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 * @(#)file_subs.c 8.1 (Berkeley) 5/31/93 38 * $FreeBSD: src/bin/pax/file_subs.c,v 1.12.2.1 2001/08/01 05:03:11 obrien Exp $ 39 * $DragonFly: src/bin/pax/file_subs.c,v 1.7 2006/08/03 16:40:45 swildner Exp $ 40 */ 41 42 #include <sys/types.h> 43 #include <sys/time.h> 44 #include <sys/stat.h> 45 #include <unistd.h> 46 #include <fcntl.h> 47 #include <string.h> 48 #include <stdio.h> 49 #include <errno.h> 50 #include <sys/uio.h> 51 #include <stdlib.h> 52 #include "pax.h" 53 #include "options.h" 54 #include "extern.h" 55 56 static int 57 mk_link (char *,struct stat *,char *, int); 58 59 /* 60 * routines that deal with file operations such as: creating, removing; 61 * and setting access modes, uid/gid and times of files 62 */ 63 64 #define FILEBITS (S_ISVTX | S_IRWXU | S_IRWXG | S_IRWXO) 65 #define SETBITS (S_ISUID | S_ISGID) 66 #define ABITS (FILEBITS | SETBITS) 67 68 /* 69 * file_creat() 70 * Create and open a file. 71 * Return: 72 * file descriptor or -1 for failure 73 */ 74 75 int 76 file_creat(ARCHD *arcn) 77 { 78 int fd = -1; 79 mode_t file_mode; 80 int oerrno; 81 82 /* 83 * assume file doesn't exist, so just try to create it, most times this 84 * works. We have to take special handling when the file does exist. To 85 * detect this, we use O_EXCL. For example when trying to create a 86 * file and a character device or fifo exists with the same name, we 87 * can accidently open the device by mistake (or block waiting to open) 88 * If we find that the open has failed, then figure spend the effort to 89 * figure out why. This strategy was found to have better average 90 * performance in common use than checking the file (and the path) 91 * first with lstat. 92 */ 93 file_mode = arcn->sb.st_mode & FILEBITS; 94 if ((fd = open(arcn->name, O_WRONLY | O_CREAT | O_TRUNC | O_EXCL, 95 file_mode)) >= 0) 96 return(fd); 97 98 /* 99 * the file seems to exist. First we try to get rid of it (found to be 100 * the second most common failure when traced). If this fails, only 101 * then we go to the expense to check and create the path to the file 102 */ 103 if (unlnk_exist(arcn->name, arcn->type) != 0) 104 return(-1); 105 106 for (;;) { 107 /* 108 * try to open it again, if this fails, check all the nodes in 109 * the path and give it a final try. if chk_path() finds that 110 * it cannot fix anything, we will skip the last attempt 111 */ 112 if ((fd = open(arcn->name, O_WRONLY | O_CREAT | O_TRUNC, 113 file_mode)) >= 0) 114 break; 115 oerrno = errno; 116 if (nodirs || chk_path(arcn->name,arcn->sb.st_uid,arcn->sb.st_gid) < 0) { 117 syswarn(1, oerrno, "Unable to create %s", arcn->name); 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 */ 131 132 void 133 file_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, "Unable to close file descriptor on %s", 141 arcn->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->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 155 */ 156 if (!pmode || res) 157 arcn->sb.st_mode &= ~(SETBITS); 158 if (pmode) 159 set_pmode(arcn->name, arcn->sb.st_mode); 160 if (patime || pmtime) 161 set_ftime(arcn->name, arcn->sb.st_mtime, arcn->sb.st_atime, 0); 162 } 163 164 /* 165 * lnk_creat() 166 * Create a hard link to arcn->ln_name from arcn->name. arcn->ln_name 167 * must exist; 168 * Return: 169 * 0 if ok, -1 otherwise 170 */ 171 172 int 173 lnk_creat(ARCHD *arcn) 174 { 175 struct stat sb; 176 177 /* 178 * we may be running as root, so we have to be sure that link target 179 * is not a directory, so we lstat and check 180 */ 181 if (lstat(arcn->ln_name, &sb) < 0) { 182 syswarn(1,errno,"Unable to link to %s from %s", arcn->ln_name, 183 arcn->name); 184 