file_subs.c revision 1.32 1 /* $NetBSD: file_subs.c,v 1.32 2003/02/10 07:49:57 grant 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. All advertising materials mentioning features or use of this software
20 * must display the following acknowledgement:
21 * This product includes software developed by the University of
22 * California, Berkeley and its contributors.
23 * 4. Neither the name of the University nor the names of its contributors
24 * may be used to endorse or promote products derived from this software
25 * without specific prior written permission.
26 *
27 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
28 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
29 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
30 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
31 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
32 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
33 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
34 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
35 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
36 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
37 * SUCH DAMAGE.
38 */
39
40 #include <sys/cdefs.h>
41 #if defined(__RCSID) && !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 1.32 2003/02/10 07:49:57 grant 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 static int
66 mk_link(char *,struct stat *,char *, int);
67
68 /*
69 * routines that deal with file operations such as: creating, removing;
70 * and setting access modes, uid/gid and times of files
71 */
72
73 #define FILEBITS (S_ISVTX | S_IRWXU | S_IRWXG | S_IRWXO)
74 #define SETBITS (S_ISUID | S_ISGID)
75 #define ABITS (FILEBITS | SETBITS)
76
77 /*
78 * file_creat()
79 * Create and open a file.
80 * Return:
81 * file descriptor or -1 for failure
82 */
83
84 int
85 file_creat(ARCHD *arcn)
86 {
87 int fd = -1;
88 mode_t file_mode;
89 int oerrno;
90
91 /*
92 * assume file doesn't exist, so just try to create it, most times this
93 * works. We have to take special handling when the file does exist. To
94 * detect this, we use O_EXCL. For example when trying to create a
95 * file and a character device or fifo exists with the same name, we
96 * can accidently open the device by mistake (or block waiting to open)
97 * If we find that the open has failed, then spend the effort to
98 * figure out why. This strategy was found to have better average
99 * performance in common use than checking the file (and the path)
100 * first with lstat.
101 */
102 file_mode = arcn->sb.st_mode & FILEBITS;
103 if ((fd = open(arcn->name, O_WRONLY | O_CREAT | O_TRUNC | O_EXCL,
104 file_mode)) >= 0)
105 return(fd);
106
107 /*
108 * the file seems to exist. First we try to get rid of it (found to be
109 * the second most common failure when traced). If this fails, only
110 * then we go to the expense to check and create the path to the file
111 */
112 if (unlnk_exist(arcn->name, arcn->type) != 0)
113 return(-1);
114
115 for (;;) {
116 /*
117 * try to open it again, if this fails, check all the nodes in
118 * the path and give it a final try. if chk_path() finds that
119 * it cannot fix anything, we will skip the last attempt
120 */
121 if ((fd = open(arcn->name, O_WRONLY | O_CREAT | O_TRUNC,
122 file_mode)) >= 0)
123 break;
124 oerrno = errno;
125 if (nodirs || chk_path(arcn->name,arcn->sb.st_uid,arcn->sb.st_gid) < 0) {
126 (void)fflush(listf);
127 syswarn(1, oerrno, "Cannot create %s", arcn->name);
128 return(-1);
129 }
130 }
131 return(fd);
132 }
133
134 /*
135 * file_close()
136 * Close file descriptor to a file just created by pax. Sets modes,
137 * ownership and times as required.
138 * Return:
139 * 0 for success, -1 for failure
140 */
141
142 void
143 file_close(ARCHD *arcn, int fd)
144 {
145 int res = 0;
146
147 if (fd < 0)
148 return;
149 if (close(fd) < 0)
150 syswarn(0, errno, "Cannot close file descriptor on %s",
151 arcn->name);
152
153 /*
154 * set owner/groups first as this may strip off mode bits we want
155 * then set file permission modes. Then set file access and
156 * modification times.
