file_subs.c revision 1.35
1/* $NetBSD: file_subs.c,v 1.35 2003/10/13 07:41:22 agc 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#include <sys/cdefs.h> 37#if defined(__RCSID) && !defined(lint) 38#if 0 39static char sccsid[] = "@(#)file_subs.c 8.1 (Berkeley) 5/31/93"; 40#else 41__RCSID("$NetBSD: file_subs.c,v 1.35 2003/10/13 07:41:22 agc Exp $"); 42#endif 43#endif /* not lint */ 44 45#include <sys/types.h> 46#include <sys/time.h> 47#include <sys/stat.h> 48#include <unistd.h> 49#include <sys/param.h> 50#include <fcntl.h> 51#include <string.h> 52#include <stdio.h> 53#include <ctype.h> 54#include <errno.h> 55#include <sys/uio.h> 56#include <stdlib.h> 57#include "pax.h" 58#include "extern.h" 59#include "options.h" 60 61static int 62mk_link(char *,struct stat *,char *, int); 63 64/* 65 * routines that deal with file operations such as: creating, removing; 66 * and setting access modes, uid/gid and times of files 67 */ 68 69#define FILEBITS (S_ISVTX | S_IRWXU | S_IRWXG | S_IRWXO) 70#define SETBITS (S_ISUID | S_ISGID) 71#define ABITS (FILEBITS | SETBITS) 72 73/* 74 * file_creat() 75 * Create and open a file. 76 * Return: 77 * file descriptor or -1 for failure 78 */ 79 80int 81file_creat(ARCHD *arcn) 82{ 83 int fd = -1; 84 mode_t file_mode; 85 int oerrno; 86 87 /* 88 * assume file doesn't exist, so just try to create it, most times this 89 * works. We have to take special handling when the file does exist. To 90 * detect this, we use O_EXCL. For example when trying to create a 91 * file and a character device or fifo exists with the same name, we 92 * can accidently open the device by mistake (or block waiting to open) 93 * If we find that the open has failed, then spend the effort to 94 * figure out why. This strategy was found to have better average 95 * performance in common use than checking the file (and the path) 96 * first with lstat. 97 */ 98 file_mode = arcn->sb.st_mode & FILEBITS; 99 if ((fd = open(arcn->name, O_WRONLY | O_CREAT | O_TRUNC | O_EXCL, 100 file_mode)) >= 0) 101 return(fd); 102 103 /* 104 * the file seems to exist. First we try to get rid of it (found to be 105 * the second most common failure when traced). If this fails, only 106 * then we go to the expense to check and create the path to the file 107 */ 108 if (unlnk_exist(arcn->name, arcn->type) != 0) 109 return(-1); 110 111 for (;;) { 112 /* 113 * try to open it again, if this fails, check all the nodes in 114 * the path and give it a final try. if chk_path() finds that 115 * it cannot fix anything, we will skip the last attempt 116 */ 117 if ((fd = open(arcn->name, O_WRONLY | O_CREAT | O_TRUNC, 118 file_mode)) >= 0) 119 break; 120 oerrno = errno; 121 if (nodirs || chk_path(arcn->name,arcn->sb.st_uid,arcn->sb.st_gid) < 0) { 122 (void)fflush(listf); 123 syswarn(1, oerrno, "Cannot create %s", arcn->name); 124 return(-1); 125 } 126 } 127 return(fd); 128} 129 130/* 131 * file_close() 132 * Close file descriptor to a file just created by pax. Sets modes, 133 * ownership and times as required. 134 * Return: 135 * 0 for success, -1 for failure 136 */ 137 138void 139file_close(ARCHD *arcn, int fd) 140{ 141 int res = 0; 142 143 if (fd < 0) 144 return; 145 if (close(fd) < 0) 146 syswarn(0, errno, "Cannot close file descriptor on %s", 147 arcn->name); 148 149 /* 150 * set owner/groups first as this may strip off mode bits we want 151 * then set file permission modes. Then set file access and 152 * modification times. 