8 files changed, 716 insertions, 414 deletions

diff --git a/lib/libarchive/archive_entry.c b/lib/libarchive/archive_entry.c
index 34c91162fc20..8da0215197c1 100644
--- a/lib/libarchive/archive_entry.c
+++ b/lib/libarchive/archive_entry.c
@@ -207,6 +207,8 @@ aes_get_mbs(struct aes *aes)
 static const wchar_t *
 aes_get_wcs(struct aes *aes)
 {
+	int r;
+
 	if (aes->aes_wcs == NULL && aes->aes_mbs == NULL)
 		return NULL;
 	if (aes->aes_wcs == NULL && aes->aes_mbs != NULL) {
@@ -221,8 +223,13 @@ aes_get_wcs(struct aes *aes)
 		aes->aes_wcs = aes->aes_wcs_alloc;
 		if (aes->aes_wcs == NULL)
 			__archive_errx(1, "No memory for aes_get_wcs()");
-		mbstowcs(aes->aes_wcs_alloc, aes->aes_mbs, wcs_length);
+		r = mbstowcs(aes->aes_wcs_alloc, aes->aes_mbs, wcs_length);
 		aes->aes_wcs_alloc[wcs_length] = 0;
+		if (r == -1) {
+			/* Conversion failed, don't lie to our clients. */
+			free(aes->aes_wcs_alloc);
+			aes->aes_wcs = aes->aes_wcs_alloc = NULL;
+		}
 	}
 	return (aes->aes_wcs);
 }
@@ -307,6 +314,8 @@ aes_copy_wcs_len(struct aes *aes, const wchar_t *wcs, size_t len)
 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);
@@ -752,6 +761,28 @@ archive_entry_set_link(struct archive_entry *entry, const char *target)
 		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_symlink.aes_mbs != NULL ||
+	    entry->ae_symlink.aes_wcs != NULL)
+		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_symlink.aes_mbs != NULL ||
+	    entry->ae_symlink.aes_wcs != NULL)
+		aes_copy_wcs(&entry->ae_symlink, target);
+	else
+		aes_copy_wcs(&entry->ae_hardlink, target);
+}
+
 void
 archive_entry_set_mode(struct archive_entry *entry, mode_t m)
 {
@@ -1124,6 +1155,11 @@ archive_entry_acl_next(struct archive_entry *entry, int want_type, int *type,
 		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;
@@ -1143,7 +1179,7 @@ const wchar_t *
 archive_entry_acl_text_w(struct archive_entry *entry, int flags)
 {
 	int count;
-	int length;
+	size_t length;
 	const wchar_t *wname;
 	const wchar_t *prefix;
 	wchar_t separator;
@@ -1505,7 +1541,7 @@ archive_entry_xattr_next(struct archive_entry * entry,
 		return (ARCHIVE_OK);
 	} else {
 		*name = NULL;
-		*name = NULL;
+		*value = NULL;
 		*size = (size_t)0;
 		return (ARCHIVE_WARN);
 	}
diff --git a/lib/libarchive/archive_entry.h b/lib/libarchive/archive_entry.h
index 173fbc30474e..8c61b9553b63 100644
--- a/lib/libarchive/archive_entry.h
+++ b/lib/libarchive/archive_entry.h
@@ -57,7 +57,25 @@ extern "C" {
 struct archive_entry;
 
 /*
- * File-type constants.  These are returned from archive_entry_filetype().
+ * File-type constants.  These are returned from archive_entry_filetype()
+ * and passed to archive_entry_set_filetype().
+ *
+ * These values match S_XXX defines on every platform I've checked,
+ * including Windows, AIX, Linux, Solaris, and BSD.  They're
+ * (re)defined here because platforms generally don't define the ones
+ * they don't support.  For example, Windows doesn't define S_IFLNK or
+ * S_IFBLK.  Instead of having a mass of conditional logic and system
+ * checks to define any S_XXX values that aren't supported locally,
+ * I've just defined a new set of such constants so that
+ * libarchive-based applications can manipulate and identify archive
+ * entries properly even if the hosting platform can't store them on
+ * disk.
+ *
+ * These values are also used directly within some portable formats,
+ * such as cpio.  If you find a platform that varies from these, the
+ * correct solution is to leave these alone and translate from these
+ * portable values to platform-native values when entries are read from
+ * or written to disk.
  */
 #define	AE_IFMT		0170000
 #define	AE_IFREG	0100000
@@ -91,7 +109,8 @@ dev_t			 archive_entry_devmajor(struct archive_entry *);
 dev_t			 archive_entry_devminor(struct archive_entry *);
 mode_t			 archive_entry_filetype(struct archive_entry *);
 void			 archive_entry_fflags(struct archive_entry *,
-			     unsigned long *set, unsigned long *clear);
+			    unsigned long * /* set */,
+			    unsigned long * /* clear */);
 const char		*archive_entry_fflags_text(struct archive_entry *);
 gid_t			 archive_entry_gid(struct archive_entry *);
 const char		*archive_entry_gname(struct archive_entry *);
@@ -130,7 +149,7 @@ void	archive_entry_set_devmajor(struct archive_entry *, dev_t);
 void	archive_entry_set_devminor(struct archive_entry *, dev_t);
 void	archive_entry_set_filetype(struct archive_entry *, unsigned int);
 void	archive_entry_set_fflags(struct archive_entry *,
-	    unsigned long set, unsigned long clear);
+	    unsigned long /* set */, unsigned long /* clear */);
 /* Returns pointer to start of first invalid token, or NULL if none. */
 /* Note that all recognized tokens are processed, regardless. */
 const wchar_t *archive_entry_copy_fflags_text_w(struct archive_entry *,
@@ -144,6 +163,8 @@ void	archive_entry_copy_hardlink(struct archive_entry *, const char *);
 void	archive_entry_copy_hardlink_w(struct archive_entry *, const wchar_t *);
 void	archive_entry_set_ino(struct archive_entry *, unsigned long);
 void	archive_entry_set_link(struct archive_entry *, const char *);
+void	archive_entry_copy_link(struct archive_entry *, const char *);
+void	archive_entry_copy_link_w(struct archive_entry *, const wchar_t *);
 void	archive_entry_set_mode(struct archive_entry *, mode_t);
 void	archive_entry_set_mtime(struct archive_entry *, time_t, long);
 void	archive_entry_set_nlink(struct archive_entry *, unsigned int);
@@ -182,6 +203,13 @@ void	archive_entry_copy_stat(struct archive_entry *, const struct stat *);
  *   = there are many different ACL text formats
  *   = would like to be able to read/convert archives containing ACLs
  *     on platforms that lack ACL libraries
+ *
+ *  This last point, in particular, forces me to implement a reasonably
+ *  complete set of ACL support routines.
+ *
+ *  TODO: Extend this to support NFSv4/NTFS permissions.  That should
+ *  allow full ACL support on Mac OS, in particular, which uses
+ *  POSIX.1e-style interfaces to manipulate NFSv4/NTFS permissions.
  */
 
 /*
@@ -216,21 +244,24 @@ void	archive_entry_copy_stat(struct archive_entry *, const struct stat *);
  */
 void	 archive_entry_acl_clear(struct archive_entry *);
 void	 archive_entry_acl_add_entry(struct archive_entry *,
-	     int type, int permset, int tag, int qual, const char *name);
+	    int /* type */, int /* permset */, int /* tag */,
+	    int /* qual */, const char * /* name */);
 void	 archive_entry_acl_add_entry_w(struct archive_entry *,
-	     int type, int permset, int tag, int qual, const wchar_t *name);
+	    int /* type */, int /* permset */, int /* tag */,
+	    int /* qual */, const wchar_t * /* name */);
 
 /*
  * To retrieve the ACL, first "reset", then repeatedly ask for the
  * "next" entry.  The want_type parameter allows you to request only
  * access entries or only default entries.
  */
-int	 archive_entry_acl_reset(struct archive_entry *, int want_type);
-int	 archive_entry_acl_next(struct archive_entry *, int want_type,
-	     int *type, int *permset, int *tag, int *qual, const char **name);
-int	 archive_entry_acl_next_w(struct archive_entry *, int want_type,
-	     int *type, int *permset, int *tag, int *qual,
-	     const wchar_t **name);
+int	 archive_entry_acl_reset(struct archive_entry *, int /* want_type */);
+int	 archive_entry_acl_next(struct archive_entry *, int /* want_type */,
+	    int * /* type */, int * /* permset */, int * /* tag */,
+	    int * /* qual */, const char ** /* name */);
+int	 archive_entry_acl_next_w(struct archive_entry *, int /* want_type */,
+	    int * /* type */, int * /* permset */, int * /* tag */,
+	    int * /* qual */, const wchar_t ** /* name */);
 
 /*
  * Construct a text-format ACL.  The flags argument is a bitmask that
@@ -245,10 +276,11 @@ int	 archive_entry_acl_next_w(struct archive_entry *, int want_type,
  */
 #define	ARCHIVE_ENTRY_ACL_STYLE_EXTRA_ID	1024
 #define	ARCHIVE_ENTRY_ACL_STYLE_MARK_DEFAULT	2048
-const wchar_t	*archive_entry_acl_text_w(struct archive_entry *, int flags);
+const wchar_t	*archive_entry_acl_text_w(struct archive_entry *,
+		    int /* flags */);
 
 /* Return a count of entries matching 'want_type' */
-int	 archive_entry_acl_count(struct archive_entry *, int want_type);
+int	 archive_entry_acl_count(struct archive_entry *, int /* want_type */);
 
 /*
  * Private ACL parser.  This is private because it handles some
@@ -259,9 +291,12 @@ int	 archive_entry_acl_count(struct archive_entry *, int want_type);
  * this interface are likely to be surprised when it changes.
  *
  * You were warned!
+ *
+ * TODO: Move this declaration out of the public header and into
+ * a private header.  Warnings above are silly.
  */
 int		 __archive_entry_acl_parse_w(struct archive_entry *,
-		     const wchar_t *, int type);
+		    const wchar_t *, int /* type */);
 
 /*
  * extended attributes
@@ -269,7 +304,8 @@ int		 __archive_entry_acl_parse_w(struct archive_entry *,
 
 void	 archive_entry_xattr_clear(struct archive_entry *);
 void	 archive_entry_xattr_add_entry(struct archive_entry *,
-	     const char *name, const void *value, size_t size);
+	    const char * /* name */, const void * /* value */,
+	    size_t /* size */);
 
