1 /*	$OpenBSD: file_subs.c,v 1.32 2009/12/22 12:08:30 jasper Exp $	*/
2 /*	$NetBSD: file_subs.c,v 1.4 1995/03/21 09:07:18 cgd Exp $	*/
3 
4 /*-
5  * Copyright (c) 2007, 2008, 2009, 2012, 2014
6  *	Thorsten Glaser <tg@mirbsd.org>
7  * Copyright (c) 2011
8  *	Svante Signell <svante.signell@telia.com>
9  * Copyright (c) 1992 Keith Muller.
10  * Copyright (c) 1992, 1993
11  *	The Regents of the University of California.  All rights reserved.
12  *
13  * This code is derived from software contributed to Berkeley by
14  * Keith Muller of the University of California, San Diego.
15  *
16  * Redistribution and use in source and binary forms, with or without
17  * modification, are permitted provided that the following conditions
18  * are met:
19  * 1. Redistributions of source code must retain the above copyright
20  *    notice, this list of conditions and the following disclaimer.
21  * 2. Redistributions in binary form must reproduce the above copyright
22  *    notice, this list of conditions and the following disclaimer in the
23  *    documentation and/or other materials provided with the distribution.
24  * 3. Neither the name of the University nor the names of its contributors
25  *    may be used to endorse or promote products derived from this software
26  *    without specific prior written permission.
27  *
28  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
29  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
30  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
31  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
32  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
33  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
34  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
35  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
36  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
37  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
38  * SUCH DAMAGE.
39  */
40 
41 #include <sys/param.h>
42 #include <sys/time.h>
43 #include <sys/stat.h>
44 #include <sys/uio.h>
45 #include <err.h>
46 #include <errno.h>
47 #include <fcntl.h>
48 #include <stdio.h>
49 #include <stdlib.h>
50 #include <string.h>
51 #include <time.h>
52 #include <unistd.h>
53 #ifdef __INTERIX
54 #include <utime.h>
55 #endif
56 #include "pax.h"
57 #include "options.h"
58 #include "extern.h"
59 
60 __RCSID("$MirOS: src/bin/pax/file_subs.c,v 1.19 2014/07/03 17:52:11 tg Exp $");
61 
62 #ifndef __GLIBC_PREREQ
63 #define __GLIBC_PREREQ(maj,min)	0
64 #endif
65 
66 #if !defined(__INTERIX) && (!defined(__GLIBC__) || __GLIBC_PREREQ(2, 3))
67 #define PAX_FUTIMES	/* we have futimes() */
68 #endif
69 
70 static int mk_link(char *, struct stat *, char *, int);
71 #ifdef PAX_FUTIMES
72 static void fset_ftime(char *, int, time_t, time_t, int);
73 #endif
74 
75 /*
76  * routines that deal with file operations such as: creating, removing;
77  * and setting access modes, uid/gid and times of files
78  */
79 
80 #define FILEBITS		(S_ISVTX | S_IRWXU | S_IRWXG | S_IRWXO)
81 #define SETBITS			(S_ISUID | S_ISGID)
82 #define ABITS			(FILEBITS | SETBITS)
83 
84 /*
85  * file_creat()
86  *	Create and open a file.
87  * Return:
88  *	file descriptor or -1 for failure
89  */
90 
91 int
file_creat(ARCHD * arcn)92 file_creat(ARCHD *arcn)
93 {
94 	int fd = -1;
95 	mode_t file_mode;
96 	int oerrno;
97 
98 	/*
99 	 * Assume file doesn't exist, so just try to create it, most times this
100 	 * works. We have to take special handling when the file does exist. To
101 	 * detect this, we use O_EXCL. For example when trying to create a
102 	 * file and a character device or fifo exists with the same name, we
103 	 * can accidently open the device by mistake (or block waiting to open).
104 	 * If we find that the open has failed, then spend the effort to
105 	 * figure out why. This strategy was found to have better average
106 	 * performance in common use than checking the file (and the path)
107 	 * first with lstat.
108 	 */
109 	file_mode = arcn->sb.st_mode & FILEBITS;
110 	if ((fd = open(arcn->name, O_WRONLY | O_CREAT | O_TRUNC | O_EXCL,
111 	    file_mode)) >= 0)
112 		return(fd);
113 
114 	/*
115 	 * the file seems to exist. First we try to get rid of it (found to be
116 	 * the second most common failure when traced). If this fails, only
117 	 * then we go to the expense to check and create the path to the file
118 	 */
119 	if (unlnk_exist(arcn->name, arcn->type) != 0)
120 		return(-1);
121 
122 	for (;;) {
123 		/*
124 		 * try to open it again, if this fails, check all the nodes in
125 		 * the path and give it a final try. if chk_path() finds that
126 		 * it cannot fix anything, we will skip the last attempt
127 		 */
128 		if ((fd = open(arcn->name, O_WRONLY | O_CREAT | O_TRUNC,
129 		    file_mode)) >= 0)
130 			break;
131 		oerrno = errno;
132 		if (nodirs || chk_path(arcn->name,arcn->sb.st_uid,arcn->sb.st_gid) < 0) {
133 			syswarn(1, oerrno, "Unable to create %s", arcn->name);
134 			return(-1);
135 		}
136 	}
137 	return(fd);
138 }
139 
140 /*
141  * file_close()
142  *	Close file descriptor to a file just created by pax. Sets modes,
143  *	ownership and times as required.
