file_cmds-287.11.1.tar.gz
[apple/file_cmds.git] / pax / file_subs.c
1 /* $OpenBSD: file_subs.c,v 1.30 2005/11/09 19:59:06 otto Exp $ */
2 /* $NetBSD: file_subs.c,v 1.4 1995/03/21 09:07:18 cgd Exp $ */
3
4 /*-
5 * Copyright (c) 1992 Keith Muller.
6 * Copyright (c) 1992, 1993
7 * The Regents of the University of California. All rights reserved.
8 *
9 * This code is derived from software contributed to Berkeley by
10 * Keith Muller of the University of California, San Diego.
11 *
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
14 * are met:
15 * 1. Redistributions of source code must retain the above copyright
16 * notice, this list of conditions and the following disclaimer.
17 * 2. Redistributions in binary form must reproduce the above copyright
18 * notice, this list of conditions and the following disclaimer in the
19 * documentation and/or other materials provided with the distribution.
20 * 3. Neither the name of the University nor the names of its contributors
21 * may be used to endorse or promote products derived from this software
22 * without specific prior written permission.
23 *
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * SUCH DAMAGE.
35 */
36
37 #include <sys/cdefs.h>
38 #ifndef lint
39 #if 0
40 static const char sccsid[] = "@(#)file_subs.c 8.1 (Berkeley) 5/31/93";
41 #else
42 __used static const char rcsid[] = "$OpenBSD: file_subs.c,v 1.30 2005/11/09 19:59:06 otto Exp $";
43 #endif
44 #endif /* not lint */
45
46 #include <sys/param.h>
47 #include <sys/time.h>
48 #include <sys/stat.h>
49 #include <sys/uio.h>
50 #include <err.h>
51 #include <errno.h>
52 #include <fcntl.h>
53 #include <stdio.h>
54 #include <stdlib.h>
55 #include <string.h>
56 #include <unistd.h>
57 #include "pax.h"
58 #include "options.h"
59 #include "extern.h"
60
61 static int
62 mk_link(char *, struct stat *, char *, int);
63
64 /*
65 * routines that deal with file operations such as: creating, removing;
66 * and setting access modes, uid/gid and times of files
67 */
68
69 #define FILEBITS (S_ISVTX | S_IRWXU | S_IRWXG | S_IRWXO)
70 #define SETBITS (S_ISUID | S_ISGID)
71 #define ABITS (FILEBITS | SETBITS)
72
73 /*
74 * file_creat()
75 * Create and open a file.
76 * Return:
77 * file descriptor or -1 for failure
78 */
79
80 int
81 file_creat(ARCHD *arcn)
82 {
83 int fd = -1;
84 mode_t file_mode;
85 int oerrno;
86 int rc = 0;
87 char *path_to_open;
88 char *new_path;
89 char *cwd;
90 char cwd_buff[MAXPATHLEN];
91
92 /*
93 * Assume file doesn't exist, so just try to create it, most times this
94 * works. We have to take special handling when the file does exist. To
95 * detect this, we use O_EXCL. For example when trying to create a
96 * file and a character device or fifo exists with the same name, we
97 * can accidently open the device by mistake (or block waiting to open).
98 * If we find that the open has failed, then spend the effort to
99 * figure out why. This strategy was found to have better average
100 * performance in common use than checking the file (and the path)
101 * first with lstat.
102 */
103 file_mode = arcn->sb.st_mode & FILEBITS;
104 if ((fd = open(arcn->name, O_WRONLY | O_CREAT | O_TRUNC | O_EXCL,
105 file_mode)) >= 0)
106 return(fd);
107
108 /*
109 * the file seems to exist. First we try to get rid of it (found to be
110 * the second most common failure when traced). If this fails, only
111 * then we go to the expense to check and create the path to the file
112 */
113 if (unlnk_exist(arcn->name, arcn->type) != 0)
114 return(-1);
115
116 path_to_open = arcn->name;
117 new_path = arcn->name;
118 cwd = getcwd(cwd_buff,sizeof(cwd_buff));
119 if (cwd==NULL) return -1;
120 for (;;) {
121 /*
122 * try to open it again, if this fails, check all the nodes in
123 * the path and give it a final try. if chk_path() finds that
124 * it cannot fix anything, we will skip the last attempt
125 */
126 if ((fd = open(path_to_open, O_WRONLY | O_CREAT | O_TRUNC,
127 file_mode)) >= 0) {
128 /* clean up the invalid_action */
129 if (pax_invalid_action>0) {
130 record_pax_invalid_action_results(arcn, path_to_open);
131 }
132 break;
133 }
134 oerrno = errno;
135 if (pax_invalid_action>0) {
136 rc = perform_pax_invalid_action(arcn, oerrno);
137 if (rc == 0) continue;
138 if (rc == 1) {
139 fd = -1;
140 break;
141 }
142 }
143 /* rc == 2 reserved for -o invalid_action=write */
144 if (nodirs || chk_path(path_to_open,arcn->sb.st_uid,arcn->sb.st_gid,
145 (rc==2) ? &new_path: NULL) < 0) {
146 syswarn((pax_invalid_action==0), oerrno, "Unable to create %s", arcn->name);
147 fd = -1;
148 break;
149 }
150 if (new_path) path_to_open = new_path; /* try again */
151 }
152 if (new_path && strcmp(new_path, arcn->name)!=0) {
153 dochdir(cwd); /* go back to original directory */
154 }
155 return(fd);
156 }
157
158 /*
159 * file_close()
160 * Close file descriptor to a file just created by pax. Sets modes,
161 * ownership and times as required.
