1 /* $NetBSD: file_subs.c,v 1.65 2023/05/28 21:42:40 lukem 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 1.65 2023/05/28 21:42:40 lukem 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
apply_umask(mode_t mode)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
file_creat(ARCHD * arcn,int write_to_hardlink)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_cleanup()
183 * Close file descriptor to a file just created by pax.
184 * Unlinks any temporary file.
185 */
186
187 void
file_cleanup(ARCHD * arcn,int fd)188 file_cleanup(ARCHD *arcn, int fd)
189 {
190 char *tmp_name;
191
192 if (fd < 0)
193 return;
194
195 tmp_name = (arcn->tmp_name != NULL) ? arcn->tmp_name : arcn->name;
196
197 if (close(fd) < 0)
198 syswarn(0, errno, "Cannot close file descriptor on %s",
199 tmp_name);
200
201 /* No temporary file to cleanup? */
202 if (arcn->tmp_name == NULL)
203 return;
204
205 /* Cleanup the temporary file. */
206 if (unlink(arcn->tmp_name) < 0) {
207 syswarn(0, errno, "Cannot unlink %s", arcn->tmp_name);
208 }
209
210 free(arcn->tmp_name);
211 arcn->tmp_name = NULL;
212 xtmp_name = NULL;
213 }
214
215 /*
216 * file_close()
217 * Close file descriptor to a file just created by pax. Sets modes,
218 * ownership and times as required.
219 */
220
221 void
file_close(ARCHD * arcn,int fd)222 file_close(ARCHD *arcn, int fd)
223 {
224 char *tmp_name;
225 int res;
226
227 if (fd < 0)
228 return;
229
230 tmp_name = (arcn->tmp_name != NULL) ? arcn->tmp_name : arcn->name;
231
232 if (close(fd) < 0)
233 syswarn(0, errno, "Cannot close file descriptor on %s",
234 tmp_name);
235
236 /*
237 * set owner/groups first as this may strip off mode bits we want
238 * then set file permission modes. Then set file access and
239 * modification times.
240 */
241 if (pids)
242 res = set_ids(tmp_name, arcn->sb.st_uid, arcn->sb.st_gid);
243 else
244 res = 0;
245
246 /*
247 * IMPORTANT SECURITY NOTE:
248 * if not preserving mode or we cannot set uid/gid, then PROHIBIT
249 * set uid/gid bits but restore the file modes (since mkstemp doesn't).
250 */
251 if (!pmode || res)
252 arcn->sb.st_mode &= ~SETBITS(0);
253 if (pmode)
254 set_pmode(tmp_name, arcn->sb.st_mode);
255 else
256 set_pmode(tmp_name,
257 apply_umask((arcn->sb.st_mode & FILEBITS(0))));
258 if (patime || pmtime)
259 set_ftime(tmp_name, arcn->sb.st_mtime,
260 arcn->sb.st_atime, 0, 0);
261
262 /* Did we write directly to the target file? */
263 if (arcn->tmp_name == NULL)
264 return;
265
266 /*
267 * Finally, now the temp file is fully instantiated rename it to
268 * the desired file name.
269 */
270 if (rename(tmp_name, arcn->name) < 0) {
271 syswarn(0, errno, "Cannot rename %s to %s",
272 tmp_name, arcn->name);
273 (void)unlink(tmp_name);
274 }
275
276 #if HAVE_STRUCT_STAT_ST_FLAGS
277 if (pfflags && arcn->type != PAX_SLK)
278 set_chflags(arcn->name, arcn->sb.st_flags);
279 #endif
280
281 free(arcn->tmp_name);
282 arcn->tmp_name = NULL;
283 xtmp_name = NULL;
284 }
285
286 /*
287 * lnk_creat()
288 * Create a hard link to arcn->ln_name from arcn->name. arcn->ln_name
289 * must exist;
290 * Return:
291 * 0 if ok, -1 otherwise
292 */
293
294 int
lnk_creat(ARCHD * arcn,int * payload)295 lnk_creat(ARCHD *arcn, int *payload)
296 {
297 struct stat sb;
298
299 /*
300 * Check if this hardlink carries the "payload". In "cpio" archives
301 * it's usually the last record of a set of hardlinks which includes
302 * the contents of the file.
