// Rust Bitcoin Library // Written in 2014 by // Andrew Poelstra // // To the extent possible under law, the author(s) have dedicated all // copyright and related and neighboring rights to this software to // the public domain worldwide. This software is distributed without // any warranty. // // You should have received a copy of the CC0 Public Domain Dedication // along with this software. // If not, see . // //! # Bitcoin Blockchain //! //! This module provides the structures and functions to maintain the //! blockchain. //! //! Note to developers: do not expose any ref-counted pointers in the public //! API of this module. Internally we do unsafe mutations of them and we need //! to make sure we are holding the only references. //! use std::io::{IoResult, IoError, OtherIoError}; use std::num::Zero; use std::kinds::marker; use blockdata::block::{Block, BlockHeader}; use blockdata::transaction::Transaction; use blockdata::constants::{DIFFCHANGE_INTERVAL, DIFFCHANGE_TIMESPAN, TARGET_BLOCK_SPACING, max_target, genesis_block}; use network::constants::{Network, BitcoinTestnet}; use network::serialize::{Serializable, SerializeIter}; use util::BitArray; use util::error::{BitcoinResult, BlockNotFound, DuplicateHash, PrevHashNotFound}; use util::uint::Uint256; use util::hash::Sha256dHash; use util::misc::prepend_err; use util::patricia_tree::PatriciaTree; type BlockTree = PatriciaTree, Uint256>; type NodePtr = *const BlockchainNode; /// A link in the blockchain pub struct BlockchainNode { /// The actual block pub block: Block, /// Total work from genesis to this point pub total_work: Uint256, /// Expected value of `block.header.bits` for this block; only changes every /// `blockdata::constants::DIFFCHANGE_INTERVAL;` blocks pub required_difficulty: Uint256, /// Height above genesis pub height: u32, /// Whether the transaction data is stored pub has_txdata: bool, /// Pointer to block's parent prev: NodePtr, /// Pointer to block's child next: NodePtr } impl BlockchainNode { /// Is the node on the main chain? fn is_on_main_chain(&self, chain: &Blockchain) -> bool { if self.block.header == unsafe { (*chain.best_tip).block.header } { return true; } unsafe { let mut scan = self.next; while scan.is_not_null() { if (*scan).block.header == (*chain.best_tip).block.header { return true; } scan = (*scan).next; } } return false; } } impl Serializable for BlockchainNode { fn serialize(&self) -> Vec { let mut ret = vec![]; ret.extend(self.block.serialize().move_iter()); ret.extend(self.total_work.serialize().move_iter()); ret.extend(self.required_difficulty.serialize().move_iter()); ret.extend(self.height.serialize().move_iter()); ret.extend(self.has_txdata.serialize().move_iter()); // Don't serialize the prev or next pointers ret } fn deserialize>(mut iter: I) -> IoResult { Ok(BlockchainNode { block: try!(prepend_err("block", Serializable::deserialize(iter.by_ref()))), total_work: try!(prepend_err("total_work", Serializable::deserialize(iter.by_ref()))), required_difficulty: try!(prepend_err("req_difficulty", Serializable::deserialize(iter.by_ref()))), height: try!(prepend_err("height", Serializable::deserialize(iter.by_ref()))), has_txdata: try!(prepend_err("has_txdata", Serializable::deserialize(iter.by_ref()))), prev: RawPtr::null(), next: RawPtr::null() }) } // Override Serialize::hash to return the blockheader hash, since the // hash of the node itself is pretty much meaningless. fn bitcoin_hash(&self) -> Sha256dHash { self.block.header.bitcoin_hash() } } /// The blockchain pub struct Blockchain { network: Network, tree: BlockTree, best_tip: NodePtr, best_hash: Sha256dHash, genesis_hash: Sha256dHash } impl Serializable for Blockchain { fn serialize(&self) -> Vec { let mut ret = vec![]; ret.extend(self.network.serialize().move_iter()); ret.extend(self.tree.serialize().move_iter()); ret.extend(self.best_hash.serialize().move_iter()); ret.extend(self.genesis_hash.serialize().move_iter()); ret } fn serialize_iter<'a>(&'a self) -> SerializeIter<'a> { SerializeIter { data_iter: None, sub_iter_iter: box vec![ &self.network as &Serializable, &self.tree as &Serializable, &self.best_hash as &Serializable, &self.genesis_hash as &Serializable ].move_iter(), sub_iter: None, sub_started: false } } fn deserialize>(mut iter: I) -> IoResult { let network: Network = try!