file_subs.c revision 1.40
1/* $NetBSD: file_subs.c,v 1.40 2004/02/13 08:27:12 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 1.40 2004/02/13 08:27:12 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 } 99 100 for (;;) { 101 /* 102 * try to create the temporary file we use to restore the 103 * contents info. if this fails, keep checking all the nodes 104 * in the path until chk_path() finds that it cannot fix 105 * anything further. if that happens we just give up. 106 */ 107 (void)snprintf(arcn->tmp_name, arcn->nlen + 8, "%s.XXXXXX", 108 arcn->name); 109 fd = mkstemp(arcn->tmp_name); 110 if (fd >= 0) 111 break; 112 oerrno = errno; 113 if (nodirs || chk_path(arcn->name,arcn->sb.st_uid,arcn->sb.st_gid) < 0) { 114 (void)fflush(listf); 115 syswarn(1, oerrno, "Cannot create %s", arcn->tmp_name); 116 free(arcn->tmp_name); 117 arcn->tmp_name = NULL; 118 return(-1); 119 } 120 } 121 return(fd); 122} 123 124/* 125 * file_close() 126 * Close file descriptor to a file just created by pax. Sets modes, 127 * ownership and times as required. 128 * Return: 129 * 0 for success, -1 for failure 130 */ 131 132void 133file_close(ARCHD *arcn, int fd) 134{ 135 int res = 0; 136 137 if (fd < 0) 138 return; 139 if (close(fd) < 0) 140 syswarn(0, errno, "Cannot close file descriptor on %s", 141 arcn->tmp_name); 142 143 /* 144 * set owner/groups first as this may strip off mode bits we want 145 * then set file permission modes. Then set file access and 146 * modification times. 147 */ 148 if (pids) 149 res = set_ids(arcn->tmp_name, arcn->sb.st_uid, arcn->sb.st_gid); 150 151 /* 152 * IMPORTANT SECURITY NOTE: 153 * if not preserving mode or we cannot set uid/gid, then PROHIBIT 154 * set uid/gid bits but restore the file modes (since mkstemp doesn't). 155 */ 156 if (!pmode || res) 157 arcn->sb.st_mode &= ~(SETBITS); 158 if (pmode) 159 set_pmode(arcn->tmp_name, arcn->sb.st_mode); 160 else 161 set_pmode(arcn->tmp_name, arcn->sb.st_mode & FILEBITS); 162 if (patime || pmtime) 163 set_ftime(arcn->tmp_name, arcn->sb.st_mtime, arcn->sb.st_atime, 0); 164#if HAVE_STRUCT_STAT_ST_FLAGS 165 if (pfflags && arcn->type != PAX_SLK) 166 set_chflags(arcn->tmp_name, arcn->sb.st_flags); 167#endif 168 169 /* 170 * Finally, now the temp file is fully instantiated rename it to 171 * the desired file name. 172 */ 173 if (rename(arcn->tmp_name, arcn->name) < 0) { 174 syswarn(0, errno, "Cannot rename %s to %s", 175 arcn->tmp_name, arcn->name); 176 (void)unlink(arcn->tmp_name); 177 } 178 179 free(arcn->tmp_name); 180 arcn->tmp_name = NULL; 181} 182 183/* 184 * lnk_creat() 185 * Create a hard link to arcn->ln_name from arcn->name. arcn->ln_name 186 * must exist; 187 * Return: 188 * 0 if ok, -1 otherwise 189 */ 190 191int 192lnk_creat(ARCHD *arcn) 193{ 194 struct stat sb; 195 196 /* 197 * we may be running as root, so we have to be sure that link target 198 * is not a directory, so we lstat and check 199 */ 200 if (lstat(arcn->ln_name, &sb) < 0) { 201 syswarn(1, errno, "Cannot link to %s from %s", arcn->ln_name, 202 arcn->name); 203 return(-1); 204 } 205 206 if (S_ISDIR(sb.st_mode)) { 207 tty_warn(1, "A hard link to the directory %s is not allowed", 208 arcn->ln_name); 209 return(-1); 210 } 211 212 return(mk_link(arcn->ln_name, &sb, arcn->name, 0)); 213} 214 215/* 216 * cross_lnk() 217 * Create a hard link to arcn->org_name from arcn->name. Only used in copy 218 * with the -l flag. No warning or error if this does not succeed (we will 219 * then just create the file) 220 * Return: 221 * 1 if copy() should try to create this file node 222 * 0 if cross_lnk() ok, -1 for fatal flaw (like linking to self). 