xref: /netbsd-src/bin/pax/file_subs.c (revision 404fbe5fb94ca1e054339640cabb2801ce52dd30)
1 /*	$NetBSD: file_subs.c,v 1.61 2008/01/10 04:24:51 tls 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 1.61 2008/01/10 04:24:51 tls 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 static mode_t
88 apply_umask(mode_t mode)
89 {
90 	static mode_t cached_umask;
91 	static int cached_umask_valid;
92 
93 	if (!cached_umask_valid) {
94 		cached_umask = umask(0);
95 		umask(cached_umask);
96 		cached_umask_valid = 1;
97 	}
98 
99 	return mode & ~cached_umask;
100 }
101 
102 /*
103  * file_creat()
104  *	Create and open a file.
105  * Return:
106  *	file descriptor or -1 for failure
107  */
108 
109 int
110 file_creat(ARCHD *arcn, int write_to_hardlink)
111 {
112 	int fd = -1;
113 	int oerrno;
114 
115 	/*
116 	 * Some horribly busted tar implementations, have directory nodes
117 	 * that end in a /, but they mark as files. Compensate for that
118 	 * by not creating a directory node at this point, but a file node,
119 	 * and not creating the temp file.
120 	 */
121 	if (arcn->nlen != 0 && arcn->name[arcn->nlen - 1] == '/') {
122 		if (!warn_broken) {
123 			tty_warn(0, "Archive was created with a broken tar;"
124 			    " file `%s' is a directory, but marked as plain.",
125 			    arcn->name);
126 			warn_broken = 1;
127 		}
128 		return -1;
129 	}
130 
131 	/*
132 	 * In "cpio" archives it's usually the last record of a set of
133 	 * hardlinks which includes the contents of the file. We cannot
134 	 * use a tempory file in that case because we couldn't link it
135 	 * with the existing other hardlinks after restoring the contents
136 	 * to it. And it's also useless to create the hardlink under a
137 	 * temporary name because the other hardlinks would have partial
138 	 * contents while restoring.
139 	 */
140 	if (write_to_hardlink)
141 		return (open(arcn->name, O_TRUNC | O_EXCL | O_RDWR, 0));
142 
143 	/*
144 	 * Create a temporary file name so that the file doesn't have partial
145 	 * contents while restoring.
146 	 */
147 	arcn->tmp_name = malloc(arcn->nlen + 8);
148 	if (arcn->tmp_name == NULL) {
149 		syswarn(1, errno, "Cannot malloc %d bytes", arcn->nlen + 8);
150 		return -1;
151 	}
152 	if (xtmp_name != NULL)
153 		abort();
154 	xtmp_name = arcn->tmp_name;
155 
156 	for (;;) {
157 		/*
158 		 * try to create the temporary file we use to restore the
159 		 * contents info.  if this fails, keep checking all the nodes
160 		 * in the path until chk_path() finds that it cannot fix
161 		 * anything further.  if that happens we just give up.
162 		 */
163 		(void)snprintf(arcn->tmp_name, arcn->nlen + 8, "%s.XXXXXX",
164 		    arcn->name);
165 		fd = mkstemp(arcn->tmp_name);
166 		if (fd >= 0)
167 			break;
168 		oerrno = errno;
169 		if (nodirs || chk_path(arcn->name,arcn->sb.st_uid,arcn->sb.st_gid) < 0) {
170 			(void)fflush(listf);
171 			syswarn(1, oerrno, "Cannot create %s", arcn->tmp_name);
172 			xtmp_name = NULL;
173 			free(arcn->tmp_name);
174 			arcn->tmp_name = NULL;
175 			return -1;
176 		}
177 	}
178 	return fd;
179 }
180 
181 /*
182  * file_close()
183  *	Close file descriptor to a file just created by pax. Sets modes,
184  *	ownership and times as required.
185  * Return:
186  *	0 for success, -1 for failure
187  */
188 
189 void
190 file_close(ARCHD *arcn, int fd)
191 {
192 	char *tmp_name;
193 	int res;
194 
195 	if (fd < 0)
196 		return;
197 
198 	tmp_name = (arcn->tmp_name != NULL) ? arcn->tmp_name : arcn->name;
199 
200 	if (close(fd) < 0)
201 		syswarn(0, errno, "Cannot close file descriptor on %s",
202 		    tmp_name);
203 
204 	/*
205 	 * set owner/groups first as this may strip off mode bits we want
206 	 * then set file permission modes. Then set file access and
207 	 * modification times.
208 	 */
209 	if (pids)
210 		res = set_ids(tmp_name, arcn->sb.st_uid, arcn->sb.st_gid);
211 	else
212 		res = 0;
213 
214 	/*
215 	 * IMPORTANT SECURITY NOTE:
216 	 * if not preserving mode or we cannot set uid/gid, then PROHIBIT
217 	 * set uid/gid bits but restore the file modes (since mkstemp doesn't).
