X-Git-Url: https://wimlib.net/git/?p=wimlib;a=blobdiff_plain;f=src%2Fencoding.c;h=9337c9a16e6a277b2f4880058103b4cdd1377eee;hp=edb24c70796ff3efc73b36d16aad812df2e12c9a;hb=4a20aae0dd8469a352517a0b107416ffa99ccc55;hpb=ded522fd0a15d740354329c5066ebd3473563e57

diff --git a/src/encoding.c b/src/encoding.c
index edb24c70..9337c9a1 100644
--- a/src/encoding.c
+++ b/src/encoding.c
@@ -1,332 +1,464 @@
 /*
- * encoding.c:  Convert UTF-8 to UTF-16LE strings and vice versa
- */
-
-/*
- * Copyright (C) 2012 Eric Biggers
+ * encoding.c - UTF-8 and UTF-16LE codecs and utility functions
  *
- * This file is part of wimlib, a library for working with WIM files.
+ * Copyright (C) 2012-2016 Eric Biggers
  *
- * wimlib is free software; you can redistribute it and/or modify it under the
- * terms of the GNU General Public License as published by the Free
- * Software Foundation; either version 3 of the License, or (at your option)
- * any later version.
+ * This file is free software; you can redistribute it and/or modify it under
+ * the terms of the GNU Lesser General Public License as published by the Free
+ * Software Foundation; either version 3 of the License, or (at your option) any
+ * later version.
  *
- * wimlib is distributed in the hope that it will be useful, but WITHOUT ANY
- * WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR
- * A PARTICULAR PURPOSE. See the GNU General Public License for more
+ * This file is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+ * FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License for more
  * details.
  *
- * You should have received a copy of the GNU General Public License
- * along with wimlib; if not, see http://www.gnu.org/licenses/.
+ * You should have received a copy of the GNU Lesser General Public License
+ * along with this file; if not, see http://www.gnu.org/licenses/.
  */
 
-#include "wimlib.h"
-#include "util.h"
-#include "endianness.h"
+#ifdef HAVE_CONFIG_H
+#  include "config.h"
+#endif
 
 #include <errno.h>
+#include <string.h>
 
-#ifdef WITH_NTFS_3G
-#include <ntfs-3g/volume.h>
-#include <ntfs-3g/unistr.h>
-#else
-#include <iconv.h>
-#endif
+#include "wimlib/encoding.h"
+#include "wimlib/endianness.h"
+#include "wimlib/error.h"
+#include "wimlib/unaligned.h"
+#include "wimlib/util.h"
 
 /*
- * NOTE:
+ * Allow unpaired surrogates, such as might exist in Windows-style filenames ---
+ * which are normally valid UTF-16LE, but are actually treated as opaque
+ * sequences of 16-bit WCHARs by Windows.  When decoding "UTF-16LE", unpaired
+ * surrogates will be decoded as their surrogate codepoints; and when encoding
+ * to and from "UTF-8", the encoding will actually be WTF-8 ("Wobbly
+ * Transformation Format - 8-bit"), a superset of UTF-8 which permits the
+ * surrogate codepoints.
+ *
+ * In combination with also allowing the "non-character" codepoints U+FFFE and
+ * U+FFFF, the result is that every Windows-style filename can be translated to
+ * a UNIX-style filename.
  *
- * utf16_to_utf8_size() and utf8_to_utf16_size() were taken from
- * libntfs-3g/unistr.c in the NTFS-3g sources.  (Modified slightly to remove
- * unneeded functionality.)
+ * Unfortunately, the converse is not true: not every UNIX filename can be
+ * translated to a Windows filename.  Only UNIX filenames that are valid "WTF-8"
+ * can be translated.  I considered ways to define a bijective mapping, but
+ * there did not seem to be a straightforward way.  The "UTF-8b" scheme, for
+ * example, would map each invalid byte 'b' to a surrogate "escape code" 'U+DC00
+ * + b'.  The problem with this was that surrogate escape codes can be combined
+ * to create a valid UTF-8 sequence, thus breaking the bijection by mapping
+ * multiple Windows filenames to a single UNIX filename.
