Newer
Older
// Copyright 2015-2018 Parity Technologies (UK) Ltd.
// This file is part of Parity.
// Parity 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, either version 3 of the License, or
// (at your option) any later version.
// Parity 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 Parity. If not, see <http://www.gnu.org/licenses/>.
use std::collections::{HashSet, BTreeMap, BTreeSet, VecDeque};
use std::sync::atomic::{AtomicUsize, AtomicBool, Ordering as AtomicOrdering};
use std::sync::{Arc, Weak};
use std::time::{Instant, Duration};
use bytes::Bytes;
use itertools::Itertools;
use trie::{TrieSpec, TrieFactory, Trie};
use block::{IsBlock, LockedBlock, Drain, ClosedBlock, OpenBlock, enact_verified, SealedBlock};
use blockchain::{BlockChain, BlockProvider, TreeRoute, ImportRoute, TransactionAddress, ExtrasInsert};
use client::ancient_import::AncientVerifier;
use client::Error as ClientError;
Nonce, Balance, ChainInfo, BlockInfo, CallContract, TransactionInfo,
RegistryInfo, ReopenBlock, PrepareOpenBlock, ScheduleInfo, ImportSealedBlock,
BroadcastProposalBlock, ImportBlock, StateOrBlock, StateInfo, StateClient, Call,
AccountData, BlockChain as BlockChainTrait, BlockProducer, SealedBlockImporter,
ClientIoMessage
BlockId, TransactionId, UncleId, TraceId, ClientConfig, BlockChainClient,
TraceFilter, CallAnalytics, BlockImportError, Mode,
ChainNotify, ChainRoute, PruningInfo, ProvingBlockChainClient, EngineInfo, ChainMessageType,
IoClient,
use engines::{EthEngine, EpochTransition, ForkChoice};
use error::{ImportErrorKind, BlockImportErrorKind, ExecutionError, CallError, BlockError, ImportResult, Error as EthcoreError};
use executive::{Executive, Executed, TransactOptions, contract_address};
use header::{BlockNumber, Header, ExtendedHeader};
use io::{IoChannel, IoError};
use log_entry::LocalizedLogEntry;
use miner::{Miner, MinerService};
use ethcore_miner::pool::VerifiedTransaction;
use rand::OsRng;
use receipt::{Receipt, LocalizedReceipt};
use snapshot::{self, io as snapshot_io};
use spec::Spec;
use trace;
use trace::{TraceDB, ImportRequest as TraceImportRequest, LocalizedTrace, Database as TraceDatabase};
use transaction::{self, LocalizedTransaction, UnverifiedTransaction, SignedTransaction, Transaction, Action};
use types::filter::Filter;
use types::mode::Mode as IpcMode;
use types::ancestry_action::AncestryAction;
use verification;
use verification::{PreverifiedBlock, Verifier};
use verification::queue::BlockQueue;
use views::BlockView;
use parity_machine::{Finalizable, WithMetadata};
// re-export
pub use types::blockchain_info::BlockChainInfo;
pub use types::block_status::BlockStatus;
pub use blockchain::CacheSize as BlockChainCacheSize;
pub use verification::queue::QueueInfo as BlockQueueInfo;
use_contract!(registry, "Registry", "res/contracts/registrar.json");
const MAX_ANCIENT_BLOCKS_QUEUE_SIZE: usize = 4096;
// Max number of blocks imported at once.
const MAX_ANCIENT_BLOCKS_TO_IMPORT: usize = 4;
const MAX_QUEUE_SIZE_TO_SLEEP_ON: usize = 2;
const MIN_HISTORY_SIZE: u64 = 8;
#[derive(Default, Clone, Debug, Eq, PartialEq)]
/// Memory used by state DB
pub state_db_mem: usize,
/// Alter internal reporting to reflect the additional `block` has been processed.