153 */ 154 if (pids) 155 res = set_ids(arcn->name, arcn->sb.st_uid, arcn->sb.st_gid); 156 157 /* 158 * IMPORTANT SECURITY NOTE: 159 * if not preserving mode or we cannot set uid/gid, then PROHIBIT 160 * set uid/gid bits 161 */ 162 if (!pmode || res) 163 arcn->sb.st_mode &= ~(SETBITS); 164 if (pmode) 165 set_pmode(arcn->name, arcn->sb.st_mode); 166 if (patime || pmtime) 167 set_ftime(arcn->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->name, arcn->sb.st_flags); 171#endif 172} 173 174/* 175 * lnk_creat() 176 * Create a hard link to arcn->ln_name from arcn->name. arcn->ln_name 177 * must exist; 178 * Return: 179 * 0 if ok, -1 otherwise 180 */ 181 182int 183lnk_creat(ARCHD *arcn) 184{ 185 struct stat sb; 186 187 /* 188 * we may be running as root, so we have to be sure that link target 189 * is not a directory, so we lstat and check 190 */ 191 if (lstat(arcn->ln_name, &sb) < 0) { 192 syswarn(1, errno, "Cannot link to %s from %s", arcn->ln_name, 193 arcn->name); 194 return(-1); 195 } 196 197 if (S_ISDIR(sb.st_mode)) { 198 tty_warn(1, "A hard link to the directory %s is not allowed", 199 arcn->ln_name); 200 return(-1); 201 } 202 203 return(mk_link(arcn->ln_name, &sb, arcn->name, 0)); 204} 205 206/* 207 * cross_lnk() 208 * Create a hard link to arcn->org_name from arcn->name. Only used in copy 209 * with the -l flag. No warning or error if this does not succeed (we will 210 * then just create the file) 211 * Return: 212 * 1 if copy() should try to create this file node 213 * 0 if cross_lnk() ok, -1 for fatal flaw (like linking to self). 214 */ 215 216int 217cross_lnk(ARCHD *arcn) 218{ 219 /* 220 * try to make a link to original file (-l flag in copy mode). make 221 * sure we do not try to link to directories in case we are running as 222 * root (and it might succeed). 223 */ 224 if (arcn->type == PAX_DIR) 225 return(1); 226 return(mk_link(arcn->org_name, &(arcn->sb), arcn->name, 1)); 227} 228 229/* 230 * chk_same() 231 * In copy mode if we are not trying to make hard links between the src 232 * and destinations, make sure we are not going to overwrite ourselves by 233 * accident. This slows things down a little, but we have to protect all 234 * those people who make typing errors. 235 * Return: 236 * 1 the target does not exist, go ahead and copy 237 * 0 skip it file exists (-k) or may be the same as source file 238 */ 239 240int 241chk_same(ARCHD *arcn) 242{ 243 struct stat sb; 244 245 /* 246 * if file does not exist, return. if file exists and -k, skip it 247 * quietly 248 */ 249 if (lstat(arcn->name, &sb) < 0) 250 return(1); 251 if (kflag) 252 return(0); 253 254 /* 255 * better make sure the user does not have src == dest by mistake 256 */ 257 if ((arcn->sb.st_dev == sb.st_dev) && (arcn->sb.st_ino == sb.st_ino)) { 258 tty_warn(1, "Unable to copy %s, file would overwrite itself", 259 arcn->name); 260 return(0); 261 } 262 return(1); 263} 264 265/* 266 * mk_link() 267 * try to make a hard link between two files. if ign set, we do not 268 * complain. 269 * Return: 270 * 0 if successful (or we are done with this file but no error, such as 271 * finding the from file exists and the user has set -k). 272 * 1 when ign was set to indicates we could not make the link but we 273 * should try to copy/extract the file as that might work (and is an 274 * allowed option). -1 an error occurred. 