 /*
  * To retrieve the xattr list, first "reset", then repeatedly ask for the
@@ -279,7 +315,7 @@ void	 archive_entry_xattr_add_entry(struct archive_entry *,
 int	archive_entry_xattr_count(struct archive_entry *);
 int	archive_entry_xattr_reset(struct archive_entry *);
 int	archive_entry_xattr_next(struct archive_entry *,
-	     const char **name, const void **value, size_t *);
+	    const char ** /* name */, const void ** /* value */, size_t *);
 
 /*
  * Utility to detect hardlinks.
diff --git a/lib/libarchive/archive_platform.h b/lib/libarchive/archive_platform.h
index 3f232803a6f1..035d27ccd30e 100644
--- a/lib/libarchive/archive_platform.h
+++ b/lib/libarchive/archive_platform.h
@@ -36,7 +36,10 @@
 #ifndef ARCHIVE_PLATFORM_H_INCLUDED
 #define	ARCHIVE_PLATFORM_H_INCLUDED
 
-#if defined(PLATFORM_CONFIG_H)
+#ifdef _WIN32
+#include "config_windows.h"
+#include "archive_windows.h"
+#elif defined(PLATFORM_CONFIG_H)
 /* Use hand-built config.h in environments that need it. */
 #include PLATFORM_CONFIG_H
 #elif defined(HAVE_CONFIG_H)
diff --git a/lib/libarchive/archive_read_support_format_mtree.c b/lib/libarchive/archive_read_support_format_mtree.c
index d7afb8c13c4a..fc806a51fc4b 100644
--- a/lib/libarchive/archive_read_support_format_mtree.c
+++ b/lib/libarchive/archive_read_support_format_mtree.c
@@ -76,12 +76,18 @@ struct mtree {
 	struct mtree_entry	*this_entry;
 	struct archive_string	 current_dir;
 	struct archive_string	 contents_name;
+
+	off_t			 cur_size, cur_offset;
 };
 
 static int	cleanup(struct archive_read *);
 static int	mtree_bid(struct archive_read *);
+static int	parse_file(struct archive_read *, struct archive_entry *,
+		    struct mtree *, struct mtree_entry *);
 static void	parse_escapes(char *, struct mtree_entry *);
-static int	parse_setting(struct archive_read *, struct mtree *,
+static int	parse_line(struct archive_read *, struct archive_entry *,
+		    struct mtree *, struct mtree_entry *);
+static int	parse_keyword(struct archive_read *, struct mtree *,
 		    struct archive_entry *, char *, char *);
 static int	read_data(struct archive_read *a,
 		    const void **buff, size_t *size, off_t *offset);
@@ -252,14 +258,16 @@ read_mtree(struct archive_read *a, struct mtree *mtree)
 	}
 }
 
+/*
+ * Read in the entire mtree file into memory on the first request.
+ * Then use the next unused file to satisfy each header request.
+ */
 static int
 read_header(struct archive_read *a, struct archive_entry *entry)
 {
-	struct stat st;
 	struct mtree *mtree;
-	struct mtree_entry *mentry, *mentry2;
-	char *p, *q;
-	int r = ARCHIVE_OK, r1;
+	char *p;
+	int r;
 
 	mtree = (struct mtree *)(a->format->data);
 
@@ -278,16 +286,10 @@ read_header(struct archive_read *a, struct archive_entry *entry)
 	a->archive.archive_format_name = mtree->archive_format_name;
 
 	for (;;) {
-		mentry = mtree->this_entry;
-		if (mentry == NULL) {
-			mtree->this_entry = NULL;
+		if (mtree->this_entry == NULL)
 			return (ARCHIVE_EOF);
-		}
-		mtree->this_entry = mentry->next;
-		if (mentry->used)
-			continue;
-		mentry->used = 1;
-		if (strcmp(mentry->name, "..") == 0) {
+		if (strcmp(mtree->this_entry->name, "..") == 0) {
+			mtree->this_entry->used = 1;
 			if (archive_strlen(&mtree->current_dir) > 0) {
 				/* Roll back current path. */
 				p = mtree->current_dir.s
@@ -299,117 +301,165 @@ read_header(struct archive_read *a, struct archive_entry *entry)
 				mtree->current_dir.length
 				    = p - mtree->current_dir.s + 1;
 			}
-			continue;
 		}
+		if (!mtree->this_entry->used) {
+			r = parse_file(a, entry, mtree, mtree->this_entry);
+			return (r);
+		}
+		mtree->this_entry = mtree->this_entry->next;
+	}
+}
+
+/*
+ * A single file can have multiple lines contribute specifications.
+ * Parse as many lines as necessary, then pull additional information
+ * from a backing file on disk as necessary.
+ */
+static int
+parse_file(struct archive_read *a, struct archive_entry *entry,
+    struct mtree *mtree, struct mtree_entry *mentry)
+{
+	struct stat st;
+	struct mtree_entry *mp;
+	int r = ARCHIVE_OK, r1;
 
-		mtree->filetype = AE_IFREG;
+	mentry->used = 1;
 
-		/* Parse options. */
-		p = mentry->option_start;
-		while (p < mentry->option_end) {
-			q = p + strlen(p);
-			r1 = parse_setting(a, mtree, entry, p, q);
-			if (r1 != ARCHIVE_OK)
-				r = r1;
-			p = q + 1;
-		}
+	/* Initialize reasonable defaults. */
+	mtree->filetype = AE_IFREG;
+	archive_entry_set_size(entry, 0);
 
-		if (mentry->full) {
-			archive_entry_copy_pathname(entry, mentry->name);
-			/*
-			 * "Full" entries are allowed to have multiple
-			 * lines and those lines aren't required to be
-			 * adjacent.  We don't support multiple lines
-			 * for "relative" entries nor do we make any
-			 * attempt to merge data from separate
-			 * "relative" and "full" entries.  (Merging
-			 * "relative" and "full" entries would require
-			 * dealing with pathname canonicalization,
-			 * which is a very tricky subject.)
-			 */
-			mentry2 = mentry->next;
-			while (mentry2 != NULL) {
-				if (mentry2->full
-				    && !mentry2->used
-				    && strcmp(mentry->name, mentry2->name) == 0) {
-					/*
-					 * Add those options as well;
-					 * later lines override
-					 * earlier ones.
-					 */
-					p = mentry2->option_start;
-					while (p < mentry2->option_end) {
-						q = p + strlen(p);
-						r1 = parse_setting(a, mtree, entry, p, q);
-						if (r1 != ARCHIVE_OK)
-							r = r1;
-						p = q + 1;
-					}
-					mentry2->used = 1;
-				}
-				mentry2 = mentry2->next;
-			}
-		} else {
-			/*
-			 * Relative entries require us to construct
-			 * the full path and possibly update the
-			 * current directory.
-			 */
-			size_t n = archive_strlen(&mtree->current_dir);
-			if (n > 0)
-				archive_strcat(&mtree->current_dir, "/");
-			archive_strcat(&mtree->current_dir, mentry->name);
-			archive_entry_copy_pathname(entry, mtree->current_dir.s);
-			if (archive_entry_filetype(entry) != AE_IFDIR)
-				mtree->current_dir.length = n;
-		}
+	/* Parse options from this line. */
+	r = parse_line(a, entry, mtree, mentry);
 
+	if (mentry->full) {
+		archive_entry_copy_pathname(entry, mentry->name);
 		/*
-		 * Try to open and stat the file to get the real size.
-		 * It would be nice to avoid this here so that getting
-		 * a listing of an mtree wouldn't require opening
-		 * every referenced contents file.  But then we
-		 * wouldn't know the actual contents size, so I don't
-		 * see a really viable way around this.  (Also, we may
-		 * want to someday pull other unspecified info from
-		 * the contents file on disk.)
+		 * "Full" entries are allowed to have multiple lines
+		 * and those lines aren't required to be adjacent.  We
+		 * don't support multiple lines for "relative" entries
+		 * nor do we make any attempt to merge data from
+		 * separate "relative" and "full" entries.  (Merging
+		 * "relative" and "full" entries would require dealing
+		 * with pathname canonicalization, which is a very
+		 * tricky subject.)
 		 */
-		if (archive_strlen(&mtree->contents_name) > 0) {
-			mtree->fd = open(mtree->contents_name.s,
-			    O_RDONLY | O_BINARY);
-			if (mtree->fd < 0) {
-				archive_set_error(&a->archive, errno,
-				    "Can't open content=\"%s\"",
-				    mtree->contents_name.s);
-				r = ARCHIVE_WARN;
+		for (mp = mentry->next; mp != NULL; mp = mp->next) {
+			if (mp->full && !mp->used
+			    && strcmp(mentry->name, mp->name) == 0) {
+				/* Later lines override earlier ones. */
+				mp->used = 1;
+				r1 = parse_line(a, entry, mtree, mp);
+				if (r1 < r)
+					r = r1;
 			}
-		} else {
-			/* If the specified path opens, use it. */
-			mtree->fd = open(mtree->current_dir.s,
-			    O_RDONLY | O_BINARY);
-			/* But don't fail if it's not there. */
 		}
-
+	} else {
 		/*
-		 * If there is a contents file on disk, use that size;
-		 * otherwise leave it as-is (it might have been set from
-		 * the mtree size= keyword).
+		 * Relative entries require us to construct
+		 * the full path and possibly update the
+		 * current directory.
 		 */
-		if (mtree->fd >= 0) {
-			fstat(mtree->fd, &st);
+		size_t n = archive_strlen(&mtree->current_dir);
+		if (n > 0)
+			archive_strcat(&mtree->current_dir, "/");
+		archive_strcat(&mtree->current_dir, mentry->name);
+		archive_entry_copy_pathname(entry, mtree->current_dir.s);
+		if (archive_entry_filetype(entry) != AE_IFDIR)
+			mtree->current_dir.length = n;
+	}
+
+	/*
+	 * Try to open and stat the file to get the real size
+	 * and other file info.  It would be nice to avoid
+	 * this here so that getting a listing of an mtree
+	 * wouldn't require opening every referenced contents
+	 * file.  But then we wouldn't know the actual
+	 * contents size, so I don't see a really viable way
+	 * around this.  (Also, we may want to someday pull
+	 * other unspecified info from the contents file on
+	 * disk.)
+	 */
+	mtree->fd = -1;
+	if (archive_strlen(&mtree->contents_name) > 0) {
+		mtree->fd = open(mtree->contents_name.s,
+		    O_RDONLY | O_BINARY);
+		if (mtree->fd < 0) {
+			archive_set_error(&a->archive, errno,
+			    "Can't open content=\"%s\"",
+			    mtree->contents_name.s);
+			r = ARCHIVE_WARN;
+		}
+	} else if (archive_entry_filetype(entry) == AE_IFREG) {
+		mtree->fd = open(archive_entry_pathname(entry),
+		    O_RDONLY | O_BINARY);
+	}
+
+	/*
+	 * If there is a contents file on disk, use that size;
+	 * otherwise leave it as-is (it might have been set from
+	 * the mtree size= keyword).
+	 */
+	if (mtree->fd >= 0) {
+		if (fstat(mtree->fd, &st) != 0) {
+			archive_set_error(&a->archive, errno,
+			    "could not stat %s",
+			    archive_entry_pathname(entry));
+			r = ARCHIVE_WARN;
+			/* If we can't stat it, don't keep it open. */
+			close(mtree->fd);
+			mtree->fd = -1;
+		} else if ((st.st_mode & S_IFMT) != S_IFREG) {
+			archive_set_error(&a->archive, errno,
+			    "%s is not a regular file",
+			    archive_entry_pathname(entry));
+			r = ARCHIVE_WARN;
+			/* Don't hold a non-regular file open. */
+			close(mtree->fd);
+			mtree->fd = -1;
+		} else {
 			archive_entry_set_size(entry, st.st_size);
+			archive_entry_set_ino(entry, st.st_ino);
+			archive_entry_set_dev(entry, st.st_dev);
+			archive_entry_set_nlink(entry, st.st_nlink);
 		}
+	}
+	mtree->cur_size = archive_entry_size(entry);
+	mtree->offset = 0;
 
