git-annex/Utility/Url.hs

791 lines
26 KiB
Haskell
Raw Normal View History

2011-08-20 20:11:42 +00:00
{- Url downloading.
-
- Copyright 2011-2023 Joey Hess <id@joeyh.name>
-
- License: BSD-2-clause
-}
{-# LANGUAGE OverloadedStrings #-}
{-# LANGUAGE RankNTypes #-}
2015-05-10 19:37:55 +00:00
{-# LANGUAGE FlexibleContexts #-}
{-# LANGUAGE BangPatterns #-}
{-# LANGUAGE CPP #-}
2011-08-20 20:11:42 +00:00
module Utility.Url (
newManager,
URLString,
UserAgent,
limit url downloads to whitelisted schemes Security fix! Allowing any schemes, particularly file: and possibly others like scp: allowed file exfiltration by anyone who had write access to the git repository, since they could add an annexed file using such an url, or using an url that redirected to such an url, and wait for the victim to get it into their repository and send them a copy. * Added annex.security.allowed-url-schemes setting, which defaults to only allowing http and https URLs. Note especially that file:/ is no longer enabled by default. * Removed annex.web-download-command, since its interface does not allow supporting annex.security.allowed-url-schemes across redirects. If you used this setting, you may want to instead use annex.web-options to pass options to curl. With annex.web-download-command removed, nearly all url accesses in git-annex are made via Utility.Url via http-client or curl. http-client only supports http and https, so no problem there. (Disabling one and not the other is not implemented.) Used curl --proto to limit the allowed url schemes. Note that this will cause git annex fsck --from web to mark files using a disallowed url scheme as not being present in the web. That seems acceptable; fsck --from web also does that when a web server is not available. youtube-dl already disabled file: itself (probably for similar reasons). The scheme check was also added to youtube-dl urls for completeness, although that check won't catch any redirects it might follow. But youtube-dl goes off and does its own thing with other protocols anyway, so that's fine. Special remotes that support other domain-specific url schemes are not affected by this change. In the bittorrent remote, aria2c can still download magnet: links. The download of the .torrent file is otherwise now limited by annex.security.allowed-url-schemes. This does not address any external special remotes that might download an url themselves. Current thinking is all external special remotes will need to be audited for this problem, although many of them will use http libraries that only support http and not curl's menagarie. The related problem of accessing private localhost and LAN urls is not addressed by this commit. This commit was sponsored by Brett Eisenberg on Patreon.
2018-06-15 20:52:24 +00:00
Scheme,
mkScheme,
allowedScheme,
UrlDownloader(..),
NonHttpUrlDownloader(..),
UrlOptions(..),
defUrlOptions,
mkUrlOptions,
check,
checkBoth,
exists,
UrlInfo(..),
getUrlInfo,
assumeUrlExists,
download,
downloadConduit,
sinkResponseFile,
downloadPartial,
parseURIPortable,
parseURIRelaxed,
matchStatusCodeException,
matchHttpExceptionContent,
BasicAuth(..),
GetBasicAuth,
noBasicAuth,
applyBasicAuth',
extractFromResourceT,
conduitUrlSchemes,
) where
2012-03-16 00:39:25 +00:00
import Common
import Utility.Debug
import Utility.Metered
import Network.HTTP.Client.Restricted
import Utility.IPAddress
import qualified Utility.RawFilePath as R
import Utility.Hash (IncrementalVerifier(..))
import Network.URI
import Network.HTTP.Types
import qualified Network.Connection as NC
import qualified Data.CaseInsensitive as CI
import qualified Data.ByteString as B
import qualified Data.ByteString.UTF8 as B8
import qualified Data.ByteString.Lazy as L
limit url downloads to whitelisted schemes Security fix! Allowing any schemes, particularly file: and possibly others like scp: allowed file exfiltration by anyone who had write access to the git repository, since they could add an annexed file using such an url, or using an url that redirected to such an url, and wait for the victim to get it into their repository and send them a copy. * Added annex.security.allowed-url-schemes setting, which defaults to only allowing http and https URLs. Note especially that file:/ is no longer enabled by default. * Removed annex.web-download-command, since its interface does not allow supporting annex.security.allowed-url-schemes across redirects. If you used this setting, you may want to instead use annex.web-options to pass options to curl. With annex.web-download-command removed, nearly all url accesses in git-annex are made via Utility.Url via http-client or curl. http-client only supports http and https, so no problem there. (Disabling one and not the other is not implemented.) Used curl --proto to limit the allowed url schemes. Note that this will cause git annex fsck --from web to mark files using a disallowed url scheme as not being present in the web. That seems acceptable; fsck --from web also does that when a web server is not available. youtube-dl already disabled file: itself (probably for similar reasons). The scheme check was also added to youtube-dl urls for completeness, although that check won't catch any redirects it might follow. But youtube-dl goes off and does its own thing with other protocols anyway, so that's fine. Special remotes that support other domain-specific url schemes are not affected by this change. In the bittorrent remote, aria2c can still download magnet: links. The download of the .torrent file is otherwise now limited by annex.security.allowed-url-schemes. This does not address any external special remotes that might download an url themselves. Current thinking is all external special remotes will need to be audited for this problem, although many of them will use http libraries that only support http and not curl's menagarie. The related problem of accessing private localhost and LAN urls is not addressed by this commit. This commit was sponsored by Brett Eisenberg on Patreon.
