path: root/src/Network/BitTorrent/DHT/Session.hs

-- |
--   Copyright   :  (c) Sam Truzjan 2013-2014
--   License     :  BSD3
--   Maintainer  :  pxqr.sta@gmail.com
--   Stability   :  experimental
--   Portability :  portable
--
--   This module defines internal state of a node instance. You can
--   have multiple nodes per application but usually you don't have
--   to. Normally, you don't need to import this module, use
--   "Network.BitTorrent.DHT" instead.
--
{-# LANGUAGE RecordWildCards            #-}
{-# LANGUAGE FlexibleContexts           #-}
{-# LANGUAGE FlexibleInstances          #-}
{-# LANGUAGE GeneralizedNewtypeDeriving #-}
{-# LANGUAGE MultiParamTypeClasses      #-}
{-# LANGUAGE ScopedTypeVariables        #-}
{-# LANGUAGE TypeFamilies               #-}
{-# LANGUAGE TemplateHaskell            #-}
module Network.BitTorrent.DHT.Session
       ( -- * Options
         -- | Use @optFooBar def@ to get default 'Alpha' or 'K'.
         Alpha
       , K
       , Options (..)

         -- * Session
       , Node
       , options
       , thisNodeId

         -- ** Initialization
       , LogFun
       , NodeHandler
       , newNode
       , closeNode

         -- * DHT
         -- | Use @asks options@ to get options passed to 'startNode'
         -- or @asks thisNodeId@ to get id of locally running node.
       , DHT
       , runDHT

         -- ** Tokens
       , grantToken
       , checkToken

         -- ** Routing table
       , getTable
       , getClosest
       , insertNode

         -- ** Peer storage
       , insertPeer
       , getPeerList
       , insertTopic
       , deleteTopic

         -- ** Messaging
       , queryNode
       , queryParallel
       , (<@>)
       ) where

import Prelude hiding (ioError)

import Control.Applicative
import Control.Concurrent.STM
import Control.Concurrent.Lifted hiding (yield)
import Control.Concurrent.Async.Lifted
import Control.Exception.Lifted hiding (Handler)
import Control.Monad.Base
import Control.Monad.Logger
import Control.Monad.Reader
import Control.Monad.Trans.Control
import Control.Monad.Trans.Resource
import Data.Default
import Data.Fixed
import Data.Hashable
import Data.List as L
import Data.Maybe
import Data.Monoid
import Data.Set as S
import Data.Text as T
import Data.Time
import Data.Time.Clock.POSIX
import Network (PortNumber)
import System.Log.FastLogger
import System.Random (randomIO)
import Text.PrettyPrint as PP hiding ((<>), ($$))
import Text.PrettyPrint.Class

import Data.Torrent.InfoHash
import Network.KRPC hiding (Options, def)
import qualified Network.KRPC as KRPC (def)
import Network.BitTorrent.Core
import Network.BitTorrent.Core.PeerAddr as P
import Network.BitTorrent.DHT.Message
import Network.BitTorrent.DHT.Routing as R
import Network.BitTorrent.DHT.Token   as T

{-----------------------------------------------------------------------
-- Options
-----------------------------------------------------------------------}

-- | Node lookups can proceed asynchronously.
type Alpha = Int

-- NOTE: libtorrent uses 5, azureus uses 10
-- | The quantity of simultaneous lookups is typically three.
defaultAlpha :: Alpha
defaultAlpha = 3

-- TODO add replication loop

-- TODO do not insert infohash -> peeraddr if infohash is too far from
-- this node id

data Order
  = NearFirst
  | FarFirst
  | Random

data Traversal
  = Greedy     -- ^ aggressive short-circuit traversal
  | Exhaustive -- ^

-- | Original Kamelia DHT uses term /publish/ for data replication
-- process. BitTorrent DHT uses term /announce/ since the purpose of
-- the DHT is peer discovery. Later in documentation, we use terms
-- /publish/ and /announce/ interchangible.
data Options = Options
  { -- | The degree of parallelism in 'find_node' queries. More
    -- parallism lead to faster bootstrapping and lookup operations,
    -- but also increase resource usage.
    --
    --   Normally this parameter should not exceed 'optK'.
    optAlpha       :: {-# UNPACK #-} !Alpha

    -- | /K/ parameter - number of nodes to return in 'find_node'
    -- responses.
  , optK           :: {-# UNPACK #-} !K