218 	 */
219 	if (!pmode || res)
220 		arcn->sb.st_mode &= ~SETBITS(0);
221 	if (pmode)
222 		set_pmode(tmp_name, arcn->sb.st_mode);
223 	else
224 		set_pmode(tmp_name,
225 		    apply_umask((arcn->sb.st_mode & FILEBITS(0))));
226 	if (patime || pmtime)
227 		set_ftime(tmp_name, arcn->sb.st_mtime,
228 		    arcn->sb.st_atime, 0, 0);
229 
230 	/* Did we write directly to the target file? */
231 	if (arcn->tmp_name == NULL)
232 		return;
233 
234 	/*
235 	 * Finally, now the temp file is fully instantiated rename it to
236 	 * the desired file name.
237 	 */
238 	if (rename(tmp_name, arcn->name) < 0) {
239 		syswarn(0, errno, "Cannot rename %s to %s",
240 		    tmp_name, arcn->name);
241 		(void)unlink(tmp_name);
242 	}
243 
244 #if HAVE_STRUCT_STAT_ST_FLAGS
245 	if (pfflags && arcn->type != PAX_SLK)
246 		set_chflags(arcn->name, arcn->sb.st_flags);
247 #endif
248 
249 	free(arcn->tmp_name);
250 	arcn->tmp_name = NULL;
251 	xtmp_name = NULL;
252 }
253 
254 /*
255  * lnk_creat()
256  *	Create a hard link to arcn->ln_name from arcn->name. arcn->ln_name
257  *	must exist;
258  * Return:
259  *	0 if ok, -1 otherwise
260  */
261 
262 int
263 lnk_creat(ARCHD *arcn, int *payload)
264 {
265 	struct stat sb;
266 
267 	/*
268 	 * Check if this hardlink carries the "payload". In "cpio" archives
269 	 * it's usually the last record of a set of hardlinks which includes
270 	 * the contents of the file.
271 	 *
272 	 */
273 	*payload = S_ISREG(arcn->sb.st_mode) &&
274 	    (arcn->sb.st_size > 0) && (arcn->sb.st_size <= arcn->skip);
275 
276 	/*
277 	 * We may be running as root, so we have to be sure that link target
278 	 * is not a directory, so we lstat and check
279 	 */
280 	if (lstat(arcn->ln_name, &sb) < 0) {
281 		syswarn(1, errno, "Cannot link to %s from %s", arcn->ln_name,
282 		    arcn->name);
283 		return -1;
284 	}
285 
286 	if (S_ISDIR(sb.st_mode)) {
287 		tty_warn(1, "A hard link to the directory %s is not allowed",
288 		    arcn->ln_name);
289 		return -1;
290 	}
291 
292 	return mk_link(arcn->ln_name, &sb, arcn->name, 0);
293 }
294 
295 /*
296  * cross_lnk()
297  *	Create a hard link to arcn->org_name from arcn->name. Only used in copy
298  *	with the -l flag. No warning or error if this does not succeed (we will
299  *	then just create the file)
300  * Return:
301  *	1 if copy() should try to create this file node
302  *	0 if cross_lnk() ok, -1 for fatal flaw (like linking to self).
303  */
304 
305 int
306 cross_lnk(ARCHD *arcn)
307 {
308 	/*
309 	 * try to make a link to original file (-l flag in copy mode). make
310 	 * sure we do not try to link to directories in case we are running as
311 	 * root (and it might succeed).
312 	 */
313 	if (arcn->type == PAX_DIR)
314 		return 1;
315 	return mk_link(arcn->org_name, &(arcn->sb), arcn->name, 1);
316 }
317 
318 /*
319  * chk_same()
320  *	In copy mode if we are not trying to make hard links between the src
321  *	and destinations, make sure we are not going to overwrite ourselves by
322  *	accident. This slows things down a little, but we have to protect all
323  *	those people who make typing errors.
324  * Return:
325  *	1 the target does not exist, go ahead and copy
326  *	0 skip it file exists (-k) or may be the same as source file
327  */
328 
329 int
330 chk_same(ARCHD *arcn)
331 {
332 	struct stat sb;
333 
334 	/*
335 	 * if file does not exist, return. if file exists and -k, skip it
336 	 * quietly
337 	 */
338 	if (lstat(arcn->name, &sb) < 0)
339 		return 1;
340 	if (kflag)
341 		return 0;
342 
343 	/*
344 	 * better make sure the user does not have src == dest by mistake
345 	 */
346 	if ((arcn->sb.st_dev == sb.st_dev) && (arcn->sb.st_ino == sb.st_ino)) {
347 		tty_warn(1, "Unable to copy %s, file would overwrite itself",
348 		    arcn->name);
349 		return 0;
350 	}
351 	return 1;
352 }
353 
354 /*
355  * mk_link()
356  *	try to make a hard link between two files. if ign set, we do not
357  *	complain.
