2 * Copyright (c) 1992 Keith Muller.
3 * Copyright (c) 1992, 1993
4 * The Regents of the University of California. All rights reserved.
6 * This code is derived from software contributed to Berkeley by
7 * Keith Muller of the University of California, San Diego.
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
17 * 3. All advertising materials mentioning features or use of this software
18 * must display the following acknowledgement:
19 * This product includes software developed by the University of
20 * California, Berkeley and its contributors.
21 * 4. Neither the name of the University nor the names of its contributors
22 * may be used to endorse or promote products derived from this software
23 * without specific prior written permission.
25 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
26 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
29 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
37 * @(#)file_subs.c 8.1 (Berkeley) 5/31/93
38 * $FreeBSD: src/bin/pax/file_subs.c,v 1.12.2.1 2001/08/01 05:03:11 obrien Exp $
39 * $DragonFly: src/bin/pax/file_subs.c,v 1.3 2003/09/21 04:24:17 drhodus Exp $
42 #include <sys/types.h>
57 mk_link (register char *,register struct stat *,register char *, int);
60 * routines that deal with file operations such as: creating, removing;
61 * and setting access modes, uid/gid and times of files
64 #define FILEBITS (S_ISVTX | S_IRWXU | S_IRWXG | S_IRWXO)
65 #define SETBITS (S_ISUID | S_ISGID)
66 #define ABITS (FILEBITS | SETBITS)
70 * Create and open a file.
72 * file descriptor or -1 for failure
77 file_creat(register ARCHD *arcn)
89 * assume file doesn't exist, so just try to create it, most times this
90 * works. We have to take special handling when the file does exist. To
91 * detect this, we use O_EXCL. For example when trying to create a
92 * file and a character device or fifo exists with the same name, we
93 * can accidently open the device by mistake (or block waiting to open)
94 * If we find that the open has failed, then figure spend the effort to
95 * figure out why. This strategy was found to have better average
96 * performance in common use than checking the file (and the path)
99 file_mode = arcn->sb.st_mode & FILEBITS;
100 if ((fd = open(arcn->name, O_WRONLY | O_CREAT | O_TRUNC | O_EXCL,
105 * the file seems to exist. First we try to get rid of it (found to be
106 * the second most common failure when traced). If this fails, only
107 * then we go to the expense to check and create the path to the file
109 if (unlnk_exist(arcn->name, arcn->type) != 0)
114 * try to open it again, if this fails, check all the nodes in
115 * the path and give it a final try. if chk_path() finds that
116 * it cannot fix anything, we will skip the last attempt
118 if ((fd = open(arcn->name, O_WRONLY | O_CREAT | O_TRUNC,
122 if (nodirs || chk_path(arcn->name,arcn->sb.st_uid,arcn->sb.st_gid) < 0) {
123 syswarn(1, oerrno, "Unable to create %s", arcn->name);
132 * Close file descriptor to a file just created by pax. Sets modes,
133 * ownership and times as required.
135 * 0 for success, -1 for failure
140 file_close(register ARCHD *arcn, int fd)
144 register ARCHD *arcn;
153 syswarn(0, errno, "Unable to close file descriptor on %s",
157 * set owner/groups first as this may strip off mode bits we want
158 * then set file permission modes. Then set file access and
159 * modification times.
162 res = set_ids(arcn->name, arcn->sb.st_uid, arcn->sb.st_gid);
165 * IMPORTANT SECURITY NOTE:
166 * if not preserving mode or we cannot set uid/gid, then PROHIBIT
170 arcn->sb.st_mode &= ~(SETBITS);
172 set_pmode(arcn->name, arcn->sb.st_mode);
173 if (patime || pmtime)
174 set_ftime(arcn->name, arcn->sb.st_mtime, arcn->sb.st_atime, 0);
179 * Create a hard link to arcn->ln_name from arcn->name. arcn->ln_name
182 * 0 if ok, -1 otherwise
187 lnk_creat(register ARCHD *arcn)
191 register ARCHD *arcn;
197 * we may be running as root, so we have to be sure that link target
198 * is not a directory, so we lstat and check
200 if (lstat(arcn->ln_name, &sb) < 0) {
201 syswarn(1,errno,"Unable to link to %s from %s", arcn->ln_name,
206 if (S_ISDIR(sb.st_mode)) {
207 paxwarn(1, "A hard link to the directory %s is not allowed",
212 return(mk_link(arcn->ln_name, &sb, arcn->name, 0));
217 * Create a hard link to arcn->org_name from arcn->name. Only used in copy
218 * with the -l flag. No warning or error if this does not succeed (we will
219 * then just create the file)
221 * 1 if copy() should try to create this file node
222 * 0 if cross_lnk() ok, -1 for fatal flaw (like linking to self).
