file_cmds-46.tar.gz
[apple/file_cmds.git] / pax / file_subs.c
1 /* $OpenBSD: file_subs.c,v 1.13 1997/09/01 18:29:48 deraadt Exp $ */
2 /* $NetBSD: file_subs.c,v 1.4 1995/03/21 09:07:18 cgd Exp $ */
3
4 /*-
5 * Copyright (c) 1992 Keith Muller.
6 * Copyright (c) 1992, 1993
7 * The Regents of the University of California. All rights reserved.
8 *
9 * This code is derived from software contributed to Berkeley by
10 * Keith Muller of the University of California, San Diego.
11 *
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
14 * are met:
15 * 1. Redistributions of source code must retain the above copyright
16 * notice, this list of conditions and the following disclaimer.
17 * 2. Redistributions in binary form must reproduce the above copyright
18 * notice, this list of conditions and the following disclaimer in the
19 * documentation and/or other materials provided with the distribution.
20 * 3. All advertising materials mentioning features or use of this software
21 * must display the following acknowledgement:
22 * This product includes software developed by the University of
23 * California, Berkeley and its contributors.
24 * 4. Neither the name of the University nor the names of its contributors
25 * may be used to endorse or promote products derived from this software
26 * without specific prior written permission.
27 *
28 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
29 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
30 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
31 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
32 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
33 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
34 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
35 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
36 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
37 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
38 * SUCH DAMAGE.
39 */
40
41 #ifndef lint
42 #if 0
43 static char sccsid[] = "@(#)file_subs.c 8.1 (Berkeley) 5/31/93";
44 #else
45 static char rcsid[] __attribute__((__unused__)) = "$OpenBSD: file_subs.c,v 1.13 1997/09/01 18:29:48 deraadt 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 <errno.h>
58 #include <sys/uio.h>
59 #include <stdlib.h>
60 #include "pax.h"
61 #include "options.h"
62 #include "extern.h"
63
64 static int
65 mk_link __P((register char *,register struct stat *,register char *, int));
66
67 /*
68 * routines that deal with file operations such as: creating, removing;
69 * and setting access modes, uid/gid and times of files
70 */
71
72 #define FILEBITS (S_ISVTX | S_IRWXU | S_IRWXG | S_IRWXO)
73 #define SETBITS (S_ISUID | S_ISGID)
74 #define ABITS (FILEBITS | SETBITS)
75
76 /*
77 * file_creat()
78 * Create and open a file.
79 * Return:
80 * file descriptor or -1 for failure
81 */
82
83 #ifdef __STDC__
84 int
85 file_creat(register ARCHD *arcn)
86 #else
87 int
88 file_creat(arcn)
89 register ARCHD *arcn;
90 #endif
91 {
92 int fd = -1;
93 mode_t file_mode;
94 int oerrno;
95
96 /*
97 * assume file doesn't exist, so just try to create it, most times this
98 * works. We have to take special handling when the file does exist. To
99 * detect this, we use O_EXCL. For example when trying to create a
100 * file and a character device or fifo exists with the same name, we
101 * can accidently open the device by mistake (or block waiting to open)
102 * If we find that the open has failed, then figure spend the effore to
103 * figure out why. This strategy was found to have better average
104 * performance in common use than checking the file (and the path)
105 * first with lstat.
106 */
107 file_mode = arcn->sb.st_mode & FILEBITS;
108 if ((fd = open(arcn->name, O_WRONLY | O_CREAT | O_TRUNC | O_EXCL,
109 file_mode)) >= 0)
110 return(fd);
111
112 /*
113 * the file seems to exist. First we try to get rid of it (found to be
114 * the second most common failure when traced). If this fails, only
115 * then we go to the expense to check and create the path to the file
116 */
117 if (unlnk_exist(arcn->name, arcn->type) != 0)
118 return(-1);
119
120 for (;;) {
121 /*
122 * try to open it again, if this fails, check all the nodes in
123 * the path and give it a final try. if chk_path() finds that
124 * it cannot fix anything, we will skip the last attempt
125 */
126 if ((fd = open(arcn->name, O_WRONLY | O_CREAT | O_TRUNC,
127 file_mode)) >= 0)
128 break;
129 oerrno = errno;
130 if (nodirs || chk_path(arcn->name,arcn->sb.st_uid,arcn->sb.st_gid) < 0) {
131 syswarn(1, oerrno, "Unable to create %s", arcn->name);
132 return(-1);
133 }
134 }
135 return(fd);
136 }
137
138 /*
139 * file_close()
140 * Close file descriptor to a file just created by pax. Sets modes,
141 * ownership and times as required.