162 * Return:
163 * 0 for success, -1 for failure
164 */
165
166 void
167 file_close(ARCHD *arcn, int fd)
168 {
169 int res = 0;
170
171 if (fd < 0)
172 return;
173
174 if (close(fd) < 0)
175 syswarn(0, errno, "Unable to close file descriptor on %s",
176 arcn->name);
177
178 /*
179 * set owner/groups first as this may strip off mode bits we want
180 * then set file permission modes. Then set file access and
181 * modification times.
182 */
183 if (pids)
184 res = set_ids(arcn->name, arcn->sb.st_uid,
185 arcn->sb.st_gid);
186 else
187 res = 1; /* without pids, pax should NOT set s bits */
188
189 /*
190 * IMPORTANT SECURITY NOTE:
191 * if not preserving mode or we cannot set uid/gid, then PROHIBIT
192 * set uid/gid bits
193 */
194 if (!pmode || res)
195 arcn->sb.st_mode &= ~(SETBITS);
196 if (pmode)
197 set_pmode(arcn->name, arcn->sb.st_mode);
198 if (patime || pmtime)
199 set_ftime(arcn->name, arcn->sb.st_mtime, arcn->sb.st_atime, 0);
200 }
201
202 /*
203 * lnk_creat()
204 * Create a hard link to arcn->ln_name from arcn->name. arcn->ln_name
205 * must exist;
206 * Return:
207 * 0 if ok, -1 otherwise
208 */
209
210 int
211 lnk_creat(ARCHD *arcn)
212 {
213 struct stat sb;
214
215 /*
216 * we may be running as root, so we have to be sure that link target
217 * is not a directory, so we lstat and check
218 */
219 if (lstat(arcn->ln_name, &sb) < 0) {
220 syswarn(1,errno,"Unable to link to %s from %s", arcn->ln_name,
221 arcn->name);
222 return(-1);
223 }
224
225 if (S_ISDIR(sb.st_mode)) {
226 paxwarn(1, "A hard link to the directory %s is not allowed",
227 arcn->ln_name);
228 return(-1);
229 }
230
231 if (S_ISLNK(sb.st_mode)) {
232 int res;
233 char buff[PATH_MAX+1];
234 /*
235 * Conformance: cannot make hard link to symlink - just make a
236 * symlink to the target of the symlink
237 */
238 if ((res = readlink(arcn->ln_name, buff, sizeof(buff)-1)) < 0) {
239 syswarn(1,errno,"Unable to symlink to %s from %s", arcn->ln_name,
240 arcn->name);
241 return(-1);
242 }
243 buff[res] = 0;
244 res = symlink(buff, arcn->name);
245 return res;
246 }
247
248 return(mk_link(arcn->ln_name, &sb, arcn->name, 0));
249 }
250
251 /*
252 * cross_lnk()
253 * Create a hard link to arcn->org_name from arcn->name. Only used in copy
254 * with the -l flag. No warning or error if this does not succeed (we will
255 * then just create the file)
256 * Return:
257 * 1 if copy() should try to create this file node
258 * 0 if cross_lnk() ok, -1 for fatal flaw (like linking to self).
259 */
260
261 int
262 cross_lnk(ARCHD *arcn)
263 {
264 /*
265 * try to make a link to original file (-l flag in copy mode). make
266 * sure we do not try to link to directories in case we are running as
267 * root (and it might succeed).
268 */
269 if (arcn->type == PAX_DIR)
270 return(1);
271 if (arcn->type == PAX_SLK) { /* for Unix 03 conformance tests 202,203 */
272 if (!Lflag)
273 return(1);
274 }
275 return(mk_link(arcn->org_name, &(arcn->sb), arcn->name, 1));
276 }
277
278 /*
279 * chk_same()
280 * In copy mode if we are not trying to make hard links between the src
281 * and destinations, make sure we are not going to overwrite ourselves by
282 * accident. This slows things down a little, but we have to protect all
283 * those people who make typing errors.