358  * Return:
359  *	0 if successful (or we are done with this file but no error, such as
360  *	finding the from file exists and the user has set -k).
361  *	1 when ign was set to indicates we could not make the link but we
362  *	should try to copy/extract the file as that might work (and is an
363  *	allowed option). -1 an error occurred.
364  */
365 
366 static int
367 mk_link(char *to, struct stat *to_sb, char *from, int ign)
368 {
369 	struct stat sb;
370 	int oerrno;
371 
372 	/*
373 	 * if from file exists, it has to be unlinked to make the link. If the
374 	 * file exists and -k is set, skip it quietly
375 	 */
376 	if (lstat(from, &sb) == 0) {
377 		if (kflag)
378 			return 0;
379 
380 		/*
381 		 * make sure it is not the same file, protect the user
382 		 */
383 		if ((to_sb->st_dev==sb.st_dev)&&(to_sb->st_ino == sb.st_ino)) {
384 			tty_warn(1, "Cannot link file %s to itself", to);
385 			return -1;
386 		}
387 
388 		/*
389 		 * try to get rid of the file, based on the type
390 		 */
391 		if (S_ISDIR(sb.st_mode) && strcmp(from, ".") != 0) {
392 			if (rmdir(from) < 0) {
393 				syswarn(1, errno, "Cannot remove %s", from);
394 				return -1;
395 			}
396 		} else if (unlink(from) < 0) {
397 			if (!ign) {
398 				syswarn(1, errno, "Cannot remove %s", from);
399 				return -1;
400 			}
401 			return 1;
402 		}
403 	}
404 
405 	/*
406 	 * from file is gone (or did not exist), try to make the hard link.
407 	 * if it fails, check the path and try it again (if chk_path() says to
408 	 * try again)
409 	 */
410 	for (;;) {
411 		if (link(to, from) == 0)
412 			break;
413 		oerrno = errno;
414 		if (chk_path(from, to_sb->st_uid, to_sb->st_gid) == 0)
415 			continue;
416 		if (!ign) {
417 			syswarn(1, oerrno, "Cannot link to %s from %s", to,
418 			    from);
419 			return -1;
420 		}
421 		return 1;
422 	}
423 
424 	/*
425 	 * all right the link was made
426 	 */
427 	return 0;
428 }
429 
430 /*
431  * node_creat()
432  *	create an entry in the file system (other than a file or hard link).
433  *	If successful, sets uid/gid modes and times as required.
434  * Return:
435  *	0 if ok, -1 otherwise
436  */
437 
438 int
439 node_creat(ARCHD *arcn)
440 {
441 	int res;
442 	int ign = 0;
443 	int oerrno;
444 	int pass = 0;
445 	mode_t file_mode;
446 	struct stat sb;
447 	char target[MAXPATHLEN];
448 	char *nm = arcn->name;
449 	int len;
450 
451 	/*
452 	 * create node based on type, if that fails try to unlink the node and
453 	 * try again. finally check the path and try again. As noted in the
454 	 * file and link creation routines, this method seems to exhibit the
455 	 * best performance in general use workloads.
456 	 */
457 	file_mode = arcn->sb.st_mode & FILEBITS(arcn->type == PAX_DIR);
458 
459 	for (;;) {
460 		switch (arcn->type) {
461 		case PAX_DIR:
462 			/*
463 			 * If -h (or -L) was given in tar-mode, follow the
464 			 * potential symlink chain before trying to create the
465 			 * directory.
