xref: /minix3/bin/pax/file_subs.c (revision 84d9c625bfea59e274550651111ae9edfdc40fbd)
1 /*	$NetBSD: file_subs.c,v 1.63 2013/07/29 17:46:36 christos Exp $	*/
2 
3 /*-
4  * Copyright (c) 1992 Keith Muller.
5  * Copyright (c) 1992, 1993
6  *	The Regents of the University of California.  All rights reserved.
7  *
8  * This code is derived from software contributed to Berkeley by
9  * Keith Muller of the University of California, San Diego.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  * 3. Neither the name of the University nor the names of its contributors
20  *    may be used to endorse or promote products derived from this software
21  *    without specific prior written permission.
22  *
23  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
24  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
31  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
32  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33  * SUCH DAMAGE.
34  */
35 
36 #if HAVE_NBTOOL_CONFIG_H
37 #include "nbtool_config.h"
38 #endif
39 
40 #include <sys/cdefs.h>
41 #if !defined(lint)
42 #if 0
43 static char sccsid[] = "@(#)file_subs.c	8.1 (Berkeley) 5/31/93";
44 #else
45 __RCSID("$NetBSD: file_subs.c,v 1.63 2013/07/29 17:46:36 christos Exp $");
46 #endif
47 #endif /* not lint */
48 
49 #include <sys/types.h>
50 #include <sys/time.h>
51 #include <sys/stat.h>
52 #include <unistd.h>
53 #include <sys/param.h>
54 #include <fcntl.h>
55 #include <string.h>
56 #include <stdio.h>
57 #include <ctype.h>
58 #include <errno.h>
59 #include <sys/uio.h>
60 #include <stdlib.h>
61 #include "pax.h"
62 #include "extern.h"
63 #include "options.h"
64 
65 #if defined(__minix)
66 #include <utime.h>
67 #endif /* defined(__minix) */
68 
69 char *xtmp_name;
70 
71 static int
72 mk_link(char *,struct stat *,char *, int);
73 
74 static int warn_broken;
75 
76 /*
77  * routines that deal with file operations such as: creating, removing;
78  * and setting access modes, uid/gid and times of files
79  */
80 #define SET_BITS		(S_ISUID | S_ISGID)
81 #define FILE_BITS		(S_IRWXU | S_IRWXG | S_IRWXO)
82 #define A_BITS			(FILE_BITS | SET_BITS | S_ISVTX)
83 
84 /*
85  * The S_ISVTX (sticky bit) can be set by non-superuser on directories
86  * but not other kinds of files.
87  */
88 #define FILEBITS(dir)		((dir) ? (FILE_BITS | S_ISVTX) : FILE_BITS)
89 #define SETBITS(dir)		((dir) ? SET_BITS : (SET_BITS | S_ISVTX))
90 
91 static mode_t
apply_umask(mode_t mode)92 apply_umask(mode_t mode)
93 {
94 	static mode_t cached_umask;
95 	static int cached_umask_valid;
96 
97 	if (!cached_umask_valid) {
98 		cached_umask = umask(0);
99 		umask(cached_umask);
100 		cached_umask_valid = 1;
101 	}
102 
103 	return mode & ~cached_umask;
104 }
105 
106 /*
107  * file_creat()
108  *	Create and open a file.
109  * Return:
110  *	file descriptor or -1 for failure
111  */
112 
113 int
file_creat(ARCHD * arcn,int write_to_hardlink)114 file_creat(ARCHD *arcn, int write_to_hardlink)
115 {
116 	int fd = -1;
117 	int oerrno;
118 
119 	/*
120 	 * Some horribly busted tar implementations, have directory nodes
121 	 * that end in a /, but they mark as files. Compensate for that
122 	 * by not creating a directory node at this point, but a file node,
123 	 * and not creating the temp file.
124 	 */
125 	if (arcn->nlen != 0 && arcn->name[arcn->nlen - 1] == '/') {
126 		if (!warn_broken) {
127 			tty_warn(0, "Archive was created with a broken tar;"
128 			    " file `%s' is a directory, but marked as plain.",
129 			    arcn->name);
130 			warn_broken = 1;
131 		}
132 		return -1;
133 	}
134 
135 	/*
136 	 * In "cpio" archives it's usually the last record of a set of
137 	 * hardlinks which includes the contents of the file. We cannot
138 	 * use a tempory file in that case because we couldn't link it
139 	 * with the existing other hardlinks after restoring the contents
140 	 * to it. And it's also useless to create the hardlink under a
141 	 * temporary name because the other hardlinks would have partial
142 	 * contents while restoring.
143 	 */
144 	if (write_to_hardlink)
145 		return (open(arcn->name, O_TRUNC | O_EXCL | O_RDWR, 0));
146 
147 	/*
148 	 * Create a temporary file name so that the file doesn't have partial
149 	 * contents while restoring.
150 	 */
151 	arcn->tmp_name = malloc(arcn->nlen + 8);
152 	if (arcn->tmp_name == NULL) {
153 		syswarn(1, errno, "Cannot malloc %d bytes", arcn->nlen + 8);
154 		return -1;
155 	}
156 	if (xtmp_name != NULL)
157 		abort();
158 	xtmp_name = arcn->tmp_name;
159 
160 	for (;;) {
161 		/*
162 		 * try to create the temporary file we use to restore the
163 		 * contents info.  if this fails, keep checking all the nodes
164 		 * in the path until chk_path() finds that it cannot fix
165 		 * anything further.  if that happens we just give up.
