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 $ */
5 * Copyright (c) 1992 Keith Muller.
6 * Copyright (c) 1992, 1993
7 * The Regents of the University of California. All rights reserved.
9 * This code is derived from software contributed to Berkeley by
10 * Keith Muller of the University of California, San Diego.
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
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.
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
37 #include <sys/cdefs.h>
40 static const char sccsid
[] = "@(#)file_subs.c 8.1 (Berkeley) 5/31/93";
42 __used
static const char rcsid
[] = "$OpenBSD: file_subs.c,v 1.30 2005/11/09 19:59:06 otto Exp $";
46 #include <sys/param.h>
62 mk_link(char *, struct stat
*, char *, int);
65 * routines that deal with file operations such as: creating, removing;
66 * and setting access modes, uid/gid and times of files
69 #define FILEBITS (S_ISVTX | S_IRWXU | S_IRWXG | S_IRWXO)
70 #define SETBITS (S_ISUID | S_ISGID)
71 #define ABITS (FILEBITS | SETBITS)
75 * Create and open a file.
77 * file descriptor or -1 for failure
81 file_creat(ARCHD
*arcn
)
90 char cwd_buff
[MAXPATHLEN
];
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)
103 file_mode
= arcn
->sb
.st_mode
& FILEBITS
;
104 if ((fd
= open(arcn
->name
, O_WRONLY
| O_CREAT
| O_TRUNC
| O_EXCL
,
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
113 if (unlnk_exist(arcn
->name
, arcn
->type
) != 0)
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;
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
126 if ((fd
= open(path_to_open
, O_WRONLY
| O_CREAT
| O_TRUNC
,
128 /* clean up the invalid_action */
129 if (pax_invalid_action
>0) {
130 record_pax_invalid_action_results(arcn
, path_to_open
);
135 if (pax_invalid_action
>0) {
136 rc
= perform_pax_invalid_action(arcn
, oerrno
);
137 if (rc
== 0) continue;
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
);
150 if (new_path
) path_to_open
= new_path
; /* try again */
152 if (new_path
&& strcmp(new_path
, arcn
->name
)!=0) {
153 dochdir(cwd
); /* go back to original directory */
160 * Close file descriptor to a file just created by pax. Sets modes,
161 * ownership and times as required.
163 * 0 for success, -1 for failure
167 file_close(ARCHD
*arcn
, int fd
)
175 syswarn(0, errno
, "Unable to close file descriptor on %s",
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.
184 res
= set_ids(arcn
->name
, arcn
->sb
.st_uid
,
187 res
= 1; /* without pids, pax should NOT set s bits */
190 * IMPORTANT SECURITY NOTE:
191 * if not preserving mode or we cannot set uid/gid, then PROHIBIT
195 arcn
->sb
.st_mode
&= ~(SETBITS
);
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);
204 * Create a hard link to arcn->ln_name from arcn->name. arcn->ln_name
207 * 0 if ok, -1 otherwise
211 lnk_creat(ARCHD
*arcn
)
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
219 if (lstat(arcn
->ln_name
, &sb
) < 0) {
220 syswarn(1,errno
,"Unable to link to %s from %s", arcn
->ln_name
,
225 if (S_ISDIR(sb
.st_mode
)) {
226 paxwarn(1, "A hard link to the directory %s is not allowed",
231 if (S_ISLNK(sb
.st_mode
)) {
233 char buff
[PATH_MAX
+1];
235 * Conformance: cannot make hard link to symlink - just make a
236 * symlink to the target of the symlink
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
,
244 res
= symlink(buff
, arcn
->name
);
248 return(mk_link(arcn
->ln_name
, &sb
, arcn
->name
, 0));
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)
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).
262 cross_lnk(ARCHD
*arcn
)
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).
269 if (arcn
->type
== PAX_DIR
)
271 if (arcn
->type
== PAX_SLK
) { /* for Unix 03 conformance tests 202,203 */
275 return(mk_link(arcn
->org_name
, &(arcn
->sb
), arcn
->name
, 1));
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.