return(-1); 185 } 186 187 if (S_ISDIR(sb.st_mode)) { 188 paxwarn(1, "A hard link to the directory %s is not allowed", 189 arcn->ln_name); 190 return(-1); 191 } 192 193 return(mk_link(arcn->ln_name, &sb, arcn->name, 0)); 194 } 195 196 /* 197 * cross_lnk() 198 * Create a hard link to arcn->org_name from arcn->name. Only used in copy 199 * with the -l flag. No warning or error if this does not succeed (we will 200 * then just create the file) 201 * Return: 202 * 1 if copy() should try to create this file node 203 * 0 if cross_lnk() ok, -1 for fatal flaw (like linking to self). 204 */ 205 206 int 207 cross_lnk(ARCHD *arcn) 208 { 209 /* 210 * try to make a link to original file (-l flag in copy mode). make sure 211 * we do not try to link to directories in case we are running as root 212 * (and it might succeed). 213 */ 214 if (arcn->type == PAX_DIR) 215 return(1); 216 return(mk_link(arcn->org_name, &(arcn->sb), arcn->name, 1)); 217 } 218 219 /* 220 * chk_same() 221 * In copy mode if we are not trying to make hard links between the src 222 * and destinations, make sure we are not going to overwrite ourselves by 223 * accident. This slows things down a little, but we have to protect all 224 * those people who make typing errors. 225 * Return: 226 * 1 the target does not exist, go ahead and copy 227 * 0 skip it file exists (-k) or may be the same as source file 228 */ 229 230 int 231 chk_same(ARCHD *arcn) 232 { 233 struct stat sb; 234 235 /* 236 * if file does not exist, return. if file exists and -k, skip it 237 * quietly 238 */ 239 if (lstat(arcn->name, &sb) < 0) 240 return(1); 241 if (kflag) 242 return(0); 243 244 /* 245 * better make sure the user does not have src == dest by mistake 246 */ 247 if ((arcn->sb.st_dev == sb.st_dev) && (arcn->sb.st_ino == sb.st_ino)) { 248 paxwarn(1, "Unable to copy %s, file would overwrite itself", 249 arcn->name); 250 return(0); 251 } 252 return(1); 253 } 254 255 /* 256 * mk_link() 257 * try to make a hard link between two files. if ign set, we do not 258 * complain. 259 * Return: 260 * 0 if successful (or we are done with this file but no error, such as 261 * finding the from file exists and the user has set -k). 262 * 1 when ign was set to indicates we could not make the link but we 263 * should try to copy/extract the file as that might work (and is an 264 * allowed option). -1 an error occurred. 265 */ 266 267 static int 268 mk_link(char *to, struct stat *to_sb, char *from, 269 int ign) 270 { 271 struct stat sb; 272 int oerrno; 273 274 /* 275 * if from file exists, it has to be unlinked to make the link. If the 276 * file exists and -k is set, skip it quietly 277 */ 278 if (lstat(from, &sb) == 0) { 279 if (kflag) 280 return(0); 281 282 /* 283 * make sure it is not the same file, protect the user 284 */ 285 if ((to_sb->st_dev==sb.st_dev)&&(to_sb->st_ino == sb.st_ino)) { 286 paxwarn(1, "Unable to link file %s to itself", to); 287 return(-1); 288 } 289 290 /* 291 * try to get rid of the file, based on the type 292 */ 293 if (S_ISDIR(sb.st_mode)) { 294 if (rmdir(from) < 0) { 295 syswarn(1, errno, "Unable to remove %s", from); 296 return(-1); 297 } 298 } else if (unlink(from) < 0) { 299 if (!ign) { 300 syswarn(1, errno, "Unable to remove %s", from); 301 return(-1); 302 } 303 return(1); 304 } 305 } 306 307 /* 308 * from file is gone (or did not exist), try to make the hard link. 309 * if it fails, check the path and try it again (if chk_path() says to 310 * try again) 311 */ 312 for (;;) { 313 if (link(to, from) == 0) 314 break; 315 oerrno = errno; 316 if (!nodirs && chk_path(from, to_sb->st_uid, to_sb->st_gid) == 0) 317 continue; 318 if (!ign) { 319 syswarn(1, oerrno, "Could not link to %s from %s", to, 320 from); 321 return(-1); 322 } 323 return(1); 324 } 325 326 /* 327 * all right the link was made 328 */ 329 return(0); 330 } 331 332 /* 333 * node_creat() 334 * create an entry in the file system (other than a file or hard link). 