166 		 */
167 #if defined(__minix)
168 		{
169 			/* For minix, generate the temporary filename
170 			 * conservatively - just write Xes into the last component.
171 			 * There is space because of the malloc().
172 			 */
173 			char *last_slash;
174 			strcpy(arcn->tmp_name, arcn->name);
175 			if(!(last_slash = strrchr(arcn->tmp_name, '/')))
176 				strcpy(arcn->tmp_name, "XXXXXX");
177 			else	strcpy(last_slash+1, "XXXXXX");
178 		}
179 #else
180 		(void)snprintf(arcn->tmp_name, arcn->nlen + 8, "%s.XXXXXX",
181 		    arcn->name);
182 #endif /* defined(__minix) */
183 		fd = mkstemp(arcn->tmp_name);
184 		if (fd >= 0)
185 			break;
186 		oerrno = errno;
187 		if (nodirs || chk_path(arcn->name,arcn->sb.st_uid,arcn->sb.st_gid) < 0) {
188 			(void)fflush(listf);
189 			syswarn(1, oerrno, "Cannot create %s", arcn->tmp_name);
190 			xtmp_name = NULL;
191 			free(arcn->tmp_name);
192 			arcn->tmp_name = NULL;
193 			return -1;
194 		}
195 	}
196 	return fd;
197 }
198 
199 /*
200  * file_close()
201  *	Close file descriptor to a file just created by pax. Sets modes,
202  *	ownership and times as required.
203  * Return:
204  *	0 for success, -1 for failure
205  */
206 
207 void
file_close(ARCHD * arcn,int fd)208 file_close(ARCHD *arcn, int fd)
209 {
210 	char *tmp_name;
211 	int res;
212 
213 	if (fd < 0)
214 		return;
215 
216 	tmp_name = (arcn->tmp_name != NULL) ? arcn->tmp_name : arcn->name;
217 
218 	if (close(fd) < 0)
219 		syswarn(0, errno, "Cannot close file descriptor on %s",
220 		    tmp_name);
221 
222 	/*
223 	 * set owner/groups first as this may strip off mode bits we want
224 	 * then set file permission modes. Then set file access and
225 	 * modification times.
226 	 */
227 	if (pids)
228 		res = set_ids(tmp_name, arcn->sb.st_uid, arcn->sb.st_gid);
229 	else
230 		res = 0;
231 
232 	/*
233 	 * IMPORTANT SECURITY NOTE:
234 	 * if not preserving mode or we cannot set uid/gid, then PROHIBIT
235 	 * set uid/gid bits but restore the file modes (since mkstemp doesn't).
236 	 */
237 	if (!pmode || res)
238 		arcn->sb.st_mode &= ~SETBITS(0);
239 	if (pmode)
240 		set_pmode(tmp_name, arcn->sb.st_mode);
241 	else
242 		set_pmode(tmp_name,
243 		    apply_umask((arcn->sb.st_mode & FILEBITS(0))));
244 	if (patime || pmtime)
245 		set_ftime(tmp_name, arcn->sb.st_mtime,
246 		    arcn->sb.st_atime, 0, 0);
247 
248 	/* Did we write directly to the target file? */
249 	if (arcn->tmp_name == NULL)
250 		return;
251 
252 	/*
253 	 * Finally, now the temp file is fully instantiated rename it to
254 	 * the desired file name.
255 	 */
256 	if (rename(tmp_name, arcn->name) < 0) {
257 		syswarn(0, errno, "Cannot rename %s to %s",
258 		    tmp_name, arcn->name);
259 		(void)unlink(tmp_name);
260 	}
261 
262 #if HAVE_STRUCT_STAT_ST_FLAGS
263 	if (pfflags && arcn->type != PAX_SLK)
264 		set_chflags(arcn->name, arcn->sb.st_flags);
265 #endif
266 
267 	free(arcn->tmp_name);
268 	arcn->tmp_name = NULL;
269 	xtmp_name = NULL;
270 }
271 
272 /*
273  * lnk_creat()
274  *	Create a hard link to arcn->ln_name from arcn->name. arcn->ln_name
275  *	must exist;
276  * Return:
277  *	0 if ok, -1 otherwise
278  */
279 
280 int
lnk_creat(ARCHD * arcn,int * payload)281 lnk_creat(ARCHD *arcn, int *payload)
282 {
283 	struct stat sb;
284 
285 	/*
286 	 * Check if this hardlink carries the "payload". In "cpio" archives
287 	 * it's usually the last record of a set of hardlinks which includes
288 	 * the contents of the file.
289 	 *
290 	 */
291 	*payload = S_ISREG(arcn->sb.st_mode) &&
292 	    (arcn->sb.st_size > 0) && (arcn->sb.st_size <= arcn->skip);
293 
294 	/*
295 	 * We may be running as root, so we have to be sure that link target
296 	 * is not a directory, so we lstat and check. XXX: This is still racy.
