to upstream.
  - revert to upstream version of function prototypes for
    setting the uid and gid fields of the archive_entry.
  - move uid/gid overflow checks into header_common().
  - use archive_set_error() instead of fprintf() for getting
    error message text back to the main program.
		
	
			
		
			
				
	
	
		
			2226 lines
		
	
	
		
			54 KiB
		
	
	
	
		
			C
		
	
	
	
	
	
			
		
		
	
	
			2226 lines
		
	
	
		
			54 KiB
		
	
	
	
		
			C
		
	
	
	
	
	
/*-
 | 
						|
 * Copyright (c) 2003-2007 Tim Kientzle
 | 
						|
 * All rights reserved.
 | 
						|
 *
 | 
						|
 * Redistribution and use in source and binary forms, with or without
 | 
						|
 * modification, are permitted provided that the following conditions
 | 
						|
 * are met:
 | 
						|
 * 1. Redistributions of source code must retain the above copyright
 | 
						|
 *    notice, this list of conditions and the following disclaimer.
 | 
						|
 * 2. Redistributions in binary form must reproduce the above copyright
 | 
						|
 *    notice, this list of conditions and the following disclaimer in the
 | 
						|
 *    documentation and/or other materials provided with the distribution.
 | 
						|
 *
 | 
						|
 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR(S) ``AS IS'' AND ANY EXPRESS OR
 | 
						|
 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
 | 
						|
 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
 | 
						|
 * IN NO EVENT SHALL THE AUTHOR(S) BE LIABLE FOR ANY DIRECT, INDIRECT,
 | 
						|
 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
 | 
						|
 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
 | 
						|
 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
 | 
						|
 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
 | 
						|
 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
 | 
						|
 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 | 
						|
 */
 | 
						|
 | 
						|
#include "archive_platform.h"
 | 
						|
__FBSDID("$FreeBSD: head/lib/libarchive/archive_entry.c 201096 2009-12-28 02:41:27Z kientzle $");
 | 
						|
 | 
						|
#ifdef HAVE_SYS_STAT_H
 | 
						|
#include <sys/stat.h>
 | 
						|
#endif
 | 
						|
#ifdef HAVE_SYS_TYPES_H
 | 
						|
#include <sys/types.h>
 | 
						|
#endif
 | 
						|
#if MAJOR_IN_MKDEV
 | 
						|
#include <sys/mkdev.h>
 | 
						|
#define HAVE_MAJOR
 | 
						|
#elif MAJOR_IN_SYSMACROS
 | 
						|
#include <sys/sysmacros.h>
 | 
						|
#define HAVE_MAJOR
 | 
						|
#endif
 | 
						|
#ifdef HAVE_LIMITS_H
 | 
						|
#include <limits.h>
 | 
						|
#endif
 | 
						|
#ifdef HAVE_LINUX_FS_H
 | 
						|
#include <linux/fs.h>	/* for Linux file flags */
 | 
						|
#endif
 | 
						|
/*
 | 
						|
 * Some Linux distributions have both linux/ext2_fs.h and ext2fs/ext2_fs.h.
 | 
						|
 * As the include guards don't agree, the order of include is important.
 | 
						|
 */
 | 
						|
#ifdef HAVE_LINUX_EXT2_FS_H
 | 
						|
#include <linux/ext2_fs.h>	/* for Linux file flags */
 | 
						|
#endif
 | 
						|
#if defined(HAVE_EXT2FS_EXT2_FS_H) && !defined(__CYGWIN__)
 | 
						|
#include <ext2fs/ext2_fs.h>	/* for Linux file flags */
 | 
						|
#endif
 | 
						|
#include <stddef.h>
 | 
						|
#include <stdio.h>
 | 
						|
#ifdef HAVE_STDLIB_H
 | 
						|
#include <stdlib.h>
 | 
						|
#endif
 | 
						|
#ifdef HAVE_STRING_H
 | 
						|
#include <string.h>
 | 
						|
#endif
 | 
						|
#ifdef HAVE_WCHAR_H
 | 
						|
#include <wchar.h>
 | 
						|
#endif
 | 
						|
 | 
						|
#include "archive.h"
 | 
						|
#include "archive_entry.h"
 | 
						|
#include "archive_private.h"
 | 
						|
#include "archive_entry_private.h"
 | 
						|
 | 
						|
#undef max
 | 
						|
#define	max(a, b)	((a)>(b)?(a):(b))
 | 
						|
 | 
						|
#if !defined(HAVE_MAJOR) && !defined(major)
 | 
						|
/* Replacement for major/minor/makedev. */
 | 
						|
#define	major(x) ((int)(0x00ff & ((x) >> 8)))
 | 
						|
#define	minor(x) ((int)(0xffff00ff & (x)))
 | 
						|
#define	makedev(maj,min) ((0xff00 & ((maj)<<8)) | (0xffff00ff & (min)))
 | 
						|
#endif
 | 
						|
 | 
						|
/* Play games to come up with a suitable makedev() definition. */
 | 
						|
#ifdef __QNXNTO__
 | 
						|
/* QNX.  <sigh> */
 | 
						|
#include <sys/netmgr.h>
 | 
						|
#define ae_makedev(maj, min) makedev(ND_LOCAL_NODE, (maj), (min))
 | 
						|
#elif defined makedev
 | 
						|
/* There's a "makedev" macro. */
 | 
						|
#define ae_makedev(maj, min) makedev((maj), (min))
 | 
						|
#elif defined mkdev || ((defined _WIN32 || defined __WIN32__) && !defined(__CYGWIN__))
 | 
						|
/* Windows. <sigh> */
 | 
						|
#define ae_makedev(maj, min) mkdev((maj), (min))
 | 
						|
#else
 | 
						|
/* There's a "makedev" function. */
 | 
						|
#define ae_makedev(maj, min) makedev((maj), (min))
 | 
						|
#endif
 | 
						|
 | 
						|
static void	aes_clean(struct aes *);
 | 
						|
static void	aes_copy(struct aes *dest, struct aes *src);
 | 
						|
static const char *	aes_get_mbs(struct aes *);
 | 
						|
static const wchar_t *	aes_get_wcs(struct aes *);
 | 
						|
static int	aes_set_mbs(struct aes *, const char *mbs);
 | 
						|
static int	aes_copy_mbs(struct aes *, const char *mbs);
 | 
						|
/* static void	aes_set_wcs(struct aes *, const wchar_t *wcs); */
 | 
						|
static int	aes_copy_wcs(struct aes *, const wchar_t *wcs);
 | 
						|
static int	aes_copy_wcs_len(struct aes *, const wchar_t *wcs, size_t);
 | 
						|
 | 
						|
static char *	 ae_fflagstostr(unsigned long bitset, unsigned long bitclear);
 | 
						|
static const wchar_t	*ae_wcstofflags(const wchar_t *stringp,
 | 
						|
		    unsigned long *setp, unsigned long *clrp);
 | 
						|
static const char	*ae_strtofflags(const char *stringp,
 | 
						|
		    unsigned long *setp, unsigned long *clrp);
 | 
						|
static void	append_entry_w(wchar_t **wp, const wchar_t *prefix, int tag,
 | 
						|
		    const wchar_t *wname, int perm, int id);
 | 
						|
static void	append_id_w(wchar_t **wp, int id);
 | 
						|
 | 
						|
static int	acl_special(struct archive_entry *entry,
 | 
						|
		    int type, int permset, int tag);
 | 
						|
static struct ae_acl *acl_new_entry(struct archive_entry *entry,
 | 
						|
		    int type, int permset, int tag, int id);
 | 
						|
static int	isint_w(const wchar_t *start, const wchar_t *end, int *result);
 | 
						|
static int	ismode_w(const wchar_t *start, const wchar_t *end, int *result);
 | 
						|
static void	next_field_w(const wchar_t **wp, const wchar_t **start,
 | 
						|
		    const wchar_t **end, wchar_t *sep);
 | 
						|
static int	prefix_w(const wchar_t *start, const wchar_t *end,
 | 
						|
		    const wchar_t *test);
 | 
						|
static void
 | 
						|
archive_entry_acl_add_entry_w_len(struct archive_entry *entry, int type,
 | 
						|
		    int permset, int tag, int id, const wchar_t *name, size_t);
 | 
						|
 | 
						|
 | 
						|
#ifndef HAVE_WCSCPY
 | 
						|
static wchar_t * wcscpy(wchar_t *s1, const wchar_t *s2)
 | 
						|
{
 | 
						|
	wchar_t *dest = s1;
 | 
						|
	while ((*s1 = *s2) != L'\0')
 | 
						|
		++s1, ++s2;
 | 
						|
	return dest;
 | 
						|
}
 | 
						|
#endif
 | 
						|
#ifndef HAVE_WCSLEN
 | 
						|
static size_t wcslen(const wchar_t *s)
 | 
						|
{
 | 
						|
	const wchar_t *p = s;
 | 
						|
	while (*p != L'\0')
 | 
						|
		++p;
 | 
						|
	return p - s;
 | 
						|
}
 | 
						|
#endif
 | 
						|
#ifndef HAVE_WMEMCMP
 | 
						|
/* Good enough for simple equality testing, but not for sorting. */
 | 
						|
#define wmemcmp(a,b,i)  memcmp((a), (b), (i) * sizeof(wchar_t))
 | 
						|
#endif
 | 
						|
#ifndef HAVE_WMEMCPY
 | 
						|
#define wmemcpy(a,b,i)  (wchar_t *)memcpy((a), (b), (i) * sizeof(wchar_t))
 | 
						|
#endif
 | 
						|
 | 
						|
static void
 | 
						|
aes_clean(struct aes *aes)
 | 
						|
{
 | 
						|
	if (aes->aes_wcs) {
 | 
						|
		free((wchar_t *)(uintptr_t)aes->aes_wcs);
 | 
						|
		aes->aes_wcs = NULL;
 | 
						|
	}
 | 
						|
	archive_string_free(&(aes->aes_mbs));
 | 
						|
	archive_string_free(&(aes->aes_utf8));
 | 
						|
	aes->aes_set = 0;
 | 
						|
}
 | 
						|
 | 
						|
static void
 | 
						|
aes_copy(struct aes *dest, struct aes *src)
 | 
						|
{
 | 
						|
	wchar_t *wp;
 | 
						|
 | 
						|
	dest->aes_set = src->aes_set;
 | 
						|
	archive_string_copy(&(dest->aes_mbs), &(src->aes_mbs));
 | 
						|
	archive_string_copy(&(dest->aes_utf8), &(src->aes_utf8));
 | 
						|
 | 
						|
	if (src->aes_wcs != NULL) {
 | 
						|
		wp = (wchar_t *)malloc((wcslen(src->aes_wcs) + 1)
 | 
						|
		    * sizeof(wchar_t));
 | 
						|
		if (wp == NULL)
 | 
						|
			__archive_errx(1, "No memory for aes_copy()");
 | 
						|
		wcscpy(wp, src->aes_wcs);
 | 
						|
		dest->aes_wcs = wp;
 | 
						|
	}
 | 
						|
}
 | 
						|
 | 
						|
static const char *
 | 
						|
aes_get_utf8(struct aes *aes)
 | 
						|
{
 | 
						|
	if (aes->aes_set & AES_SET_UTF8)
 | 
						|
		return (aes->aes_utf8.s);
 | 
						|
	if ((aes->aes_set & AES_SET_WCS)
 | 
						|
	    && archive_strappend_w_utf8(&(aes->aes_utf8), aes->aes_wcs) != NULL) {
 | 
						|
		aes->aes_set |= AES_SET_UTF8;
 | 
						|
		return (aes->aes_utf8.s);
 | 
						|
	}
 | 
						|
	return (NULL);
 | 
						|
}
 | 
						|
 | 
						|
static const char *
 | 
						|
aes_get_mbs(struct aes *aes)
 | 
						|
{
 | 
						|
	/* If we already have an MBS form, return that immediately. */
 | 
						|
	if (aes->aes_set & AES_SET_MBS)
 | 
						|
		return (aes->aes_mbs.s);
 | 
						|
	/* If there's a WCS form, try converting with the native locale. */
 | 
						|
	if ((aes->aes_set & AES_SET_WCS)
 | 
						|
	    && archive_strappend_w_mbs(&(aes->aes_mbs), aes->aes_wcs) != NULL) {
 | 
						|
		aes->aes_set |= AES_SET_MBS;
 | 
						|
		return (aes->aes_mbs.s);
 | 
						|
	}
 | 
						|
	/* We'll use UTF-8 for MBS if all else fails. */
 | 
						|
	return (aes_get_utf8(aes));
 | 
						|
}
 | 
						|
 | 
						|
static const wchar_t *
 | 
						|
aes_get_wcs(struct aes *aes)
 | 
						|
{
 | 
						|
	wchar_t *w;
 | 
						|
	size_t r;
 | 
						|
 | 
						|
	/* Return WCS form if we already have it. */
 | 
						|
	if (aes->aes_set & AES_SET_WCS)
 | 
						|
		return (aes->aes_wcs);
 | 
