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
39 static const char sccsid[] = "@(#)file_subs.c 8.1 (Berkeley) 5/31/93";
41 static const char rcsid[] = "$OpenBSD: file_subs.c,v 1.30 2005/11/09 19:59:06 otto Exp $";
45 #include <sys/param.h>
61 mk_link(char *, struct stat *, char *, int);
64 * routines that deal with file operations such as: creating, removing;
65 * and setting access modes, uid/gid and times of files
68 #define FILEBITS (S_ISVTX | S_IRWXU | S_IRWXG | S_IRWXO)
69 #define SETBITS (S_ISUID | S_ISGID)
70 #define ABITS (FILEBITS | SETBITS)
74 * Create and open a file.
76 * file descriptor or -1 for failure
80 file_creat(ARCHD *arcn)
87 * Assume file doesn't exist, so just try to create it, most times this
88 * works. We have to take special handling when the file does exist. To
89 * detect this, we use O_EXCL. For example when trying to create a
90 * file and a character device or fifo exists with the same name, we
91 * can accidently open the device by mistake (or block waiting to open).
92 * If we find that the open has failed, then spend the effort to
93 * figure out why. This strategy was found to have better average
94 * performance in common use than checking the file (and the path)
97 file_mode = arcn->sb.st_mode & FILEBITS;
98 if ((fd = open(arcn->name, O_WRONLY | O_CREAT | O_TRUNC | O_EXCL,
103 * the file seems to exist. First we try to get rid of it (found to be
104 * the second most common failure when traced). If this fails, only
105 * then we go to the expense to check and create the path to the file
107 if (unlnk_exist(arcn->name, arcn->type) != 0)
112 * try to open it again, if this fails, check all the nodes in
113 * the path and give it a final try. if chk_path() finds that
114 * it cannot fix anything, we will skip the last attempt
116 if ((fd = open(arcn->name, O_WRONLY | O_CREAT | O_TRUNC,
120 if (nodirs || chk_path(arcn->name,arcn->sb.st_uid,arcn->sb.st_gid) < 0) {
121 syswarn(1, oerrno, "Unable to create %s", arcn->name);
130 * Close file descriptor to a file just created by pax. Sets modes,
131 * ownership and times as required.
133 * 0 for success, -1 for failure
137 file_close(ARCHD *arcn, int fd)
145 * set owner/groups first as this may strip off mode bits we want
146 * then set file permission modes. Then set file access and
147 * modification times.
150 res = fset_ids(arcn->name, fd, arcn->sb.st_uid,
154 * IMPORTANT SECURITY NOTE:
155 * if not preserving mode or we cannot set uid/gid, then PROHIBIT
159 arcn->sb.st_mode &= ~(SETBITS);
161 fset_pmode(arcn->name, fd, arcn->sb.st_mode);
162 if (patime || pmtime)
163 fset_ftime(arcn->name, fd, arcn->sb.st_mtime,
164 arcn->sb.st_atime, 0);
166 syswarn(0, errno, "Unable to close file descriptor on %s",
172 * Create a hard link to arcn->ln_name from arcn->name. arcn->ln_name
175 * 0 if ok, -1 otherwise
179 lnk_creat(ARCHD *arcn)
184 * we may be running as root, so we have to be sure that link target
185 * is not a directory, so we lstat and check
187 if (lstat(arcn->ln_name, &sb) < 0) {
188 syswarn(1,errno,"Unable to link to %s from %s", arcn->ln_name,
193 if (S_ISDIR(sb.st_mode)) {
194 paxwarn(1, "A hard link to the directory %s is not allowed",
199 return(mk_link(arcn->ln_name, &sb, arcn->name, 0));
204 * Create a hard link to arcn->org_name from arcn->name. Only used in copy
205 * with the -l flag. No warning or error if this does not succeed (we will
206 * then just create the file)
208 * 1 if copy() should try to create this file node
209 * 0 if cross_lnk() ok, -1 for fatal flaw (like linking to self).