227 cross_lnk(register ARCHD *arcn)
231 register ARCHD *arcn;
235 * try to make a link to original file (-l flag in copy mode). make sure
236 * we do not try to link to directories in case we are running as root
237 * (and it might succeed).
239 if (arcn->type == PAX_DIR)
241 return(mk_link(arcn->org_name, &(arcn->sb), arcn->name, 1));
246 * In copy mode if we are not trying to make hard links between the src
247 * and destinations, make sure we are not going to overwrite ourselves by
248 * accident. This slows things down a little, but we have to protect all
249 * those people who make typing errors.
251 * 1 the target does not exist, go ahead and copy
252 * 0 skip it file exists (-k) or may be the same as source file
257 chk_same(register ARCHD *arcn)
261 register ARCHD *arcn;
267 * if file does not exist, return. if file exists and -k, skip it
270 if (lstat(arcn->name, &sb) < 0)
276 * better make sure the user does not have src == dest by mistake
278 if ((arcn->sb.st_dev == sb.st_dev) && (arcn->sb.st_ino == sb.st_ino)) {
279 paxwarn(1, "Unable to copy %s, file would overwrite itself",
288 * try to make a hard link between two files. if ign set, we do not
291 * 0 if successful (or we are done with this file but no error, such as
292 * finding the from file exists and the user has set -k).
293 * 1 when ign was set to indicates we could not make the link but we
294 * should try to copy/extract the file as that might work (and is an
295 * allowed option). -1 an error occurred.
300 mk_link(register char *to, register struct stat *to_sb, register char *from,
304 mk_link(to, to_sb, from, ign)
306 register struct stat *to_sb;
315 * if from file exists, it has to be unlinked to make the link. If the
316 * file exists and -k is set, skip it quietly
318 if (lstat(from, &sb) == 0) {
323 * make sure it is not the same file, protect the user
325 if ((to_sb->st_dev==sb.st_dev)&&(to_sb->st_ino == sb.st_ino)) {
326 paxwarn(1, "Unable to link file %s to itself", to);
331 * try to get rid of the file, based on the type
333 if (S_ISDIR(sb.st_mode)) {
334 if (rmdir(from) < 0) {
335 syswarn(1, errno, "Unable to remove %s", from);
338 } else if (unlink(from) < 0) {
340 syswarn(1, errno, "Unable to remove %s", from);
348 * from file is gone (or did not exist), try to make the hard link.
349 * if it fails, check the path and try it again (if chk_path() says to
353 if (link(to, from) == 0)
356 if (!nodirs && chk_path(from, to_sb->st_uid, to_sb->st_gid) == 0)
359 syswarn(1, oerrno, "Could not link to %s from %s", to,
367 * all right the link was made
374 * create an entry in the file system (other than a file or hard link).
375 * If successful, sets uid/gid modes and times as required.
377 * 0 if ok, -1 otherwise
382 node_creat(register ARCHD *arcn)
386 register ARCHD *arcn;
390 register int ign = 0;
392 register int pass = 0;
397 * create node based on type, if that fails try to unlink the node and
398 * try again. finally check the path and try again. As noted in the
399 * file and link creation routines, this method seems to exhibit the
400 * best performance in general use workloads.
402 file_mode = arcn->sb.st_mode & FILEBITS;
407 res = mkdir(arcn->name, file_mode);
412 file_mode |= S_IFCHR;
413 res = mknod(arcn->name, file_mode, arcn->sb.st_rdev);
416 file_mode |= S_IFBLK;
417 res = mknod(arcn->name, file_mode, arcn->sb.st_rdev);
420 res = mkfifo(arcn->name, file_mode);
424 * Skip sockets, operation has no meaning under BSD
427 "%s skipped. Sockets cannot be copied or extracted",
431 res = symlink(arcn->ln_name, arcn->name);
439 * we should never get here
441 paxwarn(0, "%s has an unknown file type, skipping",
447 * if we were able to create the node break out of the loop,
448 * otherwise try to unlink the node and try again. if that
449 * fails check the full path and try a final time.