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
290 chk_same(ARCHD
*arcn
)
295 * if file does not exist, return. if file exists and -k, skip it
298 if (lstat(arcn
->name
, &sb
) < 0)
304 * better make sure the user does not have src == dest by mistake
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",
316 * try to make a hard link between two files. if ign set, we do not
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.
327 mk_link(char *to
, struct stat
*to_sb
, char *from
, int ign
)
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
336 if (lstat(from
, &sb
) == 0) {
341 * make sure it is not the same file, protect the user
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
);
349 * try to get rid of the file, based on the type
351 if (S_ISDIR(sb
.st_mode
)) {
352 if (rmdir(from
) < 0) {
353 syswarn(1, errno
, "Unable to remove %s", from
);
356 } else if (unlink(from
) < 0) {
358 syswarn(1, errno
, "Unable to remove %s", from
);
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
371 if (link(to
, from
) == 0)
374 if (!nodirs
&& chk_path(from
, to_sb
->st_uid
, to_sb
->st_gid
, NULL
) == 0)
377 syswarn(1, oerrno
, "Could not link to %s from %s", to
,
385 * all right the link was made
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.
395 * 0 if ok, -1 otherwise
399 node_creat(ARCHD
*arcn
)
407 char target
[MAXPATHLEN
];
408 char *nm
= arcn
->name
;
409 int nmlen
= arcn
->nlen
;
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.
418 file_mode
= arcn
->sb
.st_mode
& FILEBITS
;
421 switch (arcn
->type
) {
424 * If -h (or -L) was given in tar-mode, follow the
425 * potential symlink chain before trying to create the
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
,
435 "cannot follow symlink %s in chain for %s",
445 res
= mkdir(nm
, file_mode
);
452 file_mode
|= S_IFCHR
;
453 res
= mknod(nm
, file_mode
, arcn
->sb
.st_rdev
);
456 file_mode
|= S_IFBLK
;
457 res
= mknod(nm
, file_mode
, arcn
->sb
.st_rdev
);
460 res
= mkfifo(nm
, file_mode
);
464 * Skip sockets, operation has no meaning under BSD
467 "%s skipped. Sockets cannot be copied or extracted",
471 res
= symlink(arcn
->ln_name
, nm
);
479 * we should never get here
481 paxwarn(0, "%s has an unknown file type, skipping",
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.
495 * we failed to make the node
498 if ((ign
= unlnk_exist(nm
, arcn
->type
)) < 0)
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
);
511 * we were able to create the node. set uid/gid, modes and times
514 res
= ((arcn
->type
== PAX_SLK
) ?
515 #if defined(__APPLE__)
516 /* Mac OS X doesn't have lchown, so don't bother */
519 set_lids(nm
, arcn
->sb
.st_uid
, arcn
->sb
.st_gid
) :
521 set_ids(nm
, arcn
->sb
.st_uid
, arcn
->sb
.st_gid
));
523 res
= 1; /* without pids, pax should NOT set s bits */
526 * symlinks are done now.
528 if (arcn
->type
== PAX_SLK
)
532 * IMPORTANT SECURITY NOTE:
533 * if not preserving mode or we cannot set uid/gid, then PROHIBIT any
537 arcn
->sb
.st_mode
&= ~(SETBITS
);
539 set_pmode(nm
, arcn
->sb
.st_mode
);
541 if (arcn
->type
== PAX_DIR
&& strcmp(NM_CPIO
, argv0
) != 0) {
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
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)",
555 set_pmode(nm
,file_mode
| S_IRWXU
);
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
564 ((sb
.st_mode
& FILEBITS
) | S_IRWXU
));
566 arcn
->sb
.st_mode
= sb
.st_mode
;
570 * we have to force the mode to what was set here,
571 * since we changed it from the default as created.
573 add_dir(nm
, nmlen
, &(arcn
->sb
), 1);
574 } else if (pmode
|| patime
|| pmtime
)
575 add_dir(nm
, nmlen
, &(arcn
->sb
), 0);
578 if (patime
|| pmtime
)
579 set_ftime(nm
, arcn
->sb
.st_mtime
, arcn
->sb
.st_atime
, 0);
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.