303 *
304 */
305 *payload = S_ISREG(arcn->sb.st_mode) &&
306 (arcn->sb.st_size > 0) && (arcn->sb.st_size <= arcn->skip);
307
308 /*
309 * We may be running as root, so we have to be sure that link target
310 * is not a directory, so we lstat and check. XXX: This is still racy.
311 */
312 if (lstat(arcn->ln_name, &sb) != -1 && S_ISDIR(sb.st_mode)) {
313 tty_warn(1, "A hard link to the directory %s is not allowed",
314 arcn->ln_name);
315 return -1;
316 }
317
318 return mk_link(arcn->ln_name, &sb, arcn->name, 0);
319 }
320
321 /*
322 * cross_lnk()
323 * Create a hard link to arcn->org_name from arcn->name. Only used in copy
324 * with the -l flag. No warning or error if this does not succeed (we will
325 * then just create the file)
326 * Return:
327 * 1 if copy() should try to create this file node
328 * 0 if cross_lnk() ok, -1 for fatal flaw (like linking to self).
329 */
330
331 int
cross_lnk(ARCHD * arcn)332 cross_lnk(ARCHD *arcn)
333 {
334 /*
335 * try to make a link to original file (-l flag in copy mode). make
336 * sure we do not try to link to directories in case we are running as
337 * root (and it might succeed).
338 */
339 if (arcn->type == PAX_DIR)
340 return 1;
341 return mk_link(arcn->org_name, &(arcn->sb), arcn->name, 1);
342 }
343
344 /*
345 * chk_same()
346 * In copy mode if we are not trying to make hard links between the src
347 * and destinations, make sure we are not going to overwrite ourselves by
348 * accident. This slows things down a little, but we have to protect all
349 * those people who make typing errors.
350 * Return:
351 * 1 the target does not exist, go ahead and copy
352 * 0 skip it file exists (-k) or may be the same as source file
353 */
354
355 int
chk_same(ARCHD * arcn)356 chk_same(ARCHD *arcn)
357 {
358 struct stat sb;
359
360 /*
361 * if file does not exist, return. if file exists and -k, skip it
362 * quietly
363 */
364 if (lstat(arcn->name, &sb) < 0)
365 return 1;
366 if (kflag)
367 return 0;
368
369 /*
370 * better make sure the user does not have src == dest by mistake
371 */
372 if ((arcn->sb.st_dev == sb.st_dev) && (arcn->sb.st_ino == sb.st_ino)) {
373 tty_warn(1, "Unable to copy %s, file would overwrite itself",
374 arcn->name);
375 return 0;
376 }
377 return 1;
378 }
379
380 /*
381 * mk_link()
382 * try to make a hard link between two files. if ign set, we do not
383 * complain.
384 * Return:
385 * 0 if successful (or we are done with this file but no error, such as
386 * finding the from file exists and the user has set -k).
387 * 1 when ign was set to indicates we could not make the link but we
388 * should try to copy/extract the file as that might work (and is an
389 * allowed option). -1 an error occurred.
390 */
391
392 static int
mk_link(char * to,struct stat * to_sb,char * from,int ign)393 mk_link(char *to, struct stat *to_sb, char *from, int ign)
394 {
395 struct stat sb;
396 int oerrno;
397
398 /*
399 * if from file exists, it has to be unlinked to make the link. If the
400 * file exists and -k is set, skip it quietly
401 */
402 if (lstat(from, &sb) == 0) {
403 if (kflag)
404 return 0;
405
406 /*
407 * make sure it is not the same file, protect the user
408 */
409 if ((to_sb->st_dev==sb.st_dev)&&(to_sb->st_ino == sb.st_ino)) {
410 tty_warn(1, "Cannot link file %s to itself", to);
411 return -1;
412 }
413
414 /*
415 * try to get rid of the file, based on the type
416 */
417 if (S_ISDIR(sb.st_mode) && strcmp(from, ".") != 0) {
418 if (rmdir(from) < 0) {
419 syswarn(1, errno, "Cannot remove %s", from);
420 return -1;
421 }
422 } else if (unlink(from) < 0) {
423 if (!ign) {
424 syswarn(1, errno, "Cannot remove %s", from);
425 return -1;
426 }
427 return 1;
428 }
429 }
430
431 /*
432 * from file is gone (or did not exist), try to make the hard link.