-		return r;
+	return r;
+}
+
+/*
+ * Each line contains a sequence of keywords.
+ */
+static int
+parse_line(struct archive_read *a, struct archive_entry *entry,
+    struct mtree *mtree, struct mtree_entry *mp)
+{
+	char *p, *q;
+	int r = ARCHIVE_OK, r1;
+
+	p = mp->option_start;
+	while (p < mp->option_end) {
+		q = p + strlen(p);
+		r1 = parse_keyword(a, mtree, entry, p, q);
+		if (r1 < r)
+			r = r1;
+		p = q + 1;
 	}
+	return (r);
 }
 
+/*
+ * Parse a single keyword and its value.
+ */
 static int
-parse_setting(struct archive_read *a, struct mtree *mtree, struct archive_entry *entry, char *key, char *end)
+parse_keyword(struct archive_read *a, struct mtree *mtree,
+    struct archive_entry *entry, char *key, char *end)
 {
 	char *val;
 
-
 	if (end == key)
 		return (ARCHIVE_OK);
 	if (*key == '\0')
@@ -427,7 +477,8 @@ parse_setting(struct archive_read *a, struct mtree *mtree, struct archive_entry
 
 	switch (key[0]) {
 	case 'c':
-		if (strcmp(key, "content") == 0) {
+		if (strcmp(key, "content") == 0
+		    || strcmp(key, "contents") == 0) {
 			parse_escapes(val, NULL);
 			archive_strcpy(&mtree->contents_name, val);
 			break;
@@ -441,6 +492,11 @@ parse_setting(struct archive_read *a, struct mtree *mtree, struct archive_entry
 			archive_entry_copy_gname(entry, val);
 			break;
 		}
+	case 'l':
+		if (strcmp(key, "link") == 0) {
+			archive_entry_set_link(entry, val);
+			break;
+		}
 	case 'm':
 		if (strcmp(key, "mode") == 0) {
 			if (val[0] == '0') {
@@ -452,6 +508,11 @@ parse_setting(struct archive_read *a, struct mtree *mtree, struct archive_entry
 				    "Symbolic mode \"%s\" unsupported", val);
 			break;
 		}
+	case 's':
+		if (strcmp(key, "size") == 0) {
+			archive_entry_set_size(entry, mtree_atol10(&val));
+			break;
+		}
 	case 't':
 		if (strcmp(key, "type") == 0) {
 			switch (val[0]) {
@@ -517,6 +578,7 @@ parse_setting(struct archive_read *a, struct mtree *mtree, struct archive_entry
 static int
 read_data(struct archive_read *a, const void **buff, size_t *size, off_t *offset)
 {
+	size_t bytes_to_read;
 	ssize_t bytes_read;
 	struct mtree *mtree;
 
@@ -538,7 +600,11 @@ read_data(struct archive_read *a, const void **buff, size_t *size, off_t *offset
 
 	*buff = mtree->buff;
 	*offset = mtree->offset;
-	bytes_read = read(mtree->fd, mtree->buff, mtree->buffsize);
+	if ((off_t)mtree->buffsize > mtree->cur_size - mtree->offset)
+		bytes_to_read = mtree->cur_size - mtree->offset;
+	else
+		bytes_to_read = mtree->buffsize;
+	bytes_read = read(mtree->fd, mtree->buff, bytes_to_read);
 	if (bytes_read < 0) {
 		archive_set_error(&a->archive, errno, "Can't read");
 		return (ARCHIVE_WARN);
@@ -548,7 +614,7 @@ read_data(struct archive_read *a, const void **buff, size_t *size, off_t *offset
 		return (ARCHIVE_EOF);
 	}
 	mtree->offset += bytes_read;
-	*size = (size_t)bytes_read;
+	*size = bytes_read;
 	return (ARCHIVE_OK);
 }
 
diff --git a/lib/libarchive/archive_read_support_format_tar.c b/lib/libarchive/archive_read_support_format_tar.c
index 9cfe0e762aa4..a0c434244830 100644
--- a/lib/libarchive/archive_read_support_format_tar.c
+++ b/lib/libarchive/archive_read_support_format_tar.c
@@ -144,8 +144,8 @@ struct sparse_block {
 
 struct tar {
 	struct archive_string	 acl_text;
-	struct archive_string	 entry_name;
-	struct archive_string	 entry_linkname;
+	struct archive_string	 entry_pathname;
+	struct archive_string	 entry_linkpath;
 	struct archive_string	 entry_uname;
 	struct archive_string	 entry_gname;
 	struct archive_string	 longlink;
@@ -153,6 +153,7 @@ struct tar {
 	struct archive_string	 pax_header;
 	struct archive_string	 pax_global;
 	struct archive_string	 line;
+	int			 pax_hdrcharset_binary;
 	wchar_t 		*pax_entry;
 	size_t			 pax_entry_length;
 	int			 header_recursion_depth;
@@ -169,9 +170,9 @@ struct tar {
 	char			 sparse_gnu_pending;
 };
 
-static size_t	UTF8_mbrtowc(wchar_t *pwc, const char *s, size_t n);
+static ssize_t	UTF8_mbrtowc(wchar_t *pwc, const char *s, size_t n);
 static int	archive_block_is_null(const unsigned char *p);
-static char	*base64_decode(const wchar_t *, size_t, size_t *);
+static char	*base64_decode(const char *, size_t, size_t *);
 static void	 gnu_add_sparse_entry(struct tar *,
 		    off_t offset, off_t remaining);
 static void	gnu_clear_sparse_list(struct tar *);
@@ -179,7 +180,7 @@ static int	gnu_sparse_old_read(struct archive_read *, struct tar *,
 		    const struct archive_entry_header_gnutar *header);
 static void	gnu_sparse_old_parse(struct tar *,
 		    const struct gnu_sparse *sparse, int length);
-static int	gnu_sparse_01_parse(struct tar *, const wchar_t *);
+static int	gnu_sparse_01_parse(struct tar *, const char *);
 static ssize_t	gnu_sparse_10_read(struct archive_read *, struct tar *);
 static int	header_Solaris_ACL(struct archive_read *,  struct tar *,
 		    struct archive_entry *, const void *);
@@ -210,24 +211,23 @@ static int	archive_read_format_tar_read_header(struct archive_read *,
 		    struct archive_entry *);
 static int	checksum(struct archive_read *, const void *);
 static int 	pax_attribute(struct tar *, struct archive_entry *,
-		    wchar_t *key, wchar_t *value);
+		    char *key, char *value);
 static int 	pax_header(struct archive_read *, struct tar *,
 		    struct archive_entry *, char *attr);
-static void	pax_time(const wchar_t *, int64_t *sec, long *nanos);
+static void	pax_time(const char *, int64_t *sec, long *nanos);
 static ssize_t	readline(struct archive_read *, struct tar *, const char **,
 		    ssize_t limit);
 static int	read_body_to_string(struct archive_read *, struct tar *,
 		    struct archive_string *, const void *h);
 static int64_t	tar_atol(const char *, unsigned);
-static int64_t	tar_atol10(const wchar_t *, unsigned);
+static int64_t	tar_atol10(const char *, unsigned);
 static int64_t	tar_atol256(const char *, unsigned);
 static int64_t	tar_atol8(const char *, unsigned);
 static int	tar_read_header(struct archive_read *, struct tar *,
 		    struct archive_entry *);
 static int	tohex(int c);
 static char	*url_decode(const char *);
-static int	utf8_decode(wchar_t *, const char *, size_t length);
-static char	*wide_to_narrow(const wchar_t *wval);
+static wchar_t	*utf8_decode(struct tar *, const char *, size_t length);
 
 int
 archive_read_support_format_gnutar(struct archive *a)
@@ -271,8 +271,8 @@ archive_read_format_tar_cleanup(struct archive_read *a)
 	tar = (struct tar *)(a->format->data);
 	gnu_clear_sparse_list(tar);
 	archive_string_free(&tar->acl_text);
-	archive_string_free(&tar->entry_name);
-	archive_string_free(&tar->entry_linkname);
+	archive_string_free(&tar->entry_pathname);
+	archive_string_free(&tar->entry_linkpath);
 	archive_string_free(&tar->entry_uname);
 	archive_string_free(&tar->entry_gname);
 	archive_string_free(&tar->line);
@@ -714,7 +714,7 @@ archive_block_is_null(const unsigned char *p)
 {
 	unsigned i;
 
-	for (i = 0; i < ARCHIVE_BYTES_PER_RECORD / sizeof(*p); i++)
+	for (i = 0; i < 512; i++)
 		if (*p++)
 			return (0);
 	return (1);
@@ -766,16 +766,9 @@ header_Solaris_ACL(struct archive_read *a, struct tar *tar,
 	while (*p != '\0' && p < acl + size)
 		p++;
 
