file_subs.c revision 1.41
1/* $NetBSD: file_subs.c,v 1.41 2004/02/13 23:10:14 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 1.41 2004/02/13 23:10:14 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 65char *xtmp_name; 66 67static int 68mk_link(char *,struct stat *,char *, int); 69 70/* 71 * routines that deal with file operations such as: creating, removing; 72 * and setting access modes, uid/gid and times of files 73 */ 74 75#define FILEBITS (S_ISVTX | S_IRWXU | S_IRWXG | S_IRWXO) 76#define SETBITS (S_ISUID | S_ISGID) 77#define ABITS (FILEBITS | SETBITS) 78 79/* 80 * file_creat() 81 * Create and open a file. 82 * Return: 83 * file descriptor or -1 for failure 84 */ 85 86int 87file_creat(ARCHD *arcn) 88{ 89 int fd = -1; 90 int oerrno; 91 92 /* 93 * Create a temporary file name so that the file doesn't have partial 94 * contents while restoring. 95 */ 96 arcn->tmp_name = malloc(arcn->nlen + 8); 97 if (arcn->tmp_name == NULL) { 98 syswarn(1, errno, "Cannot malloc %d bytes", arcn->nlen + 8); 99 return(-1); 100 } 101 if (xtmp_name) 102 abort(); 103 xtmp_name = arcn->tmp_name; 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 (void)snprintf(arcn->tmp_name, arcn->nlen + 8, "%s.XXXXXX", 113 arcn->name); 114 fd = mkstemp(arcn->tmp_name); 115 if (fd >= 0) 116 break; 117 oerrno = errno; 118 if (nodirs || chk_path(arcn->name,arcn->sb.st_uid,arcn->sb.st_gid) < 0) { 119 (void)fflush(listf); 120 syswarn(1, oerrno, "Cannot create %s", arcn->tmp_name); 121 free(arcn->tmp_name); 122 arcn->tmp_name = NULL; 123 return(-1); 124 } 125 } 126 return(fd); 127} 128 129/* 130 * file_close() 131 * Close file descriptor to a file just created by pax. Sets modes, 132 * ownership and times as required. 133 * Return: 134 * 0 for success, -1 for failure 135 */ 136 137void 138file_close(ARCHD *arcn, int fd) 139{ 140 int res = 0; 141 142 if (fd < 0) 143 return; 144 if (close(fd) < 0) 145 syswarn(0, errno, "Cannot close file descriptor on %s", 146 arcn->tmp_name); 147 148 /* 149 * set owner/groups first as this may strip off mode bits we want 150 * then set file permission modes. Then set file access and 151 * modification times. 152 */ 153 if (pids) 154 res = set_ids(arcn->tmp_name, arcn->sb.st_uid, arcn->sb.st_gid); 155 156 /* 157 * IMPORTANT SECURITY NOTE: 158 * if not preserving mode or we cannot set uid/gid, then PROHIBIT 159 * set uid/gid bits but restore the file modes (since mkstemp doesn't). 160 */ 161 if (!pmode || res) 162 arcn->sb.st_mode &= ~(SETBITS); 163 if (pmode) 164 set_pmode(arcn->tmp_name, arcn->sb.st_mode); 165 else 166 set_pmode(arcn->tmp_name, arcn->sb.st_mode & FILEBITS); 167 if (patime || pmtime) 168 set_ftime(arcn->tmp_name, arcn->sb.st_mtime, arcn->sb.st_atime, 0); 169#if HAVE_STRUCT_STAT_ST_FLAGS 170 if (pfflags && arcn->type != PAX_SLK) 171 set_chflags(arcn->tmp_name, arcn->sb.st_flags); 172#endif 173 174 /* 175 * Finally, now the temp file is fully instantiated rename it to 176 * the desired file name. 177 */ 178 if (rename(arcn->tmp_name, arcn->name) < 0) { 179 syswarn(0, errno, "Cannot rename %s to %s", 180 arcn->tmp_name, arcn->name); 181 (void)unlink(arcn->tmp_name); 182 } 183 184 free(arcn->tmp_name); 185 arcn->tmp_name = NULL; 186 xtmp_name = NULL; 187} 188 189/* 190 * lnk_creat() 191 * Create a hard link to arcn->ln_name from arcn->name. arcn->ln_name 