275 */ 276 277static int 278mk_link(char *to, struct stat *to_sb, char *from, int ign) 279{ 280 struct stat sb; 281 int oerrno; 282 283 /* 284 * if from file exists, it has to be unlinked to make the link. If the 285 * file exists and -k is set, skip it quietly 286 */ 287 if (lstat(from, &sb) == 0) { 288 if (kflag) 289 return(0); 290 291 /* 292 * make sure it is not the same file, protect the user 293 */ 294 if ((to_sb->st_dev==sb.st_dev)&&(to_sb->st_ino == sb.st_ino)) { 295 tty_warn(1, "Cannot link file %s to itself", to); 296 return(-1); 297 } 298 299 /* 300 * try to get rid of the file, based on the type 301 */ 302 if (S_ISDIR(sb.st_mode)) { 303 if (rmdir(from) < 0) { 304 syswarn(1, errno, "Cannot remove %s", from); 305 return(-1); 306 } 307 } else if (unlink(from) < 0) { 308 if (!ign) { 309 syswarn(1, errno, "Cannot remove %s", from); 310 return(-1); 311 } 312 return(1); 313 } 314 } 315 316 /* 317 * from file is gone (or did not exist), try to make the hard link. 318 * if it fails, check the path and try it again (if chk_path() says to 319 * try again) 320 */ 321 for (;;) { 322 if (link(to, from) == 0) 323 break; 324 oerrno = errno; 325 if (chk_path(from, to_sb->st_uid, to_sb->st_gid) == 0) 326 continue; 327 if (!ign) { 328 syswarn(1, oerrno, "Cannot link to %s from %s", to, 329 from); 330 return(-1); 331 } 332 return(1); 333 } 334 335 /* 336 * all right the link was made 337 */ 338 return(0); 339} 340 341/* 342 * node_creat() 343 * create an entry in the file system (other than a file or hard link). 344 * If successful, sets uid/gid modes and times as required. 345 * Return: 346 * 0 if ok, -1 otherwise 347 */ 348 349int 350node_creat(ARCHD *arcn) 351{ 352 int res; 353 int ign = 0; 354 int oerrno; 355 int pass = 0; 356 mode_t file_mode; 357 struct stat sb; 358 char target[MAXPATHLEN]; 359 char *nm = arcn->name; 360 int len; 361 362 /* 363 * create node based on type, if that fails try to unlink the node and 364 * try again. finally check the path and try again. As noted in the 365 * file and link creation routines, this method seems to exhibit the 366 * best performance in general use workloads. 367 */ 368 file_mode = arcn->sb.st_mode & FILEBITS; 369 370 for (;;) { 371 switch(arcn->type) { 372 case PAX_DIR: 373 /* 374 * If -h (or -L) was given in tar-mode, follow the 375 * potential symlink chain before trying to create the 376 * directory. 377 */ 378 if (strcmp(NM_TAR, argv0) == 0 && Lflag) { 379 while (lstat(nm, &sb) == 0 && 380 S_ISLNK(sb.st_mode)) { 381 len = readlink(nm, target, 382 sizeof target - 1); 383 if (len == -1) { 384 syswarn(0, errno, 385 "cannot follow symlink %s in chain for %s", 386 nm, arcn->name); 387 res = -1; 388 goto badlink; 389 } 390 target[len] = '\0'; 391 nm = target; 392 } 393 } 394 res = mkdir(nm, file_mode); 395 396badlink: 397 if (ign) 398 res = 0; 399 break; 400 case PAX_CHR: 401 file_mode |= S_IFCHR; 402 res = mknod(nm, file_mode, arcn->sb.st_rdev); 403 break; 404 case PAX_BLK: 405 file_mode |= S_IFBLK; 406 res = mknod(nm, file_mode, arcn->sb.st_rdev); 407 break; 408 case PAX_FIF: 409 res = mkfifo(nm, file_mode); 410 break; 411 case PAX_SCK: 412 /* 413 * Skip sockets, operation has no meaning under BSD 414 */ 415 tty_warn(0, 416 "%s skipped. Sockets cannot be copied or extracted", 417 nm); 418 return(-1); 419 case PAX_SLK: 420 res = symlink(arcn->ln_name, nm); 421 break; 422 case PAX_CTG: 423 case PAX_HLK: 424 case PAX_HRG: 425 case PAX_REG: 426 default: 427 /* 