0
Fork 0
mirror of https://github.com/ninenines/cowboy.git synced 2025-07-14 20:30:23 +00:00
cowboy/test/security_SUITE.erl

405 lines
15 KiB
Erlang
Raw Normal View History

2024-01-25 11:22:54 +01:00
%% Copyright (c) 2018-2024, Loïc Hoguin <essen@ninenines.eu>
%%
%% Permission to use, copy, modify, and/or distribute this software for any
%% purpose with or without fee is hereby granted, provided that the above
%% copyright notice and this permission notice appear in all copies.
%%
%% THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
%% WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
%% MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
%% ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
%% WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
%% ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
%% OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
2018-11-21 14:18:20 +01:00
-module(security_SUITE).
-compile(export_all).
-compile(nowarn_export_all).
-import(ct_helper, [config/2]).
-import(ct_helper, [doc/1]).
-import(cowboy_test, [gun_open/1]).
-import(cowboy_test, [raw_open/1]).
-import(cowboy_test, [raw_send/2]).
-import(cowboy_test, [raw_recv_head/1]).
-import(cowboy_test, [raw_recv/3]).
%% ct.
all() ->
cowboy_test:common_all().
groups() ->
Fix HTTP/2 CVEs A number of HTTP/2 CVEs were documented recently: https://www.kb.cert.org/vuls/id/605641/ This commit, along with a few changes and additions in Cowlib, fix or improve protection against all of them. For CVE-2019-9511, also known as Data Dribble, the new option stream_window_data_threshold can be used to control how little the DATA frames that Cowboy sends can get. For CVE-2019-9516, also known as 0-Length Headers Leak, Cowboy will now simply reject streams containing 0-length header names. For CVE-2019-9517, also known as Internal Data Buffering, the backpressure changes were already pretty good at preventing this issue, but a new option max_connection_buffer_size was added for even better control over how much memory we are willing to allocate. For CVE-2019-9512, also known as Ping Flood; CVE-2019-9515, also known as Settings Flood; CVE-2019-9518, also known as Empty Frame Flooding; and similar undocumented scenarios, a frame rate limiting mechanism was added. By default Cowboy will now allow 1000 frames every 10 seconds. This can be configured via max_received_frame_rate. For CVE-2019-9514, also known as Reset Flood, another rate limiting mechanism was added and can be configured via max_reset_stream_rate. By default Cowboy will do up to 10 stream resets every 10 seconds. Finally, nothing was done for CVE-2019-9513, also known as Resource Loop, because Cowboy does not currently implement the HTTP/2 priority mechanism (in parts because these issues were well known from the start). Tests were added for all cases except Internal Data Buffering, which I'm not sure how to test, and Resource Loop, which is not currently relevant.
2019-10-02 10:44:45 +02:00
Tests = [nc_rand, nc_zero],
H1Tests = [slowloris, slowloris_chunks],
H2CTests = [
http2_cancel_flood,
Fix HTTP/2 CVEs A number of HTTP/2 CVEs were documented recently: https://www.kb.cert.org/vuls/id/605641/ This commit, along with a few changes and additions in Cowlib, fix or improve protection against all of them. For CVE-2019-9511, also known as Data Dribble, the new option stream_window_data_threshold can be used to control how little the DATA frames that Cowboy sends can get. For CVE-2019-9516, also known as 0-Length Headers Leak, Cowboy will now simply reject streams containing 0-length header names. For CVE-2019-9517, also known as Internal Data Buffering, the backpressure changes were already pretty good at preventing this issue, but a new option max_connection_buffer_size was added for even better control over how much memory we are willing to allocate. For CVE-2019-9512, also known as Ping Flood; CVE-2019-9515, also known as Settings Flood; CVE-2019-9518, also known as Empty Frame Flooding; and similar undocumented scenarios, a frame rate limiting mechanism was added. By default Cowboy will now allow 1000 frames every 10 seconds. This can be configured via max_received_frame_rate. For CVE-2019-9514, also known as Reset Flood, another rate limiting mechanism was added and can be configured via max_reset_stream_rate. By default Cowboy will do up to 10 stream resets every 10 seconds. Finally, nothing was done for CVE-2019-9513, also known as Resource Loop, because Cowboy does not currently implement the HTTP/2 priority mechanism (in parts because these issues were well known from the start). Tests were added for all cases except Internal Data Buffering, which I'm not sure how to test, and Resource Loop, which is not currently relevant.
2019-10-02 10:44:45 +02:00
http2_data_dribble,
http2_empty_frame_flooding_data,
http2_empty_frame_flooding_headers_continuation,
http2_empty_frame_flooding_push_promise,
http2_infinite_continuations,
Fix HTTP/2 CVEs A number of HTTP/2 CVEs were documented recently: https://www.kb.cert.org/vuls/id/605641/ This commit, along with a few changes and additions in Cowlib, fix or improve protection against all of them. For CVE-2019-9511, also known as Data Dribble, the new option stream_window_data_threshold can be used to control how little the DATA frames that Cowboy sends can get. For CVE-2019-9516, also known as 0-Length Headers Leak, Cowboy will now simply reject streams containing 0-length header names. For CVE-2019-9517, also known as Internal Data Buffering, the backpressure changes were already pretty good at preventing this issue, but a new option max_connection_buffer_size was added for even better control over how much memory we are willing to allocate. For CVE-2019-9512, also known as Ping Flood; CVE-2019-9515, also known as Settings Flood; CVE-2019-9518, also known as Empty Frame Flooding; and similar undocumented scenarios, a frame rate limiting mechanism was added. By default Cowboy will now allow 1000 frames every 10 seconds. This can be configured via max_received_frame_rate. For CVE-2019-9514, also known as Reset Flood, another rate limiting mechanism was added and can be configured via max_reset_stream_rate. By default Cowboy will do up to 10 stream resets every 10 seconds. Finally, nothing was done for CVE-2019-9513, also known as Resource Loop, because Cowboy does not currently implement the HTTP/2 priority mechanism (in parts because these issues were well known from the start). Tests were added for all cases except Internal Data Buffering, which I'm not sure how to test, and Resource Loop, which is not currently relevant.
