file_subs.c revision 1.56 1 /* $NetBSD: file_subs.c,v 1.56 2005/09/16 16:48:18 christos 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
43 static char sccsid[] = "@(#)file_subs.c 8.1 (Berkeley) 5/31/93";
44 #else
45 __RCSID("$NetBSD: file_subs.c,v 1.56 2005/09/16 16:48:18 christos 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
65 char *xtmp_name;
66
67 static int
68 mk_link(char *,struct stat *,char *, int);
69
70 static int warn_broken;
71
72 /*
73 * routines that deal with file operations such as: creating, removing;
74 * and setting access modes, uid/gid and times of files
75 */
76 #define SET_BITS (S_ISUID | S_ISGID)
77 #define FILE_BITS (S_IRWXU | S_IRWXG | S_IRWXO)
78 #define A_BITS (FILE_BITS | SET_BITS | S_ISVTX)
79
80 /*
81 * The S_ISVTX (sticky bit) can be set by non-superuser on directories
82 * but not other kinds of files.
83 */
84 #define FILEBITS(dir) ((dir) ? (FILE_BITS | S_ISVTX) : FILE_BITS)
85 #define SETBITS(dir) ((dir) ? SET_BITS : (SET_BITS | S_ISVTX))
86
87 /*
88 * file_creat()
89 * Create and open a file.
90 * Return:
91 * file descriptor or -1 for failure
92 */
93
94 int
95 file_creat(ARCHD *arcn, int write_to_hardlink)
96 {
97 int fd = -1;
98 int oerrno;
99
100 /*
101 * Some horribly busted tar implementations, have directory nodes
102 * that end in a /, but they mark as files. Compensate for that
103 * by not creating a directory node at this point, but a file node,
104 * and not creating the temp file.
105 */
106 if (arcn->nlen != 0 && arcn->name[arcn->nlen - 1] == '/') {
107 if (!warn_broken) {
108 tty_warn(0, "Archive was created with a broken tar;"
109 " file `%s' is a directory, but marked as plain.",
110 arcn->name);
111 warn_broken = 1;
112 }
113 return -1;
114 }
115
116 /*
117 * In "cpio" archives it's usually the last record of a set of
118 * hardlinks which includes the contents of the file. We cannot
119 * use a tempory file in that case because we couldn't link it
120 * with the existing other hardlinks after restoring the contents
121 * to it. And it's also useless to create the hardlink under a
122 * temporary name because the other hardlinks would have partial
123 * contents while restoring.
124 */
125 if (write_to_hardlink)
126 return (open(arcn->name, O_TRUNC | O_EXCL | O_RDWR, 0));
127
128 /*
129 * Create a temporary file name so that the file doesn't have partial
130 * contents while restoring.
131 */
132 arcn->tmp_name = malloc(arcn->nlen + 8);
133 if (arcn->tmp_name == NULL) {
134 syswarn(1, errno, "Cannot malloc %d bytes", arcn->nlen + 8);
135 return(-1);
136 }
137 if (xtmp_name != NULL)
138 abort();
139 xtmp_name = arcn->tmp_name;
140
141 for (;;) {
142 /*
143 * try to create the temporary file we use to restore the
144 * contents info. if this fails, keep checking all the nodes
145 * in the path until chk_path() finds that it cannot fix
146 * anything further. if that happens we just give up.
147 */
148 (void)snprintf(arcn->tmp_name, arcn->nlen + 8, "%s.XXXXXX",
149 arcn->name);
150 fd = mkstemp(arcn->tmp_name);
151 if (fd >= 0)
152 break;
153 oerrno = errno;
154 if (nodirs || chk_path(arcn->name,arcn->sb.st_uid,arcn->sb.st_gid) < 0) {
155 (void)fflush(listf);
156 syswarn(1, oerrno, "Cannot create %s", arcn->tmp_name);
157 xtmp_name = NULL;
158 free(arcn->tmp_name);
159 arcn->tmp_name = NULL;
160 return(-1);
161 }
162 }
163 return(fd);
164 }
165
166 /*
167 * file_close()
168 * Close file descriptor to a file just created by pax. Sets modes,
169 * ownership and times as required.