144  * Return:
145  *	0 for success, -1 for failure
146  */
147 
148 void
file_close(ARCHD * arcn,int fd)149 file_close(ARCHD *arcn, int fd)
150 {
151 	int res = 0;
152 
153 	if (fd < 0)
154 		return;
155 
156 	/*
157 	 * set owner/groups first as this may strip off mode bits we want
158 	 * then set file permission modes. Then set file access and
159 	 * modification times.
160 	 */
161 	if (pids)
162 		res = fset_ids(arcn->name, fd, arcn->sb.st_uid,
163 		    arcn->sb.st_gid);
164 
165 	/*
166 	 * IMPORTANT SECURITY NOTE:
167 	 * if not preserving mode or we cannot set uid/gid, then PROHIBIT
168 	 * set uid/gid bits
169 	 */
170 	if (!pmode || res)
171 		arcn->sb.st_mode &= ~(SETBITS);
172 	if (pmode)
173 		fset_pmode(arcn->name, fd, arcn->sb.st_mode);
174 #ifdef PAX_FUTIMES
175 	if (patime || pmtime)
176 		fset_ftime(arcn->name, fd, arcn->sb.st_mtime,
177 		    arcn->sb.st_atime, 0);
178 #endif
179 	if (close(fd) < 0)
180 		syswarn(0, errno, "Unable to close file descriptor on %s",
181 		    arcn->name);
182 }
183 
184 /*
185  * lnk_creat()
186  *	Create a hard link to arcn->ln_name from arcn->name. arcn->ln_name
187  *	must exist;
188  * Return:
189  *	fd+2 if data should be extracted,
190  *	0 if ok, 1 if we could not make the link, -1 otherwise
191  */
192 
193 int
lnk_creat(ARCHD * arcn,int * fdp)194 lnk_creat(ARCHD *arcn, int *fdp)
195 {
196 	struct stat sb;
197 	int rv;
198 
199 	/*
200 	 * we may be running as root, so we have to be sure that link target
201 	 * is not a directory, so we lstat and check
202 	 */
203 	if (lstat(arcn->ln_name, &sb) < 0) {
204 		syswarn(1,errno,"Unable to link to %s from %s", arcn->ln_name,
205 		    arcn->name);
206 		return(-1);
207 	}
208 
209 	if (S_ISDIR(sb.st_mode)) {
210 		paxwarn(1, "A hard link to the directory %s is not allowed",
211 		    arcn->ln_name);
212 		return(-1);
213 	}
214 
215 	rv = mk_link(arcn->ln_name, &sb, arcn->name, 0);
216 	if (fdp != NULL && rv == 0 && sb.st_size == 0 && arcn->skip > 0) {
217 		/* request to write out file data late (broken archive) */
218 		if (pmode)
219 			set_pmode(arcn->name, 0600);
220 		if ((*fdp = open(arcn->name, O_WRONLY | O_TRUNC)) == -1) {
221 			rv = errno;
222 			syswarn(1, rv, "Unable to re-open %s", arcn->name);
223 			if (pmode)
224 				set_pmode(arcn->name, sb.st_mode);
225 		}
226 		rv = 0;
227 	} else if (fdp != NULL)
228 		*fdp = -1;
229 	return (rv);
230 }
231 
232 /*
233  * cross_lnk()
234  *	Create a hard link to arcn->org_name from arcn->name. Only used in copy
235  *	with the -l flag. No warning or error if this does not succeed (we will
236  *	then just create the file)
237  * Return:
238  *	1 if copy() should try to create this file node
239  *	0 if cross_lnk() ok, -1 for fatal flaw (like linking to self).
240  */
241 
242 int
cross_lnk(ARCHD * arcn)243 cross_lnk(ARCHD *arcn)
244 {
245 	/*
246 	 * try to make a link to original file (-l flag in copy mode). make
247 	 * sure we do not try to link to directories in case we are running as
248 	 * root (and it might succeed).
249 	 */
250 	if (arcn->type == PAX_DIR)
251 		return(1);
252 	return(mk_link(arcn->org_name, &(arcn->sb), arcn->name, 1));
253 }
254 
255 /*
256  * chk_same()
257  *	In copy mode if we are not trying to make hard links between the src
258  *	and destinations, make sure we are not going to overwrite ourselves by
259  *	accident. This slows things down a little, but we have to protect all
260  *	those people who make typing errors.
261  * Return:
262  *	1 the target does not exist, go ahead and copy
263  *	0 skip it file exists (-k) or may be the same as source file
264  */
265 
266 int
chk_same(ARCHD * arcn)267 chk_same(ARCHD *arcn)
268 {
269 	struct stat sb;
270 
271 	/*
272 	 * if file does not exist, return. if file exists and -k, skip it
273 	 * quietly
274 	 */
275 	if (lstat(arcn->name, &sb) < 0)
276 		return(1);
277 	if (kflag)
278 		return(0);
279 
280 	/*
281 	 * better make sure the user does not have src == dest by mistake
282 	 */
283 	if ((arcn->sb.st_dev == sb.st_dev) && (arcn->sb.st_ino == sb.st_ino)) {
284 		paxwarn(1, "Unable to copy %s, file would overwrite itself",
285 		    arcn->name);
286 		return(0);
287 	}
288 	return(1);
289 }
290 
291 /*
292  * helper function to copy a symbolic link
293  */
294 
295 static int
mk_link_symlink(const char * to,const char * from)296 mk_link_symlink(const char *to, const char *from)
297 {
298 	int cnt;
299 	char buf[PAXPATHLEN + 1];
300 
301 	if ((cnt = readlink(to, buf, PAXPATHLEN)) < 0)
302 		return (-1);
303 	/* cf. comment in ftree.c:next_file() */
304 	buf[cnt] = '\0';
305 	return (symlink(buf, from));
306 }
307 
308 /*
309  * mk_link()
310  *	try to make a hard link between two files. if ign set, we do not
311  *	complain.