466 			 */
467 			if (strcmp(NM_TAR, argv0) == 0 && Lflag) {
468 				while (lstat(nm, &sb) == 0 &&
469 				    S_ISLNK(sb.st_mode)) {
470 					len = readlink(nm, target,
471 					    sizeof target - 1);
472 					if (len == -1) {
473 						syswarn(0, errno,
474 						   "cannot follow symlink %s "
475 						   "in chain for %s",
476 						    nm, arcn->name);
477 						res = -1;
478 						goto badlink;
479 					}
480 					target[len] = '\0';
481 					nm = target;
482 				}
483 			}
484 			res = domkdir(nm, file_mode);
485 badlink:
486 			if (ign)
487 				res = 0;
488 			break;
489 		case PAX_CHR:
490 			file_mode |= S_IFCHR;
491 			res = mknod(nm, file_mode, arcn->sb.st_rdev);
492 			break;
493 		case PAX_BLK:
494 			file_mode |= S_IFBLK;
495 			res = mknod(nm, file_mode, arcn->sb.st_rdev);
496 			break;
497 		case PAX_FIF:
498 			res = mkfifo(nm, file_mode);
499 			break;
500 		case PAX_SCK:
501 			/*
502 			 * Skip sockets, operation has no meaning under BSD
503 			 */
504 			tty_warn(0,
505 			    "%s skipped. Sockets cannot be copied or extracted",
506 			    nm);
507 			return (-1);
508 		case PAX_SLK:
509 			res = symlink(arcn->ln_name, nm);
510 			break;
511 		case PAX_CTG:
512 		case PAX_HLK:
513 		case PAX_HRG:
514 		case PAX_REG:
515 		default:
516 			/*
517 			 * we should never get here
518 			 */
519 			tty_warn(0, "%s has an unknown file type, skipping",
520 			    nm);
521 			return (-1);
522 		}
523 
524 		/*
525 		 * if we were able to create the node break out of the loop,
526 		 * otherwise try to unlink the node and try again. if that
527 		 * fails check the full path and try a final time.
528 		 */
529 		if (res == 0)
530 			break;
531 
532 		/*
533 		 * we failed to make the node
534 		 */
535 		oerrno = errno;
536 		switch (pass++) {
537 		case 0:
538 			if ((ign = unlnk_exist(nm, arcn->type)) < 0)
539 				return (-1);
540 			continue;
541 
542 		case 1:
543 			if (nodirs ||
544 			    chk_path(nm, arcn->sb.st_uid,
545 			    arcn->sb.st_gid) < 0) {
546 				syswarn(1, oerrno, "Cannot create %s", nm);
547 				return (-1);
548 			}
549 			continue;
550 		}
551 
552 		/*
553 		 * it must be a file that exists but we can't create or
554 		 * remove, but we must avoid the infinite loop.
555 		 */
556 		break;
557 	}
558 
559 	/*
560 	 * we were able to create the node. set uid/gid, modes and times
561 	 */
562 	if (pids)
563 		res = set_ids(nm, arcn->sb.st_uid, arcn->sb.st_gid);
564 	else
565 		res = 0;
566 
567 	/*
568 	 * IMPORTANT SECURITY NOTE:
569 	 * if not preserving mode or we cannot set uid/gid, then PROHIBIT any
570 	 * set uid/gid bits
571 	 */
572 	if (!pmode || res)
573 		arcn->sb.st_mode &= ~SETBITS(arcn->type == PAX_DIR);
574 	if (pmode)
575 		set_pmode(arcn->name, arcn->sb.st_mode);
576 
577 	if (arcn->type == PAX_DIR && strcmp(NM_CPIO, argv0) != 0) {
578 		/*
579 		 * Dirs must be processed again at end of extract to set times
580 		 * and modes to agree with those stored in the archive. However
581 		 * to allow extract to continue, we may have to also set owner
582 		 * rights. This allows nodes in the archive that are children
583 		 * of this directory to be extracted without failure. Both time
584 		 * and modes will be fixed after the entire archive is read and
585 		 * before pax exits.
586 		 */
587 		if (access(nm, R_OK | W_OK | X_OK) < 0) {
588 			if (lstat(nm, &sb) < 0) {
589 				syswarn(0, errno,"Cannot access %s (stat)",
590 				    arcn->name);
591 				set_pmode(nm,file_mode | S_IRWXU);
592 			} else {
593 				/*
594 				 * We have to add rights to the dir, so we make
595 				 * sure to restore the mode. The mode must be
596 				 * restored AS CREATED and not as stored if
597 				 * pmode is not set.
598 				 */
599 				set_pmode(nm, ((sb.st_mode &
600 				    FILEBITS(arcn->type == PAX_DIR)) |
601 				    S_IRWXU));
602 				if (!pmode)
603 					arcn->sb.st_mode = sb.st_mode;
604 			}
605 
606 			/*
607 			 * we have to force the mode to what was set here,
608 			 * since we changed it from the default as created.
609 			 */
610 			add_dir(nm, arcn->nlen, &(arcn->sb), 1);
611 		} else if (pmode || patime || pmtime)
612 			add_dir(nm, arcn->nlen, &(arcn->sb), 0);
613 	}
614 
615 	if (patime || pmtime)
616 		set_ftime(arcn->name, arcn->sb.st_mtime,
617 		    arcn->sb.st_atime, 0, (arcn->type == PAX_SLK) ? 1 : 0);
618 
619 #if HAVE_STRUCT_STAT_ST_FLAGS
620 	if (pfflags && arcn->type != PAX_SLK)
621 		set_chflags(arcn->name, arcn->sb.st_flags);
622 #endif
623 	return 0;
624 }
625 
626 /*
627  * unlnk_exist()
628  *	Remove node from file system with the specified name. We pass the type
629  *	of the node that is going to replace it. When we try to create a
630  *	directory and find that it already exists, we allow processing to
631  *	continue as proper modes etc will always be set for it later on.
