mirror of
https://github.com/SChernykh/p2pool.git
synced 2024-12-23 03:49:23 +00:00
443 lines
13 KiB
C++
443 lines
13 KiB
C++
/*
|
|
* This file is part of the Monero P2Pool <https://github.com/SChernykh/p2pool>
|
|
* Copyright (c) 2021-2023 SChernykh <https://github.com/SChernykh>
|
|
*
|
|
* This program is free software: you can redistribute it and/or modify
|
|
* it under the terms of the GNU General Public License as published by
|
|
* the Free Software Foundation, version 3.
|
|
*
|
|
* This program is distributed in the hope that it will be useful, but
|
|
* WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
|
* General Public License for more details.
|
|
*
|
|
* You should have received a copy of the GNU General Public License
|
|
* along with this program. If not, see <http://www.gnu.org/licenses/>.
|
|
*/
|
|
|
|
#pragma once
|
|
|
|
namespace p2pool {
|
|
|
|
// Parse an arbitrary binary blob into the pool block
|
|
// Since data here can come from external and possibly malicious sources, check everything
|
|
// Only the syntax (i.e. the serialized block binary format) and the keccak hash are checked here
|
|
// Semantics must also be checked elsewhere before accepting the block (PoW, reward split between miners, difficulty calculation and so on)
|
|
int PoolBlock::deserialize(const uint8_t* data, size_t size, const SideChain& sidechain, uv_loop_t* loop, bool compact)
|
|
{
|
|
try {
|
|
// Sanity check
|
|
if (!data || (size > MAX_BLOCK_SIZE)) {
|
|
return __LINE__;
|
|
}
|
|
|
|
const uint8_t* const data_begin = data;
|
|
const uint8_t* const data_end = data + size;
|
|
|
|
auto read_byte = [&data, data_end](uint8_t& b) -> bool
|
|
{
|
|
if (data < data_end) {
|
|
b = *(data++);
|
|
return true;
|
|
}
|
|
return false;
|
|
};
|
|
|
|
#define READ_BYTE(x) do { if (!read_byte(x)) return __LINE__; } while (0)
|
|
#define EXPECT_BYTE(value) do { uint8_t tmp; READ_BYTE(tmp); if (tmp != (value)) return __LINE__; } while (0)
|
|
|
|
#define READ_VARINT(x) do { data = readVarint(data, data_end, x); if (!data) return __LINE__; } while(0)
|
|
|
|
auto read_buf = [&data, data_end](void* buf, size_t size) -> bool
|
|
{
|
|
if (static_cast<size_t>(data_end - data) < size) {
|
|
return false;
|
|
}
|
|
|
|
memcpy(buf, data, size);
|
|
data += size;
|
|
return true;
|
|
};
|
|
|
|
#define READ_BUF(buf, size) do { if (!read_buf((buf), (size))) return __LINE__; } while(0)
|
|
|
|
READ_BYTE(m_majorVersion);
|
|
if (m_majorVersion > HARDFORK_SUPPORTED_VERSION) return __LINE__;
|
|
|
|
READ_BYTE(m_minorVersion);
|
|
if (m_minorVersion < m_majorVersion) return __LINE__;
|
|
|
|
READ_VARINT(m_timestamp);
|
|
READ_BUF(m_prevId.h, HASH_SIZE);
|
|
|
|
const int nonce_offset = static_cast<int>(data - data_begin);
|
|
READ_BUF(&m_nonce, NONCE_SIZE);
|
|
|
|
EXPECT_BYTE(TX_VERSION);
|
|
|
|
uint64_t unlock_height;
|
|
READ_VARINT(unlock_height);
|
|
|
|