2019-10-02 10:44:45 +02:00
http2_ping_flood,
http2_reset_flood,
http2_settings_flood,
http2_zero_length_header_leak
],
[
{http, [parallel], Tests ++ H1Tests},
{https, [parallel], Tests ++ H1Tests},
{h2, [parallel], Tests},
{h2c, [parallel], Tests ++ H2CTests},
{h3, [], Tests},
Fix HTTP/2 CVEs A number of HTTP/2 CVEs were documented recently: https://www.kb.cert.org/vuls/id/605641/ This commit, along with a few changes and additions in Cowlib, fix or improve protection against all of them. For CVE-2019-9511, also known as Data Dribble, the new option stream_window_data_threshold can be used to control how little the DATA frames that Cowboy sends can get. For CVE-2019-9516, also known as 0-Length Headers Leak, Cowboy will now simply reject streams containing 0-length header names. For CVE-2019-9517, also known as Internal Data Buffering, the backpressure changes were already pretty good at preventing this issue, but a new option max_connection_buffer_size was added for even better control over how much memory we are willing to allocate. For CVE-2019-9512, also known as Ping Flood; CVE-2019-9515, also known as Settings Flood; CVE-2019-9518, also known as Empty Frame Flooding; and similar undocumented scenarios, a frame rate limiting mechanism was added. By default Cowboy will now allow 1000 frames every 10 seconds. This can be configured via max_received_frame_rate. For CVE-2019-9514, also known as Reset Flood, another rate limiting mechanism was added and can be configured via max_reset_stream_rate. By default Cowboy will do up to 10 stream resets every 10 seconds. Finally, nothing was done for CVE-2019-9513, also known as Resource Loop, because Cowboy does not currently implement the HTTP/2 priority mechanism (in parts because these issues were well known from the start). Tests were added for all cases except Internal Data Buffering, which I'm not sure how to test, and Resource Loop, which is not currently relevant.
2019-10-02 10:44:45 +02:00
{http_compress, [parallel], Tests ++ H1Tests},
{https_compress, [parallel], Tests ++ H1Tests},
{h2_compress, [parallel], Tests},
{h2c_compress, [parallel], Tests ++ H2CTests},
{h3_compress, [], Tests}
Fix HTTP/2 CVEs A number of HTTP/2 CVEs were documented recently: https://www.kb.cert.org/vuls/id/605641/ This commit, along with a few changes and additions in Cowlib, fix or improve protection against all of them. For CVE-2019-9511, also known as Data Dribble, the new option stream_window_data_threshold can be used to control how little the DATA frames that Cowboy sends can get. For CVE-2019-9516, also known as 0-Length Headers Leak, Cowboy will now simply reject streams containing 0-length header names. For CVE-2019-9517, also known as Internal Data Buffering, the backpressure changes were already pretty good at preventing this issue, but a new option max_connection_buffer_size was added for even better control over how much memory we are willing to allocate. For CVE-2019-9512, also known as Ping Flood; CVE-2019-9515, also known as Settings Flood; CVE-2019-9518, also known as Empty Frame Flooding; and similar undocumented scenarios, a frame rate limiting mechanism was added. By default Cowboy will now allow 1000 frames every 10 seconds. This can be configured via max_received_frame_rate. For CVE-2019-9514, also known as Reset Flood, another rate limiting mechanism was added and can be configured via max_reset_stream_rate. By default Cowboy will do up to 10 stream resets every 10 seconds. Finally, nothing was done for CVE-2019-9513, also known as Resource Loop, because Cowboy does not currently implement the HTTP/2 priority mechanism (in parts because these issues were well known from the start). Tests were added for all cases except Internal Data Buffering, which I'm not sure how to test, and Resource Loop, which is not currently relevant.
2019-10-02 10:44:45 +02:00
].
init_per_suite(Config) ->
ct_helper:create_static_dir(config(priv_dir, Config) ++ "/static"),
Config.
end_per_suite(Config) ->
ct_helper:delete_static_dir(config(priv_dir, Config) ++ "/static").
init_per_group(Name, Config) ->
cowboy_test:init_common_groups(Name, Config, ?MODULE).
end_per_group(Name, _) ->
cowboy_test:stop_group(Name).
%% Routes.
init_dispatch(_) ->
cowboy_router:compile([{"localhost", [
Fix HTTP/2 CVEs A number of HTTP/2 CVEs were documented recently: https://www.kb.cert.org/vuls/id/605641/ This commit, along with a few changes and additions in Cowlib, fix or improve protection against all of them. For CVE-2019-9511, also known as Data Dribble, the new option stream_window_data_threshold can be used to control how little the DATA frames that Cowboy sends can get. For CVE-2019-9516, also known as 0-Length Headers Leak, Cowboy will now simply reject streams containing 0-length header names. For CVE-2019-9517, also known as Internal Data Buffering, the backpressure changes were already pretty good at preventing this issue, but a new option max_connection_buffer_size was added for even better control over how much memory we are willing to allocate. For CVE-2019-9512, also known as Ping Flood; CVE-2019-9515, also known as Settings Flood; CVE-2019-9518, also known as Empty Frame Flooding; and similar undocumented scenarios, a frame rate limiting mechanism was added. By default Cowboy will now allow 1000 frames every 10 seconds. This can be configured via max_received_frame_rate. For CVE-2019-9514, also known as Reset Flood, another rate limiting mechanism was added and can be configured via max_reset_stream_rate. By default Cowboy will do up to 10 stream resets every 10 seconds. Finally, nothing was done for CVE-2019-9513, also known as Resource Loop, because Cowboy does not currently implement the HTTP/2 priority mechanism (in parts because these issues were well known from the start). Tests were added for all cases except Internal Data Buffering, which I'm not sure how to test, and Resource Loop, which is not currently relevant.