632  * Return:
633  *	0 is ok to proceed, no file with the specified name exists
634  *	-1 we were unable to remove the node, or we should not remove it (-k)
635  *	1 we found a directory and we were going to create a directory.
636  */
637 
638 int
639 unlnk_exist(char *name, int type)
640 {
641 	struct stat sb;
642 
643 	/*
644 	 * the file does not exist, or -k we are done
645 	 */
646 	if (lstat(name, &sb) < 0)
647 		return 0;
648 	if (kflag)
649 		return -1;
650 
651 	if (S_ISDIR(sb.st_mode)) {
652 		/*
653 		 * try to remove a directory, if it fails and we were going to
654 		 * create a directory anyway, tell the caller (return a 1).
655 		 *
656 		 * don't try to remove the directory if the name is "."
657 		 * otherwise later file/directory creation fails.
658 		 */
659 		if (strcmp(name, ".") == 0)
660 			return 1;
661 		if (rmdir(name) < 0) {
662 			if (type == PAX_DIR)
663 				return 1;
664 			syswarn(1, errno, "Cannot remove directory %s", name);
665 			return -1;
666 		}
667 		return 0;
668 	}
669 
670 	/*
671 	 * try to get rid of all non-directory type nodes
672 	 */
673 	if (unlink(name) < 0) {
674 		(void)fflush(listf);
675 		syswarn(1, errno, "Cannot unlink %s", name);
676 		return -1;
677 	}
678 	return 0;
679 }
680 
681 /*
682  * chk_path()
683  *	We were trying to create some kind of node in the file system and it
684  *	failed. chk_path() makes sure the path up to the node exists and is
685  *	writable. When we have to create a directory that is missing along the
686  *	path somewhere, the directory we create will be set to the same
687  *	uid/gid as the file has (when uid and gid are being preserved).
688  *	NOTE: this routine is a real performance loss. It is only used as a
689  *	last resort when trying to create entries in the file system.
690  * Return:
691  *	-1 when it could find nothing it is allowed to fix.
692  *	0 otherwise
693  */
694 
695 int
696 chk_path(char *name, uid_t st_uid, gid_t st_gid)
697 {
698 	char *spt = name;
699 	struct stat sb;
700 	int retval = -1;
701 
702 	/*
703 	 * watch out for paths with nodes stored directly in / (e.g. /bozo)
704 	 */
705 	if (*spt == '/')
706 		++spt;
707 
708 	for(;;) {
709 		/*
710 		 * work forward from the first / and check each part of
711 		 * the path
712 		 */
713 		spt = strchr(spt, '/');
714 		if (spt == NULL)
715 			break;
716 		*spt = '\0';
717 
718 		/*
719 		 * if it exists we assume it is a directory, it is not within
720 		 * the spec (at least it seems to read that way) to alter the
721 		 * file system for nodes NOT EXPLICITLY stored on the archive.
722 		 * If that assumption is changed, you would test the node here
723 		 * and figure out how to get rid of it (probably like some
724 		 * recursive unlink()) or fix up the directory permissions if
725 		 * required (do an access()).
726 		 */
727 		if (lstat(name, &sb) == 0) {
728 			*(spt++) = '/';
729 			continue;
730 		}
731 
732 		/*
733 		 * the path fails at this point, see if we can create the
734 		 * needed directory and continue on
735 		 */
736 		if (domkdir(name, S_IRWXU | S_IRWXG | S_IRWXO) == -1) {
737 			*spt = '/';
738 			retval = -1;
739 			break;
740 		}
741 
742 		/*
743 		 * we were able to create the directory. We will tell the
744 		 * caller that we found something to fix, and it is ok to try
745 		 * and create the node again.
746 		 */
747 		retval = 0;
748 		if (pids)
749 			(void)set_ids(name, st_uid, st_gid);
750 
751 		/*
752 		 * make sure the user doesn't have some strange umask that
753 		 * causes this newly created directory to be unusable. We fix
754 		 * the modes and restore them back to the creation default at
755 		 * the end of pax
756 		 */
757 		if ((access(name, R_OK | W_OK | X_OK) < 0) &&
758 		    (lstat(name, &sb) == 0)) {
759 			set_pmode(name, ((sb.st_mode & FILEBITS(0)) |
760 			    S_IRWXU));
761 			add_dir(name, spt - name, &sb, 1);
762 		}
763 		*(spt++) = '/';
764 		continue;
765 	}
766 	/*
767 	 * We perform one final check here, because if someone else
768 	 * created the directory in parallel with us, we might return
769 	 * the wrong error code, even if the directory exists now.
