xref: /netbsd-src/bin/pax/file_subs.c (revision eb7c1594f145c931049e1fd9eb056a5987e87e59)
1 /*	$NetBSD: file_subs.c,v 1.34 2003/08/07 09:05:20 agc Exp $	*/
2 
3 /*-
4  * Copyright (c) 1992, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Keith Muller of the University of California, San Diego.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  */
34 
35 /*-
36  * Copyright (c) 1992 Keith Muller.
37  *
38  * This code is derived from software contributed to Berkeley by
39  * Keith Muller of the University of California, San Diego.
40  *
41  * Redistribution and use in source and binary forms, with or without
42  * modification, are permitted provided that the following conditions
43  * are met:
44  * 1. Redistributions of source code must retain the above copyright
45  *    notice, this list of conditions and the following disclaimer.
46  * 2. Redistributions in binary form must reproduce the above copyright
47  *    notice, this list of conditions and the following disclaimer in the
48  *    documentation and/or other materials provided with the distribution.
49  * 3. All advertising materials mentioning features or use of this software
50  *    must display the following acknowledgement:
51  *	This product includes software developed by the University of
52  *	California, Berkeley and its contributors.
53  * 4. Neither the name of the University nor the names of its contributors
54  *    may be used to endorse or promote products derived from this software
55  *    without specific prior written permission.
56  *
57  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
58  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
59  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
60  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
61  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
62  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
63  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
64  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
65  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
66  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
67  * SUCH DAMAGE.
68  */
69 
70 #include <sys/cdefs.h>
71 #if defined(__RCSID) && !defined(lint)
72 #if 0
73 static char sccsid[] = "@(#)file_subs.c	8.1 (Berkeley) 5/31/93";
74 #else
75 __RCSID("$NetBSD: file_subs.c,v 1.34 2003/08/07 09:05:20 agc Exp $");
76 #endif
77 #endif /* not lint */
78 
79 #include <sys/types.h>
80 #include <sys/time.h>
81 #include <sys/stat.h>
82 #include <unistd.h>
83 #include <sys/param.h>
84 #include <fcntl.h>
85 #include <string.h>
86 #include <stdio.h>
87 #include <ctype.h>
88 #include <errno.h>
89 #include <sys/uio.h>
90 #include <stdlib.h>
91 #include "pax.h"
92 #include "extern.h"
93 #include "options.h"
94 
95 static int
96 mk_link(char *,struct stat *,char *, int);
97 
98 /*
99  * routines that deal with file operations such as: creating, removing;
100  * and setting access modes, uid/gid and times of files
101  */
102 
103 #define FILEBITS		(S_ISVTX | S_IRWXU | S_IRWXG | S_IRWXO)
104 #define SETBITS			(S_ISUID | S_ISGID)
105 #define ABITS			(FILEBITS | SETBITS)
106 
107 /*
108  * file_creat()
109  *	Create and open a file.
110  * Return:
111  *	file descriptor or -1 for failure
112  */
113 
114 int
115 file_creat(ARCHD *arcn)
116 {
117 	int fd = -1;
118 	mode_t file_mode;
119 	int oerrno;
120 
121 	/*
122 	 * assume file doesn't exist, so just try to create it, most times this
123 	 * works. We have to take special handling when the file does exist. To
124 	 * detect this, we use O_EXCL. For example when trying to create a
125 	 * file and a character device or fifo exists with the same name, we
126 	 * can accidently open the device by mistake (or block waiting to open)
127 	 * If we find that the open has failed, then spend the effort to
128 	 * figure out why. This strategy was found to have better average
129 	 * performance in common use than checking the file (and the path)
130 	 * first with lstat.
