libexec exec fix
[minix.git] / bin / pax / file_subs.c
blobea7a66a8bd238fbc7bb3dbd82bd11dd8ddd4bcda
1 /* $NetBSD: file_subs.c,v 1.62 2009/04/07 19:52:35 perry Exp $ */
3 /*-
4 * Copyright (c) 1992 Keith Muller.
5 * Copyright (c) 1992, 1993
6 * The Regents of the University of California. All rights reserved.
8 * This code is derived from software contributed to Berkeley by
9 * Keith Muller of the University of California, San Diego.
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
13 * are met:
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 * 3. Neither the name of the University nor the names of its contributors
20 * may be used to endorse or promote products derived from this software
21 * without specific prior written permission.
23 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
24 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
31 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
32 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33 * SUCH DAMAGE.
36 #if HAVE_NBTOOL_CONFIG_H
37 #include "nbtool_config.h"
38 #endif
40 #include <sys/cdefs.h>
41 #if !defined(lint)
42 #if 0
43 static char sccsid[] = "@(#)file_subs.c 8.1 (Berkeley) 5/31/93";
44 #else
45 __RCSID("$NetBSD: file_subs.c,v 1.62 2009/04/07 19:52:35 perry Exp $");
46 #endif
47 #endif /* not lint */
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"
65 char *xtmp_name;
67 static int
68 mk_link(char *,struct stat *,char *, int);
70 static int warn_broken;
73 * routines that deal with file operations such as: creating, removing;
74 * and setting access modes, uid/gid and times of files
76 #define SET_BITS (S_ISUID | S_ISGID)
77 #define FILE_BITS (S_IRWXU | S_IRWXG | S_IRWXO)
78 #define A_BITS (FILE_BITS | SET_BITS | S_ISVTX)
81 * The S_ISVTX (sticky bit) can be set by non-superuser on directories
82 * but not other kinds of files.
84 #define FILEBITS(dir) ((dir) ? (FILE_BITS | S_ISVTX) : FILE_BITS)
85 #define SETBITS(dir) ((dir) ? SET_BITS : (SET_BITS | S_ISVTX))
87 static mode_t
88 apply_umask(mode_t mode)
90 static mode_t cached_umask;
91 static int cached_umask_valid;
93 if (!cached_umask_valid) {
94 cached_umask = umask(0);
95 umask(cached_umask);
96 cached_umask_valid = 1;
99 return mode & ~cached_umask;
103 * file_creat()
104 * Create and open a file.
105 * Return:
106 * file descriptor or -1 for failure
110 file_creat(ARCHD *arcn, int write_to_hardlink)
112 int fd = -1;
113 int oerrno;
116 * Some horribly busted tar implementations, have directory nodes
117 * that end in a /, but they mark as files. Compensate for that
118 * by not creating a directory node at this point, but a file node,
119 * and not creating the temp file.
121 if (arcn->nlen != 0 && arcn->name[arcn->nlen - 1] == '/') {
122 if (!warn_broken) {
123 tty_warn(0, "Archive was created with a broken tar;"
124 " file `%s' is a directory, but marked as plain.",
125 arcn->name);
126 warn_broken = 1;
128 return -1;
132 * In "cpio" archives it's usually the last record of a set of
133 * hardlinks which includes the contents of the file. We cannot
134 * use a tempory file in that case because we couldn't link it
135 * with the existing other hardlinks after restoring the contents
136 * to it. And it's also useless to create the hardlink under a
137 * temporary name because the other hardlinks would have partial
138 * contents while restoring.
140 if (write_to_hardlink)
141 return (open(arcn->name, O_TRUNC | O_EXCL | O_RDWR, 0));
144 * Create a temporary file name so that the file doesn't have partial
145 * contents while restoring.
147 arcn->tmp_name = malloc(arcn->nlen + 8);
148 if (arcn->tmp_name == NULL) {
149 syswarn(1, errno, "Cannot malloc %d bytes", arcn->nlen + 8);
150 return -1;
152 if (xtmp_name != NULL)
153 abort();
154 xtmp_name = arcn->tmp_name;
156 for (;;) {
158 * try to create the temporary file we use to restore the
159 * contents info. if this fails, keep checking all the nodes
160 * in the path until chk_path() finds that it cannot fix
161 * anything further. if that happens we just give up.
