shaka-packager/packager/media/crypto/subsample_generator.cc

368 lines
13 KiB
C++

// Copyright 2018 Google LLC. All rights reserved.
//
// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file or at
// https://developers.google.com/open-source/licenses/bsd
#include "packager/media/crypto/subsample_generator.h"
#include <algorithm>
#include <limits>
#include "packager/media/base/decrypt_config.h"
#include "packager/media/base/video_stream_info.h"
#include "packager/media/codecs/av1_parser.h"
#include "packager/media/codecs/video_slice_header_parser.h"
#include "packager/media/codecs/vp8_parser.h"
#include "packager/media/codecs/vp9_parser.h"
namespace shaka {
namespace media {
namespace {
const size_t kAesBlockSize = 16u;
uint8_t GetNaluLengthSize(const StreamInfo& stream_info) {
if (stream_info.stream_type() != kStreamVideo)
return 0;
const VideoStreamInfo& video_stream_info =
static_cast<const VideoStreamInfo&>(stream_info);
return video_stream_info.nalu_length_size();
}
bool ShouldAlignProtectedData(Codec codec,
FourCC protection_scheme,
bool vp9_subsample_encryption) {
switch (codec) {
case kCodecAV1:
// Per AV1 in ISO-BMFF spec [1], BytesOfProtectedData SHALL be a multiple
// of 16 bytes.
// [1] https://aomediacodec.github.io/av1-isobmff/#subsample-encryption
return true;
case kCodecVP9:
// "VP Codec ISO Media File Format Binding" document requires that the
// encrypted bytes of each frame within the superframe must be block
// aligned so that the counter state can be computed for each frame
// within the superframe.
// ISO/IEC 23001-7:2016 10.2 'cbc1' 10.3 'cens'
// The BytesOfProtectedData size SHALL be a multiple of 16 bytes to
// avoid partial blocks in Subsamples.
// For consistency, apply block alignment to all frames when VP9 subsample
// encryption is enabled.
return vp9_subsample_encryption;
default:
// ISO/IEC 23001-7:2016 10.2 'cbc1' 10.3 'cens'
// The BytesOfProtectedData size SHALL be a multiple of 16 bytes to avoid
// partial blocks in Subsamples.
// CMAF requires 'cenc' scheme BytesOfProtectedData SHALL be a multiple of
// 16 bytes; while 'cbcs' scheme BytesOfProtectedData SHALL start on the
// first byte of video data following the slice header.
return protection_scheme == FOURCC_cbc1 ||
protection_scheme == FOURCC_cens ||
protection_scheme == FOURCC_cenc;
}
}
// A convenient util class to organize subsamples, e.g. combine consecutive
// subsamples with only clear bytes, split subsamples if the clear bytes exceeds
// 2^16 etc.
class SubsampleOrganizer {
public:
SubsampleOrganizer(bool align_protected_data,
std::vector<SubsampleEntry>* subsamples)
: align_protected_data_(align_protected_data), subsamples_(subsamples) {}
~SubsampleOrganizer() {
if (accumulated_clear_bytes_ > 0) {
PushSubsample(accumulated_clear_bytes_, 0);
accumulated_clear_bytes_ = 0;
}
}
void AddSubsample(size_t clear_bytes, size_t cipher_bytes) {
DCHECK_LT(clear_bytes, std::numeric_limits<uint32_t>::max());
DCHECK_LT(cipher_bytes, std::numeric_limits<uint32_t>::max());
if (align_protected_data_ && cipher_bytes != 0) {
const size_t misalign_bytes = cipher_bytes % kAesBlockSize;
clear_bytes += misalign_bytes;
cipher_bytes -= misalign_bytes;
}
accumulated_clear_bytes_ += clear_bytes;
// Accumulated clear bytes are handled later.
if (cipher_bytes == 0)
return;
PushSubsample(accumulated_clear_bytes_, cipher_bytes);
accumulated_clear_bytes_ = 0;
}
private:
SubsampleOrganizer(const SubsampleOrganizer&) = delete;
SubsampleOrganizer& operator=(const SubsampleOrganizer&) = delete;
void PushSubsample(size_t clear_bytes, size_t cipher_bytes) {
const uint16_t kUInt16Max = std::numeric_limits<uint16_t>::max();
while (clear_bytes > kUInt16Max) {
subsamples_->emplace_back(kUInt16Max, 0);
clear_bytes -= kUInt16Max;
}
subsamples_->emplace_back(static_cast<uint16_t>(clear_bytes),
static_cast<uint32_t>(cipher_bytes));
}
const bool align_protected_data_ = false;
std::vector<SubsampleEntry>* const subsamples_ = nullptr;
size_t accumulated_clear_bytes_ = 0;
};
} // namespace
SubsampleGenerator::SubsampleGenerator(bool vp9_subsample_encryption)
: vp9_subsample_encryption_(vp9_subsample_encryption) {}
SubsampleGenerator::~SubsampleGenerator() {}
Status SubsampleGenerator::Initialize(FourCC protection_scheme,
const StreamInfo& stream_info) {
codec_ = stream_info.codec();
nalu_length_size_ = GetNaluLengthSize(stream_info);
switch (codec_) {
case kCodecAV1:
av1_parser_.reset(new AV1Parser);
break;
case kCodecVP9:
if (vp9_subsample_encryption_)
vpx_parser_.reset(new VP9Parser);
break;
case kCodecH264:
header_parser_.reset(new H264VideoSliceHeaderParser);
break;
case kCodecH265:
header_parser_.reset(new H265VideoSliceHeaderParser);
break;
default:
// Other codecs should have nalu length size == 0.