  */
-#ifndef WITH_NTFS_3G
+#define ALLOW_UNPAIRED_SURROGATES	1
+
+#define INVALID_CODEPOINT	0xFFFFFFFF
+#define VALIDATE(expr)		if (validate && unlikely(!(expr))) goto invalid
+#define IS_SURROGATE(c)		((c) >= 0xD800 && (c) < 0xE000)
+#define IS_HIGH_SURROGATE(c)	((c) >= 0xD800 && (c) < 0xDC00)
+#define IS_LOW_SURROGATE(c)	((c) >= 0xDC00 && (c) < 0xE000)
+#define IS_UTF8_TAIL(c)		(((c) & 0xC0) == 0x80)
+
 /*
- * Return the amount of 8-bit elements in UTF-8 needed (without the terminating
- * null) to store a given UTF-16LE string.
+ * Decode the next Unicode codepoint from the string at @in, which has
+ * @remaining >= 1 bytes remaining.  Return the number of bytes consumed and
+ * write the decoded codepoint to *c_ret.
  *
- * Return -1 with errno set if string has invalid byte sequence or too long.
+ * If the input might not be a valid string in the source encoding, then
+ * @validate must be specified as %true, and then on invalid input the function
+ * consumes at least one byte and sets *c_ret to INVALID_CODEPOINT.  If the
+ * input is guaranteed to be valid, then @validate may be specified as %false.
  */
-static int utf16_to_utf8_size(const u16 *ins, const int ins_len)
+typedef unsigned (*decode_codepoint_fn)(const u8 *in, size_t remaining,
+					bool validate, u32 *c_ret);
+
+/* Encode the Unicode codepoint @c and return the number of bytes used. */
+typedef unsigned (*encode_codepoint_fn)(u32 c, u8 *out);
+
+static forceinline unsigned
+utf8_decode_codepoint(const u8 *in, size_t remaining, bool validate, u32 *c_ret)
 {
-	int i, ret = -1;
-	int count = 0;
-	bool surrog;
-
-	surrog = false;
-	for (i = 0; i < ins_len && ins[i]; i++) {
-		unsigned short c = le16_to_cpu(ins[i]);
-		if (surrog) {
-			if ((c >= 0xdc00) && (c < 0xe000)) {
-				surrog = false;
-				count += 4;
-			} else
-				goto fail;
-		} else
-			if (c < 0x80)
-				count++;
-			else if (c < 0x800)
-				count += 2;
-			else if (c < 0xd800)
-				count += 3;
-			else if (c < 0xdc00)
-				surrog = true;
-#if NOREVBOM
-			else if ((c >= 0xe000) && (c < 0xfffe))
-#else
-			else if (c >= 0xe000)
+	if (likely(in[0] < 0x80)) { /* U+0...U+7F */
+		*c_ret = in[0];
+		return 1;
+	}
+
+	if (in[0] < 0xE0) { /* U+80...U+7FF */
+		VALIDATE(in[0] >= 0xC2 && remaining >= 2 &&
+			 IS_UTF8_TAIL(in[1]));
+		*c_ret = ((u32)(in[0] & 0x1F) << 6) |
+			 ((u32)(in[1] & 0x3F) << 0);
+		return 2;
+	}
+
+	if (in[0] < 0xF0) { /* U+800...U+FFFF, possibly excluding surrogates */
+		VALIDATE(remaining >= 3 &&
+			 IS_UTF8_TAIL(in[1]) &&
+			 IS_UTF8_TAIL(in[2]));
+		*c_ret = ((u32)(in[0] & 0x0F) << 12) |
+			 ((u32)(in[1] & 0x3F) << 6) |
+			 ((u32)(in[2] & 0x3F) << 0);
+		VALIDATE(*c_ret >= 0x800);
+	#if !ALLOW_UNPAIRED_SURROGATES
+		VALIDATE(!IS_SURROGATE(*c_ret));
+	#endif
+		return 3;
+	}
+
+	/* U+10000...U+10FFFF */
+	VALIDATE(in[0] < 0xF8 && remaining >= 4 &&
+		 IS_UTF8_TAIL(in[1]) &&
+		 IS_UTF8_TAIL(in[2]) &&
+		 IS_UTF8_TAIL(in[3]));
+	*c_ret = ((u32)(in[0] & 0x07) << 18) |
+		 ((u32)(in[1] & 0x3F) << 12) |
+		 ((u32)(in[2] & 0x3F) << 6) |
+		 ((u32)(in[3] & 0x3F) << 0);
+	VALIDATE(*c_ret >= 0x10000 && *c_ret <= 0x10FFFF);
+	return 4;
+
+invalid:
+	*c_ret = INVALID_CODEPOINT;
+	return 1;
+}
+
+static forceinline unsigned
+utf8_encode_codepoint(u32 c, u8 *out)
+{
+	if (likely(c < 0x80)) {
+		out[0] = c;
+		return 1;
+	}
+
+	if (c < 0x800) {
+		out[0] = 0xC0 | (c >> 6);
+		out[1] = 0x80 | (c & 0x3F);
+		return 2;
+	}
+
+	if (c < 0x10000) {
+		out[0] = 0xE0 | (c >> 12);
+		out[1] = 0x80 | ((c >> 6) & 0x3F);
+		out[2] = 0x80 | (c & 0x3F);
+		return 3;
+	}
+
+	out[0] = 0xF0 | (c >> 18);
+	out[1] = 0x80 | ((c >> 12) & 0x3F);
+	out[2] = 0x80 | ((c >> 6) & 0x3F);
+	out[3] = 0x80 | (c & 0x3F);
+	return 4;
+}
+
+static forceinline unsigned
+utf16le_decode_codepoint(const u8 *in, size_t remaining, bool validate,
+			 u32 *c_ret)
+{
+	u32 h, l;
+
+	VALIDATE(remaining >= 2);
+	h = get_unaligned_le16(in);
+	if (unlikely(IS_SURROGATE(h))) {
+		/* Surrogate pairs are U+10000...U+10FFFF.