163 #ifdef __minix
165 /* For minix, generate the temporary filename
166 * conservatively - just write Xes into the last component.
167 * There is space because of the malloc().
169 int cc;
170 char *last_slash;
171 strcpy(arcn->tmp_name, arcn->name);
172 if(!(last_slash = strrchr(arcn->tmp_name, '/')))
173 strcpy(arcn->tmp_name, "XXXXXX");
174 else strcpy(last_slash+1, "XXXXXX");
176 #else
177 (void)snprintf(arcn->tmp_name, arcn->nlen + 8, "%s.XXXXXX",
178 arcn->name);
179 #endif
180 fd = mkstemp(arcn->tmp_name);
181 if (fd >= 0)
182 break;
183 oerrno = errno;
184 if (nodirs || chk_path(arcn->name,arcn->sb.st_uid,arcn->sb.st_gid) < 0) {
185 (void)fflush(listf);
186 syswarn(1, oerrno, "Cannot create %s", arcn->tmp_name);
187 xtmp_name = NULL;
188 free(arcn->tmp_name);
189 arcn->tmp_name = NULL;
190 return -1;
193 return fd;
197 * file_close()
198 * Close file descriptor to a file just created by pax. Sets modes,
199 * ownership and times as required.
200 * Return:
201 * 0 for success, -1 for failure
204 void
205 file_close(ARCHD *arcn, int fd)
207 char *tmp_name;
208 int res;
210 if (fd < 0)
211 return;
213 tmp_name = (arcn->tmp_name != NULL) ? arcn->tmp_name : arcn->name;
215 if (close(fd) < 0)
216 syswarn(0, errno, "Cannot close file descriptor on %s",
217 tmp_name);
220 * set owner/groups first as this may strip off mode bits we want
221 * then set file permission modes. Then set file access and
222 * modification times.
224 if (pids)
225 res = set_ids(tmp_name, arcn->sb.st_uid, arcn->sb.st_gid);
226 else
227 res = 0;
230 * IMPORTANT SECURITY NOTE:
231 * if not preserving mode or we cannot set uid/gid, then PROHIBIT
232 * set uid/gid bits but restore the file modes (since mkstemp doesn't).
234 if (!pmode || res)
235 arcn->sb.st_mode &= ~SETBITS(0);
236 if (pmode)
237 set_pmode(tmp_name, arcn->sb.st_mode);
238 else
239 set_pmode(tmp_name,
240 apply_umask((arcn->sb.st_mode & FILEBITS(0))));
241 if (patime || pmtime)
242 set_ftime(tmp_name, arcn->sb.st_mtime,
243 arcn->sb.st_atime, 0, 0);
245 /* Did we write directly to the target file? */
246 if (arcn->tmp_name == NULL)
247 return;
250 * Finally, now the temp file is fully instantiated rename it to
251 * the desired file name.
253 if (rename(tmp_name, arcn->name) < 0) {
254 syswarn(0, errno, "Cannot rename %s to %s",
255 tmp_name, arcn->name);
256 (void)unlink(tmp_name);
259 #if HAVE_STRUCT_STAT_ST_FLAGS
260 if (pfflags && arcn->type != PAX_SLK)
261 set_chflags(arcn->name, arcn->sb.st_flags);
262 #endif
264 free(arcn->tmp_name);
265 arcn->tmp_name = NULL;
266 xtmp_name = NULL;
270 * lnk_creat()
271 * Create a hard link to arcn->ln_name from arcn->name. arcn->ln_name
272 * must exist;
273 * Return:
274 * 0 if ok, -1 otherwise
278 lnk_creat(ARCHD *arcn, int *payload)
280 struct stat sb;
283 * Check if this hardlink carries the "payload". In "cpio" archives
284 * it's usually the last record of a set of hardlinks which includes
285 * the contents of the file.
288 *payload = S_ISREG(arcn->sb.st_mode) &&
289 (arcn->sb.st_size > 0) && (arcn->sb.st_size <= arcn->skip);
292 * We may be running as root, so we have to be sure that link target
293 * is not a directory, so we lstat and check. XXX: This is still racy.