						|
 | 
						|
	if (aes->aes_set & AES_SET_MBS) {
 | 
						|
		/* Try converting MBS to WCS using native locale. */
 | 
						|
		/*
 | 
						|
		 * No single byte will be more than one wide character,
 | 
						|
		 * so this length estimate will always be big enough.
 | 
						|
		 */
 | 
						|
		size_t wcs_length = aes->aes_mbs.length;
 | 
						|
 | 
						|
		w = (wchar_t *)malloc((wcs_length + 1) * sizeof(wchar_t));
 | 
						|
		if (w == NULL)
 | 
						|
			__archive_errx(1, "No memory for aes_get_wcs()");
 | 
						|
		r = mbstowcs(w, aes->aes_mbs.s, wcs_length);
 | 
						|
		if (r != (size_t)-1 && r != 0) {
 | 
						|
			w[r] = 0;
 | 
						|
			aes->aes_set |= AES_SET_WCS;
 | 
						|
			return (aes->aes_wcs = w);
 | 
						|
		}
 | 
						|
		free(w);
 | 
						|
	}
 | 
						|
 | 
						|
	if (aes->aes_set & AES_SET_UTF8) {
 | 
						|
		/* Try converting UTF8 to WCS. */
 | 
						|
		aes->aes_wcs = __archive_string_utf8_w(&(aes->aes_utf8));
 | 
						|
		if (aes->aes_wcs != NULL)
 | 
						|
			aes->aes_set |= AES_SET_WCS;
 | 
						|
		return (aes->aes_wcs);
 | 
						|
	}
 | 
						|
	return (NULL);
 | 
						|
}
 | 
						|
 | 
						|
static int
 | 
						|
aes_set_mbs(struct aes *aes, const char *mbs)
 | 
						|
{
 | 
						|
	return (aes_copy_mbs(aes, mbs));
 | 
						|
}
 | 
						|
 | 
						|
static int
 | 
						|
aes_copy_mbs(struct aes *aes, const char *mbs)
 | 
						|
{
 | 
						|
	if (mbs == NULL) {
 | 
						|
		aes->aes_set = 0;
 | 
						|
		return (0);
 | 
						|
	}
 | 
						|
	aes->aes_set = AES_SET_MBS; /* Only MBS form is set now. */
 | 
						|
	archive_strcpy(&(aes->aes_mbs), mbs);
 | 
						|
	archive_string_empty(&(aes->aes_utf8));
 | 
						|
	if (aes->aes_wcs) {
 | 
						|
		free((wchar_t *)(uintptr_t)aes->aes_wcs);
 | 
						|
		aes->aes_wcs = NULL;
 | 
						|
	}
 | 
						|
	return (0);
 | 
						|
}
 | 
						|
 | 
						|
/*
 | 
						|
 * The 'update' form tries to proactively update all forms of
 | 
						|
 * this string (WCS and MBS) and returns an error if any of
 | 
						|
 * them fail.  This is used by the 'pax' handler, for instance,
 | 
						|
 * to detect and report character-conversion failures early while
 | 
						|
 * still allowing clients to get potentially useful values from
 | 
						|
 * the more tolerant lazy conversions.  (get_mbs and get_wcs will
 | 
						|
 * strive to give the user something useful, so you can get hopefully
 | 
						|
 * usable values even if some of the character conversions are failing.)
 | 
						|
 */
 | 
						|
static int
 | 
						|
aes_update_utf8(struct aes *aes, const char *utf8)
 | 
						|
{
 | 
						|
	if (utf8 == NULL) {
 | 
						|
		aes->aes_set = 0;
 | 
						|
		return (1); /* Succeeded in clearing everything. */
 | 
						|
	}
 | 
						|
 | 
						|
	/* Save the UTF8 string. */
 | 
						|
	archive_strcpy(&(aes->aes_utf8), utf8);
 | 
						|
 | 
						|
	/* Empty the mbs and wcs strings. */
 | 
						|
	archive_string_empty(&(aes->aes_mbs));
 | 
						|
	if (aes->aes_wcs) {
 | 
						|
		free((wchar_t *)(uintptr_t)aes->aes_wcs);
 | 
						|
		aes->aes_wcs = NULL;
 | 
						|
	}
 | 
						|
 | 
						|
	aes->aes_set = AES_SET_UTF8;	/* Only UTF8 is set now. */
 | 
						|
 | 
						|
	/* TODO: We should just do a direct UTF-8 to MBS conversion
 | 
						|
	 * here.  That would be faster, use less space, and give the
 | 
						|
	 * same information.  (If a UTF-8 to MBS conversion succeeds,
 | 
						|
	 * then UTF-8->WCS and Unicode->MBS conversions will both
 | 
						|
	 * succeed.) */
 | 
						|
 | 
						|
	/* Try converting UTF8 to WCS, return false on failure. */
 | 
						|
	aes->aes_wcs = __archive_string_utf8_w(&(aes->aes_utf8));
 | 
						|
	if (aes->aes_wcs == NULL)
 | 
						|
		return (0);
 | 
						|
	aes->aes_set = AES_SET_UTF8 | AES_SET_WCS; /* Both UTF8 and WCS set. */
 | 
						|
 | 
						|
	/* Try converting WCS to MBS, return false on failure. */
 | 
						|
	if (archive_strappend_w_mbs(&(aes->aes_mbs), aes->aes_wcs) == NULL)
 | 
						|
		return (0);
 | 
						|
	aes->aes_set = AES_SET_UTF8 | AES_SET_WCS | AES_SET_MBS;
 | 
						|
 | 
						|
	/* All conversions succeeded. */
 | 
						|
	return (1);
 | 
						|
}
 | 
						|
 | 
						|
static int
 | 
						|
aes_copy_wcs(struct aes *aes, const wchar_t *wcs)
 | 
						|
{
 | 
						|
	return aes_copy_wcs_len(aes, wcs, wcs == NULL ? 0 : wcslen(wcs));
 | 
						|
}
 | 
						|
 | 
						|
static int
 | 
						|
aes_copy_wcs_len(struct aes *aes, const wchar_t *wcs, size_t len)
 | 
						|
{
 | 
						|
	wchar_t *w;
 | 
						|
 | 
						|
	if (wcs == NULL) {
 | 
						|
		aes->aes_set = 0;
 | 
						|
		return (0);
 | 
						|
	}
 | 
						|
	aes->aes_set = AES_SET_WCS; /* Only WCS form set. */
 | 
						|
	archive_string_empty(&(aes->aes_mbs));
 | 
						|
	archive_string_empty(&(aes->aes_utf8));
 | 
						|
	if (aes->aes_wcs) {
 | 
						|
		free((wchar_t *)(uintptr_t)aes->aes_wcs);
 | 
						|
		aes->aes_wcs = NULL;
 | 
						|
	}
 | 
						|
	w = (wchar_t *)malloc((len + 1) * sizeof(wchar_t));
 | 
						|
	if (w == NULL)
 | 
						|
		__archive_errx(1, "No memory for aes_copy_wcs()");
 | 
						|
	wmemcpy(w, wcs, len);
 | 
						|
	w[len] = L'\0';
 | 
						|
	aes->aes_wcs = w;
 | 
						|
	return (0);
 | 
						|
}
 | 
						|
 | 
						|
/****************************************************************************
 | 
						|
 *
 | 
						|
 * Public Interface
 | 
						|
 *
 | 
						|
 ****************************************************************************/
 | 
						|
 | 
						|
struct archive_entry *
 | 
						|
archive_entry_clear(struct archive_entry *entry)
 | 
						|
{
 | 
						|
	if (entry == NULL)
 | 
						|
		return (NULL);
 | 
						|
	aes_clean(&entry->ae_fflags_text);
 | 
						|
	aes_clean(&entry->ae_gname);
 | 
						|
	aes_clean(&entry->ae_hardlink);
 | 
						|
	aes_clean(&entry->ae_pathname);
 | 
						|
	aes_clean(&entry->ae_sourcepath);
 | 
						|
	aes_clean(&entry->ae_symlink);
 | 
						|
	aes_clean(&entry->ae_uname);
 | 
						|
	archive_entry_acl_clear(entry);
 | 
						|
	archive_entry_xattr_clear(entry);
 | 
						|
	free(entry->stat);
 | 
						|
	memset(entry, 0, sizeof(*entry));
 | 
						|
	return entry;
 | 
						|
}
 | 
						|
 | 
						|
struct archive_entry *
 | 
						|
archive_entry_clone(struct archive_entry *entry)
 | 
						|
{
 | 
						|
	struct archive_entry *entry2;
 | 
						|
	struct ae_acl *ap, *ap2;
 | 
						|
	struct ae_xattr *xp;
 | 
						|
 | 
						|
	/* Allocate new structure and copy over all of the fields. */
 | 
						|
	entry2 = (struct archive_entry *)malloc(sizeof(*entry2));
 | 
						|
	if (entry2 == NULL)
 | 
						|
		return (NULL);
 | 
						|
	memset(entry2, 0, sizeof(*entry2));
 | 
						|
	entry2->ae_stat = entry->ae_stat;
 | 
						|
	entry2->ae_fflags_set = entry->ae_fflags_set;
 | 
						|
	entry2->ae_fflags_clear = entry->ae_fflags_clear;
 | 
						|
 | 
						|
	aes_copy(&entry2->ae_fflags_text, &entry->ae_fflags_text);
 | 
						|
	aes_copy(&entry2->ae_gname, &entry->ae_gname);
 | 
						|
	aes_copy(&entry2->ae_hardlink, &entry->ae_hardlink);
 | 
						|
	aes_copy(&entry2->ae_pathname, &entry->ae_pathname);
 | 
						|
	aes_copy(&entry2->ae_sourcepath, &entry->ae_sourcepath);
 | 
						|
	aes_copy(&entry2->ae_symlink, &entry->ae_symlink);
 | 
						|
	entry2->ae_set = entry->ae_set;
 | 
						|
	aes_copy(&entry2->ae_uname, &entry->ae_uname);
 | 
						|
 | 
						|
	/* Copy ACL data over. */
 | 
						|
	ap = entry->acl_head;
 | 
						|
	while (ap != NULL) {
 | 
						|
		ap2 = acl_new_entry(entry2,
 | 
						|
		    ap->type, ap->permset, ap->tag, ap->id);
 | 
						|
		if (ap2 != NULL)
 | 
						|
			aes_copy(&ap2->name, &ap->name);
 | 
						|
		ap = ap->next;
 | 
						|
	}
 | 
						|
 | 
						|
	/* Copy xattr data over. */
 | 
						|
	xp = entry->xattr_head;
 | 
						|
	while (xp != NULL) {
 | 
						|
		archive_entry_xattr_add_entry(entry2,
 | 
						|
		    xp->name, xp->value, xp->size);
 | 
						|
		xp = xp->next;
 | 
						|
	}
 | 
						|
 | 
						|
	return (entry2);
 | 
						|
}
 | 
						|
 | 
						|
void
 | 
						|
archive_entry_free(struct archive_entry *entry)
 | 
						|
{
 | 
						|
	archive_entry_clear(entry);
 | 
						|
	free(entry);
 | 
						|
}
 | 
						|
 | 
						|
struct archive_entry *
 | 
						|
archive_entry_new(void)
 | 
						|
{
 | 
						|
	struct archive_entry *entry;
 | 
						|
 | 
						|
	entry = (struct archive_entry *)malloc(sizeof(*entry));
 | 
						|
	if (entry == NULL)
 | 
						|
		return (NULL);
 | 
						|
	memset(entry, 0, sizeof(*entry));
 | 
						|
	return (entry);
 | 
						|
}
 | 
						|
 | 
						|
/*
 | 
						|
 * Functions for reading fields from an archive_entry.
 | 
						|
 */
 | 
						|
 | 
						|
time_t
 | 
						|
archive_entry_atime(struct archive_entry *entry)
 | 
						|
{
 | 
						|
	return (entry->ae_stat.aest_atime);
 | 
						|
}
 | 
						|
 | 
						|
long
 | 
						|
archive_entry_atime_nsec(struct archive_entry *entry)
 | 
						|
{
 | 
						|
	return (entry->ae_stat.aest_atime_nsec);
 | 
						|
}
 | 
						|
 | 
						|
int
 | 
						|
archive_entry_atime_is_set(struct archive_entry *entry)
 | 
						|
{
 | 
						|
	return (entry->ae_set & AE_SET_ATIME);
 | 
						|
}
 | 
						|
 | 
						|
time_t
 | 
						|
archive_entry_birthtime(struct archive_entry *entry)
 | 
						|
{
 | 
						|
	return (entry->ae_stat.aest_birthtime);
 | 
						|
}
 | 
						|
 | 
						|
long
 | 
						|
archive_entry_birthtime_nsec(struct archive_entry *entry)
 | 
						|
{
 | 
						|
	return (entry->ae_stat.aest_birthtime_nsec);
 | 
						|
}
 | 
						|
 | 
						|
int
 | 
						|
archive_entry_birthtime_is_set(struct archive_entry *entry)
 | 
						|
{
 | 
						|
	return (entry->ae_set & AE_SET_BIRTHTIME);
 | 
						|
}
 | 
						|
 | 
						|
time_t
 | 
						|
archive_entry_ctime(struct archive_entry *entry)
 | 
						|
{
 | 
						|
	return (entry->ae_stat.aest_ctime);
 | 
						|
}
 | 
						|
 | 
						|
int
 | 
						|
archive_entry_ctime_is_set(struct archive_entry *entry)
 | 
						|
{
 | 
						|
	return (entry->ae_set & AE_SET_CTIME);
 | 
						|
}
 | 
						|
 | 
						|
long
 | 
						|
archive_entry_ctime_nsec(struct archive_entry *entry)