+		 * Unpaired surrogates are U+D800...U+DFFF. */
+	#if ALLOW_UNPAIRED_SURROGATES
+		if (unlikely(!IS_HIGH_SURROGATE(h) || remaining < 4))
+			goto unpaired;
+		l = get_unaligned_le16(in + 2);
+		if (unlikely(!IS_LOW_SURROGATE(l)))
+			goto unpaired;
+	#else
+		VALIDATE(IS_HIGH_SURROGATE(h) && remaining >= 4);
+		l = get_unaligned_le16(in + 2);
+		VALIDATE(IS_LOW_SURROGATE(l));
+	#endif
+		*c_ret = 0x10000 + ((h - 0xD800) << 10) + (l - 0xDC00);
+		return 4;
+	}
+#if ALLOW_UNPAIRED_SURROGATES
+unpaired:
 #endif
-				count += 3;
-			else
-				goto fail;
+	*c_ret = h;
+	return 2;
+
+invalid:
+	*c_ret = INVALID_CODEPOINT;
+	return min(remaining, 2);
+}
+
+static forceinline unsigned
+utf16le_encode_codepoint(u32 c, u8 *out)
+{
+	if (likely(c < 0x10000)) {
+		put_unaligned_le16(c, out);
+		return 2;
 	}
-	if (surrog)
-		goto fail;
-
-	ret = count;
-out:
-	return ret;
-fail:
-	errno = EILSEQ;
-	goto out;
+	c -= 0x10000;
+	put_unaligned_le16(0xD800 + (c >> 10), out);
+	put_unaligned_le16(0xDC00 + (c & 0x3FF), out + 2);
+	return 4;
 }
 
 /*
- * Return the amount of 16-bit elements in UTF-16LE needed
- * (without the terminating null) to store given UTF-8 string.
+ * Convert the string @in of size @in_nbytes from the encoding given by the
+ * @decode_codepoint function to the encoding given by the @encode_codepoint
+ * function.  @in does not need to be null-terminated, but a null terminator
+ * will be added to the output string.
  *
- * Return -1 with errno set if it's longer than PATH_MAX or string is invalid.
+ * On success, write the allocated output string to @out_ret (must not be NULL)
+ * and its size excluding the null terminator to @out_nbytes_ret (may be NULL).
  *
- * Note: This does not check whether the input sequence is a valid utf8 string,
- *	 and should be used only in context where such check is made!
+ * If the input string is malformed, return @ilseq_err with errno set to EILSEQ.
+ * If out of memory, return WIMLIB_ERR_NOMEM with errno set to ENOMEM.