295 if (lstat(arcn->ln_name, &sb) != -1 && S_ISDIR(sb.st_mode)) {
296 tty_warn(1, "A hard link to the directory %s is not allowed",
297 arcn->ln_name);
298 return -1;
301 return mk_link(arcn->ln_name, &sb, arcn->name, 0);
305 * cross_lnk()
306 * Create a hard link to arcn->org_name from arcn->name. Only used in copy
307 * with the -l flag. No warning or error if this does not succeed (we will
308 * then just create the file)
309 * Return:
310 * 1 if copy() should try to create this file node
311 * 0 if cross_lnk() ok, -1 for fatal flaw (like linking to self).
315 cross_lnk(ARCHD *arcn)
318 * try to make a link to original file (-l flag in copy mode). make
319 * sure we do not try to link to directories in case we are running as
320 * root (and it might succeed).
322 if (arcn->type == PAX_DIR)
323 return 1;
324 return mk_link(arcn->org_name, &(arcn->sb), arcn->name, 1);
328 * chk_same()
329 * In copy mode if we are not trying to make hard links between the src
330 * and destinations, make sure we are not going to overwrite ourselves by
331 * accident. This slows things down a little, but we have to protect all
332 * those people who make typing errors.
333 * Return:
334 * 1 the target does not exist, go ahead and copy
335 * 0 skip it file exists (-k) or may be the same as source file
339 chk_same(ARCHD *arcn)
341 struct stat sb;
344 * if file does not exist, return. if file exists and -k, skip it
345 * quietly
347 if (lstat(arcn->name, &sb) < 0)
348 return 1;
349 if (kflag)
350 return 0;
353 * better make sure the user does not have src == dest by mistake
355 if ((arcn->sb.st_dev == sb.st_dev) && (arcn->sb.st_ino == sb.st_ino)) {
356 tty_warn(1, "Unable to copy %s, file would overwrite itself",
357 arcn->name);
358 return 0;
360 return 1;
364 * mk_link()
365 * try to make a hard link between two files. if ign set, we do not
366 * complain.
367 * Return:
368 * 0 if successful (or we are done with this file but no error, such as
369 * finding the from file exists and the user has set -k).
370 * 1 when ign was set to indicates we could not make the link but we
371 * should try to copy/extract the file as that might work (and is an
372 * allowed option). -1 an error occurred.
375 static int
376 mk_link(char *to, struct stat *to_sb, char *from, int ign)
378 struct stat sb;
379 int oerrno;
382 * if from file exists, it has to be unlinked to make the link. If the
383 * file exists and -k is set, skip it quietly
385 if (lstat(from, &sb) == 0) {
386 if (kflag)
387 return 0;
390 * make sure it is not the same file, protect the user
392 if ((to_sb->st_dev==sb.st_dev)&&(to_sb->st_ino == sb.st_ino)) {
393 tty_warn(1, "Cannot link file %s to itself", to);
394 return -1;
398 * try to get rid of the file, based on the type
400 if (S_ISDIR(sb.st_mode) && strcmp(from, ".") != 0) {
401 if (rmdir(from) < 0) {
402 syswarn(1, errno, "Cannot remove %s", from);
403 return -1;
405 } else if (unlink(from) < 0) {
406 if (!ign) {
407 syswarn(1, errno, "Cannot remove %s", from);
408 return -1;
410 return 1;
415 * from file is gone (or did not exist), try to make the hard link.
416 * if it fails, check the path and try it again (if chk_path() says to
417 * try again)
419 for (;;) {
420 if (link(to, from) == 0)
421 break;
422 oerrno = errno;
423 if (chk_path(from, to_sb->st_uid, to_sb->st_gid) == 0)
424 continue;
425 if (!ign) {
426 syswarn(1, oerrno, "Cannot link to %s from %s", to,
427 from);
428 return -1;
430 return 1;
434 * all right the link was made
436 return 0;
440 * node_creat()
441 * create an entry in the file system (other than a file or hard link).
442 * If successful, sets uid/gid modes and times as required.
443 * Return:
444 * 0 if ok, -1 otherwise
448 node_creat(ARCHD *arcn)
450 int res;
451 int ign = 0;
452 int oerrno;
453 int pass = 0;
454 mode_t file_mode;
455 struct stat sb;
456 char target[MAXPATHLEN];
457 char *nm = arcn->name;
458 int len;
461 * create node based on type, if that fails try to unlink the node and
462 * try again. finally check the path and try again. As noted in the
463 * file and link creation routines, this method seems to exhibit the
464 * best performance in general use workloads.
