file_subs.c revision 1.33 1 /* $NetBSD: file_subs.c,v 1.33 2003/06/23 13:33:15 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 1.33 2003/06/23 13:33:15 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 (pids)
467 res = set_ids(nm, arcn->sb.st_uid, arcn->sb.st_gid);
468 else
469 res = 0;
470
471 /*
472 * IMPORTANT SECURITY NOTE:
473 * if not preserving mode or we cannot set uid/gid, then PROHIBIT any
474 * set uid/gid bits
475 */
476 if (!pmode || res)
477 arcn->sb.st_mode &= ~(SETBITS);
478 if (pmode)
479 set_pmode(arcn->name, arcn->sb.st_mode);
480
481 if (arcn->type == PAX_DIR && strcmp(NM_CPIO, argv0) != 0) {
482 /*
483 * Dirs must be processed again at end of extract to set times
484 * and modes to agree with those stored in the archive. However
485 * to allow extract to continue, we may have to also set owner
486 * rights. This allows nodes in the archive that are children
487 * of this directory to be extracted without failure. Both time
488 * and modes will be fixed after the entire archive is read and
489 * before pax exits.
490 */
491 if (access(nm, R_OK | W_OK | X_OK) < 0) {
492 if (lstat(nm, &sb) < 0) {
493 syswarn(0, errno,"Cannot access %s (stat)",
494 arcn->name);
495 set_pmode(nm,file_mode | S_IRWXU);
496 } else {
497 /*
498 * We have to add rights to the dir, so we make
499 * sure to restore the mode. The mode must be
500 * restored AS CREATED and not as stored if
501 * pmode is not set.
502 */
503 set_pmode(nm,
504 ((sb.st_mode & FILEBITS) | S_IRWXU));
505 if (!pmode)
506 arcn->sb.st_mode = sb.st_mode;
507 }
508
509 /*
510 * we have to force the mode to what was set here,
511 * since we changed it from the default as created.
512 */
513 add_dir(nm, arcn->nlen, &(arcn->sb), 1);
514 } else if (pmode || patime || pmtime)
515 add_dir(nm, arcn->nlen, &(arcn->sb), 0);
516 }
517
518 #if HAVE_LUTIMES
519 if (patime || pmtime)
520 #else
521 if ((patime || pmtime) && arcn->type != PAX_SLK)
522 #endif
523 set_ftime(arcn->name, arcn->sb.st_mtime, arcn->sb.st_atime, 0);
524
525 #if HAVE_STRUCT_STAT_ST_FLAGS
526 if (pfflags && arcn->type != PAX_SLK)
527 set_chflags(arcn->name, arcn->sb.st_flags);
528 #endif
529 return(0);
530 }
531
532 /*
533 * unlnk_exist()
534 * Remove node from file system with the specified name. We pass the type
535 * of the node that is going to replace it. When we try to create a
536 * directory and find that it already exists, we allow processing to
537 * continue as proper modes etc will always be set for it later on.
538 * Return:
539 * 0 is ok to proceed, no file with the specified name exists
540 * -1 we were unable to remove the node, or we should not remove it (-k)
541 * 1 we found a directory and we were going to create a directory.
542 */
543
544 int
545 unlnk_exist(char *name, int type)
546 {
547 struct stat sb;
548
549 /*
550 * the file does not exist, or -k we are done
551 */
552 if (lstat(name, &sb) < 0)
553 return(0);
554 if (kflag)
555 return(-1);
556
557 if (S_ISDIR(sb.st_mode)) {
558 /*
559 * try to remove a directory, if it fails and we were going to
560 * create a directory anyway, tell the caller (return a 1)
561 */
562 if (rmdir(name) < 0) {
563 if (type == PAX_DIR)
564 return(1);
565 syswarn(1, errno, "Cannot remove directory %s", name);
566 return(-1);
567 }
568 return(0);
569 }
570
571 /*
572 * try to get rid of all non-directory type nodes
573 */
574 if (unlink(name) < 0) {
575 (void)fflush(listf);
576 syswarn(1, errno, "Cannot unlink %s", name);
577 return(-1);
578 }
579 return(0);
580 }
581
582 /*
583 * chk_path()
584 * We were trying to create some kind of node in the file system and it
585 * failed. chk_path() makes sure the path up to the node exists and is
586 * writable. When we have to create a directory that is missing along the
587 * path somewhere, the directory we create will be set to the same
588 * uid/gid as the file has (when uid and gid are being preserved).
589 * NOTE: this routine is a real performance loss. It is only used as a
590 * last resort when trying to create entries in the file system.
591 * Return:
592 * -1 when it could find nothing it is allowed to fix.
