1 /* $OpenBSD: file_subs.c,v 1.13 1997/09/01 18:29:48 deraadt 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. All advertising materials mentioning features or use of this software
21 * must display the following acknowledgement:
22 * This product includes software developed by the University of
23 * California, Berkeley and its contributors.
24 * 4. Neither the name of the University nor the names of its contributors
25 * may be used to endorse or promote products derived from this software
26 * without specific prior written permission.
28 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
29 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
30 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
31 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
32 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
33 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
34 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
35 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
36 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
37 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
43 static char sccsid[] = "@(#)file_subs.c 8.1 (Berkeley) 5/31/93";
45 static char rcsid[] = "$OpenBSD: file_subs.c,v 1.13 1997/09/01 18:29:48 deraadt Exp $";
49 #include <sys/types.h>
53 #include <sys/param.h>
65 mk_link __P((register char *,register struct stat *,register char *, int));
68 * routines that deal with file operations such as: creating, removing;
69 * and setting access modes, uid/gid and times of files
72 #define FILEBITS (S_ISVTX | S_IRWXU | S_IRWXG | S_IRWXO)
73 #define SETBITS (S_ISUID | S_ISGID)
74 #define ABITS (FILEBITS | SETBITS)
78 * Create and open a file.
80 * file descriptor or -1 for failure
85 file_creat(register ARCHD *arcn)
97 * assume file doesn't exist, so just try to create it, most times this
98 * works. We have to take special handling when the file does exist. To
99 * detect this, we use O_EXCL. For example when trying to create a
100 * file and a character device or fifo exists with the same name, we
101 * can accidently open the device by mistake (or block waiting to open)
102 * If we find that the open has failed, then figure spend the effore to
103 * figure out why. This strategy was found to have better average
104 * performance in common use than checking the file (and the path)
107 file_mode = arcn->sb.st_mode & FILEBITS;
108 if ((fd = open(arcn->name, O_WRONLY | O_CREAT | O_TRUNC | O_EXCL,
113 * the file seems to exist. First we try to get rid of it (found to be
114 * the second most common failure when traced). If this fails, only
115 * then we go to the expense to check and create the path to the file
117 if (unlnk_exist(arcn->name, arcn->type) != 0)
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(arcn->name, O_WRONLY | O_CREAT | O_TRUNC,
130 if (nodirs || chk_path(arcn->name,arcn->sb.st_uid,arcn->sb.st_gid) < 0) {
131 syswarn(1, oerrno, "Unable to create %s", arcn->name);
140 * Close file descriptor to a file just created by pax. Sets modes,
141 * ownership and times as required.
143 * 0 for success, -1 for failure
148 file_close(register ARCHD *arcn, int fd)
152 register ARCHD *arcn;
161 syswarn(0, errno, "Unable to close file descriptor on %s",
165 * set owner/groups first as this may strip off mode bits we want
166 * then set file permission modes. Then set file access and
167 * modification times.
170 res = set_ids(arcn->name, arcn->sb.st_uid, arcn->sb.st_gid);
173 * IMPORTANT SECURITY NOTE:
174 * if not preserving mode or we cannot set uid/gid, then PROHIBIT
178 arcn->sb.st_mode &= ~(SETBITS);
180 set_pmode(arcn->name, arcn->sb.st_mode);
181 if (patime || pmtime)
182 set_ftime(arcn->name, arcn->sb.st_mtime, arcn->sb.st_atime, 0);
187 * Create a hard link to arcn->ln_name from arcn->name. arcn->ln_name
190 * 0 if ok, -1 otherwise
195 lnk_creat(register ARCHD *arcn)
199 register ARCHD *arcn;
205 * we may be running as root, so we have to be sure that link target
206 * is not a directory, so we lstat and check
208 if (lstat(arcn->ln_name, &sb) < 0) {
209 syswarn(1,errno,"Unable to link to %s from %s", arcn->ln_name,
214 if (S_ISDIR(sb.st_mode)) {
215 paxwarn(1, "A hard link to the directory %s is not allowed",
220 return(mk_link(arcn->ln_name, &sb, arcn->name, 0));
225 * Create a hard link to arcn->org_name from arcn->name. Only used in copy
226 * with the -l flag. No warning or error if this does not succeed (we will
227 * then just create the file)
229 * 1 if copy() should try to create this file node
230 * 0 if cross_lnk() ok, -1 for fatal flaw (like linking to self).