131 	 */
132 	file_mode = arcn->sb.st_mode & FILEBITS;
133 	if ((fd = open(arcn->name, O_WRONLY | O_CREAT | O_TRUNC | O_EXCL,
134 	    file_mode)) >= 0)
135 		return(fd);
136 
137 	/*
138 	 * the file seems to exist. First we try to get rid of it (found to be
139 	 * the second most common failure when traced). If this fails, only
140 	 * then we go to the expense to check and create the path to the file
141 	 */
142 	if (unlnk_exist(arcn->name, arcn->type) != 0)
143 		return(-1);
144 
145 	for (;;) {
146 		/*
147 		 * try to open it again, if this fails, check all the nodes in
148 		 * the path and give it a final try. if chk_path() finds that
149 		 * it cannot fix anything, we will skip the last attempt
150 		 */
151 		if ((fd = open(arcn->name, O_WRONLY | O_CREAT | O_TRUNC,
152 		    file_mode)) >= 0)
153 			break;
154 		oerrno = errno;
155 		if (nodirs || chk_path(arcn->name,arcn->sb.st_uid,arcn->sb.st_gid) < 0) {
156 			(void)fflush(listf);
157 			syswarn(1, oerrno, "Cannot create %s", arcn->name);
158 			return(-1);
159 		}
160 	}
161 	return(fd);
162 }
163 
164 /*
165  * file_close()
166  *	Close file descriptor to a file just created by pax. Sets modes,
167  *	ownership and times as required.
168  * Return:
169  *	0 for success, -1 for failure
170  */
171 
172 void
173 file_close(ARCHD *arcn, int fd)
174 {
175 	int res = 0;
176 
177 	if (fd < 0)
178 		return;
179 	if (close(fd) < 0)
180 		syswarn(0, errno, "Cannot close file descriptor on %s",
181 		    arcn->name);
182 
183 	/*
184 	 * set owner/groups first as this may strip off mode bits we want
185 	 * then set file permission modes. Then set file access and
186 	 * modification times.
187 	 */
188 	if (pids)
189 		res = set_ids(arcn->name, arcn->sb.st_uid, arcn->sb.st_gid);
190 
191 	/*
192 	 * IMPORTANT SECURITY NOTE:
193 	 * if not preserving mode or we cannot set uid/gid, then PROHIBIT
194 	 * set uid/gid bits
195 	 */
196 	if (!pmode || res)
197 		arcn->sb.st_mode &= ~(SETBITS);
198 	if (pmode)
199 		set_pmode(arcn->name, arcn->sb.st_mode);
200 	if (patime || pmtime)
201 		set_ftime(arcn->name, arcn->sb.st_mtime, arcn->sb.st_atime, 0);
202 #if HAVE_STRUCT_STAT_ST_FLAGS
203 	if (pfflags && arcn->type != PAX_SLK)
204 		set_chflags(arcn->name, arcn->sb.st_flags);
205 #endif
206 }
207 
208 /*
209  * lnk_creat()
210  *	Create a hard link to arcn->ln_name from arcn->name. arcn->ln_name
211  *	must exist;
212  * Return:
213  *	0 if ok, -1 otherwise
214  */
215 
216 int
217 lnk_creat(ARCHD *arcn)
218 {
219 	struct stat sb;
220 
221 	/*
222 	 * we may be running as root, so we have to be sure that link target
223 	 * is not a directory, so we lstat and check
224 	 */
225 	if (lstat(arcn->ln_name, &sb) < 0) {
226 		syswarn(1, errno, "Cannot link to %s from %s", arcn->ln_name,
227 		    arcn->name);
228 		return(-1);
229 	}
230 
231 	if (S_ISDIR(sb.st_mode)) {
232 		tty_warn(1, "A hard link to the directory %s is not allowed",
233 		    arcn->ln_name);
234 		return(-1);
235 	}
236 
237 	return(mk_link(arcn->ln_name, &sb, arcn->name, 0));
238 }
239 
240 /*
241  * cross_lnk()
242  *	Create a hard link to arcn->org_name from arcn->name. Only used in copy
243  *	with the -l flag. No warning or error if this does not succeed (we will
244  *	then just create the file)
245  * Return:
246  *	1 if copy() should try to create this file node
247  *	0 if cross_lnk() ok, -1 for fatal flaw (like linking to self).
248  */
249 
250 int
251 cross_lnk(ARCHD *arcn)
252 {
253 	/*
254 	 * try to make a link to original file (-l flag in copy mode). make
255 	 * sure we do not try to link to directories in case we are running as
256 	 * root (and it might succeed).
257 	 */
258 	if (arcn->type == PAX_DIR)
259 		return(1);
260 	return(mk_link(arcn->org_name, &(arcn->sb), arcn->name, 1));
261 }
262 
263 /*
264  * chk_same()
265  *	In copy mode if we are not trying to make hard links between the src
266  *	and destinations, make sure we are not going to overwrite ourselves by
267  *	accident. This slows things down a little, but we have to protect all
268  *	those people who make typing errors.