-	wp = (wchar_t *)malloc((p - acl + 1) * sizeof(wchar_t));
-	if (wp == NULL) {
-		archive_set_error(&a->archive, ENOMEM,
-		    "Can't allocate work buffer for ACL parsing");
-		return (ARCHIVE_FATAL);
-	}
-	utf8_decode(wp, acl, p - acl);
+	wp = utf8_decode(tar, acl, p - acl);
 	err = __archive_entry_acl_parse_w(entry, wp,
 	    ARCHIVE_ENTRY_ACL_TYPE_ACCESS);
-	free(wp);
 	return (err);
 }
 
@@ -795,7 +788,7 @@ header_longlink(struct archive_read *a, struct tar *tar,
 	if ((err != ARCHIVE_OK) && (err != ARCHIVE_WARN))
 		return (err);
 	/* Set symlink if symlink already set, else hardlink. */
-	archive_entry_set_link(entry, tar->longlink.s);
+	archive_entry_copy_link(entry, tar->longlink.s);
 	return (ARCHIVE_OK);
 }
 
@@ -815,7 +808,7 @@ header_longname(struct archive_read *a, struct tar *tar,
 	err = tar_read_header(a, tar, entry);
 	if ((err != ARCHIVE_OK) && (err != ARCHIVE_WARN))
 		return (err);
-	archive_entry_set_pathname(entry, tar->longname.s);
+	archive_entry_copy_pathname(entry, tar->longname.s);
 	return (ARCHIVE_OK);
 }
 
@@ -907,10 +900,10 @@ header_common(struct archive_read *a, struct tar *tar,
 
 	header = (const struct archive_entry_header_ustar *)h;
 	if (header->linkname[0])
-		archive_strncpy(&(tar->entry_linkname), header->linkname,
+		archive_strncpy(&(tar->entry_linkpath), header->linkname,
 		    sizeof(header->linkname));
 	else
-		archive_string_empty(&(tar->entry_linkname));
+		archive_string_empty(&(tar->entry_linkpath));
 
 	/* Parse out the numeric fields (all are octal) */
 	archive_entry_set_mode(entry, tar_atol(header->mode, sizeof(header->mode)));
@@ -926,7 +919,7 @@ header_common(struct archive_read *a, struct tar *tar,
 
 	switch (tartype) {
 	case '1': /* Hard link */
-		archive_entry_set_hardlink(entry, tar->entry_linkname.s);
+		archive_entry_copy_hardlink(entry, tar->entry_linkpath.s);
 		/*
 		 * The following may seem odd, but: Technically, tar
 		 * does not store the file type for a "hard link"
@@ -988,7 +981,7 @@ header_common(struct archive_read *a, struct tar *tar,
 		archive_entry_set_filetype(entry, AE_IFLNK);
 		archive_entry_set_size(entry, 0);
 		tar->entry_bytes_remaining = 0;
-		archive_entry_set_symlink(entry, tar->entry_linkname.s);
+		archive_entry_copy_symlink(entry, tar->entry_linkpath.s);
 		break;
 	case '3': /* Character device */
 		archive_entry_set_filetype(entry, AE_IFCHR);
@@ -1060,8 +1053,8 @@ header_old_tar(struct archive_read *a, struct tar *tar,
 
 	/* Copy filename over (to ensure null termination). */
 	header = (const struct archive_entry_header_ustar *)h;
-	archive_strncpy(&(tar->entry_name), header->name, sizeof(header->name));
-	archive_entry_set_pathname(entry, tar->entry_name.s);
+	archive_strncpy(&(tar->entry_pathname), header->name, sizeof(header->name));
+	archive_entry_copy_pathname(entry, tar->entry_pathname.s);
 
 	/* Grab rest of common fields */
 	header_common(a, tar, entry, h);
@@ -1132,7 +1125,7 @@ header_ustar(struct archive_read *a, struct tar *tar,
 	header = (const struct archive_entry_header_ustar *)h;
 
 	/* Copy name into an internal buffer to ensure null-termination. */
-	as = &(tar->entry_name);
+	as = &(tar->entry_pathname);
 	if (header->prefix[0]) {
 		archive_strncpy(as, header->prefix, sizeof(header->prefix));
 		if (as->s[archive_strlen(as) - 1] != '/')
@@ -1141,7 +1134,7 @@ header_ustar(struct archive_read *a, struct tar *tar,
 	} else
 		archive_strncpy(as, header->name, sizeof(header->name));
 
-	archive_entry_set_pathname(entry, as->s);
+	archive_entry_copy_pathname(entry, as->s);
 
 	/* Handle rest of common fields. */
 	header_common(a, tar, entry, h);
@@ -1149,11 +1142,11 @@ header_ustar(struct archive_read *a, struct tar *tar,
 	/* Handle POSIX ustar fields. */
 	archive_strncpy(&(tar->entry_uname), header->uname,
 	    sizeof(header->uname));
-	archive_entry_set_uname(entry, tar->entry_uname.s);
+	archive_entry_copy_uname(entry, tar->entry_uname.s);
 
 	archive_strncpy(&(tar->entry_gname), header->gname,
 	    sizeof(header->gname));
-	archive_entry_set_gname(entry, tar->entry_gname.s);
+	archive_entry_copy_gname(entry, tar->entry_gname.s);
 
 	/* Parse out device numbers only for char and block specials. */
 	if (header->typeflag[0] == '3' || header->typeflag[0] == '4') {
@@ -1180,10 +1173,16 @@ pax_header(struct archive_read *a, struct tar *tar,
 {
 	size_t attr_length, l, line_length;
 	char *line, *p;
-	wchar_t *key, *wp, *value;
+	char *key, *value;
+	wchar_t *wp;
 	int err, err2;
 
 	attr_length = strlen(attr);
+	tar->pax_hdrcharset_binary = 0;
+	archive_string_empty(&(tar->entry_gname));
+	archive_string_empty(&(tar->entry_linkpath));
+	archive_string_empty(&(tar->entry_pathname));
+	archive_string_empty(&(tar->entry_uname));
 	err = ARCHIVE_OK;
 	while (attr_length > 0) {
 		/* Parse decimal length field at start of line. */
@@ -1226,49 +1225,24 @@ pax_header(struct archive_read *a, struct tar *tar,
 			return (ARCHIVE_WARN);
 		}
 
-		/* Ensure pax_entry buffer is big enough. */
-		if (tar->pax_entry_length <= line_length) {
-			wchar_t *old_entry = tar->pax_entry;
-
-			if (tar->pax_entry_length <= 0)
-				tar->pax_entry_length = 1024;
-			while (tar->pax_entry_length <= line_length + 1)
-				tar->pax_entry_length *= 2;
-
-			old_entry = tar->pax_entry;
-			tar->pax_entry = (wchar_t *)realloc(tar->pax_entry,
-			    tar->pax_entry_length * sizeof(wchar_t));
-			if (tar->pax_entry == NULL) {
-				free(old_entry);
-				archive_set_error(&a->archive, ENOMEM,
-					"No memory");
-				return (ARCHIVE_FATAL);
-			}
-		}
-
-		/* Decode UTF-8 to wchar_t, null-terminate result. */
-		if (utf8_decode(tar->pax_entry, p,
-			line_length - (p - attr) - 1)) {
-			archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
-			   "Invalid UTF8 character in pax extended attribute");
-			err = err_combine(err, ARCHIVE_WARN);
-		}
+		/* Null-terminate the line. */
+		attr[line_length - 1] = '\0';
 
-		/* Null-terminate 'key' value. */
-		wp = key = tar->pax_entry;
-		if (key[0] == L'=')
+		/* Find end of key and null terminate it. */
+		key = p;
+		if (key[0] == '=')
 			return (-1);
-		while (*wp && *wp != L'=')
-			++wp;
-		if (*wp == L'\0') {
+		while (*p && *p != '=')
+			++p;
+		if (*p == '\0') {
 			archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
 			    "Invalid pax extended attributes");
 			return (ARCHIVE_WARN);
 		}
-		*wp = 0;
+		*p = '\0';
 
 		/* Identify null-terminated 'value' portion. */
-		value = wp + 1;
+		value = p + 1;
 
 		/* Identify this attribute and set it in the entry. */
 		err2 = pax_attribute(tar, entry, key, value);
@@ -1278,33 +1252,85 @@ pax_header(struct archive_read *a, struct tar *tar,
 		attr += line_length;
 		attr_length -= line_length;
 	}
+	if (archive_strlen(&(tar->entry_gname)) > 0) {
+		value = tar->entry_gname.s;
+		if (tar->pax_hdrcharset_binary)
+			archive_entry_copy_gname(entry, value);
+		else {
+			wp = utf8_decode(tar, value, strlen(value));
+			if (wp == NULL) {
+				archive_entry_copy_gname(entry, value);
+				if (err > ARCHIVE_WARN)
+					err = ARCHIVE_WARN;
+			} else
+				archive_entry_copy_gname_w(entry, wp);
+		}
+	}
+	if (archive_strlen(&(tar->entry_linkpath)) > 0) {
+		value = tar->entry_linkpath.s;
+		if (tar->pax_hdrcharset_binary)
+			archive_entry_copy_link(entry, value);
+		else {
+			wp = utf8_decode(tar, value, strlen(value));
+			if (wp == NULL) {
+				archive_entry_copy_link(entry, value);
+				if (err > ARCHIVE_WARN)
+					err = ARCHIVE_WARN;
+			} else
+				archive_entry_copy_link_w(entry, wp);
+		}
+	}
+	if (archive_strlen(&(tar->entry_pathname)) > 0) {
+		value = tar->entry_pathname.s;
+		if (tar->pax_hdrcharset_binary)
+			archive_entry_copy_pathname(entry, value);
+		else {
+			wp = utf8_decode(tar, value, strlen(value));
+			if (wp == NULL) {
+				archive_entry_copy_pathname(entry, value);
+				if (err > ARCHIVE_WARN)
+					err = ARCHIVE_WARN;
+			} else
+				archive_entry_copy_pathname_w(entry, wp);
+		}
+	}
+	if (archive_strlen(&(tar->entry_uname)) > 0) {
+		value = tar->entry_uname.s;
+		if (tar->pax_hdrcharset_binary)
+			archive_entry_copy_uname(entry, value);
+		else {
+			wp = utf8_decode(tar, value, strlen(value));
+			if (wp == NULL) {
+				archive_entry_copy_uname(entry, value);
+				if (err > ARCHIVE_WARN)
+					err = ARCHIVE_WARN;
+			} else
+				archive_entry_copy_uname_w(entry, wp);
+		}
+	}
 	return (err);
 }
 
 static int
 pax_attribute_xattr(struct archive_entry *entry,
-	wchar_t *name, wchar_t *value)
+	char *name, char *value)
 {
-	char *name_decoded, *name_narrow;
+	char *name_decoded;
 	void *value_decoded;
 	size_t value_len;
 