2018-06-15 20:52:24 +00:00
import qualified Data.Set as S
import Control.Exception (throwIO, evaluate)
import Control.Monad.Trans.Resource
import Control.Monad.IO.Class (MonadIO)
import Control.DeepSeq
import Network.HTTP.Conduit
import Network.HTTP.Client
import Network.HTTP.Simple (getResponseHeader)
import Network.Socket
import Network.BSD (getProtocolNumber)
import Data.Either
import Data.Conduit
import Text.Read
#ifdef mingw32_HOST_OS
import qualified System.FilePath.Windows as PW
#endif
type URLString = String
type Headers = [String]
type UserAgent = String
limit url downloads to whitelisted schemes Security fix! Allowing any schemes, particularly file: and possibly others like scp: allowed file exfiltration by anyone who had write access to the git repository, since they could add an annexed file using such an url, or using an url that redirected to such an url, and wait for the victim to get it into their repository and send them a copy. * Added annex.security.allowed-url-schemes setting, which defaults to only allowing http and https URLs. Note especially that file:/ is no longer enabled by default. * Removed annex.web-download-command, since its interface does not allow supporting annex.security.allowed-url-schemes across redirects. If you used this setting, you may want to instead use annex.web-options to pass options to curl. With annex.web-download-command removed, nearly all url accesses in git-annex are made via Utility.Url via http-client or curl. http-client only supports http and https, so no problem there. (Disabling one and not the other is not implemented.) Used curl --proto to limit the allowed url schemes. Note that this will cause git annex fsck --from web to mark files using a disallowed url scheme as not being present in the web. That seems acceptable; fsck --from web also does that when a web server is not available. youtube-dl already disabled file: itself (probably for similar reasons). The scheme check was also added to youtube-dl urls for completeness, although that check won't catch any redirects it might follow. But youtube-dl goes off and does its own thing with other protocols anyway, so that's fine. Special remotes that support other domain-specific url schemes are not affected by this change. In the bittorrent remote, aria2c can still download magnet: links. The download of the .torrent file is otherwise now limited by annex.security.allowed-url-schemes. This does not address any external special remotes that might download an url themselves. Current thinking is all external special remotes will need to be audited for this problem, although many of them will use http libraries that only support http and not curl's menagarie. The related problem of accessing private localhost and LAN urls is not addressed by this commit. This commit was sponsored by Brett Eisenberg on Patreon.
2018-06-15 20:52:24 +00:00
newtype Scheme = Scheme (CI.CI String)
deriving (Eq, Ord)
mkScheme :: String -> Scheme
mkScheme = Scheme . CI.mk
fromScheme :: Scheme -> String
fromScheme (Scheme s) = CI.original s
data UrlOptions = UrlOptions
2014-02-25 02:00:25 +00:00
{ userAgent :: Maybe UserAgent
, reqHeaders :: Headers
, urlDownloader :: UrlDownloader
, applyRequest :: Request -> Request
, httpManager :: Manager
limit url downloads to whitelisted schemes Security fix! Allowing any schemes, particularly file: and possibly others like scp: allowed file exfiltration by anyone who had write access to the git repository, since they could add an annexed file using such an url, or using an url that redirected to such an url, and wait for the victim to get it into their repository and send them a copy. * Added annex.security.allowed-url-schemes setting, which defaults to only allowing http and https URLs. Note especially that file:/ is no longer enabled by default. * Removed annex.web-download-command, since its interface does not allow supporting annex.security.allowed-url-schemes across redirects. If you used this setting, you may want to instead use annex.web-options to pass options to curl. With annex.web-download-command removed, nearly all url accesses in git-annex are made via Utility.Url via http-client or curl. http-client only supports http and https, so no problem there. (Disabling one and not the other is not implemented.) Used curl --proto to limit the allowed url schemes. Note that this will cause git annex fsck --from web to mark files using a disallowed url scheme as not being present in the web. That seems acceptable; fsck --from web also does that when a web server is not available. youtube-dl already disabled file: itself (probably for similar reasons). The scheme check was also added to youtube-dl urls for completeness, although that check won't catch any redirects it might follow. But youtube-dl goes off and does its own thing with other protocols anyway, so that's fine. Special remotes that support other domain-specific url schemes are not affected by this change. In the bittorrent remote, aria2c can still download magnet: links. The download of the .torrent file is otherwise now limited by annex.security.allowed-url-schemes. This does not address any external special remotes that might download an url themselves. Current thinking is all external special remotes will need to be audited for this problem, although many of them will use http libraries that only support http and not curl's menagarie. The related problem of accessing private localhost and LAN urls is not addressed by this commit. This commit was sponsored by Brett Eisenberg on Patreon.
2018-06-15 20:52:24 +00:00
, allowedSchemes :: S.Set Scheme
, disallowedSchemeMessage :: Maybe (URI -> String)
, getBasicAuth :: GetBasicAuth
2014-02-25 02:00:25 +00:00
}
data UrlDownloader
= DownloadWithConduit NonHttpUrlDownloader
| DownloadWithCurl [CommandParam]
data NonHttpUrlDownloader
= DownloadWithCurlRestricted Restriction
defUrlOptions :: IO UrlOptions
defUrlOptions = UrlOptions
<$> pure Nothing
<*> pure []
<*> pure (DownloadWithConduit (DownloadWithCurlRestricted mempty))
<*> pure id
<*> newManager tlsManagerSettings
<*> pure conduitUrlSchemes
<*> pure Nothing
<*> pure noBasicAuth
conduitUrlSchemes :: S.Set Scheme
conduitUrlSchemes = S.fromList $ map mkScheme ["http", "https", "ftp"]
mkUrlOptions :: Maybe UserAgent -> Headers -> UrlDownloader -> Manager -> S.Set Scheme -> Maybe (URI -> String) -> GetBasicAuth -> UrlOptions
mkUrlOptions defuseragent reqheaders urldownloader =
UrlOptions useragent reqheaders urldownloader applyrequest
where
applyrequest = \r -> r { requestHeaders = requestHeaders r ++ addedheaders }
addedheaders = uaheader ++ otherheaders
useragent = maybe defuseragent (Just . B8.toString . snd)
(headMaybe uafromheaders)
uaheader = case useragent of
Nothing -> []
Just ua -> [(hUserAgent, B8.fromString ua)]
(uafromheaders, otherheaders) = partition (\(h, _) -> h == hUserAgent)
(map toheader reqheaders)
toheader s =
let (h, v) = separate (== ':') s
h' = CI.mk (B8.fromString h)
in case v of
(' ':v') -> (h', B8.fromString v')
_ -> (h', B8.fromString v)
curlParams :: UrlOptions -> [CommandParam] -> [CommandParam]
limit url downloads to whitelisted schemes Security fix! Allowing any schemes, particularly file: and possibly others like scp: allowed file exfiltration by anyone who had write access to the git repository, since they could add an annexed file using such an url, or using an url that redirected to such an url, and wait for the victim to get it into their repository and send them a copy. * Added annex.security.allowed-url-schemes setting, which defaults to only allowing http and https URLs. Note especially that file:/ is no longer enabled by default. * Removed annex.web-download-command, since its interface does not allow supporting annex.security.allowed-url-schemes across redirects. If you used this setting, you may want to instead use annex.web-options to pass options to curl. With annex.web-download-command removed, nearly all url accesses in git-annex are made via Utility.Url via http-client or curl. http-client only supports http and https, so no problem there. (Disabling one and not the other is not implemented.) Used curl --proto to limit the allowed url schemes. Note that this will cause git annex fsck --from web to mark files using a disallowed url scheme as not being present in the web. That seems acceptable; fsck --from web also does that when a web server is not available. youtube-dl already disabled file: itself (probably for similar reasons). The scheme check was also added to youtube-dl urls for completeness, although that check won't catch any redirects it might follow. But youtube-dl goes off and does its own thing with other protocols anyway, so that's fine. Special remotes that support other domain-specific url schemes are not affected by this change. In the bittorrent remote, aria2c can still download magnet: links. The download of the .torrent file is otherwise now limited by annex.security.allowed-url-schemes. This does not address any external special remotes that might download an url themselves. Current thinking is all external special remotes will need to be audited for this problem, although many of them will use http libraries that only support http and not curl's menagarie. The related problem of accessing private localhost and LAN urls is not addressed by this commit. This commit was sponsored by Brett Eisenberg on Patreon.
