Merge pull request #1442 from nghttp2/upgrade-llhttp

Bump llhttp to 2.0.4

third-party/llhttp/README.md

@@ -14,6 +14,8 @@ This project aims to:
 * Verifiable
 * Improving benchmarks where possible
 
+More details in [Fedor Indutny's talk at JSConf EU 2019](https://youtu.be/x3k_5Mi66sY)
+
 ## How?
 
 Over time, different approaches for improving [http_parser][0]'s code base
@@ -30,11 +32,10 @@ So far llhttp outperforms http_parser:
 
 |                 | input size |  bandwidth   |  reqs/sec  |   time  |
 |:----------------|-----------:|-------------:|-----------:|--------:|
-| **llhttp** _(C)_ | 8192.00 mb | 1497.88 mb/s | 3020458.87 ops/sec | 5.47 s |
+| **llhttp** _(C)_ | 8192.00 mb | 1777.24 mb/s | 3583799.39 ops/sec | 4.61 s |
-| **llhttp** _(bitcode)_ | 8192.00 mb | 1131.75 mb/s | 2282171.24 ops/sec | 7.24 s |
 | **http_parser** | 8192.00 mb | 694.66 mb/s | 1406180.33 req/sec | 11.79 s |
 
-llhttp is faster by approximately **116%**.
+llhttp is faster by approximately **156%**.
 
 ## Maintenance
 
@@ -77,8 +78,6 @@ settings.on_message_complete = handle_on_message_complete;
  */
 llhttp_init(&parser, HTTP_BOTH, &settings);
 
-/* Use `llhttp_set_type(&parser, HTTP_REQUEST);` to override the mode */
-
 /* Parse request! */
 const char* request = "GET / HTTP/1.1\r\n\r\n";
 int request_len = strlen(request);

third-party/llhttp/include/llhttp.h

@@ -1,9 +1,9 @@
 #ifndef INCLUDE_LLHTTP_H_
 #define INCLUDE_LLHTTP_H_
 
-#define LLHTTP_VERSION_MAJOR 1
+#define LLHTTP_VERSION_MAJOR 2
-#define LLHTTP_VERSION_MINOR 1
+#define LLHTTP_VERSION_MINOR 0
-#define LLHTTP_VERSION_PATCH 3
+#define LLHTTP_VERSION_PATCH 4
 
 #ifndef INCLUDE_LLHTTP_ITSELF_H_
 #define INCLUDE_LLHTTP_ITSELF_H_
@@ -29,7 +29,7 @@ struct llhttp__internal_s {
   uint8_t http_major;
   uint8_t http_minor;
   uint8_t header_state;
-  uint8_t flags;
+  uint16_t flags;
   uint8_t upgrade;
   uint16_t status_code;
   uint8_t finish;
@@ -66,14 +66,15 @@ enum llhttp_errno {
   HPE_INVALID_CHUNK_SIZE = 12,
   HPE_INVALID_STATUS = 13,
   HPE_INVALID_EOF_STATE = 14,
-  HPE_CB_MESSAGE_BEGIN = 15,
+  HPE_INVALID_TRANSFER_ENCODING = 15,
-  HPE_CB_HEADERS_COMPLETE = 16,
+  HPE_CB_MESSAGE_BEGIN = 16,
-  HPE_CB_MESSAGE_COMPLETE = 17,
+  HPE_CB_HEADERS_COMPLETE = 17,
-  HPE_CB_CHUNK_HEADER = 18,
+  HPE_CB_MESSAGE_COMPLETE = 18,
-  HPE_CB_CHUNK_COMPLETE = 19,
+  HPE_CB_CHUNK_HEADER = 19,
-  HPE_PAUSED = 20,
+  HPE_CB_CHUNK_COMPLETE = 20,
-  HPE_PAUSED_UPGRADE = 21,
+  HPE_PAUSED = 21,
-  HPE_USER = 22
+  HPE_PAUSED_UPGRADE = 22,
+  HPE_USER = 23
 };
 typedef enum llhttp_errno llhttp_errno_t;
 
@@ -85,7 +86,9 @@ enum llhttp_flags {
   F_UPGRADE = 0x10,
   F_CONTENT_LENGTH = 0x20,
   F_SKIPBODY = 0x40,
-  F_TRAILING = 0x80
+  F_TRAILING = 0x80,
+  F_LENIENT = 0x100,
+  F_TRANSFER_ENCODING = 0x200
 };
 typedef enum llhttp_flags llhttp_flags_t;
 