455 * we failed to make the node
458 if ((ign = unlnk_exist(arcn->name, arcn->type)) < 0)
464 if (nodirs || chk_path(arcn->name,arcn->sb.st_uid,arcn->sb.st_gid) < 0) {
465 syswarn(1, oerrno, "Could not create: %s", arcn->name);
471 * we were able to create the node. set uid/gid, modes and times
474 res = ((arcn->type == PAX_SLK) ?
475 set_lids(arcn->name, arcn->sb.st_uid, arcn->sb.st_gid) :
476 set_ids(arcn->name, arcn->sb.st_uid, arcn->sb.st_gid));
481 * symlinks are done now.
483 if (arcn->type == PAX_SLK)
487 * IMPORTANT SECURITY NOTE:
488 * if not preserving mode or we cannot set uid/gid, then PROHIBIT any
492 arcn->sb.st_mode &= ~(SETBITS);
494 set_pmode(arcn->name, arcn->sb.st_mode);
496 if (arcn->type == PAX_DIR && strcmp(NM_CPIO, argv0) != 0) {
498 * Dirs must be processed again at end of extract to set times
499 * and modes to agree with those stored in the archive. However
500 * to allow extract to continue, we may have to also set owner
501 * rights. This allows nodes in the archive that are children
502 * of this directory to be extracted without failure. Both time
503 * and modes will be fixed after the entire archive is read and
506 if (access(arcn->name, R_OK | W_OK | X_OK) < 0) {
507 if (lstat(arcn->name, &sb) < 0) {
508 syswarn(0, errno,"Could not access %s (stat)",
510 set_pmode(arcn->name,file_mode | S_IRWXU);
513 * We have to add rights to the dir, so we make
514 * sure to restore the mode. The mode must be
515 * restored AS CREATED and not as stored if
518 set_pmode(arcn->name,
519 ((sb.st_mode & FILEBITS) | S_IRWXU));
521 arcn->sb.st_mode = sb.st_mode;
525 * we have to force the mode to what was set here,
526 * since we changed it from the default as created.
528 add_dir(arcn->name, arcn->nlen, &(arcn->sb), 1);
529 } else if (pmode || patime || pmtime)
530 add_dir(arcn->name, arcn->nlen, &(arcn->sb), 0);
533 if (patime || pmtime)
534 set_ftime(arcn->name, 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.
552 unlnk_exist(register char *name, register int type)
555 unlnk_exist(name, type)
563 * the file does not exist, or -k we are done
565 if (lstat(name, &sb) < 0)
570 if (S_ISDIR(sb.st_mode)) {
572 * try to remove a directory, if it fails and we were going to
573 * create a directory anyway, tell the caller (return a 1)
575 if (rmdir(name) < 0) {
578 syswarn(1,errno,"Unable to remove directory %s", name);
585 * try to get rid of all non-directory type nodes
587 if (unlink(name) < 0) {
588 syswarn(1, errno, "Could not unlink %s", name);
596 * We were trying to create some kind of node in the file system and it
597 * failed. chk_path() makes sure the path up to the node exists and is
598 * writeable. When we have to create a directory that is missing along the
599 * path somewhere, the directory we create will be set to the same
600 * uid/gid as the file has (when uid and gid are being preserved).
601 * NOTE: this routine is a real performance loss. It is only used as a
602 * last resort when trying to create entries in the file system.
604 * -1 when it could find nothing it is allowed to fix.
610 chk_path( register char *name, uid_t st_uid, gid_t st_gid)
613 chk_path(name, st_uid, st_gid)
619 register char *spt = name;
624 * watch out for paths with nodes stored directly in / (e.g. /bozo)
631 * work foward from the first / and check each part of the path
633 spt = strchr(spt, '/');
639 * if it exists we assume it is a directory, it is not within
640 * the spec (at least it seems to read that way) to alter the
641 * file system for nodes NOT EXPLICITLY stored on the archive.
642 * If that assumption is changed, you would test the node here
643 * and figure out how to get rid of it (probably like some
644 * recursive unlink()) or fix up the directory permissions if
645 * required (do an access()).
647 if (lstat(name, &sb) == 0) {
653 * the path fails at this point, see if we can create the
654 * needed directory and continue on
656 if (mkdir(name, S_IRWXU | S_IRWXG | S_IRWXO) < 0) {
663 * we were able to create the directory. We will tell the
664 * caller that we found something to fix, and it is ok to try
665 * and create the node again.