770 	 */
771 	if (retval == -1 && stat(name, &sb) == 0 && S_ISDIR(sb.st_mode))
772 		retval = 0;
773 	return retval;
774 }
775 
776 /*
777  * set_ftime()
778  *	Set the access time and modification time for a named file. If frc
779  *	is non-zero we force these times to be set even if the user did not
780  *	request access and/or modification time preservation (this is also
781  *	used by -t to reset access times).
782  *	When ign is zero, only those times the user has asked for are set, the
783  *	other ones are left alone. We do not assume the un-documented feature
784  *	of many utimes() implementations that consider a 0 time value as a do
785  *	not set request.
786  *
787  *	Unfortunately, there are systems where lutimes() is present but does
788  *	not work on some filesystem types, which cannot be detected at
789  * 	compile time.  This requires passing down symlink knowledge into
790  *	this function to obtain correct operation.  Linux with XFS is one
791  * 	example of such a system.
792  */
793 
794 void
795 set_ftime(char *fnm, time_t mtime, time_t atime, int frc, int slk)
796 {
797 	struct timeval tv[2];
798 	struct stat sb;
799 
800 	tv[0].tv_sec = (long)atime;
801 	tv[0].tv_usec = 0;
802 	tv[1].tv_sec = (long)mtime;
803 	tv[1].tv_usec = 0;
804 	if (!frc && (!patime || !pmtime)) {
805 		/*
806 		 * if we are not forcing, only set those times the user wants
807 		 * set. We get the current values of the times if we need them.
808 		 */
809 		if (lstat(fnm, &sb) == 0) {
810 #if BSD4_4 && !HAVE_NBTOOL_CONFIG_H
811 			if (!patime)
812 				TIMESPEC_TO_TIMEVAL(&tv[0], &sb.st_atimespec);
813 			if (!pmtime)
814 				TIMESPEC_TO_TIMEVAL(&tv[1], &sb.st_mtimespec);
815 #else
816 			if (!patime)
817 				tv[0].tv_sec = sb.st_atime;
818 			if (!pmtime)
819 				tv[1].tv_sec = sb.st_mtime;
820 #endif
821 		} else
822 			syswarn(0, errno, "Cannot obtain file stats %s", fnm);
823 	}
824 
825 	/*
826 	 * set the times
827 	 */
828 #if HAVE_LUTIMES
829 	if (lutimes(fnm, tv) == 0)
830 		return;
831 	if (errno != ENOSYS)	/* XXX linux: lutimes is per-FS */
832 		goto bad;
833 #endif
834 	if (slk)
835 		return;
836 	if (utimes(fnm, tv) == -1)
837 		goto bad;
838 	return;
839 bad:
840 	syswarn(1, errno, "Access/modification time set failed on: %s", fnm);
841 }
842 
843 /*
844  * set_ids()
845  *	set the uid and gid of a file system node
846  * Return:
847  *	0 when set, -1 on failure
848  */
849 
850 int
851 set_ids(char *fnm, uid_t uid, gid_t gid)
852 {
853 	if (geteuid() == 0)
854 		if (lchown(fnm, uid, gid)) {
855 			(void)fflush(listf);
856 			syswarn(1, errno, "Cannot set file uid/gid of %s",
857 			    fnm);
858 			return -1;
859 		}
860 	return 0;
861 }
862 
863 /*
864  * set_pmode()
865  *	Set file access mode
866  */
867 
868 void
869 set_pmode(char *fnm, mode_t mode)
870 {
871 	mode &= A_BITS;
872 	if (lchmod(fnm, mode)) {
873 		(void)fflush(listf);
874 		syswarn(1, errno, "Cannot set permissions on %s", fnm);
875 	}
876 	return;
877 }
878 
879 /*
880  * set_chflags()
881  *	Set 4.4BSD file flags
882  */
883 void
884 set_chflags(char *fnm, u_int32_t flags)
885 {
886 
887 #if 0
888 	if (chflags(fnm, flags) < 0 && errno != EOPNOTSUPP)
889 		syswarn(1, errno, "Cannot set file flags on %s", fnm);
890 #endif
891 	return;
892 }
893 
894 /*
895  * file_write()
896  *	Write/copy a file (during copy or archive extract). This routine knows
897  *	how to copy files with lseek holes in it. (Which are read as file
898  *	blocks containing all 0's but do not have any file blocks associated
899  *	with the data). Typical examples of these are files created by dbm
900  *	variants (.pag files). While the file size of these files are huge, the
901  *	actual storage is quite small (the files are sparse). The problem is
902  *	the holes read as all zeros so are probably stored on the archive that
903  *	way (there is no way to determine if the file block is really a hole,
904  *	we only know that a file block of all zero's can be a hole).