2019-10-02 10:44:45 +02:00
{"/", hello_h, []},
{"/echo/:key", echo_h, []},
{"/delay_hello", delay_hello_h, 1000},
2019-10-07 12:18:03 +02:00
{"/long_polling", long_polling_h, []},
Fix HTTP/2 CVEs A number of HTTP/2 CVEs were documented recently: https://www.kb.cert.org/vuls/id/605641/ This commit, along with a few changes and additions in Cowlib, fix or improve protection against all of them. For CVE-2019-9511, also known as Data Dribble, the new option stream_window_data_threshold can be used to control how little the DATA frames that Cowboy sends can get. For CVE-2019-9516, also known as 0-Length Headers Leak, Cowboy will now simply reject streams containing 0-length header names. For CVE-2019-9517, also known as Internal Data Buffering, the backpressure changes were already pretty good at preventing this issue, but a new option max_connection_buffer_size was added for even better control over how much memory we are willing to allocate. For CVE-2019-9512, also known as Ping Flood; CVE-2019-9515, also known as Settings Flood; CVE-2019-9518, also known as Empty Frame Flooding; and similar undocumented scenarios, a frame rate limiting mechanism was added. By default Cowboy will now allow 1000 frames every 10 seconds. This can be configured via max_received_frame_rate. For CVE-2019-9514, also known as Reset Flood, another rate limiting mechanism was added and can be configured via max_reset_stream_rate. By default Cowboy will do up to 10 stream resets every 10 seconds. Finally, nothing was done for CVE-2019-9513, also known as Resource Loop, because Cowboy does not currently implement the HTTP/2 priority mechanism (in parts because these issues were well known from the start). Tests were added for all cases except Internal Data Buffering, which I'm not sure how to test, and Resource Loop, which is not currently relevant.
2019-10-02 10:44:45 +02:00
{"/resp/:key[/:arg]", resp_h, []}
]}]).
%% Tests.
http2_cancel_flood(Config) ->
doc("Confirm that Cowboy detects the rapid reset attack. (CVE-2023-44487)"),
do_http2_cancel_flood(Config, 1, 500),
do_http2_cancel_flood(Config, 10, 50),
do_http2_cancel_flood(Config, 500, 1),
ok.
do_http2_cancel_flood(Config, NumStreamsPerBatch, NumBatches) ->
{ok, Socket} = rfc7540_SUITE:do_handshake(Config),
{HeadersBlock, _} = cow_hpack:encode([
{<<":method">>, <<"GET">>},
{<<":scheme">>, <<"http">>},
{<<":authority">>, <<"localhost">>}, %% @todo Correct port number.
{<<":path">>, <<"/delay_hello">>}
]),
AllStreamIDs = lists:seq(1, NumBatches * NumStreamsPerBatch * 2, 2),
_ = lists:foldl(
fun (_BatchNumber, AvailableStreamIDs) ->
%% Take a bunch of IDs from the available stream IDs.
%% Send HEADERS for all these and then cancel them.
{IDs, RemainingStreamIDs} = lists:split(NumStreamsPerBatch, AvailableStreamIDs),
_ = gen_tcp:send(Socket, [cow_http2:headers(ID, fin, HeadersBlock) || ID <- IDs]),
_ = gen_tcp:send(Socket, [<<4:24, 3:8, 0:8, ID:32, 8:32>> || ID <- IDs]),
RemainingStreamIDs
end,
AllStreamIDs,
lists:seq(1, NumBatches, 1)),
%% When Cowboy detects a flood it must close the connection.
case gen_tcp:recv(Socket, 17, 6000) of
{ok, <<_:24, 7:8, 0:8, 0:32, _LastStreamId:32, 11:32>>} ->
%% GOAWAY with error code 11 = ENHANCE_YOUR_CALM.
ok;
%% We also accept the connection being closed immediately,
%% which may happen because we send the GOAWAY right before closing.
{error, closed} ->
ok
end.
Fix HTTP/2 CVEs A number of HTTP/2 CVEs were documented recently: https://www.kb.cert.org/vuls/id/605641/ This commit, along with a few changes and additions in Cowlib, fix or improve protection against all of them. For CVE-2019-9511, also known as Data Dribble, the new option stream_window_data_threshold can be used to control how little the DATA frames that Cowboy sends can get. For CVE-2019-9516, also known as 0-Length Headers Leak, Cowboy will now simply reject streams containing 0-length header names. For CVE-2019-9517, also known as Internal Data Buffering, the backpressure changes were already pretty good at preventing this issue, but a new option max_connection_buffer_size was added for even better control over how much memory we are willing to allocate. For CVE-2019-9512, also known as Ping Flood; CVE-2019-9515, also known as Settings Flood; CVE-2019-9518, also known as Empty Frame Flooding; and similar undocumented scenarios, a frame rate limiting mechanism was added. By default Cowboy will now allow 1000 frames every 10 seconds. This can be configured via max_received_frame_rate. For CVE-2019-9514, also known as Reset Flood, another rate limiting mechanism was added and can be configured via max_reset_stream_rate. By default Cowboy will do up to 10 stream resets every 10 seconds. Finally, nothing was done for CVE-2019-9513, also known as Resource Loop, because Cowboy does not currently implement the HTTP/2 priority mechanism (in parts because these issues were well known from the start). Tests were added for all cases except Internal Data Buffering, which I'm not sure how to test, and Resource Loop, which is not currently relevant.
2019-10-02 10:44:45 +02:00
http2_data_dribble(Config) ->
doc("Request a very large response then update the window 1 byte at a time. (CVE-2019-9511)"),
{ok, Socket} = rfc7540_SUITE:do_handshake(Config),
%% Send a GET request for a very large response.
{HeadersBlock, _} = cow_hpack:encode([
{<<":method">>, <<"GET">>},
{<<":scheme">>, <<"http">>},
{<<":authority">>, <<"localhost">>}, %% @todo Correct port number.
{<<":path">>, <<"/resp/stream_body/loop">>}
]),
ok = gen_tcp:send(Socket, cow_http2:headers(1, fin, HeadersBlock)),
%% Receive a response with a few DATA frames draining the window.
