diff --git a/ext/zlib/zlib.c b/ext/zlib/zlib.c
index dc608ee..fe03072 100644
--- a/ext/zlib/zlib.c
+++ b/ext/zlib/zlib.c
@@ -3500,6 +3500,9 @@ static VALUE
 rb_gzfile_eof_p(VALUE obj)
 {
     struct gzfile *gz = get_gzfile(obj);
+    while (!ZSTREAM_IS_FINISHED(&gz->z) && ZSTREAM_BUF_FILLED(&gz->z) == 0) {
+	gzfile_read_more(gz, Qnil);
+    }
     return GZFILE_IS_FINISHED(gz) ? Qtrue : Qfalse;
 }
 
diff --git a/test/zlib/test_zlib.rb b/test/zlib/test_zlib.rb
index 779c583..502ccce 100644
--- a/test/zlib/test_zlib.rb
+++ b/test/zlib/test_zlib.rb
@@ -1205,6 +1205,38 @@ def test_double_close
       }
     end
 
+    # Various methods of Zlib::GzipReader failed when to reading files
+    # just a few bytes larger than GZFILE_READ_SIZE.
+    def test_gzfile_read_size_boundary
+      Tempfile.create("test_zlib_gzip_read_size_boundary") {|t|
+        t.close
+        # NO_COMPRESSION helps with recreating the error condition.
+        # The error happens on compressed files too, but it's harder to reproduce.
+        # For example, ~12750 bytes are needed to trigger the error using __FILE__.
+        # We avoid this because the test file will change over time.
+        Zlib::GzipWriter.open(t.path, Zlib::NO_COMPRESSION) do |gz|
+          gz.print("\n" * 2024) # range from 2024 to 2033 triggers the error
+          gz.flush
+        end
+
+        Zlib::GzipReader.open(t.path) do |f|
+          f.readpartial(1024) until f.eof?
+          assert_raise(EOFError) { f.readpartial(1) }
+        end
+
+        Zlib::GzipReader.open(t.path) do |f|
+          f.readline until f.eof?
+          assert_raise(EOFError) { f.readline }
+        end
+
+        Zlib::GzipReader.open(t.path) do |f|
+          b = f.readbyte until f.eof?
+          f.ungetbyte(b)
+          f.readbyte
+          assert_raise(EOFError) { f.readbyte }
+        end
+      }
+    end
   end
 
   class TestZlibGzipWriter < Test::Unit::TestCase