{-# LINE 1 "src/System/Socket.hsc" #-} {-# LANGUAGE TypeFamilies, FlexibleContexts, ScopedTypeVariables #-} {-# LINE 2 "src/System/Socket.hsc" #-} -------------------------------------------------------------------------------- -- | -- Module : System.Socket -- Copyright : (c) Lars Petersen 2015 -- License : MIT -- -- Maintainer : info@lars-petersen.net -- Stability : experimental -- -- > {-# LANGUAGE OverloadedStrings #-} -- > module Main where -- > -- > import Control.Exception ( bracket, catch ) -- > import Control.Monad ( forever ) -- > -- > import System.Socket -- > import System.Socket.Family.Inet6 -- > import System.Socket.Type.Stream -- > import System.Socket.Protocol.TCP -- > -- > main :: IO () -- > main = bracket -- > ( socket :: IO (Socket Inet6 Stream TCP) ) -- > ( \s-> do -- > close s -- > putStrLn "Listening socket closed." -- > ) -- > ( \s-> do -- > setSocketOption s (ReuseAddress True) -- > setSocketOption s (V6Only False) -- > bind s (SocketAddressInet6 inet6Any 8080 0 0) -- > listen s 5 -- > putStrLn "Listening socket ready..." -- > forever $ acceptAndHandle s `catch` \e-> print (e :: SocketException) -- > ) -- > -- > acceptAndHandle :: Socket Inet6 Stream TCP -> IO () -- > acceptAndHandle s = bracket -- > ( accept s ) -- > ( \(p, addr)-> do -- > close p -- > putStrLn $ "Closed connection to " ++ show addr -- > ) -- > ( \(p, addr)-> do -- > putStrLn $ "Accepted connection from " ++ show addr -- > sendAll p "Hello world!" msgNoSignal -- > ) -------------------------------------------------------------------------------- module System.Socket ( -- * Socket Socket () , SocketAddress () -- ** Family , Family (..) -- ** Type , Type (..) -- ** Protocol , Protocol (..) -- * Operations -- ** socket , socket -- ** connect , connect -- ** bind , bind -- ** listen , listen -- ** accept , accept -- ** send, sendTo , send, sendTo -- ** receive, receiveFrom , receive, receiveFrom -- ** close , close -- * Options , SocketOption (..) , Error (..) , ReuseAddress (..) -- * Name Resolution -- ** getAddressInfo , AddressInfo (..) , HasAddressInfo (..) -- ** getNameInfo , NameInfo (..) , HasNameInfo (..) -- * Flags -- ** MessageFlags , MessageFlags () , msgEndOfRecord , msgNoSignal , msgOutOfBand , msgWaitAll -- ** AddressInfoFlags , AddressInfoFlags () , aiAddressConfig , aiAll , aiCanonicalName , aiNumericHost , aiNumericService , aiPassive , aiV4Mapped -- ** NameInfoFlags , NameInfoFlags () , niNameRequired , niDatagram , niNoFullyQualifiedDomainName , niNumericHost , niNumericService -- * Exceptions -- ** SocketException , module System.Socket.Internal.Exception -- ** AddressInfoException , AddressInfoException (..) , eaiAgain , eaiBadFlags , eaiFail , eaiFamily , eaiMemory , eaiNoName , eaiSocketType , eaiService , eaiSystem ) where import Control.Exception import Control.Monad import Control.Applicative import Control.Concurrent import Data.Function import qualified Data.ByteString as BS import qualified Data.ByteString.Unsafe as BS import GHC.Conc (closeFdWith) import Foreign.Ptr import Foreign.Storable import Foreign.Marshal.Alloc import System.Socket.Unsafe import System.Socket.Internal.Socket import System.Socket.Internal.Exception import System.Socket.Internal.Message import System.Socket.Internal.AddressInfo import System.Socket.Internal.Platform {-# LINE 151 "src/System/Socket.hsc" #-} -- | Creates a new socket. -- -- Whereas the underlying POSIX socket operation takes 3 parameters, this library -- encodes this information in the type variables. This rules out several -- kinds of errors and escpecially simplifies the handling of addresses (by using -- associated type families). Examples: -- -- > -- create a IPv4-UDP-datagram socket -- > sock <- socket :: IO (Socket Inet Datagram UDP) -- > -- create a IPv6-TCP-streaming socket -- > sock6 <- socket :: IO (Socket Inet6 Stream TCP) -- -- - This operation sets up a finalizer that automatically closes the socket -- when the garbage collection decides to collect it. This is just a -- fail-safe. You might still run out of file descriptors as there's -- no guarantee about when the finalizer is run. You're advised to -- manually `close` the socket when it's no longer needed. -- If possible, use `Control.Exception.bracket` to reliably close the -- socket descriptor on exception or regular termination of your -- computation: -- -- > result <- bracket (socket :: IO (Socket Inet6 Stream TCP)) close $ \sock-> do -- > somethingWith sock -- your computation here -- > return somethingelse -- -- -- - This operation