pub fn accrue_block(&mut self, header: &Header, transactions: usize) {
self.transactions_applied += transactions;
self.gas_processed = self.gas_processed + *header.gas_used();
asynchronous rob
committed
impl<'a> ::std::ops::Sub<&'a ClientReport> for ClientReport {
type Output = Self;
fn sub(mut self, other: &'a ClientReport) -> Self {
let higher_mem = ::std::cmp::max(self.state_db_mem, other.state_db_mem);
let lower_mem = ::std::cmp::min(self.state_db_mem, other.state_db_mem);
self.blocks_imported -= other.blocks_imported;
self.transactions_applied -= other.transactions_applied;
self.gas_processed = self.gas_processed - other.gas_processed;
self.state_db_mem = higher_mem - lower_mem;
asynchronous rob
committed
self
}
}
struct SleepState {
last_activity: Option<Instant>,
last_autosleep: Option<Instant>,
}
impl SleepState {
fn new(awake: bool) -> Self {
SleepState {
last_activity: match awake { false => None, true => Some(Instant::now()) },
last_autosleep: match awake { false => Some(Instant::now()), true => None },
}
}
}
struct Importer {
/// Lock used during block import
pub import_lock: Mutex<()>, // FIXME Maybe wrap the whole `Importer` instead?
/// Used to verify blocks
pub verifier: Box<Verifier<Client>>,
/// Queue containing pending blocks
pub block_queue: BlockQueue,
/// Handles block sealing
pub miner: Arc<Miner>,
/// Ancient block verifier: import an ancient sequence of blocks in order from a starting epoch
pub ancient_verifier: AncientVerifier,
/// Ethereum engine to be used during import
pub engine: Arc<EthEngine>,
}
/// Blockchain database client backed by a persistent database. Owns and manages a blockchain and a block queue.
/// Call `import_block()` to import a block asynchronously; `flush_queue()` flushes the queue.
pub struct Client {
/// Flag used to disable the client forever. Not to be confused with `liveness`.
///
/// For example, auto-updater will disable client forever if there is a
/// hard fork registered on-chain that we don't have capability for.
/// When hard fork block rolls around, the client (if `update` is false)
/// knows it can't proceed further.
chain: RwLock<Arc<BlockChain>>,
tracedb: RwLock<TraceDB<BlockChain>>,
/// Database pruning strategy to use for StateDB
/// Client uses this to store blocks, traces, etc.
db: RwLock<Arc<KeyValueDB>>,
/// Flag changed by `sleep` and `wake_up` methods. Not to be confused with `enabled`.
io_channel: Mutex<IoChannel<ClientIoMessage>>,
/// List of actors to be notified on certain chain events
notify: RwLock<Vec<Weak<ChainNotify>>>,
/// Queued transactions from IO
queue_transactions: IoChannelQueue,
/// Ancient blocks import queue
queue_ancient_blocks: IoChannelQueue,
/// Queued ancient blocks, make sure they are imported in order.
queued_ancient_blocks: Arc<RwLock<(
HashSet<H256>,
VecDeque<(Header, Bytes, Bytes)>
)>>,
ancient_blocks_import_lock: Arc<Mutex<()>>,
/// Consensus messages import queue
queue_consensus_message: IoChannelQueue,
last_hashes: RwLock<VecDeque<H256>>,
factories: Factories,
/// Number of eras kept in a journal before they are pruned
/// An action to be done if a mode/spec_name change happens
on_user_defaults_change: Mutex<Option<Box<FnMut(Option<Mode>) + 'static + Send>>>,
/// Link to a registry object useful for looking up names
registrar: registry::Registry,
registrar_address: Option<Address>,
/// A closure to call when we want to restart the client
exit_handler: Mutex<Option<Box<Fn(String) + 'static + Send>>>,
config: &ClientConfig,
engine: Arc<EthEngine>,
message_channel: IoChannel<ClientIoMessage>,
miner: Arc<Miner>,
) -> Result<Importer, ::error::Error> {
let block_queue = BlockQueue::new(config.queue.clone(), engine.clone(), message_channel.clone(), config.verifier_type.verifying_seal());
Tomusdrw
committed
verifier: verification::new(config.verifier_type.clone()),
block_queue,
miner,
ancient_verifier: AncientVerifier::new(engine.clone()),
/// This is triggered by a message coming from a block queue when the block is ready for insertion
pub fn import_verified_blocks(&self, client: &Client) -> usize {
// Shortcut out if we know we're incapable of syncing the chain.