433 * if it fails, check the path and try it again (if chk_path() says to
434 * try again)
435 */
436 for (;;) {
437 if (link(to, from) == 0)
438 break;
439 oerrno = errno;
440 if (chk_path(from, to_sb->st_uid, to_sb->st_gid) == 0)
441 continue;
442 if (!ign) {
443 syswarn(1, oerrno, "Cannot link to %s from %s", to,
444 from);
445 return -1;
446 }
447 return 1;
448 }
449
450 /*
451 * all right the link was made
452 */
453 return 0;
454 }
455
456 /*
457 * node_creat()
458 * create an entry in the file system (other than a file or hard link).
459 * If successful, sets uid/gid modes and times as required.
460 * Return:
461 * 0 if ok, -1 otherwise
462 */
463
464 int
node_creat(ARCHD * arcn)465 node_creat(ARCHD *arcn)
466 {
467 int res;
468 int ign = 0;
469 int oerrno;
470 int pass = 0;
471 mode_t file_mode;
472 struct stat sb;
473 char target[MAXPATHLEN];
474 char *nm = arcn->name;
475 int len;
476
477 /*
478 * create node based on type, if that fails try to unlink the node and
479 * try again. finally check the path and try again. As noted in the
480 * file and link creation routines, this method seems to exhibit the
481 * best performance in general use workloads.
482 */
483 file_mode = arcn->sb.st_mode & FILEBITS(arcn->type == PAX_DIR);
484
485 for (;;) {
486 switch (arcn->type) {
487 case PAX_DIR:
488 /*
489 * If -h (or -L) was given in tar-mode, follow the
490 * potential symlink chain before trying to create the
491 * directory.
492 */
493 if (strcmp(NM_TAR, argv0) == 0 && Lflag) {
494 while (lstat(nm, &sb) == 0 &&
495 S_ISLNK(sb.st_mode)) {
496 len = readlink(nm, target,
497 sizeof target - 1);
498 if (len == -1) {
499 syswarn(0, errno,
500 "cannot follow symlink %s "
501 "in chain for %s",
502 nm, arcn->name);
503 res = -1;
504 goto badlink;
505 }
506 target[len] = '\0';
507 nm = target;
508 }
509 }
510 res = domkdir(nm, file_mode);
511 badlink:
512 if (ign)
513 res = 0;
514 break;
515 case PAX_CHR:
516 file_mode |= S_IFCHR;
517 res = mknod(nm, file_mode, arcn->sb.st_rdev);
518 break;
519 case PAX_BLK:
520 file_mode |= S_IFBLK;
521 res = mknod(nm, file_mode, arcn->sb.st_rdev);
522 break;
523 case PAX_FIF:
524 res = mkfifo(nm, file_mode);
525 break;
526 case PAX_SCK:
527 /*
528 * Skip sockets, operation has no meaning under BSD
529 */
530 tty_warn(0,
531 "%s skipped. Sockets cannot be copied or extracted",
532 nm);
533 return (-1);
534 case PAX_SLK:
535 res = symlink(arcn->ln_name, nm);
536 break;
537 case PAX_CTG:
538 case PAX_HLK:
539 case PAX_HRG:
540 case PAX_REG:
541 default:
542 /*
543 * we should never get here
544 */
545 tty_warn(0, "%s has an unknown file type, skipping",
546 nm);
547 return (-1);
548 }
549
550 /*
551 * if we were able to create the node break out of the loop,
552 * otherwise try to unlink the node and try again. if that
553 * fails check the full path and try a final time.
554 */
555 if (res == 0)
556 break;
557
558 /*
559 * we failed to make the node
560 */
561 oerrno = errno;
562 switch (pass++) {
563 case 0:
564 if ((ign = unlnk_exist(nm, arcn->type)) < 0)
565 return (-1);
566 continue;
567
568 case 1:
569 if (nodirs ||
570 chk_path(nm, arcn->sb.st_uid,
571 arcn->sb.st_gid) < 0) {
572 syswarn(1, oerrno, "Cannot create %s", nm);
573 return (-1);
574 }
575 continue;
576 }
577
578 /*
579 * it must be a file that exists but we can't create or
580 * remove, but we must avoid the infinite loop.