335 * If successful, sets uid/gid modes and times as required. 336 * Return: 337 * 0 if ok, -1 otherwise 338 */ 339 340 int 341 node_creat(ARCHD *arcn) 342 { 343 int res; 344 int ign = 0; 345 int oerrno; 346 int pass = 0; 347 mode_t file_mode; 348 struct stat sb; 349 350 /* 351 * create node based on type, if that fails try to unlink the node and 352 * try again. finally check the path and try again. As noted in the 353 * file and link creation routines, this method seems to exhibit the 354 * best performance in general use workloads. 355 */ 356 file_mode = arcn->sb.st_mode & FILEBITS; 357 358 for (;;) { 359 switch(arcn->type) { 360 case PAX_DIR: 361 res = mkdir(arcn->name, file_mode); 362 if (ign) 363 res = 0; 364 break; 365 case PAX_CHR: 366 file_mode |= S_IFCHR; 367 res = mknod(arcn->name, file_mode, arcn->sb.st_rdev); 368 break; 369 case PAX_BLK: 370 file_mode |= S_IFBLK; 371 res = mknod(arcn->name, file_mode, arcn->sb.st_rdev); 372 break; 373 case PAX_FIF: 374 res = mkfifo(arcn->name, file_mode); 375 break; 376 case PAX_SCK: 377 /* 378 * Skip sockets, operation has no meaning under BSD 379 */ 380 paxwarn(0, 381 "%s skipped. Sockets cannot be copied or extracted", 382 arcn->name); 383 return(-1); 384 case PAX_SLK: 385 res = symlink(arcn->ln_name, arcn->name); 386 break; 387 case PAX_CTG: 388 case PAX_HLK: 389 case PAX_HRG: 390 case PAX_REG: 391 default: 392 /* 393 * we should never get here 394 */ 395 paxwarn(0, "%s has an unknown file type, skipping", 396 arcn->name); 397 return(-1); 398 } 399 400 /* 401 * if we were able to create the node break out of the loop, 402 * otherwise try to unlink the node and try again. if that 403 * fails check the full path and try a final time. 404 */ 405 if (res == 0) 406 break; 407 408 /* 409 * we failed to make the node 410 */ 411 oerrno = errno; 412 if ((ign = unlnk_exist(arcn->name, arcn->type)) < 0) 413 return(-1); 414 415 if (++pass <= 1) 416 continue; 417 418 if (nodirs || chk_path(arcn->name,arcn->sb.st_uid,arcn->sb.st_gid) < 0) { 419 syswarn(1, oerrno, "Could not create: %s", arcn->name); 420 return(-1); 421 } 422 } 423 424 /* 425 * we were able to create the node. set uid/gid, modes and times 426 */ 427 if (pids) 428 res = ((arcn->type == PAX_SLK) ? 429 set_lids(arcn->name, arcn->sb.st_uid, arcn->sb.st_gid) : 430 set_ids(arcn->name, arcn->sb.st_uid, arcn->sb.st_gid)); 431 else 432 res = 0; 433 434 /* 435 * symlinks are done now. 436 */ 437 if (arcn->type == PAX_SLK) 438 return(0); 439 440 /* 441 * IMPORTANT SECURITY NOTE: 442 * if not preserving mode or we cannot set uid/gid, then PROHIBIT any 443 * set uid/gid bits 444 */ 445 if (!pmode || res) 446 arcn->sb.st_mode &= ~(SETBITS); 447 if (pmode) 448 set_pmode(arcn->name, arcn->sb.st_mode); 449 450 if (arcn->type == PAX_DIR && strcmp(NM_CPIO, argv0) != 0) { 451 /* 452 * Dirs must be processed again at end of extract to set times 453 * and modes to agree with those stored in the archive. However 454 * to allow extract to continue, we may have to also set owner 455 * rights. This allows nodes in the archive that are children 456 * of this directory to be extracted without failure. Both time 457 * and modes will be fixed after the entire archive is read and 458 * before pax exits. 