tables.c

早期freebsd实现

    bad:
	warn(1, "Unable to fix truncated inode/device field when storing %s",
	    arcn->name);
	warn(0, "Archive may create improper hard links when extracted");
	return(0);
}

/*
 * directory access/mod time reset table routines (for directories READ by pax)
 *
 * The pax -t flag requires that access times of archive files to be the same
 * before being read by pax. For regular files, access time is restored after
 * the file has been copied. This database provides the same functionality for
 * directories read during file tree traversal. Restoring directory access time
 * is more complex than files since directories may be read several times until
 * all the descendants in their subtree are visited by fts. Directory access
 * and modification times are stored during the fts pre-order visit (done
 * before any descendants in the subtree is visited) and restored after the
 * fts post-order visit (after all the descendants have been visited). In the
 * case of premature exit from a subtree (like from the effects of -n), any
 * directory entries left in this database are reset during final cleanup
 * operations of pax. Entries are hashed by inode number for fast lookup.
 */

/*
 * atdir_start()
 *	create the directory access time database for directories READ by pax.
 * Return:
 *	0 is created ok, -1 otherwise.
 */

#if __STDC__
int
atdir_start(void)
#else
int
atdir_start()
#endif
{
	if (atab != NULL)
		return(0);
 	if ((atab = (ATDIR **)calloc(A_TAB_SZ, sizeof(ATDIR *))) == NULL) {
                warn(1,"Cannot allocate space for directory access time table");
                return(-1);
        }
	return(0);
}


/*
 * atdir_end()
 *	walk through the directory access time table and reset the access time
 *	of any directory who still has an entry left in the database. These
 *	entries are for directories READ by pax
 */

#if __STDC__
void
atdir_end(void)
#else
void
atdir_end()
#endif
{
	register ATDIR *pt;
	register int i;

	if (atab == NULL)
		return;
	/*
	 * for each non-empty hash table entry reset all the directories
	 * chained there.
	 */
	for (i = 0; i < A_TAB_SZ; ++i) {
		if ((pt = atab[i]) == NULL)
			continue;
		/*
		 * remember to force the times, set_ftime() looks at pmtime
		 * and patime, which only applies to things CREATED by pax,
		 * not read by pax. Read time reset is controlled by -t.
		 */
		for (; pt != NULL; pt = pt->fow)
			set_ftime(pt->name, pt->mtime, pt->atime, 1);
	}
}

/*
 * add_atdir()
 *	add a directory to the directory access time table. Table is hashed
 *	and chained by inode number. This is for directories READ by pax
 */

#if __STDC__
void
add_atdir(char *fname, dev_t dev, ino_t ino, time_t mtime, time_t atime)
#else
void
add_atdir(fname, dev, ino, mtime, atime)
	char *fname;
	dev_t dev;
	ino_t ino;
	time_t mtime;
	time_t atime;
#endif
{
	register ATDIR *pt;
	register u_int indx;

	if (atab == NULL)
		return;

	/*
	 * make sure this directory is not already in the table, if so just 
	 * return (the older entry always has the correct time). The only
	 * way this will happen is when the same subtree can be traversed by
	 * different args to pax and the -n option is aborting fts out of a
	 * subtree before all the post-order visits have been made).
	 */
	indx = ((unsigned)ino) % A_TAB_SZ;
	if ((pt = atab[indx]) != NULL) {
		while (pt != NULL) {
			if ((pt->ino == ino) && (pt->dev == dev))
				break;
			pt = pt->fow;
		}

		/*
		 * oops, already there. Leave it alone.
		 */
		if (pt != NULL)
			return;
	}

	/*
	 * add it to the front of the hash chain
	 */
	if ((pt = (ATDIR *)malloc(sizeof(ATDIR))) != NULL) {
		if ((pt->name = strdup(fname)) != NULL) {
			pt->dev = dev;
			pt->ino = ino;
			pt->mtime = mtime;
			pt->atime = atime;
			pt->fow = atab[indx];
			atab[indx] = pt;
			return;
		}
		(void)free((char *)pt);
	}

	warn(1, "Directory access time reset table ran out of memory");
	return;
}

/*
 * get_atdir()
 *	look up a directory by inode and device number to obtain the access
 *	and modification time you want to set to. If found, the modification
 *	and access time parameters are set and the entry is removed from the
 *	table (as it is no longer needed). These are for directories READ by
 *	pax
 * Return:
 *	0 if found, -1 if not found.
 */