297 	 */
298 	if (lstat(arcn->ln_name, &sb) != -1 && S_ISDIR(sb.st_mode)) {
299 		tty_warn(1, "A hard link to the directory %s is not allowed",
300 		    arcn->ln_name);
301 		return -1;
302 	}
303 
304 	return mk_link(arcn->ln_name, &sb, arcn->name, 0);
305 }
306 
307 /*
308  * cross_lnk()
309  *	Create a hard link to arcn->org_name from arcn->name. Only used in copy
310  *	with the -l flag. No warning or error if this does not succeed (we will
311  *	then just create the file)
312  * Return:
313  *	1 if copy() should try to create this file node
314  *	0 if cross_lnk() ok, -1 for fatal flaw (like linking to self).
315  */
316 
317 int
cross_lnk(ARCHD * arcn)318 cross_lnk(ARCHD *arcn)
319 {
320 	/*
321 	 * try to make a link to original file (-l flag in copy mode). make
322 	 * sure we do not try to link to directories in case we are running as
323 	 * root (and it might succeed).
324 	 */
325 	if (arcn->type == PAX_DIR)
326 		return 1;
327 	return mk_link(arcn->org_name, &(arcn->sb), arcn->name, 1);
328 }
329 
330 /*
331  * chk_same()
332  *	In copy mode if we are not trying to make hard links between the src
333  *	and destinations, make sure we are not going to overwrite ourselves by
334  *	accident. This slows things down a little, but we have to protect all
335  *	those people who make typing errors.
336  * Return:
337  *	1 the target does not exist, go ahead and copy
338  *	0 skip it file exists (-k) or may be the same as source file
339  */
340 
341 int
chk_same(ARCHD * arcn)342 chk_same(ARCHD *arcn)
343 {
344 	struct stat sb;
345 
346 	/*
347 	 * if file does not exist, return. if file exists and -k, skip it
348 	 * quietly
349 	 */
350 	if (lstat(arcn->name, &sb) < 0)
351 		return 1;
352 	if (kflag)
353 		return 0;
354 
355 	/*
356 	 * better make sure the user does not have src == dest by mistake
357 	 */
358 	if ((arcn->sb.st_dev == sb.st_dev) && (arcn->sb.st_ino == sb.st_ino)) {
359 		tty_warn(1, "Unable to copy %s, file would overwrite itself",
360 		    arcn->name);
361 		return 0;
362 	}
363 	return 1;
364 }
365 
366 /*
367  * mk_link()
368  *	try to make a hard link between two files. if ign set, we do not
369  *	complain.
370  * Return:
371  *	0 if successful (or we are done with this file but no error, such as
372  *	finding the from file exists and the user has set -k).
373  *	1 when ign was set to indicates we could not make the link but we
374  *	should try to copy/extract the file as that might work (and is an
375  *	allowed option). -1 an error occurred.
376  */
377 
378 static int
mk_link(char * to,struct stat * to_sb,char * from,int ign)379 mk_link(char *to, struct stat *to_sb, char *from, int ign)
380 {
381 	struct stat sb;
382 	int oerrno;
383 
384 	/*
385 	 * if from file exists, it has to be unlinked to make the link. If the
386 	 * file exists and -k is set, skip it quietly
387 	 */
388 	if (lstat(from, &sb) == 0) {
389 		if (kflag)
390 			return 0;
391 
392 		/*
393 		 * make sure it is not the same file, protect the user
394 		 */
395 		if ((to_sb->st_dev==sb.st_dev)&&(to_sb->st_ino == sb.st_ino)) {
396 			tty_warn(1, "Cannot link file %s to itself", to);
397 			return -1;
398 		}
399 
400 		/*
401 		 * try to get rid of the file, based on the type
402 		 */
403 		if (S_ISDIR(sb.st_mode) && strcmp(from, ".") != 0) {
404 			if (rmdir(from) < 0) {
405 				syswarn(1, errno, "Cannot remove %s", from);
406 				return -1;
407 			}
408 		} else if (unlink(from) < 0) {
409 			if (!ign) {
410 				syswarn(1, errno, "Cannot remove %s", from);
411 				return -1;
412 			}
413 			return 1;
414 		}
415 	}
416 
417 	/*
418 	 * from file is gone (or did not exist), try to make the hard link.
419 	 * if it fails, check the path and try it again (if chk_path() says to
420 	 * try again)
421 	 */
422 	for (;;) {
423 		if (link(to, from) == 0)
424 			break;
425 		oerrno = errno;
426 		if (chk_path(from, to_sb->st_uid, to_sb->st_gid) == 0)
427 			continue;
428 		if (!ign) {
429 			syswarn(1, oerrno, "Cannot link to %s from %s", to,
430 			    from);
431 			return -1;
432 		}
433 		return 1;
434 	}
435 
436 	/*
437 	 * all right the link was made
438 	 */
439 	return 0;
440 }
441 
442 /*
443  * node_creat()
444  *	create an entry in the file system (other than a file or hard link).
445  *	If successful, sets uid/gid modes and times as required.