223 */ 224 225int 226cross_lnk(ARCHD *arcn) 227{ 228 /* 229 * try to make a link to original file (-l flag in copy mode). make 230 * sure we do not try to link to directories in case we are running as 231 * root (and it might succeed). 232 */ 233 if (arcn->type == PAX_DIR) 234 return(1); 235 return(mk_link(arcn->org_name, &(arcn->sb), arcn->name, 1)); 236} 237 238/* 239 * chk_same() 240 * In copy mode if we are not trying to make hard links between the src 241 * and destinations, make sure we are not going to overwrite ourselves by 242 * accident. This slows things down a little, but we have to protect all 243 * those people who make typing errors. 244 * Return: 245 * 1 the target does not exist, go ahead and copy 246 * 0 skip it file exists (-k) or may be the same as source file 247 */ 248 249int 250chk_same(ARCHD *arcn) 251{ 252 struct stat sb; 253 254 /* 255 * if file does not exist, return. if file exists and -k, skip it 256 * quietly 257 */ 258 if (lstat(arcn->name, &sb) < 0) 259 return(1); 260 if (kflag) 261 return(0); 262 263 /* 264 * better make sure the user does not have src == dest by mistake 265 */ 266 if ((arcn->sb.st_dev == sb.st_dev) && (arcn->sb.st_ino == sb.st_ino)) { 267 tty_warn(1, "Unable to copy %s, file would overwrite itself", 268 arcn->name); 269 return(0); 270 } 271 return(1); 272} 273 274/* 275 * mk_link() 276 * try to make a hard link between two files. if ign set, we do not 277 * complain. 278 * Return: 279 * 0 if successful (or we are done with this file but no error, such as 280 * finding the from file exists and the user has set -k). 281 * 1 when ign was set to indicates we could not make the link but we 282 * should try to copy/extract the file as that might work (and is an 283 * allowed option). -1 an error occurred. 284 */ 285 286static int 287mk_link(char *to, struct stat *to_sb, char *from, int ign) 288{ 289 struct stat sb; 290 int oerrno; 291 292 /* 293 * if from file exists, it has to be unlinked to make the link. If the 294 * file exists and -k is set, skip it quietly 295 */ 296 if (lstat(from, &sb) == 0) { 297 if (kflag) 298 return(0); 299 300 /* 301 * make sure it is not the same file, protect the user 302 */ 303 if ((to_sb->st_dev==sb.st_dev)&&(to_sb->st_ino == sb.st_ino)) { 304 tty_warn(1, "Cannot link file %s to itself", to); 305 return(-1); 306 } 307 308 /* 309 * try to get rid of the file, based on the type 310 */ 311 if (S_ISDIR(sb.st_mode)) { 312 if (rmdir(from) < 0) { 313 syswarn(1, errno, "Cannot remove %s", from); 314 return(-1); 315 } 316 } else if (unlink(from) < 0) { 317 if (!ign) { 318 syswarn(1, errno, "Cannot remove %s", from); 319 return(-1); 320 } 321 return(1); 322 } 323 } 324 325 /* 326 * from file is gone (or did not exist), try to make the hard link. 327 * if it fails, check the path and try it again (if chk_path() says to 328 * try again) 329 */ 330 for (;;) { 331 if (link(to, from) == 0) 332 break; 333 oerrno = errno; 334 if (chk_path(from, to_sb->st_uid, to_sb->st_gid) == 0) 335 continue; 336 if (!ign) { 337 syswarn(1, oerrno, "Cannot link to %s from %s", to, 338 from); 339 return(-1); 340 } 341 return(1); 342 } 343 344 /* 345 * all right the link was made 346 */ 347 return(0); 348} 349 350/* 351 * node_creat() 352 * create an entry in the file system (other than a file or hard link). 353 * If successful, sets uid/gid modes and times as required. 354 * Return: 355 * 0 if ok, -1 otherwise 356 */ 357 358int 359node_creat(ARCHD *arcn) 360{ 361 int res; 362 int ign = 0; 363 int oerrno; 364 int pass = 0; 365 mode_t file_mode; 366 struct stat sb; 367 char target[MAXPATHLEN]; 368 char *nm = arcn->name; 369 int len; 370 371 /* 372 * create node based on type, if that fails try to unlink the node and 373 * try again. finally check the path and try again. As noted in the 374 * file and link creation routines, this method seems to exhibit the 375 * best performance in general use workloads. 