459 */ 460 if (access(arcn->name, R_OK | W_OK | X_OK) < 0) { 461 if (lstat(arcn->name, &sb) < 0) { 462 syswarn(0, errno,"Could not access %s (stat)", 463 arcn->name); 464 set_pmode(arcn->name,file_mode | S_IRWXU); 465 } else { 466 /* 467 * We have to add rights to the dir, so we make 468 * sure to restore the mode. The mode must be 469 * restored AS CREATED and not as stored if 470 * pmode is not set. 471 */ 472 set_pmode(arcn->name, 473 ((sb.st_mode & FILEBITS) | S_IRWXU)); 474 if (!pmode) 475 arcn->sb.st_mode = sb.st_mode; 476 } 477 478 /* 479 * we have to force the mode to what was set here, 480 * since we changed it from the default as created. 481 */ 482 add_dir(arcn->name, arcn->nlen, &(arcn->sb), 1); 483 } else if (pmode || patime || pmtime) 484 add_dir(arcn->name, arcn->nlen, &(arcn->sb), 0); 485 } 486 487 if (patime || pmtime) 488 set_ftime(arcn->name, arcn->sb.st_mtime, arcn->sb.st_atime, 0); 489 return(0); 490 } 491 492 /* 493 * unlnk_exist() 494 * Remove node from file system with the specified name. We pass the type 495 * of the node that is going to replace it. When we try to create a 496 * directory and find that it already exists, we allow processing to 497 * continue as proper modes etc will always be set for it later on. 498 * Return: 499 * 0 is ok to proceed, no file with the specified name exists 500 * -1 we were unable to remove the node, or we should not remove it (-k) 501 * 1 we found a directory and we were going to create a directory. 502 */ 503 504 int 505 unlnk_exist(char *name, int type) 506 { 507 struct stat sb; 508 509 /* 510 * the file does not exist, or -k we are done 511 */ 512 if (lstat(name, &sb) < 0) 513 return(0); 514 if (kflag) 515 return(-1); 516 517 if (S_ISDIR(sb.st_mode)) { 518 /* 519 * try to remove a directory, if it fails and we were going to 520 * create a directory anyway, tell the caller (return a 1) 521 */ 522 if (rmdir(name) < 0) { 523 if (type == PAX_DIR) 524 return(1); 525 syswarn(1,errno,"Unable to remove directory %s", name); 526 return(-1); 527 } 528 return(0); 529 } 530 531 /* 532 * try to get rid of all non-directory type nodes 533 */ 534 if (unlink(name) < 0) { 535 syswarn(1, errno, "Could not unlink %s", name); 536 return(-1); 537 } 538 return(0); 539 } 540 541 /* 542 * chk_path() 543 * We were trying to create some kind of node in the file system and it 544 * failed. chk_path() makes sure the path up to the node exists and is 545 * writeable. When we have to create a directory that is missing along the 546 * path somewhere, the directory we create will be set to the same 547 * uid/gid as the file has (when uid and gid are being preserved). 548 * NOTE: this routine is a real performance loss. It is only used as a 549 * last resort when trying to create entries in the file system. 550 * Return: 551 * -1 when it could find nothing it is allowed to fix. 552 * 0 otherwise 553 */ 554 555 int 556 chk_path( char *name, uid_t st_uid, gid_t st_gid) 557 { 558 char *spt = name; 559 struct stat sb; 560 int retval = -1; 561 562 /* 563 * watch out for paths with nodes stored directly in / (e.g. /bozo) 564 */ 565 if (*spt == '/') 566 ++spt; 567 568 for(;;) { 569 /* 570 * work foward from the first / and check each part of the path 571 */ 572 spt = strchr(spt, '/'); 573 if (spt == NULL) 574 break; 575 *spt = '\0'; 576 577 /* 578 * if it exists we assume it is a directory, it is not within 579 * the spec (at least it seems to read that way) to alter the 580 * file system for nodes NOT EXPLICITLY stored on the archive. 581 * If that assumption is changed, you would test the node here 582 * and figure out how to get rid of it (probably like some 583 * recursive unlink()) or fix up the directory permissions if 584 * required (do an access()). 