192 * must exist; 193 * Return: 194 * 0 if ok, -1 otherwise 195 */ 196 197int 198lnk_creat(ARCHD *arcn) 199{ 200 struct stat sb; 201 202 /* 203 * we may be running as root, so we have to be sure that link target 204 * is not a directory, so we lstat and check 205 */ 206 if (lstat(arcn->ln_name, &sb) < 0) { 207 syswarn(1, errno, "Cannot link to %s from %s", arcn->ln_name, 208 arcn->name); 209 return(-1); 210 } 211 212 if (S_ISDIR(sb.st_mode)) { 213 tty_warn(1, "A hard link to the directory %s is not allowed", 214 arcn->ln_name); 215 return(-1); 216 } 217 218 return(mk_link(arcn->ln_name, &sb, arcn->name, 0)); 219} 220 221/* 222 * cross_lnk() 223 * Create a hard link to arcn->org_name from arcn->name. Only used in copy 224 * with the -l flag. No warning or error if this does not succeed (we will 225 * then just create the file) 226 * Return: 227 * 1 if copy() should try to create this file node 228 * 0 if cross_lnk() ok, -1 for fatal flaw (like linking to self). 229 */ 230 231int 232cross_lnk(ARCHD *arcn) 233{ 234 /* 235 * try to make a link to original file (-l flag in copy mode). make 236 * sure we do not try to link to directories in case we are running as 237 * root (and it might succeed). 238 */ 239 if (arcn->type == PAX_DIR) 240 return(1); 241 return(mk_link(arcn->org_name, &(arcn->sb), arcn->name, 1)); 242} 243 244/* 245 * chk_same() 246 * In copy mode if we are not trying to make hard links between the src 247 * and destinations, make sure we are not going to overwrite ourselves by 248 * accident. This slows things down a little, but we have to protect all 249 * those people who make typing errors. 250 * Return: 251 * 1 the target does not exist, go ahead and copy 252 * 0 skip it file exists (-k) or may be the same as source file 253 */ 254 255int 256chk_same(ARCHD *arcn) 257{ 258 struct stat sb; 259 260 /* 261 * if file does not exist, return. if file exists and -k, skip it 262 * quietly 263 */ 264 if (lstat(arcn->name, &sb) < 0) 265 return(1); 266 if (kflag) 267 return(0); 268 269 /* 270 * better make sure the user does not have src == dest by mistake 271 */ 272 if ((arcn->sb.st_dev == sb.st_dev) && (arcn->sb.st_ino == sb.st_ino)) { 273 tty_warn(1, "Unable to copy %s, file would overwrite itself", 274 arcn->name); 275 return(0); 276 } 277 return(1); 278} 279 280/* 281 * mk_link() 282 * try to make a hard link between two files. if ign set, we do not 283 * complain. 284 * Return: 285 * 0 if successful (or we are done with this file but no error, such as 286 * finding the from file exists and the user has set -k). 287 * 1 when ign was set to indicates we could not make the link but we 288 * should try to copy/extract the file as that might work (and is an 289 * allowed option). -1 an error occurred. 290 */ 291 292static int 293mk_link(char *to, struct stat *to_sb, char *from, int ign) 294{ 295 struct stat sb; 296 int oerrno; 297 298 /* 299 * if from file exists, it has to be unlinked to make the link. If the 300 * file exists and -k is set, skip it quietly 301 */ 302 if (lstat(from, &sb) == 0) { 303 if (kflag) 304 return(0); 305 306 /* 307 * make sure it is not the same file, protect the user 308 */ 309 if ((to_sb->st_dev==sb.st_dev)&&(to_sb->st_ino == sb.st_ino)) { 310 tty_warn(1, "Cannot link file %s to itself", to); 311 return(-1); 312 } 313 314 /* 315 * try to