file_subs.c

早期freebsd实现

	return(0);
}

/*
 * unlnk_exist()
 *	Remove node from file system with the specified name. We pass the type
 *	of the node that is going to replace it. When we try to create a
 *	directory and find that it already exists, we allow processing to
 *	continue as proper modes etc will always be set for it later on.
 * Return:
 *	0 is ok to proceed, no file with the specified name exists
 *	-1 we were unable to remove the node, or we should not remove it (-k)
 *	1 we found a directory and we were going to create a directory.
 */

#if __STDC__
int
unlnk_exist(register char *name, register int type)
#else
int
unlnk_exist(name, type)
	register char *name;
	register int type;
#endif
{
	struct stat sb;

	/*
	 * the file does not exist, or -k we are done
	 */
	if (lstat(name, &sb) < 0)
		return(0);
	if (kflag)
		return(-1);

	if (S_ISDIR(sb.st_mode)) {
		/*
		 * try to remove a directory, if it fails and we were going to
		 * create a directory anyway, tell the caller (return a 1)
		 */
		if (rmdir(name) < 0) {
			if (type == PAX_DIR)
				return(1); 
			syswarn(1,errno,"Unable to remove directory %s", name);
			return(-1);
		}
		return(0);
	}

	/*
	 * try to get rid of all non-directory type nodes
	 */
	if (unlink(name) < 0) {
		syswarn(1, errno, "Could not unlink %s", name);
		return(-1);
	}
	return(0);
}

/*
 * chk_path()
 *	We were trying to create some kind of node in the file system and it
 *	failed. chk_path() makes sure the path up to the node exists and is
 *	writeable. When we have to create a directory that is missing along the
 *	path somewhere, the directory we create will be set to the same
 *	uid/gid as the file has (when uid and gid are being preserved).
 *	NOTE: this routine is a real performance loss. It is only used as a
 *	last resort when trying to create entries in the file system.
 * Return:
 *	-1 when it could find nothing it is allowed to fix.
 *	0 otherwise
 */

#if __STDC__
int
chk_path( register char *name, uid_t st_uid, gid_t st_gid)
#else
int
chk_path(name, st_uid, st_gid)
	register char *name;
	uid_t st_uid;
	gid_t st_gid;
#endif
{
	register char *spt = name;
	struct stat sb;
	int retval = -1;

	/*
	 * watch out for paths with nodes stored directly in / (e.g. /bozo)
	 */
	if (*spt == '/')
		++spt;

	for(;;) {
		/*
		 * work foward from the first / and check each part of the path
		 */
		spt = strchr(spt, '/');
		if (spt == NULL)
			break;
		*spt = '\0';

		/*
		 * if it exists we assume it is a directory, it is not within
		 * the spec (at least it seems to read that way) to alter the
		 * file system for nodes NOT EXPLICITLY stored on the archive.
		 * If that assumption is changed, you would test the node here
		 * and figure out how to get rid of it (probably like some
		 * recursive unlink()) or fix up the directory permissions if
		 * required (do an access()).
		 */
		if (lstat(name, &sb) == 0) {
			*(spt++) = '/';
			continue;
		}

		/*
		 * the path fails at this point, see if we can create the
		 * needed directory and continue on
		 */
		if (mkdir(name, S_IRWXU | S_IRWXG | S_IRWXO) < 0) {
			*spt = '/';
			retval = -1;
			break;
		}

		/*
		 * we were able to create the directory. We will tell the
		 * caller that we found something to fix, and it is ok to try
		 * and create the node again.
		 */
		retval = 0;
		if (pids)
			(void)set_ids(name, st_uid, st_gid);

		/*
		 * make sure the user doen't have some strange umask that
		 * causes this newly created directory to be unusable. We fix
		 * the modes and restore them back to the creation default at
		 * the end of pax
		 */
		if ((access(name, R_OK | W_OK | X_OK) < 0) &&
		    (lstat(name, &sb) == 0)) {
			set_pmode(name, ((sb.st_mode & FILEBITS) | S_IRWXU));
			add_dir(name, spt - name, &sb, 1);
		}
		*(spt++) = '/';
		continue;
	}
	return(retval);
}

/*
 * set_ftime()
 *	Set the access time and modification time for a named file. If frc is
 *	non-zero we force these times to be set even if the the user did not
 *	request access and/or modification time preservation (this is also
 *	used by -t to reset access times).
 *	When ign is zero, only those times the user has asked for are set, the
 *	other ones are left alone. We do not assume the un-documented feature
 *	of many utimes() implementations that consider a 0 time value as a do
 *	not set request.
 */