if !client.enabled.load(AtomicOrdering::Relaxed) {
let (imported_blocks, import_results, invalid_blocks, imported, proposed_blocks, duration, is_empty) = {
let mut imported_blocks = Vec::with_capacity(max_blocks_to_import);
let mut invalid_blocks = HashSet::new();
let mut proposed_blocks = Vec::with_capacity(max_blocks_to_import);
let mut import_results = Vec::with_capacity(max_blocks_to_import);
let _import_lock = self.import_lock.lock();
let blocks = self.block_queue.drain(max_blocks_to_import);
if blocks.is_empty() {
return 0;
}
trace_time!("import_verified_blocks");
let start = Instant::now();
for block in blocks {
let header = block.header.clone();
let bytes = block.bytes.clone();
let hash = header.hash();
let is_invalid = invalid_blocks.contains(header.parent_hash());
invalid_blocks.insert(hash);
if let Ok(closed_block) = self.check_and_close_block(block, client) {
if self.engine.is_proposal(&header) {
self.block_queue.mark_as_good(&[hash]);
proposed_blocks.push(bytes);
imported_blocks.push(hash);
let transactions_len = closed_block.transactions().len();
let route = self.commit_block(closed_block, &header, &bytes, client);
client.report.write().accrue_block(&header, transactions_len);
} else {
invalid_blocks.insert(header.hash());
}
let imported = imported_blocks.len();
let invalid_blocks = invalid_blocks.into_iter().collect::<Vec<H256>>();
if !invalid_blocks.is_empty() {
self.block_queue.mark_as_bad(&invalid_blocks);
}
let is_empty = self.block_queue.mark_as_good(&imported_blocks);
(imported_blocks, import_results, invalid_blocks, imported, proposed_blocks, start.elapsed(), is_empty)
};
{
if !imported_blocks.is_empty() && is_empty {
let route = ChainRoute::from(import_results.as_ref());
self.miner.chain_new_blocks(client, &imported_blocks, &invalid_blocks, route.enacted(), route.retracted(), false);
}
client.notify(|notify| {
notify.new_blocks(
imported_blocks.clone(),
invalid_blocks.clone(),
Vec::new(),
proposed_blocks.clone(),
duration,
);
});
}
}
client.db.read().flush().expect("DB flush failed.");
imported
}
fn check_and_close_block(&self, block: PreverifiedBlock, client: &Client) -> Result<LockedBlock, ()> {
let header = block.header.clone();
// Check the block isn't so old we won't be able to enact it.