 | 
						|
{
 | 
						|
	return (entry->ae_stat.aest_ctime_nsec);
 | 
						|
}
 | 
						|
 | 
						|
dev_t
 | 
						|
archive_entry_dev(struct archive_entry *entry)
 | 
						|
{
 | 
						|
	if (entry->ae_stat.aest_dev_is_broken_down)
 | 
						|
		return ae_makedev(entry->ae_stat.aest_devmajor,
 | 
						|
		    entry->ae_stat.aest_devminor);
 | 
						|
	else
 | 
						|
		return (entry->ae_stat.aest_dev);
 | 
						|
}
 | 
						|
 | 
						|
dev_t
 | 
						|
archive_entry_devmajor(struct archive_entry *entry)
 | 
						|
{
 | 
						|
	if (entry->ae_stat.aest_dev_is_broken_down)
 | 
						|
		return (entry->ae_stat.aest_devmajor);
 | 
						|
	else
 | 
						|
		return major(entry->ae_stat.aest_dev);
 | 
						|
}
 | 
						|
 | 
						|
dev_t
 | 
						|
archive_entry_devminor(struct archive_entry *entry)
 | 
						|
{
 | 
						|
	if (entry->ae_stat.aest_dev_is_broken_down)
 | 
						|
		return (entry->ae_stat.aest_devminor);
 | 
						|
	else
 | 
						|
		return minor(entry->ae_stat.aest_dev);
 | 
						|
}
 | 
						|
 | 
						|
mode_t
 | 
						|
archive_entry_filetype(struct archive_entry *entry)
 | 
						|
{
 | 
						|
	return (AE_IFMT & entry->ae_stat.aest_mode);
 | 
						|
}
 | 
						|
 | 
						|
void
 | 
						|
archive_entry_fflags(struct archive_entry *entry,
 | 
						|
    unsigned long *set, unsigned long *clear)
 | 
						|
{
 | 
						|
	*set = entry->ae_fflags_set;
 | 
						|
	*clear = entry->ae_fflags_clear;
 | 
						|
}
 | 
						|
 | 
						|
/*
 | 
						|
 * Note: if text was provided, this just returns that text.  If you
 | 
						|
 * really need the text to be rebuilt in a canonical form, set the
 | 
						|
 * text, ask for the bitmaps, then set the bitmaps.  (Setting the
 | 
						|
 * bitmaps clears any stored text.)  This design is deliberate: if
 | 
						|
 * we're editing archives, we don't want to discard flags just because
 | 
						|
 * they aren't supported on the current system.  The bitmap<->text
 | 
						|
 * conversions are platform-specific (see below).
 | 
						|
 */
 | 
						|
const char *
 | 
						|
archive_entry_fflags_text(struct archive_entry *entry)
 | 
						|
{
 | 
						|
	const char *f;
 | 
						|
	char *p;
 | 
						|
 | 
						|
	f = aes_get_mbs(&entry->ae_fflags_text);
 | 
						|
	if (f != NULL)
 | 
						|
		return (f);
 | 
						|
 | 
						|
	if (entry->ae_fflags_set == 0  &&  entry->ae_fflags_clear == 0)
 | 
						|
		return (NULL);
 | 
						|
 | 
						|
	p = ae_fflagstostr(entry->ae_fflags_set, entry->ae_fflags_clear);
 | 
						|
	if (p == NULL)
 | 
						|
		return (NULL);
 | 
						|
 | 
						|
	aes_copy_mbs(&entry->ae_fflags_text, p);
 | 
						|
	free(p);
 | 
						|
	f = aes_get_mbs(&entry->ae_fflags_text);
 | 
						|
	return (f);
 | 
						|
}
 | 
						|
 | 
						|
gid_t
 | 
						|
archive_entry_gid(struct archive_entry *entry)
 | 
						|
{
 | 
						|
	return (entry->ae_stat.aest_gid);
 | 
						|
}
 | 
						|
 | 
						|
const char *
 | 
						|
archive_entry_gname(struct archive_entry *entry)
 | 
						|
{
 | 
						|
	return (aes_get_mbs(&entry->ae_gname));
 | 
						|
}
 | 
						|
 | 
						|
const wchar_t *
 | 
						|
archive_entry_gname_w(struct archive_entry *entry)
 | 
						|
{
 | 
						|
	return (aes_get_wcs(&entry->ae_gname));
 | 
						|
}
 | 
						|
 | 
						|
const char *
 | 
						|
archive_entry_hardlink(struct archive_entry *entry)
 | 
						|
{
 | 
						|
	if (entry->ae_set & AE_SET_HARDLINK)
 | 
						|
		return (aes_get_mbs(&entry->ae_hardlink));
 | 
						|
	return (NULL);
 | 
						|
}
 | 
						|
 | 
						|
const wchar_t *
 | 
						|
archive_entry_hardlink_w(struct archive_entry *entry)
 | 
						|
{
 | 
						|
	if (entry->ae_set & AE_SET_HARDLINK)
 | 
						|
		return (aes_get_wcs(&entry->ae_hardlink));
 | 
						|
	return (NULL);
 | 
						|
}
 | 
						|
 | 
						|
ino_t
 | 
						|
archive_entry_ino(struct archive_entry *entry)
 | 
						|
{
 | 
						|
	return (entry->ae_stat.aest_ino);
 | 
						|
}
 | 
						|
 | 
						|
#ifndef __minix
 | 
						|
int64_t
 | 
						|
archive_entry_ino64(struct archive_entry *entry)
 | 
						|
{
 | 
						|
	return (entry->ae_stat.aest_ino);
 | 
						|
}
 | 
						|
#endif
 | 
						|
 | 
						|
 | 
						|
mode_t
 | 
						|
archive_entry_mode(struct archive_entry *entry)
 | 
						|
{
 | 
						|
	return (entry->ae_stat.aest_mode);
 | 
						|
}
 | 
						|
 | 
						|
time_t
 | 
						|
archive_entry_mtime(struct archive_entry *entry)
 | 
						|
{
 | 
						|
	return (entry->ae_stat.aest_mtime);
 | 
						|
}
 | 
						|
 | 
						|
long
 | 
						|
archive_entry_mtime_nsec(struct archive_entry *entry)
 | 
						|
{
 | 
						|
	return (entry->ae_stat.aest_mtime_nsec);
 | 
						|
}
 | 
						|
 | 
						|
int
 | 
						|
archive_entry_mtime_is_set(struct archive_entry *entry)
 | 
						|
{
 | 
						|
	return (entry->ae_set & AE_SET_MTIME);
 | 
						|
}
 | 
						|
 | 
						|
unsigned int
 | 
						|
archive_entry_nlink(struct archive_entry *entry)
 | 
						|
{
 | 
						|
	return (entry->ae_stat.aest_nlink);
 | 
						|
}
 | 
						|
 | 
						|
const char *
 | 
						|
archive_entry_pathname(struct archive_entry *entry)
 | 
						|
{
 | 
						|
	return (aes_get_mbs(&entry->ae_pathname));
 | 
						|
}
 | 
						|
 | 
						|
const wchar_t *
 | 
						|
archive_entry_pathname_w(struct archive_entry *entry)
 | 
						|
{
 | 
						|
	return (aes_get_wcs(&entry->ae_pathname));
 | 
						|
}
 | 
						|
 | 
						|
dev_t
 | 
						|
archive_entry_rdev(struct archive_entry *entry)
 | 
						|
{
 | 
						|
	if (entry->ae_stat.aest_rdev_is_broken_down)
 | 
						|
		return ae_makedev(entry->ae_stat.aest_rdevmajor,
 | 
						|
		    entry->ae_stat.aest_rdevminor);
 | 
						|
	else
 | 
						|
		return (entry->ae_stat.aest_rdev);
 | 
						|
}
 | 
						|
 | 
						|
dev_t
 | 
						|
archive_entry_rdevmajor(struct archive_entry *entry)
 | 
						|
{
 | 
						|
	if (entry->ae_stat.aest_rdev_is_broken_down)
 | 
						|
		return (entry->ae_stat.aest_rdevmajor);
 | 
						|
	else
 | 
						|
		return major(entry->ae_stat.aest_rdev);
 | 
						|
}
 | 
						|
 | 
						|
dev_t
 | 
						|
archive_entry_rdevminor(struct archive_entry *entry)
 | 
						|
{
 | 
						|
	if (entry->ae_stat.aest_rdev_is_broken_down)
 | 
						|
		return (entry->ae_stat.aest_rdevminor);
 | 
						|
	else
 | 
						|
		return minor(entry->ae_stat.aest_rdev);
 | 
						|
}
 | 
						|
 | 
						|
#ifndef __minix
 | 
						|
int64_t
 | 
						|
archive_entry_size(struct archive_entry *entry)
 | 
						|
{
 | 
						|
	return (entry->ae_stat.aest_size);
 | 
						|
}
 | 
						|
#else
 | 
						|
ssize_t
 | 
						|
archive_entry_size(struct archive_entry *entry)
 | 
						|
{
 | 
						|
	return (entry->ae_stat.aest_size);
 | 
						|
}
 | 
						|
#endif
 | 
						|
 | 
						|
int
 | 
						|
archive_entry_size_is_set(struct archive_entry *entry)
 | 
						|
{
 | 
						|
	return (entry->ae_set & AE_SET_SIZE);
 | 
						|
}
 | 
						|
 | 
						|
const char *
 | 
						|
archive_entry_sourcepath(struct archive_entry *entry)
 | 
						|
{
 | 
						|
	return (aes_get_mbs(&entry->ae_sourcepath));
 | 
						|
}
 | 
						|
 | 
						|
const char *
 | 
						|
archive_entry_symlink(struct archive_entry *entry)
 | 
						|
{
 | 
						|
	if (entry->ae_set & AE_SET_SYMLINK)
 | 
						|
		return (aes_get_mbs(&entry->ae_symlink));
 | 
						|
	return (NULL);
 | 
						|
}
 | 
						|
 | 
						|
const wchar_t *
 | 
						|
archive_entry_symlink_w(struct archive_entry *entry)
 | 
						|
{
 | 
						|
	if (entry->ae_set & AE_SET_SYMLINK)
 | 
						|
		return (aes_get_wcs(&entry->ae_symlink));
 | 
						|
	return (NULL);
 | 
						|
}
 | 
						|
 | 
						|
uid_t
 | 
						|
archive_entry_uid(struct archive_entry *entry)
 | 
						|
{
 | 
						|
	return (entry->ae_stat.aest_uid);
 | 
						|
}
 | 
						|
 | 
						|
const char *
 | 
						|
archive_entry_uname(struct archive_entry *entry)
 | 
						|
{
 | 
						|
	return (aes_get_mbs(&entry->ae_uname));
 | 
						|
}
 | 
						|
 | 
						|
const wchar_t *
 | 
						|
archive_entry_uname_w(struct archive_entry *entry)
 | 
						|
{
 | 
						|
	return (aes_get_wcs(&entry->ae_uname));
 | 
						|
}
 | 
						|
 | 
						|
/*
 | 
						|
 * Functions to set archive_entry properties.
 | 
						|
 */
 | 
						|
 | 
						|
void
 | 
						|
archive_entry_set_filetype(struct archive_entry *entry, unsigned int type)
 | 
						|
{
 | 
						|
	entry->stat_valid = 0;
 | 
						|
	entry->ae_stat.aest_mode &= ~AE_IFMT;
 | 
						|
	entry->ae_stat.aest_mode |= AE_IFMT & type;
 | 
						|
}
 | 
						|
 | 
						|
void
 | 
						|
archive_entry_set_fflags(struct archive_entry *entry,
 | 
						|
    unsigned long set, unsigned long clear)
 | 
						|
{
 | 
						|
	aes_clean(&entry->ae_fflags_text);
 | 
						|
	entry->ae_fflags_set = set;
 | 
						|
	entry->ae_fflags_clear = clear;
 | 
						|
}
 | 
						|
 | 
						|
const char *
 | 
						|
archive_entry_copy_fflags_text(struct archive_entry *entry,
 | 
						|
    const char *flags)
 | 
						|
{
 | 
						|
	aes_copy_mbs(&entry->ae_fflags_text, flags);
 | 
						|
	return (ae_strtofflags(flags,
 | 
						|
		    &entry->ae_fflags_set, &entry->ae_fflags_clear));
 | 
						|
}
 | 
						|
 | 
						|
const wchar_t *
 | 
						|
archive_entry_copy_fflags_text_w(struct archive_entry *entry,
 | 
						|
    const wchar_t *flags)
 | 
						|
{
 | 
						|
	aes_copy_wcs(&entry->ae_fflags_text, flags);
 | 
						|
	return (ae_wcstofflags(flags,
 | 
						|
		    &entry->ae_fflags_set, &entry->ae_fflags_clear));
 | 
						|
}
 | 
						|
 | 
						|
void
 | 
						|
archive_entry_set_gid(struct archive_entry *entry, gid_t g)
 | 
						|
{
 | 
						|
	entry->stat_valid = 0;
 | 
						|
	entry->ae_stat.aest_gid = g;
 | 
						|
}
 | 
						|
 | 
						|
void
 | 
						|
archive_entry_set_gname(struct archive_entry *entry, const char *name)
 | 
						|
{
 | 
						|
	aes_set_mbs(&entry->ae_gname, name);
 | 
						|
}
 | 
						|
 | 
						|
void
 | 
						|
archive_entry_copy_gname(struct archive_entry *entry, const char *name)
 | 
						|
{
 | 
						|
	aes_copy_mbs(&entry->ae_gname, name);
 | 
						|
}
 | 
						|
 | 
						|
void
 | 
						|