if (nalu_length_size_ > 0) {
LOG(WARNING) << "Unknown video codec '" << codec_ << "'";
return Status(error::ENCRYPTION_FAILURE, "Unknown video codec.");
}
}
if (av1_parser_) {
// Parse configOBUs in AV1CodecConfigurationRecord if exists.
// https://aomediacodec.github.io/av1-isobmff/#av1codecconfigurationbox-syntax.
const size_t kConfigOBUsOffset = 4;
const bool has_config_obus =
stream_info.codec_config().size() > kConfigOBUsOffset;
std::vector<AV1Parser::Tile> tiles;
if (has_config_obus &&
!av1_parser_->Parse(
&stream_info.codec_config()[kConfigOBUsOffset],
stream_info.codec_config().size() - kConfigOBUsOffset, &tiles)) {
return Status(
error::ENCRYPTION_FAILURE,
"Failed to parse configOBUs in AV1CodecConfigurationRecord.");
}
DCHECK(tiles.empty());
}
if (header_parser_) {
CHECK_NE(nalu_length_size_, 0u) << "AnnexB stream is not supported yet";
if (!header_parser_->Initialize(stream_info.codec_config())) {
return Status(error::ENCRYPTION_FAILURE,
"Failed to read SPS and PPS data.");
}
}
align_protected_data_ = ShouldAlignProtectedData(codec_, protection_scheme,
vp9_subsample_encryption_);
if (protection_scheme == kAppleSampleAesProtectionScheme) {
const size_t kH264LeadingClearBytesSize = 32u;
const size_t kAudioLeadingClearBytesSize = 16u;
switch (codec_) {
case kCodecH264:
leading_clear_bytes_size_ = kH264LeadingClearBytesSize;
min_protected_data_size_ =
leading_clear_bytes_size_ + kAesBlockSize + 1u;
break;
case kCodecAAC:
FALLTHROUGH_INTENDED;
case kCodecAC3:
leading_clear_bytes_size_ = kAudioLeadingClearBytesSize;
min_protected_data_size_ = leading_clear_bytes_size_ + kAesBlockSize;
break;
case kCodecEAC3:
// E-AC3 encryption is handled by SampleAesEc3Cryptor, which also
// manages leading clear bytes.
leading_clear_bytes_size_ = 0;
min_protected_data_size_ = leading_clear_bytes_size_ + kAesBlockSize;
break;
default:
LOG(ERROR) << "Unexpected codec for SAMPLE-AES " << codec_;
return Status(error::ENCRYPTION_FAILURE,
"Unexpected codec for SAMPLE-AES.");
}
}
return Status::OK;
}
Status SubsampleGenerator::GenerateSubsamples(
const uint8_t* frame,
size_t frame_size,
std::vector<SubsampleEntry>* subsamples) {
subsamples->clear();
switch (codec_) {
case kCodecAV1:
return GenerateSubsamplesFromAV1Frame(frame, frame_size, subsamples);
case kCodecH264:
FALLTHROUGH_INTENDED;
case kCodecH265:
return GenerateSubsamplesFromH26xFrame(frame, frame_size, subsamples);
case kCodecVP9:
if (vp9_subsample_encryption_)
return GenerateSubsamplesFromVPxFrame(frame, frame_size, subsamples);
// Full sample encrypted so no subsamples.
break;
default:
// Other codecs are full sample encrypted unless there are clear leading
// bytes.
if (leading_clear_bytes_size_ > 0) {
SubsampleOrganizer subsample_organizer(align_protected_data_,
subsamples);
const size_t clear_bytes =
std::min(frame_size, leading_clear_bytes_size_);
const size_t cipher_bytes = frame_size - clear_bytes;
subsample_organizer.AddSubsample(clear_bytes, cipher_bytes);
} else {
// Full sample encrypted so no subsamples.