269  * Return:
270  *	1 the target does not exist, go ahead and copy
271  *	0 skip it file exists (-k) or may be the same as source file
272  */
273 
274 int
275 chk_same(ARCHD *arcn)
276 {
277 	struct stat sb;
278 
279 	/*
280 	 * if file does not exist, return. if file exists and -k, skip it
281 	 * quietly
282 	 */
283 	if (lstat(arcn->name, &sb) < 0)
284 		return(1);
285 	if (kflag)
286 		return(0);
287 
288 	/*
289 	 * better make sure the user does not have src == dest by mistake
290 	 */
291 	if ((arcn->sb.st_dev == sb.st_dev) && (arcn->sb.st_ino == sb.st_ino)) {
292 		tty_warn(1, "Unable to copy %s, file would overwrite itself",
293 		    arcn->name);
294 		return(0);
295 	}
296 	return(1);
297 }
298 
299 /*
300  * mk_link()
301  *	try to make a hard link between two files. if ign set, we do not
302  *	complain.
303  * Return:
304  *	0 if successful (or we are done with this file but no error, such as
305  *	finding the from file exists and the user has set -k).
306  *	1 when ign was set to indicates we could not make the link but we
307  *	should try to copy/extract the file as that might work (and is an
308  *	allowed option). -1 an error occurred.
309  */
310 
311 static int
312 mk_link(char *to, struct stat *to_sb, char *from, int ign)
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 			tty_warn(1, "Cannot 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 				syswarn(1, errno, "Cannot remove %s", from);
339 				return(-1);
340 			}
341 		} else if (unlink(from) < 0) {
342 			if (!ign) {
343 				syswarn(1, errno, "Cannot 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 			syswarn(1, oerrno, "Cannot 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 int
384 node_creat(ARCHD *arcn)
385 {
386 	int res;
387 	int ign = 0;
388 	int oerrno;
389 	int pass = 0;
390 	mode_t file_mode;
391 	struct stat sb;
392 	char target[MAXPATHLEN];
393 	char *nm = arcn->name;
394 	int len;
395 
396 	/*
397 	 * create node based on type, if that fails try to unlink the node and
398 	 * try again. finally check the path and try again. As noted in the
399 	 * file and link creation routines, this method seems to exhibit the
400 	 * best performance in general use workloads.
401 	 */
402 	file_mode = arcn->sb.st_mode & FILEBITS;
403 
404 	for (;;) {
405 		switch(arcn->type) {
406 		case PAX_DIR:
407 			/*
408 			 * If -h (or -L) was given in tar-mode, follow the
409 			 * potential symlink chain before trying to create the
410 			 * directory.
411 			 */
412 			if (strcmp(NM_TAR, argv0) == 0 && Lflag) {
413 				while (lstat(nm, &sb) == 0 &&
414 				    S_ISLNK(sb.st_mode)) {
415 					len = readlink(nm, target,
416 					    sizeof target - 1);
417 					if (len == -1) {
418 						syswarn(0, errno,
419 						   "cannot follow symlink %s in chain for %s",
420 						    nm, arcn->name);
421 						res = -1;
422 						goto badlink;
423 					}
424 					target[len] = '\0';
425 					nm = target;
426 				}
427 			}
428 			res = mkdir(nm, file_mode);
429 
430 badlink:
431 			if (ign)
432 				res = 0;
433 			break;
434 		case PAX_CHR:
435 			file_mode |= S_IFCHR;
436 			res = mknod(nm, file_mode, arcn->sb.st_rdev);
437 			break;
438 		case PAX_BLK:
439 			file_mode |= S_IFBLK;
440 			res = mknod(nm, file_mode, arcn->sb.st_rdev);
441 			break;
442 		case PAX_FIF:
443 			res = mkfifo(nm, file_mode);
444 			break;
445 		case PAX_SCK:
446 			/*
447 			 * Skip sockets, operation has no meaning under BSD
448 			 */
449 			tty_warn(0,
450 			    "%s skipped. Sockets cannot be copied or extracted",
451 			    nm);
452 			return(-1);
453 		case PAX_SLK:
454 			res = symlink(arcn->ln_name, nm);
455 			break;
456 		case PAX_CTG:
457 		case PAX_HLK:
458 		case PAX_HRG:
459 		case PAX_REG:
460 		default:
461 			/*
462 			 * we should never get here
463 			 */
464 			tty_warn(0, "%s has an unknown file type, skipping",
465 			    nm);
466 			return(-1);
467 		}
468 
469 		/*
470 		 * if we were able to create the node break out of the loop,
471 		 * otherwise try to unlink the node and try again. if that
472 		 * fails check the full path and try a final time.