669 (void)set_ids(name, st_uid, st_gid);
672 * make sure the user doen't have some strange umask that
673 * causes this newly created directory to be unusable. We fix
674 * the modes and restore them back to the creation default at
677 if ((access(name, R_OK | W_OK | X_OK) < 0) &&
678 (lstat(name, &sb) == 0)) {
679 set_pmode(name, ((sb.st_mode & FILEBITS) | S_IRWXU));
680 add_dir(name, spt - name, &sb, 1);
690 * Set the access time and modification time for a named file. If frc is
691 * non-zero we force these times to be set even if the user did not
692 * request access and/or modification time preservation (this is also
693 * used by -t to reset access times).
694 * When ign is zero, only those times the user has asked for are set, the
695 * other ones are left alone. We do not assume the un-documented feature
696 * of many utimes() implementations that consider a 0 time value as a do
702 set_ftime(char *fnm, time_t mtime, time_t atime, int frc)
705 set_ftime(fnm, mtime, atime, frc)
712 static struct timeval tv[2] = {{0L, 0L}, {0L, 0L}};
715 tv[0].tv_sec = (long)atime;
716 tv[1].tv_sec = (long)mtime;
717 if (!frc && (!patime || !pmtime)) {
719 * if we are not forcing, only set those times the user wants
720 * set. We get the current values of the times if we need them.
722 if (lstat(fnm, &sb) == 0) {
724 tv[0].tv_sec = (long)sb.st_atime;
726 tv[1].tv_sec = (long)sb.st_mtime;
728 syswarn(0,errno,"Unable to obtain file stats %s", fnm);
734 if (utimes(fnm, tv) < 0)
735 syswarn(1, errno, "Access/modification time set failed on: %s",
742 * set the uid and gid of a file system node
744 * 0 when set, -1 on failure
749 set_ids(char *fnm, uid_t uid, gid_t gid)
752 set_ids(fnm, uid, gid)
758 if (chown(fnm, uid, gid) < 0) {
760 * ignore EPERM unless in verbose mode or being run by root.
761 * if running as pax, POSIX requires a warning.
763 if (strcmp(NM_PAX, argv0) == 0 || errno != EPERM || vflag ||
765 syswarn(1, errno, "Unable to set file uid/gid of %s",
774 * set the uid and gid of a file system node
776 * 0 when set, -1 on failure
781 set_lids(char *fnm, uid_t uid, gid_t gid)
784 set_lids(fnm, uid, gid)
790 if (lchown(fnm, uid, gid) < 0) {
792 * ignore EPERM unless in verbose mode or being run by root.
793 * if running as pax, POSIX requires a warning.
795 if (strcmp(NM_PAX, argv0) == 0 || errno != EPERM || vflag ||
797 syswarn(1, errno, "Unable to set file uid/gid of %s",
806 * Set file access mode
811 set_pmode(char *fnm, mode_t mode)
820 if (chmod(fnm, mode) < 0)
821 syswarn(1, errno, "Could not set permissions on %s", fnm);
827 * Write/copy a file (during copy or archive extract). This routine knows
828 * how to copy files with lseek holes in it. (Which are read as file
829 * blocks containing all 0's but do not have any file blocks associated
830 * with the data). Typical examples of these are files created by dbm
831 * variants (.pag files). While the file size of these files are huge, the
832 * actual storage is quite small (the files are sparse). The problem is
833 * the holes read as all zeros so are probably stored on the archive that
834 * way (there is no way to determine if the file block is really a hole,
835 * we only know that a file block of all zero's can be a hole).
836 * At this writing, no major archive format knows how to archive files
837 * with holes. However, on extraction (or during copy, -rw) we have to
838 * deal with these files. Without detecting the holes, the files can
839 * consume a lot of file space if just written to disk. This replacement
840 * for write when passed the basic allocation size of a file system block,
841 * uses lseek whenever it detects the input data is all 0 within that
842 * file block. In more detail, the strategy is as follows:
843 * While the input is all zero keep doing an lseek. Keep track of when we
844 * pass over file block boundries. Only write when we hit a non zero
845 * input. once we have written a file block, we continue to write it to
846 * the end (we stop looking at the input). When we reach the start of the
847 * next file block, start checking for zero blocks again. Working on file
848 * block boundries significantly reduces the overhead when copying files
849 * that are NOT very sparse. This overhead (when compared to a write) is
850 * almost below the measurement resolution on many systems. Without it,
851 * files with holes cannot be safely copied. It does has a side effect as
852 * it can put holes into files that did not have them before, but that is
853 * not a problem since the file contents are unchanged (in fact it saves
854 * file space). (Except on paging files for diskless clients. But since we
855 * cannot determine one of those file from here, we ignore them). If this
856 * ever ends up on a system where CTG files are supported and the holes
857 * are not desired, just do a conditional test in those routines that
858 * call file_write() and have it call write() instead. BEFORE CLOSING THE
859 * FILE, make sure to call file_flush() when the last write finishes with
860 * an empty block. A lot of file systems will not create an lseek hole at
861 * the end. In this case we drop a single 0 at the end to force the
862 * trailing 0's in the file.