EXPECT_BYTE(1);
|
|
EXPECT_BYTE(TXIN_GEN);
|
|
|
|
READ_VARINT(m_txinGenHeight);
|
|
if (m_majorVersion != sidechain.network_major_version(m_txinGenHeight)) return __LINE__;
|
|
if (unlock_height != m_txinGenHeight + MINER_REWARD_UNLOCK_TIME) return __LINE__;
|
|
|
|
std::vector<uint8_t> outputs_blob;
|
|
const int outputs_offset = static_cast<int>(data - data_begin);
|
|
|
|
uint64_t num_outputs;
|
|
READ_VARINT(num_outputs);
|
|
|
|
uint64_t total_reward = 0;
|
|
int outputs_blob_size;
|
|
|
|
if (num_outputs > 0) {
|
|
// Outputs are in the buffer, just read them
|
|
// Each output is at least 34 bytes, exit early if there's not enough data left
|
|
// 1 byte for reward, 1 byte for tx_type, 32 bytes for eph_pub_key
|
|
constexpr uint64_t MIN_OUTPUT_SIZE = 34;
|
|
|
|
if (num_outputs > std::numeric_limits<uint64_t>::max() / MIN_OUTPUT_SIZE) return __LINE__;
|
|
if (static_cast<uint64_t>(data_end - data) < num_outputs * MIN_OUTPUT_SIZE) return __LINE__;
|
|
|
|
m_outputs.resize(num_outputs);
|
|
m_outputs.shrink_to_fit();
|
|
|
|
const uint8_t expected_tx_type = get_tx_type();
|
|
|
|
for (uint64_t i = 0; i < num_outputs; ++i) {
|
|
TxOutput& t = m_outputs[i];
|
|
|
|
uint64_t reward;
|
|
READ_VARINT(reward);
|
|
t.m_reward = reward;
|
|
total_reward += reward;
|
|
|
|
EXPECT_BYTE(expected_tx_type);
|
|
|
|
READ_BUF(t.m_ephPublicKey.h, HASH_SIZE);
|
|
|
|
if (expected_tx_type == TXOUT_TO_TAGGED_KEY) {
|
|
uint8_t view_tag;
|
|
READ_BYTE(view_tag);
|
|
t.m_viewTag = view_tag;
|
|
}
|
|
}
|
|
|
|
outputs_blob_size = static_cast<int>(data - data_begin) - outputs_offset;
|
|
outputs_blob.assign(data_begin + outputs_offset, data);
|
|
|
|
m_sidechainId.clear();
|
|
}
|
|
else {
|
|
// Outputs are not in the buffer and must be calculated from sidechain data
|
|
// We only have total reward and outputs blob size here
|
|
READ_VARINT(total_reward);
|
|
|
|
uint64_t tmp;
|
|
READ_VARINT(tmp);
|
|
|
|
// Sanity check
|
|
if ((tmp == 0) || (tmp > MAX_BLOCK_SIZE)) {
|
|
return __LINE__;
|
|
}
|
|
|
|
outputs_blob_size = static_cast<int>(tmp);
|
|
|
|
READ_BUF(m_sidechainId.h, HASH_SIZE);
|
|
}
|
|
|
|
// Technically some p2pool node could keep stuffing block with transactions until reward is less than 0.6 XMR
|
|
// But default transaction picking algorithm never does that. It's better to just ban such nodes
|
|
if (total_reward < BASE_BLOCK_REWARD) {
|
|
return __LINE__;
|
|
}
|
|
|
|
const int outputs_actual_blob_size = static_cast<int>(data - data_begin) - outputs_offset;
|
|
if (outputs_blob_size < outputs_actual_blob_size) {
|
|
return __LINE__;
|
|
}
|
|
|
|
const int outputs_blob_size_diff = outputs_blob_size - outputs_actual_blob_size;
|
|
|
|
uint64_t tx_extra_size;
|
|
READ_VARINT(tx_extra_size);
|
|
|
|
const uint8_t* tx_extra_begin = data;
|
|
|
|
EXPECT_BYTE(TX_EXTRA_TAG_PUBKEY);
|
|
READ_BUF(m_txkeyPub.h, HASH_SIZE);
|
|
|
|
EXPECT_BYTE(TX_EXTRA_NONCE);