473 		 */
474 		if (res == 0)
475 			break;
476 
477 		/*
478 		 * we failed to make the node
479 		 */
480 		oerrno = errno;
481 		if ((ign = unlnk_exist(nm, arcn->type)) < 0)
482 			return(-1);
483 
484 		if (++pass <= 1)
485 			continue;
486 
487 		if (nodirs || chk_path(nm,arcn->sb.st_uid,arcn->sb.st_gid) < 0) {
488 			syswarn(1, oerrno, "Cannot create %s", nm);
489 			return(-1);
490 		}
491 	}
492 
493 	/*
494 	 * we were able to create the node. set uid/gid, modes and times
495 	 */
496 	if (pids)
497 		res = set_ids(nm, arcn->sb.st_uid, arcn->sb.st_gid);
498 	else
499 		res = 0;
500 
501 	/*
502 	 * IMPORTANT SECURITY NOTE:
503 	 * if not preserving mode or we cannot set uid/gid, then PROHIBIT any
504 	 * set uid/gid bits
505 	 */
506 	if (!pmode || res)
507 		arcn->sb.st_mode &= ~(SETBITS);
508 	if (pmode)
509 		set_pmode(arcn->name, arcn->sb.st_mode);
510 
511 	if (arcn->type == PAX_DIR && strcmp(NM_CPIO, argv0) != 0) {
512 		/*
513 		 * Dirs must be processed again at end of extract to set times
514 		 * and modes to agree with those stored in the archive. However
515 		 * to allow extract to continue, we may have to also set owner
516 		 * rights. This allows nodes in the archive that are children
517 		 * of this directory to be extracted without failure. Both time
518 		 * and modes will be fixed after the entire archive is read and
519 		 * before pax exits.
520 		 */
521 		if (access(nm, R_OK | W_OK | X_OK) < 0) {
522 			if (lstat(nm, &sb) < 0) {
523 				syswarn(0, errno,"Cannot access %s (stat)",
524 				    arcn->name);
525 				set_pmode(nm,file_mode | S_IRWXU);
526 			} else {
527 				/*
528 				 * We have to add rights to the dir, so we make
529 				 * sure to restore the mode. The mode must be
530 				 * restored AS CREATED and not as stored if
531 				 * pmode is not set.
532 				 */
533 				set_pmode(nm,
534 				    ((sb.st_mode & FILEBITS) | S_IRWXU));
535 				if (!pmode)
536 					arcn->sb.st_mode = sb.st_mode;
537 			}
538 
539 			/*
540 			 * we have to force the mode to what was set here,
541 			 * since we changed it from the default as created.
542 			 */
543 			add_dir(nm, arcn->nlen, &(arcn->sb), 1);
544 		} else if (pmode || patime || pmtime)
545 			add_dir(nm, arcn->nlen, &(arcn->sb), 0);
546 	}
547 
548 #if HAVE_LUTIMES
549 	if (patime || pmtime)
550 #else
551 	if ((patime || pmtime) && arcn->type != PAX_SLK)
552 #endif
553 		set_ftime(arcn->name, arcn->sb.st_mtime, arcn->sb.st_atime, 0);
554 
555 #if HAVE_STRUCT_STAT_ST_FLAGS
556 	if (pfflags && arcn->type != PAX_SLK)
557 		set_chflags(arcn->name, arcn->sb.st_flags);
558 #endif
559 	return(0);
560 }
561 
562 /*
563  * unlnk_exist()
564  *	Remove node from file system with the specified name. We pass the type
565  *	of the node that is going to replace it. When we try to create a
566  *	directory and find that it already exists, we allow processing to
567  *	continue as proper modes etc will always be set for it later on.