585 */ 586 if (lstat(name, &sb) == 0) { 587 *(spt++) = '/'; 588 continue; 589 } 590 591 /* 592 * the path fails at this point, see if we can create the 593 * needed directory and continue on 594 */ 595 if (mkdir(name, S_IRWXU | S_IRWXG | S_IRWXO) < 0) { 596 *spt = '/'; 597 retval = -1; 598 break; 599 } 600 601 /* 602 * we were able to create the directory. We will tell the 603 * caller that we found something to fix, and it is ok to try 604 * and create the node again. 605 */ 606 retval = 0; 607 if (pids) 608 set_ids(name, st_uid, st_gid); 609 610 /* 611 * make sure the user doen't have some strange umask that 612 * causes this newly created directory to be unusable. We fix 613 * the modes and restore them back to the creation default at 614 * the end of pax 615 */ 616 if ((access(name, R_OK | W_OK | X_OK) < 0) && 617 (lstat(name, &sb) == 0)) { 618 set_pmode(name, ((sb.st_mode & FILEBITS) | S_IRWXU)); 619 add_dir(name, spt - name, &sb, 1); 620 } 621 *(spt++) = '/'; 622 continue; 623 } 624 return(retval); 625 } 626 627 /* 628 * set_ftime() 629 * Set the access time and modification time for a named file. If frc is 630 * non-zero we force these times to be set even if the user did not 631 * request access and/or modification time preservation (this is also 632 * used by -t to reset access times). 633 * When ign is zero, only those times the user has asked for are set, the 634 * other ones are left alone. We do not assume the un-documented feature 635 * of many utimes() implementations that consider a 0 time value as a do 636 * not set request. 637 */ 638 639 void 640 set_ftime(char *fnm, time_t mtime, time_t atime, int frc) 641 { 642 static struct timeval tv[2] = {{0L, 0L}, {0L, 0L}}; 643 struct stat sb; 644 645 tv[0].tv_sec = (long)atime; 646 tv[1].tv_sec = (long)mtime; 647 if (!frc && (!patime || !pmtime)) { 648 /* 649 * if we are not forcing, only set those times the user wants 650 * set. We get the current values of the times if we need them. 651 */ 652 if (lstat(fnm, &sb) == 0) { 653 if (!patime) 654 tv[0].tv_sec = (long)sb.st_atime; 655 if (!pmtime) 656 tv[1].tv_sec = (long)sb.st_mtime; 657 } else 658 syswarn(0,errno,"Unable to obtain file stats %s", fnm); 659 } 660 661 /* 662 * set the times 663 */ 664 if (utimes(fnm, tv) < 0) 665 syswarn(1, errno, "Access/modification time set failed on: %s", 666 fnm); 667 return; 668 } 669 670 /* 671 * set_ids() 672 * set the uid and gid of a file system node 673 * Return: 674 * 0 when set, -1 on failure 675 */ 676 677 int 678 set_ids(char *fnm, uid_t uid, gid_t gid) 679 { 680 if (chown(fnm, uid, gid) < 0) { 681 /* 682 * ignore EPERM unless in verbose mode or being run by root. 683 * if running as pax, POSIX requires a warning. 684 */ 685 if (strcmp(NM_PAX, argv0) == 0 || errno != EPERM || vflag || 686 geteuid() == 0) 687 syswarn(1, errno, "Unable to set file uid/gid of %s", 688 fnm); 689 return(-1); 690 } 691 return(0); 692 } 693 694 /* 695 * set_lids() 696 * set the uid and gid of a file system node 697 * Return: 698 * 0 when set, -1 on failure 699 */ 700 701 int 702 set_lids(char *fnm, uid_t uid, gid_t gid) 703 { 704 if (lchown(fnm, uid, gid) < 0) { 705 /* 706 * ignore EPERM unless in verbose mode or being run by root. 707 * if running as pax, POSIX requires a warning. 708 */ 709 if (strcmp(NM_PAX, argv0) == 0 || errno != EPERM || vflag || 710 geteuid() == 0) 711 syswarn(1, errno, "Unable to set file uid/gid of %s", 712 fnm); 713 return(-1); 714 } 715 return(0); 716 } 717 718 /* 719 * set_pmode() 720 * Set file access mode 721 */ 722 723 void 724 set_pmode(char *fnm, mode_t mode) 725 { 726 mode &= ABITS; 727 if (chmod(fnm, mode) < 0) 728 syswarn(1, errno, "Could not set permissions on %s", fnm); 729 return; 730 } 731 732 /* 733 * file_write() 734 * Write/copy a file (during copy or archive extract). This routine knows 735 * how