157 */
158 if (pids)
159 res = set_ids(arcn->name, arcn->sb.st_uid, arcn->sb.st_gid);
160
161 /*
162 * IMPORTANT SECURITY NOTE:
163 * if not preserving mode or we cannot set uid/gid, then PROHIBIT
164 * set uid/gid bits
165 */
166 if (!pmode || res)
167 arcn->sb.st_mode &= ~(SETBITS);
168 if (pmode)
169 set_pmode(arcn->name, arcn->sb.st_mode);
170 if (patime || pmtime)
171 set_ftime(arcn->name, arcn->sb.st_mtime, arcn->sb.st_atime, 0);
172 #if HAVE_STRUCT_STAT_ST_FLAGS
173 if (pfflags && arcn->type != PAX_SLK)
174 set_chflags(arcn->name, arcn->sb.st_flags);
175 #endif
176 }
177
178 /*
179 * lnk_creat()
180 * Create a hard link to arcn->ln_name from arcn->name. arcn->ln_name
181 * must exist;
182 * Return:
183 * 0 if ok, -1 otherwise
184 */
185
186 int
187 lnk_creat(ARCHD *arcn)
188 {
189 struct stat sb;
190
191 /*
192 * we may be running as root, so we have to be sure that link target
193 * is not a directory, so we lstat and check
194 */
195 if (lstat(arcn->ln_name, &sb) < 0) {
196 syswarn(1, errno, "Cannot link to %s from %s", arcn->ln_name,
197 arcn->name);
198 return(-1);
199 }
200
201 if (S_ISDIR(sb.st_mode)) {
202 tty_warn(1, "A hard link to the directory %s is not allowed",
203 arcn->ln_name);
204 return(-1);
205 }
206
207 return(mk_link(arcn->ln_name, &sb, arcn->name, 0));
208 }
209
210 /*
211 * cross_lnk()
212 * Create a hard link to arcn->org_name from arcn->name. Only used in copy
213 * with the -l flag. No warning or error if this does not succeed (we will
214 * then just create the file)
215 * Return:
216 * 1 if copy() should try to create this file node
217 * 0 if cross_lnk() ok, -1 for fatal flaw (like linking to self).
218 */
219
220 int
221 cross_lnk(ARCHD *arcn)
222 {
223 /*
224 * try to make a link to original file (-l flag in copy mode). make
225 * sure we do not try to link to directories in case we are running as
226 * root (and it might succeed).
227 */
228 if (arcn->type == PAX_DIR)
229 return(1);
230 return(mk_link(arcn->org_name, &(arcn->sb), arcn->name, 1));
231 }
232
233 /*
234 * chk_same()
235 * In copy mode if we are not trying to make hard links between the src
236 * and destinations, make sure we are not going to overwrite ourselves by
237 * accident. This slows things down a little, but we have to protect all
238 * those people who make typing errors.
239 * Return:
240 * 1 the target does not exist, go ahead and copy
241 * 0 skip it file exists (-k) or may be the same as source file
242 */
243
244 int
245 chk_same(ARCHD *arcn)
246 {
247 struct stat sb;
248
249 /*
250 * if file does not exist, return. if file exists and -k, skip it
251 * quietly
252 */
253 if (lstat(arcn->name, &sb) < 0)
254 return(1);
255 if (kflag)
256 return(0);
257
258 /*
259 * better make sure the user does not have src == dest by mistake
260 */
261 if ((arcn->sb.st_dev == sb.st_dev) && (arcn->sb.st_ino == sb.st_ino)) {
262 tty_warn(1, "Unable to copy %s, file would overwrite itself",
263 arcn->name);
264 return(0);
265 }
266 return(1);
267 }
268
269 /*
270 * mk_link()
271 * try to make a hard link between two files. if ign set, we do not
272 * complain.
273 * Return:
274 * 0 if successful (or we are done with this file but no error, such as
275 * finding the from file exists and the user has set -k).
276 * 1 when ign was set to indicates we could not make the link but we
277 * should try to copy/extract the file as that might work (and is an
278 * allowed option). -1 an error occurred.