(prepend_err("network", Serializable::deserialize(iter.by_ref()))); let mut tree: BlockTree = try!(prepend_err("tree", Serializable::deserialize(iter.by_ref()))); let best_hash: Sha256dHash = try!(prepend_err("best_hash", Serializable::deserialize(iter.by_ref()))); let genesis_hash: Sha256dHash = try!(prepend_err("genesis_hash", Serializable::deserialize(iter.by_ref()))); // Lookup best tip let best = match tree.lookup(&best_hash.as_uint256(), 256) { Some(node) => &**node as NodePtr, None => { return Err(IoError { kind: OtherIoError, desc: "best tip reference not found in tree", detail: Some(format!("best tip {:x} not found", best_hash)) }); } }; // Lookup genesis if tree.lookup(&genesis_hash.as_uint256(), 256).is_none() { return Err(IoError { kind: OtherIoError, desc: "genesis block not found in tree", detail: Some(format!("genesis {:x} not found", genesis_hash)) }); } // Reconnect all prev pointers let raw_tree = &tree as *const _; for node in tree.mut_iter() { let hash = node.block.header.prev_blockhash.as_uint256(); let prevptr = match unsafe { (*raw_tree).lookup(&hash, 256) } { Some(node) => &**node as NodePtr, None => RawPtr::null() }; node.prev = prevptr; } // Reconnect next pointers on the main chain unsafe { let mut scan = best; while (*scan).prev.is_not_null() { let prev = (*scan).prev as *mut BlockchainNode; (*prev).next = scan; scan = prev as NodePtr; } // Check that "genesis" is the genesis if (*scan).bitcoin_hash() != genesis_hash { return Err(IoError { kind: OtherIoError, desc: "best tip did not link back to genesis", detail: Some(format!("no path from tip {:x} to genesis {:x}", best_hash, genesis_hash)) }); } } // Return the chain Ok(Blockchain { network: network, tree: tree, best_tip: best, best_hash: best_hash, genesis_hash: genesis_hash }) } } // TODO: this should maybe be public, in which case it needs to be tagged // with a ContravariantLifetime marker tying it to the tree's lifetime. struct LocatorHashIter { index: NodePtr, count: uint, skip: uint } impl LocatorHashIter { fn new(init: NodePtr) -> LocatorHashIter { LocatorHashIter { index: init, count: 0, skip: 1 } } } impl Iterator for LocatorHashIter { fn next(&mut self) -> Option { if self.index.is_null() { return None; } let ret = Some(unsafe { (*self.index).bitcoin_hash() }); // Rewind once (if we are at the genesis, this will set self.index to None) self.index = unsafe { (*self.index).prev }; // If we are not at the genesis, rewind `self.skip` times, or until we are. if self.index.is_not_null() { for _ in range(1, self.skip) { unsafe { if (*self.index).prev.is_null() { break; } self.index = (*self.index).prev; } } } self.count += 1; if self.count > 10 { self.skip *= 2; } ret } } /// An iterator over blocks in blockheight order pub struct BlockIter<'tree> { index: NodePtr, // Note: we don't actually touch the blockchain. But we need // to keep it borrowed to prevent it being mutated, since some // mutable blockchain methods call .mut_borrow() on the block // links, which would blow up if the iterator did a regular // borrow at the same time. marker: marker::ContravariantLifetime<'tree> } /// An iterator over blocks in reverse blockheight order. Note that this /// is essentially the same as if we'd implemented `DoubleEndedIterator` /// on `BlockIter` --- but we can't do that since if `BlockIter` is started /// off the main chain, it will not reach the best tip, so the iterator /// and its `.rev()` would be iterators over different chains! To avoid /// this suprising behaviour we simply use separate iterators. pub struct RevBlockIter<'tree> { index: NodePtr, // See comment in BlockIter for why we need this marker: marker::ContravariantLifetime<'tree> } /// An iterator over blocks in reverse blockheight order, which yielding only /// stale blocks (ending at the point where it would've returned a block on /// the main chain). It does this by checking if the `next` pointer of the /// next-to-by-yielded block matches the currently-yielded block. If not, scan /// forward from next-to-be-yielded block. If we hit the best tip, set the /// next-to-by-yielded block to None instead. /// /// So to handle reorgs, you create a `RevStaleBlockIter` starting from the last /// known block, and play it until it runs out, rewinding every block except for /// the last one. Since the UtxoSet `rewind` function sets its `last_hash()` to /// the prevblockhash of the rewinded block (which will be on the main chain at /// the end of the iteration), you can then sync it up same as if you were doing /// a plain old fast-forward. pub struct RevStaleBlockIter<'tree> { index: NodePtr, chain: &'tree Blockchain } impl<'tree> Iterator<&'tree BlockchainNode> for BlockIter<'tree> { fn next(&mut self) -> Option<&'tree BlockchainNode> { if self.index.is_null() { return None; } unsafe { let ret = Some(&*self.index); self.index = (*self.index).next; ret } } } impl<'tree> Iterator<&'tree BlockchainNode> for RevBlockIter<'tree> { fn next(&mut self) -> Option<&'tree BlockchainNode> { if self.index.is_null() { return None; } unsafe { let ret = Some(&*self.index); self.index = (*self.index).prev; ret } } } impl<'tree> Iterator<&'tree Block> for RevStaleBlockIter<'tree> { fn next(&mut self) -> Option<&'tree Block> { if self.index.is_null() { return None; } unsafe { let ret = Some(&(*self.index).block); let next_index = (*self.index).prev; // Check if the next block is going to be on the main chain if next_index.is_not_null() && (*next_index).next != self.index && (&*next_index).is_on_main_chain(self.chain) { self.index = RawPtr::null(); } else { self.index = next_index; } ret } } } /// This function emulates the GetCompact(SetCompact(n)) in the satoshi code, /// which drops the precision to something that can be encoded precisely in /// the nBits block header field. Savour the perversity. This is in Bitcoin /// consensus code. What. The. Fuck. fn satoshi_the_precision(n: &Uint256) -> Uint256 { // Shift by B bits right then left to turn the low bits to zero let bits = 8 * ((n.bits() + 7) / 8 - 3); let mut ret = n >> bits; // Oh, did I say B was that fucked up formula? I meant sometimes also + 8. if ret.bit(23) { ret = (ret >> 8) << 8; } ret << bits } impl Blockchain { /// Constructs a new blockchain pub fn new(network: Network) -> Blockchain { let genesis = genesis_block(network); let genhash = genesis.header.bitcoin_hash(); let new_node = box BlockchainNode { total_work: Zero::zero(), required_difficulty: genesis.header.target(), block: genesis, height: 0, has_txdata: true, prev: RawPtr::null(), next: RawPtr::null() }; let raw_ptr = &*new_node as NodePtr; Blockchain { network: network, tree: { let mut pat = PatriciaTree::new(); pat.insert(&genhash.as_uint256(), 256, new_node); pat }, best_hash: genhash, genesis_hash: genhash, best_tip: raw_ptr } } fn replace_txdata(&mut self, hash: &Uint256, txdata: Vec, has_txdata: bool) -> BitcoinResult<()> { match self.tree.lookup_mut(hash, 256) { Some(existing_block) => { unsafe { // existing_block is an Rc. Rust will not let us mutate it under // any circumstances, since if it were to be reallocated, then // all other references to it would be destroyed. However, we // just need a mutable pointer to the txdata vector; by calling // Vec::clone_from() rather than assigning, we can be assured that // no reallocation can occur, since clone_from() takes an &mut self, // which it does not own and therefore cannot move. // // To be clear: there will undoubtedly be some reallocation within // the Vec itself. We don't care about this. What we