376 */ 377 file_mode = arcn->sb.st_mode & FILEBITS; 378 379 for (;;) { 380 switch(arcn->type) { 381 case PAX_DIR: 382 /* 383 * If -h (or -L) was given in tar-mode, follow the 384 * potential symlink chain before trying to create the 385 * directory. 386 */ 387 if (strcmp(NM_TAR, argv0) == 0 && Lflag) { 388 while (lstat(nm, &sb) == 0 && 389 S_ISLNK(sb.st_mode)) { 390 len = readlink(nm, target, 391 sizeof target - 1); 392 if (len == -1) { 393 syswarn(0, errno, 394 "cannot follow symlink %s in chain for %s", 395 nm, arcn->name); 396 res = -1; 397 goto badlink; 398 } 399 target[len] = '\0'; 400 nm = target; 401 } 402 } 403 res = mkdir(nm, file_mode); 404 405badlink: 406 if (ign) 407 res = 0; 408 break; 409 case PAX_CHR: 410 file_mode |= S_IFCHR; 411 res = mknod(nm, file_mode, arcn->sb.st_rdev); 412 break; 413 case PAX_BLK: 414 file_mode |= S_IFBLK; 415 res = mknod(nm, file_mode, arcn->sb.st_rdev); 416 break; 417 case PAX_FIF: 418 res = mkfifo(nm, file_mode); 419 break; 420 case PAX_SCK: 421 /* 422 * Skip sockets, operation has no meaning under BSD 423 */ 424 tty_warn(0, 425 "%s skipped. Sockets cannot be copied or extracted", 426 nm); 427 return(-1); 428 case PAX_SLK: 429 res = symlink(arcn->ln_name, nm); 430 break; 431 case PAX_CTG: 432 case PAX_HLK: 433 case PAX_HRG: 434 case PAX_REG: 435 default: 436 /* 437 * we should never get here 438 */ 439 tty_warn(0, "%s has an unknown file type, skipping", 440 nm); 441 return(-1); 442 } 443 444 /* 445 * if we were able to create the node break out of the loop, 446 * otherwise try to unlink the node and try again. if that 447 * fails check the full path and try a final time. 448 */ 449 if (res == 0) 450 break; 451 452 /* 453 * we failed to make the node 454 */ 455 oerrno = errno; 456 if ((ign = unlnk_exist(nm, arcn->type)) < 0) 457 return(-1); 458 459 if (++pass <= 1) 460 continue; 461 462 if (nodirs || chk_path(nm,arcn->sb.st_uid,arcn->sb.st_gid) < 0) { 463 syswarn(1, oerrno, "Cannot create %s", nm); 464 return(-1); 465 } 466 } 467 468 /* 469 * we were able to create the node. set uid/gid, modes and times 470 */ 471 if (pids) 472 res = set_ids(nm, arcn->sb.st_uid, arcn->sb.st_gid); 473 else 474 res = 0; 475 476 /* 477 * IMPORTANT SECURITY NOTE: 478 * if not preserving mode or we cannot set uid/gid, then PROHIBIT any 479 * set uid/gid bits 480 */ 481 if (!pmode || res) 482 arcn->sb.st_mode &= ~(SETBITS); 483 if (pmode) 484 set_pmode(arcn->name, arcn->sb.st_mode); 485 486 if (arcn->type == PAX_DIR && strcmp(NM_CPIO, argv0) != 0) { 487 /* 488 * Dirs must be processed again at end of extract to set times 489 * and modes to agree with those stored in the archive. However 490 * to allow extract to continue, we may have to also set owner 491 * rights. This allows nodes in the archive that are children 492 * of this directory to be extracted without failure. Both time 493 * and modes will be fixed after the entire archive is read and 494 * before pax exits. 495 */ 496 if (access(nm, R_OK | W_OK | X_OK) < 0) { 497 if (lstat(nm, &sb) < 0) { 498 syswarn(0, errno,"Cannot access %s (stat)", 499 arcn->name); 500 set_pmode(nm,file_mode | S_IRWXU); 501 } else { 502 /* 503 * We have to add rights to the dir, so we make 504 * sure to restore the mode. The mode must be 505 * restored AS CREATED and not as stored if 506 * pmode is not set. 507 */ 508 set_pmode(nm, 509 ((sb.st_mode & FILEBITS) | S_IRWXU)); 510 if (!pmode) 511 arcn->sb.st_mode = sb.st_mode; 512 } 513 514 /* 515 * we have to force the mode to what was set here, 516 * since we changed it from the default as created. 