142 * Return:
143 * 0 for success, -1 for failure
144 */
145
146 #ifdef __STDC__
147 void
148 file_close(register ARCHD *arcn, int fd)
149 #else
150 void
151 file_close(arcn, fd)
152 register ARCHD *arcn;
153 int fd;
154 #endif
155 {
156 int res = 0;
157
158 if (fd < 0)
159 return;
160 if (close(fd) < 0)
161 syswarn(0, errno, "Unable to close file descriptor on %s",
162 arcn->name);
163
164 /*
165 * set owner/groups first as this may strip off mode bits we want
166 * then set file permission modes. Then set file access and
167 * modification times.
168 */
169 if (pids)
170 res = set_ids(arcn->name, arcn->sb.st_uid, arcn->sb.st_gid);
171
172 /*
173 * IMPORTANT SECURITY NOTE:
174 * if not preserving mode or we cannot set uid/gid, then PROHIBIT
175 * set uid/gid bits
176 */
177 if (!pmode || res)
178 arcn->sb.st_mode &= ~(SETBITS);
179 if (pmode)
180 set_pmode(arcn->name, arcn->sb.st_mode);
181 if (patime || pmtime)
182 set_ftime(arcn->name, arcn->sb.st_mtime, arcn->sb.st_atime, 0);
183 }
184
185 /*
186 * lnk_creat()
187 * Create a hard link to arcn->ln_name from arcn->name. arcn->ln_name
188 * must exist;
189 * Return:
190 * 0 if ok, -1 otherwise
191 */
192
193 #ifdef __STDC__
194 int
195 lnk_creat(register ARCHD *arcn)
196 #else
197 int
198 lnk_creat(arcn)
199 register ARCHD *arcn;
200 #endif
201 {
202 struct stat sb;
203
204 /*
205 * we may be running as root, so we have to be sure that link target
206 * is not a directory, so we lstat and check
207 */
208 if (lstat(arcn->ln_name, &sb) < 0) {
209 syswarn(1,errno,"Unable to link to %s from %s", arcn->ln_name,
210 arcn->name);
211 return(-1);
212 }
213
214 if (S_ISDIR(sb.st_mode)) {
215 paxwarn(1, "A hard link to the directory %s is not allowed",
216 arcn->ln_name);
217 return(-1);
218 }
219
220 return(mk_link(arcn->ln_name, &sb, arcn->name, 0));
221 }
222
223 /*
224 * cross_lnk()
225 * Create a hard link to arcn->org_name from arcn->name. Only used in copy
226 * with the -l flag. No warning or error if this does not succeed (we will
227 * then just create the file)
228 * Return:
229 * 1 if copy() should try to create this file node
230 * 0 if cross_lnk() ok, -1 for fatal flaw (like linking to self).
231 */
232
233 #ifdef __STDC__
234 int
235 cross_lnk(register ARCHD *arcn)
236 #else
237 int
238 cross_lnk(arcn)
239 register ARCHD *arcn;
240 #endif
241 {
242 /*
243 * try to make a link to orginal file (-l flag in copy mode). make sure
244 * we do not try to link to directories in case we are running as root
245 * (and it might succeed).
246 */
247 if (arcn->type == PAX_DIR)
248 return(1);
249 return(mk_link(arcn->org_name, &(arcn->sb), arcn->name, 1));
250 }
251
252 /*
253 * chk_same()
254 * In copy mode if we are not trying to make hard links between the src
255 * and destinations, make sure we are not going to overwrite ourselves by
256 * accident. This slows things down a little, but we have to protect all
257 * those people who make typing errors.