|
|
READ_VARINT(m_extraNonceSize);
|
|
|
|
// Sanity check
|
|
if ((m_extraNonceSize < EXTRA_NONCE_SIZE) || (m_extraNonceSize > EXTRA_NONCE_MAX_SIZE)) return __LINE__;
|
|
|
|
const int extra_nonce_offset = static_cast<int>((data - data_begin) + outputs_blob_size_diff);
|
|
READ_BUF(&m_extraNonce, EXTRA_NONCE_SIZE);
|
|
for (uint64_t i = EXTRA_NONCE_SIZE; i < m_extraNonceSize; ++i) {
|
|
EXPECT_BYTE(0);
|
|
}
|
|
|
|
EXPECT_BYTE(TX_EXTRA_MERGE_MINING_TAG);
|
|
|
|
uint64_t mm_field_size;
|
|
READ_VARINT(mm_field_size);
|
|
|
|
const uint8_t* const mm_field_begin = data;
|
|
|
|
READ_VARINT(m_merkleTreeData);
|
|
|
|
m_merkleTreeDataSize = static_cast<uint32_t>(data - mm_field_begin);
|
|
|
|
uint32_t mm_n_aux_chains, mm_nonce;
|
|
decode_merkle_tree_data(mm_n_aux_chains, mm_nonce);
|
|
|
|
const int mm_root_hash_offset = static_cast<int>((data - data_begin) + outputs_blob_size_diff);
|
|
READ_BUF(m_merkleRoot.h, HASH_SIZE);
|
|
|
|
if (static_cast<uint64_t>(data - mm_field_begin) != mm_field_size) {
|
|
return __LINE__;
|
|
}
|
|
|
|
if (static_cast<uint64_t>(data - tx_extra_begin) != tx_extra_size) return __LINE__;
|
|
|
|
EXPECT_BYTE(0);
|
|
|
|
uint64_t num_transactions;
|
|
READ_VARINT(num_transactions);
|
|
|
|
const int transactions_offset = static_cast<int>(data - data_begin);
|
|
|
|
std::vector<uint64_t> parent_indices;
|
|
if (compact) {
|
|
if (static_cast<uint64_t>(data_end - data) < num_transactions) return __LINE__;
|
|
|
|
m_transactions.resize(1);
|
|
parent_indices.resize(1);
|
|
|
|
// limit reserved memory size because we can't check "num_transactions" properly here
|
|
const uint64_t k = std::min<uint64_t>(num_transactions + 1, 256);
|
|
m_transactions.reserve(k);
|
|
parent_indices.reserve(k);
|
|
|
|
for (uint64_t i = 0; i < num_transactions; ++i) {
|
|
uint64_t parent_index;
|
|
READ_VARINT(parent_index);
|
|
|
|
hash id;
|
|
if (parent_index == 0) {
|
|
READ_BUF(id.h, HASH_SIZE);
|
|
}
|
|
|
|
m_transactions.emplace_back(id);
|
|
parent_indices.emplace_back(parent_index);
|
|
}
|
|
}
|
|
else {
|
|
if (num_transactions > std::numeric_limits<uint64_t>::max() / HASH_SIZE) return __LINE__;
|
|
if (static_cast<uint64_t>(data_end - data) < num_transactions * HASH_SIZE) return __LINE__;
|
|
|
|
m_transactions.resize(1);
|
|
m_transactions.reserve(num_transactions + 1);
|
|
|
|
for (uint64_t i = 0; i < num_transactions; ++i) {
|
|
hash id;
|
|
READ_BUF(id.h, HASH_SIZE);
|
|
m_transactions.emplace_back(id);
|
|
}
|
|
}
|
|
|
|
const int transactions_actual_blob_size = static_cast<int>(data - data_begin) - transactions_offset;
|
|
const int transactions_blob_size = static_cast<int>(num_transactions) * HASH_SIZE;
|
|
const int transactions_blob_size_diff = transactions_blob_size - transactions_actual_blob_size;
|
|
|
|
m_transactions.shrink_to_fit();
|
|
|
|
#if POOL_BLOCK_DEBUG
|
|
