diff --git a/lib/go/camli/schema/filewriter.go b/lib/go/camli/schema/filewriter.go
index f96af3752..4126513f2 100644
--- a/lib/go/camli/schema/filewriter.go
+++ b/lib/go/camli/schema/filewriter.go
@@ -17,6 +17,7 @@ limitations under the License.
 package schema
 
 import (
+	"bufio"
 	"bytes"
 	"crypto"
 	"fmt"
@@ -27,6 +28,7 @@ import (
 
 	"camli/blobref"
 	"camli/blobserver"
+	"camli/rollsum"
 )
 
 var _ = log.Printf
@@ -110,5 +112,154 @@ func serverHasBlob(bs blobserver.Storage, br *blobref.BlobRef) (have bool, err o
 		have = true
 	}
 	return
-
+}
+
+type span struct {
+	from, to int64
+	bits     int
+	br       *blobref.BlobRef
+	children []span
+}
+
+func (s *span) size() int64 {
+	size := s.to - s.from
+	for _, cs := range s.children {
+		size += cs.size()
+	}
+	return size
+}
+
+// WriteFileFromReaderRolling creates and uploads a "file" JSON schema
+// composed of chunks of r, also uploading the chunks.  The returned
+// BlobRef is of the JSON file schema blob.
+func WriteFileFromReaderRolling(bs blobserver.Storage, filename string, r io.Reader) (outbr *blobref.BlobRef, outerr os.Error) {
+	bufr := bufio.NewReader(r)
+	spans := []span{} // the tree of spans, cut on interesting rollsum boundaries
+	rs := rollsum.New()
+	n := int64(0)
+	last := n
+	buf := new(bytes.Buffer)
+
+	uploadString := func(s string) (*blobref.BlobRef, os.Error) {
+		br := blobref.Sha1FromString(s)
+		hasIt, err := serverHasBlob(bs, br)
+		if err != nil {
+			return nil, err
+		}
+		if hasIt {
+			return br, nil
+		}
+		_, err = bs.ReceiveBlob(br, strings.NewReader(s))
+		if err != nil {
+			return nil, err
+		}
+		return br, nil
+	}
+
+	// TODO: keep multiple of these in-flight at a time.
+	uploadLastSpan := func() bool {
+		defer buf.Reset()
+		br, err := uploadString(buf.String())
+		if err != nil {
+			outerr = err
+			return false
+		}
+		spans[len(spans)-1].br = br
+		return true
+	}
+
+	for {
+		c, err := bufr.ReadByte()
+		if err == os.EOF {
+			if n != last {
+				spans = append(spans, span{from: last, to: n})
+				if !uploadLastSpan() {
+					return
+				}
+			}
+			break
+		}
+		if err != nil {
+			return nil, err
+		}
+		buf.WriteByte(c)
+
+		n++
+		rs.Roll(c)
+		if !rs.OnSplit() {
+			continue
+		}
+		bits := rs.Bits()
+
+		// Take any spans from the end of the spans slice that
+		// have a smaller 'bits' score and make them children
+		// of this node.
+		var children []span
+		childrenFrom := len(spans)
+		for childrenFrom > 0 && spans[childrenFrom-1].bits < bits {
+			childrenFrom--
+		}
+		if nCopy := len(spans) - childrenFrom; nCopy > 0 {
+			children = make([]span, nCopy)
+			copy(children, spans[childrenFrom:])
+			spans = spans[:childrenFrom]
+		}
+
+		spans = append(spans, span{from: last, to: n, bits: bits, children: children})
+		last = n
+		if !uploadLastSpan() {
+			return
+		}
+	}
+
+	var addContentParts func(dst *[]ContentPart, s []span) os.Error
+
+	uploadFile := func(filename string, fileSize int64, s []span) (*blobref.BlobRef, os.Error) {
+		parts := []ContentPart{}
+		err := addContentParts(&parts, s)
+		if err != nil {
+			return nil, err
+		}
+		m := NewCommonFilenameMap(filename)
+		err = PopulateRegularFileMap(m, fileSize, parts)
+		if err != nil {
+			return nil, err
+		}
+		json, err := MapToCamliJson(m)
+		if err != nil {
+			return nil, err
+		}
+		return uploadString(json)
+	}
+
+	addContentParts = func(dst *[]ContentPart, spansl []span) os.Error {
+		for _, sp := range spansl {
+			if len(sp.children) > 0 {
+				childrenSize := int64(0)
+				for _, cs := range sp.children {
+					childrenSize += cs.size()
+				}
+				br, err := uploadFile("", childrenSize, sp.children)
+				if err != nil {
+					return err
+				}
+				*dst = append(*dst, ContentPart{
+					SubBlobRefString: br.String(),
+					SubBlobRef:       br,
+					Size:             uint64(childrenSize),
+				})
+			}
+			if sp.from != sp.to {
+				*dst = append(*dst, ContentPart{
+					BlobRefString: sp.br.String(),
+					BlobRef:       sp.br,
+					Size:          uint64(sp.to - sp.from),
+				})
+			}
+		}
+		return nil
+	}
+
+	// The top-level content parts
+	return uploadFile(filename, n, spans)
 }
diff --git a/lib/go/camli/schema/schema.go b/lib/go/camli/schema/schema.go
index 1098b723f..a2cb1217a 100644
--- a/lib/go/camli/schema/schema.go
+++ b/lib/go/camli/schema/schema.go
@@ -90,10 +90,12 @@ type Superset struct {
 }
 
 type ContentPart struct {
-	BlobRefString string           "blobRef"
-	BlobRef       *blobref.BlobRef // TODO: ditch BlobRefString? use json.Unmarshaler?
-	Size          uint64           "size"
-	Offset        uint64           "offset"
+	BlobRefString    string           "blobRef"
+	BlobRef          *blobref.BlobRef // TODO: ditch BlobRefString? use json.Unmarshaler?
+	SubBlobRefString string           "subFileBlobRef"
+	SubBlobRef       *blobref.BlobRef
+	Size             uint64 "size"
+	Offset           uint64 "offset"
 }
 
 func (cp *ContentPart) blobref() *blobref.BlobRef {
@@ -242,12 +244,14 @@ func MapToCamliJson(m map[string]interface{}) (string, os.Error) {
 
 func NewCommonFilenameMap(fileName string) map[string]interface{} {
 	m := newCamliMap(1, "" /* no type yet */ )
-	lastSlash := strings.LastIndex(fileName, "/")
-	baseName := fileName[lastSlash+1:]
-	if isValidUtf8(baseName) {
-		m["fileName"] = baseName
-	} else {
-		m["fileNameBytes"] = []uint8(baseName)
+	if fileName != "" {
+		lastSlash := strings.LastIndex(fileName, "/")
+		baseName := fileName[lastSlash+1:]
+		if isValidUtf8(baseName) {
+			m["fileName"] = baseName
+		} else {
+			m["fileNameBytes"] = []uint8(baseName)
+		}
 	}
 	return m
 }
@@ -299,7 +303,11 @@ func PopulateRegularFileMap(m map[string]interface{}, size int64, parts []Conten
 	for idx, part := range parts {
 		mpart := make(map[string]interface{})
 		mparts[idx] = mpart
-		mpart["blobRef"] = part.BlobRef.String()
+		if part.BlobRef != nil {
+			mpart["blobRef"] = part.BlobRef.String()
+		} else if part.SubBlobRef != nil {
+			mpart["subFileBlobRef"] = part.SubBlobRef.String()
+		}
 		mpart["size"] = part.Size
 		sumSize += part.Size
 		if part.Offset != 0 {