428 * we should never get here 429 */ 430 tty_warn(0, "%s has an unknown file type, skipping", 431 nm); 432 return(-1); 433 } 434 435 /* 436 * if we were able to create the node break out of the loop, 437 * otherwise try to unlink the node and try again. if that 438 * fails check the full path and try a final time. 439 */ 440 if (res == 0) 441 break; 442 443 /* 444 * we failed to make the node 445 */ 446 oerrno = errno; 447 if ((ign = unlnk_exist(nm, arcn->type)) < 0) 448 return(-1); 449 450 if (++pass <= 1) 451 continue; 452 453 if (nodirs || chk_path(nm,arcn->sb.st_uid,arcn->sb.st_gid) < 0) { 454 syswarn(1, oerrno, "Cannot create %s", nm); 455 return(-1); 456 } 457 } 458 459 /* 460 * we were able to create the node. set uid/gid, modes and times 461 */ 462 if (pids) 463 res = set_ids(nm, arcn->sb.st_uid, arcn->sb.st_gid); 464 else 465 res = 0; 466 467 /* 468 * IMPORTANT SECURITY NOTE: 469 * if not preserving mode or we cannot set uid/gid, then PROHIBIT any 470 * set uid/gid bits 471 */ 472 if (!pmode || res) 473 arcn->sb.st_mode &= ~(SETBITS); 474 if (pmode) 475 set_pmode(arcn->name, arcn->sb.st_mode); 476 477 if (arcn->type == PAX_DIR && strcmp(NM_CPIO, argv0) != 0) { 478 /* 479 * Dirs must be processed again at end of extract to set times 480 * and modes to agree with those stored in the archive. However 481 * to allow extract to continue, we may have to also set owner 482 * rights. This allows nodes in the archive that are children 483 * of this directory to be extracted without failure. Both time 484 * and modes will be fixed after the entire archive is read and 485 * before pax exits. 486 */ 487 if (access(nm, R_OK | W_OK | X_OK) < 0) { 488 if (lstat(nm, &sb) < 0) { 489 syswarn(0, errno,"Cannot access %s (stat)", 490 arcn->name); 491 set_pmode(nm,file_mode | S_IRWXU); 492 } else { 493 /* 494 * We have to add rights to the dir, so we make 495 * sure to restore the mode. The mode must be 496 * restored AS CREATED and not as stored if 497 * pmode is not set. 498 */ 499 set_pmode(nm, 500 ((sb.st_mode & FILEBITS) | S_IRWXU)); 501 if (!pmode) 502 arcn->sb.st_mode = sb.st_mode; 503 } 504 505 /* 506 * we have to force the mode to what was set here, 507 * since we changed it from the default as created. 508 */ 509 add_dir(nm, arcn->nlen, &(arcn->sb), 1); 510 } else if (pmode || patime || pmtime) 511 add_dir(nm, arcn->nlen, &(arcn->sb), 0); 512 } 513 514#if HAVE_LUTIMES 515 if (patime || pmtime) 516#else 517 if ((patime || pmtime) && arcn->type != PAX_SLK) 518#endif 519 set_ftime(arcn->name, arcn->sb.st_mtime, arcn->sb.st_atime, 0); 520 521#if HAVE_STRUCT_STAT_ST_FLAGS 522 if (pfflags && arcn->type != PAX_SLK) 523 set_chflags(arcn->name, arcn->sb.st_flags); 524#endif 525 return(0); 526} 527 528/* 529 * unlnk_exist() 530 * Remove node from file system with the specified name. We pass the type 531 * of the node that is going to replace it. When we try to create a 532 * directory and find that it already exists, we allow processing to 533 * continue as proper modes etc will always be set for it later on. 534 * Return: 535 * 0 is ok to proceed, no file with the specified name exists 536 * -1 we were unable to remove the node, or we should not remove it (-k) 537 * 1 we found a directory and we were going to create a directory. 