{ok, <<SkipLen:24, 1:8, _:8, 1:32>>} = gen_tcp:recv(Socket, 9, 1000),
{ok, _} = gen_tcp:recv(Socket, SkipLen, 1000),
{ok, <<16384:24, 0:8, 0:8, 1:32, _:16384/unit:8>>} = gen_tcp:recv(Socket, 9 + 16384, 1000),
{ok, <<16384:24, 0:8, 0:8, 1:32, _:16384/unit:8>>} = gen_tcp:recv(Socket, 9 + 16384, 1000),
{ok, <<16384:24, 0:8, 0:8, 1:32, _:16384/unit:8>>} = gen_tcp:recv(Socket, 9 + 16384, 1000),
{ok, <<16383:24, 0:8, 0:8, 1:32, _:16383/unit:8>>} = gen_tcp:recv(Socket, 9 + 16383, 1000),
%% Send WINDOW_UPDATE frames with a value of 1. The server should
%% not attempt to send data until the window is over a configurable threshold.
ok = gen_tcp:send(Socket, [
cow_http2:window_update(1),
cow_http2:window_update(1, 1)
]),
{error, timeout} = gen_tcp:recv(Socket, 0, 1000),
ok.
http2_empty_frame_flooding_data(Config) ->
doc("Confirm that Cowboy detects empty DATA frame flooding. (CVE-2019-9518)"),
{ok, Socket} = rfc7540_SUITE:do_handshake(Config),
%% Send a POST request followed by many empty DATA frames.
{HeadersBlock, _} = cow_hpack:encode([
{<<":method">>, <<"POST">>},
{<<":scheme">>, <<"http">>},
{<<":authority">>, <<"localhost">>}, %% @todo Correct port number.
{<<":path">>, <<"/echo/read_body">>}
]),
ok = gen_tcp:send(Socket, cow_http2:headers(1, nofin, HeadersBlock)),
_ = [gen_tcp:send(Socket, cow_http2:data(1, nofin, <<>>)) || _ <- lists:seq(1, 20000)],
Fix HTTP/2 CVEs A number of HTTP/2 CVEs were documented recently: https://www.kb.cert.org/vuls/id/605641/ This commit, along with a few changes and additions in Cowlib, fix or improve protection against all of them. For CVE-2019-9511, also known as Data Dribble, the new option stream_window_data_threshold can be used to control how little the DATA frames that Cowboy sends can get. For CVE-2019-9516, also known as 0-Length Headers Leak, Cowboy will now simply reject streams containing 0-length header names. For CVE-2019-9517, also known as Internal Data Buffering, the backpressure changes were already pretty good at preventing this issue, but a new option max_connection_buffer_size was added for even better control over how much memory we are willing to allocate. For CVE-2019-9512, also known as Ping Flood; CVE-2019-9515, also known as Settings Flood; CVE-2019-9518, also known as Empty Frame Flooding; and similar undocumented scenarios, a frame rate limiting mechanism was added. By default Cowboy will now allow 1000 frames every 10 seconds. This can be configured via max_received_frame_rate. For CVE-2019-9514, also known as Reset Flood, another rate limiting mechanism was added and can be configured via max_reset_stream_rate. By default Cowboy will do up to 10 stream resets every 10 seconds. Finally, nothing was done for CVE-2019-9513, also known as Resource Loop, because Cowboy does not currently implement the HTTP/2 priority mechanism (in parts because these issues were well known from the start). Tests were added for all cases except Internal Data Buffering, which I'm not sure how to test, and Resource Loop, which is not currently relevant.
2019-10-02 10:44:45 +02:00
%% When Cowboy detects a flood it must close the connection.
%% We skip WINDOW_UPDATE frames sent when Cowboy starts to read the body.
case gen_tcp:recv(Socket, 43, 6000) of
{ok, <<_:26/unit:8, _:24, 7:8, _:72, 11:32>>} ->
ok;
%% We also accept the connection being closed immediately,
%% which may happen because we send the GOAWAY right before closing.
{error, closed} ->
ok;
%% At least on Windows this might also occur.
{error, enotconn} ->
Fix HTTP/2 CVEs A number of HTTP/2 CVEs were documented recently: https://www.kb.cert.org/vuls/id/605641/ This commit, along with a few changes and additions in Cowlib, fix or improve protection against all of them. For CVE-2019-9511, also known as Data Dribble, the new option stream_window_data_threshold can be used to control how little the DATA frames that Cowboy sends can get. For CVE-2019-9516, also known as 0-Length Headers Leak, Cowboy will now simply reject streams containing 0-length header names. For CVE-2019-9517, also known as Internal Data Buffering, the backpressure changes were already pretty good at preventing this issue, but a new option max_connection_buffer_size was added for even better control over how much memory we are willing to allocate. For CVE-2019-9512, also known as Ping Flood; CVE-2019-9515, also known as Settings Flood; CVE-2019-9518, also known as Empty Frame Flooding; and similar undocumented scenarios, a frame rate limiting mechanism was added. By default Cowboy will now allow 1000 frames every 10 seconds. This can be configured via max_received_frame_rate. For CVE-2019-9514, also known as Reset Flood, another rate limiting mechanism was added and can be configured via max_reset_stream_rate. By default Cowboy will do up to 10 stream resets every 10 seconds. Finally, nothing was done for CVE-2019-9513, also known as Resource Loop, because Cowboy does not currently implement the HTTP/2 priority mechanism (in parts because these issues were well known from the start). Tests were added for all cases except Internal Data Buffering, which I'm not sure how to test, and Resource Loop, which is not currently relevant.
2019-10-02 10:44:45 +02:00
ok
end.
http2_empty_frame_flooding_headers_continuation(Config) ->
doc("Confirm that Cowboy detects empty HEADERS/CONTINUATION frame flooding. (CVE-2019-9518)"),
{ok, Socket} = rfc7540_SUITE:do_handshake(Config),
%% Send many empty HEADERS/CONTINUATION frames before the headers.
ok = gen_tcp:send(Socket, <<0:24, 1:8, 0:9, 1:31>>),
_ = [gen_tcp:send(Socket, <<0:24, 9:8, 0:9, 1:31>>) || _ <- lists:seq(1, 20000)],
Fix HTTP/2 CVEs A number of HTTP/2 CVEs were documented recently: https://www.kb.cert.org/vuls/id/605641/ This commit, along with a few changes and additions in Cowlib, fix or improve protection against all of them. For CVE-2019-9511, also known as Data Dribble, the new option stream_window_data_threshold can be used to control how little the DATA frames that Cowboy sends can get. For CVE-2019-9516, also known as 0-Length Headers Leak, Cowboy will now simply reject streams containing 0-length header names. For CVE-2019-9517, also known as Internal Data Buffering, the backpressure changes were already pretty good at preventing this issue, but a new option max_connection_buffer_size was added for even better control over how much memory we are willing to allocate. For CVE-2019-9512, also known as Ping Flood; CVE-2019-9515, also known as Settings Flood; CVE-2019-9518, also known as Empty Frame Flooding; and similar undocumented scenarios, a frame rate limiting mechanism was added. By default Cowboy will now allow 1000 frames every 10 seconds. This can be configured via max_received_frame_rate. For CVE-2019-9514, also known as Reset Flood, another rate limiting mechanism was added and can be configured via max_reset_stream_rate. By default Cowboy will do up to 10 stream resets every 10 seconds. Finally, nothing was done for CVE-2019-9513, also known as Resource Loop, because Cowboy does not currently implement the HTTP/2 priority mechanism (in parts because these issues were well known from the start). Tests were added for all cases except Internal Data Buffering, which I'm not sure how to test, and Resource Loop, which is not currently relevant.
