Index: net/base/x509_cert_types_mac.cc |
=================================================================== |
--- net/base/x509_cert_types_mac.cc (revision 42829) |
+++ net/base/x509_cert_types_mac.cc (working copy) |
@@ -1,293 +0,0 @@ |
-// Copyright (c) 2006-2010 The Chromium Authors. All rights reserved. |
-// Use of this source code is governed by a BSD-style license that can be |
-// found in the LICENSE file. |
- |
-#include "net/base/x509_cert_types.h" |
- |
-#include <CoreServices/CoreServices.h> |
-#include <Security/Security.h> |
-#include <Security/SecAsn1Coder.h> |
- |
-#include "base/logging.h" |
-#include "base/i18n/icu_string_conversions.h" |
-#include "base/utf_string_conversions.h" |
- |
-namespace net { |
- |
-static const CSSM_OID* kOIDs[] = { |
- &CSSMOID_CommonName, |
- &CSSMOID_LocalityName, |
- &CSSMOID_StateProvinceName, |
- &CSSMOID_CountryName, |
- &CSSMOID_StreetAddress, |
- &CSSMOID_OrganizationName, |
- &CSSMOID_OrganizationalUnitName, |
- &CSSMOID_DNQualifier // This should be "DC" but is undoubtedly wrong. |
-}; // TODO(avi): Find the right OID. |
- |
-// Converts raw CSSM_DATA to a std::string. (Char encoding is unaltered.) |
-static std::string DataToString(CSSM_DATA data); |
- |
-// Converts raw CSSM_DATA in ISO-8859-1 to a std::string in UTF-8. |
-static std::string Latin1DataToUTF8String(CSSM_DATA data); |
- |
-// Converts big-endian UTF-16 to UTF-8 in a std::string. |
-// Note: The byte-order flipping is done in place on the input buffer! |
-static bool UTF16BigEndianToUTF8(char16* chars, size_t length, |
- std::string* out_string); |
- |
-// Converts big-endian UTF-32 to UTF-8 in a std::string. |
-// Note: The byte-order flipping is done in place on the input buffer! |
-static bool UTF32BigEndianToUTF8(char32* chars, size_t length, |
- std::string* out_string); |
- |
-// Adds a type+value pair to the appropriate vector from a C array. |
-// The array is keyed by the matching OIDs from kOIDS[]. |
- static void AddTypeValuePair(const CSSM_OID type, |
- const std::string& value, |
- std::vector<std::string>* values[]); |
- |
-// Stores the first string of the vector, if any, to *single_value. |
-static void SetSingle(const std::vector<std::string> &values, |
- std::string* single_value); |
- |
- |
-void CertPrincipal::Parse(const CSSM_X509_NAME* name) { |
- std::vector<std::string> common_names, locality_names, state_names, |
- country_names; |
- |
- std::vector<std::string>* values[] = { |
- &common_names, &locality_names, |
- &state_names, &country_names, |
- &(this->street_addresses), |
- &(this->organization_names), |
- &(this->organization_unit_names), |
- &(this->domain_components) |
- }; |
- DCHECK(arraysize(kOIDs) == arraysize(values)); |
- |
- for (size_t rdn = 0; rdn < name->numberOfRDNs; ++rdn) { |
- CSSM_X509_RDN rdn_struct = name->RelativeDistinguishedName[rdn]; |
- for (size_t pair = 0; pair < rdn_struct.numberOfPairs; ++pair) { |
- CSSM_X509_TYPE_VALUE_PAIR pair_struct = |
- rdn_struct.AttributeTypeAndValue[pair]; |
- AddTypeValuePair(pair_struct.type, |
- DataToString(pair_struct.value), |
- values); |
- } |
- } |
- |
- SetSingle(common_names, &this->common_name); |
- SetSingle(locality_names, &this->locality_name); |
- SetSingle(state_names, &this->state_or_province_name); |
- SetSingle(country_names, &this->country_name); |
-} |
- |
- |