archive_entry_copy_gname_w(struct archive_entry *entry, const wchar_t *name)
 | 
						|
{
 | 
						|
	aes_copy_wcs(&entry->ae_gname, name);
 | 
						|
}
 | 
						|
 | 
						|
int
 | 
						|
archive_entry_update_gname_utf8(struct archive_entry *entry, const char *name)
 | 
						|
{
 | 
						|
	return (aes_update_utf8(&entry->ae_gname, name));
 | 
						|
}
 | 
						|
 | 
						|
void
 | 
						|
archive_entry_set_ino(struct archive_entry *entry, unsigned long ino)
 | 
						|
{
 | 
						|
	entry->stat_valid = 0;
 | 
						|
	entry->ae_stat.aest_ino = ino;
 | 
						|
}
 | 
						|
 | 
						|
#ifndef __minix
 | 
						|
void
 | 
						|
archive_entry_set_ino64(struct archive_entry *entry, int64_t ino)
 | 
						|
{
 | 
						|
	entry->stat_valid = 0;
 | 
						|
	entry->ae_stat.aest_ino = ino;
 | 
						|
}
 | 
						|
#endif
 | 
						|
 | 
						|
void
 | 
						|
archive_entry_set_hardlink(struct archive_entry *entry, const char *target)
 | 
						|
{
 | 
						|
	aes_set_mbs(&entry->ae_hardlink, target);
 | 
						|
	if (target != NULL)
 | 
						|
		entry->ae_set |= AE_SET_HARDLINK;
 | 
						|
	else
 | 
						|
		entry->ae_set &= ~AE_SET_HARDLINK;
 | 
						|
}
 | 
						|
 | 
						|
void
 | 
						|
archive_entry_copy_hardlink(struct archive_entry *entry, const char *target)
 | 
						|
{
 | 
						|
	aes_copy_mbs(&entry->ae_hardlink, target);
 | 
						|
	if (target != NULL)
 | 
						|
		entry->ae_set |= AE_SET_HARDLINK;
 | 
						|
	else
 | 
						|
		entry->ae_set &= ~AE_SET_HARDLINK;
 | 
						|
}
 | 
						|
 | 
						|
void
 | 
						|
archive_entry_copy_hardlink_w(struct archive_entry *entry, const wchar_t *target)
 | 
						|
{
 | 
						|
	aes_copy_wcs(&entry->ae_hardlink, target);
 | 
						|
	if (target != NULL)
 | 
						|
		entry->ae_set |= AE_SET_HARDLINK;
 | 
						|
	else
 | 
						|
		entry->ae_set &= ~AE_SET_HARDLINK;
 | 
						|
}
 | 
						|
 | 
						|
int
 | 
						|
archive_entry_update_hardlink_utf8(struct archive_entry *entry, const char *target)
 | 
						|
{
 | 
						|
	if (target != NULL)
 | 
						|
		entry->ae_set |= AE_SET_HARDLINK;
 | 
						|
	else
 | 
						|
		entry->ae_set &= ~AE_SET_HARDLINK;
 | 
						|
	return (aes_update_utf8(&entry->ae_hardlink, target));
 | 
						|
}
 | 
						|
 | 
						|
void
 | 
						|
archive_entry_set_atime(struct archive_entry *entry, time_t t, long ns)
 | 
						|
{
 | 
						|
	entry->stat_valid = 0;
 | 
						|
	entry->ae_set |= AE_SET_ATIME;
 | 
						|
	entry->ae_stat.aest_atime = t;
 | 
						|
	entry->ae_stat.aest_atime_nsec = ns;
 | 
						|
}
 | 
						|
 | 
						|
void
 | 
						|
archive_entry_unset_atime(struct archive_entry *entry)
 | 
						|
{
 | 
						|
	archive_entry_set_atime(entry, 0, 0);
 | 
						|
	entry->ae_set &= ~AE_SET_ATIME;
 | 
						|
}
 | 
						|
 | 
						|
void
 | 
						|
archive_entry_set_birthtime(struct archive_entry *entry, time_t m, long ns)
 | 
						|
{
 | 
						|
	entry->stat_valid = 0;
 | 
						|
	entry->ae_set |= AE_SET_BIRTHTIME;
 | 
						|
	entry->ae_stat.aest_birthtime = m;
 | 
						|
	entry->ae_stat.aest_birthtime_nsec = ns;
 | 
						|
}
 | 
						|
 | 
						|
void
 | 
						|
archive_entry_unset_birthtime(struct archive_entry *entry)
 | 
						|
{
 | 
						|
	archive_entry_set_birthtime(entry, 0, 0);
 | 
						|
	entry->ae_set &= ~AE_SET_BIRTHTIME;
 | 
						|
}
 | 
						|
 | 
						|
void
 | 
						|
archive_entry_set_ctime(struct archive_entry *entry, time_t t, long ns)
 | 
						|
{
 | 
						|
	entry->stat_valid = 0;
 | 
						|
	entry->ae_set |= AE_SET_CTIME;
 | 
						|
	entry->ae_stat.aest_ctime = t;
 | 
						|
	entry->ae_stat.aest_ctime_nsec = ns;
 | 
						|
}
 | 
						|
 | 
						|
void
 | 
						|
archive_entry_unset_ctime(struct archive_entry *entry)
 | 
						|
{
 | 
						|
	archive_entry_set_ctime(entry, 0, 0);
 | 
						|
	entry->ae_set &= ~AE_SET_CTIME;
 | 
						|
}
 | 
						|
 | 
						|
void
 | 
						|
archive_entry_set_dev(struct archive_entry *entry, dev_t d)
 | 
						|
{
 | 
						|
	entry->stat_valid = 0;
 | 
						|
	entry->ae_stat.aest_dev_is_broken_down = 0;
 | 
						|
	entry->ae_stat.aest_dev = d;
 | 
						|
}
 | 
						|
 | 
						|
void
 | 
						|
archive_entry_set_devmajor(struct archive_entry *entry, dev_t m)
 | 
						|
{
 | 
						|
	entry->stat_valid = 0;
 | 
						|
	entry->ae_stat.aest_dev_is_broken_down = 1;
 | 
						|
	entry->ae_stat.aest_devmajor = m;
 | 
						|
}
 | 
						|
 | 
						|
void
 | 
						|
archive_entry_set_devminor(struct archive_entry *entry, dev_t m)
 | 
						|
{
 | 
						|
	entry->stat_valid = 0;
 | 
						|
	entry->ae_stat.aest_dev_is_broken_down = 1;
 | 
						|
	entry->ae_stat.aest_devminor = m;
 | 
						|
}
 | 
						|
 | 
						|
/* Set symlink if symlink is already set, else set hardlink. */
 | 
						|
void
 | 
						|
archive_entry_set_link(struct archive_entry *entry, const char *target)
 | 
						|
{
 | 
						|
	if (entry->ae_set & AE_SET_SYMLINK)
 | 
						|
		aes_set_mbs(&entry->ae_symlink, target);
 | 
						|
	else
 | 
						|
		aes_set_mbs(&entry->ae_hardlink, target);
 | 
						|
}
 | 
						|
 | 
						|
/* Set symlink if symlink is already set, else set hardlink. */
 | 
						|
void
 | 
						|
archive_entry_copy_link(struct archive_entry *entry, const char *target)
 | 
						|
{
 | 
						|
	if (entry->ae_set & AE_SET_SYMLINK)
 | 
						|
		aes_copy_mbs(&entry->ae_symlink, target);
 | 
						|
	else
 | 
						|
		aes_copy_mbs(&entry->ae_hardlink, target);
 | 
						|
}
 | 
						|
 | 
						|
/* Set symlink if symlink is already set, else set hardlink. */
 | 
						|
void
 | 
						|
archive_entry_copy_link_w(struct archive_entry *entry, const wchar_t *target)
 | 
						|
{
 | 
						|
	if (entry->ae_set & AE_SET_SYMLINK)
 | 
						|
		aes_copy_wcs(&entry->ae_symlink, target);
 | 
						|
	else
 | 
						|
		aes_copy_wcs(&entry->ae_hardlink, target);
 | 
						|
}
 | 
						|
 | 
						|
int
 | 
						|
archive_entry_update_link_utf8(struct archive_entry *entry, const char *target)
 | 
						|
{
 | 
						|
	if (entry->ae_set & AE_SET_SYMLINK)
 | 
						|
		return (aes_update_utf8(&entry->ae_symlink, target));
 | 
						|
	else
 | 
						|
		return (aes_update_utf8(&entry->ae_hardlink, target));
 | 
						|
}
 | 
						|
 | 
						|
void
 | 
						|
archive_entry_set_mode(struct archive_entry *entry, mode_t m)
 | 
						|
{
 | 
						|
	entry->stat_valid = 0;
 | 
						|
	entry->ae_stat.aest_mode = m;
 | 
						|
}
 | 
						|
 | 
						|
void
 | 
						|
archive_entry_set_mtime(struct archive_entry *entry, time_t m, long ns)
 | 
						|
{
 | 
						|
	entry->stat_valid = 0;
 | 
						|
	entry->ae_set |= AE_SET_MTIME;
 | 
						|
	entry->ae_stat.aest_mtime = m;
 | 
						|
	entry->ae_stat.aest_mtime_nsec = ns;
 | 
						|
}
 | 
						|
 | 
						|
void
 | 
						|
archive_entry_unset_mtime(struct archive_entry *entry)
 | 
						|
{
 | 
						|
	archive_entry_set_mtime(entry, 0, 0);
 | 
						|
	entry->ae_set &= ~AE_SET_MTIME;
 | 
						|
}
 | 
						|
 | 
						|
void
 | 
						|
archive_entry_set_nlink(struct archive_entry *entry, unsigned int nlink)
 | 
						|
{
 | 
						|
	entry->stat_valid = 0;
 | 
						|
	entry->ae_stat.aest_nlink = nlink;
 | 
						|
}
 | 
						|
 | 
						|
void
 | 
						|
archive_entry_set_pathname(struct archive_entry *entry, const char *name)
 | 
						|
{
 | 
						|
	aes_set_mbs(&entry->ae_pathname, name);
 | 
						|
}
 | 
						|
 | 
						|
void
 | 
						|
archive_entry_copy_pathname(struct archive_entry *entry, const char *name)
 | 
						|
{
 | 
						|
	aes_copy_mbs(&entry->ae_pathname, name);
 | 
						|
}
 | 
						|
 | 
						|
void
 | 
						|
archive_entry_copy_pathname_w(struct archive_entry *entry, const wchar_t *name)
 | 
						|
{
 | 
						|
	aes_copy_wcs(&entry->ae_pathname, name);
 | 
						|
}
 | 
						|
 | 
						|
int
 | 
						|
archive_entry_update_pathname_utf8(struct archive_entry *entry, const char *name)
 | 
						|
{
 | 
						|
	return (aes_update_utf8(&entry->ae_pathname, name));
 | 
						|
}
 | 
						|
 | 
						|
void
 | 
						|
archive_entry_set_perm(struct archive_entry *entry, mode_t p)
 | 
						|
{
 | 
						|
	entry->stat_valid = 0;
 | 
						|
	entry->ae_stat.aest_mode &= AE_IFMT;
 | 
						|
	entry->ae_stat.aest_mode |= ~AE_IFMT & p;
 | 
						|
}
 | 
						|
 | 
						|
void
 | 
						|
archive_entry_set_rdev(struct archive_entry *entry, dev_t m)
 | 
						|
{
 | 
						|
	entry->stat_valid = 0;
 | 
						|
	entry->ae_stat.aest_rdev = m;
 | 
						|
	entry->ae_stat.aest_rdev_is_broken_down = 0;
 | 
						|
}
 | 
						|
 | 
						|
void
 | 
						|
archive_entry_set_rdevmajor(struct archive_entry *entry, dev_t m)
 | 
						|
{
 | 
						|
	entry->stat_valid = 0;
 | 
						|
	entry->ae_stat.aest_rdev_is_broken_down = 1;
 | 
						|
	entry->ae_stat.aest_rdevmajor = m;
 | 
						|
}
 | 
						|
 | 
						|
void
 | 
						|
archive_entry_set_rdevminor(struct archive_entry *entry, dev_t m)
 | 
						|
{
 | 
						|
	entry->stat_valid = 0;
 | 
						|
	entry->ae_stat.aest_rdev_is_broken_down = 1;
 | 
						|
	entry->ae_stat.aest_rdevminor = m;
 | 
						|
}
 | 
						|
 | 
						|
#ifndef __minix
 | 
						|
void
 | 
						|
archive_entry_set_size(struct archive_entry *entry, int64_t s)
 | 
						|
{
 | 
						|
	entry->stat_valid = 0;
 | 
						|
	entry->ae_stat.aest_size = s;
 | 
						|
	entry->ae_set |= AE_SET_SIZE;
 | 
						|
}
 | 
						|
#else
 | 
						|
void
 | 
						|
archive_entry_set_size(struct archive_entry *entry, ssize_t s)
 | 
						|
{
 | 
						|
	entry->stat_valid = 0;
 | 
						|
	entry->ae_stat.aest_size = s;
 | 
						|
	entry->ae_set |= AE_SET_SIZE;
 | 
						|
}
 | 
						|
#endif
 | 
						|
 | 
						|
void
 | 
						|
archive_entry_unset_size(struct archive_entry *entry)
 | 
						|
{
 | 
						|
	archive_entry_set_size(entry, 0);
 | 
						|
	entry->ae_set &= ~AE_SET_SIZE;
 | 
						|
}
 | 
						|
 | 
						|
void
 | 
						|
archive_entry_copy_sourcepath(struct archive_entry *entry, const char *path)
 | 
						|
{
 | 
						|
	aes_set_mbs(&entry->ae_sourcepath, path);
 | 
						|
}
 | 
						|
 | 
						|
void
 | 
						|
archive_entry_set_symlink(struct archive_entry *entry, const char *linkname)
 | 
						|
{
 | 
						|
	aes_set_mbs(&entry->ae_symlink, linkname);