258 * Return:
259 * 1 the target does not exist, go ahead and copy
260 * 0 skip it file exists (-k) or may be the same as source file
261 */
262
263 #ifdef __STDC__
264 int
265 chk_same(register ARCHD *arcn)
266 #else
267 int
268 chk_same(arcn)
269 register ARCHD *arcn;
270 #endif
271 {
272 struct stat sb;
273
274 /*
275 * if file does not exist, return. if file exists and -k, skip it
276 * quietly
277 */
278 if (lstat(arcn->name, &sb) < 0)
279 return(1);
280 if (kflag)
281 return(0);
282
283 /*
284 * better make sure the user does not have src == dest by mistake
285 */
286 if ((arcn->sb.st_dev == sb.st_dev) && (arcn->sb.st_ino == sb.st_ino)) {
287 paxwarn(1, "Unable to copy %s, file would overwrite itself",
288 arcn->name);
289 return(0);
290 }
291 return(1);
292 }
293
294 /*
295 * mk_link()
296 * try to make a hard link between two files. if ign set, we do not
297 * complain.
298 * Return:
299 * 0 if successful (or we are done with this file but no error, such as
300 * finding the from file exists and the user has set -k).
301 * 1 when ign was set to indicates we could not make the link but we
302 * should try to copy/extract the file as that might work (and is an
303 * allowed option). -1 an error occurred.
304 */
305
306 #ifdef __STDC__
307 static int
308 mk_link(register char *to, register struct stat *to_sb, register char *from,
309 int ign)
310 #else
311 static int
312 mk_link(to, to_sb, from, ign)
313 register char *to;
314 register struct stat *to_sb;
315 register char *from;
316 int ign;
317 #endif
318 {
319 struct stat sb;
320 int oerrno;
321
322 /*
323 * if from file exists, it has to be unlinked to make the link. If the
324 * file exists and -k is set, skip it quietly
325 */
326 if (lstat(from, &sb) == 0) {
327 if (kflag)
328 return(0);
329
330 /*
331 * make sure it is not the same file, protect the user
332 */
333 if ((to_sb->st_dev==sb.st_dev)&&(to_sb->st_ino == sb.st_ino)) {
334 paxwarn(1, "Unable to link file %s to itself", to);
335 return(-1);;
336 }
337
338 /*
339 * try to get rid of the file, based on the type
340 */
341 if (S_ISDIR(sb.st_mode)) {
342 if (rmdir(from) < 0) {
343 syswarn(1, errno, "Unable to remove %s", from);
344 return(-1);
345 }
346 } else if (unlink(from) < 0) {
347 if (!ign) {
348 syswarn(1, errno, "Unable to remove %s", from);
349 return(-1);
350 }
351 return(1);
352 }
353 }
354
355 /*
356 * from file is gone (or did not exist), try to make the hard link.
357 * if it fails, check the path and try it again (if chk_path() says to
358 * try again)
359 */
360 for (;;) {
361 if (link(to, from) == 0)
362 break;
363 oerrno = errno;
364 if (!nodirs && chk_path(from, to_sb->st_uid, to_sb->st_gid) == 0)
365 continue;
366 if (!ign) {
367 syswarn(1, oerrno, "Could not link to %s from %s", to,
368 from);
369 return(-1);
370 }
371 return(1);
372 }
373
374 /*
375 * all right the link was made
376 */
377 return(0);
378 }
379
380 /*
381 * node_creat()
382 * create an entry in the file system (other than a file or hard link).
383 * If successful, sets uid/gid modes and times as required.
384 * Return:
385 * 0 if ok, -1 otherwise
386 */
387
388 #ifdef __STDC__
389 int
390 node_creat(register ARCHD *arcn)
391 #else
392 int
393 node_creat(arcn)
394 register ARCHD *arcn;
395 #endif
396 {
397 register int res;
398 register int ign = 0;
399 register int oerrno;
400 register int pass = 0;
401 mode_t file_mode;
402 struct stat sb;
403
404 /*
405 * create node based on type, if that fails try to unlink the node and
406 * try again. finally check the path and try again. As noted in the
407 * file and link creation routines, this method seems to exhibit the
408 * best performance in general use workloads.