configures the socket non-blocking to work seamlessly -- with the runtime system's event notification mechanism. -- - This operation can safely deal with asynchronous exceptions without -- leaking file descriptors. -- - This operation throws `SocketException`s. Consult your @man@ page for -- details and specific @errno@s. socket :: (Family f, Type t, Protocol p) => IO (Socket f t p) socket = socket' where socket' :: forall f t p. (Family f, Type t, Protocol p) => IO (Socket f t p) socket' = do bracketOnError -- Try to acquire the socket resource. This part has exceptions masked. ( c_socket (familyNumber (undefined :: f)) (typeNumber (undefined :: t)) (protocolNumber (undefined :: p)) ) -- On failure after the c_socket call we try to close the socket to not leak file descriptors. -- If closing fails we cannot really do something about it. We tried at least. -- This part has exceptions masked as well. c_close is an unsafe FFI call. ( \fd-> when (fd >= 0) (c_close fd >> return ()) ) -- If an exception is raised, it is reraised after the socket has been closed. -- This part has async exceptions unmasked (via restore). ( \fd-> if fd < 0 then do c_get_last_socket_error >>= throwIO else do -- setNonBlockingFD calls c_fcntl_write which is an unsafe FFI call. i <- c_setnonblocking fd if i < 0 then do c_get_last_socket_error >>= throwIO else do mfd <- newMVar fd let s = Socket mfd _ <- mkWeakMVar mfd (close s) return s ) -- | Connects to an remote address. -- -- - Calling `connect` on a `close`d socket throws `eBadFileDescriptor` even if the former file descriptor has been reassigned. -- - This operation returns as soon as a connection has been established (as -- if the socket were blocking). The connection attempt has either failed or -- succeeded after this operation threw an exception or returned. -- - The operation might throw `SocketException`s. Due to implementation quirks -- the socket should be considered in an undefined state when this operation -- failed. It should be closed then. -- - Also see [these considerations](http://cr.yp.to/docs/connect.html) on -- the problems with connecting non-blocking sockets. connect :: (Family f, Storable (SocketAddress f)) => Socket f t p -> SocketAddress f -> IO () connect (Socket mfd) addr = do alloca $ \addrPtr-> do poke addrPtr addr let addrLen = fromIntegral (sizeOf addr) mwait <- withMVar mfd $ \fd-> do when (fd < 0) (throwIO eBadFileDescriptor) -- The actual connection attempt. i <- c_connect fd addrPtr addrLen -- On non-blocking sockets we expect to get EINPROGRESS or EWOULDBLOCK. if i < 0 then do e <- c_get_last_socket_error if e == eInProgress || e == eWouldBlock || e == eInterrupted then do -- The manpage says that in this case the connection -- shall be established asynchronously and one is -- supposed to wait. wait <- unsafeSocketWaitWrite fd 0 return (Just wait) else do throwIO e else do -- This should not be the case on non-blocking socket, but better safe than sorry. return Nothing case mwait of Nothing -> do -- The connection could be established synchronously. Nothing else to do. return () Just wait -> do -- This either waits or does nothing. wait -- By here we don't know anything about the current connection status. -- It might either have succeeded, failed or is still undecided. -- The approach is to try a second connection attempt, because its error -- codes allow us to distinguish all three potential states. ( fix $ \again iteration-> do mwait' <- withMVar mfd $ \fd-> do i <- c_connect fd addrPtr addrLen if i < 0 then do e <- c_get_last_socket_error if e == eIsConnected then do -- This is what we want. The connection is established. return Nothing else if e == eAlready then do -- The previous connection attempt is still pending. Just Control.Applicative.<$> unsafeSocketWaitWrite fd iteration else do -- The previous connection failed (results in EINPROGRESS or -- EWOULBLOCK here) or something else is wrong. -- We throw eTimedOut here as we don't know and will never -- know the exact reason (better suggestions appreciated). throwIO eTimedOut else do -- This means the last connection attempt succeeded immediately. -- Linux does this when connecting to the same address when the -- unsafeSocketWaitWrite call signals writeability. return Nothing case mwait' of Nothing -> do return () Just wait' -> do wait' again $! iteration + 1 ) 1 -- | Bind a socket to an address. -- -- - Calling `bind` on a `close`d socket throws `eBadFileDescriptor` even if the former file descriptor has been reassigned. -- - It