1 /*- 2 * Copyright (c) 1992 Keith Muller. 3 * Copyright (c) 1992, 1993 4 * The Regents of the University of California. All rights reserved. 5 * 6 * This code is derived from software contributed to Berkeley by 7 * Keith Muller of the University of California, San Diego. 8 * 9 * Redistribution and use in source and binary forms, with or without 10 * modification, are permitted provided that the following conditions 11 * are met: 12 * 1. Redistributions of source code must retain the above copyright 13 * notice, this list of conditions and the following disclaimer. 14 * 2. Redistributions in binary form must reproduce the above copyright 15 * notice, this list of conditions and the following disclaimer in the 16 * documentation and/or other materials provided with the distribution. 17 * 3. All advertising materials mentioning features or use of this software 18 * must display the following acknowledgement: 19 * This product includes software developed by the University of 20 * California, Berkeley and its contributors. 21 * 4. Neither the name of the University nor the names of its contributors 22 * may be used to endorse or promote products derived from this software 23 * without specific prior written permission. 24 * 25 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 26 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 27 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 28 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 29 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 30 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 31 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 32 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 33 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 34 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 35 * SUCH DAMAGE. 36 */ 37 38 #ifndef lint 39 #if 0 40 static char sccsid[] = "@(#)file_subs.c 8.1 (Berkeley) 5/31/93"; 41 #endif 42 static const char rcsid[] = 43 "$Id$"; 44 #endif /* not lint */ 45 46 #include <sys/types.h> 47 #include <sys/time.h> 48 #include <sys/stat.h> 49 #include <unistd.h> 50 #include <fcntl.h> 51 #include <string.h> 52 #include <stdio.h> 53 #include <errno.h> 54 #include <sys/uio.h> 55 #include <stdlib.h> 56 #include "pax.h" 57 #include "extern.h" 58 59 static int 60 mk_link __P((register char *,register struct stat *,register char *, int)); 61 62 /* 63 * routines that deal with file operations such as: creating, removing; 64 * and setting access modes, uid/gid and times of files 65 */ 66 67 #define FILEBITS (S_ISVTX | S_IRWXU | S_IRWXG | S_IRWXO) 68 #define SETBITS (S_ISUID | S_ISGID) 69 #define ABITS (FILEBITS | SETBITS) 70 71 /* 72 * file_creat() 73 * Create and open a file. 74 * Return: 75 * file descriptor or -1 for failure 76 */ 77 78 #if __STDC__ 79 int 80 file_creat(register ARCHD *arcn) 81 #else 82 int 83 file_creat(arcn) 84 register ARCHD *arcn; 85 #endif 86 { 87 int fd = -1; 88 mode_t file_mode; 89 int oerrno; 90 91 /* 92 * assume file doesn't exist, so just try to create it, most times this 93 * works. We have to take special handling when the file does exist. To 94 * detect this, we use O_EXCL. For example when trying to create a 95 * file and a character device or fifo exists with the same name, we 96 * can accidently open the device by mistake (or block waiting to open) 97 * If we find that the open has failed, then figure spend the effore to 98 * figure out why. This strategy was found to have better average 99 * performance in common use than checking the file (and the path) 100 * first with lstat. 