care about is // that the Vec (and more pointedly, its containing struct) does not // move, since this would invalidate the Rc that we are snookering. use std::mem::{forget, transmute}; let mutable_vec: &mut Vec = transmute(&existing_block.block.txdata); mutable_vec.clone_from(&txdata); // If mutable_vec went out of scope unhindered, it would deallocate // the Vec it points to, since Rust assumes that a mutable vector // is a unique reference (and this one is definitely not). forget(mutable_vec); // Do the same thing with the txdata flac let mutable_bool: &mut bool = transmute(&existing_block.has_txdata); *mutable_bool = has_txdata; forget(mutable_bool); } Ok(()) }, None => Err(BlockNotFound) } } /// Looks up a block in the chain and returns a reference to it pub fn get_block<'a>(&'a self, hash: Sha256dHash) -> Option<&'a Block> { self.tree.lookup(&hash.as_uint256(), 256).as_ref().map(|node| &node.block) } /// Locates a block in the chain and overwrites its txdata pub fn add_txdata(&mut self, block: Block) -> BitcoinResult<()> { self.replace_txdata(&block.header.bitcoin_hash().as_uint256(), block.txdata, true) } /// Locates a block in the chain and removes its txdata pub fn remove_txdata(&mut self, hash: Sha256dHash) -> BitcoinResult<()> { self.replace_txdata(&hash.as_uint256(), vec![], false) } /// Adds a block header to the chain pub fn add_header(&mut self, header: BlockHeader) -> BitcoinResult<()> { self.real_add_block(Block { header: header, txdata: vec![] }, false) } /// Adds a block to the chain pub fn add_block(&mut self, block: Block) -> BitcoinResult<()> { self.real_add_block(block, true) } fn real_add_block(&mut self, block: Block, has_txdata: bool) -> BitcoinResult<()> { // get_prev optimizes the common case where we are extending the best tip #[inline] fn get_prev<'a>(chain: &'a Blockchain, hash: Sha256dHash) -> Option { if hash == chain.best_hash { Some(chain.best_tip) } else { chain.tree.lookup(&hash.as_uint256(), 256).map(|boxptr| &**boxptr as NodePtr) } } // Check for multiple inserts (bitcoind from c9a09183 to 3c85d2ec doesn't // handle locator hashes properly and may return blocks multiple times, // and this may also happen in case of a reorg. if self.tree.lookup(&block.header.bitcoin_hash().as_uint256(), 256).is_some() { return Err(DuplicateHash); } // Construct node, if possible let new_block = match get_prev(self, block.header.prev_blockhash) { Some(prev) => { let difficulty = // Compute required difficulty if this is a diffchange block if (unsafe { (*prev).height } + 1) % DIFFCHANGE_INTERVAL == 0 { let timespan = unsafe { // Scan back DIFFCHANGE_INTERVAL blocks let mut scan = prev; for _ in range(0, DIFFCHANGE_INTERVAL - 1) { scan = (*scan).prev; } // Get clamped timespan between first and last blocks match (*prev).block.header.time - (*scan).block.header.time { n if n < DIFFCHANGE_TIMESPAN / 4 => DIFFCHANGE_TIMESPAN / 4, n if n > DIFFCHANGE_TIMESPAN * 4 => DIFFCHANGE_TIMESPAN * 4, n => n } }; // Compute new target let mut target = unsafe { (*prev).block.header.target() }; target = target.mul_u32(timespan); target = target / FromPrimitive::from_u64(DIFFCHANGE_TIMESPAN as u64).unwrap(); // Clamp below MAX_TARGET (difficulty 1) let max = max_target(self.network); if target > max { target = max }; // Compactify (make expressible in the 8+24 nBits float format satoshi_the_precision(&target) // On non-diffchange blocks, Testnet has a rule that any 20-minute-long // block intervals result the difficulty } else if self.network == BitcoinTestnet && block.header.time > unsafe { (*prev).block.header.time } + 2*TARGET_BLOCK_SPACING { max_target(self.network) // On the other hand, if we are in Testnet and the block interval is less // than 20 minutes, we need to scan backward to find a block for which the // previous rule did not apply, to find the "real" difficulty. } else if self.network == BitcoinTestnet { // Scan back DIFFCHANGE_INTERVAL blocks unsafe { let mut scan = prev; while (*scan).height % DIFFCHANGE_INTERVAL != 0 && (*scan).required_difficulty == max_target(self.network) { scan = (*scan).prev; } (*scan).required_difficulty } // Otherwise just use the last block's difficulty } else { unsafe { (*prev).required_difficulty } }; // Create node let ret = box BlockchainNode { total_work: block.header.work().add(unsafe { &(*prev).total_work }), block: block, required_difficulty: difficulty, height: unsafe { (*prev).height + 1 }, has_txdata: has_txdata, prev: prev, next: RawPtr::null() }; unsafe { let prev = prev as *mut BlockchainNode; (*prev).next = &*ret as NodePtr; } ret }, None => { return Err(PrevHashNotFound); } }; // spv validate the block try!(new_block.block.header.spv_validate(&new_block.required_difficulty)); // Insert the new block let raw_ptr = &*new_block as NodePtr; self.tree.insert(&new_block.block.header.bitcoin_hash().as_uint256(), 256, new_block); // Replace the best tip if necessary if unsafe { (*raw_ptr).total_work > (*self.best_tip).total_work } { self.set_best_tip(raw_ptr); } Ok(()) } /// Sets the best tip (not public) fn set_best_tip(&mut self, tip: NodePtr) { // Fix next links unsafe { let mut scan = self.best_tip; // Scan backward while (*scan).prev.is_not_null() { // If we hit the old best, there is no need to reorg. if scan == self.best_tip { break; } // Otherwise set the next-ptr and carry on let prev = (*scan).prev as *mut BlockchainNode; (*prev).next = scan; scan = (*scan).prev; } } // Set best self.best_hash = unsafe { (*tip).bitcoin_hash() }; self.best_tip = tip; } /// Returns the best tip pub fn best_tip<'a>(&'a self) -> &'a Block { unsafe { &(*self.best_tip).block } } /// Returns the best tip's blockhash pub fn best_tip_hash(&self) -> Sha256dHash { self.best_hash } /// Returns an array of locator hashes used in `getheaders` messages pub fn locator_hashes(&self) -> Vec { LocatorHashIter::new(self.best_tip).collect() } /// An iterator over all blocks in the chain starting from `start_hash` pub fn iter<'a>(&'a self, start_hash: Sha256dHash) -> BlockIter<'a> { let start = match self.tree.lookup(&start_hash.as_uint256(), 256) { Some(boxptr) => &**boxptr as NodePtr, None => RawPtr::null() }; BlockIter { index: start, marker: marker::ContravariantLifetime::<'a> } } /// An iterator over all blocks in reverse order to the genesis, starting with `start_hash` pub fn rev_iter<'a>(&'a self, start_hash: Sha256dHash) -> RevBlockIter<'a> { let start = match self.tree.lookup(&start_hash.as_uint256(), 256) { Some(boxptr) => &**boxptr as NodePtr, None => RawPtr::null() }; RevBlockIter { index: start, marker: marker::ContravariantLifetime::<'a> } } /// An iterator over all blocks -not- in the best chain, in reverse order, starting from `start_hash` pub fn rev_stale_iter<'a>(&'a self, start_hash: Sha256dHash) -> RevStaleBlockIter<'a> { let start = match self.tree.lookup(&start_hash.as_uint256(), 256) { Some(boxptr) => { // If we are already on the main chain, we have a dead iterator if boxptr.is_on_main_chain(self) { RawPtr::null() } else { &**boxptr as NodePtr } } None => RawPtr::null() }; RevStaleBlockIter { index: start, chain: self } } } #[cfg(test)] mod tests { use std::prelude::*; use std::io::IoResult; use blockdata::blockchain::Blockchain; use blockdata::constants::genesis_block; use network::constants::Bitcoin; use network::serialize::Serializable; #[test] fn blockchain_serialize_test() { let empty_chain = Blockchain::new(Bitcoin); assert_eq!(empty_chain.best_tip().header.bitcoin_hash().serialize(), genesis_block(Bitcoin).header.bitcoin_hash().serialize()); let serial = empty_chain.serialize(); assert_eq!(serial, empty_chain.serialize_iter().collect()); let deserial: IoResult = Serializable::deserialize(serial.iter().map(|n| *n)); assert!(deserial.is_ok()); let read_chain = deserial.unwrap(); assert_eq!(read_chain.best_tip().header.bitcoin_hash().serialize(), genesis_block(Bitcoin).header.bitcoin_hash().serialize()); } }