When accessing a local remote, shut down git-cat-file processes afterwards, to ensure that remotes on removable media can be unmounted. Closes: #758630

This does mean that eg, copying multiple files to a local remote will
become slightly slower, since it now restarts git-cat-file after each copy.
Should not be significant slowdown.

The reason git-cat-file is run on the remote at all is to update its
location log. In order to add an item to it, it needs to get the current
content of the log. Finding a way to avoid needing to do that would be a
good path to avoiding this slowdown if it does become a problem somehow.

This commit was sponsored by Evan Deaubl.

Annex.hs

@@ -15,6 +15,7 @@ module Annex (
 	eval,
 	getState,
 	changeState,
+	withState,
 	setFlag,
 	setField,
 	setOutput,
@@ -214,6 +215,11 @@ changeState modifier = do
 	mvar <- ask
 	liftIO $ modifyMVar_ mvar $ return . modifier
 
+withState :: (AnnexState -> (AnnexState, b)) -> Annex b
+withState modifier = do
+	mvar <- ask
+	liftIO $ modifyMVar mvar $ return . modifier
+
 {- Sets a flag to True -}
 setFlag :: String -> Annex ()
 setFlag flag = changeState $ \s ->

Annex/CatFile.hs

@@ -12,6 +12,7 @@ module Annex.CatFile (
 	catTree,
 	catObjectDetails,
 	catFileHandle,
+	catFileStop,
 	catKey,
 	catKeyFile,
 	catKeyFileHEAD,
@@ -71,6 +72,14 @@ catFileHandle = do
 			Annex.changeState $ \s -> s { Annex.catfilehandles = m' }
 			return h
 
+{- Stops all running cat-files. Should only be run when it's known that
+ - nothing is using the handles, eg at shutdown. -}
+catFileStop :: Annex ()
+catFileStop = do
+	m <- Annex.withState $ \s ->
+		(s { Annex.catfilehandles = M.empty }, Annex.catfilehandles s)
+	liftIO $ mapM_ Git.CatFile.catFileStop (M.elems m)
+
 {- From the Sha or Ref of a symlink back to the key.
  -
  - Requires a mode witness, to guarantee that the file is a symlink.

Remote/Git.hs

@@ -51,6 +51,7 @@ import qualified Remote.Helper.Ssh as Ssh
 import qualified Remote.GCrypt
 import Config.Files
 import Creds
+import Annex.CatFile
 
 import Control.Concurrent
 import Control.Concurrent.MSampleVar
@@ -500,6 +501,8 @@ repairRemote r a = return $ do
 {- Runs an action from the perspective of a local remote.
  -
  - The AnnexState is cached for speed and to avoid resource leaks.
+ - However, catFileStop is called to avoid git-cat-file processes hanging
+ - around on removable media.
  -
  - The repository's git-annex branch is not updated, as an optimisation.
  - No caller of onLocal can query data from the branch and be ensured
@@ -520,7 +523,8 @@ onLocal r a = do
 	cache st = Annex.changeState $ \s -> s
 		{ Annex.remoteannexstate = M.insert (uuid r) st (Annex.remoteannexstate s) }
 	go st a' = do
-		(ret, st') <- liftIO $ Annex.run st a'
+		(ret, st') <- liftIO $ Annex.run st $
+			catFileStop `after` a'
 		cache st'
 		return ret
 

debian/changelog

@@ -3,6 +3,9 @@ git-annex (5.20140818) UNRELEASED; urgency=medium
   * Make --help work when not in a git repository. Closes: #758592
   * Ensure that all lock fds are close-on-exec, fixing various problems with
     them being inherited by child processes such as git commands.
+  * When accessing a local remote, shut down git-cat-file processes
+    afterwards, to ensure that remotes on removable media can be unmounted.
+    Closes: #758630
 
  -- Joey Hess <joeyh@debian.org>  Tue, 19 Aug 2014 12:52:41 -0400