to copy files with lseek holes in it. (Which are read as file 736 * blocks containing all 0's but do not have any file blocks associated 737 * with the data). Typical examples of these are files created by dbm 738 * variants (.pag files). While the file size of these files are huge, the 739 * actual storage is quite small (the files are sparse). The problem is 740 * the holes read as all zeros so are probably stored on the archive that 741 * way (there is no way to determine if the file block is really a hole, 742 * we only know that a file block of all zero's can be a hole). 743 * At this writing, no major archive format knows how to archive files 744 * with holes. However, on extraction (or during copy, -rw) we have to 745 * deal with these files. Without detecting the holes, the files can 746 * consume a lot of file space if just written to disk. This replacement 747 * for write when passed the basic allocation size of a file system block, 748 * uses lseek whenever it detects the input data is all 0 within that 749 * file block. In more detail, the strategy is as follows: 750 * While the input is all zero keep doing an lseek. Keep track of when we 751 * pass over file block boundries. Only write when we hit a non zero 752 * input. once we have written a file block, we continue to write it to 753 * the end (we stop looking at the input). When we reach the start of the 754 * next file block, start checking for zero blocks again. Working on file 755 * block boundries significantly reduces the overhead when copying files 756 * that are NOT very sparse. This overhead (when compared to a write) is 757 * almost below the measurement resolution on many systems. Without it, 758 * files with holes cannot be safely copied. It does has a side effect as 759 * it can put holes into files that did not have them before, but that is 760 * not a problem since the file contents are unchanged (in fact it saves 761 * file space). (Except on paging files for diskless clients. But since we 762 * cannot determine one of those file from here, we ignore them). If this 763 * ever ends up on a system where CTG files are supported and the holes 764 * are not desired, just do a conditional test in those routines that 765 * call file_write() and have it call write() instead. BEFORE CLOSING THE 766 * FILE, make sure to call file_flush() when the last write finishes with 767 * an empty block. A lot of file systems will not create an lseek hole at 768 * the end. In this case we drop a single 0 at the end to force the 769 * trailing 0's in the file. 770 * ---Parameters--- 771 * rem: how many bytes left in this file system block 772 * isempt: have we written to the file block yet (is it empty) 773 * sz: basic file block allocation size 774 * cnt: number of bytes on this write 775 * str: buffer to write 776 * Return: 777 * number of bytes written, -1 on write (or lseek) error. 778 */ 779 780 int 781 file_write(int fd, char *str, int cnt, int *rem, int *isempt, int sz, 782 char *name) 783 { 784 char *pt; 785 char *end; 786 int wcnt; 787 char *st = str; 788 789 /* 790 * while we have data to process 791 */ 792 while (cnt) { 793 if (!