 | 
						|
	if (linkname != NULL)
 | 
						|
		entry->ae_set |= AE_SET_SYMLINK;
 | 
						|
	else
 | 
						|
		entry->ae_set &= ~AE_SET_SYMLINK;
 | 
						|
}
 | 
						|
 | 
						|
void
 | 
						|
archive_entry_copy_symlink(struct archive_entry *entry, const char *linkname)
 | 
						|
{
 | 
						|
	aes_copy_mbs(&entry->ae_symlink, linkname);
 | 
						|
	if (linkname != NULL)
 | 
						|
		entry->ae_set |= AE_SET_SYMLINK;
 | 
						|
	else
 | 
						|
		entry->ae_set &= ~AE_SET_SYMLINK;
 | 
						|
}
 | 
						|
 | 
						|
void
 | 
						|
archive_entry_copy_symlink_w(struct archive_entry *entry, const wchar_t *linkname)
 | 
						|
{
 | 
						|
	aes_copy_wcs(&entry->ae_symlink, linkname);
 | 
						|
	if (linkname != NULL)
 | 
						|
		entry->ae_set |= AE_SET_SYMLINK;
 | 
						|
	else
 | 
						|
		entry->ae_set &= ~AE_SET_SYMLINK;
 | 
						|
}
 | 
						|
 | 
						|
int
 | 
						|
archive_entry_update_symlink_utf8(struct archive_entry *entry, const char *linkname)
 | 
						|
{
 | 
						|
	if (linkname != NULL)
 | 
						|
		entry->ae_set |= AE_SET_SYMLINK;
 | 
						|
	else
 | 
						|
		entry->ae_set &= ~AE_SET_SYMLINK;
 | 
						|
	return (aes_update_utf8(&entry->ae_symlink, linkname));
 | 
						|
}
 | 
						|
 | 
						|
void
 | 
						|
archive_entry_set_uid(struct archive_entry *entry, uid_t u)
 | 
						|
{
 | 
						|
	entry->stat_valid = 0;
 | 
						|
	entry->ae_stat.aest_uid = u;
 | 
						|
}
 | 
						|
 | 
						|
void
 | 
						|
archive_entry_set_uname(struct archive_entry *entry, const char *name)
 | 
						|
{
 | 
						|
	aes_set_mbs(&entry->ae_uname, name);
 | 
						|
}
 | 
						|
 | 
						|
void
 | 
						|
archive_entry_copy_uname(struct archive_entry *entry, const char *name)
 | 
						|
{
 | 
						|
	aes_copy_mbs(&entry->ae_uname, name);
 | 
						|
}
 | 
						|
 | 
						|
void
 | 
						|
archive_entry_copy_uname_w(struct archive_entry *entry, const wchar_t *name)
 | 
						|
{
 | 
						|
	aes_copy_wcs(&entry->ae_uname, name);
 | 
						|
}
 | 
						|
 | 
						|
int
 | 
						|
archive_entry_update_uname_utf8(struct archive_entry *entry, const char *name)
 | 
						|
{
 | 
						|
	return (aes_update_utf8(&entry->ae_uname, name));
 | 
						|
}
 | 
						|
 | 
						|
/*
 | 
						|
 * ACL management.  The following would, of course, be a lot simpler
 | 
						|
 * if: 1) the last draft of POSIX.1e were a really thorough and
 | 
						|
 * complete standard that addressed the needs of ACL archiving and 2)
 | 
						|
 * everyone followed it faithfully.  Alas, neither is true, so the
 | 
						|
 * following is a lot more complex than might seem necessary to the
 | 
						|
 * uninitiated.
 | 
						|
 */
 | 
						|
 | 
						|
void
 | 
						|
archive_entry_acl_clear(struct archive_entry *entry)
 | 
						|
{
 | 
						|
	struct ae_acl	*ap;
 | 
						|
 | 
						|
	while (entry->acl_head != NULL) {
 | 
						|
		ap = entry->acl_head->next;
 | 
						|
		aes_clean(&entry->acl_head->name);
 | 
						|
		free(entry->acl_head);
 | 
						|
		entry->acl_head = ap;
 | 
						|
	}
 | 
						|
	if (entry->acl_text_w != NULL) {
 | 
						|
		free(entry->acl_text_w);
 | 
						|
		entry->acl_text_w = NULL;
 | 
						|
	}
 | 
						|
	entry->acl_p = NULL;
 | 
						|
	entry->acl_state = 0; /* Not counting. */
 | 
						|
}
 | 
						|
 | 
						|
/*
 | 
						|
 * Add a single ACL entry to the internal list of ACL data.
 | 
						|
 */
 | 
						|
void
 | 
						|
archive_entry_acl_add_entry(struct archive_entry *entry,
 | 
						|
    int type, int permset, int tag, int id, const char *name)
 | 
						|
{
 | 
						|
	struct ae_acl *ap;
 | 
						|
 | 
						|
	if (acl_special(entry, type, permset, tag) == 0)
 | 
						|
		return;
 | 
						|
	ap = acl_new_entry(entry, type, permset, tag, id);
 | 
						|
	if (ap == NULL) {
 | 
						|
		/* XXX Error XXX */
 | 
						|
		return;
 | 
						|
	}
 | 
						|
	if (name != NULL  &&  *name != '\0')
 | 
						|
		aes_copy_mbs(&ap->name, name);
 | 
						|
	else
 | 
						|
		aes_clean(&ap->name);
 | 
						|
}
 | 
						|
 | 
						|
/*
 | 
						|
 * As above, but with a wide-character name.
 | 
						|
 */
 | 
						|
void
 | 
						|
archive_entry_acl_add_entry_w(struct archive_entry *entry,
 | 
						|
    int type, int permset, int tag, int id, const wchar_t *name)
 | 
						|
{
 | 
						|
	archive_entry_acl_add_entry_w_len(entry, type, permset, tag, id, name, wcslen(name));
 | 
						|
}
 | 
						|
 | 
						|
static void
 | 
						|
archive_entry_acl_add_entry_w_len(struct archive_entry *entry,
 | 
						|
    int type, int permset, int tag, int id, const wchar_t *name, size_t len)
 | 
						|
{
 | 
						|
	struct ae_acl *ap;
 | 
						|
 | 
						|
	if (acl_special(entry, type, permset, tag) == 0)
 | 
						|
		return;
 | 
						|
	ap = acl_new_entry(entry, type, permset, tag, id);
 | 
						|
	if (ap == NULL) {
 | 
						|
		/* XXX Error XXX */
 | 
						|
		return;
 | 
						|
	}
 | 
						|
	if (name != NULL  &&  *name != L'\0' && len > 0)
 | 
						|
		aes_copy_wcs_len(&ap->name, name, len);
 | 
						|
	else
 | 
						|
		aes_clean(&ap->name);
 | 
						|
}
 | 
						|
 | 
						|
/*
 | 
						|
 * If this ACL entry is part of the standard POSIX permissions set,
 | 
						|
 * store the permissions in the stat structure and return zero.
 | 
						|
 */
 | 
						|
static int
 | 
						|
acl_special(struct archive_entry *entry, int type, int permset, int tag)
 | 
						|
{
 | 
						|
	if (type == ARCHIVE_ENTRY_ACL_TYPE_ACCESS) {
 | 
						|
		switch (tag) {
 | 
						|
		case ARCHIVE_ENTRY_ACL_USER_OBJ:
 | 
						|
			entry->ae_stat.aest_mode &= ~0700;
 | 
						|
			entry->ae_stat.aest_mode |= (permset & 7) << 6;
 | 
						|
			return (0);
 | 
						|
		case ARCHIVE_ENTRY_ACL_GROUP_OBJ:
 | 
						|
			entry->ae_stat.aest_mode &= ~0070;
 | 
						|
			entry->ae_stat.aest_mode |= (permset & 7) << 3;
 | 
						|
			return (0);
 | 
						|
		case ARCHIVE_ENTRY_ACL_OTHER:
 | 
						|
			entry->ae_stat.aest_mode &= ~0007;
 | 
						|
			entry->ae_stat.aest_mode |= permset & 7;
 | 
						|
			return (0);
 | 
						|
		}
 | 
						|
	}
 | 
						|
	return (1);
 | 
						|
}
 | 
						|
 | 
						|
/*
 | 
						|
 * Allocate and populate a new ACL entry with everything but the
 | 
						|
 * name.
 | 
						|
 */
 | 
						|
static struct ae_acl *
 | 
						|
acl_new_entry(struct archive_entry *entry,
 | 
						|
    int type, int permset, int tag, int id)
 | 
						|
{
 | 
						|
	struct ae_acl *ap, *aq;
 | 
						|
 | 
						|
	if (type != ARCHIVE_ENTRY_ACL_TYPE_ACCESS &&
 | 
						|
	    type != ARCHIVE_ENTRY_ACL_TYPE_DEFAULT)
 | 
						|
		return (NULL);
 | 
						|
	if (entry->acl_text_w != NULL) {
 | 
						|
		free(entry->acl_text_w);
 | 
						|
		entry->acl_text_w = NULL;
 | 
						|
	}
 | 
						|
 | 
						|
	/* XXX TODO: More sanity-checks on the arguments XXX */
 | 
						|
 | 
						|
	/* If there's a matching entry already in the list, overwrite it. */
 | 
						|
	ap = entry->acl_head;
 | 
						|
	aq = NULL;
 | 
						|
	while (ap != NULL) {
 | 
						|
		if (ap->type == type && ap->tag == tag && ap->id == id) {
 | 
						|
			ap->permset = permset;
 | 
						|
			return (ap);
 | 
						|
		}
 | 
						|
		aq = ap;
 | 
						|
		ap = ap->next;
 | 
						|
	}
 | 
						|
 | 
						|
	/* Add a new entry to the end of the list. */
 | 
						|
	ap = (struct ae_acl *)malloc(sizeof(*ap));
 | 
						|
	if (ap == NULL)
 | 
						|
		return (NULL);
 | 
						|
	memset(ap, 0, sizeof(*ap));
 | 
						|
	if (aq == NULL)
 | 
						|
		entry->acl_head = ap;
 | 
						|
	else
 | 
						|
		aq->next = ap;
 | 
						|
	ap->type = type;
 | 
						|
	ap->tag = tag;
 | 
						|
	ap->id = id;
 | 
						|
	ap->permset = permset;
 | 
						|
	return (ap);
 | 
						|
}
 | 
						|
 | 
						|
/*
 | 
						|
 * Return a count of entries matching "want_type".
 | 
						|
 */
 | 
						|
int
 | 
						|
archive_entry_acl_count(struct archive_entry *entry, int want_type)
 | 
						|
{
 | 
						|
	int count;
 | 
						|
	struct ae_acl *ap;
 | 
						|
 | 
						|
	count = 0;
 | 
						|
	ap = entry->acl_head;
 | 
						|
	while (ap != NULL) {
 | 
						|
		if ((ap->type & want_type) != 0)
 | 
						|
			count++;
 | 
						|
		ap = ap->next;
 | 
						|
	}
 | 
						|
 | 
						|
	if (count > 0 && ((want_type & ARCHIVE_ENTRY_ACL_TYPE_ACCESS) != 0))
 | 
						|
		count += 3;
 | 
						|
	return (count);
 | 
						|
}
 | 
						|
 | 
						|
/*
 | 
						|
 * Prepare for reading entries from the ACL data.  Returns a count
 | 
						|
 * of entries matching "want_type", or zero if there are no
 | 
						|
 * non-extended ACL entries of that type.
 | 
						|
 */
 | 
						|
int
 | 
						|
archive_entry_acl_reset(struct archive_entry *entry, int want_type)
 | 
						|
{
 | 
						|
	int count, cutoff;
 | 
						|
 | 
						|
	count = archive_entry_acl_count(entry, want_type);
 | 
						|
 | 
						|
	/*
 | 
						|
	 * If the only entries are the three standard ones,
 | 
						|
	 * then don't return any ACL data.  (In this case,
 | 
						|
	 * client can just use chmod(2) to set permissions.)
 | 
						|
	 */
 | 
						|
	if ((want_type & ARCHIVE_ENTRY_ACL_TYPE_ACCESS) != 0)
 | 
						|
		cutoff = 3;
 | 
						|
	else
 | 
						|
		cutoff = 0;
 | 
						|
 | 
						|
	if (count > cutoff)
 | 
						|
		entry->acl_state = ARCHIVE_ENTRY_ACL_USER_OBJ;
 | 
						|
	else
 | 
						|
		entry->acl_state = 0;
 | 
						|
	entry->acl_p = entry->acl_head;
 | 
						|
	return (count);
 | 
						|
}
 | 
						|
 | 
						|
/*
 | 
						|
 * Return the next ACL entry in the list.  Fake entries for the
 | 
						|
 * standard permissions and include them in the returned list.
 | 
						|
 */
 | 
						|
 | 
						|
int
 | 
						|
archive_entry_acl_next(struct archive_entry *entry, int want_type, int *type,
 | 
						|
    int *permset, int *tag, int *id, const char **name)
 | 
						|
{
 | 
						|
	*name = NULL;
 | 
						|
	*id = -1;
 | 
						|
 | 
						|
	/*
 | 
						|
	 * The acl_state is either zero (no entries available), -1
 | 
						|
	 * (reading from list), or an entry type (retrieve that type
 | 
						|
	 * from ae_stat.aest_mode).
 | 
						|
	 */
 | 
						|
	if (entry->acl_state == 0)
 | 
						|