284 * Return:
285 * 1 the target does not exist, go ahead and copy
286 * 0 skip it file exists (-k) or may be the same as source file
287 */
288
289 int
290 chk_same(ARCHD *arcn)
291 {
292 struct stat sb;
293
294 /*
295 * if file does not exist, return. if file exists and -k, skip it
296 * quietly
297 */
298 if (lstat(arcn->name, &sb) < 0)
299 return(1);
300 if (kflag)
301 return(0);
302
303 /*
304 * better make sure the user does not have src == dest by mistake
305 */
306 if ((arcn->sb.st_dev == sb.st_dev) && (arcn->sb.st_ino == sb.st_ino)) {
307 paxwarn(1, "Unable to copy %s, file would overwrite itself",
308 arcn->name);
309 return(0);
310 }
311 return(1);
312 }
313
314 /*
315 * mk_link()
316 * try to make a hard link between two files. if ign set, we do not
317 * complain.
318 * Return:
319 * 0 if successful (or we are done with this file but no error, such as
320 * finding the from file exists and the user has set -k).
321 * 1 when ign was set to indicates we could not make the link but we
322 * should try to copy/extract the file as that might work (and is an
323 * allowed option). -1 an error occurred.
324 */
325
326 static int
327 mk_link(char *to, struct stat *to_sb, char *from, int ign)
328 {
329 struct stat sb;
330 int oerrno;
331
332 /*
333 * if from file exists, it has to be unlinked to make the link. If the
334 * file exists and -k is set, skip it quietly
335 */
336 if (lstat(from, &sb) == 0) {
337 if (kflag)
338 return(0);
339
340 /*
341 * make sure it is not the same file, protect the user
342 */
343 if ((to_sb->st_dev==sb.st_dev)&&(to_sb->st_ino == sb.st_ino)) {
344 paxwarn(1, "Unable to link file %s to itself", to);
345 return(-1);
346 }
347
348 /*
349 * try to get rid of the file, based on the type
350 */
351 if (S_ISDIR(sb.st_mode)) {
352 if (rmdir(from) < 0) {
353 syswarn(1, errno, "Unable to remove %s", from);
354 return(-1);
355 }
356 } else if (unlink(from) < 0) {
357 if (!ign) {
358 syswarn(1, errno, "Unable to remove %s", from);
359 return(-1);
360 }
361 return(1);
362 }
363 }
364
365 /*
366 * from file is gone (or did not exist), try to make the hard link.
367 * if it fails, check the path and try it again (if chk_path() says to
368 * try again)
369 */
370 for (;;) {
371 if (link(to, from) == 0)
372 break;
373 oerrno = errno;
374 if (!nodirs && chk_path(from, to_sb->st_uid, to_sb->st_gid, NULL) == 0)
375 continue;
376 if (!ign) {
377 syswarn(1, oerrno, "Could not link to %s from %s", to,
378 from);
379 return(-1);
380 }
381 return(1);
382 }
383
384 /*
385 * all right the link was made
386 */
387 return(0);
388 }
389
390 /*
391 * node_creat()
392 * create an entry in the file system (other than a file or hard link).
393 * If successful, sets uid/gid modes and times as required.
394 * Return:
395 * 0 if ok, -1 otherwise
396 */
397
398 int
399 node_creat(ARCHD *arcn)
400 {
401 int res;
402 int ign = 0;
403 int oerrno;
404 int pass = 0;
405 mode_t file_mode;
406 struct stat sb;
407 char target[MAXPATHLEN];
408 char *nm = arcn->name;
409 int nmlen = arcn->nlen;
410 int len;
411
412 /*
413 * create node based on type, if that fails try to unlink the node and
414 * try again. finally check the path and try again. As noted in the
415 * file and link creation routines, this method seems to exhibit the
416 * best performance in general use workloads.
417 */
418 file_mode = arcn->sb.st_mode & FILEBITS;
419
420 for (;;) {
421 switch (arcn->type) {
422 case PAX_DIR:
423 /*
424 * If -h (or -L) was given in tar-mode, follow the
425 * potential symlink chain before trying to create the
426 * directory.