466 file_mode = arcn->sb.st_mode & FILEBITS(arcn->type == PAX_DIR);
468 for (;;) {
469 switch (arcn->type) {
470 case PAX_DIR:
472 * If -h (or -L) was given in tar-mode, follow the
473 * potential symlink chain before trying to create the
474 * directory.
476 if (strcmp(NM_TAR, argv0) == 0 && Lflag) {
477 while (lstat(nm, &sb) == 0 &&
478 S_ISLNK(sb.st_mode)) {
479 len = readlink(nm, target,
480 sizeof target - 1);
481 if (len == -1) {
482 syswarn(0, errno,
483 "cannot follow symlink %s "
484 "in chain for %s",
485 nm, arcn->name);
486 res = -1;
487 goto badlink;
489 target[len] = '\0';
490 nm = target;
493 res = domkdir(nm, file_mode);
494 badlink:
495 if (ign)
496 res = 0;
497 break;
498 case PAX_CHR:
499 file_mode |= S_IFCHR;
500 res = mknod(nm, file_mode, arcn->sb.st_rdev);
501 break;
502 case PAX_BLK:
503 file_mode |= S_IFBLK;
504 res = mknod(nm, file_mode, arcn->sb.st_rdev);
505 break;
506 case PAX_FIF:
507 res = mkfifo(nm, file_mode);
508 break;
509 case PAX_SCK:
511 * Skip sockets, operation has no meaning under BSD
513 tty_warn(0,
514 "%s skipped. Sockets cannot be copied or extracted",
515 nm);
516 return (-1);
517 case PAX_SLK:
518 res = symlink(arcn->ln_name, nm);
519 break;
520 case PAX_CTG:
521 case PAX_HLK:
522 case PAX_HRG:
523 case PAX_REG:
524 default:
526 * we should never get here
528 tty_warn(0, "%s has an unknown file type, skipping",
529 nm);
530 return (-1);
534 * if we were able to create the node break out of the loop,
535 * otherwise try to unlink the node and try again. if that
536 * fails check the full path and try a final time.
538 if (res == 0)
539 break;
542 * we failed to make the node
544 oerrno = errno;
545 switch (pass++) {
546 case 0:
547 if ((ign = unlnk_exist(nm, arcn->type)) < 0)
548 return (-1);
549 continue;
551 case 1:
552 if (nodirs ||
553 chk_path(nm, arcn->sb.st_uid,
554 arcn->sb.st_gid) < 0) {
555 syswarn(1, oerrno, "Cannot create %s", nm);
556 return (-1);
558 continue;
562 * it must be a file that exists but we can't create or
563 * remove, but we must avoid the infinite loop.
565 break;
569 * we were able to create the node. set uid/gid, modes and times
571 if (pids)
572 res = set_ids(nm, arcn->sb.st_uid, arcn->sb.st_gid);
573 else
574 res = 0;
577 * IMPORTANT SECURITY NOTE:
578 * if not preserving mode or we cannot set uid/gid, then PROHIBIT any
579 * set uid/gid bits
581 if (!pmode || res)
582 arcn->sb.st_mode &= ~SETBITS(arcn->type == PAX_DIR);
583 if (pmode)
584 set_pmode(arcn->name, arcn->sb.st_mode);
586 if (arcn->type == PAX_DIR && strcmp(NM_CPIO, argv0) != 0) {
588 * Dirs must be processed again at end of extract to set times
589 * and modes to agree with those stored in the archive. However
590 * to allow extract to continue, we may have to also set owner
591 * rights. This allows nodes in the archive that are children
592 * of this directory to be extracted without failure. Both time
593 * and modes will be fixed after the entire archive is read and
594 * before pax exits.
596 if (access(nm, R_OK | W_OK | X_OK) < 0) {
597 if (lstat(nm, &sb) < 0) {
598 syswarn(0, errno,"Cannot access %s (stat)",
599 arcn->name);
600 set_pmode(nm,file_mode | S_IRWXU);
601 } else {
603 * We have to add rights to the dir, so we make
604 * sure to restore the mode. The mode must be
605 * restored AS CREATED and not as stored if
606 * pmode is not set.