538 */ 539 540int 541unlnk_exist(char *name, int type) 542{ 543 struct stat sb; 544 545 /* 546 * the file does not exist, or -k we are done 547 */ 548 if (lstat(name, &sb) < 0) 549 return(0); 550 if (kflag) 551 return(-1); 552 553 if (S_ISDIR(sb.st_mode)) { 554 /* 555 * try to remove a directory, if it fails and we were going to 556 * create a directory anyway, tell the caller (return a 1) 557 */ 558 if (rmdir(name) < 0) { 559 if (type == PAX_DIR) 560 return(1); 561 syswarn(1, errno, "Cannot remove directory %s", name); 562 return(-1); 563 } 564 return(0); 565 } 566 567 /* 568 * try to get rid of all non-directory type nodes 569 */ 570 if (unlink(name) < 0) { 571 (void)fflush(listf); 572 syswarn(1, errno, "Cannot unlink %s", name); 573 return(-1); 574 } 575 return(0); 576} 577 578/* 579 * chk_path() 580 * We were trying to create some kind of node in the file system and it 581 * failed. chk_path() makes sure the path up to the node exists and is 582 * writable. When we have to create a directory that is missing along the 583 * path somewhere, the directory we create will be set to the same 584 * uid/gid as the file has (when uid and gid are being preserved). 585 * NOTE: this routine is a real performance loss. It is only used as a 586 * last resort when trying to create entries in the file system. 587 * Return: 588 * -1 when it could find nothing it is allowed to fix. 589 * 0 otherwise 590 */ 591 592int 593chk_path( char *name, uid_t st_uid, gid_t st_gid) 594{ 595 char *spt = name; 596 struct stat sb; 597 int retval = -1; 598 599 /* 600 * watch out for paths with nodes stored directly in / (e.g. /bozo) 601 */ 602 if (*spt == '/') 603 ++spt; 604 605 for(;;) { 606 /* 607 * work forward from the first / and check each part of 608 * the path 609 */ 610 spt = strchr(spt, '/'); 611 if (spt == NULL) 612 break; 613 *spt = '\0'; 614 615 /* 616 * if it exists we assume it is a directory, it is not within 617 * the spec (at least it seems to read that way) to alter the 618 * file system for nodes NOT EXPLICITLY stored on the archive. 619 * If that assumption is changed, you would test the node here 620 * and figure out how to get rid of it (probably like some 621 * recursive unlink()) or fix up the directory permissions if 622 * required (do an access()). 623 */ 624 if (lstat(name, &sb) == 0) { 625 *(spt++) = '/'; 626 continue; 627 } 628 629 /* 630 * the path fails at this point, see if we can create the 631 * needed directory and continue on 632 */ 633 if (mkdir(name, S_IRWXU | S_IRWXG | S_IRWXO) < 0) { 634 *spt = '/'; 635 retval = -1; 636 break; 637 } 638 639 /* 640 * we were able to create the directory. We will tell the 641 * caller that we found something to fix, and it is ok to try 642 * and create the node again. 643 */ 644 retval = 0; 645 if (pids) 646 (void)set_ids(name, st_uid, st_gid); 647 648 /* 649 * make sure the user doesn't have some strange umask that 650 * causes this newly created directory to be unusable. We fix 651 * the modes and restore them back to the creation default at 652 * the end of pax 653 */ 654 if ((access(name, R_OK | W_OK | X_OK) < 0) && 655 (lstat(name, &sb) == 0)) { 656 set_pmode(name, ((sb.st_mode & FILEBITS) | S_IRWXU)); 657 add_dir(name, spt - name, &sb, 1); 658 } 659 *(spt++) = '/'; 660 continue; 661 } 662 return(retval); 663} 664 665/* 666 * set_ftime() 667 * Set the access time and modification time for a named file. If frc 668 * is non-zero we force these times to be set even if the user did not 669 * request access and/or modification time preservation (this is also 670 * used by -t to reset access times). 