2019-10-02 10:44:45 +02:00
{HeadersBlock, _} = cow_hpack:encode([
{<<":method">>, <<"POST">>},
{<<":scheme">>, <<"http">>},
{<<":authority">>, <<"localhost">>}, %% @todo Correct port number.
{<<":path">>, <<"/">>}
]),
Len = iolist_size(HeadersBlock),
_ = gen_tcp:send(Socket, [<<Len:24, 9:8, 0:5, 1:1, 0:1, 1:1, 0:1, 1:31>>, HeadersBlock]),
%% When Cowboy detects a flood it must close the connection.
case gen_tcp:recv(Socket, 17, 6000) of
{ok, <<_:24, 7:8, _:72, 11:32>>} ->
ok;
%% We also accept the connection being closed immediately,
%% which may happen because we send the GOAWAY right before closing.
{error, closed} ->
ok;
%% At least on Windows this might also occur.
{error, enotconn} ->
Fix HTTP/2 CVEs A number of HTTP/2 CVEs were documented recently: https://www.kb.cert.org/vuls/id/605641/ This commit, along with a few changes and additions in Cowlib, fix or improve protection against all of them. For CVE-2019-9511, also known as Data Dribble, the new option stream_window_data_threshold can be used to control how little the DATA frames that Cowboy sends can get. For CVE-2019-9516, also known as 0-Length Headers Leak, Cowboy will now simply reject streams containing 0-length header names. For CVE-2019-9517, also known as Internal Data Buffering, the backpressure changes were already pretty good at preventing this issue, but a new option max_connection_buffer_size was added for even better control over how much memory we are willing to allocate. For CVE-2019-9512, also known as Ping Flood; CVE-2019-9515, also known as Settings Flood; CVE-2019-9518, also known as Empty Frame Flooding; and similar undocumented scenarios, a frame rate limiting mechanism was added. By default Cowboy will now allow 1000 frames every 10 seconds. This can be configured via max_received_frame_rate. For CVE-2019-9514, also known as Reset Flood, another rate limiting mechanism was added and can be configured via max_reset_stream_rate. By default Cowboy will do up to 10 stream resets every 10 seconds. Finally, nothing was done for CVE-2019-9513, also known as Resource Loop, because Cowboy does not currently implement the HTTP/2 priority mechanism (in parts because these issues were well known from the start). Tests were added for all cases except Internal Data Buffering, which I'm not sure how to test, and Resource Loop, which is not currently relevant.
2019-10-02 10:44:45 +02:00
ok
end.
http2_empty_frame_flooding_push_promise(Config) ->
doc("Confirm that Cowboy detects empty PUSH_PROMISE frame flooding. (CVE-2019-9518)"),
{ok, Socket} = rfc7540_SUITE:do_handshake(Config),
%% Send a HEADERS frame to which we will attach a PUSH_PROMISE.
%% We use nofin in order to keep the stream alive.
{HeadersBlock, _} = cow_hpack:encode([
{<<":method">>, <<"GET">>},
{<<":scheme">>, <<"http">>},
{<<":authority">>, <<"localhost">>}, %% @todo Correct port number.
2019-10-07 12:18:03 +02:00
{<<":path">>, <<"/long_polling">>}
Fix HTTP/2 CVEs A number of HTTP/2 CVEs were documented recently: https://www.kb.cert.org/vuls/id/605641/ This commit, along with a few changes and additions in Cowlib, fix or improve protection against all of them. For CVE-2019-9511, also known as Data Dribble, the new option stream_window_data_threshold can be used to control how little the DATA frames that Cowboy sends can get. For CVE-2019-9516, also known as 0-Length Headers Leak, Cowboy will now simply reject streams containing 0-length header names. For CVE-2019-9517, also known as Internal Data Buffering, the backpressure changes were already pretty good at preventing this issue, but a new option max_connection_buffer_size was added for even better control over how much memory we are willing to allocate. For CVE-2019-9512, also known as Ping Flood; CVE-2019-9515, also known as Settings Flood; CVE-2019-9518, also known as Empty Frame Flooding; and similar undocumented scenarios, a frame rate limiting mechanism was added. By default Cowboy will now allow 1000 frames every 10 seconds. This can be configured via max_received_frame_rate. For CVE-2019-9514, also known as Reset Flood, another rate limiting mechanism was added and can be configured via max_reset_stream_rate. By default Cowboy will do up to 10 stream resets every 10 seconds. Finally, nothing was done for CVE-2019-9513, also known as Resource Loop, because Cowboy does not currently implement the HTTP/2 priority mechanism (in parts because these issues were well known from the start). Tests were added for all cases except Internal Data Buffering, which I'm not sure how to test, and Resource Loop, which is not currently relevant.
2019-10-02 10:44:45 +02:00
]),
ok = gen_tcp:send(Socket, cow_http2:headers(1, nofin, HeadersBlock)),
%% Send nofin PUSH_PROMISE frame without any data.
ok = gen_tcp:send(Socket, <<4:24, 5:8, 0:8, 0:1, 1:31, 0:1, 3:31>>),
%% Receive a PROTOCOL_ERROR connection error.
%%
%% Cowboy rejects all PUSH_PROMISE frames therefore no flooding
%% can take place.
{ok, <<_:24, 7:8, _:72, 1:32>>} = gen_tcp:recv(Socket, 17, 6000),
ok.
http2_infinite_continuations(Config) ->
doc("Confirm that Cowboy rejects CONTINUATION frames when the "
"total size of HEADERS + CONTINUATION(s) exceeds the limit. (VU#421644)"),
{ok, Socket} = rfc7540_SUITE:do_handshake(Config),
%% Send a HEADERS frame followed by a large number
%% of continuation frames.