608 set_pmode(nm, ((sb.st_mode &
609 FILEBITS(arcn->type == PAX_DIR)) |
610 S_IRWXU));
611 if (!pmode)
612 arcn->sb.st_mode = sb.st_mode;
616 * we have to force the mode to what was set here,
617 * since we changed it from the default as created.
619 add_dir(nm, arcn->nlen, &(arcn->sb), 1);
620 } else if (pmode || patime || pmtime)
621 add_dir(nm, arcn->nlen, &(arcn->sb), 0);
624 if (patime || pmtime)
625 set_ftime(arcn->name, arcn->sb.st_mtime,
626 arcn->sb.st_atime, 0, (arcn->type == PAX_SLK) ? 1 : 0);
628 #if HAVE_STRUCT_STAT_ST_FLAGS
629 if (pfflags && arcn->type != PAX_SLK)
630 set_chflags(arcn->name, arcn->sb.st_flags);
631 #endif
632 return 0;
636 * unlnk_exist()
637 * Remove node from file system with the specified name. We pass the type
638 * of the node that is going to replace it. When we try to create a
639 * directory and find that it already exists, we allow processing to
640 * continue as proper modes etc will always be set for it later on.
641 * Return:
642 * 0 is ok to proceed, no file with the specified name exists
643 * -1 we were unable to remove the node, or we should not remove it (-k)
644 * 1 we found a directory and we were going to create a directory.
648 unlnk_exist(char *name, int type)
650 struct stat sb;
653 * the file does not exist, or -k we are done
655 if (lstat(name, &sb) < 0)
656 return 0;
657 if (kflag)
658 return -1;
660 if (S_ISDIR(sb.st_mode)) {
662 * try to remove a directory, if it fails and we were going to
663 * create a directory anyway, tell the caller (return a 1).
665 * don't try to remove the directory if the name is "."
666 * otherwise later file/directory creation fails.
668 if (strcmp(name, ".") == 0)
669 return 1;
670 if (rmdir(name) < 0) {
671 if (type == PAX_DIR)
672 return 1;
673 syswarn(1, errno, "Cannot remove directory %s", name);
674 return -1;
676 return 0;
680 * try to get rid of all non-directory type nodes
682 if (unlink(name) < 0) {
683 (void)fflush(listf);
684 syswarn(1, errno, "Cannot unlink %s", name);
685 return -1;
687 return 0;
691 * chk_path()
692 * We were trying to create some kind of node in the file system and it
693 * failed. chk_path() makes sure the path up to the node exists and is
694 * writable. When we have to create a directory that is missing along the
695 * path somewhere, the directory we create will be set to the same
696 * uid/gid as the file has (when uid and gid are being preserved).
697 * NOTE: this routine is a real performance loss. It is only used as a
698 * last resort when trying to create entries in the file system.
699 * Return:
700 * -1 when it could find nothing it is allowed to fix.
701 * 0 otherwise
705 chk_path(char *name, uid_t st_uid, gid_t st_gid)
707 char *spt = name;
708 struct stat sb;
709 int retval = -1;
712 * watch out for paths with nodes stored directly in / (e.g. /bozo)
714 if (*spt == '/')
715 ++spt;
717 for(;;) {
719 * work forward from the first / and check each part of
720 * the path
722 spt = strchr(spt, '/');
723 if (spt == NULL)
724 break;
725 *spt = '\0';
728 * if it exists we assume it is a directory, it is not within
729 * the spec (at least it seems to read that way) to alter the
730 * file system for nodes NOT EXPLICITLY stored on the archive.
731 * If that assumption is changed, you would test the node here
732 * and figure out how to get rid of it (probably like some
733 * recursive unlink()) or fix up the directory permissions if
734 * required (do an access()).
736 if (lstat(name, &sb) == 0) {
737 *(spt++) = '/';
738 continue;
742 * the path fails at this point, see if we can create the
743 * needed directory and continue on
745 if (domkdir(name, S_IRWXU | S_IRWXG | S_IRWXO) == -1) {
746 *spt = '/';
747 retval = -1;
748 break;
752 * we were able to create the directory. We will tell the
753 * caller that we found something to fix, and it is ok to try
754 * and create the node again.