		return (ARCHIVE_WARN);
 | 
						|
 | 
						|
	/* The first three access entries are special. */
 | 
						|
	if ((want_type & ARCHIVE_ENTRY_ACL_TYPE_ACCESS) != 0) {
 | 
						|
		switch (entry->acl_state) {
 | 
						|
		case ARCHIVE_ENTRY_ACL_USER_OBJ:
 | 
						|
			*permset = (entry->ae_stat.aest_mode >> 6) & 7;
 | 
						|
			*type = ARCHIVE_ENTRY_ACL_TYPE_ACCESS;
 | 
						|
			*tag = ARCHIVE_ENTRY_ACL_USER_OBJ;
 | 
						|
			entry->acl_state = ARCHIVE_ENTRY_ACL_GROUP_OBJ;
 | 
						|
			return (ARCHIVE_OK);
 | 
						|
		case ARCHIVE_ENTRY_ACL_GROUP_OBJ:
 | 
						|
			*permset = (entry->ae_stat.aest_mode >> 3) & 7;
 | 
						|
			*type = ARCHIVE_ENTRY_ACL_TYPE_ACCESS;
 | 
						|
			*tag = ARCHIVE_ENTRY_ACL_GROUP_OBJ;
 | 
						|
			entry->acl_state = ARCHIVE_ENTRY_ACL_OTHER;
 | 
						|
			return (ARCHIVE_OK);
 | 
						|
		case ARCHIVE_ENTRY_ACL_OTHER:
 | 
						|
			*permset = entry->ae_stat.aest_mode & 7;
 | 
						|
			*type = ARCHIVE_ENTRY_ACL_TYPE_ACCESS;
 | 
						|
			*tag = ARCHIVE_ENTRY_ACL_OTHER;
 | 
						|
			entry->acl_state = -1;
 | 
						|
			entry->acl_p = entry->acl_head;
 | 
						|
			return (ARCHIVE_OK);
 | 
						|
		default:
 | 
						|
			break;
 | 
						|
		}
 | 
						|
	}
 | 
						|
 | 
						|
	while (entry->acl_p != NULL && (entry->acl_p->type & want_type) == 0)
 | 
						|
		entry->acl_p = entry->acl_p->next;
 | 
						|
	if (entry->acl_p == NULL) {
 | 
						|
		entry->acl_state = 0;
 | 
						|
		*type = 0;
 | 
						|
		*permset = 0;
 | 
						|
		*tag = 0;
 | 
						|
		*id = -1;
 | 
						|
		*name = NULL;
 | 
						|
		return (ARCHIVE_EOF); /* End of ACL entries. */
 | 
						|
	}
 | 
						|
	*type = entry->acl_p->type;
 | 
						|
	*permset = entry->acl_p->permset;
 | 
						|
	*tag = entry->acl_p->tag;
 | 
						|
	*id = entry->acl_p->id;
 | 
						|
	*name = aes_get_mbs(&entry->acl_p->name);
 | 
						|
	entry->acl_p = entry->acl_p->next;
 | 
						|
	return (ARCHIVE_OK);
 | 
						|
}
 | 
						|
 | 
						|
/*
 | 
						|
 * Generate a text version of the ACL.  The flags parameter controls
 | 
						|
 * the style of the generated ACL.
 | 
						|
 */
 | 
						|
const wchar_t *
 | 
						|
archive_entry_acl_text_w(struct archive_entry *entry, int flags)
 | 
						|
{
 | 
						|
	int count;
 | 
						|
	size_t length;
 | 
						|
	const wchar_t *wname;
 | 
						|
	const wchar_t *prefix;
 | 
						|
	wchar_t separator;
 | 
						|
	struct ae_acl *ap;
 | 
						|
	int id;
 | 
						|
	wchar_t *wp;
 | 
						|
 | 
						|
	if (entry->acl_text_w != NULL) {
 | 
						|
		free (entry->acl_text_w);
 | 
						|
		entry->acl_text_w = NULL;
 | 
						|
	}
 | 
						|
 | 
						|
	separator = L',';
 | 
						|
	count = 0;
 | 
						|
	length = 0;
 | 
						|
	ap = entry->acl_head;
 | 
						|
	while (ap != NULL) {
 | 
						|
		if ((ap->type & flags) != 0) {
 | 
						|
			count++;
 | 
						|
			if ((flags & ARCHIVE_ENTRY_ACL_STYLE_MARK_DEFAULT) &&
 | 
						|
			    (ap->type & ARCHIVE_ENTRY_ACL_TYPE_DEFAULT))
 | 
						|
				length += 8; /* "default:" */
 | 
						|
			length += 5; /* tag name */
 | 
						|
			length += 1; /* colon */
 | 
						|
			wname = aes_get_wcs(&ap->name);
 | 
						|
			if (wname != NULL)
 | 
						|
				length += wcslen(wname);
 | 
						|
			else
 | 
						|
				length += sizeof(uid_t) * 3 + 1;
 | 
						|
			length ++; /* colon */
 | 
						|
			length += 3; /* rwx */
 | 
						|
			length += 1; /* colon */
 | 
						|
			length += max(sizeof(uid_t), sizeof(gid_t)) * 3 + 1;
 | 
						|
			length ++; /* newline */
 | 
						|
		}
 | 
						|
		ap = ap->next;
 | 
						|
	}
 | 
						|
 | 
						|
	if (count > 0 && ((flags & ARCHIVE_ENTRY_ACL_TYPE_ACCESS) != 0)) {
 | 
						|
		length += 10; /* "user::rwx\n" */
 | 
						|
		length += 11; /* "group::rwx\n" */
 | 
						|
		length += 11; /* "other::rwx\n" */
 | 
						|
	}
 | 
						|
 | 
						|
	if (count == 0)
 | 
						|
		return (NULL);
 | 
						|
 | 
						|
	/* Now, allocate the string and actually populate it. */
 | 
						|
	wp = entry->acl_text_w = (wchar_t *)malloc(length * sizeof(wchar_t));
 | 
						|
	if (wp == NULL)
 | 
						|
		__archive_errx(1, "No memory to generate the text version of the ACL");
 | 
						|
	count = 0;
 | 
						|
	if ((flags & ARCHIVE_ENTRY_ACL_TYPE_ACCESS) != 0) {
 | 
						|
		append_entry_w(&wp, NULL, ARCHIVE_ENTRY_ACL_USER_OBJ, NULL,
 | 
						|
		    entry->ae_stat.aest_mode & 0700, -1);
 | 
						|
		*wp++ = ',';
 | 
						|
		append_entry_w(&wp, NULL, ARCHIVE_ENTRY_ACL_GROUP_OBJ, NULL,
 | 
						|
		    entry->ae_stat.aest_mode & 0070, -1);
 | 
						|
		*wp++ = ',';
 | 
						|
		append_entry_w(&wp, NULL, ARCHIVE_ENTRY_ACL_OTHER, NULL,
 | 
						|
		    entry->ae_stat.aest_mode & 0007, -1);
 | 
						|
		count += 3;
 | 
						|
 | 
						|
		ap = entry->acl_head;
 | 
						|
		while (ap != NULL) {
 | 
						|
			if ((ap->type & ARCHIVE_ENTRY_ACL_TYPE_ACCESS) != 0) {
 | 
						|
				wname = aes_get_wcs(&ap->name);
 | 
						|
				*wp++ = separator;
 | 
						|
				if (flags & ARCHIVE_ENTRY_ACL_STYLE_EXTRA_ID)
 | 
						|
					id = ap->id;
 | 
						|
				else
 | 
						|
					id = -1;
 | 
						|
				append_entry_w(&wp, NULL, ap->tag, wname,
 | 
						|
				    ap->permset, id);
 | 
						|
				count++;
 | 
						|
			}
 | 
						|
			ap = ap->next;
 | 
						|
		}
 | 
						|
	}
 | 
						|
 | 
						|
 | 
						|
	if ((flags & ARCHIVE_ENTRY_ACL_TYPE_DEFAULT) != 0) {
 | 
						|
		if (flags & ARCHIVE_ENTRY_ACL_STYLE_MARK_DEFAULT)
 | 
						|
			prefix = L"default:";
 | 
						|
		else
 | 
						|
			prefix = NULL;
 | 
						|
		ap = entry->acl_head;
 | 
						|
		count = 0;
 | 
						|
		while (ap != NULL) {
 | 
						|
			if ((ap->type & ARCHIVE_ENTRY_ACL_TYPE_DEFAULT) != 0) {
 | 
						|
				wname = aes_get_wcs(&ap->name);
 | 
						|
				if (count > 0)
 | 
						|
					*wp++ = separator;
 | 
						|
				if (flags & ARCHIVE_ENTRY_ACL_STYLE_EXTRA_ID)
 | 
						|
					id = ap->id;
 | 
						|
				else
 | 
						|
					id = -1;
 | 
						|
				append_entry_w(&wp, prefix, ap->tag,
 | 
						|
				    wname, ap->permset, id);
 | 
						|
				count ++;
 | 
						|
			}
 | 
						|
			ap = ap->next;
 | 
						|
		}
 | 
						|
	}
 | 
						|
 | 
						|
	return (entry->acl_text_w);
 | 
						|
}
 | 
						|
 | 
						|
static void
 | 
						|
append_id_w(wchar_t **wp, int id)
 | 
						|
{
 | 
						|
	if (id < 0)
 | 
						|
		id = 0;
 | 
						|
	if (id > 9)
 | 
						|
		append_id_w(wp, id / 10);
 | 
						|
	*(*wp)++ = L"0123456789"[id % 10];
 | 
						|
}
 | 
						|
 | 
						|
static void
 | 
						|
append_entry_w(wchar_t **wp, const wchar_t *prefix, int tag,
 | 
						|
    const wchar_t *wname, int perm, int id)
 | 
						|
{
 | 
						|
	if (prefix != NULL) {
 | 
						|
		wcscpy(*wp, prefix);
 | 
						|
		*wp += wcslen(*wp);
 | 
						|
	}
 | 
						|
	switch (tag) {
 | 
						|
	case ARCHIVE_ENTRY_ACL_USER_OBJ:
 | 
						|
		wname = NULL;
 | 
						|
		id = -1;
 | 
						|
		/* FALLTHROUGH */
 | 
						|
	case ARCHIVE_ENTRY_ACL_USER:
 | 
						|
		wcscpy(*wp, L"user");
 | 
						|
		break;
 | 
						|
	case ARCHIVE_ENTRY_ACL_GROUP_OBJ:
 | 
						|
		wname = NULL;
 | 
						|
		id = -1;
 | 
						|
		/* FALLTHROUGH */
 | 
						|
	case ARCHIVE_ENTRY_ACL_GROUP:
 | 
						|
		wcscpy(*wp, L"group");
 | 
						|
		break;
 | 
						|
	case ARCHIVE_ENTRY_ACL_MASK:
 | 
						|
		wcscpy(*wp, L"mask");
 | 
						|
		wname = NULL;
 | 
						|
		id = -1;
 | 
						|
		break;
 | 
						|
	case ARCHIVE_ENTRY_ACL_OTHER:
 | 
						|
		wcscpy(*wp, L"other");
 | 
						|
		wname = NULL;
 | 
						|
		id = -1;
 | 
						|
		break;
 | 
						|
	}
 | 
						|
	*wp += wcslen(*wp);
 | 
						|
	*(*wp)++ = L':';
 | 
						|
	if (wname != NULL) {
 | 
						|
		wcscpy(*wp, wname);
 | 
						|
		*wp += wcslen(*wp);
 | 
						|
	} else if (tag == ARCHIVE_ENTRY_ACL_USER
 | 
						|
	    || tag == ARCHIVE_ENTRY_ACL_GROUP) {
 | 
						|
		append_id_w(wp, id);
 | 
						|
		id = -1;
 | 
						|
	}
 | 
						|
	*(*wp)++ = L':';
 | 
						|
	*(*wp)++ = (perm & 0444) ? L'r' : L'-';
 | 
						|
	*(*wp)++ = (perm & 0222) ? L'w' : L'-';
 | 
						|
	*(*wp)++ = (perm & 0111) ? L'x' : L'-';
 | 
						|
	if (id != -1) {
 | 
						|
		*(*wp)++ = L':';
 | 
						|
		append_id_w(wp, id);
 | 
						|
	}
 | 
						|
	**wp = L'\0';
 | 
						|
}
 | 
						|
 | 
						|
/*
 | 
						|
 * Parse a textual ACL.  This automatically recognizes and supports
 | 
						|
 * extensions described above.  The 'type' argument is used to
 | 
						|
 * indicate the type that should be used for any entries not
 | 
						|
 * explicitly marked as "default:".
 | 
						|
 */
 | 
						|
int
 | 
						|
__archive_entry_acl_parse_w(struct archive_entry *entry,
 | 
						|
    const wchar_t *text, int default_type)
 | 
						|
{
 | 
						|
	struct {
 | 
						|
		const wchar_t *start;
 | 
						|
		const wchar_t *end;
 | 
						|
	} field[4], name;
 | 
						|
 | 
						|
	int fields, n;
 | 
						|
	int type, tag, permset, id;
 | 
						|
	wchar_t sep;
 | 
						|
 | 
						|
	while (text != NULL  &&  *text != L'\0') {
 | 
						|
		/*
 | 
						|
		 * Parse the fields out of the next entry,
 | 
						|
		 * advance 'text' to start of next entry.
 | 
						|
		 */
 | 
						|
		fields = 0;
 | 
						|
		do {
 | 
						|
			const wchar_t *start, *end;
 | 
						|
			next_field_w(&text, &start, &end, &sep);
 | 
						|
			if (fields < 4) {
 | 
						|
				field[fields].start = start;
 | 
						|
				field[fields].end = end;
 | 
						|
			}
 | 
						|
			++fields;
 | 
						|
		} while (sep == L':');
 | 
						|
 | 
						|
		/* Set remaining fields to blank. */
 | 
						|
		for (n = fields; n < 4; ++n)
 | 
						|
			field[n].start = field[n].end = NULL;
 | 
						|
 | 
						|
		/* Check for a numeric ID in field 1 or 3. */
 | 
						|
		id = -1;
 | 
						|
		isint_w(field[1].start, field[1].end, &id);
 | 
						|
		/* Field 3 is optional. */
 | 
						|
		if (id == -1 && fields > 3)
 | 
						|
			isint_w(field[3].start, field[3].end, &id);
 | 
						|
 | 
						|
		/*
 | 
						|
		 * Solaris extension:  "defaultuser::rwx" is the
 | 
						|
		 * default ACL corresponding to "user::rwx", etc.
 | 
						|
		 */
 | 
						|
		if (field[0].end - field[0].start > 7
 | 
						|