235 cross_lnk(register ARCHD *arcn)
239 register ARCHD *arcn;
243 * try to make a link to orginal file (-l flag in copy mode). make sure
244 * we do not try to link to directories in case we are running as root
245 * (and it might succeed).
247 if (arcn->type == PAX_DIR)
249 return(mk_link(arcn->org_name, &(arcn->sb), arcn->name, 1));
254 * In copy mode if we are not trying to make hard links between the src
255 * and destinations, make sure we are not going to overwrite ourselves by
256 * accident. This slows things down a little, but we have to protect all
257 * those people who make typing errors.
259 * 1 the target does not exist, go ahead and copy
260 * 0 skip it file exists (-k) or may be the same as source file
265 chk_same(register ARCHD *arcn)
269 register ARCHD *arcn;
275 * if file does not exist, return. if file exists and -k, skip it
278 if (lstat(arcn->name, &sb) < 0)
284 * better make sure the user does not have src == dest by mistake
286 if ((arcn->sb.st_dev == sb.st_dev) && (arcn->sb.st_ino == sb.st_ino)) {
287 paxwarn(1, "Unable to copy %s, file would overwrite itself",
296 * try to make a hard link between two files. if ign set, we do not
299 * 0 if successful (or we are done with this file but no error, such as
300 * finding the from file exists and the user has set -k).
301 * 1 when ign was set to indicates we could not make the link but we
302 * should try to copy/extract the file as that might work (and is an
303 * allowed option). -1 an error occurred.
308 mk_link(register char *to, register struct stat *to_sb, register char *from,
312 mk_link(to, to_sb, from, ign)
314 register struct stat *to_sb;
323 * if from file exists, it has to be unlinked to make the link. If the
324 * file exists and -k is set, skip it quietly
326 if (lstat(from, &sb) == 0) {
331 * make sure it is not the same file, protect the user
333 if ((to_sb->st_dev==sb.st_dev)&&(to_sb->st_ino == sb.st_ino)) {
334 paxwarn(1, "Unable to link file %s to itself", to);
339 * try to get rid of the file, based on the type
341 if (S_ISDIR(sb.st_mode)) {
342 if (rmdir(from) < 0) {
343 syswarn(1, errno, "Unable to remove %s", from);
346 } else if (unlink(from) < 0) {
348 syswarn(1, errno, "Unable to remove %s", from);
356 * from file is gone (or did not exist), try to make the hard link.
357 * if it fails, check the path and try it again (if chk_path() says to
361 if (link(to, from) == 0)
364 if (!nodirs && chk_path(from, to_sb->st_uid, to_sb->st_gid) == 0)
367 syswarn(1, oerrno, "Could not link to %s from %s", to,
375 * all right the link was made
382 * create an entry in the file system (other than a file or hard link).
383 * If successful, sets uid/gid modes and times as required.
385 * 0 if ok, -1 otherwise
390 node_creat(register ARCHD *arcn)
394 register ARCHD *arcn;
398 register int ign = 0;
400 register int pass = 0;
405 * create node based on type, if that fails try to unlink the node and
406 * try again. finally check the path and try again. As noted in the
407 * file and link creation routines, this method seems to exhibit the
408 * best performance in general use workloads.
410 file_mode = arcn->sb.st_mode & FILEBITS;
415 res = mkdir(arcn->name, file_mode);
420 file_mode |= S_IFCHR;
421 res = mknod(arcn->name, file_mode, arcn->sb.st_rdev);
424 file_mode |= S_IFBLK;
425 res = mknod(arcn->name, file_mode, arcn->sb.st_rdev);
428 res = mkfifo(arcn->name, file_mode);
432 * Skip sockets, operation has no meaning under BSD
435 "%s skipped. Sockets cannot be copied or extracted",
439 res = symlink(arcn->ln_name, arcn->name);
447 * we should never get here
449 paxwarn(0, "%s has an unknown file type, skipping",
455 * if we were able to create the node break out of the loop,
456 * otherwise try to unlink the node and try again. if that
457 * fails check the full path and try a final time.
463 * we failed to make the node
466 if ((ign = unlnk_exist(arcn->name, arcn->type)) < 0)
472 if (nodirs || chk_path(arcn->name,arcn->sb.st_uid,arcn->sb.st_gid) < 0) {
473 syswarn(1, oerrno, "Could not create: %s", arcn->name);
479 * we were able to create the node. set uid/gid, modes and times
482 res = ((arcn->type == PAX_SLK) ?
483 set_lids(arcn->name, arcn->sb.st_uid, arcn->sb.st_gid) :
484 set_ids(arcn->name, arcn->sb.st_uid, arcn->sb.st_gid));
489 * symlinks are done now.