get rid of the file, based on the type 316 */ 317 if (S_ISDIR(sb.st_mode)) { 318 if (rmdir(from) < 0) { 319 syswarn(1, errno, "Cannot remove %s", from); 320 return(-1); 321 } 322 } else if (unlink(from) < 0) { 323 if (!ign) { 324 syswarn(1, errno, "Cannot remove %s", from); 325 return(-1); 326 } 327 return(1); 328 } 329 } 330 331 /* 332 * from file is gone (or did not exist), try to make the hard link. 333 * if it fails, check the path and try it again (if chk_path() says to 334 * try again) 335 */ 336 for (;;) { 337 if (link(to, from) == 0) 338 break; 339 oerrno = errno; 340 if (chk_path(from, to_sb->st_uid, to_sb->st_gid) == 0) 341 continue; 342 if (!ign) { 343 syswarn(1, oerrno, "Cannot link to %s from %s", to, 344 from); 345 return(-1); 346 } 347 return(1); 348 } 349 350 /* 351 * all right the link was made 352 */ 353 return(0); 354} 355 356/* 357 * node_creat() 358 * create an entry in the file system (other than a file or hard link). 359 * If successful, sets uid/gid modes and times as required. 360 * Return: 361 * 0 if ok, -1 otherwise 362 */ 363 364int 365node_creat(ARCHD *arcn) 366{ 367 int res; 368 int ign = 0; 369 int oerrno; 370 int pass = 0; 371 mode_t file_mode; 372 struct stat sb; 373 char target[MAXPATHLEN]; 374 char *nm = arcn->name; 375 int len; 376 377 /* 378 * create node based on type, if that fails try to unlink the node and 379 * try again. finally check the path and try again. As noted in the 380 * file and link creation routines, this method seems to exhibit the 381 * best performance in general use workloads. 382 */ 383 file_mode = arcn->sb.st_mode & FILEBITS; 384 385 for (;;) { 386 switch(arcn->type) { 387 case PAX_DIR: 388 /* 389 * If -h (or -L) was given in tar-mode, follow the 390 * potential symlink chain before trying to create the 391 * directory. 392 */ 393 if (strcmp(NM_TAR, argv0) == 0 && Lflag) { 394 while (lstat(nm, &sb) == 0 && 395 S_ISLNK(sb.st_mode)) { 396 len = readlink(nm, target, 397 sizeof target - 1); 398 if (len == -1) { 399 syswarn(0, errno, 400 "cannot follow symlink %s in chain for %s", 401 nm, arcn->name); 402 res = -1; 403 goto badlink; 404 } 405 target[len] = '\0'; 406 nm = target; 407 } 408 } 409 res = mkdir(nm, file_mode); 410 411badlink: 412 if (ign) 413 res = 0; 414 break; 415 case PAX_CHR: 416 file_mode |= S_IFCHR; 417 res = mknod(nm, file_mode, arcn->sb.st_rdev); 418 break; 419 case PAX_BLK: 420 file_mode |= S_IFBLK; 421 res = mknod(nm, file_mode, arcn->sb.st_rdev); 422 break; 423 case PAX_FIF: 424 res = mkfifo(nm, file_mode); 425 break; 426 case PAX_SCK: 427 /* 428 * Skip sockets, operation has no meaning under BSD 429 */ 430 tty_warn(0, 431 "%s skipped. Sockets cannot be copied or extracted", 432 nm); 433 return(-1); 434 case PAX_SLK: 435 res = symlink(arcn->ln_name, nm); 436 break; 437 case PAX_CTG: 438 case PAX_HLK: 439 case PAX_HRG: 440 case PAX_REG: 441 default: 442 /* 443 * we should never get here 444 */ 445 tty_warn(0, "%s has an unknown file type, skipping", 446 nm); 447 return(-1); 448 } 449 450 /* 451 * if we were able to create the node break out of the loop, 452 * otherwise try to unlink the node and try again. if that 453 * fails check the full path and try a final time. 454 */ 455 if (res == 0) 456 break; 457 458 /* 459 * we failed to make the node 460 */ 461 oerrno = errno; 462 if ((ign = unlnk_exist(nm, arcn->type)) < 0) 463 return(-1); 464 465 if (++pass <= 1) 466 continue; 467 468 if (nodirs || chk_path(nm,arcn->sb.st_uid,arcn->sb.st_gid) < 0) { 469 syswarn(1, oerrno, "Cannot create %s", nm); 470 return(-1); 471 } 472 } 473 474 /* 475 * we were able to create the node. set uid/gid, modes and times 476 */ 477 if (pids) 478 res = set_ids(nm, arcn->sb.st_uid, arcn->sb.st_gid); 479 else 480 res = 0; 481 482 /* 483 * IMPORTANT SECURITY NOTE: 484 * if not preserving mode or we cannot set uid/gid, then PROHIBIT any 485 * set uid/gid bits 486 */ 487 if (!pmode || res) 488 arcn->sb.st_mode &= ~(SETBITS); 489 if (pmode) 490 set_pmode(arcn->name, arcn->sb.st_mode); 491 492 if (arcn->type == PAX_DIR && strcmp(NM_CPIO, argv0) != 0) { 493 /* 494 * Dirs must be processed again at end of extract to set times 495 * and modes to agree with those stored in the archive. However 496 * to allow extract to continue, we may have to also set owner 497 * rights. This allows nodes in the archive that are children 498 * of this directory to be extracted without failure. Both time 499 * and modes will be fixed after the entire archive is read and 500 * before pax exits. 501 */ 502 if (access(nm, R_OK | W_OK | X_OK) < 0) { 503 if (lstat(nm, &sb) < 0) { 504 syswarn(0, errno,"Cannot access %s (stat)", 505 arcn->name); 506 set_pmode(nm,file_mode | S_IRWXU); 507 } else { 508 /* 509 * We have to add rights to the dir, so we make 510 * sure to restore the mode. The mode must be 511 * restored AS CREATED and not as stored if 512 * pmode is not set. 513 */ 514 set_pmode(nm, 515 ((sb.st_mode & FILEBITS) | S_IRWXU)); 516 if (!pmode) 517 arcn->sb.st_mode = sb.st_mode; 518 } 519 520 /* 521 * we have to force the mode to what was set here, 522 * since we changed it from the default as created. 523 */ 524 add_dir(nm, arcn->nlen, &(arcn->sb), 1); 525 } else if (pmode || patime || pmtime) 526 add_dir(nm, arcn->nlen, &(arcn->sb), 0); 527 } 528 529#if HAVE_LUTIMES 530 if (patime || pmtime) 531#else 532 if ((patime || pmtime) && arcn->type != PAX_SLK) 533#endif 534 set_ftime(arcn->name, arcn->sb.st_mtime, arcn->sb.st_atime, 0); 535 536#if HAVE_STRUCT_STAT_ST_FLAGS 537 if (pfflags && arcn->type != PAX_SLK) 538 set_chflags(arcn->name, arcn->sb.st_flags); 539#endif 540 return(0); 541} 542 543/* 544 * unlnk_exist() 545 * Remove node from file system with the specified name. We pass the type 546 * of the node that is going to replace it. When we try to create a 547 * directory and find that it already exists, we allow processing to 548 * continue as proper modes etc will always be set for it later on. 549 * Return: 550 * 0 is ok to proceed, no file with the specified name exists 551 * -1 we were unable to remove the node, or we should not remove it (-k) 552 * 1 we found a directory and we were going to create a directory. 