446  * Return:
447  *	0 if ok, -1 otherwise
448  */
449 
450 int
node_creat(ARCHD * arcn)451 node_creat(ARCHD *arcn)
452 {
453 	int res;
454 	int ign = 0;
455 	int oerrno;
456 	int pass = 0;
457 	mode_t file_mode;
458 	struct stat sb;
459 	char target[MAXPATHLEN];
460 	char *nm = arcn->name;
461 	int len;
462 
463 	/*
464 	 * create node based on type, if that fails try to unlink the node and
465 	 * try again. finally check the path and try again. As noted in the
466 	 * file and link creation routines, this method seems to exhibit the
467 	 * best performance in general use workloads.
468 	 */
469 	file_mode = arcn->sb.st_mode & FILEBITS(arcn->type == PAX_DIR);
470 
471 	for (;;) {
472 		switch (arcn->type) {
473 		case PAX_DIR:
474 			/*
475 			 * If -h (or -L) was given in tar-mode, follow the
476 			 * potential symlink chain before trying to create the
477 			 * directory.
478 			 */
479 			if (strcmp(NM_TAR, argv0) == 0 && Lflag) {
480 				while (lstat(nm, &sb) == 0 &&
481 				    S_ISLNK(sb.st_mode)) {
482 					len = readlink(nm, target,
483 					    sizeof target - 1);
484 					if (len == -1) {
485 						syswarn(0, errno,
486 						   "cannot follow symlink %s "
487 						   "in chain for %s",
488 						    nm, arcn->name);
489 						res = -1;
490 						goto badlink;
491 					}
492 					target[len] = '\0';
493 					nm = target;
494 				}
495 			}
496 			res = domkdir(nm, file_mode);
497 badlink:
498 			if (ign)
499 				res = 0;
500 			break;
501 		case PAX_CHR:
502 			file_mode |= S_IFCHR;
503 			res = mknod(nm, file_mode, arcn->sb.st_rdev);
504 			break;
505 		case PAX_BLK:
506 			file_mode |= S_IFBLK;
507 			res = mknod(nm, file_mode, arcn->sb.st_rdev);
508 			break;
509 		case PAX_FIF:
510 			res = mkfifo(nm, file_mode);
511 			break;
512 		case PAX_SCK:
513 			/*
514 			 * Skip sockets, operation has no meaning under BSD
515 			 */
516 			tty_warn(0,
517 			    "%s skipped. Sockets cannot be copied or extracted",
518 			    nm);
519 			return (-1);
520 		case PAX_SLK:
521 			res = symlink(arcn->ln_name, nm);
522 			break;
523 		case PAX_CTG:
524 		case PAX_HLK:
525 		case PAX_HRG:
526 		case PAX_REG:
527 		default:
528 			/*
529 			 * we should never get here
530 			 */
531 			tty_warn(0, "%s has an unknown file type, skipping",
532 			    nm);
533 			return (-1);
534 		}
535 
536 		/*
537 		 * if we were able to create the node break out of the loop,
538 		 * otherwise try to unlink the node and try again. if that
539 		 * fails check the full path and try a final time.
540 		 */
541 		if (res == 0)
542 			break;
543 
544 		/*
545 		 * we failed to make the node
546 		 */
547 		oerrno = errno;
548 		switch (pass++) {
549 		case 0:
550 			if ((ign = unlnk_exist(nm, arcn->type)) < 0)
551 				return (-1);
552 			continue;
553 
554 		case 1:
555 			if (nodirs ||
556 			    chk_path(nm, arcn->sb.st_uid,
557 			    arcn->sb.st_gid) < 0) {
558 				syswarn(1, oerrno, "Cannot create %s", nm);
559 				return (-1);
560 			}
561 			continue;
562 		}
563 
564 		/*
565 		 * it must be a file that exists but we can't create or
566 		 * remove, but we must avoid the infinite loop.
567 		 */
568 		break;
569 	}
570 
571 	/*
572 	 * we were able to create the node. set uid/gid, modes and times
573 	 */
574 	if (pids)
575 		res = set_ids(nm, arcn->sb.st_uid, arcn->sb.st_gid);
576 	else
577 		res = 0;
578 
579 	/*
580 	 * IMPORTANT SECURITY NOTE:
581 	 * if not preserving mode or we cannot set uid/gid, then PROHIBIT any
582 	 * set uid/gid bits
583 	 */
584 	if (!pmode || res)
585 		arcn->sb.st_mode &= ~SETBITS(arcn->type == PAX_DIR);
586 	if (pmode)
587 		set_pmode(arcn->name, arcn->sb.st_mode);
588 
589 	if (arcn->type == PAX_DIR && strcmp(NM_CPIO, argv0) != 0) {
590 		/*
591 		 * Dirs must be processed again at end of extract to set times
592 		 * and modes to agree with those stored in the archive. However
593 		 * to allow extract to continue, we may have to also set owner
594 		 * rights. This allows nodes in the archive that are children
595 		 * of this directory to be extracted without failure. Both time
596 		 * and modes will be fixed after the entire archive is read and
597 		 * before pax exits.
598 		 */
599 		if (access(nm, R_OK | W_OK | X_OK) < 0) {
600 			if (lstat(nm, &sb) < 0) {
601 				syswarn(0, errno,"Cannot access %s (stat)",
602 				    arcn->name);
603 				set_pmode(nm,file_mode | S_IRWXU);
604 			} else {
605 				/*
606 				 * We have to add rights to the dir, so we make
607 				 * sure to restore the mode. The mode must be
608 				 * restored AS CREATED and not as stored if
609 				 * pmode is not set.