m_mainChainDataDebug.reserve((data - data_begin) + outputs_blob_size_diff + transactions_blob_size_diff);
|
|
m_mainChainDataDebug.assign(data_begin, data_begin + outputs_offset);
|
|
m_mainChainDataDebug.insert(m_mainChainDataDebug.end(), outputs_blob_size, 0);
|
|
m_mainChainDataDebug.insert(m_mainChainDataDebug.end(), data_begin + outputs_offset + outputs_actual_blob_size, data_begin + transactions_offset);
|
|
m_mainChainDataDebug.insert(m_mainChainDataDebug.end(), transactions_blob_size, 0);
|
|
m_mainChainDataDebug.insert(m_mainChainDataDebug.end(), data_begin + transactions_offset + transactions_actual_blob_size, data);
|
|
|
|
const uint8_t* sidechain_data_begin = data;
|
|
#endif
|
|
|
|
hash spend_pub_key;
|
|
hash view_pub_key;
|
|
READ_BUF(spend_pub_key.h, HASH_SIZE);
|
|
READ_BUF(view_pub_key.h, HASH_SIZE);
|
|
if (!m_minerWallet.assign(spend_pub_key, view_pub_key, sidechain.network_type())) {
|
|
return __LINE__;
|
|
}
|
|
|
|
READ_BUF(m_txkeySecSeed.h, HASH_SIZE);
|
|
|
|
hash pub;
|
|
get_tx_keys(pub, m_txkeySec, m_txkeySecSeed, m_prevId);
|
|
if (pub != m_txkeyPub) {
|
|
return __LINE__;
|
|
}
|
|
|
|
if (!check_keys(m_txkeyPub, m_txkeySec)) {
|
|
return __LINE__;
|
|
}
|
|
|
|
READ_BUF(m_parent.h, HASH_SIZE);
|
|
|
|
if (compact) {
|
|
const PoolBlock* parent = sidechain.find_block(m_parent);
|
|
if (!parent) {
|
|
return __LINE__;
|
|
}
|
|
|
|
for (uint64_t i = 1, n = m_transactions.size(); i < n; ++i) {
|
|
const uint64_t parent_index = parent_indices[i];
|
|
if (parent_index) {
|
|
if (parent_index >= parent->m_transactions.size()) {
|
|
return __LINE__;
|
|
}
|
|
m_transactions[i] = parent->m_transactions[parent_index];
|
|
}
|
|
}
|
|
}
|
|
|
|
uint64_t num_uncles;
|
|
READ_VARINT(num_uncles);
|
|
|
|
if (num_uncles > std::numeric_limits<uint64_t>::max() / HASH_SIZE) return __LINE__;
|
|
if (static_cast<uint64_t>(data_end - data) < num_uncles * HASH_SIZE) return __LINE__;
|
|
|
|
m_uncles.clear();
|
|
m_uncles.reserve(num_uncles);
|
|
|
|
for (uint64_t i = 0; i < num_uncles; ++i) {
|
|
hash id;
|
|
READ_BUF(id.h, HASH_SIZE);
|
|
m_uncles.emplace_back(id);
|
|
}
|
|
|
|
READ_VARINT(m_sidechainHeight);
|
|
|
|
READ_VARINT(m_difficulty.lo);
|
|
READ_VARINT(m_difficulty.hi);
|
|
|
|
READ_VARINT(m_cumulativeDifficulty.lo);
|
|
READ_VARINT(m_cumulativeDifficulty.hi);
|
|
|
|
uint8_t merkle_proof_size;
|
|
READ_BYTE(merkle_proof_size);
|
|
|
|
if (merkle_proof_size > 7) {
|
|
return __LINE__;
|
|
}
|
|
|
|
m_merkleProof.clear();
|
|
m_merkleProof.reserve(merkle_proof_size);
|
|
|
|
for (uint8_t i = 0; i < merkle_proof_size; ++i) {
|
|
hash h;
|
|
READ_BUF(h.h, HASH_SIZE);
|
|
m_merkleProof.emplace_back(h);
|
|
}
|
|
|
|
READ_BUF(m_sidechainExtraBuf, sizeof(m_sidechainExtraBuf));
|
|
|
|
#undef READ_BYTE
|
|
#undef EXPECT_BYTE
|
|
#undef READ_VARINT
|
|
#undef READ_BUF
|
|
|
|
if (data != data_end) {
|
|
return __LINE__;
|
|
}
|
|
|
|
if ((num_outputs == 0) && !sidechain.get_outputs_blob(this, total_reward, outputs_blob, loop)) {