553 */ 554 555int 556unlnk_exist(char *name, int type) 557{ 558 struct stat sb; 559 560 /* 561 * the file does not exist, or -k we are done 562 */ 563 if (lstat(name, &sb) < 0) 564 return(0); 565 if (kflag) 566 return(-1); 567 568 if (S_ISDIR(sb.st_mode)) { 569 /* 570 * try to remove a directory, if it fails and we were going to 571 * create a directory anyway, tell the caller (return a 1) 572 */ 573 if (rmdir(name) < 0) { 574 if (type == PAX_DIR) 575 return(1); 576 syswarn(1, errno, "Cannot remove directory %s", name); 577 return(-1); 578 } 579 return(0); 580 } 581 582 /* 583 * try to get rid of all non-directory type nodes 584 */ 585 if (unlink(name) < 0) { 586 (void)fflush(listf); 587 syswarn(1, errno, "Cannot unlink %s", name); 588 return(-1); 589 } 590 return(0); 591} 592 593/* 594 * chk_path() 595 * We were trying to create some kind of node in the file system and it 596 * failed. chk_path() makes sure the path up to the node exists and is 597 * writable. When we have to create a directory that is missing along the 598 * path somewhere, the directory we create will be set to the same 599 * uid/gid as the file has (when uid and gid are being preserved). 600 * NOTE: this routine is a real performance loss. It is only used as a 601 * last resort when trying to create entries in the file system. 602 * Return: 603 * -1 when it could find nothing it is allowed to fix. 604 * 0 otherwise 605 */ 606 607int 608chk_path( char *name, uid_t st_uid, gid_t st_gid) 609{ 610 char *spt = name; 611 struct stat sb; 612 int retval = -1; 613 614 /* 615 * watch out for paths with nodes stored directly in / (e.g. /bozo) 616 */ 617 if (*spt == '/') 618 ++spt; 619 620 for(;;) { 621 /* 622 * work forward from the first / and check each part of 623 * the path 624 */ 625 spt = strchr(spt, '/'); 626 if (spt == NULL) 627 break; 628 *spt = '\0'; 629 630 /* 631 * if it exists we assume it is a directory, it is not within 632 * the spec (at least it seems to read that way) to alter the 633 * file system for nodes NOT EXPLICITLY stored on the archive. 634 * If that assumption is changed, you would test the node here 635 * and figure out how to get rid of it (probably like some 636 * recursive unlink()) or fix up the directory permissions if 637 * required (do an access()). 638 */ 639 if (lstat(name, &sb) == 0) { 640 *(spt++) = '/'; 641 continue; 642 } 643 644 /* 645 * the path fails at this point, see if we can create the 646 * needed directory and continue on 647 */ 648 if (mkdir(name, S_IRWXU | S_IRWXG | S_IRWXO) < 0) { 649 *spt = '/'; 650 retval = -1; 651 break; 652 } 653 654 /* 655 * we were able to create the directory. We will tell the 656 * caller that we found something to fix, and it is ok to try 657 * and create the node again. 658 */ 659 retval = 0; 660 if (pids) 661 (void)set_ids(name, st_uid, st_gid); 662 663 /* 664 * make sure the user doesn't have some strange umask that 665 * causes this newly created directory to be unusable. We fix 666 * the modes and restore them back to the creation default at 667 * the end of pax 668 */ 669 if ((access(name, R_OK | W_OK | X_OK) < 0) && 670 (lstat(name, &sb) == 0)) { 671 set_pmode(name, ((sb.st_mode & FILEBITS) | S_IRWXU)); 672 add_dir(name, spt - name, &sb, 1); 673 } 674 *(spt++) = '/'; 675 continue; 676 } 677 return(retval); 678} 679 680/* 681 * set_ftime() 682 * Set the access time and modification time for a named file. If frc 683 * is non-zero we force these times to be set even if the user did not 684 * request access and/or modification time preservation (this is also 685 * used by -t to reset access times). 686 * When ign is zero, only those times the user has asked for are set, the 687 * other ones are left alone. We do not assume the un-documented feature 688 * of many utimes() implementations that consider a 0 time value as a do 689 * not set request. 