listen_socket: Support multiple options and allow socket options to be set before bind().

include/envoy/api/os_sys_calls.h

@@ -63,6 +63,17 @@ class OsSysCalls {
    * @see man 2 stat
    */
   virtual int stat(const char* pathname, struct stat* buf) PURE;
+
+  /**
+   * @see man 2 setsockopt
+   */
+  virtual int setsockopt(int sockfd, int level, int optname, const void* optval,
+                         socklen_t optlen) PURE;
+
+  /**
+   * @see man 2 getsockopt
+   */
+  virtual int getsockopt(int sockfd, int level, int optname, void* optval, socklen_t* optlen) PURE;
 };
 
 typedef std::unique_ptr<OsSysCalls> OsSysCallsPtr;

include/envoy/network/listen_socket.h

@@ -1,6 +1,7 @@
 #pragma once
 
 #include <memory>
+#include <vector>
 
 #include "envoy/common/pure.h"
 #include "envoy/network/address.h"
@@ -30,36 +31,41 @@ class Socket {
    */
   virtual void close() PURE;
 
+  enum class SocketState { PreBind, PostBind };
+
   /**
    * Visitor class for setting socket options.
    */
-  class Options {
+  class Option {
   public:
-    virtual ~Options() {}
+    virtual ~Option() {}
 
     /**
      * @param socket the socket on which to apply options.
+     * @param state the current state of the socket. Significant for options that can only be
+     *        set for some particular state of the socket.
      * @return true if succeeded, false otherwise.
      */
-    virtual bool setOptions(Socket& socket) const PURE;
+    virtual bool setOption(Socket& socket, SocketState state) const PURE;
 
     /**
-     * @return bits that can be used to separate connections based on the options. Should return
-     *         zero if connections with different options can be pooled together. This is limited
-     *         to 32 bits to allow these bits to be efficiently combined into a larger hash key
-     *         used in connection pool lookups.
+     * @param vector of bits that can be used to separate connections based on the options. Should
+     *        return zero if connections with different options can be pooled together. This is
+     *        limited to 32 bits to allow these bits to be efficiently combined into a larger hash
+     *        key used in connection pool lookups.
      */
-    virtual uint32_t hashKey() const PURE;
+    virtual void hashKey(std::vector<uint8_t>& key) const PURE;
   };
-  typedef std::shared_ptr<Options> OptionsSharedPtr;
+  typedef std::unique_ptr<Option> OptionPtr;
+  typedef std::shared_ptr<std::vector<OptionPtr>> OptionsSharedPtr;
 
   /**
-   * Set the socket options for later retrieval with options().
+   * Add a socket option visitor for later retrieval with options().
    */
-  virtual void setOptions(const OptionsSharedPtr&) PURE;
+  virtual void addOption(OptionPtr&&) PURE;
 
   /**
-   * @return the socket options stored earlier with setOptions(), if any.
+   * @return the socket options stored earlier with addOption() calls, if any.
    */
   virtual const OptionsSharedPtr& options() const PURE;
 };

include/envoy/server/filter_config.h

@@ -135,7 +135,7 @@ class ListenerFactoryContext : public FactoryContext {
   /**
    * Store socket options to be set on the listen socket before listening.
    */
-  virtual void setListenSocketOptions(const Network::Socket::OptionsSharedPtr& options) PURE;
+  virtual void addListenSocketOption(Network::Socket::OptionPtr&& option) PURE;
 };
 
 /**

include/envoy/server/listener_manager.h

@@ -23,11 +23,13 @@ class ListenerComponentFactory {
   /**
    * Creates a socket.
    * @param address supplies the socket's address.
+   * @param options to be set on the created socket just before calling 'bind()'.
    * @param bind_to_port supplies whether to actually bind the socket.
    * @return Network::SocketSharedPtr an initialized and potentially bound socket.
    */
   virtual Network::SocketSharedPtr
-  createListenSocket(Network::Address::InstanceConstSharedPtr address, bool bind_to_port) PURE;
+  createListenSocket(Network::Address::InstanceConstSharedPtr address,
+                     const Network::Socket::OptionsSharedPtr& options, bool bind_to_port) PURE;
 
   /**
    * Creates a list of filter factories.

source/common/api/os_sys_calls_impl.cc

@@ -35,5 +35,15 @@ void* OsSysCallsImpl::mmap(void* addr, size_t length, int prot, int flags, int f
 
 int OsSysCallsImpl::stat(const char* pathname, struct stat* buf) { return ::stat(pathname, buf); }
 