*rem) { 794 /* 795 * We are now at the start of file system block again 796 * (or what we think one is...). start looking for 797 * empty blocks again 798 */ 799 *isempt = 1; 800 *rem = sz; 801 } 802 803 /* 804 * only examine up to the end of the current file block or 805 * remaining characters to write, whatever is smaller 806 */ 807 wcnt = MIN(cnt, *rem); 808 cnt -= wcnt; 809 *rem -= wcnt; 810 if (*isempt) { 811 /* 812 * have not written to this block yet, so we keep 813 * looking for zero's 814 */ 815 pt = st; 816 end = st + wcnt; 817 818 /* 819 * look for a zero filled buffer 820 */ 821 while ((pt < end) && (*pt == '\0')) 822 ++pt; 823 824 if (pt == end) { 825 /* 826 * skip, buf is empty so far 827 */ 828 if (lseek(fd, (off_t)wcnt, SEEK_CUR) < 0) { 829 syswarn(1,errno,"File seek on %s", 830 name); 831 return(-1); 832 } 833 st = pt; 834 continue; 835 } 836 /* 837 * drat, the buf is not zero filled 838 */ 839 *isempt = 0; 840 } 841 842 /* 843 * have non-zero data in this file system block, have to write 844 */ 845 if (write(fd, st, wcnt) != wcnt) { 846 syswarn(1, errno, "Failed write to file %s", name); 847 return(-1); 848 } 849 st += wcnt; 850 } 851 return(st - str); 852 } 853 854 /* 855 * file_flush() 856 * when the last file block in a file is zero, many file systems will not 857 * let us create a hole at the end. To get the last block with zeros, we 858 * write the last BYTE with a zero (back up one byte and write a zero). 859 */ 860 861 void 862 file_flush(int fd, char *fname, int isempt) 863 { 864 static char blnk[] = "\0"; 865 866 /* 867 * silly test, but make sure we are only called when the last block is 868 * filled with all zeros. 869 */ 870 if (!isempt) 871 return; 872 873 /* 874 * move back one byte and write a zero 875 */ 876 if (lseek(fd, (off_t)-1, SEEK_CUR) < 0) { 877 syswarn(1, errno, "Failed seek on file %s", fname); 878 return; 879 } 880 881 if (write(fd, blnk, 1) < 0) 882 syswarn(1, errno, "Failed write to file %s", fname); 883 return; 884 } 885 886 /* 887 * rdfile_close() 888 * close a file we have beed reading (to copy or archive). If we have to 889 * reset access time (tflag) do so (the times are stored in arcn). 890 */ 891 892 void 893 rdfile_close(ARCHD *arcn, int *fd) 894 { 895 /* 896 * make sure the file is open 897 */ 898 if (*fd < 0) 899 return; 900 901 close(*fd); 902 *fd = -1; 903 if (!tflag) 904 return; 905 906 /* 907 * user wants last access time reset 908 */ 909 set_ftime(arcn->org_name, arcn->sb.st_mtime, arcn->sb.st_atime, 1); 910 return; 911 } 912 913 /* 914 * set_crc() 915 * read a file to calculate its crc. This is a real drag. Archive formats 916 * that have this, end up reading the file twice (we have to write the 917 * header WITH the crc before writing the file contents. Oh well... 918 * Return: 919 * 0 if was able to calculate the crc, -1 otherwise 920 */ 921 922 int 923 set_crc(ARCHD *arcn, int fd) 924 { 925 int i; 926 int res; 927 off_t cpcnt = 0L; 928 u_long size; 929 unsigned long crc = 0L; 930 char tbuf[FILEBLK]; 931 struct stat sb; 932 933 if (fd < 0) { 934 /* 935 * hmm, no fd, should never happen. well no crc then. 936 */ 937 arcn->crc = 0L; 938 return(0); 939 } 940 941 if ((size = (u_long)arcn->sb.st_blksize) > (u_long)sizeof(tbuf)) 942 size = (u_long)sizeof(tbuf); 943 944 /* 945 * read all the bytes we think that there are in the file. If the user 946 * is trying to archive an active file, forget this file. 947 */ 948 for(;;) { 949 if ((res = read(fd, tbuf, size)) <= 0) 950 break; 951 cpcnt += res; 952 for (i = 0; i < res; ++i) 953 crc += (tbuf[i] & 0xff); 954 } 955 956 /* 957 * safety check. we want to avoid archiving files that are active as 958 * they can create inconsistant archive copies. 959 */ 960 if (cpcnt != arcn->sb.st_size) 961 paxwarn(1, "File changed size %s", arcn->org_name); 962 else if (fstat(fd, &sb) < 0) 963 syswarn(1, errno, "Failed stat on %s", arcn->org_name); 964 else if (arcn->sb.st_mtime != sb.st_mtime) 965 paxwarn(1, "File %s was modified during read", arcn->org_name); 966 else if (lseek(fd, (off_t)0L, SEEK_SET) < 0) 967 syswarn(1, errno, "File rewind failed on: %s", arcn->org_name); 968 else { 969 arcn->crc = crc; 970 return(0); 971 } 972 return(-1); 973 } 974