581 */
582 break;
583 }
584
585 /*
586 * we were able to create the node. set uid/gid, modes and times
587 */
588 if (pids)
589 res = set_ids(nm, arcn->sb.st_uid, arcn->sb.st_gid);
590 else
591 res = 0;
592
593 /*
594 * IMPORTANT SECURITY NOTE:
595 * if not preserving mode or we cannot set uid/gid, then PROHIBIT any
596 * set uid/gid bits
597 */
598 if (!pmode || res)
599 arcn->sb.st_mode &= ~SETBITS(arcn->type == PAX_DIR);
600 if (pmode)
601 set_pmode(arcn->name, arcn->sb.st_mode);
602
603 if (arcn->type == PAX_DIR && strcmp(NM_CPIO, argv0) != 0) {
604 /*
605 * Dirs must be processed again at end of extract to set times
606 * and modes to agree with those stored in the archive. However
607 * to allow extract to continue, we may have to also set owner
608 * rights. This allows nodes in the archive that are children
609 * of this directory to be extracted without failure. Both time
610 * and modes will be fixed after the entire archive is read and
611 * before pax exits.
612 */
613 if (access(nm, R_OK | W_OK | X_OK) < 0) {
614 if (lstat(nm, &sb) < 0) {
615 syswarn(0, errno,"Cannot access %s (stat)",
616 arcn->name);
617 set_pmode(nm,file_mode | S_IRWXU);
618 } else {
619 /*
620 * We have to add rights to the dir, so we make
621 * sure to restore the mode. The mode must be
622 * restored AS CREATED and not as stored if
623 * pmode is not set.
624 */
625 set_pmode(nm, ((sb.st_mode &
626 FILEBITS(arcn->type == PAX_DIR)) |
627 S_IRWXU));
628 if (!pmode)
629 arcn->sb.st_mode = sb.st_mode;
630 }
631
632 /*
633 * we have to force the mode to what was set here,
634 * since we changed it from the default as created.
635 */
636 add_dir(nm, arcn->nlen, &(arcn->sb), 1);
637 } else if (pmode || patime || pmtime)
638 add_dir(nm, arcn->nlen, &(arcn->sb), 0);
639 }
640
641 if (patime || pmtime)
642 set_ftime(arcn->name, arcn->sb.st_mtime,
643 arcn->sb.st_atime, 0, (arcn->type == PAX_SLK) ? 1 : 0);
644
645 #if HAVE_STRUCT_STAT_ST_FLAGS
646 if (pfflags && arcn->type != PAX_SLK)
647 set_chflags(arcn->name, arcn->sb.st_flags);
648 #endif
649 return 0;
650 }
651
652 /*
653 * unlnk_exist()
654 * Remove node from file system with the specified name. We pass the type
655 * of the node that is going to replace it. When we try to create a
656 * directory and find that it already exists, we allow processing to
657 * continue as proper modes etc will always be set for it later on.
658 * Return:
659 * 0 is ok to proceed, no file with the specified name exists
660 * -1 we were unable to remove the node, or we should not remove it (-k)
661 * 1 we found a directory and we were going to create a directory.
662 */
663
664 int
unlnk_exist(char * name,int type)665 unlnk_exist(char *name, int type)
666 {
667 struct stat sb;
668
669 /*
670 * the file does not exist, or -k we are done
671 */
672 if (lstat(name, &sb) < 0)
673 return 0;
674 if (kflag)
675 return -1;
676
677 if (S_ISDIR(sb.st_mode)) {
678 /*
679 * try to remove a directory, if it fails and we were going to
680 * create a directory anyway, tell the caller (return a 1).
681 *
682 * don't try to remove the directory if the name is "."
683 * otherwise later file/directory creation fails.