312  * Return:
313  *	0 if successful (or we are done with this file but no error, such as
314  *	finding the from file exists and the user has set -k).
315  *	1 when ign was set to indicates we could not make the link but we
316  *	should try to copy/extract the file as that might work (and is an
317  *	allowed option). -1 an error occurred.
318  */
319 
320 static int
mk_link(char * to,struct stat * to_sb,char * from,int ign)321 mk_link(char *to, struct stat *to_sb, char *from, int ign)
322 {
323 	struct stat sb;
324 	int oerrno;
325 
326 	/*
327 	 * if from file exists, it has to be unlinked to make the link. If the
328 	 * file exists and -k is set, skip it quietly
329 	 */
330 	if (lstat(from, &sb) == 0) {
331 		if (kflag)
332 			return(0);
333 
334 		/*
335 		 * make sure it is not the same file, protect the user
336 		 */
337 		if ((to_sb->st_dev==sb.st_dev)&&(to_sb->st_ino == sb.st_ino)) {
338 			paxwarn(1, "Unable to link file %s to itself", to);
339 			return(-1);
340 		}
341 
342 		/*
343 		 * try to get rid of the file, based on the type
344 		 */
345 		if (S_ISDIR(sb.st_mode)) {
346 			if (rmdir(from) < 0) {
347 				syswarn(1, errno, "Unable to remove %s", from);
348 				return(-1);
349 			}
350 		} else if (unlink(from) < 0) {
351 			if (!ign) {
352 				syswarn(1, errno, "Unable to remove %s", from);
353 				return(-1);
354 			}
355 			return(1);
356 		}
357 	}
358 
359 	/*
360 	 * from file is gone (or did not exist), try to make the hard link.
361 	 * if it fails, check the path and try it again (if chk_path() says to
362 	 * try again)
363 	 */
364 	for (;;) {
365 		if (link(to, from) == 0)
366 			break;
367 		oerrno = errno;
368 		if (S_ISLNK(to_sb->st_mode)) {
369 			/* just copy the symlink */
370 			if (mk_link_symlink(to, from) == 0)
371 				break;
372 		}
373 		if (!nodirs && chk_path(from, to_sb->st_uid, to_sb->st_gid) == 0)
374 			continue;
375 		/*-
376 		 * non-standard (via -M lncp) cross-device link handling:
377 		 * copy if hard link fails (but what if there are several
378 		 * links for the same file mixed between several devices?
379 		 * this code copies for all non-original devices, instead
380 		 * of tracking them and linking between them on their re-
381 		 * spective target device)
382 		 */
383 		if (oerrno == EXDEV && (anonarch & ANON_LNCP)) {
384 			int fdsrc, fddest;
385 			ARCHD tarcn;
386 
387 			if ((fdsrc = open(to, O_RDONLY, 0)) < 0) {
388 				if (!ign)
389 					syswarn(1, errno,
390 					    "Unable to open %s to read", to);
391 				goto lncp_failed;
392 			}
393 			strlcpy(tarcn.name, from, sizeof(tarcn.name));
394 			memcpy(&tarcn.sb, to_sb, sizeof(struct stat));
395 			tarcn.type = PAX_REG;	/* XXX */
396 			tarcn.org_name = to;
397 			if ((fddest = file_creat(&tarcn)) < 0) {
398 				rdfile_close(&tarcn, &fdsrc);
399 				goto lncp_failed;
400 			}
401 			cp_file(&tarcn, fdsrc, fddest);
402 			file_close(&tarcn, fddest);
403 			rdfile_close(&tarcn, &fdsrc);
404 			/* file copied successfully, continue on */
405 			break;
406 		}
407 		if (!ign) {
408  lncp_failed:
409 			syswarn(1, oerrno, "Could not link to %s from %s", to,
410 			    from);
411 			return(-1);
412 		}
413 		return(1);
414 	}
415 
416 	/*
417 	 * all right the link was made
418 	 */
419 	return(0);
420 }
421 
422 /*
423  * node_creat()
424  *	create an entry in the file system (other than a file or hard link).
425  *	If successful, sets uid/gid modes and times as required.
426  * Return:
427  *	0 if ok, -1 otherwise
428  */
429 
430 int
node_creat(ARCHD * arcn)431 node_creat(ARCHD *arcn)
432 {
433 	int res;
434 	int ign = 0;
435 	int oerrno;
436 	int pass = 0;
437 	mode_t file_mode;
438 	struct stat sb;
439 	char *target = NULL;
440 	char *nm = arcn->name;
441 	int len;
442 
443 	/*
444 	 * create node based on type, if that fails try to unlink the node and
445 	 * try again. finally check the path and try again. As noted in the
446 	 * file and link creation routines, this method seems to exhibit the
447 	 * best performance in general use workloads.