593 * 0 otherwise
594 */
595
596 int
597 chk_path( char *name, uid_t st_uid, gid_t st_gid)
598 {
599 char *spt = name;
600 struct stat sb;
601 int retval = -1;
602
603 /*
604 * watch out for paths with nodes stored directly in / (e.g. /bozo)
605 */
606 if (*spt == '/')
607 ++spt;
608
609 for(;;) {
610 /*
611 * work forward from the first / and check each part of
612 * the path
613 */
614 spt = strchr(spt, '/');
615 if (spt == NULL)
616 break;
617 *spt = '\0';
618
619 /*
620 * if it exists we assume it is a directory, it is not within
621 * the spec (at least it seems to read that way) to alter the
622 * file system for nodes NOT EXPLICITLY stored on the archive.
623 * If that assumption is changed, you would test the node here
624 * and figure out how to get rid of it (probably like some
625 * recursive unlink()) or fix up the directory permissions if
626 * required (do an access()).
627 */
628 if (lstat(name, &sb) == 0) {
629 *(spt++) = '/';
630 continue;
631 }
632
633 /*
634 * the path fails at this point, see if we can create the
635 * needed directory and continue on
636 */
637 if (mkdir(name, S_IRWXU | S_IRWXG | S_IRWXO) < 0) {
638 *spt = '/';
639 retval = -1;
640 break;
641 }
642
643 /*
644 * we were able to create the directory. We will tell the
645 * caller that we found something to fix, and it is ok to try
646 * and create the node again.
647 */
648 retval = 0;
649 if (pids)
650 (void)set_ids(name, st_uid, st_gid);
651
652 /*
653 * make sure the user doesn't have some strange umask that
654 * causes this newly created directory to be unusable. We fix
655 * the modes and restore them back to the creation default at
656 * the end of pax
657 */
658 if ((access(name, R_OK | W_OK | X_OK) < 0) &&
659 (lstat(name, &sb) == 0)) {
660 set_pmode(name, ((sb.st_mode & FILEBITS) | S_IRWXU));
661 add_dir(name, spt - name, &sb, 1);
662 }
663 *(spt++) = '/';
664 continue;
665 }
666 return(retval);
667 }
668
669 /*
670 * set_ftime()
671 * Set the access time and modification time for a named file. If frc
672 * is non-zero we force these times to be set even if the user did not
673 * request access and/or modification time preservation (this is also
674 * used by -t to reset access times).
675 * When ign is zero, only those times the user has asked for are set, the
676 * other ones are left alone. We do not assume the un-documented feature
677 * of many utimes() implementations that consider a 0 time value as a do
678 * not set request.
679 */
680
681 void
682 set_ftime(char *fnm, time_t mtime, time_t atime, int frc)
683 {
684 struct timeval tv[2];
685 struct stat sb;
686
687 tv[0].tv_sec = (long)atime;
688 tv[0].tv_usec = 0;
689 tv[1].tv_sec = (long)mtime;
690 tv[1].tv_usec = 0;
691 if (!frc && (!patime || !pmtime)) {
692 /*
693 * if we are not forcing, only set those times the user wants
694 * set. We get the current values of the times if we need them.
695 */
696 if (lstat(fnm, &sb) == 0) {
697 #ifdef BSD4_4
698 if (!patime)
699 TIMESPEC_TO_TIMEVAL(&tv[0], &sb.st_atimespec);
700 if (!pmtime)
701 TIMESPEC_TO_TIMEVAL(&tv[1], &sb.st_mtimespec);
702 #else
703 if (!patime)
704 tv[0].tv_sec = sb.st_atime;
705 if (!pmtime)
706 tv[1].tv_sec = sb.st_mtime;
707 #endif
708 } else
709 syswarn(0, errno, "Cannot obtain file stats %s", fnm);
710 }
711
712 /*
713 * set the times
714 */
715 #if HAVE_LUTIMES
716 if (lutimes(fnm, tv))
717 #else
718 if (utimes(fnm, tv))
719 #endif
720 syswarn(1, errno, "Access/modification time set failed on: %s",
721 fnm);
722 return;
723 }
724
725 /*
726 * set_ids()
727 * set the uid and gid of a file system node
728 * Return:
729 * 0 when set, -1 on failure
730 */
731
732 int
733 set_ids(char *fnm, uid_t uid, gid_t gid)
734 {
735 if (geteuid() == 0)
736 if (lchown(fnm, uid, gid)) {
737 (void)fflush(listf);
738 syswarn(1, errno, "Cannot set file uid/gid of %s",
739 fnm);
740 return(-1);
741 }
742 return(0);
743 }
744
745 /*
746 * set_pmode()
747 * Set file access mode
748 */
749
750 void
751 set_pmode(char *fnm, mode_t mode)
752 {
753 mode &= ABITS;
754 if (lchmod(fnm, mode)) {
755 (void)fflush(listf);
756 syswarn(1, errno, "Cannot set permissions on %s", fnm);
757 }
758 return;
759 }
760
761 /*
762 * set_chflags()
763 * Set 4.4BSD file flags
764 */
765 void
766 set_chflags(char *fnm, u_int32_t flags)
767 {
768
769 #if 0
770 if (chflags(fnm, flags) < 0 && errno != EOPNOTSUPP)
771 syswarn(1, errno, "Cannot set file flags on %s", fnm);
772 #endif
773 return;
774 }
775
776 /*
777 * file_write()
778 * Write/copy a file (during copy or archive extract). This routine knows
779 * how to copy files with lseek holes in it. (Which are read as file
780 * blocks containing all 0's but do not have any file blocks associated
781 * with the data). Typical examples of these are files created by dbm
782 * variants (.pag files). While the file size of these files are huge, the
783 * actual storage is quite small (the files are sparse). The problem is
784 * the holes read as all zeros so are probably stored on the archive that
785 * way (there is no way to determine if the file block is really a hole,
786 * we only know that a file block of all zero's can be a hole).