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.
596 unlnk_exist(char *name
, int type
)
601 * the file does not exist, or -k we are done
603 if (lstat(name
, &sb
) < 0)
608 if(strstr(name
, "._") != NULL
) /* remove when stat works properly */
611 if (S_ISDIR(sb
.st_mode
)) {
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)
616 if (rmdir(name
) < 0) {
619 syswarn(1,errno
,"Unable to remove directory %s", name
);
626 * try to get rid of all non-directory type nodes
628 if (unlink(name
) < 0) {
629 syswarn(1, errno
, "Could not unlink %s", name
);
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.
645 * -1 when it could find nothing it is allowed to fix.
650 chk_path(char *name
, uid_t st_uid
, gid_t st_gid
, char ** new_name
)
653 int namelen
= strlen(name
);
658 * watch out for paths with nodes stored directly in / (e.g. /bozo)
665 * work forward from the first / and check each part of the path
667 spt
= strchr(spt
, '/');
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()).
681 if (lstat(name
, &sb
) == 0) {
683 if (new_name
==NULL
) continue;
684 retval
= 0; /* accept it one directory at a time */
689 * the path fails at this point, see if we can create the
690 * needed directory and continue on
692 if (mkdir(name
, S_IRWXU
| S_IRWXG
| S_IRWXO
) < 0) {
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.
705 (void)set_ids(name
, st_uid
, st_gid
);
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
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);
719 if (new_name
==NULL
) continue;
722 if ((new_name
!= NULL
) && retval
==0) {
723 /* save the new path */
726 printf ("chdir to %s\n", name);
731 printf ("remaining path: %s\n",spt);
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
753 set_ftime(char *fnm
, time_t mtime
, time_t atime
, int frc
)
755 static struct timeval tv
[2] = {{0L, 0L}, {0L, 0L}};
758 tv
[0].tv_sec
= (long)atime
;
759 tv
[1].tv_sec
= (long)mtime
;
760 if (!frc
&& (!patime
|| !pmtime
)) {
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.
765 if (lstat(fnm
, &sb
) == 0) {
767 tv
[0].tv_sec
= (long)sb
.st_atime
;
769 tv
[1].tv_sec
= (long)sb
.st_mtime
;
771 syswarn(0,errno
,"Unable to obtain file stats %s", fnm
);
777 if (pax_invalid_action_write_cwd
) {
778 char cwd_buff
[MAXPATHLEN
];
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
);
786 cleanup_pax_invalid_action();
788 if (utimes(fnm
, tv
) < 0)
789 syswarn(1, errno
, "Access/modification time set failed on: %s",
796 fset_ftime(char *fnm
, int fd
, time_t mtime
, time_t atime
, int frc
)
798 static struct timeval tv
[2] = {{0L, 0L}, {0L, 0L}};
801 tv
[0].tv_sec
= (long)atime
;
802 tv
[1].tv_sec
= (long)mtime
;
803 if (!frc
&& (!patime
|| !pmtime
)) {
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.
808 if (fstat(fd
, &sb
) == 0) {
810 tv
[0].tv_sec
= (long)sb
.st_atime
;
812 tv
[1].tv_sec
= (long)sb
.st_mtime
;
814 syswarn(0,errno
,"Unable to obtain file stats %s", fnm
);
819 if (futimes(fd
, tv
) < 0)
820 syswarn(1, errno
, "Access/modification time set failed on: %s",
827 * set the uid and gid of a file system node
829 * 0 when set, -1 on failure
833 set_ids(char *fnm
, uid_t uid
, gid_t gid
)
835 if (chown(fnm
, uid
, gid
) < 0) {
837 * ignore EPERM unless in verbose mode or being run by root.
838 * if running as pax, POSIX requires a warning.
840 if (strcmp(NM_PAX
, argv0
) == 0 || errno
!= EPERM
|| vflag
||
842 syswarn(1, errno
, "Unable to set file uid/gid of %s",
850 fset_ids(char *fnm
, int fd
, uid_t uid
, gid_t gid
)
852 if (fchown(fd
, uid
, gid
) < 0) {
854 * ignore EPERM unless in verbose mode or being run by root.