568  * Return:
569  *	0 is ok to proceed, no file with the specified name exists
570  *	-1 we were unable to remove the node, or we should not remove it (-k)
571  *	1 we found a directory and we were going to create a directory.
572  */
573 
574 int
575 unlnk_exist(char *name, int type)
576 {
577 	struct stat sb;
578 
579 	/*
580 	 * the file does not exist, or -k we are done
581 	 */
582 	if (lstat(name, &sb) < 0)
583 		return(0);
584 	if (kflag)
585 		return(-1);
586 
587 	if (S_ISDIR(sb.st_mode)) {
588 		/*
589 		 * try to remove a directory, if it fails and we were going to
590 		 * create a directory anyway, tell the caller (return a 1)
591 		 */
592 		if (rmdir(name) < 0) {
593 			if (type == PAX_DIR)
594 				return(1);
595 			syswarn(1, errno, "Cannot remove directory %s", name);
596 			return(-1);
597 		}
598 		return(0);
599 	}
600 
601 	/*
602 	 * try to get rid of all non-directory type nodes
603 	 */
604 	if (unlink(name) < 0) {
605 		(void)fflush(listf);
606 		syswarn(1, errno, "Cannot unlink %s", name);
607 		return(-1);
608 	}
609 	return(0);
610 }
611 
612 /*
613  * chk_path()
614  *	We were trying to create some kind of node in the file system and it
615  *	failed. chk_path() makes sure the path up to the node exists and is
616  *	writable. When we have to create a directory that is missing along the
617  *	path somewhere, the directory we create will be set to the same
618  *	uid/gid as the file has (when uid and gid are being preserved).
619  *	NOTE: this routine is a real performance loss. It is only used as a
620  *	last resort when trying to create entries in the file system.
621  * Return:
622  *	-1 when it could find nothing it is allowed to fix.
623  *	0 otherwise
624  */
625 
626 int
627 chk_path( char *name, uid_t st_uid, gid_t st_gid)
628 {
629 	char *spt = name;
630 	struct stat sb;
631 	int retval = -1;
632 
633 	/*
634 	 * watch out for paths with nodes stored directly in / (e.g. /bozo)
635 	 */
636 	if (*spt == '/')
637 		++spt;
638 
639 	for(;;) {
640 		/*
641 		 * work forward from the first / and check each part of
642 		 * the path
643 		 */
644 		spt = strchr(spt, '/');
645 		if (spt == NULL)
646 			break;
647 		*spt = '\0';
648 
649 		/*
650 		 * if it exists we assume it is a directory, it is not within
651 		 * the spec (at least it seems to read that way) to alter the
652 		 * file system for nodes NOT EXPLICITLY stored on the archive.
653 		 * If that assumption is changed, you would test the node here
654 		 * and figure out how to get rid of it (probably like some
655 		 * recursive unlink()) or fix up the directory permissions if
656 		 * required (do an access()).
657 		 */
658 		if (lstat(name, &sb) == 0) {
659 			*(spt++) = '/';
660 			continue;
661 		}
662 
663 		/*
664 		 * the path fails at this point, see if we can create the
665 		 * needed directory and continue on
666 		 */
667 		if (mkdir(name, S_IRWXU | S_IRWXG | S_IRWXO) < 0) {
668 			*spt = '/';
669 			retval = -1;
670 			break;
671 		}
672 
673 		/*
674 		 * we were able to create the directory. We will tell the
675 		 * caller that we found something to fix, and it is ok to try
676 		 * and create the node again.
677 		 */
678 		retval = 0;
679 		if (pids)
680 			(void)set_ids(name, st_uid, st_gid);
681 
682 		/*
683 		 * make sure the user doesn't have some strange umask that
684 		 * causes this newly created directory to be unusable. We fix
685 		 * the modes and restore them back to the creation default at
686 		 * the end of pax
687 		 */
688 		if ((access(name, R_OK | W_OK | X_OK) < 0) &&
689 		    (lstat(name, &sb) == 0)) {
690 			set_pmode(name, ((sb.st_mode & FILEBITS) | S_IRWXU));
691 			add_dir(name, spt - name, &sb, 1);
692 		}
693 		*(spt++) = '/';
694 		continue;
695 	}
696 	return(retval);
697 }
698 
699 /*
700  * set_ftime()
701  *	Set the access time and modification time for a named file. If frc
702  *	is non-zero we force these times to be set even if the user did not
703  *	request access and/or modification time preservation (this is also
704  *	used by -t to reset access times).