448 	 */
449 	file_mode = arcn->sb.st_mode & FILEBITS;
450 
451 	for (;;) {
452 		switch (arcn->type) {
453 		case PAX_DIR:
454 			/*
455 			 * If -h (or -L) was given in tar-mode, follow the
456 			 * potential symlink chain before trying to create the
457 			 * directory.
458 			 */
459 			if (strcmp(NM_TAR, argv0) == 0 && Lflag) {
460 				while (lstat(nm, &sb) == 0 &&
461 				    S_ISLNK(sb.st_mode)) {
462 					target = malloc(sb.st_size + 1);
463 					if (target == NULL) {
464 						oerrno = ENOMEM;
465 						syswarn(1, oerrno,
466 						    "Out of memory");
467 						return (-1);
468 					}
469 					len = readlink(nm, target,
470 					    sb.st_size + 1);
471 					if (len == -1) {
472 						syswarn(0, errno,
473 						   "cannot follow symlink %s in chain for %s",
474 						    nm, arcn->name);
475 						res = -1;
476 						goto badlink;
477 					}
478 					target[len] = '\0';
479 					nm = target;
480 				}
481 			}
482 			res = mkdir(nm, file_mode);
483 
484  badlink:
485 			if (ign)
486 				res = 0;
487 			break;
488 		case PAX_CHR:
489 			file_mode |= S_IFCHR;
490 			res = mknod(nm, file_mode, arcn->sb.st_rdev);
491 			break;
492 		case PAX_BLK:
493 			file_mode |= S_IFBLK;
494 			res = mknod(nm, file_mode, arcn->sb.st_rdev);
495 			break;
496 		case PAX_FIF:
497 			res = mkfifo(nm, file_mode);
498 			break;
499 		case PAX_SCK:
500 			/*
501 			 * Skip sockets, operation has no meaning under BSD
502 			 */
503 			paxwarn(0,
504 			    "%s skipped. Sockets cannot be copied or extracted",
505 			    nm);
506 			free(target);
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 			paxwarn(0, "%s has an unknown file type, skipping",
520 				nm);
521 			free(target);
522 			return (-1);
523 		}
524 
525 		/*
526 		 * if we were able to create the node break out of the loop,
527 		 * otherwise try to unlink the node and try again. if that
528 		 * fails check the full path and try a final time.
529 		 */
530 		if (res == 0)
531 			break;
532 
533 		/*
534 		 * we failed to make the node
535 		 */
536 		oerrno = errno;
537 		if ((ign = unlnk_exist(nm, arcn->type)) < 0) {
538 			free(target);
539 			return (-1);
540 		}
541 
542 		if (++pass <= 1)
543 			continue;
544 
545 		if (nodirs || chk_path(nm,arcn->sb.st_uid,arcn->sb.st_gid) < 0) {
546 			syswarn(1, oerrno, "Could not create: %s", nm);
547 			free(target);
548 			return (-1);
549 		}
550 	}
551 	free(target);
552 
553 	/*
554 	 * we were able to create the node. set uid/gid, modes and times
555 	 */
556 	if (pids)
557 		res = ((arcn->type == PAX_SLK) ?
558 		    set_lids(nm, arcn->sb.st_uid, arcn->sb.st_gid) :
559 		    set_ids(nm, arcn->sb.st_uid, arcn->sb.st_gid));
560 	else
561 		res = 0;
562 
563 	/*
564 	 * symlinks are done now.
565 	 */
566 	if (arcn->type == PAX_SLK)
567 		return (0);
568 
569 	/*
570 	 * IMPORTANT SECURITY NOTE:
571 	 * if not preserving mode or we cannot set uid/gid, then PROHIBIT any
572 	 * set uid/gid bits
573 	 */
574 	if (!pmode || res)
575 		arcn->sb.st_mode &= ~(SETBITS);
576 	if (pmode)
577 		set_pmode(nm, arcn->sb.st_mode);
578 
579 	if (arcn->type == PAX_DIR && strcmp(NM_CPIO, argv0) != 0) {
580 		/*
581 		 * Dirs must be processed again at end of extract to set times
582 		 * and modes to agree with those stored in the archive. However
583 		 * to allow extract to continue, we may have to also set owner
584 		 * rights. This allows nodes in the archive that are children
585 		 * of this directory to be extracted without failure. Both time
586 		 * and modes will be fixed after the entire archive is read and
587 		 * before pax exits.
588 		 */
589 		if (access(nm, R_OK | W_OK | X_OK) < 0) {
590 			if (lstat(nm, &sb) < 0) {
591 				syswarn(0, errno,"Could not access %s (stat)",
592 				    arcn->name);
593 				set_pmode(nm,file_mode | S_IRWXU);
594 			} else {
595 				/*
596 				 * We have to add rights to the dir, so we make
597 				 * sure to restore the mode. The mode must be
598 				 * restored AS CREATED and not as stored if
599 				 * pmode is not set.
600 				 */
601 				set_pmode(nm,
602 				    ((sb.st_mode & FILEBITS) | S_IRWXU));
603 				if (!pmode)
604 					arcn->sb.st_mode = sb.st_mode;
605 			}
606 
607 			/*
608 			 * we have to force the mode to what was set here,
609 			 * since we changed it from the default as created.