101 */ 102 file_mode = arcn->sb.st_mode & FILEBITS; 103 if ((fd = open(arcn->name, O_WRONLY | O_CREAT | O_TRUNC | O_EXCL, 104 file_mode)) >= 0) 105 return(fd); 106 107 /* 108 * the file seems to exist. First we try to get rid of it (found to be 109 * the second most common failure when traced). If this fails, only 110 * then we go to the expense to check and create the path to the file 111 */ 112 if (unlnk_exist(arcn->name, arcn->type) != 0) 113 return(-1); 114 115 for (;;) { 116 /* 117 * try to open it again, if this fails, check all the nodes in 118 * the path and give it a final try. if chk_path() finds that 119 * it cannot fix anything, we will skip the last attempt 120 */ 121 if ((fd = open(arcn->name, O_WRONLY | O_CREAT | O_TRUNC, 122 file_mode)) >= 0) 123 break; 124 oerrno = errno; 125 if (chk_path(arcn->name,arcn->sb.st_uid,arcn->sb.st_gid) < 0) { 126 sys_warn(1, oerrno, "Unable to create %s", arcn->name); 127 return(-1); 128 } 129 } 130 return(fd); 131 } 132 133 /* 134 * file_close() 135 * Close file descriptor to a file just created by pax. Sets modes, 136 * ownership and times as required. 137 * Return: 138 * 0 for success, -1 for failure 139 */ 140 141 #if __STDC__ 142 void 143 file_close(register ARCHD *arcn, int fd) 144 #else 145 void 146 file_close(arcn, fd) 147 register ARCHD *arcn; 148 int fd; 149 #endif 150 { 151 int res = 0; 152 153 if (fd < 0) 154 return; 155 if (close(fd) < 0) 156 sys_warn(0, errno, "Unable to close file descriptor on %s", 157 arcn->name); 158 159 /* 160 * set owner/groups first as this may strip off mode bits we want 161 * then set file permission modes. Then set file access and 162 * modification times. 163 */ 164 if (pids) 165 res = set_ids(arcn->name, arcn->sb.st_uid, arcn->sb.st_gid); 166 167 /* 168 * IMPORTANT SECURITY NOTE: 169 * if not preserving mode or we cannot set uid/gid, then PROHIBIT 170 * set uid/gid bits 171 */ 172 if (!pmode || res) 173 arcn->sb.st_mode &= ~(SETBITS); 174 if (pmode) 175 set_pmode(arcn->name, arcn->sb.st_mode); 176 if (patime || pmtime) 177 set_ftime(arcn->name, arcn->sb.st_mtime, arcn->sb.st_atime, 0); 178 } 179 180 /* 181 * lnk_creat() 182 * Create a hard link to arcn->ln_name from arcn->name. arcn->ln_name 183 * must exist; 184 * Return: 185 * 0 if ok, -1 otherwise 186 */ 187 188 #if __STDC__ 189 int 190 lnk_creat(register ARCHD *arcn) 191 #else 192 int 193 lnk_creat(arcn) 194 register ARCHD *arcn; 195 #endif 196 { 197 struct stat sb; 198 199 /* 200 * we may be running as root, so we have to be sure that link target 201 * is not a directory, so we lstat and check 202 */ 203 if (lstat(arcn->ln_name, &sb) < 0) { 204 sys_warn(1,errno,"Unable to link to %s from %s", arcn->ln_name, 205 arcn->name); 206 return(-1); 207 } 208 209 if (S_ISDIR(sb.st_mode)) { 210 pax_warn(1, "A hard link to the directory %s is not allowed", 211 arcn->ln_name); 212 return(-1); 213 } 214 215 return(mk_link(arcn->ln_name, &sb, arcn->name, 0)); 216 } 217 218 /* 219 * cross_lnk() 220 * Create a hard link to arcn->org_name from arcn->name. Only used in copy 221 * with the -l flag. No pax_warning or error if this does not succeed (we will 222 * then just create the file) 223 * Return: 224 * 1 if copy() should try to create this file node 225 * 0 if cross_lnk() ok, -1 for fatal flaw (like linking to self). 226 */ 227 228 #if __STDC__ 229 int 230 cross_lnk(register ARCHD *arcn) 231 #else 232 int 233 cross_lnk(arcn) 234 register ARCHD *arcn; 235 #endif 236 { 237 /* 238 * try to make a link to orginal file (-l flag in copy mode). make sure 239 * we do not try to link to directories in case we are running as root 240 * (and it might succeed). 241 */ 242 if (arcn->type == PAX_DIR) 243 return(1); 244 return(mk_link(arcn->org_name, &(arcn->sb), arcn->name, 1)); 245 } 246 247 /* 248 * chk_same() 249 * In copy mode if we are not trying to make hard links between the src 250 * and destinations, make sure we are not going to overwrite ourselves by 251 * accident. This slows things down a little, but we have to protect all 252 * those people who make typing errors. 