427 */
428 if (strcmp(NM_TAR, argv0) == 0 && Lflag) {
429 while (lstat(nm, &sb) == 0 &&
430 S_ISLNK(sb.st_mode)) {
431 len = readlink(nm, target,
432 sizeof target - 1);
433 if (len == -1) {
434 syswarn(0, errno,
435 "cannot follow symlink %s in chain for %s",
436 nm, arcn->name);
437 res = -1;
438 goto badlink;
439 }
440 target[len] = '\0';
441 nm = target;
442 nmlen = len;
443 }
444 }
445 res = mkdir(nm, file_mode);
446
447 badlink:
448 if (ign)
449 res = 0;
450 break;
451 case PAX_CHR:
452 file_mode |= S_IFCHR;
453 res = mknod(nm, file_mode, arcn->sb.st_rdev);
454 break;
455 case PAX_BLK:
456 file_mode |= S_IFBLK;
457 res = mknod(nm, file_mode, arcn->sb.st_rdev);
458 break;
459 case PAX_FIF:
460 res = mkfifo(nm, file_mode);
461 break;
462 case PAX_SCK:
463 /*
464 * Skip sockets, operation has no meaning under BSD
465 */
466 paxwarn(0,
467 "%s skipped. Sockets cannot be copied or extracted",
468 nm);
469 return(-1);
470 case PAX_SLK:
471 res = symlink(arcn->ln_name, nm);
472 break;
473 case PAX_CTG:
474 case PAX_HLK:
475 case PAX_HRG:
476 case PAX_REG:
477 default:
478 /*
479 * we should never get here
480 */
481 paxwarn(0, "%s has an unknown file type, skipping",
482 nm);
483 return(-1);
484 }
485
486 /*
487 * if we were able to create the node break out of the loop,
488 * otherwise try to unlink the node and try again. if that
489 * fails check the full path and try a final time.
490 */
491 if (res == 0)
492 break;
493
494 /*
495 * we failed to make the node
496 */
497 oerrno = errno;
498 if ((ign = unlnk_exist(nm, arcn->type)) < 0)
499 return(-1);
500
501 if (++pass <= 1)
502 continue;
503
504 if (nodirs || chk_path(nm,arcn->sb.st_uid,arcn->sb.st_gid, NULL) < 0) {
505 syswarn(1, oerrno, "Could not create: %s", nm);
506 return(-1);
507 }
508 }
509
510 /*
511 * we were able to create the node. set uid/gid, modes and times
512 */
513 if (pids)
514 res = ((arcn->type == PAX_SLK) ?
515 #if defined(__APPLE__)
516 /* Mac OS X doesn't have lchown, so don't bother */
517 0 :
518 #else
519 set_lids(nm, arcn->sb.st_uid, arcn->sb.st_gid) :
520 #endif
521 set_ids(nm, arcn->sb.st_uid, arcn->sb.st_gid));
522 else
523 res = 1; /* without pids, pax should NOT set s bits */
524
525 /*
526 * symlinks are done now.
527 */
528 if (arcn->type == PAX_SLK)
529 return(0);
530
531 /*
532 * IMPORTANT SECURITY NOTE:
533 * if not preserving mode or we cannot set uid/gid, then PROHIBIT any
534 * set uid/gid bits
535 */
536 if (!pmode || res)
537 arcn->sb.st_mode &= ~(SETBITS);
538 if (pmode)
539 set_pmode(nm, arcn->sb.st_mode);
540
541 if (arcn->type == PAX_DIR && strcmp(NM_CPIO, argv0) != 0) {
542 /*
543 * Dirs must be processed again at end of extract to set times
544 * and modes to agree with those stored in the archive. However
545 * to allow extract to continue, we may have to also set owner
546 * rights. This allows nodes in the archive that are children
547 * of this directory to be extracted without failure. Both time
548 * and modes will be fixed after the entire archive is read and
549 * before pax exits.
550 */
551 if (access(nm, R_OK | W_OK | X_OK) < 0) {
552 if (lstat(nm, &sb) < 0) {
553 syswarn(0, errno,"Could not access %s (stat)",
554 arcn->name);
555 set_pmode(nm,file_mode | S_IRWXU);
556 } else {
557 /*
558 * We have to add rights to the dir, so we make
559 * sure to restore the mode. The mode must be
560 * restored AS CREATED and not as stored if
561 * pmode is not set.
562 */
563 set_pmode(nm,
564 ((sb.st_mode & FILEBITS) | S_IRWXU));
565 if (!pmode)
566 arcn->sb.st_mode = sb.st_mode;
567 }
568
569 /*
570 * we have to force the mode to what was set here,
571 * since we changed it from the default as created.
572 */
573 add_dir(nm, nmlen, &(arcn->sb), 1);
574 } else if (pmode || patime || pmtime)
575 add_dir(nm, nmlen, &(arcn->sb), 0);
576 }
577
578 if (patime || pmtime)
579 set_ftime(nm, arcn->sb.st_mtime, arcn->sb.st_atime, 0);
580 return(0);
581 }
582
583 /*
584 * unlnk_exist()
585 * Remove node from file system with the specified name. We pass the type
586 * of the node that is going to replace it. When we try to create a
587 * directory and find that it already exists, we allow processing to
588 * continue as proper modes etc will always be set for it later on.
589 * Return:
590 * 0 is ok to proceed, no file with the specified name exists
591 * -1 we were unable to remove the node, or we should not remove it (-k)
592 * 1 we found a directory and we were going to create a directory.