170 * Return:
171 * 0 for success, -1 for failure
172 */
173
174 void
175 file_close(ARCHD *arcn, int fd)
176 {
177 char *tmp_name;
178 int res;
179
180 if (fd < 0)
181 return;
182
183 tmp_name = (arcn->tmp_name != NULL) ? arcn->tmp_name : arcn->name;
184
185 if (close(fd) < 0)
186 syswarn(0, errno, "Cannot close file descriptor on %s",
187 tmp_name);
188
189 /*
190 * set owner/groups first as this may strip off mode bits we want
191 * then set file permission modes. Then set file access and
192 * modification times.
193 */
194 if (pids)
195 res = set_ids(tmp_name, arcn->sb.st_uid, arcn->sb.st_gid);
196 else
197 res = 0;
198
199 /*
200 * IMPORTANT SECURITY NOTE:
201 * if not preserving mode or we cannot set uid/gid, then PROHIBIT
202 * set uid/gid bits but restore the file modes (since mkstemp doesn't).
203 */
204 if (!pmode || res)
205 arcn->sb.st_mode &= ~SETBITS(0);
206 if (pmode)
207 set_pmode(tmp_name, arcn->sb.st_mode);
208 else
209 set_pmode(tmp_name, arcn->sb.st_mode & FILEBITS(0));
210 if (patime || pmtime)
211 set_ftime(tmp_name, arcn->sb.st_mtime, arcn->sb.st_atime, 0);
212
213 /* Did we write directly to the target file? */
214 if (arcn->tmp_name == NULL)
215 return;
216
217 /*
218 * Finally, now the temp file is fully instantiated rename it to
219 * the desired file name.
220 */
221 if (rename(tmp_name, arcn->name) < 0) {
222 syswarn(0, errno, "Cannot rename %s to %s",
223 tmp_name, arcn->name);
224 (void)unlink(tmp_name);
225 }
226
227 #if HAVE_STRUCT_STAT_ST_FLAGS
228 if (pfflags && arcn->type != PAX_SLK)
229 set_chflags(arcn->name, arcn->sb.st_flags);
230 #endif
231
232 free(arcn->tmp_name);
233 arcn->tmp_name = NULL;
234 xtmp_name = NULL;
235 }
236
237 /*
238 * lnk_creat()
239 * Create a hard link to arcn->ln_name from arcn->name. arcn->ln_name
240 * must exist;
241 * Return:
242 * 0 if ok, -1 otherwise
243 */
244
245 int
246 lnk_creat(ARCHD *arcn, int *payload)
247 {
248 struct stat sb;
249
250 /*
251 * Check if this hardlink carries the "payload". In "cpio" archives
252 * it's usually the last record of a set of hardlinks which includes
253 * the contents of the file.
254 *
255 */
256 *payload = S_ISREG(arcn->sb.st_mode) &&
257 (arcn->sb.st_size > 0) && (arcn->sb.st_size <= arcn->skip);
258
259 /*
260 * We may be running as root, so we have to be sure that link target
261 * is not a directory, so we lstat and check
262 */
263 if (lstat(arcn->ln_name, &sb) < 0) {
264 syswarn(1, errno, "Cannot link to %s from %s", arcn->ln_name,
265 arcn->name);
266 return(-1);
267 }
268
269 if (S_ISDIR(sb.st_mode)) {
270 tty_warn(1, "A hard link to the directory %s is not allowed",
271 arcn->ln_name);
272 return(-1);
273 }
274
275 return(mk_link(arcn->ln_name, &sb, arcn->name, 0));
276 }
277
278 /*
279 * cross_lnk()
280 * Create a hard link to arcn->org_name from arcn->name. Only used in copy
281 * with the -l flag. No warning or error if this does not succeed (we will
282 * then just create the file)
283 * Return:
284 * 1 if copy() should try to create this file node
285 * 0 if cross_lnk() ok, -1 for fatal flaw (like linking to self).
286 */
287
288 int
289 cross_lnk(ARCHD *arcn)
290 {
291 /*
292 * try to make a link to original file (-l flag in copy mode). make
293 * sure we do not try to link to directories in case we are running as
294 * root (and it might succeed).
295 */
296 if (arcn->type == PAX_DIR)
297 return(1);
298 return(mk_link(arcn->org_name, &(arcn->sb), arcn->name, 1));
299 }
300
301 /*
302 * chk_same()
303 * In copy mode if we are not trying to make hard links between the src
304 * and destinations, make sure we are not going to overwrite ourselves by
305 * accident. This slows things down a little, but we have to protect all
306 * those people who make typing errors.