#if __STDC__
void
set_ftime(char *fnm, time_t mtime, time_t atime, int frc)
#else
void
set_ftime(fnm, mtime, atime, frc)
	char *fnm;
	time_t mtime;
	time_t atime;
	int frc;
#endif
{
	static struct timeval tv[2] = {{0L, 0L}, {0L, 0L}};
	struct stat sb;

	tv[0].tv_sec = (long)atime;
	tv[1].tv_sec = (long)mtime;
	if (!frc && (!patime || !pmtime)) {
		/*
		 * if we are not forcing, only set those times the user wants
		 * set. We get the current values of the times if we need them.
		 */
		if (lstat(fnm, &sb) == 0) {
			if (!patime)
				tv[0].tv_sec = (long)sb.st_atime;
			if (!pmtime)
				tv[1].tv_sec = (long)sb.st_mtime;
		} else
			syswarn(0,errno,"Unable to obtain file stats %s", fnm);
	}

	/*
	 * set the times
	 */
	if (utimes(fnm, tv) < 0)
		syswarn(1, errno, "Access/modification time set failed on: %s",
		    fnm);
	return;
}

/*
 * set_ids()
 *	set the uid and gid of a file system node
 * Return:
 *	0 when set, -1 on failure
 */

#if __STDC__
int
set_ids(char *fnm, uid_t uid, gid_t gid)
#else
int
set_ids(fnm, uid, gid)
	char *fnm;
	uid_t uid;
	gid_t gid;
#endif
{
	if (chown(fnm, uid, gid) < 0) {
		syswarn(1, errno, "Unable to set file uid/gid of %s", fnm);
		return(-1);
	}
	return(0);
}

/*
 * set_pmode()
 *	Set file access mode
 */

#if __STDC__
void
set_pmode(char *fnm, mode_t mode)
#else
void
set_pmode(fnm, mode)
	char *fnm;
	mode_t mode;
#endif
{
	mode &= ABITS;
	if (chmod(fnm, mode) < 0)
		syswarn(1, errno, "Could not set permissions on %s", fnm);
	return;
}

/*
 * file_write()
 *	Write/copy a file (during copy or archive extract). This routine knows
 *	how to copy files with lseek holes in it. (Which are read as file
 *	blocks containing all 0's but do not have any file blocks associated
 *	with the data). Typical examples of these are files created by dbm
 *	variants (.pag files). While the file size of these files are huge, the
 *	actual storage is quite small (the files are sparse). The problem is
 *	the holes read as all zeros so are probably stored on the archive that
 *	way (there is no way to determine if the file block is really a hole,
 *	we only know that a file block of all zero's can be a hole).
 *	At this writing, no major archive format knows how to archive files
 *	with holes. However, on extraction (or during copy, -rw) we have to
 *	deal with these files. Without detecting the holes, the files can
 *	consume a lot of file space if just written to disk. This replacement
 *	for write when passed the basic allocation size of a file system block,
 *	uses lseek whenever it detects the input data is all 0 within that
 *	file block. In more detail, the strategy is as follows:
 *	While the input is all zero keep doing an lseek. Keep track of when we
 *	pass over file block boundries. Only write when we hit a non zero
 *	input. once we have written a file block, we continue to write it to
 *	the end (we stop looking at the input). When we reach the start of the
 *	next file block, start checking for zero blocks again. Working on file
 *	block boundries significantly reduces the overhead when copying files
 *	that are NOT very sparse. This overhead (when compared to a write) is
 *	almost below the measurement resolution on many systems. Without it,
 *	files with holes cannot be safely copied. It does has a side effect as
 *	it can put holes into files that did not have them before, but that is
 *	not a problem since the file contents are unchanged (in fact it saves
 *	file space). (Except on paging files for diskless clients. But since we
 *	cannot determine one of those file from here, we ignore them). If this
 *	ever ends up on a system where CTG files are supported and the holes
 *	are not desired, just do a conditional test in those routines that
 *	call file_write() and have it call write() instead. BEFORE CLOSING THE
 *	FILE, make sure to call file_flush() when the last write finishes with
 *	an empty block. A lot of file systems will not create an lseek hole at
 *	the end. In this case we drop a single 0 at the end to force the
 *	trailing 0's in the file.
 *	---Parameters---
 *	rem: how many bytes left in this file system block
 *	isempt: have we written to the file block yet (is it empty)
 *	sz: basic file block allocation size
 *	cnt: number of bytes on this write
 *	str: buffer to write
 * Return:
 *	number of bytes written, -1 on write (or lseek) error.
 */

#if __STDC__
int
file_write(int fd, char *str, register int cnt, int *rem, int *isempt, int sz,
	char *name)
#else
int
file_write(fd, str, cnt, rem, isempt, sz, name)
	int fd;
	char *str;
	register int cnt;
	int *rem;
	int *isempt;
	int sz;
	char *name;
#endif
{
	register char *pt;
	register char *end;
	register int wcnt;
	register char *st = str;
	
	/*
	 * while we have data to process
	 */
	while (cnt) {
		if (!*rem) {
			/*
			 * We are now at the start of file system block again
			 * (or what we think one is...). start looking for
			 * empty blocks again
			 */
			*isempt = 1;
			*rem = sz;
		}