253 * Return: 254 * 1 the target does not exist, go ahead and copy 255 * 0 skip it file exists (-k) or may be the same as source file 256 */ 257 258 #if __STDC__ 259 int 260 chk_same(register ARCHD *arcn) 261 #else 262 int 263 chk_same(arcn) 264 register ARCHD *arcn; 265 #endif 266 { 267 struct stat sb; 268 269 /* 270 * if file does not exist, return. if file exists and -k, skip it 271 * quietly 272 */ 273 if (lstat(arcn->name, &sb) < 0) 274 return(1); 275 if (kflag) 276 return(0); 277 278 /* 279 * better make sure the user does not have src == dest by mistake 280 */ 281 if ((arcn->sb.st_dev == sb.st_dev) && (arcn->sb.st_ino == sb.st_ino)) { 282 pax_warn(1, "Unable to copy %s, file would overwrite itself", 283 arcn->name); 284 return(0); 285 } 286 return(1); 287 } 288 289 /* 290 * mk_link() 291 * try to make a hard link between two files. if ign set, we do not 292 * complain. 293 * Return: 294 * 0 if successful (or we are done with this file but no error, such as 295 * finding the from file exists and the user has set -k). 296 * 1 when ign was set to indicates we could not make the link but we 297 * should try to copy/extract the file as that might work (and is an 298 * allowed option). -1 an error occurred. 299 */ 300 301 #if __STDC__ 302 static int 303 mk_link(register char *to, register struct stat *to_sb, register char *from, 304 int ign) 305 #else 306 static int 307 mk_link(to, to_sb, from, ign) 308 register char *to; 309 register struct stat *to_sb; 310 register char *from; 311 int ign; 312 #endif 313 { 314 struct stat sb; 315 int oerrno; 316 317 /* 318 * if from file exists, it has to be unlinked to make the link. If the 319 * file exists and -k is set, skip it quietly 320 */ 321 if (lstat(from, &sb) == 0) { 322 if (kflag) 323 return(0); 324 325 /* 326 * make sure it is not the same file, protect the user 327 */ 328 if ((to_sb->st_dev==sb.st_dev)&&(to_sb->st_ino == sb.st_ino)) { 329 pax_warn(1, "Unable to link file %s to itself", to); 330 return(-1);; 331 } 332 333 /* 334 * try to get rid of the file, based on the type 335 */ 336 if (S_ISDIR(sb.st_mode)) { 337 if (rmdir(from) < 0) { 338 sys_warn(1, errno, "Unable to remove %s", from); 339 return(-1); 340 } 341 } else if (unlink(from) < 0) { 342 if (!ign) { 343 sys_warn(1, errno, "Unable to remove %s", from); 344 return(-1); 345 } 346 return(1); 347 } 348 } 349 350 /* 351 * from file is gone (or did not exist), try to make the hard link. 352 * if it fails, check the path and try it again (if chk_path() says to 353 * try again) 354 */ 355 for (;;) { 356 if (link(to, from) == 0) 357 break; 358 oerrno = errno; 359 if (chk_path(from, to_sb->st_uid, to_sb->st_gid) == 0) 360 continue; 361 if (!ign) { 362 sys_warn(1, oerrno, "Could not link to %s from %s", to, 363 from); 364 return(-1); 365 } 366 return(1); 367 } 368 369 /* 370 * all right the link was made 371 */ 372 return(0); 373 } 374 375 /* 376 * node_creat() 377 * create an entry in the file system (other than a file or hard link). 378 * If successful, sets uid/gid modes and times as required. 379 * Return: 380 * 0 if ok, -1 otherwise 381 */ 382 383 #if __STDC__ 384 int 385 node_creat(register ARCHD *arcn) 386 #else 387 int 388 node_creat(arcn) 389 register ARCHD *arcn; 390 #endif 391 { 392 register int res; 393 register int ign = 0; 394 register int oerrno; 395 register int pass = 0; 396 mode_t file_mode; 397 struct stat sb; 398 399 /* 400 * create node based on type, if that fails try to unlink the node and 401 * try again. finally check the path and try again. As noted in the 402 * file and link creation routines, this method seems to exhibit the 403 * best performance in general use workloads. 