517 */ 518 add_dir(nm, arcn->nlen, &(arcn->sb), 1); 519 } else if (pmode || patime || pmtime) 520 add_dir(nm, arcn->nlen, &(arcn->sb), 0); 521 } 522 523#if HAVE_LUTIMES 524 if (patime || pmtime) 525#else 526 if ((patime || pmtime) && arcn->type != PAX_SLK) 527#endif 528 set_ftime(arcn->name, arcn->sb.st_mtime, arcn->sb.st_atime, 0); 529 530#if HAVE_STRUCT_STAT_ST_FLAGS 531 if (pfflags && arcn->type != PAX_SLK) 532 set_chflags(arcn->name, arcn->sb.st_flags); 533#endif 534 return(0); 535} 536 537/* 538 * unlnk_exist() 539 * Remove node from file system with the specified name. We pass the type 540 * of the node that is going to replace it. When we try to create a 541 * directory and find that it already exists, we allow processing to 542 * continue as proper modes etc will always be set for it later on. 543 * Return: 544 * 0 is ok to proceed, no file with the specified name exists 545 * -1 we were unable to remove the node, or we should not remove it (-k) 546 * 1 we found a directory and we were going to create a directory. 547 */ 548 549int 550unlnk_exist(char *name, int type) 551{ 552 struct stat sb; 553 554 /* 555 * the file does not exist, or -k we are done 556 */ 557 if (lstat(name, &sb) < 0) 558 return(0); 559 if (kflag) 560 return(-1); 561 562 if (S_ISDIR(sb.st_mode)) { 563 /* 564 * try to remove a directory, if it fails and we were going to 565 * create a directory anyway, tell the caller (return a 1) 566 */ 567 if (rmdir(name) < 0) { 568 if (type == PAX_DIR) 569 return(1); 570 syswarn(1, errno, "Cannot remove directory %s", name); 571 return(-1); 572 } 573 return(0); 574 } 575 576 /* 577 * try to get rid of all non-directory type nodes 578 */ 579 if (unlink(name) < 0) { 580 (void)fflush(listf); 581 syswarn(1, errno, "Cannot unlink %s", name); 582 return(-1); 583 } 584 return(0); 585} 586 587/* 588 * chk_path() 589 * We were trying to create some kind of node in the file system and it 590 * failed. chk_path() makes sure the path up to the node exists and is 591 * writable. When we have to create a directory that is missing along the 592 * path somewhere, the directory we create will be set to the same 593 * uid/gid as the file has (when uid and gid are being preserved). 594 * NOTE: this routine is a real performance loss. It is only used as a 595 * last resort when trying to create entries in the file system. 596 * Return: 597 * -1 when it could find nothing it is allowed to fix. 598 * 0 otherwise 599 */ 600 601int 602chk_path( char *name, uid_t st_uid, gid_t st_gid) 603{ 604 char *spt = name; 605 struct stat sb; 606 int retval = -1; 607 608 /* 609 * watch out for paths with nodes stored directly in / (e.g. /bozo) 610 */ 611 if (*spt == '/') 612 ++spt; 613 614 for(;;) { 615 /* 616 * work forward from the first / and check each part of 617 * the path 618 */ 619 spt = strchr(spt, '/'); 620 if (spt == NULL) 621 break; 622 *spt = '\0'; 623 624 /* 625 * if it exists we assume it is a directory, it is not within 626 * the spec (at least it seems to read that way) to alter the 627 * file system for nodes NOT EXPLICITLY stored on the archive. 628 * If that assumption is changed, you would test the node here 629 * and figure out how to get rid of it (probably like some 630 * recursive unlink()) or fix up the directory permissions if 631 * required (do an access()). 632 */ 633 if (lstat(name, &sb) == 0) { 634 *(spt++) = '/'; 635 continue; 636 } 637 638 /* 639 * the path fails at this point, see if we can create the 640 * needed directory and continue on 641 */ 642 if (mkdir(name, S_IRWXU | S_IRWXG | S_IRWXO) < 0) { 643 *spt = '/'; 644 retval = -1; 645 break; 646 } 647 648 /* 649 * we were able to create the directory. We will tell the 650 * caller that we found something to fix, and it is ok to try 651 * and create the node again. 