610 				 */
611 				set_pmode(nm, ((sb.st_mode &
612 				    FILEBITS(arcn->type == PAX_DIR)) |
613 				    S_IRWXU));
614 				if (!pmode)
615 					arcn->sb.st_mode = sb.st_mode;
616 			}
617 
618 			/*
619 			 * we have to force the mode to what was set here,
620 			 * since we changed it from the default as created.
621 			 */
622 			add_dir(nm, arcn->nlen, &(arcn->sb), 1);
623 		} else if (pmode || patime || pmtime)
624 			add_dir(nm, arcn->nlen, &(arcn->sb), 0);
625 	}
626 
627 	if (patime || pmtime)
628 		set_ftime(arcn->name, arcn->sb.st_mtime,
629 		    arcn->sb.st_atime, 0, (arcn->type == PAX_SLK) ? 1 : 0);
630 
631 #if HAVE_STRUCT_STAT_ST_FLAGS
632 	if (pfflags && arcn->type != PAX_SLK)
633 		set_chflags(arcn->name, arcn->sb.st_flags);
634 #endif
635 	return 0;
636 }
637 
638 /*
639  * unlnk_exist()
640  *	Remove node from file system with the specified name. We pass the type
641  *	of the node that is going to replace it. When we try to create a
642  *	directory and find that it already exists, we allow processing to
643  *	continue as proper modes etc will always be set for it later on.
644  * Return:
645  *	0 is ok to proceed, no file with the specified name exists
646  *	-1 we were unable to remove the node, or we should not remove it (-k)
647  *	1 we found a directory and we were going to create a directory.
648  */
649 
650 int
unlnk_exist(char * name,int type)651 unlnk_exist(char *name, int type)
652 {
653 	struct stat sb;
654 
655 	/*
656 	 * the file does not exist, or -k we are done
657 	 */
658 	if (lstat(name, &sb) < 0)
659 		return 0;
660 	if (kflag)
661 		return -1;
662 
663 	if (S_ISDIR(sb.st_mode)) {
664 		/*
665 		 * try to remove a directory, if it fails and we were going to
666 		 * create a directory anyway, tell the caller (return a 1).
667 		 *
668 		 * don't try to remove the directory if the name is "."
669 		 * otherwise later file/directory creation fails.
670 		 */
671 		if (strcmp(name, ".") == 0)
672 			return 1;
673 		if (rmdir(name) < 0) {
674 			if (type == PAX_DIR)
675 				return 1;
676 			syswarn(1, errno, "Cannot remove directory %s", name);
677 			return -1;
678 		}
679 		return 0;
680 	}
681 
682 	/*
683 	 * try to get rid of all non-directory type nodes
684 	 */
685 	if (unlink(name) < 0) {
686 		(void)fflush(listf);
687 		syswarn(1, errno, "Cannot unlink %s", name);
688 		return -1;
689 	}
690 	return 0;
691 }
692 
693 /*
694  * chk_path()
695  *	We were trying to create some kind of node in the file system and it
696  *	failed. chk_path() makes sure the path up to the node exists and is
697  *	writable. When we have to create a directory that is missing along the
698  *	path somewhere, the directory we create will be set to the same
699  *	uid/gid as the file has (when uid and gid are being preserved).
700  *	NOTE: this routine is a real performance loss. It is only used as a
701  *	last resort when trying to create entries in the file system.
702  * Return:
703  *	-1 when it could find nothing it is allowed to fix.
704  *	0 otherwise
705  */
706 
707 int
chk_path(char * name,uid_t st_uid,gid_t st_gid)708 chk_path(char *name, uid_t st_uid, gid_t st_gid)
709 {
710 	char *spt = name;
711 	struct stat sb;
712 	int retval = -1;
713 
714 	/*
715 	 * watch out for paths with nodes stored directly in / (e.g. /bozo)
716 	 */
717 	if (*spt == '/')
718 		++spt;
719 
720 	for(;;) {
721 		/*
722 		 * work forward from the first / and check each part of
723 		 * the path
724 		 */
725 		spt = strchr(spt, '/');
726 		if (spt == NULL)
727 			break;
728 		*spt = '\0';
729 
730 		/*
731 		 * if it exists we assume it is a directory, it is not within
732 		 * the spec (at least it seems to read that way) to alter the
733 		 * file system for nodes NOT EXPLICITLY stored on the archive.
734 		 * If that assumption is changed, you would test the node here
735 		 * and figure out how to get rid of it (probably like some
736 		 * recursive unlink()) or fix up the directory permissions if
737 		 * required (do an access()).
738 		 */
739 		if (lstat(name, &sb) == 0) {
740 			*(spt++) = '/';
741 			continue;
742 		}
743 
744 		/*
745 		 * the path fails at this point, see if we can create the
746 		 * needed directory and continue on
747 		 */
748 		if (domkdir(name, S_IRWXU | S_IRWXG | S_IRWXO) == -1) {
749 			*spt = '/';
750 			retval = -1;
751 			break;
752 		}
753 
754 		/*
755 		 * we were able to create the directory. We will tell the
756 		 * caller that we found something to fix, and it is ok to try
757 		 * and create the node again.