705  *	When ign is zero, only those times the user has asked for are set, the
706  *	other ones are left alone. We do not assume the un-documented feature
707  *	of many utimes() implementations that consider a 0 time value as a do
708  *	not set request.
709  */
710 
711 void
712 set_ftime(char *fnm, time_t mtime, time_t atime, int frc)
713 {
714 	struct timeval tv[2];
715 	struct stat sb;
716 
717 	tv[0].tv_sec = (long)atime;
718 	tv[0].tv_usec = 0;
719 	tv[1].tv_sec = (long)mtime;
720 	tv[1].tv_usec = 0;
721 	if (!frc && (!patime || !pmtime)) {
722 		/*
723 		 * if we are not forcing, only set those times the user wants
724 		 * set. We get the current values of the times if we need them.
725 		 */
726 		if (lstat(fnm, &sb) == 0) {
727 #ifdef BSD4_4
728 			if (!patime)
729 				TIMESPEC_TO_TIMEVAL(&tv[0], &sb.st_atimespec);
730 			if (!pmtime)
731 				TIMESPEC_TO_TIMEVAL(&tv[1], &sb.st_mtimespec);
732 #else
733 			if (!patime)
734 				tv[0].tv_sec = sb.st_atime;
735 			if (!pmtime)
736 				tv[1].tv_sec = sb.st_mtime;
737 #endif
738 		} else
739 			syswarn(0, errno, "Cannot obtain file stats %s", fnm);
740 	}
741 
742 	/*
743 	 * set the times
744 	 */
745 #if HAVE_LUTIMES
746 	if (lutimes(fnm, tv))
747 #else
748 	if (utimes(fnm, tv))
749 #endif
750 		syswarn(1, errno, "Access/modification time set failed on: %s",
751 		    fnm);
752 	return;
753 }
754 
755 /*
756  * set_ids()
757  *	set the uid and gid of a file system node
758  * Return:
759  *	0 when set, -1 on failure
760  */
761 
762 int
763 set_ids(char *fnm, uid_t uid, gid_t gid)
764 {
765 	if (geteuid() == 0)
766 		if (lchown(fnm, uid, gid)) {
767 			(void)fflush(listf);
768 			syswarn(1, errno, "Cannot set file uid/gid of %s",
769 			    fnm);
770 			return(-1);
771 		}
772 	return(0);
773 }
774 
775 /*
776  * set_pmode()
777  *	Set file access mode
778  */
779 
780 void
781 set_pmode(char *fnm, mode_t mode)
782 {
783 	mode &= ABITS;
784 	if (lchmod(fnm, mode)) {
785 		(void)fflush(listf);
786 		syswarn(1, errno, "Cannot set permissions on %s", fnm);
787 	}
788 	return;
789 }
790 
791 /*
792  * set_chflags()
793  *	Set 4.4BSD file flags
794  */
795 void
796 set_chflags(char *fnm, u_int32_t flags)
797 {
798 
799 #if 0
800 	if (chflags(fnm, flags) < 0 && errno != EOPNOTSUPP)
801 		syswarn(1, errno, "Cannot set file flags on %s", fnm);
802 #endif
803 	return;
804 }
805 
806 /*
807  * file_write()
808  *	Write/copy a file (during copy or archive extract). This routine knows
809  *	how to copy files with lseek holes in it. (Which are read as file
810  *	blocks containing all 0's but do not have any file blocks associated
811  *	with the data). Typical examples of these are files created by dbm
812  *	variants (.pag files). While the file size of these files are huge, the
813  *	actual storage is quite small (the files are sparse). The problem is
814  *	the holes read as all zeros so are probably stored on the archive that
815  *	way (there is no way to determine if the file block is really a hole,
816  *	we only know that a file block of all zero's can be a hole).