    -- | Number of buckets to maintain. This parameter depends on
    -- amount of nodes in the DHT network.
  , optBucketCount :: {-# UNPACK #-} !BucketCount

    -- | RPC timeout.
  , optTimeout     :: !NominalDiffTime

    -- | /R/ parameter - how many target nodes the 'announce' query
    -- should affect.
    --
    --   A large replica set compensates for inconsistent routing and
    --   reduces the need to frequently republish data for
    --   persistence. This comes at an increased cost for
    --   'Network.BitTorrent.DHT.insert' in terms of time, nodes
    --   contacted, and storage.
  , optReplication :: {-# UNPACK #-} !NodeCount

    -- | How often this node should republish (or reannounce) its
    -- data.
    --
    -- Large replica set ('optReplication') should require
    -- smaller reannounce intervals ('optReannounce').
  , optReannounce  :: !NominalDiffTime

    -- | The time it takes for data to expire in the
    --   network. Publisher of the data should republish (or
    --   reannounce) data to keep it in the network.
    --
    --   The /data expired timeout/ should be more than 'optReannounce'
    --   interval.
  , optDataExpired :: !NominalDiffTime
  } deriving (Show, Eq)

-- | Optimal options for bittorrent client. For short-lifetime
-- utilities you most likely need to tune 'optAlpha' and
-- 'optBucketCount'.
instance Default Options where
  def = Options
    { optAlpha       = defaultAlpha
    , optK           = defaultK

      -- see Fig.2 from "BitTorrent Mainline DHT Measurement" paper.
    , optBucketCount = defaultBucketCount

      -- see Fig.4 from "Profiling a Million User DHT" paper.
    , optTimeout     = 5  -- seconds
    , optReplication = 20 -- nodes
    , optReannounce  = 15 * 60
    , optDataExpired = 60 * 60
    }

seconds :: NominalDiffTime -> Int
seconds dt = fromEnum (realToFrac dt :: Uni)

{-----------------------------------------------------------------------
-- Tokens policy
-----------------------------------------------------------------------}

data SessionTokens = SessionTokens
  { tokenMap    :: !TokenMap
  , lastUpdate  :: !UTCTime
  , maxInterval :: !NominalDiffTime
  }

nullSessionTokens :: IO SessionTokens
nullSessionTokens = SessionTokens
  <$> (tokens <$> liftIO randomIO)
  <*> liftIO getCurrentTime
  <*> pure defaultUpdateInterval

-- TODO invalidate *twice* if needed
invalidateTokens :: UTCTime -> SessionTokens -> SessionTokens
invalidateTokens curTime ts @ SessionTokens {..}
  | curTime `diffUTCTime` lastUpdate > maxInterval = SessionTokens
    { tokenMap    = update tokenMap
    , lastUpdate  = curTime
    , maxInterval = maxInterval
    }
  |                  otherwise                     = ts

{-----------------------------------------------------------------------
-- Session
-----------------------------------------------------------------------}

-- | A set of torrents this peer intends to share.
type AnnounceSet = Set (InfoHash, PortNumber)

-- | Logger function.
type LogFun = Loc -> LogSource -> LogLevel -> LogStr -> IO ()

-- | DHT session keep track state of /this/ node.
data Node ip = Node
  { -- | Session configuration;
    options       :: !Options

    -- | Pseudo-unique self-assigned session identifier. This value is
    -- constant during DHT session and (optionally) between sessions.
  , thisNodeId    :: !NodeId

  , resources     :: !InternalState
  , manager       :: !(Manager (DHT       ip)) -- ^ RPC manager;
  , routingTable  :: !(MVar    (Table     ip)) -- ^ search table;
  , contactInfo   :: !(TVar    (PeerStore ip)) -- ^ published by other nodes;
  , announceInfo  :: !(TVar     AnnounceSet  ) -- ^ to publish by this node;
  , sessionTokens :: !(TVar     SessionTokens) -- ^ query session IDs.
  , loggerFun     :: !LogFun
  }

-- | DHT keep track current session and proper resource allocation for
-- safe multithreading.
newtype DHT ip a = DHT { unDHT :: ReaderT (Node ip) IO a }
  deriving ( Functor, Applicative, Monad
           , MonadIO, MonadBase IO
           , MonadReader (Node ip)
           , MonadThrow, MonadUnsafeIO
           )

instance MonadBaseControl IO (DHT ip) where
  newtype StM (DHT ip) a = StM {
      unSt :: StM (ReaderT (Node ip) IO) a
    }
  liftBaseWith cc = DHT $ liftBaseWith $ \ cc' ->
      cc $ \ (DHT m) -> StM <$> cc' m
  {-# INLINE liftBaseWith #-}

  restoreM = DHT . restoreM . unSt
  {-# INLINE restoreM #-}

-- | Check is it is possible to run 'queryNode' or handle pending
-- query from remote node.
instance MonadActive (DHT ip) where
  monadActive = getManager >>= liftIO . isActive
  {-# INLINE monadActive #-}

