-- -- HTTP client for use with io-streams -- -- Copyright © 2012-2013 Operational Dynamics Consulting, Pty Ltd -- -- The code in this file, and the program it is a part of, is -- made available to you by its authors as open source software: -- you can redistribute it and/or modify it under the terms of -- the BSD licence. -- {-# LANGUAGE DeriveDataTypeable #-} {-# LANGUAGE DoAndIfThenElse #-} {-# LANGUAGE OverloadedStrings #-} module Network.Http.Connection ( Hostname, Port, Connection(..), -- constructors only for testing makeConnection, withConnection, openConnection, openConnectionSSL, closeConnection, getHostname, sendRequest, receiveResponse, emptyBody, fileBody, inputStreamBody, debugHandler, concatHandler ) where import Blaze.ByteString.Builder (Builder) import qualified Blaze.ByteString.Builder as Builder (flush, fromByteString, toByteString) import qualified Blaze.ByteString.Builder.HTTP as Builder (chunkedTransferEncoding, chunkedTransferTerminator) import Control.Exception (bracket) import Data.ByteString (ByteString) import qualified Data.ByteString.Char8 as S import Data.Monoid (mappend, mempty) import Network.Socket import OpenSSL.Session (SSL, SSLContext) import qualified OpenSSL.Session as SSL import System.IO.Streams (InputStream, OutputStream, stdout) import qualified System.IO.Streams as Streams import qualified System.IO.Streams.SSL as Streams hiding (connect) import Network.Http.ResponseParser import Network.Http.Types {- This is a String because that's what the uri package works in. There was a fairly detailed disucssion on haskell-cafe about this, with the conclusion that URLs are composed of characters, not octets. -} type Hostname = String type Port = Int -- | A connection to a web server. -- data Connection = Connection { cHost :: ByteString, -- ^ will be used as the Host: header in the HTTP request. cClose :: IO (), -- ^ called when the connection should be closed. cOut :: OutputStream Builder, cIn :: InputStream ByteString } instance Show Connection where show c = {-# SCC "Connection.show" #-} concat ["Host: ", S.unpack $ cHost c, "\n"] -- -- | Create a raw Connection object from the given parts. This is -- primarily of use when teseting, for example: -- -- > fakeConnection :: IO Connection -- > fakeConnection = do -- > o <- Streams.nullOutput -- > i <- Streams.nullInput -- > c <- makeConnection "www.example.com" (return()) o i -- > return c -- -- is an idiom we use frequently in testing and benchmarking, usually -- replacing the InputStream with something like: -- -- > x' <- S.readFile "properly-formatted-response.txt" -- > i <- Streams.fromByteString x' -- -- If you're going to do that, keep in mind that you /must/ have CR-LF -- pairs after each header line and between the header and body to -- be compliant with the HTTP protocol; otherwise, parsers will -- reject your message. -- makeConnection :: ByteString -- ^ will be used as the @Host:@ header in the HTTP request. -> IO () -- ^ an action to be called when the connection is terminated. -> OutputStream ByteString -- ^ write end of the HTTP client-server connection. -> InputStream ByteString -- ^ read end of the HTTP client-server connection. -> IO Connection makeConnection h c o1 i = do o2 <- Streams.builderStream o1 return $! Connection h c o2 i -- -- | Given an @IO@ action producing a 'Connection', and a computation -- that needs one, runs the computation, cleaning up the -- @Connection@ afterwards. -- -- > x <- withConnection (openConnection "s3.example.com" 80) $ (\c -> do -- > q <- buildRequest $ do -- > http GET "/bucket42/object/149" -- > sendRequest c q emptyBody -- > ... -- > return "blah") -- -- which can make the code making an HTTP request a lot more -- straight-forward. -- -- Wraps @Control.Exception@'s 'Control.Exception.bracket'. -- withConnection :: IO Connection -> (Connection -> IO γ) -> IO γ withConnection mkC = bracket mkC closeConnection -- -- | In order to make a request you first establish the TCP -- connection to the server over which to send it. -- -- Ordinarily you would supply the host part of the URL here and it will -- be used as the value of the HTTP 1.1 @Host:@ field. However, you can -- specify any server name or IP addresss and set the @Host:@ value -- later with 'Network.Http.Client.setHostname' when building the -- request. -- -- Usage is as follows: -- -- > c <- openConnection "localhost" 80 -- > ... -- > closeConnection c -- -- More likely, you'll use 'withConnection' to wrap the call in order -- to ensure finalization. -- -- HTTP pipelining is supported; you can reuse the connection to a -- web server, but it's up to you to ensure you match the number of -- requests sent to the number of responses read, and to process those -- responses in order. This is all assuming that the /server/ supports -- pipelining; be warned that not all do. Web browsers go to -- extraordinary lengths to probe this; you probably only want to do -- pipelining under controlled conditions. Otherwise just open a new -- connection for subsequent requests. -- openConnection :: Hostname -> Port -> IO Connection openConnection h p = do is <- getAddrInfo (Just hints) (Just h) (Just $ show p) let addr = head is let a = addrAddress addr s <- socket (addrFamily addr) Stream defaultProtocol connect s a (i,o1) <- Streams.socketToStreams s o2 <- Streams.builderStream o1 return Connection { cHost = h', cClose = close s, cOut = o2, cIn = i } where hints = defaultHints {addrFlags = [AI_ADDRCONFIG, AI_NUMERICSERV]} h' :: ByteString h' = if p == 80 then S.pack h else S.concat [ S.pack h, ":", S.pack $ show p ] -- -- | Open a secure connection to a web server. -- -- You need to wrap this (and subsequent code using this connection) -- within a call to 'OpenSSL.withOpenSSL': -- -- > import OpenSSL (withOpenSSL) -- > -- > main :: IO () -- > main = withOpenSSL $ do -- > ctx <- baselineContextSSL -- > c <- openConnectionSSL ctx "api.github.com" 443 -- > ... -- > closeConnection c -- -- If you want to tune the parameters used in making SSL connections, -- manually specify certificates, etc, then setup your own context: -- -- > import OpenSSL.Session (SSLContext) -- > import qualified OpenSSL.Session as SSL -- > -- > ... -- > ctx <- SSL.context -- > ... -- -- See "OpenSSL.Session". -- -- Crypto is as provided by the system @openssl@ library, as wrapped -- by the @HsOpenSSL@ package and @openssl-streams@. -- openConnectionSSL :: SSLContext -> Hostname -> Port -> IO Connection openConnectionSSL ctx h p = do s <- socket AF_INET Stream defaultProtocol is <- getAddrInfo Nothing (Just h) (Just $ show p) let a = addrAddress $ head is connect s a ssl <- SSL.connection ctx s SSL.connect ssl (i,o1) <- Streams.sslToStreams ssl o2 <- Streams.builderStream o1 return Connection { cHost = h', cClose = closeSSL s ssl, cOut = o2, cIn = i } where h' :: ByteString h' = if p == 443 then S.pack h else S.concat [ S.pack h, ":", S.pack $ show p ] closeSSL :: Socket -> SSL -> IO () closeSSL s ssl = do SSL.shutdown ssl SSL.Unidirectional close s -- -- | Having composed a 'Request' object with the headers and metadata for -- this connection, you can now send the request to the server, along -- with the entity body, if there is one. For the rather common case of -- HTTP requests like 'GET' that don't send data, use 'emptyBody' as the -- output stream: -- -- > sendRequest c q emptyBody -- -- For 'PUT' and 'POST' requests, you can use 'fileBody' or -- 'inputStreamBody' to send content to the server, or you can work with -- the @io-streams@ API directly: -- -- > sendRequest c q (\o -> -- > Streams.write (Just "Hello World\n") o) -- {- I would like to enforce the constraints on the Empty and Static cases shown here, but those functions take OutputStream ByteString, and we are of course working in OutputStream Builder by that point. -} sendRequest :: Connection -> Request -> (OutputStream Builder -> IO α) -> IO α sendRequest c q handler = do -- write the headers Streams.write (Just msg) o2 -- deal with the expect-continue mess e2 <- case t of Normal -> do return e Continue -> do Streams.write (Just Builder.flush) o2 p <- readResponseHeader i case getStatusCode p of 100 -> do -- ok to send return e _ -> do -- put the response back Streams.unRead (rsp p) i return Empty -- write the body, if there is one x <- case e2 of Empty -> do o3 <- Streams.nullOutput y <- handler o3 return y Chunking -> do o3 <- Streams.contramap Builder.chunkedTransferEncoding o2 y <- handler o3 Streams.write (Just Builder.chunkedTransferTerminator) o2 return y (Static _) -> do -- o3 <- Streams.giveBytes (fromIntegral n :: Int64) o2 y <- handler o2 return y -- push the stream out by flushing the output buffers Streams.write (Just Builder.flush) o2 return x where o2 = cOut c e = qBody q t = qExpect q msg = composeRequestBytes q h' h' = cHost c i = cIn c rsp p = Builder.toByteString $ composeResponseBytes p -- -- | Get the virtual hostname that will be used as the @Host:@ header in -- the HTTP 1.1 request. Per RFC 2616 § 14.23, this will be of the form -- @hostname:port@ if the port number is other than the default, ie 80 -- for HTTP. -- getHostname :: Connection -> Request -> ByteString getHostname c q = case qHost q of Just h' -> h' Nothing -> cHost c -- -- | Handle the response coming back from the server. This function -- hands control to a handler function you supply, passing you the -- 'Response' object with the response headers and an 'InputStream' -- containing the entity body. -- -- For example, if you just wanted to print the first chunk of the -- content from the server: -- -- > receiveResponse c (\p i -> do -- > m <- Streams.read b -- > case m of -- > Just bytes -> putStr bytes -- > Nothing -> return ()) -- -- Obviously, you can do more sophisticated things with the -- 'InputStream', which is the whole point of having an @io-streams@ -- based HTTP client library. -- -- The final value from the handler function. is the return value of -- @receiveResponse@, if you need it. -- {- The reponse body coming from the server MUST be fully read, even if (especially if) the users's handler doesn't consume it all. This is necessary to maintain the HTTP protocol invariants; otherwise pipelining would not work. It's not entirely clear *which* InputStream is being drained here; the underlying InputStream ByteString in Connection remains unconsumed beyond the threshold of the current response, which is exactly what we need. -} receiveResponse :: Connection -> (Response -> InputStream ByteString -> IO β) -> IO β receiveResponse c handler = do p <- readResponseHeader i i' <- readResponseBody p i x <- handler p i' Streams.skipToEof i' return x where i = cIn c -- -- | Use this for the common case of the HTTP methods that only send -- headers and which have no entity body, i.e. 