409 */
410 file_mode = arcn->sb.st_mode & FILEBITS;
411
412 for (;;) {
413 switch(arcn->type) {
414 case PAX_DIR:
415 res = mkdir(arcn->name, file_mode);
416 if (ign)
417 res = 0;
418 break;
419 case PAX_CHR:
420 file_mode |= S_IFCHR;
421 res = mknod(arcn->name, file_mode, arcn->sb.st_rdev);
422 break;
423 case PAX_BLK:
424 file_mode |= S_IFBLK;
425 res = mknod(arcn->name, file_mode, arcn->sb.st_rdev);
426 break;
427 case PAX_FIF:
428 res = mkfifo(arcn->name, file_mode);
429 break;
430 case PAX_SCK:
431 /*
432 * Skip sockets, operation has no meaning under BSD
433 */
434 paxwarn(0,
435 "%s skipped. Sockets cannot be copied or extracted",
436 arcn->name);
437 return(-1);
438 case PAX_SLK:
439 res = symlink(arcn->ln_name, arcn->name);
440 break;
441 case PAX_CTG:
442 case PAX_HLK:
443 case PAX_HRG:
444 case PAX_REG:
445 default:
446 /*
447 * we should never get here
448 */
449 paxwarn(0, "%s has an unknown file type, skipping",
450 arcn->name);
451 return(-1);
452 }
453
454 /*
455 * if we were able to create the node break out of the loop,
456 * otherwise try to unlink the node and try again. if that
457 * fails check the full path and try a final time.
458 */
459 if (res == 0)
460 break;
461
462 /*
463 * we failed to make the node
464 */
465 oerrno = errno;
466 if ((ign = unlnk_exist(arcn->name, arcn->type)) < 0)
467 return(-1);
468
469 if (++pass <= 1)
470 continue;
471
472 if (nodirs || chk_path(arcn->name,arcn->sb.st_uid,arcn->sb.st_gid) < 0) {
473 syswarn(1, oerrno, "Could not create: %s", arcn->name);
474 return(-1);
475 }
476 }
477
478 /*
479 * we were able to create the node. set uid/gid, modes and times
480 */
481 if (pids)
482 res = ((arcn->type == PAX_SLK) ?
483 #if defined(__APPLE__)
484 /* Mac OS X doesn't have lchown, so don't bother */
485 0 :
486 #else
487 set_lids(arcn->name, arcn->sb.st_uid, arcn->sb.st_gid) :
488 #endif
489 set_ids(arcn->name, arcn->sb.st_uid, arcn->sb.st_gid));
490 else
491 res = 0;
492
493 /*
494 * symlinks are done now.
495 */
496 if (arcn->type == PAX_SLK)
497 return(0);
498
499 /*
500 * IMPORTANT SECURITY NOTE:
501 * if not preserving mode or we cannot set uid/gid, then PROHIBIT any
502 * set uid/gid bits
503 */
504 if (!pmode || res)
505 arcn->sb.st_mode &= ~(SETBITS);
506 if (pmode)
507 set_pmode(arcn->name, arcn->sb.st_mode);
508
509 if (arcn->type == PAX_DIR && strcmp(NM_CPIO, argv0) != 0) {
510 /*
511 * Dirs must be processed again at end of extract to set times
512 * and modes to agree with those stored in the archive. However
513 * to allow extract to continue, we may have to also set owner
514 * rights. This allows nodes in the archive that are children
515 * of this directory to be extracted without failure. Both time
516 * and modes will be fixed after the entire archive is read and
517 * before pax exits.
518 */
519 if (access(arcn->name, R_OK | W_OK | X_OK) < 0) {
520 if (lstat(arcn->name, &sb) < 0) {
521 syswarn(0, errno,"Could not access %s (stat)",
522 arcn->name);
523 set_pmode(arcn->name,file_mode | S_IRWXU);
524 } else {
525 /*
526 * We have to add rights to the dir, so we make
527 * sure to restore the mode. The mode must be
528 * restored AS CREATED and not as stored if
529 * pmode is not set.
530 */
531 set_pmode(arcn->name,
532 ((sb.st_mode & FILEBITS) | S_IRWXU));
533 if (!pmode)
534 arcn->sb.st_mode = sb.st_mode;
535 }
536
537 /*
538 * we have to force the mode to what was set here,
539 * since we changed it from the default as created.
540 */
541 add_dir(arcn->name, arcn->nlen, &(arcn->sb), 1);
542 } else if (pmode || patime || pmtime)
543 add_dir(arcn->name, arcn->nlen, &(arcn->sb), 0);
544 }
545
546 if (patime || pmtime)
547 set_ftime(arcn->name, arcn->sb.st_mtime, arcn->sb.st_atime, 0);
548 return(0);
549 }
550
551 /*
552 * unlnk_exist()
553 * Remove node from file system with the specified name. We pass the type
554 * of the node that is going to replace it. When we try to create a
555 * directory and find that it already exists, we allow processing to
556 * continue as proper modes etc will always be set for it later on.