-- | All allocated resources will be closed at 'stopNode'.
instance MonadResource (DHT ip) where
  liftResourceT m = do
    s <- asks resources
    liftIO $ runInternalState m s

instance MonadKRPC (DHT ip) (DHT ip) where
  getManager = asks manager

instance MonadLogger (DHT ip) where
  monadLoggerLog loc src lvl msg = do
    logger <- asks loggerFun
    liftIO $ logger loc src lvl (toLogStr msg)

type NodeHandler ip = Handler (DHT ip)

-- | Run DHT session. You /must/ properly close session using
-- 'stopNode' function, otherwise socket or other scarce resources may
-- leak.
newNode :: Address ip
          => [NodeHandler ip] -- ^ handlers to run on accepted queries;
          -> Options          -- ^ various dht options;
          -> NodeAddr ip      -- ^ node address to bind;
          -> LogFun           -- ^
          -> IO (Node ip)     -- ^ a new DHT node running at given address.
newNode hs opts naddr logger = do
    s <- createInternalState
    runInternalState initNode s
      `onException` closeInternalState s
  where
    rpcOpts  = KRPC.def { optQueryTimeout = seconds (optTimeout opts) }
    nodeAddr = toSockAddr naddr
    initNode = do
      s      <- getInternalState
      (_, m) <- allocate (newManager rpcOpts nodeAddr hs) closeManager
      liftIO $ do
        myId   <- genNodeId
        node   <- Node opts myId s m
                <$> newMVar (nullTable myId (optBucketCount opts))
                <*> newTVarIO def
                <*> newTVarIO S.empty
                <*> (newTVarIO =<< nullSessionTokens)
                <*> pure logger
        runReaderT (unDHT listen) node
        return node

-- | Some resources like listener thread may live for
-- some short period of time right after this DHT session closed.
closeNode :: Node ip -> IO ()
closeNode Node {..} = closeInternalState resources

-- | Run DHT operation on the given session.
runDHT :: Node ip -> DHT ip a -> IO a
runDHT node action = runReaderT (unDHT action) node
{-# INLINE runDHT #-}

{-----------------------------------------------------------------------
--  Routing
-----------------------------------------------------------------------}

routing :: Address ip => Routing ip a -> DHT ip (Maybe a)
routing = runRouting probeNode refreshNodes getTimestamp

probeNode :: Address ip => NodeAddr ip -> DHT ip Bool
probeNode addr = do
  $(logDebugS) "routing.questionable_node" (T.pack (render (pretty addr)))
  result <- try $ Ping <@> addr
  let _ = result :: Either SomeException Ping
  return $ either (const False) (const True) result

-- /pick a random ID/ in the range of the bucket and perform a
-- find_nodes search on it.

-- FIXME do not use getClosest sinse we should /refresh/ them
refreshNodes :: Address ip => NodeId -> DHT ip [NodeInfo ip]
refreshNodes nid = do
  $(logDebugS) "routing.refresh_bucket" (T.pack (render (pretty nid)))
  nodes <- getClosest nid
  nss <- forM (nodeAddr <$> nodes) $ \ addr -> do
    NodeFound ns <- FindNode nid <@> addr
    return ns
  return $ L.concat nss

getTimestamp :: DHT ip Timestamp
getTimestamp = do
  utcTime <- liftIO $ getCurrentTime
  $(logDebugS) "routing.make_timestamp" (T.pack (render (pretty utcTime)))
  return $ utcTimeToPOSIXSeconds utcTime

{-----------------------------------------------------------------------
-- Tokens
-----------------------------------------------------------------------}

tryUpdateSecret :: DHT ip ()
tryUpdateSecret = do
  curTime <- liftIO getCurrentTime
  toks    <- asks sessionTokens
  liftIO $ atomically $ modifyTVar' toks (invalidateTokens curTime)

grantToken :: Hashable a => NodeAddr a -> DHT ip Token
grantToken addr = do
  tryUpdateSecret
  toks <- asks sessionTokens >>= liftIO . readTVarIO
  return $ T.lookup addr $ tokenMap toks