-// The following structs and templates work with Apple's very arcane and under- |
-// documented SecAsn1Parser API, which is apparently the same as NSS's ASN.1 |
-// decoder: |
-// http://www.mozilla.org/projects/security/pki/nss/tech-notes/tn1.html |
- |
-// These are used to parse the contents of a raw |
-// BER DistinguishedName structure. |
- |
-struct KeyValuePair { |
- CSSM_OID key; |
- int value_type; |
- CSSM_DATA value; |
- |
- enum { |
- kTypeOther = 0, |
- kTypePrintableString, |
- kTypeIA5String, |
- kTypeT61String, |
- kTypeUTF8String, |
- kTypeBMPString, |
- kTypeUniversalString, |
- }; |
-}; |
- |
-static const SecAsn1Template kStringValueTemplate[] = { |
- { SEC_ASN1_CHOICE, offsetof(KeyValuePair, value_type), }, |
- { SEC_ASN1_PRINTABLE_STRING, |
- offsetof(KeyValuePair, value), 0, KeyValuePair::kTypePrintableString }, |
- { SEC_ASN1_IA5_STRING, |
- offsetof(KeyValuePair, value), 0, KeyValuePair::kTypeIA5String }, |
- { SEC_ASN1_T61_STRING, |
- offsetof(KeyValuePair, value), 0, KeyValuePair::kTypeT61String }, |
- { SEC_ASN1_UTF8_STRING, |
- offsetof(KeyValuePair, value), 0, KeyValuePair::kTypeUTF8String }, |
- { SEC_ASN1_BMP_STRING, |
- offsetof(KeyValuePair, value), 0, KeyValuePair::kTypeBMPString }, |
- { SEC_ASN1_UNIVERSAL_STRING, |
- offsetof(KeyValuePair, value), 0, KeyValuePair::kTypeUniversalString }, |
- { 0, } |
-}; |
- |
-static const SecAsn1Template kKeyValuePairTemplate[] = { |
- { SEC_ASN1_SEQUENCE, 0, NULL, sizeof(KeyValuePair) }, |
- { SEC_ASN1_OBJECT_ID, offsetof(KeyValuePair, key), }, |
- { SEC_ASN1_INLINE, 0, &kStringValueTemplate, }, |
- { 0, } |
-}; |
- |
-struct KeyValuePairs { |
- KeyValuePair* pairs; |
-}; |
- |
-static const SecAsn1Template kKeyValuePairSetTemplate[] = { |
- { SEC_ASN1_SET_OF, offsetof(KeyValuePairs,pairs), |
- kKeyValuePairTemplate, sizeof(KeyValuePairs) } |
-}; |
- |
-struct X509Name { |
- KeyValuePairs** pairs_list; |
-}; |
- |
-static const SecAsn1Template kNameTemplate[] = { |
- { SEC_ASN1_SEQUENCE_OF, offsetof(X509Name,pairs_list), |
- kKeyValuePairSetTemplate, sizeof(X509Name) } |
-}; |
- |
-bool CertPrincipal::ParseDistinguishedName(const void* ber_name_data, |
- size_t length) { |
- DCHECK(ber_name_data); |
- |
- // First parse the BER |name_data| into the above structs. |
- SecAsn1CoderRef coder = NULL; |
- SecAsn1CoderCreate(&coder); |
- DCHECK(coder); |
- X509Name* name = NULL; |
- OSStatus err = SecAsn1Decode(coder, ber_name_data, length, kNameTemplate, |
- &name); |
- if (err) { |
- LOG(ERROR) << "SecAsn1Decode returned " << err << "; name=" << name; |
- SecAsn1CoderRelease(coder); |
- return false; |
- } |
- |
- // Now scan the structs and add the values to my string vectors. |
- // I don't store multiple common/locality/state/country names, so use |
- // temporary vectors for those. |
- std::vector<std::string> common_names, locality_names, state_names, |
- country_names; |
- std::vector<std::string>* values[] = { |
- &common_names, &locality_names, |
- &state_names, &country_names, |
- &this->street_addresses, |
- &this->organization_names, |
- &this->organization_unit_names, |
- &this->domain_components |
- }; |
- DCHECK(arraysize(kOIDs) == arraysize(values)); |
- |
- for (int rdn=0; name[rdn].pairs_list; ++rdn) { |
- KeyValuePair *pair; |
- for (int pair_index = 0; |
- NULL != (pair = name[rdn].pairs_list[0][pair_index].pairs); |
- ++pair_index) { |