-	if (wcslen(name) < 18 || (wcsncmp(name, L"LIBARCHIVE.xattr.", 17)) != 0)
+	if (strlen(name) < 18 || (strncmp(name, "LIBARCHIVE.xattr.", 17)) != 0)
 		return 3;
 
 	name += 17;
 
 	/* URL-decode name */
-	name_narrow = wide_to_narrow(name);
-	if (name_narrow == NULL)
-		return 2;
-	name_decoded = url_decode(name_narrow);
-	free(name_narrow);
+	name_decoded = url_decode(name);
 	if (name_decoded == NULL)
 		return 2;
 
 	/* Base-64 decode value */
-	value_decoded = base64_decode(value, wcslen(value), &value_len);
+	value_decoded = base64_decode(value, strlen(value), &value_len);
 	if (value_decoded == NULL) {
 		free(name_decoded);
 		return 1;
@@ -1333,22 +1359,23 @@ pax_attribute_xattr(struct archive_entry *entry,
  */
 static int
 pax_attribute(struct tar *tar, struct archive_entry *entry,
-    wchar_t *key, wchar_t *value)
+    char *key, char *value)
 {
 	int64_t s;
 	long n;
+	wchar_t *wp;
 
 	switch (key[0]) {
 	case 'G':
 		/* GNU "0.0" sparse pax format. */
-		if (wcscmp(key, L"GNU.sparse.numblocks") == 0) {
+		if (strcmp(key, "GNU.sparse.numblocks") == 0) {
 			tar->sparse_offset = -1;
 			tar->sparse_numbytes = -1;
 			tar->sparse_gnu_major = 0;
 			tar->sparse_gnu_minor = 0;
 		}
-		if (wcscmp(key, L"GNU.sparse.offset") == 0) {
-			tar->sparse_offset = tar_atol10(value, wcslen(value));
+		if (strcmp(key, "GNU.sparse.offset") == 0) {
+			tar->sparse_offset = tar_atol10(value, strlen(value));
 			if (tar->sparse_numbytes != -1) {
 				gnu_add_sparse_entry(tar,
 				    tar->sparse_offset, tar->sparse_numbytes);
@@ -1356,8 +1383,8 @@ pax_attribute(struct tar *tar, struct archive_entry *entry,
 				tar->sparse_numbytes = -1;
 			}
 		}
-		if (wcscmp(key, L"GNU.sparse.numbytes") == 0) {
-			tar->sparse_numbytes = tar_atol10(value, wcslen(value));
+		if (strcmp(key, "GNU.sparse.numbytes") == 0) {
+			tar->sparse_numbytes = tar_atol10(value, strlen(value));
 			if (tar->sparse_numbytes != -1) {
 				gnu_add_sparse_entry(tar,
 				    tar->sparse_offset, tar->sparse_numbytes);
@@ -1365,13 +1392,13 @@ pax_attribute(struct tar *tar, struct archive_entry *entry,
 				tar->sparse_numbytes = -1;
 			}
 		}
-		if (wcscmp(key, L"GNU.sparse.size") == 0) {
-			tar->realsize = tar_atol10(value, wcslen(value));
+		if (strcmp(key, "GNU.sparse.size") == 0) {
+			tar->realsize = tar_atol10(value, strlen(value));
 			archive_entry_set_size(entry, tar->realsize);
 		}
 
 		/* GNU "0.1" sparse pax format. */
-		if (wcscmp(key, L"GNU.sparse.map") == 0) {
+		if (strcmp(key, "GNU.sparse.map") == 0) {
 			tar->sparse_gnu_major = 0;
 			tar->sparse_gnu_minor = 1;
 			if (gnu_sparse_01_parse(tar, value) != ARCHIVE_OK)
@@ -1379,18 +1406,23 @@ pax_attribute(struct tar *tar, struct archive_entry *entry,
 		}
 
 		/* GNU "1.0" sparse pax format */
-		if (wcscmp(key, L"GNU.sparse.major") == 0) {
-			tar->sparse_gnu_major = tar_atol10(value, wcslen(value));
+		if (strcmp(key, "GNU.sparse.major") == 0) {
+			tar->sparse_gnu_major = tar_atol10(value, strlen(value));
 			tar->sparse_gnu_pending = 1;
 		}
-		if (wcscmp(key, L"GNU.sparse.minor") == 0) {
-			tar->sparse_gnu_minor = tar_atol10(value, wcslen(value));
+		if (strcmp(key, "GNU.sparse.minor") == 0) {
+			tar->sparse_gnu_minor = tar_atol10(value, strlen(value));
 			tar->sparse_gnu_pending = 1;
 		}
-		if (wcscmp(key, L"GNU.sparse.name") == 0)
-			archive_entry_copy_pathname_w(entry, value);
-		if (wcscmp(key, L"GNU.sparse.realsize") == 0) {
-			tar->realsize = tar_atol10(value, wcslen(value));
+		if (strcmp(key, "GNU.sparse.name") == 0) {
+			wp = utf8_decode(tar, value, strlen(value));
+			if (wp != NULL)
+				archive_entry_copy_pathname_w(entry, wp);
+			else
+				archive_entry_copy_pathname(entry, value);
+		}
+		if (strcmp(key, "GNU.sparse.realsize") == 0) {
+			tar->realsize = tar_atol10(value, strlen(value));
 			archive_entry_set_size(entry, tar->realsize);
 		}
 		break;
@@ -1401,85 +1433,107 @@ pax_attribute(struct tar *tar, struct archive_entry *entry,
 		if (strcmp(key, "LIBARCHIVE.xxxxxxx")==0)
 			archive_entry_set_xxxxxx(entry, value);
 */
-		if (wcsncmp(key, L"LIBARCHIVE.xattr.", 17)==0)
+		if (strncmp(key, "LIBARCHIVE.xattr.", 17)==0)
 			pax_attribute_xattr(entry, key, value);
 		break;
 	case 'S':
 		/* We support some keys used by the "star" archiver */
-		if (wcscmp(key, L"SCHILY.acl.access")==0)
-			__archive_entry_acl_parse_w(entry, value,
+		if (strcmp(key, "SCHILY.acl.access")==0) {
+			wp = utf8_decode(tar, value, strlen(value));
+			/* TODO: if (wp == NULL) */
+			__archive_entry_acl_parse_w(entry, wp,
 			    ARCHIVE_ENTRY_ACL_TYPE_ACCESS);
-		else if (wcscmp(key, L"SCHILY.acl.default")==0)
-			__archive_entry_acl_parse_w(entry, value,
+		} else if (strcmp(key, "SCHILY.acl.default")==0) {
+			wp = utf8_decode(tar, value, strlen(value));
+			/* TODO: if (wp == NULL) */
+			__archive_entry_acl_parse_w(entry, wp,
 			    ARCHIVE_ENTRY_ACL_TYPE_DEFAULT);
-		else if (wcscmp(key, L"SCHILY.devmajor")==0)
-			archive_entry_set_rdevmajor(entry, tar_atol10(value, wcslen(value)));
-		else if (wcscmp(key, L"SCHILY.devminor")==0)
-			archive_entry_set_rdevminor(entry, tar_atol10(value, wcslen(value)));
-		else if (wcscmp(key, L"SCHILY.fflags")==0)
-			archive_entry_copy_fflags_text_w(entry, value);
-		else if (wcscmp(key, L"SCHILY.dev")==0)
-			archive_entry_set_dev(entry, tar_atol10(value, wcslen(value)));
-		else if (wcscmp(key, L"SCHILY.ino")==0)
-			archive_entry_set_ino(entry, tar_atol10(value, wcslen(value)));
-		else if (wcscmp(key, L"SCHILY.nlink")==0)
-			archive_entry_set_nlink(entry, tar_atol10(value, wcslen(value)));
-		else if (wcscmp(key, L"SCHILY.realsize")==0) {
-			tar->realsize = tar_atol10(value, wcslen(value));
+		} else if (strcmp(key, "SCHILY.devmajor")==0) {
+			archive_entry_set_rdevmajor(entry,
+			    tar_atol10(value, strlen(value)));
+		} else if (strcmp(key, "SCHILY.devminor")==0) {
+			archive_entry_set_rdevminor(entry,
+			    tar_atol10(value, strlen(value)));
+		} else if (strcmp(key, "SCHILY.fflags")==0) {
+			wp = utf8_decode(tar, value, strlen(value));
+			/* TODO: if (wp == NULL) */
+			archive_entry_copy_fflags_text_w(entry, wp);
+		} else if (strcmp(key, "SCHILY.dev")==0) {
+			archive_entry_set_dev(entry,
+			    tar_atol10(value, strlen(value)));
+		} else if (strcmp(key, "SCHILY.ino")==0) {
+			archive_entry_set_ino(entry,
+			    tar_atol10(value, strlen(value)));
+		} else if (strcmp(key, "SCHILY.nlink")==0) {
+			archive_entry_set_nlink(entry,
+			    tar_atol10(value, strlen(value)));
+		} else if (strcmp(key, "SCHILY.realsize")==0) {
+			tar->realsize = tar_atol10(value, strlen(value));
 			archive_entry_set_size(entry, tar->realsize);
 		}
 		break;
 	case 'a':
-		if (wcscmp(key, L"atime")==0) {
+		if (strcmp(key, "atime")==0) {
 			pax_time(value, &s, &n);
 			archive_entry_set_atime(entry, s, n);
 		}
 		break;
 	case 'c':
-		if (wcscmp(key, L"ctime")==0) {
+		if (strcmp(key, "ctime")==0) {
 			pax_time(value, &s, &n);
 			archive_entry_set_ctime(entry, s, n);
-		} else if (wcscmp(key, L"charset")==0) {
+		} else if (strcmp(key, "charset")==0) {
 			/* TODO: Publish charset information in entry. */
-		} else if (wcscmp(key, L"comment")==0) {
+		} else if (strcmp(key, "comment")==0) {
 			/* TODO: Publish comment in entry. */
 		}
 		break;
 	case 'g':
-		if (wcscmp(key, L"gid")==0)
-			archive_entry_set_gid(entry, tar_atol10(value, wcslen(value)));
-		else if (wcscmp(key, L"gname")==0)
-			archive_entry_copy_gname_w(entry, value);
+		if (strcmp(key, "gid")==0) {
+			archive_entry_set_gid(entry,
+			    tar_atol10(value, strlen(value)));
+		} else if (strcmp(key, "gname")==0) {
+			archive_strcpy(&(tar->entry_gname), value);
+		}
+		break;
+	case 'h':
+		if (strcmp(key, "hdrcharset") == 0) {
+			if (strcmp(value, "BINARY") == 0)
+				tar->pax_hdrcharset_binary = 1;
+			else if (strcmp(value, "ISO-IR 10646 2000 UTF-8") == 0)
+				tar->pax_hdrcharset_binary = 0;
+			else {
+				/* TODO: Warn about unsupported hdrcharset */
+			}
+		}
 		break;
 	case 'l':
 		/* pax interchange doesn't distinguish hardlink vs. symlink. */
-		if (wcscmp(key, L"linkpath")==0) {
-			if (archive_entry_hardlink(entry))
-				archive_entry_copy_hardlink_w(entry, value);
-			else
-				archive_entry_copy_symlink_w(entry, value);
+		if (strcmp(key, "linkpath")==0) {
+			archive_strcpy(&(tar->entry_linkpath), value);
 		}
 		break;
 	case 'm':
-		if (wcscmp(key, L"mtime")==0) {
+		if (strcmp(key, "mtime")==0) {
 			pax_time(value, &s, &n);
 			archive_entry_set_mtime(entry, s, n);
 		}
 		break;
 	case 'p':
-		if (wcscmp(key, L"path")==0)
-			archive_entry_copy_pathname_w(entry, value);
+		if (strcmp(key, "path")==0) {
+			archive_strcpy(&(tar->entry_pathname), value);
+		}
 		break;
 	case 'r':
 		/* POSIX has reserved 'realtime.*' */
 		break;
 	case 's':
 		/* POSIX has reserved 'security.*' */
-		/* Someday: if (wcscmp(key, L"security.acl")==0) { ... } */
-		if (wcscmp(key, L"size")==0) {
+		/* Someday: if (strcmp(key, "security.acl")==0) { ... } */
+		if (strcmp(key, "size")==0) {
 			/* "size" is the size of the data in the entry. */
 			tar->entry_bytes_remaining
-			    = tar_atol10(value, wcslen(value));
+			    = tar_atol10(value, strlen(value));
 			/*
 			 * But, "size" is not necessarily the size of
 			 * the file on disk; if this is a sparse file,
@@ -1497,10 +1551,12 @@ pax_attribute(struct tar *tar, struct archive_entry *entry,
 		}
 		break;
 	case 'u':
-		if (wcscmp(key, L"uid")==0)
-			archive_entry_set_uid(entry, tar_atol10(value, wcslen(value)));
-		else if (wcscmp(key, L"uname")==0)
-			archive_entry_copy_uname_w(entry, value);
+		if (strcmp(key, "uid")==0) {
+			archive_entry_set_uid(entry,
+			    tar_atol10(value, strlen(value)));
+		} else if (strcmp(key, "uname")==0) {
+			archive_strcpy(&(tar->entry_uname), value);
+		}
 		break;
 	}
 	return (0);
@@ -1512,7 +1568,7 @@ pax_attribute(struct tar *tar, struct archive_entry *entry,
  * parse a decimal time value, which may include a fractional portion
  */
 static void
-pax_time(const wchar_t *p, int64_t *ps, long *pn)
+pax_time(const char *p, int64_t *ps, long *pn)
 {
 	char digit;
 	int64_t	s;
@@ -1580,9 +1636,9 @@ header_gnutar(struct archive_read *a, struct tar *tar,
 