-- | Throws 'HandlerError' if the token is invalid or already
-- expired. See 'TokenMap' for details.
checkToken :: Hashable a => NodeAddr a -> Token -> DHT ip Bool
checkToken addr questionableToken = do
  tryUpdateSecret
  toks <- asks sessionTokens >>= liftIO . readTVarIO
  return $ T.member addr questionableToken (tokenMap toks)

{-----------------------------------------------------------------------
-- Routing table
-----------------------------------------------------------------------}

-- | Get current routing table. Normally you don't need to use this
-- function, but it can be usefull for debugging and profiling purposes.
getTable :: DHT ip (Table ip)
getTable = do
  var <- asks routingTable
  liftIO (readMVar var)

-- | Find a set of closest nodes from routing table of this node. (in
-- no particular order)
--
--   This operation used for 'find_nodes' query.
--
getClosest :: Eq ip => TableKey k => k -> DHT ip [NodeInfo ip]
getClosest node = do
  k <- asks (optK . options)
  kclosest k node <$> getTable

-- | This operation do not block but acquire exclusive access to
--   routing table.
insertNode :: Address ip => NodeInfo ip -> DHT ip ThreadId
insertNode info = fork $ do
  var <- asks routingTable
  modifyMVar_ var $ \ t -> do
    result <- routing (R.insert info t)
    case result of
      Nothing -> do
        $(logDebugS) "insertNode" $ "Routing table is full: "
                   <> T.pack (show (pretty t))
        return t
      Just t' -> do
        let logMsg = "Routing table updated: "
                  <> pretty t <> " -> " <> pretty t'
        $(logDebugS) "insertNode" (T.pack (render logMsg))
        return t'

{-----------------------------------------------------------------------
-- Peer storage
-----------------------------------------------------------------------}

-- TODO limit dht peer store in size (probably by removing oldest peers)

refreshContacts :: DHT ip ()
refreshContacts = undefined

-- | Insert peer to peer store. Used to handle announce requests.
insertPeer :: Eq ip => InfoHash -> PeerAddr ip -> DHT ip ()
insertPeer ih addr = do
  refreshContacts
  var <- asks contactInfo
  liftIO $ atomically $ modifyTVar' var (P.insert ih addr)

-- | Get peer set for specific swarm.
lookupPeers :: InfoHash -> DHT ip [PeerAddr ip]
lookupPeers ih = do
  refreshContacts
  var <- asks contactInfo
  liftIO $ P.lookup ih <$> readTVarIO var

-- | Prepare result for 'get_peers' query.
--
--   This operation use 'getClosest' as failback so it may block.
--
getPeerList :: Eq ip => InfoHash -> DHT ip (PeerList ip)
getPeerList ih = do
  ps <- lookupPeers ih
  if L.null ps
    then Left <$> getClosest ih
    else return (Right ps)

insertTopic :: InfoHash -> PortNumber -> DHT ip ()
insertTopic ih p = do
  var <- asks announceInfo
  liftIO $ atomically $ modifyTVar' var (S.insert (ih, p))

deleteTopic :: InfoHash -> PortNumber -> DHT ip ()
deleteTopic ih p = do
  var <- asks announceInfo
  liftIO $ atomically $ modifyTVar' var (S.delete (ih, p))

{-----------------------------------------------------------------------
-- Messaging
-----------------------------------------------------------------------}

-- | Throws exception if node is not responding.
queryNode :: forall a b ip. Address ip => KRPC (Query a) (Response b)
          => NodeAddr ip -> a -> DHT ip (NodeId, b)
queryNode addr q = do
  nid <- asks thisNodeId
  Response remoteId r <- query (toSockAddr addr) (Query nid q)
  insertNode (NodeInfo remoteId addr)
  return (remoteId, r)

-- | Infix version of 'queryNode' function.
(<@>) :: Address ip => KRPC (Query a) (Response b)
      => a -> NodeAddr ip -> DHT ip b
q <@> addr = snd <$> queryNode addr q
{-# INLINE (<@>) #-}

-- TODO: use alpha
-- | Failed queries are ignored.
queryParallel :: [DHT ip a] -> DHT ip [a]
queryParallel queries = do
    alpha <- asks (optAlpha . options)
    cleanup <$> mapConcurrently try queries
  where
    cleanup :: [Either QueryFailure a] -> [a]
    cleanup = mapMaybe (either (const Nothing) Just)