652 */ 653 retval = 0; 654 if (pids) 655 (void)set_ids(name, st_uid, st_gid); 656 657 /* 658 * make sure the user doesn't have some strange umask that 659 * causes this newly created directory to be unusable. We fix 660 * the modes and restore them back to the creation default at 661 * the end of pax 662 */ 663 if ((access(name, R_OK | W_OK | X_OK) < 0) && 664 (lstat(name, &sb) == 0)) { 665 set_pmode(name, ((sb.st_mode & FILEBITS) | S_IRWXU)); 666 add_dir(name, spt - name, &sb, 1); 667 } 668 *(spt++) = '/'; 669 continue; 670 } 671 return(retval); 672} 673 674/* 675 * set_ftime() 676 * Set the access time and modification time for a named file. If frc 677 * is non-zero we force these times to be set even if the user did not 678 * request access and/or modification time preservation (this is also 679 * used by -t to reset access times). 680 * When ign is zero, only those times the user has asked for are set, the 681 * other ones are left alone. We do not assume the un-documented feature 682 * of many utimes() implementations that consider a 0 time value as a do 683 * not set request. 684 */ 685 686void 687set_ftime(char *fnm, time_t mtime, time_t atime, int frc) 688{ 689 struct timeval tv[2]; 690 struct stat sb; 691 692 tv[0].tv_sec = (long)atime; 693 tv[0].tv_usec = 0; 694 tv[1].tv_sec = (long)mtime; 695 tv[1].tv_usec = 0; 696 if (!frc && (!patime || !pmtime)) { 697 /* 698 * if we are not forcing, only set those times the user wants 699 * set. We get the current values of the times if we need them. 700 */ 701 if (lstat(fnm, &sb) == 0) { 702#ifdef BSD4_4 703 if (!patime) 704 TIMESPEC_TO_TIMEVAL(&tv[0], &sb.st_atimespec); 705 if (!pmtime) 706 TIMESPEC_TO_TIMEVAL(&tv[1], &sb.st_mtimespec); 707#else 708 if (!patime) 709 tv[0].tv_sec = sb.st_atime; 710 if (!pmtime) 711 tv[1].tv_sec = sb.st_mtime; 712#endif 713 } else 714 syswarn(0, errno, "Cannot obtain file stats %s", fnm); 715 } 716 717 /* 718 * set the times 719 */ 720#if HAVE_LUTIMES 721 if (lutimes(fnm, tv)) 722#else 723 if (utimes(fnm, tv)) 724#endif 725 syswarn(1, errno, "Access/modification time set failed on: %s", 726 fnm); 727 return; 728} 729 730/* 731 * set_ids() 732 * set the uid and gid of a file system node 733 * Return: 734 * 0 when set, -1 on failure 735 */ 736 737int 738set_ids(char *fnm, uid_t uid, gid_t gid) 739{ 740 if (geteuid() == 0) 741 if (lchown(fnm, uid, gid)) { 742 (void)fflush(listf); 743 syswarn(1, errno, "Cannot set file uid/gid of %s", 744 fnm); 745 return(-1); 746 } 747 return(0); 748} 749 750/* 751 * set_pmode() 752 * Set file access mode 753 */ 754 755void 756set_pmode(char *fnm, mode_t mode) 757{ 758 mode &= ABITS; 759 if (lchmod(fnm, mode)) { 760 (void)fflush(listf); 761 syswarn(1, errno, "Cannot set permissions on %s", fnm); 762 } 763 return; 764} 765 766/* 767 * set_chflags() 768 * Set 4.4BSD file flags 769 */ 770void 771set_chflags(char *fnm, u_int32_t flags) 772{ 773 774#if 0 775 if (chflags(fnm, flags) < 0 && errno != EOPNOTSUPP) 776 syswarn(1, errno, "Cannot set file flags on %s", fnm); 777#endif 778 return; 779} 780 781/* 782 * file_write() 783 * Write/copy a file (during copy or archive extract). This routine knows 784 * how to copy files with lseek holes in it. (Which are read as file 785 * blocks containing all 0's but do not have any file blocks associated 786 * with the data). Typical examples of these are files created by dbm 787 * variants (.pag files). While the file size of these files are huge, the 788 * actual storage is quite small (the files are sparse). The problem is 789 * the holes read as all zeros so are probably stored on the archive that 790 * way (there is no way to determine if the file block is really a hole, 791 * we only know that a file block of all zero's can be a hole). 