  */
-static int utf8_to_utf16_size(const char *s)
+static forceinline int
+convert_string(const u8 * const in, const size_t in_nbytes,
+	       u8 **out_ret, size_t *out_nbytes_ret,
+	       int ilseq_err,
+	       decode_codepoint_fn decode_codepoint,
+	       encode_codepoint_fn encode_codepoint)
 {
-	unsigned int byte;
-	size_t count = 0;
-	while ((byte = *((const unsigned char *)s++))) {
-		count++;
-		if (byte >= 0xc0) {
-			if (byte >= 0xF5) {
-				errno = EILSEQ;
-				return -1;
-			}
-			if (!*s)
-				break;
-			if (byte >= 0xC0)
-				s++;
-			if (!*s)
-				break;
-			if (byte >= 0xE0)
-				s++;
-			if (!*s)
-				break;
-			if (byte >= 0xF0) {
-				s++;
-				count++;
-			}
+	const u8 * const in_end = in + in_nbytes;
+	const u8 *p_in;
+	u8 *p_out;
+	size_t out_nbytes = 0;
+	u8 *out;
+	u8 tmp[8]; /* assuming no codepoint requires > 8 bytes to encode */
+	u32 c;
+
+	/* Validate the input string and compute the output size. */
+	for (p_in = in; p_in != in_end; ) {
+		p_in += (*decode_codepoint)(p_in, in_end - p_in, true, &c);
+		if (unlikely(c == INVALID_CODEPOINT)) {
+			errno = EILSEQ;
+			return ilseq_err;
 		}
+		out_nbytes += (*encode_codepoint)(c, tmp);
 	}
-	return count;
+
+	/* Allocate the output string, including space for a null terminator. */
+	out = MALLOC(out_nbytes + (*encode_codepoint)(0, tmp));
+	if (unlikely(!out))
+		return WIMLIB_ERR_NOMEM;
+
+	/* Do the conversion. */
+	for (p_in = in, p_out = out; p_in != in_end; ) {
+		p_in += (*decode_codepoint)(p_in, in_end - p_in, false, &c);
+		p_out += (*encode_codepoint)(c, p_out);
+	}
+
+	/* Add a null terminator. */
+	(*encode_codepoint)(0, p_out);
+
+	/* Return the output string and its size (by reference). */
+	*out_ret = out;
+	if (out_nbytes_ret)
+		*out_nbytes_ret = out_nbytes;
+	return 0;
 }
-#endif /* !WITH_NTFS_3G */
 
-/* Converts a string in the UTF-16LE encoding to a newly allocated string in the
- * UTF-8 encoding.
- *
- * If available, do so by calling a similar function from libntfs-3g.
- * Otherwise, use iconv() along with the helper function utf16_to_utf8_size().
- */
-int utf16_to_utf8(const char *utf16_str, size_t utf16_nbytes,
-		  char **utf8_str_ret, size_t *utf8_nbytes_ret)
+int
+utf8_to_utf16le(const char *in, size_t in_nbytes,
+		utf16lechar **out_ret, size_t *out_nbytes_ret)
 {
-	int ret;
+	return convert_string((const u8 *)in, in_nbytes,
+			      (u8 **)out_ret, out_nbytes_ret,
+			      WIMLIB_ERR_INVALID_UTF8_STRING,
+			      utf8_decode_codepoint, utf16le_encode_codepoint);
+}
 
-	if (utf16_nbytes == 0) {
-		*utf8_str_ret = NULL;
-		*utf8_nbytes_ret = 0;
-		return 0;
-	}
+int
+utf16le_to_utf8(const utf16lechar *in, size_t in_nbytes,
+		char **out_ret, size_t *out_nbytes_ret)
+{
+	return convert_string((const u8 *)in, in_nbytes,
+			      (u8 **)out_ret, out_nbytes_ret,
+			      WIMLIB_ERR_INVALID_UTF16_STRING,
+			      utf16le_decode_codepoint, utf8_encode_codepoint);
+}
 
-	if (utf16_nbytes & 1) {
-		ERROR("UTF-16LE string is invalid (odd number of bytes)!");
-		return WIMLIB_ERR_INVALID_UTF16_STRING;
-	}
-#ifdef WITH_NTFS_3G
-	char *outs = NULL;
-	int outs_len = ntfs_ucstombs((const ntfschar*)utf16_str,
-				     utf16_nbytes / 2, &outs, 0);
-	if (outs_len >= 0) {
-		*utf8_str_ret = outs;
-		*utf8_nbytes_ret = outs_len;
-		ret = 0;
-	} else {
-		if (errno == ENOMEM)
-			ret = WIMLIB_ERR_NOMEM;
-		else
-			ret = WIMLIB_ERR_INVALID_UTF16_STRING;
-	}
-#else /* WITH_NTFS_3G */
-	static iconv_t cd_utf16_to_utf8 = (iconv_t)(-1);
-	if (cd_utf16_to_utf8 == (iconv_t)(-1)) {
-		cd_utf16_to_utf8 = iconv_open("UTF-8", "UTF-16LE");
-		if (cd_utf16_to_utf8 == (iconv_t)-1) {
-			ERROR_WITH_ERRNO("Failed to get conversion descriptor "
-					 "for converting UTF-16LE to UTF-8");
-			if (errno == ENOMEM)
-				return WIMLIB_ERR_NOMEM;
-			else
-				return WIMLIB_ERR_ICONV_NOT_AVAILABLE;
+/*
+ * A table that maps from UCS-2 characters to their upper case equivalents.