#if __STDC__
int
get_atdir(dev_t dev, ino_t ino, time_t *mtime, time_t *atime)
#else
int
get_atdir(dev, ino, mtime, atime)
	dev_t dev;
	ino_t ino;
	time_t *mtime;
	time_t *atime;
#endif
{
	register ATDIR *pt;
	register ATDIR **ppt;
	register u_int indx;

	if (atab == NULL)
		return(-1);
	/*
	 * hash by inode and search the chain for an inode and device match
	 */
	indx = ((unsigned)ino) % A_TAB_SZ;
	if ((pt = atab[indx]) == NULL)
		return(-1);

	ppt = &(atab[indx]);
	while (pt != NULL) {
		if ((pt->ino == ino) && (pt->dev == dev))
			break;
		/*
		 * no match, go to next one
		 */
		ppt = &(pt->fow);
		pt = pt->fow;
	}

	/*
	 * return if we did not find it.
	 */
	if (pt == NULL)
		return(-1);

	/*
	 * found it. return the times and remove the entry from the table.
	 */
	*ppt = pt->fow;
	*mtime = pt->mtime;
	*atime = pt->atime;
	(void)free((char *)pt->name);
	(void)free((char *)pt);
	return(0);
}

/*
 * directory access mode and time storage routines (for directories CREATED
 * by pax).
 *
 * Pax requires that extracted directories, by default, have their access/mod
 * times and permissions set to the values specified in the archive. During the
 * actions of extracting (and creating the destination subtree during -rw copy)
 * directories extracted may be modified after being created. Even worse is
 * that these directories may have been created with file permissions which
 * prohibits any descendants of these directories from being extracted. When
 * directories are created by pax, access rights may be added to permit the
 * creation of files in their subtree. Every time pax creates a directory, the
 * times and file permissions specified by the archive are stored. After all
 * files have been extracted (or copied), these directories have their times
 * and file modes reset to the stored values. The directory info is restored in
 * reverse order as entries were added to the data file from root to leaf. To
 * restore atime properly, we must go backwards. The data file consists of
 * records with two parts, the file name followed by a DIRDATA trailer. The
 * fixed sized trailer contains the size of the name plus the off_t location in
 * the file. To restore we work backwards through the file reading the trailer
 * then the file name.
 */

/*
 * dir_start()
 *	set up the directory time and file mode storage for directories CREATED
 *	by pax.
 * Return:
 *	0 if ok, -1 otherwise
 */

#if __STDC__
int
dir_start(void)
#else
int
dir_start()
#endif
{
	char *pt;

	if (dirfd != -1)
		return(0);
	if ((pt = tempnam((char *)NULL, (char *)NULL)) == NULL)
		return(-1);

	/*
	 * unlink the file so it goes away at termination by itself
	 */
	(void)unlink(pt);
	if ((dirfd = open(pt, O_RDWR|O_CREAT, 0600)) >= 0) {
		(void)unlink(pt);
		return(0);
	}
	warn(1, "Unable to create temporary file for directory times: %s", pt);
	return(-1);
}

/*
 * add_dir()
 *	add the mode and times for a newly CREATED directory
 *	name is name of the directory, psb the stat buffer with the data in it,
 *	frc_mode is a flag that says whether to force the setting of the mode
 *	(ignoring the user set values for preserving file mode). Frc_mode is
 *	for the case where we created a file and found that the resulting 
 *	directory was not writeable and the user asked for file modes to NOT
 *	be preserved. (we have to preserve what was created by default, so we
 *	have to force the setting at the end. this is stated explicitly in the
 *	pax spec)
 */

#if __STDC__
void
add_dir(char *name, int nlen, struct stat *psb, int frc_mode)
#else
void
add_dir(name, nlen, psb, frc_mode)
	char *name;
	int nlen;
	struct stat *psb;
	int frc_mode;
#endif
{
	DIRDATA dblk;

	if (dirfd < 0)
		return;

	/*
	 * get current position (where file name will start) so we can store it
	 * in the trailer
	 */
	if ((dblk.npos = lseek(dirfd, 0L, SEEK_CUR)) < 0) {
		warn(1,"Unable to store mode and times for directory: %s",name);
		return;
	}