593 */
594
595 int
596 unlnk_exist(char *name, int type)
597 {
598 struct stat sb;
599
600 /*
601 * the file does not exist, or -k we are done
602 */
603 if (lstat(name, &sb) < 0)
604 return(0);
605 if (kflag)
606 return(-1);
607
608 if(strstr(name, "._") != NULL) /* remove when stat works properly */
609 return(0);
610
611 if (S_ISDIR(sb.st_mode)) {
612 /*
613 * try to remove a directory, if it fails and we were going to
614 * create a directory anyway, tell the caller (return a 1)
615 */
616 if (rmdir(name) < 0) {
617 if (type == PAX_DIR)
618 return(1);
619 syswarn(1,errno,"Unable to remove directory %s", name);
620 return(-1);
621 }
622 return(0);
623 }
624
625 /*
626 * try to get rid of all non-directory type nodes
627 */
628 if (unlink(name) < 0) {
629 syswarn(1, errno, "Could not unlink %s", name);
630 return(-1);
631 }
632 return(0);
633 }
634
635 /*
636 * chk_path()
637 * We were trying to create some kind of node in the file system and it
638 * failed. chk_path() makes sure the path up to the node exists and is
639 * writeable. When we have to create a directory that is missing along the
640 * path somewhere, the directory we create will be set to the same
641 * uid/gid as the file has (when uid and gid are being preserved).
642 * NOTE: this routine is a real performance loss. It is only used as a
643 * last resort when trying to create entries in the file system.
644 * Return:
645 * -1 when it could find nothing it is allowed to fix.
646 * 0 otherwise
647 */
648
649 int
650 chk_path(char *name, uid_t st_uid, gid_t st_gid, char ** new_name)
651 {
652 char *spt = name;
653 int namelen = strlen(name);
654 struct stat sb;
655 int retval = -1;
656
657 /*
658 * watch out for paths with nodes stored directly in / (e.g. /bozo)
659 */
660 if (*spt == '/')
661 ++spt;
662
663 for (;;) {
664 /*
665 * work forward from the first / and check each part of the path
666 */
667 spt = strchr(spt, '/');
668 if (spt == NULL)
669 break;
670 *spt = '\0';
671
672 /*
673 * if it exists we assume it is a directory, it is not within
674 * the spec (at least it seems to read that way) to alter the
675 * file system for nodes NOT EXPLICITLY stored on the archive.
676 * If that assumption is changed, you would test the node here
677 * and figure out how to get rid of it (probably like some
678 * recursive unlink()) or fix up the directory permissions if
679 * required (do an access()).
680 */
681 if (lstat(name, &sb) == 0) {
682 *(spt++) = '/';
683 if (new_name==NULL) continue;
684 retval = 0; /* accept it one directory at a time */
685 break;
686 }
687
688 /*
689 * the path fails at this point, see if we can create the
690 * needed directory and continue on
691 */
692 if (mkdir(name, S_IRWXU | S_IRWXG | S_IRWXO) < 0) {
693 *spt = '/';
694 retval = -1;
695 break;
696 }
697
698 /*
699 * we were able to create the directory. We will tell the
700 * caller that we found something to fix, and it is ok to try
701 * and create the node again.
702 */
703 retval = 0;
704 if (pids)
705 (void)set_ids(name, st_uid, st_gid);
706
707 /*
708 * make sure the user doesn't have some strange umask that
709 * causes this newly created directory to be unusable. We fix
710 * the modes and restore them back to the creation default at
711 * the end of pax
712 */
713 if ((access(name, R_OK | W_OK | X_OK) < 0) &&
714 (lstat(name, &sb) == 0)) {
715 set_pmode(name, ((sb.st_mode & FILEBITS) | S_IRWXU));
716 add_dir(name, namelen, &sb, 1);
717 }
718 *(spt++) = '/';
719 if (new_name==NULL) continue;
720 break;
721 }
722 if ((new_name != NULL) && retval==0) {
723 /* save the new path */
724 *(--spt) = '\0';
725 /*
726 printf ("chdir to %s\n", name);
727 */
728 if(0==chdir(name)) {
729 *spt++ = '/';
730 /*
731 printf ("remaining path: %s\n",spt);
732 */
733 *new_name = spt;
734 } else
735 *spt = '/';
736 }
737 return(retval);
738 }
739
740 /*
741 * set_ftime()
742 * Set the access time and modification time for a named file. If frc
743 * is non-zero we force these times to be set even if the user did not
744 * request access and/or modification time preservation (this is also
745 * used by -t to reset access times).
746 * When frc is zero, only those times the user has asked for are set, the
747 * other ones are left alone. We do not assume the un-documented feature
748 * of many utimes() implementations that consider a 0 time value as a do
749 * not set request.