 	/* Copy filename over (to ensure null termination). */
 	header = (const struct archive_entry_header_gnutar *)h;
-	archive_strncpy(&(tar->entry_name), header->name,
+	archive_strncpy(&(tar->entry_pathname), header->name,
 	    sizeof(header->name));
-	archive_entry_set_pathname(entry, tar->entry_name.s);
+	archive_entry_copy_pathname(entry, tar->entry_pathname.s);
 
 	/* Fields common to ustar and GNU */
 	/* XXX Can the following be factored out since it's common
@@ -1590,11 +1646,11 @@ header_gnutar(struct archive_read *a, struct tar *tar,
 	 * header_common, perhaps?  */
 	archive_strncpy(&(tar->entry_uname),
 	    header->uname, sizeof(header->uname));
-	archive_entry_set_uname(entry, tar->entry_uname.s);
+	archive_entry_copy_uname(entry, tar->entry_uname.s);
 
 	archive_strncpy(&(tar->entry_gname),
 	    header->gname, sizeof(header->gname));
-	archive_entry_set_gname(entry, tar->entry_gname.s);
+	archive_entry_copy_gname(entry, tar->entry_gname.s);
 
 	/* Parse out device numbers only for char and block specials */
 	if (header->typeflag[0] == '3' || header->typeflag[0] == '4') {
@@ -1748,9 +1804,9 @@ gnu_sparse_old_parse(struct tar *tar,
  */
 
 static int
-gnu_sparse_01_parse(struct tar *tar, const wchar_t *p)
+gnu_sparse_01_parse(struct tar *tar, const char *p)
 {
-	const wchar_t *e;
+	const char *e;
 	off_t offset = -1, size = -1;
 
 	for (;;) {
@@ -1785,12 +1841,11 @@ gnu_sparse_01_parse(struct tar *tar, const wchar_t *p)
  * don't support this format will extract the block map along with the
  * data and a separate post-process can restore the sparseness.
  *
- * Unfortunately, GNU tar 1.16 adds bogus padding to the end of the
- * entry that depends on the size of the map; this means we have to
- * parse the sparse map when we read the header (otherwise, entry_skip
- * will fail).  This is why sparse_10_read is called from read_header
- * above, instead of at the beginning of read_data, where it "should"
- * go.
+ * Unfortunately, GNU tar 1.16 had a bug that added unnecessary
+ * padding to the body of the file when using this format.  GNU tar
+ * 1.17 corrected this bug without bumping the version number, so
+ * it's not possible to support both variants.  This code supports
+ * the later variant at the expense of not supporting the former.
  *
  * This variant also replaced GNU.sparse.size with GNU.sparse.realsize
  * and introduced the GNU.sparse.major/GNU.sparse.minor attributes.
@@ -1954,7 +2009,7 @@ tar_atol8(const char *p, unsigned char_cnt)
  * it does obey locale.
  */
 static int64_t
-tar_atol10(const wchar_t *p, unsigned char_cnt)
+tar_atol10(const char *p, unsigned char_cnt)
 {
 	int64_t l, limit, last_digit_limit;
 	int base, digit, sign;
@@ -1987,10 +2042,7 @@ tar_atol10(const wchar_t *p, unsigned char_cnt)
 /*
  * Parse a base-256 integer.  This is just a straight signed binary
  * value in big-endian order, except that the high-order bit is
- * ignored.  Remember that "int64_t" may or may not be exactly 64
- * bits; the implementation here tries to avoid making any assumptions
- * about the actual size of an int64_t.  It does assume we're using
- * twos-complement arithmetic, though.
+ * ignored.
  */
 static int64_t
 tar_atol256(const char *_p, unsigned char_cnt)
@@ -2088,15 +2140,38 @@ readline(struct archive_read *a, struct tar *tar, const char **start,
 	}
 }
 
-static int
-utf8_decode(wchar_t *dest, const char *src, size_t length)
+static wchar_t *
+utf8_decode(struct tar *tar, const char *src, size_t length)
 {
-	size_t n;
+	wchar_t *dest;
+	ssize_t n;
 	int err;
 
+	/* Ensure pax_entry buffer is big enough. */
+	if (tar->pax_entry_length <= length) {
+		wchar_t *old_entry = tar->pax_entry;
+
+		if (tar->pax_entry_length <= 0)
+			tar->pax_entry_length = 1024;
+		while (tar->pax_entry_length <= length + 1)
+			tar->pax_entry_length *= 2;
+
+		old_entry = tar->pax_entry;
+		tar->pax_entry = (wchar_t *)realloc(tar->pax_entry,
+		    tar->pax_entry_length * sizeof(wchar_t));
+		if (tar->pax_entry == NULL) {
+			free(old_entry);
+			/* TODO: Handle this error. */
+			return (NULL);
+		}
+	}
+
+	dest = tar->pax_entry;
 	err = 0;
 	while (length > 0) {
 		n = UTF8_mbrtowc(dest, src, length);
+		if (n < 0)
+			return (NULL);
 		if (n == 0)
 			break;
 		dest++;
@@ -2104,13 +2179,13 @@ utf8_decode(wchar_t *dest, const char *src, size_t length)
 		length -= n;
 	}
 	*dest++ = L'\0';
-	return (err);
+	return (tar->pax_entry);
 }
 
 /*
  * Copied and simplified from FreeBSD libc/locale.
  */
-static size_t
+static ssize_t
 UTF8_mbrtowc(wchar_t *pwc, const char *s, size_t n)
 {
         int ch, i, len, mask;
@@ -2137,22 +2212,14 @@ UTF8_mbrtowc(wchar_t *pwc, const char *s, size_t n)
         } else if ((ch & 0xf8) == 0xf0) {
                 mask = 0x07;
                 len = 4;
-        } else if ((ch & 0xfc) == 0xf8) {
-                mask = 0x03;
-                len = 5;
-        } else if ((ch & 0xfe) == 0xfc) {
-                mask = 0x01;
-                len = 6;
         } else {
-		/* Invalid first byte; convert to '?' */
-		*pwc = '?';
-		return (1);
+		/* Invalid first byte. */
+		return (-1);
         }
 
         if (n < (size_t)len) {
-		/* Invalid first byte; convert to '?' */
-		*pwc = '?';
-                return (1);
+		/* Valid first byte but truncated. */
+                return (-2);
 	}
 
         /*
@@ -2198,7 +2265,7 @@ UTF8_mbrtowc(wchar_t *pwc, const char *s, size_t n)
  * omits line breaks; RFC1341 used for MIME requires both.)
  */
 static char *
-base64_decode(const wchar_t *src, size_t len, size_t *out_len)
+base64_decode(const char *s, size_t len, size_t *out_len)
 {
 	static const unsigned char digits[64] = {
 		'A','B','C','D','E','F','G','H','I','J','K','L','M','N',
@@ -2208,6 +2275,7 @@ base64_decode(const wchar_t *src, size_t len, size_t *out_len)
 		'4','5','6','7','8','9','+','/' };
 	static unsigned char decode_table[128];
 	char *out, *d;
+	const unsigned char *src = (const unsigned char *)s;
 