	/*
	 * write the file name followed by the trailer
	 */
	dblk.nlen = nlen + 1;
	dblk.mode = psb->st_mode & 0xffff;
	dblk.mtime = psb->st_mtime;
	dblk.atime = psb->st_atime;
	dblk.frc_mode = frc_mode;
	if ((write(dirfd, name, dblk.nlen) == dblk.nlen) &&
	    (write(dirfd, (char *)&dblk, sizeof(dblk)) == sizeof(dblk))) {
		++dircnt;
		return;
	}

	warn(1,"Unable to store mode and times for created directory: %s",name);
	return;
}

/*
 * proc_dir()
 *	process all file modes and times stored for directories CREATED
 *	by pax
 */

#if __STDC__
void
proc_dir(void)
#else
void
proc_dir()
#endif
{
	char name[PAXPATHLEN+1];
	DIRDATA dblk;
	u_long cnt;

	if (dirfd < 0)
		return;
	/*
	 * read backwards through the file and process each directory
	 */
	for (cnt = 0; cnt < dircnt; ++cnt) {
		/*
		 * read the trailer, then the file name, if this fails
		 * just give up.
		 */
		if (lseek(dirfd, -((off_t)sizeof(dblk)), SEEK_CUR) < 0) 
			break;
		if (read(dirfd,(char *)&dblk, sizeof(dblk)) != sizeof(dblk))
			break;
		if (lseek(dirfd, dblk.npos, SEEK_SET) < 0) 
			break;
		if (read(dirfd, name, dblk.nlen) != dblk.nlen)
			break;
		if (lseek(dirfd, dblk.npos, SEEK_SET) < 0) 
			break;

		/*
		 * frc_mode set, make sure we set the file modes even if
		 * the user didn't ask for it (see file_subs.c for more info)
		 */
		if (pmode || dblk.frc_mode)
			set_pmode(name, dblk.mode);
		if (patime || pmtime)
			set_ftime(name, dblk.mtime, dblk.atime, 0);
	}

	(void)close(dirfd);
	dirfd = -1;
	if (cnt != dircnt)
		warn(1,"Unable to set mode and times for created directories");
	return;
}

/*
 * database independent routines
 */

/*
 * st_hash()
 *	hashes filenames to a u_int for hashing into a table. Looks at the tail
 *	end of file, as this provides far better distribution than any other
 *	part of the name. For performance reasons we only care about the last
 *	MAXKEYLEN chars (should be at LEAST large enough to pick off the file
 *	name). Was tested on 500,000 name file tree traversal from the root
 *	and gave almost a perfectly uniform distribution of keys when used with
 *	prime sized tables (MAXKEYLEN was 128 in test). Hashes (sizeof int)
 *	chars at a time and pads with 0 for last addition.
 * Return:
 *	the hash value of the string MOD (%) the table size.
 */

#if __STDC__
u_int
st_hash(char *name, int len, int tabsz)
#else
u_int
st_hash(name, len, tabsz)
	char *name;
	int len;
	int tabsz;
#endif
{
	register char *pt;
	register char *dest;
	register char *end;
	register int i;
	register u_int key = 0;
	register int steps;
	register int res;
	u_int val;

	/*
	 * only look at the tail up to MAXKEYLEN, we do not need to waste
	 * time here (remember these are pathnames, the tail is what will
	 * spread out the keys)
	 */
	if (len > MAXKEYLEN) {
                pt = &(name[len - MAXKEYLEN]);
		len = MAXKEYLEN;
	} else
		pt = name;

	/*
	 * calculate the number of u_int size steps in the string and if
	 * there is a runt to deal with
	 */
	steps = len/sizeof(u_int);
	res = len % sizeof(u_int);

	/*
	 * add up the value of the string in unsigned integer sized pieces
	 * too bad we cannot have unsigned int aligned strings, then we
	 * could avoid the expensive copy.
	 */
	for (i = 0; i < steps; ++i) {
		end = pt + sizeof(u_int);
		dest = (char *)&val;
		while (pt < end)
			*dest++ = *pt++;
		key += val;
	}

	/*
	 * add in the runt padded with zero to the right
	 */
	if (res) {
		val = 0;
		end = pt + res;
		dest = (char *)&val;
		while (pt < end)
			*dest++ = *pt++;
		key += val;
	}

	/*
	 * return the result mod the table size
	 */
	return(key % tabsz);
}