690 */ 691 692void 693set_ftime(char *fnm, time_t mtime, time_t atime, int frc) 694{ 695 struct timeval tv[2]; 696 struct stat sb; 697 698 tv[0].tv_sec = (long)atime; 699 tv[0].tv_usec = 0; 700 tv[1].tv_sec = (long)mtime; 701 tv[1].tv_usec = 0; 702 if (!frc && (!patime || !pmtime)) { 703 /* 704 * if we are not forcing, only set those times the user wants 705 * set. We get the current values of the times if we need them. 706 */ 707 if (lstat(fnm, &sb) == 0) { 708#ifdef BSD4_4 709 if (!patime) 710 TIMESPEC_TO_TIMEVAL(&tv[0], &sb.st_atimespec); 711 if (!pmtime) 712 TIMESPEC_TO_TIMEVAL(&tv[1], &sb.st_mtimespec); 713#else 714 if (!patime) 715 tv[0].tv_sec = sb.st_atime; 716 if (!pmtime) 717 tv[1].tv_sec = sb.st_mtime; 718#endif 719 } else 720 syswarn(0, errno, "Cannot obtain file stats %s", fnm); 721 } 722 723 /* 724 * set the times 725 */ 726#if HAVE_LUTIMES 727 if (lutimes(fnm, tv)) 728#else 729 if (utimes(fnm, tv)) 730#endif 731 syswarn(1, errno, "Access/modification time set failed on: %s", 732 fnm); 733 return; 734} 735 736/* 737 * set_ids() 738 * set the uid and gid of a file system node 739 * Return: 740 * 0 when set, -1 on failure 741 */ 742 743int 744set_ids(char *fnm, uid_t uid, gid_t gid) 745{ 746 if (geteuid() == 0) 747 if (lchown(fnm, uid, gid)) { 748 (void)fflush(listf); 749 syswarn(1, errno, "Cannot set file uid/gid of %s", 750 fnm); 751 return(-1); 752 } 753 return(0); 754} 755 756/* 757 * set_pmode() 758 * Set file access mode 759 */ 760 761void 762set_pmode(char *fnm, mode_t mode) 763{ 764 mode &= ABITS; 765 if (lchmod(fnm, mode)) { 766 (void)fflush(listf); 767 syswarn(1, errno, "Cannot set permissions on %s", fnm); 768 } 769 return; 770} 771 772/* 773 * set_chflags() 774 * Set 4.4BSD file flags 775 */ 776void 777set_chflags(char *fnm, u_int32_t flags) 778{ 779 780#if 0 781 if (chflags(fnm, flags) < 0 && errno != EOPNOTSUPP) 782 syswarn(1, errno, "Cannot set file flags on %s", fnm); 783#endif 784 return; 785} 786 787/* 788 * file_write() 789 * Write/copy a file (during copy or archive extract). This routine knows 790 * how to copy files with lseek holes in it. (Which are read as file 791 * blocks containing all 0's but do not have any file blocks associated 792 * with the data). Typical examples of these are files created by dbm 793 * variants (.pag files). While the file size of these files are huge, the 794 * actual storage is quite small (the files are sparse). The problem is 795 * the holes read as all zeros so are probably stored on the archive that 796 * way (there is no way to determine if the file block is really a hole, 797 * we only know that a file block of all zero's can be a hole). 798 * At this writing, no major archive format knows how to archive files 799 * with holes. However, on extraction (or during copy, -rw) we have to 800 * deal with these files. Without detecting the holes, the files can 801 * consume a lot of file space if just written to disk. This replacement 802 * for write when passed the basic allocation size of a file system block, 803 * uses lseek whenever it detects the input data is all 0 within that 804 * file block. In more detail, the strategy is as follows: 805 * While the input is all zero keep doing an lseek. Keep track of when we 806 * pass over file block boundaries. Only write when we