213 cross_lnk(ARCHD *arcn)
216 * try to make a link to original file (-l flag in copy mode). make
217 * sure we do not try to link to directories in case we are running as
218 * root (and it might succeed).
220 if (arcn->type == PAX_DIR)
222 return(mk_link(arcn->org_name, &(arcn->sb), arcn->name, 1));
227 * In copy mode if we are not trying to make hard links between the src
228 * and destinations, make sure we are not going to overwrite ourselves by
229 * accident. This slows things down a little, but we have to protect all
230 * those people who make typing errors.
232 * 1 the target does not exist, go ahead and copy
233 * 0 skip it file exists (-k) or may be the same as source file
237 chk_same(ARCHD *arcn)
242 * if file does not exist, return. if file exists and -k, skip it
245 if (lstat(arcn->name, &sb) < 0)
251 * better make sure the user does not have src == dest by mistake
253 if ((arcn->sb.st_dev == sb.st_dev) && (arcn->sb.st_ino == sb.st_ino)) {
254 paxwarn(1, "Unable to copy %s, file would overwrite itself",
263 * try to make a hard link between two files. if ign set, we do not
266 * 0 if successful (or we are done with this file but no error, such as
267 * finding the from file exists and the user has set -k).
268 * 1 when ign was set to indicates we could not make the link but we
269 * should try to copy/extract the file as that might work (and is an
270 * allowed option). -1 an error occurred.
274 mk_link(char *to, struct stat *to_sb, char *from, int ign)
280 * if from file exists, it has to be unlinked to make the link. If the
281 * file exists and -k is set, skip it quietly
283 if (lstat(from, &sb) == 0) {
288 * make sure it is not the same file, protect the user
290 if ((to_sb->st_dev==sb.st_dev)&&(to_sb->st_ino == sb.st_ino)) {
291 paxwarn(1, "Unable to link file %s to itself", to);
296 * try to get rid of the file, based on the type
298 if (S_ISDIR(sb.st_mode)) {
299 if (rmdir(from) < 0) {
300 syswarn(1, errno, "Unable to remove %s", from);
303 } else if (unlink(from) < 0) {
305 syswarn(1, errno, "Unable to remove %s", from);
313 * from file is gone (or did not exist), try to make the hard link.
314 * if it fails, check the path and try it again (if chk_path() says to
318 if (link(to, from) == 0)
321 if (!nodirs && chk_path(from, to_sb->st_uid, to_sb->st_gid) == 0)
324 syswarn(1, oerrno, "Could not link to %s from %s", to,
332 * all right the link was made
339 * create an entry in the file system (other than a file or hard link).
340 * If successful, sets uid/gid modes and times as required.
342 * 0 if ok, -1 otherwise
346 node_creat(ARCHD *arcn)
355 char *nm = arcn->name;
359 * create node based on type, if that fails try to unlink the node and
360 * try again. finally check the path and try again. As noted in the
361 * file and link creation routines, this method seems to exhibit the
362 * best performance in general use workloads.
364 file_mode = arcn->sb.st_mode & FILEBITS;
367 switch (arcn->type) {
370 * If -h (or -L) was given in tar-mode, follow the
371 * potential symlink chain before trying to create the
374 if (strcmp(NM_TAR, argv0) == 0 && Lflag) {
375 while (lstat(nm, &sb) == 0 &&
376 S_ISLNK(sb.st_mode)) {
377 target = malloc(sb.st_size + 1);
378 if (target == NULL) {
381 "Insufficient memory");
384 len = readlink(nm, target,
388 "cannot follow symlink %s in chain for %s",
397 res = mkdir(nm, file_mode);
404 file_mode |= S_IFCHR;
405 res = mknod(nm, file_mode, arcn->sb.st_rdev);
408 file_mode |= S_IFBLK;
409 res = mknod(nm, file_mode, arcn->sb.st_rdev);
412 res = mkfifo(nm, file_mode);
416 * Skip sockets, operation has no meaning under BSD
419 "%s skipped. Sockets cannot be copied or extracted",
424 res = symlink(arcn->ln_name, nm);
432 * we should never get here
434 paxwarn(0, "%s has an unknown file type, skipping",
441 * if we were able to create the node break out of the loop,
442 * otherwise try to unlink the node and try again. if that
443 * fails check the full path and try a final time.