'GET' requests. -- emptyBody :: OutputStream Builder -> IO () emptyBody _ = return () -- -- | Specify a local file to be sent to the server as the body of the -- request. -- -- You use this partially applied: -- -- > sendRequest c q (fileBody "/etc/passwd") -- -- Note that the type of @(fileBody \"\/path\/to\/file\")@ is just what -- you need for the third argument to 'sendRequest', namely -- -- >>> :t filePath "hello.txt" -- :: OutputStream Builder -> IO () -- {- Relies on Streams.withFileAsInput generating (very) large chunks [which it does]. A more efficient way to do this would be interesting. -} fileBody :: FilePath -> OutputStream Builder -> IO () fileBody p o = do Streams.withFileAsInput p (\i -> inputStreamBody i o) -- -- | Read from a pre-existing 'InputStream' and pipe that through to the -- connection to the server. This is useful in the general case where -- something else has handed you stream to read from and you want to use -- it as the entity body for the request. -- -- You use this partially applied: -- -- > i <- getStreamFromVault -- magic, clearly -- > sendRequest c q (inputStreamBody i) -- -- This function maps "Builder.fromByteString" over the input, which will -- be efficient if the ByteString chunks are large. -- {- Note that this has to be 'supply' and not 'connect' as we do not want the end of stream to prematurely terminate the chunked encoding pipeline! -} inputStreamBody :: InputStream ByteString -> OutputStream Builder -> IO () inputStreamBody i1 o = do i2 <- Streams.map Builder.fromByteString i1 Streams.supply i2 o -- -- | Print the response headers and response body to @stdout@. You can -- use this with 'receiveResponse' or one of the convenience functions -- when testing. For example, doing: -- -- > c <- openConnection "kernel.operationaldynamics.com" 58080 -- > -- > q <- buildRequest $ do -- > http GET "/time" -- > -- > sendRequest c q emptyBody -- > -- > receiveResponse c debugHandler -- -- would print out: -- -- > HTTP/1.1 200 OK -- > Transfer-Encoding: chunked -- > Content-Type: text/plain -- > Vary: Accept-Encoding -- > Server: Snap/0.9.2.4 -- > Content-Encoding: gzip -- > Date: Mon, 21 Jan 2013 06:13:37 GMT -- > -- > Mon 21 Jan 13, 06:13:37.303Z -- -- or thereabouts. -- debugHandler :: Response -> InputStream ByteString -> IO () debugHandler p i = do S.putStr $ S.filter (/= '\r') $ Builder.toByteString $ composeResponseBytes p Streams.connect i stdout -- -- | Sometimes you just want the entire response body as a single blob. -- This function concatonates all the bytes from the response into a -- ByteString. If using the main @http-streams@ API, you would use it -- as follows: -- -- > ... -- > x' <- receiveResponse c concatHandler -- > ... -- -- The methods in the convenience API all take a function to handle the -- response; this function is passed directly to the 'receiveResponse' -- call underlying the request. Thus this utility function can be used -- for 'get' as well: -- -- > x' <- get "http://www.example.com/document.txt" concatHandler -- -- Either way, the usual caveats about allocating a -- single object from streaming I/O apply: do not use this if you are -- not absolutely certain that the response body will fit in a -- reasonable amount of memory. -- -- Note that this function makes no discrimination based on the -- response's HTTP status code. You're almost certainly better off -- writing your own handler function. -- {- I'd welcome a better name for this function. -} concatHandler :: Response -> InputStream ByteString -> IO ByteString concatHandler _ i1 = do i2 <- Streams.map Builder.fromByteString i1 x <- Streams.fold mappend mempty i2 return $ Builder.toByteString x -- -- | Shutdown the connection. You need to call this release the -- underlying socket file descriptor and related network resources. To -- do so reliably, use this in conjunction with 'openConnection' in a -- call to 'Control.Exception.bracket': -- -- > -- -- > -- Make connection, cleaning up afterward -- > -- -- > -- > foo :: IO ByteString -- > foo = bracket -- > (openConnection "localhost" 80) -- > (closeConnection) -- > (doStuff) -- > -- > -- -- > -- Actually use Connection to send Request and receive Response -- > -- -- > -- > doStuff :: Connection -> IO ByteString -- -- or, just use 'withConnection'. -- -- While returning a ByteString is probably the most common use case, -- you could conceivably do more processing of the response in 'doStuff' -- and have it and 'foo' return a different type. -- closeConnection :: Connection -> IO () closeConnection c = cClose c