{HeadersBlock, _} = cow_hpack:encode([
{<<":method">>, <<"GET">>},
{<<":scheme">>, <<"http">>},
{<<":authority">>, <<"localhost">>}, %% @todo Correct port number.
{<<":path">>, <<"/">>}
]),
HeadersBlockLen = iolist_size(HeadersBlock),
ok = gen_tcp:send(Socket, [
%% HEADERS frame.
<<
HeadersBlockLen:24, 1:8, 0:5,
0:1, %% END_HEADERS
0:1,
1:1, %% END_STREAM
0:1,
1:31 %% Stream ID.
>>,
HeadersBlock,
%% CONTINUATION frames.
[<<1024:24, 9:8, 0:8, 0:1, 1:31, 0:1024/unit:8>>
|| _ <- lists:seq(1, 100)]
]),
%% Receive an ENHANCE_YOUR_CALM connection error.
{ok, <<_:24, 7:8, _:72, 11:32>>} = gen_tcp:recv(Socket, 17, 6000),
ok.
Fix HTTP/2 CVEs A number of HTTP/2 CVEs were documented recently: https://www.kb.cert.org/vuls/id/605641/ This commit, along with a few changes and additions in Cowlib, fix or improve protection against all of them. For CVE-2019-9511, also known as Data Dribble, the new option stream_window_data_threshold can be used to control how little the DATA frames that Cowboy sends can get. For CVE-2019-9516, also known as 0-Length Headers Leak, Cowboy will now simply reject streams containing 0-length header names. For CVE-2019-9517, also known as Internal Data Buffering, the backpressure changes were already pretty good at preventing this issue, but a new option max_connection_buffer_size was added for even better control over how much memory we are willing to allocate. For CVE-2019-9512, also known as Ping Flood; CVE-2019-9515, also known as Settings Flood; CVE-2019-9518, also known as Empty Frame Flooding; and similar undocumented scenarios, a frame rate limiting mechanism was added. By default Cowboy will now allow 1000 frames every 10 seconds. This can be configured via max_received_frame_rate. For CVE-2019-9514, also known as Reset Flood, another rate limiting mechanism was added and can be configured via max_reset_stream_rate. By default Cowboy will do up to 10 stream resets every 10 seconds. Finally, nothing was done for CVE-2019-9513, also known as Resource Loop, because Cowboy does not currently implement the HTTP/2 priority mechanism (in parts because these issues were well known from the start). Tests were added for all cases except Internal Data Buffering, which I'm not sure how to test, and Resource Loop, which is not currently relevant.
2019-10-02 10:44:45 +02:00
%% @todo http2_internal_data_buffering(Config) -> I do not know how to test this.
% doc("Request many very large responses, with a larger than necessary window size, "
% "but do not attempt to read from the socket. (CVE-2019-9517)"),
http2_ping_flood(Config) ->
doc("Confirm that Cowboy detects PING floods. (CVE-2019-9512)"),
{ok, Socket} = rfc7540_SUITE:do_handshake(Config),
%% Flood the server with PING frames.
_ = [gen_tcp:send(Socket, cow_http2:ping(0)) || _ <- lists:seq(1, 20000)],
Fix HTTP/2 CVEs A number of HTTP/2 CVEs were documented recently: https://www.kb.cert.org/vuls/id/605641/ This commit, along with a few changes and additions in Cowlib, fix or improve protection against all of them. For CVE-2019-9511, also known as Data Dribble, the new option stream_window_data_threshold can be used to control how little the DATA frames that Cowboy sends can get. For CVE-2019-9516, also known as 0-Length Headers Leak, Cowboy will now simply reject streams containing 0-length header names. For CVE-2019-9517, also known as Internal Data Buffering, the backpressure changes were already pretty good at preventing this issue, but a new option max_connection_buffer_size was added for even better control over how much memory we are willing to allocate. For CVE-2019-9512, also known as Ping Flood; CVE-2019-9515, also known as Settings Flood; CVE-2019-9518, also known as Empty Frame Flooding; and similar undocumented scenarios, a frame rate limiting mechanism was added. By default Cowboy will now allow 1000 frames every 10 seconds. This can be configured via max_received_frame_rate. For CVE-2019-9514, also known as Reset Flood, another rate limiting mechanism was added and can be configured via max_reset_stream_rate. By default Cowboy will do up to 10 stream resets every 10 seconds. Finally, nothing was done for CVE-2019-9513, also known as Resource Loop, because Cowboy does not currently implement the HTTP/2 priority mechanism (in parts because these issues were well known from the start). Tests were added for all cases except Internal Data Buffering, which I'm not sure how to test, and Resource Loop, which is not currently relevant.
2019-10-02 10:44:45 +02:00
%% Receive a number of PING ACK frames in return, following by the closing of the connection.
try
[case gen_tcp:recv(Socket, 17, 6000) of
{ok, <<8:24, 6:8, _:7, 1:1, _:32, 0:64>>} -> ok;
{ok, <<_:24, 7:8, _:72, 11:32>>} -> throw(goaway);
%% We also accept the connection being closed immediately,
%% which may happen because we send the GOAWAY right before closing.
{error, closed} -> throw(goaway)
end || _ <- lists:seq(1, 20000)],
Fix HTTP/2 CVEs A number of HTTP/2 CVEs were documented recently: https://www.kb.cert.org/vuls/id/605641/ This commit, along with a few changes and additions in Cowlib, fix or improve protection against all of them. For CVE-2019-9511, also known as Data Dribble, the new option stream_window_data_threshold can be used to control how little the DATA frames that Cowboy sends can get. For CVE-2019-9516, also known as 0-Length Headers Leak, Cowboy will now simply reject streams containing 0-length header names. For CVE-2019-9517, also known as Internal Data Buffering, the backpressure changes were already pretty good at preventing this issue, but a new option max_connection_buffer_size was added for even better control over how much memory we are willing to allocate. For CVE-2019-9512, also known as Ping Flood; CVE-2019-9515, also known as Settings Flood; CVE-2019-9518, also known as Empty Frame Flooding; and similar undocumented scenarios, a frame rate limiting mechanism was added. By default Cowboy will now allow 1000 frames every 10 seconds. This can be configured via max_received_frame_rate. For CVE-2019-9514, also known as Reset Flood, another rate limiting mechanism was added and can be configured via max_reset_stream_rate. By default Cowboy will do up to 10 stream resets every 10 seconds. Finally, nothing was done for CVE-2019-9513, also known as Resource Loop, because Cowboy does not currently implement the HTTP/2 priority mechanism (in parts because these issues were well known from the start). Tests were added for all cases except Internal Data Buffering, which I'm not sure how to test, and Resource Loop, which is not currently relevant.