404 */ 405 file_mode = arcn->sb.st_mode & FILEBITS; 406 407 for (;;) { 408 switch(arcn->type) { 409 case PAX_DIR: 410 res = mkdir(arcn->name, file_mode); 411 if (ign) 412 res = 0; 413 break; 414 case PAX_CHR: 415 file_mode |= S_IFCHR; 416 res = mknod(arcn->name, file_mode, arcn->sb.st_rdev); 417 break; 418 case PAX_BLK: 419 file_mode |= S_IFBLK; 420 res = mknod(arcn->name, file_mode, arcn->sb.st_rdev); 421 break; 422 case PAX_FIF: 423 res = mkfifo(arcn->name, file_mode); 424 break; 425 case PAX_SCK: 426 /* 427 * Skip sockets, operation has no meaning under BSD 428 */ 429 pax_warn(0, 430 "%s skipped. Sockets cannot be copied or extracted", 431 arcn->name); 432 return(-1); 433 case PAX_SLK: 434 if ((res = symlink(arcn->ln_name, arcn->name)) == 0) 435 return(0); 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 pax_warn(0, "%s has an unknown file type, skipping", 446 arcn->name); 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(arcn->name, arcn->type)) < 0) 463 return(-1); 464 465 if (++pass <= 1) 466 continue; 467 468 if (chk_path(arcn->name,arcn->sb.st_uid,arcn->sb.st_gid) < 0) { 469 sys_warn(1, oerrno, "Could not create: %s", arcn->name); 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(arcn->name, 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) { 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(arcn->name, R_OK | W_OK | X_OK) < 0) { 503 if (lstat(arcn->name, &sb) < 0) { 504 sys_warn(0, errno,"Could not access %s (stat)", 505 arcn->name); 506 set_pmode(arcn->name,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(arcn->name, 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(arcn->name, arcn->nlen, &(arcn->sb), 1); 525 } else if (pmode || patime || pmtime) 526 add_dir(arcn->name, arcn->nlen, &(arcn->sb), 0); 527 } 528 529 if (patime || pmtime) 530 set_ftime(arcn->name, arcn->sb.st_mtime, arcn->sb.st_atime, 0); 531 return(0); 532 } 533 534 /* 535 * unlnk_exist() 536 * Remove node from file system with the specified name. We pass the type 537 * of the node that is going to replace it. When we try to create a 538 * directory and find that it already exists, we allow processing to 539 * continue as proper modes etc will always be set for it later on. 540 * Return: 541 * 0 is ok to proceed, no file with the specified name exists 542 * -1 we were unable to remove the node, or we should not remove it (-k) 543 * 1 we found a directory and we were going to create a directory. 544 */ 545 546 #if __STDC__ 547 int 548 unlnk_exist(register char *name, register int type) 549 #else 550 int 551 unlnk_exist(name, type) 552 register char *name; 553 register int type; 554 #endif 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 sys_warn(1,errno,"Unable to 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 sys_warn(1, errno, "Could not unlink %s", name); 585 return(-1); 586 } 587 return(0); 588 } 589 590 /* 591 * chk_path() 592 * We were trying to create some kind of node in the file system and it 593 * failed. chk_path() makes sure the path up to the node exists and is 594 * writeable. When we have to create a directory that is missing along the 595 * path somewhere, the directory we create will be set to the same 596 * uid/gid as the file has (when uid and gid are being preserved). 597 * NOTE: this routine is a real performance loss. It is only used as a 598 * last resort when trying to create entries in the file system. 599 * Return: 600 * -1 when it could find nothing it is allowed to fix. 601 * 0 otherwise 602 */ 603 604 #if __STDC__ 605 int 606 chk_path( register char *name, uid_t st_uid, gid_t st_gid) 607 #else 608 int 609 chk_path(name, st_uid, st_gid) 610 register char *name; 611 uid_t st_uid; 612 gid_t st_gid; 613 #endif 614 { 615 register char *spt = name; 616 struct stat sb; 617 int retval = -1; 618 619 /* 620 * watch out for paths with nodes stored directly in / (e.g. /bozo) 621 */ 622 if (*spt == '/') 623 ++spt; 624 625 for(;;) { 626 /* 627 * work foward from the first / and check each part of the path 628 */ 629 spt = strchr(spt, '/'); 630 if (spt == NULL) 631 break; 632 *spt = '\0'; 633 634 /* 635 * if it exists we assume it is a directory, it is not within 636 * the spec (at least it seems to read that way) to alter the 637 * file system for nodes NOT EXPLICITLY stored on the archive. 