684 */
685 if (strcmp(name, ".") == 0)
686 return 1;
687 if (rmdir(name) < 0) {
688 if (type == PAX_DIR)
689 return 1;
690 syswarn(1, errno, "Cannot remove directory %s", name);
691 return -1;
692 }
693 return 0;
694 }
695
696 /*
697 * try to get rid of all non-directory type nodes
698 */
699 if (unlink(name) < 0) {
700 (void)fflush(listf);
701 syswarn(1, errno, "Cannot unlink %s", name);
702 return -1;
703 }
704 return 0;
705 }
706
707 /*
708 * chk_path()
709 * We were trying to create some kind of node in the file system and it
710 * failed. chk_path() makes sure the path up to the node exists and is
711 * writable. When we have to create a directory that is missing along the
712 * path somewhere, the directory we create will be set to the same
713 * uid/gid as the file has (when uid and gid are being preserved).
714 * NOTE: this routine is a real performance loss. It is only used as a
715 * last resort when trying to create entries in the file system.
716 * Return:
717 * -1 when it could find nothing it is allowed to fix.
718 * 0 otherwise
719 */
720
721 int
chk_path(char * name,uid_t st_uid,gid_t st_gid)722 chk_path(char *name, uid_t st_uid, gid_t st_gid)
723 {
724 char *spt = name;
725 struct stat sb;
726 int retval = -1;
727
728 /*
729 * watch out for paths with nodes stored directly in / (e.g. /bozo)
730 */
731 if (*spt == '/')
732 ++spt;
733
734 for(;;) {
735 /*
736 * work forward from the first / and check each part of
737 * the path
738 */
739 spt = strchr(spt, '/');
740 if (spt == NULL)
741 break;
742 *spt = '\0';
743
744 /*
745 * if it exists we assume it is a directory, it is not within
746 * the spec (at least it seems to read that way) to alter the
747 * file system for nodes NOT EXPLICITLY stored on the archive.
748 * If that assumption is changed, you would test the node here
749 * and figure out how to get rid of it (probably like some
750 * recursive unlink()) or fix up the directory permissions if
751 * required (do an access()).
752 */
753 if (lstat(name, &sb) == 0) {
754 *(spt++) = '/';
755 continue;
756 }
757
758 /*
759 * the path fails at this point, see if we can create the
760 * needed directory and continue on
761 */
762 if (domkdir(name, S_IRWXU | S_IRWXG | S_IRWXO) == -1) {
763 *spt = '/';
764 retval = -1;
765 break;
766 }
767
768 /*
769 * we were able to create the directory. We will tell the
770 * caller that we found something to fix, and it is ok to try
771 * and create the node again.
772 */
773 retval = 0;
774 if (pids)
775 (void)set_ids(name, st_uid, st_gid);
776
777 /*
778 * make sure the user doesn't have some strange umask that
779 * causes this newly created directory to be unusable. We fix
780 * the modes and restore them back to the creation default at
781 * the end of pax
782 */
783 if ((access(name, R_OK | W_OK | X_OK) < 0) &&
784 (lstat(name, &sb) == 0)) {
785 set_pmode(name, ((sb.st_mode & FILEBITS(0)) |
786 S_IRWXU));
787 add_dir(name, spt - name, &sb, 1);
788 }
789 *(spt++) = '/';
790 continue;
791 }
792 /*
793 * We perform one final check here, because if someone else
794 * created the directory in parallel with us, we might return
795 * the wrong error code, even if the directory exists now.
796 */
797 if (retval == -1 && stat(name, &sb) == 0 && S_ISDIR(sb.st_mode))
798 retval = 0;
799 return retval;
800 }
801
802 /*
803 * set_ftime()
804 * Set the access time and modification time for a named file. If frc
805 * is non-zero we force these times to be set even if the user did not
806 * request access and/or modification time preservation (this is also
807 * used by -t to reset access times).
808 * When ign is zero, only those times the user has asked for are set, the
809 * other ones are left alone. We do not assume the un-documented feature
810 * of many utimes() implementations that consider a 0 time value as a do
811 * not set request.
812 *
813 * Unfortunately, there are systems where lutimes() is present but does
814 * not work on some filesystem types, which cannot be detected at
815 * compile time. This requires passing down symlink knowledge into
816 * this function to obtain correct operation. Linux with XFS is one
817 * example of such a system.