557 * Return:
558 * 0 is ok to proceed, no file with the specified name exists
559 * -1 we were unable to remove the node, or we should not remove it (-k)
560 * 1 we found a directory and we were going to create a directory.
561 */
562
563 #ifdef __STDC__
564 int
565 unlnk_exist(register char *name, register int type)
566 #else
567 int
568 unlnk_exist(name, type)
569 register char *name;
570 register int type;
571 #endif
572 {
573 struct stat sb;
574
575 /*
576 * the file does not exist, or -k we are done
577 */
578 if (lstat(name, &sb) < 0)
579 return(0);
580 if (kflag)
581 return(-1);
582
583 if (S_ISDIR(sb.st_mode)) {
584 /*
585 * try to remove a directory, if it fails and we were going to
586 * create a directory anyway, tell the caller (return a 1)
587 */
588 if (rmdir(name) < 0) {
589 if (type == PAX_DIR)
590 return(1);
591 syswarn(1,errno,"Unable to remove directory %s", name);
592 return(-1);
593 }
594 return(0);
595 }
596
597 /*
598 * try to get rid of all non-directory type nodes
599 */
600 if (unlink(name) < 0) {
601 syswarn(1, errno, "Could not unlink %s", name);
602 return(-1);
603 }
604 return(0);
605 }
606
607 /*
608 * chk_path()
609 * We were trying to create some kind of node in the file system and it
610 * failed. chk_path() makes sure the path up to the node exists and is
611 * writeable. When we have to create a directory that is missing along the
612 * path somewhere, the directory we create will be set to the same
613 * uid/gid as the file has (when uid and gid are being preserved).
614 * NOTE: this routine is a real performance loss. It is only used as a
615 * last resort when trying to create entries in the file system.
616 * Return:
617 * -1 when it could find nothing it is allowed to fix.
618 * 0 otherwise
619 */
620
621 #ifdef __STDC__
622 int
623 chk_path( register char *name, uid_t st_uid, gid_t st_gid)
624 #else
625 int
626 chk_path(name, st_uid, st_gid)
627 register char *name;
628 uid_t st_uid;
629 gid_t st_gid;
630 #endif
631 {
632 register char *spt = name;
633 struct stat sb;
634 int retval = -1;
635
636 /*
637 * watch out for paths with nodes stored directly in / (e.g. /bozo)
638 */
639 if (*spt == '/')
640 ++spt;
641
642 for(;;) {
643 /*
644 * work foward from the first / and check each part of the path
645 */
646 spt = strchr(spt, '/');
647 if (spt == NULL)
648 break;
649 *spt = '\0';
650
651 /*
652 * if it exists we assume it is a directory, it is not within
653 * the spec (at least it seems to read that way) to alter the
654 * file system for nodes NOT EXPLICITLY stored on the archive.
655 * If that assumption is changed, you would test the node here
656 * and figure out how to get rid of it (probably like some
657 * recursive unlink()) or fix up the directory permissions if
658 * required (do an access()).
659 */
660 if (lstat(name, &sb) == 0) {
661 *(spt++) = '/';
662 continue;
663 }
664
665 /*
666 * the path fails at this point, see if we can create the
667 * needed directory and continue on
668 */
669 if (mkdir(name, S_IRWXU | S_IRWXG | S_IRWXO) < 0) {
670 *spt = '/';
671 retval = -1;
672 break;
673 }
674
675 /*
676 * we were able to create the directory. We will tell the
677 * caller that we found something to fix, and it is ok to try
678 * and create the node again.
679 */
680 retval = 0;
681 if (pids)
682 (void)set_ids(name, st_uid, st_gid);
683
684 /*
685 * make sure the user doen't have some strange umask that
686 * causes this newly created directory to be unusable. We fix
687 * the modes and restore them back to the creation default at
688 * the end of pax
689 */
690 if ((access(name, R_OK | W_OK | X_OK) < 0) &&
691 (lstat(name, &sb) == 0)) {
692 set_pmode(name, ((sb.st_mode & FILEBITS) | S_IRWXU));
693 add_dir(name, spt - name, &sb, 1);
694 }
695 *(spt++) = '/';
696 continue;
697 }
698 return(retval);
699 }
700
701 /*
702 * set_ftime()
703 * Set the access time and modification time for a named file. If frc is
704 * non-zero we force these times to be set even if the user did not
705 * request access and/or modification time preservation (this is also
706 * used by -t to reset access times).