		    && wmemcmp(field[0].start, L"default", 7) == 0) {
 | 
						|
			type = ARCHIVE_ENTRY_ACL_TYPE_DEFAULT;
 | 
						|
			field[0].start += 7;
 | 
						|
		} else
 | 
						|
			type = default_type;
 | 
						|
 | 
						|
		name.start = name.end = NULL;
 | 
						|
		if (prefix_w(field[0].start, field[0].end, L"user")) {
 | 
						|
			if (!ismode_w(field[2].start, field[2].end, &permset))
 | 
						|
				return (ARCHIVE_WARN);
 | 
						|
			if (id != -1 || field[1].start < field[1].end) {
 | 
						|
				tag = ARCHIVE_ENTRY_ACL_USER;
 | 
						|
				name = field[1];
 | 
						|
			} else
 | 
						|
				tag = ARCHIVE_ENTRY_ACL_USER_OBJ;
 | 
						|
		} else if (prefix_w(field[0].start, field[0].end, L"group")) {
 | 
						|
			if (!ismode_w(field[2].start, field[2].end, &permset))
 | 
						|
				return (ARCHIVE_WARN);
 | 
						|
			if (id != -1 || field[1].start < field[1].end) {
 | 
						|
				tag = ARCHIVE_ENTRY_ACL_GROUP;
 | 
						|
				name = field[1];
 | 
						|
			} else
 | 
						|
				tag = ARCHIVE_ENTRY_ACL_GROUP_OBJ;
 | 
						|
		} else if (prefix_w(field[0].start, field[0].end, L"other")) {
 | 
						|
			if (fields == 2
 | 
						|
			    && field[1].start < field[1].end
 | 
						|
			    && ismode_w(field[1].start, field[1].end, &permset)) {
 | 
						|
				/* This is Solaris-style "other:rwx" */
 | 
						|
			} else if (fields == 3
 | 
						|
			    && field[1].start == field[1].end
 | 
						|
			    && field[2].start < field[2].end
 | 
						|
			    && ismode_w(field[2].start, field[2].end, &permset)) {
 | 
						|
				/* This is FreeBSD-style "other::rwx" */
 | 
						|
			} else
 | 
						|
				return (ARCHIVE_WARN);
 | 
						|
			tag = ARCHIVE_ENTRY_ACL_OTHER;
 | 
						|
		} else if (prefix_w(field[0].start, field[0].end, L"mask")) {
 | 
						|
			if (fields == 2
 | 
						|
			    && field[1].start < field[1].end
 | 
						|
			    && ismode_w(field[1].start, field[1].end, &permset)) {
 | 
						|
				/* This is Solaris-style "mask:rwx" */
 | 
						|
			} else if (fields == 3
 | 
						|
			    && field[1].start == field[1].end
 | 
						|
			    && field[2].start < field[2].end
 | 
						|
			    && ismode_w(field[2].start, field[2].end, &permset)) {
 | 
						|
				/* This is FreeBSD-style "mask::rwx" */
 | 
						|
			} else
 | 
						|
				return (ARCHIVE_WARN);
 | 
						|
			tag = ARCHIVE_ENTRY_ACL_MASK;
 | 
						|
		} else
 | 
						|
			return (ARCHIVE_WARN);
 | 
						|
 | 
						|
		/* Add entry to the internal list. */
 | 
						|
		archive_entry_acl_add_entry_w_len(entry, type, permset,
 | 
						|
		    tag, id, name.start, name.end - name.start);
 | 
						|
	}
 | 
						|
	return (ARCHIVE_OK);
 | 
						|
}
 | 
						|
 | 
						|
/*
 | 
						|
 * Parse a string to a positive decimal integer.  Returns true if
 | 
						|
 * the string is non-empty and consists only of decimal digits,
 | 
						|
 * false otherwise.
 | 
						|
 */
 | 
						|
static int
 | 
						|
isint_w(const wchar_t *start, const wchar_t *end, int *result)
 | 
						|
{
 | 
						|
	int n = 0;
 | 
						|
	if (start >= end)
 | 
						|
		return (0);
 | 
						|
	while (start < end) {
 | 
						|
		if (*start < '0' || *start > '9')
 | 
						|
			return (0);
 | 
						|
		if (n > (INT_MAX / 10))
 | 
						|
			n = INT_MAX;
 | 
						|
		else {
 | 
						|
			n *= 10;
 | 
						|
			n += *start - '0';
 | 
						|
		}
 | 
						|
		start++;
 | 
						|
	}
 | 
						|
	*result = n;
 | 
						|
	return (1);
 | 
						|
}
 | 
						|
 | 
						|
/*
 | 
						|
 * Parse a string as a mode field.  Returns true if
 | 
						|
 * the string is non-empty and consists only of mode characters,
 | 
						|
 * false otherwise.
 | 
						|
 */
 | 
						|
static int
 | 
						|
ismode_w(const wchar_t *start, const wchar_t *end, int *permset)
 | 
						|
{
 | 
						|
	const wchar_t *p;
 | 
						|
 | 
						|
	if (start >= end)
 | 
						|
		return (0);
 | 
						|
	p = start;
 | 
						|
	*permset = 0;
 | 
						|
	while (p < end) {
 | 
						|
		switch (*p++) {
 | 
						|
		case 'r': case 'R':
 | 
						|
			*permset |= ARCHIVE_ENTRY_ACL_READ;
 | 
						|
			break;
 | 
						|
		case 'w': case 'W':
 | 
						|
			*permset |= ARCHIVE_ENTRY_ACL_WRITE;
 | 
						|
			break;
 | 
						|
		case 'x': case 'X':
 | 
						|
			*permset |= ARCHIVE_ENTRY_ACL_EXECUTE;
 | 
						|
			break;
 | 
						|
		case '-':
 | 
						|
			break;
 | 
						|
		default:
 | 
						|
			return (0);
 | 
						|
		}
 | 
						|
	}
 | 
						|
	return (1);
 | 
						|
}
 | 
						|
 | 
						|
/*
 | 
						|
 * Match "[:whitespace:]*(.*)[:whitespace:]*[:,\n]".  *wp is updated
 | 
						|
 * to point to just after the separator.  *start points to the first
 | 
						|
 * character of the matched text and *end just after the last
 | 
						|
 * character of the matched identifier.  In particular *end - *start
 | 
						|
 * is the length of the field body, not including leading or trailing
 | 
						|
 * whitespace.
 | 
						|
 */
 | 
						|
static void
 | 
						|
next_field_w(const wchar_t **wp, const wchar_t **start,
 | 
						|
    const wchar_t **end, wchar_t *sep)
 | 
						|
{
 | 
						|
	/* Skip leading whitespace to find start of field. */
 | 
						|
	while (**wp == L' ' || **wp == L'\t' || **wp == L'\n') {
 | 
						|
		(*wp)++;
 | 
						|
	}
 | 
						|
	*start = *wp;
 | 
						|
 | 
						|
	/* Scan for the separator. */
 | 
						|
	while (**wp != L'\0' && **wp != L',' && **wp != L':' &&
 | 
						|
	    **wp != L'\n') {
 | 
						|
		(*wp)++;
 | 
						|
	}
 | 
						|
	*sep = **wp;
 | 
						|
 | 
						|
	/* Trim trailing whitespace to locate end of field. */
 | 
						|
	*end = *wp - 1;
 | 
						|
	while (**end == L' ' || **end == L'\t' || **end == L'\n') {
 | 
						|
		(*end)--;
 | 
						|
	}
 | 
						|
	(*end)++;
 | 
						|
 | 
						|
	/* Adjust scanner location. */
 | 
						|
	if (**wp != L'\0')
 | 
						|
		(*wp)++;
 | 
						|
}
 | 
						|
 | 
						|
/*
 | 
						|
 * Return true if the characters [start...end) are a prefix of 'test'.
 | 
						|
 * This makes it easy to handle the obvious abbreviations: 'u' for 'user', etc.
 | 
						|
 */
 | 
						|
static int
 | 
						|
prefix_w(const wchar_t *start, const wchar_t *end, const wchar_t *test)
 | 
						|
{
 | 
						|
	if (start == end)
 | 
						|
		return (0);
 | 
						|
 | 
						|
	if (*start++ != *test++)
 | 
						|
		return (0);
 | 
						|
 | 
						|
	while (start < end  &&  *start++ == *test++)
 | 
						|
		;
 | 
						|
 | 
						|
	if (start < end)
 | 
						|
		return (0);
 | 
						|
 | 
						|
	return (1);
 | 
						|
}
 | 
						|
 | 
						|
 | 
						|
/*
 | 
						|
 * Following code is modified from UC Berkeley sources, and
 | 
						|
 * is subject to the following copyright notice.
 | 
						|
 */
 | 
						|
 | 
						|
/*-
 | 
						|
 * Copyright (c) 1993
 | 
						|
 *	The Regents of the University of California.  All rights reserved.
 | 
						|
 *
 | 
						|
 * Redistribution and use in source and binary forms, with or without
 | 
						|
 * modification, are permitted provided that the following conditions
 | 
						|
 * are met:
 | 
						|
 * 1. Redistributions of source code must retain the above copyright
 | 
						|
 *    notice, this list of conditions and the following disclaimer.
 | 
						|
 * 2. Redistributions in binary form must reproduce the above copyright
 | 
						|
 *    notice, this list of conditions and the following disclaimer in the
 | 
						|
 *    documentation and/or other materials provided with the distribution.
 | 
						|
 * 4. Neither the name of the University nor the names of its contributors
 | 
						|
 *    may be used to endorse or promote products derived from this software
 | 
						|
 *    without specific prior written permission.
 | 
						|
 *
 | 
						|
 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
 | 
						|
 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
 | 
						|
 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
 | 
						|
 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
 | 
						|
 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
 | 
						|
 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
 | 
						|
 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
 | 
						|
 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
 | 
						|
 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
 | 
						|
 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
 | 
						|
 * SUCH DAMAGE.
 | 
						|
 */
 | 
						|
 | 
						|
static struct flag {
 | 
						|
	const char	*name;
 | 
						|
	const wchar_t	*wname;
 | 
						|
	unsigned long	 set;
 | 
						|
	unsigned long	 clear;
 | 
						|
} flags[] = {
 | 
						|
	/* Preferred (shorter) names per flag first, all prefixed by "no" */
 | 
						|
#ifdef SF_APPEND
 | 
						|
	{ "nosappnd",	L"nosappnd",		SF_APPEND,	0 },
 | 
						|
	{ "nosappend",	L"nosappend",		SF_APPEND,	0 },
 | 
						|
#endif
 | 
						|
#ifdef  EXT2_APPEND_FL				/* 'a' */
 | 
						|
	{ "nosappnd",	L"nosappnd",		EXT2_APPEND_FL,	0 },
 | 
						|
	{ "nosappend",	L"nosappend",		EXT2_APPEND_FL,	0 },
 | 
						|
#endif
 | 
						|
#ifdef SF_ARCHIVED
 | 
						|
	{ "noarch",	L"noarch",		SF_ARCHIVED,	0 },
 | 
						|
	{ "noarchived",	L"noarchived",       	SF_ARCHIVED,	0 },
 | 
						|
#endif
 | 
						|
#ifdef SF_IMMUTABLE
 | 
						|
	{ "noschg",	L"noschg",		SF_IMMUTABLE,	0 },
 | 
						|
	{ "noschange",	L"noschange",		SF_IMMUTABLE,	0 },
 | 
						|
	{ "nosimmutable",	L"nosimmutable",	SF_IMMUTABLE,	0 },
 | 
						|
#endif
 | 
						|
#ifdef EXT2_IMMUTABLE_FL			/* 'i' */
 | 
						|
	{ "noschg",	L"noschg",		EXT2_IMMUTABLE_FL,	0 },
 | 
						|
	{ "noschange",	L"noschange",		EXT2_IMMUTABLE_FL,	0 },
 | 
						|
	{ "nosimmutable",	L"nosimmutable",	EXT2_IMMUTABLE_FL,	0 },
 | 
						|
#endif
 | 
						|
#ifdef SF_NOUNLINK
 | 
						|
	{ "nosunlnk",	L"nosunlnk",		SF_NOUNLINK,	0 },
 | 
						|
	{ "nosunlink",	L"nosunlink",		SF_NOUNLINK,	0 },
 | 
						|
#endif
 | 
						|
#ifdef SF_SNAPSHOT
 | 
						|
	{ "nosnapshot",	L"nosnapshot",	SF_SNAPSHOT,	0 },
 | 
						|
#endif
 | 
						|
#ifdef UF_APPEND
 | 
						|
	{ "nouappnd",	L"nouappnd",		UF_APPEND,	0 },
 | 
						|
	{ "nouappend",	L"nouappend",		UF_APPEND,	0 },