2018-06-15 20:52:24 +00:00
curlParams uo ps = ps ++ uaparams ++ headerparams ++ addedparams ++ schemeparams
where
uaparams = case userAgent uo of
Nothing -> []
Just ua -> [Param "--user-agent", Param ua]
headerparams = concatMap (\h -> [Param "-H", Param h]) (reqHeaders uo)
addedparams = case urlDownloader uo of
DownloadWithConduit _ -> []
DownloadWithCurl l -> l
limit url downloads to whitelisted schemes Security fix! Allowing any schemes, particularly file: and possibly others like scp: allowed file exfiltration by anyone who had write access to the git repository, since they could add an annexed file using such an url, or using an url that redirected to such an url, and wait for the victim to get it into their repository and send them a copy. * Added annex.security.allowed-url-schemes setting, which defaults to only allowing http and https URLs. Note especially that file:/ is no longer enabled by default. * Removed annex.web-download-command, since its interface does not allow supporting annex.security.allowed-url-schemes across redirects. If you used this setting, you may want to instead use annex.web-options to pass options to curl. With annex.web-download-command removed, nearly all url accesses in git-annex are made via Utility.Url via http-client or curl. http-client only supports http and https, so no problem there. (Disabling one and not the other is not implemented.) Used curl --proto to limit the allowed url schemes. Note that this will cause git annex fsck --from web to mark files using a disallowed url scheme as not being present in the web. That seems acceptable; fsck --from web also does that when a web server is not available. youtube-dl already disabled file: itself (probably for similar reasons). The scheme check was also added to youtube-dl urls for completeness, although that check won't catch any redirects it might follow. But youtube-dl goes off and does its own thing with other protocols anyway, so that's fine. Special remotes that support other domain-specific url schemes are not affected by this change. In the bittorrent remote, aria2c can still download magnet: links. The download of the .torrent file is otherwise now limited by annex.security.allowed-url-schemes. This does not address any external special remotes that might download an url themselves. Current thinking is all external special remotes will need to be audited for this problem, although many of them will use http libraries that only support http and not curl's menagarie. The related problem of accessing private localhost and LAN urls is not addressed by this commit. This commit was sponsored by Brett Eisenberg on Patreon.
2018-06-15 20:52:24 +00:00
schemeparams =
[ Param "--proto"
, Param $ intercalate "," ("-all" : schemelist)
]
schemelist = map fromScheme $ S.toList $ allowedSchemes uo
checkPolicy :: UrlOptions -> URI -> IO (Either String a) -> IO (Either String a)
checkPolicy uo u a
limit url downloads to whitelisted schemes Security fix! Allowing any schemes, particularly file: and possibly others like scp: allowed file exfiltration by anyone who had write access to the git repository, since they could add an annexed file using such an url, or using an url that redirected to such an url, and wait for the victim to get it into their repository and send them a copy. * Added annex.security.allowed-url-schemes setting, which defaults to only allowing http and https URLs. Note especially that file:/ is no longer enabled by default. * Removed annex.web-download-command, since its interface does not allow supporting annex.security.allowed-url-schemes across redirects. If you used this setting, you may want to instead use annex.web-options to pass options to curl. With annex.web-download-command removed, nearly all url accesses in git-annex are made via Utility.Url via http-client or curl. http-client only supports http and https, so no problem there. (Disabling one and not the other is not implemented.) Used curl --proto to limit the allowed url schemes. Note that this will cause git annex fsck --from web to mark files using a disallowed url scheme as not being present in the web. That seems acceptable; fsck --from web also does that when a web server is not available. youtube-dl already disabled file: itself (probably for similar reasons). The scheme check was also added to youtube-dl urls for completeness, although that check won't catch any redirects it might follow. But youtube-dl goes off and does its own thing with other protocols anyway, so that's fine. Special remotes that support other domain-specific url schemes are not affected by this change. In the bittorrent remote, aria2c can still download magnet: links. The download of the .torrent file is otherwise now limited by annex.security.allowed-url-schemes. This does not address any external special remotes that might download an url themselves. Current thinking is all external special remotes will need to be audited for this problem, although many of them will use http libraries that only support http and not curl's menagarie. The related problem of accessing private localhost and LAN urls is not addressed by this commit. This commit was sponsored by Brett Eisenberg on Patreon.
