[http] add llhttp parser implementation

bazel/BUILD

@@ -527,6 +527,11 @@ config_setting(
     define_values = {"FUZZING_ENGINE": "oss-fuzz"},
 )
 
+config_setting(
+    name = "enable_new_http1_parser_in_integration_tests",
+    values = {"define": "use_new_http1_parser_in_integration_tests=true"},
+)
+
 alias(
     name = "fuzzing_engine",
     actual = select({

bazel/envoy_build_system.bzl

@@ -21,6 +21,7 @@ load(
     _envoy_select_enable_http3 = "envoy_select_enable_http3",
     _envoy_select_google_grpc = "envoy_select_google_grpc",
     _envoy_select_hot_restart = "envoy_select_hot_restart",
+    _envoy_select_new_http1_parser_in_integration_tests = "envoy_select_new_http1_parser_in_integration_tests",
     _envoy_select_wasm_cpp_tests = "envoy_select_wasm_cpp_tests",
     _envoy_select_wasm_rust_tests = "envoy_select_wasm_rust_tests",
     _envoy_select_wasm_v8 = "envoy_select_wasm_v8",
@@ -205,6 +206,7 @@ def envoy_google_grpc_external_deps():
 envoy_select_boringssl = _envoy_select_boringssl
 envoy_select_google_grpc = _envoy_select_google_grpc
 envoy_select_enable_http3 = _envoy_select_enable_http3
+envoy_select_new_http1_parser_in_integration_tests = _envoy_select_new_http1_parser_in_integration_tests
 envoy_select_hot_restart = _envoy_select_hot_restart
 envoy_select_wasm_cpp_tests = _envoy_select_wasm_cpp_tests
 envoy_select_wasm_rust_tests = _envoy_select_wasm_rust_tests

bazel/envoy_select.bzl

@@ -79,3 +79,9 @@ def envoy_select_wasm_wasmtime(xs):
         "@envoy//bazel:wasm_wasmtime": xs,
         "//conditions:default": [],
     })
+
+def envoy_select_new_http1_parser_in_integration_tests(xs):
+    return select({
+        "@envoy//bazel:enable_new_http1_parser_in_integration_tests": xs,
+        "//conditions:default": [],
+    })

bazel/external/llhttp.BUILD

@@ -0,0 +1,14 @@
+licenses(["notice"])  # Apache 2
+
+cc_library(
+    name = "llhttp",
+    srcs = [
+        "src/api.c",
+        "src/http.c",
+        "src/llhttp.c",
+    ],
+    hdrs = ["include/llhttp.h"],
+    defines = ["LLHTTP_STRICT_MODE"],
+    includes = ["include"],
+    visibility = ["//visibility:public"],
+)

bazel/repositories.bzl

@@ -144,6 +144,7 @@ def envoy_dependencies(skip_targets = []):
     _com_github_nghttp2_nghttp2()
     _com_github_skyapm_cpp2sky()
     _com_github_nodejs_http_parser()
+    _com_github_nodejs_llhttp()
     _com_github_alibaba_hessian2_codec()
     _com_github_tencent_rapidjson()
     _com_github_nlohmann_json()
@@ -479,6 +480,16 @@ def _com_github_nodejs_http_parser():
         actual = "@com_github_nodejs_http_parser//:http_parser",
     )
 
+def _com_github_nodejs_llhttp():
+    external_http_archive(
+        name = "com_github_nodejs_llhttp",
+        build_file = "@envoy//bazel/external:llhttp.BUILD",
+    )
+    native.bind(
+        name = "llhttp",
+        actual = "@com_github_nodejs_llhttp//:llhttp",
+    )
+
 def _com_github_alibaba_hessian2_codec():
     external_http_archive("com_github_alibaba_hessian2_codec")
     native.bind(

bazel/repository_locations.bzl

@@ -490,6 +490,18 @@ REPOSITORY_LOCATIONS_SPEC = dict(
         release_date = "2020-07-10",
         cpe = "cpe:2.3:a:nodejs:node.js:*",
     ),
+    com_github_nodejs_llhttp = dict(
+        project_name = "llhttp",
+        project_desc = "Parser for HTTP messages written in C",
+        project_url = "https://github.com/nodejs/llhttp",
+        version = "5.1.0",
+        sha256 = "7140bbafa30f9cad4d20a0065953bd19bcc11203ab9a901ce7dea1b8182c8c78",
+        strip_prefix = "llhttp-release-v{version}",
+        urls = ["https://github.com/nodejs/llhttp/archive/release/v{version}.tar.gz"],
+        use_category = ["dataplane_core"],
+        release_date = "2021-04-06",
+        cpe = "cpe:2.3:a:nodejs:node.js:*",
+    ),
     com_github_alibaba_hessian2_codec = dict(
         project_name = "hessian2-codec",
         project_desc = "hessian2-codec is a C++ library for hessian2 codec",

ci/do_ci.sh

@@ -312,6 +312,7 @@ elif [[ "$CI_TARGET" == "bazel.compile_time_options" ]]; then
     "--define" "deprecated_features=disabled"
     "--define" "tcmalloc=gperftools"
     "--define" "zlib=ng"
+    "--define" "use_new_http1_parser_in_integration_tests=true"
     "--@envoy//bazel:http3=False"
     "--@envoy//source/extensions/filters/http/kill_request:enabled"
     "--test_env=ENVOY_HAS_EXTRA_EXTENSIONS=true")