279 */
280
281 static int
282 mk_link(char *to, struct stat *to_sb, char *from, int ign)
283 {
284 struct stat sb;
285 int oerrno;
286
287 /*
288 * if from file exists, it has to be unlinked to make the link. If the
289 * file exists and -k is set, skip it quietly
290 */
291 if (lstat(from, &sb) == 0) {
292 if (kflag)
293 return(0);
294
295 /*
296 * make sure it is not the same file, protect the user
297 */
298 if ((to_sb->st_dev==sb.st_dev)&&(to_sb->st_ino == sb.st_ino)) {
299 tty_warn(1, "Cannot link file %s to itself", to);
300 return(-1);
301 }
302
303 /*
304 * try to get rid of the file, based on the type
305 */
306 if (S_ISDIR(sb.st_mode)) {
307 if (rmdir(from) < 0) {
308 syswarn(1, errno, "Cannot remove %s", from);
309 return(-1);
310 }
311 } else if (unlink(from) < 0) {
312 if (!ign) {
313 syswarn(1, errno, "Cannot remove %s", from);
314 return(-1);
315 }
316 return(1);
317 }
318 }
319
320 /*
321 * from file is gone (or did not exist), try to make the hard link.
322 * if it fails, check the path and try it again (if chk_path() says to
323 * try again)
324 */
325 for (;;) {
326 if (link(to, from) == 0)
327 break;
328 oerrno = errno;
329 if (chk_path(from, to_sb->st_uid, to_sb->st_gid) == 0)
330 continue;
331 if (!ign) {
332 syswarn(1, oerrno, "Cannot link to %s from %s", to,
333 from);
334 return(-1);
335 }
336 return(1);
337 }
338
339 /*
340 * all right the link was made
341 */
342 return(0);
343 }
344
345 /*
346 * node_creat()
347 * create an entry in the file system (other than a file or hard link).
348 * If successful, sets uid/gid modes and times as required.
349 * Return:
350 * 0 if ok, -1 otherwise
351 */
352
353 int
354 node_creat(ARCHD *arcn)
355 {
356 int res;
357 int ign = 0;
358 int oerrno;
359 int pass = 0;
360 mode_t file_mode;
361 struct stat sb;
362 char target[MAXPATHLEN];
363 char *nm = arcn->name;
364 int len;
365
366 /*
367 * create node based on type, if that fails try to unlink the node and
368 * try again. finally check the path and try again. As noted in the
369 * file and link creation routines, this method seems to exhibit the
370 * best performance in general use workloads.
371 */
372 file_mode = arcn->sb.st_mode & FILEBITS;
373
374 for (;;) {
375 switch(arcn->type) {
376 case PAX_DIR:
377 /*
378 * If -h (or -L) was given in tar-mode, follow the
379 * potential symlink chain before trying to create the
380 * directory.
381 */
382 if (strcmp(NM_TAR, argv0) == 0 && Lflag) {
383 while (lstat(nm, &sb) == 0 &&
384 S_ISLNK(sb.st_mode)) {
385 len = readlink(nm, target,
386 sizeof target - 1);
387 if (len == -1) {
388 syswarn(0, errno,
389 "cannot follow symlink %s in chain for %s",
390 nm, arcn->name);
391 res = -1;
392 goto badlink;
393 }
394 target[len] = '\0';
395 nm = target;
396 }
397 }
398 res = mkdir(nm, file_mode);
399
400 badlink:
401 if (ign)
402 res = 0;
403 break;
404 case PAX_CHR:
405 file_mode |= S_IFCHR;
406 res = mknod(nm, file_mode, arcn->sb.st_rdev);
407 break;
408 case PAX_BLK:
409 file_mode |= S_IFBLK;
410 res = mknod(nm, file_mode, arcn->sb.st_rdev);
411 break;
412 case PAX_FIF:
413 res = mkfifo(nm, file_mode);
414 break;
415 case PAX_SCK:
416 /*
417 * Skip sockets, operation has no meaning under BSD
418 */
419 tty_warn(0,
420 "%s skipped. Sockets cannot be copied or extracted",
421 nm);
422 return(-1);
423 case PAX_SLK:
424 res = symlink(arcn->ln_name, nm);
425 break;
426 case PAX_CTG:
427 case PAX_HLK:
428 case PAX_HRG:
429 case PAX_REG:
430 default:
431 /*
432 * we should never get here
433 */
434 tty_warn(0, "%s has an unknown file type, skipping",
435 nm);
436 return(-1);
437 }
438
439 /*
440 * if we were able to create the node break out of the loop,
441 * otherwise try to unlink the node and try again. if that
442 * fails check the full path and try a final time.