307 * Return:
308 * 1 the target does not exist, go ahead and copy
309 * 0 skip it file exists (-k) or may be the same as source file
310 */
311
312 int
313 chk_same(ARCHD *arcn)
314 {
315 struct stat sb;
316
317 /*
318 * if file does not exist, return. if file exists and -k, skip it
319 * quietly
320 */
321 if (lstat(arcn->name, &sb) < 0)
322 return(1);
323 if (kflag)
324 return(0);
325
326 /*
327 * better make sure the user does not have src == dest by mistake
328 */
329 if ((arcn->sb.st_dev == sb.st_dev) && (arcn->sb.st_ino == sb.st_ino)) {
330 tty_warn(1, "Unable to copy %s, file would overwrite itself",
331 arcn->name);
332 return(0);
333 }
334 return(1);
335 }
336
337 /*
338 * mk_link()
339 * try to make a hard link between two files. if ign set, we do not
340 * complain.
341 * Return:
342 * 0 if successful (or we are done with this file but no error, such as
343 * finding the from file exists and the user has set -k).
344 * 1 when ign was set to indicates we could not make the link but we
345 * should try to copy/extract the file as that might work (and is an
346 * allowed option). -1 an error occurred.
347 */
348
349 static int
350 mk_link(char *to, struct stat *to_sb, char *from, int ign)
351 {
352 struct stat sb;
353 int oerrno;
354
355 /*
356 * if from file exists, it has to be unlinked to make the link. If the
357 * file exists and -k is set, skip it quietly
358 */
359 if (lstat(from, &sb) == 0) {
360 if (kflag)
361 return(0);
362
363 /*
364 * make sure it is not the same file, protect the user
365 */
366 if ((to_sb->st_dev==sb.st_dev)&&(to_sb->st_ino == sb.st_ino)) {
367 tty_warn(1, "Cannot link file %s to itself", to);
368 return(-1);
369 }
370
371 /*
372 * try to get rid of the file, based on the type
373 */
374 if (S_ISDIR(sb.st_mode) && strcmp(from, ".") != 0) {
375 if (rmdir(from) < 0) {
376 syswarn(1, errno, "Cannot remove %s", from);
377 return(-1);
378 }
379 } else if (unlink(from) < 0) {
380 if (!ign) {
381 syswarn(1, errno, "Cannot remove %s", from);
382 return(-1);
383 }
384 return(1);
385 }
386 }
387
388 /*
389 * from file is gone (or did not exist), try to make the hard link.
390 * if it fails, check the path and try it again (if chk_path() says to
391 * try again)
392 */
393 for (;;) {
394 if (link(to, from) == 0)
395 break;
396 oerrno = errno;
397 if (chk_path(from, to_sb->st_uid, to_sb->st_gid) == 0)
398 continue;
399 if (!ign) {
400 syswarn(1, oerrno, "Cannot link to %s from %s", to,
401 from);
402 return(-1);
403 }
404 return(1);
405 }
406
407 /*
408 * all right the link was made
409 */
410 return(0);
411 }
412
413 /*
414 * node_creat()
415 * create an entry in the file system (other than a file or hard link).
416 * If successful, sets uid/gid modes and times as required.
417 * Return:
418 * 0 if ok, -1 otherwise
419 */
420
421 int
422 node_creat(ARCHD *arcn)
423 {
424 int res;
425 int ign = 0;
426 int oerrno;
427 int pass = 0;
428 mode_t file_mode;
429 struct stat sb;
430 char target[MAXPATHLEN];
431 char *nm = arcn->name;
432 int len;
433
434 /*
435 * create node based on type, if that fails try to unlink the node and
436 * try again. finally check the path and try again. As noted in the
437 * file and link creation routines, this method seems to exhibit the
438 * best performance in general use workloads.
439 */
440 file_mode = arcn->sb.st_mode & FILEBITS(arcn->type == PAX_DIR);
441
442 for (;;) {
443 switch (arcn->type) {
444 case PAX_DIR:
445 /*
446 * If -h (or -L) was given in tar-mode, follow the
447 * potential symlink chain before trying to create the
448 * directory.