787 * At this writing, no major archive format knows how to archive files
788 * with holes. However, on extraction (or during copy, -rw) we have to
789 * deal with these files. Without detecting the holes, the files can
790 * consume a lot of file space if just written to disk. This replacement
791 * for write when passed the basic allocation size of a file system block,
792 * uses lseek whenever it detects the input data is all 0 within that
793 * file block. In more detail, the strategy is as follows:
794 * While the input is all zero keep doing an lseek. Keep track of when we
795 * pass over file block boundaries. Only write when we hit a non zero
796 * input. once we have written a file block, we continue to write it to
797 * the end (we stop looking at the input). When we reach the start of the
798 * next file block, start checking for zero blocks again. Working on file
799 * block boundaries significantly reduces the overhead when copying files
800 * that are NOT very sparse. This overhead (when compared to a write) is
801 * almost below the measurement resolution on many systems. Without it,
802 * files with holes cannot be safely copied. It does has a side effect as
803 * it can put holes into files that did not have them before, but that is
804 * not a problem since the file contents are unchanged (in fact it saves
805 * file space). (Except on paging files for diskless clients. But since we
806 * cannot determine one of those file from here, we ignore them). If this
807 * ever ends up on a system where CTG files are supported and the holes
808 * are not desired, just do a conditional test in those routines that
809 * call file_write() and have it call write() instead. BEFORE CLOSING THE
810 * FILE, make sure to call file_flush() when the last write finishes with
811 * an empty block. A lot of file systems will not create an lseek hole at
812 * the end. In this case we drop a single 0 at the end to force the
813 * trailing 0's in the file.
814 * ---Parameters---
815 * rem: how many bytes left in this file system block
816 * isempt: have we written to the file block yet (is it empty)
817 * sz: basic file block allocation size
818 * cnt: number of bytes on this write
819 * str: buffer to write
820 * Return:
821 * number of bytes written, -1 on write (or lseek) error.
822 */
823
824 int
825 file_write(int fd, char *str, int cnt, int *rem, int *isempt, int sz,
826 char *name)
827 {
828 char *pt;
829 char *end;
830 int wcnt;
831 char *st = str;
832 char **strp;
833
834 /*
835 * while we have data to process
836 */
837 while (cnt) {
838 if (!*rem) {
839 /*
840 * We are now at the start of file system block again
841 * (or what we think one is...). start looking for
842 * empty blocks again
843 */
844 *isempt = 1;
845 *rem = sz;
846 }
847
848 /*
849 * only examine up to the end of the current file block or
850 * remaining characters to write, whatever is smaller
851 */
852 wcnt = MIN(cnt, *rem);
853 cnt -= wcnt;
854 *rem -= wcnt;
855 if (*isempt) {
856 /*
857 * have not written to this block yet, so we keep
858 * looking for zero's
859 */
860 pt = st;
861 end = st + wcnt;
862
863 /*
864 * look for a zero filled buffer
865 */
866 while ((pt < end) && (*pt == '\0'))
867 ++pt;
868
869 if (pt == end) {
870 /*
871 * skip, buf is empty so far
872 */
873 if (fd > -1 &&
874 lseek(fd, (off_t)wcnt, SEEK_CUR) < 0) {
875 syswarn(1, errno, "File seek on %s",
876 name);
877 return(-1);
878 }
879 st = pt;
880 continue;
881 }
882 /*
883 * drat, the buf is not zero filled
884 */
885 *isempt = 0;
886 }
887
888 /*
889 * have non-zero data in this file system block, have to write
890 */
891 switch (fd) {
892 case -1:
893 strp = &gnu_name_string;
894 break;
895 case -2:
896 strp = &gnu_link_string;
897 break;
898 default:
899 strp = NULL;
900 break;
901 }
902 if (strp) {
903 if (*strp)
904 err(1, "WARNING! Major Internal Error! GNU hack Failing!");
905 *strp = malloc(wcnt + 1);
906 if (*strp == NULL) {
907 tty_warn(1, "Out of memory");
908 return(-1);
909 }
910 strlcpy(*strp, st, wcnt);
911 break;
912 } else if (xwrite(fd, st, wcnt) != wcnt) {
913 syswarn(1, errno, "Failed write to file %s", name);
914 return(-1);
915 }
916 st += wcnt;
917 }
918 return(st - str);
919 }
920
921 /*
922 * file_flush()
923 * when the last file block in a file is zero, many file systems will not
924 * let us create a hole at the end. To get the last block with zeros, we
925 * write the last BYTE with a zero (back up one byte and write a zero).