449 * we failed to make the node
452 if ((ign = unlnk_exist(nm, arcn->type)) < 0) {
460 if (nodirs || chk_path(nm,arcn->sb.st_uid,arcn->sb.st_gid) < 0) {
461 syswarn(1, oerrno, "Could not create: %s", nm);
468 * we were able to create the node. set uid/gid, modes and times
471 res = ((arcn->type == PAX_SLK) ?
472 set_lids(nm, arcn->sb.st_uid, arcn->sb.st_gid) :
473 set_ids(nm, arcn->sb.st_uid, arcn->sb.st_gid));
478 * symlinks are done now.
480 if (arcn->type == PAX_SLK) {
486 * IMPORTANT SECURITY NOTE:
487 * if not preserving mode or we cannot set uid/gid, then PROHIBIT any
491 arcn->sb.st_mode &= ~(SETBITS);
493 set_pmode(nm, arcn->sb.st_mode);
495 if (arcn->type == PAX_DIR && strcmp(NM_CPIO, argv0) != 0) {
497 * Dirs must be processed again at end of extract to set times
498 * and modes to agree with those stored in the archive. However
499 * to allow extract to continue, we may have to also set owner
500 * rights. This allows nodes in the archive that are children
501 * of this directory to be extracted without failure. Both time
502 * and modes will be fixed after the entire archive is read and
505 if (access(nm, R_OK | W_OK | X_OK) < 0) {
506 if (lstat(nm, &sb) < 0) {
507 syswarn(0, errno,"Could not access %s (stat)",
509 set_pmode(nm,file_mode | S_IRWXU);
512 * We have to add rights to the dir, so we make
513 * sure to restore the mode. The mode must be
514 * restored AS CREATED and not as stored if
518 ((sb.st_mode & FILEBITS) | S_IRWXU));
520 arcn->sb.st_mode = sb.st_mode;
524 * we have to force the mode to what was set here,
525 * since we changed it from the default as created.
527 add_dir(nm, &(arcn->sb), 1);
528 } else if (pmode || patime || pmtime)
529 add_dir(nm, &(arcn->sb), 0);
532 if (patime || pmtime)
533 set_ftime(nm, arcn->sb.st_mtime, arcn->sb.st_atime, 0);
540 * Remove node from file system with the specified name. We pass the type
541 * of the node that is going to replace it. When we try to create a
542 * directory and find that it already exists, we allow processing to
543 * continue as proper modes etc will always be set for it later on.
545 * 0 is ok to proceed, no file with the specified name exists
546 * -1 we were unable to remove the node, or we should not remove it (-k)
547 * 1 we found a directory and we were going to create a directory.
551 unlnk_exist(char *name, int type)
556 * the file does not exist, or -k we are done
558 if (lstat(name, &sb) < 0)
563 if (S_ISDIR(sb.st_mode)) {
565 * try to remove a directory, if it fails and we were going to
566 * create a directory anyway, tell the caller (return a 1)
568 if (rmdir(name) < 0) {
571 syswarn(1,errno,"Unable to remove directory %s", name);
578 * try to get rid of all non-directory type nodes
580 if (unlink(name) < 0) {
581 syswarn(1, errno, "Could not unlink %s", name);
589 * We were trying to create some kind of node in the file system and it
590 * failed. chk_path() makes sure the path up to the node exists and is
591 * writeable. When we have to create a directory that is missing along the
592 * path somewhere, the directory we create will be set to the same
593 * uid/gid as the file has (when uid and gid are being preserved).