756 retval = 0;
757 if (pids)
758 (void)set_ids(name, st_uid, st_gid);
761 * make sure the user doesn't have some strange umask that
762 * causes this newly created directory to be unusable. We fix
763 * the modes and restore them back to the creation default at
764 * the end of pax
766 if ((access(name, R_OK | W_OK | X_OK) < 0) &&
767 (lstat(name, &sb) == 0)) {
768 set_pmode(name, ((sb.st_mode & FILEBITS(0)) |
769 S_IRWXU));
770 add_dir(name, spt - name, &sb, 1);
772 *(spt++) = '/';
773 continue;
776 * We perform one final check here, because if someone else
777 * created the directory in parallel with us, we might return
778 * the wrong error code, even if the directory exists now.
780 if (retval == -1 && stat(name, &sb) == 0 && S_ISDIR(sb.st_mode))
781 retval = 0;
782 return retval;
786 * set_ftime()
787 * Set the access time and modification time for a named file. If frc
788 * is non-zero we force these times to be set even if the user did not
789 * request access and/or modification time preservation (this is also
790 * used by -t to reset access times).
791 * When ign is zero, only those times the user has asked for are set, the
792 * other ones are left alone. We do not assume the un-documented feature
793 * of many utimes() implementations that consider a 0 time value as a do
794 * not set request.
796 * Unfortunately, there are systems where lutimes() is present but does
797 * not work on some filesystem types, which cannot be detected at
798 * compile time. This requires passing down symlink knowledge into
799 * this function to obtain correct operation. Linux with XFS is one
800 * example of such a system.
803 void
804 set_ftime(char *fnm, time_t mtime, time_t atime, int frc, int slk)
806 struct timeval tv[2];
807 struct stat sb;
809 tv[0].tv_sec = (long)atime;
810 tv[0].tv_usec = 0;
811 tv[1].tv_sec = (long)mtime;
812 tv[1].tv_usec = 0;
813 if (!frc && (!patime || !pmtime)) {
815 * if we are not forcing, only set those times the user wants
816 * set. We get the current values of the times if we need them.
818 if (lstat(fnm, &sb) == 0) {
819 #if BSD4_4 && !HAVE_NBTOOL_CONFIG_H
820 if (!patime)
821 TIMESPEC_TO_TIMEVAL(&tv[0], &sb.st_atimespec);
822 if (!pmtime)
823 TIMESPEC_TO_TIMEVAL(&tv[1], &sb.st_mtimespec);
824 #else
825 if (!patime)
826 tv[0].tv_sec = sb.st_atime;
827 if (!pmtime)
828 tv[1].tv_sec = sb.st_mtime;
829 #endif
830 } else
831 syswarn(0, errno, "Cannot obtain file stats %s", fnm);
835 * set the times
837 #if HAVE_LUTIMES
838 if (lutimes(fnm, tv) == 0)
839 return;
840 if (errno != ENOSYS) /* XXX linux: lutimes is per-FS */
841 goto bad;
842 #endif
843 if (slk)
844 return;
845 #ifndef __minix
846 /* LSC: FIXME UGLY Hack */
847 if (utimes(fnm, tv) == -1)
848 goto bad;
849 #endif
850 return;
851 bad:
852 syswarn(1, errno, "Access/modification time set failed on: %s", fnm);
856 * set_ids()
857 * set the uid and gid of a file system node
858 * Return:
859 * 0 when set, -1 on failure
863 set_ids(char *fnm, uid_t uid, gid_t gid)
865 if (geteuid() == 0)
866 if (lchown(fnm, uid, gid)) {
867 (void)fflush(listf);
868 syswarn(1, errno, "Cannot set file uid/gid of %s",
869 fnm);
870 return -1;
872 return 0;
876 * set_pmode()
877 * Set file access mode
880 void
881 set_pmode(char *fnm, mode_t mode)
883 mode &= A_BITS;
884 if (lchmod(fnm, mode)) {
885 (void)fflush(listf);
886 syswarn(1, errno, "Cannot set permissions on %s", fnm);
888 return;
892 * set_chflags()
893 * Set 4.4BSD file flags
895 void
896 set_chflags(char *fnm, u_int32_t flags)
899 #if 0
900 if (chflags(fnm, flags) < 0 && errno != EOPNOTSUPP)
901 syswarn(1, errno, "Cannot set file flags on %s", fnm);
902 #endif
903 return;
907 * file_write()
908 * Write/copy a file (during copy or archive extract). This routine knows
909 * how to copy files with lseek holes in it. (Which are read as file
910 * blocks containing all 0's but do not have any file blocks associated
911 * with the data). Typical examples of these are files created by dbm
912 * variants (.pag files). While the file size of these files are huge, the
913 * actual storage is quite small (the files are sparse). The problem is
914 * the holes read as all zeros so are probably stored on the archive that
915 * way (there is no way to determine if the file block is really a hole,
916 * we only know that a file block of all zero's can be a hole).
