Reimplement -serialize_reads flag using new SyncRead mount flag

Let the kernel do the work for us.

See hanwen/go-fuse@15a8bb0
for more info.

Author: rfjakob

internal/fusefrontend/args.go

@@ -26,8 +26,6 @@ type Args struct {
 	ConfigCustom bool
 	// NoPrealloc disables automatic preallocation before writing
 	NoPrealloc bool
-	// Try to serialize read operations, "-serialize_reads"
-	SerializeReads bool
 	// Force decode even if integrity check fails (openSSL only)
 	ForceDecode bool
 	// Exclude is a list of paths to make inaccessible, starting match at

mount.go

@@ -273,7 +273,6 @@ func initFuseFrontend(args *argContainer) (rootNode fs.InodeEmbedder, wipeKeys f
 		LongNames:          args.longnames,
 		ConfigCustom:       args._configCustom,
 		NoPrealloc:         args.noprealloc,
-		SerializeReads:     args.serialize_reads,
 		ForceDecode:        args.forcedecode,
 		ForceOwner:         args._forceOwner,
 		Exclude:            args.exclude,
@@ -377,6 +376,14 @@ func initGoFuse(rootNode fs.InodeEmbedder, args *argContainer) *fuse.Server {
 		MaxWrite: fuse.MAX_KERNEL_WRITE,
 		Options:  []string{fmt.Sprintf("max_read=%d", fuse.MAX_KERNEL_WRITE)},
 		Debug:    args.fusedebug,
+		// The kernel usually submits multiple read requests in parallel,
+		// which means we serve them in any order. Out-of-order reads are
+		// expensive on some backing network filesystems
+		// ( https://github.com/rfjakob/gocryptfs/issues/92 ).
+		//
+		// Setting SyncRead disables FUSE_CAP_ASYNC_READ. This makes the kernel
+		// do everything in-order without parallelism.
+		SyncRead: args.serialize_reads,
 	}
 
 	mOpts := &fuseOpts.MountOptions