792 * At this writing, no major archive format knows how to archive files 793 * with holes. However, on extraction (or during copy, -rw) we have to 794 * deal with these files. Without detecting the holes, the files can 795 * consume a lot of file space if just written to disk. This replacement 796 * for write when passed the basic allocation size of a file system block, 797 * uses lseek whenever it detects the input data is all 0 within that 798 * file block. In more detail, the strategy is as follows: 799 * While the input is all zero keep doing an lseek. Keep track of when we 800 * pass over file block boundaries. Only write when we hit a non zero 801 * input. once we have written a file block, we continue to write it to 802 * the end (we stop looking at the input). When we reach the start of the 803 * next file block, start checking for zero blocks again. Working on file 804 * block boundaries significantly reduces the overhead when copying files 805 * that are NOT very sparse. This overhead (when compared to a write) is 806 * almost below the measurement resolution on many systems. Without it, 807 * files with holes cannot be safely copied. It does has a side effect as 808 * it can put holes into files that did not have them before, but that is 809 * not a problem since the file contents are unchanged (in fact it saves 810 * file space). (Except on paging files for diskless clients. But since we 811 * cannot determine one of those file from here, we ignore them). If this 812 * ever ends up on a system where CTG files are supported and the holes 813 * are not desired, just do a conditional test in those routines that 814 * call file_write() and have it call write() instead. BEFORE CLOSING THE 815 * FILE, make sure to call file_flush() when the last write finishes with 816 * an empty block. A lot of file systems will not create an lseek hole at 817 * the end. In this case we drop a single 0 at the end to force the 818 * trailing 0's in the file. 819 * ---Parameters--- 820 * rem: how many bytes left in this file system block 821 * isempt: have we written to the file block yet (is it empty) 822 * sz: basic file block allocation size 823 * cnt: number of bytes on this write 824 * str: buffer to write 825 * Return: 826 * number of bytes written, -1 on write (or lseek) error. 827 */ 828 829int 830file_write(int fd, char *str, int cnt, int *rem, int *isempt, int sz, 831 char *name) 832{ 833 char *pt; 834 char *end; 835 int wcnt; 836 char *st = str; 837 char **strp; 838 839 /* 840 * while we have data to process 841 */ 842 while (cnt) { 843 if (!*rem) { 844 /* 845 * We are now at the start of file system block again 846 * (or what we think one is...). start looking for 847 * empty blocks again 848 */ 849 *isempt = 1; 850 *rem = sz; 851 } 852 853 /* 854 * only examine up to the end of the current file block or 855 * remaining characters to write, whatever is smaller 856 */ 857 wcnt = MIN(cnt, *rem); 858 cnt -= wcnt; 859 *rem -= wcnt; 860 if (*isempt) { 861 /* 862 * have not written to this block yet, so we keep 863 * looking for zero's 864 */ 865 pt = st; 866 end = st + wcnt; 867 868 /* 869 * look for a zero filled buffer 870 */ 871 while ((pt < end) && (*pt == '\0')) 872 ++pt; 873 874 if (pt == end) { 875 /* 876 * skip, buf is empty so far 877 */ 878 if (fd > -1 && 879 lseek(fd, (off_t)wcnt, SEEK_CUR) < 0) { 880 syswarn(1, errno, "File seek on %s", 881 name); 882 return(-1); 883 } 884 st = pt; 885 continue; 886 } 887 /* 888 * drat, the buf is not zero filled 889 */ 890 *isempt = 0; 891 } 892 893 /* 894 * have non-zero data in this