855 * if running as pax, POSIX requires a warning.
857 if (strcmp(NM_PAX
, argv0
) == 0 || errno
!= EPERM
|| vflag
||
859 syswarn(1, errno
, "Unable to set file uid/gid of %s",
866 #if !defined(__APPLE__)
869 * set the uid and gid of a file system node
871 * 0 when set, -1 on failure
875 set_lids(char *fnm
, uid_t uid
, gid_t gid
)
877 if (lchown(fnm
, uid
, gid
) < 0) {
879 * ignore EPERM unless in verbose mode or being run by root.
880 * if running as pax, POSIX requires a warning.
882 if (strcmp(NM_PAX
, argv0
) == 0 || errno
!= EPERM
|| vflag
||
884 syswarn(1, errno
, "Unable to set file uid/gid of %s",
890 #endif /* !__APPLE__ */
894 * Set file access mode
898 set_pmode(char *fnm
, mode_t mode
)
901 if (chmod(fnm
, mode
) < 0)
902 syswarn(1, errno
, "Could not set permissions on %s", fnm
);
907 fset_pmode(char *fnm
, int fd
, mode_t mode
)
910 if (fchmod(fd
, mode
) < 0)
911 syswarn(1, errno
, "Could not set permissions on %s", fnm
);
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.
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
960 * number of bytes written, -1 on write (or lseek) error.
964 file_write(int fd
, char *str
, int cnt
, int *rem
, int *isempt
, int sz
,
974 * while we have data to process
979 * We are now at the start of file system block again
980 * (or what we think one is...). start looking for
988 * only examine up to the end of the current file block or
989 * remaining characters to write, whatever is smaller
991 wcnt
= MIN(cnt
, *rem
);
996 * have not written to this block yet, so we keep
1003 * look for a zero filled buffer
1005 while ((pt
< end
) && (*pt
== '\0'))
1010 * skip, buf is empty so far
1013 lseek(fd
, (off_t
)wcnt
, SEEK_CUR
) < 0) {
1014 syswarn(1,errno
,"File seek on %s",
1022 * drat, the buf is not zero filled
1028 * have non-zero data in this file system block, have to write
1032 strp
= &gnu_name_string
;
1035 strp
= &gnu_link_string
;
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");
1049 memcpy(*strp
, st
, wcnt
);
1050 (*strp
)[wcnt
] = '\0';
1052 } else if (write(fd
, st
, wcnt
) != wcnt
) {
1053 syswarn(1, errno
, "Failed write to file %s", name
);
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).
1069 file_flush(int fd
, char *fname
, int isempt
)
1071 static char blnk
[] = "\0";
1074 * silly test, but make sure we are only called when the last block is
1075 * filled with all zeros.
1081 * move back one byte and write a zero
1083 if (lseek(fd
, (off_t
)-1, SEEK_CUR
) < 0) {
1084 syswarn(1, errno
, "Failed seek on file %s", fname
);
1088 if (write(fd
, blnk
, 1) < 0)
1089 syswarn(1, errno
, "Failed write to file %s", fname
);
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).
1100 rdfile_close(ARCHD
*arcn
, int *fd
)
1103 * make sure the file is open
1114 * user wants last access time reset
1116 set_ftime(arcn
->org_name
, arcn
->sb
.st_mtime
, arcn
->sb
.st_atime
, 1);
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...
1126 * 0 if was able to calculate the crc, -1 otherwise
1130 set_crc(ARCHD
*arcn
, int fd
)
1142 * hmm, no fd, should never happen. well no crc then.
1148 if ((size
= (u_long
)arcn
->sb
.st_blksize
) > (u_long
)sizeof(tbuf
))
1149 size
= (u_long
)sizeof(tbuf
);
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.
1156 if ((res
= read(fd
, tbuf
, size
)) <= 0)
1159 for (i
= 0; i
< res
; ++i
)
1160 crc
+= (tbuf
[i
] & 0xff);
1164 * safety check. we want to avoid archiving files that are active as
1165 * they can create inconsistent archive copies.
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
);