2018-06-15 20:52:24 +00:00
| allowedScheme uo u = a
| otherwise = return $ Left $ case disallowedSchemeMessage uo of
Nothing -> "Configuration does not allow accessing" ++ show u
Just f -> f u
limit url downloads to whitelisted schemes Security fix! Allowing any schemes, particularly file: and possibly others like scp: allowed file exfiltration by anyone who had write access to the git repository, since they could add an annexed file using such an url, or using an url that redirected to such an url, and wait for the victim to get it into their repository and send them a copy. * Added annex.security.allowed-url-schemes setting, which defaults to only allowing http and https URLs. Note especially that file:/ is no longer enabled by default. * Removed annex.web-download-command, since its interface does not allow supporting annex.security.allowed-url-schemes across redirects. If you used this setting, you may want to instead use annex.web-options to pass options to curl. With annex.web-download-command removed, nearly all url accesses in git-annex are made via Utility.Url via http-client or curl. http-client only supports http and https, so no problem there. (Disabling one and not the other is not implemented.) Used curl --proto to limit the allowed url schemes. Note that this will cause git annex fsck --from web to mark files using a disallowed url scheme as not being present in the web. That seems acceptable; fsck --from web also does that when a web server is not available. youtube-dl already disabled file: itself (probably for similar reasons). The scheme check was also added to youtube-dl urls for completeness, although that check won't catch any redirects it might follow. But youtube-dl goes off and does its own thing with other protocols anyway, so that's fine. Special remotes that support other domain-specific url schemes are not affected by this change. In the bittorrent remote, aria2c can still download magnet: links. The download of the .torrent file is otherwise now limited by annex.security.allowed-url-schemes. This does not address any external special remotes that might download an url themselves. Current thinking is all external special remotes will need to be audited for this problem, although many of them will use http libraries that only support http and not curl's menagarie. The related problem of accessing private localhost and LAN urls is not addressed by this commit. This commit was sponsored by Brett Eisenberg on Patreon.
2018-06-15 20:52:24 +00:00
unsupportedUrlScheme :: URI -> String
unsupportedUrlScheme u = "Unsupported url scheme " ++ show u
limit url downloads to whitelisted schemes Security fix! Allowing any schemes, particularly file: and possibly others like scp: allowed file exfiltration by anyone who had write access to the git repository, since they could add an annexed file using such an url, or using an url that redirected to such an url, and wait for the victim to get it into their repository and send them a copy. * Added annex.security.allowed-url-schemes setting, which defaults to only allowing http and https URLs. Note especially that file:/ is no longer enabled by default. * Removed annex.web-download-command, since its interface does not allow supporting annex.security.allowed-url-schemes across redirects. If you used this setting, you may want to instead use annex.web-options to pass options to curl. With annex.web-download-command removed, nearly all url accesses in git-annex are made via Utility.Url via http-client or curl. http-client only supports http and https, so no problem there. (Disabling one and not the other is not implemented.) Used curl --proto to limit the allowed url schemes. Note that this will cause git annex fsck --from web to mark files using a disallowed url scheme as not being present in the web. That seems acceptable; fsck --from web also does that when a web server is not available. youtube-dl already disabled file: itself (probably for similar reasons). The scheme check was also added to youtube-dl urls for completeness, although that check won't catch any redirects it might follow. But youtube-dl goes off and does its own thing with other protocols anyway, so that's fine. Special remotes that support other domain-specific url schemes are not affected by this change. In the bittorrent remote, aria2c can still download magnet: links. The download of the .torrent file is otherwise now limited by annex.security.allowed-url-schemes. This does not address any external special remotes that might download an url themselves. Current thinking is all external special remotes will need to be audited for this problem, although many of them will use http libraries that only support http and not curl's menagarie. The related problem of accessing private localhost and LAN urls is not addressed by this commit. This commit was sponsored by Brett Eisenberg on Patreon.
2018-06-15 20:52:24 +00:00
allowedScheme :: UrlOptions -> URI -> Bool
allowedScheme uo u = uscheme `S.member` allowedSchemes uo
where
uscheme = mkScheme $ takeWhile (/=':') (uriScheme u)
{- Checks that an url exists and could be successfully downloaded,
- also checking that its size, if available, matches a specified size.
-
- The Left error is returned if policy or the restricted http manager
- does not allow accessing the url or the url scheme is not supported.
-}
checkBoth :: URLString -> Maybe Integer -> UrlOptions -> IO (Either String Bool)
checkBoth url expected_size uo = fmap go <$> check url expected_size uo
where
go v = fst v && snd v
check :: URLString -> Maybe Integer -> UrlOptions -> IO (Either String (Bool, Bool))
check url expected_size uo = fmap go <$> getUrlInfo url uo
2012-12-13 04:24:19 +00:00
where
go (UrlInfo False _ _) = (False, False)
go (UrlInfo True Nothing _) = (True, True)
go (UrlInfo True s _) = case expected_size of
Just _ -> (True, expected_size == s)
Nothing -> (True, True)
exists :: URLString -> UrlOptions -> IO (Either String Bool)
exists url uo = fmap urlExists <$> getUrlInfo url uo
data UrlInfo = UrlInfo
{ urlExists :: Bool
, urlSize :: Maybe Integer
, urlSuggestedFile :: Maybe FilePath
}
deriving (Show)
assumeUrlExists :: UrlInfo
assumeUrlExists = UrlInfo True Nothing Nothing
{- Checks that an url exists and could be successfully downloaded,
- also returning its size and suggested filename if available.
-
- The Left error is returned if policy or the restricted http manages
- does not allow accessing the url or the url scheme is not supported.