443 */
444 if (res == 0)
445 break;
446
447 /*
448 * we failed to make the node
449 */
450 oerrno = errno;
451 if ((ign = unlnk_exist(nm, arcn->type)) < 0)
452 return(-1);
453
454 if (++pass <= 1)
455 continue;
456
457 if (nodirs || chk_path(nm,arcn->sb.st_uid,arcn->sb.st_gid) < 0) {
458 syswarn(1, oerrno, "Cannot create %s", nm);
459 return(-1);
460 }
461 }
462
463 /*
464 * we were able to create the node. set uid/gid, modes and times
465 */
466 #if HAVE_LCHOWN
467 if (pids)
468 #else
469 if (pids && arcn->type != PAX_SLK)
470 #endif
471 res = set_ids(nm, arcn->sb.st_uid, arcn->sb.st_gid);
472 else
473 res = 0;
474
475 /*
476 * IMPORTANT SECURITY NOTE:
477 * if not preserving mode or we cannot set uid/gid, then PROHIBIT any
478 * set uid/gid bits
479 */
480 if (!pmode || res)
481 arcn->sb.st_mode &= ~(SETBITS);
482 #if HAVE_LCHMOD
483 if (pmode)
484 #else
485 if (pmode && arcn->type != PAX_SLK)
486 #endif
487 set_pmode(arcn->name, arcn->sb.st_mode);
488
489 if (arcn->type == PAX_DIR && strcmp(NM_CPIO, argv0) != 0) {
490 /*
491 * Dirs must be processed again at end of extract to set times
492 * and modes to agree with those stored in the archive. However
493 * to allow extract to continue, we may have to also set owner
494 * rights. This allows nodes in the archive that are children
495 * of this directory to be extracted without failure. Both time
496 * and modes will be fixed after the entire archive is read and
497 * before pax exits.
498 */
499 if (access(nm, R_OK | W_OK | X_OK) < 0) {
500 if (lstat(nm, &sb) < 0) {
501 syswarn(0, errno,"Cannot access %s (stat)",
502 arcn->name);
503 set_pmode(nm,file_mode | S_IRWXU);
504 } else {
505 /*
506 * We have to add rights to the dir, so we make
507 * sure to restore the mode. The mode must be
508 * restored AS CREATED and not as stored if
509 * pmode is not set.
510 */
511 set_pmode(nm,
512 ((sb.st_mode & FILEBITS) | S_IRWXU));
513 if (!pmode)
514 arcn->sb.st_mode = sb.st_mode;
515 }
516
517 /*
518 * we have to force the mode to what was set here,
519 * since we changed it from the default as created.
520 */
521 add_dir(nm, arcn->nlen, &(arcn->sb), 1);
522 } else if (pmode || patime || pmtime)
523 add_dir(nm, arcn->nlen, &(arcn->sb), 0);
524 }
525
526 #if HAVE_LUTIMES
527 if (patime || pmtime)
528 #else
529 if ((patime || pmtime) && arcn->type != PAX_SLK)
530 #endif
531 set_ftime(arcn->name, arcn->sb.st_mtime, arcn->sb.st_atime, 0);
532
533 #if HAVE_STRUCT_STAT_ST_FLAGS
534 if (pfflags && arcn->type != PAX_SLK)
535 set_chflags(arcn->name, arcn->sb.st_flags);
536 #endif
537 return(0);
538 }
539
540 /*
541 * unlnk_exist()
542 * Remove node from file system with the specified name. We pass the type
543 * of the node that is going to replace it. When we try to create a
544 * directory and find that it already exists, we allow processing to
545 * continue as proper modes etc will always be set for it later on.
