crypto/pkcs12: add UTF8 support.

Reviewed-by: Richard Levitte <levitte@openssl.org>

3 files changed, 174 insertions, 1 deletions

diff --git a/crypto/pkcs12/p12_key.c b/crypto/pkcs12/p12_key.c
index 4f1d29bb6d..9c13a451e0 100644
--- a/crypto/pkcs12/p12_key.c
+++ b/crypto/pkcs12/p12_key.c
@@ -50,6 +50,29 @@ int PKCS12_key_gen_asc(const char *pass, int passlen, unsigned char *salt,
     return ret;
 }
 
+int PKCS12_key_gen_utf8(const char *pass, int passlen, unsigned char *salt,
+                        int saltlen, int id, int iter, int n,
+                        unsigned char *out, const EVP_MD *md_type)
+{
+    int ret;
+    unsigned char *unipass;
+    int uniplen;
+
+    if (!pass) {
+        unipass = NULL;
+        uniplen = 0;
+    } else if (!OPENSSL_utf82uni(pass, passlen, &unipass, &uniplen)) {
+        PKCS12err(PKCS12_F_PKCS12_KEY_GEN_UTF8, ERR_R_MALLOC_FAILURE);
+        return 0;
+    }
+    ret = PKCS12_key_gen_uni(unipass, uniplen, salt, saltlen,
+                             id, iter, n, out, md_type);
+    if (ret <= 0)
+        return 0;
+    OPENSSL_clear_free(unipass, uniplen);
+    return ret;
+}
+
 int PKCS12_key_gen_uni(unsigned char *pass, int passlen, unsigned char *salt,
                        int saltlen, int id, int iter, int n,
                        unsigned char *out, const EVP_MD *md_type)
diff --git a/crypto/pkcs12/p12_utl.c b/crypto/pkcs12/p12_utl.c
index c4feb90b9e..7d471f5c0d 100644
--- a/crypto/pkcs12/p12_utl.c
+++ b/crypto/pkcs12/p12_utl.c
@@ -38,7 +38,7 @@ unsigned char *OPENSSL_asc2uni(const char *asc, int asclen,
     return unitmp;
 }
 
-char *OPENSSL_uni2asc(unsigned char *uni, int unilen)
+char *OPENSSL_uni2asc(const unsigned char *uni, int unilen)
 {
     int asclen, i;
     char *asctmp;
@@ -58,6 +58,155 @@ char *OPENSSL_uni2asc(unsigned char *uni, int unilen)
     return asctmp;
 }
 