917 * At this writing, no major archive format knows how to archive files
918 * with holes. However, on extraction (or during copy, -rw) we have to
919 * deal with these files. Without detecting the holes, the files can
920 * consume a lot of file space if just written to disk. This replacement
921 * for write when passed the basic allocation size of a file system block,
922 * uses lseek whenever it detects the input data is all 0 within that
923 * file block. In more detail, the strategy is as follows:
924 * While the input is all zero keep doing an lseek. Keep track of when we
925 * pass over file block boundaries. Only write when we hit a non zero
926 * input. once we have written a file block, we continue to write it to
927 * the end (we stop looking at the input). When we reach the start of the
928 * next file block, start checking for zero blocks again. Working on file
929 * block boundaries significantly reduces the overhead when copying files
930 * that are NOT very sparse. This overhead (when compared to a write) is
931 * almost below the measurement resolution on many systems. Without it,
932 * files with holes cannot be safely copied. It does has a side effect as
933 * it can put holes into files that did not have them before, but that is
934 * not a problem since the file contents are unchanged (in fact it saves
935 * file space). (Except on paging files for diskless clients. But since we
936 * cannot determine one of those file from here, we ignore them). If this
937 * ever ends up on a system where CTG files are supported and the holes
938 * are not desired, just do a conditional test in those routines that
939 * call file_write() and have it call write() instead. BEFORE CLOSING THE
940 * FILE, make sure to call file_flush() when the last write finishes with
941 * an empty block. A lot of file systems will not create an lseek hole at
942 * the end. In this case we drop a single 0 at the end to force the
943 * trailing 0's in the file.
944 * ---Parameters---
945 * rem: how many bytes left in this file system block
946 * isempt: have we written to the file block yet (is it empty)
947 * sz: basic file block allocation size
948 * cnt: number of bytes on this write
949 * str: buffer to write
950 * Return:
951 * number of bytes written, -1 on write (or lseek) error.
955 file_write(int fd, char *str, int cnt, int *rem, int *isempt, int sz,
956 char *name)
958 char *pt;
959 char *end;
960 int wcnt;
961 char *st = str;
962 char **strp;
963 size_t *lenp;
966 * while we have data to process
968 while (cnt) {
969 if (!*rem) {
971 * We are now at the start of file system block again
972 * (or what we think one is...). start looking for
973 * empty blocks again
975 *isempt = 1;
976 *rem = sz;
980 * only examine up to the end of the current file block or
981 * remaining characters to write, whatever is smaller
983 wcnt = MIN(cnt, *rem);
984 cnt -= wcnt;
985 *rem -= wcnt;
986 if (*isempt) {
988 * have not written to this block yet, so we keep
989 * looking for zero's
991 pt = st;
992 end = st + wcnt;
995 * look for a zero filled buffer
997 while ((pt < end) && (*pt == '\0'))
998 ++pt;
1000 if (pt == end) {
1002 * skip, buf is empty so far
1004 if (fd > -1 &&
1005 lseek(fd, (off_t)wcnt, SEEK_CUR) < 0) {
1006 syswarn(1, errno, "File seek on %s",
1007 name);
1008 return -1;
1010 st = pt;
1011 continue;
1014 * drat, the buf is not zero filled
1016 *isempt = 0;
1020 * have non-zero data in this file system block, have to write
1022 switch (fd) {
1023 case -PAX_GLF:
1024 strp = &gnu_name_string;
1025 lenp = &gnu_name_length;
1026 break;
1027 case -PAX_GLL:
1028 strp = &gnu_link_string;
1029 lenp = &gnu_link_length;
1030 break;
1031 default:
1032 strp = NULL;
1033 lenp = NULL;
1034 break;
1036 if (strp) {
1037 char *nstr = *strp ? realloc(*strp, *lenp + wcnt + 1) :
1038 malloc(wcnt + 1);
1039 if (nstr == NULL) {
1040 tty_warn(1, "Out of memory");
1041 return -1;
1043 (void)strlcpy(&nstr[*lenp], st, wcnt + 1);
1044 *strp = nstr;
1045 *lenp += wcnt;
1046 } else if (xwrite(fd, st, wcnt) != wcnt) {
1047 syswarn(1, errno, "Failed write to file %s", name);
1048 return -1;
1050 st += wcnt;
1052 return st - str;
1056 * file_flush()
1057 * when the last file block in a file is zero, many file systems will not
1058 * let us create a hole at the end. To get the last block with zeros, we
1059 * write the last BYTE with a zero (back up one byte and write a zero).