905  *	At this writing, no major archive format knows how to archive files
906  *	with holes. However, on extraction (or during copy, -rw) we have to
907  *	deal with these files. Without detecting the holes, the files can
908  *	consume a lot of file space if just written to disk. This replacement
909  *	for write when passed the basic allocation size of a file system block,
910  *	uses lseek whenever it detects the input data is all 0 within that
911  *	file block. In more detail, the strategy is as follows:
912  *	While the input is all zero keep doing an lseek. Keep track of when we
913  *	pass over file block boundaries. Only write when we hit a non zero
914  *	input. once we have written a file block, we continue to write it to
915  *	the end (we stop looking at the input). When we reach the start of the
916  *	next file block, start checking for zero blocks again. Working on file
917  *	block boundaries significantly reduces the overhead when copying files
918  *	that are NOT very sparse. This overhead (when compared to a write) is
919  *	almost below the measurement resolution on many systems. Without it,
920  *	files with holes cannot be safely copied. It does has a side effect as
921  *	it can put holes into files that did not have them before, but that is
922  *	not a problem since the file contents are unchanged (in fact it saves
923  *	file space). (Except on paging files for diskless clients. But since we
924  *	cannot determine one of those file from here, we ignore them). If this
925  *	ever ends up on a system where CTG files are supported and the holes
926  *	are not desired, just do a conditional test in those routines that
927  *	call file_write() and have it call write() instead. BEFORE CLOSING THE
928  *	FILE, make sure to call file_flush() when the last write finishes with
929  *	an empty block. A lot of file systems will not create an lseek hole at
930  *	the end. In this case we drop a single 0 at the end to force the
931  *	trailing 0's in the file.
932  *	---Parameters---
933  *	rem: how many bytes left in this file system block
934  *	isempt: have we written to the file block yet (is it empty)
935  *	sz: basic file block allocation size
936  *	cnt: number of bytes on this write
937  *	str: buffer to write
938  * Return:
939  *	number of bytes written, -1 on write (or lseek) error.
940  */
941 
942 int
943 file_write(int fd, char *str, int cnt, int *rem, int *isempt, int sz,
944 	char *name)
945 {
946 	char *pt;
947 	char *end;
948 	int wcnt;
949 	char *st = str;
950 	char **strp;
951 	size_t *lenp;
952 
953 	/*
954 	 * while we have data to process
955 	 */
956 	while (cnt) {
957 		if (!*rem) {
958 			/*
959 			 * We are now at the start of file system block again
960 			 * (or what we think one is...). start looking for
961 			 * empty blocks again
962 			 */
963 			*isempt = 1;
964 			*rem = sz;
965 		}
966 
967 		/*
968 		 * only examine up to the end of the current file block or
969 		 * remaining characters to write, whatever is smaller
970 		 */
971 		wcnt = MIN(cnt, *rem);
972 		cnt -= wcnt;
973 		*rem -= wcnt;
974 		if (*isempt) {
975 			/*
976 			 * have not written to this block yet, so we keep
977 			 * looking for zero's
978 			 */
979 			pt = st;
980 			end = st + wcnt;
981 
982 			/*
983 			 * look for a zero filled buffer
984 			 */
985 			while ((pt < end) && (*pt == '\0'))
986 				++pt;
987 
988 			if (pt == end) {
989 				/*
990 				 * skip, buf is empty so far
991 				 */
992 				if (fd > -1 &&
993 				    lseek(fd, (off_t)wcnt, SEEK_CUR) < 0) {
994 					syswarn(1, errno, "File seek on %s",
995 					    name);
996 					return -1;
997 				}
998 				st = pt;
999 				continue;
1000 			}
1001 			/*
1002 			 * drat, the buf is not zero filled
1003 			 */
1004 			*isempt = 0;
1005 		}
1006 
1007 		/*
1008 		 * have non-zero data in this file system block, have to write
1009 		 */
1010 		switch (fd) {
1011 		case -PAX_GLF:
1012 			strp = &gnu_name_string;
1013 			lenp = &gnu_name_length;
1014 			break;
1015 		case -PAX_GLL:
1016 			strp = &gnu_link_string;
1017 			lenp = &gnu_link_length;
1018 			break;
1019 		default:
1020 			strp = NULL;
1021 			lenp = NULL;
1022 			break;
1023 		}
1024 		if (strp) {
1025 			char *nstr = *strp ? realloc(*strp, *lenp + wcnt + 1) :
1026 				malloc(wcnt + 1);
1027 			if (nstr == NULL) {
1028 				tty_warn(1, "Out of memory");
1029 				return -1;
1030 			}
1031 			(void)strlcpy(&nstr[*lenp], st, wcnt + 1);
1032 			*strp = nstr;
1033 			*lenp += wcnt;
1034 		} else if (xwrite(fd, st, wcnt) != wcnt) {
1035 			syswarn(1, errno, "Failed write to file %s", name);
1036 			return -1;
1037 		}
1038 		st += wcnt;
1039 	}
1040 	return st - str;
1041 }
1042 
1043 /*
1044  * file_flush()
1045  *	when the last file block in a file is zero, many file systems will not
1046  *	let us create a hole at the end. To get the last block with zeros, we
1047  *	write the last BYTE with a zero (back up one byte and write a zero).