707 * When ign is zero, only those times the user has asked for are set, the
708 * other ones are left alone. We do not assume the un-documented feature
709 * of many utimes() implementations that consider a 0 time value as a do
710 * not set request.
711 */
712
713 #ifdef __STDC__
714 void
715 set_ftime(char *fnm, time_t mtime, time_t atime, int frc)
716 #else
717 void
718 set_ftime(fnm, mtime, atime, frc)
719 char *fnm;
720 time_t mtime;
721 time_t atime;
722 int frc;
723 #endif
724 {
725 static struct timeval tv[2] = {{0L, 0L}, {0L, 0L}};
726 struct stat sb;
727
728 tv[0].tv_sec = (long)atime;
729 tv[1].tv_sec = (long)mtime;
730 if (!frc && (!patime || !pmtime)) {
731 /*
732 * if we are not forcing, only set those times the user wants
733 * set. We get the current values of the times if we need them.
734 */
735 if (lstat(fnm, &sb) == 0) {
736 if (!patime)
737 tv[0].tv_sec = (long)sb.st_atime;
738 if (!pmtime)
739 tv[1].tv_sec = (long)sb.st_mtime;
740 } else
741 syswarn(0,errno,"Unable to obtain file stats %s", fnm);
742 }
743
744 /*
745 * set the times
746 */
747 if (utimes(fnm, tv) < 0)
748 syswarn(1, errno, "Access/modification time set failed on: %s",
749 fnm);
750 return;
751 }
752
753 /*
754 * set_ids()
755 * set the uid and gid of a file system node
756 * Return:
757 * 0 when set, -1 on failure
758 */
759
760 #ifdef __STDC__
761 int
762 set_ids(char *fnm, uid_t uid, gid_t gid)
763 #else
764 int
765 set_ids(fnm, uid, gid)
766 char *fnm;
767 uid_t uid;
768 gid_t gid;
769 #endif
770 {
771 if (chown(fnm, uid, gid) < 0) {
772 /*
773 * ignore EPERM unless in verbose mode or being run by root.
774 * if running as pax, POSIX requires a warning.
775 */
776 if (strcmp(NM_PAX, argv0) == 0 || errno != EPERM || vflag ||
777 geteuid() == 0)
778 syswarn(1, errno, "Unable to set file uid/gid of %s",
779 fnm);
780 return(-1);
781 }
782 return(0);
783 }
784
785 #if !defined(__APPLE__)
786 /* Mac OS X doesn't have lchown */
787
788 /*
789 * set_lids()
790 * set the uid and gid of a file system node
791 * Return:
792 * 0 when set, -1 on failure
793 */
794
795 #ifdef __STDC__
796 int
797 set_lids(char *fnm, uid_t uid, gid_t gid)
798 #else
799 int
800 set_lids(fnm, uid, gid)
801 char *fnm;
802 uid_t uid;
803 gid_t gid;
804 #endif
805 {
806 if (lchown(fnm, uid, gid) < 0) {
807 /*
808 * ignore EPERM unless in verbose mode or being run by root.
809 * if running as pax, POSIX requires a warning.
810 */
811 if (strcmp(NM_PAX, argv0) == 0 || errno != EPERM || vflag ||
812 geteuid() == 0)
813 syswarn(1, errno, "Unable to set file uid/gid of %s",
814 fnm);
815 return(-1);
816 }
817 return(0);
818 }
819 #endif
820
821 /*
822 * set_pmode()
823 * Set file access mode
824 */
825
826 #ifdef __STDC__
827 void
828 set_pmode(char *fnm, mode_t mode)
829 #else
830 void
831 set_pmode(fnm, mode)
832 char *fnm;
833 mode_t mode;
834 #endif
835 {
836 mode &= ABITS;
837 if (chmod(fnm, mode) < 0)
838 syswarn(1, errno, "Could not set permissions on %s", fnm);
839 return;
840 }
841
842 /*
843 * file_write()
844 * Write/copy a file (during copy or archive extract). This routine knows
845 * how to copy files with lseek holes in it. (Which are read as file
846 * blocks containing all 0's but do not have any file blocks associated
847 * with the data). Typical examples of these are files created by dbm
848 * variants (.pag files). While the file size of these files are huge, the
849 * actual storage is quite small (the files are sparse). The problem is
850 * the holes read as all zeros so are probably stored on the archive that
851 * way (there is no way to determine if the file block is really a hole,
852 * we only know that a file block of all zero's can be a hole).
