Github user ogoodman commented on a diff in the pull request:
https://github.com/apache/trafficserver/pull/624#discussion_r62778716
--- Diff: lib/atscppapi/examples/websocket/WebSocket.cc ---
@@ -0,0 +1,246 @@
+#include "WebSocket.hh"
+
+#include <iostream>
+#include "ts/ink_base64.h"
+#include "openssl/evp.h"
+
+using namespace atscppapi;
+
+#define SAY(a) std::cout << a << std::endl;
+#define SHOW(a) std::cout << #a << " = " << a << std::endl
+
+void TSPluginInit(int argc, const char* argv[])
+{
+ RegisterGlobalPlugin("WebSocket", "Apache", "supp...@example.com");
+ new WebSocketInstaller();
+}
+
+
+// WebSocketInstaller
+
+WebSocketInstaller::WebSocketInstaller()
+ : GlobalPlugin(true /* ignore internal transactions */)
+{
+
GlobalPlugin::registerHook(Plugin::HOOK_READ_REQUEST_HEADERS_PRE_REMAP);
+}
+
+#define WS_DIGEST_MAX ATS_BASE64_ENCODE_DSTLEN(20)
+static const std::string magic = "258EAFA5-E914-47DA-95CA-C5AB0DC85B11";
+
+static std::string ws_digest(std::string const& key) {
+ EVP_MD_CTX digest;
+ EVP_MD_CTX_init(&digest);
+
+ if (!EVP_DigestInit_ex(&digest, EVP_sha1(), NULL)) {
+ EVP_MD_CTX_cleanup(&digest);
+ return "init-failed";
+ }
+ if (!EVP_DigestUpdate(&digest, key.data(), key.length())) {
+ EVP_MD_CTX_cleanup(&digest);
+ return "update1-failed";
+ }
+ if (!EVP_DigestUpdate(&digest, magic.data(), magic.length())) {
+ EVP_MD_CTX_cleanup(&digest);
+ return "update2-failed";
+ }
+
+ unsigned char hash_buf[EVP_MAX_MD_SIZE];
+ unsigned int hash_len = 0;
+ if (!EVP_DigestFinal_ex(&digest, hash_buf, &hash_len)) {
+ EVP_MD_CTX_cleanup(&digest);
+ return "final-failed";
+ }
+ EVP_MD_CTX_cleanup(&digest);
+ if (hash_len != 20) {
+ return "bad-hash-length";
+ }
+
+ char digest_buf[WS_DIGEST_MAX];
+ size_t digest_len = 0;
+
+ ats_base64_encode(hash_buf, hash_len, digest_buf, WS_DIGEST_MAX,
&digest_len);
+
+ return std::string((char*)digest_buf, digest_len);
+}
+
+
+void WebSocketInstaller::handleReadRequestHeadersPreRemap(Transaction
&transaction)
+{
+ SAY("Incoming request.");
+ transaction.addPlugin(new WebSocket(transaction));
+ transaction.resume();
+}
+
+// WebSocket implementation.
+
+WebSocket::WebSocket(Transaction& transaction)
+ : InterceptPlugin(transaction, InterceptPlugin::SERVER_INTERCEPT)
+ , pos_(0)
+ , ctrl_(0)
+ , msg_len_(0)
+ , ws_handshake_done_(false)
+{
+ ws_key_ =
transaction.getClientRequest().getHeaders().values("sec-websocket-key");
+ if (ws_key_.size()) {
+ printf("setting timeouts\n");
+ transaction.setTimeout(Transaction::TIMEOUT_ACTIVE, 844000000);
+ transaction.setTimeout(Transaction::TIMEOUT_NO_ACTIVITY, 84400000);
+ }
+}
+
+WebSocket::~WebSocket()
+{
+ SAY("WebSocket finished.");
+}
+
+void WebSocket::ws_send(std::string const& data, int code)
+{
+ std::string frame(1, char(code));
+ size_t len = data.length();
+ if (len <= 125) {
+ frame += char(len);
+ } else {
+ int len_len;
+ if (len <= 65535) {
+ frame += char(126);
+ len_len = 2;
+ } else {
+ frame += char(127);
+ len_len = 8;
+ }
+ while (--len_len >= 0) {
+ frame += char((len >> (8*len_len)) & 0xFF);
+ }
+ }
+ produce(frame + data);
+}
+
+void WebSocket::consume(const std::string &data,
InterceptPlugin::RequestDataType type)
+{
+ if (isWebsocket()) {
+ if (!ws_handshake_done_) {
+ std::string digest = ws_digest(ws_key_);
+
+ printf("WS key digest: %s %s\n", ws_key_.c_str(),
digest.c_str());
+
+ std::string headers;
+ headers +=
+ "HTTP/1.1 101 Switching Protocols\r\n"
+ "Upgrade: websocket\r\n"
+ "Connection: Upgrade\r\n"
+ "Sec-WebSocket-Accept: " + digest + "\r\n\r\n";
+ produce(headers);
+ ws_handshake_done_ = true;
+ }
+ }
+
+ // auto tid = std::this_thread::get_id();
+ if (type == InterceptPlugin::REQUEST_HEADER) {
+ headers_ += data;
+ } else {
+ // cout << tid << ": Read request body" << endl << data << endl;
+ if (isWebsocket()) {
+
+ // When we've read as much as we can for any incoming
+ // data chunk we remove the head of ws_buf_ and reset pos_
+ // to 0.
+ //
+ // If we can't read a complete length + masks, we leave
+ // pos_ unchanged and just append what we got to ws_buf_.
+ //
+ // When we get the length, we update pos_ and length.
+ // While length is non-zero we wait for ws_buf_ to have
+ // msg_len_ bytes beyond pos_.
+
+ ws_buf_ += data;
+ while (true) {
+ if (!msg_len_) {
+ size_t avail = ws_buf_.size() - pos_;
+
+ // Read the msg_length if we have enough data.
+ if (avail < 6) break; // 2 + 4 mask
+ ctrl_ = ws_buf_[pos_] & 0xF;
+ size_t msg_len = ws_buf_[pos_ + 1] & 0x7F;
+ if (msg_len == 0x7E) {
+ if (avail < 8) { // 2 + 2 + length bytes + 4 mask.
+ break;
+ }
+ msg_len = 0;
+ pos_ += 2;
+ for (int i = 0; i < 2; ++i, ++pos_) {
+ msg_len <<= 8;
+ msg_len += (unsigned char)(ws_buf_[pos_]);
--- End diff --
The code did handle network byte order correctly, but it is true that ntohs
and friends are more efficient on big-endian machines.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at infrastruct...@apache.org or file a JIRA ticket
with INFRA.
---