-}
getUrlInfo :: URLString -> UrlOptions -> IO (Either String UrlInfo)
getUrlInfo url uo = case parseURIRelaxed url of
Just u -> checkPolicy uo u (go u)
Nothing -> return (Right dne)
where
go :: URI -> IO (Either String UrlInfo)
go u = case (urlDownloader uo, parseRequestRelaxed u) of
(DownloadWithConduit (DownloadWithCurlRestricted r), Just req) ->
existsconduit r req
(DownloadWithConduit (DownloadWithCurlRestricted r), Nothing)
| isfileurl u -> Right <$> existsfile u
| isftpurl u -> (Right <$> existscurlrestricted r u url ftpport)
`catchNonAsync` (const $ return $ Right dne)
| otherwise -> return $ Left $ unsupportedUrlScheme u
(DownloadWithCurl _, _)
| isfileurl u -> Right <$> existsfile u
| otherwise -> Right <$> existscurl u (basecurlparams url)
limit url downloads to whitelisted schemes Security fix! Allowing any schemes, particularly file: and possibly others like scp: allowed file exfiltration by anyone who had write access to the git repository, since they could add an annexed file using such an url, or using an url that redirected to such an url, and wait for the victim to get it into their repository and send them a copy. * Added annex.security.allowed-url-schemes setting, which defaults to only allowing http and https URLs. Note especially that file:/ is no longer enabled by default. * Removed annex.web-download-command, since its interface does not allow supporting annex.security.allowed-url-schemes across redirects. If you used this setting, you may want to instead use annex.web-options to pass options to curl. With annex.web-download-command removed, nearly all url accesses in git-annex are made via Utility.Url via http-client or curl. http-client only supports http and https, so no problem there. (Disabling one and not the other is not implemented.) Used curl --proto to limit the allowed url schemes. Note that this will cause git annex fsck --from web to mark files using a disallowed url scheme as not being present in the web. That seems acceptable; fsck --from web also does that when a web server is not available. youtube-dl already disabled file: itself (probably for similar reasons). The scheme check was also added to youtube-dl urls for completeness, although that check won't catch any redirects it might follow. But youtube-dl goes off and does its own thing with other protocols anyway, so that's fine. Special remotes that support other domain-specific url schemes are not affected by this change. In the bittorrent remote, aria2c can still download magnet: links. The download of the .torrent file is otherwise now limited by annex.security.allowed-url-schemes. This does not address any external special remotes that might download an url themselves. Current thinking is all external special remotes will need to be audited for this problem, although many of them will use http libraries that only support http and not curl's menagarie. The related problem of accessing private localhost and LAN urls is not addressed by this commit. This commit was sponsored by Brett Eisenberg on Patreon.
2018-06-15 20:52:24 +00:00
dne = UrlInfo False Nothing Nothing
found sz f = return $ UrlInfo True sz f
isfileurl u = uriScheme u == "file:"
isftpurl u = uriScheme u == "ftp:"
ftpport = 21
basecurlparams u = curlParams uo $
[ Param "-s"
, Param "--head"
, Param "-L", Param u
, Param "-w", Param "%{http_code}"
]
extractlencurl s = case lastMaybe $ filter ("Content-Length:" `isPrefixOf`) (lines s) of
Just l -> case lastMaybe $ words l of
Just sz -> readish sz
_ -> Nothing
_ -> Nothing
extractlen = readish . B8.toString
<=< lookup hContentLength . responseHeaders
extractfilename = contentDispositionFilename . B8.toString
<=< lookup hContentDisposition . responseHeaders
existsconduit r req =
2020-05-04 16:44:26 +00:00
let a = catchcrossprotoredir r (existsconduit' req uo)
in catchJust matchconnectionrestricted a retconnectionrestricted
matchconnectionrestricted he@(HttpExceptionRequest _ (InternalException ie)) =
case fromException ie of
2020-05-04 16:44:26 +00:00
Just (ConnectionRestricted _why) -> Just he
_ -> Nothing
matchconnectionrestricted _ = Nothing
retconnectionrestricted he@(HttpExceptionRequest _ (InternalException ie)) =
case fromException ie of
Just (ConnectionRestricted why) -> return (Left why)
_ -> throwM he
retconnectionrestricted he = throwM he
existsconduit' req uo' = do
2015-01-22 17:47:06 +00:00
let req' = headRequest (applyRequest uo req)
debug "Utility.Url" (show req')
join $ runResourceT $ do
resp <- http req' (httpManager uo)
if responseStatus resp == ok200
then do
len <- extractFromResourceT (extractlen resp)
fn <- extractFromResourceT (extractfilename resp)
return $ found len fn
else if responseStatus resp == unauthorized401
then return $ getBasicAuth uo' (show (getUri req)) >>= \case
Nothing -> return dne
Just (ba, signalsuccess) -> do
ui <- existsconduit'
(applyBasicAuth' ba req)
(uo' { getBasicAuth = noBasicAuth })
signalsuccess (urlExists ui)
return ui
else return $ return dne
existscurl u curlparams = do
output <- catchDefaultIO "" $
readProcess "curl" $ toCommand curlparams
let len = extractlencurl output
let good = found len Nothing
let isftp = or
[ "ftp" `isInfixOf` uriScheme u
-- Check to see if http redirected to ftp.
, "Location: ftp://" `isInfixOf` output
]
case lastMaybe (lines output) of
Just ('2':_:_) -> good
-- don't try to parse ftp status codes; if curl
-- got a length, it's good
_ | isftp && isJust len -> good
limit url downloads to whitelisted schemes Security fix! Allowing any schemes, particularly file: and possibly others like scp: allowed file exfiltration by anyone who had write access to the git repository, since they could add an annexed file using such an url, or using an url that redirected to such an url, and wait for the victim to get it into their repository and send them a copy. * Added annex.security.allowed-url-schemes setting, which defaults to only allowing http and https URLs. Note especially that file:/ is no longer enabled by default. * Removed annex.web-download-command, since its interface does not allow supporting annex.security.allowed-url-schemes across redirects. If you used this setting, you may want to instead use annex.web-options to pass options to curl. With annex.web-download-command removed, nearly all url accesses in git-annex are made via Utility.Url via http-client or curl. http-client only supports http and https, so no problem there. (Disabling one and not the other is not implemented.) Used curl --proto to limit the allowed url schemes. Note that this will cause git annex fsck --from web to mark files using a disallowed url scheme as not being present in the web. That seems acceptable; fsck --from web also does that when a web server is not available. youtube-dl already disabled file: itself (probably for similar reasons). The scheme check was also added to youtube-dl urls for completeness, although that check won't catch any redirects it might follow. But youtube-dl goes off and does its own thing with other protocols anyway, so that's fine. Special remotes that support other domain-specific url schemes are not affected by this change. In the bittorrent remote, aria2c can still download magnet: links. The download of the .torrent file is otherwise now limited by annex.security.allowed-url-schemes. This does not address any external special remotes that might download an url themselves. Current thinking is all external special remotes will need to be audited for this problem, although many of them will use http libraries that only support http and not curl's menagarie. The related problem of accessing private localhost and LAN urls is not addressed by this commit. This commit was sponsored by Brett Eisenberg on Patreon.
