|  | // Copyright 2010 The Chromium Authors | 
|  | // Use of this source code is governed by a BSD-style license that can be | 
|  | // found in the LICENSE file. | 
|  |  | 
|  | #include "pem.h" | 
|  | #include "string_util.h" | 
|  |  | 
|  | #include <string_view> | 
|  |  | 
|  | namespace { | 
|  |  | 
|  | constexpr std::string_view kPEMHeaderBeginBlock = "-----BEGIN "; | 
|  | constexpr std::string_view kPEMHeaderEndBlock = "-----END "; | 
|  | constexpr std::string_view kPEMHeaderTail = "-----"; | 
|  |  | 
|  | }  // namespace | 
|  |  | 
|  | BSSL_NAMESPACE_BEGIN | 
|  |  | 
|  |  | 
|  |  | 
|  | struct PEMTokenizer::PEMType { | 
|  | std::string type; | 
|  | std::string header; | 
|  | std::string footer; | 
|  | }; | 
|  |  | 
|  | PEMTokenizer::PEMTokenizer( | 
|  | std::string_view str, const std::vector<std::string> &allowed_block_types) { | 
|  | Init(str, allowed_block_types); | 
|  | } | 
|  |  | 
|  | PEMTokenizer::~PEMTokenizer() = default; | 
|  |  | 
|  | bool PEMTokenizer::GetNext() { | 
|  | while (pos_ != std::string_view::npos) { | 
|  | // Scan for the beginning of the next PEM encoded block. | 
|  | pos_ = str_.find(kPEMHeaderBeginBlock, pos_); | 
|  | if (pos_ == std::string_view::npos) { | 
|  | return false;  // No more PEM blocks | 
|  | } | 
|  |  | 
|  | std::vector<PEMType>::const_iterator it; | 
|  | // Check to see if it is of an acceptable block type. | 
|  | for (it = block_types_.begin(); it != block_types_.end(); ++it) { | 
|  | if (!bssl::string_util::StartsWith(str_.substr(pos_), it->header)) { | 
|  | continue; | 
|  | } | 
|  |  | 
|  | // Look for a footer matching the header. If none is found, then all | 
|  | // data following this point is invalid and should not be parsed. | 
|  | std::string_view::size_type footer_pos = str_.find(it->footer, pos_); | 
|  | if (footer_pos == std::string_view::npos) { | 
|  | pos_ = std::string_view::npos; | 
|  | return false; | 
|  | } | 
|  |  | 
|  | // Chop off the header and footer and parse the data in between. | 
|  | std::string_view::size_type data_begin = pos_ + it->header.size(); | 
|  | pos_ = footer_pos + it->footer.size(); | 
|  | block_type_ = it->type; | 
|  |  | 
|  | std::string_view encoded = | 
|  | str_.substr(data_begin, footer_pos - data_begin); | 
|  | if (!string_util::Base64Decode( | 
|  | string_util::CollapseWhitespaceASCII(encoded, true), &data_)) { | 
|  | // The most likely cause for a decode failure is a datatype that | 
|  | // includes PEM headers, which are not supported. | 
|  | break; | 
|  | } | 
|  |  | 
|  | return true; | 
|  | } | 
|  |  | 
|  | // If the block did not match any acceptable type, move past it and | 
|  | // continue the search. Otherwise, |pos_| has been updated to the most | 
|  | // appropriate search position to continue searching from and should not | 
|  | // be adjusted. | 
|  | if (it == block_types_.end()) { | 
|  | pos_ += kPEMHeaderBeginBlock.size(); | 
|  | } | 
|  | } | 
|  |  | 
|  | return false; | 
|  | } | 
|  |  | 
|  | void PEMTokenizer::Init(std::string_view str, | 
|  | const std::vector<std::string> &allowed_block_types) { | 
|  | str_ = str; | 
|  | pos_ = 0; | 
|  |  | 
|  | // Construct PEM header/footer strings for all the accepted types, to | 
|  | // reduce parsing later. | 
|  | for (const auto &allowed_block_type : allowed_block_types) { | 
|  | PEMType allowed_type; | 
|  | allowed_type.type = allowed_block_type; | 
|  | allowed_type.header = kPEMHeaderBeginBlock; | 
|  | allowed_type.header.append(allowed_block_type); | 
|  | allowed_type.header.append(kPEMHeaderTail); | 
|  | allowed_type.footer = kPEMHeaderEndBlock; | 
|  | allowed_type.footer.append(allowed_block_type); | 
|  | allowed_type.footer.append(kPEMHeaderTail); | 
|  | block_types_.push_back(allowed_type); | 
|  | } | 
|  | } | 
|  |  | 
|  | std::string PEMEncode(std::string_view data, const std::string &type) { | 
|  | std::string b64_encoded; | 
|  | string_util::Base64Encode(data, &b64_encoded); | 
|  |  | 
|  | // Divide the Base-64 encoded data into 64-character chunks, as per | 
|  | // 4.3.2.4 of RFC 1421. | 
|  | static const size_t kChunkSize = 64; | 
|  | size_t chunks = (b64_encoded.size() + (kChunkSize - 1)) / kChunkSize; | 
|  |  | 
|  | std::string pem_encoded; | 
|  | pem_encoded.reserve( | 
|  | // header & footer | 
|  | 17 + 15 + type.size() * 2 + | 
|  | // encoded data | 
|  | b64_encoded.size() + | 
|  | // newline characters for line wrapping in encoded data | 
|  | chunks); | 
|  |  | 
|  | pem_encoded = kPEMHeaderBeginBlock; | 
|  | pem_encoded.append(type); | 
|  | pem_encoded.append(kPEMHeaderTail); | 
|  | pem_encoded.append("\n"); | 
|  |  | 
|  | for (size_t i = 0, chunk_offset = 0; i < chunks; | 
|  | ++i, chunk_offset += kChunkSize) { | 
|  | pem_encoded.append(b64_encoded, chunk_offset, kChunkSize); | 
|  | pem_encoded.append("\n"); | 
|  | } | 
|  |  | 
|  | pem_encoded.append(kPEMHeaderEndBlock); | 
|  | pem_encoded.append(type); | 
|  | pem_encoded.append(kPEMHeaderTail); | 
|  | pem_encoded.append("\n"); | 
|  | return pem_encoded; | 
|  | } | 
|  |  | 
|  | BSSL_NAMESPACE_END |