		/*
		 * only examine up to the end of the current file block or
		 * remaining characters to write, whatever is smaller
		 */
		wcnt = MIN(cnt, *rem);
		cnt -= wcnt;
		*rem -= wcnt;
		if (*isempt) {
			/*
			 * have not written to this block yet, so we keep
			 * looking for zero's
			 */
			pt = st;
			end = st + wcnt;

			/*
			 * look for a zero filled buffer
			 */
			while ((pt < end) && (*pt == '\0'))
				++pt;

			if (pt == end) {
				/*
				 * skip, buf is empty so far
				 */
				if (lseek(fd, (off_t)wcnt, SEEK_CUR) < 0) {
					syswarn(1,errno,"File seek on %s",
					    name);
					return(-1);
				}
				st = pt;
				continue;
			}
			/*
			 * drat, the buf is not zero filled
			 */
			*isempt = 0;
		}

		/*
		 * have non-zero data in this file system block, have to write
		 */
		if (write(fd, st, wcnt) != wcnt) {
			syswarn(1, errno, "Failed write to file %s", name);
			return(-1);
		}
		st += wcnt;
	}
	return(st - str);
}

/*
 * file_flush()
 *	when the last file block in a file is zero, many file systems will not
 *	let us create a hole at the end. To get the last block with zeros, we
 *	write the last BYTE with a zero (back up one byte and write a zero).
 */

#if __STDC__
void
file_flush(int fd, char *fname, int isempt)
#else
void
file_flush(fd, fname, isempt)
	int fd;
	char *fname;
	int isempt;
#endif
{
	static char blnk[] = "\0";

	/*
	 * silly test, but make sure we are only called when the last block is
	 * filled with all zeros.
	 */
	if (!isempt)
		return;

	/*
	 * move back one byte and write a zero
	 */
	if (lseek(fd, (off_t)-1, SEEK_CUR) < 0) {
		syswarn(1, errno, "Failed seek on file %s", fname);
		return;
	}

	if (write(fd, blnk, 1) < 0)
		syswarn(1, errno, "Failed write to file %s", fname);
	return;
}

/*
 * rdfile_close()
 *	close a file we have beed reading (to copy or archive). If we have to
 *	reset access time (tflag) do so (the times are stored in arcn).
 */

#if __STDC__
void
rdfile_close(register ARCHD *arcn, register int *fd)
#else
void
rdfile_close(arcn, fd)
	register ARCHD *arcn;
	register int *fd;
#endif
{
	/*
	 * make sure the file is open
	 */
	if (*fd < 0)
		return;

	(void)close(*fd);
	*fd = -1;
	if (!tflag)
		return;

	/*
	 * user wants last access time reset
	 */
	set_ftime(arcn->org_name, arcn->sb.st_mtime, arcn->sb.st_atime, 1);
	return;
}

/*
 * set_crc()
 *	read a file to calculate its crc. This is a real drag. Archive formats
 *	that have this, end up reading the file twice (we have to write the
 *	header WITH the crc before writing the file contents. Oh well...
 * Return:
 *	0 if was able to calculate the crc, -1 otherwise
 */

#if __STDC__
int
set_crc(register ARCHD *arcn, register int fd)
#else
int
set_crc(arcn, fd)
	register ARCHD *arcn;
	register int fd;
#endif
{
	register int i;
	register int res;
	off_t cpcnt = 0L;
	u_long size;
	unsigned long crc = 0L;
	char tbuf[FILEBLK];
	struct stat sb;

	if (fd < 0) {
		/*
		 * hmm, no fd, should never happen. well no crc then.
		 */
		arcn->crc = 0L;
		return(0);
	}

	if ((size = (u_long)arcn->sb.st_blksize) > (u_long)sizeof(tbuf))
		size = (u_long)sizeof(tbuf);

	/*
	 * read all the bytes we think that there are in the file. If the user
	 * is trying to archive an active file, forget this file.
	 */
	for(;;) {
		if ((res = read(fd, tbuf, size)) <= 0)
			break;
		cpcnt += res;
		for (i = 0; i < res; ++i)
			crc += (tbuf[i] & 0xff);
	}

	/*
	 * safety check. we want to avoid archiving files that are active as
	 * they can create inconsistant archive copies.
	 */
	if (cpcnt != arcn->sb.st_size)
		warn(1, "File changed size %s", arcn->org_name);
	else if (fstat(fd, &sb) < 0)
		syswarn(1, errno, "Failed stat on %s", arcn->org_name);
	else if (arcn->sb.st_mtime != sb.st_mtime)
		warn(1, "File %s was modified during read", arcn->org_name);
	else if (lseek(fd, (off_t)0L, SEEK_SET) < 0)
		syswarn(1, errno, "File rewind failed on: %s", arcn->org_name);
	else {
		arcn->crc = crc;
		return(0);
	}
	return(-1);
}