let best_block_number = client.chain.read().best_block_number();
if client.pruning_info().earliest_state > header.number() {
warn!(target: "client", "Block import failed for #{} ({})\nBlock is ancient (current best block: #{}).", header.number(), header.hash(), best_block_number);
return Err(());
}
// Check if parent is in chain
let parent = match client.block_header_decoded(BlockId::Hash(*header.parent_hash())) {
Some(h) => h,
None => {
warn!(target: "client", "Block import failed for #{} ({}): Parent not found ({}) ", header.number(), header.hash(), header.parent_hash());
return Err(());
}
};
// Verify Block Family
let verify_family_result = self.verifier.verify_block_family(
&parent,
engine,
Some(verification::FullFamilyParams {
block_bytes: &block.bytes,
transactions: &block.transactions,
block_provider: &**chain,
client
}),
);
if let Err(e) = verify_family_result {
warn!(target: "client", "Stage 3 block verification failed for #{} ({})\nError: {:?}", header.number(), header.hash(), e);
return Err(());
let verify_external_result = self.verifier.verify_block_external(&header, engine);
if let Err(e) = verify_external_result {
warn!(target: "client", "Stage 4 block verification failed for #{} ({})\nError: {:?}", header.number(), header.hash(), e);
return Err(());
};
// Enact Verified Block
let last_hashes = client.build_last_hashes(header.parent_hash());
let db = client.state_db.read().boxed_clone_canon(header.parent_hash());
let is_epoch_begin = chain.epoch_transition(parent.number(), *header.parent_hash()).is_some();
let enact_result = enact_verified(
block,
engine,
client.tracedb.read().tracing_enabled(),
db,
&parent,
last_hashes,
client.factories.clone(),
is_epoch_begin,
&mut chain.ancestry_with_metadata_iter(*header.parent_hash()),
let mut locked_block = enact_result.map_err(|e| {
warn!(target: "client", "Block import failed for #{} ({})\nError: {:?}", header.number(), header.hash(), e);
})?;
// Strip receipts for blocks before validate_receipts_transition,
// if the expected receipts root header does not match.
// (i.e. allow inconsistency in receipts outcome before the transition block)
if header.number() < engine.params().validate_receipts_transition
&& header.receipts_root() != locked_block.block().header().receipts_root()
{
locked_block.strip_receipts_outcomes();
}
if let Err(e) = self.verifier.verify_block_final(&header, locked_block.block().header()) {
warn!(target: "client", "Stage 5 block verification failed for #{} ({})\nError: {:?}", header.number(), header.hash(), e);
return Err(());
}
Ok(locked_block)
/// Import a block with transaction receipts.
///
/// The block is guaranteed to be the next best blocks in the
/// first block sequence. Does no sealing or transaction validation.
fn import_old_block(&self, header: &Header, block_bytes: &[u8], receipts_bytes: &[u8], db: &KeyValueDB, chain: &BlockChain) -> Result<H256, ::error::Error> {
let receipts = ::rlp::decode_list(receipts_bytes);
let _import_lock = self.import_lock.lock();
trace_time!("import_old_block");
// verify the block, passing the chain for updating the epoch verifier.
let mut rng = OsRng::new().map_err(UtilError::from)?;
self.ancient_verifier.verify(&mut rng, &header, &chain)?;
let mut batch = DBTransaction::new();
chain.insert_unordered_block(&mut batch, block_bytes, receipts, None, false, true);
// NOTE: the header of the block passed here is not necessarily sealed, as
// it is for reconstructing the state transition.
//
// The header passed is from the original block data and is sealed.
fn commit_block<B>(&self, block: B, header: &Header, block_data: &[u8], client: &Client) -> ImportRoute where B: IsBlock + Drain {
let hash = &header.hash();
let number = header.number();
let parent = header.parent_hash();
let receipts = block.receipts().to_owned();
let traces = block.traces().clone().drain();
assert_eq!(header.hash(), view!(BlockView, block_data).header_view().hash());
//let traces = From::from(block.traces().clone().unwrap_or_else(Vec::new));
let mut batch = DBTransaction::new();
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
let ancestry_actions = self.engine.ancestry_actions(block.block(), &mut chain.ancestry_with_metadata_iter(*parent));
let best_hash = chain.best_block_hash();
let metadata = block.block().metadata().map(Into::into);
let is_finalized = block.block().is_finalized();
let new = ExtendedHeader {
header: header.clone(),
is_finalized: is_finalized,
metadata: metadata,
parent_total_difficulty: chain.block_details(&parent).expect("Parent block is in the database; qed").total_difficulty
};
let best = {
let hash = best_hash;
let header = chain.block_header_data(&hash)
.expect("Best block is in the database; qed")
.decode()
.expect("Stored block header is valid RLP; qed");
let details = chain.block_details(&hash)
.expect("Best block is in the database; qed");
ExtendedHeader {
parent_total_difficulty: details.total_difficulty - *header.difficulty(),
is_finalized: details.is_finalized,
metadata: details.metadata,
header: header,
}
};
let route = chain.tree_route(best_hash, *parent).expect("forks are only kept when it has common ancestors; tree route from best to prospective's parent always exists; qed");
let fork_choice = if route.is_from_route_finalized {
ForkChoice::Old
} else {
self.engine.fork_choice(&new, &best)
};
// CHECK! I *think* this is fine, even if the state_root is equal to another
// already-imported block of the same number.