2019-10-02 10:44:45 +02:00
error(flood_successful)
catch throw:goaway ->
ok
end.
http2_reset_flood(Config) ->
doc("Confirm that Cowboy detects reset floods. (CVE-2019-9514)"),
{ok, Socket} = rfc7540_SUITE:do_handshake(Config),
%% Flood the server with HEADERS frames without a :method pseudo-header.
{HeadersBlock, _} = cow_hpack:encode([
{<<":scheme">>, <<"http">>},
{<<":authority">>, <<"localhost">>}, %% @todo Correct port number.
{<<":path">>, <<"/">>}
]),
_ = [gen_tcp:send(Socket, cow_http2:headers(ID, fin, HeadersBlock)) || ID <- lists:seq(1, 100, 2)],
%% Receive a number of RST_STREAM frames in return, following by the closing of the connection.
try
[case gen_tcp:recv(Socket, 13, 6000) of
{ok, <<_:24, 3:8, _:8, ID:32, 1:32>>} -> ok;
{ok, <<_:24, 7:8, _:72>>} ->
{ok, <<11:32>>} = gen_tcp:recv(Socket, 4, 1000),
throw(goaway);
%% We also accept the connection being closed immediately,
%% which may happen because we send the GOAWAY right before closing.
{error, closed} ->
throw(goaway)
end || ID <- lists:seq(1, 100, 2)],
error(flood_successful)
catch throw:goaway ->
ok
end.
%% @todo If we ever implement the PRIORITY mechanism, this test should
%% be implemented as well. CVE-2019-9513 https://www.kb.cert.org/vuls/id/605641/
%% http2_resource_loop
http2_settings_flood(Config) ->
doc("Confirm that Cowboy detects SETTINGS floods. (CVE-2019-9515)"),
{ok, Socket} = rfc7540_SUITE:do_handshake(Config),
%% Flood the server with empty SETTINGS frames.
_ = [gen_tcp:send(Socket, cow_http2:settings(#{})) || _ <- lists:seq(1, 20000)],
Fix HTTP/2 CVEs A number of HTTP/2 CVEs were documented recently: https://www.kb.cert.org/vuls/id/605641/ This commit, along with a few changes and additions in Cowlib, fix or improve protection against all of them. For CVE-2019-9511, also known as Data Dribble, the new option stream_window_data_threshold can be used to control how little the DATA frames that Cowboy sends can get. For CVE-2019-9516, also known as 0-Length Headers Leak, Cowboy will now simply reject streams containing 0-length header names. For CVE-2019-9517, also known as Internal Data Buffering, the backpressure changes were already pretty good at preventing this issue, but a new option max_connection_buffer_size was added for even better control over how much memory we are willing to allocate. For CVE-2019-9512, also known as Ping Flood; CVE-2019-9515, also known as Settings Flood; CVE-2019-9518, also known as Empty Frame Flooding; and similar undocumented scenarios, a frame rate limiting mechanism was added. By default Cowboy will now allow 1000 frames every 10 seconds. This can be configured via max_received_frame_rate. For CVE-2019-9514, also known as Reset Flood, another rate limiting mechanism was added and can be configured via max_reset_stream_rate. By default Cowboy will do up to 10 stream resets every 10 seconds. Finally, nothing was done for CVE-2019-9513, also known as Resource Loop, because Cowboy does not currently implement the HTTP/2 priority mechanism (in parts because these issues were well known from the start). Tests were added for all cases except Internal Data Buffering, which I'm not sure how to test, and Resource Loop, which is not currently relevant.
2019-10-02 10:44:45 +02:00
%% Receive a number of SETTINGS ACK frames in return, following by the closing of the connection.
try
[case gen_tcp:recv(Socket, 9, 6000) of
{ok, <<0:24, 4:8, 0:7, 1:1, 0:32>>} -> ok;
{ok, <<_:24, 7:8, _:40>>} ->
{ok, <<_:32, 11:32>>} = gen_tcp:recv(Socket, 8, 1000),
throw(goaway);
%% We also accept the connection being closed immediately,
%% which may happen because we send the GOAWAY right before closing.
{error, closed} ->
throw(goaway)
end || _ <- lists:seq(1, 20000)],
Fix HTTP/2 CVEs A number of HTTP/2 CVEs were documented recently: https://www.kb.cert.org/vuls/id/605641/ This commit, along with a few changes and additions in Cowlib, fix or improve protection against all of them. For CVE-2019-9511, also known as Data Dribble, the new option stream_window_data_threshold can be used to control how little the DATA frames that Cowboy sends can get. For CVE-2019-9516, also known as 0-Length Headers Leak, Cowboy will now simply reject streams containing 0-length header names. For CVE-2019-9517, also known as Internal Data Buffering, the backpressure changes were already pretty good at preventing this issue, but a new option max_connection_buffer_size was added for even better control over how much memory we are willing to allocate. For CVE-2019-9512, also known as Ping Flood; CVE-2019-9515, also known as Settings Flood; CVE-2019-9518, also known as Empty Frame Flooding; and similar undocumented scenarios, a frame rate limiting mechanism was added. By default Cowboy will now allow 1000 frames every 10 seconds. This can be configured via max_received_frame_rate. For CVE-2019-9514, also known as Reset Flood, another rate limiting mechanism was added and can be configured via max_reset_stream_rate. By default Cowboy will do up to 10 stream resets every 10 seconds. Finally, nothing was done for CVE-2019-9513, also known as Resource Loop, because Cowboy does not currently implement the HTTP/2 priority mechanism (in parts because these issues were well known from the start). Tests were added for all cases except Internal Data Buffering, which I'm not sure how to test, and Resource Loop, which is not currently relevant.