546 * Return:
547 * 0 is ok to proceed, no file with the specified name exists
548 * -1 we were unable to remove the node, or we should not remove it (-k)
549 * 1 we found a directory and we were going to create a directory.
550 */
551
552 int
553 unlnk_exist(char *name, int type)
554 {
555 struct stat sb;
556
557 /*
558 * the file does not exist, or -k we are done
559 */
560 if (lstat(name, &sb) < 0)
561 return(0);
562 if (kflag)
563 return(-1);
564
565 if (S_ISDIR(sb.st_mode)) {
566 /*
567 * try to remove a directory, if it fails and we were going to
568 * create a directory anyway, tell the caller (return a 1)
569 */
570 if (rmdir(name) < 0) {
571 if (type == PAX_DIR)
572 return(1);
573 syswarn(1, errno, "Cannot remove directory %s", name);
574 return(-1);
575 }
576 return(0);
577 }
578
579 /*
580 * try to get rid of all non-directory type nodes
581 */
582 if (unlink(name) < 0) {
583 (void)fflush(listf);
584 syswarn(1, errno, "Cannot unlink %s", name);
585 return(-1);
586 }
587 return(0);
588 }
589
590 /*
591 * chk_path()
592 * We were trying to create some kind of node in the file system and it
593 * failed. chk_path() makes sure the path up to the node exists and is
594 * writable. When we have to create a directory that is missing along the
595 * path somewhere, the directory we create will be set to the same
596 * uid/gid as the file has (when uid and gid are being preserved).
597 * NOTE: this routine is a real performance loss. It is only used as a
598 * last resort when trying to create entries in the file system.
599 * Return:
600 * -1 when it could find nothing it is allowed to fix.
601 * 0 otherwise
602 */
603
604 int
605 chk_path( char *name, uid_t st_uid, gid_t st_gid)
606 {
607 char *spt = name;
608 struct stat sb;
609 int retval = -1;
610
611 /*
612 * watch out for paths with nodes stored directly in / (e.g. /bozo)
613 */
614 if (*spt == '/')
615 ++spt;
616
617 for(;;) {
618 /*
619 * work forward from the first / and check each part of
620 * the path
621 */
622 spt = strchr(spt, '/');
623 if (spt == NULL)
624 break;
625 *spt = '\0';
626
627 /*
628 * if it exists we assume it is a directory, it is not within
629 * the spec (at least it seems to read that way) to alter the
630 * file system for nodes NOT EXPLICITLY stored on the archive.
631 * If that assumption is changed, you would test the node here
632 * and figure out how to get rid of it (probably like some
633 * recursive unlink()) or fix up the directory permissions if
634 * required (do an access()).
635 */
636 if (lstat(name, &sb) == 0) {
637 *(spt++) = '/';
638 continue;
639 }
640
641 /*
642 * the path fails at this point, see if we can create the
643 * needed directory and continue on
644 */
645 if (mkdir(name, S_IRWXU | S_IRWXG | S_IRWXO) < 0) {
646 *spt = '/';
647 retval = -1;
648 break;
649 }
650
651 /*
652 * we were able to create the directory. We will tell the
653 * caller that we found something to fix, and it is ok to try
654 * and create the node again.
655 */
656 retval = 0;
657 if (pids)
658 (void)set_ids(name, st_uid, st_gid);
659
660 /*
661 * make sure the user doesn't have some strange umask that
662 * causes this newly created directory to be unusable. We fix
663 * the modes and restore them back to the creation default at
664 * the end of pax
665 */
666 if ((access(name, R_OK | W_OK | X_OK) < 0) &&
667 (lstat(name, &sb) == 0)) {
668 set_pmode(name, ((sb.st_mode & FILEBITS) | S_IRWXU));
669 add_dir(name, spt - name, &sb, 1);
670 }
671 *(spt++) = '/';
672 continue;
673 }
674 return(retval);
675 }
676
677 /*
678 * set_ftime()
679 * Set the access time and modification time for a named file. If frc
680 * is non-zero we force these times to be set even if the user did not
681 * request access and/or modification time preservation (this is also
682 * used by -t to reset access times).