+ * Index and array values are both CPU endian.
+ * Note: this is only an *approximation* of real UTF-16 case folding.
+ */
+u16 upcase[65536];
+
+void
+init_upcase(void)
+{
+	/* This is the table used in NTFS volumes formatted by Windows 10.
+	 * It was compressed by tools/compress_upcase_table.c.  */
+	static const u16 upcase_compressed[] = {
+		0x0000, 0x0000, 0x0060, 0x0000, 0x0000, 0xffe0, 0x0019, 0x0061,
+		0x0061, 0x0000, 0x001b, 0x005d, 0x0008, 0x0060, 0x0000, 0x0079,
+		0x0000, 0x0000, 0x0000, 0xffff, 0x002f, 0x0100, 0x0002, 0x0000,
+		0x0007, 0x012b, 0x0011, 0x0121, 0x002f, 0x0103, 0x0006, 0x0101,
+		0x0000, 0x00c3, 0x0006, 0x0131, 0x0007, 0x012e, 0x0004, 0x0000,
+		0x0003, 0x012f, 0x0000, 0x0061, 0x0004, 0x0130, 0x0000, 0x00a3,
+		0x0003, 0x0000, 0x0000, 0x0082, 0x000b, 0x0131, 0x0006, 0x0189,
+		0x0008, 0x012f, 0x0007, 0x012e, 0x0000, 0x0038, 0x0006, 0x0000,
+		0x0000, 0xfffe, 0x0007, 0x01c4, 0x000f, 0x0101, 0x0000, 0xffb1,
+		0x0015, 0x011e, 0x0004, 0x01cc, 0x002a, 0x0149, 0x0014, 0x0149,
+		0x0007, 0x0000, 0x0009, 0x018c, 0x000b, 0x0138, 0x0000, 0x2a1f,
+		0x0000, 0x2a1c, 0x0000, 0x0000, 0x0000, 0xff2e, 0x0000, 0xff32,
+		0x0000, 0x0000, 0x0000, 0xff33, 0x0000, 0xff33, 0x0000, 0x0000,
+		0x0000, 0xff36, 0x0000, 0x0000, 0x0000, 0xff35, 0x0004, 0x0000,
+		0x0002, 0x0257, 0x0000, 0x0000, 0x0000, 0xff31, 0x0004, 0x0000,
+		0x0000, 0xff2f, 0x0000, 0xff2d, 0x0000, 0x0000, 0x0000, 0x29f7,
+		0x0003, 0x0000, 0x0002, 0x0269, 0x0000, 0x29fd, 0x0000, 0xff2b,
+		0x0002, 0x0000, 0x0000, 0xff2a, 0x0007, 0x0000, 0x0000, 0x29e7,
+		0x0002, 0x0000, 0x0000, 0xff26, 0x0005, 0x027e, 0x0003, 0x027e,
+		0x0000, 0xffbb, 0x0000, 0xff27, 0x0000, 0xff27, 0x0000, 0xffb9,
+		0x0005, 0x0000, 0x0000, 0xff25, 0x0065, 0x007b, 0x0079, 0x0293,
+		0x0008, 0x012d, 0x0003, 0x019c, 0x0002, 0x037b, 0x002e, 0x0000,
+		0x0000, 0xffda, 0x0000, 0xffdb, 0x0002, 0x03ad, 0x0012, 0x0060,
+		0x000a, 0x0060, 0x0000, 0xffc0, 0x0000, 0xffc1, 0x0000, 0xffc1,
+		0x0008, 0x0000, 0x0000, 0xfff8, 0x001a, 0x0118, 0x0000, 0x0007,
+		0x0008, 0x018d, 0x0009, 0x0233, 0x0046, 0x0035, 0x0006, 0x0061,
+		0x0000, 0xffb0, 0x000f, 0x0450, 0x0025, 0x010e, 0x000a, 0x036b,
+		0x0032, 0x048b, 0x000e, 0x0100, 0x0000, 0xfff1, 0x0037, 0x048a,
+		0x0026, 0x0465, 0x0034, 0x0000, 0x0000, 0xffd0, 0x0025, 0x0561,
+		0x00de, 0x0293, 0x1714, 0x0587, 0x0000, 0x8a04, 0x0003, 0x0000,
+		0x0000, 0x0ee6, 0x0087, 0x02ee, 0x0092, 0x1e01, 0x0069, 0x1df7,
+		0x0000, 0x0008, 0x0007, 0x1f00, 0x0008, 0x0000, 0x000e, 0x1f02,
+		0x0008, 0x1f0e, 0x0010, 0x1f06, 0x001a, 0x1f06, 0x0002, 0x1f0f,
+		0x0007, 0x1f50, 0x0017, 0x1f19, 0x0000, 0x004a, 0x0000, 0x004a,
+		0x0000, 0x0056, 0x0003, 0x1f72, 0x0000, 0x0064, 0x0000, 0x0064,
+		0x0000, 0x0080, 0x0000, 0x0080, 0x0000, 0x0070, 0x0000, 0x0070,