2019-10-02 10:44:45 +02:00
error(flood_successful)
catch throw:goaway ->
ok
end.
http2_zero_length_header_leak(Config) ->
doc("Confirm that Cowboy rejects HEADERS frame with a 0-length header name. (CVE-2019-9516)"),
{ok, Socket} = rfc7540_SUITE:do_handshake(Config),
%% Send a GET request with a 0-length header name.
{HeadersBlock, _} = cow_hpack:encode([
{<<":method">>, <<"GET">>},
{<<":scheme">>, <<"http">>},
{<<":authority">>, <<"localhost">>}, %% @todo Correct port number.
{<<":path">>, <<"/">>},
{<<>>, <<"CVE-2019-9516">>}
]),
ok = gen_tcp:send(Socket, cow_http2:headers(1, fin, HeadersBlock)),
%% Receive a PROTOCOL_ERROR stream error.
{ok, <<_:24, 3:8, _:8, 1:32, 1:32>>} = gen_tcp:recv(Socket, 13, 6000),
ok.
nc_rand(Config) ->
doc("Throw random garbage at the server, then check if it's still up."),
do_nc(Config, "/dev/urandom").
nc_zero(Config) ->
doc("Throw zeroes at the server, then check if it's still up."),
do_nc(Config, "/dev/zero").
do_nc(Config, Input) ->
Cat = os:find_executable("cat"),
Nc = os:find_executable("nc"),
case {Cat, Nc} of
{false, _} ->
Fix HTTP/2 CVEs A number of HTTP/2 CVEs were documented recently: https://www.kb.cert.org/vuls/id/605641/ This commit, along with a few changes and additions in Cowlib, fix or improve protection against all of them. For CVE-2019-9511, also known as Data Dribble, the new option stream_window_data_threshold can be used to control how little the DATA frames that Cowboy sends can get. For CVE-2019-9516, also known as 0-Length Headers Leak, Cowboy will now simply reject streams containing 0-length header names. For CVE-2019-9517, also known as Internal Data Buffering, the backpressure changes were already pretty good at preventing this issue, but a new option max_connection_buffer_size was added for even better control over how much memory we are willing to allocate. For CVE-2019-9512, also known as Ping Flood; CVE-2019-9515, also known as Settings Flood; CVE-2019-9518, also known as Empty Frame Flooding; and similar undocumented scenarios, a frame rate limiting mechanism was added. By default Cowboy will now allow 1000 frames every 10 seconds. This can be configured via max_received_frame_rate. For CVE-2019-9514, also known as Reset Flood, another rate limiting mechanism was added and can be configured via max_reset_stream_rate. By default Cowboy will do up to 10 stream resets every 10 seconds. Finally, nothing was done for CVE-2019-9513, also known as Resource Loop, because Cowboy does not currently implement the HTTP/2 priority mechanism (in parts because these issues were well known from the start). Tests were added for all cases except Internal Data Buffering, which I'm not sure how to test, and Resource Loop, which is not currently relevant.
2019-10-02 10:44:45 +02:00
{skip, "The cat executable was not found."};
{_, false} ->
Fix HTTP/2 CVEs A number of HTTP/2 CVEs were documented recently: https://www.kb.cert.org/vuls/id/605641/ This commit, along with a few changes and additions in Cowlib, fix or improve protection against all of them. For CVE-2019-9511, also known as Data Dribble, the new option stream_window_data_threshold can be used to control how little the DATA frames that Cowboy sends can get. For CVE-2019-9516, also known as 0-Length Headers Leak, Cowboy will now simply reject streams containing 0-length header names. For CVE-2019-9517, also known as Internal Data Buffering, the backpressure changes were already pretty good at preventing this issue, but a new option max_connection_buffer_size was added for even better control over how much memory we are willing to allocate. For CVE-2019-9512, also known as Ping Flood; CVE-2019-9515, also known as Settings Flood; CVE-2019-9518, also known as Empty Frame Flooding; and similar undocumented scenarios, a frame rate limiting mechanism was added. By default Cowboy will now allow 1000 frames every 10 seconds. This can be configured via max_received_frame_rate. For CVE-2019-9514, also known as Reset Flood, another rate limiting mechanism was added and can be configured via max_reset_stream_rate. By default Cowboy will do up to 10 stream resets every 10 seconds. Finally, nothing was done for CVE-2019-9513, also known as Resource Loop, because Cowboy does not currently implement the HTTP/2 priority mechanism (in parts because these issues were well known from the start). Tests were added for all cases except Internal Data Buffering, which I'm not sure how to test, and Resource Loop, which is not currently relevant.
2019-10-02 10:44:45 +02:00
{skip, "The nc executable was not found."};
_ ->
StrPort = integer_to_list(config(port, Config)),
_ = [
os:cmd("cat " ++ Input ++ " | nc localhost " ++ StrPort)
|| _ <- lists:seq(1, 100)],
ConnPid = gun_open(Config),
Ref = gun:get(ConnPid, "/"),
{response, _, 200, _} = gun:await(ConnPid, Ref),
ok
end.
slowloris(Config) ->
doc("Send request headers one byte at a time. "
"Confirm that the connection gets closed."),
Client = raw_open(Config),
try
[begin
ok = raw_send(Client, [C]),
timer:sleep(250)
end || C <- "GET / HTTP/1.1\r\nHost: localhost\r\n"
"User-Agent: Mozilla/5.0 (Windows; U; Windows NT 6.0; en-US)\r\n"
"Cookie: name=aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa\r\n\r\n"],
error(failure)
catch error:{badmatch, _} ->
ok
end.
slowloris_chunks(Config) ->
doc("Send request headers one line at a time. "
"Confirm that the connection gets closed."),
Client = raw_open(Config),
ok = raw_send(Client, "GET / HTTP/1.1\r\n"),
timer:sleep(300),
ok = raw_send(Client, "Host: localhost\r\n"),
timer:sleep(300),
Data = raw_recv_head(Client),
{'HTTP/1.1', 408, _, Rest} = cow_http:parse_status_line(Data),
{Headers, _} = cow_http:parse_headers(Rest),
{_, <<"close">>} = lists:keyfind(<<"connection">>, 1, Headers),
{error, closed} = raw_recv(Client, 0, 1000).