594 * NOTE: this routine is a real performance loss. It is only used as a
595 * last resort when trying to create entries in the file system.
597 * -1 when it could find nothing it is allowed to fix.
602 chk_path(char *name, uid_t st_uid, gid_t st_gid)
609 * watch out for paths with nodes stored directly in / (e.g. /bozo)
616 * work forward from the first / and check each part of the path
618 spt = strchr(spt, '/');
624 * if it exists we assume it is a directory, it is not within
625 * the spec (at least it seems to read that way) to alter the
626 * file system for nodes NOT EXPLICITLY stored on the archive.
627 * If that assumption is changed, you would test the node here
628 * and figure out how to get rid of it (probably like some
629 * recursive unlink()) or fix up the directory permissions if
630 * required (do an access()).
632 if (lstat(name, &sb) == 0) {
638 * the path fails at this point, see if we can create the
639 * needed directory and continue on
641 if (mkdir(name, S_IRWXU | S_IRWXG | S_IRWXO) < 0) {
648 * we were able to create the directory. We will tell the
649 * caller that we found something to fix, and it is ok to try
650 * and create the node again.
654 (void)set_ids(name, st_uid, st_gid);
657 * make sure the user doesn't have some strange umask that
658 * causes this newly created directory to be unusable. We fix
659 * the modes and restore them back to the creation default at
662 if ((access(name, R_OK | W_OK | X_OK) < 0) &&
663 (lstat(name, &sb) == 0)) {
664 set_pmode(name, ((sb.st_mode & FILEBITS) | S_IRWXU));
665 add_dir(name, &sb, 1);
675 * Set the access time and modification time for a named file. If frc
676 * is non-zero we force these times to be set even if the user did not
677 * request access and/or modification time preservation (this is also
678 * used by -t to reset access times).
679 * When ign is zero, only those times the user has asked for are set, the
680 * other ones are left alone. We do not assume the un-documented feature
681 * of many utimes() implementations that consider a 0 time value as a do
686 set_ftime(char *fnm, time_t mtime, time_t atime, int frc)
688 static struct timeval tv[2] = {{0L, 0L}, {0L, 0L}};
691 tv[0].tv_sec = (long)atime;
692 tv[1].tv_sec = (long)mtime;
693 if (!frc && (!patime || !pmtime)) {
695 * if we are not forcing, only set those times the user wants
696 * set. We get the current values of the times if we need them.
698 if (lstat(fnm, &sb) == 0) {
700 tv[0].tv_sec = (long)sb.st_atime;
702 tv[1].tv_sec = (long)sb.st_mtime;
704 syswarn(0,errno,"Unable to obtain file stats %s", fnm);
710 if (utimes(fnm, tv) < 0)
711 syswarn(1, errno, "Access/modification time set failed on: %s",
717 fset_ftime(char *fnm, int fd, time_t mtime, time_t atime, int frc)
719 static struct timeval tv[2] = {{0L, 0L}, {0L, 0L}};
722 tv[0].tv_sec = (long)atime;
723 tv[1].tv_sec = (long)mtime;
724 if (!frc && (!patime || !pmtime)) {
726 * if we are not forcing, only set those times the user wants
727 * set. We get the current values of the times if we need them.
729 if (fstat(fd, &sb) == 0) {
731 tv[0].tv_sec = (long)sb.st_atime;
733 tv[1].tv_sec = (long)sb.st_mtime;
735 syswarn(0,errno,"Unable to obtain file stats %s", fnm);
740 if (futimes(fd, tv) < 0)
741 syswarn(1, errno, "Access/modification time set failed on: %s",
748 * set the uid and gid of a file system node
750 * 0 when set, -1 on failure
754 set_ids(char *fnm, uid_t uid, gid_t gid)
756 if (chown(fnm, uid, gid) < 0) {
758 * ignore EPERM unless in verbose mode or being run by root.