610 			 */
611 			add_dir(nm, &(arcn->sb), 1);
612 		} else if (pmode || patime || pmtime)
613 			add_dir(nm, &(arcn->sb), 0);
614 	}
615 
616 	if (patime || pmtime)
617 		set_ftime(nm, arcn->sb.st_mtime, arcn->sb.st_atime, 0);
618 	return (0);
619 }
620 
621 /*
622  * unlnk_exist()
623  *	Remove node from file system with the specified name. We pass the type
624  *	of the node that is going to replace it. When we try to create a
625  *	directory and find that it already exists, we allow processing to
626  *	continue as proper modes etc will always be set for it later on.
627  * Return:
628  *	0 is ok to proceed, no file with the specified name exists
629  *	-1 we were unable to remove the node, or we should not remove it (-k)
630  *	1 we found a directory and we were going to create a directory.
631  */
632 
633 int
unlnk_exist(char * name,int type)634 unlnk_exist(char *name, int type)
635 {
636 	struct stat sb;
637 
638 	/*
639 	 * the file does not exist, or -k we are done
640 	 */
641 	if (lstat(name, &sb) < 0)
642 		return(0);
643 	if (kflag)
644 		return(-1);
645 
646 	if (S_ISDIR(sb.st_mode)) {
647 		/*
648 		 * try to remove a directory, if it fails and we were going to
649 		 * create a directory anyway, tell the caller (return a 1)
650 		 */
651 		if (rmdir(name) < 0) {
652 			if (type == PAX_DIR)
653 				return(1);
654 			syswarn(1,errno,"Unable to remove directory %s", name);
655 			return(-1);
656 		}
657 		return(0);
658 	}
659 
660 	/*
661 	 * try to get rid of all non-directory type nodes
662 	 */
663 	if (unlink(name) < 0) {
664 		syswarn(1, errno, "Could not unlink %s", name);
665 		return(-1);
666 	}
667 	return(0);
668 }
669 
670 /*
671  * chk_path()
672  *	We were trying to create some kind of node in the file system and it
673  *	failed. chk_path() makes sure the path up to the node exists and is
674  *	writeable. When we have to create a directory that is missing along the
675  *	path somewhere, the directory we create will be set to the same
676  *	uid/gid as the file has (when uid and gid are being preserved).
677  *	NOTE: this routine is a real performance loss. It is only used as a
678  *	last resort when trying to create entries in the file system.
679  * Return:
680  *	-1 when it could find nothing it is allowed to fix.
681  *	0 otherwise
682  */
683 
684 int
chk_path(char * name,uid_t st_uid,gid_t st_gid)685 chk_path(char *name, uid_t st_uid, gid_t st_gid)
686 {
687 	char *spt = name;
688 	struct stat sb;
689 	int retval = -1;
690 
691 	/*
692 	 * watch out for paths with nodes stored directly in / (e.g. /bozo)
693 	 */
694 	if (*spt == '/')
695 		++spt;
696 
697 	for (;;) {
698 		/*
699 		 * work forward from the first / and check each part of the path
700 		 */
701 		spt = strchr(spt, '/');
702 		if (spt == NULL)
703 			break;
704 		*spt = '\0';
705 
706 		/*
707 		 * if it exists we assume it is a directory, it is not within
708 		 * the spec (at least it seems to read that way) to alter the
709 		 * file system for nodes NOT EXPLICITLY stored on the archive.
710 		 * If that assumption is changed, you would test the node here
711 		 * and figure out how to get rid of it (probably like some
712 		 * recursive unlink()) or fix up the directory permissions if
713 		 * required (do an access()).
714 		 */
715 		if (lstat(name, &sb) == 0) {
716 			*(spt++) = '/';
717 			continue;
718 		}
719 
720 		/*
721 		 * the path fails at this point, see if we can create the
722 		 * needed directory and continue on
723 		 */
724 		if (mkdir(name, S_IRWXU | S_IRWXG | S_IRWXO) < 0) {
725 			*spt = '/';
726 			retval = -1;
727 			break;
728 		}
729 
730 		/*
731 		 * we were able to create the directory. We will tell the
732 		 * caller that we found something to fix, and it is ok to try
733 		 * and create the node again.
734 		 */
735 		retval = 0;
736 		if (pids)
737 			(void)set_ids(name, st_uid, st_gid);
738 
739 		/*
740 		 * make sure the user doesn't have some strange umask that
741 		 * causes this newly created directory to be unusable. We fix
742 		 * the modes and restore them back to the creation default at
743 		 * the end of pax
744 		 */
745 		if ((access(name, R_OK | W_OK | X_OK) < 0) &&
746 		    (lstat(name, &sb) == 0)) {
747 			set_pmode(name, ((sb.st_mode & FILEBITS) | S_IRWXU));
748 			add_dir(name, &sb, 1);
749 		}
750 		*(spt++) = '/';
751 		continue;
752 	}
753 	return(retval);
754 }
755 
756 /*
757  * set_ftime()
758  *	Set the access time and modification time for a named file. If frc
759  *	is non-zero we force these times to be set even if the user did not
760  *	request access and/or modification time preservation (this is also
761  *	used by -t to reset access times).
762  *	When ign is zero, only those times the user has asked for are set, the
763  *	other ones are left alone. We do not assume the un-documented feature
764  *	of many utimes() implementations that consider a 0 time value as a do
765  *	not set request.