853 * At this writing, no major archive format knows how to archive files
854 * with holes. However, on extraction (or during copy, -rw) we have to
855 * deal with these files. Without detecting the holes, the files can
856 * consume a lot of file space if just written to disk. This replacement
857 * for write when passed the basic allocation size of a file system block,
858 * uses lseek whenever it detects the input data is all 0 within that
859 * file block. In more detail, the strategy is as follows:
860 * While the input is all zero keep doing an lseek. Keep track of when we
861 * pass over file block boundries. Only write when we hit a non zero
862 * input. once we have written a file block, we continue to write it to
863 * the end (we stop looking at the input). When we reach the start of the
864 * next file block, start checking for zero blocks again. Working on file
865 * block boundries significantly reduces the overhead when copying files
866 * that are NOT very sparse. This overhead (when compared to a write) is
867 * almost below the measurement resolution on many systems. Without it,
868 * files with holes cannot be safely copied. It does has a side effect as
869 * it can put holes into files that did not have them before, but that is
870 * not a problem since the file contents are unchanged (in fact it saves
871 * file space). (Except on paging files for diskless clients. But since we
872 * cannot determine one of those file from here, we ignore them). If this
873 * ever ends up on a system where CTG files are supported and the holes
874 * are not desired, just do a conditional test in those routines that
875 * call file_write() and have it call write() instead. BEFORE CLOSING THE
876 * FILE, make sure to call file_flush() when the last write finishes with
877 * an empty block. A lot of file systems will not create an lseek hole at
878 * the end. In this case we drop a single 0 at the end to force the
879 * trailing 0's in the file.
880 * ---Parameters---
881 * rem: how many bytes left in this file system block
882 * isempt: have we written to the file block yet (is it empty)
883 * sz: basic file block allocation size
884 * cnt: number of bytes on this write
885 * str: buffer to write
886 * Return:
887 * number of bytes written, -1 on write (or lseek) error.
888 */
889
890 #ifdef __STDC__
891 int
892 file_write(int fd, char *str, register int cnt, int *rem, int *isempt, int sz,
893 char *name)
894 #else
895 int
896 file_write(fd, str, cnt, rem, isempt, sz, name)
897 int fd;
898 char *str;
899 register int cnt;
900 int *rem;
901 int *isempt;
902 int sz;
903 char *name;
904 #endif
905 {
906 register char *pt;
907 register char *end;
908 register int wcnt;
909 register char *st = str;
910
911 /*
912 * while we have data to process
913 */
914 while (cnt) {
915 if (!*rem) {
916 /*
917 * We are now at the start of file system block again
918 * (or what we think one is...). start looking for
919 * empty blocks again
920 */
921 *isempt = 1;
922 *rem = sz;
923 }
924
925 /*
926 * only examine up to the end of the current file block or
927 * remaining characters to write, whatever is smaller
928 */
929 wcnt = MIN(cnt, *rem);
930 cnt -= wcnt;
931 *rem -= wcnt;
932 if (*isempt) {
933 /*
934 * have not written to this block yet, so we keep
935 * looking for zero's
936 */
937 pt = st;
938 end = st + wcnt;
939
940 /*
941 * look for a zero filled buffer
942 */
943 while ((pt < end) && (*pt == '\0'))
944 ++pt;
945
946 if (pt == end) {
947 /*
948 * skip, buf is empty so far
949 */
950 if (lseek(fd, (off_t)wcnt, SEEK_CUR) < 0) {
951 syswarn(1,errno,"File seek on %s",
952 name);
953 return(-1);
954 }
955 st = pt;
956 continue;
957 }
958 /*
959 * drat, the buf is not zero filled
960 */
961 *isempt = 0;
962 }
963
964 /*
965 * have non-zero data in this file system block, have to write
966 */
967 if (write(fd, st, wcnt) != wcnt) {
968 syswarn(1, errno, "Failed write to file %s", name);
969 return(-1);
970 }
971 st += wcnt;
972 }
973 return(st - str);
974 }
975
976 /*
977 * file_flush()
978 * when the last file block in a file is zero, many file systems will not
979 * let us create a hole at the end. To get the last block with zeros, we
980 * write the last BYTE with a zero (back up one byte and write a zero).