449 */
450 if (strcmp(NM_TAR, argv0) == 0 && Lflag) {
451 while (lstat(nm, &sb) == 0 &&
452 S_ISLNK(sb.st_mode)) {
453 len = readlink(nm, target,
454 sizeof target - 1);
455 if (len == -1) {
456 syswarn(0, errno,
457 "cannot follow symlink %s "
458 "in chain for %s",
459 nm, arcn->name);
460 res = -1;
461 goto badlink;
462 }
463 target[len] = '\0';
464 nm = target;
465 }
466 }
467 res = domkdir(nm, file_mode);
468 badlink:
469 if (ign)
470 res = 0;
471 break;
472 case PAX_CHR:
473 file_mode |= S_IFCHR;
474 res = mknod(nm, file_mode, arcn->sb.st_rdev);
475 break;
476 case PAX_BLK:
477 file_mode |= S_IFBLK;
478 res = mknod(nm, file_mode, arcn->sb.st_rdev);
479 break;
480 case PAX_FIF:
481 res = mkfifo(nm, file_mode);
482 break;
483 case PAX_SCK:
484 /*
485 * Skip sockets, operation has no meaning under BSD
486 */
487 tty_warn(0,
488 "%s skipped. Sockets cannot be copied or extracted",
489 nm);
490 return (-1);
491 case PAX_SLK:
492 res = symlink(arcn->ln_name, nm);
493 break;
494 case PAX_CTG:
495 case PAX_HLK:
496 case PAX_HRG:
497 case PAX_REG:
498 default:
499 /*
500 * we should never get here
501 */
502 tty_warn(0, "%s has an unknown file type, skipping",
503 nm);
504 return (-1);
505 }
506
507 /*
508 * if we were able to create the node break out of the loop,
509 * otherwise try to unlink the node and try again. if that
510 * fails check the full path and try a final time.
511 */
512 if (res == 0)
513 break;
514
515 /*
516 * we failed to make the node
517 */
518 oerrno = errno;
519 switch (pass++) {
520 case 0:
521 if ((ign = unlnk_exist(nm, arcn->type)) < 0)
522 return (-1);
523 continue;
524
525 case 1:
526 if (nodirs ||
527 chk_path(nm, arcn->sb.st_uid,
528 arcn->sb.st_gid) < 0) {
529 syswarn(1, oerrno, "Cannot create %s", nm);
530 return (-1);
531 }
532 continue;
533 }
534
535 /*
536 * it must be a file that exists but we can't create or
537 * remove, but we must avoid the infinite loop.
538 */
539 break;
540 }
541
542 /*
543 * we were able to create the node. set uid/gid, modes and times
544 */
545 if (pids)
546 res = set_ids(nm, arcn->sb.st_uid, arcn->sb.st_gid);
547 else
548 res = 0;
549
550 /*
551 * IMPORTANT SECURITY NOTE:
552 * if not preserving mode or we cannot set uid/gid, then PROHIBIT any
553 * set uid/gid bits
554 */
555 if (!pmode || res)
556 arcn->sb.st_mode &= ~SETBITS(arcn->type == PAX_DIR);
557 if (pmode)
558 set_pmode(arcn->name, arcn->sb.st_mode);
559
560 if (arcn->type == PAX_DIR && strcmp(NM_CPIO, argv0) != 0) {
561 /*
562 * Dirs must be processed again at end of extract to set times
563 * and modes to agree with those stored in the archive. However
564 * to allow extract to continue, we may have to also set owner
565 * rights. This allows nodes in the archive that are children
566 * of this directory to be extracted without failure. Both time
567 * and modes will be fixed after the entire archive is read and
568 * before pax exits.
569 */
570 if (access(nm, R_OK | W_OK | X_OK) < 0) {
571 if (lstat(nm, &sb) < 0) {
572 syswarn(0, errno,"Cannot access %s (stat)",
573 arcn->name);
574 set_pmode(nm,file_mode | S_IRWXU);
575 } else {
576 /*
577 * We have to add rights to the dir, so we make
578 * sure to restore the mode. The mode must be
579 * restored AS CREATED and not as stored if
580 * pmode is not set.
581 */
582 set_pmode(nm, ((sb.st_mode &
583 FILEBITS(arcn->type == PAX_DIR)) |
584 S_IRWXU));
585 if (!pmode)
586 arcn->sb.st_mode = sb.st_mode;
587 }
588
589 /*
590 * we have to force the mode to what was set here,
591 * since we changed it from the default as created.