491 if (arcn->type == PAX_SLK)
495 * IMPORTANT SECURITY NOTE:
496 * if not preserving mode or we cannot set uid/gid, then PROHIBIT any
500 arcn->sb.st_mode &= ~(SETBITS);
502 set_pmode(arcn->name, arcn->sb.st_mode);
504 if (arcn->type == PAX_DIR && strcmp(NM_CPIO, argv0) != 0) {
506 * Dirs must be processed again at end of extract to set times
507 * and modes to agree with those stored in the archive. However
508 * to allow extract to continue, we may have to also set owner
509 * rights. This allows nodes in the archive that are children
510 * of this directory to be extracted without failure. Both time
511 * and modes will be fixed after the entire archive is read and
514 if (access(arcn->name, R_OK | W_OK | X_OK) < 0) {
515 if (lstat(arcn->name, &sb) < 0) {
516 syswarn(0, errno,"Could not access %s (stat)",
518 set_pmode(arcn->name,file_mode | S_IRWXU);
521 * We have to add rights to the dir, so we make
522 * sure to restore the mode. The mode must be
523 * restored AS CREATED and not as stored if
526 set_pmode(arcn->name,
527 ((sb.st_mode & FILEBITS) | S_IRWXU));
529 arcn->sb.st_mode = sb.st_mode;
533 * we have to force the mode to what was set here,
534 * since we changed it from the default as created.
536 add_dir(arcn->name, arcn->nlen, &(arcn->sb), 1);
537 } else if (pmode || patime || pmtime)
538 add_dir(arcn->name, arcn->nlen, &(arcn->sb), 0);
541 if (patime || pmtime)
542 set_ftime(arcn->name, arcn->sb.st_mtime, arcn->sb.st_atime, 0);
548 * Remove node from file system with the specified name. We pass the type
549 * of the node that is going to replace it. When we try to create a
550 * directory and find that it already exists, we allow processing to
551 * continue as proper modes etc will always be set for it later on.
553 * 0 is ok to proceed, no file with the specified name exists
554 * -1 we were unable to remove the node, or we should not remove it (-k)
555 * 1 we found a directory and we were going to create a directory.
560 unlnk_exist(register char *name, register int type)
563 unlnk_exist(name, type)
571 * the file does not exist, or -k we are done
573 if (lstat(name, &sb) < 0)
578 if (S_ISDIR(sb.st_mode)) {
580 * try to remove a directory, if it fails and we were going to
581 * create a directory anyway, tell the caller (return a 1)
583 if (rmdir(name) < 0) {
586 syswarn(1,errno,"Unable to remove directory %s", name);
593 * try to get rid of all non-directory type nodes
595 if (unlink(name) < 0) {
596 syswarn(1, errno, "Could not unlink %s", name);
604 * We were trying to create some kind of node in the file system and it
605 * failed. chk_path() makes sure the path up to the node exists and is
606 * writeable. When we have to create a directory that is missing along the
607 * path somewhere, the directory we create will be set to the same
608 * uid/gid as the file has (when uid and gid are being preserved).
609 * NOTE: this routine is a real performance loss. It is only used as a
610 * last resort when trying to create entries in the file system.
612 * -1 when it could find nothing it is allowed to fix.
618 chk_path( register char *name, uid_t st_uid, gid_t st_gid)
621 chk_path(name, st_uid, st_gid)
627 register char *spt = name;
632 * watch out for paths with nodes stored directly in / (e.g. /bozo)
639 * work foward from the first / and check each part of the path
641 spt = strchr(spt, '/');
647 * if it exists we assume it is a directory, it is not within
648 * the spec (at least it seems to read that way) to alter the
649 * file system for nodes NOT EXPLICITLY stored on the archive.
650 * If that assumption is changed, you would test the node here
651 * and figure out how to get rid of it (probably like some
652 * recursive unlink()) or fix up the directory permissions if
653 * required (do an access()).
655 if (lstat(name, &sb) == 0) {
661 * the path fails at this point, see if we can create the
662 * needed directory and continue on
664 if (mkdir(name, S_IRWXU | S_IRWXG | S_IRWXO) < 0) {
671 * we were able to create the directory. We will tell the
672 * caller that we found something to fix, and it is ok to try
673 * and create the node again.