750 */
751
752 void
753 set_ftime(char *fnm, time_t mtime, time_t atime, int frc)
754 {
755 static struct timeval tv[2] = {{0L, 0L}, {0L, 0L}};
756 struct stat sb;
757
758 tv[0].tv_sec = (long)atime;
759 tv[1].tv_sec = (long)mtime;
760 if (!frc && (!patime || !pmtime)) {
761 /*
762 * if we are not forcing, only set those times the user wants
763 * set. We get the current values of the times if we need them.
764 */
765 if (lstat(fnm, &sb) == 0) {
766 if (!patime)
767 tv[0].tv_sec = (long)sb.st_atime;
768 if (!pmtime)
769 tv[1].tv_sec = (long)sb.st_mtime;
770 } else
771 syswarn(0,errno,"Unable to obtain file stats %s", fnm);
772 }
773
774 /*
775 * set the times
776 */
777 if (pax_invalid_action_write_cwd) {
778 char cwd_buff[MAXPATHLEN];
779 char * cwd;
780 cwd = getcwd(&cwd_buff[0],MAXPATHLEN);
781 chdir(pax_invalid_action_write_cwd);
782 if (utimes(pax_invalid_action_write_path, tv) < 0)
783 syswarn(1, errno, "Access/modification time set failed on: %s",
784 pax_invalid_action_write_path);
785 chdir(cwd);
786 cleanup_pax_invalid_action();
787 } else {
788 if (utimes(fnm, tv) < 0)
789 syswarn(1, errno, "Access/modification time set failed on: %s",
790 fnm);
791 }
792 return;
793 }
794
795 void
796 fset_ftime(char *fnm, int fd, time_t mtime, time_t atime, int frc)
797 {
798 static struct timeval tv[2] = {{0L, 0L}, {0L, 0L}};
799 struct stat sb;
800
801 tv[0].tv_sec = (long)atime;
802 tv[1].tv_sec = (long)mtime;
803 if (!frc && (!patime || !pmtime)) {
804 /*
805 * if we are not forcing, only set those times the user wants
806 * set. We get the current values of the times if we need them.
807 */
808 if (fstat(fd, &sb) == 0) {
809 if (!patime)
810 tv[0].tv_sec = (long)sb.st_atime;
811 if (!pmtime)
812 tv[1].tv_sec = (long)sb.st_mtime;
813 } else
814 syswarn(0,errno,"Unable to obtain file stats %s", fnm);
815 }
816 /*
817 * set the times
818 */
819 if (futimes(fd, tv) < 0)
820 syswarn(1, errno, "Access/modification time set failed on: %s",
821 fnm);
822 return;
823 }
824
825 /*
826 * set_ids()
827 * set the uid and gid of a file system node
828 * Return:
829 * 0 when set, -1 on failure
830 */
831
832 int
833 set_ids(char *fnm, uid_t uid, gid_t gid)
834 {
835 if (chown(fnm, uid, gid) < 0) {
836 /*
837 * ignore EPERM unless in verbose mode or being run by root.
838 * if running as pax, POSIX requires a warning.
839 */
840 if (strcmp(NM_PAX, argv0) == 0 || errno != EPERM || vflag ||
841 geteuid() == 0)
842 syswarn(1, errno, "Unable to set file uid/gid of %s",
843 fnm);
844 return(-1);
845 }
846 return(0);
847 }
848
849 int
850 fset_ids(char *fnm, int fd, uid_t uid, gid_t gid)
851 {
852 if (fchown(fd, uid, gid) < 0) {
853 /*
854 * ignore EPERM unless in verbose mode or being run by root.
855 * if running as pax, POSIX requires a warning.
856 */
857 if (strcmp(NM_PAX, argv0) == 0 || errno != EPERM || vflag ||
858 geteuid() == 0)
859 syswarn(1, errno, "Unable to set file uid/gid of %s",
860 fnm);
861 return(-1);
862 }
863 return(0);
864 }
865
866 #if !defined(__APPLE__)
867 /*
868 * set_lids()
869 * set the uid and gid of a file system node
870 * Return:
871 * 0 when set, -1 on failure
872 */
873
874 int
875 set_lids(char *fnm, uid_t uid, gid_t gid)
876 {
877 if (lchown(fnm, uid, gid) < 0) {
878 /*
879 * ignore EPERM unless in verbose mode or being run by root.
880 * if running as pax, POSIX requires a warning.
