diff options
Diffstat (limited to 'src/cowboy_http.erl')
-rw-r--r-- | src/cowboy_http.erl | 38 |
1 files changed, 32 insertions, 6 deletions
diff --git a/src/cowboy_http.erl b/src/cowboy_http.erl index 7e20615..f2defdc 100644 --- a/src/cowboy_http.erl +++ b/src/cowboy_http.erl @@ -963,8 +963,17 @@ te_chunked(Data, {0, Streamed}) -> %% @todo We are expecting an hex size, not a general token. token(Data, fun (<< "\r\n", Rest/binary >>, BinLen) -> - Len = list_to_integer(binary_to_list(BinLen), 16), - te_chunked(Rest, {Len, Streamed}); + case list_to_integer(binary_to_list(BinLen), 16) of + %% Final chunk is parsed in one go above. Rest would be + %% <<\r\n">> if complete. + 0 when byte_size(Rest) < 2 -> + more; + %% Normal chunk. Add 2 to Len for trailing <<"\r\n">>. Note + %% that repeated <<"-2\r\n">> would be streamed, and + %% accumulated, until out of memory if Len could be -2. + Len when Len > 0 -> + te_chunked(Rest, {Len + 2, Streamed}) + end; %% Chunk size shouldn't take too many bytes, %% don't try to stream forever. (Rest, _) when byte_size(Rest) < 16 -> @@ -972,11 +981,28 @@ te_chunked(Data, {0, Streamed}) -> (_, _) -> {error, badarg} end); -te_chunked(Data, {ChunkRem, Streamed}) when byte_size(Data) >= ChunkRem + 2 -> - << Chunk:ChunkRem/binary, "\r\n", Rest/binary >> = Data, - {ok, Chunk, Rest, {0, Streamed + byte_size(Chunk)}}; +%% <<"\n">> from trailing <<"\r\n">>. +te_chunked(<< "\n", Rest/binary>>, {1, Streamed}) -> + {ok, <<>>, Rest, {0, Streamed}}; +te_chunked(<<>>, State={1, _Streamed}) -> + {more, 1, <<>>, State}; +%% Remainder of chunk (if any) and as much of trailing <<"\r\n">> as possible. +te_chunked(Data, {ChunkRem, Streamed}) when byte_size(Data) >= ChunkRem - 2 -> + ChunkSize = ChunkRem - 2, + Streamed2 = Streamed + ChunkSize, + case Data of + << Chunk:ChunkSize/binary, "\r\n", Rest/binary >> -> + {ok, Chunk, Rest, {0, Streamed2}}; + << Chunk:ChunkSize/binary, "\r" >> -> + {more, 1, Chunk, {1, Streamed2}}; + << Chunk:ChunkSize/binary >> -> + {more, 2, Chunk, {2, Streamed2}} + end; +%% Incomplete chunk. te_chunked(Data, {ChunkRem, Streamed}) -> - {more, ChunkRem + 2, Data, {ChunkRem, Streamed}}. + ChunkRem2 = ChunkRem - byte_size(Data), + Streamed2 = Streamed + byte_size(Data), + {more, ChunkRem2, Data, {ChunkRem2, Streamed2}}. %% @doc Decode an identity stream. -spec te_identity(Bin, TransferState) |