hit a non zero 807 * input. once we have written a file block, we continue to write it to 808 * the end (we stop looking at the input). When we reach the start of the 809 * next file block, start checking for zero blocks again. Working on file 810 * block boundaries significantly reduces the overhead when copying files 811 * that are NOT very sparse. This overhead (when compared to a write) is 812 * almost below the measurement resolution on many systems. Without it, 813 * files with holes cannot be safely copied. It does has a side effect as 814 * it can put holes into files that did not have them before, but that is 815 * not a problem since the file contents are unchanged (in fact it saves 816 * file space). (Except on paging files for diskless clients. But since we 817 * cannot determine one of those file from here, we ignore them). If this 818 * ever ends up on a system where CTG files are supported and the holes 819 * are not desired, just do a conditional test in those routines that 820 * call file_write() and have it call write() instead. BEFORE CLOSING THE 821 * FILE, make sure to call file_flush() when the last write finishes with 822 * an empty block. A lot of file systems will not create an lseek hole at 823 * the end. In this case we drop a single 0 at the end to force the 824 * trailing 0's in the file. 825 * ---Parameters--- 826 * rem: how many bytes left in this file system block 827 * isempt: have we written to the file block yet (is it empty) 828 * sz: basic file block allocation size 829 * cnt: number of bytes on this write 830 * str: buffer to write 831 * Return: 832 * number of bytes written, -1 on write (or lseek) error. 833 */ 834 835int 836file_write(int fd, char *str, int cnt, int *rem, int *isempt, int sz, 837 char *name) 838{ 839 char *pt; 840 char *end; 841 int wcnt; 842 char *st = str; 843 char **strp; 844 845 /* 846 * while we have data to process 847 */ 848 while (cnt) { 849 if (!*rem) { 850 /* 851 * We are now at the start of file system block again 852 * (or what we think one is...). start looking for 853 * empty blocks again 854 */ 855 *isempt = 1; 856 *rem = sz; 857 } 858 859 /* 860 * only examine up to the end of the current file block or 861 * remaining characters to write, whatever is smaller 862 */ 863 wcnt = MIN(cnt, *rem); 864 cnt -= wcnt; 865 *rem -= wcnt; 866 if (*isempt) { 867 /* 868 * have not written to this block yet, so we keep 869 * looking for zero's 870 */ 871 pt = st; 872 end = st + wcnt; 873 874 /* 875 * look for a zero filled buffer 876 */ 877 while ((pt < end) && (*pt == '\0')) 878 ++pt; 879 880 if (pt == end) { 881 /* 882 * skip, buf is empty so far 883 */ 884 if (fd > -1 && 885 lseek(fd, (off_t)wcnt, SEEK_CUR) < 0) { 886 syswarn(1, errno, "File seek on %s", 887 name); 888 return(-1); 889 } 890 st = pt; 891 continue; 892 } 893 /* 894 * drat, the buf is not zero filled 895 */ 896 *isempt = 0; 897 } 898 899 /* 900 * have non-zero data in this file system block, have to write 901 */ 902 switch (fd) { 903 case -1: 904 strp = &gnu_name_string; 905 break; 906 case -2: 907 strp = &gnu_link_string; 908 break; 909 default: 910 strp = NULL; 911 break; 912 } 913 if (strp) { 914 if (*strp) 915 err(1, "WARNING! Major Internal Error! GNU hack Failing!"); 916 *strp = malloc(wcnt + 1); 917 if (*strp == NULL) { 918 tty_warn(1, "Out