+int OsSysCallsImpl::setsockopt(int sockfd, int level, int optname, const void* optval,
+                               socklen_t optlen) {
+  return ::setsockopt(sockfd, level, optname, optval, optlen);
+}
+
+int OsSysCallsImpl::getsockopt(int sockfd, int level, int optname, void* optval,
+                               socklen_t* optlen) {
+  return ::getsockopt(sockfd, level, optname, optval, optlen);
+}
+
 } // namespace Api
 } // namespace Envoy

source/common/api/os_sys_calls_impl.h

@@ -19,6 +19,8 @@ class OsSysCallsImpl : public OsSysCalls {
   int ftruncate(int fd, off_t length) override;
   void* mmap(void* addr, size_t length, int prot, int flags, int fd, off_t offset) override;
   int stat(const char* pathname, struct stat* buf) override;
+  int setsockopt(int sockfd, int level, int optname, const void* optval, socklen_t optlen) override;
+  int getsockopt(int sockfd, int level, int optname, void* optval, socklen_t* optlen) override;
 };
 
 typedef ThreadSafeSingleton<OsSysCallsImpl> OsSysCallsSingleton;

source/common/network/connection_impl.cc

@@ -539,13 +539,15 @@ ClientConnectionImpl::ClientConnectionImpl(
     : ConnectionImpl(dispatcher, std::make_unique<ClientSocketImpl>(remote_address),
                      std::move(transport_socket), false) {
   if (options) {
-    if (!options->setOptions(*socket_)) {
-      // Set a special error state to ensure asynchronous close to give the owner of the
-      // ConnectionImpl a chance to add callbacks and detect the "disconnect".
-      immediate_error_event_ = ConnectionEvent::LocalClose;
-      // Trigger a write event to close this connection out-of-band.
-      file_event_->activate(Event::FileReadyType::Write);
-      return;
+    for (const auto& option : *options) {
+      if (!option->setOption(*socket_, Socket::SocketState::PreBind)) {
+        // Set a special error state to ensure asynchronous close to give the owner of the
+        // ConnectionImpl a chance to add callbacks and detect the "disconnect".
+        immediate_error_event_ = ConnectionEvent::LocalClose;
+        // Trigger a write event to close this connection out-of-band.
+        file_event_->activate(Event::FileReadyType::Write);
+        return;
+      }
     }
   }
   if (source_address != nullptr) {

source/common/network/listen_socket_impl.cc

@@ -29,21 +29,38 @@ void ListenSocketImpl::doBind() {
   }
 }
 
-TcpListenSocket::TcpListenSocket(const Address::InstanceConstSharedPtr& address, bool bind_to_port)
+void ListenSocketImpl::setListenSocketOptions(const Network::Socket::OptionsSharedPtr& options) {
+  if (options) {
+    for (const auto& option : *options) {
+      if (!option->setOption(*this, SocketState::PreBind)) {
+        throw EnvoyException("ListenSocket: Setting socket options failed");
+      }
+    }
+  }
+}
+
+TcpListenSocket::TcpListenSocket(const Address::InstanceConstSharedPtr& address,
+                                 const Network::Socket::OptionsSharedPtr& options,
+                                 bool bind_to_port)
     : ListenSocketImpl(address->socket(Address::SocketType::Stream), address) {
   RELEASE_ASSERT(fd_ != -1);
 
   int on = 1;
   int rc = setsockopt(fd_, SOL_SOCKET, SO_REUSEADDR, &on, sizeof(on));
   RELEASE_ASSERT(rc != -1);
 
+  setListenSocketOptions(options);
+
   if (bind_to_port) {
     doBind();
   }
 }
 
-TcpListenSocket::TcpListenSocket(int fd, const Address::InstanceConstSharedPtr& address)
-    : ListenSocketImpl(fd, address) {}
+TcpListenSocket::TcpListenSocket(int fd, const Address::InstanceConstSharedPtr& address,
+                                 const Network::Socket::OptionsSharedPtr& options)
+    : ListenSocketImpl(fd, address) {
+  setListenSocketOptions(options);
+}
 
 UdsListenSocket::UdsListenSocket(const Address::InstanceConstSharedPtr& address)
     : ListenSocketImpl(address->socket(Address::SocketType::Stream), address) {

source/common/network/listen_socket_impl.h

@@ -26,7 +26,12 @@ class SocketImpl : public virtual Socket {
       fd_ = -1;
     }
   }
-  void setOptions(const OptionsSharedPtr& options) override { options_ = options; }
+  void addOption(OptionPtr&& option) override {
+    if (!options_) {
+      options_ = std::make_shared<std::vector<OptionPtr>>();
+    }
+    options_->emplace_back(std::move(option));
+  }
   const OptionsSharedPtr& options() const override { return options_; }
 
 protected:
@@ -44,15 +49,18 @@ class ListenSocketImpl : public SocketImpl {
       : SocketImpl(fd, local_address) {}
 
   void doBind();
+  void setListenSocketOptions(const Network::Socket::OptionsSharedPtr& options);
 };
 
 /**
  * Wraps a unix socket.
  */
 class TcpListenSocket : public ListenSocketImpl {
 public:
-  TcpListenSocket(const Address::InstanceConstSharedPtr& address, bool bind_to_port);
-  TcpListenSocket(int fd, const Address::InstanceConstSharedPtr& address);
+  TcpListenSocket(const Address::InstanceConstSharedPtr& address,
+                  const Network::Socket::OptionsSharedPtr& options, bool bind_to_port);
+  TcpListenSocket(int fd, const Address::InstanceConstSharedPtr& address,
+                  const Network::Socket::OptionsSharedPtr& options);
 };
 
 typedef std::unique_ptr<TcpListenSocket> TcpListenSocketPtr;

source/common/upstream/cluster_manager_impl.cc

@@ -813,28 +813,31 @@ ClusterManagerImpl::ThreadLocalClusterManagerImpl::ClusterEntry::connPool(
   }
 