 	/* If the decode table is not yet initialized, prepare it. */
 	if (decode_table[digits[1]] != 1) {
@@ -2268,43 +2336,6 @@ base64_decode(const wchar_t *src, size_t len, size_t *out_len)
 	return (out);
 }
 
-/*
- * This is a little tricky because the C99 standard wcstombs()
- * function returns the number of bytes that were converted,
- * not the number that should be converted.  As a result,
- * we can never accurately size the output buffer (without
- * doing a tedious output size calculation in advance).
- * This approach (try a conversion, then try again if it fails)
- * will almost always succeed on the first try, and is thus
- * much faster, at the cost of sometimes requiring multiple
- * passes while we expand the buffer.
- */
-static char *
-wide_to_narrow(const wchar_t *wval)
-{
-	int converted_length;
-	/* Guess an output buffer size and try the conversion. */
-	int alloc_length = wcslen(wval) * 3;
-	char *mbs_val = (char *)malloc(alloc_length + 1);
-	if (mbs_val == NULL)
-		return (NULL);
-	converted_length = wcstombs(mbs_val, wval, alloc_length);
-
-	/* If we exhausted the buffer, resize and try again. */
-	while (converted_length >= alloc_length) {
-		free(mbs_val);
-		alloc_length *= 2;
-		mbs_val = (char *)malloc(alloc_length + 1);
-		if (mbs_val == NULL)
-			return (NULL);
-		converted_length = wcstombs(mbs_val, wval, alloc_length);
-	}
-
-	/* Ensure a trailing null and return the final string. */
-	mbs_val[alloc_length] = '\0';
-	return (mbs_val);
-}
-
 static char *
 url_decode(const char *in)
 {
diff --git a/lib/libarchive/archive_util.c b/lib/libarchive/archive_util.c
index e80cabb642db..cad14f4b810b 100644
--- a/lib/libarchive/archive_util.c
+++ b/lib/libarchive/archive_util.c
@@ -38,29 +38,71 @@ __FBSDID("$FreeBSD$");
 
 #include "archive.h"
 #include "archive_private.h"
+#include "archive_string.h"
 
+#if ARCHIVE_VERSION_NUMBER < 3000000
+/* These disappear in libarchive 3.0 */
+/* Deprecated. */
 int
 archive_api_feature(void)
 {
 	return (ARCHIVE_API_FEATURE);
 }
 
+/* Deprecated. */
 int
 archive_api_version(void)
 {
 	return (ARCHIVE_API_VERSION);
 }
 
+/* Deprecated synonym for archive_version_number() */
 int
 archive_version_stamp(void)
 {
-	return (ARCHIVE_VERSION_STAMP);
+	return (archive_version_number());
 }
 
+/* Deprecated synonym for archive_version_string() */
 const char *
 archive_version(void)
 {
-	return (ARCHIVE_LIBRARY_VERSION);
+	return (archive_version_string());
+}
+#endif
+
+int
+archive_version_number(void)
+{
+	return (ARCHIVE_VERSION_NUMBER);
+}
+
+/*
+ * Format a version string of the form "libarchive x.y.z", where x, y,
+ * z are the correct parts of the version ID from
+ * archive_version_number().
+ *
+ * I used to do all of this at build time in shell scripts but that
+ * proved to be a portability headache.
+ */
+
+const char *
+archive_version_string(void)
+{
+	static char buff[128];
+	struct archive_string as;
+	int n;
+
+	if (buff[0] == '\0') {
+		n = archive_version_number();
+		memset(&as, 0, sizeof(as));
+		archive_string_sprintf(&as, "libarchive %d.%d.%d",
+		    n / 1000000, (n / 1000) % 1000, n % 1000);
+		strncpy(buff, as.s, sizeof(buff));
+		buff[sizeof(buff) - 1] = '\0';
+		archive_string_free(&as);
+	}
+	return (buff);
 }
 
 int
diff --git a/lib/libarchive/archive_write_private.h b/lib/libarchive/archive_write_private.h
index 7764fdd9b4d9..8deabbf0e0e3 100644
--- a/lib/libarchive/archive_write_private.h
+++ b/lib/libarchive/archive_write_private.h
@@ -82,13 +82,6 @@ struct archive_write {
 	} compressor;
 
 	/*
-	 * Again, write support is considerably simpler because there's
-	 * no need for an auction.
-	 */
-	int		  archive_format;
-	const char	 *archive_format_name;
-
-	/*
 	 * Pointers to format-specific functions for writing.  They're
 	 * initialized by archive_write_set_format_XXX() calls.
 	 */
diff --git a/lib/libarchive/archive_write_set_format_pax.c b/lib/libarchive/archive_write_set_format_pax.c
index fbec7eb316c8..6f05a99292ad 100644
--- a/lib/libarchive/archive_write_set_format_pax.c
+++ b/lib/libarchive/archive_write_set_format_pax.c
@@ -85,8 +85,8 @@ archive_write_set_format_pax_restricted(struct archive *_a)
 	struct archive_write *a = (struct archive_write *)_a;
 	int r;
 	r = archive_write_set_format_pax(&a->archive);
-	a->archive_format = ARCHIVE_FORMAT_TAR_PAX_RESTRICTED;
-	a->archive_format_name = "restricted POSIX pax interchange";
+	a->archive.archive_format = ARCHIVE_FORMAT_TAR_PAX_RESTRICTED;
+	a->archive.archive_format_name = "restricted POSIX pax interchange";
 	return (r);
 }
 
@@ -116,8 +116,8 @@ archive_write_set_format_pax(struct archive *_a)
 	a->format_finish = archive_write_pax_finish;
 	a->format_destroy = archive_write_pax_destroy;
 	a->format_finish_entry = archive_write_pax_finish_entry;
-	a->archive_format = ARCHIVE_FORMAT_TAR_PAX_INTERCHANGE;
-	a->archive_format_name = "POSIX pax interchange";
+	a->archive.archive_format = ARCHIVE_FORMAT_TAR_PAX_INTERCHANGE;
+	a->archive.archive_format_name = "POSIX pax interchange";
 	return (ARCHIVE_OK);
 }
 
@@ -383,19 +383,25 @@ archive_write_pax_header(struct archive_write *a,
     struct archive_entry *entry_original)
 {
 	struct archive_entry *entry_main;
-	const char *linkname, *p;
+	const char *p;
 	char *t;
-	const char *hardlink;
 	const wchar_t *wp;
 	const char *suffix_start;
 	int need_extension, r, ret;
 	struct pax *pax;
+	const char *hdrcharset = NULL;
+	const char *hardlink;
+	const char *path = NULL, *linkpath = NULL;
+	const char *uname = NULL, *gname = NULL;
+	const wchar_t *path_w = NULL, *linkpath_w = NULL;
+	const wchar_t *uname_w = NULL, *gname_w = NULL;
 
 	char paxbuff[512];
 	char ustarbuff[512];
 	char ustar_entry_name[256];
 	char pax_entry_name[256];
 
+	ret = ARCHIVE_OK;
 	need_extension = 0;
 	pax = (struct pax *)a->format_data;
 
@@ -442,53 +448,109 @@ archive_write_pax_header(struct archive_write *a,
 	archive_string_empty(&(pax->pax_header)); /* Blank our work area. */
 
 	/*
+	 * First, check the name fields and see if any of them
+	 * require binary coding.  If any of them does, then all of
+	 * them do.
+	 */
+	hdrcharset = NULL;
+	path = archive_entry_pathname(entry_main);
+	path_w = archive_entry_pathname_w(entry_main);
+	if (path != NULL && path_w == NULL) {
+		archive_set_error(&a->archive, EILSEQ,
+		    "Can't translate pathname '%s' to UTF-8", path);
+		ret = ARCHIVE_WARN;
+		hdrcharset = "BINARY";
+	}
+	uname = archive_entry_uname(entry_main);
+	uname_w = archive_entry_uname_w(entry_main);
+	if (uname != NULL && uname_w == NULL) {
+		archive_set_error(&a->archive, EILSEQ,
+		    "Can't translate uname '%s' to UTF-8", uname);
+		ret = ARCHIVE_WARN;
+		hdrcharset = "BINARY";
+	}
+	gname = archive_entry_gname(entry_main);
+	gname_w = archive_entry_gname_w(entry_main);
+	if (gname != NULL && gname_w == NULL) {
+		archive_set_error(&a->archive, EILSEQ,
+		    "Can't translate gname '%s' to UTF-8", gname);
+		ret = ARCHIVE_WARN;
+		hdrcharset = "BINARY";
+	}
+	linkpath = hardlink;
+	if (linkpath != NULL) {
+		linkpath_w = archive_entry_hardlink_w(entry_main);
+	} else {
+		linkpath = archive_entry_symlink(entry_main);
+		if (linkpath != NULL)
+			linkpath_w = archive_entry_symlink_w(entry_main);
+	}
+	if (linkpath != NULL && linkpath_w == NULL) {
+		archive_set_error(&a->archive, EILSEQ,
+		    "Can't translate linkpath '%s' to UTF-8", linkpath);
+		ret = ARCHIVE_WARN;
+		hdrcharset = "BINARY";
+	}
+
+	/* Store the header encoding first, to be nice to readers. */
+	if (hdrcharset != NULL)
+		add_pax_attr(&(pax->pax_header), "hdrcharset", hdrcharset);
+
+	/*
 	 * Determining whether or not the name is too big is ugly
 	 * because of the rules for dividing names between 'name' and
 	 * 'prefix' fields.  Here, I pick out the longest possible
 	 * suffix, then test whether the remaining prefix is too long.
 	 */
-	wp = archive_entry_pathname_w(entry_main);
-	p = archive_entry_pathname(entry_main);
-	if (strlen(p) <= 100)	/* Short enough for just 'name' field */
-		suffix_start = p;	/* Record a zero-length prefix */
+	if (strlen(path) <= 100)    /* Short enough for just 'name' field */
+		suffix_start = path;	/* Record a zero-length prefix */
 	else
 		/* Find the largest suffix that fits in 'name' field. */
-		suffix_start = strchr(p + strlen(p) - 100 - 1, '/');
+		suffix_start = strchr(path + strlen(path) - 100 - 1, '/');
 