638 * If that assumption is changed, you would test the node here 639 * and figure out how to get rid of it (probably like some 640 * recursive unlink()) or fix up the directory permissions if 641 * required (do an access()). 642 */ 643 if (lstat(name, &sb) == 0) { 644 *(spt++) = '/'; 645 continue; 646 } 647 648 /* 649 * the path fails at this point, see if we can create the 650 * needed directory and continue on 651 */ 652 if (mkdir(name, S_IRWXU | S_IRWXG | S_IRWXO) < 0) { 653 *spt = '/'; 654 retval = -1; 655 break; 656 } 657 658 /* 659 * we were able to create the directory. We will tell the 660 * caller that we found something to fix, and it is ok to try 661 * and create the node again. 662 */ 663 retval = 0; 664 if (pids) 665 (void)set_ids(name, st_uid, st_gid); 666 667 /* 668 * make sure the user doen't have some strange umask that 669 * causes this newly created directory to be unusable. We fix 670 * the modes and restore them back to the creation default at 671 * the end of pax 672 */ 673 if ((access(name, R_OK | W_OK | X_OK) < 0) && 674 (lstat(name, &sb) == 0)) { 675 set_pmode(name, ((sb.st_mode & FILEBITS) | S_IRWXU)); 676 add_dir(name, spt - name, &sb, 1); 677 } 678 *(spt++) = '/'; 679 continue; 680 } 681 return(retval); 682 } 683 684 /* 685 * set_ftime() 686 * Set the access time and modification time for a named file. If frc is 687 * non-zero we force these times to be set even if the the user did not 688 * request access and/or modification time preservation (this is also 689 * used by -t to reset access times). 690 * When ign is zero, only those times the user has asked for are set, the 691 * other ones are left alone. We do not assume the un-documented feature 692 * of many utimes() implementations that consider a 0 time value as a do 693 * not set request. 694 */ 695 696 #if __STDC__ 697 void 698 set_ftime(char *fnm, time_t mtime, time_t atime, int frc) 699 #else 700 void 701 set_ftime(fnm, mtime, atime, frc) 702 char *fnm; 703 time_t mtime; 704 time_t atime; 705 int frc; 706 #endif 707 { 708 static struct timeval tv[2] = {{0L, 0L}, {0L, 0L}}; 709 struct stat sb; 710 711 tv[0].tv_sec = (long)atime; 712 tv[1].tv_sec = (long)mtime; 713 if (!frc && (!patime || !pmtime)) { 714 /* 715 * if we are not forcing, only set those times the user wants 716 * set. We get the current values of the times if we need them. 717 */ 718 if (lstat(fnm, &sb) == 0) { 719 if (!patime) 720 tv[0].tv_sec = (long)sb.st_atime; 721 if (!pmtime) 722 tv[1].tv_sec = (long)sb.st_mtime; 723 } else 724 sys_warn(0,errno,"Unable to obtain file stats %s", fnm); 725 } 726 727 /* 728 * set the times 729 */ 730 if (utimes(fnm, tv) < 0) 731 sys_warn(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 */ 742 743 #if __STDC__ 744 int 745 set_ids(char *fnm, uid_t uid, gid_t gid) 746 #else 747 int 748 set_ids(fnm, uid, gid) 749 char *fnm; 750 uid_t uid; 751 gid_t gid; 752 #endif 753 { 754 if (chown(fnm, uid, gid) < 0) { 755 sys_warn(1, errno, "Unable to set file uid/gid of %s", fnm); 756 return(-1); 757 } 758 return(0); 759 } 760 761 /* 762 * set_pmode() 763 * Set file access mode 764 */ 765 766 #if __STDC__ 767 void 768 set_pmode(char *fnm, mode_t mode) 769 #else 770 void 771 set_pmode(fnm, mode) 772 char *fnm; 773 mode_t mode; 774 #endif 775 { 776 mode &= ABITS; 777 if (chmod(fnm, mode) < 0) 778 sys_warn(1, errno, "Could not set permissions on %s", fnm); 779 return; 780 } 781 782 /* 783 * file_write() 784 * Write/copy a file (during copy or archive extract). This routine knows 785 * how to copy files with lseek holes in it. (Which are read as file 786 * blocks containing all 0's but do not have any file blocks associated 787 * with the data). Typical examples of these are files created by dbm 788 * variants (.pag files). While the file size of these files are huge, the 789 * actual storage is quite small (the files are sparse). The problem is 790 * the holes read as all zeros so are probably stored on the archive that 791 * way (there is no way to determine if the file block is really a hole, 792 * we only know that a file block of all zero's can be a hole). 