this, end up reading the file twice (we have to write the 993 * header WITH the crc before writing the file contents. Oh well... 994 * Return: 995 * 0 if was able to calculate the crc, -1 otherwise 996 */ 997 998int 999set_crc(ARCHD *arcn, int fd) 1000{ 1001 int i; 1002 int res; 1003 off_t cpcnt = 0L; 1004 u_long size; 1005 unsigned long crc = 0L; 1006 char tbuf[FILEBLK]; 1007 struct stat sb; 1008 1009 if (fd < 0) { 1010 /* 1011 * hmm, no fd, should never happen. well no crc then. 1012 */ 1013 arcn->crc = 0L; 1014 return(0); 1015 } 1016 1017 if ((size = (u_long)arcn->sb.st_blksize) > (u_long)sizeof(tbuf)) 1018 size = (u_long)sizeof(tbuf); 1019 1020 /* 1021 * read all the bytes we think that there are in the file. If the user 1022 * is trying to archive an active file, forget this file. 1023 */ 1024 for(;;) { 1025 if ((res = read(fd, tbuf, size)) <= 0) 1026 break; 1027 cpcnt += res; 1028 for (i = 0; i < res; ++i) 1029 crc += (tbuf[i] & 0xff); 1030 } 1031 1032 /* 1033 * safety check. we want to avoid archiving files that are active as 1034 * they can create inconsistant archive copies. 1035 */ 1036 if (cpcnt != arcn->sb.st_size) 1037 tty_warn(1, "File changed size %s", arcn->org_name); 1038 else if (fstat(fd, &sb) < 0) 1039 syswarn(1, errno, "Failed stat on %s", arcn->org_name); 1040 else if (arcn->sb.st_mtime != sb.st_mtime) 1041 tty_warn(1, "File %s was modified during read", arcn->org_name); 1042 else if (lseek(fd, (off_t)0L, SEEK_SET) < 0) 1043 syswarn(1, errno, "File rewind failed on: %s", arcn->org_name); 1044 else { 1045 arcn->crc = crc; 1046 return(0); 1047 } 1048 return(-1); 1049} 1050