+		0x0000, 0x007e, 0x0000, 0x007e, 0x0028, 0x1f1e, 0x000c, 0x1f06,
+		0x0000, 0x0000, 0x0000, 0x0009, 0x000f, 0x0000, 0x000d, 0x1fb3,
+		0x000d, 0x1f44, 0x0008, 0x1fcd, 0x0006, 0x03f2, 0x0015, 0x1fbb,
+		0x014e, 0x0587, 0x0000, 0xffe4, 0x0021, 0x0000, 0x0000, 0xfff0,
+		0x000f, 0x2170, 0x000a, 0x0238, 0x0346, 0x0587, 0x0000, 0xffe6,
+		0x0019, 0x24d0, 0x0746, 0x0587, 0x0026, 0x0561, 0x000b, 0x057e,
+		0x0004, 0x012f, 0x0000, 0xd5d5, 0x0000, 0xd5d8, 0x000c, 0x022e,
+		0x000e, 0x03f8, 0x006e, 0x1e33, 0x0011, 0x0000, 0x0000, 0xe3a0,
+		0x0025, 0x2d00, 0x17f2, 0x0587, 0x6129, 0x2d26, 0x002e, 0x0201,
+		0x002a, 0x1def, 0x0098, 0xa5b7, 0x0040, 0x1dff, 0x000e, 0x0368,
+		0x000d, 0x022b, 0x034c, 0x2184, 0x5469, 0x2d26, 0x007f, 0x0061,
+		0x0040, 0x0000,
+	};
+
+	/* Simple LZ decoder  */
+	const u16 *in_next = upcase_compressed;
+	for (u32 i = 0; i < ARRAY_LEN(upcase); ) {
+		u16 length = *in_next++;
+		u16 src_pos = *in_next++;
+		if (length == 0) {
+			/* Literal */
+			upcase[i++] = src_pos;
+		} else {
+			/* Match */
+			do {
+				upcase[i++] = upcase[src_pos++];
+			} while (--length);
 		}
 	}
-	ret = utf16_to_utf8_size((const u16*)utf16_str, utf16_nbytes / 2);
-	if (ret >= 0) {
-		size_t utf8_expected_nbytes;
-		char  *utf8_str;
-		size_t utf8_bytes_left;
-		size_t utf16_bytes_left;
-		size_t num_chars_converted;
-		char  *utf8_str_save;
-		const char *utf16_str_save;
-
-		utf8_expected_nbytes = ret;
- 		utf8_str = MALLOC(utf8_expected_nbytes + 1);
-		if (utf8_str) {
-			utf8_bytes_left = utf8_expected_nbytes;
-			utf16_bytes_left = utf16_nbytes;
-			utf8_str_save = utf8_str;
-			utf16_str_save = utf16_str;
-			num_chars_converted = iconv(cd_utf16_to_utf8,
-						    (char**)&utf16_str,
-						    &utf16_bytes_left,
-						    &utf8_str,
-						    &utf8_bytes_left);
-			utf8_str = utf8_str_save;
-			utf16_str = utf16_str_save;
-			if (utf16_bytes_left == 0 &&
-			    utf8_bytes_left == 0 &&
-			    num_chars_converted != (size_t)(-1))
-			{
-				utf8_str[utf8_expected_nbytes] = '\0';
-				*utf8_str_ret = utf8_str;
-				*utf8_nbytes_ret = utf8_expected_nbytes;
-				ret = 0;
-			} else {
-				FREE(utf8_str);
-				ret = WIMLIB_ERR_INVALID_UTF16_STRING;
-			}
-		} else
-			ret = WIMLIB_ERR_NOMEM;
-	} else
-		ret = WIMLIB_ERR_INVALID_UTF16_STRING;
-#endif /* WITH_NTFS_3G */
-
-#ifdef ENABLE_ERROR_MESSAGES
-	if (ret != 0) {
-		ERROR_WITH_ERRNO("Error converting UTF-16LE string to UTF-8");
-		ERROR("The failing string was:");
-		print_string(utf16_str, utf16_nbytes);
-		putchar('\n');
-	}
-#endif /* ENABLE_ERROR_MESSAGES */
-	return ret;
-}
 
+	/* Delta filter  */
+	for (u32 i = 0; i < ARRAY_LEN(upcase); i++)
+		upcase[i] += i;
+}
 
-/* Converts a string in the UTF-8 encoding to a newly allocated string in the
- * UTF-16 encoding.