1048  */
1049 
1050 void
1051 file_flush(int fd, char *fname, int isempt)
1052 {
1053 	static char blnk[] = "\0";
1054 
1055 	/*
1056 	 * silly test, but make sure we are only called when the last block is
1057 	 * filled with all zeros.
1058 	 */
1059 	if (!isempt)
1060 		return;
1061 
1062 	/*
1063 	 * move back one byte and write a zero
1064 	 */
1065 	if (lseek(fd, (off_t)-1, SEEK_CUR) < 0) {
1066 		syswarn(1, errno, "Failed seek on file %s", fname);
1067 		return;
1068 	}
1069 
1070 	if (write_with_restart(fd, blnk, 1) < 0)
1071 		syswarn(1, errno, "Failed write to file %s", fname);
1072 	return;
1073 }
1074 
1075 /*
1076  * rdfile_close()
1077  *	close a file we have been reading (to copy or archive). If we have to
1078  *	reset access time (tflag) do so (the times are stored in arcn).
1079  */
1080 
1081 void
1082 rdfile_close(ARCHD *arcn, int *fd)
1083 {
1084 	/*
1085 	 * make sure the file is open
1086 	 */
1087 	if (*fd < 0)
1088 		return;
1089 
1090 	(void)close(*fd);
1091 	*fd = -1;
1092 	if (!tflag)
1093 		return;
1094 
1095 	/*
1096 	 * user wants last access time reset
1097 	 */
1098 	set_ftime(arcn->org_name, arcn->sb.st_mtime, arcn->sb.st_atime, 1, 0);
1099 	return;
1100 }
1101 
1102 /*
1103  * set_crc()
1104  *	read a file to calculate its crc. This is a real drag. Archive formats
1105  *	that have this, end up reading the file twice (we have to write the
1106  *	header WITH the crc before writing the file contents. Oh well...
1107  * Return:
1108  *	0 if was able to calculate the crc, -1 otherwise
1109  */
1110 
1111 int
1112 set_crc(ARCHD *arcn, int fd)
1113 {
1114 	int i;
1115 	int res;
1116 	off_t cpcnt = 0L;
1117 	u_long size;
1118 	unsigned long crc = 0L;
1119 	char tbuf[FILEBLK];
1120 	struct stat sb;
1121 
1122 	if (fd < 0) {
1123 		/*
1124 		 * hmm, no fd, should never happen. well no crc then.
1125 		 */
1126 		arcn->crc = 0L;
1127 		return 0;
1128 	}
1129 
1130 	if ((size = (u_long)arcn->sb.st_blksize) > (u_long)sizeof(tbuf))
1131 		size = (u_long)sizeof(tbuf);
1132 
1133 	/*
1134 	 * read all the bytes we think that there are in the file. If the user
1135 	 * is trying to archive an active file, forget this file.
1136 	 */
1137 	for(;;) {
1138 		if ((res = read(fd, tbuf, size)) <= 0)
1139 			break;
1140 		cpcnt += res;
1141 		for (i = 0; i < res; ++i)
1142 			crc += (tbuf[i] & 0xff);
1143 	}
1144 
1145 	/*
1146 	 * safety check. we want to avoid archiving files that are active as
1147 	 * they can create inconsistent archive copies.
1148 	 */
1149 	if (cpcnt != arcn->sb.st_size)
1150 		tty_warn(1, "File changed size %s", arcn->org_name);
1151 	else if (fstat(fd, &sb) < 0)
1152 		syswarn(1, errno, "Failed stat on %s", arcn->org_name);
1153 	else if (arcn->sb.st_mtime != sb.st_mtime)
1154 		tty_warn(1, "File %s was modified during read", arcn->org_name);
1155 	else if (lseek(fd, (off_t)0L, SEEK_SET) < 0)
1156 		syswarn(1, errno, "File rewind failed on: %s", arcn->org_name);
1157 	else {
1158 		arcn->crc = crc;
1159 		return 0;
1160 	}
1161 	return -1;
1162 }
1163