817  *	At this writing, no major archive format knows how to archive files
818  *	with holes. However, on extraction (or during copy, -rw) we have to
819  *	deal with these files. Without detecting the holes, the files can
820  *	consume a lot of file space if just written to disk. This replacement
821  *	for write when passed the basic allocation size of a file system block,
822  *	uses lseek whenever it detects the input data is all 0 within that
823  *	file block. In more detail, the strategy is as follows:
824  *	While the input is all zero keep doing an lseek. Keep track of when we
825  *	pass over file block boundaries. Only write when we hit a non zero
826  *	input. once we have written a file block, we continue to write it to
827  *	the end (we stop looking at the input). When we reach the start of the
828  *	next file block, start checking for zero blocks again. Working on file
829  *	block boundaries significantly reduces the overhead when copying files
830  *	that are NOT very sparse. This overhead (when compared to a write) is
831  *	almost below the measurement resolution on many systems. Without it,
832  *	files with holes cannot be safely copied. It does has a side effect as
833  *	it can put holes into files that did not have them before, but that is
834  *	not a problem since the file contents are unchanged (in fact it saves
835  *	file space). (Except on paging files for diskless clients. But since we
836  *	cannot determine one of those file from here, we ignore them). If this
837  *	ever ends up on a system where CTG files are supported and the holes
838  *	are not desired, just do a conditional test in those routines that
839  *	call file_write() and have it call write() instead. BEFORE CLOSING THE
840  *	FILE, make sure to call file_flush() when the last write finishes with
841  *	an empty block. A lot of file systems will not create an lseek hole at
842  *	the end. In this case we drop a single 0 at the end to force the
843  *	trailing 0's in the file.
844  *	---Parameters---
845  *	rem: how many bytes left in this file system block
846  *	isempt: have we written to the file block yet (is it empty)
847  *	sz: basic file block allocation size
848  *	cnt: number of bytes on this write
849  *	str: buffer to write
850  * Return:
851  *	number of bytes written, -1 on write (or lseek) error.
852  */
853 
854 int
855 file_write(int fd, char *str, int cnt, int *rem, int *isempt, int sz,
856 	char *name)
857 {
858 	char *pt;
859 	char *end;
860 	int wcnt;
861 	char *st = str;
862 	char **strp;
863 
864 	/*
865 	 * while we have data to process
866 	 */
867 	while (cnt) {
868 		if (!*rem) {
869 			/*
870 			 * We are now at the start of file system block again
871 			 * (or what we think one is...). start looking for
872 			 * empty blocks again
873 			 */
874 			*isempt = 1;
875 			*rem = sz;
876 		}
877 
878 		/*
879 		 * only examine up to the end of the current file block or
880 		 * remaining characters to write, whatever is smaller
881 		 */
882 		wcnt = MIN(cnt, *rem);
883 		cnt -= wcnt;
884 		*rem -= wcnt;
885 		if (*isempt) {
886 			/*
887 			 * have not written to this block yet, so we keep
888 			 * looking for zero's
889 			 */
890 			pt = st;
891 			end = st + wcnt;
892 
893 			/*
894 			 * look for a zero filled buffer
895 			 */
896 			while ((pt < end) && (*pt == '\0'))
897 				++pt;
898 
899 			if (pt == end) {
900 				/*
901 				 * skip, buf is empty so far
902 				 */
903 				if (fd > -1 &&
904 				    lseek(fd, (off_t)wcnt, SEEK_CUR) < 0) {
905 					syswarn(1, errno, "File seek on %s",
906 					    name);
907 					return(-1);
908 				}
909 				st = pt;
910 				continue;
911 			}
912 			/*
913 			 * drat, the buf is not zero filled
914 			 */
915 			*isempt = 0;
916 		}
917 
918 		/*
919 		 * have non-zero data in this file system block, have to write
920 		 */
921 		switch (fd) {
922 		case -1:
923 			strp = &gnu_name_string;
924 			break;
925 		case -2:
926 			strp = &gnu_link_string;
927 			break;
928 		default:
929 			strp = NULL;
930 			break;
931 		}
932 		if (strp) {
933 			if (*strp)
934 				err(1, "WARNING! Major Internal Error! GNU hack Failing!");
935 			*strp = malloc(wcnt + 1);
936 			if (*strp == NULL) {
937 				tty_warn(1, "Out of memory");
938 				return(-1);
939 			}
940 			strlcpy(*strp, st, wcnt);
941 			break;
942 		} else if (xwrite(fd, st, wcnt) != wcnt) {
943 			syswarn(1, errno, "Failed write to file %s", name);
944 			return(-1);
945 		}
946 		st += wcnt;
947 	}
948 	return(st - str);
949 }
950 
951 /*
952  * file_flush()
953  *	when the last file block in a file is zero, many file systems will not
954  *	let us create a hole at the end. To get the last block with zeros, we
955  *	write the last BYTE with a zero (back up one byte and write a zero).