+/*
+ * Compare UTF-16LE strings case-sensitively (%ignore_case == false) or
+ * case-insensitively (%ignore_case == true).
  *
- * If available, do so by calling a similar function from libntfs-3g.
- * Otherwise, use iconv() along with the helper function utf8_to_utf16_size().
+ * This is implemented using the default upper-case table used by NTFS.  It does
+ * not handle all possible cases allowed by UTF-16LE.  For example, different
+ * normalizations of the same sequence of "characters" are not considered equal.
+ * It hopefully does the right thing most of the time though.
  */
-int utf8_to_utf16(const char *utf8_str, size_t utf8_nbytes,
-		  char **utf16_str_ret, size_t *utf16_nbytes_ret)
+int
+cmp_utf16le_strings(const utf16lechar *s1, size_t n1,
+		    const utf16lechar *s2, size_t n2,
+		    bool ignore_case)
 {
-	int ret;
-	if (utf8_nbytes == 0) {
-		*utf16_str_ret = NULL;
-		*utf16_nbytes_ret = 0;
-		return 0;
-	}
-#ifdef WITH_NTFS_3G
-	char *outs = NULL;
-	int outs_nchars = ntfs_mbstoucs(utf8_str, (ntfschar**)&outs);
-	if (outs_nchars >= 0) {
-		*utf16_str_ret = outs;
-		*utf16_nbytes_ret = (size_t)outs_nchars * 2;
-		ret = 0;
+	size_t n = min(n1, n2);
+
+	if (ignore_case) {
+		for (size_t i = 0; i < n; i++) {
+			u16 c1 = upcase[le16_to_cpu(s1[i])];
+			u16 c2 = upcase[le16_to_cpu(s2[i])];
+			if (c1 != c2)
+				return (c1 < c2) ? -1 : 1;
+		}
 	} else {
-		if (errno == ENOMEM)
-			ret = WIMLIB_ERR_NOMEM;
-		else
-			ret = WIMLIB_ERR_INVALID_UTF8_STRING;
+		for (size_t i = 0; i < n; i++) {
+			u16 c1 = le16_to_cpu(s1[i]);
+			u16 c2 = le16_to_cpu(s2[i]);
+			if (c1 != c2)
+				return (c1 < c2) ? -1 : 1;
+		}
 	}
-#else /* WITH_NTFS_3G */
-	static iconv_t cd_utf8_to_utf16 = (iconv_t)(-1);
-	if (cd_utf8_to_utf16 == (iconv_t)(-1)) {
-		cd_utf8_to_utf16 = iconv_open("UTF-16LE", "UTF-8");
-		if (cd_utf8_to_utf16 == (iconv_t)-1) {
-			ERROR_WITH_ERRNO("Failed to get conversion descriptor "
-					 "for converting UTF-8 to UTF-16LE");
-			if (errno == ENOMEM)
-				return WIMLIB_ERR_NOMEM;
-			else
-				return WIMLIB_ERR_ICONV_NOT_AVAILABLE;
+	if (n1 == n2)
+		return 0;
+	return (n1 < n2) ? -1 : 1;
+}
+
+/* Like cmp_utf16le_strings(), but assumes the strings are null terminated.  */
+int
+cmp_utf16le_strings_z(const utf16lechar *s1, const utf16lechar *s2,
+		      bool ignore_case)
+{
+	if (ignore_case) {
+		for (;;) {
+			u16 c1 = upcase[le16_to_cpu(*s1)];
+			u16 c2 = upcase[le16_to_cpu(*s2)];
+			if (c1 != c2)
+				return (c1 < c2) ? -1 : 1;
+			if (c1 == 0)
+				return 0;
+			s1++, s2++;
 		}
+	} else {
+		while (*s1 && *s1 == *s2)
+			s1++, s2++;
+		if (*s1 == *s2)
+			return 0;