1062 void
1063 file_flush(int fd, char *fname, int isempt)
1065 static char blnk[] = "\0";
1068 * silly test, but make sure we are only called when the last block is
1069 * filled with all zeros.
1071 if (!isempt)
1072 return;
1075 * move back one byte and write a zero
1077 if (lseek(fd, (off_t)-1, SEEK_CUR) < 0) {
1078 syswarn(1, errno, "Failed seek on file %s", fname);
1079 return;
1082 if (write_with_restart(fd, blnk, 1) < 0)
1083 syswarn(1, errno, "Failed write to file %s", fname);
1084 return;
1088 * rdfile_close()
1089 * close a file we have been reading (to copy or archive). If we have to
1090 * reset access time (tflag) do so (the times are stored in arcn).
1093 void
1094 rdfile_close(ARCHD *arcn, int *fd)
1097 * make sure the file is open
1099 if (*fd < 0)
1100 return;
1102 (void)close(*fd);
1103 *fd = -1;
1104 if (!tflag)
1105 return;
1108 * user wants last access time reset
1110 set_ftime(arcn->org_name, arcn->sb.st_mtime, arcn->sb.st_atime, 1, 0);
1111 return;
1115 * set_crc()
1116 * read a file to calculate its crc. This is a real drag. Archive formats
1117 * that have this, end up reading the file twice (we have to write the
1118 * header WITH the crc before writing the file contents. Oh well...
1119 * Return:
1120 * 0 if was able to calculate the crc, -1 otherwise
1124 set_crc(ARCHD *arcn, int fd)
1126 int i;
1127 int res;
1128 off_t cpcnt = 0L;
1129 u_long size;
1130 unsigned long crc = 0L;
1131 char tbuf[FILEBLK];
1132 struct stat sb;
1134 if (fd < 0) {
1136 * hmm, no fd, should never happen. well no crc then.
1138 arcn->crc = 0L;
1139 return 0;
1142 if ((size = (u_long)arcn->sb.st_blksize) > (u_long)sizeof(tbuf))
1143 size = (u_long)sizeof(tbuf);
1146 * read all the bytes we think that there are in the file. If the user
1147 * is trying to archive an active file, forget this file.
1149 for(;;) {
1150 if ((res = read(fd, tbuf, size)) <= 0)
1151 break;
1152 cpcnt += res;
1153 for (i = 0; i < res; ++i)
1154 crc += (tbuf[i] & 0xff);
1158 * safety check. we want to avoid archiving files that are active as
1159 * they can create inconsistent archive copies.
1161 if (cpcnt != arcn->sb.st_size)
1162 tty_warn(1, "File changed size %s", arcn->org_name);
1163 else if (fstat(fd, &sb) < 0)
1164 syswarn(1, errno, "Failed stat on %s", arcn->org_name);
1165 else if (arcn->sb.st_mtime != sb.st_mtime)
1166 tty_warn(1, "File %s was modified during read", arcn->org_name);
1167 else if (lseek(fd, (off_t)0L, SEEK_SET) < 0)
1168 syswarn(1, errno, "File rewind failed on: %s", arcn->org_name);
1169 else {
1170 arcn->crc = crc;
1171 return 0;
1173 return -1;