956  */
957 
958 void
959 file_flush(int fd, char *fname, int isempt)
960 {
961 	static char blnk[] = "\0";
962 
963 	/*
964 	 * silly test, but make sure we are only called when the last block is
965 	 * filled with all zeros.
966 	 */
967 	if (!isempt)
968 		return;
969 
970 	/*
971 	 * move back one byte and write a zero
972 	 */
973 	if (lseek(fd, (off_t)-1, SEEK_CUR) < 0) {
974 		syswarn(1, errno, "Failed seek on file %s", fname);
975 		return;
976 	}
977 
978 	if (write_with_restart(fd, blnk, 1) < 0)
979 		syswarn(1, errno, "Failed write to file %s", fname);
980 	return;
981 }
982 
983 /*
984  * rdfile_close()
985  *	close a file we have been reading (to copy or archive). If we have to
986  *	reset access time (tflag) do so (the times are stored in arcn).
987  */
988 
989 void
990 rdfile_close(ARCHD *arcn, int *fd)
991 {
992 	/*
993 	 * make sure the file is open
994 	 */
995 	if (*fd < 0)
996 		return;
997 
998 	(void)close(*fd);
999 	*fd = -1;
1000 	if (!tflag)
1001 		return;
1002 
1003 	/*
1004 	 * user wants last access time reset
1005 	 */
1006 	set_ftime(arcn->org_name, arcn->sb.st_mtime, arcn->sb.st_atime, 1);
1007 	return;
1008 }
1009 
1010 /*
1011  * set_crc()
1012  *	read a file to calculate its crc. This is a real drag. Archive formats
1013  *	that have this, end up reading the file twice (we have to write the
1014  *	header WITH the crc before writing the file contents. Oh well...
1015  * Return:
1016  *	0 if was able to calculate the crc, -1 otherwise
1017  */
1018 
1019 int
1020 set_crc(ARCHD *arcn, int fd)
1021 {
1022 	int i;
1023 	int res;
1024 	off_t cpcnt = 0L;
1025 	u_long size;
1026 	unsigned long crc = 0L;
1027 	char tbuf[FILEBLK];
1028 	struct stat sb;
1029 
1030 	if (fd < 0) {
1031 		/*
1032 		 * hmm, no fd, should never happen. well no crc then.
1033 		 */
1034 		arcn->crc = 0L;
1035 		return(0);
1036 	}
1037 
1038 	if ((size = (u_long)arcn->sb.st_blksize) > (u_long)sizeof(tbuf))
1039 		size = (u_long)sizeof(tbuf);
1040 
1041 	/*
1042 	 * read all the bytes we think that there are in the file. If the user
1043 	 * is trying to archive an active file, forget this file.
1044 	 */
1045 	for(;;) {
1046 		if ((res = read(fd, tbuf, size)) <= 0)
1047 			break;
1048 		cpcnt += res;
1049 		for (i = 0; i < res; ++i)
1050 			crc += (tbuf[i] & 0xff);
1051 	}
1052 
1053 	/*
1054 	 * safety check. we want to avoid archiving files that are active as
1055 	 * they can create inconsistant archive copies.
1056 	 */
1057 	if (cpcnt != arcn->sb.st_size)
1058 		tty_warn(1, "File changed size %s", arcn->org_name);
1059 	else if (fstat(fd, &sb) < 0)
1060 		syswarn(1, errno, "Failed stat on %s", arcn->org_name);
1061 	else if (arcn->sb.st_mtime != sb.st_mtime)
1062 		tty_warn(1, "File %s was modified during read", arcn->org_name);
1063 	else if (lseek(fd, (off_t)0L, SEEK_SET) < 0)
1064 		syswarn(1, errno, "File rewind failed on: %s", arcn->org_name);
1065 	else {
1066 		arcn->crc = crc;
1067 		return(0);
1068 	}
1069 	return(-1);
1070 }
1071