671 * When ign is zero, only those times the user has asked for are set, the 672 * other ones are left alone. We do not assume the un-documented feature 673 * of many utimes() implementations that consider a 0 time value as a do 674 * not set request. 675 */ 676 677void 678set_ftime(char *fnm, time_t mtime, time_t atime, int frc) 679{ 680 struct timeval tv[2]; 681 struct stat sb; 682 683 tv[0].tv_sec = (long)atime; 684 tv[0].tv_usec = 0; 685 tv[1].tv_sec = (long)mtime; 686 tv[1].tv_usec = 0; 687 if (!frc && (!patime || !pmtime)) { 688 /* 689 * if we are not forcing, only set those times the user wants 690 * set. We get the current values of the times if we need them. 691 */ 692 if (lstat(fnm, &sb) == 0) { 693#ifdef BSD4_4 694 if (!patime) 695 TIMESPEC_TO_TIMEVAL(&tv[0], &sb.st_atimespec); 696 if (!pmtime) 697 TIMESPEC_TO_TIMEVAL(&tv[1], &sb.st_mtimespec); 698#else 699 if (!patime) 700 tv[0].tv_sec = sb.st_atime; 701 if (!pmtime) 702 tv[1].tv_sec = sb.st_mtime; 703#endif 704 } else 705 syswarn(0, errno, "Cannot obtain file stats %s", fnm); 706 } 707 708 /* 709 * set the times 710 */ 711#if HAVE_LUTIMES 712 if (lutimes(fnm, tv)) 713#else 714 if (utimes(fnm, tv)) 715#endif 716 syswarn(1, errno, "Access/modification time set failed on: %s", 717 fnm); 718 return; 719} 720 721/* 722 * set_ids() 723 * set the uid and gid of a file system node 724 * Return: 725 * 0 when set, -1 on failure 726 */ 727 728int 729set_ids(char *fnm, uid_t uid, gid_t gid) 730{ 731 if (geteuid() == 0) 732 if (lchown(fnm, uid, gid)) { 733 (void)fflush(listf); 734 syswarn(1, errno, "Cannot set file uid/gid of %s", 735 fnm); 736 return(-1); 737 } 738 return(0); 739} 740 741/* 742 * set_pmode() 743 * Set file access mode 744 */ 745 746void 747set_pmode(char *fnm, mode_t mode) 748{ 749 mode &= ABITS; 750 if (lchmod(fnm, mode)) { 751 (void)fflush(listf); 752 syswarn(1, errno, "Cannot set permissions on %s", fnm); 753 } 754 return; 755} 756 757/* 758 * set_chflags() 759 * Set 4.4BSD file flags 760 */ 761void 762set_chflags(char *fnm, u_int32_t flags) 763{ 764 765#if 0 766 if (chflags(fnm, flags) < 0 && errno != EOPNOTSUPP) 767 syswarn(1, errno, "Cannot set file flags on %s", fnm); 768#endif 769 return; 770} 771 772/* 773 * file_write() 774 * Write/copy a file (during copy or archive extract). This routine knows 775 * how to copy files with lseek holes in it. (Which are read as file 776 * blocks containing all 0's but do not have any file blocks associated 777 * with the data). Typical examples of these are files created by dbm 778 * variants (.pag files). While the file size of these files are huge, the 779 * actual storage is quite small (the files are sparse). The problem is 780 * the holes read as all zeros so are probably stored on the archive that 781 * way (there is no way to determine if the file block is really a hole, 782 * we only know that a file block of all zero's can be a hole). 