758 		 */
759 		retval = 0;
760 		if (pids)
761 			(void)set_ids(name, st_uid, st_gid);
762 
763 		/*
764 		 * make sure the user doesn't have some strange umask that
765 		 * causes this newly created directory to be unusable. We fix
766 		 * the modes and restore them back to the creation default at
767 		 * the end of pax
768 		 */
769 		if ((access(name, R_OK | W_OK | X_OK) < 0) &&
770 		    (lstat(name, &sb) == 0)) {
771 			set_pmode(name, ((sb.st_mode & FILEBITS(0)) |
772 			    S_IRWXU));
773 			add_dir(name, spt - name, &sb, 1);
774 		}
775 		*(spt++) = '/';
776 		continue;
777 	}
778 	/*
779 	 * We perform one final check here, because if someone else
780 	 * created the directory in parallel with us, we might return
781 	 * the wrong error code, even if the directory exists now.
782 	 */
783 	if (retval == -1 && stat(name, &sb) == 0 && S_ISDIR(sb.st_mode))
784 		retval = 0;
785 	return retval;
786 }
787 
788 /*
789  * set_ftime()
790  *	Set the access time and modification time for a named file. If frc
791  *	is non-zero we force these times to be set even if the user did not
792  *	request access and/or modification time preservation (this is also
793  *	used by -t to reset access times).
794  *	When ign is zero, only those times the user has asked for are set, the
795  *	other ones are left alone. We do not assume the un-documented feature
796  *	of many utimes() implementations that consider a 0 time value as a do
797  *	not set request.
798  *
799  *	Unfortunately, there are systems where lutimes() is present but does
800  *	not work on some filesystem types, which cannot be detected at
801  * 	compile time.  This requires passing down symlink knowledge into
802  *	this function to obtain correct operation.  Linux with XFS is one
803  * 	example of such a system.
804  */
805 
806 void
set_ftime(char * fnm,time_t mtime,time_t atime,int frc,int slk)807 set_ftime(char *fnm, time_t mtime, time_t atime, int frc, int slk)
808 {
809 	struct timeval tv[2];
810 	struct stat sb;
811 
812 	tv[0].tv_sec = atime;
813 	tv[0].tv_usec = 0;
814 	tv[1].tv_sec = mtime;
815 	tv[1].tv_usec = 0;
816 	if (!frc && (!patime || !pmtime)) {
817 		/*
818 		 * if we are not forcing, only set those times the user wants
819 		 * set. We get the current values of the times if we need them.
820 		 */
821 		if (lstat(fnm, &sb) == 0) {
822 #if BSD4_4 && !HAVE_NBTOOL_CONFIG_H
823 			if (!patime)
824 				TIMESPEC_TO_TIMEVAL(&tv[0], &sb.st_atimespec);
825 			if (!pmtime)
826 				TIMESPEC_TO_TIMEVAL(&tv[1], &sb.st_mtimespec);
827 #else
828 			if (!patime)
829 				tv[0].tv_sec = sb.st_atime;
830 			if (!pmtime)
831 				tv[1].tv_sec = sb.st_mtime;
832 #endif
833 		} else
834 			syswarn(0, errno, "Cannot obtain file stats %s", fnm);
835 	}
836 
837 	/*
838 	 * set the times
839 	 */
840 #if HAVE_LUTIMES
841 	if (lutimes(fnm, tv) == 0)
842 		return;
843 	if (errno != ENOSYS)	/* XXX linux: lutimes is per-FS */
844 		goto bad;
845 #endif
846 	if (slk)
847 		return;
848 	if (utimes(fnm, tv) == -1)
849 		goto bad;
850 	return;
851 bad:
852 	syswarn(1, errno, "Access/modification time set failed on: %s", fnm);
853 }
854 
855 /*
856  * set_ids()
857  *	set the uid and gid of a file system node
858  * Return:
859  *	0 when set, -1 on failure
860  */
861 
862 int
set_ids(char * fnm,uid_t uid,gid_t gid)863 set_ids(char *fnm, uid_t uid, gid_t gid)
864 {
865 	if (geteuid() == 0)
866 		if (lchown(fnm, uid, gid)) {
867 			(void)fflush(listf);
868 			syswarn(1, errno, "Cannot set file uid/gid of %s",
869 			    fnm);
870 			return -1;
871 		}
872 	return 0;
873 }
874 
875 /*
876  * set_pmode()
877  *	Set file access mode
878  */
879 
880 void
set_pmode(char * fnm,mode_t mode)881 set_pmode(char *fnm, mode_t mode)
882 {
883 	mode &= A_BITS;
884 	if (lchmod(fnm, mode)) {
885 		(void)fflush(listf);
886 		syswarn(1, errno, "Cannot set permissions on %s", fnm);
887 	}
888 	return;
889 }
890 
891 /*
892  * set_chflags()
893  *	Set 4.4BSD file flags
894  */
895 void
set_chflags(char * fnm,u_int32_t flags)896 set_chflags(char *fnm, u_int32_t flags)
897 {
898 
899 #if 0
900 	if (chflags(fnm, flags) < 0 && errno != EOPNOTSUPP)
901 		syswarn(1, errno, "Cannot set file flags on %s", fnm);
902 #endif
903 	return;
904 }
905 
906 /*
907  * file_write()
908  *	Write/copy a file (during copy or archive extract). This routine knows
909  *	how to copy files with lseek holes in it. (Which are read as file
910  *	blocks containing all 0's but do not have any file blocks associated
911  *	with the data). Typical examples of these are files created by dbm
912  *	variants (.pag files). While the file size of these files are huge, the
913  *	actual storage is quite small (the files are sparse). The problem is
914  *	the holes read as all zeros so are probably stored on the archive that
915  *	way (there is no way to determine if the file block is really a hole,
916  *	we only know that a file block of all zero's can be a hole).