881 */
882 if (strcmp(NM_PAX, argv0) == 0 || errno != EPERM || vflag ||
883 geteuid() == 0)
884 syswarn(1, errno, "Unable to set file uid/gid of %s",
885 fnm);
886 return(-1);
887 }
888 return(0);
889 }
890 #endif /* !__APPLE__ */
891
892 /*
893 * set_pmode()
894 * Set file access mode
895 */
896
897 void
898 set_pmode(char *fnm, mode_t mode)
899 {
900 mode &= ABITS;
901 if (chmod(fnm, mode) < 0)
902 syswarn(1, errno, "Could not set permissions on %s", fnm);
903 return;
904 }
905
906 void
907 fset_pmode(char *fnm, int fd, mode_t mode)
908 {
909 mode &= ABITS;
910 if (fchmod(fd, mode) < 0)
911 syswarn(1, errno, "Could not set permissions on %s", fnm);
912 return;
913 }
914
915 /*
916 * file_write()
917 * Write/copy a file (during copy or archive extract). This routine knows
918 * how to copy files with lseek holes in it. (Which are read as file
919 * blocks containing all 0's but do not have any file blocks associated
920 * with the data). Typical examples of these are files created by dbm
921 * variants (.pag files). While the file size of these files are huge, the
922 * actual storage is quite small (the files are sparse). The problem is
923 * the holes read as all zeros so are probably stored on the archive that
924 * way (there is no way to determine if the file block is really a hole,
925 * we only know that a file block of all zero's can be a hole).
926 * At this writing, no major archive format knows how to archive files
927 * with holes. However, on extraction (or during copy, -rw) we have to
928 * deal with these files. Without detecting the holes, the files can
929 * consume a lot of file space if just written to disk. This replacement
930 * for write when passed the basic allocation size of a file system block,
931 * uses lseek whenever it detects the input data is all 0 within that
932 * file block. In more detail, the strategy is as follows:
933 * While the input is all zero keep doing an lseek. Keep track of when we
934 * pass over file block boundaries. Only write when we hit a non zero
935 * input. once we have written a file block, we continue to write it to
936 * the end (we stop looking at the input). When we reach the start of the
937 * next file block, start checking for zero blocks again. Working on file
938 * block boundaries significantly reduces the overhead when copying files
939 * that are NOT very sparse. This overhead (when compared to a write) is
940 * almost below the measurement resolution on many systems. Without it,
941 * files with holes cannot be safely copied. It does has a side effect as
942 * it can put holes into files that did not have them before, but that is
943 * not a problem since the file contents are unchanged (in fact it saves
944 * file space). (Except on paging files for diskless clients. But since we
945 * cannot determine one of those file from here, we ignore them). If this
946 * ever ends up on a system where CTG files are supported and the holes
947 * are not desired, just do a conditional test in those routines that
948 * call file_write() and have it call write() instead. BEFORE CLOSING THE
949 * FILE, make sure to call file_flush() when the last write finishes with
950 * an empty block. A lot of file systems will not create an lseek hole at
951 * the end. In this case we drop a single 0 at the end to force the
952 * trailing 0's in the file.
953 * ---Parameters---
954 * rem: how many bytes left in this file system block
955 * isempt: have we written to the file block yet (is it empty)
956 * sz: basic file block allocation size
957 * cnt: number of bytes on this write
958 * str: buffer to write
959 * Return:
960 * number of bytes written, -1 on write (or lseek) error.
961 */
962
963 int
964 file_write(int fd, char *str, int cnt, int *rem, int *isempt, int sz,
965 char *name)
966 {
967 char *pt;
968 char *end;
969 int wcnt;
970 char *st = str;
971 char **strp;
972
973 /*
974 * while we have data to process
975 */
976 while (cnt) {
977 if (!*rem) {
978 /*
979 * We are now at the start of file system block again
980 * (or what we think one is...). start looking for
981 * empty blocks again
982 */
983 *isempt = 1;
984 *rem = sz;
985 }
986
987 /*
988 * only examine up to the end of the current file block or
989 * remaining characters to write, whatever is smaller
990 */
991 wcnt = MIN(cnt, *rem);
992 cnt -= wcnt;
993 *rem -= wcnt;
994 if (*isempt) {
995 /*
996 * have not written to this block yet, so we keep
997 * looking for zero's
998 */
999 pt = st;
1000 end = st + wcnt;
1001
1002 /*
1003 * look for a zero filled buffer
1004 */
1005 while ((pt < end) && (*pt == '\0'))
1006 ++pt;
1007
1008 if (pt == end) {
1009 /*
1010 * skip, buf is empty so far
1011 */
1012 if (fd > -1 &&
1013 lseek(fd, (off_t)wcnt, SEEK_CUR) < 0) {
1014 syswarn(1,errno,"File seek on %s",
1015 name);
1016 return(-1);
1017 }
1018 st = pt;
1019 continue;
1020 }
1021 /*
1022 * drat, the buf is not zero filled
1023 */
1024 *isempt = 0;
1025 }
1026
1027 /*
1028 * have non-zero data in this file system block, have to write
1029 */
1030 switch (fd) {
1031 case -1:
1032 strp = &gnu_name_string;
1033 break;
1034 case -2:
1035 strp = &gnu_link_string;
1036 break;
1037 default:
1038 strp = NULL;
1039 break;
1040 }
1041 if (strp) {
1042 if (*strp)
1043 err(1, "WARNING! Major Internal Error! GNU hack Failing!");
1044 *strp = malloc(wcnt + 1);
1045 if (*strp == NULL) {
1046 paxwarn(1, "Out of memory");
1047 return(-1);
1048 }
1049 memcpy(*strp, st, wcnt);
1050 (*strp)[wcnt] = '\0';
1051 break;
1052 } else if (write(fd, st, wcnt) != wcnt) {
1053 syswarn(1, errno, "Failed write to file %s", name);
1054 return(-1);
1055 }
1056 st += wcnt;
1057 }
1058 return(st - str);
1059 }
1060
1061 /*
1062 * file_flush()
1063 * when the last file block in a file is zero, many file systems will not
1064 * let us create a hole at the end. To get the last block with zeros, we
1065 * write the last BYTE with a zero (back up one byte and write a zero).