818 */
819
820 void
set_ftime(char * fnm,time_t mtime,time_t atime,int frc,int slk)821 set_ftime(char *fnm, time_t mtime, time_t atime, int frc, int slk)
822 {
823 struct timeval tv[2];
824 struct stat sb;
825
826 tv[0].tv_sec = atime;
827 tv[0].tv_usec = 0;
828 tv[1].tv_sec = mtime;
829 tv[1].tv_usec = 0;
830 if (!frc && (!patime || !pmtime)) {
831 /*
832 * if we are not forcing, only set those times the user wants
833 * set. We get the current values of the times if we need them.
834 */
835 if (lstat(fnm, &sb) == 0) {
836 #if BSD4_4 && !HAVE_NBTOOL_CONFIG_H
837 if (!patime)
838 TIMESPEC_TO_TIMEVAL(&tv[0], &sb.st_atimespec);
839 if (!pmtime)
840 TIMESPEC_TO_TIMEVAL(&tv[1], &sb.st_mtimespec);
841 #else
842 if (!patime)
843 tv[0].tv_sec = sb.st_atime;
844 if (!pmtime)
845 tv[1].tv_sec = sb.st_mtime;
846 #endif
847 } else
848 syswarn(0, errno, "Cannot obtain file stats %s", fnm);
849 }
850
851 /*
852 * set the times
853 */
854 #if HAVE_LUTIMES
855 if (lutimes(fnm, tv) == 0)
856 return;
857 if (errno != ENOSYS) /* XXX linux: lutimes is per-FS */
858 goto bad;
859 #endif
860 if (slk)
861 return;
862 if (utimes(fnm, tv) == -1)
863 goto bad;
864 return;
865 bad:
866 syswarn(1, errno, "Access/modification time set failed on: %s", fnm);
867 }
868
869 /*
870 * set_ids()
871 * set the uid and gid of a file system node
872 * Return:
873 * 0 when set, -1 on failure
874 */
875
876 int
set_ids(char * fnm,uid_t uid,gid_t gid)877 set_ids(char *fnm, uid_t uid, gid_t gid)
878 {
879 if (geteuid() == 0)
880 if (lchown(fnm, uid, gid)) {
881 (void)fflush(listf);
882 syswarn(1, errno, "Cannot set file uid/gid of %s",
883 fnm);
884 return -1;
885 }
886 return 0;
887 }
888
889 /*
890 * set_pmode()
891 * Set file access mode
892 */
893
894 void
set_pmode(char * fnm,mode_t mode)895 set_pmode(char *fnm, mode_t mode)
896 {
897 mode &= A_BITS;
898 if (lchmod(fnm, mode)) {
899 (void)fflush(listf);
900 syswarn(1, errno, "Cannot set permissions on %s", fnm);
901 }
902 return;
903 }
904
905 /*
906 * set_chflags()
907 * Set 4.4BSD file flags
908 */
909 void
set_chflags(char * fnm,u_int32_t flags)910 set_chflags(char *fnm, u_int32_t flags)
911 {
912
913 #if 0
914 if (chflags(fnm, flags) < 0 && errno != EOPNOTSUPP)
915 syswarn(1, errno, "Cannot set file flags on %s", fnm);
916 #endif
917 return;
918 }
919
920 /*
921 * file_write()
922 * Write/copy a file (during copy or archive extract). This routine knows
923 * how to copy files with lseek holes in it. (Which are read as file
924 * blocks containing all 0's but do not have any file blocks associated
925 * with the data). Typical examples of these are files created by dbm
926 * variants (.pag files). While the file size of these files are huge, the
927 * actual storage is quite small (the files are sparse). The problem is
928 * the holes read as all zeros so are probably stored on the archive that
929 * way (there is no way to determine if the file block is really a hole,
930 * we only know that a file block of all zeros can be a hole).