@@ -157,14 +160,15 @@ typedef enum llhttp_method llhttp_method_t;
   XX(12, INVALID_CHUNK_SIZE, INVALID_CHUNK_SIZE) \
   XX(13, INVALID_STATUS, INVALID_STATUS) \
   XX(14, INVALID_EOF_STATE, INVALID_EOF_STATE) \
-  XX(15, CB_MESSAGE_BEGIN, CB_MESSAGE_BEGIN) \
+  XX(15, INVALID_TRANSFER_ENCODING, INVALID_TRANSFER_ENCODING) \
-  XX(16, CB_HEADERS_COMPLETE, CB_HEADERS_COMPLETE) \
+  XX(16, CB_MESSAGE_BEGIN, CB_MESSAGE_BEGIN) \
-  XX(17, CB_MESSAGE_COMPLETE, CB_MESSAGE_COMPLETE) \
+  XX(17, CB_HEADERS_COMPLETE, CB_HEADERS_COMPLETE) \
-  XX(18, CB_CHUNK_HEADER, CB_CHUNK_HEADER) \
+  XX(18, CB_MESSAGE_COMPLETE, CB_MESSAGE_COMPLETE) \
-  XX(19, CB_CHUNK_COMPLETE, CB_CHUNK_COMPLETE) \
+  XX(19, CB_CHUNK_HEADER, CB_CHUNK_HEADER) \
-  XX(20, PAUSED, PAUSED) \
+  XX(20, CB_CHUNK_COMPLETE, CB_CHUNK_COMPLETE) \
-  XX(21, PAUSED_UPGRADE, PAUSED_UPGRADE) \
+  XX(21, PAUSED, PAUSED) \
-  XX(22, USER, USER) \
+  XX(22, PAUSED_UPGRADE, PAUSED_UPGRADE) \
+  XX(23, USER, USER) \
 
 
 #define HTTP_METHOD_MAP(XX) \
@@ -297,7 +301,7 @@ llhttp_errno_t llhttp_finish(llhttp_t* parser);
 int llhttp_message_needs_eof(const llhttp_t* parser);
 
 /* Returns `1` if there might be any other messages following the last that was
- * successfuly parsed.
+ * successfully parsed.
  */
 int llhttp_should_keep_alive(const llhttp_t* parser);
 
@@ -353,6 +357,18 @@ const char* llhttp_errno_name(llhttp_errno_t err);
 /* Returns textual name of HTTP method */
 const char* llhttp_method_name(llhttp_method_t method);
 
+
+/* Enables/disables lenient header value parsing (disabled by default).
+ *
+ * Lenient parsing disables header value token checks, extending llhttp's
+ * protocol support to highly non-compliant clients/server. No
+ * `HPE_INVALID_HEADER_TOKEN` will be raised for incorrect header values when
+ * lenient parsing is "on".
+ *
+ * **(USE AT YOUR OWN RISK)**
+ */
+void llhttp_set_lenient(llhttp_t* parser, int enabled);
+
 #ifdef __cplusplus
 }  /* extern "C" */
 #endif

third-party/llhttp/src/api.c

@@ -127,6 +127,15 @@ const char* llhttp_method_name(llhttp_method_t method) {
 }
 
 
+void llhttp_set_lenient(llhttp_t* parser, int enabled) {
+  if (enabled) {
+    parser->flags |= F_LENIENT;
+  } else {
+    parser->flags &= ~F_LENIENT;
+  }
+}
+
+
 /* Callbacks */
 
 

third-party/llhttp/src/http.c

@@ -32,6 +32,7 @@ int llhttp__before_headers_complete(llhttp_t* parser, const char* p,
  * 2 - chunk_size_start
  * 3 - body_identity
  * 4 - body_identity_eof
+ * 5 - invalid transfer-encoding for request
  */
 int llhttp__after_headers_complete(llhttp_t* parser, const char* p,
                                    const char* endp) {
@@ -47,8 +48,29 @@ int llhttp__after_headers_complete(llhttp_t* parser, const char* p,
   if (parser->flags & F_SKIPBODY) {
     return 0;
   } else if (parser->flags & F_CHUNKED) {
-    /* chunked encoding - ignore Content-Length header */
+    /* chunked encoding - ignore Content-Length header, prepare for a chunk */
     return 2;
+  } else if (parser->flags & F_TRANSFER_ENCODING) {
+    if (parser->type == HTTP_REQUEST && (parser->flags & F_LENIENT) == 0) {
+      /* RFC 7230 3.3.3 */
+
+      /* If a Transfer-Encoding header field
+       * is present in a request and the chunked transfer coding is not
+       * the final encoding, the message body length cannot be determined
+       * reliably; the server MUST respond with the 400 (Bad Request)
+       * status code and then close the connection.
+       */
+      return 5;
+    } else {
+      /* RFC 7230 3.3.3 */
+
+      /* If a Transfer-Encoding header field is present in a response and
+       * the chunked transfer coding is not the final encoding, the
+       * message body length is determined by reading the connection until
+       * it is closed by the server.
+       */
+      return 4;
+    }
   } else {
     if (!(parser->flags & F_CONTENT_LENGTH)) {
       if (!llhttp_message_needs_eof(parser)) {
@@ -74,9 +96,11 @@ int llhttp__after_message_complete(llhttp_t* parser, const char* p,
   int should_keep_alive;
 
   should_keep_alive = llhttp_should_keep_alive(parser);
-  parser->flags = 0;
   parser->finish = HTTP_FINISH_SAFE;
 
+  /* Keep `F_LENIENT` flag between messages, but reset every other flag */
+  parser->flags &= F_LENIENT;
+
   /* NOTE: this is ignored in loose parsing mode */
   return should_keep_alive;
 }
@@ -95,6 +119,12 @@ int llhttp_message_needs_eof(const llhttp_t* parser) {
     return 0;
   }
 
+  /* RFC 7230 3.3.3, see `llhttp__after_headers_complete` */
+  if ((parser->flags & F_TRANSFER_ENCODING) &&
+      (parser->flags & F_CHUNKED) == 0) {
+    return 1;
+  }
+
   if (parser->flags & (F_CHUNKED | F_CONTENT_LENGTH)) {
     return 0;
   }