- switch (pair->value_type) { |
- case KeyValuePair::kTypeIA5String: // ASCII (that means 7-bit!) |
- case KeyValuePair::kTypePrintableString: // a subset of ASCII |
- case KeyValuePair::kTypeUTF8String: // UTF-8 |
- AddTypeValuePair(pair->key, DataToString(pair->value), values); |
- break; |
- case KeyValuePair::kTypeT61String: // T61, pretend it's Latin-1 |
- AddTypeValuePair(pair->key, |
- Latin1DataToUTF8String(pair->value), |
- values); |
- break; |
- case KeyValuePair::kTypeBMPString: { // UTF-16, big-endian |
- std::string value; |
- UTF16BigEndianToUTF8(reinterpret_cast<char16*>(pair->value.Data), |
- pair->value.Length / sizeof(char16), |
- &value); |
- AddTypeValuePair(pair->key, value, values); |
- break; |
- } |
- case KeyValuePair::kTypeUniversalString: { // UTF-32, big-endian |
- std::string value; |
- UTF32BigEndianToUTF8(reinterpret_cast<char32*>(pair->value.Data), |
- pair->value.Length / sizeof(char32), |
- &value); |
- AddTypeValuePair(pair->key, value, values); |
- break; |
- } |
- default: |
- DCHECK_EQ(pair->value_type, KeyValuePair::kTypeOther); |
- // We don't know what data type this is, but we'll store it as a blob. |
- // Displaying the string may not work, but at least it can be compared |
- // byte-for-byte by a Matches() call. |
- AddTypeValuePair(pair->key, DataToString(pair->value), values); |
- break; |
- } |
- } |
- } |
- |
- SetSingle(common_names, &this->common_name); |
- SetSingle(locality_names, &this->locality_name); |
- SetSingle(state_names, &this->state_or_province_name); |
- SetSingle(country_names, &this->country_name); |
- |
- // Releasing |coder| frees all the memory pointed to via |name|. |
- SecAsn1CoderRelease(coder); |
- return true; |
-} |
- |
- |
-// SUBROUTINES: |
- |
-static std::string DataToString(CSSM_DATA data) { |
- return std::string( |
- reinterpret_cast<std::string::value_type*>(data.Data), |
- data.Length); |
-} |
- |
-static std::string Latin1DataToUTF8String(CSSM_DATA data) { |
- string16 utf16; |
- if (!CodepageToUTF16(DataToString(data), base::kCodepageLatin1, |
- base::OnStringConversionError::FAIL, &utf16)) |
- return ""; |
- return UTF16ToUTF8(utf16); |
-} |
- |
-bool UTF16BigEndianToUTF8(char16* chars, size_t length, |
- std::string* out_string) { |
- for (size_t i = 0; i < length; i++) |
- chars[i] = EndianU16_BtoN(chars[i]); |
- return UTF16ToUTF8(chars, length, out_string); |
-} |
- |
-bool UTF32BigEndianToUTF8(char32* chars, size_t length, |
- std::string* out_string) { |
- for (size_t i = 0; i < length; i++) |
- chars[i] = EndianS32_BtoN(chars[i]); |
-#if defined(WCHAR_T_IS_UTF32) |
- return WideToUTF8(reinterpret_cast<const wchar_t*>(chars), |
- length, out_string); |
-#else |
-#error This code doesn't handle 16-bit wchar_t. |
-#endif |
-} |
- |
- static void AddTypeValuePair(const CSSM_OID type, |
- const std::string& value, |
- std::vector<std::string>* values[]) { |
- for (size_t oid = 0; oid < arraysize(kOIDs); ++oid) { |
- if (CSSMOIDEqual(&type, kOIDs[oid])) { |
- values[oid]->push_back(value); |
- break; |
- } |
- } |
-} |
- |
-static void SetSingle(const std::vector<std::string> &values, |
- std::string* single_value) { |
- // We don't expect to have more than one CN, L, S, and C. |
- LOG_IF(WARNING, values.size() > 1) << "Didn't expect multiple values"; |
- if (values.size() > 0) |
- *single_value = values[0]; |
-} |
- |
-} // namespace net |