1066 */
1067
1068 void
1069 file_flush(int fd, char *fname, int isempt)
1070 {
1071 static char blnk[] = "\0";
1072
1073 /*
1074 * silly test, but make sure we are only called when the last block is
1075 * filled with all zeros.
1076 */
1077 if (!isempt)
1078 return;
1079
1080 /*
1081 * move back one byte and write a zero
1082 */
1083 if (lseek(fd, (off_t)-1, SEEK_CUR) < 0) {
1084 syswarn(1, errno, "Failed seek on file %s", fname);
1085 return;
1086 }
1087
1088 if (write(fd, blnk, 1) < 0)
1089 syswarn(1, errno, "Failed write to file %s", fname);
1090 return;
1091 }
1092
1093 /*
1094 * rdfile_close()
1095 * close a file we have beed reading (to copy or archive). If we have to
1096 * reset access time (tflag) do so (the times are stored in arcn).
1097 */
1098
1099 void
1100 rdfile_close(ARCHD *arcn, int *fd)
1101 {
1102 /*
1103 * make sure the file is open
1104 */
1105 if (*fd < 0)
1106 return;
1107
1108 (void)close(*fd);
1109 *fd = -1;
1110 if (!tflag)
1111 return;
1112
1113 /*
1114 * user wants last access time reset
1115 */
1116 set_ftime(arcn->org_name, arcn->sb.st_mtime, arcn->sb.st_atime, 1);
1117 return;
1118 }
1119
1120 /*
1121 * set_crc()
1122 * read a file to calculate its crc. This is a real drag. Archive formats
1123 * that have this, end up reading the file twice (we have to write the
1124 * header WITH the crc before writing the file contents. Oh well...
1125 * Return:
1126 * 0 if was able to calculate the crc, -1 otherwise
1127 */
1128
1129 int
1130 set_crc(ARCHD *arcn, int fd)
1131 {
1132 int i;
1133 int res;
1134 off_t cpcnt = 0L;
1135 u_long size;
1136 u_int32_t crc = 0;
1137 char tbuf[FILEBLK];
1138 struct stat sb;
1139
1140 if (fd < 0) {
1141 /*
1142 * hmm, no fd, should never happen. well no crc then.
1143 */
1144 arcn->crc = 0L;
1145 return(0);
1146 }
1147
1148 if ((size = (u_long)arcn->sb.st_blksize) > (u_long)sizeof(tbuf))
1149 size = (u_long)sizeof(tbuf);
1150
1151 /*
1152 * read all the bytes we think that there are in the file. If the user
1153 * is trying to archive an active file, forget this file.
1154 */
1155 for (;;) {
1156 if ((res = read(fd, tbuf, size)) <= 0)
1157 break;
1158 cpcnt += res;
1159 for (i = 0; i < res; ++i)
1160 crc += (tbuf[i] & 0xff);
1161 }
1162
1163 /*
1164 * safety check. we want to avoid archiving files that are active as
1165 * they can create inconsistent archive copies.
1166 */
1167 if (cpcnt != arcn->sb.st_size)
1168 paxwarn(1, "File changed size %s", arcn->org_name);
1169 else if (fstat(fd, &sb) < 0)
1170 syswarn(1, errno, "Failed stat on %s", arcn->org_name);
1171 else if (arcn->sb.st_mtime != sb.st_mtime)
1172 paxwarn(1, "File %s was modified during read", arcn->org_name);
1173 else if (lseek(fd, (off_t)0L, SEEK_SET) < 0)
1174 syswarn(1, errno, "File rewind failed on: %s", arcn->org_name);
1175 else {
1176 arcn->crc = crc;
1177 return(0);
1178 }
1179 return(-1);
1180 }