793 * At this writing, no major archive format knows how to archive files 794 * with holes. However, on extraction (or during copy, -rw) we have to 795 * deal with these files. Without detecting the holes, the files can 796 * consume a lot of file space if just written to disk. This replacement 797 * for write when passed the basic allocation size of a file system block, 798 * uses lseek whenever it detects the input data is all 0 within that 799 * file block. In more detail, the strategy is as follows: 800 * While the input is all zero keep doing an lseek. Keep track of when we 801 * pass over file block boundries. Only write when we hit a non zero 802 * input. once we have written a file block, we continue to write it to 803 * the end (we stop looking at the input). When we reach the start of the 804 * next file block, start checking for zero blocks again. Working on file 805 * block boundries significantly reduces the overhead when copying files 806 * that are NOT very sparse. This overhead (when compared to a write) is 807 * almost below the measurement resolution on many systems. Without it, 808 * files with holes cannot be safely copied. It does has a side effect as 809 * it can put holes into files that did not have them before, but that is 810 * not a problem since the file contents are unchanged (in fact it saves 811 * file space). (Except on paging files for diskless clients. But since we 812 * cannot determine one of those file from here, we ignore them). If this 813 * ever ends up on a system where CTG files are supported and the holes 814 * are not desired, just do a conditional test in those routines that 815 * call file_write() and have it call write() instead. BEFORE CLOSING THE 816 * FILE, make sure to call file_flush() when the last write finishes with 817 * an empty block. A lot of file systems will not create an lseek hole at 818 * the end. In this case we drop a single 0 at the end to force the 819 * trailing 0's in the file. 820 * ---Parameters--- 821 * rem: how many bytes left in this file system block 822 * isempt: have we written to the file block yet (is it empty) 823 * sz: basic file block allocation size 824 * cnt: number of bytes on this write 825 * str: buffer to write 826 * Return: 827 * number of bytes written, -1 on write (or lseek) error. 828 */ 829 830 #if __STDC__ 831 int 832 file_write(int fd, char *str, register int cnt, int *rem, int *isempt, int sz, 833 char *name) 834 #else 835 int 836 file_write(fd, str, cnt, rem, isempt, sz, name) 837 int fd; 838 char *str; 839 register int cnt; 840 int *rem; 841 int *isempt; 842 int sz; 843 char *name; 844 #endif 845 { 846 register char *pt; 847 register char *end; 848 register int wcnt; 849 register char *st = str; 850 851 /* 852 * while we have data to process 853 */ 854 while (cnt) { 855 if (!*rem) { 856 /* 857 * We are now at the start of file system block again 858 * (or what we think one is...). start looking for 859 * empty blocks again 860 */ 861 *isempt = 1; 862 *rem = sz; 863 } 864 865 /* 866 * only examine up to the end of the current file block or 867 * remaining characters to write, whatever is smaller 868 */ 869 wcnt = MIN(cnt, *rem); 870 cnt -= wcnt; 871 *rem -= wcnt; 872 if (*isempt) { 873 /* 874 * have not written to this block yet, so we keep 875 * looking for zero's 876 */ 877 pt = st; 878 end = st + wcnt; 879 880 /* 881 * look for a zero filled buffer 882 */ 883 while ((pt < end) && (*pt == '\0')) 884 ++pt; 885 886 if (pt == end) { 887 /* 888 * skip, buf is empty so far 889 */ 890 if (lseek(fd, (off_t)wcnt, SEEK_CUR) < 0) { 891 sys_warn(1,errno,"File seek on %s", 892 name); 