766  */
767 
768 void
set_ftime(char * fnm,time_t mtime,time_t atime,int frc)769 set_ftime(char *fnm, time_t mtime, time_t atime, int frc)
770 {
771 	static struct timeval tv[2] = {{0L, 0L}, {0L, 0L}};
772 	struct stat sb;
773 #ifdef __INTERIX
774 	struct utimbuf u;
775 #endif
776 
777 	tv[0].tv_sec = (long)atime;
778 	tv[1].tv_sec = (long)mtime;
779 	if (!frc && (!patime || !pmtime)) {
780 		/*
781 		 * if we are not forcing, only set those times the user wants
782 		 * set. We get the current values of the times if we need them.
783 		 */
784 		if (lstat(fnm, &sb) == 0) {
785 			if (!patime)
786 				tv[0].tv_sec = (long)sb.st_atime;
787 			if (!pmtime)
788 				tv[1].tv_sec = (long)sb.st_mtime;
789 		} else
790 			syswarn(0,errno,"Unable to obtain file stats %s", fnm);
791 	}
792 
793 	/*
794 	 * set the times
795 	 */
796 #ifdef __INTERIX
797 	u.actime = tv[0].tv_sec;
798 	u.modtime = tv[1].tv_sec;
799 	if (utime(fnm, &u) < 0)
800 #else
801 	if (utimes(fnm, tv) < 0)
802 #endif
803 		syswarn(1, errno, "Access/modification time set failed on: %s",
804 		    fnm);
805 	return;
806 }
807 
808 #ifdef PAX_FUTIMES
809 static void
fset_ftime(char * fnm,int fd,time_t mtime,time_t atime,int frc)810 fset_ftime(char *fnm, int fd, time_t mtime, time_t atime, int frc)
811 {
812 	static struct timeval tv[2] = {{0L, 0L}, {0L, 0L}};
813 	struct stat sb;
814 
815 	tv[0].tv_sec = (long)atime;
816 	tv[1].tv_sec = (long)mtime;
817 	if (!frc && (!patime || !pmtime)) {
818 		/*
819 		 * if we are not forcing, only set those times the user wants
820 		 * set. We get the current values of the times if we need them.
821 		 */
822 		if (fstat(fd, &sb) == 0) {
823 			if (!patime)
824 				tv[0].tv_sec = (long)sb.st_atime;
825 			if (!pmtime)
826 				tv[1].tv_sec = (long)sb.st_mtime;
827 		} else
828 			syswarn(0,errno,"Unable to obtain file stats %s", fnm);
829 	}
830 	/*
831 	 * set the times
832 	 */
833 	if (futimes(fd, tv) < 0)
834 		syswarn(1, errno, "Access/modification time set failed on: %s",
835 		    fnm);
836 	return;
837 }
838 #endif
839 
840 /*
841  * set_ids()
842  *	set the uid and gid of a file system node
843  * Return:
844  *	0 when set, -1 on failure
845  */
846 
847 int
set_ids(char * fnm,uid_t uid,gid_t gid)848 set_ids(char *fnm, uid_t uid, gid_t gid)
849 {
850 	if (chown(fnm, uid, gid) < 0) {
851 		/*
852 		 * ignore EPERM unless in verbose mode or being run by root.
853 		 * if running as pax, POSIX requires a warning.
854 		 */
855 		if (strcmp(NM_PAX, argv0) == 0
856 #ifndef __INTERIX
857 		    || errno != EPERM || vflag ||
858 		    geteuid() == 0
859 #endif
860 		    )
861 			syswarn(1, errno, "Unable to set file uid/gid of %s",
862 			    fnm);
863 		return(-1);
864 	}
865 	return(0);
866 }
867 
868 int
fset_ids(char * fnm,int fd,uid_t uid,gid_t gid)869 fset_ids(char *fnm, int fd, uid_t uid, gid_t gid)
870 {
871 	if (fchown(fd, uid, gid) < 0) {
872 		/*
873 		 * ignore EPERM unless in verbose mode or being run by root.
874 		 * if running as pax, POSIX requires a warning.
875 		 */
876 		if (strcmp(NM_PAX, argv0) == 0 || errno != EPERM || vflag ||
877 		    geteuid() == 0)
878 			syswarn(1, errno, "Unable to set file uid/gid of %s",
879 			    fnm);
880 		return(-1);
881 	}
882 	return(0);
883 }
884 
885 /*
886  * set_lids()
887  *	set the uid and gid of a file system node
888  * Return:
889  *	0 when set, -1 on failure
890  */
891 
892 int
set_lids(char * fnm,uid_t uid,gid_t gid)893 set_lids(char *fnm, uid_t uid, gid_t gid)
894 {
895 #ifndef __APPLE__
896 	if (lchown(fnm, uid, gid) < 0) {
897 		/*
898 		 * ignore EPERM unless in verbose mode or being run by root.
899 		 * if running as pax, POSIX requires a warning.