931 * At this writing, no major archive format knows how to archive files
932 * with holes. However, on extraction (or during copy, -rw) we have to
933 * deal with these files. Without detecting the holes, the files can
934 * consume a lot of file space if just written to disk. This replacement
935 * for write when passed the basic allocation size of a file system block,
936 * uses lseek whenever it detects the input data is all 0 within that
937 * file block. In more detail, the strategy is as follows:
938 * While the input is all zero keep doing an lseek. Keep track of when we
939 * pass over file block boundaries. Only write when we hit a non zero
940 * input. once we have written a file block, we continue to write it to
941 * the end (we stop looking at the input). When we reach the start of the
942 * next file block, start checking for zero blocks again. Working on file
943 * block boundaries significantly reduces the overhead when copying files
944 * that are NOT very sparse. This overhead (when compared to a write) is
945 * almost below the measurement resolution on many systems. Without it,
946 * files with holes cannot be safely copied. It does has a side effect as
947 * it can put holes into files that did not have them before, but that is
948 * not a problem since the file contents are unchanged (in fact it saves
949 * file space). (Except on paging files for diskless clients. But since we
950 * cannot determine one of those file from here, we ignore them). If this
951 * ever ends up on a system where CTG files are supported and the holes
952 * are not desired, just do a conditional test in those routines that
953 * call file_write() and have it call write() instead. BEFORE CLOSING THE
954 * FILE, make sure to call file_flush() when the last write finishes with
955 * an empty block. A lot of file systems will not create an lseek hole at
956 * the end. In this case we drop a single 0 at the end to force the
957 * trailing 0's in the file.
958 * ---Parameters---
959 * rem: how many bytes left in this file system block
960 * isempt: have we written to the file block yet (is it empty)
961 * sz: basic file block allocation size
962 * cnt: number of bytes on this write
963 * str: buffer to write
964 * Return:
965 * number of bytes written, -1 on write (or lseek) error.
966 */
967
968 int
file_write(int fd,char * str,int cnt,int * rem,int * isempt,int sz,char * name)969 file_write(int fd, char *str, int cnt, int *rem, int *isempt, int sz,
970 char *name)
971 {
972 char *pt;
973 char *end;
974 int wcnt;
975 char *st = str;
976 char **strp;
977 size_t *lenp;
978
979 /*
980 * while we have data to process
981 */
982 while (cnt) {
983 if (!*rem) {
984 /*
985 * We are now at the start of file system block again
986 * (or what we think one is...). start looking for
987 * empty blocks again
988 */
989 *isempt = 1;
990 *rem = sz;
991 }
992
993 /*
994 * only examine up to the end of the current file block or
995 * remaining characters to write, whatever is smaller
996 */
997 wcnt = MIN(cnt, *rem);
998 cnt -= wcnt;
999 *rem -= wcnt;
1000 if (*isempt) {
1001 /*
1002 * have not written to this block yet, so we keep
1003 * looking for zeros
1004 */
1005 pt = st;
1006 end = st + wcnt;
1007
1008 /*
1009 * look for a zero filled buffer
1010 */
1011 while ((pt < end) && (*pt == '\0'))
1012 ++pt;
1013
1014 if (pt == end) {
1015 /*
1016 * skip, buf is empty so far
1017 */
1018 if (fd > -1 &&
1019 lseek(fd, (off_t)wcnt, SEEK_CUR) < 0) {
1020 syswarn(1, errno, "File seek on %s",
1021 name);
1022 return -1;
1023 }
1024 st = pt;
1025 continue;
1026 }
1027 /*
1028 * drat, the buf is not zero filled
1029 */
1030 *isempt = 0;
1031 }
1032
1033 /*
1034 * have non-zero data in this file system block, have to write
1035 */
1036 switch (fd) {
1037 case -PAX_GLF:
1038 strp = &gnu_name_string;
1039 lenp = &gnu_name_length;
1040 break;
1041 case -PAX_GLL:
1042 strp = &gnu_link_string;
1043 lenp = &gnu_link_length;
1044 break;
1045 default:
1046 strp = NULL;
1047 lenp = NULL;
1048 break;
1049 }
1050 if (strp) {
1051 char *nstr = *strp ? realloc(*strp, *lenp + wcnt + 1) :
1052 malloc(wcnt + 1);
1053 if (nstr == NULL) {
1054 tty_warn(1, "Out of memory");
1055 return -1;
1056 }
1057 (void)strlcpy(&nstr[*lenp], st, wcnt + 1);
1058 *strp = nstr;
1059 *lenp += wcnt;
1060 } else if (xwrite(fd, st, wcnt) != wcnt) {
1061 syswarn(1, errno, "Failed write to file %s", name);
1062 return -1;
1063 }
1064 st += wcnt;
1065 }
1066 return st - str;
1067 }
1068
1069 /*
1070 * file_flush()
1071 * when the last file block in a file is zero, many file systems will not
1072 * let us create a hole at the end. To get the last block with zeros, we
1073 * write the last BYTE with a zero (back up one byte and write a zero).