981 */
982
983 #ifdef __STDC__
984 void
985 file_flush(int fd, char *fname, int isempt)
986 #else
987 void
988 file_flush(fd, fname, isempt)
989 int fd;
990 char *fname;
991 int isempt;
992 #endif
993 {
994 static char blnk[] = "\0";
995
996 /*
997 * silly test, but make sure we are only called when the last block is
998 * filled with all zeros.
999 */
1000 if (!isempt)
1001 return;
1002
1003 /*
1004 * move back one byte and write a zero
1005 */
1006 if (lseek(fd, (off_t)-1, SEEK_CUR) < 0) {
1007 syswarn(1, errno, "Failed seek on file %s", fname);
1008 return;
1009 }
1010
1011 if (write(fd, blnk, 1) < 0)
1012 syswarn(1, errno, "Failed write to file %s", fname);
1013 return;
1014 }
1015
1016 /*
1017 * rdfile_close()
1018 * close a file we have beed reading (to copy or archive). If we have to
1019 * reset access time (tflag) do so (the times are stored in arcn).
1020 */
1021
1022 #ifdef __STDC__
1023 void
1024 rdfile_close(register ARCHD *arcn, register int *fd)
1025 #else
1026 void
1027 rdfile_close(arcn, fd)
1028 register ARCHD *arcn;
1029 register int *fd;
1030 #endif
1031 {
1032 /*
1033 * make sure the file is open
1034 */
1035 if (*fd < 0)
1036 return;
1037
1038 (void)close(*fd);
1039 *fd = -1;
1040 if (!tflag)
1041 return;
1042
1043 /*
1044 * user wants last access time reset
1045 */
1046 set_ftime(arcn->org_name, arcn->sb.st_mtime, arcn->sb.st_atime, 1);
1047 return;
1048 }
1049
1050 /*
1051 * set_crc()
1052 * read a file to calculate its crc. This is a real drag. Archive formats
1053 * that have this, end up reading the file twice (we have to write the
1054 * header WITH the crc before writing the file contents. Oh well...
1055 * Return:
1056 * 0 if was able to calculate the crc, -1 otherwise
1057 */
1058
1059 #ifdef __STDC__
1060 int
1061 set_crc(register ARCHD *arcn, register int fd)
1062 #else
1063 int
1064 set_crc(arcn, fd)
1065 register ARCHD *arcn;
1066 register int fd;
1067 #endif
1068 {
1069 register int i;
1070 register int res;
1071 off_t cpcnt = 0L;
1072 u_long size;
1073 unsigned long crc = 0L;
1074 char tbuf[FILEBLK];
1075 struct stat sb;
1076
1077 if (fd < 0) {
1078 /*
1079 * hmm, no fd, should never happen. well no crc then.
1080 */
1081 arcn->crc = 0L;
1082 return(0);
1083 }
1084
1085 if ((size = (u_long)arcn->sb.st_blksize) > (u_long)sizeof(tbuf))
1086 size = (u_long)sizeof(tbuf);
1087
1088 /*
1089 * read all the bytes we think that there are in the file. If the user
1090 * is trying to archive an active file, forget this file.
1091 */
1092 for(;;) {
1093 if ((res = read(fd, tbuf, size)) <= 0)
1094 break;
1095 cpcnt += res;
1096 for (i = 0; i < res; ++i)
1097 crc += (tbuf[i] & 0xff);
1098 }
1099
1100 /*
1101 * safety check. we want to avoid archiving files that are active as
1102 * they can create inconsistant archive copies.
1103 */
1104 if (cpcnt != arcn->sb.st_size)
1105 paxwarn(1, "File changed size %s", arcn->org_name);
1106 else if (fstat(fd, &sb) < 0)
1107 syswarn(1, errno, "Failed stat on %s", arcn->org_name);
1108 else if (arcn->sb.st_mtime != sb.st_mtime)
1109 paxwarn(1, "File %s was modified during read", arcn->org_name);
1110 else if (lseek(fd, (off_t)0L, SEEK_SET) < 0)
1111 syswarn(1, errno, "File rewind failed on: %s", arcn->org_name);
1112 else {
1113 arcn->crc = crc;
1114 return(0);
1115 }
1116 return(-1);
1117 }