// TODO: Prove it with a test.
let mut state = block.drain();
// check epoch end signal, potentially generating a proof on the current
// state.
self.check_epoch_end_signal(
&header,
block_data,
&receipts,
&state,
&chain,
&mut batch,
state.journal_under(&mut batch, number, hash).expect("DB commit failed");
for ancestry_action in ancestry_actions {
let AncestryAction::MarkFinalized(ancestry) = ancestry_action;
chain.mark_finalized(&mut batch, ancestry).expect("Engine's ancestry action must be known blocks; qed");
}
let route = chain.insert_block(&mut batch, block_data, receipts.clone(), ExtrasInsert {
fork_choice: fork_choice,
is_finalized: is_finalized,
metadata: new.metadata,
});
client.tracedb.read().import(&mut batch, TraceImportRequest {
block_hash: hash.clone(),
block_number: number,
enacted: route.enacted.clone(),
retracted: route.retracted.len()
});
let is_canon = route.enacted.last().map_or(false, |h| h == hash);
state.sync_cache(&route.enacted, &route.retracted, is_canon);
self.check_epoch_end(&header, &chain, client);
client.update_last_hashes(&parent, hash);
if let Err(e) = client.prune_ancient(state, &chain) {
warn!("Failed to prune ancient state data: {}", e);
}
// check for epoch end signal and write pending transition if it occurs.
// state for the given block must be available.
fn check_epoch_end_signal(
&self,
header: &Header,
block_bytes: &[u8],
receipts: &[Receipt],
state_db: &StateDB,
chain: &BlockChain,
batch: &mut DBTransaction,
use engines::EpochChange;
let auxiliary = ::machine::AuxiliaryData {
bytes: Some(block_bytes),
receipts: Some(&receipts),
};
match self.engine.signals_epoch_end(header, auxiliary) {
EpochChange::Yes(proof) => {
use engines::epoch::PendingTransition;
use engines::Proof;
let proof = match proof {
Proof::Known(proof) => proof,
Proof::WithState(with_state) => {
let env_info = EnvInfo {
number: header.number(),
author: header.author().clone(),
timestamp: header.timestamp(),
difficulty: header.difficulty().clone(),
last_hashes: client.build_last_hashes(header.parent_hash()),
gas_used: U256::default(),
gas_limit: u64::max_value().into(),
};
let call = move |addr, data| {
let mut state_db = state_db.boxed_clone();
let backend = ::state::backend::Proving::new(state_db.as_hashdb_mut());
let transaction =
client.contract_call_tx(BlockId::Hash(*header.parent_hash()), addr, data);
let mut state = State::from_existing(
backend,
header.state_root().clone(),
self.engine.account_start_nonce(header.number()),
).expect("state known to be available for just-imported block; qed");
let options = TransactOptions::with_no_tracing().dont_check_nonce();
let res = Executive::new(&mut state, &env_info, self.engine.machine())
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
let res = match res {
Err(ExecutionError::Internal(e)) =>
Err(format!("Internal error: {}", e)),
Err(e) => {
trace!(target: "client", "Proved call failed: {}", e);
Ok((Vec::new(), state.drop().1.extract_proof()))
}
Ok(res) => Ok((res.output, state.drop().1.extract_proof())),
};
res.map(|(output, proof)| (output, proof.into_iter().map(|x| x.into_vec()).collect()))
};
match with_state.generate_proof(&call) {
Ok(proof) => proof,
Err(e) => {
warn!(target: "client", "Failed to generate transition proof for block {}: {}", hash, e);
warn!(target: "client", "Snapshots produced by this client may be incomplete");
Vec::new()
}
}
}
};
debug!(target: "client", "Block {} signals epoch end.", hash);
let pending = PendingTransition { proof: proof };
chain.insert_pending_transition(batch, hash, pending);
},
EpochChange::No => {},
EpochChange::Unsure(_) => {
warn!(target: "client", "Detected invalid engine implementation.");
warn!(target: "client", "Engine claims to require more block data, but everything provided.");
}
}
}
// check for ending of epoch and write transition if it occurs.
