assistant: Fix a race condition that could cause a pointer file to get ingested into the annex

This was caused by commit fb8ab2469d putting
an isPointerFile check in the wrong place. So if the file was not a pointer
file at that point, but got replaced by one before the file got locked
down, the pointer file would be ingested into the annex.

The fix is simply to move the isPointerFile check to after safeToAdd locks
down the file. Now if the file changes to a pointer file after the
isPointerFile check, ingestion will see that it changed after lockdown,
and will refuse to add it to the annex.

Sponsored-by: the NIH-funded NICEMAN (ReproNim TR&D3) project

Assistant/Threads/Committer.hs

@@ -290,12 +290,14 @@ handleAdds lockdowndir havelsof largefilematcher annexdotfiles delayadd cs = ret
 		refillChanges postponed
 
 	returnWhen (null toadd) $ do
+		(addedpointerfiles, toaddrest) <- partitionEithers
+			<$> mapM checkpointerfile toadd
 		(toaddannexed, toaddsmall) <- partitionEithers
-			<$> mapM checksmall toadd
+			<$> mapM checksmall toaddrest
 		addsmall toaddsmall
 		addedannexed <- addaction toadd $
 			catMaybes <$> addannexed toaddannexed
-		return $ addedannexed ++ toaddsmall ++ otherchanges
+		return $ addedannexed ++ toaddsmall ++ addedpointerfiles ++ otherchanges
   where
 	(incomplete, otherchanges) = partition (\c -> isPendingAddChange c || isInProcessAddChange c) cs
 
@@ -303,6 +305,19 @@ handleAdds lockdowndir havelsof largefilematcher annexdotfiles delayadd cs = ret
 		| c = return otherchanges
 		| otherwise = a
 	
+	checkpointerfile change = do
+		let file = toRawFilePath $ changeFile change
+		mk <- liftIO $ isPointerFile file
+		case mk of
+			Nothing -> return (Right change)
+			Just key -> do
+				mode <- liftIO $ catchMaybeIO $ fileMode <$> R.getFileStatus file
+				liftAnnex $ stagePointerFile file mode =<< hashPointerFile key
+				return $ Left $ Change
+					(changeTime change)
+					(changeFile change)
+					(LinkChange (Just key))
+	
 	checksmall change
 		| not annexdotfiles && dotfile f =
 			return (Right change)

Assistant/Threads/Watcher.hs

@@ -196,11 +196,8 @@ shouldRestage :: DaemonStatus -> Bool
 shouldRestage ds = scanComplete ds || forceRestage ds
 
 onAddFile :: Bool -> Handler
-onAddFile symlinkssupported f fs = do
+onAddFile symlinkssupported f fs =
-	mk <- liftIO $ isPointerFile $ toRawFilePath f
+	onAddFile' contentchanged addassociatedfile addlink samefilestatus symlinkssupported f fs
-	case mk of
-		Nothing -> onAddFile' contentchanged addassociatedfile addlink samefilestatus symlinkssupported f fs
-		Just k -> addlink f k
   where
 	addassociatedfile key file = 
 		Database.Keys.addAssociatedFile key

CHANGELOG

@@ -1,3 +1,10 @@
+git-annex (10.20240702) UNRELEASED; urgency=medium
+
+  * assistant: Fix a race condition that could cause a pointer file to
+    get ingested into the annex.
+
+ -- Joey Hess <id@joeyh.name>  Tue, 02 Jul 2024 12:14:53 -0400
+
 git-annex (10.20240701) upstream; urgency=medium
 
   * git-annex remotes can now act as proxies that provide access to

doc/bugs/assistant___40__webapp__41___commited_unlocked_link_to_annex.mdwn

@@ -104,3 +104,6 @@ on laptop where I dive into inception: 10.20240129
 
 [[!meta author=yoh]]
 [[!tag projects/repronim]]
+
+[[!meta title="inception: pointer file can be ingested into the annex due to assistant bug or manually"]]
+

doc/bugs/assistant___40__webapp__41___commited_unlocked_link_to_annex/comment_7_44ab02fd027efc56ce8bb701e5513b1d._comment

@@ -0,0 +1,15 @@
+[[!comment format=mdwn
+ username="joey"
+ subject="""comment 7"""
+ date="2024-07-02T16:15:28Z"
+ content="""
+I've fixed this race in the assistant.
+
+Question now is, can this bug be closed, or does it need to be left open,
+and git-annex made to recover from this situation? Given the complexity
+of making git-annex notice this, I'm sort of inclined to not have it
+auto-recover. Manual recovery seems pretty simple, just delete the file and
+re-add it with the right key. 
+
+Thoughts?
+"""]]