1074 * Return:
1075 * 0 if was able to flush the file, -1 otherwise
1076 */
1077
1078 int
file_flush(int fd,char * fname,int isempt)1079 file_flush(int fd, char *fname, int isempt)
1080 {
1081 static char blnk[] = "\0";
1082
1083 /*
1084 * silly test, but make sure we are only called when the last block is
1085 * filled with all zeros.
1086 */
1087 if (!isempt)
1088 return 0;
1089
1090 /*
1091 * move back one byte and write a zero
1092 */
1093 if (lseek(fd, (off_t)-1, SEEK_CUR) < 0) {
1094 syswarn(1, errno, "Failed seek on file %s", fname);
1095 return -1;
1096 }
1097
1098 if (write_with_restart(fd, blnk, 1) < 0) {
1099 syswarn(1, errno, "Failed write to file %s", fname);
1100 return -1;
1101 }
1102 return 0;
1103 }
1104
1105 /*
1106 * rdfile_close()
1107 * close a file we have been reading (to copy or archive). If we have to
1108 * reset access time (tflag) do so (the times are stored in arcn).
1109 */
1110
1111 void
rdfile_close(ARCHD * arcn,int * fd)1112 rdfile_close(ARCHD *arcn, int *fd)
1113 {
1114 /*
1115 * make sure the file is open
1116 */
1117 if (*fd < 0)
1118 return;
1119
1120 (void)close(*fd);
1121 *fd = -1;
1122 if (!tflag)
1123 return;
1124
1125 /*
1126 * user wants last access time reset
1127 */
1128 set_ftime(arcn->org_name, arcn->sb.st_mtime, arcn->sb.st_atime, 1, 0);
1129 return;
1130 }
1131
1132 /*
1133 * set_crc()
1134 * read a file to calculate its crc. This is a real drag. Archive formats
1135 * that have this, end up reading the file twice (we have to write the
1136 * header WITH the crc before writing the file contents. Oh well...
1137 * Return:
1138 * 0 if was able to calculate the crc, -1 otherwise
1139 */
1140
1141 int
set_crc(ARCHD * arcn,int fd)1142 set_crc(ARCHD *arcn, int fd)
1143 {
1144 int i;
1145 int res;
1146 off_t cpcnt = 0L;
1147 u_long size;
1148 unsigned long crc = 0L;
1149 char tbuf[FILEBLK];
1150 struct stat sb;
1151
1152 if (fd < 0) {
1153 /*
1154 * hmm, no fd, should never happen. well no crc then.
1155 */
1156 arcn->crc = 0L;
1157 return 0;
1158 }
1159
1160 if ((size = (u_long)arcn->sb.st_blksize) > (u_long)sizeof(tbuf))
1161 size = (u_long)sizeof(tbuf);
1162
1163 /*
1164 * read all the bytes we think that there are in the file. If the user
1165 * is trying to archive an active file, forget this file.
1166 */
1167 for(;;) {
1168 if ((res = read(fd, tbuf, size)) <= 0)
1169 break;
1170 cpcnt += res;
1171 for (i = 0; i < res; ++i)
1172 crc += (tbuf[i] & 0xff);
1173 }
1174
1175 /*
1176 * safety check. we want to avoid archiving files that are active as
1177 * they can create inconsistent archive copies.
1178 */
1179 if (cpcnt != arcn->sb.st_size)
1180 tty_warn(1, "File changed size %s", arcn->org_name);
1181 else if (fstat(fd, &sb) < 0)
1182 syswarn(1, errno, "Failed stat on %s", arcn->org_name);
1183 else if (arcn->sb.st_mtime != sb.st_mtime)
1184 tty_warn(1, "File %s was modified during read", arcn->org_name);
1185 else if (lseek(fd, (off_t)0L, SEEK_SET) < 0)
1186 syswarn(1, errno, "File rewind failed on: %s", arcn->org_name);
1187 else {
1188 arcn->crc = crc;
1189 return 0;
1190 }
1191 return -1;
1192 }
1193