683 * When ign is zero, only those times the user has asked for are set, the
684 * other ones are left alone. We do not assume the un-documented feature
685 * of many utimes() implementations that consider a 0 time value as a do
686 * not set request.
687 */
688
689 void
690 set_ftime(char *fnm, time_t mtime, time_t atime, int frc)
691 {
692 struct timeval tv[2];
693 struct stat sb;
694
695 tv[0].tv_sec = (long)atime;
696 tv[0].tv_usec = 0;
697 tv[1].tv_sec = (long)mtime;
698 tv[1].tv_usec = 0;
699 if (!frc && (!patime || !pmtime)) {
700 /*
701 * if we are not forcing, only set those times the user wants
702 * set. We get the current values of the times if we need them.
703 */
704 if (lstat(fnm, &sb) == 0) {
705 #ifdef BSD4_4
706 if (!patime)
707 TIMESPEC_TO_TIMEVAL(&tv[0], &sb.st_atimespec);
708 if (!pmtime)
709 TIMESPEC_TO_TIMEVAL(&tv[1], &sb.st_mtimespec);
710 #else
711 if (!patime)
712 tv[0].tv_sec = sb.st_atime;
713 if (!pmtime)
714 tv[1].tv_sec = sb.st_mtime;
715 #endif
716 } else
717 syswarn(0, errno, "Cannot obtain file stats %s", fnm);
718 }
719
720 /*
721 * set the times
722 */
723 #if HAVE_LUTIMES
724 if (lutimes(fnm, tv))
725 #else
726 if (utimes(fnm, tv))
727 #endif
728 syswarn(1, errno, "Access/modification time set failed on: %s",
729 fnm);
730 return;
731 }
732
733 /*
734 * set_ids()
735 * set the uid and gid of a file system node
736 * Return:
737 * 0 when set, -1 on failure
738 */
739
740 int
741 set_ids(char *fnm, uid_t uid, gid_t gid)
742 {
743 if (geteuid() == 0)
744 #if HAVE_LCHOWN
745 if (lchown(fnm, uid, gid))
746 #else
747 if (chown(fnm, uid, gid))
748 #endif
749 {
750 (void)fflush(listf);
751 syswarn(1, errno, "Cannot set file uid/gid of %s",
752 fnm);
753 return(-1);
754 }
755 return(0);
756 }
757
758 /*
759 * set_pmode()
760 * Set file access mode
761 */
762
763 void
764 set_pmode(char *fnm, mode_t mode)
765 {
766 mode &= ABITS;
767 #if HAVE_LCHMOD
768 if (lchmod(fnm, mode))
769 #else
770 if (chmod(fnm, mode))
771 #endif
772 {
773 (void)fflush(listf);
774 syswarn(1, errno, "Cannot set permissions on %s", fnm);
775 }
776 return;
777 }
778
779 /*
780 * set_chflags()
781 * Set 4.4BSD file flags
782 */
783 void
784 set_chflags(char *fnm, u_int32_t flags)
785 {
786
787 #if 0
788 if (chflags(fnm, flags) < 0 && errno != EOPNOTSUPP)
789 syswarn(1, errno, "Cannot set file flags on %s", fnm);
790 #endif
791 return;
792 }
793
794 /*
795 * file_write()
796 * Write/copy a file (during copy or archive extract). This routine knows
797 * how to copy files with lseek holes in it. (Which are read as file
798 * blocks containing all 0's but do not have any file blocks associated
799 * with the data). Typical examples of these are files created by dbm
800 * variants (.pag files). While the file size of these files are huge, the
801 * actual storage is quite small (the files are sparse). The problem is
802 * the holes read as all zeros so are probably stored on the archive that
803 * way (there is no way to determine if the file block is really a hole,
804 * we only know that a file block of all zero's can be a hole).