900 		 */
901 		if (strcmp(NM_PAX, argv0) == 0
902 #ifndef __INTERIX
903 		    || errno != EPERM || vflag ||
904 		    geteuid() == 0
905 #endif
906 		    )
907 			syswarn(1, errno, "Unable to set file uid/gid of %s",
908 			    fnm);
909 		return(-1);
910 	}
911 #endif
912 	return(0);
913 }
914 
915 /*
916  * set_pmode()
917  *	Set file access mode
918  */
919 
920 void
set_pmode(char * fnm,mode_t mode)921 set_pmode(char *fnm, mode_t mode)
922 {
923 	mode &= ABITS;
924 	if (chmod(fnm, mode) < 0)
925 		syswarn(1, errno, "Could not set permissions on %s", fnm);
926 	return;
927 }
928 
929 void
fset_pmode(char * fnm,int fd,mode_t mode)930 fset_pmode(char *fnm, int fd, mode_t mode)
931 {
932 	mode &= ABITS;
933 	if (fchmod(fd, mode) < 0)
934 		syswarn(1, errno, "Could not set permissions on %s", fnm);
935 	return;
936 }
937 
938 /*
939  * file_write()
940  *	Write/copy a file (during copy or archive extract). This routine knows
941  *	how to copy files with lseek holes in it. (Which are read as file
942  *	blocks containing all 0's but do not have any file blocks associated
943  *	with the data). Typical examples of these are files created by dbm
944  *	variants (.pag files). While the file size of these files are huge, the
945  *	actual storage is quite small (the files are sparse). The problem is
946  *	the holes read as all zeros so are probably stored on the archive that
947  *	way (there is no way to determine if the file block is really a hole,
948  *	we only know that a file block of all zero's can be a hole).
949  *	At this writing, no major archive format knows how to archive files
950  *	with holes. However, on extraction (or during copy, -rw) we have to
951  *	deal with these files. Without detecting the holes, the files can
952  *	consume a lot of file space if just written to disk. This replacement
953  *	for write when passed the basic allocation size of a file system block,
954  *	uses lseek whenever it detects the input data is all 0 within that
955  *	file block. In more detail, the strategy is as follows:
956  *	While the input is all zero keep doing an lseek. Keep track of when we
957  *	pass over file block boundaries. Only write when we hit a non zero
958  *	input. once we have written a file block, we continue to write it to
959  *	the end (we stop looking at the input). When we reach the start of the
960  *	next file block, start checking for zero blocks again. Working on file
961  *	block boundaries significantly reduces the overhead when copying files
962  *	that are NOT very sparse. This overhead (when compared to a write) is
963  *	almost below the measurement resolution on many systems. Without it,
964  *	files with holes cannot be safely copied. It does has a side effect as
965  *	it can put holes into files that did not have them before, but that is
966  *	not a problem since the file contents are unchanged (in fact it saves
967  *	file space). (Except on paging files for diskless clients. But since we
968  *	cannot determine one of those file from here, we ignore them). If this
969  *	ever ends up on a system where CTG files are supported and the holes
970  *	are not desired, just do a conditional test in those routines that
971  *	call file_write() and have it call write() instead. BEFORE CLOSING THE
972  *	FILE, make sure to call file_flush() when the last write finishes with
973  *	an empty block. A lot of file systems will not create an lseek hole at
974  *	the end. In this case we drop a single 0 at the end to force the
975  *	trailing 0's in the file.
976  *	---Parameters---
977  *	rem: how many bytes left in this file system block
978  *	isempt: have we written to the file block yet (is it empty)
979  *	sz: basic file block allocation size
980  *	cnt: number of bytes on this write
981  *	str: buffer to write
982  * Return:
983  *	number of bytes written, -1 on write (or lseek) error.
984  */
985 
986 int
file_write(int fd,char * str,int cnt,int * rem,int * isempt,int sz,char * name)987 file_write(int fd, char *str, int cnt, int *rem, int *isempt, int sz,
988 	char *name)
989 {
990 	char *pt;
991 	char *end;
992 	int wcnt;
993 	char *st = str;
994 	char **strp;
995 
996 	/*
997 	 * while we have data to process
998 	 */
999 	while (cnt) {
1000 		if (!*rem) {
1001 			/*
1002 			 * We are now at the start of file system block again
1003 			 * (or what we think one is...). start looking for
1004 			 * empty blocks again
1005 			 */
1006 			*isempt = 1;
1007 			*rem = sz;
1008 		}
1009 
1010 		/*
1011 		 * only examine up to the end of the current file block or
1012 		 * remaining characters to write, whatever is smaller
1013 		 */
1014 		wcnt = MIN(cnt, *rem);
1015 		cnt -= wcnt;
1016 		*rem -= wcnt;
1017 		if (*isempt) {
1018 			/*
1019 			 * have not written to this block yet, so we keep
1020 			 * looking for zero's
1021 			 */
1022 			pt = st;
1023 			end = st + wcnt;
1024 
1025 			/*
1026 			 * look for a zero filled buffer
1027 			 */
1028 			while ((pt < end) && (*pt == '\0'))
1029 				++pt;
1030 
1031 			if (pt == end) {
1032 				/*
1033 				 * skip, buf is empty so far
1034 				 */
1035 				if (fd > -1 &&
1036 				    lseek(fd, (off_t)wcnt, SEEK_CUR) < 0) {
1037 					if (errno == ESPIPE)
1038 						goto isapipe;
1039 					syswarn(1,errno,"File seek on %s",
1040 					    name);
1041 					return(-1);
1042 				}
1043 				st = pt;
1044 				continue;
1045 			}
1046  isapipe:
1047 			/*
1048 			 * drat, the buf is not zero filled
1049 			 */
1050 			*isempt = 0;
1051 		}
1052 
1053 		/*
1054 		 * have non-zero data in this file system block, have to write
1055 		 */
1056 		switch (fd) {
1057 		case -1:
1058 			strp = &gnu_name_string;
1059 			break;
1060 		case -2:
1061 			strp = &gnu_link_string;
1062 			break;
1063 		default:
1064 			strp = NULL;
1065 			break;
1066 		}
1067 		if (strp) {
1068 			if (*strp)
1069 				err(1, "WARNING! Major Internal Error! GNU hack Failing!");
1070 			*strp = malloc(wcnt + 1);
1071 			if (*strp == NULL) {
1072 				paxwarn(1, "Out of memory");
1073 				return(-1);
1074 			}
1075 			memcpy(*strp, st, wcnt);
1076 			(*strp)[wcnt] = '\0';
1077 			break;
1078 		} else if (write(fd, st, wcnt) != wcnt) {
1079 			syswarn(1, errno, "Failed write to file %s", name);
1080 			return(-1);
1081 		}
1082 		st += wcnt;
1083 	}
1084 	return(st - str);
1085 }
1086 
1087 /*
1088  * file_flush()
1089  *	when the last file block in a file is zero, many file systems will not
1090  *	let us create a hole at the end. To get the last block with zeros, we
1091  *	write the last BYTE with a zero (back up one byte and write a zero).