759 * if running as pax, POSIX requires a warning.
761 if (strcmp(NM_PAX, argv0) == 0 || errno != EPERM || vflag ||
763 syswarn(1, errno, "Unable to set file uid/gid of %s",
771 fset_ids(char *fnm, int fd, uid_t uid, gid_t gid)
773 if (fchown(fd, uid, gid) < 0) {
775 * ignore EPERM unless in verbose mode or being run by root.
776 * if running as pax, POSIX requires a warning.
778 if (strcmp(NM_PAX, argv0) == 0 || errno != EPERM || vflag ||
780 syswarn(1, errno, "Unable to set file uid/gid of %s",
789 * set the uid and gid of a file system node
791 * 0 when set, -1 on failure
795 set_lids(char *fnm, uid_t uid, gid_t gid)
797 if (lchown(fnm, uid, gid) < 0) {
799 * ignore EPERM unless in verbose mode or being run by root.
800 * if running as pax, POSIX requires a warning.
802 if (strcmp(NM_PAX, argv0) == 0 || errno != EPERM || vflag ||
804 syswarn(1, errno, "Unable to set file uid/gid of %s",
813 * Set file access mode
817 set_pmode(char *fnm, mode_t mode)
820 if (chmod(fnm, mode) < 0)
821 syswarn(1, errno, "Could not set permissions on %s", fnm);
826 fset_pmode(char *fnm, int fd, mode_t mode)
829 if (fchmod(fd, mode) < 0)
830 syswarn(1, errno, "Could not set permissions on %s", fnm);
836 * Write/copy a file (during copy or archive extract). This routine knows
837 * how to copy files with lseek holes in it. (Which are read as file
838 * blocks containing all 0's but do not have any file blocks associated
839 * with the data). Typical examples of these are files created by dbm
840 * variants (.pag files). While the file size of these files are huge, the
841 * actual storage is quite small (the files are sparse). The problem is
842 * the holes read as all zeros so are probably stored on the archive that
843 * way (there is no way to determine if the file block is really a hole,
844 * we only know that a file block of all zero's can be a hole).
845 * At this writing, no major archive format knows how to archive files
846 * with holes. However, on extraction (or during copy, -rw) we have to
847 * deal with these files. Without detecting the holes, the files can
848 * consume a lot of file space if just written to disk. This replacement
849 * for write when passed the basic allocation size of a file system block,
850 * uses lseek whenever it detects the input data is all 0 within that
851 * file block. In more detail, the strategy is as follows:
852 * While the input is all zero keep doing an lseek. Keep track of when we
853 * pass over file block boundaries. Only write when we hit a non zero
854 * input. once we have written a file block, we continue to write it to
855 * the end (we stop looking at the input). When we reach the start of the
856 * next file block, start checking for zero blocks again. Working on file
857 * block boundaries significantly reduces the overhead when copying files
858 * that are NOT very sparse. This overhead (when compared to a write) is
859 * almost below the measurement resolution on many systems. Without it,
860 * files with holes cannot be safely copied. It does has a side effect as
861 * it can put holes into files that did not have them before, but that is
862 * not a problem since the file contents are unchanged (in fact it saves
863 * file space). (Except on paging files for diskless clients. But since we
864 * cannot determine one of those file from here, we ignore them). If this
865 * ever ends up on a system where CTG files are supported and the holes
866 * are not desired, just do a conditional test in those routines that
867 * call file_write() and have it call write() instead. BEFORE CLOSING THE
868 * FILE, make sure to call file_flush() when the last write finishes with
869 * an empty block. A lot of file systems will not create an lseek hole at
870 * the end. In this case we drop a single 0 at the end to force the
871 * trailing 0's in the file.
873 * rem: how many bytes left in this file system block
874 * isempt: have we written to the file block yet (is it empty)
875 * sz: basic file block allocation size
876 * cnt: number of bytes on this write
877 * str: buffer to write
879 * number of bytes written, -1 on write (or lseek) error.