592 */
593 add_dir(nm, arcn->nlen, &(arcn->sb), 1);
594 } else if (pmode || patime || pmtime)
595 add_dir(nm, arcn->nlen, &(arcn->sb), 0);
596 }
597
598 #if HAVE_LUTIMES
599 if (patime || pmtime)
600 #else
601 if ((patime || pmtime) && arcn->type != PAX_SLK)
602 #endif
603 set_ftime(arcn->name, arcn->sb.st_mtime, arcn->sb.st_atime, 0);
604
605 #if HAVE_STRUCT_STAT_ST_FLAGS
606 if (pfflags && arcn->type != PAX_SLK)
607 set_chflags(arcn->name, arcn->sb.st_flags);
608 #endif
609 return(0);
610 }
611
612 /*
613 * unlnk_exist()
614 * Remove node from file system with the specified name. We pass the type
615 * of the node that is going to replace it. When we try to create a
616 * directory and find that it already exists, we allow processing to
617 * continue as proper modes etc will always be set for it later on.
618 * Return:
619 * 0 is ok to proceed, no file with the specified name exists
620 * -1 we were unable to remove the node, or we should not remove it (-k)
621 * 1 we found a directory and we were going to create a directory.
622 */
623
624 int
625 unlnk_exist(char *name, int type)
626 {
627 struct stat sb;
628
629 /*
630 * the file does not exist, or -k we are done
631 */
632 if (lstat(name, &sb) < 0)
633 return(0);
634 if (kflag)
635 return(-1);
636
637 if (S_ISDIR(sb.st_mode)) {
638 /*
639 * try to remove a directory, if it fails and we were going to
640 * create a directory anyway, tell the caller (return a 1).
641 *
642 * don't try to remove the directory if the name is "."
643 * otherwise later file/directory creation fails.
644 */
645 if (strcmp(name, ".") == 0)
646 return(1);
647 if (rmdir(name) < 0) {
648 if (type == PAX_DIR)
649 return(1);
650 syswarn(1, errno, "Cannot remove directory %s", name);
651 return(-1);
652 }
653 return(0);
654 }
655
656 /*
657 * try to get rid of all non-directory type nodes
658 */
659 if (unlink(name) < 0) {
660 (void)fflush(listf);
661 syswarn(1, errno, "Cannot unlink %s", name);
662 return(-1);
663 }
664 return(0);
665 }
666
667 /*
668 * chk_path()
669 * We were trying to create some kind of node in the file system and it
670 * failed. chk_path() makes sure the path up to the node exists and is
671 * writable. When we have to create a directory that is missing along the
672 * path somewhere, the directory we create will be set to the same
673 * uid/gid as the file has (when uid and gid are being preserved).
674 * NOTE: this routine is a real performance loss. It is only used as a
675 * last resort when trying to create entries in the file system.
676 * Return:
677 * -1 when it could find nothing it is allowed to fix.
678 * 0 otherwise
679 */
680
681 int
682 chk_path(char *name, uid_t st_uid, gid_t st_gid)
683 {
684 char *spt = name;
685 struct stat sb;
686 int retval = -1;
687
688 /*
689 * watch out for paths with nodes stored directly in / (e.g. /bozo)
690 */
691 if (*spt == '/')
692 ++spt;
693
694 for(;;) {
695 /*
696 * work forward from the first / and check each part of
697 * the path
698 */
699 spt = strchr(spt, '/');
700 if (spt == NULL)
701 break;
702 *spt = '\0';
703
704 /*
705 * if it exists we assume it is a directory, it is not within
706 * the spec (at least it seems to read that way) to alter the
707 * file system for nodes NOT EXPLICITLY stored on the archive.
708 * If that assumption is changed, you would test the node here
709 * and figure out how to get rid of it (probably like some
710 * recursive unlink()) or fix up the directory permissions if
711 * required (do an access()).
712 */
713 if (lstat(name, &sb) == 0) {
714 *(spt++) = '/';
715 continue;
716 }
717
718 /*
719 * the path fails at this point, see if we can create the
720 * needed directory and continue on
721 */
722 if (domkdir(name, S_IRWXU | S_IRWXG | S_IRWXO) == -1) {
723 *spt = '/';
724 retval = -1;
725 break;
726 }
727
728 /*
729 * we were able to create the directory. We will tell the
730 * caller that we found something to fix, and it is ok to try
731 * and create the node again.