 | 
						|
#endif
 | 
						|
#ifdef UF_IMMUTABLE
 | 
						|
	{ "nouchg",	L"nouchg",		UF_IMMUTABLE,	0 },
 | 
						|
	{ "nouchange",	L"nouchange",		UF_IMMUTABLE,	0 },
 | 
						|
	{ "nouimmutable",	L"nouimmutable",	UF_IMMUTABLE,	0 },
 | 
						|
#endif
 | 
						|
#ifdef UF_NODUMP
 | 
						|
	{ "nodump",	L"nodump",		0,		UF_NODUMP},
 | 
						|
#endif
 | 
						|
#ifdef EXT2_NODUMP_FL				/* 'd' */
 | 
						|
	{ "nodump",	L"nodump",		0,		EXT2_NODUMP_FL},
 | 
						|
#endif
 | 
						|
#ifdef UF_OPAQUE
 | 
						|
	{ "noopaque",	L"noopaque",		UF_OPAQUE,	0 },
 | 
						|
#endif
 | 
						|
#ifdef UF_NOUNLINK
 | 
						|
	{ "nouunlnk",	L"nouunlnk",		UF_NOUNLINK,	0 },
 | 
						|
	{ "nouunlink",	L"nouunlink",		UF_NOUNLINK,	0 },
 | 
						|
#endif
 | 
						|
#ifdef EXT2_UNRM_FL
 | 
						|
        { "nouunlink",	L"nouunlink",		EXT2_UNRM_FL,	0},
 | 
						|
#endif
 | 
						|
 | 
						|
#ifdef EXT2_BTREE_FL
 | 
						|
        { "nobtree",	L"nobtree",       	EXT2_BTREE_FL,	0 },
 | 
						|
#endif
 | 
						|
 | 
						|
#ifdef EXT2_ECOMPR_FL
 | 
						|
        { "nocomperr",	L"nocomperr",       	EXT2_ECOMPR_FL,	0 },
 | 
						|
#endif
 | 
						|
 | 
						|
#ifdef EXT2_COMPR_FL				/* 'c' */
 | 
						|
        { "nocompress",	L"nocompress",       	EXT2_COMPR_FL,	0 },
 | 
						|
#endif
 | 
						|
 | 
						|
#ifdef EXT2_NOATIME_FL				/* 'A' */
 | 
						|
        { "noatime",	L"noatime",		0,		EXT2_NOATIME_FL},
 | 
						|
#endif
 | 
						|
 | 
						|
#ifdef EXT2_DIRTY_FL
 | 
						|
        { "nocompdirty",L"nocompdirty",		EXT2_DIRTY_FL,		0},
 | 
						|
#endif
 | 
						|
 | 
						|
#ifdef EXT2_COMPRBLK_FL
 | 
						|
#ifdef EXT2_NOCOMPR_FL
 | 
						|
        { "nocomprblk",	L"nocomprblk",		EXT2_COMPRBLK_FL, EXT2_NOCOMPR_FL},
 | 
						|
#else
 | 
						|
        { "nocomprblk",	L"nocomprblk",		EXT2_COMPRBLK_FL,	0},
 | 
						|
#endif
 | 
						|
#endif
 | 
						|
#ifdef EXT2_DIRSYNC_FL
 | 
						|
        { "nodirsync",	L"nodirsync",		EXT2_DIRSYNC_FL,	0},
 | 
						|
#endif
 | 
						|
#ifdef EXT2_INDEX_FL
 | 
						|
        { "nohashidx",	L"nohashidx",		EXT2_INDEX_FL,		0},
 | 
						|
#endif
 | 
						|
#ifdef EXT2_IMAGIC_FL
 | 
						|
        { "noimagic",	L"noimagic",		EXT2_IMAGIC_FL,		0},
 | 
						|
#endif
 | 
						|
#ifdef EXT3_JOURNAL_DATA_FL
 | 
						|
        { "nojournal",	L"nojournal",		EXT3_JOURNAL_DATA_FL,	0},
 | 
						|
#endif
 | 
						|
#ifdef EXT2_SECRM_FL
 | 
						|
        { "nosecuredeletion",L"nosecuredeletion",EXT2_SECRM_FL,		0},
 | 
						|
#endif
 | 
						|
#ifdef EXT2_SYNC_FL
 | 
						|
        { "nosync",	L"nosync",		EXT2_SYNC_FL,		0},
 | 
						|
#endif
 | 
						|
#ifdef EXT2_NOTAIL_FL
 | 
						|
        { "notail",	L"notail",		0,		EXT2_NOTAIL_FL},
 | 
						|
#endif
 | 
						|
#ifdef EXT2_TOPDIR_FL
 | 
						|
        { "notopdir",	L"notopdir",		EXT2_TOPDIR_FL,		0},
 | 
						|
#endif
 | 
						|
#ifdef EXT2_RESERVED_FL
 | 
						|
        { "noreserved",	L"noreserved",		EXT2_RESERVED_FL,	0},
 | 
						|
#endif
 | 
						|
 | 
						|
	{ NULL,		NULL,			0,		0 }
 | 
						|
};
 | 
						|
 | 
						|
/*
 | 
						|
 * fflagstostr --
 | 
						|
 *	Convert file flags to a comma-separated string.  If no flags
 | 
						|
 *	are set, return the empty string.
 | 
						|
 */
 | 
						|
static char *
 | 
						|
ae_fflagstostr(unsigned long bitset, unsigned long bitclear)
 | 
						|
{
 | 
						|
	char *string, *dp;
 | 
						|
	const char *sp;
 | 
						|
	unsigned long bits;
 | 
						|
	struct flag *flag;
 | 
						|
	size_t	length;
 | 
						|
 | 
						|
	bits = bitset | bitclear;
 | 
						|
	length = 0;
 | 
						|
	for (flag = flags; flag->name != NULL; flag++)
 | 
						|
		if (bits & (flag->set | flag->clear)) {
 | 
						|
			length += strlen(flag->name) + 1;
 | 
						|
			bits &= ~(flag->set | flag->clear);
 | 
						|
		}
 | 
						|
 | 
						|
	if (length == 0)
 | 
						|
		return (NULL);
 | 
						|
	string = (char *)malloc(length);
 | 
						|
	if (string == NULL)
 | 
						|
		return (NULL);
 | 
						|
 | 
						|
	dp = string;
 | 
						|
	for (flag = flags; flag->name != NULL; flag++) {
 | 
						|
		if (bitset & flag->set || bitclear & flag->clear) {
 | 
						|
			sp = flag->name + 2;
 | 
						|
		} else if (bitset & flag->clear  ||  bitclear & flag->set) {
 | 
						|
			sp = flag->name;
 | 
						|
		} else
 | 
						|
			continue;
 | 
						|
		bitset &= ~(flag->set | flag->clear);
 | 
						|
		bitclear &= ~(flag->set | flag->clear);
 | 
						|
		if (dp > string)
 | 
						|
			*dp++ = ',';
 | 
						|
		while ((*dp++ = *sp++) != '\0')
 | 
						|
			;
 | 
						|
		dp--;
 | 
						|
	}
 | 
						|
 | 
						|
	*dp = '\0';
 | 
						|
	return (string);
 | 
						|
}
 | 
						|
 | 
						|
/*
 | 
						|
 * strtofflags --
 | 
						|
 *	Take string of arguments and return file flags.  This
 | 
						|
 *	version works a little differently than strtofflags(3).
 | 
						|
 *	In particular, it always tests every token, skipping any
 | 
						|
 *	unrecognized tokens.  It returns a pointer to the first
 | 
						|
 *	unrecognized token, or NULL if every token was recognized.
 | 
						|
 *	This version is also const-correct and does not modify the
 | 
						|
 *	provided string.
 | 
						|
 */
 | 
						|
static const char *
 | 
						|
ae_strtofflags(const char *s, unsigned long *setp, unsigned long *clrp)
 | 
						|
{
 | 
						|
	const char *start, *end;
 | 
						|
	struct flag *flag;
 | 
						|
	unsigned long set, clear;
 | 
						|
	const char *failed;
 | 
						|
 | 
						|
	set = clear = 0;
 | 
						|
	start = s;
 | 
						|
	failed = NULL;
 | 
						|
	/* Find start of first token. */
 | 
						|
	while (*start == '\t'  ||  *start == ' '  ||  *start == ',')
 | 
						|
		start++;
 | 
						|
	while (*start != '\0') {
 | 
						|
		/* Locate end of token. */
 | 
						|
		end = start;
 | 
						|
		while (*end != '\0'  &&  *end != '\t'  &&
 | 
						|
		    *end != ' '  &&  *end != ',')
 | 
						|
			end++;
 | 
						|
		for (flag = flags; flag->name != NULL; flag++) {
 | 
						|
			if (memcmp(start, flag->name, end - start) == 0) {
 | 
						|
				/* Matched "noXXXX", so reverse the sense. */
 | 
						|
				clear |= flag->set;
 | 
						|
				set |= flag->clear;
 | 
						|
				break;
 | 
						|
			} else if (memcmp(start, flag->name + 2, end - start)
 | 
						|
			    == 0) {
 | 
						|
				/* Matched "XXXX", so don't reverse. */
 | 
						|
				set |= flag->set;
 | 
						|
				clear |= flag->clear;
 | 
						|
				break;
 | 
						|
			}
 | 
						|
		}
 | 
						|
		/* Ignore unknown flag names. */
 | 
						|
		if (flag->name == NULL  &&  failed == NULL)
 | 
						|
			failed = start;
 | 
						|
 | 
						|
		/* Find start of next token. */
 | 
						|
		start = end;
 | 
						|
		while (*start == '\t'  ||  *start == ' '  ||  *start == ',')
 | 
						|
			start++;
 | 
						|
 | 
						|
	}
 | 
						|
 | 
						|
	if (setp)
 | 
						|
		*setp = set;
 | 
						|
	if (clrp)
 | 
						|
		*clrp = clear;
 | 
						|
 | 
						|
	/* Return location of first failure. */
 | 
						|
	return (failed);
 | 
						|
}
 | 
						|
 | 
						|
/*
 | 
						|
 * wcstofflags --
 | 
						|
 *	Take string of arguments and return file flags.  This
 | 
						|
 *	version works a little differently than strtofflags(3).
 | 
						|
 *	In particular, it always tests every token, skipping any
 | 
						|
 *	unrecognized tokens.  It returns a pointer to the first
 | 
						|
 *	unrecognized token, or NULL if every token was recognized.
 | 
						|
 *	This version is also const-correct and does not modify the
 | 
						|
 *	provided string.
 | 
						|
 */
 | 
						|
static const wchar_t *
 | 
						|
ae_wcstofflags(const wchar_t *s, unsigned long *setp, unsigned long *clrp)
 | 
						|
{
 | 
						|
	const wchar_t *start, *end;
 | 
						|
	struct flag *flag;
 | 
						|
	unsigned long set, clear;
 | 
						|
	const wchar_t *failed;
 | 
						|
 | 
						|
	set = clear = 0;
 | 
						|
	start = s;
 | 
						|
	failed = NULL;
 | 
						|
	/* Find start of first token. */
 | 
						|
	while (*start == L'\t'  ||  *start == L' '  ||  *start == L',')
 | 
						|
		start++;
 | 
						|
	while (*start != L'\0') {
 | 
						|
		/* Locate end of token. */
 | 
						|
		end = start;
 | 
						|
		while (*end != L'\0'  &&  *end != L'\t'  &&
 | 
						|
		    *end != L' '  &&  *end != L',')
 | 
						|
			end++;
 | 
						|
		for (flag = flags; flag->wname != NULL; flag++) {
 | 
						|
			if (wmemcmp(start, flag->wname, end - start) == 0) {
 | 
						|
				/* Matched "noXXXX", so reverse the sense. */
 | 
						|
				clear |= flag->set;
 | 
						|
				set |= flag->clear;
 | 
						|
				break;
 | 
						|
			} else if (wmemcmp(start, flag->wname + 2, end - start)
 | 
						|
			    == 0) {
 | 
						|
				/* Matched "XXXX", so don't reverse. */
 | 
						|
				set |= flag->set;
 | 
						|
				clear |= flag->clear;
 | 
						|
				break;
 | 
						|
			}
 | 
						|
		}
 | 
						|
		/* Ignore unknown flag names. */
 | 
						|
		if (flag->wname == NULL  &&  failed == NULL)
 | 
						|
			failed = start;
 | 
						|
 | 
						|
		/* Find start of next token. */
 | 
						|
		start = end;
 | 
						|
		while (*start == L'\t'  ||  *start == L' '  ||  *start == L',')
 | 
						|
			start++;
 | 
						|
 | 
						|
	}
 | 
						|
 | 
						|
	if (setp)
 | 
						|
		*setp = set;
 | 
						|
	if (clrp)
 | 
						|
		*clrp = clear;
 | 
						|
 | 
						|
	/* Return location of first failure. */
 | 
						|
	return (failed);
 | 
						|
}
 | 
						|
 | 
						|
 | 
						|
#ifdef TEST
 | 
						|
#include <stdio.h>
 | 
						|
int
 | 
						|
main(int argc, char **argv)
 | 
						|
{
 | 
						|
	struct archive_entry *entry = archive_entry_new();
 | 
						|
	unsigned long set, clear;
 | 
						|
	const wchar_t *remainder;
 | 
						|
 | 
						|
	remainder = archive_entry_copy_fflags_text_w(entry, L"nosappnd dump archive,,,,,,,");
 | 
						|
	archive_entry_fflags(entry, &set, &clear);
 | 
						|
 | 
						|
	wprintf(L"set=0x%lX clear=0x%lX remainder='%ls'\n", set, clear, remainder);
 | 
						|
 | 
						|
	wprintf(L"new flags='%s'\n", archive_entry_fflags_text(entry));
 | 
						|
	return (0);
 | 
						|
}
 | 
						|
#endif
 |