677 (void)set_ids(name, st_uid, st_gid);
680 * make sure the user doen't have some strange umask that
681 * causes this newly created directory to be unusable. We fix
682 * the modes and restore them back to the creation default at
685 if ((access(name, R_OK | W_OK | X_OK) < 0) &&
686 (lstat(name, &sb) == 0)) {
687 set_pmode(name, ((sb.st_mode & FILEBITS) | S_IRWXU));
688 add_dir(name, spt - name, &sb, 1);
698 * Set the access time and modification time for a named file. If frc is
699 * non-zero we force these times to be set even if the user did not
700 * request access and/or modification time preservation (this is also
701 * used by -t to reset access times).
702 * When ign is zero, only those times the user has asked for are set, the
703 * other ones are left alone. We do not assume the un-documented feature
704 * of many utimes() implementations that consider a 0 time value as a do
710 set_ftime(char *fnm, time_t mtime, time_t atime, int frc)
713 set_ftime(fnm, mtime, atime, frc)
720 static struct timeval tv[2] = {{0L, 0L}, {0L, 0L}};
723 tv[0].tv_sec = (long)atime;
724 tv[1].tv_sec = (long)mtime;
725 if (!frc && (!patime || !pmtime)) {
727 * if we are not forcing, only set those times the user wants
728 * set. We get the current values of the times if we need them.
730 if (lstat(fnm, &sb) == 0) {
732 tv[0].tv_sec = (long)sb.st_atime;
734 tv[1].tv_sec = (long)sb.st_mtime;
736 syswarn(0,errno,"Unable to obtain file stats %s", fnm);
742 if (utimes(fnm, tv) < 0)
743 syswarn(1, errno, "Access/modification time set failed on: %s",
750 * set the uid and gid of a file system node
752 * 0 when set, -1 on failure
757 set_ids(char *fnm, uid_t uid, gid_t gid)
760 set_ids(fnm, uid, gid)
766 if (chown(fnm, uid, gid) < 0) {
768 * ignore EPERM unless in verbose mode or being run by root.
769 * if running as pax, POSIX requires a warning.
771 if (strcmp(NM_PAX, argv0) == 0 || errno != EPERM || vflag ||
773 syswarn(1, errno, "Unable to set file uid/gid of %s",
782 * set the uid and gid of a file system node
784 * 0 when set, -1 on failure
789 set_lids(char *fnm, uid_t uid, gid_t gid)
792 set_lids(fnm, uid, gid)
798 if (lchown(fnm, uid, gid) < 0) {
800 * ignore EPERM unless in verbose mode or being run by root.
801 * if running as pax, POSIX requires a warning.
803 if (strcmp(NM_PAX, argv0) == 0 || errno != EPERM || vflag ||
805 syswarn(1, errno, "Unable to set file uid/gid of %s",
814 * Set file access mode
819 set_pmode(char *fnm, mode_t mode)
828 if (chmod(fnm, mode) < 0)
829 syswarn(1, errno, "Could not set permissions on %s", fnm);
835 * Write/copy a file (during copy or archive extract). This routine knows
836 * how to copy files with lseek holes in it. (Which are read as file
837 * blocks containing all 0's but do not have any file blocks associated
838 * with the data). Typical examples of these are files created by dbm
839 * variants (.pag files). While the file size of these files are huge, the
840 * actual storage is quite small (the files are sparse). The problem is
841 * the holes read as all zeros so are probably stored on the archive that
842 * way (there is no way to determine if the file block is really a hole,
843 * we only know that a file block of all zero's can be a hole).
844 * At this writing, no major archive format knows how to archive files
845 * with holes. However, on extraction (or during copy, -rw) we have to
846 * deal with these files. Without detecting the holes, the files can
847 * consume a lot of file space if just written to disk. This replacement
848 * for write when passed the basic allocation size of a file system block,
849 * uses lseek whenever it detects the input data is all 0 within that
850 * file block. In more detail, the strategy is as follows:
851 * While the input is all zero keep doing an lseek. Keep track of when we
852 * pass over file block boundries. Only write when we hit a non zero
853 * input. once we have written a file block, we continue to write it to
854 * the end (we stop looking at the input). When we reach the start of the
855 * next file block, start checking for zero blocks again. Working on file
856 * block boundries significantly reduces the overhead when copying files
857 * that are NOT very sparse. This overhead (when compared to a write) is
858 * almost below the measurement resolution on many systems. Without it,
859 * files with holes cannot be safely copied. It does has a side effect as
860 * it can put holes into files that did not have them before, but that is
861 * not a problem since the file contents are unchanged (in fact it saves
862 * file space). (Except on paging files for diskless clients. But since we
863 * cannot determine one of those file from here, we ignore them). If this
864 * ever ends up on a system where CTG files are supported and the holes
865 * are not desired, just do a conditional test in those routines that
866 * call file_write() and have it call write() instead. BEFORE CLOSING THE
867 * FILE, make sure to call file_flush() when the last write finishes with
868 * an empty block. A lot of file systems will not create an lseek hole at
869 * the end. In this case we drop a single 0 at the end to force the
870 * trailing 0's in the file.