fn check_epoch_end<'a>(&self, header: &'a Header, chain: &BlockChain, client: &Client) {
let is_epoch_end = self.engine.is_epoch_end(
header,
&(|hash| client.block_header_decoded(BlockId::Hash(hash))),
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
&(|hash| chain.get_pending_transition(hash)), // TODO: limit to current epoch.
);
if let Some(proof) = is_epoch_end {
debug!(target: "client", "Epoch transition at block {}", header.hash());
let mut batch = DBTransaction::new();
chain.insert_epoch_transition(&mut batch, header.number(), EpochTransition {
block_hash: header.hash(),
block_number: header.number(),
proof: proof,
});
// always write the batch directly since epoch transition proofs are
// fetched from a DB iterator and DB iterators are only available on
// flushed data.
client.db.read().write(batch).expect("DB flush failed");
}
}
}
impl Client {
/// Create a new client with given parameters.
/// The database is assumed to have been initialized with the correct columns.
pub fn new(
config: ClientConfig,
spec: &Spec,
db: Arc<KeyValueDB>,
miner: Arc<Miner>,
message_channel: IoChannel<ClientIoMessage>,
) -> Result<Arc<Client>, ::error::Error> {
let trie_spec = match config.fat_db {
true => TrieSpec::Fat,
false => TrieSpec::Secure,
};
let trie_factory = TrieFactory::new(trie_spec);
let factories = Factories {
vm: VmFactory::new(config.vm_type.clone(), config.jump_table_size),
trie: trie_factory,
accountdb: Default::default(),
};
let journal_db = journaldb::new(db.clone(), config.pruning, ::db::COL_STATE);
let mut state_db = StateDB::new(journal_db, config.state_cache_size);
if state_db.journal_db().is_empty() {
// Sets the correct state root.
state_db = spec.ensure_db_good(state_db, &factories)?;
let mut batch = DBTransaction::new();
state_db.journal_under(&mut batch, 0, &spec.genesis_header().hash())?;
db.write(batch).map_err(ClientError::Database)?;
}
let gb = spec.genesis_block();
let chain = Arc::new(BlockChain::new(config.blockchain.clone(), &gb, db.clone()));
let tracedb = RwLock::new(TraceDB::new(config.tracing.clone(), db.clone(), chain.clone()));
trace!("Cleanup journal: DB Earliest = {:?}, Latest = {:?}", state_db.journal_db().earliest_era(), state_db.journal_db().latest_era());
let history = if config.history < MIN_HISTORY_SIZE {
info!(target: "client", "Ignoring pruning history parameter of {}\
, falling back to minimum of {}",
config.history, MIN_HISTORY_SIZE);
MIN_HISTORY_SIZE
} else {
config.history
};
if !chain.block_header_data(&chain.best_block_hash()).map_or(true, |h| state_db.journal_db().contains(&h.state_root())) {
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
warn!("State root not found for block #{} ({:x})", chain.best_block_number(), chain.best_block_hash());
}
let engine = spec.engine.clone();
let awake = match config.mode { Mode::Dark(..) | Mode::Off => false, _ => true };
let importer = Importer::new(&config, engine.clone(), message_channel.clone(), miner)?;
let registrar_address = engine.additional_params().get("registrar").and_then(|s| Address::from_str(s).ok());
if let Some(ref addr) = registrar_address {
trace!(target: "client", "Found registrar at {}", addr);
}
let client = Arc::new(Client {
enabled: AtomicBool::new(true),
sleep_state: Mutex::new(SleepState::new(awake)),
liveness: AtomicBool::new(awake),
mode: Mutex::new(config.mode.clone()),
chain: RwLock::new(chain),
tracedb: tracedb,
engine: engine,
pruning: config.pruning.clone(),