783 * At this writing, no major archive format knows how to archive files 784 * with holes. However, on extraction (or during copy, -rw) we have to 785 * deal with these files. Without detecting the holes, the files can 786 * consume a lot of file space if just written to disk. This replacement 787 * for write when passed the basic allocation size of a file system block, 788 * uses lseek whenever it detects the input data is all 0 within that 789 * file block. In more detail, the strategy is as follows: 790 * While the input is all zero keep doing an lseek. Keep track of when we 791 * pass over file block boundaries. Only write when we hit a non zero 792 * input. once we have written a file block, we continue to write it to 793 * the end (we stop looking at the input). When we reach the start of the 794 * next file block, start checking for zero blocks again. Working on file 795 * block boundaries significantly reduces the overhead when copying files 796 * that are NOT very sparse. This overhead (when compared to a write) is 797 * almost below the measurement resolution on many systems. Without it, 798 * files with holes cannot be safely copied. It does has a side effect as 799 * it can put holes into files that did not have them before, but that is 800 * not a problem since the file contents are unchanged (in fact it saves 801 * file space). (Except on paging files for diskless clients. But since we 802 * cannot determine one of those file from here, we ignore them). If this 803 * ever ends up on a system where CTG files are supported and the holes 804 * are not desired, just do a conditional test in those routines that 805 * call file_write() and have it call write() instead. BEFORE CLOSING THE 806 * FILE, make sure to call file_flush() when the last write finishes with 807 * an empty block. A lot of file systems will not create an lseek hole at 808 * the end. In this case we drop a single 0 at the end to force the 809 * trailing 0's in the file. 810 * ---Parameters--- 811 * rem: how many bytes left in this file system block 812 * isempt: have we written to the file block yet (is it empty) 813 * sz: basic file block allocation size 814 * cnt: number of bytes on this write 815 * str: buffer to write 816 * Return: 817 * number of bytes written, -1 on write (or lseek) error. 818 */ 819 820int 821file_write(int fd, char *str, int cnt, int *rem, int *isempt, int sz, 822 char *name) 823{ 824 char *pt; 825 char *end; 826 int wcnt; 827 char *st = str; 828 char **strp; 829 830 /* 831 * while we have data to process 832 */ 833 while (cnt) { 834 if (!*rem) { 835 /* 836 * We are now at the start of file system block again 837 * (or what we think one is...). start looking for 838 * empty blocks again 839 */ 840 *isempt = 1; 841 *rem = sz; 842 } 843 844 /* 845 * only examine up to the end of the current file block or 846 * remaining characters to write, whatever is smaller 847 */ 848 wcnt = MIN(cnt, *rem); 849 cnt -= wcnt; 850 *rem -= wcnt; 851 if (*isempt) { 852 /* 853 * have not written to this block yet, so we keep 854 * looking for zero's 855 */ 856 pt = st; 857 end = st + wcnt; 858 859 /* 860 * look for a zero filled buffer 861 */ 862 while ((pt < end) && (*pt == '\0')) 863 ++pt; 864 865 if (pt == end) { 866 /* 867 * skip, buf is empty so far 868 */ 869 if (fd > -1 && 870 lseek(fd, (off_t)wcnt, SEEK_CUR) < 0) { 871 syswarn(1, errno, "File seek on %s", 872 name); 873 return(-1); 874 } 875 st = pt; 876 continue; 877 } 878 /* 879 * drat, the buf is not zero filled 880 */ 881 *isempt = 0; 882 } 883 884 /* 885 * have non-zero data in this file system block, have to write 886 */ 887 switch (fd) { 888 case -1: 889 strp = &gnu_name_string; 890 break; 891 case -2: 892 strp = &gnu_link_string; 893 break; 894 default: 895 strp = NULL; 896 break; 897 } 898 if (strp) { 899 if (*strp) 900 err(1, "WARNING! Major Internal