805 * At this writing, no major archive format knows how to archive files
806 * with holes. However, on extraction (or during copy, -rw) we have to
807 * deal with these files. Without detecting the holes, the files can
808 * consume a lot of file space if just written to disk. This replacement
809 * for write when passed the basic allocation size of a file system block,
810 * uses lseek whenever it detects the input data is all 0 within that
811 * file block. In more detail, the strategy is as follows:
812 * While the input is all zero keep doing an lseek. Keep track of when we
813 * pass over file block boundaries. Only write when we hit a non zero
814 * input. once we have written a file block, we continue to write it to
815 * the end (we stop looking at the input). When we reach the start of the
816 * next file block, start checking for zero blocks again. Working on file
817 * block boundaries significantly reduces the overhead when copying files
818 * that are NOT very sparse. This overhead (when compared to a write) is
819 * almost below the measurement resolution on many systems. Without it,
820 * files with holes cannot be safely copied. It does has a side effect as
821 * it can put holes into files that did not have them before, but that is
822 * not a problem since the file contents are unchanged (in fact it saves
823 * file space). (Except on paging files for diskless clients. But since we
824 * cannot determine one of those file from here, we ignore them). If this
825 * ever ends up on a system where CTG files are supported and the holes
826 * are not desired, just do a conditional test in those routines that
827 * call file_write() and have it call write() instead. BEFORE CLOSING THE
828 * FILE, make sure to call file_flush() when the last write finishes with
829 * an empty block. A lot of file systems will not create an lseek hole at
830 * the end. In this case we drop a single 0 at the end to force the
831 * trailing 0's in the file.
832 * ---Parameters---
833 * rem: how many bytes left in this file system block
834 * isempt: have we written to the file block yet (is it empty)
835 * sz: basic file block allocation size
836 * cnt: number of bytes on this write
837 * str: buffer to write
838 * Return:
839 * number of bytes written, -1 on write (or lseek) error.
840 */
841
842 int
843 file_write(int fd, char *str, int cnt, int *rem, int *isempt, int sz,
844 char *name)
845 {
846 char *pt;
847 char *end;
848 int wcnt;
849 char *st = str;
850 char **strp;
851
852 /*
853 * while we have data to process
854 */
855 while (cnt) {
856 if (!*rem) {
857 /*
858 * We are now at the start of file system block again
859 * (or what we think one is...). start looking for
860 * empty blocks again
861 */
862 *isempt = 1;
863 *rem = sz;
864 }
865
866 /*
867 * only examine up to the end of the current file block or
868 * remaining characters to write, whatever is smaller
869 */
870 wcnt = MIN(cnt, *rem);
871 cnt -= wcnt;
872 *rem -= wcnt;
873 if (*isempt) {
874 /*
875 * have not written to this block yet, so we keep
876 * looking for zero's
877 */
878 pt = st;
879 end = st + wcnt;
880
881 /*
882 * look for a zero filled buffer
883 */
884 while ((pt < end) && (*pt == '\0'))
885 ++pt;
886
887 if (pt == end) {
888 /*
889 * skip, buf is empty so far
890 */
891 if (fd > -1 &&
892 lseek(fd, (off_t)wcnt, SEEK_CUR) < 0) {
893 syswarn(1, errno, "File seek on %s",
894 name);
895 return(-1);
896 }
897 st = pt;
898 continue;
899 }
900 /*
901 * drat, the buf is not zero filled
902 */
903 *isempt = 0;
904 }
905
906 /*
907 * have non-zero data in this file system block, have to write
908 */
909 switch (fd) {
910 case -1:
911 strp = &gnu_name_string;
912 break;
913 case -2:
914 strp = &gnu_link_string;
915 break;
916 default:
917 strp = NULL;
918 break;
919 }
920 if (strp) {
921 if (*strp)
922 err(1, "WARNING! Major Internal Error! GNU hack Failing!");
923 *strp = malloc(wcnt + 1);
924 if (*strp == NULL) {
925 tty_warn(1, "Out of memory");
926 return(-1);
927 }
928 strlcpy(*strp, st, wcnt);
929 break;
930 } else if (xwrite(fd, st, wcnt) != wcnt) {
931 syswarn(1, errno, "Failed write to file %s", name);
932 return(-1);
933 }
934 st += wcnt;
935 }
936 return(st - str);
937 }
938
939 /*
940 * file_flush()
941 * when the last file block in a file is zero, many file systems will not
942 * let us create a hole at the end. To get the last block with zeros, we
943 * write the last BYTE with a zero (back up one byte and write a zero).