is assumed that `c_bind` never blocks and therefore `eInProgress`, `eAlready` and `eInterrupted` don't occur. -- This assumption is supported by the fact that the Linux manpage doesn't mention any of these errors, -- the Posix manpage doesn't mention the last one and even MacOS' implementation will never -- fail with any of these when the socket is configured non-blocking as -- [argued here](http://stackoverflow.com/a/14485305). -- - This operation throws `SocketException`s. Consult your @man@ page for -- details and specific @errno@s. bind :: (Family f, Storable (SocketAddress f)) => Socket f t p -> SocketAddress f -> IO () bind (Socket mfd) addr = do alloca $ \addrPtr-> do poke addrPtr addr withMVar mfd $ \fd-> do i <- c_bind fd addrPtr (fromIntegral $ sizeOf addr) if i < 0 then c_get_last_socket_error >>= throwIO else return () -- | Starts listening and queueing connection requests on a connection-mode -- socket. -- -- - Calling `listen` on a `close`d socket throws `eBadFileDescriptor` even if the former -- file descriptor has been reassigned. -- - The second parameter is called /backlog/ and sets a limit on how many -- unaccepted connections the socket implementation shall queue. A value -- of @0@ leaves the decision to the implementation. -- - This operation throws `SocketException`s. Consult your @man@ page for -- details and specific @errno@s. listen :: Socket f t p -> Int -> IO () listen (Socket ms) backlog = do i <- withMVar ms $ \s-> do c_listen s (fromIntegral backlog) if i < 0 then do c_get_last_socket_error >>= throwIO else do return () -- | Accept a new connection. -- -- - Calling `accept` on a `close`d socket throws `eBadFileDescriptor` even if the former -- file descriptor has been reassigned. -- - This operation configures the new socket non-blocking (TODO: use `accept4` if available). -- - This operation sets up a finalizer for the new socket that automatically -- closes the new socket when the garbage collection decides to collect it. -- This is just a fail-safe. You might still run out of file descriptors as -- there's no guarantee about when the finalizer is run. You're advised to -- manually `close` the socket when it's no longer needed. -- - This operation throws `SocketException`s. Consult your @man@ page for -- details and specific @errno@s. -- - This operation catches `eAgain`, `eWouldBlock` and `eInterrupted` internally -- and retries automatically. accept :: (Family f, Storable (SocketAddress f)) => Socket f t p -> IO (Socket f t p, SocketAddress f) accept s@(Socket mfd) = accept' where accept' :: forall f t p. (Family f, Storable (SocketAddress f)) => IO (Socket f t p, SocketAddress f) accept' = do -- Allocate local (!) memory for the address. alloca $ \addrPtr-> do alloca $ \addrPtrLen-> do poke addrPtrLen (fromIntegral $ sizeOf (undefined :: SocketAddress f)) ( fix $ \again iteration-> do -- We mask asynchronous exceptions during this critical section. ews <- withMVarMasked mfd $ \fd-> do fix $ \retry-> do ft <- c_accept fd addrPtr addrPtrLen if ft < 0 then do e <- c_get_last_socket_error if e == eWouldBlock || e == eAgain then do unsafeSocketWaitRead fd iteration >>= return . Left else if e == eInterrupted -- On EINTR it is good practice to just retry. then retry else throwIO e -- This is the critical section: We got a valid descriptor we have not yet returned. else do i <- c_setnonblocking ft if i < 0 then do c_get_last_socket_error >>= throwIO else do -- This peek operation might be a little expensive, but I don't see an alternative. addr <- peek addrPtr :: IO (SocketAddress f) -- newMVar is guaranteed to be not interruptible. mft <- newMVar ft -- Register a finalizer on the new socket. _ <- mkWeakMVar mft (close (Socket mft `asTypeOf` s)) return (Right (Socket mft, addr)) -- If ews is Left we got EAGAIN or EWOULDBLOCK and retry after the next event. case ews of Left wait -> wait >> (again $! iteration + 1) Right sock -> return sock ) 0 -- This is the initial iteration value. -- | Send a message on a connected socket. -- -- - Calling `send` on a `close`d socket throws `eBadFileDescriptor` even if the former -- file descriptor has been reassigned. -- - The operation returns the number of bytes sent. On `Datagram` and -- `SequentialPacket` sockets certain assurances on atomicity exist and `eAgain` or -- `eWouldBlock` are returned until the whole message would fit -- into the send buffer. -- - This operation throws `SocketException`s. Consult @man 3p send@ for -- details and specific @errno@s. -- - `eAgain`, `eWouldBlock` and `eInterrupted` and handled internally and won't -- be thrown. For performance reasons the operation first tries a write -- on the socket and then waits when it got `eAgain` or `eWouldBlock`. send :: Socket f t p -> BS.ByteString -> MessageFlags -> IO Int send s bs flags = do bytesSent <- BS.unsafeUseAsCStringLen bs $ \(bufPtr,bufSize)-> unsafeSend s (castPtr bufPtr) (fromIntegral bufSize) flags return (fromIntegral bytesSent) -- | Like `send`, but allows to specify a destination address. sendTo ::(Family f, Storable (SocketAddress f)) => Socket f t p -> BS.ByteString -> MessageFlags -> SocketAddress f -> IO Int sendTo s bs flags addr = do bytesSent <- alloca $ \addrPtr-> do poke addrPtr addr BS.unsafeUseAsCStringLen bs $ \(bufPtr,bufSize)-> unsafeSendTo s bufPtr (fromIntegral bufSize) flags addrPtr (fromIntegral $ sizeOf addr) return (fromIntegral bytesSent) -- | Receive a message on a connected socket. -- -- - Calling `receive` on a `close`d socket throws `eBadFileDescriptor` even if the former file descriptor has been reassigned. -- - The operation takes a buffer size in bytes a first parameter which -- limits the maximum length of the returned `Data.ByteString.ByteString`. -- - This operation throws `SocketException`s. Consult @man 3p receive@ for -- details and specific @errno@s. -- - `eAgain`, `eWouldBlock` and `eInterrupted` and handled internally and won't be thrown. -- For performance reasons the operation first tries a read -- on the socket and then waits when it got `eAgain` or `eWouldBlock`. receive :: Socket f t p -> Int -> MessageFlags -> IO BS.ByteString receive s bufSize flags = bracketOnError ( mallocBytes bufSize ) (\bufPtr-> free bufPtr ) (\bufPtr-> do bytesReceived <- unsafeReceive s bufPtr (fromIntegral bufSize) flags BS.unsafePackMallocCStringLen (bufPtr, fromIntegral bytesReceived) ) -- | Like `receive`, but additionally yields the peer address. receiveFrom :: (Family f, Storable (SocketAddress f)) => Socket f t p -> Int -> MessageFlags -> IO (BS.ByteString, SocketAddress f) receiveFrom = receiveFrom' where receiveFrom' :: forall f t p. (Family f, Storable (SocketAddress f)) => Socket f t p -> Int -> MessageFlags -> IO (BS.ByteString, SocketAddress f) receiveFrom' s bufSize flags = do alloca $ \addrPtr-> do alloca $ \addrSizePtr-> do poke addrSizePtr (fromIntegral $ sizeOf (undefined :: SocketAddress f)) bracketOnError ( mallocBytes bufSize ) (\bufPtr-> free bufPtr ) (\bufPtr-> do bytesReceived <- unsafeReceiveFrom s bufPtr (fromIntegral bufSize) flags addrPtr addrSizePtr addr <- peek addrPtr bs <- BS.unsafePackMallocCStringLen (bufPtr, fromIntegral bytesReceived) return (bs, addr) ) -- | Closes a socket. -- -- - This operation is idempotent and thus can be performed more than once without throwing an exception. -- If it throws an exception it is presumably a not recoverable situation and the process should exit. -- - This operation does not block. -- - This operation wakes up all threads that are currently blocking on this -- socket. All other threads are guaranteed not to block on operations on this socket in the future. -- Threads that perform operations other than `close` on this socket will fail with `eBadFileDescriptor` -- after the socket has been closed (`close` replaces the -- `System.Posix.Types.Fd` in the `Control.Concurrent.MVar.MVar` with @-1@ -- to reliably avoid use-after-free situations). -- - This operation potentially throws `SocketException`s (only @EIO@ is -- documented). `eInterrupted` is catched internally and retried automatically, so won't be thrown. close :: Socket f t p -> IO () close (Socket mfd) = do modifyMVarMasked_ mfd $ \fd-> do if fd < 0 then do return fd else do -- closeFdWith does not throw even on invalid file descriptors. -- It just assures no thread is blocking on the fd anymore and then executes the IO action. closeFdWith -- The c_close operation may (according to Posix documentation) fails with EINTR or EBADF or EIO. -- EBADF: Should be ruled out by the library's design. -- EINTR: It is best practice to just retry the operation what we do here. -- EIO: Only occurs when filesystem is involved (?). -- Conclusion: Our close should never fail. If it does, something is horribly wrong. ( const $ fix $ \retry-> do i <- c_close fd if i < 0 then do e <- c_get_last_socket_error if e == eInterrupted then retry else throwIO e else return () ) fd -- When we arrive here, no exception has been thrown and the descriptor has been closed. -- We put an invalid file descriptor into the MVar. return (-1)