+/*
+ * OPENSSL_{utf82uni|uni2utf8} perform conversion between UTF-8 and
+ * PKCS#12 BMPString format, which is specified as big-endian UTF-16.
+ * One should keep in mind that even though BMPString is passed as
+ * unsigned char *, it's not the kind of string you can exercise e.g.
+ * strlen on. Caller also has to keep in mind that its length is
+ * expressed not in number of UTF-16 characters, but in number of
+ * bytes the string occupies, and treat it, the length, accordingly.
+ */
+unsigned char *OPENSSL_utf82uni(const char *asc, int asclen,
+                                unsigned char **uni, int *unilen)
+{
+    int ulen, i, j;
+    unsigned char *unitmp, *ret;
+    unsigned long utf32chr = 0;
+
+    if (asclen == -1)
+        asclen = strlen(asc);
+
+    for (ulen = 0, i = 0; i < asclen; i += j) {
+        j = UTF8_getc((const unsigned char *)asc+i, asclen-i, &utf32chr);
+
+        /*
+         * Following condition is somewhat opportunistic is sense that
+         * decoding failure is used as *indirect* indication that input
+         * string might in fact be extended ASCII/ANSI/ISO-8859-X. The
+         * fallback is taken in hope that it would allow to process
+         * files created with previous OpenSSL version, which used the
+         * naive OPENSSL_asc2uni all along. It might be worth noting
+         * that probability of false positive depends on language. In
+         * cases covered by ISO Latin 1 probability is very low, because
+         * any printable non-ASCII alphabet letter followed by another
+         * or any ASCII character will trigger failure and fallback.
+         * In other cases situation can be intensified by the fact that
+         * English letters are not part of alternative keyboard layout,
+         * but even then there should be plenty of pairs that trigger
+         * decoding failure...
+         */
+        if (j < 0)
+	    return OPENSSL_asc2uni(asc, asclen, uni, unilen);
+
+        if (utf32chr > 0x10FFFF)        /* UTF-16 cap */
+	    return NULL;
+
+        if (utf32chr >= 0x10000)        /* pair of UTF-16 characters */
+            ulen += 2*2;
+        else                            /* or just one */
+            ulen += 2;
+    }
+
+    ulen += 2;  /* for trailing UTF16 zero */
+
+    if ((ret = OPENSSL_malloc(ulen)) == NULL)
+        return NULL;
+
+    /* re-run the loop writing down UTF-16 characters in big-endian order */
+    for (unitmp = ret, i = 0; i < asclen; i += j) {
+        j = UTF8_getc((const unsigned char *)asc+i, asclen-i, &utf32chr);
+        if (utf32chr >= 0x10000) {      /* pair if UTF-16 characters */
+            unsigned int hi, lo;
+
+            utf32chr -= 0x10000;
+            hi = 0xD800 + (utf32chr>>10);
+            lo = 0xDC00 + (utf32chr&0x3ff);
+            *unitmp++ = (unsigned char)(hi>>8);
+            *unitmp++ = (unsigned char)(hi);
+            *unitmp++ = (unsigned char)(lo>>8);
+            *unitmp++ = (unsigned char)(lo);
+        } else {                        /* or just one */
+            *unitmp++ = (unsigned char)(utf32chr>>8);
+            *unitmp++ = (unsigned char)(utf32chr);
+        }
+    }
+    /* Make result double null terminated */
+    *unitmp++ = 0;
+    *unitmp++ = 0;
+    if (unilen)
+        *unilen = ulen;
+    if (uni)
+        *uni = ret;
+    return ret;
+}
+
+static int bmp_to_utf8(char *str, const unsigned char *utf16, int len)
+{
+    unsigned long utf32chr;
+
+    if (len == 0) return 0;
+
+    if (len < 2) return -1;
+
+    /* pull UTF-16 character in big-endian order */
+    utf32chr = (utf16[0]<<8) | utf16[1];
+
+    if (utf32chr >= 0xD800 && utf32chr < 0xE000) {   /* two chars */
+        unsigned int lo;
+
+        if (len < 4) return -1;
+
+        utf32chr -= 0xD800;
+        utf32chr <<= 10;
+        lo = (utf16[2]<<8) | utf16[3];
+        if (lo < 0xDC00 || lo >= 0xE000) return -1;
+        utf32chr |= lo-0xDC00;
+        utf32chr += 0x10000;
+    }
+
+    return UTF8_putc((unsigned char *)str, len > 4 ? 4 : len, utf32chr);
+}
+
+char *OPENSSL_uni2utf8(const unsigned char *uni, int unilen)
+{
+    int asclen, i, j;
+    char *asctmp;
+
+    for (asclen = 0, i = 0; i < unilen; ) {
+        j = bmp_to_utf8(NULL, uni+i, unilen-i);
+        /*
+         * falling back to OPENSSL_uni2asc makes lesser sense, it's
+         * done rather to maintain symmetry...
+         */
+        if (j < 0) return OPENSSL_uni2asc(uni, unilen);
+        if (j == 4) i += 4;
+        else        i += 2;
+        asclen += j;
+    }
+
+    /* If no terminating zero allow for one */
+    if (!unilen || (uni[unilen-2]||uni[unilen - 1]))
+        asclen++;
+
+    if ((asctmp = OPENSSL_malloc(asclen)) == NULL)
+        return NULL;
+
+    /* re-run the loop emitting UTF-8 string */
+    for (asclen = 0, i = 0; i < unilen; ) {
+        j = bmp_to_utf8(asctmp+asclen, uni+i, unilen-i);
+        if (j == 4) i += 4;
+        else        i += 2;
+        asclen += j;
+    }
+
+    /* If no terminating zero write one */
+    if (!unilen || (uni[unilen-2]||uni[unilen - 1]))
+        asctmp[asclen] = '\0';
+
+    return asctmp;
+}
+
 int i2d_PKCS12_bio(BIO *bp, PKCS12 *p12)
 {
     return ASN1_item_i2d_bio(ASN1_ITEM_rptr(PKCS12), bp, p12);
diff --git a/crypto/pkcs12/pk12err.c b/crypto/pkcs12/pk12err.c
index f15a6951a2..f705084a2a 100644
--- a/crypto/pkcs12/pk12err.c
+++ b/crypto/pkcs12/pk12err.c
@@ -27,6 +27,7 @@ static ERR_STRING_DATA PKCS12_str_functs[] = {
     {ERR_FUNC(PKCS12_F_PKCS12_ITEM_PACK_SAFEBAG), "PKCS12_item_pack_safebag"},
     {ERR_FUNC(PKCS12_F_PKCS12_KEY_GEN_ASC), "PKCS12_key_gen_asc"},
     {ERR_FUNC(PKCS12_F_PKCS12_KEY_GEN_UNI), "PKCS12_key_gen_uni"},
+    {ERR_FUNC(PKCS12_F_PKCS12_KEY_GEN_UTF8), "PKCS12_key_gen_utf8"},
     {ERR_FUNC(PKCS12_F_PKCS12_NEWPASS), "PKCS12_newpass"},
     {ERR_FUNC(PKCS12_F_PKCS12_PACK_P7DATA), "PKCS12_pack_p7data"},
     {ERR_FUNC(PKCS12_F_PKCS12_PACK_P7ENCDATA), "PKCS12_pack_p7encdata"},