732 */
733 retval = 0;
734 if (pids)
735 (void)set_ids(name, st_uid, st_gid);
736
737 /*
738 * make sure the user doesn't have some strange umask that
739 * causes this newly created directory to be unusable. We fix
740 * the modes and restore them back to the creation default at
741 * the end of pax
742 */
743 if ((access(name, R_OK | W_OK | X_OK) < 0) &&
744 (lstat(name, &sb) == 0)) {
745 set_pmode(name, ((sb.st_mode & FILEBITS(0)) |
746 S_IRWXU));
747 add_dir(name, spt - name, &sb, 1);
748 }
749 *(spt++) = '/';
750 continue;
751 }
752 return(retval);
753 }
754
755 /*
756 * set_ftime()
757 * Set the access time and modification time for a named file. If frc
758 * is non-zero we force these times to be set even if the user did not
759 * request access and/or modification time preservation (this is also
760 * used by -t to reset access times).
761 * When ign is zero, only those times the user has asked for are set, the
762 * other ones are left alone. We do not assume the un-documented feature
763 * of many utimes() implementations that consider a 0 time value as a do
764 * not set request.
765 */
766
767 void
768 set_ftime(char *fnm, time_t mtime, time_t atime, int frc)
769 {
770 struct timeval tv[2];
771 struct stat sb;
772
773 tv[0].tv_sec = (long)atime;
774 tv[0].tv_usec = 0;
775 tv[1].tv_sec = (long)mtime;
776 tv[1].tv_usec = 0;
777 if (!frc && (!patime || !pmtime)) {
778 /*
779 * if we are not forcing, only set those times the user wants
780 * set. We get the current values of the times if we need them.
781 */
782 if (lstat(fnm, &sb) == 0) {
783 #if BSD4_4 && !HAVE_NBTOOL_CONFIG_H
784 if (!patime)
785 TIMESPEC_TO_TIMEVAL(&tv[0], &sb.st_atimespec);
786 if (!pmtime)
787 TIMESPEC_TO_TIMEVAL(&tv[1], &sb.st_mtimespec);
788 #else
789 if (!patime)
790 tv[0].tv_sec = sb.st_atime;
791 if (!pmtime)
792 tv[1].tv_sec = sb.st_mtime;
793 #endif
794 } else
795 syswarn(0, errno, "Cannot obtain file stats %s", fnm);
796 }
797
798 /*
799 * set the times
800 */
801 #if HAVE_LUTIMES
802 if (lutimes(fnm, tv))
803 #else
804 if (utimes(fnm, tv))
805 #endif
806 syswarn(1, errno, "Access/modification time set failed on: %s",
807 fnm);
808 return;
809 }
810
811 /*
812 * set_ids()
813 * set the uid and gid of a file system node
814 * Return:
815 * 0 when set, -1 on failure
816 */
817
818 int
819 set_ids(char *fnm, uid_t uid, gid_t gid)
820 {
821 if (geteuid() == 0)
822 if (lchown(fnm, uid, gid)) {
823 (void)fflush(listf);
824 syswarn(1, errno, "Cannot set file uid/gid of %s",
825 fnm);
826 return(-1);
827 }
828 return(0);
829 }
830
831 /*
832 * set_pmode()
833 * Set file access mode
834 */
835
836 void
837 set_pmode(char *fnm, mode_t mode)
838 {
839 mode &= A_BITS;
840 if (lchmod(fnm, mode)) {
841 (void)fflush(listf);
842 syswarn(1, errno, "Cannot set permissions on %s", fnm);
843 }
844 return;
845 }
846
847 /*
848 * set_chflags()
849 * Set 4.4BSD file flags
850 */
851 void
852 set_chflags(char *fnm, u_int32_t flags)
853 {
854
855 #if 0
856 if (chflags(fnm, flags) < 0 && errno != EOPNOTSUPP)
857 syswarn(1, errno, "Cannot set file flags on %s", fnm);
858 #endif
859 return;
860 }
861
862 /*
863 * file_write()
864 * Write/copy a file (during copy or archive extract). This routine knows
865 * how to copy files with lseek holes in it. (Which are read as file
866 * blocks containing all 0's but do not have any file blocks associated
867 * with the data). Typical examples of these are files created by dbm
868 * variants (.pag files). While the file size of these files are huge, the
869 * actual storage is quite small (the files are sparse). The problem is
870 * the holes read as all zeros so are probably stored on the archive that
871 * way (there is no way to determine if the file block is really a hole,
872 * we only know that a file block of all zero's can be a hole).