+		return (le16_to_cpu(*s1) < le16_to_cpu(*s2)) ? -1 : 1;
 	}
+}
 
-	ret = utf8_to_utf16_size(utf8_str);
-	if (ret >= 0) {
-		size_t utf16_expected_nbytes;
-		char  *utf16_str;
-		size_t utf16_bytes_left;
-		size_t utf8_bytes_left;
-		size_t num_chars_converted;
-		const char *utf8_str_save;
-		char  *utf16_str_save;
-
-		utf16_expected_nbytes = (size_t)ret * 2;
- 		utf16_str = MALLOC(utf16_expected_nbytes + 2);
-		if (utf16_str) {
-			utf16_bytes_left = utf16_expected_nbytes;
-			utf8_bytes_left = utf8_nbytes;
-			utf8_str_save = utf8_str;
-			utf16_str_save = utf16_str;
-			num_chars_converted = iconv(cd_utf8_to_utf16,
-						    (char**)&utf8_str,
-						    &utf8_bytes_left,
-						    &utf16_str,
-						    &utf16_bytes_left);
-			utf8_str = utf8_str_save;
-			utf16_str = utf16_str_save;
-			if (utf16_bytes_left == 0 &&
-			    utf8_bytes_left == 0 &&
-			    num_chars_converted != (size_t)(-1))
-			{
-				utf16_str[utf16_expected_nbytes] = '\0';
-				utf16_str[utf16_expected_nbytes + 1] = '\0';
-				*utf16_str_ret = utf16_str;
-				*utf16_nbytes_ret = utf16_expected_nbytes;
-				ret = 0;
-			} else {
-				FREE(utf16_str);
-				ret = WIMLIB_ERR_INVALID_UTF8_STRING;
-			}
-		} else
-			ret = WIMLIB_ERR_NOMEM;
-	} else
-		ret = WIMLIB_ERR_INVALID_UTF8_STRING;
-#endif /* WITH_NTFS_3G */
-
-#ifdef ENABLE_ERROR_MESSAGES
-	if (ret != 0) {
-		ERROR_WITH_ERRNO("Error converting UTF-8 string to UTF-16LE");
-		ERROR("The failing string was:");
-		print_string(utf8_str, utf8_nbytes);
-		putchar('\n');
-		ERROR("Length: %zu bytes", utf8_nbytes);
+/* Duplicate a UTF-16 string.  The input string might not be null terminated and
+ * might be misaligned, but the returned string is guaranteed to be null
+ * terminated and properly aligned.  */
+utf16lechar *
+utf16le_dupz(const void *s, size_t size)
+{
+	utf16lechar *dup = MALLOC(size + sizeof(utf16lechar));
+	if (dup) {
+		memcpy(dup, s, size);
+		dup[size / sizeof(utf16lechar)] = 0;
 	}
-#endif /* ENABLE_ERROR_MESSAGES */
-	return ret;
+	return dup;
+}
+
+/* Duplicate a null-terminated UTF-16 string.  */
+utf16lechar *
+utf16le_dup(const utf16lechar *s)
+{
+	return memdup(s, utf16le_len_bytes(s) + sizeof(utf16lechar));
+}
+
+/* Return the length, in bytes, of a null terminated UTF-16 string, excluding
+ * the null terminator.  */
+size_t
+utf16le_len_bytes(const utf16lechar *s)
+{
+	const utf16lechar *p = s;
+	while (*p)
+		p++;
+	return (p - s) * sizeof(utf16lechar);
+}
+
+/* Return the length, in UTF-16 coding units, of a null terminated UTF-16
+ * string, excluding the null terminator.  */
+size_t
+utf16le_len_chars(const utf16lechar *s)
+{
+	return utf16le_len_bytes(s) / sizeof(utf16lechar);
 }