config: config,
db: RwLock::new(db),
state_db: RwLock::new(state_db),
report: RwLock::new(Default::default()),
io_channel: Mutex::new(message_channel),
notify: RwLock::new(Vec::new()),
queue_transactions: IoChannelQueue::new(MAX_TX_QUEUE_SIZE),
queue_ancient_blocks: IoChannelQueue::new(MAX_ANCIENT_BLOCKS_QUEUE_SIZE),
queued_ancient_blocks: Default::default(),
ancient_blocks_import_lock: Default::default(),
queue_consensus_message: IoChannelQueue::new(usize::max_value()),
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
last_hashes: RwLock::new(VecDeque::new()),
factories: factories,
history: history,
on_user_defaults_change: Mutex::new(None),
registrar: registry::Registry::default(),
registrar_address,
exit_handler: Mutex::new(None),
importer,
});
// prune old states.
{
let state_db = client.state_db.read().boxed_clone();
let chain = client.chain.read();
client.prune_ancient(state_db, &chain)?;
}
// ensure genesis epoch proof in the DB.
{
let chain = client.chain.read();
let gh = spec.genesis_header();
if chain.epoch_transition(0, gh.hash()).is_none() {
trace!(target: "client", "No genesis transition found.");
let proof = client.with_proving_caller(
BlockId::Number(0),
|call| client.engine.genesis_epoch_data(&gh, call)
);
let proof = match proof {
Ok(proof) => proof,
Err(e) => {
warn!(target: "client", "Error generating genesis epoch data: {}. Snapshots generated may not be complete.", e);
Vec::new()
}
};
debug!(target: "client", "Obtained genesis transition proof: {:?}", proof);
let mut batch = DBTransaction::new();
chain.insert_epoch_transition(&mut batch, 0, EpochTransition {
block_hash: gh.hash(),
block_number: 0,
proof: proof,
});
client.db.read().write_buffered(batch);
}
}
// ensure buffered changes are flushed.
client.db.read().flush().map_err(ClientError::Database)?;
Ok(client)
}
/// Wakes up client if it's a sleep.
pub fn keep_alive(&self) {
let should_wake = match *self.mode.lock() {
Mode::Dark(..) | Mode::Passive(..) => true,
_ => false,
};
if should_wake {
self.wake_up();
(*self.sleep_state.lock()).last_activity = Some(Instant::now());
}
}
/// Adds an actor to be notified on certain events
pub fn add_notify(&self, target: Arc<ChainNotify>) {
self.notify.write().push(Arc::downgrade(&target));
}
/// Set a closure to call when the client wants to be restarted.
///
/// The parameter passed to the callback is the name of the new chain spec to use after
/// the restart.
pub fn set_exit_handler<F>(&self, f: F) where F: Fn(String) + 'static + Send {
*self.exit_handler.lock() = Some(Box::new(f));
}
/// Returns engine reference.
pub fn engine(&self) -> &EthEngine {
&*self.engine
}
fn notify<F>(&self, f: F) where F: Fn(&ChainNotify) {
for np in &*self.notify.read() {
if let Some(n) = np.upgrade() {
f(&*n);
/// Register an action to be done if a mode/spec_name change happens.
pub fn on_user_defaults_change<F>(&self, f: F) where F: 'static + FnMut(Option<Mode>) + Send {
*self.on_user_defaults_change.lock() = Some(Box::new(f));
}
/// Flush the block import queue.
pub fn flush_queue(&self) {
self.importer.block_queue.flush();
while !self.importer.block_queue.queue_info().is_empty() {
self.import_verified_blocks();
}
}
/// The env info as of the best block.
pub fn latest_env_info(&self) -> EnvInfo {
self.env_info(BlockId::Latest).expect("Best block header always stored; qed")
}
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
/// The env info as of a given block.