917  *	At this writing, no major archive format knows how to archive files
918  *	with holes. However, on extraction (or during copy, -rw) we have to
919  *	deal with these files. Without detecting the holes, the files can
920  *	consume a lot of file space if just written to disk. This replacement
921  *	for write when passed the basic allocation size of a file system block,
922  *	uses lseek whenever it detects the input data is all 0 within that
923  *	file block. In more detail, the strategy is as follows:
924  *	While the input is all zero keep doing an lseek. Keep track of when we
925  *	pass over file block boundaries. Only write when we hit a non zero
926  *	input. once we have written a file block, we continue to write it to
927  *	the end (we stop looking at the input). When we reach the start of the
928  *	next file block, start checking for zero blocks again. Working on file
929  *	block boundaries significantly reduces the overhead when copying files
930  *	that are NOT very sparse. This overhead (when compared to a write) is
931  *	almost below the measurement resolution on many systems. Without it,
932  *	files with holes cannot be safely copied. It does has a side effect as
933  *	it can put holes into files that did not have them before, but that is
934  *	not a problem since the file contents are unchanged (in fact it saves
935  *	file space). (Except on paging files for diskless clients. But since we
936  *	cannot determine one of those file from here, we ignore them). If this
937  *	ever ends up on a system where CTG files are supported and the holes
938  *	are not desired, just do a conditional test in those routines that
939  *	call file_write() and have it call write() instead. BEFORE CLOSING THE
940  *	FILE, make sure to call file_flush() when the last write finishes with
941  *	an empty block. A lot of file systems will not create an lseek hole at
942  *	the end. In this case we drop a single 0 at the end to force the
943  *	trailing 0's in the file.
944  *	---Parameters---
945  *	rem: how many bytes left in this file system block
946  *	isempt: have we written to the file block yet (is it empty)
947  *	sz: basic file block allocation size
948  *	cnt: number of bytes on this write
949  *	str: buffer to write
950  * Return:
951  *	number of bytes written, -1 on write (or lseek) error.
952  */
953 
954 int
file_write(int fd,char * str,int cnt,int * rem,int * isempt,int sz,char * name)955 file_write(int fd, char *str, int cnt, int *rem, int *isempt, int sz,
956 	char *name)
957 {
958 	char *pt;
959 	char *end;
960 	int wcnt;
961 	char *st = str;
962 	char **strp;
963 	size_t *lenp;
964 
965 	/*
966 	 * while we have data to process
967 	 */
968 	while (cnt) {
969 		if (!*rem) {
970 			/*
971 			 * We are now at the start of file system block again
972 			 * (or what we think one is...). start looking for
973 			 * empty blocks again
974 			 */
975 			*isempt = 1;
976 			*rem = sz;
977 		}
978 
979 		/*
980 		 * only examine up to the end of the current file block or
981 		 * remaining characters to write, whatever is smaller
982 		 */
983 		wcnt = MIN(cnt, *rem);
984 		cnt -= wcnt;
985 		*rem -= wcnt;
986 		if (*isempt) {
987 			/*
988 			 * have not written to this block yet, so we keep
989 			 * looking for zero's
990 			 */
991 			pt = st;
992 			end = st + wcnt;
993 
994 			/*
995 			 * look for a zero filled buffer
996 			 */
997 			while ((pt < end) && (*pt == '\0'))
998 				++pt;
999 
1000 			if (pt == end) {
1001 				/*
1002 				 * skip, buf is empty so far
1003 				 */
1004 				if (fd > -1 &&
1005 				    lseek(fd, (off_t)wcnt, SEEK_CUR) < 0) {
1006 					syswarn(1, errno, "File seek on %s",
1007 					    name);
1008 					return -1;
1009 				}
1010 				st = pt;
1011 				continue;
1012 			}
1013 			/*
1014 			 * drat, the buf is not zero filled
1015 			 */
1016 			*isempt = 0;
1017 		}
1018 
1019 		/*
1020 		 * have non-zero data in this file system block, have to write
1021 		 */
1022 		switch (fd) {
1023 		case -PAX_GLF:
1024 			strp = &gnu_name_string;
1025 			lenp = &gnu_name_length;
1026 			break;
1027 		case -PAX_GLL:
1028 			strp = &gnu_link_string;
1029 			lenp = &gnu_link_length;
1030 			break;
1031 		default:
1032 			strp = NULL;
1033 			lenp = NULL;
1034 			break;
1035 		}
1036 		if (strp) {
1037 			char *nstr = *strp ? realloc(*strp, *lenp + wcnt + 1) :
1038 				malloc(wcnt + 1);
1039 			if (nstr == NULL) {
1040 				tty_warn(1, "Out of memory");
1041 				return -1;
1042 			}
1043 			(void)strlcpy(&nstr[*lenp], st, wcnt + 1);
1044 			*strp = nstr;
1045 			*lenp += wcnt;
1046 		} else if (xwrite(fd, st, wcnt) != wcnt) {
1047 			syswarn(1, errno, "Failed write to file %s", name);
1048 			return -1;
1049 		}
1050 		st += wcnt;
1051 	}
1052 	return st - str;
1053 }
1054 
1055 /*
1056  * file_flush()
1057  *	when the last file block in a file is zero, many file systems will not
1058  *	let us create a hole at the end. To get the last block with zeros, we
1059  *	write the last BYTE with a zero (back up one byte and write a zero).