864 * rem: how many bytes left in this file system block
865 * isempt: have we written to the file block yet (is it empty)
866 * sz: basic file block allocation size
867 * cnt: number of bytes on this write
868 * str: buffer to write
870 * number of bytes written, -1 on write (or lseek) error.
875 file_write(int fd, char *str, register int cnt, int *rem, int *isempt, int sz,
879 file_write(fd, str, cnt, rem, isempt, sz, name)
892 register char *st = str;
895 * while we have data to process
900 * We are now at the start of file system block again
901 * (or what we think one is...). start looking for
909 * only examine up to the end of the current file block or
910 * remaining characters to write, whatever is smaller
912 wcnt = MIN(cnt, *rem);
917 * have not written to this block yet, so we keep
924 * look for a zero filled buffer
926 while ((pt < end) && (*pt == '\0'))
931 * skip, buf is empty so far
933 if (lseek(fd, (off_t)wcnt, SEEK_CUR) < 0) {
934 syswarn(1,errno,"File seek on %s",
942 * drat, the buf is not zero filled
948 * have non-zero data in this file system block, have to write
950 if (write(fd, st, wcnt) != wcnt) {
951 syswarn(1, errno, "Failed write to file %s", name);
961 * when the last file block in a file is zero, many file systems will not
962 * let us create a hole at the end. To get the last block with zeros, we
963 * write the last BYTE with a zero (back up one byte and write a zero).
968 file_flush(int fd, char *fname, int isempt)
971 file_flush(fd, fname, isempt)
977 static char blnk[] = "\0";
980 * silly test, but make sure we are only called when the last block is
981 * filled with all zeros.
987 * move back one byte and write a zero
989 if (lseek(fd, (off_t)-1, SEEK_CUR) < 0) {
990 syswarn(1, errno, "Failed seek on file %s", fname);
994 if (write(fd, blnk, 1) < 0)
995 syswarn(1, errno, "Failed write to file %s", fname);
1001 * close a file we have beed reading (to copy or archive). If we have to
1002 * reset access time (tflag) do so (the times are stored in arcn).
1007 rdfile_close(register ARCHD *arcn, register int *fd)
1010 rdfile_close(arcn, fd)
1011 register ARCHD *arcn;
1016 * make sure the file is open
1027 * user wants last access time reset
1029 set_ftime(arcn->org_name, arcn->sb.st_mtime, arcn->sb.st_atime, 1);
1035 * read a file to calculate its crc. This is a real drag. Archive formats
1036 * that have this, end up reading the file twice (we have to write the
1037 * header WITH the crc before writing the file contents. Oh well...
1039 * 0 if was able to calculate the crc, -1 otherwise
1044 set_crc(register ARCHD *arcn, register int fd)
1048 register ARCHD *arcn;
1056 unsigned long crc = 0L;
1062 * hmm, no fd, should never happen. well no crc then.
1068 if ((size = (u_long)arcn->sb.st_blksize) > (u_long)sizeof(tbuf))
1069 size = (u_long)sizeof(tbuf);
1072 * read all the bytes we think that there are in the file. If the user
1073 * is trying to archive an active file, forget this file.
1076 if ((res = read(fd, tbuf, size)) <= 0)
1079 for (i = 0; i < res; ++i)
1080 crc += (tbuf[i] & 0xff);
1084 * safety check. we want to avoid archiving files that are active as
1085 * they can create inconsistant archive copies.
1087 if (cpcnt != arcn->sb.st_size)
1088 paxwarn(1, "File changed size %s", arcn->org_name);
1089 else if (fstat(fd, &sb) < 0)
1090 syswarn(1, errno, "Failed stat on %s", arcn->org_name);
1091 else if (arcn->sb.st_mtime != sb.st_mtime)
1092 paxwarn(1, "File %s was modified during read", arcn->org_name);
1093 else if (lseek(fd, (off_t)0L, SEEK_SET) < 0)
1094 syswarn(1, errno, "File rewind failed on: %s", arcn->org_name);