Account for HTTP/2 padding in receive windows.

net/spdy/spdy_framer.cc

Index: net/spdy/spdy_framer.cc
diff --git a/net/spdy/spdy_framer.cc b/net/spdy/spdy_framer.cc
index e7603ff7bd445d6ee723db28f17a0ee7badba672..a706c56ab0540a3dc7ef7e35faa63854983e4b96 100644
--- a/net/spdy/spdy_framer.cc
+++ b/net/spdy/spdy_framer.cc
@@ -2155,6 +2155,8 @@ size_t SpdyFramer::ProcessDataFramePaddingLength(const char* data, size_t len) {
   size_t original_len = len;
   if (current_frame_flags_ & DATA_FLAG_PADDED) {
     if (len != 0) {
+      static_assert(kPadLengthFieldSize == 1,
+                    "Unexpected pad length field size.");
       if (remaining_data_length_ < kPadLengthFieldSize) {
         set_error(SPDY_INVALID_DATA_FRAME_FLAGS);
         return 0;
@@ -2164,6 +2166,11 @@ size_t SpdyFramer::ProcessDataFramePaddingLength(const char* data, size_t len) {
       ++data;
       --len;
       --remaining_data_length_;
+
+      // Notify visitor that |kPadLengthFieldSize| bytes of payload are
+      // consumed.
+      visitor_->OnStreamFrameData(current_frame_stream_id_, NULL,
+                                  kPadLengthFieldSize, false);

[Ryan Hamilton, 2015/02/25 21:59:15]

Should this change land internally first? In any case, I'm not in love with this
approach. I think I'd prefer either:
1) and explicit OnStreamPadding() or
2) and extra size_t num_padding_bytes argument to OnStreamFrameData()

     } else {
       // We don't have the data available for parsing the pad length field. Keep
       // waiting.