/// returns `None` if the block unknown.
pub fn env_info(&self, id: BlockId) -> Option<EnvInfo> {
self.block_header(id).map(|header| {
EnvInfo {
number: header.number(),
author: header.author(),
timestamp: header.timestamp(),
difficulty: header.difficulty(),
last_hashes: self.build_last_hashes(&header.parent_hash()),
gas_used: U256::default(),
gas_limit: header.gas_limit(),
}
})
}
fn build_last_hashes(&self, parent_hash: &H256) -> Arc<LastHashes> {
{
let hashes = self.last_hashes.read();
if hashes.front().map_or(false, |h| h == parent_hash) {
let mut res = Vec::from(hashes.clone());
res.resize(256, H256::default());
return Arc::new(res);
}
}
let mut last_hashes = LastHashes::new();
last_hashes.resize(256, H256::default());
last_hashes[0] = parent_hash.clone();
let chain = self.chain.read();
for i in 0..255 {
match chain.block_details(&last_hashes[i]) {
Some(details) => {
last_hashes[i + 1] = details.parent.clone();
},
None => break,
}
let mut cached_hashes = self.last_hashes.write();
*cached_hashes = VecDeque::from(last_hashes.clone());
Arc::new(last_hashes)
}
/// This is triggered by a message coming from a block queue when the block is ready for insertion
pub fn import_verified_blocks(&self) -> usize {
self.importer.import_verified_blocks(self)
}
// use a state-proving closure for the given block.
fn with_proving_caller<F, T>(&self, id: BlockId, with_call: F) -> T
where F: FnOnce(&::machine::Call) -> T
{
let call = |a, d| {
let tx = self.contract_call_tx(id, a, d);
let (result, items) = self.prove_transaction(tx, id)
.ok_or_else(|| format!("Unable to make call. State unavailable?"))?;
let items = items.into_iter().map(|x| x.to_vec()).collect();
Ok((result, items))
};
with_call(&call)
}
// prune ancient states until below the memory limit or only the minimum amount remain.
fn prune_ancient(&self, mut state_db: StateDB, chain: &BlockChain) -> Result<(), ClientError> {
let number = match state_db.journal_db().latest_era() {
Some(n) => n,
None => return Ok(()),
};
// prune all ancient eras until we're below the memory target,
// but have at least the minimum number of states.
loop {
let needs_pruning = state_db.journal_db().is_pruned() &&
state_db.journal_db().journal_size() >= self.config.history_mem;
if !needs_pruning { break }
match state_db.journal_db().earliest_era() {
Some(era) if era + self.history <= number => {
trace!(target: "client", "Pruning state for ancient era {}", era);
match chain.block_hash(era) {
Some(ancient_hash) => {
let mut batch = DBTransaction::new();
state_db.mark_canonical(&mut batch, era, &ancient_hash)?;
self.db.read().write_buffered(batch);
state_db.journal_db().flush();
}
None =>
debug!(target: "client", "Missing expected hash for block {}", era),
}
}
_ => break, // means that every era is kept, no pruning necessary.
}
}
Ok(())
}
fn update_last_hashes(&self, parent: &H256, hash: &H256) {
let mut hashes = self.last_hashes.write();
if hashes.front().map_or(false, |h| h == parent) {
if hashes.len() > 255 {
hashes.pop_back();
}
hashes.push_front(hash.clone());
}
}
/// Get shared miner reference.
pub fn miner(&self) -> Arc<Miner> {
/// Replace io channel. Useful for testing.
pub fn set_io_channel(&self, io_channel: IoChannel<ClientIoMessage>) {
*self.io_channel.lock() = io_channel;