873 * At this writing, no major archive format knows how to archive files
874 * with holes. However, on extraction (or during copy, -rw) we have to
875 * deal with these files. Without detecting the holes, the files can
876 * consume a lot of file space if just written to disk. This replacement
877 * for write when passed the basic allocation size of a file system block,
878 * uses lseek whenever it detects the input data is all 0 within that
879 * file block. In more detail, the strategy is as follows:
880 * While the input is all zero keep doing an lseek. Keep track of when we
881 * pass over file block boundaries. Only write when we hit a non zero
882 * input. once we have written a file block, we continue to write it to
883 * the end (we stop looking at the input). When we reach the start of the
884 * next file block, start checking for zero blocks again. Working on file
885 * block boundaries significantly reduces the overhead when copying files
886 * that are NOT very sparse. This overhead (when compared to a write) is
887 * almost below the measurement resolution on many systems. Without it,
888 * files with holes cannot be safely copied. It does has a side effect as
889 * it can put holes into files that did not have them before, but that is
890 * not a problem since the file contents are unchanged (in fact it saves
891 * file space). (Except on paging files for diskless clients. But since we
892 * cannot determine one of those file from here, we ignore them). If this
893 * ever ends up on a system where CTG files are supported and the holes
894 * are not desired, just do a conditional test in those routines that
895 * call file_write() and have it call write() instead. BEFORE CLOSING THE
896 * FILE, make sure to call file_flush() when the last write finishes with
897 * an empty block. A lot of file systems will not create an lseek hole at
898 * the end. In this case we drop a single 0 at the end to force the
899 * trailing 0's in the file.
900 * ---Parameters---
901 * rem: how many bytes left in this file system block
902 * isempt: have we written to the file block yet (is it empty)
903 * sz: basic file block allocation size
904 * cnt: number of bytes on this write
905 * str: buffer to write
906 * Return:
907 * number of bytes written, -1 on write (or lseek) error.
908 */
909
910 int
911 file_write(int fd, char *str, int cnt, int *rem, int *isempt, int sz,
912 char *name)
913 {
914 char *pt;
915 char *end;
916 int wcnt;
917 char *st = str;
918 char **strp;
919 size_t *lenp;
920
921 /*
922 * while we have data to process
923 */
924 while (cnt) {
925 if (!*rem) {
926 /*
927 * We are now at the start of file system block again
928 * (or what we think one is...). start looking for
929 * empty blocks again
930 */
931 *isempt = 1;
932 *rem = sz;
933 }
934
935 /*
936 * only examine up to the end of the current file block or
937 * remaining characters to write, whatever is smaller
938 */
939 wcnt = MIN(cnt, *rem);
940 cnt -= wcnt;
941 *rem -= wcnt;
942 if (*isempt) {
943 /*
944 * have not written to this block yet, so we keep
945 * looking for zero's
946 */
947 pt = st;
948 end = st + wcnt;
949
950 /*
951 * look for a zero filled buffer
952 */
953 while ((pt < end) && (*pt == '\0'))
954 ++pt;
955
956 if (pt == end) {
957 /*
958 * skip, buf is empty so far
959 */
960 if (fd > -1 &&
961 lseek(fd, (off_t)wcnt, SEEK_CUR) < 0) {
962 syswarn(1, errno, "File seek on %s",
963 name);
964 return(-1);
965 }
966 st = pt;
967 continue;
968 }
969 /*
970 * drat, the buf is not zero filled
971 */
972 *isempt = 0;
973 }
974
975 /*
976 * have non-zero data in this file system block, have to write
977 */
978 switch (fd) {
979 case -PAX_GLF:
980 strp = &gnu_name_string;
981 lenp = &gnu_name_length;
982 break;
983 case -PAX_GLL:
984 strp = &gnu_link_string;
985 lenp = &gnu_link_length;
986 break;
987 default:
988 strp = NULL;
989 lenp = NULL;
990 break;
991 }
992 if (strp) {
993 char *nstr = *strp ? realloc(*strp, *lenp + wcnt + 1) :
994 malloc(wcnt + 1);
995 if (nstr == NULL) {
996 tty_warn(1, "Out of memory");
997 return(-1);
998 }
999 (void)strlcpy(&nstr[*lenp], st, wcnt + 1);
1000 *strp = nstr;
1001 *lenp += wcnt;
1002 } else if (xwrite(fd, st, wcnt) != wcnt) {
1003 syswarn(1, errno, "Failed write to file %s", name);
1004 return(-1);
1005 }
1006 st += wcnt;
1007 }
1008 return(st - str);
1009 }
1010
1011 /*
1012 * file_flush()
1013 * when the last file block in a file is zero, many file systems will not
1014 * let us create a hole at the end. To get the last block with zeros, we
1015 * write the last BYTE with a zero (back up one byte and write a zero).