1060  */
1061 
1062 void
file_flush(int fd,char * fname,int isempt)1063 file_flush(int fd, char *fname, int isempt)
1064 {
1065 	static char blnk[] = "\0";
1066 
1067 	/*
1068 	 * silly test, but make sure we are only called when the last block is
1069 	 * filled with all zeros.
1070 	 */
1071 	if (!isempt)
1072 		return;
1073 
1074 	/*
1075 	 * move back one byte and write a zero
1076 	 */
1077 	if (lseek(fd, (off_t)-1, SEEK_CUR) < 0) {
1078 		syswarn(1, errno, "Failed seek on file %s", fname);
1079 		return;
1080 	}
1081 
1082 	if (write_with_restart(fd, blnk, 1) < 0)
1083 		syswarn(1, errno, "Failed write to file %s", fname);
1084 	return;
1085 }
1086 
1087 /*
1088  * rdfile_close()
1089  *	close a file we have been reading (to copy or archive). If we have to
1090  *	reset access time (tflag) do so (the times are stored in arcn).
1091  */
1092 
1093 void
rdfile_close(ARCHD * arcn,int * fd)1094 rdfile_close(ARCHD *arcn, int *fd)
1095 {
1096 	/*
1097 	 * make sure the file is open
1098 	 */
1099 	if (*fd < 0)
1100 		return;
1101 
1102 	(void)close(*fd);
1103 	*fd = -1;
1104 	if (!tflag)
1105 		return;
1106 
1107 	/*
1108 	 * user wants last access time reset
1109 	 */
1110 	set_ftime(arcn->org_name, arcn->sb.st_mtime, arcn->sb.st_atime, 1, 0);
1111 	return;
1112 }
1113 
1114 /*
1115  * set_crc()
1116  *	read a file to calculate its crc. This is a real drag. Archive formats
1117  *	that have this, end up reading the file twice (we have to write the
1118  *	header WITH the crc before writing the file contents. Oh well...
1119  * Return:
1120  *	0 if was able to calculate the crc, -1 otherwise
1121  */
1122 
1123 int
set_crc(ARCHD * arcn,int fd)1124 set_crc(ARCHD *arcn, int fd)
1125 {
1126 	int i;
1127 	int res;
1128 	off_t cpcnt = 0L;
1129 	u_long size;
1130 	unsigned long crc = 0L;
1131 	char tbuf[FILEBLK];
1132 	struct stat sb;
1133 
1134 	if (fd < 0) {
1135 		/*
1136 		 * hmm, no fd, should never happen. well no crc then.
1137 		 */
1138 		arcn->crc = 0L;
1139 		return 0;
1140 	}
1141 
1142 	if ((size = (u_long)arcn->sb.st_blksize) > (u_long)sizeof(tbuf))
1143 		size = (u_long)sizeof(tbuf);
1144 
1145 	/*
1146 	 * read all the bytes we think that there are in the file. If the user
1147 	 * is trying to archive an active file, forget this file.
1148 	 */
1149 	for(;;) {
1150 		if ((res = read(fd, tbuf, size)) <= 0)
1151 			break;
1152 		cpcnt += res;
1153 		for (i = 0; i < res; ++i)
1154 			crc += (tbuf[i] & 0xff);
1155 	}
1156 
1157 	/*
1158 	 * safety check. we want to avoid archiving files that are active as
1159 	 * they can create inconsistent archive copies.
1160 	 */
1161 	if (cpcnt != arcn->sb.st_size)
1162 		tty_warn(1, "File changed size %s", arcn->org_name);
1163 	else if (fstat(fd, &sb) < 0)
1164 		syswarn(1, errno, "Failed stat on %s", arcn->org_name);
1165 	else if (arcn->sb.st_mtime != sb.st_mtime)
1166 		tty_warn(1, "File %s was modified during read", arcn->org_name);
1167 	else if (lseek(fd, (off_t)0L, SEEK_SET) < 0)
1168 		syswarn(1, errno, "File rewind failed on: %s", arcn->org_name);
1169 	else {
1170 		arcn->crc = crc;
1171 		return 0;
1172 	}
1173 	return -1;
1174 }
1175