1092  */
1093 
1094 void
file_flush(int fd,char * fname,int isempt)1095 file_flush(int fd, char *fname, int isempt)
1096 {
1097 	static char blnk[] = "\0";
1098 
1099 	/*
1100 	 * silly test, but make sure we are only called when the last block is
1101 	 * filled with all zeros.
1102 	 */
1103 	if (!isempt)
1104 		return;
1105 
1106 	/*
1107 	 * move back one byte and write a zero
1108 	 */
1109 	if (lseek(fd, (off_t)-1, SEEK_CUR) < 0) {
1110 		syswarn(1, errno, "Failed seek on file %s", fname);
1111 		return;
1112 	}
1113 
1114 	if (write(fd, blnk, 1) < 0)
1115 		syswarn(1, errno, "Failed write to file %s", fname);
1116 	return;
1117 }
1118 
1119 /*
1120  * rdfile_close()
1121  *	close a file we have been reading (to copy or archive). If we have to
1122  *	reset access time (tflag) do so (the times are stored in arcn).
1123  */
1124 
1125 void
rdfile_close(ARCHD * arcn,int * fd)1126 rdfile_close(ARCHD *arcn, int *fd)
1127 {
1128 	/*
1129 	 * make sure the file is open
1130 	 */
1131 	if (*fd < 0)
1132 		return;
1133 
1134 	(void)close(*fd);
1135 	*fd = -1;
1136 	if (!tflag)
1137 		return;
1138 
1139 	/*
1140 	 * user wants last access time reset
1141 	 */
1142 	set_ftime(arcn->org_name, arcn->sb.st_mtime, arcn->sb.st_atime, 1);
1143 	return;
1144 }
1145 
1146 /*
1147  * set_crc()
1148  *	read a file to calculate its crc. This is a real drag. Archive formats
1149  *	that have this, end up reading the file twice (we have to write the
1150  *	header WITH the crc before writing the file contents. Oh well...
1151  * Return:
1152  *	0 if was able to calculate the crc, -1 otherwise
1153  */
1154 
1155 int
set_crc(ARCHD * arcn,int fd)1156 set_crc(ARCHD *arcn, int fd)
1157 {
1158 	int i;
1159 	int res;
1160 	off_t cpcnt = 0L;
1161 	u_long size;
1162 	u_int32_t crc = 0;
1163 	char tbuf[FILEBLK];
1164 	struct stat sb;
1165 
1166 	if (fd < 0) {
1167 		/*
1168 		 * hmm, no fd, should never happen. well no crc then.
1169 		 */
1170 		arcn->crc = 0L;
1171 		return(0);
1172 	}
1173 
1174 	if ((size = (u_long)arcn->sb.st_blksize) > (u_long)sizeof(tbuf))
1175 		size = (u_long)sizeof(tbuf);
1176 
1177 	/*
1178 	 * read all the bytes we think that there are in the file. If the user
1179 	 * is trying to archive an active file, forget this file.
1180 	 */
1181 	for (;;) {
1182 		if ((res = read(fd, tbuf, size)) <= 0)
1183 			break;
1184 		cpcnt += res;
1185 		for (i = 0; i < res; ++i)
1186 			crc += (tbuf[i] & 0xff);
1187 	}
1188 
1189 	/*
1190 	 * safety check. we want to avoid archiving files that are active as
1191 	 * they can create inconsistent archive copies.
1192 	 */
1193 	if (cpcnt != arcn->sb.st_size)
1194 		paxwarn(1, "File changed size %s", arcn->org_name);
1195 	else if (fstat(fd, &sb) < 0)
1196 		syswarn(1, errno, "Failed stat on %s", arcn->org_name);
1197 	else if (arcn->sb.st_mtime != sb.st_mtime)
1198 		paxwarn(1, "File %s was modified during read", arcn->org_name);
1199 	else if (lseek(fd, (off_t)0L, SEEK_SET) < 0)
1200 		syswarn(1, errno, "File rewind failed on: %s", arcn->org_name);
1201 	else {
1202 		arcn->crc = crc;
1203 		return(0);
1204 	}
1205 	return(-1);
1206 }
1207