872 * rem: how many bytes left in this file system block
873 * isempt: have we written to the file block yet (is it empty)
874 * sz: basic file block allocation size
875 * cnt: number of bytes on this write
876 * str: buffer to write
878 * number of bytes written, -1 on write (or lseek) error.
883 file_write(int fd, char *str, register int cnt, int *rem, int *isempt, int sz,
887 file_write(fd, str, cnt, rem, isempt, sz, name)
900 register char *st = str;
903 * while we have data to process
908 * We are now at the start of file system block again
909 * (or what we think one is...). start looking for
917 * only examine up to the end of the current file block or
918 * remaining characters to write, whatever is smaller
920 wcnt = MIN(cnt, *rem);
925 * have not written to this block yet, so we keep
932 * look for a zero filled buffer
934 while ((pt < end) && (*pt == '\0'))
939 * skip, buf is empty so far
941 if (lseek(fd, (off_t)wcnt, SEEK_CUR) < 0) {
942 syswarn(1,errno,"File seek on %s",
950 * drat, the buf is not zero filled
956 * have non-zero data in this file system block, have to write
958 if (write(fd, st, wcnt) != wcnt) {
959 syswarn(1, errno, "Failed write to file %s", name);
969 * when the last file block in a file is zero, many file systems will not
970 * let us create a hole at the end. To get the last block with zeros, we
971 * write the last BYTE with a zero (back up one byte and write a zero).
976 file_flush(int fd, char *fname, int isempt)
979 file_flush(fd, fname, isempt)
985 static char blnk[] = "\0";
988 * silly test, but make sure we are only called when the last block is
989 * filled with all zeros.
995 * move back one byte and write a zero
997 if (lseek(fd, (off_t)-1, SEEK_CUR) < 0) {
998 syswarn(1, errno, "Failed seek on file %s", fname);
1002 if (write(fd, blnk, 1) < 0)
1003 syswarn(1, errno, "Failed write to file %s", fname);
1009 * close a file we have beed reading (to copy or archive). If we have to
1010 * reset access time (tflag) do so (the times are stored in arcn).
1015 rdfile_close(register ARCHD *arcn, register int *fd)
1018 rdfile_close(arcn, fd)
1019 register ARCHD *arcn;
1024 * make sure the file is open
1035 * user wants last access time reset
1037 set_ftime(arcn->org_name, arcn->sb.st_mtime, arcn->sb.st_atime, 1);
1043 * read a file to calculate its crc. This is a real drag. Archive formats
1044 * that have this, end up reading the file twice (we have to write the
1045 * header WITH the crc before writing the file contents. Oh well...
1047 * 0 if was able to calculate the crc, -1 otherwise
1052 set_crc(register ARCHD *arcn, register int fd)
1056 register ARCHD *arcn;
1064 unsigned long crc = 0L;
1070 * hmm, no fd, should never happen. well no crc then.
1076 if ((size = (u_long)arcn->sb.st_blksize) > (u_long)sizeof(tbuf))
1077 size = (u_long)sizeof(tbuf);
1080 * read all the bytes we think that there are in the file. If the user
1081 * is trying to archive an active file, forget this file.
1084 if ((res = read(fd, tbuf, size)) <= 0)
1087 for (i = 0; i < res; ++i)
1088 crc += (tbuf[i] & 0xff);
1092 * safety check. we want to avoid archiving files that are active as
1093 * they can create inconsistant archive copies.
1095 if (cpcnt != arcn->sb.st_size)
1096 paxwarn(1, "File changed size %s", arcn->org_name);
1097 else if (fstat(fd, &sb) < 0)
1098 syswarn(1, errno, "Failed stat on %s", arcn->org_name);
1099 else if (arcn->sb.st_mtime != sb.st_mtime)
1100 paxwarn(1, "File %s was modified during read", arcn->org_name);
1101 else if (lseek(fd, (off_t)0L, SEEK_SET) < 0)
1102 syswarn(1, errno, "File rewind failed on: %s", arcn->org_name);