docs/root/version_history/current.rst

@@ -140,6 +140,7 @@ New Features
 * ext_authz: added :ref:`allowed_client_headers_on_success <envoy_v3_api_field_extensions.filters.http.ext_authz.v3.AuthorizationResponse.allowed_client_headers_on_success>` to support sending response headers to downstream clients on OK external authorization checks via HTTP.
 * grpc_json_transcoder: added :ref:`request_validation_options <envoy_v3_api_field_extensions.filters.http.grpc_json_transcoder.v3.GrpcJsonTranscoder.request_validation_options>` to reject invalid requests early.
 * grpc_json_transcoder: filter can now be configured on per-route/per-vhost level as well. Leaving empty list of services in the filter configuration disables transcoding on the specific route.
+* http: added new HTTP/1.1 parser llhttp. This is off by default and can be enabled by setting the `envoy.reloadable_features.enable_new_http1_parser` runtime feature to true.
 * http: added support for `Envoy::ScopeTrackedObject` for HTTP/1 and HTTP/2 dispatching. Crashes while inside the dispatching loop should dump debug information. Furthermore, HTTP/1 and HTTP/2 clients now dumps the originating request whose response from the upstream caused Envoy to crash.
 * http: added support for :ref:`preconnecting <envoy_v3_api_msg_config.cluster.v3.Cluster.PreconnectPolicy>`. Preconnecting is off by default, but recommended for clusters serving latency-sensitive traffic, especially if using HTTP/1.1.
 * http: added support for stream filters to mutate the cached route set by HCM route resolution. Useful for filters in a filter chain that want to override specific methods/properties of a route. See :ref:`http route mutation <arch_overview_http_filters_route_mutation>` docs for more information.

source/common/http/http1/BUILD

@@ -36,6 +36,7 @@ envoy_cc_library(
         ":header_formatter_lib",
         ":legacy_parser_lib",
         ":parser_interface",
+        ":parser_lib",
         "//include/envoy/buffer:buffer_interface",
         "//include/envoy/common:scope_tracker_interface",
         "//include/envoy/http:codec_interface",
@@ -116,3 +117,14 @@ envoy_cc_library(
         "//source/common/common:assert_lib",
     ],
 )
+
+envoy_cc_library(
+    name = "parser_lib",
+    srcs = ["parser_impl.cc"],
+    hdrs = ["parser_impl.h"],
+    external_deps = ["llhttp"],
+    deps = [
+        ":parser_interface",
+        "//source/common/common:assert_lib",
+    ],
+)

source/common/http/http1/codec_impl.cc

@@ -21,6 +21,7 @@
 #include "common/http/headers.h"
 #include "common/http/http1/header_formatter.h"
 #include "common/http/http1/legacy_parser_impl.h"
+#include "common/http/http1/parser_impl.h"
 #include "common/http/utility.h"
 #include "common/runtime/runtime_features.h"
 
@@ -466,7 +467,11 @@ ConnectionImpl::ConnectionImpl(Network::Connection& connection, CodecStats& stat
                                []() -> void { /* TODO(adisuissa): Handle overflow watermark */ })),
       max_headers_kb_(max_headers_kb), max_headers_count_(max_headers_count) {
   output_buffer_->setWatermarks(connection.bufferLimit());
-  parser_ = std::make_unique<LegacyHttpParserImpl>(type, this);
+  if (Runtime::runtimeFeatureEnabled("envoy.reloadable_features.enable_new_http1_parser")) {
+    parser_ = std::make_unique<HttpParserImpl>(type, this);
+  } else {
+    parser_ = std::make_unique<LegacyHttpParserImpl>(type, this);
+  }
 }
 
 Status ConnectionImpl::completeLastHeader() {
@@ -659,6 +664,8 @@ Status ConnectionImpl::onHeaderValue(const char* data, size_t length) {
   }
 
   absl::string_view header_value{data, length};
+  // TODO(5155): This can be removed when http_parser is removed. llhttp enables strict validation
+  // of HTTP header values.
   if (!Http::HeaderUtility::headerValueIsValid(header_value)) {
     ENVOY_CONN_LOG(debug, "invalid header value: {}", connection_, header_value);
     error_code_ = Http::Code::BadRequest;
@@ -767,6 +774,7 @@ StatusOr<ParserStatus> ConnectionImpl::onHeadersComplete() {
       return codecProtocolError("http/1.1 protocol error: unsupported transfer encoding");
     }
   }
+  parser_->setHasContentLength(request_or_response_headers.ContentLength() != nullptr);
 
   auto statusor = onHeadersCompleteBase();
   if (!statusor.ok()) {

source/common/http/http1/legacy_parser_impl.h

@@ -22,6 +22,7 @@ class LegacyHttpParserImpl : public Parser {
   int httpMajor() const override;
   int httpMinor() const override;
   absl::optional<uint64_t> contentLength() const override;
+  void setHasContentLength(bool) override{};
   bool isChunked() const override;
   absl::string_view methodName() const override;
   absl::string_view errnoName(int rc) const override;

source/common/http/http1/parser.h

@@ -140,6 +140,10 @@ class Parser {
   // Returns the number of bytes in the body. absl::nullopt if no Content-Length header
   virtual absl::optional<uint64_t> contentLength() const PURE;
 
+  // Indicated that Content-Length header is present. This is used to differentiate between an unset
+  // Content-Length and a 0 value.
+  virtual void setHasContentLength(bool val) PURE;
+
   // Returns whether headers are chunked.
   virtual bool isChunked() const PURE;