}
break;
}
return Status::OK;
}
void SubsampleGenerator::InjectVpxParserForTesting(
std::unique_ptr<VPxParser> vpx_parser) {
vpx_parser_ = std::move(vpx_parser);
}
void SubsampleGenerator::InjectVideoSliceHeaderParserForTesting(
std::unique_ptr<VideoSliceHeaderParser> header_parser) {
header_parser_ = std::move(header_parser);
}
void SubsampleGenerator::InjectAV1ParserForTesting(
std::unique_ptr<AV1Parser> av1_parser) {
av1_parser_ = std::move(av1_parser);
}
Status SubsampleGenerator::GenerateSubsamplesFromVPxFrame(
const uint8_t* frame,
size_t frame_size,
std::vector<SubsampleEntry>* subsamples) {
DCHECK(vpx_parser_);
std::vector<VPxFrameInfo> vpx_frames;
if (!vpx_parser_->Parse(frame, frame_size, &vpx_frames))
return Status(error::ENCRYPTION_FAILURE, "Failed to parse vpx frame.");
SubsampleOrganizer subsample_organizer(align_protected_data_, subsamples);
size_t total_size = 0;
for (const VPxFrameInfo& frame : vpx_frames) {
subsample_organizer.AddSubsample(
frame.uncompressed_header_size,
frame.frame_size - frame.uncompressed_header_size);
total_size += frame.frame_size;
}
// Add subsample for the superframe index if exists.
const bool is_superframe = vpx_frames.size() > 1;
if (is_superframe) {
const size_t index_size = frame_size - total_size;
DCHECK_LE(index_size, 2 + vpx_frames.size() * 4);
DCHECK_GE(index_size, 2 + vpx_frames.size() * 1);
subsample_organizer.AddSubsample(index_size, 0);
} else {
DCHECK_EQ(total_size, frame_size);
}
return Status::OK;
}
Status SubsampleGenerator::GenerateSubsamplesFromH26xFrame(
const uint8_t* frame,
size_t frame_size,
std::vector<SubsampleEntry>* subsamples) {
DCHECK_NE(nalu_length_size_, 0u);
DCHECK(header_parser_);
SubsampleOrganizer subsample_organizer(align_protected_data_, subsamples);
const Nalu::CodecType nalu_type =
(codec_ == kCodecH265) ? Nalu::kH265 : Nalu::kH264;
NaluReader reader(nalu_type, nalu_length_size_, frame, frame_size);
Nalu nalu;
NaluReader::Result result;
while ((result = reader.Advance(&nalu)) == NaluReader::kOk) {
const size_t nalu_total_size = nalu.header_size() + nalu.payload_size();
size_t clear_bytes = 0;
if (nalu.is_video_slice() && nalu_total_size >= min_protected_data_size_) {
clear_bytes = leading_clear_bytes_size_;
if (clear_bytes == 0) {
// For video-slice NAL units, encrypt the video slice. This skips
// the frame header.
const int64_t video_slice_header_size =
header_parser_->GetHeaderSize(nalu);
if (video_slice_header_size < 0) {
LOG(ERROR) << "Failed to read slice header.";
return Status(error::ENCRYPTION_FAILURE,
"Failed to read slice header.");
}
clear_bytes = nalu.header_size() + video_slice_header_size;
}
} else {
// For non-video-slice or small NAL units, don't encrypt.
clear_bytes = nalu_total_size;
}
const size_t cipher_bytes = nalu_total_size - clear_bytes;
subsample_organizer.AddSubsample(nalu_length_size_ + clear_bytes,
cipher_bytes);
}
if (result != NaluReader::kEOStream) {
LOG(ERROR) << "Failed to parse NAL units.";
return Status(error::ENCRYPTION_FAILURE, "Failed to parse NAL units.");
}
return Status::OK;
}
Status SubsampleGenerator::GenerateSubsamplesFromAV1Frame(
const uint8_t* frame,
size_t frame_size,
std::vector<SubsampleEntry>* subsamples) {
DCHECK(av1_parser_);
std::vector<AV1Parser::Tile> av1_tiles;
if (!av1_parser_->Parse(frame, frame_size, &av1_tiles))
return Status(error::ENCRYPTION_FAILURE, "Failed to parse AV1 frame.");
SubsampleOrganizer subsample_organizer(align_protected_data_, subsamples);
size_t last_tile_end_offset = 0;
for (const AV1Parser::Tile& tile : av1_tiles) {
DCHECK_LE(last_tile_end_offset, tile.start_offset_in_bytes);
// Per AV1 in ISO-BMFF spec [1], only decode_tile is encrypted.
// [1] https://aomediacodec.github.io/av1-isobmff/#subsample-encryption
subsample_organizer.AddSubsample(
tile.start_offset_in_bytes - last_tile_end_offset, tile.size_in_bytes);
last_tile_end_offset = tile.start_offset_in_bytes + tile.size_in_bytes;
}
DCHECK_LE(last_tile_end_offset, frame_size);
if (last_tile_end_offset < frame_size)
subsample_organizer.AddSubsample(frame_size - last_tile_end_offset, 0);
return Status::OK;
}
} // namespace media
} // namespace shaka