883 file_write(int fd, char *str, int cnt, int *rem, int *isempt, int sz,
893 * while we have data to process
898 * We are now at the start of file system block again
899 * (or what we think one is...). start looking for
907 * only examine up to the end of the current file block or
908 * remaining characters to write, whatever is smaller
910 wcnt = MIN(cnt, *rem);
915 * have not written to this block yet, so we keep
922 * look for a zero filled buffer
924 while ((pt < end) && (*pt == '\0'))
929 * skip, buf is empty so far
932 lseek(fd, (off_t)wcnt, SEEK_CUR) < 0) {
933 syswarn(1,errno,"File seek on %s",
941 * drat, the buf is not zero filled
947 * have non-zero data in this file system block, have to write
951 strp = &gnu_name_string;
954 strp = &gnu_link_string;
962 err(1, "WARNING! Major Internal Error! GNU hack Failing!");
963 *strp = malloc(wcnt + 1);
965 paxwarn(1, "Out of memory");
968 memcpy(*strp, st, wcnt);
969 (*strp)[wcnt] = '\0';
971 } else if (write(fd, st, wcnt) != wcnt) {
972 syswarn(1, errno, "Failed write to file %s", name);
982 * when the last file block in a file is zero, many file systems will not
983 * let us create a hole at the end. To get the last block with zeros, we
984 * write the last BYTE with a zero (back up one byte and write a zero).
988 file_flush(int fd, char *fname, int isempt)
990 static char blnk[] = "\0";
993 * silly test, but make sure we are only called when the last block is
994 * filled with all zeros.
1000 * move back one byte and write a zero
1002 if (lseek(fd, (off_t)-1, SEEK_CUR) < 0) {
1003 syswarn(1, errno, "Failed seek on file %s", fname);
1007 if (write(fd, blnk, 1) < 0)
1008 syswarn(1, errno, "Failed write to file %s", fname);
1014 * close a file we have beed reading (to copy or archive). If we have to
1015 * reset access time (tflag) do so (the times are stored in arcn).
1019 rdfile_close(ARCHD *arcn, int *fd)
1022 * make sure the file is open
1033 * user wants last access time reset
1035 set_ftime(arcn->org_name, arcn->sb.st_mtime, arcn->sb.st_atime, 1);
1041 * read a file to calculate its crc. This is a real drag. Archive formats
1042 * that have this, end up reading the file twice (we have to write the
1043 * header WITH the crc before writing the file contents. Oh well...
1045 * 0 if was able to calculate the crc, -1 otherwise
1049 set_crc(ARCHD *arcn, int fd)
1061 * hmm, no fd, should never happen. well no crc then.
1067 if ((size = (u_long)arcn->sb.st_blksize) > (u_long)sizeof(tbuf))
1068 size = (u_long)sizeof(tbuf);
1071 * read all the bytes we think that there are in the file. If the user
1072 * is trying to archive an active file, forget this file.
1075 if ((res = read(fd, tbuf, size)) <= 0)
1078 for (i = 0; i < res; ++i)
1079 crc += (tbuf[i] & 0xff);
1083 * safety check. we want to avoid archiving files that are active as
1084 * they can create inconsistent archive copies.
1086 if (cpcnt != arcn->sb.st_size)
1087 paxwarn(1, "File changed size %s", arcn->org_name);
1088 else if (fstat(fd, &sb) < 0)
1089 syswarn(1, errno, "Failed stat on %s", arcn->org_name);
1090 else if (arcn->sb.st_mtime != sb.st_mtime)
1091 paxwarn(1, "File %s was modified during read", arcn->org_name);
1092 else if (lseek(fd, (off_t)0L, SEEK_SET) < 0)
1093 syswarn(1, errno, "File rewind failed on: %s", arcn->org_name);