893 return(-1); 894 } 895 st = pt; 896 continue; 897 } 898 /* 899 * drat, the buf is not zero filled 900 */ 901 *isempt = 0; 902 } 903 904 /* 905 * have non-zero data in this file system block, have to write 906 */ 907 if (write(fd, st, wcnt) != wcnt) { 908 sys_warn(1, errno, "Failed write to file %s", name); 909 return(-1); 910 } 911 st += wcnt; 912 } 913 return(st - str); 914 } 915 916 /* 917 * file_flush() 918 * when the last file block in a file is zero, many file systems will not 919 * let us create a hole at the end. To get the last block with zeros, we 920 * write the last BYTE with a zero (back up one byte and write a zero). 921 */ 922 923 #if __STDC__ 924 void 925 file_flush(int fd, char *fname, int isempt) 926 #else 927 void 928 file_flush(fd, fname, isempt) 929 int fd; 930 char *fname; 931 int isempt; 932 #endif 933 { 934 static char blnk[] = "\0"; 935 936 /* 937 * silly test, but make sure we are only called when the last block is 938 * filled with all zeros. 939 */ 940 if (!isempt) 941 return; 942 943 /* 944 * move back one byte and write a zero 945 */ 946 if (lseek(fd, (off_t)-1, SEEK_CUR) < 0) { 947 sys_warn(1, errno, "Failed seek on file %s", fname); 948 return; 949 } 950 951 if (write(fd, blnk, 1) < 0) 952 sys_warn(1, errno, "Failed write to file %s", fname); 953 return; 954 } 955 956 /* 957 * rdfile_close() 958 * close a file we have beed reading (to copy or archive). If we have to 959 * reset access time (tflag) do so (the times are stored in arcn). 960 */ 961 962 #if __STDC__ 963 void 964 rdfile_close(register ARCHD *arcn, register int *fd) 965 #else 966 void 967 rdfile_close(arcn, fd) 968 register ARCHD *arcn; 969 register int *fd; 970 #endif 971 { 972 /* 973 * make sure the file is open 974 */ 975 if (*fd < 0) 976 return; 977 978 (void)close(*fd); 979 *fd = -1; 980 if (!tflag) 981 return; 982 983 /* 984 * user wants last access time reset 985 */ 986 set_ftime(arcn->org_name, arcn->sb.st_mtime, arcn->sb.st_atime, 1); 987 return; 988 } 989 990 /* 991 * set_crc() 992 * read a file to calculate its crc. This is a real drag. Archive formats 993 * that have this, end up reading the file twice (we have to write the 994 * header WITH the crc before writing the file contents. Oh well... 995 * Return: 996 * 0 if was able to calculate the crc, -1 otherwise 997 */ 998 999 #if __STDC__ 1000 int 1001 set_crc(register ARCHD *arcn, register int fd) 1002 #else 1003 int 1004 set_crc(arcn, fd) 1005 register ARCHD *arcn; 1006 register int fd; 1007 #endif 1008 { 1009 register int i; 1010 register int res; 1011 off_t cpcnt = 0L; 1012 u_long size; 1013 unsigned long crc = 0L; 1014 char tbuf[FILEBLK]; 1015 struct stat sb; 1016 1017 if (fd < 0) { 1018 /* 1019 * hmm, no fd, should never happen. well no crc then. 1020 */ 1021 arcn->crc = 0L; 1022 return(0); 1023 } 1024 1025 if ((size = (u_long)arcn->sb.st_blksize) > (u_long)sizeof(tbuf)) 1026 size = (u_long)sizeof(tbuf); 1027 1028 /* 1029 * read all the bytes we think that there are in the file. If the user 1030 * is trying to archive an active file, forget this file. 1031 */ 1032 for(;;) { 1033 if ((res = read(fd, tbuf, size)) <= 0) 1034 break; 1035 cpcnt += res; 1036 for (i = 0; i < res; ++i) 1037 crc += (tbuf[i] & 0xff); 1038 } 1039 1040 /* 1041 * safety check. we want to avoid archiving files that are active as 1042 * they can create inconsistant archive copies. 1043 */ 1044 if (cpcnt != arcn->sb.st_size) 1045 pax_warn(1, "File changed size %s", arcn->org_name); 1046 else if (fstat(fd, &sb) < 0) 1047 sys_warn(1, errno, "Failed stat on %s", arcn->org_name); 1048 else if (arcn->sb.st_mtime != sb.st_mtime) 1049 pax_warn(1, "File %s was modified during read", arcn->org_name); 1050 else if (lseek(fd, (off_t)0L, SEEK_SET) < 0) 1051 sys_warn(1, errno, "File rewind failed on: %s", arcn->org_name); 1052 else { 1053 arcn->crc = crc; 1054 return(0); 1055 } 1056 return(-1); 1057 } 1058