file system block, have to write 895 */ 896 switch (fd) { 897 case -1: 898 strp = &gnu_name_string; 899 break; 900 case -2: 901 strp = &gnu_link_string; 902 break; 903 default: 904 strp = NULL; 905 break; 906 } 907 if (strp) { 908 if (*strp) 909 err(1, "WARNING! Major Internal Error! GNU hack Failing!"); 910 *strp = malloc(wcnt + 1); 911 if (*strp == NULL) { 912 tty_warn(1, "Out of memory"); 913 return(-1); 914 } 915 strlcpy(*strp, st, wcnt); 916 break; 917 } else if (xwrite(fd, st, wcnt) != wcnt) { 918 syswarn(1, errno, "Failed write to file %s", name); 919 return(-1); 920 } 921 st += wcnt; 922 } 923 return(st - str); 924} 925 926/* 927 * file_flush() 928 * when the last file block in a file is zero, many file systems will not 929 * let us create a hole at the end. To get the last block with zeros, we 930 * write the last BYTE with a zero (back up one byte and write a zero). 931 */ 932 933void 934file_flush(int fd, char *fname, int isempt) 935{ 936 static char blnk[] = "\0"; 937 938 /* 939 * silly test, but make sure we are only called when the last block is 940 * filled with all zeros. 941 */ 942 if (!isempt) 943 return; 944 945 /* 946 * move back one byte and write a zero 947 */ 948 if (lseek(fd, (off_t)-1, SEEK_CUR) < 0) { 949 syswarn(1, errno, "Failed seek on file %s", fname); 950 return; 951 } 952 953 if (write_with_restart(fd, blnk, 1) < 0) 954 syswarn(1, errno, "Failed write to file %s", fname); 955 return; 956} 957 958/* 959 * rdfile_close() 960 * close a file we have been reading (to copy or archive). If we have to 961 * reset access time (tflag) do so (the times are stored in arcn). 962 */ 963 964void 965rdfile_close(ARCHD *arcn, int *fd) 966{ 967 /* 968 * make sure the file is open 969 */ 970 if (*fd < 0) 971 return; 972 973 (void)close(*fd); 974 *fd = -1; 975 if (!tflag) 976 return; 977 978 /* 979 * user wants last access time reset 980 */ 981 set_ftime(arcn->org_name, arcn->sb.st_mtime, arcn->sb.st_atime, 1); 982 return; 983} 984 985/* 986 * set_crc() 987 * read a file to calculate its crc. This is a real drag. Archive formats 988 * that have this, end up reading the file twice (we have to write the 989 * header WITH the crc before writing the file contents. Oh well... 990 * Return: 991 * 0 if was able to calculate the crc, -1 otherwise 992 */ 993 994int 995set_crc(ARCHD *arcn, int fd) 996{ 997 int i; 998 int res; 999 off_t cpcnt = 0L; 1000 u_long size; 1001 unsigned long crc = 0L; 1002 char tbuf[FILEBLK]; 1003 struct stat sb; 1004 1005 if (fd < 0) { 1006 /* 1007 * hmm, no fd, should never happen. well no crc then. 1008 */ 1009 arcn->crc = 0L; 1010 return(0); 1011 } 1012 1013 if ((size = (u_long)arcn->sb.st_blksize) > (u_long)sizeof(tbuf)) 1014 size = (u_long)sizeof(tbuf); 1015 1016 /* 1017 * read all the bytes we think that there are in the file. If the user 1018 * is trying to archive an active file, forget this file. 1019 */ 1020 for(;;) { 1021 if ((res = read(fd, tbuf, size)) <= 0) 1022 break; 1023 cpcnt += res; 1024 for (i = 0; i < res; ++i) 1025 crc += (tbuf[i] & 0xff); 1026 } 1027 1028 /* 1029 * safety check. we want to avoid archiving files that are active as 1030 * they can create inconsistant archive copies. 1031 */ 1032 if (cpcnt != arcn->sb.st_size) 1033 tty_warn(1, "File changed size %s", arcn->org_name); 1034 else if (fstat(fd, &sb) < 0) 1035 syswarn(1, errno, "Failed stat on %s", arcn->org_name); 1036 else if (arcn->sb.st_mtime != sb.st_mtime) 1037 tty_warn(1, "File %s was modified during read", arcn->org_name); 1038 else if (lseek(fd, (off_t)0L, SEEK_SET) < 0) 1039 syswarn(1, errno, "File rewind failed on: %s", arcn->org_name); 1040 else { 1041 arcn->crc = crc; 1042 return(0); 1043 } 1044 return(-1); 1045} 1046