of memory"); 919 return(-1); 920 } 921 strlcpy(*strp, st, wcnt); 922 break; 923 } else if (xwrite(fd, st, wcnt) != wcnt) { 924 syswarn(1, errno, "Failed write to file %s", name); 925 return(-1); 926 } 927 st += wcnt; 928 } 929 return(st - str); 930} 931 932/* 933 * file_flush() 934 * when the last file block in a file is zero, many file systems will not 935 * let us create a hole at the end. To get the last block with zeros, we 936 * write the last BYTE with a zero (back up one byte and write a zero). 937 */ 938 939void 940file_flush(int fd, char *fname, int isempt) 941{ 942 static char blnk[] = "\0"; 943 944 /* 945 * silly test, but make sure we are only called when the last block is 946 * filled with all zeros. 947 */ 948 if (!isempt) 949 return; 950 951 /* 952 * move back one byte and write a zero 953 */ 954 if (lseek(fd, (off_t)-1, SEEK_CUR) < 0) { 955 syswarn(1, errno, "Failed seek on file %s", fname); 956 return; 957 } 958 959 if (write_with_restart(fd, blnk, 1) < 0) 960 syswarn(1, errno, "Failed write to file %s", fname); 961 return; 962} 963 964/* 965 * rdfile_close() 966 * close a file we have been reading (to copy or archive). If we have to 967 * reset access time (tflag) do so (the times are stored in arcn). 968 */ 969 970void 971rdfile_close(ARCHD *arcn, int *fd) 972{ 973 /* 974 * make sure the file is open 975 */ 976 if (*fd < 0) 977 return; 978 979 (void)close(*fd); 980 *fd = -1; 981 if (!tflag) 982 return; 983 984 /* 985 * user wants last access time reset 986 */ 987 set_ftime(arcn->org_name, arcn->sb.st_mtime, arcn->sb.st_atime, 1); 988 return; 989} 990 991/* 992 * set_crc() 993 * read a file to calculate its crc. This is a real drag. Archive formats 994 * that have this, end up reading the file twice (we have to write the 995 * header WITH the crc before writing the file contents. Oh well... 996 * Return: 997 * 0 if was able to calculate the crc, -1 otherwise 998 */ 999 1000int 1001set_crc(ARCHD *arcn, int fd) 1002{ 1003 int i; 1004 int res; 1005 off_t cpcnt = 0L; 1006 u_long size; 1007 unsigned long crc = 0L; 1008 char tbuf[FILEBLK]; 1009 struct stat sb; 1010 1011 if (fd < 0) { 1012 /* 1013 * hmm, no fd, should never happen. well no crc then. 1014 */ 1015 arcn->crc = 0L; 1016 return(0); 1017 } 1018 1019 if ((size = (u_long)arcn->sb.st_blksize) > (u_long)sizeof(tbuf)) 1020 size = (u_long)sizeof(tbuf); 1021 1022 /* 1023 * read all the bytes we think that there are in the file. If the user 1024 * is trying to archive an active file, forget this file. 1025 */ 1026 for(;;) { 1027 if ((res = read(fd, tbuf, size)) <= 0) 1028 break; 1029 cpcnt += res; 1030 for (i = 0; i < res; ++i) 1031 crc += (tbuf[i] & 0xff); 1032 } 1033 1034 /* 1035 * safety check. we want to avoid archiving files that are active as 1036 * they can create inconsistant archive copies. 1037 */ 1038 if (cpcnt != arcn->sb.st_size) 1039 tty_warn(1, "File changed size %s", arcn->org_name); 1040 else if (fstat(fd, &sb) < 0) 1041 syswarn(1, errno, "Failed stat on %s", arcn->org_name); 1042 else if (arcn->sb.st_mtime != sb.st_mtime) 1043 tty_warn(1, "File %s was modified during read", arcn->org_name); 1044 else if (lseek(fd, (off_t)0L, SEEK_SET) < 0) 1045 syswarn(1, errno, "File rewind failed on: %s", arcn->org_name); 1046 else { 1047 arcn->crc = crc; 1048 return(0); 1049 } 1050 return(-1); 1051} 1052