926 */
927
928 void
929 file_flush(int fd, char *fname, int isempt)
930 {
931 static char blnk[] = "\0";
932
933 /*
934 * silly test, but make sure we are only called when the last block is
935 * filled with all zeros.
936 */
937 if (!isempt)
938 return;
939
940 /*
941 * move back one byte and write a zero
942 */
943 if (lseek(fd, (off_t)-1, SEEK_CUR) < 0) {
944 syswarn(1, errno, "Failed seek on file %s", fname);
945 return;
946 }
947
948 if (write_with_restart(fd, blnk, 1) < 0)
949 syswarn(1, errno, "Failed write to file %s", fname);
950 return;
951 }
952
953 /*
954 * rdfile_close()
955 * close a file we have been reading (to copy or archive). If we have to
956 * reset access time (tflag) do so (the times are stored in arcn).
957 */
958
959 void
960 rdfile_close(ARCHD *arcn, int *fd)
961 {
962 /*
963 * make sure the file is open
964 */
965 if (*fd < 0)
966 return;
967
968 (void)close(*fd);
969 *fd = -1;
970 if (!tflag)
971 return;
972
973 /*
974 * user wants last access time reset
975 */
976 set_ftime(arcn->org_name, arcn->sb.st_mtime, arcn->sb.st_atime, 1);
977 return;
978 }
979
980 /*
981 * set_crc()
982 * read a file to calculate its crc. This is a real drag. Archive formats
983 * that have this, end up reading the file twice (we have to write the
984 * header WITH the crc before writing the file contents. Oh well...
985 * Return:
986 * 0 if was able to calculate the crc, -1 otherwise
987 */
988
989 int
990 set_crc(ARCHD *arcn, int fd)
991 {
992 int i;
993 int res;
994 off_t cpcnt = 0L;
995 u_long size;
996 unsigned long crc = 0L;
997 char tbuf[FILEBLK];
998 struct stat sb;
999
1000 if (fd < 0) {
1001 /*
1002 * hmm, no fd, should never happen. well no crc then.
1003 */
1004 arcn->crc = 0L;
1005 return(0);
1006 }
1007
1008 if ((size = (u_long)arcn->sb.st_blksize) > (u_long)sizeof(tbuf))
1009 size = (u_long)sizeof(tbuf);
1010
1011 /*
1012 * read all the bytes we think that there are in the file. If the user
1013 * is trying to archive an active file, forget this file.
1014 */
1015 for(;;) {
1016 if ((res = read(fd, tbuf, size)) <= 0)
1017 break;
1018 cpcnt += res;
1019 for (i = 0; i < res; ++i)
1020 crc += (tbuf[i] & 0xff);
1021 }
1022
1023 /*
1024 * safety check. we want to avoid archiving files that are active as
1025 * they can create inconsistant archive copies.
1026 */
1027 if (cpcnt != arcn->sb.st_size)
1028 tty_warn(1, "File changed size %s", arcn->org_name);
1029 else if (fstat(fd, &sb) < 0)
1030 syswarn(1, errno, "Failed stat on %s", arcn->org_name);
1031 else if (arcn->sb.st_mtime != sb.st_mtime)
1032 tty_warn(1, "File %s was modified during read", arcn->org_name);
1033 else if (lseek(fd, (off_t)0L, SEEK_SET) < 0)
1034 syswarn(1, errno, "File rewind failed on: %s", arcn->org_name);
1035 else {
1036 arcn->crc = crc;
1037 return(0);
1038 }
1039 return(-1);
1040 }
1041