2018-06-15 20:52:24 +00:00
_ -> return dne
existscurlrestricted r u url' defport = existscurl u
=<< curlRestrictedParams r u defport (basecurlparams url')
existsfile u = do
let f = toRawFilePath (unEscapeString (uriPath u))
s <- catchMaybeIO $ R.getSymbolicLinkStatus f
case s of
Just stat -> do
sz <- getFileSize' f stat
found (Just sz) Nothing
Nothing -> return dne
-- When http server redirects to a protocol which conduit does not
-- support, it will throw a StatusCodeException with found302
-- and a Response with the redir Location.
catchcrossprotoredir r a =
catchJust (matchStatusCodeException (== found302))
(Right <$> a)
(followredir r)
followredir r (HttpExceptionRequest _ (StatusCodeException resp _)) =
case headMaybe $ map decodeBS $ getResponseHeader hLocation resp of
Just url' -> case parseURIRelaxed url' of
-- only follow http to ftp redirects;
-- http to file redirect would not be secure,
-- and http-conduit follows http to http.
Just u' | isftpurl u' ->
checkPolicy uo u' $ Right <$>
existscurlrestricted r u' url' ftpport
_ -> return (Right dne)
Nothing -> return (Right dne)
followredir _ _ = return (Right dne)
-- Parse eg: attachment; filename="fname.ext"
-- per RFC 2616
contentDispositionFilename :: String -> Maybe FilePath
contentDispositionFilename s
| "attachment; filename=\"" `isPrefixOf` s && "\"" `isSuffixOf` s =
2018-11-23 15:24:05 +00:00
Just $ dropFromEnd 1 $ drop 1 $ dropWhile (/= '"') s
| otherwise = Nothing
headRequest :: Request -> Request
headRequest r = r
{ method = methodHead
2023-03-14 02:39:16 +00:00
-- remove default Accept-Encoding header, to get actual,
-- not gzip compressed size.
, requestHeaders = (hAcceptEncoding, B.empty) :
filter (\(h, _) -> h /= hAcceptEncoding)
(requestHeaders r)
}
{- Download a perhaps large file, with auto-resume of incomplete downloads.
-
- When the download fails, returns an error message.
-}
download :: MeterUpdate -> Maybe IncrementalVerifier -> URLString -> FilePath -> UrlOptions -> IO (Either String ())
download = download' False
download' :: Bool -> MeterUpdate -> Maybe IncrementalVerifier -> URLString -> FilePath -> UrlOptions -> IO (Either String ())
download' nocurlerror meterupdate iv url file uo =
catchJust matchHttpException go showhttpexception
2018-10-03 15:56:52 +00:00
`catchNonAsync` (dlfailed . show)
where
go = case parseURIRelaxed url of
Just u -> checkPolicy uo u $
case (urlDownloader uo, parseRequestRelaxed u) of
(DownloadWithConduit (DownloadWithCurlRestricted r), Just req) -> catchJust
(matchStatusCodeException (== found302))
(downloadConduit meterupdate iv req file uo >> return (Right ()))
(followredir r)
(DownloadWithConduit (DownloadWithCurlRestricted r), Nothing)
| isfileurl u -> downloadfile u
| isftpurl u -> downloadcurlrestricted r u url ftpport
| otherwise -> dlfailed $ unsupportedUrlScheme u
(DownloadWithCurl _, _)
| isfileurl u -> downloadfile u
| otherwise -> downloadcurl url basecurlparams
Nothing -> do
liftIO $ debug "Utility.Url" url
2018-10-03 15:56:52 +00:00
dlfailed "invalid url"
isfileurl u = uriScheme u == "file:"
isftpurl u = uriScheme u == "ftp:"
ftpport = 21
showhttpexception he = dlfailed $ case he of
HttpExceptionRequest _ (StatusCodeException r _) ->
B8.toString $ statusMessage $ responseStatus r
HttpExceptionRequest _ (InternalException ie) ->
case fromException ie of
Nothing -> show ie
Just (ConnectionRestricted why) -> why
HttpExceptionRequest _ other -> show other
_ -> show he
dlfailed msg = do
noverification
return $ Left $ "download failed: " ++ msg
basecurlparams = curlParams uo
[ if nocurlerror
then Param "-S"
else Param "-sS"
, Param "-f"
, Param "-L"
, Param "-C", Param "-"
]
downloadcurl rawurl curlparams = do
noverification
-- curl does not create destination file
-- if the url happens to be empty, so pre-create.
unlessM (doesFileExist file) $
writeFile file ""
ifM (boolSystem "curl" (curlparams ++ [Param "-o", File file, File rawurl]))
( return $ Right ()
, return $ Left "download failed"
)
downloadcurlrestricted r u rawurl defport =
downloadcurl rawurl =<< curlRestrictedParams r u defport basecurlparams
downloadfile u = do
noverification
let src = unEscapeString (uriPath u)
withMeteredFile src meterupdate $
L.writeFile file
return $ Right ()
-- Conduit does not support ftp, so will throw an exception on a
-- redirect to a ftp url; fall back to curl.
followredir r ex@(HttpExceptionRequest _ (StatusCodeException resp _)) =
case headMaybe $ map decodeBS $ getResponseHeader hLocation resp of
Just url' -> case parseURIRelaxed url' of
Just u' | isftpurl u' ->
checkPolicy uo u' $
downloadcurlrestricted r u' url' ftpport
_ -> throwIO ex
Nothing -> throwIO ex
followredir _ ex = throwIO ex
noverification = maybe noop unableIncrementalVerifier iv
{- Download a perhaps large file using conduit, with auto-resume
- of incomplete downloads.
-
- A Request can be configured to throw exceptions for non-2xx http
- status codes, or not. That configuration is overridden by this,
- and if it is unable to download, it throws an exception containing
- a user-visible explanation of the problem. (However, exceptions
- thrown for reasons other than http status codes will still be thrown
- as usual.)