   // Inherit socket options from downstream connection, if set.
-  absl::optional<uint32_t> hash_key;
+  std::vector<uint8_t> hash_key = {uint8_t(protocol), uint8_t(priority)};
 
   // Use downstream connection socket options for computing connection pool hash key, if any.
   // This allows socket options to control connection pooling so that connections with
   // different options are not pooled together.
+  bool have_options = false;
   if (context && context->downstreamConnection()) {
     const Network::ConnectionSocket::OptionsSharedPtr& options =
         context->downstreamConnection()->socketOptions();
     if (options) {
-      hash_key = options->hashKey();
+      for (const auto& option : *options) {
+        have_options = true;
+        option->hashKey(hash_key);
+      }
     }
   }
 
   ConnPoolsContainer& container = parent_.host_http_conn_pool_map_[host];
-  const auto key = container.key(priority, protocol, hash_key ? hash_key.value() : 0);
-  if (!container.pools_[key]) {
-    container.pools_[key] = parent_.parent_.factory_.allocateConnPool(
+  if (!container.pools_[hash_key]) {
+    container.pools_[hash_key] = parent_.parent_.factory_.allocateConnPool(
         parent_.thread_local_dispatcher_, host, priority, protocol,
-        hash_key ? context->downstreamConnection()->socketOptions() : nullptr);
+        have_options ? context->downstreamConnection()->socketOptions() : nullptr);
   }
 
-  return container.pools_[key].get();
+  return container.pools_[hash_key].get();
 }
 
 ClusterManagerPtr ProdClusterManagerFactory::clusterManagerFromProto(

source/common/upstream/cluster_manager_impl.h

@@ -4,9 +4,9 @@
 #include <cstdint>
 #include <functional>
 #include <list>
+#include <map>
 #include <memory>
 #include <string>
-#include <unordered_map>
 #include <vector>
 
 #include "envoy/config/bootstrap/v2/bootstrap.pb.h"
@@ -199,17 +199,7 @@ class ClusterManagerImpl : public ClusterManager, Logger::Loggable<Logger::Id::u
    */
   struct ThreadLocalClusterManagerImpl : public ThreadLocal::ThreadLocalObject {
     struct ConnPoolsContainer {
-      typedef std::unordered_map<uint64_t, Http::ConnectionPool::InstancePtr> ConnPools;
-
-      uint64_t key(ResourcePriority priority, Http::Protocol protocol, uint32_t hash_key) {
-        // One bit needed for priority
-        static_assert(NumResourcePriorities == 2,
-                      "Fix shifts below to match number of bits needed for 'priority'");
-        // Two bits needed for protocol
-        static_assert(Http::NumProtocols <= 4,
-                      "Fix shifts below to match number of bits needed for 'protocol'");
-        return uint64_t(hash_key) << 3 | uint64_t(protocol) << 1 | uint64_t(priority);
-      }
+      typedef std::map<std::vector<uint8_t>, Http::ConnectionPool::InstancePtr> ConnPools;
 
       ConnPools pools_;
       uint64_t drains_remaining_{};

source/server/BUILD

@@ -202,6 +202,7 @@ envoy_cc_library(
         "//include/envoy/server:listener_manager_interface",
         "//include/envoy/server:transport_socket_config_interface",
         "//include/envoy/server:worker_interface",
+        "//source/common/api:os_sys_calls_lib",
         "//source/common/config:utility_lib",
         "//source/common/config:well_known_names",
         "//source/common/network:listen_socket_lib",

source/server/config_validation/server.h

@@ -101,6 +101,7 @@ class ValidationInstance : Logger::Loggable<Logger::Id::main>,
     return ProdListenerComponentFactory::createListenerFilterFactoryList_(filters, context);
   }
   Network::SocketSharedPtr createListenSocket(Network::Address::InstanceConstSharedPtr,
+                                              const Network::Socket::OptionsSharedPtr&,
                                               bool) override {
     // Returned sockets are not currently used so we can return nothing here safely vs. a
     // validation mock.

source/server/http/admin.cc

@@ -620,7 +620,7 @@ AdminImpl::AdminImpl(const std::string& access_log_path, const std::string& prof
                      Network::Address::InstanceConstSharedPtr address, Server::Instance& server,
                      Stats::ScopePtr&& listener_scope)
     : server_(server), profile_path_(profile_path),
-      socket_(new Network::TcpListenSocket(address, true)),
+      socket_(new Network::TcpListenSocket(address, nullptr, true)),
       stats_(Http::ConnectionManagerImpl::generateStats("http.admin.", server_.stats())),
       tracing_stats_(Http::ConnectionManagerImpl::generateTracingStats("http.admin.tracing.",
                                                                        server_.stats())),