 	/*
 	 * If name is too long, or has non-ASCII characters, add
-	 * 'path' to pax extended attrs.
+	 * 'path' to pax extended attrs.  (Note that an unconvertible
+	 * name must have non-ASCII characters.)
 	 */
-	if (suffix_start == NULL || suffix_start - p > 155 || has_non_ASCII(wp)) {
-		add_pax_attr_w(&(pax->pax_header), "path", wp);
+	if (suffix_start == NULL || suffix_start - path > 155
+	    || path_w == NULL || has_non_ASCII(path_w)) {
+		if (path_w == NULL || hdrcharset != NULL)
+			/* Can't do UTF-8, so store it raw. */
+			add_pax_attr(&(pax->pax_header), "path", path);
+		else
+			add_pax_attr_w(&(pax->pax_header), "path", path_w);
 		archive_entry_set_pathname(entry_main,
-		    build_ustar_entry_name(ustar_entry_name, p, strlen(p), NULL));
+		    build_ustar_entry_name(ustar_entry_name,
+			path, strlen(path), NULL));
 		need_extension = 1;
 	}
 
-	/* If link name is too long or has non-ASCII characters, add
-	 * 'linkpath' to pax extended attrs. */
-	linkname = hardlink;
-	if (linkname == NULL)
-		linkname = archive_entry_symlink(entry_main);
-
-	if (linkname != NULL) {
-		/* There is a link name, get the wide version as well. */
-		if (hardlink != NULL)
-			wp = archive_entry_hardlink_w(entry_main);
-		else
-			wp = archive_entry_symlink_w(entry_main);
-
-		/* If the link is long or has a non-ASCII character,
-		 * store it as a pax extended attribute. */
-		if (strlen(linkname) > 100 || has_non_ASCII(wp)) {
-			add_pax_attr_w(&(pax->pax_header), "linkpath", wp);
-			if (hardlink != NULL)
-				archive_entry_set_hardlink(entry_main,
-				    "././@LongHardLink");
+	if (linkpath != NULL) {
+		/* If link name is too long or has non-ASCII characters, add
+		 * 'linkpath' to pax extended attrs. */
+		if (strlen(linkpath) > 100 || linkpath_w == NULL
+		    || linkpath_w == NULL || has_non_ASCII(linkpath_w)) {
+			if (linkpath_w == NULL || hdrcharset != NULL)
+				/* If the linkpath is not convertible
+				 * to wide, or we're encoding in
+				 * binary anyway, store it raw. */
+				add_pax_attr(&(pax->pax_header),
+				    "linkpath", linkpath);
 			else
-				archive_entry_set_symlink(entry_main,
-				    "././@LongSymLink");
+				/* If the link is long or has a
+				 * non-ASCII character, store it as a
+				 * pax extended attribute. */
+				add_pax_attr_w(&(pax->pax_header),
+				    "linkpath", linkpath_w);
+			if (strlen(linkpath) > 100) {
+				if (hardlink != NULL)
+					archive_entry_set_hardlink(entry_main,
+					    "././@LongHardLink");
+				else
+					archive_entry_set_symlink(entry_main,
+					    "././@LongSymLink");
+			}
 			need_extension = 1;
 		}
 	}
@@ -509,12 +571,20 @@ archive_write_pax_header(struct archive_write *a,
 
 	/* If group name is too large or has non-ASCII characters, add
 	 * 'gname' to pax extended attrs. */
-	p = archive_entry_gname(entry_main);
-	wp = archive_entry_gname_w(entry_main);
-	if (p != NULL && (strlen(p) > 31 || has_non_ASCII(wp))) {
-		add_pax_attr_w(&(pax->pax_header), "gname", wp);
-		archive_entry_set_gname(entry_main, NULL);
-		need_extension = 1;
+	if (gname != NULL) {
+		if (strlen(gname) > 31
+		    || gname_w == NULL
+		    || has_non_ASCII(gname_w))
+		{
+			if (gname_w == NULL || hdrcharset != NULL) {
+				add_pax_attr(&(pax->pax_header),
+				    "gname", gname);
+			} else  {
+				add_pax_attr_w(&(pax->pax_header),
+				    "gname", gname_w);
+			}
+			need_extension = 1;
+		}
 	}
 
 	/* If numeric UID is too large, add 'uid' to pax extended attrs. */
@@ -524,14 +594,21 @@ archive_write_pax_header(struct archive_write *a,
 		need_extension = 1;
 	}
 
-	/* If user name is too large, add 'uname' to pax extended attrs. */
-	/* TODO: If uname has non-ASCII characters, use pax attribute. */
-	p = archive_entry_uname(entry_main);
-	wp = archive_entry_uname_w(entry_main);
-	if (p != NULL && (strlen(p) > 31 || has_non_ASCII(wp))) {
-		add_pax_attr_w(&(pax->pax_header), "uname", wp);
-		archive_entry_set_uname(entry_main, NULL);
-		need_extension = 1;
+	/* Add 'uname' to pax extended attrs if necessary. */
+	if (uname != NULL) {
+		if (strlen(uname) > 31
+		    || uname_w == NULL
+		    || has_non_ASCII(uname_w))
+		{
+			if (uname_w == NULL || hdrcharset != NULL) {
+				add_pax_attr(&(pax->pax_header),
+				    "uname", uname);
+			} else {
+				add_pax_attr_w(&(pax->pax_header),
+				    "uname", uname_w);
+			}
+			need_extension = 1;
+		}
 	}
 
 	/*
@@ -624,7 +701,7 @@ archive_write_pax_header(struct archive_write *a,
 	 * already set (we're already generating an extended header, so
 	 * may as well include these).
 	 */
-	if (a->archive_format != ARCHIVE_FORMAT_TAR_PAX_RESTRICTED ||
+	if (a->archive.archive_format != ARCHIVE_FORMAT_TAR_PAX_RESTRICTED ||
 	    need_extension) {
 
 		if (archive_entry_mtime(entry_main) < 0  ||
@@ -687,7 +764,7 @@ archive_write_pax_header(struct archive_write *a,
 	 * Pax-restricted does not store data for hardlinks, in order
 	 * to improve compatibility with ustar.
 	 */
-	if (a->archive_format != ARCHIVE_FORMAT_TAR_PAX_INTERCHANGE &&
+	if (a->archive.archive_format != ARCHIVE_FORMAT_TAR_PAX_INTERCHANGE &&
 	    hardlink != NULL)
 		archive_entry_set_size(entry_main, 0);
 
@@ -733,7 +810,6 @@ archive_write_pax_header(struct archive_write *a,
 	__archive_write_format_header_ustar(a, ustarbuff, entry_main, -1, 0);
 
 	/* If we built any extended attributes, write that entry first. */
-	ret = ARCHIVE_OK;
 	if (archive_strlen(&(pax->pax_header)) > 0) {
 		struct archive_entry *pax_attr_entry;
 		time_t s;
@@ -793,13 +869,13 @@ archive_write_pax_header(struct archive_write *a,
 		/* Standard ustar doesn't support ctime. */
 		archive_entry_set_ctime(pax_attr_entry, 0, 0);
 
-		ret = __archive_write_format_header_ustar(a, paxbuff,
+		r = __archive_write_format_header_ustar(a, paxbuff,
 		    pax_attr_entry, 'x', 1);
 
 		archive_entry_free(pax_attr_entry);
 
 		/* Note that the 'x' header shouldn't ever fail to format */
-		if (ret != 0) {
+		if (r != 0) {
 			const char *msg = "archive_write_pax_header: "
 			    "'x' header failed?!  This can't happen.\n";
 			write(2, msg, strlen(msg));
@@ -984,23 +1060,31 @@ build_ustar_entry_name(char *dest, const char *src, size_t src_length,
 
 /*
  * The ustar header for the pax extended attributes must have a
- * reasonable name:  SUSv3 suggests 'dirname'/PaxHeader/'filename'
+ * reasonable name:  SUSv3 requires 'dirname'/PaxHeader.'pid'/'filename'
+ * where 'pid' is the PID of the archiving process.  Unfortunately,
+ * that makes testing a pain since the output varies for each run,
+ * so I'm sticking with the simpler 'dirname'/PaxHeader/'filename'
+ * for now.  (Someday, I'll make this settable.  Then I can use the
+ * SUS recommendation as default and test harnesses can override it
+ * to get predictable results.)
  *
- * Joerg Schiling has argued that this is unnecessary because, in practice,
- * if the pax extended attributes get extracted as regular files, noone is
- * going to bother reading those attributes to manually restore them.
- * Based on this, 'star' uses /tmp/PaxHeader/'basename' as the ustar header
- * name.  This is a tempting argument, but I'm not entirely convinced.
- * I'm also uncomfortable with the fact that "/tmp" is a Unix-ism.
+ * Joerg Schilling has argued that this is unnecessary because, in
+ * practice, if the pax extended attributes get extracted as regular
+ * files, noone is going to bother reading those attributes to
+ * manually restore them.  Based on this, 'star' uses
+ * /tmp/PaxHeader/'basename' as the ustar header name.  This is a
+ * tempting argument, in part because it's simpler than the SUSv3
+ * recommendation, but I'm not entirely convinced.  I'm also
+ * uncomfortable with the fact that "/tmp" is a Unix-ism.
  *
- * The following routine implements the SUSv3 recommendation, and is
- * much simpler because build_ustar_entry_name() above already does
- * most of the work (we just need to give it an extra path element to
- * insert and handle a few pathological cases).
+ * The following routine leverages build_ustar_entry_name() above and
+ * so is simpler than you might think.  It just needs to provide the
+ * additional path element and handle a few pathological cases).
  */
 static char *
 build_pax_attribute_name(char *dest, const char *src)
 {
+	char buff[64];
 	const char *p;
 
 	/* Handle the null filename case. */
@@ -1039,8 +1123,19 @@ build_pax_attribute_name(char *dest, const char *src)
 		return (dest);
 	}
 
+	/*
+	 * TODO: Push this string into the 'pax' structure to avoid
+	 * recomputing it every time.  That will also open the door
+	 * to having clients override it.
+	 */
+#if HAVE_GETPID && 0  /* Disable this for now; see above comment. */
+	sprintf(buff, "PaxHeader.%d", getpid());
+#else
+	/* If the platform can't fetch the pid, don't include it. */
+	strcpy(buff, "PaxHeader");
+#endif
 	/* General case: build a ustar-compatible name adding "/PaxHeader/". */
-	build_ustar_entry_name(dest, src, p - src, "PaxHeader");
+	build_ustar_entry_name(dest, src, p - src, buff);
 
 	return (dest);
 }