-}
downloadConduit :: MeterUpdate -> Maybe IncrementalVerifier -> Request -> FilePath -> UrlOptions -> IO ()
downloadConduit meterupdate iv req file uo =
catchMaybeIO (getFileSize (toRawFilePath file)) >>= \case
Just sz | sz > 0 -> resumedownload sz
_ -> join $ runResourceT $ do
liftIO $ debug "Utility.Url" (show req')
resp <- http req' (httpManager uo)
if responseStatus resp == ok200
then do
store zeroBytesProcessed WriteMode resp
return (return ())
else do
rf <- extractFromResourceT (respfailure resp)
if responseStatus resp == unauthorized401
then return $ getBasicAuth uo (show (getUri req')) >>= \case
Nothing -> giveup rf
Just ba -> retryauthed ba
else return $ giveup rf
where
req' = applyRequest uo $ req
-- Override http-client's default decompression of gzip
-- compressed files. We want the unmodified file content.
{ requestHeaders = (hAcceptEncoding, "identity") :
filter ((/= hAcceptEncoding) . fst)
(requestHeaders req)
, decompress = const False
-- Avoid throwing exceptions non-2xx http status codes,
-- since we rely on parsing the Response to handle
-- several such codes.
, checkResponse = \_ _ -> return ()
}
-- Resume download from where a previous download was interrupted,
-- when supported by the http server. The server may also opt to
-- send the whole file rather than resuming.
resumedownload sz = join $ runResourceT $ do
let req'' = req' { requestHeaders = resumeFromHeader sz : requestHeaders req' }
liftIO $ debug "Urility.Url" (show req'')
resp <- http req'' (httpManager uo)
if responseStatus resp == partialContent206
then do
store (toBytesProcessed sz) AppendMode resp
return (return ())
else if responseStatus resp == ok200
then do
store zeroBytesProcessed WriteMode resp
return (return ())
else if alreadydownloaded sz resp
then do
liftIO noverification
return (return ())
else do
rf <- extractFromResourceT (respfailure resp)
if responseStatus resp == unauthorized401
then return $ getBasicAuth uo (show (getUri req'')) >>= \case
Nothing -> giveup rf
Just ba -> retryauthed ba
else return $ giveup rf
alreadydownloaded sz resp
| responseStatus resp /= requestedRangeNotSatisfiable416 = False
| otherwise = case lookup hContentRange (responseHeaders resp) of
-- This could be improved by fixing
-- https://github.com/aristidb/http-types/issues/87
Just crh -> crh == B8.fromString ("bytes */" ++ show sz)
-- Some http servers send no Content-Range header when
-- the range extends beyond the end of the file.
-- There is no way to distinguish between the file
-- being the same size on the http server, vs
-- it being shorter than the file we already have.
-- So assume we have the whole content of the file
-- already, the same as wget and curl do.
Nothing -> True
store initialp mode resp =
sinkResponseFile meterupdate iv initialp file mode resp
respfailure = B8.toString . statusMessage . responseStatus
retryauthed (ba, signalsuccess) = do
r <- tryNonAsync $ downloadConduit
meterupdate iv
(applyBasicAuth' ba req)
file
(uo { getBasicAuth = noBasicAuth })
case r of
Right () -> signalsuccess True
Left e -> do
() <- signalsuccess False
throwM e
noverification = maybe noop unableIncrementalVerifier iv
{- Sinks a Response's body to a file. The file can either be appended to
- (AppendMode), or written from the start of the response (WriteMode).
- Updates the meter and incremental verifier as data is received,
- when not appending.
-
- Note that the responseStatus is not checked by this function.
-}
2018-04-25 01:23:40 +00:00
sinkResponseFile
:: MonadResource m
=> MeterUpdate
-> Maybe IncrementalVerifier
2018-04-25 01:23:40 +00:00
-> BytesProcessed
-> FilePath
-> IOMode
-> Response (ConduitM () B8.ByteString m ())
-> m ()
sinkResponseFile meterupdate iv initialp file mode resp = do
ui <- case (iv, mode) of
(Just iv', AppendMode) -> do
liftIO $ unableIncrementalVerifier iv'
return (const noop)
(Just iv', _) -> return (updateIncrementalVerifier iv')
(Nothing, _) -> return (const noop)
(fr, fh) <- allocate (openBinaryFile file mode) hClose
runConduit $ responseBody resp .| go ui initialp fh
release fr
where
go ui sofar fh = await >>= \case
Nothing -> return ()
Just bs -> do
let sofar' = addBytesProcessed sofar (B.length bs)
liftIO $ do
void $ meterupdate sofar'
() <- ui bs
B.hPut fh bs
go ui sofar' fh
{- Downloads at least the specified number of bytes from an url. -}
downloadPartial :: URLString -> UrlOptions -> Int -> IO (Maybe L.ByteString)
downloadPartial url uo n = case parseURIRelaxed url of
Nothing -> return Nothing
Just u -> go u `catchNonAsync` const (return Nothing)
where
go u = case parseRequestRelaxed u of
Nothing -> return Nothing
Just req -> do
let req' = applyRequest uo req
liftIO $ debug "Utility.Url" (show req')
withResponse req' (httpManager uo) $ \resp ->
if responseStatus resp == ok200
then Just <$> brReadSome (responseBody resp) n
else return Nothing
{- On unix this is the same as parseURI. But on Windows,
- it can parse urls such as file:///C:/path/to/file
- parseURI normally parses that as a path /C:/path/to/file
- and this simply removes the excess leading slash when there is a
- drive letter after it. -}
parseURIPortable :: URLString -> Maybe URI
#ifndef mingw32_HOST_OS
parseURIPortable = parseURI
#else
parseURIPortable s
| "file:" `isPrefixOf` s = do
u <- parseURI s
return $ case PW.splitDirectories (uriPath u) of
(p:d:_) | all PW.isPathSeparator p && PW.isDrive d ->
u { uriPath = dropWhile PW.isPathSeparator (uriPath u) }
_ -> u
| otherwise = parseURI s
#endif
{- Allows for spaces and other stuff in urls, properly escaping them. -}
parseURIRelaxed :: URLString -> Maybe URI
parseURIRelaxed s = maybe (parseURIRelaxed' s) Just $
parseURIPortable $ escapeURIString isAllowedInURI s
{- Generate a http-conduit Request for an URI. This is able
- to deal with some urls that parseRequest would usually reject.