1016 */
1017
1018 void
1019 file_flush(int fd, char *fname, int isempt)
1020 {
1021 static char blnk[] = "\0";
1022
1023 /*
1024 * silly test, but make sure we are only called when the last block is
1025 * filled with all zeros.
1026 */
1027 if (!isempt)
1028 return;
1029
1030 /*
1031 * move back one byte and write a zero
1032 */
1033 if (lseek(fd, (off_t)-1, SEEK_CUR) < 0) {
1034 syswarn(1, errno, "Failed seek on file %s", fname);
1035 return;
1036 }
1037
1038 if (write_with_restart(fd, blnk, 1) < 0)
1039 syswarn(1, errno, "Failed write to file %s", fname);
1040 return;
1041 }
1042
1043 /*
1044 * rdfile_close()
1045 * close a file we have been reading (to copy or archive). If we have to
1046 * reset access time (tflag) do so (the times are stored in arcn).
1047 */
1048
1049 void
1050 rdfile_close(ARCHD *arcn, int *fd)
1051 {
1052 /*
1053 * make sure the file is open
1054 */
1055 if (*fd < 0)
1056 return;
1057
1058 (void)close(*fd);
1059 *fd = -1;
1060 if (!tflag)
1061 return;
1062
1063 /*
1064 * user wants last access time reset
1065 */
1066 set_ftime(arcn->org_name, arcn->sb.st_mtime, arcn->sb.st_atime, 1);
1067 return;
1068 }
1069
1070 /*
1071 * set_crc()
1072 * read a file to calculate its crc. This is a real drag. Archive formats
1073 * that have this, end up reading the file twice (we have to write the
1074 * header WITH the crc before writing the file contents. Oh well...
1075 * Return:
1076 * 0 if was able to calculate the crc, -1 otherwise
1077 */
1078
1079 int
1080 set_crc(ARCHD *arcn, int fd)
1081 {
1082 int i;
1083 int res;
1084 off_t cpcnt = 0L;
1085 u_long size;
1086 unsigned long crc = 0L;
1087 char tbuf[FILEBLK];
1088 struct stat sb;
1089
1090 if (fd < 0) {
1091 /*
1092 * hmm, no fd, should never happen. well no crc then.
1093 */
1094 arcn->crc = 0L;
1095 return(0);
1096 }
1097
1098 if ((size = (u_long)arcn->sb.st_blksize) > (u_long)sizeof(tbuf))
1099 size = (u_long)sizeof(tbuf);
1100
1101 /*
1102 * read all the bytes we think that there are in the file. If the user
1103 * is trying to archive an active file, forget this file.
1104 */
1105 for(;;) {
1106 if ((res = read(fd, tbuf, size)) <= 0)
1107 break;
1108 cpcnt += res;
1109 for (i = 0; i < res; ++i)
1110 crc += (tbuf[i] & 0xff);
1111 }
1112
1113 /*
1114 * safety check. we want to avoid archiving files that are active as
1115 * they can create inconsistant archive copies.
1116 */
1117 if (cpcnt != arcn->sb.st_size)
1118 tty_warn(1, "File changed size %s", arcn->org_name);
1119 else if (fstat(fd, &sb) < 0)
1120 syswarn(1, errno, "Failed stat on %s", arcn->org_name);
1121 else if (arcn->sb.st_mtime != sb.st_mtime)
1122 tty_warn(1, "File %s was modified during read", arcn->org_name);
1123 else if (lseek(fd, (off_t)0L, SEEK_SET) < 0)
1124 syswarn(1, errno, "File rewind failed on: %s", arcn->org_name);
1125 else {
1126 arcn->crc = crc;
1127 return(0);
1128 }
1129 return(-1);
1130 }
1131