charset-iconv.c

/* Copyright (c) 2002-2018 Dovecot authors, see the included COPYING file */

#include "lib.h"
#include "buffer.h"
#include "charset-utf8-private.h"

#ifdef HAVE_ICONV

#include <iconv.h>
#include <ctype.h>

struct charset_translation {
	iconv_t cd;
	normalizer_func_t *normalizer;
};

static int
iconv_charset_to_utf8_begin(const char *charset, normalizer_func_t *normalizer,
			    struct charset_translation **t_r)
{
	struct charset_translation *t;
	iconv_t cd;

	if (charset_is_utf8(charset))
		cd = (iconv_t)-1;
	else {
		if (strcmp(charset, "UTF-8//TEST") == 0)
			charset = "UTF-8";
		cd = iconv_open("UTF-8", charset);
		if (cd == (iconv_t)-1)
			return -1;
	}

	t = i_new(struct charset_translation, 1);
	t->cd = cd;
	t->normalizer = normalizer;
	*t_r = t;
	return 0;
}

static void iconv_charset_to_utf8_end(struct charset_translation *t)
{
	if (t->cd != (iconv_t)-1)
		iconv_close(t->cd);
	i_free(t);
}

static void iconv_charset_to_utf8_reset(struct charset_translation *t)
{
	if (t->cd != (iconv_t)-1)
		(void)iconv(t->cd, NULL, NULL, NULL, NULL);
}

static bool
charset_to_utf8_try(struct charset_translation *t,
		    const unsigned char *src, size_t *src_size, buffer_t *dest,
		    enum charset_result *result)
{
	ICONV_CONST char *ic_srcbuf;
	char tmpbuf[8192], *ic_destbuf;
	size_t srcleft, destleft, tmpbuf_used;
	bool ret = TRUE;

	if (t->cd == (iconv_t)-1) {
		/* input is already supposed to be UTF-8 */
		*result = charset_utf8_to_utf8(t->normalizer, src, src_size, dest);
		return TRUE;
	}
	destleft = sizeof(tmpbuf);
	ic_destbuf = tmpbuf;
	srcleft = *src_size;
	ic_srcbuf = (ICONV_CONST char *) src;

	if (iconv(t->cd, &ic_srcbuf, &srcleft,
		  &ic_destbuf, &destleft) != (size_t)-1) {
		i_assert(srcleft == 0);
		*result = CHARSET_RET_OK;
	} else if (errno == E2BIG) {
		/* set result just to avoid compiler warning */
		*result = CHARSET_RET_INCOMPLETE_INPUT;
		ret = FALSE;
	} else if (errno == EINVAL) {
		i_assert(srcleft <= CHARSET_MAX_PENDING_BUF_SIZE);
		*result = CHARSET_RET_INCOMPLETE_INPUT;
	} else {
		/* should be EILSEQ */
		*result = CHARSET_RET_INVALID_INPUT;
		ret = FALSE;
	}
	*src_size -= srcleft;

	/* we just converted data to UTF-8. it shouldn't be invalid, but
	   Solaris iconv appears to pass invalid data through sometimes
	   (e.g. 8 bit characters with UTF-7) */
	tmpbuf_used = sizeof(tmpbuf) - destleft;
	if (charset_utf8_to_utf8(t->normalizer, (void *)tmpbuf,
				 &tmpbuf_used, dest) != CHARSET_RET_OK)
		*result = CHARSET_RET_INVALID_INPUT;
	return ret;
}

static enum charset_result
iconv_charset_to_utf8(struct charset_translation *t,
		      const unsigned char *src, size_t *src_size,
		      buffer_t *dest)
{
	enum charset_result result;
	size_t pos, size;
	size_t prev_invalid_pos = (size_t)-1;
	bool ret;

	for (pos = 0;;) {
		i_assert(pos <= *src_size);
		size = *src_size - pos;
		ret = charset_to_utf8_try(t, src + pos, &size, dest, &result);
		pos += size;

		if (ret)
			break;

		if (result == CHARSET_RET_INVALID_INPUT) {
			if (prev_invalid_pos != dest->used) {
				buffer_append(dest, UNICODE_REPLACEMENT_CHAR_UTF8,
					      strlen(UNICODE_REPLACEMENT_CHAR_UTF8));
				prev_invalid_pos = dest->used;
			}
			if (pos < *src_size)
				pos++;
		}
	}

	if (prev_invalid_pos != (size_t)-1)
		result = CHARSET_RET_INVALID_INPUT;

	i_assert(*src_size - pos <= CHARSET_MAX_PENDING_BUF_SIZE);
	*src_size = pos;
	return result;
}

const struct charset_utf8_vfuncs charset_iconv = {
	.to_utf8_begin = iconv_charset_to_utf8_begin,
	.to_utf8_end = iconv_charset_to_utf8_end,
	.to_utf8_reset = iconv_charset_to_utf8_reset,
	.to_utf8 = iconv_charset_to_utf8,
};

#endif