-}
parseRequestRelaxed :: MonadThrow m => URI -> m Request
parseRequestRelaxed u = case uriAuthority u of
Just ua
-- parseURI can handle an empty port value, but
-- parseRequest cannot. So remove the ':' to
-- make it work.
| uriPort ua == ":" -> parseRequest $ show $
u { uriAuthority = Just $ ua { uriPort = "" } }
_ -> parseRequest (show u)
{- Some characters like '[' are allowed in eg, the address of
- an uri, but cannot appear unescaped further along in the uri.
- This handles that, expensively, by successively escaping each character
- from the back of the url until the url parses.
-}
parseURIRelaxed' :: URLString -> Maybe URI
parseURIRelaxed' s = go [] (reverse s)
where
go back [] = parseURI back
go back (c:cs) = case parseURI (escapeURIString isAllowedInURI (reverse (c:cs)) ++ back) of
Just u -> Just u
Nothing -> go (escapeURIChar escapemore c ++ back) cs
escapemore '[' = False
escapemore ']' = False
escapemore c = isAllowedInURI c
hAcceptEncoding :: CI.CI B.ByteString
hAcceptEncoding = "Accept-Encoding"
hContentDisposition :: CI.CI B.ByteString
hContentDisposition = "Content-Disposition"
hContentRange :: CI.CI B.ByteString
hContentRange = "Content-Range"
resumeFromHeader :: FileSize -> Header
resumeFromHeader sz = (hRange, renderByteRanges [ByteRangeFrom sz])
{- Use with eg:
-
- > catchJust (matchStatusCodeException (== notFound404))
-}
matchStatusCodeException :: (Status -> Bool) -> HttpException -> Maybe HttpException
matchStatusCodeException want = matchStatusCodeHeadersException (\s _h -> want s)
matchStatusCodeHeadersException :: (Status -> ResponseHeaders -> Bool) -> HttpException -> Maybe HttpException
matchStatusCodeHeadersException want e@(HttpExceptionRequest _ (StatusCodeException r _))
| want (responseStatus r) (responseHeaders r) = Just e
| otherwise = Nothing
matchStatusCodeHeadersException _ _ = Nothing
{- Use with eg:
-
- > catchJust matchHttpException
-}
matchHttpException :: HttpException -> Maybe HttpException
matchHttpException = Just
matchHttpExceptionContent :: (HttpExceptionContent -> Bool) -> HttpException -> Maybe HttpException
matchHttpExceptionContent want e@(HttpExceptionRequest _ hec)
| want hec = Just e
| otherwise = Nothing
matchHttpExceptionContent _ _ = Nothing
{- Constructs parameters that prevent curl from accessing any IP addresses
- blocked by the Restriction. These are added to the input parameters,
- which should tell curl what to do.
-
- This has to disable redirects because it looks up the IP addresses
- of the host and after limiting to those allowed by the Restriction,
- makes curl resolve the host to those IP addresses. It doesn't make sense
- to use this for http anyway, only for ftp or perhaps other protocols
- supported by curl.
-
- Throws an exception if the Restriction blocks all addresses, or
- if the dns lookup fails. A malformed url will also cause an exception.
-}
curlRestrictedParams :: Restriction -> URI -> Int -> [CommandParam] -> IO [CommandParam]
curlRestrictedParams r u defport ps = case uriAuthority u of
Nothing -> giveup "malformed url"
Just uath -> case uriPort uath of
"" -> go (uriRegName uath) defport
-- ignore an empty port, same as
-- parseRequestRelaxed does.
":" -> go (uriRegName uath) defport
-- strict parser because the port we provide to curl
-- needs to match the port in the url
(':':s) -> case readMaybe s :: Maybe Int of
Just p -> go (uriRegName uath) p
Nothing -> giveup "malformed url"
_ -> giveup "malformed url"
where
go hostname p = do
proto <- getProtocolNumber "tcp"
let serv = show p
let hints = defaultHints
{ addrFlags = [AI_ADDRCONFIG]
, addrProtocol = proto
, addrSocketType = Stream
}
addrs <- getAddrInfo (Just hints) (Just hostname) (Just serv)
case partitionEithers (map checkrestriction addrs) of
((e:_es), []) -> throwIO e
(_, as)
| null as -> throwIO $
NC.HostNotResolved hostname
| otherwise -> return $
(limitresolve p) as ++ ps
checkrestriction addr = maybe (Right addr) Left $
checkAddressRestriction r addr
limitresolve p addrs =
[ Param "--resolve"
, Param $ "*:" ++ show p ++ ":" ++ intercalate ":"
(mapMaybe (bracketaddr <$$> extractIPAddress . addrAddress) addrs)
-- Don't let a ftp server provide an IP address.
, Param "--ftp-skip-pasv-ip"
-- Prevent all http redirects.
, Param "--max-redirs", Param "0"
]
bracketaddr a = "[" ++ a ++ "]"
data BasicAuth = BasicAuth
{ basicAuthUser :: String
, basicAuthPassword :: String
}
-- Note that this is only used when using conduit, not curl.
--
-- The returned IO action is run after trying to use the BasicAuth,
-- indicating if the password worked.
type GetBasicAuth = URLString -> IO (Maybe (BasicAuth, Bool -> IO ()))
noBasicAuth :: GetBasicAuth
noBasicAuth = const $ pure Nothing
applyBasicAuth' :: BasicAuth -> Request -> Request
applyBasicAuth' ba = applyBasicAuth
(encodeBS (basicAuthUser ba))
(encodeBS (basicAuthPassword ba))
{- Make sure whatever is returned is fully evaluated. Avoids any possible
- issues with laziness deferring processing until a time when the resource
- has been freed. -}
extractFromResourceT :: (MonadIO m, NFData a) => a -> ResourceT m a
extractFromResourceT v = do
liftIO $ evaluate (rnf v)
return v