|
|
return __LINE__;
|
|
}
|
|
|
|
if (static_cast<int>(outputs_blob.size()) != outputs_blob_size) {
|
|
return __LINE__;
|
|
}
|
|
|
|
const uint8_t* transactions_blob = reinterpret_cast<uint8_t*>(m_transactions.data() + 1);
|
|
|
|
#if POOL_BLOCK_DEBUG
|
|
memcpy(m_mainChainDataDebug.data() + outputs_offset, outputs_blob.data(), outputs_blob_size);
|
|
memcpy(m_mainChainDataDebug.data() + transactions_offset + outputs_blob_size_diff, transactions_blob, transactions_blob_size);
|
|
#endif
|
|
|
|
hash check;
|
|
const std::vector<uint8_t>& consensus_id = sidechain.consensus_id();
|
|
const int data_size = static_cast<int>((data_end - data_begin) + outputs_blob_size_diff + transactions_blob_size_diff);
|
|
|
|
if (data_size > static_cast<int>(MAX_BLOCK_SIZE)) {
|
|
return __LINE__;
|
|
}
|
|
|
|
keccak_custom(
|
|
[nonce_offset, extra_nonce_offset, mm_root_hash_offset, data_begin, data_size, &consensus_id, &outputs_blob, outputs_blob_size_diff, outputs_offset, outputs_blob_size, transactions_blob, transactions_blob_size_diff, transactions_offset, transactions_blob_size](int offset) -> uint8_t
|
|
{
|
|
uint32_t k = static_cast<uint32_t>(offset - nonce_offset);
|
|
if (k < NONCE_SIZE) {
|
|
return 0;
|
|
}
|
|
|
|
k = static_cast<uint32_t>(offset - extra_nonce_offset);
|
|
if (k < EXTRA_NONCE_SIZE) {
|
|
return 0;
|
|
}
|
|
|
|
k = static_cast<uint32_t>(offset - mm_root_hash_offset);
|
|
if (k < HASH_SIZE) {
|
|
return 0;
|
|
}
|
|
|
|
if (offset < data_size) {
|
|
if (offset < outputs_offset) {
|
|
return data_begin[offset];
|
|
}
|
|
else if (offset < outputs_offset + outputs_blob_size) {
|
|
return outputs_blob[offset - outputs_offset];
|
|
}
|
|
else if (offset < transactions_offset + outputs_blob_size_diff) {
|
|
return data_begin[offset - outputs_blob_size_diff];
|
|
}
|
|
else if (offset < transactions_offset + outputs_blob_size_diff + transactions_blob_size) {
|
|
return transactions_blob[offset - (transactions_offset + outputs_blob_size_diff)];
|
|
}
|
|
return data_begin[offset - outputs_blob_size_diff - transactions_blob_size_diff];
|
|
}
|
|
|
|
return consensus_id[offset - data_size];
|
|
},
|
|
static_cast<int>(size + outputs_blob_size_diff + transactions_blob_size_diff + consensus_id.size()), check.h, HASH_SIZE);
|
|
|
|
#if POOL_BLOCK_DEBUG
|
|
m_sideChainDataDebug.assign(sidechain_data_begin, data_end);
|
|
#endif
|
|
|
|
const uint32_t mm_aux_slot = get_aux_slot(sidechain.consensus_hash(), mm_nonce, mm_n_aux_chains);
|
|
|
|
if (!verify_merkle_proof(check, m_merkleProof, mm_aux_slot, mm_n_aux_chains, m_merkleRoot)) {
|
|
return __LINE__;
|
|
}
|
|
|
|
m_sidechainId = check;
|
|
}
|
|
catch (std::exception& e) {
|
|
LOGERR(0, "Exception in PoolBlock::deserialize(): " << e.what());
|
|
return __LINE__;
|
|
}
|
|
|
|
reset_offchain_data();
|
|
return 0;
|
|
}
|
|
|
|
} // namespace p2pool
|