Error! GNU hack Failing!"); 901 *strp = malloc(wcnt + 1); 902 if (*strp == NULL) { 903 tty_warn(1, "Out of memory"); 904 return(-1); 905 } 906 strlcpy(*strp, st, wcnt); 907 break; 908 } else if (xwrite(fd, st, wcnt) != wcnt) { 909 syswarn(1, errno, "Failed write to file %s", name); 910 return(-1); 911 } 912 st += wcnt; 913 } 914 return(st - str); 915} 916 917/* 918 * file_flush() 919 * when the last file block in a file is zero, many file systems will not 920 * let us create a hole at the end. To get the last block with zeros, we 921 * write the last BYTE with a zero (back up one byte and write a zero). 922 */ 923 924void 925file_flush(int fd, char *fname, int isempt) 926{ 927 static char blnk[] = "\0"; 928 929 /* 930 * silly test, but make sure we are only called when the last block is 931 * filled with all zeros. 932 */ 933 if (!isempt) 934 return; 935 936 /* 937 * move back one byte and write a zero 938 */ 939 if (lseek(fd, (off_t)-1, SEEK_CUR) < 0) { 940 syswarn(1, errno, "Failed seek on file %s", fname); 941 return; 942 } 943 944 if (write_with_restart(fd, blnk, 1) < 0) 945 syswarn(1, errno, "Failed write to file %s", fname); 946 return; 947} 948 949/* 950 * rdfile_close() 951 * close a file we have been reading (to copy or archive). If we have to 952 * reset access time (tflag) do so (the times are stored in arcn). 953 */ 954 955void 956rdfile_close(ARCHD *arcn, int *fd) 957{ 958 /* 959 * make sure the file is open 960 */ 961 if (*fd < 0) 962 return; 963 964 (void)close(*fd); 965 *fd = -1; 966 if (!tflag) 967 return; 968 969 /* 970 * user wants last access time reset 971 */ 972 set_ftime(arcn->org_name, arcn->sb.st_mtime, arcn->sb.st_atime, 1); 973 return; 974} 975 976/* 977 * set_crc() 978 * read a file to calculate its crc. This is a real drag. Archive formats 979 * that have this, end up reading the file twice (we have to write the 980 * header WITH the crc before writing the file contents. Oh well... 981 * Return: 982 * 0 if was able to calculate the crc, -1 otherwise 983 */ 984 985int 986set_crc(ARCHD *arcn, int fd) 987{ 988 int i; 989 int res; 990 off_t cpcnt = 0L; 991 u_long size; 992 unsigned long crc = 0L; 993 char tbuf[FILEBLK]; 994 struct stat sb; 995 996 if (fd < 0) { 997 /* 998 * hmm, no fd, should never happen. well no crc then. 999 */ 1000 arcn->crc = 0L; 1001 return(0); 1002 } 1003 1004 if ((size = (u_long)arcn->sb.st_blksize) > (u_long)sizeof(tbuf)) 1005 size = (u_long)sizeof(tbuf); 1006 1007 /* 1008 * read all the bytes we think that there are in the file. If the user 1009 * is trying to archive an active file, forget this file. 1010 */ 1011 for(;;) { 1012 if ((res = read(fd, tbuf, size)) <= 0) 1013 break; 1014 cpcnt += res; 1015 for (i = 0; i < res; ++i) 1016 crc += (tbuf[i] & 0xff); 1017 } 1018 1019 /* 1020 * safety check. we want to avoid archiving files that are active as 1021 * they can create inconsistant archive copies. 1022 */ 1023 if (cpcnt != arcn->sb.st_size) 1024 tty_warn(1, "File changed size %s", arcn->org_name); 1025 else if (fstat(fd, &sb) < 0) 1026 syswarn(1, errno, "Failed stat on %s", arcn->org_name); 1027 else if (arcn->sb.st_mtime != sb.st_mtime) 1028 tty_warn(1, "File %s was modified during read", arcn->org_name); 1029 else if (lseek(fd, (off_t)0L, SEEK_SET) < 0) 1030 syswarn(1, errno, "File rewind failed on: %s", arcn->org_name); 1031 else { 1032 arcn->crc = crc; 1033 return(0); 1034 } 1035 return(-1); 1036} 1037