944 */
945
946 void
947 file_flush(int fd, char *fname, int isempt)
948 {
949 static char blnk[] = "\0";
950
951 /*
952 * silly test, but make sure we are only called when the last block is
953 * filled with all zeros.
954 */
955 if (!isempt)
956 return;
957
958 /*
959 * move back one byte and write a zero
960 */
961 if (lseek(fd, (off_t)-1, SEEK_CUR) < 0) {
962 syswarn(1, errno, "Failed seek on file %s", fname);
963 return;
964 }
965
966 if (write_with_restart(fd, blnk, 1) < 0)
967 syswarn(1, errno, "Failed write to file %s", fname);
968 return;
969 }
970
971 /*
972 * rdfile_close()
973 * close a file we have been reading (to copy or archive). If we have to
974 * reset access time (tflag) do so (the times are stored in arcn).
975 */
976
977 void
978 rdfile_close(ARCHD *arcn, int *fd)
979 {
980 /*
981 * make sure the file is open
982 */
983 if (*fd < 0)
984 return;
985
986 (void)close(*fd);
987 *fd = -1;
988 if (!tflag)
989 return;
990
991 /*
992 * user wants last access time reset
993 */
994 set_ftime(arcn->org_name, arcn->sb.st_mtime, arcn->sb.st_atime, 1);
995 return;
996 }
997
998 /*
999 * set_crc()
1000 * read a file to calculate its crc. This is a real drag. Archive formats
1001 * that have this, end up reading the file twice (we have to write the
1002 * header WITH the crc before writing the file contents. Oh well...
1003 * Return:
1004 * 0 if was able to calculate the crc, -1 otherwise
1005 */
1006
1007 int
1008 set_crc(ARCHD *arcn, int fd)
1009 {
1010 int i;
1011 int res;
1012 off_t cpcnt = 0L;
1013 u_long size;
1014 unsigned long crc = 0L;
1015 char tbuf[FILEBLK];
1016 struct stat sb;
1017
1018 if (fd < 0) {
1019 /*
1020 * hmm, no fd, should never happen. well no crc then.
1021 */
1022 arcn->crc = 0L;
1023 return(0);
1024 }
1025
1026 if ((size = (u_long)arcn->sb.st_blksize) > (u_long)sizeof(tbuf))
1027 size = (u_long)sizeof(tbuf);
1028
1029 /*
1030 * read all the bytes we think that there are in the file. If the user
1031 * is trying to archive an active file, forget this file.
1032 */
1033 for(;;) {
1034 if ((res = read(fd, tbuf, size)) <= 0)
1035 break;
1036 cpcnt += res;
1037 for (i = 0; i < res; ++i)
1038 crc += (tbuf[i] & 0xff);
1039 }
1040
1041 /*
1042 * safety check. we want to avoid archiving files that are active as
1043 * they can create inconsistant archive copies.
1044 */
1045 if (cpcnt != arcn->sb.st_size)
1046 tty_warn(1, "File changed size %s", arcn->org_name);
1047 else if (fstat(fd, &sb) < 0)
1048 syswarn(1, errno, "Failed stat on %s", arcn->org_name);
1049 else if (arcn->sb.st_mtime != sb.st_mtime)
1050 tty_warn(1, "File %s was modified during read", arcn->org_name);
1051 else if (lseek(fd, (off_t)0L, SEEK_SET) < 0)
1052 syswarn(1, errno, "File rewind failed on: %s", arcn->org_name);
1053 else {
1054 arcn->crc = crc;
1055 return(0);
1056 }
1057 return(-1);
1058 }
1059